commit 3c6a37b0e1934660f10f507a2edf0120cd0ceb64 Author: Cheng Date: Wed Feb 16 15:53:01 2022 +1000 leaving potentially inconvenient history behind diff --git a/.gitattributes b/.gitattributes new file mode 100644 index 0000000..745d764 --- /dev/null +++ b/.gitattributes @@ -0,0 +1,96 @@ +############################################################################### +# Handle line endings automatically for files detected as text +# and leave all files detected as binary untouched. +* text=auto +# Force the following filetypes to have unix eols and encoding, so that Windows does not break them. +# If a file is going to be used on linux and windows, we want it invariant, +# rather than automatically translated, because automatic translation always screw things up. +.gitignore text eol=lf encoding=utf-8 whitespace=trailing-space,space-before-tab,tabwidth=4 +.gitattributes text eol=lf encoding=utf-8 whitespace=trailing-space,space-before-tab,tabwidth=4 +.gitmodules text eol=lf encoding=utf-8 whitespace=trailing-space,space-before-tab,tabwidth=4 +*.sh text eol=lf encoding=utf-8 whitespace=trailing-space,space-before-tab,tabwidth=4 +*.c text eol=lf encoding=utf-8 whitespace=trailing-space,space-before-tab,tabwidth=4 +*.cpp text eol=lf encoding=utf-8 whitespace=trailing-space,space-before-tab,tabwidth=4 +*.h text eol=lf encoding=utf-8 whitespace=trailing-space,space-before-tab,tabwidth=4 +*.txt text eol=lf encoding=utf-8 whitespace=trailing-space,space-before-tab,tabwidth=4 +*.html text eol=lf encoding=utf-8 whitespace=trailing-space,space-before-tab,tabwidth=4 +*.htm text eol=lf encoding=utf-8 whitespace=trailing-space,space-before-tab,tabwidth=4 +*.md text eol=lf encoding=utf-8 whitespace=trailing-space,space-before-tab,tabwidth=4 +*.pandoc text eol=lf encoding=utf-8 whitespace=trailing-space,space-before-tab,tabwidth=4 +*.css text eol=lf encoding=utf-8 whitespace=trailing-space,space-before-tab,tabwidth=4 +*.manifest text eol=lf encoding=utf-8 whitespace=trailing-space,space-before-tab,tabwidth=4 + + +makefile text eol=lf encoding=utf-8 +Makefile text eol=lf encoding=utf-8 + +# Force the following Visual Studio specific filetypes to have Windows eols, +# so that Git does not break them +*.bat text eol=crlf encoding=utf-8 whitespace=trailing-space,space-before-tab,tabwidth=4 +*.cmd text eol=crlf encoding=utf-8 whitespace=trailing-space,space-before-tab,tabwidth=4 +*.rc text eol=crlf encoding=utf-8 whitespace=trailing-space,space-before-tab,tabwidth=4 +*.sln text eol=crlf encoding=utf-8 whitespace=trailing-space,space-before-tab,tabwidth=4 +*.vcproj text eol=crlf encoding=utf-8 whitespace=trailing-space,space-before-tab,tabwidth=4 +*.vcxproj text eol=crlf encoding=utf-8 whitespace=trailing-space,space-before-tab,tabwidth=4 +*.vcxproj.filters text eol=crlf encoding=utf-8 whitespace=trailing-space,space-before-tab,tabwidth=4 +*.vcxproj.user text eol=crlf encoding=utf-8 whitespace=trailing-space,space-before-tab,tabwidth=4 + +#Don't let git screw with pdf files +*.pdf -text + +# Force binary files to be binary +*.gif -textn -diff +*.jpg -textn -diff +*.jepg -textn -diff +*.png -textn -diff +*.webp -textn -diff + +############################################################################### +# Set default behavior for command prompt diff. +# +# This is need for earlier builds of msysgit that does not have it on by +# default for csharp files. +# Note: This is only used by command line +############################################################################### +#*.cs diff=csharp + +############################################################################### +# Set the merge driver for project and solution files +# +# Merging from the command prompt will add diff markers to the files if there +# are conflicts (Merging from VS is not affected by the settings below, in VS +# the diff markers are never inserted). Diff markers may cause the following +# file extensions to fail to load in VS. An alternative would be to treat +# these files as binary and thus will always conflict and require user +# intervention with every merge. To do so, just uncomment the entries below +############################################################################### +#*.sln merge=binary +#*.csproj merge=binary +#*.vbproj merge=binary +#*.vcxproj merge=binary +#*.vcproj merge=binary +#*.dbproj merge=binary +#*.fsproj merge=binary +#*.lsproj merge=binary +#*.wixproj merge=binary +#*.modelproj merge=binary +#*.sqlproj merge=binary +#*.wwaproj merge=binary + +############################################################################### +# diff behavior for common document formats +# +# Convert binary document formats to text before diffing them. This feature +# is only available from the command line. Turn it on by uncommenting the +# entries below. +############################################################################### +#*.doc diff=astextplain +#*.DOC diff=astextplain +#*.docx diff=astextplain +#*.DOCX diff=astextplain +#*.dot diff=astextplain +#*.DOT diff=astextplain +#*.pdf diff=astextplain +#*.PDF diff=astextplain +#*.rtf diff=astextplain +#*.RTF diff=astextplain diff --git a/.gitconfig b/.gitconfig new file mode 100644 index 0000000..e50d385 --- /dev/null +++ b/.gitconfig @@ -0,0 +1,29 @@ +[core] + autocrlf = input + whitespace = trailing-space,space-before-tab,tabwidth=4 + safecrlf +[alias] + lg = log --reverse --max-count=4 --oneline --pretty='format:%C(yellow)%h %d %Creset%p %C("#60A0FF")%cr %Cgreen %cn %GT trust%Creset%n%s%n' + graph = log --max-count=20 --graph --pretty=format:'%C(yellow)%h%Creset %s %Cgreen(%cr) %C(bold blue)%cn %GT%Creset' --abbrev-commit + alias = ! git config --get-regexp ^alias\\. | sed -e s/^alias\\.// -e s/\\ /\\ =\\ / | grep -v ^'alias ' | sort + fixws = !"\ + if (! git diff-files --quiet .) && \ + (! git diff-index --quiet --cached HEAD) ; then \ + git commit -m FIXWS_SAVE_INDEX && \ + git add -u :/ && \ + git commit -m FIXWS_SAVE_TREE && \ + git rebase --whitespace=fix HEAD~2 && \ + git reset HEAD~ && \ + git reset --soft HEAD~ ; \ + elif (! git diff-files --quiet .) ; then \ + git add -u :/ && \ + git commit -m FIXWS_SAVE_TREE && \ + git rebase --whitespace=fix HEAD~ && \ + git reset HEAD~ ; \ + an elif (! git diff-index --quiet --cached HEAD) ; then \ + git commit -m FIXWS_SAVE_INDEX && \ + git rebase --whitespace=fix HEAD~ && \ + git reset --soft HEAD~ ; \ + fi" +[commit] + gpgSign = true diff --git a/.gitignore b/.gitignore new file mode 100644 index 0000000..9edb00c --- /dev/null +++ b/.gitignore @@ -0,0 +1,271 @@ +*.bat +## Ignore Visual Studio temporary files, build results, and +## files generated by popular Visual Studio add-ons. +*.bak + +# User-specific files +*.suo +*.user +*.userosscache +*.sln.docstates +*.exe +*.idb +*.vcxproj.filters +*.html +*.htm +wallet.cppcheck + +# User-specific files (MonoDevelop/Xamarin Studio) +*.userprefs + +# Build results +[Dd]ebug/ +[Dd]ebugPublic/ +[Rr]elease/ +[Rr]eleases/ +x64/ +x86/ +bld/ +[Bb]in/ +[Oo]bj/ +[Ll]og/ + +.vscode + +# Visual Studio 2015 cache/options directory +.vs/ +# Uncomment if you have tasks that create the project's static files in wwwroot +#wwwroot/ + +# MSTest test Results +[Tt]est[Rr]esult*/ +[Bb]uild[Ll]og.* + +# NUNIT +*.VisualState.xml +TestResult.xml + +# Build Results of an ATL Project +[Dd]ebugPS/ +[Rr]eleasePS/ +dlldata.c + +# DNX +project.lock.json +project.fragment.lock.json +artifacts/ + +*_i.c +*_p.c +*_i.h +*.ilk +*.meta +*.obj +*.pch +*.pdb +*.pgc +*.pgd +*.rsp +*.sbr +*.tlb +*.tli +*.tlh +*.tmp +*.tmp_proj +*.log +*.vspscc +*.vssscc +.builds +*.pidb +*.svclog +*.scc + +# Chutzpah Test files +_Chutzpah* + +# Visual C++ cache files +ipch/ +*.aps +*.ncb +*.opendb +*.opensdf +*.sdf +*.cachefile +*.VC.db +*.VC.VC.opendb + +# Visual Studio profiler +*.psess +*.vsp +*.vspx +*.sap + +# TFS 2012 Local Workspace +$tf/ + +# Guidance Automation Toolkit +*.gpState + +# ReSharper is a .NET coding add-in +_ReSharper*/ +*.[Rr]e[Ss]harper +*.DotSettings.user + +# JustCode is a .NET coding add-in +.JustCode + +# TeamCity is a build add-in +_TeamCity* + +# DotCover is a Code Coverage Tool +*.dotCover + +# NCrunch +_NCrunch_* +.*crunch*.local.xml +nCrunchTemp_* + +# MightyMoose +*.mm.* +AutoTest.Net/ + +# Web workbench (sass) +.sass-cache/ + +# Installshield output folder +[Ee]xpress/ + +# DocProject is a documentation generator add-in +DocProject/buildhelp/ +DocProject/Help/*.HxT +DocProject/Help/*.HxC +DocProject/Help/*.hhc +DocProject/Help/*.hhk +DocProject/Help/*.hhp +DocProject/Help/Html2 +DocProject/Help/html + +# Click-Once directory +publish/ + +# Publish Web Output +*.[Pp]ublish.xml +*.azurePubxml +# TODO: Comment the next line if you want to checkin your web deploy settings +# but database connection strings (with potential passwords) will be unencrypted +#*.pubxml +*.publishproj + +# Microsoft Azure Web App publish settings. Comment the next line if you want to +# checkin your Azure Web App publish settings, but sensitive information contained +# in these scripts will be unencrypted +PublishScripts/ + +# NuGet Packages +*.nupkg +# The packages folder can be ignored because of Package Restore +**/packages/* +# except build/, which is used as an MSBuild target. +!**/packages/build/ +# Uncomment if necessary however generally it will be regenerated when needed +#!**/packages/repositories.config +# NuGet v3's project.json files produces more ignoreable files +*.nuget.props +*.nuget.targets + +# Microsoft Azure Build Output +csx/ +*.build.csdef + +# Microsoft Azure Emulator +ecf/ +rcf/ + +# Windows Store app package directories and files +AppPackages/ +BundleArtifacts/ +Package.StoreAssociation.xml +_pkginfo.txt + +# Visual Studio cache files +# files ending in .cache can be ignored +*.[Cc]ache +# but keep track of directories ending in .cache +!*.[Cc]ache/ + +# Others +ClientBin/ +~$* +*~ +*.dbmdl +*.dbproj.schemaview +*.jfm +*.pfx +*.publishsettings +node_modules/ +orleans.codegen.cs + +# Since there are multiple workflows, uncomment next line to ignore bower_components +# (https://github.com/github/gitignore/pull/1529#issuecomment-104372622) +#bower_components/ + +# RIA/Silverlight projects +Generated_Code/ + +# Backup & report files from converting an old project file +# to a newer Visual Studio version. Backup files are not needed, +# because we have git ;-) +_UpgradeReport_Files/ +Backup*/ +UpgradeLog*.XML +UpgradeLog*.htm + +# SQL Server files +*.mdf +*.ldf + +# Business Intelligence projects +*.rdl.data +*.bim.layout +*.bim_*.settings + +# Microsoft Fakes +FakesAssemblies/ + +# GhostDoc plugin setting file +*.GhostDoc.xml + +# Node.js Tools for Visual Studio +.ntvs_analysis.dat + +# Visual Studio 6 build log +*.plg + +# Visual Studio 6 workspace options file +*.opt + +# Visual Studio LightSwitch build output +**/*.HTMLClient/GeneratedArtifacts +**/*.DesktopClient/GeneratedArtifacts +**/*.DesktopClient/ModelManifest.xml +**/*.Server/GeneratedArtifacts +**/*.Server/ModelManifest.xml +_Pvt_Extensions + +# Paket dependency manager +.paket/paket.exe +paket-files/ + +# FAKE - F# Make +.fake/ + +# JetBrains Rider +.idea/ +*.sln.iml + +# CodeRush +.cr/ + +# Python Tools for Visual Studio (PTVS) +__pycache__/ +*.pyc \ No newline at end of file diff --git a/.gitmodules b/.gitmodules new file mode 100644 index 0000000..71093de --- /dev/null +++ b/.gitmodules @@ -0,0 +1,11 @@ +[submodule "libsodium"] + path = libsodium + url = https://github.com/jedisct1/libsodium.git + ignore = dirty +[submodule "wxWidgets"] + path = wxWidgets + url = https://github.com/wxWidgets/wxWidgets.git + ignore = dirty +[submodule "mpir"] + path = mpir + url = git://github.com/BrianGladman/mpir.git diff --git a/ILog.cpp b/ILog.cpp new file mode 100644 index 0000000..572015a --- /dev/null +++ b/ILog.cpp @@ -0,0 +1,41 @@ +#include "stdafx.h" + +void ILogFatalError(const char* sz) { + wxLogFatalError(_wx("%s"), _wx(sz)); +} // which is like wxLogError(), but also terminates the program with the exit code 3 (using abort() standard function).Unlike for all the other logging functions, this function can't be overridden by a log target. +void ILogError(const char* sz) { + wxLogError(_wx("%s"), _wx(sz)); +} + //is the function to use for error messages, i.e.the messages that must be shown to the user.The default processing is to pop up a message box to inform the user about it. +void ILogWarning(const char* sz) { + wxLogWarning(_wx("%s"), _wx(sz)); +} //for warnings.They are also normally shown to the user, but don't interrupt the program work. +void ILogMessage(const char* sz) { + wxLogMessage(_wx("%s"), _wx(sz)); +} // is for all normal, informational messages.*/ +void ILogVerbose(const char* sz) { + wxLogVerbose(_wx("%s"), _wx(sz)); +} + ; // is for verbose output.Normally, it is suppressed, but might be activated if the user wishes to know more details about the program progress(another, but possibly confusing name for the same function is wxLogInfo). +void ILogDebug(const char* sz) { + wxLogDebug(_wx("%s"), _wx(sz)); +} //is the right function for debug output. It only does anything at all in the + //debug mode(when the preprocessor symbol WXDEBUG is defined) and expands to + //nothing in release mode(otherwise).Note that under Windows, you must either + //run the program under debugger or use a 3rd party program such as DebugView + +void queue_error_message(const char* psz) { + // Used where throwing immediately would be disastrous, as in a destructor. + auto event = new wxCommandEvent(wxEVT_MENU, myID_ERRORMESSAGE); + event->SetString(_wx(psz)); + // wxQueueEvent(singletonFrame->GetMenuBar(), event); + wxQueueEvent(singletonApp, event); +} + +void queue_fatal_error(const char* psz) { + // Used where throwing immediately would be disastrous, as in a destructor or when constructing the main frame + if (!errorCode)errorCode = 10; + queue_error_message(psz); + singletonFrame->Close(); +} + diff --git a/ILog.h b/ILog.h new file mode 100644 index 0000000..2a55c96 --- /dev/null +++ b/ILog.h @@ -0,0 +1,88 @@ +#pragma once +extern int errorCode; +extern std::string szError; +void ILogFatalError(const char*); +void ILogError(const char*); +void ILogWarning(const char*); +void ILogMessage(const char* format); +void ILogVerbose(const char*); +void ILogDebug(const char*); +void queue_error_message(const char*); //Used for error conditions within a destructor because you cannot throw within a destructor +void queue_fatal_error(const char*); //Used for fatal error conditions within a destructor in place of FatalException because you cannot throw within a destructor + +class MyException: public std::exception { +private: + std::string err; +public: + virtual ~MyException() override = default; + MyException() = delete; + explicit MyException(const std::string &m) noexcept :err(m){} + explicit MyException(const char* sz) noexcept :err(sz) {} + virtual const char* what() const override { + return err.c_str(); + } +}; + +class FatalException : public MyException { +public: + using MyException::MyException; + FatalException() noexcept; +}; + +class HashReuseException : public MyException { +public: + using MyException::MyException; + HashReuseException() noexcept; +}; + +class SQLexception : public MyException { +public: + using MyException::MyException; + SQLexception() noexcept; +}; + +class NonUtf8DataInDatabase : public MyException { +public: + using MyException::MyException; + NonUtf8DataInDatabase() noexcept; +}; + +class BadDataException : public MyException { +public: + using MyException::MyException; + BadDataException() noexcept; +}; + +class NonRandomScalarException : public MyException { +public: + using MyException::MyException; + NonRandomScalarException() noexcept; +}; + +class BadScalarException : public MyException { +public: + using MyException::MyException; + BadScalarException() noexcept; +}; + +class OversizeBase58String : public MyException { +public: + using MyException::MyException; + OversizeBase58String() noexcept; +}; + +// This exception is obviously far too generic, because the routine throwing it knows nothing of the context. +// does not know what the cryptographic id identifies. +// higher level code that does know the context needs to catch the exception and issue a more +// relevant errror message, possibly with by more informative rethrow. +class BadStringRepresentationOfCryptoIdException : public MyException { +public: + using MyException::MyException; + BadStringRepresentationOfCryptoIdException() noexcept; +}; + +class NotBase58Exception : public MyException { +public: + using MyException::MyException; + NotBase58Exception() noexcept; +}; diff --git a/ISqlit3Impl.cpp b/ISqlit3Impl.cpp new file mode 100644 index 0000000..cebd543 --- /dev/null +++ b/ISqlit3Impl.cpp @@ -0,0 +1,239 @@ +// this is implementation class of pure virtual interface base class between sqlite3, +// which speaks only C and utf8 char[] +// and wxWidgets which speaks only C++ and unicode strings. +// +// In this code I continually declare stuff public that should be private, +// but that is OK, because declared in a cpp file, not a header file, +// and thus they remain private to any code outside this particular cpp file. +// When the compiler complains that something is inaccessible, I don't muck +// around with friend functions and suchlike, which rapidly gets surprisingly +// complicated, I just make it public, but only public to this one file. +#include +#include // for basic_string, allocator, char_traits +#include // for initializer_list +#include // for shared_ptr, unique_ptr +#include +#include "ISqlite3.h" +#include "sqlite3.h" + +static auto error_message(int rc, sqlite3* pdb) { + return std::string("Sqlite3 Error: ") + sqlite3_errmsg(pdb) + ". Sqlite3 error number=" + std::to_string(rc); +} + +void sqlite3_init() { + if (sqlite3_initialize() != SQLITE_OK) { + errorCode = 7; + szError = "Fatal Error: Sqlite library did not init."; +// Cannot log the error, because logging not set up yet, so logging itself causes an exception + throw FatalException(szError.c_str()); + } +} + +class IcompiledImpl_sql; + +static int callback(void* NotUsed, int argc, char** argv, char** azColName) { + std::string str; + str.reserve(256); + for (int i = 0; i < argc; i++) { + str =str + "\t\"" + azColName[i]+ R"|("=)|" + (argv[i]!=nullptr ? argv[i] : "NULL"); + } + ILogMessage(str.c_str()); + return 0; +} + +class ISqlite3Impl : + public ISqlite3 +{ +public: + sqlite3* pdb; + ISqlite3Impl() = delete; + ISqlite3Impl(const char* dbName, int flags) { +#ifndef NDEBUG + pdb = nullptr; +#endif + int rc =sqlite3_open_v2(dbName, &pdb, flags, nullptr); + if (rc != SQLITE_OK) throw SQLexception(error_message(rc, pdb)); + assert(pdb != nullptr); + // pdb can never be nullptr, since the sqlite3_open_v2 command always initializes + // it even if open fails + } + + void exec(const char* szsql) override { + char* zErrMsg = nullptr; + int rc = sqlite3_exec(pdb, szsql, callback, nullptr, &zErrMsg); + if (rc != SQLITE_OK) { + SQLexception e(std::string("SQL Exec Error: ") + zErrMsg); + sqlite3_free(zErrMsg); + throw e; + } + } + + ~ISqlite3Impl() override { + exec("PRAGMA optimize;"); //If we have multiple threads, will want to call this only once, and we will also want to call the check pragma in a separate thread with a separate connection. + int rc{ sqlite3_close(pdb) }; + if (rc == SQLITE_OK) {} + else { + std::string err(error_message(rc, pdb) + ". Bad destruction of ISqlite3Impl"); + ILogError(err.c_str()); + queue_error_message(err.c_str()); //Does not actually pop up a message, which would be extremely bad in a destructor, instead queues an event which causes the message to pop up. + } + // If called before everything is finalized, will return SQL_BUSY + // but that is a coding error. + // sqlite3_close_v2 sets everything to shutdown when everything is finalized, + // but this is C++. We do our own memory management, and if we need + // sqlite3_close_v2 we are doing it wrong. + } +}; + +// Factory method to open database. +ISqlite3* Sqlite3_open(const char * db_name) { + return new ISqlite3Impl(db_name, SQLITE_OPEN_READWRITE); +} + +// Factory method to create database. +ISqlite3 * Sqlite3_create(const char* db_name) { + return new ISqlite3Impl(db_name, SQLITE_OPEN_READWRITE | SQLITE_OPEN_CREATE); +} + +class IcompiledImpl_sql : + public Icompiled_sql +{ + friend class ISqlite3Impl; +private: + sqlite3_stmt *pStmt; + ISqlite3Impl *pdbImplOwn; + auto e(int rc) { + assert(rc != SQLITE_OK); + return SQLexception(error_message(rc, pdbImplOwn->pdb)); + } +public: + IcompiledImpl_sql() = delete; + IcompiledImpl_sql( + ISqlite3 * pdbImpl, /* wrapped database handle */ + const char *zSql /* SQL statement, UTF-8 encoded */ + ) { + assert(pdbImpl!=nullptr); + assert(zSql!=nullptr); + pdbImplOwn = static_cast (pdbImpl); //static downcast + // Static downcast is safe because the base is pure virtual, + // and we have only one derived type, + // If we ever have multiple derived types, I will use an enum + // in the pure virtual base class and continue to use static downcasts. + // Or, better, derive additional implementation classes from ISqlite3Impl so that + // downcast to ISqlite3Impl must always work. + assert(pdbImplOwn); + // If an instance of the derived class exists, has to a valid upcast. + const char *pzTail; + int rc = sqlite3_prepare_v3( + pdbImplOwn->pdb, /* Database handle */ + zSql, /* SQL statement, UTF-8 encoded */ + -1, /* Maximum length of zSql in bytes. */ + SQLITE_PREPARE_PERSISTENT, + &pStmt, /* OUT: Statement handle */ + &pzTail /* OUT: Pointer to unused portion of zSql */ + ); + if (rc != SQLITE_OK) throw e(rc); + } + + ~IcompiledImpl_sql() override{ + int rc=sqlite3_finalize(pStmt); + if (rc == SQLITE_OK) {} + else { + std::string err(error_message(rc, pdbImplOwn->pdb) + ". Bad destruction of Icompiled_sql"); + ILogError(err.c_str()); + // This error should only ever happen if object failed to compile, in which case we have already handled the error + // Hence we do not queue an event to pop up a message, only log the error. (Unless ILogError pops up a message, which it might, but normally does not) + } + } + + virtual void Isqlite3_bind(int param, std::span blob) override { + int rc = sqlite3_bind_blob(pStmt, param, &blob[0], static_cast(blob.size_bytes()), SQLITE_STATIC); + if (rc != SQLITE_OK) throw e(rc); + } + + virtual void Isqlite3_bind(int param, int i) override { + int rc = sqlite3_bind_int(pStmt, param, i); + if (rc != SQLITE_OK) throw e(rc); + } + + virtual void Isqlite3_bind(int param, int64_t i) override { + int rc = sqlite3_bind_int64(pStmt, param, i); + if (rc != SQLITE_OK) throw e(rc); + } + + virtual void Isqlite3_bind(int param) override { + int rc = sqlite3_bind_null(pStmt, param); + if (rc != SQLITE_OK) throw e(rc); + } + + virtual void Isqlite3_bind(int param, const char* str) override { + int rc = sqlite3_bind_text(pStmt, param, str, -1, SQLITE_STATIC); + if (rc != SQLITE_OK) throw e(rc); + } + + virtual sql_result Isqlite3_step() override { + int rc = sqlite3_step(pStmt); + sql_result ret; + switch (rc & 0xFF) { + case SQLITE_DONE: + ret = DONE; + break; + case SQLITE_ROW: + ret = ROW; + break; + case SQLITE_BUSY: + //ret = BUSY; + // As handling busy is hard, we will always use WAL mode and only allow one thread in one process write to the database. + // If we need many threads, perhaps in many processes, to write, they will all channel through a single thread + // in a single process whose transactions are never substantially larger than thirty two kilobytes. + // As a result, we should never get SQL_BUSY codes except in pathological cases that are OK to handle by + // terminating or reporting to the user that his operation has failed because database abnormally busy. + // When we are building the blockchain, every process but one will see the blockchain as instantaneously changing + // from n blocks to n+1 blocks when a single transaction updates the root and anciliary data. + // We will build the blockchain hash table in postfix format, with patricia tree nodes that + // have skiplink format stored only in memory and rebuilt each startup so that it grows append only, + throw SQLexception("Abnormal busy database"); + break; + case SQLITE_MISUSE: + //ret = MISUSE; + throw e(rc); + break; + default: + //ret = SQL_ERROR; + throw e(rc); + } + return ret; + } + virtual std::span Isqlite3_column_blob(int iCol) const override { + return std::span((const uint8_t*)sqlite3_column_blob(pStmt, iCol), sqlite3_column_bytes(pStmt, iCol)); + // returns the null pointer if null + } + virtual int Isqlite3_column_int(int iCol) const override { + return sqlite3_column_int(pStmt, iCol); + } + virtual int64_t Isqlite3_column_int64(int iCol) const override { + return sqlite3_column_int64(pStmt, iCol); + } + virtual char *Isqlite3_column_text(int iCol) const override { + return static_cast(sqlite3_column_text(pStmt, iCol)); + /* returns pointer to zero length string if null. If we need to distinguish between zero length strings and nulls, need the type function. + We can store any type in any column, and read any type from any column, but if something unexpected is in a column, it gets converted to the expected type on being read back. For example an integer gets converted a decimal string if read as a blob or as text. + It is very rarely valid to store different types in the same column, except that null is permissible. The difference between null and zero matters, but the case of null is usually dealt with by sql code, not C code. */ + } + virtual void Isqlite3_reset()override { + int rc = sqlite3_reset(pStmt); + if (rc != SQLITE_OK) throw e(rc); + // sqlite3_reset returns extended error codes + // https://sqlite.org/c3ref/reset.html + } + +}; + +/* Factory method to prepare a compiled sql statement +Uses automatic upcast. You always want to start with the most derived smart pointer if you can, and let the compiler take care of default upcasting.*/ +Icompiled_sql* sqlite3_prepare(ISqlite3 *pdbImpl, const char * zSql) { + return new IcompiledImpl_sql( + pdbImpl, /* Database handle */ + zSql /* SQL statement, UTF-8 encoded */ + ); +} \ No newline at end of file diff --git a/ISqlite3.h b/ISqlite3.h new file mode 100644 index 0000000..74f21c4 --- /dev/null +++ b/ISqlite3.h @@ -0,0 +1,96 @@ +#pragma once +#include "ILog.h" +// this is pure virtual interface base class between sqlite3, which speaks only C and utf8 char[] +// and wxWidgets which speaks only C++ and unicode strings. + +// Usage: Call the factory function std::shared_ptr sqlite3_open(const char *) to get a shared +// pointer to the // Sqlite3 database object. Then call the factory function +// sqlite3_prepare(std::shared_ptr, const char *) to get a unique pointer to +// a compiled SQL statement + +// Its primary purpose is to avoid code that needs both the wxWidgets header files, +// and the sqlite3.h header file. +// +// It speaks only utf8 char[], and needs to be called in wxWidgets code using +// wxString.utf8_str() and its return values need to be interpreted in wxWidgets code +// using wxString::FromUTF8(). +// +// This header file can be included in code that has the sqlite3.h header file +// and in code that has the wxWidgets header file, for it has no dependencies on either one +// +// In code that has wxWidgets headers, we call members of this interface class, +// rather than directly calling sqlite3 functions. +// +// I originally implemented the pimpl idiom, but it turns out that pimpl has become +// substantially more difficult in C++14, because one is effectively rolling one's own +// unique pointer. +// +// It is therefore easier to implement a pure virtual base class with a virtual destructor and +// factory function that returns a smart pointer to a member of the derived implementation +// +/* This code is at a low level abstraction, because it provides low level C++ interface to inherently low level C + It is intended to be wrapped in higher level code that does not know about the nuts and bolts of sqlite3, but which supports throwing, templated functions, and all that.*/ + +// +//___________________________________ + +// This class wraps a compiled sql statement. +class Icompiled_sql +{ +protected: + Icompiled_sql() = default; // needed for derived constructor +public: + virtual ~Icompiled_sql() = default; // needed for derived destructor + // Bind is used when writing stuff into the database. These objects should continue to exist until the write is finalized or reset. + virtual void Isqlite3_bind( int, const std::span) = 0; // https://sqlite.org/c3ref/bind.html + virtual void Isqlite3_bind(int, int) = 0; + virtual void Isqlite3_bind(int, int64_t) = 0; + virtual void Isqlite3_bind(int) = 0; + virtual void Isqlite3_bind(int, const char*) = 0; + enum sql_result { DONE, ROW, BUSY, SQL_ERROR, MISUSE }; + virtual sql_result Isqlite3_step() = 0; + // when reading, you don't use bind. Sqlite creates a temporary in the memory that it manages. If you want the object to live beyond the next step operation, need to make a copy + // When writing objects, we reinterpret a pointer to a typed object as a blob pointer, when reading them, we need a typed copy, otherwise calling the destructor could be bad. + // We don't want Sqlite3 calling destructors on our objects, hence write them as static, and create them from raw bytes on reading. + virtual std::span Isqlite3_column_blob (int) const = 0; // returns the null pointer and zero length if null. + virtual int Isqlite3_column_int (int) const = 0; + virtual int64_t Isqlite3_column_int64 (int) const = 0; + virtual char* Isqlite3_column_text (int) const = 0; // returns pointer to zero length + // string if null. If we need to distinguish betweem zero length strings and nulls, need the + // type function. + // We can store any type in any column, and read any type from any column, but if something + // unexpected is in a column, it gets coerced to the expected type on being read back. + // Thus something stored as a number and read back as blob will come back as the decimal character string. + // It is very rarely valid to store different types in the same column, except that + // null is permissible. The difference between null and zero matters, but the case of + // null is usually dealt with by sql code, not C code. + virtual void Isqlite3_reset() = 0; // https://sqlite.org/c3ref/reset.html +}; + +//___________________________________ + +// This class wraps a database. Its derived implementation will hold an old type C pointer +// to an opened database object, which is destroyed when the class object is destroyed +class ISqlite3 +{ +protected: + ISqlite3() = default; // needed for derived constructor +public: + virtual ~ISqlite3() = default; // needed for derived destructor + virtual void exec(const char*) = 0; +}; + +// Factory method to open a database and produce a shared object wrapping the database +ISqlite3* Sqlite3_open(const char*); + +// Factory method to create a database and produce a shared object wrapping the database +ISqlite3* Sqlite3_create(const char*); + +// Factory method to prepare a compiled sql statement +Icompiled_sql* sqlite3_prepare(ISqlite3*, const char *); + +void sqlite3_init(); +extern "C" { + int sqlite3_shutdown(void); +} + diff --git a/Icon.ico b/Icon.ico new file mode 100644 index 0000000..5d06b9f Binary files /dev/null and b/Icon.ico differ diff --git a/LICENSE.md b/LICENSE.md new file mode 100644 index 0000000..697eb22 --- /dev/null +++ b/LICENSE.md @@ -0,0 +1,24 @@ +--- +generator: +title: LICENSE +--- +Copyright © 2021 reaction.la gpg key 154588427F2709CD9D7146B01C99BB982002C39F + +This distribution of free software contains numerous other +distributions with other compatible free software licenses and copyrights. +Those files and directories are governed by their own license, and their +combination and integration into this project by this license and this +copyright, and anything in this distribution not otherwise licensed and +copyrighted in this distribution is governed by this license, and this +copyright. + + Licensed under the Apache License, Version 2.0 (the "License"); +you may not use this distribution of software except in compliance with the License. +You may obtain a copy of the License at + + +Unless required by applicable law or agreed to in writing, software +distributed under the License is distributed on an "AS IS" BASIS, +WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +See the License for the specific language governing permissions and +limitations under the License. diff --git a/NOTICE.md b/NOTICE.md new file mode 100644 index 0000000..4eb8f95 --- /dev/null +++ b/NOTICE.md @@ -0,0 +1,22 @@ +--- +title: NOTICE +--- +Copyright © 2021 reaction.la gpg key 154588427F2709CD9D7146B01C99BB982002C39F + +The license of this software, and the licenses of the packages on which it + relies, grant the four software freedoms: + +0. The freedom to run the program as you wish, for any purpose. +1. The freedom to study how the program works, and change it so it + does your computing as you wish. +2. The freedom to redistribute copies so you can help others. +3. The freedom to distribute copies of your modified versions to + others. + +This software is licensed under the [apache 2.0 license](LICENSE.html). + +This product includes several packages, each with their own free software licence, referenced in the relevant files or subdirectories. + +Or, in the case of Sqlite, the Sqlite blessing in place of a license, which is +morally though not legally obligatory on those that obey the +commandments of Gnon. See also the [contributor code of conduct](docs/contributor_code_of_conduct.html). diff --git a/README.md b/README.md new file mode 100644 index 0000000..9540f8b --- /dev/null +++ b/README.md @@ -0,0 +1,74 @@ +--- +title: >- + README +--- +[pre alpha documentation (mostly a wish list)](docs/index.htm) + +[copyright © and license](./license.txt) + +pre-requisite, Pandoc to build the html documentation from the markdown files. + +Windows pre-requisites: Visual Studio and git-bash + +To obtain the source code from which the project can be built, including +this README, from the bash command line (git-bash in windows). + +```bash2 +git clone missing url +cd wallet +./winConfigure.sh +``` + +To configure and build the required third party libraries in windows, then +build the program and run unit test for the first time, launch the Visual +Studio X64 native tools command prompt in the cloned directory, then: + +```bat +winConfigure.bat +``` + +[cryptographic software is under attack]:./docs/contributor_code_of_conduct.html#code-will-be-cryptographically-signed +"Contributor Code of Conduct" +{target="_blank"} + +winConfigure.bat also configures the repository you just created to use +`.gitconfig` in the repository, causing git to to implement GPG signed +commits -- because [cryptographic software is under attack] from NSA +entryists, and shills, who seek to introduce backdoors. + +This may be inconvenient if you do not have `gpg` installed and set up. + +`.gitconfig` adds several git aliases: + +1. `git lg` to display the gpg trust information for the last four commits. + For this to be useful you need to import the repository public key + `public_key.gpg` into gpg, and locally sign that key. +1. `git fixws` to standardise white space to the project standards +1. `git graph` to graph the commit tree +1. `git alias` to display the git aliases. + +```bash +# To verify that the signature on future pulls is unchanged. +gpg --import public_key.gpg +gpg --lsign 096EAE16FB8D62E75D243199BC4482E49673711C +# We ignore the Gpg Web of Trust model and instead use +# the Zooko identity model. +# We use Gpg signatures to verify that remote repository +# code is coming from an unchanging entity, not for +# Gpg Web of Trust. Web of Trust is too complicated +# and too user hostile to be workable or safe. +# Never --sign any Gpg key related to this project. --lsign it. +# Never check any Gpg key related to this project against a +# public gpg key repository. It should not be there. +# Never use any email address on a gpg key related to this project +# unless it is only used for project purposes, or a fake email, +# or the email of someone whom you do not like. +``` + +To build the documentation in its intended html form from the markdown +files, execute the bash script file `docs/mkdocs.sh`, in an environment where +`pandoc` is available. On Windows, if Git Bash and Pandoc has bee +installed, you should be able to run a shell file in bash by double clicking on it. + +[Pre alpha release](./RELEASE_NOTES.html), which means it does not yet work even well enough for +it to be apparent what it would do if it did work. diff --git a/RELEASE_NOTES.md b/RELEASE_NOTES.md new file mode 100644 index 0000000..3ae279d --- /dev/null +++ b/RELEASE_NOTES.md @@ -0,0 +1,8 @@ +--- +title: Release Notes +--- +To build and run [README](./README.html) + +[pre alpha documentation (mostly a wish list)](docs/index.htm) + +This software is pre alpha and should not yet be released. It does not work well enough to even show what it would do if it was working diff --git a/app.cpp b/app.cpp new file mode 100644 index 0000000..b6bae26 --- /dev/null +++ b/app.cpp @@ -0,0 +1,206 @@ +#include "stdafx.h" + +thread_local thread_local__* thl{ nullptr }; +wxIMPLEMENT_APP(App); + +App::App() +{ + assert (singletonApp == nullptr); + singletonApp = this; + if (thl == nullptr)thl = new thread_local__(); +} + +App::~App() +{ + assert(singletonApp == this); + singletonApp = nullptr; + if (thl != nullptr)delete thl; + thl = nullptr; +} + +bool App::OnInit() +{ if (wxApp::OnInit()) { + SetVendorName(_T("rho")); /* This causes the non volatile config data to be stored under the rho on + windows.*/ + SetAppName(_T("wallet")); /* This causes the non volatile config data to be stored under rho\wallet + We will generally place data in the database, and if additional executables need their own data + in the config, they will create their own subkey under Computer\HKEY_CURRENT_USER\Software\rho */ + pConfig = std::unique_ptr(wxConfigBase::Get()); + pConfig->SetRecordDefaults(false); + /* pConfig corresponds to the Windows Registry entry + Computer\HKEY_CURRENT_USER\Software\rho\wallet + + Contrary to wxWidgets documentation, the config data on windows is by default stored in + HKCU, HKEY_CURRENT_USER, not in HKLM, HKEY_LOCAL_MACHINE. + + We probably should have placed per user data in an sqlite3 file in + wxStandardPaths::GetUserDataDir() + + Data global to all users has to go in an sqlite3 file in wxStandardPaths::GetAppDocumentsDir() + or wxStandardPaths::GetLocalDataDir() + + User local database will record the derivation of all secrets, and what wallets along the path + are logged in. The local machine database will hold the global consensus blockchain, which contains + no privacy sensitive information, and will also hold data global to all users on a particular + machine. + + A wallet's secret can be stored in a file - we will eventually provide passwords for files, + but passwords provide a false sense of security, because if someone gets a copy of that file, + a sophisticated attacker can perform an offline brute force attack, thus a human memorable + password only provides protection against casual and opportunistic attackers. + If the file is insecure, password needs to impossible to remember, and stored somewhere secure..*/ + + Frame* frame = new Frame(pConfig->GetAppName()); + frame->Show(true); //Frame, being top level unowned window, is owned by the one and only message pump + if (m_display_in_front && singletonFrame != nullptr && singletonFrame->m_pLogWindow != nullptr) singletonFrame->m_pLogWindow->GetFrame()->Raise(); + return true; +} +else return false; +} + +int App::OnRun() +{ + Bind(wxEVT_MENU, &App::OnError, this, myID_ERRORMESSAGE); + int exitcode = wxApp::OnRun(); + //wxTheClipboard->Flush(); + return exitcode ? exitcode : errorCode; +} + +bool App::OnExceptionInMainLoop() +{ + bool handled{ false }; + wxString error; + try { + throw; // Rethrow the current exception. + } + catch (const FatalException& e) { + error = wsz_program + _wx(e.what()); + if (!errorCode)errorCode = 10; + } + catch (const MyException& e) { + // If we handle an error at this level, the current action has been abruptly terminated, + // and we need to inform the user, but we are not going to terminate the program, + // nor set an error number for exit. + handled = true; + error = wsz_operation + _wx(e.what()); + } + catch (const std::exception& e) { + error = wsz_program + _wx(e.what()); + errorCode = 9; + } + catch (...) { + error = wsz_program + _wx(sz_unknown_error); + errorCode = 8; + } + wxLogError(_T("%s"), error); + wxMessageDialog dlg(singletonFrame, error, wsz_error, wxICON_ERROR); + dlg.SetId(myID_ERRORMESSAGE); + dlg.ShowModal(); + // returning false to exit the main loop and thus terminate the program. + return handled; +} + +void App::OnInitCmdLine(wxCmdLineParser& parser) +{ + parser.SetDesc(g_cmdLineDesc); + // must refuse '/' as parameter starter or cannot use "/path" style paths + parser.SetSwitchChars(_T("-")); + //Command line parameters + parser.SetLogo(wsz_commandLineLogo); + parser.AddUsageText(wsz_usageText); +} + +bool App::OnCmdLineParsed(wxCmdLineParser& parser) +{ + for (const auto& arg : parser.GetArguments()) { + wxString optionName; + switch (arg.GetKind()) + { + case wxCMD_LINE_SWITCH: + optionName = arg.GetShortName(); + if (optionName == _T("t")) { + m_unit_test = !arg.IsNegated(); + } + else if (optionName == _T("l")) { + m_display = !arg.IsNegated(); + } + else if (optionName == _T("d")) { + m_display |= m_display_in_front = !arg.IsNegated(); + } + else if (optionName == _T("f")) { + m_log_focus_events = !arg.IsNegated(); + if (m_log_focus_events) { + Bind( + wxEVT_IDLE, + +[](wxIdleEvent& event) { //Since this function is only ever used once, never being unbound, using a lambda to avoid naming it. + static wxWindow* lastFocus = (wxWindow*)NULL; + //wxLogMessage(_T("OnIdle")); + wxWindow* curFocus = ::wxWindow::FindFocus(); + if (curFocus != lastFocus && curFocus) + { + lastFocus = curFocus; + wxString name{ "" }; + do { + name = wxString(_T("/")) + curFocus->GetClassInfo()->GetClassName() + _T(":") + curFocus->GetName() + name; + } while (curFocus = curFocus->GetParent()); + wxLogMessage(name); + } + event.Skip(); //Called so we can bind multiple tasks to idle, and they will all be handled. + } + ); + } + } + else if (optionName == _T("q")) { + m_quick_unit_test = !arg.IsNegated(); + m_complete_unit_test = m_complete_unit_test && !m_quick_unit_test; + } + else if (optionName == _T("c")) { + m_complete_unit_test = !arg.IsNegated(); + m_quick_unit_test = m_quick_unit_test && !m_complete_unit_test; + } + break; + case wxCMD_LINE_OPTION: + assert(false); +/* switch (arg.GetType()) { + case wxCMD_LINE_VAL_NUMBER: + // do something with itarg->GetLongVal(); + break; + case wxCMD_LINE_VAL_DOUBLE: + // do something with itarg->GetDoubleVal(); + break; + case wxCMD_LINE_VAL_DATE: + // do something with itarg->GetDateVal(); + break; + case wxCMD_LINE_VAL_STRING: + // do something with itarg->GetStrVal(); + break; + }*/ + break; + case wxCMD_LINE_PARAM: + m_params.push_back(arg.GetStrVal()); + // This intended to support subcommand processing, but not handling subcommands yet + // g_cmdLineDesc has been set to disallow multiple arguments. + break; + default: + assert(0); + break; + } + } + return true; +} + +void App::OnError(wxCommandEvent& event) +{ + // We use this to display errors where throwing would cause problems, as in a destructor + // Instead we post an event to be handled in due course. + wxMessageDialog dlg(singletonFrame, event.GetString(), _T("Error"), wxICON_ERROR); + dlg.SetId(myID_ERRORMESSAGE); + dlg.ShowModal(); +} + +int App::OnExit() +{ + assert(pConfig.get()); + if (errorCode)wxLogDebug("%s", szError); + return 0; +} diff --git a/app.h b/app.h new file mode 100644 index 0000000..4ec8189 --- /dev/null +++ b/app.h @@ -0,0 +1,58 @@ +#pragma once + +class App : public wxApp +{ +public: + std::unique_ptrpConfig; + // pConfig corresponds to the Windows Registry entry Computer\HKEY_CURRENT_USER\Software\ro\wallet + // Don't use the registry for stuff better served by wxStandardPaths and sqlit3 files located + // in locations specified by wxStandardPaths + App(); + virtual ~App(); + virtual bool OnInit() wxOVERRIDE; + virtual int OnExit() wxOVERRIDE; + virtual int OnRun() wxOVERRIDE; + void OnError(wxCommandEvent&); + virtual void OnInitCmdLine(wxCmdLineParser& parser) wxOVERRIDE; + virtual bool OnCmdLineParsed(wxCmdLineParser& parser) wxOVERRIDE; + virtual bool OnExceptionInMainLoop() wxOVERRIDE; + bool m_unit_test{ false }; + bool m_display{ false }; + bool m_display_in_front{ false }; + bool m_log_focus_events{ false }; + bool m_quick_unit_test{ false }; + bool m_complete_unit_test{ false }; + wxVector m_params; + }; + +void UnitTest(wxIdleEvent& event); + +static constexpr wxCmdLineEntryDesc g_cmdLineDesc[] = +{ + { wxCMD_LINE_SWITCH, "h", "help", "displays help on the command line parameters.", + wxCMD_LINE_VAL_NONE, wxCMD_LINE_OPTION_HELP }, + { wxCMD_LINE_SWITCH, "t", "test", "-t or --test performs unit test, exits on completion of " + "unit test returning error value.", + wxCMD_LINE_VAL_NONE, wxCMD_LINE_SWITCH_NEGATABLE}, + { wxCMD_LINE_SWITCH, "q", "quick", "-qt or --quick --test performs those unit tests that do not cause noticeable startup delay.", + wxCMD_LINE_VAL_NONE, wxCMD_LINE_SWITCH_NEGATABLE}, + { wxCMD_LINE_SWITCH, "c", "complete", "-ct or --complete --test tests everything.", + wxCMD_LINE_VAL_NONE, wxCMD_LINE_SWITCH_NEGATABLE}, + { wxCMD_LINE_SWITCH, "d", "display", "-d or --display enables display of log in front. " + "Usually used with unit test as -dct. " + "If the log is displayed, then does not exit on completion of unit test.", + wxCMD_LINE_VAL_NONE, wxCMD_LINE_SWITCH_NEGATABLE}, + { wxCMD_LINE_SWITCH, "l", "log", "-l or --log enables display of log behind. " + "Usually used with unit test as -lt. " + "If the log is displayed, then does not exit on completion of unit test.", + wxCMD_LINE_VAL_NONE, wxCMD_LINE_SWITCH_NEGATABLE}, + { wxCMD_LINE_SWITCH, "f", "focus", "-f or --focus causes focus events to be logged for debugging purposes. " + "Usually used as -lf or -lft, as logging them without displaying them is useless.", + wxCMD_LINE_VAL_NONE, wxCMD_LINE_SWITCH_NEGATABLE}, + { wxCMD_LINE_PARAM, "", "", "mywallet.wallet", + wxCMD_LINE_VAL_NONE, /*wxCMD_LINE_PARAM_MULTIPLE|*/wxCMD_LINE_PARAM_OPTIONAL}, + { wxCMD_LINE_NONE } +}; + +DECLARE_APP(App) +inline App *singletonApp{nullptr}; diff --git a/bit_hacks.h b/bit_hacks.h new file mode 100644 index 0000000..8e84b7e --- /dev/null +++ b/bit_hacks.h @@ -0,0 +1,102 @@ +#pragma once + +// We should template this to use __popcnt64 if available +// but that is premature optimization +inline uint64_t bitcount(uint64_t c) { + c = c - ((c >> 1) & 0x5555555555555555); + c = ((c >> 2) & 0x3333333333333333) + + (c & 0x3333333333333333); + c = ((c >> 4) + c) & 0x0F0F0F0F0F0F0F0F; + c = ((c >> 8) + c) & 0x00FF00FF00FF00FF; + c = ((c >> 16) + c) & 0x0000FFFF0000FFFF; + c = ((c >> 32) + c) & 0x00000000FFFFFFFF; + return c; +} + +// http://graphics.stanford.edu/~seander/bithacks.html#IntegerLog "Bit Hacks" +// Find ⌊log2⌋ +// We should template this to use lzcnt64, __builtin_clz or _BitScanReverse if available, +// but that is premature optimization. +inline auto rounded_log2(uint32_t v) { + // This algorithm extends to 64 bits, by adding a step, shrinks to sixteen bits by removing a step. + decltype(v) r{ 0 }, s; + // This redundant initialization and redundant |= of r can be eliminated, + // but eliminating it obfuscates the simplicity of the algorithm. + s = (v > 0xFFFF) << 4; v >>= s; r |= s; + s = (v > 0x00FF) << 3; v >>= s; r |= s; + s = (v > 0x000F) << 2; v >>= s; r |= s; + s = (v > 0x0003) << 1; v >>= s; r |= s; + r |= (v >> 1); + // result of ⌊log2(v)⌋ is in r + return r; +} + +// For trailing bits, consider int __builtin_ctz (unsigned int x) +// http://graphics.stanford.edu/~seander/bithacks.html#ZerosOnRightLinear + +// Count the consecutive trailing zero bits +inline auto trailing_zero_bits(uint64_t v) { + unsigned int c; + if (v & 0x3F) { + v = (v ^ (v - 1)) >> 1; // Set v's trailing 0s to 1s and zero rest + for (c = 0; v; c++) { + v >>= 1; + } + } + else { + c = 1; + if ((v & 0xffffffff) == 0) { + v >>= 32; + c += 32; + } + if ((v & 0xffff) == 0) { + v >>= 16; + c += 16; + } + if ((v & 0xff) == 0){ + v >>= 8; + c += 8; + } + if ((v & 0xf) == 0){ + v >>= 4; + c += 4; + } + if ((v & 0x3) == 0) { + v >>= 2; + c += 2; + } + if ((v & 0x1) == 0) { + v >>= 1; + c += 1; + } + c -= v & 0x01; + } + return c; +} + + + + + + + + + + + + + + + + + + + + + + + + + + + \ No newline at end of file diff --git a/db_accessors.h b/db_accessors.h new file mode 100644 index 0000000..166b3db --- /dev/null +++ b/db_accessors.h @@ -0,0 +1,220 @@ +#pragma once +namespace ro { + // Compile time test to see if a type can be directly read from or written to an sqlite3 file + // This can be used in if constexpr (is_sqlite3_field_type::value) + template struct is_sqlite3_field_type + { + template static constexpr decltype(std::declval().Isqlite3_bind(1, std::declval()), bool()) test() { + return true; + } + template static constexpr bool test(int = 0) { + return false; + } + static constexpr bool value = is_sqlite3_field_type::template test(); + }; + + static_assert(is_sqlite3_field_type::value); + + //Owns a compiled sql statement and destroys it when it is deconstructed. + //Has move semantics. + class sql : public std::unique_ptr { + public: + class null {}; + sql(ISqlite3* p, const char* sz) :std::unique_ptr(sqlite3_prepare(p, sz)) {} + sql(const std::unique_ptr& p, const char* sz) :std::unique_ptr(sqlite3_prepare(p.get(), sz)) {} + // copy constructor + sql(const sql& a) = delete; + // move constructor + sql(sql&& p) :std::unique_ptr(p.release()) { } + // copy assignment + sql& operator=(const sql) = delete; + // Move assignment + sql& operator=(sql&& p) { + std::unique_ptr::reset(p.release()); + } + sql(Icompiled_sql* p) :std::unique_ptr(p) {} + sql(std::unique_ptr&& p) :std::unique_ptr(p.release()) { } + ~sql() = default; + template auto column(int i) const { + if constexpr (ro::is_blob_field_type::value) { + auto st = (*this)->Isqlite3_column_blob(i); + if (st.size_bytes() != sizeof(T)) throw BadDataException(); + static_assert (std::is_standard_layout(), "not standard layout"); + static_assert (std::is_trivial(), "not trivial"); + return reinterpret_cast(&st[0]); + } + else if constexpr (std::is_integral::value) { + if constexpr (sizeof(T) > sizeof(int_least32_t)) { + T retval = (*this)->Isqlite3_column_int64(i); + return retval; + } + else { + T retval = (*this)->Isqlite3_column_int(i); + return retval; + } + } + else if constexpr (std::is_same< T, std::span>::value) { + return (*this)->Isqlite3_column_blob(i); + } + else if constexpr (std::is_same< T, const char*>::value) { + auto sz{ (*this)->Isqlite3_column_text(i) }; +// if (!IsValidUtf8String(sz)) throw NonUtf8DataInDatabase(); + return sz; + } + else { + static_assert(false, "Don't know how to read this datatype from database"); + return null(); + } + } + + template const sql& read(int i, T& j) const{ + if constexpr (ro::is_blob_field_type::value) { + auto st = (*this)->Isqlite3_column_blob(i); + if (st.size_bytes() != sizeof(T)) throw BadDataException(); + static_assert (std::is_standard_layout(), "not standard layout"); + static_assert (std::is_trivial(), "not trivial"); + j = *reinterpret_cast(&st[0]); + } + else if constexpr (std::is_integral::value) { + if constexpr (sizeof(T) > sizeof(int_least32_t)) { + j = (*this)->Isqlite3_column_int64(i); + } + else { + j = (*this)->Isqlite3_column_int(i); + } + } + else if constexpr (std::is_same< T, std::span>::value) { + j = (*this)->Isqlite3_column_blob(i); + } + else if constexpr (std::is_same< T, const char*>::value) { + j = (*this)->Isqlite3_column_text(i); + } + else { + static_assert(false, "Don't know how to read this datatype from database"); + } + return *this; + } + + void bind(int i, null) { (*this)->Isqlite3_bind(i); } + template < typename T, + typename std::enable_if::value, int >::type dummy_arg = 0 > + void bind(int i, T j) { + (*this)->Isqlite3_bind(i, j); + } + template < typename T, + typename std::enable_if::value, int >::type dummy_arg = 0 > + void bind(int i, const T& j) { + static_assert(ro::is_serializable::value, "Don't know how to store this type in a database"); + (*this)->Isqlite3_bind(i, ro::serialize(j)); + } + typedef Icompiled_sql::sql_result result; + result step() { + return (*this)->Isqlite3_step(); + } + void final_step() { + if (step() != result::DONE) throw SQLexception("SQL: Unexpected rows remaining"); + } + void reset() { + (*this)->Isqlite3_reset(); + }// https://sqlite.org/c3ref/reset.html + + template void bind(int i, const T& first, const Args& ...args) { + bind(i, first); + bind(i + 1, args...); + } + template void do_one(const Args& ...args) { + reset(); + if constexpr (sizeof...(args) > 0) { + bind(1, args...); + } + final_step(); + } + template auto read_one(const Args& ...args) { + reset(); + if constexpr (sizeof...(args) > 0) { + bind(1, args...); + } + return step() == result::ROW; + } + }; + + class sql_update :sql { + public: + using sql::sql; + sql_update(ISqlite3* p, const char* sz) : sql(p, sz) {} + template void operator()(const Args& ...args) { + do_one(args...); + } + }; +} + +class sql_update_to_misc :ro::sql { +public: + sql_update_to_misc(ISqlite3* p) : sql(p, R"|(REPLACE INTO "Misc" VALUES(?1, ?2);)|") {} + sql_update_to_misc(const std::unique_ptr& p) : sql_update_to_misc(p.get()) {} + templatevoid operator()(int i, const T& j) { + do_one(i, j); + } + templatevoid operator()(int i, T* j) { + do_one(i, j); + } +}; + +class sql_read_from_misc :ro::sql { +public: + sql_read_from_misc(ISqlite3 *p) : sql(p, R"|(SELECT "m" FROM "Misc" WHERE "index" = ?1;)|") {} + sql_read_from_misc(const std::unique_ptr& p) : sql_read_from_misc(p.get()){} + auto operator()(int i) { + return read_one(i); + } + templateauto value() { + return column(0); + } + template void read(T& j) const { + sql::read(0,j); + } +}; + +class sql_insert_name { +public: + ro::sql csql_begin; + ro::sql csql_into_names; + ro::sql csql_namekey_into_keys; + ro::sql csql_commit; + sql_insert_name(ISqlite3* p) : + csql_begin(p, R"|(BEGIN;)|"), + csql_into_names(p, R"|(INSERT OR ROLLBACK INTO "Names" VALUES(?1);)|"), + csql_namekey_into_keys(p, R"|(INSERT OR ROLLBACK INTO "Keys" VALUES(?1, last_insert_rowid(), 1);)|"), + csql_commit(p, R"|(COMMIT;)|") { + } + sql_insert_name(const std::unique_ptr& p) : sql_insert_name(p.get()) {} + void operator()(const char* psz, const ristretto255::point& pt) { + csql_begin.do_one(); + try { + csql_into_names.do_one(psz); + csql_namekey_into_keys.do_one(pt); + } + catch (const std::exception & e) { + csql_commit.do_one(); + throw e; + } + csql_commit.do_one(); + } +}; + +class sql_read_name :ro::sql { +public: + sql_read_name(ISqlite3* p) : sql(p, R"|(SELECT * FROM "Names" WHERE OID = ?1;)|") {} + sql_read_name(const std::unique_ptr& p) : sql_read_name(p.get()) {} + bool operator()(int i) { + return read_one(i) == Icompiled_sql::ROW; + } + auto name() const { + return sql::column(0); + } +}; + + + +constexpr auto WALLET_FILE_IDENTIFIER (0x56d34bc5a655dd1fi64); +constexpr auto WALLET_FILE_SCHEMA_VERSION_0_0(1); diff --git a/display_wallet.cpp b/display_wallet.cpp new file mode 100644 index 0000000..bf6c864 --- /dev/null +++ b/display_wallet.cpp @@ -0,0 +1,62 @@ +#include "stdafx.h" +using ro::base58; +display_wallet::display_wallet(wxWindow* parent, wxFileName& walletfile) : + wxPanel(parent, myID_WALLET_UI, wxDefaultPosition, wxDefaultSize, wxTAB_TRAVERSAL, _T("Wallet")), + m_db(nullptr) +{ + wxLogMessage(_T("Loading %s"), walletfile.GetFullPath()); + if (!walletfile.IsOk() || !walletfile.HasName() || !walletfile.HasExt()) throw MyException("unexpected file name"); + if (!walletfile.FileExists())throw MyException( + walletfile.GetFullPath().append(" does not exist.").ToUTF8() + ); + m_db.reset(Sqlite3_open(walletfile.GetFullPath().ToUTF8())); + sql_read_from_misc read_from_misc(m_db); + if (!read_from_misc(1) || read_from_misc.value() != WALLET_FILE_IDENTIFIER)throw MyException(sz_unrecognizable_wallet_file_format); + if (!read_from_misc(2) || read_from_misc.value() != WALLET_FILE_SCHEMA_VERSION_0_0 || !read_from_misc(4))throw MyException(sz_unrecognized_wallet_schema); + read_from_misc.read(m_MasterSecret); + if (!m_MasterSecret.valid()) throw MyException(sz_cold_wallets_not_yet_implemented); + auto sizer = new wxBoxSizer(wxHORIZONTAL); + m_lSizer = new wxBoxSizer(wxVERTICAL); + m_rSizer = new wxBoxSizer(wxVERTICAL); + sizer->Add(m_lSizer,0, wxGROW, 4); + sizer->Add(m_rSizer, 50, wxGROW, 4); + SetSizer(sizer); + ro::sql read_keys(m_db, R"|(SELECT * FROM "Keys";)|"); + sql_read_name read_name(m_db); + // m_db.reset(nullptr);// Force error of premature destruction of Isqlite3 + while (read_keys.step() == Icompiled_sql::ROW) { + auto pubkey = read_keys.column(0); + auto id = read_keys.column(1); + auto use = read_keys.column(2); + if (use != 1)throw MyException(sz_unknown_secret_key_algorithm); + if (!read_name(id)) throw MyException(sz_no_corresponding_entry); + const char* name = read_name.name(); + if (m_MasterSecret(name).timesBase() != *pubkey)throw MyException(std::string(sz_public_key_of) + name + sz_fails_to_correspond); + m_lSizer->Add( + new wxStaticText( + this, + wxID_ANY, + name, + wxDefaultPosition, wxDefaultSize, wxALIGN_RIGHT|wxST_ELLIPSIZE_END + ), + 10, + wxEXPAND | // make horizontally stretchable + wxALL, // and make border all around + 2); + m_rSizer->Add( + new wxStaticText( + this, + wxID_ANY, + "#" + base58(*pubkey).operator std::string(), + wxDefaultPosition, wxDefaultSize, wxALIGN_LEFT | wxST_ELLIPSIZE_END + ), + 10, + wxEXPAND | // make horizontally stretchable + wxALL, // and make border all around + 2); + } + this->SetSize(this->GetParent()->GetClientSize()); + singletonFrame->m_LastUsedSqlite.Assign(walletfile); +} +display_wallet::~display_wallet() { +} \ No newline at end of file diff --git a/display_wallet.h b/display_wallet.h new file mode 100644 index 0000000..bc5a9b6 --- /dev/null +++ b/display_wallet.h @@ -0,0 +1,16 @@ +#pragma once +class display_wallet : public wxPanel +{ +public: + display_wallet(wxWindow*, wxFileName&); + ~display_wallet(); +private: + std::unique_ptr m_db; + ristretto255::CMasterSecret m_MasterSecret; + wxBoxSizer* m_lSizer; + wxBoxSizer* m_rSizer; +}; + + + + diff --git a/docs/BlackHat-DC-09-Marlinspike-Defeating-SSL.pdf.URL b/docs/BlackHat-DC-09-Marlinspike-Defeating-SSL.pdf.URL new file mode 100644 index 0000000..dfe8a82 --- /dev/null +++ b/docs/BlackHat-DC-09-Marlinspike-Defeating-SSL.pdf.URL @@ -0,0 +1,2 @@ +[InternetShortcut] +URL=http://www.blackhat.com/presentations/bh-dc-09/Marlinspike/BlackHat-DC-09-Marlinspike-Defeating-SSL.pdf diff --git a/docs/Breaking_out_of_the_browser.pdf b/docs/Breaking_out_of_the_browser.pdf new file mode 100644 index 0000000..3c3f4c7 Binary files /dev/null and b/docs/Breaking_out_of_the_browser.pdf differ diff --git a/docs/Efficient_Error-Propagating_Block_Chaining.pdf b/docs/Efficient_Error-Propagating_Block_Chaining.pdf new file mode 100644 index 0000000..3c1cd86 Binary files /dev/null and b/docs/Efficient_Error-Propagating_Block_Chaining.pdf differ diff --git a/docs/How to be a program manager - Joel on Software.URL b/docs/How to be a program manager - Joel on Software.URL new file mode 100644 index 0000000..814dbb7 --- /dev/null +++ b/docs/How to be a program manager - Joel on Software.URL @@ -0,0 +1,2 @@ +[InternetShortcut] +URL=http://www.joelonsoftware.com/items/2009/03/09.html diff --git a/docs/InternetProtocol.pdf b/docs/InternetProtocol.pdf new file mode 100644 index 0000000..c224027 Binary files /dev/null and b/docs/InternetProtocol.pdf differ diff --git a/docs/PracticalLargeScaleDistributedKeyGeneration.pdf b/docs/PracticalLargeScaleDistributedKeyGeneration.pdf new file mode 100644 index 0000000..d7bd97f Binary files /dev/null and b/docs/PracticalLargeScaleDistributedKeyGeneration.pdf differ diff --git a/docs/SecureDistributedKeyGeneration.pdf b/docs/SecureDistributedKeyGeneration.pdf new file mode 100644 index 0000000..3025040 Binary files /dev/null and b/docs/SecureDistributedKeyGeneration.pdf differ diff --git a/docs/SoK_Diving_into_DAG-based_Blockchain_Systems.pdf b/docs/SoK_Diving_into_DAG-based_Blockchain_Systems.pdf new file mode 100644 index 0000000..fec4450 Binary files /dev/null and b/docs/SoK_Diving_into_DAG-based_Blockchain_Systems.pdf differ diff --git a/docs/ThresholdSignatures.pdf b/docs/ThresholdSignatures.pdf new file mode 100644 index 0000000..d0c2af1 Binary files /dev/null and b/docs/ThresholdSignatures.pdf differ diff --git a/docs/anonymous_blockchain_transactions.pdf b/docs/anonymous_blockchain_transactions.pdf new file mode 100644 index 0000000..e1ba97d Binary files /dev/null and b/docs/anonymous_blockchain_transactions.pdf differ diff --git a/docs/anonymous_multihop_locks_lightning_network.pdf b/docs/anonymous_multihop_locks_lightning_network.pdf new file mode 100644 index 0000000..7be6d6d Binary files /dev/null and b/docs/anonymous_multihop_locks_lightning_network.pdf differ diff --git a/docs/bitcoin_vulnerable_to_currency_controls.md b/docs/bitcoin_vulnerable_to_currency_controls.md new file mode 100644 index 0000000..f96b704 --- /dev/null +++ b/docs/bitcoin_vulnerable_to_currency_controls.md @@ -0,0 +1,153 @@ +--- +lang: 'en-US' +title: How could regulators successfully introduce Bitcoin censorship and other dystopias +--- +[Original document](https://juraj.bednar.io/en/blog-en/2020/11/12/how-could-regulators-successfully-introduce-bitcoin-censorship-and-other-dystopias/) by [Juraj Bednar](https://juraj.bednar.io/en/juraj-bednar-2/) + +Publishing this a violation of copyright. Needs to be summarized and paraphrased. + +**Note:** A lot of people think this is purely about \>50% attack. Not true, [here’s how this unfolds with 10% of censoring hashrate.](https://threadreaderapp.com/thread/1327206062437621760.html) + +Bitcoin is often said to be anonymous and uncensorable. Thanks to chain analysis, anonymity is to some extent a disputed wishful thinking from the past. And it looks like it won’t be so nice with censorship resistance either. + +My reasoning begins with this quote from Twitter of fluffypony: +Note: A lot of people think this is purely about >50% attack. Not true, here’s how this unfolds with 10% of censoring hashrate. + +![blockseer](blockseer.jpg){width=100%} + +( [More information here, for example](https://cointelegraph.com/news/slippery-slope-as-new-bitcoin-mining-pool-censors-transactions) ) + +This mining pool censors transactions that are included in the government +blacklist. For the time being, the pool just leaves money on the table, so +if the pool decides not to include “dirty” transactions, the end result is +that they do not to earn transaction fees for that transaction (and go for +cheaper transactions) and the “dirty” transaction is mined in another block +by a different miner. But I think it’s still a dangerous precedent and it +gets scary when you think it through. + +I think if governments or anti-money laundering organizations want to censor Bitcoin, that’s exactly the first step. Try it out on one pool. But if at least one pool mines these transactions, we’re fine, right? Not really. + +Let’s think about what these organizations might do next. Spoiler alert, these steps lead to successful censorship of Bitcoin: + +- Miners have invested a lot of money in the mining hardware, data centers, they are paying electricity and taxes – especially large mining operations. They are mostly **not cypherpunks**, but corporations with shareholders, the CEO wears a suit and a tie, the company has a business permit, all stamps in order, ... At the same time they need a bank account and an account at some Bitcoin exchange, because they have to pay suppliers (for energy, rent, taxes, ...) . +- If the government comes in and says, “You can’t mine the blocks that spend these UTXOs”, or you’ll lose either bank account, exchange account, business permit or go to jail for money laundering, most of the big miners would comply. Blockseer is just a first example. They have shareholders that are awaiting dividends, they are not rebels against establishment. By not mining certain transactions, they are only losing some transaction fees and at this point, no one would know. Some miners even occasionally mine empty blocks even if mempool is not empty, so **not including a transaction is not an unusual thing to do**. +- Very important note here is that **anti-money laundering regulations and blacklists are mostly global** and they are not approved by states’ parliaments or governments. The enforcement is done through network effects. If you want to be connected in the payment network of the world (SEPA, SWIFT, ACH, ...) is dependent on how well you fight money laundering and how well you are implementing AML standards. These standards are created by organizations such as the [FATF-GAFI](https://translate.googleusercontent.com/translate_c?depth=1&pto=aue&rurl=translate.google.com&sl=sk&sp=nmt4&tl=en&u=https://en.wikipedia.org/wiki/Financial_Action_Task_Force&usg=ALkJrhg3fRHeA2Jvu1BnXtmrZfFfIwnzxg) (remember [crypto travel rule](http://www.fatf-gafi.org/publications/fatfrecommendations/documents/regulation-virtual-assets-interpretive-note.html)?). Thus, it is quite possible that such an organization will start publishing blacklists and they will be accepted by miners in all countries under the threat of jail / loss of business license / loss of bank account / loss of exchange account. For the last two, you only need FATF AML network effects and not local law! The FATF-GAFI rules are considered an international standard of fighting against AML. States are just saying that some entities need to fight money laundering by referring to international standards that are not approved by any parliament in any country. +- Of course, there will always be small scale miners who have bought ASIC and are mining in their kitchens or balconies. They could not care less about FATF-GAFI travel rules. But it won’t help much... +- If these rules are followed by more than 50% of hashrate, there may be a simple addition to the rules: “If you build upon a block that contains prohibited transactions, you are laundering money.“ This is actually a soft fork introduced by the regulator (beware, not even directly by a state). +- This creates a weird [Schelling point](https://en.wikipedia.org/wiki/Focal_point_(game_theory)) situation. Even if I am a miner mining in my kitchen and don’t care about any transaction blacklist, if there is even a double digit probability that if I (or my pool rather) find a block and it will become orphaned, because the next block will be found by a big miner, I will think twice about including a blacklisted transaction in the blockchain. The math is simple – I either get an additional \~ 5 USD transaction fee, but I could lose the whole block reward, or I don’t include this transaction and can keep the whole block reward (coinbase + all tx fees). Including a tainted transaction is low upside, huge downside decision. Even if I am not under a jurisdiction of such a rule and I don’t particularly want to censor transactions, if I understand risk-reward properly, I will just omit it. No harm done, someone else can try. +- Majority hashrate can introduce such soft fork. In addition to the fact that the company can continue to do business and ensure a return on its investment in hardware, the second reason is economic. You don’t want to mine later orphaned blocks. But it gets even better for the censors. The effect would be the same as if the miners who do not meet the new soft fork rules shut down their mining machines (if there is at least one transaction with “dirty coins” in the mempool). Why? If a lot of miners decide to refuse tainted transactions and build upon the blocks with tainted transactions, every miner that mines them into block is basically burning electricity for nothing. It is the same as if they just stopped mining – the “softfork” hashrate is lower, so the reward per hashrate is higher. Introducing a softfork means more money for complying miners and no money for non-compliant miners. The compliant miners mine more blocks, it is as if the non-compliant miners did not exist. This means that **the miners have an economic incentive to enforce this rule**. Let me repeat this: Even the miners that are morally against this rule are economically motivated to comply. +- If governments or FATF-GAFI do not want to wait for the majority of the hashrate, they can implement this rule even faster. Just tell the exchanges “if you want to be connected to the fiat payment network, your Bitcoin nodes can only accept compliant blocks, otherwise you will be laundering money and we will shut you off from the fiat network and send nice SEC agents to your shiny office”. Exchangers are dependent on connectivity to the fiat network, because new capital flows through it to the crypto economy and that’s how they make money – from the fees on trades. Kraken, Coinbase, Binance,... will gradually start running full nodes with a blacklist. They are using blacklists already, but they only refuse deposits from tainted addresses. I am talking about refusing blocks with tainted transactions. Guess what the miners do? At the end of the month, they need to send the mined coins to an exchange to get fiat to pay for electricity. For this, it is absolutely essential, that the exchange “sees” the blocks. The miners need to be on the same chain as the exchange, otherwise they are screwed. + +The first step of this dystopian scenario has already taken place. We have the first (albeit minority) pool, which does not include some transactions. At this point, it means nothing, at worst, the transaction is mined a bit later. After the introduction of full soft fork (whether by the hashate majority or the economic majority of exchangers), Bitcoin’s non-censorship practically ended. + +# Lightning network and tainted coins + +We will talk about one more dystopian scenario. Imagine for a moment that you are running a node of the Bitcoin Lightning Network (BTW if you have never tried, [check out my intro course](https://hackyourself.io/product/174/)). So you have installed something like [Umbrel](https://getumbrel.com/) or [BtcPayServer](https://getumbrel.com/), you are a good Bitcoiner – you run a full-node Bitcoin, some Lightning daemon, you even run it all through Tor. You’ve opened a few channels, providing liquidity to route payments, and earning some fees. You do it all to help the network and verify transactions. So far so good. Or – so far, great! + +One day, a local drug dealer on the dark market will check out your node. He needs to launder the drug Bitcoins. He will do this as follows: + +- He will install two Lightning nodes, node A and node B. We will mark your node L. +- He moves his dirty coins to node A onchain. +- He buys incoming liquidity on node B (for example, he buys it from [Bitrefill](https://www.bitrefill.com/buy/lightning-channel/?hl=en)) +- On node A, after the coins are confirmed on chain, he will create a high-capacity channel(s) with your node L. +- On node B, he creates a lightning invoice to receive coins. He will pay the invoice from node A. +- Node A after using all sending capacity is turned off and deleted – it doesn’t even have to close the channel, as there is no capacity on his side, he will only lose the channel reserve. Or he can close the channel and deal with the reserve in a later transaction. +- He closes the channel on node B (cooperative close) and sends nice clean money to his wallet. +- When the channel between node A and node L closes (for example you force close it), you get dirty money from drug sales – tainted coins. You can’t even send them away through Lightning because node A no longer exists. They will end up in your on-chain wallet. + +If there are common chain analysis issues and these “dirty” coins are a problem – either legal, problems with depositing them to an exchange or even the fact that these coins are so tainted that no miner will mine the transaction – letting other nodes open channels with you (with their UTXO) is a serious security risk. If someone succeeds and the coins are marked as “dirty” only after the attacker does this operation, it is quite possible that you will not be able to move the coins anymore. + +Of course, people will probably not be satisfied with this situation and will rightly complain to the state or regulator that they have nothing to do with drug sales and that someone has just opened a channel with them. One possible solution is for the state and anti-money-laundering organizations to say “sorry, our bad, this censorship thing was a bad idea”. Another solution is much more likely: + +*“Dear users of the lightning network, we see that you often get dirty coins. We’ve passed a new law that addresses your issue. We therefore recommend that you install this open-souce module in your Lightning node. Through the API, it verifies that the UTXO through which the other party wants to open the channel is clean. If it is clean, we return a state-signed proof of purity as a result of an API call. If you attach this proof of purity to the transaction as supplementary data, the compliant miner will happily mine it for you, because of course you could not know that the coins were dirty – we did not know either! Thank you for you cooperation in fighting money laundering!* + +***API Call parameters** : KYC ID of the caller, list of UTXOs in the onchain wallet of the node (why not collect extra data that the state does not need? Have you ever seen a government form that did not ask you when and where you were born? Of course they want to know the age and purity of your UTXOs as well), unsigned transaction by which the other party wants to open the channel* + +***Output** : Answer yes-no, State digitally signed certificate of transaction purity* + +You can get your KYC ID at any branch of the Ministry of the Interior, SEC, just bring two documents and proof of ownership of UTXO – a message signed with your identity with the address keys ” + +(*Crypto-anti money laundering lightning enablement act of 202*1) + +(Why enablement? Because when government wants to regulate something, meaning ban something, they always sell it to you that they are enabling you to do something. You know, if it is not forbidden, it is enabled by default, but for some reason, in 2020, if government regulates it, it enables it... Weird, right?) + +OK, they probably won’t be able to pass such a law in 2021, the soft fork dystopia must happen first. But a similar approach has already been taken by the European Union when verifying reverse charge VAT numbers – if you are a VAT paying entity and the customer is a VAT payer in another EU country and therefore you do not invoice VAT, you can verify their VAT ID on the European Union website (or via an API) and **save the call result**. If it is not valid and you do not have stored evidence that you tried to verify it (and it was valid then), you have (perhaps) a problem. But I don’t know that anyone would enforce this rule. + +# Tadaaaa, I’ll do a coinjoin + +If you have “dirty” coins and the miners refuse to mine transactions containing dirty coins, you will most certainly not do a coinjoin. + +Coinjoin is a standard transaction that has inputs – if they are already marked as dirty, then you will not get such a coinjoin transaction into the blockchain (soft-forked away!). + +If you get it into blockchain and the coins are marked later, you have a problem – you could even put completely clean coins in the coinjoin and suddenly you are marked as a drug dealer on the dark market because some other coinjoin participant was marked and tried to launder money. + +If anyone could just use coinjoin to avoid all this censorship, they would. So let’s do it the other way around – coinjoin is an act of money laundering and if any input is tainted, all outputs are tainted. + +Ironically enough, the only safe coinjoin is if the coinjoin provider (and preferably also use) uses and enforces a blacklist. I’ve heard that some coinjoin providers already do this. I don’t know what is worse – if you enforce the blacklist, you are censoring and hurting fungibility. If you are not enforcing blacklist, you taint all your users’ coins and they will be pissed when they want to use them and are not able to. + +Of course, this topic is already relevant now, because many services (such as exchanges) reject dirty coins – and many also reject coinjoin outputs. Even if you withdraw crypto from an exchange to a Wasabi or Samourai and then send it directly to a mixer, you will get a love letter from your exchange, telling you nicely to stop doing that, or they will close your account next time. Of course, they know your name and you have shown your ID, so if you piss them off, they will also report you to your local anti money laundering unit (in my country, that would be financial police). + +# Change it to Monero and back + +If someone has Bitcoins that are not exactly clean and wants to keep Bitcoins, they can exchange Bitcoin for Monero using a decentralized exchange and then after some time (and gradually) change Monero back for Bitcoin, through a reputable exchange (eg [xmr.to](https://xmr.to/)). This will of course cost a few percent in exchange fees and you are also exposed to XMR/BTC exchange rate risk (although it can be both upside risk). + +If many people solve this problem in this way, there will be a lot of tainted coins left in the wallets of the exchanges and their clients. I don’t know how people will deal with it. + +The key is to do it before the coins are marked tainted of course (similar to lightning strategy). + +# Possible solutions to censorship issues + +Anonymous cryptocurrencies such as Monero do not suffer from this problem, at least not so much. The sender, recipient, and amount sent are not visible in the Monero transaction. The Monero transaction refers to your input and ten other inputs. + +This might look similar to a bitcoin coinjoin transaction, but there are key differences: + +- In Monero, this is how you make any transaction. Miners will not mine transactions that do not have decoy inputs. That’s just how Monero works. Changing this would not be a soft fork, but a hard fork (you would need to relax the consensus rules). +- In Coinjoin, you are signing the coinjoin transaction with your key. You have seen it, understand what you are doing and all the parties approve and sign. In Monero, only you sign, the other people are unwilling participants that your wallet chose. If you appear as one possible input of a transaction in a drug deal, you did not even have to know it – you did not even needed to be online when it happened. +- Monero uses stealth addresses, so you cannot blacklist an address. You can only blacklist a particular transaction (without view key of the address). If one address of a dark market is revealed, in Bitcoin, you could cluster many more dark market addresses. In Monero, if a transaction is revealed, it can be blacklisted, but that’s about it. You can maybe blacklist a few transactions with a 90% probability that you are wrong about them. +- Although I’m not a fan of ASIC resistance, because network security depends on proof of work, Monero is much more likely to decentralize miners and have someone mining “in their kitchen”. Mining takes place on regular computers, not on dedicated devices. At the same time, miners include botnets, which is sad, but such miners are motivated not to censor transactions, because I assume they use Monero themselves and need to anonymously spend the rewards. I assume it is much harder to come to the majority of the Monero hashrate and demand they do something (I am not sure about this though). + +So should we just ditch Bitcoin and switch to Monero? Well, there is a different kind of censorship happening – exchanges are kicking privacy coins out. Most [recently ShapeShift](https://decrypt.co/47508/shapeshift-quietly-delists-monero-privacy-coin). + +Here comes the Bitcoin network effect. It is enough if there is one exchange in the world that exchanges Monero for clean untainted Bitcoin without KYC and then any Bitcoin exchange can change it to fiat or anything else. Such exchange, of course, involves two fees (Monero for Bitcoin and Bitcoin for fiat), but it is still possible. + +I call this rule “crypto to crypto fungibility”. Crypto to crypto exchanges are not so easily regulated and all it takes is one that works reasonably well and it does not matter if someone bans one cryptocurrency. It is a “ban all or none” effect in practice. + +# Two Bitcoins + +It is very likely that hard core Bitcoiners will try to resist such censorship. And that’s good. One question is: how is it possible technically? A soft fork is a completely valid chain, with following the consensus rules and a majority soft fork will just be Bitcoin. It is hard to enforce that a miner **includes a transaction**. Consensus rules are good for excluding transactions. Even if there is a hard fork or a checkpoint that all nodes agree on and that includes a tainted transaction, right from the next block a soft fork can continue and censor transactions, including the outputs of the mined tainted transaction. So you **can not easily “fork yourself off” to a censorship resistant fork**. Censorship decisions are made in each new block. You have to win this fight one block at a time, forever, until the end of timechain. + +All this can result in two types of Bitcoin – KYCed and clean vs “black market” Bitcoin. Whether they will live on one blockchain or Bitcoin will be divided into two forked chains depends mainly on the miners and exchanges and their willingness to succumb to the regulatory pressure of the regulators and violent coercion if they fail to comply. + +A paradoxical solution might be to change the hashing algorithm, which would significantly reduce network security (Bitcoin’s Proof of Work currently makes Bitcoin the safest blockchain on the planet). In this way, two Bitcoins would also probably be created – less safe, less regulated and mined in the kitchens all over the world and on the other hand safer but heavily regulated. Where it goes from there, no one knows. Is this enough to avoid censorship? Probably not. Introducing better privacy might help, but then why not just use Monero? + +Thus, the majority hashrate (i.e. miners who control more than half of the power of the network) decide on censorship. These are companies that have their managers, buildings and state licenses. If decentralized mining pools do not have an absolute majority, it will pay off financially to mine on a regulated pool, as we said above. + +The idea that a large miner will “rebel” and move to p2pool (or use Stratum2 and create their own blocks, not dictated by a pool) and problem solved is very naive. Mining companies that control significant hashrates need to achieve a return on their investment in the first place. They are very conservative, don’t want to risk losing rewards by mining blocks that are later orphaned. So the main incentive is not “the government will kick our door if we mine this transaction”. The incentive is simple “let’s kick out all the hashrate that does not comply, more block rewards for us and make sure that no one will kick out our hashrate”. + +Bitcoiners like to signal the virtue of running their own node and how this makes sure that all rules are followed and helping to decentralize the network. While this is nice and I applaud everyone who runs their own node, decentralization from the point of view of censorship is mainly about miners, and running one’s own node will not help in any way. + +(Of course, we can create new rules – blocks that do not involve censored transactions with a sufficient fee to reject as blocks of censors. This has several problems though – how do you know that everyone sees this transaction? If there is already a consensus about transactions, you would not need miners. So this is nicely said, but very difficult to actually achieve. It would probably also lead to two Bitcoins – guess which version would Coinbase, Kraken, Binance, Bitstamp,... and for that matter Microstrategy run? + +# Conclusion + +The idea of ​​the unstoppability and uncontrollability of Bitcoin is, in my opinion, an outdated concept. In the past, we could not imagine what censors and regulators could do. We thought that a rule like the crypto travel rule from FATF that is already in force was pure sci-fi – how could states agree to regulate all exchanges in the world the same way? They cannot even agree on the type of power outlet! Yet, it happened. FATF rules are enforced globally through network effects. These rules apply in Europe, the US and China as well. Without any need for elected officials to pass it through democratic rituals. The way that power and enforcement works in the last few years has changed dramatically. While Bitcoiners still believe it is not possible, we are being regulated more and more – and using power structures that have nothing to do with the ideals of democracy. One office in OECD office in Paris is writing worldwide AML regulations. Another office in the same building created the reporting standards that invade our privacy (the Common Recording Standard – CRS). Payment networks create and enforce their own regulations – even outside their users! + +What can we do to make sure this dystopia does not happen? Build a parallel society that does not rely on regulated services (shops, courts, exchanges, ...). Treat anonymity and privacy as a feature. A core feature. [Reject](https://kycnot.me/) any KYC-requiring service in principle and become an ethical crypto dealer. Buy and sell crypto. Support any services that do not ask for our identity. Promote, build and use decentralized exchanges, ATMs, and local in-person crypto exchange communities. And build a crypto economy that blatantly rejects these ideas, but not only on social media, but in reality. + +If the split of Bitcoin into regulated and unregulated really occurs, the unregulated one should have the greatest network effect, the greatest economic power. It should be the Bitcoin, in which we settle small debts with friends and family. The Bitcoin with which we buy vegetables that someone else grew in their garden. And we should also support cryptocurrencies like Monero, which are not traceable and their censorship is much more difficult to achieve. It is not that hard to admit, that privacy is a good thing to have, even if you are a “Bitcoin is hard money maximalist”. They play along nicely. + +If this Bitcoin’s global censorship really takes place under the leadership of states or other AML organizations, we should have the strength to say “we don’t want this centralized coin, it’s the same shit as your central bank issued digital fiat money.” And “no, thank you.” + +And the time to start building this situation and this network effect is now. + +# Learn more + +A [Twitter thread](https://threadreaderapp.com/thread/1327206062437621760.html) about how this attack unfolds with 10% hashrate enforcing censorship and what is the cost-benefit analysis for individual miners. + +I made a [course](https://hackyourself.io/product/174/) about how to settle small debts among friends and family and use Lightning network to pay through non-KYC exchanges. If you have never tried Lightning network and don’t know where to start, this might be a good start. Open channels when fees are low, you can thank me later. + +I also produce a podcast dedicated to increasing our options, thus increasing our freedom. It’s called [Option Plus Podcast](https://optionplus.io/). There are episodes about opting out, strategies for being more free here and now. If you want to learn more about strategy of parallel societies, I recommend a [Cypherpunk Bitstream episode](https://taz0.org/bitstream/0x0b-the-roots-of-parallel-polis/), where Smuggler and Frank invited me and Martin to talk about Parallel Polis – a strategy to achieve more liberty in a communist dictatorship of former communist Czechoslovakia. Yes, we can use this strategy today. + +If you want to learn more about financial surveillance and how it applies to crypto – and especially how it is made and enforced outside of parliaments and governments, check out [my talk from HCPP on Financial Surveillance and Crypto Utopias](https://juraj.bednar.io/en/talk-en/2019/10/16/financial-surveillance-and-crypto-utopias-recording-from-hcpp19/). + +You can also [follow me on Twitter \@jurbed](https://twitter.com/jurbed). diff --git a/docs/block_chain_scaling.md b/docs/block_chain_scaling.md new file mode 100644 index 0000000..02848b2 --- /dev/null +++ b/docs/block_chain_scaling.md @@ -0,0 +1,106 @@ +--- +title: Blockchain Scaling +--- +A blockchain is an immutable append only ledger, which ensures that +everyone sees the same unchanging account of the past. A principal +purpose of blockchain technology is to track ownership of assets on a +ledger that is distributed among a large number of computers in such a +way that it is not vulnerable to alteration, destruction, or loss at any single location. + +The peers that determine the consensus on each new block only need to +have the unspent transaction outputs in fast storage, whereupon they +generate a new consensus of unspent transaction outputs. They could +throw away all the transactions as soon as there is consensus on the +current state that is the result of applying those transactions to the previous +state. Or some of them could throw away all the transactions, while others +transfer them to distributed and slow storage. + +But this creates the opportunity to inject a fake history with no past +through a fifty one attack. + +At scale you have a lot of transactions, a lot of clients, and considerably +fewer peers, so you worry about peers conspiring to quietly introduce new +unspent transactions with no past through the fifty one percent attack. + +Any dilution has to take place through a process that leaves clearly in the +blockchain evidence of dilution that everyone can see. One way to make +sure of this is that when any peer asserts that a transaction set leads to a +mutable state, and another peer does not agree, peers that persistently +agree will never reach consensus with peers that agree, and we get an +automatic fork. + +When there are many transactions, the computers constructing the final +consensus hash of the final block, which testifies to the entire immutable +consensus past, are necessarily rather few, rather large, and owned by a +rather small number of rather wealthy people. + +To keep them honest, need a widely distributed history of at least the past +few weeks. + +We need a large number of people making sure, and able to make sure, that the +history is consistent from day to day, not just from block to block. + +Everyone should keep the transactions to which he is a party, and the subset +of the Merkle–patricia tree linking them to the past consensus on unspent +transactions, and to the current consensus of the entire history of the +blockchain, but this is not enough to prevent a 51% attack from injecting new +history with no past. + +The full list of unspent transaction outputs needs to be kept locally in very +fast storage sorted and accessed by a primary key that keeps the transaction +approximately in temporal order, so that older, and less frequently needed, +transaction outputs are stored together, and newer and more likely to be +needed transaction outputs are stored together. + +As this ledger can potentially grow quite large, it needs to be subdivided +into general ledger and subledgers.  When the general ledger is +maintained on a blockchain, the chain without the subledgers directly on it in +full is known as the mainchain.  Where a subledger, like the mainchain, is +maintained by multiple entities, the subledger is called a “sidechain” The +mainchain contains aggregated and summarized data about the sidechains, +and the sidechains can themselves have sidechains. + +Ultimately we want a mainchain that functions like a central bank, with +several hundred peers that function like banks, in that many peers on the +mainchain maintain a sidechain.  Each peer hosts hundreds of thousands of +client wallets. + +When the mainchain runs into scaling limits, transactions between +individuals will be pushed down into sidechains.  A transaction between +sidechains will typically have a very large number of inputs from a very +large number of sidechains, and a very large number of outputs to a very +large number of sidechains.  In such a transaction each sidechain usually +provides only one or two inputs, usually one input, and receives only on or +two outputs, usually one output, that one input and one output representing +the aggregate of many transactions with many other sidechains, and each +transaction between two sidechains representing the aggregate of many +transactions between client wallets. + +But for an input and an output to or from a sidechain to be reasonably +short, rather than proportional to the number of peers on the sidechain, we +are going to have to have linear chain of signatures, + +You make a payment with a client wallet that works like a bill of exchange. +  Your host is a peer on the mainchain.  It sends your bill of +exchange to the other guys host.  What appears on the mainchain is the +root of a Merkle tree of all the bills of exchange, and the settlements +between peers, each such payment being the sum of many bills of exchange, each +such payment representing the Merkle tree of many bills of exchange.  + +The mainchain records that each sidechain has such and such an amount of money, and owes such and such an amount of money to its client wallets, but only knows totals over all the client wallets of a sidechain.  Does not know individual client wallets.  + +The individual client wallet has a chain of hashes leading to the root hash of the mainchain consensus that proves it has the money.  But the lower levels in this chain of hashes do not appear on the mainchain. + +When one client wallet makes a payment to another client wallet, that +payment is final when it is a leaf on the Merkle tree of the consensus hash, +but only the upper nodes of the tree, the aggregate payments between +mainchain peers, appear on the blockchain. + +The lower nodes of the tree are held in the sidechains, and the very lowest nodes of the tree are held in the client wallets. + +When a transaction between sidechains occurs on the mainchain, the root +of the sidechain Merkle tree is placed or referenced on the mainchain.  +But this does not in itself prove that the sidechain transactions that it +summarizes are valid or authorized.  + +If any one sidechain peer in good standing on the sidechain objects to the proposed hash of the sidechain state which is to be incorporated on the mainchain, it can demand that transactions necessary to derive the new hash from values attested to by a recent older hash be lifted from the sidechain to the mainchain, putting up some money for what is in effect a bet that the proposed mainchain transaction cannot be justified by the underlying sidechain transactions.  If the proposed hash of the sidechain state is supported by valid transactions, and the mainchain peers validate the proposed hash, then the peer that insisted on the sidechain data being raised to the mainchain loses its good standing, and has to pay a fee reflecting the cost of pushing all those transactions onto the mainchain. If not, those sidechain peers who signed the proposed, but unsupported, hash lose their good standing and have to pay the costs of pushing all those transactions onto the mainchain.  It should be rare that portions of a sidechain are raised into the mainchain.  Trust, to save bandwidth, storage space, and time, but verify. diff --git a/docs/blockchain_structure_on_disk.md b/docs/blockchain_structure_on_disk.md new file mode 100644 index 0000000..a592810 --- /dev/null +++ b/docs/blockchain_structure_on_disk.md @@ -0,0 +1,69 @@ +--- +title: Block chain structure on disk. +--- + +The question is: One enormous SQLite file, or actually store the chain as a collection of files? + +In the minimum viable product, the blockchain will be quite small, and it will be workable to put it one big SQLite file. +The trouble with one enormous SQLite file is that when it gets big enough, we face a high and steadily increasing risk of one sector on the enormous disk going bad, corrupting the entire database. SQLite does not handle the loss of a single sector gracefully. + +We will eventually need our own database structure designed around +Merkle-patricia trees, append only data structures, and accommodating a near +certainty of sectors and entire disks continually going bad. When one hundred +disks have to be added every year, entire disks will be failing every day or +so, and sectors will be failing every second. + +Eventually, a typical peer will have several big racks of disks. When we +replace the world monetary system, twenty servers each with twenty disks, two +hundred thousand transaction inputs and outputs a second, (for each +transaction minimally involves one input and two outputs, a change output and +a payment output, and usually a lot more. Each signature is sixty four bytes. +Each input and output is at least forty bytes. So, say, on average two inputs +and two outputs per payment – say, perhaps 288 bytes per payment, and we will +want to do one hundred thousand payments per second. So, about nine hundred +terabytes a year. With 2020 disk technology, that is about seventy five twelve +terabyte hard drives per year, costing about one hundred and fifty hard drives +per year costing fifty five thousand dollars per year, to store all the +transactions of the world forever. + +If we are constructing one block per five minutes, each block is about ten +gigabytes. Sqlite3 cannot possibly handle that – the blocks are going to have +to be dispersed over many drives and many physical computers. We are going to +have to go to our own custom low level format, in which a block is distributed +over many drives and many servers, the upper part of the block Merkle-patricia +tree duplicated on every shard, but the the lower branches of the tree each in +a separate shard. Instead of a file structure with many files on one enormous +disk, we have one enormous data structure on servers, each server with many +disks. + +Optimal solution is to store recently accessed data in one big SQLite file, +while also storing the data in a large collection of blocks, once it has become +subject to wide consensus. Older blocks, fully incorporated in the current +consensus, get written to disk in our own custom Merkle-patricia tree format, +with append only Merkle-patricia tree node locations, [a sequential append only +collection of binary trees in postfix tree format]( +merkle_patricia-dac.html#a-sequential-append-only-collection-of-postfix-binary-trees). + +Each file, incorporating a +range of blocks, has its location on disk, time, size, and the roots of its +Merkle-patricia trees recorded in the SQL database. On program launch, the +size, touch time, and root has of newest block in the file are checked. If +there is a discrepancy, we do a full check of the Merkle-patricia tree, editing +it as necessary to an incomplete Merkle-patricia tree, download missing data +from peers, and rebuild the blocks, thus winding up with a newer touch dates. +Our per peer configuration file tells us where to find the block files, and if +they are not stored where expected, we rebuild. If stored where expected, but +touch dates unavailable or incorrect (perhaps because this is the first time the +program launched) then the entire system of Merkle-patricia trees is validated, +making sure the data on disk is consistent. + +How do we tell the one true blockchain, from some other evil blockchain? +Well, the running definition is consensus, that you can interact with other +peers because they agree on the running root hash. So you downloaded this +software from somewhere, and when you downloaded it, you got the means to +contact a bunch of peers, whom we suppose agree, and each have evidence that +other peers agree. And, having downloaded what they agree on, you then treat +it as gospel and as more authoritative that what others say, so long a touch +dates, file sizes, locations, and the hash of the most recent block in the file +are consistent, and the internal contents of each file are consistent with root +of the most recent tree. diff --git a/docs/blockdag_consensus.md b/docs/blockdag_consensus.md new file mode 100644 index 0000000..02f1294 --- /dev/null +++ b/docs/blockdag_consensus.md @@ -0,0 +1,476 @@ +--- +title: Blockdag Consensus +--- + +# Hedera, Bitcoin Proof of Work, and Paxos + +## Paxos + +All consensus algorithms that work are equivalent to Paxos. + +All consensus algorithms that continue to work despite Byzantine Fault +and Brigading are equivalent to Byzantine Fault Tolerant Paxos. + +But Paxos is not in fact an algorithm. It rather is an idea that underlies +actual useful algorithms, and in so far as it is described as algorithm, it is +wrong, for the algorithm as described describes many different things that +you are unlikely to be interested in doing, or even comprehending, and the +algorithm as described is incapable of doing all sorts of things that you are +likely to need done. Even worse it is totally specific to one particular +common use case, which it studiously avoids mentioning, and does not +mention any of the things that you actually need to couple it in to this +specific case, making the description utterly mysterious, because the +writer has all the specific details of this common case in mind, but is carefully avoiding any mention of what he has in mind. These things are +out of scope of the algorithm as given in the interests of maximum +generality, but the algorithm as given is not in fact very general and makes +no sense and is no use without them. + +Despite the studious effort to be as generic as possible by omitting all of +the details required to make it actually do anything useful, the algorithm as +given is the simplest and most minimal example of the concept, +implementing one specific form of Paxos in one specific way, and as +given, will very likely not accomplish you need to do. + +Paxos assumes that each peer knows exactly how many peers there should +be, though some of them may be permanently or temporarily unresponsive +or permanently or temporarily out of contact. + +In Paxos, every peer repeatedly sends messages to every other peer, and +every peer keeps track of those messages, which if you have a lot of peers +adds up to a lot of overhead. + +Hedera assumes that each peer knows exactly how many peers there +should be, *and that each peer eventually gets through*. + +Which is a much stronger assumption than that made by Paxos or Bitcoin. + +In Hedera, each peer's state eventually becomes known to every other +peer, even though it does not necessarily communicate directly with every +other peer, which if you have a whole lot of peers still adds up to a whole +lot of overhead, though not as much as Paxos. It can handle more peers +than Paxos, but if too many peers, still going to bite. + +A blockdag algorithm such as Hedera functions by in effect forking all the +time, and resolving those forks very fast, but if you have almost as many +forks as you have peers, resolving all those forks is still going to require +receiving a great deal of data, processing a great deal of data, and sending +a great deal of data. + +Hedera and Paxos can handle a whole lot of transactions very fast, but +they cannot reach consensus among a very large number of peers in a +reasonable time. + +Bitcoin does not know or care how many peers there are, though it does +know and care roughly how much hashing power there is, but this is +roughly guesstimated over time, over a long time, over a very long time, +over a very very long time. It does not need to know exactly how much +hashing power there is at any one time. + +If there are a very large number of peers, this only slows Bitcoin +consensus time down logarithmically, not linearly, while the amount of +data per round that any one peer has to handle under Hedera is roughly + $\bigcirc\big(N\log(N)\big)$ where N is the number of peers. Bitcoin can handle an +astronomically large number of peers, unlike Hedera and Paxos, because +Bitcoin does not attempt to produce a definitive, known and well defined +consensus. It just provides a plausible guess of the current consensus, and +over time you get exponentially greater certainty about the long past +consensuses. No peer ever knows the current consensus for sure, it just +operates on the recent best guess of its immediate neighbours in the +network of what the recent consensus likely is. If it is wrong, it eventually +finds out. + +## Equivalence of Proof of Work and Paxos + +Bitcoin is of course equivalent to Byzantine Fault Tolerant Paxos, but I +compare it to Paxos because Paxos is difficult to understand, and Byzantine +Fault Tolerant Paxos is nigh incomprehensible. + +In Paxos, before a peer suggests a value to its peers, it must obtain +permission from a majority of peers for that suggestion. And when it seeks +permission from each peer, it learns if a value has already been accepted +by that peer. If so, it has to accept that value, only propose that value in +future, and never propose a different value. Which if everyone always gets +through, means that the first time someone proposes a value, that value, +being the first his peers have seen, will be accepted by someone, if only by +that peer himself. + +Paxos is effect a method for figuring out who was "first", in an +environment where, due to network delays and lost packets, it is difficult +to figure out, or even define, who was first. But if most packets mostly get +through quickly enough, the peer that was first by clock time will usually +get his way. Similarly Bitcoin, the first miner to construct a valid block at +block height $N$ usually winds up defining the consensus for the block at +block height $N$. + +This permission functionality of Paxos is equivalent to the gossip process +in Bitcoin, where a peer learns what the current block height is, and seeks +to add another block, rather than attempt to replace an existing block. + +In Paxos, once one peer accepts one value, it will eventually become the +consensus value, assuming that everyone eventually gets through and that +the usual network problems do not foul things up. Thus Paxos can provide +a definitive result eventually, while Bitcoin's results are never definitive, +merely exponentially probable. + +In Paxos, a peer learns of the definitive and final consensus when it +discovers that a majority of peers have accepted one value. Which if +several values are in play can take a while, but eventually it is going to +happen. In Bitcoin, when the blockchain forks, eventually more hashing +power piles on one branch of the fork than the other, and eventually +everyone can see that more hashing power has piled on one fork than the +other, but there is no moment when a peer discovers than one branch is +definitive and final. It just finds that one branch is becoming more and +more likely, and all the other branches less and less likely. + +Thus paxos has a stronger liveness property than bitcoin, but this +difference is in practice not important, for paxos may take an indefinitely +long time before it can report a definite and final consensus, while Bitcoin +takes a fairly definite time to report it is nearly certain about the consensus +value and that value is unlikely is unlikely to change. + +# Bitcoin does not scale to competing with fiat currency + +Bitcoin is limited to ten transactions per second. Credit card networks +handle about ten thousand transactions per second. + +We will need a crypto coin that enables seven billion people to buy a lollipop. + +Blockdag consensus can achieve sufficient speed. + +There are thirty or more proposed blockdag systems, and the number grows rapidly. + +While blockdags can handle very large numbers of transactions, it is not +obvious to me that any of the existing blockdag algorithms can handle +very large numbers of peers. When actually implemented, they always +wind up privileging a small number of special peers, resulting in hidden +centralization, as somehow these special and privileged peers all seem to +be in the same data centre as the organization operating the blockchain. + +Cardano has a very clever, too clever by half, algorithm to generate +random numbers known to everyone and unpredictable and uncontrollable +by anyone, with which to distribute specialness fairly and uniformly over +time, but this algorithm runs in one centre, rather than using speed of light +delay based fair randomness algorithms, which makes me wonder if it is +distributing specialness fairly, or operating at all. + +I have become inclined to believe that there is no way around making +some peers special, but we need to distribute the specialness fairly and +uniformly, so that every peer get his turn being special at a certain block +height, with the proportion of block heights at which he is special being +proportional to his stake. + +If the number of peers that have a special role in forming the next block is +very small, and the selection and organization of those peers is not +furtively centralized to make sure that only one such group forms, but +rather organized directly those special peers themselves we wind up with +forks sometimes, I hope infrequently, because the special peers should +most of the time successfully self organize into a single group that +contains almost all of the most special peers. If however, we have another, +somewhat larger group of peers that have a special role in deciding which +branch of the fork is the most popular, two phase blockdag, I think we can +preserve blockdag speed without blockdag de-facto concentration of power. + +The algorithm will only have bitcoin liveness, rather than paxos liveness, +which is the liveness most blockdag algorithms seek to achieve. + +I will have to test this empirically, because it is hard to predict, or even to +comprehend, limits on consensus bandwidth. + +## Bitcoin is limited by its consensus bandwidth + +Not by its network bandwidth. + +Bitcoin makes the miners wade through molasses. Very thick molasses. +That is what proof of work is. If there is a fork, it discovers consensus by +noticing which fork has made the most progress through the molasses. + +This takes a while. And if there are more forks, it takes longer. To slow +down the rate of forks, it makes the molasses thicker. If the molasses is +thicker, this slows down fork formation more than it slows down the +resolution of forks. It needs to keep the rate of new blocks down slow +enough that a miner usually discovers the most recent block before it +attempts to add a new block. And if a miner does add a new block at +roughly the same time as another miner adds a new block, quite a few +more blocks have to be added before the fork is resolved. And as the +blocks get bigger, it takes longer for them to circulate. So bigger blocks +need thicker molasses. If forks form faster than they can be resolved, no +consensus. + +## The network bandwidth limit + +The net bandwidth limit on adding transactions is not a problem. + +What bites every blockchain is consensus bandwidth limit, how fast all the +peers can agree on the total order of transactions, when transactions are +coming in fast. + +Suppose a typical transaction consists to two input coins, a change output +coin, and the actual payment. (I use the term coin to refer to transaction +inputs and outputs, although they don’t come in any fixed denominations +except as part of anti tracking measures) + +Each output coin consists of payment amount, suppose around sixty four bits, +and a public key, two hundred and fifty six bits. It also has a script +reference on any special conditions as to what constitutes a valid spend, +which might have a lot of long arguments, but it generally will not, so the +script reference will normally be one byte. + +The input coins can be a hash reference to a coin in the consensus +blockchain, two fifty six bits, or they can be a reference by total order +within the blockchain, sixty four bits. + +We can use a Schnorr group signature, which is five hundred and twelve +bits no matter how many coins are being signed, no matter how many +people are signing, and no matter if it is an n of m signature. + +So a typical transaction, assuming we have a good compact representation +of transactions, should be around 1680 bits, maybe less. + +At scale you inevitably have a large number of clients and a small number +of full peers. Say several hundred peers, a few billion clients, most of them +lightning gateways. So we can assume every peer has a good connection. + +A typical, moderately good, home connection is thirty Mbps download but +its upload connection is only ten Mbs or so. + +So if our peers are typical decent home connections, and they will be a lot +better than that, bandwidth limits them to adding transactions at 10Mbps, +six thousand transactions per second, Visa card magnitude. Though if such +a large number of transactions are coming in so fast, blockchain storage +requirements will be very large, around 24 TiB, about three or four +standard home desktop system disk drives. But by the time we get to that +scale all peers will be expensive dedicated systems, rather than a +background process using its owners spare storage and spare bandwidth, +running on the same desktop that its owner uses to +shop at Amazon. + +Which if everyone in the world is buying their lollipops on the blockchain +will still need most people using the lightning network layer, rather than +the blockchain layer, but everyone will still routinely access the blockchain +layer directly, thus ensuring that problems with their lightning +gateways are resolved by a peer they can choose, rather than resolved by +their lightning network wallet provider, thus ensuring that we can have a +truly decentralized lightning network. + +We will not necessarily *get* a truly decentralized lightning layer, but a base +layer capable of handling a lot of transactions makes it physically possible. + +So if bandwidth is not a problem, why is bitcoin so slow? + +The bottleneck in bitcoin is that to avoid too many forks, which waste time +with fork resolution, you need a fair bit of consensus on the previous block +before you form the next block. + +And bitcoin consensus is slow, because the way a fork is resolved is that +blocks that received one branch fork first continue to work on that branch, +while blocks that received the other branch first continue to work on that +branch, until one branch gets ahead of the other branch, whereupon the +leading branch spreads rapidly through the peers. With proof of stake, that +is not going work, one can lengthen a branch as fast as you please. Instead, +each branch has to be accompanied by evidence of the weight of stake of +peers on that branch. Which means the winning branch can start spreading +immediately. + +# Blockdag to the rescue + +On a blockdag, you don’t need a fair bit of consensus on the previous +block to avoid too many forks forming. Every peer is continually forming +his own fork, and these forks reach consensus about their left great grand +child, or left great great … great grandchild. The blocks that eventually +become the consensus as leftmost blocks form a blockchain. So we can +roll right ahead, and groups of blocks that deviate from the consensus, +which is all of them but one, eventually get included, but later in the total +order than they initially thought they were. + +In a blockdag, each block has several children, instead of just one. Total +order starting from any one block is depth first search. The left blocks +come before the right blocks, and the child blocks come before the parent +block. Each block may be referenced by several different parent blocks, but +only the first reference in the total order matters. + +Each leftmost block defines the total order of all previous blocks, the +total order being the dag in depth first order. + +Each peer disagrees with all the other peers about the total order of recent +blocks and recent transactions, each is its own fork, but they all agree +about the total order of older blocks and older transactions. + +## previous work + +[There are umpteen proposals for blockdags](./SoK_Diving_into_DAG-based_Blockchain_Systems) most of them garbage, but the general principle is sound. + +For a bunch of algorithms that plausibly claim to approach the upload +limit, see: + +* [Scalable and probabilistic leaderless bft consensus through metastability](https://files.avalabs.org/papers/consensus.pdf) + + This explains the underlying concept, that a peer looks at the dag, + make its best guess as to which way consensus is going, and joins + the seeming consensus, which make it more likely to become the + actual consensus. + + Which is a good way of making arbitrary choices where it does not + matter which choice everyone makes, provided that they all make + the same choice, even though it is an utterly disastrous way of + making choices where the choice matters. + + This uses an algorithm that rewards fast mixing peers by making + their blocks appear earlier in the total order. This algorithm does + not look incentive compatible to me. It looks to me that if all the + peers are using that algorithm, then any one peer has an incentive + to use a slightly different algorithm. + + The authors use the term Byzantine fault incorrectly, referring to + behavior that suggests the unpredictable failures of an unreliable + data network as Byzantine failure. No, a Byzantine fault suggests + Byzantine defection, treachery, and failure to follow process. It is + named after Byzantium because of the stuff that happened during + the decline of the Byzantine empire. + +* [Prism: Deconstructing the blockchain to approach physical limits](https://arxiv.org/pdf/1810.08092.pdf) + + A messy, unclear, and overly complicated proposed implementation + of the blockdag algorithm, which, however, makes the important + point that it can go mighty fast, that the physical limits on + consensus are bandwidth, storage, and communication delay, and + that we can approach these limits. + +* [Blockmania: from block dags to consensus](https://arxiv.org/pdf/1809.01620.pdf) + + This brings the important concept, that the tree structure created by + gossiping the blockdag around _is_ the blockdag, and also is the data + you need to create consensus, bringing together things that were + separate in Prism, radically simplifying what is complicated in + Prism by uniting data and functionality that Prism divided. + + This study shows that the Blockmania implementation of the + blockdag is equivalent to the Practical Byzantine Fault Tolerant + consensus algorithm, only a great deal faster, more efficient, and + considerably easier to understand. + + The Practical Byzantine Fault Tolerant consensus algorithm is an + implementation of the Paxos protocol in the presence of Byzantine + faults, and the Paxos protocol is already hard enough to understand. + + So anyone who wants to implement consensus in a system where + Byzantine failure and Byzantine defection is possible should forget + about Paxos, and study blockdags. + +* [A highly scalable, decentralized dag–based consensus algorithm](https://eprint.iacr.org/2018/1112.pdf) + + Another blockdag algorithm, but one whose performance has been tested. Can handle high bandwidth, lots of transactions, and achieves fast Byzantine fault resistant total order consensus in time $O(6λ)$, where λ is the upper bound of the network’s gossip period. + +* [Blockchai–free cryptocurrencies: A framework for truly decentralised fast transactions](https://eprint.iacr.org/2016/871.pdf) + + These transactions are indeed truly decentralized, fast, and free from + blocks, assuming all participants download the entire set of + transactions all the time. + + The problem with this algorithm is that when the blockchain grows enormous, most participants will become clients, and only a few giant peers will keep the whole transaction set, and this system, because it does not provide a total order of all transactions, will then place all the power in the hands of the peers. + + We would like the clients to have control of their private + keys, thus must publish their public keys with the money they + spend, in which case the giant peers must exchange blocks of + information containing those keys, and it is back to having blocks. + + The defect of this proposal is that convergence does not + converge to a total order on all past transactions, but merely a total + set of all past transactions. Since the graph is a graph of + transactions, not blocks, double spends are simply excluded, so a + total order is not needed. While you can get by with a total set, a + total order enables you to do many things a total set does not let + you do. Such as publish two conflicting transactions and resolve them. + + Total order can represent consensus decisions that total set cannot + easily represent, perhaps cannot represent at all. We need a + blockdag algorithm that gives us consensus on the total order of + blocks, not just the set of blocks. + + In a total order, you do not just converge to the same set, you + converge to the same order of the set. Having the same total order + of the set makes makes it, among other things, a great deal easier + and faster to check that you have the same set. Plus your set can + contain double spends, which you are going to need if the clients + themselves can commit transactions through the peers, if the clients + themselves hold the secret keys and do not need to trust the peers. + +# Proposed blockdag implementation + +The specific details of many of these proposed systems are rather silly and +often vague, typical academic exercises unconcerned with real world +issues, but the general idea that the academics intend to illustrate is sound +and should work, certainly can be made to work. They need to be +understood as academic illustrations of the idea of the general algorithm +for fast and massive blockdag consensus, and not necessarily intended as +ready to roll implementations of that idea. + +Here is an even more vague outline of my variant of this idea, I name +Yabca “Yet another blockdag consensus algorithm”, + +I propose proof of stake. The stake of a peer is not the stake it owns, but +the stake that it has injected into the blockchain on behalf of its clients +and that its clients have not spent yet. Each peer pays on behalf of its +clients for the amount of space it takes up on the blockchain, though it does +not pay in each block. It makes an advance payment that will cover many +transactions in many blocks. The money disappears, built in deflation, +instead of built in inflation. Each block is a record of what a peer has +injected + +The system does not pay the peers for generating a total order of +transactions. Clients pay peers for injecting transactions. We want the +power to be in the hands of people who own the money, thus governance will +have a built in bias towards appreciation and deflation, rather than +inflation. + +The special sauce that makes each proposed blockdag different from each +of the others is how each peer decides what consensus is forming about +the leftmost edge of the dag, the graph analysis that each peer performs. +And this, my special sauce, I will explain when I have something running. + +Each peer adopts as its leftmost child for its latest block, a previous block +that looks like a good candidate for consensus, which looks like a good +candidate for consensus because the left child has a left child that looks +like consensus actually is forming around that grandchild , in part because +the left child has a … left child has a … left child that looks like it might +have consensus, until eventually, as new blocks pile on top of old blocks, we +actually do get consensus about the left most child sufficiently deep in +the dag from the latest blocks. + +The blockdag can run fast because all the forks that are continually +forming eventually get stuffed into the consensus total order somewhere. +So we don’t have to impose a speed limit to prevent excessive forking. + +# Cost of storage on the blockchain. + +Tardigrade charges $120 per year for per terabyte of storage, $45 per terabyte of download + +We have a pile of theory, though no practical experience, that a blockdag can approach the physical limits, that its limits are going to be bandwidth and storage.. + +Storage on the blockdag is going to cost more, because massively +replicated, so say three hundred times as much, and is going to be +optimized for tiny fragments of data while Tardigrade is optimized for +enormous blocks of data, so say three times as much on top of that, a +thousand times as expensive to store should be in the right ballpark. + +When you download, you are downloading from only a single peer on the blockdag, but you are downloading tiny fragments dispersed over a large pile of data, so again, a thousand times as expensive to download sounds like it might be in the right ballpark. + + Then storing a chain of keys and the accompanying roots of total state, + with one new key per day for ten years will cost about two dollars over ten + years. + +Ten megabytes is a pretty big pile of human readable documentation. Let +us suppose you want to store ten megabytes of human readable data and +read and write access costs a thousand times what tardigrade costs, will +cost about twelve dollars. + +So, we should consider the blockdag as an immutable store of arbitrary +typed data, a reliable broadcast channel, where some types are executable, +and, when executed, cause a change in mutable total state, typically that +a new unspent coin record is added, and an old unspent coin record is +deleted. + +In another use, a valid update to a chain of signatures should cause a +change in the signature associated with a name, the association being +mutable state controlled by immutable data. Thus we can implement +corporations on the blockdag by a chain of signatures, each of which +represents [an n of m multisig](./PracticalLargeScaleDistributedKeyGeneration.pdf “Practical Large Scale Distributed Key Generation”). diff --git a/docs/blockseer.jpg b/docs/blockseer.jpg new file mode 100644 index 0000000..c23a494 Binary files /dev/null and b/docs/blockseer.jpg differ diff --git a/docs/byzantine_paxos.pdf b/docs/byzantine_paxos.pdf new file mode 100644 index 0000000..5b21a43 --- /dev/null +++ b/docs/byzantine_paxos.pdf @@ -0,0 +1,46117 @@ +%PDF-1.1 +3 0 obj +<< /Type /Font /Name /R3 /Subtype /Type1 /BaseFont /Times-Roman >> +endobj +4 0 obj +<< /Length 5 0 R >> +stream +q +BT +/R3 8 Tf +1 0 0 1 84.12 740.76 Tm +(Appears) Tj +1 0 0 1 112.68 740.76 Tm +(in) Tj +1 0 0 1 120.84 740.76 Tm +(the) Tj +ET +endstream +endobj +5 0 obj +121 +endobj +6 0 obj +<< /Type /Font /Name /R6 /Subtype /Type1 /BaseFont /Times-Italic >> +endobj +7 0 obj +<< /Length 8 0 R >> +stream +BT +/R6 8 Tf +1 0 0 1 132.6 740.76 Tm +(Pr) Tj +1 0 0 1 140.28 740.76 Tm +(oceedings) Tj +1 0 0 1 174.12 740.76 Tm +(of) Tj +1 0 0 1 182.28 740.76 Tm +(the) Tj +1 0 0 1 194.039 740.76 Tm +(Thir) Tj +1 0 0 1 207.479 740.76 Tm +(d) Tj +1 0 0 1 213.479 740.76 Tm +(Symposium) Tj +1 0 0 1 251.759 740.76 Tm +(on) Tj +1 0 0 1 261.719 740.76 Tm +(Oper) Tj +1 0 0 1 277.919 740.76 Tm +(ating) Tj +1 0 0 1 296.399 740.76 Tm +(Systems) Tj +1 0 0 1 323.519 740.76 Tm +(Design) Tj +1 0 0 1 348.119 740.76 Tm +(and) Tj +1 0 0 1 362.039 740.76 Tm +(Implementation,) Tj +1 0 0 1 416.159 740.76 Tm +(Ne) Tj +1 0 0 1 424.799 740.76 Tm +(w) Tj +1 0 0 1 432.119 740.76 Tm +(Orleans,) Tj +1 0 0 1 461.758 740.76 Tm +(USA,) Tj +1 0 0 1 480.358 740.76 Tm +(F) Tj +1 0 0 1 484.678 740.76 Tm +(ebruary) Tj +1 0 0 1 511.918 740.76 Tm +(1999) Tj +ET +endstream +endobj +8 0 obj +781 +endobj +9 0 obj +<< /Type /Font /Name /R9 /Subtype /Type1 /BaseFont /Times-Bold >> +endobj +10 0 obj +<< /Length 11 0 R >> +stream +BT +/R9 14 Tf +1 0 0 1 199.32 702.96 Tm +(Practical) Tj +1 0 0 1 256.2 702.96 Tm +(Byzantine) Tj +1 0 0 1 319.44 702.96 Tm +(F) Tj +1 0 0 1 327.6 702.96 Tm +(ault) Tj +1 0 0 1 354.36 702.96 Tm +(T) Tj +1 0 0 1 362.4 702.96 Tm +(olerance) Tj +/R3 12 Tf +1 0 0 1 223.08 669.12 Tm +(Miguel) Tj +1 0 0 1 260.64 669.12 Tm +(Castro) Tj +1 0 0 1 294.84 669.12 Tm +(and) Tj +1 0 0 1 315.12 669.12 Tm +(Barbara) Tj +1 0 0 1 355.919 669.12 Tm +(Lisk) Tj +1 0 0 1 377.159 669.12 Tm +(o) Tj +1 0 0 1 383.039 669.12 Tm +(v) Tj +/R6 12 Tf +1 0 0 1 223.8 655.2 Tm +(Labor) Tj +1 0 0 1 252.96 655.2 Tm +(atory) Tj +1 0 0 1 281.04 655.2 Tm +(for) Tj +1 0 0 1 297.96 655.2 Tm +(Computer) Tj +1 0 0 1 348.839 655.2 Tm +(Science) Tj +1 0 0 1 385.199 655.2 Tm +(,) Tj +1 0 0 1 214.08 641.16 Tm +(Massac) Tj +1 0 0 1 250.56 641.16 Tm +(husetts) Tj +1 0 0 1 286.68 641.16 Tm +(Institute) Tj +1 0 0 1 328.799 641.16 Tm +(of) Tj +1 0 0 1 341.159 641.16 Tm +(T) Tj +1 0 0 1 346.679 641.16 Tm +(ec) Tj +1 0 0 1 357.119 641.16 Tm +(hnolo) Tj +1 0 0 1 384.359 641.16 Tm +(gy) Tj +1 0 0 1 395.039 641.16 Tm +(,) Tj +1 0 0 1 192.36 627.24 Tm +(545) Tj +1 0 0 1 213.24 627.24 Tm +(T) Tj +1 0 0 1 218.76 627.24 Tm +(ec) Tj +1 0 0 1 229.2 627.24 Tm +(hnolo) Tj +1 0 0 1 256.44 627.24 Tm +(gy) Tj +1 0 0 1 270.72 627.24 Tm +(Squar) Tj +1 0 0 1 298.92 627.24 Tm +(e) Tj +1 0 0 1 304.079 627.24 Tm +(,) Tj +1 0 0 1 309.959 627.24 Tm +(Cambridg) Tj +1 0 0 1 358.439 627.24 Tm +(e) Tj +1 0 0 1 363.599 627.24 Tm +(,) Tj +1 0 0 1 369.479 627.24 Tm +(MA) Tj +1 0 0 1 389.759 627.24 Tm +(02139) Tj +ET +Q +q +W +0 0 612 792 re +n +endstream +endobj +11 0 obj +1497 +endobj +12 0 obj +<< /Type /XObject /Name /R12 /Subtype /Image /Length 13 0 R +/ImageMask true /Width 39 /Height 100 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 39 /BlackIs1 true >>] +>> +stream ++CYskJO*-9kn*)Wn:LN: +s8W-!s8W-!s4?PIKnQ+Ne;bBe<0A()%X^A[(i\@1^OQ:`s8W-!s8W,ur&u1hf`^Jj+bNVe!tYGt"9~> +endstream +endobj +13 0 obj +101 +endobj +14 0 obj +<< /Length 15 0 R >> +stream +q +3.9 0 0 -10 226.8 620.8 cm +/R12 Do +Q +endstream +endobj +15 0 obj +39 +endobj +16 0 obj +<< /Type /Font /Name /R16 /Subtype /Type1 /BaseFont /Courier >> +endobj +17 0 obj +<< /Length 18 0 R >> +stream +BT +/R16 10 Tf +1 0 0 1 231.36 613.32 Tm +(castro,liskov) Tj +ET +endstream +endobj +18 0 obj +61 +endobj +19 0 obj +<< /Type /XObject /Name /R19 /Subtype /Image /Length 20 0 R +/ImageMask true /Width 39 /Height 100 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 39 /BlackIs1 true >>] +>> +stream +,DhM)&Fel!i:(XA +\*s@ss8W-!s8W-!hhT+`nDT#4)$`#/'GV2X"UH\]_@,]0s8W-!s8W-!s'MeW#sg*k*#HKf&-r8$"9~> +endstream +endobj +20 0 obj +96 +endobj +21 0 obj +<< /Length 22 0 R >> +stream +q +3.9 0 0 -10 309.9 620.8 cm +/R19 Do +Q +BT +1 0 0 1 314.4 613.32 Tm +(@lcs.mit.edu) Tj +/R9 12 Tf +1 0 0 1 72 568.68 Tm +(Abstract) Tj +/R3 9 Tf +1 0 0 1 72 552.6 Tm +(This) Tj +1 0 0 1 91.1999 552.6 Tm +(paper) Tj +1 0 0 1 114.36 552.6 Tm +(describes) Tj +1 0 0 1 151.08 552.6 Tm +(a) Tj +1 0 0 1 158.28 552.6 Tm +(ne) Tj +1 0 0 1 166.44 552.6 Tm +(w) Tj +1 0 0 1 176.28 552.6 Tm +(replication) Tj +1 0 0 1 217.919 552.6 Tm +(algorithm) Tj +1 0 0 1 256.079 552.6 Tm +(that) Tj +1 0 0 1 272.759 552.6 Tm +(is) Tj +1 0 0 1 282.119 552.6 Tm +(able) Tj +1 0 0 1 72 541.68 Tm +(to) Tj +1 0 0 1 84.12 541.68 Tm +(tolerate) Tj +1 0 0 1 116.16 541.68 Tm +(Byzantine) Tj +1 0 0 1 157.56 541.68 Tm +(f) Tj +1 0 0 1 160.44 541.68 Tm +(aults.) Tj +1 0 0 1 191.52 541.68 Tm +(W) Tj +1 0 0 1 199.32 541.68 Tm +(e) Tj +1 0 0 1 208.44 541.68 Tm +(belie) Tj +1 0 0 1 225.6 541.68 Tm +(v) Tj +1 0 0 1 229.92 541.68 Tm +(e) Tj +1 0 0 1 239.039 541.68 Tm +(that) Tj +1 0 0 1 257.639 541.68 Tm +(Byzantine-) Tj +1 0 0 1 72 530.64 Tm +(f) Tj +1 0 0 1 74.88 530.64 Tm +(ault-tolerant) Tj +1 0 0 1 124.08 530.64 Tm +(algorithms) Tj +1 0 0 1 167.64 530.64 Tm +(will) Tj +1 0 0 1 186.96 530.64 Tm +(be) Tj +1 0 0 1 200.76 530.64 Tm +(increasingly) Tj +1 0 0 1 249.839 530.64 Tm +(important) Tj +1 0 0 1 290.039 530.64 Tm +(in) Tj +1 0 0 1 72 519.72 Tm +(the) Tj +1 0 0 1 86.5199 519.72 Tm +(future) Tj +1 0 0 1 111.6 519.72 Tm +(because) Tj +1 0 0 1 143.64 519.72 Tm +(malicious) Tj +1 0 0 1 182.04 519.72 Tm +(attacks) Tj +1 0 0 1 210.6 519.72 Tm +(and) Tj +1 0 0 1 227.159 519.72 Tm +(softw) Tj +1 0 0 1 246.959 519.72 Tm +(are) Tj +1 0 0 1 261.479 519.72 Tm +(errors) Tj +1 0 0 1 286.079 519.72 Tm +(are) Tj +1 0 0 1 72 508.8 Tm +(increasingly) Tj +1 0 0 1 119.76 508.8 Tm +(common) Tj +1 0 0 1 154.92 508.8 Tm +(and) Tj +1 0 0 1 171.72 508.8 Tm +(can) Tj +1 0 0 1 188.04 508.8 Tm +(cause) Tj +1 0 0 1 211.919 508.8 Tm +(f) Tj +1 0 0 1 214.799 508.8 Tm +(aulty) Tj +1 0 0 1 236.639 508.8 Tm +(nodes) Tj +1 0 0 1 261.359 508.8 Tm +(to) Tj +1 0 0 1 272.279 508.8 Tm +(e) Tj +1 0 0 1 276.119 508.8 Tm +(xhibit) Tj +1 0 0 1 72 497.76 Tm +(arbitrary) Tj +1 0 0 1 106.56 497.76 Tm +(beha) Tj +1 0 0 1 123.24 497.76 Tm +(vior) Tj +1 0 0 1 137.16 497.76 Tm +(.) Tj +1 0 0 1 147 497.76 Tm +(Whereas) Tj +1 0 0 1 182.04 497.76 Tm +(pre) Tj +1 0 0 1 193.2 497.76 Tm +(vious) Tj +1 0 0 1 216.479 497.76 Tm +(algorithms) Tj +1 0 0 1 258.479 497.76 Tm +(assumed) Tj +1 0 0 1 293.039 497.76 Tm +(a) Tj +1 0 0 1 72 486.84 Tm +(synchronous) Tj +1 0 0 1 120.6 486.84 Tm +(system) Tj +1 0 0 1 148.68 486.84 Tm +(or) Tj +1 0 0 1 159.36 486.84 Tm +(were) Tj +1 0 0 1 180 486.84 Tm +(too) Tj +1 0 0 1 194.64 486.84 Tm +(slo) Tj +1 0 0 1 204.84 486.84 Tm +(w) Tj +1 0 0 1 214.679 486.84 Tm +(to) Tj +1 0 0 1 224.879 486.84 Tm +(be) Tj +1 0 0 1 236.519 486.84 Tm +(used) Tj +1 0 0 1 256.199 486.84 Tm +(in) Tj +1 0 0 1 266.399 486.84 Tm +(practice,) Tj +1 0 0 1 72 475.92 Tm +(the) Tj +1 0 0 1 85.9199 475.92 Tm +(algorithm) Tj +1 0 0 1 123.84 475.92 Tm +(described) Tj +1 0 0 1 161.16 475.92 Tm +(in) Tj +1 0 0 1 171.24 475.92 Tm +(this) Tj +1 0 0 1 187.2 475.92 Tm +(paper) Tj +1 0 0 1 210.119 475.92 Tm +(is) Tj +1 0 0 1 219.119 475.92 Tm +(practical:) Tj +1 0 0 1 257.159 475.92 Tm +(it) Tj +1 0 0 1 265.199 475.92 Tm +(w) Tj +1 0 0 1 271.559 475.92 Tm +(orks) Tj +1 0 0 1 290.039 475.92 Tm +(in) Tj +1 0 0 1 72 464.88 Tm +(asynchronous) Tj +1 0 0 1 124.32 464.88 Tm +(en) Tj +1 0 0 1 132.36 464.88 Tm +(vironments) Tj +1 0 0 1 175.68 464.88 Tm +(lik) Tj +1 0 0 1 185.04 464.88 Tm +(e) Tj +1 0 0 1 192 464.88 Tm +(the) Tj +1 0 0 1 205.92 464.88 Tm +(Internet) Tj +1 0 0 1 236.759 464.88 Tm +(and) Tj +1 0 0 1 252.719 464.88 Tm +(incorporates) Tj +1 0 0 1 72 453.96 Tm +(se) Tj +1 0 0 1 79.2 453.96 Tm +(v) Tj +1 0 0 1 83.5199 453.96 Tm +(eral) Tj +1 0 0 1 99.1198 453.96 Tm +(important) Tj +1 0 0 1 136.2 453.96 Tm +(optimizations) Tj +1 0 0 1 187.08 453.96 Tm +(that) Tj +1 0 0 1 202.68 453.96 Tm +(impro) Tj +1 0 0 1 223.919 453.96 Tm +(v) Tj +1 0 0 1 228.239 453.96 Tm +(e) Tj +1 0 0 1 234.479 453.96 Tm +(the) Tj +1 0 0 1 247.559 453.96 Tm +(response) Tj +1 0 0 1 281.039 453.96 Tm +(time) Tj +1 0 0 1 72 443.04 Tm +(of) Tj +1 0 0 1 81.2399 443.04 Tm +(pre) Tj +1 0 0 1 92.3999 443.04 Tm +(vious) Tj +1 0 0 1 113.64 443.04 Tm +(algorithms) Tj +1 0 0 1 153.84 443.04 Tm +(by) Tj +1 0 0 1 164.52 443.04 Tm +(more) Tj +1 0 0 1 184.68 443.04 Tm +(than) Tj +1 0 0 1 201.959 443.04 Tm +(an) Tj +1 0 0 1 212.159 443.04 Tm +(order) Tj +1 0 0 1 232.799 443.04 Tm +(of) Tj +1 0 0 1 242.039 443.04 Tm +(magnitude.) Tj +1 0 0 1 285.239 443.04 Tm +(W) Tj +1 0 0 1 293.039 443.04 Tm +(e) Tj +1 0 0 1 72 432 Tm +(implemented) Tj +1 0 0 1 121.32 432 Tm +(a) Tj +1 0 0 1 127.8 432 Tm +(Byzantine-f) Tj +1 0 0 1 169.92 432 Tm +(ault-tolerant) Tj +1 0 0 1 216.359 432 Tm +(NFS) Tj +1 0 0 1 235.199 432 Tm +(service) Tj +1 0 0 1 263.159 432 Tm +(using) Tj +1 0 0 1 284.999 432 Tm +(our) Tj +1 0 0 1 72 421.08 Tm +(algorithm) Tj +1 0 0 1 108.84 421.08 Tm +(and) Tj +1 0 0 1 123.84 421.08 Tm +(measured) Tj +1 0 0 1 160.2 421.08 Tm +(its) Tj +1 0 0 1 170.76 421.08 Tm +(performance.) Tj +1 0 0 1 221.399 421.08 Tm +(The) Tj +1 0 0 1 237.359 421.08 Tm +(results) Tj +1 0 0 1 262.799 421.08 Tm +(sho) Tj +1 0 0 1 274.919 421.08 Tm +(w) Tj +1 0 0 1 283.559 421.08 Tm +(that) Tj +1 0 0 1 72 410.16 Tm +(our) Tj +1 0 0 1 85.6799 410.16 Tm +(service) Tj +1 0 0 1 112.8 410.16 Tm +(is) Tj +1 0 0 1 120.48 410.16 Tm +(only) Tj +1 0 0 1 138.24 410.16 Tm +(3%) Tj +1 0 0 1 151.92 410.16 Tm +(slo) Tj +1 0 0 1 162.12 410.16 Tm +(wer) Tj +1 0 0 1 177.24 410.16 Tm +(than) Tj +1 0 0 1 194.399 410.16 Tm +(a) Tj +1 0 0 1 200.159 410.16 Tm +(standard) Tj +1 0 0 1 232.319 410.16 Tm +(unreplicated) Tj +1 0 0 1 278.279 410.16 Tm +(NFS.) Tj +/R9 12 Tf +1 0 0 1 72 383.04 Tm +(1) Tj +1 0 0 1 89.8799 383.04 Tm +(Intr) Tj +1 0 0 1 110.16 383.04 Tm +(oduction) Tj +/R3 10 Tf +1 0 0 1 72 365.88 Tm +(Malicious) Tj +1 0 0 1 116.04 365.88 Tm +(attacks) Tj +1 0 0 1 147.84 365.88 Tm +(and) Tj +1 0 0 1 166.32 365.88 Tm +(softw) Tj +1 0 0 1 188.4 365.88 Tm +(are) Tj +1 0 0 1 204.6 365.88 Tm +(errors) Tj +1 0 0 1 231.96 365.88 Tm +(are) Tj +1 0 0 1 248.279 365.88 Tm +(increasingly) Tj +1 0 0 1 72 354 Tm +(common.) Tj +1 0 0 1 118.2 354 Tm +(The) Tj +1 0 0 1 137.88 354 Tm +(gro) Tj +1 0 0 1 151.08 354 Tm +(wing) Tj +1 0 0 1 175.08 354 Tm +(reliance) Tj +1 0 0 1 210.84 354 Tm +(of) Tj +1 0 0 1 223.44 354 Tm +(industry) Tj +1 0 0 1 260.28 354 Tm +(and) Tj +1 0 0 1 278.879 354 Tm +(go) Tj +1 0 0 1 288.839 354 Tm +(v-) Tj +1 0 0 1 72 342 Tm +(ernment) Tj +1 0 0 1 107.04 342 Tm +(on) Tj +1 0 0 1 119.4 342 Tm +(online) Tj +1 0 0 1 146.64 342 Tm +(information) Tj +1 0 0 1 196.08 342 Tm +(services) Tj +1 0 0 1 230.52 342 Tm +(mak) Tj +1 0 0 1 247.679 342 Tm +(es) Tj +1 0 0 1 258.239 342 Tm +(malicious) Tj +1 0 0 1 72 330.12 Tm +(attacks) Tj +1 0 0 1 103.32 330.12 Tm +(more) Tj +1 0 0 1 127.32 330.12 Tm +(attracti) Tj +1 0 0 1 154.8 330.12 Tm +(v) Tj +1 0 0 1 159.72 330.12 Tm +(e) Tj +1 0 0 1 167.64 330.12 Tm +(and) Tj +1 0 0 1 185.64 330.12 Tm +(mak) Tj +1 0 0 1 202.8 330.12 Tm +(es) Tj +1 0 0 1 214.56 330.12 Tm +(the) Tj +1 0 0 1 230.279 330.12 Tm +(consequences) Tj +1 0 0 1 288.719 330.12 Tm +(of) Tj +1 0 0 1 72 318.12 Tm +(successful) Tj +1 0 0 1 115.56 318.12 Tm +(attacks) Tj +1 0 0 1 145.68 318.12 Tm +(more) Tj +1 0 0 1 168.72 318.12 Tm +(serious.) Tj +1 0 0 1 203.28 318.12 Tm +(In) Tj +1 0 0 1 214.079 318.12 Tm +(addition,) Tj +1 0 0 1 251.879 318.12 Tm +(the) Tj +1 0 0 1 266.519 318.12 Tm +(number) Tj +1 0 0 1 72 306.24 Tm +(of) Tj +1 0 0 1 82.92 306.24 Tm +(softw) Tj +1 0 0 1 105 306.24 Tm +(are) Tj +1 0 0 1 119.76 306.24 Tm +(errors) Tj +1 0 0 1 145.56 306.24 Tm +(is) Tj +1 0 0 1 154.8 306.24 Tm +(increasing) Tj +1 0 0 1 198.36 306.24 Tm +(due) Tj +1 0 0 1 215.4 306.24 Tm +(to) Tj +1 0 0 1 225.719 306.24 Tm +(the) Tj +1 0 0 1 240.599 306.24 Tm +(gro) Tj +1 0 0 1 253.799 306.24 Tm +(wth) Tj +1 0 0 1 271.199 306.24 Tm +(in) Tj +1 0 0 1 281.519 306.24 Tm +(size) Tj +1 0 0 1 72 294.24 Tm +(and) Tj +1 0 0 1 88.8 294.24 Tm +(comple) Tj +1 0 0 1 118.2 294.24 Tm +(xity) Tj +1 0 0 1 135.84 294.24 Tm +(of) Tj +1 0 0 1 146.64 294.24 Tm +(softw) Tj +1 0 0 1 168.72 294.24 Tm +(are.) Tj +1 0 0 1 186.84 294.24 Tm +(Since) Tj +1 0 0 1 211.44 294.24 Tm +(malicious) Tj +1 0 0 1 252.599 294.24 Tm +(attacks) Tj +1 0 0 1 282.599 294.24 Tm +(and) Tj +1 0 0 1 72 282.36 Tm +(softw) Tj +1 0 0 1 94.0799 282.36 Tm +(are) Tj +1 0 0 1 108.84 282.36 Tm +(errors) Tj +1 0 0 1 134.76 282.36 Tm +(can) Tj +1 0 0 1 151.2 282.36 Tm +(cause) Tj +1 0 0 1 175.92 282.36 Tm +(f) Tj +1 0 0 1 179.16 282.36 Tm +(aulty) Tj +1 0 0 1 201.72 282.36 Tm +(nodes) Tj +1 0 0 1 227.64 282.36 Tm +(to) Tj +1 0 0 1 238.079 282.36 Tm +(e) Tj +1 0 0 1 242.399 282.36 Tm +(xhibit) Tj +1 0 0 1 268.199 282.36 Tm +(Byzan-) Tj +1 0 0 1 72 270.36 Tm +(tine) Tj +1 0 0 1 89.2799 270.36 Tm +(\(i.e.,) Tj +1 0 0 1 109.56 270.36 Tm +(arbitrary\)) Tj +1 0 0 1 149.52 270.36 Tm +(beha) Tj +1 0 0 1 168.24 270.36 Tm +(vior) Tj +1 0 0 1 184.08 270.36 Tm +(,) Tj +1 0 0 1 188.52 270.36 Tm +(Byzantine-f) Tj +1 0 0 1 235.679 270.36 Tm +(ault-tolerant) Tj +1 0 0 1 286.439 270.36 Tm +(al-) Tj +1 0 0 1 72 258.48 Tm +(gorithms) Tj +1 0 0 1 109.92 258.48 Tm +(are) Tj +1 0 0 1 124.56 258.48 Tm +(increasingly) Tj +1 0 0 1 175.8 258.48 Tm +(important.) Tj +1 0 0 1 81.96 246.12 Tm +(This) Tj +1 0 0 1 104.4 246.12 Tm +(paper) Tj +1 0 0 1 131.28 246.12 Tm +(presents) Tj +1 0 0 1 168.72 246.12 Tm +(a) Tj +1 0 0 1 177.84 246.12 Tm +(ne) Tj +1 0 0 1 187.08 246.12 Tm +(w) Tj +1 0 0 1 193.68 246.12 Tm +(,) Tj +/R6 10 Tf +1 0 0 1 201.36 246.12 Tm +(pr) Tj +1 0 0 1 210.12 246.12 Tm +(actical) Tj +/R3 10 Tf +1 0 0 1 241.92 246.12 Tm +(algorithm) Tj +1 0 0 1 285.359 246.12 Tm +(for) Tj +1 0 0 1 72 234.24 Tm +(state) Tj +1 0 0 1 92.0399 234.24 Tm +(machine) Tj +1 0 0 1 127.56 234.24 Tm +(replication) Tj +1 0 0 1 171.84 234.24 Tm +([17) Tj +1 0 0 1 185.16 234.24 Tm +(,) Tj +1 0 0 1 189.36 234.24 Tm +(34) Tj +1 0 0 1 199.32 234.24 Tm +(]) Tj +1 0 0 1 204.36 234.24 Tm +(that) Tj +1 0 0 1 221.159 234.24 Tm +(tolerates) Tj +1 0 0 1 256.559 234.24 Tm +(Byzantine) Tj +1 0 0 1 72 222.24 Tm +(f) Tj +1 0 0 1 75.24 222.24 Tm +(aults.) Tj +1 0 0 1 105.84 222.24 Tm +(The) Tj +1 0 0 1 125.76 222.24 Tm +(algorithm) Tj +1 0 0 1 168.84 222.24 Tm +(of) Tj +1 0 0 1 177 222.24 Tm +(fers) Tj +1 0 0 1 196.2 222.24 Tm +(both) Tj +1 0 0 1 218.28 222.24 Tm +(li) Tj +1 0 0 1 223.56 222.24 Tm +(v) Tj +1 0 0 1 228.48 222.24 Tm +(eness) Tj +1 0 0 1 254.399 222.24 Tm +(and) Tj +1 0 0 1 273.239 222.24 Tm +(safety) Tj +1 0 0 1 72 210.36 Tm +(pro) Tj +1 0 0 1 85.32 210.36 Tm +(vided) Tj +1 0 0 1 110.64 210.36 Tm +(at) Tj +1 0 0 1 121.08 210.36 Tm +(most) Tj +ET +endstream +endobj +22 0 obj +10155 +endobj +23 0 obj +<< /Type /XObject /Name /R23 /Subtype /Image /Length 24 0 R +/ImageMask true /Width 26 /Height 99 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 26 /BlackIs1 true >>] +>> +stream +/da'+AAiuts8W-!s8W-!s8W-!s8W-!s8W-!s8W-!s8W-!s8W-!qbR:n&-~> +endstream +endobj +24 0 obj +60 +endobj +25 0 obj +<< /Length 26 0 R >> +stream +q +2.6 0 0 -9.9 145.5 217.9 cm +/R23 Do +Q +endstream +endobj +26 0 obj +40 +endobj +27 0 obj +<< /Type /XObject /Name /R27 /Subtype /Image /Length 28 0 R +/ImageMask true /Width 43 /Height 33 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 43 /BlackIs1 true >>] +>> +stream +366ZtBKi$Kj)nf1pi$0"h;="Vq3Z%=rV-?Z\+qmGnGV5SiW&fRKsuPqJ89Nq=JdRb$l@h'!$D~> +endstream +endobj +28 0 obj +76 +endobj +29 0 obj +<< /Length 30 0 R >> +stream +q +4.3 0 0 -3.3 149.8 217.4 cm +/R27 Do +Q +endstream +endobj +30 0 obj +40 +endobj +31 0 obj +<< /Type /XObject /Name /R31 /Subtype /Image /Length 32 0 R +/ImageMask true /Width 47 /Height 5 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 47 /BlackIs1 true >>] +>> +stream +-D7RA!$D=~> +endstream +endobj +32 0 obj +12 +endobj +33 0 obj +<< /Length 34 0 R >> +stream +q +4.7 0 0 -0.5 155.1 216.2 cm +/R31 Do +Q +BT +/R3 7 Tf +1 0 0 1 160.68 214.2 Tm +(1) Tj +ET +q +14.64 0 0 -0.48 149.508 213.012 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +1 0 0 1 155.04 206.88 Tm +(3) Tj +ET +endstream +endobj +34 0 obj +213 +endobj +35 0 obj +<< /Type /XObject /Name /R35 /Subtype /Image /Length 36 0 R +/ImageMask true /Width 26 /Height 99 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 26 /BlackIs1 true >>] +>> +stream +,Ct/ij?\:qs8W-!s8W-!s8W-!s8W-!s8W-!s+[0J!$D~> +endstream +endobj +36 0 obj +46 +endobj +37 0 obj +<< /Length 38 0 R >> +stream +q +2.6 0 0 -9.9 165.3 217.9 cm +/R35 Do +Q +BT +/R3 10 Tf +1 0 0 1 173.04 210.36 Tm +(out) Tj +1 0 0 1 189.12 210.36 Tm +(of) Tj +1 0 0 1 200.76 210.36 Tm +(a) Tj +1 0 0 1 208.56 210.36 Tm +(total) Tj +1 0 0 1 229.56 210.36 Tm +(of) Tj +ET +endstream +endobj +38 0 obj +224 +endobj +39 0 obj +<< /Type /XObject /Name /R39 /Subtype /Image /Length 40 0 R +/ImageMask true /Width 54 /Height 46 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 54 /BlackIs1 true >>] +>> +stream +4N]t,$m>7rl/"o@d!re, +pYTohhg]lis4_o!aocV_^Yur>rVsXfR^`8 +endstream +endobj +40 0 obj +102 +endobj +41 0 obj +<< /Length 42 0 R >> +stream +q +5.4 0 0 -4.6 241.4 214.9 cm +/R39 Do +Q +BT +1 0 0 1 250.56 210.36 Tm +(replicas) Tj +1 0 0 1 284.88 210.36 Tm +(are) Tj +1 0 0 1 72 198.36 Tm +(simultaneously) Tj +1 0 0 1 134.52 198.36 Tm +(f) Tj +1 0 0 1 137.76 198.36 Tm +(aulty) Tj +1 0 0 1 157.2 198.36 Tm +(.) Tj +1 0 0 1 162.96 198.36 Tm +(This) Tj +1 0 0 1 182.88 198.36 Tm +(means) Tj +1 0 0 1 210.6 198.36 Tm +(that) Tj +1 0 0 1 227.64 198.36 Tm +(clients) Tj +1 0 0 1 255.839 198.36 Tm +(e) Tj +1 0 0 1 260.039 198.36 Tm +(v) Tj +1 0 0 1 264.959 198.36 Tm +(entually) Tj +1 0 0 1 72 186.48 Tm +(recei) Tj +1 0 0 1 91.2 186.48 Tm +(v) Tj +1 0 0 1 96.1199 186.48 Tm +(e) Tj +1 0 0 1 105 186.48 Tm +(replies) Tj +1 0 0 1 136.2 186.48 Tm +(to) Tj +1 0 0 1 148.68 186.48 Tm +(their) Tj +1 0 0 1 171.6 186.48 Tm +(requests) Tj +1 0 0 1 208.92 186.48 Tm +(and) Tj +1 0 0 1 227.879 186.48 Tm +(those) Tj +1 0 0 1 253.559 186.48 Tm +(replies) Tj +1 0 0 1 284.879 186.48 Tm +(are) Tj +1 0 0 1 72 174.48 Tm +(correct) Tj +1 0 0 1 101.4 174.48 Tm +(according) Tj +1 0 0 1 142.56 174.48 Tm +(to) Tj +1 0 0 1 152.04 174.48 Tm +(linearizability) Tj +1 0 0 1 209.28 174.48 Tm +([14) Tj +1 0 0 1 222.48 174.48 Tm +(,) Tj +1 0 0 1 226.799 174.48 Tm +(4].) Tj +1 0 0 1 240.959 174.48 Tm +(The) Tj +1 0 0 1 258.239 174.48 Tm +(algorithm) Tj +1 0 0 1 72 162.48 Tm +(w) Tj +1 0 0 1 79.08 162.48 Tm +(orks) Tj +1 0 0 1 99.8399 162.48 Tm +(in) Tj +1 0 0 1 111.24 162.48 Tm +(asynchronous) Tj +1 0 0 1 169.56 162.48 Tm +(systems) Tj +1 0 0 1 204.72 162.48 Tm +(lik) Tj +1 0 0 1 215.16 162.48 Tm +(e) Tj +1 0 0 1 223.2 162.48 Tm +(the) Tj +1 0 0 1 238.919 162.48 Tm +(Internet) Tj +1 0 0 1 273.479 162.48 Tm +(and) Tj +1 0 0 1 291.479 162.48 Tm +(it) Tj +1 0 0 1 72 150.6 Tm +(incorporates) Tj +1 0 0 1 126 150.6 Tm +(important) Tj +1 0 0 1 169.44 150.6 Tm +(optimizations) Tj +1 0 0 1 228.48 150.6 Tm +(that) Tj +1 0 0 1 248.159 150.6 Tm +(enable) Tj +1 0 0 1 278.999 150.6 Tm +(it) Tj +1 0 0 1 289.199 150.6 Tm +(to) Tj +1 0 0 1 72 138.6 Tm +(perform) Tj +1 0 0 1 106.56 138.6 Tm +(ef) Tj +1 0 0 1 114.12 138.6 Tm +(\256ciently) Tj +1 0 0 1 146.28 138.6 Tm +(.) Tj +1 0 0 1 81.96 126.36 Tm +(There) Tj +1 0 0 1 110.04 126.36 Tm +(is) Tj +1 0 0 1 121.68 126.36 Tm +(a) Tj +1 0 0 1 130.92 126.36 Tm +(signi\256cant) Tj +1 0 0 1 177.36 126.36 Tm +(body) Tj +1 0 0 1 202.2 126.36 Tm +(of) Tj +1 0 0 1 215.4 126.36 Tm +(w) Tj +1 0 0 1 222.48 126.36 Tm +(ork) Tj +1 0 0 1 240.599 126.36 Tm +(on) Tj +1 0 0 1 255.479 126.36 Tm +(agreement) Tj +ET +q +90 0 0 -0.48 71.988 116.892 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +/R3 8 Tf +1 0 0 1 72 108.84 Tm +(This) Tj +1 0 0 1 87.3599 108.84 Tm +(research) Tj +1 0 0 1 114.84 108.84 Tm +(w) Tj +1 0 0 1 120.48 108.84 Tm +(as) Tj +1 0 0 1 128.28 108.84 Tm +(supported) Tj +1 0 0 1 160.679 108.84 Tm +(in) Tj +1 0 0 1 167.999 108.84 Tm +(p) Tj +1 0 0 1 172.079 108.84 Tm +(art) Tj +1 0 0 1 181.679 108.84 Tm +(by) Tj +1 0 0 1 190.799 108.84 Tm +(D) Tj +1 0 0 1 196.319 108.84 Tm +(ARP) Tj +1 0 0 1 211.079 108.84 Tm +(A) Tj +1 0 0 1 218.039 108.84 Tm +(unde) Tj +1 0 0 1 233.519 108.84 Tm +(r) Tj +1 0 0 1 237.359 108.84 Tm +(c) Tj +1 0 0 1 240.959 108.84 Tm +(ont) Tj +1 0 0 1 251.159 108.84 Tm +(rac) Tj +1 0 0 1 260.879 108.84 Tm +(t) Tj +1 0 0 1 264.239 108.84 Tm +(D) Tj +1 0 0 1 269.759 108.84 Tm +(ABT63-) Tj +1 0 0 1 72 99.36 Tm +(95-C-005,) Tj +1 0 0 1 108.72 99.36 Tm +(monitored) Tj +1 0 0 1 145.2 99.36 Tm +(by) Tj +1 0 0 1 156.96 99.36 Tm +(Army) Tj +1 0 0 1 179.28 99.36 Tm +(F) Tj +1 0 0 1 183.6 99.36 Tm +(ort) Tj +1 0 0 1 196.2 99.36 Tm +(Huachuca,) Tj +1 0 0 1 234.239 99.36 Tm +(and) Tj +1 0 0 1 249.479 99.36 Tm +(under) Tj +1 0 0 1 271.319 99.36 Tm +(contract) Tj +1 0 0 1 72 89.88 Tm +(F30602-98-1-0237,) Tj +1 0 0 1 136.8 89.88 Tm +(monitored) Tj +1 0 0 1 172.08 89.88 Tm +(by) Tj +1 0 0 1 182.52 89.88 Tm +(the) Tj +1 0 0 1 194.759 89.88 Tm +(Air) Tj +1 0 0 1 207.839 89.88 Tm +(F) Tj +1 0 0 1 212.159 89.88 Tm +(orce) Tj +1 0 0 1 228.359 89.88 Tm +(Research) Tj +1 0 0 1 260.159 89.88 Tm +(Laboratory) Tj +1 0 0 1 294.839 89.88 Tm +(,) Tj +1 0 0 1 72 80.52 Tm +(and) Tj +1 0 0 1 84.5999 80.52 Tm +(in) Tj +1 0 0 1 91.9198 80.52 Tm +(part) Tj +1 0 0 1 105.36 80.52 Tm +(by) Tj +1 0 0 1 114.48 80.52 Tm +(NEC.) Tj +1 0 0 1 133.68 80.52 Tm +(Miguel) Tj +1 0 0 1 157.799 80.52 Tm +(Ca) Tj +1 0 0 1 166.679 80.52 Tm +(stro) Tj +1 0 0 1 179.879 80.52 Tm +(w) Tj +1 0 0 1 185.519 80.52 Tm +(a) Tj +1 0 0 1 189.119 80.52 Tm +(s) Tj +1 0 0 1 193.439 80.52 Tm +(pa) Tj +1 0 0 1 200.999 80.52 Tm +(rti) Tj +1 0 0 1 208.079 80.52 Tm +(al) Tj +1 0 0 1 213.839 80.52 Tm +(ly) Tj +1 0 0 1 221.279 80.52 Tm +(suppor) Tj +1 0 0 1 242.999 80.52 Tm +(ted) Tj +1 0 0 1 253.919 80.52 Tm +(by) Tj +1 0 0 1 263.159 80.52 Tm +(a) Tj +1 0 0 1 267.959 80.52 Tm +(PRAXIS) Tj +1 0 0 1 72 71.04 Tm +(XXI) Tj +1 0 0 1 88.1999 71.04 Tm +(fello) Tj +1 0 0 1 102.36 71.04 Tm +(wship.) Tj +/R3 10 Tf +1 0 0 1 315 568.68 Tm +(and) Tj +1 0 0 1 332.4 568.68 Tm +(replication) Tj +1 0 0 1 378 568.68 Tm +(techniques) Tj +1 0 0 1 423.72 568.68 Tm +(that) Tj +1 0 0 1 441.6 568.68 Tm +(tolerate) Tj +1 0 0 1 474.599 568.68 Tm +(Byzantine) Tj +1 0 0 1 517.919 568.68 Tm +(f) Tj +1 0 0 1 521.159 568.68 Tm +(aults) Tj +1 0 0 1 315 556.8 Tm +(\(starting) Tj +1 0 0 1 351.72 556.8 Tm +(with) Tj +1 0 0 1 372.96 556.8 Tm +([19) Tj +1 0 0 1 386.28 556.8 Tm +(]\).) Tj +1 0 0 1 402.24 556.8 Tm +(Ho) Tj +1 0 0 1 414.24 556.8 Tm +(we) Tj +1 0 0 1 425.64 556.8 Tm +(v) Tj +1 0 0 1 430.56 556.8 Tm +(er) Tj +1 0 0 1 438 556.8 Tm +(,) Tj +1 0 0 1 444 556.8 Tm +(most) Tj +1 0 0 1 466.92 556.8 Tm +(earlier) Tj +1 0 0 1 495.839 556.8 Tm +(w) Tj +1 0 0 1 502.919 556.8 Tm +(ork) Tj +1 0 0 1 519.839 556.8 Tm +(\(e.g.,) Tj +1 0 0 1 315 544.8 Tm +([3) Tj +1 0 0 1 323.28 544.8 Tm +(,) Tj +1 0 0 1 331.68 544.8 Tm +(24) Tj +1 0 0 1 341.64 544.8 Tm +(,) Tj +1 0 0 1 350.04 544.8 Tm +(10) Tj +1 0 0 1 360 544.8 Tm +(]\)) Tj +1 0 0 1 372.48 544.8 Tm +(either) Tj +1 0 0 1 401.16 544.8 Tm +(concerns) Tj +1 0 0 1 442.44 544.8 Tm +(techniques) Tj +1 0 0 1 490.919 544.8 Tm +(designed) Tj +1 0 0 1 532.199 544.8 Tm +(to) Tj +1 0 0 1 315 532.92 Tm +(demonstrate) Tj +1 0 0 1 366.12 532.92 Tm +(theoretical) Tj +1 0 0 1 410.52 532.92 Tm +(feasibility) Tj +1 0 0 1 452.76 532.92 Tm +(that) Tj +1 0 0 1 470.159 532.92 Tm +(are) Tj +1 0 0 1 484.679 532.92 Tm +(too) Tj +1 0 0 1 499.799 532.92 Tm +(inef) Tj +1 0 0 1 515.159 532.92 Tm +(\256cient) Tj +1 0 0 1 315 520.92 Tm +(to) Tj +1 0 0 1 328.56 520.92 Tm +(be) Tj +1 0 0 1 343.8 520.92 Tm +(used) Tj +1 0 0 1 367.92 520.92 Tm +(in) Tj +1 0 0 1 381.6 520.92 Tm +(practice,) Tj +1 0 0 1 422.28 520.92 Tm +(or) Tj +1 0 0 1 436.44 520.92 Tm +(assumes) Tj +1 0 0 1 475.44 520.92 Tm +(synchron) Tj +1 0 0 1 512.159 520.92 Tm +(y) Tj +1 0 0 1 516.599 520.92 Tm +(,) Tj +1 0 0 1 525.359 520.92 Tm +(i.e.,) Tj +1 0 0 1 315 509.04 Tm +(relies) Tj +1 0 0 1 339.6 509.04 Tm +(on) Tj +1 0 0 1 352.68 509.04 Tm +(kno) Tj +1 0 0 1 367.56 509.04 Tm +(wn) Tj +1 0 0 1 382.56 509.04 Tm +(bounds) Tj +1 0 0 1 414.36 509.04 Tm +(on) Tj +1 0 0 1 427.44 509.04 Tm +(message) Tj +1 0 0 1 464.159 509.04 Tm +(delays) Tj +1 0 0 1 492.719 509.04 Tm +(and) Tj +1 0 0 1 510.119 509.04 Tm +(process) Tj +1 0 0 1 315 497.04 Tm +(speeds.) Tj +1 0 0 1 350.04 497.04 Tm +(The) Tj +1 0 0 1 368.76 497.04 Tm +(systems) Tj +1 0 0 1 403.56 497.04 Tm +(closest) Tj +1 0 0 1 434.04 497.04 Tm +(to) Tj +1 0 0 1 445.08 497.04 Tm +(ours,) Tj +1 0 0 1 468.119 497.04 Tm +(Rampart) Tj +1 0 0 1 505.679 497.04 Tm +([30) Tj +1 0 0 1 518.999 497.04 Tm +(]) Tj +1 0 0 1 525.599 497.04 Tm +(and) Tj +1 0 0 1 315 485.16 Tm +(SecureRing) Tj +1 0 0 1 364.08 485.16 Tm +([16) Tj +1 0 0 1 377.4 485.16 Tm +(],) Tj +1 0 0 1 385.92 485.16 Tm +(were) Tj +1 0 0 1 407.88 485.16 Tm +(designed) Tj +1 0 0 1 445.92 485.16 Tm +(to) Tj +1 0 0 1 456.36 485.16 Tm +(be) Tj +1 0 0 1 468.36 485.16 Tm +(practical,) Tj +1 0 0 1 507.839 485.16 Tm +(b) Tj +1 0 0 1 512.639 485.16 Tm +(ut) Tj +1 0 0 1 522.959 485.16 Tm +(the) Tj +1 0 0 1 535.079 485.16 Tm +(y) Tj +1 0 0 1 315 473.16 Tm +(rely) Tj +1 0 0 1 333 473.16 Tm +(on) Tj +1 0 0 1 345.36 473.16 Tm +(the) Tj +1 0 0 1 360.12 473.16 Tm +(synchron) Tj +1 0 0 1 396.84 473.16 Tm +(y) Tj +1 0 0 1 403.92 473.16 Tm +(assumption) Tj +1 0 0 1 451.8 473.16 Tm +(for) Tj +1 0 0 1 465.84 473.16 Tm +(correctness,) Tj +1 0 0 1 515.639 473.16 Tm +(which) Tj +1 0 0 1 315 461.28 Tm +(is) Tj +1 0 0 1 325.32 461.28 Tm +(dangerous) Tj +1 0 0 1 369.96 461.28 Tm +(in) Tj +1 0 0 1 381.48 461.28 Tm +(the) Tj +1 0 0 1 397.32 461.28 Tm +(presence) Tj +1 0 0 1 435.96 461.28 Tm +(of) Tj +1 0 0 1 447.96 461.28 Tm +(malicious) Tj +1 0 0 1 490.439 461.28 Tm +(attacks.) Tj +1 0 0 1 527.879 461.28 Tm +(An) Tj +1 0 0 1 315 449.28 Tm +(attack) Tj +1 0 0 1 338.76 449.28 Tm +(er) Tj +1 0 0 1 351.48 449.28 Tm +(may) Tj +1 0 0 1 373.68 449.28 Tm +(compromise) Tj +1 0 0 1 427.92 449.28 Tm +(the) Tj +1 0 0 1 445.2 449.28 Tm +(safety) Tj +1 0 0 1 474 449.28 Tm +(of) Tj +1 0 0 1 487.319 449.28 Tm +(a) Tj +1 0 0 1 496.799 449.28 Tm +(service) Tj +1 0 0 1 530.039 449.28 Tm +(by) Tj +1 0 0 1 315 437.28 Tm +(delaying) Tj +1 0 0 1 351 437.28 Tm +(non-f) Tj +1 0 0 1 372.72 437.28 Tm +(aulty) Tj +1 0 0 1 394.2 437.28 Tm +(nodes) Tj +1 0 0 1 419.16 437.28 Tm +(or) Tj +1 0 0 1 429.24 437.28 Tm +(the) Tj +1 0 0 1 443.04 437.28 Tm +(communication) Tj +1 0 0 1 506.759 437.28 Tm +(between) Tj +1 0 0 1 315 425.4 Tm +(them) Tj +1 0 0 1 336.72 425.4 Tm +(until) Tj +1 0 0 1 356.76 425.4 Tm +(the) Tj +1 0 0 1 368.88 425.4 Tm +(y) Tj +1 0 0 1 375.6 425.4 Tm +(are) Tj +1 0 0 1 389.52 425.4 Tm +(tagged) Tj +1 0 0 1 417.84 425.4 Tm +(as) Tj +1 0 0 1 427.92 425.4 Tm +(f) Tj +1 0 0 1 431.16 425.4 Tm +(aulty) Tj +1 0 0 1 452.879 425.4 Tm +(and) Tj +1 0 0 1 469.079 425.4 Tm +(e) Tj +1 0 0 1 473.399 425.4 Tm +(xcluded) Tj +1 0 0 1 506.639 425.4 Tm +(from) Tj +1 0 0 1 527.879 425.4 Tm +(the) Tj +1 0 0 1 315 413.4 Tm +(replica) Tj +1 0 0 1 343.8 413.4 Tm +(group.) Tj +1 0 0 1 372.72 413.4 Tm +(Such) Tj +1 0 0 1 394.32 413.4 Tm +(a) Tj +1 0 0 1 400.32 413.4 Tm +(denial-of-service) Tj +1 0 0 1 469.799 413.4 Tm +(attack) Tj +1 0 0 1 495.239 413.4 Tm +(is) Tj +1 0 0 1 503.399 413.4 Tm +(generally) Tj +1 0 0 1 315 401.52 Tm +(easier) Tj +1 0 0 1 340.68 401.52 Tm +(than) Tj +1 0 0 1 360.36 401.52 Tm +(gaining) Tj +1 0 0 1 392.64 401.52 Tm +(control) Tj +1 0 0 1 423.36 401.52 Tm +(o) Tj +1 0 0 1 428.28 401.52 Tm +(v) Tj +1 0 0 1 433.2 401.52 Tm +(er) Tj +1 0 0 1 443.279 401.52 Tm +(a) Tj +1 0 0 1 450.24 401.52 Tm +(non-f) Tj +1 0 0 1 471.959 401.52 Tm +(aulty) Tj +1 0 0 1 494.159 401.52 Tm +(node.) Tj +1 0 0 1 324.96 389.04 Tm +(Our) Tj +1 0 0 1 346.92 389.04 Tm +(algorithm) Tj +1 0 0 1 392.16 389.04 Tm +(is) Tj +1 0 0 1 405.24 389.04 Tm +(not) Tj +1 0 0 1 424.44 389.04 Tm +(vulnerable) Tj +1 0 0 1 473.04 389.04 Tm +(to) Tj +1 0 0 1 487.2 389.04 Tm +(this) Tj +1 0 0 1 508.079 389.04 Tm +(type) Tj +1 0 0 1 531.719 389.04 Tm +(of) Tj +1 0 0 1 315 377.04 Tm +(attack) Tj +1 0 0 1 345.6 377.04 Tm +(because) Tj +1 0 0 1 384 377.04 Tm +(it) Tj +1 0 0 1 396.36 377.04 Tm +(does) Tj +1 0 0 1 421.44 377.04 Tm +(not) Tj +1 0 0 1 441 377.04 Tm +(rely) Tj +1 0 0 1 463.44 377.04 Tm +(on) Tj +1 0 0 1 480.24 377.04 Tm +(synchron) Tj +1 0 0 1 516.959 377.04 Tm +(y) Tj +1 0 0 1 528.359 377.04 Tm +(for) Tj +1 0 0 1 315 365.16 Tm +(safety) Tj +1 0 0 1 338.28 365.16 Tm +(.) Tj +1 0 0 1 352.08 365.16 Tm +(In) Tj +1 0 0 1 365.52 365.16 Tm +(addition,) Tj +1 0 0 1 406.44 365.16 Tm +(it) Tj +1 0 0 1 417.12 365.16 Tm +(impro) Tj +1 0 0 1 441 365.16 Tm +(v) Tj +1 0 0 1 445.92 365.16 Tm +(es) Tj +1 0 0 1 458.999 365.16 Tm +(the) Tj +1 0 0 1 476.279 365.16 Tm +(performance) Tj +1 0 0 1 531.719 365.16 Tm +(of) Tj +1 0 0 1 315 353.16 Tm +(Rampart) Tj +1 0 0 1 354.72 353.16 Tm +(and) Tj +1 0 0 1 374.52 353.16 Tm +(SecureRing) Tj +1 0 0 1 426.48 353.16 Tm +(by) Tj +1 0 0 1 441.84 353.16 Tm +(more) Tj +1 0 0 1 467.76 353.16 Tm +(than) Tj +1 0 0 1 490.44 353.16 Tm +(an) Tj +1 0 0 1 505.199 353.16 Tm +(order) Tj +1 0 0 1 531.719 353.16 Tm +(of) Tj +1 0 0 1 315 341.28 Tm +(magnitude) Tj +1 0 0 1 360.6 341.28 Tm +(as) Tj +1 0 0 1 372.48 341.28 Tm +(e) Tj +1 0 0 1 376.8 341.28 Tm +(xplained) Tj +1 0 0 1 414.72 341.28 Tm +(in) Tj +1 0 0 1 426 341.28 Tm +(Section) Tj +1 0 0 1 459.48 341.28 Tm +(7.) Tj +1 0 0 1 473.879 341.28 Tm +(It) Tj +1 0 0 1 483.599 341.28 Tm +(uses) Tj +1 0 0 1 504.359 341.28 Tm +(only) Tj +1 0 0 1 525.599 341.28 Tm +(one) Tj +1 0 0 1 315 329.28 Tm +(message) Tj +1 0 0 1 351.96 329.28 Tm +(round) Tj +1 0 0 1 378.36 329.28 Tm +(trip) Tj +1 0 0 1 395.4 329.28 Tm +(to) Tj +1 0 0 1 406.32 329.28 Tm +(e) Tj +1 0 0 1 410.64 329.28 Tm +(x) Tj +1 0 0 1 415.56 329.28 Tm +(ecute) Tj +1 0 0 1 439.56 329.28 Tm +(read-only) Tj +1 0 0 1 480.959 329.28 Tm +(operations) Tj +1 0 0 1 525.599 329.28 Tm +(and) Tj +1 0 0 1 315 317.4 Tm +(tw) Tj +1 0 0 1 324.84 317.4 Tm +(o) Tj +1 0 0 1 335.04 317.4 Tm +(to) Tj +1 0 0 1 347.88 317.4 Tm +(e) Tj +1 0 0 1 352.2 317.4 Tm +(x) Tj +1 0 0 1 357.12 317.4 Tm +(ecute) Tj +1 0 0 1 383.28 317.4 Tm +(read-write) Tj +1 0 0 1 429.36 317.4 Tm +(operations.) Tj +1 0 0 1 485.039 317.4 Tm +(Also,) Tj +1 0 0 1 512.159 317.4 Tm +(it) Tj +1 0 0 1 522.839 317.4 Tm +(uses) Tj +1 0 0 1 315 305.4 Tm +(an) Tj +1 0 0 1 329.52 305.4 Tm +(ef) Tj +1 0 0 1 337.08 305.4 Tm +(\256cient) Tj +1 0 0 1 366.96 305.4 Tm +(authentication) Tj +1 0 0 1 428.52 305.4 Tm +(scheme) Tj +1 0 0 1 463.44 305.4 Tm +(based) Tj +1 0 0 1 491.159 305.4 Tm +(on) Tj +1 0 0 1 506.279 305.4 Tm +(message) Tj +1 0 0 1 315 293.52 Tm +(authentication) Tj +1 0 0 1 373.2 293.52 Tm +(codes) Tj +1 0 0 1 397.56 293.52 Tm +(during) Tj +1 0 0 1 425.28 293.52 Tm +(normal) Tj +1 0 0 1 455.28 293.52 Tm +(operation;) Tj +1 0 0 1 497.639 293.52 Tm +(public-k) Tj +1 0 0 1 530.999 293.52 Tm +(e) Tj +1 0 0 1 535.319 293.52 Tm +(y) Tj +1 0 0 1 315 281.52 Tm +(cryptography) Tj +1 0 0 1 368.04 281.52 Tm +(,) Tj +1 0 0 1 372.72 281.52 Tm +(which) Tj +1 0 0 1 399.84 281.52 Tm +(w) Tj +1 0 0 1 406.92 281.52 Tm +(as) Tj +1 0 0 1 417.96 281.52 Tm +(cited) Tj +1 0 0 1 440.04 281.52 Tm +(as) Tj +1 0 0 1 451.08 281.52 Tm +(the) Tj +1 0 0 1 466.079 281.52 Tm +(major) Tj +1 0 0 1 491.999 281.52 Tm +(latenc) Tj +1 0 0 1 515.759 281.52 Tm +(y) Tj +1 0 0 1 523.439 281.52 Tm +([29) Tj +1 0 0 1 536.639 281.52 Tm +(]) Tj +1 0 0 1 315 269.64 Tm +(and) Tj +1 0 0 1 332.04 269.64 Tm +(throughput) Tj +1 0 0 1 378.48 269.64 Tm +([22) Tj +1 0 0 1 391.8 269.64 Tm +(]) Tj +1 0 0 1 397.8 269.64 Tm +(bottleneck) Tj +1 0 0 1 441.96 269.64 Tm +(in) Tj +1 0 0 1 452.4 269.64 Tm +(Rampart,) Tj +1 0 0 1 491.999 269.64 Tm +(is) Tj +1 0 0 1 501.359 269.64 Tm +(used) Tj +1 0 0 1 522.239 269.64 Tm +(only) Tj +1 0 0 1 315 257.64 Tm +(when) Tj +1 0 0 1 339.12 257.64 Tm +(there) Tj +1 0 0 1 361.44 257.64 Tm +(are) Tj +1 0 0 1 376.08 257.64 Tm +(f) Tj +1 0 0 1 379.32 257.64 Tm +(aults.) Tj +1 0 0 1 324.96 245.16 Tm +(T) Tj +1 0 0 1 330.24 245.16 Tm +(o) Tj +1 0 0 1 338.28 245.16 Tm +(e) Tj +1 0 0 1 342.48 245.16 Tm +(v) Tj +1 0 0 1 347.28 245.16 Tm +(aluate) Tj +1 0 0 1 373.92 245.16 Tm +(our) Tj +1 0 0 1 390.24 245.16 Tm +(approach,) Tj +1 0 0 1 432.48 245.16 Tm +(we) Tj +1 0 0 1 447.12 245.16 Tm +(implemented) Tj +1 0 0 1 502.079 245.16 Tm +(a) Tj +1 0 0 1 509.519 245.16 Tm +(replica-) Tj +1 0 0 1 315 233.16 Tm +(tion) Tj +1 0 0 1 334.2 233.16 Tm +(library) Tj +1 0 0 1 364.56 233.16 Tm +(and) Tj +1 0 0 1 382.68 233.16 Tm +(used) Tj +1 0 0 1 404.64 233.16 Tm +(it) Tj +1 0 0 1 413.88 233.16 Tm +(to) Tj +1 0 0 1 425.4 233.16 Tm +(implement) Tj +1 0 0 1 471.719 233.16 Tm +(a) Tj +1 0 0 1 479.879 233.16 Tm +(real) Tj +1 0 0 1 498.479 233.16 Tm +(service:) Tj +1 0 0 1 535.559 233.16 Tm +(a) Tj +1 0 0 1 315 221.28 Tm +(Byzantine-f) Tj +1 0 0 1 362.16 221.28 Tm +(ault-tolerant) Tj +1 0 0 1 413.52 221.28 Tm +(distrib) Tj +1 0 0 1 438.84 221.28 Tm +(uted) Tj +1 0 0 1 458.879 221.28 Tm +(\256le) Tj +1 0 0 1 474.479 221.28 Tm +(system) Tj +1 0 0 1 504.959 221.28 Tm +(that) Tj +1 0 0 1 522.839 221.28 Tm +(sup-) Tj +1 0 0 1 315 209.28 Tm +(ports) Tj +1 0 0 1 339 209.28 Tm +(the) Tj +1 0 0 1 355.2 209.28 Tm +(NFS) Tj +1 0 0 1 377.52 209.28 Tm +(protocol.) Tj +1 0 0 1 421.44 209.28 Tm +(W) Tj +1 0 0 1 429.96 209.28 Tm +(e) Tj +1 0 0 1 438.48 209.28 Tm +(used) Tj +1 0 0 1 460.8 209.28 Tm +(the) Tj +1 0 0 1 476.999 209.28 Tm +(Andre) Tj +1 0 0 1 501.839 209.28 Tm +(w) Tj +1 0 0 1 512.879 209.28 Tm +(bench-) Tj +1 0 0 1 315 197.4 Tm +(mark) Tj +1 0 0 1 337.32 197.4 Tm +([15) Tj +1 0 0 1 350.52 197.4 Tm +(]) Tj +1 0 0 1 355.68 197.4 Tm +(to) Tj +1 0 0 1 365.16 197.4 Tm +(e) Tj +1 0 0 1 369.36 197.4 Tm +(v) Tj +1 0 0 1 374.16 197.4 Tm +(aluate) Tj +1 0 0 1 399.72 197.4 Tm +(the) Tj +1 0 0 1 413.64 197.4 Tm +(performance) Tj +1 0 0 1 465.84 197.4 Tm +(of) Tj +1 0 0 1 475.919 197.4 Tm +(our) Tj +1 0 0 1 490.919 197.4 Tm +(system.) Tj +1 0 0 1 524.519 197.4 Tm +(The) Tj +1 0 0 1 315 185.4 Tm +(results) Tj +1 0 0 1 343.92 185.4 Tm +(sho) Tj +1 0 0 1 357.6 185.4 Tm +(w) Tj +1 0 0 1 367.56 185.4 Tm +(that) Tj +1 0 0 1 385.44 185.4 Tm +(our) Tj +1 0 0 1 401.64 185.4 Tm +(system) Tj +1 0 0 1 432.24 185.4 Tm +(is) Tj +1 0 0 1 441.72 185.4 Tm +(only) Tj +1 0 0 1 462.359 185.4 Tm +(3%) Tj +1 0 0 1 478.559 185.4 Tm +(slo) Tj +1 0 0 1 489.959 185.4 Tm +(wer) Tj +1 0 0 1 507.719 185.4 Tm +(than) Tj +1 0 0 1 527.879 185.4 Tm +(the) Tj +1 0 0 1 315 173.52 Tm +(standard) Tj +1 0 0 1 351.6 173.52 Tm +(NFS) Tj +1 0 0 1 372.84 173.52 Tm +(daemon) Tj +1 0 0 1 407.28 173.52 Tm +(in) Tj +1 0 0 1 417.84 173.52 Tm +(the) Tj +1 0 0 1 432.96 173.52 Tm +(Digital) Tj +1 0 0 1 463.44 173.52 Tm +(Unix) Tj +1 0 0 1 486.359 173.52 Tm +(k) Tj +1 0 0 1 491.279 173.52 Tm +(ernel) Tj +1 0 0 1 513.959 173.52 Tm +(during) Tj +1 0 0 1 315 161.52 Tm +(normal-case) Tj +1 0 0 1 366.12 161.52 Tm +(operation.) Tj +1 0 0 1 324.96 149.04 Tm +(Thus,) Tj +1 0 0 1 349.92 149.04 Tm +(the) Tj +1 0 0 1 364.56 149.04 Tm +(paper) Tj +1 0 0 1 389.16 149.04 Tm +(mak) Tj +1 0 0 1 406.32 149.04 Tm +(es) Tj +1 0 0 1 417 149.04 Tm +(the) Tj +1 0 0 1 431.64 149.04 Tm +(follo) Tj +1 0 0 1 450.36 149.04 Tm +(wing) Tj +1 0 0 1 472.679 149.04 Tm +(contrib) Tj +1 0 0 1 500.879 149.04 Tm +(utions:) Tj +ET +endstream +endobj +42 0 obj +17137 +endobj +43 0 obj +<< /Type /XObject /Name /R43 /Subtype /Image /Length 44 0 R +/ImageMask true /Width 39 /Height 39 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 39 /BlackIs1 true >>] +>> +stream +3f*u*F9LP-kW3[R5O&26s8VF=rNC"SR~> +endstream +endobj +44 0 obj +48 +endobj +45 0 obj +<< /Length 46 0 R >> +stream +q +3.9 0 0 -3.9 320.5 138.1 cm +/R43 Do +Q +BT +1 0 0 1 330 133.56 Tm +(It) Tj +1 0 0 1 338.88 133.56 Tm +(describes) Tj +1 0 0 1 378.72 133.56 Tm +(the) Tj +1 0 0 1 393.72 133.56 Tm +(\256rst) Tj +1 0 0 1 412.08 133.56 Tm +(state-machine) Tj +1 0 0 1 470.16 133.56 Tm +(replication) Tj +1 0 0 1 515.639 133.56 Tm +(proto-) Tj +1 0 0 1 330 121.56 Tm +(col) Tj +1 0 0 1 345.36 121.56 Tm +(that) Tj +1 0 0 1 363.6 121.56 Tm +(correctly) Tj +1 0 0 1 402.36 121.56 Tm +(survi) Tj +1 0 0 1 422.16 121.56 Tm +(v) Tj +1 0 0 1 427.08 121.56 Tm +(es) Tj +1 0 0 1 438.36 121.56 Tm +(Byzantine) Tj +1 0 0 1 482.04 121.56 Tm +(f) Tj +1 0 0 1 485.279 121.56 Tm +(aults) Tj +1 0 0 1 507.359 121.56 Tm +(in) Tj +1 0 0 1 518.399 121.56 Tm +(asyn-) Tj +1 0 0 1 330 109.56 Tm +(chronous) Tj +1 0 0 1 369 109.56 Tm +(netw) Tj +1 0 0 1 388.32 109.56 Tm +(orks.) Tj +ET +q +3.9 0 0 -3.9 320.5 99.4 cm +/R43 Do +Q +BT +1 0 0 1 330 94.92 Tm +(It) Tj +1 0 0 1 338.28 94.92 Tm +(describes) Tj +1 0 0 1 377.64 94.92 Tm +(a) Tj +1 0 0 1 384.24 94.92 Tm +(number) Tj +1 0 0 1 417 94.92 Tm +(of) Tj +1 0 0 1 427.56 94.92 Tm +(important) Tj +1 0 0 1 468.599 94.92 Tm +(optimizations) Tj +1 0 0 1 524.999 94.92 Tm +(that) Tj +1 0 0 1 330 82.92 Tm +(allo) Tj +1 0 0 1 344.76 82.92 Tm +(w) Tj +1 0 0 1 354.6 82.92 Tm +(the) Tj +1 0 0 1 369.6 82.92 Tm +(algorithm) Tj +1 0 0 1 411 82.92 Tm +(to) Tj +1 0 0 1 421.56 82.92 Tm +(perform) Tj +1 0 0 1 456.48 82.92 Tm +(well) Tj +1 0 0 1 476.4 82.92 Tm +(so) Tj +1 0 0 1 487.919 82.92 Tm +(that) Tj +1 0 0 1 505.679 82.92 Tm +(it) Tj +1 0 0 1 513.959 82.92 Tm +(can) Tj +1 0 0 1 530.639 82.92 Tm +(be) Tj +1 0 0 1 330 71.04 Tm +(used) Tj +1 0 0 1 350.64 71.04 Tm +(in) Tj +1 0 0 1 360.96 71.04 Tm +(real) Tj +1 0 0 1 378.36 71.04 Tm +(systems.) Tj +1 0 0 1 303.48 35.16 Tm +(1) Tj +ET +Q +endstream +endobj +46 0 obj +1695 +endobj +47 0 obj +<< +/Type /Page +/MediaBox [0 0 612 792] +/Parent 2 0 R +/Resources << /ProcSet [/PDF /ImageB /Text] +/Font << +/R16 16 0 R +/R9 9 0 R +/R6 6 0 R +/R3 3 0 R +>> +/XObject << +/R43 43 0 R +/R39 39 0 R +/R35 35 0 R +/R31 31 0 R +/R27 27 0 R +/R23 23 0 R +/R19 19 0 R +/R12 12 0 R +>> +>> +/Contents [ +4 0 R +7 0 R +10 0 R +14 0 R +17 0 R +21 0 R +25 0 R +29 0 R +33 0 R +37 0 R +41 0 R +45 0 R +] +>> +endobj +48 0 obj +<< /Length 49 0 R >> +stream +q +Q +q +W +0 0 612 792 re +n +endstream +endobj +49 0 obj +25 +endobj +50 0 obj +<< /Type /XObject /Name /R50 /Subtype /Image /Length 51 0 R +/ImageMask true /Width 39 /Height 39 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 39 /BlackIs1 true >>] +>> +stream +3f*u*F9LP-kW3[R5O&26s8VF=rNC"SR~> +endstream +endobj +51 0 obj +48 +endobj +52 0 obj +<< /Length 53 0 R >> +stream +q +3.9 0 0 -3.9 77.5 713.6 cm +/R50 Do +Q +endstream +endobj +53 0 obj +39 +endobj +54 0 obj +<< /Type /Font /Name /R54 /Subtype /Type1 /BaseFont /Times-Roman >> +endobj +55 0 obj +<< /Length 56 0 R >> +stream +BT +/R54 10 Tf +1 0 0 1 87 709.08 Tm +(It) Tj +1 0 0 1 95.3999 709.08 Tm +(describes) Tj +1 0 0 1 134.88 709.08 Tm +(the) Tj +1 0 0 1 149.4 709.08 Tm +(implementation) Tj +1 0 0 1 214.32 709.08 Tm +(of) Tj +1 0 0 1 225 709.08 Tm +(a) Tj +1 0 0 1 231.839 709.08 Tm +(Byzantine-f) Tj +1 0 0 1 278.999 709.08 Tm +(ault-) Tj +1 0 0 1 87 697.08 Tm +(tolerant) Tj +1 0 0 1 119.88 697.08 Tm +(distrib) Tj +1 0 0 1 145.2 697.08 Tm +(uted) Tj +1 0 0 1 164.76 697.08 Tm +(\256le) Tj +1 0 0 1 180 697.08 Tm +(system.) Tj +ET +q +3.9 0 0 -3.9 77.5 687.2 cm +/R50 Do +Q +BT +1 0 0 1 87 682.68 Tm +(It) Tj +1 0 0 1 95.2799 682.68 Tm +(pro) Tj +1 0 0 1 108.6 682.68 Tm +(vides) Tj +1 0 0 1 131.64 682.68 Tm +(e) Tj +1 0 0 1 135.96 682.68 Tm +(xperimental) Tj +1 0 0 1 185.76 682.68 Tm +(results) Tj +1 0 0 1 213.96 682.68 Tm +(that) Tj +1 0 0 1 231.12 682.68 Tm +(quantify) Tj +1 0 0 1 266.519 682.68 Tm +(the) Tj +1 0 0 1 280.919 682.68 Tm +(cost) Tj +1 0 0 1 87 670.8 Tm +(of) Tj +1 0 0 1 97.6799 670.8 Tm +(the) Tj +1 0 0 1 112.44 670.8 Tm +(replication) Tj +1 0 0 1 157.44 670.8 Tm +(technique.) Tj +1 0 0 1 81.96 653.76 Tm +(The) Tj +1 0 0 1 101.28 653.76 Tm +(remainder) Tj +1 0 0 1 145.44 653.76 Tm +(of) Tj +1 0 0 1 157.56 653.76 Tm +(the) Tj +1 0 0 1 173.52 653.76 Tm +(paper) Tj +1 0 0 1 199.44 653.76 Tm +(is) Tj +1 0 0 1 209.88 653.76 Tm +(or) Tj +1 0 0 1 218.16 653.76 Tm +(ganized) Tj +1 0 0 1 252.719 653.76 Tm +(as) Tj +1 0 0 1 264.839 653.76 Tm +(follo) Tj +1 0 0 1 283.559 653.76 Tm +(ws.) Tj +1 0 0 1 72 641.88 Tm +(W) Tj +1 0 0 1 80.52 641.88 Tm +(e) Tj +1 0 0 1 87.48 641.88 Tm +(be) Tj +1 0 0 1 96.8399 641.88 Tm +(gin) Tj +1 0 0 1 111.84 641.88 Tm +(by) Tj +1 0 0 1 124.2 641.88 Tm +(describing) Tj +1 0 0 1 168.12 641.88 Tm +(our) Tj +1 0 0 1 183.84 641.88 Tm +(system) Tj +1 0 0 1 213.84 641.88 Tm +(model,) Tj +1 0 0 1 243.719 641.88 Tm +(including) Tj +1 0 0 1 283.679 641.88 Tm +(our) Tj +1 0 0 1 72 629.88 Tm +(f) Tj +1 0 0 1 75.24 629.88 Tm +(ailure) Tj +1 0 0 1 102.12 629.88 Tm +(assumptions.) Tj +1 0 0 1 162.72 629.88 Tm +(Section) Tj +1 0 0 1 196.8 629.88 Tm +(3) Tj +1 0 0 1 206.04 629.88 Tm +(describes) Tj +1 0 0 1 247.439 629.88 Tm +(the) Tj +1 0 0 1 263.759 629.88 Tm +(problem) Tj +1 0 0 1 72 618 Tm +(solv) Tj +1 0 0 1 88.56 618 Tm +(ed) Tj +1 0 0 1 99.7199 618 Tm +(by) Tj +1 0 0 1 111.48 618 Tm +(the) Tj +1 0 0 1 125.52 618 Tm +(algorithm) Tj +1 0 0 1 166.08 618 Tm +(and) Tj +1 0 0 1 182.4 618 Tm +(states) Tj +1 0 0 1 206.279 618 Tm +(correctness) Tj +1 0 0 1 252.959 618 Tm +(conditions.) Tj +1 0 0 1 72 606 Tm +(The) Tj +1 0 0 1 93.4799 606 Tm +(algorithm) Tj +1 0 0 1 138.24 606 Tm +(is) Tj +1 0 0 1 150.84 606 Tm +(described) Tj +1 0 0 1 195 606 Tm +(in) Tj +1 0 0 1 208.8 606 Tm +(Section) Tj +1 0 0 1 244.679 606 Tm +(4) Tj +1 0 0 1 255.599 606 Tm +(and) Tj +1 0 0 1 275.999 606 Tm +(some) Tj +1 0 0 1 72 594.12 Tm +(important) Tj +1 0 0 1 116.88 594.12 Tm +(optimizations) Tj +1 0 0 1 177.12 594.12 Tm +(are) Tj +1 0 0 1 195.48 594.12 Tm +(described) Tj +1 0 0 1 239.76 594.12 Tm +(in) Tj +1 0 0 1 253.559 594.12 Tm +(Section) Tj +1 0 0 1 289.559 594.12 Tm +(5.) Tj +1 0 0 1 72 582.12 Tm +(Section) Tj +1 0 0 1 107.52 582.12 Tm +(6) Tj +1 0 0 1 118.08 582.12 Tm +(describes) Tj +1 0 0 1 160.8 582.12 Tm +(our) Tj +1 0 0 1 179.64 582.12 Tm +(replication) Tj +1 0 0 1 227.88 582.12 Tm +(library) Tj +1 0 0 1 260.039 582.12 Tm +(and) Tj +1 0 0 1 280.079 582.12 Tm +(ho) Tj +1 0 0 1 289.919 582.12 Tm +(w) Tj +1 0 0 1 72 570.24 Tm +(we) Tj +1 0 0 1 90.1199 570.24 Tm +(used) Tj +1 0 0 1 114.96 570.24 Tm +(it) Tj +1 0 0 1 126.96 570.24 Tm +(to) Tj +1 0 0 1 141.24 570.24 Tm +(implement) Tj +1 0 0 1 190.44 570.24 Tm +(a) Tj +1 0 0 1 201.36 570.24 Tm +(Byzantine-f) Tj +1 0 0 1 248.519 570.24 Tm +(ault-tolerant) Tj +1 0 0 1 72 558.24 Tm +(NFS.) Tj +1 0 0 1 96 558.24 Tm +(Section) Tj +1 0 0 1 129.12 558.24 Tm +(7) Tj +1 0 0 1 137.4 558.24 Tm +(presents) Tj +1 0 0 1 173.28 558.24 Tm +(the) Tj +1 0 0 1 188.64 558.24 Tm +(results) Tj +1 0 0 1 217.92 558.24 Tm +(of) Tj +1 0 0 1 229.44 558.24 Tm +(our) Tj +1 0 0 1 246 558.24 Tm +(e) Tj +1 0 0 1 250.32 558.24 Tm +(xperiments.) Tj +1 0 0 1 72 546.24 Tm +(Section) Tj +1 0 0 1 105.48 546.24 Tm +(8) Tj +1 0 0 1 114 546.24 Tm +(discusses) Tj +1 0 0 1 154.68 546.24 Tm +(related) Tj +1 0 0 1 185.4 546.24 Tm +(w) Tj +1 0 0 1 192.48 546.24 Tm +(ork.) Tj +1 0 0 1 215.16 546.24 Tm +(W) Tj +1 0 0 1 223.68 546.24 Tm +(e) Tj +1 0 0 1 231.72 546.24 Tm +(conclude) Tj +1 0 0 1 271.319 546.24 Tm +(with) Tj +1 0 0 1 292.559 546.24 Tm +(a) Tj +1 0 0 1 72 534.36 Tm +(summary) Tj +1 0 0 1 111 534.36 Tm +(of) Tj +1 0 0 1 121.2 534.36 Tm +(what) Tj +1 0 0 1 142.44 534.36 Tm +(we) Tj +1 0 0 1 155.88 534.36 Tm +(ha) Tj +1 0 0 1 165.12 534.36 Tm +(v) Tj +1 0 0 1 170.04 534.36 Tm +(e) Tj +1 0 0 1 176.28 534.36 Tm +(accomplished) Tj +1 0 0 1 232.92 534.36 Tm +(and) Tj +1 0 0 1 249.119 534.36 Tm +(a) Tj +1 0 0 1 255.479 534.36 Tm +(discussion) Tj +1 0 0 1 72 522.36 Tm +(of) Tj +1 0 0 1 82.8 522.36 Tm +(future) Tj +1 0 0 1 109.08 522.36 Tm +(research) Tj +1 0 0 1 144.72 522.36 Tm +(directions.) Tj +ET +endstream +endobj +56 0 obj +4797 +endobj +57 0 obj +<< /Type /Font /Name /R57 /Subtype /Type1 /BaseFont /Times-Bold >> +endobj +58 0 obj +<< /Length 59 0 R >> +stream +BT +/R57 12 Tf +1 0 0 1 72 495.24 Tm +(2) Tj +1 0 0 1 89.8799 495.24 Tm +(System) Tj +1 0 0 1 129.48 495.24 Tm +(Model) Tj +/R54 10 Tf +1 0 0 1 72 478.2 Tm +(W) Tj +1 0 0 1 80.52 478.2 Tm +(e) Tj +1 0 0 1 89.04 478.2 Tm +(assume) Tj +1 0 0 1 122.4 478.2 Tm +(an) Tj +1 0 0 1 135.72 478.2 Tm +(asynchronous) Tj +1 0 0 1 194.52 478.2 Tm +(distrib) Tj +1 0 0 1 219.84 478.2 Tm +(uted) Tj +1 0 0 1 240.96 478.2 Tm +(system) Tj +1 0 0 1 272.639 478.2 Tm +(where) Tj +1 0 0 1 72 466.2 Tm +(nodes) Tj +1 0 0 1 99.1199 466.2 Tm +(are) Tj +1 0 0 1 115.08 466.2 Tm +(connected) Tj +1 0 0 1 159.36 466.2 Tm +(by) Tj +1 0 0 1 173.16 466.2 Tm +(a) Tj +1 0 0 1 181.44 466.2 Tm +(netw) Tj +1 0 0 1 200.76 466.2 Tm +(ork.) Tj +1 0 0 1 224.16 466.2 Tm +(The) Tj +1 0 0 1 243.48 466.2 Tm +(netw) Tj +1 0 0 1 262.799 466.2 Tm +(ork) Tj +1 0 0 1 279.839 466.2 Tm +(may) Tj +1 0 0 1 72 454.32 Tm +(f) Tj +1 0 0 1 75.24 454.32 Tm +(ail) Tj +1 0 0 1 88.3199 454.32 Tm +(to) Tj +1 0 0 1 99.1199 454.32 Tm +(deli) Tj +1 0 0 1 113.88 454.32 Tm +(v) Tj +1 0 0 1 118.8 454.32 Tm +(er) Tj +1 0 0 1 129.6 454.32 Tm +(messages,) Tj +1 0 0 1 172.92 454.32 Tm +(delay) Tj +1 0 0 1 197.64 454.32 Tm +(them,) Tj +1 0 0 1 223.32 454.32 Tm +(duplicate) Tj +1 0 0 1 263.039 454.32 Tm +(them,) Tj +1 0 0 1 288.719 454.32 Tm +(or) Tj +1 0 0 1 72 442.32 Tm +(deli) Tj +1 0 0 1 86.76 442.32 Tm +(v) Tj +1 0 0 1 91.68 442.32 Tm +(er) Tj +1 0 0 1 101.76 442.32 Tm +(them) Tj +1 0 0 1 124.2 442.32 Tm +(out) Tj +1 0 0 1 139.44 442.32 Tm +(of) Tj +1 0 0 1 150.24 442.32 Tm +(order) Tj +1 0 0 1 170.88 442.32 Tm +(.) Tj +1 0 0 1 81.96 430.08 Tm +(W) Tj +1 0 0 1 90.48 430.08 Tm +(e) Tj +1 0 0 1 98.64 430.08 Tm +(use) Tj +1 0 0 1 115.56 430.08 Tm +(a) Tj +1 0 0 1 123.6 430.08 Tm +(Byzantine) Tj +1 0 0 1 167.64 430.08 Tm +(f) Tj +1 0 0 1 170.88 430.08 Tm +(ailure) Tj +1 0 0 1 197.16 430.08 Tm +(model,) Tj +1 0 0 1 228.359 430.08 Tm +(i.e.,) Tj +1 0 0 1 246.959 430.08 Tm +(f) Tj +1 0 0 1 250.199 430.08 Tm +(aulty) Tj +1 0 0 1 273.719 430.08 Tm +(nodes) Tj +1 0 0 1 72 418.08 Tm +(may) Tj +1 0 0 1 93.6 418.08 Tm +(beha) Tj +1 0 0 1 112.32 418.08 Tm +(v) Tj +1 0 0 1 117.24 418.08 Tm +(e) Tj +1 0 0 1 126 418.08 Tm +(arbitrarily) Tj +1 0 0 1 165.48 418.08 Tm +(,) Tj +1 0 0 1 172.68 418.08 Tm +(subject) Tj +1 0 0 1 205.44 418.08 Tm +(only) Tj +1 0 0 1 227.64 418.08 Tm +(to) Tj +1 0 0 1 239.88 418.08 Tm +(the) Tj +1 0 0 1 256.559 418.08 Tm +(restriction) Tj +1 0 0 1 72 406.08 Tm +(mentioned) Tj +1 0 0 1 116.28 406.08 Tm +(belo) Tj +1 0 0 1 133.32 406.08 Tm +(w) Tj +1 0 0 1 139.92 406.08 Tm +(.) Tj +1 0 0 1 145.68 406.08 Tm +(W) Tj +1 0 0 1 154.2 406.08 Tm +(e) Tj +1 0 0 1 160.8 406.08 Tm +(assume) Tj +1 0 0 1 192.36 406.08 Tm +(independent) Tj +1 0 0 1 243.24 406.08 Tm +(node) Tj +1 0 0 1 264.719 406.08 Tm +(f) Tj +1 0 0 1 267.959 406.08 Tm +(ailures.) Tj +1 0 0 1 72 394.2 Tm +(F) Tj +1 0 0 1 77.4 394.2 Tm +(or) Tj +1 0 0 1 87.3599 394.2 Tm +(this) Tj +1 0 0 1 103.44 394.2 Tm +(assumption) Tj +1 0 0 1 150.6 394.2 Tm +(to) Tj +1 0 0 1 160.08 394.2 Tm +(be) Tj +1 0 0 1 171.12 394.2 Tm +(true) Tj +1 0 0 1 188.4 394.2 Tm +(in) Tj +1 0 0 1 197.76 394.2 Tm +(the) Tj +1 0 0 1 211.679 394.2 Tm +(presence) Tj +1 0 0 1 248.279 394.2 Tm +(of) Tj +1 0 0 1 258.239 394.2 Tm +(malicious) Tj +1 0 0 1 72 382.2 Tm +(attacks,) Tj +1 0 0 1 106.8 382.2 Tm +(some) Tj +1 0 0 1 132.12 382.2 Tm +(steps) Tj +1 0 0 1 156.24 382.2 Tm +(need) Tj +1 0 0 1 179.28 382.2 Tm +(to) Tj +1 0 0 1 191.28 382.2 Tm +(be) Tj +1 0 0 1 204.96 382.2 Tm +(tak) Tj +1 0 0 1 217.08 382.2 Tm +(en,) Tj +1 0 0 1 233.64 382.2 Tm +(e.g.,) Tj +1 0 0 1 255.119 382.2 Tm +(each) Tj +1 0 0 1 277.679 382.2 Tm +(node) Tj +1 0 0 1 72 370.32 Tm +(should) Tj +1 0 0 1 101.04 370.32 Tm +(run) Tj +1 0 0 1 116.76 370.32 Tm +(dif) Tj +1 0 0 1 127.68 370.32 Tm +(ferent) Tj +1 0 0 1 153.24 370.32 Tm +(implementations) Tj +1 0 0 1 222.12 370.32 Tm +(of) Tj +1 0 0 1 232.919 370.32 Tm +(the) Tj +1 0 0 1 247.559 370.32 Tm +(service) Tj +1 0 0 1 278.159 370.32 Tm +(code) Tj +1 0 0 1 72 358.32 Tm +(and) Tj +1 0 0 1 90.36 358.32 Tm +(operating) Tj +1 0 0 1 131.88 358.32 Tm +(system) Tj +1 0 0 1 163.44 358.32 Tm +(and) Tj +1 0 0 1 181.8 358.32 Tm +(should) Tj +1 0 0 1 212.28 358.32 Tm +(ha) Tj +1 0 0 1 221.52 358.32 Tm +(v) Tj +1 0 0 1 226.44 358.32 Tm +(e) Tj +1 0 0 1 234.6 358.32 Tm +(a) Tj +1 0 0 1 243 358.32 Tm +(dif) Tj +1 0 0 1 253.92 358.32 Tm +(ferent) Tj +1 0 0 1 280.919 358.32 Tm +(root) Tj +1 0 0 1 72 346.44 Tm +(passw) Tj +1 0 0 1 96.2399 346.44 Tm +(ord) Tj +1 0 0 1 113.76 346.44 Tm +(and) Tj +1 0 0 1 132.36 346.44 Tm +(a) Tj +1 0 0 1 141 346.44 Tm +(dif) Tj +1 0 0 1 151.92 346.44 Tm +(ferent) Tj +1 0 0 1 179.28 346.44 Tm +(administrator) Tj +1 0 0 1 232.08 346.44 Tm +(.) Tj +1 0 0 1 243.119 346.44 Tm +(It) Tj +1 0 0 1 253.439 346.44 Tm +(is) Tj +1 0 0 1 264.359 346.44 Tm +(possible) Tj +1 0 0 1 72 334.44 Tm +(to) Tj +1 0 0 1 82.8 334.44 Tm +(obtain) Tj +1 0 0 1 110.64 334.44 Tm +(dif) Tj +1 0 0 1 121.56 334.44 Tm +(ferent) Tj +1 0 0 1 147.72 334.44 Tm +(implementations) Tj +1 0 0 1 217.2 334.44 Tm +(from) Tj +1 0 0 1 239.519 334.44 Tm +(the) Tj +1 0 0 1 254.759 334.44 Tm +(same) Tj +1 0 0 1 278.159 334.44 Tm +(code) Tj +1 0 0 1 72 322.56 Tm +(base) Tj +1 0 0 1 92.2799 322.56 Tm +([28) Tj +1 0 0 1 105.6 322.56 Tm +(]) Tj +1 0 0 1 111.48 322.56 Tm +(and) Tj +1 0 0 1 128.4 322.56 Tm +(for) Tj +1 0 0 1 142.68 322.56 Tm +(lo) Tj +1 0 0 1 150.24 322.56 Tm +(w) Tj +1 0 0 1 159.96 322.56 Tm +(de) Tj +1 0 0 1 169.32 322.56 Tm +(grees) Tj +1 0 0 1 192.84 322.56 Tm +(of) Tj +1 0 0 1 203.64 322.56 Tm +(replication) Tj +1 0 0 1 248.879 322.56 Tm +(one) Tj +1 0 0 1 265.799 322.56 Tm +(can) Tj +1 0 0 1 282.239 322.56 Tm +(b) Tj +1 0 0 1 287.039 322.56 Tm +(uy) Tj +1 0 0 1 72 310.56 Tm +(operating) Tj +1 0 0 1 114.36 310.56 Tm +(systems) Tj +1 0 0 1 150.48 310.56 Tm +(from) Tj +1 0 0 1 174.6 310.56 Tm +(dif) Tj +1 0 0 1 185.52 310.56 Tm +(ferent) Tj +1 0 0 1 213.24 310.56 Tm +(v) Tj +1 0 0 1 218.16 310.56 Tm +(endors.) Tj +1 0 0 1 257.279 310.56 Tm +(N-v) Tj +1 0 0 1 272.759 310.56 Tm +(ersion) Tj +1 0 0 1 72 298.68 Tm +(programming,) Tj +1 0 0 1 136.32 298.68 Tm +(i.e.,) Tj +1 0 0 1 158.52 298.68 Tm +(dif) Tj +1 0 0 1 169.44 298.68 Tm +(ferent) Tj +1 0 0 1 199.2 298.68 Tm +(teams) Tj +1 0 0 1 228.96 298.68 Tm +(of) Tj +1 0 0 1 243.84 298.68 Tm +(programmers) Tj +1 0 0 1 72 286.68 Tm +(produce) Tj +1 0 0 1 106.8 286.68 Tm +(dif) Tj +1 0 0 1 117.72 286.68 Tm +(ferent) Tj +1 0 0 1 143.52 286.68 Tm +(implementations,) Tj +1 0 0 1 215.28 286.68 Tm +(is) Tj +1 0 0 1 224.64 286.68 Tm +(another) Tj +1 0 0 1 257.159 286.68 Tm +(option) Tj +1 0 0 1 285.359 286.68 Tm +(for) Tj +1 0 0 1 72 274.8 Tm +(some) Tj +1 0 0 1 95.5199 274.8 Tm +(services.) Tj +1 0 0 1 81.96 262.44 Tm +(W) Tj +1 0 0 1 90.48 262.44 Tm +(e) Tj +1 0 0 1 98.16 262.44 Tm +(use) Tj +1 0 0 1 114.72 262.44 Tm +(cryptographic) Tj +1 0 0 1 173.28 262.44 Tm +(techniques) Tj +1 0 0 1 219 262.44 Tm +(to) Tj +1 0 0 1 230.04 262.44 Tm +(pre) Tj +1 0 0 1 242.64 262.44 Tm +(v) Tj +1 0 0 1 247.56 262.44 Tm +(ent) Tj +1 0 0 1 262.679 262.44 Tm +(spoo\256ng) Tj +1 0 0 1 72 250.56 Tm +(and) Tj +1 0 0 1 91.44 250.56 Tm +(replays) Tj +1 0 0 1 125.4 250.56 Tm +(and) Tj +1 0 0 1 144.84 250.56 Tm +(to) Tj +1 0 0 1 157.8 250.56 Tm +(detect) Tj +1 0 0 1 186.6 250.56 Tm +(corrupted) Tj +1 0 0 1 229.92 250.56 Tm +(messages.) Tj +1 0 0 1 281.519 250.56 Tm +(Our) Tj +1 0 0 1 72 238.56 Tm +(messages) Tj +1 0 0 1 114 238.56 Tm +(contain) Tj +1 0 0 1 147.72 238.56 Tm +(public-k) Tj +1 0 0 1 181.08 238.56 Tm +(e) Tj +1 0 0 1 185.4 238.56 Tm +(y) Tj +1 0 0 1 194.52 238.56 Tm +(signatures) Tj +1 0 0 1 239.279 238.56 Tm +([33) Tj +1 0 0 1 252.599 238.56 Tm +(],) Tj +1 0 0 1 263.279 238.56 Tm +(message) Tj +1 0 0 1 72 226.68 Tm +(authentication) Tj +1 0 0 1 130.8 226.68 Tm +(codes) Tj +1 0 0 1 155.76 226.68 Tm +([36) Tj +1 0 0 1 169.08 226.68 Tm +(],) Tj +1 0 0 1 177.24 226.68 Tm +(and) Tj +1 0 0 1 193.92 226.68 Tm +(message) Tj +1 0 0 1 229.92 226.68 Tm +(digests) Tj +1 0 0 1 259.919 226.68 Tm +(produced) Tj +1 0 0 1 72 214.68 Tm +(by) Tj +1 0 0 1 85.2 214.68 Tm +(collision-resistant) Tj +1 0 0 1 159.36 214.68 Tm +(hash) Tj +1 0 0 1 180.96 214.68 Tm +(functions) Tj +1 0 0 1 221.28 214.68 Tm +([32) Tj +1 0 0 1 234.6 214.68 Tm +(].) Tj +1 0 0 1 246.359 214.68 Tm +(W) Tj +1 0 0 1 254.879 214.68 Tm +(e) Tj +1 0 0 1 262.679 214.68 Tm +(denote) Tj +1 0 0 1 292.559 214.68 Tm +(a) Tj +1 0 0 1 72 202.68 Tm +(message) Tj +ET +endstream +endobj +59 0 obj +7978 +endobj +60 0 obj +<< /Type /XObject /Name /R60 /Subtype /Image /Length 61 0 R +/ImageMask true /Width 81 /Height 46 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 81 /BlackIs1 true >>] +>> +stream +4N`CV%M3$8612U>:Ag"NnFZb +endstream +endobj +61 0 obj +139 +endobj +62 0 obj +<< /Length 63 0 R >> +stream +q +8.1 0 0 -4.6 108.9 207.2 cm +/R60 Do +Q +BT +1 0 0 1 120.36 202.68 Tm +(signed) Tj +1 0 0 1 149.28 202.68 Tm +(by) Tj +1 0 0 1 162.12 202.68 Tm +(node) Tj +ET +endstream +endobj +63 0 obj +151 +endobj +64 0 obj +<< /Type /XObject /Name /R64 /Subtype /Image /Length 65 0 R +/ImageMask true /Width 26 /Height 69 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 26 /BlackIs1 true >>] +>> +stream +-R^*Zl/YrmIf:DfhtVWo^UNqc%K6'1s*addX2"DHrSQB/+4u(tp(mUa[G\kks2oQE"5iR%(OubkJ,~> +endstream +endobj +65 0 obj +80 +endobj +66 0 obj +<< /Length 67 0 R >> +stream +q +2.6 0 0 -6.9 184.6 209.5 cm +/R64 Do +Q +BT +1 0 0 1 190.68 202.68 Tm +(as) Tj +ET +endstream +endobj +67 0 obj +79 +endobj +68 0 obj +<< /Type /XObject /Name /R68 /Subtype /Image /Length 69 0 R +/ImageMask true /Width 24 /Height 99 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 24 /BlackIs1 true >>] +>> +stream ++t-07X8N5#0B"SKInTVGqHrM(f3_PN_YUJ4L&^prs)7cFqg[B")ufa0GPC"Nr'(,Kmt^kVHo1m8!<~> +endstream +endobj +69 0 obj +80 +endobj +70 0 obj +<< /Length 71 0 R >> +stream +q +2.4 0 0 -9.9 202.9 210.2 cm +/R68 Do +Q +q +8.1 0 0 -4.6 206 207.2 cm +/R60 Do +Q +endstream +endobj +71 0 obj +78 +endobj +72 0 obj +<< /Type /XObject /Name /R72 /Subtype /Image /Length 73 0 R +/ImageMask true /Width 24 /Height 99 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 24 /BlackIs1 true >>] +>> +stream +02pd&htA/cs0qnbrP.\`mJhV5_u&o]hqrq@J+[15ImJQWpe1HjlFsc8R~> +endstream +endobj +73 0 obj +78 +endobj +74 0 obj +<< /Length 75 0 R >> +stream +q +2.4 0 0 -9.9 215 210.2 cm +/R72 Do +Q +endstream +endobj +75 0 obj +38 +endobj +76 0 obj +<< /Type /XObject /Name /R76 /Subtype /Image /Length 77 0 R +/ImageMask true /Width 42 /Height 31 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 42 /BlackIs1 true >>] +>> +stream +1`R)4"qZ&4%Y>/aKS+]fJ#a+Sp>>oFIt)n@>7^\DYeoBbLCXeP2$%q%J-Z~> +endstream +endobj +77 0 obj +61 +endobj +78 0 obj +<< /Length 79 0 R >> +stream +q +4.2 0 0 -3.1 218.6 204.2 cm +/R76 Do +Q +endstream +endobj +79 0 obj +40 +endobj +80 0 obj +<< /Type /XObject /Name /R80 /Subtype /Image /Length 81 0 R +/ImageMask true /Width 17 /Height 35 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 17 /BlackIs1 true >>] +>> +stream +4B*=`S,;]mpj7Y*s1(">iS6Nohi?6cs!N02p(RKl"9~> +endstream +endobj +81 0 obj +45 +endobj +82 0 obj +<< /Length 83 0 R >> +stream +q +1.7 0 0 -3.5 223.3 203.8 cm +/R80 Do +Q +BT +1 0 0 1 229.56 202.68 Tm +(and) Tj +1 0 0 1 246.96 202.68 Tm +(the) Tj +1 0 0 1 261.96 202.68 Tm +(digest) Tj +1 0 0 1 288.72 202.68 Tm +(of) Tj +1 0 0 1 72 190.8 Tm +(message) Tj +ET +q +8.1 0 0 -4.6 109.2 195.3 cm +/R60 Do +Q +BT +1 0 0 1 120.84 190.8 Tm +(by) Tj +ET +endstream +endobj +83 0 obj +295 +endobj +84 0 obj +<< /Type /XObject /Name /R84 /Subtype /Image /Length 85 0 R +/ImageMask true /Width 75 /Height 69 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 75 /BlackIs1 true >>] +>> +stream +-D@i/="t3q:)'T:EY'Q7@JfCQK73IN +DqmDb@DDO(rqCj@Fs$I.iOf--It,63^V:<:qnMn*^\@b0s+Zm:ht=@d?EiAlYY3Kq%(IruTE5*9~> +endstream +endobj +85 0 obj +109 +endobj +86 0 obj +<< /Length 87 0 R >> +stream +q +7.5 0 0 -6.9 134.2 197.7 cm +/R84 Do +Q +endstream +endobj +87 0 obj +40 +endobj +88 0 obj +<< /Type /XObject /Name /R88 /Subtype /Image /Length 89 0 R +/ImageMask true /Width 24 /Height 99 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 24 /BlackIs1 true >>] +>> +stream ++rF]%*LO8IHL@*j8[taqp!<~> +endstream +endobj +89 0 obj +80 +endobj +90 0 obj +<< /Length 91 0 R >> +stream +q +2.4 0 0 -9.9 143.3 198.3 cm +/R88 Do +Q +q +8.1 0 0 -4.6 146.5 195.3 cm +/R60 Do +Q +endstream +endobj +91 0 obj +80 +endobj +92 0 obj +<< /Type /XObject /Name /R92 /Subtype /Image /Length 93 0 R +/ImageMask true /Width 24 /Height 99 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 24 /BlackIs1 true >>] +>> +stream +,JEO$loTRL4niC`hqS$`rnmKps6nkFp\Y!js6KW5s0(&RK)W@\(KK(`l\2@U?Bm&k<5,",!$D~> +endstream +endobj +93 0 obj +76 +endobj +94 0 obj +<< /Length 95 0 R >> +stream +q +2.4 0 0 -9.9 155.4 198.3 cm +/R92 Do +Q +BT +1 0 0 1 158.88 190.8 Tm +(.) Tj +1 0 0 1 167.04 190.8 Tm +(W) Tj +1 0 0 1 175.56 190.8 Tm +(e) Tj +1 0 0 1 183.24 190.8 Tm +(follo) Tj +1 0 0 1 201.96 190.8 Tm +(w) Tj +1 0 0 1 212.16 190.8 Tm +(the) Tj +1 0 0 1 227.4 190.8 Tm +(common) Tj +1 0 0 1 265.44 190.8 Tm +(practice) Tj +1 0 0 1 72 178.8 Tm +(of) Tj +1 0 0 1 84.6 178.8 Tm +(signing) Tj +1 0 0 1 118.32 178.8 Tm +(a) Tj +1 0 0 1 127.08 178.8 Tm +(digest) Tj +1 0 0 1 155.28 178.8 Tm +(of) Tj +1 0 0 1 167.88 178.8 Tm +(a) Tj +1 0 0 1 176.64 178.8 Tm +(message) Tj +1 0 0 1 214.8 178.8 Tm +(and) Tj +1 0 0 1 233.519 178.8 Tm +(appending) Tj +1 0 0 1 279.359 178.8 Tm +(it) Tj +1 0 0 1 289.199 178.8 Tm +(to) Tj +1 0 0 1 72 166.92 Tm +(the) Tj +1 0 0 1 87.3599 166.92 Tm +(plainte) Tj +1 0 0 1 114.48 166.92 Tm +(xt) Tj +1 0 0 1 125.16 166.92 Tm +(of) Tj +1 0 0 1 136.56 166.92 Tm +(the) Tj +1 0 0 1 151.92 166.92 Tm +(message) Tj +1 0 0 1 188.76 166.92 Tm +(rather) Tj +1 0 0 1 215.159 166.92 Tm +(than) Tj +1 0 0 1 235.439 166.92 Tm +(signing) Tj +1 0 0 1 267.839 166.92 Tm +(the) Tj +1 0 0 1 283.199 166.92 Tm +(full) Tj +1 0 0 1 72 154.92 Tm +(message) Tj +1 0 0 1 108.84 154.92 Tm +(\() Tj +ET +q +2.4 0 0 -9.9 113.1 162.4 cm +/R68 Do +Q +q +8.1 0 0 -4.6 116.2 159.4 cm +/R60 Do +Q +q +2.4 0 0 -9.9 125.2 162.4 cm +/R72 Do +Q +q +4.2 0 0 -3.1 128.8 156.5 cm +/R76 Do +Q +q +1.7 0 0 -3.5 133.5 156 cm +/R80 Do +Q +BT +1 0 0 1 140.04 154.92 Tm +(should) Tj +1 0 0 1 169.68 154.92 Tm +(be) Tj +1 0 0 1 182.16 154.92 Tm +(interpreted) Tj +1 0 0 1 228.36 154.92 Tm +(in) Tj +1 0 0 1 239.16 154.92 Tm +(this) Tj +1 0 0 1 256.68 154.92 Tm +(w) Tj +1 0 0 1 263.76 154.92 Tm +(ay\).) Tj +1 0 0 1 284.279 154.92 Tm +(All) Tj +1 0 0 1 72 143.04 Tm +(replicas) Tj +1 0 0 1 104.88 143.04 Tm +(kno) Tj +1 0 0 1 119.76 143.04 Tm +(w) Tj +1 0 0 1 128.64 143.04 Tm +(the) Tj +1 0 0 1 142.68 143.04 Tm +(others') Tj +1 0 0 1 172.2 143.04 Tm +(public) Tj +1 0 0 1 199.079 143.04 Tm +(k) Tj +1 0 0 1 203.999 143.04 Tm +(e) Tj +1 0 0 1 208.319 143.04 Tm +(ys) Tj +1 0 0 1 218.999 143.04 Tm +(to) Tj +1 0 0 1 228.599 143.04 Tm +(v) Tj +1 0 0 1 233.519 143.04 Tm +(erify) Tj +1 0 0 1 254.159 143.04 Tm +(signatures.) Tj +1 0 0 1 81.96 130.68 Tm +(W) Tj +1 0 0 1 90.48 130.68 Tm +(e) Tj +1 0 0 1 101.76 130.68 Tm +(allo) Tj +1 0 0 1 116.52 130.68 Tm +(w) Tj +1 0 0 1 130.44 130.68 Tm +(for) Tj +1 0 0 1 148.8 130.68 Tm +(a) Tj +1 0 0 1 160.08 130.68 Tm +(v) Tj +1 0 0 1 165 130.68 Tm +(ery) Tj +1 0 0 1 184.44 130.68 Tm +(strong) Tj +1 0 0 1 216.12 130.68 Tm +(adv) Tj +1 0 0 1 230.52 130.68 Tm +(ersary) Tj +1 0 0 1 261.479 130.68 Tm +(that) Tj +1 0 0 1 283.199 130.68 Tm +(can) Tj +1 0 0 1 72 118.8 Tm +(coordinate) Tj +1 0 0 1 117 118.8 Tm +(f) Tj +1 0 0 1 120.24 118.8 Tm +(aulty) Tj +1 0 0 1 143.16 118.8 Tm +(nodes,) Tj +1 0 0 1 171.96 118.8 Tm +(delay) Tj +1 0 0 1 196.56 118.8 Tm +(communication,) Tj +1 0 0 1 264.119 118.8 Tm +(or) Tj +1 0 0 1 275.399 118.8 Tm +(delay) Tj +1 0 0 1 72 106.8 Tm +(correct) Tj +1 0 0 1 102.96 106.8 Tm +(nodes) Tj +1 0 0 1 129.48 106.8 Tm +(in) Tj +1 0 0 1 140.52 106.8 Tm +(order) Tj +1 0 0 1 164.88 106.8 Tm +(to) Tj +1 0 0 1 175.92 106.8 Tm +(cause) Tj +1 0 0 1 201.36 106.8 Tm +(the) Tj +1 0 0 1 216.84 106.8 Tm +(most) Tj +1 0 0 1 239.52 106.8 Tm +(damage) Tj +1 0 0 1 273.72 106.8 Tm +(to) Tj +1 0 0 1 284.879 106.8 Tm +(the) Tj +1 0 0 1 72 94.92 Tm +(replicated) Tj +1 0 0 1 115.8 94.92 Tm +(service.) Tj +1 0 0 1 156.24 94.92 Tm +(W) Tj +1 0 0 1 164.76 94.92 Tm +(e) Tj +1 0 0 1 173.88 94.92 Tm +(do) Tj +1 0 0 1 188.4 94.92 Tm +(assume) Tj +1 0 0 1 222.24 94.92 Tm +(that) Tj +1 0 0 1 241.68 94.92 Tm +(the) Tj +1 0 0 1 258.479 94.92 Tm +(adv) Tj +1 0 0 1 272.879 94.92 Tm +(ersary) Tj +1 0 0 1 72 82.92 Tm +(cannot) Tj +1 0 0 1 101.28 82.92 Tm +(delay) Tj +1 0 0 1 125.52 82.92 Tm +(correct) Tj +1 0 0 1 155.88 82.92 Tm +(nodes) Tj +1 0 0 1 181.8 82.92 Tm +(inde\256nitely) Tj +1 0 0 1 226.8 82.92 Tm +(.) Tj +1 0 0 1 233.16 82.92 Tm +(W) Tj +1 0 0 1 241.68 82.92 Tm +(e) Tj +1 0 0 1 248.88 82.92 Tm +(also) Tj +1 0 0 1 267.719 82.92 Tm +(assume) Tj +1 0 0 1 72 71.04 Tm +(that) Tj +1 0 0 1 92.1599 71.04 Tm +(the) Tj +1 0 0 1 109.68 71.04 Tm +(adv) Tj +1 0 0 1 124.08 71.04 Tm +(ersary) Tj +1 0 0 1 153.48 71.04 Tm +(\(and) Tj +1 0 0 1 176.4 71.04 Tm +(the) Tj +1 0 0 1 193.92 71.04 Tm +(f) Tj +1 0 0 1 197.16 71.04 Tm +(aulty) Tj +1 0 0 1 222.24 71.04 Tm +(nodes) Tj +1 0 0 1 250.799 71.04 Tm +(it) Tj +1 0 0 1 261.599 71.04 Tm +(controls\)) Tj +1 0 0 1 315 709.08 Tm +(are) Tj +1 0 0 1 333.36 709.08 Tm +(computationally) Tj +1 0 0 1 404.28 709.08 Tm +(bound) Tj +1 0 0 1 435.24 709.08 Tm +(so) Tj +1 0 0 1 450.24 709.08 Tm +(that) Tj +1 0 0 1 471.36 709.08 Tm +(\(with) Tj +1 0 0 1 498.6 709.08 Tm +(v) Tj +1 0 0 1 503.52 709.08 Tm +(ery) Tj +1 0 0 1 522.239 709.08 Tm +(high) Tj +1 0 0 1 315 697.08 Tm +(probability\)) Tj +1 0 0 1 367.08 697.08 Tm +(it) Tj +1 0 0 1 377.52 697.08 Tm +(is) Tj +1 0 0 1 389.16 697.08 Tm +(unable) Tj +1 0 0 1 420.72 697.08 Tm +(to) Tj +1 0 0 1 433.44 697.08 Tm +(sub) Tj +1 0 0 1 447.24 697.08 Tm +(v) Tj +1 0 0 1 452.159 697.08 Tm +(ert) Tj +1 0 0 1 467.519 697.08 Tm +(the) Tj +1 0 0 1 484.679 697.08 Tm +(cryptographic) Tj +1 0 0 1 315 685.2 Tm +(techniques) Tj +1 0 0 1 359.64 685.2 Tm +(mentioned) Tj +1 0 0 1 403.8 685.2 Tm +(abo) Tj +1 0 0 1 418.2 685.2 Tm +(v) Tj +1 0 0 1 423.12 685.2 Tm +(e.) Tj +1 0 0 1 433.2 685.2 Tm +(F) Tj +1 0 0 1 438.6 685.2 Tm +(or) Tj +1 0 0 1 448.92 685.2 Tm +(e) Tj +1 0 0 1 453.24 685.2 Tm +(xample,) Tj +1 0 0 1 487.199 685.2 Tm +(the) Tj +1 0 0 1 501.479 685.2 Tm +(adv) Tj +1 0 0 1 515.879 685.2 Tm +(ersary) Tj +1 0 0 1 315 673.2 Tm +(cannot) Tj +1 0 0 1 345.24 673.2 Tm +(produce) Tj +1 0 0 1 381.12 673.2 Tm +(a) Tj +1 0 0 1 389.28 673.2 Tm +(v) Tj +1 0 0 1 394.08 673.2 Tm +(alid) Tj +1 0 0 1 412.68 673.2 Tm +(signature) Tj +1 0 0 1 452.879 673.2 Tm +(of) Tj +1 0 0 1 464.879 673.2 Tm +(a) Tj +1 0 0 1 473.039 673.2 Tm +(non-f) Tj +1 0 0 1 494.759 673.2 Tm +(aulty) Tj +1 0 0 1 518.159 673.2 Tm +(node,) Tj +1 0 0 1 315 661.32 Tm +(compute) Tj +1 0 0 1 352.92 661.32 Tm +(the) Tj +1 0 0 1 368.64 661.32 Tm +(information) Tj +1 0 0 1 419.28 661.32 Tm +(summarized) Tj +1 0 0 1 471.6 661.32 Tm +(by) Tj +1 0 0 1 485.16 661.32 Tm +(a) Tj +1 0 0 1 493.199 661.32 Tm +(digest) Tj +1 0 0 1 520.679 661.32 Tm +(from) Tj +1 0 0 1 315 649.32 Tm +(the) Tj +1 0 0 1 330.84 649.32 Tm +(digest,) Tj +1 0 0 1 361.2 649.32 Tm +(or) Tj +1 0 0 1 373.2 649.32 Tm +(\256nd) Tj +1 0 0 1 392.4 649.32 Tm +(tw) Tj +1 0 0 1 402.24 649.32 Tm +(o) Tj +1 0 0 1 410.88 649.32 Tm +(messages) Tj +1 0 0 1 452.279 649.32 Tm +(with) Tj +1 0 0 1 473.639 649.32 Tm +(the) Tj +1 0 0 1 489.479 649.32 Tm +(same) Tj +1 0 0 1 513.719 649.32 Tm +(digest.) Tj +1 0 0 1 315 637.44 Tm +(The) Tj +1 0 0 1 332.76 637.44 Tm +(cryptographic) Tj +1 0 0 1 390.24 637.44 Tm +(techniques) Tj +1 0 0 1 435.12 637.44 Tm +(we) Tj +1 0 0 1 448.92 637.44 Tm +(use) Tj +1 0 0 1 464.52 637.44 Tm +(are) Tj +1 0 0 1 478.919 637.44 Tm +(thought) Tj +1 0 0 1 511.559 637.44 Tm +(to) Tj +1 0 0 1 521.519 637.44 Tm +(ha) Tj +1 0 0 1 530.759 637.44 Tm +(v) Tj +1 0 0 1 535.679 637.44 Tm +(e) Tj +1 0 0 1 315 625.44 Tm +(these) Tj +1 0 0 1 337.92 625.44 Tm +(properties) Tj +1 0 0 1 380.28 625.44 Tm +([33) Tj +1 0 0 1 393.6 625.44 Tm +(,) Tj +1 0 0 1 398.52 625.44 Tm +(36) Tj +1 0 0 1 408.48 625.44 Tm +(,) Tj +1 0 0 1 413.52 625.44 Tm +(32) Tj +1 0 0 1 423.48 625.44 Tm +(].) Tj +/R57 12 Tf +1 0 0 1 315 600.96 Tm +(3) Tj +1 0 0 1 332.88 600.96 Tm +(Ser) Tj +1 0 0 1 349.92 600.96 Tm +(vice) Tj +1 0 0 1 372.96 600.96 Tm +(Pr) Tj +1 0 0 1 385.32 600.96 Tm +(operties) Tj +/R54 10 Tf +1 0 0 1 315 584.52 Tm +(Our) Tj +1 0 0 1 332.16 584.52 Tm +(algorithm) Tj +1 0 0 1 372.72 584.52 Tm +(can) Tj +1 0 0 1 388.2 584.52 Tm +(be) Tj +1 0 0 1 399.24 584.52 Tm +(used) Tj +1 0 0 1 419.16 584.52 Tm +(to) Tj +1 0 0 1 428.52 584.52 Tm +(implemen) Tj +1 0 0 1 468.48 584.52 Tm +(t) Tj +1 0 0 1 472.8 584.52 Tm +(an) Tj +1 0 0 1 482.16 584.52 Tm +(y) Tj +1 0 0 1 488.639 584.52 Tm +(deter) Tj +1 0 0 1 508.559 584.52 Tm +(ministic) Tj +1 0 0 1 315 572.64 Tm +(replicated) Tj +ET +endstream +endobj +95 0 obj +7671 +endobj +96 0 obj +<< /Type /Font /Name /R96 /Subtype /Type1 /BaseFont /Times-Italic >> +endobj +97 0 obj +<< /Length 98 0 R >> +stream +BT +/R96 10 Tf +1 0 0 1 356.76 572.64 Tm +(service) Tj +/R54 10 Tf +1 0 0 1 387.36 572.64 Tm +(with) Tj +1 0 0 1 407.52 572.64 Tm +(a) Tj +/R96 10 Tf +1 0 0 1 414.36 572.64 Tm +(state) Tj +/R54 10 Tf +1 0 0 1 435.72 572.64 Tm +(and) Tj +1 0 0 1 452.52 572.64 Tm +(some) Tj +/R96 10 Tf +1 0 0 1 475.919 572.64 Tm +(oper) Tj +1 0 0 1 494.159 572.64 Tm +(ations) Tj +/R54 10 Tf +1 0 0 1 518.639 572.64 Tm +(.) Tj +1 0 0 1 524.519 572.64 Tm +(The) Tj +1 0 0 1 315 560.64 Tm +(operations) Tj +1 0 0 1 359.28 560.64 Tm +(are) Tj +1 0 0 1 374.28 560.64 Tm +(not) Tj +1 0 0 1 389.76 560.64 Tm +(restricted) Tj +1 0 0 1 429.72 560.64 Tm +(to) Tj +1 0 0 1 440.28 560.64 Tm +(simple) Tj +1 0 0 1 469.56 560.64 Tm +(reads) Tj +1 0 0 1 493.439 560.64 Tm +(or) Tj +1 0 0 1 504.599 560.64 Tm +(writes) Tj +1 0 0 1 531.719 560.64 Tm +(of) Tj +1 0 0 1 315 548.76 Tm +(portions) Tj +1 0 0 1 350.76 548.76 Tm +(of) Tj +1 0 0 1 362.28 548.76 Tm +(the) Tj +1 0 0 1 377.52 548.76 Tm +(service) Tj +1 0 0 1 408.84 548.76 Tm +(state;) Tj +1 0 0 1 433.32 548.76 Tm +(the) Tj +1 0 0 1 445.44 548.76 Tm +(y) Tj +1 0 0 1 453.479 548.76 Tm +(can) Tj +1 0 0 1 470.519 548.76 Tm +(perform) Tj +1 0 0 1 505.679 548.76 Tm +(arbitrary) Tj +1 0 0 1 315 536.76 Tm +(deterministic) Tj +1 0 0 1 369.6 536.76 Tm +(computations) Tj +1 0 0 1 425.76 536.76 Tm +(using) Tj +1 0 0 1 449.88 536.76 Tm +(the) Tj +1 0 0 1 464.64 536.76 Tm +(state) Tj +1 0 0 1 485.399 536.76 Tm +(and) Tj +1 0 0 1 502.319 536.76 Tm +(operation) Tj +1 0 0 1 315 524.88 Tm +(ar) Tj +1 0 0 1 322.68 524.88 Tm +(guments.) Tj +1 0 0 1 362.16 524.88 Tm +(Clients) Tj +1 0 0 1 391.92 524.88 Tm +(issue) Tj +1 0 0 1 413.4 524.88 Tm +(requests) Tj +1 0 0 1 447.72 524.88 Tm +(to) Tj +1 0 0 1 457.079 524.88 Tm +(the) Tj +1 0 0 1 470.879 524.88 Tm +(replicated) Tj +1 0 0 1 511.919 524.88 Tm +(service) Tj +1 0 0 1 315 512.88 Tm +(to) Tj +1 0 0 1 325.8 512.88 Tm +(in) Tj +1 0 0 1 333.24 512.88 Tm +(v) Tj +1 0 0 1 338.04 512.88 Tm +(ok) Tj +1 0 0 1 348 512.88 Tm +(e) Tj +1 0 0 1 355.2 512.88 Tm +(operations) Tj +1 0 0 1 399.72 512.88 Tm +(and) Tj +1 0 0 1 417.12 512.88 Tm +(block) Tj +1 0 0 1 442.2 512.88 Tm +(w) Tj +1 0 0 1 449.279 512.88 Tm +(aiting) Tj +1 0 0 1 474.959 512.88 Tm +(for) Tj +1 0 0 1 489.599 512.88 Tm +(a) Tj +1 0 0 1 497.039 512.88 Tm +(reply) Tj +1 0 0 1 517.079 512.88 Tm +(.) Tj +1 0 0 1 524.519 512.88 Tm +(The) Tj +1 0 0 1 315 501 Tm +(replicated) Tj +1 0 0 1 357.12 501 Tm +(service) Tj +1 0 0 1 388.08 501 Tm +(is) Tj +1 0 0 1 397.44 501 Tm +(implemented) Tj +1 0 0 1 452.28 501 Tm +(by) Tj +ET +endstream +endobj +98 0 obj +2414 +endobj +99 0 obj +<< /Type /XObject /Name /R99 /Subtype /Image /Length 100 0 R +/ImageMask true /Width 54 /Height 46 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 54 /BlackIs1 true >>] +>> +stream +4N]t,$m>7rl/"o@d!re, +pYTohhg]lis4_o!aocV_^Yur>rVsXfR^`8 +endstream +endobj +100 0 obj +102 +endobj +101 0 obj +<< /Length 102 0 R >> +stream +q +5.4 0 0 -4.6 465.2 505.5 cm +/R99 Do +Q +BT +1 0 0 1 473.76 501 Tm +(replicas.) Tj +1 0 0 1 511.799 501 Tm +(Clients) Tj +1 0 0 1 315 489 Tm +(and) Tj +1 0 0 1 332.64 489 Tm +(replicas) Tj +1 0 0 1 366.96 489 Tm +(are) Tj +1 0 0 1 382.44 489 Tm +(non-f) Tj +1 0 0 1 404.16 489 Tm +(aulty) Tj +1 0 0 1 427.08 489 Tm +(if) Tj +1 0 0 1 436.44 489 Tm +(the) Tj +1 0 0 1 448.559 489 Tm +(y) Tj +1 0 0 1 456.719 489 Tm +(follo) Tj +1 0 0 1 475.439 489 Tm +(w) Tj +1 0 0 1 485.759 489 Tm +(the) Tj +1 0 0 1 501.239 489 Tm +(algorithm) Tj +1 0 0 1 315 477.12 Tm +(in) Tj +1 0 0 1 325.2 477.12 Tm +(Section) Tj +1 0 0 1 357.6 477.12 Tm +(4) Tj +1 0 0 1 365.04 477.12 Tm +(and) Tj +1 0 0 1 381.96 477.12 Tm +(if) Tj +1 0 0 1 390.6 477.12 Tm +(no) Tj +1 0 0 1 402.96 477.12 Tm +(attack) Tj +1 0 0 1 426.72 477.12 Tm +(er) Tj +1 0 0 1 436.919 477.12 Tm +(can) Tj +1 0 0 1 453.239 477.12 Tm +(for) Tj +1 0 0 1 464.879 477.12 Tm +(ge) Tj +1 0 0 1 476.519 477.12 Tm +(their) Tj +1 0 0 1 497.279 477.12 Tm +(signature.) Tj +1 0 0 1 324.96 465.12 Tm +(The) Tj +1 0 0 1 342.24 465.12 Tm +(algorithm) Tj +1 0 0 1 382.92 465.12 Tm +(pro) Tj +1 0 0 1 396.24 465.12 Tm +(vides) Tj +1 0 0 1 419.04 465.12 Tm +(b) Tj +1 0 0 1 423.96 465.12 Tm +(oth) Tj +/R96 10 Tf +1 0 0 1 438.36 465.12 Tm +(safety) Tj +/R54 10 Tf +1 0 0 1 463.319 465.12 Tm +(a) Tj +1 0 0 1 467.64 465.12 Tm +(nd) Tj +/R96 10 Tf +1 0 0 1 479.279 465.12 Tm +(livene) Tj +1 0 0 1 503.039 465.12 Tm +(ss) Tj +/R54 10 Tf +1 0 0 1 512.399 465.12 Tm +(assum-) Tj +1 0 0 1 315 453.24 Tm +(ing) Tj +1 0 0 1 329.76 453.24 Tm +(no) Tj +1 0 0 1 341.76 453.24 Tm +(more) Tj +1 0 0 1 364.32 453.24 Tm +(than) Tj +ET +endstream +endobj +102 0 obj +1556 +endobj +103 0 obj +<< /Type /XObject /Name /R103 /Subtype /Image /Length 104 0 R +/ImageMask true /Width 26 /Height 99 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 26 /BlackIs1 true >>] +>> +stream +/da'+AAiuts8W-!s8W-!s8W-!s8W-!s8W-!s8W-!s8W-!s8W-!qbR:n&-~> +endstream +endobj +104 0 obj +60 +endobj +105 0 obj +<< /Length 106 0 R >> +stream +q +2.6 0 0 -9.9 385.1 460.7 cm +/R103 Do +Q +endstream +endobj +106 0 obj +41 +endobj +107 0 obj +<< /Type /XObject /Name /R107 /Subtype /Image /Length 108 0 R +/ImageMask true /Width 43 /Height 33 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 43 /BlackIs1 true >>] +>> +stream +366ZtBKi$Kj)nf1pi$0"h;="Vq3Z%=rV-?Z\+qmGnGV5SiW&fRKsuPqJ89Nq=JdRb$l@h'!$D~> +endstream +endobj +108 0 obj +76 +endobj +109 0 obj +<< /Length 110 0 R >> +stream +q +4.3 0 0 -3.3 389.5 460.3 cm +/R107 Do +Q +endstream +endobj +110 0 obj +41 +endobj +111 0 obj +<< /Type /XObject /Name /R111 /Subtype /Image /Length 112 0 R +/ImageMask true /Width 47 /Height 5 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 47 /BlackIs1 true >>] +>> +stream +-D7RA!$D=~> +endstream +endobj +112 0 obj +12 +endobj +113 0 obj +<< /Length 114 0 R >> +stream +q +4.7 0 0 -0.5 394.8 459.1 cm +/R111 Do +Q +BT +/R54 7 Tf +1 0 0 1 400.32 457.08 Tm +(1) Tj +ET +q +14.64 0 0 -0.48 389.028 455.892 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +1 0 0 1 394.68 449.76 Tm +(3) Tj +ET +endstream +endobj +114 0 obj +216 +endobj +115 0 obj +<< /Type /XObject /Name /R115 /Subtype /Image /Length 116 0 R +/ImageMask true /Width 26 /Height 99 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 26 /BlackIs1 true >>] +>> +stream +,Ct/ij?\:qs8W-!s8W-!s8W-!s8W-!s8W-!s+[0J!$D~> +endstream +endobj +116 0 obj +46 +endobj +117 0 obj +<< /Length 118 0 R >> +stream +q +2.6 0 0 -9.9 405 460.7 cm +/R115 Do +Q +BT +/R54 10 Tf +1 0 0 1 411.36 453.24 Tm +(replicas) Tj +1 0 0 1 444.36 453.24 Tm +(are) Tj +1 0 0 1 458.64 453.24 Tm +(f) Tj +1 0 0 1 461.88 453.24 Tm +(aulty) Tj +1 0 0 1 481.32 453.24 Tm +(.) Tj +1 0 0 1 487.08 453.24 Tm +(Safety) Tj +1 0 0 1 514.56 453.24 Tm +(means) Tj +1 0 0 1 315 441.24 Tm +(that) Tj +1 0 0 1 334.44 441.24 Tm +(the) Tj +1 0 0 1 351.12 441.24 Tm +(replicated) Tj +1 0 0 1 394.8 441.24 Tm +(service) Tj +1 0 0 1 427.56 441.24 Tm +(satis\256es) Tj +1 0 0 1 463.559 441.24 Tm +(linearizability) Tj +1 0 0 1 523.439 441.24 Tm +([14) Tj +1 0 0 1 536.639 441.24 Tm +(]) Tj +1 0 0 1 315 429.24 Tm +(\(modi\256ed) Tj +1 0 0 1 356.4 429.24 Tm +(to) Tj +1 0 0 1 366.84 429.24 Tm +(account) Tj +1 0 0 1 400.44 429.24 Tm +(for) Tj +1 0 0 1 414.72 429.24 Tm +(Byzantine-f) Tj +1 0 0 1 461.88 429.24 Tm +(aulty) Tj +1 0 0 1 484.199 429.24 Tm +(clients) Tj +1 0 0 1 512.879 429.24 Tm +([4) Tj +1 0 0 1 521.159 429.24 Tm +(]\):) Tj +1 0 0 1 534.479 429.24 Tm +(it) Tj +1 0 0 1 315 417.36 Tm +(beha) Tj +1 0 0 1 333.72 417.36 Tm +(v) Tj +1 0 0 1 338.64 417.36 Tm +(es) Tj +1 0 0 1 349.92 417.36 Tm +(lik) Tj +1 0 0 1 360.36 417.36 Tm +(e) Tj +1 0 0 1 367.8 417.36 Tm +(a) Tj +1 0 0 1 375.36 417.36 Tm +(centralized) Tj +1 0 0 1 422.28 417.36 Tm +(implementation) Tj +1 0 0 1 487.919 417.36 Tm +(that) Tj +1 0 0 1 506.039 417.36 Tm +(e) Tj +1 0 0 1 510.359 417.36 Tm +(x) Tj +1 0 0 1 515.279 417.36 Tm +(ecutes) Tj +1 0 0 1 315 405.36 Tm +(operations) Tj +1 0 0 1 359.4 405.36 Tm +(atomically) Tj +1 0 0 1 404.4 405.36 Tm +(one) Tj +1 0 0 1 421.8 405.36 Tm +(at) Tj +1 0 0 1 431.88 405.36 Tm +(a) Tj +1 0 0 1 439.32 405.36 Tm +(time.) Tj +1 0 0 1 464.399 405.36 Tm +(Safety) Tj +1 0 0 1 492.839 405.36 Tm +(requires) Tj +1 0 0 1 527.879 405.36 Tm +(the) Tj +1 0 0 1 315 393.48 Tm +(bound) Tj +1 0 0 1 342.84 393.48 Tm +(on) Tj +1 0 0 1 355.68 393.48 Tm +(the) Tj +1 0 0 1 370.68 393.48 Tm +(number) Tj +1 0 0 1 404.04 393.48 Tm +(of) Tj +1 0 0 1 415.2 393.48 Tm +(f) Tj +1 0 0 1 418.44 393.48 Tm +(aulty) Tj +1 0 0 1 441.24 393.48 Tm +(replicas) Tj +1 0 0 1 475.079 393.48 Tm +(because) Tj +1 0 0 1 509.519 393.48 Tm +(a) Tj +1 0 0 1 516.839 393.48 Tm +(f) Tj +1 0 0 1 520.079 393.48 Tm +(aulty) Tj +1 0 0 1 315 381.48 Tm +(replica) Tj +1 0 0 1 344.16 381.48 Tm +(can) Tj +1 0 0 1 360.12 381.48 Tm +(beha) Tj +1 0 0 1 378.84 381.48 Tm +(v) Tj +1 0 0 1 383.76 381.48 Tm +(e) Tj +1 0 0 1 390 381.48 Tm +(arbitrarily) Tj +1 0 0 1 429.48 381.48 Tm +(,) Tj +1 0 0 1 433.92 381.48 Tm +(e.g.,) Tj +1 0 0 1 452.879 381.48 Tm +(it) Tj +1 0 0 1 460.559 381.48 Tm +(can) Tj +1 0 0 1 476.399 381.48 Tm +(destro) Tj +1 0 0 1 500.759 381.48 Tm +(y) Tj +1 0 0 1 507.719 381.48 Tm +(its) Tj +1 0 0 1 519.239 381.48 Tm +(state.) Tj +1 0 0 1 324.96 369.6 Tm +(Safety) Tj +1 0 0 1 355.68 369.6 Tm +(is) Tj +1 0 0 1 367.68 369.6 Tm +(pro) Tj +1 0 0 1 381 369.6 Tm +(vided) Tj +1 0 0 1 408.24 369.6 Tm +(re) Tj +1 0 0 1 415.92 369.6 Tm +(gardless) Tj +1 0 0 1 453.72 369.6 Tm +(of) Tj +1 0 0 1 467.4 369.6 Tm +(ho) Tj +1 0 0 1 477.24 369.6 Tm +(w) Tj +1 0 0 1 489.599 369.6 Tm +(man) Tj +1 0 0 1 506.759 369.6 Tm +(y) Tj +1 0 0 1 516.839 369.6 Tm +(f) Tj +1 0 0 1 520.079 369.6 Tm +(aulty) Tj +1 0 0 1 315 357.6 Tm +(clients) Tj +1 0 0 1 344.64 357.6 Tm +(are) Tj +1 0 0 1 360.48 357.6 Tm +(using) Tj +1 0 0 1 385.8 357.6 Tm +(the) Tj +1 0 0 1 401.64 357.6 Tm +(service) Tj +1 0 0 1 433.44 357.6 Tm +(\(e) Tj +1 0 0 1 440.999 357.6 Tm +(v) Tj +1 0 0 1 445.919 357.6 Tm +(en) Tj +1 0 0 1 458.879 357.6 Tm +(if) Tj +1 0 0 1 468.599 357.6 Tm +(the) Tj +1 0 0 1 480.719 357.6 Tm +(y) Tj +1 0 0 1 489.239 357.6 Tm +(collude) Tj +1 0 0 1 522.239 357.6 Tm +(with) Tj +1 0 0 1 315 345.72 Tm +(f) Tj +1 0 0 1 318.24 345.72 Tm +(aulty) Tj +1 0 0 1 339.84 345.72 Tm +(replicas\):) Tj +1 0 0 1 379.92 345.72 Tm +(all) Tj +1 0 0 1 391.44 345.72 Tm +(operations) Tj +1 0 0 1 434.76 345.72 Tm +(performed) Tj +1 0 0 1 478.199 345.72 Tm +(by) Tj +1 0 0 1 489.839 345.72 Tm +(f) Tj +1 0 0 1 493.079 345.72 Tm +(aulty) Tj +1 0 0 1 514.679 345.72 Tm +(clients) Tj +1 0 0 1 315 333.72 Tm +(are) Tj +1 0 0 1 330.72 333.72 Tm +(observ) Tj +1 0 0 1 357.36 333.72 Tm +(ed) Tj +1 0 0 1 370.08 333.72 Tm +(in) Tj +1 0 0 1 381.48 333.72 Tm +(a) Tj +1 0 0 1 389.52 333.72 Tm +(consistent) Tj +1 0 0 1 432.84 333.72 Tm +(w) Tj +1 0 0 1 439.92 333.72 Tm +(ay) Tj +1 0 0 1 452.999 333.72 Tm +(by) Tj +1 0 0 1 466.559 333.72 Tm +(non-f) Tj +1 0 0 1 488.279 333.72 Tm +(aulty) Tj +1 0 0 1 511.439 333.72 Tm +(clients.) Tj +1 0 0 1 315 321.84 Tm +(In) Tj +1 0 0 1 327.36 321.84 Tm +(particular) Tj +1 0 0 1 365.4 321.84 Tm +(,) Tj +1 0 0 1 372.12 321.84 Tm +(if) Tj +1 0 0 1 382.32 321.84 Tm +(the) Tj +1 0 0 1 398.64 321.84 Tm +(service) Tj +1 0 0 1 430.92 321.84 Tm +(operations) Tj +1 0 0 1 476.519 321.84 Tm +(are) Tj +1 0 0 1 492.719 321.84 Tm +(designed) Tj +1 0 0 1 532.199 321.84 Tm +(to) Tj +1 0 0 1 315 309.84 Tm +(preserv) Tj +1 0 0 1 344.4 309.84 Tm +(e) Tj +1 0 0 1 353.88 309.84 Tm +(some) Tj +1 0 0 1 380.16 309.84 Tm +(in) Tj +1 0 0 1 387.6 309.84 Tm +(v) Tj +1 0 0 1 392.4 309.84 Tm +(ariants) Tj +1 0 0 1 424.08 309.84 Tm +(on) Tj +1 0 0 1 439.2 309.84 Tm +(the) Tj +1 0 0 1 456.72 309.84 Tm +(service) Tj +1 0 0 1 490.199 309.84 Tm +(state,) Tj +1 0 0 1 516.839 309.84 Tm +(f) Tj +1 0 0 1 520.079 309.84 Tm +(aulty) Tj +1 0 0 1 315 297.96 Tm +(clients) Tj +1 0 0 1 343.56 297.96 Tm +(cannot) Tj +1 0 0 1 372.6 297.96 Tm +(break) Tj +1 0 0 1 397.2 297.96 Tm +(those) Tj +1 0 0 1 420.72 297.96 Tm +(in) Tj +1 0 0 1 428.16 297.96 Tm +(v) Tj +1 0 0 1 432.96 297.96 Tm +(ariants.) Tj +1 0 0 1 324.96 285.96 Tm +(The) Tj +1 0 0 1 345 285.96 Tm +(safety) Tj +1 0 0 1 373.44 285.96 Tm +(property) Tj +1 0 0 1 411.84 285.96 Tm +(is) Tj +1 0 0 1 423 285.96 Tm +(insuf) Tj +1 0 0 1 442.8 285.96 Tm +(\256cient) Tj +1 0 0 1 472.199 285.96 Tm +(to) Tj +1 0 0 1 484.559 285.96 Tm +(guard) Tj +1 0 0 1 511.799 285.96 Tm +(against) Tj +1 0 0 1 315 274.08 Tm +(f) Tj +1 0 0 1 318.24 274.08 Tm +(aulty) Tj +1 0 0 1 342.12 274.08 Tm +(clients,) Tj +1 0 0 1 374.88 274.08 Tm +(e.g.,) Tj +1 0 0 1 396.12 274.08 Tm +(in) Tj +1 0 0 1 407.88 274.08 Tm +(a) Tj +1 0 0 1 416.28 274.08 Tm +(\256le) Tj +1 0 0 1 432.959 274.08 Tm +(system) Tj +1 0 0 1 464.639 274.08 Tm +(a) Tj +1 0 0 1 472.919 274.08 Tm +(f) Tj +1 0 0 1 476.159 274.08 Tm +(aulty) Tj +1 0 0 1 500.039 274.08 Tm +(client) Tj +1 0 0 1 526.199 274.08 Tm +(can) Tj +1 0 0 1 315 262.08 Tm +(write) Tj +1 0 0 1 339.48 262.08 Tm +(garbage) Tj +1 0 0 1 375 262.08 Tm +(data) Tj +1 0 0 1 395.64 262.08 Tm +(to) Tj +1 0 0 1 407.4 262.08 Tm +(some) Tj +1 0 0 1 432.48 262.08 Tm +(shared) Tj +1 0 0 1 462.6 262.08 Tm +(\256le.) Tj +1 0 0 1 485.999 262.08 Tm +(Ho) Tj +1 0 0 1 497.999 262.08 Tm +(we) Tj +1 0 0 1 509.399 262.08 Tm +(v) Tj +1 0 0 1 514.319 262.08 Tm +(er) Tj +1 0 0 1 521.759 262.08 Tm +(,) Tj +1 0 0 1 528.359 262.08 Tm +(we) Tj +1 0 0 1 315 250.2 Tm +(limit) Tj +1 0 0 1 337.92 250.2 Tm +(the) Tj +1 0 0 1 354.24 250.2 Tm +(amount) Tj +1 0 0 1 388.32 250.2 Tm +(of) Tj +1 0 0 1 400.68 250.2 Tm +(damage) Tj +1 0 0 1 435.84 250.2 Tm +(a) Tj +1 0 0 1 444.36 250.2 Tm +(f) Tj +1 0 0 1 447.6 250.2 Tm +(aulty) Tj +1 0 0 1 471.719 250.2 Tm +(client) Tj +1 0 0 1 497.999 250.2 Tm +(can) Tj +1 0 0 1 515.879 250.2 Tm +(do) Tj +1 0 0 1 530.039 250.2 Tm +(by) Tj +1 0 0 1 315 238.2 Tm +(pro) Tj +1 0 0 1 328.32 238.2 Tm +(viding) Tj +1 0 0 1 358.08 238.2 Tm +(access) Tj +1 0 0 1 387.96 238.2 Tm +(control:) Tj +1 0 0 1 426.48 238.2 Tm +(we) Tj +1 0 0 1 442.56 238.2 Tm +(authenticate) Tj +1 0 0 1 495.119 238.2 Tm +(clients) Tj +1 0 0 1 525.599 238.2 Tm +(and) Tj +1 0 0 1 315 226.32 Tm +(den) Tj +1 0 0 1 329.4 226.32 Tm +(y) Tj +1 0 0 1 337.08 226.32 Tm +(access) Tj +1 0 0 1 365.52 226.32 Tm +(if) Tj +1 0 0 1 374.52 226.32 Tm +(the) Tj +1 0 0 1 389.52 226.32 Tm +(client) Tj +1 0 0 1 414.6 226.32 Tm +(issuing) Tj +1 0 0 1 445.68 226.32 Tm +(a) Tj +1 0 0 1 453 226.32 Tm +(request) Tj +1 0 0 1 484.679 226.32 Tm +(does) Tj +1 0 0 1 505.919 226.32 Tm +(not) Tj +1 0 0 1 521.519 226.32 Tm +(ha) Tj +1 0 0 1 530.759 226.32 Tm +(v) Tj +1 0 0 1 535.679 226.32 Tm +(e) Tj +1 0 0 1 315 214.32 Tm +(the) Tj +1 0 0 1 331.2 214.32 Tm +(right) Tj +1 0 0 1 354.12 214.32 Tm +(to) Tj +1 0 0 1 365.88 214.32 Tm +(in) Tj +1 0 0 1 373.32 214.32 Tm +(v) Tj +1 0 0 1 378.12 214.32 Tm +(ok) Tj +1 0 0 1 388.08 214.32 Tm +(e) Tj +1 0 0 1 396.36 214.32 Tm +(the) Tj +1 0 0 1 412.56 214.32 Tm +(operation.) Tj +1 0 0 1 460.919 214.32 Tm +(Also,) Tj +1 0 0 1 486.719 214.32 Tm +(services) Tj +1 0 0 1 522.839 214.32 Tm +(may) Tj +1 0 0 1 315 202.32 Tm +(pro) Tj +1 0 0 1 328.32 202.32 Tm +(vide) Tj +1 0 0 1 348.24 202.32 Tm +(operations) Tj +1 0 0 1 392.64 202.32 Tm +(to) Tj +1 0 0 1 403.32 202.32 Tm +(change) Tj +1 0 0 1 434.4 202.32 Tm +(the) Tj +1 0 0 1 449.52 202.32 Tm +(access) Tj +1 0 0 1 477.839 202.32 Tm +(permissions) Tj +1 0 0 1 528.359 202.32 Tm +(for) Tj +1 0 0 1 315 190.44 Tm +(a) Tj +1 0 0 1 322.68 190.44 Tm +(client.) Tj +1 0 0 1 352.92 190.44 Tm +(Since) Tj +1 0 0 1 378.24 190.44 Tm +(the) Tj +1 0 0 1 393.72 190.44 Tm +(algorithm) Tj +1 0 0 1 435.6 190.44 Tm +(ensures) Tj +1 0 0 1 468.719 190.44 Tm +(that) Tj +1 0 0 1 486.839 190.44 Tm +(the) Tj +1 0 0 1 502.199 190.44 Tm +(ef) Tj +1 0 0 1 509.759 190.44 Tm +(fects) Tj +1 0 0 1 531.719 190.44 Tm +(of) Tj +1 0 0 1 315 178.44 Tm +(access) Tj +1 0 0 1 342.24 178.44 Tm +(re) Tj +1 0 0 1 349.8 178.44 Tm +(v) Tj +1 0 0 1 354.6 178.44 Tm +(ocation) Tj +1 0 0 1 385.68 178.44 Tm +(operations) Tj +1 0 0 1 429 178.44 Tm +(are) Tj +1 0 0 1 443.04 178.44 Tm +(observ) Tj +1 0 0 1 469.679 178.44 Tm +(ed) Tj +1 0 0 1 480.599 178.44 Tm +(consistently) Tj +1 0 0 1 530.039 178.44 Tm +(by) Tj +1 0 0 1 315 166.56 Tm +(all) Tj +1 0 0 1 326.88 166.56 Tm +(clients,) Tj +1 0 0 1 357.36 166.56 Tm +(this) Tj +1 0 0 1 373.56 166.56 Tm +(pro) Tj +1 0 0 1 386.88 166.56 Tm +(vides) Tj +1 0 0 1 409.68 166.56 Tm +(a) Tj +1 0 0 1 415.92 166.56 Tm +(po) Tj +1 0 0 1 425.76 166.56 Tm +(werful) Tj +1 0 0 1 453.599 166.56 Tm +(mechanism) Tj +1 0 0 1 500.759 166.56 Tm +(to) Tj +1 0 0 1 510.479 166.56 Tm +(reco) Tj +1 0 0 1 527.639 166.56 Tm +(v) Tj +1 0 0 1 532.559 166.56 Tm +(er) Tj +1 0 0 1 315 154.56 Tm +(from) Tj +1 0 0 1 336.84 154.56 Tm +(attacks) Tj +1 0 0 1 366.96 154.56 Tm +(by) Tj +1 0 0 1 379.44 154.56 Tm +(f) Tj +1 0 0 1 382.68 154.56 Tm +(aulty) Tj +1 0 0 1 405.12 154.56 Tm +(clients.) Tj +1 0 0 1 324.96 142.68 Tm +(The) Tj +1 0 0 1 343.56 142.68 Tm +(algorithm) Tj +1 0 0 1 385.44 142.68 Tm +(does) Tj +1 0 0 1 406.8 142.68 Tm +(not) Tj +1 0 0 1 422.64 142.68 Tm +(rely) Tj +1 0 0 1 441.24 142.68 Tm +(on) Tj +1 0 0 1 454.319 142.68 Tm +(synchron) Tj +1 0 0 1 491.039 142.68 Tm +(y) Tj +1 0 0 1 498.839 142.68 Tm +(to) Tj +1 0 0 1 509.759 142.68 Tm +(pro) Tj +1 0 0 1 523.079 142.68 Tm +(vide) Tj +1 0 0 1 315 130.68 Tm +(safety) Tj +1 0 0 1 338.28 130.68 Tm +(.) Tj +1 0 0 1 346.2 130.68 Tm +(Therefore,) Tj +1 0 0 1 391.32 130.68 Tm +(it) Tj +1 0 0 1 399.96 130.68 Tm +(must) Tj +1 0 0 1 422.52 130.68 Tm +(rely) Tj +1 0 0 1 441.24 130.68 Tm +(on) Tj +1 0 0 1 454.319 130.68 Tm +(synchron) Tj +1 0 0 1 491.039 130.68 Tm +(y) Tj +1 0 0 1 498.839 130.68 Tm +(to) Tj +1 0 0 1 509.759 130.68 Tm +(pro) Tj +1 0 0 1 523.079 130.68 Tm +(vide) Tj +1 0 0 1 315 118.8 Tm +(li) Tj +1 0 0 1 320.28 118.8 Tm +(v) Tj +1 0 0 1 325.2 118.8 Tm +(eness;) Tj +1 0 0 1 357.72 118.8 Tm +(otherwise) Tj +1 0 0 1 402.72 118.8 Tm +(it) Tj +1 0 0 1 414.6 118.8 Tm +(could) Tj +1 0 0 1 443.04 118.8 Tm +(be) Tj +1 0 0 1 458.76 118.8 Tm +(used) Tj +1 0 0 1 483.36 118.8 Tm +(to) Tj +1 0 0 1 497.399 118.8 Tm +(implement) Tj +1 0 0 1 315 106.8 Tm +(consensus) Tj +1 0 0 1 360.84 106.8 Tm +(in) Tj +1 0 0 1 374.04 106.8 Tm +(an) Tj +1 0 0 1 388.8 106.8 Tm +(asynchronous) Tj +1 0 0 1 449.04 106.8 Tm +(system,) Tj +1 0 0 1 485.4 106.8 Tm +(which) Tj +1 0 0 1 515.159 106.8 Tm +(is) Tj +1 0 0 1 527.279 106.8 Tm +(not) Tj +1 0 0 1 315 94.92 Tm +(possible) Tj +1 0 0 1 354.36 94.92 Tm +([9) Tj +1 0 0 1 362.64 94.92 Tm +(].) Tj +1 0 0 1 384.6 94.92 Tm +(W) Tj +1 0 0 1 393.12 94.92 Tm +(e) Tj +1 0 0 1 404.28 94.92 Tm +(guarantee) Tj +1 0 0 1 449.64 94.92 Tm +(li) Tj +1 0 0 1 454.92 94.92 Tm +(v) Tj +1 0 0 1 459.84 94.92 Tm +(eness,) Tj +1 0 0 1 491.64 94.92 Tm +(i.e.,) Tj +1 0 0 1 513.959 94.92 Tm +(clients) Tj +1 0 0 1 315 82.92 Tm +(e) Tj +1 0 0 1 319.2 82.92 Tm +(v) Tj +1 0 0 1 324.12 82.92 Tm +(entually) Tj +1 0 0 1 359.64 82.92 Tm +(recei) Tj +1 0 0 1 378.84 82.92 Tm +(v) Tj +1 0 0 1 383.76 82.92 Tm +(e) Tj +1 0 0 1 391.68 82.92 Tm +(replies) Tj +1 0 0 1 421.8 82.92 Tm +(to) Tj +1 0 0 1 433.079 82.92 Tm +(their) Tj +1 0 0 1 454.919 82.92 Tm +(requests,) Tj +1 0 0 1 493.919 82.92 Tm +(pro) Tj +1 0 0 1 507.239 82.92 Tm +(vided) Tj +1 0 0 1 532.799 82.92 Tm +(at) Tj +1 0 0 1 315 71.04 Tm +(most) Tj +ET +q +2.6 0 0 -9.9 340.4 78.5 cm +/R103 Do +Q +q +4.3 0 0 -3.3 344.7 78.1 cm +/R107 Do +Q +q +4.7 0 0 -0.5 350 76.9 cm +/R111 Do +Q +BT +/R54 7 Tf +1 0 0 1 355.56 74.88 Tm +(1) Tj +ET +q +14.64 0 0 -0.48 344.388 73.692 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +1 0 0 1 349.92 67.56 Tm +(3) Tj +ET +q +2.6 0 0 -9.9 360.3 78.5 cm +/R115 Do +Q +BT +/R54 10 Tf +1 0 0 1 369 71.04 Tm +(replicas) Tj +1 0 0 1 404.4 71.04 Tm +(are) Tj +1 0 0 1 420.96 71.04 Tm +(f) Tj +1 0 0 1 424.2 71.04 Tm +(aulty) Tj +1 0 0 1 448.56 71.04 Tm +(and) Tj +/R96 10 Tf +1 0 0 1 467.28 71.04 Tm +(delay) Tj +ET +q +2.4 0 0 -9.9 489.8 78.5 cm +/R88 Do +Q +endstream +endobj +118 0 obj +12699 +endobj +119 0 obj +<< /Type /XObject /Name /R119 /Subtype /Image /Length 120 0 R +/ImageMask true /Width 31 /Height 64 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 31 /BlackIs1 true >>] +>> +stream +-R^*Z*BGB0H4Soj^)si_^$kpon\=tqs*c'f5C`EWjO>)lnj&fHn&,0\n,L/,TE5*9~> +endstream +endobj +120 0 obj +68 +endobj +121 0 obj +<< /Length 122 0 R >> +stream +q +3.1 0 0 -6.4 492.9 77.3 cm +/R119 Do +Q +q +2.4 0 0 -9.9 496.7 78.5 cm +/R92 Do +Q +BT +/R54 10 Tf +1 0 0 1 504.6 71.04 Tm +(does) Tj +1 0 0 1 527.279 71.04 Tm +(not) Tj +1 0 0 1 303.48 35.16 Tm +(2) Tj +ET +Q +endstream +endobj +122 0 obj +196 +endobj +123 0 obj +<< +/Type /Page +/MediaBox [0 0 612 792] +/Parent 2 0 R +/Resources << /ProcSet [/PDF /ImageB /Text] +/Font << +/R96 96 0 R +/R57 57 0 R +/R54 54 0 R +>> +/XObject << +/R92 92 0 R +/R119 119 0 R +/R88 88 0 R +/R115 115 0 R +/R111 111 0 R +/R107 107 0 R +/R103 103 0 R +/R99 99 0 R +/R80 80 0 R +/R76 76 0 R +/R72 72 0 R +/R60 60 0 R +/R68 68 0 R +/R84 84 0 R +/R64 64 0 R +/R50 50 0 R +>> +>> +/Contents [ +48 0 R +52 0 R +55 0 R +58 0 R +62 0 R +66 0 R +70 0 R +74 0 R +78 0 R +82 0 R +86 0 R +90 0 R +94 0 R +97 0 R +101 0 R +105 0 R +109 0 R +113 0 R +117 0 R +121 0 R +] +>> +endobj +124 0 obj +<< /Type /Font /Name /R124 /Subtype /Type1 /BaseFont /Times-Roman >> +endobj +125 0 obj +<< /Length 126 0 R >> +stream +q +BT +/R124 10 Tf +1 0 0 1 72 709.08 Tm +(gro) Tj +1 0 0 1 85.2 709.08 Tm +(w) Tj +1 0 0 1 96.7199 709.08 Tm +(f) Tj +1 0 0 1 99.9599 709.08 Tm +(aster) Tj +1 0 0 1 123.12 709.08 Tm +(than) Tj +ET +Q +q +W +0 0 612 792 re +n +endstream +endobj +126 0 obj +208 +endobj +127 0 obj +<< /Type /XObject /Name /R127 /Subtype /Image /Length 128 0 R +/ImageMask true /Width 31 /Height 64 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 31 /BlackIs1 true >>] +>> +stream +-R^*Z*BGB0H4Soj^)si_^$kpon\=tqs*c'f5C`EWjO>)lnj&fHn&,0\n,L/,TE5*9~> +endstream +endobj +128 0 obj +68 +endobj +129 0 obj +<< /Length 130 0 R >> +stream +q +3.1 0 0 -6.4 144.9 715.4 cm +/R127 Do +Q +BT +/R124 10 Tf +1 0 0 1 152.88 709.08 Tm +(inde\256nitely) Tj +1 0 0 1 197.88 709.08 Tm +(.) Tj +1 0 0 1 209.52 709.08 Tm +(Here,) Tj +ET +endstream +endobj +130 0 obj +172 +endobj +131 0 obj +<< /Type /Font /Name /R131 /Subtype /Type1 /BaseFont /Times-Italic >> +endobj +132 0 obj +<< /Length 133 0 R >> +stream +BT +/R131 10 Tf +1 0 0 1 236.4 709.08 Tm +(delay) Tj +ET +endstream +endobj +133 0 obj +53 +endobj +134 0 obj +<< /Type /XObject /Name /R134 /Subtype /Image /Length 135 0 R +/ImageMask true /Width 24 /Height 99 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 24 /BlackIs1 true >>] +>> +stream ++rF]%*LO8IHL@*j8[taqp!<~> +endstream +endobj +135 0 obj +80 +endobj +136 0 obj +<< /Length 137 0 R >> +stream +q +2.4 0 0 -9.9 258.8 716.6 cm +/R134 Do +Q +q +3.1 0 0 -6.4 261.9 715.4 cm +/R127 Do +Q +endstream +endobj +137 0 obj +82 +endobj +138 0 obj +<< /Type /XObject /Name /R138 /Subtype /Image /Length 139 0 R +/ImageMask true /Width 24 /Height 99 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 24 /BlackIs1 true >>] +>> +stream +,JEO$loTRL4niC`hqS$`rnmKps6nkFp\Y!js6KW5s0(&RK)W@\(KK(`l\2@U?Bm&k<5,",!$D~> +endstream +endobj +139 0 obj +76 +endobj +140 0 obj +<< /Length 141 0 R >> +stream +q +2.4 0 0 -9.9 265.7 716.6 cm +/R138 Do +Q +BT +/R124 10 Tf +1 0 0 1 273.719 709.08 Tm +(is) Tj +1 0 0 1 284.879 709.08 Tm +(the) Tj +1 0 0 1 72 697.08 Tm +(time) Tj +1 0 0 1 92.8799 697.08 Tm +(between) Tj +1 0 0 1 129.24 697.08 Tm +(the) Tj +1 0 0 1 144.6 697.08 Tm +(moment) Tj +ET +q +3.1 0 0 -6.4 180.6 703.4 cm +/R127 Do +Q +BT +1 0 0 1 187.08 697.08 Tm +(when) Tj +1 0 0 1 211.92 697.08 Tm +(a) Tj +1 0 0 1 219.48 697.08 Tm +(message) Tj +1 0 0 1 256.32 697.08 Tm +(is) Tj +1 0 0 1 266.159 697.08 Tm +(sent) Tj +1 0 0 1 285.359 697.08 Tm +(for) Tj +1 0 0 1 72 685.2 Tm +(the) Tj +1 0 0 1 87.2399 685.2 Tm +(\256rst) Tj +1 0 0 1 105.84 685.2 Tm +(time) Tj +1 0 0 1 126.6 685.2 Tm +(and) Tj +1 0 0 1 144.12 685.2 Tm +(the) Tj +1 0 0 1 159.36 685.2 Tm +(moment) Tj +1 0 0 1 195.12 685.2 Tm +(when) Tj +1 0 0 1 219.72 685.2 Tm +(it) Tj +1 0 0 1 228.359 685.2 Tm +(is) Tj +1 0 0 1 238.079 685.2 Tm +(recei) Tj +1 0 0 1 257.279 685.2 Tm +(v) Tj +1 0 0 1 262.199 685.2 Tm +(ed) Tj +1 0 0 1 274.559 685.2 Tm +(by) Tj +1 0 0 1 287.639 685.2 Tm +(its) Tj +1 0 0 1 72 673.2 Tm +(destination) Tj +1 0 0 1 117.6 673.2 Tm +(\(assuming) Tj +1 0 0 1 160.56 673.2 Tm +(the) Tj +1 0 0 1 174.6 673.2 Tm +(sender) Tj +1 0 0 1 202.56 673.2 Tm +(k) Tj +1 0 0 1 207.479 673.2 Tm +(eeps) Tj +1 0 0 1 227.039 673.2 Tm +(retransmitting) Tj +1 0 0 1 284.879 673.2 Tm +(the) Tj +1 0 0 1 72 661.32 Tm +(message) Tj +1 0 0 1 108.6 661.32 Tm +(until) Tj +1 0 0 1 129.6 661.32 Tm +(it) Tj +1 0 0 1 138 661.32 Tm +(is) Tj +1 0 0 1 147.48 661.32 Tm +(recei) Tj +1 0 0 1 166.68 661.32 Tm +(v) Tj +1 0 0 1 171.6 661.32 Tm +(ed\).) Tj +1 0 0 1 191.16 661.32 Tm +(\(A) Tj +1 0 0 1 204.48 661.32 Tm +(more) Tj +1 0 0 1 227.76 661.32 Tm +(precise) Tj +1 0 0 1 258.839 661.32 Tm +(de\256nition) Tj +1 0 0 1 72 649.32 Tm +(can) Tj +1 0 0 1 89.4 649.32 Tm +(be) Tj +1 0 0 1 102.36 649.32 Tm +(found) Tj +1 0 0 1 129.24 649.32 Tm +(in) Tj +1 0 0 1 140.52 649.32 Tm +([4) Tj +1 0 0 1 148.8 649.32 Tm +(].\)) Tj +1 0 0 1 164.76 649.32 Tm +(This) Tj +1 0 0 1 186 649.32 Tm +(is) Tj +1 0 0 1 196.2 649.32 Tm +(a) Tj +1 0 0 1 204.24 649.32 Tm +(rather) Tj +1 0 0 1 230.999 649.32 Tm +(weak) Tj +1 0 0 1 255.599 649.32 Tm +(synchron) Tj +1 0 0 1 292.319 649.32 Tm +(y) Tj +1 0 0 1 72 637.44 Tm +(assumption) Tj +1 0 0 1 121.2 637.44 Tm +(that) Tj +1 0 0 1 139.92 637.44 Tm +(is) Tj +1 0 0 1 150.48 637.44 Tm +(lik) Tj +1 0 0 1 160.92 637.44 Tm +(ely) Tj +1 0 0 1 176.88 637.44 Tm +(to) Tj +1 0 0 1 188.4 637.44 Tm +(be) Tj +1 0 0 1 201.6 637.44 Tm +(true) Tj +1 0 0 1 220.92 637.44 Tm +(in) Tj +1 0 0 1 232.56 637.44 Tm +(an) Tj +1 0 0 1 241.92 637.44 Tm +(y) Tj +1 0 0 1 250.559 637.44 Tm +(real) Tj +1 0 0 1 269.279 637.44 Tm +(system) Tj +1 0 0 1 72 625.44 Tm +(pro) Tj +1 0 0 1 85.32 625.44 Tm +(vided) Tj +1 0 0 1 111.36 625.44 Tm +(netw) Tj +1 0 0 1 130.68 625.44 Tm +(ork) Tj +1 0 0 1 147.96 625.44 Tm +(f) Tj +1 0 0 1 151.2 625.44 Tm +(aults) Tj +1 0 0 1 174.12 625.44 Tm +(are) Tj +1 0 0 1 190.44 625.44 Tm +(e) Tj +1 0 0 1 194.64 625.44 Tm +(v) Tj +1 0 0 1 199.56 625.44 Tm +(entually) Tj +1 0 0 1 235.56 625.44 Tm +(repaired,) Tj +1 0 0 1 275.159 625.44 Tm +(yet) Tj +1 0 0 1 291.479 625.44 Tm +(it) Tj +1 0 0 1 72 613.44 Tm +(enables) Tj +1 0 0 1 104.4 613.44 Tm +(us) Tj +1 0 0 1 115.68 613.44 Tm +(to) Tj +1 0 0 1 126 613.44 Tm +(circumv) Tj +1 0 0 1 158.76 613.44 Tm +(ent) Tj +1 0 0 1 173.16 613.44 Tm +(the) Tj +1 0 0 1 187.8 613.44 Tm +(impossibility) Tj +1 0 0 1 242.279 613.44 Tm +(result) Tj +1 0 0 1 266.999 613.44 Tm +(in) Tj +1 0 0 1 277.199 613.44 Tm +([9) Tj +1 0 0 1 285.479 613.44 Tm +(].) Tj +1 0 0 1 81.96 601.56 Tm +(The) Tj +1 0 0 1 99.2399 601.56 Tm +(resilienc) Tj +1 0 0 1 132.96 601.56 Tm +(y) Tj +1 0 0 1 139.56 601.56 Tm +(of) Tj +1 0 0 1 149.52 601.56 Tm +(our) Tj +1 0 0 1 164.64 601.56 Tm +(algorithm) Tj +1 0 0 1 205.08 601.56 Tm +(is) Tj +1 0 0 1 213.479 601.56 Tm +(optimal:) Tj +1 0 0 1 249.959 601.56 Tm +(3) Tj +ET +endstream +endobj +141 0 obj +3700 +endobj +142 0 obj +<< /Type /XObject /Name /R142 /Subtype /Image /Length 143 0 R +/ImageMask true /Width 50 /Height 92 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 50 /BlackIs1 true >>] +>> +stream +5$eL7-b:K3p0<%? +h1u#R]2'Mug].5is7$!eqtL-hqu?EZs5/V3iQXZVs7/ihqu?EZs4QRbJe-ans*o2/4DdB_@6aqq5Q~> +endstream +endobj +143 0 obj +96 +endobj +144 0 obj +<< /Length 145 0 R >> +stream +q +5 0 0 -9.2 255.4 608.7 cm +/R142 Do +Q +endstream +endobj +145 0 obj +39 +endobj +146 0 obj +<< /Type /XObject /Name /R146 /Subtype /Image /Length 147 0 R +/ImageMask true /Width 66 /Height 65 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 66 /BlackIs1 true >>] +>> +stream +,BZ6Ns8W-!s8W-!s-Eq9Iaj?!s8W-!s8W-!s8Uf'#QTA~> +endstream +endobj +147 0 obj +47 +endobj +148 0 obj +<< /Length 149 0 R >> +stream +q +6.6 0 0 -6.5 262.1 607.4 cm +/R146 Do +Q +BT +1 0 0 1 269.999 601.56 Tm +(1) Tj +1 0 0 1 276.479 601.56 Tm +(is) Tj +1 0 0 1 284.879 601.56 Tm +(the) Tj +1 0 0 1 72 589.56 Tm +(minimum) Tj +1 0 0 1 112.92 589.56 Tm +(number) Tj +1 0 0 1 145.68 589.56 Tm +(of) Tj +1 0 0 1 156.12 589.56 Tm +(replicas) Tj +1 0 0 1 189.36 589.56 Tm +(that) Tj +1 0 0 1 206.52 589.56 Tm +(allo) Tj +1 0 0 1 221.28 589.56 Tm +(w) Tj +1 0 0 1 230.64 589.56 Tm +(an) Tj +1 0 0 1 242.159 589.56 Tm +(asynchronous) Tj +1 0 0 1 72 577.68 Tm +(system) Tj +1 0 0 1 101.64 577.68 Tm +(to) Tj +1 0 0 1 111.36 577.68 Tm +(pro) Tj +1 0 0 1 124.68 577.68 Tm +(vide) Tj +1 0 0 1 143.64 577.68 Tm +(the) Tj +1 0 0 1 157.8 577.68 Tm +(safety) Tj +1 0 0 1 183.6 577.68 Tm +(and) Tj +1 0 0 1 199.92 577.68 Tm +(li) Tj +1 0 0 1 205.2 577.68 Tm +(v) Tj +1 0 0 1 210.12 577.68 Tm +(eness) Tj +1 0 0 1 233.64 577.68 Tm +(properties) Tj +1 0 0 1 275.399 577.68 Tm +(when) Tj +1 0 0 1 72 565.68 Tm +(up) Tj +1 0 0 1 85.8 565.68 Tm +(to) Tj +ET +q +5 0 0 -9.2 97.8 572.8 cm +/R142 Do +Q +BT +1 0 0 1 107.28 565.68 Tm +(replicas) Tj +1 0 0 1 142.08 565.68 Tm +(are) Tj +1 0 0 1 158.04 565.68 Tm +(f) Tj +1 0 0 1 161.28 565.68 Tm +(aulty) Tj +1 0 0 1 185.04 565.68 Tm +(\(see) Tj +1 0 0 1 204.96 565.68 Tm +([2) Tj +1 0 0 1 213.24 565.68 Tm +(]) Tj +1 0 0 1 220.44 565.68 Tm +(for) Tj +1 0 0 1 235.919 565.68 Tm +(a) Tj +1 0 0 1 244.199 565.68 Tm +(proof\).) Tj +1 0 0 1 279.239 565.68 Tm +(This) Tj +1 0 0 1 72 553.8 Tm +(man) Tj +1 0 0 1 89.16 553.8 Tm +(y) Tj +1 0 0 1 96.8399 553.8 Tm +(replicas) Tj +1 0 0 1 130.56 553.8 Tm +(are) Tj +1 0 0 1 145.56 553.8 Tm +(needed) Tj +1 0 0 1 176.64 553.8 Tm +(because) Tj +1 0 0 1 210.96 553.8 Tm +(it) Tj +1 0 0 1 219.36 553.8 Tm +(must) Tj +1 0 0 1 241.559 553.8 Tm +(be) Tj +1 0 0 1 253.799 553.8 Tm +(possible) Tj +1 0 0 1 289.199 553.8 Tm +(to) Tj +1 0 0 1 72 541.8 Tm +(proceed) Tj +1 0 0 1 106.32 541.8 Tm +(after) Tj +1 0 0 1 127.44 541.8 Tm +(communicating) Tj +1 0 0 1 192.24 541.8 Tm +(with) Tj +ET +endstream +endobj +149 0 obj +1895 +endobj +150 0 obj +<< /Type /XObject /Name /R150 /Subtype /Image /Length 151 0 R +/ImageMask true /Width 54 /Height 46 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 54 /BlackIs1 true >>] +>> +stream +4N]t,$m>7rl/"o@d!re, +pYTohhg]lis4_o!aocV_^Yur>rVsXfR^`8 +endstream +endobj +151 0 obj +102 +endobj +152 0 obj +<< /Length 153 0 R >> +stream +q +5.4 0 0 -4.6 213 546.3 cm +/R150 Do +Q +endstream +endobj +153 0 obj +39 +endobj +154 0 obj +<< /Type /XObject /Name /R154 /Subtype /Image /Length 155 0 R +/ImageMask true /Width 61 /Height 5 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 61 /BlackIs1 true >>] +>> +stream +,Cs,bC&f@X~> +endstream +endobj +155 0 obj +13 +endobj +156 0 obj +<< /Length 157 0 R >> +stream +q +6.1 0 0 -0.5 222 544.6 cm +/R154 Do +Q +q +5 0 0 -9.2 231.8 548.9 cm +/R142 Do +Q +BT +1 0 0 1 240.12 541.8 Tm +(replicas,) Tj +1 0 0 1 276.479 541.8 Tm +(since) Tj +ET +q +5 0 0 -9.2 72.4 537 cm +/R142 Do +Q +BT +1 0 0 1 80.4 529.92 Tm +(replicas) Tj +1 0 0 1 113.76 529.92 Tm +(might) Tj +1 0 0 1 139.44 529.92 Tm +(be) Tj +1 0 0 1 151.2 529.92 Tm +(f) Tj +1 0 0 1 154.44 529.92 Tm +(aulty) Tj +1 0 0 1 176.76 529.92 Tm +(and) Tj +1 0 0 1 193.56 529.92 Tm +(not) Tj +1 0 0 1 208.68 529.92 Tm +(responding.) Tj +1 0 0 1 259.079 529.92 Tm +(Ho) Tj +1 0 0 1 271.079 529.92 Tm +(we) Tj +1 0 0 1 282.479 529.92 Tm +(v) Tj +1 0 0 1 287.399 529.92 Tm +(er) Tj +1 0 0 1 294.839 529.92 Tm +(,) Tj +1 0 0 1 72 517.92 Tm +(it) Tj +1 0 0 1 80.4 517.92 Tm +(is) Tj +1 0 0 1 89.9999 517.92 Tm +(possible) Tj +1 0 0 1 125.52 517.92 Tm +(that) Tj +1 0 0 1 143.4 517.92 Tm +(the) Tj +ET +q +5 0 0 -9.2 158.9 525 cm +/R142 Do +Q +BT +1 0 0 1 167.4 517.92 Tm +(replicas) Tj +1 0 0 1 201.24 517.92 Tm +(that) Tj +1 0 0 1 219.119 517.92 Tm +(did) Tj +1 0 0 1 234.719 517.92 Tm +(not) Tj +1 0 0 1 250.439 517.92 Tm +(respond) Tj +1 0 0 1 284.879 517.92 Tm +(are) Tj +1 0 0 1 72 506.04 Tm +(not) Tj +1 0 0 1 86.88 506.04 Tm +(f) Tj +1 0 0 1 90.1199 506.04 Tm +(aulty) Tj +1 0 0 1 112.2 506.04 Tm +(and,) Tj +1 0 0 1 131.28 506.04 Tm +(therefore,) Tj +ET +q +5 0 0 -9.2 172.2 513.1 cm +/R142 Do +Q +BT +1 0 0 1 180 506.04 Tm +(of) Tj +1 0 0 1 190.44 506.04 Tm +(those) Tj +1 0 0 1 213.6 506.04 Tm +(that) Tj +1 0 0 1 230.639 506.04 Tm +(responded) Tj +1 0 0 1 273.719 506.04 Tm +(might) Tj +1 0 0 1 72 494.04 Tm +(be) Tj +1 0 0 1 84.2399 494.04 Tm +(f) Tj +1 0 0 1 87.4799 494.04 Tm +(aulty) Tj +1 0 0 1 106.92 494.04 Tm +(.) Tj +1 0 0 1 113.76 494.04 Tm +(Ev) Tj +1 0 0 1 124.8 494.04 Tm +(en) Tj +1 0 0 1 136.92 494.04 Tm +(so,) Tj +1 0 0 1 151.2 494.04 Tm +(there) Tj +1 0 0 1 173.88 494.04 Tm +(must) Tj +1 0 0 1 196.08 494.04 Tm +(still) Tj +1 0 0 1 213.84 494.04 Tm +(be) Tj +1 0 0 1 226.079 494.04 Tm +(enough) Tj +1 0 0 1 258.239 494.04 Tm +(responses) Tj +1 0 0 1 72 482.16 Tm +(that) Tj +1 0 0 1 88.9199 482.16 Tm +(those) Tj +1 0 0 1 111.84 482.16 Tm +(from) Tj +1 0 0 1 133.2 482.16 Tm +(non-f) Tj +1 0 0 1 154.92 482.16 Tm +(aulty) Tj +1 0 0 1 176.52 482.16 Tm +(replicas) Tj +1 0 0 1 209.52 482.16 Tm +(outnumber) Tj +1 0 0 1 254.639 482.16 Tm +(those) Tj +1 0 0 1 277.679 482.16 Tm +(from) Tj +1 0 0 1 72 470.16 Tm +(f) Tj +1 0 0 1 75.24 470.16 Tm +(aulty) Tj +1 0 0 1 97.6799 470.16 Tm +(ones,) Tj +1 0 0 1 120.84 470.16 Tm +(i.e.,) Tj +ET +q +5.4 0 0 -4.6 138.2 474.7 cm +/R150 Do +Q +q +6.1 0 0 -0.5 147.1 473 cm +/R154 Do +Q +BT +1 0 0 1 156.24 470.16 Tm +(2) Tj +ET +q +5 0 0 -9.2 161.7 477.3 cm +/R142 Do +Q +endstream +endobj +157 0 obj +2543 +endobj +158 0 obj +<< /Type /XObject /Name /R158 /Subtype /Image /Length 159 0 R +/ImageMask true /Width 61 /Height 61 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 61 /BlackIs1 true >>] +>> +stream +,Ld57i8DMjL(5&>G[TII_hJX +endstream +endobj +159 0 obj +114 +endobj +160 0 obj +<< /Length 161 0 R >> +stream +q +6.1 0 0 -6.1 170.7 475.8 cm +/R158 Do +Q +q +5 0 0 -9.2 180.8 477.3 cm +/R142 Do +Q +BT +1 0 0 1 186.36 470.16 Tm +(.) Tj +1 0 0 1 192.48 470.16 Tm +(Therefore) Tj +ET +q +5.4 0 0 -4.6 234.4 474.7 cm +/R150 Do +Q +q +6.1 0 0 -6.1 243.8 475.8 cm +/R158 Do +Q +BT +1 0 0 1 253.439 470.16 Tm +(3) Tj +ET +q +5 0 0 -9.2 258.9 477.3 cm +/R142 Do +Q +BT +1 0 0 1 264.479 470.16 Tm +(.) Tj +1 0 0 1 81.96 458.28 Tm +(The) Tj +1 0 0 1 100.56 458.28 Tm +(algorithm) Tj +1 0 0 1 142.44 458.28 Tm +(does) Tj +1 0 0 1 163.8 458.28 Tm +(not) Tj +1 0 0 1 179.52 458.28 Tm +(address) Tj +1 0 0 1 212.52 458.28 Tm +(the) Tj +1 0 0 1 227.879 458.28 Tm +(problem) Tj +1 0 0 1 264.119 458.28 Tm +(of) Tj +1 0 0 1 275.519 458.28 Tm +(f) Tj +1 0 0 1 278.759 458.28 Tm +(ault-) Tj +1 0 0 1 72 446.28 Tm +(tolerant) Tj +1 0 0 1 104.76 446.28 Tm +(pri) Tj +1 0 0 1 115.68 446.28 Tm +(v) Tj +1 0 0 1 120.48 446.28 Tm +(ac) Tj +1 0 0 1 129.24 446.28 Tm +(y:) Tj +1 0 0 1 140.16 446.28 Tm +(a) Tj +1 0 0 1 146.88 446.28 Tm +(f) Tj +1 0 0 1 150.12 446.28 Tm +(aulty) Tj +1 0 0 1 172.32 446.28 Tm +(replica) Tj +1 0 0 1 201.6 446.28 Tm +(may) Tj +1 0 0 1 221.039 446.28 Tm +(leak) Tj +1 0 0 1 239.999 446.28 Tm +(information) Tj +1 0 0 1 289.199 446.28 Tm +(to) Tj +1 0 0 1 72 434.4 Tm +(an) Tj +1 0 0 1 83.1599 434.4 Tm +(attack) Tj +1 0 0 1 106.92 434.4 Tm +(er) Tj +1 0 0 1 114.12 434.4 Tm +(.) Tj +1 0 0 1 120 434.4 Tm +(It) Tj +1 0 0 1 127.8 434.4 Tm +(is) Tj +1 0 0 1 136.32 434.4 Tm +(not) Tj +1 0 0 1 150.84 434.4 Tm +(feasible) Tj +1 0 0 1 183.6 434.4 Tm +(to) Tj +1 0 0 1 193.2 434.4 Tm +(of) Tj +1 0 0 1 201.359 434.4 Tm +(fer) Tj +1 0 0 1 214.079 434.4 Tm +(f) Tj +1 0 0 1 217.319 434.4 Tm +(ault-tolerant) Tj +1 0 0 1 267.719 434.4 Tm +(pri) Tj +1 0 0 1 278.639 434.4 Tm +(v) Tj +1 0 0 1 283.439 434.4 Tm +(ac) Tj +1 0 0 1 292.199 434.4 Tm +(y) Tj +1 0 0 1 72 422.4 Tm +(in) Tj +1 0 0 1 85.44 422.4 Tm +(the) Tj +1 0 0 1 103.44 422.4 Tm +(general) Tj +1 0 0 1 138.48 422.4 Tm +(case) Tj +1 0 0 1 161.4 422.4 Tm +(because) Tj +1 0 0 1 198.6 422.4 Tm +(service) Tj +1 0 0 1 232.56 422.4 Tm +(operations) Tj +1 0 0 1 279.839 422.4 Tm +(may) Tj +1 0 0 1 72 410.52 Tm +(perform) Tj +1 0 0 1 105.96 410.52 Tm +(arbitrary) Tj +1 0 0 1 142.08 410.52 Tm +(compu) Tj +1 0 0 1 169.32 410.52 Tm +(tation) Tj +1 0 0 1 192 410.52 Tm +(s) Tj +1 0 0 1 197.4 410.52 Tm +(using) Tj +1 0 0 1 220.68 410.52 Tm +(th) Tj +1 0 0 1 228.36 410.52 Tm +(eir) Tj +1 0 0 1 240.479 410.52 Tm +(ar) Tj +1 0 0 1 248.039 410.52 Tm +(g) Tj +1 0 0 1 252.959 410.52 Tm +(um) Tj +1 0 0 1 265.679 410.52 Tm +(ents) Tj +1 0 0 1 283.319 410.52 Tm +(an) Tj +1 0 0 1 292.679 410.52 Tm +(d) Tj +1 0 0 1 72 398.52 Tm +(the) Tj +1 0 0 1 88.0799 398.52 Tm +(service) Tj +1 0 0 1 120.36 398.52 Tm +(state;) Tj +1 0 0 1 146.04 398.52 Tm +(replicas) Tj +1 0 0 1 180.96 398.52 Tm +(need) Tj +1 0 0 1 203.76 398.52 Tm +(this) Tj +1 0 0 1 222.12 398.52 Tm +(information) Tj +1 0 0 1 273.119 398.52 Tm +(in) Tj +1 0 0 1 284.879 398.52 Tm +(the) Tj +1 0 0 1 72 386.52 Tm +(clear) Tj +1 0 0 1 93.8399 386.52 Tm +(to) Tj +1 0 0 1 103.92 386.52 Tm +(e) Tj +1 0 0 1 108.24 386.52 Tm +(x) Tj +1 0 0 1 113.16 386.52 Tm +(ecute) Tj +1 0 0 1 136.56 386.52 Tm +(such) Tj +1 0 0 1 157.2 386.52 Tm +(operations) Tj +1 0 0 1 201.12 386.52 Tm +(ef) Tj +1 0 0 1 208.68 386.52 Tm +(\256ciently) Tj +1 0 0 1 240.839 386.52 Tm +(.) Tj +1 0 0 1 246.719 386.52 Tm +(It) Tj +1 0 0 1 255.239 386.52 Tm +(is) Tj +1 0 0 1 264.359 386.52 Tm +(possible) Tj +1 0 0 1 72 374.64 Tm +(to) Tj +1 0 0 1 82.08 374.64 Tm +(use) Tj +1 0 0 1 97.6799 374.64 Tm +(secret) Tj +1 0 0 1 123.24 374.64 Tm +(sharing) Tj +1 0 0 1 154.8 374.64 Tm +(schemes) Tj +1 0 0 1 190.92 374.64 Tm +([35) Tj +1 0 0 1 204.12 374.64 Tm +(]) Tj +1 0 0 1 209.76 374.64 Tm +(to) Tj +1 0 0 1 219.84 374.64 Tm +(obtain) Tj +1 0 0 1 247.079 374.64 Tm +(pri) Tj +1 0 0 1 257.999 374.64 Tm +(v) Tj +1 0 0 1 262.799 374.64 Tm +(ac) Tj +1 0 0 1 271.559 374.64 Tm +(y) Tj +1 0 0 1 278.639 374.64 Tm +(e) Tj +1 0 0 1 282.839 374.64 Tm +(v) Tj +1 0 0 1 287.759 374.64 Tm +(en) Tj +1 0 0 1 72 362.64 Tm +(in) Tj +1 0 0 1 82.56 362.64 Tm +(the) Tj +1 0 0 1 97.6799 362.64 Tm +(presence) Tj +1 0 0 1 135.36 362.64 Tm +(of) Tj +1 0 0 1 146.52 362.64 Tm +(a) Tj +1 0 0 1 153.84 362.64 Tm +(threshold) Tj +1 0 0 1 193.8 362.64 Tm +(of) Tj +1 0 0 1 204.96 362.64 Tm +(malicious) Tj +1 0 0 1 246.599 362.64 Tm +(replicas) Tj +1 0 0 1 280.439 362.64 Tm +([13) Tj +1 0 0 1 293.639 362.64 Tm +(]) Tj +1 0 0 1 72 350.76 Tm +(for) Tj +1 0 0 1 85.4399 350.76 Tm +(the) Tj +1 0 0 1 99.4799 350.76 Tm +(ar) Tj +1 0 0 1 107.16 350.76 Tm +(guments) Tj +1 0 0 1 142.56 350.76 Tm +(and) Tj +1 0 0 1 158.76 350.76 Tm +(portions) Tj +1 0 0 1 193.32 350.76 Tm +(of) Tj +1 0 0 1 203.4 350.76 Tm +(the) Tj +1 0 0 1 217.439 350.76 Tm +(state) Tj +1 0 0 1 237.479 350.76 Tm +(that) Tj +1 0 0 1 254.279 350.76 Tm +(are) Tj +1 0 0 1 268.199 350.76 Tm +(opaque) Tj +1 0 0 1 72 338.76 Tm +(to) Tj +1 0 0 1 83.28 338.76 Tm +(the) Tj +1 0 0 1 99.1199 338.76 Tm +(service) Tj +1 0 0 1 130.92 338.76 Tm +(operations.) Tj +1 0 0 1 181.68 338.76 Tm +(W) Tj +1 0 0 1 190.2 338.76 Tm +(e) Tj +1 0 0 1 198.36 338.76 Tm +(plan) Tj +1 0 0 1 218.999 338.76 Tm +(to) Tj +1 0 0 1 230.399 338.76 Tm +(in) Tj +1 0 0 1 237.839 338.76 Tm +(v) Tj +1 0 0 1 242.759 338.76 Tm +(estigate) Tj +1 0 0 1 276.479 338.76 Tm +(these) Tj +1 0 0 1 72 326.88 Tm +(techniques) Tj +1 0 0 1 117.12 326.88 Tm +(in) Tj +1 0 0 1 127.32 326.88 Tm +(the) Tj +1 0 0 1 141.96 326.88 Tm +(future.) Tj +ET +endstream +endobj +161 0 obj +5184 +endobj +162 0 obj +<< /Type /Font /Name /R162 /Subtype /Type1 /BaseFont /Times-Bold >> +endobj +163 0 obj +<< /Length 164 0 R >> +stream +BT +/R162 12 Tf +1 0 0 1 72 302.4 Tm +(4) Tj +1 0 0 1 89.8799 302.4 Tm +(The) Tj +1 0 0 1 112.8 302.4 Tm +(Algorithm) Tj +/R124 10 Tf +1 0 0 1 72 285.96 Tm +(Our) Tj +1 0 0 1 89.8799 285.96 Tm +(algorithm) Tj +1 0 0 1 130.92 285.96 Tm +(is) Tj +1 0 0 1 139.92 285.96 Tm +(a) Tj +1 0 0 1 146.76 285.96 Tm +(form) Tj +1 0 0 1 168.48 285.96 Tm +(of) Tj +/R131 10 Tf +1 0 0 1 179.16 285.96 Tm +(state) Tj +1 0 0 1 200.279 285.96 Tm +(mac) Tj +1 0 0 1 216.839 285.96 Tm +(hine) Tj +/R124 10 Tf +1 0 0 1 236.279 285.96 Tm +(replication) Tj +1 0 0 1 281.279 285.96 Tm +([17) Tj +1 0 0 1 294.479 285.96 Tm +(,) Tj +1 0 0 1 72 274.08 Tm +(34) Tj +1 0 0 1 81.96 274.08 Tm +(]:) Tj +1 0 0 1 94.4399 274.08 Tm +(the) Tj +1 0 0 1 110.52 274.08 Tm +(service) Tj +1 0 0 1 142.56 274.08 Tm +(is) Tj +1 0 0 1 153.12 274.08 Tm +(modeled) Tj +1 0 0 1 191.28 274.08 Tm +(as) Tj +1 0 0 1 203.4 274.08 Tm +(a) Tj +1 0 0 1 211.8 274.08 Tm +(state) Tj +1 0 0 1 233.879 274.08 Tm +(machine) Tj +1 0 0 1 271.559 274.08 Tm +(that) Tj +1 0 0 1 290.399 274.08 Tm +(is) Tj +1 0 0 1 72 262.08 Tm +(replicated) Tj +1 0 0 1 114 262.08 Tm +(across) Tj +1 0 0 1 141.6 262.08 Tm +(dif) Tj +1 0 0 1 152.52 262.08 Tm +(ferent) Tj +1 0 0 1 178.32 262.08 Tm +(nodes) Tj +1 0 0 1 204.24 262.08 Tm +(in) Tj +1 0 0 1 214.68 262.08 Tm +(a) Tj +1 0 0 1 221.76 262.08 Tm +(distrib) Tj +1 0 0 1 247.079 262.08 Tm +(uted) Tj +1 0 0 1 266.879 262.08 Tm +(system.) Tj +1 0 0 1 72 250.2 Tm +(Each) Tj +1 0 0 1 96 250.2 Tm +(state) Tj +1 0 0 1 118.32 250.2 Tm +(machine) Tj +1 0 0 1 156.12 250.2 Tm +(replica) Tj +1 0 0 1 187.32 250.2 Tm +(maintains) Tj +1 0 0 1 230.16 250.2 Tm +(the) Tj +1 0 0 1 246.479 250.2 Tm +(service) Tj +1 0 0 1 278.759 250.2 Tm +(state) Tj +1 0 0 1 72 238.2 Tm +(and) Tj +1 0 0 1 89.88 238.2 Tm +(implements) Tj +1 0 0 1 139.92 238.2 Tm +(the) Tj +1 0 0 1 155.64 238.2 Tm +(service) Tj +1 0 0 1 187.44 238.2 Tm +(operations.) Tj +1 0 0 1 238.2 238.2 Tm +(W) Tj +1 0 0 1 246.719 238.2 Tm +(e) Tj +1 0 0 1 254.759 238.2 Tm +(denote) Tj +1 0 0 1 284.879 238.2 Tm +(the) Tj +1 0 0 1 72 226.32 Tm +(set) Tj +1 0 0 1 86.5199 226.32 Tm +(of) Tj +1 0 0 1 98.2799 226.32 Tm +(replicas) Tj +1 0 0 1 132.72 226.32 Tm +(by) Tj +ET +endstream +endobj +164 0 obj +2073 +endobj +165 0 obj +<< /Type /XObject /Name /R165 /Subtype /Image /Length 166 0 R +/ImageMask true /Width 82 /Height 72 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 82 /BlackIs1 true >>] +>> +stream +-Y/ktlnh'J_1WlVIN>d;OF5k,Do-F.&+/,KrBCW.%mf0XjHsl[XfmH0hq-EWKY$j0J0P"&G5OH\ +%NI1uW$_fE0RLEdqg:VHE;eVks1FCucp-FBHiJ$ama`KH4`C&`>7V$N*[EFX"s^Mb8oRjd(Qr1T!"]/~> +endstream +endobj +166 0 obj +158 +endobj +167 0 obj +<< /Length 168 0 R >> +stream +q +8.2 0 0 -7.2 146.3 233.2 cm +/R165 Do +Q +BT +1 0 0 1 158.04 226.32 Tm +(and) Tj +1 0 0 1 175.8 226.32 Tm +(identify) Tj +1 0 0 1 210.239 226.32 Tm +(each) Tj +1 0 0 1 231.959 226.32 Tm +(replica) Tj +1 0 0 1 262.559 226.32 Tm +(using) Tj +1 0 0 1 287.639 226.32 Tm +(an) Tj +1 0 0 1 72 214.32 Tm +(inte) Tj +1 0 0 1 86.88 214.32 Tm +(ger) Tj +1 0 0 1 103.2 214.32 Tm +(in) Tj +ET +endstream +endobj +168 0 obj +361 +endobj +169 0 obj +<< /Type /XObject /Name /R169 /Subtype /Image /Length 170 0 R +/ImageMask true /Width 39 /Height 100 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 39 /BlackIs1 true >>] +>> +stream ++CYskJO*-9kn*)Wn:LN: +s8W-!s8W-!s4?PIKnQ+Ne;bBe<0A()%X^A[(i\@1^OQ:`s8W-!s8W,ur&u1hf`^Jj+bNVe!tYGt"9~> +endstream +endobj +170 0 obj +101 +endobj +171 0 obj +<< /Length 172 0 R >> +stream +q +3.9 0 0 -10 115.2 221.8 cm +/R169 Do +Q +BT +1 0 0 1 119.76 214.32 Tm +(0) Tj +ET +endstream +endobj +172 0 obj +78 +endobj +173 0 obj +<< /Type /XObject /Name /R173 /Subtype /Image /Length 174 0 R +/ImageMask true /Width 13 /Height 30 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 13 /BlackIs1 true >>] +>> +stream +0g^A/h=&*equ,LR2Xj(@lh_"t-iec'~> +endstream +endobj +174 0 obj +33 +endobj +175 0 obj +<< /Length 176 0 R >> +stream +q +1.3 0 0 -3 125.6 215.4 cm +/R173 Do +Q +endstream +endobj +176 0 obj +39 +endobj +177 0 obj +<< /Type /XObject /Name /R177 /Subtype /Image /Length 178 0 R +/ImageMask true /Width 12 /Height 11 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 12 /BlackIs1 true >>] +>> +stream +2W04\s8@WU'ECB$~> +endstream +endobj +178 0 obj +18 +endobj +179 0 obj +<< /Length 180 0 R >> +stream +q +1.2 0 0 -1.1 130 215.4 cm +/R177 Do +Q +q +1.2 0 0 -1.1 132.8 215.4 cm +/R177 Do +Q +q +1.2 0 0 -1.1 135.6 215.4 cm +/R177 Do +Q +q +1.3 0 0 -3 138.3 215.4 cm +/R173 Do +Q +endstream +endobj +180 0 obj +160 +endobj +181 0 obj +<< /Type /XObject /Name /R181 /Subtype /Image /Length 182 0 R +/ImageMask true /Width 6 /Height 100 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 6 /BlackIs1 true >>] +>> +stream +02qoNs8W-!s8W-!s8W-!s8W-!s8W-!s8Uf'#QTA~> +endstream +endobj +182 0 obj +42 +endobj +183 0 obj +<< /Length 184 0 R >> +stream +q +0.6 0 0 -10 143 221.8 cm +/R181 Do +Q +q +8.2 0 0 -7.2 144.8 221.2 cm +/R165 Do +Q +q +0.6 0 0 -10 154.1 221.8 cm +/R181 Do +Q +q +6.1 0 0 -0.5 159.6 217.1 cm +/R154 Do +Q +BT +1 0 0 1 169.44 214.32 Tm +(1) Tj +ET +endstream +endobj +184 0 obj +198 +endobj +185 0 obj +<< /Type /XObject /Name /R185 /Subtype /Image /Length 186 0 R +/ImageMask true /Width 39 /Height 100 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 39 /BlackIs1 true >>] +>> +stream +,DhM)&Fel!i:(XA +\*s@ss8W-!s8W-!hhT+`nDT#4)$`#/'GV2X"UH\]_@,]0s8W-!s8W-!s'MeW#sg*k*#HKf&-r8$"9~> +endstream +endobj +186 0 obj +96 +endobj +187 0 obj +<< /Length 188 0 R >> +stream +q +3.9 0 0 -10 175 221.8 cm +/R185 Do +Q +BT +1 0 0 1 179.52 214.32 Tm +(.) Tj +1 0 0 1 189.12 214.32 Tm +(F) Tj +1 0 0 1 194.52 214.32 Tm +(or) Tj +1 0 0 1 206.519 214.32 Tm +(simplicity) Tj +1 0 0 1 245.879 214.32 Tm +(,) Tj +1 0 0 1 252.359 214.32 Tm +(we) Tj +1 0 0 1 267.719 214.32 Tm +(assume) Tj +ET +q +0.6 0 0 -10 73 209.8 cm +/R181 Do +Q +q +8.2 0 0 -7.2 74.9 209.2 cm +/R165 Do +Q +q +0.6 0 0 -10 84.2 209.8 cm +/R181 Do +Q +endstream +endobj +188 0 obj +404 +endobj +189 0 obj +<< /Type /XObject /Name /R189 /Subtype /Image /Length 190 0 R +/ImageMask true /Width 66 /Height 27 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 66 /BlackIs1 true >>] +>> +stream +/e%CP4)j*Wq&BoPm6CdS5Q~> +endstream +endobj +190 0 obj +25 +endobj +191 0 obj +<< /Length 192 0 R >> +stream +q +6.6 0 0 -2.7 92.3 206.2 cm +/R189 Do +Q +BT +1 0 0 1 105.36 202.32 Tm +(3) Tj +ET +q +5 0 0 -9.2 110.8 209.4 cm +/R142 Do +Q +q +6.6 0 0 -6.5 120.3 208.1 cm +/R146 Do +Q +BT +1 0 0 1 130.92 202.32 Tm +(1) Tj +1 0 0 1 140.64 202.32 Tm +(where) Tj +ET +q +5 0 0 -9.2 170.1 209.4 cm +/R142 Do +Q +BT +1 0 0 1 180.24 202.32 Tm +(is) Tj +1 0 0 1 191.64 202.32 Tm +(the) Tj +1 0 0 1 208.44 202.32 Tm +(maximum) Tj +1 0 0 1 253.56 202.32 Tm +(number) Tj +1 0 0 1 288.719 202.32 Tm +(of) Tj +1 0 0 1 72 190.44 Tm +(replicas) Tj +1 0 0 1 107.88 190.44 Tm +(that) Tj +1 0 0 1 127.68 190.44 Tm +(may) Tj +1 0 0 1 149.76 190.44 Tm +(be) Tj +1 0 0 1 164.04 190.44 Tm +(f) Tj +1 0 0 1 167.28 190.44 Tm +(aulty;) Tj +1 0 0 1 196.08 190.44 Tm +(although) Tj +1 0 0 1 235.799 190.44 Tm +(there) Tj +1 0 0 1 260.519 190.44 Tm +(could) Tj +1 0 0 1 287.639 190.44 Tm +(be) Tj +1 0 0 1 72 178.44 Tm +(more) Tj +1 0 0 1 94.4399 178.44 Tm +(than) Tj +1 0 0 1 113.52 178.44 Tm +(3) Tj +ET +q +5 0 0 -9.2 119 185.5 cm +/R142 Do +Q +q +6.6 0 0 -6.5 126 184.2 cm +/R146 Do +Q +BT +1 0 0 1 134.4 178.44 Tm +(1) Tj +1 0 0 1 141.24 178.44 Tm +(replicas,) Tj +1 0 0 1 176.76 178.44 Tm +(the) Tj +1 0 0 1 190.92 178.44 Tm +(additional) Tj +1 0 0 1 232.679 178.44 Tm +(replicas) Tj +1 0 0 1 265.559 178.44 Tm +(de) Tj +1 0 0 1 274.919 178.44 Tm +(grade) Tj +1 0 0 1 72 166.56 Tm +(performance) Tj +1 0 0 1 125.04 166.56 Tm +(\(since) Tj +1 0 0 1 151.44 166.56 Tm +(more) Tj +1 0 0 1 174.6 166.56 Tm +(and) Tj +1 0 0 1 191.64 166.56 Tm +(bigger) Tj +1 0 0 1 219.72 166.56 Tm +(messages) Tj +1 0 0 1 260.039 166.56 Tm +(are) Tj +1 0 0 1 274.919 166.56 Tm +(being) Tj +1 0 0 1 72 154.56 Tm +(e) Tj +1 0 0 1 76.32 154.56 Tm +(xchanged\)) Tj +1 0 0 1 120.24 154.56 Tm +(without) Tj +1 0 0 1 153.24 154.56 Tm +(pro) Tj +1 0 0 1 166.56 154.56 Tm +(viding) Tj +1 0 0 1 194.28 154.56 Tm +(impro) Tj +1 0 0 1 218.16 154.56 Tm +(v) Tj +1 0 0 1 223.079 154.56 Tm +(ed) Tj +1 0 0 1 234.599 154.56 Tm +(resilienc) Tj +1 0 0 1 268.319 154.56 Tm +(y) Tj +1 0 0 1 272.759 154.56 Tm +(.) Tj +1 0 0 1 81.96 142.68 Tm +(The) Tj +1 0 0 1 100.2 142.68 Tm +(replicas) Tj +1 0 0 1 134.04 142.68 Tm +(mo) Tj +1 0 0 1 146.76 142.68 Tm +(v) Tj +1 0 0 1 151.68 142.68 Tm +(e) Tj +1 0 0 1 158.64 142.68 Tm +(through) Tj +1 0 0 1 192.36 142.68 Tm +(a) Tj +1 0 0 1 199.56 142.68 Tm +(succession) Tj +1 0 0 1 244.92 142.68 Tm +(of) Tj +1 0 0 1 256.079 142.68 Tm +(con\256gura-) Tj +1 0 0 1 72 130.68 Tm +(tions) Tj +1 0 0 1 94.4399 130.68 Tm +(called) Tj +/R131 10 Tf +1 0 0 1 121.32 130.68 Tm +(vie) Tj +1 0 0 1 132.84 130.68 Tm +(ws) Tj +/R124 10 Tf +1 0 0 1 143.28 130.68 Tm +(.) Tj +1 0 0 1 151.2 130.68 Tm +(In) Tj +1 0 0 1 162.6 130.68 Tm +(a) Tj +1 0 0 1 170.04 130.68 Tm +(vie) Tj +1 0 0 1 182.04 130.68 Tm +(w) Tj +1 0 0 1 192.239 130.68 Tm +(one) Tj +1 0 0 1 209.759 130.68 Tm +(replica) Tj +1 0 0 1 239.879 130.68 Tm +(is) Tj +1 0 0 1 249.599 130.68 Tm +(the) Tj +/R131 10 Tf +1 0 0 1 264.959 130.68 Tm +(primary) Tj +/R124 10 Tf +1 0 0 1 72 118.8 Tm +(and) Tj +1 0 0 1 90 118.8 Tm +(the) Tj +1 0 0 1 105.84 118.8 Tm +(others) Tj +1 0 0 1 133.92 118.8 Tm +(are) Tj +/R131 10 Tf +1 0 0 1 149.64 118.8 Tm +(bac) Tj +1 0 0 1 163.92 118.8 Tm +(kups) Tj +/R124 10 Tf +1 0 0 1 182.28 118.8 Tm +(.) Tj +1 0 0 1 191.76 118.8 Tm +(V) Tj +1 0 0 1 198.36 118.8 Tm +(ie) Tj +1 0 0 1 205.32 118.8 Tm +(ws) Tj +1 0 0 1 219.96 118.8 Tm +(are) Tj +1 0 0 1 235.8 118.8 Tm +(numbered) Tj +1 0 0 1 279.239 118.8 Tm +(con-) Tj +1 0 0 1 72 106.8 Tm +(secuti) Tj +1 0 0 1 95.0399 106.8 Tm +(v) Tj +1 0 0 1 99.9599 106.8 Tm +(ely) Tj +1 0 0 1 111.6 106.8 Tm +(.) Tj +1 0 0 1 119.04 106.8 Tm +(The) Tj +1 0 0 1 137.52 106.8 Tm +(primary) Tj +1 0 0 1 172.08 106.8 Tm +(of) Tj +1 0 0 1 183.36 106.8 Tm +(a) Tj +1 0 0 1 190.8 106.8 Tm +(vie) Tj +1 0 0 1 202.8 106.8 Tm +(w) Tj +1 0 0 1 213 106.8 Tm +(is) Tj +1 0 0 1 222.599 106.8 Tm +(replica) Tj +ET +endstream +endobj +192 0 obj +3645 +endobj +193 0 obj +<< /Type /XObject /Name /R193 /Subtype /Image /Length 194 0 R +/ImageMask true /Width 54 /Height 64 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 54 /BlackIs1 true >>] +>> +stream +-D98r4=0b"4s0Orro]!o&:3^Ke&7l2 +[/(7*rS^,RIfB4oLV,SG]>+(NrPSU5I/eEj^Aki-s4fFGn+Z5e95',@28).TTZ(Ru?\8Ii-iec'~> +endstream +endobj +194 0 obj +109 +endobj +195 0 obj +<< /Length 196 0 R >> +stream +q +5.4 0 0 -6.4 252.3 111.3 cm +/R193 Do +Q +BT +1 0 0 1 260.759 106.8 Tm +(such) Tj +1 0 0 1 281.999 106.8 Tm +(that) Tj +ET +q +5.4 0 0 -6.4 71.6 99.4 cm +/R193 Do +Q +q +6.6 0 0 -2.7 82.5 98.8 cm +/R189 Do +Q +endstream +endobj +196 0 obj +195 +endobj +197 0 obj +<< /Type /XObject /Name /R197 /Subtype /Image /Length 198 0 R +/ImageMask true /Width 44 /Height 46 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 44 /BlackIs1 true >>] +>> +stream +2$jBq +&9R,hkFhTE_G0fRqg[T's0obU5O\AciSX+=pUN/*r',Z@_Z&BgfC8fKf@@X?]DE=DTC?6WJ-Z.*~> +endstream +endobj +198 0 obj +84 +endobj +199 0 obj +<< /Length 200 0 R >> +stream +q +4.4 0 0 -4.6 94.9 99.4 cm +/R197 Do +Q +BT +/R131 10 Tf +1 0 0 1 104.76 94.92 Tm +(mod) Tj +ET +q +0.6 0 0 -10 127.8 102.4 cm +/R181 Do +Q +q +8.2 0 0 -7.2 129.7 101.8 cm +/R165 Do +Q +q +0.6 0 0 -10 139 102.4 cm +/R181 Do +Q +BT +/R124 10 Tf +1 0 0 1 140.76 94.92 Tm +(,) Tj +1 0 0 1 147.84 94.92 Tm +(where) Tj +ET +q +4.4 0 0 -4.6 176.4 99.4 cm +/R197 Do +Q +BT +1 0 0 1 185.52 94.92 Tm +(is) Tj +1 0 0 1 196.2 94.92 Tm +(the) Tj +1 0 0 1 212.4 94.92 Tm +(vie) Tj +1 0 0 1 224.4 94.92 Tm +(w) Tj +1 0 0 1 235.559 94.92 Tm +(number) Tj +1 0 0 1 265.679 94.92 Tm +(.) Tj +1 0 0 1 276.239 94.92 Tm +(V) Tj +1 0 0 1 282.839 94.92 Tm +(ie) Tj +1 0 0 1 289.799 94.92 Tm +(w) Tj +1 0 0 1 72 82.92 Tm +(changes) Tj +1 0 0 1 106.8 82.92 Tm +(are) Tj +1 0 0 1 121.68 82.92 Tm +(carried) Tj +1 0 0 1 151.92 82.92 Tm +(out) Tj +1 0 0 1 167.4 82.92 Tm +(when) Tj +1 0 0 1 191.64 82.92 Tm +(it) Tj +1 0 0 1 199.92 82.92 Tm +(appears) Tj +1 0 0 1 232.92 82.92 Tm +(that) Tj +1 0 0 1 250.559 82.92 Tm +(the) Tj +1 0 0 1 265.439 82.92 Tm +(primary) Tj +1 0 0 1 72 71.04 Tm +(has) Tj +1 0 0 1 87.24 71.04 Tm +(f) Tj +1 0 0 1 90.48 71.04 Tm +(ailed.) Tj +1 0 0 1 115.8 71.04 Tm +(V) Tj +1 0 0 1 122.4 71.04 Tm +(ie) Tj +1 0 0 1 129.36 71.04 Tm +(wstamped) Tj +1 0 0 1 171.72 71.04 Tm +(Replication) Tj +1 0 0 1 219.72 71.04 Tm +([26) Tj +1 0 0 1 233.04 71.04 Tm +(]) Tj +1 0 0 1 238.32 71.04 Tm +(and) Tj +1 0 0 1 254.759 71.04 Tm +(P) Tj +1 0 0 1 260.159 71.04 Tm +(axos) Tj +1 0 0 1 280.439 71.04 Tm +([18) Tj +1 0 0 1 293.639 71.04 Tm +(]) Tj +1 0 0 1 315 709.08 Tm +(used) Tj +1 0 0 1 336 709.08 Tm +(a) Tj +1 0 0 1 343.2 709.08 Tm +(similar) Tj +1 0 0 1 373.68 709.08 Tm +(approach) Tj +1 0 0 1 412.92 709.08 Tm +(to) Tj +1 0 0 1 423.36 709.08 Tm +(tolerate) Tj +1 0 0 1 456 709.08 Tm +(benign) Tj +1 0 0 1 485.879 709.08 Tm +(f) Tj +1 0 0 1 489.119 709.08 Tm +(aults) Tj +1 0 0 1 510.719 709.08 Tm +(\(as) Tj +1 0 0 1 524.999 709.08 Tm +(dis-) Tj +1 0 0 1 315 697.08 Tm +(cussed) Tj +1 0 0 1 344.04 697.08 Tm +(in) Tj +1 0 0 1 354.24 697.08 Tm +(Section) Tj +1 0 0 1 386.64 697.08 Tm +(8.\)) Tj +1 0 0 1 324.96 685.2 Tm +(The) Tj +1 0 0 1 342.96 685.2 Tm +(algorithm) Tj +1 0 0 1 384.24 685.2 Tm +(w) Tj +1 0 0 1 391.32 685.2 Tm +(orks) Tj +1 0 0 1 410.88 685.2 Tm +(roughly) Tj +1 0 0 1 444.36 685.2 Tm +(as) Tj +1 0 0 1 455.16 685.2 Tm +(follo) Tj +1 0 0 1 473.879 685.2 Tm +(ws:) Tj +1 0 0 1 317.52 673.92 Tm +(1.) Tj +1 0 0 1 330 673.92 Tm +(A) Tj +1 0 0 1 339.48 673.92 Tm +(client) Tj +1 0 0 1 363.96 673.92 Tm +(sends) Tj +1 0 0 1 388.44 673.92 Tm +(a) Tj +1 0 0 1 395.16 673.92 Tm +(request) Tj +1 0 0 1 426.24 673.92 Tm +(to) Tj +1 0 0 1 436.32 673.92 Tm +(in) Tj +1 0 0 1 443.759 673.92 Tm +(v) Tj +1 0 0 1 448.559 673.92 Tm +(ok) Tj +1 0 0 1 458.519 673.92 Tm +(e) Tj +1 0 0 1 465.119 673.92 Tm +(a) Tj +1 0 0 1 471.839 673.92 Tm +(service) Tj +1 0 0 1 502.319 673.92 Tm +(operation) Tj +1 0 0 1 330 662.04 Tm +(to) Tj +1 0 0 1 340.2 662.04 Tm +(the) Tj +1 0 0 1 354.84 662.04 Tm +(primary) Tj +1 0 0 1 317.52 648.48 Tm +(2.) Tj +1 0 0 1 330 648.48 Tm +(The) Tj +1 0 0 1 347.88 648.48 Tm +(primary) Tj +1 0 0 1 381.96 648.48 Tm +(multicasts) Tj +1 0 0 1 424.8 648.48 Tm +(the) Tj +1 0 0 1 439.56 648.48 Tm +(request) Tj +1 0 0 1 470.759 648.48 Tm +(to) Tj +1 0 0 1 481.079 648.48 Tm +(the) Tj +1 0 0 1 495.719 648.48 Tm +(backups) Tj +1 0 0 1 317.52 635.04 Tm +(3.) Tj +1 0 0 1 330 635.04 Tm +(Replicas) Tj +1 0 0 1 367.32 635.04 Tm +(e) Tj +1 0 0 1 371.64 635.04 Tm +(x) Tj +1 0 0 1 376.56 635.04 Tm +(ecute) Tj +1 0 0 1 400.44 635.04 Tm +(the) Tj +1 0 0 1 415.68 635.04 Tm +(request) Tj +1 0 0 1 447.359 635.04 Tm +(and) Tj +1 0 0 1 464.879 635.04 Tm +(send) Tj +1 0 0 1 486.119 635.04 Tm +(a) Tj +1 0 0 1 493.559 635.04 Tm +(reply) Tj +1 0 0 1 517.079 635.04 Tm +(to) Tj +1 0 0 1 527.879 635.04 Tm +(the) Tj +1 0 0 1 330 623.04 Tm +(client) Tj +1 0 0 1 317.52 609.6 Tm +(4.) Tj +1 0 0 1 330 609.6 Tm +(The) Tj +1 0 0 1 350.28 609.6 Tm +(client) Tj +1 0 0 1 377.28 609.6 Tm +(w) Tj +1 0 0 1 384.36 609.6 Tm +(aits) Tj +1 0 0 1 403.08 609.6 Tm +(for) Tj +ET +q +5 0 0 -9.2 420 616.7 cm +/R142 Do +Q +q +6.6 0 0 -6.5 429.6 615.4 cm +/R146 Do +Q +BT +1 0 0 1 440.4 609.6 Tm +(1) Tj +1 0 0 1 450.24 609.6 Tm +(replies) Tj +1 0 0 1 481.679 609.6 Tm +(from) Tj +1 0 0 1 505.919 609.6 Tm +(dif) Tj +1 0 0 1 516.839 609.6 Tm +(ferent) Tj +1 0 0 1 330 597.6 Tm +(replicas) Tj +1 0 0 1 365.28 597.6 Tm +(with) Tj +1 0 0 1 387.36 597.6 Tm +(the) Tj +1 0 0 1 403.92 597.6 Tm +(same) Tj +1 0 0 1 428.76 597.6 Tm +(result;) Tj +1 0 0 1 458.88 597.6 Tm +(this) Tj +1 0 0 1 477.719 597.6 Tm +(is) Tj +1 0 0 1 488.639 597.6 Tm +(the) Tj +1 0 0 1 505.199 597.6 Tm +(result) Tj +1 0 0 1 531.719 597.6 Tm +(of) Tj +1 0 0 1 330 585.72 Tm +(the) Tj +1 0 0 1 344.64 585.72 Tm +(operation.) Tj +1 0 0 1 324.96 571.44 Tm +(Lik) Tj +1 0 0 1 338.76 571.44 Tm +(e) Tj +1 0 0 1 348.24 571.44 Tm +(all) Tj +1 0 0 1 363.24 571.44 Tm +(state) Tj +1 0 0 1 386.64 571.44 Tm +(machine) Tj +1 0 0 1 425.52 571.44 Tm +(replication) Tj +1 0 0 1 473.159 571.44 Tm +(techniques) Tj +1 0 0 1 520.919 571.44 Tm +([34) Tj +1 0 0 1 534.239 571.44 Tm +(],) Tj +1 0 0 1 315 559.56 Tm +(we) Tj +1 0 0 1 331.44 559.56 Tm +(impose) Tj +1 0 0 1 365.16 559.56 Tm +(tw) Tj +1 0 0 1 375 559.56 Tm +(o) Tj +1 0 0 1 384.84 559.56 Tm +(requirements) Tj +1 0 0 1 441.72 559.56 Tm +(on) Tj +1 0 0 1 456.6 559.56 Tm +(replicas:) Tj +1 0 0 1 498.719 559.56 Tm +(the) Tj +1 0 0 1 510.839 559.56 Tm +(y) Tj +1 0 0 1 520.679 559.56 Tm +(must) Tj +1 0 0 1 315 547.56 Tm +(be) Tj +/R131 10 Tf +1 0 0 1 328.32 547.56 Tm +(deterministic) Tj +/R124 10 Tf +1 0 0 1 384.36 547.56 Tm +(\(i.e.,) Tj +1 0 0 1 406.56 547.56 Tm +(the) Tj +1 0 0 1 422.76 547.56 Tm +(e) Tj +1 0 0 1 427.08 547.56 Tm +(x) Tj +1 0 0 1 432 547.56 Tm +(ecution) Tj +1 0 0 1 465.119 547.56 Tm +(of) Tj +1 0 0 1 477.359 547.56 Tm +(an) Tj +1 0 0 1 490.679 547.56 Tm +(operation) Tj +1 0 0 1 532.199 547.56 Tm +(in) Tj +1 0 0 1 315 535.68 Tm +(a) Tj +1 0 0 1 323.64 535.68 Tm +(gi) Tj +1 0 0 1 331.2 535.68 Tm +(v) Tj +1 0 0 1 336.12 535.68 Tm +(en) Tj +1 0 0 1 349.56 535.68 Tm +(state) Tj +1 0 0 1 372.12 535.68 Tm +(and) Tj +1 0 0 1 390.72 535.68 Tm +(with) Tj +1 0 0 1 412.68 535.68 Tm +(a) Tj +1 0 0 1 421.32 535.68 Tm +(gi) Tj +1 0 0 1 428.879 535.68 Tm +(v) Tj +1 0 0 1 433.8 535.68 Tm +(en) Tj +1 0 0 1 447.239 535.68 Tm +(set) Tj +1 0 0 1 462.599 535.68 Tm +(of) Tj +1 0 0 1 475.079 535.68 Tm +(ar) Tj +1 0 0 1 482.759 535.68 Tm +(guments) Tj +1 0 0 1 520.679 535.68 Tm +(must) Tj +1 0 0 1 315 523.68 Tm +(al) Tj +1 0 0 1 322.08 523.68 Tm +(w) Tj +1 0 0 1 329.16 523.68 Tm +(ays) Tj +1 0 0 1 344.4 523.68 Tm +(produce) Tj +1 0 0 1 378.48 523.68 Tm +(the) Tj +1 0 0 1 392.52 523.68 Tm +(same) Tj +1 0 0 1 414.96 523.68 Tm +(result\)) Tj +1 0 0 1 442.44 523.68 Tm +(and) Tj +1 0 0 1 458.759 523.68 Tm +(the) Tj +1 0 0 1 470.879 523.68 Tm +(y) Tj +1 0 0 1 477.719 523.68 Tm +(must) Tj +1 0 0 1 499.079 523.68 Tm +(start) Tj +1 0 0 1 518.159 523.68 Tm +(in) Tj +1 0 0 1 527.879 523.68 Tm +(the) Tj +1 0 0 1 315 511.8 Tm +(same) Tj +1 0 0 1 338.04 511.8 Tm +(state.) Tj +1 0 0 1 362.52 511.8 Tm +(Gi) Tj +1 0 0 1 372.24 511.8 Tm +(v) Tj +1 0 0 1 377.16 511.8 Tm +(en) Tj +1 0 0 1 389.04 511.8 Tm +(these) Tj +1 0 0 1 412.08 511.8 Tm +(tw) Tj +1 0 0 1 421.92 511.8 Tm +(o) Tj +1 0 0 1 429.48 511.8 Tm +(requirements,) Tj +1 0 0 1 486.479 511.8 Tm +(the) Tj +1 0 0 1 501.239 511.8 Tm +(algorithm) Tj +1 0 0 1 315 499.8 Tm +(ensures) Tj +1 0 0 1 347.4 499.8 Tm +(the) Tj +1 0 0 1 362.16 499.8 Tm +(safety) Tj +1 0 0 1 388.44 499.8 Tm +(property) Tj +1 0 0 1 424.68 499.8 Tm +(by) Tj +1 0 0 1 437.279 499.8 Tm +(guaranteeing) Tj +1 0 0 1 491.159 499.8 Tm +(that) Tj +/R131 10 Tf +1 0 0 1 508.679 499.8 Tm +(all) Tj +1 0 0 1 521.759 499.8 Tm +(non-) Tj +1 0 0 1 315 487.92 Tm +(faulty) Tj +1 0 0 1 340.08 487.92 Tm +(r) Tj +1 0 0 1 343.56 487.92 Tm +(eplicas) Tj +1 0 0 1 374.16 487.92 Tm +(a) Tj +1 0 0 1 379.08 487.92 Tm +(gr) Tj +1 0 0 1 387.6 487.92 Tm +(ee) Tj +1 0 0 1 398.76 487.92 Tm +(on) Tj +1 0 0 1 411.12 487.92 Tm +(a) Tj +1 0 0 1 418.56 487.92 Tm +(total) Tj +1 0 0 1 439.199 487.92 Tm +(or) Tj +1 0 0 1 447.719 487.92 Tm +(der) Tj +1 0 0 1 463.319 487.92 Tm +(for) Tj +1 0 0 1 477.359 487.92 Tm +(the) Tj +1 0 0 1 491.879 487.92 Tm +(e) Tj +1 0 0 1 496.079 487.92 Tm +(xecution) Tj +1 0 0 1 532.199 487.92 Tm +(of) Tj +1 0 0 1 315 475.92 Tm +(r) Tj +1 0 0 1 318.48 475.92 Tm +(equests) Tj +1 0 0 1 350.28 475.92 Tm +(despite) Tj +1 0 0 1 381 475.92 Tm +(failur) Tj +1 0 0 1 402.84 475.92 Tm +(es) Tj +/R124 10 Tf +1 0 0 1 411.12 475.92 Tm +(.) Tj +1 0 0 1 324.96 464.04 Tm +(The) Tj +1 0 0 1 344.28 464.04 Tm +(remainder) Tj +1 0 0 1 388.56 464.04 Tm +(of) Tj +1 0 0 1 400.68 464.04 Tm +(this) Tj +1 0 0 1 418.92 464.04 Tm +(section) Tj +1 0 0 1 450.96 464.04 Tm +(describes) Tj +1 0 0 1 491.879 464.04 Tm +(a) Tj +1 0 0 1 500.159 464.04 Tm +(simpli\256ed) Tj +1 0 0 1 315 452.04 Tm +(v) Tj +1 0 0 1 319.92 452.04 Tm +(ersion) Tj +1 0 0 1 348.12 452.04 Tm +(of) Tj +1 0 0 1 360.48 452.04 Tm +(the) Tj +1 0 0 1 376.68 452.04 Tm +(algorithm.) Tj +1 0 0 1 426 452.04 Tm +(W) Tj +1 0 0 1 434.52 452.04 Tm +(e) Tj +1 0 0 1 443.04 452.04 Tm +(omit) Tj +1 0 0 1 465.24 452.04 Tm +(discussion) Tj +1 0 0 1 510.839 452.04 Tm +(of) Tj +1 0 0 1 523.079 452.04 Tm +(ho) Tj +1 0 0 1 532.919 452.04 Tm +(w) Tj +1 0 0 1 315 440.16 Tm +(nodes) Tj +1 0 0 1 342.84 440.16 Tm +(reco) Tj +1 0 0 1 360 440.16 Tm +(v) Tj +1 0 0 1 364.92 440.16 Tm +(er) Tj +1 0 0 1 376.92 440.16 Tm +(from) Tj +1 0 0 1 400.8 440.16 Tm +(f) Tj +1 0 0 1 404.04 440.16 Tm +(aults) Tj +1 0 0 1 427.44 440.16 Tm +(due) Tj +1 0 0 1 446.28 440.16 Tm +(to) Tj +1 0 0 1 458.64 440.16 Tm +(lack) Tj +1 0 0 1 479.759 440.16 Tm +(of) Tj +1 0 0 1 492.599 440.16 Tm +(space.) Tj +1 0 0 1 526.919 440.16 Tm +(W) Tj +1 0 0 1 535.439 440.16 Tm +(e) Tj +1 0 0 1 315 428.16 Tm +(also) Tj +1 0 0 1 336.36 428.16 Tm +(omit) Tj +1 0 0 1 359.88 428.16 Tm +(details) Tj +1 0 0 1 391.08 428.16 Tm +(related) Tj +1 0 0 1 423.48 428.16 Tm +(to) Tj +1 0 0 1 436.56 428.16 Tm +(message) Tj +1 0 0 1 475.559 428.16 Tm +(retransmissions.) Tj +1 0 0 1 315 416.28 Tm +(Furthermore,) Tj +1 0 0 1 370.8 416.28 Tm +(we) Tj +1 0 0 1 385.68 416.28 Tm +(assume) Tj +1 0 0 1 418.32 416.28 Tm +(that) Tj +1 0 0 1 436.56 416.28 Tm +(message) Tj +1 0 0 1 473.64 416.28 Tm +(authentication) Tj +1 0 0 1 533.399 416.28 Tm +(is) Tj +1 0 0 1 315 404.28 Tm +(achie) Tj +1 0 0 1 335.88 404.28 Tm +(v) Tj +1 0 0 1 340.8 404.28 Tm +(ed) Tj +1 0 0 1 354.24 404.28 Tm +(using) Tj +1 0 0 1 380.04 404.28 Tm +(digital) Tj +1 0 0 1 409.68 404.28 Tm +(signatures) Tj +1 0 0 1 454.32 404.28 Tm +(rather) Tj +1 0 0 1 481.799 404.28 Tm +(than) Tj +1 0 0 1 503.159 404.28 Tm +(the) Tj +1 0 0 1 519.479 404.28 Tm +(more) Tj +1 0 0 1 315 392.4 Tm +(ef) Tj +1 0 0 1 322.56 392.4 Tm +(\256cient) Tj +1 0 0 1 349.8 392.4 Tm +(scheme) Tj +1 0 0 1 382.08 392.4 Tm +(based) Tj +1 0 0 1 407.16 392.4 Tm +(on) Tj +1 0 0 1 419.52 392.4 Tm +(message) Tj +1 0 0 1 455.64 392.4 Tm +(authentication) Tj +1 0 0 1 514.559 392.4 Tm +(codes;) Tj +1 0 0 1 315 380.4 Tm +(Section) Tj +1 0 0 1 351 380.4 Tm +(5) Tj +1 0 0 1 362.04 380.4 Tm +(discusses) Tj +1 0 0 1 405.24 380.4 Tm +(this) Tj +1 0 0 1 425.76 380.4 Tm +(issue) Tj +1 0 0 1 451.68 380.4 Tm +(further) Tj +1 0 0 1 478.439 380.4 Tm +(.) Tj +1 0 0 1 495.119 380.4 Tm +(A) Tj +1 0 0 1 508.439 380.4 Tm +(detailed) Tj +1 0 0 1 315 368.4 Tm +(formalization) Tj +1 0 0 1 372 368.4 Tm +(of) Tj +1 0 0 1 383.52 368.4 Tm +(the) Tj +1 0 0 1 399 368.4 Tm +(algorithm) Tj +1 0 0 1 441 368.4 Tm +(using) Tj +1 0 0 1 465.96 368.4 Tm +(the) Tj +1 0 0 1 481.319 368.4 Tm +(I/O) Tj +1 0 0 1 497.879 368.4 Tm +(automaton) Tj +1 0 0 1 315 356.52 Tm +(model) Tj +1 0 0 1 342.36 356.52 Tm +([21) Tj +1 0 0 1 355.68 356.52 Tm +(]) Tj +1 0 0 1 361.44 356.52 Tm +(is) Tj +1 0 0 1 370.68 356.52 Tm +(presented) Tj +1 0 0 1 411.24 356.52 Tm +(in) Tj +1 0 0 1 421.56 356.52 Tm +([4) Tj +1 0 0 1 429.84 356.52 Tm +(].) Tj +/R162 10 Tf +1 0 0 1 315 336 Tm +(4.1) Tj +1 0 0 1 337.44 336 Tm +(The) Tj +1 0 0 1 356.52 336 Tm +(Client) Tj +/R124 10 Tf +1 0 0 1 315 321.84 Tm +(A) Tj +1 0 0 1 328.2 321.84 Tm +(client) Tj +ET +endstream +endobj +200 0 obj +11470 +endobj +201 0 obj +<< /Type /XObject /Name /R201 /Subtype /Image /Length 202 0 R +/ImageMask true /Width 39 /Height 46 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 39 /BlackIs1 true >>] +>> +stream +,R&cB5lcJ_F@&^\*D2E$qfTOoCb+sSa7''-+,BlFrn/.V#X.#4LY-rlHOLpI:jTBC_X\*\!'gY~> +endstream +endobj +202 0 obj +77 +endobj +203 0 obj +<< /Length 204 0 R >> +stream +q +3.9 0 0 -4.6 356.6 326.3 cm +/R201 Do +Q +BT +1 0 0 1 366.6 321.84 Tm +(requests) Tj +1 0 0 1 405.24 321.84 Tm +(the) Tj +1 0 0 1 423.36 321.84 Tm +(e) Tj +1 0 0 1 427.68 321.84 Tm +(x) Tj +1 0 0 1 432.6 321.84 Tm +(ecution) Tj +1 0 0 1 467.759 321.84 Tm +(of) Tj +1 0 0 1 482.039 321.84 Tm +(state) Tj +1 0 0 1 506.279 321.84 Tm +(machine) Tj +1 0 0 1 315 309.84 Tm +(operation) Tj +ET +endstream +endobj +204 0 obj +367 +endobj +205 0 obj +<< /Type /XObject /Name /R205 /Subtype /Image /Length 206 0 R +/ImageMask true /Width 44 /Height 46 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 44 /BlackIs1 true >>] +>> +stream +,RFT> +EI`ineN+j94Q'iSTFbqu-P3hnIY1ojRRBhdNX>"JG7c +endstream +endobj +206 0 obj +87 +endobj +207 0 obj +<< /Length 208 0 R >> +stream +q +4.4 0 0 -4.6 357.1 314.3 cm +/R205 Do +Q +BT +1 0 0 1 365.76 309.84 Tm +(by) Tj +1 0 0 1 379.92 309.84 Tm +(sending) Tj +1 0 0 1 415.08 309.84 Tm +(a) Tj +ET +endstream +endobj +208 0 obj +150 +endobj +209 0 obj +<< /Type /XObject /Name /R209 /Subtype /Image /Length 210 0 R +/ImageMask true /Width 24 /Height 99 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 24 /BlackIs1 true >>] +>> +stream ++t-07X8N5#0B"SKInTVGqHrM(f3_PN_YUJ4L&^prs)7cFqg[B")ufa0GPC"Nr'(,Kmt^kVHo1m8!<~> +endstream +endobj +210 0 obj +80 +endobj +211 0 obj +<< /Length 212 0 R >> +stream +q +2.4 0 0 -9.9 424.6 317.3 cm +/R209 Do +Q +BT +/R124 8 Tf +1 0 0 1 427.56 309.84 Tm +(REQ) Tj +1 0 0 1 443.4 309.84 Tm +(UEST) Tj +ET +q +1.3 0 0 -3 464.1 310.9 cm +/R173 Do +Q +q +4.4 0 0 -4.6 468.1 314.3 cm +/R205 Do +Q +q +1.3 0 0 -3 473.4 310.9 cm +/R173 Do +Q +q +3.1 0 0 -6.4 477.2 316.1 cm +/R127 Do +Q +q +1.3 0 0 -3 481.4 310.9 cm +/R173 Do +Q +q +3.9 0 0 -4.6 485.3 314.3 cm +/R201 Do +Q +endstream +endobj +212 0 obj +366 +endobj +213 0 obj +<< /Type /XObject /Name /R213 /Subtype /Image /Length 214 0 R +/ImageMask true /Width 24 /Height 99 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 24 /BlackIs1 true >>] +>> +stream +02pd&htA/cs0qnbrP.\`mJhV5_u&o]hqrq@J+[15ImJQWpe1HjlFsc8R~> +endstream +endobj +214 0 obj +78 +endobj +215 0 obj +<< /Length 216 0 R >> +stream +q +2.4 0 0 -9.9 489.8 317.3 cm +/R213 Do +Q +endstream +endobj +216 0 obj +41 +endobj +217 0 obj +<< /Type /XObject /Name /R217 /Subtype /Image /Length 218 0 R +/ImageMask true /Width 42 /Height 31 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 42 /BlackIs1 true >>] +>> +stream +1`R)4"qZ&4%Y>/aKS+]fJ#a+Sp>>oFIt)n@>7^\DYeoBbLCXeP2$%q%J-Z~> +endstream +endobj +218 0 obj +61 +endobj +219 0 obj +<< /Length 220 0 R >> +stream +q +4.2 0 0 -3.1 493.4 311.4 cm +/R217 Do +Q +endstream +endobj +220 0 obj +41 +endobj +221 0 obj +<< /Type /XObject /Name /R221 /Subtype /Image /Length 222 0 R +/ImageMask true /Width 24 /Height 22 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 24 /BlackIs1 true >>] +>> +stream +4q\,A"/8b:&%,:7s)7KqYh&c]_[`rh_Z^#O!<~> +endstream +endobj +222 0 obj +40 +endobj +223 0 obj +<< /Length 224 0 R >> +stream +q +2.4 0 0 -2.2 498.2 309.6 cm +/R221 Do +Q +BT +/R124 10 Tf +1 0 0 1 506.28 309.84 Tm +(message) Tj +1 0 0 1 315 297.96 Tm +(to) Tj +1 0 0 1 326.16 297.96 Tm +(the) Tj +1 0 0 1 341.76 297.96 Tm +(primary) Tj +1 0 0 1 372.96 297.96 Tm +(.) Tj +1 0 0 1 381.48 297.96 Tm +(T) Tj +1 0 0 1 387.24 297.96 Tm +(imestamp) Tj +ET +q +3.1 0 0 -6.4 429.6 304.3 cm +/R127 Do +Q +BT +1 0 0 1 436.32 297.96 Tm +(is) Tj +1 0 0 1 446.4 297.96 Tm +(used) Tj +1 0 0 1 468.119 297.96 Tm +(to) Tj +1 0 0 1 479.279 297.96 Tm +(ensure) Tj +/R131 10 Tf +1 0 0 1 508.679 297.96 Tm +(e) Tj +1 0 0 1 512.879 297.96 Tm +(xactly-) Tj +1 0 0 1 315 285.96 Tm +(once) Tj +/R124 10 Tf +1 0 0 1 339.36 285.96 Tm +(semantics) Tj +1 0 0 1 384.24 285.96 Tm +(for) Tj +1 0 0 1 401.52 285.96 Tm +(the) Tj +1 0 0 1 419.28 285.96 Tm +(e) Tj +1 0 0 1 423.6 285.96 Tm +(x) Tj +1 0 0 1 428.52 285.96 Tm +(ecution) Tj +1 0 0 1 463.2 285.96 Tm +(of) Tj +1 0 0 1 477.119 285.96 Tm +(client) Tj +1 0 0 1 504.839 285.96 Tm +(requests.) Tj +1 0 0 1 315 274.08 Tm +(T) Tj +1 0 0 1 320.76 274.08 Tm +(imestamps) Tj +1 0 0 1 365.88 274.08 Tm +(for) Tj +ET +q +3.9 0 0 -4.6 380.3 278.6 cm +/R201 Do +Q +BT +1 0 0 1 384.36 274.08 Tm +(') Tj +1 0 0 1 387.12 274.08 Tm +(s) Tj +1 0 0 1 393.48 274.08 Tm +(requests) Tj +1 0 0 1 428.64 274.08 Tm +(are) Tj +1 0 0 1 443.4 274.08 Tm +(totally) Tj +1 0 0 1 471.359 274.08 Tm +(ordered) Tj +1 0 0 1 504.239 274.08 Tm +(such) Tj +1 0 0 1 524.999 274.08 Tm +(that) Tj +1 0 0 1 315 262.08 Tm +(later) Tj +1 0 0 1 335.76 262.08 Tm +(requests) Tj +1 0 0 1 371.52 262.08 Tm +(ha) Tj +1 0 0 1 380.76 262.08 Tm +(v) Tj +1 0 0 1 385.68 262.08 Tm +(e) Tj +1 0 0 1 393.12 262.08 Tm +(higher) Tj +1 0 0 1 421.68 262.08 Tm +(timestamps) Tj +1 0 0 1 470.159 262.08 Tm +(than) Tj +1 0 0 1 490.439 262.08 Tm +(earlier) Tj +1 0 0 1 518.999 262.08 Tm +(ones;) Tj +1 0 0 1 315 250.2 Tm +(for) Tj +1 0 0 1 330.84 250.2 Tm +(e) Tj +1 0 0 1 335.16 250.2 Tm +(xample,) Tj +1 0 0 1 371.52 250.2 Tm +(the) Tj +1 0 0 1 387.84 250.2 Tm +(timestamp) Tj +1 0 0 1 433.56 250.2 Tm +(could) Tj +1 0 0 1 459.84 250.2 Tm +(be) Tj +1 0 0 1 473.52 250.2 Tm +(the) Tj +1 0 0 1 489.839 250.2 Tm +(v) Tj +1 0 0 1 494.639 250.2 Tm +(alue) Tj +1 0 0 1 515.399 250.2 Tm +(of) Tj +1 0 0 1 527.879 250.2 Tm +(the) Tj +1 0 0 1 315 238.2 Tm +(client') Tj +1 0 0 1 339.96 238.2 Tm +(s) Tj +1 0 0 1 346.32 238.2 Tm +(local) Tj +1 0 0 1 368.16 238.2 Tm +(clock) Tj +1 0 0 1 392.16 238.2 Tm +(when) Tj +1 0 0 1 416.28 238.2 Tm +(the) Tj +1 0 0 1 430.92 238.2 Tm +(request) Tj +1 0 0 1 462.239 238.2 Tm +(is) Tj +1 0 0 1 471.359 238.2 Tm +(issued.) Tj +1 0 0 1 324.96 226.32 Tm +(Each) Tj +1 0 0 1 346.8 226.32 Tm +(message) Tj +1 0 0 1 382.44 226.32 Tm +(sent) Tj +1 0 0 1 400.32 226.32 Tm +(by) Tj +1 0 0 1 412.2 226.32 Tm +(the) Tj +1 0 0 1 426.24 226.32 Tm +(replicas) Tj +1 0 0 1 459.119 226.32 Tm +(to) Tj +1 0 0 1 468.719 226.32 Tm +(the) Tj +1 0 0 1 482.759 226.32 Tm +(client) Tj +1 0 0 1 506.759 226.32 Tm +(includes) Tj +1 0 0 1 315 214.32 Tm +(the) Tj +1 0 0 1 329.88 214.32 Tm +(current) Tj +1 0 0 1 360.72 214.32 Tm +(vie) Tj +1 0 0 1 372.72 214.32 Tm +(w) Tj +1 0 0 1 382.56 214.32 Tm +(number) Tj +1 0 0 1 412.92 214.32 Tm +(,) Tj +1 0 0 1 417.84 214.32 Tm +(allo) Tj +1 0 0 1 432.6 214.32 Tm +(wing) Tj +1 0 0 1 455.16 214.32 Tm +(the) Tj +1 0 0 1 469.92 214.32 Tm +(client) Tj +1 0 0 1 494.759 214.32 Tm +(to) Tj +1 0 0 1 505.199 214.32 Tm +(track) Tj +1 0 0 1 527.879 214.32 Tm +(the) Tj +1 0 0 1 315 202.32 Tm +(vie) Tj +1 0 0 1 327 202.32 Tm +(w) Tj +1 0 0 1 338.76 202.32 Tm +(and) Tj +1 0 0 1 357.84 202.32 Tm +(hence) Tj +1 0 0 1 385.8 202.32 Tm +(the) Tj +1 0 0 1 402.72 202.32 Tm +(current) Tj +1 0 0 1 435.6 202.32 Tm +(primary) Tj +1 0 0 1 466.8 202.32 Tm +(.) Tj +1 0 0 1 479.16 202.32 Tm +(A) Tj +1 0 0 1 491.04 202.32 Tm +(client) Tj +1 0 0 1 517.919 202.32 Tm +(sends) Tj +1 0 0 1 315 190.44 Tm +(a) Tj +1 0 0 1 324.84 190.44 Tm +(request) Tj +1 0 0 1 358.92 190.44 Tm +(to) Tj +1 0 0 1 372 190.44 Tm +(what) Tj +1 0 0 1 396.84 190.44 Tm +(it) Tj +1 0 0 1 407.64 190.44 Tm +(belie) Tj +1 0 0 1 426.84 190.44 Tm +(v) Tj +1 0 0 1 431.76 190.44 Tm +(es) Tj +1 0 0 1 445.32 190.44 Tm +(is) Tj +1 0 0 1 457.319 190.44 Tm +(the) Tj +1 0 0 1 474.839 190.44 Tm +(current) Tj +1 0 0 1 508.439 190.44 Tm +(primary) Tj +1 0 0 1 315 178.44 Tm +(using) Tj +1 0 0 1 339.36 178.44 Tm +(a) Tj +1 0 0 1 346.56 178.44 Tm +(point-to-point) Tj +1 0 0 1 404.64 178.44 Tm +(message.) Tj +1 0 0 1 445.32 178.44 Tm +(The) Tj +1 0 0 1 463.68 178.44 Tm +(primary) Tj +1 0 0 1 497.879 178.44 Tm +(atomically) Tj +1 0 0 1 315 166.56 Tm +(multicasts) Tj +1 0 0 1 357 166.56 Tm +(the) Tj +1 0 0 1 370.8 166.56 Tm +(request) Tj +1 0 0 1 401.28 166.56 Tm +(to) Tj +1 0 0 1 410.64 166.56 Tm +(all) Tj +1 0 0 1 422.16 166.56 Tm +(the) Tj +1 0 0 1 435.96 166.56 Tm +(backups) Tj +1 0 0 1 470.399 166.56 Tm +(using) Tj +1 0 0 1 493.679 166.56 Tm +(the) Tj +1 0 0 1 507.479 166.56 Tm +(pr) Tj +1 0 0 1 515.759 166.56 Tm +(otoc) Tj +1 0 0 1 532.919 166.56 Tm +(ol) Tj +1 0 0 1 315 154.56 Tm +(described) Tj +1 0 0 1 355.68 154.56 Tm +(in) Tj +1 0 0 1 365.88 154.56 Tm +(the) Tj +1 0 0 1 380.52 154.56 Tm +(ne) Tj +1 0 0 1 389.88 154.56 Tm +(xt) Tj +1 0 0 1 400.08 154.56 Tm +(section.) Tj +1 0 0 1 324.96 142.68 Tm +(A) Tj +1 0 0 1 336.48 142.68 Tm +(replica) Tj +1 0 0 1 367.92 142.68 Tm +(sends) Tj +1 0 0 1 394.44 142.68 Tm +(the) Tj +1 0 0 1 410.88 142.68 Tm +(reply) Tj +1 0 0 1 435.72 142.68 Tm +(to) Tj +1 0 0 1 447.84 142.68 Tm +(the) Tj +1 0 0 1 464.399 142.68 Tm +(request) Tj +1 0 0 1 497.519 142.68 Tm +(directly) Tj +1 0 0 1 532.199 142.68 Tm +(to) Tj +1 0 0 1 315 130.68 Tm +(the) Tj +1 0 0 1 330.24 130.68 Tm +(client.) Tj +1 0 0 1 360.12 130.68 Tm +(The) Tj +1 0 0 1 378.6 130.68 Tm +(reply) Tj +1 0 0 1 402.12 130.68 Tm +(has) Tj +1 0 0 1 418.44 130.68 Tm +(the) Tj +1 0 0 1 433.68 130.68 Tm +(form) Tj +ET +q +2.4 0 0 -9.9 457 138.2 cm +/R209 Do +Q +BT +/R124 8 Tf +1 0 0 1 459.959 130.68 Tm +(REPL) Tj +1 0 0 1 478.679 130.68 Tm +(Y) Tj +ET +q +1.3 0 0 -3 485.2 131.8 cm +/R173 Do +Q +q +4.4 0 0 -4.6 489.1 135.2 cm +/R197 Do +Q +q +1.3 0 0 -3 494.8 131.8 cm +/R173 Do +Q +q +3.1 0 0 -6.4 498.7 137 cm +/R127 Do +Q +q +1.3 0 0 -3 502.9 131.8 cm +/R173 Do +Q +q +3.9 0 0 -4.6 506.8 135.2 cm +/R201 Do +Q +q +1.3 0 0 -3 511.6 131.8 cm +/R173 Do +Q +endstream +endobj +224 0 obj +5932 +endobj +225 0 obj +<< /Type /XObject /Name /R225 /Subtype /Image /Length 226 0 R +/ImageMask true /Width 26 /Height 69 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 26 /BlackIs1 true >>] +>> +stream +-R^*Zl/YrmIf:DfhtVWo^UNqc%K6'1s*addX2"DHrSQB/+4u(tp(mUa[G\kks2oQE"5iR%(OubkJ,~> +endstream +endobj +226 0 obj +80 +endobj +227 0 obj +<< /Length 228 0 R >> +stream +q +2.6 0 0 -6.9 515.5 137.5 cm +/R225 Do +Q +q +1.3 0 0 -3 519.6 131.8 cm +/R173 Do +Q +endstream +endobj +228 0 obj +80 +endobj +229 0 obj +<< /Type /XObject /Name /R229 /Subtype /Image /Length 230 0 R +/ImageMask true /Width 39 /Height 46 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 39 /BlackIs1 true >>] +>> +stream +4Q-O[J+rF2n,E5LpjW83afBpkfg4TU#MZ`fmdT`'^@/Es*&)T7Ft& +endstream +endobj +230 0 obj +68 +endobj +231 0 obj +<< /Length 232 0 R >> +stream +q +3.9 0 0 -4.6 523.4 135.2 cm +/R229 Do +Q +q +2.4 0 0 -9.9 528.3 138.2 cm +/R213 Do +Q +q +4.2 0 0 -3.1 531.9 132.2 cm +/R217 Do +Q +endstream +endobj +232 0 obj +123 +endobj +233 0 obj +<< /Type /XObject /Name /R233 /Subtype /Image /Length 234 0 R +/ImageMask true /Width 17 /Height 35 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 17 /BlackIs1 true >>] +>> +stream +4B*=`S,;]mpj7Y*s1(">iS6Nohi?6cs!N02p(RKl"9~> +endstream +endobj +234 0 obj +45 +endobj +235 0 obj +<< /Length 236 0 R >> +stream +q +1.7 0 0 -3.5 536.6 131.8 cm +/R233 Do +Q +BT +/R124 10 Tf +1 0 0 1 315 118.8 Tm +(where) Tj +ET +q +4.4 0 0 -4.6 341.7 123.3 cm +/R197 Do +Q +BT +1 0 0 1 348.96 118.8 Tm +(is) Tj +1 0 0 1 357.84 118.8 Tm +(the) Tj +1 0 0 1 372.24 118.8 Tm +(current) Tj +1 0 0 1 402.72 118.8 Tm +(vie) Tj +1 0 0 1 414.72 118.8 Tm +(w) Tj +1 0 0 1 424.08 118.8 Tm +(number) Tj +1 0 0 1 454.44 118.8 Tm +(,) Tj +ET +q +3.1 0 0 -6.4 459.1 125.1 cm +/R127 Do +Q +BT +1 0 0 1 464.639 118.8 Tm +(is) Tj +1 0 0 1 473.519 118.8 Tm +(the) Tj +1 0 0 1 487.919 118.8 Tm +(timestamp) Tj +1 0 0 1 531.719 118.8 Tm +(of) Tj +1 0 0 1 315 106.8 Tm +(the) Tj +1 0 0 1 329.76 106.8 Tm +(corresponding) Tj +1 0 0 1 389.4 106.8 Tm +(request,) Tj +ET +q +2.6 0 0 -6.9 423.4 113.6 cm +/R225 Do +Q +BT +1 0 0 1 429.36 106.8 Tm +(is) Tj +1 0 0 1 438.6 106.8 Tm +(the) Tj +1 0 0 1 453.36 106.8 Tm +(replica) Tj +1 0 0 1 483.119 106.8 Tm +(number) Tj +1 0 0 1 513.479 106.8 Tm +(,) Tj +1 0 0 1 518.279 106.8 Tm +(and) Tj +ET +q +3.9 0 0 -4.6 535.4 111.3 cm +/R229 Do +Q +BT +1 0 0 1 315 94.92 Tm +(is) Tj +1 0 0 1 324.12 94.92 Tm +(the) Tj +1 0 0 1 338.76 94.92 Tm +(result) Tj +1 0 0 1 363.48 94.92 Tm +(of) Tj +1 0 0 1 374.16 94.92 Tm +(e) Tj +1 0 0 1 378.48 94.92 Tm +(x) Tj +1 0 0 1 383.4 94.92 Tm +(ecuting) Tj +1 0 0 1 415.2 94.92 Tm +(the) Tj +1 0 0 1 429.84 94.92 Tm +(requested) Tj +1 0 0 1 470.519 94.92 Tm +(operation.) Tj +1 0 0 1 324.96 82.92 Tm +(The) Tj +1 0 0 1 342.72 82.92 Tm +(client) Tj +1 0 0 1 367.2 82.92 Tm +(w) Tj +1 0 0 1 374.28 82.92 Tm +(aits) Tj +1 0 0 1 390.36 82.92 Tm +(for) Tj +ET +q +5 0 0 -9.2 404.7 90 cm +/R142 Do +Q +q +6.6 0 0 -6.5 412.5 88.7 cm +/R146 Do +Q +BT +1 0 0 1 421.56 82.92 Tm +(1) Tj +1 0 0 1 428.76 82.92 Tm +(replies) Tj +1 0 0 1 457.679 82.92 Tm +(with) Tj +1 0 0 1 477.599 82.92 Tm +(v) Tj +1 0 0 1 482.399 82.92 Tm +(alid) Tj +1 0 0 1 499.559 82.92 Tm +(signatures) Tj +1 0 0 1 315 71.04 Tm +(from) Tj +1 0 0 1 336.6 71.04 Tm +(dif) Tj +1 0 0 1 347.52 71.04 Tm +(ferent) Tj +1 0 0 1 372.96 71.04 Tm +(replicas,) Tj +1 0 0 1 408.72 71.04 Tm +(and) Tj +1 0 0 1 425.4 71.04 Tm +(with) Tj +1 0 0 1 445.32 71.04 Tm +(the) Tj +1 0 0 1 459.719 71.04 Tm +(same) Tj +ET +q +3.1 0 0 -6.4 482.7 77.3 cm +/R127 Do +Q +BT +1 0 0 1 488.399 71.04 Tm +(and) Tj +ET +q +3.9 0 0 -4.6 505.2 75.5 cm +/R229 Do +Q +BT +1 0 0 1 509.639 71.04 Tm +(,) Tj +1 0 0 1 514.559 71.04 Tm +(before) Tj +1 0 0 1 303.48 35.16 Tm +(3) Tj +ET +Q +endstream +endobj +236 0 obj +2265 +endobj +237 0 obj +<< +/Type /Page +/MediaBox [0 0 612 792] +/Parent 2 0 R +/Resources << /ProcSet [/PDF /ImageB /Text] +/Font << +/R162 162 0 R +/R131 131 0 R +/R124 124 0 R +>> +/XObject << +/R229 229 0 R +/R127 127 0 R +/R146 146 0 R +/R142 142 0 R +/R225 225 0 R +/R197 197 0 R +/R233 233 0 R +/R217 217 0 R +/R213 213 0 R +/R173 173 0 R +/R201 201 0 R +/R209 209 0 R +/R221 221 0 R +/R205 205 0 R +/R181 181 0 R +/R165 165 0 R +/R189 189 0 R +/R193 193 0 R +/R185 185 0 R +/R154 154 0 R +/R177 177 0 R +/R169 169 0 R +/R158 158 0 R +/R150 150 0 R +/R138 138 0 R +/R134 134 0 R +>> +>> +/Contents [ +125 0 R +129 0 R +132 0 R +136 0 R +140 0 R +144 0 R +148 0 R +152 0 R +156 0 R +160 0 R +163 0 R +167 0 R +171 0 R +175 0 R +179 0 R +183 0 R +187 0 R +191 0 R +195 0 R +199 0 R +203 0 R +207 0 R +211 0 R +215 0 R +219 0 R +223 0 R +227 0 R +231 0 R +235 0 R +] +>> +endobj +238 0 obj +<< /Type /Font /Name /R238 /Subtype /Type1 /BaseFont /Times-Roman >> +endobj +239 0 obj +<< /Length 240 0 R >> +stream +q +BT +/R238 10 Tf +1 0 0 1 72 709.08 Tm +(accepting) Tj +1 0 0 1 111.96 709.08 Tm +(the) Tj +1 0 0 1 126 709.08 Tm +(result) Tj +ET +Q +q +W +0 0 612 792 re +n +endstream +endobj +240 0 obj +147 +endobj +241 0 obj +<< /Type /XObject /Name /R241 /Subtype /Image /Length 242 0 R +/ImageMask true /Width 39 /Height 46 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 39 /BlackIs1 true >>] +>> +stream +4Q-O[J+rF2n,E5LpjW83afBpkfg4TU#MZ`fmdT`'^@/Es*&)T7Ft& +endstream +endobj +242 0 obj +68 +endobj +243 0 obj +<< /Length 244 0 R >> +stream +q +3.9 0 0 -4.6 150.1 713.6 cm +/R241 Do +Q +BT +/R238 10 Tf +1 0 0 1 154.56 709.08 Tm +(.) Tj +1 0 0 1 160.56 709.08 Tm +(This) Tj +1 0 0 1 180 709.08 Tm +(ensures) Tj +1 0 0 1 211.679 709.08 Tm +(that) Tj +1 0 0 1 228.479 709.08 Tm +(the) Tj +1 0 0 1 242.399 709.08 Tm +(result) Tj +1 0 0 1 266.399 709.08 Tm +(is) Tj +1 0 0 1 274.799 709.08 Tm +(v) Tj +1 0 0 1 279.599 709.08 Tm +(alid,) Tj +1 0 0 1 72 697.08 Tm +(since) Tj +1 0 0 1 94.9199 697.08 Tm +(at) Tj +1 0 0 1 104.64 697.08 Tm +(most) Tj +ET +endstream +endobj +244 0 obj +475 +endobj +245 0 obj +<< /Type /XObject /Name /R245 /Subtype /Image /Length 246 0 R +/ImageMask true /Width 50 /Height 92 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 50 /BlackIs1 true >>] +>> +stream +5$eL7-b:K3p0<%? +h1u#R]2'Mug].5is7$!eqtL-hqu?EZs5/V3iQXZVs7/ihqu?EZs4QRbJe-ans*o2/4DdB_@6aqq5Q~> +endstream +endobj +246 0 obj +96 +endobj +247 0 obj +<< /Length 248 0 R >> +stream +q +5 0 0 -9.2 126.9 704.2 cm +/R245 Do +Q +BT +1 0 0 1 135 697.08 Tm +(replicas) Tj +1 0 0 1 168.48 697.08 Tm +(can) Tj +1 0 0 1 184.8 697.08 Tm +(be) Tj +1 0 0 1 196.68 697.08 Tm +(f) Tj +1 0 0 1 199.92 697.08 Tm +(aulty) Tj +1 0 0 1 219.359 697.08 Tm +(.) Tj +1 0 0 1 81.96 684.72 Tm +(If) Tj +1 0 0 1 92.1599 684.72 Tm +(the) Tj +1 0 0 1 107.88 684.72 Tm +(client) Tj +1 0 0 1 133.44 684.72 Tm +(does) Tj +1 0 0 1 155.28 684.72 Tm +(not) Tj +1 0 0 1 171.48 684.72 Tm +(recei) Tj +1 0 0 1 190.68 684.72 Tm +(v) Tj +1 0 0 1 195.6 684.72 Tm +(e) Tj +1 0 0 1 203.4 684.72 Tm +(replies) Tj +1 0 0 1 233.519 684.72 Tm +(soon) Tj +1 0 0 1 255.839 684.72 Tm +(enough,) Tj +1 0 0 1 291.479 684.72 Tm +(it) Tj +1 0 0 1 72 672.72 Tm +(broadcasts) Tj +1 0 0 1 117.12 672.72 Tm +(the) Tj +1 0 0 1 132.36 672.72 Tm +(request) Tj +1 0 0 1 164.28 672.72 Tm +(to) Tj +1 0 0 1 175.08 672.72 Tm +(all) Tj +1 0 0 1 188.039 672.72 Tm +(replicas.) Tj +1 0 0 1 226.919 672.72 Tm +(If) Tj +1 0 0 1 236.639 672.72 Tm +(the) Tj +1 0 0 1 251.879 672.72 Tm +(request) Tj +1 0 0 1 283.679 672.72 Tm +(has) Tj +1 0 0 1 72 660.84 Tm +(already) Tj +1 0 0 1 104.64 660.84 Tm +(been) Tj +1 0 0 1 126.72 660.84 Tm +(processed,) Tj +1 0 0 1 171.96 660.84 Tm +(the) Tj +1 0 0 1 187.44 660.84 Tm +(replicas) Tj +1 0 0 1 221.76 660.84 Tm +(simply) Tj +1 0 0 1 252.24 660.84 Tm +(re-send) Tj +1 0 0 1 284.879 660.84 Tm +(the) Tj +1 0 0 1 72 648.84 Tm +(reply;) Tj +1 0 0 1 97.4399 648.84 Tm +(replicas) Tj +1 0 0 1 130.56 648.84 Tm +(remember) Tj +1 0 0 1 173.16 648.84 Tm +(the) Tj +1 0 0 1 187.44 648.84 Tm +(last) Tj +1 0 0 1 203.4 648.84 Tm +(reply) Tj +1 0 0 1 225.96 648.84 Tm +(message) Tj +1 0 0 1 261.839 648.84 Tm +(the) Tj +1 0 0 1 273.959 648.84 Tm +(y) Tj +1 0 0 1 280.919 648.84 Tm +(sent) Tj +1 0 0 1 72 636.96 Tm +(to) Tj +1 0 0 1 81.6 636.96 Tm +(each) Tj +1 0 0 1 101.76 636.96 Tm +(client.) Tj +1 0 0 1 129.84 636.96 Tm +(Otherwise,) Tj +1 0 0 1 175.32 636.96 Tm +(if) Tj +1 0 0 1 183.24 636.96 Tm +(the) Tj +1 0 0 1 197.4 636.96 Tm +(replica) Tj +1 0 0 1 226.319 636.96 Tm +(is) Tj +1 0 0 1 234.959 636.96 Tm +(not) Tj +1 0 0 1 249.479 636.96 Tm +(the) Tj +1 0 0 1 263.639 636.96 Tm +(primary) Tj +1 0 0 1 294.839 636.96 Tm +(,) Tj +1 0 0 1 72 624.96 Tm +(it) Tj +1 0 0 1 80.64 624.96 Tm +(relays) Tj +1 0 0 1 107.4 624.96 Tm +(the) Tj +1 0 0 1 122.64 624.96 Tm +(request) Tj +1 0 0 1 154.56 624.96 Tm +(to) Tj +1 0 0 1 165.36 624.96 Tm +(the) Tj +1 0 0 1 180.48 624.96 Tm +(primary) Tj +1 0 0 1 211.679 624.96 Tm +(.) Tj +1 0 0 1 219.239 624.96 Tm +(If) Tj +1 0 0 1 228.959 624.96 Tm +(the) Tj +1 0 0 1 244.199 624.96 Tm +(primary) Tj +1 0 0 1 278.759 624.96 Tm +(does) Tj +1 0 0 1 72 613.08 Tm +(not) Tj +1 0 0 1 87.7199 613.08 Tm +(multicast) Tj +1 0 0 1 127.2 613.08 Tm +(the) Tj +1 0 0 1 142.44 613.08 Tm +(request) Tj +1 0 0 1 174.12 613.08 Tm +(to) Tj +1 0 0 1 184.92 613.08 Tm +(the) Tj +1 0 0 1 200.04 613.08 Tm +(group,) Tj +1 0 0 1 228.84 613.08 Tm +(it) Tj +1 0 0 1 237.359 613.08 Tm +(will) Tj +1 0 0 1 255.839 613.08 Tm +(e) Tj +1 0 0 1 260.039 613.08 Tm +(v) Tj +1 0 0 1 264.959 613.08 Tm +(entually) Tj +1 0 0 1 72 601.08 Tm +(be) Tj +1 0 0 1 84.6 601.08 Tm +(suspected) Tj +1 0 0 1 126.48 601.08 Tm +(to) Tj +1 0 0 1 137.52 601.08 Tm +(be) Tj +1 0 0 1 150.12 601.08 Tm +(f) Tj +1 0 0 1 153.36 601.08 Tm +(aulty) Tj +1 0 0 1 176.4 601.08 Tm +(by) Tj +1 0 0 1 189.6 601.08 Tm +(enough) Tj +1 0 0 1 222.119 601.08 Tm +(replicas) Tj +1 0 0 1 256.319 601.08 Tm +(to) Tj +1 0 0 1 267.239 601.08 Tm +(cause) Tj +1 0 0 1 292.559 601.08 Tm +(a) Tj +1 0 0 1 72 589.2 Tm +(vie) Tj +1 0 0 1 84 589.2 Tm +(w) Tj +1 0 0 1 93.5999 589.2 Tm +(change.) Tj +1 0 0 1 81.96 576.84 Tm +(In) Tj +1 0 0 1 93.5999 576.84 Tm +(this) Tj +1 0 0 1 111.24 576.84 Tm +(paper) Tj +1 0 0 1 136.68 576.84 Tm +(we) Tj +1 0 0 1 151.68 576.84 Tm +(assume) Tj +1 0 0 1 184.32 576.84 Tm +(that) Tj +1 0 0 1 202.56 576.84 Tm +(the) Tj +1 0 0 1 218.04 576.84 Tm +(client) Tj +1 0 0 1 243.479 576.84 Tm +(w) Tj +1 0 0 1 250.559 576.84 Tm +(aits) Tj +1 0 0 1 267.719 576.84 Tm +(for) Tj +1 0 0 1 282.599 576.84 Tm +(one) Tj +1 0 0 1 72 564.84 Tm +(request) Tj +1 0 0 1 103.2 564.84 Tm +(to) Tj +1 0 0 1 113.52 564.84 Tm +(complete) Tj +1 0 0 1 152.52 564.84 Tm +(before) Tj +1 0 0 1 180.36 564.84 Tm +(sending) Tj +1 0 0 1 213.84 564.84 Tm +(the) Tj +1 0 0 1 228.599 564.84 Tm +(ne) Tj +1 0 0 1 237.959 564.84 Tm +(xt) Tj +1 0 0 1 248.039 564.84 Tm +(one.) Tj +1 0 0 1 268.559 564.84 Tm +(But) Tj +1 0 0 1 285.359 564.84 Tm +(we) Tj +1 0 0 1 72 552.96 Tm +(can) Tj +1 0 0 1 89.76 552.96 Tm +(allo) Tj +1 0 0 1 104.52 552.96 Tm +(w) Tj +1 0 0 1 115.44 552.96 Tm +(a) Tj +1 0 0 1 123.84 552.96 Tm +(client) Tj +1 0 0 1 149.76 552.96 Tm +(to) Tj +1 0 0 1 161.4 552.96 Tm +(mak) Tj +1 0 0 1 178.56 552.96 Tm +(e) Tj +1 0 0 1 186.84 552.96 Tm +(asynchronous) Tj +1 0 0 1 245.4 552.96 Tm +(requests,) Tj +1 0 0 1 284.879 552.96 Tm +(yet) Tj +1 0 0 1 72 540.96 Tm +(preserv) Tj +1 0 0 1 101.4 540.96 Tm +(e) Tj +1 0 0 1 108.12 540.96 Tm +(ordering) Tj +1 0 0 1 144.36 540.96 Tm +(constraints) Tj +1 0 0 1 189.96 540.96 Tm +(on) Tj +1 0 0 1 202.44 540.96 Tm +(them.) Tj +ET +endstream +endobj +248 0 obj +4845 +endobj +249 0 obj +<< /Type /Font /Name /R249 /Subtype /Type1 /BaseFont /Times-Bold >> +endobj +250 0 obj +<< /Length 251 0 R >> +stream +BT +/R249 10 Tf +1 0 0 1 72 517.68 Tm +(4.2) Tj +1 0 0 1 94.4399 517.68 Tm +(Normal-Case) Tj +1 0 0 1 153.36 517.68 Tm +(Operation) Tj +/R238 10 Tf +1 0 0 1 72 502.68 Tm +(The) Tj +1 0 0 1 93.4799 502.68 Tm +(state) Tj +1 0 0 1 117.6 502.68 Tm +(of) Tj +1 0 0 1 131.88 502.68 Tm +(each) Tj +1 0 0 1 156.12 502.68 Tm +(replica) Tj +1 0 0 1 189.12 502.68 Tm +(includes) Tj +1 0 0 1 228.24 502.68 Tm +(the) Tj +1 0 0 1 246.359 502.68 Tm +(state) Tj +1 0 0 1 270.599 502.68 Tm +(of) Tj +1 0 0 1 284.879 502.68 Tm +(the) Tj +1 0 0 1 72 490.8 Tm +(service,) Tj +1 0 0 1 106.44 490.8 Tm +(a) Tj +ET +endstream +endobj +251 0 obj +559 +endobj +252 0 obj +<< /Type /Font /Name /R252 /Subtype /Type1 /BaseFont /Times-Italic >> +endobj +253 0 obj +<< /Length 254 0 R >> +stream +BT +/R252 10 Tf +1 0 0 1 114.36 490.8 Tm +(messa) Tj +1 0 0 1 138.6 490.8 Tm +(g) Tj +1 0 0 1 143.52 490.8 Tm +(e) Tj +1 0 0 1 151.44 490.8 Tm +(lo) Tj +1 0 0 1 159.12 490.8 Tm +(g) Tj +/R238 10 Tf +1 0 0 1 167.52 490.8 Tm +(containing) Tj +1 0 0 1 213.12 490.8 Tm +(messages) Tj +1 0 0 1 254.279 490.8 Tm +(the) Tj +1 0 0 1 269.879 490.8 Tm +(replica) Tj +1 0 0 1 72 478.8 Tm +(has) Tj +1 0 0 1 87 478.8 Tm +(accepted,) Tj +1 0 0 1 126.24 478.8 Tm +(and) Tj +1 0 0 1 142.32 478.8 Tm +(an) Tj +1 0 0 1 153.48 478.8 Tm +(inte) Tj +1 0 0 1 168.36 478.8 Tm +(ger) Tj +1 0 0 1 182.76 478.8 Tm +(denoting) Tj +1 0 0 1 219.36 478.8 Tm +(the) Tj +1 0 0 1 233.279 478.8 Tm +(replica') Tj +1 0 0 1 263.279 478.8 Tm +(s) Tj +1 0 0 1 268.799 478.8 Tm +(current) Tj +1 0 0 1 72 466.8 Tm +(vie) Tj +1 0 0 1 84 466.8 Tm +(w) Tj +1 0 0 1 90.6 466.8 Tm +(.) Tj +1 0 0 1 96.48 466.8 Tm +(W) Tj +1 0 0 1 105 466.8 Tm +(e) Tj +1 0 0 1 111.72 466.8 Tm +(describe) Tj +1 0 0 1 147.12 466.8 Tm +(ho) Tj +1 0 0 1 156.96 466.8 Tm +(w) Tj +1 0 0 1 166.32 466.8 Tm +(to) Tj +1 0 0 1 176.28 466.8 Tm +(truncate) Tj +1 0 0 1 210.6 466.8 Tm +(the) Tj +1 0 0 1 225 466.8 Tm +(log) Tj +1 0 0 1 239.999 466.8 Tm +(in) Tj +1 0 0 1 249.959 466.8 Tm +(Section) Tj +1 0 0 1 281.999 466.8 Tm +(4.3.) Tj +1 0 0 1 81.96 454.44 Tm +(When) Tj +1 0 0 1 110.16 454.44 Tm +(the) Tj +1 0 0 1 126.84 454.44 Tm +(primary) Tj +1 0 0 1 158.04 454.44 Tm +(,) Tj +ET +endstream +endobj +254 0 obj +1335 +endobj +255 0 obj +<< /Type /XObject /Name /R255 /Subtype /Image /Length 256 0 R +/ImageMask true /Width 54 /Height 64 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 54 /BlackIs1 true >>] +>> +stream +-D98r4=0b"4s0Orro]!o&:3^Ke&7l2 +[/(7*rS^,RIfB4oLV,SG]>+(NrPSU5I/eEj^Aki-s4fFGn+Z5e95',@28).TTZ(Ru?\8Ii-iec'~> +endstream +endobj +256 0 obj +109 +endobj +257 0 obj +<< /Length 258 0 R >> +stream +q +5.4 0 0 -6.4 164.7 458.9 cm +/R255 Do +Q +BT +1 0 0 1 170.16 454.44 Tm +(,) Tj +1 0 0 1 177.48 454.44 Tm +(recei) Tj +1 0 0 1 196.68 454.44 Tm +(v) Tj +1 0 0 1 201.6 454.44 Tm +(es) Tj +1 0 0 1 214.2 454.44 Tm +(a) Tj +1 0 0 1 223.08 454.44 Tm +(client) Tj +1 0 0 1 249.6 454.44 Tm +(request,) Tj +ET +endstream +endobj +258 0 obj +285 +endobj +259 0 obj +<< /Type /XObject /Name /R259 /Subtype /Image /Length 260 0 R +/ImageMask true /Width 81 /Height 46 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 81 /BlackIs1 true >>] +>> +stream +4N`CV%M3$8612U>:Ag"NnFZb +endstream +endobj +260 0 obj +139 +endobj +261 0 obj +<< /Length 262 0 R >> +stream +q +8.1 0 0 -4.6 285.9 458.9 cm +/R259 Do +Q +BT +1 0 0 1 294.479 454.44 Tm +(,) Tj +1 0 0 1 72 442.56 Tm +(it) Tj +1 0 0 1 81.6 442.56 Tm +(starts) Tj +1 0 0 1 106.68 442.56 Tm +(a) Tj +1 0 0 1 115.2 442.56 Tm +(three-phase) Tj +1 0 0 1 165.24 442.56 Tm +(protocol) Tj +1 0 0 1 202.56 442.56 Tm +(to) Tj +1 0 0 1 214.319 442.56 Tm +(atomically) Tj +1 0 0 1 260.519 442.56 Tm +(multicast) Tj +1 0 0 1 72 430.56 Tm +(the) Tj +1 0 0 1 89.6399 430.56 Tm +(request) Tj +1 0 0 1 123.84 430.56 Tm +(to) Tj +1 0 0 1 137.04 430.56 Tm +(the) Tj +1 0 0 1 154.56 430.56 Tm +(replicas.) Tj +1 0 0 1 200.52 430.56 Tm +(The) Tj +1 0 0 1 221.4 430.56 Tm +(primary) Tj +1 0 0 1 258.359 430.56 Tm +(starts) Tj +1 0 0 1 284.879 430.56 Tm +(the) Tj +1 0 0 1 72 418.68 Tm +(protocol) Tj +1 0 0 1 109.92 418.68 Tm +(immediately) Tj +1 0 0 1 164.52 418.68 Tm +(unless) Tj +1 0 0 1 194.16 418.68 Tm +(the) Tj +1 0 0 1 211.08 418.68 Tm +(number) Tj +1 0 0 1 246.359 418.68 Tm +(of) Tj +1 0 0 1 259.319 418.68 Tm +(messages) Tj +1 0 0 1 72 406.68 Tm +(for) Tj +1 0 0 1 87.8399 406.68 Tm +(which) Tj +1 0 0 1 116.4 406.68 Tm +(the) Tj +1 0 0 1 132.84 406.68 Tm +(protocol) Tj +1 0 0 1 170.16 406.68 Tm +(is) Tj +1 0 0 1 181.08 406.68 Tm +(in) Tj +1 0 0 1 193.08 406.68 Tm +(progress) Tj +1 0 0 1 230.999 406.68 Tm +(e) Tj +1 0 0 1 235.319 406.68 Tm +(xceeds) Tj +1 0 0 1 266.639 406.68 Tm +(a) Tj +1 0 0 1 275.279 406.68 Tm +(gi) Tj +1 0 0 1 282.839 406.68 Tm +(v) Tj +1 0 0 1 287.759 406.68 Tm +(en) Tj +1 0 0 1 72 394.8 Tm +(maximum.) Tj +1 0 0 1 119.88 394.8 Tm +(In) Tj +1 0 0 1 131.16 394.8 Tm +(this) Tj +1 0 0 1 148.56 394.8 Tm +(case,) Tj +1 0 0 1 171.24 394.8 Tm +(it) Tj +1 0 0 1 179.76 394.8 Tm +(b) Tj +1 0 0 1 184.56 394.8 Tm +(uf) Tj +1 0 0 1 192.72 394.8 Tm +(fers) Tj +1 0 0 1 210.48 394.8 Tm +(the) Tj +1 0 0 1 225.6 394.8 Tm +(request.) Tj +1 0 0 1 261.839 394.8 Tm +(Buf) Tj +1 0 0 1 276.599 394.8 Tm +(fered) Tj +1 0 0 1 72 382.8 Tm +(requests) Tj +1 0 0 1 108.48 382.8 Tm +(are) Tj +1 0 0 1 124.44 382.8 Tm +(multicast) Tj +1 0 0 1 164.76 382.8 Tm +(later) Tj +1 0 0 1 186.36 382.8 Tm +(as) Tj +1 0 0 1 198.48 382.8 Tm +(a) Tj +1 0 0 1 206.64 382.8 Tm +(group) Tj +1 0 0 1 233.76 382.8 Tm +(to) Tj +1 0 0 1 245.279 382.8 Tm +(cut) Tj +1 0 0 1 261.359 382.8 Tm +(do) Tj +1 0 0 1 271.199 382.8 Tm +(wn) Tj +1 0 0 1 287.039 382.8 Tm +(on) Tj +1 0 0 1 72 370.92 Tm +(message) Tj +1 0 0 1 107.52 370.92 Tm +(traf) Tj +1 0 0 1 121.2 370.92 Tm +(\256c) Tj +1 0 0 1 132.84 370.92 Tm +(and) Tj +1 0 0 1 149.04 370.92 Tm +(CPU) Tj +1 0 0 1 169.92 370.92 Tm +(o) Tj +1 0 0 1 174.84 370.92 Tm +(v) Tj +1 0 0 1 179.76 370.92 Tm +(erheads) Tj +1 0 0 1 211.92 370.92 Tm +(un) Tj +1 0 0 1 221.88 370.92 Tm +(der) Tj +1 0 0 1 236.28 370.92 Tm +(h) Tj +1 0 0 1 241.199 370.92 Tm +(ea) Tj +1 0 0 1 249.839 370.92 Tm +(vy) Tj +1 0 0 1 261.479 370.92 Tm +(load) Tj +1 0 0 1 278.639 370.92 Tm +(;) Tj +1 0 0 1 282.599 370.92 Tm +(this) Tj +1 0 0 1 72 358.92 Tm +(optimization) Tj +1 0 0 1 124.32 358.92 Tm +(is) Tj +1 0 0 1 132.96 358.92 Tm +(similar) Tj +1 0 0 1 162.6 358.92 Tm +(to) Tj +1 0 0 1 172.32 358.92 Tm +(a) Tj +1 0 0 1 178.68 358.92 Tm +(group) Tj +1 0 0 1 204 358.92 Tm +(commit) Tj +1 0 0 1 236.4 358.92 Tm +(in) Tj +1 0 0 1 246.119 358.92 Tm +(transactional) Tj +1 0 0 1 72 347.04 Tm +(systems) Tj +1 0 0 1 105.84 347.04 Tm +([11) Tj +1 0 0 1 119.04 347.04 Tm +(].) Tj +1 0 0 1 128.4 347.04 Tm +(F) Tj +1 0 0 1 133.8 347.04 Tm +(or) Tj +1 0 0 1 144.36 347.04 Tm +(simplicity) Tj +1 0 0 1 183.72 347.04 Tm +(,) Tj +1 0 0 1 188.4 347.04 Tm +(we) Tj +1 0 0 1 202.2 347.04 Tm +(ignore) Tj +1 0 0 1 229.919 347.04 Tm +(this) Tj +1 0 0 1 246.599 347.04 Tm +(optimization) Tj +1 0 0 1 72 335.04 Tm +(in) Tj +1 0 0 1 82.2 335.04 Tm +(the) Tj +1 0 0 1 96.9599 335.04 Tm +(description) Tj +1 0 0 1 143.64 335.04 Tm +(belo) Tj +1 0 0 1 160.68 335.04 Tm +(w) Tj +1 0 0 1 167.28 335.04 Tm +(.) Tj +1 0 0 1 81.96 322.68 Tm +(The) Tj +1 0 0 1 99.1199 322.68 Tm +(three) Tj +1 0 0 1 120.72 322.68 Tm +(phases) Tj +1 0 0 1 148.92 322.68 Tm +(are) Tj +/R252 10 Tf +1 0 0 1 162.84 322.68 Tm +(pr) Tj +1 0 0 1 171.36 322.68 Tm +(e-pr) Tj +1 0 0 1 187.68 322.68 Tm +(epar) Tj +1 0 0 1 205.68 322.68 Tm +(e) Tj +1 0 0 1 210 322.68 Tm +(,) Tj +1 0 0 1 214.08 322.68 Tm +(pr) Tj +1 0 0 1 222.599 322.68 Tm +(epar) Tj +1 0 0 1 240.599 322.68 Tm +(e) Tj +1 0 0 1 244.919 322.68 Tm +(,) Tj +/R238 10 Tf +1 0 0 1 249.119 322.68 Tm +(and) Tj +/R252 10 Tf +1 0 0 1 265.199 322.68 Tm +(commit) Tj +/R238 10 Tf +1 0 0 1 294.599 322.68 Tm +(.) Tj +1 0 0 1 72 310.68 Tm +(The) Tj +1 0 0 1 91.0799 310.68 Tm +(pre-prepare) Tj +1 0 0 1 140.64 310.68 Tm +(and) Tj +1 0 0 1 158.76 310.68 Tm +(prepare) Tj +1 0 0 1 192.24 310.68 Tm +(phases) Tj +1 0 0 1 222.48 310.68 Tm +(are) Tj +1 0 0 1 238.2 310.68 Tm +(used) Tj +1 0 0 1 260.159 310.68 Tm +(to) Tj +1 0 0 1 271.559 310.68 Tm +(totally) Tj +1 0 0 1 72 298.8 Tm +(order) Tj +1 0 0 1 97.9199 298.8 Tm +(requests) Tj +1 0 0 1 135.36 298.8 Tm +(sent) Tj +1 0 0 1 156.24 298.8 Tm +(in) Tj +1 0 0 1 168.84 298.8 Tm +(the) Tj +1 0 0 1 185.88 298.8 Tm +(same) Tj +1 0 0 1 211.2 298.8 Tm +(vie) Tj +1 0 0 1 223.199 298.8 Tm +(w) Tj +1 0 0 1 235.199 298.8 Tm +(e) Tj +1 0 0 1 239.399 298.8 Tm +(v) Tj +1 0 0 1 244.319 298.8 Tm +(en) Tj +1 0 0 1 258.359 298.8 Tm +(when) Tj +1 0 0 1 284.879 298.8 Tm +(the) Tj +1 0 0 1 72 286.8 Tm +(primary) Tj +1 0 0 1 103.2 286.8 Tm +(,) Tj +1 0 0 1 111 286.8 Tm +(which) Tj +1 0 0 1 140.28 286.8 Tm +(proposes) Tj +1 0 0 1 180.6 286.8 Tm +(the) Tj +1 0 0 1 197.76 286.8 Tm +(ordering) Tj +1 0 0 1 236.4 286.8 Tm +(of) Tj +1 0 0 1 249.599 286.8 Tm +(requests,) Tj +1 0 0 1 290.399 286.8 Tm +(is) Tj +1 0 0 1 72 274.92 Tm +(f) Tj +1 0 0 1 75.24 274.92 Tm +(aulty) Tj +1 0 0 1 94.6799 274.92 Tm +(.) Tj +1 0 0 1 100.44 274.92 Tm +(The) Tj +1 0 0 1 117.84 274.92 Tm +(prepare) Tj +1 0 0 1 149.64 274.92 Tm +(and) Tj +1 0 0 1 166.08 274.92 Tm +(commit) Tj +1 0 0 1 198.48 274.92 Tm +(phases) Tj +1 0 0 1 226.92 274.92 Tm +(are) Tj +1 0 0 1 241.079 274.92 Tm +(used) Tj +1 0 0 1 261.239 274.92 Tm +(to) Tj +1 0 0 1 270.959 274.92 Tm +(ensure) Tj +1 0 0 1 72 262.92 Tm +(that) Tj +1 0 0 1 88.6799 262.92 Tm +(requests) Tj +1 0 0 1 123 262.92 Tm +(that) Tj +1 0 0 1 139.68 262.92 Tm +(commit) Tj +1 0 0 1 171.72 262.92 Tm +(are) Tj +1 0 0 1 185.64 262.92 Tm +(totally) Tj +1 0 0 1 212.759 262.92 Tm +(ordered) Tj +1 0 0 1 244.919 262.92 Tm +(across) Tj +1 0 0 1 271.559 262.92 Tm +(vie) Tj +1 0 0 1 283.559 262.92 Tm +(ws.) Tj +1 0 0 1 81.96 250.56 Tm +(In) Tj +1 0 0 1 96.7199 250.56 Tm +(the) Tj +1 0 0 1 115.44 250.56 Tm +(pre-prepare) Tj +1 0 0 1 167.88 250.56 Tm +(phase,) Tj +1 0 0 1 200.52 250.56 Tm +(the) Tj +1 0 0 1 219.24 250.56 Tm +(primary) Tj +1 0 0 1 257.279 250.56 Tm +(assigns) Tj +1 0 0 1 292.559 250.56 Tm +(a) Tj +1 0 0 1 72 238.68 Tm +(sequence) Tj +1 0 0 1 112.8 238.68 Tm +(number) Tj +1 0 0 1 143.16 238.68 Tm +(,) Tj +ET +endstream +endobj +262 0 obj +6490 +endobj +263 0 obj +<< /Type /XObject /Name /R263 /Subtype /Image /Length 264 0 R +/ImageMask true /Width 54 /Height 46 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 54 /BlackIs1 true >>] +>> +stream +4N]t,$m>7rl/"o@d!re, +pYTohhg]lis4_o!aocV_^Yur>rVsXfR^`8 +endstream +endobj +264 0 obj +102 +endobj +265 0 obj +<< /Length 266 0 R >> +stream +q +5.4 0 0 -4.6 150.2 243.2 cm +/R263 Do +Q +BT +1 0 0 1 156 238.68 Tm +(,) Tj +1 0 0 1 163.2 238.68 Tm +(to) Tj +1 0 0 1 175.2 238.68 Tm +(the) Tj +1 0 0 1 191.64 238.68 Tm +(request,) Tj +1 0 0 1 227.64 238.68 Tm +(multicasts) Tj +1 0 0 1 272.279 238.68 Tm +(a) Tj +1 0 0 1 280.919 238.68 Tm +(pre-) Tj +1 0 0 1 72 226.68 Tm +(prepare) Tj +1 0 0 1 104.28 226.68 Tm +(message) Tj +1 0 0 1 140.4 226.68 Tm +(with) Tj +ET +q +8.1 0 0 -4.6 160.6 231.2 cm +/R259 Do +Q +BT +1 0 0 1 171.6 226.68 Tm +(piggyback) Tj +1 0 0 1 213.36 226.68 Tm +(ed) Tj +1 0 0 1 224.76 226.68 Tm +(to) Tj +1 0 0 1 234.96 226.68 Tm +(all) Tj +1 0 0 1 247.319 226.68 Tm +(the) Tj +1 0 0 1 261.839 226.68 Tm +(backups,) Tj +1 0 0 1 72 214.8 Tm +(and) Tj +1 0 0 1 88.8 214.8 Tm +(appends) Tj +1 0 0 1 123.84 214.8 Tm +(the) Tj +1 0 0 1 138.36 214.8 Tm +(message) Tj +1 0 0 1 174.48 214.8 Tm +(to) Tj +1 0 0 1 184.56 214.8 Tm +(its) Tj +1 0 0 1 196.44 214.8 Tm +(log.) Tj +1 0 0 1 215.16 214.8 Tm +(The) Tj +1 0 0 1 233.04 214.8 Tm +(message) Tj +1 0 0 1 269.159 214.8 Tm +(has) Tj +1 0 0 1 284.879 214.8 Tm +(the) Tj +1 0 0 1 72 202.8 Tm +(form) Tj +ET +endstream +endobj +266 0 obj +1058 +endobj +267 0 obj +<< /Type /XObject /Name /R267 /Subtype /Image /Length 268 0 R +/ImageMask true /Width 24 /Height 99 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 24 /BlackIs1 true >>] +>> +stream ++t-07X8N5#0B"SKInTVGqHrM(f3_PN_YUJ4L&^prs)7cFqg[B")ufa0GPC"Nr'(,Kmt^kVHo1m8!<~> +endstream +endobj +268 0 obj +80 +endobj +269 0 obj +<< /Length 270 0 R >> +stream +q +2.4 0 0 -9.9 96.8 210.3 cm +/R267 Do +Q +q +2.4 0 0 -9.9 100.6 210.3 cm +/R267 Do +Q +BT +/R238 8 Tf +1 0 0 1 103.56 202.8 Tm +(PRE) Tj +/R238 10 Tf +1 0 0 1 118.2 202.8 Tm +(-) Tj +/R238 8 Tf +1 0 0 1 121.56 202.8 Tm +(PREP) Tj +1 0 0 1 139.92 202.8 Tm +(ARE) Tj +ET +endstream +endobj +270 0 obj +251 +endobj +271 0 obj +<< /Type /XObject /Name /R271 /Subtype /Image /Length 272 0 R +/ImageMask true /Width 13 /Height 30 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 13 /BlackIs1 true >>] +>> +stream +0g^A/h=&*equ,LR2Xj(@lh_"t-iec'~> +endstream +endobj +272 0 obj +33 +endobj +273 0 obj +<< /Length 274 0 R >> +stream +q +1.3 0 0 -3 156.6 203.9 cm +/R271 Do +Q +endstream +endobj +274 0 obj +39 +endobj +275 0 obj +<< /Type /XObject /Name /R275 /Subtype /Image /Length 276 0 R +/ImageMask true /Width 44 /Height 46 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 44 /BlackIs1 true >>] +>> +stream +2$jBq +&9R,hkFhTE_G0fRqg[T's0obU5O\AciSX+=pUN/*r',Z@_Z&BgfC8fKf@@X?]DE=DTC?6WJ-Z.*~> +endstream +endobj +276 0 obj +84 +endobj +277 0 obj +<< /Length 278 0 R >> +stream +q +4.4 0 0 -4.6 160.4 207.3 cm +/R275 Do +Q +q +1.3 0 0 -3 166.2 203.9 cm +/R271 Do +Q +q +5.4 0 0 -4.6 170 207.3 cm +/R263 Do +Q +q +1.3 0 0 -3 176.6 203.9 cm +/R271 Do +Q +endstream +endobj +278 0 obj +158 +endobj +279 0 obj +<< /Type /XObject /Name /R279 /Subtype /Image /Length 280 0 R +/ImageMask true /Width 49 /Height 71 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 49 /BlackIs1 true >>] +>> +stream +-PYiBJV"+0)UN3G',i&VcojQ(@sYoeKmuDG +s)&q9s*o7jJ)C$Fpj^'QJ%rnUr&sfAs+VQL0)H)pV[&D-@^p.P-,mCns*qra5O\@FpQ6S*nl.";J-Z~> +endstream +endobj +280 0 obj +117 +endobj +281 0 obj +<< /Length 282 0 R >> +stream +q +4.9 0 0 -7.1 180.5 209.8 cm +/R279 Do +Q +endstream +endobj +282 0 obj +41 +endobj +283 0 obj +<< /Type /XObject /Name /R283 /Subtype /Image /Length 284 0 R +/ImageMask true /Width 24 /Height 99 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 24 /BlackIs1 true >>] +>> +stream +02pd&htA/cs0qnbrP.\`mJhV5_u&o]hqrq@J+[15ImJQWpe1HjlFsc8R~> +endstream +endobj +284 0 obj +78 +endobj +285 0 obj +<< /Length 286 0 R >> +stream +q +2.4 0 0 -9.9 185.8 210.3 cm +/R283 Do +Q +endstream +endobj +286 0 obj +41 +endobj +287 0 obj +<< /Type /XObject /Name /R287 /Subtype /Image /Length 288 0 R +/ImageMask true /Width 42 /Height 31 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 42 /BlackIs1 true >>] +>> +stream +1`R)4"qZ&4%Y>/aKS+]fJ#a+Sp>>oFIt)n@>7^\DYeoBbLCXeP2$%q%J-Z~> +endstream +endobj +288 0 obj +61 +endobj +289 0 obj +<< /Length 290 0 R >> +stream +q +4.2 0 0 -3.1 189.4 204.4 cm +/R287 Do +Q +endstream +endobj +290 0 obj +41 +endobj +291 0 obj +<< /Type /XObject /Name /R291 /Subtype /Image /Length 292 0 R +/ImageMask true /Width 31 /Height 32 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 31 /BlackIs1 true >>] +>> +stream +-D:A;E<"bo0Kp%mK +endstream +endobj +292 0 obj +58 +endobj +293 0 obj +<< /Length 294 0 R >> +stream +q +3.1 0 0 -3.2 193.9 202.6 cm +/R291 Do +Q +q +1.3 0 0 -3 199.4 203.9 cm +/R271 Do +Q +q +8.1 0 0 -4.6 203.2 207.3 cm +/R259 Do +Q +q +2.4 0 0 -9.9 212.2 210.3 cm +/R283 Do +Q +BT +/R238 10 Tf +1 0 0 1 215.64 202.8 Tm +(,) Tj +1 0 0 1 223.08 202.8 Tm +(where) Tj +ET +q +4.4 0 0 -4.6 252.1 207.3 cm +/R275 Do +Q +BT +1 0 0 1 261.6 202.8 Tm +(indicates) Tj +1 0 0 1 72 190.92 Tm +(the) Tj +1 0 0 1 87.8399 190.92 Tm +(vie) Tj +1 0 0 1 99.8399 190.92 Tm +(w) Tj +1 0 0 1 110.64 190.92 Tm +(in) Tj +1 0 0 1 122.04 190.92 Tm +(which) Tj +1 0 0 1 150.12 190.92 Tm +(the) Tj +1 0 0 1 165.96 190.92 Tm +(message) Tj +1 0 0 1 203.4 190.92 Tm +(is) Tj +1 0 0 1 213.719 190.92 Tm +(being) Tj +1 0 0 1 239.519 190.92 Tm +(sent,) Tj +ET +q +8.1 0 0 -4.6 262.3 195.4 cm +/R259 Do +Q +BT +1 0 0 1 274.439 190.92 Tm +(is) Tj +1 0 0 1 284.879 190.92 Tm +(the) Tj +1 0 0 1 72 178.92 Tm +(client') Tj +1 0 0 1 96.9599 178.92 Tm +(s) Tj +1 0 0 1 103.32 178.92 Tm +(request) Tj +1 0 0 1 134.52 178.92 Tm +(message,) Tj +1 0 0 1 173.28 178.92 Tm +(and) Tj +ET +q +4.9 0 0 -7.1 190.5 185.9 cm +/R279 Do +Q +BT +1 0 0 1 197.88 178.92 Tm +(is) Tj +ET +q +8.1 0 0 -4.6 207.2 183.4 cm +/R259 Do +Q +BT +1 0 0 1 215.76 178.92 Tm +(') Tj +1 0 0 1 218.52 178.92 Tm +(s) Tj +1 0 0 1 224.88 178.92 Tm +(digest.) Tj +1 0 0 1 81.96 166.56 Tm +(Requests) Tj +1 0 0 1 122.64 166.56 Tm +(are) Tj +1 0 0 1 139.68 166.56 Tm +(not) Tj +1 0 0 1 157.08 166.56 Tm +(included) Tj +1 0 0 1 196.2 166.56 Tm +(in) Tj +1 0 0 1 208.68 166.56 Tm +(pre-prepare) Tj +1 0 0 1 259.319 166.56 Tm +(messages) Tj +1 0 0 1 72 154.56 Tm +(to) Tj +1 0 0 1 84.48 154.56 Tm +(k) Tj +1 0 0 1 89.3999 154.56 Tm +(eep) Tj +1 0 0 1 108 154.56 Tm +(them) Tj +1 0 0 1 132.72 154.56 Tm +(small.) Tj +1 0 0 1 167.28 154.56 Tm +(This) Tj +1 0 0 1 189.72 154.56 Tm +(is) Tj +1 0 0 1 201.12 154.56 Tm +(important) Tj +1 0 0 1 244.679 154.56 Tm +(because) Tj +1 0 0 1 280.919 154.56 Tm +(pre-) Tj +1 0 0 1 72 142.68 Tm +(prepare) Tj +1 0 0 1 104.04 142.68 Tm +(messages) Tj +1 0 0 1 143.76 142.68 Tm +(are) Tj +1 0 0 1 158.16 142.68 Tm +(used) Tj +1 0 0 1 178.56 142.68 Tm +(as) Tj +1 0 0 1 189 142.68 Tm +(a) Tj +1 0 0 1 195.6 142.68 Tm +(proof) Tj +1 0 0 1 219.24 142.68 Tm +(that) Tj +1 0 0 1 236.399 142.68 Tm +(the) Tj +1 0 0 1 250.679 142.68 Tm +(request) Tj +1 0 0 1 281.639 142.68 Tm +(w) Tj +1 0 0 1 288.719 142.68 Tm +(as) Tj +1 0 0 1 72 130.68 Tm +(assigned) Tj +1 0 0 1 108.96 130.68 Tm +(sequence) Tj +1 0 0 1 148.2 130.68 Tm +(number) Tj +ET +q +5.4 0 0 -4.6 181.5 135.2 cm +/R263 Do +Q +BT +1 0 0 1 189.96 130.68 Tm +(in) Tj +1 0 0 1 200.52 130.68 Tm +(vie) Tj +1 0 0 1 212.52 130.68 Tm +(w) Tj +ET +q +4.4 0 0 -4.6 222.4 135.2 cm +/R275 Do +Q +BT +1 0 0 1 230.16 130.68 Tm +(in) Tj +1 0 0 1 240.6 130.68 Tm +(vie) Tj +1 0 0 1 252.6 130.68 Tm +(w) Tj +1 0 0 1 262.439 130.68 Tm +(changes.) Tj +1 0 0 1 72 118.8 Tm +(Additionally) Tj +1 0 0 1 122.04 118.8 Tm +(,) Tj +1 0 0 1 128.52 118.8 Tm +(it) Tj +1 0 0 1 138 118.8 Tm +(decouples) Tj +1 0 0 1 181.68 118.8 Tm +(the) Tj +1 0 0 1 197.76 118.8 Tm +(protocol) Tj +1 0 0 1 234.96 118.8 Tm +(to) Tj +1 0 0 1 246.599 118.8 Tm +(totally) Tj +1 0 0 1 275.999 118.8 Tm +(order) Tj +1 0 0 1 72 106.8 Tm +(requests) Tj +1 0 0 1 107.52 106.8 Tm +(from) Tj +1 0 0 1 129.84 106.8 Tm +(the) Tj +1 0 0 1 144.96 106.8 Tm +(protocol) Tj +1 0 0 1 181.08 106.8 Tm +(to) Tj +1 0 0 1 191.76 106.8 Tm +(transmit) Tj +1 0 0 1 227.399 106.8 Tm +(the) Tj +1 0 0 1 242.399 106.8 Tm +(request) Tj +1 0 0 1 274.199 106.8 Tm +(to) Tj +1 0 0 1 284.879 106.8 Tm +(the) Tj +1 0 0 1 72 94.92 Tm +(replicas;) Tj +1 0 0 1 111 94.92 Tm +(allo) Tj +1 0 0 1 125.76 94.92 Tm +(wing) Tj +1 0 0 1 149.88 94.92 Tm +(us) Tj +1 0 0 1 163.08 94.92 Tm +(to) Tj +1 0 0 1 175.2 94.92 Tm +(use) Tj +1 0 0 1 192.72 94.92 Tm +(a) Tj +1 0 0 1 201.48 94.92 Tm +(transport) Tj +1 0 0 1 241.199 94.92 Tm +(optimized) Tj +1 0 0 1 285.359 94.92 Tm +(for) Tj +1 0 0 1 72 82.92 Tm +(small) Tj +1 0 0 1 97.7999 82.92 Tm +(messages) Tj +1 0 0 1 139.56 82.92 Tm +(for) Tj +1 0 0 1 155.28 82.92 Tm +(protocol) Tj +1 0 0 1 192.72 82.92 Tm +(messages) Tj +1 0 0 1 234.48 82.92 Tm +(and) Tj +1 0 0 1 252.959 82.92 Tm +(a) Tj +1 0 0 1 261.599 82.92 Tm +(transport) Tj +1 0 0 1 72 71.04 Tm +(optimized) Tj +1 0 0 1 114.36 71.04 Tm +(for) Tj +1 0 0 1 128.4 71.04 Tm +(lar) Tj +1 0 0 1 138.84 71.04 Tm +(ge) Tj +1 0 0 1 150.72 71.04 Tm +(messages) Tj +1 0 0 1 190.8 71.04 Tm +(for) Tj +1 0 0 1 204.96 71.04 Tm +(lar) Tj +1 0 0 1 215.399 71.04 Tm +(ge) Tj +1 0 0 1 227.159 71.04 Tm +(requests.) Tj +1 0 0 1 324.96 709.08 Tm +(A) Tj +1 0 0 1 334.68 709.08 Tm +(backup) Tj +1 0 0 1 365.88 709.08 Tm +(accepts) Tj +1 0 0 1 397.68 709.08 Tm +(a) Tj +1 0 0 1 404.64 709.08 Tm +(pre-prepare) Tj +1 0 0 1 453 709.08 Tm +(message) Tj +1 0 0 1 489.239 709.08 Tm +(pro) Tj +1 0 0 1 502.559 709.08 Tm +(vided:) Tj +ET +endstream +endobj +294 0 obj +4543 +endobj +295 0 obj +<< /Type /XObject /Name /R295 /Subtype /Image /Length 296 0 R +/ImageMask true /Width 39 /Height 39 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 39 /BlackIs1 true >>] +>> +stream +3f*u*F9LP-kW3[R5O&26s8VF=rNC"SR~> +endstream +endobj +296 0 obj +48 +endobj +297 0 obj +<< /Length 298 0 R >> +stream +q +3.9 0 0 -3.9 320.5 700.9 cm +/R295 Do +Q +BT +1 0 0 1 330 696.36 Tm +(the) Tj +1 0 0 1 347.28 696.36 Tm +(signatures) Tj +1 0 0 1 392.88 696.36 Tm +(in) Tj +1 0 0 1 405.84 696.36 Tm +(the) Tj +1 0 0 1 423.24 696.36 Tm +(request) Tj +1 0 0 1 457.2 696.36 Tm +(and) Tj +1 0 0 1 476.76 696.36 Tm +(the) Tj +1 0 0 1 494.039 696.36 Tm +(pre-prepare) Tj +1 0 0 1 330 684.36 Tm +(message) Tj +1 0 0 1 366.24 684.36 Tm +(are) Tj +1 0 0 1 380.88 684.36 Tm +(correct) Tj +1 0 0 1 411 684.36 Tm +(and) Tj +ET +q +4.9 0 0 -7.1 428.2 691.4 cm +/R279 Do +Q +BT +1 0 0 1 435.6 684.36 Tm +(is) Tj +1 0 0 1 444.72 684.36 Tm +(the) Tj +1 0 0 1 459.36 684.36 Tm +(digest) Tj +1 0 0 1 485.639 684.36 Tm +(for) Tj +ET +q +8.1 0 0 -4.6 500 688.9 cm +/R259 Do +Q +BT +1 0 0 1 508.559 684.36 Tm +(;) Tj +ET +q +3.9 0 0 -3.9 320.5 675.8 cm +/R295 Do +Q +BT +1 0 0 1 330 671.28 Tm +(it) Tj +1 0 0 1 337.92 671.28 Tm +(is) Tj +1 0 0 1 347.16 671.28 Tm +(in) Tj +1 0 0 1 357.36 671.28 Tm +(vie) Tj +1 0 0 1 369.36 671.28 Tm +(w) Tj +ET +q +4.4 0 0 -4.6 379.2 675.8 cm +/R275 Do +Q +BT +1 0 0 1 384.12 671.28 Tm +(;) Tj +ET +q +3.9 0 0 -3.9 320.5 662.7 cm +/R295 Do +Q +BT +1 0 0 1 330 658.2 Tm +(it) Tj +1 0 0 1 338.16 658.2 Tm +(has) Tj +1 0 0 1 354 658.2 Tm +(not) Tj +1 0 0 1 369.36 658.2 Tm +(accepted) Tj +1 0 0 1 406.8 658.2 Tm +(a) Tj +1 0 0 1 413.88 658.2 Tm +(pre-prepare) Tj +1 0 0 1 462.479 658.2 Tm +(message) Tj +1 0 0 1 498.839 658.2 Tm +(for) Tj +1 0 0 1 513.119 658.2 Tm +(vie) Tj +1 0 0 1 525.119 658.2 Tm +(w) Tj +ET +q +4.4 0 0 -4.6 535 662.7 cm +/R275 Do +Q +BT +1 0 0 1 330 646.2 Tm +(and) Tj +1 0 0 1 346.56 646.2 Tm +(sequence) Tj +1 0 0 1 385.32 646.2 Tm +(number) Tj +ET +q +5.4 0 0 -4.6 418.2 650.7 cm +/R263 Do +Q +BT +1 0 0 1 426.24 646.2 Tm +(containing) Tj +1 0 0 1 470.52 646.2 Tm +(a) Tj +1 0 0 1 477.12 646.2 Tm +(dif) Tj +1 0 0 1 488.04 646.2 Tm +(ferent) Tj +1 0 0 1 513.479 646.2 Tm +(digest;) Tj +ET +q +3.9 0 0 -3.9 320.5 637.6 cm +/R295 Do +Q +BT +1 0 0 1 330 633.12 Tm +(the) Tj +1 0 0 1 345.6 633.12 Tm +(sequence) Tj +1 0 0 1 385.68 633.12 Tm +(number) Tj +1 0 0 1 419.64 633.12 Tm +(in) Tj +1 0 0 1 430.92 633.12 Tm +(the) Tj +1 0 0 1 446.64 633.12 Tm +(pre-prepare) Tj +1 0 0 1 496.079 633.12 Tm +(message) Tj +1 0 0 1 533.399 633.12 Tm +(is) Tj +1 0 0 1 330 621.12 Tm +(between) Tj +1 0 0 1 365.4 621.12 Tm +(a) Tj +1 0 0 1 372.12 621.12 Tm +(lo) Tj +1 0 0 1 379.68 621.12 Tm +(w) Tj +1 0 0 1 389.16 621.12 Tm +(w) Tj +1 0 0 1 396.24 621.12 Tm +(ater) Tj +1 0 0 1 413.4 621.12 Tm +(mark,) Tj +ET +endstream +endobj +298 0 obj +2351 +endobj +299 0 obj +<< /Type /XObject /Name /R299 /Subtype /Image /Length 300 0 R +/ImageMask true /Width 50 /Height 71 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 50 /BlackIs1 true >>] +>> +stream +0Zl\u@0_OZ#Wp,3rBJ3g^\>2: +rV-&ursUj$gmgWn*E:5fO9!_pbg-FlJ[WfIqQ^[&,mZcqtL,7s*eVhs5pRo!WX>~> +endstream +endobj +300 0 obj +108 +endobj +301 0 obj +<< /Length 302 0 R >> +stream +q +5 0 0 -7.1 439.1 628.1 cm +/R299 Do +Q +BT +1 0 0 1 444.48 621.12 Tm +(,) Tj +1 0 0 1 449.28 621.12 Tm +(and) Tj +1 0 0 1 465.96 621.12 Tm +(a) Tj +1 0 0 1 472.679 621.12 Tm +(high) Tj +1 0 0 1 492.719 621.12 Tm +(w) Tj +1 0 0 1 499.799 621.12 Tm +(ater) Tj +1 0 0 1 517.079 621.12 Tm +(mark,) Tj +ET +endstream +endobj +302 0 obj +285 +endobj +303 0 obj +<< /Type /XObject /Name /R303 /Subtype /Image /Length 304 0 R +/ImageMask true /Width 84 /Height 69 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 84 /BlackIs1 true >>] +>> +stream +-D8e7)F+8D_$Y]c$*6AK +mN;?)rZ1b5h;A,.rpTgb+7B0HrVjk:r%\j3J,S4'^YeJZhr4S'r'0*,+8c$9*VY&PJY_cls1 +endstream +endobj +304 0 obj +102 +endobj +305 0 obj +<< /Length 306 0 R >> +stream +q +8.4 0 0 -6.9 330.3 616.1 cm +/R303 Do +Q +BT +1 0 0 1 339.12 609.24 Tm +(.) Tj +1 0 0 1 315 596.52 Tm +(The) Tj +1 0 0 1 337.08 596.52 Tm +(last) Tj +1 0 0 1 357.48 596.52 Tm +(condition) Tj +1 0 0 1 401.76 596.52 Tm +(pre) Tj +1 0 0 1 414.36 596.52 Tm +(v) Tj +1 0 0 1 419.28 596.52 Tm +(ents) Tj +1 0 0 1 441.72 596.52 Tm +(a) Tj +1 0 0 1 452.76 596.52 Tm +(f) Tj +1 0 0 1 456 596.52 Tm +(aulty) Tj +1 0 0 1 482.519 596.52 Tm +(primary) Tj +1 0 0 1 520.679 596.52 Tm +(from) Tj +1 0 0 1 315 584.52 Tm +(e) Tj +1 0 0 1 319.32 584.52 Tm +(xhausting) Tj +1 0 0 1 361.32 584.52 Tm +(the) Tj +1 0 0 1 376.8 584.52 Tm +(space) Tj +1 0 0 1 402.24 584.52 Tm +(of) Tj +1 0 0 1 413.88 584.52 Tm +(sequence) Tj +1 0 0 1 453.72 584.52 Tm +(numbers) Tj +1 0 0 1 491.279 584.52 Tm +(by) Tj +1 0 0 1 504.599 584.52 Tm +(selecting) Tj +1 0 0 1 315 572.64 Tm +(a) Tj +1 0 0 1 322.68 572.64 Tm +(v) Tj +1 0 0 1 327.6 572.64 Tm +(ery) Tj +1 0 0 1 343.44 572.64 Tm +(lar) Tj +1 0 0 1 353.88 572.64 Tm +(ge) Tj +1 0 0 1 366.48 572.64 Tm +(one.) Tj +1 0 0 1 389.16 572.64 Tm +(W) Tj +1 0 0 1 397.68 572.64 Tm +(e) Tj +1 0 0 1 405.48 572.64 Tm +(discuss) Tj +1 0 0 1 437.52 572.64 Tm +(ho) Tj +1 0 0 1 447.36 572.64 Tm +(w) Tj +ET +q +8.4 0 0 -6.9 458 579.5 cm +/R303 Do +Q +BT +1 0 0 1 470.04 572.64 Tm +(and) Tj +ET +q +5 0 0 -7.1 488.1 579.6 cm +/R299 Do +Q +BT +1 0 0 1 496.559 572.64 Tm +(adv) Tj +1 0 0 1 510.839 572.64 Tm +(ance) Tj +1 0 0 1 532.199 572.64 Tm +(in) Tj +1 0 0 1 315 560.64 Tm +(Section) Tj +1 0 0 1 347.4 560.64 Tm +(4.3.) Tj +1 0 0 1 324.96 548.76 Tm +(If) Tj +1 0 0 1 334.56 548.76 Tm +(backup) Tj +ET +endstream +endobj +306 0 obj +1512 +endobj +307 0 obj +<< /Type /XObject /Name /R307 /Subtype /Image /Length 308 0 R +/ImageMask true /Width 26 /Height 69 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 26 /BlackIs1 true >>] +>> +stream +-R^*Zl/YrmIf:DfhtVWo^UNqc%K6'1s*addX2"DHrSQB/+4u(tp(mUa[G\kks2oQE"5iR%(OubkJ,~> +endstream +endobj +308 0 obj +80 +endobj +309 0 obj +<< /Length 310 0 R >> +stream +q +2.6 0 0 -6.9 366.6 555.6 cm +/R307 Do +Q +BT +1 0 0 1 372.72 548.76 Tm +(accepts) Tj +1 0 0 1 405 548.76 Tm +(the) Tj +ET +q +2.4 0 0 -9.9 421.1 556.3 cm +/R267 Do +Q +q +2.4 0 0 -9.9 425 556.3 cm +/R267 Do +Q +BT +/R238 8 Tf +1 0 0 1 427.92 548.76 Tm +(PRE) Tj +/R238 10 Tf +1 0 0 1 442.56 548.76 Tm +(-) Tj +/R238 8 Tf +1 0 0 1 445.92 548.76 Tm +(PREP) Tj +1 0 0 1 464.279 548.76 Tm +(ARE) Tj +ET +q +1.3 0 0 -3 480.9 549.9 cm +/R271 Do +Q +q +4.4 0 0 -4.6 484.8 553.3 cm +/R275 Do +Q +q +1.3 0 0 -3 490.5 549.9 cm +/R271 Do +Q +q +5.4 0 0 -4.6 494.4 553.3 cm +/R263 Do +Q +q +1.3 0 0 -3 501 549.9 cm +/R271 Do +Q +q +4.9 0 0 -7.1 504.9 555.8 cm +/R279 Do +Q +q +2.4 0 0 -9.9 510.2 556.3 cm +/R283 Do +Q +q +4.2 0 0 -3.1 513.8 550.3 cm +/R287 Do +Q +q +3.1 0 0 -3.2 518.3 548.6 cm +/R291 Do +Q +q +1.3 0 0 -3 523.8 549.9 cm +/R271 Do +Q +q +8.1 0 0 -4.6 527.6 553.3 cm +/R259 Do +Q +q +2.4 0 0 -9.9 536.6 556.3 cm +/R283 Do +Q +BT +/R238 10 Tf +1 0 0 1 315 536.76 Tm +(message,) Tj +1 0 0 1 355.32 536.76 Tm +(it) Tj +1 0 0 1 364.56 536.76 Tm +(enters) Tj +1 0 0 1 392.04 536.76 Tm +(the) Tj +/R252 10 Tf +1 0 0 1 408 536.76 Tm +(pr) Tj +1 0 0 1 416.52 536.76 Tm +(epar) Tj +1 0 0 1 434.52 536.76 Tm +(e) Tj +/R238 10 Tf +1 0 0 1 442.44 536.76 Tm +(phase) Tj +1 0 0 1 468.84 536.76 Tm +(by) Tj +1 0 0 1 482.639 536.76 Tm +(multicasting) Tj +1 0 0 1 535.559 536.76 Tm +(a) Tj +ET +q +2.4 0 0 -9.9 315.9 532.4 cm +/R267 Do +Q +BT +/R238 8 Tf +1 0 0 1 318.84 524.88 Tm +(PREP) Tj +1 0 0 1 337.2 524.88 Tm +(ARE) Tj +ET +q +1.3 0 0 -3 354 526 cm +/R271 Do +Q +q +4.4 0 0 -4.6 357.8 529.4 cm +/R275 Do +Q +q +1.3 0 0 -3 363.6 526 cm +/R271 Do +Q +q +5.4 0 0 -4.6 367.4 529.4 cm +/R263 Do +Q +q +1.3 0 0 -3 374 526 cm +/R271 Do +Q +q +4.9 0 0 -7.1 377.9 531.9 cm +/R279 Do +Q +q +1.3 0 0 -3 383.6 526 cm +/R271 Do +Q +q +2.6 0 0 -6.9 387.4 531.7 cm +/R307 Do +Q +q +2.4 0 0 -9.9 391.1 532.4 cm +/R283 Do +Q +q +4.2 0 0 -3.1 394.8 526.4 cm +/R287 Do +Q +endstream +endobj +310 0 obj +1795 +endobj +311 0 obj +<< /Type /XObject /Name /R311 /Subtype /Image /Length 312 0 R +/ImageMask true /Width 17 /Height 35 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 17 /BlackIs1 true >>] +>> +stream +4B*=`S,;]mpj7Y*s1(">iS6Nohi?6cs!N02p(RKl"9~> +endstream +endobj +312 0 obj +45 +endobj +313 0 obj +<< /Length 314 0 R >> +stream +q +1.7 0 0 -3.5 399.4 526 cm +/R311 Do +Q +BT +/R238 10 Tf +1 0 0 1 406.08 524.88 Tm +(message) Tj +1 0 0 1 443.16 524.88 Tm +(to) Tj +1 0 0 1 454.2 524.88 Tm +(all) Tj +1 0 0 1 467.52 524.88 Tm +(other) Tj +1 0 0 1 491.28 524.88 Tm +(replicas) Tj +1 0 0 1 525.6 524.88 Tm +(and) Tj +1 0 0 1 315 512.88 Tm +(adds) Tj +1 0 0 1 335.16 512.88 Tm +(both) Tj +1 0 0 1 354.84 512.88 Tm +(messages) Tj +1 0 0 1 394.32 512.88 Tm +(to) Tj +1 0 0 1 404.04 512.88 Tm +(its) Tj +1 0 0 1 415.32 512.88 Tm +(log.) Tj +1 0 0 1 433.92 512.88 Tm +(Otherwise,) Tj +1 0 0 1 479.399 512.88 Tm +(it) Tj +1 0 0 1 486.839 512.88 Tm +(does) Tj +1 0 0 1 507.119 512.88 Tm +(nothing.) Tj +1 0 0 1 324.96 501 Tm +(A) Tj +1 0 0 1 337.8 501 Tm +(replica) Tj +1 0 0 1 370.44 501 Tm +(\(including) Tj +1 0 0 1 417 501 Tm +(the) Tj +1 0 0 1 434.76 501 Tm +(primary\)) Tj +1 0 0 1 475.2 501 Tm +(accepts) Tj +1 0 0 1 510.119 501 Tm +(prepare) Tj +1 0 0 1 315 489 Tm +(messages) Tj +1 0 0 1 359.04 489 Tm +(and) Tj +1 0 0 1 379.8 489 Tm +(adds) Tj +1 0 0 1 404.52 489 Tm +(them) Tj +1 0 0 1 430.92 489 Tm +(to) Tj +1 0 0 1 445.08 489 Tm +(its) Tj +1 0 0 1 460.92 489 Tm +(log) Tj +1 0 0 1 479.999 489 Tm +(pro) Tj +1 0 0 1 493.319 489 Tm +(vided) Tj +1 0 0 1 521.759 489 Tm +(their) Tj +1 0 0 1 315 477.12 Tm +(signatures) Tj +1 0 0 1 360.6 477.12 Tm +(are) Tj +1 0 0 1 377.88 477.12 Tm +(correct,) Tj +1 0 0 1 413.88 477.12 Tm +(their) Tj +1 0 0 1 437.28 477.12 Tm +(vie) Tj +1 0 0 1 449.28 477.12 Tm +(w) Tj +1 0 0 1 461.64 477.12 Tm +(number) Tj +1 0 0 1 497.159 477.12 Tm +(equals) Tj +1 0 0 1 527.879 477.12 Tm +(the) Tj +1 0 0 1 315 465.12 Tm +(replica') Tj +1 0 0 1 345 465.12 Tm +(s) Tj +1 0 0 1 353.76 465.12 Tm +(current) Tj +1 0 0 1 387 465.12 Tm +(vie) Tj +1 0 0 1 399 465.12 Tm +(w) Tj +1 0 0 1 405.6 465.12 Tm +(,) Tj +1 0 0 1 413.64 465.12 Tm +(and) Tj +1 0 0 1 433.08 465.12 Tm +(their) Tj +1 0 0 1 456.36 465.12 Tm +(sequence) Tj +1 0 0 1 497.879 465.12 Tm +(number) Tj +1 0 0 1 533.399 465.12 Tm +(is) Tj +1 0 0 1 315 453.24 Tm +(between) Tj +ET +q +5 0 0 -7.1 351 460.2 cm +/R299 Do +Q +BT +1 0 0 1 358.92 453.24 Tm +(and) Tj +ET +q +8.4 0 0 -6.9 376.1 460.1 cm +/R303 Do +Q +BT +1 0 0 1 384.96 453.24 Tm +(.) Tj +1 0 0 1 324.96 441.24 Tm +(W) Tj +1 0 0 1 333.48 441.24 Tm +(e) Tj +1 0 0 1 340.32 441.24 Tm +(de\256ne) Tj +1 0 0 1 366.96 441.24 Tm +(the) Tj +1 0 0 1 381.48 441.24 Tm +(predicate) Tj +/R252 10 Tf +1 0 0 1 420.24 441.24 Tm +(pr) Tj +1 0 0 1 428.76 441.24 Tm +(epar) Tj +1 0 0 1 446.76 441.24 Tm +(ed) Tj +ET +endstream +endobj +314 0 obj +2362 +endobj +315 0 obj +<< /Type /XObject /Name /R315 /Subtype /Image /Length 316 0 R +/ImageMask true /Width 24 /Height 99 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 24 /BlackIs1 true >>] +>> +stream ++rF]%*LO8IHL@*j8[taqp!<~> +endstream +endobj +316 0 obj +80 +endobj +317 0 obj +<< /Length 318 0 R >> +stream +q +2.4 0 0 -9.9 457.1 448.7 cm +/R315 Do +Q +q +8.1 0 0 -4.6 460.3 445.7 cm +/R259 Do +Q +q +1.3 0 0 -3 469.6 442.3 cm +/R271 Do +Q +q +4.4 0 0 -4.6 473.5 445.7 cm +/R275 Do +Q +q +1.3 0 0 -3 479.2 442.3 cm +/R271 Do +Q +q +5.4 0 0 -4.6 483.1 445.7 cm +/R263 Do +Q +q +1.3 0 0 -3 489.7 442.3 cm +/R271 Do +Q +q +2.6 0 0 -6.9 493.5 448 cm +/R307 Do +Q +endstream +endobj +318 0 obj +320 +endobj +319 0 obj +<< /Type /XObject /Name /R319 /Subtype /Image /Length 320 0 R +/ImageMask true /Width 24 /Height 99 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 24 /BlackIs1 true >>] +>> +stream +,JEO$loTRL4niC`hqS$`rnmKps6nkFp\Y!js6KW5s0(&RK)W@\(KK(`l\2@U?Bm&k<5,",!$D~> +endstream +endobj +320 0 obj +76 +endobj +321 0 obj +<< /Length 322 0 R >> +stream +q +2.4 0 0 -9.9 497.2 448.7 cm +/R319 Do +Q +BT +/R238 10 Tf +1 0 0 1 502.679 441.24 Tm +(to) Tj +1 0 0 1 512.759 441.24 Tm +(be) Tj +1 0 0 1 524.519 441.24 Tm +(true) Tj +1 0 0 1 315 429.24 Tm +(if) Tj +1 0 0 1 323.64 429.24 Tm +(and) Tj +1 0 0 1 340.56 429.24 Tm +(only) Tj +1 0 0 1 360.84 429.24 Tm +(if) Tj +1 0 0 1 369.48 429.24 Tm +(replica) Tj +ET +q +2.6 0 0 -6.9 399.3 436 cm +/R307 Do +Q +BT +1 0 0 1 405.12 429.24 Tm +(has) Tj +1 0 0 1 420.96 429.24 Tm +(inserted) Tj +1 0 0 1 455.039 429.24 Tm +(in) Tj +1 0 0 1 465.359 429.24 Tm +(its) Tj +1 0 0 1 477.239 429.24 Tm +(log:) Tj +1 0 0 1 496.559 429.24 Tm +(the) Tj +1 0 0 1 511.199 429.24 Tm +(request) Tj +ET +q +8.1 0 0 -4.6 315.2 421.9 cm +/R259 Do +Q +BT +1 0 0 1 323.76 417.36 Tm +(,) Tj +1 0 0 1 328.92 417.36 Tm +(a) Tj +1 0 0 1 336 417.36 Tm +(pre-prepare) Tj +1 0 0 1 384.48 417.36 Tm +(for) Tj +ET +q +8.1 0 0 -4.6 399 421.9 cm +/R259 Do +Q +BT +1 0 0 1 410.16 417.36 Tm +(in) Tj +1 0 0 1 420.48 417.36 Tm +(vie) Tj +1 0 0 1 432.48 417.36 Tm +(w) Tj +ET +q +4.4 0 0 -4.6 442.4 421.9 cm +/R275 Do +Q +BT +1 0 0 1 450.12 417.36 Tm +(with) Tj +1 0 0 1 470.399 417.36 Tm +(sequence) Tj +1 0 0 1 509.519 417.36 Tm +(number) Tj +ET +q +5.4 0 0 -4.6 315.2 409.9 cm +/R263 Do +Q +BT +1 0 0 1 321 405.36 Tm +(,) Tj +1 0 0 1 327.84 405.36 Tm +(and) Tj +1 0 0 1 346.2 405.36 Tm +(2) Tj +ET +q +5 0 0 -9.2 351.6 412.5 cm +/R245 Do +Q +BT +1 0 0 1 361.08 405.36 Tm +(prepares) Tj +1 0 0 1 398.76 405.36 Tm +(from) Tj +1 0 0 1 422.16 405.36 Tm +(dif) Tj +1 0 0 1 433.08 405.36 Tm +(ferent) Tj +1 0 0 1 460.08 405.36 Tm +(backups) Tj +1 0 0 1 496.679 405.36 Tm +(that) Tj +1 0 0 1 515.639 405.36 Tm +(match) Tj +1 0 0 1 315 393.48 Tm +(the) Tj +1 0 0 1 329.28 393.48 Tm +(pre-prepare.) Tj +1 0 0 1 381.24 393.48 Tm +(The) Tj +1 0 0 1 398.76 393.48 Tm +(replicas) Tj +1 0 0 1 431.88 393.48 Tm +(v) Tj +1 0 0 1 436.8 393.48 Tm +(erify) Tj +1 0 0 1 457.68 393.48 Tm +(whether) Tj +1 0 0 1 492 393.48 Tm +(the) Tj +1 0 0 1 506.279 393.48 Tm +(prepares) Tj +1 0 0 1 315 381.48 Tm +(match) Tj +1 0 0 1 343.56 381.48 Tm +(the) Tj +1 0 0 1 360 381.48 Tm +(pre-prepare) Tj +1 0 0 1 410.16 381.48 Tm +(by) Tj +1 0 0 1 424.44 381.48 Tm +(checking) Tj +1 0 0 1 464.64 381.48 Tm +(that) Tj +1 0 0 1 483.839 381.48 Tm +(the) Tj +1 0 0 1 495.959 381.48 Tm +(y) Tj +1 0 0 1 505.079 381.48 Tm +(ha) Tj +1 0 0 1 514.319 381.48 Tm +(v) Tj +1 0 0 1 519.239 381.48 Tm +(e) Tj +1 0 0 1 527.879 381.48 Tm +(the) Tj +1 0 0 1 315 369.6 Tm +(same) Tj +1 0 0 1 337.92 369.6 Tm +(vie) Tj +1 0 0 1 349.92 369.6 Tm +(w) Tj +1 0 0 1 356.52 369.6 Tm +(,) Tj +1 0 0 1 361.44 369.6 Tm +(sequence) Tj +1 0 0 1 400.44 369.6 Tm +(number) Tj +1 0 0 1 430.8 369.6 Tm +(,) Tj +1 0 0 1 435.48 369.6 Tm +(and) Tj +1 0 0 1 452.28 369.6 Tm +(digest.) Tj +1 0 0 1 324.96 357.6 Tm +(The) Tj +1 0 0 1 343.8 357.6 Tm +(pre-prepare) Tj +1 0 0 1 393.12 357.6 Tm +(and) Tj +1 0 0 1 410.88 357.6 Tm +(prepare) Tj +1 0 0 1 444.12 357.6 Tm +(phases) Tj +1 0 0 1 473.999 357.6 Tm +(of) Tj +1 0 0 1 485.759 357.6 Tm +(the) Tj +1 0 0 1 501.239 357.6 Tm +(algorithm) Tj +1 0 0 1 315 345.72 Tm +(guarantee) Tj +1 0 0 1 356.76 345.72 Tm +(that) Tj +1 0 0 1 374.76 345.72 Tm +(non-f) Tj +1 0 0 1 396.48 345.72 Tm +(aulty) Tj +1 0 0 1 419.16 345.72 Tm +(replicas) Tj +1 0 0 1 453.24 345.72 Tm +(agree) Tj +1 0 0 1 477.839 345.72 Tm +(on) Tj +1 0 0 1 490.799 345.72 Tm +(a) Tj +1 0 0 1 498.239 345.72 Tm +(total) Tj +1 0 0 1 518.999 345.72 Tm +(order) Tj +1 0 0 1 315 333.72 Tm +(for) Tj +1 0 0 1 331.08 333.72 Tm +(the) Tj +1 0 0 1 347.76 333.72 Tm +(requests) Tj +1 0 0 1 384.84 333.72 Tm +(within) Tj +1 0 0 1 414.84 333.72 Tm +(a) Tj +1 0 0 1 423.72 333.72 Tm +(vie) Tj +1 0 0 1 435.72 333.72 Tm +(w) Tj +1 0 0 1 442.32 333.72 Tm +(.) Tj +1 0 0 1 454.2 333.72 Tm +(More) Tj +1 0 0 1 480.239 333.72 Tm +(precisely) Tj +1 0 0 1 515.759 333.72 Tm +(,) Tj +1 0 0 1 522.959 333.72 Tm +(the) Tj +1 0 0 1 535.079 333.72 Tm +(y) Tj +1 0 0 1 315 321.84 Tm +(ensure) Tj +1 0 0 1 343.8 321.84 Tm +(the) Tj +1 0 0 1 358.68 321.84 Tm +(follo) Tj +1 0 0 1 377.4 321.84 Tm +(wing) Tj +1 0 0 1 399.96 321.84 Tm +(in) Tj +1 0 0 1 407.4 321.84 Tm +(v) Tj +1 0 0 1 412.2 321.84 Tm +(ariant:) Tj +1 0 0 1 441.6 321.84 Tm +(if) Tj +/R252 10 Tf +1 0 0 1 450.48 321.84 Tm +(pr) Tj +1 0 0 1 459 321.84 Tm +(epar) Tj +1 0 0 1 476.999 321.84 Tm +(ed) Tj +ET +q +2.4 0 0 -9.9 487.4 329.3 cm +/R315 Do +Q +q +8.1 0 0 -4.6 490.5 326.3 cm +/R259 Do +Q +q +1.3 0 0 -3 499.9 322.9 cm +/R271 Do +Q +q +4.4 0 0 -4.6 503.7 326.3 cm +/R275 Do +Q +q +1.3 0 0 -3 509.5 322.9 cm +/R271 Do +Q +q +5.4 0 0 -4.6 513.3 326.3 cm +/R263 Do +Q +q +1.3 0 0 -3 519.9 322.9 cm +/R271 Do +Q +q +2.6 0 0 -6.9 523.8 328.6 cm +/R307 Do +Q +q +2.4 0 0 -9.9 527.4 329.3 cm +/R319 Do +Q +BT +/R238 10 Tf +1 0 0 1 533.399 321.84 Tm +(is) Tj +1 0 0 1 315 309.84 Tm +(true) Tj +1 0 0 1 332.76 309.84 Tm +(then) Tj +/R252 10 Tf +1 0 0 1 352.08 309.84 Tm +(pr) Tj +1 0 0 1 360.6 309.84 Tm +(epar) Tj +1 0 0 1 378.6 309.84 Tm +(ed) Tj +ET +q +2.4 0 0 -9.9 389 317.3 cm +/R315 Do +Q +q +8.1 0 0 -4.6 392.1 314.3 cm +/R259 Do +Q +endstream +endobj +322 0 obj +4746 +endobj +323 0 obj +<< /Type /XObject /Name /R323 /Subtype /Image /Length 324 0 R +/ImageMask true /Width 17 /Height 37 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 17 /BlackIs1 true >>] +>> +stream +2p&%Irsf"qs+g(;pODr+p\l"Z@=(fO!"]~> +endstream +endobj +324 0 obj +36 +endobj +325 0 obj +<< /Length 326 0 R >> +stream +q +1.7 0 0 -3.7 401 317.3 cm +/R323 Do +Q +q +1.3 0 0 -3 404.1 310.9 cm +/R271 Do +Q +q +4.4 0 0 -4.6 408 314.3 cm +/R275 Do +Q +q +1.3 0 0 -3 413.7 310.9 cm +/R271 Do +Q +q +5.4 0 0 -4.6 417.6 314.3 cm +/R263 Do +Q +q +1.3 0 0 -3 424.2 310.9 cm +/R271 Do +Q +endstream +endobj +326 0 obj +236 +endobj +327 0 obj +<< /Type /XObject /Name /R327 /Subtype /Image /Length 328 0 R +/ImageMask true /Width 41 /Height 89 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 41 /BlackIs1 true >>] +>> +stream +3%oqNe0+KR +8.TS3gNRT8m0C&=*WH)RhtR$Is*qrFs*qfPpZ+:-rpTR[p\3 +endstream +endobj +328 0 obj +92 +endobj +329 0 obj +<< /Length 330 0 R >> +stream +q +4.1 0 0 -8.9 427.6 316.6 cm +/R327 Do +Q +q +2.4 0 0 -9.9 432.9 317.3 cm +/R319 Do +Q +BT +/R238 10 Tf +1 0 0 1 438.48 309.84 Tm +(is) Tj +1 0 0 1 447.36 309.84 Tm +(f) Tj +1 0 0 1 450.6 309.84 Tm +(alse) Tj +1 0 0 1 468.36 309.84 Tm +(for) Tj +1 0 0 1 482.16 309.84 Tm +(an) Tj +1 0 0 1 491.52 309.84 Tm +(y) Tj +1 0 0 1 498.6 309.84 Tm +(non-f) Tj +1 0 0 1 520.319 309.84 Tm +(aulty) Tj +1 0 0 1 315 297.96 Tm +(replica) Tj +ET +q +4.1 0 0 -8.9 343.7 304.8 cm +/R327 Do +Q +BT +1 0 0 1 350.28 297.96 Tm +(\(including) Tj +ET +q +2.6 0 0 -6.9 393.2 304.8 cm +/R307 Do +Q +endstream +endobj +330 0 obj +535 +endobj +331 0 obj +<< /Type /XObject /Name /R331 /Subtype /Image /Length 332 0 R +/ImageMask true /Width 66 /Height 27 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 66 /BlackIs1 true >>] +>> +stream +/e%CP4)j*Wq&BoPm6CdS5Q~> +endstream +endobj +332 0 obj +25 +endobj +333 0 obj +<< /Length 334 0 R >> +stream +q +6.6 0 0 -2.7 399.7 301.9 cm +/R331 Do +Q +q +4.1 0 0 -8.9 409.5 304.8 cm +/R327 Do +Q +BT +1 0 0 1 414.36 297.96 Tm +(\)) Tj +1 0 0 1 419.52 297.96 Tm +(and) Tj +1 0 0 1 435.599 297.96 Tm +(an) Tj +1 0 0 1 444.959 297.96 Tm +(y) Tj +ET +q +8.1 0 0 -4.6 451.9 302.5 cm +/R259 Do +Q +q +1.7 0 0 -3.7 460.7 305.5 cm +/R323 Do +Q +BT +1 0 0 1 464.88 297.96 Tm +(such) Tj +1 0 0 1 484.92 297.96 Tm +(that) Tj +ET +endstream +endobj +334 0 obj +380 +endobj +335 0 obj +<< /Type /XObject /Name /R335 /Subtype /Image /Length 336 0 R +/ImageMask true /Width 75 /Height 69 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 75 /BlackIs1 true >>] +>> +stream +-D@i/="t3q:)'T:EY'Q7@JfCQK73IN +DqmDb@DDO(rqCj@Fs$I.iOf--It,63^V:<:qnMn*^\@b0s+Zm:ht=@d?EiAlYY3Kq%(IruTE5*9~> +endstream +endobj +336 0 obj +109 +endobj +337 0 obj +<< /Length 338 0 R >> +stream +q +7.5 0 0 -6.9 502 304.9 cm +/R335 Do +Q +q +2.4 0 0 -9.9 511.1 305.5 cm +/R315 Do +Q +q +8.1 0 0 -4.6 514.3 302.5 cm +/R259 Do +Q +q +1.7 0 0 -3.7 523.1 305.5 cm +/R323 Do +Q +q +2.4 0 0 -9.9 526 305.5 cm +/R319 Do +Q +endstream +endobj +338 0 obj +201 +endobj +339 0 obj +<< /Type /XObject /Name /R339 /Subtype /Image /Length 340 0 R +/ImageMask true /Width 51 /Height 94 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 51 /BlackIs1 true >>] +>> +stream +,Ll]7 +h=lC.rSQ7WGMW0U^L?$SqnImbmtZC2mf"1&^)d0e]D-=4pilH2n):tW%fYAApCHZh48XMGLM?f3!W~> +endstream +endobj +340 0 obj +86 +endobj +341 0 obj +<< /Length 342 0 R >> +stream +q +5.1 0 0 -9.4 533.5 305.2 cm +/R339 Do +Q +q +6.6 0 0 -2.7 532.7 301.9 cm +/R331 Do +Q +q +7.5 0 0 -6.9 315.3 292.9 cm +/R335 Do +Q +q +2.4 0 0 -9.9 324.4 293.5 cm +/R315 Do +Q +q +8.1 0 0 -4.6 327.6 290.5 cm +/R259 Do +Q +q +2.4 0 0 -9.9 336.5 293.5 cm +/R319 Do +Q +BT +1 0 0 1 339.96 285.96 Tm +(.) Tj +1 0 0 1 351.96 285.96 Tm +(This) Tj +1 0 0 1 374.04 285.96 Tm +(is) Tj +1 0 0 1 385.08 285.96 Tm +(true) Tj +1 0 0 1 405.12 285.96 Tm +(because) Tj +/R252 10 Tf +1 0 0 1 441 285.96 Tm +(pr) Tj +1 0 0 1 449.52 285.96 Tm +(epar) Tj +1 0 0 1 467.52 285.96 Tm +(ed) Tj +ET +q +2.4 0 0 -9.9 477.9 293.5 cm +/R315 Do +Q +q +8.1 0 0 -4.6 481 290.5 cm +/R259 Do +Q +q +1.3 0 0 -3 490.4 287.1 cm +/R271 Do +Q +q +4.4 0 0 -4.6 494.2 290.5 cm +/R275 Do +Q +q +1.3 0 0 -3 500 287.1 cm +/R271 Do +Q +q +5.4 0 0 -4.6 503.8 290.5 cm +/R263 Do +Q +q +1.3 0 0 -3 510.4 287.1 cm +/R271 Do +Q +q +2.6 0 0 -6.9 514.3 292.8 cm +/R307 Do +Q +q +2.4 0 0 -9.9 518 293.5 cm +/R319 Do +Q +BT +/R238 10 Tf +1 0 0 1 525.599 285.96 Tm +(and) Tj +ET +endstream +endobj +342 0 obj +945 +endobj +343 0 obj +<< /Type /XObject /Name /R343 /Subtype /Image /Length 344 0 R +/ImageMask true /Width 6 /Height 100 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 6 /BlackIs1 true >>] +>> +stream +02qoNs8W-!s8W-!s8W-!s8W-!s8W-!s8Uf'#QTA~> +endstream +endobj +344 0 obj +42 +endobj +345 0 obj +<< /Length 346 0 R >> +stream +q +0.6 0 0 -10 316 281.6 cm +/R343 Do +Q +endstream +endobj +346 0 obj +38 +endobj +347 0 obj +<< /Type /XObject /Name /R347 /Subtype /Image /Length 348 0 R +/ImageMask true /Width 82 /Height 72 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 82 /BlackIs1 true >>] +>> +stream +-Y/ktlnh'J_1WlVIN>d;OF5k,Do-F.&+/,KrBCW.%mf0XjHsl[XfmH0hq-EWKY$j0J0P"&G5OH\ +%NI1uW$_fE0RLEdqg:VHE;eVks1FCucp-FBHiJ$ama`KH4`C&`>7V$N*[EFX"s^Mb8oRjd(Qr1T!"]/~> +endstream +endobj +348 0 obj +158 +endobj +349 0 obj +<< /Length 350 0 R >> +stream +q +8.2 0 0 -7.2 317.9 281 cm +/R347 Do +Q +q +0.6 0 0 -10 327.2 281.6 cm +/R343 Do +Q +q +6.6 0 0 -2.7 332.3 278 cm +/R331 Do +Q +BT +1 0 0 1 342.24 274.08 Tm +(3) Tj +ET +q +5 0 0 -9.2 347.7 281.2 cm +/R245 Do +Q +endstream +endobj +350 0 obj +195 +endobj +351 0 obj +<< /Type /XObject /Name /R351 /Subtype /Image /Length 352 0 R +/ImageMask true /Width 66 /Height 65 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 66 /BlackIs1 true >>] +>> +stream +,BZ6Ns8W-!s8W-!s-Eq9Iaj?!s8W-!s8W-!s8Uf'#QTA~> +endstream +endobj +352 0 obj +47 +endobj +353 0 obj +<< /Length 354 0 R >> +stream +q +6.6 0 0 -6.5 355 279.9 cm +/R351 Do +Q +BT +1 0 0 1 363.48 274.08 Tm +(1) Tj +1 0 0 1 370.44 274.08 Tm +(imply) Tj +1 0 0 1 395.76 274.08 Tm +(that) Tj +1 0 0 1 412.8 274.08 Tm +(at) Tj +1 0 0 1 422.04 274.08 Tm +(least) Tj +ET +q +5 0 0 -9.2 442.7 281.2 cm +/R245 Do +Q +q +6.6 0 0 -6.5 450 279.9 cm +/R351 Do +Q +BT +1 0 0 1 458.519 274.08 Tm +(1) Tj +1 0 0 1 465.599 274.08 Tm +(non-f) Tj +1 0 0 1 487.319 274.08 Tm +(aulty) Tj +1 0 0 1 509.039 274.08 Tm +(replicas) Tj +1 0 0 1 315 262.08 Tm +(ha) Tj +1 0 0 1 324.24 262.08 Tm +(v) Tj +1 0 0 1 329.16 262.08 Tm +(e) Tj +1 0 0 1 336.36 262.08 Tm +(sent) Tj +1 0 0 1 355.32 262.08 Tm +(a) Tj +1 0 0 1 362.64 262.08 Tm +(pre-prepare) Tj +1 0 0 1 411.48 262.08 Tm +(or) Tj +1 0 0 1 422.64 262.08 Tm +(prepare) Tj +1 0 0 1 455.399 262.08 Tm +(for) Tj +ET +q +8.1 0 0 -4.6 470.1 266.6 cm +/R259 Do +Q +BT +1 0 0 1 481.559 262.08 Tm +(in) Tj +1 0 0 1 492.239 262.08 Tm +(vie) Tj +1 0 0 1 504.239 262.08 Tm +(w) Tj +ET +q +4.4 0 0 -4.6 514.4 266.6 cm +/R275 Do +Q +BT +1 0 0 1 522.239 262.08 Tm +(with) Tj +1 0 0 1 315 250.2 Tm +(sequence) Tj +1 0 0 1 356.4 250.2 Tm +(number) Tj +ET +q +5.4 0 0 -4.6 392 254.7 cm +/R263 Do +Q +BT +1 0 0 1 397.8 250.2 Tm +(.) Tj +1 0 0 1 411.24 250.2 Tm +(Thus,) Tj +1 0 0 1 439.2 250.2 Tm +(for) Tj +/R252 10 Tf +1 0 0 1 455.76 250.2 Tm +(pr) Tj +1 0 0 1 464.28 250.2 Tm +(epar) Tj +1 0 0 1 482.28 250.2 Tm +(ed) Tj +ET +q +2.4 0 0 -9.9 492.7 257.7 cm +/R315 Do +Q +q +8.1 0 0 -4.6 495.8 254.7 cm +/R259 Do +Q +q +1.7 0 0 -3.7 504.7 257.7 cm +/R323 Do +Q +q +1.3 0 0 -3 507.8 251.3 cm +/R271 Do +Q +q +4.4 0 0 -4.6 511.6 254.7 cm +/R275 Do +Q +q +1.3 0 0 -3 517.4 251.3 cm +/R271 Do +Q +q +5.4 0 0 -4.6 521.2 254.7 cm +/R263 Do +Q +q +1.3 0 0 -3 527.8 251.3 cm +/R271 Do +Q +q +4.1 0 0 -8.9 531.3 257 cm +/R327 Do +Q +q +2.4 0 0 -9.9 536.6 257.7 cm +/R319 Do +Q +BT +/R238 10 Tf +1 0 0 1 315 238.2 Tm +(to) Tj +1 0 0 1 326.76 238.2 Tm +(be) Tj +1 0 0 1 340.08 238.2 Tm +(true) Tj +1 0 0 1 359.64 238.2 Tm +(at) Tj +1 0 0 1 370.8 238.2 Tm +(least) Tj +1 0 0 1 393 238.2 Tm +(one) Tj +1 0 0 1 411.36 238.2 Tm +(of) Tj +1 0 0 1 423.72 238.2 Tm +(these) Tj +1 0 0 1 448.2 238.2 Tm +(replicas) Tj +1 0 0 1 483.119 238.2 Tm +(needs) Tj +1 0 0 1 509.759 238.2 Tm +(to) Tj +1 0 0 1 521.519 238.2 Tm +(ha) Tj +1 0 0 1 530.759 238.2 Tm +(v) Tj +1 0 0 1 535.679 238.2 Tm +(e) Tj +1 0 0 1 315 226.32 Tm +(sent) Tj +1 0 0 1 333.96 226.32 Tm +(tw) Tj +1 0 0 1 343.8 226.32 Tm +(o) Tj +1 0 0 1 351.6 226.32 Tm +(con\257icting) Tj +1 0 0 1 397.2 226.32 Tm +(prepares) Tj +1 0 0 1 433.8 226.32 Tm +(\(or) Tj +1 0 0 1 448.319 226.32 Tm +(pre-prepares) Tj +1 0 0 1 500.879 226.32 Tm +(if) Tj +1 0 0 1 509.879 226.32 Tm +(it) Tj +1 0 0 1 518.279 226.32 Tm +(is) Tj +1 0 0 1 527.879 226.32 Tm +(the) Tj +1 0 0 1 315 214.32 Tm +(primary) Tj +1 0 0 1 350.64 214.32 Tm +(for) Tj +ET +q +4.4 0 0 -4.6 366.7 218.8 cm +/R275 Do +Q +BT +1 0 0 1 371.64 214.32 Tm +(\),) Tj +1 0 0 1 381.96 214.32 Tm +(i.e.,) Tj +1 0 0 1 401.16 214.32 Tm +(tw) Tj +1 0 0 1 411 214.32 Tm +(o) Tj +1 0 0 1 420.24 214.32 Tm +(prepares) Tj +1 0 0 1 458.04 214.32 Tm +(with) Tj +1 0 0 1 479.999 214.32 Tm +(the) Tj +1 0 0 1 496.319 214.32 Tm +(same) Tj +1 0 0 1 520.919 214.32 Tm +(vie) Tj +1 0 0 1 532.919 214.32 Tm +(w) Tj +1 0 0 1 315 202.32 Tm +(and) Tj +1 0 0 1 332.4 202.32 Tm +(sequence) Tj +1 0 0 1 371.88 202.32 Tm +(number) Tj +1 0 0 1 405.36 202.32 Tm +(and) Tj +1 0 0 1 422.76 202.32 Tm +(a) Tj +1 0 0 1 430.08 202.32 Tm +(dif) Tj +1 0 0 1 441 202.32 Tm +(ferent) Tj +1 0 0 1 467.159 202.32 Tm +(digest.) Tj +1 0 0 1 498.599 202.32 Tm +(But) Tj +1 0 0 1 515.999 202.32 Tm +(this) Tj +1 0 0 1 533.399 202.32 Tm +(is) Tj +1 0 0 1 315 190.44 Tm +(not) Tj +1 0 0 1 329.88 190.44 Tm +(possible) Tj +1 0 0 1 364.8 190.44 Tm +(because) Tj +1 0 0 1 398.52 190.44 Tm +(the) Tj +1 0 0 1 412.8 190.44 Tm +(replica) Tj +1 0 0 1 442.2 190.44 Tm +(is) Tj +1 0 0 1 450.959 190.44 Tm +(not) Tj +1 0 0 1 465.959 190.44 Tm +(f) Tj +1 0 0 1 469.199 190.44 Tm +(aulty) Tj +1 0 0 1 488.639 190.44 Tm +(.) Tj +1 0 0 1 494.399 190.44 Tm +(Finally) Tj +1 0 0 1 522.119 190.44 Tm +(,) Tj +1 0 0 1 526.679 190.44 Tm +(our) Tj +1 0 0 1 315 178.44 Tm +(assumption) Tj +1 0 0 1 362.28 178.44 Tm +(about) Tj +1 0 0 1 386.4 178.44 Tm +(the) Tj +1 0 0 1 400.56 178.44 Tm +(strength) Tj +1 0 0 1 434.52 178.44 Tm +(of) Tj +1 0 0 1 444.84 178.44 Tm +(message) Tj +1 0 0 1 480.479 178.44 Tm +(digests) Tj +1 0 0 1 510.119 178.44 Tm +(ensures) Tj +1 0 0 1 315 166.56 Tm +(that) Tj +1 0 0 1 332.52 166.56 Tm +(the) Tj +1 0 0 1 347.28 166.56 Tm +(probability) Tj +1 0 0 1 393.6 166.56 Tm +(that) Tj +ET +q +8.1 0 0 -4.6 411.3 171.1 cm +/R259 Do +Q +q +5.1 0 0 -9.4 424.1 173.8 cm +/R339 Do +Q +q +6.6 0 0 -2.7 423.3 170.5 cm +/R331 Do +Q +q +8.1 0 0 -4.6 433.5 171.1 cm +/R259 Do +Q +q +1.7 0 0 -3.7 442.4 174.1 cm +/R323 Do +Q +BT +1 0 0 1 447.48 166.56 Tm +(and) Tj +ET +q +7.5 0 0 -6.9 464.7 173.5 cm +/R335 Do +Q +q +2.4 0 0 -9.9 473.8 174.1 cm +/R315 Do +Q +q +8.1 0 0 -4.6 477 171.1 cm +/R259 Do +Q +q +2.4 0 0 -9.9 485.9 174.1 cm +/R319 Do +Q +q +6.6 0 0 -2.7 492.9 170.5 cm +/R331 Do +Q +q +7.5 0 0 -6.9 503.2 173.5 cm +/R335 Do +Q +q +2.4 0 0 -9.9 512.3 174.1 cm +/R315 Do +Q +q +8.1 0 0 -4.6 515.5 171.1 cm +/R259 Do +Q +q +1.7 0 0 -3.7 524.3 174.1 cm +/R323 Do +Q +q +2.4 0 0 -9.9 527.3 174.1 cm +/R319 Do +Q +BT +1 0 0 1 533.4 166.56 Tm +(is) Tj +1 0 0 1 315 154.56 Tm +(ne) Tj +1 0 0 1 324.36 154.56 Tm +(gligible.) Tj +1 0 0 1 324.96 142.68 Tm +(Replica) Tj +ET +q +2.6 0 0 -6.9 357.3 149.5 cm +/R307 Do +Q +BT +1 0 0 1 362.28 142.68 Tm +(multicasts) Tj +1 0 0 1 404.4 142.68 Tm +(a) Tj +ET +q +2.4 0 0 -9.9 411.4 150.2 cm +/R267 Do +Q +BT +/R238 8 Tf +1 0 0 1 414.36 142.68 Tm +(COMMIT) Tj +ET +q +1.3 0 0 -3 447.9 143.8 cm +/R271 Do +Q +q +4.4 0 0 -4.6 451.8 147.2 cm +/R275 Do +Q +q +1.3 0 0 -3 457.5 143.8 cm +/R271 Do +Q +q +5.4 0 0 -4.6 461.4 147.2 cm +/R263 Do +Q +q +1.3 0 0 -3 468 143.8 cm +/R271 Do +Q +q +7.5 0 0 -6.9 471.9 149.6 cm +/R335 Do +Q +q +2.4 0 0 -9.9 481 150.2 cm +/R315 Do +Q +q +8.1 0 0 -4.6 484.2 147.2 cm +/R259 Do +Q +q +2.4 0 0 -9.9 493.1 150.2 cm +/R319 Do +Q +q +1.3 0 0 -3 497.4 143.8 cm +/R271 Do +Q +q +2.6 0 0 -6.9 501.2 149.5 cm +/R307 Do +Q +q +2.4 0 0 -9.9 504.9 150.2 cm +/R283 Do +Q +q +4.2 0 0 -3.1 508.5 144.2 cm +/R287 Do +Q +q +1.7 0 0 -3.5 513.2 143.8 cm +/R311 Do +Q +BT +/R238 10 Tf +1 0 0 1 518.4 142.68 Tm +(to) Tj +1 0 0 1 527.88 142.68 Tm +(the) Tj +1 0 0 1 315 130.68 Tm +(other) Tj +1 0 0 1 339.24 130.68 Tm +(replicas) Tj +1 0 0 1 374.04 130.68 Tm +(when) Tj +/R252 10 Tf +1 0 0 1 399.36 130.68 Tm +(pr) Tj +1 0 0 1 407.88 130.68 Tm +(epar) Tj +1 0 0 1 425.88 130.68 Tm +(ed) Tj +ET +q +2.4 0 0 -9.9 436.3 138.2 cm +/R315 Do +Q +q +8.1 0 0 -4.6 439.4 135.2 cm +/R259 Do +Q +q +1.3 0 0 -3 448.8 131.8 cm +/R271 Do +Q +q +4.4 0 0 -4.6 452.6 135.2 cm +/R275 Do +Q +q +1.3 0 0 -3 458.4 131.8 cm +/R271 Do +Q +q +5.4 0 0 -4.6 462.2 135.2 cm +/R263 Do +Q +q +1.3 0 0 -3 468.8 131.8 cm +/R271 Do +Q +q +2.6 0 0 -6.9 472.6 137.5 cm +/R307 Do +Q +q +2.4 0 0 -9.9 476.3 138.2 cm +/R319 Do +Q +BT +/R238 10 Tf +1 0 0 1 483.359 130.68 Tm +(becomes) Tj +1 0 0 1 521.999 130.68 Tm +(true.) Tj +1 0 0 1 315 118.8 Tm +(This) Tj +1 0 0 1 336.24 118.8 Tm +(starts) Tj +1 0 0 1 360.84 118.8 Tm +(the) Tj +1 0 0 1 376.68 118.8 Tm +(commit) Tj +1 0 0 1 410.64 118.8 Tm +(phase.) Tj +1 0 0 1 442.68 118.8 Tm +(Replicas) Tj +1 0 0 1 480.479 118.8 Tm +(accept) Tj +1 0 0 1 509.519 118.8 Tm +(commit) Tj +1 0 0 1 315 106.8 Tm +(messages) Tj +1 0 0 1 355.68 106.8 Tm +(and) Tj +1 0 0 1 373.2 106.8 Tm +(insert) Tj +1 0 0 1 398.4 106.8 Tm +(them) Tj +1 0 0 1 421.44 106.8 Tm +(in) Tj +1 0 0 1 432.24 106.8 Tm +(their) Tj +1 0 0 1 453.599 106.8 Tm +(log) Tj +1 0 0 1 469.439 106.8 Tm +(pro) Tj +1 0 0 1 482.759 106.8 Tm +(vided) Tj +1 0 0 1 507.719 106.8 Tm +(the) Tj +1 0 0 1 519.839 106.8 Tm +(y) Tj +1 0 0 1 527.879 106.8 Tm +(are) Tj +1 0 0 1 315 94.92 Tm +(properly) Tj +1 0 0 1 350.88 94.92 Tm +(signed,) Tj +1 0 0 1 381.6 94.92 Tm +(the) Tj +1 0 0 1 395.88 94.92 Tm +(vie) Tj +1 0 0 1 407.88 94.92 Tm +(w) Tj +1 0 0 1 417.12 94.92 Tm +(number) Tj +1 0 0 1 449.64 94.92 Tm +(in) Tj +1 0 0 1 459.479 94.92 Tm +(the) Tj +1 0 0 1 473.759 94.92 Tm +(message) Tj +1 0 0 1 509.639 94.92 Tm +(is) Tj +1 0 0 1 518.399 94.92 Tm +(equal) Tj +1 0 0 1 315 82.92 Tm +(to) Tj +1 0 0 1 325.2 82.92 Tm +(the) Tj +1 0 0 1 339.84 82.92 Tm +(replica') Tj +1 0 0 1 369.84 82.92 Tm +(s) Tj +1 0 0 1 375.96 82.92 Tm +(current) Tj +1 0 0 1 406.68 82.92 Tm +(vie) Tj +1 0 0 1 418.68 82.92 Tm +(w) Tj +1 0 0 1 425.28 82.92 Tm +(,) Tj +1 0 0 1 430.08 82.92 Tm +(and) Tj +1 0 0 1 446.879 82.92 Tm +(the) Tj +1 0 0 1 461.519 82.92 Tm +(sequence) Tj +1 0 0 1 500.519 82.92 Tm +(number) Tj +1 0 0 1 533.399 82.92 Tm +(is) Tj +1 0 0 1 315 71.04 Tm +(between) Tj +ET +q +5 0 0 -7.1 351 78 cm +/R299 Do +Q +BT +1 0 0 1 358.92 71.04 Tm +(and) Tj +ET +q +8.4 0 0 -6.9 376.1 77.9 cm +/R303 Do +Q +BT +1 0 0 1 303.48 35.16 Tm +(4) Tj +ET +Q +endstream +endobj +354 0 obj +8348 +endobj +355 0 obj +<< +/Type /Page +/MediaBox [0 0 612 792] +/Parent 2 0 R +/Resources << /ProcSet [/PDF /ImageB /Text] +/Font << +/R252 252 0 R +/R249 249 0 R +/R238 238 0 R +>> +/XObject << +/R303 303 0 R +/R299 299 0 R +/R319 319 0 R +/R307 307 0 R +/R271 271 0 R +/R263 263 0 R +/R275 275 0 R +/R259 259 0 R +/R315 315 0 R +/R311 311 0 R +/R287 287 0 R +/R283 283 0 R +/R335 335 0 R +/R267 267 0 R +/R323 323 0 R +/R331 331 0 R +/R339 339 0 R +/R327 327 0 R +/R351 351 0 R +/R245 245 0 R +/R343 343 0 R +/R347 347 0 R +/R279 279 0 R +/R291 291 0 R +/R295 295 0 R +/R255 255 0 R +/R241 241 0 R +>> +>> +/Contents [ +239 0 R +243 0 R +247 0 R +250 0 R +253 0 R +257 0 R +261 0 R +265 0 R +269 0 R +273 0 R +277 0 R +281 0 R +285 0 R +289 0 R +293 0 R +297 0 R +301 0 R +305 0 R +309 0 R +313 0 R +317 0 R +321 0 R +325 0 R +329 0 R +333 0 R +337 0 R +341 0 R +345 0 R +349 0 R +353 0 R +] +>> +endobj +356 0 obj +<< /Type /Font /Name /R356 /Subtype /Type1 /BaseFont /Times-Roman >> +endobj +357 0 obj +<< /Length 358 0 R >> +stream +q +BT +/R356 10 Tf +1 0 0 1 81.96 709.08 Tm +(W) Tj +1 0 0 1 90.48 709.08 Tm +(e) Tj +1 0 0 1 98.5199 709.08 Tm +(de\256ne) Tj +1 0 0 1 126.36 709.08 Tm +(the) Tj +ET +endstream +endobj +358 0 obj +156 +endobj +359 0 obj +<< /Type /Font /Name /R359 /Subtype /Type1 /BaseFont /Times-Italic >> +endobj +360 0 obj +<< /Length 361 0 R >> +stream +BT +/R359 10 Tf +1 0 0 1 142.08 709.08 Tm +(committed) Tj +/R356 10 Tf +1 0 0 1 187.08 709.08 Tm +(and) Tj +/R359 10 Tf +1 0 0 1 204.96 709.08 Tm +(committed-local) Tj +/R356 10 Tf +1 0 0 1 273.239 709.08 Tm +(predi-) Tj +1 0 0 1 72 697.08 Tm +(cates) Tj +1 0 0 1 94.9199 697.08 Tm +(as) Tj +1 0 0 1 106.32 697.08 Tm +(follo) Tj +1 0 0 1 125.04 697.08 Tm +(ws:) Tj +/R359 10 Tf +1 0 0 1 143.4 697.08 Tm +(committed) Tj +ET +Q +q +W +0 0 612 792 re +n +endstream +endobj +361 0 obj +422 +endobj +362 0 obj +<< /Type /XObject /Name /R362 /Subtype /Image /Length 363 0 R +/ImageMask true /Width 24 /Height 99 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 24 /BlackIs1 true >>] +>> +stream ++rF]%*LO8IHL@*j8[taqp!<~> +endstream +endobj +363 0 obj +80 +endobj +364 0 obj +<< /Length 365 0 R >> +stream +q +2.4 0 0 -9.9 185.9 704.6 cm +/R362 Do +Q +endstream +endobj +365 0 obj +41 +endobj +366 0 obj +<< /Type /XObject /Name /R366 /Subtype /Image /Length 367 0 R +/ImageMask true /Width 81 /Height 46 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 81 /BlackIs1 true >>] +>> +stream +4N`CV%M3$8612U>:Ag"NnFZb +endstream +endobj +367 0 obj +139 +endobj +368 0 obj +<< /Length 369 0 R >> +stream +q +8.1 0 0 -4.6 189.1 701.6 cm +/R366 Do +Q +endstream +endobj +369 0 obj +41 +endobj +370 0 obj +<< /Type /XObject /Name /R370 /Subtype /Image /Length 371 0 R +/ImageMask true /Width 13 /Height 30 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 13 /BlackIs1 true >>] +>> +stream +0g^A/h=&*equ,LR2Xj(@lh_"t-iec'~> +endstream +endobj +371 0 obj +33 +endobj +372 0 obj +<< /Length 373 0 R >> +stream +q +1.3 0 0 -3 198.4 698.2 cm +/R370 Do +Q +endstream +endobj +373 0 obj +39 +endobj +374 0 obj +<< /Type /XObject /Name /R374 /Subtype /Image /Length 375 0 R +/ImageMask true /Width 44 /Height 46 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 44 /BlackIs1 true >>] +>> +stream +2$jBq +&9R,hkFhTE_G0fRqg[T's0obU5O\AciSX+=pUN/*r',Z@_Z&BgfC8fKf@@X?]DE=DTC?6WJ-Z.*~> +endstream +endobj +375 0 obj +84 +endobj +376 0 obj +<< /Length 377 0 R >> +stream +q +4.4 0 0 -4.6 202.3 701.6 cm +/R374 Do +Q +q +1.3 0 0 -3 208 698.2 cm +/R370 Do +Q +endstream +endobj +377 0 obj +78 +endobj +378 0 obj +<< /Type /XObject /Name /R378 /Subtype /Image /Length 379 0 R +/ImageMask true /Width 54 /Height 46 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 54 /BlackIs1 true >>] +>> +stream +4N]t,$m>7rl/"o@d!re, +pYTohhg]lis4_o!aocV_^Yur>rVsXfR^`8 +endstream +endobj +379 0 obj +102 +endobj +380 0 obj +<< /Length 381 0 R >> +stream +q +5.4 0 0 -4.6 211.9 701.6 cm +/R378 Do +Q +endstream +endobj +381 0 obj +41 +endobj +382 0 obj +<< /Type /XObject /Name /R382 /Subtype /Image /Length 383 0 R +/ImageMask true /Width 24 /Height 99 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 24 /BlackIs1 true >>] +>> +stream +,JEO$loTRL4niC`hqS$`rnmKps6nkFp\Y!js6KW5s0(&RK)W@\(KK(`l\2@U?Bm&k<5,",!$D~> +endstream +endobj +383 0 obj +76 +endobj +384 0 obj +<< /Length 385 0 R >> +stream +q +2.4 0 0 -9.9 218.1 704.6 cm +/R382 Do +Q +BT +/R356 10 Tf +1 0 0 1 224.519 697.08 Tm +(is) Tj +1 0 0 1 234.239 697.08 Tm +(true) Tj +1 0 0 1 252.719 697.08 Tm +(if) Tj +1 0 0 1 261.839 697.08 Tm +(and) Tj +1 0 0 1 279.239 697.08 Tm +(only) Tj +1 0 0 1 72 685.2 Tm +(if) Tj +/R359 10 Tf +1 0 0 1 82.44 685.2 Tm +(pr) Tj +1 0 0 1 90.9599 685.2 Tm +(epar) Tj +1 0 0 1 108.96 685.2 Tm +(ed) Tj +ET +q +2.4 0 0 -9.9 119.3 692.7 cm +/R362 Do +Q +q +8.1 0 0 -4.6 122.5 689.7 cm +/R366 Do +Q +q +1.3 0 0 -3 131.8 686.3 cm +/R370 Do +Q +q +4.4 0 0 -4.6 135.7 689.7 cm +/R374 Do +Q +q +1.3 0 0 -3 141.4 686.3 cm +/R370 Do +Q +q +5.4 0 0 -4.6 145.3 689.7 cm +/R378 Do +Q +q +1.3 0 0 -3 151.9 686.3 cm +/R370 Do +Q +endstream +endobj +385 0 obj +653 +endobj +386 0 obj +<< /Type /XObject /Name /R386 /Subtype /Image /Length 387 0 R +/ImageMask true /Width 26 /Height 69 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 26 /BlackIs1 true >>] +>> +stream +-R^*Zl/YrmIf:DfhtVWo^UNqc%K6'1s*addX2"DHrSQB/+4u(tp(mUa[G\kks2oQE"5iR%(OubkJ,~> +endstream +endobj +387 0 obj +80 +endobj +388 0 obj +<< /Length 389 0 R >> +stream +q +2.6 0 0 -6.9 155.7 692 cm +/R386 Do +Q +q +2.4 0 0 -9.9 159.4 692.7 cm +/R382 Do +Q +BT +/R356 10 Tf +1 0 0 1 166.92 685.2 Tm +(is) Tj +1 0 0 1 177.96 685.2 Tm +(true) Tj +1 0 0 1 197.76 685.2 Tm +(for) Tj +1 0 0 1 213.719 685.2 Tm +(all) Tj +ET +q +2.6 0 0 -6.9 228.2 692 cm +/R386 Do +Q +BT +1 0 0 1 235.799 685.2 Tm +(in) Tj +1 0 0 1 247.919 685.2 Tm +(some) Tj +1 0 0 1 273.239 685.2 Tm +(set) Tj +1 0 0 1 288.719 685.2 Tm +(of) Tj +ET +endstream +endobj +389 0 obj +411 +endobj +390 0 obj +<< /Type /XObject /Name /R390 /Subtype /Image /Length 391 0 R +/ImageMask true /Width 50 /Height 92 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 50 /BlackIs1 true >>] +>> +stream +5$eL7-b:K3p0<%? +h1u#R]2'Mug].5is7$!eqtL-hqu?EZs5/V3iQXZVs7/ihqu?EZs4QRbJe-ans*o2/4DdB_@6aqq5Q~> +endstream +endobj +391 0 obj +96 +endobj +392 0 obj +<< /Length 393 0 R >> +stream +q +5 0 0 -9.2 72.4 680.3 cm +/R390 Do +Q +endstream +endobj +393 0 obj +38 +endobj +394 0 obj +<< /Type /XObject /Name /R394 /Subtype /Image /Length 395 0 R +/ImageMask true /Width 66 /Height 65 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 66 /BlackIs1 true >>] +>> +stream +,BZ6Ns8W-!s8W-!s-Eq9Iaj?!s8W-!s8W-!s8Uf'#QTA~> +endstream +endobj +395 0 obj +47 +endobj +396 0 obj +<< /Length 397 0 R >> +stream +q +6.6 0 0 -6.5 78.9 679 cm +/R394 Do +Q +BT +1 0 0 1 86.5199 673.2 Tm +(1) Tj +1 0 0 1 93.2399 673.2 Tm +(non-f) Tj +1 0 0 1 114.96 673.2 Tm +(aulty) Tj +1 0 0 1 136.68 673.2 Tm +(replicas;) Tj +1 0 0 1 171.84 673.2 Tm +(and) Tj +/R359 10 Tf +1 0 0 1 188.04 673.2 Tm +(committed-local) Tj +ET +q +2.4 0 0 -9.9 254 680.7 cm +/R362 Do +Q +q +8.1 0 0 -4.6 257.1 677.7 cm +/R366 Do +Q +q +1.3 0 0 -3 266.5 674.3 cm +/R370 Do +Q +q +4.4 0 0 -4.6 270.3 677.7 cm +/R374 Do +Q +q +1.3 0 0 -3 276.1 674.3 cm +/R370 Do +Q +q +5.4 0 0 -4.6 279.9 677.7 cm +/R378 Do +Q +q +1.3 0 0 -3 286.5 674.3 cm +/R370 Do +Q +q +2.6 0 0 -6.9 290.4 680 cm +/R386 Do +Q +q +2.4 0 0 -9.9 294 680.7 cm +/R382 Do +Q +BT +/R356 10 Tf +1 0 0 1 72 661.32 Tm +(is) Tj +1 0 0 1 80.88 661.32 Tm +(true) Tj +1 0 0 1 98.7599 661.32 Tm +(if) Tj +1 0 0 1 107.04 661.32 Tm +(and) Tj +1 0 0 1 123.72 661.32 Tm +(only) Tj +1 0 0 1 143.76 661.32 Tm +(if) Tj +/R359 10 Tf +1 0 0 1 152.16 661.32 Tm +(pr) Tj +1 0 0 1 160.68 661.32 Tm +(epar) Tj +1 0 0 1 178.68 661.32 Tm +(ed) Tj +ET +q +2.4 0 0 -9.9 189.1 668.8 cm +/R362 Do +Q +q +8.1 0 0 -4.6 192.2 665.8 cm +/R366 Do +Q +q +1.3 0 0 -3 201.6 662.4 cm +/R370 Do +Q +q +4.4 0 0 -4.6 205.4 665.8 cm +/R374 Do +Q +q +1.3 0 0 -3 211.2 662.4 cm +/R370 Do +Q +q +5.4 0 0 -4.6 215 665.8 cm +/R378 Do +Q +q +1.3 0 0 -3 221.6 662.4 cm +/R370 Do +Q +q +2.6 0 0 -6.9 225.4 668.1 cm +/R386 Do +Q +q +2.4 0 0 -9.9 229.1 668.8 cm +/R382 Do +Q +BT +/R356 10 Tf +1 0 0 1 234.599 661.32 Tm +(is) Tj +1 0 0 1 243.479 661.32 Tm +(true) Tj +1 0 0 1 261.359 661.32 Tm +(and) Tj +ET +q +2.6 0 0 -6.9 278.2 668.1 cm +/R386 Do +Q +BT +1 0 0 1 283.679 661.32 Tm +(has) Tj +1 0 0 1 72 649.32 Tm +(accepted) Tj +1 0 0 1 110.64 649.32 Tm +(2) Tj +ET +q +5 0 0 -9.2 116.1 656.4 cm +/R390 Do +Q +q +6.6 0 0 -6.5 124.9 655.1 cm +/R394 Do +Q +BT +1 0 0 1 135.12 649.32 Tm +(1) Tj +1 0 0 1 143.88 649.32 Tm +(commits) Tj +1 0 0 1 181.92 649.32 Tm +(\(possibly) Tj +1 0 0 1 222.24 649.32 Tm +(including) Tj +1 0 0 1 263.639 649.32 Tm +(its) Tj +1 0 0 1 276.719 649.32 Tm +(o) Tj +1 0 0 1 281.519 649.32 Tm +(wn\)) Tj +1 0 0 1 72 637.44 Tm +(from) Tj +1 0 0 1 93.8399 637.44 Tm +(dif) Tj +1 0 0 1 104.76 637.44 Tm +(ferent) Tj +1 0 0 1 130.44 637.44 Tm +(replicas) Tj +1 0 0 1 163.92 637.44 Tm +(that) Tj +1 0 0 1 181.44 637.44 Tm +(match) Tj +1 0 0 1 208.2 637.44 Tm +(the) Tj +1 0 0 1 222.959 637.44 Tm +(pre-prepare) Tj +1 0 0 1 271.319 637.44 Tm +(for) Tj +ET +q +8.1 0 0 -4.6 285.7 641.9 cm +/R366 Do +Q +BT +1 0 0 1 294.239 637.44 Tm +(;) Tj +1 0 0 1 72 625.44 Tm +(a) Tj +1 0 0 1 80.0399 625.44 Tm +(commit) Tj +1 0 0 1 114.24 625.44 Tm +(matches) Tj +1 0 0 1 150.48 625.44 Tm +(a) Tj +1 0 0 1 158.64 625.44 Tm +(pre-prepare) Tj +1 0 0 1 208.2 625.44 Tm +(if) Tj +1 0 0 1 217.92 625.44 Tm +(the) Tj +1 0 0 1 230.04 625.44 Tm +(y) Tj +1 0 0 1 238.559 625.44 Tm +(ha) Tj +1 0 0 1 247.799 625.44 Tm +(v) Tj +1 0 0 1 252.719 625.44 Tm +(e) Tj +1 0 0 1 260.639 625.44 Tm +(the) Tj +1 0 0 1 276.479 625.44 Tm +(same) Tj +1 0 0 1 72 613.44 Tm +(vie) Tj +1 0 0 1 84 613.44 Tm +(w) Tj +1 0 0 1 90.6 613.44 Tm +(,) Tj +1 0 0 1 95.4 613.44 Tm +(sequence) Tj +1 0 0 1 134.4 613.44 Tm +(number) Tj +1 0 0 1 164.76 613.44 Tm +(,) Tj +1 0 0 1 169.44 613.44 Tm +(and) Tj +1 0 0 1 186.36 613.44 Tm +(digest.) Tj +1 0 0 1 81.96 601.56 Tm +(The) Tj +1 0 0 1 100.44 601.56 Tm +(commit) Tj +1 0 0 1 133.8 601.56 Tm +(phase) Tj +1 0 0 1 159.48 601.56 Tm +(ensures) Tj +1 0 0 1 192.24 601.56 Tm +(the) Tj +1 0 0 1 207.36 601.56 Tm +(follo) Tj +1 0 0 1 226.08 601.56 Tm +(wing) Tj +1 0 0 1 248.879 601.56 Tm +(in) Tj +1 0 0 1 256.319 601.56 Tm +(v) Tj +1 0 0 1 261.119 601.56 Tm +(ariant:) Tj +1 0 0 1 290.879 601.56 Tm +(if) Tj +/R359 10 Tf +1 0 0 1 72 589.56 Tm +(committed-local) Tj +ET +q +2.4 0 0 -9.9 137.9 597.1 cm +/R362 Do +Q +q +8.1 0 0 -4.6 141.1 594.1 cm +/R366 Do +Q +q +1.3 0 0 -3 150.4 590.7 cm +/R370 Do +Q +q +4.4 0 0 -4.6 154.3 594.1 cm +/R374 Do +Q +q +1.3 0 0 -3 160 590.7 cm +/R370 Do +Q +q +5.4 0 0 -4.6 163.9 594.1 cm +/R378 Do +Q +q +1.3 0 0 -3 170.5 590.7 cm +/R370 Do +Q +q +2.6 0 0 -6.9 174.3 596.4 cm +/R386 Do +Q +q +2.4 0 0 -9.9 178 597.1 cm +/R382 Do +Q +BT +/R356 10 Tf +1 0 0 1 185.04 589.56 Tm +(is) Tj +1 0 0 1 195.6 589.56 Tm +(true) Tj +1 0 0 1 215.039 589.56 Tm +(for) Tj +1 0 0 1 230.639 589.56 Tm +(some) Tj +1 0 0 1 255.599 589.56 Tm +(non-f) Tj +1 0 0 1 277.319 589.56 Tm +(aulty) Tj +ET +q +2.6 0 0 -6.9 72.2 584.5 cm +/R386 Do +Q +BT +1 0 0 1 79.68 577.68 Tm +(then) Tj +/R359 10 Tf +1 0 0 1 101.16 577.68 Tm +(committed) Tj +ET +q +2.4 0 0 -9.9 143.7 585.2 cm +/R362 Do +Q +q +8.1 0 0 -4.6 146.8 582.2 cm +/R366 Do +Q +q +1.3 0 0 -3 156.2 578.8 cm +/R370 Do +Q +q +4.4 0 0 -4.6 160 582.2 cm +/R374 Do +Q +q +1.3 0 0 -3 165.8 578.8 cm +/R370 Do +Q +q +5.4 0 0 -4.6 169.6 582.2 cm +/R378 Do +Q +q +2.4 0 0 -9.9 175.8 585.2 cm +/R382 Do +Q +BT +/R356 10 Tf +1 0 0 1 183.48 577.68 Tm +(is) Tj +1 0 0 1 194.52 577.68 Tm +(true.) Tj +1 0 0 1 221.519 577.68 Tm +(This) Tj +1 0 0 1 243.479 577.68 Tm +(in) Tj +1 0 0 1 250.919 577.68 Tm +(v) Tj +1 0 0 1 255.719 577.68 Tm +(ariant) Tj +1 0 0 1 282.599 577.68 Tm +(and) Tj +1 0 0 1 72 565.68 Tm +(the) Tj +1 0 0 1 86.2799 565.68 Tm +(vie) Tj +1 0 0 1 98.2799 565.68 Tm +(w-change) Tj +1 0 0 1 139.08 565.68 Tm +(protocol) Tj +1 0 0 1 174.36 565.68 Tm +(described) Tj +1 0 0 1 214.56 565.68 Tm +(in) Tj +1 0 0 1 224.399 565.68 Tm +(Section) Tj +1 0 0 1 256.439 565.68 Tm +(4.4) Tj +1 0 0 1 270.959 565.68 Tm +(ensure) Tj +1 0 0 1 72 553.8 Tm +(that) Tj +1 0 0 1 90.2399 553.8 Tm +(non-f) Tj +1 0 0 1 111.96 553.8 Tm +(aulty) Tj +1 0 0 1 135 553.8 Tm +(replicas) Tj +1 0 0 1 169.2 553.8 Tm +(agree) Tj +1 0 0 1 194.16 553.8 Tm +(on) Tj +1 0 0 1 207.36 553.8 Tm +(the) Tj +1 0 0 1 222.84 553.8 Tm +(sequence) Tj +1 0 0 1 262.679 553.8 Tm +(numbers) Tj +1 0 0 1 72 541.8 Tm +(of) Tj +1 0 0 1 83.88 541.8 Tm +(requests) Tj +1 0 0 1 120 541.8 Tm +(that) Tj +1 0 0 1 138.48 541.8 Tm +(commit) Tj +1 0 0 1 172.44 541.8 Tm +(locally) Tj +1 0 0 1 203.16 541.8 Tm +(e) Tj +1 0 0 1 207.36 541.8 Tm +(v) Tj +1 0 0 1 212.28 541.8 Tm +(en) Tj +1 0 0 1 225.12 541.8 Tm +(if) Tj +1 0 0 1 234.72 541.8 Tm +(the) Tj +1 0 0 1 246.839 541.8 Tm +(y) Tj +1 0 0 1 255.239 541.8 Tm +(commit) Tj +1 0 0 1 289.199 541.8 Tm +(in) Tj +1 0 0 1 72 529.92 Tm +(dif) Tj +1 0 0 1 82.9199 529.92 Tm +(ferent) Tj +1 0 0 1 109.32 529.92 Tm +(vie) Tj +1 0 0 1 121.32 529.92 Tm +(ws) Tj +1 0 0 1 135.48 529.92 Tm +(at) Tj +1 0 0 1 145.92 529.92 Tm +(each) Tj +1 0 0 1 167.28 529.92 Tm +(replica.) Tj +1 0 0 1 202.68 529.92 Tm +(Furthermore,) Tj +1 0 0 1 258.359 529.92 Tm +(it) Tj +1 0 0 1 267.119 529.92 Tm +(ensures) Tj +1 0 0 1 72 517.92 Tm +(that) Tj +1 0 0 1 91.6799 517.92 Tm +(an) Tj +1 0 0 1 101.04 517.92 Tm +(y) Tj +1 0 0 1 110.64 517.92 Tm +(request) Tj +1 0 0 1 144.12 517.92 Tm +(that) Tj +1 0 0 1 163.68 517.92 Tm +(commits) Tj +1 0 0 1 202.8 517.92 Tm +(locally) Tj +1 0 0 1 234.599 517.92 Tm +(at) Tj +1 0 0 1 246.479 517.92 Tm +(a) Tj +1 0 0 1 255.599 517.92 Tm +(non-f) Tj +1 0 0 1 277.319 517.92 Tm +(aulty) Tj +1 0 0 1 72 506.04 Tm +(replica) Tj +1 0 0 1 101.64 506.04 Tm +(will) Tj +1 0 0 1 119.76 506.04 Tm +(commit) Tj +1 0 0 1 152.76 506.04 Tm +(at) Tj +ET +q +5 0 0 -9.2 162.9 513.1 cm +/R390 Do +Q +q +6.6 0 0 -6.5 171.1 511.8 cm +/R394 Do +Q +BT +1 0 0 1 180.6 506.04 Tm +(1) Tj +1 0 0 1 188.16 506.04 Tm +(or) Tj +1 0 0 1 199.08 506.04 Tm +(more) Tj +1 0 0 1 222 506.04 Tm +(non-f) Tj +1 0 0 1 243.719 506.04 Tm +(aulty) Tj +1 0 0 1 266.039 506.04 Tm +(replicas) Tj +1 0 0 1 72 494.04 Tm +(e) Tj +1 0 0 1 76.2 494.04 Tm +(v) Tj +1 0 0 1 81.1199 494.04 Tm +(entually) Tj +1 0 0 1 112.8 494.04 Tm +(.) Tj +1 0 0 1 81.96 482.16 Tm +(Each) Tj +1 0 0 1 106.56 482.16 Tm +(replica) Tj +ET +q +2.6 0 0 -6.9 138.6 489 cm +/R386 Do +Q +BT +1 0 0 1 146.4 482.16 Tm +(e) Tj +1 0 0 1 150.72 482.16 Tm +(x) Tj +1 0 0 1 155.64 482.16 Tm +(ecutes) Tj +1 0 0 1 185.04 482.16 Tm +(the) Tj +1 0 0 1 201.96 482.16 Tm +(operation) Tj +1 0 0 1 244.199 482.16 Tm +(requested) Tj +1 0 0 1 287.039 482.16 Tm +(by) Tj +ET +q +8.1 0 0 -4.6 72.2 474.7 cm +/R366 Do +Q +BT +1 0 0 1 84.12 470.16 Tm +(after) Tj +/R359 10 Tf +1 0 0 1 105.84 470.16 Tm +(committed-local) Tj +ET +q +2.4 0 0 -9.9 171.8 477.7 cm +/R362 Do +Q +q +8.1 0 0 -4.6 174.9 474.7 cm +/R366 Do +Q +q +1.3 0 0 -3 184.3 471.3 cm +/R370 Do +Q +q +4.4 0 0 -4.6 188.1 474.7 cm +/R374 Do +Q +q +1.3 0 0 -3 193.9 471.3 cm +/R370 Do +Q +q +5.4 0 0 -4.6 197.7 474.7 cm +/R378 Do +Q +q +1.3 0 0 -3 204.3 471.3 cm +/R370 Do +Q +q +2.6 0 0 -6.9 208.2 477 cm +/R386 Do +Q +q +2.4 0 0 -9.9 211.8 477.7 cm +/R382 Do +Q +BT +/R356 10 Tf +1 0 0 1 218.399 470.16 Tm +(is) Tj +1 0 0 1 228.479 470.16 Tm +(true) Tj +1 0 0 1 247.439 470.16 Tm +(and) Tj +ET +q +2.6 0 0 -6.9 265.4 477 cm +/R386 Do +Q +BT +1 0 0 1 268.679 470.16 Tm +(') Tj +1 0 0 1 271.439 470.16 Tm +(s) Tj +1 0 0 1 278.759 470.16 Tm +(state) Tj +1 0 0 1 72 458.28 Tm +(re\257ects) Tj +1 0 0 1 105.96 458.28 Tm +(the) Tj +1 0 0 1 123.36 458.28 Tm +(sequential) Tj +1 0 0 1 169.08 458.28 Tm +(e) Tj +1 0 0 1 173.4 458.28 Tm +(x) Tj +1 0 0 1 178.32 458.28 Tm +(ecution) Tj +1 0 0 1 212.64 458.28 Tm +(of) Tj +1 0 0 1 226.199 458.28 Tm +(all) Tj +1 0 0 1 241.439 458.28 Tm +(requests) Tj +1 0 0 1 279.239 458.28 Tm +(with) Tj +1 0 0 1 72 446.28 Tm +(lo) Tj +1 0 0 1 79.56 446.28 Tm +(wer) Tj +1 0 0 1 99.1199 446.28 Tm +(sequence) Tj +1 0 0 1 140.4 446.28 Tm +(numbers.) Tj +1 0 0 1 187.44 446.28 Tm +(This) Tj +1 0 0 1 209.88 446.28 Tm +(ensures) Tj +1 0 0 1 244.44 446.28 Tm +(that) Tj +1 0 0 1 264.119 446.28 Tm +(all) Tj +1 0 0 1 278.759 446.28 Tm +(non-) Tj +1 0 0 1 72 434.4 Tm +(f) Tj +1 0 0 1 75.24 434.4 Tm +(aulty) Tj +1 0 0 1 99.9599 434.4 Tm +(replicas) Tj +1 0 0 1 135.72 434.4 Tm +(e) Tj +1 0 0 1 140.04 434.4 Tm +(x) Tj +1 0 0 1 144.96 434.4 Tm +(ecute) Tj +1 0 0 1 170.64 434.4 Tm +(requests) Tj +1 0 0 1 208.08 434.4 Tm +(in) Tj +1 0 0 1 220.679 434.4 Tm +(the) Tj +1 0 0 1 237.599 434.4 Tm +(same) Tj +1 0 0 1 262.919 434.4 Tm +(order) Tj +1 0 0 1 288.719 434.4 Tm +(as) Tj +1 0 0 1 72 422.4 Tm +(required) Tj +1 0 0 1 108.12 422.4 Tm +(to) Tj +1 0 0 1 118.92 422.4 Tm +(pro) Tj +1 0 0 1 132.24 422.4 Tm +(vide) Tj +1 0 0 1 152.16 422.4 Tm +(the) Tj +1 0 0 1 167.28 422.4 Tm +(safety) Tj +1 0 0 1 194.04 422.4 Tm +(property) Tj +1 0 0 1 227.519 422.4 Tm +(.) Tj +1 0 0 1 234.599 422.4 Tm +(After) Tj +1 0 0 1 258.599 422.4 Tm +(e) Tj +1 0 0 1 262.919 422.4 Tm +(x) Tj +1 0 0 1 267.839 422.4 Tm +(ecuting) Tj +1 0 0 1 72 410.52 Tm +(the) Tj +1 0 0 1 85.9199 410.52 Tm +(requested) Tj +1 0 0 1 125.88 410.52 Tm +(operation,) Tj +1 0 0 1 167.88 410.52 Tm +(replicas) Tj +1 0 0 1 200.52 410.52 Tm +(send) Tj +1 0 0 1 220.559 410.52 Tm +(a) Tj +1 0 0 1 226.799 410.52 Tm +(reply) Tj +1 0 0 1 248.999 410.52 Tm +(to) Tj +1 0 0 1 258.479 410.52 Tm +(the) Tj +1 0 0 1 272.399 410.52 Tm +(client.) Tj +1 0 0 1 72 398.52 Tm +(Replicas) Tj +1 0 0 1 108.6 398.52 Tm +(discard) Tj +1 0 0 1 139.68 398.52 Tm +(requests) Tj +1 0 0 1 174.6 398.52 Tm +(whose) Tj +1 0 0 1 202.32 398.52 Tm +(timestamp) Tj +1 0 0 1 246.119 398.52 Tm +(is) Tj +1 0 0 1 255.119 398.52 Tm +(lo) Tj +1 0 0 1 262.679 398.52 Tm +(wer) Tj +1 0 0 1 279.839 398.52 Tm +(than) Tj +1 0 0 1 72 386.52 Tm +(the) Tj +1 0 0 1 87.2399 386.52 Tm +(timestamp) Tj +1 0 0 1 131.88 386.52 Tm +(in) Tj +1 0 0 1 142.8 386.52 Tm +(the) Tj +1 0 0 1 158.04 386.52 Tm +(last) Tj +1 0 0 1 174.96 386.52 Tm +(reply) Tj +1 0 0 1 198.6 386.52 Tm +(the) Tj +1 0 0 1 210.72 386.52 Tm +(y) Tj +1 0 0 1 218.639 386.52 Tm +(sent) Tj +1 0 0 1 237.839 386.52 Tm +(to) Tj +1 0 0 1 248.759 386.52 Tm +(the) Tj +1 0 0 1 263.999 386.52 Tm +(client) Tj +1 0 0 1 289.199 386.52 Tm +(to) Tj +1 0 0 1 72 374.64 Tm +(guarantee) Tj +1 0 0 1 113.16 374.64 Tm +(e) Tj +1 0 0 1 117.48 374.64 Tm +(xactly-once) Tj +1 0 0 1 166.44 374.64 Tm +(semantics.) Tj +1 0 0 1 81.96 362.64 Tm +(W) Tj +1 0 0 1 90.48 362.64 Tm +(e) Tj +1 0 0 1 100.08 362.64 Tm +(do) Tj +1 0 0 1 115.08 362.64 Tm +(not) Tj +1 0 0 1 132.84 362.64 Tm +(rely) Tj +1 0 0 1 153.48 362.64 Tm +(on) Tj +1 0 0 1 168.48 362.64 Tm +(ordered) Tj +1 0 0 1 204 362.64 Tm +(message) Tj +1 0 0 1 242.76 362.64 Tm +(deli) Tj +1 0 0 1 257.52 362.64 Tm +(v) Tj +1 0 0 1 262.439 362.64 Tm +(ery) Tj +1 0 0 1 274.679 362.64 Tm +(,) Tj +1 0 0 1 282.599 362.64 Tm +(and) Tj +1 0 0 1 72 350.76 Tm +(therefore) Tj +1 0 0 1 111.36 350.76 Tm +(it) Tj +1 0 0 1 120.24 350.76 Tm +(is) Tj +1 0 0 1 130.2 350.76 Tm +(possible) Tj +1 0 0 1 166.2 350.76 Tm +(for) Tj +1 0 0 1 181.2 350.76 Tm +(a) Tj +1 0 0 1 188.999 350.76 Tm +(replica) Tj +1 0 0 1 219.479 350.76 Tm +(to) Tj +1 0 0 1 230.519 350.76 Tm +(commit) Tj +1 0 0 1 264.359 350.76 Tm +(requests) Tj +1 0 0 1 72 338.76 Tm +(out) Tj +1 0 0 1 87.36 338.76 Tm +(of) Tj +1 0 0 1 98.2799 338.76 Tm +(order) Tj +1 0 0 1 118.92 338.76 Tm +(.) Tj +1 0 0 1 125.16 338.76 Tm +(This) Tj +1 0 0 1 145.44 338.76 Tm +(does) Tj +1 0 0 1 166.32 338.76 Tm +(not) Tj +1 0 0 1 181.68 338.76 Tm +(matter) Tj +1 0 0 1 209.76 338.76 Tm +(since) Tj +1 0 0 1 232.8 338.76 Tm +(it) Tj +1 0 0 1 240.959 338.76 Tm +(k) Tj +1 0 0 1 245.879 338.76 Tm +(eeps) Tj +1 0 0 1 266.159 338.76 Tm +(the) Tj +1 0 0 1 280.919 338.76 Tm +(pre-) Tj +1 0 0 1 72 326.88 Tm +(prepare,) Tj +1 0 0 1 107.28 326.88 Tm +(prepare,) Tj +1 0 0 1 142.68 326.88 Tm +(and) Tj +1 0 0 1 159.84 326.88 Tm +(commit) Tj +1 0 0 1 193.2 326.88 Tm +(messages) Tj +1 0 0 1 233.76 326.88 Tm +(logged) Tj +1 0 0 1 263.64 326.88 Tm +(until) Tj +1 0 0 1 284.879 326.88 Tm +(the) Tj +1 0 0 1 72 314.88 Tm +(corresponding) Tj +1 0 0 1 131.52 314.88 Tm +(request) Tj +1 0 0 1 162.72 314.88 Tm +(can) Tj +1 0 0 1 179.04 314.88 Tm +(be) Tj +1 0 0 1 190.92 314.88 Tm +(e) Tj +1 0 0 1 195.24 314.88 Tm +(x) Tj +1 0 0 1 200.16 314.88 Tm +(ecuted.) Tj +1 0 0 1 81.96 303 Tm +(Figure) Tj +1 0 0 1 111.36 303 Tm +(1) Tj +1 0 0 1 119.76 303 Tm +(sho) Tj +1 0 0 1 133.44 303 Tm +(ws) Tj +1 0 0 1 147.72 303 Tm +(the) Tj +1 0 0 1 163.32 303 Tm +(operation) Tj +1 0 0 1 204.36 303 Tm +(of) Tj +1 0 0 1 216 303 Tm +(the) Tj +1 0 0 1 231.6 303 Tm +(algorithm) Tj +1 0 0 1 273.719 303 Tm +(in) Tj +1 0 0 1 284.879 303 Tm +(the) Tj +1 0 0 1 72 291 Tm +(normal) Tj +1 0 0 1 102 291 Tm +(case) Tj +1 0 0 1 120.72 291 Tm +(of) Tj +1 0 0 1 130.68 291 Tm +(no) Tj +1 0 0 1 142.32 291 Tm +(primary) Tj +1 0 0 1 175.68 291 Tm +(f) Tj +1 0 0 1 178.92 291 Tm +(aults.) Tj +1 0 0 1 203.04 291 Tm +(Replica) Tj +1 0 0 1 235.079 291 Tm +(0) Tj +1 0 0 1 241.679 291 Tm +(is) Tj +1 0 0 1 249.839 291 Tm +(the) Tj +1 0 0 1 263.639 291 Tm +(primary) Tj +1 0 0 1 294.839 291 Tm +(,) Tj +1 0 0 1 72 279.12 Tm +(replica) Tj +1 0 0 1 101.64 279.12 Tm +(3) Tj +1 0 0 1 109.08 279.12 Tm +(is) Tj +1 0 0 1 118.2 279.12 Tm +(f) Tj +1 0 0 1 121.44 279.12 Tm +(aulty) Tj +1 0 0 1 140.88 279.12 Tm +(,) Tj +1 0 0 1 145.68 279.12 Tm +(and) Tj +ET +endstream +endobj +397 0 obj +13859 +endobj +398 0 obj +<< /Type /XObject /Name /R398 /Subtype /Image /Length 399 0 R +/ImageMask true /Width 71 /Height 74 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 71 /BlackIs1 true >>] +>> +stream +-,TJZ1dVEQ3$Jkuk_6*$#G?]kK5eSH^CS).)nksrhgWI<)tp]Y_[[^Ja2*B2Sf%[*I!cbX_1^mB-u]1e&-~> +endstream +endobj +399 0 obj +131 +endobj +400 0 obj +<< /Length 401 0 R >> +stream +q +7.1 0 0 -7.4 163 286.2 cm +/R398 Do +Q +BT +1 0 0 1 172.92 279.12 Tm +(is) Tj +1 0 0 1 182.04 279.12 Tm +(the) Tj +1 0 0 1 196.679 279.12 Tm +(client.) Tj +ET +1 i +83.3164 227.163 211.705 1.07319 re +f +83.315 211.063 211.705 1.07319 re +f +83.3161 193.363 211.702 1.07319 re +f +83.3161 176.763 211.702 1.07319 re +f +endstream +endobj +401 0 obj +302 +endobj +402 0 obj +<< /Type /Font /Name /R402 /Subtype /Type1 /BaseFont /Helvetica-Bold >> +endobj +403 0 obj +<< /Length 404 0 R >> +stream +BT +/R402 7.65182 Tf +1 0 0 0.981817 102.068 175.27 Tm +(X) Tj +ET +83.3258 244.863 211.56 1.07319 re +f +125.841 263.126 m +125.294 263.129 l +125.288 262.056 l +125.834 262.053 l +f +125.828 260.98 m +125.282 260.983 l +125.276 259.909 l +125.822 259.906 l +f +125.816 258.833 m +125.27 258.836 l +125.264 257.763 l +125.81 257.76 l +f +125.804 256.687 m +125.257 256.69 l +125.251 255.616 l +125.798 255.614 l +f +125.792 254.54 m +125.245 254.543 l +125.239 253.47 l +125.786 253.467 l +f +125.78 252.394 m +125.233 252.397 l +125.227 251.324 l +125.773 251.321 l +f +125.767 250.247 m +125.221 250.25 l +125.215 249.177 l +125.761 249.174 l +f +125.755 248.101 m +125.209 248.104 l +125.202 247.031 l +125.749 247.028 l +f +125.743 245.955 m +125.196 245.958 l +125.19 244.884 l +125.737 244.881 l +f +125.731 243.808 m +125.184 243.811 l +125.178 242.738 l +125.725 242.735 l +f +125.718 241.662 m +125.172 241.665 l +125.166 240.591 l +125.712 240.588 l +f +125.706 239.515 m +125.16 239.518 l +125.154 238.445 l +125.7 238.442 l +f +125.694 237.369 m +125.148 237.372 l +125.141 236.299 l +125.688 236.296 l +f +125.682 235.222 m +125.135 235.225 l +125.129 234.152 l +125.676 234.149 l +f +125.67 233.076 m +125.123 233.079 l +125.117 232.006 l +125.664 232.003 l +f +125.657 230.929 m +125.111 230.932 l +125.105 229.859 l +125.651 229.856 l +f +125.645 228.783 m +125.099 228.786 l +125.093 227.713 l +125.639 227.71 l +f +125.633 226.637 m +125.086 226.64 l +125.08 225.566 l +125.627 225.563 l +f +125.621 224.49 m +125.074 224.493 l +125.068 223.42 l +125.615 223.417 l +f +125.609 222.344 m +125.062 222.347 l +125.056 221.274 l +125.603 221.271 l +f +125.596 220.197 m +125.05 220.2 l +125.044 219.127 l +125.59 219.124 l +f +125.584 218.051 m +125.038 218.054 l +125.032 216.981 l +125.578 216.978 l +f +125.572 215.904 m +125.025 215.907 l +125.019 214.834 l +125.566 214.831 l +f +125.56 213.758 m +125.013 213.761 l +125.007 212.688 l +125.554 212.685 l +f +125.548 211.612 m +125.001 211.615 l +124.995 210.541 l +125.541 210.538 l +f +125.535 209.465 m +124.989 209.468 l +124.983 208.395 l +125.529 208.392 l +f +125.523 207.319 m +124.977 207.322 l +124.971 206.248 l +125.517 206.245 l +f +125.511 205.172 m +124.964 205.175 l +124.958 204.102 l +125.505 204.099 l +f +125.499 203.026 m +124.952 203.029 l +124.946 201.956 l +125.493 201.953 l +f +125.487 200.879 m +124.94 200.882 l +124.934 199.809 l +125.48 199.806 l +f +125.474 198.733 m +124.928 198.736 l +124.922 197.663 l +125.468 197.66 l +f +125.462 196.587 m +124.916 196.59 l +124.909 195.516 l +125.456 195.513 l +f +125.45 194.44 m +124.903 194.443 l +124.897 193.37 l +125.444 193.367 l +f +125.438 192.294 m +124.891 192.297 l +124.885 191.223 l +125.432 191.22 l +f +125.426 190.147 m +124.879 190.15 l +124.873 189.077 l +125.419 189.074 l +f +125.413 188.001 m +124.867 188.004 l +124.861 186.931 l +125.407 186.928 l +f +125.401 185.854 m +124.855 185.857 l +124.848 184.784 l +125.395 184.781 l +f +125.389 183.708 m +124.842 183.711 l +124.836 182.638 l +125.383 182.635 l +f +125.377 181.561 m +124.83 181.564 l +124.824 180.491 l +125.371 180.488 l +f +125.364 179.415 m +124.818 179.418 l +124.812 178.345 l +125.358 178.342 l +f +125.352 177.269 m +124.806 177.272 l +124.8 176.198 l +125.346 176.195 l +f +125.34 175.122 m +124.794 175.125 l +124.787 174.052 l +125.334 174.049 l +f +125.328 172.976 m +124.781 172.979 l +124.775 171.905 l +125.322 171.903 l +f +125.316 170.829 m +124.769 170.832 l +124.763 169.759 l +125.31 169.756 l +f +125.303 168.683 m +124.757 168.686 l +124.751 167.613 l +125.297 167.61 l +f +168.427 261.518 0.546533 1.07322 re +f +168.427 259.371 0.546533 1.07322 re +f +168.427 257.225 0.546533 1.07322 re +f +168.427 255.078 0.546533 1.07322 re +f +168.427 252.932 0.546533 1.07322 re +f +168.427 250.785 0.546533 1.07322 re +f +168.427 248.639 0.546533 1.07322 re +f +168.427 246.493 0.546533 1.07322 re +f +168.427 244.346 0.546533 1.07322 re +f +168.427 242.2 0.546533 1.07322 re +f +168.427 240.053 0.546533 1.07322 re +f +168.427 237.907 0.546533 1.07322 re +f +168.427 235.76 0.546533 1.07322 re +f +168.427 233.614 0.546533 1.07322 re +f +168.427 231.468 0.546533 1.07322 re +f +168.427 229.321 0.546533 1.07322 re +f +168.427 227.175 0.546533 1.07322 re +f +168.427 225.028 0.546533 1.07322 re +f +168.427 222.882 0.546533 1.07322 re +f +168.427 220.735 0.546533 1.07322 re +f +168.427 218.589 0.546533 1.07322 re +f +168.427 216.443 0.546533 1.07322 re +f +168.427 214.296 0.546533 1.07322 re +f +168.427 212.15 0.546533 1.07322 re +f +168.427 210.003 0.546533 1.07322 re +f +168.427 207.857 0.546533 1.07322 re +f +168.427 205.71 0.546533 1.07322 re +f +168.427 203.564 0.546533 1.07322 re +f +168.427 201.417 0.546533 1.07322 re +f +168.427 199.271 0.546533 1.07322 re +f +168.427 197.125 0.546533 1.07322 re +f +168.427 194.978 0.546533 1.07322 re +f +168.427 192.832 0.546533 1.07322 re +f +168.427 190.685 0.546533 1.07322 re +f +168.427 188.539 0.546533 1.07322 re +f +168.427 186.392 0.546533 1.07322 re +f +168.427 184.246 0.546533 1.07322 re +f +168.427 182.1 0.546533 1.07322 re +f +168.427 179.953 0.546533 1.07322 re +f +168.427 177.807 0.546533 1.07322 re +f +168.427 175.66 0.546533 1.07322 re +f +168.427 173.514 0.546533 1.07322 re +f +168.427 171.367 0.546533 1.07322 re +f +168.427 169.221 0.546533 1.07322 re +f +168.427 167.075 0.546533 1.07322 re +f +168.427 165.999 0.546533 0.00219727 re +f +208.924 262.592 m +208.377 262.589 l +208.383 261.516 l +208.93 261.519 l +f +208.936 260.446 m +208.389 260.443 l +208.395 259.37 l +208.942 259.373 l +f +208.948 258.299 m +208.401 258.297 l +208.407 257.223 l +208.954 257.226 l +f +208.96 256.153 m +208.413 256.15 l +208.419 255.077 l +208.966 255.08 l +f +208.972 254.007 m +208.425 254.004 l +208.431 252.93 l +208.978 252.933 l +f +208.984 251.86 m +208.437 251.857 l +208.443 250.784 l +208.99 250.787 l +f +208.996 249.714 m +208.449 249.711 l +208.455 248.638 l +209.002 248.641 l +f +209.008 247.567 m +208.461 247.564 l +208.467 246.491 l +209.014 246.494 l +f +209.02 245.421 m +208.474 245.418 l +208.48 244.345 l +209.026 244.348 l +f +209.032 243.274 m +208.486 243.272 l +208.492 242.198 l +209.038 242.201 l +f +209.044 241.128 m +208.498 241.125 l +208.504 240.052 l +209.05 240.055 l +f +209.056 238.982 m +208.51 238.979 l +208.516 237.905 l +209.062 237.908 l +f +209.068 236.835 m +208.522 236.832 l +208.528 235.759 l +209.074 235.762 l +f +209.08 234.689 m +208.534 234.686 l +208.54 233.613 l +209.086 233.615 l +f +209.092 232.542 m +208.546 232.539 l +208.552 231.466 l +209.098 231.469 l +f +209.104 230.396 m +208.558 230.393 l +208.564 229.32 l +209.11 229.323 l +f +209.117 228.249 m +208.57 228.246 l +208.576 227.173 l +209.123 227.176 l +f +209.129 226.103 m +208.582 226.1 l +208.588 225.027 l +209.135 225.03 l +f +209.141 223.957 m +208.594 223.954 l +208.6 222.88 l +209.147 222.883 l +f +209.153 221.81 m +208.606 221.807 l +208.612 220.734 l +209.159 220.737 l +f +209.165 219.664 m +208.618 219.661 l +208.624 218.588 l +209.171 218.59 l +f +209.177 217.517 m +208.63 217.514 l +208.636 216.441 l +209.183 216.444 l +f +209.189 215.371 m +208.642 215.368 l +208.648 214.295 l +209.195 214.298 l +f +209.201 213.224 m +208.654 213.221 l +208.66 212.148 l +209.207 212.151 l +f +209.213 211.078 m +208.666 211.075 l +208.673 210.002 l +209.219 210.005 l +f +209.225 208.931 m +208.679 208.929 l +208.685 207.855 l +209.231 207.858 l +f +209.237 206.785 m +208.691 206.782 l +208.697 205.709 l +209.243 205.712 l +f +209.249 204.639 m +208.703 204.636 l +208.709 203.562 l +209.255 203.565 l +f +209.261 202.492 m +208.715 202.489 l +208.721 201.416 l +209.267 201.419 l +f +209.273 200.346 m +208.727 200.343 l +208.733 199.27 l +209.279 199.273 l +f +209.285 198.199 m +208.739 198.196 l +208.745 197.123 l +209.291 197.126 l +f +209.297 196.053 m +208.751 196.05 l +208.757 194.977 l +209.303 194.98 l +f +209.309 193.906 m +208.763 193.903 l +208.769 192.83 l +209.316 192.833 l +f +209.322 191.76 m +208.775 191.757 l +208.781 190.684 l +209.328 190.687 l +f +209.334 189.614 m +208.787 189.611 l +208.793 188.537 l +209.34 188.54 l +f +209.346 187.467 m +208.799 187.464 l +208.805 186.391 l +209.352 186.394 l +f +209.358 185.321 m +208.811 185.318 l +208.817 184.245 l +209.364 184.247 l +f +209.37 183.174 m +208.823 183.171 l +208.829 182.098 l +209.376 182.101 l +f +209.382 181.028 m +208.835 181.025 l +208.841 179.952 l +209.388 179.955 l +f +209.394 178.881 m +208.847 178.878 l +208.853 177.805 l +209.4 177.808 l +f +209.406 176.735 m +208.859 176.732 l +208.865 175.659 l +209.412 175.662 l +f +209.418 174.589 m +208.872 174.586 l +208.878 173.512 l +209.424 173.515 l +f +209.43 172.442 m +208.884 172.439 l +208.89 171.366 l +209.436 171.369 l +f +209.442 170.296 m +208.896 170.293 l +208.902 169.219 l +209.448 169.222 l +f +209.454 168.149 m +208.908 168.146 l +208.914 167.073 l +209.46 167.076 l +f +209.449 166.095 m +208.937 165.908 l +208.938 165.906 l +209.45 166.092 l +f +249.369 261.519 m +248.822 261.516 l +248.829 260.443 l +249.375 260.446 l +f +249.381 259.373 m +248.835 259.37 l +248.841 258.297 l +249.387 258.299 l +f +249.393 257.226 m +248.847 257.223 l +248.853 256.15 l +249.399 256.153 l +f +249.406 255.08 m +248.859 255.077 l +248.865 254.004 l +249.412 254.007 l +f +249.418 252.933 m +248.871 252.93 l +248.877 251.857 l +249.424 251.86 l +f +249.43 250.787 m +248.883 250.784 l +248.89 249.711 l +249.436 249.714 l +f +249.442 248.641 m +248.896 248.638 l +248.902 247.564 l +249.448 247.567 l +f +249.454 246.494 m +248.908 246.491 l +248.914 245.418 l +249.461 245.421 l +f +249.467 244.348 m +248.92 244.345 l +248.926 243.271 l +249.473 243.274 l +f +249.479 242.201 m +248.932 242.198 l +248.938 241.125 l +249.485 241.128 l +f +249.491 240.055 m +248.945 240.052 l +248.951 238.979 l +249.497 238.982 l +f +249.503 237.908 m +248.957 237.905 l +248.963 236.832 l +249.509 236.835 l +f +249.515 235.762 m +248.969 235.759 l +248.975 234.686 l +249.522 234.689 l +f +249.528 233.615 m +248.981 233.613 l +248.987 232.539 l +249.534 232.542 l +f +249.54 231.469 m +248.993 231.466 l +248.999 230.393 l +249.546 230.396 l +f +249.552 229.323 m +249.006 229.32 l +249.012 228.246 l +249.558 228.249 l +f +249.564 227.176 m +249.018 227.173 l +249.024 226.1 l +249.57 226.103 l +f +249.576 225.03 m +249.03 225.027 l +249.036 223.954 l +249.583 223.957 l +f +249.589 222.883 m +249.042 222.88 l +249.048 221.807 l +249.595 221.81 l +f +249.601 220.737 m +249.054 220.734 l +249.06 219.661 l +249.607 219.664 l +f +249.613 218.59 m +249.067 218.587 l +249.073 217.514 l +249.619 217.517 l +f +249.625 216.444 m +249.079 216.441 l +249.085 215.368 l +249.631 215.371 l +f +249.638 214.298 m +249.091 214.295 l +249.097 213.221 l +249.644 213.224 l +f +249.65 212.151 m +249.103 212.148 l +249.109 211.075 l +249.656 211.078 l +f +249.662 210.005 m +249.115 210.002 l +249.122 208.928 l +249.668 208.931 l +f +249.674 207.858 m +249.128 207.855 l +249.134 206.782 l +249.68 206.785 l +f +249.686 205.712 m +249.14 205.709 l +249.146 204.636 l +249.692 204.639 l +f +249.699 203.565 m +249.152 203.562 l +249.158 202.489 l +249.705 202.492 l +f +249.711 201.419 m +249.164 201.416 l +249.17 200.343 l +249.717 200.346 l +f +249.723 199.273 m +249.176 199.27 l +249.183 198.196 l +249.729 198.199 l +f +249.735 197.126 m +249.189 197.123 l +249.195 196.05 l +249.741 196.053 l +f +249.747 194.98 m +249.201 194.977 l +249.207 193.903 l +249.753 193.906 l +f +249.76 192.833 m +249.213 192.83 l +249.219 191.757 l +249.766 191.76 l +f +249.772 190.687 m +249.225 190.684 l +249.231 189.611 l +249.778 189.614 l +f +249.784 188.54 m +249.237 188.537 l +249.244 187.464 l +249.79 187.467 l +f +249.796 186.394 m +249.25 186.391 l +249.256 185.318 l +249.802 185.321 l +f +249.808 184.247 m +249.262 184.244 l +249.268 183.171 l +249.815 183.174 l +f +249.821 182.101 m +249.274 182.098 l +249.28 181.025 l +249.827 181.028 l +f +249.833 179.955 m +249.286 179.952 l +249.292 178.878 l +249.839 178.881 l +f +249.845 177.808 m +249.299 177.805 l +249.305 176.732 l +249.851 176.735 l +f +249.857 175.662 m +249.311 175.659 l +249.317 174.586 l +249.863 174.589 l +f +249.869 173.515 m +249.323 173.512 l +249.329 172.439 l +249.876 172.442 l +f +249.882 171.369 m +249.335 171.366 l +249.341 170.293 l +249.888 170.296 l +f +249.894 169.222 m +249.347 169.219 l +249.353 168.146 l +249.9 168.149 l +f +249.906 167.076 m +249.36 167.073 l +249.366 166 l +249.912 166.003 l +f +88.5281 245.658 m +88.2806 245.18 l +122.714 228.008 l +122.961 228.486 l +f +123.842 227.746 m +119.349 231.576 l +118.042 229.049 l +f* +122.658 228.045 m +123.017 228.45 l +119.349 231.576 l +119.191 231.27 l +119.434 231.148 l +119.255 230.946 l +f +119.678 231.027 m +119.191 231.27 l +118.042 229.05 l +118.383 228.973 l +118.444 229.234 l +118.688 229.113 l +f +118.505 229.496 m +118.383 228.973 l +123.842 227.746 l +123.017 228.45 l +122.838 228.247 l +122.899 228.509 l +f +253.235 228.474 m +253.528 228.021 l +281.066 245.193 l +280.773 245.645 l +f +281.87 246.012 m +276.224 244.17 l +277.771 241.777 l +f* +281.005 245.164 m +280.833 245.674 l +276.224 244.17 l +276.412 243.88 l +276.642 244.023 l +276.728 243.769 l +f +276.873 244.167 m +276.412 243.88 l +277.771 241.778 l +278.012 242.027 l +277.814 242.211 l +278.045 242.355 l +f +277.616 242.396 m +278.012 242.027 l +281.87 246.012 l +280.833 245.674 l +280.919 245.419 l +280.721 245.604 l +f +258.351 211.776 m +258.784 211.448 l +285.312 245.255 l +284.879 245.583 l +f +285.783 246.295 m +281.117 242.67 l +283.403 240.941 l +f* +285.265 245.209 m +284.926 245.629 l +281.117 242.67 l +281.394 242.461 l +281.561 242.673 l +281.731 242.463 l +f +281.728 242.886 m +281.394 242.461 l +283.403 240.941 l +283.543 241.256 l +283.293 241.363 l +283.46 241.576 l +f +283.042 241.471 m +283.543 241.256 l +285.783 246.295 l +284.926 245.629 l +285.095 245.419 l +284.845 245.526 l +f +252.325 194.595 m +252.771 194.285 l +289.677 245.264 l +289.232 245.574 l +f +290.107 246.321 m +285.587 242.522 l +287.939 240.88 l +f* +289.632 245.215 m +289.277 245.623 l +285.587 242.522 l +285.872 242.323 l +286.031 242.542 l +286.208 242.338 l +f +286.189 242.76 m +285.872 242.323 l +287.939 240.88 l +288.067 241.2 l +287.812 241.298 l +287.971 241.517 l +f +287.558 241.396 m +288.067 241.2 l +290.107 246.321 l +289.277 245.623 l +289.455 245.419 l +289.2 245.517 l +f +endstream +endobj +404 0 obj +13606 +endobj +405 0 obj +<< /Type /Font /Name /R405 /Subtype /Type1 /BaseFont /Times-Bold >> +endobj +406 0 obj +<< /Length 407 0 R >> +stream +BT +/R405 7.65182 Tf +1 0 0 0.981817 92.7768 250.277 Tm +(request) Tj +1 0 0 0.981817 127.757 250.277 Tm +(pre-prepare) Tj +1 0 0 0.981817 175.854 250.277 Tm +(prepare) Tj +1 0 0 0.981817 215.206 250.277 Tm +(commit) Tj +1 0 0 0.981817 258.384 250.277 Tm +(reply) Tj +1 0 0 0.981817 74.1938 241.154 Tm +(C) Tj +1 0 0 0.981817 74.7404 224.519 Tm +(0) Tj +1 0 0 0.981817 74.7404 207.347 Tm +(1) Tj +1 0 0 0.981817 74.7404 190.175 Tm +(2) Tj +1 0 0 0.981817 74.7404 173.54 Tm +(3) Tj +ET +130.241 228.493 m +130.021 228.002 l +168.471 211.366 l +168.692 211.857 l +f +169.612 211.166 m +164.912 214.748 l +163.748 212.155 l +f* +168.414 211.4 m +168.749 211.824 l +164.912 214.748 l +164.771 214.433 l +165.021 214.325 l +164.853 214.113 l +f +165.271 214.217 m +164.771 214.433 l +163.748 212.155 l +164.093 212.096 l +164.139 212.361 l +164.389 212.253 l +f +164.186 212.625 m +164.093 212.096 l +169.612 211.166 l +168.749 211.824 l +168.581 211.612 l +168.628 211.876 l +f +131.79 228.433 m +131.395 228.062 l +164.352 194.254 l +164.747 194.626 l +f +165.329 193.641 m +162.38 198.715 l +160.295 196.755 l +f* +164.312 194.307 m +164.787 194.573 l +162.38 198.715 l +162.127 198.477 l +162.316 198.283 l +162.079 198.15 l +f +162.505 198.09 m +162.127 198.477 l +160.295 196.755 l +160.591 196.572 l +160.737 196.799 l +160.926 196.605 l +f +160.882 197.026 m +160.591 196.572 l +165.329 193.641 l +164.787 194.573 l +164.55 194.44 l +164.695 194.667 l +f +131.861 228.42 m +131.402 228.129 l +164.3 178.15 l +164.759 178.441 l +f +165.141 177.366 m +163.22 182.895 l +160.796 181.357 l +f* +164.271 178.209 m +164.788 178.382 l +163.22 182.895 l +162.926 182.709 l +163.074 182.483 l +162.816 182.397 l +f +163.223 182.258 m +162.926 182.709 l +160.796 181.357 l +161.052 181.123 l +161.238 181.318 l +161.386 181.093 l +f +161.425 181.514 m +161.052 181.123 l +165.141 177.366 l +164.788 178.382 l +164.53 178.295 l +164.716 178.491 l +f +173.006 211.766 m +173.294 211.31 l +199.185 227.069 l +198.897 227.525 l +f +199.998 227.88 m +194.332 226.098 l +195.853 223.69 l +f* +199.124 227.042 m +198.957 227.553 l +194.332 226.098 l +194.516 225.806 l +194.748 225.948 l +194.832 225.692 l +f +194.981 226.089 m +194.516 225.806 l +195.853 223.69 l +196.096 223.936 l +195.9 224.123 l +196.133 224.265 l +f +195.704 224.31 m +196.096 223.936 l +199.998 227.88 l +198.957 227.553 l +199.041 227.297 l +198.845 227.484 l +f +174.657 211.304 m +174.371 210.847 l +202.902 193.675 l +203.187 194.132 l +f +204.005 193.325 m +199.838 197.495 l +198.33 195.079 l +f* +202.85 193.715 m +203.24 194.091 l +199.838 197.495 l +199.655 197.202 l +199.888 197.062 l +199.693 196.874 l +f +200.121 196.922 m +199.655 197.202 l +198.33 195.079 l +198.664 194.976 l +198.746 195.232 l +198.979 195.092 l +f +198.828 195.488 m +198.664 194.976 l +204.005 193.325 l +203.24 194.091 l +203.045 193.903 l +203.127 194.159 l +f +173.774 211.835 m +173.376 211.467 l +205.35 178.091 l +205.748 178.458 l +f +206.321 177.469 m +203.417 182.568 l +201.315 180.626 l +f* +205.31 178.143 m +205.787 178.405 l +203.417 182.568 l +203.162 182.332 l +203.349 182.137 l +203.111 182.006 l +f +203.537 181.941 m +203.162 182.332 l +201.315 180.626 l +201.609 180.44 l +201.757 180.666 l +201.944 180.471 l +f +201.905 180.892 m +201.609 180.44 l +206.321 177.469 l +205.787 178.405 l +205.549 178.274 l +205.696 178.5 l +f +175.145 194.679 m +175.395 194.202 l +209.337 211.373 l +209.086 211.85 l +f +210.213 212.119 m +204.421 210.782 l +205.743 208.263 l +f* +209.274 211.351 m +209.149 211.873 l +204.421 210.782 l +204.581 210.477 l +204.824 210.599 l +204.887 210.338 l +f +205.067 210.722 m +204.581 210.477 l +205.743 208.263 l +206.006 208.49 l +205.825 208.691 l +206.068 208.814 l +f +205.645 208.893 m +206.006 208.49 l +210.213 212.119 l +209.149 211.873 l +209.212 211.612 l +209.031 211.813 l +f +174.451 194.603 m +174.852 194.238 l +206.334 227.562 l +205.933 227.927 l +f +206.9 228.556 m +201.916 225.364 l +204.032 223.437 l +f* +206.283 227.52 m +205.984 227.969 l +201.916 225.364 l +202.173 225.13 l +202.358 225.327 l +202.508 225.102 l +f +202.544 225.524 m +202.173 225.13 l +204.032 223.437 l +204.201 223.738 l +203.961 223.867 l +204.147 224.064 l +f +203.722 223.996 m +204.201 223.738 l +206.899 228.555 l +205.984 227.969 l +206.133 227.744 l +205.894 227.874 l +f +174.182 194.665 m +173.885 194.215 l +200.856 177.043 l +201.153 177.493 l +f +201.95 176.666 m +197.892 180.939 l +196.322 178.561 l +f* +200.805 177.085 m +201.205 177.451 l +197.892 180.939 l +197.702 180.65 l +197.931 180.504 l +197.731 180.321 l +f +198.16 180.358 m +197.702 180.65 l +196.322 178.561 l +196.653 178.45 l +196.741 178.703 l +196.971 178.557 l +f +196.83 178.957 m +196.653 178.45 l +201.95 176.666 l +201.205 177.451 l +201.005 177.268 l +201.093 177.522 l +f +211.899 228.481 m +211.63 228.014 l +241.618 211.378 l +241.887 211.845 l +f +242.733 211.068 m +238.417 215.089 l +236.997 212.622 l +f* +241.564 211.417 m +241.94 211.806 l +238.417 215.089 l +238.245 214.79 l +238.483 214.658 l +238.295 214.463 l +f +238.721 214.526 m +238.245 214.79 l +236.997 212.622 l +237.335 212.531 l +237.408 212.789 l +237.645 212.657 l +f +237.48 213.048 m +237.335 212.531 l +242.733 211.068 l +241.94 211.806 l +241.752 211.612 l +241.825 211.87 l +f +212.666 228.437 m +212.28 228.057 l +246.713 194.25 l +247.1 194.63 l +f +247.703 193.658 m +244.642 198.667 l +242.601 196.663 l +f* +246.672 194.302 m +247.141 194.578 l +244.642 198.667 l +244.394 198.424 l +244.588 198.234 l +244.353 198.096 l +f +244.781 198.045 m +244.394 198.424 l +242.601 196.663 l +242.901 196.487 l +243.042 196.717 l +243.235 196.527 l +f +243.182 196.947 m +242.901 196.487 l +247.703 193.658 l +247.141 194.578 l +246.906 194.44 l +247.047 194.67 l +f +212.7 228.425 m +212.247 228.124 l +246.68 178.145 l +247.133 178.446 l +f +247.538 177.379 m +245.499 182.867 l +243.108 181.28 l +f* +246.65 178.203 m +247.163 178.387 l +245.499 182.867 l +245.209 182.675 l +245.362 182.453 l +245.105 182.361 l +f +245.515 182.23 m +245.209 182.675 l +243.108 181.28 l +243.369 181.05 l +243.551 181.25 l +243.704 181.028 l +f +243.734 181.45 m +243.369 181.05 l +247.538 177.379 l +247.163 178.387 l +246.906 178.295 l +247.089 178.495 l +f +213.052 211.765 m +213.342 211.31 l +238.921 227.07 l +238.631 227.525 l +f +239.73 227.885 m +234.074 226.073 l +235.608 223.673 l +f* +238.861 227.042 m +238.691 227.552 l +234.074 226.073 l +234.26 225.782 l +234.491 225.925 l +234.576 225.67 l +f +234.723 226.067 m +234.26 225.782 l +235.608 223.673 l +235.85 223.921 l +235.653 224.107 l +235.885 224.249 l +f +235.456 224.292 m +235.85 223.921 l +239.73 227.885 l +238.691 227.552 l +238.776 227.297 l +238.579 227.483 l +f +213.458 211.838 m +213.165 211.385 l +240.664 194.214 l +240.957 194.666 l +f +241.761 193.846 m +237.665 198.083 l +236.116 195.692 l +f* +240.612 194.255 m +241.009 194.625 l +237.665 198.083 l +237.477 197.793 l +237.708 197.649 l +237.51 197.465 l +f +237.938 197.505 m +237.477 197.793 l +236.116 195.692 l +236.448 195.584 l +236.535 195.838 l +236.765 195.694 l +f +236.621 196.093 m +236.448 195.584 l +241.761 193.846 l +241.009 194.625 l +240.811 194.44 l +240.897 194.695 l +f +213.469 211.819 m +213.042 211.485 l +240.105 178.108 l +240.533 178.443 l +f +241.021 177.41 m +238.554 182.726 l +236.296 180.962 l +f* +240.07 178.164 m +240.568 178.387 l +238.554 182.726 l +238.281 182.512 l +238.451 182.303 l +238.202 182.191 l +f +238.621 182.093 m +238.281 182.512 l +236.296 180.962 l +236.574 180.753 l +236.74 180.966 l +236.91 180.756 l +f +236.906 181.179 m +236.574 180.753 l +241.02 177.41 l +240.568 178.387 l +240.319 178.276 l +240.485 178.489 l +f +210.777 194.681 m +211.016 194.199 l +246.925 211.371 l +246.686 211.853 l +f +247.819 212.096 m +241.997 210.888 l +243.261 208.341 l +f* +246.862 211.349 m +246.749 211.874 l +241.997 210.888 l +242.15 210.579 l +242.396 210.697 l +242.453 210.434 l +f +242.642 210.814 m +242.15 210.579 l +243.261 208.341 l +243.529 208.562 l +243.353 208.767 l +243.599 208.885 l +f +243.178 208.973 m +243.529 208.562 l +247.819 212.096 l +246.749 211.874 l +246.806 211.612 l +246.63 211.817 l +f +211.097 195.101 m +211.492 194.731 l +242.905 227.089 l +242.509 227.459 l +f +243.485 228.075 m +238.457 224.95 l +240.547 222.995 l +f* +242.853 227.047 m +242.561 227.501 l +238.457 224.95 l +238.711 224.713 l +238.899 224.907 l +239.045 224.68 l +f +239.088 225.101 m +238.711 224.713 l +240.547 222.995 l +240.719 223.293 l +240.482 223.426 l +240.67 223.62 l +f +240.244 223.558 m +240.719 223.293 l +243.485 228.075 l +242.561 227.501 l +242.707 227.274 l +242.47 227.406 l +f +210.014 194.693 m +209.698 194.255 l +233.801 177.524 l +234.117 177.962 l +f +234.879 177.104 m +230.999 181.533 l +229.333 179.219 l +f* +233.752 177.568 m +234.166 177.918 l +230.999 181.533 l +230.797 181.252 l +231.02 181.097 l +230.813 180.922 l +f +231.243 180.942 m +230.797 181.252 l +229.333 179.219 l +229.659 179.094 l +229.758 179.345 l +229.981 179.19 l +f +229.857 179.595 m +229.659 179.094 l +234.879 177.104 l +234.166 177.918 l +233.959 177.743 l +234.058 177.993 l +f +BT +/R356 10 Tf +1 0 0 1 117.24 141.84 Tm +(Figure) Tj +1 0 0 1 145.8 141.84 Tm +(1:) Tj +1 0 0 1 157.08 141.84 Tm +(Normal) Tj +1 0 0 1 190.08 141.84 Tm +(Case) Tj +1 0 0 1 211.92 141.84 Tm +(Operation) Tj +/R405 10 Tf +1 0 0 1 72 109.08 Tm +(4.3) Tj +1 0 0 1 94.4399 109.08 Tm +(Garbage) Tj +1 0 0 1 134.04 109.08 Tm +(Collection) Tj +/R356 10 Tf +1 0 0 1 72 94.92 Tm +(This) Tj +1 0 0 1 93.9599 94.92 Tm +(section) Tj +1 0 0 1 126.36 94.92 Tm +(discusses) Tj +1 0 0 1 167.76 94.92 Tm +(the) Tj +1 0 0 1 184.08 94.92 Tm +(mechanism) Tj +1 0 0 1 233.759 94.92 Tm +(used) Tj +1 0 0 1 256.199 94.92 Tm +(to) Tj +1 0 0 1 268.199 94.92 Tm +(discard) Tj +1 0 0 1 72 82.92 Tm +(messages) Tj +1 0 0 1 112.32 82.92 Tm +(from) Tj +1 0 0 1 134.28 82.92 Tm +(the) Tj +1 0 0 1 149.16 82.92 Tm +(log.) Tj +1 0 0 1 168.48 82.92 Tm +(F) Tj +1 0 0 1 173.88 82.92 Tm +(or) Tj +1 0 0 1 184.8 82.92 Tm +(the) Tj +1 0 0 1 199.68 82.92 Tm +(safety) Tj +1 0 0 1 226.08 82.92 Tm +(condition) Tj +1 0 0 1 266.399 82.92 Tm +(to) Tj +1 0 0 1 276.839 82.92 Tm +(hold,) Tj +1 0 0 1 72 71.04 Tm +(messages) Tj +1 0 0 1 111.24 71.04 Tm +(must) Tj +1 0 0 1 132.36 71.04 Tm +(be) Tj +1 0 0 1 143.4 71.04 Tm +(k) Tj +1 0 0 1 148.32 71.04 Tm +(ept) Tj +1 0 0 1 162.12 71.04 Tm +(in) Tj +1 0 0 1 171.6 71.04 Tm +(a) Tj +1 0 0 1 177.6 71.04 Tm +(replica') Tj +1 0 0 1 207.6 71.04 Tm +(s) Tj +1 0 0 1 213.12 71.04 Tm +(log) Tj +1 0 0 1 227.519 71.04 Tm +(until) Tj +1 0 0 1 247.439 71.04 Tm +(it) Tj +1 0 0 1 254.639 71.04 Tm +(kno) Tj +1 0 0 1 269.519 71.04 Tm +(ws) Tj +1 0 0 1 281.999 71.04 Tm +(that) Tj +1 0 0 1 315 709.08 Tm +(the) Tj +1 0 0 1 329.88 709.08 Tm +(requests) Tj +1 0 0 1 365.16 709.08 Tm +(the) Tj +1 0 0 1 377.28 709.08 Tm +(y) Tj +1 0 0 1 384.72 709.08 Tm +(concern) Tj +1 0 0 1 418.92 709.08 Tm +(ha) Tj +1 0 0 1 428.16 709.08 Tm +(v) Tj +1 0 0 1 433.08 709.08 Tm +(e) Tj +1 0 0 1 440.04 709.08 Tm +(been) Tj +1 0 0 1 461.519 709.08 Tm +(e) Tj +1 0 0 1 465.84 709.08 Tm +(x) Tj +1 0 0 1 470.759 709.08 Tm +(ecuted) Tj +1 0 0 1 499.319 709.08 Tm +(by) Tj +1 0 0 1 511.919 709.08 Tm +(at) Tj +1 0 0 1 521.759 709.08 Tm +(least) Tj +ET +q +5 0 0 -9.2 315.4 704.2 cm +/R390 Do +Q +q +6.6 0 0 -6.5 323.9 702.9 cm +/R394 Do +Q +BT +1 0 0 1 333.6 697.08 Tm +(1) Tj +1 0 0 1 341.52 697.08 Tm +(non-f) Tj +1 0 0 1 363.24 697.08 Tm +(aulty) Tj +1 0 0 1 385.92 697.08 Tm +(replicas) Tj +1 0 0 1 419.76 697.08 Tm +(and) Tj +1 0 0 1 437.04 697.08 Tm +(it) Tj +1 0 0 1 445.559 697.08 Tm +(can) Tj +1 0 0 1 462.359 697.08 Tm +(pro) Tj +1 0 0 1 475.679 697.08 Tm +(v) Tj +1 0 0 1 480.599 697.08 Tm +(e) Tj +1 0 0 1 487.679 697.08 Tm +(this) Tj +1 0 0 1 504.959 697.08 Tm +(to) Tj +1 0 0 1 515.639 697.08 Tm +(others) Tj +1 0 0 1 315 685.2 Tm +(in) Tj +1 0 0 1 327 685.2 Tm +(vie) Tj +1 0 0 1 339 685.2 Tm +(w) Tj +1 0 0 1 350.4 685.2 Tm +(changes.) Tj +1 0 0 1 393.96 685.2 Tm +(In) Tj +1 0 0 1 406.56 685.2 Tm +(addition,) Tj +1 0 0 1 446.4 685.2 Tm +(if) Tj +1 0 0 1 456.719 685.2 Tm +(some) Tj +1 0 0 1 482.039 685.2 Tm +(replica) Tj +1 0 0 1 513.479 685.2 Tm +(misses) Tj +1 0 0 1 315 673.2 Tm +(messages) Tj +1 0 0 1 355.44 673.2 Tm +(that) Tj +1 0 0 1 373.32 673.2 Tm +(were) Tj +1 0 0 1 395.52 673.2 Tm +(discarded) Tj +1 0 0 1 436.56 673.2 Tm +(by) Tj +1 0 0 1 449.4 673.2 Tm +(all) Tj +1 0 0 1 462.24 673.2 Tm +(non-f) Tj +1 0 0 1 483.959 673.2 Tm +(aulty) Tj +1 0 0 1 506.519 673.2 Tm +(replicas,) Tj +1 0 0 1 315 661.32 Tm +(it) Tj +1 0 0 1 323.88 661.32 Tm +(will) Tj +1 0 0 1 342.6 661.32 Tm +(need) Tj +1 0 0 1 364.8 661.32 Tm +(to) Tj +1 0 0 1 375.84 661.32 Tm +(be) Tj +1 0 0 1 388.56 661.32 Tm +(brought) Tj +1 0 0 1 422.76 661.32 Tm +(up) Tj +1 0 0 1 436.08 661.32 Tm +(to) Tj +1 0 0 1 447.119 661.32 Tm +(date) Tj +1 0 0 1 467.039 661.32 Tm +(by) Tj +1 0 0 1 480.239 661.32 Tm +(transferring) Tj +1 0 0 1 530.039 661.32 Tm +(all) Tj +1 0 0 1 315 649.32 Tm +(or) Tj +1 0 0 1 325.92 649.32 Tm +(a) Tj +1 0 0 1 333 649.32 Tm +(portion) Tj +1 0 0 1 364.44 649.32 Tm +(of) Tj +1 0 0 1 375.36 649.32 Tm +(the) Tj +1 0 0 1 390.24 649.32 Tm +(service) Tj +1 0 0 1 421.08 649.32 Tm +(state.) Tj +1 0 0 1 445.92 649.32 Tm +(Therefore,) Tj +1 0 0 1 490.319 649.32 Tm +(replicas) Tj +1 0 0 1 523.919 649.32 Tm +(also) Tj +1 0 0 1 315 637.44 Tm +(need) Tj +1 0 0 1 336.36 637.44 Tm +(some) Tj +1 0 0 1 359.88 637.44 Tm +(proof) Tj +1 0 0 1 383.88 637.44 Tm +(that) Tj +1 0 0 1 401.28 637.44 Tm +(the) Tj +1 0 0 1 416.04 637.44 Tm +(state) Tj +1 0 0 1 436.8 637.44 Tm +(is) Tj +1 0 0 1 445.92 637.44 Tm +(correct.) Tj +1 0 0 1 324.96 624.96 Tm +(Generating) Tj +1 0 0 1 372.72 624.96 Tm +(these) Tj +1 0 0 1 396.72 624.96 Tm +(proofs) Tj +1 0 0 1 425.64 624.96 Tm +(after) Tj +1 0 0 1 447.36 624.96 Tm +(e) Tj +1 0 0 1 451.68 624.96 Tm +(x) Tj +1 0 0 1 456.6 624.96 Tm +(ecuting) Tj +1 0 0 1 489.359 624.96 Tm +(e) Tj +1 0 0 1 493.559 624.96 Tm +(v) Tj +1 0 0 1 498.479 624.96 Tm +(ery) Tj +1 0 0 1 514.559 624.96 Tm +(opera-) Tj +1 0 0 1 315 613.08 Tm +(tion) Tj +1 0 0 1 334.68 613.08 Tm +(w) Tj +1 0 0 1 341.76 613.08 Tm +(ould) Tj +1 0 0 1 363.72 613.08 Tm +(be) Tj +1 0 0 1 377.28 613.08 Tm +(e) Tj +1 0 0 1 381.6 613.08 Tm +(xpensi) Tj +1 0 0 1 407.52 613.08 Tm +(v) Tj +1 0 0 1 412.44 613.08 Tm +(e.) Tj +1 0 0 1 427.8 613.08 Tm +(Instead,) Tj +1 0 0 1 463.679 613.08 Tm +(the) Tj +1 0 0 1 475.799 613.08 Tm +(y) Tj +1 0 0 1 484.919 613.08 Tm +(are) Tj +1 0 0 1 501.239 613.08 Tm +(generated) Tj +1 0 0 1 315 601.08 Tm +(periodically) Tj +1 0 0 1 362.28 601.08 Tm +(,) Tj +1 0 0 1 367.08 601.08 Tm +(when) Tj +1 0 0 1 391.2 601.08 Tm +(a) Tj +1 0 0 1 398.16 601.08 Tm +(request) Tj +1 0 0 1 429.48 601.08 Tm +(with) Tj +1 0 0 1 449.88 601.08 Tm +(a) Tj +1 0 0 1 456.84 601.08 Tm +(sequence) Tj +1 0 0 1 495.959 601.08 Tm +(number) Tj +1 0 0 1 528.959 601.08 Tm +(di-) Tj +1 0 0 1 315 589.2 Tm +(visible) Tj +1 0 0 1 344.04 589.2 Tm +(by) Tj +1 0 0 1 356.4 589.2 Tm +(some) Tj +1 0 0 1 379.8 589.2 Tm +(constant) Tj +1 0 0 1 415.44 589.2 Tm +(\(e.g.,) Tj +1 0 0 1 438.12 589.2 Tm +(100\)) Tj +1 0 0 1 458.759 589.2 Tm +(is) Tj +1 0 0 1 467.879 589.2 Tm +(e) Tj +1 0 0 1 472.199 589.2 Tm +(x) Tj +1 0 0 1 477.119 589.2 Tm +(ecuted.) Tj +1 0 0 1 509.039 589.2 Tm +(W) Tj +1 0 0 1 517.559 589.2 Tm +(e) Tj +1 0 0 1 524.519 589.2 Tm +(will) Tj +1 0 0 1 315 577.2 Tm +(refer) Tj +1 0 0 1 336.48 577.2 Tm +(to) Tj +1 0 0 1 346.92 577.2 Tm +(the) Tj +1 0 0 1 361.68 577.2 Tm +(states) Tj +1 0 0 1 386.52 577.2 Tm +(produced) Tj +1 0 0 1 426.24 577.2 Tm +(by) Tj +1 0 0 1 438.84 577.2 Tm +(the) Tj +1 0 0 1 453.72 577.2 Tm +(e) Tj +1 0 0 1 458.04 577.2 Tm +(x) Tj +1 0 0 1 462.959 577.2 Tm +(ecution) Tj +1 0 0 1 494.879 577.2 Tm +(of) Tj +1 0 0 1 505.799 577.2 Tm +(these) Tj +1 0 0 1 528.959 577.2 Tm +(re-) Tj +1 0 0 1 315 565.32 Tm +(quests) Tj +1 0 0 1 342.96 565.32 Tm +(as) Tj +/R359 10 Tf +1 0 0 1 354.24 565.32 Tm +(c) Tj +1 0 0 1 358.56 565.32 Tm +(hec) Tj +1 0 0 1 372.24 565.32 Tm +(kpoints) Tj +/R356 10 Tf +1 0 0 1 403.92 565.32 Tm +(and) Tj +1 0 0 1 421.44 565.32 Tm +(we) Tj +1 0 0 1 436.079 565.32 Tm +(will) Tj +1 0 0 1 454.559 565.32 Tm +(say) Tj +1 0 0 1 470.879 565.32 Tm +(that) Tj +1 0 0 1 488.879 565.32 Tm +(a) Tj +1 0 0 1 496.319 565.32 Tm +(checkpoint) Tj +1 0 0 1 315 553.32 Tm +(with) Tj +1 0 0 1 335.16 553.32 Tm +(a) Tj +1 0 0 1 342.12 553.32 Tm +(proof) Tj +1 0 0 1 366.24 553.32 Tm +(is) Tj +1 0 0 1 375.36 553.32 Tm +(a) Tj +/R359 10 Tf +1 0 0 1 382.2 553.32 Tm +(stable) Tj +1 0 0 1 408.48 553.32 Tm +(c) Tj +1 0 0 1 412.8 553.32 Tm +(hec) Tj +1 0 0 1 426.48 553.32 Tm +(kpoint) Tj +/R356 10 Tf +1 0 0 1 451.559 553.32 Tm +(.) Tj +1 0 0 1 324.96 540.96 Tm +(A) Tj +1 0 0 1 333.72 540.96 Tm +(replica) Tj +1 0 0 1 362.52 540.96 Tm +(maintains) Tj +1 0 0 1 402.96 540.96 Tm +(se) Tj +1 0 0 1 411 540.96 Tm +(v) Tj +1 0 0 1 415.92 540.96 Tm +(eral) Tj +1 0 0 1 432.48 540.96 Tm +(logical) Tj +1 0 0 1 461.279 540.96 Tm +(copies) Tj +1 0 0 1 488.399 540.96 Tm +(of) Tj +1 0 0 1 498.359 540.96 Tm +(the) Tj +1 0 0 1 512.159 540.96 Tm +(service) Tj +1 0 0 1 315 529.08 Tm +(state:) Tj +1 0 0 1 339.24 529.08 Tm +(the) Tj +1 0 0 1 353.04 529.08 Tm +(last) Tj +1 0 0 1 368.52 529.08 Tm +(stable) Tj +1 0 0 1 393.48 529.08 Tm +(checkpoint,) Tj +1 0 0 1 441.479 529.08 Tm +(zero) Tj +1 0 0 1 460.319 529.08 Tm +(or) Tj +1 0 0 1 470.279 529.08 Tm +(more) Tj +1 0 0 1 492.359 529.08 Tm +(checkpoints) Tj +1 0 0 1 315 517.08 Tm +(that) Tj +1 0 0 1 333.6 517.08 Tm +(are) Tj +1 0 0 1 349.44 517.08 Tm +(not) Tj +1 0 0 1 365.88 517.08 Tm +(stable,) Tj +1 0 0 1 395.52 517.08 Tm +(and) Tj +1 0 0 1 413.64 517.08 Tm +(a) Tj +1 0 0 1 421.679 517.08 Tm +(current) Tj +1 0 0 1 453.599 517.08 Tm +(state.) Tj +1 0 0 1 481.439 517.08 Tm +(Cop) Tj +1 0 0 1 497.999 517.08 Tm +(y-on-write) Tj +1 0 0 1 315 505.2 Tm +(techniques) Tj +1 0 0 1 362.16 505.2 Tm +(can) Tj +1 0 0 1 380.4 505.2 Tm +(be) Tj +1 0 0 1 394.32 505.2 Tm +(used) Tj +1 0 0 1 417.12 505.2 Tm +(to) Tj +1 0 0 1 429.36 505.2 Tm +(reduce) Tj +1 0 0 1 460.44 505.2 Tm +(the) Tj +1 0 0 1 477.119 505.2 Tm +(space) Tj +1 0 0 1 503.759 505.2 Tm +(o) Tj +1 0 0 1 508.679 505.2 Tm +(v) Tj +1 0 0 1 513.599 505.2 Tm +(erhead) Tj +1 0 0 1 315 493.2 Tm +(to) Tj +1 0 0 1 327.24 493.2 Tm +(store) Tj +1 0 0 1 351 493.2 Tm +(the) Tj +1 0 0 1 367.68 493.2 Tm +(e) Tj +1 0 0 1 372 493.2 Tm +(xtra) Tj +1 0 0 1 391.92 493.2 Tm +(copies) Tj +1 0 0 1 421.8 493.2 Tm +(of) Tj +1 0 0 1 434.52 493.2 Tm +(the) Tj +1 0 0 1 451.2 493.2 Tm +(state,) Tj +1 0 0 1 476.879 493.2 Tm +(as) Tj +1 0 0 1 489.599 493.2 Tm +(discussed) Tj +1 0 0 1 532.199 493.2 Tm +(in) Tj +1 0 0 1 315 481.2 Tm +(Section) Tj +1 0 0 1 347.4 481.2 Tm +(6.3.) Tj +1 0 0 1 324.96 468.84 Tm +(The) Tj +1 0 0 1 342.96 468.84 Tm +(proof) Tj +1 0 0 1 366.96 468.84 Tm +(of) Tj +1 0 0 1 377.76 468.84 Tm +(correctness) Tj +1 0 0 1 425.04 468.84 Tm +(for) Tj +1 0 0 1 439.08 468.84 Tm +(a) Tj +1 0 0 1 446.04 468.84 Tm +(checkpoint) Tj +1 0 0 1 492.119 468.84 Tm +(is) Tj +1 0 0 1 501.239 468.84 Tm +(generated) Tj +1 0 0 1 315 456.96 Tm +(as) Tj +1 0 0 1 327.6 456.96 Tm +(follo) Tj +1 0 0 1 346.32 456.96 Tm +(ws.) Tj +1 0 0 1 368.76 456.96 Tm +(When) Tj +1 0 0 1 396.84 456.96 Tm +(a) Tj +1 0 0 1 405.6 456.96 Tm +(replica) Tj +ET +q +2.6 0 0 -6.9 437.2 463.8 cm +/R386 Do +Q +BT +1 0 0 1 444.84 456.96 Tm +(produces) Tj +1 0 0 1 485.04 456.96 Tm +(a) Tj +1 0 0 1 493.799 456.96 Tm +(checkpoint,) Tj +1 0 0 1 315 444.96 Tm +(it) Tj +1 0 0 1 324.96 444.96 Tm +(multicasts) Tj +1 0 0 1 369.84 444.96 Tm +(a) Tj +1 0 0 1 378.6 444.96 Tm +(message) Tj +ET +endstream +endobj +407 0 obj +18527 +endobj +408 0 obj +<< /Type /XObject /Name /R408 /Subtype /Image /Length 409 0 R +/ImageMask true /Width 24 /Height 99 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 24 /BlackIs1 true >>] +>> +stream ++t-07X8N5#0B"SKInTVGqHrM(f3_PN_YUJ4L&^prs)7cFqg[B")ufa0GPC"Nr'(,Kmt^kVHo1m8!<~> +endstream +endobj +409 0 obj +80 +endobj +410 0 obj +<< /Length 411 0 R >> +stream +q +2.4 0 0 -9.9 417.8 452.5 cm +/R408 Do +Q +BT +/R356 8 Tf +1 0 0 1 420.72 444.96 Tm +(CHECKPOINT) Tj +ET +q +1.3 0 0 -3 472 446.1 cm +/R370 Do +Q +q +5.4 0 0 -4.6 475.9 449.5 cm +/R378 Do +Q +q +1.3 0 0 -3 482.5 446.1 cm +/R370 Do +Q +endstream +endobj +411 0 obj +216 +endobj +412 0 obj +<< /Type /XObject /Name /R412 /Subtype /Image /Length 413 0 R +/ImageMask true /Width 49 /Height 71 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 49 /BlackIs1 true >>] +>> +stream +-PYiBJV"+0)UN3G',i&VcojQ(@sYoeKmuDG +s)&q9s*o7jJ)C$Fpj^'QJ%rnUr&sfAs+VQL0)H)pV[&D-@^p.P-,mCns*qra5O\@FpQ6S*nl.";J-Z~> +endstream +endobj +413 0 obj +117 +endobj +414 0 obj +<< /Length 415 0 R >> +stream +q +4.9 0 0 -7.1 486.4 452 cm +/R412 Do +Q +q +1.3 0 0 -3 492.1 446.1 cm +/R370 Do +Q +q +2.6 0 0 -6.9 495.9 451.8 cm +/R386 Do +Q +endstream +endobj +415 0 obj +119 +endobj +416 0 obj +<< /Type /XObject /Name /R416 /Subtype /Image /Length 417 0 R +/ImageMask true /Width 24 /Height 99 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 24 /BlackIs1 true >>] +>> +stream +02pd&htA/cs0qnbrP.\`mJhV5_u&o]hqrq@J+[15ImJQWpe1HjlFsc8R~> +endstream +endobj +417 0 obj +78 +endobj +418 0 obj +<< /Length 419 0 R >> +stream +q +2.4 0 0 -9.9 499.6 452.5 cm +/R416 Do +Q +endstream +endobj +419 0 obj +41 +endobj +420 0 obj +<< /Type /XObject /Name /R420 /Subtype /Image /Length 421 0 R +/ImageMask true /Width 42 /Height 31 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 42 /BlackIs1 true >>] +>> +stream +1`R)4"qZ&4%Y>/aKS+]fJ#a+Sp>>oFIt)n@>7^\DYeoBbLCXeP2$%q%J-Z~> +endstream +endobj +421 0 obj +61 +endobj +422 0 obj +<< /Length 423 0 R >> +stream +q +4.2 0 0 -3.1 503.2 446.5 cm +/R420 Do +Q +endstream +endobj +423 0 obj +41 +endobj +424 0 obj +<< /Type /XObject /Name /R424 /Subtype /Image /Length 425 0 R +/ImageMask true /Width 17 /Height 35 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 17 /BlackIs1 true >>] +>> +stream +4B*=`S,;]mpj7Y*s1(">iS6Nohi?6cs!N02p(RKl"9~> +endstream +endobj +425 0 obj +45 +endobj +426 0 obj +<< /Length 427 0 R >> +stream +q +1.7 0 0 -3.5 507.9 446.1 cm +/R424 Do +Q +BT +/R356 10 Tf +1 0 0 1 515.64 444.96 Tm +(to) Tj +1 0 0 1 527.88 444.96 Tm +(the) Tj +1 0 0 1 315 433.08 Tm +(other) Tj +1 0 0 1 339.24 433.08 Tm +(replicas,) Tj +1 0 0 1 376.92 433.08 Tm +(where) Tj +ET +q +5.4 0 0 -4.6 405.2 437.6 cm +/R378 Do +Q +BT +1 0 0 1 414.72 433.08 Tm +(is) Tj +1 0 0 1 425.16 433.08 Tm +(the) Tj +1 0 0 1 441.24 433.08 Tm +(sequence) Tj +1 0 0 1 481.44 433.08 Tm +(number) Tj +1 0 0 1 515.759 433.08 Tm +(of) Tj +1 0 0 1 527.879 433.08 Tm +(the) Tj +1 0 0 1 315 421.08 Tm +(last) Tj +1 0 0 1 333.72 421.08 Tm +(request) Tj +1 0 0 1 367.44 421.08 Tm +(whose) Tj +1 0 0 1 397.8 421.08 Tm +(e) Tj +1 0 0 1 402.12 421.08 Tm +(x) Tj +1 0 0 1 407.04 421.08 Tm +(ecution) Tj +1 0 0 1 441.24 421.08 Tm +(is) Tj +1 0 0 1 452.76 421.08 Tm +(re\257ected) Tj +1 0 0 1 491.999 421.08 Tm +(in) Tj +1 0 0 1 504.599 421.08 Tm +(the) Tj +1 0 0 1 521.759 421.08 Tm +(state) Tj +1 0 0 1 315 409.2 Tm +(and) Tj +ET +q +4.9 0 0 -7.1 333.7 416.2 cm +/R412 Do +Q +BT +1 0 0 1 342.6 409.2 Tm +(is) Tj +1 0 0 1 353.16 409.2 Tm +(the) Tj +1 0 0 1 369.36 409.2 Tm +(digest) Tj +1 0 0 1 397.2 409.2 Tm +(of) Tj +1 0 0 1 409.44 409.2 Tm +(the) Tj +1 0 0 1 425.64 409.2 Tm +(state.) Tj +1 0 0 1 454.559 409.2 Tm +(Each) Tj +1 0 0 1 478.439 409.2 Tm +(replica) Tj +1 0 0 1 509.519 409.2 Tm +(collects) Tj +1 0 0 1 315 397.2 Tm +(checkpoint) Tj +1 0 0 1 363 397.2 Tm +(messages) Tj +1 0 0 1 404.88 397.2 Tm +(in) Tj +1 0 0 1 416.88 397.2 Tm +(its) Tj +1 0 0 1 430.56 397.2 Tm +(log) Tj +1 0 0 1 447.48 397.2 Tm +(until) Tj +1 0 0 1 470.04 397.2 Tm +(it) Tj +1 0 0 1 479.879 397.2 Tm +(has) Tj +1 0 0 1 497.399 397.2 Tm +(2) Tj +ET +q +5 0 0 -9.2 502.8 404.3 cm +/R390 Do +Q +q +6.6 0 0 -6.5 512.1 403 cm +/R394 Do +Q +BT +1 0 0 1 522.479 397.2 Tm +(1) Tj +1 0 0 1 531.719 397.2 Tm +(of) Tj +1 0 0 1 315 385.32 Tm +(them) Tj +1 0 0 1 339.48 385.32 Tm +(for) Tj +1 0 0 1 355.68 385.32 Tm +(sequence) Tj +1 0 0 1 396.84 385.32 Tm +(number) Tj +ET +q +5.4 0 0 -4.6 432 389.8 cm +/R378 Do +Q +BT +1 0 0 1 442.32 385.32 Tm +(with) Tj +1 0 0 1 464.64 385.32 Tm +(the) Tj +1 0 0 1 481.44 385.32 Tm +(same) Tj +1 0 0 1 506.4 385.32 Tm +(digest) Tj +ET +q +4.9 0 0 -7.1 535.1 392.3 cm +/R412 Do +Q +BT +1 0 0 1 315 373.32 Tm +(signed) Tj +1 0 0 1 344.4 373.32 Tm +(by) Tj +1 0 0 1 357.72 373.32 Tm +(dif) Tj +1 0 0 1 368.64 373.32 Tm +(ferent) Tj +1 0 0 1 395.04 373.32 Tm +(replicas) Tj +1 0 0 1 429.48 373.32 Tm +(\(including) Tj +1 0 0 1 473.759 373.32 Tm +(possibly) Tj +1 0 0 1 510.359 373.32 Tm +(its) Tj +1 0 0 1 523.079 373.32 Tm +(o) Tj +1 0 0 1 527.879 373.32 Tm +(wn) Tj +1 0 0 1 315 361.44 Tm +(such) Tj +1 0 0 1 336 361.44 Tm +(message\).) Tj +1 0 0 1 379.8 361.44 Tm +(These) Tj +1 0 0 1 406.32 361.44 Tm +(2) Tj +ET +q +5 0 0 -9.2 411.8 368.5 cm +/R390 Do +Q +q +6.6 0 0 -6.5 420 367.2 cm +/R394 Do +Q +BT +1 0 0 1 429.6 361.44 Tm +(1) Tj +1 0 0 1 437.28 361.44 Tm +(messages) Tj +1 0 0 1 477.6 361.44 Tm +(are) Tj +1 0 0 1 492.48 361.44 Tm +(the) Tj +1 0 0 1 507.359 361.44 Tm +(proof) Tj +1 0 0 1 531.719 361.44 Tm +(of) Tj +1 0 0 1 315 349.44 Tm +(correctness) Tj +1 0 0 1 362.28 349.44 Tm +(for) Tj +1 0 0 1 376.44 349.44 Tm +(the) Tj +1 0 0 1 391.08 349.44 Tm +(checkpoint.) Tj +1 0 0 1 324.96 337.08 Tm +(A) Tj +1 0 0 1 336.6 337.08 Tm +(checkpoint) Tj +1 0 0 1 384.84 337.08 Tm +(with) Tj +1 0 0 1 407.04 337.08 Tm +(a) Tj +1 0 0 1 415.92 337.08 Tm +(proof) Tj +1 0 0 1 441.96 337.08 Tm +(becomes) Tj +1 0 0 1 481.2 337.08 Tm +(stable) Tj +1 0 0 1 508.919 337.08 Tm +(and) Tj +1 0 0 1 527.879 337.08 Tm +(the) Tj +1 0 0 1 315 325.08 Tm +(replica) Tj +1 0 0 1 346.92 325.08 Tm +(discards) Tj +1 0 0 1 384.24 325.08 Tm +(all) Tj +1 0 0 1 399 325.08 Tm +(pre-prepare,) Tj +1 0 0 1 452.76 325.08 Tm +(prepare,) Tj +1 0 0 1 490.439 325.08 Tm +(and) Tj +1 0 0 1 509.519 325.08 Tm +(commit) Tj +1 0 0 1 315 313.2 Tm +(messages) Tj +1 0 0 1 356.76 313.2 Tm +(with) Tj +1 0 0 1 378.6 313.2 Tm +(sequence) Tj +1 0 0 1 419.28 313.2 Tm +(number) Tj +1 0 0 1 453.84 313.2 Tm +(less) Tj +1 0 0 1 472.8 313.2 Tm +(than) Tj +1 0 0 1 494.159 313.2 Tm +(or) Tj +1 0 0 1 506.519 313.2 Tm +(equal) Tj +1 0 0 1 532.199 313.2 Tm +(to) Tj +ET +q +5.4 0 0 -4.6 315.2 305.7 cm +/R378 Do +Q +BT +1 0 0 1 323.28 301.2 Tm +(from) Tj +1 0 0 1 345 301.2 Tm +(its) Tj +1 0 0 1 356.64 301.2 Tm +(log;) Tj +1 0 0 1 374.52 301.2 Tm +(it) Tj +1 0 0 1 382.44 301.2 Tm +(also) Tj +1 0 0 1 400.8 301.2 Tm +(discards) Tj +1 0 0 1 435.719 301.2 Tm +(all) Tj +1 0 0 1 447.959 301.2 Tm +(earlier) Tj +1 0 0 1 475.679 301.2 Tm +(checkpoints) Tj +1 0 0 1 525.599 301.2 Tm +(and) Tj +1 0 0 1 315 289.32 Tm +(checkpoint) Tj +1 0 0 1 361.2 289.32 Tm +(messages.) Tj +1 0 0 1 324.96 276.96 Tm +(Computing) Tj +1 0 0 1 373.56 276.96 Tm +(the) Tj +1 0 0 1 389.4 276.96 Tm +(proofs) Tj +1 0 0 1 418.56 276.96 Tm +(is) Tj +1 0 0 1 429 276.96 Tm +(ef) Tj +1 0 0 1 436.56 276.96 Tm +(\256cient) Tj +1 0 0 1 465 276.96 Tm +(because) Tj +1 0 0 1 500.279 276.96 Tm +(the) Tj +1 0 0 1 516.239 276.96 Tm +(digest) Tj +1 0 0 1 315 264.96 Tm +(can) Tj +1 0 0 1 331.8 264.96 Tm +(be) Tj +1 0 0 1 344.16 264.96 Tm +(computed) Tj +1 0 0 1 386.52 264.96 Tm +(using) Tj +1 0 0 1 411 264.96 Tm +(incremental) Tj +1 0 0 1 461.04 264.96 Tm +(cryptography) Tj +1 0 0 1 517.079 264.96 Tm +([1) Tj +1 0 0 1 525.359 264.96 Tm +(]) Tj +1 0 0 1 531.719 264.96 Tm +(as) Tj +1 0 0 1 315 253.08 Tm +(discussed) Tj +1 0 0 1 355.44 253.08 Tm +(in) Tj +1 0 0 1 365.52 253.08 Tm +(Section) Tj +1 0 0 1 397.68 253.08 Tm +(6.3,) Tj +1 0 0 1 414.96 253.08 Tm +(and) Tj +1 0 0 1 431.64 253.08 Tm +(proofs) Tj +1 0 0 1 459.479 253.08 Tm +(are) Tj +1 0 0 1 473.879 253.08 Tm +(generated) Tj +1 0 0 1 514.919 253.08 Tm +(rarely) Tj +1 0 0 1 537.719 253.08 Tm +(.) Tj +1 0 0 1 324.96 240.72 Tm +(The) Tj +1 0 0 1 344.28 240.72 Tm +(checkpoint) Tj +1 0 0 1 391.8 240.72 Tm +(protocol) Tj +1 0 0 1 428.88 240.72 Tm +(is) Tj +1 0 0 1 439.32 240.72 Tm +(used) Tj +1 0 0 1 461.52 240.72 Tm +(to) Tj +1 0 0 1 473.04 240.72 Tm +(adv) Tj +1 0 0 1 487.319 240.72 Tm +(ance) Tj +1 0 0 1 509.279 240.72 Tm +(the) Tj +1 0 0 1 525.359 240.72 Tm +(lo) Tj +1 0 0 1 532.919 240.72 Tm +(w) Tj +1 0 0 1 315 228.72 Tm +(and) Tj +1 0 0 1 332.88 228.72 Tm +(high) Tj +1 0 0 1 354 228.72 Tm +(w) Tj +1 0 0 1 361.08 228.72 Tm +(ater) Tj +1 0 0 1 379.44 228.72 Tm +(marks) Tj +1 0 0 1 407.28 228.72 Tm +(\(which) Tj +1 0 0 1 438.36 228.72 Tm +(limit) Tj +1 0 0 1 460.56 228.72 Tm +(what) Tj +1 0 0 1 483.479 228.72 Tm +(messages) Tj +1 0 0 1 524.519 228.72 Tm +(will) Tj +1 0 0 1 315 216.84 Tm +(be) Tj +1 0 0 1 328.8 216.84 Tm +(accepted\).) Tj +1 0 0 1 378.96 216.84 Tm +(The) Tj +1 0 0 1 398.88 216.84 Tm +(lo) Tj +1 0 0 1 406.44 216.84 Tm +(w-w) Tj +1 0 0 1 424.08 216.84 Tm +(ater) Tj +1 0 0 1 443.4 216.84 Tm +(mark) Tj +ET +endstream +endobj +427 0 obj +6379 +endobj +428 0 obj +<< /Type /XObject /Name /R428 /Subtype /Image /Length 429 0 R +/ImageMask true /Width 50 /Height 71 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 50 /BlackIs1 true >>] +>> +stream +0Zl\u@0_OZ#Wp,3rBJ3g^\>2: +rV-&ursUj$gmgWn*E:5fO9!_pbg-FlJ[WfIqQ^[&,mZcqtL,7s*eVhs5pRo!WX>~> +endstream +endobj +429 0 obj +108 +endobj +430 0 obj +<< /Length 431 0 R >> +stream +q +5 0 0 -7.1 468.8 223.8 cm +/R428 Do +Q +BT +1 0 0 1 478.56 216.84 Tm +(is) Tj +1 0 0 1 489.6 216.84 Tm +(equal) Tj +1 0 0 1 515.639 216.84 Tm +(to) Tj +1 0 0 1 527.879 216.84 Tm +(the) Tj +1 0 0 1 315 204.84 Tm +(sequence) Tj +1 0 0 1 354 204.84 Tm +(number) Tj +1 0 0 1 387 204.84 Tm +(of) Tj +1 0 0 1 397.68 204.84 Tm +(the) Tj +1 0 0 1 412.44 204.84 Tm +(last) Tj +1 0 0 1 428.76 204.84 Tm +(stable) Tj +1 0 0 1 454.44 204.84 Tm +(checkpoint.) Tj +1 0 0 1 504.239 204.84 Tm +(The) Tj +1 0 0 1 522.239 204.84 Tm +(high) Tj +1 0 0 1 315 192.84 Tm +(w) Tj +1 0 0 1 322.08 192.84 Tm +(ater) Tj +1 0 0 1 340.2 192.84 Tm +(mark) Tj +ET +endstream +endobj +431 0 obj +599 +endobj +432 0 obj +<< /Type /XObject /Name /R432 /Subtype /Image /Length 433 0 R +/ImageMask true /Width 84 /Height 69 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 84 /BlackIs1 true >>] +>> +stream +-D8e7)F+8D_$Y]c$*6AK +mN;?)rZ1b5h;A,.rpTgb+7B0HrVjk:r%\j3J,S4'^YeJZhr4S'r'0*,+8c$9*VY&PJY_cls1 +endstream +endobj +433 0 obj +102 +endobj +434 0 obj +<< /Length 435 0 R >> +stream +q +8.4 0 0 -6.9 364.3 199.7 cm +/R432 Do +Q +endstream +endobj +435 0 obj +41 +endobj +436 0 obj +<< /Type /XObject /Name /R436 /Subtype /Image /Length 437 0 R +/ImageMask true /Width 66 /Height 27 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 66 /BlackIs1 true >>] +>> +stream +/e%CP4)j*Wq&BoPm6CdS5Q~> +endstream +endobj +437 0 obj +25 +endobj +438 0 obj +<< /Length 439 0 R >> +stream +q +6.6 0 0 -2.7 377.3 196.7 cm +/R436 Do +Q +q +5 0 0 -7.1 388.7 199.8 cm +/R428 Do +Q +q +6.6 0 0 -6.5 397.1 198.6 cm +/R394 Do +Q +endstream +endobj +439 0 obj +121 +endobj +440 0 obj +<< /Type /XObject /Name /R440 /Subtype /Image /Length 441 0 R +/ImageMask true /Width 47 /Height 71 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 47 /BlackIs1 true >>] +>> +stream +0ZYuc@0_O>l/"oP624[@s1A +endstream +endobj +441 0 obj +119 +endobj +442 0 obj +<< /Length 443 0 R >> +stream +q +4.7 0 0 -7.1 407.4 199.8 cm +/R440 Do +Q +BT +1 0 0 1 412.56 192.84 Tm +(,) Tj +1 0 0 1 418.44 192.84 Tm +(where) Tj +ET +q +4.7 0 0 -7.1 446.3 199.8 cm +/R440 Do +Q +BT +1 0 0 1 454.679 192.84 Tm +(is) Tj +1 0 0 1 464.519 192.84 Tm +(big) Tj +1 0 0 1 480.479 192.84 Tm +(enough) Tj +1 0 0 1 512.999 192.84 Tm +(so) Tj +1 0 0 1 524.999 192.84 Tm +(that) Tj +1 0 0 1 315 180.96 Tm +(replicas) Tj +1 0 0 1 348.96 180.96 Tm +(do) Tj +1 0 0 1 361.8 180.96 Tm +(not) Tj +1 0 0 1 377.52 180.96 Tm +(stall) Tj +1 0 0 1 397.08 180.96 Tm +(w) Tj +1 0 0 1 404.16 180.96 Tm +(aiting) Tj +1 0 0 1 429.72 180.96 Tm +(for) Tj +1 0 0 1 444.359 180.96 Tm +(a) Tj +1 0 0 1 451.679 180.96 Tm +(checkpoint) Tj +1 0 0 1 498.359 180.96 Tm +(to) Tj +1 0 0 1 509.039 180.96 Tm +(become) Tj +1 0 0 1 315 168.96 Tm +(stable.) Tj +1 0 0 1 345.6 168.96 Tm +(F) Tj +1 0 0 1 351 168.96 Tm +(or) Tj +1 0 0 1 362.28 168.96 Tm +(e) Tj +1 0 0 1 366.6 168.96 Tm +(xample,) Tj +1 0 0 1 401.4 168.96 Tm +(if) Tj +1 0 0 1 410.4 168.96 Tm +(checkpoints) Tj +1 0 0 1 460.919 168.96 Tm +(are) Tj +1 0 0 1 476.039 168.96 Tm +(tak) Tj +1 0 0 1 488.159 168.96 Tm +(en) Tj +1 0 0 1 500.399 168.96 Tm +(e) Tj +1 0 0 1 504.599 168.96 Tm +(v) Tj +1 0 0 1 509.519 168.96 Tm +(ery) Tj +1 0 0 1 524.999 168.96 Tm +(100) Tj +1 0 0 1 315 157.08 Tm +(requests,) Tj +ET +q +4.7 0 0 -7.1 353.1 164.1 cm +/R440 Do +Q +BT +1 0 0 1 360.6 157.08 Tm +(might) Tj +1 0 0 1 386.4 157.08 Tm +(be) Tj +1 0 0 1 398.28 157.08 Tm +(200.) Tj +/R405 10 Tf +1 0 0 1 315 133.8 Tm +(4.4) Tj +1 0 0 1 337.44 133.8 Tm +(V) Tj +1 0 0 1 344.28 133.8 Tm +(iew) Tj +1 0 0 1 361.08 133.8 Tm +(Changes) Tj +/R356 10 Tf +1 0 0 1 315 118.8 Tm +(The) Tj +1 0 0 1 333 118.8 Tm +(vie) Tj +1 0 0 1 345 118.8 Tm +(w-change) Tj +1 0 0 1 386.16 118.8 Tm +(protocol) Tj +1 0 0 1 421.92 118.8 Tm +(pro) Tj +1 0 0 1 435.24 118.8 Tm +(vides) Tj +1 0 0 1 458.64 118.8 Tm +(li) Tj +1 0 0 1 463.92 118.8 Tm +(v) Tj +1 0 0 1 468.84 118.8 Tm +(eness) Tj +1 0 0 1 492.839 118.8 Tm +(by) Tj +1 0 0 1 505.319 118.8 Tm +(allo) Tj +1 0 0 1 520.079 118.8 Tm +(wing) Tj +1 0 0 1 315 106.8 Tm +(the) Tj +1 0 0 1 328.92 106.8 Tm +(system) Tj +1 0 0 1 358.32 106.8 Tm +(to) Tj +1 0 0 1 367.8 106.8 Tm +(mak) Tj +1 0 0 1 384.96 106.8 Tm +(e) Tj +1 0 0 1 391.08 106.8 Tm +(progress) Tj +1 0 0 1 426.72 106.8 Tm +(when) Tj +1 0 0 1 450.12 106.8 Tm +(th) Tj +1 0 0 1 457.799 106.8 Tm +(e) Tj +1 0 0 1 463.919 106.8 Tm +(pr) Tj +1 0 0 1 472.199 106.8 Tm +(imar) Tj +1 0 0 1 490.439 106.8 Tm +(y) Tj +1 0 0 1 497.159 106.8 Tm +(f) Tj +1 0 0 1 500.279 106.8 Tm +(ails.) Tj +1 0 0 1 519.239 106.8 Tm +(V) Tj +1 0 0 1 525.839 106.8 Tm +(ie) Tj +1 0 0 1 532.799 106.8 Tm +(w) Tj +1 0 0 1 315 94.92 Tm +(changes) Tj +1 0 0 1 350.4 94.92 Tm +(are) Tj +1 0 0 1 366 94.92 Tm +(triggered) Tj +1 0 0 1 405.24 94.92 Tm +(by) Tj +1 0 0 1 418.56 94.92 Tm +(timeouts) Tj +1 0 0 1 456.24 94.92 Tm +(that) Tj +1 0 0 1 474.479 94.92 Tm +(pre) Tj +1 0 0 1 487.079 94.92 Tm +(v) Tj +1 0 0 1 491.999 94.92 Tm +(ent) Tj +1 0 0 1 507.359 94.92 Tm +(backups) Tj +1 0 0 1 315 82.92 Tm +(from) Tj +1 0 0 1 339.12 82.92 Tm +(w) Tj +1 0 0 1 346.2 82.92 Tm +(aiting) Tj +1 0 0 1 373.56 82.92 Tm +(inde\256nitely) Tj +1 0 0 1 423.72 82.92 Tm +(for) Tj +1 0 0 1 440.04 82.92 Tm +(requests) Tj +1 0 0 1 477.479 82.92 Tm +(to) Tj +1 0 0 1 489.839 82.92 Tm +(e) Tj +1 0 0 1 494.159 82.92 Tm +(x) Tj +1 0 0 1 499.079 82.92 Tm +(ecute.) Tj +1 0 0 1 532.799 82.92 Tm +(A) Tj +1 0 0 1 315 71.04 Tm +(backup) Tj +1 0 0 1 345.6 71.04 Tm +(is) Tj +/R359 10 Tf +1 0 0 1 353.76 71.04 Tm +(waiting) Tj +/R356 10 Tf +1 0 0 1 384.96 71.04 Tm +(for) Tj +1 0 0 1 398.28 71.04 Tm +(a) Tj +1 0 0 1 404.28 71.04 Tm +(request) Tj +1 0 0 1 434.76 71.04 Tm +(if) Tj +1 0 0 1 442.44 71.04 Tm +(it) Tj +1 0 0 1 449.519 71.04 Tm +(recei) Tj +1 0 0 1 468.719 71.04 Tm +(v) Tj +1 0 0 1 473.639 71.04 Tm +(ed) Tj +1 0 0 1 484.679 71.04 Tm +(a) Tj +1 0 0 1 490.679 71.04 Tm +(v) Tj +1 0 0 1 495.479 71.04 Tm +(alid) Tj +1 0 0 1 511.919 71.04 Tm +(req) Tj +1 0 0 1 524.639 71.04 Tm +(uest) Tj +1 0 0 1 303.48 35.16 Tm +(5) Tj +ET +Q +endstream +endobj +443 0 obj +3824 +endobj +444 0 obj +<< +/Type /Page +/MediaBox [0 0 612 792] +/Parent 2 0 R +/Resources << /ProcSet [/PDF /ImageB /Text] +/Font << +/R405 405 0 R +/R402 402 0 R +/R359 359 0 R +/R356 356 0 R +>> +/XObject << +/R440 440 0 R +/R394 394 0 R +/R428 428 0 R +/R436 436 0 R +/R432 432 0 R +/R378 378 0 R +/R390 390 0 R +/R412 412 0 R +/R424 424 0 R +/R420 420 0 R +/R416 416 0 R +/R386 386 0 R +/R370 370 0 R +/R408 408 0 R +/R398 398 0 R +/R382 382 0 R +/R374 374 0 R +/R366 366 0 R +/R362 362 0 R +>> +>> +/Contents [ +357 0 R +360 0 R +364 0 R +368 0 R +372 0 R +376 0 R +380 0 R +384 0 R +388 0 R +392 0 R +396 0 R +400 0 R +403 0 R +406 0 R +410 0 R +414 0 R +418 0 R +422 0 R +426 0 R +430 0 R +434 0 R +438 0 R +442 0 R +] +>> +endobj +445 0 obj +<< /Type /Font /Name /R445 /Subtype /Type1 /BaseFont /Times-Roman >> +endobj +446 0 obj +<< /Length 447 0 R >> +stream +q +BT +/R445 10 Tf +1 0 0 1 72 709.08 Tm +(and) Tj +1 0 0 1 89.16 709.08 Tm +(has) Tj +1 0 0 1 105.12 709.08 Tm +(not) Tj +1 0 0 1 120.6 709.08 Tm +(e) Tj +1 0 0 1 124.92 709.08 Tm +(x) Tj +1 0 0 1 129.84 709.08 Tm +(ecuted) Tj +1 0 0 1 158.52 709.08 Tm +(it.) Tj +1 0 0 1 170.88 709.08 Tm +(A) Tj +1 0 0 1 180.84 709.08 Tm +(backup) Tj +1 0 0 1 212.4 709.08 Tm +(starts) Tj +1 0 0 1 236.159 709.08 Tm +(a) Tj +1 0 0 1 243.359 709.08 Tm +(timer) Tj +1 0 0 1 267.119 709.08 Tm +(when) Tj +1 0 0 1 291.479 709.08 Tm +(it) Tj +1 0 0 1 72 697.08 Tm +(recei) Tj +1 0 0 1 91.2 697.08 Tm +(v) Tj +1 0 0 1 96.1199 697.08 Tm +(es) Tj +1 0 0 1 107.52 697.08 Tm +(a) Tj +1 0 0 1 115.32 697.08 Tm +(request) Tj +1 0 0 1 147.36 697.08 Tm +(and) Tj +1 0 0 1 165.12 697.08 Tm +(the) Tj +1 0 0 1 180.6 697.08 Tm +(timer) Tj +1 0 0 1 204.84 697.08 Tm +(is) Tj +1 0 0 1 214.8 697.08 Tm +(not) Tj +1 0 0 1 230.879 697.08 Tm +(already) Tj +1 0 0 1 263.519 697.08 Tm +(running.) Tj +1 0 0 1 72 685.2 Tm +(It) Tj +1 0 0 1 80.64 685.2 Tm +(stops) Tj +1 0 0 1 103.68 685.2 Tm +(the) Tj +1 0 0 1 118.44 685.2 Tm +(timer) Tj +1 0 0 1 142.08 685.2 Tm +(when) Tj +1 0 0 1 166.2 685.2 Tm +(it) Tj +1 0 0 1 174.36 685.2 Tm +(is) Tj +1 0 0 1 183.6 685.2 Tm +(no) Tj +1 0 0 1 196.08 685.2 Tm +(longer) Tj +1 0 0 1 224.159 685.2 Tm +(w) Tj +1 0 0 1 231.239 685.2 Tm +(aiting) Tj +1 0 0 1 256.559 685.2 Tm +(to) Tj +1 0 0 1 266.879 685.2 Tm +(e) Tj +1 0 0 1 271.199 685.2 Tm +(x) Tj +1 0 0 1 276.119 685.2 Tm +(ecute) Tj +1 0 0 1 72 673.2 Tm +(the) Tj +1 0 0 1 87.2399 673.2 Tm +(request,) Tj +1 0 0 1 121.56 673.2 Tm +(b) Tj +1 0 0 1 126.36 673.2 Tm +(ut) Tj +1 0 0 1 137.16 673.2 Tm +(restarts) Tj +1 0 0 1 168.96 673.2 Tm +(it) Tj +1 0 0 1 177.48 673.2 Tm +(if) Tj +1 0 0 1 186.599 673.2 Tm +(at) Tj +1 0 0 1 196.799 673.2 Tm +(that) Tj +1 0 0 1 214.799 673.2 Tm +(point) Tj +1 0 0 1 238.199 673.2 Tm +(it) Tj +1 0 0 1 246.839 673.2 Tm +(is) Tj +1 0 0 1 256.439 673.2 Tm +(w) Tj +1 0 0 1 263.519 673.2 Tm +(aiting) Tj +1 0 0 1 289.199 673.2 Tm +(to) Tj +1 0 0 1 72 661.32 Tm +(e) Tj +1 0 0 1 76.32 661.32 Tm +(x) Tj +1 0 0 1 81.24 661.32 Tm +(ecute) Tj +1 0 0 1 104.64 661.32 Tm +(some) Tj +1 0 0 1 128.16 661.32 Tm +(other) Tj +1 0 0 1 151.08 661.32 Tm +(request.) Tj +1 0 0 1 81.96 649.32 Tm +(If) Tj +1 0 0 1 94.4399 649.32 Tm +(the) Tj +1 0 0 1 112.44 649.32 Tm +(timer) Tj +1 0 0 1 139.32 649.32 Tm +(of) Tj +1 0 0 1 153.48 649.32 Tm +(backup) Tj +ET +Q +q +W +0 0 612 792 re +n +endstream +endobj +447 0 obj +2311 +endobj +448 0 obj +<< /Type /XObject /Name /R448 /Subtype /Image /Length 449 0 R +/ImageMask true /Width 26 /Height 69 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 26 /BlackIs1 true >>] +>> +stream +-R^*Zl/YrmIf:DfhtVWo^UNqc%K6'1s*addX2"DHrSQB/+4u(tp(mUa[G\kks2oQE"5iR%(OubkJ,~> +endstream +endobj +449 0 obj +80 +endobj +450 0 obj +<< /Length 451 0 R >> +stream +q +2.6 0 0 -6.9 188.4 656.1 cm +/R448 Do +Q +BT +/R445 10 Tf +1 0 0 1 197.4 649.32 Tm +(e) Tj +1 0 0 1 201.72 649.32 Tm +(xpires) Tj +1 0 0 1 231.84 649.32 Tm +(in) Tj +1 0 0 1 245.519 649.32 Tm +(vie) Tj +1 0 0 1 257.519 649.32 Tm +(w) Tj +ET +endstream +endobj +451 0 obj +228 +endobj +452 0 obj +<< /Type /XObject /Name /R452 /Subtype /Image /Length 453 0 R +/ImageMask true /Width 44 /Height 46 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 44 /BlackIs1 true >>] +>> +stream +2$jBq +&9R,hkFhTE_G0fRqg[T's0obU5O\AciSX+=pUN/*r',Z@_Z&BgfC8fKf@@X?]DE=DTC?6WJ-Z.*~> +endstream +endobj +453 0 obj +84 +endobj +454 0 obj +<< /Length 455 0 R >> +stream +q +4.4 0 0 -4.6 270.7 653.8 cm +/R452 Do +Q +BT +1 0 0 1 275.639 649.32 Tm +(,) Tj +1 0 0 1 284.879 649.32 Tm +(the) Tj +1 0 0 1 72 637.44 Tm +(backup) Tj +1 0 0 1 105.96 637.44 Tm +(starts) Tj +1 0 0 1 132.12 637.44 Tm +(a) Tj +1 0 0 1 141.6 637.44 Tm +(vie) Tj +1 0 0 1 153.6 637.44 Tm +(w) Tj +1 0 0 1 165.96 637.44 Tm +(change) Tj +1 0 0 1 199.2 637.44 Tm +(to) Tj +1 0 0 1 212.16 637.44 Tm +(mo) Tj +1 0 0 1 224.88 637.44 Tm +(v) Tj +1 0 0 1 229.8 637.44 Tm +(e) Tj +1 0 0 1 239.16 637.44 Tm +(the) Tj +1 0 0 1 256.44 637.44 Tm +(system) Tj +1 0 0 1 289.199 637.44 Tm +(to) Tj +1 0 0 1 72 625.44 Tm +(vie) Tj +1 0 0 1 84 625.44 Tm +(w) Tj +ET +q +4.4 0 0 -4.6 95.7 629.9 cm +/R452 Do +Q +endstream +endobj +455 0 obj +649 +endobj +456 0 obj +<< /Type /XObject /Name /R456 /Subtype /Image /Length 457 0 R +/ImageMask true /Width 66 /Height 65 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 66 /BlackIs1 true >>] +>> +stream +,BZ6Ns8W-!s8W-!s-Eq9Iaj?!s8W-!s8W-!s8Uf'#QTA~> +endstream +endobj +457 0 obj +47 +endobj +458 0 obj +<< /Length 459 0 R >> +stream +q +6.6 0 0 -6.5 104.4 631.2 cm +/R456 Do +Q +BT +1 0 0 1 114.96 625.44 Tm +(1.) Tj +1 0 0 1 131.76 625.44 Tm +(It) Tj +1 0 0 1 142.2 625.44 Tm +(stops) Tj +1 0 0 1 167.04 625.44 Tm +(accepting) Tj +1 0 0 1 209.64 625.44 Tm +(messages) Tj +1 0 0 1 251.639 625.44 Tm +(\(other) Tj +1 0 0 1 279.839 625.44 Tm +(than) Tj +1 0 0 1 72 613.44 Tm +(checkpoint,) Tj +1 0 0 1 121.8 613.44 Tm +(vie) Tj +1 0 0 1 133.8 613.44 Tm +(w-change,) Tj +1 0 0 1 178.44 613.44 Tm +(and) Tj +1 0 0 1 196.2 613.44 Tm +(ne) Tj +1 0 0 1 205.44 613.44 Tm +(w-vie) Tj +1 0 0 1 228 613.44 Tm +(w) Tj +1 0 0 1 238.32 613.44 Tm +(messages\)) Tj +1 0 0 1 282.6 613.44 Tm +(and) Tj +1 0 0 1 72 601.56 Tm +(multicasts) Tj +1 0 0 1 114.72 601.56 Tm +(a) Tj +ET +endstream +endobj +459 0 obj +686 +endobj +460 0 obj +<< /Type /XObject /Name /R460 /Subtype /Image /Length 461 0 R +/ImageMask true /Width 24 /Height 99 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 24 /BlackIs1 true >>] +>> +stream ++t-07X8N5#0B"SKInTVGqHrM(f3_PN_YUJ4L&^prs)7cFqg[B")ufa0GPC"Nr'(,Kmt^kVHo1m8!<~> +endstream +endobj +461 0 obj +80 +endobj +462 0 obj +<< /Length 463 0 R >> +stream +q +2.4 0 0 -9.9 122.3 609.1 cm +/R460 Do +Q +BT +/R445 8 Tf +1 0 0 1 125.28 601.56 Tm +(VIEW) Tj +/R445 10 Tf +1 0 0 1 145.68 601.56 Tm +(-) Tj +/R445 8 Tf +1 0 0 1 149.04 601.56 Tm +(CHANGE) Tj +ET +endstream +endobj +463 0 obj +185 +endobj +464 0 obj +<< /Type /XObject /Name /R464 /Subtype /Image /Length 465 0 R +/ImageMask true /Width 13 /Height 30 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 13 /BlackIs1 true >>] +>> +stream +0g^A/h=&*equ,LR2Xj(@lh_"t-iec'~> +endstream +endobj +465 0 obj +33 +endobj +466 0 obj +<< /Length 467 0 R >> +stream +q +1.3 0 0 -3 183 602.7 cm +/R464 Do +Q +q +4.4 0 0 -4.6 186.8 606.1 cm +/R452 Do +Q +q +6.6 0 0 -6.5 194.1 607.4 cm +/R456 Do +Q +BT +/R445 10 Tf +1 0 0 1 203.16 601.56 Tm +(1) Tj +ET +q +1.3 0 0 -3 209 602.7 cm +/R464 Do +Q +endstream +endobj +467 0 obj +206 +endobj +468 0 obj +<< /Type /XObject /Name /R468 /Subtype /Image /Length 469 0 R +/ImageMask true /Width 54 /Height 46 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 54 /BlackIs1 true >>] +>> +stream +4N]t,$m>7rl/"o@d!re, +pYTohhg]lis4_o!aocV_^Yur>rVsXfR^`8 +endstream +endobj +469 0 obj +102 +endobj +470 0 obj +<< /Length 471 0 R >> +stream +q +5.4 0 0 -4.6 212.8 606.1 cm +/R468 Do +Q +q +1.3 0 0 -3 219.4 602.7 cm +/R464 Do +Q +endstream +endobj +471 0 obj +80 +endobj +472 0 obj +<< /Type /XObject /Name /R472 /Subtype /Image /Length 473 0 R +/ImageMask true /Width 51 /Height 74 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 51 /BlackIs1 true >>] +>> +stream +,Q`3Ui.96s*~> +endstream +endobj +473 0 obj +108 +endobj +474 0 obj +<< /Length 475 0 R >> +stream +q +5.1 0 0 -7.4 223.2 608.7 cm +/R472 Do +Q +q +1.3 0 0 -3 229.8 602.7 cm +/R464 Do +Q +endstream +endobj +475 0 obj +80 +endobj +476 0 obj +<< /Type /XObject /Name /R476 /Subtype /Image /Length 477 0 R +/ImageMask true /Width 71 /Height 75 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 71 /BlackIs1 true >>] +>> +stream +-a<+0n+Z5LIJ`KNiW!ir]91&&#RC*7(FNXHi:o#"8-c],:)i/ImgL%P +mJ$#;hX9n1s0oc#Hf+iJh4SuL\$#UXrTYHSLP2l)n?sB\%/C-b5U_6s +endstream +endobj +477 0 obj +134 +endobj +478 0 obj +<< /Length 479 0 R >> +stream +q +7.1 0 0 -7.5 233.5 608.5 cm +/R476 Do +Q +q +1.3 0 0 -3 242 602.7 cm +/R464 Do +Q +q +2.6 0 0 -6.9 245.8 608.4 cm +/R448 Do +Q +endstream +endobj +479 0 obj +119 +endobj +480 0 obj +<< /Type /XObject /Name /R480 /Subtype /Image /Length 481 0 R +/ImageMask true /Width 24 /Height 99 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 24 /BlackIs1 true >>] +>> +stream +02pd&htA/cs0qnbrP.\`mJhV5_u&o]hqrq@J+[15ImJQWpe1HjlFsc8R~> +endstream +endobj +481 0 obj +78 +endobj +482 0 obj +<< /Length 483 0 R >> +stream +q +2.4 0 0 -9.9 249.5 609.1 cm +/R480 Do +Q +endstream +endobj +483 0 obj +41 +endobj +484 0 obj +<< /Type /XObject /Name /R484 /Subtype /Image /Length 485 0 R +/ImageMask true /Width 42 /Height 31 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 42 /BlackIs1 true >>] +>> +stream +1`R)4"qZ&4%Y>/aKS+]fJ#a+Sp>>oFIt)n@>7^\DYeoBbLCXeP2$%q%J-Z~> +endstream +endobj +485 0 obj +61 +endobj +486 0 obj +<< /Length 487 0 R >> +stream +q +4.2 0 0 -3.1 253.2 603.1 cm +/R484 Do +Q +endstream +endobj +487 0 obj +41 +endobj +488 0 obj +<< /Type /XObject /Name /R488 /Subtype /Image /Length 489 0 R +/ImageMask true /Width 17 /Height 35 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 17 /BlackIs1 true >>] +>> +stream +4B*=`S,;]mpj7Y*s1(">iS6Nohi?6cs!N02p(RKl"9~> +endstream +endobj +489 0 obj +45 +endobj +490 0 obj +<< /Length 491 0 R >> +stream +q +1.7 0 0 -3.5 257.8 602.7 cm +/R488 Do +Q +BT +1 0 0 1 263.28 601.56 Tm +(message) Tj +1 0 0 1 72 589.56 Tm +(to) Tj +1 0 0 1 82.2 589.56 Tm +(all) Tj +1 0 0 1 94.6799 589.56 Tm +(replicas.) Tj +1 0 0 1 131.76 589.56 Tm +(Here) Tj +ET +q +5.4 0 0 -4.6 153.8 594.1 cm +/R468 Do +Q +BT +1 0 0 1 162 589.56 Tm +(is) Tj +1 0 0 1 171.12 589.56 Tm +(the) Tj +1 0 0 1 185.76 589.56 Tm +(sequence) Tj +1 0 0 1 224.759 589.56 Tm +(number) Tj +1 0 0 1 257.759 589.56 Tm +(of) Tj +1 0 0 1 268.559 589.56 Tm +(the) Tj +1 0 0 1 283.199 589.56 Tm +(last) Tj +1 0 0 1 72 577.68 Tm +(stable) Tj +1 0 0 1 97.6799 577.68 Tm +(checkpoint) Tj +ET +endstream +endobj +491 0 obj +590 +endobj +492 0 obj +<< /Type /XObject /Name /R492 /Subtype /Image /Length 493 0 R +/ImageMask true /Width 37 /Height 46 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 37 /BlackIs1 true >>] +>> +stream +1^aSA +!J1f4d1VKfY*42Ai;KnWs7a;-)HcZabs@p6K&KV(n>M'uOFo&O1?^[smruILGVDte$reoe$n)++#Q~> +endstream +endobj +493 0 obj +86 +endobj +494 0 obj +<< /Length 495 0 R >> +stream +q +3.7 0 0 -4.6 144.2 582.2 cm +/R492 Do +Q +BT +1 0 0 1 150.84 577.68 Tm +(kno) Tj +1 0 0 1 165.72 577.68 Tm +(wn) Tj +1 0 0 1 180.24 577.68 Tm +(to) Tj +ET +q +2.6 0 0 -6.9 190.5 584.5 cm +/R448 Do +Q +BT +1 0 0 1 193.8 577.68 Tm +(,) Tj +ET +q +5.1 0 0 -7.4 198.8 584.8 cm +/R472 Do +Q +BT +1 0 0 1 207 577.68 Tm +(is) Tj +1 0 0 1 215.999 577.68 Tm +(a) Tj +1 0 0 1 222.839 577.68 Tm +(set) Tj +1 0 0 1 236.399 577.68 Tm +(of) Tj +1 0 0 1 247.079 577.68 Tm +(2) Tj +ET +endstream +endobj +495 0 obj +437 +endobj +496 0 obj +<< /Type /XObject /Name /R496 /Subtype /Image /Length 497 0 R +/ImageMask true /Width 50 /Height 92 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 50 /BlackIs1 true >>] +>> +stream +5$eL7-b:K3p0<%? +h1u#R]2'Mug].5is7$!eqtL-hqu?EZs5/V3iQXZVs7/ihqu?EZs4QRbJe-ans*o2/4DdB_@6aqq5Q~> +endstream +endobj +497 0 obj +96 +endobj +498 0 obj +<< /Length 499 0 R >> +stream +q +5 0 0 -9.2 252.5 584.8 cm +/R496 Do +Q +q +6.6 0 0 -6.5 260.7 583.5 cm +/R456 Do +Q +BT +1 0 0 1 269.879 577.68 Tm +(1) Tj +1 0 0 1 277.319 577.68 Tm +(v) Tj +1 0 0 1 282.119 577.68 Tm +(alid) Tj +1 0 0 1 72 565.68 Tm +(checkpoint) Tj +1 0 0 1 119.4 565.68 Tm +(messages) Tj +1 0 0 1 160.68 565.68 Tm +(pro) Tj +1 0 0 1 174 565.68 Tm +(ving) Tj +1 0 0 1 195.24 565.68 Tm +(the) Tj +1 0 0 1 211.08 565.68 Tm +(correctness) Tj +1 0 0 1 259.559 565.68 Tm +(of) Tj +ET +q +3.7 0 0 -4.6 271.8 570.2 cm +/R492 Do +Q +BT +1 0 0 1 276.119 565.68 Tm +(,) Tj +1 0 0 1 282.599 565.68 Tm +(and) Tj +ET +q +7.1 0 0 -7.5 72.1 560.7 cm +/R476 Do +Q +BT +1 0 0 1 83.16 553.8 Tm +(is) Tj +1 0 0 1 93.2399 553.8 Tm +(a) Tj +1 0 0 1 101.04 553.8 Tm +(set) Tj +1 0 0 1 115.44 553.8 Tm +(containing) Tj +1 0 0 1 160.92 553.8 Tm +(a) Tj +1 0 0 1 168.72 553.8 Tm +(set) Tj +ET +q +7.1 0 0 -7.5 183.3 560.7 cm +/R476 Do +Q +endstream +endobj +499 0 obj +841 +endobj +500 0 obj +<< /Type /XObject /Name /R500 /Subtype /Image /Length 501 0 R +/ImageMask true /Width 63 /Height 33 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 63 /BlackIs1 true >>] +>> +stream +36HD6'aC$jrgrOM2g''= +rP/?jh7L;trr;+h\8B"TnGiOCD85XVJ,"6UrZ/nkn+R@d'CFS6G.p]PWt2ocSubj)(omor5f3O$&-2~> +endstream +endobj +501 0 obj +102 +endobj +502 0 obj +<< /Length 503 0 R >> +stream +q +6.3 0 0 -3.3 190.5 555.6 cm +/R500 Do +Q +BT +1 0 0 1 201.12 553.8 Tm +(for) Tj +1 0 0 1 216.12 553.8 Tm +(each) Tj +1 0 0 1 237.72 553.8 Tm +(request) Tj +ET +endstream +endobj +503 0 obj +151 +endobj +504 0 obj +<< /Type /XObject /Name /R504 /Subtype /Image /Length 505 0 R +/ImageMask true /Width 81 /Height 46 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 81 /BlackIs1 true >>] +>> +stream +4N`CV%M3$8612U>:Ag"NnFZb +endstream +endobj +505 0 obj +139 +endobj +506 0 obj +<< /Length 507 0 R >> +stream +q +8.1 0 0 -4.6 270.1 558.3 cm +/R504 Do +Q +BT +1 0 0 1 282 553.8 Tm +(that) Tj +1 0 0 1 72 541.8 Tm +(prepared) Tj +1 0 0 1 108.6 541.8 Tm +(at) Tj +ET +q +2.6 0 0 -6.9 117.8 548.6 cm +/R448 Do +Q +BT +1 0 0 1 122.88 541.8 Tm +(with) Tj +1 0 0 1 142.32 541.8 Tm +(a) Tj +1 0 0 1 148.56 541.8 Tm +(sequence) Tj +1 0 0 1 186.84 541.8 Tm +(number) Tj +1 0 0 1 219.12 541.8 Tm +(higher) Tj +1 0 0 1 246.359 541.8 Tm +(than) Tj +ET +q +5.4 0 0 -4.6 265.4 546.3 cm +/R468 Do +Q +BT +1 0 0 1 271.199 541.8 Tm +(.) Tj +1 0 0 1 277.079 541.8 Tm +(Each) Tj +1 0 0 1 72 529.92 Tm +(set) Tj +ET +q +7.1 0 0 -7.5 85.2 536.8 cm +/R476 Do +Q +q +6.3 0 0 -3.3 92.3 531.7 cm +/R500 Do +Q +BT +1 0 0 1 101.64 529.92 Tm +(contains) Tj +1 0 0 1 136.92 529.92 Tm +(a) Tj +1 0 0 1 143.4 529.92 Tm +(v) Tj +1 0 0 1 148.2 529.92 Tm +(alid) Tj +1 0 0 1 165.24 529.92 Tm +(pre-prepare) Tj +1 0 0 1 213.12 529.92 Tm +(message) Tj +1 0 0 1 249 529.92 Tm +(\(without) Tj +1 0 0 1 284.879 529.92 Tm +(the) Tj +1 0 0 1 72 517.92 Tm +(corresponding) Tj +1 0 0 1 132.96 517.92 Tm +(client) Tj +1 0 0 1 159.12 517.92 Tm +(message\)) Tj +1 0 0 1 200.16 517.92 Tm +(and) Tj +1 0 0 1 218.52 517.92 Tm +(2) Tj +ET +q +5 0 0 -9.2 224 525 cm +/R496 Do +Q +BT +1 0 0 1 233.4 517.92 Tm +(matching,) Tj +1 0 0 1 277.319 517.92 Tm +(v) Tj +1 0 0 1 282.119 517.92 Tm +(alid) Tj +1 0 0 1 72 506.04 Tm +(prepare) Tj +1 0 0 1 105.48 506.04 Tm +(messages) Tj +1 0 0 1 146.64 506.04 Tm +(signed) Tj +1 0 0 1 176.16 506.04 Tm +(by) Tj +1 0 0 1 189.72 506.04 Tm +(dif) Tj +1 0 0 1 200.64 506.04 Tm +(ferent) Tj +1 0 0 1 227.279 506.04 Tm +(backups) Tj +1 0 0 1 263.519 506.04 Tm +(with) Tj +1 0 0 1 284.879 506.04 Tm +(the) Tj +1 0 0 1 72 494.04 Tm +(same) Tj +1 0 0 1 94.9199 494.04 Tm +(vie) Tj +1 0 0 1 106.92 494.04 Tm +(w) Tj +1 0 0 1 113.52 494.04 Tm +(,) Tj +1 0 0 1 118.44 494.04 Tm +(sequence) Tj +1 0 0 1 157.44 494.04 Tm +(number) Tj +1 0 0 1 187.8 494.04 Tm +(,) Tj +1 0 0 1 192.48 494.04 Tm +(and) Tj +1 0 0 1 209.28 494.04 Tm +(the) Tj +1 0 0 1 224.04 494.04 Tm +(digest) Tj +1 0 0 1 250.319 494.04 Tm +(of) Tj +ET +q +8.1 0 0 -4.6 261.3 498.5 cm +/R504 Do +Q +BT +1 0 0 1 269.879 494.04 Tm +(.) Tj +1 0 0 1 81.96 482.16 Tm +(When) Tj +1 0 0 1 108.96 482.16 Tm +(the) Tj +1 0 0 1 124.44 482.16 Tm +(primary) Tj +ET +endstream +endobj +507 0 obj +2136 +endobj +508 0 obj +<< /Type /XObject /Name /R508 /Subtype /Image /Length 509 0 R +/ImageMask true /Width 54 /Height 64 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 54 /BlackIs1 true >>] +>> +stream +-D98r4=0b"4s0Orro]!o&:3^Ke&7l2 +[/(7*rS^,RIfB4oLV,SG]>+(NrPSU5I/eEj^Aki-s4fFGn+Z5e95',@28).TTZ(Ru?\8Ii-iec'~> +endstream +endobj +509 0 obj +109 +endobj +510 0 obj +<< /Length 511 0 R >> +stream +q +5.4 0 0 -6.4 158.7 486.7 cm +/R508 Do +Q +BT +1 0 0 1 167.4 482.16 Tm +(of) Tj +1 0 0 1 178.92 482.16 Tm +(vie) Tj +1 0 0 1 190.92 482.16 Tm +(w) Tj +ET +q +4.4 0 0 -4.6 201.4 486.7 cm +/R452 Do +Q +q +6.6 0 0 -6.5 209.7 488 cm +/R456 Do +Q +BT +1 0 0 1 219.48 482.16 Tm +(1) Tj +1 0 0 1 227.64 482.16 Tm +(recei) Tj +1 0 0 1 246.84 482.16 Tm +(v) Tj +1 0 0 1 251.76 482.16 Tm +(es) Tj +1 0 0 1 263.159 482.16 Tm +(2) Tj +ET +q +5 0 0 -9.2 268.6 489.3 cm +/R496 Do +Q +BT +1 0 0 1 277.319 482.16 Tm +(v) Tj +1 0 0 1 282.119 482.16 Tm +(alid) Tj +1 0 0 1 72 470.16 Tm +(vie) Tj +1 0 0 1 84 470.16 Tm +(w-change) Tj +1 0 0 1 124.56 470.16 Tm +(messages) Tj +1 0 0 1 164.04 470.16 Tm +(for) Tj +1 0 0 1 177.48 470.16 Tm +(vie) Tj +1 0 0 1 189.48 470.16 Tm +(w) Tj +ET +q +4.4 0 0 -4.6 198.7 474.7 cm +/R452 Do +Q +q +6.6 0 0 -6.5 205 476 cm +/R456 Do +Q +BT +1 0 0 1 213.119 470.16 Tm +(1) Tj +1 0 0 1 219.959 470.16 Tm +(from) Tj +1 0 0 1 241.199 470.16 Tm +(other) Tj +1 0 0 1 263.519 470.16 Tm +(replicas,) Tj +1 0 0 1 72 458.28 Tm +(it) Tj +1 0 0 1 79.9199 458.28 Tm +(multicasts) Tj +1 0 0 1 122.64 458.28 Tm +(a) Tj +ET +q +2.4 0 0 -9.9 130.4 465.8 cm +/R460 Do +Q +BT +/R445 8 Tf +1 0 0 1 133.32 458.28 Tm +(NEW) Tj +/R445 10 Tf +1 0 0 1 151.08 458.28 Tm +(-) Tj +/R445 8 Tf +1 0 0 1 154.44 458.28 Tm +(VIEW) Tj +ET +q +1.3 0 0 -3 175.9 459.4 cm +/R464 Do +Q +q +4.4 0 0 -4.6 179.7 462.8 cm +/R452 Do +Q +q +6.6 0 0 -6.5 187.1 464.1 cm +/R456 Do +Q +BT +/R445 10 Tf +1 0 0 1 196.32 458.28 Tm +(1) Tj +ET +q +1.3 0 0 -3 202.2 459.4 cm +/R464 Do +Q +endstream +endobj +511 0 obj +1441 +endobj +512 0 obj +<< /Type /XObject /Name /R512 /Subtype /Image /Length 513 0 R +/ImageMask true /Width 63 /Height 74 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 63 /BlackIs1 true >>] +>> +stream +3nSfS4mq2+%std@nf[D1mtU]MLV%uL +HZ-Y0fR:M_^AL36_uFYahhUt*h-^,9hr!Z's1\N8deI/VX$K^`$prPKWY^jqg$5s +endstream +endobj +513 0 obj +113 +endobj +514 0 obj +<< /Length 515 0 R >> +stream +q +6.3 0 0 -7.4 206 465.2 cm +/R512 Do +Q +q +1.3 0 0 -3 213.6 459.4 cm +/R464 Do +Q +endstream +endobj +515 0 obj +78 +endobj +516 0 obj +<< /Type /XObject /Name /R516 /Subtype /Image /Length 517 0 R +/ImageMask true /Width 72 /Height 74 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 72 /BlackIs1 true >>] +>> +stream +.D\.1%HncC3X8G(Ssge8@F>AX*$@u(%Hp!FloO!m_Jf)Vl23boo`"`;p\l8Ps6eeEqfi&> +rP.2d@Gpp)^[++umqqs0GJjBg_puYOE5G\*g[Ns3DUu^A/VY)r9(0U#`%,\4qd@o5@6r1i?ig +endstream +endobj +517 0 obj +149 +endobj +518 0 obj +<< /Length 519 0 R >> +stream +q +7.2 0 0 -7.4 217.7 465.4 cm +/R516 Do +Q +q +2.4 0 0 -9.9 225.8 465.8 cm +/R480 Do +Q +q +4.2 0 0 -3.1 229.4 459.8 cm +/R484 Do +Q +endstream +endobj +519 0 obj +123 +endobj +520 0 obj +<< /Type /XObject /Name /R520 /Subtype /Image /Length 521 0 R +/ImageMask true /Width 31 /Height 32 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 31 /BlackIs1 true >>] +>> +stream +-D:A;E<"bo0Kp%mK +endstream +endobj +521 0 obj +58 +endobj +522 0 obj +<< /Length 523 0 R >> +stream +q +3.1 0 0 -3.2 233.9 458.1 cm +/R520 Do +Q +BT +1 0 0 1 240.84 458.28 Tm +(message) Tj +1 0 0 1 276.96 458.28 Tm +(to) Tj +1 0 0 1 287.04 458.28 Tm +(all) Tj +1 0 0 1 72 446.28 Tm +(other) Tj +1 0 0 1 94.5599 446.28 Tm +(replicas,) Tj +1 0 0 1 130.32 446.28 Tm +(where) Tj +ET +q +6.3 0 0 -7.4 156.9 453.2 cm +/R512 Do +Q +BT +1 0 0 1 165.72 446.28 Tm +(is) Tj +1 0 0 1 174.6 446.28 Tm +(a) Tj +1 0 0 1 181.08 446.28 Tm +(set) Tj +1 0 0 1 194.28 446.28 Tm +(containing) Tj +1 0 0 1 238.439 446.28 Tm +(the) Tj +1 0 0 1 252.719 446.28 Tm +(v) Tj +1 0 0 1 257.519 446.28 Tm +(alid) Tj +1 0 0 1 274.559 446.28 Tm +(vie) Tj +1 0 0 1 286.559 446.28 Tm +(w-) Tj +1 0 0 1 72 434.4 Tm +(change) Tj +1 0 0 1 102.84 434.4 Tm +(messages) Tj +1 0 0 1 143.04 434.4 Tm +(recei) Tj +1 0 0 1 162.24 434.4 Tm +(v) Tj +1 0 0 1 167.16 434.4 Tm +(ed) Tj +1 0 0 1 179.04 434.4 Tm +(by) Tj +1 0 0 1 191.64 434.4 Tm +(the) Tj +1 0 0 1 206.4 434.4 Tm +(primary) Tj +1 0 0 1 240.6 434.4 Tm +(plus) Tj +1 0 0 1 259.799 434.4 Tm +(the) Tj +1 0 0 1 274.559 434.4 Tm +(vie) Tj +1 0 0 1 286.559 434.4 Tm +(w-) Tj +1 0 0 1 72 422.4 Tm +(change) Tj +1 0 0 1 102.12 422.4 Tm +(message) Tj +1 0 0 1 137.76 422.4 Tm +(for) Tj +ET +q +4.4 0 0 -4.6 151.5 426.9 cm +/R452 Do +Q +q +6.6 0 0 -6.5 157.8 428.2 cm +/R456 Do +Q +BT +1 0 0 1 165.96 422.4 Tm +(1) Tj +1 0 0 1 172.92 422.4 Tm +(the) Tj +1 0 0 1 186.96 422.4 Tm +(primary) Tj +1 0 0 1 220.319 422.4 Tm +(sent) Tj +1 0 0 1 238.319 422.4 Tm +(\(or) Tj +1 0 0 1 251.879 422.4 Tm +(w) Tj +1 0 0 1 258.959 422.4 Tm +(ould) Tj +1 0 0 1 278.519 422.4 Tm +(ha) Tj +1 0 0 1 287.759 422.4 Tm +(v) Tj +1 0 0 1 292.679 422.4 Tm +(e) Tj +1 0 0 1 72 410.52 Tm +(sent\),) Tj +1 0 0 1 95.9999 410.52 Tm +(and) Tj +ET +q +7.2 0 0 -7.4 112.9 417.6 cm +/R516 Do +Q +BT +1 0 0 1 122.76 410.52 Tm +(is) Tj +1 0 0 1 131.4 410.52 Tm +(a) Tj +1 0 0 1 137.88 410.52 Tm +(set) Tj +1 0 0 1 150.96 410.52 Tm +(of) Tj +1 0 0 1 161.4 410.52 Tm +(pre-prepare) Tj +1 0 0 1 209.279 410.52 Tm +(messages) Tj +1 0 0 1 248.999 410.52 Tm +(\(without) Tj +1 0 0 1 284.879 410.52 Tm +(the) Tj +1 0 0 1 72 398.52 Tm +(piggyback) Tj +1 0 0 1 113.76 398.52 Tm +(ed) Tj +1 0 0 1 125.28 398.52 Tm +(request\).) Tj +ET +q +7.2 0 0 -7.4 163.9 405.6 cm +/R516 Do +Q +BT +1 0 0 1 174.12 398.52 Tm +(is) Tj +1 0 0 1 183.36 398.52 Tm +(computed) Tj +1 0 0 1 225.12 398.52 Tm +(as) Tj +1 0 0 1 235.92 398.52 Tm +(follo) Tj +1 0 0 1 254.639 398.52 Tm +(ws:) Tj +1 0 0 1 74.52 386.4 Tm +(1.) Tj +1 0 0 1 87 386.4 Tm +(The) Tj +1 0 0 1 105.48 386.4 Tm +(primary) Tj +1 0 0 1 140.04 386.4 Tm +(determines) Tj +1 0 0 1 186.72 386.4 Tm +(the) Tj +1 0 0 1 201.84 386.4 Tm +(sequence) Tj +1 0 0 1 241.439 386.4 Tm +(number) Tj +ET +endstream +endobj +523 0 obj +2530 +endobj +524 0 obj +<< /Type /Font /Name /R524 /Subtype /Type1 /BaseFont /Times-Italic >> +endobj +525 0 obj +<< /Length 526 0 R >> +stream +BT +/R524 10 Tf +1 0 0 1 274.919 386.4 Tm +(min-s) Tj +/R445 10 Tf +1 0 0 1 87 374.52 Tm +(of) Tj +1 0 0 1 98.7599 374.52 Tm +(the) Tj +1 0 0 1 114.48 374.52 Tm +(latest) Tj +1 0 0 1 138.96 374.52 Tm +(stable) Tj +1 0 0 1 165.72 374.52 Tm +(checkpoint) Tj +1 0 0 1 213 374.52 Tm +(in) Tj +ET +q +6.3 0 0 -7.4 224.4 381.4 cm +/R512 Do +Q +BT +1 0 0 1 234.599 374.52 Tm +(and) Tj +1 0 0 1 252.479 374.52 Tm +(the) Tj +1 0 0 1 268.199 374.52 Tm +(highest) Tj +1 0 0 1 87 362.52 Tm +(sequence) Tj +1 0 0 1 126 362.52 Tm +(number) Tj +/R524 10 Tf +1 0 0 1 158.88 362.52 Tm +(max-s) Tj +/R445 10 Tf +1 0 0 1 185.16 362.52 Tm +(in) Tj +1 0 0 1 195.36 362.52 Tm +(a) Tj +1 0 0 1 202.32 362.52 Tm +(prepare) Tj +1 0 0 1 234.72 362.52 Tm +(message) Tj +1 0 0 1 270.959 362.52 Tm +(in) Tj +ET +q +6.3 0 0 -7.4 281.4 369.4 cm +/R512 Do +Q +BT +1 0 0 1 288.119 362.52 Tm +(.) Tj +1 0 0 1 74.52 348.84 Tm +(2.) Tj +1 0 0 1 87 348.84 Tm +(The) Tj +1 0 0 1 105.72 348.84 Tm +(primary) Tj +1 0 0 1 140.64 348.84 Tm +(creates) Tj +1 0 0 1 171.6 348.84 Tm +(a) Tj +1 0 0 1 179.4 348.84 Tm +(ne) Tj +1 0 0 1 188.64 348.84 Tm +(w) Tj +1 0 0 1 199.08 348.84 Tm +(pre-prepare) Tj +1 0 0 1 248.279 348.84 Tm +(message) Tj +1 0 0 1 285.359 348.84 Tm +(for) Tj +1 0 0 1 87 336.84 Tm +(vie) Tj +1 0 0 1 99 336.84 Tm +(w) Tj +ET +q +4.4 0 0 -4.6 108 341.3 cm +/R452 Do +Q +q +6.6 0 0 -6.5 113.9 342.6 cm +/R456 Do +Q +BT +1 0 0 1 121.68 336.84 Tm +(1) Tj +1 0 0 1 128.52 336.84 Tm +(for) Tj +1 0 0 1 141.84 336.84 Tm +(each) Tj +1 0 0 1 161.76 336.84 Tm +(sequence) Tj +1 0 0 1 200.04 336.84 Tm +(number) Tj +ET +q +5.4 0 0 -4.6 232.4 341.3 cm +/R468 Do +Q +BT +1 0 0 1 239.879 336.84 Tm +(between) Tj +/R524 10 Tf +1 0 0 1 274.919 336.84 Tm +(min-s) Tj +/R445 10 Tf +1 0 0 1 87 324.96 Tm +(and) Tj +/R524 10 Tf +1 0 0 1 104.16 324.96 Tm +(max-s) Tj +/R445 10 Tf +1 0 0 1 127.92 324.96 Tm +(.) Tj +1 0 0 1 135.12 324.96 Tm +(There) Tj +1 0 0 1 161.16 324.96 Tm +(are) Tj +1 0 0 1 176.16 324.96 Tm +(tw) Tj +1 0 0 1 186 324.96 Tm +(o) Tj +1 0 0 1 193.92 324.96 Tm +(cases:) Tj +1 0 0 1 222 324.96 Tm +(\(1\)) Tj +1 0 0 1 236.4 324.96 Tm +(there) Tj +1 0 0 1 259.199 324.96 Tm +(is) Tj +1 0 0 1 268.679 324.96 Tm +(at) Tj +1 0 0 1 278.759 324.96 Tm +(least) Tj +1 0 0 1 87 312.96 Tm +(one) Tj +1 0 0 1 105.36 312.96 Tm +(set) Tj +1 0 0 1 120.6 312.96 Tm +(in) Tj +1 0 0 1 132.36 312.96 Tm +(the) Tj +ET +q +7.1 0 0 -7.5 148.8 319.9 cm +/R476 Do +Q +BT +1 0 0 1 160.44 312.96 Tm +(component) Tj +1 0 0 1 208.8 312.96 Tm +(of) Tj +1 0 0 1 221.28 312.96 Tm +(some) Tj +1 0 0 1 246.359 312.96 Tm +(vie) Tj +1 0 0 1 258.359 312.96 Tm +(w-change) Tj +1 0 0 1 87 301.08 Tm +(message) Tj +1 0 0 1 123.84 301.08 Tm +(in) Tj +ET +q +6.3 0 0 -7.4 135 308 cm +/R512 Do +Q +BT +1 0 0 1 144.84 301.08 Tm +(with) Tj +1 0 0 1 165.72 301.08 Tm +(sequence) Tj +1 0 0 1 205.44 301.08 Tm +(number) Tj +ET +q +5.4 0 0 -4.6 239.2 305.6 cm +/R468 Do +Q +BT +1 0 0 1 245.04 301.08 Tm +(,) Tj +1 0 0 1 250.799 301.08 Tm +(or) Tj +1 0 0 1 262.319 301.08 Tm +(\(2\)) Tj +1 0 0 1 277.079 301.08 Tm +(there) Tj +1 0 0 1 87 289.08 Tm +(is) Tj +1 0 0 1 96.8399 289.08 Tm +(no) Tj +1 0 0 1 109.92 289.08 Tm +(such) Tj +1 0 0 1 131.4 289.08 Tm +(set.) Tj +1 0 0 1 150.72 289.08 Tm +(In) Tj +1 0 0 1 162.12 289.08 Tm +(the) Tj +1 0 0 1 177.48 289.08 Tm +(\256rst) Tj +1 0 0 1 196.2 289.08 Tm +(case,) Tj +1 0 0 1 219.24 289.08 Tm +(the) Tj +1 0 0 1 234.6 289.08 Tm +(primary) Tj +1 0 0 1 269.399 289.08 Tm +(creates) Tj +1 0 0 1 87 277.2 Tm +(a) Tj +1 0 0 1 94.3199 277.2 Tm +(ne) Tj +1 0 0 1 103.56 277.2 Tm +(w) Tj +1 0 0 1 113.76 277.2 Tm +(message) Tj +ET +q +2.4 0 0 -9.9 151.4 284.7 cm +/R460 Do +Q +BT +/R445 8 Tf +1 0 0 1 154.32 277.2 Tm +(PRE) Tj +/R445 10 Tf +1 0 0 1 168.96 277.2 Tm +(-) Tj +/R445 8 Tf +1 0 0 1 172.32 277.2 Tm +(PREP) Tj +1 0 0 1 190.68 277.2 Tm +(ARE) Tj +ET +q +1.3 0 0 -3 207.3 278.3 cm +/R464 Do +Q +q +4.4 0 0 -4.6 211.2 281.7 cm +/R452 Do +Q +q +6.6 0 0 -6.5 219.1 283 cm +/R456 Do +Q +BT +/R445 10 Tf +1 0 0 1 228.839 277.2 Tm +(1) Tj +ET +q +1.3 0 0 -3 234.7 278.3 cm +/R464 Do +Q +q +5.4 0 0 -4.6 238.5 281.7 cm +/R468 Do +Q +q +1.3 0 0 -3 245.1 278.3 cm +/R464 Do +Q +endstream +endobj +526 0 obj +3873 +endobj +527 0 obj +<< /Type /XObject /Name /R527 /Subtype /Image /Length 528 0 R +/ImageMask true /Width 49 /Height 71 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 49 /BlackIs1 true >>] +>> +stream +-PYiBJV"+0)UN3G',i&VcojQ(@sYoeKmuDG +s)&q9s*o7jJ)C$Fpj^'QJ%rnUr&sfAs+VQL0)H)pV[&D-@^p.P-,mCns*qra5O\@FpQ6S*nl.";J-Z~> +endstream +endobj +528 0 obj +117 +endobj +529 0 obj +<< /Length 530 0 R >> +stream +q +4.9 0 0 -7.1 249.1 284.2 cm +/R527 Do +Q +q +2.4 0 0 -9.9 254.3 284.7 cm +/R480 Do +Q +q +4.2 0 0 -3.1 258 278.8 cm +/R484 Do +Q +q +3.1 0 0 -3.2 262.4 277 cm +/R520 Do +Q +BT +1 0 0 1 267 277.2 Tm +(,) Tj +1 0 0 1 272.64 277.2 Tm +(where) Tj +ET +q +4.9 0 0 -7.1 87.3 272.2 cm +/R527 Do +Q +BT +1 0 0 1 94.5599 265.2 Tm +(is) Tj +1 0 0 1 103.68 265.2 Tm +(the) Tj +1 0 0 1 118.32 265.2 Tm +(request) Tj +1 0 0 1 149.64 265.2 Tm +(digest) Tj +1 0 0 1 175.92 265.2 Tm +(in) Tj +1 0 0 1 186.12 265.2 Tm +(the) Tj +1 0 0 1 200.759 265.2 Tm +(pre-prepare) Tj +1 0 0 1 249.119 265.2 Tm +(message) Tj +1 0 0 1 285.359 265.2 Tm +(for) Tj +1 0 0 1 87 253.32 Tm +(sequence) Tj +1 0 0 1 127.56 253.32 Tm +(number) Tj +ET +q +5.4 0 0 -4.6 162.3 257.8 cm +/R468 Do +Q +BT +1 0 0 1 172.2 253.32 Tm +(with) Tj +1 0 0 1 194.16 253.32 Tm +(the) Tj +1 0 0 1 210.36 253.32 Tm +(highest) Tj +1 0 0 1 243.359 253.32 Tm +(vie) Tj +1 0 0 1 255.359 253.32 Tm +(w) Tj +1 0 0 1 266.519 253.32 Tm +(number) Tj +1 0 0 1 87 241.32 Tm +(in) Tj +ET +q +6.3 0 0 -7.4 99.8 248.2 cm +/R512 Do +Q +BT +1 0 0 1 106.56 241.32 Tm +(.) Tj +1 0 0 1 119.64 241.32 Tm +(In) Tj +1 0 0 1 132.84 241.32 Tm +(the) Tj +1 0 0 1 149.88 241.32 Tm +(second) Tj +1 0 0 1 182.4 241.32 Tm +(case,) Tj +1 0 0 1 207.48 241.32 Tm +(it) Tj +1 0 0 1 217.92 241.32 Tm +(creates) Tj +1 0 0 1 250.439 241.32 Tm +(a) Tj +1 0 0 1 259.799 241.32 Tm +(ne) Tj +1 0 0 1 269.039 241.32 Tm +(w) Tj +1 0 0 1 280.919 241.32 Tm +(pre-) Tj +1 0 0 1 87 229.44 Tm +(prepare) Tj +1 0 0 1 121.8 229.44 Tm +(message) Tj +ET +q +2.4 0 0 -9.9 161.3 236.9 cm +/R460 Do +Q +BT +/R445 8 Tf +1 0 0 1 164.28 229.44 Tm +(PRE) Tj +/R445 10 Tf +1 0 0 1 178.92 229.44 Tm +(-) Tj +/R445 8 Tf +1 0 0 1 182.28 229.44 Tm +(PREP) Tj +1 0 0 1 200.64 229.44 Tm +(ARE) Tj +ET +q +1.3 0 0 -3 217.4 230.5 cm +/R464 Do +Q +q +4.4 0 0 -4.6 221.2 233.9 cm +/R452 Do +Q +q +6.6 0 0 -6.5 230.3 235.2 cm +/R456 Do +Q +BT +/R445 10 Tf +1 0 0 1 241.08 229.44 Tm +(1) Tj +ET +q +1.3 0 0 -3 246.9 230.5 cm +/R464 Do +Q +q +5.4 0 0 -4.6 250.8 233.9 cm +/R468 Do +Q +q +1.3 0 0 -3 257.4 230.5 cm +/R464 Do +Q +q +4.9 0 0 -7.1 261.3 236.4 cm +/R527 Do +Q +endstream +endobj +530 0 obj +1995 +endobj +531 0 obj +<< /Type /XObject /Name /R531 /Subtype /Image /Length 532 0 R +/ImageMask true /Width 43 /Height 33 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 43 /BlackIs1 true >>] +>> +stream +366ZtBKi$Kj)nf1pi$0"h;="Vq3Z%=rV-?Z\+qmGnGV5SiW&fRKsuPqJ89Nq=JdRb$l@h'!$D~> +endstream +endobj +532 0 obj +76 +endobj +533 0 obj +<< /Length 534 0 R >> +stream +q +4.3 0 0 -3.3 266.5 236.2 cm +/R531 Do +Q +endstream +endobj +534 0 obj +41 +endobj +535 0 obj +<< /Type /XObject /Name /R535 /Subtype /Image /Length 536 0 R +/ImageMask true /Width 41 /Height 33 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 41 /BlackIs1 true >>] +>> +stream +,RS3]JV5%#k`6^IoGFXX]IX$\_o'7hq4MUEDu\^EYl?&J&+AtsIJ\HmJ,?^gl`g^Q%1dsp!.Y~> +endstream +endobj +536 0 obj +76 +endobj +537 0 obj +<< /Length 538 0 R >> +stream +q +4.1 0 0 -3.3 271.4 236.2 cm +/R535 Do +Q +endstream +endobj +538 0 obj +41 +endobj +539 0 obj +<< /Type /XObject /Name /R539 /Subtype /Image /Length 540 0 R +/ImageMask true /Width 18 /Height 49 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 18 /BlackIs1 true >>] +>> +stream +3,jM'nkeZHp\iq*s4I>^^[L[l^Yee_rr7Jcn,(B6_t&A/#QTA~> +endstream +endobj +540 0 obj +52 +endobj +541 0 obj +<< /Length 542 0 R >> +stream +q +1.8 0 0 -4.9 276 237.8 cm +/R539 Do +Q +q +1.8 0 0 -4.9 278.5 237.8 cm +/R539 Do +Q +q +2.4 0 0 -9.9 281.7 236.9 cm +/R480 Do +Q +q +4.2 0 0 -3.1 285.3 231 cm +/R484 Do +Q +q +3.1 0 0 -3.2 289.8 229.2 cm +/R520 Do +Q +BT +1 0 0 1 294.48 229.44 Tm +(,) Tj +1 0 0 1 87 217.44 Tm +(where) Tj +ET +q +4.9 0 0 -7.1 115.5 224.4 cm +/R527 Do +Q +q +4.3 0 0 -3.3 120.7 224.2 cm +/R531 Do +Q +q +4.1 0 0 -3.3 125.6 224.2 cm +/R535 Do +Q +q +1.8 0 0 -4.9 130.2 225.8 cm +/R539 Do +Q +q +1.8 0 0 -4.9 132.7 225.8 cm +/R539 Do +Q +BT +1 0 0 1 139.56 217.44 Tm +(is) Tj +1 0 0 1 150.12 217.44 Tm +(the) Tj +1 0 0 1 166.2 217.44 Tm +(digest) Tj +1 0 0 1 193.92 217.44 Tm +(of) Tj +1 0 0 1 206.16 217.44 Tm +(a) Tj +1 0 0 1 214.44 217.44 Tm +(special) Tj +/R524 10 Tf +1 0 0 1 246 217.44 Tm +(null) Tj +/R445 10 Tf +1 0 0 1 265.44 217.44 Tm +(request;) Tj +1 0 0 1 87 205.56 Tm +(a) Tj +1 0 0 1 95.28 205.56 Tm +(null) Tj +1 0 0 1 114.72 205.56 Tm +(request) Tj +1 0 0 1 147.36 205.56 Tm +(goes) Tj +1 0 0 1 169.56 205.56 Tm +(through) Tj +1 0 0 1 204.48 205.56 Tm +(the) Tj +1 0 0 1 220.68 205.56 Tm +(protocol) Tj +1 0 0 1 257.759 205.56 Tm +(lik) Tj +1 0 0 1 268.199 205.56 Tm +(e) Tj +1 0 0 1 276.479 205.56 Tm +(other) Tj +1 0 0 1 87 193.56 Tm +(requests,) Tj +1 0 0 1 126.12 193.56 Tm +(b) Tj +1 0 0 1 130.92 193.56 Tm +(ut) Tj +1 0 0 1 142.44 193.56 Tm +(its) Tj +1 0 0 1 155.64 193.56 Tm +(e) Tj +1 0 0 1 159.96 193.56 Tm +(x) Tj +1 0 0 1 164.88 193.56 Tm +(ecution) Tj +1 0 0 1 197.88 193.56 Tm +(is) Tj +1 0 0 1 208.32 193.56 Tm +(a) Tj +1 0 0 1 216.48 193.56 Tm +(no-op.) Tj +1 0 0 1 249.719 193.56 Tm +(\(P) Tj +1 0 0 1 258.479 193.56 Tm +(axos) Tj +1 0 0 1 280.439 193.56 Tm +([18) Tj +1 0 0 1 293.639 193.56 Tm +(]) Tj +1 0 0 1 87 181.68 Tm +(used) Tj +1 0 0 1 107.64 181.68 Tm +(a) Tj +1 0 0 1 114.6 181.68 Tm +(similar) Tj +1 0 0 1 144.72 181.68 Tm +(technique) Tj +1 0 0 1 186 181.68 Tm +(to) Tj +1 0 0 1 196.2 181.68 Tm +(\256ll) Tj +1 0 0 1 209.76 181.68 Tm +(in) Tj +1 0 0 1 220.08 181.68 Tm +(gaps.\)) Tj +1 0 0 1 81.96 166.56 Tm +(Ne) Tj +1 0 0 1 93.48 166.56 Tm +(xt) Tj +1 0 0 1 105.24 166.56 Tm +(the) Tj +1 0 0 1 121.44 166.56 Tm +(primary) Tj +1 0 0 1 157.08 166.56 Tm +(appends) Tj +1 0 0 1 193.8 166.56 Tm +(the) Tj +1 0 0 1 210 166.56 Tm +(messages) Tj +1 0 0 1 251.64 166.56 Tm +(in) Tj +ET +q +7.2 0 0 -7.4 264 173.7 cm +/R516 Do +Q +BT +1 0 0 1 275.759 166.56 Tm +(to) Tj +1 0 0 1 287.639 166.56 Tm +(its) Tj +1 0 0 1 72 154.56 Tm +(log.) Tj +1 0 0 1 92.5199 154.56 Tm +(If) Tj +/R524 10 Tf +1 0 0 1 102.12 154.56 Tm +(min-s) Tj +/R445 10 Tf +1 0 0 1 127.32 154.56 Tm +(is) Tj +1 0 0 1 137.04 154.56 Tm +(greater) Tj +1 0 0 1 167.76 154.56 Tm +(than) Tj +1 0 0 1 187.919 154.56 Tm +(the) Tj +1 0 0 1 203.159 154.56 Tm +(sequence) Tj +1 0 0 1 242.759 154.56 Tm +(number) Tj +1 0 0 1 276.239 154.56 Tm +(of) Tj +1 0 0 1 287.639 154.56 Tm +(its) Tj +1 0 0 1 72 142.68 Tm +(latest) Tj +1 0 0 1 94.9199 142.68 Tm +(stable) Tj +1 0 0 1 120 142.68 Tm +(checkpoint,) Tj +1 0 0 1 168.24 142.68 Tm +(the) Tj +1 0 0 1 182.16 142.68 Tm +(primary) Tj +1 0 0 1 215.64 142.68 Tm +(also) Tj +1 0 0 1 233.519 142.68 Tm +(inserts) Tj +1 0 0 1 261.359 142.68 Tm +(the) Tj +1 0 0 1 275.399 142.68 Tm +(proof) Tj +1 0 0 1 72 130.68 Tm +(of) Tj +1 0 0 1 84.96 130.68 Tm +(stability) Tj +1 0 0 1 121.68 130.68 Tm +(for) Tj +1 0 0 1 137.88 130.68 Tm +(the) Tj +1 0 0 1 154.68 130.68 Tm +(checkpoint) Tj +1 0 0 1 203.04 130.68 Tm +(with) Tj +1 0 0 1 225.36 130.68 Tm +(sequence) Tj +1 0 0 1 266.519 130.68 Tm +(number) Tj +/R524 10 Tf +1 0 0 1 72 118.8 Tm +(min-s) Tj +/R445 10 Tf +1 0 0 1 97.6799 118.8 Tm +(in) Tj +1 0 0 1 108.96 118.8 Tm +(its) Tj +1 0 0 1 121.92 118.8 Tm +(log,) Tj +1 0 0 1 141 118.8 Tm +(and) Tj +1 0 0 1 158.88 118.8 Tm +(discards) Tj +1 0 0 1 195.12 118.8 Tm +(information) Tj +1 0 0 1 245.639 118.8 Tm +(from) Tj +1 0 0 1 268.559 118.8 Tm +(the) Tj +1 0 0 1 284.279 118.8 Tm +(log) Tj +1 0 0 1 72 106.8 Tm +(as) Tj +1 0 0 1 82.5599 106.8 Tm +(discussed) Tj +1 0 0 1 123 106.8 Tm +(in) Tj +1 0 0 1 133.08 106.8 Tm +(Section) Tj +1 0 0 1 165.24 106.8 Tm +(4.3.) Tj +1 0 0 1 183.72 106.8 Tm +(Then) Tj +1 0 0 1 206.4 106.8 Tm +(it) Tj +/R524 10 Tf +1 0 0 1 214.199 106.8 Tm +(enter) Tj +1 0 0 1 234.599 106.8 Tm +(s) Tj +/R445 10 Tf +1 0 0 1 240.719 106.8 Tm +(vie) Tj +1 0 0 1 252.719 106.8 Tm +(w) Tj +ET +q +4.4 0 0 -4.6 262.4 111.3 cm +/R452 Do +Q +q +6.6 0 0 -6.5 269.5 112.6 cm +/R456 Do +Q +BT +1 0 0 1 278.519 106.8 Tm +(1:) Tj +1 0 0 1 289.799 106.8 Tm +(at) Tj +1 0 0 1 72 94.92 Tm +(this) Tj +1 0 0 1 88.9199 94.92 Tm +(point) Tj +1 0 0 1 111.84 94.92 Tm +(it) Tj +1 0 0 1 119.88 94.92 Tm +(is) Tj +1 0 0 1 129 94.92 Tm +(able) Tj +1 0 0 1 148.08 94.92 Tm +(to) Tj +1 0 0 1 158.4 94.92 Tm +(accept) Tj +1 0 0 1 186.24 94.92 Tm +(messages) Tj +1 0 0 1 226.44 94.92 Tm +(for) Tj +1 0 0 1 240.479 94.92 Tm +(vie) Tj +1 0 0 1 252.479 94.92 Tm +(w) Tj +ET +q +4.4 0 0 -4.6 262.3 99.4 cm +/R452 Do +Q +q +6.6 0 0 -6.5 270 100.7 cm +/R456 Do +Q +BT +1 0 0 1 279.479 94.92 Tm +(1.) Tj +1 0 0 1 81.96 82.92 Tm +(A) Tj +1 0 0 1 91.68 82.92 Tm +(backup) Tj +1 0 0 1 122.88 82.92 Tm +(accepts) Tj +1 0 0 1 154.68 82.92 Tm +(a) Tj +1 0 0 1 161.52 82.92 Tm +(ne) Tj +1 0 0 1 170.76 82.92 Tm +(w-vie) Tj +1 0 0 1 193.32 82.92 Tm +(w) Tj +1 0 0 1 202.8 82.92 Tm +(message) Tj +1 0 0 1 239.04 82.92 Tm +(for) Tj +1 0 0 1 253.199 82.92 Tm +(vie) Tj +1 0 0 1 265.199 82.92 Tm +(w) Tj +ET +q +4.4 0 0 -4.6 274.9 87.4 cm +/R452 Do +Q +q +6.6 0 0 -6.5 282.6 88.7 cm +/R456 Do +Q +BT +1 0 0 1 291.959 82.92 Tm +(1) Tj +1 0 0 1 72 71.04 Tm +(if) Tj +1 0 0 1 81.7199 71.04 Tm +(it) Tj +1 0 0 1 90.9599 71.04 Tm +(is) Tj +1 0 0 1 101.16 71.04 Tm +(signed) Tj +1 0 0 1 130.8 71.04 Tm +(properly) Tj +1 0 0 1 164.28 71.04 Tm +(,) Tj +1 0 0 1 170.4 71.04 Tm +(if) Tj +1 0 0 1 180.12 71.04 Tm +(the) Tj +1 0 0 1 195.959 71.04 Tm +(vie) Tj +1 0 0 1 207.959 71.04 Tm +(w-change) Tj +1 0 0 1 250.199 71.04 Tm +(messages) Tj +1 0 0 1 291.479 71.04 Tm +(it) Tj +1 0 0 1 315 709.08 Tm +(contains) Tj +1 0 0 1 352.2 709.08 Tm +(are) Tj +1 0 0 1 368.28 709.08 Tm +(v) Tj +1 0 0 1 373.08 709.08 Tm +(alid) Tj +1 0 0 1 392.04 709.08 Tm +(for) Tj +1 0 0 1 407.64 709.08 Tm +(vie) Tj +1 0 0 1 419.64 709.08 Tm +(w) Tj +ET +q +4.4 0 0 -4.6 430.9 713.6 cm +/R452 Do +Q +q +6.6 0 0 -6.5 439.5 714.9 cm +/R456 Do +Q +BT +1 0 0 1 449.76 709.08 Tm +(1,) Tj +1 0 0 1 461.519 709.08 Tm +(and) Tj +1 0 0 1 479.879 709.08 Tm +(if) Tj +1 0 0 1 489.959 709.08 Tm +(the) Tj +1 0 0 1 506.159 709.08 Tm +(set) Tj +ET +q +7.2 0 0 -7.4 521.7 716.2 cm +/R516 Do +Q +BT +1 0 0 1 533.399 709.08 Tm +(is) Tj +1 0 0 1 315 697.08 Tm +(correct;) Tj +1 0 0 1 348.72 697.08 Tm +(it) Tj +1 0 0 1 357.24 697.08 Tm +(v) Tj +1 0 0 1 362.16 697.08 Tm +(eri\256es) Tj +1 0 0 1 389.4 697.08 Tm +(the) Tj +1 0 0 1 404.52 697.08 Tm +(correctness) Tj +1 0 0 1 452.279 697.08 Tm +(of) Tj +ET +q +7.2 0 0 -7.4 464.1 704.2 cm +/R516 Do +Q +BT +1 0 0 1 474.839 697.08 Tm +(by) Tj +1 0 0 1 487.799 697.08 Tm +(performing) Tj +1 0 0 1 535.559 697.08 Tm +(a) Tj +1 0 0 1 315 685.2 Tm +(computation) Tj +1 0 0 1 368.52 685.2 Tm +(similar) Tj +1 0 0 1 399.96 685.2 Tm +(to) Tj +1 0 0 1 411.36 685.2 Tm +(the) Tj +1 0 0 1 427.32 685.2 Tm +(one) Tj +1 0 0 1 445.44 685.2 Tm +(used) Tj +1 0 0 1 467.4 685.2 Tm +(by) Tj +1 0 0 1 481.079 685.2 Tm +(the) Tj +1 0 0 1 497.039 685.2 Tm +(primary) Tj +1 0 0 1 532.199 685.2 Tm +(to) Tj +1 0 0 1 315 673.2 Tm +(create) Tj +ET +q +7.2 0 0 -7.4 342 680.3 cm +/R516 Do +Q +BT +1 0 0 1 349.68 673.2 Tm +(.) Tj +1 0 0 1 356.76 673.2 Tm +(Then) Tj +1 0 0 1 380.04 673.2 Tm +(it) Tj +1 0 0 1 388.44 673.2 Tm +(adds) Tj +1 0 0 1 409.44 673.2 Tm +(the) Tj +1 0 0 1 424.44 673.2 Tm +(ne) Tj +1 0 0 1 433.68 673.2 Tm +(w) Tj +1 0 0 1 443.52 673.2 Tm +(information) Tj +1 0 0 1 493.439 673.2 Tm +(to) Tj +1 0 0 1 503.999 673.2 Tm +(its) Tj +1 0 0 1 516.119 673.2 Tm +(log) Tj +1 0 0 1 531.719 673.2 Tm +(as) Tj +1 0 0 1 315 661.32 Tm +(described) Tj +1 0 0 1 356.28 661.32 Tm +(for) Tj +1 0 0 1 371.04 661.32 Tm +(the) Tj +1 0 0 1 386.28 661.32 Tm +(primary) Tj +1 0 0 1 417.48 661.32 Tm +(,) Tj +1 0 0 1 423 661.32 Tm +(multicasts) Tj +1 0 0 1 466.439 661.32 Tm +(a) Tj +1 0 0 1 473.999 661.32 Tm +(prepare) Tj +1 0 0 1 506.999 661.32 Tm +(for) Tj +1 0 0 1 521.759 661.32 Tm +(each) Tj +1 0 0 1 315 649.32 Tm +(message) Tj +1 0 0 1 350.64 649.32 Tm +(in) Tj +ET +q +7.2 0 0 -7.4 360.7 656.4 cm +/R516 Do +Q +BT +1 0 0 1 370.32 649.32 Tm +(to) Tj +1 0 0 1 380.04 649.32 Tm +(all) Tj +1 0 0 1 391.8 649.32 Tm +(the) Tj +1 0 0 1 405.96 649.32 Tm +(other) Tj +1 0 0 1 428.28 649.32 Tm +(replicas,) Tj +1 0 0 1 463.799 649.32 Tm +(adds) Tj +1 0 0 1 483.839 649.32 Tm +(these) Tj +1 0 0 1 506.279 649.32 Tm +(prepares) Tj +1 0 0 1 315 637.44 Tm +(to) Tj +1 0 0 1 325.2 637.44 Tm +(its) Tj +1 0 0 1 337.2 637.44 Tm +(log,) Tj +1 0 0 1 354.84 637.44 Tm +(and) Tj +1 0 0 1 371.76 637.44 Tm +(enters) Tj +1 0 0 1 398.04 637.44 Tm +(vie) Tj +1 0 0 1 410.04 637.44 Tm +(w) Tj +ET +q +4.4 0 0 -4.6 419.8 641.9 cm +/R452 Do +Q +q +6.6 0 0 -6.5 427.6 643.2 cm +/R456 Do +Q +BT +1 0 0 1 437.04 637.44 Tm +(1.) Tj +1 0 0 1 324.96 625.08 Tm +(Thereafter) Tj +1 0 0 1 366.36 625.08 Tm +(,) Tj +1 0 0 1 375.36 625.08 Tm +(the) Tj +1 0 0 1 393.48 625.08 Tm +(protocol) Tj +1 0 0 1 432.6 625.08 Tm +(proceeds) Tj +1 0 0 1 473.879 625.08 Tm +(as) Tj +1 0 0 1 488.159 625.08 Tm +(described) Tj +1 0 0 1 532.199 625.08 Tm +(in) Tj +1 0 0 1 315 613.2 Tm +(Section) Tj +1 0 0 1 349.08 613.2 Tm +(4.2.) Tj +1 0 0 1 372.6 613.2 Tm +(Replicas) Tj +1 0 0 1 411 613.2 Tm +(redo) Tj +1 0 0 1 432.96 613.2 Tm +(the) Tj +1 0 0 1 449.28 613.2 Tm +(protocol) Tj +1 0 0 1 486.599 613.2 Tm +(for) Tj +1 0 0 1 502.319 613.2 Tm +(messages) Tj +1 0 0 1 315 601.2 Tm +(between) Tj +/R524 10 Tf +1 0 0 1 352.56 601.2 Tm +(min-s) Tj +/R445 10 Tf +1 0 0 1 379.08 601.2 Tm +(and) Tj +/R524 10 Tf +1 0 0 1 397.92 601.2 Tm +(max-s) Tj +/R445 10 Tf +1 0 0 1 426.12 601.2 Tm +(b) Tj +1 0 0 1 430.92 601.2 Tm +(ut) Tj +1 0 0 1 443.04 601.2 Tm +(the) Tj +1 0 0 1 455.16 601.2 Tm +(y) Tj +1 0 0 1 464.4 601.2 Tm +(a) Tj +1 0 0 1 468.6 601.2 Tm +(v) Tj +1 0 0 1 473.4 601.2 Tm +(oid) Tj +1 0 0 1 490.56 601.2 Tm +(re-e) Tj +1 0 0 1 506.039 601.2 Tm +(x) Tj +1 0 0 1 510.959 601.2 Tm +(ecuting) Tj +1 0 0 1 315 589.32 Tm +(client) Tj +1 0 0 1 340.56 589.32 Tm +(requests) Tj +1 0 0 1 376.56 589.32 Tm +(\(by) Tj +1 0 0 1 393.12 589.32 Tm +(using) Tj +1 0 0 1 418.08 589.32 Tm +(their) Tj +1 0 0 1 439.8 589.32 Tm +(stored) Tj +1 0 0 1 467.399 589.32 Tm +(information) Tj +1 0 0 1 517.919 589.32 Tm +(about) Tj +1 0 0 1 315 577.32 Tm +(the) Tj +1 0 0 1 329.64 577.32 Tm +(last) Tj +1 0 0 1 345.96 577.32 Tm +(reply) Tj +1 0 0 1 369 577.32 Tm +(sent) Tj +1 0 0 1 387.48 577.32 Tm +(to) Tj +1 0 0 1 397.68 577.32 Tm +(each) Tj +1 0 0 1 418.44 577.32 Tm +(client\).) Tj +1 0 0 1 324.96 565.08 Tm +(A) Tj +1 0 0 1 336 565.08 Tm +(replica) Tj +1 0 0 1 366.96 565.08 Tm +(may) Tj +1 0 0 1 387.96 565.08 Tm +(be) Tj +1 0 0 1 401.28 565.08 Tm +(missing) Tj +1 0 0 1 436.08 565.08 Tm +(some) Tj +1 0 0 1 461.04 565.08 Tm +(request) Tj +1 0 0 1 493.679 565.08 Tm +(message) Tj +ET +q +8.1 0 0 -4.6 531.4 569.6 cm +/R504 Do +Q +BT +1 0 0 1 315 553.08 Tm +(or) Tj +1 0 0 1 326.52 553.08 Tm +(a) Tj +1 0 0 1 334.32 553.08 Tm +(stable) Tj +1 0 0 1 360.72 553.08 Tm +(checkpoint) Tj +1 0 0 1 407.76 553.08 Tm +(\(since) Tj +1 0 0 1 434.76 553.08 Tm +(these) Tj +1 0 0 1 458.519 553.08 Tm +(are) Tj +1 0 0 1 473.999 553.08 Tm +(not) Tj +1 0 0 1 489.959 553.08 Tm +(sent) Tj +1 0 0 1 509.279 553.08 Tm +(in) Tj +1 0 0 1 520.319 553.08 Tm +(ne) Tj +1 0 0 1 529.559 553.08 Tm +(w-) Tj +1 0 0 1 315 541.2 Tm +(vie) Tj +1 0 0 1 327 541.2 Tm +(w) Tj +1 0 0 1 336.84 541.2 Tm +(messages.\)) Tj +1 0 0 1 384.36 541.2 Tm +(It) Tj +1 0 0 1 393.12 541.2 Tm +(can) Tj +1 0 0 1 409.68 541.2 Tm +(obtain) Tj +1 0 0 1 437.28 541.2 Tm +(missing) Tj +1 0 0 1 470.88 541.2 Tm +(information) Tj +1 0 0 1 520.679 541.2 Tm +(from) Tj +1 0 0 1 315 529.2 Tm +(another) Tj +1 0 0 1 349.44 529.2 Tm +(replica.) Tj +1 0 0 1 388.8 529.2 Tm +(F) Tj +1 0 0 1 394.2 529.2 Tm +(or) Tj +1 0 0 1 407.04 529.2 Tm +(e) Tj +1 0 0 1 411.36 529.2 Tm +(xample,) Tj +1 0 0 1 448.08 529.2 Tm +(replica) Tj +ET +q +2.6 0 0 -6.9 480 536 cm +/R448 Do +Q +BT +1 0 0 1 487.799 529.2 Tm +(can) Tj +1 0 0 1 506.159 529.2 Tm +(obtain) Tj +1 0 0 1 535.559 529.2 Tm +(a) Tj +1 0 0 1 315 517.32 Tm +(missing) Tj +1 0 0 1 351.24 517.32 Tm +(checkpoint) Tj +1 0 0 1 400.2 517.32 Tm +(state) Tj +ET +q +3.7 0 0 -4.6 424.2 521.8 cm +/R492 Do +Q +BT +1 0 0 1 433.68 517.32 Tm +(from) Tj +1 0 0 1 458.4 517.32 Tm +(one) Tj +1 0 0 1 477.96 517.32 Tm +(of) Tj +1 0 0 1 491.519 517.32 Tm +(the) Tj +1 0 0 1 509.039 517.32 Tm +(replicas) Tj +1 0 0 1 315 505.32 Tm +(whose) Tj +1 0 0 1 344.4 505.32 Tm +(checkpoint) Tj +1 0 0 1 392.04 505.32 Tm +(messages) Tj +1 0 0 1 433.56 505.32 Tm +(certi\256ed) Tj +1 0 0 1 470.16 505.32 Tm +(its) Tj +1 0 0 1 483.479 505.32 Tm +(correctness) Tj +1 0 0 1 532.199 505.32 Tm +(in) Tj +ET +q +6.3 0 0 -7.4 315.2 500.3 cm +/R512 Do +Q +BT +1 0 0 1 321.96 493.44 Tm +(.) Tj +1 0 0 1 327.96 493.44 Tm +(Since) Tj +ET +q +5 0 0 -9.2 352.6 500.5 cm +/R496 Do +Q +q +6.6 0 0 -6.5 360.1 499.2 cm +/R456 Do +Q +BT +1 0 0 1 369 493.44 Tm +(1) Tj +1 0 0 1 376.08 493.44 Tm +(of) Tj +1 0 0 1 386.52 493.44 Tm +(those) Tj +1 0 0 1 409.8 493.44 Tm +(replicas) Tj +1 0 0 1 442.92 493.44 Tm +(are) Tj +1 0 0 1 457.199 493.44 Tm +(correct,) Tj +1 0 0 1 489.599 493.44 Tm +(replica) Tj +ET +q +2.6 0 0 -6.9 519.1 500.2 cm +/R448 Do +Q +BT +1 0 0 1 524.519 493.44 Tm +(will) Tj +1 0 0 1 315 481.44 Tm +(al) Tj +1 0 0 1 322.08 481.44 Tm +(w) Tj +1 0 0 1 329.16 481.44 Tm +(ays) Tj +1 0 0 1 344.88 481.44 Tm +(obtain) Tj +ET +q +3.7 0 0 -4.6 372.5 485.9 cm +/R492 Do +Q +BT +1 0 0 1 379.08 481.44 Tm +(or) Tj +1 0 0 1 389.76 481.44 Tm +(a) Tj +1 0 0 1 396.6 481.44 Tm +(later) Tj +1 0 0 1 416.64 481.44 Tm +(certi\256ed) Tj +1 0 0 1 451.559 481.44 Tm +(stable) Tj +1 0 0 1 477.239 481.44 Tm +(checkpoint.) Tj +1 0 0 1 526.919 481.44 Tm +(W) Tj +1 0 0 1 535.439 481.44 Tm +(e) Tj +1 0 0 1 315 469.56 Tm +(can) Tj +1 0 0 1 332.16 469.56 Tm +(a) Tj +1 0 0 1 336.36 469.56 Tm +(v) Tj +1 0 0 1 341.16 469.56 Tm +(oid) Tj +1 0 0 1 357.24 469.56 Tm +(sending) Tj +1 0 0 1 391.56 469.56 Tm +(the) Tj +1 0 0 1 407.16 469.56 Tm +(entire) Tj +1 0 0 1 433.08 469.56 Tm +(checkpoint) Tj +1 0 0 1 480.24 469.56 Tm +(by) Tj +1 0 0 1 493.559 469.56 Tm +(partitioning) Tj +1 0 0 1 315 457.56 Tm +(the) Tj +1 0 0 1 330.36 457.56 Tm +(state) Tj +1 0 0 1 351.72 457.56 Tm +(and) Tj +1 0 0 1 369.24 457.56 Tm +(stamping) Tj +1 0 0 1 408.96 457.56 Tm +(each) Tj +1 0 0 1 430.32 457.56 Tm +(partition) Tj +1 0 0 1 467.279 457.56 Tm +(with) Tj +1 0 0 1 488.159 457.56 Tm +(the) Tj +1 0 0 1 503.519 457.56 Tm +(sequence) Tj +1 0 0 1 315 445.68 Tm +(number) Tj +1 0 0 1 349.56 445.68 Tm +(of) Tj +1 0 0 1 362.04 445.68 Tm +(the) Tj +1 0 0 1 378.36 445.68 Tm +(last) Tj +1 0 0 1 396.36 445.68 Tm +(request) Tj +1 0 0 1 429.24 445.68 Tm +(that) Tj +1 0 0 1 448.32 445.68 Tm +(modi\256ed) Tj +1 0 0 1 487.919 445.68 Tm +(it.) Tj +1 0 0 1 504.599 445.68 Tm +(T) Tj +1 0 0 1 509.879 445.68 Tm +(o) Tj +1 0 0 1 518.999 445.68 Tm +(bring) Tj +1 0 0 1 315 433.68 Tm +(a) Tj +1 0 0 1 323.04 433.68 Tm +(replica) Tj +1 0 0 1 353.88 433.68 Tm +(up) Tj +1 0 0 1 367.56 433.68 Tm +(to) Tj +1 0 0 1 378.96 433.68 Tm +(date,) Tj +1 0 0 1 402 433.68 Tm +(it) Tj +1 0 0 1 411.12 433.68 Tm +(is) Tj +1 0 0 1 421.44 433.68 Tm +(only) Tj +1 0 0 1 442.919 433.68 Tm +(necessary) Tj +1 0 0 1 485.279 433.68 Tm +(to) Tj +1 0 0 1 496.679 433.68 Tm +(send) Tj +1 0 0 1 518.639 433.68 Tm +(it) Tj +1 0 0 1 527.879 433.68 Tm +(the) Tj +1 0 0 1 315 421.8 Tm +(partitions) Tj +1 0 0 1 356.04 421.8 Tm +(where) Tj +1 0 0 1 383.64 421.8 Tm +(it) Tj +1 0 0 1 392.64 421.8 Tm +(is) Tj +1 0 0 1 402.6 421.8 Tm +(out) Tj +1 0 0 1 418.68 421.8 Tm +(of) Tj +1 0 0 1 430.32 421.8 Tm +(date,) Tj +1 0 0 1 453 421.8 Tm +(rather) Tj +1 0 0 1 479.639 421.8 Tm +(than) Tj +1 0 0 1 500.159 421.8 Tm +(the) Tj +1 0 0 1 515.639 421.8 Tm +(whole) Tj +1 0 0 1 315 409.8 Tm +(checkpoint.) Tj +ET +endstream +endobj +542 0 obj +15114 +endobj +543 0 obj +<< /Type /Font /Name /R543 /Subtype /Type1 /BaseFont /Times-Bold >> +endobj +544 0 obj +<< /Length 545 0 R >> +stream +BT +/R543 10 Tf +1 0 0 1 315 387.12 Tm +(4.5) Tj +1 0 0 1 337.44 387.12 Tm +(Corr) Tj +1 0 0 1 358.44 387.12 Tm +(ectness) Tj +/R445 10 Tf +1 0 0 1 315 372.24 Tm +(This) Tj +1 0 0 1 339.24 372.24 Tm +(section) Tj +1 0 0 1 374.04 372.24 Tm +(sk) Tj +1 0 0 1 382.8 372.24 Tm +(etches) Tj +1 0 0 1 414.24 372.24 Tm +(the) Tj +1 0 0 1 432.96 372.24 Tm +(proof) Tj +1 0 0 1 461.04 372.24 Tm +(that) Tj +1 0 0 1 482.519 372.24 Tm +(the) Tj +1 0 0 1 501.239 372.24 Tm +(algorithm) Tj +1 0 0 1 315 360.36 Tm +(pro) Tj +1 0 0 1 328.32 360.36 Tm +(vides) Tj +1 0 0 1 351.6 360.36 Tm +(safety) Tj +1 0 0 1 377.88 360.36 Tm +(and) Tj +1 0 0 1 394.8 360.36 Tm +(li) Tj +1 0 0 1 400.08 360.36 Tm +(v) Tj +1 0 0 1 405 360.36 Tm +(eness;) Tj +1 0 0 1 431.76 360.36 Tm +(details) Tj +1 0 0 1 460.319 360.36 Tm +(can) Tj +1 0 0 1 476.639 360.36 Tm +(be) Tj +1 0 0 1 488.519 360.36 Tm +(found) Tj +1 0 0 1 514.199 360.36 Tm +(in) Tj +1 0 0 1 524.519 360.36 Tm +([4) Tj +1 0 0 1 532.799 360.36 Tm +(].) Tj +/R543 10 Tf +1 0 0 1 315 337.56 Tm +(4.5.1) Tj +1 0 0 1 344.88 337.56 Tm +(Safety) Tj +/R445 10 Tf +1 0 0 1 315 322.8 Tm +(As) Tj +1 0 0 1 328.92 322.8 Tm +(discussed) Tj +1 0 0 1 370.08 322.8 Tm +(earlier) Tj +1 0 0 1 395.28 322.8 Tm +(,) Tj +1 0 0 1 400.56 322.8 Tm +(the) Tj +1 0 0 1 415.68 322.8 Tm +(algorithm) Tj +1 0 0 1 457.319 322.8 Tm +(pro) Tj +1 0 0 1 470.639 322.8 Tm +(vides) Tj +1 0 0 1 494.399 322.8 Tm +(safety) Tj +1 0 0 1 521.039 322.8 Tm +(if) Tj +1 0 0 1 530.039 322.8 Tm +(all) Tj +1 0 0 1 315 310.8 Tm +(non-f) Tj +1 0 0 1 336.72 310.8 Tm +(aulty) Tj +1 0 0 1 360.6 310.8 Tm +(replicas) Tj +1 0 0 1 395.88 310.8 Tm +(agree) Tj +1 0 0 1 421.68 310.8 Tm +(on) Tj +1 0 0 1 435.96 310.8 Tm +(the) Tj +1 0 0 1 452.4 310.8 Tm +(sequence) Tj +1 0 0 1 493.079 310.8 Tm +(numbers) Tj +1 0 0 1 531.719 310.8 Tm +(of) Tj +1 0 0 1 315 298.92 Tm +(requests) Tj +1 0 0 1 350.16 298.92 Tm +(that) Tj +1 0 0 1 367.56 298.92 Tm +(commit) Tj +1 0 0 1 400.56 298.92 Tm +(locally) Tj +1 0 0 1 427.2 298.92 Tm +(.) Tj +1 0 0 1 324.96 286.68 Tm +(In) Tj +1 0 0 1 336 286.68 Tm +(Section) Tj +1 0 0 1 368.52 286.68 Tm +(4.2,) Tj +1 0 0 1 386.28 286.68 Tm +(we) Tj +1 0 0 1 400.56 286.68 Tm +(sho) Tj +1 0 0 1 414.24 286.68 Tm +(wed) Tj +1 0 0 1 433.44 286.68 Tm +(that) Tj +1 0 0 1 451.08 286.68 Tm +(if) Tj +/R524 10 Tf +1 0 0 1 459.84 286.68 Tm +(pr) Tj +1 0 0 1 468.36 286.68 Tm +(epar) Tj +1 0 0 1 486.36 286.68 Tm +(ed) Tj +ET +endstream +endobj +545 0 obj +2285 +endobj +546 0 obj +<< /Type /XObject /Name /R546 /Subtype /Image /Length 547 0 R +/ImageMask true /Width 24 /Height 99 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 24 /BlackIs1 true >>] +>> +stream ++rF]%*LO8IHL@*j8[taqp!<~> +endstream +endobj +547 0 obj +80 +endobj +548 0 obj +<< /Length 549 0 R >> +stream +q +2.4 0 0 -9.9 496.7 294.2 cm +/R546 Do +Q +q +8.1 0 0 -4.6 499.9 291.2 cm +/R504 Do +Q +q +1.3 0 0 -3 509.2 287.8 cm +/R464 Do +Q +q +4.4 0 0 -4.6 513.1 291.2 cm +/R452 Do +Q +q +1.3 0 0 -3 518.8 287.8 cm +/R464 Do +Q +q +5.4 0 0 -4.6 522.7 291.2 cm +/R468 Do +Q +q +1.3 0 0 -3 529.3 287.8 cm +/R464 Do +Q +q +2.6 0 0 -6.9 533.1 293.5 cm +/R448 Do +Q +endstream +endobj +549 0 obj +322 +endobj +550 0 obj +<< /Type /XObject /Name /R550 /Subtype /Image /Length 551 0 R +/ImageMask true /Width 24 /Height 99 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 24 /BlackIs1 true >>] +>> +stream +,JEO$loTRL4niC`hqS$`rnmKps6nkFp\Y!js6KW5s0(&RK)W@\(KK(`l\2@U?Bm&k<5,",!$D~> +endstream +endobj +551 0 obj +76 +endobj +552 0 obj +<< /Length 553 0 R >> +stream +q +2.4 0 0 -9.9 536.8 294.2 cm +/R550 Do +Q +BT +/R445 10 Tf +1 0 0 1 315 274.68 Tm +(is) Tj +1 0 0 1 324.96 274.68 Tm +(true,) Tj +/R524 10 Tf +1 0 0 1 346.56 274.68 Tm +(pr) Tj +1 0 0 1 355.08 274.68 Tm +(epar) Tj +1 0 0 1 373.08 274.68 Tm +(ed) Tj +ET +q +2.4 0 0 -9.9 383.5 282.2 cm +/R546 Do +Q +q +8.1 0 0 -4.6 386.6 279.2 cm +/R504 Do +Q +endstream +endobj +553 0 obj +320 +endobj +554 0 obj +<< /Type /XObject /Name /R554 /Subtype /Image /Length 555 0 R +/ImageMask true /Width 17 /Height 37 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 17 /BlackIs1 true >>] +>> +stream +2p&%Irsf"qs+g(;pODr+p\l"Z@=(fO!"]~> +endstream +endobj +555 0 obj +36 +endobj +556 0 obj +<< /Length 557 0 R >> +stream +q +1.7 0 0 -3.7 395.5 282.2 cm +/R554 Do +Q +q +1.3 0 0 -3 398.5 275.8 cm +/R464 Do +Q +q +4.4 0 0 -4.6 402.3 279.2 cm +/R452 Do +Q +q +1.3 0 0 -3 408.1 275.8 cm +/R464 Do +Q +q +5.4 0 0 -4.6 411.9 279.2 cm +/R468 Do +Q +q +1.3 0 0 -3 418.5 275.8 cm +/R464 Do +Q +endstream +endobj +557 0 obj +240 +endobj +558 0 obj +<< /Type /XObject /Name /R558 /Subtype /Image /Length 559 0 R +/ImageMask true /Width 41 /Height 89 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 41 /BlackIs1 true >>] +>> +stream +3%oqNe0+KR +8.TS3gNRT8m0C&=*WH)RhtR$Is*qrFs*qfPpZ+:-rpTR[p\3 +endstream +endobj +559 0 obj +92 +endobj +560 0 obj +<< /Length 561 0 R >> +stream +q +4.1 0 0 -8.9 422 281.5 cm +/R558 Do +Q +q +2.4 0 0 -9.9 427.2 282.2 cm +/R550 Do +Q +BT +/R445 10 Tf +1 0 0 1 434.04 274.68 Tm +(is) Tj +1 0 0 1 444 274.68 Tm +(f) Tj +1 0 0 1 447.24 274.68 Tm +(alse) Tj +1 0 0 1 466.08 274.68 Tm +(for) Tj +1 0 0 1 481.079 274.68 Tm +(an) Tj +1 0 0 1 490.44 274.68 Tm +(y) Tj +1 0 0 1 498.599 274.68 Tm +(non-f) Tj +1 0 0 1 520.319 274.68 Tm +(aulty) Tj +1 0 0 1 315 262.8 Tm +(replica) Tj +ET +q +4.1 0 0 -8.9 347.6 269.6 cm +/R558 Do +Q +BT +1 0 0 1 358.08 262.8 Tm +(\(including) Tj +ET +q +2.6 0 0 -6.9 405 269.6 cm +/R448 Do +Q +endstream +endobj +561 0 obj +530 +endobj +562 0 obj +<< /Type /XObject /Name /R562 /Subtype /Image /Length 563 0 R +/ImageMask true /Width 66 /Height 27 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 66 /BlackIs1 true >>] +>> +stream +/e%CP4)j*Wq&BoPm6CdS5Q~> +endstream +endobj +563 0 obj +25 +endobj +564 0 obj +<< /Length 565 0 R >> +stream +q +6.6 0 0 -2.7 415.8 266.7 cm +/R562 Do +Q +q +4.1 0 0 -8.9 430.1 269.6 cm +/R558 Do +Q +BT +1 0 0 1 435 262.8 Tm +(\)) Tj +1 0 0 1 444 262.8 Tm +(and) Tj +1 0 0 1 464.04 262.8 Tm +(an) Tj +1 0 0 1 473.399 262.8 Tm +(y) Tj +ET +q +8.1 0 0 -4.6 484.2 267.3 cm +/R504 Do +Q +q +1.7 0 0 -3.7 493 270.3 cm +/R554 Do +Q +BT +1 0 0 1 501.12 262.8 Tm +(such) Tj +1 0 0 1 525 262.8 Tm +(that) Tj +ET +endstream +endobj +565 0 obj +362 +endobj +566 0 obj +<< /Type /XObject /Name /R566 /Subtype /Image /Length 567 0 R +/ImageMask true /Width 75 /Height 69 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 75 /BlackIs1 true >>] +>> +stream +-D@i/="t3q:)'T:EY'Q7@JfCQK73IN +DqmDb@DDO(rqCj@Fs$I.iOf--It,63^V:<:qnMn*^\@b0s+Zm:ht=@d?EiAlYY3Kq%(IruTE5*9~> +endstream +endobj +567 0 obj +109 +endobj +568 0 obj +<< /Length 569 0 R >> +stream +q +7.5 0 0 -6.9 315.3 257.7 cm +/R566 Do +Q +q +2.4 0 0 -9.9 324.4 258.3 cm +/R546 Do +Q +q +8.1 0 0 -4.6 327.6 255.3 cm +/R504 Do +Q +q +1.7 0 0 -3.7 336.4 258.3 cm +/R554 Do +Q +q +2.4 0 0 -9.9 339.4 258.3 cm +/R550 Do +Q +endstream +endobj +569 0 obj +205 +endobj +570 0 obj +<< /Type /XObject /Name /R570 /Subtype /Image /Length 571 0 R +/ImageMask true /Width 51 /Height 94 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 51 /BlackIs1 true >>] +>> +stream +,Ll]7 +h=lC.rSQ7WGMW0U^L?$SqnImbmtZC2mf"1&^)d0e]D-=4pilH2n):tW%fYAApCHZh48XMGLM?f3!W~> +endstream +endobj +571 0 obj +86 +endobj +572 0 obj +<< /Length 573 0 R >> +stream +q +5.1 0 0 -9.4 350.4 258 cm +/R570 Do +Q +q +6.6 0 0 -2.7 349.6 254.7 cm +/R562 Do +Q +q +7.5 0 0 -6.9 363.4 257.7 cm +/R566 Do +Q +q +2.4 0 0 -9.9 372.5 258.3 cm +/R546 Do +Q +q +8.1 0 0 -4.6 375.7 255.3 cm +/R504 Do +Q +q +2.4 0 0 -9.9 384.6 258.3 cm +/R550 Do +Q +BT +1 0 0 1 388.08 250.8 Tm +(.) Tj +1 0 0 1 401.76 250.8 Tm +(This) Tj +1 0 0 1 424.44 250.8 Tm +(implies) Tj +1 0 0 1 458.76 250.8 Tm +(that) Tj +1 0 0 1 478.8 250.8 Tm +(tw) Tj +1 0 0 1 488.64 250.8 Tm +(o) Tj +1 0 0 1 498.6 250.8 Tm +(non-f) Tj +1 0 0 1 520.32 250.8 Tm +(aulty) Tj +1 0 0 1 315 238.92 Tm +(replicas) Tj +1 0 0 1 349.44 238.92 Tm +(agree) Tj +1 0 0 1 374.4 238.92 Tm +(on) Tj +1 0 0 1 387.84 238.92 Tm +(the) Tj +1 0 0 1 403.44 238.92 Tm +(sequence) Tj +1 0 0 1 443.4 238.92 Tm +(number) Tj +1 0 0 1 477.239 238.92 Tm +(of) Tj +1 0 0 1 488.999 238.92 Tm +(requests) Tj +1 0 0 1 524.999 238.92 Tm +(that) Tj +1 0 0 1 315 226.92 Tm +(commit) Tj +1 0 0 1 347.88 226.92 Tm +(locally) Tj +1 0 0 1 377.52 226.92 Tm +(in) Tj +1 0 0 1 387.84 226.92 Tm +(the) Tj +1 0 0 1 402.48 226.92 Tm +(same) Tj +1 0 0 1 425.4 226.92 Tm +(vie) Tj +1 0 0 1 437.4 226.92 Tm +(w) Tj +1 0 0 1 447 226.92 Tm +(at) Tj +1 0 0 1 456.719 226.92 Tm +(the) Tj +1 0 0 1 471.359 226.92 Tm +(tw) Tj +1 0 0 1 481.199 226.92 Tm +(o) Tj +1 0 0 1 488.759 226.92 Tm +(replicas.) Tj +1 0 0 1 324.96 214.68 Tm +(The) Tj +1 0 0 1 346.32 214.68 Tm +(vie) Tj +1 0 0 1 358.32 214.68 Tm +(w-change) Tj +1 0 0 1 402.84 214.68 Tm +(protocol) Tj +1 0 0 1 441.96 214.68 Tm +(ensures) Tj +1 0 0 1 477.72 214.68 Tm +(that) Tj +1 0 0 1 498.6 214.68 Tm +(non-f) Tj +1 0 0 1 520.319 214.68 Tm +(aulty) Tj +1 0 0 1 315 202.68 Tm +(replicas) Tj +1 0 0 1 349.32 202.68 Tm +(also) Tj +1 0 0 1 368.64 202.68 Tm +(agree) Tj +1 0 0 1 393.48 202.68 Tm +(on) Tj +1 0 0 1 406.8 202.68 Tm +(the) Tj +1 0 0 1 422.16 202.68 Tm +(sequence) Tj +1 0 0 1 462 202.68 Tm +(number) Tj +1 0 0 1 495.719 202.68 Tm +(of) Tj +1 0 0 1 507.359 202.68 Tm +(requests) Tj +1 0 0 1 315 190.8 Tm +(that) Tj +1 0 0 1 331.68 190.8 Tm +(commit) Tj +1 0 0 1 363.96 190.8 Tm +(locally) Tj +1 0 0 1 392.88 190.8 Tm +(in) Tj +1 0 0 1 402.36 190.8 Tm +(dif) Tj +1 0 0 1 413.28 190.8 Tm +(fer) Tj +1 0 0 1 424.32 190.8 Tm +(ent) Tj +1 0 0 1 438.119 190.8 Tm +(vie) Tj +1 0 0 1 450.119 190.8 Tm +(w) Tj +1 0 0 1 457.199 190.8 Tm +(s) Tj +1 0 0 1 462.719 190.8 Tm +(at) Tj +1 0 0 1 471.599 190.8 Tm +(d) Tj +1 0 0 1 476.519 190.8 Tm +(if) Tj +1 0 0 1 482.399 190.8 Tm +(fe) Tj +1 0 0 1 490.079 190.8 Tm +(ren) Tj +1 0 0 1 502.799 190.8 Tm +(t) Tj +1 0 0 1 507.239 190.8 Tm +(rep) Tj +1 0 0 1 519.959 190.8 Tm +(licas.) Tj +1 0 0 1 315 178.8 Tm +(A) Tj +1 0 0 1 324.48 178.8 Tm +(request) Tj +ET +q +8.1 0 0 -4.6 355.9 183.3 cm +/R504 Do +Q +BT +1 0 0 1 366.72 178.8 Tm +(commits) Tj +1 0 0 1 403.32 178.8 Tm +(locally) Tj +1 0 0 1 432.84 178.8 Tm +(at) Tj +1 0 0 1 442.32 178.8 Tm +(a) Tj +1 0 0 1 449.16 178.8 Tm +(non-f) Tj +1 0 0 1 470.879 178.8 Tm +(aulty) Tj +1 0 0 1 492.839 178.8 Tm +(replica) Tj +1 0 0 1 522.239 178.8 Tm +(with) Tj +1 0 0 1 315 166.92 Tm +(sequence) Tj +1 0 0 1 353.76 166.92 Tm +(number) Tj +ET +q +5.4 0 0 -4.6 386.7 171.4 cm +/R468 Do +Q +BT +1 0 0 1 394.8 166.92 Tm +(in) Tj +1 0 0 1 404.76 166.92 Tm +(vie) Tj +1 0 0 1 416.76 166.92 Tm +(w) Tj +ET +q +4.4 0 0 -4.6 426.3 171.4 cm +/R452 Do +Q +BT +1 0 0 1 433.68 166.92 Tm +(only) Tj +1 0 0 1 453.6 166.92 Tm +(if) Tj +/R524 10 Tf +1 0 0 1 461.999 166.92 Tm +(committed) Tj +ET +q +2.4 0 0 -9.9 504.5 174.4 cm +/R546 Do +Q +q +8.1 0 0 -4.6 507.7 171.4 cm +/R504 Do +Q +q +1.3 0 0 -3 517 168 cm +/R464 Do +Q +q +4.4 0 0 -4.6 520.9 171.4 cm +/R452 Do +Q +q +1.3 0 0 -3 526.6 168 cm +/R464 Do +Q +q +5.4 0 0 -4.6 530.5 171.4 cm +/R468 Do +Q +q +2.4 0 0 -9.9 536.7 174.4 cm +/R550 Do +Q +BT +/R445 10 Tf +1 0 0 1 315 154.92 Tm +(is) Tj +1 0 0 1 323.28 154.92 Tm +(true.) Tj +1 0 0 1 344.52 154.92 Tm +(This) Tj +1 0 0 1 363.96 154.92 Tm +(means) Tj +1 0 0 1 391.2 154.92 Tm +(that) Tj +1 0 0 1 407.88 154.92 Tm +(there) Tj +1 0 0 1 429.6 154.92 Tm +(is) Tj +1 0 0 1 437.879 154.92 Tm +(a) Tj +1 0 0 1 443.999 154.92 Tm +(set) Tj +ET +endstream +endobj +573 0 obj +3843 +endobj +574 0 obj +<< /Type /XObject /Name /R574 /Subtype /Image /Length 575 0 R +/ImageMask true /Width 72 /Height 72 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 72 /BlackIs1 true >>] +>> +stream +,bfN+EbZ%+T220fDo-cZ,s?pcnFun\pj__WTt^P^p]'u;n,MhU +gAgoM\7one3%l@pc*52@)Lt#%cP0LqgZ5-hr4=Er'14o0CHFELJ)8VIoB/o@='0"=9*jW~> +endstream +endobj +575 0 obj +129 +endobj +576 0 obj +<< /Length 577 0 R >> +stream +q +7.2 0 0 -7.2 457 161.8 cm +/R574 Do +Q +BT +/R445 7 Tf +1 0 0 1 464.28 153.48 Tm +(1) Tj +/R445 10 Tf +1 0 0 1 469.2 154.92 Tm +(containing) Tj +1 0 0 1 513.24 154.92 Tm +(at) Tj +1 0 0 1 522.12 154.92 Tm +(least) Tj +ET +q +5 0 0 -9.2 315.4 150.1 cm +/R496 Do +Q +q +6.6 0 0 -6.5 324 148.8 cm +/R456 Do +Q +BT +1 0 0 1 333.84 143.04 Tm +(1) Tj +1 0 0 1 341.88 143.04 Tm +(non-f) Tj +1 0 0 1 363.6 143.04 Tm +(aulty) Tj +1 0 0 1 386.4 143.04 Tm +(replicas) Tj +1 0 0 1 420.48 143.04 Tm +(such) Tj +1 0 0 1 441.84 143.04 Tm +(that) Tj +/R524 10 Tf +1 0 0 1 459.839 143.04 Tm +(pr) Tj +1 0 0 1 468.359 143.04 Tm +(epar) Tj +1 0 0 1 486.359 143.04 Tm +(ed) Tj +ET +q +2.4 0 0 -9.9 496.7 150.5 cm +/R546 Do +Q +q +8.1 0 0 -4.6 499.9 147.5 cm +/R504 Do +Q +q +1.3 0 0 -3 509.2 144.1 cm +/R464 Do +Q +q +4.4 0 0 -4.6 513.1 147.5 cm +/R452 Do +Q +q +1.3 0 0 -3 518.8 144.1 cm +/R464 Do +Q +q +5.4 0 0 -4.6 522.7 147.5 cm +/R468 Do +Q +q +1.3 0 0 -3 529.3 144.1 cm +/R464 Do +Q +q +2.6 0 0 -6.9 533.1 149.8 cm +/R448 Do +Q +q +2.4 0 0 -9.9 536.8 150.5 cm +/R550 Do +Q +BT +/R445 10 Tf +1 0 0 1 315 131.04 Tm +(is) Tj +1 0 0 1 324.12 131.04 Tm +(true) Tj +1 0 0 1 342.12 131.04 Tm +(for) Tj +1 0 0 1 356.28 131.04 Tm +(e) Tj +1 0 0 1 360.48 131.04 Tm +(v) Tj +1 0 0 1 365.4 131.04 Tm +(ery) Tj +1 0 0 1 380.4 131.04 Tm +(replica) Tj +ET +q +2.6 0 0 -6.9 410.2 137.8 cm +/R448 Do +Q +BT +1 0 0 1 415.919 131.04 Tm +(in) Tj +1 0 0 1 426.239 131.04 Tm +(the) Tj +1 0 0 1 440.879 131.04 Tm +(set.) Tj +1 0 0 1 324.96 118.8 Tm +(Non-f) Tj +1 0 0 1 348.84 118.8 Tm +(aulty) Tj +1 0 0 1 372.12 118.8 Tm +(replicas) Tj +1 0 0 1 406.68 118.8 Tm +(will) Tj +1 0 0 1 425.64 118.8 Tm +(not) Tj +1 0 0 1 441.96 118.8 Tm +(accept) Tj +1 0 0 1 470.999 118.8 Tm +(a) Tj +1 0 0 1 478.919 118.8 Tm +(pre-prepare) Tj +1 0 0 1 528.359 118.8 Tm +(for) Tj +1 0 0 1 315 106.8 Tm +(vie) Tj +1 0 0 1 327 106.8 Tm +(w) Tj +ET +q +4.4 0 0 -4.6 336 111.3 cm +/R452 Do +Q +q +1.7 0 0 -3.7 341.2 114.3 cm +/R554 Do +Q +endstream +endobj +577 0 obj +1843 +endobj +578 0 obj +<< /Type /XObject /Name /R578 /Subtype /Image /Length 579 0 R +/ImageMask true /Width 61 /Height 61 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 61 /BlackIs1 true >>] +>> +stream +,Ld57i8DMjL(5&>G[TII_hJX +endstream +endobj +579 0 obj +114 +endobj +580 0 obj +<< /Length 581 0 R >> +stream +q +6.1 0 0 -6.1 347.4 112.4 cm +/R578 Do +Q +q +4.4 0 0 -4.6 357.3 111.3 cm +/R452 Do +Q +BT +1 0 0 1 363.96 106.8 Tm +(without) Tj +1 0 0 1 396 106.8 Tm +(ha) Tj +1 0 0 1 405.24 106.8 Tm +(ving) Tj +1 0 0 1 424.68 106.8 Tm +(recei) Tj +1 0 0 1 443.88 106.8 Tm +(v) Tj +1 0 0 1 448.8 106.8 Tm +(ed) Tj +1 0 0 1 459.72 106.8 Tm +(a) Tj +1 0 0 1 465.84 106.8 Tm +(ne) Tj +1 0 0 1 475.079 106.8 Tm +(w-vie) Tj +1 0 0 1 497.639 106.8 Tm +(w) Tj +1 0 0 1 506.279 106.8 Tm +(message) Tj +1 0 0 1 315 94.92 Tm +(for) Tj +ET +q +4.4 0 0 -4.6 328.6 99.4 cm +/R452 Do +Q +q +1.7 0 0 -3.7 333.9 102.4 cm +/R554 Do +Q +BT +1 0 0 1 338.16 94.92 Tm +(\(since) Tj +1 0 0 1 363.72 94.92 Tm +(only) Tj +1 0 0 1 383.16 94.92 Tm +(at) Tj +1 0 0 1 392.16 94.92 Tm +(that) Tj +1 0 0 1 408.84 94.92 Tm +(point) Tj +1 0 0 1 431.16 94.92 Tm +(do) Tj +1 0 0 1 442.92 94.92 Tm +(the) Tj +1 0 0 1 455.04 94.92 Tm +(y) Tj +1 0 0 1 461.639 94.92 Tm +(enter) Tj +1 0 0 1 483.359 94.92 Tm +(the) Tj +1 0 0 1 497.279 94.92 Tm +(vie) Tj +1 0 0 1 509.279 94.92 Tm +(w\).) Tj +1 0 0 1 525.599 94.92 Tm +(But) Tj +1 0 0 1 315 82.92 Tm +(an) Tj +1 0 0 1 324.36 82.92 Tm +(y) Tj +1 0 0 1 332.04 82.92 Tm +(correct) Tj +1 0 0 1 362.4 82.92 Tm +(ne) Tj +1 0 0 1 371.64 82.92 Tm +(w-vie) Tj +1 0 0 1 394.2 82.92 Tm +(w) Tj +1 0 0 1 404.04 82.92 Tm +(message) Tj +1 0 0 1 440.52 82.92 Tm +(for) Tj +1 0 0 1 454.92 82.92 Tm +(vie) Tj +1 0 0 1 466.92 82.92 Tm +(w) Tj +ET +q +4.4 0 0 -4.6 477 87.4 cm +/R452 Do +Q +q +1.7 0 0 -3.7 482.2 90.4 cm +/R554 Do +Q +q +6.1 0 0 -6.1 488.7 88.5 cm +/R578 Do +Q +q +4.4 0 0 -4.6 499 87.4 cm +/R452 Do +Q +BT +1 0 0 1 506.76 82.92 Tm +(contains) Tj +1 0 0 1 315 71.04 Tm +(correct) Tj +1 0 0 1 345.84 71.04 Tm +(vie) Tj +1 0 0 1 357.84 71.04 Tm +(w-change) Tj +1 0 0 1 399.6 71.04 Tm +(messages) Tj +1 0 0 1 440.4 71.04 Tm +(from) Tj +1 0 0 1 462.96 71.04 Tm +(e) Tj +1 0 0 1 467.16 71.04 Tm +(v) Tj +1 0 0 1 472.08 71.04 Tm +(ery) Tj +1 0 0 1 487.8 71.04 Tm +(replica) Tj +ET +q +2.6 0 0 -6.9 518.2 77.8 cm +/R448 Do +Q +BT +1 0 0 1 524.639 71.04 Tm +(in) Tj +1 0 0 1 535.559 71.04 Tm +(a) Tj +1 0 0 1 303.48 35.16 Tm +(6) Tj +ET +Q +endstream +endobj +581 0 obj +1989 +endobj +582 0 obj +<< +/Type /Page +/MediaBox [0 0 612 792] +/Parent 2 0 R +/Resources << /ProcSet [/PDF /ImageB /Text] +/Font << +/R543 543 0 R +/R524 524 0 R +/R445 445 0 R +>> +/XObject << +/R448 448 0 R +/R452 452 0 R +/R578 578 0 R +/R554 554 0 R +/R550 550 0 R +/R464 464 0 R +/R468 468 0 R +/R504 504 0 R +/R546 546 0 R +/R456 456 0 R +/R496 496 0 R +/R574 574 0 R +/R566 566 0 R +/R562 562 0 R +/R570 570 0 R +/R558 558 0 R +/R492 492 0 R +/R512 512 0 R +/R516 516 0 R +/R539 539 0 R +/R535 535 0 R +/R531 531 0 R +/R527 527 0 R +/R520 520 0 R +/R484 484 0 R +/R480 480 0 R +/R460 460 0 R +/R476 476 0 R +/R508 508 0 R +/R500 500 0 R +/R472 472 0 R +/R488 488 0 R +>> +>> +/Contents [ +446 0 R +450 0 R +454 0 R +458 0 R +462 0 R +466 0 R +470 0 R +474 0 R +478 0 R +482 0 R +486 0 R +490 0 R +494 0 R +498 0 R +502 0 R +506 0 R +510 0 R +514 0 R +518 0 R +522 0 R +525 0 R +529 0 R +533 0 R +537 0 R +541 0 R +544 0 R +548 0 R +552 0 R +556 0 R +560 0 R +564 0 R +568 0 R +572 0 R +576 0 R +580 0 R +] +>> +endobj +583 0 obj +<< /Type /Font /Name /R583 /Subtype /Type1 /BaseFont /Times-Roman >> +endobj +584 0 obj +<< /Length 585 0 R >> +stream +q +BT +/R583 10 Tf +1 0 0 1 72 709.08 Tm +(set) Tj +ET +Q +q +W +0 0 612 792 re +n +endstream +endobj +585 0 obj +73 +endobj +586 0 obj +<< /Type /XObject /Name /R586 /Subtype /Image /Length 587 0 R +/ImageMask true /Width 72 /Height 72 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 72 /BlackIs1 true >>] +>> +stream +,bfN+EbZ%+T220fDo-cZ,s?pcnFun\pj__WTt^P^p]'u;n,MhU +gAgoM\7one3%l@pc*52@)Lt#%cP0LqgZ5-hr4=Er'14o0CHFELJ)8VIoB/o@='0"=9*jW~> +endstream +endobj +587 0 obj +129 +endobj +588 0 obj +<< /Length 589 0 R >> +stream +q +7.2 0 0 -7.2 85.5 716 cm +/R586 Do +Q +BT +/R583 7 Tf +1 0 0 1 92.76 707.64 Tm +(2) Tj +/R583 10 Tf +1 0 0 1 98.76 709.08 Tm +(of) Tj +1 0 0 1 109.2 709.08 Tm +(2) Tj +ET +endstream +endobj +589 0 obj +161 +endobj +590 0 obj +<< /Type /XObject /Name /R590 /Subtype /Image /Length 591 0 R +/ImageMask true /Width 50 /Height 92 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 50 /BlackIs1 true >>] +>> +stream +5$eL7-b:K3p0<%? +h1u#R]2'Mug].5is7$!eqtL-hqu?EZs5/V3iQXZVs7/ihqu?EZs4QRbJe-ans*o2/4DdB_@6aqq5Q~> +endstream +endobj +591 0 obj +96 +endobj +592 0 obj +<< /Length 593 0 R >> +stream +q +5 0 0 -9.2 114.6 716.2 cm +/R590 Do +Q +endstream +endobj +593 0 obj +39 +endobj +594 0 obj +<< /Type /XObject /Name /R594 /Subtype /Image /Length 595 0 R +/ImageMask true /Width 66 /Height 65 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 66 /BlackIs1 true >>] +>> +stream +,BZ6Ns8W-!s8W-!s-Eq9Iaj?!s8W-!s8W-!s8Uf'#QTA~> +endstream +endobj +595 0 obj +47 +endobj +596 0 obj +<< /Length 597 0 R >> +stream +q +6.6 0 0 -6.5 121.9 714.9 cm +/R594 Do +Q +BT +1 0 0 1 130.56 709.08 Tm +(1) Tj +1 0 0 1 137.64 709.08 Tm +(replicas.) Tj +1 0 0 1 174.6 709.08 Tm +(Since) Tj +1 0 0 1 198.84 709.08 Tm +(there) Tj +1 0 0 1 220.8 709.08 Tm +(are) Tj +1 0 0 1 235.08 709.08 Tm +(3) Tj +ET +q +5 0 0 -9.2 240.5 716.2 cm +/R590 Do +Q +q +6.6 0 0 -6.5 247.8 714.9 cm +/R594 Do +Q +BT +1 0 0 1 256.439 709.08 Tm +(1) Tj +1 0 0 1 263.519 709.08 Tm +(replicas,) Tj +ET +q +7.2 0 0 -7.2 72.3 704 cm +/R586 Do +Q +BT +/R583 7 Tf +1 0 0 1 79.56 695.64 Tm +(1) Tj +/R583 10 Tf +1 0 0 1 86.28 697.08 Tm +(and) Tj +ET +q +7.2 0 0 -7.2 103.6 704 cm +/R586 Do +Q +BT +/R583 7 Tf +1 0 0 1 110.88 695.64 Tm +(2) Tj +/R583 10 Tf +1 0 0 1 117.6 697.08 Tm +(must) Tj +1 0 0 1 139.68 697.08 Tm +(intersect) Tj +1 0 0 1 176.16 697.08 Tm +(in) Tj +1 0 0 1 186.72 697.08 Tm +(at) Tj +1 0 0 1 196.56 697.08 Tm +(least) Tj +1 0 0 1 217.56 697.08 Tm +(one) Tj +1 0 0 1 234.6 697.08 Tm +(replica) Tj +ET +endstream +endobj +597 0 obj +893 +endobj +598 0 obj +<< /Type /XObject /Name /R598 /Subtype /Image /Length 599 0 R +/ImageMask true /Width 47 /Height 71 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 47 /BlackIs1 true >>] +>> +stream +0ZYuc@0_O>l/"oP624[@s1A +endstream +endobj +599 0 obj +119 +endobj +600 0 obj +<< /Length 601 0 R >> +stream +q +4.7 0 0 -7.1 264.9 704.1 cm +/R598 Do +Q +BT +1 0 0 1 272.639 697.08 Tm +(that) Tj +1 0 0 1 290.399 697.08 Tm +(is) Tj +1 0 0 1 72 685.2 Tm +(not) Tj +1 0 0 1 87.36 685.2 Tm +(f) Tj +1 0 0 1 90.5999 685.2 Tm +(aulty) Tj +1 0 0 1 110.04 685.2 Tm +(.) Tj +ET +q +4.7 0 0 -7.1 116.7 692.2 cm +/R598 Do +Q +BT +1 0 0 1 121.8 685.2 Tm +(') Tj +1 0 0 1 124.56 685.2 Tm +(s) Tj +1 0 0 1 131.04 685.2 Tm +(vie) Tj +1 0 0 1 143.04 685.2 Tm +(w-change) Tj +1 0 0 1 184.2 685.2 Tm +(message) Tj +1 0 0 1 220.56 685.2 Tm +(will) Tj +1 0 0 1 238.68 685.2 Tm +(ensure) Tj +1 0 0 1 267.239 685.2 Tm +(that) Tj +1 0 0 1 284.879 685.2 Tm +(the) Tj +1 0 0 1 72 673.2 Tm +(f) Tj +1 0 0 1 75.24 673.2 Tm +(act) Tj +1 0 0 1 89.6399 673.2 Tm +(that) Tj +ET +endstream +endobj +601 0 obj +691 +endobj +602 0 obj +<< /Type /XObject /Name /R602 /Subtype /Image /Length 603 0 R +/ImageMask true /Width 81 /Height 46 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 81 /BlackIs1 true >>] +>> +stream +4N`CV%M3$8612U>:Ag"NnFZb +endstream +endobj +603 0 obj +139 +endobj +604 0 obj +<< /Length 605 0 R >> +stream +q +8.1 0 0 -4.6 107.5 677.7 cm +/R602 Do +Q +BT +1 0 0 1 118.8 673.2 Tm +(prepared) Tj +1 0 0 1 156.48 673.2 Tm +(in) Tj +1 0 0 1 166.92 673.2 Tm +(a) Tj +1 0 0 1 174.12 673.2 Tm +(pre) Tj +1 0 0 1 186.72 673.2 Tm +(vious) Tj +1 0 0 1 210.959 673.2 Tm +(vie) Tj +1 0 0 1 222.959 673.2 Tm +(w) Tj +1 0 0 1 232.799 673.2 Tm +(is) Tj +1 0 0 1 242.279 673.2 Tm +(propagated) Tj +1 0 0 1 289.199 673.2 Tm +(to) Tj +1 0 0 1 72 661.32 Tm +(subsequent) Tj +1 0 0 1 118.32 661.32 Tm +(vie) Tj +1 0 0 1 130.32 661.32 Tm +(ws,) Tj +1 0 0 1 145.92 661.32 Tm +(unless) Tj +1 0 0 1 172.92 661.32 Tm +(the) Tj +1 0 0 1 187.2 661.32 Tm +(ne) Tj +1 0 0 1 196.44 661.32 Tm +(w-vie) Tj +1 0 0 1 219 661.32 Tm +(w) Tj +1 0 0 1 228 661.32 Tm +(message) Tj +1 0 0 1 263.76 661.32 Tm +(contains) Tj +1 0 0 1 72 649.32 Tm +(a) Tj +1 0 0 1 79.56 649.32 Tm +(vie) Tj +1 0 0 1 91.5599 649.32 Tm +(w-change) Tj +1 0 0 1 133.32 649.32 Tm +(message) Tj +1 0 0 1 170.16 649.32 Tm +(with) Tj +1 0 0 1 191.04 649.32 Tm +(a) Tj +1 0 0 1 198.6 649.32 Tm +(stable) Tj +1 0 0 1 224.88 649.32 Tm +(checkpoint) Tj +1 0 0 1 271.799 649.32 Tm +(with) Tj +1 0 0 1 292.559 649.32 Tm +(a) Tj +1 0 0 1 72 637.44 Tm +(sequence) Tj +1 0 0 1 112.44 637.44 Tm +(number) Tj +1 0 0 1 146.76 637.44 Tm +(higher) Tj +1 0 0 1 176.04 637.44 Tm +(than) Tj +ET +endstream +endobj +605 0 obj +1231 +endobj +606 0 obj +<< /Type /XObject /Name /R606 /Subtype /Image /Length 607 0 R +/ImageMask true /Width 54 /Height 46 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 54 /BlackIs1 true >>] +>> +stream +4N]t,$m>7rl/"o@d!re, +pYTohhg]lis4_o!aocV_^Yur>rVsXfR^`8 +endstream +endobj +607 0 obj +102 +endobj +608 0 obj +<< /Length 609 0 R >> +stream +q +5.4 0 0 -4.6 197.2 641.9 cm +/R606 Do +Q +BT +1 0 0 1 203.04 637.44 Tm +(.) Tj +1 0 0 1 213.36 637.44 Tm +(In) Tj +1 0 0 1 225.48 637.44 Tm +(the) Tj +1 0 0 1 241.559 637.44 Tm +(\256rst) Tj +1 0 0 1 260.999 637.44 Tm +(case,) Tj +1 0 0 1 284.879 637.44 Tm +(the) Tj +1 0 0 1 72 625.44 Tm +(algorithm) Tj +1 0 0 1 112.92 625.44 Tm +(redoes) Tj +1 0 0 1 141.12 625.44 Tm +(the) Tj +1 0 0 1 155.52 625.44 Tm +(three) Tj +1 0 0 1 177.6 625.44 Tm +(phases) Tj +1 0 0 1 206.28 625.44 Tm +(of) Tj +1 0 0 1 216.84 625.44 Tm +(the) Tj +1 0 0 1 231.12 625.44 Tm +(atomic) Tj +1 0 0 1 260.519 625.44 Tm +(multicast) Tj +1 0 0 1 72 613.44 Tm +(protocol) Tj +1 0 0 1 107.4 613.44 Tm +(for) Tj +ET +q +8.1 0 0 -4.6 121.3 617.9 cm +/R602 Do +Q +BT +1 0 0 1 132 613.44 Tm +(with) Tj +1 0 0 1 151.92 613.44 Tm +(the) Tj +1 0 0 1 166.2 613.44 Tm +(same) Tj +1 0 0 1 188.88 613.44 Tm +(sequence) Tj +1 0 0 1 227.52 613.44 Tm +(number) Tj +ET +q +5.4 0 0 -4.6 260.4 617.9 cm +/R606 Do +Q +BT +1 0 0 1 268.319 613.44 Tm +(and) Tj +1 0 0 1 284.879 613.44 Tm +(the) Tj +1 0 0 1 72 601.56 Tm +(ne) Tj +1 0 0 1 81.24 601.56 Tm +(w) Tj +1 0 0 1 91.08 601.56 Tm +(vie) Tj +1 0 0 1 103.08 601.56 Tm +(w) Tj +1 0 0 1 112.92 601.56 Tm +(number) Tj +1 0 0 1 143.04 601.56 Tm +(.) Tj +1 0 0 1 149.76 601.56 Tm +(This) Tj +1 0 0 1 170.16 601.56 Tm +(is) Tj +1 0 0 1 179.52 601.56 Tm +(important) Tj +1 0 0 1 221.04 601.56 Tm +(because) Tj +1 0 0 1 255.359 601.56 Tm +(it) Tj +1 0 0 1 263.639 601.56 Tm +(pre) Tj +1 0 0 1 276.239 601.56 Tm +(v) Tj +1 0 0 1 281.159 601.56 Tm +(ents) Tj +1 0 0 1 72 589.56 Tm +(an) Tj +1 0 0 1 81.36 589.56 Tm +(y) Tj +1 0 0 1 91.1999 589.56 Tm +(dif) Tj +1 0 0 1 102.12 589.56 Tm +(ferent) Tj +1 0 0 1 130.2 589.56 Tm +(request) Tj +1 0 0 1 163.92 589.56 Tm +(that) Tj +1 0 0 1 183.72 589.56 Tm +(w) Tj +1 0 0 1 190.8 589.56 Tm +(as) Tj +1 0 0 1 204.12 589.56 Tm +(assigned) Tj +1 0 0 1 243.359 589.56 Tm +(the) Tj +1 0 0 1 260.519 589.56 Tm +(sequence) Tj +1 0 0 1 72 577.68 Tm +(number) Tj +ET +q +5.4 0 0 -4.6 105.8 582.2 cm +/R606 Do +Q +BT +1 0 0 1 114.6 577.68 Tm +(in) Tj +1 0 0 1 125.52 577.68 Tm +(a) Tj +1 0 0 1 133.08 577.68 Tm +(pre) Tj +1 0 0 1 145.68 577.68 Tm +(vious) Tj +1 0 0 1 170.28 577.68 Tm +(vie) Tj +1 0 0 1 182.28 577.68 Tm +(w) Tj +1 0 0 1 192.48 577.68 Tm +(from) Tj +1 0 0 1 214.92 577.68 Tm +(e) Tj +1 0 0 1 219.12 577.68 Tm +(v) Tj +1 0 0 1 224.04 577.68 Tm +(er) Tj +1 0 0 1 234.839 577.68 Tm +(committing.) Tj +1 0 0 1 288.719 577.68 Tm +(In) Tj +1 0 0 1 72 565.68 Tm +(the) Tj +1 0 0 1 86.0399 565.68 Tm +(second) Tj +1 0 0 1 115.56 565.68 Tm +(case) Tj +1 0 0 1 134.52 565.68 Tm +(no) Tj +1 0 0 1 146.4 565.68 Tm +(replica) Tj +1 0 0 1 175.32 565.68 Tm +(in) Tj +1 0 0 1 184.92 565.68 Tm +(the) Tj +1 0 0 1 198.959 565.68 Tm +(ne) Tj +1 0 0 1 208.199 565.68 Tm +(w) Tj +1 0 0 1 217.199 565.68 Tm +(vie) Tj +1 0 0 1 229.199 565.68 Tm +(w) Tj +1 0 0 1 238.079 565.68 Tm +(will) Tj +1 0 0 1 255.479 565.68 Tm +(accept) Tj +1 0 0 1 282.719 565.68 Tm +(an) Tj +1 0 0 1 292.079 565.68 Tm +(y) Tj +1 0 0 1 72 553.8 Tm +(message) Tj +1 0 0 1 108.84 553.8 Tm +(with) Tj +1 0 0 1 129.72 553.8 Tm +(sequence) Tj +1 0 0 1 169.44 553.8 Tm +(number) Tj +1 0 0 1 203.04 553.8 Tm +(lo) Tj +1 0 0 1 210.6 553.8 Tm +(wer) Tj +1 0 0 1 228.6 553.8 Tm +(than) Tj +ET +q +5.4 0 0 -4.6 249.1 558.3 cm +/R606 Do +Q +BT +1 0 0 1 254.879 553.8 Tm +(.) Tj +1 0 0 1 262.919 553.8 Tm +(In) Tj +1 0 0 1 274.319 553.8 Tm +(either) Tj +1 0 0 1 72 541.8 Tm +(case,) Tj +1 0 0 1 94.4399 541.8 Tm +(the) Tj +1 0 0 1 109.44 541.8 Tm +(replicas) Tj +1 0 0 1 143.16 541.8 Tm +(will) Tj +1 0 0 1 161.4 541.8 Tm +(agree) Tj +1 0 0 1 185.76 541.8 Tm +(on) Tj +1 0 0 1 198.48 541.8 Tm +(the) Tj +1 0 0 1 213.48 541.8 Tm +(request) Tj +1 0 0 1 244.919 541.8 Tm +(that) Tj +1 0 0 1 262.679 541.8 Tm +(commits) Tj +1 0 0 1 72 529.92 Tm +(locally) Tj +1 0 0 1 101.64 529.92 Tm +(with) Tj +1 0 0 1 121.8 529.92 Tm +(sequence) Tj +1 0 0 1 160.8 529.92 Tm +(number) Tj +ET +q +5.4 0 0 -4.6 194 534.4 cm +/R606 Do +Q +BT +1 0 0 1 199.8 529.92 Tm +(.) Tj +ET +endstream +endobj +609 0 obj +3806 +endobj +610 0 obj +<< /Type /Font /Name /R610 /Subtype /Type1 /BaseFont /Times-Bold >> +endobj +611 0 obj +<< /Length 612 0 R >> +stream +BT +/R610 10 Tf +1 0 0 1 72 500.4 Tm +(4.5.2) Tj +1 0 0 1 101.88 500.4 Tm +(Li) Tj +1 0 0 1 111.12 500.4 Tm +(v) Tj +1 0 0 1 116.04 500.4 Tm +(eness) Tj +/R583 10 Tf +1 0 0 1 72 483.24 Tm +(T) Tj +1 0 0 1 77.28 483.24 Tm +(o) Tj +1 0 0 1 84.7199 483.24 Tm +(pro) Tj +1 0 0 1 98.0399 483.24 Tm +(vide) Tj +1 0 0 1 117.48 483.24 Tm +(li) Tj +1 0 0 1 122.76 483.24 Tm +(v) Tj +1 0 0 1 127.68 483.24 Tm +(eness,) Tj +1 0 0 1 154.08 483.24 Tm +(replicas) Tj +1 0 0 1 187.44 483.24 Tm +(must) Tj +1 0 0 1 209.28 483.24 Tm +(mo) Tj +1 0 0 1 222 483.24 Tm +(v) Tj +1 0 0 1 226.92 483.24 Tm +(e) Tj +1 0 0 1 233.519 483.24 Tm +(to) Tj +1 0 0 1 243.719 483.24 Tm +(a) Tj +1 0 0 1 250.559 483.24 Tm +(ne) Tj +1 0 0 1 259.799 483.24 Tm +(w) Tj +1 0 0 1 269.399 483.24 Tm +(vie) Tj +1 0 0 1 281.399 483.24 Tm +(w) Tj +1 0 0 1 290.879 483.24 Tm +(if) Tj +1 0 0 1 72 471.36 Tm +(the) Tj +1 0 0 1 84.1199 471.36 Tm +(y) Tj +1 0 0 1 91.9199 471.36 Tm +(are) Tj +1 0 0 1 106.92 471.36 Tm +(unable) Tj +1 0 0 1 136.32 471.36 Tm +(to) Tj +1 0 0 1 146.88 471.36 Tm +(e) Tj +1 0 0 1 151.2 471.36 Tm +(x) Tj +1 0 0 1 156.12 471.36 Tm +(ecute) Tj +1 0 0 1 179.88 471.36 Tm +(a) Tj +1 0 0 1 187.2 471.36 Tm +(request.) Tj +1 0 0 1 223.079 471.36 Tm +(But) Tj +1 0 0 1 240.359 471.36 Tm +(it) Tj +1 0 0 1 248.759 471.36 Tm +(is) Tj +1 0 0 1 258.239 471.36 Tm +(important) Tj +1 0 0 1 72 459.36 Tm +(to) Tj +1 0 0 1 83.88 459.36 Tm +(maximize) Tj +1 0 0 1 127.32 459.36 Tm +(the) Tj +1 0 0 1 143.64 459.36 Tm +(period) Tj +1 0 0 1 173.28 459.36 Tm +(of) Tj +1 0 0 1 185.76 459.36 Tm +(time) Tj +1 0 0 1 207.6 459.36 Tm +(when) Tj +1 0 0 1 233.28 459.36 Tm +(at) Tj +1 0 0 1 244.679 459.36 Tm +(least) Tj +1 0 0 1 267.119 459.36 Tm +(2) Tj +ET +q +5 0 0 -9.2 272.6 466.5 cm +/R590 Do +Q +q +6.6 0 0 -6.5 281.7 465.2 cm +/R594 Do +Q +BT +1 0 0 1 291.959 459.36 Tm +(1) Tj +1 0 0 1 72 447.48 Tm +(non-f) Tj +1 0 0 1 93.7199 447.48 Tm +(aulty) Tj +1 0 0 1 116.76 447.48 Tm +(replicas) Tj +1 0 0 1 151.2 447.48 Tm +(are) Tj +1 0 0 1 166.8 447.48 Tm +(in) Tj +1 0 0 1 177.96 447.48 Tm +(the) Tj +1 0 0 1 193.56 447.48 Tm +(same) Tj +1 0 0 1 217.44 447.48 Tm +(vie) Tj +1 0 0 1 229.439 447.48 Tm +(w) Tj +1 0 0 1 236.039 447.48 Tm +(,) Tj +1 0 0 1 242.039 447.48 Tm +(and) Tj +1 0 0 1 259.799 447.48 Tm +(to) Tj +1 0 0 1 270.959 447.48 Tm +(ensure) Tj +1 0 0 1 72 435.48 Tm +(that) Tj +1 0 0 1 88.6799 435.48 Tm +(this) Tj +1 0 0 1 104.88 435.48 Tm +(period) Tj +1 0 0 1 132.12 435.48 Tm +(of) Tj +1 0 0 1 142.2 435.48 Tm +(time) Tj +1 0 0 1 161.76 435.48 Tm +(increases) Tj +1 0 0 1 200.04 435.48 Tm +(e) Tj +1 0 0 1 204.36 435.48 Tm +(xponentially) Tj +1 0 0 1 255.959 435.48 Tm +(until) Tj +1 0 0 1 275.999 435.48 Tm +(some) Tj +1 0 0 1 72 423.6 Tm +(requested) Tj +1 0 0 1 112.56 423.6 Tm +(operation) Tj +1 0 0 1 152.64 423.6 Tm +(e) Tj +1 0 0 1 156.96 423.6 Tm +(x) Tj +1 0 0 1 161.88 423.6 Tm +(ecutes.) Tj +1 0 0 1 192.72 423.6 Tm +(W) Tj +1 0 0 1 201.24 423.6 Tm +(e) Tj +1 0 0 1 208.2 423.6 Tm +(achie) Tj +1 0 0 1 229.08 423.6 Tm +(v) Tj +1 0 0 1 234 423.6 Tm +(e) Tj +1 0 0 1 240.719 423.6 Tm +(these) Tj +1 0 0 1 263.519 423.6 Tm +(goals) Tj +1 0 0 1 287.039 423.6 Tm +(by) Tj +1 0 0 1 72 411.6 Tm +(three) Tj +1 0 0 1 94.4399 411.6 Tm +(means.) Tj +1 0 0 1 81.96 398.16 Tm +(First,) Tj +1 0 0 1 104.64 398.16 Tm +(to) Tj +1 0 0 1 114.12 398.16 Tm +(a) Tj +1 0 0 1 118.32 398.16 Tm +(v) Tj +1 0 0 1 123.12 398.16 Tm +(oid) Tj +1 0 0 1 137.52 398.16 Tm +(starting) Tj +1 0 0 1 169.2 398.16 Tm +(a) Tj +1 0 0 1 175.32 398.16 Tm +(vie) Tj +1 0 0 1 187.32 398.16 Tm +(w) Tj +1 0 0 1 196.2 398.16 Tm +(change) Tj +1 0 0 1 226.079 398.16 Tm +(too) Tj +1 0 0 1 240.599 398.16 Tm +(soon,) Tj +1 0 0 1 263.759 398.16 Tm +(a) Tj +1 0 0 1 269.879 398.16 Tm +(replica) Tj +1 0 0 1 72 386.16 Tm +(that) Tj +1 0 0 1 90.7199 386.16 Tm +(multicasts) Tj +1 0 0 1 134.88 386.16 Tm +(a) Tj +1 0 0 1 143.04 386.16 Tm +(vie) Tj +1 0 0 1 155.04 386.16 Tm +(w-change) Tj +1 0 0 1 197.52 386.16 Tm +(message) Tj +1 0 0 1 234.96 386.16 Tm +(for) Tj +1 0 0 1 250.32 386.16 Tm +(vie) Tj +1 0 0 1 262.319 386.16 Tm +(w) Tj +ET +endstream +endobj +612 0 obj +3847 +endobj +613 0 obj +<< /Type /XObject /Name /R613 /Subtype /Image /Length 614 0 R +/ImageMask true /Width 44 /Height 46 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 44 /BlackIs1 true >>] +>> +stream +2$jBq +&9R,hkFhTE_G0fRqg[T's0obU5O\AciSX+=pUN/*r',Z@_Z&BgfC8fKf@@X?]DE=DTC?6WJ-Z.*~> +endstream +endobj +614 0 obj +84 +endobj +615 0 obj +<< /Length 616 0 R >> +stream +q +4.4 0 0 -4.6 273.4 390.7 cm +/R613 Do +Q +q +6.6 0 0 -6.5 281.9 392 cm +/R594 Do +Q +BT +1 0 0 1 291.959 386.16 Tm +(1) Tj +1 0 0 1 72 374.28 Tm +(w) Tj +1 0 0 1 79.08 374.28 Tm +(aits) Tj +1 0 0 1 96 374.28 Tm +(for) Tj +1 0 0 1 110.76 374.28 Tm +(2) Tj +ET +q +5 0 0 -9.2 116.2 381.4 cm +/R590 Do +Q +q +6.6 0 0 -6.5 124.7 380.1 cm +/R594 Do +Q +BT +1 0 0 1 134.52 374.28 Tm +(1) Tj +1 0 0 1 142.56 374.28 Tm +(vie) Tj +1 0 0 1 154.56 374.28 Tm +(w-change) Tj +1 0 0 1 196.32 374.28 Tm +(messages) Tj +1 0 0 1 237.12 374.28 Tm +(for) Tj +1 0 0 1 251.759 374.28 Tm +(vie) Tj +1 0 0 1 263.759 374.28 Tm +(w) Tj +ET +q +4.4 0 0 -4.6 274.2 378.8 cm +/R613 Do +Q +q +6.6 0 0 -6.5 282.3 380.1 cm +/R594 Do +Q +BT +1 0 0 1 291.959 374.28 Tm +(1) Tj +1 0 0 1 72 362.28 Tm +(and) Tj +1 0 0 1 90.7199 362.28 Tm +(then) Tj +1 0 0 1 112.08 362.28 Tm +(starts) Tj +1 0 0 1 137.4 362.28 Tm +(its) Tj +1 0 0 1 151.08 362.28 Tm +(timer) Tj +1 0 0 1 176.4 362.28 Tm +(to) Tj +1 0 0 1 188.4 362.28 Tm +(e) Tj +1 0 0 1 192.72 362.28 Tm +(xpire) Tj +1 0 0 1 217.44 362.28 Tm +(after) Tj +1 0 0 1 239.999 362.28 Tm +(some) Tj +1 0 0 1 265.319 362.28 Tm +(time) Tj +ET +endstream +endobj +616 0 obj +1075 +endobj +617 0 obj +<< /Type /XObject /Name /R617 /Subtype /Image /Length 618 0 R +/ImageMask true /Width 67 /Height 69 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 67 /BlackIs1 true >>] +>> +stream +,DNu& +o;Eol^[L[l^Aiucrr7K*n,*#In%[i\jON<1s53j3J,=`QLZ.\$qtn/.E;L1?qfLL$)^G[f,C]\)!<~> +endstream +endobj +618 0 obj +86 +endobj +619 0 obj +<< /Length 620 0 R >> +stream +q +6.7 0 0 -6.9 287.5 369.2 cm +/R617 Do +Q +BT +1 0 0 1 294.599 362.28 Tm +(.) Tj +1 0 0 1 72 350.4 Tm +(If) Tj +1 0 0 1 81.96 350.4 Tm +(the) Tj +1 0 0 1 97.5599 350.4 Tm +(timer) Tj +1 0 0 1 122.04 350.4 Tm +(e) Tj +1 0 0 1 126.36 350.4 Tm +(xpires) Tj +1 0 0 1 153.96 350.4 Tm +(before) Tj +1 0 0 1 182.76 350.4 Tm +(it) Tj +1 0 0 1 191.76 350.4 Tm +(recei) Tj +1 0 0 1 210.959 350.4 Tm +(v) Tj +1 0 0 1 215.879 350.4 Tm +(es) Tj +1 0 0 1 227.399 350.4 Tm +(a) Tj +1 0 0 1 235.199 350.4 Tm +(v) Tj +1 0 0 1 239.999 350.4 Tm +(alid) Tj +1 0 0 1 258.239 350.4 Tm +(ne) Tj +1 0 0 1 267.479 350.4 Tm +(w-vie) Tj +1 0 0 1 290.039 350.4 Tm +(w) Tj +1 0 0 1 72 338.4 Tm +(message) Tj +1 0 0 1 108.84 338.4 Tm +(for) Tj +ET +q +4.4 0 0 -4.6 123.7 342.9 cm +/R613 Do +Q +q +6.6 0 0 -6.5 131.7 344.2 cm +/R594 Do +Q +BT +1 0 0 1 141.36 338.4 Tm +(1) Tj +1 0 0 1 149.4 338.4 Tm +(or) Tj +1 0 0 1 160.68 338.4 Tm +(before) Tj +1 0 0 1 189.24 338.4 Tm +(it) Tj +1 0 0 1 197.759 338.4 Tm +(e) Tj +1 0 0 1 202.079 338.4 Tm +(x) Tj +1 0 0 1 206.999 338.4 Tm +(ecutes) Tj +1 0 0 1 234.839 338.4 Tm +(a) Tj +1 0 0 1 242.279 338.4 Tm +(request) Tj +1 0 0 1 274.079 338.4 Tm +(in) Tj +1 0 0 1 284.879 338.4 Tm +(the) Tj +1 0 0 1 72 326.52 Tm +(ne) Tj +1 0 0 1 81.24 326.52 Tm +(w) Tj +1 0 0 1 90.8399 326.52 Tm +(vie) Tj +1 0 0 1 102.84 326.52 Tm +(w) Tj +1 0 0 1 112.32 326.52 Tm +(that) Tj +1 0 0 1 129.72 326.52 Tm +(it) Tj +1 0 0 1 137.76 326.52 Tm +(had) Tj +1 0 0 1 154.56 326.52 Tm +(not) Tj +1 0 0 1 169.68 326.52 Tm +(e) Tj +1 0 0 1 174 326.52 Tm +(x) Tj +1 0 0 1 178.92 326.52 Tm +(ecuted) Tj +1 0 0 1 207.24 326.52 Tm +(pre) Tj +1 0 0 1 219.839 326.52 Tm +(viously) Tj +1 0 0 1 248.759 326.52 Tm +(,) Tj +1 0 0 1 253.319 326.52 Tm +(it) Tj +1 0 0 1 261.359 326.52 Tm +(starts) Tj +1 0 0 1 284.879 326.52 Tm +(the) Tj +1 0 0 1 72 314.52 Tm +(vie) Tj +1 0 0 1 84 314.52 Tm +(w) Tj +1 0 0 1 93.8399 314.52 Tm +(change) Tj +1 0 0 1 124.68 314.52 Tm +(for) Tj +1 0 0 1 138.96 314.52 Tm +(vie) Tj +1 0 0 1 150.96 314.52 Tm +(w) Tj +ET +q +4.4 0 0 -4.6 161 319 cm +/R613 Do +Q +q +6.6 0 0 -6.5 168.7 320.3 cm +/R594 Do +Q +BT +1 0 0 1 178.32 314.52 Tm +(2) Tj +1 0 0 1 186 314.52 Tm +(b) Tj +1 0 0 1 190.8 314.52 Tm +(ut) Tj +1 0 0 1 201.24 314.52 Tm +(this) Tj +1 0 0 1 218.28 314.52 Tm +(time) Tj +1 0 0 1 238.679 314.52 Tm +(it) Tj +1 0 0 1 246.839 314.52 Tm +(will) Tj +1 0 0 1 265.079 314.52 Tm +(w) Tj +1 0 0 1 272.159 314.52 Tm +(ait) Tj +1 0 0 1 284.759 314.52 Tm +(2) Tj +ET +q +6.7 0 0 -6.9 290 321.4 cm +/R617 Do +Q +BT +1 0 0 1 72 302.64 Tm +(before) Tj +1 0 0 1 99.9599 302.64 Tm +(starting) Tj +1 0 0 1 132.36 302.64 Tm +(a) Tj +1 0 0 1 139.2 302.64 Tm +(vie) Tj +1 0 0 1 151.2 302.64 Tm +(w) Tj +1 0 0 1 160.8 302.64 Tm +(change) Tj +1 0 0 1 191.52 302.64 Tm +(for) Tj +1 0 0 1 205.679 302.64 Tm +(vie) Tj +1 0 0 1 217.679 302.64 Tm +(w) Tj +ET +q +4.4 0 0 -4.6 227.5 307.1 cm +/R613 Do +Q +q +6.6 0 0 -6.5 235.1 308.4 cm +/R594 Do +Q +BT +1 0 0 1 244.679 302.64 Tm +(3.) Tj +1 0 0 1 81.96 289.08 Tm +(Second,) Tj +1 0 0 1 116.28 289.08 Tm +(if) Tj +1 0 0 1 124.8 289.08 Tm +(a) Tj +1 0 0 1 131.64 289.08 Tm +(replica) Tj +1 0 0 1 161.16 289.08 Tm +(recei) Tj +1 0 0 1 180.36 289.08 Tm +(v) Tj +1 0 0 1 185.28 289.08 Tm +(es) Tj +1 0 0 1 195.96 289.08 Tm +(a) Tj +1 0 0 1 202.8 289.08 Tm +(set) Tj +1 0 0 1 216.359 289.08 Tm +(of) Tj +ET +q +5 0 0 -9.2 227.4 296.2 cm +/R590 Do +Q +q +6.6 0 0 -6.5 235.6 294.9 cm +/R594 Do +Q +BT +1 0 0 1 245.039 289.08 Tm +(1) Tj +1 0 0 1 252.479 289.08 Tm +(v) Tj +1 0 0 1 257.279 289.08 Tm +(alid) Tj +1 0 0 1 274.559 289.08 Tm +(vie) Tj +1 0 0 1 286.559 289.08 Tm +(w-) Tj +1 0 0 1 72 277.2 Tm +(change) Tj +1 0 0 1 103.92 277.2 Tm +(messages) Tj +1 0 0 1 145.32 277.2 Tm +(from) Tj +1 0 0 1 168.36 277.2 Tm +(other) Tj +1 0 0 1 192.6 277.2 Tm +(replicas) Tj +1 0 0 1 227.28 277.2 Tm +(for) Tj +1 0 0 1 242.639 277.2 Tm +(vie) Tj +1 0 0 1 254.639 277.2 Tm +(ws) Tj +1 0 0 1 269.399 277.2 Tm +(greater) Tj +1 0 0 1 72 265.2 Tm +(than) Tj +1 0 0 1 93.2399 265.2 Tm +(its) Tj +1 0 0 1 106.68 265.2 Tm +(current) Tj +1 0 0 1 138.96 265.2 Tm +(vie) Tj +1 0 0 1 150.96 265.2 Tm +(w) Tj +1 0 0 1 157.56 265.2 Tm +(,) Tj +1 0 0 1 164.28 265.2 Tm +(it) Tj +1 0 0 1 173.88 265.2 Tm +(sends) Tj +1 0 0 1 200.04 265.2 Tm +(a) Tj +1 0 0 1 208.56 265.2 Tm +(vie) Tj +1 0 0 1 220.559 265.2 Tm +(w-change) Tj +1 0 0 1 263.279 265.2 Tm +(message) Tj +1 0 0 1 72 253.32 Tm +(for) Tj +1 0 0 1 87.9599 253.32 Tm +(the) Tj +1 0 0 1 104.4 253.32 Tm +(smallest) Tj +1 0 0 1 141.36 253.32 Tm +(vie) Tj +1 0 0 1 153.36 253.32 Tm +(w) Tj +1 0 0 1 164.76 253.32 Tm +(in) Tj +1 0 0 1 176.76 253.32 Tm +(the) Tj +1 0 0 1 193.32 253.32 Tm +(set,) Tj +1 0 0 1 211.559 253.32 Tm +(e) Tj +1 0 0 1 215.759 253.32 Tm +(v) Tj +1 0 0 1 220.679 253.32 Tm +(en) Tj +1 0 0 1 234.239 253.32 Tm +(if) Tj +1 0 0 1 244.679 253.32 Tm +(its) Tj +1 0 0 1 258.359 253.32 Tm +(timer) Tj +1 0 0 1 283.679 253.32 Tm +(has) Tj +1 0 0 1 72 241.32 Tm +(not) Tj +1 0 0 1 87.96 241.32 Tm +(e) Tj +1 0 0 1 92.28 241.32 Tm +(xpired;) Tj +1 0 0 1 123.96 241.32 Tm +(this) Tj +1 0 0 1 141.48 241.32 Tm +(pre) Tj +1 0 0 1 154.08 241.32 Tm +(v) Tj +1 0 0 1 159 241.32 Tm +(ents) Tj +1 0 0 1 178.08 241.32 Tm +(it) Tj +1 0 0 1 186.72 241.32 Tm +(from) Tj +1 0 0 1 209.28 241.32 Tm +(starting) Tj +1 0 0 1 242.399 241.32 Tm +(the) Tj +1 0 0 1 257.639 241.32 Tm +(ne) Tj +1 0 0 1 266.999 241.32 Tm +(xt) Tj +1 0 0 1 277.919 241.32 Tm +(vie) Tj +1 0 0 1 289.919 241.32 Tm +(w) Tj +1 0 0 1 72 229.32 Tm +(change) Tj +1 0 0 1 102.72 229.32 Tm +(too) Tj +1 0 0 1 117.96 229.32 Tm +(late.) Tj +1 0 0 1 81.96 215.88 Tm +(Third,) Tj +1 0 0 1 110.64 215.88 Tm +(f) Tj +1 0 0 1 113.88 215.88 Tm +(aulty) Tj +1 0 0 1 137.64 215.88 Tm +(replicas) Tj +1 0 0 1 172.32 215.88 Tm +(are) Tj +1 0 0 1 188.28 215.88 Tm +(unable) Tj +1 0 0 1 218.64 215.88 Tm +(to) Tj +1 0 0 1 230.159 215.88 Tm +(impede) Tj +1 0 0 1 263.279 215.88 Tm +(progress) Tj +1 0 0 1 72 203.88 Tm +(by) Tj +1 0 0 1 87 203.88 Tm +(forcing) Tj +1 0 0 1 120.96 203.88 Tm +(frequent) Tj +1 0 0 1 159.24 203.88 Tm +(vie) Tj +1 0 0 1 171.24 203.88 Tm +(w) Tj +1 0 0 1 183.36 203.88 Tm +(changes.) Tj +1 0 0 1 229.32 203.88 Tm +(A) Tj +1 0 0 1 241.68 203.88 Tm +(f) Tj +1 0 0 1 244.92 203.88 Tm +(aulty) Tj +1 0 0 1 269.879 203.88 Tm +(replica) Tj +1 0 0 1 72 192 Tm +(cannot) Tj +1 0 0 1 102.12 192 Tm +(cause) Tj +1 0 0 1 127.8 192 Tm +(a) Tj +1 0 0 1 135.84 192 Tm +(vie) Tj +1 0 0 1 147.84 192 Tm +(w) Tj +1 0 0 1 158.52 192 Tm +(change) Tj +1 0 0 1 190.32 192 Tm +(by) Tj +1 0 0 1 203.76 192 Tm +(sending) Tj +1 0 0 1 238.32 192 Tm +(a) Tj +1 0 0 1 246.359 192 Tm +(vie) Tj +1 0 0 1 258.359 192 Tm +(w-change) Tj +1 0 0 1 72 180 Tm +(message,) Tj +1 0 0 1 111.84 180 Tm +(because) Tj +1 0 0 1 146.64 180 Tm +(a) Tj +1 0 0 1 154.44 180 Tm +(vie) Tj +1 0 0 1 166.44 180 Tm +(w) Tj +1 0 0 1 176.88 180 Tm +(change) Tj +1 0 0 1 208.44 180 Tm +(will) Tj +1 0 0 1 227.28 180 Tm +(happen) Tj +1 0 0 1 259.32 180 Tm +(only) Tj +1 0 0 1 280.439 180 Tm +(if) Tj +1 0 0 1 289.799 180 Tm +(at) Tj +1 0 0 1 72 168.12 Tm +(least) Tj +ET +q +5 0 0 -9.2 94.1 175.2 cm +/R590 Do +Q +q +6.6 0 0 -6.5 103 173.9 cm +/R594 Do +Q +BT +1 0 0 1 112.92 168.12 Tm +(1) Tj +1 0 0 1 121.32 168.12 Tm +(replicas) Tj +1 0 0 1 155.76 168.12 Tm +(send) Tj +1 0 0 1 177.48 168.12 Tm +(vie) Tj +1 0 0 1 189.48 168.12 Tm +(w-change) Tj +1 0 0 1 231.6 168.12 Tm +(messages,) Tj +1 0 0 1 275.519 168.12 Tm +(b) Tj +1 0 0 1 280.319 168.12 Tm +(ut) Tj +1 0 0 1 291.479 168.12 Tm +(it) Tj +1 0 0 1 72 156.12 Tm +(can) Tj +1 0 0 1 88.9199 156.12 Tm +(cause) Tj +1 0 0 1 114.12 156.12 Tm +(a) Tj +1 0 0 1 121.56 156.12 Tm +(vie) Tj +1 0 0 1 133.56 156.12 Tm +(w) Tj +1 0 0 1 143.76 156.12 Tm +(change) Tj +1 0 0 1 175.08 156.12 Tm +(when) Tj +1 0 0 1 199.68 156.12 Tm +(it) Tj +1 0 0 1 208.32 156.12 Tm +(is) Tj +1 0 0 1 218.04 156.12 Tm +(the) Tj +1 0 0 1 233.279 156.12 Tm +(primary) Tj +1 0 0 1 267.959 156.12 Tm +(\(by) Tj +1 0 0 1 284.279 156.12 Tm +(not) Tj +1 0 0 1 72 144.24 Tm +(sending) Tj +1 0 0 1 106.2 144.24 Tm +(messages) Tj +1 0 0 1 146.88 144.24 Tm +(or) Tj +1 0 0 1 158.4 144.24 Tm +(sending) Tj +1 0 0 1 192.48 144.24 Tm +(bad) Tj +1 0 0 1 210 144.24 Tm +(messages\).) Tj +1 0 0 1 259.079 144.24 Tm +(Ho) Tj +1 0 0 1 271.079 144.24 Tm +(we) Tj +1 0 0 1 282.479 144.24 Tm +(v) Tj +1 0 0 1 287.399 144.24 Tm +(er) Tj +1 0 0 1 294.839 144.24 Tm +(,) Tj +1 0 0 1 72 132.24 Tm +(because) Tj +1 0 0 1 106.56 132.24 Tm +(the) Tj +1 0 0 1 121.68 132.24 Tm +(primary) Tj +1 0 0 1 156.24 132.24 Tm +(of) Tj +1 0 0 1 167.52 132.24 Tm +(vie) Tj +1 0 0 1 179.52 132.24 Tm +(w) Tj +ET +q +4.4 0 0 -4.6 189.9 136.7 cm +/R613 Do +Q +BT +1 0 0 1 197.88 132.24 Tm +(is) Tj +1 0 0 1 207.48 132.24 Tm +(the) Tj +1 0 0 1 222.72 132.24 Tm +(replica) Tj +ET +endstream +endobj +620 0 obj +8223 +endobj +621 0 obj +<< /Type /XObject /Name /R621 /Subtype /Image /Length 622 0 R +/ImageMask true /Width 54 /Height 64 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 54 /BlackIs1 true >>] +>> +stream +-D98r4=0b"4s0Orro]!o&:3^Ke&7l2 +[/(7*rS^,RIfB4oLV,SG]>+(NrPSU5I/eEj^Aki-s4fFGn+Z5e95',@28).TTZ(Ru?\8Ii-iec'~> +endstream +endobj +622 0 obj +109 +endobj +623 0 obj +<< /Length 624 0 R >> +stream +q +5.4 0 0 -6.4 252.4 136.7 cm +/R621 Do +Q +BT +1 0 0 1 260.759 132.24 Tm +(such) Tj +1 0 0 1 281.999 132.24 Tm +(that) Tj +ET +q +5.4 0 0 -6.4 71.6 124.9 cm +/R621 Do +Q +endstream +endobj +624 0 obj +159 +endobj +625 0 obj +<< /Type /XObject /Name /R625 /Subtype /Image /Length 626 0 R +/ImageMask true /Width 66 /Height 27 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 66 /BlackIs1 true >>] +>> +stream +/e%CP4)j*Wq&BoPm6CdS5Q~> +endstream +endobj +626 0 obj +25 +endobj +627 0 obj +<< /Length 628 0 R >> +stream +q +6.6 0 0 -2.7 81.4 124.3 cm +/R625 Do +Q +q +4.4 0 0 -4.6 92.7 124.9 cm +/R613 Do +Q +endstream +endobj +628 0 obj +80 +endobj +629 0 obj +<< /Type /Font /Name /R629 /Subtype /Type1 /BaseFont /Times-Italic >> +endobj +630 0 obj +<< /Length 631 0 R >> +stream +BT +/R629 10 Tf +1 0 0 1 101.52 120.36 Tm +(mod) Tj +ET +endstream +endobj +631 0 obj +52 +endobj +632 0 obj +<< /Type /XObject /Name /R632 /Subtype /Image /Length 633 0 R +/ImageMask true /Width 6 /Height 100 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 6 /BlackIs1 true >>] +>> +stream +02qoNs8W-!s8W-!s8W-!s8W-!s8W-!s8Uf'#QTA~> +endstream +endobj +633 0 obj +42 +endobj +634 0 obj +<< /Length 635 0 R >> +stream +q +0.6 0 0 -10 123.5 127.9 cm +/R632 Do +Q +endstream +endobj +635 0 obj +40 +endobj +636 0 obj +<< /Type /XObject /Name /R636 /Subtype /Image /Length 637 0 R +/ImageMask true /Width 82 /Height 72 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 82 /BlackIs1 true >>] +>> +stream +-Y/ktlnh'J_1WlVIN>d;OF5k,Do-F.&+/,KrBCW.%mf0XjHsl[XfmH0hq-EWKY$j0J0P"&G5OH\ +%NI1uW$_fE0RLEdqg:VHE;eVks1FCucp-FBHiJ$ama`KH4`C&`>7V$N*[EFX"s^Mb8oRjd(Qr1T!"]/~> +endstream +endobj +637 0 obj +158 +endobj +638 0 obj +<< /Length 639 0 R >> +stream +q +8.2 0 0 -7.2 125.4 127.3 cm +/R636 Do +Q +q +0.6 0 0 -10 134.7 127.9 cm +/R632 Do +Q +BT +/R583 10 Tf +1 0 0 1 136.44 120.36 Tm +(,) Tj +1 0 0 1 142.44 120.36 Tm +(the) Tj +1 0 0 1 157.92 120.36 Tm +(primary) Tj +1 0 0 1 192.72 120.36 Tm +(cannot) Tj +1 0 0 1 222.6 120.36 Tm +(be) Tj +1 0 0 1 235.2 120.36 Tm +(f) Tj +1 0 0 1 238.439 120.36 Tm +(aulty) Tj +1 0 0 1 261.599 120.36 Tm +(for) Tj +1 0 0 1 276.479 120.36 Tm +(more) Tj +1 0 0 1 72 108.36 Tm +(than) Tj +ET +q +5 0 0 -9.2 92.1 115.5 cm +/R590 Do +Q +BT +1 0 0 1 100.08 108.36 Tm +(consecuti) Tj +1 0 0 1 137.64 108.36 Tm +(v) Tj +1 0 0 1 142.56 108.36 Tm +(e) Tj +1 0 0 1 149.16 108.36 Tm +(vie) Tj +1 0 0 1 161.16 108.36 Tm +(ws.) Tj +1 0 0 1 81.96 94.92 Tm +(These) Tj +1 0 0 1 112.44 94.92 Tm +(three) Tj +1 0 0 1 139.08 94.92 Tm +(techniques) Tj +1 0 0 1 188.28 94.92 Tm +(guarantee) Tj +1 0 0 1 233.76 94.92 Tm +(li) Tj +1 0 0 1 239.04 94.92 Tm +(v) Tj +1 0 0 1 243.96 94.92 Tm +(eness) Tj +1 0 0 1 272.039 94.92 Tm +(unless) Tj +1 0 0 1 72 82.92 Tm +(message) Tj +1 0 0 1 111.24 82.92 Tm +(delays) Tj +1 0 0 1 142.2 82.92 Tm +(gro) Tj +1 0 0 1 155.4 82.92 Tm +(w) Tj +1 0 0 1 167.88 82.92 Tm +(f) Tj +1 0 0 1 171.12 82.92 Tm +(aster) Tj +1 0 0 1 195.36 82.92 Tm +(than) Tj +1 0 0 1 217.92 82.92 Tm +(the) Tj +1 0 0 1 235.68 82.92 Tm +(timeout) Tj +1 0 0 1 271.559 82.92 Tm +(period) Tj +1 0 0 1 72 71.04 Tm +(inde\256nitely) Tj +1 0 0 1 117 71.04 Tm +(,) Tj +1 0 0 1 121.68 71.04 Tm +(which) Tj +1 0 0 1 148.56 71.04 Tm +(is) Tj +1 0 0 1 157.68 71.04 Tm +(unlik) Tj +1 0 0 1 178.2 71.04 Tm +(ely) Tj +1 0 0 1 192.72 71.04 Tm +(in) Tj +1 0 0 1 203.04 71.04 Tm +(a) Tj +1 0 0 1 209.88 71.04 Tm +(real) Tj +1 0 0 1 227.279 71.04 Tm +(system.) Tj +/R610 10 Tf +1 0 0 1 315 709.08 Tm +(4.6) Tj +1 0 0 1 337.44 709.08 Tm +(Non-Determinism) Tj +/R583 10 Tf +1 0 0 1 315 693.84 Tm +(State) Tj +1 0 0 1 338.52 693.84 Tm +(machine) Tj +1 0 0 1 375.84 693.84 Tm +(replicas) Tj +1 0 0 1 410.4 693.84 Tm +(must) Tj +1 0 0 1 433.32 693.84 Tm +(be) Tj +1 0 0 1 446.28 693.84 Tm +(deterministic) Tj +1 0 0 1 501.959 693.84 Tm +(b) Tj +1 0 0 1 506.759 693.84 Tm +(ut) Tj +1 0 0 1 518.039 693.84 Tm +(man) Tj +1 0 0 1 535.199 693.84 Tm +(y) Tj +1 0 0 1 315 681.96 Tm +(services) Tj +1 0 0 1 351.48 681.96 Tm +(in) Tj +1 0 0 1 358.92 681.96 Tm +(v) Tj +1 0 0 1 363.72 681.96 Tm +(olv) Tj +1 0 0 1 376.44 681.96 Tm +(e) Tj +1 0 0 1 384.96 681.96 Tm +(some) Tj +1 0 0 1 410.28 681.96 Tm +(form) Tj +1 0 0 1 433.92 681.96 Tm +(of) Tj +1 0 0 1 446.64 681.96 Tm +(non-determinism.) Tj +1 0 0 1 526.319 681.96 Tm +(F) Tj +1 0 0 1 531.719 681.96 Tm +(or) Tj +1 0 0 1 315 669.96 Tm +(e) Tj +1 0 0 1 319.32 669.96 Tm +(xample,) Tj +1 0 0 1 353.52 669.96 Tm +(the) Tj +1 0 0 1 368.16 669.96 Tm +(time-last-modi\256ed) Tj +1 0 0 1 444.24 669.96 Tm +(in) Tj +1 0 0 1 454.44 669.96 Tm +(NFS) Tj +1 0 0 1 475.079 669.96 Tm +(is) Tj +1 0 0 1 484.199 669.96 Tm +(set) Tj +1 0 0 1 497.759 669.96 Tm +(by) Tj +1 0 0 1 510.119 669.96 Tm +(reading) Tj +1 0 0 1 315 658.08 Tm +(the) Tj +1 0 0 1 330 658.08 Tm +(serv) Tj +1 0 0 1 346.56 658.08 Tm +(er') Tj +1 0 0 1 357.12 658.08 Tm +(s) Tj +1 0 0 1 363.6 658.08 Tm +(local) Tj +1 0 0 1 385.8 658.08 Tm +(clock;) Tj +1 0 0 1 413.16 658.08 Tm +(if) Tj +1 0 0 1 422.04 658.08 Tm +(this) Tj +1 0 0 1 439.2 658.08 Tm +(were) Tj +1 0 0 1 461.399 658.08 Tm +(done) Tj +1 0 0 1 483.599 658.08 Tm +(independently) Tj +1 0 0 1 315 646.08 Tm +(at) Tj +1 0 0 1 325.92 646.08 Tm +(each) Tj +1 0 0 1 347.88 646.08 Tm +(replica,) Tj +1 0 0 1 381.6 646.08 Tm +(the) Tj +1 0 0 1 397.44 646.08 Tm +(states) Tj +1 0 0 1 423.36 646.08 Tm +(of) Tj +1 0 0 1 435.36 646.08 Tm +(non-f) Tj +1 0 0 1 457.079 646.08 Tm +(aulty) Tj +1 0 0 1 480.479 646.08 Tm +(replicas) Tj +1 0 0 1 515.159 646.08 Tm +(w) Tj +1 0 0 1 522.239 646.08 Tm +(ould) Tj +1 0 0 1 315 634.2 Tm +(di) Tj +1 0 0 1 322.56 634.2 Tm +(v) Tj +1 0 0 1 327.48 634.2 Tm +(er) Tj +1 0 0 1 335.16 634.2 Tm +(ge.) Tj +1 0 0 1 352.92 634.2 Tm +(Therefore,) Tj +1 0 0 1 398.28 634.2 Tm +(some) Tj +1 0 0 1 422.64 634.2 Tm +(mechanism) Tj +1 0 0 1 471.36 634.2 Tm +(to) Tj +1 0 0 1 482.4 634.2 Tm +(ensure) Tj +1 0 0 1 511.8 634.2 Tm +(that) Tj +1 0 0 1 530.039 634.2 Tm +(all) Tj +1 0 0 1 315 622.2 Tm +(replicas) Tj +1 0 0 1 348.84 622.2 Tm +(select) Tj +1 0 0 1 374.4 622.2 Tm +(the) Tj +1 0 0 1 389.4 622.2 Tm +(same) Tj +1 0 0 1 412.8 622.2 Tm +(v) Tj +1 0 0 1 417.6 622.2 Tm +(alue) Tj +1 0 0 1 436.92 622.2 Tm +(is) Tj +1 0 0 1 446.52 622.2 Tm +(needed.) Tj +1 0 0 1 481.919 622.2 Tm +(In) Tj +1 0 0 1 493.079 622.2 Tm +(general,) Tj +1 0 0 1 527.879 622.2 Tm +(the) Tj +1 0 0 1 315 610.32 Tm +(client) Tj +1 0 0 1 340.92 610.32 Tm +(cannot) Tj +1 0 0 1 371.16 610.32 Tm +(select) Tj +1 0 0 1 397.56 610.32 Tm +(the) Tj +1 0 0 1 413.52 610.32 Tm +(v) Tj +1 0 0 1 418.32 610.32 Tm +(alue) Tj +1 0 0 1 438.6 610.32 Tm +(because) Tj +1 0 0 1 473.759 610.32 Tm +(it) Tj +1 0 0 1 483.119 610.32 Tm +(does) Tj +1 0 0 1 505.079 610.32 Tm +(not) Tj +1 0 0 1 521.519 610.32 Tm +(ha) Tj +1 0 0 1 530.759 610.32 Tm +(v) Tj +1 0 0 1 535.679 610.32 Tm +(e) Tj +1 0 0 1 315 598.32 Tm +(enough) Tj +1 0 0 1 347.16 598.32 Tm +(information;) Tj +1 0 0 1 399.96 598.32 Tm +(for) Tj +1 0 0 1 414.36 598.32 Tm +(e) Tj +1 0 0 1 418.68 598.32 Tm +(xample,) Tj +1 0 0 1 453.36 598.32 Tm +(it) Tj +1 0 0 1 461.76 598.32 Tm +(does) Tj +1 0 0 1 482.879 598.32 Tm +(not) Tj +1 0 0 1 498.359 598.32 Tm +(kno) Tj +1 0 0 1 513.239 598.32 Tm +(w) Tj +1 0 0 1 523.079 598.32 Tm +(ho) Tj +1 0 0 1 532.919 598.32 Tm +(w) Tj +1 0 0 1 315 586.32 Tm +(its) Tj +1 0 0 1 326.76 586.32 Tm +(request) Tj +1 0 0 1 357.96 586.32 Tm +(will) Tj +1 0 0 1 375.84 586.32 Tm +(be) Tj +1 0 0 1 387.6 586.32 Tm +(ordered) Tj +1 0 0 1 420.36 586.32 Tm +(relati) Tj +1 0 0 1 440.64 586.32 Tm +(v) Tj +1 0 0 1 445.559 586.32 Tm +(e) Tj +1 0 0 1 452.279 586.32 Tm +(to) Tj +1 0 0 1 462.359 586.32 Tm +(concurrent) Tj +1 0 0 1 507.359 586.32 Tm +(requests) Tj +1 0 0 1 315 574.44 Tm +(by) Tj +1 0 0 1 327.6 574.44 Tm +(other) Tj +1 0 0 1 350.76 574.44 Tm +(clients.) Tj +1 0 0 1 383.52 574.44 Tm +(Instead,) Tj +1 0 0 1 417.48 574.44 Tm +(the) Tj +1 0 0 1 432.36 574.44 Tm +(primary) Tj +1 0 0 1 466.679 574.44 Tm +(needs) Tj +1 0 0 1 491.999 574.44 Tm +(to) Tj +1 0 0 1 502.439 574.44 Tm +(select) Tj +1 0 0 1 527.879 574.44 Tm +(the) Tj +1 0 0 1 315 562.44 Tm +(v) Tj +1 0 0 1 319.8 562.44 Tm +(alue) Tj +1 0 0 1 339.6 562.44 Tm +(either) Tj +1 0 0 1 365.64 562.44 Tm +(independently) Tj +1 0 0 1 425.4 562.44 Tm +(or) Tj +1 0 0 1 436.92 562.44 Tm +(based) Tj +1 0 0 1 462.96 562.44 Tm +(on) Tj +1 0 0 1 476.279 562.44 Tm +(v) Tj +1 0 0 1 481.079 562.44 Tm +(alues) Tj +1 0 0 1 504.719 562.44 Tm +(pro) Tj +1 0 0 1 518.039 562.44 Tm +(vided) Tj +1 0 0 1 315 550.56 Tm +(by) Tj +1 0 0 1 327.48 550.56 Tm +(the) Tj +1 0 0 1 342.12 550.56 Tm +(backups.) Tj +1 0 0 1 324.96 538.08 Tm +(If) Tj +1 0 0 1 333.6 538.08 Tm +(the) Tj +1 0 0 1 347.76 538.08 Tm +(primary) Tj +1 0 0 1 381.24 538.08 Tm +(selects) Tj +1 0 0 1 409.8 538.08 Tm +(the) Tj +1 0 0 1 423.96 538.08 Tm +(non-deterministic) Tj +1 0 0 1 496.199 538.08 Tm +(v) Tj +1 0 0 1 500.999 538.08 Tm +(alue) Tj +1 0 0 1 519.479 538.08 Tm +(inde-) Tj +1 0 0 1 315 526.2 Tm +(pendently) Tj +1 0 0 1 354 526.2 Tm +(,) Tj +1 0 0 1 359.64 526.2 Tm +(it) Tj +1 0 0 1 368.4 526.2 Tm +(concatenates) Tj +1 0 0 1 422.52 526.2 Tm +(the) Tj +1 0 0 1 438 526.2 Tm +(v) Tj +1 0 0 1 442.8 526.2 Tm +(alue) Tj +1 0 0 1 462.599 526.2 Tm +(with) Tj +1 0 0 1 483.599 526.2 Tm +(the) Tj +1 0 0 1 499.079 526.2 Tm +(associated) Tj +1 0 0 1 315 514.2 Tm +(request) Tj +1 0 0 1 347.04 514.2 Tm +(and) Tj +1 0 0 1 364.68 514.2 Tm +(e) Tj +1 0 0 1 369 514.2 Tm +(x) Tj +1 0 0 1 373.92 514.2 Tm +(ecutes) Tj +1 0 0 1 402 514.2 Tm +(the) Tj +1 0 0 1 417.36 514.2 Tm +(three) Tj +1 0 0 1 440.52 514.2 Tm +(phase) Tj +1 0 0 1 466.56 514.2 Tm +(protocol) Tj +1 0 0 1 503.039 514.2 Tm +(to) Tj +1 0 0 1 513.959 514.2 Tm +(ensure) Tj +1 0 0 1 315 502.2 Tm +(that) Tj +1 0 0 1 332.88 502.2 Tm +(non-f) Tj +1 0 0 1 354.6 502.2 Tm +(aulty) Tj +1 0 0 1 377.16 502.2 Tm +(replicas) Tj +1 0 0 1 411 502.2 Tm +(agree) Tj +1 0 0 1 435.48 502.2 Tm +(on) Tj +1 0 0 1 448.32 502.2 Tm +(a) Tj +1 0 0 1 455.64 502.2 Tm +(sequence) Tj +1 0 0 1 494.999 502.2 Tm +(number) Tj +1 0 0 1 528.359 502.2 Tm +(for) Tj +1 0 0 1 315 490.32 Tm +(the) Tj +1 0 0 1 328.92 490.32 Tm +(request) Tj +1 0 0 1 359.52 490.32 Tm +(and) Tj +1 0 0 1 375.72 490.32 Tm +(v) Tj +1 0 0 1 380.52 490.32 Tm +(alue.) Tj +1 0 0 1 402.36 490.32 Tm +(This) Tj +1 0 0 1 421.8 490.32 Tm +(pre) Tj +1 0 0 1 434.4 490.32 Tm +(v) Tj +1 0 0 1 439.32 490.32 Tm +(ents) Tj +1 0 0 1 457.079 490.32 Tm +(a) Tj +1 0 0 1 463.199 490.32 Tm +(f) Tj +1 0 0 1 466.439 490.32 Tm +(aulty) Tj +1 0 0 1 488.159 490.32 Tm +(pr) Tj +1 0 0 1 496.439 490.32 Tm +(imary) Tj +1 0 0 1 521.399 490.32 Tm +(f) Tj +1 0 0 1 524.639 490.32 Tm +(ro) Tj +1 0 0 1 532.919 490.32 Tm +(m) Tj +1 0 0 1 315 478.32 Tm +(causing) Tj +1 0 0 1 348.12 478.32 Tm +(replica) Tj +1 0 0 1 378 478.32 Tm +(state) Tj +1 0 0 1 399 478.32 Tm +(to) Tj +1 0 0 1 409.56 478.32 Tm +(di) Tj +1 0 0 1 417.12 478.32 Tm +(v) Tj +1 0 0 1 422.04 478.32 Tm +(er) Tj +1 0 0 1 429.72 478.32 Tm +(ge) Tj +1 0 0 1 441.479 478.32 Tm +(by) Tj +1 0 0 1 454.199 478.32 Tm +(sending) Tj +1 0 0 1 487.919 478.32 Tm +(dif) Tj +1 0 0 1 498.839 478.32 Tm +(ferent) Tj +1 0 0 1 524.759 478.32 Tm +(v) Tj +1 0 0 1 529.559 478.32 Tm +(al-) Tj +1 0 0 1 315 466.44 Tm +(ues) Tj +1 0 0 1 330.12 466.44 Tm +(to) Tj +1 0 0 1 339.6 466.44 Tm +(dif) Tj +1 0 0 1 350.52 466.44 Tm +(ferent) Tj +1 0 0 1 375.48 466.44 Tm +(replicas.) Tj +1 0 0 1 412.32 466.44 Tm +(Ho) Tj +1 0 0 1 424.32 466.44 Tm +(we) Tj +1 0 0 1 435.72 466.44 Tm +(v) Tj +1 0 0 1 440.64 466.44 Tm +(er) Tj +1 0 0 1 448.079 466.44 Tm +(,) Tj +1 0 0 1 452.279 466.44 Tm +(a) Tj +1 0 0 1 458.399 466.44 Tm +(f) Tj +1 0 0 1 461.639 466.44 Tm +(aulty) Tj +1 0 0 1 483.359 466.44 Tm +(primary) Tj +1 0 0 1 516.719 466.44 Tm +(might) Tj +1 0 0 1 315 454.44 Tm +(send) Tj +1 0 0 1 335.52 454.44 Tm +(the) Tj +1 0 0 1 349.92 454.44 Tm +(same,) Tj +1 0 0 1 375.24 454.44 Tm +(incorrect,) Tj +1 0 0 1 415.44 454.44 Tm +(v) Tj +1 0 0 1 420.24 454.44 Tm +(alue) Tj +1 0 0 1 438.96 454.44 Tm +(to) Tj +1 0 0 1 449.04 454.44 Tm +(all) Tj +1 0 0 1 461.279 454.44 Tm +(replicas.) Tj +1 0 0 1 498.239 454.44 Tm +(Therefore,) Tj +1 0 0 1 315 442.56 Tm +(replicas) Tj +1 0 0 1 348.24 442.56 Tm +(must) Tj +1 0 0 1 369.84 442.56 Tm +(be) Tj +1 0 0 1 381.6 442.56 Tm +(able) Tj +1 0 0 1 400.44 442.56 Tm +(to) Tj +1 0 0 1 410.4 442.56 Tm +(decide) Tj +1 0 0 1 438.72 442.56 Tm +(deterministically) Tj +1 0 0 1 507.959 442.56 Tm +(whether) Tj +1 0 0 1 315 430.56 Tm +(the) Tj +1 0 0 1 329.28 430.56 Tm +(v) Tj +1 0 0 1 334.08 430.56 Tm +(alue) Tj +1 0 0 1 352.8 430.56 Tm +(is) Tj +1 0 0 1 361.56 430.56 Tm +(correct) Tj +1 0 0 1 391.32 430.56 Tm +(\(and) Tj +1 0 0 1 411.12 430.56 Tm +(what) Tj +1 0 0 1 432.6 430.56 Tm +(to) Tj +1 0 0 1 442.56 430.56 Tm +(do) Tj +1 0 0 1 454.679 430.56 Tm +(if) Tj +1 0 0 1 462.839 430.56 Tm +(it) Tj +1 0 0 1 470.519 430.56 Tm +(is) Tj +1 0 0 1 479.279 430.56 Tm +(not\)) Tj +1 0 0 1 497.519 430.56 Tm +(based) Tj +1 0 0 1 522.239 430.56 Tm +(only) Tj +1 0 0 1 315 418.68 Tm +(on) Tj +1 0 0 1 327.48 418.68 Tm +(the) Tj +1 0 0 1 342.12 418.68 Tm +(service) Tj +1 0 0 1 372.84 418.68 Tm +(state.) Tj +1 0 0 1 324.96 406.2 Tm +(This) Tj +1 0 0 1 345.24 406.2 Tm +(protocol) Tj +1 0 0 1 381 406.2 Tm +(is) Tj +1 0 0 1 390.24 406.2 Tm +(adequate) Tj +1 0 0 1 428.28 406.2 Tm +(for) Tj +1 0 0 1 442.44 406.2 Tm +(most) Tj +1 0 0 1 464.399 406.2 Tm +(services) Tj +1 0 0 1 499.079 406.2 Tm +(\(including) Tj +1 0 0 1 315 394.2 Tm +(NFS\)) Tj +1 0 0 1 343.44 394.2 Tm +(b) Tj +1 0 0 1 348.24 394.2 Tm +(ut) Tj +1 0 0 1 362.76 394.2 Tm +(occasionally) Tj +1 0 0 1 419.4 394.2 Tm +(replicas) Tj +1 0 0 1 457.2 394.2 Tm +(must) Tj +1 0 0 1 483.36 394.2 Tm +(participate) Tj +1 0 0 1 532.199 394.2 Tm +(in) Tj +1 0 0 1 315 382.32 Tm +(selecting) Tj +1 0 0 1 354.72 382.32 Tm +(the) Tj +1 0 0 1 371.16 382.32 Tm +(v) Tj +1 0 0 1 375.96 382.32 Tm +(alue) Tj +1 0 0 1 396.84 382.32 Tm +(to) Tj +1 0 0 1 408.96 382.32 Tm +(satisfy) Tj +1 0 0 1 439.199 382.32 Tm +(a) Tj +1 0 0 1 447.959 382.32 Tm +(service') Tj +1 0 0 1 479.039 382.32 Tm +(s) Tj +1 0 0 1 487.199 382.32 Tm +(speci\256cation.) Tj +1 0 0 1 315 370.32 Tm +(This) Tj +1 0 0 1 336.36 370.32 Tm +(can) Tj +1 0 0 1 353.76 370.32 Tm +(be) Tj +1 0 0 1 366.84 370.32 Tm +(accomplished) Tj +1 0 0 1 425.16 370.32 Tm +(by) Tj +1 0 0 1 438.84 370.32 Tm +(adding) Tj +1 0 0 1 469.56 370.32 Tm +(an) Tj +1 0 0 1 482.519 370.32 Tm +(e) Tj +1 0 0 1 486.839 370.32 Tm +(xtra) Tj +1 0 0 1 505.919 370.32 Tm +(phase) Tj +1 0 0 1 532.199 370.32 Tm +(to) Tj +1 0 0 1 315 358.44 Tm +(the) Tj +1 0 0 1 331.08 358.44 Tm +(protocol:) Tj +1 0 0 1 373.56 358.44 Tm +(the) Tj +1 0 0 1 389.64 358.44 Tm +(primary) Tj +1 0 0 1 425.04 358.44 Tm +(obtains) Tj +1 0 0 1 457.8 358.44 Tm +(authenticated) Tj +1 0 0 1 514.799 358.44 Tm +(v) Tj +1 0 0 1 519.599 358.44 Tm +(alues) Tj +1 0 0 1 315 346.44 Tm +(proposed) Tj +1 0 0 1 354.84 346.44 Tm +(by) Tj +1 0 0 1 368.04 346.44 Tm +(the) Tj +1 0 0 1 383.52 346.44 Tm +(backups,) Tj +1 0 0 1 422.04 346.44 Tm +(concatenates) Tj +1 0 0 1 476.279 346.44 Tm +(2) Tj +ET +q +5 0 0 -9.2 481.7 353.5 cm +/R590 Do +Q +q +6.6 0 0 -6.5 490.3 352.2 cm +/R594 Do +Q +BT +1 0 0 1 500.279 346.44 Tm +(1) Tj +1 0 0 1 508.559 346.44 Tm +(of) Tj +1 0 0 1 520.079 346.44 Tm +(them) Tj +1 0 0 1 315 334.56 Tm +(with) Tj +1 0 0 1 336.72 334.56 Tm +(the) Tj +1 0 0 1 352.92 334.56 Tm +(associated) Tj +1 0 0 1 397.92 334.56 Tm +(request,) Tj +1 0 0 1 433.68 334.56 Tm +(and) Tj +1 0 0 1 452.04 334.56 Tm +(starts) Tj +1 0 0 1 477.119 334.56 Tm +(the) Tj +1 0 0 1 493.319 334.56 Tm +(three) Tj +1 0 0 1 517.319 334.56 Tm +(phase) Tj +1 0 0 1 315 322.56 Tm +(protocol) Tj +1 0 0 1 350.88 322.56 Tm +(for) Tj +1 0 0 1 365.28 322.56 Tm +(the) Tj +1 0 0 1 380.16 322.56 Tm +(concatenated) Tj +1 0 0 1 434.88 322.56 Tm +(message.) Tj +1 0 0 1 475.319 322.56 Tm +(Replicas) Tj +1 0 0 1 512.279 322.56 Tm +(choose) Tj +1 0 0 1 315 310.68 Tm +(the) Tj +1 0 0 1 330.24 310.68 Tm +(v) Tj +1 0 0 1 335.04 310.68 Tm +(alue) Tj +1 0 0 1 354.6 310.68 Tm +(by) Tj +1 0 0 1 367.68 310.68 Tm +(a) Tj +1 0 0 1 375.12 310.68 Tm +(deterministic) Tj +1 0 0 1 430.2 310.68 Tm +(computation) Tj +1 0 0 1 482.999 310.68 Tm +(on) Tj +1 0 0 1 496.079 310.68 Tm +(the) Tj +1 0 0 1 511.319 310.68 Tm +(2) Tj +ET +q +5 0 0 -9.2 516.8 317.8 cm +/R590 Do +Q +q +6.6 0 0 -6.5 525.3 316.5 cm +/R594 Do +Q +BT +1 0 0 1 534.959 310.68 Tm +(1) Tj +1 0 0 1 315 298.68 Tm +(v) Tj +1 0 0 1 319.8 298.68 Tm +(alues) Tj +1 0 0 1 342.84 298.68 Tm +(and) Tj +1 0 0 1 359.88 298.68 Tm +(their) Tj +1 0 0 1 380.76 298.68 Tm +(state,) Tj +1 0 0 1 404.28 298.68 Tm +(e.g.,) Tj +1 0 0 1 423.84 298.68 Tm +(taking) Tj +1 0 0 1 451.32 298.68 Tm +(the) Tj +1 0 0 1 466.199 298.68 Tm +(median.) Tj +1 0 0 1 502.079 298.68 Tm +(The) Tj +1 0 0 1 520.199 298.68 Tm +(e) Tj +1 0 0 1 524.519 298.68 Tm +(xtra) Tj +1 0 0 1 315 286.8 Tm +(phase) Tj +1 0 0 1 340.8 286.8 Tm +(can) Tj +1 0 0 1 357.72 286.8 Tm +(be) Tj +1 0 0 1 370.2 286.8 Tm +(optimized) Tj +1 0 0 1 413.16 286.8 Tm +(a) Tj +1 0 0 1 417.48 286.8 Tm +(w) Tj +1 0 0 1 424.56 286.8 Tm +(ay) Tj +1 0 0 1 437.04 286.8 Tm +(in) Tj +1 0 0 1 447.96 286.8 Tm +(the) Tj +1 0 0 1 463.199 286.8 Tm +(common) Tj +1 0 0 1 501.239 286.8 Tm +(case.) Tj +1 0 0 1 526.319 286.8 Tm +(F) Tj +1 0 0 1 531.719 286.8 Tm +(or) Tj +1 0 0 1 315 274.8 Tm +(e) Tj +1 0 0 1 319.32 274.8 Tm +(xample,) Tj +1 0 0 1 354.24 274.8 Tm +(if) Tj +1 0 0 1 363.24 274.8 Tm +(replicas) Tj +1 0 0 1 397.2 274.8 Tm +(need) Tj +1 0 0 1 419.04 274.8 Tm +(a) Tj +1 0 0 1 426.48 274.8 Tm +(v) Tj +1 0 0 1 431.28 274.8 Tm +(alue) Tj +1 0 0 1 450.84 274.8 Tm +(that) Tj +1 0 0 1 468.719 274.8 Tm +(is) Tj +1 0 0 1 478.319 274.8 Tm +(\252close) Tj +1 0 0 1 506.279 274.8 Tm +(enough\272) Tj +1 0 0 1 315 262.92 Tm +(to) Tj +1 0 0 1 324.84 262.92 Tm +(that) Tj +1 0 0 1 341.88 262.92 Tm +(of) Tj +1 0 0 1 352.32 262.92 Tm +(their) Tj +1 0 0 1 372.72 262.92 Tm +(local) Tj +1 0 0 1 394.2 262.92 Tm +(clock,) Tj +1 0 0 1 420.48 262.92 Tm +(the) Tj +1 0 0 1 434.76 262.92 Tm +(e) Tj +1 0 0 1 439.079 262.92 Tm +(xtra) Tj +1 0 0 1 456.599 262.92 Tm +(phase) Tj +1 0 0 1 481.439 262.92 Tm +(can) Tj +1 0 0 1 497.399 262.92 Tm +(be) Tj +1 0 0 1 508.799 262.92 Tm +(a) Tj +1 0 0 1 512.999 262.92 Tm +(v) Tj +1 0 0 1 517.799 262.92 Tm +(oided) Tj +1 0 0 1 315 250.92 Tm +(when) Tj +1 0 0 1 339.12 250.92 Tm +(their) Tj +1 0 0 1 359.88 250.92 Tm +(clocks) Tj +1 0 0 1 387.72 250.92 Tm +(are) Tj +1 0 0 1 402.48 250.92 Tm +(synchronized) Tj +1 0 0 1 458.04 250.92 Tm +(within) Tj +1 0 0 1 485.999 250.92 Tm +(some) Tj +1 0 0 1 509.519 250.92 Tm +(delta.) Tj +/R610 12 Tf +1 0 0 1 315 222.96 Tm +(5) Tj +1 0 0 1 332.88 222.96 Tm +(Optimizations) Tj +/R583 10 Tf +1 0 0 1 315 205.44 Tm +(This) Tj +1 0 0 1 335.88 205.44 Tm +(section) Tj +1 0 0 1 367.2 205.44 Tm +(describes) Tj +1 0 0 1 407.4 205.44 Tm +(some) Tj +1 0 0 1 431.64 205.44 Tm +(optimizations) Tj +1 0 0 1 488.999 205.44 Tm +(that) Tj +1 0 0 1 507.119 205.44 Tm +(impro) Tj +1 0 0 1 530.999 205.44 Tm +(v) Tj +1 0 0 1 535.919 205.44 Tm +(e) Tj +1 0 0 1 315 193.56 Tm +(the) Tj +1 0 0 1 331.92 193.56 Tm +(performance) Tj +1 0 0 1 387 193.56 Tm +(of) Tj +1 0 0 1 400.08 193.56 Tm +(the) Tj +1 0 0 1 417 193.56 Tm +(algorithm) Tj +1 0 0 1 460.56 193.56 Tm +(during) Tj +1 0 0 1 491.279 193.56 Tm +(normal-case) Tj +1 0 0 1 315 181.56 Tm +(operation.) Tj +1 0 0 1 363.24 181.56 Tm +(All) Tj +1 0 0 1 380.04 181.56 Tm +(the) Tj +1 0 0 1 396.24 181.56 Tm +(optimizations) Tj +1 0 0 1 454.44 181.56 Tm +(preserv) Tj +1 0 0 1 483.839 181.56 Tm +(e) Tj +1 0 0 1 492.119 181.56 Tm +(the) Tj +1 0 0 1 508.319 181.56 Tm +(li) Tj +1 0 0 1 513.599 181.56 Tm +(v) Tj +1 0 0 1 518.519 181.56 Tm +(eness) Tj +1 0 0 1 315 169.68 Tm +(and) Tj +1 0 0 1 331.92 169.68 Tm +(safety) Tj +1 0 0 1 358.2 169.68 Tm +(properties.) Tj +/R610 10 Tf +1 0 0 1 315 145.92 Tm +(5.1) Tj +1 0 0 1 337.44 145.92 Tm +(Reducing) Tj +1 0 0 1 380.28 145.92 Tm +(Communication) Tj +/R583 10 Tf +1 0 0 1 315 130.68 Tm +(W) Tj +1 0 0 1 323.52 130.68 Tm +(e) Tj +1 0 0 1 334.8 130.68 Tm +(use) Tj +1 0 0 1 354.72 130.68 Tm +(three) Tj +1 0 0 1 381.36 130.68 Tm +(optimizations) Tj +1 0 0 1 442.32 130.68 Tm +(to) Tj +1 0 0 1 456.84 130.68 Tm +(reduce) Tj +1 0 0 1 490.079 130.68 Tm +(the) Tj +1 0 0 1 508.919 130.68 Tm +(cost) Tj +1 0 0 1 531.719 130.68 Tm +(of) Tj +1 0 0 1 315 118.8 Tm +(communication.) Tj +1 0 0 1 389.76 118.8 Tm +(The) Tj +1 0 0 1 409.92 118.8 Tm +(\256rst) Tj +1 0 0 1 430.2 118.8 Tm +(a) Tj +1 0 0 1 434.4 118.8 Tm +(v) Tj +1 0 0 1 439.2 118.8 Tm +(oids) Tj +1 0 0 1 460.44 118.8 Tm +(sending) Tj +1 0 0 1 496.2 118.8 Tm +(most) Tj +1 0 0 1 520.319 118.8 Tm +(lar) Tj +1 0 0 1 530.759 118.8 Tm +(ge) Tj +1 0 0 1 315 106.8 Tm +(replies.) Tj +1 0 0 1 352.92 106.8 Tm +(A) Tj +1 0 0 1 364.44 106.8 Tm +(client) Tj +1 0 0 1 390.84 106.8 Tm +(request) Tj +1 0 0 1 423.84 106.8 Tm +(designates) Tj +1 0 0 1 469.679 106.8 Tm +(a) Tj +1 0 0 1 478.319 106.8 Tm +(replica) Tj +1 0 0 1 509.759 106.8 Tm +(to) Tj +1 0 0 1 521.759 106.8 Tm +(send) Tj +1 0 0 1 315 94.92 Tm +(the) Tj +1 0 0 1 330.24 94.92 Tm +(result;) Tj +1 0 0 1 358.56 94.92 Tm +(all) Tj +1 0 0 1 371.64 94.92 Tm +(other) Tj +1 0 0 1 395.28 94.92 Tm +(replicas) Tj +1 0 0 1 429.36 94.92 Tm +(send) Tj +1 0 0 1 450.72 94.92 Tm +(replies) Tj +1 0 0 1 480.479 94.92 Tm +(containing) Tj +1 0 0 1 525.599 94.92 Tm +(just) Tj +1 0 0 1 315 82.92 Tm +(the) Tj +1 0 0 1 330.6 82.92 Tm +(digest) Tj +1 0 0 1 357.72 82.92 Tm +(of) Tj +1 0 0 1 369.48 82.92 Tm +(the) Tj +1 0 0 1 384.96 82.92 Tm +(result.) Tj +1 0 0 1 415.92 82.92 Tm +(The) Tj +1 0 0 1 434.76 82.92 Tm +(digests) Tj +1 0 0 1 465.84 82.92 Tm +(allo) Tj +1 0 0 1 480.599 82.92 Tm +(w) Tj +1 0 0 1 491.159 82.92 Tm +(the) Tj +1 0 0 1 506.759 82.92 Tm +(client) Tj +1 0 0 1 532.199 82.92 Tm +(to) Tj +1 0 0 1 315 71.04 Tm +(check) Tj +1 0 0 1 339.96 71.04 Tm +(the) Tj +1 0 0 1 353.76 71.04 Tm +(correctness) Tj +1 0 0 1 400.32 71.04 Tm +(of) Tj +1 0 0 1 410.28 71.04 Tm +(the) Tj +1 0 0 1 424.08 71.04 Tm +(result) Tj +1 0 0 1 447.84 71.04 Tm +(while) Tj +1 0 0 1 471.599 71.04 Tm +(reducin) Tj +1 0 0 1 501.599 71.04 Tm +(g) Tj +1 0 0 1 508.199 71.04 Tm +(n) Tj +1 0 0 1 513.119 71.04 Tm +(etw) Tj +1 0 0 1 527.399 71.04 Tm +(or) Tj +1 0 0 1 535.679 71.04 Tm +(k) Tj +1 0 0 1 303.48 35.16 Tm +(7) Tj +ET +Q +endstream +endobj +639 0 obj +19357 +endobj +640 0 obj +<< +/Type /Page +/MediaBox [0 0 612 792] +/Parent 2 0 R +/Resources << /ProcSet [/PDF /ImageB /Text] +/Font << +/R629 629 0 R +/R610 610 0 R +/R583 583 0 R +>> +/XObject << +/R594 594 0 R +/R590 590 0 R +/R632 632 0 R +/R636 636 0 R +/R613 613 0 R +/R625 625 0 R +/R621 621 0 R +/R617 617 0 R +/R606 606 0 R +/R602 602 0 R +/R598 598 0 R +/R586 586 0 R +>> +>> +/Contents [ +584 0 R +588 0 R +592 0 R +596 0 R +600 0 R +604 0 R +608 0 R +611 0 R +615 0 R +619 0 R +623 0 R +627 0 R +630 0 R +634 0 R +638 0 R +] +>> +endobj +641 0 obj +<< /Type /Font /Name /R641 /Subtype /Type1 /BaseFont /Times-Roman >> +endobj +642 0 obj +<< /Length 643 0 R >> +stream +q +BT +/R641 10 Tf +1 0 0 1 72 709.08 Tm +(bandwidth) Tj +1 0 0 1 116.52 709.08 Tm +(consumption) Tj +1 0 0 1 170.4 709.08 Tm +(and) Tj +1 0 0 1 187.2 709.08 Tm +(CPU) Tj +1 0 0 1 209.04 709.08 Tm +(o) Tj +1 0 0 1 213.96 709.08 Tm +(v) Tj +1 0 0 1 218.88 709.08 Tm +(erhead) Tj +1 0 0 1 247.68 709.08 Tm +(signi\256cantly) Tj +1 0 0 1 72 697.08 Tm +(for) Tj +1 0 0 1 86.6399 697.08 Tm +(lar) Tj +1 0 0 1 97.0799 697.08 Tm +(ge) Tj +1 0 0 1 109.32 697.08 Tm +(replies.) Tj +1 0 0 1 143.52 697.08 Tm +(If) Tj +1 0 0 1 153.24 697.08 Tm +(the) Tj +1 0 0 1 168.36 697.08 Tm +(client) Tj +1 0 0 1 193.559 697.08 Tm +(does) Tj +1 0 0 1 214.799 697.08 Tm +(not) Tj +1 0 0 1 230.519 697.08 Tm +(recei) Tj +1 0 0 1 249.719 697.08 Tm +(v) Tj +1 0 0 1 254.639 697.08 Tm +(e) Tj +1 0 0 1 261.959 697.08 Tm +(a) Tj +1 0 0 1 269.399 697.08 Tm +(correct) Tj +1 0 0 1 72 685.2 Tm +(result) Tj +1 0 0 1 99.3599 685.2 Tm +(from) Tj +1 0 0 1 123.96 685.2 Tm +(the) Tj +1 0 0 1 141.24 685.2 Tm +(designated) Tj +1 0 0 1 189.12 685.2 Tm +(replica,) Tj +1 0 0 1 224.64 685.2 Tm +(it) Tj +1 0 0 1 235.32 685.2 Tm +(retransmits) Tj +1 0 0 1 284.879 685.2 Tm +(the) Tj +1 0 0 1 72 673.2 Tm +(request) Tj +1 0 0 1 105.96 673.2 Tm +(as) Tj +1 0 0 1 119.4 673.2 Tm +(usual,) Tj +1 0 0 1 148.8 673.2 Tm +(requesting) Tj +1 0 0 1 195.48 673.2 Tm +(all) Tj +1 0 0 1 210.6 673.2 Tm +(replicas) Tj +1 0 0 1 246.84 673.2 Tm +(to) Tj +1 0 0 1 259.679 673.2 Tm +(send) Tj +1 0 0 1 283.199 673.2 Tm +(full) Tj +1 0 0 1 72 661.32 Tm +(replies.) Tj +1 0 0 1 81.96 648.84 Tm +(The) Tj +1 0 0 1 104.16 648.84 Tm +(second) Tj +1 0 0 1 138.48 648.84 Tm +(optimization) Tj +1 0 0 1 195.6 648.84 Tm +(reduces) Tj +1 0 0 1 232.68 648.84 Tm +(the) Tj +1 0 0 1 251.519 648.84 Tm +(number) Tj +1 0 0 1 288.719 648.84 Tm +(of) Tj +1 0 0 1 72 636.84 Tm +(message) Tj +1 0 0 1 111.6 636.84 Tm +(delays) Tj +1 0 0 1 143.04 636.84 Tm +(for) Tj +1 0 0 1 160.56 636.84 Tm +(an) Tj +1 0 0 1 175.8 636.84 Tm +(operation) Tj +1 0 0 1 219.36 636.84 Tm +(in) Tj +1 0 0 1 226.8 636.84 Tm +(v) Tj +1 0 0 1 231.6 636.84 Tm +(ocation) Tj +1 0 0 1 266.759 636.84 Tm +(from) Tj +1 0 0 1 291.959 636.84 Tm +(5) Tj +1 0 0 1 72 624.96 Tm +(to) Tj +1 0 0 1 84.48 624.96 Tm +(4.) Tj +1 0 0 1 102.36 624.96 Tm +(Replicas) Tj +1 0 0 1 141.36 624.96 Tm +(e) Tj +1 0 0 1 145.68 624.96 Tm +(x) Tj +1 0 0 1 150.6 624.96 Tm +(ecute) Tj +1 0 0 1 176.28 624.96 Tm +(a) Tj +1 0 0 1 185.4 624.96 Tm +(request) Tj +ET +endstream +endobj +643 0 obj +2300 +endobj +644 0 obj +<< /Type /Font /Name /R644 /Subtype /Type1 /BaseFont /Times-Italic >> +endobj +645 0 obj +<< /Length 646 0 R >> +stream +BT +/R644 10 Tf +1 0 0 1 218.88 624.96 Tm +(tentatively) Tj +/R641 10 Tf +1 0 0 1 265.199 624.96 Tm +(as) Tj +1 0 0 1 278.159 624.96 Tm +(soon) Tj +1 0 0 1 72 612.96 Tm +(as) Tj +1 0 0 1 84.9599 612.96 Tm +(the) Tj +1 0 0 1 101.88 612.96 Tm +(prepared) Tj +1 0 0 1 141.48 612.96 Tm +(predicate) Tj +1 0 0 1 182.64 612.96 Tm +(holds) Tj +1 0 0 1 209.04 612.96 Tm +(for) Tj +1 0 0 1 225.359 612.96 Tm +(the) Tj +1 0 0 1 242.159 612.96 Tm +(request,) Tj +1 0 0 1 278.759 612.96 Tm +(their) Tj +1 0 0 1 72 601.08 Tm +(state) Tj +1 0 0 1 94.7999 601.08 Tm +(re\257ects) Tj +1 0 0 1 128.16 601.08 Tm +(the) Tj +1 0 0 1 144.84 601.08 Tm +(e) Tj +1 0 0 1 149.16 601.08 Tm +(x) Tj +1 0 0 1 154.08 601.08 Tm +(ecution) Tj +1 0 0 1 187.8 601.08 Tm +(of) Tj +1 0 0 1 200.64 601.08 Tm +(all) Tj +1 0 0 1 215.159 601.08 Tm +(requests) Tj +1 0 0 1 252.359 601.08 Tm +(with) Tj +1 0 0 1 274.559 601.08 Tm +(lo) Tj +1 0 0 1 282.119 601.08 Tm +(wer) Tj +1 0 0 1 72 589.08 Tm +(sequence) Tj +1 0 0 1 112.44 589.08 Tm +(number) Tj +1 0 0 1 142.8 589.08 Tm +(,) Tj +1 0 0 1 149.28 589.08 Tm +(and) Tj +1 0 0 1 167.52 589.08 Tm +(these) Tj +1 0 0 1 191.88 589.08 Tm +(requests) Tj +1 0 0 1 228.48 589.08 Tm +(are) Tj +1 0 0 1 244.56 589.08 Tm +(all) Tj +1 0 0 1 258.479 589.08 Tm +(kno) Tj +1 0 0 1 273.359 589.08 Tm +(wn) Tj +1 0 0 1 289.199 589.08 Tm +(to) Tj +1 0 0 1 72 577.2 Tm +(ha) Tj +1 0 0 1 81.24 577.2 Tm +(v) Tj +1 0 0 1 86.16 577.2 Tm +(e) Tj +1 0 0 1 92.3999 577.2 Tm +(committed.) Tj +1 0 0 1 141 577.2 Tm +(After) Tj +1 0 0 1 163.92 577.2 Tm +(e) Tj +1 0 0 1 168.24 577.2 Tm +(x) Tj +1 0 0 1 173.16 577.2 Tm +(ecuting) Tj +1 0 0 1 204.36 577.2 Tm +(the) Tj +1 0 0 1 218.52 577.2 Tm +(request,) Tj +1 0 0 1 251.879 577.2 Tm +(the) Tj +1 0 0 1 266.039 577.2 Tm +(replicas) Tj +1 0 0 1 72 565.2 Tm +(send) Tj +1 0 0 1 93.4799 565.2 Tm +(tentati) Tj +1 0 0 1 118.2 565.2 Tm +(v) Tj +1 0 0 1 123.12 565.2 Tm +(e) Tj +1 0 0 1 130.56 565.2 Tm +(replies) Tj +1 0 0 1 160.44 565.2 Tm +(to) Tj +1 0 0 1 171.36 565.2 Tm +(the) Tj +1 0 0 1 186.72 565.2 Tm +(client.) Tj +1 0 0 1 217.199 565.2 Tm +(The) Tj +1 0 0 1 235.919 565.2 Tm +(client) Tj +1 0 0 1 261.239 565.2 Tm +(w) Tj +1 0 0 1 268.319 565.2 Tm +(aits) Tj +1 0 0 1 285.359 565.2 Tm +(for) Tj +1 0 0 1 72 553.32 Tm +(2) Tj +ET +Q +q +W +0 0 612 792 re +n +endstream +endobj +646 0 obj +2166 +endobj +647 0 obj +<< /Type /XObject /Name /R647 /Subtype /Image /Length 648 0 R +/ImageMask true /Width 50 /Height 92 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 50 /BlackIs1 true >>] +>> +stream +5$eL7-b:K3p0<%? +h1u#R]2'Mug].5is7$!eqtL-hqu?EZs5/V3iQXZVs7/ihqu?EZs4QRbJe-ans*o2/4DdB_@6aqq5Q~> +endstream +endobj +648 0 obj +96 +endobj +649 0 obj +<< /Length 650 0 R >> +stream +q +5 0 0 -9.2 77.4 560.4 cm +/R647 Do +Q +endstream +endobj +650 0 obj +38 +endobj +651 0 obj +<< /Type /XObject /Name /R651 /Subtype /Image /Length 652 0 R +/ImageMask true /Width 66 /Height 65 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 66 /BlackIs1 true >>] +>> +stream +,BZ6Ns8W-!s8W-!s-Eq9Iaj?!s8W-!s8W-!s8Uf'#QTA~> +endstream +endobj +652 0 obj +47 +endobj +653 0 obj +<< /Length 654 0 R >> +stream +q +6.6 0 0 -6.5 86.7 559.1 cm +/R651 Do +Q +BT +/R641 10 Tf +1 0 0 1 97.1999 553.32 Tm +(1) Tj +1 0 0 1 106.68 553.32 Tm +(matching) Tj +1 0 0 1 148.2 553.32 Tm +(tentati) Tj +1 0 0 1 172.92 553.32 Tm +(v) Tj +1 0 0 1 177.84 553.32 Tm +(e) Tj +1 0 0 1 186.48 553.32 Tm +(replies.) Tj +1 0 0 1 224.879 553.32 Tm +(If) Tj +1 0 0 1 236.039 553.32 Tm +(it) Tj +1 0 0 1 245.999 553.32 Tm +(recei) Tj +1 0 0 1 265.199 553.32 Tm +(v) Tj +1 0 0 1 270.119 553.32 Tm +(es) Tj +1 0 0 1 282.599 553.32 Tm +(this) Tj +1 0 0 1 72 541.32 Tm +(man) Tj +1 0 0 1 89.16 541.32 Tm +(y) Tj +1 0 0 1 93.6 541.32 Tm +(,) Tj +1 0 0 1 100.32 541.32 Tm +(the) Tj +1 0 0 1 116.52 541.32 Tm +(request) Tj +1 0 0 1 149.28 541.32 Tm +(is) Tj +1 0 0 1 159.96 541.32 Tm +(guaranteed) Tj +1 0 0 1 207.72 541.32 Tm +(to) Tj +1 0 0 1 219.6 541.32 Tm +(commit) Tj +1 0 0 1 254.04 541.32 Tm +(e) Tj +1 0 0 1 258.239 541.32 Tm +(v) Tj +1 0 0 1 263.159 541.32 Tm +(entually) Tj +1 0 0 1 294.839 541.32 Tm +(.) Tj +1 0 0 1 72 529.44 Tm +(Otherwise,) Tj +1 0 0 1 119.52 529.44 Tm +(the) Tj +1 0 0 1 135.48 529.44 Tm +(client) Tj +1 0 0 1 161.4 529.44 Tm +(retransmits) Tj +1 0 0 1 209.4 529.44 Tm +(the) Tj +1 0 0 1 225.36 529.44 Tm +(request) Tj +1 0 0 1 257.879 529.44 Tm +(and) Tj +1 0 0 1 276.119 529.44 Tm +(w) Tj +1 0 0 1 283.199 529.44 Tm +(aits) Tj +1 0 0 1 72 517.44 Tm +(for) Tj +ET +q +5 0 0 -9.2 86.6 524.5 cm +/R647 Do +Q +q +6.6 0 0 -6.5 94.8 523.2 cm +/R651 Do +Q +BT +1 0 0 1 104.28 517.44 Tm +(1) Tj +1 0 0 1 111.72 517.44 Tm +(non-tentati) Tj +1 0 0 1 154.92 517.44 Tm +(v) Tj +1 0 0 1 159.84 517.44 Tm +(e) Tj +1 0 0 1 166.32 517.44 Tm +(replies.) Tj +1 0 0 1 81.96 504.96 Tm +(A) Tj +1 0 0 1 93 504.96 Tm +(request) Tj +1 0 0 1 125.52 504.96 Tm +(that) Tj +1 0 0 1 144.24 504.96 Tm +(has) Tj +1 0 0 1 161.28 504.96 Tm +(e) Tj +1 0 0 1 165.6 504.96 Tm +(x) Tj +1 0 0 1 170.52 504.96 Tm +(ecuted) Tj +1 0 0 1 200.28 504.96 Tm +(tentati) Tj +1 0 0 1 225 504.96 Tm +(v) Tj +1 0 0 1 229.92 504.96 Tm +(ely) Tj +1 0 0 1 245.639 504.96 Tm +(may) Tj +1 0 0 1 266.639 504.96 Tm +(abort) Tj +1 0 0 1 290.879 504.96 Tm +(if) Tj +1 0 0 1 72 493.08 Tm +(there) Tj +1 0 0 1 97.0799 493.08 Tm +(is) Tj +1 0 0 1 108.84 493.08 Tm +(a) Tj +1 0 0 1 118.44 493.08 Tm +(vie) Tj +1 0 0 1 130.44 493.08 Tm +(w) Tj +1 0 0 1 142.68 493.08 Tm +(change) Tj +1 0 0 1 176.04 493.08 Tm +(and) Tj +1 0 0 1 195.48 493.08 Tm +(it) Tj +1 0 0 1 206.16 493.08 Tm +(is) Tj +1 0 0 1 217.92 493.08 Tm +(replaced) Tj +1 0 0 1 256.799 493.08 Tm +(by) Tj +1 0 0 1 271.919 493.08 Tm +(a) Tj +/R644 10 Tf +1 0 0 1 281.519 493.08 Tm +(null) Tj +/R641 10 Tf +1 0 0 1 72 481.08 Tm +(request.) Tj +1 0 0 1 110.4 481.08 Tm +(In) Tj +1 0 0 1 122.4 481.08 Tm +(this) Tj +1 0 0 1 140.52 481.08 Tm +(case) Tj +1 0 0 1 161.28 481.08 Tm +(the) Tj +1 0 0 1 177.12 481.08 Tm +(replica) Tj +1 0 0 1 207.96 481.08 Tm +(re) Tj +1 0 0 1 215.519 481.08 Tm +(v) Tj +1 0 0 1 220.439 481.08 Tm +(erts) Tj +1 0 0 1 238.319 481.08 Tm +(its) Tj +1 0 0 1 251.399 481.08 Tm +(state) Tj +1 0 0 1 273.359 481.08 Tm +(to) Tj +1 0 0 1 284.879 481.08 Tm +(the) Tj +1 0 0 1 72 469.2 Tm +(last) Tj +1 0 0 1 88.7999 469.2 Tm +(stable) Tj +1 0 0 1 114.96 469.2 Tm +(checkpoint) Tj +1 0 0 1 161.52 469.2 Tm +(in) Tj +1 0 0 1 172.2 469.2 Tm +(the) Tj +1 0 0 1 187.32 469.2 Tm +(ne) Tj +1 0 0 1 196.56 469.2 Tm +(w-vie) Tj +1 0 0 1 219.12 469.2 Tm +(w) Tj +1 0 0 1 229.08 469.2 Tm +(message) Tj +1 0 0 1 265.679 469.2 Tm +(or) Tj +1 0 0 1 276.959 469.2 Tm +(to) Tj +1 0 0 1 287.639 469.2 Tm +(its) Tj +1 0 0 1 72 457.2 Tm +(last) Tj +1 0 0 1 88.4399 457.2 Tm +(checkpointed) Tj +1 0 0 1 144.12 457.2 Tm +(state) Tj +1 0 0 1 165 457.2 Tm +(\(depending) Tj +1 0 0 1 212.4 457.2 Tm +(on) Tj +1 0 0 1 225 457.2 Tm +(which) Tj +1 0 0 1 251.999 457.2 Tm +(one) Tj +1 0 0 1 268.919 457.2 Tm +(has) Tj +1 0 0 1 284.879 457.2 Tm +(the) Tj +1 0 0 1 72 445.32 Tm +(higher) Tj +1 0 0 1 99.9599 445.32 Tm +(sequence) Tj +1 0 0 1 138.96 445.32 Tm +(number\).) Tj +1 0 0 1 81.96 432.84 Tm +(The) Tj +1 0 0 1 101.28 432.84 Tm +(third) Tj +1 0 0 1 123.84 432.84 Tm +(optimization) Tj +1 0 0 1 178.08 432.84 Tm +(impro) Tj +1 0 0 1 201.96 432.84 Tm +(v) Tj +1 0 0 1 206.88 432.84 Tm +(es) Tj +1 0 0 1 218.64 432.84 Tm +(the) Tj +1 0 0 1 234.6 432.84 Tm +(performance) Tj +1 0 0 1 288.719 432.84 Tm +(of) Tj +1 0 0 1 72 420.84 Tm +(read-only) Tj +1 0 0 1 115.68 420.84 Tm +(operations) Tj +1 0 0 1 162.72 420.84 Tm +(that) Tj +1 0 0 1 183.12 420.84 Tm +(do) Tj +1 0 0 1 198.6 420.84 Tm +(not) Tj +1 0 0 1 216.84 420.84 Tm +(modify) Tj +1 0 0 1 251.159 420.84 Tm +(the) Tj +1 0 0 1 268.799 420.84 Tm +(service) Tj +1 0 0 1 72 408.96 Tm +(state.) Tj +1 0 0 1 103.68 408.96 Tm +(A) Tj +1 0 0 1 115.8 408.96 Tm +(client) Tj +1 0 0 1 142.8 408.96 Tm +(multicasts) Tj +1 0 0 1 188.16 408.96 Tm +(a) Tj +1 0 0 1 197.52 408.96 Tm +(read-only) Tj +1 0 0 1 240.599 408.96 Tm +(request) Tj +1 0 0 1 274.319 408.96 Tm +(to) Tj +1 0 0 1 287.039 408.96 Tm +(all) Tj +1 0 0 1 72 396.96 Tm +(replicas.) Tj +1 0 0 1 113.76 396.96 Tm +(Replicas) Tj +1 0 0 1 152.16 396.96 Tm +(e) Tj +1 0 0 1 156.48 396.96 Tm +(x) Tj +1 0 0 1 161.4 396.96 Tm +(ecute) Tj +1 0 0 1 186.36 396.96 Tm +(the) Tj +1 0 0 1 202.56 396.96 Tm +(request) Tj +1 0 0 1 235.319 396.96 Tm +(immediately) Tj +1 0 0 1 289.199 396.96 Tm +(in) Tj +1 0 0 1 72 385.08 Tm +(their) Tj +1 0 0 1 95.0399 385.08 Tm +(tentati) Tj +1 0 0 1 119.76 385.08 Tm +(v) Tj +1 0 0 1 124.68 385.08 Tm +(e) Tj +1 0 0 1 133.68 385.08 Tm +(state) Tj +1 0 0 1 156.6 385.08 Tm +(after) Tj +1 0 0 1 179.64 385.08 Tm +(checking) Tj +1 0 0 1 220.32 385.08 Tm +(that) Tj +1 0 0 1 239.999 385.08 Tm +(the) Tj +1 0 0 1 256.799 385.08 Tm +(request) Tj +1 0 0 1 290.399 385.08 Tm +(is) Tj +1 0 0 1 72 373.08 Tm +(properly) Tj +1 0 0 1 110.04 373.08 Tm +(authenticated,) Tj +1 0 0 1 170.4 373.08 Tm +(that) Tj +1 0 0 1 189.6 373.08 Tm +(the) Tj +1 0 0 1 206.04 373.08 Tm +(client) Tj +1 0 0 1 232.439 373.08 Tm +(has) Tj +1 0 0 1 249.959 373.08 Tm +(access,) Tj +1 0 0 1 282.599 373.08 Tm +(and) Tj +1 0 0 1 72 361.2 Tm +(that) Tj +1 0 0 1 89.6399 361.2 Tm +(the) Tj +1 0 0 1 104.52 361.2 Tm +(request) Tj +1 0 0 1 135.96 361.2 Tm +(is) Tj +1 0 0 1 145.32 361.2 Tm +(in) Tj +1 0 0 1 155.88 361.2 Tm +(f) Tj +1 0 0 1 159.12 361.2 Tm +(act) Tj +1 0 0 1 173.4 361.2 Tm +(read-only) Tj +1 0 0 1 211.32 361.2 Tm +(.) Tj +1 0 0 1 217.679 361.2 Tm +(The) Tj +1 0 0 1 233.159 361.2 Tm +(y) Tj +1 0 0 1 240.719 361.2 Tm +(send) Tj +1 0 0 1 261.599 361.2 Tm +(the) Tj +1 0 0 1 276.479 361.2 Tm +(reply) Tj +1 0 0 1 72 349.2 Tm +(only) Tj +1 0 0 1 92.1599 349.2 Tm +(after) Tj +1 0 0 1 112.8 349.2 Tm +(all) Tj +1 0 0 1 125.16 349.2 Tm +(requests) Tj +1 0 0 1 160.2 349.2 Tm +(re\257ected) Tj +1 0 0 1 196.8 349.2 Tm +(in) Tj +1 0 0 1 207 349.2 Tm +(the) Tj +1 0 0 1 221.519 349.2 Tm +(tentati) Tj +1 0 0 1 246.239 349.2 Tm +(v) Tj +1 0 0 1 251.159 349.2 Tm +(e) Tj +1 0 0 1 257.879 349.2 Tm +(state) Tj +1 0 0 1 278.519 349.2 Tm +(ha) Tj +1 0 0 1 287.759 349.2 Tm +(v) Tj +1 0 0 1 292.679 349.2 Tm +(e) Tj +1 0 0 1 72 337.32 Tm +(committed;) Tj +1 0 0 1 121.44 337.32 Tm +(this) Tj +1 0 0 1 139.44 337.32 Tm +(is) Tj +1 0 0 1 149.64 337.32 Tm +(necessary) Tj +1 0 0 1 191.88 337.32 Tm +(to) Tj +1 0 0 1 203.16 337.32 Tm +(pre) Tj +1 0 0 1 215.759 337.32 Tm +(v) Tj +1 0 0 1 220.679 337.32 Tm +(ent) Tj +1 0 0 1 236.159 337.32 Tm +(the) Tj +1 0 0 1 251.999 337.32 Tm +(client) Tj +1 0 0 1 277.679 337.32 Tm +(from) Tj +1 0 0 1 72 325.32 Tm +(observing) Tj +1 0 0 1 113.4 325.32 Tm +(uncommitted) Tj +1 0 0 1 168 325.32 Tm +(state.) Tj +1 0 0 1 192.24 325.32 Tm +(The) Tj +1 0 0 1 209.76 325.32 Tm +(client) Tj +1 0 0 1 234 325.32 Tm +(w) Tj +1 0 0 1 241.08 325.32 Tm +(aits) Tj +1 0 0 1 257.039 325.32 Tm +(for) Tj +1 0 0 1 270.719 325.32 Tm +(2) Tj +ET +q +5 0 0 -9.2 276.2 332.4 cm +/R647 Do +Q +q +6.6 0 0 -6.5 283.5 331.1 cm +/R651 Do +Q +BT +1 0 0 1 292.079 325.32 Tm +(1) Tj +1 0 0 1 72 313.44 Tm +(replies) Tj +1 0 0 1 101.28 313.44 Tm +(from) Tj +1 0 0 1 123.36 313.44 Tm +(dif) Tj +1 0 0 1 134.28 313.44 Tm +(ferent) Tj +1 0 0 1 160.2 313.44 Tm +(replicas) Tj +1 0 0 1 193.92 313.44 Tm +(with) Tj +1 0 0 1 214.44 313.44 Tm +(the) Tj +1 0 0 1 229.32 313.44 Tm +(same) Tj +1 0 0 1 252.479 313.44 Tm +(result.) Tj +1 0 0 1 281.519 313.44 Tm +(The) Tj +1 0 0 1 72 301.44 Tm +(client) Tj +1 0 0 1 95.9999 301.44 Tm +(may) Tj +1 0 0 1 115.08 301.44 Tm +(be) Tj +1 0 0 1 126.48 301.44 Tm +(unable) Tj +1 0 0 1 154.92 301.44 Tm +(to) Tj +1 0 0 1 164.52 301.44 Tm +(collect) Tj +1 0 0 1 193.08 301.44 Tm +(2) Tj +ET +q +5 0 0 -9.2 198.5 308.5 cm +/R647 Do +Q +q +6.6 0 0 -6.5 205.3 307.2 cm +/R651 Do +Q +BT +1 0 0 1 213.599 301.44 Tm +(1) Tj +1 0 0 1 220.439 301.44 Tm +(such) Tj +1 0 0 1 240.599 301.44 Tm +(replies) Tj +1 0 0 1 269.039 301.44 Tm +(if) Tj +1 0 0 1 277.079 301.44 Tm +(there) Tj +1 0 0 1 72 289.56 Tm +(are) Tj +1 0 0 1 86.6399 289.56 Tm +(concurrent) Tj +1 0 0 1 131.76 289.56 Tm +(writes) Tj +1 0 0 1 158.64 289.56 Tm +(to) Tj +1 0 0 1 168.84 289.56 Tm +(data) Tj +1 0 0 1 187.92 289.56 Tm +(that) Tj +1 0 0 1 205.319 289.56 Tm +(af) Tj +1 0 0 1 212.879 289.56 Tm +(fect) Tj +1 0 0 1 230.279 289.56 Tm +(the) Tj +1 0 0 1 244.919 289.56 Tm +(result;) Tj +1 0 0 1 272.399 289.56 Tm +(in) Tj +1 0 0 1 282.599 289.56 Tm +(this) Tj +1 0 0 1 72 277.56 Tm +(case,) Tj +1 0 0 1 96.1199 277.56 Tm +(it) Tj +1 0 0 1 105.72 277.56 Tm +(retransmits) Tj +1 0 0 1 154.08 277.56 Tm +(the) Tj +1 0 0 1 170.28 277.56 Tm +(request) Tj +1 0 0 1 203.04 277.56 Tm +(as) Tj +1 0 0 1 215.4 277.56 Tm +(a) Tj +1 0 0 1 223.92 277.56 Tm +(re) Tj +1 0 0 1 231.6 277.56 Tm +(gular) Tj +1 0 0 1 256.079 277.56 Tm +(read-write) Tj +1 0 0 1 72 265.68 Tm +(request) Tj +1 0 0 1 103.32 265.68 Tm +(after) Tj +1 0 0 1 123.96 265.68 Tm +(its) Tj +1 0 0 1 135.96 265.68 Tm +(retransmission) Tj +1 0 0 1 196.56 265.68 Tm +(timer) Tj +1 0 0 1 220.079 265.68 Tm +(e) Tj +1 0 0 1 224.399 265.68 Tm +(xpires.) Tj +ET +endstream +endobj +654 0 obj +9324 +endobj +655 0 obj +<< /Type /Font /Name /R655 /Subtype /Type1 /BaseFont /Times-Bold >> +endobj +656 0 obj +<< /Length 657 0 R >> +stream +BT +/R655 10 Tf +1 0 0 1 72 241.92 Tm +(5.2) Tj +1 0 0 1 94.4399 241.92 Tm +(Cryptograph) Tj +1 0 0 1 150.48 241.92 Tm +(y) Tj +/R641 10 Tf +1 0 0 1 72 226.8 Tm +(In) Tj +1 0 0 1 86.4 226.8 Tm +(Section) Tj +1 0 0 1 122.28 226.8 Tm +(4,) Tj +1 0 0 1 136.68 226.8 Tm +(we) Tj +1 0 0 1 154.32 226.8 Tm +(described) Tj +1 0 0 1 198.6 226.8 Tm +(an) Tj +1 0 0 1 214.08 226.8 Tm +(algorithm) Tj +1 0 0 1 258.839 226.8 Tm +(that) Tj +1 0 0 1 279.839 226.8 Tm +(uses) Tj +1 0 0 1 72 214.8 Tm +(digital) Tj +1 0 0 1 99.8399 214.8 Tm +(signatures) Tj +1 0 0 1 142.56 214.8 Tm +(to) Tj +1 0 0 1 152.64 214.8 Tm +(authenticate) Tj +1 0 0 1 203.04 214.8 Tm +(all) Tj +1 0 0 1 215.399 214.8 Tm +(messages.) Tj +1 0 0 1 259.079 214.8 Tm +(Ho) Tj +1 0 0 1 271.079 214.8 Tm +(we) Tj +1 0 0 1 282.479 214.8 Tm +(v) Tj +1 0 0 1 287.399 214.8 Tm +(er) Tj +1 0 0 1 294.839 214.8 Tm +(,) Tj +1 0 0 1 72 202.92 Tm +(we) Tj +1 0 0 1 90.8399 202.92 Tm +(actually) Tj +1 0 0 1 129.72 202.92 Tm +(use) Tj +1 0 0 1 150.24 202.92 Tm +(digital) Tj +1 0 0 1 183 202.92 Tm +(signatures) Tj +1 0 0 1 230.639 202.92 Tm +(only) Tj +1 0 0 1 255.719 202.92 Tm +(for) Tj +1 0 0 1 274.559 202.92 Tm +(vie) Tj +1 0 0 1 286.559 202.92 Tm +(w-) Tj +1 0 0 1 72 190.92 Tm +(change) Tj +1 0 0 1 103.92 190.92 Tm +(and) Tj +1 0 0 1 121.92 190.92 Tm +(ne) Tj +1 0 0 1 131.16 190.92 Tm +(w-vie) Tj +1 0 0 1 153.72 190.92 Tm +(w) Tj +1 0 0 1 164.4 190.92 Tm +(messages,) Tj +1 0 0 1 208.44 190.92 Tm +(which) Tj +1 0 0 1 236.4 190.92 Tm +(are) Tj +1 0 0 1 252.239 190.92 Tm +(sent) Tj +1 0 0 1 271.919 190.92 Tm +(rarely) Tj +1 0 0 1 294.719 190.92 Tm +(,) Tj +1 0 0 1 72 179.04 Tm +(and) Tj +1 0 0 1 92.88 179.04 Tm +(authenticate) Tj +1 0 0 1 147.6 179.04 Tm +(all) Tj +1 0 0 1 164.04 179.04 Tm +(other) Tj +1 0 0 1 191.04 179.04 Tm +(messages) Tj +1 0 0 1 235.2 179.04 Tm +(using) Tj +1 0 0 1 263.279 179.04 Tm +(message) Tj +1 0 0 1 72 167.04 Tm +(authentication) Tj +1 0 0 1 131.4 167.04 Tm +(codes) Tj +1 0 0 1 157.08 167.04 Tm +(\(MA) Tj +1 0 0 1 176.16 167.04 Tm +(Cs\).) Tj +1 0 0 1 197.4 167.04 Tm +(This) Tj +1 0 0 1 218.04 167.04 Tm +(eliminates) Tj +1 0 0 1 261.959 167.04 Tm +(the) Tj +1 0 0 1 277.079 167.04 Tm +(main) Tj +1 0 0 1 72 155.16 Tm +(performance) Tj +1 0 0 1 124.8 155.16 Tm +(bottleneck) Tj +1 0 0 1 168.84 155.16 Tm +(in) Tj +1 0 0 1 179.04 155.16 Tm +(pre) Tj +1 0 0 1 191.64 155.16 Tm +(vious) Tj +1 0 0 1 215.64 155.16 Tm +(systems) Tj +1 0 0 1 249.719 155.16 Tm +([29) Tj +1 0 0 1 262.919 155.16 Tm +(,) Tj +1 0 0 1 267.959 155.16 Tm +(22) Tj +1 0 0 1 277.919 155.16 Tm +(].) Tj +1 0 0 1 81.96 142.68 Tm +(Ho) Tj +1 0 0 1 93.96 142.68 Tm +(we) Tj +1 0 0 1 105.36 142.68 Tm +(v) Tj +1 0 0 1 110.28 142.68 Tm +(er) Tj +1 0 0 1 117.72 142.68 Tm +(,) Tj +1 0 0 1 123.72 142.68 Tm +(MA) Tj +1 0 0 1 139.44 142.68 Tm +(Cs) Tj +1 0 0 1 153.36 142.68 Tm +(ha) Tj +1 0 0 1 162.6 142.68 Tm +(v) Tj +1 0 0 1 167.52 142.68 Tm +(e) Tj +1 0 0 1 175.32 142.68 Tm +(a) Tj +1 0 0 1 183.24 142.68 Tm +(fundamental) Tj +1 0 0 1 236.52 142.68 Tm +(limitation) Tj +1 0 0 1 278.759 142.68 Tm +(rela-) Tj +1 0 0 1 72 130.68 Tm +(ti) Tj +1 0 0 1 77.28 130.68 Tm +(v) Tj +1 0 0 1 82.2 130.68 Tm +(e) Tj +1 0 0 1 90.6 130.68 Tm +(to) Tj +1 0 0 1 102.36 130.68 Tm +(digital) Tj +1 0 0 1 131.88 130.68 Tm +(signatures) Tj +1 0 0 1 176.4 130.68 Tm +(\320) Tj +1 0 0 1 190.32 130.68 Tm +(the) Tj +1 0 0 1 206.52 130.68 Tm +(inability) Tj +1 0 0 1 243.839 130.68 Tm +(to) Tj +1 0 0 1 255.599 130.68 Tm +(pro) Tj +1 0 0 1 268.919 130.68 Tm +(v) Tj +1 0 0 1 273.839 130.68 Tm +(e) Tj +1 0 0 1 281.999 130.68 Tm +(that) Tj +1 0 0 1 72 118.8 Tm +(a) Tj +1 0 0 1 80.0399 118.8 Tm +(message) Tj +1 0 0 1 117.48 118.8 Tm +(is) Tj +1 0 0 1 127.68 118.8 Tm +(authentic) Tj +1 0 0 1 167.88 118.8 Tm +(to) Tj +1 0 0 1 179.28 118.8 Tm +(a) Tj +1 0 0 1 187.32 118.8 Tm +(third) Tj +1 0 0 1 209.759 118.8 Tm +(party) Tj +1 0 0 1 229.799 118.8 Tm +(.) Tj +1 0 0 1 239.159 118.8 Tm +(The) Tj +1 0 0 1 258.239 118.8 Tm +(algorithm) Tj +1 0 0 1 72 106.8 Tm +(in) Tj +1 0 0 1 82.68 106.8 Tm +(Section) Tj +1 0 0 1 115.56 106.8 Tm +(4) Tj +1 0 0 1 123.48 106.8 Tm +(and) Tj +1 0 0 1 140.88 106.8 Tm +(pre) Tj +1 0 0 1 153.48 106.8 Tm +(vious) Tj +1 0 0 1 177.96 106.8 Tm +(Byzantine-f) Tj +1 0 0 1 225.12 106.8 Tm +(ault-tolerant) Tj +1 0 0 1 276.479 106.8 Tm +(algo-) Tj +1 0 0 1 72 94.92 Tm +(rithms) Tj +1 0 0 1 100.44 94.92 Tm +([31) Tj +1 0 0 1 113.76 94.92 Tm +(,) Tj +1 0 0 1 119.16 94.92 Tm +(16) Tj +1 0 0 1 129.12 94.92 Tm +(]) Tj +1 0 0 1 135.36 94.92 Tm +(for) Tj +1 0 0 1 150 94.92 Tm +(state) Tj +1 0 0 1 171.12 94.92 Tm +(machine) Tj +1 0 0 1 207.839 94.92 Tm +(replication) Tj +1 0 0 1 253.439 94.92 Tm +(rely) Tj +1 0 0 1 271.919 94.92 Tm +(on) Tj +1 0 0 1 284.879 94.92 Tm +(the) Tj +1 0 0 1 72 82.92 Tm +(e) Tj +1 0 0 1 76.32 82.92 Tm +(xtra) Tj +1 0 0 1 94.3199 82.92 Tm +(po) Tj +1 0 0 1 104.16 82.92 Tm +(wer) Tj +1 0 0 1 121.56 82.92 Tm +(of) Tj +1 0 0 1 132.36 82.92 Tm +(digital) Tj +1 0 0 1 160.44 82.92 Tm +(signatures.) Tj +1 0 0 1 207.12 82.92 Tm +(W) Tj +1 0 0 1 215.64 82.92 Tm +(e) Tj +1 0 0 1 222.719 82.92 Tm +(modi\256ed) Tj +1 0 0 1 260.639 82.92 Tm +(our) Tj +1 0 0 1 276.479 82.92 Tm +(algo-) Tj +1 0 0 1 72 71.04 Tm +(rithm) Tj +1 0 0 1 96.3599 71.04 Tm +(to) Tj +1 0 0 1 106.92 71.04 Tm +(circumv) Tj +1 0 0 1 139.68 71.04 Tm +(ent) Tj +1 0 0 1 154.32 71.04 Tm +(the) Tj +1 0 0 1 169.32 71.04 Tm +(problem) Tj +1 0 0 1 205.32 71.04 Tm +(by) Tj +1 0 0 1 218.04 71.04 Tm +(taking) Tj +1 0 0 1 245.759 71.04 Tm +(adv) Tj +1 0 0 1 260.039 71.04 Tm +(antage) Tj +1 0 0 1 288.719 71.04 Tm +(of) Tj +1 0 0 1 315 709.08 Tm +(speci\256c) Tj +1 0 0 1 347.76 709.08 Tm +(in) Tj +1 0 0 1 355.2 709.08 Tm +(v) Tj +1 0 0 1 360 709.08 Tm +(ariants,) Tj +1 0 0 1 391.2 709.08 Tm +(e.g,) Tj +1 0 0 1 407.88 709.08 Tm +(the) Tj +1 0 0 1 422.28 709.08 Tm +(in) Tj +1 0 0 1 429.72 709.08 Tm +(v) Tj +1 0 0 1 434.52 709.08 Tm +(ariant) Tj +1 0 0 1 459.359 709.08 Tm +(that) Tj +1 0 0 1 476.639 709.08 Tm +(no) Tj +1 0 0 1 488.879 709.08 Tm +(tw) Tj +1 0 0 1 498.719 709.08 Tm +(o) Tj +1 0 0 1 505.919 709.08 Tm +(dif) Tj +1 0 0 1 516.839 709.08 Tm +(ferent) Tj +1 0 0 1 315 697.08 Tm +(requests) Tj +1 0 0 1 350.28 697.08 Tm +(prepare) Tj +1 0 0 1 382.8 697.08 Tm +(with) Tj +1 0 0 1 403.08 697.08 Tm +(the) Tj +1 0 0 1 417.96 697.08 Tm +(same) Tj +1 0 0 1 441 697.08 Tm +(vie) Tj +1 0 0 1 453 697.08 Tm +(w) Tj +1 0 0 1 462.84 697.08 Tm +(and) Tj +1 0 0 1 479.879 697.08 Tm +(sequence) Tj +1 0 0 1 518.999 697.08 Tm +(num-) Tj +1 0 0 1 315 685.2 Tm +(ber) Tj +1 0 0 1 330 685.2 Tm +(at) Tj +1 0 0 1 339.36 685.2 Tm +(tw) Tj +1 0 0 1 349.2 685.2 Tm +(o) Tj +1 0 0 1 356.4 685.2 Tm +(non-f) Tj +1 0 0 1 378.12 685.2 Tm +(aulty) Tj +1 0 0 1 400.08 685.2 Tm +(replicas.) Tj +1 0 0 1 437.04 685.2 Tm +(The) Tj +1 0 0 1 454.799 685.2 Tm +(modi\256ed) Tj +1 0 0 1 492.359 685.2 Tm +(algorithm) Tj +1 0 0 1 533.399 685.2 Tm +(is) Tj +1 0 0 1 315 673.2 Tm +(described) Tj +1 0 0 1 356.28 673.2 Tm +(in) Tj +1 0 0 1 367.2 673.2 Tm +([5) Tj +1 0 0 1 375.48 673.2 Tm +(].) Tj +1 0 0 1 386.76 673.2 Tm +(Here) Tj +1 0 0 1 409.2 673.2 Tm +(we) Tj +1 0 0 1 423.96 673.2 Tm +(sk) Tj +1 0 0 1 432.72 673.2 Tm +(etch) Tj +1 0 0 1 452.399 673.2 Tm +(the) Tj +1 0 0 1 467.639 673.2 Tm +(main) Tj +1 0 0 1 490.679 673.2 Tm +(implications) Tj +1 0 0 1 315 661.32 Tm +(of) Tj +1 0 0 1 325.8 661.32 Tm +(using) Tj +1 0 0 1 349.92 661.32 Tm +(MA) Tj +1 0 0 1 365.64 661.32 Tm +(Cs.) Tj +1 0 0 1 324.96 648.36 Tm +(MA) Tj +1 0 0 1 340.68 648.36 Tm +(Cs) Tj +1 0 0 1 355.56 648.36 Tm +(can) Tj +1 0 0 1 373.92 648.36 Tm +(be) Tj +1 0 0 1 387.72 648.36 Tm +(computed) Tj +1 0 0 1 431.52 648.36 Tm +(three) Tj +1 0 0 1 455.88 648.36 Tm +(orders) Tj +1 0 0 1 485.16 648.36 Tm +(of) Tj +1 0 0 1 497.879 648.36 Tm +(magnitude) Tj +1 0 0 1 315 636.36 Tm +(f) Tj +1 0 0 1 318.24 636.36 Tm +(aster) Tj +1 0 0 1 340.56 636.36 Tm +(than) Tj +1 0 0 1 361.2 636.36 Tm +(digital) Tj +1 0 0 1 390.12 636.36 Tm +(signatures.) Tj +1 0 0 1 439.679 636.36 Tm +(F) Tj +1 0 0 1 445.079 636.36 Tm +(or) Tj +1 0 0 1 456.839 636.36 Tm +(e) Tj +1 0 0 1 461.159 636.36 Tm +(xample,) Tj +1 0 0 1 496.679 636.36 Tm +(a) Tj +1 0 0 1 504.599 636.36 Tm +(200MHz) Tj +1 0 0 1 315 624.48 Tm +(Pentium) Tj +1 0 0 1 351 624.48 Tm +(Pro) Tj +1 0 0 1 367.56 624.48 Tm +(tak) Tj +1 0 0 1 379.68 624.48 Tm +(es) Tj +1 0 0 1 390.72 624.48 Tm +(43ms) Tj +1 0 0 1 414.96 624.48 Tm +(to) Tj +1 0 0 1 425.52 624.48 Tm +(generate) Tj +1 0 0 1 462 624.48 Tm +(a) Tj +1 0 0 1 469.199 624.48 Tm +(1024-bit) Tj +1 0 0 1 505.679 624.48 Tm +(modulus) Tj +1 0 0 1 315 612.48 Tm +(RSA) Tj +1 0 0 1 338.4 612.48 Tm +(signature) Tj +1 0 0 1 379.08 612.48 Tm +(of) Tj +1 0 0 1 391.44 612.48 Tm +(an) Tj +1 0 0 1 404.88 612.48 Tm +(MD5) Tj +1 0 0 1 430.08 612.48 Tm +(digest) Tj +1 0 0 1 457.92 612.48 Tm +(and) Tj +1 0 0 1 476.399 612.48 Tm +(0.6ms) Tj +1 0 0 1 504.479 612.48 Tm +(to) Tj +1 0 0 1 516.359 612.48 Tm +(v) Tj +1 0 0 1 521.279 612.48 Tm +(erify) Tj +1 0 0 1 315 600.6 Tm +(the) Tj +1 0 0 1 332.76 600.6 Tm +(signature) Tj +1 0 0 1 375 600.6 Tm +([37) Tj +1 0 0 1 388.2 600.6 Tm +(],) Tj +1 0 0 1 400.44 600.6 Tm +(whereas) Tj +1 0 0 1 438.72 600.6 Tm +(it) Tj +1 0 0 1 449.88 600.6 Tm +(tak) Tj +1 0 0 1 462 600.6 Tm +(es) Tj +1 0 0 1 475.92 600.6 Tm +(only) Tj +1 0 0 1 499.319 600.6 Tm +(10.3) Tj +ET +endstream +endobj +657 0 obj +8780 +endobj +658 0 obj +<< /Type /XObject /Name /R658 /Subtype /Image /Length 659 0 R +/ImageMask true /Width 55 /Height 67 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 55 /BlackIs1 true >>] +>> +stream +,Fo)=s1dI\pj_2iqjRk5*s`]L +iYYt!ON4^rGA'-'R,85M_r:L)s6\^!r&aE:no3LSs5!_L&,jEGmf2,CpODW2pijXUhVkZj+b0guJ,~> +endstream +endobj +659 0 obj +106 +endobj +660 0 obj +<< /Length 661 0 R >> +stream +q +5.5 0 0 -6.7 517.2 605.1 cm +/R658 Do +Q +BT +1 0 0 1 522.959 600.6 Tm +(s) Tj +1 0 0 1 532.199 600.6 Tm +(to) Tj +1 0 0 1 315 588.6 Tm +(compute) Tj +1 0 0 1 353.4 588.6 Tm +(the) Tj +1 0 0 1 369.6 588.6 Tm +(MA) Tj +1 0 0 1 385.32 588.6 Tm +(C) Tj +1 0 0 1 396 588.6 Tm +(of) Tj +1 0 0 1 408.36 588.6 Tm +(a) Tj +1 0 0 1 416.88 588.6 Tm +(64-byte) Tj +1 0 0 1 451.44 588.6 Tm +(message) Tj +1 0 0 1 489.239 588.6 Tm +(on) Tj +1 0 0 1 503.279 588.6 Tm +(the) Tj +1 0 0 1 519.479 588.6 Tm +(same) Tj +1 0 0 1 315 576.6 Tm +(hardw) Tj +1 0 0 1 339.96 576.6 Tm +(are) Tj +1 0 0 1 354.36 576.6 Tm +(in) Tj +1 0 0 1 364.44 576.6 Tm +(our) Tj +1 0 0 1 380.16 576.6 Tm +(implementation.) Tj +1 0 0 1 448.8 576.6 Tm +(There) Tj +1 0 0 1 474.359 576.6 Tm +(are) Tj +1 0 0 1 488.879 576.6 Tm +(other) Tj +1 0 0 1 511.799 576.6 Tm +(public-) Tj +1 0 0 1 315 564.72 Tm +(k) Tj +1 0 0 1 319.92 564.72 Tm +(e) Tj +1 0 0 1 324.24 564.72 Tm +(y) Tj +1 0 0 1 333 564.72 Tm +(cryptosystems) Tj +1 0 0 1 393.96 564.72 Tm +(that) Tj +1 0 0 1 412.8 564.72 Tm +(generate) Tj +1 0 0 1 450.48 564.72 Tm +(signatures) Tj +1 0 0 1 494.759 564.72 Tm +(f) Tj +1 0 0 1 497.999 564.72 Tm +(aster) Tj +1 0 0 1 516.479 564.72 Tm +(,) Tj +1 0 0 1 523.079 564.72 Tm +(e.g.,) Tj +1 0 0 1 315 552.72 Tm +(elliptic) Tj +1 0 0 1 347.4 552.72 Tm +(curv) Tj +1 0 0 1 365.16 552.72 Tm +(e) Tj +1 0 0 1 374.16 552.72 Tm +(public-k) Tj +1 0 0 1 407.52 552.72 Tm +(e) Tj +1 0 0 1 411.84 552.72 Tm +(y) Tj +1 0 0 1 421.32 552.72 Tm +(cryptosystems,) Tj +1 0 0 1 486.119 552.72 Tm +(b) Tj +1 0 0 1 490.919 552.72 Tm +(ut) Tj +1 0 0 1 503.519 552.72 Tm +(signature) Tj +1 0 0 1 315 540.84 Tm +(v) Tj +1 0 0 1 319.92 540.84 Tm +(eri\256cation) Tj +1 0 0 1 364.92 540.84 Tm +(is) Tj +1 0 0 1 376.32 540.84 Tm +(slo) Tj +1 0 0 1 387.72 540.84 Tm +(wer) Tj +1 0 0 1 407.4 540.84 Tm +([37) Tj +1 0 0 1 420.6 540.84 Tm +(]) Tj +1 0 0 1 428.64 540.84 Tm +(and) Tj +1 0 0 1 447.84 540.84 Tm +(in) Tj +1 0 0 1 460.32 540.84 Tm +(our) Tj +1 0 0 1 478.319 540.84 Tm +(algorithm) Tj +1 0 0 1 521.759 540.84 Tm +(each) Tj +1 0 0 1 315 528.84 Tm +(signature) Tj +1 0 0 1 354 528.84 Tm +(is) Tj +1 0 0 1 363.12 528.84 Tm +(v) Tj +1 0 0 1 368.04 528.84 Tm +(eri\256ed) Tj +1 0 0 1 395.88 528.84 Tm +(man) Tj +1 0 0 1 413.04 528.84 Tm +(y) Tj +1 0 0 1 420.36 528.84 Tm +(times.) Tj +1 0 0 1 324.96 515.88 Tm +(Each) Tj +1 0 0 1 348.12 515.88 Tm +(node) Tj +1 0 0 1 370.68 515.88 Tm +(\(including) Tj +1 0 0 1 414.72 515.88 Tm +(acti) Tj +1 0 0 1 428.88 515.88 Tm +(v) Tj +1 0 0 1 433.8 515.88 Tm +(e) Tj +1 0 0 1 441.36 515.88 Tm +(clients\)) Tj +1 0 0 1 473.879 515.88 Tm +(shares) Tj +1 0 0 1 501.959 515.88 Tm +(a) Tj +1 0 0 1 509.519 515.88 Tm +(16-byte) Tj +1 0 0 1 315 504 Tm +(secret) Tj +1 0 0 1 343.8 504 Tm +(session) Tj +1 0 0 1 378.12 504 Tm +(k) Tj +1 0 0 1 383.04 504 Tm +(e) Tj +1 0 0 1 387.36 504 Tm +(y) Tj +1 0 0 1 397.8 504 Tm +(with) Tj +1 0 0 1 420.96 504 Tm +(each) Tj +1 0 0 1 444.84 504 Tm +(replica.) Tj +1 0 0 1 487.079 504 Tm +(W) Tj +1 0 0 1 495.599 504 Tm +(e) Tj +1 0 0 1 505.679 504 Tm +(compute) Tj +1 0 0 1 315 492 Tm +(message) Tj +1 0 0 1 352.56 492 Tm +(authentication) Tj +1 0 0 1 412.68 492 Tm +(codes) Tj +1 0 0 1 439.2 492 Tm +(by) Tj +1 0 0 1 452.88 492 Tm +(applying) Tj +1 0 0 1 491.519 492 Tm +(MD5) Tj +1 0 0 1 516.359 492 Tm +(to) Tj +1 0 0 1 527.879 492 Tm +(the) Tj +1 0 0 1 315 480.12 Tm +(concatenation) Tj +1 0 0 1 372.6 480.12 Tm +(of) Tj +1 0 0 1 383.28 480.12 Tm +(the) Tj +1 0 0 1 397.8 480.12 Tm +(message) Tj +1 0 0 1 433.8 480.12 Tm +(with) Tj +1 0 0 1 453.84 480.12 Tm +(the) Tj +1 0 0 1 468.36 480.12 Tm +(secret) Tj +1 0 0 1 493.919 480.12 Tm +(k) Tj +1 0 0 1 498.839 480.12 Tm +(e) Tj +1 0 0 1 503.159 480.12 Tm +(y) Tj +1 0 0 1 507.599 480.12 Tm +(.) Tj +1 0 0 1 513.479 480.12 Tm +(Rather) Tj +1 0 0 1 315 468.12 Tm +(than) Tj +1 0 0 1 335.16 468.12 Tm +(using) Tj +1 0 0 1 359.64 468.12 Tm +(the) Tj +1 0 0 1 374.76 468.12 Tm +(16) Tj +1 0 0 1 387.72 468.12 Tm +(bytes) Tj +1 0 0 1 411.72 468.12 Tm +(of) Tj +1 0 0 1 423 468.12 Tm +(the) Tj +1 0 0 1 438.119 468.12 Tm +(\256nal) Tj +1 0 0 1 458.759 468.12 Tm +(MD5) Tj +1 0 0 1 482.759 468.12 Tm +(digest,) Tj +1 0 0 1 512.159 468.12 Tm +(we) Tj +1 0 0 1 526.679 468.12 Tm +(use) Tj +1 0 0 1 315 456.24 Tm +(only) Tj +1 0 0 1 336.12 456.24 Tm +(the) Tj +1 0 0 1 351.72 456.24 Tm +(10) Tj +1 0 0 1 365.16 456.24 Tm +(least) Tj +1 0 0 1 386.88 456.24 Tm +(signi\256cant) Tj +1 0 0 1 431.76 456.24 Tm +(bytes.) Tj +1 0 0 1 461.759 456.24 Tm +(This) Tj +1 0 0 1 482.879 456.24 Tm +(truncation) Tj +1 0 0 1 526.679 456.24 Tm +(has) Tj +1 0 0 1 315 444.24 Tm +(the) Tj +1 0 0 1 329.52 444.24 Tm +(ob) Tj +1 0 0 1 339.48 444.24 Tm +(vious) Tj +1 0 0 1 363.36 444.24 Tm +(adv) Tj +1 0 0 1 377.64 444.24 Tm +(antage) Tj +1 0 0 1 405.96 444.24 Tm +(of) Tj +1 0 0 1 416.64 444.24 Tm +(reducing) Tj +1 0 0 1 453.84 444.24 Tm +(the) Tj +1 0 0 1 468.479 444.24 Tm +(size) Tj +1 0 0 1 486.359 444.24 Tm +(of) Tj +1 0 0 1 497.039 444.24 Tm +(MA) Tj +1 0 0 1 512.759 444.24 Tm +(Cs) Tj +1 0 0 1 525.599 444.24 Tm +(and) Tj +1 0 0 1 315 432.36 Tm +(it) Tj +1 0 0 1 324.12 432.36 Tm +(also) Tj +1 0 0 1 343.68 432.36 Tm +(impro) Tj +1 0 0 1 367.56 432.36 Tm +(v) Tj +1 0 0 1 372.48 432.36 Tm +(es) Tj +1 0 0 1 384.12 432.36 Tm +(their) Tj +1 0 0 1 405.96 432.36 Tm +(resilience) Tj +1 0 0 1 447.719 432.36 Tm +(to) Tj +1 0 0 1 458.999 432.36 Tm +(certain) Tj +1 0 0 1 489.719 432.36 Tm +(attacks) Tj +1 0 0 1 520.919 432.36 Tm +([27) Tj +1 0 0 1 534.239 432.36 Tm +(].) Tj +1 0 0 1 315 420.36 Tm +(This) Tj +1 0 0 1 335.64 420.36 Tm +(is) Tj +1 0 0 1 345.24 420.36 Tm +(a) Tj +1 0 0 1 352.56 420.36 Tm +(v) Tj +1 0 0 1 357.36 420.36 Tm +(ariant) Tj +1 0 0 1 382.92 420.36 Tm +(of) Tj +1 0 0 1 394.08 420.36 Tm +(the) Tj +1 0 0 1 409.2 420.36 Tm +(secret) Tj +1 0 0 1 435.359 420.36 Tm +(suf) Tj +1 0 0 1 447.359 420.36 Tm +(\256x) Tj +1 0 0 1 460.679 420.36 Tm +(method) Tj +1 0 0 1 493.559 420.36 Tm +([36) Tj +1 0 0 1 506.759 420.36 Tm +(],) Tj +1 0 0 1 515.639 420.36 Tm +(which) Tj +1 0 0 1 315 408.48 Tm +(is) Tj +1 0 0 1 324.12 408.48 Tm +(secure) Tj +1 0 0 1 352.08 408.48 Tm +(as) Tj +1 0 0 1 362.88 408.48 Tm +(long) Tj +1 0 0 1 383.04 408.48 Tm +(as) Tj +1 0 0 1 393.84 408.48 Tm +(MD5) Tj +1 0 0 1 417.36 408.48 Tm +(is) Tj +1 0 0 1 426.48 408.48 Tm +(collision) Tj +1 0 0 1 463.319 408.48 Tm +(resistant) Tj +1 0 0 1 498.959 408.48 Tm +([27) Tj +1 0 0 1 512.279 408.48 Tm +(,) Tj +1 0 0 1 517.319 408.48 Tm +(8) Tj +1 0 0 1 522.239 408.48 Tm +(].) Tj +1 0 0 1 324.96 395.52 Tm +(The) Tj +1 0 0 1 342.24 395.52 Tm +(digital) Tj +1 0 0 1 369.48 395.52 Tm +(signature) Tj +1 0 0 1 407.88 395.52 Tm +(in) Tj +1 0 0 1 417.36 395.52 Tm +(a) Tj +1 0 0 1 423.48 395.52 Tm +(rep) Tj +1 0 0 1 436.2 395.52 Tm +(ly) Tj +1 0 0 1 445.679 395.52 Tm +(m) Tj +1 0 0 1 453.359 395.52 Tm +(essage) Tj +1 0 0 1 481.079 395.52 Tm +(is) Tj +1 0 0 1 489.359 395.52 Tm +(r) Tj +1 0 0 1 492.599 395.52 Tm +(eplac) Tj +1 0 0 1 513.599 395.52 Tm +(ed) Tj +1 0 0 1 524.639 395.52 Tm +(by) Tj +1 0 0 1 536.279 395.52 Tm +(a) Tj +1 0 0 1 315 383.52 Tm +(single) Tj +1 0 0 1 341.52 383.52 Tm +(MA) Tj +1 0 0 1 357.24 383.52 Tm +(C,) Tj +1 0 0 1 369 383.52 Tm +(which) Tj +1 0 0 1 396 383.52 Tm +(is) Tj +1 0 0 1 405.36 383.52 Tm +(suf) Tj +1 0 0 1 417.36 383.52 Tm +(\256cient) Tj +1 0 0 1 444.96 383.52 Tm +(because) Tj +1 0 0 1 479.159 383.52 Tm +(these) Tj +1 0 0 1 502.319 383.52 Tm +(messages) Tj +1 0 0 1 315 371.64 Tm +(ha) Tj +1 0 0 1 324.24 371.64 Tm +(v) Tj +1 0 0 1 329.16 371.64 Tm +(e) Tj +1 0 0 1 337.2 371.64 Tm +(a) Tj +1 0 0 1 345.24 371.64 Tm +(single) Tj +1 0 0 1 372.72 371.64 Tm +(intended) Tj +1 0 0 1 410.76 371.64 Tm +(recipient.) Tj +1 0 0 1 455.279 371.64 Tm +(The) Tj +1 0 0 1 474.479 371.64 Tm +(signatures) Tj +1 0 0 1 518.639 371.64 Tm +(in) Tj +1 0 0 1 530.039 371.64 Tm +(all) Tj +1 0 0 1 315 359.64 Tm +(other) Tj +1 0 0 1 338.76 359.64 Tm +(messages) Tj +1 0 0 1 379.56 359.64 Tm +(\(including) Tj +1 0 0 1 423.84 359.64 Tm +(client) Tj +1 0 0 1 449.16 359.64 Tm +(requests) Tj +1 0 0 1 485.039 359.64 Tm +(b) Tj +1 0 0 1 489.839 359.64 Tm +(ut) Tj +1 0 0 1 500.879 359.64 Tm +(e) Tj +1 0 0 1 505.199 359.64 Tm +(xcluding) Tj +1 0 0 1 315 347.76 Tm +(vie) Tj +1 0 0 1 327 347.76 Tm +(w) Tj +1 0 0 1 336.84 347.76 Tm +(changes\)) Tj +1 0 0 1 375.12 347.76 Tm +(are) Tj +1 0 0 1 390 347.76 Tm +(replaced) Tj +1 0 0 1 426.48 347.76 Tm +(by) Tj +1 0 0 1 439.2 347.76 Tm +(v) Tj +1 0 0 1 444.12 347.76 Tm +(ectors) Tj +1 0 0 1 470.64 347.76 Tm +(of) Tj +1 0 0 1 481.679 347.76 Tm +(MA) Tj +1 0 0 1 497.4 347.76 Tm +(Cs) Tj +1 0 0 1 510.719 347.76 Tm +(that) Tj +1 0 0 1 528.359 347.76 Tm +(we) Tj +1 0 0 1 315 335.76 Tm +(call) Tj +/R644 10 Tf +1 0 0 1 333.36 335.76 Tm +(authenticator) Tj +1 0 0 1 387.24 335.76 Tm +(s) Tj +/R641 10 Tf +1 0 0 1 391.08 335.76 Tm +(.) Tj +1 0 0 1 401.28 335.76 Tm +(An) Tj +1 0 0 1 417.36 335.76 Tm +(authenticator) Tj +1 0 0 1 473.399 335.76 Tm +(has) Tj +1 0 0 1 490.559 335.76 Tm +(an) Tj +1 0 0 1 503.999 335.76 Tm +(entry) Tj +1 0 0 1 528.359 335.76 Tm +(for) Tj +1 0 0 1 315 323.88 Tm +(e) Tj +1 0 0 1 319.2 323.88 Tm +(v) Tj +1 0 0 1 324.12 323.88 Tm +(ery) Tj +1 0 0 1 341.04 323.88 Tm +(replica) Tj +1 0 0 1 372.48 323.88 Tm +(other) Tj +1 0 0 1 397.32 323.88 Tm +(than) Tj +1 0 0 1 418.8 323.88 Tm +(the) Tj +1 0 0 1 435.36 323.88 Tm +(sender;) Tj +1 0 0 1 469.439 323.88 Tm +(each) Tj +1 0 0 1 491.999 323.88 Tm +(entry) Tj +1 0 0 1 516.839 323.88 Tm +(is) Tj +1 0 0 1 527.879 323.88 Tm +(the) Tj +1 0 0 1 315 311.88 Tm +(MA) Tj +1 0 0 1 330.72 311.88 Tm +(C) Tj +1 0 0 1 340.68 311.88 Tm +(computed) Tj +1 0 0 1 383.4 311.88 Tm +(with) Tj +1 0 0 1 404.64 311.88 Tm +(the) Tj +1 0 0 1 420.24 311.88 Tm +(k) Tj +1 0 0 1 425.16 311.88 Tm +(e) Tj +1 0 0 1 429.48 311.88 Tm +(y) Tj +1 0 0 1 437.76 311.88 Tm +(shared) Tj +1 0 0 1 467.16 311.88 Tm +(by) Tj +1 0 0 1 480.6 311.88 Tm +(the) Tj +1 0 0 1 496.199 311.88 Tm +(sender) Tj +1 0 0 1 525.599 311.88 Tm +(and) Tj +1 0 0 1 315 300 Tm +(the) Tj +1 0 0 1 329.64 300 Tm +(replica) Tj +1 0 0 1 359.28 300 Tm +(corresponding) Tj +1 0 0 1 418.8 300 Tm +(to) Tj +1 0 0 1 429 300 Tm +(the) Tj +1 0 0 1 443.64 300 Tm +(entry) Tj +1 0 0 1 463.679 300 Tm +(.) Tj +1 0 0 1 324.96 287.04 Tm +(The) Tj +1 0 0 1 343.2 287.04 Tm +(time) Tj +1 0 0 1 363.6 287.04 Tm +(to) Tj +1 0 0 1 374.04 287.04 Tm +(v) Tj +1 0 0 1 378.96 287.04 Tm +(erify) Tj +1 0 0 1 400.44 287.04 Tm +(an) Tj +1 0 0 1 412.56 287.04 Tm +(authenticator) Tj +1 0 0 1 467.279 287.04 Tm +(is) Tj +1 0 0 1 476.639 287.04 Tm +(constant) Tj +1 0 0 1 512.519 287.04 Tm +(b) Tj +1 0 0 1 517.319 287.04 Tm +(ut) Tj +1 0 0 1 527.879 287.04 Tm +(the) Tj +1 0 0 1 315 275.04 Tm +(time) Tj +1 0 0 1 335.88 275.04 Tm +(to) Tj +1 0 0 1 346.8 275.04 Tm +(generate) Tj +1 0 0 1 383.64 275.04 Tm +(one) Tj +1 0 0 1 401.16 275.04 Tm +(gro) Tj +1 0 0 1 414.36 275.04 Tm +(ws) Tj +1 0 0 1 428.4 275.04 Tm +(linearly) Tj +1 0 0 1 461.999 275.04 Tm +(with) Tj +1 0 0 1 482.759 275.04 Tm +(the) Tj +1 0 0 1 498.119 275.04 Tm +(number) Tj +1 0 0 1 531.719 275.04 Tm +(of) Tj +1 0 0 1 315 263.16 Tm +(replicas.) Tj +1 0 0 1 352.08 263.16 Tm +(This) Tj +1 0 0 1 372.12 263.16 Tm +(is) Tj +1 0 0 1 381.24 263.16 Tm +(not) Tj +1 0 0 1 396.36 263.16 Tm +(a) Tj +1 0 0 1 403.08 263.16 Tm +(problem) Tj +1 0 0 1 438.72 263.16 Tm +(because) Tj +1 0 0 1 472.679 263.16 Tm +(we) Tj +1 0 0 1 486.719 263.16 Tm +(do) Tj +1 0 0 1 499.079 263.16 Tm +(not) Tj +1 0 0 1 514.199 263.16 Tm +(e) Tj +1 0 0 1 518.519 263.16 Tm +(xpect) Tj +1 0 0 1 315 251.16 Tm +(to) Tj +1 0 0 1 326.4 251.16 Tm +(ha) Tj +1 0 0 1 335.64 251.16 Tm +(v) Tj +1 0 0 1 340.56 251.16 Tm +(e) Tj +1 0 0 1 348.48 251.16 Tm +(a) Tj +1 0 0 1 356.64 251.16 Tm +(lar) Tj +1 0 0 1 367.08 251.16 Tm +(ge) Tj +1 0 0 1 379.92 251.16 Tm +(number) Tj +1 0 0 1 414.12 251.16 Tm +(of) Tj +1 0 0 1 426 251.16 Tm +(replicas) Tj +1 0 0 1 460.679 251.16 Tm +(and) Tj +1 0 0 1 478.679 251.16 Tm +(there) Tj +1 0 0 1 502.199 251.16 Tm +(is) Tj +1 0 0 1 512.519 251.16 Tm +(a) Tj +1 0 0 1 520.679 251.16 Tm +(huge) Tj +1 0 0 1 315 239.16 Tm +(performance) Tj +1 0 0 1 370.08 239.16 Tm +(gap) Tj +1 0 0 1 389.16 239.16 Tm +(between) Tj +1 0 0 1 427.08 239.16 Tm +(MA) Tj +1 0 0 1 442.8 239.16 Tm +(C) Tj +1 0 0 1 454.08 239.16 Tm +(and) Tj +1 0 0 1 473.28 239.16 Tm +(digital) Tj +1 0 0 1 503.519 239.16 Tm +(signature) Tj +1 0 0 1 315 227.28 Tm +(computation.) Tj +1 0 0 1 374.52 227.28 Tm +(Furthermore,) Tj +1 0 0 1 430.8 227.28 Tm +(we) Tj +1 0 0 1 446.16 227.28 Tm +(compute) Tj +1 0 0 1 484.08 227.28 Tm +(authenticators) Tj +1 0 0 1 315 215.28 Tm +(ef) Tj +1 0 0 1 322.56 215.28 Tm +(\256ciently;) Tj +1 0 0 1 360.96 215.28 Tm +(MD5) Tj +1 0 0 1 384.84 215.28 Tm +(is) Tj +1 0 0 1 394.44 215.28 Tm +(applied) Tj +1 0 0 1 426.6 215.28 Tm +(to) Tj +1 0 0 1 437.16 215.28 Tm +(the) Tj +1 0 0 1 452.279 215.28 Tm +(message) Tj +1 0 0 1 488.879 215.28 Tm +(once) Tj +1 0 0 1 510.599 215.28 Tm +(and) Tj +1 0 0 1 527.879 215.28 Tm +(the) Tj +1 0 0 1 315 203.4 Tm +(resulting) Tj +1 0 0 1 353.64 203.4 Tm +(conte) Tj +1 0 0 1 375.24 203.4 Tm +(xt) Tj +1 0 0 1 386.64 203.4 Tm +(is) Tj +1 0 0 1 397.08 203.4 Tm +(used) Tj +1 0 0 1 419.16 203.4 Tm +(to) Tj +1 0 0 1 430.8 203.4 Tm +(compute) Tj +1 0 0 1 468.959 203.4 Tm +(each) Tj +1 0 0 1 490.919 203.4 Tm +(v) Tj +1 0 0 1 495.839 203.4 Tm +(ector) Tj +1 0 0 1 519.479 203.4 Tm +(entry) Tj +1 0 0 1 315 191.4 Tm +(by) Tj +1 0 0 1 327.72 191.4 Tm +(applying) Tj +1 0 0 1 365.4 191.4 Tm +(MD5) Tj +1 0 0 1 389.16 191.4 Tm +(to) Tj +1 0 0 1 399.72 191.4 Tm +(the) Tj +1 0 0 1 414.6 191.4 Tm +(corresponding) Tj +1 0 0 1 474.36 191.4 Tm +(session) Tj +1 0 0 1 505.919 191.4 Tm +(k) Tj +1 0 0 1 510.839 191.4 Tm +(e) Tj +1 0 0 1 515.159 191.4 Tm +(y) Tj +1 0 0 1 519.599 191.4 Tm +(.) Tj +1 0 0 1 526.319 191.4 Tm +(F) Tj +1 0 0 1 531.719 191.4 Tm +(or) Tj +1 0 0 1 315 179.52 Tm +(e) Tj +1 0 0 1 319.32 179.52 Tm +(xample,) Tj +1 0 0 1 355.92 179.52 Tm +(in) Tj +1 0 0 1 368.04 179.52 Tm +(a) Tj +1 0 0 1 376.92 179.52 Tm +(system) Tj +1 0 0 1 408.96 179.52 Tm +(with) Tj +1 0 0 1 431.04 179.52 Tm +(37) Tj +1 0 0 1 445.32 179.52 Tm +(replicas) Tj +1 0 0 1 480.719 179.52 Tm +(\(i.e.,) Tj +1 0 0 1 503.519 179.52 Tm +(a) Tj +1 0 0 1 512.279 179.52 Tm +(system) Tj +1 0 0 1 315 167.52 Tm +(that) Tj +1 0 0 1 332.4 167.52 Tm +(can) Tj +1 0 0 1 348.6 167.52 Tm +(tolerate) Tj +1 0 0 1 381 167.52 Tm +(12) Tj +1 0 0 1 393.36 167.52 Tm +(simultaneous) Tj +1 0 0 1 448.319 167.52 Tm +(f) Tj +1 0 0 1 451.559 167.52 Tm +(aults\)) Tj +1 0 0 1 476.159 167.52 Tm +(an) Tj +1 0 0 1 488.039 167.52 Tm +(authenticator) Tj +1 0 0 1 315 155.64 Tm +(can) Tj +1 0 0 1 333.24 155.64 Tm +(still) Tj +1 0 0 1 352.56 155.64 Tm +(be) Tj +1 0 0 1 366.48 155.64 Tm +(computed) Tj +1 0 0 1 410.16 155.64 Tm +(much) Tj +1 0 0 1 436.68 155.64 Tm +(more) Tj +1 0 0 1 461.52 155.64 Tm +(than) Tj +1 0 0 1 483.119 155.64 Tm +(tw) Tj +1 0 0 1 492.959 155.64 Tm +(o) Tj +1 0 0 1 502.439 155.64 Tm +(orders) Tj +1 0 0 1 531.719 155.64 Tm +(of) Tj +1 0 0 1 315 143.64 Tm +(magnitude) Tj +1 0 0 1 358.92 143.64 Tm +(f) Tj +1 0 0 1 362.16 143.64 Tm +(aster) Tj +1 0 0 1 382.8 143.64 Tm +(than) Tj +1 0 0 1 401.76 143.64 Tm +(a) Tj +1 0 0 1 408 143.64 Tm +(1024-bit) Tj +1 0 0 1 443.64 143.64 Tm +(modulus) Tj +1 0 0 1 479.759 143.64 Tm +(RSA) Tj +1 0 0 1 500.999 143.64 Tm +(signature.) Tj +1 0 0 1 324.96 130.68 Tm +(The) Tj +1 0 0 1 345.48 130.68 Tm +(size) Tj +1 0 0 1 366.12 130.68 Tm +(of) Tj +1 0 0 1 379.44 130.68 Tm +(authenticators) Tj +1 0 0 1 440.4 130.68 Tm +(gro) Tj +1 0 0 1 453.6 130.68 Tm +(ws) Tj +1 0 0 1 469.56 130.68 Tm +(linearly) Tj +1 0 0 1 505.079 130.68 Tm +(with) Tj +1 0 0 1 527.879 130.68 Tm +(the) Tj +1 0 0 1 315 118.8 Tm +(number) Tj +1 0 0 1 349.44 118.8 Tm +(of) Tj +1 0 0 1 361.68 118.8 Tm +(replicas) Tj +1 0 0 1 396.72 118.8 Tm +(b) Tj +1 0 0 1 401.52 118.8 Tm +(ut) Tj +1 0 0 1 413.16 118.8 Tm +(it) Tj +1 0 0 1 422.76 118.8 Tm +(gro) Tj +1 0 0 1 435.96 118.8 Tm +(ws) Tj +1 0 0 1 450.84 118.8 Tm +(slo) Tj +1 0 0 1 462.24 118.8 Tm +(wly:) Tj +1 0 0 1 486.479 118.8 Tm +(it) Tj +1 0 0 1 496.079 118.8 Tm +(is) Tj +1 0 0 1 506.639 118.8 Tm +(equal) Tj +1 0 0 1 532.199 118.8 Tm +(to) Tj +1 0 0 1 315 106.8 Tm +(30) Tj +ET +endstream +endobj +661 0 obj +15291 +endobj +662 0 obj +<< /Type /XObject /Name /R662 /Subtype /Image /Length 663 0 R +/ImageMask true /Width 49 /Height 49 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 49 /BlackIs1 true >>] +>> +stream +,K-UT@67h56YTu2Y_'iL)8\I86km$G`&5CSk:J&,`]#017Bg"U@"<.1<5Qh~> +endstream +endobj +663 0 obj +102 +endobj +664 0 obj +<< /Length 665 0 R >> +stream +q +4.9 0 0 -4.9 329 111.8 cm +/R662 Do +Q +endstream +endobj +665 0 obj +39 +endobj +666 0 obj +<< /Type /XObject /Name /R666 /Subtype /Image /Length 667 0 R +/ImageMask true /Width 26 /Height 99 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 26 /BlackIs1 true >>] +>> +stream +/da'+AAiuts8W-!s8W-!s8W-!s8W-!s8W-!s8W-!s8W-!s8W-!qbR:n&-~> +endstream +endobj +667 0 obj +60 +endobj +668 0 obj +<< /Length 669 0 R >> +stream +q +2.6 0 0 -9.9 339.6 114.3 cm +/R666 Do +Q +endstream +endobj +669 0 obj +41 +endobj +670 0 obj +<< /Type /XObject /Name /R670 /Subtype /Image /Length 671 0 R +/ImageMask true /Width 43 /Height 33 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 43 /BlackIs1 true >>] +>> +stream +366ZtBKi$Kj)nf1pi$0"h;="Vq3Z%=rV-?Z\+qmGnGV5SiW&fRKsuPqJ89Nq=JdRb$l@h'!$D~> +endstream +endobj +671 0 obj +76 +endobj +672 0 obj +<< /Length 673 0 R >> +stream +q +4.3 0 0 -3.3 344 114 cm +/R670 Do +Q +endstream +endobj +673 0 obj +37 +endobj +674 0 obj +<< /Type /XObject /Name /R674 /Subtype /Image /Length 675 0 R +/ImageMask true /Width 47 /Height 5 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 47 /BlackIs1 true >>] +>> +stream +-D7RA!$D=~> +endstream +endobj +675 0 obj +12 +endobj +676 0 obj +<< /Length 677 0 R >> +stream +q +4.7 0 0 -0.5 349.3 112.8 cm +/R674 Do +Q +BT +/R641 7 Tf +1 0 0 1 354.84 110.76 Tm +(1) Tj +ET +q +14.64 0 0 -0.48 343.668 109.572 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +1 0 0 1 349.2 103.44 Tm +(3) Tj +ET +endstream +endobj +677 0 obj +216 +endobj +678 0 obj +<< /Type /XObject /Name /R678 /Subtype /Image /Length 679 0 R +/ImageMask true /Width 26 /Height 99 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 26 /BlackIs1 true >>] +>> +stream +,Ct/ij?\:qs8W-!s8W-!s8W-!s8W-!s8W-!s+[0J!$D~> +endstream +endobj +679 0 obj +46 +endobj +680 0 obj +<< /Length 681 0 R >> +stream +q +2.6 0 0 -9.9 359.6 114.3 cm +/R678 Do +Q +BT +/R641 10 Tf +1 0 0 1 367.2 106.8 Tm +(bytes.) Tj +1 0 0 1 396.72 106.8 Tm +(An) Tj +1 0 0 1 412.2 106.8 Tm +(authenticator) Tj +1 0 0 1 467.52 106.8 Tm +(is) Tj +1 0 0 1 477.48 106.8 Tm +(smaller) Tj +1 0 0 1 510.119 106.8 Tm +(than) Tj +1 0 0 1 530.639 106.8 Tm +(an) Tj +1 0 0 1 315 94.92 Tm +(RSA) Tj +1 0 0 1 336.84 94.92 Tm +(signature) Tj +1 0 0 1 375.96 94.92 Tm +(with) Tj +1 0 0 1 396.12 94.92 Tm +(a) Tj +1 0 0 1 403.08 94.92 Tm +(1024-bit) Tj +1 0 0 1 439.32 94.92 Tm +(modulus) Tj +1 0 0 1 476.159 94.92 Tm +(for) Tj +ET +endstream +endobj +681 0 obj +548 +endobj +682 0 obj +<< /Type /XObject /Name /R682 /Subtype /Image /Length 683 0 R +/ImageMask true /Width 54 /Height 46 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 54 /BlackIs1 true >>] +>> +stream +4N]t,$m>7rl/"o@d!re, +pYTohhg]lis4_o!aocV_^Yur>rVsXfR^`8 +endstream +endobj +683 0 obj +102 +endobj +684 0 obj +<< /Length 685 0 R >> +stream +q +5.4 0 0 -4.6 490.4 99.4 cm +/R682 Do +Q +endstream +endobj +685 0 obj +40 +endobj +686 0 obj +<< /Type /XObject /Name /R686 /Subtype /Image /Length 687 0 R +/ImageMask true /Width 61 /Height 80 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 61 /BlackIs1 true >>] +>> +stream +,Cs,bCWHJN=?&u&-qa^H'Hg5V5sbq_";DtTYX(S` +"U#M^:p<5P$O$YI+At1.K#5?3i5G[TII_hJX +endstream +endobj +687 0 obj +123 +endobj +688 0 obj +<< /Length 689 0 R >> +stream +q +6.1 0 0 -8 499.8 101.3 cm +/R686 Do +Q +BT +1 0 0 1 509.519 94.92 Tm +(13) Tj +1 0 0 1 521.999 94.92 Tm +(\(i.e.,) Tj +1 0 0 1 315 82.92 Tm +(systems) Tj +1 0 0 1 350.52 82.92 Tm +(that) Tj +1 0 0 1 369.36 82.92 Tm +(can) Tj +1 0 0 1 387.24 82.92 Tm +(tolerate) Tj +1 0 0 1 421.08 82.92 Tm +(up) Tj +1 0 0 1 435 82.92 Tm +(to) Tj +1 0 0 1 446.64 82.92 Tm +(4) Tj +1 0 0 1 455.64 82.92 Tm +(simultaneous) Tj +1 0 0 1 512.159 82.92 Tm +(f) Tj +1 0 0 1 515.399 82.92 Tm +(aults\),) Tj +1 0 0 1 315 71.04 Tm +(which) Tj +1 0 0 1 341.88 71.04 Tm +(we) Tj +1 0 0 1 355.92 71.04 Tm +(e) Tj +1 0 0 1 360.24 71.04 Tm +(xpect) Tj +1 0 0 1 384.24 71.04 Tm +(to) Tj +1 0 0 1 394.56 71.04 Tm +(be) Tj +1 0 0 1 406.44 71.04 Tm +(true) Tj +1 0 0 1 424.44 71.04 Tm +(in) Tj +1 0 0 1 434.64 71.04 Tm +(most) Tj +1 0 0 1 456.479 71.04 Tm +(con\256gurations.) Tj +1 0 0 1 303.48 35.16 Tm +(8) Tj +ET +Q +endstream +endobj +689 0 obj +835 +endobj +690 0 obj +<< +/Type /Page +/MediaBox [0 0 612 792] +/Parent 2 0 R +/Resources << /ProcSet [/PDF /ImageB /Text] +/Font << +/R655 655 0 R +/R644 644 0 R +/R641 641 0 R +>> +/XObject << +/R686 686 0 R +/R682 682 0 R +/R678 678 0 R +/R674 674 0 R +/R670 670 0 R +/R666 666 0 R +/R662 662 0 R +/R658 658 0 R +/R651 651 0 R +/R647 647 0 R +>> +>> +/Contents [ +642 0 R +645 0 R +649 0 R +653 0 R +656 0 R +660 0 R +664 0 R +668 0 R +672 0 R +676 0 R +680 0 R +684 0 R +688 0 R +] +>> +endobj +691 0 obj +<< /Type /Font /Name /R691 /Subtype /Type1 /BaseFont /Times-Bold >> +endobj +692 0 obj +<< /Length 693 0 R >> +stream +q +BT +/R691 12 Tf +1 0 0 1 72 709.08 Tm +(6) Tj +1 0 0 1 89.8799 709.08 Tm +(Implementation) Tj +ET +endstream +endobj +693 0 obj +94 +endobj +694 0 obj +<< /Type /Font /Name /R694 /Subtype /Type1 /BaseFont /Times-Roman >> +endobj +695 0 obj +<< /Length 696 0 R >> +stream +BT +/R694 10 Tf +1 0 0 1 72 692.4 Tm +(This) Tj +1 0 0 1 94.4399 692.4 Tm +(section) Tj +1 0 0 1 127.32 692.4 Tm +(describes) Tj +1 0 0 1 169.08 692.4 Tm +(our) Tj +1 0 0 1 187.08 692.4 Tm +(implementation.) Tj +1 0 0 1 262.439 692.4 Tm +(First) Tj +1 0 0 1 285.359 692.4 Tm +(we) Tj +1 0 0 1 72 680.52 Tm +(discuss) Tj +1 0 0 1 105.6 680.52 Tm +(the) Tj +1 0 0 1 122.52 680.52 Tm +(replication) Tj +1 0 0 1 169.92 680.52 Tm +(library) Tj +1 0 0 1 196.08 680.52 Tm +(,) Tj +1 0 0 1 203.76 680.52 Tm +(which) Tj +1 0 0 1 232.8 680.52 Tm +(can) Tj +1 0 0 1 251.399 680.52 Tm +(be) Tj +1 0 0 1 265.679 680.52 Tm +(used) Tj +1 0 0 1 288.719 680.52 Tm +(as) Tj +1 0 0 1 72 668.52 Tm +(a) Tj +1 0 0 1 80.6399 668.52 Tm +(basis) Tj +1 0 0 1 104.76 668.52 Tm +(for) Tj +1 0 0 1 120.6 668.52 Tm +(an) Tj +1 0 0 1 129.96 668.52 Tm +(y) Tj +1 0 0 1 139.08 668.52 Tm +(replicated) Tj +1 0 0 1 182.64 668.52 Tm +(service.) Tj +1 0 0 1 222.119 668.52 Tm +(In) Tj +1 0 0 1 234.599 668.52 Tm +(Section) Tj +1 0 0 1 268.679 668.52 Tm +(6.2) Tj +1 0 0 1 285.359 668.52 Tm +(we) Tj +1 0 0 1 72 656.64 Tm +(describe) Tj +1 0 0 1 108.48 656.64 Tm +(ho) Tj +1 0 0 1 118.32 656.64 Tm +(w) Tj +1 0 0 1 128.76 656.64 Tm +(we) Tj +1 0 0 1 143.64 656.64 Tm +(implemented) Tj +1 0 0 1 199.08 656.64 Tm +(a) Tj +1 0 0 1 206.76 656.64 Tm +(replicated) Tj +1 0 0 1 249.36 656.64 Tm +(NFS) Tj +1 0 0 1 270.959 656.64 Tm +(on) Tj +1 0 0 1 284.279 656.64 Tm +(top) Tj +1 0 0 1 72 644.64 Tm +(of) Tj +1 0 0 1 84.6 644.64 Tm +(the) Tj +1 0 0 1 101.16 644.64 Tm +(replication) Tj +1 0 0 1 148.2 644.64 Tm +(library) Tj +1 0 0 1 174.36 644.64 Tm +(.) Tj +1 0 0 1 185.76 644.64 Tm +(Then) Tj +1 0 0 1 210.6 644.64 Tm +(we) Tj +1 0 0 1 226.559 644.64 Tm +(describe) Tj +1 0 0 1 264.119 644.64 Tm +(ho) Tj +1 0 0 1 273.959 644.64 Tm +(w) Tj +1 0 0 1 285.359 644.64 Tm +(we) Tj +1 0 0 1 72 632.76 Tm +(maintain) Tj +1 0 0 1 111.36 632.76 Tm +(checkpoints) Tj +1 0 0 1 163.44 632.76 Tm +(and) Tj +1 0 0 1 182.28 632.76 Tm +(compute) Tj +1 0 0 1 221.16 632.76 Tm +(checkpoint) Tj +1 0 0 1 269.279 632.76 Tm +(digests) Tj +1 0 0 1 72 620.76 Tm +(ef) Tj +1 0 0 1 79.56 620.76 Tm +(\256ciently) Tj +1 0 0 1 111.72 620.76 Tm +(.) Tj +/R691 10 Tf +1 0 0 1 72 599.28 Tm +(6.1) Tj +1 0 0 1 94.4399 599.28 Tm +(The) Tj +1 0 0 1 113.52 599.28 Tm +(Replication) Tj +1 0 0 1 164.76 599.28 Tm +(Library) Tj +/R694 10 Tf +1 0 0 1 72 585 Tm +(The) Tj +1 0 0 1 90.4799 585 Tm +(client) Tj +1 0 0 1 115.56 585 Tm +(interf) Tj +1 0 0 1 137.16 585 Tm +(ace) Tj +1 0 0 1 153.24 585 Tm +(to) Tj +1 0 0 1 163.92 585 Tm +(the) Tj +1 0 0 1 179.16 585 Tm +(replication) Tj +1 0 0 1 224.639 585 Tm +(library) Tj +1 0 0 1 254.159 585 Tm +(consists) Tj +1 0 0 1 288.719 585 Tm +(of) Tj +1 0 0 1 72 573 Tm +(a) Tj +1 0 0 1 79.4399 573 Tm +(single) Tj +1 0 0 1 106.2 573 Tm +(procedure,) Tj +ET +endstream +endobj +696 0 obj +2668 +endobj +697 0 obj +<< /Type /Font /Name /R697 /Subtype /Type1 /BaseFont /Times-Italic >> +endobj +698 0 obj +<< /Length 699 0 R >> +stream +BT +/R697 10 Tf +1 0 0 1 151.68 573 Tm +(in) Tj +1 0 0 1 159.12 573 Tm +(vok) Tj +1 0 0 1 172.92 573 Tm +(e) Tj +/R694 10 Tf +1 0 0 1 177.36 573 Tm +(,) Tj +1 0 0 1 182.88 573 Tm +(with) Tj +1 0 0 1 203.639 573 Tm +(one) Tj +1 0 0 1 221.039 573 Tm +(ar) Tj +1 0 0 1 228.719 573 Tm +(gument,) Tj +1 0 0 1 264.119 573 Tm +(an) Tj +1 0 0 1 276.479 573 Tm +(input) Tj +1 0 0 1 72 561.12 Tm +(b) Tj +1 0 0 1 76.8 561.12 Tm +(uf) Tj +1 0 0 1 84.96 561.12 Tm +(fer) Tj +1 0 0 1 100.32 561.12 Tm +(containing) Tj +1 0 0 1 146.76 561.12 Tm +(a) Tj +1 0 0 1 155.64 561.12 Tm +(request) Tj +1 0 0 1 188.76 561.12 Tm +(to) Tj +1 0 0 1 200.88 561.12 Tm +(in) Tj +1 0 0 1 208.32 561.12 Tm +(v) Tj +1 0 0 1 213.12 561.12 Tm +(ok) Tj +1 0 0 1 223.079 561.12 Tm +(e) Tj +1 0 0 1 231.719 561.12 Tm +(a) Tj +1 0 0 1 240.599 561.12 Tm +(state) Tj +1 0 0 1 263.279 561.12 Tm +(machine) Tj +1 0 0 1 72 549.12 Tm +(operation.) Tj +1 0 0 1 120.48 549.12 Tm +(The) Tj +/R697 10 Tf +1 0 0 1 139.92 549.12 Tm +(in) Tj +1 0 0 1 147.36 549.12 Tm +(vok) Tj +1 0 0 1 161.16 549.12 Tm +(e) Tj +/R694 10 Tf +1 0 0 1 169.56 549.12 Tm +(procedure) Tj +1 0 0 1 213.48 549.12 Tm +(uses) Tj +1 0 0 1 234.6 549.12 Tm +(our) Tj +1 0 0 1 251.999 549.12 Tm +(protocol) Tj +1 0 0 1 289.199 549.12 Tm +(to) Tj +1 0 0 1 72 537.12 Tm +(e) Tj +1 0 0 1 76.32 537.12 Tm +(x) Tj +1 0 0 1 81.24 537.12 Tm +(ecute) Tj +1 0 0 1 104.52 537.12 Tm +(the) Tj +1 0 0 1 119.16 537.12 Tm +(requested) Tj +1 0 0 1 159.72 537.12 Tm +(operation) Tj +1 0 0 1 199.8 537.12 Tm +(at) Tj +1 0 0 1 209.52 537.12 Tm +(the) Tj +1 0 0 1 224.039 537.12 Tm +(replicas) Tj +1 0 0 1 257.519 537.12 Tm +(and) Tj +1 0 0 1 274.319 537.12 Tm +(select) Tj +1 0 0 1 72 525.24 Tm +(the) Tj +1 0 0 1 86.2799 525.24 Tm +(correct) Tj +1 0 0 1 116.04 525.24 Tm +(reply) Tj +1 0 0 1 138.6 525.24 Tm +(from) Tj +1 0 0 1 160.08 525.24 Tm +(among) Tj +1 0 0 1 189.24 525.24 Tm +(the) Tj +1 0 0 1 203.52 525.24 Tm +(replies) Tj +1 0 0 1 232.199 525.24 Tm +(of) Tj +1 0 0 1 242.519 525.24 Tm +(the) Tj +1 0 0 1 256.799 525.24 Tm +(indi) Tj +1 0 0 1 272.159 525.24 Tm +(vidual) Tj +1 0 0 1 72 513.24 Tm +(replicas.) Tj +1 0 0 1 112.2 513.24 Tm +(It) Tj +1 0 0 1 121.8 513.24 Tm +(returns) Tj +1 0 0 1 153 513.24 Tm +(a) Tj +1 0 0 1 160.92 513.24 Tm +(pointer) Tj +1 0 0 1 192.6 513.24 Tm +(to) Tj +1 0 0 1 203.879 513.24 Tm +(a) Tj +1 0 0 1 211.799 513.24 Tm +(b) Tj +1 0 0 1 216.599 513.24 Tm +(uf) Tj +1 0 0 1 224.759 513.24 Tm +(fer) Tj +1 0 0 1 239.279 513.24 Tm +(containing) Tj +1 0 0 1 284.879 513.24 Tm +(the) Tj +1 0 0 1 72 501.36 Tm +(operation) Tj +1 0 0 1 112.08 501.36 Tm +(result.) Tj +1 0 0 1 81.96 489.36 Tm +(On) Tj +1 0 0 1 99.5999 489.36 Tm +(the) Tj +1 0 0 1 117.24 489.36 Tm +(serv) Tj +1 0 0 1 133.8 489.36 Tm +(er) Tj +1 0 0 1 147 489.36 Tm +(side,) Tj +1 0 0 1 171.72 489.36 Tm +(the) Tj +1 0 0 1 189.36 489.36 Tm +(replication) Tj +1 0 0 1 237.479 489.36 Tm +(code) Tj +1 0 0 1 261.719 489.36 Tm +(mak) Tj +1 0 0 1 278.879 489.36 Tm +(es) Tj +1 0 0 1 292.559 489.36 Tm +(a) Tj +1 0 0 1 72 477.36 Tm +(number) Tj +1 0 0 1 105.96 477.36 Tm +(of) Tj +1 0 0 1 117.72 477.36 Tm +(upcalls) Tj +1 0 0 1 149.4 477.36 Tm +(to) Tj +1 0 0 1 160.68 477.36 Tm +(procedures) Tj +1 0 0 1 207.96 477.36 Tm +(that) Tj +1 0 0 1 226.32 477.36 Tm +(the) Tj +1 0 0 1 242.04 477.36 Tm +(serv) Tj +1 0 0 1 258.599 477.36 Tm +(er) Tj +1 0 0 1 269.759 477.36 Tm +(part) Tj +1 0 0 1 288.719 477.36 Tm +(of) Tj +1 0 0 1 72 465.48 Tm +(the) Tj +1 0 0 1 87.7199 465.48 Tm +(application) Tj +1 0 0 1 135.6 465.48 Tm +(must) Tj +1 0 0 1 158.64 465.48 Tm +(implement.) Tj +1 0 0 1 210.72 465.48 Tm +(There) Tj +1 0 0 1 237.48 465.48 Tm +(are) Tj +1 0 0 1 253.319 465.48 Tm +(procedures) Tj +1 0 0 1 72 453.48 Tm +(to) Tj +1 0 0 1 82.2 453.48 Tm +(e) Tj +1 0 0 1 86.5199 453.48 Tm +(x) Tj +1 0 0 1 91.4399 453.48 Tm +(ecute) Tj +1 0 0 1 114.84 453.48 Tm +(requests) Tj +1 0 0 1 149.88 453.48 Tm +(\() Tj +/R697 10 Tf +1 0 0 1 153.24 453.48 Tm +(e) Tj +1 0 0 1 157.44 453.48 Tm +(xecute) Tj +/R694 10 Tf +1 0 0 1 183 453.48 Tm +(\),) Tj +1 0 0 1 191.16 453.48 Tm +(to) Tj +1 0 0 1 201.36 453.48 Tm +(maintain) Tj +1 0 0 1 238.679 453.48 Tm +(checkpoints) Tj +1 0 0 1 288.719 453.48 Tm +(of) Tj +1 0 0 1 72 441.48 Tm +(the) Tj +1 0 0 1 86.0399 441.48 Tm +(service) Tj +1 0 0 1 116.04 441.48 Tm +(state) Tj +1 0 0 1 136.2 441.48 Tm +(\() Tj +/R697 10 Tf +1 0 0 1 139.56 441.48 Tm +(mak) Tj +1 0 0 1 156.12 441.48 Tm +(e) Tj +ET +Q +q +W +0 0 612 792 re +n +q +3 0 0 -0.48 161.148 441.972 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +/R697 10 Tf +1 0 0 1 164.04 441.48 Tm +(c) Tj +1 0 0 1 168.36 441.48 Tm +(hec) Tj +1 0 0 1 182.04 441.48 Tm +(kpoint) Tj +/R694 10 Tf +1 0 0 1 207.12 441.48 Tm +(,) Tj +/R697 10 Tf +1 0 0 1 211.319 441.48 Tm +(delete) Tj +ET +q +3 0 0 -0.48 235.788 441.972 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +1 0 0 1 238.679 441.48 Tm +(c) Tj +1 0 0 1 242.999 441.48 Tm +(hec) Tj +1 0 0 1 256.679 441.48 Tm +(kpoint) Tj +/R694 10 Tf +1 0 0 1 281.759 441.48 Tm +(\),) Tj +1 0 0 1 289.199 441.48 Tm +(to) Tj +1 0 0 1 72 429.6 Tm +(obtain) Tj +1 0 0 1 100.32 429.6 Tm +(the) Tj +1 0 0 1 115.92 429.6 Tm +(digest) Tj +1 0 0 1 143.04 429.6 Tm +(of) Tj +1 0 0 1 154.8 429.6 Tm +(a) Tj +1 0 0 1 162.6 429.6 Tm +(speci\256ed) Tj +1 0 0 1 201.36 429.6 Tm +(checkpoint) Tj +1 0 0 1 248.519 429.6 Tm +(\() Tj +/R697 10 Tf +1 0 0 1 251.879 429.6 Tm +(g) Tj +1 0 0 1 256.799 429.6 Tm +(et) Tj +ET +q +3 0 0 -0.48 264.588 430.092 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +1 0 0 1 267.479 429.6 Tm +(dig) Tj +1 0 0 1 280.199 429.6 Tm +(est) Tj +/R694 10 Tf +1 0 0 1 291.239 429.6 Tm +(\),) Tj +1 0 0 1 72 417.6 Tm +(and) Tj +1 0 0 1 93.48 417.6 Tm +(to) Tj +1 0 0 1 108.36 417.6 Tm +(obtain) Tj +1 0 0 1 140.4 417.6 Tm +(missing) Tj +1 0 0 1 178.56 417.6 Tm +(information) Tj +1 0 0 1 232.68 417.6 Tm +(\() Tj +/R697 10 Tf +1 0 0 1 236.04 417.6 Tm +(g) Tj +1 0 0 1 240.96 417.6 Tm +(et) Tj +ET +q +3 0 0 -0.48 248.748 418.092 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +1 0 0 1 251.639 417.6 Tm +(c) Tj +1 0 0 1 255.959 417.6 Tm +(hec) Tj +1 0 0 1 269.639 417.6 Tm +(kpoint) Tj +/R694 10 Tf +1 0 0 1 294.719 417.6 Tm +(,) Tj +/R697 10 Tf +1 0 0 1 72 405.72 Tm +(set) Tj +ET +q +3 0 0 -0.48 83.628 406.212 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +1 0 0 1 86.6399 405.72 Tm +(c) Tj +1 0 0 1 90.9599 405.72 Tm +(hec) Tj +1 0 0 1 104.64 405.72 Tm +(kpoint) Tj +/R694 10 Tf +1 0 0 1 129.72 405.72 Tm +(\).) Tj +1 0 0 1 138.84 405.72 Tm +(The) Tj +/R697 10 Tf +1 0 0 1 156.72 405.72 Tm +(e) Tj +1 0 0 1 160.92 405.72 Tm +(xecute) Tj +/R694 10 Tf +1 0 0 1 188.88 405.72 Tm +(procedure) Tj +1 0 0 1 231.12 405.72 Tm +(recei) Tj +1 0 0 1 250.319 405.72 Tm +(v) Tj +1 0 0 1 255.239 405.72 Tm +(es) Tj +1 0 0 1 265.799 405.72 Tm +(as) Tj +1 0 0 1 276.479 405.72 Tm +(input) Tj +1 0 0 1 72 393.72 Tm +(a) Tj +1 0 0 1 78.96 393.72 Tm +(b) Tj +1 0 0 1 83.7599 393.72 Tm +(uf) Tj +1 0 0 1 91.9199 393.72 Tm +(fer) Tj +1 0 0 1 105.48 393.72 Tm +(containing) Tj +1 0 0 1 150.12 393.72 Tm +(the) Tj +1 0 0 1 164.76 393.72 Tm +(requested) Tj +1 0 0 1 205.56 393.72 Tm +(operation,) Tj +1 0 0 1 248.279 393.72 Tm +(e) Tj +1 0 0 1 252.599 393.72 Tm +(x) Tj +1 0 0 1 257.519 393.72 Tm +(ecutes) Tj +1 0 0 1 284.879 393.72 Tm +(the) Tj +1 0 0 1 72 381.84 Tm +(operation,) Tj +1 0 0 1 115.08 381.84 Tm +(and) Tj +1 0 0 1 132.24 381.84 Tm +(places) Tj +1 0 0 1 159.96 381.84 Tm +(the) Tj +1 0 0 1 174.96 381.84 Tm +(result) Tj +1 0 0 1 199.92 381.84 Tm +(in) Tj +1 0 0 1 210.479 381.84 Tm +(an) Tj +1 0 0 1 222.719 381.84 Tm +(output) Tj +1 0 0 1 251.039 381.84 Tm +(b) Tj +1 0 0 1 255.839 381.84 Tm +(uf) Tj +1 0 0 1 263.999 381.84 Tm +(fer) Tj +1 0 0 1 274.559 381.84 Tm +(.) Tj +1 0 0 1 281.519 381.84 Tm +(The) Tj +1 0 0 1 72 369.84 Tm +(other) Tj +1 0 0 1 96.7199 369.84 Tm +(procedures) Tj +1 0 0 1 144.6 369.84 Tm +(are) Tj +1 0 0 1 160.92 369.84 Tm +(discussed) Tj +1 0 0 1 203.28 369.84 Tm +(further) Tj +1 0 0 1 234.599 369.84 Tm +(in) Tj +1 0 0 1 246.599 369.84 Tm +(Sections) Tj +1 0 0 1 284.519 369.84 Tm +(6.3) Tj +1 0 0 1 72 357.96 Tm +(and) Tj +1 0 0 1 88.9199 357.96 Tm +(6.4.) Tj +1 0 0 1 81.96 345.84 Tm +(Point-to-point) Tj +1 0 0 1 139.8 345.84 Tm +(communication) Tj +1 0 0 1 203.76 345.84 Tm +(betwee) Tj +1 0 0 1 231.96 345.84 Tm +(n) Tj +1 0 0 1 238.56 345.84 Tm +(no) Tj +1 0 0 1 248.52 345.84 Tm +(des) Tj +1 0 0 1 263.399 345.84 Tm +(is) Tj +1 0 0 1 271.559 345.84 Tm +(imple-) Tj +1 0 0 1 72 333.96 Tm +(mented) Tj +1 0 0 1 103.44 333.96 Tm +(using) Tj +1 0 0 1 127.2 333.96 Tm +(UDP) Tj +1 0 0 1 146.04 333.96 Tm +(,) Tj +1 0 0 1 150.72 333.96 Tm +(and) Tj +1 0 0 1 167.28 333.96 Tm +(multicast) Tj +1 0 0 1 205.92 333.96 Tm +(to) Tj +1 0 0 1 215.88 333.96 Tm +(the) Tj +1 0 0 1 230.16 333.96 Tm +(group) Tj +1 0 0 1 255.599 333.96 Tm +(of) Tj +1 0 0 1 266.039 333.96 Tm +(replicas) Tj +1 0 0 1 72 321.96 Tm +(is) Tj +1 0 0 1 81.48 321.96 Tm +(implemented) Tj +1 0 0 1 136.44 321.96 Tm +(using) Tj +1 0 0 1 160.8 321.96 Tm +(UDP) Tj +1 0 0 1 183.6 321.96 Tm +(o) Tj +1 0 0 1 188.52 321.96 Tm +(v) Tj +1 0 0 1 193.44 321.96 Tm +(er) Tj +1 0 0 1 203.88 321.96 Tm +(IP) Tj +1 0 0 1 215.52 321.96 Tm +(multicast) Tj +1 0 0 1 254.999 321.96 Tm +([7) Tj +1 0 0 1 263.279 321.96 Tm +(].) Tj +1 0 0 1 273.719 321.96 Tm +(There) Tj +1 0 0 1 72 310.08 Tm +(is) Tj +1 0 0 1 80.52 310.08 Tm +(a) Tj +1 0 0 1 86.7599 310.08 Tm +(single) Tj +1 0 0 1 112.44 310.08 Tm +(IP) Tj +1 0 0 1 123.24 310.08 Tm +(multicast) Tj +1 0 0 1 161.64 310.08 Tm +(group) Tj +1 0 0 1 186.72 310.08 Tm +(for) Tj +1 0 0 1 200.279 310.08 Tm +(each) Tj +1 0 0 1 220.319 310.08 Tm +(service,) Tj +1 0 0 1 253.079 310.08 Tm +(which) Tj +1 0 0 1 279.239 310.08 Tm +(con-) Tj +1 0 0 1 72 298.08 Tm +(tains) Tj +1 0 0 1 92.8799 298.08 Tm +(all) Tj +1 0 0 1 104.76 298.08 Tm +(the) Tj +1 0 0 1 119.04 298.08 Tm +(replicas.) Tj +1 0 0 1 155.88 298.08 Tm +(These) Tj +1 0 0 1 181.68 298.08 Tm +(communication) Tj +1 0 0 1 245.759 298.08 Tm +(protocols) Tj +1 0 0 1 284.879 298.08 Tm +(are) Tj +1 0 0 1 72 286.2 Tm +(unreliable;) Tj +1 0 0 1 116.64 286.2 Tm +(the) Tj +1 0 0 1 128.76 286.2 Tm +(y) Tj +1 0 0 1 135.36 286.2 Tm +(may) Tj +1 0 0 1 154.2 286.2 Tm +(duplicate) Tj +1 0 0 1 192.48 286.2 Tm +(or) Tj +1 0 0 1 202.56 286.2 Tm +(lose) Tj +1 0 0 1 220.319 286.2 Tm +(messages) Tj +1 0 0 1 259.679 286.2 Tm +(or) Tj +1 0 0 1 269.759 286.2 Tm +(deli) Tj +1 0 0 1 284.519 286.2 Tm +(v) Tj +1 0 0 1 289.439 286.2 Tm +(er) Tj +1 0 0 1 72 274.2 Tm +(them) Tj +1 0 0 1 94.4399 274.2 Tm +(out) Tj +1 0 0 1 109.68 274.2 Tm +(of) Tj +1 0 0 1 120.48 274.2 Tm +(order) Tj +1 0 0 1 141.12 274.2 Tm +(.) Tj +1 0 0 1 81.96 262.2 Tm +(The) Tj +1 0 0 1 103.8 262.2 Tm +(algorithm) Tj +1 0 0 1 148.92 262.2 Tm +(tolerates) Tj +1 0 0 1 189 262.2 Tm +(out-of-order) Tj +1 0 0 1 243.959 262.2 Tm +(deli) Tj +1 0 0 1 258.719 262.2 Tm +(v) Tj +1 0 0 1 263.639 262.2 Tm +(ery) Tj +1 0 0 1 282.599 262.2 Tm +(and) Tj +1 0 0 1 72 250.2 Tm +(rejects) Tj +1 0 0 1 100.8 250.2 Tm +(duplicates.) Tj +1 0 0 1 148.32 250.2 Tm +(V) Tj +1 0 0 1 154.92 250.2 Tm +(ie) Tj +1 0 0 1 161.88 250.2 Tm +(w) Tj +1 0 0 1 171.96 250.2 Tm +(changes) Tj +1 0 0 1 206.88 250.2 Tm +(can) Tj +1 0 0 1 223.56 250.2 Tm +(be) Tj +1 0 0 1 235.8 250.2 Tm +(used) Tj +1 0 0 1 256.799 250.2 Tm +(to) Tj +1 0 0 1 267.479 250.2 Tm +(reco) Tj +1 0 0 1 284.639 250.2 Tm +(v) Tj +1 0 0 1 289.559 250.2 Tm +(er) Tj +1 0 0 1 72 238.32 Tm +(from) Tj +1 0 0 1 93.8399 238.32 Tm +(lost) Tj +1 0 0 1 110.76 238.32 Tm +(messages,) Tj +1 0 0 1 153.36 238.32 Tm +(b) Tj +1 0 0 1 158.16 238.32 Tm +(ut) Tj +1 0 0 1 168.36 238.32 Tm +(this) Tj +1 0 0 1 185.28 238.32 Tm +(is) Tj +1 0 0 1 194.4 238.32 Tm +(e) Tj +1 0 0 1 198.72 238.32 Tm +(xpensi) Tj +1 0 0 1 224.64 238.32 Tm +(v) Tj +1 0 0 1 229.56 238.32 Tm +(e) Tj +1 0 0 1 236.159 238.32 Tm +(and) Tj +1 0 0 1 253.079 238.32 Tm +(therefore) Tj +1 0 0 1 291.479 238.32 Tm +(it) Tj +1 0 0 1 72 226.32 Tm +(is) Tj +1 0 0 1 81.48 226.32 Tm +(important) Tj +1 0 0 1 123.12 226.32 Tm +(to) Tj +1 0 0 1 133.68 226.32 Tm +(perform) Tj +1 0 0 1 168.6 226.32 Tm +(retransmissions.) Tj +1 0 0 1 237.719 226.32 Tm +(During) Tj +1 0 0 1 268.799 226.32 Tm +(normal) Tj +1 0 0 1 72 214.44 Tm +(operation) Tj +1 0 0 1 115.32 214.44 Tm +(reco) Tj +1 0 0 1 132.48 214.44 Tm +(v) Tj +1 0 0 1 137.4 214.44 Tm +(ery) Tj +1 0 0 1 155.52 214.44 Tm +(from) Tj +1 0 0 1 180.6 214.44 Tm +(lost) Tj +1 0 0 1 200.64 214.44 Tm +(messages) Tj +1 0 0 1 243.96 214.44 Tm +(is) Tj +1 0 0 1 256.319 214.44 Tm +(dri) Tj +1 0 0 1 267.239 214.44 Tm +(v) Tj +1 0 0 1 272.159 214.44 Tm +(en) Tj +1 0 0 1 287.039 214.44 Tm +(by) Tj +1 0 0 1 72 202.44 Tm +(the) Tj +1 0 0 1 88.1999 202.44 Tm +(recei) Tj +1 0 0 1 107.4 202.44 Tm +(v) Tj +1 0 0 1 112.32 202.44 Tm +(er:) Tj +1 0 0 1 129.24 202.44 Tm +(backups) Tj +1 0 0 1 165.84 202.44 Tm +(send) Tj +1 0 0 1 188.16 202.44 Tm +(ne) Tj +1 0 0 1 197.52 202.44 Tm +(gati) Tj +1 0 0 1 212.28 202.44 Tm +(v) Tj +1 0 0 1 217.2 202.44 Tm +(e) Tj +1 0 0 1 225.359 202.44 Tm +(ackno) Tj +1 0 0 1 249.119 202.44 Tm +(wledgments) Tj +1 0 0 1 72 190.56 Tm +(to) Tj +1 0 0 1 85.56 190.56 Tm +(the) Tj +1 0 0 1 103.68 190.56 Tm +(primary) Tj +1 0 0 1 141 190.56 Tm +(when) Tj +1 0 0 1 168.48 190.56 Tm +(the) Tj +1 0 0 1 180.6 190.56 Tm +(y) Tj +1 0 0 1 191.4 190.56 Tm +(are) Tj +1 0 0 1 209.4 190.56 Tm +(out) Tj +1 0 0 1 228 190.56 Tm +(of) Tj +1 0 0 1 242.159 190.56 Tm +(date) Tj +1 0 0 1 264.599 190.56 Tm +(and) Tj +1 0 0 1 284.879 190.56 Tm +(the) Tj +1 0 0 1 72 178.56 Tm +(primary) Tj +1 0 0 1 107.76 178.56 Tm +(retransmits) Tj +1 0 0 1 156.24 178.56 Tm +(pre-prepare) Tj +1 0 0 1 206.4 178.56 Tm +(messages) Tj +1 0 0 1 248.159 178.56 Tm +(after) Tj +1 0 0 1 270.599 178.56 Tm +(a) Tj +1 0 0 1 279.239 178.56 Tm +(long) Tj +1 0 0 1 72 166.68 Tm +(timeout.) Tj +1 0 0 1 113.76 166.68 Tm +(A) Tj +1 0 0 1 125.16 166.68 Tm +(reply) Tj +1 0 0 1 149.88 166.68 Tm +(to) Tj +1 0 0 1 161.76 166.68 Tm +(a) Tj +1 0 0 1 170.4 166.68 Tm +(ne) Tj +1 0 0 1 179.76 166.68 Tm +(gati) Tj +1 0 0 1 194.52 166.68 Tm +(v) Tj +1 0 0 1 199.44 166.68 Tm +(e) Tj +1 0 0 1 207.84 166.68 Tm +(ackno) Tj +1 0 0 1 231.6 166.68 Tm +(wledgment) Tj +1 0 0 1 279.839 166.68 Tm +(may) Tj +1 0 0 1 72 154.68 Tm +(include) Tj +1 0 0 1 103.56 154.68 Tm +(both) Tj +1 0 0 1 123.36 154.68 Tm +(a) Tj +1 0 0 1 130.08 154.68 Tm +(portion) Tj +1 0 0 1 161.04 154.68 Tm +(of) Tj +1 0 0 1 171.48 154.68 Tm +(a) Tj +1 0 0 1 178.08 154.68 Tm +(stable) Tj +1 0 0 1 203.52 154.68 Tm +(checkpoint) Tj +1 0 0 1 249.479 154.68 Tm +(and) Tj +1 0 0 1 266.039 154.68 Tm +(missing) Tj +1 0 0 1 72 142.8 Tm +(messages.) Tj +1 0 0 1 123.12 142.8 Tm +(During) Tj +1 0 0 1 156.36 142.8 Tm +(vie) Tj +1 0 0 1 168.36 142.8 Tm +(w) Tj +1 0 0 1 180.48 142.8 Tm +(changes,) Tj +1 0 0 1 220.68 142.8 Tm +(replicas) Tj +1 0 0 1 256.56 142.8 Tm +(retransmit) Tj +1 0 0 1 72 130.8 Tm +(vie) Tj +1 0 0 1 84 130.8 Tm +(w-change) Tj +1 0 0 1 124.32 130.8 Tm +(messages) Tj +1 0 0 1 163.68 130.8 Tm +(until) Tj +1 0 0 1 183.6 130.8 Tm +(the) Tj +1 0 0 1 195.72 130.8 Tm +(y) Tj +1 0 0 1 202.32 130.8 Tm +(recei) Tj +1 0 0 1 221.52 130.8 Tm +(v) Tj +1 0 0 1 226.44 130.8 Tm +(e) Tj +1 0 0 1 232.44 130.8 Tm +(a) Tj +1 0 0 1 238.559 130.8 Tm +(matching) Tj +1 0 0 1 277.319 130.8 Tm +(ne) Tj +1 0 0 1 286.559 130.8 Tm +(w-) Tj +1 0 0 1 72 118.92 Tm +(vie) Tj +1 0 0 1 84 118.92 Tm +(w) Tj +1 0 0 1 93.5999 118.92 Tm +(message) Tj +1 0 0 1 129.84 118.92 Tm +(or) Tj +1 0 0 1 140.64 118.92 Tm +(the) Tj +1 0 0 1 152.76 118.92 Tm +(y) Tj +1 0 0 1 160.08 118.92 Tm +(mo) Tj +1 0 0 1 172.8 118.92 Tm +(v) Tj +1 0 0 1 177.72 118.92 Tm +(e) Tj +1 0 0 1 184.44 118.92 Tm +(on) Tj +1 0 0 1 196.92 118.92 Tm +(to) Tj +1 0 0 1 207.12 118.92 Tm +(a) Tj +1 0 0 1 214.079 118.92 Tm +(later) Tj +1 0 0 1 234.239 118.92 Tm +(vie) Tj +1 0 0 1 246.239 118.92 Tm +(w) Tj +1 0 0 1 252.839 118.92 Tm +(.) Tj +1 0 0 1 81.96 106.8 Tm +(The) Tj +1 0 0 1 103.68 106.8 Tm +(replication) Tj +1 0 0 1 152.64 106.8 Tm +(library) Tj +1 0 0 1 185.4 106.8 Tm +(does) Tj +1 0 0 1 210 106.8 Tm +(not) Tj +1 0 0 1 228.96 106.8 Tm +(implement) Tj +1 0 0 1 277.919 106.8 Tm +(vie) Tj +1 0 0 1 289.919 106.8 Tm +(w) Tj +1 0 0 1 72 94.92 Tm +(changes) Tj +1 0 0 1 110.64 94.92 Tm +(or) Tj +1 0 0 1 125.4 94.92 Tm +(retransmissions) Tj +1 0 0 1 193.92 94.92 Tm +(at) Tj +1 0 0 1 207.6 94.92 Tm +(present.) Tj +1 0 0 1 254.52 94.92 Tm +(This) Tj +1 0 0 1 278.759 94.92 Tm +(does) Tj +1 0 0 1 72 82.92 Tm +(not) Tj +1 0 0 1 91.4399 82.92 Tm +(compromise) Tj +1 0 0 1 147.48 82.92 Tm +(the) Tj +1 0 0 1 166.44 82.92 Tm +(accurac) Tj +1 0 0 1 196.92 82.92 Tm +(y) Tj +1 0 0 1 208.44 82.92 Tm +(of) Tj +1 0 0 1 223.56 82.92 Tm +(the) Tj +1 0 0 1 242.52 82.92 Tm +(results) Tj +1 0 0 1 275.279 82.92 Tm +(gi) Tj +1 0 0 1 282.839 82.92 Tm +(v) Tj +1 0 0 1 287.759 82.92 Tm +(en) Tj +1 0 0 1 72 71.04 Tm +(in) Tj +1 0 0 1 86.28 71.04 Tm +(Section) Tj +1 0 0 1 122.64 71.04 Tm +(7) Tj +1 0 0 1 134.04 71.04 Tm +(because) Tj +1 0 0 1 172.08 71.04 Tm +(the) Tj +1 0 0 1 190.8 71.04 Tm +(rest) Tj +1 0 0 1 211.68 71.04 Tm +(of) Tj +1 0 0 1 226.44 71.04 Tm +(the) Tj +1 0 0 1 245.159 71.04 Tm +(algorithm) Tj +1 0 0 1 290.399 71.04 Tm +(is) Tj +1 0 0 1 315 709.08 Tm +(completely) Tj +1 0 0 1 362.16 709.08 Tm +(implemented) Tj +1 0 0 1 417.12 709.08 Tm +(\(including) Tj +1 0 0 1 461.04 709.08 Tm +(the) Tj +1 0 0 1 476.16 709.08 Tm +(manipulation) Tj +1 0 0 1 531.719 709.08 Tm +(of) Tj +1 0 0 1 315 697.08 Tm +(the) Tj +1 0 0 1 331.44 697.08 Tm +(timers) Tj +1 0 0 1 360.72 697.08 Tm +(that) Tj +1 0 0 1 379.92 697.08 Tm +(trigger) Tj +1 0 0 1 410.76 697.08 Tm +(vie) Tj +1 0 0 1 422.76 697.08 Tm +(w) Tj +1 0 0 1 434.16 697.08 Tm +(changes\)) Tj +1 0 0 1 473.879 697.08 Tm +(and) Tj +1 0 0 1 492.599 697.08 Tm +(because) Tj +1 0 0 1 528.359 697.08 Tm +(we) Tj +1 0 0 1 315 685.2 Tm +(ha) Tj +1 0 0 1 324.24 685.2 Tm +(v) Tj +1 0 0 1 329.16 685.2 Tm +(e) Tj +1 0 0 1 336.96 685.2 Tm +(formalized) Tj +1 0 0 1 383.64 685.2 Tm +(the) Tj +1 0 0 1 399.36 685.2 Tm +(complete) Tj +1 0 0 1 439.44 685.2 Tm +(algorithm) Tj +1 0 0 1 481.799 685.2 Tm +(and) Tj +1 0 0 1 499.679 685.2 Tm +(pro) Tj +1 0 0 1 512.999 685.2 Tm +(v) Tj +1 0 0 1 517.919 685.2 Tm +(ed) Tj +1 0 0 1 530.639 685.2 Tm +(its) Tj +1 0 0 1 315 673.2 Tm +(correctness) Tj +1 0 0 1 362.28 673.2 Tm +([4) Tj +1 0 0 1 370.56 673.2 Tm +(].) Tj +/R691 10 Tf +1 0 0 1 315 650.04 Tm +(6.2) Tj +1 0 0 1 337.44 650.04 Tm +(BFS:) Tj +1 0 0 1 361.44 650.04 Tm +(A) Tj +1 0 0 1 371.16 650.04 Tm +(Byzantine-F) Tj +1 0 0 1 423.12 650.04 Tm +(ault-tolerant) Tj +1 0 0 1 479.16 650.04 Tm +(File) Tj +1 0 0 1 497.639 650.04 Tm +(System) Tj +/R694 10 Tf +1 0 0 1 315 635.04 Tm +(W) Tj +1 0 0 1 323.52 635.04 Tm +(e) Tj +1 0 0 1 332.28 635.04 Tm +(implemented) Tj +1 0 0 1 388.56 635.04 Tm +(BFS,) Tj +1 0 0 1 413.16 635.04 Tm +(a) Tj +1 0 0 1 421.8 635.04 Tm +(Byzantine-f) Tj +1 0 0 1 468.96 635.04 Tm +(ault-tolerant) Tj +1 0 0 1 521.759 635.04 Tm +(NFS) Tj +1 0 0 1 315 623.16 Tm +(service,) Tj +1 0 0 1 348.24 623.16 Tm +(using) Tj +1 0 0 1 372.48 623.16 Tm +(the) Tj +1 0 0 1 387.12 623.16 Tm +(replication) Tj +1 0 0 1 432.36 623.16 Tm +(library) Tj +1 0 0 1 458.519 623.16 Tm +(.) Tj +1 0 0 1 464.519 623.16 Tm +(Figure) Tj +1 0 0 1 493.079 623.16 Tm +(2) Tj +1 0 0 1 500.639 623.16 Tm +(sho) Tj +1 0 0 1 514.319 623.16 Tm +(ws) Tj +1 0 0 1 527.879 623.16 Tm +(the) Tj +1 0 0 1 315 611.16 Tm +(architecture) Tj +1 0 0 1 365.16 611.16 Tm +(of) Tj +1 0 0 1 376.56 611.16 Tm +(BFS.) Tj +1 0 0 1 399.84 611.16 Tm +(W) Tj +1 0 0 1 408.36 611.16 Tm +(e) Tj +1 0 0 1 416.04 611.16 Tm +(opted) Tj +1 0 0 1 441.24 611.16 Tm +(not) Tj +1 0 0 1 457.199 611.16 Tm +(to) Tj +1 0 0 1 467.999 611.16 Tm +(modify) Tj +1 0 0 1 499.919 611.16 Tm +(the) Tj +1 0 0 1 515.159 611.16 Tm +(k) Tj +1 0 0 1 520.079 611.16 Tm +(ernel) Tj +1 0 0 1 315 599.28 Tm +(NFS) Tj +1 0 0 1 334.92 599.28 Tm +(client) Tj +1 0 0 1 358.8 599.28 Tm +(and) Tj +1 0 0 1 374.88 599.28 Tm +(serv) Tj +1 0 0 1 391.44 599.28 Tm +(er) Tj +1 0 0 1 400.68 599.28 Tm +(because) Tj +1 0 0 1 433.92 599.28 Tm +(we) Tj +1 0 0 1 447.24 599.28 Tm +(did) Tj +1 0 0 1 461.639 599.28 Tm +(not) Tj +1 0 0 1 476.159 599.28 Tm +(ha) Tj +1 0 0 1 485.399 599.28 Tm +(v) Tj +1 0 0 1 490.319 599.28 Tm +(e) Tj +1 0 0 1 496.199 599.28 Tm +(the) Tj +1 0 0 1 510.119 599.28 Tm +(sources) Tj +1 0 0 1 315 587.28 Tm +(for) Tj +1 0 0 1 329.16 587.28 Tm +(the) Tj +1 0 0 1 343.8 587.28 Tm +(Digital) Tj +1 0 0 1 373.92 587.28 Tm +(Unix) Tj +1 0 0 1 396.36 587.28 Tm +(k) Tj +1 0 0 1 401.28 587.28 Tm +(ernel.) Tj +1 0 0 1 324.96 574.92 Tm +(A) Tj +1 0 0 1 333.72 574.92 Tm +(\256le) Tj +1 0 0 1 348 574.92 Tm +(system) Tj +1 0 0 1 377.28 574.92 Tm +(e) Tj +1 0 0 1 381.6 574.92 Tm +(xported) Tj +1 0 0 1 413.88 574.92 Tm +(by) Tj +1 0 0 1 425.52 574.92 Tm +(the) Tj +1 0 0 1 439.32 574.92 Tm +(f) Tj +1 0 0 1 442.44 574.92 Tm +(ault-to) Tj +1 0 0 1 468.479 574.92 Tm +(leran) Tj +1 0 0 1 488.399 574.92 Tm +(t) Tj +1 0 0 1 492.719 574.92 Tm +(NFS) Tj +1 0 0 1 512.519 574.92 Tm +(serv) Tj +1 0 0 1 529.079 574.92 Tm +(ice) Tj +1 0 0 1 315 563.04 Tm +(is) Tj +1 0 0 1 324.96 563.04 Tm +(mounted) Tj +1 0 0 1 363.24 563.04 Tm +(on) Tj +1 0 0 1 376.44 563.04 Tm +(the) Tj +1 0 0 1 392.04 563.04 Tm +(client) Tj +1 0 0 1 417.48 563.04 Tm +(machine) Tj +1 0 0 1 454.56 563.04 Tm +(lik) Tj +1 0 0 1 465 563.04 Tm +(e) Tj +1 0 0 1 472.8 563.04 Tm +(an) Tj +1 0 0 1 482.16 563.04 Tm +(y) Tj +1 0 0 1 490.319 563.04 Tm +(re) Tj +1 0 0 1 497.999 563.04 Tm +(gular) Tj +1 0 0 1 521.759 563.04 Tm +(NFS) Tj +1 0 0 1 315 551.04 Tm +(\256le) Tj +1 0 0 1 331.2 551.04 Tm +(system.) Tj +1 0 0 1 367.92 551.04 Tm +(Application) Tj +1 0 0 1 418.32 551.04 Tm +(processes) Tj +1 0 0 1 459.96 551.04 Tm +(run) Tj +1 0 0 1 476.76 551.04 Tm +(unmodi\256ed) Tj +1 0 0 1 525.599 551.04 Tm +(and) Tj +1 0 0 1 315 539.16 Tm +(interact) Tj +1 0 0 1 348.48 539.16 Tm +(with) Tj +1 0 0 1 369.72 539.16 Tm +(the) Tj +1 0 0 1 385.44 539.16 Tm +(mounted) Tj +1 0 0 1 423.84 539.16 Tm +(\256le) Tj +1 0 0 1 440.16 539.16 Tm +(system) Tj +1 0 0 1 471.479 539.16 Tm +(through) Tj +1 0 0 1 506.039 539.16 Tm +(the) Tj +1 0 0 1 521.759 539.16 Tm +(NFS) Tj +1 0 0 1 315 527.16 Tm +(client) Tj +1 0 0 1 339.36 527.16 Tm +(in) Tj +1 0 0 1 349.32 527.16 Tm +(the) Tj +1 0 0 1 363.72 527.16 Tm +(k) Tj +1 0 0 1 368.64 527.16 Tm +(ernel.) Tj +1 0 0 1 394.44 527.16 Tm +(W) Tj +1 0 0 1 402.96 527.16 Tm +(e) Tj +1 0 0 1 409.68 527.16 Tm +(rely) Tj +1 0 0 1 427.44 527.16 Tm +(on) Tj +1 0 0 1 439.559 527.16 Tm +(user) Tj +1 0 0 1 458.399 527.16 Tm +(le) Tj +1 0 0 1 465.359 527.16 Tm +(v) Tj +1 0 0 1 470.279 527.16 Tm +(el) Tj +/R697 10 Tf +1 0 0 1 479.519 527.16 Tm +(r) Tj +1 0 0 1 482.999 527.16 Tm +(elay) Tj +/R694 10 Tf +1 0 0 1 501.839 527.16 Tm +(processes) Tj +1 0 0 1 315 515.28 Tm +(to) Tj +1 0 0 1 327.12 515.28 Tm +(mediate) Tj +1 0 0 1 363.12 515.28 Tm +(communication) Tj +1 0 0 1 429.48 515.28 Tm +(between) Tj +1 0 0 1 467.04 515.28 Tm +(the) Tj +1 0 0 1 483.6 515.28 Tm +(standard) Tj +1 0 0 1 521.759 515.28 Tm +(NFS) Tj +1 0 0 1 315 503.28 Tm +(client) Tj +1 0 0 1 340.68 503.28 Tm +(and) Tj +1 0 0 1 358.68 503.28 Tm +(the) Tj +1 0 0 1 374.52 503.28 Tm +(replicas.) Tj +1 0 0 1 414.84 503.28 Tm +(A) Tj +1 0 0 1 425.64 503.28 Tm +(relay) Tj +1 0 0 1 449.16 503.28 Tm +(recei) Tj +1 0 0 1 468.359 503.28 Tm +(v) Tj +1 0 0 1 473.279 503.28 Tm +(es) Tj +1 0 0 1 484.919 503.28 Tm +(NFS) Tj +1 0 0 1 506.759 503.28 Tm +(protocol) Tj +1 0 0 1 315 491.4 Tm +(requests,) Tj +1 0 0 1 354.72 491.4 Tm +(calls) Tj +1 0 0 1 377.28 491.4 Tm +(the) Tj +/R697 10 Tf +1 0 0 1 393.6 491.4 Tm +(in) Tj +1 0 0 1 401.04 491.4 Tm +(vok) Tj +1 0 0 1 414.84 491.4 Tm +(e) Tj +/R694 10 Tf +1 0 0 1 423.36 491.4 Tm +(procedure) Tj +1 0 0 1 467.399 491.4 Tm +(of) Tj +1 0 0 1 479.879 491.4 Tm +(our) Tj +1 0 0 1 497.399 491.4 Tm +(replication) Tj +1 0 0 1 315 479.4 Tm +(library) Tj +1 0 0 1 341.16 479.4 Tm +(,) Tj +1 0 0 1 345.84 479.4 Tm +(and) Tj +1 0 0 1 362.76 479.4 Tm +(sends) Tj +1 0 0 1 387.36 479.4 Tm +(the) Tj +1 0 0 1 402.12 479.4 Tm +(result) Tj +1 0 0 1 426.72 479.4 Tm +(back) Tj +1 0 0 1 447.959 479.4 Tm +(to) Tj +1 0 0 1 458.279 479.4 Tm +(the) Tj +1 0 0 1 472.919 479.4 Tm +(NFS) Tj +1 0 0 1 493.679 479.4 Tm +(client.) Tj +ET +1 i +346.202 420.898 m +346.52 419.464 l +347.898 419.767 l +347.739 420.484 l +347.65 421.213 l +347.615 421.209 l +f +347.65 421.213 m +347.827 419.755 l +349.122 419.911 l +349.034 420.64 l +349 421.373 l +348.972 421.372 l +f +349 421.373 m +349.068 419.906 l +350.309 419.964 l +350.274 420.697 l +350.269 421.431 l +350.255 421.431 l +f +350.269 421.431 m +350.279 419.963 l +351.492 419.971 l +351.487 420.7 l +351.487 421.434 l +351.482 421.44 l +f +351.487 421.434 m +351.487 419.966 l +356.32 419.966 l +356.32 420.7 l +356.32 421.434 l +356.32 421.434 l +f +356.32 421.434 m +356.32 419.966 l +361.153 419.966 l +361.153 420.7 l +361.153 421.434 l +361.153 421.434 l +f +361.153 421.434 m +361.153 419.966 l +365.987 419.966 l +365.987 420.7 l +365.987 421.434 l +365.987 421.434 l +f +365.987 421.434 m +365.987 419.966 l +370.746 419.966 l +370.746 420.7 l +370.746 421.434 l +370.746 421.434 l +f +370.746 421.434 m +370.746 419.966 l +375.287 419.966 l +375.287 420.7 l +375.287 421.434 l +375.287 421.434 l +f +375.287 421.434 m +375.287 419.966 l +380.046 419.966 l +380.046 420.7 l +380.046 421.434 l +380.046 421.434 l +f +380.046 421.434 m +380.046 419.966 l +384.88 419.966 l +384.88 420.7 l +384.88 421.434 l +384.88 421.434 l +f +384.88 421.434 m +384.88 419.966 l +389.713 419.966 l +389.713 420.7 l +389.713 421.434 l +389.713 421.434 l +f +389.713 421.434 m +389.713 419.966 l +394.546 419.966 l +394.546 420.7 l +394.546 421.434 l +394.546 421.434 l +f +394.546 421.434 m +394.546 419.966 l +399.379 419.966 l +399.379 420.7 l +399.379 421.434 l +399.379 421.434 l +f +399.379 421.434 m +399.379 419.966 l +404.139 419.966 l +404.139 420.7 l +404.139 421.434 l +404.139 421.434 l +f +404.139 421.434 m +404.139 419.966 l +408.679 419.966 l +408.679 420.7 l +408.679 421.434 l +408.679 421.434 l +f +408.679 421.434 m +408.679 419.966 l +413.439 419.966 l +413.439 420.7 l +413.439 421.434 l +413.439 421.434 l +f +413.439 421.434 m +413.439 419.966 l +418.272 419.966 l +418.272 420.7 l +418.272 421.434 l +418.272 421.434 l +f +418.272 421.434 m +418.272 419.966 l +423.105 419.966 l +423.105 420.7 l +423.105 421.434 l +423.105 421.434 l +f +423.105 421.434 m +423.105 419.966 l +427.938 419.966 l +427.938 420.7 l +427.938 421.434 l +427.938 421.434 l +f +427.938 421.434 m +427.938 419.966 l +432.698 419.966 l +432.698 420.705 l +432.703 421.44 l +432.698 421.434 l +f +432.703 421.44 m +432.693 419.971 l +433.839 419.963 l +433.845 420.697 l +433.881 421.431 l +433.866 421.431 l +f +433.881 421.431 m +433.808 419.964 l +434.968 419.906 l +435.005 420.64 l +435.099 421.368 l +435.07 421.372 l +f +435.099 421.368 m +434.911 419.911 l +436.126 419.756 l +436.22 420.484 l +436.387 421.199 l +436.351 421.208 l +f +436.387 421.199 m +436.053 419.769 l +437.364 419.466 l +437.531 420.181 l +437.773 420.875 l +437.736 420.887 l +f +437.773 420.875 m +437.289 419.487 l +438.716 418.994 l +438.958 419.688 l +439.274 420.352 l +439.238 420.369 l +f +439.274 420.352 m +438.642 419.025 l +440.117 418.328 l +440.433 418.992 l +440.829 419.611 l +440.79 419.636 l +f +440.829 419.611 m +440.037 418.372 l +441.471 417.465 l +441.867 418.085 l +442.35 418.639 l +442.308 418.675 l +f +442.35 418.639 m +441.383 417.53 l +442.687 416.405 l +443.17 416.96 l +443.74 417.426 l +443.7 417.474 l +f +443.74 417.426 m +442.6 416.494 l +443.699 415.161 l +444.269 415.626 l +444.901 416.005 l +444.874 416.051 l +f +444.901 416.005 m +443.637 415.248 l +444.524 413.781 l +445.156 414.16 l +445.829 414.461 l +445.811 414.501 l +f +445.829 414.461 m +444.483 413.859 l +445.164 412.35 l +445.837 412.651 l +446.538 412.88 l +446.526 412.917 l +f +446.538 412.88 m +445.136 412.422 l +445.618 410.962 l +446.319 411.191 l +447.039 411.349 l +447.031 411.385 l +f +447.039 411.349 m +445.598 411.034 l +445.894 409.692 l +446.615 409.85 l +447.347 409.939 l +447.343 409.974 l +f +447.347 409.939 m +445.883 409.761 l +446.035 408.518 l +446.767 408.607 l +447.504 408.641 l +447.503 408.669 l +f +447.504 408.641 m +446.03 408.572 l +446.086 407.386 l +446.823 407.42 l +447.561 407.425 l +447.561 407.44 l +f +447.561 407.425 m +446.086 407.415 l +446.094 406.242 l +446.8 406.247 l +447.538 406.247 l +447.569 406.252 l +f +447.538 406.247 m +446.062 406.247 l +446.062 401.378 l +446.8 401.378 l +447.538 401.378 l +447.538 401.378 l +f +447.538 401.378 m +446.062 401.378 l +446.062 396.434 l +446.8 396.434 l +447.538 396.434 l +447.538 396.434 l +f +447.538 396.434 m +446.062 396.434 l +446.062 391.489 l +446.8 391.489 l +447.538 391.489 l +447.538 391.489 l +f +447.538 391.489 m +446.062 391.489 l +446.062 386.545 l +446.8 386.545 l +447.538 386.545 l +447.538 386.545 l +f +447.538 386.545 m +446.062 386.545 l +446.062 381.601 l +446.8 381.601 l +447.538 381.601 l +447.538 381.601 l +f +447.538 381.601 m +446.062 381.601 l +446.062 376.732 l +446.8 376.732 l +447.538 376.732 l +447.538 376.732 l +f +447.538 376.732 m +446.062 376.732 l +446.062 372.087 l +446.831 372.087 l +447.569 372.082 l +447.538 372.087 l +f +447.569 372.082 m +446.094 372.092 l +446.086 370.895 l +446.823 370.89 l +447.56 370.857 l +447.561 370.885 l +f +447.56 370.857 m +446.086 370.922 l +446.03 369.676 l +446.767 369.643 l +447.5 369.559 l +447.503 369.585 l +f +447.5 369.559 m +446.034 369.727 l +445.882 368.41 l +446.615 368.326 l +447.337 368.176 l +447.344 368.209 l +f +447.337 368.176 m +445.893 368.477 l +445.597 367.069 l +446.319 366.918 l +447.021 366.695 l +447.033 366.731 l +f +447.021 366.695 m +445.616 367.141 l +445.134 365.638 l +445.837 365.415 l +446.511 365.118 l +446.528 365.154 l +f +446.511 365.118 m +445.162 365.712 l +444.482 364.181 l +445.156 363.884 l +445.789 363.507 l +445.812 363.546 l +f +445.789 363.507 m +444.523 364.261 l +443.636 362.786 l +444.269 362.409 l +444.84 361.944 l +444.874 361.986 l +f +444.84 361.944 m +443.699 362.875 l +442.599 361.541 l +443.17 361.075 l +443.653 360.52 l +443.7 360.561 l +f +443.653 360.52 m +442.687 361.63 l +441.383 360.505 l +441.867 359.95 l +442.262 359.33 l +442.308 359.359 l +f +442.262 359.33 m +441.471 360.57 l +440.037 359.663 l +440.433 359.043 l +440.749 358.379 l +440.79 358.399 l +f +440.749 358.379 m +440.117 359.706 l +438.642 359.01 l +438.958 358.346 l +439.2 357.653 l +439.238 357.666 l +f +439.2 357.653 m +438.716 359.04 l +437.289 358.547 l +437.531 357.854 l +437.698 357.138 l +437.736 357.147 l +f +437.698 357.138 m +437.364 358.569 l +436.053 358.266 l +436.22 357.551 l +436.314 356.822 l +436.351 356.827 l +f +436.314 356.822 m +436.126 358.279 l +434.911 358.123 l +435.005 357.395 l +435.041 356.661 l +435.07 356.663 l +f +435.041 356.661 m +434.968 358.128 l +433.808 358.071 l +433.845 357.337 l +433.85 356.603 l +433.866 356.603 l +f +433.85 356.603 m +433.839 358.072 l +432.693 358.064 l +432.698 357.3 l +432.698 356.566 l +432.703 356.595 l +f +432.698 356.566 m +432.698 358.034 l +427.938 358.034 l +427.938 357.3 l +427.938 356.566 l +427.938 356.566 l +f +427.938 356.566 m +427.938 358.034 l +423.105 358.034 l +423.105 357.3 l +423.105 356.566 l +423.105 356.566 l +f +423.105 356.566 m +423.105 358.034 l +418.272 358.034 l +418.272 357.3 l +418.272 356.566 l +418.272 356.566 l +f +418.272 356.566 m +418.272 358.034 l +413.439 358.034 l +413.439 357.3 l +413.439 356.566 l +413.439 356.566 l +f +413.439 356.566 m +413.439 358.034 l +408.679 358.034 l +408.679 357.3 l +408.679 356.566 l +408.679 356.566 l +f +408.679 356.566 m +408.679 358.034 l +404.139 358.034 l +404.139 357.3 l +404.139 356.566 l +404.139 356.566 l +f +404.139 356.566 m +404.139 358.034 l +399.379 358.034 l +399.379 357.3 l +399.379 356.566 l +399.379 356.566 l +f +399.379 356.566 m +399.379 358.034 l +394.546 358.034 l +394.546 357.3 l +394.546 356.566 l +394.546 356.566 l +f +394.546 356.566 m +394.546 358.034 l +389.713 358.034 l +389.713 357.3 l +389.713 356.566 l +389.713 356.566 l +f +389.713 356.566 m +389.713 358.034 l +384.88 358.034 l +384.88 357.3 l +384.88 356.566 l +384.88 356.566 l +f +384.88 356.566 m +384.88 358.034 l +380.046 358.034 l +380.046 357.3 l +380.046 356.566 l +380.046 356.566 l +f +380.046 356.566 m +380.046 358.034 l +375.287 358.034 l +375.287 357.3 l +375.287 356.566 l +375.287 356.566 l +f +375.287 356.566 m +375.287 358.034 l +370.746 358.034 l +370.746 357.3 l +370.746 356.566 l +370.746 356.566 l +f +370.746 356.566 m +370.746 358.034 l +365.987 358.034 l +365.987 357.3 l +365.987 356.566 l +365.987 356.566 l +f +365.987 356.566 m +365.987 358.034 l +361.153 358.034 l +361.153 357.3 l +361.153 356.566 l +361.153 356.566 l +f +361.153 356.566 m +361.153 358.034 l +356.32 358.034 l +356.32 357.3 l +356.32 356.566 l +356.32 356.566 l +f +356.32 356.566 m +356.32 358.034 l +351.487 358.034 l +351.487 357.329 l +351.482 356.595 l +351.487 356.566 l +f +351.482 356.595 m +351.492 358.064 l +350.279 358.072 l +350.274 357.337 l +350.24 356.604 l +350.255 356.603 l +f +350.24 356.604 m +350.309 358.071 l +349.068 358.128 l +349.034 357.395 l +348.946 356.666 l +348.972 356.663 l +f +348.946 356.666 m +349.122 358.124 l +347.827 358.28 l +347.739 357.551 l +347.58 356.833 l +347.615 356.826 l +f +347.58 356.833 m +347.898 358.268 l +346.52 358.571 l +346.361 357.854 l +346.125 357.158 l +346.163 357.145 l +f +346.125 357.158 m +346.597 358.55 l +345.129 359.042 l +344.893 358.346 l +344.579 357.682 l +344.618 357.664 l +f +344.579 357.682 m +345.207 359.011 l +343.718 359.708 l +343.405 359.043 l +343.006 358.425 l +343.047 358.399 l +f +343.006 358.425 m +343.803 359.661 l +342.383 360.568 l +341.985 359.95 l +341.493 359.403 l +341.537 359.364 l +f +341.493 359.403 m +342.477 360.497 l +341.215 361.622 l +340.723 361.075 l +340.138 360.627 l +340.18 360.573 l +f +340.138 360.627 m +341.307 361.523 l +340.274 362.858 l +339.689 362.409 l +339.041 362.058 l +339.069 362.007 l +f +339.041 362.058 m +340.337 362.76 l +339.531 364.235 l +338.883 363.884 l +338.196 363.617 l +338.212 363.574 l +f +338.196 363.617 m +339.57 364.151 l +338.969 365.682 l +338.282 365.415 l +337.571 365.22 l +337.581 365.184 l +f +337.571 365.22 m +338.993 365.61 l +338.578 367.113 l +337.867 366.918 l +337.141 366.788 l +337.147 366.755 l +f +337.141 366.788 m +338.593 367.048 l +338.339 368.456 l +337.613 368.326 l +336.879 368.254 l +336.881 368.225 l +f +336.879 368.254 m +338.347 368.398 l +338.216 369.715 l +337.482 369.643 l +336.745 369.615 l +336.746 369.593 l +f +336.745 369.615 m +338.219 369.671 l +338.171 370.918 l +337.434 370.89 l +336.696 370.885 l +336.697 370.861 l +f +336.696 370.885 m +338.172 370.894 l +338.165 372.091 l +337.4 372.087 l +336.662 372.087 l +336.69 372.083 l +f +336.662 372.087 m +338.138 372.087 l +338.138 376.732 l +337.4 376.732 l +336.662 376.732 l +336.662 376.732 l +f +336.662 376.732 m +338.138 376.732 l +338.138 381.601 l +337.4 381.601 l +336.662 381.601 l +336.662 381.601 l +f +336.662 381.601 m +338.138 381.601 l +338.138 386.545 l +337.4 386.545 l +336.662 386.545 l +336.662 386.545 l +f +336.662 386.545 m +338.138 386.545 l +338.138 391.489 l +337.4 391.489 l +336.662 391.489 l +336.662 391.489 l +f +336.662 391.489 m +338.138 391.489 l +338.138 396.434 l +337.4 396.434 l +336.662 396.434 l +336.662 396.434 l +f +336.662 396.434 m +338.138 396.434 l +338.138 401.378 l +337.4 401.378 l +336.662 401.378 l +336.662 401.378 l +f +336.662 401.378 m +338.138 401.378 l +338.138 406.247 l +337.427 406.247 l +336.69 406.251 l +336.662 406.247 l +f +336.69 406.251 m +338.165 406.243 l +338.172 407.416 l +337.434 407.42 l +336.697 407.45 l +336.696 407.424 l +f +336.697 407.45 m +338.171 407.39 l +338.219 408.577 l +337.482 408.607 l +336.749 408.683 l +336.746 408.66 l +f +336.749 408.683 m +338.216 408.531 l +338.346 409.774 l +337.613 409.85 l +336.888 409.986 l +336.882 409.956 l +f +336.888 409.986 m +338.338 409.714 l +338.592 411.055 l +337.867 411.191 l +337.157 411.392 l +337.148 411.36 l +f +337.157 411.392 m +338.576 410.991 l +338.992 412.451 l +338.282 412.651 l +337.596 412.922 l +337.582 412.887 l +f +337.596 412.922 m +338.968 412.381 l +339.568 413.889 l +338.883 414.16 l +338.236 414.512 l +338.214 414.473 l +f +338.236 414.512 m +339.53 413.807 l +340.336 415.274 l +339.689 415.626 l +339.105 416.075 l +339.07 416.03 l +f +339.105 416.075 m +340.273 415.178 l +341.307 416.511 l +340.723 416.96 l +340.23 417.507 l +340.18 417.462 l +f +340.23 417.507 m +341.215 416.413 l +342.477 417.537 l +341.985 418.085 l +341.586 418.703 l +341.536 418.671 l +f +341.586 418.703 m +342.383 417.467 l +343.803 418.374 l +343.405 418.992 l +343.091 419.656 l +343.047 419.636 l +f +343.091 419.656 m +343.718 418.327 l +345.207 419.024 l +344.893 419.688 l +344.657 420.384 l +344.617 420.371 l +f +344.657 420.384 m +345.129 418.992 l +346.597 419.485 l +346.125 420.877 l +f +348.232 409.726 m +348.43 409.019 l +350.086 409.48 l +349.986 409.834 l +349.971 410.201 l +349.928 410.199 l +f +349.971 410.201 m +350.001 409.467 l +351.656 409.533 l +351.641 409.9 l +351.641 410.267 l +351.627 410.266 l +f +351.641 410.267 m +351.641 409.533 l +354.952 409.533 l +354.952 409.9 l +354.952 410.267 l +354.952 410.267 l +f +354.952 410.267 m +354.952 409.533 l +358.262 409.533 l +358.262 409.9 l +358.262 410.267 l +358.262 410.267 l +f +358.262 410.267 m +358.262 409.533 l +361.572 409.533 l +361.572 409.9 l +361.572 410.267 l +361.572 410.267 l +f +361.572 410.267 m +361.572 409.533 l +364.883 409.533 l +364.883 409.9 l +364.883 410.267 l +364.883 410.267 l +f +364.883 410.267 m +364.883 409.533 l +368.193 409.533 l +368.193 409.9 l +368.193 410.267 l +368.193 410.267 l +f +368.193 410.267 m +368.193 409.533 l +371.503 409.533 l +371.503 409.9 l +371.503 410.267 l +371.503 410.267 l +f +371.503 410.267 m +371.503 409.533 l +373.176 409.533 l +373.176 409.9 l +373.176 410.267 l +373.176 410.267 l +f +373.176 410.267 m +373.176 409.533 l +374.952 409.533 l +374.952 409.9 l +374.952 410.267 l +374.952 410.267 l +f +374.952 410.267 m +374.952 409.533 l +376.883 409.533 l +376.883 409.9 l +376.883 410.267 l +376.883 410.267 l +f +376.883 410.267 m +376.883 409.533 l +378.814 409.533 l +378.814 409.9 l +378.817 410.267 l +378.814 410.267 l +f +378.817 410.267 m +378.81 409.532 l +379.718 409.524 l +379.721 409.891 l +379.746 410.258 l +379.724 410.258 l +f +379.746 410.258 m +379.697 409.525 l +380.565 409.467 l +380.59 409.834 l +380.657 410.195 l +380.636 410.199 l +f +380.657 410.195 m +380.522 409.473 l +381.365 409.316 l +381.432 409.677 l +381.56 410.022 l +381.531 410.033 l +f +381.56 410.022 m +381.305 409.333 l +382.135 409.028 l +382.262 409.373 l +382.451 409.688 l +382.422 409.706 l +f +382.451 409.688 m +382.073 409.057 l +382.892 408.572 l +383.081 408.887 l +383.317 409.17 l +383.294 409.188 l +f +383.317 409.17 m +382.846 408.605 l +383.613 407.971 l +383.848 408.254 l +384.122 408.5 l +384.104 408.519 l +f +384.122 408.5 m +383.574 408.008 l +384.238 407.275 l +384.512 407.521 l +384.822 407.721 l +384.806 407.745 l +f +384.822 407.721 m +384.202 407.322 l +384.711 406.54 l +385.021 406.739 l +385.363 406.876 l +385.35 406.909 l +f +385.363 406.876 m +384.678 406.602 l +384.997 405.812 l +385.34 405.949 l +385.701 406.023 l +385.694 406.056 l +f +385.701 406.023 m +384.978 405.875 l +385.142 405.085 l +385.503 405.159 l +385.871 405.187 l +385.869 405.21 l +f +385.871 405.187 m +385.136 405.131 l +385.196 404.341 l +385.564 404.369 l +385.933 404.373 l +385.932 404.397 l +f +385.933 404.373 m +385.195 404.365 l +385.204 403.575 l +385.6 403.579 l +385.969 403.579 l +385.941 403.583 l +f +385.969 403.579 m +385.231 403.579 l +385.231 400.419 l +385.6 400.419 l +385.969 400.419 l +385.969 400.419 l +f +385.969 400.419 m +385.231 400.419 l +385.231 398.905 l +385.6 398.905 l +385.969 398.905 l +385.969 398.905 l +f +385.969 398.905 m +385.231 398.905 l +385.231 397.785 l +385.6 397.785 l +385.969 397.785 l +385.969 397.785 l +f +385.969 397.785 m +385.231 397.785 l +385.231 397.473 l +385.6 397.473 l +385.969 397.473 l +385.969 397.473 l +f +385.969 397.473 m +385.231 397.473 l +385.231 397.045 l +385.6 397.045 l +385.969 397.045 l +385.969 397.045 l +f +385.969 397.045 m +385.231 397.045 l +385.231 396.732 l +385.6 396.732 l +385.969 396.732 l +385.969 396.732 l +f +385.969 396.732 m +385.231 396.732 l +385.231 395.613 l +385.6 395.613 l +385.969 395.613 l +385.969 395.613 l +f +385.969 395.613 m +385.231 395.613 l +385.231 394.098 l +385.6 394.098 l +385.969 394.098 l +385.969 394.098 l +f +385.969 394.098 m +385.231 394.098 l +385.231 390.938 l +385.6 390.938 l +385.969 390.938 l +385.969 390.938 l +f +385.969 390.938 m +385.231 390.938 l +385.231 387.778 l +385.6 387.778 l +385.969 387.778 l +385.969 387.778 l +f +385.969 387.778 m +385.231 387.778 l +385.231 384.618 l +385.572 384.618 l +385.941 384.614 l +385.969 384.618 l +f +385.941 384.614 m +385.204 384.622 l +385.195 383.832 l +385.564 383.828 l +385.932 383.8 l +385.932 383.812 l +f +385.932 383.8 m +385.196 383.856 l +385.136 383.066 l +385.503 383.038 l +385.865 382.964 l +385.869 382.986 l +f +385.865 382.964 m +385.142 383.112 l +384.978 382.322 l +385.34 382.248 l +385.682 382.111 l +385.694 382.141 l +f +385.682 382.111 m +384.997 382.385 l +384.678 381.595 l +385.021 381.458 l +385.33 381.258 l +385.35 381.288 l +f +385.33 381.258 m +384.711 381.657 l +384.202 380.875 l +384.512 380.676 l +384.786 380.43 l +384.806 380.452 l +f +384.786 380.43 m +384.238 380.922 l +383.574 380.189 l +383.848 379.943 l +384.084 379.661 l +384.104 379.678 l +f +384.084 379.661 m +383.613 380.226 l +382.846 379.592 l +383.081 379.31 l +383.27 378.994 l +383.294 379.009 l +f +383.27 378.994 m +382.892 379.625 l +382.073 379.14 l +382.262 378.824 l +382.39 378.48 l +382.422 378.491 l +f +382.39 378.48 m +382.135 379.169 l +381.305 378.864 l +381.432 378.52 l +381.5 378.159 l +381.531 378.164 l +f +381.5 378.159 m +381.365 378.881 l +380.522 378.724 l +380.59 378.363 l +380.614 377.997 l +380.636 377.998 l +f +380.614 377.997 m +380.565 378.73 l +379.697 378.672 l +379.721 378.306 l +379.724 377.939 l +379.746 377.939 l +f +379.724 377.939 m +379.718 378.673 l +378.81 378.665 l +378.814 378.3 l +378.814 377.933 l +378.817 377.93 l +f +378.814 377.933 m +378.814 378.667 l +376.883 378.667 l +376.883 378.3 l +376.883 377.933 l +376.883 377.933 l +f +376.883 377.933 m +376.883 378.667 l +374.952 378.667 l +374.952 378.3 l +374.952 377.933 l +374.952 377.933 l +f +374.952 377.933 m +374.952 378.667 l +373.176 378.667 l +373.176 378.3 l +373.176 377.933 l +373.176 377.933 l +f +373.176 377.933 m +373.176 378.667 l +371.503 378.667 l +371.503 378.3 l +371.503 377.933 l +371.503 377.933 l +f +371.503 377.933 m +371.503 378.667 l +368.193 378.667 l +368.193 378.3 l +368.193 377.933 l +368.193 377.933 l +f +368.193 377.933 m +368.193 378.667 l +364.883 378.667 l +364.883 378.3 l +364.883 377.933 l +364.883 377.933 l +f +364.883 377.933 m +364.883 378.667 l +361.572 378.667 l +361.572 378.3 l +361.572 377.933 l +361.572 377.933 l +f +361.572 377.933 m +361.572 378.667 l +358.262 378.667 l +358.262 378.3 l +358.262 377.933 l +358.262 377.933 l +f +358.262 377.933 m +358.262 378.667 l +354.952 378.667 l +354.952 378.3 l +354.952 377.933 l +354.952 377.933 l +f +354.952 377.933 m +354.952 378.667 l +351.641 378.667 l +351.641 378.297 l +351.627 377.931 l +351.641 377.933 l +f +351.627 377.931 m +351.656 378.664 l +350.001 378.73 l +349.986 378.363 l +349.887 378.01 l +349.928 377.998 l +f +349.887 378.01 m +350.086 378.717 l +348.43 379.178 l +348.331 378.824 l +348.142 378.509 l +348.184 378.484 l +f +348.142 378.509 m +348.52 379.14 l +347.701 379.625 l +347.512 379.31 l +347.277 379.027 l +347.299 379.009 l +f +347.277 379.027 m +347.748 379.592 l +346.98 380.226 l +346.745 379.943 l +346.471 379.697 l +346.489 379.678 l +f +346.471 379.697 m +347.019 380.189 l +346.355 380.922 l +346.081 380.676 l +345.771 380.476 l +345.787 380.452 l +f +345.771 380.476 m +346.391 380.875 l +345.882 381.657 l +345.572 381.458 l +345.23 381.321 l +345.243 381.288 l +f +345.23 381.321 m +345.915 381.595 l +345.596 382.385 l +345.253 382.248 l +344.892 382.173 l +344.899 382.141 l +f +344.892 382.173 m +345.615 382.322 l +345.451 383.112 l +345.09 383.038 l +344.722 383.01 l +344.724 382.986 l +f +344.722 383.01 m +345.457 383.066 l +345.397 383.856 l +345.029 383.828 l +344.66 383.824 l +344.661 383.812 l +f +344.66 383.824 m +345.398 383.832 l +345.39 384.622 l +345 384.618 l +344.631 384.618 l +344.652 384.614 l +f +344.631 384.618 m +345.369 384.618 l +345.369 387.778 l +345 387.778 l +344.631 387.778 l +344.631 387.778 l +f +344.631 387.778 m +345.369 387.778 l +345.369 390.938 l +345 390.938 l +344.631 390.938 l +344.631 390.938 l +f +344.631 390.938 m +345.369 390.938 l +345.369 394.098 l +345 394.098 l +344.631 394.098 l +344.631 394.098 l +f +344.631 394.098 m +345.369 394.098 l +345.369 397.259 l +345 397.259 l +344.631 397.259 l +344.631 397.259 l +f +344.631 397.259 m +345.369 397.259 l +345.369 400.419 l +345 400.419 l +344.631 400.419 l +344.631 400.419 l +f +344.631 400.419 m +345.369 400.419 l +345.369 403.579 l +345.021 403.579 l +344.652 403.583 l +344.631 403.579 l +f +344.652 403.583 m +345.39 403.575 l +345.398 404.365 l +345.029 404.369 l +344.662 404.397 l +344.66 404.373 l +f +344.662 404.397 m +345.397 404.341 l +345.457 405.131 l +345.09 405.159 l +344.728 405.233 l +344.724 405.21 l +f +344.728 405.233 m +345.451 405.085 l +345.615 405.875 l +345.253 405.949 l +344.911 406.086 l +344.899 406.056 l +f +344.911 406.086 m +345.596 405.812 l +345.915 406.602 l +345.572 406.739 l +345.263 406.939 l +345.243 406.909 l +f +345.263 406.939 m +345.882 406.54 l +346.391 407.322 l +346.081 407.521 l +345.807 407.767 l +345.787 407.745 l +f +345.807 407.767 m +346.355 407.275 l +347.019 408.008 l +346.745 408.254 l +346.509 408.536 l +346.489 408.519 l +f +346.509 408.536 m +346.98 407.971 l +347.748 408.605 l +347.512 408.887 l +347.323 409.203 l +347.299 409.188 l +f +347.323 409.203 m +347.701 408.572 l +348.52 409.057 l +348.142 409.688 l +f +BT +/R691 6.82548 Tf +1 0 0 0.995457 353.162 395.957 Tm +(Andrew) Tj +1 0 0 0.995457 348.233 388.03 Tm +(benchmark) Tj +ET +337.243 369.633 109.586 0.734326 re +f +BT +/R691 6.82544 Tf +1 0 0 0.995457 366.062 360.576 Tm +(kernel NFS client) Tj +ET +401.994 409.726 m +402.193 409.019 l +403.848 409.48 l +403.749 409.834 l +403.734 410.201 l +403.691 410.199 l +f +403.734 410.201 m +403.764 409.467 l +405.419 409.533 l +405.404 409.9 l +405.404 410.267 l +405.389 410.266 l +f +405.404 410.267 m +405.404 409.533 l +408.715 409.533 l +408.715 409.9 l +408.715 410.267 l +408.715 410.267 l +f +408.715 410.267 m +408.715 409.533 l +412.025 409.533 l +412.025 409.9 l +412.025 410.267 l +412.025 410.267 l +f +412.025 410.267 m +412.025 409.533 l +415.335 409.533 l +415.335 409.9 l +415.335 410.267 l +415.335 410.267 l +f +415.335 410.267 m +415.335 409.533 l +418.646 409.533 l +418.646 409.9 l +418.646 410.267 l +418.646 410.267 l +f +418.646 410.267 m +418.646 409.533 l +421.956 409.533 l +421.956 409.9 l +421.956 410.267 l +421.956 410.267 l +f +421.956 410.267 m +421.956 409.533 l +425.266 409.533 l +425.266 409.9 l +425.266 410.267 l +425.266 410.267 l +f +425.266 410.267 m +425.266 409.533 l +426.939 409.533 l +426.939 409.9 l +426.939 410.267 l +426.939 410.267 l +f +426.939 410.267 m +426.939 409.533 l +428.715 409.533 l +428.715 409.9 l +428.715 410.267 l +428.715 410.267 l +f +428.715 410.267 m +428.715 409.533 l +430.646 409.533 l +430.646 409.9 l +430.646 410.267 l +430.646 410.267 l +f +430.646 410.267 m +430.646 409.533 l +432.577 409.533 l +432.577 409.9 l +432.58 410.267 l +432.577 410.267 l +f +432.58 410.267 m +432.573 409.532 l +433.481 409.524 l +433.484 409.891 l +433.508 410.258 l +433.498 410.258 l +f +433.508 410.258 m +433.459 409.525 l +434.328 409.467 l +434.352 409.834 l +434.42 410.195 l +434.399 410.199 l +f +434.42 410.195 m +434.285 409.473 l +435.128 409.316 l +435.195 409.677 l +435.323 410.022 l +435.293 410.033 l +f +435.323 410.022 m +435.068 409.333 l +435.897 409.028 l +436.025 409.373 l +436.214 409.688 l +436.184 409.706 l +f +436.214 409.688 m +435.836 409.057 l +436.655 408.572 l +436.844 408.887 l +437.079 409.17 l +437.057 409.188 l +f +437.079 409.17 m +436.608 408.605 l +437.376 407.971 l +437.611 408.254 l +437.885 408.5 l +437.867 408.519 l +f +437.885 408.5 m +437.337 408.008 l +438.001 407.275 l +438.275 407.521 l +438.584 407.721 l +438.568 407.745 l +f +438.584 407.721 m +437.965 407.322 l +438.474 406.54 l +438.783 406.739 l +439.126 406.876 l +439.112 406.909 l +f +439.126 406.876 m +438.441 406.602 l +438.76 405.812 l +439.102 405.949 l +439.464 406.023 l +439.457 406.056 l +f +439.464 406.023 m +438.741 405.875 l +438.905 405.085 l +439.266 405.159 l +439.634 405.187 l +439.632 405.21 l +f +439.634 405.187 m +438.898 405.131 l +438.959 404.341 l +439.327 404.369 l +439.695 404.373 l +439.694 404.397 l +f +439.695 404.373 m +438.958 404.365 l +438.966 403.575 l +439.3 403.579 l +439.669 403.579 l +439.704 403.583 l +f +439.669 403.579 m +438.931 403.579 l +438.931 400.419 l +439.3 400.419 l +439.669 400.419 l +439.669 400.419 l +f +439.669 400.419 m +438.931 400.419 l +438.931 398.905 l +439.3 398.905 l +439.669 398.905 l +439.669 398.905 l +f +439.669 398.905 m +438.931 398.905 l +438.931 397.785 l +439.3 397.785 l +439.669 397.785 l +439.669 397.785 l +f +439.669 397.785 m +438.931 397.785 l +438.931 397.473 l +439.3 397.473 l +439.669 397.473 l +439.669 397.473 l +f +439.669 397.473 m +438.931 397.473 l +438.931 397.045 l +439.3 397.045 l +439.669 397.045 l +439.669 397.045 l +f +439.669 397.045 m +438.931 397.045 l +438.931 396.732 l +439.3 396.732 l +439.669 396.732 l +439.669 396.732 l +f +439.669 396.732 m +438.931 396.732 l +438.931 395.613 l +439.3 395.613 l +439.669 395.613 l +439.669 395.613 l +f +439.669 395.613 m +438.931 395.613 l +438.931 394.098 l +439.3 394.098 l +439.669 394.098 l +439.669 394.098 l +f +439.669 394.098 m +438.931 394.098 l +438.931 390.938 l +439.3 390.938 l +439.669 390.938 l +439.669 390.938 l +f +439.669 390.938 m +438.931 390.938 l +438.931 387.778 l +439.3 387.778 l +439.669 387.778 l +439.669 387.778 l +f +439.669 387.778 m +438.931 387.778 l +438.931 384.618 l +439.335 384.618 l +439.704 384.614 l +439.669 384.618 l +f +439.704 384.614 m +438.966 384.622 l +438.958 383.832 l +439.327 383.828 l +439.694 383.8 l +439.695 383.812 l +f +439.694 383.8 m +438.959 383.856 l +438.898 383.066 l +439.266 383.038 l +439.627 382.964 l +439.632 382.986 l +f +439.627 382.964 m +438.905 383.112 l +438.741 382.322 l +439.102 382.248 l +439.445 382.111 l +439.457 382.141 l +f +439.445 382.111 m +438.76 382.385 l +438.441 381.595 l +438.783 381.458 l +439.093 381.258 l +439.112 381.288 l +f +439.093 381.258 m +438.474 381.657 l +437.965 380.875 l +438.275 380.676 l +438.549 380.43 l +438.569 380.452 l +f +438.549 380.43 m +438.001 380.922 l +437.337 380.189 l +437.611 379.943 l +437.847 379.661 l +437.867 379.678 l +f +437.847 379.661 m +437.376 380.226 l +436.608 379.592 l +436.844 379.31 l +437.033 378.994 l +437.057 379.009 l +f +437.033 378.994 m +436.655 379.625 l +435.836 379.14 l +436.025 378.824 l +436.152 378.48 l +436.184 378.491 l +f +436.152 378.48 m +435.897 379.169 l +435.068 378.864 l +435.195 378.52 l +435.263 378.159 l +435.293 378.164 l +f +435.263 378.159 m +435.128 378.881 l +434.285 378.724 l +434.352 378.363 l +434.377 377.997 l +434.399 377.998 l +f +434.377 377.997 m +434.328 378.73 l +433.459 378.672 l +433.484 378.306 l +433.487 377.939 l +433.498 377.939 l +f +433.487 377.939 m +433.481 378.673 l +432.573 378.665 l +432.577 378.3 l +432.577 377.933 l +432.58 377.93 l +f +432.577 377.933 m +432.577 378.667 l +430.646 378.667 l +430.646 378.3 l +430.646 377.933 l +430.646 377.933 l +f +430.646 377.933 m +430.646 378.667 l +428.715 378.667 l +428.715 378.3 l +428.715 377.933 l +428.715 377.933 l +f +428.715 377.933 m +428.715 378.667 l +426.939 378.667 l +426.939 378.3 l +426.939 377.933 l +426.939 377.933 l +f +426.939 377.933 m +426.939 378.667 l +425.266 378.667 l +425.266 378.3 l +425.266 377.933 l +425.266 377.933 l +f +425.266 377.933 m +425.266 378.667 l +421.956 378.667 l +421.956 378.3 l +421.956 377.933 l +421.956 377.933 l +f +421.956 377.933 m +421.956 378.667 l +418.646 378.667 l +418.646 378.3 l +418.646 377.933 l +418.646 377.933 l +f +418.646 377.933 m +418.646 378.667 l +415.335 378.667 l +415.335 378.3 l +415.335 377.933 l +415.335 377.933 l +f +415.335 377.933 m +415.335 378.667 l +412.025 378.667 l +412.025 378.3 l +412.025 377.933 l +412.025 377.933 l +f +412.025 377.933 m +412.025 378.667 l +408.715 378.667 l +408.715 378.3 l +408.715 377.933 l +408.715 377.933 l +f +408.715 377.933 m +408.715 378.667 l +405.404 378.667 l +405.404 378.297 l +405.389 377.931 l +405.404 377.933 l +f +405.389 377.931 m +405.419 378.664 l +403.764 378.73 l +403.749 378.363 l +403.65 378.01 l +403.691 377.998 l +f +403.65 378.01 m +403.848 378.717 l +402.193 379.178 l +402.094 378.824 l +401.905 378.509 l +401.947 378.484 l +f +401.905 378.509 m +402.283 379.14 l +401.464 379.625 l +401.275 379.31 l +401.039 379.027 l +401.061 379.009 l +f +401.039 379.027 m +401.51 379.592 l +400.743 380.226 l +400.508 379.943 l +400.234 379.697 l +400.252 379.678 l +f +400.234 379.697 m +400.781 380.189 l +400.118 380.922 l +399.844 380.676 l +399.534 380.476 l +399.55 380.452 l +f +399.534 380.476 m +400.153 380.875 l +399.645 381.657 l +399.335 381.458 l +398.993 381.321 l +399.006 381.288 l +f +398.993 381.321 m +399.677 381.595 l +399.358 382.385 l +399.016 382.248 l +398.655 382.173 l +398.662 382.141 l +f +398.655 382.173 m +399.377 382.322 l +399.214 383.112 l +398.852 383.038 l +398.485 383.01 l +398.486 382.986 l +f +398.485 383.01 m +399.22 383.066 l +399.16 383.856 l +398.792 383.828 l +398.423 383.824 l +398.423 383.812 l +f +398.423 383.824 m +399.161 383.832 l +399.152 384.622 l +398.8 384.618 l +398.431 384.618 l +398.415 384.614 l +f +398.431 384.618 m +399.169 384.618 l +399.169 387.778 l +398.8 387.778 l +398.431 387.778 l +398.431 387.778 l +f +398.431 387.778 m +399.169 387.778 l +399.169 390.938 l +398.8 390.938 l +398.431 390.938 l +398.431 390.938 l +f +398.431 390.938 m +399.169 390.938 l +399.169 394.098 l +398.8 394.098 l +398.431 394.098 l +398.431 394.098 l +f +398.431 394.098 m +399.169 394.098 l +399.169 397.259 l +398.8 397.259 l +398.431 397.259 l +398.431 397.259 l +f +398.431 397.259 m +399.169 397.259 l +399.169 400.419 l +398.8 400.419 l +398.431 400.419 l +398.431 400.419 l +f +398.431 400.419 m +399.169 400.419 l +399.169 403.579 l +398.783 403.579 l +398.415 403.583 l +398.431 403.579 l +f +398.415 403.583 m +399.152 403.575 l +399.161 404.365 l +398.792 404.369 l +398.424 404.397 l +398.423 404.373 l +f +398.424 404.397 m +399.16 404.341 l +399.22 405.131 l +398.852 405.159 l +398.491 405.233 l +398.486 405.21 l +f +398.491 405.233 m +399.214 405.085 l +399.377 405.875 l +399.016 405.949 l +398.674 406.086 l +398.662 406.056 l +f +398.674 406.086 m +399.358 405.812 l +399.677 406.602 l +399.335 406.739 l +399.026 406.939 l +399.006 406.909 l +f +399.026 406.939 m +399.645 406.54 l +400.153 407.322 l +399.844 407.521 l +399.57 407.767 l +399.55 407.745 l +f +399.57 407.767 m +400.118 407.275 l +400.781 408.008 l +400.508 408.254 l +400.272 408.536 l +400.252 408.519 l +f +400.272 408.536 m +400.743 407.971 l +401.51 408.605 l +401.275 408.887 l +401.086 409.203 l +401.061 409.188 l +f +401.086 409.203 m +401.464 408.572 l +402.283 409.057 l +401.905 409.688 l +f +BT +/R691 6.82548 Tf +1 0 0 0.995457 402.843 390.528 Tm +(replication) Tj +1 0 0 0.995457 408.531 383.356 Tm +(library) Tj +ET +398.673 398.533 40.5739 0.734326 re +f +BT +/R691 6.82551 Tf +1 0 0 0.995457 410.805 402.593 Tm +(relay) Tj +/R691 6.84901 Tf +1 0 0 0.995457 349.611 423.354 Tm +(client) Tj +ET +0.501961 g +365.048 382.807 m +367.324 379.862 l +362.772 379.862 l +365.048 382.807 l +f* +365.341 383.031 m +364.756 382.583 l +367.032 379.638 l +367.324 379.9 l +367.324 379.533 l +368.043 379.533 l +f +367.324 379.533 m +367.324 380.267 l +362.772 380.267 l +362.772 379.862 l +362.48 380.085 l +362.053 379.533 l +f +362.48 380.085 m +363.065 379.638 l +365.341 382.583 l +365.048 382.807 l +365.341 383.031 l +365.048 383.409 l +f +367.324 368.59 m +365.048 365.532 l +362.772 368.59 l +f* +367.621 368.372 m +367.028 368.809 l +364.752 365.75 l +365.048 365.532 l +364.752 365.313 l +365.048 364.915 l +f +364.752 365.313 m +365.345 365.75 l +363.069 368.809 l +362.772 368.6 l +362.772 368.967 l +362.033 368.967 l +f +362.772 368.967 m +362.772 368.233 l +367.324 368.233 l +367.324 368.59 l +367.621 368.372 l +368.064 368.967 l +f +364.309 368.961 1.4793 10.7807 re +f* +363.931 368.961 m +364.669 368.961 l +364.669 379.742 l +364.309 379.7 l +364.309 380.067 l +363.931 380.067 l +f +364.309 380.067 m +364.309 379.333 l +365.788 379.333 l +365.8 379.742 l +366.169 379.742 l +366.169 380.067 l +f +366.169 379.742 m +365.431 379.742 l +365.431 368.961 l +365.788 369 l +365.788 368.633 l +366.169 368.633 l +f +365.788 368.633 m +365.788 369.367 l +364.309 369.367 l +364.3 368.961 l +363.931 368.961 l +363.931 368.633 l +f +418.998 382.726 m +421.274 379.78 l +416.722 379.78 l +418.998 382.726 l +f* +419.29 382.949 m +418.705 382.502 l +420.981 379.557 l +421.274 379.8 l +421.274 379.433 l +422.007 379.433 l +f +421.274 379.433 m +421.274 380.167 l +416.722 380.167 l +416.722 379.78 l +416.429 380.004 l +415.988 379.433 l +f +416.429 380.004 m +417.014 379.557 l +419.29 382.502 l +418.998 382.726 l +419.29 382.949 l +418.998 383.328 l +f +421.274 368.509 m +418.998 365.451 l +416.722 368.509 l +f* +421.57 368.291 m +420.977 368.728 l +418.701 365.669 l +418.998 365.451 l +418.701 365.232 l +418.998 364.834 l +f +418.701 365.232 m +419.294 365.669 l +417.018 368.728 l +416.722 368.5 l +416.722 368.867 l +415.996 368.867 l +f +416.722 368.867 m +416.722 368.133 l +421.274 368.133 l +421.274 368.509 l +421.57 368.291 l +421.999 368.867 l +f +418.258 368.88 1.4793 10.7807 re +f* +417.931 368.88 m +418.669 368.88 l +418.669 379.661 l +418.258 379.7 l +418.258 380.067 l +417.931 380.067 l +f +418.258 380.067 m +418.258 379.333 l +419.737 379.333 l +419.7 379.661 l +420.069 379.661 l +420.069 380.067 l +f +420.069 379.661 m +419.331 379.661 l +419.331 368.88 l +419.737 368.9 l +419.737 368.533 l +420.069 368.533 l +f +419.737 368.533 m +419.737 369.267 l +418.258 369.267 l +418.3 368.88 l +417.931 368.88 l +417.931 368.533 l +f +0.764706 g +489.669 384.661 m +489.669 385.498 488.988 386.176 488.147 386.176 c +487.307 386.176 486.625 385.498 486.625 384.661 c +486.625 383.825 487.307 383.146 488.147 383.146 c +488.988 383.146 489.669 383.825 489.669 384.661 c +f* +489.302 384.625 m +490.036 384.698 l +490.002 385.04 l +489.991 385.076 l +489.638 384.967 l +489.272 384.93 l +f +489.286 384.858 m +489.991 385.076 l +489.891 385.395 l +489.874 385.426 l +489.55 385.251 l +489.198 385.142 l +f +489.226 385.076 m +489.874 385.426 l +489.717 385.715 l +489.694 385.742 l +489.409 385.508 l +489.085 385.333 l +f +489.125 385.275 m +489.694 385.742 l +489.486 385.993 l +489.458 386.016 l +489.224 385.733 l +488.939 385.499 l +f +488.989 385.449 m +489.458 386.016 l +489.206 386.223 l +489.174 386.24 l +488.998 385.918 l +488.764 385.634 l +f +488.822 385.595 m +489.174 386.24 l +488.884 386.397 l +488.849 386.408 l +488.74 386.057 l +488.564 385.735 l +f +488.63 385.707 m +488.849 386.408 l +488.528 386.507 l +488.491 386.511 l +488.454 386.146 l +488.345 385.795 l +f +488.417 385.78 m +488.491 386.511 l +488.147 386.545 l +488.11 386.542 l +488.147 386.176 l +488.11 385.811 l +f +488.184 385.811 m +488.11 386.542 l +487.767 386.507 l +487.731 386.496 l +487.841 386.146 l +487.878 385.78 l +f +487.95 385.795 m +487.731 386.496 l +487.411 386.397 l +487.379 386.38 l +487.555 386.057 l +487.664 385.707 l +f +487.731 385.735 m +487.379 386.38 l +487.089 386.223 l +487.062 386.201 l +487.296 385.918 l +487.472 385.595 l +f +487.531 385.634 m +487.062 386.201 l +486.809 385.994 l +486.787 385.966 l +487.071 385.733 l +487.306 385.449 l +f +487.356 385.499 m +486.787 385.966 l +486.578 385.715 l +486.561 385.684 l +486.885 385.508 l +487.17 385.275 l +f +487.21 385.333 m +486.561 385.684 l +486.404 385.395 l +486.393 385.36 l +486.745 385.251 l +487.069 385.076 l +f +487.097 385.142 m +486.393 385.36 l +486.293 385.04 l +486.289 385.003 l +486.656 384.967 l +487.009 384.858 l +f +487.023 384.93 m +486.289 385.003 l +486.255 384.661 l +486.258 384.625 l +486.625 384.661 l +486.992 384.625 l +f +486.992 384.698 m +486.258 384.625 l +486.293 384.283 l +486.304 384.247 l +486.656 384.356 l +487.023 384.393 l +f +487.009 384.465 m +486.304 384.247 l +486.404 383.928 l +486.421 383.896 l +486.745 384.072 l +487.097 384.18 l +f +487.069 384.247 m +486.421 383.896 l +486.578 383.608 l +486.601 383.581 l +486.885 383.814 l +487.21 383.989 l +f +487.17 384.048 m +486.601 383.581 l +486.809 383.329 l +486.836 383.307 l +487.071 383.59 l +487.356 383.824 l +f +487.306 383.873 m +486.836 383.307 l +487.089 383.099 l +487.12 383.082 l +487.296 383.405 l +487.531 383.688 l +f +487.472 383.728 m +487.12 383.082 l +487.411 382.925 l +487.446 382.915 l +487.555 383.265 l +487.731 383.588 l +f +487.664 383.616 m +487.446 382.915 l +487.767 382.815 l +487.804 382.812 l +487.841 383.177 l +487.95 383.528 l +f +487.878 383.542 m +487.804 382.812 l +488.147 382.777 l +488.184 382.781 l +488.147 383.146 l +488.184 383.512 l +f +488.11 383.512 m +488.184 382.781 l +488.528 382.815 l +488.563 382.826 l +488.454 383.177 l +488.417 383.542 l +f +488.345 383.528 m +488.563 382.826 l +488.884 382.925 l +488.916 382.943 l +488.74 383.265 l +488.63 383.616 l +f +488.564 383.588 m +488.916 382.943 l +489.206 383.099 l +489.233 383.122 l +488.998 383.405 l +488.822 383.728 l +f +488.764 383.688 m +489.233 383.122 l +489.486 383.329 l +489.508 383.356 l +489.224 383.59 l +488.989 383.873 l +f +488.939 383.824 m +489.508 383.356 l +489.717 383.608 l +489.734 383.639 l +489.409 383.814 l +489.125 384.048 l +f +489.085 383.989 m +489.734 383.639 l +489.891 383.928 l +489.902 383.963 l +489.55 384.072 l +489.226 384.247 l +f +489.198 384.18 m +489.902 383.963 l +490.002 384.282 l +490.005 384.319 l +489.638 384.356 l +489.286 384.465 l +f +489.272 384.393 m +490.005 384.319 l +490.036 384.625 l +489.302 384.698 l +f +489.669 390.262 m +489.669 391.099 488.988 391.777 488.147 391.777 c +487.307 391.777 486.625 391.099 486.625 390.262 c +486.625 389.425 487.307 388.747 488.147 388.747 c +488.988 388.747 489.669 389.425 489.669 390.262 c +f* +489.302 390.225 m +490.036 390.299 l +490.002 390.641 l +489.991 390.676 l +489.638 390.567 l +489.271 390.53 l +f +489.286 390.458 m +489.991 390.676 l +489.891 390.995 l +489.874 391.027 l +489.55 390.851 l +489.197 390.743 l +f +489.226 390.676 m +489.874 391.027 l +489.717 391.315 l +489.694 391.343 l +489.409 391.109 l +489.085 390.934 l +f +489.125 390.875 m +489.694 391.343 l +489.486 391.594 l +489.458 391.616 l +489.224 391.333 l +488.939 391.099 l +f +488.989 391.05 m +489.458 391.616 l +489.206 391.824 l +489.174 391.841 l +488.998 391.518 l +488.764 391.235 l +f +488.822 391.195 m +489.174 391.841 l +488.884 391.998 l +488.849 392.008 l +488.74 391.658 l +488.564 391.335 l +f +488.63 391.307 m +488.849 392.008 l +488.528 392.108 l +488.491 392.111 l +488.454 391.746 l +488.345 391.395 l +f +488.417 391.381 m +488.491 392.111 l +488.147 392.146 l +488.11 392.142 l +488.147 391.777 l +488.11 391.412 l +f +488.184 391.412 m +488.11 392.142 l +487.767 392.108 l +487.731 392.097 l +487.841 391.746 l +487.878 391.381 l +f +487.95 391.395 m +487.731 392.097 l +487.411 391.998 l +487.379 391.98 l +487.555 391.658 l +487.664 391.307 l +f +487.731 391.335 m +487.379 391.98 l +487.089 391.824 l +487.062 391.801 l +487.296 391.518 l +487.472 391.195 l +f +487.531 391.235 m +487.062 391.801 l +486.809 391.594 l +486.787 391.567 l +487.071 391.333 l +487.306 391.05 l +f +487.356 391.099 m +486.787 391.567 l +486.578 391.315 l +486.561 391.284 l +486.885 391.109 l +487.17 390.875 l +f +487.209 390.934 m +486.561 391.284 l +486.404 390.995 l +486.393 390.96 l +486.745 390.851 l +487.069 390.676 l +f +487.097 390.743 m +486.393 390.96 l +486.293 390.641 l +486.289 390.604 l +486.656 390.567 l +487.009 390.458 l +f +487.023 390.53 m +486.289 390.604 l +486.255 390.262 l +486.258 390.225 l +486.625 390.262 l +486.992 390.225 l +f +486.992 390.299 m +486.258 390.225 l +486.293 389.883 l +486.304 389.848 l +486.656 389.956 l +487.023 389.993 l +f +487.009 390.065 m +486.304 389.848 l +486.403 389.529 l +486.421 389.497 l +486.745 389.672 l +487.097 389.781 l +f +487.069 389.847 m +486.421 389.497 l +486.578 389.208 l +486.601 389.181 l +486.885 389.415 l +487.209 389.59 l +f +487.17 389.648 m +486.601 389.181 l +486.809 388.93 l +486.836 388.907 l +487.071 389.19 l +487.356 389.424 l +f +487.306 389.474 m +486.836 388.907 l +487.089 388.7 l +487.12 388.683 l +487.296 389.005 l +487.531 389.289 l +f +487.472 389.328 m +487.12 388.683 l +487.411 388.526 l +487.446 388.515 l +487.555 388.866 l +487.731 389.188 l +f +487.664 389.216 m +487.446 388.515 l +487.767 388.416 l +487.804 388.412 l +487.841 388.777 l +487.95 389.128 l +f +487.878 389.143 m +487.804 388.412 l +488.147 388.378 l +488.184 388.381 l +488.147 388.747 l +488.184 389.112 l +f +488.11 389.112 m +488.184 388.381 l +488.528 388.416 l +488.563 388.427 l +488.454 388.777 l +488.417 389.143 l +f +488.345 389.128 m +488.563 388.427 l +488.884 388.526 l +488.916 388.543 l +488.74 388.866 l +488.63 389.216 l +f +488.564 389.188 m +488.916 388.543 l +489.206 388.7 l +489.233 388.722 l +488.998 389.005 l +488.822 389.328 l +f +488.764 389.289 m +489.233 388.722 l +489.486 388.93 l +489.508 388.957 l +489.224 389.19 l +488.989 389.474 l +f +488.939 389.424 m +489.508 388.957 l +489.717 389.208 l +489.734 389.24 l +489.409 389.415 l +489.125 389.648 l +f +489.085 389.59 m +489.734 389.24 l +489.891 389.529 l +489.902 389.563 l +489.55 389.672 l +489.226 389.847 l +f +489.197 389.781 m +489.902 389.563 l +490.002 389.883 l +490.005 389.92 l +489.638 389.956 l +489.286 390.065 l +f +489.271 389.993 m +490.005 389.92 l +490.036 390.225 l +489.302 390.299 l +f +489.669 396.241 m +489.669 397.078 488.988 397.756 488.147 397.756 c +487.307 397.756 486.625 397.078 486.625 396.241 c +486.625 395.404 487.307 394.726 488.147 394.726 c +488.988 394.726 489.669 395.404 489.669 396.241 c +f* +489.302 396.204 m +490.036 396.278 l +490.002 396.62 l +489.991 396.655 l +489.638 396.546 l +489.271 396.509 l +f +489.286 396.437 m +489.991 396.655 l +489.891 396.974 l +489.874 397.006 l +489.55 396.831 l +489.197 396.722 l +f +489.226 396.656 m +489.874 397.006 l +489.717 397.294 l +489.694 397.322 l +489.409 397.088 l +489.085 396.913 l +f +489.125 396.854 m +489.694 397.322 l +489.486 397.573 l +489.458 397.595 l +489.224 397.312 l +488.939 397.079 l +f +488.989 397.029 m +489.458 397.595 l +489.206 397.803 l +489.174 397.82 l +488.998 397.497 l +488.764 397.214 l +f +488.822 397.175 m +489.174 397.82 l +488.884 397.977 l +488.849 397.988 l +488.74 397.637 l +488.564 397.314 l +f +488.63 397.286 m +488.849 397.988 l +488.528 398.087 l +488.491 398.091 l +488.454 397.725 l +488.345 397.375 l +f +488.417 397.36 m +488.491 398.091 l +488.147 398.125 l +488.11 398.121 l +488.147 397.756 l +488.11 397.391 l +f +488.184 397.391 m +488.11 398.121 l +487.767 398.087 l +487.731 398.076 l +487.841 397.725 l +487.878 397.36 l +f +487.95 397.375 m +487.731 398.076 l +487.411 397.977 l +487.379 397.96 l +487.555 397.637 l +487.664 397.286 l +f +487.731 397.314 m +487.379 397.96 l +487.089 397.803 l +487.062 397.78 l +487.296 397.497 l +487.472 397.175 l +f +487.531 397.214 m +487.062 397.78 l +486.809 397.573 l +486.787 397.546 l +487.071 397.312 l +487.306 397.029 l +f +487.356 397.079 m +486.787 397.546 l +486.578 397.294 l +486.561 397.263 l +486.885 397.088 l +487.17 396.854 l +f +487.209 396.913 m +486.561 397.263 l +486.404 396.974 l +486.393 396.94 l +486.745 396.831 l +487.069 396.656 l +f +487.097 396.722 m +486.393 396.94 l +486.293 396.62 l +486.289 396.583 l +486.656 396.546 l +487.009 396.437 l +f +487.023 396.509 m +486.289 396.583 l +486.255 396.241 l +486.258 396.204 l +486.625 396.241 l +486.992 396.204 l +f +486.992 396.278 m +486.258 396.204 l +486.293 395.862 l +486.304 395.827 l +486.656 395.936 l +487.023 395.972 l +f +487.009 396.044 m +486.304 395.827 l +486.403 395.508 l +486.421 395.476 l +486.745 395.651 l +487.097 395.76 l +f +487.069 395.826 m +486.421 395.476 l +486.578 395.188 l +486.601 395.16 l +486.885 395.394 l +487.209 395.569 l +f +487.17 395.628 m +486.601 395.16 l +486.809 394.909 l +486.836 394.886 l +487.071 395.17 l +487.356 395.403 l +f +487.306 395.453 m +486.836 394.886 l +487.089 394.679 l +487.12 394.662 l +487.296 394.985 l +487.531 395.268 l +f +487.472 395.307 m +487.12 394.662 l +487.411 394.505 l +487.446 394.494 l +487.555 394.845 l +487.731 395.168 l +f +487.664 395.196 m +487.446 394.494 l +487.767 394.395 l +487.804 394.391 l +487.841 394.757 l +487.95 395.107 l +f +487.878 395.122 m +487.804 394.391 l +488.147 394.357 l +488.184 394.361 l +488.147 394.726 l +488.184 395.091 l +f +488.11 395.091 m +488.184 394.361 l +488.528 394.395 l +488.563 394.406 l +488.454 394.757 l +488.417 395.122 l +f +488.345 395.107 m +488.563 394.406 l +488.884 394.505 l +488.916 394.522 l +488.74 394.845 l +488.63 395.196 l +f +488.564 395.168 m +488.916 394.522 l +489.206 394.679 l +489.233 394.701 l +488.998 394.985 l +488.822 395.307 l +f +488.764 395.268 m +489.233 394.701 l +489.486 394.909 l +489.508 394.936 l +489.224 395.17 l +488.989 395.453 l +f +488.939 395.403 m +489.508 394.936 l +489.717 395.188 l +489.734 395.219 l +489.409 395.394 l +489.125 395.628 l +f +489.085 395.569 m +489.734 395.219 l +489.891 395.508 l +489.902 395.542 l +489.55 395.651 l +489.226 395.826 l +f +489.197 395.76 m +489.902 395.542 l +490.002 395.862 l +490.005 395.899 l +489.638 395.936 l +489.286 396.044 l +f +489.271 395.972 m +490.005 395.899 l +490.036 396.204 l +489.302 396.278 l +f +BT +0 g +1 0 0 0.995457 463.599 462.633 Tm +(replica 0) Tj +ET +466.484 453.617 m +466.683 452.91 l +468.338 453.37 l +468.239 453.724 l +468.224 454.09 l +468.181 454.089 l +f +468.224 454.09 m +468.253 453.357 l +469.909 453.422 l +469.894 453.8 l +469.894 454.167 l +469.879 454.156 l +f +469.894 454.167 m +469.894 453.433 l +473.204 453.433 l +473.204 453.8 l +473.204 454.167 l +473.204 454.167 l +f +473.204 454.167 m +473.204 453.433 l +476.515 453.433 l +476.515 453.8 l +476.515 454.167 l +476.515 454.167 l +f +476.515 454.167 m +476.515 453.433 l +479.825 453.433 l +479.825 453.8 l +479.825 454.167 l +479.825 454.167 l +f +479.825 454.167 m +479.825 453.433 l +483.135 453.433 l +483.135 453.8 l +483.135 454.167 l +483.135 454.167 l +f +483.135 454.167 m +483.135 453.433 l +486.446 453.433 l +486.446 453.8 l +486.446 454.167 l +486.446 454.167 l +f +486.446 454.167 m +486.446 453.433 l +489.756 453.433 l +489.756 453.8 l +489.756 454.167 l +489.756 454.167 l +f +489.756 454.167 m +489.756 453.433 l +491.428 453.433 l +491.428 453.8 l +491.428 454.167 l +491.428 454.167 l +f +491.428 454.167 m +491.428 453.433 l +493.204 453.433 l +493.204 453.8 l +493.204 454.167 l +493.204 454.167 l +f +493.204 454.167 m +493.204 453.433 l +495.135 453.433 l +495.135 453.8 l +495.135 454.167 l +495.135 454.167 l +f +495.135 454.167 m +495.135 453.433 l +497.066 453.433 l +497.066 453.789 l +497.07 454.156 l +497.066 454.167 l +f +497.07 454.156 m +497.063 453.422 l +497.97 453.414 l +497.974 453.781 l +497.998 454.147 l +497.977 454.148 l +f +497.998 454.147 m +497.949 453.415 l +498.818 453.357 l +498.842 453.724 l +498.91 454.085 l +498.888 454.089 l +f +498.91 454.085 m +498.775 453.363 l +499.617 453.206 l +499.685 453.567 l +499.812 453.912 l +499.783 453.923 l +f +499.812 453.912 m +499.558 453.223 l +500.387 452.919 l +500.515 453.263 l +500.703 453.579 l +500.674 453.596 l +f +500.703 453.579 m +500.326 452.948 l +501.145 452.463 l +501.334 452.779 l +501.569 453.061 l +501.547 453.08 l +f +501.569 453.061 m +501.098 452.496 l +501.866 451.863 l +502.101 452.146 l +502.375 452.392 l +502.357 452.412 l +f +502.375 452.392 m +501.827 451.9 l +502.491 451.168 l +502.765 451.415 l +503.074 451.614 l +503.058 451.639 l +f +503.074 451.614 m +502.455 451.215 l +502.964 450.434 l +503.273 450.634 l +503.615 450.771 l +503.602 450.804 l +f +503.615 450.771 m +502.931 450.497 l +503.25 449.708 l +503.592 449.845 l +503.953 449.919 l +503.947 449.952 l +f +503.953 449.919 m +503.231 449.771 l +503.395 448.982 l +503.756 449.056 l +504.124 449.084 l +504.122 449.108 l +f +504.124 449.084 m +503.388 449.028 l +503.449 448.24 l +503.816 448.268 l +504.185 448.272 l +504.184 448.295 l +f +504.185 448.272 m +503.448 448.264 l +503.456 447.475 l +503.8 447.479 l +504.169 447.479 l +504.194 447.483 l +f +504.169 447.479 m +503.431 447.479 l +503.431 444.323 l +503.8 444.323 l +504.169 444.323 l +504.169 444.323 l +f +504.169 444.323 m +503.431 444.323 l +503.431 442.811 l +503.8 442.811 l +504.169 442.811 l +504.169 442.811 l +f +504.169 442.811 m +503.431 442.811 l +503.431 441.694 l +503.8 441.694 l +504.169 441.694 l +504.169 441.694 l +f +504.169 441.694 m +503.431 441.694 l +503.431 441.382 l +503.8 441.382 l +504.169 441.382 l +504.169 441.382 l +f +504.169 441.382 m +503.431 441.382 l +503.431 440.954 l +503.8 440.954 l +504.169 440.954 l +504.169 440.954 l +f +504.169 440.954 m +503.431 440.954 l +503.431 440.642 l +503.8 440.642 l +504.169 440.642 l +504.169 440.642 l +f +504.169 440.642 m +503.431 440.642 l +503.431 439.525 l +503.8 439.525 l +504.169 439.525 l +504.169 439.525 l +f +504.169 439.525 m +503.431 439.525 l +503.431 438.013 l +503.8 438.013 l +504.169 438.013 l +504.169 438.013 l +f +504.169 438.013 m +503.431 438.013 l +503.431 434.857 l +503.8 434.857 l +504.169 434.857 l +504.169 434.857 l +f +504.169 434.857 m +503.431 434.857 l +503.431 431.702 l +503.8 431.702 l +504.169 431.702 l +504.169 431.702 l +f +504.169 431.702 m +503.431 431.702 l +503.431 428.547 l +503.825 428.547 l +504.194 428.543 l +504.169 428.547 l +f +504.194 428.543 m +503.456 428.551 l +503.448 427.762 l +503.816 427.758 l +504.184 427.73 l +504.185 427.742 l +f +504.184 427.73 m +503.449 427.786 l +503.388 426.997 l +503.756 426.969 l +504.117 426.895 l +504.122 426.918 l +f +504.117 426.895 m +503.395 427.043 l +503.231 426.255 l +503.592 426.18 l +503.934 426.043 l +503.947 426.074 l +f +503.934 426.043 m +503.25 426.317 l +502.931 425.528 l +503.273 425.391 l +503.583 425.192 l +503.602 425.221 l +f +503.583 425.192 m +502.964 425.591 l +502.455 424.811 l +502.765 424.611 l +503.038 424.365 l +503.058 424.386 l +f +503.038 424.365 m +502.491 424.857 l +501.827 424.126 l +502.101 423.879 l +502.336 423.597 l +502.357 423.614 l +f +502.336 423.597 m +501.866 424.162 l +501.098 423.529 l +501.334 423.247 l +501.522 422.931 l +501.547 422.946 l +f +501.522 422.931 m +501.145 423.562 l +500.326 423.078 l +500.515 422.762 l +500.642 422.417 l +500.674 422.429 l +f +500.642 422.417 m +500.387 423.106 l +499.558 422.802 l +499.685 422.458 l +499.752 422.097 l +499.783 422.103 l +f +499.752 422.097 m +499.617 422.819 l +498.775 422.663 l +498.842 422.302 l +498.867 421.935 l +498.888 421.937 l +f +498.867 421.935 m +498.818 422.668 l +497.949 422.611 l +497.974 422.244 l +497.977 421.877 l +497.998 421.878 l +f +497.977 421.877 m +497.97 422.611 l +497.063 422.603 l +497.066 422.2 l +497.066 421.833 l +497.07 421.869 l +f +497.066 421.833 m +497.066 422.567 l +495.135 422.567 l +495.135 422.2 l +495.135 421.833 l +495.135 421.833 l +f +495.135 421.833 m +495.135 422.567 l +493.204 422.567 l +493.204 422.2 l +493.204 421.833 l +493.204 421.833 l +f +493.204 421.833 m +493.204 422.567 l +491.428 422.567 l +491.428 422.2 l +491.428 421.833 l +491.428 421.833 l +f +491.428 421.833 m +491.428 422.567 l +489.756 422.567 l +489.756 422.2 l +489.756 421.833 l +489.756 421.833 l +f +489.756 421.833 m +489.756 422.567 l +486.446 422.567 l +486.446 422.2 l +486.446 421.833 l +486.446 421.833 l +f +486.446 421.833 m +486.446 422.567 l +483.135 422.567 l +483.135 422.2 l +483.135 421.833 l +483.135 421.833 l +f +483.135 421.833 m +483.135 422.567 l +479.825 422.567 l +479.825 422.2 l +479.825 421.833 l +479.825 421.833 l +f +479.825 421.833 m +479.825 422.567 l +476.515 422.567 l +476.515 422.2 l +476.515 421.833 l +476.515 421.833 l +f +476.515 421.833 m +476.515 422.567 l +473.204 422.567 l +473.204 422.2 l +473.204 421.833 l +473.204 421.833 l +f +473.204 421.833 m +473.204 422.567 l +469.894 422.567 l +469.894 422.236 l +469.879 421.869 l +469.894 421.833 l +f +469.879 421.869 m +469.909 422.603 l +468.253 422.669 l +468.239 422.302 l +468.14 421.948 l +468.181 421.937 l +f +468.14 421.948 m +468.338 422.655 l +466.683 423.116 l +466.584 422.762 l +466.395 422.446 l +466.437 422.421 l +f +466.395 422.446 m +466.772 423.078 l +465.953 423.562 l +465.765 423.247 l +465.529 422.964 l +465.551 422.946 l +f +465.529 422.964 m +466 423.529 l +465.233 424.162 l +464.997 423.879 l +464.724 423.633 l +464.742 423.614 l +f +464.724 423.633 m +465.271 424.126 l +464.607 424.857 l +464.334 424.611 l +464.024 424.411 l +464.04 424.386 l +f +464.024 424.411 m +464.643 424.811 l +464.134 425.591 l +463.825 425.391 l +463.483 425.254 l +463.496 425.221 l +f +463.483 425.254 m +464.167 425.528 l +463.848 426.317 l +463.506 426.18 l +463.145 426.106 l +463.152 426.074 l +f +463.145 426.106 m +463.867 426.255 l +463.703 427.043 l +463.342 426.969 l +462.974 426.941 l +462.976 426.918 l +f +462.974 426.941 m +463.71 426.997 l +463.65 427.786 l +463.282 427.758 l +462.913 427.754 l +462.913 427.742 l +f +462.913 427.754 m +463.651 427.762 l +463.642 428.551 l +463.3 428.547 l +462.931 428.547 l +462.904 428.543 l +f +462.931 428.547 m +463.669 428.547 l +463.669 431.702 l +463.3 431.702 l +462.931 431.702 l +462.931 431.702 l +f +462.931 431.702 m +463.669 431.702 l +463.669 434.857 l +463.3 434.857 l +462.931 434.857 l +462.931 434.857 l +f +462.931 434.857 m +463.669 434.857 l +463.669 438.013 l +463.3 438.013 l +462.931 438.013 l +462.931 438.013 l +f +462.931 438.013 m +463.669 438.013 l +463.669 441.168 l +463.3 441.168 l +462.931 441.168 l +462.931 441.168 l +f +462.931 441.168 m +463.669 441.168 l +463.669 444.323 l +463.3 444.323 l +462.931 444.323 l +462.931 444.323 l +f +462.931 444.323 m +463.669 444.323 l +463.669 447.479 l +463.273 447.479 l +462.904 447.483 l +462.931 447.479 l +f +462.904 447.483 m +463.642 447.475 l +463.651 448.264 l +463.282 448.268 l +462.914 448.295 l +462.913 448.271 l +f +462.914 448.295 m +463.65 448.24 l +463.71 449.028 l +463.342 449.056 l +462.981 449.131 l +462.976 449.108 l +f +462.981 449.131 m +463.703 448.982 l +463.867 449.771 l +463.506 449.845 l +463.164 449.982 l +463.152 449.952 l +f +463.164 449.982 m +463.848 449.708 l +464.167 450.497 l +463.825 450.634 l +463.516 450.834 l +463.496 450.804 l +f +463.516 450.834 m +464.134 450.434 l +464.643 451.215 l +464.334 451.415 l +464.06 451.661 l +464.04 451.639 l +f +464.06 451.661 m +464.607 451.168 l +465.271 451.9 l +464.997 452.146 l +464.762 452.429 l +464.742 452.412 l +f +464.762 452.429 m +465.233 451.863 l +466 452.496 l +465.765 452.779 l +465.576 453.094 l +465.551 453.08 l +f +465.576 453.094 m +465.953 452.463 l +466.772 452.948 l +466.395 453.579 l +f +BT +/R691 6.82548 Tf +1 0 0 0.995457 467.333 434.089 Tm +(replication) Tj +1 0 0 0.995457 473.021 426.917 Tm +(library) Tj +ET +463.161 442.433 40.5735 0.734326 re +f +BT +/R691 6.82544 Tf +1 0 0 0.995457 475.295 447.29 Tm +(snfsd) Tj +ET +462.708 459.455 m +463.307 458.113 l +464.535 458.656 l +464.236 459.327 l +464.072 460.043 l +464.002 460.027 l +f +464.072 460.043 m +464.4 458.611 l +465.629 458.89 l +465.465 459.606 l +465.403 460.338 l +465.351 460.333 l +f +465.403 460.338 m +465.526 458.874 l +466.755 458.977 l +466.693 459.709 l +466.684 460.443 l +466.658 460.443 l +f +466.684 460.443 m +466.702 458.974 l +467.931 458.989 l +467.922 459.7 l +467.922 460.434 l +467.913 460.458 l +f +467.922 460.434 m +467.922 458.966 l +472.836 458.966 l +472.836 459.7 l +472.836 460.434 l +472.836 460.434 l +f +472.836 460.434 m +472.836 458.966 l +477.751 458.966 l +477.751 459.7 l +477.751 460.434 l +477.751 460.434 l +f +477.751 460.434 m +477.751 458.966 l +482.665 458.966 l +482.665 459.7 l +482.665 460.434 l +482.665 460.434 l +f +482.665 460.434 m +482.665 458.966 l +487.58 458.966 l +487.58 459.7 l +487.58 460.434 l +487.58 460.434 l +f +487.58 460.434 m +487.58 458.966 l +492.494 458.966 l +492.494 459.7 l +492.494 460.434 l +492.494 460.434 l +f +492.494 460.434 m +492.494 458.966 l +497.409 458.966 l +497.409 459.7 l +497.409 460.434 l +497.409 460.434 l +f +497.409 460.434 m +497.409 458.966 l +499.891 458.966 l +499.891 459.7 l +499.891 460.434 l +499.891 460.434 l +f +499.891 460.434 m +499.891 458.966 l +502.528 458.966 l +502.528 459.7 l +502.528 460.434 l +502.528 460.434 l +f +502.528 460.434 m +502.528 458.966 l +505.395 458.966 l +505.395 459.7 l +505.395 460.434 l +505.395 460.434 l +f +505.395 460.434 m +505.395 458.966 l +508.261 458.966 l +508.261 459.723 l +508.269 460.458 l +508.261 460.434 l +f +508.269 460.458 m +508.253 458.989 l +509.6 458.974 l +509.608 459.709 l +509.667 460.441 l +509.642 460.443 l +f +509.667 460.441 m +509.549 458.977 l +510.839 458.874 l +510.898 459.606 l +511.059 460.323 l +511.008 460.334 l +f +511.059 460.323 m +510.736 458.889 l +511.988 458.61 l +512.149 459.327 l +512.447 459.998 l +512.381 460.028 l +f +512.447 459.998 m +511.85 458.656 l +513.082 458.112 l +513.381 458.784 l +513.81 459.381 l +513.748 459.425 l +f +513.81 459.381 m +512.951 458.187 l +514.167 457.321 l +514.596 457.918 l +515.117 458.438 l +515.075 458.48 l +f +515.117 458.438 m +514.076 457.398 l +515.215 456.267 l +515.735 456.787 l +516.325 457.228 l +516.294 457.27 l +f +516.325 457.228 m +515.145 456.347 l +516.131 455.04 l +516.721 455.481 l +517.37 455.829 l +517.344 455.878 l +f +517.37 455.829 m +516.071 455.133 l +516.827 453.738 l +517.476 454.086 l +518.176 454.319 l +518.155 454.379 l +f +518.176 454.319 m +516.776 453.854 l +517.25 452.444 l +517.949 452.677 l +518.676 452.802 l +518.667 452.857 l +f +518.676 452.802 m +517.222 452.553 l +517.466 451.144 l +518.193 451.268 l +518.929 451.314 l +518.926 451.354 l +f +518.929 451.314 m +517.456 451.222 l +517.546 449.812 l +518.282 449.859 l +519.02 449.865 l +519.02 449.885 l +f +519.02 449.865 m +517.545 449.852 l +517.557 448.443 l +518.3 448.45 l +519.038 448.45 l +519.033 448.456 l +f +519.038 448.45 m +517.562 448.45 l +517.562 442.813 l +518.3 442.813 l +519.038 442.813 l +519.038 442.813 l +f +519.038 442.813 m +517.562 442.813 l +517.562 441.418 l +518.3 441.418 l +519.038 441.418 l +519.038 441.418 l +f +519.038 441.418 m +517.562 441.418 l +517.562 438.981 l +518.3 438.981 l +519.038 438.981 l +519.038 438.981 l +f +519.038 438.981 m +517.562 438.981 l +517.562 438.115 l +518.3 438.115 l +519.038 438.115 l +519.038 438.115 l +f +519.038 438.115 m +517.562 438.115 l +517.562 437.557 l +518.3 437.557 l +519.038 437.557 l +519.038 437.557 l +f +519.038 437.557 m +517.562 437.557 l +517.562 436.794 l +518.3 436.794 l +519.038 436.794 l +519.038 436.794 l +f +519.038 436.794 m +517.562 436.794 l +517.562 436.236 l +518.3 436.236 l +519.038 436.236 l +519.038 436.236 l +f +519.038 436.236 m +517.562 436.236 l +517.562 435.37 l +518.3 435.37 l +519.038 435.37 l +519.038 435.37 l +f +519.038 435.37 m +517.562 435.37 l +517.562 432.933 l +518.3 432.933 l +519.038 432.933 l +519.038 432.933 l +f +519.038 432.933 m +517.562 432.933 l +517.562 431.539 l +518.3 431.539 l +519.038 431.539 l +519.038 431.539 l +f +519.038 431.539 m +517.562 431.539 l +517.562 425.902 l +518.3 425.902 l +519.038 425.902 l +519.038 425.902 l +f +519.038 425.902 m +517.562 425.902 l +517.562 420.265 l +518.3 420.265 l +519.038 420.265 l +519.038 420.265 l +f +519.038 420.265 m +517.562 420.265 l +517.562 414.628 l +518.295 414.628 l +519.033 414.622 l +519.038 414.628 l +f +519.033 414.622 m +517.557 414.635 l +517.545 413.226 l +518.282 413.219 l +519.018 413.173 l +519.02 413.193 l +f +519.018 413.173 m +517.546 413.265 l +517.456 411.856 l +518.193 411.81 l +518.92 411.685 l +518.926 411.724 l +f +518.92 411.685 m +517.466 411.934 l +517.222 410.525 l +517.949 410.401 l +518.649 410.168 l +518.667 410.221 l +f +518.649 410.168 m +517.25 410.634 l +516.776 409.224 l +517.476 408.991 l +518.125 408.643 l +518.155 408.699 l +f +518.125 408.643 m +516.827 409.34 l +516.071 407.945 l +516.721 407.597 l +517.311 407.156 l +517.344 407.2 l +f +517.311 407.156 m +516.131 408.038 l +515.145 406.731 l +515.735 406.29 l +516.256 405.77 l +516.294 405.807 l +f +516.256 405.77 m +515.215 406.81 l +514.076 405.68 l +514.596 405.16 l +515.026 404.563 l +515.075 404.598 l +f +515.026 404.563 m +514.167 405.757 l +512.951 404.891 l +513.381 404.294 l +513.679 403.623 l +513.748 403.653 l +f +513.679 403.623 m +513.082 404.965 l +511.85 404.422 l +512.149 403.751 l +512.31 403.034 l +512.381 403.05 l +f +512.31 403.034 m +511.988 404.467 l +510.736 404.189 l +510.898 403.472 l +510.957 402.74 l +511.008 402.744 l +f +510.957 402.74 m +510.839 404.204 l +509.549 404.101 l +509.608 403.369 l +509.616 402.635 l +509.642 402.635 l +f +509.616 402.635 m +509.6 404.103 l +508.253 404.089 l +508.261 403.4 l +508.261 402.666 l +508.269 402.62 l +f +508.261 402.666 m +508.261 404.134 l +505.395 404.134 l +505.395 403.4 l +505.395 402.666 l +505.395 402.666 l +f +505.395 402.666 m +505.395 404.134 l +502.528 404.134 l +502.528 403.4 l +502.528 402.666 l +502.528 402.666 l +f +502.528 402.666 m +502.528 404.134 l +499.891 404.134 l +499.891 403.4 l +499.891 402.666 l +499.891 402.666 l +f +499.891 402.666 m +499.891 404.134 l +497.409 404.134 l +497.409 403.4 l +497.409 402.666 l +497.409 402.666 l +f +497.409 402.666 m +497.409 404.134 l +492.494 404.134 l +492.494 403.4 l +492.494 402.666 l +492.494 402.666 l +f +492.494 402.666 m +492.494 404.134 l +487.58 404.134 l +487.58 403.4 l +487.58 402.666 l +487.58 402.666 l +f +487.58 402.666 m +487.58 404.134 l +482.665 404.134 l +482.665 403.4 l +482.665 402.666 l +482.665 402.666 l +f +482.665 402.666 m +482.665 404.134 l +477.751 404.134 l +477.751 403.4 l +477.751 402.666 l +477.751 402.666 l +f +477.751 402.666 m +477.751 404.134 l +472.836 404.134 l +472.836 403.4 l +472.836 402.666 l +472.836 402.666 l +f +472.836 402.666 m +472.836 404.134 l +467.922 404.134 l +467.922 403.354 l +467.913 402.62 l +467.922 402.666 l +f +467.913 402.62 m +467.931 404.089 l +466.702 404.103 l +466.693 403.369 l +466.631 402.637 l +466.658 402.635 l +f +466.631 402.637 m +466.755 404.101 l +465.526 404.204 l +465.465 403.472 l +465.301 402.756 l +465.351 402.744 l +f +465.301 402.756 m +465.629 404.188 l +464.4 404.467 l +464.236 403.751 l +463.937 403.08 l +464.002 403.051 l +f +463.937 403.08 m +464.535 404.422 l +463.307 404.965 l +463.007 404.294 l +462.578 403.697 l +462.639 403.653 l +f +462.578 403.697 m +463.437 404.891 l +462.221 405.757 l +461.792 405.16 l +461.271 404.64 l +461.313 404.598 l +f +461.271 404.64 m +462.312 405.68 l +461.173 406.81 l +460.653 406.29 l +460.063 405.849 l +460.094 405.807 l +f +460.063 405.849 m +461.242 406.731 l +460.257 408.038 l +459.667 407.597 l +459.018 407.248 l +459.044 407.2 l +f +459.018 407.248 m +460.316 407.945 l +459.561 409.34 l +458.912 408.991 l +458.212 408.758 l +458.233 408.699 l +f +458.212 408.758 m +459.612 409.224 l +459.138 410.634 l +458.438 410.401 l +457.711 410.276 l +457.721 410.221 l +f +457.711 410.276 m +459.166 410.525 l +458.922 411.934 l +458.195 411.81 l +457.459 411.763 l +457.462 411.724 l +f +457.459 411.763 m +458.932 411.856 l +458.842 413.265 l +458.106 413.219 l +457.368 413.212 l +457.368 413.192 l +f +457.368 413.212 m +458.843 413.226 l +458.831 414.635 l +458.1 414.628 l +457.362 414.628 l +457.355 414.622 l +f +457.362 414.628 m +458.838 414.628 l +458.838 420.265 l +458.1 420.265 l +457.362 420.265 l +457.362 420.265 l +f +457.362 420.265 m +458.838 420.265 l +458.838 425.902 l +458.1 425.902 l +457.362 425.902 l +457.362 425.902 l +f +457.362 425.902 m +458.838 425.902 l +458.838 431.539 l +458.1 431.539 l +457.362 431.539 l +457.362 431.539 l +f +457.362 431.539 m +458.838 431.539 l +458.838 437.176 l +458.1 437.176 l +457.362 437.176 l +457.362 437.176 l +f +457.362 437.176 m +458.838 437.176 l +458.838 442.813 l +458.1 442.813 l +457.362 442.813 l +457.362 442.813 l +f +457.362 442.813 m +458.838 442.813 l +458.838 448.45 l +458.093 448.45 l +457.355 448.456 l +457.362 448.45 l +f +457.355 448.456 m +458.831 448.443 l +458.843 449.852 l +458.106 449.859 l +457.37 449.905 l +457.368 449.885 l +f +457.37 449.905 m +458.842 449.812 l +458.932 451.222 l +458.195 451.268 l +457.468 451.392 l +457.462 451.354 l +f +457.468 451.392 m +458.922 451.144 l +459.166 452.553 l +458.438 452.677 l +457.739 452.91 l +457.721 452.857 l +f +457.739 452.91 m +459.138 452.444 l +459.612 453.854 l +458.912 454.086 l +458.263 454.435 l +458.233 454.379 l +f +458.263 454.435 m +459.561 453.738 l +460.316 455.133 l +459.667 455.481 l +459.077 455.922 l +459.044 455.878 l +f +459.077 455.922 m +460.257 455.04 l +461.242 456.347 l +460.653 456.787 l +460.132 457.308 l +460.094 457.27 l +f +460.132 457.308 m +461.173 456.267 l +462.312 457.398 l +461.792 457.918 l +461.362 458.515 l +461.313 458.48 l +f +461.362 458.515 m +462.221 457.321 l +463.437 458.187 l +462.578 459.381 l +f +458 415.733 59.9384 0.734326 re +f +BT +/R691 6.84901 Tf +1 0 0 0.995457 472.398 408.469 Tm +(kernel VM) Tj +ET +0.501961 g +513.356 414.391 m +511.08 411.332 l +508.805 414.391 l +f* +513.653 414.172 m +513.06 414.609 l +510.784 411.551 l +511.08 411.332 l +510.784 411.114 l +511.08 410.715 l +f +510.784 411.114 m +511.377 411.551 l +509.101 414.609 l +508.805 414.4 l +508.805 414.767 l +508.065 414.767 l +f +508.805 414.767 m +508.805 414.033 l +513.356 414.033 l +513.356 414.391 l +513.653 414.172 l +514.096 414.767 l +f +510.341 414.698 1.4793 33.7112 re +f* +509.931 414.698 m +510.669 414.698 l +510.669 448.409 l +510.341 448.4 l +510.341 448.767 l +509.931 448.767 l +f +510.341 448.767 m +510.341 448.033 l +511.82 448.033 l +511.8 448.409 l +512.169 448.409 l +512.169 448.767 l +f +512.169 448.409 m +511.431 448.409 l +511.431 414.698 l +511.82 414.7 l +511.82 414.333 l +512.169 414.333 l +f +511.82 414.333 m +511.82 415.067 l +510.341 415.067 l +510.3 414.698 l +509.931 414.698 l +509.931 414.333 l +f +500.547 445.715 m +497.475 447.981 l +500.547 450.246 l +f* +500.328 445.42 m +500.767 446.01 l +497.694 448.276 l +497.475 447.981 l +497.255 448.276 l +496.855 447.981 l +f +497.255 448.276 m +497.694 447.686 l +500.767 449.951 l +500.5 450.246 l +500.869 450.246 l +500.869 450.94 l +f +500.869 450.246 m +500.131 450.246 l +500.131 445.715 l +500.547 445.715 l +500.328 445.42 l +500.869 445.021 l +f +500.919 447.245 10.83 1.47258 re +f* +500.919 449.067 m +500.919 448.333 l +511.749 448.333 l +511.7 448.717 l +512.069 448.717 l +512.069 449.067 l +f +512.069 448.717 m +511.331 448.717 l +511.331 447.245 l +511.749 447.2 l +511.749 446.833 l +512.069 446.833 l +f +511.749 446.833 m +511.749 447.567 l +500.919 447.567 l +500.9 447.245 l +500.531 447.245 l +500.531 446.833 l +f +500.531 447.245 m +501.269 447.245 l +501.269 448.717 l +500.919 448.7 l +500.919 449.067 l +500.531 449.067 l +f +BT +0 g +1 0 0 0.995457 463.599 377.329 Tm +(replica n) Tj +ET +466.485 368.312 m +466.683 367.605 l +468.338 368.065 l +468.239 368.419 l +468.224 368.786 l +468.181 368.784 l +f +468.224 368.786 m +468.254 368.052 l +469.909 368.118 l +469.894 368.5 l +469.894 368.867 l +469.88 368.852 l +f +469.894 368.867 m +469.894 368.133 l +473.205 368.133 l +473.205 368.5 l +473.205 368.867 l +473.205 368.867 l +f +473.205 368.867 m +473.205 368.133 l +476.515 368.133 l +476.515 368.5 l +476.515 368.867 l +476.515 368.867 l +f +476.515 368.867 m +476.515 368.133 l +479.825 368.133 l +479.825 368.5 l +479.825 368.867 l +479.825 368.867 l +f +479.825 368.867 m +479.825 368.133 l +483.136 368.133 l +483.136 368.5 l +483.136 368.867 l +483.136 368.867 l +f +483.136 368.867 m +483.136 368.133 l +486.446 368.133 l +486.446 368.5 l +486.446 368.867 l +486.446 368.867 l +f +486.446 368.867 m +486.446 368.133 l +489.756 368.133 l +489.756 368.5 l +489.756 368.867 l +489.756 368.867 l +f +489.756 368.867 m +489.756 368.133 l +491.429 368.133 l +491.429 368.5 l +491.429 368.867 l +491.429 368.867 l +f +491.429 368.867 m +491.429 368.133 l +493.205 368.133 l +493.205 368.5 l +493.205 368.867 l +493.205 368.867 l +f +493.205 368.867 m +493.205 368.133 l +495.136 368.133 l +495.136 368.5 l +495.136 368.867 l +495.136 368.867 l +f +495.136 368.867 m +495.136 368.133 l +497.067 368.133 l +497.067 368.485 l +497.07 368.852 l +497.067 368.867 l +f +497.07 368.852 m +497.063 368.118 l +497.971 368.109 l +497.974 368.476 l +497.998 368.843 l +497.977 368.844 l +f +497.998 368.843 m +497.95 368.11 l +498.818 368.053 l +498.843 368.419 l +498.91 368.78 l +498.889 368.784 l +f +498.91 368.78 m +498.775 368.058 l +499.618 367.902 l +499.685 368.263 l +499.813 368.607 l +499.783 368.618 l +f +499.813 368.607 m +499.558 367.918 l +500.388 367.614 l +500.515 367.959 l +500.703 368.274 l +500.674 368.292 l +f +500.703 368.274 m +500.326 367.643 l +501.145 367.158 l +501.334 367.474 l +501.569 367.757 l +501.547 367.775 l +f +501.569 367.757 m +501.099 367.191 l +501.866 366.558 l +502.101 366.841 l +502.375 367.087 l +502.357 367.107 l +f +502.375 367.087 m +501.827 366.595 l +502.491 365.864 l +502.765 366.11 l +503.074 366.31 l +503.058 366.334 l +f +503.074 366.31 m +502.455 365.91 l +502.964 365.13 l +503.274 365.329 l +503.616 365.466 l +503.602 365.499 l +f +503.616 365.466 m +502.931 365.192 l +503.25 364.403 l +503.593 364.54 l +503.954 364.615 l +503.947 364.647 l +f +503.954 364.615 m +503.231 364.466 l +503.395 363.677 l +503.756 363.752 l +504.124 363.779 l +504.122 363.803 l +f +504.124 363.779 m +503.389 363.724 l +503.449 362.935 l +503.817 362.963 l +504.185 362.967 l +504.184 362.991 l +f +504.185 362.967 m +503.448 362.959 l +503.456 362.17 l +503.8 362.174 l +504.169 362.174 l +504.194 362.178 l +f +504.169 362.174 m +503.431 362.174 l +503.431 359.019 l +503.8 359.019 l +504.169 359.019 l +504.169 359.019 l +f +504.169 359.019 m +503.431 359.019 l +503.431 357.507 l +503.8 357.507 l +504.169 357.507 l +504.169 357.507 l +f +504.169 357.507 m +503.431 357.507 l +503.431 356.389 l +503.8 356.389 l +504.169 356.389 l +504.169 356.389 l +f +504.169 356.389 m +503.431 356.389 l +503.431 356.077 l +503.8 356.077 l +504.169 356.077 l +504.169 356.077 l +f +504.169 356.077 m +503.431 356.077 l +503.431 355.65 l +503.8 355.65 l +504.169 355.65 l +504.169 355.65 l +f +504.169 355.65 m +503.431 355.65 l +503.431 355.337 l +503.8 355.337 l +504.169 355.337 l +504.169 355.337 l +f +504.169 355.337 m +503.431 355.337 l +503.431 354.22 l +503.8 354.22 l +504.169 354.22 l +504.169 354.22 l +f +504.169 354.22 m +503.431 354.22 l +503.431 352.708 l +503.8 352.708 l +504.169 352.708 l +504.169 352.708 l +f +504.169 352.708 m +503.431 352.708 l +503.431 349.553 l +503.8 349.553 l +504.169 349.553 l +504.169 349.553 l +f +504.169 349.553 m +503.431 349.553 l +503.431 346.397 l +503.8 346.397 l +504.169 346.397 l +504.169 346.397 l +f +504.169 346.397 m +503.431 346.397 l +503.431 343.242 l +503.825 343.242 l +504.194 343.238 l +504.169 343.242 l +f +504.194 343.238 m +503.456 343.246 l +503.448 342.457 l +503.817 342.453 l +504.184 342.425 l +504.185 342.449 l +f +504.184 342.425 m +503.449 342.481 l +503.389 341.692 l +503.756 341.664 l +504.118 341.59 l +504.122 341.613 l +f +504.118 341.59 m +503.395 341.739 l +503.231 340.95 l +503.593 340.876 l +503.935 340.738 l +503.947 340.769 l +f +503.935 340.738 m +503.25 341.013 l +502.931 340.224 l +503.274 340.087 l +503.583 339.887 l +503.602 339.917 l +f +503.583 339.887 m +502.964 340.286 l +502.455 339.506 l +502.765 339.306 l +503.039 339.06 l +503.058 339.082 l +f +503.039 339.06 m +502.491 339.552 l +501.827 338.821 l +502.101 338.575 l +502.336 338.292 l +502.357 338.309 l +f +502.336 338.292 m +501.866 338.857 l +501.099 338.225 l +501.334 337.942 l +501.522 337.626 l +501.547 337.641 l +f +501.522 337.626 m +501.145 338.258 l +500.326 337.773 l +500.515 337.457 l +500.642 337.113 l +500.674 337.124 l +f +500.642 337.113 m +500.388 337.802 l +499.558 337.498 l +499.685 337.153 l +499.753 336.792 l +499.783 336.798 l +f +499.753 336.792 m +499.618 337.514 l +498.775 337.358 l +498.843 336.997 l +498.867 336.631 l +498.889 336.632 l +f +498.867 336.631 m +498.818 337.363 l +497.95 337.306 l +497.974 336.94 l +497.977 336.572 l +497.998 336.573 l +f +497.977 336.572 m +497.971 337.307 l +497.063 337.299 l +497.067 336.9 l +497.067 336.533 l +497.07 336.564 l +f +497.067 336.533 m +497.067 337.267 l +495.136 337.267 l +495.136 336.9 l +495.136 336.533 l +495.136 336.533 l +f +495.136 336.533 m +495.136 337.267 l +493.205 337.267 l +493.205 336.9 l +493.205 336.533 l +493.205 336.533 l +f +493.205 336.533 m +493.205 337.267 l +491.429 337.267 l +491.429 336.9 l +491.429 336.533 l +491.429 336.533 l +f +491.429 336.533 m +491.429 337.267 l +489.756 337.267 l +489.756 336.9 l +489.756 336.533 l +489.756 336.533 l +f +489.756 336.533 m +489.756 337.267 l +486.446 337.267 l +486.446 336.9 l +486.446 336.533 l +486.446 336.533 l +f +486.446 336.533 m +486.446 337.267 l +483.136 337.267 l +483.136 336.9 l +483.136 336.533 l +483.136 336.533 l +f +483.136 336.533 m +483.136 337.267 l +479.825 337.267 l +479.825 336.9 l +479.825 336.533 l +479.825 336.533 l +f +479.825 336.533 m +479.825 337.267 l +476.515 337.267 l +476.515 336.9 l +476.515 336.533 l +476.515 336.533 l +f +476.515 336.533 m +476.515 337.267 l +473.205 337.267 l +473.205 336.9 l +473.205 336.533 l +473.205 336.533 l +f +473.205 336.533 m +473.205 337.267 l +469.894 337.267 l +469.894 336.931 l +469.88 336.565 l +469.894 336.533 l +f +469.88 336.565 m +469.909 337.298 l +468.254 337.364 l +468.239 336.997 l +468.14 336.643 l +468.181 336.632 l +f +468.14 336.643 m +468.338 337.351 l +466.683 337.811 l +466.584 337.457 l +466.395 337.142 l +466.438 337.117 l +f +466.395 337.142 m +466.772 337.773 l +465.953 338.258 l +465.765 337.942 l +465.53 337.659 l +465.552 337.641 l +f +465.53 337.659 m +466 338.225 l +465.233 338.857 l +464.998 338.575 l +464.724 338.329 l +464.742 338.309 l +f +464.724 338.329 m +465.271 338.821 l +464.608 339.552 l +464.334 339.306 l +464.024 339.106 l +464.041 339.082 l +f +464.024 339.106 m +464.643 339.506 l +464.135 340.286 l +463.825 340.087 l +463.483 339.95 l +463.496 339.917 l +f +463.483 339.95 m +464.167 340.224 l +463.848 341.013 l +463.506 340.876 l +463.145 340.801 l +463.152 340.769 l +f +463.145 340.801 m +463.867 340.95 l +463.704 341.739 l +463.342 341.664 l +462.975 341.636 l +462.977 341.613 l +f +462.975 341.636 m +463.71 341.692 l +463.65 342.481 l +463.282 342.453 l +462.913 342.449 l +462.914 342.425 l +f +462.913 342.449 m +463.651 342.457 l +463.642 343.246 l +463.3 343.242 l +462.931 343.242 l +462.905 343.238 l +f +462.931 343.242 m +463.669 343.242 l +463.669 346.397 l +463.3 346.397 l +462.931 346.397 l +462.931 346.397 l +f +462.931 346.397 m +463.669 346.397 l +463.669 349.553 l +463.3 349.553 l +462.931 349.553 l +462.931 349.553 l +f +462.931 349.553 m +463.669 349.553 l +463.669 352.708 l +463.3 352.708 l +462.931 352.708 l +462.931 352.708 l +f +462.931 352.708 m +463.669 352.708 l +463.669 355.863 l +463.3 355.863 l +462.931 355.863 l +462.931 355.863 l +f +462.931 355.863 m +463.669 355.863 l +463.669 359.019 l +463.3 359.019 l +462.931 359.019 l +462.931 359.019 l +f +462.931 359.019 m +463.669 359.019 l +463.669 362.174 l +463.274 362.174 l +462.905 362.178 l +462.931 362.174 l +f +462.905 362.178 m +463.642 362.17 l +463.651 362.959 l +463.282 362.963 l +462.914 362.991 l +462.913 362.967 l +f +462.914 362.991 m +463.65 362.935 l +463.71 363.724 l +463.342 363.752 l +462.981 363.826 l +462.976 363.803 l +f +462.981 363.826 m +463.704 363.677 l +463.867 364.466 l +463.506 364.54 l +463.164 364.678 l +463.152 364.647 l +f +463.164 364.678 m +463.848 364.403 l +464.167 365.192 l +463.825 365.329 l +463.516 365.529 l +463.496 365.499 l +f +463.516 365.529 m +464.135 365.13 l +464.643 365.91 l +464.334 366.11 l +464.06 366.356 l +464.04 366.334 l +f +464.06 366.356 m +464.608 365.864 l +465.271 366.595 l +464.998 366.841 l +464.762 367.124 l +464.742 367.107 l +f +464.762 367.124 m +465.233 366.558 l +466 367.191 l +465.765 367.474 l +465.576 367.79 l +465.552 367.775 l +f +465.576 367.79 m +465.953 367.158 l +466.772 367.643 l +466.395 368.274 l +f +BT +/R691 6.82548 Tf +1 0 0 0.995457 467.333 348.784 Tm +(replication) Tj +1 0 0 0.995457 473.021 341.612 Tm +(library) Tj +ET +463.162 357.133 40.5735 0.734326 re +f +BT +/R691 6.82544 Tf +1 0 0 0.995457 475.295 361.985 Tm +(snfsd) Tj +ET +462.708 374.15 m +463.307 372.808 l +464.535 373.351 l +464.236 374.022 l +464.072 374.738 l +464.002 374.722 l +f +464.072 374.738 m +464.4 373.306 l +465.629 373.585 l +465.465 374.301 l +465.403 375.033 l +465.351 375.028 l +f +465.403 375.033 m +465.526 373.569 l +466.755 373.672 l +466.693 374.404 l +466.684 375.138 l +466.658 375.138 l +f +466.684 375.138 m +466.702 373.669 l +467.931 373.684 l +467.922 374.4 l +467.922 375.134 l +467.913 375.153 l +f +467.922 375.134 m +467.922 373.666 l +472.836 373.666 l +472.836 374.4 l +472.836 375.134 l +472.836 375.134 l +f +472.836 375.134 m +472.836 373.666 l +477.751 373.666 l +477.751 374.4 l +477.751 375.134 l +477.751 375.134 l +f +477.751 375.134 m +477.751 373.666 l +482.665 373.666 l +482.665 374.4 l +482.665 375.134 l +482.665 375.134 l +f +482.665 375.134 m +482.665 373.666 l +487.58 373.666 l +487.58 374.4 l +487.58 375.134 l +487.58 375.134 l +f +487.58 375.134 m +487.58 373.666 l +492.494 373.666 l +492.494 374.4 l +492.494 375.134 l +492.494 375.134 l +f +492.494 375.134 m +492.494 373.666 l +497.409 373.666 l +497.409 374.4 l +497.409 375.134 l +497.409 375.134 l +f +497.409 375.134 m +497.409 373.666 l +499.891 373.666 l +499.891 374.4 l +499.891 375.134 l +499.891 375.134 l +f +499.891 375.134 m +499.891 373.666 l +502.528 373.666 l +502.528 374.4 l +502.528 375.134 l +502.528 375.134 l +f +502.528 375.134 m +502.528 373.666 l +505.395 373.666 l +505.395 374.4 l +505.395 375.134 l +505.395 375.134 l +f +505.395 375.134 m +505.395 373.666 l +508.261 373.666 l +508.261 374.418 l +508.269 375.153 l +508.261 375.134 l +f +508.269 375.153 m +508.253 373.684 l +509.6 373.669 l +509.608 374.404 l +509.667 375.136 l +509.642 375.138 l +f +509.667 375.136 m +509.549 373.672 l +510.839 373.569 l +510.898 374.301 l +511.059 375.018 l +511.008 375.029 l +f +511.059 375.018 m +510.736 373.584 l +511.988 373.306 l +512.149 374.022 l +512.447 374.694 l +512.381 374.723 l +f +512.447 374.694 m +511.85 373.351 l +513.082 372.808 l +513.381 373.479 l +513.81 374.076 l +513.748 374.12 l +f +513.81 374.076 m +512.951 372.882 l +514.167 372.016 l +514.596 372.613 l +515.117 373.133 l +515.075 373.175 l +f +515.117 373.133 m +514.076 372.093 l +515.215 370.962 l +515.735 371.483 l +516.325 371.923 l +516.294 371.965 l +f +516.325 371.923 m +515.145 371.042 l +516.131 369.735 l +516.721 370.176 l +517.37 370.524 l +517.344 370.573 l +f +517.37 370.524 m +516.071 369.828 l +516.827 368.433 l +517.476 368.782 l +518.175 369.014 l +518.155 369.074 l +f +518.175 369.014 m +516.776 368.549 l +517.25 367.139 l +517.949 367.372 l +518.676 367.497 l +518.667 367.552 l +f +518.676 367.497 m +517.222 367.248 l +517.466 365.839 l +518.193 365.963 l +518.929 366.009 l +518.926 366.049 l +f +518.929 366.009 m +517.456 365.917 l +517.546 364.508 l +518.282 364.554 l +519.02 364.561 l +519.02 364.58 l +f +519.02 364.561 m +517.544 364.547 l +517.557 363.138 l +518.3 363.145 l +519.038 363.145 l +519.033 363.151 l +f +519.038 363.145 m +517.562 363.145 l +517.562 357.508 l +518.3 357.508 l +519.038 357.508 l +519.038 357.508 l +f +519.038 357.508 m +517.562 357.508 l +517.562 356.113 l +518.3 356.113 l +519.038 356.113 l +519.038 356.113 l +f +519.038 356.113 m +517.562 356.113 l +517.562 353.676 l +518.3 353.676 l +519.038 353.676 l +519.038 353.676 l +f +519.038 353.676 m +517.562 353.676 l +517.562 352.81 l +518.3 352.81 l +519.038 352.81 l +519.038 352.81 l +f +519.038 352.81 m +517.562 352.81 l +517.562 352.253 l +518.3 352.253 l +519.038 352.253 l +519.038 352.253 l +f +519.038 352.253 m +517.562 352.253 l +517.562 351.489 l +518.3 351.489 l +519.038 351.489 l +519.038 351.489 l +f +519.038 351.489 m +517.562 351.489 l +517.562 350.931 l +518.3 350.931 l +519.038 350.931 l +519.038 350.931 l +f +519.038 350.931 m +517.562 350.931 l +517.562 350.065 l +518.3 350.065 l +519.038 350.065 l +519.038 350.065 l +f +519.038 350.065 m +517.562 350.065 l +517.562 347.629 l +518.3 347.629 l +519.038 347.629 l +519.038 347.629 l +f +519.038 347.629 m +517.562 347.629 l +517.562 346.234 l +518.3 346.234 l +519.038 346.234 l +519.038 346.234 l +f +519.038 346.234 m +517.562 346.234 l +517.562 340.597 l +518.3 340.597 l +519.038 340.597 l +519.038 340.597 l +f +519.038 340.597 m +517.562 340.597 l +517.562 334.96 l +518.3 334.96 l +519.038 334.96 l +519.038 334.96 l +f +519.038 334.96 m +517.562 334.96 l +517.562 329.323 l +518.295 329.323 l +519.033 329.317 l +519.038 329.323 l +f +519.033 329.317 m +517.557 329.33 l +517.544 327.921 l +518.282 327.914 l +519.018 327.868 l +519.02 327.888 l +f +519.018 327.868 m +517.546 327.96 l +517.456 326.551 l +518.193 326.505 l +518.92 326.381 l +518.926 326.419 l +f +518.92 326.381 m +517.466 326.629 l +517.222 325.22 l +517.949 325.096 l +518.649 324.863 l +518.667 324.916 l +f +518.649 324.863 m +517.25 325.329 l +516.776 323.919 l +517.476 323.686 l +518.125 323.338 l +518.155 323.394 l +f +518.125 323.338 m +516.827 324.035 l +516.071 322.64 l +516.721 322.292 l +517.311 321.851 l +517.344 321.895 l +f +517.311 321.851 m +516.131 322.733 l +515.145 321.426 l +515.735 320.985 l +516.256 320.465 l +516.294 320.503 l +f +516.256 320.465 m +515.215 321.505 l +514.076 320.375 l +514.596 319.855 l +515.026 319.258 l +515.074 319.293 l +f +515.026 319.258 m +514.167 320.452 l +512.951 319.586 l +513.381 318.989 l +513.679 318.318 l +513.748 318.348 l +f +513.679 318.318 m +513.082 319.66 l +511.85 319.117 l +512.149 318.446 l +512.31 317.729 l +512.381 317.745 l +f +512.31 317.729 m +511.988 319.162 l +510.736 318.884 l +510.898 318.167 l +510.957 317.435 l +511.008 317.439 l +f +510.957 317.435 m +510.839 318.899 l +509.549 318.796 l +509.608 318.064 l +509.616 317.33 l +509.642 317.33 l +f +509.616 317.33 m +509.6 318.799 l +508.253 318.784 l +508.261 318 l +508.261 317.266 l +508.269 317.315 l +f +508.261 317.266 m +508.261 318.734 l +505.395 318.734 l +505.395 318 l +505.395 317.266 l +505.395 317.266 l +f +505.395 317.266 m +505.395 318.734 l +502.528 318.734 l +502.528 318 l +502.528 317.266 l +502.528 317.266 l +f +502.528 317.266 m +502.528 318.734 l +499.891 318.734 l +499.891 318 l +499.891 317.266 l +499.891 317.266 l +f +499.891 317.266 m +499.891 318.734 l +497.409 318.734 l +497.409 318 l +497.409 317.266 l +497.409 317.266 l +f +497.409 317.266 m +497.409 318.734 l +492.494 318.734 l +492.494 318 l +492.494 317.266 l +492.494 317.266 l +f +492.494 317.266 m +492.494 318.734 l +487.58 318.734 l +487.58 318 l +487.58 317.266 l +487.58 317.266 l +f +487.58 317.266 m +487.58 318.734 l +482.665 318.734 l +482.665 318 l +482.665 317.266 l +482.665 317.266 l +f +482.665 317.266 m +482.665 318.734 l +477.751 318.734 l +477.751 318 l +477.751 317.266 l +477.751 317.266 l +f +477.751 317.266 m +477.751 318.734 l +472.836 318.734 l +472.836 318 l +472.836 317.266 l +472.836 317.266 l +f +472.836 317.266 m +472.836 318.734 l +467.922 318.734 l +467.922 318.05 l +467.913 317.315 l +467.922 317.266 l +f +467.913 317.315 m +467.931 318.784 l +466.702 318.799 l +466.693 318.064 l +466.631 317.332 l +466.658 317.33 l +f +466.631 317.332 m +466.755 318.796 l +465.526 318.899 l +465.465 318.167 l +465.301 317.451 l +465.351 317.44 l +f +465.301 317.451 m +465.629 318.883 l +464.4 319.162 l +464.236 318.446 l +463.937 317.775 l +464.002 317.746 l +f +463.937 317.775 m +464.535 319.117 l +463.307 319.66 l +463.007 318.989 l +462.578 318.392 l +462.639 318.348 l +f +462.578 318.392 m +463.437 319.586 l +462.221 320.452 l +461.792 319.855 l +461.271 319.335 l +461.313 319.293 l +f +461.271 319.335 m +462.312 320.375 l +461.173 321.505 l +460.653 320.985 l +460.063 320.545 l +460.094 320.502 l +f +460.063 320.545 m +461.242 321.426 l +460.257 322.733 l +459.667 322.292 l +459.018 321.943 l +459.044 321.895 l +f +459.018 321.943 m +460.316 322.64 l +459.561 324.035 l +458.912 323.686 l +458.212 323.453 l +458.233 323.394 l +f +458.212 323.453 m +459.612 323.919 l +459.138 325.329 l +458.438 325.096 l +457.711 324.971 l +457.721 324.916 l +f +457.711 324.971 m +459.166 325.22 l +458.922 326.629 l +458.195 326.505 l +457.459 326.459 l +457.462 326.419 l +f +457.459 326.459 m +458.932 326.551 l +458.842 327.96 l +458.106 327.914 l +457.368 327.907 l +457.368 327.888 l +f +457.368 327.907 m +458.843 327.921 l +458.831 329.33 l +458.1 329.323 l +457.362 329.323 l +457.355 329.317 l +f +457.362 329.323 m +458.838 329.323 l +458.838 334.96 l +458.1 334.96 l +457.362 334.96 l +457.362 334.96 l +f +457.362 334.96 m +458.838 334.96 l +458.838 340.597 l +458.1 340.597 l +457.362 340.597 l +457.362 340.597 l +f +457.362 340.597 m +458.838 340.597 l +458.838 346.234 l +458.1 346.234 l +457.362 346.234 l +457.362 346.234 l +f +457.362 346.234 m +458.838 346.234 l +458.838 351.871 l +458.1 351.871 l +457.362 351.871 l +457.362 351.871 l +f +457.362 351.871 m +458.838 351.871 l +458.838 357.508 l +458.1 357.508 l +457.362 357.508 l +457.362 357.508 l +f +457.362 357.508 m +458.838 357.508 l +458.838 363.145 l +458.093 363.145 l +457.355 363.151 l +457.362 363.145 l +f +457.355 363.151 m +458.831 363.138 l +458.843 364.547 l +458.106 364.554 l +457.369 364.6 l +457.368 364.58 l +f +457.369 364.6 m +458.842 364.508 l +458.932 365.917 l +458.195 365.963 l +457.468 366.087 l +457.462 366.049 l +f +457.468 366.087 m +458.922 365.839 l +459.166 367.248 l +458.438 367.372 l +457.739 367.605 l +457.721 367.552 l +f +457.739 367.605 m +459.138 367.139 l +459.612 368.549 l +458.912 368.782 l +458.263 369.13 l +458.233 369.074 l +f +458.263 369.13 m +459.561 368.433 l +460.316 369.828 l +459.667 370.176 l +459.077 370.617 l +459.044 370.573 l +f +459.077 370.617 m +460.257 369.735 l +461.242 371.042 l +460.653 371.483 l +460.132 372.003 l +460.094 371.965 l +f +460.132 372.003 m +461.173 370.962 l +462.312 372.093 l +461.792 372.613 l +461.362 373.21 l +461.313 373.175 l +f +461.362 373.21 m +462.221 372.016 l +463.437 372.882 l +462.578 374.076 l +f +458 330.433 59.9384 0.734326 re +f +BT +/R691 6.84901 Tf +1 0 0 0.995457 472.399 323.164 Tm +(kernel VM) Tj +ET +0.501961 g +513.356 329.086 m +511.08 326.027 l +508.804 329.086 l +f* +513.653 328.867 m +513.06 329.304 l +510.784 326.246 l +511.08 326.027 l +510.784 325.809 l +511.08 325.41 l +f +510.784 325.809 m +511.377 326.246 l +509.101 329.304 l +508.804 329.1 l +508.804 329.467 l +508.061 329.467 l +f +508.804 329.467 m +508.804 328.733 l +513.356 328.733 l +513.356 329.086 l +513.653 328.867 l +514.099 329.467 l +f +510.341 329.393 1.4793 33.7112 re +f* +509.931 329.393 m +510.669 329.393 l +510.669 363.105 l +510.341 363.1 l +510.341 363.467 l +509.931 363.467 l +f +510.341 363.467 m +510.341 362.733 l +511.82 362.733 l +511.8 363.105 l +512.169 363.105 l +512.169 363.467 l +f +512.169 363.105 m +511.431 363.105 l +511.431 329.393 l +511.82 329.4 l +511.82 329.033 l +512.169 329.033 l +f +511.82 329.033 m +511.82 329.767 l +510.341 329.767 l +510.3 329.393 l +509.931 329.393 l +509.931 329.033 l +f +500.547 360.411 m +497.475 362.676 l +500.547 364.942 l +f* +500.328 360.116 m +500.767 360.706 l +497.694 362.971 l +497.475 362.676 l +497.255 362.971 l +496.855 362.676 l +f +497.255 362.971 m +497.694 362.381 l +500.767 364.647 l +500.5 364.942 l +500.869 364.942 l +500.869 365.636 l +f +500.869 364.942 m +500.131 364.942 l +500.131 360.411 l +500.547 360.411 l +500.328 360.116 l +500.869 359.717 l +f +500.919 361.94 10.83 1.47258 re +f* +500.919 363.767 m +500.919 363.033 l +511.749 363.033 l +511.7 363.413 l +512.069 363.413 l +512.069 363.767 l +f +512.069 363.413 m +511.331 363.413 l +511.331 361.94 l +511.749 361.9 l +511.749 361.533 l +512.069 361.533 l +f +511.749 361.533 m +511.749 362.267 l +500.919 362.267 l +500.9 361.94 l +500.531 361.94 l +500.531 361.533 l +f +500.531 361.94 m +501.269 361.94 l +501.269 363.413 l +500.919 363.4 l +500.919 363.767 l +500.531 363.767 l +f +434.045 385.282 m +434.166 388.996 l +437.682 386.118 l +434.045 385.282 l +f* +433.676 385.294 m +434.413 385.27 l +434.535 388.984 l +434.166 388.996 l +434.4 389.279 l +433.822 389.753 l +f +434.4 389.279 m +433.932 388.712 l +437.447 385.834 l +437.682 386.118 l +437.765 385.76 l +438.494 385.928 l +f +437.765 385.76 m +437.599 386.476 l +433.962 385.64 l +434.045 385.282 l +433.676 385.294 l +433.66 384.817 l +f +465.525 426.958 m +469.234 427.881 l +469.04 424.08 l +f* +465.435 427.314 m +465.614 426.601 l +469.323 427.525 l +469.234 427.881 l +469.602 427.862 l +469.627 428.357 l +f +469.602 427.862 m +468.866 427.9 l +468.672 424.098 l +469.04 424.08 l +468.806 423.796 l +469.372 423.333 l +f +468.806 423.796 m +469.275 424.363 l +465.759 427.241 l +465.525 426.958 l +465.435 427.314 l +464.725 427.137 l +f +467.933 425.147 m +436.638 387.262 l +435.496 388.198 l +466.791 426.082 l +f* +468.218 424.914 m +467.648 425.38 l +436.353 387.496 l +436.638 387.262 l +436.404 386.979 l +436.689 386.746 l +f +436.404 386.979 m +436.873 387.546 l +435.73 388.481 l +435.496 388.198 l +435.211 388.431 l +434.977 388.147 l +f +435.211 388.431 m +435.781 387.964 l +467.075 425.849 l +466.791 426.082 l +467.025 426.366 l +466.74 426.599 l +f +467.025 426.366 m +466.556 425.799 l +467.699 424.864 l +467.933 425.147 l +468.218 424.914 l +468.452 425.197 l +f +468.299 340.127 m +464.681 341.042 l +468.259 343.843 l +468.299 340.127 l +f* +468.208 339.772 m +468.39 340.483 l +464.772 341.398 l +464.681 341.042 l +464.453 341.331 l +463.864 340.87 l +f +464.453 341.331 m +464.909 340.754 l +468.487 343.554 l +468.259 343.843 l +468.628 343.847 l +468.62 344.592 l +f +468.628 343.847 m +467.891 343.839 l +467.93 340.124 l +468.299 340.127 l +468.208 339.772 l +468.673 339.654 l +f +434.165 379.679 m +434.055 383.483 l +437.744 382.479 l +f* +433.797 379.668 m +434.534 379.689 l +434.424 383.494 l +434.055 383.483 l +434.153 383.838 l +433.673 383.968 l +f +434.153 383.838 m +433.958 383.129 l +437.646 382.125 l +437.744 382.479 l +437.972 382.191 l +438.547 382.641 l +f +437.972 382.191 m +437.516 382.768 l +433.937 379.968 l +434.165 379.679 l +433.797 379.668 l +433.818 378.94 l +f +436.459 381.632 m +466.913 343.074 l +465.75 342.164 l +435.296 380.722 l +f* +436.749 381.859 m +436.169 381.405 l +466.623 342.847 l +466.913 343.074 l +467.141 342.786 l +467.431 343.012 l +f +467.141 342.786 m +466.685 343.363 l +465.522 342.453 l +465.75 342.164 l +465.46 341.937 l +465.688 341.648 l +f +465.46 341.937 m +466.04 342.391 l +435.586 380.949 l +435.296 380.722 l +435.068 381.011 l +434.778 380.784 l +f +435.068 381.011 m +435.524 380.433 l +436.687 381.344 l +436.459 381.632 l +436.749 381.859 l +436.521 382.148 l +f +BT +0 g +/R694 10 Tf +1 0 0 1 333.96 294.6 Tm +(Figure) Tj +1 0 0 1 362.52 294.6 Tm +(2:) Tj +1 0 0 1 373.8 294.6 Tm +(Replicated) Tj +1 0 0 1 418.92 294.6 Tm +(File) Tj +1 0 0 1 436.92 294.6 Tm +(System) Tj +1 0 0 1 468.72 294.6 Tm +(Architecture.) Tj +1 0 0 1 324.96 274.8 Tm +(Each) Tj +1 0 0 1 351.72 274.8 Tm +(replica) Tj +1 0 0 1 385.68 274.8 Tm +(runs) Tj +1 0 0 1 409.68 274.8 Tm +(a) Tj +1 0 0 1 420.96 274.8 Tm +(user) Tj +1 0 0 1 437.4 274.8 Tm +(-le) Tj +1 0 0 1 447.72 274.8 Tm +(v) Tj +1 0 0 1 452.64 274.8 Tm +(el) Tj +1 0 0 1 466.56 274.8 Tm +(process) Tj +1 0 0 1 503.279 274.8 Tm +(with) Tj +1 0 0 1 527.879 274.8 Tm +(the) Tj +1 0 0 1 315 262.92 Tm +(replication) Tj +1 0 0 1 361.08 262.92 Tm +(library) Tj +1 0 0 1 391.08 262.92 Tm +(and) Tj +1 0 0 1 408.84 262.92 Tm +(our) Tj +1 0 0 1 425.52 262.92 Tm +(NFS) Tj +1 0 0 1 447.24 262.92 Tm +(V2) Tj +1 0 0 1 462.84 262.92 Tm +(daemon,) Tj +1 0 0 1 500.639 262.92 Tm +(which) Tj +1 0 0 1 528.359 262.92 Tm +(we) Tj +1 0 0 1 315 250.92 Tm +(will) Tj +1 0 0 1 333.72 250.92 Tm +(refer) Tj +1 0 0 1 355.8 250.92 Tm +(to) Tj +1 0 0 1 366.84 250.92 Tm +(as) Tj +/R697 10 Tf +1 0 0 1 378.48 250.92 Tm +(snfsd) Tj +/R694 10 Tf +1 0 0 1 402.24 250.92 Tm +(\(for) Tj +1 0 0 1 420.36 250.92 Tm +(simple) Tj +/R697 10 Tf +1 0 0 1 450.24 250.92 Tm +(nfsd) Tj +/R694 10 Tf +1 0 0 1 466.92 250.92 Tm +(\).) Tj +1 0 0 1 478.679 250.92 Tm +(The) Tj +1 0 0 1 497.399 250.92 Tm +(replication) Tj +1 0 0 1 315 239.04 Tm +(library) Tj +1 0 0 1 345.6 239.04 Tm +(recei) Tj +1 0 0 1 364.8 239.04 Tm +(v) Tj +1 0 0 1 369.72 239.04 Tm +(es) Tj +1 0 0 1 381.96 239.04 Tm +(requests) Tj +1 0 0 1 418.56 239.04 Tm +(from) Tj +1 0 0 1 442.08 239.04 Tm +(the) Tj +1 0 0 1 458.279 239.04 Tm +(relay) Tj +1 0 0 1 477.719 239.04 Tm +(,) Tj +1 0 0 1 484.439 239.04 Tm +(interacts) Tj +1 0 0 1 522.239 239.04 Tm +(with) Tj +/R697 10 Tf +1 0 0 1 315 227.04 Tm +(snfsd) Tj +/R694 10 Tf +1 0 0 1 338.28 227.04 Tm +(by) Tj +1 0 0 1 351.12 227.04 Tm +(making) Tj +1 0 0 1 383.88 227.04 Tm +(upcalls,) Tj +1 0 0 1 417.48 227.04 Tm +(and) Tj +1 0 0 1 434.64 227.04 Tm +(packages) Tj +1 0 0 1 474 227.04 Tm +(NFS) Tj +1 0 0 1 495.119 227.04 Tm +(replies) Tj +1 0 0 1 524.519 227.04 Tm +(into) Tj +1 0 0 1 315 215.16 Tm +(replication) Tj +1 0 0 1 360.12 215.16 Tm +(protocol) Tj +1 0 0 1 395.76 215.16 Tm +(replies) Tj +1 0 0 1 424.8 215.16 Tm +(that) Tj +1 0 0 1 442.319 215.16 Tm +(it) Tj +1 0 0 1 450.359 215.16 Tm +(sends) Tj +1 0 0 1 474.959 215.16 Tm +(to) Tj +1 0 0 1 485.159 215.16 Tm +(the) Tj +1 0 0 1 499.799 215.16 Tm +(relay) Tj +1 0 0 1 519.239 215.16 Tm +(.) Tj +1 0 0 1 324.96 202.8 Tm +(W) Tj +1 0 0 1 333.48 202.8 Tm +(e) Tj +1 0 0 1 342.72 202.8 Tm +(implemented) Tj +/R697 10 Tf +1 0 0 1 399.48 202.8 Tm +(snfsd) Tj +/R694 10 Tf +1 0 0 1 424.68 202.8 Tm +(using) Tj +1 0 0 1 450.96 202.8 Tm +(a) Tj +1 0 0 1 460.2 202.8 Tm +(\256x) Tj +1 0 0 1 470.64 202.8 Tm +(ed-size) Tj +1 0 0 1 503.519 202.8 Tm +(memory-) Tj +1 0 0 1 315 190.8 Tm +(mapped) Tj +1 0 0 1 350.76 190.8 Tm +(\256le.) Tj +1 0 0 1 374.52 190.8 Tm +(All) Tj +1 0 0 1 391.44 190.8 Tm +(the) Tj +1 0 0 1 407.76 190.8 Tm +(\256le) Tj +1 0 0 1 424.68 190.8 Tm +(system) Tj +1 0 0 1 456.6 190.8 Tm +(data) Tj +1 0 0 1 477.359 190.8 Tm +(structures,) Tj +1 0 0 1 523.079 190.8 Tm +(e.g.,) Tj +1 0 0 1 315 178.92 Tm +(inodes,) Tj +1 0 0 1 345.96 178.92 Tm +(blocks) Tj +1 0 0 1 374.52 178.92 Tm +(and) Tj +1 0 0 1 391.44 178.92 Tm +(their) Tj +1 0 0 1 412.2 178.92 Tm +(free) Tj +1 0 0 1 430.08 178.92 Tm +(lists,) Tj +1 0 0 1 451.199 178.92 Tm +(are) Tj +1 0 0 1 465.839 178.92 Tm +(in) Tj +1 0 0 1 476.039 178.92 Tm +(the) Tj +1 0 0 1 490.799 178.92 Tm +(mapped) Tj +1 0 0 1 524.759 178.92 Tm +(\256le.) Tj +1 0 0 1 315 166.92 Tm +(W) Tj +1 0 0 1 323.52 166.92 Tm +(e) Tj +1 0 0 1 330.72 166.92 Tm +(rely) Tj +1 0 0 1 348.96 166.92 Tm +(on) Tj +1 0 0 1 361.68 166.92 Tm +(the) Tj +1 0 0 1 376.56 166.92 Tm +(operating) Tj +1 0 0 1 416.88 166.92 Tm +(system) Tj +1 0 0 1 447.24 166.92 Tm +(to) Tj +1 0 0 1 457.679 166.92 Tm +(manage) Tj +1 0 0 1 491.399 166.92 Tm +(the) Tj +1 0 0 1 506.279 166.92 Tm +(cache) Tj +1 0 0 1 531.719 166.92 Tm +(of) Tj +1 0 0 1 315 155.04 Tm +(memory-mapped) Tj +1 0 0 1 386.04 155.04 Tm +(\256le) Tj +1 0 0 1 401.76 155.04 Tm +(pages) Tj +1 0 0 1 427.44 155.04 Tm +(and) Tj +1 0 0 1 444.72 155.04 Tm +(to) Tj +1 0 0 1 455.52 155.04 Tm +(write) Tj +1 0 0 1 478.919 155.04 Tm +(modi\256ed) Tj +1 0 0 1 517.319 155.04 Tm +(pages) Tj +1 0 0 1 315 143.04 Tm +(to) Tj +1 0 0 1 327.6 143.04 Tm +(disk) Tj +1 0 0 1 349.08 143.04 Tm +(asynchronously) Tj +1 0 0 1 411.48 143.04 Tm +(.) Tj +1 0 0 1 424.08 143.04 Tm +(The) Tj +1 0 0 1 444.36 143.04 Tm +(current) Tj +1 0 0 1 477.48 143.04 Tm +(implementation) Tj +1 0 0 1 315 131.16 Tm +(uses) Tj +1 0 0 1 337.2 131.16 Tm +(8KB) Tj +1 0 0 1 360.96 131.16 Tm +(blocks) Tj +1 0 0 1 392.04 131.16 Tm +(and) Tj +1 0 0 1 411.48 131.16 Tm +(inodes) Tj +1 0 0 1 442.44 131.16 Tm +(contain) Tj +1 0 0 1 476.88 131.16 Tm +(the) Tj +1 0 0 1 494.039 131.16 Tm +(NFS) Tj +1 0 0 1 517.319 131.16 Tm +(status) Tj +1 0 0 1 315 119.16 Tm +(information) Tj +1 0 0 1 364.2 119.16 Tm +(plus) Tj +1 0 0 1 382.92 119.16 Tm +(256) Tj +1 0 0 1 400.08 119.16 Tm +(bytes) Tj +1 0 0 1 423.24 119.16 Tm +(of) Tj +1 0 0 1 433.68 119.16 Tm +(data,) Tj +1 0 0 1 455.039 119.16 Tm +(which) Tj +1 0 0 1 481.559 119.16 Tm +(is) Tj +1 0 0 1 490.319 119.16 Tm +(used) Tj +1 0 0 1 510.719 119.16 Tm +(to) Tj +1 0 0 1 520.679 119.16 Tm +(store) Tj +1 0 0 1 315 107.28 Tm +(directory) Tj +1 0 0 1 353.16 107.28 Tm +(entries) Tj +1 0 0 1 381.84 107.28 Tm +(in) Tj +1 0 0 1 391.8 107.28 Tm +(directories,) Tj +1 0 0 1 438.599 107.28 Tm +(pointers) Tj +1 0 0 1 472.919 107.28 Tm +(to) Tj +1 0 0 1 482.759 107.28 Tm +(blocks) Tj +1 0 0 1 510.959 107.28 Tm +(in) Tj +1 0 0 1 520.919 107.28 Tm +(\256les,) Tj +1 0 0 1 315 95.28 Tm +(and) Tj +1 0 0 1 331.56 95.28 Tm +(te) Tj +1 0 0 1 338.64 95.28 Tm +(xt) Tj +1 0 0 1 348.6 95.28 Tm +(in) Tj +1 0 0 1 358.56 95.28 Tm +(symbolic) Tj +1 0 0 1 397.32 95.28 Tm +(links.) Tj +1 0 0 1 422.64 95.28 Tm +(Directories) Tj +1 0 0 1 469.199 95.28 Tm +(and) Tj +1 0 0 1 485.759 95.28 Tm +(\256les) Tj +1 0 0 1 504.599 95.28 Tm +(may) Tj +1 0 0 1 523.919 95.28 Tm +(also) Tj +1 0 0 1 315 83.4 Tm +(use) Tj +1 0 0 1 330.72 83.4 Tm +(indirect) Tj +1 0 0 1 363.72 83.4 Tm +(blocks) Tj +1 0 0 1 392.16 83.4 Tm +(in) Tj +1 0 0 1 402.48 83.4 Tm +(a) Tj +1 0 0 1 409.32 83.4 Tm +(w) Tj +1 0 0 1 416.4 83.4 Tm +(ay) Tj +1 0 0 1 428.4 83.4 Tm +(similar) Tj +1 0 0 1 458.519 83.4 Tm +(to) Tj +1 0 0 1 468.719 83.4 Tm +(Unix.) Tj +1 0 0 1 324.96 71.04 Tm +(Our) Tj +1 0 0 1 345.48 71.04 Tm +(implementation) Tj +1 0 0 1 413.04 71.04 Tm +(ensures) Tj +1 0 0 1 447.96 71.04 Tm +(that) Tj +1 0 0 1 468 71.04 Tm +(all) Tj +1 0 0 1 483 71.04 Tm +(state) Tj +1 0 0 1 506.279 71.04 Tm +(machine) Tj +1 0 0 1 303.48 35.16 Tm +(9) Tj +ET +Q +endstream +endobj +699 0 obj +115962 +endobj +700 0 obj +<< +/Type /Page +/MediaBox [0 0 612 792] +/Parent 2 0 R +/Resources << /ProcSet [/PDF /ImageB /Text] +/Font << +/R697 697 0 R +/R694 694 0 R +/R691 691 0 R +>> +>> +/Contents [ +692 0 R +695 0 R +698 0 R +] +>> +endobj +701 0 obj +<< /Type /Font /Name /R701 /Subtype /Type1 /BaseFont /Times-Roman >> +endobj +702 0 obj +<< /Length 703 0 R >> +stream +q +BT +/R701 10 Tf +1 0 0 1 72 709.08 Tm +(replicas) Tj +1 0 0 1 104.64 709.08 Tm +(start) Tj +1 0 0 1 123.36 709.08 Tm +(in) Tj +1 0 0 1 132.72 709.08 Tm +(the) Tj +1 0 0 1 146.52 709.08 Tm +(same) Tj +1 0 0 1 168.6 709.08 Tm +(initial) Tj +1 0 0 1 193.439 709.08 Tm +(state) Tj +1 0 0 1 213.239 709.08 Tm +(and) Tj +1 0 0 1 229.319 709.08 Tm +(are) Tj +1 0 0 1 243.119 709.08 Tm +(determ) Tj +1 0 0 1 270.839 709.08 Tm +(inistic,) Tj +1 0 0 1 72 697.08 Tm +(which) Tj +1 0 0 1 99.7199 697.08 Tm +(are) Tj +1 0 0 1 115.2 697.08 Tm +(necessary) Tj +1 0 0 1 157.32 697.08 Tm +(conditions) Tj +1 0 0 1 202.2 697.08 Tm +(for) Tj +1 0 0 1 217.2 697.08 Tm +(the) Tj +1 0 0 1 232.68 697.08 Tm +(correctness) Tj +1 0 0 1 280.919 697.08 Tm +(of) Tj +1 0 0 1 292.559 697.08 Tm +(a) Tj +1 0 0 1 72 685.2 Tm +(service) Tj +1 0 0 1 104.04 685.2 Tm +(implemented) Tj +1 0 0 1 159.96 685.2 Tm +(using) Tj +1 0 0 1 185.52 685.2 Tm +(our) Tj +1 0 0 1 202.68 685.2 Tm +(protocol.) Tj +1 0 0 1 246.12 685.2 Tm +(The) Tj +1 0 0 1 265.439 685.2 Tm +(primary) Tj +1 0 0 1 72 673.2 Tm +(proposes) Tj +1 0 0 1 112.68 673.2 Tm +(the) Tj +1 0 0 1 130.08 673.2 Tm +(v) Tj +1 0 0 1 134.88 673.2 Tm +(alues) Tj +1 0 0 1 160.56 673.2 Tm +(for) Tj +1 0 0 1 177.48 673.2 Tm +(time-last-modi\256ed) Tj +1 0 0 1 256.319 673.2 Tm +(and) Tj +1 0 0 1 275.999 673.2 Tm +(time-) Tj +1 0 0 1 72 661.32 Tm +(last-accessed,) Tj +1 0 0 1 134.04 661.32 Tm +(and) Tj +1 0 0 1 154.92 661.32 Tm +(replicas) Tj +1 0 0 1 192.48 661.32 Tm +(select) Tj +1 0 0 1 221.64 661.32 Tm +(the) Tj +1 0 0 1 240.359 661.32 Tm +(lar) Tj +1 0 0 1 250.799 661.32 Tm +(ger) Tj +1 0 0 1 269.999 661.32 Tm +(of) Tj +1 0 0 1 284.879 661.32 Tm +(the) Tj +1 0 0 1 72 649.32 Tm +(proposed) Tj +1 0 0 1 111 649.32 Tm +(v) Tj +1 0 0 1 115.8 649.32 Tm +(alue) Tj +1 0 0 1 134.88 649.32 Tm +(and) Tj +1 0 0 1 151.8 649.32 Tm +(one) Tj +1 0 0 1 168.72 649.32 Tm +(greater) Tj +1 0 0 1 198.96 649.32 Tm +(than) Tj +1 0 0 1 218.639 649.32 Tm +(the) Tj +1 0 0 1 233.279 649.32 Tm +(maximum) Tj +1 0 0 1 276.239 649.32 Tm +(of) Tj +1 0 0 1 287.039 649.32 Tm +(all) Tj +1 0 0 1 72 637.44 Tm +(v) Tj +1 0 0 1 76.8 637.44 Tm +(alues) Tj +1 0 0 1 100.8 637.44 Tm +(selected) Tj +1 0 0 1 136.44 637.44 Tm +(for) Tj +1 0 0 1 151.56 637.44 Tm +(earlier) Tj +1 0 0 1 180.48 637.44 Tm +(requests.) Tj +1 0 0 1 222.479 637.44 Tm +(W) Tj +1 0 0 1 230.999 637.44 Tm +(e) Tj +1 0 0 1 239.039 637.44 Tm +(do) Tj +1 0 0 1 252.479 637.44 Tm +(not) Tj +1 0 0 1 268.799 637.44 Tm +(require) Tj +1 0 0 1 72 625.44 Tm +(synchronous) Tj +1 0 0 1 128.4 625.44 Tm +(writes) Tj +1 0 0 1 158.76 625.44 Tm +(to) Tj +1 0 0 1 172.56 625.44 Tm +(implement) Tj +1 0 0 1 221.28 625.44 Tm +(NFS) Tj +1 0 0 1 245.52 625.44 Tm +(V2) Tj +1 0 0 1 263.759 625.44 Tm +(protocol) Tj +1 0 0 1 72 613.44 Tm +(semantics) Tj +1 0 0 1 115.8 613.44 Tm +(because) Tj +1 0 0 1 151.68 613.44 Tm +(BFS) Tj +1 0 0 1 173.88 613.44 Tm +(achie) Tj +1 0 0 1 194.76 613.44 Tm +(v) Tj +1 0 0 1 199.68 613.44 Tm +(es) Tj +1 0 0 1 212.28 613.44 Tm +(stability) Tj +1 0 0 1 248.88 613.44 Tm +(of) Tj +1 0 0 1 261.599 613.44 Tm +(modi\256ed) Tj +1 0 0 1 72 601.56 Tm +(data) Tj +1 0 0 1 91.0799 601.56 Tm +(and) Tj +1 0 0 1 108 601.56 Tm +(meta-data) Tj +1 0 0 1 149.76 601.56 Tm +(through) Tj +1 0 0 1 183.24 601.56 Tm +(replication) Tj +1 0 0 1 228.359 601.56 Tm +([20) Tj +1 0 0 1 241.559 601.56 Tm +(].) Tj +ET +endstream +endobj +703 0 obj +3195 +endobj +704 0 obj +<< /Type /Font /Name /R704 /Subtype /Type1 /BaseFont /Times-Bold >> +endobj +705 0 obj +<< /Length 706 0 R >> +stream +BT +/R704 10 Tf +1 0 0 1 72 577.8 Tm +(6.3) Tj +1 0 0 1 94.4399 577.8 Tm +(Maintaining) Tj +1 0 0 1 149.52 577.8 Tm +(Checkpoints) Tj +/R701 10 Tf +1 0 0 1 72 562.68 Tm +(This) Tj +1 0 0 1 92.7599 562.68 Tm +(section) Tj +1 0 0 1 124.08 562.68 Tm +(describes) Tj +1 0 0 1 164.16 562.68 Tm +(ho) Tj +1 0 0 1 174 562.68 Tm +(w) Tj +ET +endstream +endobj +706 0 obj +314 +endobj +707 0 obj +<< /Type /Font /Name /R707 /Subtype /Type1 /BaseFont /Times-Italic >> +endobj +708 0 obj +<< /Length 709 0 R >> +stream +BT +/R707 10 Tf +1 0 0 1 184.08 562.68 Tm +(snfsd) Tj +/R701 10 Tf +1 0 0 1 207.6 562.68 Tm +(maintains) Tj +1 0 0 1 249.359 562.68 Tm +(checkpoints) Tj +1 0 0 1 72 550.8 Tm +(of) Tj +1 0 0 1 82.2 550.8 Tm +(the) Tj +1 0 0 1 96.3599 550.8 Tm +(\256le) Tj +1 0 0 1 111 550.8 Tm +(system) Tj +1 0 0 1 140.64 550.8 Tm +(state.) Tj +1 0 0 1 164.76 550.8 Tm +(Recall) Tj +1 0 0 1 192.12 550.8 Tm +(that) Tj +1 0 0 1 209.04 550.8 Tm +(each) Tj +1 0 0 1 229.199 550.8 Tm +(replica) Tj +1 0 0 1 258.239 550.8 Tm +(maintains) Tj +1 0 0 1 72 538.8 Tm +(se) Tj +1 0 0 1 80.04 538.8 Tm +(v) Tj +1 0 0 1 84.96 538.8 Tm +(eral) Tj +1 0 0 1 102 538.8 Tm +(logical) Tj +1 0 0 1 131.28 538.8 Tm +(copies) Tj +1 0 0 1 158.88 538.8 Tm +(of) Tj +1 0 0 1 169.44 538.8 Tm +(the) Tj +1 0 0 1 183.72 538.8 Tm +(state:) Tj +1 0 0 1 208.199 538.8 Tm +(the) Tj +1 0 0 1 222.599 538.8 Tm +(current) Tj +1 0 0 1 252.959 538.8 Tm +(state,) Tj +1 0 0 1 275.999 538.8 Tm +(some) Tj +1 0 0 1 72 526.92 Tm +(number) Tj +1 0 0 1 104.52 526.92 Tm +(of) Tj +1 0 0 1 114.84 526.92 Tm +(checkpoints) Tj +1 0 0 1 164.4 526.92 Tm +(that) Tj +1 0 0 1 181.44 526.92 Tm +(are) Tj +1 0 0 1 195.6 526.92 Tm +(not) Tj +1 0 0 1 210.48 526.92 Tm +(yet) Tj +1 0 0 1 224.64 526.92 Tm +(stable,) Tj +1 0 0 1 252.479 526.92 Tm +(and) Tj +1 0 0 1 268.919 526.92 Tm +(the) Tj +1 0 0 1 283.199 526.92 Tm +(last) Tj +1 0 0 1 72 514.92 Tm +(stable) Tj +1 0 0 1 97.6799 514.92 Tm +(checkpoint.) Tj +/R707 10 Tf +1 0 0 1 81.96 502.44 Tm +(snfsd) Tj +/R701 10 Tf +1 0 0 1 106.44 502.44 Tm +(e) Tj +1 0 0 1 110.76 502.44 Tm +(x) Tj +1 0 0 1 115.68 502.44 Tm +(ecutes) Tj +1 0 0 1 144.48 502.44 Tm +(\256le) Tj +1 0 0 1 161.28 502.44 Tm +(system) Tj +1 0 0 1 192.96 502.44 Tm +(operations) Tj +1 0 0 1 238.559 502.44 Tm +(directly) Tj +1 0 0 1 272.999 502.44 Tm +(in) Tj +1 0 0 1 284.879 502.44 Tm +(the) Tj +1 0 0 1 72 490.56 Tm +(memory) Tj +1 0 0 1 107.16 490.56 Tm +(mapped) Tj +1 0 0 1 140.64 490.56 Tm +(\256le) Tj +1 0 0 1 155.04 490.56 Tm +(to) Tj +1 0 0 1 164.4 490.56 Tm +(pre) Tj +1 0 0 1 177.12 490.56 Tm +(serv) Tj +1 0 0 1 193.56 490.56 Tm +(e) Tj +1 0 0 1 199.68 490.56 Tm +(lo) Tj +1 0 0 1 207.36 490.56 Tm +(cality) Tj +1 0 0 1 228.839 490.56 Tm +(,) Tj +1 0 0 1 232.439 490.56 Tm +(and) Tj +1 0 0 1 248.639 490.56 Tm +(it) Tj +1 0 0 1 255.839 490.56 Tm +(uses) Tj +1 0 0 1 274.679 490.56 Tm +(cop) Tj +1 0 0 1 289.079 490.56 Tm +(y-) Tj +1 0 0 1 72 478.56 Tm +(on-write) Tj +1 0 0 1 107.4 478.56 Tm +(to) Tj +1 0 0 1 116.88 478.56 Tm +(reduce) Tj +1 0 0 1 145.08 478.56 Tm +(the) Tj +1 0 0 1 158.88 478.56 Tm +(space) Tj +1 0 0 1 182.76 478.56 Tm +(and) Tj +1 0 0 1 198.72 478.56 Tm +(time) Tj +1 0 0 1 218.159 478.56 Tm +(o) Tj +1 0 0 1 223.079 478.56 Tm +(v) Tj +1 0 0 1 227.999 478.56 Tm +(erhead) Tj +1 0 0 1 256.079 478.56 Tm +(associated) Tj +1 0 0 1 72 466.68 Tm +(with) Tj +1 0 0 1 93.1199 466.68 Tm +(maintaining) Tj +1 0 0 1 144.12 466.68 Tm +(checkpoints.) Tj +/R707 10 Tf +1 0 0 1 200.52 466.68 Tm +(snfsd) Tj +/R701 10 Tf +1 0 0 1 224.4 466.68 Tm +(maintains) Tj +1 0 0 1 266.639 466.68 Tm +(a) Tj +1 0 0 1 274.439 466.68 Tm +(cop) Tj +1 0 0 1 288.839 466.68 Tm +(y-) Tj +1 0 0 1 72 454.68 Tm +(on-write) Tj +1 0 0 1 110.4 454.68 Tm +(bit) Tj +1 0 0 1 125.4 454.68 Tm +(for) Tj +1 0 0 1 141.6 454.68 Tm +(e) Tj +1 0 0 1 145.8 454.68 Tm +(v) Tj +1 0 0 1 150.72 454.68 Tm +(ery) Tj +1 0 0 1 168 454.68 Tm +(512-byte) Tj +1 0 0 1 207.96 454.68 Tm +(block) Tj +1 0 0 1 234.72 454.68 Tm +(in) Tj +1 0 0 1 247.079 454.68 Tm +(the) Tj +1 0 0 1 263.759 454.68 Tm +(memory) Tj +1 0 0 1 72 442.8 Tm +(mapped) Tj +1 0 0 1 108.24 442.8 Tm +(\256le.) Tj +1 0 0 1 133.68 442.8 Tm +(When) Tj +1 0 0 1 162.12 442.8 Tm +(the) Tj +1 0 0 1 179.04 442.8 Tm +(replication) Tj +1 0 0 1 226.32 442.8 Tm +(code) Tj +1 0 0 1 249.84 442.8 Tm +(in) Tj +1 0 0 1 257.279 442.8 Tm +(v) Tj +1 0 0 1 262.079 442.8 Tm +(ok) Tj +1 0 0 1 272.039 442.8 Tm +(es) Tj +1 0 0 1 284.879 442.8 Tm +(the) Tj +/R707 10 Tf +1 0 0 1 72 430.8 Tm +(mak) Tj +1 0 0 1 88.56 430.8 Tm +(e) Tj +ET +Q +q +W +0 0 612 792 re +n +q +3 0 0 -0.48 93.588 431.292 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +/R707 10 Tf +1 0 0 1 96.4799 430.8 Tm +(c) Tj +1 0 0 1 100.8 430.8 Tm +(hec) Tj +1 0 0 1 114.48 430.8 Tm +(kpoint) Tj +/R701 10 Tf +1 0 0 1 142.08 430.8 Tm +(upcall,) Tj +/R707 10 Tf +1 0 0 1 171.84 430.8 Tm +(snfsd) Tj +/R701 10 Tf +1 0 0 1 195.12 430.8 Tm +(sets) Tj +1 0 0 1 212.88 430.8 Tm +(all) Tj +1 0 0 1 225.719 430.8 Tm +(the) Tj +1 0 0 1 240.719 430.8 Tm +(cop) Tj +1 0 0 1 255.119 430.8 Tm +(y-on-write) Tj +1 0 0 1 72 418.92 Tm +(bits) Tj +1 0 0 1 88.4399 418.92 Tm +(and) Tj +1 0 0 1 104.76 418.92 Tm +(creates) Tj +1 0 0 1 134.4 418.92 Tm +(a) Tj +1 0 0 1 140.88 418.92 Tm +(\(v) Tj +1 0 0 1 149.04 418.92 Tm +(olatile\)) Tj +1 0 0 1 179.16 418.92 Tm +(checkpoint) Tj +1 0 0 1 224.88 418.92 Tm +(record,) Tj +1 0 0 1 254.879 418.92 Tm +(containing) Tj +1 0 0 1 72 406.92 Tm +(the) Tj +1 0 0 1 88.5599 406.92 Tm +(current) Tj +1 0 0 1 121.2 406.92 Tm +(sequence) Tj +1 0 0 1 162.12 406.92 Tm +(number) Tj +1 0 0 1 192.48 406.92 Tm +(,) Tj +1 0 0 1 199.56 406.92 Tm +(which) Tj +1 0 0 1 228.24 406.92 Tm +(it) Tj +1 0 0 1 238.2 406.92 Tm +(recei) Tj +1 0 0 1 257.399 406.92 Tm +(v) Tj +1 0 0 1 262.319 406.92 Tm +(es) Tj +1 0 0 1 274.919 406.92 Tm +(as) Tj +1 0 0 1 287.639 406.92 Tm +(an) Tj +1 0 0 1 72 395.04 Tm +(ar) Tj +1 0 0 1 79.68 395.04 Tm +(gument) Tj +1 0 0 1 113.28 395.04 Tm +(to) Tj +1 0 0 1 124.92 395.04 Tm +(the) Tj +1 0 0 1 141 395.04 Tm +(upcall,) Tj +1 0 0 1 172.08 395.04 Tm +(and) Tj +1 0 0 1 190.32 395.04 Tm +(a) Tj +1 0 0 1 198.72 395.04 Tm +(list) Tj +1 0 0 1 214.8 395.04 Tm +(of) Tj +1 0 0 1 226.92 395.04 Tm +(blocks.) Tj +1 0 0 1 263.159 395.04 Tm +(This) Tj +1 0 0 1 284.759 395.04 Tm +(list) Tj +1 0 0 1 72 383.04 Tm +(contains) Tj +1 0 0 1 109.92 383.04 Tm +(the) Tj +1 0 0 1 126.84 383.04 Tm +(copies) Tj +1 0 0 1 157.08 383.04 Tm +(of) Tj +1 0 0 1 170.16 383.04 Tm +(the) Tj +1 0 0 1 187.08 383.04 Tm +(blocks) Tj +1 0 0 1 217.799 383.04 Tm +(that) Tj +1 0 0 1 237.479 383.04 Tm +(were) Tj +1 0 0 1 261.599 383.04 Tm +(modi\256ed) Tj +1 0 0 1 72 371.16 Tm +(since) Tj +1 0 0 1 97.7999 371.16 Tm +(the) Tj +1 0 0 1 115.32 371.16 Tm +(checkpoint) Tj +1 0 0 1 164.4 371.16 Tm +(w) Tj +1 0 0 1 171.48 371.16 Tm +(as) Tj +1 0 0 1 185.16 371.16 Tm +(tak) Tj +1 0 0 1 197.28 371.16 Tm +(en,) Tj +1 0 0 1 215.16 371.16 Tm +(and) Tj +1 0 0 1 234.959 371.16 Tm +(therefore,) Tj +1 0 0 1 279.479 371.16 Tm +(it) Tj +1 0 0 1 290.399 371.16 Tm +(is) Tj +1 0 0 1 72 359.16 Tm +(initially) Tj +1 0 0 1 106.68 359.16 Tm +(empty) Tj +1 0 0 1 131.16 359.16 Tm +(.) Tj +1 0 0 1 140.52 359.16 Tm +(The) Tj +1 0 0 1 159.72 359.16 Tm +(record) Tj +1 0 0 1 188.76 359.16 Tm +(also) Tj +1 0 0 1 208.56 359.16 Tm +(contains) Tj +1 0 0 1 245.399 359.16 Tm +(the) Tj +1 0 0 1 261.239 359.16 Tm +(digest) Tj +1 0 0 1 288.719 359.16 Tm +(of) Tj +1 0 0 1 72 347.16 Tm +(the) Tj +1 0 0 1 86.8799 347.16 Tm +(current) Tj +1 0 0 1 117.72 347.16 Tm +(state;) Tj +1 0 0 1 141.6 347.16 Tm +(we) Tj +1 0 0 1 155.88 347.16 Tm +(discuss) Tj +1 0 0 1 187.44 347.16 Tm +(ho) Tj +1 0 0 1 197.28 347.16 Tm +(w) Tj +1 0 0 1 207 347.16 Tm +(the) Tj +1 0 0 1 221.879 347.16 Tm +(digest) Tj +1 0 0 1 248.399 347.16 Tm +(is) Tj +1 0 0 1 257.759 347.16 Tm +(computed) Tj +1 0 0 1 72 335.28 Tm +(in) Tj +1 0 0 1 82.2 335.28 Tm +(Section) Tj +1 0 0 1 114.6 335.28 Tm +(6.4.) Tj +1 0 0 1 81.96 322.8 Tm +(When) Tj +1 0 0 1 108.48 322.8 Tm +(a) Tj +1 0 0 1 115.68 322.8 Tm +(block) Tj +1 0 0 1 140.52 322.8 Tm +(of) Tj +1 0 0 1 151.56 322.8 Tm +(the) Tj +1 0 0 1 166.44 322.8 Tm +(memory) Tj +1 0 0 1 202.44 322.8 Tm +(mapped) Tj +1 0 0 1 236.76 322.8 Tm +(\256le) Tj +1 0 0 1 252.24 322.8 Tm +(is) Tj +1 0 0 1 261.599 322.8 Tm +(modi\256ed) Tj +1 0 0 1 72 310.92 Tm +(while) Tj +1 0 0 1 97.1999 310.92 Tm +(e) Tj +1 0 0 1 101.52 310.92 Tm +(x) Tj +1 0 0 1 106.44 310.92 Tm +(ecuting) Tj +1 0 0 1 138.6 310.92 Tm +(a) Tj +1 0 0 1 146.04 310.92 Tm +(client) Tj +1 0 0 1 171.24 310.92 Tm +(request,) Tj +/R707 10 Tf +1 0 0 1 205.559 310.92 Tm +(snfsd) Tj +/R701 10 Tf +1 0 0 1 229.079 310.92 Tm +(checks) Tj +1 0 0 1 259.199 310.92 Tm +(the) Tj +1 0 0 1 274.439 310.92 Tm +(cop) Tj +1 0 0 1 288.839 310.92 Tm +(y-) Tj +1 0 0 1 72 298.92 Tm +(on-write) Tj +1 0 0 1 107.52 298.92 Tm +(bit) Tj +1 0 0 1 119.88 298.92 Tm +(for) Tj +1 0 0 1 133.32 298.92 Tm +(the) Tj +1 0 0 1 147.24 298.92 Tm +(block) Tj +1 0 0 1 171.24 298.92 Tm +(and,) Tj +1 0 0 1 190.08 298.92 Tm +(if) Tj +1 0 0 1 197.879 298.92 Tm +(it) Tj +1 0 0 1 205.199 298.92 Tm +(is) Tj +1 0 0 1 213.719 298.92 Tm +(set,) Tj +1 0 0 1 229.199 298.92 Tm +(stores) Tj +1 0 0 1 254.279 298.92 Tm +(the) Tj +1 0 0 1 268.199 298.92 Tm +(block') Tj +1 0 0 1 293.279 298.92 Tm +(s) Tj +1 0 0 1 72 287.04 Tm +(current) Tj +1 0 0 1 102 287.04 Tm +(contents) Tj +1 0 0 1 137.04 287.04 Tm +(and) Tj +1 0 0 1 153.36 287.04 Tm +(its) Tj +1 0 0 1 164.52 287.04 Tm +(identi\256er) Tj +1 0 0 1 202.439 287.04 Tm +(in) Tj +1 0 0 1 212.039 287.04 Tm +(the) Tj +1 0 0 1 225.959 287.04 Tm +(checkpoint) Tj +1 0 0 1 271.559 287.04 Tm +(record) Tj +1 0 0 1 72 275.04 Tm +(for) Tj +1 0 0 1 87.5999 275.04 Tm +(the) Tj +1 0 0 1 103.8 275.04 Tm +(last) Tj +1 0 0 1 121.68 275.04 Tm +(checkpoint.) Tj +1 0 0 1 176.04 275.04 Tm +(Then,) Tj +1 0 0 1 203.4 275.04 Tm +(it) Tj +1 0 0 1 212.879 275.04 Tm +(o) Tj +1 0 0 1 217.799 275.04 Tm +(v) Tj +1 0 0 1 222.719 275.04 Tm +(erwrites) Tj +1 0 0 1 258.719 275.04 Tm +(the) Tj +1 0 0 1 274.919 275.04 Tm +(block) Tj +1 0 0 1 72 263.16 Tm +(with) Tj +1 0 0 1 95.2799 263.16 Tm +(its) Tj +1 0 0 1 110.16 263.16 Tm +(ne) Tj +1 0 0 1 119.4 263.16 Tm +(w) Tj +1 0 0 1 132.12 263.16 Tm +(v) Tj +1 0 0 1 136.92 263.16 Tm +(alue) Tj +1 0 0 1 159 263.16 Tm +(and) Tj +1 0 0 1 178.92 263.16 Tm +(resets) Tj +1 0 0 1 207.12 263.16 Tm +(its) Tj +1 0 0 1 222.12 263.16 Tm +(cop) Tj +1 0 0 1 236.519 263.16 Tm +(y-on-write) Tj +1 0 0 1 284.039 263.16 Tm +(bit.) Tj +/R707 10 Tf +1 0 0 1 72 251.16 Tm +(snfsd) Tj +/R701 10 Tf +1 0 0 1 96.7199 251.16 Tm +(retains) Tj +1 0 0 1 127.56 251.16 Tm +(a) Tj +1 0 0 1 136.2 251.16 Tm +(checkpoint) Tj +1 0 0 1 184.2 251.16 Tm +(record) Tj +1 0 0 1 213.959 251.16 Tm +(until) Tj +1 0 0 1 236.399 251.16 Tm +(told) Tj +1 0 0 1 256.199 251.16 Tm +(to) Tj +1 0 0 1 268.199 251.16 Tm +(discard) Tj +1 0 0 1 72 239.28 Tm +(it) Tj +1 0 0 1 81 239.28 Tm +(via) Tj +1 0 0 1 96.5999 239.28 Tm +(a) Tj +/R707 10 Tf +1 0 0 1 104.4 239.28 Tm +(delete) Tj +ET +q +3 0 0 -0.48 128.868 239.772 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +1 0 0 1 131.76 239.28 Tm +(c) Tj +1 0 0 1 136.08 239.28 Tm +(hec) Tj +1 0 0 1 149.76 239.28 Tm +(kpoint) Tj +/R701 10 Tf +1 0 0 1 178.08 239.28 Tm +(upcall,) Tj +1 0 0 1 208.56 239.28 Tm +(which) Tj +1 0 0 1 236.399 239.28 Tm +(is) Tj +1 0 0 1 246.359 239.28 Tm +(made) Tj +1 0 0 1 271.439 239.28 Tm +(by) Tj +1 0 0 1 284.879 239.28 Tm +(the) Tj +1 0 0 1 72 227.28 Tm +(replication) Tj +1 0 0 1 116.88 227.28 Tm +(code) Tj +1 0 0 1 137.88 227.28 Tm +(when) Tj +1 0 0 1 161.64 227.28 Tm +(a) Tj +1 0 0 1 168.36 227.28 Tm +(later) Tj +1 0 0 1 188.28 227.28 Tm +(checkpoint) Tj +1 0 0 1 234.24 227.28 Tm +(becomes) Tj +1 0 0 1 271.319 227.28 Tm +(stable.) Tj +1 0 0 1 81.96 214.8 Tm +(If) Tj +1 0 0 1 92.1599 214.8 Tm +(the) Tj +1 0 0 1 107.88 214.8 Tm +(replication) Tj +1 0 0 1 154.08 214.8 Tm +(code) Tj +1 0 0 1 176.52 214.8 Tm +(requires) Tj +1 0 0 1 212.16 214.8 Tm +(a) Tj +1 0 0 1 220.08 214.8 Tm +(checkpoint) Tj +1 0 0 1 267.359 214.8 Tm +(to) Tj +1 0 0 1 278.759 214.8 Tm +(send) Tj +1 0 0 1 72 202.92 Tm +(to) Tj +1 0 0 1 82.8 202.92 Tm +(another) Tj +1 0 0 1 115.68 202.92 Tm +(replica,) Tj +1 0 0 1 148.32 202.92 Tm +(it) Tj +1 0 0 1 156.84 202.92 Tm +(calls) Tj +1 0 0 1 178.08 202.92 Tm +(the) Tj +/R707 10 Tf +1 0 0 1 193.32 202.92 Tm +(g) Tj +1 0 0 1 198.239 202.92 Tm +(et) Tj +ET +q +3 0 0 -0.48 206.028 203.412 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +1 0 0 1 208.919 202.92 Tm +(c) Tj +1 0 0 1 213.239 202.92 Tm +(hec) Tj +1 0 0 1 226.919 202.92 Tm +(kpoint) Tj +/R701 10 Tf +1 0 0 1 254.759 202.92 Tm +(upcall.) Tj +1 0 0 1 286.679 202.92 Tm +(T) Tj +1 0 0 1 291.959 202.92 Tm +(o) Tj +1 0 0 1 72 190.92 Tm +(obtain) Tj +1 0 0 1 100.08 190.92 Tm +(the) Tj +1 0 0 1 115.56 190.92 Tm +(v) Tj +1 0 0 1 120.36 190.92 Tm +(alue) Tj +1 0 0 1 140.04 190.92 Tm +(for) Tj +1 0 0 1 154.92 190.92 Tm +(a) Tj +1 0 0 1 162.6 190.92 Tm +(block,) Tj +/R707 10 Tf +1 0 0 1 190.56 190.92 Tm +(snfsd) Tj +/R701 10 Tf +1 0 0 1 214.319 190.92 Tm +(\256rst) Tj +1 0 0 1 233.039 190.92 Tm +(searches) Tj +1 0 0 1 269.999 190.92 Tm +(for) Tj +1 0 0 1 284.879 190.92 Tm +(the) Tj +1 0 0 1 72 179.04 Tm +(block) Tj +1 0 0 1 97.0799 179.04 Tm +(in) Tj +1 0 0 1 107.88 179.04 Tm +(the) Tj +1 0 0 1 123 179.04 Tm +(checkpoint) Tj +1 0 0 1 169.68 179.04 Tm +(record) Tj +1 0 0 1 198.12 179.04 Tm +(of) Tj +1 0 0 1 209.4 179.04 Tm +(the) Tj +1 0 0 1 224.519 179.04 Tm +(stable) Tj +1 0 0 1 250.799 179.04 Tm +(checkpoint,) Tj +1 0 0 1 72 167.04 Tm +(and) Tj +1 0 0 1 90.7199 167.04 Tm +(then) Tj +1 0 0 1 112.08 167.04 Tm +(searches) Tj +1 0 0 1 150.12 167.04 Tm +(the) Tj +1 0 0 1 166.56 167.04 Tm +(checkpoint) Tj +1 0 0 1 214.56 167.04 Tm +(records) Tj +1 0 0 1 248.159 167.04 Tm +(of) Tj +1 0 0 1 260.759 167.04 Tm +(an) Tj +1 0 0 1 270.119 167.04 Tm +(y) Tj +1 0 0 1 279.239 167.04 Tm +(later) Tj +1 0 0 1 72 155.16 Tm +(checkpoints.) Tj +1 0 0 1 125.52 155.16 Tm +(If) Tj +1 0 0 1 134.4 155.16 Tm +(the) Tj +1 0 0 1 148.8 155.16 Tm +(block) Tj +1 0 0 1 173.04 155.16 Tm +(is) Tj +1 0 0 1 181.92 155.16 Tm +(not) Tj +1 0 0 1 196.8 155.16 Tm +(in) Tj +1 0 0 1 206.76 155.16 Tm +(an) Tj +1 0 0 1 216.12 155.16 Tm +(y) Tj +1 0 0 1 223.199 155.16 Tm +(checkpoint) Tj +1 0 0 1 269.039 155.16 Tm +(record,) Tj +1 0 0 1 72 143.16 Tm +(it) Tj +1 0 0 1 80.04 143.16 Tm +(returns) Tj +1 0 0 1 110.16 143.16 Tm +(the) Tj +1 0 0 1 124.8 143.16 Tm +(v) Tj +1 0 0 1 129.6 143.16 Tm +(alue) Tj +1 0 0 1 148.68 143.16 Tm +(from) Tj +1 0 0 1 170.52 143.16 Tm +(the) Tj +1 0 0 1 185.16 143.16 Tm +(current) Tj +1 0 0 1 215.879 143.16 Tm +(state.) Tj +1 0 0 1 81.96 130.68 Tm +(The) Tj +1 0 0 1 101.16 130.68 Tm +(use) Tj +1 0 0 1 118.08 130.68 Tm +(of) Tj +1 0 0 1 130.08 130.68 Tm +(the) Tj +1 0 0 1 145.92 130.68 Tm +(cop) Tj +1 0 0 1 160.32 130.68 Tm +(y-on-write) Tj +1 0 0 1 205.92 130.68 Tm +(technique) Tj +1 0 0 1 248.279 130.68 Tm +(and) Tj +1 0 0 1 266.279 130.68 Tm +(the) Tj +1 0 0 1 282.119 130.68 Tm +(f) Tj +1 0 0 1 285.359 130.68 Tm +(act) Tj +1 0 0 1 72 118.8 Tm +(that) Tj +1 0 0 1 89.5199 118.8 Tm +(we) Tj +1 0 0 1 103.68 118.8 Tm +(k) Tj +1 0 0 1 108.6 118.8 Tm +(eep) Tj +1 0 0 1 124.92 118.8 Tm +(at) Tj +1 0 0 1 134.64 118.8 Tm +(most) Tj +1 0 0 1 156.48 118.8 Tm +(2) Tj +1 0 0 1 164.04 118.8 Tm +(checkpoints) Tj +1 0 0 1 214.2 118.8 Tm +(ensure) Tj +1 0 0 1 242.639 118.8 Tm +(that) Tj +1 0 0 1 260.159 118.8 Tm +(the) Tj +1 0 0 1 274.919 118.8 Tm +(space) Tj +1 0 0 1 72 106.8 Tm +(and) Tj +1 0 0 1 91.08 106.8 Tm +(time) Tj +1 0 0 1 113.4 106.8 Tm +(o) Tj +1 0 0 1 118.32 106.8 Tm +(v) Tj +1 0 0 1 123.24 106.8 Tm +(erheads) Tj +1 0 0 1 158.16 106.8 Tm +(of) Tj +1 0 0 1 171.12 106.8 Tm +(k) Tj +1 0 0 1 176.04 106.8 Tm +(eeping) Tj +1 0 0 1 207.24 106.8 Tm +(se) Tj +1 0 0 1 215.28 106.8 Tm +(v) Tj +1 0 0 1 220.2 106.8 Tm +(eral) Tj +1 0 0 1 239.76 106.8 Tm +(logical) Tj +1 0 0 1 271.559 106.8 Tm +(copies) Tj +1 0 0 1 72 94.92 Tm +(of) Tj +1 0 0 1 85.68 94.92 Tm +(the) Tj +1 0 0 1 103.2 94.92 Tm +(state) Tj +1 0 0 1 126.84 94.92 Tm +(are) Tj +1 0 0 1 144.36 94.92 Tm +(lo) Tj +1 0 0 1 151.92 94.92 Tm +(w) Tj +1 0 0 1 158.52 94.92 Tm +(.) Tj +1 0 0 1 173.16 94.92 Tm +(F) Tj +1 0 0 1 178.56 94.92 Tm +(or) Tj +1 0 0 1 192.24 94.92 Tm +(e) Tj +1 0 0 1 196.56 94.92 Tm +(xample,) Tj +1 0 0 1 234.48 94.92 Tm +(in) Tj +1 0 0 1 247.559 94.92 Tm +(the) Tj +1 0 0 1 265.199 94.92 Tm +(Andre) Tj +1 0 0 1 290.039 94.92 Tm +(w) Tj +1 0 0 1 72 82.92 Tm +(benchmark) Tj +1 0 0 1 122.28 82.92 Tm +(e) Tj +1 0 0 1 126.6 82.92 Tm +(xperiments) Tj +1 0 0 1 176.76 82.92 Tm +(described) Tj +1 0 0 1 220.92 82.92 Tm +(in) Tj +1 0 0 1 234.6 82.92 Tm +(Section) Tj +1 0 0 1 270.479 82.92 Tm +(7,) Tj +1 0 0 1 284.879 82.92 Tm +(the) Tj +1 0 0 1 72 71.04 Tm +(a) Tj +1 0 0 1 76.2 71.04 Tm +(v) Tj +1 0 0 1 81.1199 71.04 Tm +(erage) Tj +1 0 0 1 105.12 71.04 Tm +(checkpoint) Tj +1 0 0 1 151.32 71.04 Tm +(record) Tj +1 0 0 1 179.16 71.04 Tm +(size) Tj +1 0 0 1 197.16 71.04 Tm +(is) Tj +1 0 0 1 206.28 71.04 Tm +(only) Tj +1 0 0 1 226.439 71.04 Tm +(182) Tj +1 0 0 1 243.959 71.04 Tm +(blocks) Tj +1 0 0 1 272.399 71.04 Tm +(with) Tj +1 0 0 1 292.559 71.04 Tm +(a) Tj +1 0 0 1 315 709.08 Tm +(maximum) Tj +1 0 0 1 357.84 709.08 Tm +(of) Tj +1 0 0 1 368.64 709.08 Tm +(500.) Tj +/R704 10 Tf +1 0 0 1 315 687.96 Tm +(6.4) Tj +1 0 0 1 337.44 687.96 Tm +(Computing) Tj +1 0 0 1 388.08 687.96 Tm +(Checkpoint) Tj +1 0 0 1 439.8 687.96 Tm +(Digests) Tj +/R707 10 Tf +1 0 0 1 315 673.68 Tm +(snfsd) Tj +/R701 10 Tf +1 0 0 1 339.84 673.68 Tm +(computes) Tj +1 0 0 1 382.44 673.68 Tm +(a) Tj +1 0 0 1 391.32 673.68 Tm +(digest) Tj +1 0 0 1 419.52 673.68 Tm +(of) Tj +1 0 0 1 432.24 673.68 Tm +(a) Tj +1 0 0 1 441 673.68 Tm +(checkpoint) Tj +1 0 0 1 489.119 673.68 Tm +(state) Tj +1 0 0 1 511.799 673.68 Tm +(as) Tj +1 0 0 1 524.519 673.68 Tm +(part) Tj +1 0 0 1 315 661.8 Tm +(of) Tj +1 0 0 1 328.56 661.8 Tm +(a) Tj +/R707 10 Tf +1 0 0 1 338.28 661.8 Tm +(mak) Tj +1 0 0 1 354.84 661.8 Tm +(e) Tj +ET +q +3 0 0 -0.48 359.868 662.292 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +1 0 0 1 362.76 661.8 Tm +(c) Tj +1 0 0 1 367.08 661.8 Tm +(hec) Tj +1 0 0 1 380.76 661.8 Tm +(kpoint) Tj +/R701 10 Tf +1 0 0 1 410.76 661.8 Tm +(upcall.) Tj +1 0 0 1 449.52 661.8 Tm +(Although) Tj +1 0 0 1 492.36 661.8 Tm +(checkpoints) Tj +1 0 0 1 315 649.8 Tm +(are) Tj +1 0 0 1 330.72 649.8 Tm +(only) Tj +1 0 0 1 352.08 649.8 Tm +(tak) Tj +1 0 0 1 364.2 649.8 Tm +(en) Tj +1 0 0 1 377.16 649.8 Tm +(occasionally) Tj +1 0 0 1 426.6 649.8 Tm +(,) Tj +1 0 0 1 432.6 649.8 Tm +(it) Tj +1 0 0 1 441.72 649.8 Tm +(is) Tj +1 0 0 1 452.039 649.8 Tm +(important) Tj +1 0 0 1 494.399 649.8 Tm +(to) Tj +1 0 0 1 505.679 649.8 Tm +(compute) Tj +1 0 0 1 315 637.92 Tm +(the) Tj +1 0 0 1 330.6 637.92 Tm +(state) Tj +1 0 0 1 352.32 637.92 Tm +(digest) Tj +1 0 0 1 379.56 637.92 Tm +(incrementally) Tj +1 0 0 1 437.76 637.92 Tm +(because) Tj +1 0 0 1 472.679 637.92 Tm +(the) Tj +1 0 0 1 488.279 637.92 Tm +(state) Tj +1 0 0 1 509.999 637.92 Tm +(may) Tj +1 0 0 1 530.639 637.92 Tm +(be) Tj +1 0 0 1 315 625.92 Tm +(lar) Tj +1 0 0 1 325.44 625.92 Tm +(ge.) Tj +/R707 10 Tf +1 0 0 1 342.36 625.92 Tm +(snfsd) Tj +/R701 10 Tf +1 0 0 1 365.88 625.92 Tm +(uses) Tj +1 0 0 1 386.04 625.92 Tm +(an) Tj +1 0 0 1 398.4 625.92 Tm +(incremental) Tj +1 0 0 1 448.44 625.92 Tm +(collision-resistant) Tj +1 0 0 1 522.239 625.92 Tm +(one-) Tj +1 0 0 1 315 614.04 Tm +(w) Tj +1 0 0 1 322.08 614.04 Tm +(ay) Tj +1 0 0 1 335.88 614.04 Tm +(hash) Tj +1 0 0 1 358.56 614.04 Tm +(function) Tj +1 0 0 1 396.12 614.04 Tm +(called) Tj +1 0 0 1 424.32 614.04 Tm +(AdHash) Tj +1 0 0 1 461.28 614.04 Tm +([1) Tj +1 0 0 1 469.56 614.04 Tm +(].) Tj +1 0 0 1 484.68 614.04 Tm +(This) Tj +1 0 0 1 506.76 614.04 Tm +(function) Tj +1 0 0 1 315 602.04 Tm +(di) Tj +1 0 0 1 322.56 602.04 Tm +(vides) Tj +1 0 0 1 347.76 602.04 Tm +(the) Tj +1 0 0 1 364.08 602.04 Tm +(state) Tj +1 0 0 1 386.52 602.04 Tm +(into) Tj +1 0 0 1 406.2 602.04 Tm +(\256x) Tj +1 0 0 1 416.64 602.04 Tm +(ed-size) Tj +1 0 0 1 448.92 602.04 Tm +(blocks) Tj +1 0 0 1 479.039 602.04 Tm +(and) Tj +1 0 0 1 497.639 602.04 Tm +(uses) Tj +1 0 0 1 518.999 602.04 Tm +(some) Tj +1 0 0 1 315 590.16 Tm +(other) Tj +1 0 0 1 339.24 590.16 Tm +(hash) Tj +1 0 0 1 361.32 590.16 Tm +(function) Tj +1 0 0 1 398.28 590.16 Tm +(\(e.g.,) Tj +1 0 0 1 422.52 590.16 Tm +(MD5\)) Tj +1 0 0 1 450.6 590.16 Tm +(to) Tj +1 0 0 1 462.12 590.16 Tm +(compute) Tj +1 0 0 1 500.279 590.16 Tm +(the) Tj +1 0 0 1 516.239 590.16 Tm +(digest) Tj +1 0 0 1 315 578.16 Tm +(of) Tj +1 0 0 1 326.52 578.16 Tm +(the) Tj +1 0 0 1 342 578.16 Tm +(string) Tj +1 0 0 1 367.92 578.16 Tm +(obtained) Tj +1 0 0 1 405.48 578.16 Tm +(by) Tj +1 0 0 1 418.68 578.16 Tm +(concatenating) Tj +1 0 0 1 477.24 578.16 Tm +(the) Tj +1 0 0 1 492.599 578.16 Tm +(block) Tj +1 0 0 1 518.039 578.16 Tm +(inde) Tj +1 0 0 1 535.199 578.16 Tm +(x) Tj +1 0 0 1 315 566.28 Tm +(with) Tj +1 0 0 1 336.12 566.28 Tm +(the) Tj +1 0 0 1 351.72 566.28 Tm +(block) Tj +1 0 0 1 377.28 566.28 Tm +(v) Tj +1 0 0 1 382.08 566.28 Tm +(alue) Tj +1 0 0 1 402.12 566.28 Tm +(for) Tj +1 0 0 1 417.12 566.28 Tm +(each) Tj +1 0 0 1 438.84 566.28 Tm +(block.) Tj +1 0 0 1 469.92 566.28 Tm +(The) Tj +1 0 0 1 488.879 566.28 Tm +(digest) Tj +1 0 0 1 516.119 566.28 Tm +(of) Tj +1 0 0 1 527.879 566.28 Tm +(the) Tj +1 0 0 1 315 554.28 Tm +(state) Tj +1 0 0 1 335.52 554.28 Tm +(is) Tj +1 0 0 1 344.52 554.28 Tm +(the) Tj +1 0 0 1 358.92 554.28 Tm +(sum) Tj +1 0 0 1 377.88 554.28 Tm +(of) Tj +1 0 0 1 388.44 554.28 Tm +(the) Tj +1 0 0 1 402.96 554.28 Tm +(digests) Tj +1 0 0 1 432.839 554.28 Tm +(of) Tj +1 0 0 1 443.519 554.28 Tm +(the) Tj +1 0 0 1 457.919 554.28 Tm +(blocks) Tj +1 0 0 1 486.239 554.28 Tm +(modulo) Tj +1 0 0 1 518.999 554.28 Tm +(some) Tj +1 0 0 1 315 542.28 Tm +(lar) Tj +1 0 0 1 325.44 542.28 Tm +(ge) Tj +1 0 0 1 337.56 542.28 Tm +(inte) Tj +1 0 0 1 352.44 542.28 Tm +(ger) Tj +1 0 0 1 364.68 542.28 Tm +(.) Tj +1 0 0 1 371.4 542.28 Tm +(In) Tj +1 0 0 1 382.44 542.28 Tm +(our) Tj +1 0 0 1 398.52 542.28 Tm +(current) Tj +1 0 0 1 429.479 542.28 Tm +(implementation,) Tj +1 0 0 1 497.399 542.28 Tm +(we) Tj +1 0 0 1 511.799 542.28 Tm +(use) Tj +1 0 0 1 527.879 542.28 Tm +(the) Tj +1 0 0 1 315 530.4 Tm +(512-byte) Tj +1 0 0 1 354.12 530.4 Tm +(blocks) Tj +1 0 0 1 384 530.4 Tm +(from) Tj +1 0 0 1 407.04 530.4 Tm +(the) Tj +1 0 0 1 423 530.4 Tm +(cop) Tj +1 0 0 1 437.4 530.4 Tm +(y-on-write) Tj +1 0 0 1 483.119 530.4 Tm +(technique) Tj +1 0 0 1 525.599 530.4 Tm +(and) Tj +1 0 0 1 315 518.4 Tm +(compute) Tj +1 0 0 1 351.84 518.4 Tm +(their) Tj +1 0 0 1 372.6 518.4 Tm +(digest) Tj +1 0 0 1 398.88 518.4 Tm +(using) Tj +1 0 0 1 422.88 518.4 Tm +(MD5.) Tj +1 0 0 1 324.96 506.52 Tm +(T) Tj +1 0 0 1 330.24 506.52 Tm +(o) Tj +1 0 0 1 337.08 506.52 Tm +(compute) Tj +1 0 0 1 373.32 506.52 Tm +(the) Tj +1 0 0 1 387.36 506.52 Tm +(digest) Tj +1 0 0 1 413.16 506.52 Tm +(for) Tj +1 0 0 1 426.6 506.52 Tm +(the) Tj +1 0 0 1 440.76 506.52 Tm +(state) Tj +1 0 0 1 460.919 506.52 Tm +(incrementally) Tj +1 0 0 1 515.399 506.52 Tm +(,) Tj +/R707 10 Tf +1 0 0 1 519.479 506.52 Tm +(snfsd) Tj +/R701 10 Tf +1 0 0 1 315 494.52 Tm +(maintains) Tj +1 0 0 1 357.72 494.52 Tm +(a) Tj +1 0 0 1 366 494.52 Tm +(table) Tj +1 0 0 1 389.4 494.52 Tm +(with) Tj +1 0 0 1 411 494.52 Tm +(a) Tj +1 0 0 1 419.4 494.52 Tm +(hash) Tj +1 0 0 1 441.6 494.52 Tm +(v) Tj +1 0 0 1 446.4 494.52 Tm +(alue) Tj +1 0 0 1 466.8 494.52 Tm +(for) Tj +1 0 0 1 482.4 494.52 Tm +(each) Tj +1 0 0 1 504.599 494.52 Tm +(512-byte) Tj +1 0 0 1 315 482.64 Tm +(block.) Tj +1 0 0 1 347.76 482.64 Tm +(This) Tj +1 0 0 1 369.48 482.64 Tm +(hash) Tj +1 0 0 1 391.8 482.64 Tm +(v) Tj +1 0 0 1 396.6 482.64 Tm +(alue) Tj +1 0 0 1 417.12 482.64 Tm +(is) Tj +1 0 0 1 427.8 482.64 Tm +(obtained) Tj +1 0 0 1 466.08 482.64 Tm +(by) Tj +1 0 0 1 480.12 482.64 Tm +(applying) Tj +1 0 0 1 518.999 482.64 Tm +(MD5) Tj +1 0 0 1 315 470.64 Tm +(to) Tj +1 0 0 1 325.92 470.64 Tm +(the) Tj +1 0 0 1 341.28 470.64 Tm +(block) Tj +1 0 0 1 366.48 470.64 Tm +(inde) Tj +1 0 0 1 383.64 470.64 Tm +(x) Tj +1 0 0 1 391.68 470.64 Tm +(concatenated) Tj +1 0 0 1 446.88 470.64 Tm +(with) Tj +1 0 0 1 467.64 470.64 Tm +(the) Tj +1 0 0 1 482.999 470.64 Tm +(block) Tj +1 0 0 1 508.319 470.64 Tm +(v) Tj +1 0 0 1 513.119 470.64 Tm +(alue) Tj +1 0 0 1 532.799 470.64 Tm +(at) Tj +1 0 0 1 315 458.76 Tm +(the) Tj +1 0 0 1 330.24 458.76 Tm +(time) Tj +1 0 0 1 351 458.76 Tm +(of) Tj +1 0 0 1 362.28 458.76 Tm +(the) Tj +1 0 0 1 377.52 458.76 Tm +(last) Tj +1 0 0 1 394.44 458.76 Tm +(checkpoint.) Tj +1 0 0 1 445.8 458.76 Tm +(When) Tj +/R707 10 Tf +1 0 0 1 472.679 458.76 Tm +(mak) Tj +1 0 0 1 489.239 458.76 Tm +(e) Tj +ET +q +3 0 0 -0.48 494.268 459.252 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +1 0 0 1 497.159 458.76 Tm +(c) Tj +1 0 0 1 501.479 458.76 Tm +(hec) Tj +1 0 0 1 515.159 458.76 Tm +(kpoint) Tj +/R701 10 Tf +1 0 0 1 315 446.76 Tm +(is) Tj +1 0 0 1 326.4 446.76 Tm +(called,) Tj +/R707 10 Tf +1 0 0 1 358.08 446.76 Tm +(snfsd) Tj +/R701 10 Tf +1 0 0 1 383.4 446.76 Tm +(obtains) Tj +1 0 0 1 417 446.76 Tm +(the) Tj +1 0 0 1 433.92 446.76 Tm +(digest) Tj +ET +endstream +endobj +709 0 obj +23289 +endobj +710 0 obj +<< /Type /XObject /Name /R710 /Subtype /Image /Length 711 0 R +/ImageMask true /Width 49 /Height 71 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 49 /BlackIs1 true >>] +>> +stream +-PYiBJV"+0)UN3G',i&VcojQ(@sYoeKmuDG +s)&q9s*o7jJ)C$Fpj^'QJ%rnUr&sfAs+VQL0)H)pV[&D-@^p.P-,mCns*qra5O\@FpQ6S*nl.";J-Z~> +endstream +endobj +711 0 obj +117 +endobj +712 0 obj +<< /Length 713 0 R >> +stream +q +4.9 0 0 -7.1 462.9 453.8 cm +/R710 Do +Q +BT +1 0 0 1 472.559 446.76 Tm +(for) Tj +1 0 0 1 488.999 446.76 Tm +(the) Tj +1 0 0 1 505.919 446.76 Tm +(pre) Tj +1 0 0 1 518.519 446.76 Tm +(vious) Tj +1 0 0 1 315 434.76 Tm +(checkpoint) Tj +1 0 0 1 360.72 434.76 Tm +(state) Tj +1 0 0 1 381.12 434.76 Tm +(\(from) Tj +1 0 0 1 405.84 434.76 Tm +(the) Tj +1 0 0 1 420 434.76 Tm +(associated) Tj +1 0 0 1 462.96 434.76 Tm +(checkpoint) Tj +1 0 0 1 508.799 434.76 Tm +(record\).) Tj +1 0 0 1 315 422.88 Tm +(It) Tj +1 0 0 1 323.4 422.88 Tm +(computes) Tj +1 0 0 1 363.96 422.88 Tm +(ne) Tj +1 0 0 1 373.2 422.88 Tm +(w) Tj +1 0 0 1 382.56 422.88 Tm +(hash) Tj +1 0 0 1 403.2 422.88 Tm +(v) Tj +1 0 0 1 408 422.88 Tm +(alues) Tj +1 0 0 1 430.68 422.88 Tm +(for) Tj +1 0 0 1 444.6 422.88 Tm +(each) Tj +1 0 0 1 465.239 422.88 Tm +(block) Tj +1 0 0 1 489.599 422.88 Tm +(whose) Tj +1 0 0 1 517.439 422.88 Tm +(cop) Tj +1 0 0 1 531.839 422.88 Tm +(y-) Tj +1 0 0 1 315 410.88 Tm +(on-write) Tj +1 0 0 1 351.24 410.88 Tm +(bit) Tj +1 0 0 1 364.32 410.88 Tm +(is) Tj +1 0 0 1 373.56 410.88 Tm +(reset) Tj +1 0 0 1 394.92 410.88 Tm +(by) Tj +1 0 0 1 407.4 410.88 Tm +(applying) Tj +1 0 0 1 444.72 410.88 Tm +(MD5) Tj +1 0 0 1 468.36 410.88 Tm +(to) Tj +1 0 0 1 478.679 410.88 Tm +(the) Tj +1 0 0 1 493.319 410.88 Tm +(block) Tj +1 0 0 1 518.039 410.88 Tm +(inde) Tj +1 0 0 1 535.199 410.88 Tm +(x) Tj +1 0 0 1 315 399 Tm +(concatenated) Tj +1 0 0 1 369.6 399 Tm +(with) Tj +1 0 0 1 389.88 399 Tm +(the) Tj +1 0 0 1 404.64 399 Tm +(current) Tj +1 0 0 1 435.48 399 Tm +(block) Tj +1 0 0 1 460.2 399 Tm +(v) Tj +1 0 0 1 465 399 Tm +(alue.) Tj +1 0 0 1 488.039 399 Tm +(Then,) Tj +1 0 0 1 513.599 399 Tm +(it) Tj +1 0 0 1 521.759 399 Tm +(adds) Tj +1 0 0 1 315 387 Tm +(the) Tj +1 0 0 1 331.32 387 Tm +(ne) Tj +1 0 0 1 340.56 387 Tm +(w) Tj +1 0 0 1 351.72 387 Tm +(hash) Tj +1 0 0 1 374.16 387 Tm +(v) Tj +1 0 0 1 378.96 387 Tm +(alue) Tj +1 0 0 1 399.6 387 Tm +(to) Tj +ET +q +4.9 0 0 -7.1 411.8 394 cm +/R710 Do +Q +BT +1 0 0 1 416.64 387 Tm +(,) Tj +1 0 0 1 423.6 387 Tm +(subtracts) Tj +1 0 0 1 463.2 387 Tm +(the) Tj +1 0 0 1 479.399 387 Tm +(old) Tj +1 0 0 1 496.319 387 Tm +(hash) Tj +1 0 0 1 518.639 387 Tm +(v) Tj +1 0 0 1 523.439 387 Tm +(alue) Tj +1 0 0 1 315 375.12 Tm +(from) Tj +ET +q +4.9 0 0 -7.1 338.5 382.1 cm +/R710 Do +Q +BT +1 0 0 1 343.32 375.12 Tm +(,) Tj +1 0 0 1 349.8 375.12 Tm +(and) Tj +1 0 0 1 367.92 375.12 Tm +(updates) Tj +1 0 0 1 402.12 375.12 Tm +(the) Tj +1 0 0 1 418.08 375.12 Tm +(table) Tj +1 0 0 1 441.12 375.12 Tm +(to) Tj +1 0 0 1 452.64 375.12 Tm +(contain) Tj +1 0 0 1 485.759 375.12 Tm +(the) Tj +1 0 0 1 501.599 375.12 Tm +(ne) Tj +1 0 0 1 510.839 375.12 Tm +(w) Tj +1 0 0 1 521.759 375.12 Tm +(hash) Tj +1 0 0 1 315 363.12 Tm +(v) Tj +1 0 0 1 319.8 363.12 Tm +(alue.) Tj +1 0 0 1 344.64 363.12 Tm +(This) Tj +1 0 0 1 365.52 363.12 Tm +(process) Tj +1 0 0 1 398.64 363.12 Tm +(is) Tj +1 0 0 1 408.48 363.12 Tm +(ef) Tj +1 0 0 1 416.04 363.12 Tm +(\256cient) Tj +1 0 0 1 444.12 363.12 Tm +(pro) Tj +1 0 0 1 457.44 363.12 Tm +(vided) Tj +1 0 0 1 482.639 363.12 Tm +(the) Tj +1 0 0 1 497.999 363.12 Tm +(number) Tj +1 0 0 1 531.719 363.12 Tm +(of) Tj +1 0 0 1 315 351.24 Tm +(modi\256ed) Tj +1 0 0 1 352.32 351.24 Tm +(blocks) Tj +1 0 0 1 380.28 351.24 Tm +(is) Tj +1 0 0 1 388.8 351.24 Tm +(small;) Tj +1 0 0 1 415.32 351.24 Tm +(as) Tj +1 0 0 1 425.52 351.24 Tm +(mentioned) Tj +1 0 0 1 469.44 351.24 Tm +(abo) Tj +1 0 0 1 483.839 351.24 Tm +(v) Tj +1 0 0 1 488.759 351.24 Tm +(e,) Tj +1 0 0 1 497.519 351.24 Tm +(on) Tj +1 0 0 1 509.399 351.24 Tm +(a) Tj +1 0 0 1 513.599 351.24 Tm +(v) Tj +1 0 0 1 518.519 351.24 Tm +(erage) Tj +1 0 0 1 315 339.24 Tm +(182) Tj +1 0 0 1 333 339.24 Tm +(blocks) Tj +1 0 0 1 362.04 339.24 Tm +(are) Tj +1 0 0 1 377.28 339.24 Tm +(modi\256ed) Tj +1 0 0 1 415.68 339.24 Tm +(per) Tj +1 0 0 1 431.52 339.24 Tm +(checkpoint) Tj +1 0 0 1 478.319 339.24 Tm +(for) Tj +1 0 0 1 492.959 339.24 Tm +(the) Tj +1 0 0 1 508.199 339.24 Tm +(Andre) Tj +1 0 0 1 533.039 339.24 Tm +(w) Tj +1 0 0 1 315 327.36 Tm +(benchmark.) Tj +/R704 12 Tf +1 0 0 1 315 302.04 Tm +(7) Tj +1 0 0 1 332.88 302.04 Tm +(P) Tj +1 0 0 1 339.96 302.04 Tm +(erf) Tj +1 0 0 1 354.24 302.04 Tm +(ormance) Tj +1 0 0 1 401.76 302.04 Tm +(Ev) Tj +1 0 0 1 415.56 302.04 Tm +(aluation) Tj +/R701 10 Tf +1 0 0 1 315 285.48 Tm +(This) Tj +1 0 0 1 337.08 285.48 Tm +(section) Tj +1 0 0 1 369.72 285.48 Tm +(e) Tj +1 0 0 1 373.92 285.48 Tm +(v) Tj +1 0 0 1 378.72 285.48 Tm +(aluates) Tj +1 0 0 1 410.64 285.48 Tm +(the) Tj +1 0 0 1 427.2 285.48 Tm +(performance) Tj +1 0 0 1 481.919 285.48 Tm +(of) Tj +1 0 0 1 494.639 285.48 Tm +(our) Tj +1 0 0 1 512.279 285.48 Tm +(system) Tj +1 0 0 1 315 273.6 Tm +(using) Tj +1 0 0 1 341.88 273.6 Tm +(tw) Tj +1 0 0 1 351.72 273.6 Tm +(o) Tj +1 0 0 1 362.04 273.6 Tm +(benchmarks:) Tj +1 0 0 1 422.28 273.6 Tm +(a) Tj +1 0 0 1 432 273.6 Tm +(micro-benchmark) Tj +1 0 0 1 508.079 273.6 Tm +(and) Tj +1 0 0 1 527.879 273.6 Tm +(the) Tj +1 0 0 1 315 261.6 Tm +(Andre) Tj +1 0 0 1 339.84 261.6 Tm +(w) Tj +1 0 0 1 348.6 261.6 Tm +(benchmark) Tj +1 0 0 1 394.8 261.6 Tm +([15) Tj +1 0 0 1 407.4 261.6 Tm +(].) Tj +1 0 0 1 416.52 261.6 Tm +(The) Tj +1 0 0 1 433.68 261.6 Tm +(micro-benchmark) Tj +1 0 0 1 506.519 261.6 Tm +(pr) Tj +1 0 0 1 514.799 261.6 Tm +(o) Tj +1 0 0 1 519.719 261.6 Tm +(v) Tj +1 0 0 1 524.639 261.6 Tm +(ides) Tj +1 0 0 1 315 249.72 Tm +(a) Tj +1 0 0 1 322.8 249.72 Tm +(service-independent) Tj +1 0 0 1 406.44 249.72 Tm +(e) Tj +1 0 0 1 410.64 249.72 Tm +(v) Tj +1 0 0 1 415.44 249.72 Tm +(aluation) Tj +1 0 0 1 450.72 249.72 Tm +(of) Tj +1 0 0 1 462.479 249.72 Tm +(the) Tj +1 0 0 1 477.959 249.72 Tm +(performance) Tj +1 0 0 1 531.719 249.72 Tm +(of) Tj +1 0 0 1 315 237.72 Tm +(the) Tj +1 0 0 1 330.12 237.72 Tm +(replication) Tj +1 0 0 1 375.6 237.72 Tm +(library;) Tj +1 0 0 1 408 237.72 Tm +(it) Tj +1 0 0 1 416.4 237.72 Tm +(measures) Tj +1 0 0 1 456.359 237.72 Tm +(the) Tj +1 0 0 1 471.479 237.72 Tm +(latenc) Tj +1 0 0 1 495.239 237.72 Tm +(y) Tj +1 0 0 1 502.919 237.72 Tm +(to) Tj +1 0 0 1 513.599 237.72 Tm +(in) Tj +1 0 0 1 521.039 237.72 Tm +(v) Tj +1 0 0 1 525.839 237.72 Tm +(ok) Tj +1 0 0 1 535.799 237.72 Tm +(e) Tj +1 0 0 1 315 225.84 Tm +(a) Tj +1 0 0 1 321.96 225.84 Tm +(null) Tj +1 0 0 1 339.96 225.84 Tm +(operation,) Tj +1 0 0 1 382.56 225.84 Tm +(i.e.,) Tj +1 0 0 1 399.72 225.84 Tm +(an) Tj +1 0 0 1 411.6 225.84 Tm +(operation) Tj +1 0 0 1 451.679 225.84 Tm +(that) Tj +1 0 0 1 469.079 225.84 Tm +(does) Tj +1 0 0 1 489.839 225.84 Tm +(nothing.) Tj +1 0 0 1 324.96 213.84 Tm +(The) Tj +1 0 0 1 343.08 213.84 Tm +(Andre) Tj +1 0 0 1 367.92 213.84 Tm +(w) Tj +1 0 0 1 377.52 213.84 Tm +(benchmark) Tj +1 0 0 1 424.44 213.84 Tm +(is) Tj +1 0 0 1 433.68 213.84 Tm +(used) Tj +1 0 0 1 454.56 213.84 Tm +(to) Tj +1 0 0 1 465 213.84 Tm +(compare) Tj +1 0 0 1 501.96 213.84 Tm +(BFS) Tj +1 0 0 1 522.239 213.84 Tm +(with) Tj +1 0 0 1 315 201.84 Tm +(tw) Tj +1 0 0 1 324.84 201.84 Tm +(o) Tj +1 0 0 1 331.56 201.84 Tm +(other) Tj +1 0 0 1 353.88 201.84 Tm +(\256le) Tj +1 0 0 1 368.28 201.84 Tm +(systems:) Tj +1 0 0 1 405.48 201.84 Tm +(one) Tj +1 0 0 1 421.68 201.84 Tm +(is) Tj +1 0 0 1 429.96 201.84 Tm +(the) Tj +1 0 0 1 443.879 201.84 Tm +(NFS) Tj +1 0 0 1 463.799 201.84 Tm +(V2) Tj +1 0 0 1 477.719 201.84 Tm +(implementation) Tj +1 0 0 1 315 189.96 Tm +(in) Tj +1 0 0 1 325.56 189.96 Tm +(Digital) Tj +1 0 0 1 356.04 189.96 Tm +(Unix,) Tj +1 0 0 1 381.24 189.96 Tm +(and) Tj +1 0 0 1 398.4 189.96 Tm +(the) Tj +1 0 0 1 413.4 189.96 Tm +(other) Tj +1 0 0 1 436.56 189.96 Tm +(is) Tj +1 0 0 1 446.04 189.96 Tm +(identical) Tj +1 0 0 1 483.119 189.96 Tm +(to) Tj +1 0 0 1 493.679 189.96 Tm +(BFS) Tj +1 0 0 1 514.199 189.96 Tm +(e) Tj +1 0 0 1 518.519 189.96 Tm +(xcept) Tj +1 0 0 1 315 177.96 Tm +(without) Tj +1 0 0 1 348.72 177.96 Tm +(replication.) Tj +1 0 0 1 399.96 177.96 Tm +(The) Tj +1 0 0 1 418.8 177.96 Tm +(\256rst) Tj +1 0 0 1 437.64 177.96 Tm +(comparison) Tj +1 0 0 1 487.439 177.96 Tm +(demonstrates) Tj +1 0 0 1 315 166.08 Tm +(that) Tj +1 0 0 1 332.4 166.08 Tm +(our) Tj +1 0 0 1 348.24 166.08 Tm +(system) Tj +1 0 0 1 378.36 166.08 Tm +(is) Tj +1 0 0 1 387.6 166.08 Tm +(practical) Tj +1 0 0 1 424.32 166.08 Tm +(by) Tj +1 0 0 1 436.799 166.08 Tm +(sho) Tj +1 0 0 1 450.479 166.08 Tm +(wing) Tj +1 0 0 1 472.799 166.08 Tm +(that) Tj +1 0 0 1 490.319 166.08 Tm +(its) Tj +1 0 0 1 502.199 166.08 Tm +(latenc) Tj +1 0 0 1 525.959 166.08 Tm +(y) Tj +1 0 0 1 533.399 166.08 Tm +(is) Tj +1 0 0 1 315 154.08 Tm +(similar) Tj +1 0 0 1 344.88 154.08 Tm +(to) Tj +1 0 0 1 354.72 154.08 Tm +(the) Tj +1 0 0 1 369 154.08 Tm +(latenc) Tj +1 0 0 1 392.76 154.08 Tm +(y) Tj +1 0 0 1 399.84 154.08 Tm +(of) Tj +1 0 0 1 410.28 154.08 Tm +(a) Tj +1 0 0 1 416.88 154.08 Tm +(commercial) Tj +1 0 0 1 466.079 154.08 Tm +(system) Tj +1 0 0 1 495.839 154.08 Tm +(that) Tj +1 0 0 1 512.999 154.08 Tm +(is) Tj +1 0 0 1 521.759 154.08 Tm +(used) Tj +1 0 0 1 315 142.2 Tm +(daily) Tj +1 0 0 1 336.72 142.2 Tm +(by) Tj +1 0 0 1 348.6 142.2 Tm +(man) Tj +1 0 0 1 365.76 142.2 Tm +(y) Tj +1 0 0 1 372.36 142.2 Tm +(users.) Tj +1 0 0 1 398.76 142.2 Tm +(The) Tj +1 0 0 1 416.04 142.2 Tm +(second) Tj +1 0 0 1 445.56 142.2 Tm +(comparison) Tj +1 0 0 1 493.919 142.2 Tm +(allo) Tj +1 0 0 1 508.679 142.2 Tm +(ws) Tj +1 0 0 1 521.519 142.2 Tm +(us) Tj +1 0 0 1 532.199 142.2 Tm +(to) Tj +1 0 0 1 315 130.2 Tm +(e) Tj +1 0 0 1 319.2 130.2 Tm +(v) Tj +1 0 0 1 324 130.2 Tm +(aluate) Tj +1 0 0 1 350.16 130.2 Tm +(the) Tj +1 0 0 1 364.8 130.2 Tm +(o) Tj +1 0 0 1 369.72 130.2 Tm +(v) Tj +1 0 0 1 374.64 130.2 Tm +(erhead) Tj +1 0 0 1 403.44 130.2 Tm +(of) Tj +1 0 0 1 414.24 130.2 Tm +(our) Tj +1 0 0 1 429.96 130.2 Tm +(algorithm) Tj +1 0 0 1 471.119 130.2 Tm +(accurately) Tj +1 0 0 1 514.559 130.2 Tm +(within) Tj +1 0 0 1 315 118.32 Tm +(an) Tj +1 0 0 1 326.88 118.32 Tm +(implementation) Tj +1 0 0 1 391.92 118.32 Tm +(of) Tj +1 0 0 1 402.72 118.32 Tm +(a) Tj +1 0 0 1 409.68 118.32 Tm +(real) Tj +1 0 0 1 427.08 118.32 Tm +(service.) Tj +/R704 10 Tf +1 0 0 1 315 97.2 Tm +(7.1) Tj +1 0 0 1 337.44 97.2 Tm +(Experimental) Tj +1 0 0 1 398.04 97.2 Tm +(Setup) Tj +/R701 10 Tf +1 0 0 1 315 82.92 Tm +(The) Tj +1 0 0 1 335.76 82.92 Tm +(e) Tj +1 0 0 1 340.08 82.92 Tm +(xperiments) Tj +1 0 0 1 389.64 82.92 Tm +(measure) Tj +1 0 0 1 428.04 82.92 Tm +(normal-case) Tj +1 0 0 1 482.04 82.92 Tm +(beha) Tj +1 0 0 1 500.76 82.92 Tm +(vior) Tj +1 0 0 1 521.999 82.92 Tm +(\(i.e.,) Tj +1 0 0 1 315 71.04 Tm +(there) Tj +1 0 0 1 337.8 71.04 Tm +(are) Tj +1 0 0 1 352.92 71.04 Tm +(no) Tj +1 0 0 1 365.88 71.04 Tm +(vie) Tj +1 0 0 1 377.88 71.04 Tm +(w) Tj +1 0 0 1 387.96 71.04 Tm +(changes\),) Tj +1 0 0 1 428.88 71.04 Tm +(because) Tj +1 0 0 1 463.319 71.04 Tm +(this) Tj +1 0 0 1 480.599 71.04 Tm +(is) Tj +1 0 0 1 490.199 71.04 Tm +(the) Tj +1 0 0 1 505.319 71.04 Tm +(beha) Tj +1 0 0 1 524.039 71.04 Tm +(vior) Tj +1 0 0 1 300.96 35.16 Tm +(10) Tj +ET +Q +endstream +endobj +713 0 obj +10322 +endobj +714 0 obj +<< +/Type /Page +/MediaBox [0 0 612 792] +/Parent 2 0 R +/Resources << /ProcSet [/PDF /ImageB /Text] +/Font << +/R707 707 0 R +/R704 704 0 R +/R701 701 0 R +>> +/XObject << +/R710 710 0 R +>> +>> +/Contents [ +702 0 R +705 0 R +708 0 R +712 0 R +] +>> +endobj +715 0 obj +<< /Type /Font /Name /R715 /Subtype /Type1 /BaseFont /Times-Roman >> +endobj +716 0 obj +<< /Length 717 0 R >> +stream +q +BT +/R715 10 Tf +1 0 0 1 72 709.08 Tm +(that) Tj +1 0 0 1 91.9199 709.08 Tm +(determines) Tj +1 0 0 1 140.52 709.08 Tm +(the) Tj +1 0 0 1 157.56 709.08 Tm +(performance) Tj +1 0 0 1 212.88 709.08 Tm +(of) Tj +1 0 0 1 226.08 709.08 Tm +(the) Tj +1 0 0 1 243.239 709.08 Tm +(system.) Tj +1 0 0 1 284.279 709.08 Tm +(All) Tj +1 0 0 1 72 697.08 Tm +(e) Tj +1 0 0 1 76.32 697.08 Tm +(xperiments) Tj +1 0 0 1 126.84 697.08 Tm +(ran) Tj +1 0 0 1 145.8 697.08 Tm +(with) Tj +1 0 0 1 169.68 697.08 Tm +(one) Tj +1 0 0 1 190.32 697.08 Tm +(client) Tj +1 0 0 1 218.76 697.08 Tm +(running) Tj +1 0 0 1 255.959 697.08 Tm +(tw) Tj +1 0 0 1 265.799 697.08 Tm +(o) Tj +1 0 0 1 277.079 697.08 Tm +(relay) Tj +1 0 0 1 72 685.2 Tm +(processes,) Tj +1 0 0 1 116.28 685.2 Tm +(and) Tj +1 0 0 1 133.92 685.2 Tm +(four) Tj +1 0 0 1 153.96 685.2 Tm +(replicas.) Tj +1 0 0 1 193.56 685.2 Tm +(F) Tj +1 0 0 1 198.96 685.2 Tm +(our) Tj +1 0 0 1 215.64 685.2 Tm +(replicas) Tj +1 0 0 1 249.959 685.2 Tm +(can) Tj +1 0 0 1 267.119 685.2 Tm +(tolerate) Tj +1 0 0 1 72 673.2 Tm +(one) Tj +1 0 0 1 90.24 673.2 Tm +(Byzantine) Tj +1 0 0 1 134.64 673.2 Tm +(f) Tj +1 0 0 1 137.88 673.2 Tm +(ault;) Tj +1 0 0 1 160.2 673.2 Tm +(we) Tj +1 0 0 1 175.68 673.2 Tm +(e) Tj +1 0 0 1 180 673.2 Tm +(xpect) Tj +1 0 0 1 205.44 673.2 Tm +(this) Tj +1 0 0 1 223.68 673.2 Tm +(reliability) Tj +1 0 0 1 266.399 673.2 Tm +(le) Tj +1 0 0 1 273.359 673.2 Tm +(v) Tj +1 0 0 1 278.279 673.2 Tm +(el) Tj +1 0 0 1 289.199 673.2 Tm +(to) Tj +1 0 0 1 72 661.32 Tm +(suf) Tj +1 0 0 1 84 661.32 Tm +(\256ce) Tj +1 0 0 1 103.68 661.32 Tm +(for) Tj +1 0 0 1 120.6 661.32 Tm +(most) Tj +1 0 0 1 145.32 661.32 Tm +(applications.) Tj +1 0 0 1 207.96 661.32 Tm +(The) Tj +1 0 0 1 228.84 661.32 Tm +(replicas) Tj +1 0 0 1 265.079 661.32 Tm +(and) Tj +1 0 0 1 284.879 661.32 Tm +(the) Tj +1 0 0 1 72 649.32 Tm +(client) Tj +1 0 0 1 95.7599 649.32 Tm +(ran) Tj +1 0 0 1 110.16 649.32 Tm +(on) Tj +1 0 0 1 121.8 649.32 Tm +(identical) Tj +1 0 0 1 157.8 649.32 Tm +(DEC) Tj +1 0 0 1 179.28 649.32 Tm +(300) Tj +1 0 0 1 194.28 649.32 Tm +(0/4) Tj +1 0 0 1 207 649.32 Tm +(00) Tj +1 0 0 1 218.52 649.32 Tm +(Alph) Tj +1 0 0 1 238.439 649.32 Tm +(a) Tj +1 0 0 1 244.439 649.32 Tm +(w) Tj +1 0 0 1 251.519 649.32 Tm +(o) Tj +1 0 0 1 256.439 649.32 Tm +(rkstatio) Tj +1 0 0 1 286.319 649.32 Tm +(ns.) Tj +1 0 0 1 72 637.44 Tm +(These) Tj +1 0 0 1 101.88 637.44 Tm +(w) Tj +1 0 0 1 108.96 637.44 Tm +(orkstations) Tj +1 0 0 1 158.88 637.44 Tm +(ha) Tj +1 0 0 1 168.12 637.44 Tm +(v) Tj +1 0 0 1 173.04 637.44 Tm +(e) Tj +1 0 0 1 183.48 637.44 Tm +(a) Tj +1 0 0 1 193.92 637.44 Tm +(133) Tj +1 0 0 1 215.04 637.44 Tm +(MHz) Tj +1 0 0 1 241.68 637.44 Tm +(Alpha) Tj +1 0 0 1 272.039 637.44 Tm +(21064) Tj +1 0 0 1 72 625.44 Tm +(processor) Tj +1 0 0 1 110.04 625.44 Tm +(,) Tj +1 0 0 1 117.12 625.44 Tm +(128) Tj +1 0 0 1 136.32 625.44 Tm +(MB) Tj +1 0 0 1 156.24 625.44 Tm +(of) Tj +1 0 0 1 168.84 625.44 Tm +(memory) Tj +1 0 0 1 201.72 625.44 Tm +(,) Tj +1 0 0 1 208.68 625.44 Tm +(and) Tj +1 0 0 1 227.4 625.44 Tm +(run) Tj +1 0 0 1 245.04 625.44 Tm +(Digital) Tj +1 0 0 1 277.079 625.44 Tm +(Unix) Tj +1 0 0 1 72 613.44 Tm +(v) Tj +1 0 0 1 76.92 613.44 Tm +(ersion) Tj +1 0 0 1 104.28 613.44 Tm +(4.0.) Tj +1 0 0 1 124.56 613.44 Tm +(The) Tj +1 0 0 1 143.04 613.44 Tm +(\256le) Tj +1 0 0 1 158.88 613.44 Tm +(system) Tj +1 0 0 1 189.6 613.44 Tm +(w) Tj +1 0 0 1 196.68 613.44 Tm +(as) Tj +1 0 0 1 208.08 613.44 Tm +(stored) Tj +1 0 0 1 235.559 613.44 Tm +(by) Tj +1 0 0 1 248.519 613.44 Tm +(each) Tj +1 0 0 1 269.879 613.44 Tm +(replica) Tj +1 0 0 1 72 601.56 Tm +(on) Tj +1 0 0 1 87.7199 601.56 Tm +(a) Tj +1 0 0 1 97.9199 601.56 Tm +(DEC) Tj +1 0 0 1 123.6 601.56 Tm +(RZ26) Tj +1 0 0 1 152.04 601.56 Tm +(disk.) Tj +1 0 0 1 184.56 601.56 Tm +(All) Tj +1 0 0 1 203.16 601.56 Tm +(the) Tj +1 0 0 1 221.04 601.56 Tm +(w) Tj +1 0 0 1 228.12 601.56 Tm +(orkstations) Tj +1 0 0 1 277.679 601.56 Tm +(were) Tj +1 0 0 1 72 589.56 Tm +(connected) Tj +1 0 0 1 114.48 589.56 Tm +(by) Tj +1 0 0 1 126.48 589.56 Tm +(a) Tj +1 0 0 1 132.96 589.56 Tm +(10Mbit/s) Tj +1 0 0 1 171 589.56 Tm +(switched) Tj +1 0 0 1 208.44 589.56 Tm +(Ethernet) Tj +1 0 0 1 244.199 589.56 Tm +(and) Tj +1 0 0 1 260.639 589.56 Tm +(had) Tj +1 0 0 1 277.079 589.56 Tm +(DEC) Tj +1 0 0 1 72 577.68 Tm +(LANCE) Tj +1 0 0 1 109.56 577.68 Tm +(Ethernet) Tj +1 0 0 1 147.72 577.68 Tm +(interf) Tj +1 0 0 1 169.32 577.68 Tm +(aces.) Tj +1 0 0 1 198.12 577.68 Tm +(The) Tj +1 0 0 1 218.04 577.68 Tm +(switch) Tj +1 0 0 1 248.52 577.68 Tm +(w) Tj +1 0 0 1 255.6 577.68 Tm +(as) Tj +1 0 0 1 268.2 577.68 Tm +(a) Tj +1 0 0 1 277.079 577.68 Tm +(DEC) Tj +1 0 0 1 72 565.68 Tm +(EtherW) Tj +1 0 0 1 102.96 565.68 Tm +(ORKS) Tj +1 0 0 1 132.6 565.68 Tm +(8T/TX.) Tj +1 0 0 1 165.36 565.68 Tm +(The) Tj +1 0 0 1 183.96 565.68 Tm +(e) Tj +1 0 0 1 188.28 565.68 Tm +(xperiments) Tj +1 0 0 1 235.56 565.68 Tm +(were) Tj +1 0 0 1 258.12 565.68 Tm +(run) Tj +1 0 0 1 274.559 565.68 Tm +(on) Tj +1 0 0 1 287.639 565.68 Tm +(an) Tj +1 0 0 1 72 553.8 Tm +(isolated) Tj +1 0 0 1 105.48 553.8 Tm +(netw) Tj +1 0 0 1 124.8 553.8 Tm +(ork.) Tj +1 0 0 1 81.96 541.8 Tm +(The) Tj +1 0 0 1 101.28 541.8 Tm +(interv) Tj +1 0 0 1 124.44 541.8 Tm +(al) Tj +1 0 0 1 135.36 541.8 Tm +(between) Tj +1 0 0 1 172.44 541.8 Tm +(checkpoints) Tj +1 0 0 1 223.8 541.8 Tm +(w) Tj +1 0 0 1 230.88 541.8 Tm +(as) Tj +1 0 0 1 243.12 541.8 Tm +(128) Tj +1 0 0 1 261.839 541.8 Tm +(requests,) Tj +1 0 0 1 72 529.92 Tm +(which) Tj +1 0 0 1 98.4 529.92 Tm +(causes) Tj +1 0 0 1 126.6 529.92 Tm +(garbage) Tj +1 0 0 1 160.2 529.92 Tm +(collection) Tj +1 0 0 1 201.6 529.92 Tm +(to) Tj +1 0 0 1 211.44 529.92 Tm +(occur) Tj +1 0 0 1 235.68 529.92 Tm +(se) Tj +1 0 0 1 243.72 529.92 Tm +(v) Tj +1 0 0 1 248.639 529.92 Tm +(eral) Tj +1 0 0 1 265.559 529.92 Tm +(times) Tj +1 0 0 1 289.199 529.92 Tm +(in) Tj +1 0 0 1 72 517.92 Tm +(an) Tj +1 0 0 1 81.36 517.92 Tm +(y) Tj +1 0 0 1 88.4399 517.92 Tm +(of) Tj +1 0 0 1 98.8799 517.92 Tm +(the) Tj +1 0 0 1 113.16 517.92 Tm +(e) Tj +1 0 0 1 117.48 517.92 Tm +(xperiments.) Tj +1 0 0 1 167.64 517.92 Tm +(The) Tj +1 0 0 1 185.28 517.92 Tm +(maximum) Tj +1 0 0 1 227.88 517.92 Tm +(sequence) Tj +1 0 0 1 266.519 517.92 Tm +(number) Tj +1 0 0 1 72 506.04 Tm +(accepted) Tj +1 0 0 1 110.76 506.04 Tm +(by) Tj +1 0 0 1 124.68 506.04 Tm +(replicas) Tj +1 0 0 1 159.6 506.04 Tm +(in) Tj +1 0 0 1 171.36 506.04 Tm +(pre-prepare) Tj +1 0 0 1 221.16 506.04 Tm +(messages) Tj +1 0 0 1 262.679 506.04 Tm +(w) Tj +1 0 0 1 269.759 506.04 Tm +(as) Tj +1 0 0 1 281.999 506.04 Tm +(256) Tj +1 0 0 1 72 494.04 Tm +(plus) Tj +1 0 0 1 91.0799 494.04 Tm +(the) Tj +1 0 0 1 105.72 494.04 Tm +(sequence) Tj +1 0 0 1 144.72 494.04 Tm +(number) Tj +1 0 0 1 177.72 494.04 Tm +(of) Tj +1 0 0 1 188.52 494.04 Tm +(the) Tj +1 0 0 1 203.16 494.04 Tm +(last) Tj +1 0 0 1 219.48 494.04 Tm +(stable) Tj +1 0 0 1 245.159 494.04 Tm +(checkpoint.) Tj +ET +endstream +endobj +717 0 obj +6532 +endobj +718 0 obj +<< /Type /Font /Name /R718 /Subtype /Type1 /BaseFont /Times-Bold >> +endobj +719 0 obj +<< /Length 720 0 R >> +stream +BT +/R718 10 Tf +1 0 0 1 72 474 Tm +(7.2) Tj +1 0 0 1 94.4399 474 Tm +(Micr) Tj +1 0 0 1 115.32 474 Tm +(o-Benchmark) Tj +/R715 10 Tf +1 0 0 1 72 459.84 Tm +(The) Tj +1 0 0 1 91.9199 459.84 Tm +(micro-benchmark) Tj +1 0 0 1 167.16 459.84 Tm +(measures) Tj +1 0 0 1 208.68 459.84 Tm +(the) Tj +1 0 0 1 225.36 459.84 Tm +(latenc) Tj +1 0 0 1 249.12 459.84 Tm +(y) Tj +1 0 0 1 258.36 459.84 Tm +(to) Tj +1 0 0 1 270.599 459.84 Tm +(in) Tj +1 0 0 1 278.039 459.84 Tm +(v) Tj +1 0 0 1 282.839 459.84 Tm +(ok) Tj +1 0 0 1 292.799 459.84 Tm +(e) Tj +1 0 0 1 72 447.96 Tm +(a) Tj +1 0 0 1 80.5199 447.96 Tm +(null) Tj +1 0 0 1 100.08 447.96 Tm +(operation.) Tj +1 0 0 1 148.44 447.96 Tm +(It) Tj +1 0 0 1 158.52 447.96 Tm +(e) Tj +1 0 0 1 162.72 447.96 Tm +(v) Tj +1 0 0 1 167.52 447.96 Tm +(aluates) Tj +1 0 0 1 199.2 447.96 Tm +(the) Tj +1 0 0 1 215.4 447.96 Tm +(performance) Tj +1 0 0 1 269.759 447.96 Tm +(of) Tj +1 0 0 1 282.119 447.96 Tm +(tw) Tj +1 0 0 1 291.959 447.96 Tm +(o) Tj +1 0 0 1 72 435.96 Tm +(implementations) Tj +1 0 0 1 142.2 435.96 Tm +(of) Tj +1 0 0 1 154.32 435.96 Tm +(a) Tj +1 0 0 1 162.48 435.96 Tm +(simple) Tj +1 0 0 1 192.84 435.96 Tm +(service) Tj +1 0 0 1 224.76 435.96 Tm +(with) Tj +1 0 0 1 246.24 435.96 Tm +(no) Tj +1 0 0 1 260.039 435.96 Tm +(state) Tj +1 0 0 1 281.999 435.96 Tm +(that) Tj +1 0 0 1 72 423.96 Tm +(implements) Tj +1 0 0 1 122.16 423.96 Tm +(null) Tj +1 0 0 1 141.36 423.96 Tm +(operations) Tj +1 0 0 1 186.48 423.96 Tm +(with) Tj +1 0 0 1 207.96 423.96 Tm +(ar) Tj +1 0 0 1 215.64 423.96 Tm +(guments) Tj +1 0 0 1 252.959 423.96 Tm +(and) Tj +1 0 0 1 270.959 423.96 Tm +(results) Tj +1 0 0 1 72 412.08 Tm +(of) Tj +1 0 0 1 83.04 412.08 Tm +(dif) Tj +1 0 0 1 93.9599 412.08 Tm +(ferent) Tj +1 0 0 1 120 412.08 Tm +(sizes.) Tj +1 0 0 1 146.28 412.08 Tm +(The) Tj +1 0 0 1 164.64 412.08 Tm +(\256rst) Tj +1 0 0 1 182.88 412.08 Tm +(implementation) Tj +1 0 0 1 248.279 412.08 Tm +(is) Tj +1 0 0 1 257.759 412.08 Tm +(replicated) Tj +1 0 0 1 72 400.08 Tm +(using) Tj +1 0 0 1 98.5199 400.08 Tm +(our) Tj +1 0 0 1 116.64 400.08 Tm +(library) Tj +1 0 0 1 148.08 400.08 Tm +(and) Tj +1 0 0 1 167.4 400.08 Tm +(the) Tj +1 0 0 1 184.44 400.08 Tm +(second) Tj +1 0 0 1 216.959 400.08 Tm +(is) Tj +1 0 0 1 228.479 400.08 Tm +(unreplicated) Tj +1 0 0 1 282.599 400.08 Tm +(and) Tj +1 0 0 1 72 388.2 Tm +(uses) Tj +1 0 0 1 92.7599 388.2 Tm +(UDP) Tj +1 0 0 1 116.4 388.2 Tm +(directly) Tj +1 0 0 1 146.4 388.2 Tm +(.) Tj +1 0 0 1 155.64 388.2 Tm +(T) Tj +1 0 0 1 160.92 388.2 Tm +(able) Tj +1 0 0 1 181.2 388.2 Tm +(1) Tj +1 0 0 1 189.84 388.2 Tm +(reports) Tj +1 0 0 1 221.159 388.2 Tm +(the) Tj +1 0 0 1 236.879 388.2 Tm +(response) Tj +1 0 0 1 275.399 388.2 Tm +(times) Tj +1 0 0 1 72 376.2 Tm +(measured) Tj +1 0 0 1 115.56 376.2 Tm +(at) Tj +1 0 0 1 128.16 376.2 Tm +(the) Tj +1 0 0 1 145.68 376.2 Tm +(client) Tj +1 0 0 1 173.16 376.2 Tm +(for) Tj +1 0 0 1 190.2 376.2 Tm +(both) Tj +1 0 0 1 213.24 376.2 Tm +(read-only) Tj +1 0 0 1 256.799 376.2 Tm +(and) Tj +1 0 0 1 276.479 376.2 Tm +(read-) Tj +1 0 0 1 72 364.32 Tm +(write) Tj +1 0 0 1 95.5199 364.32 Tm +(operations.) Tj +1 0 0 1 144.6 364.32 Tm +(The) Tj +1 0 0 1 160.08 364.32 Tm +(y) Tj +1 0 0 1 167.88 364.32 Tm +(were) Tj +1 0 0 1 190.32 364.32 Tm +(obtained) Tj +1 0 0 1 227.639 364.32 Tm +(by) Tj +1 0 0 1 240.599 364.32 Tm +(timing) Tj +1 0 0 1 269.639 364.32 Tm +(10,000) Tj +1 0 0 1 72 352.32 Tm +(operation) Tj +1 0 0 1 111.48 352.32 Tm +(in) Tj +1 0 0 1 118.92 352.32 Tm +(v) Tj +1 0 0 1 123.72 352.32 Tm +(ocations) Tj +1 0 0 1 158.64 352.32 Tm +(in) Tj +1 0 0 1 168.24 352.32 Tm +(three) Tj +1 0 0 1 190.08 352.32 Tm +(separate) Tj +1 0 0 1 224.519 352.32 Tm +(runs) Tj +1 0 0 1 243.479 352.32 Tm +(and) Tj +1 0 0 1 259.799 352.32 Tm +(we) Tj +1 0 0 1 273.239 352.32 Tm +(report) Tj +1 0 0 1 72 340.44 Tm +(the) Tj +1 0 0 1 89.1599 340.44 Tm +(median) Tj +1 0 0 1 123.48 340.44 Tm +(v) Tj +1 0 0 1 128.28 340.44 Tm +(alue) Tj +1 0 0 1 149.88 340.44 Tm +(of) Tj +1 0 0 1 163.2 340.44 Tm +(the) Tj +1 0 0 1 180.36 340.44 Tm +(three) Tj +1 0 0 1 205.32 340.44 Tm +(runs.) Tj +1 0 0 1 236.04 340.44 Tm +(The) Tj +1 0 0 1 256.56 340.44 Tm +(maximum) Tj +1 0 0 1 72 328.44 Tm +(de) Tj +1 0 0 1 81.24 328.44 Tm +(viation) Tj +1 0 0 1 112.8 328.44 Tm +(from) Tj +1 0 0 1 136.08 328.44 Tm +(the) Tj +1 0 0 1 152.16 328.44 Tm +(median) Tj +1 0 0 1 185.4 328.44 Tm +(w) Tj +1 0 0 1 192.48 328.44 Tm +(as) Tj +1 0 0 1 204.6 328.44 Tm +(al) Tj +1 0 0 1 211.68 328.44 Tm +(w) Tj +1 0 0 1 218.76 328.44 Tm +(ays) Tj +1 0 0 1 236.04 328.44 Tm +(belo) Tj +1 0 0 1 253.08 328.44 Tm +(w) Tj +1 0 0 1 264 328.44 Tm +(0.3%) Tj +1 0 0 1 288.719 328.44 Tm +(of) Tj +1 0 0 1 72 316.56 Tm +(the) Tj +1 0 0 1 88.0799 316.56 Tm +(reported) Tj +1 0 0 1 125.16 316.56 Tm +(v) Tj +1 0 0 1 129.96 316.56 Tm +(alue.) Tj +1 0 0 1 156.84 316.56 Tm +(W) Tj +1 0 0 1 165.36 316.56 Tm +(e) Tj +1 0 0 1 173.76 316.56 Tm +(denote) Tj +1 0 0 1 204.24 316.56 Tm +(each) Tj +1 0 0 1 226.44 316.56 Tm +(operation) Tj +1 0 0 1 267.959 316.56 Tm +(by) Tj +ET +endstream +endobj +720 0 obj +4750 +endobj +721 0 obj +<< /Type /Font /Name /R721 /Subtype /Type1 /BaseFont /Times-Italic >> +endobj +722 0 obj +<< /Length 723 0 R >> +stream +BT +/R721 10 Tf +1 0 0 1 281.759 316.56 Tm +(a/b) Tj +/R715 10 Tf +1 0 0 1 294.599 316.56 Tm +(,) Tj +1 0 0 1 72 304.56 Tm +(where) Tj +/R721 10 Tf +1 0 0 1 98.4 304.56 Tm +(a) Tj +/R715 10 Tf +1 0 0 1 105.48 304.56 Tm +(and) Tj +/R721 10 Tf +1 0 0 1 121.92 304.56 Tm +(b) Tj +/R715 10 Tf +1 0 0 1 128.88 304.56 Tm +(are) Tj +1 0 0 1 143.16 304.56 Tm +(the) Tj +1 0 0 1 157.44 304.56 Tm +(sizes) Tj +1 0 0 1 178.8 304.56 Tm +(of) Tj +1 0 0 1 189.12 304.56 Tm +(the) Tj +1 0 0 1 203.4 304.56 Tm +(operation) Tj +1 0 0 1 243.119 304.56 Tm +(ar) Tj +1 0 0 1 250.799 304.56 Tm +(gument) Tj +1 0 0 1 282.599 304.56 Tm +(and) Tj +1 0 0 1 72 292.68 Tm +(result) Tj +1 0 0 1 96.5999 292.68 Tm +(in) Tj +1 0 0 1 106.92 292.68 Tm +(KBytes.) Tj +ET +Q +q +W +0 0 612 792 re +n +q +204.6 0 0 -0.48 82.188 281.892 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +q +0.48 0 0 -11.04 81.948 281.532 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +/R715 9 Tf +1 0 0 1 88.2 273.72 Tm +(ar) Tj +1 0 0 1 95.0399 273.72 Tm +(g./res.) Tj +ET +q +0.48 0 0 -11.04 122.748 281.532 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +q +0.48 0 0 -11.04 122.748 281.532 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +1 0 0 1 162 273.72 Tm +(replicated) Tj +ET +q +0.48 0 0 -11.04 236.268 281.532 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +1 0 0 1 247.92 273.72 Tm +(without) Tj +ET +q +0.48 0 0 -11.04 286.548 281.532 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +q +113.64 0 0 -0.48 122.868 270.612 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +q +0.48 0 0 -11.04 81.948 270.612 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +1 0 0 1 93.36 262.8 Tm +(\(KB\)) Tj +ET +q +0.48 0 0 -11.04 122.748 270.612 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +1 0 0 1 135.12 262.8 Tm +(read-write) Tj +ET +q +0.48 0 0 -11.04 183.948 270.612 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +1 0 0 1 193.08 262.8 Tm +(read-only) Tj +ET +q +0.48 0 0 -11.04 236.268 270.612 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +1 0 0 1 242.4 262.8 Tm +(replication) Tj +ET +q +0.48 0 0 -11.04 286.548 270.612 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +q +204.6 0 0 -0.48 82.188 259.572 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +q +0.48 0 0 -11.04 81.948 259.212 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +1 0 0 1 96.84 251.4 Tm +(0/0) Tj +ET +q +0.48 0 0 -11.04 122.748 259.212 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +1 0 0 1 131.16 251.4 Tm +(3.35) Tj +1 0 0 1 149.04 251.4 Tm +(\(309%\)) Tj +ET +q +0.48 0 0 -11.04 183.948 259.212 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +1 0 0 1 190.2 251.4 Tm +(1.62) Tj +1 0 0 1 208.08 251.4 Tm +(\(98%\)) Tj +ET +q +0.48 0 0 -11.04 236.268 259.212 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +1 0 0 1 253.8 251.4 Tm +(0.82) Tj +ET +q +0.48 0 0 -11.04 286.548 259.212 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +q +0.48 0 0 -11.04 81.948 248.292 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +1 0 0 1 96.84 240.48 Tm +(4/0) Tj +ET +q +0.48 0 0 -11.04 122.748 248.292 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +1 0 0 1 128.88 240.48 Tm +(14.19) Tj +1 0 0 1 151.32 240.48 Tm +(\(207%\)) Tj +ET +q +0.48 0 0 -11.04 183.948 248.292 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +1 0 0 1 190.2 240.48 Tm +(6.98) Tj +1 0 0 1 208.08 240.48 Tm +(\(51%\)) Tj +ET +q +0.48 0 0 -11.04 236.268 248.292 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +1 0 0 1 253.8 240.48 Tm +(4.62) Tj +ET +q +0.48 0 0 -11.04 286.548 248.292 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +q +0.48 0 0 -11.04 81.948 237.252 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +1 0 0 1 96.84 229.56 Tm +(0/4) Tj +ET +q +0.48 0 0 -11.04 122.748 237.252 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +1 0 0 1 133.32 229.56 Tm +(8.01) Tj +1 0 0 1 151.32 229.56 Tm +(\(72%\)) Tj +ET +q +0.48 0 0 -11.04 183.948 237.252 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +1 0 0 1 190.2 229.56 Tm +(5.94) Tj +1 0 0 1 208.08 229.56 Tm +(\(27%\)) Tj +ET +q +0.48 0 0 -11.04 236.268 237.252 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +1 0 0 1 253.8 229.56 Tm +(4.66) Tj +ET +q +0.48 0 0 -11.04 286.548 237.252 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +q +204.6 0 0 -0.48 82.188 226.332 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +/R715 10 Tf +1 0 0 1 72 212.4 Tm +(T) Tj +1 0 0 1 77.28 212.4 Tm +(able) Tj +1 0 0 1 96.5999 212.4 Tm +(1:) Tj +1 0 0 1 108.36 212.4 Tm +(Micro-benchmark) Tj +1 0 0 1 183 212.4 Tm +(results) Tj +1 0 0 1 211.8 212.4 Tm +(\(in) Tj +1 0 0 1 225.599 212.4 Tm +(milliseconds\);) Tj +1 0 0 1 284.879 212.4 Tm +(the) Tj +1 0 0 1 72 200.4 Tm +(percentage) Tj +1 0 0 1 117.6 200.4 Tm +(o) Tj +1 0 0 1 122.52 200.4 Tm +(v) Tj +1 0 0 1 127.44 200.4 Tm +(erhead) Tj +1 0 0 1 156.36 200.4 Tm +(is) Tj +1 0 0 1 165.48 200.4 Tm +(relati) Tj +1 0 0 1 185.76 200.4 Tm +(v) Tj +1 0 0 1 190.68 200.4 Tm +(e) Tj +1 0 0 1 197.4 200.4 Tm +(to) Tj +1 0 0 1 207.719 200.4 Tm +(the) Tj +1 0 0 1 222.359 200.4 Tm +(unreplicated) Tj +1 0 0 1 274.079 200.4 Tm +(case.) Tj +1 0 0 1 81.96 178.44 Tm +(The) Tj +1 0 0 1 100.68 178.44 Tm +(o) Tj +1 0 0 1 105.6 178.44 Tm +(v) Tj +1 0 0 1 110.52 178.44 Tm +(erhead) Tj +1 0 0 1 140.16 178.44 Tm +(introduced) Tj +1 0 0 1 186 178.44 Tm +(by) Tj +1 0 0 1 199.2 178.44 Tm +(the) Tj +1 0 0 1 214.68 178.44 Tm +(replication) Tj +1 0 0 1 260.519 178.44 Tm +(library) Tj +1 0 0 1 290.399 178.44 Tm +(is) Tj +1 0 0 1 72 166.56 Tm +(due) Tj +1 0 0 1 88.44 166.56 Tm +(to) Tj +1 0 0 1 98.2799 166.56 Tm +(e) Tj +1 0 0 1 102.6 166.56 Tm +(xtra) Tj +1 0 0 1 120.12 166.56 Tm +(computation) Tj +1 0 0 1 172.08 166.56 Tm +(and) Tj +1 0 0 1 188.52 166.56 Tm +(communication.) Tj +1 0 0 1 256.44 166.56 Tm +(F) Tj +1 0 0 1 261.839 166.56 Tm +(or) Tj +1 0 0 1 272.279 166.56 Tm +(e) Tj +1 0 0 1 276.599 166.56 Tm +(xam-) Tj +1 0 0 1 72 154.56 Tm +(ple,) Tj +1 0 0 1 89.3999 154.56 Tm +(the) Tj +1 0 0 1 104.28 154.56 Tm +(computation) Tj +1 0 0 1 156.72 154.56 Tm +(o) Tj +1 0 0 1 161.64 154.56 Tm +(v) Tj +1 0 0 1 166.56 154.56 Tm +(erhead) Tj +1 0 0 1 195.6 154.56 Tm +(for) Tj +1 0 0 1 209.88 154.56 Tm +(the) Tj +1 0 0 1 224.759 154.56 Tm +(read-write) Tj +1 0 0 1 268.319 154.56 Tm +(0/0) Tj +1 0 0 1 283.679 154.56 Tm +(op-) Tj +1 0 0 1 72 142.68 Tm +(eration) Tj +1 0 0 1 102.12 142.68 Tm +(is) Tj +1 0 0 1 111.36 142.68 Tm +(approximately) Tj +1 0 0 1 171.36 142.68 Tm +(1.06ms,) Tj +1 0 0 1 205.44 142.68 Tm +(which) Tj +1 0 0 1 232.2 142.68 Tm +(includes) Tj +1 0 0 1 267.959 142.68 Tm +(0.55ms) Tj +1 0 0 1 72 130.68 Tm +(spent) Tj +1 0 0 1 96.3599 130.68 Tm +(e) Tj +1 0 0 1 100.68 130.68 Tm +(x) Tj +1 0 0 1 105.6 130.68 Tm +(ecuting) Tj +1 0 0 1 138.24 130.68 Tm +(cryptographic) Tj +1 0 0 1 196.92 130.68 Tm +(operations.) Tj +1 0 0 1 247.199 130.68 Tm +(The) Tj +1 0 0 1 266.039 130.68 Tm +(remain-) Tj +1 0 0 1 72 118.8 Tm +(ing) Tj +1 0 0 1 87.2399 118.8 Tm +(1.47ms) Tj +1 0 0 1 118.8 118.8 Tm +(of) Tj +1 0 0 1 129.6 118.8 Tm +(o) Tj +1 0 0 1 134.52 118.8 Tm +(v) Tj +1 0 0 1 139.44 118.8 Tm +(erhead) Tj +1 0 0 1 168.24 118.8 Tm +(are) Tj +1 0 0 1 182.88 118.8 Tm +(due) Tj +1 0 0 1 199.8 118.8 Tm +(to) Tj +1 0 0 1 210 118.8 Tm +(e) Tj +1 0 0 1 214.32 118.8 Tm +(xtra) Tj +1 0 0 1 232.199 118.8 Tm +(communication;) Tj +1 0 0 1 72 106.8 Tm +(the) Tj +1 0 0 1 86.0399 106.8 Tm +(replication) Tj +1 0 0 1 130.44 106.8 Tm +(library) Tj +1 0 0 1 158.76 106.8 Tm +(introduces) Tj +1 0 0 1 202.079 106.8 Tm +(an) Tj +1 0 0 1 213.359 106.8 Tm +(e) Tj +1 0 0 1 217.679 106.8 Tm +(xtra) Tj +1 0 0 1 234.839 106.8 Tm +(message) Tj +1 0 0 1 270.479 106.8 Tm +(round-) Tj +1 0 0 1 72 94.92 Tm +(trip,) Tj +1 0 0 1 90.4799 94.92 Tm +(it) Tj +1 0 0 1 98.0399 94.92 Tm +(sends) Tj +1 0 0 1 122.28 94.92 Tm +(lar) Tj +1 0 0 1 132.72 94.92 Tm +(ger) Tj +1 0 0 1 147.36 94.92 Tm +(messages,) Tj +1 0 0 1 189.72 94.92 Tm +(and) Tj +1 0 0 1 206.16 94.92 Tm +(it) Tj +1 0 0 1 213.72 94.92 Tm +(increases) Tj +1 0 0 1 252.359 94.92 Tm +(the) Tj +1 0 0 1 266.519 94.92 Tm +(number) Tj +1 0 0 1 72 82.92 Tm +(of) Tj +1 0 0 1 82.56 82.92 Tm +(messages) Tj +1 0 0 1 122.52 82.92 Tm +(recei) Tj +1 0 0 1 141.72 82.92 Tm +(v) Tj +1 0 0 1 146.64 82.92 Tm +(ed) Tj +1 0 0 1 158.16 82.92 Tm +(by) Tj +1 0 0 1 170.4 82.92 Tm +(each) Tj +1 0 0 1 190.92 82.92 Tm +(node) Tj +1 0 0 1 212.52 82.92 Tm +(relati) Tj +1 0 0 1 232.8 82.92 Tm +(v) Tj +1 0 0 1 237.72 82.92 Tm +(e) Tj +1 0 0 1 244.319 82.92 Tm +(to) Tj +1 0 0 1 254.279 82.92 Tm +(the) Tj +1 0 0 1 268.799 82.92 Tm +(service) Tj +1 0 0 1 72 71.04 Tm +(without) Tj +1 0 0 1 104.88 71.04 Tm +(replication.) Tj +1 0 0 1 324.96 709.08 Tm +(The) Tj +1 0 0 1 343.2 709.08 Tm +(o) Tj +1 0 0 1 348.12 709.08 Tm +(v) Tj +1 0 0 1 353.04 709.08 Tm +(erhead) Tj +1 0 0 1 382.08 709.08 Tm +(for) Tj +1 0 0 1 396.36 709.08 Tm +(read-only) Tj +1 0 0 1 437.16 709.08 Tm +(operations) Tj +1 0 0 1 481.44 709.08 Tm +(is) Tj +1 0 0 1 490.679 709.08 Tm +(signi\256cantly) Tj +1 0 0 1 315 697.08 Tm +(lo) Tj +1 0 0 1 322.56 697.08 Tm +(wer) Tj +1 0 0 1 340.32 697.08 Tm +(because) Tj +1 0 0 1 374.76 697.08 Tm +(the) Tj +1 0 0 1 389.76 697.08 Tm +(optimization) Tj +1 0 0 1 443.04 697.08 Tm +(discussed) Tj +1 0 0 1 484.079 697.08 Tm +(in) Tj +1 0 0 1 494.759 697.08 Tm +(Section) Tj +1 0 0 1 527.519 697.08 Tm +(5.1) Tj +1 0 0 1 315 685.2 Tm +(reduces) Tj +1 0 0 1 347.16 685.2 Tm +(both) Tj +1 0 0 1 366.6 685.2 Tm +(computation) Tj +1 0 0 1 418.32 685.2 Tm +(and) Tj +1 0 0 1 434.4 685.2 Tm +(commu) Tj +1 0 0 1 464.4 685.2 Tm +(nica) Tj +1 0 0 1 480.959 685.2 Tm +(tion) Tj +1 0 0 1 498.119 685.2 Tm +(o) Tj +1 0 0 1 502.919 685.2 Tm +(v) Tj +1 0 0 1 507.839 685.2 Tm +(er) Tj +1 0 0 1 515.519 685.2 Tm +(hea) Tj +1 0 0 1 529.319 685.2 Tm +(ds.) Tj +1 0 0 1 315 673.2 Tm +(F) Tj +1 0 0 1 320.4 673.2 Tm +(or) Tj +1 0 0 1 330.84 673.2 Tm +(e) Tj +1 0 0 1 335.16 673.2 Tm +(xample,) Tj +1 0 0 1 369.12 673.2 Tm +(the) Tj +1 0 0 1 383.4 673.2 Tm +(computation) Tj +1 0 0 1 435.36 673.2 Tm +(o) Tj +1 0 0 1 440.28 673.2 Tm +(v) Tj +1 0 0 1 445.2 673.2 Tm +(erhead) Tj +1 0 0 1 473.759 673.2 Tm +(for) Tj +1 0 0 1 487.559 673.2 Tm +(the) Tj +1 0 0 1 501.839 673.2 Tm +(read-only) Tj +1 0 0 1 315 661.32 Tm +(0/0) Tj +1 0 0 1 331.2 661.32 Tm +(operation) Tj +1 0 0 1 372.36 661.32 Tm +(is) Tj +1 0 0 1 382.56 661.32 Tm +(approximately) Tj +1 0 0 1 443.64 661.32 Tm +(0.43ms,) Tj +1 0 0 1 478.92 661.32 Tm +(which) Tj +1 0 0 1 506.759 661.32 Tm +(includes) Tj +1 0 0 1 315 649.32 Tm +(0.23ms) Tj +1 0 0 1 348.48 649.32 Tm +(spent) Tj +1 0 0 1 374.04 649.32 Tm +(e) Tj +1 0 0 1 378.36 649.32 Tm +(x) Tj +1 0 0 1 383.28 649.32 Tm +(ecuting) Tj +1 0 0 1 416.88 649.32 Tm +(cryptographic) Tj +1 0 0 1 476.64 649.32 Tm +(operations,) Tj +1 0 0 1 525.599 649.32 Tm +(and) Tj +1 0 0 1 315 637.44 Tm +(the) Tj +1 0 0 1 329.64 637.44 Tm +(communication) Tj +1 0 0 1 394.2 637.44 Tm +(o) Tj +1 0 0 1 399.12 637.44 Tm +(v) Tj +1 0 0 1 404.04 637.44 Tm +(erhead) Tj +1 0 0 1 432.84 637.44 Tm +(is) Tj +1 0 0 1 442.08 637.44 Tm +(only) Tj +1 0 0 1 462.24 637.44 Tm +(0.37ms) Tj +1 0 0 1 493.8 637.44 Tm +(because) Tj +1 0 0 1 527.879 637.44 Tm +(the) Tj +1 0 0 1 315 625.44 Tm +(protocol) Tj +1 0 0 1 352.32 625.44 Tm +(to) Tj +1 0 0 1 364.2 625.44 Tm +(e) Tj +1 0 0 1 368.52 625.44 Tm +(x) Tj +1 0 0 1 373.44 625.44 Tm +(ecute) Tj +1 0 0 1 398.4 625.44 Tm +(read-only) Tj +1 0 0 1 440.76 625.44 Tm +(operations) Tj +1 0 0 1 486.359 625.44 Tm +(uses) Tj +1 0 0 1 507.599 625.44 Tm +(a) Tj +1 0 0 1 516.239 625.44 Tm +(single) Tj +1 0 0 1 315 613.44 Tm +(round-trip.) Tj +1 0 0 1 324.96 599.88 Tm +(T) Tj +1 0 0 1 330.24 599.88 Tm +(able) Tj +1 0 0 1 350.16 599.88 Tm +(1) Tj +1 0 0 1 358.44 599.88 Tm +(sho) Tj +1 0 0 1 372.12 599.88 Tm +(ws) Tj +1 0 0 1 386.52 599.88 Tm +(that) Tj +1 0 0 1 404.76 599.88 Tm +(the) Tj +1 0 0 1 420.24 599.88 Tm +(relati) Tj +1 0 0 1 440.52 599.88 Tm +(v) Tj +1 0 0 1 445.44 599.88 Tm +(e) Tj +1 0 0 1 453.119 599.88 Tm +(o) Tj +1 0 0 1 458.039 599.88 Tm +(v) Tj +1 0 0 1 462.959 599.88 Tm +(erhead) Tj +1 0 0 1 492.599 599.88 Tm +(is) Tj +1 0 0 1 502.559 599.88 Tm +(lo) Tj +1 0 0 1 510.119 599.88 Tm +(wer) Tj +1 0 0 1 528.359 599.88 Tm +(for) Tj +1 0 0 1 315 587.88 Tm +(the) Tj +1 0 0 1 330 587.88 Tm +(4/0) Tj +1 0 0 1 345.6 587.88 Tm +(and) Tj +1 0 0 1 362.76 587.88 Tm +(0/4) Tj +1 0 0 1 378.36 587.88 Tm +(operations.) Tj +1 0 0 1 426.84 587.88 Tm +(This) Tj +1 0 0 1 447.48 587.88 Tm +(is) Tj +1 0 0 1 456.84 587.88 Tm +(because) Tj +1 0 0 1 491.279 587.88 Tm +(a) Tj +1 0 0 1 498.479 587.88 Tm +(signi\256cant) Tj +1 0 0 1 315 576 Tm +(fraction) Tj +1 0 0 1 350.28 576 Tm +(of) Tj +1 0 0 1 362.88 576 Tm +(the) Tj +1 0 0 1 379.32 576 Tm +(o) Tj +1 0 0 1 384.24 576 Tm +(v) Tj +1 0 0 1 389.16 576 Tm +(erhead) Tj +1 0 0 1 419.76 576 Tm +(introduced) Tj +1 0 0 1 466.679 576 Tm +(by) Tj +1 0 0 1 480.959 576 Tm +(the) Tj +1 0 0 1 497.399 576 Tm +(replication) Tj +1 0 0 1 315 564 Tm +(library) Tj +1 0 0 1 344.04 564 Tm +(is) Tj +1 0 0 1 353.16 564 Tm +(independent) Tj +1 0 0 1 404.4 564 Tm +(of) Tj +1 0 0 1 415.2 564 Tm +(the) Tj +1 0 0 1 429.84 564 Tm +(size) Tj +1 0 0 1 447.719 564 Tm +(of) Tj +1 0 0 1 458.519 564 Tm +(operation) Tj +1 0 0 1 498.719 564 Tm +(ar) Tj +1 0 0 1 506.399 564 Tm +(guments) Tj +1 0 0 1 315 552.12 Tm +(and) Tj +1 0 0 1 331.56 552.12 Tm +(results.) Tj +1 0 0 1 363.48 552.12 Tm +(F) Tj +1 0 0 1 368.88 552.12 Tm +(or) Tj +1 0 0 1 379.32 552.12 Tm +(e) Tj +1 0 0 1 383.64 552.12 Tm +(xample,) Tj +1 0 0 1 417.72 552.12 Tm +(in) Tj +1 0 0 1 427.56 552.12 Tm +(the) Tj +1 0 0 1 441.96 552.12 Tm +(read-write) Tj +1 0 0 1 485.039 552.12 Tm +(0/4) Tj +1 0 0 1 499.919 552.12 Tm +(operation,) Tj +1 0 0 1 315 540.12 Tm +(the) Tj +1 0 0 1 332.16 540.12 Tm +(lar) Tj +1 0 0 1 342.6 540.12 Tm +(ge) Tj +1 0 0 1 356.88 540.12 Tm +(message) Tj +1 0 0 1 395.52 540.12 Tm +(\(the) Tj +1 0 0 1 416.04 540.12 Tm +(reply\)) Tj +1 0 0 1 444.72 540.12 Tm +(goes) Tj +1 0 0 1 468 540.12 Tm +(o) Tj +1 0 0 1 472.92 540.12 Tm +(v) Tj +1 0 0 1 477.84 540.12 Tm +(er) Tj +1 0 0 1 490.319 540.12 Tm +(the) Tj +1 0 0 1 507.479 540.12 Tm +(netw) Tj +1 0 0 1 526.799 540.12 Tm +(ork) Tj +1 0 0 1 315 528.24 Tm +(only) Tj +1 0 0 1 337.32 528.24 Tm +(once) Tj +1 0 0 1 360.6 528.24 Tm +(\(as) Tj +1 0 0 1 376.8 528.24 Tm +(discussed) Tj +1 0 0 1 419.52 528.24 Tm +(in) Tj +1 0 0 1 431.88 528.24 Tm +(Section) Tj +1 0 0 1 466.319 528.24 Tm +(5.1\)) Tj +1 0 0 1 486.599 528.24 Tm +(and) Tj +1 0 0 1 505.559 528.24 Tm +(only) Tj +1 0 0 1 527.879 528.24 Tm +(the) Tj +1 0 0 1 315 516.24 Tm +(cryptographic) Tj +1 0 0 1 373.56 516.24 Tm +(o) Tj +1 0 0 1 378.48 516.24 Tm +(v) Tj +1 0 0 1 383.4 516.24 Tm +(erhead) Tj +1 0 0 1 413.04 516.24 Tm +(to) Tj +1 0 0 1 424.08 516.24 Tm +(process) Tj +1 0 0 1 457.2 516.24 Tm +(the) Tj +1 0 0 1 472.679 516.24 Tm +(reply) Tj +1 0 0 1 496.319 516.24 Tm +(message) Tj +1 0 0 1 533.399 516.24 Tm +(is) Tj +1 0 0 1 315 504.36 Tm +(increased.) Tj +1 0 0 1 359.52 504.36 Tm +(The) Tj +1 0 0 1 377.76 504.36 Tm +(o) Tj +1 0 0 1 382.68 504.36 Tm +(v) Tj +1 0 0 1 387.6 504.36 Tm +(erhead) Tj +1 0 0 1 416.64 504.36 Tm +(is) Tj +1 0 0 1 426.12 504.36 Tm +(higher) Tj +1 0 0 1 454.319 504.36 Tm +(for) Tj +1 0 0 1 468.599 504.36 Tm +(the) Tj +1 0 0 1 483.599 504.36 Tm +(read-write) Tj +1 0 0 1 527.279 504.36 Tm +(4/0) Tj +1 0 0 1 315 492.36 Tm +(operation) Tj +1 0 0 1 356.16 492.36 Tm +(because) Tj +1 0 0 1 391.08 492.36 Tm +(the) Tj +1 0 0 1 406.8 492.36 Tm +(lar) Tj +1 0 0 1 417.24 492.36 Tm +(ge) Tj +1 0 0 1 429.96 492.36 Tm +(message) Tj +1 0 0 1 467.16 492.36 Tm +(\(the) Tj +1 0 0 1 486.119 492.36 Tm +(request\)) Tj +1 0 0 1 521.759 492.36 Tm +(goes) Tj +1 0 0 1 315 480.48 Tm +(o) Tj +1 0 0 1 319.92 480.48 Tm +(v) Tj +1 0 0 1 324.84 480.48 Tm +(er) Tj +1 0 0 1 335.64 480.48 Tm +(the) Tj +1 0 0 1 351.12 480.48 Tm +(netw) Tj +1 0 0 1 370.44 480.48 Tm +(ork) Tj +1 0 0 1 386.88 480.48 Tm +(twice) Tj +1 0 0 1 411.84 480.48 Tm +(and) Tj +1 0 0 1 429.48 480.48 Tm +(increases) Tj +1 0 0 1 469.199 480.48 Tm +(the) Tj +1 0 0 1 484.679 480.48 Tm +(cryptographic) Tj +1 0 0 1 315 468.48 Tm +(o) Tj +1 0 0 1 319.92 468.48 Tm +(v) Tj +1 0 0 1 324.84 468.48 Tm +(erhead) Tj +1 0 0 1 355.8 468.48 Tm +(for) Tj +1 0 0 1 372.12 468.48 Tm +(processing) Tj +1 0 0 1 419.28 468.48 Tm +(both) Tj +1 0 0 1 441.72 468.48 Tm +(request) Tj +1 0 0 1 475.079 468.48 Tm +(and) Tj +1 0 0 1 494.039 468.48 Tm +(pre-prepare) Tj +1 0 0 1 315 456.6 Tm +(messages.) Tj +1 0 0 1 324.96 442.92 Tm +(It) Tj +1 0 0 1 336.6 442.92 Tm +(is) Tj +1 0 0 1 348.72 442.92 Tm +(important) Tj +1 0 0 1 392.88 442.92 Tm +(to) Tj +1 0 0 1 406.2 442.92 Tm +(note) Tj +1 0 0 1 428.88 442.92 Tm +(that) Tj +1 0 0 1 449.279 442.92 Tm +(this) Tj +1 0 0 1 469.199 442.92 Tm +(micro-benchmark) Tj +1 0 0 1 315 431.04 Tm +(represents) Tj +1 0 0 1 360.12 431.04 Tm +(the) Tj +1 0 0 1 377.16 431.04 Tm +(w) Tj +1 0 0 1 384.24 431.04 Tm +(orst) Tj +1 0 0 1 403.92 431.04 Tm +(case) Tj +1 0 0 1 425.88 431.04 Tm +(o) Tj +1 0 0 1 430.8 431.04 Tm +(v) Tj +1 0 0 1 435.72 431.04 Tm +(erhead) Tj +1 0 0 1 466.799 431.04 Tm +(for) Tj +1 0 0 1 483.239 431.04 Tm +(our) Tj +1 0 0 1 501.239 431.04 Tm +(algorithm) Tj +1 0 0 1 315 419.04 Tm +(because) Tj +1 0 0 1 353.76 419.04 Tm +(the) Tj +1 0 0 1 373.2 419.04 Tm +(operations) Tj +1 0 0 1 422.04 419.04 Tm +(perform) Tj +1 0 0 1 461.28 419.04 Tm +(no) Tj +1 0 0 1 478.559 419.04 Tm +(w) Tj +1 0 0 1 485.639 419.04 Tm +(ork) Tj +1 0 0 1 506.159 419.04 Tm +(and) Tj +1 0 0 1 527.879 419.04 Tm +(the) Tj +1 0 0 1 315 407.04 Tm +(unreplicated) Tj +1 0 0 1 371.04 407.04 Tm +(serv) Tj +1 0 0 1 387.6 407.04 Tm +(er) Tj +1 0 0 1 401.88 407.04 Tm +(pro) Tj +1 0 0 1 415.2 407.04 Tm +(vides) Tj +1 0 0 1 442.8 407.04 Tm +(v) Tj +1 0 0 1 447.72 407.04 Tm +(ery) Tj +1 0 0 1 467.159 407.04 Tm +(weak) Tj +1 0 0 1 494.879 407.04 Tm +(guarantees.) Tj +1 0 0 1 315 395.16 Tm +(Most) Tj +1 0 0 1 341.04 395.16 Tm +(services) Tj +1 0 0 1 378.72 395.16 Tm +(will) Tj +1 0 0 1 399.72 395.16 Tm +(require) Tj +1 0 0 1 433.56 395.16 Tm +(stronger) Tj +1 0 0 1 471.72 395.16 Tm +(guarantees,) Tj +1 0 0 1 523.079 395.16 Tm +(e.g.,) Tj +1 0 0 1 315 383.16 Tm +(authenticated) Tj +1 0 0 1 371.4 383.16 Tm +(connections,) Tj +1 0 0 1 424.92 383.16 Tm +(and) Tj +1 0 0 1 442.56 383.16 Tm +(the) Tj +1 0 0 1 457.92 383.16 Tm +(o) Tj +1 0 0 1 462.84 383.16 Tm +(v) Tj +1 0 0 1 467.76 383.16 Tm +(erhead) Tj +1 0 0 1 497.399 383.16 Tm +(introduced) Tj +1 0 0 1 315 371.28 Tm +(by) Tj +1 0 0 1 326.64 371.28 Tm +(our) Tj +1 0 0 1 341.64 371.28 Tm +(algorithm) Tj +1 0 0 1 382.08 371.28 Tm +(relati) Tj +1 0 0 1 402.36 371.28 Tm +(v) Tj +1 0 0 1 407.28 371.28 Tm +(e) Tj +1 0 0 1 413.28 371.28 Tm +(to) Tj +1 0 0 1 422.64 371.28 Tm +(a) Tj +1 0 0 1 428.76 371.28 Tm +(serv) Tj +1 0 0 1 445.32 371.28 Tm +(er) Tj +1 0 0 1 454.679 371.28 Tm +(that) Tj +1 0 0 1 471.359 371.28 Tm +(implements) Tj +1 0 0 1 519.479 371.28 Tm +(these) Tj +1 0 0 1 315 359.28 Tm +(guarantees) Tj +1 0 0 1 361.92 359.28 Tm +(will) Tj +1 0 0 1 381.6 359.28 Tm +(be) Tj +1 0 0 1 395.28 359.28 Tm +(lo) Tj +1 0 0 1 402.84 359.28 Tm +(wer) Tj +1 0 0 1 417.24 359.28 Tm +(.) Tj +1 0 0 1 428.64 359.28 Tm +(F) Tj +1 0 0 1 434.04 359.28 Tm +(or) Tj +1 0 0 1 446.52 359.28 Tm +(e) Tj +1 0 0 1 450.84 359.28 Tm +(xample,) Tj +1 0 0 1 487.319 359.28 Tm +(the) Tj +1 0 0 1 503.759 359.28 Tm +(o) Tj +1 0 0 1 508.679 359.28 Tm +(v) Tj +1 0 0 1 513.599 359.28 Tm +(erhead) Tj +1 0 0 1 315 347.4 Tm +(of) Tj +1 0 0 1 328.2 347.4 Tm +(the) Tj +1 0 0 1 345.24 347.4 Tm +(replication) Tj +1 0 0 1 392.76 347.4 Tm +(library) Tj +1 0 0 1 424.2 347.4 Tm +(relati) Tj +1 0 0 1 444.479 347.4 Tm +(v) Tj +1 0 0 1 449.4 347.4 Tm +(e) Tj +1 0 0 1 458.64 347.4 Tm +(to) Tj +1 0 0 1 471.239 347.4 Tm +(a) Tj +1 0 0 1 480.599 347.4 Tm +(v) Tj +1 0 0 1 485.519 347.4 Tm +(ersion) Tj +1 0 0 1 514.679 347.4 Tm +(of) Tj +1 0 0 1 527.879 347.4 Tm +(the) Tj +1 0 0 1 315 335.4 Tm +(unreplicated) Tj +1 0 0 1 367.8 335.4 Tm +(service) Tj +1 0 0 1 399.48 335.4 Tm +(that) Tj +1 0 0 1 417.96 335.4 Tm +(uses) Tj +1 0 0 1 438.72 335.4 Tm +(MA) Tj +1 0 0 1 454.44 335.4 Tm +(Cs) Tj +1 0 0 1 468.359 335.4 Tm +(for) Tj +1 0 0 1 483.599 335.4 Tm +(authentication) Tj +1 0 0 1 315 323.52 Tm +(is) Tj +1 0 0 1 324 323.52 Tm +(only) Tj +1 0 0 1 343.92 323.52 Tm +(243%) Tj +1 0 0 1 369.48 323.52 Tm +(for) Tj +1 0 0 1 383.4 323.52 Tm +(the) Tj +1 0 0 1 397.92 323.52 Tm +(read-write) Tj +1 0 0 1 441.12 323.52 Tm +(0/0) Tj +1 0 0 1 456.24 323.52 Tm +(operation) Tj +1 0 0 1 496.079 323.52 Tm +(and) Tj +1 0 0 1 512.759 323.52 Tm +(4%) Tj +1 0 0 1 528.359 323.52 Tm +(for) Tj +1 0 0 1 315 311.52 Tm +(the) Tj +1 0 0 1 329.64 311.52 Tm +(read-only) Tj +1 0 0 1 370.32 311.52 Tm +(4/0) Tj +1 0 0 1 385.56 311.52 Tm +(operation.) Tj +1 0 0 1 324.96 297.96 Tm +(W) Tj +1 0 0 1 333.48 297.96 Tm +(e) Tj +1 0 0 1 345.24 297.96 Tm +(can) Tj +1 0 0 1 366.24 297.96 Tm +(estimate) Tj +1 0 0 1 406.68 297.96 Tm +(a) Tj +1 0 0 1 418.32 297.96 Tm +(rough) Tj +1 0 0 1 448.8 297.96 Tm +(lo) Tj +1 0 0 1 456.36 297.96 Tm +(wer) Tj +1 0 0 1 478.44 297.96 Tm +(bound) Tj +1 0 0 1 510.599 297.96 Tm +(on) Tj +1 0 0 1 527.879 297.96 Tm +(the) Tj +1 0 0 1 315 285.96 Tm +(performance) Tj +1 0 0 1 368.88 285.96 Tm +(gain) Tj +1 0 0 1 389.52 285.96 Tm +(af) Tj +1 0 0 1 397.08 285.96 Tm +(forded) Tj +1 0 0 1 426.6 285.96 Tm +(by) Tj +1 0 0 1 440.04 285.96 Tm +(our) Tj +1 0 0 1 456.96 285.96 Tm +(algorithm) Tj +1 0 0 1 499.199 285.96 Tm +(relati) Tj +1 0 0 1 519.479 285.96 Tm +(v) Tj +1 0 0 1 524.399 285.96 Tm +(e) Tj +1 0 0 1 532.199 285.96 Tm +(to) Tj +1 0 0 1 315 274.08 Tm +(Rampart) Tj +1 0 0 1 352.08 274.08 Tm +([30) Tj +1 0 0 1 365.4 274.08 Tm +(].) Tj +1 0 0 1 375.72 274.08 Tm +(Reiter) Tj +1 0 0 1 402.84 274.08 Tm +(reports) Tj +1 0 0 1 433.32 274.08 Tm +(that) Tj +1 0 0 1 450.96 274.08 Tm +(Rampart) Tj +1 0 0 1 488.039 274.08 Tm +(has) Tj +1 0 0 1 504.119 274.08 Tm +(a) Tj +1 0 0 1 511.319 274.08 Tm +(latenc) Tj +1 0 0 1 535.079 274.08 Tm +(y) Tj +1 0 0 1 315 262.08 Tm +(of) Tj +1 0 0 1 325.44 262.08 Tm +(45ms) Tj +1 0 0 1 349.2 262.08 Tm +(for) Tj +1 0 0 1 363 262.08 Tm +(a) Tj +1 0 0 1 369.48 262.08 Tm +(multi-RPC) Tj +1 0 0 1 414.84 262.08 Tm +(of) Tj +1 0 0 1 425.28 262.08 Tm +(a) Tj +1 0 0 1 431.88 262.08 Tm +(null) Tj +1 0 0 1 449.519 262.08 Tm +(message) Tj +1 0 0 1 485.399 262.08 Tm +(in) Tj +1 0 0 1 495.359 262.08 Tm +(a) Tj +1 0 0 1 501.839 262.08 Tm +(10) Tj +1 0 0 1 513.959 262.08 Tm +(Mbit/s) Tj +1 0 0 1 315 250.2 Tm +(Ethernet) Tj +1 0 0 1 351.36 250.2 Tm +(netw) Tj +1 0 0 1 370.68 250.2 Tm +(ork) Tj +1 0 0 1 386.4 250.2 Tm +(of) Tj +1 0 0 1 397.32 250.2 Tm +(4) Tj +1 0 0 1 404.88 250.2 Tm +(SparcStation) Tj +1 0 0 1 458.28 250.2 Tm +(10s) Tj +1 0 0 1 474.719 250.2 Tm +([30) Tj +1 0 0 1 487.919 250.2 Tm +(].) Tj +1 0 0 1 497.639 250.2 Tm +(The) Tj +1 0 0 1 515.639 250.2 Tm +(multi-) Tj +1 0 0 1 315 238.2 Tm +(RPC) Tj +1 0 0 1 335.52 238.2 Tm +(is) Tj +1 0 0 1 343.8 238.2 Tm +(suf) Tj +1 0 0 1 355.8 238.2 Tm +(\256cient) Tj +1 0 0 1 382.32 238.2 Tm +(for) Tj +1 0 0 1 395.64 238.2 Tm +(the) Tj +1 0 0 1 409.44 238.2 Tm +(primary) Tj +1 0 0 1 442.679 238.2 Tm +(to) Tj +1 0 0 1 452.159 238.2 Tm +(in) Tj +1 0 0 1 459.599 238.2 Tm +(v) Tj +1 0 0 1 464.399 238.2 Tm +(ok) Tj +1 0 0 1 474.359 238.2 Tm +(e) Tj +1 0 0 1 480.239 238.2 Tm +(a) Tj +1 0 0 1 486.359 238.2 Tm +(state) Tj +1 0 0 1 506.279 238.2 Tm +(machine) Tj +1 0 0 1 315 226.32 Tm +(operation) Tj +1 0 0 1 354.48 226.32 Tm +(b) Tj +1 0 0 1 359.28 226.32 Tm +(ut) Tj +1 0 0 1 368.76 226.32 Tm +(for) Tj +1 0 0 1 382.2 226.32 Tm +(an) Tj +1 0 0 1 393.36 226.32 Tm +(arbitrary) Tj +1 0 0 1 429.479 226.32 Tm +(client) Tj +1 0 0 1 453.359 226.32 Tm +(to) Tj +1 0 0 1 462.959 226.32 Tm +(in) Tj +1 0 0 1 470.399 226.32 Tm +(v) Tj +1 0 0 1 475.199 226.32 Tm +(ok) Tj +1 0 0 1 485.159 226.32 Tm +(e) Tj +1 0 0 1 491.159 226.32 Tm +(an) Tj +1 0 0 1 502.319 226.32 Tm +(operation) Tj +1 0 0 1 315 214.32 Tm +(it) Tj +1 0 0 1 323.04 214.32 Tm +(w) Tj +1 0 0 1 330.12 214.32 Tm +(ould) Tj +1 0 0 1 350.4 214.32 Tm +(be) Tj +1 0 0 1 362.4 214.32 Tm +(necessary) Tj +1 0 0 1 403.68 214.32 Tm +(to) Tj +1 0 0 1 414 214.32 Tm +(add) Tj +1 0 0 1 430.92 214.32 Tm +(an) Tj +1 0 0 1 442.92 214.32 Tm +(e) Tj +1 0 0 1 447.24 214.32 Tm +(xtra) Tj +1 0 0 1 465.239 214.32 Tm +(message) Tj +1 0 0 1 501.479 214.32 Tm +(delay) Tj +1 0 0 1 525.599 214.32 Tm +(and) Tj +1 0 0 1 315 202.32 Tm +(an) Tj +1 0 0 1 327.84 202.32 Tm +(e) Tj +1 0 0 1 332.16 202.32 Tm +(xtra) Tj +1 0 0 1 351.12 202.32 Tm +(RSA) Tj +1 0 0 1 374.04 202.32 Tm +(signature) Tj +1 0 0 1 414 202.32 Tm +(and) Tj +1 0 0 1 431.88 202.32 Tm +(v) Tj +1 0 0 1 436.8 202.32 Tm +(eri\256cation) Tj +1 0 0 1 480.599 202.32 Tm +(to) Tj +1 0 0 1 491.879 202.32 Tm +(authenticate) Tj +1 0 0 1 315 190.44 Tm +(the) Tj +1 0 0 1 330.36 190.44 Tm +(client;) Tj +1 0 0 1 358.92 190.44 Tm +(this) Tj +1 0 0 1 376.56 190.44 Tm +(w) Tj +1 0 0 1 383.64 190.44 Tm +(ould) Tj +1 0 0 1 404.64 190.44 Tm +(lead) Tj +1 0 0 1 424.44 190.44 Tm +(to) Tj +1 0 0 1 435.48 190.44 Tm +(a) Tj +1 0 0 1 443.04 190.44 Tm +(latenc) Tj +1 0 0 1 466.799 190.44 Tm +(y) Tj +1 0 0 1 474.959 190.44 Tm +(of) Tj +1 0 0 1 486.479 190.44 Tm +(at) Tj +1 0 0 1 496.919 190.44 Tm +(least) Tj +1 0 0 1 518.399 190.44 Tm +(65ms) Tj +1 0 0 1 315 178.44 Tm +(\(using) Tj +1 0 0 1 343.68 178.44 Tm +(the) Tj +1 0 0 1 359.52 178.44 Tm +(RSA) Tj +1 0 0 1 382.68 178.44 Tm +(timings) Tj +1 0 0 1 416.28 178.44 Tm +(reported) Tj +1 0 0 1 453.24 178.44 Tm +(in) Tj +1 0 0 1 464.759 178.44 Tm +([29) Tj +1 0 0 1 478.079 178.44 Tm +(].\)) Tj +1 0 0 1 494.519 178.44 Tm +(Ev) Tj +1 0 0 1 505.559 178.44 Tm +(en) Tj +1 0 0 1 518.519 178.44 Tm +(if) Tj +1 0 0 1 528.359 178.44 Tm +(we) Tj +1 0 0 1 315 166.56 Tm +(di) Tj +1 0 0 1 322.56 166.56 Tm +(vide) Tj +1 0 0 1 343.8 166.56 Tm +(this) Tj +1 0 0 1 362.28 166.56 Tm +(latenc) Tj +1 0 0 1 386.04 166.56 Tm +(y) Tj +1 0 0 1 394.92 166.56 Tm +(by) Tj +1 0 0 1 408.96 166.56 Tm +(1.7,) Tj +1 0 0 1 428.4 166.56 Tm +(the) Tj +1 0 0 1 444.72 166.56 Tm +(ratio) Tj +1 0 0 1 467.039 166.56 Tm +(of) Tj +1 0 0 1 479.399 166.56 Tm +(the) Tj +1 0 0 1 495.719 166.56 Tm +(SPECint92) Tj +1 0 0 1 315 154.56 Tm +(ratings) Tj +1 0 0 1 343.92 154.56 Tm +(of) Tj +1 0 0 1 354.12 154.56 Tm +(the) Tj +1 0 0 1 368.04 154.56 Tm +(DEC) Tj +1 0 0 1 389.88 154.56 Tm +(3000/400) Tj +1 0 0 1 429.36 154.56 Tm +(and) Tj +1 0 0 1 445.56 154.56 Tm +(the) Tj +1 0 0 1 459.599 154.56 Tm +(SparcStation) Tj +1 0 0 1 512.279 154.56 Tm +(10,) Tj +1 0 0 1 526.679 154.56 Tm +(our) Tj +1 0 0 1 315 142.68 Tm +(algorithm) Tj +1 0 0 1 355.56 142.68 Tm +(still) Tj +1 0 0 1 372 142.68 Tm +(reduces) Tj +1 0 0 1 404.16 142.68 Tm +(the) Tj +1 0 0 1 417.96 142.68 Tm +(latenc) Tj +1 0 0 1 441.72 142.68 Tm +(y) Tj +1 0 0 1 448.319 142.68 Tm +(to) Tj +1 0 0 1 457.679 142.68 Tm +(in) Tj +1 0 0 1 465.119 142.68 Tm +(v) Tj +1 0 0 1 469.919 142.68 Tm +(ok) Tj +1 0 0 1 479.879 142.68 Tm +(e) Tj +1 0 0 1 485.879 142.68 Tm +(the) Tj +1 0 0 1 499.679 142.68 Tm +(read-) Tj +1 0 0 1 520.199 142.68 Tm +(write) Tj +1 0 0 1 315 130.68 Tm +(and) Tj +1 0 0 1 332.16 130.68 Tm +(read-only) Tj +1 0 0 1 373.2 130.68 Tm +(0/0) Tj +1 0 0 1 388.8 130.68 Tm +(operations) Tj +1 0 0 1 433.08 130.68 Tm +(by) Tj +1 0 0 1 445.8 130.68 Tm +(f) Tj +1 0 0 1 449.04 130.68 Tm +(actors) Tj +1 0 0 1 475.679 130.68 Tm +(of) Tj +1 0 0 1 486.719 130.68 Tm +(more) Tj +1 0 0 1 510.119 130.68 Tm +(than) Tj +1 0 0 1 530.039 130.68 Tm +(10) Tj +1 0 0 1 315 118.8 Tm +(and) Tj +1 0 0 1 331.08 118.8 Tm +(20,) Tj +1 0 0 1 345.36 118.8 Tm +(respecti) Tj +1 0 0 1 376.2 118.8 Tm +(v) Tj +1 0 0 1 381.12 118.8 Tm +(ely) Tj +1 0 0 1 392.76 118.8 Tm +(.) Tj +1 0 0 1 398.28 118.8 Tm +(Note) Tj +1 0 0 1 419.4 118.8 Tm +(that) Tj +1 0 0 1 435.96 118.8 Tm +(this) Tj +1 0 0 1 452.04 118.8 Tm +(scaling) Tj +1 0 0 1 481.919 118.8 Tm +(is) Tj +1 0 0 1 490.319 118.8 Tm +(conserv) Tj +1 0 0 1 521.279 118.8 Tm +(ati) Tj +1 0 0 1 530.999 118.8 Tm +(v) Tj +1 0 0 1 535.919 118.8 Tm +(e) Tj +1 0 0 1 315 106.8 Tm +(because) Tj +1 0 0 1 350.16 106.8 Tm +(the) Tj +1 0 0 1 366 106.8 Tm +(netw) Tj +1 0 0 1 385.32 106.8 Tm +(ork) Tj +1 0 0 1 402.12 106.8 Tm +(accounts) Tj +1 0 0 1 440.64 106.8 Tm +(for) Tj +1 0 0 1 455.88 106.8 Tm +(a) Tj +1 0 0 1 463.92 106.8 Tm +(signi\256cant) Tj +1 0 0 1 509.039 106.8 Tm +(fraction) Tj +1 0 0 1 315 94.92 Tm +(of) Tj +1 0 0 1 327.6 94.92 Tm +(Rampart') Tj +1 0 0 1 364.8 94.92 Tm +(s) Tj +1 0 0 1 372.72 94.92 Tm +(latenc) Tj +1 0 0 1 396.48 94.92 Tm +(y) Tj +1 0 0 1 405.6 94.92 Tm +([29) Tj +1 0 0 1 418.92 94.92 Tm +(]) Tj +1 0 0 1 426.48 94.92 Tm +(and) Tj +1 0 0 1 445.2 94.92 Tm +(Rampart') Tj +1 0 0 1 482.399 94.92 Tm +(s) Tj +1 0 0 1 490.319 94.92 Tm +(results) Tj +1 0 0 1 520.679 94.92 Tm +(were) Tj +1 0 0 1 315 82.92 Tm +(obtained) Tj +1 0 0 1 352.56 82.92 Tm +(using) Tj +1 0 0 1 377.28 82.92 Tm +(300-bit) Tj +1 0 0 1 409.32 82.92 Tm +(modulus) Tj +1 0 0 1 446.76 82.92 Tm +(RSA) Tj +1 0 0 1 469.44 82.92 Tm +(signatures,) Tj +1 0 0 1 515.639 82.92 Tm +(which) Tj +1 0 0 1 315 71.04 Tm +(are) Tj +1 0 0 1 330.36 71.04 Tm +(not) Tj +1 0 0 1 346.2 71.04 Tm +(considered) Tj +1 0 0 1 392.52 71.04 Tm +(secure) Tj +1 0 0 1 421.08 71.04 Tm +(today) Tj +1 0 0 1 446.28 71.04 Tm +(unless) Tj +1 0 0 1 474.359 71.04 Tm +(the) Tj +1 0 0 1 489.719 71.04 Tm +(k) Tj +1 0 0 1 494.639 71.04 Tm +(e) Tj +1 0 0 1 498.959 71.04 Tm +(ys) Tj +1 0 0 1 510.839 71.04 Tm +(used) Tj +1 0 0 1 532.199 71.04 Tm +(to) Tj +1 0 0 1 300.96 35.16 Tm +(11) Tj +ET +Q +endstream +endobj +723 0 obj +28018 +endobj +724 0 obj +<< +/Type /Page +/MediaBox [0 0 612 792] +/Parent 2 0 R +/Resources << /ProcSet [/PDF /ImageB /Text] +/Font << +/R721 721 0 R +/R718 718 0 R +/R715 715 0 R +>> +>> +/Contents [ +716 0 R +719 0 R +722 0 R +] +>> +endobj +725 0 obj +<< /Type /Font /Name /R725 /Subtype /Type1 /BaseFont /Times-Roman >> +endobj +726 0 obj +<< /Length 727 0 R >> +stream +q +BT +/R725 10 Tf +1 0 0 1 72 709.08 Tm +(generate) Tj +1 0 0 1 108.24 709.08 Tm +(them) Tj +1 0 0 1 130.68 709.08 Tm +(are) Tj +1 0 0 1 145.32 709.08 Tm +(refreshed) Tj +1 0 0 1 184.8 709.08 Tm +(v) Tj +1 0 0 1 189.72 709.08 Tm +(ery) Tj +1 0 0 1 204.96 709.08 Tm +(frequently) Tj +1 0 0 1 245.639 709.08 Tm +(.) Tj +1 0 0 1 81.96 696.72 Tm +(There) Tj +1 0 0 1 111 696.72 Tm +(are) Tj +1 0 0 1 128.88 696.72 Tm +(no) Tj +1 0 0 1 144.6 696.72 Tm +(published) Tj +1 0 0 1 189.12 696.72 Tm +(performance) Tj +1 0 0 1 245.28 696.72 Tm +(numbers) Tj +1 0 0 1 285.359 696.72 Tm +(for) Tj +1 0 0 1 72 684.72 Tm +(SecureRing) Tj +1 0 0 1 122.52 684.72 Tm +([16) Tj +1 0 0 1 135.72 684.72 Tm +(]) Tj +1 0 0 1 143.04 684.72 Tm +(b) Tj +1 0 0 1 147.84 684.72 Tm +(ut) Tj +1 0 0 1 159.6 684.72 Tm +(it) Tj +1 0 0 1 169.08 684.72 Tm +(w) Tj +1 0 0 1 176.16 684.72 Tm +(ould) Tj +1 0 0 1 197.88 684.72 Tm +(be) Tj +1 0 0 1 211.2 684.72 Tm +(slo) Tj +1 0 0 1 222.6 684.72 Tm +(wer) Tj +1 0 0 1 241.56 684.72 Tm +(than) Tj +1 0 0 1 262.679 684.72 Tm +(Rampart) Tj +1 0 0 1 72 672.72 Tm +(because) Tj +1 0 0 1 108.96 672.72 Tm +(its) Tj +1 0 0 1 123.84 672.72 Tm +(algorithm) Tj +1 0 0 1 168 672.72 Tm +(has) Tj +1 0 0 1 186.72 672.72 Tm +(more) Tj +1 0 0 1 212.52 672.72 Tm +(message) Tj +1 0 0 1 251.76 672.72 Tm +(delays) Tj +1 0 0 1 282.599 672.72 Tm +(and) Tj +1 0 0 1 72 660.84 Tm +(signature) Tj +1 0 0 1 111 660.84 Tm +(operations) Tj +1 0 0 1 155.04 660.84 Tm +(in) Tj +1 0 0 1 165.24 660.84 Tm +(the) Tj +1 0 0 1 179.88 660.84 Tm +(critical) Tj +1 0 0 1 210.12 660.84 Tm +(path.) Tj +ET +endstream +endobj +727 0 obj +1496 +endobj +728 0 obj +<< /Type /Font /Name /R728 /Subtype /Type1 /BaseFont /Times-Bold >> +endobj +729 0 obj +<< /Length 730 0 R >> +stream +BT +/R728 10 Tf +1 0 0 1 72 637.56 Tm +(7.3) Tj +1 0 0 1 94.4399 637.56 Tm +(Andr) Tj +1 0 0 1 117 637.56 Tm +(ew) Tj +1 0 0 1 131.04 637.56 Tm +(Benchmark) Tj +/R725 10 Tf +1 0 0 1 72 622.56 Tm +(The) Tj +1 0 0 1 94.5599 622.56 Tm +(Andre) Tj +1 0 0 1 119.4 622.56 Tm +(w) Tj +1 0 0 1 133.56 622.56 Tm +(benchmark) Tj +1 0 0 1 184.92 622.56 Tm +([15) Tj +1 0 0 1 198.24 622.56 Tm +(]) Tj +1 0 0 1 208.68 622.56 Tm +(emulates) Tj +1 0 0 1 251.28 622.56 Tm +(a) Tj +1 0 0 1 262.8 622.56 Tm +(softw) Tj +1 0 0 1 284.879 622.56 Tm +(are) Tj +1 0 0 1 72 610.56 Tm +(de) Tj +1 0 0 1 81.24 610.56 Tm +(v) Tj +1 0 0 1 86.16 610.56 Tm +(elopment) Tj +1 0 0 1 126.72 610.56 Tm +(w) Tj +1 0 0 1 133.8 610.56 Tm +(orkload.) Tj +1 0 0 1 173.76 610.56 Tm +(It) Tj +1 0 0 1 183.48 610.56 Tm +(has) Tj +1 0 0 1 200.4 610.56 Tm +(\256v) Tj +1 0 0 1 210.84 610.56 Tm +(e) Tj +1 0 0 1 218.88 610.56 Tm +(phases:) Tj +1 0 0 1 254.039 610.56 Tm +(\(1\)) Tj +1 0 0 1 269.399 610.56 Tm +(creates) Tj +1 0 0 1 72 598.68 Tm +(subdirectories) Tj +1 0 0 1 131.4 598.68 Tm +(recursi) Tj +1 0 0 1 158.4 598.68 Tm +(v) Tj +1 0 0 1 163.32 598.68 Tm +(ely;) Tj +1 0 0 1 182.04 598.68 Tm +(\(2\)) Tj +1 0 0 1 197.16 598.68 Tm +(copies) Tj +1 0 0 1 226.2 598.68 Tm +(a) Tj +1 0 0 1 234.119 598.68 Tm +(source) Tj +1 0 0 1 263.639 598.68 Tm +(tree;) Tj +1 0 0 1 285.359 598.68 Tm +(\(3\)) Tj +1 0 0 1 72 586.68 Tm +(e) Tj +1 0 0 1 76.32 586.68 Tm +(xamines) Tj +1 0 0 1 113.52 586.68 Tm +(the) Tj +1 0 0 1 129.72 586.68 Tm +(status) Tj +1 0 0 1 156.48 586.68 Tm +(of) Tj +1 0 0 1 168.72 586.68 Tm +(all) Tj +1 0 0 1 182.76 586.68 Tm +(the) Tj +1 0 0 1 198.96 586.68 Tm +(\256les) Tj +1 0 0 1 219.599 586.68 Tm +(in) Tj +1 0 0 1 231.359 586.68 Tm +(the) Tj +1 0 0 1 247.559 586.68 Tm +(tree) Tj +1 0 0 1 266.519 586.68 Tm +(without) Tj +1 0 0 1 72 574.8 Tm +(e) Tj +1 0 0 1 76.32 574.8 Tm +(xamining) Tj +1 0 0 1 116.64 574.8 Tm +(their) Tj +1 0 0 1 137.64 574.8 Tm +(data;) Tj +1 0 0 1 159.96 574.8 Tm +(\(4\)) Tj +1 0 0 1 174.24 574.8 Tm +(e) Tj +1 0 0 1 178.56 574.8 Tm +(xamines) Tj +1 0 0 1 214.44 574.8 Tm +(e) Tj +1 0 0 1 218.64 574.8 Tm +(v) Tj +1 0 0 1 223.56 574.8 Tm +(ery) Tj +1 0 0 1 238.919 574.8 Tm +(byte) Tj +1 0 0 1 258.839 574.8 Tm +(of) Tj +1 0 0 1 269.879 574.8 Tm +(data) Tj +1 0 0 1 289.199 574.8 Tm +(in) Tj +1 0 0 1 72 562.8 Tm +(all) Tj +1 0 0 1 84.48 562.8 Tm +(the) Tj +1 0 0 1 99.1199 562.8 Tm +(\256les;) Tj +1 0 0 1 120.96 562.8 Tm +(and) Tj +1 0 0 1 137.88 562.8 Tm +(\(5\)) Tj +1 0 0 1 151.92 562.8 Tm +(compiles) Tj +1 0 0 1 190.44 562.8 Tm +(and) Tj +1 0 0 1 207.359 562.8 Tm +(links) Tj +1 0 0 1 229.199 562.8 Tm +(the) Tj +1 0 0 1 243.839 562.8 Tm +(\256les.) Tj +1 0 0 1 81.96 550.44 Tm +(W) Tj +1 0 0 1 90.48 550.44 Tm +(e) Tj +1 0 0 1 97.9199 550.44 Tm +(use) Tj +1 0 0 1 114 550.44 Tm +(the) Tj +1 0 0 1 129 550.44 Tm +(Andre) Tj +1 0 0 1 153.84 550.44 Tm +(w) Tj +1 0 0 1 163.8 550.44 Tm +(benchmark) Tj +1 0 0 1 210.96 550.44 Tm +(to) Tj +1 0 0 1 221.52 550.44 Tm +(compare) Tj +1 0 0 1 258.72 550.44 Tm +(BFS) Tj +1 0 0 1 279.24 550.44 Tm +(with) Tj +1 0 0 1 72 538.56 Tm +(tw) Tj +1 0 0 1 81.84 538.56 Tm +(o) Tj +1 0 0 1 89.88 538.56 Tm +(other) Tj +1 0 0 1 113.28 538.56 Tm +(\256le) Tj +1 0 0 1 129.12 538.56 Tm +(system) Tj +1 0 0 1 159.72 538.56 Tm +(con\256gurations:) Tj +1 0 0 1 224.16 538.56 Tm +(NFS-std,) Tj +1 0 0 1 263.04 538.56 Tm +(which) Tj +1 0 0 1 290.399 538.56 Tm +(is) Tj +1 0 0 1 72 526.56 Tm +(the) Tj +1 0 0 1 86.1599 526.56 Tm +(NFS) Tj +1 0 0 1 106.44 526.56 Tm +(V2) Tj +1 0 0 1 120.6 526.56 Tm +(implementation) Tj +1 0 0 1 185.16 526.56 Tm +(in) Tj +1 0 0 1 195 526.56 Tm +(Digital) Tj +1 0 0 1 224.64 526.56 Tm +(Unix,) Tj +1 0 0 1 249.12 526.56 Tm +(and) Tj +1 0 0 1 265.559 526.56 Tm +(BFS-nr) Tj +1 0 0 1 294.599 526.56 Tm +(,) Tj +1 0 0 1 72 514.68 Tm +(which) Tj +1 0 0 1 98.4 514.68 Tm +(is) Tj +1 0 0 1 107.04 514.68 Tm +(identical) Tj +1 0 0 1 143.4 514.68 Tm +(to) Tj +1 0 0 1 153.12 514.68 Tm +(BFS) Tj +1 0 0 1 172.92 514.68 Tm +(b) Tj +1 0 0 1 177.72 514.68 Tm +(ut) Tj +1 0 0 1 187.44 514.68 Tm +(with) Tj +1 0 0 1 207.12 514.68 Tm +(no) Tj +1 0 0 1 219.12 514.68 Tm +(replication.) Tj +1 0 0 1 267.719 514.68 Tm +(BFS-nr) Tj +1 0 0 1 72 502.68 Tm +(ran) Tj +1 0 0 1 86.5199 502.68 Tm +(tw) Tj +1 0 0 1 96.3599 502.68 Tm +(o) Tj +1 0 0 1 103.08 502.68 Tm +(simple) Tj +1 0 0 1 131.52 502.68 Tm +(UDP) Tj +1 0 0 1 153.12 502.68 Tm +(relays) Tj +1 0 0 1 178.68 502.68 Tm +(on) Tj +1 0 0 1 190.44 502.68 Tm +(the) Tj +1 0 0 1 204.36 502.68 Tm +(client,) Tj +1 0 0 1 230.999 502.68 Tm +(and) Tj +1 0 0 1 247.079 502.68 Tm +(on) Tj +1 0 0 1 258.839 502.68 Tm +(the) Tj +1 0 0 1 272.759 502.68 Tm +(serv) Tj +1 0 0 1 289.319 502.68 Tm +(er) Tj +1 0 0 1 72 490.8 Tm +(it) Tj +1 0 0 1 80.88 490.8 Tm +(ran) Tj +1 0 0 1 96.8399 490.8 Tm +(a) Tj +1 0 0 1 104.52 490.8 Tm +(thin) Tj +1 0 0 1 123.36 490.8 Tm +(v) Tj +1 0 0 1 128.28 490.8 Tm +(eneer) Tj +1 0 0 1 153 490.8 Tm +(link) Tj +1 0 0 1 168.48 490.8 Tm +(ed) Tj +1 0 0 1 181.08 490.8 Tm +(with) Tj +1 0 0 1 202.08 490.8 Tm +(a) Tj +1 0 0 1 209.76 490.8 Tm +(v) Tj +1 0 0 1 214.679 490.8 Tm +(ersion) Tj +1 0 0 1 242.279 490.8 Tm +(of) Tj +ET +endstream +endobj +730 0 obj +4859 +endobj +731 0 obj +<< /Type /Font /Name /R731 /Subtype /Type1 /BaseFont /Times-Italic >> +endobj +732 0 obj +<< /Length 733 0 R >> +stream +BT +/R731 10 Tf +1 0 0 1 253.919 490.8 Tm +(snfsd) Tj +/R725 10 Tf +1 0 0 1 277.679 490.8 Tm +(from) Tj +1 0 0 1 72 478.8 Tm +(which) Tj +1 0 0 1 98.1599 478.8 Tm +(all) Tj +1 0 0 1 109.92 478.8 Tm +(the) Tj +1 0 0 1 123.96 478.8 Tm +(checkpoint) Tj +1 0 0 1 169.44 478.8 Tm +(management) Tj +1 0 0 1 222.12 478.8 Tm +(code) Tj +1 0 0 1 242.759 478.8 Tm +(w) Tj +1 0 0 1 249.839 478.8 Tm +(as) Tj +1 0 0 1 259.919 478.8 Tm +(remo) Tj +1 0 0 1 280.439 478.8 Tm +(v) Tj +1 0 0 1 285.359 478.8 Tm +(ed.) Tj +1 0 0 1 72 466.8 Tm +(This) Tj +1 0 0 1 93.5999 466.8 Tm +(con\256guration) Tj +1 0 0 1 150.48 466.8 Tm +(does) Tj +/R731 10 Tf +1 0 0 1 172.56 466.8 Tm +(not) Tj +/R725 10 Tf +1 0 0 1 189.24 466.8 Tm +(write) Tj +1 0 0 1 213.48 466.8 Tm +(modi\256ed) Tj +1 0 0 1 252.72 466.8 Tm +(\256le) Tj +1 0 0 1 269.279 466.8 Tm +(system) Tj +1 0 0 1 72 454.92 Tm +(state) Tj +1 0 0 1 93.4799 454.92 Tm +(to) Tj +1 0 0 1 104.52 454.92 Tm +(disk) Tj +1 0 0 1 124.32 454.92 Tm +(before) Tj +1 0 0 1 153 454.92 Tm +(replying) Tj +1 0 0 1 189.48 454.92 Tm +(to) Tj +1 0 0 1 200.4 454.92 Tm +(the) Tj +1 0 0 1 215.879 454.92 Tm +(client.) Tj +1 0 0 1 246.239 454.92 Tm +(Therefore,) Tj +1 0 0 1 291.479 454.92 Tm +(it) Tj +1 0 0 1 72 442.92 Tm +(does) Tj +1 0 0 1 92.1599 442.92 Tm +(not) Tj +1 0 0 1 106.8 442.92 Tm +(implement) Tj +1 0 0 1 151.32 442.92 Tm +(NFS) Tj +1 0 0 1 171.48 442.92 Tm +(V2) Tj +1 0 0 1 185.52 442.92 Tm +(protocol) Tj +1 0 0 1 220.56 442.92 Tm +(semantics,) Tj +1 0 0 1 264.359 442.92 Tm +(whereas) Tj +1 0 0 1 72 431.04 Tm +(both) Tj +1 0 0 1 92.1599 431.04 Tm +(BFS) Tj +1 0 0 1 112.44 431.04 Tm +(and) Tj +1 0 0 1 129.24 431.04 Tm +(NFS-std) Tj +1 0 0 1 165 431.04 Tm +(do.) Tj +1 0 0 1 81.96 418.68 Tm +(Out) Tj +1 0 0 1 99.5999 418.68 Tm +(of) Tj +1 0 0 1 110.52 418.68 Tm +(the) Tj +1 0 0 1 125.4 418.68 Tm +(18) Tj +1 0 0 1 138.12 418.68 Tm +(operations) Tj +1 0 0 1 182.28 418.68 Tm +(in) Tj +1 0 0 1 192.72 418.68 Tm +(the) Tj +1 0 0 1 207.6 418.68 Tm +(NFS) Tj +1 0 0 1 228.479 418.68 Tm +(V2) Tj +1 0 0 1 243.359 418.68 Tm +(protocol) Tj +1 0 0 1 279.239 418.68 Tm +(only) Tj +ET +endstream +endobj +733 0 obj +1993 +endobj +734 0 obj +<< /Type /Font /Name /R734 /Subtype /Type1 /BaseFont /Courier >> +endobj +735 0 obj +<< /Length 736 0 R >> +stream +BT +/R734 10 Tf +1 0 0 1 72 406.68 Tm +(getattr) Tj +/R725 10 Tf +1 0 0 1 118.2 406.68 Tm +(is) Tj +1 0 0 1 129.12 406.68 Tm +(read-only) Tj +1 0 0 1 171.6 406.68 Tm +(because) Tj +1 0 0 1 207.479 406.68 Tm +(the) Tj +1 0 0 1 223.919 406.68 Tm +(time-last-accessed) Tj +1 0 0 1 72 394.8 Tm +(attrib) Tj +1 0 0 1 92.8799 394.8 Tm +(ute) Tj +1 0 0 1 110.16 394.8 Tm +(of) Tj +1 0 0 1 123.72 394.8 Tm +(\256les) Tj +1 0 0 1 145.44 394.8 Tm +(and) Tj +1 0 0 1 165 394.8 Tm +(directories) Tj +1 0 0 1 212.28 394.8 Tm +(is) Tj +1 0 0 1 224.04 394.8 Tm +(set) Tj +1 0 0 1 240.359 394.8 Tm +(by) Tj +1 0 0 1 255.479 394.8 Tm +(operations) Tj +1 0 0 1 72 382.8 Tm +(that) Tj +1 0 0 1 92.6399 382.8 Tm +(w) Tj +1 0 0 1 99.7199 382.8 Tm +(ould) Tj +1 0 0 1 123.24 382.8 Tm +(otherwise) Tj +1 0 0 1 167.76 382.8 Tm +(be) Tj +1 0 0 1 182.88 382.8 Tm +(read-only) Tj +1 0 0 1 220.8 382.8 Tm +(,) Tj +1 0 0 1 229.56 382.8 Tm +(e.g.,) Tj +/R734 10 Tf +1 0 0 1 252.96 382.8 Tm +(read) Tj +/R725 10 Tf +1 0 0 1 282.599 382.8 Tm +(and) Tj +/R734 10 Tf +1 0 0 1 72 370.92 Tm +(lookup) Tj +/R725 10 Tf +1 0 0 1 108 370.92 Tm +(.) Tj +1 0 0 1 117.12 370.92 Tm +(The) Tj +1 0 0 1 136.2 370.92 Tm +(result) Tj +1 0 0 1 161.88 370.92 Tm +(is) Tj +1 0 0 1 172.08 370.92 Tm +(that) Tj +1 0 0 1 190.559 370.92 Tm +(our) Tj +1 0 0 1 207.479 370.92 Tm +(optimization) Tj +1 0 0 1 261.359 370.92 Tm +(for) Tj +1 0 0 1 276.479 370.92 Tm +(read-) Tj +1 0 0 1 72 358.92 Tm +(only) Tj +1 0 0 1 92.6399 358.92 Tm +(operations) Tj +1 0 0 1 137.04 358.92 Tm +(can) Tj +1 0 0 1 153.84 358.92 Tm +(rarely) Tj +1 0 0 1 180 358.92 Tm +(be) Tj +1 0 0 1 192.24 358.92 Tm +(used.) Tj +1 0 0 1 217.92 358.92 Tm +(T) Tj +1 0 0 1 223.2 358.92 Tm +(o) Tj +1 0 0 1 231.119 358.92 Tm +(sho) Tj +1 0 0 1 244.799 358.92 Tm +(w) Tj +1 0 0 1 254.759 358.92 Tm +(the) Tj +1 0 0 1 269.879 358.92 Tm +(impact) Tj +1 0 0 1 72 347.04 Tm +(of) Tj +1 0 0 1 82.92 347.04 Tm +(this) Tj +1 0 0 1 99.8399 347.04 Tm +(optimization,) Tj +1 0 0 1 155.4 347.04 Tm +(we) Tj +1 0 0 1 169.56 347.04 Tm +(also) Tj +1 0 0 1 188.16 347.04 Tm +(ran) Tj +1 0 0 1 203.52 347.04 Tm +(the) Tj +1 0 0 1 218.28 347.04 Tm +(Andre) Tj +1 0 0 1 243.119 347.04 Tm +(w) Tj +1 0 0 1 252.719 347.04 Tm +(benchmark) Tj +1 0 0 1 72 335.04 Tm +(on) Tj +1 0 0 1 85.3199 335.04 Tm +(a) Tj +1 0 0 1 93.1199 335.04 Tm +(second) Tj +1 0 0 1 124.2 335.04 Tm +(v) Tj +1 0 0 1 129.12 335.04 Tm +(ersion) Tj +1 0 0 1 156.72 335.04 Tm +(of) Tj +1 0 0 1 168.48 335.04 Tm +(BFS) Tj +1 0 0 1 189.6 335.04 Tm +(that) Tj +1 0 0 1 207.84 335.04 Tm +(modi\256es) Tj +1 0 0 1 245.519 335.04 Tm +(the) Tj +/R734 10 Tf +1 0 0 1 261.119 335.04 Tm +(lookup) Tj +/R725 10 Tf +1 0 0 1 72 323.16 Tm +(operation) Tj +1 0 0 1 114 323.16 Tm +(to) Tj +1 0 0 1 126.12 323.16 Tm +(be) Tj +1 0 0 1 139.92 323.16 Tm +(read-only) Tj +1 0 0 1 177.84 323.16 Tm +(.) Tj +1 0 0 1 189.12 323.16 Tm +(This) Tj +1 0 0 1 211.32 323.16 Tm +(modi\256cation) Tj +1 0 0 1 266.039 323.16 Tm +(violates) Tj +1 0 0 1 72 311.16 Tm +(strict) Tj +1 0 0 1 95.0399 311.16 Tm +(Unix) Tj +1 0 0 1 117.96 311.16 Tm +(\256le) Tj +1 0 0 1 133.8 311.16 Tm +(system) Tj +1 0 0 1 164.52 311.16 Tm +(semantics) Tj +1 0 0 1 206.88 311.16 Tm +(b) Tj +1 0 0 1 211.68 311.16 Tm +(ut) Tj +1 0 0 1 222.36 311.16 Tm +(is) Tj +1 0 0 1 232.08 311.16 Tm +(unlik) Tj +1 0 0 1 252.599 311.16 Tm +(ely) Tj +1 0 0 1 267.719 311.16 Tm +(to) Tj +1 0 0 1 278.519 311.16 Tm +(ha) Tj +1 0 0 1 287.759 311.16 Tm +(v) Tj +1 0 0 1 292.679 311.16 Tm +(e) Tj +1 0 0 1 72 299.16 Tm +(adv) Tj +1 0 0 1 86.4 299.16 Tm +(erse) Tj +1 0 0 1 104.76 299.16 Tm +(ef) Tj +1 0 0 1 112.32 299.16 Tm +(fects) Tj +1 0 0 1 133.56 299.16 Tm +(in) Tj +1 0 0 1 143.76 299.16 Tm +(practice.) Tj +1 0 0 1 81.96 286.8 Tm +(F) Tj +1 0 0 1 87.36 286.8 Tm +(or) Tj +1 0 0 1 98.5199 286.8 Tm +(all) Tj +1 0 0 1 111.36 286.8 Tm +(con\256gurations,) Tj +1 0 0 1 173.76 286.8 Tm +(the) Tj +1 0 0 1 188.88 286.8 Tm +(actual) Tj +1 0 0 1 215.52 286.8 Tm +(benchmark) Tj +1 0 0 1 262.559 286.8 Tm +(code) Tj +1 0 0 1 284.279 286.8 Tm +(ran) Tj +1 0 0 1 72 274.92 Tm +(at) Tj +1 0 0 1 82.8 274.92 Tm +(the) Tj +1 0 0 1 98.5199 274.92 Tm +(client) Tj +1 0 0 1 124.2 274.92 Tm +(w) Tj +1 0 0 1 131.28 274.92 Tm +(orkstation) Tj +1 0 0 1 174.72 274.92 Tm +(using) Tj +1 0 0 1 199.92 274.92 Tm +(the) Tj +1 0 0 1 215.639 274.92 Tm +(standard) Tj +1 0 0 1 252.959 274.92 Tm +(NFS) Tj +1 0 0 1 274.919 274.92 Tm +(client) Tj +1 0 0 1 72 262.92 Tm +(implementation) Tj +1 0 0 1 137.04 262.92 Tm +(in) Tj +1 0 0 1 147.24 262.92 Tm +(the) Tj +1 0 0 1 161.88 262.92 Tm +(Digital) Tj +1 0 0 1 192 262.92 Tm +(Unix) Tj +1 0 0 1 214.44 262.92 Tm +(k) Tj +1 0 0 1 219.36 262.92 Tm +(ernel) Tj +1 0 0 1 241.679 262.92 Tm +(with) Tj +1 0 0 1 261.839 262.92 Tm +(the) Tj +1 0 0 1 276.479 262.92 Tm +(same) Tj +/R734 10 Tf +1 0 0 1 72 251.04 Tm +(mount) Tj +/R725 10 Tf +1 0 0 1 105 251.04 Tm +(options.) Tj +1 0 0 1 142.2 251.04 Tm +(The) Tj +1 0 0 1 160.8 251.04 Tm +(most) Tj +1 0 0 1 183.24 251.04 Tm +(rele) Tj +1 0 0 1 198 251.04 Tm +(v) Tj +1 0 0 1 202.8 251.04 Tm +(ant) Tj +1 0 0 1 218.04 251.04 Tm +(of) Tj +1 0 0 1 229.319 251.04 Tm +(these) Tj +1 0 0 1 252.959 251.04 Tm +(options) Tj +1 0 0 1 285.359 251.04 Tm +(for) Tj +1 0 0 1 72 239.04 Tm +(the) Tj +1 0 0 1 87.3599 239.04 Tm +(benchmark) Tj +1 0 0 1 134.76 239.04 Tm +(are:) Tj +1 0 0 1 154.56 239.04 Tm +(UDP) Tj +1 0 0 1 177.6 239.04 Tm +(transport,) Tj +1 0 0 1 218.76 239.04 Tm +(4096-byte) Tj +1 0 0 1 262.319 239.04 Tm +(read) Tj +1 0 0 1 282.599 239.04 Tm +(and) Tj +1 0 0 1 72 227.16 Tm +(write) Tj +1 0 0 1 95.9999 227.16 Tm +(b) Tj +1 0 0 1 100.8 227.16 Tm +(uf) Tj +1 0 0 1 108.96 227.16 Tm +(fers,) Tj +1 0 0 1 129.96 227.16 Tm +(allo) Tj +1 0 0 1 144.72 227.16 Tm +(wing) Tj +1 0 0 1 168.12 227.16 Tm +(asynchronous) Tj +1 0 0 1 226.44 227.16 Tm +(client) Tj +1 0 0 1 251.999 227.16 Tm +(writes,) Tj +1 0 0 1 282.599 227.16 Tm +(and) Tj +1 0 0 1 72 215.16 Tm +(allo) Tj +1 0 0 1 86.76 215.16 Tm +(wing) Tj +1 0 0 1 109.08 215.16 Tm +(attrib) Tj +1 0 0 1 129.96 215.16 Tm +(ute) Tj +1 0 0 1 144.6 215.16 Tm +(caching.) Tj +1 0 0 1 81.96 202.8 Tm +(W) Tj +1 0 0 1 90.48 202.8 Tm +(e) Tj +1 0 0 1 98.28 202.8 Tm +(report) Tj +1 0 0 1 125.28 202.8 Tm +(the) Tj +1 0 0 1 140.76 202.8 Tm +(mean) Tj +1 0 0 1 165.6 202.8 Tm +(of) Tj +1 0 0 1 177.24 202.8 Tm +(10) Tj +1 0 0 1 190.44 202.8 Tm +(runs) Tj +1 0 0 1 210.84 202.8 Tm +(of) Tj +1 0 0 1 222.36 202.8 Tm +(the) Tj +1 0 0 1 237.84 202.8 Tm +(benchmark) Tj +1 0 0 1 285.359 202.8 Tm +(for) Tj +1 0 0 1 72 190.92 Tm +(each) Tj +1 0 0 1 93.8399 190.92 Tm +(con\256guration.) Tj +1 0 0 1 156.48 190.92 Tm +(The) Tj +1 0 0 1 175.68 190.92 Tm +(sample) Tj +1 0 0 1 207.48 190.92 Tm +(standard) Tj +1 0 0 1 244.919 190.92 Tm +(de) Tj +1 0 0 1 254.159 190.92 Tm +(viation) Tj +1 0 0 1 285.359 190.92 Tm +(for) Tj +1 0 0 1 72 178.92 Tm +(the) Tj +1 0 0 1 87.8399 178.92 Tm +(total) Tj +1 0 0 1 109.2 178.92 Tm +(time) Tj +1 0 0 1 130.56 178.92 Tm +(to) Tj +1 0 0 1 141.96 178.92 Tm +(run) Tj +1 0 0 1 159 178.92 Tm +(the) Tj +1 0 0 1 174.84 178.92 Tm +(benchmark) Tj +1 0 0 1 222.719 178.92 Tm +(w) Tj +1 0 0 1 229.799 178.92 Tm +(as) Tj +1 0 0 1 241.799 178.92 Tm +(al) Tj +1 0 0 1 248.879 178.92 Tm +(w) Tj +1 0 0 1 255.959 178.92 Tm +(ays) Tj +1 0 0 1 272.879 178.92 Tm +(belo) Tj +1 0 0 1 289.919 178.92 Tm +(w) Tj +1 0 0 1 72 167.04 Tm +(2.6%) Tj +1 0 0 1 95.4 167.04 Tm +(of) Tj +1 0 0 1 106.32 167.04 Tm +(the) Tj +1 0 0 1 121.08 167.04 Tm +(reported) Tj +1 0 0 1 156.96 167.04 Tm +(v) Tj +1 0 0 1 161.76 167.04 Tm +(alue) Tj +1 0 0 1 180.96 167.04 Tm +(b) Tj +1 0 0 1 185.76 167.04 Tm +(ut) Tj +1 0 0 1 196.08 167.04 Tm +(it) Tj +1 0 0 1 204.24 167.04 Tm +(w) Tj +1 0 0 1 211.32 167.04 Tm +(as) Tj +1 0 0 1 222.24 167.04 Tm +(as) Tj +1 0 0 1 233.159 167.04 Tm +(high) Tj +1 0 0 1 253.559 167.04 Tm +(as) Tj +1 0 0 1 264.479 167.04 Tm +(14%) Tj +1 0 0 1 285.359 167.04 Tm +(for) Tj +1 0 0 1 72 155.04 Tm +(the) Tj +1 0 0 1 87.5999 155.04 Tm +(indi) Tj +1 0 0 1 102.96 155.04 Tm +(vidual) Tj +1 0 0 1 131.28 155.04 Tm +(times) Tj +1 0 0 1 156.24 155.04 Tm +(of) Tj +1 0 0 1 168 155.04 Tm +(the) Tj +1 0 0 1 183.6 155.04 Tm +(\256rst) Tj +1 0 0 1 202.56 155.04 Tm +(four) Tj +1 0 0 1 222.599 155.04 Tm +(phases.) Tj +1 0 0 1 258.119 155.04 Tm +(This) Tj +1 0 0 1 279.239 155.04 Tm +(high) Tj +1 0 0 1 72 143.04 Tm +(v) Tj +1 0 0 1 76.8 143.04 Tm +(ariance) Tj +1 0 0 1 108.72 143.04 Tm +(w) Tj +1 0 0 1 115.8 143.04 Tm +(as) Tj +1 0 0 1 127.32 143.04 Tm +(also) Tj +1 0 0 1 146.64 143.04 Tm +(present) Tj +1 0 0 1 178.56 143.04 Tm +(in) Tj +1 0 0 1 189.6 143.04 Tm +(the) Tj +1 0 0 1 204.96 143.04 Tm +(NFS-std) Tj +1 0 0 1 241.319 143.04 Tm +(con\256guration.) Tj +1 0 0 1 72 131.16 Tm +(The) Tj +1 0 0 1 91.6799 131.16 Tm +(estimated) Tj +1 0 0 1 134.16 131.16 Tm +(error) Tj +1 0 0 1 157.8 131.16 Tm +(for) Tj +1 0 0 1 173.64 131.16 Tm +(the) Tj +1 0 0 1 189.96 131.16 Tm +(reported) Tj +1 0 0 1 227.399 131.16 Tm +(mean) Tj +1 0 0 1 253.319 131.16 Tm +(w) Tj +1 0 0 1 260.399 131.16 Tm +(as) Tj +1 0 0 1 272.879 131.16 Tm +(belo) Tj +1 0 0 1 289.919 131.16 Tm +(w) Tj +1 0 0 1 72 119.16 Tm +(4.5%) Tj +1 0 0 1 95.2799 119.16 Tm +(for) Tj +1 0 0 1 109.32 119.16 Tm +(the) Tj +1 0 0 1 123.96 119.16 Tm +(indi) Tj +1 0 0 1 139.32 119.16 Tm +(vidual) Tj +1 0 0 1 166.68 119.16 Tm +(phases) Tj +1 0 0 1 195.72 119.16 Tm +(and) Tj +1 0 0 1 212.639 119.16 Tm +(0.8%) Tj +1 0 0 1 235.799 119.16 Tm +(for) Tj +1 0 0 1 249.959 119.16 Tm +(the) Tj +1 0 0 1 264.599 119.16 Tm +(total.) Tj +1 0 0 1 81.96 106.8 Tm +(T) Tj +1 0 0 1 87.24 106.8 Tm +(able) Tj +1 0 0 1 107.4 106.8 Tm +(2) Tj +1 0 0 1 115.92 106.8 Tm +(sho) Tj +1 0 0 1 129.6 106.8 Tm +(ws) Tj +1 0 0 1 144 106.8 Tm +(the) Tj +1 0 0 1 159.72 106.8 Tm +(results) Tj +1 0 0 1 189.24 106.8 Tm +(for) Tj +1 0 0 1 204.36 106.8 Tm +(BFS) Tj +1 0 0 1 225.6 106.8 Tm +(and) Tj +1 0 0 1 243.599 106.8 Tm +(BFS-nr) Tj +1 0 0 1 272.399 106.8 Tm +(.) Tj +1 0 0 1 281.519 106.8 Tm +(The) Tj +1 0 0 1 72 94.92 Tm +(comparison) Tj +1 0 0 1 122.04 94.92 Tm +(between) Tj +1 0 0 1 158.64 94.92 Tm +(BFS-strict) Tj +1 0 0 1 203.16 94.92 Tm +(and) Tj +1 0 0 1 221.04 94.92 Tm +(BFS-nr) Tj +1 0 0 1 253.919 94.92 Tm +(sho) Tj +1 0 0 1 267.599 94.92 Tm +(ws) Tj +1 0 0 1 281.999 94.92 Tm +(that) Tj +1 0 0 1 72 82.92 Tm +(the) Tj +1 0 0 1 86.5199 82.92 Tm +(o) Tj +1 0 0 1 91.4399 82.92 Tm +(v) Tj +1 0 0 1 96.3599 82.92 Tm +(erhead) Tj +1 0 0 1 125.16 82.92 Tm +(of) Tj +1 0 0 1 135.84 82.92 Tm +(Byzantine) Tj +1 0 0 1 178.56 82.92 Tm +(f) Tj +1 0 0 1 181.8 82.92 Tm +(ault) Tj +1 0 0 1 199.08 82.92 Tm +(tolerance) Tj +1 0 0 1 238.079 82.92 Tm +(for) Tj +1 0 0 1 251.999 82.92 Tm +(this) Tj +1 0 0 1 268.799 82.92 Tm +(service) Tj +1 0 0 1 72 71.04 Tm +(is) Tj +1 0 0 1 82.32 71.04 Tm +(lo) Tj +1 0 0 1 89.8799 71.04 Tm +(w) Tj +1 0 0 1 100.8 71.04 Tm +(\320) Tj +1 0 0 1 114.48 71.04 Tm +(BFS-strict) Tj +1 0 0 1 159.12 71.04 Tm +(tak) Tj +1 0 0 1 171.24 71.04 Tm +(es) Tj +1 0 0 1 183.24 71.04 Tm +(only) Tj +1 0 0 1 204.6 71.04 Tm +(26%) Tj +1 0 0 1 226.68 71.04 Tm +(more) Tj +1 0 0 1 250.799 71.04 Tm +(time) Tj +1 0 0 1 272.279 71.04 Tm +(to) Tj +1 0 0 1 283.679 71.04 Tm +(run) Tj +ET +Q +q +W +0 0 612 792 re +n +q +184.32 0 0 -0.48 335.388 709.092 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +q +0.48 0 0 -11.04 335.148 708.732 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +q +0.48 0 0 -11.04 367.548 708.732 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +q +0.48 0 0 -11.04 367.548 708.732 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +/R725 9 Tf +1 0 0 1 416.52 700.92 Tm +(BFS) Tj +ET +q +0.48 0 0 -11.04 481.068 708.732 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +q +0.48 0 0 -11.04 519.468 708.732 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +q +113.64 0 0 -0.48 367.668 697.812 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +q +0.48 0 0 -11.04 335.148 697.812 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +1 0 0 1 341.28 690 Tm +(phase) Tj +ET +q +0.48 0 0 -11.04 367.548 697.812 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +1 0 0 1 387.12 690 Tm +(strict) Tj +ET +q +0.48 0 0 -11.04 424.308 697.812 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +1 0 0 1 434.28 690 Tm +(r/o) Tj +1 0 0 1 446.52 690 Tm +(lookup) Tj +ET +q +0.48 0 0 -11.04 481.068 697.812 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +1 0 0 1 487.32 690 Tm +(BFS-nr) Tj +ET +q +0.48 0 0 -11.04 519.468 697.812 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +q +184.32 0 0 -0.48 335.388 686.772 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +q +0.48 0 0 -11.04 335.148 686.412 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +1 0 0 1 349.32 678.6 Tm +(1) Tj +ET +q +0.48 0 0 -11.04 367.548 686.412 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +1 0 0 1 375.96 678.6 Tm +(0.55) Tj +1 0 0 1 393.84 678.6 Tm +(\(57%\)) Tj +ET +q +0.48 0 0 -11.04 424.308 686.412 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +1 0 0 1 432.72 678.6 Tm +(0.47) Tj +1 0 0 1 450.6 678.6 Tm +(\(34%\)) Tj +ET +q +0.48 0 0 -11.04 481.068 686.412 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +1 0 0 1 492.6 678.6 Tm +(0.35) Tj +ET +q +0.48 0 0 -11.04 519.468 686.412 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +q +0.48 0 0 -11.04 335.148 675.492 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +1 0 0 1 349.32 667.68 Tm +(2) Tj +ET +q +0.48 0 0 -11.04 367.548 675.492 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +1 0 0 1 375.96 667.68 Tm +(9.24) Tj +1 0 0 1 393.84 667.68 Tm +(\(82%\)) Tj +ET +q +0.48 0 0 -11.04 424.308 675.492 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +1 0 0 1 432.72 667.68 Tm +(7.91) Tj +1 0 0 1 450.6 667.68 Tm +(\(56%\)) Tj +ET +q +0.48 0 0 -11.04 481.068 675.492 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +1 0 0 1 492.6 667.68 Tm +(5.08) Tj +ET +q +0.48 0 0 -11.04 519.468 675.492 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +q +0.48 0 0 -11.04 335.148 664.452 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +1 0 0 1 349.32 656.76 Tm +(3) Tj +ET +q +0.48 0 0 -11.04 367.548 664.452 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +1 0 0 1 375.96 656.76 Tm +(7.24) Tj +1 0 0 1 393.84 656.76 Tm +(\(18%\)) Tj +ET +q +0.48 0 0 -11.04 424.308 664.452 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +1 0 0 1 435 656.76 Tm +(6.45) Tj +1 0 0 1 452.88 656.76 Tm +(\(6%\)) Tj +ET +q +0.48 0 0 -11.04 481.068 664.452 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +1 0 0 1 492.6 656.76 Tm +(6.11) Tj +ET +q +0.48 0 0 -11.04 519.468 664.452 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +q +0.48 0 0 -11.04 335.148 653.532 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +1 0 0 1 349.32 645.72 Tm +(4) Tj +ET +q +0.48 0 0 -11.04 367.548 653.532 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +1 0 0 1 375.96 645.72 Tm +(8.77) Tj +1 0 0 1 393.84 645.72 Tm +(\(18%\)) Tj +ET +q +0.48 0 0 -11.04 424.308 653.532 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +1 0 0 1 435 645.72 Tm +(7.87) Tj +1 0 0 1 452.88 645.72 Tm +(\(6%\)) Tj +ET +q +0.48 0 0 -11.04 481.068 653.532 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +1 0 0 1 492.6 645.72 Tm +(7.41) Tj +ET +q +0.48 0 0 -11.04 519.468 653.532 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +q +0.48 0 0 -11.04 335.148 642.612 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +1 0 0 1 349.32 634.8 Tm +(5) Tj +ET +q +0.48 0 0 -11.04 367.548 642.612 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +1 0 0 1 373.68 634.8 Tm +(38.68) Tj +1 0 0 1 396.12 634.8 Tm +(\(20%\)) Tj +ET +q +0.48 0 0 -11.04 424.308 642.612 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +1 0 0 1 430.44 634.8 Tm +(38.38) Tj +1 0 0 1 452.88 634.8 Tm +(\(19%\)) Tj +ET +q +0.48 0 0 -11.04 481.068 642.612 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +1 0 0 1 490.319 634.8 Tm +(32.12) Tj +ET +q +0.48 0 0 -11.04 519.468 642.612 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +q +184.32 0 0 -0.48 335.388 631.572 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +q +0.48 0 0 -11.04 335.148 631.212 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +1 0 0 1 343.56 623.52 Tm +(total) Tj +ET +q +0.48 0 0 -11.04 367.548 631.212 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +1 0 0 1 373.68 623.52 Tm +(64.48) Tj +1 0 0 1 396.12 623.52 Tm +(\(26%\)) Tj +ET +q +0.48 0 0 -11.04 424.308 631.212 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +1 0 0 1 430.44 623.52 Tm +(61.07) Tj +1 0 0 1 452.88 623.52 Tm +(\(20%\)) Tj +ET +q +0.48 0 0 -11.04 481.068 631.212 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +1 0 0 1 490.319 623.52 Tm +(51.07) Tj +ET +q +0.48 0 0 -11.04 519.468 631.212 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +q +184.32 0 0 -0.48 335.388 620.292 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +/R725 10 Tf +1 0 0 1 315 606.36 Tm +(T) Tj +1 0 0 1 320.28 606.36 Tm +(able) Tj +1 0 0 1 339.12 606.36 Tm +(2:) Tj +1 0 0 1 350.28 606.36 Tm +(Andre) Tj +1 0 0 1 375.12 606.36 Tm +(w) Tj +1 0 0 1 384.36 606.36 Tm +(benchmark:) Tj +1 0 0 1 434.88 606.36 Tm +(BFS) Tj +1 0 0 1 454.8 606.36 Tm +(vs) Tj +1 0 0 1 465.96 606.36 Tm +(BFS-nr) Tj +1 0 0 1 494.759 606.36 Tm +(.) Tj +1 0 0 1 500.639 606.36 Tm +(The) Tj +1 0 0 1 518.399 606.36 Tm +(times) Tj +1 0 0 1 315 594.36 Tm +(are) Tj +1 0 0 1 329.64 594.36 Tm +(in) Tj +1 0 0 1 339.84 594.36 Tm +(seconds.) Tj +1 0 0 1 315 569.16 Tm +(the) Tj +1 0 0 1 331.08 569.16 Tm +(complete) Tj +1 0 0 1 371.64 569.16 Tm +(benchmark.) Tj +1 0 0 1 426.24 569.16 Tm +(The) Tj +1 0 0 1 445.68 569.16 Tm +(o) Tj +1 0 0 1 450.6 569.16 Tm +(v) Tj +1 0 0 1 455.52 569.16 Tm +(erhead) Tj +1 0 0 1 485.88 569.16 Tm +(is) Tj +1 0 0 1 496.44 569.16 Tm +(lo) Tj +1 0 0 1 503.999 569.16 Tm +(wer) Tj +1 0 0 1 522.839 569.16 Tm +(than) Tj +1 0 0 1 315 557.28 Tm +(what) Tj +1 0 0 1 338.4 557.28 Tm +(w) Tj +1 0 0 1 345.48 557.28 Tm +(as) Tj +1 0 0 1 357.84 557.28 Tm +(observ) Tj +1 0 0 1 384.48 557.28 Tm +(ed) Tj +1 0 0 1 397.8 557.28 Tm +(for) Tj +1 0 0 1 413.4 557.28 Tm +(the) Tj +1 0 0 1 429.72 557.28 Tm +(micro-benchmarks) Tj +1 0 0 1 508.439 557.28 Tm +(because) Tj +1 0 0 1 315 545.28 Tm +(the) Tj +1 0 0 1 329.28 545.28 Tm +(client) Tj +1 0 0 1 353.4 545.28 Tm +(spends) Tj +1 0 0 1 382.68 545.28 Tm +(a) Tj +1 0 0 1 389.16 545.28 Tm +(signi\256cant) Tj +1 0 0 1 432.72 545.28 Tm +(fraction) Tj +1 0 0 1 465.72 545.28 Tm +(of) Tj +1 0 0 1 476.159 545.28 Tm +(the) Tj +1 0 0 1 490.319 545.28 Tm +(elapsed) Tj +1 0 0 1 522.239 545.28 Tm +(time) Tj +1 0 0 1 315 533.4 Tm +(computing) Tj +1 0 0 1 361.56 533.4 Tm +(between) Tj +1 0 0 1 398.76 533.4 Tm +(operations,) Tj +1 0 0 1 447 533.4 Tm +(i.e.,) Tj +1 0 0 1 466.08 533.4 Tm +(between) Tj +1 0 0 1 503.16 533.4 Tm +(recei) Tj +1 0 0 1 522.359 533.4 Tm +(ving) Tj +1 0 0 1 315 521.4 Tm +(the) Tj +1 0 0 1 330.96 521.4 Tm +(reply) Tj +1 0 0 1 355.32 521.4 Tm +(to) Tj +1 0 0 1 366.84 521.4 Tm +(an) Tj +1 0 0 1 380.16 521.4 Tm +(operation) Tj +1 0 0 1 421.56 521.4 Tm +(and) Tj +1 0 0 1 439.8 521.4 Tm +(issuing) Tj +1 0 0 1 471.84 521.4 Tm +(the) Tj +1 0 0 1 487.919 521.4 Tm +(ne) Tj +1 0 0 1 497.279 521.4 Tm +(xt) Tj +1 0 0 1 508.679 521.4 Tm +(request,) Tj +1 0 0 1 315 509.52 Tm +(and) Tj +1 0 0 1 332.28 509.52 Tm +(operations) Tj +1 0 0 1 376.68 509.52 Tm +(at) Tj +1 0 0 1 386.76 509.52 Tm +(the) Tj +1 0 0 1 401.76 509.52 Tm +(serv) Tj +1 0 0 1 418.32 509.52 Tm +(er) Tj +1 0 0 1 428.88 509.52 Tm +(perform) Tj +1 0 0 1 463.799 509.52 Tm +(some) Tj +1 0 0 1 487.679 509.52 Tm +(computation.) Tj +1 0 0 1 315 497.52 Tm +(But) Tj +1 0 0 1 333.12 497.52 Tm +(the) Tj +1 0 0 1 348.96 497.52 Tm +(o) Tj +1 0 0 1 353.88 497.52 Tm +(v) Tj +1 0 0 1 358.8 497.52 Tm +(erhead) Tj +1 0 0 1 388.8 497.52 Tm +(is) Tj +1 0 0 1 399.12 497.52 Tm +(not) Tj +1 0 0 1 415.56 497.52 Tm +(uniform) Tj +1 0 0 1 451.32 497.52 Tm +(across) Tj +1 0 0 1 479.879 497.52 Tm +(the) Tj +1 0 0 1 495.719 497.52 Tm +(benchmark) Tj +1 0 0 1 315 485.52 Tm +(phases.) Tj +1 0 0 1 351.72 485.52 Tm +(The) Tj +1 0 0 1 371.04 485.52 Tm +(main) Tj +1 0 0 1 394.8 485.52 Tm +(reason) Tj +1 0 0 1 424.68 485.52 Tm +(for) Tj +1 0 0 1 440.16 485.52 Tm +(this) Tj +1 0 0 1 458.4 485.52 Tm +(is) Tj +1 0 0 1 468.96 485.52 Tm +(a) Tj +1 0 0 1 477.24 485.52 Tm +(v) Tj +1 0 0 1 482.04 485.52 Tm +(ariation) Tj +1 0 0 1 516.239 485.52 Tm +(in) Tj +1 0 0 1 527.879 485.52 Tm +(the) Tj +1 0 0 1 315 473.64 Tm +(amount) Tj +1 0 0 1 349.44 473.64 Tm +(of) Tj +1 0 0 1 362.28 473.64 Tm +(time) Tj +1 0 0 1 384.6 473.64 Tm +(the) Tj +1 0 0 1 401.28 473.64 Tm +(client) Tj +1 0 0 1 427.92 473.64 Tm +(spends) Tj +1 0 0 1 459.6 473.64 Tm +(computing) Tj +1 0 0 1 506.759 473.64 Tm +(between) Tj +1 0 0 1 315 461.64 Tm +(operations;) Tj +1 0 0 1 364.2 461.64 Tm +(the) Tj +1 0 0 1 380.52 461.64 Tm +(\256rst) Tj +1 0 0 1 400.2 461.64 Tm +(tw) Tj +1 0 0 1 410.04 461.64 Tm +(o) Tj +1 0 0 1 419.16 461.64 Tm +(phases) Tj +1 0 0 1 449.76 461.64 Tm +(ha) Tj +1 0 0 1 459 461.64 Tm +(v) Tj +1 0 0 1 463.92 461.64 Tm +(e) Tj +1 0 0 1 472.44 461.64 Tm +(a) Tj +1 0 0 1 480.959 461.64 Tm +(higher) Tj +1 0 0 1 510.479 461.64 Tm +(relati) Tj +1 0 0 1 530.759 461.64 Tm +(v) Tj +1 0 0 1 535.679 461.64 Tm +(e) Tj +1 0 0 1 315 449.76 Tm +(o) Tj +1 0 0 1 319.92 449.76 Tm +(v) Tj +1 0 0 1 324.84 449.76 Tm +(erhead) Tj +1 0 0 1 354.6 449.76 Tm +(because) Tj +1 0 0 1 389.4 449.76 Tm +(the) Tj +1 0 0 1 404.88 449.76 Tm +(client) Tj +1 0 0 1 430.44 449.76 Tm +(spends) Tj +1 0 0 1 460.799 449.76 Tm +(approximately) Tj +1 0 0 1 521.759 449.76 Tm +(40%) Tj +1 0 0 1 315 437.76 Tm +(of) Tj +1 0 0 1 325.68 437.76 Tm +(the) Tj +1 0 0 1 340.2 437.76 Tm +(total) Tj +1 0 0 1 360.36 437.76 Tm +(time) Tj +1 0 0 1 380.4 437.76 Tm +(computing) Tj +1 0 0 1 425.4 437.76 Tm +(between) Tj +1 0 0 1 460.919 437.76 Tm +(operations,) Tj +1 0 0 1 507.359 437.76 Tm +(whereas) Tj +1 0 0 1 315 425.88 Tm +(it) Tj +1 0 0 1 322.44 425.88 Tm +(spends) Tj +1 0 0 1 351.6 425.88 Tm +(approximately) Tj +1 0 0 1 411 425.88 Tm +(70%) Tj +1 0 0 1 431.28 425.88 Tm +(during) Tj +1 0 0 1 459.24 425.88 Tm +(the) Tj +1 0 0 1 473.279 425.88 Tm +(last) Tj +1 0 0 1 489.119 425.88 Tm +(three) Tj +1 0 0 1 510.959 425.88 Tm +(phases.) Tj +1 0 0 1 324.96 413.52 Tm +(The) Tj +1 0 0 1 342.96 413.52 Tm +(table) Tj +1 0 0 1 364.68 413.52 Tm +(sho) Tj +1 0 0 1 378.36 413.52 Tm +(ws) Tj +1 0 0 1 391.8 413.52 Tm +(that) Tj +1 0 0 1 409.2 413.52 Tm +(applying) Tj +1 0 0 1 446.519 413.52 Tm +(the) Tj +1 0 0 1 461.159 413.52 Tm +(read-only) Tj +1 0 0 1 501.839 413.52 Tm +(optimiza-) Tj +1 0 0 1 315 401.52 Tm +(tion) Tj +1 0 0 1 333.72 401.52 Tm +(to) Tj +/R734 10 Tf +1 0 0 1 344.64 401.52 Tm +(lookup) Tj +/R725 10 Tf +1 0 0 1 383.76 401.52 Tm +(impro) Tj +1 0 0 1 407.64 401.52 Tm +(v) Tj +1 0 0 1 412.56 401.52 Tm +(es) Tj +1 0 0 1 423.72 401.52 Tm +(the) Tj +1 0 0 1 439.079 401.52 Tm +(performance) Tj +1 0 0 1 492.599 401.52 Tm +(of) Tj +1 0 0 1 504.119 401.52 Tm +(BFS) Tj +1 0 0 1 524.999 401.52 Tm +(sig-) Tj +1 0 0 1 315 389.64 Tm +(ni\256cantly) Tj +1 0 0 1 356.64 389.64 Tm +(and) Tj +1 0 0 1 374.88 389.64 Tm +(reduces) Tj +1 0 0 1 409.32 389.64 Tm +(the) Tj +1 0 0 1 425.4 389.64 Tm +(o) Tj +1 0 0 1 430.32 389.64 Tm +(v) Tj +1 0 0 1 435.24 389.64 Tm +(erhead) Tj +1 0 0 1 465.6 389.64 Tm +(relati) Tj +1 0 0 1 485.879 389.64 Tm +(v) Tj +1 0 0 1 490.799 389.64 Tm +(e) Tj +1 0 0 1 498.959 389.64 Tm +(to) Tj +1 0 0 1 510.719 389.64 Tm +(BFS-nr) Tj +1 0 0 1 315 377.64 Tm +(to) Tj +1 0 0 1 326.52 377.64 Tm +(20%.) Tj +1 0 0 1 354.96 377.64 Tm +(This) Tj +1 0 0 1 376.44 377.64 Tm +(optimization) Tj +1 0 0 1 430.68 377.64 Tm +(has) Tj +1 0 0 1 447.72 377.64 Tm +(a) Tj +1 0 0 1 456 377.64 Tm +(signi\256cant) Tj +1 0 0 1 501.359 377.64 Tm +(impact) Tj +1 0 0 1 532.199 377.64 Tm +(in) Tj +1 0 0 1 315 365.76 Tm +(the) Tj +1 0 0 1 330.6 365.76 Tm +(\256rst) Tj +1 0 0 1 349.44 365.76 Tm +(four) Tj +1 0 0 1 369.36 365.76 Tm +(phases) Tj +1 0 0 1 399.36 365.76 Tm +(because) Tj +1 0 0 1 434.16 365.76 Tm +(the) Tj +1 0 0 1 449.76 365.76 Tm +(time) Tj +1 0 0 1 470.879 365.76 Tm +(spent) Tj +1 0 0 1 495.239 365.76 Tm +(w) Tj +1 0 0 1 502.319 365.76 Tm +(aiting) Tj +1 0 0 1 528.359 365.76 Tm +(for) Tj +/R734 10 Tf +1 0 0 1 315 353.76 Tm +(lookup) Tj +/R725 10 Tf +1 0 0 1 353.64 353.76 Tm +(operations) Tj +1 0 0 1 398.04 353.76 Tm +(to) Tj +1 0 0 1 408.6 353.76 Tm +(complete) Tj +1 0 0 1 447.96 353.76 Tm +(in) Tj +1 0 0 1 458.519 353.76 Tm +(BFS-strict) Tj +1 0 0 1 502.319 353.76 Tm +(is) Tj +1 0 0 1 511.679 353.76 Tm +(at) Tj +1 0 0 1 521.759 353.76 Tm +(least) Tj +1 0 0 1 315 341.88 Tm +(20%) Tj +1 0 0 1 336.48 341.88 Tm +(of) Tj +1 0 0 1 348.12 341.88 Tm +(the) Tj +1 0 0 1 363.48 341.88 Tm +(elapsed) Tj +1 0 0 1 396.6 341.88 Tm +(time) Tj +1 0 0 1 417.6 341.88 Tm +(for) Tj +1 0 0 1 432.48 341.88 Tm +(these) Tj +1 0 0 1 456.24 341.88 Tm +(phases,) Tj +1 0 0 1 488.639 341.88 Tm +(whereas) Tj +1 0 0 1 524.519 341.88 Tm +(it) Tj +1 0 0 1 533.399 341.88 Tm +(is) Tj +1 0 0 1 315 329.88 Tm +(less) Tj +1 0 0 1 332.4 329.88 Tm +(than) Tj +1 0 0 1 352.08 329.88 Tm +(5%) Tj +1 0 0 1 367.8 329.88 Tm +(of) Tj +1 0 0 1 378.6 329.88 Tm +(the) Tj +1 0 0 1 393.36 329.88 Tm +(elapsed) Tj +1 0 0 1 425.64 329.88 Tm +(time) Tj +1 0 0 1 445.919 329.88 Tm +(for) Tj +1 0 0 1 459.959 329.88 Tm +(the) Tj +1 0 0 1 474.599 329.88 Tm +(last) Tj +1 0 0 1 491.039 329.88 Tm +(phase.) Tj +ET +q +184.8 0 0 -0.48 335.148 316.452 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +q +0.48 0 0 -11.04 334.908 315.972 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +q +0.48 0 0 -11.04 367.308 315.972 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +q +0.48 0 0 -11.04 367.308 315.972 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +/R725 9 Tf +1 0 0 1 414.84 308.28 Tm +(BFS) Tj +ET +q +0.48 0 0 -11.04 477.828 315.972 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +q +0.48 0 0 -11.04 519.708 315.972 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +q +110.64 0 0 -0.48 367.428 305.052 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +q +0.48 0 0 -11.04 334.908 305.052 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +1 0 0 1 341.04 297.24 Tm +(phase) Tj +ET +q +0.48 0 0 -11.04 367.308 305.052 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +1 0 0 1 386.16 297.24 Tm +(strict) Tj +ET +q +0.48 0 0 -11.04 422.508 305.052 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +1 0 0 1 431.88 297.24 Tm +(r/o) Tj +1 0 0 1 444 297.24 Tm +(lookup) Tj +ET +q +0.48 0 0 -11.04 477.828 305.052 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +1 0 0 1 483.96 297.24 Tm +(NFS-std) Tj +ET +q +0.48 0 0 -11.04 519.708 305.052 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +q +184.8 0 0 -0.48 335.148 294.132 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +q +0.48 0 0 -11.04 334.908 293.652 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +1 0 0 1 349.08 285.96 Tm +(1) Tj +ET +q +0.48 0 0 -11.04 367.308 293.652 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +1 0 0 1 373.44 285.96 Tm +(0.55) Tj +1 0 0 1 391.44 285.96 Tm +(\(-69%\)) Tj +ET +q +0.48 0 0 -11.04 422.508 293.652 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +1 0 0 1 428.76 285.96 Tm +(0.47) Tj +1 0 0 1 446.64 285.96 Tm +(\(-73%\)) Tj +ET +q +0.48 0 0 -11.04 477.828 293.652 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +1 0 0 1 491.16 285.96 Tm +(1.75) Tj +ET +q +0.48 0 0 -11.04 519.708 293.652 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +q +0.48 0 0 -11.04 334.908 282.732 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +1 0 0 1 349.08 274.92 Tm +(2) Tj +ET +q +0.48 0 0 -11.04 367.308 282.732 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +1 0 0 1 375.72 274.92 Tm +(9.24) Tj +1 0 0 1 393.6 274.92 Tm +(\(-2%\)) Tj +ET +q +0.48 0 0 -11.04 422.508 282.732 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +1 0 0 1 428.76 274.92 Tm +(7.91) Tj +1 0 0 1 446.64 274.92 Tm +(\(-16%\)) Tj +ET +q +0.48 0 0 -11.04 477.828 282.732 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +1 0 0 1 491.16 274.92 Tm +(9.46) Tj +ET +q +0.48 0 0 -11.04 519.708 282.732 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +q +0.48 0 0 -11.04 334.908 271.812 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +1 0 0 1 349.08 264 Tm +(3) Tj +ET +q +0.48 0 0 -11.04 367.308 271.812 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +1 0 0 1 375 264 Tm +(7.24) Tj +1 0 0 1 392.88 264 Tm +(\(35%\)) Tj +ET +q +0.48 0 0 -11.04 422.508 271.812 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +1 0 0 1 430.2 264 Tm +(6.45) Tj +1 0 0 1 448.2 264 Tm +(\(20%\)) Tj +ET +q +0.48 0 0 -11.04 477.828 271.812 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +1 0 0 1 491.16 264 Tm +(5.36) Tj +ET +q +0.48 0 0 -11.04 519.708 271.812 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +q +0.48 0 0 -11.04 334.908 260.772 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +1 0 0 1 349.08 253.08 Tm +(4) Tj +ET +q +0.48 0 0 -11.04 367.308 260.772 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +1 0 0 1 375 253.08 Tm +(8.77) Tj +1 0 0 1 392.88 253.08 Tm +(\(32%\)) Tj +ET +q +0.48 0 0 -11.04 422.508 260.772 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +1 0 0 1 430.2 253.08 Tm +(7.87) Tj +1 0 0 1 448.2 253.08 Tm +(\(19%\)) Tj +ET +q +0.48 0 0 -11.04 477.828 260.772 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +1 0 0 1 491.16 253.08 Tm +(6.60) Tj +ET +q +0.48 0 0 -11.04 519.708 260.772 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +q +0.48 0 0 -11.04 334.908 249.852 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +1 0 0 1 349.08 242.04 Tm +(5) Tj +ET +q +0.48 0 0 -11.04 367.308 249.852 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +1 0 0 1 373.44 242.04 Tm +(38.68) Tj +1 0 0 1 395.88 242.04 Tm +(\(-2%\)) Tj +ET +q +0.48 0 0 -11.04 422.508 249.852 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +1 0 0 1 428.76 242.04 Tm +(38.38) Tj +1 0 0 1 451.2 242.04 Tm +(\(-2%\)) Tj +ET +q +0.48 0 0 -11.04 477.828 249.852 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +1 0 0 1 488.88 242.04 Tm +(39.35) Tj +ET +q +0.48 0 0 -11.04 519.708 249.852 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +q +184.8 0 0 -0.48 335.148 238.932 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +q +0.48 0 0 -11.04 334.908 238.452 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +1 0 0 1 343.32 230.76 Tm +(total) Tj +ET +q +0.48 0 0 -11.04 367.308 238.452 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +1 0 0 1 375 230.76 Tm +(64.48) Tj +1 0 0 1 397.32 230.76 Tm +(\(3%\)) Tj +ET +q +0.48 0 0 -11.04 422.508 238.452 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +1 0 0 1 428.76 230.76 Tm +(61.07) Tj +1 0 0 1 451.2 230.76 Tm +(\(-2%\)) Tj +ET +q +0.48 0 0 -11.04 477.828 238.452 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +1 0 0 1 488.879 230.76 Tm +(62.52) Tj +ET +q +0.48 0 0 -11.04 519.708 238.452 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +q +184.8 0 0 -0.48 335.148 227.532 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +BT +/R725 10 Tf +1 0 0 1 315 213.6 Tm +(T) Tj +1 0 0 1 320.28 213.6 Tm +(able) Tj +1 0 0 1 341.04 213.6 Tm +(3:) Tj +1 0 0 1 355.68 213.6 Tm +(Andre) Tj +1 0 0 1 380.52 213.6 Tm +(w) Tj +1 0 0 1 391.56 213.6 Tm +(benchmark:) Tj +1 0 0 1 445.44 213.6 Tm +(BFS) Tj +1 0 0 1 467.28 213.6 Tm +(vs) Tj +1 0 0 1 480.36 213.6 Tm +(NFS-std.) Tj +1 0 0 1 524.519 213.6 Tm +(The) Tj +1 0 0 1 315 201.6 Tm +(times) Tj +1 0 0 1 339.12 201.6 Tm +(are) Tj +1 0 0 1 353.76 201.6 Tm +(in) Tj +1 0 0 1 363.96 201.6 Tm +(seconds.) Tj +1 0 0 1 324.96 178.92 Tm +(T) Tj +1 0 0 1 330.24 178.92 Tm +(able) Tj +1 0 0 1 349.68 178.92 Tm +(3) Tj +1 0 0 1 357.6 178.92 Tm +(sho) Tj +1 0 0 1 371.28 178.92 Tm +(ws) Tj +1 0 0 1 385.08 178.92 Tm +(the) Tj +1 0 0 1 400.08 178.92 Tm +(results) Tj +1 0 0 1 429 178.92 Tm +(for) Tj +1 0 0 1 443.519 178.92 Tm +(BFS) Tj +1 0 0 1 464.039 178.92 Tm +(vs) Tj +1 0 0 1 475.799 178.92 Tm +(NFS-std.) Tj +1 0 0 1 516.239 178.92 Tm +(These) Tj +1 0 0 1 315 166.92 Tm +(results) Tj +1 0 0 1 344.4 166.92 Tm +(sho) Tj +1 0 0 1 358.08 166.92 Tm +(w) Tj +1 0 0 1 368.52 166.92 Tm +(that) Tj +1 0 0 1 386.88 166.92 Tm +(BFS) Tj +1 0 0 1 408 166.92 Tm +(can) Tj +1 0 0 1 425.16 166.92 Tm +(be) Tj +1 0 0 1 438 166.92 Tm +(used) Tj +1 0 0 1 459.6 166.92 Tm +(in) Tj +1 0 0 1 470.759 166.92 Tm +(practice) Tj +1 0 0 1 505.679 166.92 Tm +(\320) Tj +1 0 0 1 518.999 166.92 Tm +(BFS-) Tj +1 0 0 1 315 155.04 Tm +(strict) Tj +1 0 0 1 339.96 155.04 Tm +(tak) Tj +1 0 0 1 352.08 155.04 Tm +(es) Tj +1 0 0 1 365.52 155.04 Tm +(only) Tj +1 0 0 1 388.32 155.04 Tm +(3%) Tj +1 0 0 1 406.68 155.04 Tm +(more) Tj +1 0 0 1 432.24 155.04 Tm +(time) Tj +1 0 0 1 455.04 155.04 Tm +(to) Tj +1 0 0 1 467.879 155.04 Tm +(run) Tj +1 0 0 1 486.239 155.04 Tm +(the) Tj +1 0 0 1 503.519 155.04 Tm +(complete) Tj +1 0 0 1 315 143.04 Tm +(benchmark.) Tj +1 0 0 1 374.88 143.04 Tm +(Thus,) Tj +1 0 0 1 403.8 143.04 Tm +(one) Tj +1 0 0 1 423.84 143.04 Tm +(could) Tj +1 0 0 1 451.56 143.04 Tm +(replace) Tj +1 0 0 1 486 143.04 Tm +(the) Tj +1 0 0 1 503.879 143.04 Tm +(NFS) Tj +1 0 0 1 527.879 143.04 Tm +(V2) Tj +1 0 0 1 315 131.16 Tm +(implementation) Tj +1 0 0 1 382.56 131.16 Tm +(in) Tj +1 0 0 1 395.28 131.16 Tm +(Digital) Tj +1 0 0 1 427.92 131.16 Tm +(Unix,) Tj +1 0 0 1 455.88 131.16 Tm +(which) Tj +1 0 0 1 485.279 131.16 Tm +(is) Tj +1 0 0 1 496.799 131.16 Tm +(used) Tj +1 0 0 1 520.079 131.16 Tm +(daily) Tj +1 0 0 1 315 119.16 Tm +(by) Tj +1 0 0 1 329.4 119.16 Tm +(man) Tj +1 0 0 1 346.56 119.16 Tm +(y) Tj +1 0 0 1 355.8 119.16 Tm +(users,) Tj +1 0 0 1 383.76 119.16 Tm +(by) Tj +1 0 0 1 398.16 119.16 Tm +(BFS) Tj +1 0 0 1 420.24 119.16 Tm +(without) Tj +1 0 0 1 455.16 119.16 Tm +(af) Tj +1 0 0 1 462.72 119.16 Tm +(fecting) Tj +1 0 0 1 494.76 119.16 Tm +(the) Tj +1 0 0 1 511.319 119.16 Tm +(latenc) Tj +1 0 0 1 535.079 119.16 Tm +(y) Tj +1 0 0 1 315 107.28 Tm +(percei) Tj +1 0 0 1 339.24 107.28 Tm +(v) Tj +1 0 0 1 344.16 107.28 Tm +(ed) Tj +1 0 0 1 357.36 107.28 Tm +(by) Tj +1 0 0 1 371.4 107.28 Tm +(those) Tj +1 0 0 1 396.48 107.28 Tm +(users.) Tj +1 0 0 1 427.56 107.28 Tm +(Furthermore,) Tj +1 0 0 1 484.319 107.28 Tm +(BFS) Tj +1 0 0 1 506.039 107.28 Tm +(with) Tj +1 0 0 1 527.879 107.28 Tm +(the) Tj +1 0 0 1 315 95.28 Tm +(read-only) Tj +1 0 0 1 358.56 95.28 Tm +(optimization) Tj +1 0 0 1 414.36 95.28 Tm +(for) Tj +1 0 0 1 431.4 95.28 Tm +(the) Tj +/R734 10 Tf +1 0 0 1 449.04 95.28 Tm +(lookup) Tj +/R725 10 Tf +1 0 0 1 490.319 95.28 Tm +(operation) Tj +1 0 0 1 533.399 95.28 Tm +(is) Tj +1 0 0 1 315 83.4 Tm +(actually) Tj +1 0 0 1 349.08 83.4 Tm +(2%) Tj +1 0 0 1 364.8 83.4 Tm +(f) Tj +1 0 0 1 368.04 83.4 Tm +(aster) Tj +1 0 0 1 389.28 83.4 Tm +(than) Tj +1 0 0 1 408.96 83.4 Tm +(NFS-std.) Tj +1 0 0 1 324.96 71.04 Tm +(The) Tj +1 0 0 1 344.4 71.04 Tm +(o) Tj +1 0 0 1 349.32 71.04 Tm +(v) Tj +1 0 0 1 354.24 71.04 Tm +(erhead) Tj +1 0 0 1 384.48 71.04 Tm +(of) Tj +1 0 0 1 396.72 71.04 Tm +(BFS) Tj +1 0 0 1 418.44 71.04 Tm +(relati) Tj +1 0 0 1 438.72 71.04 Tm +(v) Tj +1 0 0 1 443.64 71.04 Tm +(e) Tj +1 0 0 1 451.8 71.04 Tm +(to) Tj +1 0 0 1 463.44 71.04 Tm +(NFS-std) Tj +1 0 0 1 500.639 71.04 Tm +(is) Tj +1 0 0 1 511.199 71.04 Tm +(not) Tj +1 0 0 1 527.879 71.04 Tm +(the) Tj +1 0 0 1 300.96 35.16 Tm +(12) Tj +ET +Q +endstream +endobj +736 0 obj +33893 +endobj +737 0 obj +<< +/Type /Page +/MediaBox [0 0 612 792] +/Parent 2 0 R +/Resources << /ProcSet [/PDF /ImageB /Text] +/Font << +/R734 734 0 R +/R731 731 0 R +/R728 728 0 R +/R725 725 0 R +>> +>> +/Contents [ +726 0 R +729 0 R +732 0 R +735 0 R +] +>> +endobj +738 0 obj +<< /Type /Font /Name /R738 /Subtype /Type1 /BaseFont /Times-Roman >> +endobj +739 0 obj +<< /Length 740 0 R >> +stream +q +BT +/R738 10 Tf +1 0 0 1 72 709.08 Tm +(same) Tj +1 0 0 1 96.6 709.08 Tm +(for) Tj +1 0 0 1 112.32 709.08 Tm +(all) Tj +1 0 0 1 126.36 709.08 Tm +(phases.) Tj +1 0 0 1 163.8 709.08 Tm +(Both) Tj +1 0 0 1 187.32 709.08 Tm +(v) Tj +1 0 0 1 192.24 709.08 Tm +(ersions) Tj +1 0 0 1 224.52 709.08 Tm +(of) Tj +1 0 0 1 236.879 709.08 Tm +(BFS) Tj +1 0 0 1 258.719 709.08 Tm +(are) Tj +1 0 0 1 274.919 709.08 Tm +(f) Tj +1 0 0 1 278.159 709.08 Tm +(aster) Tj +1 0 0 1 72 697.08 Tm +(than) Tj +1 0 0 1 92.6399 697.08 Tm +(NFS-std) Tj +1 0 0 1 129.48 697.08 Tm +(for) Tj +1 0 0 1 144.6 697.08 Tm +(phases) Tj +1 0 0 1 174.72 697.08 Tm +(1,) Tj +1 0 0 1 186 697.08 Tm +(2,) Tj +1 0 0 1 197.28 697.08 Tm +(and) Tj +1 0 0 1 215.16 697.08 Tm +(5) Tj +1 0 0 1 223.68 697.08 Tm +(b) Tj +1 0 0 1 228.48 697.08 Tm +(ut) Tj +1 0 0 1 239.759 697.08 Tm +(slo) Tj +1 0 0 1 251.159 697.08 Tm +(wer) Tj +1 0 0 1 269.639 697.08 Tm +(for) Tj +1 0 0 1 284.879 697.08 Tm +(the) Tj +1 0 0 1 72 685.2 Tm +(other) Tj +1 0 0 1 94.9199 685.2 Tm +(phases.) Tj +1 0 0 1 127.56 685.2 Tm +(This) Tj +1 0 0 1 147.72 685.2 Tm +(is) Tj +1 0 0 1 156.84 685.2 Tm +(because) Tj +1 0 0 1 190.92 685.2 Tm +(during) Tj +1 0 0 1 219.359 685.2 Tm +(phases) Tj +1 0 0 1 248.399 685.2 Tm +(1,) Tj +1 0 0 1 258.359 685.2 Tm +(2,) Tj +1 0 0 1 268.319 685.2 Tm +(and) Tj +1 0 0 1 285.119 685.2 Tm +(5) Tj +1 0 0 1 292.559 685.2 Tm +(a) Tj +1 0 0 1 72 673.2 Tm +(lar) Tj +1 0 0 1 82.4399 673.2 Tm +(ge) Tj +1 0 0 1 94.4399 673.2 Tm +(fraction) Tj +1 0 0 1 128.16 673.2 Tm +(\(between) Tj +1 0 0 1 167.28 673.2 Tm +(21%) Tj +1 0 0 1 188.28 673.2 Tm +(and) Tj +1 0 0 1 205.32 673.2 Tm +(40%\)) Tj +1 0 0 1 229.68 673.2 Tm +(of) Tj +1 0 0 1 240.6 673.2 Tm +(the) Tj +1 0 0 1 255.479 673.2 Tm +(operations) Tj +1 0 0 1 72 661.32 Tm +(issued) Tj +1 0 0 1 99.4799 661.32 Tm +(by) Tj +1 0 0 1 111.96 661.32 Tm +(the) Tj +1 0 0 1 126.6 661.32 Tm +(client) Tj +1 0 0 1 151.32 661.32 Tm +(are) Tj +ET +endstream +endobj +740 0 obj +1832 +endobj +741 0 obj +<< /Type /Font /Name /R741 /Subtype /Type1 /BaseFont /Times-Italic >> +endobj +742 0 obj +<< /Length 743 0 R >> +stream +BT +/R741 10 Tf +1 0 0 1 165.96 661.32 Tm +(sync) Tj +1 0 0 1 183.6 661.32 Tm +(hr) Tj +1 0 0 1 192 661.32 Tm +(onous) Tj +/R738 10 Tf +1 0 0 1 216 661.32 Tm +(,) Tj +1 0 0 1 220.8 661.32 Tm +(i.e.,) Tj +1 0 0 1 237.959 661.32 Tm +(operations) Tj +1 0 0 1 281.999 661.32 Tm +(that) Tj +1 0 0 1 72 649.32 Tm +(require) Tj +1 0 0 1 104.4 649.32 Tm +(the) Tj +1 0 0 1 120.84 649.32 Tm +(NFS) Tj +1 0 0 1 143.4 649.32 Tm +(implementation) Tj +1 0 0 1 210.12 649.32 Tm +(to) Tj +1 0 0 1 222.12 649.32 Tm +(ensure) Tj +1 0 0 1 252.359 649.32 Tm +(stability) Tj +1 0 0 1 288.719 649.32 Tm +(of) Tj +1 0 0 1 72 637.44 Tm +(modi\256ed) Tj +1 0 0 1 110.88 637.44 Tm +(\256le) Tj +1 0 0 1 127.08 637.44 Tm +(system) Tj +1 0 0 1 158.28 637.44 Tm +(state) Tj +1 0 0 1 180 637.44 Tm +(before) Tj +1 0 0 1 208.8 637.44 Tm +(replying) Tj +1 0 0 1 245.519 637.44 Tm +(to) Tj +1 0 0 1 256.679 637.44 Tm +(the) Tj +1 0 0 1 272.399 637.44 Tm +(client.) Tj +1 0 0 1 72 625.44 Tm +(NFS-std) Tj +1 0 0 1 108.84 625.44 Tm +(achie) Tj +1 0 0 1 129.72 625.44 Tm +(v) Tj +1 0 0 1 134.64 625.44 Tm +(es) Tj +1 0 0 1 146.4 625.44 Tm +(stability) Tj +1 0 0 1 182.16 625.44 Tm +(by) Tj +1 0 0 1 195.84 625.44 Tm +(writing) Tj +1 0 0 1 228.24 625.44 Tm +(modi\256ed) Tj +1 0 0 1 267.359 625.44 Tm +(state) Tj +1 0 0 1 289.199 625.44 Tm +(to) Tj +1 0 0 1 72 613.44 Tm +(disk) Tj +1 0 0 1 91.9199 613.44 Tm +(whereas) Tj +1 0 0 1 127.92 613.44 Tm +(BFS) Tj +1 0 0 1 148.92 613.44 Tm +(achie) Tj +1 0 0 1 169.8 613.44 Tm +(v) Tj +1 0 0 1 174.72 613.44 Tm +(es) Tj +1 0 0 1 186.12 613.44 Tm +(stability) Tj +1 0 0 1 221.64 613.44 Tm +(with) Tj +1 0 0 1 242.64 613.44 Tm +(lo) Tj +1 0 0 1 250.199 613.44 Tm +(wer) Tj +1 0 0 1 268.319 613.44 Tm +(latenc) Tj +1 0 0 1 292.079 613.44 Tm +(y) Tj +1 0 0 1 72 601.56 Tm +(using) Tj +1 0 0 1 95.6399 601.56 Tm +(replication) Tj +1 0 0 1 140.4 601.56 Tm +(\(as) Tj +1 0 0 1 154.08 601.56 Tm +(in) Tj +1 0 0 1 163.92 601.56 Tm +(Harp) Tj +1 0 0 1 185.88 601.56 Tm +([20) Tj +1 0 0 1 199.2 601.56 Tm +(]\).) Tj +1 0 0 1 211.68 601.56 Tm +(NFS-std) Tj +1 0 0 1 246.959 601.56 Tm +(is) Tj +1 0 0 1 255.719 601.56 Tm +(f) Tj +1 0 0 1 258.959 601.56 Tm +(aster) Tj +1 0 0 1 279.839 601.56 Tm +(than) Tj +1 0 0 1 72 589.56 Tm +(BFS) Tj +1 0 0 1 93 589.56 Tm +(\(and) Tj +1 0 0 1 113.88 589.56 Tm +(BFS-nr\)) Tj +1 0 0 1 149.76 589.56 Tm +(in) Tj +1 0 0 1 160.8 589.56 Tm +(phases) Tj +1 0 0 1 190.56 589.56 Tm +(3) Tj +1 0 0 1 198.84 589.56 Tm +(and) Tj +1 0 0 1 216.48 589.56 Tm +(4) Tj +1 0 0 1 224.64 589.56 Tm +(because) Tj +1 0 0 1 259.439 589.56 Tm +(the) Tj +1 0 0 1 274.919 589.56 Tm +(client) Tj +1 0 0 1 72 577.68 Tm +(issues) Tj +1 0 0 1 98.2799 577.68 Tm +(no) Tj +1 0 0 1 110.76 577.68 Tm +(synchronous) Tj +1 0 0 1 163.56 577.68 Tm +(operations) Tj +1 0 0 1 207.6 577.68 Tm +(during) Tj +1 0 0 1 236.04 577.68 Tm +(these) Tj +1 0 0 1 259.079 577.68 Tm +(phases.) Tj +ET +endstream +endobj +743 0 obj +2727 +endobj +744 0 obj +<< /Type /Font /Name /R744 /Subtype /Type1 /BaseFont /Times-Bold >> +endobj +745 0 obj +<< /Length 746 0 R >> +stream +BT +/R744 12 Tf +1 0 0 1 72 546.6 Tm +(8) Tj +1 0 0 1 89.8799 546.6 Tm +(Related) Tj +1 0 0 1 132.12 546.6 Tm +(W) Tj +1 0 0 1 143.28 546.6 Tm +(ork) Tj +/R738 10 Tf +1 0 0 1 72 528.12 Tm +(Most) Tj +1 0 0 1 96.5999 528.12 Tm +(pre) Tj +1 0 0 1 109.2 528.12 Tm +(vious) Tj +1 0 0 1 134.64 528.12 Tm +(w) Tj +1 0 0 1 141.72 528.12 Tm +(ork) Tj +1 0 0 1 159.12 528.12 Tm +(on) Tj +1 0 0 1 173.16 528.12 Tm +(replication) Tj +1 0 0 1 219.84 528.12 Tm +(techniques) Tj +1 0 0 1 266.519 528.12 Tm +(ignored) Tj +1 0 0 1 72 516.12 Tm +(Byzantine) Tj +1 0 0 1 118.68 516.12 Tm +(f) Tj +1 0 0 1 121.92 516.12 Tm +(aults) Tj +1 0 0 1 146.88 516.12 Tm +(or) Tj +1 0 0 1 161.52 516.12 Tm +(assumed) Tj +1 0 0 1 202.08 516.12 Tm +(a) Tj +1 0 0 1 212.76 516.12 Tm +(synchronous) Tj +1 0 0 1 269.28 516.12 Tm +(system) Tj +1 0 0 1 72 504.24 Tm +(model) Tj +1 0 0 1 101.88 504.24 Tm +(\(e.g.,) Tj +1 0 0 1 127.68 504.24 Tm +([17) Tj +1 0 0 1 140.88 504.24 Tm +(,) Tj +1 0 0 1 148.32 504.24 Tm +(26) Tj +1 0 0 1 158.28 504.24 Tm +(,) Tj +1 0 0 1 165.72 504.24 Tm +(18) Tj +1 0 0 1 175.68 504.24 Tm +(,) Tj +1 0 0 1 183.12 504.24 Tm +(34) Tj +1 0 0 1 193.08 504.24 Tm +(,) Tj +1 0 0 1 200.52 504.24 Tm +(6,) Tj +1 0 0 1 213 504.24 Tm +(10) Tj +1 0 0 1 222.96 504.24 Tm +(]\).) Tj +1 0 0 1 243.12 504.24 Tm +(V) Tj +1 0 0 1 249.72 504.24 Tm +(ie) Tj +1 0 0 1 256.68 504.24 Tm +(wstamped) Tj +1 0 0 1 72 492.24 Tm +(replication) Tj +1 0 0 1 117 492.24 Tm +([26) Tj +1 0 0 1 130.2 492.24 Tm +(]) Tj +1 0 0 1 135.96 492.24 Tm +(and) Tj +1 0 0 1 152.64 492.24 Tm +(P) Tj +1 0 0 1 158.04 492.24 Tm +(axos) Tj +1 0 0 1 178.68 492.24 Tm +([18) Tj +1 0 0 1 192 492.24 Tm +(]) Tj +1 0 0 1 197.64 492.24 Tm +(use) Tj +1 0 0 1 213.239 492.24 Tm +(vie) Tj +1 0 0 1 225.239 492.24 Tm +(ws) Tj +1 0 0 1 238.559 492.24 Tm +(with) Tj +1 0 0 1 258.719 492.24 Tm +(a) Tj +1 0 0 1 265.439 492.24 Tm +(primary) Tj +1 0 0 1 72 480.36 Tm +(and) Tj +1 0 0 1 88.8 480.36 Tm +(backups) Tj +1 0 0 1 123.84 480.36 Tm +(to) Tj +1 0 0 1 134.04 480.36 Tm +(tolerate) Tj +1 0 0 1 166.32 480.36 Tm +(benign) Tj +1 0 0 1 195.84 480.36 Tm +(f) Tj +1 0 0 1 199.08 480.36 Tm +(aults) Tj +1 0 0 1 220.199 480.36 Tm +(in) Tj +1 0 0 1 230.399 480.36 Tm +(an) Tj +1 0 0 1 242.159 480.36 Tm +(asynchronous) Tj +1 0 0 1 72 468.36 Tm +(system.) Tj +1 0 0 1 105.48 468.36 Tm +(T) Tj +1 0 0 1 110.76 468.36 Tm +(olerating) Tj +1 0 0 1 147.96 468.36 Tm +(Byzantine) Tj +1 0 0 1 190.08 468.36 Tm +(f) Tj +1 0 0 1 193.32 468.36 Tm +(aults) Tj +1 0 0 1 213.72 468.36 Tm +(requires) Tj +1 0 0 1 247.559 468.36 Tm +(a) Tj +1 0 0 1 253.559 468.36 Tm +(mu) Tj +1 0 0 1 266.279 468.36 Tm +(ch) Tj +1 0 0 1 277.199 468.36 Tm +(mo) Tj +1 0 0 1 289.919 468.36 Tm +(re) Tj +1 0 0 1 72 456.48 Tm +(comple) Tj +1 0 0 1 101.4 456.48 Tm +(x) Tj +1 0 0 1 109.32 456.48 Tm +(protocol) Tj +1 0 0 1 145.8 456.48 Tm +(with) Tj +1 0 0 1 166.68 456.48 Tm +(cryptographic) Tj +1 0 0 1 225.24 456.48 Tm +(authentication,) Tj +1 0 0 1 287.639 456.48 Tm +(an) Tj +1 0 0 1 72 444.48 Tm +(e) Tj +1 0 0 1 76.32 444.48 Tm +(xtra) Tj +1 0 0 1 96.5999 444.48 Tm +(pre-prepare) Tj +1 0 0 1 147.48 444.48 Tm +(phase,) Tj +1 0 0 1 178.08 444.48 Tm +(and) Tj +1 0 0 1 197.4 444.48 Tm +(a) Tj +1 0 0 1 206.64 444.48 Tm +(dif) Tj +1 0 0 1 217.559 444.48 Tm +(ferent) Tj +1 0 0 1 245.639 444.48 Tm +(technique) Tj +1 0 0 1 289.199 444.48 Tm +(to) Tj +1 0 0 1 72 432.6 Tm +(trigger) Tj +1 0 0 1 101.28 432.6 Tm +(vie) Tj +1 0 0 1 113.28 432.6 Tm +(w) Tj +1 0 0 1 123.12 432.6 Tm +(changes) Tj +1 0 0 1 157.92 432.6 Tm +(and) Tj +1 0 0 1 174.96 432.6 Tm +(select) Tj +1 0 0 1 200.4 432.6 Tm +(primaries.) Tj +1 0 0 1 244.679 432.6 Tm +(Furthermore,) Tj +1 0 0 1 72 420.6 Tm +(our) Tj +1 0 0 1 87 420.6 Tm +(system) Tj +1 0 0 1 116.28 420.6 Tm +(uses) Tj +1 0 0 1 135 420.6 Tm +(vie) Tj +1 0 0 1 147 420.6 Tm +(w) Tj +1 0 0 1 155.76 420.6 Tm +(changes) Tj +1 0 0 1 189.6 420.6 Tm +(on) Tj +1 0 0 1 199.56 420.6 Tm +(ly) Tj +1 0 0 1 208.92 420.6 Tm +(to) Tj +1 0 0 1 218.159 420.6 Tm +(select) Tj +1 0 0 1 242.399 420.6 Tm +(a) Tj +1 0 0 1 248.279 420.6 Tm +(ne) Tj +1 0 0 1 257.519 420.6 Tm +(w) Tj +1 0 0 1 266.159 420.6 Tm +(pr) Tj +1 0 0 1 274.439 420.6 Tm +(imar) Tj +1 0 0 1 292.679 420.6 Tm +(y) Tj +1 0 0 1 72 408.6 Tm +(b) Tj +1 0 0 1 76.8 408.6 Tm +(ut) Tj +1 0 0 1 87.4799 408.6 Tm +(ne) Tj +1 0 0 1 96.7199 408.6 Tm +(v) Tj +1 0 0 1 101.64 408.6 Tm +(er) Tj +1 0 0 1 112.08 408.6 Tm +(to) Tj +1 0 0 1 122.76 408.6 Tm +(select) Tj +1 0 0 1 148.44 408.6 Tm +(a) Tj +1 0 0 1 155.76 408.6 Tm +(dif) Tj +1 0 0 1 166.68 408.6 Tm +(ferent) Tj +1 0 0 1 192.719 408.6 Tm +(set) Tj +1 0 0 1 206.759 408.6 Tm +(of) Tj +1 0 0 1 217.919 408.6 Tm +(replicas) Tj +1 0 0 1 251.879 408.6 Tm +(to) Tj +1 0 0 1 262.559 408.6 Tm +(form) Tj +1 0 0 1 284.879 408.6 Tm +(the) Tj +1 0 0 1 72 396.72 Tm +(ne) Tj +1 0 0 1 81.24 396.72 Tm +(w) Tj +1 0 0 1 90.8399 396.72 Tm +(vie) Tj +1 0 0 1 102.84 396.72 Tm +(w) Tj +1 0 0 1 112.44 396.72 Tm +(as) Tj +1 0 0 1 123.24 396.72 Tm +(in) Tj +1 0 0 1 133.44 396.72 Tm +([26) Tj +1 0 0 1 146.76 396.72 Tm +(,) Tj +1 0 0 1 151.8 396.72 Tm +(18) Tj +1 0 0 1 161.76 396.72 Tm +(].) Tj +1 0 0 1 81.96 383.64 Tm +(Some) Tj +1 0 0 1 109.32 383.64 Tm +(agreement) Tj +1 0 0 1 155.52 383.64 Tm +(and) Tj +1 0 0 1 174.72 383.64 Tm +(consensus) Tj +1 0 0 1 219.84 383.64 Tm +(algorithms) Tj +1 0 0 1 267.12 383.64 Tm +(tolerate) Tj +1 0 0 1 72 371.76 Tm +(Byzantine) Tj +1 0 0 1 114.36 371.76 Tm +(f) Tj +1 0 0 1 117.6 371.76 Tm +(aults) Tj +1 0 0 1 138.24 371.76 Tm +(in) Tj +1 0 0 1 147.96 371.76 Tm +(asynchronous) Tj +1 0 0 1 204.6 371.76 Tm +(systems) Tj +1 0 0 1 238.079 371.76 Tm +(\(e.g,[2) Tj +1 0 0 1 264.119 371.76 Tm +(,) Tj +1 0 0 1 268.559 371.76 Tm +(3) Tj +1 0 0 1 273.479 371.76 Tm +(,) Tj +1 0 0 1 277.919 371.76 Tm +(24) Tj +1 0 0 1 287.879 371.76 Tm +(]\).) Tj +1 0 0 1 72 359.76 Tm +(Ho) Tj +1 0 0 1 84 359.76 Tm +(we) Tj +1 0 0 1 95.4 359.76 Tm +(v) Tj +1 0 0 1 100.32 359.76 Tm +(er) Tj +1 0 0 1 107.76 359.76 Tm +(,) Tj +1 0 0 1 114.36 359.76 Tm +(the) Tj +1 0 0 1 126.48 359.76 Tm +(y) Tj +1 0 0 1 135.36 359.76 Tm +(do) Tj +1 0 0 1 149.28 359.76 Tm +(not) Tj +1 0 0 1 166.08 359.76 Tm +(pro) Tj +1 0 0 1 179.4 359.76 Tm +(vide) Tj +1 0 0 1 200.28 359.76 Tm +(a) Tj +1 0 0 1 208.8 359.76 Tm +(complete) Tj +1 0 0 1 249.239 359.76 Tm +(solution) Tj +1 0 0 1 285.359 359.76 Tm +(for) Tj +1 0 0 1 72 347.88 Tm +(state) Tj +1 0 0 1 92.5199 347.88 Tm +(machine) Tj +1 0 0 1 128.52 347.88 Tm +(replication,) Tj +1 0 0 1 175.92 347.88 Tm +(and) Tj +1 0 0 1 192.48 347.88 Tm +(furthermore,) Tj +1 0 0 1 244.919 347.88 Tm +(most) Tj +1 0 0 1 266.519 347.88 Tm +(of) Tj +1 0 0 1 277.079 347.88 Tm +(them) Tj +1 0 0 1 72 335.88 Tm +(were) Tj +1 0 0 1 94.3199 335.88 Tm +(designed) Tj +1 0 0 1 132.6 335.88 Tm +(to) Tj +1 0 0 1 143.28 335.88 Tm +(demonstrate) Tj +1 0 0 1 194.88 335.88 Tm +(theoretical) Tj +1 0 0 1 239.879 335.88 Tm +(feasibility) Tj +1 0 0 1 282.599 335.88 Tm +(and) Tj +1 0 0 1 72 324 Tm +(are) Tj +1 0 0 1 89.0399 324 Tm +(too) Tj +1 0 0 1 106.68 324 Tm +(slo) Tj +1 0 0 1 118.08 324 Tm +(w) Tj +1 0 0 1 130.2 324 Tm +(to) Tj +1 0 0 1 142.8 324 Tm +(be) Tj +1 0 0 1 157.2 324 Tm +(used) Tj +1 0 0 1 180.36 324 Tm +(in) Tj +1 0 0 1 192.96 324 Tm +(practice.) Tj +1 0 0 1 237.839 324 Tm +(Our) Tj +1 0 0 1 258.239 324 Tm +(algorithm) Tj +1 0 0 1 72 312 Tm +(during) Tj +1 0 0 1 100.56 312 Tm +(normal-case) Tj +1 0 0 1 151.8 312 Tm +(operation) Tj +1 0 0 1 192 312 Tm +(is) Tj +1 0 0 1 201.24 312 Tm +(similar) Tj +1 0 0 1 231.48 312 Tm +(to) Tj +1 0 0 1 241.799 312 Tm +(the) Tj +1 0 0 1 256.559 312 Tm +(Byzantine) Tj +1 0 0 1 72 300.12 Tm +(agreement) Tj +1 0 0 1 116.76 300.12 Tm +(algorithm) Tj +1 0 0 1 158.76 300.12 Tm +(in) Tj +1 0 0 1 169.68 300.12 Tm +([2]) Tj +1 0 0 1 184.56 300.12 Tm +(b) Tj +1 0 0 1 189.36 300.12 Tm +(ut) Tj +1 0 0 1 200.4 300.12 Tm +(that) Tj +1 0 0 1 218.52 300.12 Tm +(algorithm) Tj +1 0 0 1 260.519 300.12 Tm +(is) Tj +1 0 0 1 270.479 300.12 Tm +(unable) Tj +1 0 0 1 72 288.12 Tm +(to) Tj +1 0 0 1 82.2 288.12 Tm +(survi) Tj +1 0 0 1 102 288.12 Tm +(v) Tj +1 0 0 1 106.92 288.12 Tm +(e) Tj +1 0 0 1 113.64 288.12 Tm +(primary) Tj +1 0 0 1 147.72 288.12 Tm +(f) Tj +1 0 0 1 150.96 288.12 Tm +(ailures.) Tj +1 0 0 1 81.96 275.16 Tm +(The) Tj +1 0 0 1 100.8 275.16 Tm +(tw) Tj +1 0 0 1 110.64 275.16 Tm +(o) Tj +1 0 0 1 119.04 275.16 Tm +(systems) Tj +1 0 0 1 153.96 275.16 Tm +(that) Tj +1 0 0 1 172.32 275.16 Tm +(are) Tj +1 0 0 1 187.8 275.16 Tm +(most) Tj +1 0 0 1 210.48 275.16 Tm +(closely) Tj +1 0 0 1 242.159 275.16 Tm +(related) Tj +1 0 0 1 272.639 275.16 Tm +(to) Tj +1 0 0 1 283.679 275.16 Tm +(our) Tj +1 0 0 1 72 263.16 Tm +(w) Tj +1 0 0 1 79.08 263.16 Tm +(ork) Tj +1 0 0 1 95.04 263.16 Tm +(are) Tj +1 0 0 1 109.92 263.16 Tm +(Rampart) Tj +1 0 0 1 146.88 263.16 Tm +([29) Tj +1 0 0 1 160.2 263.16 Tm +(,) Tj +1 0 0 1 165.36 263.16 Tm +(30) Tj +1 0 0 1 175.32 263.16 Tm +(,) Tj +1 0 0 1 180.48 263.16 Tm +(31) Tj +1 0 0 1 190.44 263.16 Tm +(,) Tj +1 0 0 1 195.72 263.16 Tm +(22) Tj +1 0 0 1 205.68 263.16 Tm +(]) Tj +1 0 0 1 211.68 263.16 Tm +(and) Tj +1 0 0 1 228.72 263.16 Tm +(SecureRing) Tj +1 0 0 1 277.919 263.16 Tm +([16) Tj +1 0 0 1 291.239 263.16 Tm +(].) Tj +1 0 0 1 72 251.28 Tm +(The) Tj +1 0 0 1 87.48 251.28 Tm +(y) Tj +1 0 0 1 95.5199 251.28 Tm +(implement) Tj +1 0 0 1 141.36 251.28 Tm +(state) Tj +1 0 0 1 162.72 251.28 Tm +(machine) Tj +1 0 0 1 199.68 251.28 Tm +(replication) Tj +1 0 0 1 245.519 251.28 Tm +(b) Tj +1 0 0 1 250.319 251.28 Tm +(ut) Tj +1 0 0 1 261.239 251.28 Tm +(are) Tj +1 0 0 1 276.479 251.28 Tm +(more) Tj +1 0 0 1 72 239.28 Tm +(than) Tj +1 0 0 1 92.0399 239.28 Tm +(an) Tj +1 0 0 1 104.28 239.28 Tm +(order) Tj +1 0 0 1 128.16 239.28 Tm +(of) Tj +1 0 0 1 139.32 239.28 Tm +(magnitude) Tj +1 0 0 1 184.2 239.28 Tm +(slo) Tj +1 0 0 1 195.6 239.28 Tm +(wer) Tj +1 0 0 1 213.48 239.28 Tm +(than) Tj +1 0 0 1 233.4 239.28 Tm +(our) Tj +1 0 0 1 249.599 239.28 Tm +(system) Tj +1 0 0 1 280.079 239.28 Tm +(and,) Tj +1 0 0 1 72 227.4 Tm +(most) Tj +1 0 0 1 93.8399 227.4 Tm +(importantly) Tj +1 0 0 1 140.04 227.4 Tm +(,) Tj +1 0 0 1 144.72 227.4 Tm +(the) Tj +1 0 0 1 156.84 227.4 Tm +(y) Tj +1 0 0 1 164.16 227.4 Tm +(rely) Tj +1 0 0 1 182.16 227.4 Tm +(on) Tj +1 0 0 1 194.64 227.4 Tm +(synchron) Tj +1 0 0 1 231.36 227.4 Tm +(y) Tj +1 0 0 1 238.439 227.4 Tm +(assumptions.) Tj +1 0 0 1 81.96 214.32 Tm +(Both) Tj +1 0 0 1 105.84 214.32 Tm +(Rampart) Tj +1 0 0 1 144.72 214.32 Tm +(and) Tj +1 0 0 1 163.56 214.32 Tm +(SecureRing) Tj +1 0 0 1 214.56 214.32 Tm +(must) Tj +1 0 0 1 238.56 214.32 Tm +(e) Tj +1 0 0 1 242.88 214.32 Tm +(xclude) Tj +1 0 0 1 273.839 214.32 Tm +(f) Tj +1 0 0 1 277.079 214.32 Tm +(aulty) Tj +1 0 0 1 72 202.32 Tm +(replicas) Tj +1 0 0 1 105 202.32 Tm +(from) Tj +1 0 0 1 126.48 202.32 Tm +(the) Tj +1 0 0 1 140.76 202.32 Tm +(group) Tj +1 0 0 1 166.08 202.32 Tm +(to) Tj +1 0 0 1 175.92 202.32 Tm +(mak) Tj +1 0 0 1 193.08 202.32 Tm +(e) Tj +1 0 0 1 199.44 202.32 Tm +(progress) Tj +1 0 0 1 235.199 202.32 Tm +(\(e.g.,) Tj +1 0 0 1 257.639 202.32 Tm +(to) Tj +1 0 0 1 267.479 202.32 Tm +(remo) Tj +1 0 0 1 287.999 202.32 Tm +(v) Tj +1 0 0 1 292.919 202.32 Tm +(e) Tj +1 0 0 1 72 190.44 Tm +(a) Tj +1 0 0 1 80.16 190.44 Tm +(f) Tj +1 0 0 1 83.3999 190.44 Tm +(aulty) Tj +1 0 0 1 107.04 190.44 Tm +(primary) Tj +1 0 0 1 142.32 190.44 Tm +(and) Tj +1 0 0 1 160.32 190.44 Tm +(elect) Tj +1 0 0 1 182.88 190.44 Tm +(a) Tj +1 0 0 1 191.04 190.44 Tm +(ne) Tj +1 0 0 1 200.28 190.44 Tm +(w) Tj +1 0 0 1 211.079 190.44 Tm +(one\),) Tj +1 0 0 1 235.319 190.44 Tm +(and) Tj +1 0 0 1 253.439 190.44 Tm +(to) Tj +1 0 0 1 264.959 190.44 Tm +(perform) Tj +1 0 0 1 72 178.44 Tm +(garbage) Tj +1 0 0 1 109.44 178.44 Tm +(collection.) Tj +1 0 0 1 165.36 178.44 Tm +(The) Tj +1 0 0 1 180.84 178.44 Tm +(y) Tj +1 0 0 1 191.64 178.44 Tm +(rely) Tj +1 0 0 1 213.12 178.44 Tm +(on) Tj +1 0 0 1 229.08 178.44 Tm +(f) Tj +1 0 0 1 232.32 178.44 Tm +(ailure) Tj +1 0 0 1 260.999 178.44 Tm +(detectors) Tj +1 0 0 1 72 166.56 Tm +(to) Tj +1 0 0 1 86.16 166.56 Tm +(determine) Tj +1 0 0 1 132.36 166.56 Tm +(which) Tj +1 0 0 1 162.96 166.56 Tm +(replicas) Tj +1 0 0 1 200.4 166.56 Tm +(are) Tj +1 0 0 1 218.88 166.56 Tm +(f) Tj +1 0 0 1 222.12 166.56 Tm +(aulty) Tj +1 0 0 1 241.56 166.56 Tm +(.) Tj +1 0 0 1 259.08 166.56 Tm +(Ho) Tj +1 0 0 1 271.079 166.56 Tm +(we) Tj +1 0 0 1 282.479 166.56 Tm +(v) Tj +1 0 0 1 287.399 166.56 Tm +(er) Tj +1 0 0 1 294.839 166.56 Tm +(,) Tj +1 0 0 1 72 154.56 Tm +(f) Tj +1 0 0 1 75.24 154.56 Tm +(ailure) Tj +1 0 0 1 101.04 154.56 Tm +(detectors) Tj +1 0 0 1 140.16 154.56 Tm +(cannot) Tj +1 0 0 1 169.92 154.56 Tm +(be) Tj +1 0 0 1 182.4 154.56 Tm +(accurate) Tj +1 0 0 1 218.759 154.56 Tm +(in) Tj +1 0 0 1 229.679 154.56 Tm +(an) Tj +1 0 0 1 242.159 154.56 Tm +(asynchronous) Tj +1 0 0 1 72 142.68 Tm +(system) Tj +1 0 0 1 101.88 142.68 Tm +([21) Tj +1 0 0 1 115.08 142.68 Tm +(],) Tj +1 0 0 1 123.24 142.68 Tm +(i.e.,) Tj +1 0 0 1 140.04 142.68 Tm +(the) Tj +1 0 0 1 152.16 142.68 Tm +(y) Tj +1 0 0 1 159.24 142.68 Tm +(may) Tj +1 0 0 1 178.68 142.68 Tm +(misclassify) Tj +1 0 0 1 225.6 142.68 Tm +(a) Tj +1 0 0 1 232.199 142.68 Tm +(replica) Tj +1 0 0 1 261.479 142.68 Tm +(as) Tj +1 0 0 1 272.039 142.68 Tm +(f) Tj +1 0 0 1 275.279 142.68 Tm +(aulty) Tj +1 0 0 1 294.719 142.68 Tm +(.) Tj +1 0 0 1 72 130.68 Tm +(Since) Tj +1 0 0 1 97.1999 130.68 Tm +(correctness) Tj +1 0 0 1 145.2 130.68 Tm +(requires) Tj +1 0 0 1 180.36 130.68 Tm +(that) Tj +1 0 0 1 198.48 130.68 Tm +(fe) Tj +1 0 0 1 206.04 130.68 Tm +(wer) Tj +1 0 0 1 224.04 130.68 Tm +(than) Tj +1 0 0 1 244.319 130.68 Tm +(1) Tj +ET +Q +q +W +0 0 612 792 re +n +endstream +endobj +746 0 obj +12837 +endobj +747 0 obj +<< /Type /XObject /Name /R747 /Subtype /Image /Length 748 0 R +/ImageMask true /Width 39 /Height 100 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 39 /BlackIs1 true >>] +>> +stream +02pd&_uAQfhsfOthtP%fhtQ$thtQ%&_uAQfhqUH1s02O;s1\Mfs1\N;s1\NIs+^Qfs02Iq'ECB$~> +endstream +endobj +748 0 obj +78 +endobj +749 0 obj +<< /Length 750 0 R >> +stream +q +3.9 0 0 -10 249.9 138.2 cm +/R747 Do +Q +BT +/R738 10 Tf +1 0 0 1 254.399 130.68 Tm +(3) Tj +1 0 0 1 262.319 130.68 Tm +(of) Tj +1 0 0 1 273.719 130.68 Tm +(group) Tj +1 0 0 1 72 118.8 Tm +(members) Tj +1 0 0 1 111.84 118.8 Tm +(be) Tj +1 0 0 1 124.56 118.8 Tm +(f) Tj +1 0 0 1 127.8 118.8 Tm +(aulty) Tj +1 0 0 1 147.24 118.8 Tm +(,) Tj +1 0 0 1 153.12 118.8 Tm +(a) Tj +1 0 0 1 160.8 118.8 Tm +(misclassi\256cation) Tj +1 0 0 1 230.64 118.8 Tm +(can) Tj +1 0 0 1 247.8 118.8 Tm +(compromise) Tj +1 0 0 1 72 106.8 Tm +(correctness) Tj +1 0 0 1 120.72 106.8 Tm +(by) Tj +1 0 0 1 134.64 106.8 Tm +(remo) Tj +1 0 0 1 155.16 106.8 Tm +(ving) Tj +1 0 0 1 176.64 106.8 Tm +(a) Tj +1 0 0 1 185.04 106.8 Tm +(non-f) Tj +1 0 0 1 206.76 106.8 Tm +(aulty) Tj +1 0 0 1 230.399 106.8 Tm +(replica) Tj +1 0 0 1 261.479 106.8 Tm +(from) Tj +1 0 0 1 284.879 106.8 Tm +(the) Tj +1 0 0 1 72 94.92 Tm +(group.) Tj +1 0 0 1 107.64 94.92 Tm +(This) Tj +1 0 0 1 129.96 94.92 Tm +(opens) Tj +1 0 0 1 157.8 94.92 Tm +(an) Tj +1 0 0 1 171.84 94.92 Tm +(a) Tj +1 0 0 1 176.04 94.92 Tm +(v) Tj +1 0 0 1 180.96 94.92 Tm +(enue) Tj +1 0 0 1 204.24 94.92 Tm +(of) Tj +1 0 0 1 217.2 94.92 Tm +(attack:) Tj +1 0 0 1 251.52 94.92 Tm +(an) Tj +1 0 0 1 265.559 94.92 Tm +(attack) Tj +1 0 0 1 289.319 94.92 Tm +(er) Tj +1 0 0 1 72 82.92 Tm +(gains) Tj +1 0 0 1 96.4799 82.92 Tm +(control) Tj +1 0 0 1 128.16 82.92 Tm +(o) Tj +1 0 0 1 133.08 82.92 Tm +(v) Tj +1 0 0 1 138 82.92 Tm +(er) Tj +1 0 0 1 149.04 82.92 Tm +(a) Tj +1 0 0 1 156.96 82.92 Tm +(single) Tj +1 0 0 1 184.2 82.92 Tm +(replica) Tj +1 0 0 1 214.799 82.92 Tm +(b) Tj +1 0 0 1 219.599 82.92 Tm +(ut) Tj +1 0 0 1 230.879 82.92 Tm +(does) Tj +1 0 0 1 252.599 82.92 Tm +(not) Tj +1 0 0 1 268.799 82.92 Tm +(change) Tj +1 0 0 1 72 71.04 Tm +(its) Tj +1 0 0 1 84.3599 71.04 Tm +(beha) Tj +1 0 0 1 103.08 71.04 Tm +(vior) Tj +1 0 0 1 121.92 71.04 Tm +(in) Tj +1 0 0 1 132.6 71.04 Tm +(an) Tj +1 0 0 1 141.96 71.04 Tm +(y) Tj +1 0 0 1 149.76 71.04 Tm +(detectable) Tj +1 0 0 1 193.08 71.04 Tm +(w) Tj +1 0 0 1 200.16 71.04 Tm +(ay;) Tj +1 0 0 1 215.4 71.04 Tm +(then) Tj +1 0 0 1 235.559 71.04 Tm +(it) Tj +1 0 0 1 243.959 71.04 Tm +(slo) Tj +1 0 0 1 255.359 71.04 Tm +(ws) Tj +1 0 0 1 269.399 71.04 Tm +(correct) Tj +1 0 0 1 315 709.08 Tm +(replicas) Tj +1 0 0 1 347.64 709.08 Tm +(or) Tj +1 0 0 1 357.6 709.08 Tm +(the) Tj +1 0 0 1 371.4 709.08 Tm +(communication) Tj +1 0 0 1 435.36 709.08 Tm +(b) Tj +1 0 0 1 440.28 709.08 Tm +(etween) Tj +1 0 0 1 470.04 709.08 Tm +(them) Tj +1 0 0 1 491.519 709.08 Tm +(un) Tj +1 0 0 1 501.479 709.08 Tm +(til) Tj +1 0 0 1 511.319 709.08 Tm +(eno) Tj +1 0 0 1 525.719 709.08 Tm +(ug) Tj +1 0 0 1 535.679 709.08 Tm +(h) Tj +1 0 0 1 315 697.08 Tm +(are) Tj +1 0 0 1 329.64 697.08 Tm +(e) Tj +1 0 0 1 333.96 697.08 Tm +(xcluded) Tj +1 0 0 1 367.92 697.08 Tm +(from) Tj +1 0 0 1 389.76 697.08 Tm +(the) Tj +1 0 0 1 404.52 697.08 Tm +(group.) Tj +1 0 0 1 324.96 685.2 Tm +(T) Tj +1 0 0 1 330.24 685.2 Tm +(o) Tj +1 0 0 1 338.4 685.2 Tm +(reduce) Tj +1 0 0 1 368.16 685.2 Tm +(the) Tj +1 0 0 1 383.52 685.2 Tm +(probability) Tj +1 0 0 1 430.32 685.2 Tm +(of) Tj +1 0 0 1 441.84 685.2 Tm +(misclassi\256cation,) Tj +1 0 0 1 514.079 685.2 Tm +(f) Tj +1 0 0 1 517.319 685.2 Tm +(ailure) Tj +1 0 0 1 315 673.2 Tm +(detectors) Tj +1 0 0 1 353.76 673.2 Tm +(can) Tj +1 0 0 1 370.44 673.2 Tm +(be) Tj +1 0 0 1 382.68 673.2 Tm +(calibrated) Tj +1 0 0 1 424.68 673.2 Tm +(to) Tj +1 0 0 1 435.24 673.2 Tm +(delay) Tj +1 0 0 1 459.719 673.2 Tm +(classifying) Tj +1 0 0 1 505.679 673.2 Tm +(a) Tj +1 0 0 1 512.879 673.2 Tm +(replica) Tj +1 0 0 1 315 661.32 Tm +(as) Tj +1 0 0 1 326.64 661.32 Tm +(f) Tj +1 0 0 1 329.88 661.32 Tm +(aulty) Tj +1 0 0 1 349.32 661.32 Tm +(.) Tj +1 0 0 1 357.6 661.32 Tm +(Ho) Tj +1 0 0 1 369.6 661.32 Tm +(we) Tj +1 0 0 1 381 661.32 Tm +(v) Tj +1 0 0 1 385.92 661.32 Tm +(er) Tj +1 0 0 1 393.36 661.32 Tm +(,) Tj +1 0 0 1 399.12 661.32 Tm +(for) Tj +1 0 0 1 414 661.32 Tm +(the) Tj +1 0 0 1 429.48 661.32 Tm +(probability) Tj +1 0 0 1 476.52 661.32 Tm +(to) Tj +1 0 0 1 487.559 661.32 Tm +(be) Tj +1 0 0 1 500.279 661.32 Tm +(ne) Tj +1 0 0 1 509.639 661.32 Tm +(gligible) Tj +1 0 0 1 315 649.32 Tm +(the) Tj +1 0 0 1 330.24 649.32 Tm +(delay) Tj +1 0 0 1 354.84 649.32 Tm +(must) Tj +1 0 0 1 377.16 649.32 Tm +(be) Tj +1 0 0 1 389.64 649.32 Tm +(v) Tj +1 0 0 1 394.56 649.32 Tm +(ery) Tj +1 0 0 1 410.28 649.32 Tm +(lar) Tj +1 0 0 1 420.72 649.32 Tm +(ge,) Tj +1 0 0 1 435.6 649.32 Tm +(which) Tj +1 0 0 1 462.959 649.32 Tm +(is) Tj +1 0 0 1 472.679 649.32 Tm +(undesirable.) Tj +1 0 0 1 526.319 649.32 Tm +(F) Tj +1 0 0 1 531.719 649.32 Tm +(or) Tj +1 0 0 1 315 637.44 Tm +(e) Tj +1 0 0 1 319.32 637.44 Tm +(xample,) Tj +1 0 0 1 353.16 637.44 Tm +(if) Tj +1 0 0 1 361.32 637.44 Tm +(the) Tj +1 0 0 1 375.48 637.44 Tm +(primary) Tj +1 0 0 1 409.08 637.44 Tm +(has) Tj +1 0 0 1 424.32 637.44 Tm +(actually) Tj +1 0 0 1 457.919 637.44 Tm +(f) Tj +1 0 0 1 461.159 637.44 Tm +(ailed,) Tj +1 0 0 1 485.039 637.44 Tm +(the) Tj +1 0 0 1 499.199 637.44 Tm +(group) Tj +1 0 0 1 524.519 637.44 Tm +(will) Tj +1 0 0 1 315 625.44 Tm +(be) Tj +1 0 0 1 327.84 625.44 Tm +(unable) Tj +1 0 0 1 357.96 625.44 Tm +(to) Tj +1 0 0 1 369.12 625.44 Tm +(process) Tj +1 0 0 1 402.48 625.44 Tm +(client) Tj +1 0 0 1 428.16 625.44 Tm +(requests) Tj +1 0 0 1 464.279 625.44 Tm +(until) Tj +1 0 0 1 485.999 625.44 Tm +(the) Tj +1 0 0 1 501.719 625.44 Tm +(delay) Tj +1 0 0 1 526.679 625.44 Tm +(has) Tj +1 0 0 1 315 613.44 Tm +(e) Tj +1 0 0 1 319.32 613.44 Tm +(xpired.) Tj +1 0 0 1 351 613.44 Tm +(Our) Tj +1 0 0 1 369.12 613.44 Tm +(algorithm) Tj +1 0 0 1 410.4 613.44 Tm +(is) Tj +1 0 0 1 419.64 613.44 Tm +(not) Tj +1 0 0 1 434.88 613.44 Tm +(vulnerable) Tj +1 0 0 1 479.519 613.44 Tm +(to) Tj +1 0 0 1 489.839 613.44 Tm +(this) Tj +1 0 0 1 506.759 613.44 Tm +(problem) Tj +1 0 0 1 315 601.56 Tm +(because) Tj +1 0 0 1 348.36 601.56 Tm +(it) Tj +1 0 0 1 355.8 601.56 Tm +(ne) Tj +1 0 0 1 365.04 601.56 Tm +(v) Tj +1 0 0 1 369.96 601.56 Tm +(er) Tj +1 0 0 1 379.32 601.56 Tm +(needs) Tj +1 0 0 1 403.92 601.56 Tm +(to) Tj +1 0 0 1 413.52 601.56 Tm +(e) Tj +1 0 0 1 417.84 601.56 Tm +(xclude) Tj +1 0 0 1 446.159 601.56 Tm +(replicas) Tj +1 0 0 1 479.039 601.56 Tm +(from) Tj +1 0 0 1 500.279 601.56 Tm +(the) Tj +1 0 0 1 514.319 601.56 Tm +(group.) Tj +1 0 0 1 324.96 589.56 Tm +(Phalanx) Tj +1 0 0 1 362.4 589.56 Tm +([23) Tj +1 0 0 1 375.72 589.56 Tm +(,) Tj +1 0 0 1 383.52 589.56 Tm +(25) Tj +1 0 0 1 393.48 589.56 Tm +(]) Tj +1 0 0 1 402.24 589.56 Tm +(applies) Tj +1 0 0 1 435.72 589.56 Tm +(quorum) Tj +1 0 0 1 472.079 589.56 Tm +(replication) Tj +1 0 0 1 520.079 589.56 Tm +(tech-) Tj +1 0 0 1 315 577.68 Tm +(niques) Tj +1 0 0 1 343.44 577.68 Tm +([12) Tj +1 0 0 1 356.76 577.68 Tm +(]) Tj +1 0 0 1 362.4 577.68 Tm +(to) Tj +1 0 0 1 372.6 577.68 Tm +(achie) Tj +1 0 0 1 393.48 577.68 Tm +(v) Tj +1 0 0 1 398.4 577.68 Tm +(e) Tj +1 0 0 1 405 577.68 Tm +(Byzantine) Tj +1 0 0 1 447.84 577.68 Tm +(f) Tj +1 0 0 1 451.079 577.68 Tm +(ault-tolerance) Tj +1 0 0 1 508.199 577.68 Tm +(in) Tj +1 0 0 1 518.399 577.68 Tm +(asyn-) Tj +1 0 0 1 315 565.68 Tm +(chronous) Tj +1 0 0 1 354.84 565.68 Tm +(systems.) Tj +1 0 0 1 394.8 565.68 Tm +(This) Tj +1 0 0 1 415.92 565.68 Tm +(w) Tj +1 0 0 1 423 565.68 Tm +(ork) Tj +1 0 0 1 439.56 565.68 Tm +(does) Tj +1 0 0 1 461.04 565.68 Tm +(not) Tj +1 0 0 1 477.12 565.68 Tm +(pro) Tj +1 0 0 1 490.44 565.68 Tm +(vide) Tj +1 0 0 1 510.719 565.68 Tm +(generic) Tj +1 0 0 1 315 553.8 Tm +(state) Tj +1 0 0 1 335.76 553.8 Tm +(machine) Tj +1 0 0 1 371.88 553.8 Tm +(replication;) Tj +1 0 0 1 419.76 553.8 Tm +(instead,) Tj +1 0 0 1 453 553.8 Tm +(it) Tj +1 0 0 1 460.919 553.8 Tm +(of) Tj +1 0 0 1 469.079 553.8 Tm +(fers) Tj +1 0 0 1 486.359 553.8 Tm +(a) Tj +1 0 0 1 493.319 553.8 Tm +(data) Tj +1 0 0 1 512.279 553.8 Tm +(reposi-) Tj +1 0 0 1 315 541.8 Tm +(tory) Tj +1 0 0 1 332.76 541.8 Tm +(with) Tj +1 0 0 1 352.32 541.8 Tm +(operations) Tj +1 0 0 1 395.52 541.8 Tm +(to) Tj +1 0 0 1 405.12 541.8 Tm +(read) Tj +1 0 0 1 423.96 541.8 Tm +(and) Tj +1 0 0 1 440.16 541.8 Tm +(write) Tj +1 0 0 1 462.359 541.8 Tm +(indi) Tj +1 0 0 1 477.719 541.8 Tm +(vidual) Tj +1 0 0 1 504.239 541.8 Tm +(v) Tj +1 0 0 1 509.039 541.8 Tm +(ariables) Tj +1 0 0 1 315 529.92 Tm +(and) Tj +1 0 0 1 332.16 529.92 Tm +(to) Tj +1 0 0 1 342.72 529.92 Tm +(acquire) Tj +1 0 0 1 374.76 529.92 Tm +(locks.) Tj +1 0 0 1 402.84 529.92 Tm +(The) Tj +1 0 0 1 421.08 529.92 Tm +(semantics) Tj +1 0 0 1 463.2 529.92 Tm +(it) Tj +1 0 0 1 471.479 529.92 Tm +(pro) Tj +1 0 0 1 484.799 529.92 Tm +(vides) Tj +1 0 0 1 508.439 529.92 Tm +(for) Tj +1 0 0 1 522.839 529.92 Tm +(read) Tj +1 0 0 1 315 517.92 Tm +(and) Tj +1 0 0 1 331.32 517.92 Tm +(write) Tj +1 0 0 1 353.88 517.92 Tm +(operations) Tj +1 0 0 1 397.32 517.92 Tm +(are) Tj +1 0 0 1 411.48 517.92 Tm +(weak) Tj +1 0 0 1 432.48 517.92 Tm +(er) Tj +1 0 0 1 442.08 517.92 Tm +(than) Tj +1 0 0 1 461.279 517.92 Tm +(those) Tj +1 0 0 1 484.319 517.92 Tm +(of) Tj +1 0 0 1 492.479 517.92 Tm +(fered) Tj +1 0 0 1 514.799 517.92 Tm +(by) Tj +1 0 0 1 526.679 517.92 Tm +(our) Tj +1 0 0 1 315 506.04 Tm +(algorithm;) Tj +1 0 0 1 358.56 506.04 Tm +(we) Tj +1 0 0 1 371.88 506.04 Tm +(can) Tj +1 0 0 1 387.48 506.04 Tm +(implement) Tj +1 0 0 1 431.88 506.04 Tm +(arbitrary) Tj +1 0 0 1 467.879 506.04 Tm +(operations) Tj +1 0 0 1 511.199 506.04 Tm +(that) Tj +1 0 0 1 527.879 506.04 Tm +(ac-) Tj +1 0 0 1 315 494.04 Tm +(cess) Tj +1 0 0 1 333.12 494.04 Tm +(an) Tj +1 0 0 1 342.48 494.04 Tm +(y) Tj +1 0 0 1 349.08 494.04 Tm +(number) Tj +1 0 0 1 381.36 494.04 Tm +(of) Tj +1 0 0 1 391.32 494.04 Tm +(v) Tj +1 0 0 1 396.12 494.04 Tm +(ariables,) Tj +1 0 0 1 430.8 494.04 Tm +(whereas) Tj +1 0 0 1 465.119 494.04 Tm +(in) Tj +1 0 0 1 474.479 494.04 Tm +(Phalanx) Tj +1 0 0 1 508.319 494.04 Tm +(it) Tj +1 0 0 1 515.399 494.04 Tm +(w) Tj +1 0 0 1 522.479 494.04 Tm +(ould) Tj +1 0 0 1 315 482.16 Tm +(be) Tj +1 0 0 1 326.64 482.16 Tm +(necessary) Tj +1 0 0 1 367.68 482.16 Tm +(to) Tj +1 0 0 1 377.76 482.16 Tm +(acquire) Tj +1 0 0 1 409.32 482.16 Tm +(and) Tj +1 0 0 1 426 482.16 Tm +(release) Tj +1 0 0 1 455.999 482.16 Tm +(locks) Tj +1 0 0 1 479.279 482.16 Tm +(to) Tj +1 0 0 1 489.359 482.16 Tm +(e) Tj +1 0 0 1 493.679 482.16 Tm +(x) Tj +1 0 0 1 498.599 482.16 Tm +(ecute) Tj +1 0 0 1 521.759 482.16 Tm +(such) Tj +1 0 0 1 315 470.16 Tm +(operations.) Tj +1 0 0 1 366.12 470.16 Tm +(There) Tj +1 0 0 1 393 470.16 Tm +(are) Tj +1 0 0 1 408.84 470.16 Tm +(no) Tj +1 0 0 1 422.52 470.16 Tm +(published) Tj +1 0 0 1 465 470.16 Tm +(performance) Tj +1 0 0 1 518.999 470.16 Tm +(num-) Tj +1 0 0 1 315 458.28 Tm +(bers) Tj +1 0 0 1 335.04 458.28 Tm +(for) Tj +1 0 0 1 350.16 458.28 Tm +(Phalanx) Tj +1 0 0 1 385.8 458.28 Tm +(b) Tj +1 0 0 1 390.6 458.28 Tm +(ut) Tj +1 0 0 1 401.88 458.28 Tm +(we) Tj +1 0 0 1 417 458.28 Tm +(belie) Tj +1 0 0 1 436.2 458.28 Tm +(v) Tj +1 0 0 1 441.12 458.28 Tm +(e) Tj +1 0 0 1 448.8 458.28 Tm +(our) Tj +1 0 0 1 465.599 458.28 Tm +(algorithm) Tj +1 0 0 1 507.839 458.28 Tm +(is) Tj +1 0 0 1 517.919 458.28 Tm +(f) Tj +1 0 0 1 521.159 458.28 Tm +(aster) Tj +1 0 0 1 315 446.28 Tm +(because) Tj +1 0 0 1 349.68 446.28 Tm +(it) Tj +1 0 0 1 358.32 446.28 Tm +(has) Tj +1 0 0 1 374.76 446.28 Tm +(fe) Tj +1 0 0 1 382.32 446.28 Tm +(wer) Tj +1 0 0 1 400.32 446.28 Tm +(message) Tj +1 0 0 1 437.28 446.28 Tm +(delays) Tj +1 0 0 1 465.84 446.28 Tm +(in) Tj +1 0 0 1 476.76 446.28 Tm +(the) Tj +1 0 0 1 491.999 446.28 Tm +(critical) Tj +1 0 0 1 522.839 446.28 Tm +(path) Tj +1 0 0 1 315 434.4 Tm +(and) Tj +1 0 0 1 332.4 434.4 Tm +(because) Tj +1 0 0 1 366.96 434.4 Tm +(of) Tj +1 0 0 1 378.36 434.4 Tm +(our) Tj +1 0 0 1 394.68 434.4 Tm +(use) Tj +1 0 0 1 410.88 434.4 Tm +(of) Tj +1 0 0 1 422.28 434.4 Tm +(MA) Tj +1 0 0 1 438 434.4 Tm +(Cs) Tj +1 0 0 1 451.44 434.4 Tm +(rather) Tj +1 0 0 1 477.719 434.4 Tm +(than) Tj +1 0 0 1 497.879 434.4 Tm +(public) Tj +1 0 0 1 525.839 434.4 Tm +(k) Tj +1 0 0 1 530.759 434.4 Tm +(e) Tj +1 0 0 1 535.079 434.4 Tm +(y) Tj +1 0 0 1 315 422.4 Tm +(cryptography) Tj +1 0 0 1 368.04 422.4 Tm +(.) Tj +1 0 0 1 373.44 422.4 Tm +(The) Tj +1 0 0 1 391.32 422.4 Tm +(approach) Tj +1 0 0 1 430.2 422.4 Tm +(in) Tj +1 0 0 1 440.28 422.4 Tm +(Phalanx) Tj +1 0 0 1 474.719 422.4 Tm +(of) Tj +1 0 0 1 482.879 422.4 Tm +(fers) Tj +1 0 0 1 500.039 422.4 Tm +(the) Tj +1 0 0 1 514.559 422.4 Tm +(poten-) Tj +1 0 0 1 315 410.52 Tm +(tial) Tj +1 0 0 1 330.24 410.52 Tm +(for) Tj +1 0 0 1 344.28 410.52 Tm +(impro) Tj +1 0 0 1 368.16 410.52 Tm +(v) Tj +1 0 0 1 373.08 410.52 Tm +(ed) Tj +1 0 0 1 384.6 410.52 Tm +(scalability;) Tj +1 0 0 1 430.8 410.52 Tm +(each) Tj +1 0 0 1 451.559 410.52 Tm +(operation) Tj +1 0 0 1 491.639 410.52 Tm +(is) Tj +1 0 0 1 500.759 410.52 Tm +(processed) Tj +1 0 0 1 315 398.52 Tm +(by) Tj +1 0 0 1 327.72 398.52 Tm +(only) Tj +1 0 0 1 348.36 398.52 Tm +(a) Tj +1 0 0 1 355.56 398.52 Tm +(subset) Tj +1 0 0 1 383.28 398.52 Tm +(of) Tj +1 0 0 1 394.44 398.52 Tm +(replicas.) Tj +1 0 0 1 432.48 398.52 Tm +(But) Tj +1 0 0 1 449.639 398.52 Tm +(this) Tj +1 0 0 1 466.919 398.52 Tm +(approach) Tj +1 0 0 1 506.159 398.52 Tm +(to) Tj +1 0 0 1 516.719 398.52 Tm +(scala-) Tj +1 0 0 1 315 386.52 Tm +(bility) Tj +1 0 0 1 339 386.52 Tm +(is) Tj +1 0 0 1 348.6 386.52 Tm +(e) Tj +1 0 0 1 352.92 386.52 Tm +(xpensi) Tj +1 0 0 1 378.84 386.52 Tm +(v) Tj +1 0 0 1 383.76 386.52 Tm +(e:) Tj +1 0 0 1 395.04 386.52 Tm +(it) Tj +1 0 0 1 403.56 386.52 Tm +(requires) Tj +ET +endstream +endobj +750 0 obj +12702 +endobj +751 0 obj +<< /Type /XObject /Name /R751 /Subtype /Image /Length 752 0 R +/ImageMask true /Width 54 /Height 46 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 54 /BlackIs1 true >>] +>> +stream +4N]t,$m>7rl/"o@d!re, +pYTohhg]lis4_o!aocV_^Yur>rVsXfR^`8 +endstream +endobj +752 0 obj +102 +endobj +753 0 obj +<< /Length 754 0 R >> +stream +q +5.4 0 0 -4.6 438.8 391 cm +/R751 Do +Q +endstream +endobj +754 0 obj +39 +endobj +755 0 obj +<< /Type /XObject /Name /R755 /Subtype /Image /Length 756 0 R +/ImageMask true /Width 61 /Height 61 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 61 /BlackIs1 true >>] +>> +stream +,Ld57i8DMjL(5&>G[TII_hJX +endstream +endobj +756 0 obj +114 +endobj +757 0 obj +<< /Length 758 0 R >> +stream +q +6.1 0 0 -6.1 448.8 392.1 cm +/R755 Do +Q +BT +1 0 0 1 458.999 386.52 Tm +(4) Tj +ET +endstream +endobj +758 0 obj +80 +endobj +759 0 obj +<< /Type /XObject /Name /R759 /Subtype /Image /Length 760 0 R +/ImageMask true /Width 50 /Height 92 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 50 /BlackIs1 true >>] +>> +stream +5$eL7-b:K3p0<%? +h1u#R]2'Mug].5is7$!eqtL-hqu?EZs5/V3iQXZVs7/ihqu?EZs4QRbJe-ans*o2/4DdB_@6aqq5Q~> +endstream +endobj +760 0 obj +96 +endobj +761 0 obj +<< /Length 762 0 R >> +stream +q +5 0 0 -9.2 464.4 393.6 cm +/R759 Do +Q +endstream +endobj +762 0 obj +39 +endobj +763 0 obj +<< /Type /XObject /Name /R763 /Subtype /Image /Length 764 0 R +/ImageMask true /Width 66 /Height 65 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 66 /BlackIs1 true >>] +>> +stream +,BZ6Ns8W-!s8W-!s-Eq9Iaj?!s8W-!s8W-!s8Uf'#QTA~> +endstream +endobj +764 0 obj +47 +endobj +765 0 obj +<< /Length 766 0 R >> +stream +q +6.6 0 0 -6.5 472.9 392.3 cm +/R763 Do +Q +BT +1 0 0 1 482.639 386.52 Tm +(1) Tj +1 0 0 1 490.559 386.52 Tm +(to) Tj +1 0 0 1 501.239 386.52 Tm +(tolerate) Tj +ET +q +5 0 0 -9.2 534.4 393.6 cm +/R759 Do +Q +BT +1 0 0 1 315 374.64 Tm +(f) Tj +1 0 0 1 318.24 374.64 Tm +(aults;) Tj +1 0 0 1 342 374.64 Tm +(each) Tj +1 0 0 1 362.4 374.64 Tm +(replica) Tj +1 0 0 1 391.56 374.64 Tm +(needs) Tj +1 0 0 1 416.28 374.64 Tm +(a) Tj +1 0 0 1 422.88 374.64 Tm +(cop) Tj +1 0 0 1 437.279 374.64 Tm +(y) Tj +1 0 0 1 444.239 374.64 Tm +(of) Tj +1 0 0 1 454.559 374.64 Tm +(the) Tj +1 0 0 1 468.839 374.64 Tm +(state;) Tj +1 0 0 1 492.119 374.64 Tm +(and) Tj +1 0 0 1 508.559 374.64 Tm +(the) Tj +1 0 0 1 522.839 374.64 Tm +(load) Tj +1 0 0 1 315 362.64 Tm +(on) Tj +1 0 0 1 327.24 362.64 Tm +(each) Tj +1 0 0 1 347.88 362.64 Tm +(replica) Tj +1 0 0 1 377.28 362.64 Tm +(decreases) Tj +1 0 0 1 417.84 362.64 Tm +(slo) Tj +1 0 0 1 429.24 362.64 Tm +(wly) Tj +1 0 0 1 446.52 362.64 Tm +(with) Tj +ET +q +5.4 0 0 -4.6 466.8 367.1 cm +/R751 Do +Q +BT +1 0 0 1 474.839 362.64 Tm +(\(it) Tj +1 0 0 1 485.999 362.64 Tm +(is) Tj +1 0 0 1 494.999 362.64 Tm +(O) Tj +ET +endstream +endobj +766 0 obj +1078 +endobj +767 0 obj +<< /Type /XObject /Name /R767 /Subtype /Image /Length 768 0 R +/ImageMask true /Width 24 /Height 99 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 24 /BlackIs1 true >>] +>> +stream ++rF]%*LO8IHL@*j8[taqp!<~> +endstream +endobj +768 0 obj +80 +endobj +769 0 obj +<< /Length 770 0 R >> +stream +q +2.4 0 0 -9.9 503.1 370.1 cm +/R767 Do +Q +BT +1 0 0 1 506.039 362.64 Tm +(1) Tj +ET +q +3.9 0 0 -10 511.6 370.1 cm +/R747 Do +Q +endstream +endobj +770 0 obj +120 +endobj +771 0 obj +<< /Type /XObject /Name /R771 /Subtype /Image /Length 772 0 R +/ImageMask true /Width 81 /Height 101 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 81 /BlackIs1 true >>] +>> +stream +-4b4uVs4$gpeTs@FarU6?em+!o_%k6iI9W"?U;:& +ks+gXYO>7?Ktk;+$eg@*p9#FpY3\,`pNM8F?c.A;p)i,bIguA6fRMZThqUEalp&,ZIrP;"Yl33N!< +endstream +endobj +772 0 obj +123 +endobj +773 0 obj +<< /Length 774 0 R >> +stream +q +8.1 0 0 -10.1 516.6 370.3 cm +/R771 Do +Q +q +6 0 0 -0.48 524.268 370.332 cm +BI +/IM true /W 1 /H 1 /BPC 1 /F [/A85] +ID !!~> +EI +Q +q +5.4 0 0 -4.6 524.5 367.1 cm +/R751 Do +Q +endstream +endobj +774 0 obj +168 +endobj +775 0 obj +<< /Type /XObject /Name /R775 /Subtype /Image /Length 776 0 R +/ImageMask true /Width 24 /Height 99 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 24 /BlackIs1 true >>] +>> +stream +,JEO$loTRL4niC`hqS$`rnmKps6nkFp\Y!js6KW5s0(&RK)W@\(KK(`l\2@U?Bm&k<5,",!$D~> +endstream +endobj +776 0 obj +76 +endobj +777 0 obj +<< /Length 778 0 R >> +stream +q +2.4 0 0 -9.9 530.7 370.1 cm +/R775 Do +Q +BT +1 0 0 1 534.12 362.64 Tm +(\).) Tj +/R744 12 Tf +1 0 0 1 315 338.28 Tm +(9) Tj +1 0 0 1 332.88 338.28 Tm +(Conclusions) Tj +/R738 10 Tf +1 0 0 1 315 321.84 Tm +(This) Tj +1 0 0 1 334.8 321.84 Tm +(paper) Tj +1 0 0 1 359.16 321.84 Tm +(has) Tj +1 0 0 1 374.52 321.84 Tm +(described) Tj +1 0 0 1 414.84 321.84 Tm +(a) Tj +1 0 0 1 421.44 321.84 Tm +(ne) Tj +1 0 0 1 430.68 321.84 Tm +(w) Tj +1 0 0 1 439.92 321.84 Tm +(state-machine) Tj +1 0 0 1 497.399 321.84 Tm +(replication) Tj +1 0 0 1 315 309.84 Tm +(algorithm) Tj +1 0 0 1 355.92 309.84 Tm +(that) Tj +1 0 0 1 373.08 309.84 Tm +(is) Tj +1 0 0 1 381.96 309.84 Tm +(able) Tj +1 0 0 1 400.68 309.84 Tm +(to) Tj +1 0 0 1 410.64 309.84 Tm +(tolerate) Tj +1 0 0 1 442.799 309.84 Tm +(Byzantine) Tj +1 0 0 1 485.399 309.84 Tm +(f) Tj +1 0 0 1 488.639 309.84 Tm +(aults) Tj +1 0 0 1 509.639 309.84 Tm +(and) Tj +1 0 0 1 526.199 309.84 Tm +(can) Tj +1 0 0 1 315 297.96 Tm +(be) Tj +1 0 0 1 328.2 297.96 Tm +(used) Tj +1 0 0 1 350.28 297.96 Tm +(in) Tj +1 0 0 1 361.8 297.96 Tm +(practice:) Tj +1 0 0 1 402.36 297.96 Tm +(it) Tj +1 0 0 1 411.6 297.96 Tm +(is) Tj +1 0 0 1 422.04 297.96 Tm +(the) Tj +1 0 0 1 438 297.96 Tm +(\256rst) Tj +1 0 0 1 457.32 297.96 Tm +(to) Tj +1 0 0 1 468.84 297.96 Tm +(w) Tj +1 0 0 1 475.92 297.96 Tm +(ork) Tj +1 0 0 1 493.079 297.96 Tm +(correctly) Tj +1 0 0 1 532.199 297.96 Tm +(in) Tj +1 0 0 1 315 285.96 Tm +(an) Tj +1 0 0 1 326.76 285.96 Tm +(asynchronous) Tj +1 0 0 1 383.76 285.96 Tm +(system) Tj +1 0 0 1 413.76 285.96 Tm +(lik) Tj +1 0 0 1 424.2 285.96 Tm +(e) Tj +1 0 0 1 430.92 285.96 Tm +(the) Tj +1 0 0 1 445.44 285.96 Tm +(Internet) Tj +1 0 0 1 478.679 285.96 Tm +(and) Tj +1 0 0 1 495.359 285.96 Tm +(it) Tj +1 0 0 1 503.159 285.96 Tm +(impro) Tj +1 0 0 1 527.039 285.96 Tm +(v) Tj +1 0 0 1 531.959 285.96 Tm +(es) Tj +1 0 0 1 315 274.08 Tm +(the) Tj +1 0 0 1 329.76 274.08 Tm +(performance) Tj +1 0 0 1 382.68 274.08 Tm +(of) Tj +1 0 0 1 393.48 274.08 Tm +(pre) Tj +1 0 0 1 406.08 274.08 Tm +(vious) Tj +1 0 0 1 430.08 274.08 Tm +(algorithms) Tj +1 0 0 1 475.319 274.08 Tm +(by) Tj +1 0 0 1 487.799 274.08 Tm +(more) Tj +1 0 0 1 510.839 274.08 Tm +(than) Tj +1 0 0 1 530.639 274.08 Tm +(an) Tj +1 0 0 1 315 262.08 Tm +(order) Tj +1 0 0 1 338.52 262.08 Tm +(of) Tj +1 0 0 1 349.32 262.08 Tm +(magnitude.) Tj +1 0 0 1 324.96 250.2 Tm +(The) Tj +1 0 0 1 345.96 250.2 Tm +(paper) Tj +1 0 0 1 373.68 250.2 Tm +(also) Tj +1 0 0 1 395.28 250.2 Tm +(described) Tj +1 0 0 1 439.08 250.2 Tm +(BFS,) Tj +1 0 0 1 464.88 250.2 Tm +(a) Tj +1 0 0 1 474.84 250.2 Tm +(Byzantine-f) Tj +1 0 0 1 521.999 250.2 Tm +(ault-) Tj +1 0 0 1 315 238.2 Tm +(tolerant) Tj +1 0 0 1 348.36 238.2 Tm +(implementation) Tj +1 0 0 1 413.88 238.2 Tm +(of) Tj +1 0 0 1 425.16 238.2 Tm +(NFS.) Tj +1 0 0 1 448.8 238.2 Tm +(BFS) Tj +1 0 0 1 469.56 238.2 Tm +(demonstrates) Tj +1 0 0 1 524.999 238.2 Tm +(that) Tj +1 0 0 1 315 226.32 Tm +(it) Tj +1 0 0 1 325.2 226.32 Tm +(is) Tj +1 0 0 1 336.48 226.32 Tm +(possible) Tj +1 0 0 1 373.8 226.32 Tm +(to) Tj +1 0 0 1 386.16 226.32 Tm +(use) Tj +1 0 0 1 404.04 226.32 Tm +(our) Tj +1 0 0 1 422.04 226.32 Tm +(algorithm) Tj +1 0 0 1 465.359 226.32 Tm +(to) Tj +1 0 0 1 477.839 226.32 Tm +(implement) Tj +1 0 0 1 525.119 226.32 Tm +(real) Tj +1 0 0 1 315 214.32 Tm +(services) Tj +1 0 0 1 348.96 214.32 Tm +(with) Tj +1 0 0 1 368.52 214.32 Tm +(performance) Tj +1 0 0 1 420.72 214.32 Tm +(close) Tj +1 0 0 1 443.04 214.32 Tm +(to) Tj +1 0 0 1 452.64 214.32 Tm +(that) Tj +1 0 0 1 469.319 214.32 Tm +(of) Tj +1 0 0 1 479.519 214.32 Tm +(an) Tj +1 0 0 1 490.799 214.32 Tm +(unreplicated) Tj +1 0 0 1 315 202.32 Tm +(service) Tj +1 0 0 1 345.24 202.32 Tm +(\320) Tj +1 0 0 1 357.12 202.32 Tm +(the) Tj +1 0 0 1 371.28 202.32 Tm +(performance) Tj +1 0 0 1 423.6 202.32 Tm +(of) Tj +1 0 0 1 433.92 202.32 Tm +(BFS) Tj +1 0 0 1 453.6 202.32 Tm +(is) Tj +1 0 0 1 462.24 202.32 Tm +(only) Tj +1 0 0 1 481.919 202.32 Tm +(3%) Tj +1 0 0 1 497.159 202.32 Tm +(w) Tj +1 0 0 1 504.239 202.32 Tm +(orse) Tj +1 0 0 1 522.839 202.32 Tm +(than) Tj +1 0 0 1 315 190.44 Tm +(that) Tj +1 0 0 1 332.04 190.44 Tm +(of) Tj +1 0 0 1 342.48 190.44 Tm +(the) Tj +1 0 0 1 356.88 190.44 Tm +(standard) Tj +1 0 0 1 392.76 190.44 Tm +(NFS) Tj +1 0 0 1 413.16 190.44 Tm +(implementation) Tj +1 0 0 1 477.839 190.44 Tm +(in) Tj +1 0 0 1 487.799 190.44 Tm +(Digital) Tj +1 0 0 1 517.559 190.44 Tm +(Unix.) Tj +1 0 0 1 315 178.44 Tm +(This) Tj +1 0 0 1 335.64 178.44 Tm +(good) Tj +1 0 0 1 358.44 178.44 Tm +(performance) Tj +1 0 0 1 411.72 178.44 Tm +(is) Tj +1 0 0 1 421.32 178.44 Tm +(due) Tj +1 0 0 1 438.6 178.44 Tm +(to) Tj +1 0 0 1 449.28 178.44 Tm +(a) Tj +1 0 0 1 456.72 178.44 Tm +(number) Tj +1 0 0 1 490.079 178.44 Tm +(of) Tj +1 0 0 1 501.239 178.44 Tm +(important) Tj +1 0 0 1 315 166.56 Tm +(optimizations,) Tj +1 0 0 1 374.52 166.56 Tm +(including) Tj +1 0 0 1 414.84 166.56 Tm +(replacing) Tj +1 0 0 1 454.56 166.56 Tm +(public-k) Tj +1 0 0 1 487.919 166.56 Tm +(e) Tj +1 0 0 1 492.239 166.56 Tm +(y) Tj +1 0 0 1 499.559 166.56 Tm +(signatures) Tj +1 0 0 1 315 154.56 Tm +(by) Tj +1 0 0 1 329.52 154.56 Tm +(v) Tj +1 0 0 1 334.44 154.56 Tm +(ectors) Tj +1 0 0 1 362.76 154.56 Tm +(of) Tj +1 0 0 1 375.6 154.56 Tm +(message) Tj +1 0 0 1 413.88 154.56 Tm +(authentication) Tj +1 0 0 1 474.84 154.56 Tm +(codes,) Tj +1 0 0 1 505.199 154.56 Tm +(reducing) Tj +1 0 0 1 315 142.68 Tm +(the) Tj +1 0 0 1 330.96 142.68 Tm +(size) Tj +1 0 0 1 350.16 142.68 Tm +(and) Tj +1 0 0 1 368.4 142.68 Tm +(number) Tj +1 0 0 1 402.6 142.68 Tm +(of) Tj +1 0 0 1 414.6 142.68 Tm +(messages,) Tj +1 0 0 1 458.88 142.68 Tm +(and) Tj +1 0 0 1 477 142.68 Tm +(the) Tj +1 0 0 1 492.96 142.68 Tm +(incremental) Tj +1 0 0 1 315 130.68 Tm +(checkpoint-management) Tj +1 0 0 1 415.44 130.68 Tm +(techniques.) Tj +1 0 0 1 324.96 118.8 Tm +(One) Tj +1 0 0 1 345.84 118.8 Tm +(reason) Tj +1 0 0 1 376.08 118.8 Tm +(why) Tj +1 0 0 1 397.44 118.8 Tm +(Byzantine-f) Tj +1 0 0 1 444.6 118.8 Tm +(ault-tolerant) Tj +1 0 0 1 497.399 118.8 Tm +(algorithms) Tj +1 0 0 1 315 106.8 Tm +(will) Tj +1 0 0 1 334.8 106.8 Tm +(be) Tj +1 0 0 1 348.6 106.8 Tm +(important) Tj +1 0 0 1 391.68 106.8 Tm +(in) Tj +1 0 0 1 403.68 106.8 Tm +(the) Tj +1 0 0 1 420.24 106.8 Tm +(future) Tj +1 0 0 1 448.32 106.8 Tm +(is) Tj +1 0 0 1 459.36 106.8 Tm +(that) Tj +1 0 0 1 478.559 106.8 Tm +(the) Tj +1 0 0 1 490.679 106.8 Tm +(y) Tj +1 0 0 1 499.919 106.8 Tm +(can) Tj +1 0 0 1 518.159 106.8 Tm +(allo) Tj +1 0 0 1 532.919 106.8 Tm +(w) Tj +1 0 0 1 315 94.92 Tm +(systems) Tj +1 0 0 1 350.04 94.92 Tm +(to) Tj +1 0 0 1 361.32 94.92 Tm +(continue) Tj +1 0 0 1 399.12 94.92 Tm +(to) Tj +1 0 0 1 410.28 94.92 Tm +(w) Tj +1 0 0 1 417.36 94.92 Tm +(ork) Tj +1 0 0 1 434.16 94.92 Tm +(correctly) Tj +1 0 0 1 473.159 94.92 Tm +(e) Tj +1 0 0 1 477.359 94.92 Tm +(v) Tj +1 0 0 1 482.279 94.92 Tm +(en) Tj +1 0 0 1 494.999 94.92 Tm +(when) Tj +1 0 0 1 520.079 94.92 Tm +(there) Tj +1 0 0 1 315 82.92 Tm +(are) Tj +1 0 0 1 333 82.92 Tm +(softw) Tj +1 0 0 1 355.08 82.92 Tm +(are) Tj +1 0 0 1 373.08 82.92 Tm +(errors.) Tj +1 0 0 1 412.44 82.92 Tm +(Not) Tj +1 0 0 1 433.32 82.92 Tm +(all) Tj +1 0 0 1 449.16 82.92 Tm +(errors) Tj +1 0 0 1 478.199 82.92 Tm +(are) Tj +1 0 0 1 496.199 82.92 Tm +(survi) Tj +1 0 0 1 515.999 82.92 Tm +(v) Tj +1 0 0 1 520.799 82.92 Tm +(able;) Tj +1 0 0 1 315 71.04 Tm +(our) Tj +1 0 0 1 331.92 71.04 Tm +(approach) Tj +1 0 0 1 372 71.04 Tm +(cannot) Tj +1 0 0 1 402.12 71.04 Tm +(mask) Tj +1 0 0 1 426.72 71.04 Tm +(a) Tj +1 0 0 1 434.76 71.04 Tm +(softw) Tj +1 0 0 1 456.84 71.04 Tm +(are) Tj +1 0 0 1 472.559 71.04 Tm +(error) Tj +1 0 0 1 495.479 71.04 Tm +(that) Tj +1 0 0 1 513.959 71.04 Tm +(occurs) Tj +1 0 0 1 300.96 35.16 Tm +(13) Tj +ET +Q +endstream +endobj +778 0 obj +7318 +endobj +779 0 obj +<< +/Type /Page +/MediaBox [0 0 612 792] +/Parent 2 0 R +/Resources << /ProcSet [/PDF /ImageB /Text] +/Font << +/R744 744 0 R +/R741 741 0 R +/R738 738 0 R +>> +/XObject << +/R775 775 0 R +/R751 751 0 R +/R771 771 0 R +/R747 747 0 R +/R767 767 0 R +/R759 759 0 R +/R763 763 0 R +/R755 755 0 R +>> +>> +/Contents [ +739 0 R +742 0 R +745 0 R +749 0 R +753 0 R +757 0 R +761 0 R +765 0 R +769 0 R +773 0 R +777 0 R +] +>> +endobj +780 0 obj +<< /Type /Font /Name /R780 /Subtype /Type1 /BaseFont /Times-Roman >> +endobj +781 0 obj +<< /Length 782 0 R >> +stream +q +BT +/R780 10 Tf +1 0 0 1 72 709.08 Tm +(at) Tj +1 0 0 1 85.32 709.08 Tm +(all) Tj +1 0 0 1 101.28 709.08 Tm +(replicas.) Tj +1 0 0 1 149.16 709.08 Tm +(Ho) Tj +1 0 0 1 161.16 709.08 Tm +(we) Tj +1 0 0 1 172.56 709.08 Tm +(v) Tj +1 0 0 1 177.48 709.08 Tm +(er) Tj +1 0 0 1 184.92 709.08 Tm +(,) Tj +1 0 0 1 194.04 709.08 Tm +(it) Tj +1 0 0 1 205.68 709.08 Tm +(can) Tj +1 0 0 1 225.6 709.08 Tm +(mask) Tj +1 0 0 1 252.719 709.08 Tm +(errors) Tj +1 0 0 1 281.999 709.08 Tm +(that) Tj +1 0 0 1 72 697.08 Tm +(occur) Tj +1 0 0 1 100.68 697.08 Tm +(independently) Tj +1 0 0 1 163.8 697.08 Tm +(at) Tj +1 0 0 1 177.6 697.08 Tm +(dif) Tj +1 0 0 1 188.52 697.08 Tm +(ferent) Tj +1 0 0 1 218.28 697.08 Tm +(replicas,) Tj +1 0 0 1 259.319 697.08 Tm +(including) Tj +1 0 0 1 72 685.2 Tm +(nondeterministic) Tj +1 0 0 1 143.88 685.2 Tm +(softw) Tj +1 0 0 1 165.96 685.2 Tm +(are) Tj +1 0 0 1 183 685.2 Tm +(errors,) Tj +1 0 0 1 214.2 685.2 Tm +(which) Tj +1 0 0 1 243.479 685.2 Tm +(are) Tj +1 0 0 1 260.519 685.2 Tm +(the) Tj +1 0 0 1 277.679 685.2 Tm +(most) Tj +1 0 0 1 72 673.2 Tm +(problematic) Tj +1 0 0 1 125.28 673.2 Tm +(and) Tj +1 0 0 1 145.32 673.2 Tm +(persistent) Tj +1 0 0 1 189.24 673.2 Tm +(errors) Tj +1 0 0 1 218.16 673.2 Tm +(since) Tj +1 0 0 1 244.32 673.2 Tm +(the) Tj +1 0 0 1 256.439 673.2 Tm +(y) Tj +1 0 0 1 266.999 673.2 Tm +(are) Tj +1 0 0 1 284.879 673.2 Tm +(the) Tj +1 0 0 1 72 661.32 Tm +(hardest) Tj +1 0 0 1 102.72 661.32 Tm +(to) Tj +1 0 0 1 112.44 661.32 Tm +(detect.) Tj +1 0 0 1 142.2 661.32 Tm +(In) Tj +1 0 0 1 152.52 661.32 Tm +(f) Tj +1 0 0 1 155.76 661.32 Tm +(act,) Tj +1 0 0 1 171.84 661.32 Tm +(we) Tj +1 0 0 1 185.52 661.32 Tm +(encountered) Tj +1 0 0 1 236.159 661.32 Tm +(such) Tj +1 0 0 1 256.439 661.32 Tm +(a) Tj +1 0 0 1 262.799 661.32 Tm +(softw) Tj +1 0 0 1 284.879 661.32 Tm +(are) Tj +1 0 0 1 72 649.32 Tm +(b) Tj +1 0 0 1 76.8 649.32 Tm +(ug) Tj +1 0 0 1 88.4399 649.32 Tm +(while) Tj +1 0 0 1 112.2 649.32 Tm +(running) Tj +1 0 0 1 144.96 649.32 Tm +(our) Tj +1 0 0 1 159.84 649.32 Tm +(system,) Tj +1 0 0 1 191.88 649.32 Tm +(and) Tj +1 0 0 1 207.96 649.32 Tm +(our) Tj +1 0 0 1 222.959 649.32 Tm +(algorithm) Tj +1 0 0 1 263.279 649.32 Tm +(w) Tj +1 0 0 1 270.359 649.32 Tm +(as) Tj +1 0 0 1 280.439 649.32 Tm +(able) Tj +1 0 0 1 72 637.44 Tm +(to) Tj +1 0 0 1 82.2 637.44 Tm +(continue) Tj +1 0 0 1 119.04 637.44 Tm +(running) Tj +1 0 0 1 152.52 637.44 Tm +(correctly) Tj +1 0 0 1 190.44 637.44 Tm +(in) Tj +1 0 0 1 200.64 637.44 Tm +(spite) Tj +1 0 0 1 221.999 637.44 Tm +(of) Tj +1 0 0 1 232.799 637.44 Tm +(it.) Tj +1 0 0 1 81.96 625.44 Tm +(There) Tj +1 0 0 1 106.92 625.44 Tm +(is) Tj +1 0 0 1 115.08 625.44 Tm +(still) Tj +1 0 0 1 131.52 625.44 Tm +(much) Tj +1 0 0 1 155.4 625.44 Tm +(w) Tj +1 0 0 1 162.48 625.44 Tm +(ork) Tj +1 0 0 1 177.48 625.44 Tm +(to) Tj +1 0 0 1 186.84 625.44 Tm +(do) Tj +1 0 0 1 198.36 625.44 Tm +(on) Tj +1 0 0 1 209.88 625.44 Tm +(imp) Tj +1 0 0 1 225.36 625.44 Tm +(ro) Tj +1 0 0 1 233.519 625.44 Tm +(v) Tj +1 0 0 1 238.439 625.44 Tm +(ing) Tj +1 0 0 1 252.719 625.44 Tm +(ou) Tj +1 0 0 1 262.679 625.44 Tm +(r) Tj +1 0 0 1 267.599 625.44 Tm +(sy) Tj +1 0 0 1 276.359 625.44 Tm +(stem.) Tj +1 0 0 1 72 613.44 Tm +(One) Tj +1 0 0 1 91.92 613.44 Tm +(problem) Tj +1 0 0 1 128.4 613.44 Tm +(of) Tj +1 0 0 1 139.92 613.44 Tm +(special) Tj +1 0 0 1 170.88 613.44 Tm +(interest) Tj +1 0 0 1 203.52 613.44 Tm +(is) Tj +1 0 0 1 213.48 613.44 Tm +(reducing) Tj +1 0 0 1 251.639 613.44 Tm +(the) Tj +1 0 0 1 267.119 613.44 Tm +(amount) Tj +1 0 0 1 72 601.56 Tm +(of) Tj +1 0 0 1 83.52 601.56 Tm +(resources) Tj +1 0 0 1 124.44 601.56 Tm +(required) Tj +1 0 0 1 160.92 601.56 Tm +(to) Tj +1 0 0 1 171.96 601.56 Tm +(implement) Tj +1 0 0 1 217.8 601.56 Tm +(our) Tj +1 0 0 1 234.359 601.56 Tm +(algorithm.) Tj +1 0 0 1 281.519 601.56 Tm +(The) Tj +1 0 0 1 72 589.56 Tm +(number) Tj +1 0 0 1 105.96 589.56 Tm +(of) Tj +1 0 0 1 117.84 589.56 Tm +(replicas) Tj +1 0 0 1 152.4 589.56 Tm +(can) Tj +1 0 0 1 169.8 589.56 Tm +(be) Tj +1 0 0 1 182.76 589.56 Tm +(reduced) Tj +1 0 0 1 217.8 589.56 Tm +(by) Tj +1 0 0 1 231.359 589.56 Tm +(using) Tj +ET +Q +q +W +0 0 612 792 re +n +endstream +endobj +782 0 obj +3920 +endobj +783 0 obj +<< /Type /XObject /Name /R783 /Subtype /Image /Length 784 0 R +/ImageMask true /Width 50 /Height 92 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 50 /BlackIs1 true >>] +>> +stream +5$eL7-b:K3p0<%? +h1u#R]2'Mug].5is7$!eqtL-hqu?EZs5/V3iQXZVs7/ihqu?EZs4QRbJe-ans*o2/4DdB_@6aqq5Q~> +endstream +endobj +784 0 obj +96 +endobj +785 0 obj +<< /Length 786 0 R >> +stream +q +5 0 0 -9.2 257 596.7 cm +/R783 Do +Q +BT +/R780 10 Tf +1 0 0 1 266.039 589.56 Tm +(replicas) Tj +1 0 0 1 72 577.68 Tm +(as) Tj +1 0 0 1 83.0399 577.68 Tm +(witnesses) Tj +1 0 0 1 123.96 577.68 Tm +(that) Tj +1 0 0 1 141.72 577.68 Tm +(are) Tj +1 0 0 1 156.6 577.68 Tm +(in) Tj +1 0 0 1 164.04 577.68 Tm +(v) Tj +1 0 0 1 168.84 577.68 Tm +(olv) Tj +1 0 0 1 181.56 577.68 Tm +(ed) Tj +1 0 0 1 193.56 577.68 Tm +(in) Tj +1 0 0 1 204 577.68 Tm +(the) Tj +1 0 0 1 219 577.68 Tm +(protocol) Tj +1 0 0 1 254.999 577.68 Tm +(only) Tj +1 0 0 1 275.399 577.68 Tm +(when) Tj +1 0 0 1 72 565.68 Tm +(some) Tj +1 0 0 1 95.5199 565.68 Tm +(full) Tj +1 0 0 1 111.84 565.68 Tm +(replica) Tj +1 0 0 1 141.48 565.68 Tm +(f) Tj +1 0 0 1 144.72 565.68 Tm +(ails.) Tj +1 0 0 1 164.76 565.68 Tm +(W) Tj +1 0 0 1 173.28 565.68 Tm +(e) Tj +1 0 0 1 180.24 565.68 Tm +(also) Tj +1 0 0 1 198.84 565.68 Tm +(belie) Tj +1 0 0 1 218.039 565.68 Tm +(v) Tj +1 0 0 1 222.959 565.68 Tm +(e) Tj +1 0 0 1 229.679 565.68 Tm +(that) Tj +1 0 0 1 247.079 565.68 Tm +(it) Tj +1 0 0 1 255.239 565.68 Tm +(is) Tj +1 0 0 1 264.359 565.68 Tm +(possible) Tj +1 0 0 1 72 553.8 Tm +(to) Tj +1 0 0 1 82.68 553.8 Tm +(reduce) Tj +1 0 0 1 112.08 553.8 Tm +(the) Tj +1 0 0 1 127.08 553.8 Tm +(number) Tj +1 0 0 1 160.44 553.8 Tm +(of) Tj +1 0 0 1 171.6 553.8 Tm +(copies) Tj +1 0 0 1 200.04 553.8 Tm +(of) Tj +1 0 0 1 211.2 553.8 Tm +(the) Tj +1 0 0 1 226.199 553.8 Tm +(state) Tj +1 0 0 1 247.439 553.8 Tm +(to) Tj +ET +q +5 0 0 -9.2 258.4 560.9 cm +/R783 Do +Q +endstream +endobj +786 0 obj +1426 +endobj +787 0 obj +<< /Type /XObject /Name /R787 /Subtype /Image /Length 788 0 R +/ImageMask true /Width 66 /Height 65 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 66 /BlackIs1 true >>] +>> +stream +,BZ6Ns8W-!s8W-!s-Eq9Iaj?!s8W-!s8W-!s8Uf'#QTA~> +endstream +endobj +788 0 obj +47 +endobj +789 0 obj +<< /Length 790 0 R >> +stream +q +6.6 0 0 -6.5 266.9 559.6 cm +/R787 Do +Q +BT +1 0 0 1 276.599 553.8 Tm +(1) Tj +1 0 0 1 284.519 553.8 Tm +(b) Tj +1 0 0 1 289.319 553.8 Tm +(ut) Tj +1 0 0 1 72 541.8 Tm +(the) Tj +1 0 0 1 86.6399 541.8 Tm +(details) Tj +1 0 0 1 115.2 541.8 Tm +(remain) Tj +1 0 0 1 145.32 541.8 Tm +(to) Tj +1 0 0 1 155.52 541.8 Tm +(be) Tj +1 0 0 1 167.52 541.8 Tm +(w) Tj +1 0 0 1 174.6 541.8 Tm +(ork) Tj +1 0 0 1 187.92 541.8 Tm +(ed) Tj +1 0 0 1 199.68 541.8 Tm +(out.) Tj +ET +endstream +endobj +790 0 obj +439 +endobj +791 0 obj +<< /Type /Font /Name /R791 /Subtype /Type1 /BaseFont /Times-Bold >> +endobj +792 0 obj +<< /Length 793 0 R >> +stream +BT +/R791 10 Tf +1 0 0 1 72 521.16 Tm +(Ackno) Tj +1 0 0 1 99.6 521.16 Tm +(wledgments) Tj +/R780 10 Tf +1 0 0 1 72 507 Tm +(W) Tj +1 0 0 1 80.52 507 Tm +(e) Tj +1 0 0 1 90.72 507 Tm +(w) Tj +1 0 0 1 97.8 507 Tm +(ould) Tj +1 0 0 1 121.32 507 Tm +(lik) Tj +1 0 0 1 131.76 507 Tm +(e) Tj +1 0 0 1 141.84 507 Tm +(to) Tj +1 0 0 1 155.28 507 Tm +(thank) Tj +1 0 0 1 183.24 507 Tm +(Atul) Tj +1 0 0 1 206.64 507 Tm +(Adya,) Tj +1 0 0 1 237.24 507 Tm +(Chandrasekhar) Tj +1 0 0 1 72 495 Tm +(Bo) Tj +1 0 0 1 83.5199 495 Tm +(yapati,) Tj +1 0 0 1 112.2 495 Tm +(Nanc) Tj +1 0 0 1 133.2 495 Tm +(y) Tj +1 0 0 1 139.8 495 Tm +(L) Tj +1 0 0 1 145.32 495 Tm +(ynch,) Tj +1 0 0 1 168.96 495 Tm +(Sape) Tj +1 0 0 1 190.08 495 Tm +(Mullender) Tj +1 0 0 1 231.48 495 Tm +(,) Tj +1 0 0 1 235.56 495 Tm +(Andre) Tj +1 0 0 1 260.4 495 Tm +(w) Tj +1 0 0 1 269.039 495 Tm +(Myers,) Tj +1 0 0 1 72 483 Tm +(Liuba) Tj +1 0 0 1 96.9599 483 Tm +(Shrira,) Tj +1 0 0 1 125.4 483 Tm +(and) Tj +1 0 0 1 141.48 483 Tm +(the) Tj +1 0 0 1 155.28 483 Tm +(anon) Tj +1 0 0 1 174.72 483 Tm +(ymous) Tj +1 0 0 1 203.04 483 Tm +(referees) Tj +1 0 0 1 236.279 483 Tm +(for) Tj +1 0 0 1 249.599 483 Tm +(their) Tj +1 0 0 1 269.519 483 Tm +(h) Tj +1 0 0 1 274.439 483 Tm +(elpf) Tj +1 0 0 1 289.919 483 Tm +(ul) Tj +1 0 0 1 72 471.12 Tm +(comments) Tj +1 0 0 1 115.44 471.12 Tm +(on) Tj +1 0 0 1 127.92 471.12 Tm +(drafts) Tj +1 0 0 1 153.12 471.12 Tm +(of) Tj +1 0 0 1 163.92 471.12 Tm +(this) Tj +1 0 0 1 180.72 471.12 Tm +(paper) Tj +1 0 0 1 202.44 471.12 Tm +(.) Tj +/R791 12 Tf +1 0 0 1 72 448.32 Tm +(Refer) Tj +1 0 0 1 100.2 448.32 Tm +(ences) Tj +/R780 8 Tf +1 0 0 1 75.96 435.36 Tm +([1]) Tj +1 0 0 1 90.2399 435.36 Tm +(M.) Tj +1 0 0 1 102 435.36 Tm +(Bellare) Tj +1 0 0 1 127.56 435.36 Tm +(and) Tj +1 0 0 1 141.72 435.36 Tm +(D.) Tj +1 0 0 1 152.04 435.36 Tm +(Micciancio.) Tj +1 0 0 1 194.399 435.36 Tm +(A) Tj +1 0 0 1 202.679 435.36 Tm +(Ne) Tj +1 0 0 1 211.679 435.36 Tm +(w) Tj +1 0 0 1 220.199 435.36 Tm +(P) Tj +1 0 0 1 224.519 435.36 Tm +(aradigm) Tj +1 0 0 1 253.199 435.36 Tm +(for) Tj +1 0 0 1 265.079 435.36 Tm +(Collision-) Tj +1 0 0 1 90.24 425.88 Tm +(free) Tj +1 0 0 1 105.36 425.88 Tm +(Hashing:) Tj +1 0 0 1 138.48 425.88 Tm +(Incrementality) Tj +1 0 0 1 187.68 425.88 Tm +(at) Tj +1 0 0 1 196.199 425.88 Tm +(Reduced) Tj +1 0 0 1 226.799 425.88 Tm +(Cost.) Tj +1 0 0 1 248.159 425.88 Tm +(In) Tj +ET +endstream +endobj +793 0 obj +2256 +endobj +794 0 obj +<< /Type /Font /Name /R794 /Subtype /Type1 /BaseFont /Times-Italic >> +endobj +795 0 obj +<< /Length 796 0 R >> +stream +BT +/R794 8 Tf +1 0 0 1 257.519 425.88 Tm +(Advances) Tj +1 0 0 1 290.759 425.88 Tm +(in) Tj +1 0 0 1 90.24 416.4 Tm +(Cryptolo) Tj +1 0 0 1 118.2 416.4 Tm +(gy) Tj +1 0 0 1 128.04 416.4 Tm +(\261) Tj +1 0 0 1 134.04 416.4 Tm +(Eur) Tj +1 0 0 1 145.68 416.4 Tm +(ocrypt) Tj +1 0 0 1 168 416.4 Tm +(97) Tj +/R780 8 Tf +1 0 0 1 175.92 416.4 Tm +(,) Tj +1 0 0 1 179.88 416.4 Tm +(1997.) Tj +1 0 0 1 75.96 403.32 Tm +([2]) Tj +1 0 0 1 90.2399 403.32 Tm +(G.) Tj +1 0 0 1 99.5999 403.32 Tm +(Bracha) Tj +1 0 0 1 123.72 403.32 Tm +(and) Tj +1 0 0 1 136.92 403.32 Tm +(S.) Tj +1 0 0 1 144.84 403.32 Tm +(T) Tj +1 0 0 1 149.16 403.32 Tm +(oue) Tj +1 0 0 1 160.44 403.32 Tm +(g.) Tj +1 0 0 1 168.72 403.32 Tm +(Asynchronous) Tj +1 0 0 1 216.359 403.32 Tm +(Consensus) Tj +1 0 0 1 251.999 403.32 Tm +(and) Tj +1 0 0 1 265.079 403.32 Tm +(Broadcast) Tj +1 0 0 1 90.24 393.96 Tm +(Protocols.) Tj +/R794 8 Tf +1 0 0 1 125.28 393.96 Tm +(J) Tj +1 0 0 1 128.52 393.96 Tm +(ournal) Tj +1 0 0 1 151.8 393.96 Tm +(of) Tj +1 0 0 1 159.96 393.96 Tm +(the) Tj +1 0 0 1 171.72 393.96 Tm +(A) Tj +1 0 0 1 176.4 393.96 Tm +(CM) Tj +/R780 8 Tf +1 0 0 1 188.28 393.96 Tm +(,) Tj +1 0 0 1 192.359 393.96 Tm +(32\(4\),) Tj +1 0 0 1 213.599 393.96 Tm +(1995.) Tj +1 0 0 1 75.96 380.88 Tm +([3]) Tj +1 0 0 1 90.2399 380.88 Tm +(R.) Tj +1 0 0 1 101.52 380.88 Tm +(Canneti) Tj +1 0 0 1 130.2 380.88 Tm +(and) Tj +1 0 0 1 145.56 380.88 Tm +(T) Tj +1 0 0 1 149.88 380.88 Tm +(.) Tj +1 0 0 1 155.76 380.88 Tm +(Rabin.) Tj +1 0 0 1 184.68 380.88 Tm +(Optimal) Tj +1 0 0 1 214.679 380.88 Tm +(Asynchronous) Tj +1 0 0 1 264.719 380.88 Tm +(Byzantine) Tj +1 0 0 1 90.24 371.4 Tm +(Agreement.) Tj +1 0 0 1 140.04 371.4 Tm +(T) Tj +1 0 0 1 144.36 371.4 Tm +(echnical) Tj +1 0 0 1 176.4 371.4 Tm +(Report) Tj +1 0 0 1 203.76 371.4 Tm +(#92-15,) Tj +1 0 0 1 234.719 371.4 Tm +(Computer) Tj +1 0 0 1 272.159 371.4 Tm +(Science) Tj +1 0 0 1 90.24 361.92 Tm +(Department,) Tj +1 0 0 1 131.88 361.92 Tm +(Hebre) Tj +1 0 0 1 150.96 361.92 Tm +(w) Tj +1 0 0 1 158.88 361.92 Tm +(Uni) Tj +1 0 0 1 170.52 361.92 Tm +(v) Tj +1 0 0 1 174.36 361.92 Tm +(ersity) Tj +1 0 0 1 191.4 361.92 Tm +(,) Tj +1 0 0 1 195.719 361.92 Tm +(1992.) Tj +1 0 0 1 75.96 348.96 Tm +([4]) Tj +1 0 0 1 90.2399 348.96 Tm +(M.) Tj +1 0 0 1 102.96 348.96 Tm +(Castro) Tj +1 0 0 1 127.32 348.96 Tm +(and) Tj +1 0 0 1 142.44 348.96 Tm +(B.) Tj +1 0 0 1 153.36 348.96 Tm +(Lisk) Tj +1 0 0 1 167.4 348.96 Tm +(o) Tj +1 0 0 1 171.24 348.96 Tm +(v) Tj +1 0 0 1 174.72 348.96 Tm +(.) Tj +1 0 0 1 183.84 348.96 Tm +(A) Tj +1 0 0 1 193.2 348.96 Tm +(Correctness) Tj +1 0 0 1 234.479 348.96 Tm +(Proof) Tj +1 0 0 1 255.719 348.96 Tm +(for) Tj +1 0 0 1 268.559 348.96 Tm +(a) Tj +1 0 0 1 275.759 348.96 Tm +(Practi-) Tj +1 0 0 1 90.24 339.48 Tm +(cal) Tj +1 0 0 1 102.48 339.48 Tm +(Byzantine-F) Tj +1 0 0 1 141.36 339.48 Tm +(ault-T) Tj +1 0 0 1 160.08 339.48 Tm +(olera) Tj +1 0 0 1 175.92 339.48 Tm +(nt) Tj +1 0 0 1 185.76 339.48 Tm +(Replication) Tj +1 0 0 1 225.479 339.48 Tm +(Algorithm.) Tj +1 0 0 1 266.159 339.48 Tm +(T) Tj +1 0 0 1 270.479 339.48 Tm +(echnical) Tj +1 0 0 1 90.24 330 Tm +(Memo) Tj +1 0 0 1 114.24 330 Tm +(MIT/LCS/TM-590,) Tj +1 0 0 1 180 330 Tm +(MIT) Tj +1 0 0 1 197.88 330 Tm +(Laboratory) Tj +1 0 0 1 236.519 330 Tm +(for) Tj +1 0 0 1 248.999 330 Tm +(Computer) Tj +1 0 0 1 284.159 330 Tm +(Sci-) Tj +1 0 0 1 90.24 320.52 Tm +(ence,) Tj +1 0 0 1 108.84 320.52 Tm +(1999.) Tj +1 0 0 1 75.96 307.56 Tm +([5]) Tj +1 0 0 1 90.2399 307.56 Tm +(M.) Tj +1 0 0 1 104.4 307.56 Tm +(Castro) Tj +1 0 0 1 130.2 307.56 Tm +(and) Tj +1 0 0 1 146.88 307.56 Tm +(B.) Tj +1 0 0 1 159.24 307.56 Tm +(Lisk) Tj +1 0 0 1 173.28 307.56 Tm +(o) Tj +1 0 0 1 177.12 307.56 Tm +(v) Tj +1 0 0 1 180.6 307.56 Tm +(.) Tj +1 0 0 1 193.56 307.56 Tm +(Authenticated) Tj +1 0 0 1 243.359 307.56 Tm +(Byzantine) Tj +1 0 0 1 280.679 307.56 Tm +(F) Tj +1 0 0 1 284.999 307.56 Tm +(ault) Tj +1 0 0 1 90.24 298.08 Tm +(T) Tj +1 0 0 1 94.56 298.08 Tm +(olerance) Tj +1 0 0 1 124.2 298.08 Tm +(W) Tj +1 0 0 1 131.4 298.08 Tm +(ithout) Tj +1 0 0 1 152.76 298.08 Tm +(Public-K) Tj +1 0 0 1 181.08 298.08 Tm +(e) Tj +1 0 0 1 184.44 298.08 Tm +(y) Tj +1 0 0 1 191.52 298.08 Tm +(Cryptography.) Tj +1 0 0 1 242.639 298.08 Tm +(T) Tj +1 0 0 1 246.959 298.08 Tm +(echnical) Tj +1 0 0 1 276.239 298.08 Tm +(Memo) Tj +1 0 0 1 90.24 288.6 Tm +(MIT/LCS/TM-589,) Tj +1 0 0 1 153.96 288.6 Tm +(MIT) Tj +1 0 0 1 170.04 288.6 Tm +(Laboratory) Tj +1 0 0 1 206.76 288.6 Tm +(for) Tj +1 0 0 1 217.439 288.6 Tm +(Computer) Tj +1 0 0 1 250.799 288.6 Tm +(Science,) Tj +1 0 0 1 279.119 288.6 Tm +(1999.) Tj +1 0 0 1 75.96 275.64 Tm +([6]) Tj +1 0 0 1 90.2399 275.64 Tm +(F) Tj +1 0 0 1 94.0799 275.64 Tm +(.) Tj +1 0 0 1 97.4399 275.64 Tm +(Cristian,) Tj +1 0 0 1 126.12 275.64 Tm +(H.) Tj +1 0 0 1 135.24 275.64 Tm +(Aghili,) Tj +1 0 0 1 159.12 275.64 Tm +(H.) Tj +1 0 0 1 168.24 275.64 Tm +(Strong,) Tj +1 0 0 1 193.079 275.64 Tm +(and) Tj +1 0 0 1 205.919 275.64 Tm +(D.) Tj +1 0 0 1 215.039 275.64 Tm +(Dole) Tj +1 0 0 1 230.159 275.64 Tm +(v) Tj +1 0 0 1 233.639 275.64 Tm +(.) Tj +1 0 0 1 237.599 275.64 Tm +(Atomic) Tj +1 0 0 1 262.919 275.64 Tm +(Broadcast:) Tj +1 0 0 1 90.24 266.16 Tm +(From) Tj +1 0 0 1 110.88 266.16 Tm +(Simple) Tj +1 0 0 1 136.68 266.16 Tm +(Message) Tj +1 0 0 1 167.88 266.16 Tm +(Dif) Tj +1 0 0 1 178.2 266.16 Tm +(fusion) Tj +1 0 0 1 201.48 266.16 Tm +(to) Tj +1 0 0 1 211.079 266.16 Tm +(Byzantine) Tj +1 0 0 1 246.599 266.16 Tm +(Agreement.) Tj +1 0 0 1 290.399 266.16 Tm +(In) Tj +/R794 8 Tf +1 0 0 1 90.24 256.68 Tm +(International) Tj +1 0 0 1 134.28 256.68 Tm +(Confer) Tj +1 0 0 1 156 256.68 Tm +(ence) Tj +1 0 0 1 172.8 256.68 Tm +(on) Tj +1 0 0 1 182.76 256.68 Tm +(F) Tj +1 0 0 1 187.08 256.68 Tm +(ault) Tj +1 0 0 1 201.359 256.68 Tm +(T) Tj +1 0 0 1 205.079 256.68 Tm +(oler) Tj +1 0 0 1 217.679 256.68 Tm +(ant) Tj +1 0 0 1 230.039 256.68 Tm +(Computing) Tj +/R780 8 Tf +1 0 0 1 265.199 256.68 Tm +(,) Tj +1 0 0 1 269.399 256.68 Tm +(1985.) Tj +1 0 0 1 75.96 243.72 Tm +([7]) Tj +1 0 0 1 90.2399 243.72 Tm +(S.) Tj +1 0 0 1 100.2 243.72 Tm +(Deering) Tj +1 0 0 1 129.48 243.72 Tm +(and) Tj +1 0 0 1 144.6 243.72 Tm +(D.) Tj +1 0 0 1 155.88 243.72 Tm +(Cheriton.) Tj +1 0 0 1 192.84 243.72 Tm +(Multicast) Tj +1 0 0 1 226.559 243.72 Tm +(Routing) Tj +1 0 0 1 255.839 243.72 Tm +(in) Tj +1 0 0 1 265.559 243.72 Tm +(Datagram) Tj +1 0 0 1 90.24 234.24 Tm +(Internetw) Tj +1 0 0 1 120.36 234.24 Tm +(orks) Tj +1 0 0 1 139.2 234.24 Tm +(and) Tj +1 0 0 1 155.4 234.24 Tm +(Extended) Tj +1 0 0 1 190.32 234.24 Tm +(LANs.) Tj +/R794 8 Tf +1 0 0 1 221.88 234.24 Tm +(A) Tj +1 0 0 1 226.559 234.24 Tm +(CM) Tj +1 0 0 1 243.239 234.24 Tm +(T) Tj +1 0 0 1 247.199 234.24 Tm +(r) Tj +1 0 0 1 250.199 234.24 Tm +(ansactions) Tj +1 0 0 1 289.079 234.24 Tm +(on) Tj +1 0 0 1 90.24 224.76 Tm +(Computer) Tj +1 0 0 1 124.08 224.76 Tm +(Systems) Tj +/R780 8 Tf +1 0 0 1 149.16 224.76 Tm +(,) Tj +1 0 0 1 153.36 224.76 Tm +(8\(2\),) Tj +1 0 0 1 170.64 224.76 Tm +(1990.) Tj +1 0 0 1 75.96 211.8 Tm +([8]) Tj +1 0 0 1 90.2399 211.8 Tm +(H.) Tj +1 0 0 1 100.56 211.8 Tm +(Dobbertin.) Tj +1 0 0 1 139.44 211.8 Tm +(The) Tj +1 0 0 1 154.44 211.8 Tm +(Status) Tj +1 0 0 1 176.52 211.8 Tm +(of) Tj +1 0 0 1 185.759 211.8 Tm +(MD5) Tj +1 0 0 1 205.199 211.8 Tm +(After) Tj +1 0 0 1 224.639 211.8 Tm +(a) Tj +1 0 0 1 230.759 211.8 Tm +(Recent) Tj +1 0 0 1 255.479 211.8 Tm +(Attack.) Tj +/R794 8 Tf +1 0 0 1 283.319 211.8 Tm +(RSA) Tj +1 0 0 1 90.24 202.32 Tm +(Labor) Tj +1 0 0 1 109.56 202.32 Tm +(atories') Tj +1 0 0 1 136.44 202.32 Tm +(CryptoBytes) Tj +/R780 8 Tf +1 0 0 1 175.56 202.32 Tm +(,) Tj +1 0 0 1 179.76 202.32 Tm +(2\(2\),) Tj +1 0 0 1 197.04 202.32 Tm +(1996.) Tj +1 0 0 1 75.96 189.24 Tm +([9]) Tj +1 0 0 1 90.2399 189.24 Tm +(M.) Tj +1 0 0 1 104.04 189.24 Tm +(Fischer) Tj +1 0 0 1 126.96 189.24 Tm +(,) Tj +1 0 0 1 134.52 189.24 Tm +(N.) Tj +1 0 0 1 147 189.24 Tm +(L) Tj +1 0 0 1 151.44 189.24 Tm +(ynch,) Tj +1 0 0 1 174.24 189.24 Tm +(and) Tj +1 0 0 1 190.44 189.24 Tm +(M.) Tj +1 0 0 1 204.24 189.24 Tm +(P) Tj +1 0 0 1 208.559 189.24 Tm +(aterson.) Tj +1 0 0 1 243.599 189.24 Tm +(Impossibility) Tj +1 0 0 1 290.399 189.24 Tm +(of) Tj +1 0 0 1 90.24 179.76 Tm +(Distrib) Tj +1 0 0 1 112.08 179.76 Tm +(uted) Tj +1 0 0 1 128.4 179.76 Tm +(Consensus) Tj +1 0 0 1 164.88 179.76 Tm +(W) Tj +1 0 0 1 172.08 179.76 Tm +(ith) Tj +1 0 0 1 182.88 179.76 Tm +(One) Tj +1 0 0 1 198.6 179.76 Tm +(F) Tj +1 0 0 1 202.92 179.76 Tm +(aulty) Tj +1 0 0 1 221.279 179.76 Tm +(Process.) Tj +/R794 8 Tf +1 0 0 1 251.639 179.76 Tm +(J) Tj +1 0 0 1 254.879 179.76 Tm +(ournal) Tj +1 0 0 1 278.639 179.76 Tm +(of) Tj +1 0 0 1 287.279 179.76 Tm +(the) Tj +1 0 0 1 90.24 170.4 Tm +(A) Tj +1 0 0 1 94.92 170.4 Tm +(CM) Tj +/R780 8 Tf +1 0 0 1 106.8 170.4 Tm +(,) Tj +1 0 0 1 110.88 170.4 Tm +(32\(2\),) Tj +1 0 0 1 132.12 170.4 Tm +(1985.) Tj +1 0 0 1 72 157.32 Tm +([10]) Tj +1 0 0 1 90.2399 157.32 Tm +(J.) Tj +1 0 0 1 97.6799 157.32 Tm +(Garay) Tj +1 0 0 1 119.4 157.32 Tm +(and) Tj +1 0 0 1 133.2 157.32 Tm +(Y) Tj +1 0 0 1 137.88 157.32 Tm +(.) Tj +1 0 0 1 142.32 157.32 Tm +(Moses.) Tj +1 0 0 1 168.72 157.32 Tm +(Fully) Tj +1 0 0 1 187.919 157.32 Tm +(Polynomial) Tj +1 0 0 1 226.919 157.32 Tm +(Byzantine) Tj +1 0 0 1 261.599 157.32 Tm +(Agreement) Tj +1 0 0 1 90.24 147.84 Tm +(for) Tj +1 0 0 1 101.16 147.84 Tm +(n) Tj +ET +endstream +endobj +796 0 obj +8830 +endobj +797 0 obj +<< /Type /XObject /Name /R797 /Subtype /Image /Length 798 0 R +/ImageMask true /Width 51 /Height 52 /BitsPerComponent 1 /Decode [1 0] /Filter [/ASCII85Decode /CCITTFaxDecode] /DecodeParms [null << /K -1 /Columns 51 /BlackIs1 true >>] +>> +stream +,P3#f%e*lbLPH9X +(i\tPiJ30s*#NYa@,Zlr$lg+'n6k41iF,"RKG515-p_9-!e_7'&1BhQ5lu]R((_l[@6jsf^'_`-#QTA~> +endstream +endobj +798 0 obj +98 +endobj +799 0 obj +<< /Length 800 0 R >> +stream +q +5.1 0 0 -5.2 107.5 152.4 cm +/R797 Do +Q +BT +1 0 0 1 113.4 147.84 Tm +(3t) Tj +1 0 0 1 121.2 147.84 Tm +(Processors) Tj +1 0 0 1 156.96 147.84 Tm +(in) Tj +1 0 0 1 164.76 147.84 Tm +(t+1) Tj +1 0 0 1 177.12 147.84 Tm +(Rounds.) Tj +/R794 8 Tf +1 0 0 1 205.679 147.84 Tm +(SIAM) Tj +1 0 0 1 225.479 147.84 Tm +(J) Tj +1 0 0 1 228.719 147.84 Tm +(ournal) Tj +1 0 0 1 251.759 147.84 Tm +(of) Tj +1 0 0 1 259.559 147.84 Tm +(Computing) Tj +/R780 8 Tf +1 0 0 1 294.719 147.84 Tm +(,) Tj +1 0 0 1 90.24 138.36 Tm +(27\(1\),) Tj +1 0 0 1 111.48 138.36 Tm +(1998.) Tj +1 0 0 1 72 125.4 Tm +([11]) Tj +1 0 0 1 90.2399 125.4 Tm +(D.) Tj +1 0 0 1 99.8399 125.4 Tm +(Ga) Tj +1 0 0 1 108.96 125.4 Tm +(wlick) Tj +1 0 0 1 128.64 125.4 Tm +(and) Tj +1 0 0 1 142.08 125.4 Tm +(D.) Tj +1 0 0 1 151.68 125.4 Tm +(Kinkade.) Tj +1 0 0 1 183.359 125.4 Tm +(V) Tj +1 0 0 1 188.279 125.4 Tm +(arieties) Tj +1 0 0 1 213.119 125.4 Tm +(of) Tj +1 0 0 1 221.639 125.4 Tm +(Concurrenc) Tj +1 0 0 1 258.359 125.4 Tm +(y) Tj +1 0 0 1 264.599 125.4 Tm +(Control) Tj +1 0 0 1 290.759 125.4 Tm +(in) Tj +1 0 0 1 90.24 115.92 Tm +(IMS/VS) Tj +1 0 0 1 118.8 115.92 Tm +(F) Tj +1 0 0 1 123.12 115.92 Tm +(ast) Tj +1 0 0 1 134.04 115.92 Tm +(P) Tj +1 0 0 1 138.36 115.92 Tm +(ath.) Tj +/R794 8 Tf +1 0 0 1 152.88 115.92 Tm +(Database) Tj +1 0 0 1 185.4 115.92 Tm +(Engineering) Tj +/R780 8 Tf +1 0 0 1 224.519 115.92 Tm +(,) Tj +1 0 0 1 228.839 115.92 Tm +(8\(2\),) Tj +1 0 0 1 246.119 115.92 Tm +(1985.) Tj +1 0 0 1 72 102.96 Tm +([12]) Tj +1 0 0 1 90.2399 102.96 Tm +(D.) Tj +1 0 0 1 100.08 102.96 Tm +(Gif) Tj +1 0 0 1 110.4 102.96 Tm +(ford.) Tj +1 0 0 1 128.76 102.96 Tm +(W) Tj +1 0 0 1 135.72 102.96 Tm +(eighted) Tj +1 0 0 1 161.28 102.96 Tm +(V) Tj +1 0 0 1 165.96 102.96 Tm +(oting) Tj +1 0 0 1 184.44 102.96 Tm +(for) Tj +1 0 0 1 195.719 102.96 Tm +(Replicated) Tj +1 0 0 1 231.959 102.96 Tm +(Data.) Tj +1 0 0 1 251.999 102.96 Tm +(In) Tj +/R794 8 Tf +1 0 0 1 260.639 102.96 Tm +(Symposium) Tj +1 0 0 1 90.24 93.48 Tm +(on) Tj +1 0 0 1 100.2 93.48 Tm +(Oper) Tj +1 0 0 1 116.4 93.48 Tm +(ating) Tj +1 0 0 1 134.88 93.48 Tm +(Systems) Tj +1 0 0 1 162.12 93.48 Tm +(Principles) Tj +/R780 8 Tf +1 0 0 1 194.64 93.48 Tm +(,) Tj +1 0 0 1 198.84 93.48 Tm +(1979.) Tj +1 0 0 1 72 80.52 Tm +([13]) Tj +1 0 0 1 90.2399 80.52 Tm +(M.) Tj +1 0 0 1 102.24 80.52 Tm +(Herlihy) Tj +1 0 0 1 129.36 80.52 Tm +(and) Tj +1 0 0 1 143.76 80.52 Tm +(J.) Tj +1 0 0 1 151.8 80.52 Tm +(T) Tj +1 0 0 1 156.12 80.52 Tm +(ygar) Tj +1 0 0 1 169.68 80.52 Tm +(.) Tj +1 0 0 1 176.88 80.52 Tm +(Ho) Tj +1 0 0 1 186.359 80.52 Tm +(w) Tj +1 0 0 1 194.999 80.52 Tm +(to) Tj +1 0 0 1 204.119 80.52 Tm +(mak) Tj +1 0 0 1 217.679 80.52 Tm +(e) Tj +1 0 0 1 224.159 80.52 Tm +(replicated) Tj +1 0 0 1 258.479 80.52 Tm +(data) Tj +1 0 0 1 274.679 80.52 Tm +(secure.) Tj +/R794 8 Tf +1 0 0 1 90.24 71.04 Tm +(Advances) Tj +1 0 0 1 122.76 71.04 Tm +(in) Tj +1 0 0 1 131.04 71.04 Tm +(Cryptolo) Tj +1 0 0 1 159 71.04 Tm +(gy) Tj +1 0 0 1 168.72 71.04 Tm +(\(LNCS) Tj +1 0 0 1 192.48 71.04 Tm +(293\)) Tj +/R780 8 Tf +1 0 0 1 206.999 71.04 Tm +(,) Tj +1 0 0 1 211.079 71.04 Tm +(1988.) Tj +1 0 0 1 315 709.08 Tm +([14]) Tj +1 0 0 1 333.24 709.08 Tm +(M.) Tj +1 0 0 1 343.92 709.08 Tm +(Herlihy) Tj +1 0 0 1 369.84 709.08 Tm +(and) Tj +1 0 0 1 383.04 709.08 Tm +(J.) Tj +1 0 0 1 389.76 709.08 Tm +(W) Tj +1 0 0 1 396.96 709.08 Tm +(ing.) Tj +1 0 0 1 411.359 709.08 Tm +(Axioms) Tj +1 0 0 1 438.239 709.08 Tm +(for) Tj +1 0 0 1 449.159 709.08 Tm +(Concurrent) Tj +1 0 0 1 486.599 709.08 Tm +(Objects.) Tj +1 0 0 1 515.159 709.08 Tm +(In) Tj +/R794 8 Tf +1 0 0 1 523.439 709.08 Tm +(A) Tj +1 0 0 1 528.119 709.08 Tm +(CM) Tj +1 0 0 1 333.24 699.6 Tm +(Symposium) Tj +1 0 0 1 371.52 699.6 Tm +(on) Tj +1 0 0 1 381.48 699.6 Tm +(Principles) Tj +1 0 0 1 416.28 699.6 Tm +(of) Tj +1 0 0 1 424.44 699.6 Tm +(Pr) Tj +1 0 0 1 432.12 699.6 Tm +(o) Tj +1 0 0 1 435.96 699.6 Tm +(gr) Tj +1 0 0 1 442.919 699.6 Tm +(amming) Tj +1 0 0 1 470.639 699.6 Tm +(Langua) Tj +1 0 0 1 494.759 699.6 Tm +(g) Tj +1 0 0 1 498.599 699.6 Tm +(es) Tj +/R780 8 Tf +1 0 0 1 505.199 699.6 Tm +(,) Tj +1 0 0 1 509.399 699.6 Tm +(1987.) Tj +1 0 0 1 315 686.52 Tm +([15]) Tj +1 0 0 1 333.24 686.52 Tm +(J.) Tj +1 0 0 1 339.72 686.52 Tm +(Ho) Tj +1 0 0 1 349.2 686.52 Tm +(w) Tj +1 0 0 1 354.84 686.52 Tm +(ard) Tj +1 0 0 1 366.48 686.52 Tm +(et) Tj +1 0 0 1 373.68 686.52 Tm +(al.) Tj +1 0 0 1 383.28 686.52 Tm +(Scale) Tj +1 0 0 1 401.879 686.52 Tm +(and) Tj +1 0 0 1 414.839 686.52 Tm +(performance) Tj +1 0 0 1 456.479 686.52 Tm +(in) Tj +1 0 0 1 464.039 686.52 Tm +(a) Tj +1 0 0 1 468.959 686.52 Tm +(distrib) Tj +1 0 0 1 488.999 686.52 Tm +(uted) Tj +1 0 0 1 504.239 686.52 Tm +(\256le) Tj +1 0 0 1 515.879 686.52 Tm +(system.) Tj +/R794 8 Tf +1 0 0 1 333.24 677.04 Tm +(A) Tj +1 0 0 1 337.92 677.04 Tm +(CM) Tj +1 0 0 1 351.84 677.04 Tm +(T) Tj +1 0 0 1 355.8 677.04 Tm +(r) Tj +1 0 0 1 358.8 677.04 Tm +(ansactions) Tj +1 0 0 1 394.92 677.04 Tm +(on) Tj +1 0 0 1 404.88 677.04 Tm +(Computer) Tj +1 0 0 1 438.719 677.04 Tm +(Systems) Tj +/R780 8 Tf +1 0 0 1 463.799 677.04 Tm +(,) Tj +1 0 0 1 467.999 677.04 Tm +(6\(1\),) Tj +1 0 0 1 485.159 677.04 Tm +(1988.) Tj +1 0 0 1 315 663.96 Tm +([16]) Tj +1 0 0 1 333.24 663.96 Tm +(K.) Tj +1 0 0 1 343.56 663.96 Tm +(Kihlstrom,) Tj +1 0 0 1 380.52 663.96 Tm +(L.) Tj +1 0 0 1 389.88 663.96 Tm +(Moser) Tj +1 0 0 1 409.8 663.96 Tm +(,) Tj +1 0 0 1 414.6 663.96 Tm +(and) Tj +1 0 0 1 428.64 663.96 Tm +(P) Tj +1 0 0 1 432.24 663.96 Tm +(.) Tj +1 0 0 1 436.679 663.96 Tm +(Melliar) Tj +1 0 0 1 459.719 663.96 Tm +(-Smith.) Tj +1 0 0 1 487.919 663.96 Tm +(The) Tj +1 0 0 1 502.799 663.96 Tm +(SecureRing) Tj +1 0 0 1 333.24 654.48 Tm +(Protocols) Tj +1 0 0 1 368.28 654.48 Tm +(for) Tj +1 0 0 1 382.44 654.48 Tm +(Securing) Tj +1 0 0 1 415.68 654.48 Tm +(Group) Tj +1 0 0 1 441 654.48 Tm +(Communication.) Tj +1 0 0 1 504.959 654.48 Tm +(In) Tj +/R794 8 Tf +1 0 0 1 516.479 654.48 Tm +(Hawaii) Tj +1 0 0 1 333.24 645 Tm +(International) Tj +1 0 0 1 377.28 645 Tm +(Confer) Tj +1 0 0 1 399 645 Tm +(ence) Tj +1 0 0 1 415.8 645 Tm +(on) Tj +1 0 0 1 425.76 645 Tm +(System) Tj +1 0 0 1 449.879 645 Tm +(Sciences) Tj +/R780 8 Tf +1 0 0 1 476.999 645 Tm +(,) Tj +1 0 0 1 481.319 645 Tm +(1998.) Tj +1 0 0 1 315 632.04 Tm +([17]) Tj +1 0 0 1 333.24 632.04 Tm +(L.) Tj +1 0 0 1 343.68 632.04 Tm +(Lamport.) Tj +1 0 0 1 380.16 632.04 Tm +(T) Tj +1 0 0 1 384.84 632.04 Tm +(ime,) Tj +1 0 0 1 402.6 632.04 Tm +(Clocks,) Tj +1 0 0 1 430.679 632.04 Tm +(and) Tj +1 0 0 1 445.799 632.04 Tm +(the) Tj +1 0 0 1 458.999 632.04 Tm +(Ordering) Tj +1 0 0 1 491.399 632.04 Tm +(of) Tj +1 0 0 1 501.599 632.04 Tm +(Ev) Tj +1 0 0 1 510.359 632.04 Tm +(ents) Tj +1 0 0 1 526.679 632.04 Tm +(in) Tj +1 0 0 1 536.519 632.04 Tm +(a) Tj +1 0 0 1 333.24 622.56 Tm +(Distrib) Tj +1 0 0 1 355.08 622.56 Tm +(uted) Tj +1 0 0 1 370.92 622.56 Tm +(System.) Tj +/R794 8 Tf +1 0 0 1 399.24 622.56 Tm +(Commun.) Tj +1 0 0 1 432 622.56 Tm +(A) Tj +1 0 0 1 436.679 622.56 Tm +(CM) Tj +/R780 8 Tf +1 0 0 1 448.559 622.56 Tm +(,) Tj +1 0 0 1 452.639 622.56 Tm +(21\(7\),) Tj +1 0 0 1 473.879 622.56 Tm +(1978.) Tj +1 0 0 1 315 609.48 Tm +([18]) Tj +1 0 0 1 333.24 609.48 Tm +(L.) Tj +1 0 0 1 343.2 609.48 Tm +(Lamport.) Tj +1 0 0 1 378.36 609.48 Tm +(The) Tj +1 0 0 1 393.72 609.48 Tm +(Part-Time) Tj +1 0 0 1 429.119 609.48 Tm +(Parliament.) Tj +1 0 0 1 471.359 609.48 Tm +(T) Tj +1 0 0 1 475.679 609.48 Tm +(echnical) Tj +1 0 0 1 505.319 609.48 Tm +(Report) Tj +1 0 0 1 530.039 609.48 Tm +(49,) Tj +1 0 0 1 333.24 600 Tm +(DEC) Tj +1 0 0 1 351.24 600 Tm +(Systems) Tj +1 0 0 1 379.8 600 Tm +(Research) Tj +1 0 0 1 411 600 Tm +(Center) Tj +1 0 0 1 431.64 600 Tm +(,) Tj +1 0 0 1 435.84 600 Tm +(1989.) Tj +1 0 0 1 315 586.92 Tm +([19]) Tj +1 0 0 1 333.24 586.92 Tm +(L.) Tj +1 0 0 1 342.36 586.92 Tm +(Lamport,) Tj +1 0 0 1 374.16 586.92 Tm +(R.) Tj +1 0 0 1 383.76 586.92 Tm +(Shostak,) Tj +1 0 0 1 413.4 586.92 Tm +(and) Tj +1 0 0 1 427.199 586.92 Tm +(M.) Tj +1 0 0 1 438.599 586.92 Tm +(Pease.) Tj +1 0 0 1 462.359 586.92 Tm +(The) Tj +1 0 0 1 477.119 586.92 Tm +(Byzantine) Tj +1 0 0 1 511.679 586.92 Tm +(Generals) Tj +1 0 0 1 333.24 577.44 Tm +(Problem.) Tj +/R794 8 Tf +1 0 0 1 368.04 577.44 Tm +(A) Tj +1 0 0 1 372.72 577.44 Tm +(CM) Tj +1 0 0 1 387.72 577.44 Tm +(T) Tj +1 0 0 1 391.68 577.44 Tm +(r) Tj +1 0 0 1 394.68 577.44 Tm +(ansactions) Tj +1 0 0 1 431.88 577.44 Tm +(on) Tj +1 0 0 1 442.92 577.44 Tm +(Pr) Tj +1 0 0 1 450.6 577.44 Tm +(o) Tj +1 0 0 1 454.44 577.44 Tm +(gr) Tj +1 0 0 1 461.399 577.44 Tm +(amming) Tj +1 0 0 1 490.199 577.44 Tm +(Langua) Tj +1 0 0 1 514.319 577.44 Tm +(g) Tj +1 0 0 1 518.159 577.44 Tm +(es) Tj +1 0 0 1 527.999 577.44 Tm +(and) Tj +1 0 0 1 333.24 567.96 Tm +(Systems) Tj +/R780 8 Tf +1 0 0 1 358.32 567.96 Tm +(,) Tj +1 0 0 1 362.52 567.96 Tm +(4\(3\),) Tj +1 0 0 1 379.8 567.96 Tm +(1982.) Tj +1 0 0 1 315 554.88 Tm +([20]) Tj +1 0 0 1 333.24 554.88 Tm +(B.) Tj +1 0 0 1 343.32 554.88 Tm +(Lisk) Tj +1 0 0 1 357.36 554.88 Tm +(o) Tj +1 0 0 1 361.2 554.88 Tm +(v) Tj +1 0 0 1 367.92 554.88 Tm +(et) Tj +1 0 0 1 376.44 554.88 Tm +(al.) Tj +1 0 0 1 389.04 554.88 Tm +(Replication) Tj +1 0 0 1 428.519 554.88 Tm +(in) Tj +1 0 0 1 437.519 554.88 Tm +(the) Tj +1 0 0 1 449.999 554.88 Tm +(Harp) Tj +1 0 0 1 468.599 554.88 Tm +(File) Tj +1 0 0 1 483.719 554.88 Tm +(System.) Tj +1 0 0 1 514.079 554.88 Tm +(In) Tj +/R794 8 Tf +1 0 0 1 523.439 554.88 Tm +(A) Tj +1 0 0 1 528.119 554.88 Tm +(CM) Tj +1 0 0 1 333.24 545.4 Tm +(Symposium) Tj +1 0 0 1 371.52 545.4 Tm +(on) Tj +1 0 0 1 381.48 545.4 Tm +(Oper) Tj +1 0 0 1 397.68 545.4 Tm +(ating) Tj +1 0 0 1 416.16 545.4 Tm +(System) Tj +1 0 0 1 440.279 545.4 Tm +(Principles) Tj +/R780 8 Tf +1 0 0 1 472.799 545.4 Tm +(,) Tj +1 0 0 1 476.999 545.4 Tm +(1991.) Tj +1 0 0 1 315 532.44 Tm +([21]) Tj +1 0 0 1 333.24 532.44 Tm +(N.) Tj +1 0 0 1 342.48 532.44 Tm +(L) Tj +1 0 0 1 346.92 532.44 Tm +(ynch.) Tj +/R794 8 Tf +1 0 0 1 366.24 532.44 Tm +(Distrib) Tj +1 0 0 1 388.56 532.44 Tm +(uted) Tj +1 0 0 1 403.92 532.44 Tm +(Algorithms) Tj +/R780 8 Tf +1 0 0 1 439.199 532.44 Tm +(.) Tj +1 0 0 1 443.279 532.44 Tm +(Mor) Tj +1 0 0 1 456.839 532.44 Tm +(gan) Tj +1 0 0 1 469.679 532.44 Tm +(Kaufmann) Tj +1 0 0 1 504.839 532.44 Tm +(Publishers,) Tj +1 0 0 1 333.24 522.96 Tm +(1996.) Tj +1 0 0 1 315 509.88 Tm +([22]) Tj +1 0 0 1 333.24 509.88 Tm +(D.) Tj +1 0 0 1 343.8 509.88 Tm +(Malkhi) Tj +1 0 0 1 369.48 509.88 Tm +(and) Tj +1 0 0 1 383.76 509.88 Tm +(M.) Tj +1 0 0 1 395.64 509.88 Tm +(Reiter) Tj +1 0 0 1 414.359 509.88 Tm +(.) Tj +1 0 0 1 421.559 509.88 Tm +(A) Tj +1 0 0 1 430.079 509.88 Tm +(High-Throughput) Tj +1 0 0 1 488.999 509.88 Tm +(Secure) Tj +1 0 0 1 513.479 509.88 Tm +(Reliable) Tj +1 0 0 1 333.24 500.4 Tm +(Multicast) Tj +1 0 0 1 364.92 500.4 Tm +(Protocol.) Tj +1 0 0 1 396.12 500.4 Tm +(In) Tj +/R794 8 Tf +1 0 0 1 404.28 500.4 Tm +(Computer) Tj +1 0 0 1 437.759 500.4 Tm +(Security) Tj +1 0 0 1 465.359 500.4 Tm +(F) Tj +1 0 0 1 469.439 500.4 Tm +(oundations) Tj +1 0 0 1 506.399 500.4 Tm +(W) Tj +1 0 0 1 512.279 500.4 Tm +(orkshop) Tj +/R780 8 Tf +1 0 0 1 537.839 500.4 Tm +(,) Tj +1 0 0 1 333.24 490.92 Tm +(1996.) Tj +1 0 0 1 315 477.84 Tm +([23]) Tj +1 0 0 1 333.24 477.84 Tm +(D.) Tj +1 0 0 1 343.32 477.84 Tm +(Malkhi) Tj +1 0 0 1 368.64 477.84 Tm +(and) Tj +1 0 0 1 382.44 477.84 Tm +(M.) Tj +1 0 0 1 393.84 477.84 Tm +(Reiter) Tj +1 0 0 1 412.56 477.84 Tm +(.) Tj +1 0 0 1 418.679 477.84 Tm +(Byzantine) Tj +1 0 0 1 453.239 477.84 Tm +(Quorum) Tj +1 0 0 1 482.159 477.84 Tm +(Systems.) Tj +1 0 0 1 514.439 477.84 Tm +(In) Tj +/R794 8 Tf +1 0 0 1 523.439 477.84 Tm +(A) Tj +1 0 0 1 528.119 477.84 Tm +(CM) Tj +1 0 0 1 333.24 468.36 Tm +(Symposium) Tj +1 0 0 1 371.52 468.36 Tm +(on) Tj +1 0 0 1 381.48 468.36 Tm +(Theory) Tj +1 0 0 1 406.08 468.36 Tm +(of) Tj +1 0 0 1 414.24 468.36 Tm +(Computing) Tj +/R780 8 Tf +1 0 0 1 449.399 468.36 Tm +(,) Tj +1 0 0 1 453.719 468.36 Tm +(1997.) Tj +1 0 0 1 315 455.28 Tm +([24]) Tj +1 0 0 1 333.24 455.28 Tm +(D.) Tj +1 0 0 1 344.88 455.28 Tm +(Malkhi) Tj +1 0 0 1 371.76 455.28 Tm +(and) Tj +1 0 0 1 387.12 455.28 Tm +(M.) Tj +1 0 0 1 399.96 455.28 Tm +(Reiter) Tj +1 0 0 1 418.679 455.28 Tm +(.) Tj +1 0 0 1 428.879 455.28 Tm +(Unreliable) Tj +1 0 0 1 466.319 455.28 Tm +(Intrusion) Tj +1 0 0 1 498.959 455.28 Tm +(Detection) Tj +1 0 0 1 533.759 455.28 Tm +(in) Tj +1 0 0 1 333.24 445.92 Tm +(Distrib) Tj +1 0 0 1 355.08 445.92 Tm +(uted) Tj +1 0 0 1 372.36 445.92 Tm +(Computations.) Tj +1 0 0 1 425.64 445.92 Tm +(In) Tj +/R794 8 Tf +1 0 0 1 435.72 445.92 Tm +(Computer) Tj +1 0 0 1 470.999 445.92 Tm +(Security) Tj +1 0 0 1 500.519 445.92 Tm +(F) Tj +1 0 0 1 504.599 445.92 Tm +(oundations) Tj +1 0 0 1 333.24 436.44 Tm +(W) Tj +1 0 0 1 339.12 436.44 Tm +(orkshop) Tj +/R780 8 Tf +1 0 0 1 364.68 436.44 Tm +(,) Tj +1 0 0 1 368.88 436.44 Tm +(1997.) Tj +1 0 0 1 315 423.36 Tm +([25]) Tj +1 0 0 1 333.24 423.36 Tm +(D.) Tj +1 0 0 1 344.16 423.36 Tm +(Malkhi) Tj +1 0 0 1 370.32 423.36 Tm +(and) Tj +1 0 0 1 384.96 423.36 Tm +(M.) Tj +1 0 0 1 397.2 423.36 Tm +(Reiter) Tj +1 0 0 1 415.92 423.36 Tm +(.) Tj +1 0 0 1 424.199 423.36 Tm +(Secure) Tj +1 0 0 1 449.039 423.36 Tm +(and) Tj +1 0 0 1 463.679 423.36 Tm +(Scalable) Tj +1 0 0 1 493.919 423.36 Tm +(Replication) Tj +1 0 0 1 533.759 423.36 Tm +(in) Tj +1 0 0 1 333.24 413.88 Tm +(Phalanx.) Tj +1 0 0 1 365.64 413.88 Tm +(In) Tj +/R794 8 Tf +1 0 0 1 375 413.88 Tm +(IEEE) Tj +1 0 0 1 395.04 413.88 Tm +(Symposium) Tj +1 0 0 1 434.04 413.88 Tm +(on) Tj +1 0 0 1 444.6 413.88 Tm +(Reliable) Tj +1 0 0 1 473.879 413.88 Tm +(Distrib) Tj +1 0 0 1 496.199 413.88 Tm +(uted) Tj +1 0 0 1 512.759 413.88 Tm +(Systems) Tj +/R780 8 Tf +1 0 0 1 537.839 413.88 Tm +(,) Tj +1 0 0 1 333.24 404.4 Tm +(1998.) Tj +1 0 0 1 315 391.32 Tm +([26]) Tj +1 0 0 1 333.24 391.32 Tm +(B.) Tj +1 0 0 1 342.24 391.32 Tm +(Oki) Tj +1 0 0 1 355.92 391.32 Tm +(and) Tj +1 0 0 1 369.12 391.32 Tm +(B.) Tj +1 0 0 1 378.12 391.32 Tm +(Lisk) Tj +1 0 0 1 392.16 391.32 Tm +(o) Tj +1 0 0 1 396 391.32 Tm +(v) Tj +1 0 0 1 399.48 391.32 Tm +(.) Tj +1 0 0 1 403.8 391.32 Tm +(V) Tj +1 0 0 1 409.08 391.32 Tm +(ie) Tj +1 0 0 1 414.479 391.32 Tm +(wstamped) Tj +1 0 0 1 448.679 391.32 Tm +(Replication:) Tj +1 0 0 1 490.319 391.32 Tm +(A) Tj +1 0 0 1 497.759 391.32 Tm +(Ne) Tj +1 0 0 1 506.759 391.32 Tm +(w) Tj +1 0 0 1 514.319 391.32 Tm +(Primary) Tj +1 0 0 1 333.24 381.84 Tm +(Cop) Tj +1 0 0 1 346.32 381.84 Tm +(y) Tj +1 0 0 1 352.8 381.84 Tm +(Method) Tj +1 0 0 1 379.92 381.84 Tm +(to) Tj +1 0 0 1 388.44 381.84 Tm +(Support) Tj +1 0 0 1 416.04 381.84 Tm +(Highly-A) Tj +1 0 0 1 445.799 381.84 Tm +(v) Tj +1 0 0 1 449.519 381.84 Tm +(ailable) Tj +1 0 0 1 473.399 381.84 Tm +(Distrib) Tj +1 0 0 1 495.239 381.84 Tm +(uted) Tj +1 0 0 1 511.439 381.84 Tm +(Systems.) Tj +1 0 0 1 333.24 372.36 Tm +(In) Tj +/R794 8 Tf +1 0 0 1 343.68 372.36 Tm +(A) Tj +1 0 0 1 348.36 372.36 Tm +(CM) Tj +1 0 0 1 364.08 372.36 Tm +(Symposium) Tj +1 0 0 1 404.28 372.36 Tm +(on) Tj +1 0 0 1 416.04 372.36 Tm +(Principles) Tj +1 0 0 1 452.64 372.36 Tm +(of) Tj +1 0 0 1 462.599 372.36 Tm +(Distrib) Tj +1 0 0 1 484.919 372.36 Tm +(uted) Tj +1 0 0 1 502.559 372.36 Tm +(Computing) Tj +/R780 8 Tf +1 0 0 1 537.719 372.36 Tm +(,) Tj +1 0 0 1 333.24 363 Tm +(1988.) Tj +1 0 0 1 315 349.92 Tm +([27]) Tj +1 0 0 1 333.24 349.92 Tm +(B.) Tj +1 0 0 1 342.36 349.92 Tm +(Preneel) Tj +1 0 0 1 367.92 349.92 Tm +(and) Tj +1 0 0 1 381.24 349.92 Tm +(P) Tj +1 0 0 1 384.84 349.92 Tm +(.) Tj +1 0 0 1 388.56 349.92 Tm +(Oorschot.) Tj +1 0 0 1 422.159 349.92 Tm +(MDx-MA) Tj +1 0 0 1 454.079 349.92 Tm +(C) Tj +1 0 0 1 461.279 349.92 Tm +(and) Tj +1 0 0 1 474.479 349.92 Tm +(Building) Tj +1 0 0 1 504.119 349.92 Tm +(F) Tj +1 0 0 1 508.439 349.92 Tm +(ast) Tj +1 0 0 1 519.119 349.92 Tm +(MA) Tj +1 0 0 1 531.599 349.92 Tm +(Cs) Tj +1 0 0 1 333.24 340.44 Tm +(from) Tj +1 0 0 1 350.76 340.44 Tm +(Hash) Tj +1 0 0 1 369.12 340.44 Tm +(Functions.) Tj +1 0 0 1 405.48 340.44 Tm +(In) Tj +/R794 8 Tf +1 0 0 1 414 340.44 Tm +(Crypto) Tj +1 0 0 1 438.24 340.44 Tm +(95) Tj +/R780 8 Tf +1 0 0 1 446.159 340.44 Tm +(,) Tj +1 0 0 1 450.119 340.44 Tm +(1995.) Tj +1 0 0 1 315 327.36 Tm +([28]) Tj +1 0 0 1 333.24 327.36 Tm +(C.) Tj +1 0 0 1 343.56 327.36 Tm +(Pu,) Tj +1 0 0 1 357.24 327.36 Tm +(A.) Tj +1 0 0 1 367.92 327.36 Tm +(Black,) Tj +1 0 0 1 391.8 327.36 Tm +(C.) Tj +1 0 0 1 402.12 327.36 Tm +(Co) Tj +1 0 0 1 411.12 327.36 Tm +(w) Tj +1 0 0 1 416.759 327.36 Tm +(an,) Tj +1 0 0 1 429.599 327.36 Tm +(and) Tj +1 0 0 1 444.119 327.36 Tm +(J.) Tj +1 0 0 1 452.159 327.36 Tm +(W) Tj +1 0 0 1 459.119 327.36 Tm +(alpole.) Tj +1 0 0 1 486.119 327.36 Tm +(A) Tj +1 0 0 1 494.879 327.36 Tm +(Specialization) Tj +1 0 0 1 333.24 317.88 Tm +(T) Tj +1 0 0 1 337.56 317.88 Tm +(oolkit) Tj +1 0 0 1 357.72 317.88 Tm +(to) Tj +1 0 0 1 365.52 317.88 Tm +(Increase) Tj +1 0 0 1 393.6 317.88 Tm +(the) Tj +1 0 0 1 404.88 317.88 Tm +(Di) Tj +1 0 0 1 412.56 317.88 Tm +(v) Tj +1 0 0 1 416.4 317.88 Tm +(ersity) Tj +1 0 0 1 435.839 317.88 Tm +(of) Tj +1 0 0 1 444.119 317.88 Tm +(Operating) Tj +1 0 0 1 477.599 317.88 Tm +(Systems.) Tj +1 0 0 1 508.319 317.88 Tm +(In) Tj +/R794 8 Tf +1 0 0 1 516.479 317.88 Tm +(ICMAS) Tj +1 0 0 1 333.24 308.4 Tm +(W) Tj +1 0 0 1 339.12 308.4 Tm +(orkshop) Tj +1 0 0 1 366.84 308.4 Tm +(on) Tj +1 0 0 1 376.8 308.4 Tm +(Immunity-Based) Tj +1 0 0 1 431.04 308.4 Tm +(Systems) Tj +/R780 8 Tf +1 0 0 1 456.119 308.4 Tm +(,) Tj +1 0 0 1 460.199 308.4 Tm +(1996.) Tj +1 0 0 1 315 295.32 Tm +([29]) Tj +1 0 0 1 333.24 295.32 Tm +(M.) Tj +1 0 0 1 344.52 295.32 Tm +(Reiter) Tj +1 0 0 1 363.24 295.32 Tm +(.) Tj +1 0 0 1 368.88 295.32 Tm +(Secure) Tj +1 0 0 1 392.76 295.32 Tm +(Agreement) Tj +1 0 0 1 430.319 295.32 Tm +(Protocols.) Tj +1 0 0 1 465.839 295.32 Tm +(In) Tj +/R794 8 Tf +1 0 0 1 474.599 295.32 Tm +(A) Tj +1 0 0 1 479.279 295.32 Tm +(CM) Tj +1 0 0 1 493.439 295.32 Tm +(Confer) Tj +1 0 0 1 515.159 295.32 Tm +(ence) Tj +1 0 0 1 532.079 295.32 Tm +(on) Tj +1 0 0 1 333.24 285.84 Tm +(Computer) Tj +1 0 0 1 367.08 285.84 Tm +(and) Tj +1 0 0 1 381.12 285.84 Tm +(Communication) Tj +1 0 0 1 434.04 285.84 Tm +(Security) Tj +/R780 8 Tf +1 0 0 1 459.839 285.84 Tm +(,) Tj +1 0 0 1 464.039 285.84 Tm +(1994.) Tj +1 0 0 1 315 272.76 Tm +([30]) Tj +1 0 0 1 333.24 272.76 Tm +(M.) Tj +1 0 0 1 346.2 272.76 Tm +(Reiter) Tj +1 0 0 1 364.92 272.76 Tm +(.) Tj +1 0 0 1 375 272.76 Tm +(The) Tj +1 0 0 1 391.2 272.76 Tm +(Rampart) Tj +1 0 0 1 422.52 272.76 Tm +(T) Tj +1 0 0 1 426.84 272.76 Tm +(oolkit) Tj +1 0 0 1 449.159 272.76 Tm +(for) Tj +1 0 0 1 462.359 272.76 Tm +(Building) Tj +1 0 0 1 494.039 272.76 Tm +(High-Inte) Tj +1 0 0 1 524.639 272.76 Tm +(grity) Tj +1 0 0 1 333.24 263.4 Tm +(Services.) Tj +/R794 8 Tf +1 0 0 1 368.88 263.4 Tm +(Theory) Tj +1 0 0 1 394.8 263.4 Tm +(and) Tj +1 0 0 1 410.16 263.4 Tm +(Pr) Tj +1 0 0 1 418.08 263.4 Tm +(actice) Tj +1 0 0 1 440.519 263.4 Tm +(in) Tj +1 0 0 1 450.119 263.4 Tm +(Distrib) Tj +1 0 0 1 472.439 263.4 Tm +(uted) Tj +1 0 0 1 489.599 263.4 Tm +(Systems) Tj +1 0 0 1 518.279 263.4 Tm +(\(LNCS) Tj +1 0 0 1 333.24 253.92 Tm +(938\)) Tj +/R780 8 Tf +1 0 0 1 347.76 253.92 Tm +(,) Tj +1 0 0 1 351.84 253.92 Tm +(1995.) Tj +1 0 0 1 315 240.84 Tm +([31]) Tj +1 0 0 1 333.24 240.84 Tm +(M.) Tj +1 0 0 1 346.92 240.84 Tm +(Reiter) Tj +1 0 0 1 365.64 240.84 Tm +(.) Tj +1 0 0 1 377.64 240.84 Tm +(A) Tj +1 0 0 1 387.96 240.84 Tm +(Secure) Tj +1 0 0 1 414.24 240.84 Tm +(Group) Tj +1 0 0 1 439.199 240.84 Tm +(Membership) Tj +1 0 0 1 484.079 240.84 Tm +(Protocol.) Tj +/R794 8 Tf +1 0 0 1 522.719 240.84 Tm +(IEEE) Tj +1 0 0 1 333.24 231.36 Tm +(T) Tj +1 0 0 1 337.2 231.36 Tm +(r) Tj +1 0 0 1 340.2 231.36 Tm +(ansactions) Tj +1 0 0 1 376.32 231.36 Tm +(on) Tj +1 0 0 1 386.28 231.36 Tm +(Softwar) Tj +1 0 0 1 410.64 231.36 Tm +(e) Tj +1 0 0 1 416.28 231.36 Tm +(Engineering) Tj +/R780 8 Tf +1 0 0 1 455.399 231.36 Tm +(,) Tj +1 0 0 1 459.719 231.36 Tm +(22\(1\),) Tj +1 0 0 1 480.959 231.36 Tm +(1996.) Tj +1 0 0 1 315 218.28 Tm +([32]) Tj +1 0 0 1 333.24 218.28 Tm +(R.) Tj +1 0 0 1 342.84 218.28 Tm +(Ri) Tj +1 0 0 1 350.04 218.28 Tm +(v) Tj +1 0 0 1 353.88 218.28 Tm +(est.) Tj +1 0 0 1 368.64 218.28 Tm +(The) Tj +1 0 0 1 383.4 218.28 Tm +(MD5) Tj +1 0 0 1 402.6 218.28 Tm +(Message-Digest) Tj +1 0 0 1 456.239 218.28 Tm +(Algorithm.) Tj +1 0 0 1 495.119 218.28 Tm +(Internet) Tj +1 0 0 1 522.239 218.28 Tm +(RFC-) Tj +1 0 0 1 333.24 208.8 Tm +(1321,) Tj +1 0 0 1 353.16 208.8 Tm +(1992.) Tj +1 0 0 1 315 195.72 Tm +([33]) Tj +1 0 0 1 333.24 195.72 Tm +(R.) Tj +1 0 0 1 345.84 195.72 Tm +(Ri) Tj +1 0 0 1 353.04 195.72 Tm +(v) Tj +1 0 0 1 356.88 195.72 Tm +(est,) Tj +1 0 0 1 374.04 195.72 Tm +(A.) Tj +1 0 0 1 387 195.72 Tm +(Shamir) Tj +1 0 0 1 409.56 195.72 Tm +(,) Tj +1 0 0 1 417.84 195.72 Tm +(and) Tj +1 0 0 1 434.64 195.72 Tm +(L.) Tj +1 0 0 1 446.759 195.72 Tm +(Adleman.) Tj +1 0 0 1 489.599 195.72 Tm +(A) Tj +1 0 0 1 500.639 195.72 Tm +(Method) Tj +1 0 0 1 530.759 195.72 Tm +(for) Tj +1 0 0 1 333.24 186.24 Tm +(Obtaining) Tj +1 0 0 1 369.6 186.24 Tm +(Digital) Tj +1 0 0 1 396.12 186.24 Tm +(Signatures) Tj +1 0 0 1 434.279 186.24 Tm +(and) Tj +1 0 0 1 450.239 186.24 Tm +(Public-K) Tj +1 0 0 1 478.559 186.24 Tm +(e) Tj +1 0 0 1 481.919 186.24 Tm +(y) Tj +1 0 0 1 490.679 186.24 Tm +(Cryptosystems.) Tj +/R794 8 Tf +1 0 0 1 333.24 176.88 Tm +(Communications) Tj +1 0 0 1 389.28 176.88 Tm +(of) Tj +1 0 0 1 397.44 176.88 Tm +(the) Tj +1 0 0 1 409.2 176.88 Tm +(A) Tj +1 0 0 1 413.88 176.88 Tm +(CM) Tj +/R780 8 Tf +1 0 0 1 425.76 176.88 Tm +(,) Tj +1 0 0 1 429.719 176.88 Tm +(21\(2\),) Tj +1 0 0 1 450.959 176.88 Tm +(1978.) Tj +1 0 0 1 315 163.8 Tm +([34]) Tj +1 0 0 1 333.24 163.8 Tm +(F) Tj +1 0 0 1 337.08 163.8 Tm +(.) Tj +1 0 0 1 341.88 163.8 Tm +(Schneider) Tj +1 0 0 1 372.96 163.8 Tm +(.) Tj +1 0 0 1 380.28 163.8 Tm +(Implementing) Tj +1 0 0 1 427.799 163.8 Tm +(F) Tj +1 0 0 1 432.119 163.8 Tm +(ault-T) Tj +1 0 0 1 450.839 163.8 Tm +(olerant) Tj +1 0 0 1 475.919 163.8 Tm +(Services) Tj +1 0 0 1 505.799 163.8 Tm +(Using) Tj +1 0 0 1 527.639 163.8 Tm +(The) Tj +1 0 0 1 333.24 154.32 Tm +(State) Tj +1 0 0 1 351.36 154.32 Tm +(Machine) Tj +1 0 0 1 381.48 154.32 Tm +(Approach:) Tj +1 0 0 1 418.44 154.32 Tm +(A) Tj +1 0 0 1 426.36 154.32 Tm +(T) Tj +1 0 0 1 430.92 154.32 Tm +(utorial.) Tj +/R794 8 Tf +1 0 0 1 457.079 154.32 Tm +(A) Tj +1 0 0 1 461.759 154.32 Tm +(CM) Tj +1 0 0 1 475.799 154.32 Tm +(Computing) Tj +1 0 0 1 513.479 154.32 Tm +(Surve) Tj +1 0 0 1 531.239 154.32 Tm +(ys) Tj +/R780 8 Tf +1 0 0 1 537.839 154.32 Tm +(,) Tj +1 0 0 1 333.24 144.84 Tm +(22\(4\),) Tj +1 0 0 1 354.48 144.84 Tm +(1990.) Tj +1 0 0 1 315 131.76 Tm +([35]) Tj +1 0 0 1 333.24 131.76 Tm +(A.) Tj +1 0 0 1 343.08 131.76 Tm +(Shamir) Tj +1 0 0 1 365.52 131.76 Tm +(.) Tj +1 0 0 1 370.92 131.76 Tm +(Ho) Tj +1 0 0 1 380.4 131.76 Tm +(w) Tj +1 0 0 1 388.32 131.76 Tm +(to) Tj +1 0 0 1 396.72 131.76 Tm +(share) Tj +1 0 0 1 415.679 131.76 Tm +(a) Tj +1 0 0 1 421.319 131.76 Tm +(secret.) Tj +/R794 8 Tf +1 0 0 1 445.079 131.76 Tm +(Communications) Tj +1 0 0 1 501.239 131.76 Tm +(of) Tj +1 0 0 1 509.519 131.76 Tm +(the) Tj +1 0 0 1 521.399 131.76 Tm +(A) Tj +1 0 0 1 526.079 131.76 Tm +(CM) Tj +/R780 8 Tf +1 0 0 1 537.959 131.76 Tm +(,) Tj +1 0 0 1 333.24 122.28 Tm +(22\(11\),) Tj +1 0 0 1 358.44 122.28 Tm +(1979.) Tj +1 0 0 1 315 109.2 Tm +([36]) Tj +1 0 0 1 333.24 109.2 Tm +(G.) Tj +1 0 0 1 346.44 109.2 Tm +(Tsudik.) Tj +1 0 0 1 382.32 109.2 Tm +(Message) Tj +1 0 0 1 415.56 109.2 Tm +(Authentication) Tj +1 0 0 1 468.359 109.2 Tm +(with) Tj +1 0 0 1 487.919 109.2 Tm +(One-W) Tj +1 0 0 1 510.719 109.2 Tm +(ay) Tj +1 0 0 1 523.679 109.2 Tm +(Hash) Tj +1 0 0 1 333.24 99.72 Tm +(Functions.) Tj +/R794 8 Tf +1 0 0 1 368.52 99.72 Tm +(A) Tj +1 0 0 1 373.2 99.72 Tm +(CM) Tj +1 0 0 1 386.4 99.72 Tm +(Computer) Tj +1 0 0 1 419.64 99.72 Tm +(Communications) Tj +1 0 0 1 475.079 99.72 Tm +(Re) Tj +1 0 0 1 483.359 99.72 Tm +(vie) Tj +1 0 0 1 492.359 99.72 Tm +(w) Tj +/R780 8 Tf +1 0 0 1 497.639 99.72 Tm +(,) Tj +1 0 0 1 501.359 99.72 Tm +(22\(5\),) Tj +1 0 0 1 522.119 99.72 Tm +(1992.) Tj +1 0 0 1 315 86.76 Tm +([37]) Tj +1 0 0 1 333.24 86.76 Tm +(M.) Tj +1 0 0 1 344.76 86.76 Tm +(W) Tj +1 0 0 1 351.96 86.76 Tm +(iener) Tj +1 0 0 1 367.2 86.76 Tm +(.) Tj +1 0 0 1 373.2 86.76 Tm +(Performance) Tj +1 0 0 1 416.28 86.76 Tm +(Comparison) Tj +1 0 0 1 457.679 86.76 Tm +(of) Tj +1 0 0 1 466.679 86.76 Tm +(Public-K) Tj +1 0 0 1 494.999 86.76 Tm +(e) Tj +1 0 0 1 498.359 86.76 Tm +(y) Tj +1 0 0 1 504.959 86.76 Tm +(Cryptosys-) Tj +1 0 0 1 333.24 77.28 Tm +(tems.) Tj +/R794 8 Tf +1 0 0 1 353.16 77.28 Tm +(RSA) Tj +1 0 0 1 368.88 77.28 Tm +(Labor) Tj +1 0 0 1 388.2 77.28 Tm +(atories') Tj +1 0 0 1 415.08 77.28 Tm +(CryptoBytes) Tj +/R780 8 Tf +1 0 0 1 454.199 77.28 Tm +(,) Tj +1 0 0 1 458.399 77.28 Tm +(4\(1\),) Tj +1 0 0 1 475.679 77.28 Tm +(1998.) Tj +/R780 10 Tf +1 0 0 1 300.96 35.16 Tm +(14) Tj +ET +Q +endstream +endobj +800 0 obj +23490 +endobj +801 0 obj +<< +/Type /Page +/MediaBox [0 0 612 792] +/Parent 2 0 R +/Resources << /ProcSet [/PDF /ImageB /Text] +/Font << +/R794 794 0 R +/R791 791 0 R +/R780 780 0 R +>> +/XObject << +/R797 797 0 R +/R787 787 0 R +/R783 783 0 R +>> +>> +/Contents [ +781 0 R +785 0 R +789 0 R +792 0 R +795 0 R +799 0 R +] +>> +endobj +2 0 obj +<< /Type /Pages /Kids [ +47 0 R +123 0 R +237 0 R +355 0 R +444 0 R +582 0 R +640 0 R +690 0 R +700 0 R +714 0 R +724 0 R +737 0 R +779 0 R +801 0 R +] /Count 14 +>> +endobj +1 0 obj +<< /Type /Catalog /Pages 2 0 R +>> +endobj +802 0 obj +<< /CreationDate (D:19990701175312) +/Producer (Aladdin Ghostscript 4.03) +>> +endobj +xref +0 803 +0000000000 65535 f +0000745281 00000 n +0000745116 00000 n +0000000009 00000 n +0000000091 00000 n +0000000264 00000 n +0000000283 00000 n +0000000366 00000 n +0000001199 00000 n +0000001218 00000 n +0000001299 00000 n +0000002850 00000 n +0000002871 00000 n +0000003240 00000 n +0000003260 00000 n +0000003353 00000 n +0000003372 00000 n +0000003452 00000 n +0000003567 00000 n +0000003586 00000 n +0000003950 00000 n +0000003969 00000 n +0000014178 00000 n +0000014200 00000 n +0000014527 00000 n +0000014546 00000 n +0000014640 00000 n +0000014659 00000 n +0000015002 00000 n +0000015021 00000 n +0000015115 00000 n +0000015134 00000 n +0000015412 00000 n +0000015431 00000 n +0000015698 00000 n +0000015718 00000 n +0000016031 00000 n +0000016050 00000 n +0000016328 00000 n +0000016348 00000 n +0000016717 00000 n +0000016737 00000 n +0000033928 00000 n +0000033950 00000 n +0000034265 00000 n +0000034284 00000 n +0000036033 00000 n +0000036054 00000 n +0000036434 00000 n +0000036513 00000 n +0000036532 00000 n +0000036847 00000 n +0000036866 00000 n +0000036959 00000 n +0000036978 00000 n +0000037062 00000 n +0000041913 00000 n +0000041934 00000 n +0000042017 00000 n +0000050049 00000 n +0000050070 00000 n +0000050476 00000 n +0000050496 00000 n +0000050701 00000 n +0000050721 00000 n +0000051068 00000 n +0000051087 00000 n +0000051220 00000 n +0000051239 00000 n +0000051586 00000 n +0000051605 00000 n +0000051737 00000 n +0000051756 00000 n +0000052101 00000 n +0000052120 00000 n +0000052212 00000 n +0000052231 00000 n +0000052559 00000 n +0000052578 00000 n +0000052672 00000 n +0000052691 00000 n +0000053003 00000 n +0000053022 00000 n +0000053371 00000 n +0000053391 00000 n +0000053767 00000 n +0000053787 00000 n +0000053881 00000 n +0000053900 00000 n +0000054247 00000 n +0000054266 00000 n +0000054400 00000 n +0000054419 00000 n +0000054762 00000 n +0000054781 00000 n +0000062506 00000 n +0000062527 00000 n +0000062612 00000 n +0000065080 00000 n +0000065101 00000 n +0000065471 00000 n +0000065492 00000 n +0000067104 00000 n +0000067126 00000 n +0000067456 00000 n +0000067476 00000 n +0000067573 00000 n +0000067593 00000 n +0000067939 00000 n +0000067959 00000 n +0000068056 00000 n +0000068076 00000 n +0000068357 00000 n +0000068377 00000 n +0000068649 00000 n +0000068670 00000 n +0000068986 00000 n +0000069006 00000 n +0000081761 00000 n +0000081784 00000 n +0000082122 00000 n +0000082142 00000 n +0000082394 00000 n +0000082415 00000 n +0000082960 00000 n +0000083046 00000 n +0000083310 00000 n +0000083331 00000 n +0000083669 00000 n +0000083689 00000 n +0000083917 00000 n +0000083938 00000 n +0000084025 00000 n +0000084134 00000 n +0000084154 00000 n +0000084504 00000 n +0000084524 00000 n +0000084662 00000 n +0000084682 00000 n +0000085028 00000 n +0000085048 00000 n +0000088804 00000 n +0000088826 00000 n +0000089192 00000 n +0000089212 00000 n +0000089307 00000 n +0000089327 00000 n +0000089644 00000 n +0000089664 00000 n +0000091615 00000 n +0000091637 00000 n +0000092009 00000 n +0000092030 00000 n +0000092125 00000 n +0000092145 00000 n +0000092427 00000 n +0000092447 00000 n +0000095046 00000 n +0000095068 00000 n +0000095452 00000 n +0000095473 00000 n +0000100713 00000 n +0000100735 00000 n +0000100820 00000 n +0000102949 00000 n +0000102971 00000 n +0000103399 00000 n +0000103420 00000 n +0000103837 00000 n +0000103858 00000 n +0000104230 00000 n +0000104251 00000 n +0000104385 00000 n +0000104405 00000 n +0000104708 00000 n +0000104728 00000 n +0000104823 00000 n +0000104843 00000 n +0000105131 00000 n +0000105151 00000 n +0000105367 00000 n +0000105388 00000 n +0000105699 00000 n +0000105719 00000 n +0000105973 00000 n +0000105994 00000 n +0000106361 00000 n +0000106381 00000 n +0000106841 00000 n +0000106862 00000 n +0000107157 00000 n +0000107177 00000 n +0000110878 00000 n +0000110900 00000 n +0000111279 00000 n +0000111300 00000 n +0000111551 00000 n +0000111572 00000 n +0000111926 00000 n +0000111946 00000 n +0000123472 00000 n +0000123495 00000 n +0000123842 00000 n +0000123862 00000 n +0000124285 00000 n +0000124306 00000 n +0000124663 00000 n +0000124683 00000 n +0000124889 00000 n +0000124910 00000 n +0000125260 00000 n +0000125280 00000 n +0000125702 00000 n +0000125723 00000 n +0000126071 00000 n +0000126091 00000 n +0000126188 00000 n +0000126208 00000 n +0000126539 00000 n +0000126559 00000 n +0000126656 00000 n +0000126676 00000 n +0000126986 00000 n +0000127006 00000 n +0000132994 00000 n +0000133016 00000 n +0000133366 00000 n +0000133386 00000 n +0000133522 00000 n +0000133542 00000 n +0000133880 00000 n +0000133900 00000 n +0000134079 00000 n +0000134100 00000 n +0000134415 00000 n +0000134435 00000 n +0000136756 00000 n +0000136778 00000 n +0000137577 00000 n +0000137663 00000 n +0000137866 00000 n +0000137887 00000 n +0000138225 00000 n +0000138245 00000 n +0000138776 00000 n +0000138797 00000 n +0000139163 00000 n +0000139183 00000 n +0000144084 00000 n +0000144106 00000 n +0000144191 00000 n +0000144806 00000 n +0000144827 00000 n +0000144914 00000 n +0000146305 00000 n +0000146327 00000 n +0000146706 00000 n +0000146727 00000 n +0000147068 00000 n +0000147089 00000 n +0000147498 00000 n +0000147519 00000 n +0000154065 00000 n +0000154087 00000 n +0000154459 00000 n +0000154480 00000 n +0000155594 00000 n +0000155616 00000 n +0000155966 00000 n +0000155986 00000 n +0000156293 00000 n +0000156314 00000 n +0000156617 00000 n +0000156637 00000 n +0000156732 00000 n +0000156752 00000 n +0000157106 00000 n +0000157126 00000 n +0000157340 00000 n +0000157361 00000 n +0000157748 00000 n +0000157769 00000 n +0000157866 00000 n +0000157886 00000 n +0000158234 00000 n +0000158254 00000 n +0000158351 00000 n +0000158371 00000 n +0000158702 00000 n +0000158722 00000 n +0000158819 00000 n +0000158839 00000 n +0000159167 00000 n +0000159187 00000 n +0000163786 00000 n +0000163808 00000 n +0000164126 00000 n +0000164146 00000 n +0000166553 00000 n +0000166575 00000 n +0000166953 00000 n +0000166974 00000 n +0000167315 00000 n +0000167336 00000 n +0000167708 00000 n +0000167729 00000 n +0000169297 00000 n +0000169319 00000 n +0000169669 00000 n +0000169689 00000 n +0000171540 00000 n +0000171562 00000 n +0000171877 00000 n +0000171897 00000 n +0000174315 00000 n +0000174337 00000 n +0000174687 00000 n +0000174707 00000 n +0000175083 00000 n +0000175104 00000 n +0000175450 00000 n +0000175470 00000 n +0000180272 00000 n +0000180294 00000 n +0000180600 00000 n +0000180620 00000 n +0000180912 00000 n +0000180933 00000 n +0000181295 00000 n +0000181315 00000 n +0000181906 00000 n +0000181927 00000 n +0000182222 00000 n +0000182242 00000 n +0000182678 00000 n +0000182699 00000 n +0000183078 00000 n +0000183099 00000 n +0000183356 00000 n +0000183377 00000 n +0000183733 00000 n +0000183753 00000 n +0000184754 00000 n +0000184775 00000 n +0000185086 00000 n +0000185106 00000 n +0000185200 00000 n +0000185220 00000 n +0000185648 00000 n +0000185669 00000 n +0000185920 00000 n +0000185941 00000 n +0000186258 00000 n +0000186278 00000 n +0000194682 00000 n +0000194704 00000 n +0000195525 00000 n +0000195611 00000 n +0000195823 00000 n +0000195844 00000 n +0000195931 00000 n +0000196409 00000 n +0000196430 00000 n +0000196780 00000 n +0000196800 00000 n +0000196897 00000 n +0000196917 00000 n +0000197326 00000 n +0000197347 00000 n +0000197444 00000 n +0000197464 00000 n +0000197767 00000 n +0000197787 00000 n +0000197882 00000 n +0000197902 00000 n +0000198256 00000 n +0000198276 00000 n +0000198410 00000 n +0000198430 00000 n +0000198802 00000 n +0000198823 00000 n +0000198920 00000 n +0000198940 00000 n +0000199286 00000 n +0000199306 00000 n +0000200015 00000 n +0000200036 00000 n +0000200386 00000 n +0000200406 00000 n +0000200873 00000 n +0000200894 00000 n +0000201260 00000 n +0000201280 00000 n +0000201374 00000 n +0000201394 00000 n +0000201711 00000 n +0000201731 00000 n +0000215646 00000 n +0000215669 00000 n +0000216070 00000 n +0000216091 00000 n +0000216449 00000 n +0000216470 00000 n +0000216559 00000 n +0000230221 00000 n +0000230244 00000 n +0000230329 00000 n +0000248912 00000 n +0000248935 00000 n +0000249285 00000 n +0000249305 00000 n +0000249577 00000 n +0000249598 00000 n +0000249985 00000 n +0000250006 00000 n +0000250181 00000 n +0000250202 00000 n +0000250550 00000 n +0000250570 00000 n +0000250667 00000 n +0000250687 00000 n +0000251018 00000 n +0000251038 00000 n +0000251135 00000 n +0000251155 00000 n +0000251470 00000 n +0000251490 00000 n +0000257925 00000 n +0000257947 00000 n +0000258325 00000 n +0000258346 00000 n +0000259001 00000 n +0000259022 00000 n +0000259394 00000 n +0000259415 00000 n +0000259512 00000 n +0000259532 00000 n +0000259827 00000 n +0000259847 00000 n +0000260024 00000 n +0000260045 00000 n +0000260434 00000 n +0000260455 00000 n +0000264335 00000 n +0000264357 00000 n +0000265024 00000 n +0000265110 00000 n +0000267477 00000 n +0000267499 00000 n +0000267849 00000 n +0000267869 00000 n +0000268153 00000 n +0000268174 00000 n +0000268528 00000 n +0000268548 00000 n +0000269253 00000 n +0000269274 00000 n +0000269591 00000 n +0000269611 00000 n +0000270353 00000 n +0000270374 00000 n +0000270724 00000 n +0000270744 00000 n +0000270985 00000 n +0000271006 00000 n +0000271309 00000 n +0000271329 00000 n +0000271591 00000 n +0000271612 00000 n +0000271984 00000 n +0000272005 00000 n +0000272141 00000 n +0000272161 00000 n +0000272539 00000 n +0000272560 00000 n +0000272696 00000 n +0000272716 00000 n +0000273120 00000 n +0000273141 00000 n +0000273316 00000 n +0000273337 00000 n +0000273685 00000 n +0000273705 00000 n +0000273802 00000 n +0000273822 00000 n +0000274153 00000 n +0000274173 00000 n +0000274270 00000 n +0000274290 00000 n +0000274605 00000 n +0000274625 00000 n +0000275271 00000 n +0000275292 00000 n +0000275648 00000 n +0000275668 00000 n +0000276161 00000 n +0000276182 00000 n +0000276548 00000 n +0000276568 00000 n +0000277465 00000 n +0000277486 00000 n +0000277858 00000 n +0000277879 00000 n +0000278086 00000 n +0000278107 00000 n +0000278516 00000 n +0000278537 00000 n +0000280729 00000 n +0000280751 00000 n +0000281130 00000 n +0000281151 00000 n +0000282648 00000 n +0000282670 00000 n +0000283053 00000 n +0000283074 00000 n +0000283208 00000 n +0000283228 00000 n +0000283647 00000 n +0000283668 00000 n +0000283847 00000 n +0000283868 00000 n +0000284196 00000 n +0000284216 00000 n +0000286802 00000 n +0000286824 00000 n +0000286911 00000 n +0000290840 00000 n +0000290862 00000 n +0000291249 00000 n +0000291270 00000 n +0000293321 00000 n +0000293343 00000 n +0000293689 00000 n +0000293709 00000 n +0000293806 00000 n +0000293826 00000 n +0000294172 00000 n +0000294192 00000 n +0000294289 00000 n +0000294309 00000 n +0000294631 00000 n +0000294651 00000 n +0000309821 00000 n +0000309844 00000 n +0000309929 00000 n +0000312270 00000 n +0000312292 00000 n +0000312642 00000 n +0000312662 00000 n +0000313040 00000 n +0000313061 00000 n +0000313407 00000 n +0000313427 00000 n +0000313803 00000 n +0000313824 00000 n +0000314130 00000 n +0000314150 00000 n +0000314446 00000 n +0000314467 00000 n +0000314829 00000 n +0000314849 00000 n +0000315435 00000 n +0000315456 00000 n +0000315751 00000 n +0000315771 00000 n +0000316189 00000 n +0000316210 00000 n +0000316589 00000 n +0000316610 00000 n +0000316871 00000 n +0000316892 00000 n +0000317248 00000 n +0000317268 00000 n +0000321167 00000 n +0000321189 00000 n +0000321588 00000 n +0000321609 00000 n +0000323508 00000 n +0000323530 00000 n +0000323914 00000 n +0000323935 00000 n +0000325980 00000 n +0000326002 00000 n +0000326933 00000 n +0000327019 00000 n +0000327148 00000 n +0000327168 00000 n +0000327567 00000 n +0000327588 00000 n +0000327805 00000 n +0000327826 00000 n +0000328192 00000 n +0000328212 00000 n +0000328307 00000 n +0000328327 00000 n +0000328644 00000 n +0000328664 00000 n +0000329613 00000 n +0000329634 00000 n +0000330023 00000 n +0000330044 00000 n +0000330791 00000 n +0000330812 00000 n +0000331221 00000 n +0000331242 00000 n +0000332529 00000 n +0000332551 00000 n +0000332923 00000 n +0000332944 00000 n +0000336806 00000 n +0000336828 00000 n +0000336913 00000 n +0000340816 00000 n +0000340838 00000 n +0000341192 00000 n +0000341212 00000 n +0000342343 00000 n +0000342365 00000 n +0000342721 00000 n +0000342741 00000 n +0000351020 00000 n +0000351042 00000 n +0000351421 00000 n +0000351442 00000 n +0000351657 00000 n +0000351678 00000 n +0000351973 00000 n +0000351993 00000 n +0000352129 00000 n +0000352149 00000 n +0000352236 00000 n +0000352344 00000 n +0000352364 00000 n +0000352675 00000 n +0000352695 00000 n +0000352791 00000 n +0000352811 00000 n +0000353239 00000 n +0000353260 00000 n +0000372673 00000 n +0000372696 00000 n +0000373187 00000 n +0000373273 00000 n +0000375629 00000 n +0000375651 00000 n +0000375738 00000 n +0000377960 00000 n +0000377982 00000 n +0000378348 00000 n +0000378368 00000 n +0000378462 00000 n +0000378482 00000 n +0000378799 00000 n +0000378819 00000 n +0000388199 00000 n +0000388221 00000 n +0000388306 00000 n +0000397142 00000 n +0000397164 00000 n +0000397540 00000 n +0000397561 00000 n +0000412908 00000 n +0000412931 00000 n +0000413303 00000 n +0000413324 00000 n +0000413419 00000 n +0000413439 00000 n +0000413769 00000 n +0000413789 00000 n +0000413886 00000 n +0000413906 00000 n +0000414252 00000 n +0000414272 00000 n +0000414365 00000 n +0000414385 00000 n +0000414666 00000 n +0000414686 00000 n +0000414958 00000 n +0000414979 00000 n +0000415295 00000 n +0000415315 00000 n +0000415919 00000 n +0000415940 00000 n +0000416312 00000 n +0000416333 00000 n +0000416429 00000 n +0000416449 00000 n +0000416842 00000 n +0000416863 00000 n +0000417754 00000 n +0000417775 00000 n +0000418222 00000 n +0000418307 00000 n +0000418457 00000 n +0000418477 00000 n +0000418563 00000 n +0000421287 00000 n +0000421309 00000 n +0000421396 00000 n +0000537414 00000 n +0000537438 00000 n +0000537650 00000 n +0000537736 00000 n +0000540987 00000 n +0000541009 00000 n +0000541094 00000 n +0000541464 00000 n +0000541485 00000 n +0000541572 00000 n +0000564917 00000 n +0000564940 00000 n +0000565327 00000 n +0000565348 00000 n +0000575726 00000 n +0000575749 00000 n +0000575998 00000 n +0000576084 00000 n +0000582672 00000 n +0000582694 00000 n +0000582779 00000 n +0000587585 00000 n +0000587607 00000 n +0000587694 00000 n +0000615768 00000 n +0000615791 00000 n +0000616003 00000 n +0000616089 00000 n +0000617641 00000 n +0000617663 00000 n +0000617748 00000 n +0000622663 00000 n +0000622685 00000 n +0000622772 00000 n +0000624821 00000 n +0000624843 00000 n +0000624925 00000 n +0000658874 00000 n +0000658897 00000 n +0000659131 00000 n +0000659217 00000 n +0000661105 00000 n +0000661127 00000 n +0000661214 00000 n +0000663997 00000 n +0000664019 00000 n +0000664104 00000 n +0000676997 00000 n +0000677020 00000 n +0000677369 00000 n +0000677389 00000 n +0000690147 00000 n +0000690170 00000 n +0000690542 00000 n +0000690563 00000 n +0000690658 00000 n +0000690678 00000 n +0000691062 00000 n +0000691083 00000 n +0000691219 00000 n +0000691239 00000 n +0000691605 00000 n +0000691625 00000 n +0000691720 00000 n +0000691740 00000 n +0000692057 00000 n +0000692077 00000 n +0000693211 00000 n +0000693233 00000 n +0000693583 00000 n +0000693603 00000 n +0000693779 00000 n +0000693800 00000 n +0000694194 00000 n +0000694215 00000 n +0000694439 00000 n +0000694460 00000 n +0000694806 00000 n +0000694826 00000 n +0000702200 00000 n +0000702222 00000 n +0000702625 00000 n +0000702711 00000 n +0000706687 00000 n +0000706709 00000 n +0000707075 00000 n +0000707095 00000 n +0000708577 00000 n +0000708599 00000 n +0000708916 00000 n +0000708936 00000 n +0000709431 00000 n +0000709452 00000 n +0000709537 00000 n +0000711849 00000 n +0000711871 00000 n +0000711958 00000 n +0000720844 00000 n +0000720866 00000 n +0000721234 00000 n +0000721254 00000 n +0000744800 00000 n +0000744823 00000 n +0000745330 00000 n +trailer +<< /Size 803 /Root 1 0 R /Info 802 0 R +>> +startxref +745423 +%%EOF diff --git a/docs/check_html.sh b/docs/check_html.sh new file mode 100644 index 0000000..5dbbc63 --- /dev/null +++ b/docs/check_html.sh @@ -0,0 +1,11 @@ +#!/bin/bash +cd `dirname $0` +for f in ../*.html *.html libraries/*.html +do + echo " " $f + tidy -lang en_us --doctype html5 -utf8 -access 4 -e -q -o $TMP/fred.html "$f" +done +echo "checked all html files for html5 compliance." + + + diff --git a/docs/client_server.md b/docs/client_server.md new file mode 100644 index 0000000..7212be4 --- /dev/null +++ b/docs/client_server.md @@ -0,0 +1,1149 @@ +--- +title: Client Server Data Representation +--- +# clients and hosts, masters and slaves + +A slave does the same things for a master as a host does for a client. + +The difference is how identity is seen by third parties. The slaves identity +is granted by the master, and if the master switches slaves, third parties +scarcely notice. It the same identity. The client's identity is granted by the +host, and if the client switches hosts, the client gets a new identity, as for +example a new email address. + +If we use [Pake and Opaque](libraries.html#opaque-password-protocol) for client login, then all other functionality of +the server is unchanged, regardless of whether the server is a host or a +slave. It is just that in the client case, changing servers is going to change +your public key. + +Experience with bitcoin is that a division of responsibilities, as between Wasabi wallet and Bitcoin core, is the way to go - that the peer to peer networking functions belong in another process, possibly running on +another machine, possibly running on the cloud. + +You want a peer on the blockchain to be well connected with a well +known network address. You want a wallet that contains substantial value +to be locked away and seldom on the internet. These are contradictory +desires, and contradictory functions. Ideally one would be in a basement +and generally turned off, the other in the cloud and always on. + +Plus, I have come to the conclusion that C and C++ just suck for +networking apps. Probably a good idea to go Rust for the slave or host. +The wallet is event oriented, but only has a small number of concurrent +tasks. A host or slave is event oriented, but has a potentially very large +number of concurrent tasks. Rust has no good gui system, there is no +wxWidgets framework for Rust. C++ has no good massive concurrency +system, there is no Tokio for C++. + +Where do we put the gui for controlling the slave? In the master, of +course. + +# the select problem + +To despatch an `io` event, the standard is `select()`. Which standard sucks +when you have a lot of sockets to manage. + +The recommended method for servers with massive numbers of clients is overlapped IO, of which Wikipedia says: + +> Utilizing overlapped I/O requires passing an `OVERLAPPED` structure to API functions that normally block, including ReadFile(), WriteFile(), and Winsock's WSASend() and WSARecv(). The requested operation is initiated by a function call which returns immediately, and is completed by the OS in the background. The caller may optionally specify a Win32 event handle to be raised when the operation completes. Alternatively, a program may receive notification of an event via an I/O completion port, *which is the preferred method of receiving notification when used in symmetric multiprocessing environments or when handling I/O on a large number of files or sockets*. The third and the last method to get the I/O completion notification with overlapped IO is to use ReadFileEx() and WriteFileEx(), which allow the User APC routine to be provided, which will be fired on the same thread on completion (User APC is the thing very similar to UNIX signal, with the main difference being that the signals are using signal numbers from the historically predefined enumeration, while the User APC can be any function declared as "void f(void* context)"). The so-called overlapped API presents some differences depending on the Windows version used.[1] +> +> Asynchronous I/O is particularly useful for sockets and pipes. +> +> Unix and Linux implement the POSIX asynchronous I/O API (AIO) + +Which kind of hints that there might be a clean mapping between Windows `OVERLAPPED` and Linux `AIO*` + +Because generating and reading the select() bit arrays takes time +proportional to the largest fd that you provided for `select()`, the `select()` +scales terribly when the number of sockets is high. + +Different operating systems have provided different replacement functions +for select. These include `WSApoll()`, `epoll()`, `kqueue()`, and `evports()`. All of these give better performance than select(), all give O(1) performance +for adding a socket, removing a socket, and for noticing that a socket is +ready for IO. (Well, `epoll()` does when used in edge triggered (`EPOLLET`) +mode. It has a `poll()` compatibility mode which fails to perform when you +have a large number of file descriptors) + +Windows has `WSAPoll()`, which can be a blocking call, but if it blocks +indefinitely, the OS will send an alert callback to the paused thread +(asynchronous procedure call, APC) when something happens. The +callback cannot do another blocking call without crashing, but it can do a +nonblocking poll, followed by a nonblocking read or write as appropriate. +This analogous to the Linux `epoll()`, except that `epoll()` becomes ungodly +slow, rather than crashing. The practical effect is that "wait forever" +becomes "wait until something happens that the APC did not handle, or +that the APC deliberately provoked") + +Using the APC in Windows gets you behavior somewhat similar in effect +to using `epoll()` with `EPOLLET` in Linux. Not using the APC gets you +behavior somewhat similar in effect to Linux `poll()` compatibility mode. + +Unfortunately, none of the efficient interfaces is a ubiquitous standard. Windows has `WSAPoll()`, Linux has `epoll()`, the BSDs (including Darwin) have `kqueue`(), … and none of these operating systems has any of the others. So if you want to write a portable high-performance asynchronous application, you’ll need an abstraction that wraps all of these interfaces, and provides whichever one of them is the most efficient. + +The Libevent api wraps various unix like operating system efficient replacements, but unfortunately missing from its list is the windows efficient replacement. + +The way to make them all look alike is to make them look like event +handlers that have a pool of threads that fish stuff out of a lock free +priority queue of events, create more threads capable of handling this kind +of event if there is a lot of stuff in the queue and more threads are needed, +and release all threads but one that sleeps on the queue if the queue is +empty and stays empty. + +Trouble is that windows and linux are just different. Except both support +select, but everyone agrees that select really sucks, and sucks worse the +more connections. + +A windows gui program with a moderate number of connections should use windows asynchronous sockets, which are designed to deliver events on the main windows gui event loop, designed to give you the benefits of a separate networking thread without the need for a separate networking thread. Linux does not have asynchronous sockets. Windows servers should use overlapped io, because they are going to need ten thousand sockets, they do not have a window + +Linux people recommended a small number of threads, reflecting real hardware threads, and one edge triggered `epoll()` per thread, which sounds vastly simpler than what windows does. + +I pray that that wxWidgets takes care of mapping windows asynchronous sockets to their near equivalent functionality on Linux. + +But writing a server/host/slave for Linux is fundamentally different to +writing one for windows. Maybe we can isolate the differences by having +pure windows sockets, startup and shutdown code, pure Linux sockets, +startup and shutdown code, having the sockets code stuff data to and from +lockless priority queues (which revert to locking when a thread needs to +sleep or startup) Or maybe we can use wxWidgets. Perhaps worrying +about this stuff is premature optimization. But the samples directory has +no service examples, which suggests that writing services in wxWidgets is +a bad idea. And it is an impossible idea if we are going to write in Rust. + +Tokio, however, is a Rust framework for writing services, which runs on +both Windows and Linux. Likely Tokio hides the differences, in a way +optimal for servers, as wxWidgets hides them in a way optimal for guis. + +# the equivalent of RAII in event oriented code + +Futures, promises, and cooperative multi tasking. + +Is asynch await. Implemented in a Rust library. + +This is how a server can have ten thousand tasks dealing with ten thousand clients. + +Implemented, in C++20 as co_return, co_await, and co_yield, co_yield +being the C++ equivalent of Rust’s poll. But C++20 has no standard +coroutine libraries, and various people’s half baked ideas for a coroutine +library don’t seem to be in actual use solving real problems just yet, while +actual people are using the Rust library to solve real world problems. + +I have read reviews by people attempting to use C++20 co-routines, and +the verdict is that they are useless and unusable, + +And we should use fibres instead. Fibres? + +Boost fibres provide multiple stacks on a single thread of execution. But +the consensus is that [fibres just massively suck](https://devblogs.microsoft.com/oldnewthing/20191011-00/?p=102989). + +But, suppose we don't use stack. We just put everything into a struct and disallow recursion (except you create a new struct) Then we have the functionality of fibres and coroutines, with code continuations. + +Word is that co_return, co_await, and co_yield do stuff that is +complicated, difficult to understand, and frequently surprising and not +what you want, but with std::future, you can reasonably straightforwardly +do massive concurrency, provided you have your own machinery for +scheduling tasks. Maybe we do massive concurrency with neither fibres, +nor coroutines -- code continuations or close equivalent. + +> we get not coroutines with C++20; we get a coroutines framework. +> This difference means, if you want to use coroutines in C++20, +> you are on your own. You have to create your coroutines based on +> the C++20 coroutines framework. + +C++20 coroutines seem to be designed for the case of two tasks each of +which sees the other as a subroutine, while the case that actually matters +in practice is a thousand tasks holding a relationship with a thousand clients. +(Javascript’s async). It is far from obvious how one might do what +Javascript does using C++20 coroutines, while it is absolutely obvious +how to do it with Goroutines. + +## Massive concurrency in Rust + +Well supported, works, widely used. + +The way Rust does things is that the input that you are waiting for is itself a +future, and that is what drives the cooperative multi tasking engine. + +When the event happens, the future gets flagged as fulfilled, so the next +time the polling loop is called, co_yield never gets called. And the polling +loop in your await should never get called, except the event arrives on the +event queue. The Tokio tutorial explains the implementation in full detail. + +From the point of view of procedural code, await is a loop that endlessly +checks a condition, calls yield if the condition is not fulfilled, and exits the +loop when the condition is fulfilled. But you would rather it does not +return from yield/poll until the condition is likely to have changed. And +you would rather the outermost future pauses the thread if nothing has +changed, if the event queue is empty. + +The right way to implement this is have the stack as a tree. Not sure if +Tokio does that. C++20 definitely does not – but then it does not do +anything. It is a pile of matchsticks and glue, and they tell you to build +your own boat. + +[Tokio tutorial]:https://tokio.rs/tokio/tutorial/async + +The [Tokio tutorial] discusses this and tells us how they dealt with it. + +> Ideally, we want mini-tokio to only poll futures when the future is +> able to make progress. This happens when a resource that the task +> is blocked on becomes ready to perform the requested operation. If +> the task wants to read data from a TCP socket, then we only want +> to poll the task when the TCP socket has received data. In our case, +> the task is blocked on the given Instant being reached. Ideally, +> mini-tokio would only poll the task once that instant in time has +> passed. + +> To achieve this, when a resource is polled, and the resource is not +> ready, the resource will send a notification once it transitions into a +> ready state. + +The mini Tokio tutorial shows you how to implement your own efficient +futures in Rust, and, because at the bottom you are always awaiting an +efficient future, all your futures will be efficient. You have, however +all the tools to implement an inefficient future, and if you do, there will be a +lot of spinning. So if everyone is inefficiently waiting on a future that is +inefficiently waiting on a future that is waiting on a network event or +timeout, and the network events and timeout futures are implemented +efficiently, you are done. + +If you cheerfully implement an inefficient future, which however calls an +efficient future, it stops spinning. + +> When a future returns Poll::Pending, it must ensure that the wake +> is signalled at some point. Forgetting to do this results in the task +> hanging indefinitely + +Multithreading, as implemented in C++, Rust and Julia do not scale to +huge numbers of concurrent processes the way Go does. + +notglowing, a big fan of Rust, tells me, + +> No, but like in most other languages, you can solve that with +> asynchronous code for I/O bound operations. Which is the kind +> of situation where you’d consider Go anyways. + +> With Tokio, I can spawn an obscene number of Tasks doing I/O +> work asynchronously, and only use a few threads. + +> It works really well, and I have been writing async code using +> Tokio for a project I am working on. + +> Async/await semantics are the next best thing after Goroutines. + +> Frameworks like Actix-web leverage Tokio to get web +> server performance superior to any Go framework I know of. + +> Go’s concurrency model might be superior, but Rust’s lightweight +> runtime, lack of GC pauses that can cause inconsistent +> performance, and overall low level control give it the edge it needs +> to beat Go in practical scenarios. + +I looked up Tokio and Actix, looks like exactly what the doctor ordered. + +So, if you need asynch, you need Rust. C++ is build your own boat out of +matchsticks and glue. + +The await asynch syntax and semantics are, in effect, multithreading on cheap +threads that only have cooperative yielding. + +So you have four real threads, and ten thousand tasks, the effective +equivalent of ten thousand cheap "threads". + +I conjecture that the underlying implementation is that the asynch await +keywords turn your stack into a tree, and each time a branch is created and +destroyed, it costs a small memory allocation/deallocation. + +With real threads, each thread has its own full stack, and stacks can be +costly, while with await/asynch, each task is just a small branch of the tree. +Instead of having one top of stack, you have a thousand leaves with one +root at start of thread, while having ten thousand full stacks would bring +your program to a grinding halt. + +It works like an event oriented program, except the message pumps do not +have to wait for events to complete. Tasks that are waiting around, such as +the message pump itself, can get started on the next thing, while the +messages it dispatched are waiting around. + +As recursing piles more stuff on the stack, asynching branches the stack, +while masses of threads give you masses of stacks, which can quickly +bring your computer to a grinding halt. + +Resource acquisition, disposition, and release depend on network and timer +events. + +RAII guarantees that the resource is available to any function that may +access the object (resource availability is a class invariant, +eliminating redundant runtime tests). It also guarantees that all +resources are released when the lifetime of their controlling object +ends, in reverse order of acquisition. + +In a situation where a multitude of things can go wrong, but seldom do, +you would, without RAII, wind up with an exponentially large number of +seldom tested code paths for backing out of the situation. RAII means that +all the possibilities are automagically taken care of in a consistent way, +and you don’t have to think about all the possible combinations and +permutations. + +RAII plus exceptions shuts down an potentially exponential explosion of +code paths. + +Our analog of the situation that RAII deals with is that we dispatch +messages A and B, and create the response handler for B in the response +handler for A. But A might fail, and B might get a response before A. + +With await asynch, we await A, then await B, and if B has already arrived, +our await for B just goes right ahead, never calling yield, but removing +itself from the awake notifications. + +Go already has technology and idiom for message handling. Maybe the +solution for this problem is not to re-invent Go technology in C++ using +[perfect forwarding] and lambda functions, but to provide a message +interface to Go in C. + +But there is less language mismatch between Rust and C++ than between +Go and C++. + +And maybe C++20 has arrived in time, have not checked the availability +of co_await, co_return, and co_yield. + +[perfect forwarding]:https://cpptruths.blogspot.com/2012/06/perfect-forwarding-of-parameter-groups.html + +On the other hand, Go’s substitute for RAII is the defer statement, which +presupposes that a resource is going to be released at the same stack level +as it was acquired, whereas when I use RAII I seldom know, and it is often +impossible to predict, at what stack level a resource should be released, +because the resource is owned by a return value, typically created by a +constructor. + +On checking out Go’s implementation of message handling, it is all things +for which C++ provides the primitives, and Go has assembled the +primitives into very clean and easy to use facilities. Which facilities are +not hard to write in C++. + +The clever solution used by Go is typed, potentially bounded [channels], +with a channel being able to transmit channels, and the select statement. +You can also do all the hairy shared memory things you do in C++, with +less control and elegance. But you should not. + +[channels]:https://golang.org/doc/effective_go#concurrency +"concurrency" + +What makes Go multithreading easy is channels and select. + +This an implementation of Communicating Sequential Processes, which is +that input, output, and concurrency are useful and effective primitives, that +can elegantly and cleanly express algorithms, even if they are running on a +computer that physically can only execute a single thread, that +concurrency as expressed by channels is not merely a safe way of +multithreading, but a clean way of expressing intent and procedure to the +computer. + +Goroutines are less than a thread, because they are using some multiplexed +thread’s stack. They live in an environment where a stable and small pool +of threads is despatched to function calls, and when a goroutine is stopped, +because it is attempting to communicate, its stack state, which is +usually quite small, is stashed somewhere without destroying and creating an +entire thread. They need to lightweight, because used to express +algorithms, with parallelism not necessarily being an intended side effect. + +The relationship between goroutines and node.js continuations is that a +continuation is a small packet of state that will receive an event, and a +paused goroutine is a small packet of state that will receive an event. Both +approaches seem comparably successful in expressing concurrent algorithms, +though node.js is single threaded. + +Node.js uses async/await, and by and large, more idiots are successfully +using node.js than successfully using Go, though go solutions are far more +lightweight than node.js solutions, and in theory Rust should be still +lighter. + +But Rust solutions should be even lighter weight than Go solutions. + +So maybe I do need to invent a C++ idiom for this problem. Well, a Rust +library already has the necessary idiom. Use Tokio in Rust. Score for +powerful macro language and sum types. Language is more expandable. + +# Unit Test + +It is hard to unit test a client server system, therefore, most people +unit test using mocks: Fake classes that do not really interact with the +external world replace the real classes, if you perform that part of the +unit test that deals with external interaction with clients and servers. +Your unit test runs against a dummy client and a dummy server – thus the +unit test code necessarily differs from the real code. + +But this will not detect bugs in the real class being mocked, which therefore +has to be relatively simple and unchanging – not that it is necessarily all +that practical to keep it simple and unchanging, consider the messy +irregularity of TCP. + +Any message is an event, and it is a message between an entity identified +by one elliptic point, and an entity identified by another elliptic point. + +We intend that a process is identified by many elliptic points, so that it has +a stable but separate identity on every server. Which implies that it can +send messages to itself, and these will look like messages from outside. +The network address will be an opaque object. Which is not going to help +us unit test code that has to access real network addresses, though our +program undergoing unit test can perform client operations on itself, +assuming in process loopback is handled correctly. Or maybe we just have +to assume a test network, and our unit test program makes real accesses +over the real internet. + +But our basic architecture is that we have an opaque object representing a +communication node, it has a method that creates a connection, and you +can send a message on a connection, and receive a reply event on that +connection. + +Sending a message on a connection creates the local object that will handle +the reply, and this local object’s lifetime is managed by hash code tables -- +or else this local object is stored in the database, written to disk in the +event that sends the message, and read from disk in the event that handles +the reply to the message. + +Object representing server 🢥 Object representing connection to server 🢥 +object representing request-response. + +We send messages between entities identifed by their elliptic points, we +get events on the receiving entity when these events arrive, generate +replies, and get an event on the sending entity when the reply is received. + +And one of the things in these messages will be these entities and +information about these entities. + +So we create our universal class, which may be mocked, whereby a client +takes an opaque data structure representing a server, and makes a request, +thereby creating a channel, on which channel it can create additional +requests. It can then receive a reply on this channel, and make further +requests, or replies to replies, sequentially on this channel. + +We then layer this class on top of this class – as for example setting up a +shared secret, timing out channels and establishing new ones, so we have +as much real code as possible, implementing request object classes in +terms of request object classes, so that we can mock any one layer in the +hierarchy, + +At the top layer, we don’t know we are re-using channels, and don’t know we +are re-using secrets – we don’t even keep track of the transient secret +scalar and transient shared secret point, because that might be discarded and +reconstructed. All this stuff lives in an opaque object representing the +current state of our communication with the server, which is, at the topmost +level, identified a database record, and/or an objected instantiated from a +database record and/or a handle to that object and/or a hash code to that +handle. + +Since we are using an opaque object of an opaque type, we can freely mix +fake objects with real ones. Unit test will result in fake communications +over fake channels with fake external clients and servers. + +# Factorizing the problem + +Why am I reinventing OMEMO, XMPP, and OTR? + +These projects are quite large, and have a vast pile of existing code. + +On the other hand OTR seems an unreasonably complicated way of +adding on what you get for free with perfect forward secrecy, +authentication without signing is just the natural default for perfect +forward secrecy, and signing has to be added on top. You get OTR (Off the +Record) for free just by leaving stuff out. XMPP is a presence protocol is +just name service, which is integral to any social networking system. Its +pile of existing code supports Jitsi’s wonderful video conferencing system, +which would be intolerably painful to reinvent. + +And OMEMO just does not do the job. It guarantees you have a private +room with the people you think you have a private room with, but how did +you come to decide you wanted a private room with those people and not +others? It leaves the hard part of the problem out of scope. + +The problem is factoring a heap of problems that lack obvious boundaries +between one problem and the next. You need to find the smallest factors +that are factors of all these big problems – find a solution to your problems +that is a suitable module of a solution to all these big problems. + +But you don’t want to factorize all the way down,otherwise when you +want a banana, you will get a banana tree, a monkey, and a jungle. You +want the largest factors that are common factors of more than one problem +that you have to solve. + +And a factor that we identify is that we create a shared secret with a +lifetime of around twenty minutes or so, longer than the lifetime of the +TCP connections and longer than the query-response interactions, that +ensures: + +* Encryption (eavesdroppers learn almost nothing) +* Authentication (the stable identity of the two wallets, no man in the + middle attack) +* Deniability (the wallets have proof of authentication, but no proof + that they can show to someone else) +* Perfect forward secrecy (if the wallet secrets get exposed, their + communications remain secret) + +Another factor we identify is binding a group of remote object method +calls together to a single one that must fail together, of which problem +a reliability layer on top of UDP is a special case. But we do not want +to implement our own UDP reliability layer, when [QUIC], has already been +developed and widely deployed. We notice that to handle this case, we +need not an event object referenced by an event handle and an event +hashcode, but rather an array of event objects referenced by an event +handle, an event hashcode, and the sequence number within that vector. + +## streams, secrets, messages, and authentication + +To leak minimal metadata, we should encrypt the packets with +XChaCha20-SIV, or use a random nonce. (Random nonces +are conveniently the default case libsodium's `crypto_box_easy`). The port +should be random for any one server and any one client, to make it slightly +more difficult to sweep up all packets using our encryption. Any time we +distribute new IP information for a server, also distribute new open port information. + +XChaCha20-SIV is deterministic encryption, and deterministic encryption +will leak information unless every message sent with a given key is +guaranteed to be unique - in effect, we have the nonce inside the +encryption instead of outside. Each packet must contain a forever +incrementing packet number, which gets repeated but with a different send +time, and perhaps an incremented resend count, on reliable messaging +resends. This gets potentially complicated, hard to maintain, and easy to break. + +Neither protocol includes authentication. The crypto_box wraps the +authentication with the encryption. You need to add the authenticator after +encryption and before decryption, as crypto_box does. The principle of +cryptographic doom is that if you don't, someone will find some clever +way of using the error messages your higher level protocol generates to +turn it into a decryption/encryption oracle. + +Irritatingly, `crypto_box_easy_*` defaults to XSalsa, and I prefer XChaCha. + +However `crypto_box_curve25519xchacha20poly1305.*easy.*` in +`crypto_box_curve25519xchacha20poly1305.h` wraps it all together. You +just have to call those instead of `crypto_box_easy.*` Which is likely to be +a whole lot easier and safer than wrapping XChaCha20-SIV. + +For each `crypto_box` function, there is a corresponding +`crypto_box_curve25519xchacha20poly1305` function, apart from +some special cases that you probably should not be using anyway. + +So, redefine each crypto_box function to use XChaCha20 + +```C++ +namespace crypto_box{ + const auto& «whatever» = crypto_box_curve25519xchacha20poly1305_«whatever»; +} +``` + +Nonces are intended to be communicated in the clear, thus sequential +nonces inevitably leak metadata. Don't use sequential nonces. Put the +packet number and message number or numbers inside the authenticated encryption. + +Each packet of a packetized message will contain the windowed +message id of the larger message of which it is part, the id of the thread or +thread pool that will ultimately consume it, the size of the larger message +of which it is part, the number of packets in the larger message of which it +is part, and its packet and byte position within that larger message. The +repetition is required to handle out of order messages and messages with +lost packets. + +For a single packet message, or a multi message packet, each message +similarly. + +Message ids will be windowed sequential, and messages lost in entirety will be reliably resent because their packets will be reliably resent. + +If we allocate each packet buffer from the heap, and free it when it is used, +this does not make much of a dent in performance until we are processing +well over a Gib/s. + +So we can worry about efficient allocation after we have released software +and it is coming under heavy load. + +Another more efficient way would be to have a pool of 16KiB blocks, +allocate one of them to a connection whenever that connection needs it, +allocate packet buffers sequentially in a 16KiB block, incrementing a +count, free up packet buffers in the bloc when a packet is dealt with, +decrementing the count. When the count returns to zero, it goes back to the +free pool, which is accessed in lifo order. Every few seconds the pool is +checked, and if there are number of buffers that have not been used in the +last few minutes, we free them. We organize things that inactive +connection has no packet buffers associated with it. But this is fine tuning +and premature optimization. + +The recipient will nack the sender about any missing packets within a +multipacket message. The sender will not free up any memory containing +packets that have not been acked, and the receiver will not free up any +memory that has not been handled by the thread that ultimately receives +the data. + +Experimenting with memory allocation and deallocation times, looks like +a sweet spot is to allocate in 16KiB blocks, with the initial fifo queue +being allocated with two 16KiB blocks as soon as activity starts, and the +entire fifo queue deallocated when it is empty. If we allocated, deallocated +when activity stops, and re-allocated every millisecond, it would not +matter much, and we will be doing it far less often than that, because we +will keeping the buffer around for at least one round trip time. If every +active queue has on average sixty four KiB, and we have sixteen thousand +simultaneous active connections, only costs a gigabyte. This rather +arbitrary guesstimated value seems good enough that it does not waste too +much memory, nor too much time. Memory for input output streams +seems cheap, might as well cheerfully spend plenty, perhaps a lot more +than necessary, so as to avoid hitting other limits. + +We want connections, the shared secrets, identity data, and connection parameters, hanging around for a very long time of inactivity, because +they are something like logins. We don't want their empty data stream +buffers hanging around. Re-establishing a connection takes hundreds +of times longer that allocating and deallocating a buffer. + +We also want, in a situation of resource starvation, to cut back the +connections that are the heaviest users to wait. They should not send, until +told space is available, and we just don't make it available, because their +buffer got emptied out, then thrown away, and they just have to wait their +turn till the server clears them to get a new one allocated when they send data. + +If the server has too much work, a whole lot of connections get idled for +longer and longer periods, and while idled, their buffers are discarded. + +When we have a real world application facing real world heavy load, then +we can fuss about fine tuning the parameters. + +The packet stream that is being resolved (the packets, their time of arrival and +sending, that they were acked, nacked, ack status, and all that, goes into a +first in first out random access queue, composed of fixed size blocks larger than +the packet size. + +We hope that C++ implements large random access fifo queues with +mmap. If it does not, will eventually have to write our own. + +Each block starts with metadata that enables the stream of fixed sized +blocks to be interpreted as a stream of variable sized packets and the +metadata about those packets. The block size in bits, and the size of the +block and packet metadata, but initially only 4K byte, 32K kilobit blocks +will be supported. The format of metadata that is referenced or defined +within packets is also negotiated, though initially the only format will be +format number one. Obviously each side is free to define its own format for +the metadata outside of packets, but it has to be the same size at both +ends. Each party can therefore demand any metadata size it wants, subject +to some limit, for metadata outside the packets. + +The packets are aligned within the blocks so that 512 bit blocks to be +encrypted or decrypted are aligned with the blocks of the queue so the +blocks of the queue are always a multiple of 512 bits, 32 bytes, and block +size is given as a multiple of 32 bytes. This will result in an average of +sixteen bytes of space wasted positioning each packet to a boundary. + +The pseudo random streams of encrypting information are applied with an +offset that depends on the absolute position in the queue, which is why +the queues have to have packets in identical position in both queues. +Each block header contains unwindowing values for any windowed values in +the packets and packet metadata, which unwindowing data is a mere 64 bits, +but, since block and packet metadata size gets negotiated on each +connection, this can be expanded without breaking backwards +compatibility. The format number for packet references to metadata +implies an unwindow size, but we initially assume that any connection only +sends less that 2^64 512 bit packets, rather packets plus the metadata +required to describe those packets takes up less than 2^73 bits, +corresponding to a thousand Mbps + +The packet position in the queue is the same at both ends, and is +unwindowed in the block header. + +The fundamental architecture of QUIC is that each packet has its own +nonce, which is an integer of potentially sixty two bits, expressed in +a form that is short for smaller integers, which is essentially my +design, so I expect that I can use a whole lot of QUIC code. + +It negotiates the AES session once per connection, and thereafter, it +is sequential nonces all the way. + +Make a new one time secret from a new one time public key every time +you start a stream (pair of one way streams). Keeping one time secrets +around for multiple streams, although it can in theory be done safely, gets +startlingly complicated really fast, with the result that nine times out of ten +it gets done unsafely. + +Each two way stream is a pair of one way streams. Each encryption packet +within a udp packet will have in the clear its stream number and a window +into its stream position, the window size being log base two of the position +difference between all packets in play, plus two, rounded up to the nearest +multiple of seven. Its stream number is an index into shared secrets and stream +states associated with this IP and port number. + +If initiating a connection in the clear (and thus unauthenticated) Alice +sends Bob (in a packet that is not considered part of a stream) a konce (key used once, single use +elliptic point $A_o$). She follows it, in the same packet and in a new +encrypted but unauthenticated stream, proving knowledge of the scalar +corresponding to the elliptic point by using the the shared secret +$a_oB_d = b_dA_o$, where $B_d$ is Bob’s durable public key and $b_d$ his +durable secret key. In the encrypted but unauthenticated stream, she sends +$A_d$, her durable public key, (which may only be durable until the +application is shut down) initiating a stream encrypted with +$(a_o+a_d)B_d =b_d(A_o+A_d)$, or more precisely, symmetrically encrypted +with the 384 bit hash of that elliptic point and one way stream number). + +All this stuff happens during the handshake, and when we allocate a +receive buffer, we have a shared secret. The sender may only send up to +the size of the receive buffer, and has to wait for acks which will +announce more receive buffer. + +There is no immediate reason to provide the capability to create a new +differently authenticated stream from within an authenticated stream, for +the use cases for that are better dealt with by sending authorizations for +them existing authentication signed by the other party. Hence one to one +mapping between port number and durable authenticating elliptic point, +with each authenticated stream within that port number deriving its shared +secret from a konce covers all the use cases that occur to +me. We don’t care about making creating a login relationship efficient. + +When the OS gives you a packet, it gives you the handle you associated +with that network address and port number, and the protocol layer of +application then has to expand that into the receive stream number and +packet position in the stream. After decrypting the streams within a packet, +it then maps stream id and message id to the application layer message +handler id. It passes the position of data within the message, but not the +position within the stream because you don’t want too many copies of the +shared secret floating around, and because the application does not care. + +Message data may arrive out of sequence within a message, but the +protocol layer always sends the data in sequence to the application, and +usually the application only wants complete messages, and does not +register a partial message handler anyway. + +Each application runs its own instance of the protocol layer, and each +application is, as far as it knows or cares, sending messages identified by +their receiver message handler and reply message handler to a party +identified by its zooko id. A message always receives a reply, even if the +reply is only “message acknowledged”, “message abandoned”, “message +not acknowledged” “timeout”, “graceful shutdown of connection”, or +“ungraceful shutdown of connection”, The protocol layer maps these into +encrypted sequential streams and onto message numbers within the stream +when sending them out, and onto application ids, application message +handlers and receiving zooko ids when receiving them. + +But, if a message always receives a reply, the sender may want to know +which message is being replied to. Which implies it always receives a +handle to the sent message when it gets the reply. Which implies that the +protocol layer has to provide unique reply ids for all messages in play +where a substantive reply is expected from the recipient. (“Message +received” does not need a reply id, because implicit in the reliable +transport layer, but special casing such messages to save a few bytes per +message adds substantially to complexity. Easier to have the recipient ack +all packets and all messages every round trip time, even though acking +messages is redundant, and identifying every message is redundant.) + +This is implies that the protocol layer gives every message a unique sixty +four bit windowed id, with the window size sufficient to cover all +messages in play, all messages that have neither been acked nor +abandoned. + +Suppose we are transferring one dozen eight terabyte disks in tiny fifty +byte messages. And suppose that all these messages are in play, which +seems unlikely unless we are communicating with someone on Pluto. Well, +then we will run out of storage for tracking every message in play, +but suppose we did not. Then forty bits would suffice, a sixty four bit +message id suffices. And, since it is windowed, using the same windowing +as we are using for stream packet 384 bit ids, we can always increase it +without changing the protocol on the wire when we get around to sending +messages between galaxies. + +A windowed value represents an indefinitely large unsigned integer, but +since we are usually interested in tracking the difference between two such +values, we define substraction and comparison on windowed values to +give us ordinary signed integers, the largest precision integer than we can +conveniently represent on our machine. Which will always suffice, for by +the time we get around to enormous tasks, we will have enormous +machines. + +Because each application runs its own protocol layer, it is simpler, though +not essential, for each application to have its own port number on its +network address and thus its own streams on that port number. All +protocol layers use a single operating system udp layer. All messages +coming from a single application in a single session are authenticated with +at least that session and application, or with an id durable between +sessions of the application, or with an id durable between the user using +different applications on the same machine, or with an id durable to the +user and used on different machines in different applications, though the +latter requires a fair bit of potentially hostile user interface. + +If the application wants to use multiple identities during a session, it +initiates a new connection on a new port number in the clear. One session, +one port number, at most one identity. Multiple port numbers, however, do +not need nor routinely have, multiple identities for the same run of the +application. + +[QUIC]: https://github.com/private-octopus/picoquic + +If we implement a [QUIC] large object layer, (and we really should not do +this until we have working code out there that runs without it) it will +consist of reliable request responses on top of groups of unreliable request +responses, in which case the unreliable request responses will have a +group request object that maps from their UDP origin and port numbers, +and a sequence number within that group request object that maps to an +item in an array in the group request operator. + +### speed + +The fastest authenticated encryption algorithm is OCB - and on high end +hardware, AES256OCB. + +AES256OCB, despite having a block cipher underneath, has properties +that make it possible to have the same API as xchacha20poly1305. +(Encrypts and authenticates arbitrary length, rather than block sized, messages.) + +[OCB patents were abandoned in February 2021](https://www.metzdowd.com/pipermail/cryptography/2021-February/036762.html) + +One of these days I will produce a fork of libsodium that supports ``crypto_box_ristretto25519aes256ocb.\*easy.\*`, but that is hardly urgent. +Just make sure the protocol negotiation allows new ciphers to be dropped in. + +# Getting something up and running + +I need to get a minimal system up that operates a database, does +encryption, has a gui, does unit test, and synchronizes data with other +system. + +So we will start with a self licking icecream: + +We aim for a system that has a per user database identifying public keys +related to user controlled secrets, and a local machine database relating +public keys to IP numbers and port numbers. A port and IP address +identifies a process, and a process may know the underlying secrets of +many public key. + +The gui, the user interface, will allow you to enter a secret so that it is hot and online, optionally allow you to make a subordinate wallet, a ready wallet. + +The system will be able to handle encryption, authentication, signatures, +and perfect forward secrecy. + +The system will be able to merge and floodfill the data relating public +keys to IP addresses. + +We will not at first implement capabilities equivalent to ready wallets, +subordinate wallets, and Domain Name Service. We will add that in once +we have flood fill working. + +Floodfill will be implemented on top of a Merkle-patricia tree +implemented with, perhaps, grotesque inefficiency by having nodes in the +database where the address of each node consists of the bit length of the +address as the primary sort key, then the address, and then the record, the +content of the node identified by this is hashes, the type, and the addresses +of the two children, and the hashes of the two children. The hash of the +node is the hash of the hashes of its two children, ignoring its address. +(The hash of the leaf nodes take account of the leaf node’s address, but the +hashes of the tree nodes do not) + +Initially we will get this working without network communication, merely +with copy paste communication. + +An event always consists of a bitstream, starting with a schema identifier. +The schema identifier might be followed by a shared secret identifiers, +which identifies the source and destination key, or followed by direct +identification of the source and destination key, plus stuff to set up a +shared secret. + +# Terminology + +Handle: +: A handle is short opaque identifier that corresponds to quite small +positive integer that points to an object in an `std::vector` containing a +sequence of identically sized objects. A handle is reused almost +immediately. When a handle is released, the storage it references goes +into a `std::priority_queue` for reuse, with handles at the start of the +of the vector being reused first. If the priority queue is empty, the +vector grows to provide space for another handle. The vector never +shrinks, though unused space at the end will eventually get paged out. A +handle is a member of a class with a static member that points to that +vector, and it has a member that provides a reference to an object in the +vector. It is faster than fully dynamic allocation and deallocation, but +still substantially slower than static or stack allocation. It provides +the advantages of shared pointer, with far lower cost. Copying or +destroying handles has no consequences, they are just integers, but +releasing a handle still has the problem that there may be other copies of +it hanging around, referencing the same underlying storage. Handles are +nonowning – they inherent from unsigned integers, they are just unsigned +integers plus some additional methods, and the static members +`std::vectortable;` and `std::priority_queuetable;>unused_handles;` + +Hashcode: +: A rather larger identifier that references an `std::unordered_map`, +which maps the hashcode to underlying storage, usually through a handle, +though it might map to an `std::unique_ptr`. Hashcodes are sparse, unlike +handles, and are reused infrequently, or never, so if your only reference +to the underlying storage is through a hashcode, you will not get +unintended re-use of the underlying storage, and if you do reference after +release, you get an error – the hashcode will complain it no longer maps +to a handle. Hashcodes are owning, and the hashmap has the semantics of +`unique_ptr` or `shared_ptr`. When an event is fired, it supplies a +hashcode that will be associated with the result of that fire and +constructs the object that hashcode will reference. When the response +happens, the object referenced by the hashcode is found, and the command +corresponding to the event type executed. In a procedural program, the +stack is the root data structure driving RAII, but in an event oriented +program, the stack gets unwound between events, so for data structures +that persist between events, but do not persist for the life of the +program, we need some other data structure driving RAII, and that data +structure is the database and the hashtables, the hashtables being the +database for stuff that is ephemeral, so we don’t want the overheads of +actually doing data to disk operations. + +Hot Wallet: +: The wallet secret is in memory, or the secret from which it is derived and chain of links by which that secret is derived is in memory. + +Cold Wallet, paper wallet: +: The wallet secret is not in memory nor on non volatile storage in a computer connected to the internet. High value that is intended to be kept for a long time should be controlled by a cold wallet. + +Online Wallet: +: Hot and online. Should usually be a subordinate wallet for your cold wallet. Your online subordinate wallet will commonly recieve value for your cold wallet, and will only itself control funds of moderate value. An online wallet should only be online in one machine in one place at any one time, but many online wallets can speak on behalf of one master wallet, possibly a cold wallet, and receive value for that wallet + +Ready Wallet: +: Hot and online, and when you startup, you don’t have to perform the difficult task of entering the secret because when it is not running, the secret is on disk. The wallet secret remains in non volatile storage when you switch off the computer, and therefore is potentially vulnerable to theft. It is automatically loaded into memory as the wallet, the identity, with which you communicate. + +Subordinate Wallet: +: Can generate public keys to receive value on behalf of another wallet, +but cannot generate the corresponding secret keys, while that other wallet, +perhaps currently offline, perhaps currently existing only in the form of a +cold wallet, that the other wallet has can generate the secret keys for. +Usually has an authorization lasting three months to speak in that other +wallet’s name, or until that other wallet issues a new authorization. A +wallet can receive value for any other wallet that has given it a secret and +authorization but only spend value for itself. + +# The problem + +Getting a client and a server to communicate is apt to be surprisingly complicated. This is because the basic network architecture for passing data around does not correspond to actual usage. + +TCP-IP assumes a small computer with little or no non volatile storage, and infinite streams, but actual usage is request-response, with the requests and responses going into non volatile storage. + +When a bitcoin wallet is synchronizing with fourteen other bitcoin wallets, there are a whole lot of requests and replies floating around all at the same time. We need a model based on events and message objects, rather than continuous streams of data. + +IP addresses and port numbers act as handles and hashcodes to get data from one process on one computer to another process on another computer, but within the process, in user address space, we need a representation that throws away the IP address, the port number, and the positional information and sequence within the TCP-IP streams, replacing it with information that models the process in ways that are more in line with actual usage. + +# Message objects and events + +Any time we fire an event, send a request, we create a local data structure identified by a handle and by the twofiftysix bit hashcode of the request, the pair of entities communicating. The response to the event references either the hashcode, or the handle, or both. Because handles are local, transient, live only in ram, and are not POD, handles never form part of the hash describing the message object, even though the reply to a request will contain the handle. + +We don’t store a conversation as between me and the other guy. Rather, we +store a conversation as between Ann and Bob, with the parties in lexicographic +order. When Ann sees the records on her computer, she knows she is Ann, when +Bob sees the conversation on his computer, he knows he is Bob, and Carol sees +the records, because they have been made public as part of a review, she +knows that Ann is reviewing Bob, but the records have the same form, and lead +to the same Merkle root, on everyone’s computer. + +Associated with each pair of communicating entities is a durable secret +elliptic point, formed from the wallet secrets of the parties communicating, +and a transient and frequently changing secret elliptic point. These secrets +never leave ram, and are erased from ram as soon as they cease to be +needed. A hash formed from the durable secret elliptic point is associated +with each record, and that hash goes into non volatile storage, where it is +unlikely to remain very secret for very long, and is associated with the +public keys, in lexicographic order, of the wallets communicating. The +encryption secret formed from the transient point hides the public key +associated with the durable point from eves droppers, but the public key +that is used to generate the secret point goes into nonvolatiles storage, +where it is unlikely to remain very secret for very long. + +This ensures that the guy handing out information gets information about who is interested in his information. It is a privacy leak, but we observe that sites that hand out free information on the internet go to great lengths to get this information, and if the protocol does not provide it, will engage in hacks to get it, such as Google Analytics, which hacks lead to massive privacy violation, and the accumulation of intrusive spying data in vast centralized databases. Most internet sites use Google Analytics, which downloads an enormous pile of JavaScript on your browser, which systematically probes your system for one thousand and one privacy holes and weaknesses and reports back to Google Analytics, which then shares some of their spy data with the site that surreptitiously downloaded their enormous pile of hostile spy attack code onto your computer. + +***[Block Google Analytics](./block_google_analytics.html)*** + +We can preserve some privacy on a client by the wallet initiating the connection deterministically generating a different derived wallet for each host that it wants to initate connection with, but if we want push, if we want peers that can be contacted by other peers, have to use the same wallet for all of them. + +A peer, or logged in, connection uses one wallet for all peers. A client connection without login, uses an unchanging, deterministically generated, probabilistically unique, wallet for each server. If the client has ever logged in, the peer records the association between the deterministically generated wallet, and wallet used for peer or logged in connections, so that if the client has ever logged in, that widely used wallet remains logged in forever -albeit the client can throw away that wallet, which is derived from his master secret, and use a new wallet with a different derivation from his master secret. + +The owner of a wallet has, in non volatile storage, the chain by which each wallet is derived from his master secret, and can regenerate all secrets from any link in that chain. His master secret may well be off line, on paper, while some the secrets corresponding to links in that chain are in non volatile storage, and therefore not very secure. If he wants to store a large amount of value, or final control of valuable names, he has them controlled by the secret of a cold wallet. + +When an encrypted message object enters user memory, it is associated with a handle to a shared transient volatile secret, and its decryption position in the decryption stream, and thus with a pair of communicating entities. How this association is made depends on the details of the network connection, on the messy complexities of IP and of TCP-IP position in the data stream, but once the association is made, we ignore that mess, and treat all encrypted message objects alike, regardless of how they arrived. + +Within a single TCP-IP connection, we have a message that says “subsequent +encrypted message objects will be associated with this shared secret and thus +this pair of communicating entities, with the encryption stream starting at +the following multiple of 4096 bytes, and subsequent encryption stream +positions for subsequent records are assumed to start at the next block of a +power of two bytes where the block is large enough to contain the entire +record.”, but on receiving records following that message, we associate it +with the shared secret and the encryption stream position, and pay no further +attention to IP numbers and position within the stream. Once the association +has been made, we don’t worry which TCP stream or UDP port number the record +came in on or its position within the stream. We identify the communicating +entities involved by their public keys, not their IP address. When we decrypt +the message, if it is a response to a request, it has the handle and/or the +hash of the request. + +A large record object could take quite a long time downloading. So when the +first part arrives, we decrypt the first part, to find the event handler, +and call the progress event of the handler, which may do nothing, every time +data arrives. This may cause the timeout on the handler to be reset. + +If we are sending a message object after long delay, we construct a new shared secret, so the response to a request may come over a new TCP connection, different from the one on which it was sent, with a new shared secret, and a position in the decryption stream, unrelated to the shared secret, the position in the decryption stream, and the IP stream, under which a request was sent. Our message object identity is unrelated to the underlying internet protocol transport. Its destination is a wallet, and its ID on the process of the wallet is its hashtag. + +# Handles + +I have above suggested various ad hoc measures for preventing references to reused handles, but a more robust and generic solution is hash codes. You generate fresh hash codes cyclicly, checking each fresh hash code to see if it is already in use, so that each communication referencing a new event handle or new shared secret also references a new hash code. The old hash code is de-allocated when the handle is re-used, so a new hashcode will reference the new entity pointed to by the handle, and the old hashcode fail immediately and explicitly. + +Make all hashcodes thirty two bits. That will suffice, and if scaling bites, we are going to have to go to multiple host processes anyway. Our planned protocol already allows you to be redirected to an arbitrary host wallet speaking on behalf of a master wallet that may well be in cold storage. When we have enormous peers on the internet hosting hundreds of millions of cients, they are going to have to run tens of thousands of processes. Our hashtags only have meaning within a single process and our wallet identifier address space is enormous. Further, a single process can have multiple wallets associated with it, and we could differentiate hashes by their target wallet. + +Every message object has a destination wallet, which is an online wallet, which should only be online in one host process in one machine, and an encrypted destination event hashcode. The fully general form of a message object has a source public key, a hashcode indicating a shared secret plus a decryption offset, or is prefixed by data to generate a shared secret and decryption offset, and, if a response to a previous event, an event hashcode that has meaning on the destination wallet. However, on the wire, when the object is travelling by IP protocol, some of these values are redundant, because defaults will have already been created associated with the IP connection. On the disk and inside the host process, it is kept in the clear, so does not have the associated encryption data. At the user process level, and in the database, we are not talking to IP addresses, but to wallets. The connection between a wallet and an IP address is only dealt with when we are telling the operating system to put message objects on the wire, or they are being delivered to a user process by the operating system from the wire. On the wire, having found the destination IP and port of the target wallet, the public key of the target wallet is not in the clear, and may be implicit in the port (dry). + +Any shared secret is associated with two hash codes, one being its value on the other machine, and two public keys. But under the dry principle, we don’t keep redundant data around, so the redundant data is virtual or implicit. + +# Very long lived events + +If the event handler refers to a very long lived event (maybe we are waiting for a client to download waiting message objects from his host, email style, and expect to get his response through our host, email style) it stores its associated pod data in the database, deletes it from the database when the event is completed, and if the program restarts, the program reloads it from the database with the original hashtag, but probably a new handle. Obviously database access would be an intolerable overhead in the normal case, where the event is received or timed out quickly. + +# Practical message size limits + +Even a shitty internet connection over a single TCP-IP connection can usually manage 0.3Mbps, 0,035Mps, and we try to avoid message objects larger than one hundred KB. If we want to communicate a very large data structure, we use a lot of one hundred KB objects, and if we are communicating the blockchain, we are probably communicating with a peer who has at least a 10Mbps connection, so use a lot of two MB message objects. + +1Mbps download, 0.3 Mbps upload, Third world cell phone connection, third world roach hotel connection, erratically usable.\ +2-4 Mbps Basic Email Web Surfing Video Not Recommended\ +4--6 Mbps Good Web Surfing Experience, Low Quality Video Streaming (720p)\ +6--10 Mbps Excellent Web Surfing, High Quality Video Streaming (1080p)\ +10-20 Mbps High Quality Video Streaming, High Speed Downloads / Business-Grade Speed + +A transaction involving a single individual and a single recipient will at +a minimum have one signature (which identifies one UTXO, rhocoin, making it +a TXO, hence $4*32$ bytes, two utxos, unused rocoins, hence $2*40$ bytes, and +a hash referencing the underlying contract, hence 32 bytes – say 256 bytes, +2048 bits. Likely to fit in a single datagram, and you can download six +thousand of them per second on a 12Mbs connection. + +On a third world cell phone connection, downloading a one hundred kilobyte object has high risk of failure, and busy TCP_IP connection has short life expectancy. + +For communication with client wallets, we aim that message objects received from a client should generally be smaller than 20KB, and records sent to a client wallet should generally be smaller than one hundred KB. For peer wallets and server wallets, generally smaller than 2MB. Note that bittorrent relies on 10KB message objects to communicate potentially enormous and complex data structures, and that the git protocol communicates short chunks of a few KB. Even when you are accessing a packed file over git, you access it in relatively small chunks, though when you access a git repository holding packed files over https protocol, you download the whole, potentially enormous, packed file as one potentially enormous object. But even with git over https, you have the alternative of packing it into a moderate number of moderately large packed files, so it looks as if there is a widespread allergy to very large message objects. Ten K is the sweet spot, big enough for context information overheads to be small, small enough for retries to be non disruptive, though with modern high bandwidth long fat pipes, big objects are less of a problem, and streamline communication overheads. + +# How many shared secrets, how often constructed + +The overhead to construct a shared secret is 256 bits and 1.25 milliseconds, so, on a ten Megabit per second connection, if the CPU spent half its time establishing shared secrets, it could establish one secret every three hundred microseconds, eg, one secret every three thousand bits. + +Since a minimal packet is already a couple of hundred bits, this does not give a huge amount of room for a DDoS attack. But it does give some room. We really should be seriously DDoS resistant, which implies that every single incoming packet needs to be quickly testable for validity, or cheap to respond to. A packet that requires the generation of a shared secret it not terribly expensive, but it is not cheap. + +So, we probably want to impose a cost on a client for setting up a shared +secret, And since the server could have a lot of clients, we want the cost +per server to be small, which means cost per client to be mighty small in +the legitimate non DDoS scenario – it only is going to bite in the DDoS +scenario. Suppose the server might have a hundred thousand clients, each +with sixteen kilobytes of connection data, for a total of about two +gigabyes of ram in use managing client connections. Well then, setting up +shared secrets for all those clients is going to take twelve and a half +seconds, which is quite a bit. So we want a shared secret, once set up, to +last for at least ten to twenty minutes or so. We don’t want clients +glibly setting up shared secrets at whim, particularly as this could be a +relatively high cost on the server for a relatively low cost on the +client, since the server has many clients, but the client does not have +many servers. + +We want shared secrets to be long lived enough that the cost in memory is +roughly comparable to the cost in time to set them up. A gigabyte of +shared secrets is probably around ten million shared secrets, so would +take three hours to set up. Therefore, we don’t need to worry about +throwing shared secrets away to save memory – it is far more important to +keep them around to save computation time. This implies a system where we +keep a pile of shared secrets, and the accompanying network addresses in +memory. Hashtable that hashes wallets existing in other processes, to +handles to shared secrets and network addresses of existing in this +process. So each process has the ability to speak to a lot of other +processes cached, and probably has some durable connections to a few other +processes. Which immediately makes us think about flood filling data +through the system without being vulnerable to spam. + +Setting up tcp connections and tearing them down is also costly, but it looks as though, for some reason, existing code can only handle a small number of tcp connections, so they encourage you to cotinually tear them down and recreate them. Maybe we should shut down a tcp connection after eighteen seconds of nonuse. Check them every multiple of 8 seconds past epoch, refrain from reuse twenth four seconds past the epoch, and shut them down altogether after thirty two seconds. (The reason for checking them at certain time since the epoch is that shutdown is apt to go more efficienty if initiated at both ends. + +Which means it would be intolerable to have a shared secret generation in +every UDP packet, or even very many UDP packets, so to prevent DDoS attack, +and just to have efficient communications, have to have a deal where you +cheaply for the server, but potentially expensively for the client, establish +a connection before you construct a shared secret. + +A five hundred and twelve bit hash however takes 1.5 microseconds – which +is cheap. We can use hashes to resist dos attacks, making the client +return to us the state cookie unchanged. If we have a ten megabit +connection, then every packet is roughly the size of a hash, in which case +the hash time is roughly three hundred megabits per second, not that +costly to hash everything. + +How big a hash code do we need to identify the shared secret? Suppose we generate one shared secret every millisecond microseconds. Then thirty two bit hashcodes are going to roll over in forty days. If we have a reasonable timeout on inactive shared secrets, reuse is never going to happen, and if it does happen, the connection fails, Well, connections are always failing for one reason or another, and a connection inappropriately failing is not likely to be very harmful, whereas a connection seemingly succeeding, while both sides make incorrect and different assumptions about it could be very harmful. + +# Message UDP protocol for messages that fit in a single packet + +When I look at [the existing TCP state machine] +(https://www.ietf.org/rfc/rfc0793.txt), it is hideously complicated. Why +am I thinking of reinventing that? [Cookies] +(http://cr.yp.to/syncookies.html) turn out to be less tricky than I +thought – the server just sends a secret short hash of the client data and +the server response, which the client cannot predict, and the client +response to the server response has to be consistent with that secret +short hash. + +Well, maybe it needs to be that complicated, but I feel it does not. If I find that it really does need to be that complicated, well, then I should not consider re-inventing the wheel. + +Every packet has the source port and the destination port, and in tcp initiation, the client chooses its source port at random (bind with port zero) in order to avoid session hijacking attacks. Range of source ports up to 65535 + +Notice that this gives us $2^{64}$ possible channels, and then on top of that we have the 32 bit sequence number. + +IP eats up twenty bytes, and then the source and destination ports eat four more bytes. I am guessing that NAT just looks at the port numbers and address of outgoing, and then if a packet comes in equivalent incoming, just cheerfully lets it through. TCP and UDP ports look rather similar, every packet has a specific server destination port, and a random client port. Random ports are sometimes restricted to 0xC000-0xFFFF, and sometimes mighty random (starting at 0x0800 and working upwards seems popular) But 0xC000-0xFFFF viewed as a hashcode seems ample scope. Bind for port 0 returns a random port that is not in use, use that as a hashcode. + +The three phase handshake is: + +1. Client: SYN my sequence number is X, my port number is random port A, and your port number is well known port B. +1. Server: ACK/SYN your sequence number is X, my Sequence number is Y, my Port number is well known B, and your port number is random port A. +1. Client: ACK your sequence number is Y, my sequence number is X+1, my port number is random port A, and your port number is well known port B. + +Sequence number is something like your event hashcode – or perhaps event +hashcode for grouped events, with the tcp header being the group. + +Assume the process somehow has an accessible and somehow known open UDP port. +Client low level code somehow can get hold of the process port and IP +address associated with the target elliptic point, by some mechanism we are +not thinking about yet. + +We don’t want the server to be wide open to starting any number of new +shared secrets. Shared secrets are costly enough that we want them to last +as long as cookies. But at the same time, recommended practice is that +ports in use do not last long at all. We also might well want a redirect to +another wallet in the same process on the same server, or a nearby process +on a nearby server. But if so, let us first set up a shared secret that is +associated with the shared secret on this port number, and then we can talk +about shared secrets associated with other port numbers. Life is simpler if +a one to one mapping between access ports and durable public and private keys, +even if behind that port are many durable public and private keys. + +# UDP protocol for potentially big objects + +The tcp protocol can be thought of as the tcp header, which appears in every packet of the stream, being a hashcode event object, and the sequence number, which is distinct and sequential in every packet of the unidirectional stream, being a std:dequeue event object, which fifo queue is associated with hashcode event object. + +This suggests that we handle a group of events, where we want to have an event that fires when all the members of the group have successfully fired, or one of them has unrecoverably failed, with the group being handled as one event by a hashcode event object, and the the members of the group with event objects associated with a fifo queue for the group. + +When a member of the group is triggered, it is added to the queue. When it is fired, it is marked as fired, and if it is the last element of the queue, it is removed from the queue, and if the next element is also marked as fired, that also is removed from the queue, until the last element of the queue is marked as triggered but not yet fired. +In the common case where we have a very large number of members, which are fired in the order, or approximately the order, that they are triggered, this is efficient. When the group event is marked as all elements triggered and all elements fired, and the fifo queue empty then that fires the group event. + +Well, that would be the efficient way to handle things if we were implementing TCP, a potentially infinite stream, all over again, but we are not. + +Rather, we are representing a big object as a stream of objects, and we know the size in advance, so might as well have an array that remains fixed size for the entire lifetime of the group event. The member event identifiers are indexes into this one big fixed size array. + +The event identifier is run time detected as a group event identifier, so it expects its event identifier to be followed by an index into the array, much as the sequence number immediately follows the TCP header. + +I would kind of like to have a [QUIC] protocol eventually, but that can +wait.If we have a UDP protocol, the communicating parties will negotiate a +UDP port that uniquely identifies the processes on both computers. Associated +with this UDP port will be the default public keys and the hash of the +shared secret derived from those public keys, and a default decryption shared +secret. The connection will have a keep alive heartbeat of small packets, +and a data flow of standard sized large packets, each the same size. Each +message will have a sequence number identifying the message, and each UDP +packet of the message will have the sender sequence number of its message, +its position within the message, and, redundantly, the power of two size of +the encrypted message object. Each message object, but not each packet +containing a fragment of the message object, contains the unencrypted hashtag +of the shared secret, the hashtag of the event object of the sender, which +may be null if it is the final message, and, if it is a reply, the hashtag of +event object of the message to which it is a reply, and the position within +the decryption stream as a multiple of the power of two size of the encrypted +message. This data gets converted back into standard message format when it +is taken off the UDP stream. + +Every data packet has a sequence number, and each one gets an ack, though only when the input queue is empty, so several data packets get a group ack. If an ack is not received, the sender sends a nack. If the sender responds with a nack (huh, what packets?) resends the packets. If the sender persistently fails to respond, sending the message object failed, and the connection is shut down. If the sender can respond to nacks, but not to data packets, maybe our data packet size is too big, so we halve it. If that does not work, sending the message object failed, and the connection is shut down. + +[QUIC] streams will be created and shut down fairly often, each time with a new shared secret, and message object reply may well arrive on a new stream distinct from the stream on which it was sent. + +Message objects, other than nacks and acks, intended to manage the UDP stream +are treated like any other message object, passed up to the message layer, +except that their result gets sent back down to the code managing the UDP +stream. A UDP stream is initiated by a regular message object, with its own +data to initiate a shared secret, small enough to fit in a single UDP packet, +it is just that this message object says “prepare the way for bigger message +objects” – the UDP protocol for big message objects is built on top of a UDP +protocol for message objects small enough to fit in a single packet. diff --git a/docs/code.ico b/docs/code.ico new file mode 100644 index 0000000..a049a92 Binary files /dev/null and b/docs/code.ico differ diff --git a/docs/contracts_on_blockchain.md b/docs/contracts_on_blockchain.md new file mode 100644 index 0000000..821e35d --- /dev/null +++ b/docs/contracts_on_blockchain.md @@ -0,0 +1,334 @@ +--- +title: + Contracts on the blockchain +--- +# Terminology + +A rhocoin is an unspent transaction output, and it is the public key +that identifies that unspent transaction output, the private key that +can sign with that public key, the point which is the mathematical +object of which that public key is the text representation, and the +scalar which is the mathematical object that the secret can construct. + +A public key and and its associated secret key can do all sorts of +things as well as control rocoins – logons, identifying participants in +a conversation, and signing a message, among them. + +We talk about points and scalars, meaning points on the elliptic curve +and scalars, large numbers modulo the order of the curve, the +mathematical objects underlying these keys, when we combine them +mathematically in interesting ways, for example adding several points to +create a public key that requires several secrets, possessed by several +different people, to use. Scalars can be added and multiplied, points +can be added or subtracted from other points, and points can be +multiplied by scalars. When we design contracts on the blockchain, we +should refer to scalars and points, but the rest of the time, we should +talk about coins, public keys, and private keys. Normal people should +never need to hear of points and scalars, and should not need to know +what they are. Only people writing software to manage blockchain based +interactions need to talk about or think about points and scalars. + +# Instant irrevocable transctions + +## Why we need Schnorr signatures + +In order that people can meet in person to exchange fiat for blockchain +money in person, they need a transfer that is instant and irrevocable. + +Bob wants to buy blockchain money for cash. Ann wants to sell for cash. +They agree to meet in person, and do the exchange – paper money in a +brown paper bag. + +Ann and Bob cooperate over the network to create a postdated transaction +spending an as yet nonexistent coin whose public key is the sum of a +public key whose secret key is known only to Bob, and a public key whose +secret key is known only to Ann, and *after* that transaction is +created, Ann issues a transaction placing value an unspent transaction +output in that coin, creates a rhocoin for that public key, knowing that +if nothing further is done, the coin eventually comes back to her. Then, +before the postdated transaction becomes placeable on the blockchain, +they meet in person, and once the cash is in her hands and she has +counted it, she then reveals her secret key to Bob, and his wallet can +instantly tell him that the coin is in his wallet right and spendable +right now – but will become spendable by Ann at block number so and so. +Bob can now spend that coin, but Ann still cannot spend it and needs to +spend it before the postdated transaction becomes spendable on the +blockchain. He presumably spends it to a coin wholly controlled by him, +or uses it in some other transaction, and Ann discards the now useless +postdated transaction. + +Note that such a joint shnorr signature is absolutely indistinguishable +on the blockchain from any other signature, so no one, except for Ann +and Bob, can tell there was anything different about this transaction. + +To represent this contract to normies, we call it a coin commitment, a +coin locked to someone else’s wallet for a certain period. We refain +from talking about points and scalars. Ann creates a coin that she +cannot use for anything except paying Bob, until a certain time has +passed, and once this coin is registered on the blockchain, Bob knows +she has created this coin and what it is worth, (though no one except +Ann and Bob knows it is for this purpose) but once the coin is on the +blockchain, she can use it to instantly pay Bob, in a message over the +network that his wallet will immediately register as payment completed, +without needing a third party escrow agent that governments will want to +register and regulate, as you have use in order to do instant payments +with bitcoin, or offline by Bob manually typing in the secret that makes +the coin spendable by Bob, though it can be texted in the clear, since +no one but Bob can make use of it. It only needs to be kept secret until +the time comes for Bob to receive the payment. But if she does not +instantly pay Bob, she can convert it into a regular coin, spendable by +her and only by her at any time, once its lockup time has expired. +conversely, once she has given Bob the secret, Bob can use it any +transaction, such as a transaction spending it to himself, before its +time is up. The blockchain contains no information showing this +committed coin is any different from any other, the information that +makes this coin different being in the secrets in Ann’s and Bob’s +wallets, not in the blockchain where governments are likely to look for +someone to regulate. The information that makes this coin different is +that the secret that controls it is split between Ann’s wallet and +Bob’s wallet, and Ann and Bob have already created a secret +transaction, stored in Ann’s wallet, spending the coin to a destination +determined by Ann, which transaction remains in her wallet until the +time is up, or until Bob, having received the secret from Ann, spends +the coin. This transaction is made possible, not by any terribly clever +cryptographic mathematics, but by the fact that our blockchain, unlike +the others, is organized around client wallets chatting privately with +other client wallets. Every other blockchain has necessary cryptographic +mathematics to do the equivalent, usually more powerfull and general +than anything on the rhocoin blockchain, and Monaro has immensely +superior cryptographic capabilities, but somehow, they don’t, the +difference being that rhocoin is designed to avoid uses of the internet +that render a blockchain vulnerable to regulation, rather than to use +clever cryptography to avoid regulation. The attack points whereby +government is getting hold of crypto currencies are not the +cryptography, which is usually bulletproof, but the domain name system, +the certificate authorities, and the world wide web, which is completely +vulnerable. + +# General purpose scripts + +Ethereum has a general purpose script language, which seems like +overkill, and indeed it was overkill, since humans started to misbehave, +and other humans started to judge transactions, so that the laws of +mathematics failed to dictate the outcomes, and those human judges are +predictably misbehaving. + +Bitcoin has a also has a fully general stackbased language +pay-to-scripthash (P2SH) also called Bitcoin script.  But this led to +problems, so they effectively disabled it, by making only a small set of +scripts permissible.  Seems that we have not yet figured out how to +safely enable run time user designed contracts on the blockchain.  We +instead need a short list of payment rules fixed at compile time.  We +shall not create a script language capable of embedding arbitrary +contracts in the blockchain at runtime, as it would be impossible for one of +the parties to figure out the gotchas cooked up by the other party. + +Rather than our model being the infamous click through contract and shrink +wrap contract, our model should be the returnable writs of The Lion of +Justice, Henry the First. The Anglo Saxon legal system has been going +downhill since the death of Henry the second. It is time for a restoration. +If we cannot restore, bypass. People wanted to use the legal system of The +Lion of Justice, rather than the ecclesiastical legal system, and if we do +things right, people will want to use our legal system. + +A returnable writ is a royal command that has blanks that the parties to a +dispute or a contract can fill in, but they cannot write their own writ +ad hoc. + +The trouble with excessive flexibility is that the parties to a dispute are +likely to have asymmetric knowledge of the implications of the contract, which +problem can be mitigated by having as few possible contracts as possible, and +those contracts authorized by the consensus of the blockchain. We can +increase flexibility by having multi transaction transactions, where different +elements of the aggregate transaction invoke different writs, but too much +flexibility is likely to bite people. + +# Atomic Swaps on separate blockchains + +A proof of stake currency is like a corporation, like shares in a +corporation.  So we are going to have many corporations, and individuals +will want to exchange shares in one corporation, with shares in +another.  We would like to do this without direct linking of +blockchains, without trusted intermediaries, because a trusted +intermediary is a target for regulation, where the state imposes +mandatory trust, while protecting profoundly untrustworthy behavior. + +Bob is owns some crypto currency in the foo blockchain, and wants to +exchange it with Carol’s crypto carrency in the bar blockchain.  + +Bob agrees with Carol to give her three units of his foo currency, for +five units of her bar currency.  + +But, how do we make it so that the transaction is completed? We don’t +want Carol giving five units, and Bob replying “Hah hah fooled you”. + +So Carol creates an output of five units that can be spent by a secret +key that only she knows after ten blocks, but until then can be spent by +Bob’s secret, plus a preimage that only she knows. The output contains +Bob’s public key, Carol’s public key, and the public hash of a secret +preimage known only to Carol. Bob waits until that output becomes final +and spendable in the bar blockchain. Bob then creates an output of three +units that can be spent by a secret key that Bob knows after five blocks +in the foo chain, but until then, can be spent by a carol’s secret key, +plus the preimage of a value known only to Carol. The output also +contains Carol’s public key, Bob’s public key, and the pubic hash of a +secret preimage known only to Carol, except that the public keys in +Bob’s output are in the opposite order to Carol’s, and the times are +different.  After a while, both outputs become final and spendable in +their respective blockchains.  Carol then uses her preimage to spend +those three units in the foo chain, thereby automatically revealing it +to Bob, which preimage Bob immediately employs to spend the output on +the bar chain. + +To spend an output, it has to be an input, one of many, to a +transaction, and the whole transaction has to be signed by every +signature required by every input, as each input defines a valid +signature. So an input/output carries information amounting to a +quantity of money, possibly a user name, and a signature definition. In +the simplest and most common case, a public key defines what would +constitute a valid signature. + +The immutability of the output comes from the fact that is part of +transaction, and the entire transaction has to be as signed, that the +transaction has to be signed by the signatures for all the inputs. For +this case, contract conditional on pre-image for a certain range of +block numbers, the signature block has to the pre-image as well as the +regular signature. + +# Micro and Nanopayments. + +The blockchain is heavy, thus unsuitable for small and fast payments, +such as the payments needed for storage and bandwidth while pirating +files. + +Small fast payments require trust, thus trusted intermediaries backed up +by the blockchain. + +How do we know if we can trust an intermediary? + +Because he has a history of signed contracts, that it is easy to prove +whether a contract has been honored, and no one has produced a contract +that he signed, alleged that contract was dishonored, and he could not +prove it was honored. + +Assume Bob is a trusted intermediary between Ann and Carol. Ann wants to +pay people, Carol among them, probabilistically – in lottery tickets +that if won will result in payments on the blockchain. + +The protocol is that Ann creates the unspent transaction output, which +comes to exist, unspent, on the blockchain, and no one can spend it +except Bob says so, since any transaction spending that output will need +a Bob signature.  So if Bob is an OK guy, no double spending shall ever +happen. If no double spending has ever happened, we can probably trust +Bob for transactions similar to past transactions. If no past double +spends, likely no future double spends. + +Ann promises to give Carol a lottery ticket for services rendered, Carol +gives Ann a signed hash of a secret preimage, and renders those +services. Ann issues a lottery ticket by creating a random number, and +giving Carol a signed hash of Carol’s hash and the lottery identifier. +The ticket will be valid if the hash of Ann’s secret preimage, and +Carol’s secret preimage has certain properties – typically that its +value in big endian order modulo 2^64^ is less than a certain amount.  +The ticket commits Ann to the lottery conditions, being a hash of their +secrets, and the conditions agreed to. + +Carol shows Bob the lottery ticket, and asks + +> “will I get paid if the secrets under the hashes meet the required +> condition.” + +Bob tells Carol + +> “Sure. I will issue a payment for block number such of the blockchain, +> the current block, if the preimage meets the conditions.” + +thereby assuring Carol that Ann is on the up and up and providing the +data needed to create a valid transaction if the lottery ticket is +valid.  Bob provides a link to the transaction output that will be used +showing that Carol has put real money where her mouth is, asserts it is +unspent, and promises not to validate any further potentially winning +lottery tickets for this block period, unless shown evidence that +previously validated lottery tickets were losing tickets. + +Carol, who has some reason to trust Bob, now knows that Ann is issuing +valid lottery tickets – even if this was a losing lottery ticket.  Bob +will refuse to issue more validations for this lottery and this block +period. unless Carol provides proof that this was a losing lottery +ticket. + +If Carol goes dark at this point, the money is frozen till the next +block period, which causes minor and temporary inconvenience for Ann and +Bob but does not benefit Carol. + +Suppose it is a winning lottery ticket – she informs Bob and Carol so +that they know to issue a new lottery, and now injects it into the +blockchain with the required data, and hashes that links to the +conversations between herself, Bob, and Alice. If the payment goes +through – the transaction inputs are real and have not been previously +spent.  then done. If the transaction fails for block n of the block +chain, when Bob said it would succeed, she uses the proof of failure - +that blockchain input was invalid or spent in this block when Bob said +it would be valid and unspent for this block, plus the previous +conversations between herself, Bob, and Carol, to prove to the world +that Bob was cheating. + +If, on the other hand, there are lots of transactions of this form that +have successfully gone through, and none that failed in this fashion, +these provide compelling evidence that Bob is an honest dealer. + +If Bob goes dark at some point (going dark means ceasing to respond, or +issuing responses that will be rejected and ignored as invalid) the +money remains unspent and unspendable, Ann and Carol are inconvenienced, +no one wins, and no proof of bad behavior is generated.  But Bob cannot +stay up for losing lottery tickets, but then conveniently go dark for +winning lottery tickets because he issues the data required for a +winning lottery ticket (or data that would prove he is a cheater) before +he knows whether the ticket is winning or not. + +If Carol goes dark, she does not get paid, but does not get proof of bad +behavior by Ann or Bob. + +If Ann goes dark, she does not pay Carol. Carol knows Ann is behaving +badly, ceases to deal with her, but may not get proof of bad behavior by +Ann that she can show to other people.  But getting such proof is not +very useful anyway, since Carol’s identity is cheap and expendable, +while Bob’s identity is durable and valuable. + +Random stranger Ann contacts random stranger Carol, offers to pay in +lottery tickets for data. Perhaps she wants to watch some movies. Shows +proof that lottery prize recently existed in a recent block on the +blockchain, which shows that Ann is not a total flake. Gets some data on +trust. Provides lottery ticket. Bob says lottery ticket could win, +depending on a secret that Carol has committed to, but not revealed. +Carol checks that in the past Bob has paid out on lots of lottery +tickets and not defected on any lottery ticket. More trust ensues.  Ann +now has reason to trust Carol, Carol has reason to trust Ann, without +anyone being exposed to large risks and without any actual blockchain +transactions taking place. It takes an expensive blockchain transaction +to create the lottery prize, but having created it, Ann and Bob can do +very large numbers of transactions off blockchain on trust. + +# Institutions on the blockchain + +A lot of people, instead of controlling outputs on the bitcoin +blockchain by secret keys, have a login account, username and password, +with an “exchange”.  That “exchange” (the institution, not the +particular transaction mediated by the institution) owes them bitcoins, +payable on demand.  And from time to time an exchange goes belly up, +blaming the federal government or hackers, and just does not pay its +clients the bitcoins it owes. + +We call them “exchanges”, not “banks”, because the word “banks” implies +a government guarantee, and a long past of honoring transactions, which +these exchanges seldom have, but they are performing bank like +functions. + +We would like to have evidence of the institutions reserve fraction, of +its term transformation (maturity transformation). + +Solution: Institution runs its own side chain, with a small number of +peers. The Merkle-patricia dac of unspent transaction outputs has in +each node the sum of the money in each subtree, and the hash of subtree +hashes and sums. Thus the standard client wallet will report the total +owing/shares on issue diff --git a/docs/contributor_code_of_conduct.md b/docs/contributor_code_of_conduct.md new file mode 100644 index 0000000..aa044a0 --- /dev/null +++ b/docs/contributor_code_of_conduct.md @@ -0,0 +1,154 @@ +--- +title: Contributor Code of Conduct +--- + +# Peace on Earth to all men of good will + +May you do good and not evil. May you find forgiveness for yourself and +forgive others. May you share freely, never taking more than you give. + +# Operational Security + +A huge problem with software that relates to privacy and/or to money is +that frequently strange and overcomplicated design decisions are made, +(passive tense because it is strangely difficult to find who made those +decisions), decisions whose only apparent utility is to provide paths for +hostile organizations to exploit subtle, complex, and unobvious security holes. + +These holes are often designed so that they can only be utilized efficiently +by a huge organization with a huge datacentre that collects enormous +numbers of hashes and enormous amounts of data, and checks enormous +numbers of hashes against an even more enormous number of potential +pre-images generated from that data. + +Another huge problem is that if we get penetrated by enemy shills, +entryists, and buggers, as the Patriot Front is and the Jan Sixth protestors +were, we are likely to wind up like the January sixth protestors, who as I +write this are imprisoned indefinitely being tortured by black guards +recently imported from the northern part of black Africa, awaiting +trial with no likelihood of any actual trial for years. + +## No namefags + +A participant who can be targeted is likely to introduce unobvious security +flaws into the software architecture. All contributors should make some +effort to protect themselves against a third party subsequently coercing +them to use the reputation that they have obtained by contributing to make +subsequent harmful contributions. + +All contributors will use a unique name and avatar for the purpose of +contributing to this project, and shall not link it to other names of theirs +that are potentially subject to pressure. In the event of videoconferencing, +the participants shall wear a mask over the lower part of their face that +conceals the shape of their mouth and jaw and a rigid hat like a fedora that +conceals the shape of the upper part their head. + +Apart from your mouth, the parts of your face that communicate non +verbal information turn out to be surprisingly useless for identifying +individuals. + +If you wear glasses, should not wear your usual glasses, because facial +recognition software is very good at recognizing glasses, and easily +distracted, confused, and thrown off by unusual glasses. + +Even if there are gaping holes in our security, which there will be, and +even if everyone knows another name of a participant, which they will, no +need to make the hole even bigger by mentioning it in public. People who lack +security are likely to result in code that lacks security. They come under +pressure to introduce an odd architecture for inexplicable reasons. We see +this happening all the time in cryptographic products. + +# Code will be cryptographically signed + +Of necessity, we will rest our developer identities on GPG keys, until we +can eat our own dogfood and use our own system's cryptographic keys. +Login identities shall have no password reset, because that is a security +hole. If people forget their password, they should just create a new login +that uses the same GPG key. + +# No race, sex, religion, nationality, or sexual preference + +![On the internet nobody knows you are a dog](./images/nobody_know_you_are_a_dog.webp) + +Everyone shall be white, male, heterosexual, and vaguely Christian, even +if they quite obviously are not, but no one shall unnecessarily and +irrelevantly reveal their actual race, sex, religion, or political orientation. + +All faiths shall be referred to respectfully. Even if they happen to be +making war on us, this software may not be very relevant to that kind of +warfare, in which case that discussion can be held elsewhere. + +All sovereigns shall be referred to respectfully, if they are referred to at all, +which they should not be. If this software is likely to frustrate their +objectives, or even contribute to their overthrow, no need to make it +personal, no need to trigger our enemies. War will come to us soon +enough, no need to go looking for it. + +# No preaching supererogation + +Status must be on the basis of code, good code, and clever code, not on +cheap claims of superior virtue. + +When someone plays the holier than thou card, he does not intend to share +what we are sharing. Out of envy and covetousness, he intends to deny us +what we are sharing, to deny us that which is ours. + +If he is holier than we are, he not only wants what we have, which we will +gladly share. He wants us to not have what we have. + +Christians are required to turn the other cheek, and people attempting to +maintain a politically neutral environment need to turn the other cheek. +But you very quickly run out of cheeks, and then it is on. You cannot be +politically neutral when the other guy is not being neutral. You have to +bring a gun to a gunfight and a faith to a holy war. People who start +politics in an environment intended to be politically neutral have to be +purged, and a purge cannot be conducted in a politically neutral manner. +You have to target the enemy faith and purge it as the demon worshiping +heresy that it is, or else those attempting to maintain political neutrality +will themselves be purged as heretics, as happened to the Open Source and +Free Software movements. You may not be interested in war, but war is +interested in you. + +We want to maintain a politically, racially, religiously, and ethnically +neutral environment, but it takes two to tango. You cannot maintain a +politically neutral environment in a space where an enemy faction wants +their politics to rule. Neutrality cannot actually be neutral. It merely means +that the quietly ruling faction is quiet, tolerant of its opponents, and does +not demand affirmations of faith. If an enemy faith wants to take over, +the ruling faith can no longer be quiet and tolerant of that opponent. + +## No claims of doing good to random unknown beneficiaries + +We are doing this for ourselves, our friends, our kin, and our posterity, not +for strangers a thousand miles away, and we only care about strangers a +thousand miles away to the extent that they are likely to enable us to make +money by making them secure. + +If someone mentions the troubles of people a thousand miles away, it +should only be in the frame that we will likely have similar troubles soon +enough, or that those people a thousand miles away, of a different race, +religion, and language, could use our product to their, and our, mutual +advantage, not because he cares deeply for the welfare of far away +strangers that he has never met in places he could not find on a map. + +## No victim classes, no identity politics, and no globalism + +The Open Source and Free Software movements were destroyed by +official victimhood. Status and leadership must be on the basis of code, +good code, and clever code, not on cheap claims of superior oppression. + +The experience of the Open Source and Free Software movement +demonstrates that if victimhood is high status, code and code quality must +be low status. If victimhood is high status then “you did not build that”. +Rather, if victimhood is high status, then good code, silicon fabs, and +rockets spontaneously emerged from the fertile soil of sub-Saharan Africa, +and was stolen by white male rapists from the brave and stunning black +warrior women of sub-Saharan Africa, and social justice demands that the +courageous advocate for the brave and stunning black warrior women of +sub-Saharan Africa takes what you have, what you gladly would share, +away from you. + +Unless, when a female contributor unnecessarily and irrelevantly informs +everyone she is female, she is told that she is seeking special treatment on +account of sex, and is not going to get it, no organization or group that +attempts to develop software is going to survive. Linux is a dead man walking. diff --git a/docs/crypto_currency.md b/docs/crypto_currency.md new file mode 100644 index 0000000..88da764 --- /dev/null +++ b/docs/crypto_currency.md @@ -0,0 +1,368 @@ +--- +title: Crypto currency +--- + +The objective is to implement the blockchain in a way that scales to one hundred thousand transactions per second, so that it can replace the dollar, while being less centralized than bitcoin currently is, though not as decentralized as purists would like, and preserving privacy better than bitcoin now does, though not as well as Monaro does. It is a bitcoin with minor fixes to privacy and centralization, major fixes to client host trust, and major fixes to scaling. + +The problem of bitcoin clients getting scammed by bitcoin peers will be fixed through Merkle-patricia, which is a a well known and already widely deployed fix – though people keep getting scammed due to lack of a planned bitcoin client-host architecture. Bitcoin was never designed to be client host, but it just tends to happen, usually in a way that quite unnecessarily violates privacy, client control, and client safety. + +Monaro’s brilliant and ingenious cryptography makes scaling harder, and all mining based blockchains tend to the same centralization problem as afflicts bitcoin. Getting decisions quickly about a big pile of data necessarily involves a fair bit of centralization, but the Paxos proof of stake protocol means the center can move at the speed of light in fiber, and from time to time, will do so, sometimes to locations unknown and not easy to find. We cannot avoid having a center, but we can make the center ephemeral, and we can make it so that not everyone, or even all peers, know the network address of the processes holding the secrets that signed the most recent block. + +Scaling accomplished by a client host hierarchy, where each host has many clients, and each host is a blockchain peer. + +A hundred or so big peers, who do not trust each other, each manage a copy of the blockchain. + +The latest block is signed by peers representing a majority of the stake, which is likely to be considerably less than a hundred or so peers. + +Peer stake is delegated from clients – probably a small minority of big clients – not all clients will delegate. Delegation makes privacy more complicated and leakier. Delegations will be infrequent – you can delegate the stake held by an offline cold wallet, whose secret lives in pencil on paper in a cardboard file in a safe, but a peer to which the stake was delegated has to have its secret on line. + +Each peer’s copy of the blockchain is managed, within a rack on the premises of a peer, by a hundred or so shards. The shards trust each other, but that trust does not extend outside the rack, which is probably in a room with a lock on the door and a security camera watching the rack. + +Most people transacting on the blockchain are clients of a peer. The blockchain is in the form of a sharded Merkle-patricia tree, hence the clients do not have to trust their host – they can verify any small fact about the blockchain in that they can verify that peers reflecting a majority of stake assert that so and so is true, and each client can verify that the peers have not rewritten the past. + +Scale is achieved through the client peer hierarchy, and, within each peer, by sharding the blockchain. + +Clients verify those transactions that concern them, but cannot verify that all transactions are valid, because the blockchain is too big. Each peer verifies the entire blockchain from beginning to end. If the blockchain replaces the US dollar as the world currency, then it will rapidly become far too large for any one computer to verify the whole thing, so will have to be verified by a group of mutually trusting and trusted shards, but each such group of shards is a peer. The shards trust shards of the same peer, which are likely running on the same rack in the same locked room under the gaze of the same security camera, but they don’t trust shards of some other peer. + +In each transaction, each client verifies that the other client is seeing the same history and recent state of the blockchain, and in this sense, the blockchain is a consensus of all clients, albeit that consensus is mediated through a small number of large entities that have a lot of power. + +The architecture of power is rather like a corporation, with stake as shares. +In a corporation CEO can do anything, except the board can fire him and +choose a new CEO at any time. The shareholders could in theory fire the +board at any time, but in practice, if less than happy with the board, have +to act by transacting through a small number of big shareholders. +Centralization is inevitable, but in practice, by and large corporations do +an adequate job of pursuing shareholder interests, and when they fail to do +so, as with woke capital, Star Wars, or the great minority mortgage +meltdown, it is usually due to heavy handed state intervention. Google’s +board is mighty woke, but in the Damore affair, human resources decided +that they were not woke enough, and in the Soy wars debacle, the board +was not woke at all but gave power over Star Wars brand name to wome +who threatened them with \#metoo. And if this form of distributed power +does not always work all that well, it fails less badly than anything else we +have tried. Delegated power representing assets, rather than people, results +in centralized power that, by and large, mostly, pursues the interests of +those assets. Delegated power representing people, not so much. + +In bitcoin, power is in the hands of a very small number of very large miners. This is a problem, both in concentration of power, which seems difficult to avoid if making decisions rapidly about very large amounts of data, and in that miner interests differ from stakeholder interests. Miners consume very large amounts of power, so have fixed locations vulnerable to state power. They have generally relocated to places outside the US hegemony, into the Chinese or Russian hegemonies, or the periphery of those hegemonies, but this is not a whole lot of security. + +Proof of stake has the advantage that stake is ultimately knowledge of secret keys, and while the state could find the peers representing a majority of stake, they are more mobile than miners, and the state cannot easily find the clients that have delegated stake to one peer, and could easily delegate it to a different peer, the underlying secret likely being offline on pencil and paper in someone’s safe, and hard to figure out whose safe. + +Obviously, at full scale we are always going to have immensely more clients than full peers, likely by a factor of hundreds of thousands, but we need to have enough peers, which means we need to reward peers for being peers, for providing the service of storing blockchain data, propagating transactions, verifying the blockchain, and making the data readily available, rather than for the current pointless bit crunching and waste of electricity employed by current mining. + +Bitcoin proposes to solve the scaling problem by the [Lightning Network, which is a re-invention of correspondent banking and the General Ledger, SubLedger system](https://www.forbes.com/sites/francescoppola/2016/06/17/thunder-and-lightning-in-the-bitcoin-world/). Obviously re-inventing General Ledger and Subledger will improve scaling, but [Central Clearing houses are also needed](https://gendal.me/2013/11/24/a-simple-explanation-of-how-money-moves-around-the-banking-system/).  + +The power over the blockchain, and the revenues coming from transaction and storage fees, have to go to this large number of peers, rather than, as at present, mostly to four miners located in China. + +Also, at scale, we are going to have to shard, so that a peer is actually a pool of machines, each with a shard of the blockchain, perhaps with all the machines run by one person, perhaps run by a group of people who trust each other, each of whom runs one machine managing one shard of the blockchain. + +Rewards, and the decision as to which chain is final, has to go to weight of stake, but also to proof of service – to peers, who store and check the blockchain and make it available. For the two to be connected, the peers have to get stake delegated to them by providing services to clients. + +All durable keys should live in client wallets, because they can be secured off the internet.  So how do we implement weight of stake, since only peers are sufficiently well connected to actually participate in governance? + +To solve this problem, stakes are held by client wallets.  Stakes that are in the clear get registered with a peer, the registration gets recorded in the blockchain, and the peer gets influence, and to some +extent rewards, proportional to the stake registered with it, conditional on the part it is doing to supply data storage, verification, and bandwidth. + +My original plan was to produce a better bitcoin from pair based +cryptography.  But pair based cryptography is slow.  Peers would need a +blade of computers when the volume surpassed bitcoin levels. + +Maybe not so slow.  [There is an assembly library](https://github.com/herumi/mcl) that promises three ops per millisecond + +So instead, swipe, I mean build upon, the cryptonote foundation.  (Which already implements the split between network node and wallet.) Two substantial currencies have been built from cryptonote: Monero and bytecoin.  Also Boolberry. +But, on the other hand [MimbleWimble clearly has the best cryptography – at the bleeding edge](https://github.com/ignopeverell/grin/blob/master/doc/grin4bitcoiners.md). + +> no address. All outputs in Grin are unique and have +> no common data with any previous output. Instead of relying on a +> known address to send money, transactions have to be built interactively, +> with 2 (or more) wallets exchanging data with one +> another. Practically, this isn’t so much of a problem as there +> are multiple ways for 2 programs to interact privately and securely. +>  And this interaction could even take place over email or Signal +> (or carrier pigeons). + +For example, suppose each peer has a thousand client wallets, and the capacity to connect to any other peer, that peers have fully accessible ports, and that the client wallets, who being behind consumer grade NATS generally do not have fully accessible ports, set up a direct client wallet encrypted connection through their NATS using their peer connections to initialize the connection. + +But obviously this software is not written yet.  Still vaporware, but vaporware that sounds very promising. + +Mimblewimble solves the problem of disk storage limiting scale. + +How does it go on bandwidth limiting scale? + +On bandwidth, it kind of sucks.  We are going to need shardable peers. + +We need a client peer host architecture that is future compatible with people who have serious money using a special purpose microcomputer with an lcd touchscreen, like an android but incapable of being reprogrammed, because it runs code in rom, and whose only essential functions are: Enter password, copy wallet from one memory card to another, show you what you are signing, and allow you to sign it.  Or perhaps a walled garden computer incapable of running any code except code signed by the builder, (except your adversary has physically got at it and replaced it by an evil twin) but otherwise a full internet capable androidish device. From which it follows that not only our host, but our client needs to be accessible through socket io. + +Bitcoin can do about 3 transactions per second That’s a far cry from the 2000 TPS that Visa rams through every second of every day. + +Bitcoin takes at least ten minutes to confirm your transaction. + +Inside the computer, transaction amounts will be represented as big +integers with a fixed precision limit, initially sixty four bits. On the +blockchain, in the canonical representation, they will be represented as +arbitrary precision integers times one thousand raised to a signed arbitrary +precision quantity, which for a very long time will be a one byte quantity. +The initial smallest representable unit, corresponding to the internal +representation inside the computer, $1µρ$, will be represented on the +blockchain as $1*1000^{96}$, so that we do not have to think about\ +whether that byte is +signed or unsigned. If, after millennia of deflation, which I think and hope +likely, it approaches zero, we will have to start thinking of it as a signed +quantity, and if, after millennia of inflation, which I hope is far less +likely, it approaches 128, we will start thinking of it as unsigned quantity. + +If rhocoin takes over the world, and the smallest unit is initially worth ten +trillion dollars 2^-64^economic growth and various engineered and +inadvertent currency leaks will result in slow deflation. If it deflates at +two percent a year, then in six hundred years of so, there is going to be a +problem with the smallest currency unit becoming too large. I would like my +works to outlast those of Ozymandias. But by that time the equivalent of banks +will have appeared, and banks can issue the equivalent of banknotes in a +arbitrarily small units. Entities will appear that aggregate large numbers of +small transactions on their network into a small number of large transaction +on the main network. As the network comes to span the stars, transaction +global to several stars will necessarily become too slow, leading to systems +that aggregate transactions in local currency over time and space into larger, +slower, and less frequent transactions on the main network. We don’t have to +worry about that kind of scaling for a very long time. The deflation problem +will likely be rendered irrelevant by the decentralization problem as we go +into space. Figure that one out later – need the Merkle-patricia blockchain +and paxos consensus formation on digital assets, and once we can construct and +prove arbitrary consensus on arbitrary digital assets, we can build anything. + +But trouble is, I want my data format to outlast the stars. Ozymandias +merely built in stone, which probably lasted a millennia or two. I +have more durable building materials at hand than Ozymandias did. + +I intend to initially define the smallest representable quantity as something +larger than $2^{-62}$ of the currency at issue, and then drop it to the lowest +value the ui can handle, probably yoctorho, $yρ$, when the sofware +supports that. And, having dropped, it is unlikely to change further for +many millenia or so. + +If someone reads this in a few millennia, and the exponent, still eight bits +on the blockchain, wraps through zero or one hundred and twenty eight, +drink to me as the first builder who truly built to live forever. + +One solution is to have the canonical blockchain format, and the base +communication format that every client and peer must support, even +though obviously any two peers can agree to any format to communicate +between each other, represent money in binary form as variable precision +base one thousand floating point, and to the users in units of the metric +prefixes tera giga, mega, kilo ... milli, micro, (and eventually nano, pico, +femto). When deflation runs us out of prefixes, in a few millennia or so, +perhaps the prefixes will wrap back to from zepto to yotta, but we can +worry about that UI detail in the far future, supposing that the language +has not radically changed by then. + +We have a configurable limit on the smallest representable quantity, which +just happens to correspond to translating everything to sixty four bit +integers, but that limit can be changed as necessary without breaking the +canonical format - thus the canonical format will suffice forever. The sixty +four bit integers will be an internal implementation detail, a particular +internal representation of unsigned arbitrary precision floating point base +one thousand, which can change in any one peer without breaking +anything, and with machines using different internal representations still +able to communicate with each other. + +M2, total US money supply, is about ten trillion, MB, the hard central bank +issuance that roots M2, the base money, is about three trillion, the difference +being term transformation. + +Assuming we want to replace all money everywhere, and support +transactions down to one thousandth of a cent, $2^{64}-1$ millicents is over +one hundred trillion, which will suffice.  (We don’t allow negative account +values in the base money.) + +So, assuming at full volume, the currency is worth ten trillion, the base +unit will be worth around 0.005 millicents. And at initial release, we want +the total value of the currency to be about twenty million, so the base unit +of the currency will be initially worth about 1E-10 cents. We want plenty of +headroom for additional currency issue, so will initially issue only one +sixty fourth of the possible currency, and want to initially issue sixteen +million worth, so want the smallest unity of currency to be\ +$2^{-64}*64*\$16\,000\,000$, which is approximately $\$2*10^{-10}$ + +Assuming we only have $2^{60}$ of the smallest base unit, and that when we +are competing on an equal footing with other media of exchange, it has a +capitalization of two trillion, then again the smallest base unit will be +worth about two millicents, which is inconveniently small. So our named +unit has to be a million or a billion times larger, + +If my plans work out, the currency will be controlled by the whales, who +have net positive value in the currency, hence want permanent deflation, +rather than the bankers, who owe a lot of promises to pay in the currency, +backed by promises that when push comes to shove are likely to result in +the delivery of property, rather than currency, and therefore have regular +banking crises, resulting in regular demands to debase the currency, +resulting in permanent inflation. So, assuming permanent deflation, make +the smallest base unit the microrho, $1µρ$. So, when we are competing in +the big leagues, our named unit will be worth about two dollars. Which is +inconveniently small, but I anticipate further deflation eventually. + +Traditional coinage had as its lowest value coin the half reale, or the +maravedi, one third of a reale. The most common coin in use was the peso, +the piece of eight rendered so famous in pirate lore, which was eight reales +or twenty four maravedi, subsequently divided into one hundred cents. +The famous doubloon was sixteen reales, or forty eight maravedi. + +An eight reale coin, peso, was worth about a hundred dollars in today's +money, so people were disinclined to use coins for small transaction, or +disinclined to be minutely precise about the value of a transaction. So we +want, after taking over the world economy, our standard unit of currency +to be worth about four dollars, which is about 80000 times our smallest +unit. But we want to use powers of a thousand, milli, kilo, mega, etc, So +our base unit is going to be the microrho, or $μρ$, and our standard unit, the +rho or $ρ$, is going to be worth about ten trillion$*1000000*2^{-64}$ which +is about half a dollar. Or we could make our smallest representable unit the +$nρ$, with might leave us with an inconveniently large value of the rho, and +everyone using millirho followed by a decimal point and the rest in $μρ$, +which is inconvenient. But, if we always display quantities in the metric +unit such that the quantity is less than a thousand of that unit, but equal to +or greater than one of that unit, it is OK. + +If we make our smallest possible base unit the $nρ$, then the maximum +possible currency on issue, until we go to internally representing values +within the computer as 128 bit, which is not that hard, since our +representation on the blockchain and to humans is arbitrary precision +times powers of a thousand, then the maximum transaction of which +computers are capable is going to be eighteen billion rho, which is not a +limitation. What is a limitation is that at scale, people will commonly be +transacting in $mρ$. On the other hand, if we start out transacting in $ρ$, and +end up transacting in $mρ$, that is continual propaganda for the currency as +a store of value. At scale, the $mρ$ will be a roughly right sized value to get +your mind around in small transactions, and the $ρ$ the right sized value for +asking how your solvency is going and how much a car or a house is +worth. + +We need to strongly support sidechains and chaumian cash, sidechains so that we can have competing protocols and higher volumes. Cryptonote has something arguably better than Chaumian cash. + +Our financial system is corrupt and oppressive. Cryptocurrencies represent an opportunity to route around that system, and make lots of money doing so. + +Cryptocurrency is real, and presents the opportunity to make enormous amounts of money.  Also, cryptocurrency scams are real, and present the opportunity to lose enormous amounts of money.  + +The successful altcoin will be genuinely decentralized, as bitcoin was designed to be, originally was, and to some extent still is.  Most of the altcoins, possibly all of them except the Bitcoins and Ethereum, are furtively centralized. + +It will use, or at least offer the option, of Zooko type wallet names. + +It will be scalable to enormous numbers of transactions with low transaction costs, as Steemit and Ripple are, but Bitcoin and Ethereum are not. + +It will support sidechains, and exchanges will be sidechained. + +It will be a blogging and tweeting platform, as Steemit is, and will be a decentralized blogging and tweeting platform, as Steemit is not.  + +Every website [reporting on the altcoin boom and the initial coin offering boom](https://coinmarketcap.com/coins/) has an incentive to not look too closely at the claimed numbers.  Looks to me that only Bitcoin and Steemit.com have substantial numbers of real users making real arms length transactions.  Maybe Ethereum and Ripple also.  The rest are unlikely to have any significant number of real, arms length, users. + +The crypto coin business is full of scammers, and there is no social pressure against scammers, no one wants to look too closely, because a close look would depress the market. + +Most of the alt currencies are just me-too copies of bitcoin, not adding any substantial value, and/or they cannot scale, and they are deceptive about how centralized and how vulnerable to state attack they are.  Nearly all of them are furtively centralized, as Bitcoin never was.  They all claim to be decentralized, but when you read the white paper, as with Waves, or observe actual practice, as with Steemit, they are usually completely centralized, and thus completely vulnerable to state pressure, and quite likely state seizure as an unregulated financial product, thus offer no real advantage over conventional financial products. + +The numbers [show](https://coinmarketcap.com/coins/) that Bitcoin is number one, ethereum number two, ripple number four, and steemit.com number eighteen, but my wild assed guess is that Bitcoin is number one, steemit number two, ethereum number three.  I have absolutely no idea where ripple stands.  No one is providing data that would enable us to estimate real, arms length users. + +Bitcoin exchanges are banks, and banks naturally become fractional reserve institutions.  Bitcoin exchanges are furtively and secretly investing customer deposits, without reporting the resulting term transformation. + +Genuinely free market banks, and bitcoin exchanges are genuinely free market banks, have a financial incentive to engage in term transformation – borrow short, lend long.  Which is great for everyone until a rainy day comes, rains on everyone, and everyone withdraws their deposits all at the same time, and suddenly all those long term loans cannot be liquidated except at a loss, whereupon the ~~banks~~exchanges turn to the state, and so begin the transition from a backed currency to a state currency, ceasing to be free market banks. + +The trouble with fractional reserve is that free market banks, banks trading in a backed, rather than state, currency, tend to deny, understate and misrepresent the term transformation risk, making them slowly, and often unintentionally, drift into becoming scams.  If the reserve fraction is visible to customers, then we could rely on caveat emptor.  Right now, however, every bitcoin exchange is drifting into becoming a scam. + +We need, and we could easily have but do not have, a system where the amount of bitcoins owed to customers by an exchange is knowable and provable, and the amount of bitcoins owned by an exchange is knowable and provable, so that the reserve fraction is visible, whereupon the exchange would have to provide information about the extent and nature of its term transformation, or else would likely lose customers, or at least would lose large, long term customers.  This would involve the decentralized cryptocurrency making each exchange a sidechain operating a centralized cryptocurrency backed by the decentralized cryptocurrency.  Which would also help mightily with scaling. + +Bitcoin and ethereum is truly decentralized, in that it is a protocol that any entity can use, and that in practice lots of entities do use.  If the government grabs some hosts, or some hosts do bad things, they can just be ignored, and the system continues elsewhere.  They also use Zooko type identities, which in practice means your wallet name looks like line noise.  This is outstandingly user hostile, and a reason so many people use exchanges, but it provides the core of resistance to state power. + +Unfortunately, Bitcoin and Ethereum face scaling limits.  Maybe ethereum will fix its scaling limits.  Bitcoin does not seem to be fixing them.  This makes Bitcoin and Ethereum transactions inherently expensive, which is likely to prevent them from replacing the corrupt and oppressive US government controlled financial system. + +Steemit.com has a far superior design which does not result in scaling limits – although we have yet to see how its witness election system will perform at scale – as the system scales, money holders have less incentive to vote, less incentive to vote responsibly, and voting will inherently cost more. + +Steemit.com is also highly centralized.  The altcoin that will win will be the one needs to be scalable all the way to Visa and Mastercard levels, and needs to be visibly decentralized, visibly resistant to state seizure, and needs to have a mechanism that makes the fractional reserves of exchanges visible to exchange users. + +Bitcoin was genuinely decentralized from the beginning, and over time became more centralized.  Big exchanges and a small number of big miners are on the path to inadvertently turning it into another branch of the oppressive and corrupt government fiat money system. + +The new altcoin offering are for the most part not genuinely decentralized.  They have a plan for becoming genuinely decentralized some time in the future, but the will and ability to carry the plan through has not been demonstrated. + +I like the steemit design.  The witness system is scalable, the witness election system has problems which may be fixable, or may be inherent. + +But I have a suspicion that investing in steemit is only going to profit whoever owns steemit.com, not the owners of steemit currency. + +According to Steemit documentation, it looks like a well designed cryptocurrency that deserves to replace Bitcoin, because it is more scalable, more user friendly, and more immediately usable. + +Well, that is what it looks like.  Except its front end is the steemit.com website, and any one website can easily be seized by the feds.  If actually decentralized, it should be a bunch of websites using a common crypto currency and a common identity system, + +Remember usenet: A common protocol, and an internal name system.  The particular host through which you accessed it did not matter all that much, because all hosts had to behave much the same. Steemit should be something like usenet with money, and it is not. + +The way usenet worked, anyone (meaning anyone’s computer and his client program) could join as a client by having an agreement with a host, and anyone (meaning anyone’s powerful and well connected computer system) could join as a host by having an agreement with a few existing members. + +A successful altcoin needs to be a blogging platform like Steemit, but it also needs to be a federation, like Usenet or Mastodon.  Many of the blogs will be offering goods or services for cryptocurrency. + +Then one could be more sure that success of the federation currency would benefit owners of the currency, rather than owners of a single central website. + +Needs to be Mastodon with the ability to support a blog like post, and like Steemit, and unlike Mastodon, to send and receive money.  Steemit.com is wordpress.com with the ability to send and receive money. + +Bitcoin has a decentralized name system, rooted in Zooko style names that are not human intelligible.  Its resistance to state power comes partly from the fact that there are several miners and anyone can be a miner, and partly from its decentralized name system. + +Steemit has a communication and blogging system.  But if I hold steemit currency, steemit.com connects that to my phone number, which the government connects to my true name.  All that handy dandy data that the government would like all in one place that you can serve a warrant on or mount a raid on.  Or just sell for profit. + +Need a decentralizedd communication, identity, name, and blogging system, unlike Steemit.com’s centralized communication and blogging system, and a name system that is resistant to government intervention and control, like Bitcoin’s name system. Thus the blogs offering goods and services for crypto currency will be resistant to regulation or seizure by the state.  When a ruler meddles as much as our state does, he gives dangerously great power to those dangerously close to him.  The regulatory state inevitably drifts into anarcho tyranny, or, like Venezuela, into violent and chaotic anarchy. + +But we also want human readable names.  How can we square Zooko’s triangle? (As Aaron Schwarz famously asked, and then infamously gave a very stupid answer.) I will give my answer as to how a crypto currency can square Zooko’s triangle in a following post.  (The answer being, much as namecoin does it.) + +Now since any crypto currency system is a generalized secure name system with money, how do we make this system available for general access between computers? + +Our wallet client will provide an interface to something that looks and acts very much like your browser bookmarks system.  Except that links in the system correspond to a new kind of url, perhaps ro: This will be registered the same way magnet, https, mailto, and http are registered.  In windows they are registry entryies of the form + +> Computer\\HKEY_CLASSES_ROOT\\http\\shell\\open\\command + +except, of course, that ours shall be + +> Computer\\HKEY_CLASSES_ROOT\\ro\\shell\\open\\command + +In our name system, links consist of a wallet name followed by a path.  The target wallet maps these names to a server somewhere, likely on his system, and a client protocol, such as http, on your system. + +The target may want a client walletname, or the client username and shared secret, which is usually stored in the link, but if it is not, has to be typed into the wallet software when you are opening the link.  Any required user name and password negotiation is done in the wallet UI, not in the UI of the client being launched. + +If the client protocol is http, this results in the wallet creating on your system a port which maps to a port on the destination system, and then launching your browser.  If a username and password is needed, then the wallet does the negotiation and launches the browser with a transient cookie. + +Thus, suppose the url ro:example_name/foo maps to http protocol with some target system determined by the owner of example_name. + +Then some port, perhaps 3237 on your system, will be mapped to port 80 on the target system, then the url ro:example_name/foo/bar will result in the command to launch your browser to http://localhost:3237/bar + +This is not a system for attaching to our legacy browser system.  It is global connection and protocol negotiation system which can be used for legacy systems such as http.  That browsers will mishandle these translated urls is a browser bug.  They should talk directly to the wallet client, and say "give me a socket for this ro protocol url." + +TCP identified protocols by small numbers, and target machines by rather larger numbers.  This totally failed to scale, and we have to replace it with a [better scheme](./protocol_specification.html), with support for urls such as "magnet" and "http" as a degenerate special case of this more general and more powerful scheme.  + +------------------------------------------------------------------------ + +The coin to invest in, the coin that I will invest in both in money and as a software contributor, will solve the scaling problem, will be capable of scaling all the way to wiping out the US\$ as a world currency.  It will have integral support for sidechains with payments out of one sidechain to another sidechain being endorsed by sidechain signature which can be generated by arbitrarily complex rules idiosyncratic to that sidechain provided that conformity to the rules has verification of bounded computational time that the central chain can evaluate.  It will have an efficient system for securing history in which Merkle trees do not grow to enormous depth, so that it is possible to efficiently verify any one small part of history without needing to verify all transactions that have ever taken place.  (Because scalability implies we abandon everyone verifying everything down to the last byte.) + +It will be decentralized in the sense that if the police grab every single major contributor, software writer, and server, they cannot change the rules and make the currency act differently, they can only seize the money of the people that they have grabbed. + +A Merkle tree is a tree where every node contains the hash of its immediate children.  Thus the hash of the root of any subtree guarantees the contents of all its descendants, just as the hash of a file guarantees the contents of the entire file. + +This means that we can keep on adding to the tree, while keeping the past immutable, which is a useful feature for tracking who owns what, and who owes what.  If many people see the current hash at time X, you cannot change details about the past of time X without revealing what you have been up to. + +Any tree can be severely unbalanced, for example a binary tree where every node has a right hand child, and very few nodes have a left hand child, in which case the depth of the tree is approximately proportional to the total number of nodes in the tree – and the tree grows to enormous depth when the total number of node is enormous. + +Or it can be approximately balanced, in which case the depth of the tree is approximately proportional to the log of the number of nodes, which is always a reasonably small number even if the number of nodes is enormous. + +And a hash that testifies to every transaction that anyone ever did is going to be the hash of an enormous number of nodes.  But if it is at the root of a tree of moderate depth, then we can validate any part of the tree for conformity with the rules without validating the entire tree for conformity to the rules.  + +A blockchain is a Merkle tree that is chain like, rather than tree like.  Its depth grows linearly with its size, thus in time it becomes very deep.  Every node must store or at least have processed and summaried, the entire tree.  Thus if many equal nodes, cost of adding transactions is proportional to the number of nodes + +Thus, if we want a decentralized system, this can get very expensive. + +We want a system that can resist state power, a system where if the state grabs a few individuals and coerces them, it can seize their money, and perhaps all the money that they manage for other people, but cannot seize the entire system.  If it wants to grab control of everyone’s money, has to grab everyone, or at least grab most people.  Thus reducing the cost by having a few people authorized to validate the blockchain is a bad option, since the state could grab those people, or those people could conspire together to scam everyone. + +A blockchain runs on a set of nodes, each of which may be under the control of a separate company or organization. These nodes connect to each other in a dense peer-to-peer network, so that no individual node acts as a central point of control or failure. Each node can generate and digitally sign transactions which represent operations in some kind of ledger or database, and these transactions rapidly propagate to other nodes across the network in a gossip-like way. + +## The way bitcoin works + +Each node independently verifies every new incoming transaction for validity, in terms of: (a) its compliance with the blockchain’s rules, (b) its digital signature and (c) any conflicts with previously seen transactions. If a transaction passes these tests, it enters that node’s local list of provisional unconfirmed transactions (the “memory pool”), and will be forwarded on to its peers. Transactions which fail are rejected outright, while others whose evaluation depends on unseen transactions are placed in a temporary holding area (the “orphan pool”). + +At periodic intervals, a new block is generated by one of the “validator” nodes on the network, containing a set of as-yet unconfirmed transactions. Every block has a unique 32-byte identifier called a “hash”, which is determined entirely by the block’s contents. Each block also includes a timestamp and a link to a previous block via its hash, creating a literal “block chain” going back to the very beginning. + +Just like transactions, blocks propagate across the network in a peer-to-peer fashion and are independently verified by each node. To be accepted by a node, a block must contain a set of valid transactions which do not conflict with each other or with those in the previous blocks linked. If a block passes this and other tests, it is added to that node’s local copy of the blockchain, and the transactions within are “confirmed”. Any transactions in the node’s memory pool or orphan pool which conflict with those in the new block are immediately discarded. + +Every chain employs some sort of strategy to ensure that blocks are generated by a plurality of its participants. This ensures that no individual or small group of nodes can seize control of the blockchain’s contents. Most public blockchains like bitcoin use “proof-of-work” which allows blocks to be created by anyone on the Internet who can solve a pointless and fiendishly difficult mathematical puzzle. By contrast, in private blockchains, blocks tend to be signed by one or more permitted validators, using an appropriate scheme to prevent minority control. + +Depending on the consensus mechanism used, two different validator nodes might simultaneously generate conflicting blocks, both of which point to the same previous one. When such a “fork” happens, different nodes in the network will see different blocks first, leading them to have different opinions about the chain’s recent history. These forks are automatically resolved by the blockchain software.  In bitcoin, the probability of this conflict continuing drops rapidly and exponentially, but never goes to zero. + +This document is licensed under the [CreativeCommons Attribution-Share Alike 3.0 License](http://creativecommons.org/licenses/by-sa/3.0/) diff --git a/docs/cyperhpunk_manifesto.md b/docs/cyperhpunk_manifesto.md new file mode 100644 index 0000000..a15a884 --- /dev/null +++ b/docs/cyperhpunk_manifesto.md @@ -0,0 +1,35 @@ +--- +description: >- + “A Cypherpunk’s Manifesto” was written by Eric Hughes and published on March 9, 1993. +robots: 'index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1' +title: >- + Eric Hughes: A Cypherpunk’s Manifesto +viewport: 'width=device-width, initial-scale=1.0' +--- +**Privacy is necessary for an open society in the electronic age. Privacy is not secrecy. A private matter is something one doesn’t want the whole world to know, but a secret matter is something one doesn’t want anybody to know. Privacy is the power to selectively reveal oneself to the world.** + +![The following essay was written by Eric Hughes and published on March 9, 1993. A Cypherpunk’s Manifesto was originally published on activism.net](./eric.jpg "Eric Hughes: A Cypherpunk’s Manifesto"){width="100%"} + +If two parties have some sort of dealings, then each has a memory of their interaction. Each party can speak about their own memory of this; how could anyone prevent it? One could pass laws against it, but the freedom of speech, even more than privacy, is fundamental to an open society; we seek not to restrict any speech at all. If many parties speak together in the same forum, each can speak to all the others and aggregate together knowledge about individuals and other parties. The power of electronic communications has enabled such group speech, and it will not go away merely because we might want it to. + +Since we desire privacy, we must ensure that each party to a transaction have knowledge only of that which is directly necessary for that transaction. Since any information can be spoken of, we must ensure that we reveal as little as possible. In most cases personal identity is not salient. When I purchase a magazine at a store and hand cash to the clerk, there is no need to know who I am. When I ask my electronic mail provider to send and receive messages, my provider need not know to whom I am speaking or what I am saying or what others are saying to me; my provider only need know how to get the message there and how much I owe them in fees. When my identity is revealed by the underlying mechanism of the transaction, I have no privacy. I cannot here selectively reveal myself; I must *always* reveal myself. + +Therefore, privacy in an open society requires anonymous transaction systems. Until now, cash has been the primary such system. An anonymous transaction system is not a secret transaction system. An anonymous system empowers individuals to reveal their identity when desired and only when desired; this is the essence of privacy. + +Privacy in an open society also requires cryptography. If I say something, I want it heard only by those for whom I intend it. If the content of my speech is available to the world, I have no privacy. To encrypt is to indicate the desire for privacy, and to encrypt with weak cryptography is to indicate not too much desire for privacy. Furthermore, to reveal one’s identity with assurance when the default is anonymity requires the cryptographic signature. + +We cannot expect governments, corporations, or other large, faceless organizations to grant us privacy out of their beneficence. It is to their advantage to speak of us, and we should expect that they will speak. To try to prevent their speech is to fight against the realities of information. Information does not just want to be free, it longs to be free. Information expands to fill the available storage space. Information is Rumor’s younger, stronger cousin; Information is fleeter of foot, has more eyes, knows more, and understands less than Rumor. + +We must defend our own privacy if we expect to have any. We must come together and create systems, which allow anonymous transactions to take place. People have been defending their own privacy for centuries with whispers, darkness, envelopes, closed doors, secret handshakes, and couriers. The technologies of the past did not allow for strong privacy, but electronic technologies do. + +We the Cypherpunks are dedicated to building anonymous systems. We are defending our privacy with cryptography, with anonymous mail forwarding systems, with digital signatures, and with electronic money. + +Cypherpunks write code. We know that someone has to write software to defend privacy, and since we can’t get privacy unless we all do, we're going to write it. We publish our code so that our fellow Cypherpunks may practice and play with it. Our code is free for all to use, worldwide. We don’t much care if you don’t approve of the software we write. We know that software can’t be destroyed and that a widely dispersed system can’t be shut down. + +Cypherpunks deplore regulations on cryptography, for encryption is fundamentally a private act. The act of encryption, in fact, removes information from the public realm. Even laws against cryptography reach only so far as a nation’s border and the arm of its violence. Cryptography will ineluctably spread over the whole globe, and with it the anonymous transactions systems that it makes possible. + +For privacy to be widespread it must be part of a social contract. People must come and together deploy these systems for the common good. Privacy only extends so far as the cooperation of one’s fellows in society. We the Cypherpunks seek your questions and your concerns and hope we may engage you so that we do not deceive ourselves. We will not, however, be moved out of our course because some may disagree with our goals. + +The Cypherpunks are actively engaged in making the networks safer for privacy. Let us proceed together apace. + +Onward. diff --git a/docs/dovecot.md b/docs/dovecot.md new file mode 100644 index 0000000..e3a10e4 --- /dev/null +++ b/docs/dovecot.md @@ -0,0 +1,245 @@ +--- +lang: en +title: Install Dovecot on Debian 10 +--- +# Purpose + +We want postfix working with Dovecot so that we can send and access our emails from email client such as thunderbird client on another computer. + +# Enable SMTPS in postfix + +## prerequisite + +You have already enabled [postfix TLS] and made sure that it is working by checking your logs of emails successfully sent and received. + +[postfix TLS]:set_up_build_environments.html#tls + +## setup postfix to talk to dovecot + +We are going to enable `smtps`, port 465, which your email client probably +refers to as `SSL/TLS` and `ufw` refers to as `'Postfix SMTPS'` + +We are *not* going to enable `submission`, port 587, which your email client +probably refers to as `STARTTLS`, and `ufw` refers to as `'Postfix Submission'`, +because `STARTTLS` is vulnerable to downgrade attacks if +your enemies have substantial power over the network, and many major +email clients do not support it for that reason. Since we are using normal +passwords, a successful downgrade attack will leak the password, enabling +the enemy to read and modify mail from that client, and to send spearphish, +shill, scam, and spam emails as the client identity. + +Passwords are a vulnerability, and in a hostile, untrustworthy, and +untrusting world need to be replaced by ZKA resting on a BIPS style +wallet secret, but we have to make do with `smtps` until we create something better. + +```bash +nano /etc/postfix/master.cf +``` + +You will find the lines we are about to change already in the `master.cf` file, +but commented out, and some of them need to be amended. + +```default +smtps inet n - y - - smtpd + -o syslog_name=postfix/smtps + -o smtpd_tls_wrappermode=yes + -o smtpd_sasl_auth_enable=yes + -o smtpd_relay_restrictions=permit_sasl_authenticated,reject + -o smtpd_recipient_restrictions=permit_mynetworks,permit_sasl_authenticated,reject + -o smtpd_sasl_type=dovecot + -o smtpd_sasl_path=private/auth +``` + +Now we tell postfix to talk to dovecot over lmtp + +```bash +postconf -e mailbox_transport=lmtp:unix:private/dovecot-lmtp +postconf -e smtputf8_enable=no +``` + +Obviously this is not going to work until after we install and configure +dovecot, so don't restart and test postfix yet. + +# Install Dovecot + +```bash +apt -qy update && apt -qy upgrade +apt -qy install dovecot-imapd dovecot-pop3d dovecot-lmtpd +dovecot --version +# These instructions assume version 2.3 or above +nano /etc/dovecot/dovecot.conf +``` + +```default +protocols = imap pop3 lmtp +!include_try /usr/share/dovecot/protocols.d/*.protocol +``` + +## Authentication + +Edit the authentication file for Dovecot and update following values. + +```bash +nano /etc/dovecot/conf.d/10-auth.conf +``` + +```default +disable_plaintext_auth = yes +auth_mechanisms = plain +auth_username_format = %n +``` + +## Setup Mailbox Directory + +After that, edit mail configuration file to configure location of the Mailbox. Make sure to set this to correct location where your email server is configure to save users emails. + +```bash +nano /etc/dovecot/conf.d/10-mail.conf +``` + +```default +mail_location = maildir:~/Maildir +mail_privileged_group = mail +``` + +```bash +adduser dovecot mail +``` + +We already told postfix to talk to dovecot. Now we must tell dovecot to talk to postfix using lmtp. + +```bash +nano /etc/dovecot/conf.d/10-master.conf +``` + +Delete the old `service lmtp` definition`, and replace it with: + +```default +service lmtp { + unix_listener /var/spool/postfix/private/dovecot-lmtp { + mode = 0600 + user = postfix + group = postfix + } +} +``` + +Delete the old `service auth` definition, and replace it with: + +```bash +# Postfix smtp-auth +service auth { + unix_listener /var/spool/postfix/private/auth { + mode = 0660 + user = postfix + group = postfix + } +} +``` + +## Setup SSL + +```bash +nano /etc/dovecot/conf.d/10-ssl.conf +``` + +```default +ssl=required +ssl_cert = 2021-03-30 +To: Cryptography Mailing List +``` + +Entropy is important to RNGs but unfortunately RNG people are at risk of +devoutly worshiping at the alter of "Countable Entropy", blinded to +realities, ready with jeers for anyone who does not share their extreme +theology. + +These people are so in the thrall of the theoretical that they are blinded to +the practical and any other theories. + +And for practical purposes, it is the unguessability of the RNG that +matters. Any source of unguessable data is a good thing to use to drive an +RNG. Even sources that are dismissed as "squish" by the most devout and +most blinded can be good. And there is a great example that these disciples +can't see. + +# Time Distribution is Hard + +NTP is really great, I love it, so cool. It can set my computer's clock with a +precision measured in milliseconds. Very impressive it can do this just by +applying algorithms to hardware that is already present. + +If one wants better, the next option is to get time from GPS. According to +gps.gov a specialized receiver, at a fixed location, can know the time +within 40ns. This is pretty cool, too. It is good enough to synchronize RF +signals between CDMA cell sites so phones can communicate with more +than one site as the same time. + +GPS also depends on billions of dollars of infrastructure with an annual +budget that must be in the millions. People think GPS is about location, +but at its core it is really about time distribution. From end-to-end a design +where every component is doing its best to carefully keep and distribute +precise time. If one pays attention to details and has the money for good +hardware (far more than just a smartphone), to get 40ns is very cool. + +# Guessing Time is Even Harder + +With all that money can constructive effort, one can do 40ns. What are you +going to do to do better? Get specific. (Warning, it's not going to be easy.) + +# Cheap Unguessability + +A 1 GHz clock has a cycle time of 1ns. (Is it even possible to buy an Intel-based +machine that runs slower than 1GHz these days?) 1ns is a lot +smaller than 40ns. You don't know the value of my clock. + +Intel chips have a timestamp counter that increments with every tick of the +system clock. You don't know the value of my counter. + +The system clock isn't fed to the CPU, the CPU is fed a much lower +frequency, that it then multiplies up using analog on-chip PLL circuitry. +That clock is then (carefully) distributed on-chip. And even then, different +parts of the chip are in different clock domains, because clock distribution +and synchronization is hard. + +So the "system clock" doesn't exist outside the CPU, it is only a "CPU clock", +and not all parts of the CPU are even privy to it. + +No one at any distance outside that chip knows the value of the timestamp +counter. A program might contain the instruction to read the timestamp +counter, but by the time anything is done with that value, it will have +changed. + +Is there "entropy" in that system clock? Some, but only some. The PLL will +have some jitter, the precision of the lower frequency input clock will have +iffy precision and be subject to drift. + +Is there "unguessability" in that system clock? Plenty! At least to any +observer at any distance (i.e., outside the computer). + +Remember, it takes billions of dollars and lots of careful design and +cooperation to distribute 40ns. time. No such effort nor expense has been made +to tell the world the precise value of my 1ns period (or less) CPU clock. + +No one outside my computer knows its precise value. + +# Back on Topic + +Intel hardware has a great source of unguessability in its timestamp +counter. All you need is an uncorrelated sampling of this clock. Say, a +network interrupt. + +I know the squish patrol is now all upset, because external observers can +be the one's sending these packets with careful timing. So what? The +timing can't be careful enough. The value that is read from the timestamp +counter in servicing that interrupt depends on knowing edge timings far +more closely than 1ns, for every time the observer guesses a value on the +wrong side of one of these edges, one bit of unguessability slips by. + +# RNGs are Still Hard + +A (1) uncorrelated sampling of a (2) fast clock is, indeed, a good source of +unguessability. + +But, make sure both those things be true. + +Is virtualization messing with how these things work? Is variable clock +scaling messing with it? Have interrupts been virtualized in some +predictable way? Is the timestamp counter being messed with in an +attempt to have it not appear to be warped by clock scaling and effectively +running much slower? Is some OS scheduling algorithm synchronizing +interrupt servicing with timestamp values? + +Just because there is an underappreciated way to feed an RNG doesn't +mean there aren't plenty of ways to still mess it up. ("Um, it turns out the +RNG isn't in production builds." Who will notice?) + +Implementation matters. + +But the fact remains time distribution is hard, the period of a gigahertz clock is small. No one at any distance knows its value. An awful lot of computers out there can use this to drive their RNGs. + +-kb, the Kent who laments that Arm CPUs didn't have something like a timestamp counter last he looked. + +# Attacks + +```default +From: Barney Wolff 2021-05-31 +To: Cryptography Mailing List +``` + +Surely this depends on how many guesses an attacker is +allowed before being detected and blocked. If there's +no penalty for guessing wrong, as with an offline attack, +I doubt the GHz ticker can contribute more than about 20 +bits or so. + +# Implementation + +```default +From: jrzx 2021-06-06 +To: Cryptography Mailing List +``` + +Every network or disk event provides several bits of unguessability. You +are going to accumulate a 128 bits in a hundred milliseconds or so. + +Accumulate the bits into Knuth's additive number generator 3.2.2, then +hash the seed. + +Continue accumulating randomness into the seed when you get +uncorrelated events. Continue hashing the seed when you need more +random numbers. + +The attacker performing an offline attack will have to guess all 128 bits. diff --git a/docs/identity.md b/docs/identity.md new file mode 100644 index 0000000..8a4c730 --- /dev/null +++ b/docs/identity.md @@ -0,0 +1,1137 @@ +--- +title: + Identity +--- +# Syntax and semantics of identity + +The problem is, we need a general syntax and semantics to express +identity. + +Our use cases are likely to include a big pile of documents signed by diverse +people, with no contact information, some of them encrypted so that they can +only be read by people with certain private keys, with no indication of the +public key corresponding to that private key. + +So, what is our ascii armoured signature going to look like? + +If we ascii armouring, we are likely signing a utf8 string. Which will be +hashed as a count based string introduced by an arbitrary precision integer +and followed by a null that is not included in the count, even if it is a null +terminated string with no count, or a count based string that normally has no +null terminator. This is to ensure that it is impossible to concoct a +multiple string sequence that will have the same hash for a group of strings +as for a group of strings grouped differently, and so that different computers +with different word lengths and different endianness will generate the same +hash for the same string or sequence of separate strings. + +A sig block consists of: +> `{sig` 252 bitstring as base sixty four characters, arbitrary + sequence of non base sixty four characters, 252 bit bitstring as base sixty + four characters, optional arbitrary sequence of non base sixty four + characters, 256 bit bitstring as base sixty four characters representing + the public key, optionally followed by whitespace or linefeed characters, + followed by arbitrary utf8 characters representing the nickname of that + public key, which must start with a non whitespace character, the name being + followed by `}` and may not contain `}` + + Or alternatively the nickname may be represented by a nickname block + composed of `{nick`, optionally followed by an arbitrary sequence of + bracketing or whitespace or symmetric ascii characters followed by `“`, + followed by the nickname, followed by `”` followed by the reverse sequence, + followed by `}` + +A single signed string may have several different but equivalent ascii armorings: + +* unquoted + + * blank line or start of document. + * `:::` on a line by itself + * string to be signed on a line by itself (or lines by itself if it contains + line feeds)\ + * `:::` followed by sig block. + * blank line + +* quoted + + * blank line or start of document. + * `:::` followed by arbitrary sequence of bracketing or whitespace or symmetric ascii characters followed by `“`, on a line by itself + * string to be signed on a line by itself (or lines by itself if it contains line feeds) + * `”` followed by the reverse arbitrary sequence followed by `:::` followed + by sig block. + * blank line + +* inline unquoted + * `[` string to be signed `]` followed by sig block. + +* inline quoted + * `[` followed by arbitrary sequence of bracketing or symmetric ascii + characters followed by `“`, followed by string to be signed followed by `”` + followed by reverse arbitrary sequence of bracketing or symmetric ascii + characters, followed by`]` followed by sig block. + +A signature represents an identity. If a means of contacting that identity +is to be represented, it will be represented outside of and separately +from that signature. + +Very commonly we want to sign not just one arbitrary string, but an arbitrary +string and or a nickname and or a public key. This is done similarly to the +above, with the public key introduced by a hash sign, and the nickname bare or +in a nick block. + +For example: + +> `:::`\ +> Hi\ +> `:::` John Hancock `#0123456789ABCDEFGHJKLMNPQRSTUVWXYZabcdefghi` +> `{sig 0123456789ABCDEFGHJKLMNPQRSTUVWXYZabcdefgh 0123456789ABCDEFGHJKLMNPQRSTUVWXYZabcdefgh}` + +Or: + +>`[`Hi`]` John Hancock `#0123456789ABCDEFGHJKLMNPQRSTUVWXYZabcdefghi` +> `{sig 0123456789ABCDEFGHJKLMNPQRSTUVWXYZabcdefgh 0123456789ABCDEFGHJKLMNPQRSTUVWXYZabcdefgh}` + +We cheerfully assume that strings have no semantically and syntactacly +significant characters, and if they do have semantically significant +characters, we bracket the string with angle quotes, or if the string +contains angle quotes, with angle quotes and an arbitrary string of +non significant bracketing, symmetric, or whitespace ascii characters +around the angle quotes. + +`#`, `“`, `”`, `"`, `:::`, `[`, `]`, `{`, and `}` are semantically and +syntactically significant in certain contexts, and to distinguish between +the endless variety of uses to which they will be put, the closing `:::` or +the opening `{` will generally be immediately followed by a short label +identifying the particular use to which it is being put. A starting `:::` is +preceded by a blank line or start of document, and an ending `:::` is +followed by a `{label ...}` identifying the use to which it is put, followed +by a blank line. The ending `:::` is a ternary operator, linking several +fields + +Thus, if someone obnoxiously wanted line feeds, angle quotes, and curly +brackets in nickname, perhaps: + +> John Hancock {prince of “darkness”} + +a signed block of text might then look like: + +> `:::`\ +> Hi\ +> `::: (|“`John Hancock {prince of “darkness”}`”|) #0123456789ABCDEFGHJKLMNPQRSTUVWXYZabcdefghi` +> `{sig 0123456789ABCDEFGHJKLMNPQRSTUVWXYZabcdefgh 0123456789ABCDEFGHJKLMNPQRSTUVWXYZabcdefgh}` + +with the `(|“` … `”|)` acting as quote marks that will cause whatever is +inside them to be treated as a single string by the surrounding +representation. To avoid the need for escaping special characters, we allow +an infinite variety of quotation symbols. + +Our goal in breaking from uri syntax, json syntax, and even yaml and markdown +syntax, is a syntax that allows arbitrarily complex cryptographic expressions +to be represented to the end user in a way that is as intuitive as they can be. + +This syntax is inspired by Pandoc markdown, which extended the syntax of +markdown to allow better combination of markdown with html and css, while +still holding on to as much readability as possible. + +Whitespace characters can be liberally inserted and will generally +be ignored, except that they have to be balanced when demarking quotes or +that they part of the string or the nickname. Literal strings will only +occur within these labelled bracketing operators, or within quotes inside +these labelled bracketing operators + +In an environment where lines are represented by something other than line +feed characters, the lines shall be converted into line feed characters before +being hashed and signed, and line feeds in the signed string may be +represented by lines native to the environment when the ascii armored signed +string is displayed in an environment where lines are represented by something +other than line feed characters. + +# Primary and most urgent use case + +Our primary use case, however, is not mere identity, but is a link that will +bring you to a shopping cart, containing a link to a checkout, containing a +link that will say “Your order has been placed” and will generate a record +that the payment has been attempted, containing a link that can check on your +payment status, and generate a review or respond to the vendor, which should +be followed eventually by an email like message from the vendor that your +payment has gone through, also containing a link that could generate a review +and respond to the vendor. + +Our secondary case is signed messages subject to usenet style +authentication and hashtag style pooling, distribution, and grouping +(rather than usenet hierarchical grouping), and to spam filtering by the +distributors (which is necessarily indistinguishable and inseparable +from political filtering, because a lot of political messaging *is* +spam, containing repetitiously similar messages, which messages are apt +to be misleading and manipulative). And with social media style +following, liking, downvoting, and reposting (where a repost should not +result in the same message appearing in someone’s feed multiple times, +but in the feeds of the followers of the reposting party who did not +follow the posting party. A repost should appear with the reposters that +you are following appearing. Friending should be a request to the +client’s server for permission to communicate, but following should +grant permission and capability for one way communication, while +successfully sending a message automatically by default grants +permission to communicate one way, with one way one on one communication +permitting two way one on one communication. Thus mutual following +should by default permit mutual two way one on one communication. + +Mutual following should by default imply friending, and friending should +by default imply mutual following. + +We should also allow exploding messages, which are authenticated by the +client to his server, and authenticated by his server, but you have to +login, perhaps anonymously or under a throwaway identity, to his server +in order to see them – they cannot go in usenet style pooling. + +An identity should be able to function as host and server if the owner +of that identify chooses to set that identity to continually publish its +signed network address (obtained from its host). This enables anyone to +send messages directly to that identity, but normally most such messages +are automatically ignored and discarded. + +Thus if an identity is selling something controversial or arguably +illegal, and is functioning as a server, in order that people can make +purchases, anyone in the world can discover and monitor its network +address. Therefore, purchases have to be possible between client non +host identities, whose network address is not widely known. The +transaction sequence (link, shopping cart, purchase screens) have to be +email like. + +Our primary use case for exploding messages is discussing the +manufacture, sale, and purchase of goods, such as guns, which may +attract attacks. + +In this case you have a non radioactive server carrying signed messages, +some of them mildly radioactive and carrying links to a mildly +radioactive server, carrying highly radioactive messages, some of them +exploding, which contain links to contactable identities through which +radioactive goods and services can be bought and sold, leading to signed +or exploding reviews supporting or detracting from the reputations of +sellers and purchasers of highly radioactive goods. + +This implies a cryptographic resource identifier has names and a public +key, or a blockchain transaction output identifier, but may be useless +without a locator which identifies a host. As a general rule, your host +will know where to find that host. + +Cryptographic resource identifiers should be visible to the human as per +message petnames of the entity embedding the resource identifier, +qualified by the users petname, if any, and the entities nickname, with +the embedding defaulting to the entities nickname, unless the composer +provides a per message petname. + +Because messages are widely distributed, using authentication rather +than signing would not make a difference and would be difficult to +implement. The identities signing therefore would need to be +discardable, but they will need to permit the possibility of one on one +contact, usually through a peer that hosts the client identity, perhaps +directly, if the sender wishes it. If it is a sales message, which is +our primary use case, it will frequently be convenient to be contactable +directly, usually through a named blockchain transaction output, +supported by the pool of signed network addresses associated with +blockchain outputs, which typically contains not the network addresses +of the keys signed by the blockchain key, but the network addresses of +keys signed by that key as having authority to communicate on its +behalf. + +We aim for a pseudonymous currency supporting reputations, not an +anonymous currency. We worry first about identity, and actually locating +that entity is an ad hoc afterthought cobbled on top. We want +cryptographic resource identifiers, which in some cases have the side +effect of locating the resource, in other cases have the effect of +identifying a byte string that has somehow arrived on your computer, or +discovering that you don’t have certain items of evidence that should +have arrived on your computer, but have not, and should be searched for +on the cloud. + +The foundation of our system is probabilistically unique cryptographic +identifiers, twofiftysix bit identifiers. + +# Cannot comply with uri syntax + +I would like to conform to the Uniform Resource Locator and Uniform +Resource Identifier syntax and semantics, but they are too married to +the existing system, and we are going to have subtly different semantics +and much larger semantics. They have reserved no end of characters for +their own syntax, with their own semantics, and expressing complicated +semantics within a syntax designed for different and more restricted +semantics and useful characters reserved that syntax just does not fit +too well. And they never got hip to utf8. And they prohibit spaces. + +There is an obvious problem in permitting unicode in identifiers, since +the confusible identifier problem is insoluble even with ascii only +identifiers, and becomes enormously harder with unicode identifiers. So, +since no end of people have tried unsuccessfully to solve it with ascii +identifiers, we will make absolutely no attempt to solve it. Instead, we +make it more difficult to use confusible identifiers maliciously, and +trust people to do their best to make their identifiers distinctive. + +The usual malicious use of confusible identifiers is a man in the middle +attack: “This is an urgent notice from BigImportantFinancialInstituion. +You need to login immediately or all your money will be frozen or lost.” + +Then you login onto an entity that has a name that can be confused with +that BigImportantFinancialInstitution, give it all the secrets that you +share with the actual BigImportantFinancialInstitution, and lose all +your money. + +Obviously this is not going to work with cryptographic secrets, since +they are unshared. You login with a strong private secret, instead of a +weak shared secret. So we should just go on living with the possibility +and likelihood of confusible identifiers, as we have already been doing +for a very long time, and expect that they will cause far fewer +problems. + +If you a contact a confusible entity, and it looks confusible to your +client program, the interface will detect the incompatibility, and +demand a distinctive petname, rendering it no longer confusible. Thus +confusion becomes a client problem in a small space of names, rather +than a system problem in a very large space of names. People who create +clients will worry about it when and if people who use clients are +bothered by it. Which they probably will not be. We will not worry about +it unless people start demanding a fix. Which is only going to happen if +malicious people find a use for a confusible. Which, if we do other +parts of our job right, is not going to happen. Experience has +demonstrated that it is always easy to maliciously create a confusible +name. The solution is to set things up so that the malicious person can +do little harm with it, so that no one wants to create confusible names. + +Spaces are forbidden in a uri, because one routinely passes uris as +arguments on the command line, and arguments are separated by +whitespace. But lack of spaces is s severe blow to intelligibility, and +in particular a severe blow against [Zooko](./zookos_triangle.html) nicknames. One way around this +rule is to have a rule for interpreting command line arguments that if +an item is contained in angle quotes or brackets, it is one item, and to +have as part of the scheme syntax and schema a rule for interpreting the +object, that if it begins with an angle quote or bracket, it ends at the +matching angle quote or bracket, and similarly for any item within it. +If an operator expects an argument, and there is a bracket or angled +quote mark immediately after it or before it, then everything inside the +brackets to the next matching bracket, is one argument. + +The no spaces rule is a reflection of the widespread use of lexers when +we should be using parsers. Lexers fail to provide sufficient expressive +power. We should break compatibility by demanding that anything that can +handle our expressions uses a command line parser rather than a lexer, +because you are just not going to be able to handle +[Zooko](./zookos_triangle.html) nicknames in a +lexer. + +The uri syntax is written for a lexer, not a parser, and the command +line syntax is written for a lexer. Bad! + +A program that accepts cryptographic identifiers on its command line is +just going to have to parse, rather lex, its command line. Which is a +pain because all the standard libraries for command line handling are +lexers, not parsers. + +In order to allow people to identify themselves and their computers on +the cloud with distinct and intelligible cryptographic resource +identifiers, we are going to assume parsers everywhere and unicode +everywhere. If parsers are far from everywhere, they should be. + +We could easily conform to the uniform resource identifier syntax and +semantics by adding arbitrarily complex syntax and semantics within the +authority field [RFC2396 Section +3.2](https://tools.ietf.org/html/rfc2396). + +Oracle’s way of dealing with this problem was to have a pile of +alphanumeric keywords separated by dots in the authority. And then they +could have all the syntax and semantics they needed on the keywords, +which were lexed by the dot separators within the authority field. This, +of course, resulted in long and unintelligible authority fields full of +verbose boilerplate. For intelligibility, need to have the cryptographic +resource locator look as if it is an unknown scheme that has no +authority component. + +Trouble is that the restricted character set (no spaces, no unicode +beyond ascii seven bit, and restrictions on the use of ascii seven bit +punctuation characters) means that the resulting uris would not be +visually distinctive – it would be hard for end users to make their uri +look different and memorable. One way out of this conundrum is to have +our own non conformant cryptographic resource identifiers, which can be +made visually distinctive, and which can be, if needed, mapped to +indecipherable gibberish that all looks alike using % byte codes (I +refuse to call them octet codes), but which conforms to the permitted +character set within the authority field. + +Html tag restrictions prevent us from using \" or \' marks inside an +attribute (because an attribute may be contained inside \" or \') and +html attributes prohibit newlines, and provide no way of escaping +newlines into an attribute. At least we can escape \" inside an +attribute with `"` and \' and with `&039;`, or with % encoded +octets but this is useless because we don’t want our entities to look +different inside html than they do to the parser. + +[RFC9386 Section 2.2](https://tools.ietf.org/html/rfc3986) effectively +restricts lexing and parsing to ninety three characters, *excluding +spaces* because likely to be in a context that lexes on spaces, \", and +\' and then lexes and parses as one operation by reserving no end of +special characters, forgetting decades of study of the lexing and +parsing problem. + +As they attempt to squeeze more and more semantics into this +impoverished syntax, the syntax and semantics becomes ever more +impoverished. And we plan to permit a whole lot of new semantics into +the authority field. + +The impoverished syntax and semantics led to the triple backslash in the +file scheme. Before they could get to the field with file syntax and +semantics, they had to have an empty field with authority syntax. We +want visually distinctive cryptographic resource identifiers, without +distracting rubbish wrapping them. + +We do not need the // to distinguish the authority field, because we +have several different bases for authority, and need to distinguish them +within the authority field. Which means that anything parsing the +cryptographic resource identifier as uri is going to demand relative url +syntax and character set for the whole thing. + +# [Digression on parsers](./parsers.html) + +There is no real difference between a resource locator and a resource +identifier, because usually a resource identity depends on a name +assigned by some authority, and when you contact that authority, it is +likely to know where the resource is, and a resource locator may well +refer to a resource that cannot be located, resulting in the ubiquitous +403 and 404 messages, thus necessarily have the same syntax and +semantics. So, they are best all called resource identifiers, +particularly as the purpose of cryptography is to identify, not to +locate. + +But still, we are going to have be able to recognize normal uris, so we +will have to start an absolute resource identifier with a scheme, and +might as well end the scheme name with a colon, though we might piss on +them by having our scheme name be a unicode character that is not +permitted, under some circumstances followed by a space which is not +permitted either. + +But, on the other hand, we would like environments that do not know +about cryptographic resource identifiers to at least be able to +recognize it as an unknown scheme, so I guess `rho:` it is, at least as +an option. But after that colon, it is our playing field, and no longer +“universal”. Past the colon, a much bigger syntax than the overly +grandiose “universal” will apply. + +The universal scheme is inherently far from “universal”, because it as +soon as you put anything in a “universal” scheme other than how to +recognize the scheme identifier, it ceases to be universal. + +The “universal” resource identifier scheme contains no end of stuff that +belongs not in the “universal” scheme but in particular schemes. + +The reason for the irritating double backslash is because of protocol +relative authorities. You could potentially leave out the `http:`, in +which case you have to distinguish the authority from a local directory +name. + +The reason for the triple backslash `file:///` makes little sense, it is +there because `file://`(implied authority)`/`. The backslash is there +because of “universal” syntax and semantics that are not applicable to +file systems. Unfortunately, however, while file systems have a subset +of universal semantics, we have a superset. + +In our case, we might reference keys by name anywhere, and they might +have local names or blockchain names. So we could have explicit keys, +keys referenced by local name (which might well be a chain of names +specifying a path through a local tree of bookmarks and contacts) keys +referenced by unspent transaction output sequence number (which are +supposed to generate an exception if already spent) keys referenced by +transaction sequence number (the number is unchanged by spending, but +should not generate an exception if spent) and keys referenced by the +name of the string that they own on the blockchain. And since local and +blockchain names are just strings, you have to identify what the name +refers to. + +So, rather than the // system for designating authorities, we will have +a label, which may be a single reserved character, or a string followed +by a “:” + +We do not want to restrict possible names, since we want maximum +distinctness, intelligibility, and recognizability. We want names to +have available to them spaces and the full scope of characters. + +In order that we can use brackets to denote a string entity containing +terminal that could be interpreted as syntactically significant, rather +than just more string, our syntax will have to denote different kinds of +strings. Obviously a bracket enclosed string can contain operators that, +if interpreted as operators rather than part of the string, produce a +symbol that is not of string type, so if the parser is looking at the +expression to see if it can be a string, can accept symbols as strings +that in some contexts would be operators. + +Any sequence of strings is a string, whose value is that of all the +strings concatenated. + +Any sequence of non whitespace unreserved characters is a string. + +Any whitespace between two strings is a string. + +Reserved characters can be used as part of a string in contexts where, +if interpreted as syntactically significant, the resulting non terminal +would be of unexpected type. + +A [html +entity](https://www.w3.org/wiki/Common_HTML_entities_used_for_typography) +like `'”`' is a single character string, in this case is a string +containing an unbalanced quote mark. + +A unicode entity may also be represented by one or more % encoded bytes +(or as the standards people, who still bow down to the tyranny of punched +card machines long turned to rust call them, “octets”). + +Of course many environments will throw up if you have spaces or unicode +characters within something that looks like a uniform resource +identifier, in which case you can make your string out of html entities +or % encoded bytes corresponding to utf8 encoded characters. It will +look ugly and incomprehensible, impossible to read and very hard to +write, but the ascii armor will protect it in an environment that does +not like non ascii. Tidy throws up at spaces or unicode beyond ascii in +a domain name inside a uri, but is happy with unknown schemes, and happy +with html entities representing non ascii unicode inside a domain name +inside a uri, so you can armor any string into something that tidy will +happily accept as an unknown scheme. But the parser is written for the +twenty first century, and is intended to accept cryptographic resource +identifiers that are human intelligible. We still are under the tyranny +of standards set to accommodate punch card machines and it is long past +time that we broke compatibility with those standards. + +Arbitrary precision integers, arbitrary width windowed integers, public +keys, bitstrings, and the like, are presented by a sequence of base +sixty four digits immediately following, without intervening spaces, +something that the parser knows should be immediately followed by a +public key or whatever. They may also be represented by h% followed by +base hexadecimal digits, b% followed by base two digits, but base sixty +four is the default. An integer must be represented by %d followed by +decimal digits. Thus we can reference names that are distinctive and +unique, and reference them even cryptographic resource identifiers that +have to be transported in text that restricts available characters. + +Relative resource identifiers that may be relative to cryptographic +resource identifier or a universal resource identifiers will have to be +valid. + +The left hand hand part of a cryptographic resource identifier has to be +a scheme, and the right hand part is likely to be a standard relative +uri. The central part however is going to be a cryptographic authority, +and for that we need a much broader syntax than “universal”. + +And that cryptographic authority might well have several hosts with +several temporary subkeys and several network addresses, and several human +agents with several temporary keys and several network addresses. And +each of those network addresses should have the up to date public keys +of all of them. Everyone should be the equivalent of a domain name +server for the groups of which he is part. + +The parser that parses a cryptographic resource identifier is going to +encounter no end of things that require look up over large local +databases, databases in the cloud, and finally, lookup on the target +host. Thus parsing a cryptographic resource identifier tends to be +equivalent to locating the resource. A 403 or 404 is parsing failure +because of undefined symbol. + +If an entity referenced in the identifier has existence, and perhaps +class and parse type, some host, possibly far away, has given it that +existence and parse type then for the parse to succeed, the parser has +to connect to that host. + +The straightforward case of [Zooko](./zookos_triangle.html) cryptographic resource identifiers is +that your equivalent of a domain name is a public key. You look up the +network address and the public key actually used for communication in +the equivalent of the domain name system, and get a public key signed by +the domain name key. Then you communicate with that address, with your +communications being authenticated by that key. There are likely three +destination public keys involved in making the connection: a durable +public key whose corresponding private key is likely not on any +computer, but written down on a page in an old book on the bookshelf of +the rightful owner of that public key, a durable but replaceable key +signed by that master key, and a session key that provides perfect +forward secrecy in that the corresponding secret key is discarded when +the connection is closed. + +> `#4397439879483774378943798` + +Represents the public key of an entity that knows the corresponding +secret key, or a block of data, commonly a widely shared and distributed +block of data, that when hashed according to its schema has that hash. +If a public key, you use it to make an authenticated connection or to +authenticate a signature, if a hash, authenticate data. If you can find +it, you probably already know which it is and what you are going to do +with it. + +If it is a public key, and you somehow locate the network address of +that entity, it will either authenticate itself with that key, or +authenticate itself with a public key signed by that entity authorizing +it to act as agent in some capacity. + +> Bob `#4397439879483774378943798` + +Represents the entity known to `#4397439879483774378943798` as Bob +(though if it is part of signature, it means that that entity knows +itself as Bob, that that is the nickname of the identity that this +key reprsents.) Likely the the entity itself, possibly with a +different public key, possibly an agent authorized to speak and act +on behalf of +`#4397439879483774378943798`, or possibly some random guy on his +friend’s list. If you have contacted `#4397439879483774378943798`, you +have likely contacted `Bob`, and if not `#4397439879483774378943798` may +be of help in contacting him. If the unlikely event that the entity has +a Bob on its contact list, and a different Bob on its authorized agent +list, this will bring up the agent. + +> Bob `#4397439879483774378943798` + +The entity known to `#4397439879483774378943798` as “Bob”, but not +authorized act on behalf of `#4397439879483774378943798`. For another +entity, Bob is probably a different Bob, and the same Bob probably has a +different name at that entity. + +> Receivables `#4397439879483774378943798` + +The entity known to `#4397439879483774378943798` as “Receivables”, and +authorized to act for that entity in some role. Possibly the entity +itself. + + > `#4397439879483774378943798/foo` + +A data object on the computer that identifies itself on the network with +this public key, or a public key authorized by this public key. The data +object is typically itself a name table, hence +`#4397439879483774378943798/foo/bar/program_name/arbitrary data for program.` + +uris have tied themselves in knots distinguishing between a file, and +data passed to the program represented by that file to execute. Probably +better to just say that anything to the right of the slash is entirely +up to entity to the immediate left of the slash to interpret, and if it +contains spaces and suchlike, use windows command line string +representation rules, quote marks and escape codes. + + rho:#4397439879483774378943798 + rho:Bob#4397439879483774378943798 + Bob@#4397439879483774378943798 + Receivables.#4397439879483774378943798 + +fit into the Uniform Resource Identifier scheme, poorly. + + #4397439879483774378943798/foo + +fits into the catchall leftover part of the Uniform Resource Identifier +scheme. + + rho:Bob@Carol.Dave#4397439879483774378943798/foo + +Does not fit into it in the slightest, and I think the idea of +compatibility with the URN system is a lost cause. + +But public keys are non memorable and difficult to humanly distinguish. +We need Zooko’s quadrangle. + +# Zooko’s Quadrangle + +Obviously you need to be able to reference human readable names on the +blockchain, which is the fourth corner of [Zooko’s triangle]. + +[Zooko’s triangle] : ./zookos_triangle.html + +# Location + +We want any identity to be able to send end to end encrypted messages to +any other identity, but we don’t want ten thousand scammers to be able +to spam an identity of which they know nothing other than that he can +send money over the internet, nor do we want to allow distributed denial +of service attacks on ordinary users (big peers can take care of +themselves). + +Universal Resource locators, urls, are not semantically or syntactically +distinct from Uniform Resource Identifiers that are wrapped around +methods for finding stuff on the internet, and methods for finding stuff +on the internet are wrapped around Uniform Resource Identifiers, but +this was designed for a smaller and more trusting world. Today, the +problem is not finding data, but rather preventing hostile and ill +willed people from finding data and from interfering with communication +by supplying falsified data So we need names that are rooted on +cryptographic foundations. + +If you have a system of naming that can securely identify that you are +getting the right connection and the right data, you can wrap location +around it and attach location to it ad hoc. SQL is a language for +generating on the fly ad hoc efficient methods for accessing data that +is specified in ways that bear a very indirect relationship to its +location. + +The equivalent of a web page in the cloud obviously has to have a +globally unique human readable name, being the website page and the +identifier on the website, which is associated with an network address +that hostile parties can find, to mount DDoS attack or sent the cops +around, and associated with public key, or chain of public keys. But we +do not want the equivalent capability to send messages to humans, or to +find them. Yet we want everyone to be able to talk privately to +everyone. + +Suppose someone is publishing information which he wants widely known. +He wants to cooperate with people who will cooperate with them, supply +them with good information that is arguably grounds to act in his +interest and their own. Well, often unpleasant people, or the +government, do not want that to happen. For example people cooperating +to build guns, and providing information on building guns. The +government might well prefer that people do not have guns. Or a wealthy +Chinese man wants to use the Chinese diaspora to move his assets to +where the party cannot get at those assets. Or someone simply has money, +and other people are hoping to scam him into giving them some, or give +him a hard time so that he pays them to go away, or just hate him +because he has money. Then he likely wants to pay for and operate the +server publishing that information disconnected from his tax number, his +real address, and a face that can be beaten in. So he wants the server +identity and network address widely known, but he does not want the +network address through which he makes payments for that server known to +anyone except the people he is paying for the server. + +In order to send a message directly to an identity, you are going to +need its network address (which might be intermediated by a peer) but we +don’t want to make the network address of every identity public. +Sometimes, often, the network address associated with an identity needs +to be a narrowly shared secret. + +We can fix distributed denial of service attacks by inserting a proof of +work demand in the first syn-ack of the three way connection setup +handshake (syn, ack-syn, ack), and the work has to be supplied in the +ack, before the server allocates any memory or performs any expensive +asymmetric cryptography operations. + +Most of the time a server is responding with a message that is intended +for human attention. This is OK, because client requesting the message +is under human control, so the human initiates. + +If the server could send web pages uninvited, that would be extremely +bad. + +So, should not be able to get a human readable message unless it is a +response to something, such as another human readable message. + +So how do we get things started? + +Everything is message/response, which makes a stream nature of TCP a bad +idea. And, being engineers, we are apt to break things down to messages, +which presupposes we can send any message to anyone in isolation, but a +message has to happen in a connection, and a connection may require a +relationship. + +Once we have a relationship, we can figure out a connection within this +relationship, and once we have a connection in the context of this +relationship (which may be many to many), then we can conceptualize +messages as isolated units within the context of that connection, within +the context of that relationship. + +We want groups of people to be able to securely communicate, which the +Jami UI does: Name of room is weakly durable shared secret – but the +people have to use this secret to find each other, which means that +their meeting point location has to be a publicly known network address, +and their meeting point location then knows all of their network +addresses. But UI for this is not urgent. Rather, we want buyers to be +able to find sellers, and buyers and sellers to acquire reputation. + +For reviews, creating reputation, we are going to need usenet like +distribution. For conversations regarding the transaction, email like +distribution. The client logs in with the people he is transacting with +to get and receive personal messages regarding the good or service, or +he logs in with his message server from time to time to get or receive +messages. This limits the number of people that can see the connection +between his network address and his public key. + +# Implementation + +signed +: anyone can check that some data is signed by key, and such data can + be passed around in a pool, usenet style. + +authenticated +: You got the data directly from an entity that has the key. You know + it came from that key, but cannot prove it to anyone else. + +access +: A key with authorization from another key does something. + +authorization +: a key is given authorization by a key with authority. + +authority +: A key with authority can give other keys authorization. Every key + has unlimited authority to do whatever it wants on its own + computers, and with its own reputation. It may grant other keys + authorization to access certain services on its computers and to + perform certain acts in the name of its reputation. + +We do not want the key on the server to be the master key that owns the +server name, because keys on servers are too easily stolen. So we want +it to be a key granted authority by the key that owns the server name. + +There are no end of cases that we will eventually need to handle where +one key grants some authority to another key, so we need a general +mechanism and general format for this, with the particular cases we are +now implementing being particular cases of this general mechanism and +general format. + +We will have authority to respond to automatic and anonymous queries, +analogous to hitting a web page, authority to receive crypto currency, +authority to promise goods and services in return for crypto currency +(which authorities will often belong to different keys), authority to +receive messages intended for human consumption, and authority to +authenticate messages from a human identity (which will typically belong +to the same key) + +And a data structure that associates a network address or rendezvous +server with a key, which data structure may be widely distributed, or +narrowly distributed. + +A data structure corresponds to a record or structure of records in a +database, we are talking about a way of synchronizing databases, and all +databases start as a human readable human writable text file. So, we +need a network database that has public information, and network +database that has private information. + +So, we need a collection of data akin to + +`/etc/hosts` +: public data, the broad consensus, agreed data known to the wider + community. + +`~/.ssh/known_hosts` +: privately known data about the community that cannot be widely + shared because others might not trust it, and you might not trust + others. You may want to share this with those you trust, and get it + from those you trust, but your set of people that you trust is + unlikely to agree with someone else’s and needs to be curated by a + human. + +`~/.ssh/config` +: And there is data you want to keep secret. + +Public data has to rest on a foundation of idiosyncratic data, which has +to rest on a foundation of secret data. If you do it the other way +around, as with peer review, then public data gets manipulated for +hostile purposes, as with certificate authorities and domain name +service. + +So we are going to start with idiosyncratic and narrowly shared contact +information in private databases. The typical operation is look up a +petname (guaranteed locally unique) find the controlling key for that +petname (probabilistically unique) and find a publicly accessible key +(probabilistically unique) which has the petname key as its root master +key. + +## Format for a key granting authority to a key + +### types + +We have, as always everywhere, additive and multiplicative types, with no general and universal way of ascertaining the type of data. You generally know from context. After all, if you know where to find the data, you probably know what the type is. + +Any universal type, or universal way of discovering the type, is apt to turn into a constraint, and people find, as with json, clever ways of working around that constraint, which break everything + +But situations do arise where one has to discover the type of an opaque record. + +In general, the type of value represented by a hash may be unknown. A communication channel is an indefinitely long sequence of records of unknown additive type. A file is a large object of unknown type, that is apt to be part of a big pile of objects of miscellaneous type. + +To open a connection, need protocol negotiation. + +Unix originally planned to have executable files identified in the directory field, but found it had far too many types of executable file, and started to supplement this with the file header. + +zip files, tar files, and tar.gz files are in unix identified to the user by the file name, but various utilities by the file header. + +So, we have a type that identifies type, and we have a type that consists of such an identifier, followed by the object so identified - which is the equivalent of unix's file header. + +We need a name for such a data structure It is not a datum, it is a way +of making data out of an unending string of bytes, it is a way of +identifying the data that bytes represent. It is a blob (Binary Large +OBject) beginning with a schema identifier. It is record with schema id. +It is a typed object. Having trouble naming it. + +OK, we will call a schema identifier with the fields it identifies +following a schere (SCHema REcord). It is a schema attached to the +record whose fields are implied by the schema. It can contain strings, +arbitrary precision integers, hashes, elliptic points, elliptic scalars, +and other scheres. A collection of scheres can only be parsed from left +to right, because if we do not know what to expect, we cannot know where +one ends and the next begins. + + +Well, that is OK for objects at rest, but protocol negotiation is a different animal + +We may well in future want to implement an extensive capability system, +so we have to have a data format that gives room for future extensions +to do all sorts of unforeseeable things – for example one key might want +to issue a transferable time limited right to access a particular file +to another key. + +A byte stream preceded by schema identifier that defines what those +bytes mean is not a byte string. It is a string of integers, scalars, +points, hashes, strings, and whatnot. The schema identifier tells us +what fields follow, thereby enabling us to parse them out an otherwise +endless stream of otherwise meaningless data. + +When one is dealing with scheres in C++ one is probably going to have a +pointer that could point to a variety of different structs, (additive type that +points to a variety of multiplicative types) which you would represent in +C++ as an `std::variant type`. + +Which implies that the party writing and the party reading have to have +agreement on the mapping between schemas and the integers identifying +schemas. When two people share data, the protocol they agree upon +implies a complete set of integers identifiers of a set of schemas, +which means we will frequently wind up issuing new protocols, and wind +up with a fair bit of protocol negotiation. + +Suppose we have a file just sitting by itself, not part of a connection. +Well, it just has to start with the schema id that identifies the +protocol it would have if it was part of a connection – so a protocol id +is itself a schema id, which like any schema id gives the mapping of the +schema ids whose scheres it contains. + +### Finding the network address of a master key, and the authority of the subkey that represents that master key at that network address + +Suppose someone has a high value reputation, and he wants anyone +anywhere to be able to connect to his server, and obtain an +authenticated connection, a connection that the person connecting knows +is controlled by the entity that has this reputation. + +He has the master secret corresponding to the public key connected to this +reputation, but because it is a high value secret, it is not on any +computer anywhere. It is written in the margin of a page in a bible kept +on his bookshelf, so cannot be used to authenticate the connection. + +His secret master key is used to sign another a public subkey, which resides on a +little used computer seldom if ever connected to the network, which +signs another public key on a computer generally connected to the +internet, which signs the public keys on his servers. When a client +computer connects to one of his servers, and asks for a connection +authenticated with this reputation, the client gets a connection +authenticated by a key signed by a key signed by a key signed by the key +whose secret is no longer in any computer, but is in the margin of a +bible on someone’s bookshelf. + +If the client computer has no copy of a certificate signed by this +master key, it cannot connect because it does not know the network +address. If it has a copy that testifies to the network address, but the +certificate is timed out or nonexistent, it asks for a connection +authenticated by the master key, and then, if it does not get a +connection authenticated by the master key, gets a certificate +authenticating a key communicated over the initial encrypted but +unauthenticated connection, and forms a connection authenticated by the +certified key. If it does not get either a certificate and +authentication by the certified key or authentication by the master key, +the connection fails, and it goes looking for a certificate. + +But how does the entity that signs its network address know what its +network address is? + +If the master contacts it over the internet, no problem. The master +knows network address at which he contacted it and has authority to tell +it – but that might be a merely local network address. How does random +computer find its network address? + +The master on his client that he controls probably knows the network +address of the server that he also controls, that being how he likely +controls it, but this is not guaranteed to also be the situation. + +If an entity wants to sign a network address so that others can contact +it, it is publishing its network address into a pool of network +addresses, so it already has other entities it can talk to and ask +“What is my IP?” So it tries some entities at random, asks for its IP +over the authenticated connection, and asks them to to open a connection +on its port number to that IP. + +Its prior is a high probability that they will all give the same answer +and few if any give a different answer, and a prior that all of them +will fail to call back or most of them will callback. It updates its +priors after each call, until it has a high probability that the +majority agree on its IP, and it is definitely contactable at this port, +or uncontactable at this port. + +## Finally, the format + +After long, long, long, discussion on the requirements for a format and +the meaning of the format: + +A signature consists of a schema id for signatures, followed by an +arbitrary schere, followed by the public key, and the two elliptic +scalars that form the Schnorr signature. The null schere, whose schema +id is zero, is permissible, in which case the signature is a proof that +the secret corresponding to this public key is known, which matters if +it is used in a multisignature. + +We also have schema ids for multisignatures – one for signature with two +keys with two distinct roles, and one for a variable number of keys with +symmetric and equivalent roles. + +When a schere is signed, it is public and goes into public shared data. +When it is merely communicated over an authenticated channel, the +recipient handles it as if signed by the sender, but cannot prove to +anyone else it originated from the sender, so stores the same data in +its private database, rather than its public database. + +Public information on network addresses and key authorities will be +signed, and that is signed implies it is public and should be widely +distributed. But often we do not want this information distributed, in +which case these data structures should be authenticated but not signed, +in which case it gets stored in the wallet and is not routinely and +automatically distributed. + +We will by default couple information on network addresses to +information on key authorization, so that we do not run into the DNS +fake network address problem (lack of certification) nor the CA key +repudiation problem (failure to get up to date certificates). + +But if authorization is distributed with network addresses, rather than +being provided by the authorized key on request, then the authorization +has to be signed by both the key authorized, and the key doing the +authorizing. We don’t want random scammers to be able to claim to be +the real power behind the power, neither do we want captured websites +hanging on to authorizations that have been obsoleted because of +capture. + +A network address will be signed by the key whose network address it is, +not by the key granting it authority. A key needs no authorization to +tell us where it can be contacted, and any key giving us the network +address of some other key would need evidence of authority to do so, +which would get us deep in the weeds. + +However, because the final link in the chain is jointly signed, it may +contain the network address as well. Or we may two scheres, one signed +by the last vertex in the chain giving authorization, and one signed +only by the final leaf giving the network address. The former would be +best for stable network addresses, the latter best for unstable network +addresses. Both should be permissible. + +When the network address changes, the key probably changes, and vice +versa, so they should usually and normally be distributed together. Or +maybe no contact information at all, implying that the owner of the +master key only wants to be contacted by people who have received +contact information by another, less widely shared channel. Maybe, as +with product or supplier reviews, he wants everyone to be able to check +his signature on data widely redistributed by someone else, but does not +want everyone who reads that widely redistributed data to be able to +send him messages. + +OK, the arbitrary signed schere in the case we are here discussing is an +authentication schere. Which is only meaningful as part of a signature, +so will only ever be hashed as part of a full authentication, so has no +hash rule as an isolated schere, so we can re-use schema ids, +authentication data being useless without being contained within a +signature. + +The authentication schema consists of the key that is being granted +authority, an arbitrary precision integer that represents a set of +flags, an authentication date as a multiple of 256 seconds which +signifies that the authentication supersedes all earlier +authentications, and also that the authentication does not take effect +until the indicated date. (If you have a bunch of servers with a bunch +of keys all representing one master key, you have to re-authenticate all +of them with the same start date), and an end date to encourage people +to reauthenticate every now and then. An end date earlier than the start +date is invalid, shall be rejected, and have no effect, except that an +end date of zero indicates the authentication remains valid till +superseded, till the client sees a certificate with a later start date. +A certificate with an end date some astronomical time into the future +may be rejected, or may be silently discarded and have no effect. Use an +end date of zero to represent “never expires”. A certificate with a +start date some unreasonable time into the future will not spread +through the network till its start date draws nigh. + +A key that has been authenticated has authority to grant the same +authority or a subset of the flags that have been set to it for another +key, with an end date less than or equal to its end date, and a start +date greater than or equal to its start date. + +0. Bit revokes all other keys, signifies that this chain of +authorizations invalidates all previous chains of authorizations from +the same root, where one chain is previous to the other if the start +times in its chain, considered as a Dewey number, are earlier than the +other. A chain that is identical except for the times gets invalidated +anyway, but this bit is a key revocation bit – when you don’t want some +*other* key trusted any more. Most of the time there will be one and +only one valid key chain for one root, and most of the time this bit +will be set. + +0. Bit indicates that this key may be used to authenticate a connection +as under the control of the entity at the root of the chain. +This subkey can do anything the master key can do, +except extend its timeout, or create subsubkeys with a timeout +beyond its own (the equivalent of gpg subkey, can sign, +authenticate, accept payment, whatever.) + +0. Bit is the signing bit. It indicates that this key can sign data on +behalf of this master key, sign as the entity at the root of the chain. +One typically signs data that will be delivered to the recipient +through an untrusted intermediary, as for example, downloading a +rhocoin wallet, or a peer making an assertion about the most recent +root or block of the blockchain. + +0. Bit indicates that this key may be used to make an offer in the +identity of the entity at the root of the chain. + +0. Bit indicates that this key may be used to accept crypto currency as +the entity at the root of the chain. An offer is likely to be made the +contactable key at the leaf of the chain which has no authority to +accept payment, which requests payment to an uncontactable key +closer to the root of the chain which does have authority to accept +payment. A payment request identifies the rhocoin public keys it +wants by an elliptic scalar, and the public key of the rhocoin is the +accepting key multiplied by that scalar. The payer therefore has +proof he paid that entity and is owed something, even if the money +goes astray. + +0. Bit indicates that more authorities follow, in the form of an ordered +sequence of arbitrary precision integers, terminated by zero, thus enabling +people to roll their own authorities, ad hoc. + +For authorities whose number is less than 64, the bitstring representation +and the list of integers representation are equivalent - we may provide a +long bit mask, or a zero terminated list of integers. Some +implementations may refuse to accept bitstrings longer than 64 bits, +generating a bad data exception. + +We will have a variety of contact information schemas. The contact +information needs to be signed by the key to be contacted, but that +turns out to have surprisingly messy logistics in getting things +rolling. When you are setting up a server, you want to both grant +authority to its secret key (which is what “Lets Encrypt” does), and +publish its network address which is what the DNS does. With “lets +encrypt” you publish the network address insecurely, then “lets +encrypt” insecurely finds a host key claiming the name at that network +address. Which system works only because of the good behavior of the +centralized authority authorizing domain name service. + +The entity that has the master secret somehow controls both machines. +That is, after all, what we want to prove to third parties. To generate +the necessary certificates, the machine without the master secret has to +have a connection to the machine that has the master secret. One +initiates a connection to the other, then they generate the necessary +certificates. The master at that point learns the slave public key and +proves to himself that network address works. But the machine with the +key does not necessarily know its own network address. The party +authorizing the key of machine being authorized *does* know it, and the +two machines can trust each other because under the control of the same +party. + +Whenever two parties communicate, they can verify the external network +addresses associated with the other’s key, but not the external network +address associated with their own key. And if we are talking +authorization, we have a trust relationship that can be used to prove +the network address to key at that network address. To avoid the key +revocation problem, it is easier and safer if network address +information is distributed with authorization information. + +The key actually used is authenticated by the master key through a chain +of certificates, which are normally gathered together in yet another +higher level schere, which to be valid must have a valid link in each +link in the chain. This higher level chain contains all the signatures, +plus network address information, but the individual links in the chain, +and the signed network address, are independently valid, and do not have +to be actually contiguous and in order in a chain to be useful. The +higher level schere is useful merely because it is sometimes convenient +to pack related data as one big ordered bundle, a pile of facts only +useful because they prove one fact, the beginning and end of the chain. + +If someone is relying on a chain of authorities, any key in that chain +can sign network address for itself or any descendants in the chain. But +this requires yet another schema, which should combine a grant of +authority with a network address. diff --git a/docs/images/nobody_know_you_are_a_dog.webp b/docs/images/nobody_know_you_are_a_dog.webp new file mode 100644 index 0000000..e87596e Binary files /dev/null and b/docs/images/nobody_know_you_are_a_dog.webp differ diff --git a/docs/images/postfix_cfg.png b/docs/images/postfix_cfg.png new file mode 100644 index 0000000..752269b Binary files /dev/null and b/docs/images/postfix_cfg.png differ diff --git a/docs/images/postfix_cfg1.png b/docs/images/postfix_cfg1.png new file mode 100644 index 0000000..41e8ec5 Binary files /dev/null and b/docs/images/postfix_cfg1.png differ diff --git a/docs/images/postfix_cfg1.webp b/docs/images/postfix_cfg1.webp new file mode 100644 index 0000000..d5e19b5 Binary files /dev/null and b/docs/images/postfix_cfg1.webp differ diff --git a/docs/images/postfix_cfg2.png b/docs/images/postfix_cfg2.png new file mode 100644 index 0000000..18420f9 Binary files /dev/null and b/docs/images/postfix_cfg2.png differ diff --git a/docs/images/postfix_cfg2.webp b/docs/images/postfix_cfg2.webp new file mode 100644 index 0000000..c80216a Binary files /dev/null and b/docs/images/postfix_cfg2.webp differ diff --git a/docs/images/postfix_cfg3.png b/docs/images/postfix_cfg3.png new file mode 100644 index 0000000..5174061 Binary files /dev/null and b/docs/images/postfix_cfg3.png differ diff --git a/docs/images/postfix_cfg3.webp b/docs/images/postfix_cfg3.webp new file mode 100644 index 0000000..a493e7a Binary files /dev/null and b/docs/images/postfix_cfg3.webp differ diff --git a/docs/images/working_mailersever.webp b/docs/images/working_mailersever.webp new file mode 100644 index 0000000..35011c4 Binary files /dev/null and b/docs/images/working_mailersever.webp differ diff --git a/docs/index.md b/docs/index.md new file mode 100644 index 0000000..287bdb2 --- /dev/null +++ b/docs/index.md @@ -0,0 +1,148 @@ +--- +title: How to Save the World +--- +I have almost completed an enormous design document for an uncensorable social network intended to contain a non evil scalable proof of stake currency, and I have a wallet that can generate secrets, but the wallet is missing no end of critical features – it is pre-pre alpha. When it is early pre alpha, I am going to publish it on Gitea, and call for assistance. + +Here is a link to one version of the [white paper](social_networking.html), focusing primarily on social media. (But though information wants to be free, programmers need to get paid.) + +Here is a link to [another version](white_paper.html) of the white paper, focusing primarily on money and getting rich by protecting capitalism from the state. + +# Speech and commerce + +As the internet goes, so goes the world. For freedom of speech to +exist, there must be freedom of speech on the internet, and if there is +freedom of speech on the internet, there is freedom of speech, for +governments will find it very hard to stop it. If freedom of information, +file sharing and open source code on the internet, then there is freedom +of information, if there is freedom of association on the internet, then +there is freedom of association – and, the big one, the one we have least, +the one under most severe threat, if there is freedom of commerce on the +internet … + +We can establish these freedoms by technological and business means +instead of political means. These means turned out to be more difficult +than expected in the heady days of the [cypherpunk](cypherpunk_program.html) +movement. + +To secure all these, we need the right software, software that +successfully applies the cryptographic tools that have been developed. + +Governments are getting worse, governments *always* get worse, +yet what is outside the government’s power is getting stronger. + +It is the nature of governments to always get worse over time, resulting +in them either collapsing or being bypassed by new forms of government. + +The cypherpunk program was that governments would be bypassed, as +organization moved to the internet, hidden behind cryptography. The +cypherpunk program died, yet lives – for China’s industrialization is +being organized through the VPNs of firms whose servers are located in the +cayman islands. These firms do transactions largely by trading each other’s +IOUs in private conversations rather than through regular bank +’t. Cypherpunks imagined that they would be living in tropical +paradises running businesses nominally located in tax havens. It has not +come true for them, but an increasing proportion of the world’s business +does work that way. + +In the cypherpunk vision, people of moderate wealth would escape the +power of government – unfortunately what is happening is merely +billionaires escaping the power of government. To revive and accomplish +the cypherpunk vision, we need to make these capabilities and methods more +widely available – available not just to the super rich but to the better +off middle class – not necessarily the ordinary middle class, but rather +the sort of middle class person who has a passport in more than one +country and does not need to show up at the office at 9AM every +morning. From thence it will eventually trickle down to the regular +middle class. + +At the same time as we see a billion people industrializing in an +industrialization run from islands on the internet, we also see a variety +of private use of force organizations also organized over the internet +popping up – thus for example the extortion operation against oil +companies in Nigeria was in part run over the internet from South Africa. +Somali pirates were largely eradicated by private security firms whose +home nation is far from clear. + +We are seeing entirely legal and government approved mercenaries, not +quite legal and sort of government approved mercenaries, illegal but +government tolerated militias and armed mosques, illegal distributors of +recreational chemicals very successfully resisting government power, and +assorted extortionists and terrorists. Yes, extortionists and terrorists +are bad things, but that people are ever less inclined to rely on +government provision of protection against them is a good thing. + +The power of states is increasing, in the sense that taxes and +regulation is increasing, that government ownership is increasing, that +large firms function by special privilege granted by the government to +those firms to the detriment of those less privileged – but at the same +time, that which is outside the power of the state is growing +stronger. It is a pattern that recurs every few hundred years, +leading to the renewal, or the collapse, of civilization. + +# Major concepts + +- PKI and SSL needs to be obsoleted and replaced. As Bruce + Schneier said in Secrets and Lies: 〝SSL is just simply a (very + slow) Diffie-Hellman key-exchange method. Digital certificates + provide no actual security for electronic commerce; it’s a complete sham〞 + + The underlying problem is that our mental name handling + mechanism is intended for the relatively small social groups of the + Neolithic. True names fail when we attempt to scale to the internet. + The current name system is rooted in governmental and quasi + governmental entities, who use this power to gently encourage + nominally private institutions to censor the internet. Similarly, the + encryption system of https allows the government to intercept any + website with a man in the middle attack. To fix this, we need a + name system rooted in the blockchain, with encryption rooted in + Zooko’s triangle, as with crypto currency + +- [Zooko’s triangle](zookos_triangle.html), The solution is an ID system based on Zooko’s +triangle, allowing everyone to have as many IDs as they want, but +no one else can forge their IDs, ensuring that each identity has a +corresponding public key, thus making end to end encryption easy. +These identities may correspond to people you can instant message, +or web sites, particularly secure web sites that require logon, such +as banks, or indeed any service. Thus, they also correspond to +bank accounts, that work like Swiss numbered bank account, in that your identity is a secret. +- Protocol negotiation at the levels equivalent to TCP and UDP, and +default encryption and authentication at those levels, as with ssh. +- Ability to introduce new protocols and upgrade old protocols without central coordination, just as Zooko allows us to introduce +new identities without central coordination. Central authority is failing, has become an obstacle, instead of the fast way to get things done. +- File sharing with upload credits. +- Single signon, buddy list user interface for web page logon. +- Messaging system integrated with single signon – message +authentication, all messages end to end encrypted. Zooko identity +means yurls, which means a problem in getting people onto our buddy list. +- Money transfer integrated with instant messaging. + - Money transfer uses ripple. + - Each money transfer creates a record of accompanying obligation, + equivalent record on both sides of the transaction. You can put put + money in a message, and for the recipient to get it out of the + message, he has to sign a receipt that says this money is for such + and such, and he took the money – a receipt that only the person who + sent the money and the person who received the money can read, and + any financial intermediaries cannot read, though they will need + proof that the requested receipt exists, without them being able to + read what the receipt is for. The records provide a basis for + generating reputation of Zooko based identities. + +This web page is intended to keep track of the various technologies +needed to implement liberty on the internet. There are lots of them, and +they are all fairly complex and many of them subtle and very difficult to +understand, so this web page will always be severely incomplete. Right now +it is almost totally incomplete, I have just got started listing stuff: + +# Details + +This list severely incomplete, when finished will be at least a screen’s +worth, probably several screens. + +- [how to build an operating system that is largely immune to viruses, Trojans and spyware](safe_operating_system.html) +- [how to stop + phishing and browser session hijacking, how to do browser security + right.](how_browser_security_should_be_done.html) +- [How to do VPNs right](how_to_do_VPNs.html) +- [How to prevent malware](safe_operating_system.html) +- [The cypherpunk program](cypherpunk_program.html) +- [Replacing TCP and UDP](replacing_TCP.html) diff --git a/docs/katex.min.css b/docs/katex.min.css new file mode 100644 index 0000000..c0cd145 --- /dev/null +++ b/docs/katex.min.css @@ -0,0 +1 @@ +@font-face{font-family:KaTeX_AMS;src:url(fonts/KaTeX_AMS-Regular.woff2) format("woff2"),url(fonts/KaTeX_AMS-Regular.woff) format("woff"),url(fonts/KaTeX_AMS-Regular.ttf) format("truetype");font-weight:400;font-style:normal}@font-face{font-family:KaTeX_Caligraphic;src:url(fonts/KaTeX_Caligraphic-Bold.woff2) format("woff2"),url(fonts/KaTeX_Caligraphic-Bold.woff) format("woff"),url(fonts/KaTeX_Caligraphic-Bold.ttf) format("truetype");font-weight:700;font-style:normal}@font-face{font-family:KaTeX_Caligraphic;src:url(fonts/KaTeX_Caligraphic-Regular.woff2) format("woff2"),url(fonts/KaTeX_Caligraphic-Regular.woff) format("woff"),url(fonts/KaTeX_Caligraphic-Regular.ttf) format("truetype");font-weight:400;font-style:normal}@font-face{font-family:KaTeX_Fraktur;src:url(fonts/KaTeX_Fraktur-Bold.woff2) format("woff2"),url(fonts/KaTeX_Fraktur-Bold.woff) format("woff"),url(fonts/KaTeX_Fraktur-Bold.ttf) format("truetype");font-weight:700;font-style:normal}@font-face{font-family:KaTeX_Fraktur;src:url(fonts/KaTeX_Fraktur-Regular.woff2) format("woff2"),url(fonts/KaTeX_Fraktur-Regular.woff) format("woff"),url(fonts/KaTeX_Fraktur-Regular.ttf) format("truetype");font-weight:400;font-style:normal}@font-face{font-family:KaTeX_Main;src:url(fonts/KaTeX_Main-Bold.woff2) format("woff2"),url(fonts/KaTeX_Main-Bold.woff) format("woff"),url(fonts/KaTeX_Main-Bold.ttf) format("truetype");font-weight:700;font-style:normal}@font-face{font-family:KaTeX_Main;src:url(fonts/KaTeX_Main-BoldItalic.woff2) format("woff2"),url(fonts/KaTeX_Main-BoldItalic.woff) format("woff"),url(fonts/KaTeX_Main-BoldItalic.ttf) format("truetype");font-weight:700;font-style:italic}@font-face{font-family:KaTeX_Main;src:url(fonts/KaTeX_Main-Italic.woff2) format("woff2"),url(fonts/KaTeX_Main-Italic.woff) format("woff"),url(fonts/KaTeX_Main-Italic.ttf) format("truetype");font-weight:400;font-style:italic}@font-face{font-family:KaTeX_Main;src:url(fonts/KaTeX_Main-Regular.woff2) format("woff2"),url(fonts/KaTeX_Main-Regular.woff) format("woff"),url(fonts/KaTeX_Main-Regular.ttf) format("truetype");font-weight:400;font-style:normal}@font-face{font-family:KaTeX_Math;src:url(fonts/KaTeX_Math-BoldItalic.woff2) format("woff2"),url(fonts/KaTeX_Math-BoldItalic.woff) format("woff"),url(fonts/KaTeX_Math-BoldItalic.ttf) format("truetype");font-weight:700;font-style:italic}@font-face{font-family:KaTeX_Math;src:url(fonts/KaTeX_Math-Italic.woff2) format("woff2"),url(fonts/KaTeX_Math-Italic.woff) format("woff"),url(fonts/KaTeX_Math-Italic.ttf) format("truetype");font-weight:400;font-style:italic}@font-face{font-family:"KaTeX_SansSerif";src:url(fonts/KaTeX_SansSerif-Bold.woff2) format("woff2"),url(fonts/KaTeX_SansSerif-Bold.woff) format("woff"),url(fonts/KaTeX_SansSerif-Bold.ttf) format("truetype");font-weight:700;font-style:normal}@font-face{font-family:"KaTeX_SansSerif";src:url(fonts/KaTeX_SansSerif-Italic.woff2) format("woff2"),url(fonts/KaTeX_SansSerif-Italic.woff) format("woff"),url(fonts/KaTeX_SansSerif-Italic.ttf) format("truetype");font-weight:400;font-style:italic}@font-face{font-family:"KaTeX_SansSerif";src:url(fonts/KaTeX_SansSerif-Regular.woff2) format("woff2"),url(fonts/KaTeX_SansSerif-Regular.woff) format("woff"),url(fonts/KaTeX_SansSerif-Regular.ttf) format("truetype");font-weight:400;font-style:normal}@font-face{font-family:KaTeX_Script;src:url(fonts/KaTeX_Script-Regular.woff2) format("woff2"),url(fonts/KaTeX_Script-Regular.woff) format("woff"),url(fonts/KaTeX_Script-Regular.ttf) format("truetype");font-weight:400;font-style:normal}@font-face{font-family:KaTeX_Size1;src:url(fonts/KaTeX_Size1-Regular.woff2) format("woff2"),url(fonts/KaTeX_Size1-Regular.woff) format("woff"),url(fonts/KaTeX_Size1-Regular.ttf) format("truetype");font-weight:400;font-style:normal}@font-face{font-family:KaTeX_Size2;src:url(fonts/KaTeX_Size2-Regular.woff2) format("woff2"),url(fonts/KaTeX_Size2-Regular.woff) format("woff"),url(fonts/KaTeX_Size2-Regular.ttf) format("truetype");font-weight:400;font-style:normal}@font-face{font-family:KaTeX_Size3;src:url(fonts/KaTeX_Size3-Regular.woff2) format("woff2"),url(fonts/KaTeX_Size3-Regular.woff) format("woff"),url(fonts/KaTeX_Size3-Regular.ttf) format("truetype");font-weight:400;font-style:normal}@font-face{font-family:KaTeX_Size4;src:url(fonts/KaTeX_Size4-Regular.woff2) format("woff2"),url(fonts/KaTeX_Size4-Regular.woff) format("woff"),url(fonts/KaTeX_Size4-Regular.ttf) format("truetype");font-weight:400;font-style:normal}@font-face{font-family:KaTeX_Typewriter;src:url(fonts/KaTeX_Typewriter-Regular.woff2) format("woff2"),url(fonts/KaTeX_Typewriter-Regular.woff) format("woff"),url(fonts/KaTeX_Typewriter-Regular.ttf) format("truetype");font-weight:400;font-style:normal}.katex{font:normal 1.21em KaTeX_Main,Times New Roman,serif;line-height:1.2;text-indent:0;text-rendering:auto}.katex *{-ms-high-contrast-adjust:none!important}.katex .katex-version:after{content:"0.11.1"}.katex .katex-mathml{position:absolute;clip:rect(1px,1px,1px,1px);padding:0;border:0;height:1px;width:1px;overflow:hidden}.katex .katex-html>.newline{display:block}.katex .base{position:relative;white-space:nowrap;width:min-content}.katex .base,.katex .strut{display:inline-block}.katex .textbf{font-weight:700}.katex .textit{font-style:italic}.katex .textrm{font-family:KaTeX_Main}.katex .textsf{font-family:KaTeX_SansSerif}.katex .texttt{font-family:KaTeX_Typewriter}.katex .mathdefault{font-family:KaTeX_Math;font-style:italic}.katex .mathit{font-family:KaTeX_Main;font-style:italic}.katex .mathrm{font-style:normal}.katex .mathbf{font-family:KaTeX_Main;font-weight:700}.katex .boldsymbol{font-family:KaTeX_Math;font-weight:700;font-style:italic}.katex .amsrm,.katex .mathbb,.katex .textbb{font-family:KaTeX_AMS}.katex .mathcal{font-family:KaTeX_Caligraphic}.katex .mathfrak,.katex .textfrak{font-family:KaTeX_Fraktur}.katex .mathtt{font-family:KaTeX_Typewriter}.katex .mathscr,.katex .textscr{font-family:KaTeX_Script}.katex .mathsf,.katex .textsf{font-family:KaTeX_SansSerif}.katex .mathboldsf,.katex .textboldsf{font-family:KaTeX_SansSerif;font-weight:700}.katex .mathitsf,.katex .textitsf{font-family:KaTeX_SansSerif;font-style:italic}.katex .mainrm{font-family:KaTeX_Main;font-style:normal}.katex .vlist-t{display:inline-table;table-layout:fixed}.katex .vlist-r{display:table-row}.katex .vlist{display:table-cell;vertical-align:bottom;position:relative}.katex .vlist>span{display:block;height:0;position:relative}.katex .vlist>span>span{display:inline-block}.katex .vlist>span>.pstrut{overflow:hidden;width:0}.katex .vlist-t2{margin-right:-2px}.katex .vlist-s{display:table-cell;vertical-align:bottom;font-size:1px;width:2px;min-width:2px}.katex .msupsub{text-align:left}.katex .mfrac>span>span{text-align:center}.katex .mfrac .frac-line{display:inline-block;width:100%;border-bottom-style:solid}.katex .hdashline,.katex .hline,.katex .mfrac .frac-line,.katex .overline .overline-line,.katex .rule,.katex .underline .underline-line{min-height:1px}.katex .mspace{display:inline-block}.katex .clap,.katex .llap,.katex .rlap{width:0;position:relative}.katex .clap>.inner,.katex .llap>.inner,.katex .rlap>.inner{position:absolute}.katex .clap>.fix,.katex .llap>.fix,.katex .rlap>.fix{display:inline-block}.katex .llap>.inner{right:0}.katex .clap>.inner,.katex .rlap>.inner{left:0}.katex .clap>.inner>span{margin-left:-50%;margin-right:50%}.katex .rule{display:inline-block;border:0 solid;position:relative}.katex .hline,.katex .overline .overline-line,.katex .underline .underline-line{display:inline-block;width:100%;border-bottom-style:solid}.katex .hdashline{display:inline-block;width:100%;border-bottom-style:dashed}.katex .sqrt>.root{margin-left:.27777778em;margin-right:-.55555556em}.katex .fontsize-ensurer.reset-size1.size1,.katex .sizing.reset-size1.size1{font-size:1em}.katex .fontsize-ensurer.reset-size1.size2,.katex .sizing.reset-size1.size2{font-size:1.2em}.katex .fontsize-ensurer.reset-size1.size3,.katex .sizing.reset-size1.size3{font-size:1.4em}.katex .fontsize-ensurer.reset-size1.size4,.katex .sizing.reset-size1.size4{font-size:1.6em}.katex .fontsize-ensurer.reset-size1.size5,.katex .sizing.reset-size1.size5{font-size:1.8em}.katex .fontsize-ensurer.reset-size1.size6,.katex .sizing.reset-size1.size6{font-size:2em}.katex .fontsize-ensurer.reset-size1.size7,.katex .sizing.reset-size1.size7{font-size:2.4em}.katex .fontsize-ensurer.reset-size1.size8,.katex .sizing.reset-size1.size8{font-size:2.88em}.katex .fontsize-ensurer.reset-size1.size9,.katex .sizing.reset-size1.size9{font-size:3.456em}.katex .fontsize-ensurer.reset-size1.size10,.katex .sizing.reset-size1.size10{font-size:4.148em}.katex .fontsize-ensurer.reset-size1.size11,.katex .sizing.reset-size1.size11{font-size:4.976em}.katex .fontsize-ensurer.reset-size2.size1,.katex .sizing.reset-size2.size1{font-size:.83333333em}.katex .fontsize-ensurer.reset-size2.size2,.katex .sizing.reset-size2.size2{font-size:1em}.katex .fontsize-ensurer.reset-size2.size3,.katex .sizing.reset-size2.size3{font-size:1.16666667em}.katex .fontsize-ensurer.reset-size2.size4,.katex .sizing.reset-size2.size4{font-size:1.33333333em}.katex .fontsize-ensurer.reset-size2.size5,.katex .sizing.reset-size2.size5{font-size:1.5em}.katex .fontsize-ensurer.reset-size2.size6,.katex .sizing.reset-size2.size6{font-size:1.66666667em}.katex .fontsize-ensurer.reset-size2.size7,.katex .sizing.reset-size2.size7{font-size:2em}.katex .fontsize-ensurer.reset-size2.size8,.katex .sizing.reset-size2.size8{font-size:2.4em}.katex .fontsize-ensurer.reset-size2.size9,.katex .sizing.reset-size2.size9{font-size:2.88em}.katex .fontsize-ensurer.reset-size2.size10,.katex .sizing.reset-size2.size10{font-size:3.45666667em}.katex .fontsize-ensurer.reset-size2.size11,.katex .sizing.reset-size2.size11{font-size:4.14666667em}.katex .fontsize-ensurer.reset-size3.size1,.katex .sizing.reset-size3.size1{font-size:.71428571em}.katex .fontsize-ensurer.reset-size3.size2,.katex .sizing.reset-size3.size2{font-size:.85714286em}.katex .fontsize-ensurer.reset-size3.size3,.katex .sizing.reset-size3.size3{font-size:1em}.katex .fontsize-ensurer.reset-size3.size4,.katex .sizing.reset-size3.size4{font-size:1.14285714em}.katex .fontsize-ensurer.reset-size3.size5,.katex .sizing.reset-size3.size5{font-size:1.28571429em}.katex .fontsize-ensurer.reset-size3.size6,.katex .sizing.reset-size3.size6{font-size:1.42857143em}.katex .fontsize-ensurer.reset-size3.size7,.katex .sizing.reset-size3.size7{font-size:1.71428571em}.katex .fontsize-ensurer.reset-size3.size8,.katex .sizing.reset-size3.size8{font-size:2.05714286em}.katex .fontsize-ensurer.reset-size3.size9,.katex .sizing.reset-size3.size9{font-size:2.46857143em}.katex .fontsize-ensurer.reset-size3.size10,.katex .sizing.reset-size3.size10{font-size:2.96285714em}.katex .fontsize-ensurer.reset-size3.size11,.katex .sizing.reset-size3.size11{font-size:3.55428571em}.katex .fontsize-ensurer.reset-size4.size1,.katex .sizing.reset-size4.size1{font-size:.625em}.katex .fontsize-ensurer.reset-size4.size2,.katex .sizing.reset-size4.size2{font-size:.75em}.katex .fontsize-ensurer.reset-size4.size3,.katex .sizing.reset-size4.size3{font-size:.875em}.katex .fontsize-ensurer.reset-size4.size4,.katex .sizing.reset-size4.size4{font-size:1em}.katex .fontsize-ensurer.reset-size4.size5,.katex .sizing.reset-size4.size5{font-size:1.125em}.katex .fontsize-ensurer.reset-size4.size6,.katex .sizing.reset-size4.size6{font-size:1.25em}.katex .fontsize-ensurer.reset-size4.size7,.katex .sizing.reset-size4.size7{font-size:1.5em}.katex .fontsize-ensurer.reset-size4.size8,.katex .sizing.reset-size4.size8{font-size:1.8em}.katex .fontsize-ensurer.reset-size4.size9,.katex .sizing.reset-size4.size9{font-size:2.16em}.katex .fontsize-ensurer.reset-size4.size10,.katex .sizing.reset-size4.size10{font-size:2.5925em}.katex .fontsize-ensurer.reset-size4.size11,.katex .sizing.reset-size4.size11{font-size:3.11em}.katex .fontsize-ensurer.reset-size5.size1,.katex .sizing.reset-size5.size1{font-size:.55555556em}.katex .fontsize-ensurer.reset-size5.size2,.katex .sizing.reset-size5.size2{font-size:.66666667em}.katex .fontsize-ensurer.reset-size5.size3,.katex .sizing.reset-size5.size3{font-size:.77777778em}.katex .fontsize-ensurer.reset-size5.size4,.katex .sizing.reset-size5.size4{font-size:.88888889em}.katex .fontsize-ensurer.reset-size5.size5,.katex .sizing.reset-size5.size5{font-size:1em}.katex .fontsize-ensurer.reset-size5.size6,.katex .sizing.reset-size5.size6{font-size:1.11111111em}.katex .fontsize-ensurer.reset-size5.size7,.katex .sizing.reset-size5.size7{font-size:1.33333333em}.katex .fontsize-ensurer.reset-size5.size8,.katex .sizing.reset-size5.size8{font-size:1.6em}.katex .fontsize-ensurer.reset-size5.size9,.katex .sizing.reset-size5.size9{font-size:1.92em}.katex .fontsize-ensurer.reset-size5.size10,.katex .sizing.reset-size5.size10{font-size:2.30444444em}.katex .fontsize-ensurer.reset-size5.size11,.katex .sizing.reset-size5.size11{font-size:2.76444444em}.katex .fontsize-ensurer.reset-size6.size1,.katex .sizing.reset-size6.size1{font-size:.5em}.katex .fontsize-ensurer.reset-size6.size2,.katex .sizing.reset-size6.size2{font-size:.6em}.katex .fontsize-ensurer.reset-size6.size3,.katex .sizing.reset-size6.size3{font-size:.7em}.katex .fontsize-ensurer.reset-size6.size4,.katex .sizing.reset-size6.size4{font-size:.8em}.katex .fontsize-ensurer.reset-size6.size5,.katex .sizing.reset-size6.size5{font-size:.9em}.katex .fontsize-ensurer.reset-size6.size6,.katex .sizing.reset-size6.size6{font-size:1em}.katex .fontsize-ensurer.reset-size6.size7,.katex .sizing.reset-size6.size7{font-size:1.2em}.katex .fontsize-ensurer.reset-size6.size8,.katex .sizing.reset-size6.size8{font-size:1.44em}.katex .fontsize-ensurer.reset-size6.size9,.katex .sizing.reset-size6.size9{font-size:1.728em}.katex .fontsize-ensurer.reset-size6.size10,.katex .sizing.reset-size6.size10{font-size:2.074em}.katex .fontsize-ensurer.reset-size6.size11,.katex .sizing.reset-size6.size11{font-size:2.488em}.katex .fontsize-ensurer.reset-size7.size1,.katex .sizing.reset-size7.size1{font-size:.41666667em}.katex .fontsize-ensurer.reset-size7.size2,.katex .sizing.reset-size7.size2{font-size:.5em}.katex .fontsize-ensurer.reset-size7.size3,.katex .sizing.reset-size7.size3{font-size:.58333333em}.katex .fontsize-ensurer.reset-size7.size4,.katex .sizing.reset-size7.size4{font-size:.66666667em}.katex .fontsize-ensurer.reset-size7.size5,.katex .sizing.reset-size7.size5{font-size:.75em}.katex .fontsize-ensurer.reset-size7.size6,.katex .sizing.reset-size7.size6{font-size:.83333333em}.katex .fontsize-ensurer.reset-size7.size7,.katex .sizing.reset-size7.size7{font-size:1em}.katex .fontsize-ensurer.reset-size7.size8,.katex .sizing.reset-size7.size8{font-size:1.2em}.katex .fontsize-ensurer.reset-size7.size9,.katex .sizing.reset-size7.size9{font-size:1.44em}.katex .fontsize-ensurer.reset-size7.size10,.katex .sizing.reset-size7.size10{font-size:1.72833333em}.katex .fontsize-ensurer.reset-size7.size11,.katex .sizing.reset-size7.size11{font-size:2.07333333em}.katex .fontsize-ensurer.reset-size8.size1,.katex .sizing.reset-size8.size1{font-size:.34722222em}.katex .fontsize-ensurer.reset-size8.size2,.katex .sizing.reset-size8.size2{font-size:.41666667em}.katex .fontsize-ensurer.reset-size8.size3,.katex .sizing.reset-size8.size3{font-size:.48611111em}.katex .fontsize-ensurer.reset-size8.size4,.katex .sizing.reset-size8.size4{font-size:.55555556em}.katex .fontsize-ensurer.reset-size8.size5,.katex .sizing.reset-size8.size5{font-size:.625em}.katex .fontsize-ensurer.reset-size8.size6,.katex .sizing.reset-size8.size6{font-size:.69444444em}.katex .fontsize-ensurer.reset-size8.size7,.katex .sizing.reset-size8.size7{font-size:.83333333em}.katex .fontsize-ensurer.reset-size8.size8,.katex .sizing.reset-size8.size8{font-size:1em}.katex .fontsize-ensurer.reset-size8.size9,.katex .sizing.reset-size8.size9{font-size:1.2em}.katex .fontsize-ensurer.reset-size8.size10,.katex .sizing.reset-size8.size10{font-size:1.44027778em}.katex .fontsize-ensurer.reset-size8.size11,.katex .sizing.reset-size8.size11{font-size:1.72777778em}.katex .fontsize-ensurer.reset-size9.size1,.katex .sizing.reset-size9.size1{font-size:.28935185em}.katex .fontsize-ensurer.reset-size9.size2,.katex .sizing.reset-size9.size2{font-size:.34722222em}.katex .fontsize-ensurer.reset-size9.size3,.katex .sizing.reset-size9.size3{font-size:.40509259em}.katex .fontsize-ensurer.reset-size9.size4,.katex .sizing.reset-size9.size4{font-size:.46296296em}.katex .fontsize-ensurer.reset-size9.size5,.katex .sizing.reset-size9.size5{font-size:.52083333em}.katex .fontsize-ensurer.reset-size9.size6,.katex .sizing.reset-size9.size6{font-size:.5787037em}.katex .fontsize-ensurer.reset-size9.size7,.katex .sizing.reset-size9.size7{font-size:.69444444em}.katex .fontsize-ensurer.reset-size9.size8,.katex .sizing.reset-size9.size8{font-size:.83333333em}.katex .fontsize-ensurer.reset-size9.size9,.katex .sizing.reset-size9.size9{font-size:1em}.katex .fontsize-ensurer.reset-size9.size10,.katex .sizing.reset-size9.size10{font-size:1.20023148em}.katex .fontsize-ensurer.reset-size9.size11,.katex .sizing.reset-size9.size11{font-size:1.43981481em}.katex .fontsize-ensurer.reset-size10.size1,.katex .sizing.reset-size10.size1{font-size:.24108004em}.katex .fontsize-ensurer.reset-size10.size2,.katex .sizing.reset-size10.size2{font-size:.28929605em}.katex .fontsize-ensurer.reset-size10.size3,.katex .sizing.reset-size10.size3{font-size:.33751205em}.katex .fontsize-ensurer.reset-size10.size4,.katex .sizing.reset-size10.size4{font-size:.38572806em}.katex .fontsize-ensurer.reset-size10.size5,.katex .sizing.reset-size10.size5{font-size:.43394407em}.katex .fontsize-ensurer.reset-size10.size6,.katex .sizing.reset-size10.size6{font-size:.48216008em}.katex .fontsize-ensurer.reset-size10.size7,.katex .sizing.reset-size10.size7{font-size:.57859209em}.katex .fontsize-ensurer.reset-size10.size8,.katex .sizing.reset-size10.size8{font-size:.69431051em}.katex .fontsize-ensurer.reset-size10.size9,.katex .sizing.reset-size10.size9{font-size:.83317261em}.katex .fontsize-ensurer.reset-size10.size10,.katex .sizing.reset-size10.size10{font-size:1em}.katex .fontsize-ensurer.reset-size10.size11,.katex .sizing.reset-size10.size11{font-size:1.19961427em}.katex .fontsize-ensurer.reset-size11.size1,.katex .sizing.reset-size11.size1{font-size:.20096463em}.katex .fontsize-ensurer.reset-size11.size2,.katex .sizing.reset-size11.size2{font-size:.24115756em}.katex .fontsize-ensurer.reset-size11.size3,.katex .sizing.reset-size11.size3{font-size:.28135048em}.katex .fontsize-ensurer.reset-size11.size4,.katex .sizing.reset-size11.size4{font-size:.32154341em}.katex .fontsize-ensurer.reset-size11.size5,.katex .sizing.reset-size11.size5{font-size:.36173633em}.katex .fontsize-ensurer.reset-size11.size6,.katex .sizing.reset-size11.size6{font-size:.40192926em}.katex .fontsize-ensurer.reset-size11.size7,.katex .sizing.reset-size11.size7{font-size:.48231511em}.katex .fontsize-ensurer.reset-size11.size8,.katex .sizing.reset-size11.size8{font-size:.57877814em}.katex .fontsize-ensurer.reset-size11.size9,.katex .sizing.reset-size11.size9{font-size:.69453376em}.katex .fontsize-ensurer.reset-size11.size10,.katex .sizing.reset-size11.size10{font-size:.83360129em}.katex .fontsize-ensurer.reset-size11.size11,.katex .sizing.reset-size11.size11{font-size:1em}.katex .delimsizing.size1{font-family:KaTeX_Size1}.katex .delimsizing.size2{font-family:KaTeX_Size2}.katex .delimsizing.size3{font-family:KaTeX_Size3}.katex .delimsizing.size4{font-family:KaTeX_Size4}.katex .delimsizing.mult .delim-size1>span{font-family:KaTeX_Size1}.katex .delimsizing.mult .delim-size4>span{font-family:KaTeX_Size4}.katex .nulldelimiter{display:inline-block;width:.12em}.katex .delimcenter,.katex .op-symbol{position:relative}.katex .op-symbol.small-op{font-family:KaTeX_Size1}.katex .op-symbol.large-op{font-family:KaTeX_Size2}.katex .op-limits>.vlist-t{text-align:center}.katex .accent>.vlist-t{text-align:center}.katex .accent .accent-body{position:relative}.katex .accent .accent-body:not(.accent-full){width:0}.katex .overlay{display:block}.katex .mtable .vertical-separator{display:inline-block;min-width:1px}.katex .mtable .arraycolsep{display:inline-block}.katex .mtable .col-align-c>.vlist-t{text-align:center}.katex .mtable .col-align-l>.vlist-t{text-align:left}.katex .mtable .col-align-r>.vlist-t{text-align:right}.katex .svg-align{text-align:left}.katex svg{display:block;position:absolute;width:100%;height:inherit;fill:currentColor;stroke:currentColor;fill-rule:nonzero;fill-opacity:1;stroke-width:1;stroke-linecap:butt;stroke-linejoin:miter;stroke-miterlimit:4;stroke-dasharray:none;stroke-dashoffset:0;stroke-opacity:1}.katex svg path{stroke:none}.katex img{border-style:none;min-width:0;min-height:0;max-width:none;max-height:none}.katex .stretchy{width:100%;display:block;position:relative;overflow:hidden}.katex .stretchy:after,.katex .stretchy:before{content:""}.katex .hide-tail{width:100%;position:relative;overflow:hidden}.katex .halfarrow-left{position:absolute;left:0;width:50.2%;overflow:hidden}.katex .halfarrow-right{position:absolute;right:0;width:50.2%;overflow:hidden}.katex .brace-left{position:absolute;left:0;width:25.1%;overflow:hidden}.katex .brace-center{position:absolute;left:25%;width:50%;overflow:hidden}.katex .brace-right{position:absolute;right:0;width:25.1%;overflow:hidden}.katex .x-arrow-pad{padding:0 .5em}.katex .mover,.katex .munder,.katex .x-arrow{text-align:center}.katex .boxpad{padding:0 .3em}.katex .fbox,.katex .fcolorbox{box-sizing:border-box;border:.04em solid}.katex .cancel-pad{padding:0 .2em}.katex .cancel-lap{margin-left:-.2em;margin-right:-.2em}.katex .sout{border-bottom-style:solid;border-bottom-width:.08em}.katex-display{display:block;margin:1em 0;text-align:center}.katex-display>.katex{display:block;text-align:center;white-space:nowrap}.katex-display>.katex>.katex-html{display:block;position:relative}.katex-display>.katex>.katex-html>.tag{position:absolute;right:0}.katex-display.leqno>.katex>.katex-html>.tag{left:0;right:auto}.katex-display.fleqn>.katex{text-align:left} diff --git a/docs/katex.min.js b/docs/katex.min.js new file mode 100644 index 0000000..906ce12 --- /dev/null +++ b/docs/katex.min.js @@ -0,0 +1 @@ +!function(t,e){"object"==typeof exports&&"object"==typeof module?module.exports=e():"function"==typeof define&&define.amd?define([],e):"object"==typeof exports?exports.katex=e():t.katex=e()}("undefined"!=typeof self?self:this,function(){return function(t){var e={};function r(a){if(e[a])return e[a].exports;var n=e[a]={i:a,l:!1,exports:{}};return t[a].call(n.exports,n,n.exports,r),n.l=!0,n.exports}return r.m=t,r.c=e,r.d=function(t,e,a){r.o(t,e)||Object.defineProperty(t,e,{enumerable:!0,get:a})},r.r=function(t){"undefined"!=typeof Symbol&&Symbol.toStringTag&&Object.defineProperty(t,Symbol.toStringTag,{value:"Module"}),Object.defineProperty(t,"__esModule",{value:!0})},r.t=function(t,e){if(1&e&&(t=r(t)),8&e)return t;if(4&e&&"object"==typeof t&&t&&t.__esModule)return t;var a=Object.create(null);if(r.r(a),Object.defineProperty(a,"default",{enumerable:!0,value:t}),2&e&&"string"!=typeof t)for(var n in t)r.d(a,n,function(e){return t[e]}.bind(null,n));return a},r.n=function(t){var e=t&&t.__esModule?function(){return t.default}:function(){return t};return r.d(e,"a",e),e},r.o=function(t,e){return Object.prototype.hasOwnProperty.call(t,e)},r.p="",r(r.s=1)}([function(t,e,r){},function(t,e,r){"use strict";r.r(e);r(0);var a=function(){function t(t,e,r){this.lexer=void 0,this.start=void 0,this.end=void 0,this.lexer=t,this.start=e,this.end=r}return t.range=function(e,r){return r?e&&e.loc&&r.loc&&e.loc.lexer===r.loc.lexer?new t(e.loc.lexer,e.loc.start,r.loc.end):null:e&&e.loc},t}(),n=function(){function t(t,e){this.text=void 0,this.loc=void 0,this.text=t,this.loc=e}return t.prototype.range=function(e,r){return new t(r,a.range(this,e))},t}(),i=function t(e,r){this.position=void 0;var a,n="KaTeX parse error: "+e,i=r&&r.loc;if(i&&i.start<=i.end){var o=i.lexer.input;a=i.start;var s=i.end;a===o.length?n+=" at end of input: ":n+=" at position "+(a+1)+": ";var h=o.slice(a,s).replace(/[^]/g,"$&\u0332");n+=(a>15?"\u2026"+o.slice(a-15,a):o.slice(0,a))+h+(s+15":">","<":"<",'"':""","'":"'"},l=/[&><"']/g;var m=function t(e){return"ordgroup"===e.type?1===e.body.length?t(e.body[0]):e:"color"===e.type?1===e.body.length?t(e.body[0]):e:"font"===e.type?t(e.body):e},c={contains:function(t,e){return-1!==t.indexOf(e)},deflt:function(t,e){return void 0===t?e:t},escape:function(t){return String(t).replace(l,function(t){return h[t]})},hyphenate:function(t){return t.replace(s,"-$1").toLowerCase()},getBaseElem:m,isCharacterBox:function(t){var e=m(t);return"mathord"===e.type||"textord"===e.type||"atom"===e.type},protocolFromUrl:function(t){var e=/^\s*([^\\\/#]*?)(?::|�*58|�*3a)/i.exec(t);return null!=e?e[1]:"_relative"}},u=function(){function t(t){this.displayMode=void 0,this.output=void 0,this.leqno=void 0,this.fleqn=void 0,this.throwOnError=void 0,this.errorColor=void 0,this.macros=void 0,this.minRuleThickness=void 0,this.colorIsTextColor=void 0,this.strict=void 0,this.trust=void 0,this.maxSize=void 0,this.maxExpand=void 0,t=t||{},this.displayMode=c.deflt(t.displayMode,!1),this.output=c.deflt(t.output,"htmlAndMathml"),this.leqno=c.deflt(t.leqno,!1),this.fleqn=c.deflt(t.fleqn,!1),this.throwOnError=c.deflt(t.throwOnError,!0),this.errorColor=c.deflt(t.errorColor,"#cc0000"),this.macros=t.macros||{},this.minRuleThickness=Math.max(0,c.deflt(t.minRuleThickness,0)),this.colorIsTextColor=c.deflt(t.colorIsTextColor,!1),this.strict=c.deflt(t.strict,"warn"),this.trust=c.deflt(t.trust,!1),this.maxSize=Math.max(0,c.deflt(t.maxSize,1/0)),this.maxExpand=Math.max(0,c.deflt(t.maxExpand,1e3))}var e=t.prototype;return e.reportNonstrict=function(t,e,r){var a=this.strict;if("function"==typeof a&&(a=a(t,e,r)),a&&"ignore"!==a){if(!0===a||"error"===a)throw new o("LaTeX-incompatible input and strict mode is set to 'error': "+e+" ["+t+"]",r);"warn"===a?"undefined"!=typeof console&&console.warn("LaTeX-incompatible input and strict mode is set to 'warn': "+e+" ["+t+"]"):"undefined"!=typeof console&&console.warn("LaTeX-incompatible input and strict mode is set to unrecognized '"+a+"': "+e+" ["+t+"]")}},e.useStrictBehavior=function(t,e,r){var a=this.strict;if("function"==typeof a)try{a=a(t,e,r)}catch(t){a="error"}return!(!a||"ignore"===a)&&(!0===a||"error"===a||("warn"===a?("undefined"!=typeof console&&console.warn("LaTeX-incompatible input and strict mode is set to 'warn': "+e+" ["+t+"]"),!1):("undefined"!=typeof console&&console.warn("LaTeX-incompatible input and strict mode is set to unrecognized '"+a+"': "+e+" ["+t+"]"),!1)))},e.isTrusted=function(t){t.url&&!t.protocol&&(t.protocol=c.protocolFromUrl(t.url));var e="function"==typeof this.trust?this.trust(t):this.trust;return Boolean(e)},t}(),p=function(){function t(t,e,r){this.id=void 0,this.size=void 0,this.cramped=void 0,this.id=t,this.size=e,this.cramped=r}var e=t.prototype;return e.sup=function(){return d[f[this.id]]},e.sub=function(){return d[g[this.id]]},e.fracNum=function(){return d[x[this.id]]},e.fracDen=function(){return d[v[this.id]]},e.cramp=function(){return d[b[this.id]]},e.text=function(){return d[y[this.id]]},e.isTight=function(){return this.size>=2},t}(),d=[new p(0,0,!1),new p(1,0,!0),new p(2,1,!1),new p(3,1,!0),new p(4,2,!1),new p(5,2,!0),new p(6,3,!1),new p(7,3,!0)],f=[4,5,4,5,6,7,6,7],g=[5,5,5,5,7,7,7,7],x=[2,3,4,5,6,7,6,7],v=[3,3,5,5,7,7,7,7],b=[1,1,3,3,5,5,7,7],y=[0,1,2,3,2,3,2,3],w={DISPLAY:d[0],TEXT:d[2],SCRIPT:d[4],SCRIPTSCRIPT:d[6]},k=[{name:"latin",blocks:[[256,591],[768,879]]},{name:"cyrillic",blocks:[[1024,1279]]},{name:"brahmic",blocks:[[2304,4255]]},{name:"georgian",blocks:[[4256,4351]]},{name:"cjk",blocks:[[12288,12543],[19968,40879],[65280,65376]]},{name:"hangul",blocks:[[44032,55215]]}];var S=[];function M(t){for(var e=0;e=S[e]&&t<=S[e+1])return!0;return!1}k.forEach(function(t){return t.blocks.forEach(function(t){return S.push.apply(S,t)})});var z={doubleleftarrow:"M262 157\nl10-10c34-36 62.7-77 86-123 3.3-8 5-13.3 5-16 0-5.3-6.7-8-20-8-7.3\n 0-12.2.5-14.5 1.5-2.3 1-4.8 4.5-7.5 10.5-49.3 97.3-121.7 169.3-217 216-28\n 14-57.3 25-88 33-6.7 2-11 3.8-13 5.5-2 1.7-3 4.2-3 7.5s1 5.8 3 7.5\nc2 1.7 6.3 3.5 13 5.5 68 17.3 128.2 47.8 180.5 91.5 52.3 43.7 93.8 96.2 124.5\n 157.5 9.3 8 15.3 12.3 18 13h6c12-.7 18-4 18-10 0-2-1.7-7-5-15-23.3-46-52-87\n-86-123l-10-10h399738v-40H218c328 0 0 0 0 0l-10-8c-26.7-20-65.7-43-117-69 2.7\n-2 6-3.7 10-5 36.7-16 72.3-37.3 107-64l10-8h399782v-40z\nm8 0v40h399730v-40zm0 194v40h399730v-40z",doublerightarrow:"M399738 392l\n-10 10c-34 36-62.7 77-86 123-3.3 8-5 13.3-5 16 0 5.3 6.7 8 20 8 7.3 0 12.2-.5\n 14.5-1.5 2.3-1 4.8-4.5 7.5-10.5 49.3-97.3 121.7-169.3 217-216 28-14 57.3-25 88\n-33 6.7-2 11-3.8 13-5.5 2-1.7 3-4.2 3-7.5s-1-5.8-3-7.5c-2-1.7-6.3-3.5-13-5.5-68\n-17.3-128.2-47.8-180.5-91.5-52.3-43.7-93.8-96.2-124.5-157.5-9.3-8-15.3-12.3-18\n-13h-6c-12 .7-18 4-18 10 0 2 1.7 7 5 15 23.3 46 52 87 86 123l10 10H0v40h399782\nc-328 0 0 0 0 0l10 8c26.7 20 65.7 43 117 69-2.7 2-6 3.7-10 5-36.7 16-72.3 37.3\n-107 64l-10 8H0v40zM0 157v40h399730v-40zm0 194v40h399730v-40z",leftarrow:"M400000 241H110l3-3c68.7-52.7 113.7-120\n 135-202 4-14.7 6-23 6-25 0-7.3-7-11-21-11-8 0-13.2.8-15.5 2.5-2.3 1.7-4.2 5.8\n-5.5 12.5-1.3 4.7-2.7 10.3-4 17-12 48.7-34.8 92-68.5 130S65.3 228.3 18 247\nc-10 4-16 7.7-18 11 0 8.7 6 14.3 18 17 47.3 18.7 87.8 47 121.5 85S196 441.3 208\n 490c.7 2 1.3 5 2 9s1.2 6.7 1.5 8c.3 1.3 1 3.3 2 6s2.2 4.5 3.5 5.5c1.3 1 3.3\n 1.8 6 2.5s6 1 10 1c14 0 21-3.7 21-11 0-2-2-10.3-6-25-20-79.3-65-146.7-135-202\n l-3-3h399890zM100 241v40h399900v-40z",leftbrace:"M6 548l-6-6v-35l6-11c56-104 135.3-181.3 238-232 57.3-28.7 117\n-45 179-50h399577v120H403c-43.3 7-81 15-113 26-100.7 33-179.7 91-237 174-2.7\n 5-6 9-10 13-.7 1-7.3 1-20 1H6z",leftbraceunder:"M0 6l6-6h17c12.688 0 19.313.3 20 1 4 4 7.313 8.3 10 13\n 35.313 51.3 80.813 93.8 136.5 127.5 55.688 33.7 117.188 55.8 184.5 66.5.688\n 0 2 .3 4 1 18.688 2.7 76 4.3 172 5h399450v120H429l-6-1c-124.688-8-235-61.7\n-331-161C60.687 138.7 32.312 99.3 7 54L0 41V6z",leftgroup:"M400000 80\nH435C64 80 168.3 229.4 21 260c-5.9 1.2-18 0-18 0-2 0-3-1-3-3v-38C76 61 257 0\n 435 0h399565z",leftgroupunder:"M400000 262\nH435C64 262 168.3 112.6 21 82c-5.9-1.2-18 0-18 0-2 0-3 1-3 3v38c76 158 257 219\n 435 219h399565z",leftharpoon:"M0 267c.7 5.3 3 10 7 14h399993v-40H93c3.3\n-3.3 10.2-9.5 20.5-18.5s17.8-15.8 22.5-20.5c50.7-52 88-110.3 112-175 4-11.3 5\n-18.3 3-21-1.3-4-7.3-6-18-6-8 0-13 .7-15 2s-4.7 6.7-8 16c-42 98.7-107.3 174.7\n-196 228-6.7 4.7-10.7 8-12 10-1.3 2-2 5.7-2 11zm100-26v40h399900v-40z",leftharpoonplus:"M0 267c.7 5.3 3 10 7 14h399993v-40H93c3.3-3.3 10.2-9.5\n 20.5-18.5s17.8-15.8 22.5-20.5c50.7-52 88-110.3 112-175 4-11.3 5-18.3 3-21-1.3\n-4-7.3-6-18-6-8 0-13 .7-15 2s-4.7 6.7-8 16c-42 98.7-107.3 174.7-196 228-6.7 4.7\n-10.7 8-12 10-1.3 2-2 5.7-2 11zm100-26v40h399900v-40zM0 435v40h400000v-40z\nm0 0v40h400000v-40z",leftharpoondown:"M7 241c-4 4-6.333 8.667-7 14 0 5.333.667 9 2 11s5.333\n 5.333 12 10c90.667 54 156 130 196 228 3.333 10.667 6.333 16.333 9 17 2 .667 5\n 1 9 1h5c10.667 0 16.667-2 18-6 2-2.667 1-9.667-3-21-32-87.333-82.667-157.667\n-152-211l-3-3h399907v-40zM93 281 H400000 v-40L7 241z",leftharpoondownplus:"M7 435c-4 4-6.3 8.7-7 14 0 5.3.7 9 2 11s5.3 5.3 12\n 10c90.7 54 156 130 196 228 3.3 10.7 6.3 16.3 9 17 2 .7 5 1 9 1h5c10.7 0 16.7\n-2 18-6 2-2.7 1-9.7-3-21-32-87.3-82.7-157.7-152-211l-3-3h399907v-40H7zm93 0\nv40h399900v-40zM0 241v40h399900v-40zm0 0v40h399900v-40z",lefthook:"M400000 281 H103s-33-11.2-61-33.5S0 197.3 0 164s14.2-61.2 42.5\n-83.5C70.8 58.2 104 47 142 47 c16.7 0 25 6.7 25 20 0 12-8.7 18.7-26 20-40 3.3\n-68.7 15.7-86 37-10 12-15 25.3-15 40 0 22.7 9.8 40.7 29.5 54 19.7 13.3 43.5 21\n 71.5 23h399859zM103 281v-40h399897v40z",leftlinesegment:"M40 281 V428 H0 V94 H40 V241 H400000 v40z\nM40 281 V428 H0 V94 H40 V241 H400000 v40z",leftmapsto:"M40 281 V448H0V74H40V241H400000v40z\nM40 281 V448H0V74H40V241H400000v40z",leftToFrom:"M0 147h400000v40H0zm0 214c68 40 115.7 95.7 143 167h22c15.3 0 23\n-.3 23-1 0-1.3-5.3-13.7-16-37-18-35.3-41.3-69-70-101l-7-8h399905v-40H95l7-8\nc28.7-32 52-65.7 70-101 10.7-23.3 16-35.7 16-37 0-.7-7.7-1-23-1h-22C115.7 265.3\n 68 321 0 361zm0-174v-40h399900v40zm100 154v40h399900v-40z",longequal:"M0 50 h400000 v40H0z m0 194h40000v40H0z\nM0 50 h400000 v40H0z m0 194h40000v40H0z",midbrace:"M200428 334\nc-100.7-8.3-195.3-44-280-108-55.3-42-101.7-93-139-153l-9-14c-2.7 4-5.7 8.7-9 14\n-53.3 86.7-123.7 153-211 199-66.7 36-137.3 56.3-212 62H0V214h199568c178.3-11.7\n 311.7-78.3 403-201 6-8 9.7-12 11-12 .7-.7 6.7-1 18-1s17.3.3 18 1c1.3 0 5 4 11\n 12 44.7 59.3 101.3 106.3 170 141s145.3 54.3 229 60h199572v120z",midbraceunder:"M199572 214\nc100.7 8.3 195.3 44 280 108 55.3 42 101.7 93 139 153l9 14c2.7-4 5.7-8.7 9-14\n 53.3-86.7 123.7-153 211-199 66.7-36 137.3-56.3 212-62h199568v120H200432c-178.3\n 11.7-311.7 78.3-403 201-6 8-9.7 12-11 12-.7.7-6.7 1-18 1s-17.3-.3-18-1c-1.3 0\n-5-4-11-12-44.7-59.3-101.3-106.3-170-141s-145.3-54.3-229-60H0V214z",oiintSize1:"M512.6 71.6c272.6 0 320.3 106.8 320.3 178.2 0 70.8-47.7 177.6\n-320.3 177.6S193.1 320.6 193.1 249.8c0-71.4 46.9-178.2 319.5-178.2z\nm368.1 178.2c0-86.4-60.9-215.4-368.1-215.4-306.4 0-367.3 129-367.3 215.4 0 85.8\n60.9 214.8 367.3 214.8 307.2 0 368.1-129 368.1-214.8z",oiintSize2:"M757.8 100.1c384.7 0 451.1 137.6 451.1 230 0 91.3-66.4 228.8\n-451.1 228.8-386.3 0-452.7-137.5-452.7-228.8 0-92.4 66.4-230 452.7-230z\nm502.4 230c0-111.2-82.4-277.2-502.4-277.2s-504 166-504 277.2\nc0 110 84 276 504 276s502.4-166 502.4-276z",oiiintSize1:"M681.4 71.6c408.9 0 480.5 106.8 480.5 178.2 0 70.8-71.6 177.6\n-480.5 177.6S202.1 320.6 202.1 249.8c0-71.4 70.5-178.2 479.3-178.2z\nm525.8 178.2c0-86.4-86.8-215.4-525.7-215.4-437.9 0-524.7 129-524.7 215.4 0\n85.8 86.8 214.8 524.7 214.8 438.9 0 525.7-129 525.7-214.8z",oiiintSize2:"M1021.2 53c603.6 0 707.8 165.8 707.8 277.2 0 110-104.2 275.8\n-707.8 275.8-606 0-710.2-165.8-710.2-275.8C311 218.8 415.2 53 1021.2 53z\nm770.4 277.1c0-131.2-126.4-327.6-770.5-327.6S248.4 198.9 248.4 330.1\nc0 130 128.8 326.4 772.7 326.4s770.5-196.4 770.5-326.4z",rightarrow:"M0 241v40h399891c-47.3 35.3-84 78-110 128\n-16.7 32-27.7 63.7-33 95 0 1.3-.2 2.7-.5 4-.3 1.3-.5 2.3-.5 3 0 7.3 6.7 11 20\n 11 8 0 13.2-.8 15.5-2.5 2.3-1.7 4.2-5.5 5.5-11.5 2-13.3 5.7-27 11-41 14.7-44.7\n 39-84.5 73-119.5s73.7-60.2 119-75.5c6-2 9-5.7 9-11s-3-9-9-11c-45.3-15.3-85\n-40.5-119-75.5s-58.3-74.8-73-119.5c-4.7-14-8.3-27.3-11-40-1.3-6.7-3.2-10.8-5.5\n-12.5-2.3-1.7-7.5-2.5-15.5-2.5-14 0-21 3.7-21 11 0 2 2 10.3 6 25 20.7 83.3 67\n 151.7 139 205zm0 0v40h399900v-40z",rightbrace:"M400000 542l\n-6 6h-17c-12.7 0-19.3-.3-20-1-4-4-7.3-8.3-10-13-35.3-51.3-80.8-93.8-136.5-127.5\ns-117.2-55.8-184.5-66.5c-.7 0-2-.3-4-1-18.7-2.7-76-4.3-172-5H0V214h399571l6 1\nc124.7 8 235 61.7 331 161 31.3 33.3 59.7 72.7 85 118l7 13v35z",rightbraceunder:"M399994 0l6 6v35l-6 11c-56 104-135.3 181.3-238 232-57.3\n 28.7-117 45-179 50H-300V214h399897c43.3-7 81-15 113-26 100.7-33 179.7-91 237\n-174 2.7-5 6-9 10-13 .7-1 7.3-1 20-1h17z",rightgroup:"M0 80h399565c371 0 266.7 149.4 414 180 5.9 1.2 18 0 18 0 2 0\n 3-1 3-3v-38c-76-158-257-219-435-219H0z",rightgroupunder:"M0 262h399565c371 0 266.7-149.4 414-180 5.9-1.2 18 0 18\n 0 2 0 3 1 3 3v38c-76 158-257 219-435 219H0z",rightharpoon:"M0 241v40h399993c4.7-4.7 7-9.3 7-14 0-9.3\n-3.7-15.3-11-18-92.7-56.7-159-133.7-199-231-3.3-9.3-6-14.7-8-16-2-1.3-7-2-15-2\n-10.7 0-16.7 2-18 6-2 2.7-1 9.7 3 21 15.3 42 36.7 81.8 64 119.5 27.3 37.7 58\n 69.2 92 94.5zm0 0v40h399900v-40z",rightharpoonplus:"M0 241v40h399993c4.7-4.7 7-9.3 7-14 0-9.3-3.7-15.3-11\n-18-92.7-56.7-159-133.7-199-231-3.3-9.3-6-14.7-8-16-2-1.3-7-2-15-2-10.7 0-16.7\n 2-18 6-2 2.7-1 9.7 3 21 15.3 42 36.7 81.8 64 119.5 27.3 37.7 58 69.2 92 94.5z\nm0 0v40h399900v-40z m100 194v40h399900v-40zm0 0v40h399900v-40z",rightharpoondown:"M399747 511c0 7.3 6.7 11 20 11 8 0 13-.8 15-2.5s4.7-6.8\n 8-15.5c40-94 99.3-166.3 178-217 13.3-8 20.3-12.3 21-13 5.3-3.3 8.5-5.8 9.5\n-7.5 1-1.7 1.5-5.2 1.5-10.5s-2.3-10.3-7-15H0v40h399908c-34 25.3-64.7 57-92 95\n-27.3 38-48.7 77.7-64 119-3.3 8.7-5 14-5 16zM0 241v40h399900v-40z",rightharpoondownplus:"M399747 705c0 7.3 6.7 11 20 11 8 0 13-.8\n 15-2.5s4.7-6.8 8-15.5c40-94 99.3-166.3 178-217 13.3-8 20.3-12.3 21-13 5.3-3.3\n 8.5-5.8 9.5-7.5 1-1.7 1.5-5.2 1.5-10.5s-2.3-10.3-7-15H0v40h399908c-34 25.3\n-64.7 57-92 95-27.3 38-48.7 77.7-64 119-3.3 8.7-5 14-5 16zM0 435v40h399900v-40z\nm0-194v40h400000v-40zm0 0v40h400000v-40z",righthook:"M399859 241c-764 0 0 0 0 0 40-3.3 68.7-15.7 86-37 10-12 15-25.3\n 15-40 0-22.7-9.8-40.7-29.5-54-19.7-13.3-43.5-21-71.5-23-17.3-1.3-26-8-26-20 0\n-13.3 8.7-20 26-20 38 0 71 11.2 99 33.5 0 0 7 5.6 21 16.7 14 11.2 21 33.5 21\n 66.8s-14 61.2-42 83.5c-28 22.3-61 33.5-99 33.5L0 241z M0 281v-40h399859v40z",rightlinesegment:"M399960 241 V94 h40 V428 h-40 V281 H0 v-40z\nM399960 241 V94 h40 V428 h-40 V281 H0 v-40z",rightToFrom:"M400000 167c-70.7-42-118-97.7-142-167h-23c-15.3 0-23 .3-23\n 1 0 1.3 5.3 13.7 16 37 18 35.3 41.3 69 70 101l7 8H0v40h399905l-7 8c-28.7 32\n-52 65.7-70 101-10.7 23.3-16 35.7-16 37 0 .7 7.7 1 23 1h23c24-69.3 71.3-125 142\n-167z M100 147v40h399900v-40zM0 341v40h399900v-40z",twoheadleftarrow:"M0 167c68 40\n 115.7 95.7 143 167h22c15.3 0 23-.3 23-1 0-1.3-5.3-13.7-16-37-18-35.3-41.3-69\n-70-101l-7-8h125l9 7c50.7 39.3 85 86 103 140h46c0-4.7-6.3-18.7-19-42-18-35.3\n-40-67.3-66-96l-9-9h399716v-40H284l9-9c26-28.7 48-60.7 66-96 12.7-23.333 19\n-37.333 19-42h-46c-18 54-52.3 100.7-103 140l-9 7H95l7-8c28.7-32 52-65.7 70-101\n 10.7-23.333 16-35.7 16-37 0-.7-7.7-1-23-1h-22C115.7 71.3 68 127 0 167z",twoheadrightarrow:"M400000 167\nc-68-40-115.7-95.7-143-167h-22c-15.3 0-23 .3-23 1 0 1.3 5.3 13.7 16 37 18 35.3\n 41.3 69 70 101l7 8h-125l-9-7c-50.7-39.3-85-86-103-140h-46c0 4.7 6.3 18.7 19 42\n 18 35.3 40 67.3 66 96l9 9H0v40h399716l-9 9c-26 28.7-48 60.7-66 96-12.7 23.333\n-19 37.333-19 42h46c18-54 52.3-100.7 103-140l9-7h125l-7 8c-28.7 32-52 65.7-70\n 101-10.7 23.333-16 35.7-16 37 0 .7 7.7 1 23 1h22c27.3-71.3 75-127 143-167z",tilde1:"M200 55.538c-77 0-168 73.953-177 73.953-3 0-7\n-2.175-9-5.437L2 97c-1-2-2-4-2-6 0-4 2-7 5-9l20-12C116 12 171 0 207 0c86 0\n 114 68 191 68 78 0 168-68 177-68 4 0 7 2 9 5l12 19c1 2.175 2 4.35 2 6.525 0\n 4.35-2 7.613-5 9.788l-19 13.05c-92 63.077-116.937 75.308-183 76.128\n-68.267.847-113-73.952-191-73.952z",tilde2:"M344 55.266c-142 0-300.638 81.316-311.5 86.418\n-8.01 3.762-22.5 10.91-23.5 5.562L1 120c-1-2-1-3-1-4 0-5 3-9 8-10l18.4-9C160.9\n 31.9 283 0 358 0c148 0 188 122 331 122s314-97 326-97c4 0 8 2 10 7l7 21.114\nc1 2.14 1 3.21 1 4.28 0 5.347-3 9.626-7 10.696l-22.3 12.622C852.6 158.372 751\n 181.476 676 181.476c-149 0-189-126.21-332-126.21z",tilde3:"M786 59C457 59 32 175.242 13 175.242c-6 0-10-3.457\n-11-10.37L.15 138c-1-7 3-12 10-13l19.2-6.4C378.4 40.7 634.3 0 804.3 0c337 0\n 411.8 157 746.8 157 328 0 754-112 773-112 5 0 10 3 11 9l1 14.075c1 8.066-.697\n 16.595-6.697 17.492l-21.052 7.31c-367.9 98.146-609.15 122.696-778.15 122.696\n -338 0-409-156.573-744-156.573z",tilde4:"M786 58C457 58 32 177.487 13 177.487c-6 0-10-3.345\n-11-10.035L.15 143c-1-7 3-12 10-13l22-6.7C381.2 35 637.15 0 807.15 0c337 0 409\n 177 744 177 328 0 754-127 773-127 5 0 10 3 11 9l1 14.794c1 7.805-3 13.38-9\n 14.495l-20.7 5.574c-366.85 99.79-607.3 139.372-776.3 139.372-338 0-409\n -175.236-744-175.236z",vec:"M377 20c0-5.333 1.833-10 5.5-14S391 0 397 0c4.667 0 8.667 1.667 12 5\n3.333 2.667 6.667 9 10 19 6.667 24.667 20.333 43.667 41 57 7.333 4.667 11\n10.667 11 18 0 6-1 10-3 12s-6.667 5-14 9c-28.667 14.667-53.667 35.667-75 63\n-1.333 1.333-3.167 3.5-5.5 6.5s-4 4.833-5 5.5c-1 .667-2.5 1.333-4.5 2s-4.333 1\n-7 1c-4.667 0-9.167-1.833-13.5-5.5S337 184 337 178c0-12.667 15.667-32.333 47-59\nH213l-171-1c-8.667-6-13-12.333-13-19 0-4.667 4.333-11.333 13-20h359\nc-16-25.333-24-45-24-59z",widehat1:"M529 0h5l519 115c5 1 9 5 9 10 0 1-1 2-1 3l-4 22\nc-1 5-5 9-11 9h-2L532 67 19 159h-2c-5 0-9-4-11-9l-5-22c-1-6 2-12 8-13z",widehat2:"M1181 0h2l1171 176c6 0 10 5 10 11l-2 23c-1 6-5 10\n-11 10h-1L1182 67 15 220h-1c-6 0-10-4-11-10l-2-23c-1-6 4-11 10-11z",widehat3:"M1181 0h2l1171 236c6 0 10 5 10 11l-2 23c-1 6-5 10\n-11 10h-1L1182 67 15 280h-1c-6 0-10-4-11-10l-2-23c-1-6 4-11 10-11z",widehat4:"M1181 0h2l1171 296c6 0 10 5 10 11l-2 23c-1 6-5 10\n-11 10h-1L1182 67 15 340h-1c-6 0-10-4-11-10l-2-23c-1-6 4-11 10-11z",widecheck1:"M529,159h5l519,-115c5,-1,9,-5,9,-10c0,-1,-1,-2,-1,-3l-4,-22c-1,\n-5,-5,-9,-11,-9h-2l-512,92l-513,-92h-2c-5,0,-9,4,-11,9l-5,22c-1,6,2,12,8,13z",widecheck2:"M1181,220h2l1171,-176c6,0,10,-5,10,-11l-2,-23c-1,-6,-5,-10,\n-11,-10h-1l-1168,153l-1167,-153h-1c-6,0,-10,4,-11,10l-2,23c-1,6,4,11,10,11z",widecheck3:"M1181,280h2l1171,-236c6,0,10,-5,10,-11l-2,-23c-1,-6,-5,-10,\n-11,-10h-1l-1168,213l-1167,-213h-1c-6,0,-10,4,-11,10l-2,23c-1,6,4,11,10,11z",widecheck4:"M1181,340h2l1171,-296c6,0,10,-5,10,-11l-2,-23c-1,-6,-5,-10,\n-11,-10h-1l-1168,273l-1167,-273h-1c-6,0,-10,4,-11,10l-2,23c-1,6,4,11,10,11z",baraboveleftarrow:"M400000 620h-399890l3 -3c68.7 -52.7 113.7 -120 135 -202\nc4 -14.7 6 -23 6 -25c0 -7.3 -7 -11 -21 -11c-8 0 -13.2 0.8 -15.5 2.5\nc-2.3 1.7 -4.2 5.8 -5.5 12.5c-1.3 4.7 -2.7 10.3 -4 17c-12 48.7 -34.8 92 -68.5 130\ns-74.2 66.3 -121.5 85c-10 4 -16 7.7 -18 11c0 8.7 6 14.3 18 17c47.3 18.7 87.8 47\n121.5 85s56.5 81.3 68.5 130c0.7 2 1.3 5 2 9s1.2 6.7 1.5 8c0.3 1.3 1 3.3 2 6\ns2.2 4.5 3.5 5.5c1.3 1 3.3 1.8 6 2.5s6 1 10 1c14 0 21 -3.7 21 -11\nc0 -2 -2 -10.3 -6 -25c-20 -79.3 -65 -146.7 -135 -202l-3 -3h399890z\nM100 620v40h399900v-40z M0 241v40h399900v-40zM0 241v40h399900v-40z",rightarrowabovebar:"M0 241v40h399891c-47.3 35.3-84 78-110 128-16.7 32\n-27.7 63.7-33 95 0 1.3-.2 2.7-.5 4-.3 1.3-.5 2.3-.5 3 0 7.3 6.7 11 20 11 8 0\n13.2-.8 15.5-2.5 2.3-1.7 4.2-5.5 5.5-11.5 2-13.3 5.7-27 11-41 14.7-44.7 39\n-84.5 73-119.5s73.7-60.2 119-75.5c6-2 9-5.7 9-11s-3-9-9-11c-45.3-15.3-85-40.5\n-119-75.5s-58.3-74.8-73-119.5c-4.7-14-8.3-27.3-11-40-1.3-6.7-3.2-10.8-5.5\n-12.5-2.3-1.7-7.5-2.5-15.5-2.5-14 0-21 3.7-21 11 0 2 2 10.3 6 25 20.7 83.3 67\n151.7 139 205zm96 379h399894v40H0zm0 0h399904v40H0z",baraboveshortleftharpoon:"M507,435c-4,4,-6.3,8.7,-7,14c0,5.3,0.7,9,2,11\nc1.3,2,5.3,5.3,12,10c90.7,54,156,130,196,228c3.3,10.7,6.3,16.3,9,17\nc2,0.7,5,1,9,1c0,0,5,0,5,0c10.7,0,16.7,-2,18,-6c2,-2.7,1,-9.7,-3,-21\nc-32,-87.3,-82.7,-157.7,-152,-211c0,0,-3,-3,-3,-3l399351,0l0,-40\nc-398570,0,-399437,0,-399437,0z M593 435 v40 H399500 v-40z\nM0 281 v-40 H399908 v40z M0 281 v-40 H399908 v40z",rightharpoonaboveshortbar:"M0,241 l0,40c399126,0,399993,0,399993,0\nc4.7,-4.7,7,-9.3,7,-14c0,-9.3,-3.7,-15.3,-11,-18c-92.7,-56.7,-159,-133.7,-199,\n-231c-3.3,-9.3,-6,-14.7,-8,-16c-2,-1.3,-7,-2,-15,-2c-10.7,0,-16.7,2,-18,6\nc-2,2.7,-1,9.7,3,21c15.3,42,36.7,81.8,64,119.5c27.3,37.7,58,69.2,92,94.5z\nM0 241 v40 H399908 v-40z M0 475 v-40 H399500 v40z M0 475 v-40 H399500 v40z",shortbaraboveleftharpoon:"M7,435c-4,4,-6.3,8.7,-7,14c0,5.3,0.7,9,2,11\nc1.3,2,5.3,5.3,12,10c90.7,54,156,130,196,228c3.3,10.7,6.3,16.3,9,17c2,0.7,5,1,9,\n1c0,0,5,0,5,0c10.7,0,16.7,-2,18,-6c2,-2.7,1,-9.7,-3,-21c-32,-87.3,-82.7,-157.7,\n-152,-211c0,0,-3,-3,-3,-3l399907,0l0,-40c-399126,0,-399993,0,-399993,0z\nM93 435 v40 H400000 v-40z M500 241 v40 H400000 v-40z M500 241 v40 H400000 v-40z",shortrightharpoonabovebar:"M53,241l0,40c398570,0,399437,0,399437,0\nc4.7,-4.7,7,-9.3,7,-14c0,-9.3,-3.7,-15.3,-11,-18c-92.7,-56.7,-159,-133.7,-199,\n-231c-3.3,-9.3,-6,-14.7,-8,-16c-2,-1.3,-7,-2,-15,-2c-10.7,0,-16.7,2,-18,6\nc-2,2.7,-1,9.7,3,21c15.3,42,36.7,81.8,64,119.5c27.3,37.7,58,69.2,92,94.5z\nM500 241 v40 H399408 v-40z M500 435 v40 H400000 v-40z"},A=function(){function t(t){this.children=void 0,this.classes=void 0,this.height=void 0,this.depth=void 0,this.maxFontSize=void 0,this.style=void 0,this.children=t,this.classes=[],this.height=0,this.depth=0,this.maxFontSize=0,this.style={}}var e=t.prototype;return e.hasClass=function(t){return c.contains(this.classes,t)},e.toNode=function(){for(var t=document.createDocumentFragment(),e=0;e"},N=function(){function t(t,e,r,a){this.children=void 0,this.attributes=void 0,this.classes=void 0,this.height=void 0,this.depth=void 0,this.width=void 0,this.maxFontSize=void 0,this.style=void 0,B.call(this,t,r,a),this.children=e||[]}var e=t.prototype;return e.setAttribute=function(t,e){this.attributes[t]=e},e.hasClass=function(t){return c.contains(this.classes,t)},e.toNode=function(){return C.call(this,"span")},e.toMarkup=function(){return q.call(this,"span")},t}(),I=function(){function t(t,e,r,a){this.children=void 0,this.attributes=void 0,this.classes=void 0,this.height=void 0,this.depth=void 0,this.maxFontSize=void 0,this.style=void 0,B.call(this,e,a),this.children=r||[],this.setAttribute("href",t)}var e=t.prototype;return e.setAttribute=function(t,e){this.attributes[t]=e},e.hasClass=function(t){return c.contains(this.classes,t)},e.toNode=function(){return C.call(this,"a")},e.toMarkup=function(){return q.call(this,"a")},t}(),R=function(){function t(t,e,r){this.src=void 0,this.alt=void 0,this.classes=void 0,this.height=void 0,this.depth=void 0,this.maxFontSize=void 0,this.style=void 0,this.alt=e,this.src=t,this.classes=["mord"],this.style=r}var e=t.prototype;return e.hasClass=function(t){return c.contains(this.classes,t)},e.toNode=function(){var t=document.createElement("img");for(var e in t.src=this.src,t.alt=this.alt,t.className="mord",this.style)this.style.hasOwnProperty(e)&&(t.style[e]=this.style[e]);return t},e.toMarkup=function(){var t=""+this.alt+"=n[0]&&t<=n[1])return r.name}return null}(this.text.charCodeAt(0));h&&this.classes.push(h+"_fallback"),/[\xee\xef\xed\xec]/.test(this.text)&&(this.text=O[this.text])}var e=t.prototype;return e.hasClass=function(t){return c.contains(this.classes,t)},e.toNode=function(){var t=document.createTextNode(this.text),e=null;for(var r in this.italic>0&&((e=document.createElement("span")).style.marginRight=this.italic+"em"),this.classes.length>0&&((e=e||document.createElement("span")).className=T(this.classes)),this.style)this.style.hasOwnProperty(r)&&((e=e||document.createElement("span")).style[r]=this.style[r]);return e?(e.appendChild(t),e):t},e.toMarkup=function(){var t=!1,e="0&&(r+="margin-right:"+this.italic+"em;"),this.style)this.style.hasOwnProperty(a)&&(r+=c.hyphenate(a)+":"+this.style[a]+";");r&&(t=!0,e+=' style="'+c.escape(r)+'"');var n=c.escape(this.text);return t?(e+=">",e+=n,e+=""):n},t}(),L=function(){function t(t,e){this.children=void 0,this.attributes=void 0,this.children=t||[],this.attributes=e||{}}var e=t.prototype;return e.toNode=function(){var t=document.createElementNS("http://www.w3.org/2000/svg","svg");for(var e in this.attributes)Object.prototype.hasOwnProperty.call(this.attributes,e)&&t.setAttribute(e,this.attributes[e]);for(var r=0;r":""},t}(),P=function(){function t(t){this.attributes=void 0,this.attributes=t||{}}var e=t.prototype;return e.toNode=function(){var t=document.createElementNS("http://www.w3.org/2000/svg","line");for(var e in this.attributes)Object.prototype.hasOwnProperty.call(this.attributes,e)&&t.setAttribute(e,this.attributes[e]);return t},e.toMarkup=function(){var t="",">"),$("math",Z,et,":",":"),$("math",Z,et,"\u2248","\\approx",!0),$("math",Z,et,"\u2245","\\cong",!0),$("math",Z,et,"\u2265","\\ge"),$("math",Z,et,"\u2265","\\geq",!0),$("math",Z,et,"\u2190","\\gets"),$("math",Z,et,">","\\gt"),$("math",Z,et,"\u2208","\\in",!0),$("math",Z,et,"\ue020","\\@not"),$("math",Z,et,"\u2282","\\subset",!0),$("math",Z,et,"\u2283","\\supset",!0),$("math",Z,et,"\u2286","\\subseteq",!0),$("math",Z,et,"\u2287","\\supseteq",!0),$("math",K,et,"\u2288","\\nsubseteq",!0),$("math",K,et,"\u2289","\\nsupseteq",!0),$("math",Z,et,"\u22a8","\\models"),$("math",Z,et,"\u2190","\\leftarrow",!0),$("math",Z,et,"\u2264","\\le"),$("math",Z,et,"\u2264","\\leq",!0),$("math",Z,et,"<","\\lt"),$("math",Z,et,"\u2192","\\rightarrow",!0),$("math",Z,et,"\u2192","\\to"),$("math",K,et,"\u2271","\\ngeq",!0),$("math",K,et,"\u2270","\\nleq",!0),$("math",Z,"spacing","\xa0","\\ "),$("math",Z,"spacing","\xa0","~"),$("math",Z,"spacing","\xa0","\\space"),$("math",Z,"spacing","\xa0","\\nobreakspace"),$("text",Z,"spacing","\xa0","\\ "),$("text",Z,"spacing","\xa0","~"),$("text",Z,"spacing","\xa0","\\space"),$("text",Z,"spacing","\xa0","\\nobreakspace"),$("math",Z,"spacing",null,"\\nobreak"),$("math",Z,"spacing",null,"\\allowbreak"),$("math",Z,"punct",",",","),$("math",Z,"punct",";",";"),$("math",K,J,"\u22bc","\\barwedge",!0),$("math",K,J,"\u22bb","\\veebar",!0),$("math",Z,J,"\u2299","\\odot",!0),$("math",Z,J,"\u2295","\\oplus",!0),$("math",Z,J,"\u2297","\\otimes",!0),$("math",Z,"textord","\u2202","\\partial",!0),$("math",Z,J,"\u2298","\\oslash",!0),$("math",K,J,"\u229a","\\circledcirc",!0),$("math",K,J,"\u22a1","\\boxdot",!0),$("math",Z,J,"\u25b3","\\bigtriangleup"),$("math",Z,J,"\u25bd","\\bigtriangledown"),$("math",Z,J,"\u2020","\\dagger"),$("math",Z,J,"\u22c4","\\diamond"),$("math",Z,J,"\u22c6","\\star"),$("math",Z,J,"\u25c3","\\triangleleft"),$("math",Z,J,"\u25b9","\\triangleright"),$("math",Z,"open","{","\\{"),$("text",Z,"textord","{","\\{"),$("text",Z,"textord","{","\\textbraceleft"),$("math",Z,"close","}","\\}"),$("text",Z,"textord","}","\\}"),$("text",Z,"textord","}","\\textbraceright"),$("math",Z,"open","{","\\lbrace"),$("math",Z,"close","}","\\rbrace"),$("math",Z,"open","[","\\lbrack"),$("text",Z,"textord","[","\\lbrack"),$("math",Z,"close","]","\\rbrack"),$("text",Z,"textord","]","\\rbrack"),$("math",Z,"open","(","\\lparen"),$("math",Z,"close",")","\\rparen"),$("text",Z,"textord","<","\\textless"),$("text",Z,"textord",">","\\textgreater"),$("math",Z,"open","\u230a","\\lfloor",!0),$("math",Z,"close","\u230b","\\rfloor",!0),$("math",Z,"open","\u2308","\\lceil",!0),$("math",Z,"close","\u2309","\\rceil",!0),$("math",Z,"textord","\\","\\backslash"),$("math",Z,"textord","\u2223","|"),$("math",Z,"textord","\u2223","\\vert"),$("text",Z,"textord","|","\\textbar"),$("math",Z,"textord","\u2225","\\|"),$("math",Z,"textord","\u2225","\\Vert"),$("text",Z,"textord","\u2225","\\textbardbl"),$("text",Z,"textord","~","\\textasciitilde"),$("text",Z,"textord","\\","\\textbackslash"),$("text",Z,"textord","^","\\textasciicircum"),$("math",Z,et,"\u2191","\\uparrow",!0),$("math",Z,et,"\u21d1","\\Uparrow",!0),$("math",Z,et,"\u2193","\\downarrow",!0),$("math",Z,et,"\u21d3","\\Downarrow",!0),$("math",Z,et,"\u2195","\\updownarrow",!0),$("math",Z,et,"\u21d5","\\Updownarrow",!0),$("math",Z,tt,"\u2210","\\coprod"),$("math",Z,tt,"\u22c1","\\bigvee"),$("math",Z,tt,"\u22c0","\\bigwedge"),$("math",Z,tt,"\u2a04","\\biguplus"),$("math",Z,tt,"\u22c2","\\bigcap"),$("math",Z,tt,"\u22c3","\\bigcup"),$("math",Z,tt,"\u222b","\\int"),$("math",Z,tt,"\u222b","\\intop"),$("math",Z,tt,"\u222c","\\iint"),$("math",Z,tt,"\u222d","\\iiint"),$("math",Z,tt,"\u220f","\\prod"),$("math",Z,tt,"\u2211","\\sum"),$("math",Z,tt,"\u2a02","\\bigotimes"),$("math",Z,tt,"\u2a01","\\bigoplus"),$("math",Z,tt,"\u2a00","\\bigodot"),$("math",Z,tt,"\u222e","\\oint"),$("math",Z,tt,"\u222f","\\oiint"),$("math",Z,tt,"\u2230","\\oiiint"),$("math",Z,tt,"\u2a06","\\bigsqcup"),$("math",Z,tt,"\u222b","\\smallint"),$("text",Z,"inner","\u2026","\\textellipsis"),$("math",Z,"inner","\u2026","\\mathellipsis"),$("text",Z,"inner","\u2026","\\ldots",!0),$("math",Z,"inner","\u2026","\\ldots",!0),$("math",Z,"inner","\u22ef","\\@cdots",!0),$("math",Z,"inner","\u22f1","\\ddots",!0),$("math",Z,"textord","\u22ee","\\varvdots"),$("math",Z,"accent-token","\u02ca","\\acute"),$("math",Z,"accent-token","\u02cb","\\grave"),$("math",Z,"accent-token","\xa8","\\ddot"),$("math",Z,"accent-token","~","\\tilde"),$("math",Z,"accent-token","\u02c9","\\bar"),$("math",Z,"accent-token","\u02d8","\\breve"),$("math",Z,"accent-token","\u02c7","\\check"),$("math",Z,"accent-token","^","\\hat"),$("math",Z,"accent-token","\u20d7","\\vec"),$("math",Z,"accent-token","\u02d9","\\dot"),$("math",Z,"accent-token","\u02da","\\mathring"),$("math",Z,Q,"\u0131","\\imath",!0),$("math",Z,Q,"\u0237","\\jmath",!0),$("text",Z,"textord","\u0131","\\i",!0),$("text",Z,"textord","\u0237","\\j",!0),$("text",Z,"textord","\xdf","\\ss",!0),$("text",Z,"textord","\xe6","\\ae",!0),$("text",Z,"textord","\xe6","\\ae",!0),$("text",Z,"textord","\u0153","\\oe",!0),$("text",Z,"textord","\xf8","\\o",!0),$("text",Z,"textord","\xc6","\\AE",!0),$("text",Z,"textord","\u0152","\\OE",!0),$("text",Z,"textord","\xd8","\\O",!0),$("text",Z,"accent-token","\u02ca","\\'"),$("text",Z,"accent-token","\u02cb","\\`"),$("text",Z,"accent-token","\u02c6","\\^"),$("text",Z,"accent-token","\u02dc","\\~"),$("text",Z,"accent-token","\u02c9","\\="),$("text",Z,"accent-token","\u02d8","\\u"),$("text",Z,"accent-token","\u02d9","\\."),$("text",Z,"accent-token","\u02da","\\r"),$("text",Z,"accent-token","\u02c7","\\v"),$("text",Z,"accent-token","\xa8",'\\"'),$("text",Z,"accent-token","\u02dd","\\H"),$("text",Z,"accent-token","\u25ef","\\textcircled");var rt={"--":!0,"---":!0,"``":!0,"''":!0};$("text",Z,"textord","\u2013","--"),$("text",Z,"textord","\u2013","\\textendash"),$("text",Z,"textord","\u2014","---"),$("text",Z,"textord","\u2014","\\textemdash"),$("text",Z,"textord","\u2018","`"),$("text",Z,"textord","\u2018","\\textquoteleft"),$("text",Z,"textord","\u2019","'"),$("text",Z,"textord","\u2019","\\textquoteright"),$("text",Z,"textord","\u201c","``"),$("text",Z,"textord","\u201c","\\textquotedblleft"),$("text",Z,"textord","\u201d","''"),$("text",Z,"textord","\u201d","\\textquotedblright"),$("math",Z,"textord","\xb0","\\degree",!0),$("text",Z,"textord","\xb0","\\degree"),$("text",Z,"textord","\xb0","\\textdegree",!0),$("math",Z,Q,"\xa3","\\pounds"),$("math",Z,Q,"\xa3","\\mathsterling",!0),$("text",Z,Q,"\xa3","\\pounds"),$("text",Z,Q,"\xa3","\\textsterling",!0),$("math",K,"textord","\u2720","\\maltese"),$("text",K,"textord","\u2720","\\maltese"),$("text",Z,"spacing","\xa0","\\ "),$("text",Z,"spacing","\xa0"," "),$("text",Z,"spacing","\xa0","~");for(var at=0;at<'0123456789/@."'.length;at++){var nt='0123456789/@."'.charAt(at);$("math",Z,"textord",nt,nt)}for(var it=0;it<'0123456789!@*()-=+[]<>|";:?/.,'.length;it++){var ot='0123456789!@*()-=+[]<>|";:?/.,'.charAt(it);$("text",Z,"textord",ot,ot)}for(var st="ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz",ht=0;ht=5?0:t>=3?1:2]){var r=Y[e]={cssEmPerMu:V.quad[e]/18};for(var a in V)V.hasOwnProperty(a)&&(r[a]=V[a][e])}return Y[e]}(this.size)),this._fontMetrics},e.getColor=function(){return this.phantom?"transparent":this.color},t}();kt.BASESIZE=6;var St=kt,Mt={pt:1,mm:7227/2540,cm:7227/254,in:72.27,bp:1.00375,pc:12,dd:1238/1157,cc:14856/1157,nd:685/642,nc:1370/107,sp:1/65536,px:1.00375},zt={ex:!0,em:!0,mu:!0},At=function(t){return"string"!=typeof t&&(t=t.unit),t in Mt||t in zt||"ex"===t},Tt=function(t,e){var r;if(t.unit in Mt)r=Mt[t.unit]/e.fontMetrics().ptPerEm/e.sizeMultiplier;else if("mu"===t.unit)r=e.fontMetrics().cssEmPerMu;else{var a;if(a=e.style.isTight()?e.havingStyle(e.style.text()):e,"ex"===t.unit)r=a.fontMetrics().xHeight;else{if("em"!==t.unit)throw new o("Invalid unit: '"+t.unit+"'");r=a.fontMetrics().quad}a!==e&&(r*=a.sizeMultiplier/e.sizeMultiplier)}return Math.min(t.number*r,e.maxSize)},Bt=["\\imath","\u0131","\\jmath","\u0237","\\pounds","\\mathsterling","\\textsterling","\xa3"],Ct=function(t,e,r){return j[r][t]&&j[r][t].replace&&(t=j[r][t].replace),{value:t,metrics:G(t,e,r)}},qt=function(t,e,r,a,n){var i,o=Ct(t,e,r),s=o.metrics;if(t=o.value,s){var h=s.italic;("text"===r||a&&"mathit"===a.font)&&(h=0),i=new E(t,s.height,s.depth,h,s.skew,s.width,n)}else"undefined"!=typeof console&&console.warn("No character metrics for '"+t+"' in style '"+e+"' and mode '"+r+"'"),i=new E(t,0,0,0,0,0,n);if(a){i.maxFontSize=a.sizeMultiplier,a.style.isTight()&&i.classes.push("mtight");var l=a.getColor();l&&(i.style.color=l)}return i},Nt=function(t,e){if(T(t.classes)!==T(e.classes)||t.skew!==e.skew||t.maxFontSize!==e.maxFontSize)return!1;for(var r in t.style)if(t.style.hasOwnProperty(r)&&t.style[r]!==e.style[r])return!1;for(var a in e.style)if(e.style.hasOwnProperty(a)&&t.style[a]!==e.style[a])return!1;return!0},It=function(t){for(var e=0,r=0,a=0,n=0;ne&&(e=i.height),i.depth>r&&(r=i.depth),i.maxFontSize>a&&(a=i.maxFontSize)}t.height=e,t.depth=r,t.maxFontSize=a},Rt=function(t,e,r,a){var n=new N(t,e,r,a);return It(n),n},Ot=function(t,e,r,a){return new N(t,e,r,a)},Et=function(t){var e=new A(t);return It(e),e},Lt=function(t,e,r){var a="";switch(t){case"amsrm":a="AMS";break;case"textrm":a="Main";break;case"textsf":a="SansSerif";break;case"texttt":a="Typewriter";break;default:a=t}return a+"-"+("textbf"===e&&"textit"===r?"BoldItalic":"textbf"===e?"Bold":"textit"===e?"Italic":"Regular")},Ht={mathbf:{variant:"bold",fontName:"Main-Bold"},mathrm:{variant:"normal",fontName:"Main-Regular"},textit:{variant:"italic",fontName:"Main-Italic"},mathit:{variant:"italic",fontName:"Main-Italic"},mathbb:{variant:"double-struck",fontName:"AMS-Regular"},mathcal:{variant:"script",fontName:"Caligraphic-Regular"},mathfrak:{variant:"fraktur",fontName:"Fraktur-Regular"},mathscr:{variant:"script",fontName:"Script-Regular"},mathsf:{variant:"sans-serif",fontName:"SansSerif-Regular"},mathtt:{variant:"monospace",fontName:"Typewriter-Regular"}},Pt={vec:["vec",.471,.714],oiintSize1:["oiintSize1",.957,.499],oiintSize2:["oiintSize2",1.472,.659],oiiintSize1:["oiiintSize1",1.304,.499],oiiintSize2:["oiiintSize2",1.98,.659]},Dt={fontMap:Ht,makeSymbol:qt,mathsym:function(t,e,r,a){return void 0===a&&(a=[]),"boldsymbol"===r.font&&Ct(t,"Main-Bold",e).metrics?qt(t,"Main-Bold",e,r,a.concat(["mathbf"])):"\\"===t||"main"===j[e][t].font?qt(t,"Main-Regular",e,r,a):qt(t,"AMS-Regular",e,r,a.concat(["amsrm"]))},makeSpan:Rt,makeSvgSpan:Ot,makeLineSpan:function(t,e,r){var a=Rt([t],[],e);return a.height=Math.max(r||e.fontMetrics().defaultRuleThickness,e.minRuleThickness),a.style.borderBottomWidth=a.height+"em",a.maxFontSize=1,a},makeAnchor:function(t,e,r,a){var n=new I(t,e,r,a);return It(n),n},makeFragment:Et,wrapFragment:function(t,e){return t instanceof A?Rt([],[t],e):t},makeVList:function(t,e){for(var r=function(t){if("individualShift"===t.positionType){for(var e=t.children,r=[e[0]],a=-e[0].shift-e[0].elem.depth,n=a,i=1;i0&&(i.push(pe(o,e)),o=[]),i.push(n[s]));o.length>0&&i.push(pe(o,e)),r&&((a=pe(se(r,e,!0))).classes=["tag"],i.push(a));var l=re(["katex-html"],i);if(l.setAttribute("aria-hidden","true"),a){var m=a.children[0];m.style.height=l.height+l.depth+"em",m.style.verticalAlign=-l.depth+"em"}return l}function fe(t){return new A(t)}var ge=function(){function t(t,e){this.type=void 0,this.attributes=void 0,this.children=void 0,this.type=t,this.attributes={},this.children=e||[]}var e=t.prototype;return e.setAttribute=function(t,e){this.attributes[t]=e},e.getAttribute=function(t){return this.attributes[t]},e.toNode=function(){var t=document.createElementNS("http://www.w3.org/1998/Math/MathML",this.type);for(var e in this.attributes)Object.prototype.hasOwnProperty.call(this.attributes,e)&&t.setAttribute(e,this.attributes[e]);for(var r=0;r"},e.toText=function(){return this.children.map(function(t){return t.toText()}).join("")},t}(),xe=function(){function t(t){this.text=void 0,this.text=t}var e=t.prototype;return e.toNode=function(){return document.createTextNode(this.text)},e.toMarkup=function(){return c.escape(this.toText())},e.toText=function(){return this.text},t}(),ve={MathNode:ge,TextNode:xe,SpaceNode:function(){function t(t){this.width=void 0,this.character=void 0,this.width=t,this.character=t>=.05555&&t<=.05556?"\u200a":t>=.1666&&t<=.1667?"\u2009":t>=.2222&&t<=.2223?"\u2005":t>=.2777&&t<=.2778?"\u2005\u200a":t>=-.05556&&t<=-.05555?"\u200a\u2063":t>=-.1667&&t<=-.1666?"\u2009\u2063":t>=-.2223&&t<=-.2222?"\u205f\u2063":t>=-.2778&&t<=-.2777?"\u2005\u2063":null}var e=t.prototype;return e.toNode=function(){if(this.character)return document.createTextNode(this.character);var t=document.createElementNS("http://www.w3.org/1998/Math/MathML","mspace");return t.setAttribute("width",this.width+"em"),t},e.toMarkup=function(){return this.character?""+this.character+"":''},e.toText=function(){return this.character?this.character:" "},t}(),newDocumentFragment:fe},be=function(t,e,r){return!j[e][t]||!j[e][t].replace||55349===t.charCodeAt(0)||rt.hasOwnProperty(t)&&r&&(r.fontFamily&&"tt"===r.fontFamily.substr(4,2)||r.font&&"tt"===r.font.substr(4,2))||(t=j[e][t].replace),new ve.TextNode(t)},ye=function(t){return 1===t.length?t[0]:new ve.MathNode("mrow",t)},we=function(t,e){if("texttt"===e.fontFamily)return"monospace";if("textsf"===e.fontFamily)return"textit"===e.fontShape&&"textbf"===e.fontWeight?"sans-serif-bold-italic":"textit"===e.fontShape?"sans-serif-italic":"textbf"===e.fontWeight?"bold-sans-serif":"sans-serif";if("textit"===e.fontShape&&"textbf"===e.fontWeight)return"bold-italic";if("textit"===e.fontShape)return"italic";if("textbf"===e.fontWeight)return"bold";var r=e.font;if(!r||"mathnormal"===r)return null;var a=t.mode;if("mathit"===r)return"italic";if("boldsymbol"===r)return"bold-italic";if("mathbf"===r)return"bold";if("mathbb"===r)return"double-struck";if("mathfrak"===r)return"fraktur";if("mathscr"===r||"mathcal"===r)return"script";if("mathsf"===r)return"sans-serif";if("mathtt"===r)return"monospace";var n=t.text;return c.contains(["\\imath","\\jmath"],n)?null:(j[a][n]&&j[a][n].replace&&(n=j[a][n].replace),G(n,Dt.fontMap[r].fontName,a)?Dt.fontMap[r].variant:null)},ke=function(t,e,r){if(1===t.length){var a=Me(t[0],e);return r&&a instanceof ge&&"mo"===a.type&&(a.setAttribute("lspace","0em"),a.setAttribute("rspace","0em")),[a]}for(var n,i=[],o=0;o0&&(p.text=p.text.slice(0,1)+"\u0338"+p.text.slice(1),i.pop())}}}i.push(s),n=s}return i},Se=function(t,e,r){return ye(ke(t,e,r))},Me=function(t,e){if(!t)return new ve.MathNode("mrow");if(Jt[t.type])return Jt[t.type](t,e);throw new o("Got group of unknown type: '"+t.type+"'")};function ze(t,e,r,a){var n,i=ke(t,r);n=1===i.length&&i[0]instanceof ge&&c.contains(["mrow","mtable"],i[0].type)?i[0]:new ve.MathNode("mrow",i);var o=new ve.MathNode("annotation",[new ve.TextNode(e)]);o.setAttribute("encoding","application/x-tex");var s=new ve.MathNode("semantics",[n,o]),h=new ve.MathNode("math",[s]);h.setAttribute("xmlns","http://www.w3.org/1998/Math/MathML");var l=a?"katex":"katex-mathml";return Dt.makeSpan([l],[h])}var Ae=function(t){return new St({style:t.displayMode?w.DISPLAY:w.TEXT,maxSize:t.maxSize,minRuleThickness:t.minRuleThickness})},Te=function(t,e){if(e.displayMode){var r=["katex-display"];e.leqno&&r.push("leqno"),e.fleqn&&r.push("fleqn"),t=Dt.makeSpan(r,[t])}return t},Be=function(t,e,r){var a,n=Ae(r);if("mathml"===r.output)return ze(t,e,n,!0);if("html"===r.output){var i=de(t,n);a=Dt.makeSpan(["katex"],[i])}else{var o=ze(t,e,n,!1),s=de(t,n);a=Dt.makeSpan(["katex"],[o,s])}return Te(a,r)},Ce={widehat:"^",widecheck:"\u02c7",widetilde:"~",utilde:"~",overleftarrow:"\u2190",underleftarrow:"\u2190",xleftarrow:"\u2190",overrightarrow:"\u2192",underrightarrow:"\u2192",xrightarrow:"\u2192",underbrace:"\u23df",overbrace:"\u23de",overgroup:"\u23e0",undergroup:"\u23e1",overleftrightarrow:"\u2194",underleftrightarrow:"\u2194",xleftrightarrow:"\u2194",Overrightarrow:"\u21d2",xRightarrow:"\u21d2",overleftharpoon:"\u21bc",xleftharpoonup:"\u21bc",overrightharpoon:"\u21c0",xrightharpoonup:"\u21c0",xLeftarrow:"\u21d0",xLeftrightarrow:"\u21d4",xhookleftarrow:"\u21a9",xhookrightarrow:"\u21aa",xmapsto:"\u21a6",xrightharpoondown:"\u21c1",xleftharpoondown:"\u21bd",xrightleftharpoons:"\u21cc",xleftrightharpoons:"\u21cb",xtwoheadleftarrow:"\u219e",xtwoheadrightarrow:"\u21a0",xlongequal:"=",xtofrom:"\u21c4",xrightleftarrows:"\u21c4",xrightequilibrium:"\u21cc",xleftequilibrium:"\u21cb"},qe={overrightarrow:[["rightarrow"],.888,522,"xMaxYMin"],overleftarrow:[["leftarrow"],.888,522,"xMinYMin"],underrightarrow:[["rightarrow"],.888,522,"xMaxYMin"],underleftarrow:[["leftarrow"],.888,522,"xMinYMin"],xrightarrow:[["rightarrow"],1.469,522,"xMaxYMin"],xleftarrow:[["leftarrow"],1.469,522,"xMinYMin"],Overrightarrow:[["doublerightarrow"],.888,560,"xMaxYMin"],xRightarrow:[["doublerightarrow"],1.526,560,"xMaxYMin"],xLeftarrow:[["doubleleftarrow"],1.526,560,"xMinYMin"],overleftharpoon:[["leftharpoon"],.888,522,"xMinYMin"],xleftharpoonup:[["leftharpoon"],.888,522,"xMinYMin"],xleftharpoondown:[["leftharpoondown"],.888,522,"xMinYMin"],overrightharpoon:[["rightharpoon"],.888,522,"xMaxYMin"],xrightharpoonup:[["rightharpoon"],.888,522,"xMaxYMin"],xrightharpoondown:[["rightharpoondown"],.888,522,"xMaxYMin"],xlongequal:[["longequal"],.888,334,"xMinYMin"],xtwoheadleftarrow:[["twoheadleftarrow"],.888,334,"xMinYMin"],xtwoheadrightarrow:[["twoheadrightarrow"],.888,334,"xMaxYMin"],overleftrightarrow:[["leftarrow","rightarrow"],.888,522],overbrace:[["leftbrace","midbrace","rightbrace"],1.6,548],underbrace:[["leftbraceunder","midbraceunder","rightbraceunder"],1.6,548],underleftrightarrow:[["leftarrow","rightarrow"],.888,522],xleftrightarrow:[["leftarrow","rightarrow"],1.75,522],xLeftrightarrow:[["doubleleftarrow","doublerightarrow"],1.75,560],xrightleftharpoons:[["leftharpoondownplus","rightharpoonplus"],1.75,716],xleftrightharpoons:[["leftharpoonplus","rightharpoondownplus"],1.75,716],xhookleftarrow:[["leftarrow","righthook"],1.08,522],xhookrightarrow:[["lefthook","rightarrow"],1.08,522],overlinesegment:[["leftlinesegment","rightlinesegment"],.888,522],underlinesegment:[["leftlinesegment","rightlinesegment"],.888,522],overgroup:[["leftgroup","rightgroup"],.888,342],undergroup:[["leftgroupunder","rightgroupunder"],.888,342],xmapsto:[["leftmapsto","rightarrow"],1.5,522],xtofrom:[["leftToFrom","rightToFrom"],1.75,528],xrightleftarrows:[["baraboveleftarrow","rightarrowabovebar"],1.75,901],xrightequilibrium:[["baraboveshortleftharpoon","rightharpoonaboveshortbar"],1.75,716],xleftequilibrium:[["shortbaraboveleftharpoon","shortrightharpoonabovebar"],1.75,716]},Ne=function(t){return"ordgroup"===t.type?t.body.length:1},Ie=function(t,e,r,a){var n,i=t.height+t.depth+2*r;if(/fbox|color/.test(e)){if(n=Dt.makeSpan(["stretchy",e],[],a),"fbox"===e){var o=a.color&&a.getColor();o&&(n.style.borderColor=o)}}else{var s=[];/^[bx]cancel$/.test(e)&&s.push(new P({x1:"0",y1:"0",x2:"100%",y2:"100%","stroke-width":"0.046em"})),/^x?cancel$/.test(e)&&s.push(new P({x1:"0",y1:"100%",x2:"100%",y2:"0","stroke-width":"0.046em"}));var h=new L(s,{width:"100%",height:i+"em"});n=Dt.makeSvgSpan([],[h],a)}return n.height=i,n.style.height=i+"em",n},Re=function(t){var e=new ve.MathNode("mo",[new ve.TextNode(Ce[t.substr(1)])]);return e.setAttribute("stretchy","true"),e},Oe=function(t,e){var r=function(){var r=4e5,a=t.label.substr(1);if(c.contains(["widehat","widecheck","widetilde","utilde"],a)){var n,i,o,s=Ne(t.base);if(s>5)"widehat"===a||"widecheck"===a?(n=420,r=2364,o=.42,i=a+"4"):(n=312,r=2340,o=.34,i="tilde4");else{var h=[1,1,2,2,3,3][s];"widehat"===a||"widecheck"===a?(r=[0,1062,2364,2364,2364][h],n=[0,239,300,360,420][h],o=[0,.24,.3,.3,.36,.42][h],i=a+h):(r=[0,600,1033,2339,2340][h],n=[0,260,286,306,312][h],o=[0,.26,.286,.3,.306,.34][h],i="tilde"+h)}var l=new H(i),m=new L([l],{width:"100%",height:o+"em",viewBox:"0 0 "+r+" "+n,preserveAspectRatio:"none"});return{span:Dt.makeSvgSpan([],[m],e),minWidth:0,height:o}}var u,p,d=[],f=qe[a],g=f[0],x=f[1],v=f[2],b=v/1e3,y=g.length;if(1===y)u=["hide-tail"],p=[f[3]];else if(2===y)u=["halfarrow-left","halfarrow-right"],p=["xMinYMin","xMaxYMin"];else{if(3!==y)throw new Error("Correct katexImagesData or update code here to support\n "+y+" children.");u=["brace-left","brace-center","brace-right"],p=["xMinYMin","xMidYMin","xMaxYMin"]}for(var w=0;w0&&(a.style.minWidth=n+"em"),a},Ee=function(t,e){var r,a,n,i=Vt(t,"supsub");i?(r=(a=Ft(i.base,"accent")).base,i.base=r,n=function(t){if(t instanceof N)return t;throw new Error("Expected span but got "+String(t)+".")}(ue(i,e)),i.base=a):r=(a=Ft(t,"accent")).base;var o=ue(r,e.havingCrampedStyle()),s=0;if(a.isShifty&&c.isCharacterBox(r)){var h=c.getBaseElem(r);s=D(ue(h,e.havingCrampedStyle())).skew}var l,m=Math.min(o.height,e.fontMetrics().xHeight);if(a.isStretchy)l=Oe(a,e),l=Dt.makeVList({positionType:"firstBaseline",children:[{type:"elem",elem:o},{type:"elem",elem:l,wrapperClasses:["svg-align"],wrapperStyle:s>0?{width:"calc(100% - "+2*s+"em)",marginLeft:2*s+"em"}:void 0}]},e);else{var u,p;"\\vec"===a.label?(u=Dt.staticSvg("vec",e),p=Dt.svgData.vec[1]):((u=D(u=Dt.makeOrd({mode:a.mode,text:a.label},e,"textord"))).italic=0,p=u.width),l=Dt.makeSpan(["accent-body"],[u]);var d="\\textcircled"===a.label;d&&(l.classes.push("accent-full"),m=o.height);var f=s;d||(f-=p/2),l.style.left=f+"em","\\textcircled"===a.label&&(l.style.top=".2em"),l=Dt.makeVList({positionType:"firstBaseline",children:[{type:"elem",elem:o},{type:"kern",size:-m},{type:"elem",elem:l}]},e)}var g=Dt.makeSpan(["mord","accent"],[l],e);return n?(n.children[0]=g,n.height=Math.max(g.height,n.height),n.classes[0]="mord",n):g},Le=function(t,e){var r=t.isStretchy?Re(t.label):new ve.MathNode("mo",[be(t.label,t.mode)]),a=new ve.MathNode("mover",[Me(t.base,e),r]);return a.setAttribute("accent","true"),a},He=new RegExp(["\\acute","\\grave","\\ddot","\\tilde","\\bar","\\breve","\\check","\\hat","\\vec","\\dot","\\mathring"].map(function(t){return"\\"+t}).join("|"));Qt({type:"accent",names:["\\acute","\\grave","\\ddot","\\tilde","\\bar","\\breve","\\check","\\hat","\\vec","\\dot","\\mathring","\\widecheck","\\widehat","\\widetilde","\\overrightarrow","\\overleftarrow","\\Overrightarrow","\\overleftrightarrow","\\overgroup","\\overlinesegment","\\overleftharpoon","\\overrightharpoon"],props:{numArgs:1},handler:function(t,e){var r=e[0],a=!He.test(t.funcName),n=!a||"\\widehat"===t.funcName||"\\widetilde"===t.funcName||"\\widecheck"===t.funcName;return{type:"accent",mode:t.parser.mode,label:t.funcName,isStretchy:a,isShifty:n,base:r}},htmlBuilder:Ee,mathmlBuilder:Le}),Qt({type:"accent",names:["\\'","\\`","\\^","\\~","\\=","\\u","\\.",'\\"',"\\r","\\H","\\v","\\textcircled"],props:{numArgs:1,allowedInText:!0,allowedInMath:!1},handler:function(t,e){var r=e[0];return{type:"accent",mode:t.parser.mode,label:t.funcName,isStretchy:!1,isShifty:!0,base:r}},htmlBuilder:Ee,mathmlBuilder:Le}),Qt({type:"accentUnder",names:["\\underleftarrow","\\underrightarrow","\\underleftrightarrow","\\undergroup","\\underlinesegment","\\utilde"],props:{numArgs:1},handler:function(t,e){var r=t.parser,a=t.funcName,n=e[0];return{type:"accentUnder",mode:r.mode,label:a,base:n}},htmlBuilder:function(t,e){var r=ue(t.base,e),a=Oe(t,e),n="\\utilde"===t.label?.12:0,i=Dt.makeVList({positionType:"bottom",positionData:a.height+n,children:[{type:"elem",elem:a,wrapperClasses:["svg-align"]},{type:"kern",size:n},{type:"elem",elem:r}]},e);return Dt.makeSpan(["mord","accentunder"],[i],e)},mathmlBuilder:function(t,e){var r=Re(t.label),a=new ve.MathNode("munder",[Me(t.base,e),r]);return a.setAttribute("accentunder","true"),a}});var Pe=function(t){var e=new ve.MathNode("mpadded",t?[t]:[]);return e.setAttribute("width","+0.6em"),e.setAttribute("lspace","0.3em"),e};Qt({type:"xArrow",names:["\\xleftarrow","\\xrightarrow","\\xLeftarrow","\\xRightarrow","\\xleftrightarrow","\\xLeftrightarrow","\\xhookleftarrow","\\xhookrightarrow","\\xmapsto","\\xrightharpoondown","\\xrightharpoonup","\\xleftharpoondown","\\xleftharpoonup","\\xrightleftharpoons","\\xleftrightharpoons","\\xlongequal","\\xtwoheadrightarrow","\\xtwoheadleftarrow","\\xtofrom","\\xrightleftarrows","\\xrightequilibrium","\\xleftequilibrium"],props:{numArgs:1,numOptionalArgs:1},handler:function(t,e,r){var a=t.parser,n=t.funcName;return{type:"xArrow",mode:a.mode,label:n,body:e[0],below:r[0]}},htmlBuilder:function(t,e){var r,a=e.style,n=e.havingStyle(a.sup()),i=Dt.wrapFragment(ue(t.body,n,e),e);i.classes.push("x-arrow-pad"),t.below&&(n=e.havingStyle(a.sub()),(r=Dt.wrapFragment(ue(t.below,n,e),e)).classes.push("x-arrow-pad"));var o,s=Oe(t,e),h=-e.fontMetrics().axisHeight+.5*s.height,l=-e.fontMetrics().axisHeight-.5*s.height-.111;if((i.depth>.25||"\\xleftequilibrium"===t.label)&&(l-=i.depth),r){var m=-e.fontMetrics().axisHeight+r.height+.5*s.height+.111;o=Dt.makeVList({positionType:"individualShift",children:[{type:"elem",elem:i,shift:l},{type:"elem",elem:s,shift:h},{type:"elem",elem:r,shift:m}]},e)}else o=Dt.makeVList({positionType:"individualShift",children:[{type:"elem",elem:i,shift:l},{type:"elem",elem:s,shift:h}]},e);return o.children[0].children[0].children[1].classes.push("svg-align"),Dt.makeSpan(["mrel","x-arrow"],[o],e)},mathmlBuilder:function(t,e){var r,a=Re(t.label);if(t.body){var n=Pe(Me(t.body,e));if(t.below){var i=Pe(Me(t.below,e));r=new ve.MathNode("munderover",[a,i,n])}else r=new ve.MathNode("mover",[a,n])}else if(t.below){var o=Pe(Me(t.below,e));r=new ve.MathNode("munder",[a,o])}else r=Pe(),r=new ve.MathNode("mover",[a,r]);return r}}),Qt({type:"textord",names:["\\@char"],props:{numArgs:1,allowedInText:!0},handler:function(t,e){for(var r=t.parser,a=Ft(e[0],"ordgroup").body,n="",i=0;i","\\langle","\\rangle","/","\\backslash","\\lt","\\gt"],Je=[0,1.2,1.8,2.4,3],Qe=[{type:"small",style:w.SCRIPTSCRIPT},{type:"small",style:w.SCRIPT},{type:"small",style:w.TEXT},{type:"large",size:1},{type:"large",size:2},{type:"large",size:3},{type:"large",size:4}],tr=[{type:"small",style:w.SCRIPTSCRIPT},{type:"small",style:w.SCRIPT},{type:"small",style:w.TEXT},{type:"stack"}],er=[{type:"small",style:w.SCRIPTSCRIPT},{type:"small",style:w.SCRIPT},{type:"small",style:w.TEXT},{type:"large",size:1},{type:"large",size:2},{type:"large",size:3},{type:"large",size:4},{type:"stack"}],rr=function(t){if("small"===t.type)return"Main-Regular";if("large"===t.type)return"Size"+t.size+"-Regular";if("stack"===t.type)return"Size4-Regular";throw new Error("Add support for delim type '"+t.type+"' here.")},ar=function(t,e,r,a){for(var n=Math.min(2,3-a.style.size);ne)return r[n]}return r[r.length-1]},nr=function(t,e,r,a,n,i){var o;"<"===t||"\\lt"===t||"\u27e8"===t?t="\\langle":">"!==t&&"\\gt"!==t&&"\u27e9"!==t||(t="\\rangle"),o=c.contains(Ke,t)?Qe:c.contains($e,t)?er:tr;var s=ar(t,e,o,a);return"small"===s.type?function(t,e,r,a,n,i){var o=Dt.makeSymbol(t,"Main-Regular",n,a),s=Ue(o,e,a,i);return r&&Ge(s,a,e),s}(t,s.style,r,a,n,i):"large"===s.type?Ye(t,s.size,r,a,n,i):_e(t,e,r,a,n,i)},ir=function(t,e){var r,a,n=e.havingBaseSizing(),i=ar("\\surd",t*n.sizeMultiplier,er,n),o=n.sizeMultiplier,s=Math.max(0,e.minRuleThickness-e.fontMetrics().sqrtRuleThickness),h=0,l=0,m=0;return"small"===i.type?(t<1?o=1:t<1.4&&(o=.7),l=(1+s)/o,(r=je("sqrtMain",h=(1+s+.08)/o,m=1e3+1e3*s+80,s,e)).style.minWidth="0.853em",a=.833/o):"large"===i.type?(m=1080*Je[i.size],l=(Je[i.size]+s)/o,h=(Je[i.size]+s+.08)/o,(r=je("sqrtSize"+i.size,h,m,s,e)).style.minWidth="1.02em",a=1/o):(h=t+s+.08,l=t+s,m=Math.floor(1e3*t+s)+80,(r=je("sqrtTall",h,m,s,e)).style.minWidth="0.742em",a=1.056),r.height=l,r.style.height=h+"em",{span:r,advanceWidth:a,ruleWidth:(e.fontMetrics().sqrtRuleThickness+s)*o}},or=function(t,e,r,a,n){if("<"===t||"\\lt"===t||"\u27e8"===t?t="\\langle":">"!==t&&"\\gt"!==t&&"\u27e9"!==t||(t="\\rangle"),c.contains($e,t)||c.contains(Ke,t))return Ye(t,e,!1,r,a,n);if(c.contains(Ze,t))return _e(t,Je[e],!1,r,a,n);throw new o("Illegal delimiter: '"+t+"'")},sr=nr,hr=function(t,e,r,a,n,i){var o=a.fontMetrics().axisHeight*a.sizeMultiplier,s=5/a.fontMetrics().ptPerEm,h=Math.max(e-o,r+o),l=Math.max(h/500*901,2*h-s);return nr(t,l,!0,a,n,i)},lr={"\\bigl":{mclass:"mopen",size:1},"\\Bigl":{mclass:"mopen",size:2},"\\biggl":{mclass:"mopen",size:3},"\\Biggl":{mclass:"mopen",size:4},"\\bigr":{mclass:"mclose",size:1},"\\Bigr":{mclass:"mclose",size:2},"\\biggr":{mclass:"mclose",size:3},"\\Biggr":{mclass:"mclose",size:4},"\\bigm":{mclass:"mrel",size:1},"\\Bigm":{mclass:"mrel",size:2},"\\biggm":{mclass:"mrel",size:3},"\\Biggm":{mclass:"mrel",size:4},"\\big":{mclass:"mord",size:1},"\\Big":{mclass:"mord",size:2},"\\bigg":{mclass:"mord",size:3},"\\Bigg":{mclass:"mord",size:4}},mr=["(","\\lparen",")","\\rparen","[","\\lbrack","]","\\rbrack","\\{","\\lbrace","\\}","\\rbrace","\\lfloor","\\rfloor","\u230a","\u230b","\\lceil","\\rceil","\u2308","\u2309","<",">","\\langle","\u27e8","\\rangle","\u27e9","\\lt","\\gt","\\lvert","\\rvert","\\lVert","\\rVert","\\lgroup","\\rgroup","\u27ee","\u27ef","\\lmoustache","\\rmoustache","\u23b0","\u23b1","/","\\backslash","|","\\vert","\\|","\\Vert","\\uparrow","\\Uparrow","\\downarrow","\\Downarrow","\\updownarrow","\\Updownarrow","."];function cr(t,e){var r=Yt(t);if(r&&c.contains(mr,r.text))return r;throw new o("Invalid delimiter: '"+(r?r.text:JSON.stringify(t))+"' after '"+e.funcName+"'",t)}function ur(t){if(!t.body)throw new Error("Bug: The leftright ParseNode wasn't fully parsed.")}Qt({type:"delimsizing",names:["\\bigl","\\Bigl","\\biggl","\\Biggl","\\bigr","\\Bigr","\\biggr","\\Biggr","\\bigm","\\Bigm","\\biggm","\\Biggm","\\big","\\Big","\\bigg","\\Bigg"],props:{numArgs:1},handler:function(t,e){var r=cr(e[0],t);return{type:"delimsizing",mode:t.parser.mode,size:lr[t.funcName].size,mclass:lr[t.funcName].mclass,delim:r.text}},htmlBuilder:function(t,e){return"."===t.delim?Dt.makeSpan([t.mclass]):or(t.delim,t.size,e,t.mode,[t.mclass])},mathmlBuilder:function(t){var e=[];"."!==t.delim&&e.push(be(t.delim,t.mode));var r=new ve.MathNode("mo",e);return"mopen"===t.mclass||"mclose"===t.mclass?r.setAttribute("fence","true"):r.setAttribute("fence","false"),r}}),Qt({type:"leftright-right",names:["\\right"],props:{numArgs:1},handler:function(t,e){var r=t.parser.gullet.macros.get("\\current@color");if(r&&"string"!=typeof r)throw new o("\\current@color set to non-string in \\right");return{type:"leftright-right",mode:t.parser.mode,delim:cr(e[0],t).text,color:r}}}),Qt({type:"leftright",names:["\\left"],props:{numArgs:1},handler:function(t,e){var r=cr(e[0],t),a=t.parser;++a.leftrightDepth;var n=a.parseExpression(!1);--a.leftrightDepth,a.expect("\\right",!1);var i=Ft(a.parseFunction(),"leftright-right");return{type:"leftright",mode:a.mode,body:n,left:r.text,right:i.delim,rightColor:i.color}},htmlBuilder:function(t,e){ur(t);for(var r,a,n=se(t.body,e,!0,["mopen","mclose"]),i=0,o=0,s=!1,h=0;h-1?"mpadded":"menclose",[Me(t.body,e)]);switch(t.label){case"\\cancel":a.setAttribute("notation","updiagonalstrike");break;case"\\bcancel":a.setAttribute("notation","downdiagonalstrike");break;case"\\sout":a.setAttribute("notation","horizontalstrike");break;case"\\fbox":a.setAttribute("notation","box");break;case"\\fcolorbox":case"\\colorbox":if(r=e.fontMetrics().fboxsep*e.fontMetrics().ptPerEm,a.setAttribute("width","+"+2*r+"pt"),a.setAttribute("height","+"+2*r+"pt"),a.setAttribute("lspace",r+"pt"),a.setAttribute("voffset",r+"pt"),"\\fcolorbox"===t.label){var n=Math.max(e.fontMetrics().fboxrule,e.minRuleThickness);a.setAttribute("style","border: "+n+"em solid "+String(t.borderColor))}break;case"\\xcancel":a.setAttribute("notation","updiagonalstrike downdiagonalstrike")}return t.backgroundColor&&a.setAttribute("mathbackground",t.backgroundColor),a};Qt({type:"enclose",names:["\\colorbox"],props:{numArgs:2,allowedInText:!0,greediness:3,argTypes:["color","text"]},handler:function(t,e,r){var a=t.parser,n=t.funcName,i=Ft(e[0],"color-token").color,o=e[1];return{type:"enclose",mode:a.mode,label:n,backgroundColor:i,body:o}},htmlBuilder:pr,mathmlBuilder:dr}),Qt({type:"enclose",names:["\\fcolorbox"],props:{numArgs:3,allowedInText:!0,greediness:3,argTypes:["color","color","text"]},handler:function(t,e,r){var a=t.parser,n=t.funcName,i=Ft(e[0],"color-token").color,o=Ft(e[1],"color-token").color,s=e[2];return{type:"enclose",mode:a.mode,label:n,backgroundColor:o,borderColor:i,body:s}},htmlBuilder:pr,mathmlBuilder:dr}),Qt({type:"enclose",names:["\\fbox"],props:{numArgs:1,argTypes:["hbox"],allowedInText:!0},handler:function(t,e){return{type:"enclose",mode:t.parser.mode,label:"\\fbox",body:e[0]}}}),Qt({type:"enclose",names:["\\cancel","\\bcancel","\\xcancel","\\sout"],props:{numArgs:1},handler:function(t,e,r){var a=t.parser,n=t.funcName,i=e[0];return{type:"enclose",mode:a.mode,label:n,body:i}},htmlBuilder:pr,mathmlBuilder:dr});var fr={};function gr(t){for(var e=t.type,r=t.names,a=t.props,n=t.handler,i=t.htmlBuilder,o=t.mathmlBuilder,s={type:e,numArgs:a.numArgs||0,greediness:1,allowedInText:!1,numOptionalArgs:0,handler:n},h=0;h0&&(b+=.25),l.push({pos:b,isDashed:t[e]})}for(y(i[0]),r=0;r0&&(M<(B+=v)&&(M=B),B=0),t.addJot&&(M+=f),z.height=S,z.depth=M,b+=S,z.pos=b,b+=M+B,h[r]=z,y(i[r+1])}var C,q,N=b/2+e.fontMetrics().axisHeight,I=t.cols||[],R=[];for(a=0,q=0;a=s)){var P=void 0;(a>0||t.hskipBeforeAndAfter)&&0!==(P=c.deflt(O.pregap,p))&&((C=Dt.makeSpan(["arraycolsep"],[])).style.width=P+"em",R.push(C));var D=[];for(r=0;r0){for(var G=Dt.makeLineSpan("hline",e,m),Y=Dt.makeLineSpan("hdashline",e,m),W=[{type:"elem",elem:h,shift:0}];l.length>0;){var X=l.pop(),_=X.pos-N;X.isDashed?W.push({type:"elem",elem:Y,shift:_}):W.push({type:"elem",elem:G,shift:_})}h=Dt.makeVList({positionType:"individualShift",children:W},e)}return Dt.makeSpan(["mord"],[h],e)},wr={c:"center ",l:"left ",r:"right "},kr=function(t,e){var r=new ve.MathNode("mtable",t.body.map(function(t){return new ve.MathNode("mtr",t.map(function(t){return new ve.MathNode("mtd",[Me(t,e)])}))})),a=.5===t.arraystretch?.1:.16+t.arraystretch-1+(t.addJot?.09:0);r.setAttribute("rowspacing",a+"em");var n="",i="";if(t.cols){var o=t.cols,s="",h=!1,l=0,m=o.length;"separator"===o[0].type&&(n+="top ",l=1),"separator"===o[o.length-1].type&&(n+="bottom ",m-=1);for(var c=l;c0?"left ":"",n+=g[g.length-1].length>0?"right ":"";for(var x=1;x0&&c&&(d=1),a[u]={type:"align",align:p,pregap:d,postgap:0}}return n.colSeparationType=c?"align":"alignat",n};gr({type:"array",names:["array","darray"],props:{numArgs:1},handler:function(t,e){var r={cols:(Yt(e[0])?[e[0]]:Ft(e[0],"ordgroup").body).map(function(t){var e=Gt(t).text;if(-1!=="lcr".indexOf(e))return{type:"align",align:e};if("|"===e)return{type:"separator",separator:"|"};if(":"===e)return{type:"separator",separator:":"};throw new o("Unknown column alignment: "+e,t)}),hskipBeforeAndAfter:!0};return vr(t.parser,r,br(t.envName))},htmlBuilder:yr,mathmlBuilder:kr}),gr({type:"array",names:["matrix","pmatrix","bmatrix","Bmatrix","vmatrix","Vmatrix"],props:{numArgs:0},handler:function(t){var e={matrix:null,pmatrix:["(",")"],bmatrix:["[","]"],Bmatrix:["\\{","\\}"],vmatrix:["|","|"],Vmatrix:["\\Vert","\\Vert"]}[t.envName],r=vr(t.parser,{hskipBeforeAndAfter:!1},br(t.envName));return e?{type:"leftright",mode:t.mode,body:[r],left:e[0],right:e[1],rightColor:void 0}:r},htmlBuilder:yr,mathmlBuilder:kr}),gr({type:"array",names:["smallmatrix"],props:{numArgs:0},handler:function(t){var e=vr(t.parser,{arraystretch:.5},"script");return e.colSeparationType="small",e},htmlBuilder:yr,mathmlBuilder:kr}),gr({type:"array",names:["subarray"],props:{numArgs:1},handler:function(t,e){var r=(Yt(e[0])?[e[0]]:Ft(e[0],"ordgroup").body).map(function(t){var e=Gt(t).text;if(-1!=="lc".indexOf(e))return{type:"align",align:e};throw new o("Unknown column alignment: "+e,t)});if(r.length>1)throw new o("{subarray} can contain only one column");var a={cols:r,hskipBeforeAndAfter:!1,arraystretch:.5};if((a=vr(t.parser,a,"script")).body[0].length>1)throw new o("{subarray} can contain only one column");return a},htmlBuilder:yr,mathmlBuilder:kr}),gr({type:"array",names:["cases","dcases"],props:{numArgs:0},handler:function(t){var e=vr(t.parser,{arraystretch:1.2,cols:[{type:"align",align:"l",pregap:0,postgap:1},{type:"align",align:"l",pregap:0,postgap:0}]},br(t.envName));return{type:"leftright",mode:t.mode,body:[e],left:"\\{",right:".",rightColor:void 0}},htmlBuilder:yr,mathmlBuilder:kr}),gr({type:"array",names:["aligned"],props:{numArgs:0},handler:Sr,htmlBuilder:yr,mathmlBuilder:kr}),gr({type:"array",names:["gathered"],props:{numArgs:0},handler:function(t){return vr(t.parser,{cols:[{type:"align",align:"c"}],addJot:!0},"display")},htmlBuilder:yr,mathmlBuilder:kr}),gr({type:"array",names:["alignedat"],props:{numArgs:1},handler:Sr,htmlBuilder:yr,mathmlBuilder:kr}),Qt({type:"text",names:["\\hline","\\hdashline"],props:{numArgs:0,allowedInText:!0,allowedInMath:!0},handler:function(t,e){throw new o(t.funcName+" valid only within array environment")}});var Mr=fr;Qt({type:"environment",names:["\\begin","\\end"],props:{numArgs:1,argTypes:["text"]},handler:function(t,e){var r=t.parser,a=t.funcName,n=e[0];if("ordgroup"!==n.type)throw new o("Invalid environment name",n);for(var i="",s=0;s=w.SCRIPT.id?r.text():w.DISPLAY:"text"===t&&r.size===w.DISPLAY.size?r=w.TEXT:"script"===t?r=w.SCRIPT:"scriptscript"===t&&(r=w.SCRIPTSCRIPT),r},Rr=function(t,e){var r,a=Ir(t.size,e.style),n=a.fracNum(),i=a.fracDen();r=e.havingStyle(n);var o=ue(t.numer,r,e);if(t.continued){var s=8.5/e.fontMetrics().ptPerEm,h=3.5/e.fontMetrics().ptPerEm;o.height=o.height0?3*c:7*c,d=e.fontMetrics().denom1):(m>0?(u=e.fontMetrics().num2,p=c):(u=e.fontMetrics().num3,p=3*c),d=e.fontMetrics().denom2),l){var y=e.fontMetrics().axisHeight;u-o.depth-(y+.5*m)0&&(e="."===(e=t)?null:e),e};Qt({type:"genfrac",names:["\\genfrac"],props:{numArgs:6,greediness:6,argTypes:["math","math","size","text","math","math"]},handler:function(t,e){var r=t.parser,a=e[4],n=e[5],i=Vt(e[0],"atom");i&&(i=Ut(e[0],"open"));var o=i?Lr(i.text):null,s=Vt(e[1],"atom");s&&(s=Ut(e[1],"close"));var h,l=s?Lr(s.text):null,m=Ft(e[2],"size"),c=null;h=!!m.isBlank||(c=m.value).number>0;var u="auto",p=Vt(e[3],"ordgroup");if(p){if(p.body.length>0){var d=Ft(p.body[0],"textord");u=Er[Number(d.text)]}}else p=Ft(e[3],"textord"),u=Er[Number(p.text)];return{type:"genfrac",mode:r.mode,numer:a,denom:n,continued:!1,hasBarLine:h,barSize:c,leftDelim:o,rightDelim:l,size:u}},htmlBuilder:Rr,mathmlBuilder:Or}),Qt({type:"infix",names:["\\above"],props:{numArgs:1,argTypes:["size"],infix:!0},handler:function(t,e){var r=t.parser,a=(t.funcName,t.token);return{type:"infix",mode:r.mode,replaceWith:"\\\\abovefrac",size:Ft(e[0],"size").value,token:a}}}),Qt({type:"genfrac",names:["\\\\abovefrac"],props:{numArgs:3,argTypes:["math","size","math"]},handler:function(t,e){var r=t.parser,a=(t.funcName,e[0]),n=function(t){if(!t)throw new Error("Expected non-null, but got "+String(t));return t}(Ft(e[1],"infix").size),i=e[2],o=n.number>0;return{type:"genfrac",mode:r.mode,numer:a,denom:i,continued:!1,hasBarLine:o,barSize:n,leftDelim:null,rightDelim:null,size:"auto"}},htmlBuilder:Rr,mathmlBuilder:Or});var Hr=function(t,e){var r,a,n=e.style,i=Vt(t,"supsub");i?(r=i.sup?ue(i.sup,e.havingStyle(n.sup()),e):ue(i.sub,e.havingStyle(n.sub()),e),a=Ft(i.base,"horizBrace")):a=Ft(t,"horizBrace");var o,s=ue(a.base,e.havingBaseStyle(w.DISPLAY)),h=Oe(a,e);if(a.isOver?(o=Dt.makeVList({positionType:"firstBaseline",children:[{type:"elem",elem:s},{type:"kern",size:.1},{type:"elem",elem:h}]},e)).children[0].children[0].children[1].classes.push("svg-align"):(o=Dt.makeVList({positionType:"bottom",positionData:s.depth+.1+h.height,children:[{type:"elem",elem:h},{type:"kern",size:.1},{type:"elem",elem:s}]},e)).children[0].children[0].children[0].classes.push("svg-align"),r){var l=Dt.makeSpan(["mord",a.isOver?"mover":"munder"],[o],e);o=a.isOver?Dt.makeVList({positionType:"firstBaseline",children:[{type:"elem",elem:l},{type:"kern",size:.2},{type:"elem",elem:r}]},e):Dt.makeVList({positionType:"bottom",positionData:l.depth+.2+r.height+r.depth,children:[{type:"elem",elem:r},{type:"kern",size:.2},{type:"elem",elem:l}]},e)}return Dt.makeSpan(["mord",a.isOver?"mover":"munder"],[o],e)};Qt({type:"horizBrace",names:["\\overbrace","\\underbrace"],props:{numArgs:1},handler:function(t,e){var r=t.parser,a=t.funcName;return{type:"horizBrace",mode:r.mode,label:a,isOver:/^\\over/.test(a),base:e[0]}},htmlBuilder:Hr,mathmlBuilder:function(t,e){var r=Re(t.label);return new ve.MathNode(t.isOver?"mover":"munder",[Me(t.base,e),r])}}),Qt({type:"href",names:["\\href"],props:{numArgs:2,argTypes:["url","original"],allowedInText:!0},handler:function(t,e){var r=t.parser,a=e[1],n=Ft(e[0],"url").url;return r.settings.isTrusted({command:"\\href",url:n})?{type:"href",mode:r.mode,href:n,body:ee(a)}:r.formatUnsupportedCmd("\\href")},htmlBuilder:function(t,e){var r=se(t.body,e,!1);return Dt.makeAnchor(t.href,[],r,e)},mathmlBuilder:function(t,e){var r=Se(t.body,e);return r instanceof ge||(r=new ge("mrow",[r])),r.setAttribute("href",t.href),r}}),Qt({type:"href",names:["\\url"],props:{numArgs:1,argTypes:["url"],allowedInText:!0},handler:function(t,e){var r=t.parser,a=Ft(e[0],"url").url;if(!r.settings.isTrusted({command:"\\url",url:a}))return r.formatUnsupportedCmd("\\url");for(var n=[],i=0;i0&&(a=Tt(t.totalheight,e)-r,a=Number(a.toFixed(2)));var n=0;t.width.number>0&&(n=Tt(t.width,e));var i={height:r+a+"em"};n>0&&(i.width=n+"em"),a>0&&(i.verticalAlign=-a+"em");var o=new R(t.src,t.alt,i);return o.height=r,o.depth=a,o},mathmlBuilder:function(t,e){var r=new ve.MathNode("mglyph",[]);r.setAttribute("alt",t.alt);var a=Tt(t.height,e),n=0;if(t.totalheight.number>0&&(n=(n=Tt(t.totalheight,e)-a).toFixed(2),r.setAttribute("valign","-"+n+"em")),r.setAttribute("height",a+n+"em"),t.width.number>0){var i=Tt(t.width,e);r.setAttribute("width",i+"em")}return r.setAttribute("src",t.src),r}}),Qt({type:"kern",names:["\\kern","\\mkern","\\hskip","\\mskip"],props:{numArgs:1,argTypes:["size"],allowedInText:!0},handler:function(t,e){var r=t.parser,a=t.funcName,n=Ft(e[0],"size");if(r.settings.strict){var i="m"===a[1],o="mu"===n.value.unit;i?(o||r.settings.reportNonstrict("mathVsTextUnits","LaTeX's "+a+" supports only mu units, not "+n.value.unit+" units"),"math"!==r.mode&&r.settings.reportNonstrict("mathVsTextUnits","LaTeX's "+a+" works only in math mode")):o&&r.settings.reportNonstrict("mathVsTextUnits","LaTeX's "+a+" doesn't support mu units")}return{type:"kern",mode:r.mode,dimension:n.value}},htmlBuilder:function(t,e){return Dt.makeGlue(t.dimension,e)},mathmlBuilder:function(t,e){var r=Tt(t.dimension,e);return new ve.SpaceNode(r)}}),Qt({type:"lap",names:["\\mathllap","\\mathrlap","\\mathclap"],props:{numArgs:1,allowedInText:!0},handler:function(t,e){var r=t.parser,a=t.funcName,n=e[0];return{type:"lap",mode:r.mode,alignment:a.slice(5),body:n}},htmlBuilder:function(t,e){var r;"clap"===t.alignment?(r=Dt.makeSpan([],[ue(t.body,e)]),r=Dt.makeSpan(["inner"],[r],e)):r=Dt.makeSpan(["inner"],[ue(t.body,e)]);var a=Dt.makeSpan(["fix"],[]),n=Dt.makeSpan([t.alignment],[r,a],e),i=Dt.makeSpan(["strut"]);return i.style.height=n.height+n.depth+"em",i.style.verticalAlign=-n.depth+"em",n.children.unshift(i),n=Dt.makeVList({positionType:"firstBaseline",children:[{type:"elem",elem:n}]},e),Dt.makeSpan(["mord"],[n],e)},mathmlBuilder:function(t,e){var r=new ve.MathNode("mpadded",[Me(t.body,e)]);if("rlap"!==t.alignment){var a="llap"===t.alignment?"-1":"-0.5";r.setAttribute("lspace",a+"width")}return r.setAttribute("width","0px"),r}}),Qt({type:"styling",names:["\\(","$"],props:{numArgs:0,allowedInText:!0,allowedInMath:!1},handler:function(t,e){var r=t.funcName,a=t.parser,n=a.mode;a.switchMode("math");var i="\\("===r?"\\)":"$",o=a.parseExpression(!1,i);return a.expect(i),a.switchMode(n),{type:"styling",mode:a.mode,style:"text",body:o}}}),Qt({type:"text",names:["\\)","\\]"],props:{numArgs:0,allowedInText:!0,allowedInMath:!1},handler:function(t,e){throw new o("Mismatched "+t.funcName)}});var Dr=function(t,e){switch(e.style.size){case w.DISPLAY.size:return t.display;case w.TEXT.size:return t.text;case w.SCRIPT.size:return t.script;case w.SCRIPTSCRIPT.size:return t.scriptscript;default:return t.text}};Qt({type:"mathchoice",names:["\\mathchoice"],props:{numArgs:4},handler:function(t,e){return{type:"mathchoice",mode:t.parser.mode,display:ee(e[0]),text:ee(e[1]),script:ee(e[2]),scriptscript:ee(e[3])}},htmlBuilder:function(t,e){var r=Dr(t,e),a=se(r,e,!1);return Dt.makeFragment(a)},mathmlBuilder:function(t,e){var r=Dr(t,e);return Se(r,e)}});var Fr=function(t,e,r,a,n,i,o){var s,h,l;if(t=Dt.makeSpan([],[t]),e){var m=ue(e,a.havingStyle(n.sup()),a);h={elem:m,kern:Math.max(a.fontMetrics().bigOpSpacing1,a.fontMetrics().bigOpSpacing3-m.depth)}}if(r){var c=ue(r,a.havingStyle(n.sub()),a);s={elem:c,kern:Math.max(a.fontMetrics().bigOpSpacing2,a.fontMetrics().bigOpSpacing4-c.height)}}if(h&&s){var u=a.fontMetrics().bigOpSpacing5+s.elem.height+s.elem.depth+s.kern+t.depth+o;l=Dt.makeVList({positionType:"bottom",positionData:u,children:[{type:"kern",size:a.fontMetrics().bigOpSpacing5},{type:"elem",elem:s.elem,marginLeft:-i+"em"},{type:"kern",size:s.kern},{type:"elem",elem:t},{type:"kern",size:h.kern},{type:"elem",elem:h.elem,marginLeft:i+"em"},{type:"kern",size:a.fontMetrics().bigOpSpacing5}]},a)}else if(s){var p=t.height-o;l=Dt.makeVList({positionType:"top",positionData:p,children:[{type:"kern",size:a.fontMetrics().bigOpSpacing5},{type:"elem",elem:s.elem,marginLeft:-i+"em"},{type:"kern",size:s.kern},{type:"elem",elem:t}]},a)}else{if(!h)return t;var d=t.depth+o;l=Dt.makeVList({positionType:"bottom",positionData:d,children:[{type:"elem",elem:t},{type:"kern",size:h.kern},{type:"elem",elem:h.elem,marginLeft:i+"em"},{type:"kern",size:a.fontMetrics().bigOpSpacing5}]},a)}return Dt.makeSpan(["mop","op-limits"],[l],a)},Vr=["\\smallint"],Ur=function(t,e){var r,a,n,i=!1,o=Vt(t,"supsub");o?(r=o.sup,a=o.sub,n=Ft(o.base,"op"),i=!0):n=Ft(t,"op");var s,h=e.style,l=!1;if(h.size===w.DISPLAY.size&&n.symbol&&!c.contains(Vr,n.name)&&(l=!0),n.symbol){var m=l?"Size2-Regular":"Size1-Regular",u="";if("\\oiint"!==n.name&&"\\oiiint"!==n.name||(u=n.name.substr(1),n.name="oiint"===u?"\\iint":"\\iiint"),s=Dt.makeSymbol(n.name,m,"math",e,["mop","op-symbol",l?"large-op":"small-op"]),u.length>0){var p=s.italic,d=Dt.staticSvg(u+"Size"+(l?"2":"1"),e);s=Dt.makeVList({positionType:"individualShift",children:[{type:"elem",elem:s,shift:0},{type:"elem",elem:d,shift:l?.08:0}]},e),n.name="\\"+u,s.classes.unshift("mop"),s.italic=p}}else if(n.body){var f=se(n.body,e,!0);1===f.length&&f[0]instanceof E?(s=f[0]).classes[0]="mop":s=Dt.makeSpan(["mop"],Dt.tryCombineChars(f),e)}else{for(var g=[],x=1;x0){for(var h=n.body.map(function(t){var e=t.text;return"string"==typeof e?{type:"textord",mode:t.mode,text:e}:t}),l=se(h,e.withFont("mathrm"),!0),m=0;m=0?s.setAttribute("height","+"+n+"em"):(s.setAttribute("height",n+"em"),s.setAttribute("depth","+"+-n+"em")),s.setAttribute("voffset",n+"em"),s}});var jr=["\\tiny","\\sixptsize","\\scriptsize","\\footnotesize","\\small","\\normalsize","\\large","\\Large","\\LARGE","\\huge","\\Huge"];Qt({type:"sizing",names:jr,props:{numArgs:0,allowedInText:!0},handler:function(t,e){var r=t.breakOnTokenText,a=t.funcName,n=t.parser,i=n.parseExpression(!1,r);return{type:"sizing",mode:n.mode,size:jr.indexOf(a)+1,body:i}},htmlBuilder:function(t,e){var r=e.havingSize(t.size);return _r(t.body,r,e)},mathmlBuilder:function(t,e){var r=e.havingSize(t.size),a=ke(t.body,r),n=new ve.MathNode("mstyle",a);return n.setAttribute("mathsize",r.sizeMultiplier+"em"),n}}),Qt({type:"smash",names:["\\smash"],props:{numArgs:1,numOptionalArgs:1,allowedInText:!0},handler:function(t,e,r){var a=t.parser,n=!1,i=!1,o=r[0]&&Ft(r[0],"ordgroup");if(o)for(var s="",h=0;hr.height+r.depth+i&&(i=(i+c-r.height-r.depth)/2);var u=h.height-r.height-i-l;r.style.paddingLeft=m+"em";var p=Dt.makeVList({positionType:"firstBaseline",children:[{type:"elem",elem:r,wrapperClasses:["svg-align"]},{type:"kern",size:-(r.height+u)},{type:"elem",elem:h},{type:"kern",size:l}]},e);if(t.index){var d=e.havingStyle(w.SCRIPTSCRIPT),f=ue(t.index,d,e),g=.6*(p.height-p.depth),x=Dt.makeVList({positionType:"shift",positionData:-g,children:[{type:"elem",elem:f}]},e),v=Dt.makeSpan(["root"],[x]);return Dt.makeSpan(["mord","sqrt"],[v,p],e)}return Dt.makeSpan(["mord","sqrt"],[p],e)},mathmlBuilder:function(t,e){var r=t.body,a=t.index;return a?new ve.MathNode("mroot",[Me(r,e),Me(a,e)]):new ve.MathNode("msqrt",[Me(r,e)])}});var $r={display:w.DISPLAY,text:w.TEXT,script:w.SCRIPT,scriptscript:w.SCRIPTSCRIPT};Qt({type:"styling",names:["\\displaystyle","\\textstyle","\\scriptstyle","\\scriptscriptstyle"],props:{numArgs:0,allowedInText:!0},handler:function(t,e){var r=t.breakOnTokenText,a=t.funcName,n=t.parser,i=n.parseExpression(!0,r),o=a.slice(1,a.length-5);return{type:"styling",mode:n.mode,style:o,body:i}},htmlBuilder:function(t,e){var r=$r[t.style],a=e.havingStyle(r).withFont("");return _r(t.body,a,e)},mathmlBuilder:function(t,e){var r=$r[t.style],a=e.havingStyle(r),n=ke(t.body,a),i=new ve.MathNode("mstyle",n),o={display:["0","true"],text:["0","false"],script:["1","false"],scriptscript:["2","false"]}[t.style];return i.setAttribute("scriptlevel",o[0]),i.setAttribute("displaystyle",o[1]),i}});te({type:"supsub",htmlBuilder:function(t,e){var r=function(t,e){var r=t.base;return r?"op"===r.type?r.limits&&(e.style.size===w.DISPLAY.size||r.alwaysHandleSupSub)?Ur:null:"operatorname"===r.type?r.alwaysHandleSupSub&&(e.style.size===w.DISPLAY.size||r.limits)?Xr:null:"accent"===r.type?c.isCharacterBox(r.base)?Ee:null:"horizBrace"===r.type&&!t.sub===r.isOver?Hr:null:null}(t,e);if(r)return r(t,e);var a,n,i,o=t.base,s=t.sup,h=t.sub,l=ue(o,e),m=e.fontMetrics(),u=0,p=0,d=o&&c.isCharacterBox(o);if(s){var f=e.havingStyle(e.style.sup());a=ue(s,f,e),d||(u=l.height-f.fontMetrics().supDrop*f.sizeMultiplier/e.sizeMultiplier)}if(h){var g=e.havingStyle(e.style.sub());n=ue(h,g,e),d||(p=l.depth+g.fontMetrics().subDrop*g.sizeMultiplier/e.sizeMultiplier)}i=e.style===w.DISPLAY?m.sup1:e.style.cramped?m.sup3:m.sup2;var x,v=e.sizeMultiplier,b=.5/m.ptPerEm/v+"em",y=null;if(n){var k=t.base&&"op"===t.base.type&&t.base.name&&("\\oiint"===t.base.name||"\\oiiint"===t.base.name);(l instanceof E||k)&&(y=-l.italic+"em")}if(a&&n){u=Math.max(u,i,a.depth+.25*m.xHeight),p=Math.max(p,m.sub2);var S=4*m.defaultRuleThickness;if(u-a.depth-(n.height-p)0&&(u+=M,p-=M)}var z=[{type:"elem",elem:n,shift:p,marginRight:b,marginLeft:y},{type:"elem",elem:a,shift:-u,marginRight:b}];x=Dt.makeVList({positionType:"individualShift",children:z},e)}else if(n){p=Math.max(p,m.sub1,n.height-.8*m.xHeight);var A=[{type:"elem",elem:n,marginLeft:y,marginRight:b}];x=Dt.makeVList({positionType:"shift",positionData:p,children:A},e)}else{if(!a)throw new Error("supsub must have either sup or sub.");u=Math.max(u,i,a.depth+.25*m.xHeight),x=Dt.makeVList({positionType:"shift",positionData:-u,children:[{type:"elem",elem:a,marginRight:b}]},e)}var T=me(l,"right")||"mord";return Dt.makeSpan([T],[l,Dt.makeSpan(["msupsub"],[x])],e)},mathmlBuilder:function(t,e){var r,a=!1,n=Vt(t.base,"horizBrace");n&&!!t.sup===n.isOver&&(a=!0,r=n.isOver),!t.base||"op"!==t.base.type&&"operatorname"!==t.base.type||(t.base.parentIsSupSub=!0);var i,o=[Me(t.base,e)];if(t.sub&&o.push(Me(t.sub,e)),t.sup&&o.push(Me(t.sup,e)),a)i=r?"mover":"munder";else if(t.sub)if(t.sup){var s=t.base;i=s&&"op"===s.type&&s.limits&&e.style===w.DISPLAY?"munderover":s&&"operatorname"===s.type&&s.alwaysHandleSupSub&&(e.style===w.DISPLAY||s.limits)?"munderover":"msubsup"}else{var h=t.base;i=h&&"op"===h.type&&h.limits&&(e.style===w.DISPLAY||h.alwaysHandleSupSub)?"munder":h&&"operatorname"===h.type&&h.alwaysHandleSupSub&&(h.limits||e.style===w.DISPLAY)?"munder":"msub"}else{var l=t.base;i=l&&"op"===l.type&&l.limits&&(e.style===w.DISPLAY||l.alwaysHandleSupSub)?"mover":l&&"operatorname"===l.type&&l.alwaysHandleSupSub&&(l.limits||e.style===w.DISPLAY)?"mover":"msup"}return new ve.MathNode(i,o)}}),te({type:"atom",htmlBuilder:function(t,e){return Dt.mathsym(t.text,t.mode,e,["m"+t.family])},mathmlBuilder:function(t,e){var r=new ve.MathNode("mo",[be(t.text,t.mode)]);if("bin"===t.family){var a=we(t,e);"bold-italic"===a&&r.setAttribute("mathvariant",a)}else"punct"===t.family?r.setAttribute("separator","true"):"open"!==t.family&&"close"!==t.family||r.setAttribute("stretchy","false");return r}});var Zr={mi:"italic",mn:"normal",mtext:"normal"};te({type:"mathord",htmlBuilder:function(t,e){return Dt.makeOrd(t,e,"mathord")},mathmlBuilder:function(t,e){var r=new ve.MathNode("mi",[be(t.text,t.mode,e)]),a=we(t,e)||"italic";return a!==Zr[r.type]&&r.setAttribute("mathvariant",a),r}}),te({type:"textord",htmlBuilder:function(t,e){return Dt.makeOrd(t,e,"textord")},mathmlBuilder:function(t,e){var r,a=be(t.text,t.mode,e),n=we(t,e)||"normal";return r="text"===t.mode?new ve.MathNode("mtext",[a]):/[0-9]/.test(t.text)?new ve.MathNode("mn",[a]):"\\prime"===t.text?new ve.MathNode("mo",[a]):new ve.MathNode("mi",[a]),n!==Zr[r.type]&&r.setAttribute("mathvariant",n),r}});var Kr={"\\nobreak":"nobreak","\\allowbreak":"allowbreak"},Jr={" ":{},"\\ ":{},"~":{className:"nobreak"},"\\space":{},"\\nobreakspace":{className:"nobreak"}};te({type:"spacing",htmlBuilder:function(t,e){if(Jr.hasOwnProperty(t.text)){var r=Jr[t.text].className||"";if("text"===t.mode){var a=Dt.makeOrd(t,e,"textord");return a.classes.push(r),a}return Dt.makeSpan(["mspace",r],[Dt.mathsym(t.text,t.mode,e)],e)}if(Kr.hasOwnProperty(t.text))return Dt.makeSpan(["mspace",Kr[t.text]],[],e);throw new o('Unknown type of space "'+t.text+'"')},mathmlBuilder:function(t,e){if(!Jr.hasOwnProperty(t.text)){if(Kr.hasOwnProperty(t.text))return new ve.MathNode("mspace");throw new o('Unknown type of space "'+t.text+'"')}return new ve.MathNode("mtext",[new ve.TextNode("\xa0")])}});var Qr=function(){var t=new ve.MathNode("mtd",[]);return t.setAttribute("width","50%"),t};te({type:"tag",mathmlBuilder:function(t,e){var r=new ve.MathNode("mtable",[new ve.MathNode("mtr",[Qr(),new ve.MathNode("mtd",[Se(t.body,e)]),Qr(),new ve.MathNode("mtd",[Se(t.tag,e)])])]);return r.setAttribute("width","100%"),r}});var ta={"\\text":void 0,"\\textrm":"textrm","\\textsf":"textsf","\\texttt":"texttt","\\textnormal":"textrm"},ea={"\\textbf":"textbf","\\textmd":"textmd"},ra={"\\textit":"textit","\\textup":"textup"},aa=function(t,e){var r=t.font;return r?ta[r]?e.withTextFontFamily(ta[r]):ea[r]?e.withTextFontWeight(ea[r]):e.withTextFontShape(ra[r]):e};Qt({type:"text",names:["\\text","\\textrm","\\textsf","\\texttt","\\textnormal","\\textbf","\\textmd","\\textit","\\textup"],props:{numArgs:1,argTypes:["text"],greediness:2,allowedInText:!0},handler:function(t,e){var r=t.parser,a=t.funcName,n=e[0];return{type:"text",mode:r.mode,body:ee(n),font:a}},htmlBuilder:function(t,e){var r=aa(t,e),a=se(t.body,r,!0);return Dt.makeSpan(["mord","text"],Dt.tryCombineChars(a),r)},mathmlBuilder:function(t,e){var r=aa(t,e);return Se(t.body,r)}}),Qt({type:"underline",names:["\\underline"],props:{numArgs:1,allowedInText:!0},handler:function(t,e){return{type:"underline",mode:t.parser.mode,body:e[0]}},htmlBuilder:function(t,e){var r=ue(t.body,e),a=Dt.makeLineSpan("underline-line",e),n=e.fontMetrics().defaultRuleThickness,i=Dt.makeVList({positionType:"top",positionData:r.height,children:[{type:"kern",size:n},{type:"elem",elem:a},{type:"kern",size:3*n},{type:"elem",elem:r}]},e);return Dt.makeSpan(["mord","underline"],[i],e)},mathmlBuilder:function(t,e){var r=new ve.MathNode("mo",[new ve.TextNode("\u203e")]);r.setAttribute("stretchy","true");var a=new ve.MathNode("munder",[Me(t.body,e),r]);return a.setAttribute("accentunder","true"),a}}),Qt({type:"verb",names:["\\verb"],props:{numArgs:0,allowedInText:!0},handler:function(t,e,r){throw new o("\\verb ended by end of line instead of matching delimiter")},htmlBuilder:function(t,e){for(var r=na(t),a=[],n=e.havingStyle(e.style.text()),i=0;i0&&(this.undefStack[this.undefStack.length-1][t]=e)}else{var n=this.undefStack[this.undefStack.length-1];n&&!n.hasOwnProperty(t)&&(n[t]=this.current[t])}this.current[t]=e},t}(),ca={},ua=ca;function pa(t,e){ca[t]=e}pa("\\@firstoftwo",function(t){return{tokens:t.consumeArgs(2)[0],numArgs:0}}),pa("\\@secondoftwo",function(t){return{tokens:t.consumeArgs(2)[1],numArgs:0}}),pa("\\@ifnextchar",function(t){var e=t.consumeArgs(3),r=t.future();return 1===e[0].length&&e[0][0].text===r.text?{tokens:e[1],numArgs:0}:{tokens:e[2],numArgs:0}}),pa("\\@ifstar","\\@ifnextchar *{\\@firstoftwo{#1}}"),pa("\\TextOrMath",function(t){var e=t.consumeArgs(2);return"text"===t.mode?{tokens:e[0],numArgs:0}:{tokens:e[1],numArgs:0}});var da={0:0,1:1,2:2,3:3,4:4,5:5,6:6,7:7,8:8,9:9,a:10,A:10,b:11,B:11,c:12,C:12,d:13,D:13,e:14,E:14,f:15,F:15};pa("\\char",function(t){var e,r=t.popToken(),a="";if("'"===r.text)e=8,r=t.popToken();else if('"'===r.text)e=16,r=t.popToken();else if("`"===r.text)if("\\"===(r=t.popToken()).text[0])a=r.text.charCodeAt(1);else{if("EOF"===r.text)throw new o("\\char` missing argument");a=r.text.charCodeAt(0)}else e=10;if(e){if(null==(a=da[r.text])||a>=e)throw new o("Invalid base-"+e+" digit "+r.text);for(var n;null!=(n=da[t.future().text])&&n":"\\dotsb","-":"\\dotsb","*":"\\dotsb",":":"\\dotsb","\\DOTSB":"\\dotsb","\\coprod":"\\dotsb","\\bigvee":"\\dotsb","\\bigwedge":"\\dotsb","\\biguplus":"\\dotsb","\\bigcap":"\\dotsb","\\bigcup":"\\dotsb","\\prod":"\\dotsb","\\sum":"\\dotsb","\\bigotimes":"\\dotsb","\\bigoplus":"\\dotsb","\\bigodot":"\\dotsb","\\bigsqcup":"\\dotsb","\\And":"\\dotsb","\\longrightarrow":"\\dotsb","\\Longrightarrow":"\\dotsb","\\longleftarrow":"\\dotsb","\\Longleftarrow":"\\dotsb","\\longleftrightarrow":"\\dotsb","\\Longleftrightarrow":"\\dotsb","\\mapsto":"\\dotsb","\\longmapsto":"\\dotsb","\\hookrightarrow":"\\dotsb","\\doteq":"\\dotsb","\\mathbin":"\\dotsb","\\mathrel":"\\dotsb","\\relbar":"\\dotsb","\\Relbar":"\\dotsb","\\xrightarrow":"\\dotsb","\\xleftarrow":"\\dotsb","\\DOTSI":"\\dotsi","\\int":"\\dotsi","\\oint":"\\dotsi","\\iint":"\\dotsi","\\iiint":"\\dotsi","\\iiiint":"\\dotsi","\\idotsint":"\\dotsi","\\DOTSX":"\\dotsx"};pa("\\dots",function(t){var e="\\dotso",r=t.expandAfterFuture().text;return r in xa?e=xa[r]:"\\not"===r.substr(0,4)?e="\\dotsb":r in j.math&&c.contains(["bin","rel"],j.math[r].group)&&(e="\\dotsb"),e});var va={")":!0,"]":!0,"\\rbrack":!0,"\\}":!0,"\\rbrace":!0,"\\rangle":!0,"\\rceil":!0,"\\rfloor":!0,"\\rgroup":!0,"\\rmoustache":!0,"\\right":!0,"\\bigr":!0,"\\biggr":!0,"\\Bigr":!0,"\\Biggr":!0,$:!0,";":!0,".":!0,",":!0};pa("\\dotso",function(t){return t.future().text in va?"\\ldots\\,":"\\ldots"}),pa("\\dotsc",function(t){var e=t.future().text;return e in va&&","!==e?"\\ldots\\,":"\\ldots"}),pa("\\cdots",function(t){return t.future().text in va?"\\@cdots\\,":"\\@cdots"}),pa("\\dotsb","\\cdots"),pa("\\dotsm","\\cdots"),pa("\\dotsi","\\!\\cdots"),pa("\\dotsx","\\ldots\\,"),pa("\\DOTSI","\\relax"),pa("\\DOTSB","\\relax"),pa("\\DOTSX","\\relax"),pa("\\tmspace","\\TextOrMath{\\kern#1#3}{\\mskip#1#2}\\relax"),pa("\\,","\\tmspace+{3mu}{.1667em}"),pa("\\thinspace","\\,"),pa("\\>","\\mskip{4mu}"),pa("\\:","\\tmspace+{4mu}{.2222em}"),pa("\\medspace","\\:"),pa("\\;","\\tmspace+{5mu}{.2777em}"),pa("\\thickspace","\\;"),pa("\\!","\\tmspace-{3mu}{.1667em}"),pa("\\negthinspace","\\!"),pa("\\negmedspace","\\tmspace-{4mu}{.2222em}"),pa("\\negthickspace","\\tmspace-{5mu}{.277em}"),pa("\\enspace","\\kern.5em "),pa("\\enskip","\\hskip.5em\\relax"),pa("\\quad","\\hskip1em\\relax"),pa("\\qquad","\\hskip2em\\relax"),pa("\\tag","\\@ifstar\\tag@literal\\tag@paren"),pa("\\tag@paren","\\tag@literal{({#1})}"),pa("\\tag@literal",function(t){if(t.macros.get("\\df@tag"))throw new o("Multiple \\tag");return"\\gdef\\df@tag{\\text{#1}}"}),pa("\\bmod","\\mathchoice{\\mskip1mu}{\\mskip1mu}{\\mskip5mu}{\\mskip5mu}\\mathbin{\\rm mod}\\mathchoice{\\mskip1mu}{\\mskip1mu}{\\mskip5mu}{\\mskip5mu}"),pa("\\pod","\\allowbreak\\mathchoice{\\mkern18mu}{\\mkern8mu}{\\mkern8mu}{\\mkern8mu}(#1)"),pa("\\pmod","\\pod{{\\rm mod}\\mkern6mu#1}"),pa("\\mod","\\allowbreak\\mathchoice{\\mkern18mu}{\\mkern12mu}{\\mkern12mu}{\\mkern12mu}{\\rm mod}\\,\\,#1"),pa("\\pmb","\\html@mathml{\\@binrel{#1}{\\mathrlap{#1}\\kern0.5px#1}}{\\mathbf{#1}}"),pa("\\\\","\\newline"),pa("\\TeX","\\textrm{\\html@mathml{T\\kern-.1667em\\raisebox{-.5ex}{E}\\kern-.125emX}{TeX}}");var ba=F["Main-Regular"]["T".charCodeAt(0)][1]-.7*F["Main-Regular"]["A".charCodeAt(0)][1]+"em";pa("\\LaTeX","\\textrm{\\html@mathml{L\\kern-.36em\\raisebox{"+ba+"}{\\scriptstyle A}\\kern-.15em\\TeX}{LaTeX}}"),pa("\\KaTeX","\\textrm{\\html@mathml{K\\kern-.17em\\raisebox{"+ba+"}{\\scriptstyle A}\\kern-.15em\\TeX}{KaTeX}}"),pa("\\hspace","\\@ifstar\\@hspacer\\@hspace"),pa("\\@hspace","\\hskip #1\\relax"),pa("\\@hspacer","\\rule{0pt}{0pt}\\hskip #1\\relax"),pa("\\ordinarycolon",":"),pa("\\vcentcolon","\\mathrel{\\mathop\\ordinarycolon}"),pa("\\dblcolon",'\\html@mathml{\\mathrel{\\vcentcolon\\mathrel{\\mkern-.9mu}\\vcentcolon}}{\\mathop{\\char"2237}}'),pa("\\coloneqq",'\\html@mathml{\\mathrel{\\vcentcolon\\mathrel{\\mkern-1.2mu}=}}{\\mathop{\\char"2254}}'),pa("\\Coloneqq",'\\html@mathml{\\mathrel{\\dblcolon\\mathrel{\\mkern-1.2mu}=}}{\\mathop{\\char"2237\\char"3d}}'),pa("\\coloneq",'\\html@mathml{\\mathrel{\\vcentcolon\\mathrel{\\mkern-1.2mu}\\mathrel{-}}}{\\mathop{\\char"3a\\char"2212}}'),pa("\\Coloneq",'\\html@mathml{\\mathrel{\\dblcolon\\mathrel{\\mkern-1.2mu}\\mathrel{-}}}{\\mathop{\\char"2237\\char"2212}}'),pa("\\eqqcolon",'\\html@mathml{\\mathrel{=\\mathrel{\\mkern-1.2mu}\\vcentcolon}}{\\mathop{\\char"2255}}'),pa("\\Eqqcolon",'\\html@mathml{\\mathrel{=\\mathrel{\\mkern-1.2mu}\\dblcolon}}{\\mathop{\\char"3d\\char"2237}}'),pa("\\eqcolon",'\\html@mathml{\\mathrel{\\mathrel{-}\\mathrel{\\mkern-1.2mu}\\vcentcolon}}{\\mathop{\\char"2239}}'),pa("\\Eqcolon",'\\html@mathml{\\mathrel{\\mathrel{-}\\mathrel{\\mkern-1.2mu}\\dblcolon}}{\\mathop{\\char"2212\\char"2237}}'),pa("\\colonapprox",'\\html@mathml{\\mathrel{\\vcentcolon\\mathrel{\\mkern-1.2mu}\\approx}}{\\mathop{\\char"3a\\char"2248}}'),pa("\\Colonapprox",'\\html@mathml{\\mathrel{\\dblcolon\\mathrel{\\mkern-1.2mu}\\approx}}{\\mathop{\\char"2237\\char"2248}}'),pa("\\colonsim",'\\html@mathml{\\mathrel{\\vcentcolon\\mathrel{\\mkern-1.2mu}\\sim}}{\\mathop{\\char"3a\\char"223c}}'),pa("\\Colonsim",'\\html@mathml{\\mathrel{\\dblcolon\\mathrel{\\mkern-1.2mu}\\sim}}{\\mathop{\\char"2237\\char"223c}}'),pa("\u2237","\\dblcolon"),pa("\u2239","\\eqcolon"),pa("\u2254","\\coloneqq"),pa("\u2255","\\eqqcolon"),pa("\u2a74","\\Coloneqq"),pa("\\ratio","\\vcentcolon"),pa("\\coloncolon","\\dblcolon"),pa("\\colonequals","\\coloneqq"),pa("\\coloncolonequals","\\Coloneqq"),pa("\\equalscolon","\\eqqcolon"),pa("\\equalscoloncolon","\\Eqqcolon"),pa("\\colonminus","\\coloneq"),pa("\\coloncolonminus","\\Coloneq"),pa("\\minuscolon","\\eqcolon"),pa("\\minuscoloncolon","\\Eqcolon"),pa("\\coloncolonapprox","\\Colonapprox"),pa("\\coloncolonsim","\\Colonsim"),pa("\\simcolon","\\mathrel{\\sim\\mathrel{\\mkern-1.2mu}\\vcentcolon}"),pa("\\simcoloncolon","\\mathrel{\\sim\\mathrel{\\mkern-1.2mu}\\dblcolon}"),pa("\\approxcolon","\\mathrel{\\approx\\mathrel{\\mkern-1.2mu}\\vcentcolon}"),pa("\\approxcoloncolon","\\mathrel{\\approx\\mathrel{\\mkern-1.2mu}\\dblcolon}"),pa("\\notni","\\html@mathml{\\not\\ni}{\\mathrel{\\char`\u220c}}"),pa("\\limsup","\\DOTSB\\operatorname*{lim\\,sup}"),pa("\\liminf","\\DOTSB\\operatorname*{lim\\,inf}"),pa("\\gvertneqq","\\html@mathml{\\@gvertneqq}{\u2269}"),pa("\\lvertneqq","\\html@mathml{\\@lvertneqq}{\u2268}"),pa("\\ngeqq","\\html@mathml{\\@ngeqq}{\u2271}"),pa("\\ngeqslant","\\html@mathml{\\@ngeqslant}{\u2271}"),pa("\\nleqq","\\html@mathml{\\@nleqq}{\u2270}"),pa("\\nleqslant","\\html@mathml{\\@nleqslant}{\u2270}"),pa("\\nshortmid","\\html@mathml{\\@nshortmid}{\u2224}"),pa("\\nshortparallel","\\html@mathml{\\@nshortparallel}{\u2226}"),pa("\\nsubseteqq","\\html@mathml{\\@nsubseteqq}{\u2288}"),pa("\\nsupseteqq","\\html@mathml{\\@nsupseteqq}{\u2289}"),pa("\\varsubsetneq","\\html@mathml{\\@varsubsetneq}{\u228a}"),pa("\\varsubsetneqq","\\html@mathml{\\@varsubsetneqq}{\u2acb}"),pa("\\varsupsetneq","\\html@mathml{\\@varsupsetneq}{\u228b}"),pa("\\varsupsetneqq","\\html@mathml{\\@varsupsetneqq}{\u2acc}"),pa("\\llbracket","\\html@mathml{\\mathopen{[\\mkern-3.2mu[}}{\\mathopen{\\char`\u27e6}}"),pa("\\rrbracket","\\html@mathml{\\mathclose{]\\mkern-3.2mu]}}{\\mathclose{\\char`\u27e7}}"),pa("\u27e6","\\llbracket"),pa("\u27e7","\\rrbracket"),pa("\\lBrace","\\html@mathml{\\mathopen{\\{\\mkern-3.2mu[}}{\\mathopen{\\char`\u2983}}"),pa("\\rBrace","\\html@mathml{\\mathclose{]\\mkern-3.2mu\\}}}{\\mathclose{\\char`\u2984}}"),pa("\u2983","\\lBrace"),pa("\u2984","\\rBrace"),pa("\\darr","\\downarrow"),pa("\\dArr","\\Downarrow"),pa("\\Darr","\\Downarrow"),pa("\\lang","\\langle"),pa("\\rang","\\rangle"),pa("\\uarr","\\uparrow"),pa("\\uArr","\\Uparrow"),pa("\\Uarr","\\Uparrow"),pa("\\N","\\mathbb{N}"),pa("\\R","\\mathbb{R}"),pa("\\Z","\\mathbb{Z}"),pa("\\alef","\\aleph"),pa("\\alefsym","\\aleph"),pa("\\Alpha","\\mathrm{A}"),pa("\\Beta","\\mathrm{B}"),pa("\\bull","\\bullet"),pa("\\Chi","\\mathrm{X}"),pa("\\clubs","\\clubsuit"),pa("\\cnums","\\mathbb{C}"),pa("\\Complex","\\mathbb{C}"),pa("\\Dagger","\\ddagger"),pa("\\diamonds","\\diamondsuit"),pa("\\empty","\\emptyset"),pa("\\Epsilon","\\mathrm{E}"),pa("\\Eta","\\mathrm{H}"),pa("\\exist","\\exists"),pa("\\harr","\\leftrightarrow"),pa("\\hArr","\\Leftrightarrow"),pa("\\Harr","\\Leftrightarrow"),pa("\\hearts","\\heartsuit"),pa("\\image","\\Im"),pa("\\infin","\\infty"),pa("\\Iota","\\mathrm{I}"),pa("\\isin","\\in"),pa("\\Kappa","\\mathrm{K}"),pa("\\larr","\\leftarrow"),pa("\\lArr","\\Leftarrow"),pa("\\Larr","\\Leftarrow"),pa("\\lrarr","\\leftrightarrow"),pa("\\lrArr","\\Leftrightarrow"),pa("\\Lrarr","\\Leftrightarrow"),pa("\\Mu","\\mathrm{M}"),pa("\\natnums","\\mathbb{N}"),pa("\\Nu","\\mathrm{N}"),pa("\\Omicron","\\mathrm{O}"),pa("\\plusmn","\\pm"),pa("\\rarr","\\rightarrow"),pa("\\rArr","\\Rightarrow"),pa("\\Rarr","\\Rightarrow"),pa("\\real","\\Re"),pa("\\reals","\\mathbb{R}"),pa("\\Reals","\\mathbb{R}"),pa("\\Rho","\\mathrm{P}"),pa("\\sdot","\\cdot"),pa("\\sect","\\S"),pa("\\spades","\\spadesuit"),pa("\\sub","\\subset"),pa("\\sube","\\subseteq"),pa("\\supe","\\supseteq"),pa("\\Tau","\\mathrm{T}"),pa("\\thetasym","\\vartheta"),pa("\\weierp","\\wp"),pa("\\Zeta","\\mathrm{Z}"),pa("\\argmin","\\DOTSB\\operatorname*{arg\\,min}"),pa("\\argmax","\\DOTSB\\operatorname*{arg\\,max}"),pa("\\plim","\\DOTSB\\mathop{\\operatorname{plim}}\\limits"),pa("\\blue","\\textcolor{##6495ed}{#1}"),pa("\\orange","\\textcolor{##ffa500}{#1}"),pa("\\pink","\\textcolor{##ff00af}{#1}"),pa("\\red","\\textcolor{##df0030}{#1}"),pa("\\green","\\textcolor{##28ae7b}{#1}"),pa("\\gray","\\textcolor{gray}{#1}"),pa("\\purple","\\textcolor{##9d38bd}{#1}"),pa("\\blueA","\\textcolor{##ccfaff}{#1}"),pa("\\blueB","\\textcolor{##80f6ff}{#1}"),pa("\\blueC","\\textcolor{##63d9ea}{#1}"),pa("\\blueD","\\textcolor{##11accd}{#1}"),pa("\\blueE","\\textcolor{##0c7f99}{#1}"),pa("\\tealA","\\textcolor{##94fff5}{#1}"),pa("\\tealB","\\textcolor{##26edd5}{#1}"),pa("\\tealC","\\textcolor{##01d1c1}{#1}"),pa("\\tealD","\\textcolor{##01a995}{#1}"),pa("\\tealE","\\textcolor{##208170}{#1}"),pa("\\greenA","\\textcolor{##b6ffb0}{#1}"),pa("\\greenB","\\textcolor{##8af281}{#1}"),pa("\\greenC","\\textcolor{##74cf70}{#1}"),pa("\\greenD","\\textcolor{##1fab54}{#1}"),pa("\\greenE","\\textcolor{##0d923f}{#1}"),pa("\\goldA","\\textcolor{##ffd0a9}{#1}"),pa("\\goldB","\\textcolor{##ffbb71}{#1}"),pa("\\goldC","\\textcolor{##ff9c39}{#1}"),pa("\\goldD","\\textcolor{##e07d10}{#1}"),pa("\\goldE","\\textcolor{##a75a05}{#1}"),pa("\\redA","\\textcolor{##fca9a9}{#1}"),pa("\\redB","\\textcolor{##ff8482}{#1}"),pa("\\redC","\\textcolor{##f9685d}{#1}"),pa("\\redD","\\textcolor{##e84d39}{#1}"),pa("\\redE","\\textcolor{##bc2612}{#1}"),pa("\\maroonA","\\textcolor{##ffbde0}{#1}"),pa("\\maroonB","\\textcolor{##ff92c6}{#1}"),pa("\\maroonC","\\textcolor{##ed5fa6}{#1}"),pa("\\maroonD","\\textcolor{##ca337c}{#1}"),pa("\\maroonE","\\textcolor{##9e034e}{#1}"),pa("\\purpleA","\\textcolor{##ddd7ff}{#1}"),pa("\\purpleB","\\textcolor{##c6b9fc}{#1}"),pa("\\purpleC","\\textcolor{##aa87ff}{#1}"),pa("\\purpleD","\\textcolor{##7854ab}{#1}"),pa("\\purpleE","\\textcolor{##543b78}{#1}"),pa("\\mintA","\\textcolor{##f5f9e8}{#1}"),pa("\\mintB","\\textcolor{##edf2df}{#1}"),pa("\\mintC","\\textcolor{##e0e5cc}{#1}"),pa("\\grayA","\\textcolor{##f6f7f7}{#1}"),pa("\\grayB","\\textcolor{##f0f1f2}{#1}"),pa("\\grayC","\\textcolor{##e3e5e6}{#1}"),pa("\\grayD","\\textcolor{##d6d8da}{#1}"),pa("\\grayE","\\textcolor{##babec2}{#1}"),pa("\\grayF","\\textcolor{##888d93}{#1}"),pa("\\grayG","\\textcolor{##626569}{#1}"),pa("\\grayH","\\textcolor{##3b3e40}{#1}"),pa("\\grayI","\\textcolor{##21242c}{#1}"),pa("\\kaBlue","\\textcolor{##314453}{#1}"),pa("\\kaGreen","\\textcolor{##71B307}{#1}");var ya={"\\relax":!0,"^":!0,_:!0,"\\limits":!0,"\\nolimits":!0},wa=function(){function t(t,e,r){this.settings=void 0,this.expansionCount=void 0,this.lexer=void 0,this.macros=void 0,this.stack=void 0,this.mode=void 0,this.settings=e,this.expansionCount=0,this.feed(t),this.macros=new ma(ua,e.macros),this.mode=r,this.stack=[]}var e=t.prototype;return e.feed=function(t){this.lexer=new la(t,this.settings)},e.switchMode=function(t){this.mode=t},e.beginGroup=function(){this.macros.beginGroup()},e.endGroup=function(){this.macros.endGroup()},e.future=function(){return 0===this.stack.length&&this.pushToken(this.lexer.lex()),this.stack[this.stack.length-1]},e.popToken=function(){return this.future(),this.stack.pop()},e.pushToken=function(t){this.stack.push(t)},e.pushTokens=function(t){var e;(e=this.stack).push.apply(e,t)},e.consumeSpaces=function(){for(;;){if(" "!==this.future().text)break;this.stack.pop()}},e.consumeArgs=function(t){for(var e=[],r=0;rthis.settings.maxExpand)throw new o("Too many expansions: infinite loop or need to increase maxExpand setting");var a=r.tokens;if(r.numArgs)for(var n=this.consumeArgs(r.numArgs),i=(a=a.slice()).length-1;i>=0;--i){var s=a[i];if("#"===s.text){if(0===i)throw new o("Incomplete placeholder at end of macro body",s);if("#"===(s=a[--i]).text)a.splice(i+1,1);else{if(!/^[1-9]$/.test(s.text))throw new o("Not a valid argument number",s);var h;(h=a).splice.apply(h,[i,2].concat(n[+s.text-1]))}}}return this.pushTokens(a),a},e.expandAfterFuture=function(){return this.expandOnce(),this.future()},e.expandNextToken=function(){for(;;){var t=this.expandOnce();if(t instanceof n){if("\\relax"!==t.text)return this.stack.pop();this.stack.pop()}}throw new Error},e.expandMacro=function(t){if(this.macros.get(t)){var e=[],r=this.stack.length;for(this.pushToken(new n(t));this.stack.length>r;){this.expandOnce()instanceof n&&e.push(this.stack.pop())}return e}},e.expandMacroAsText=function(t){var e=this.expandMacro(t);return e?e.map(function(t){return t.text}).join(""):e},e._getExpansion=function(t){var e=this.macros.get(t);if(null==e)return e;var r="function"==typeof e?e(this):e;if("string"==typeof r){var a=0;if(-1!==r.indexOf("#"))for(var n=r.replace(/##/g,"");-1!==n.indexOf("#"+(a+1));)++a;for(var i=new la(r,this.settings),o=[],s=i.lex();"EOF"!==s.text;)o.push(s),s=i.lex();return o.reverse(),{tokens:o,numArgs:a}}return r},e.isDefined=function(t){return this.macros.has(t)||ia.hasOwnProperty(t)||j.math.hasOwnProperty(t)||j.text.hasOwnProperty(t)||ya.hasOwnProperty(t)},t}(),ka={"\u0301":{text:"\\'",math:"\\acute"},"\u0300":{text:"\\`",math:"\\grave"},"\u0308":{text:'\\"',math:"\\ddot"},"\u0303":{text:"\\~",math:"\\tilde"},"\u0304":{text:"\\=",math:"\\bar"},"\u0306":{text:"\\u",math:"\\breve"},"\u030c":{text:"\\v",math:"\\check"},"\u0302":{text:"\\^",math:"\\hat"},"\u0307":{text:"\\.",math:"\\dot"},"\u030a":{text:"\\r",math:"\\mathring"},"\u030b":{text:"\\H"}},Sa={"\xe1":"a\u0301","\xe0":"a\u0300","\xe4":"a\u0308","\u01df":"a\u0308\u0304","\xe3":"a\u0303","\u0101":"a\u0304","\u0103":"a\u0306","\u1eaf":"a\u0306\u0301","\u1eb1":"a\u0306\u0300","\u1eb5":"a\u0306\u0303","\u01ce":"a\u030c","\xe2":"a\u0302","\u1ea5":"a\u0302\u0301","\u1ea7":"a\u0302\u0300","\u1eab":"a\u0302\u0303","\u0227":"a\u0307","\u01e1":"a\u0307\u0304","\xe5":"a\u030a","\u01fb":"a\u030a\u0301","\u1e03":"b\u0307","\u0107":"c\u0301","\u010d":"c\u030c","\u0109":"c\u0302","\u010b":"c\u0307","\u010f":"d\u030c","\u1e0b":"d\u0307","\xe9":"e\u0301","\xe8":"e\u0300","\xeb":"e\u0308","\u1ebd":"e\u0303","\u0113":"e\u0304","\u1e17":"e\u0304\u0301","\u1e15":"e\u0304\u0300","\u0115":"e\u0306","\u011b":"e\u030c","\xea":"e\u0302","\u1ebf":"e\u0302\u0301","\u1ec1":"e\u0302\u0300","\u1ec5":"e\u0302\u0303","\u0117":"e\u0307","\u1e1f":"f\u0307","\u01f5":"g\u0301","\u1e21":"g\u0304","\u011f":"g\u0306","\u01e7":"g\u030c","\u011d":"g\u0302","\u0121":"g\u0307","\u1e27":"h\u0308","\u021f":"h\u030c","\u0125":"h\u0302","\u1e23":"h\u0307","\xed":"i\u0301","\xec":"i\u0300","\xef":"i\u0308","\u1e2f":"i\u0308\u0301","\u0129":"i\u0303","\u012b":"i\u0304","\u012d":"i\u0306","\u01d0":"i\u030c","\xee":"i\u0302","\u01f0":"j\u030c","\u0135":"j\u0302","\u1e31":"k\u0301","\u01e9":"k\u030c","\u013a":"l\u0301","\u013e":"l\u030c","\u1e3f":"m\u0301","\u1e41":"m\u0307","\u0144":"n\u0301","\u01f9":"n\u0300","\xf1":"n\u0303","\u0148":"n\u030c","\u1e45":"n\u0307","\xf3":"o\u0301","\xf2":"o\u0300","\xf6":"o\u0308","\u022b":"o\u0308\u0304","\xf5":"o\u0303","\u1e4d":"o\u0303\u0301","\u1e4f":"o\u0303\u0308","\u022d":"o\u0303\u0304","\u014d":"o\u0304","\u1e53":"o\u0304\u0301","\u1e51":"o\u0304\u0300","\u014f":"o\u0306","\u01d2":"o\u030c","\xf4":"o\u0302","\u1ed1":"o\u0302\u0301","\u1ed3":"o\u0302\u0300","\u1ed7":"o\u0302\u0303","\u022f":"o\u0307","\u0231":"o\u0307\u0304","\u0151":"o\u030b","\u1e55":"p\u0301","\u1e57":"p\u0307","\u0155":"r\u0301","\u0159":"r\u030c","\u1e59":"r\u0307","\u015b":"s\u0301","\u1e65":"s\u0301\u0307","\u0161":"s\u030c","\u1e67":"s\u030c\u0307","\u015d":"s\u0302","\u1e61":"s\u0307","\u1e97":"t\u0308","\u0165":"t\u030c","\u1e6b":"t\u0307","\xfa":"u\u0301","\xf9":"u\u0300","\xfc":"u\u0308","\u01d8":"u\u0308\u0301","\u01dc":"u\u0308\u0300","\u01d6":"u\u0308\u0304","\u01da":"u\u0308\u030c","\u0169":"u\u0303","\u1e79":"u\u0303\u0301","\u016b":"u\u0304","\u1e7b":"u\u0304\u0308","\u016d":"u\u0306","\u01d4":"u\u030c","\xfb":"u\u0302","\u016f":"u\u030a","\u0171":"u\u030b","\u1e7d":"v\u0303","\u1e83":"w\u0301","\u1e81":"w\u0300","\u1e85":"w\u0308","\u0175":"w\u0302","\u1e87":"w\u0307","\u1e98":"w\u030a","\u1e8d":"x\u0308","\u1e8b":"x\u0307","\xfd":"y\u0301","\u1ef3":"y\u0300","\xff":"y\u0308","\u1ef9":"y\u0303","\u0233":"y\u0304","\u0177":"y\u0302","\u1e8f":"y\u0307","\u1e99":"y\u030a","\u017a":"z\u0301","\u017e":"z\u030c","\u1e91":"z\u0302","\u017c":"z\u0307","\xc1":"A\u0301","\xc0":"A\u0300","\xc4":"A\u0308","\u01de":"A\u0308\u0304","\xc3":"A\u0303","\u0100":"A\u0304","\u0102":"A\u0306","\u1eae":"A\u0306\u0301","\u1eb0":"A\u0306\u0300","\u1eb4":"A\u0306\u0303","\u01cd":"A\u030c","\xc2":"A\u0302","\u1ea4":"A\u0302\u0301","\u1ea6":"A\u0302\u0300","\u1eaa":"A\u0302\u0303","\u0226":"A\u0307","\u01e0":"A\u0307\u0304","\xc5":"A\u030a","\u01fa":"A\u030a\u0301","\u1e02":"B\u0307","\u0106":"C\u0301","\u010c":"C\u030c","\u0108":"C\u0302","\u010a":"C\u0307","\u010e":"D\u030c","\u1e0a":"D\u0307","\xc9":"E\u0301","\xc8":"E\u0300","\xcb":"E\u0308","\u1ebc":"E\u0303","\u0112":"E\u0304","\u1e16":"E\u0304\u0301","\u1e14":"E\u0304\u0300","\u0114":"E\u0306","\u011a":"E\u030c","\xca":"E\u0302","\u1ebe":"E\u0302\u0301","\u1ec0":"E\u0302\u0300","\u1ec4":"E\u0302\u0303","\u0116":"E\u0307","\u1e1e":"F\u0307","\u01f4":"G\u0301","\u1e20":"G\u0304","\u011e":"G\u0306","\u01e6":"G\u030c","\u011c":"G\u0302","\u0120":"G\u0307","\u1e26":"H\u0308","\u021e":"H\u030c","\u0124":"H\u0302","\u1e22":"H\u0307","\xcd":"I\u0301","\xcc":"I\u0300","\xcf":"I\u0308","\u1e2e":"I\u0308\u0301","\u0128":"I\u0303","\u012a":"I\u0304","\u012c":"I\u0306","\u01cf":"I\u030c","\xce":"I\u0302","\u0130":"I\u0307","\u0134":"J\u0302","\u1e30":"K\u0301","\u01e8":"K\u030c","\u0139":"L\u0301","\u013d":"L\u030c","\u1e3e":"M\u0301","\u1e40":"M\u0307","\u0143":"N\u0301","\u01f8":"N\u0300","\xd1":"N\u0303","\u0147":"N\u030c","\u1e44":"N\u0307","\xd3":"O\u0301","\xd2":"O\u0300","\xd6":"O\u0308","\u022a":"O\u0308\u0304","\xd5":"O\u0303","\u1e4c":"O\u0303\u0301","\u1e4e":"O\u0303\u0308","\u022c":"O\u0303\u0304","\u014c":"O\u0304","\u1e52":"O\u0304\u0301","\u1e50":"O\u0304\u0300","\u014e":"O\u0306","\u01d1":"O\u030c","\xd4":"O\u0302","\u1ed0":"O\u0302\u0301","\u1ed2":"O\u0302\u0300","\u1ed6":"O\u0302\u0303","\u022e":"O\u0307","\u0230":"O\u0307\u0304","\u0150":"O\u030b","\u1e54":"P\u0301","\u1e56":"P\u0307","\u0154":"R\u0301","\u0158":"R\u030c","\u1e58":"R\u0307","\u015a":"S\u0301","\u1e64":"S\u0301\u0307","\u0160":"S\u030c","\u1e66":"S\u030c\u0307","\u015c":"S\u0302","\u1e60":"S\u0307","\u0164":"T\u030c","\u1e6a":"T\u0307","\xda":"U\u0301","\xd9":"U\u0300","\xdc":"U\u0308","\u01d7":"U\u0308\u0301","\u01db":"U\u0308\u0300","\u01d5":"U\u0308\u0304","\u01d9":"U\u0308\u030c","\u0168":"U\u0303","\u1e78":"U\u0303\u0301","\u016a":"U\u0304","\u1e7a":"U\u0304\u0308","\u016c":"U\u0306","\u01d3":"U\u030c","\xdb":"U\u0302","\u016e":"U\u030a","\u0170":"U\u030b","\u1e7c":"V\u0303","\u1e82":"W\u0301","\u1e80":"W\u0300","\u1e84":"W\u0308","\u0174":"W\u0302","\u1e86":"W\u0307","\u1e8c":"X\u0308","\u1e8a":"X\u0307","\xdd":"Y\u0301","\u1ef2":"Y\u0300","\u0178":"Y\u0308","\u1ef8":"Y\u0303","\u0232":"Y\u0304","\u0176":"Y\u0302","\u1e8e":"Y\u0307","\u0179":"Z\u0301","\u017d":"Z\u030c","\u1e90":"Z\u0302","\u017b":"Z\u0307","\u03ac":"\u03b1\u0301","\u1f70":"\u03b1\u0300","\u1fb1":"\u03b1\u0304","\u1fb0":"\u03b1\u0306","\u03ad":"\u03b5\u0301","\u1f72":"\u03b5\u0300","\u03ae":"\u03b7\u0301","\u1f74":"\u03b7\u0300","\u03af":"\u03b9\u0301","\u1f76":"\u03b9\u0300","\u03ca":"\u03b9\u0308","\u0390":"\u03b9\u0308\u0301","\u1fd2":"\u03b9\u0308\u0300","\u1fd1":"\u03b9\u0304","\u1fd0":"\u03b9\u0306","\u03cc":"\u03bf\u0301","\u1f78":"\u03bf\u0300","\u03cd":"\u03c5\u0301","\u1f7a":"\u03c5\u0300","\u03cb":"\u03c5\u0308","\u03b0":"\u03c5\u0308\u0301","\u1fe2":"\u03c5\u0308\u0300","\u1fe1":"\u03c5\u0304","\u1fe0":"\u03c5\u0306","\u03ce":"\u03c9\u0301","\u1f7c":"\u03c9\u0300","\u038e":"\u03a5\u0301","\u1fea":"\u03a5\u0300","\u03ab":"\u03a5\u0308","\u1fe9":"\u03a5\u0304","\u1fe8":"\u03a5\u0306","\u038f":"\u03a9\u0301","\u1ffa":"\u03a9\u0300"},Ma=function(){function t(t,e){this.mode=void 0,this.gullet=void 0,this.settings=void 0,this.leftrightDepth=void 0,this.nextToken=void 0,this.mode="math",this.gullet=new wa(t,e,this.mode),this.settings=e,this.leftrightDepth=0}var e=t.prototype;return e.expect=function(t,e){if(void 0===e&&(e=!0),this.fetch().text!==t)throw new o("Expected '"+t+"', got '"+this.fetch().text+"'",this.fetch());e&&this.consume()},e.consume=function(){this.nextToken=null},e.fetch=function(){return null==this.nextToken&&(this.nextToken=this.gullet.expandNextToken()),this.nextToken},e.switchMode=function(t){this.mode=t,this.gullet.switchMode(t)},e.parse=function(){this.gullet.beginGroup(),this.settings.colorIsTextColor&&this.gullet.macros.set("\\color","\\textcolor");var t=this.parseExpression(!1);return this.expect("EOF"),this.gullet.endGroup(),t},e.parseExpression=function(e,r){for(var a=[];;){"math"===this.mode&&this.consumeSpaces();var n=this.fetch();if(-1!==t.endOfExpression.indexOf(n.text))break;if(r&&n.text===r)break;if(e&&ia[n.text]&&ia[n.text].infix)break;var i=this.parseAtom(r);if(!i)break;a.push(i)}return"text"===this.mode&&this.formLigatures(a),this.handleInfixNodes(a)},e.handleInfixNodes=function(t){for(var e,r=-1,a=0;a0&&!l||0===s&&!l&&"math"===this.mode,c=this.parseGroupOfType("argument to '"+t+"'",h,l,a,m);if(!c){if(l){i.push(null);continue}throw new o("Expected group after '"+t+"'",this.fetch())}(l?i:n).push(c)}return{args:n,optArgs:i}},e.parseGroupOfType=function(t,e,r,a,n){switch(e){case"color":return n&&this.consumeSpaces(),this.parseColorGroup(r);case"size":return n&&this.consumeSpaces(),this.parseSizeGroup(r);case"url":return this.parseUrlGroup(r,n);case"math":case"text":return this.parseGroup(t,r,a,void 0,e,n);case"hbox":var i=this.parseGroup(t,r,a,void 0,"text",n);return i?{type:"styling",mode:i.mode,body:[i],style:"text"}:i;case"raw":if(n&&this.consumeSpaces(),r&&"{"===this.fetch().text)return null;var s=this.parseStringGroup("raw",r,!0);if(s)return{type:"raw",mode:"text",string:s.text};throw new o("Expected raw group",this.fetch());case"original":case null:case void 0:return this.parseGroup(t,r,a,void 0,void 0,n);default:throw new o("Unknown group type as "+t,this.fetch())}},e.consumeSpaces=function(){for(;" "===this.fetch().text;)this.consume()},e.parseStringGroup=function(t,e,r){var a=e?"[":"{",n=e?"]":"}",i=this.fetch();if(i.text!==a){if(e)return null;if(r&&"EOF"!==i.text&&/[^{}[\]]/.test(i.text))return this.consume(),i}var s=this.mode;this.mode="text",this.expect(a);for(var h,l="",m=this.fetch(),c=0,u=m;(h=this.fetch()).text!==n||r&&c>0;){switch(h.text){case"EOF":throw new o("Unexpected end of input in "+t,m.range(u,l));case a:c++;break;case n:c--}l+=(u=h).text,this.consume()}return this.expect(n),this.mode=s,m.range(u,l)},e.parseRegexGroup=function(t,e){var r=this.mode;this.mode="text";for(var a,n=this.fetch(),i=n,s="";"EOF"!==(a=this.fetch()).text&&t.test(s+a.text);)s+=(i=a).text,this.consume();if(""===s)throw new o("Invalid "+e+": '"+n.text+"'",n);return this.mode=r,n.range(i,s)},e.parseColorGroup=function(t){var e=this.parseStringGroup("color",t);if(!e)return null;var r=/^(#[a-f0-9]{3}|#?[a-f0-9]{6}|[a-z]+)$/i.exec(e.text);if(!r)throw new o("Invalid color: '"+e.text+"'",e);var a=r[0];return/^[0-9a-f]{6}$/i.test(a)&&(a="#"+a),{type:"color-token",mode:this.mode,color:a}},e.parseSizeGroup=function(t){var e,r=!1;if(!(e=t||"{"===this.fetch().text?this.parseStringGroup("size",t):this.parseRegexGroup(/^[-+]? *(?:$|\d+|\d+\.\d*|\.\d*) *[a-z]{0,2} *$/,"size")))return null;t||0!==e.text.length||(e.text="0pt",r=!0);var a=/([-+]?) *(\d+(?:\.\d*)?|\.\d+) *([a-z]{2})/.exec(e.text);if(!a)throw new o("Invalid size: '"+e.text+"'",e);var n={number:+(a[1]+a[2]),unit:a[3]};if(!At(n))throw new o("Invalid unit: '"+n.unit+"'",e);return{type:"size",mode:this.mode,value:n,isBlank:r}},e.parseUrlGroup=function(t,e){this.gullet.lexer.setCatcode("%",13);var r=this.parseStringGroup("url",t,!0);if(this.gullet.lexer.setCatcode("%",14),!r)return null;var a=r.text.replace(/\\([#$%&~_^{}])/g,"$1");return{type:"url",mode:this.mode,url:a}},e.parseGroup=function(e,r,n,i,s,h){var l=this.mode;s&&this.switchMode(s),h&&this.consumeSpaces();var m,c=this.fetch(),u=c.text;if(r?"["===u:"{"===u||"\\begingroup"===u){this.consume();var p=t.endOfGroup[u];this.gullet.beginGroup();var d=this.parseExpression(!1,p),f=this.fetch();this.expect(p),this.gullet.endGroup(),m={type:"ordgroup",mode:this.mode,loc:a.range(c,f),body:d,semisimple:"\\begingroup"===u||void 0}}else if(r)m=null;else if(null==(m=this.parseFunction(i,e,n)||this.parseSymbol())&&"\\"===u[0]&&!ya.hasOwnProperty(u)){if(this.settings.throwOnError)throw new o("Undefined control sequence: "+u,c);m=this.formatUnsupportedCmd(u),this.consume()}return s&&this.switchMode(l),m},e.formLigatures=function(t){for(var e=t.length-1,r=0;r=0&&this.settings.reportNonstrict("unicodeTextInMathMode",'Latin-1/Unicode text character "'+e[0]+'" used in math mode',t);var h,l=j[this.mode][e].group,m=a.range(t);if(W.hasOwnProperty(l)){var c=l;h={type:"atom",mode:this.mode,family:c,loc:m,text:e}}else h={type:l,mode:this.mode,loc:m,text:e};i=h}else{if(!(e.charCodeAt(0)>=128))return null;this.settings.strict&&(M(e.charCodeAt(0))?"math"===this.mode&&this.settings.reportNonstrict("unicodeTextInMathMode",'Unicode text character "'+e[0]+'" used in math mode',t):this.settings.reportNonstrict("unknownSymbol",'Unrecognized Unicode character "'+e[0]+'" ('+e.charCodeAt(0)+")",t)),i={type:"textord",mode:"text",loc:a.range(t),text:e}}if(this.consume(),s)for(var u=0;u Tools-> Extensions&updates ->search Visual Studio Installer Projects + +Which is the Microsoft utility for building wix files. It creates a quite adequate wix setup by gui, in the spirit of the skeleton windows gui app. + +## [NSIS](https://nsis.sourceforge.io/Download) Nullsoft Scriptable Install System. + +People who know what they are doing seem to use this install system, and they + write nice installs with it. + +To build setup program: + +1. Build both x64 and Win32 Release configs +1. When you construct wallet.nsi in nullsoft, add it to your project. +1. When building a deliverable, Right click on the WalletSetup.nsi file in Visual Studio project and select properties. +1. Set Excluded from Build to No +1. OK Properties +1. Right click .nsi file again and choose Compile. +1. Set the .nsi file properties back to Excluded from Build. + +This manual building of the setup is due to the fact that we need both x64 +and Win32 exes for the setup program and Visual Studio doesn’t provide a +way to do this easily. + +# Package managers + +Lately, however, package managers have appeared: Conan and [vcPkg](https://blog.kitware.com/vcpkg-a-tool-to-build-open-source-libraries-on-windows/). Conan lacks wxWidgets, and has far fewer packages than [vcpkg](https://libraries.io/github/Microsoft/vcpkg). + +I have attempted to use package managers, and not found them very useful. It +is easier to deal with each package as its own unique special case. The +uniform abstraction that a package manager attempts to provide invariably +leaks badly, while piling cruft on top of the library. Rather than +simplifying library use, piles its own idiosyncratic complexification on top +of the complexities of the library, often inducing multiplicative complexity, + as one attempts to deal with the irregularities and particulars of a +particular library though a package manager that is unaware of and incapable +of dealing with the particularity of that particular package, and is +unshakeably convinced that the library is organized in way that is different +from the way it is in fact organized. + +# Multiprecision Arithmetic + +I will need multiprecision arithmetic if I represent information in a base or + dictionary that is not a power of two. + + [MPIR]:]http://mpir.org/ + {target="_blank"} + + [GMP]:https://gmplib.org +{target="_blank"} + +The best libraries are [GMP] for Linux and +[MPIR] for windows. These are reasonably +compatible, and generally only require very trivial changes to produce a Linux +version and a windows version. Boost attempts to make the changes invisible, +but adds needless complexity and overhead in doing so, and obstructs control. +MPIR has a Visual Studio repository on Github, and a separate Linux repository +on Github. GMP builds on a lot of obscure platforms, but not really supported +on Windows. + +For supporting Windows and Linux only, MPIR all the way is the way to go. For + compatibility with little used and obscure environments, you might want to +have your own custom thin layer that maps GMP integers and MPIR integers to +your integers, but that can wait till we have conquered the world. + +My most immediate need for MPIR is the extended Euclidean algorithm +for modular multiplicative inverse, which it, of course, supports, +`mpz_gcdext`, greatest common divisor extended, but which is deeply +hidden in the [documentation](http://www.mpir.org/mpir-3.0.0.pdf). + +# [wxWidgets](./libraries/building_and_using_libraries.html#instructions-for-wxwidgets) + +# Networking + +## notbit client + +A bitmessage client written in C. Designed to run on a linux mail server +and interface bitmessage to mail. Has no UI, intended to be used with the linux mail UI. + +Unfortunately, setting up a linux mail server is a pain in the ass. Needs the Zooko UI. + +But its library contains everything you need to share data around a group of people, many of them behind NATs. + +Does not implement NAT penetration. Participants behind a NAT are second class unless they implement port forwarding, but participants with unstable IPs are not second class. + +## Game Networking sockets + +[Game Networking Sockets](https://github.com/ValveSoftware/GameNetworkingSockets) + +A reliable udp library with congestion control which has vastly more development work done on it than any other reliable udp networking library, but which is largely used to work with Steam gaming, and Steam's closed source code. Has no end of hooks to closed source built into it, but works fine without those hooks. + +Written in C++. Architecture overly specific and married to Steam. Would +have to be married to Tokio to have massive concurrency. But you don't +need to support hundreds of clients right away. + +Well, perhaps I do, because in the face of DDOS attack, you need to keep +a lot of long lived inactive connections around for a long time, any of +which could receive a packet at any time. I need to look at the +GameNetworkingSockets code and see how it listens on lots and lots of +sockets. If it uses [overlapped IO], then it is golden. Get it up first, and it put inside a service later. + +[Overlapped IO]:client_server.html#the-select-problem +{target="_blank"} + +The nearest equivalent Rust application gave up on congestion control, having programmed themselves into a blind alley. + +## Tokio + +Tokio is a Rust framework for writing highly efficient highly scalable +services. Writing networking for a service with large numbers of clients is +very different between Windows and Linux, and I expect Tokio to take care +of the differences. + +There really is not any good C or C++ environment for writing services +except Wt, which is completely specialized for the case of writing a web +service whose client is the browser, and which runs only on Linux. + +## wxWidgets +wxWidgets has basic networking capability built in and integrated with its +event loop, but it is a bit basic, and is designed for a gui app, not for a +server – though probably more than adequate for initial release. It only +supports http, but not https and websockets. +[LibSourcery](https://sourcey.com/libsourcey) is a far more powerful +networking library, which supports https and websockets, and is designed to +interoperate with nginx and node.js. But integrating it with wxWidgets is +likely to be nontrivial. + +WxWidgets sample code for sockets is in %WXWIN%/samples/sockets. There is a +[recently updated version on github]. Their example code supports TCP and +UDP. But some people argue that the sampling is insufficiently responsive - +you really need a second thread that damned well sits on the socket, rather +than polling it. And that second thread cannot use wxSockets. + +[recently updated version on github]:https://github.com/wxWidgets/wxWidgets/tree/master/samples/sockets + +Programming sockets and networking in C is a mess. The [much praised guide +to sockets](https://beej.us/guide/bgnet/html/single/bgnet.html) goes on for +pages and pages describing a “simple” example client server. Trouble is that +C, and old type Cish C++ exposes all the dangly bits. The [QT client server +example](https://stackoverflow.com/questions/5773390/c-network-programming), +on the other hand, is elegant, short, and self explanatory. + +The code project has [example code written in C++](https://www.codeproject.com/Articles/13071/Programming-Windows-TCP-Sockets-in-C-for-the-Begin), but it is still mighty intimidating compared to the QT client server example. I have yet to look at the wxWidgets client server examples – but looking for wxWidgets networking code has me worried that it is a casual afterthought, not adequately supported or adequately used. + +ZeroMQ is Linux, C, and Cish C++. + +Boost Asio is highly praised, but I tried it, and concluded its architecture +is broken, trying to make simplicity and elegance where it cannot be made, +resulting in leaky abstractions which leak incomprehensible complexity the +moment you stray off the beaten path – I feel they have lost control of their + design, and are just throwing crap at it trying to make something that +cannot work, work. I similarly found the Boost time libraries failed, leaking + complexity that they tried to hide, with the hiding merely adding complexity. + +[cpp-httplib](https://github.com/yhirose/cpp-httplib) is wonderful in its +elegance, simplicity, and ease of integration. You just include a single +header. Unfortunately, it is strictly http/https, and we need something that +can deal with the inherently messy lower levels. + +[Poco](http://pocoproject.org/) does everything, and is C++, but hey, let us first see how far we can get with wxWidgets. + +Further, the main reason for doing https integration with the existing +browser web ecosystem, whose security is fundamentally broken, due the +state’s capacity to seize names, and the capacity of lots of entities to +intercept ssl. It might well be easier to fork opera or embed chromium. I +notice that Chromium has features supporting payment built into it, a bunch +of “PaymentMethod\*\*\*\*\*Event” + +The best open source browser, and best privacy browser, is Opera, in that it comes from an entity less evil than Google. + +[Opera](https://bit.ly/2UpSTFy) needs to be configured with [a bunch of privacy add ons](https://gab.com/PatriotKracker80/posts/c3kvL3pBbE54NEFaRGVhK1ZiWCsxZz09) [HTTPS Everywhere Add-on](https://bit.ly/2ODbPeE), +[uBlock](https://bit.ly/2nUJLqd), [DisconnectMe](https://bit.ly/2HXEEks), [Privacy-Badger](https://bit.ly/2K5d7R1), [AdBlock Plus](https://bit.ly/2U81ddo), [AdBlock for YouTube](https://bit.ly/2YBzqRh), two tracker blockers, and three ad blockers. + +It would be great if we could make our software another addon, possibly chatting by websocket to the wallet. + +The way it would work be to add another protocol to the browser: +ro://name1.name2.name3/directory/directory/endpoint. When you connect to such + an endpoint, your wallet, possibly a wallet with no global name, connects to + the named wallet, and gets IP, a port, a virtual server name, a cookie +unique for your wallet, and the hash of the valid ssl certificate for that +name, and then the browser makes a connection to the that server, ignoring +the CA system and the DNS system. The name could be a DNS name and the +certificate a CA certificate, in which case the connection looks to the +server like any other, except for the cookie which enables it to send +messages, typically a payment request, to the wallet. + +# Safe maths + +[Safeint]:https://github.com/dcleblanc/SafeInt +{target="_blank"} + +We could implement transaction outputs and inputs as a fixed amount of +fungible tokens, limited to $2^{64}-1$ tokens, using [Safeint] That will be +future proof for a long time, but not forever. + +Indeed, anything that does not use Zksnarks is not future proof for the +indefinite future. + +Or we could implement decimal floating point with unlimited exponents +and mantissa implemented on top of [MPIR] + +Or we could go ahead with the canonical representation being unlimited +decimal exponent and unlimited mantissa, but the wallet initially only +generates, and only can handle, transactions that can be represented by[Safeint], and always converts the mantissa plus decimal exponent to and +from a safeint. + +if we rely on safeint, and our smallest unit is the microrho, that is room for +eighteen trillion rho. We can start actually using the unlimited precision of +the exponent and the mantissa in times to come - not urgent, merely +architect it into the canonical format. + +From the point of view of the end user, this will merely be an upgrade that +allows nanorho, picorho, femptorho, attorho, zeptorho, yoctorho, and allows a decimal point in yoctorho quantities. And then we go to a new unit, the jim, with one thousand yottajim equals one yoctorho, a billion yoctojim equals one attorho, a trillion exajim equals one attorho. + +To go all the way around to two byte exponents, for testing purposes, will +need some additional new units after the jim. (And we should impose a +minimum unit size of $10^{-195}$ rho or $10{-6} rho, thereby ensuring +that transaction size is bounded while allowing compatibility for future expansion.) + +Except in test and development code, any attempt to form a transaction +involving quantities with exponents less than $1000^{-2}$ will cause a +gracefully handled exception, and in all code any attempt to display +or perform calculations on transaction inputs and outputs for which no +display units exist will cause an ungracefully handled exception. + +In the first release configuration parameters, the lowest allowed exponent +will be $1000^{-2}$, corresponding to microrho, and the highest allowed +exponent $1000^4$, corresponding to terarho, and machines will be +programmed to vote "incapable" and "no" on any proposal to change those +parameters. However they will correctly handle transactions beyond those +limits provided that when quantities are expressed in the smallest unit of +any of the inputs and outputs, the sum of all the inputs and of all the +outputs remains below $2^{64}$. To ensure that all releases are future +compatible, the blockchain should have some exajim transactions, and +unspent transaction outputs but the peers should refuse to form any more +of them. The documentation will say that arbitrarily small and large new +transaction outputs used to be allowed, but are currently not allowed, to +reduce the user interface attack surface that needs to be security checked +and to limit blockchain bloat, and since there is unlikely to be demand for +this, this will probably not be fixed for a very long time. + +Or perhaps it would be less work to support humungous transactions from +the beginning, subject to some mighty large arbitrary limit to prevent +denial of service attack, and eventually implementing native integer +handling of normal sized transactions as an optimization, for transactions where all quantities fit within machine sized words, and rescaled intermediate outputs will be less than $64 - \lceil log_2($number of inputs and outputs$) \rceil$ bits. + +Which leads me to digress how we are going to handle protocol updates: + +## handling protocol updates + +1. Distribute software capable of handling the update. +1. A proposed protocol update transaction is placed on the blockchain. +1. Peers indicate capability to handle the protocol update. Or ignore it, + or indicate that they cannot. If a significant number of peers + indicate capability, peers that lack capability push their owners for + an update. +1. A proposal to start emitting data that can only handled by more + recent peers is placed on the blockchain. +1. If a significant number of peers vote yes, older peers push more + vigorously for an update. +1. If a substantial supermajority votes yes by a date specified in the + proposal, then they start emitting data in the new format on a date + shortly afterwards. If no supermajority by the due date, the + proposal is dead. + +# [Zlib compression libraries.](./libraries/zlib.html) + +Built it, easy to use, easy to build, easy to link to. Useful for large amounts of text, provides, but does not use, CRC32 + +[Cap\'n Proto](./libraries/capnproto.html) + +[Crypto libraries](./libraries/crypto_library.html) + +[Memory Safety](./libraries/memory_safety.html). + +[C++ Automatic Memory Management](./libraries/cpp_automatic_memory_management.html) + +[C++ Multithreading](./libraries/cpp_multithreading.html) + +[Catch testing library](https://github.com/catchorg/Catch2) +[Boost](https://github.com/boostorg/boost) + +------------------------------------------------------------------------ + +## Boost + +My experience with Boost is that it is no damned good: They have an over +elaborate pile of stuff on top of the underlying abstractions, which pile has high runtime cost, and specializes the underlying stuff in ways that only +work with boost example programs and are not easily generalized to do what +one actually wishes done. + +Their abstractions leak. + +[Boost high precision arithmetic `gmp_int`]:https://gmplib.org/ + +[Boost high precision arithmetic `gmp_int`] A messy pile built on top of + GMP. Its primary benefit is that it makes `gmp` look like `mpir` Easier to use [MPIR] directly. + + The major benefit of boost `gmp` is that it runs on some machines and + operating systems that `mpir` does not, and is for the most part source code +compatible with `mpir`. + + A major difference is that boost `gmp` uses long integers, which are on sixty + four bit windows `int32_t`, where `mpir` uses `mpir_ui` and `mpir_si`, which are + on sixty four bit windows `uint64_t` and `int64_t`. This is apt to induce no + end of major porting issues between operating systems. + + Boost `gmp` code running on windows is apt to produce radically different + results to the same boost `gmp` code running on linux. Long `int` is just not + portable, and should never be used. This kind of issue is absolutely typical + of boost. + + In addition to the portability issue, it is also a typical example of boost + abstractions denying you access to the full capability of the thing being + abstracted away. It is silly to have a thirty two bit interface between sixty + four bit hardware and unlimited arithmetic precision software. + +------------------------------------------------------------------------ + +## Database + +The blockchain is a massively distributed database built on top of a pile of +single machine, single disk, databases communicating over the network. If you + want a single machine, single disk, database, go with SQLite, which in WAL +mode implements synch interaction on top of hidden asynch. + +[SQLite](https://www.Sqlite.org/src/doc/trunk/README.md) have their own way of doing things, that does not play nice with Github. + +The efficient and simple way to handle interaction with the network is via +callbacks rather than multithreading, but you usually need to handle +databases, and under the hood, all databases are multithreaded and blocking. +If they implement callbacks, it is usually on top of a multithreaded layer, +and the abstraction is apt to leak, apt to result in unexpected blocking on a + supposedly asynchronous callback. + +SQLite recommends at most one thread that writes to the database, and +preferably only one thread that interacts with the database. + +## The Invisible Internet Project (I2P) + +[Comes](https://geti2p.net/en/) with an I2P webserver, and the full api for streaming stuff. These +appear as local ports on your system. They are not tcp ports, but higher +level protocols, *and* UDP. (Sort of UDP - obviously you have to create a +durable tunnel, and one end is the server, the other the client.) + +Inconveniently, written in java. + +## Internet Protocol + +[QUIC] UDP with flow control and reliability. Intimately married to http/2, +https/2, and google chrome. Cannot call as library, have to analyze code, +extract their ideas, and rewrite. And, looking at their code, I think they +have written their way into a blind alley. + +But QUIC is http/2, and there is a gigantic ecosystem supporting http/2. + +We really have no alternative but to somehow interface to that ecosystem. + +[QUIC]: https://github.com/private-octopus/picoquic + +[QUIC] is UDP with flow control, reliability, and SSL/TLS encryption, but no +DDoS resistance, and total insecurity against CA attack.) + +## Boost Asynch + +Boost implements event oriented multithreading in IO service, but don’t like +it because it fails to interface with Microsoft’s implementation of asynch +internet protocol, WSAAsync, and WSAEvent. Also because brittle, +incomprehensible, and their example programs do not easily generalize to +anything other than that particular example. + +To the extent that you need to interact with a database, you need to process +connections from clients in many concurrent threads. Connection handlers are +run in thread, that called `io_service::run()`. + +You can create a pool of threads processing connection handlers (and waiting +for finalizing database connection), by running `io_service::run()` from +multiple threads. See Boost.Asio docs. + +## Asynch Database access + +MySQL 5.7 supports [X Plugin / X Protocol, which allows asynchronous query execution and NoSQL But X devapi was created to support node.js and stuff. The basic idea is that you send text messages to mysql on a certain port, and asynchronously get text messages back, in google protobuffs, in php, JavaScript, or sql. No one has bothered to create a C++ wrapper for this, it being primarily designed for php or node.js](https://dev.mysql.com/doc/refman/5.7/en/document-store-setting-up.html) + +SQLite nominally has synchronous access, and the use of one read/write +thread, many read threads is recommended. But under the hood, if you enable +WAL mode, access is asynchronous. The nominal synchrony sometimes leaks into +the underlying asynchrony. + +By default, each `INSERT` is its own transaction, and transactions are +excruciatingly slow. Wal normal mode fixes this. All writes are writes to the + writeahead file, which gets cleaned up later. + +The authors of SQLite recommend against multithreading writes, but we +do not want the network waiting on the disk, nor the disk waiting on the +network, therefore, one thread with asynch for the network, one purely +synchronous thread for the SQLite database, and a few number crunching +threads for encryption, decryption, and hashing. This implies shared +nothing message passing between threads. + +------------------------------------------------------------------------ + +[Facebook Folly library]provides many tools, with such documentation as +exists amounting to “read the f\*\*\*\*\*g header files”. They are reputed +to have the highest efficiency queuing for interthread communication, and it +is plausible that they do, because facebook views efficiency as critical. +Their [queuing header file] +(https://github.com/facebook/folly/blob/master/folly/MPMCQueue.h) gives us +`MPMCQueue`. + +[Facebook Folly library]:https://github.com/facebook/folly/blob/master/folly/ + +On the other hand, boost gives us a lockless interthread queue, which should +be very efficient. Assuming each thread is an event handler, rather than +pseudo synchronous, we queue up events in the boost queue, and handle all +unhandled exceptions from the event handler before getting the next item from + the queue. We keep enough threads going that we do not mind threads blocking + sometimes. The queue owns objects not currently being handled by a +particular thread. Objects are allocated in a particular thread, and freed in + a particular thread, which process very likely blocks briefly. Graphic +events are passed to the master thread by the wxWindows event code, but we +use our own mutltithreaded event code to handle everything else. Posting an +event to the gui code will block briefly. + +I was looking at boost’s queues and lockless mechanisms from the point of +view of implementing my own thread pool, but this is kind of stupid, since +boost already has a thread pool mechanism written to handle the asynch IO +problem. Thread pools are likely overkill. Node.js does not need them, +because its single thread does memory to memory operations. + +Boost provides us with an [`io_service` and `boost::thread` group], used to +give effect to asynchronous IO with a thread pool. `io_service` was specifically written to perform io, but can be used for any +thread pool activity whatsoever. You can “post” tasks to the io_service, +which will get executed by one of the threads in the pool. Each such task has + to be a functor. + +[`io_service` and `boost::thread` group]:http://thisthread.blogspot.com/2011/04/multithreading-with-asio.html + +Since supposedly nonblocking operations always leak and block, all we can do +is try to have blocking minimal. For example nonblocking database operations +always block. Thus our threadpool needs to be many times larger than our set +of hardware threads, because we will always wind up doing blocking operations. + +The C++11 multithreading model assumes you want to do some task in parallel, +for example you are multiplying two enormous matrices, so you spawn a bunch +of threads, then you wait for them all to complete using `join`, or all to +deliver their payload using futures and promises. This does not seem all that + useful, since the major practical issue is that you want your system to +continue to be responsive while it is waiting for some external hardware to +reply. When you are dealing with external events, rather than grinding a +matrix in parallel, event oriented architecture, rather than futures, +promises, and joins is what you need. + +Futures, promises, and joins are useful in the rather artificial case that +responding to an remote procedure call requires you to make two or more +remote procedure calls, and wait for them to complete, so that you then have +the data to respond to a remote procedure call. + +Futures, promises, and joins are useful on a server that launches one thread +per client, which is often a sensible way to do things, but does not fit that + well to the request response pattern, where you don’t have a great deal of +client state hanging around, and you may well have ten thousand clients If +you can be pretty sure you are only going to have a reasonably small number +of clients at any one time, or and significant interaction between clients, +one thread per client may well make a lot of sense. + +I was planning to use boost asynch, but upon reading the boost user threads, +sounds fragile, a great pile of complicated unintelligible code that does +only one thing, and if you attempt to do something slightly different, +everything falls apart, and you have to understand a lot of arcane details, +and rewrite them. + +[Nanomsg](http://nanomsg.org/)is a socket library, that provides a layer on +top of everything that makes everything look like sockets, and provides +sockets specialized to various communication patterns, avoiding the roll your + own problem. In the zeroMQ thread, people complained that [a simple hello +world TCP-IP program tended to be disturbingly large and complex] +Looks to me that [Nanomsg] wraps a lot of that complexity. + +[a simple hello world TCP-IP program tended to be disturbingly large and complex]:http://250bpm.com/blog + +# Sockets + +A simple hello world TCP-IP program tends to be disturbingly large and +complex, and windows TCP-IP is significantly different from posix TCP-IP. + +Waiting on network events is deadly, because they can take arbitrarily large +time, but multithreading always bites. People who succeed tend to go with +single thread asynch, similar to, [or part of, the window event handling +loop]. + +[or part of, the window event handling loop]:https://www.codeproject.com/Articles/13071/Programming-Windows-TCP-Sockets-in-C-for-the-Begin + +Asynch code should take the form of calling a routine that returns +immediately, but passing it a lambda callback, which gets executed in the +most recently used thread. + +Interthread communication bites – you don’t want several threads accessing +one object, as synch will slow you down, so if you multithread, better to +have a specialist thread for any one object, with lockless queues passing +data between threads. One thread for all writes to SQLite, one thread for +waiting on select. + +Boost Asynch supposedly makes sockets all look alike, but I am frightened of +their work guard stuff – looks to me fragile and incomprehensible. Looks to +me that no one understands boost asynch work guard, not even the man who +wrote it. And they should not be using boost bind, which has been obsolete +since lambdas have been available, indicating bitrot. + +Because work guard is incomprehensible and subject to change, will just keep +the boost io object busy with a polling timer. + +And I am having trouble finding boost asynch documented as a sockets library. + Maybe I am just looking in the wrong place. + +[A nice clean tutorial depicting strictly synchronous tcp.](https://www.binarytides.com/winsock-socket-programming-tutorial/) + +[Libpcap and Win10PCap](https://en.wikipedia.org/wiki/Pcap#Wrapper_libraries_for_libpcap) provide very low level, OS independent, access to packets, OS independent because they are below the OS, rather than above it. [Example code for visual studio.](https://www.csie.nuk.edu.tw/~wuch/course/csc521/lab/ex1-winpcap/) + +[Simple sequential procedural socket programming for windows sockets.](https://www.binarytides.com/winsock-socket-programming-tutorial/) + +If I program from the base upwards, the bottom most level would be a single +thread sitting on a select statement. Whenever the select fired, would +execute a corresponding functor transfering data between userspace and system + space. + +One thread, and only one thread, responsible for timer events and +transferring network data between userspace and systemspace. + +If further work required in userspace that could take significant time (disk +operations, database operations, cryptographic operations) that functor under + that thread would stuff another functor into a waitless stack, and a bunch +of threads would be waiting for that waitless stack to be signaled, and one +of those other threads would execute that functor. + +The reason we have a single userpace thread handling the select and transfers + between userpace and systemspace is that that is a very fast and very common + operation, and we don’t want to have unnecessary thread switches, wherein +one thread does something, then immediately afterwards another thread does +almost the same thing. All quickie tasks should be handled sequentially by +one thread that works a state machine of functors. + +The way to do asynch is to wrap sockets in classes that reflect the intended +use and function of the socket. Call each instance of such a class a +connection. Each connection has its own state machine state and its own +**message dispatcher, event handler, event pump, message pump**. + +A single thread calls select and poll, and drives all connection instances in + all transfers of data between userspace and systemspace. Connections also +have access to a thread pool for doing operations (such as file, database and + cryptography, that may involve waits. + +The hello world program for this system is to create a derived server class +that does a trivial transformation on input, and has a path in server name +space, and a client class that sends a trivial input, and displays the result. + +Microsoft WSAAsync\[Socketprocedure\] is a family of socket procedures +designed to operate with, and be driven by, the Window ui system, wherein +sockets are linked to windows, and driven by the windows message loop. Could +benefit considerably by being wrapped in connection classes. + +I am guessing that wxWidgets has a similar system for driving sockets, +wherein a wxSocket is plugged in to the wxWidget message loop. On windows, +wxWidget wraps WSASelect, which is the behavior we need. + +Microsoft has written the asynch sockets you need, and wxWidgets has wrapped +them in an OS independent fashion. + + WSAAsyncSelect + + WSAEventSelect + + select + +Using wxSockets commits us to having a single thread managing everything. To +get around the power limit inherent in that, have multiple peers under +multiple names accessing the same database, and have a temporary and +permanent redirect facility – so that if you access `peername,` your +connection, and possibly your link, get rewritten to `p2.peername` by peers +trying to balance load. + +Microsoft tells us: + +> receiving, applications use the WSARecv or WSARecvFrom functions to supply +buffers into which data is to be received. If one or more buffers are posted +prior to the time when data has been received by the network, that data could + be placed in the user’s buffers immediately as it arrives. Thus, it can +avoid the copy operation that would otherwise occur at the time the recv or +recvfrom function is invoked. + +Moral is, we should use the sockets that wrap WSA. + +# Tcl + +Tcl is a really great language, and I wish it would become the language of my new web, as JavaScript is the language of the existing web. + +But it has been semi abandoned for twenty years. + +It consists of a string (which is implemented under the hood as a copy on +write rope, with some substrings of the rope actually being run time typed +C++ types that can be serialized and deserialized to strings) and a name +table, one name table per interpreter, and at least one interpreter per +thread. The entries in the name table can be strings, C++ functions, or run +time typed C++ types, which may or may not be serializable or deserializable, + but conceptually, it is all one big string, and the name table is used to +find C and C++ functions which interpret the string following the command. +Execution consists of executing commands found in the string, which transform + it into a new string, which in turn gets transformed into a new string, +until it gets transformed into the final result. All code is metacode. If +elements of the string need to be deserialized to and from a C++ run time +type, (because the command does not expect that run time type) but cannot be, + because there is no deserialization for that run time type, you get a run +time error, but most of the time you get, under the hood, C++ code executing +C++ types – it is only conceptually a string being continually transformed +into another string. The default integer is infinite precision, because +integers are conceptually arbitrary length strings of numbers. + +To sandbox third party code, including third party gui code, just restrict +the nametable to have no dangerous commands, and to be unable to load c++ +modules that could provide dangerous commands. + +It is faster to bring up a UI in Tcl than in C. We get, for free, OS +independence. + +Tcl used to be the best level language for attaching C programs to, and for +testing C programs, or it would be if SWIG actually worked. The various C +components of Tcl provide an OS independent layer on top of both Linux and +Windows, and it has the best multithread and asynch system. + +It is also a metaprogramming language. Every Tcl program is a metaprogram – you always write code that writes code. + +The Gui is necessarily implemented as asynch, something like the JavaScript +dom in html, but with explicit calls to the event/idle loop. Multithreading +is implemented as multiple interpreters, at least one interpreter per thread, + sending messages to each other. + +# Time + +After spending far too much time on this issue, which is has sucked in far +too many engineers and far too much thought, and generated far too many +libraries, I found the solution was c++11 Chrono: For short durations, we +use the steady time in milliseconds, where each machine has its own +epoch, and no two machines have exactly the same milliseconds. For +longer durations, we use the system time in seconds, where all machines +are expected to be within a couple of seconds of each other. For the human +readable system time in seconds to be displayed on a particular machine, +we use the ISO format 2012‑01‑14_15:39:34+10:00 (timezone with 10 +hour offset equivalent to Greenwich time 2012‑01‑14_05:39:34+00:00) + +[For long durations, we use signed system time in seconds, for short durations unsigned steady time in milliseconds.](./libraries/rotime.cpp) + +Windows and Unix both use time in seconds, but accessed and manipulated in +incompatible ways. + +Boost has numerous different and not altogether compatible time libraries, +all of them overly clever and all of them overly complicated. + +wxWidgets has OS independent time based on milliseconds past the epoch, which + however fails to compress under Cap\'n Proto. + +I was favourably impressed by the approach to time taken in tcp packets, +that the time had to be approximately linear, and in milliseconds or larger, +but they were entirely relaxed about the two ends of a tcp connection +using different clocks with different, and variable, speeds. + +It turns out you can go a mighty long way without a global time, and to the +extent that you do need a global time, should be equivalent to that used in +email, which magically hides the leap seconds issue. + +# UTF‑8 strings + +Are supported by the wxWidgets wxString, which provide support to and +from wide character variants and locale variants. (We don't want locale +variants, they are obsolete. The whole world is switching to UTF, but +our software and operating environments lag) + +`wString::ToUTF8()` and `wString::FromUTF8()` do what you would expect. + +On visual studio, need to set your source files to have bom, so that Visual +Studio knows that they are UTF‑8, need to set the compiler environment in +Visual Studio to UTF‑8 with `/Zc:__cplusplus /utf-8 %(AdditionalOptions)` + +And you need to set the run time environment of the program to UTF‑8 +with a manifest. + +You will need to place all UTF‑8 string literals and string constants in a +resource file, which you will use for translated versions. + +If you fail to set the compilation and run time environment to UTF‑8 then +for extra confusion, your debugger and compiler will *look* as if they are +handling UTF‑8 characters correctly as single byte characters, while at +least wxString alerts you that something bad is happening by run time +translating to the null string. + +Automatic string conversion in wxWidgets is *not* UTF‑8, and if you have +any unusual symbols in your string, you get a run time error and the empty +string. So wxString automagic conversions will rape you in the ass at +runtime, and for double the confusion, your correctly translated UTF‑8 +strings will look like errors. Hence the need to make sure that the whole +environment from source code to run time execution is consistently UTF‑8, +which has to be separately ensured in three separate place. + +When wxWidgets is compiled using `#define wxUSE_UNICODE_UTF8 1`, +it provides UTF‑8 iterators and caches a character index, so that accessing +a character by index near a recently used character is fast. The usual +iterators `wx.begin()`, `wx.end()`, const and reverse iterators are available. +I assume something bad happens if you advance a reverse iterator after +writing to it. + +wxWidgets compiled with `#define wxUSE_UNICODE_UTF8 1` is the +way of the future, but not the way of the present. Still a work in progress +Does not build under Windows. Windows now provide UTF8 entries to all +its system functions, which should make it easy. + +wxWidgets provides `wxRegEx` which, because wxWidgets provides index +by entity, should just work. Eventually. Maybe the next release. + +# [UTF8-CPP](http://utfcpp.sourceforge.net/ "UTF-8 with C++ in a Portable Way") + +A powerful library for handling UTF‑8. This somewhat duplicates the +facilities provided by wxWidgets with `wxUSE_UNICODE_UTF8==1` + +For most purposes, wxString should suffice, when it actually works with +UTF8. Which it does not yet on windows. We shall see. wxWidgets +recommends not using wxString except to communicate with wxWidgets, +and not using it as general UTF‑8 system. Which is certainly the current +state of play with wxWidgets. + + For regex to work correctly, probably need to do it on wxString's native + UTF‑16 (windows) or UTF‑32 (unix), but it supposedly works on `UTF8`, + assuming you can successfully compile it, which you cannot. + +# Cap\'n Proto + +[Designed for a download from github and run cmake install.](https://capnproto.org/install.html) As all software should be. + +But for mere serialization to of data to a form invariant between machine +architectures and different compilers and different compilers on the same +machine, overkill for our purposes. Too much capability. + +# Awesome C++ + +[Awesome C++] A curated list of awesome C/C++ frameworks, libraries, resources, and shiny things + +[Awesome C++]:https://cpp.libhunt.com +"A curated list of awesome C/C++ frameworks, libraries, resources, and shiny things" +{target="_blank"} + +I encountered this when looking at the Wt C++ Web framework, which seems to be mighty cool except I don't think I have any use for a web framework. But [Awesome C++] has a very pile of things that I might use. + +Wt has the interesting design principle that every open web page maps to a +windows class, every widget on the web page, maps to a windows class, +every row in the sql table maps to a windows class. Cool design. + +# Opaque password protocol + +[Opaque] is PAKE done right. + +[Opaque]:https://blog.cryptographyengineering.com/2018/10/19/lets-talk-about-pake/ +"Let’s talk about PAKE" {target="_blank"} + +Server stores a per user salt, the users public key, and the user's secret key +encrypted with a secret that only the user ever learns. + +Secret is generated by the user from the salt and his password by +interaction with the server without the the user learning the salt, nor the hash of the salt, nor the server the password or the hash of the password. +User then strengthens the secret generated from salt and password +applying a large work factor to it, and decrypts the private key with it. +User and server then proceed with standard public key cryptography. + +If the server is evil, or the bad guys seize the server, everything is still +encrypted and they have to run, not a hundred million trial passwords +against all users, but a hundred million passwords against *each* user. And +user can make the process of trying a password far more costly and slow than +just generating a hash. Opaque zero knowledge is designed to be as +unfriendly as possible to big organizations harvesting data on an industrial +scale. The essential design principle of this password protocol is that +breaking a hundred million passwords by password guessing should be a +hundred million times as costly as breaking one password by password +guessing. The protocol is primarily designed to obstruct the NSA's mass +harvesting. + +It has the enormous advantage that if you have one strong password which +you use for many accounts, one evil server cannot easily attack your +accounts on other servers. To do that, it has to try every password - which +runs into your password strengthening. diff --git a/docs/libraries/app.cpp b/docs/libraries/app.cpp new file mode 100644 index 0000000..0f39562 --- /dev/null +++ b/docs/libraries/app.cpp @@ -0,0 +1,21 @@ +#include "stdafx.h" + + +app::app() +{ +} + +app::~app() +{ +} + +bool app::OnInit() +{ + wxFrame* mainFrame = new wxFrame(nullptr, wxID_ANY, L"Hello World C"); + //wxFrame* mainFrame = new wxFrame(nullptr, wxID_ANY, wcp); + mainFrame->Show(true); + return true; +} + +wxIMPLEMENT_APP(app); + diff --git a/docs/libraries/app.h b/docs/libraries/app.h new file mode 100644 index 0000000..4073568 --- /dev/null +++ b/docs/libraries/app.h @@ -0,0 +1,10 @@ +#pragma once + +class app : + public wxApp +{ +public: + app(); + virtual ~app(); + virtual bool OnInit() override; +}; \ No newline at end of file diff --git a/docs/libraries/building_and_using_libraries.md b/docs/libraries/building_and_using_libraries.md new file mode 100644 index 0000000..a69374b --- /dev/null +++ b/docs/libraries/building_and_using_libraries.md @@ -0,0 +1,616 @@ +--- +title: Building the project and its libraries in Visual Studio +--- + +# General instructions + +At present the project requires the environment to be set up by hand, with a +lot of needed libraries separately configured and separately built. + +We need to eventually make it so that it is one git project with submodules +which can be build with one autotools command with submodules, and one visual +studio command with subprojects, so that + +```bash +git clone --recursive git://example.com/foo/rhocoin.git +cd rhocoin +devenv rhocoin.sln /build +``` + +will build all the required libraries. + +And similarly we want autotools to build all the submodules + +```bash + git clone --recursive git://example.com/foo/rhocoin.git + cd rhocoin + ./configure; make && make install +``` + +so that the top level configure and make does the `./configure; make && make install` in each submodule. + +which might also create a deb file that could be used in + +```bash +apt-get -qy install ./rhocoin.deb +``` + +But we are a long way from being there yet. At present the build environment +is painfully hand made, and has to be painfully remade every time some updates +a library on which it relies. + +To build in Visual Studio under Microsoft windows + +- Set the environment variable `SODIUM` to point to the Libsodium directory containing the directory `src/libsodium/include` and build the static linking, not the DLL, library following the libsodium instructions. +- Set the environment variable `WXWIN` to point to the wxWidgets directory containing the directory `include/wx` and build the static linking library from the ide using the provided project files + +If you are building this project using the Visual Studio ide, you should use the ide to build the libraries, and if you are building this project using the makefiles, you should use the provided make files to build the libraries. In theory this should not matter, but all too often it does matter. + +When building libsodium and wxWidgets in Visual Studio, have to retarget the +solution to use the current Microsoft libraries, retarget to use x64, and +change the code generation default in every project versions from +Multithreaded Dll to Multithreaded + +Sqlite is not incorporated as an already built library, but as source code., +as the sqlite3 amalgamation file, one very big C file. + +# Instructions for wxWidgets + +## Setting wxWidgets project in Visual Studio + +First set up your environment variables as described in [Directory Structure +Microsoft Windows](../set_up_build_environments.html#dirstruct). + +Run the wxWidgets windows setup, wxMSW-X.X.X-Setup.exe. The project will +build with wxMSW-3.1.2, and will not build with earlier versions. Or just +unzip the file into the target directory, as the install does not in fact do +any useful configuration. + +Build instructions are in `%WXWIN%\docs\msw\install.txt` The setup program +for wxWidgets should set the environment variable WXWIN correctly, but does +not do so. Manually set the WXWIN environment variable + +When building in Visual Studio, have to retarget the solution to use the +current libraries, retarget to use x64, and change the code generation +default in every project versions from Multithreaded Dll to Multithreaded +(select all projects except the custom build project, then go to +properties/C++/code generation/Runtime). The DLL change has to be done +separately for release and debug builds, since Debug uses the MTd libraries, +and Release uses the MT libraries. + +A Visual Studio project needs the library build by the wxWidget Visual Studio +project, an nmake project needs the library built by the wxWidget makefile.vs + +If you build roWallet under nmake, using Visual Studio tools, you should +build your wxWidgets libraries using the using nmake – fmakefile.vc, not the +wxWidget Visual Studio project files. + +If you build roWallet using the Visual Studio project, you should build your +wxWidgets libraries using Visual Studio and the wxWidgets Visual Studio +project files. + +UDP sockets seem seriously under supported and undocumented in +wxWidgets, though theoretically present. + +The [discussion](http://www.wxwidgets.org/search/?q=datagrams "wx Widgets Datagrams"), +however, makes up for the paucity of documentation. + +wxWidgets somehow neglects to mention that you need to use the +different and entirely incompatible UDP specific system calls, `recvfrom()` +and `sendto()` and instead of `read()` and `write()` + +If using C sockets, need to pull in completely different header files on +Unix than on Windows, including Winsock2 rather than Winsock on +windows, but these completely different header files pull in almost the +same routines that work in almost the same way. + +```C +#ifdef _WIN64 + #include +#else + #include + #include + #include + #include + #include +#endif +``` + +If using wxWidgets, need to build with + +```c++ +#define WXWIN_COMPATIBILITY_3_0 0 +#define wxUSE_COMPILER_TLS 2 +#define wxUSE_STD_CONTAINERS 1 +#define wxUSE_IPV6 1 +``` + +in `%WXWIN%\include\wx\msw\setup.h` + +And for gnu builds, `./configure && make && make install`, have to +change them separately and again in configure. What `configure` actually +does is difficult to understand and predict, so I have put asserts in the code +to detect and complain about unsatisfactory settings. + +```C++ +#ifdef _WIN64 +constexpr bool b_WINDOWS = true; +#else +constexpr bool b_WINDOWS = false; +#endif +static_assert( __cplusplus >= 201703l, "Out of date C syntax"); +static_assert(wxUSE_IPV6 == 1, "IP6 unavailable in wxWidgets"); +static_assert(WXWIN_COMPATIBILITY_3_0 == 0, "wxWidgets api out of date"); +static_assert(wxUSE_COMPILER_TLS == (b_WINDOWS ? 2 : 1), "out of date workarounds in wxWidgets for windows bugs"); +static_assert(wxUSE_STD_CONTAINERS_COMPATIBLY == 1, "interoperability between stl and wxWidgets broken"); +static_assert(wxUSE_STD_CONTAINERS == 1, "wxWidgets api out of date"); +``` + +The two wxWidgets libraries that you build can co-exist because stored in +different directories of `%WXWIN%`. Unfortunately the visual studio build +projects default to the multithreaded dll, which breaks every other library, +because multithreaded, rather than multithreaded dll, is the usual windows +default used by statically linked libraries. so each subproject in the +wxWidgets Visual Studio project has to be changed to link to multithreaded, +rather than multithreaded DLL. This is a bug, or at least an inconvenient +deviation from usual practice, in the current release of wxWidgets. + +If built by a visual studio project, the wxWidgets build constructs a header +file in a build location for visual studio projects to use, if built by nmake +under visual studio tools, the build constructs a header file in another +build location for nmake projects to use. + +```c++ +#ifdef _DEBUG +#pragma comment(lib, "wxbase31ud.lib") +#else +#pragma comment(lib, "wxbase31u.lib") +#endif +``` + +You are going to have to setup the environment variables %GSL%, %SODIUM% and +%WXWIN% on your windows machines for my visual studio project files to work +and going to have to build libsodium and wxWidgets in Visual Studio. + +## Moving a sample from the samples directory + +To an unrelated directory tree. + +Create an empty desktop project using Visual Studio Wizard. Change the build +type displayed at the top of Visual Studio from Debug X86 to Debug X64 + +In Solution Explorer (a palette-window of the VisualStudio-mainwindow), click +on the project name, then click on the properties icon, the little wrench, +and then in the left pane of the Property Pages dialog box, expand +Configuration Properties and select VC++ Directories. Additional include- or +lib-paths are specifyable there. + +Set General/C++ Language Standard to ISO(S++17) Standard (std::c++17) + +Add to the include paths (`properties/configuration properties/VC++ +Directories/Include Directories` + +``` +$(GSL)\include +$(WXWIN)\include\msvc +$(WXWIN)\include + $(SODIUM)\src\libsodium\include + ``` + +Add to the lib path (`properties/configuration properties/VC++ +Directories/Library Directories`) the location of the wxWidgets libraries + +``` +$(WXWIN)\lib\vc_x64_lib\ +$(SODIUM)\Build\Debug\x64 +``` + +Set Linker/System to Windows (/SUBSYSTEM:WINDOWS). This is always set to +CONSOLE, for no sane reason, even if you tell it to create an empty windows +project. + +Put unit test command line arguments (-dt) in Configuration/Debugging/Command +Arguments. + +Add the header, source, and resource files. The C++ option will then become +available in properties. Be mindful that if you edit a wxWidgets \*.rc file +in Visual Studio, Visual Studio destroys it. + +Set C/++/Code Generation/Runtime Library to Multi-threaded Debug(/MTd) in +place of multi threaded DLL. When you switch back and forth between release +and debug, this is apt to be set incorrectly, for reasons that are hard to +keep track of. + +Set C++/Preprocessor to \_DEBUG, \_WINDOWS, in place of \_DEBUG, \_CONSOLE +for the Debug version, and \_NDEBUG, \_WINDOWS for the release version. If +you compile a release version without defining \_NDEBUG, a flood of mystery +linker error messages ensue, caused by the fact that I use \_NDEBUG to select +library version, which is an improvement on the previous Visual Studio +behavior, where Visual Studio cheerfully generated an executable that +mysteriously just died at runtime because of mismatched libraries. + +These instructions lifted wholesale from:\ +[Creating wxWidgets Programs with Visual Studio 2017](https://usingcpp.wordpress.com/2018/02/15/creating-wxwidgets-programs-with-visual-studio-2017-part-1/)\ +[Hello World Example](https://docs.wxwidgets.org/stable/overview_helloworld.html) + +When you add the sqlite3.c amalgamation, make sure to mark it as not using +precompiled headers before the first attempt to compile it, otherwise it +breaks horribly, (C++ format precompiled headers being incompatible with C +precompiled headers) and when you belatedly turn off the precompiled headers, +some data that Visual Studio has generated hangs around, so that turning off +the precompiled headers fails fix the problem. + +Similarly, if you do a typo in the include paths. Remains stuck on the old +include paths even if you fix it. To do a real clean, close down visual +studio, delete the directories generated by visual studio, and *then* your +edits to the configuration will propagate properly + +wxWidgets for visual studio should be [installed by the windows installer] +(https://www.wxwidgets.org/downloads/), then follow the instructions in +`%WXWIN%\docs\msw\install.md`, after adjusting the visual studio projectg +files to build with multithreaded, rather than multithreaded DLL, for to +avoid DLL hell, we are not building with Microsoft DLLs. Set Project to X64. + +Visual Studio resource editor will destroy a resource file that is intended +for wxWidgets. WxWidgets resource files should only be edited in a text +editor. + +The native Icon format of wxWidgets is xpm, which does not have built in +support for multiple icon sizes, multiple color resolutions, and partial +transparency. The greenfish icon editor works to convert any icon format to +any other, (though it forgets to include the const qualifier) but, sad to +say, native wxWidgets icons, xpm, are lowest common denominator, therefore, +crap. And xpm format is also crap, a crap representation of crap data, making +windows native bitmap format look good by comparison. + +Most of the documentation for wxWidgets is the samples directory, which is +set up to build only in the samples directory. To move a project out of the +samples directory, and get it compiling in the environment you have set up +for your code in Visual Studio, copy the sample, and copy the resources its +needs into your target directory, then correct the sample, so that instead of +looking for resources relative to itself, in the wxWidgets directory, it +looks for its rc file and stuff in its new home directory. You will need the +resource files from the root of the samples, and possibly some files from +samples/images. + +Due to a persistent bug in visual studio, probably need to delete the +generated project studio files so that the new values will take effect. Do +not touch \*.rc files with Visual Studio resource editor, or it will break +them. + +## wxWidgets on Windows with Mingw + +I never succeeded in doing this, so there are probably no end of gotchas of +which I am unaware. + +1. [Run the wxWidgets windows setup, wxMSW-X.X.X-Setup.exe] +(https://github.com/wxWidgets/wxWidgets/releases/). The project will build +with wxMSW-3.1.2, and will not build with earlier versions  + +2. [Build wxWidgets] +(http://wiki.codeblocks.org/index.php/WxWindowsQuickRef).  See also the +instructions in wxWidgets-X.X.X/docs/msw/install.txt, or +docs/msw-X.X.X/install.md.  CodeBlocks on windows wants you to build in the +command window.  + +3. Install [Code::Blocks](http://www.codeblocks.org/downloads).  + +4. Do the Code::Blocks [hello world tutorial] +(http://wiki.codeblocks.org/index.php?title=WxSmith_tutorial%3a_Hello_world). + +But Code::Blocks looks suspiciously dead. + +## wxWidgets on Ubuntu + +Install build essentials, and gtk (gtk being the build essentials for Ubuntu +GUI) The native and home environment of wxWidgets is the Mate fork of Gnome, +and it is behavior in all other environments is derived from this behavior. + + sudo apt-get install libgtk-3-dev build-essential checkinstall + +These instructions copied from +[How to compile and install wxWidgets on Ubuntu/Debian/Linux Mint](https://www.binarytides.com/install-wxwidgets-Ubuntu/) more or +less wholesale.  + +Download [`Source code tar.gz`](https://github.com/wxWidgets/wxWidgets/releases/) from the latest releases.  + +Place the source in the directory `code/wxWidgets/`.  + +See the instructions in +`code/wxWidgets/docs/gtk/install.txt`  + +```bash +cd code/wxWidgets +mkdir gtk-build +cd gtk-build/ +./configure --disable-shared --enable-unicode +make +``` + +## Memory management in wxWidgets + +In C++17, if you are using exceptions, everything is owned by the stack, and +heap values are owned by stack variables which release the heap value in +their destructor. `std::unique_ptr` and `std::shared_ptr`, and your own +custom heap owning objects. + +In wxWidgets, ownership is complicated, because different aspects of +ownership are owned in different hierarchies. + +The destructors of modal windows get called in the normal fashion, but modeless windows are always referenced in derived code by non owning +pointers pointing into the heap. They are owned by their parent window +when their constructor calls its base constructor, and their derived +destructor somehow mysteriously gets called by wxWidgets after +wxWidgets finishes destroying all the windows that the higher level +window owns. + +Suppose you have a complicated window with a complicated hierarchy of +modeless windows in a complicated hierarchy of sizers. + +Suppose it wants to destroy one of its many children? + +Then it asks the child for the child's sizer, detaches it, calls destroy + on the child, which somehow eventually calls the highly derived + child destructor, and tells the sizers to do layout again. + +A sizer full of windows can be treated as if it was an owning window by + calling wxSizer::Clear on it with the argument to delete child windows. +But otherwise the owning window that wants to delete one particular +window by derived code has to do so by non owning pointers pointing into +the heap + +wxWidgets has a windows hierarchy, with windows owning windows, with memory +ownership by windows, and events being passed up the hierarchy. + +wxWidgets has the bind hierarchy, where an object can get events from an +object with a different lifetime, so it has to unbind in its destructor. +(Exept in the normal case where all graphical objects get constructed at the +start of the program and destructed at the end of the program.) + +And wxWidgets has the sizer hierarchy, which allocates screen space to +child windows within the parent window. + +Sizer objects have to be detached, and then explicitly deleted, and child +windows have to be detached from the sizer hierarchy, then explicitly +destroyed. And then you call layout on the master sizer of the owning window. + +And then you have the tab hierachy. You have to tell the main window the tab +order of its child windows, which is by default the order of their creation. + +If you are potentially managing an indefinitely large number of windows, +like a browser, you need the wxNotebook control. (Which lacks a close +button on the tabs) + +So, most complex case: You have sizer that acts like a window within a +larger window. It is a sizer within a larger sizer. It gets constructed in +response to user events, and destructed in response to user events, while +the main window continues. The windows within the frame are direct +children of larger window, and they are also children of the sizer, which is +a child of the main sizer, which is owned and controlled by the larger +window. + +But, for future integration with wxNotebook, probably best to put it into an +object of type wxPanel. So you define an object that will contain ordinary +dumb pointers to all these objects, which takes the parent window and the +parent sizer as a arguments. On construction it constructs all the wxWidgets +child windows that it needs, with the main window as their window parent, +puts them into its sizer and prepends its sizer to the main sizer as their +sizer grandparent, binds them, calls layout on the main sizer, and on +destruction unbinds them, detaches them from their sizer, calls delete on the +sizer objects, then destroy on the window objects and then calls layout on +the main sizer. + +But the way the experts do it is to make that object a wxPanel window. + +On construction, first construct windows, then put them into the sizer, then +bind them, then call layout. On destruction, first unbind, then detach from +sizer, then destroy sizer objects, then destroy windows objects, then call +layout on the main sizer. + +## wxAppConsole + +Generates an event based program with no gui, console input and console +output. Should use only objects in wxBase. wxBase is a separate library, and +you should avoid linking to all the other libraries, except wxNet, and not +even wxCore, because if you are linking to another library, the other +routines in the other library will expect a gui. + +Our program that synchronizes with other peers is going to need to run as a +daemon on Linux and service on windows, with no UI except it talks to the +client wallet, which will run as a regular gui program, possibly on another +machine far away. + +It is probably easier if we use the same library, wxWidgets, for OS +independence for both daemon and client. + +## wxWidgets support + +[wxWidgets mailing list.] +(https://www.wxwidgets.org/support/mailing-lists/guide/) I have never used +this. I have often had my share of grief on wxWidgets, but by the time I +reduced it to a problem capable of being addressed on the mailing list, the +problem was always solved. + +Which tells me that wxWidgets is pretty good, and the considerable grief that +it generates is only that which is inevitable on a multiplatform gui. + +# Instructions for Libsodium + +Supports Ristretto25519, the crypto tool that I am competent to use. Nothing else seems to support this. + +[Libsodium](https://download.libsodium.org/libsodium/content/) is the authoritative library, derived most directly from Bernstein. + + git clone https://github.com/jedisct1/libsodium.git + cd libsodium + +Compiled as a separate project under visual studio. Retarget the solution to use the current libraries, retarget to use x64, and change the code generation default (properties/C++/code For the your visual studio project to include the sodium.h file, have to define SODIUM_STATIC before including the header file when linking to the static sodium library. + +```h + #define SODIUM_STATIC + #include + #pragma comment(lib, "libsodium.lib") +``` + +To link to libsodium.lib have to add `$(SODIUM)\Build\Debug\x64` to Visual Studio/project properties/linker/General/Additional Library Directories, where SODIUM is the environment variable pointing to the root libsodium directory, and similarly for the Release version of your code `$(SODIUM)\Build\Release\x64`. Assembly code is disabled by default, need to re-enable it once unit test works with encrypted utd communication. + +We need to eventually re-organize as a git subproject and autotools subproject, but, for the moment, it is a separate library project, hence the environment variable SODIUM. + +Might be better to not use environment variables $(SODIUM) and to use Git submodules and autotools submodules instead, was wxWidgets does internally. + +Set the libsodium project properties to DebugLib x64 + +Set the libsodium project properties/General/configuration type to static (should be static already if you have selected Debug Lib and ReleaseLib) + +It provides all the crypto algorithms you need, including Argon2 for password stretching, and freedom from NSA. Neglects to include base 58 encoding. Wrote my own base 64 library, which differs from the standard base 64 library in being oriented to small items, addressing arbitrary bit fields rather than blocks of three bytes, in being url safe, and in mapping 0, o, and O to zero, and mapping 1, i, I, and l to one, to allow for human entry. + +Going to need an algorithm for generating passphrases, but passphrases will be hashed and argoned, + +# Instructions for Sqlite + +Sqlite is not incorporated as an already built library, nor as a library at +all, but as source code. + +When you add the sqlite3.c amalgamation, make sure to mark it as not using +precompiled headers before the first attempt to compile it, otherwise it +breaks horribly, (C++ format precompiled headers being incompatible with C +precompiled headers) and when you belatedly turn off the precompiled headers, +some data that Visual Studio has generated hangs around, so that turning off +the precompiled headers fails fix the problem. + +Similarly, if you do a typo in the include paths. Remains stuck on the old +include paths even if you fix it. To do a real clean, close down visual +studio, delete the directories generated by visual studio, and *then* your +edits to the configuration will propagate properly + +[SQLite](https://sqlite.org/index.html) + +[Download](https://sqlite.org/download.html) + +When creating the database, should turn WAL on:`PRAGMA journal_mode=WAL; PRAGMA synchronous=1;` but before distributing it, make sure the WAL file has been cleaned out with sqlite3_close(). + +Wal mode ensures that we never get a busy error from normal business, only under abnormal conditions, provided that only one thread in one process ever writes to the the database. Which means we do not face the hard problem of handling the busy error. + +We have also chosen our threading model so that database connections and their associated compiled sql statements are thread local, but one database connection can be connected to many databases, and each database can have many database connections from many threads and many processes + +Wal mode in the default settings means that writes are normally instantaneous, because they do not hit the disk, but every so often, on completing a transaction, four megabytes get written to disk (which saves a whole lot of read/writes to disk, replacing them with sequential writes without too many reads). So it is probably not a good idea to download large amounts of data from the internet on the UI thread, because every so often the write thread is going to do four megabyes of actual write to actual disk. + +In Wal mode, recently “written” transaction data will be read from memory. +So it is a premature optimization to keep data in program memory, when the +same data is probably in Wal memory or in cache. + +Because Wal writes are near instantaneous, multiple threads writing seldom +block each other – but if one thread is busy writing while the previous +thread is busy flushing four megabytes of actual data to disk, the Wal file +may grow excessively. According to the documentation, a write thread does the +flushing, but it looks to me that a background thread does the flushing. + +But the UI thread will typically only be doing reads and writes every now and +then, so if one has only one continually writing thread, you don’t have to +use restart or full which means you very rarely get busy calls, so do not +have to handle them elegantly and efficiently. + +I suspect that the wal file gets cleaned mighty fast, so before doing anything clever, measure.by running `PRAGMA wal_checkpoint` every minute or so, which returns a table of one row and three columns telling you how much stuff needs to be written when the checkpoint started, and how much was going to be written when it completed, leaving another thread working on actually writing it. (Two PRAGMA wal_checkpoints in rapid succession are apt to give the same answer) Also log how long the checkpoint took. If the second column is substantially larger than a thousand, and the third column is kind of small, you have a checkpoint problem. This problem is unlikely to happen, because the disk has a much bigger pipe than the network, so it looks to me that with the usual settings, wal and passive checkpointing, all the actual disk write IO is going to take place in a backround thread, because we are just not going to be disk bound on updates and writes. We may find ourselves disk bound on random access reads, but Wal, transactions, and checkpoints are not relevant to that problem. + +```bash +del C:\Users\studi\AppData\Local\RoBlockchain\RoData3cx8bdx.sqlite +sqlite3 C:\Users\studi\AppData\Local\RoBlockchain\RoData3cx8bdx.sqlite +``` + +Where the file `RoData3cx8bdx.sqlite` contains: + +```sql +PRAGMA encoding = "UTF-8"; +PRAGMA journal_mode=WAL; +PRAGMA foreign_keys = OFF; +PRAGMA synchronous = 1; +CREATE TABLE w( --wallets + PublicKey BLOB PRIMARY KEY NOT NULL UNIQUE, + Name TEXT NOT NULL UNIQUE, – automatically creates index + PetName TEXT, + title TEXT +); +CREATE TABLE MasterSecret( + PublicKey BLOB PRIMARY KEY NOT NULL UNIQUE, + Secret BLOB, + Expires INTEGER NOT NULL, + FOREIGN KEY(PublicKey) REFERENCES w(PublicKey) +); +INSERT INTO w VALUES(X'deadbeef','Adam',NULL,NULL); + .dump + .output RoData3cx8bdx.sql + .dump + .exit +``` + +We are stashing everything in the user specific directory `wxStandardPaths::GetAppDocumentsDir()` +and the user specific database, which is wrong. Eventually we will need two databases, one global to all users on a particular machine, which you select when you install, and one for each particular user, that gets generated when the particular user first runs his wallet. + +Further confusing matters, wxConfigBase settings are always per user on windows. + +At the moment, only one database and one config object. + +In our own code, we don’t need to issue `PRAGMA synchronous = 1; PRAGMA +foreign_keys = ON;` because we modify the sqlite3.c with the following: + +I need to customize the sqlite3.c almalgamation with my own custom compile options as follows. + +```C +//My custom compile options +#define SQLITE_DQS 0 //Doublequote names, single quote strings. This setting disables the double – quoted string literal misfeature. +#define SQLITE_THREADSAFE 2 //One thread, one database connection. Data structures such as compiled SQL are threadlocal. But sqlite3 is empowered to do its own multithreading. Many databases per database connection. Database connection and compiled sql statements are threadlocal. last_insert_rowid() is not subject to race conditions in this mode, returning the most recent rowid generated by the thread. +#define SQLITE_DEFAULT_MEMSTATUS 0 //Don’t track memory usage. Disables the ability of the program using sqlite3 to monitor its memory usage. This setting causes the sqlite3_status() interfaces that track memory usage to be disabled. This helps the sqlite3_malloc() routines run much faster, and since SQLite uses sqlite3_malloc() internally, this helps to make the entire library faster. +#define SQLITE_DEFAULT_WAL_SYNCHRONOUS 1 // in WAL mode, recent changes to the database might be rolled back by a power loss, but the database will not be corrupted. Furthermore, transaction commit is much faster in WAL mode using synchronous=NORMAL than with the default synchronous=FULL. For these reasons, it is recommended that the synchronous setting be changed from FULL to NORMAL when switching to WAL mode. This compile-time option will accomplish that. +#define SQLITE_DEFAULT_FOREIGN_KEYS 0 //Dont handle foreign key constraints. Programmer has to do it himself. +#define SQLITE_LIKE_DOESNT_MATCH_BLOBS 1 //Blobs are not strings. Historically, SQLite has allowed BLOB operands to the LIKE and GLOB operators. But having a BLOB as an operand of LIKE or GLOB complicates and slows the LIKE optimization. When this option is set, it means that the LIKE and GLOB operators always return FALSE if either operand is a BLOB. That simplifies the implementation of the LIKE optimization and allows queries that use the LIKE optimization to run faster. +#define SQLITE_MAX_EXPR_DEPTH 0 //Setting the maximum expression parse-tree depth to zero disables all checking of the expression parse-tree depth, which simplifies the code resulting in faster execution, and helps the parse tree to use less memory. +#define SQLITE_OMIT_DECLTYPE 1 // By omitting the (seldom-needed) ability to return the declared type of columns from the result set of query, prepared statements can be made to consume less memory. +#define SQLITE_OMIT_DEPRECATED 1 +#define SQLITE_DQS 0 //Don’t accept double quoted string literals. +#define SQLITE_OMIT_PROGRESS_CALLBACK 1 +#define SQLITE_OMIT_SHARED_CACHE 1 +#define SQLITE_OMIT_UTF16 1 +#define SQLITE_USE_ALLOCA 1 //Make use of alloca() for dynamically allocating temporary stack space for use within a single function, on systems that support alloca(). Without this option, temporary space is allocated from the heap +#define SQLITE_OMIT_LOAD_EXTENSION 1 +#define SQLITE_TEMP_STORE 3 //Temporary files are in memory +#define SQLITE_OMIT_AUTOINIT 1 //.The SQLite library needs to be initialized using a call to sqlite3_initialize() before certain interfaces are used.This initialization normally happens automatically the first time it is needed.However, with the SQLITE_OMIT_AUTOINIT option, the automatic initialization is omitted.This helps many API calls to run a little faster(since they do not have to check to see if initialization has already occurredand then run initialization if it has not previously been invoked) but it also means that the application must call sqlite3_initialize() manually.If SQLite is compiled with – DSQLITE_OMIT_AUTOINIT and a routine like sqlite3_malloc() or sqlite3_vfs_find() or sqlite3_open() is invoked without first calling sqlite3_initialize(), the likely result will be a segfault +//end my custom compile options*/ +``` + +[To compile the standard sqlite3.exe tool](https://www.sqlite.org/howtocompile.html). + +We don’t need to issue `PRAGMA journal_mode=WAL; PRAGMA schema.synchronous += 1;` in our own code except if we create a new database, which our code will + typically not do, because our install will copy a working database. + +In our code, we do need to issue `PRAGMA optimize;` on close. + +Missing from their short summary of the C interface is: + +```C + /* + ** Return UTF-8 encoded English language explanation of the most recent + ** error. + */ + SQLITE_API const char *sqlite3_errmsg(sqlite3 *db) +``` + +The general rule being that for any unexpected value of rc, any value other than `SQLITE_OK`, `SQLITE_ROW`, and `SQLITE_DONE`, display that message. `sqlite3_exec` wraps this, but we should probably not use `sqlite3_exec`. + +SQlite’s pragma for identifying that something is the right application file format does not seem to work, or maybe it does work and there is some magic that I am not aware of. But this does not matter, because it only protects against Murphy, not Machiavelli, and we are going to need to guard against Machiavelli. + +We need to make sure that compiled sql statements are only compiled after the database connection is live, and destroyed before the database is closed. + +The order of construction and destruction within an object is + +- First, and only for the constructor of the most derived class as described below, virtual base classes shall be initialized in the order they appear on a depth-first left-to-right traversal of the directed acyclic graph of base classes, where “left-to-right” is the order of appearance of the base class names in the derived class base-specifier-list. +- Then, direct base classes shall be initialized in declaration order as they appear in the base-specifier-list (regardless of the order of the mem-initializers). +- Then, non-static data members shall be initialized in the order they were declared in the class definition (again regardless of the order of the mem-initializers). +- Finally, the compound-statement of the constructor body is executed. \[ Note: the declaration order is mandated to ensure that base and member subobjects are destroyed in the reverse order of initialization. + +Objects on the stack are destroyed in the reverse of the order that they were declared. + +The gui object that enables the user to manipulate the contents of the database should contain the database connection, and the compiled sql objects that manipulate the database, declared in that order, so that they get destroyed in that reverse order, compiled sql being destroyed before the database connection is destroyed, and so that in order to create the gui to manipulate the database, we have to first construction a database connection. Since the creation could throw, and we need to handle the throw in the gui, we need a gui already constructed, which then attempts to construct more gui, and informs the user if the construction fails. + +So the outer gui figures out a plausible database name, and then attempts to construct the inner gui whose constructor then attempts to open a database connection and make sure the database is in a good state, and throws if it cannot construct a good connection to a good database. + +To protect against Machiavelli, peers have to by default check the early part of the block chain to make sure it has the correct hash, preferably in obfuscated and undocumented code, accessed through a define located in a rather random header file so that Machiavelli will have to audit my code, then supply his victims with the new code as well as the new database. Or just hard code the genesis block, though Machiavelli will have an easier time finding that one. Maybe both. Hard code the genesis block, and have obfuscated hard code for the early blocks. diff --git a/docs/libraries/cpp_automatic_memory_management.md b/docs/libraries/cpp_automatic_memory_management.md new file mode 100644 index 0000000..5382aef --- /dev/null +++ b/docs/libraries/cpp_automatic_memory_management.md @@ -0,0 +1,611 @@ +--- +title: + C++ Automatic Memory Management +--- +# Memory Safety +Modern, mostly memory safe C++, is enforced by:\ + +- gsl +- Microsoft safety checker +- Guidelines +- language checker + +`$ clang-tidy test.cpp -checks=clang-analyzer-cplusplus*, cppcoreguidelines-*, modernize-*` will catch most of the issues that esr +complains about, in practice usually all of them, though I suppose that as +the project gets bigger, some will slip through. + + static_assert(__cplusplus >= 201703, "C version of out of date"); + +The gsl adds span for pointer arithmetic, where the +size of the array pointed to is kept with the pointer for safe iteration and +bounds checking during pointer maths. This should be available in the standard template library with C20. + +Modern C++ as handles arrays as arrays where possible, but they quickly +decay to pointers – which you avoid using spans. std::array is a C array +whose size is known at compile time, and which is protected from decay to +a pointer. std::vector is a dynamically resizable and insertable array +protected from decay to a pointer – which can have significant overheads. +std::make_unique, std::make_shared create pointers to memory managed +objects. (But single objects, not an array, use spans for pointer +arithmetic) + + auto sp = std::make_shared(42); + std::weak_ptr wp{sp}; + +# Array sizing and allocation + + /* This code creates a bunch of "brown dog" strings on the heap to test automatic memory management. */ + char ca[]{ "red dog" }; //Automatic array sizing + std::array arr{"red dog"}; //Requires #include + /* No automatic array sizing, going to have to count your initializer list. */ + /* The pointer of the underlying array is referenced by &arr[0] but arr is not the underlying array, nor a pointer to it. */ + /* [0] invokes operator[], and operator[] is the member function that accesses the underlying array.*/ + /* The size of the underlying array is referenced by arr.size();*/ + /* size known at compile time, array can be returned from a function getting the benefits of stack allocation.*/ + // can be passed around like POD + char *p = new char[10]{ "brown dog" }; //No automatic array + // sizing for new + std::unique_ptrpuc{ p }; // Now you do not have + // to remember to delete p + auto puc2 = std::move(puc); /* No copy constructor. Pass by reference, or pass a view, such as a span.*/ + std::unique_ptr puc3{ new char[10]{ "brown dog" } }; + /* Array size unknown at compile or run time, needs a span, and you have to manually count the initialization list. */ + /* Compiler guards against overflow, but does not default to the correct size.*/ + /* You can just guess a way too small size, and the compiler in its error message will tell you what the size should be. */ + auto pu = std::make_unique(10); // uninitialized, + // needs procedural initialization. + + /* span can be trivially created from a compile time declared array, an std:array or from a run time std:: vector, but then these things already have the characteristics of a span, and they own their own storage. */ + /* You would use a span to point into an array, for example a large blob containing smaller blobs.*/ + + // Placement New: + char *buf = new char[1000]; //pre-allocated buffer + char *p = buf; + MyObject *pMyObject = new (p) MyObject(); + p += (sizeof(MyObject+7)/8)*8 + /* Problem is that you will have to explictly call the destructor on each object before freeing your buffer. */ + /* If your objects are POD plus code for operating on POD, you don’t have to worry about destructors.*/ + // A POD object cannot do run time polymorphism. + /* The pointer referencing it has to be of the correct compile time type, and it has to explicitly have the default constructor when constructed with no arguments.*/ + /* If, however, you are building a tree in the pre-allocated buffer, no sweat. */ + /* You just destruct the root of the tree, and it recursively destructs all its children. */ + /* If you want an arbitrary graph, just make sure you have owning and non owning pointers, and the owning pointers form a tree. */ + /* Anything you can do with run time polymorphism, you can likely do with a type flag.*/ + + static_assert ( std::is_pod() , "MyType for some reason is not POD" ); + class MyClass + { + public: + MyClass()=default; // Otherwise unlikely to be POD + MyClass& operator=(const MyClass&) = default; // default assignment Not actually needed, but just a reminder. + }; + + ### alignment + + ```c++ + // every object of type struct_float will be aligned to alignof(float) boundary +// (usually 4) +struct alignas(float) struct_float { + // your definition here +}; + +// every object of type sse_t will be aligned to 256-byte boundary +struct alignas(256) sse_t +{ + float sse_data[4]; +}; + +// the array "cacheline" will be aligned to 128-byte boundary +alignas(128) char cacheline[128]; + ``` + +# Construction, assignment, and destruction + +six things: ([default +constructor](https://en.cppreference.com/w/cpp/language/default_constructor), +[copy +constructor](https://en.cppreference.com/w/cpp/language/copy_constructor), +[move +constructor](https://en.cppreference.com/w/cpp/language/move_constructor), +[copy +assignment](https://en.cppreference.com/w/cpp/language/copy_assignment), +[move +assignment](https://en.cppreference.com/w/cpp/language/move_assignment) +and [destructor](https://en.cppreference.com/w/cpp/language/destructor)) +are generated by default – except when they are not. + +So it is arguably a good idea to explicitly declare them as default or +deleted. + +Copy constructors + + A(const A& a) + +Copy assignment + + A& operator=(const A other) + +Move constructors + + class_name ( class_name && other) + A(A&& o) + D(D&&) = default; + +Move assignment operator + + V& operator=(V&& other) + +Move constructors + + class_name ( class_name && ) + +## rvalue references + +Move constructors and copy constructors primarily exist to tell the +compiler how to handle temporary values, rvalues, that have references to possibly +costly resources. + +`class_name&&` is rvalue reference, the canonical example being a reference to a compiler generated temporary. + +The primary purpose of rvalue references is to support move semantics in +objects that reference resources, primarily unique_pointer. + +`std::move(t)` is equivalent to `static_cast(t)`, causing move +semantics to be generated by the compiler. + +`t`, the compiler assumes is converted by your move constructor or move assignment into a valid state where your destructor will not need to anything very costly. + +`std::forward(t)` causes move semantics to be invoked iff the thing referenced +is an rvalue, typically a compiler generated temporary, *conditionally* +forwarding the resources. + +where `std::forward` is defined as follows: + + template< class T > struct remove_reference { + typedef T type; + }; + template< class T > struct remove_reference { + typedef T type; + }; + template< class T > struct remove_reference { + typedef T type; + }; + + template + S&& forward(typename std::remove_reference::type& a) noexcept + { + return static_cast(a); + } + +`std::move(t)` and `std::forward(t)` don't actually perform any action +in themselves, rather they cause the code referencing `t` to use the intended +copy and intended assignment. + +## constructors and destructors + +If you declare the destructor deleted that prevents the compiler from +generating its own, possibly disastrous, destructor, but then, of +course, you have to define your own destructor with the exact same +signature, which would ordinarily stop the compiler from doing that +anyway. + +When you declare your own constructors, copiers, movers, and deleters, +you should generally mark them noexcept. + + struct foo { + foo() noexcept {} + foo( const foo & ) noexcept { } + foo( foo && ) noexcept { } + ~foo() {} + }; + +Destructors are noexcept by default. If a destructor throws an exception as +a result of a destruction caused by an exception, the result is undefined, +and usually very bad. This problem is resolved in complicated ad hoc +ways that are unlikely to be satisfactory. + +If you need to define a copy constructor, probably also need to define +an assignment operator. + + t2 = t1; /* calls assignment operator, same as "t2.operator=(t1);" */ + Test t3 = t1; /* calls copy constructor, same as "Test t3(t1);" */ + +## casts + +You probably also want casts. The surprise thing about a cast operator +is that its return type is not declared, nor permitted to be declared, +DRY. Operator casts are the same thing as constructors, except declared +in the source class instead of the destination class, hence most useful +when you are converting to a generic C type, or to the type of an +external library that you do not want to change. + + struct X { + int y; + operator int(){ return y; } + operator const int&(){ return y; } /* C habits would lead you to incorrectly expect "return &y;", which is what is implied under the hood. */ + operator int*(){ return &y; } // Hood is opened. + }; + +Mpir, the Visual Studio skew of GMP infinite precision library, has some +useful and ingenious template code for converting C type functions of +the form `SetAtoBplusC(void * a, void * b, void * c);` into C++ +expressions of the form `a = b+c*d;`. It has a bunch of intermediate +types with no real existence, `__gmp_expr<>` and `__gmp_binary_expr<>` +and methods with no real existence, which generate the appropriate +calls, a templated function of potentially unlimited complexity, to +convert such an expression into the relevant C type calls using +pointers. See section mpir-3.0.0.pdf, section 17.5 “C++ Internals”. + +I don’t understand the Mpir code, but I think what is happening is that +at run time, the binary expression operating on two base types creates a +transient object on the stack containing pointers to the two base types, +and the assignment operator and copy create operator then call the +appropriate C code, and the operator for entities of indefinite +complexity creates base type values on the stack and a binary expression +operator pointing to them. + +Simpler, but introducing a redundant copy, to always generate +intermediate values on the stack, since we have fixed length objects +that do not need dynamic heap memory allocation, not that costly, and +they are not that big, at worst thirty two bytes, so clever code is apt +to cost in overheads of pointer management + +That just means we are putting 256 bits of intermediate data on the +stack instead of 128, hardly a cost worth worrying about. And in the +common bad case, (a+b)\*(c+d) clever coding would only save one stack +allocation and redundant copy. + +# Template specialization + + namespace N { + template class Y { /*...*/ }; // primary template + template<> class Y ; // forward declare specialization for double + } + template<> + class N::Y { /*...*/ }; // OK: specialization in same namespace + +is used when you have sophisticated template code, because you have to +use recursion for looping as the Mpir system uses it to evaluate an +arbitrarily complex recursive expression – but I think my rather crude +implementation will not be nearly so clever. + + extern template int fun(int); + /*prevents redundant instantiation of fun in this compilation unit – and thus renders the code for fun unnecessary in this compilation unit.*/ + +# Template traits, introspection + +Template traits: C++ has no syntactic sugar to ensure that your template +is only called using the classes you intend it to be called with. + +Often you want different templates for classes that implement similar functionality in different ways. + +This is the entire very large topic of template time, compile time code, which is a whole new ball of wax that needs to be dealt with elsewhere + +# Abstract and virtual + +An abstract base class is a base class that contains a pure virtual +function ` virtual void features() = 0;`. + +A class can have a virtual destructor, but not a virtual constructor. + +If a class contains virtual functions, then the default constructor has +to initialize the pointer to the vtable. Otherwise, the default +constructor for a POD class is empty, which implies that the default +destructor is empty. + +The copy and swap copy assignment operator, a rather slow and elaborate +method of guaranteeing that an exception will leave the system in a good +state, is never generated by default, since it always relates to rather +clever RAII. + +An interface class is a class that has no member variables, and where +all of the functions are pure virtual! In other words, the class is +purely a definition, and has no actual implementation. Interfaces are +useful when you want to define the functionality that derived classes +must implement, but leave the details of how the derived class +implements that functionality entirely up to the derived class. + +Interface classes are often named beginning with an I. Here’s a sample +interface class:. + + class IErrorLog + { + public: + virtual bool openLog(const char *filename) = 0; + virtual bool closeLog() = 0; + + virtual bool writeError(const char *errorMessage) = 0; + + virtual ~IErrorLog() {} // make a virtual destructor in case we delete an IErrorLog pointer, so the proper derived destructor is called + // Notice that the virtual destructor is declared to be trivial, but not declared =0; + }; + +[Override +specifier](https://en.cppreference.com/w/cpp/language/override) + + struct A + { + virtual void foo(); + void bar(); + }; + + struct B : A + { + void foo() const override; // Error: B::foo does not override A::foo + // (signature mismatch) + void foo() override; // OK: B::foo overrides A::foo + void bar() override; // Error: A::bar is not virtual + }; + +Similarly [Final +specifier](https://en.cppreference.com/w/cpp/language/final) + +[To obtain aligned +storage](http://www.cplusplus.com/reference/type_traits/aligned_storage/)for +use with placement new + + void* p = aligned_alloc(sizeof(NotMyClass)); + MyClass* pmc = new (p) MyClass; //Placement new. + // ... + pmc->~MyClass(); //Explicit call to destructor. + aligned_free(p);. + +# GSL: Guideline Support Library + +The Guideline Support Library (GSL) contains functions and types that +are suggested for use by the C++ Core Guidelines maintained by the +Standard C++ Foundation. This repo contains [Microsoft’s implementation +of GSL](https://github.com/Microsoft/GSL). + + git clone https://github.com/Microsoft/GSL.git + cd gsl + git tag + git checkout tags/v2.0.0 + +Which implementation mostly works on gcc/Linux, but is canonical on +Visual Studio. + +For usage of spans ([the replacement for bare naked non owning pointers +subject to pointer +arithmetic)](http://codexpert.ro/blog/2016/03/07/guidelines-support-library-review-spant/) + +For usage of string spans ([String +spans](http://codexpert.ro/blog/2016/03/21/guidelines-support-library-review-string_span/) +These are pointers to char arrays. There does not seem to be a UTF‑8 +string_span. + +GSL is a preview of C++20, as boost contained a preview of C++11. + +It is disturbingly lacking in official documentation, perhaps because +still subject to change. + +[Unofficial +documentation](http://modernescpp.com/index.php/c-core-guideline-the-guidelines-support-library) + +It provides an optional fix for C’s memory management problems, while +still retaining backward compatibility to the existing pile of rusty +razor blades and broken glass. + +# The Curiously Recurring Template Pattern + +[CRTP](https://www.fluentcpp.com/2017/05/16/what-the-crtp-brings-to-code/), +makes the relationship between the templated base class or classes and +the derived class cyclic, so that the derived class tends to function as +real base class. Useful for mixin classes. + + template class Mixin1{ + public: + // ... + void doSomething() //using the other mixin classes and the derived class T + { + T& derived = static_cast(*this); + // use derived... + } + private: + mixin1(){}; // prevents the class from being used outside the mix) + friend T; + }; + + template class Mixin2{ + { + public: + // ... + void doSomethingElse() + { + T& derived = static_cast(*this); + // use derived... + } + private: + Mixin2(){}; + friend T; + }; + + class composite: public mixin1, public mixin2{ + composite( int x, char * y): mixin1(x), mixin2(y[0]) { ...} + composite():composite(7,"a" ){ ...} + } + +# Aggregate initialization + +A class of aggregate type has no constructors – the aggregate +constructor is implied default. + +A class can be explicitly defined to take aggregate initialization + + Class T{ + T(std::initializer_list in){ + for (auto i{in.begin); i + void myFunction(F&& lambda){ + //some things + +You can put a lambda in a class using decltype,and pass it around for +continuations, though you would probably need to template the class: + + templateclass foo { + public: + T func; + foo(T in) :func{ in } {} + auto test(int x) { return func(x); } + }; + .... + auto bar = [](int x)->int {return x + 1; }; + foo<(bar)>foobar(bar); + +But we had to introduce a name, bar, so that decltype would have +something to work with, which lambdas are intended to avoid. If we are +going to have to introduce a compile time name, easier to do it as an +old fashioned function, method, or functor, as a method of a class that +is very possibly pod. + +If we are sticking a lambda around to be called later, might copy it by +value into a templated class, or might put it on the heap. + + auto bar = []() {return 5;}; + +You can give it to a std::function: + + auto func_bar = std::function(bar); + +In this case, it will get a copy of the value of bar. If bar had +captured anything by value, there would be two copies of those values on +the stack; one in bar, and one in func_bar. + +When the current scope ends, func_bar will be destroyed, followed by +bar, as per the rules of cleaning up stack variables. + +You could just as easily allocate one on the heap: + + auto bar_ptr = std::make_unique(bar); + + std::function increm{[](int arg{return arg+1;}} + +presumably does this behind the scenes + +On reflection we could probably use this method to produce a +templated function that stored a lambda somewhere in a templated class +derived from a virtual base class for execution when the event triggered +by the method fired, and returned a hashcode to the templated object for +the event to use when the event fired. The event gets the event handler +from the hashcode, and the virtual base class in the event handler fires +the lambda in the derived class, and the lambda works as a continuation, +operating in the context wherein it was defined, making event oriented +programming almost as intuitive as procedural programming. + +But then we have a problem, because we would like to store event +handlers in the database, and restore them when program restarts, which +requires pod event handlers, or event handlers constructible from POD +data, which a lambda is not. + +We could always have some event handlers which are inherently not POD +and are never sent to a database, while other event handlers are, but +this violates the dry design principle. To do full on functional +programming, use std::function and std::bind, which can encapsulate +lambdas and functors, but are slow because of dynamic allocation + +C++ does not play well with functional programming. Most of the time you +can do what you want with lambdas and functors, using a pod class that +defines operator(\...) + +# auto and decltype(variable) + +In good c++, a tremendous amount of code behavior is specified by type +information, often rather complex type information, and the more one’s +code description is in types, the better. + +But specifying types everywhere violates the dry principle, hence, +wherever possible, use auto and decltype(variable) to avoid redundant +and repeated type information. Wherever you can use an auto or a +decltype for a type, use it. + +In good event oriented code, events are not triggered procedurally, but +by type information or data structures, and they are not handled +procedurally, as by defining a lambda, but by defining a derived type. + +# Variable length Data Structures + +C++ just does not handle them well, except you embed a vector in them, +which can result in messy reallocations. + +One way is to drop back into old style C, and tell C++ not to fuck +around. + + struct Packet + { + unsigned int bytelength; + unsigned int data[]; + + private: + // Will cause compiler error if you misuse this struct + void Packet(const Packet&); + void operator=(const Packet&); + }; + Packet* CreatePacket(unsigned int length) + { + Packet *output = (Packet*) malloc((length+1)*sizeof(Packet)); + output->bytelength = length; + return output; + } + +Another solution is to work around C++’s inability to handle variable +sized objects by fixing your hash function to handle disconnected data. + +# for_each + + template + Function for_each(InputIterator first, InputIterator last, Function fn){ + while (first!=last) { + fn (*first); + ++first; + } + return move(fn); + } + +# Range-based for loop + + for(auto x: temporary_with_begin_and_end_members{ code;} + for(auto& x: temporary_with_begin_and_end_members{ code;} + for(auto&& x: temporary_with_begin_and_end_members{ code;} + for (T thing = foo(); auto& x : thing.items()) { code; } + +The types of the begin_expr and the end_expr do not have to be the same, +and in fact the type of the end_expr does not have to be an iterator: it +just needs to be able to be compared for inequality with one. This makes +it possible to delimit a range by a predicate (e.g. “the iterator +points at a null character”). + +If range_expression is an expression of a class type C that has both a +member named begin and a member named end (regardless of the type or +accessibility of such member), then begin_expr is \_\_range.begin() and +end_expr is \_\_range.end(); + + for (T thing = foo(); auto x : thing.items()) { code; } + +Produces code equivalent to: + + T thing = foo(); + auto bar = thing.items(); + auto enditer = bar.end; + for (auto iter = bar.begin(); iter != enditer; ++iter) { + x = *iter; + code; + } diff --git a/docs/libraries/cpp_multithreading.md b/docs/libraries/cpp_multithreading.md new file mode 100644 index 0000000..3fbb289 --- /dev/null +++ b/docs/libraries/cpp_multithreading.md @@ -0,0 +1,519 @@ +--- +title: C++ Multithreading +--- +Computers have to handle many different things at once, for example +screen, keyboard, drives, database, internet. + +These are best represented as communicating concurrent processes, with +channels, as in Go routines. Even algorithms that are not really handling +many things at once, but are doing a single thing, such as everyone’s +sample program, the sieve of Eratosthenes, are cleanly represented as +communicating concurrent processes with channels. + +[asynch await]:../client_server.html#the-equivalent-of-raii-in-event-oriented-code + +On the other hand, also, not quite so cleanly, represented by [asynch await] which makes for much lighter weight code, more cleanly interfaceable with C++. + +Concurrency is not the same thing as parallelism. + +A node.js program is typically thousands of communicating concurrent +processes, with absolutely no parallelism, in the sense that node.js is single +threaded, but a node.js program typically has an enormous number of code +continuations, each of which is in effect the state of a concurrent +communicating process. Lightweight threads as in Go are threads that on +hitting a pause get their stack state stashed into an event handler and +executed by event oriented code, so one can always accomplish the same +effect more efficiently by writing directly in event oriented code. + +And it is frequently the case that when you cleverly implement many +concurrent processes with more than one thread of execution, so that some +of your many concurrent processes are executed in parallel, your program +runs slower, rather than faster. + +C++ multithreading is written around a way of coding that in practice does +not seem all that useful – parallel bitbashing. The idea is that you are +doing one thing, but dividing that one thing up between several threads to get +more bits bashed per second, the archetypical example being a for loop +performed in parallel, and then all the threads join after the loop is +complete. + +The normal case however is that you want to manage a thousand things at +once, for example a thousand connections to the server. You are not +worried about how many millions of floating point operations per second, +but you are worried about processes sitting around doing nothing while +waiting for network or disk operations to complete. + +For this, you need concurrent communicating processes, as in Go or event +orientation as in node.js or nginx, node.js, not necessarily parallelism, +which C++ threads are designed around. + +The need to deal with many peers and a potentially enormous number of +clients suggests multiprocessing in the style of Go and node.js, rather than +what C++ multiprocessing is designed around, suggests a very large +number of processes that are concurrent, but not all that parallel, rather +than a small number of processes that are concurrent and also substantially +parallel. Representing a process by a thread runs into troubles at around +sixty four threads. + +It is probably efficient to represent interactions between peers as threads, +but client/peer are going to need either events or Go lightweight threads, +and client/client interactions are going to need events. + +Existing operating systems run far more than sixty four threads, but this +only works because grouped into processes, and most of those processes +inactive. If you have more than sixty four concurrently active threads in an +active process, with the intent that half a dozen or so of those active +concurrent threads will be actually executing in parallel, as for example a +browser with a thread for each tab, and sixty four tabs, that active process +is likely to be not very active. + +Thus scaling Apache, whether as threads on windows or processes under +Linux, is apt to die. + +# Need the solutions implemented by Tokio, Actix, Node.js and Go + +Not the solutions supplied by the C++ libraries, because we are worrying +about servers, not massive bit bashing. + +Go routines and channels can cleanly express both the kind of problems +that node.js addresses, and also address the kind of problem that C++ +threads address, typically that you divide a task into a dozen subtasks, and +then wait for them all to complete before you take the next step, which are +hard to express as node.js continuations. Goroutines are a more flexible +and general solution, that make it easier to express a wider range of +algorithms concisely and transparently, but I am not seeing any mass rush +from node.js to Go. Most of the time, it is easy enough to write in code +continuations inside an event handler. + +The general concurrent task that Google’s massively distributed database +is intended to express is that you have a thousand tasks each of which +generate a thousand outputs, which get sorted, and each of the enormous +number of items that sort into the same equivalence group gets aggregated +in a commutative operation, which can therefore be handled by any +number of processes in any order, and possibly the entire sort sequence +gets aggregated in an associative operation, which can therefore be +handled by any number of processes in any order. + +The magic in the Google massively parallel database is that one can define a +a massively parallel operation on a large number of items in a database +simultaneously, much as one defines a join in SQL, and one can define +another massively parallel operation as commutative and or associative +operations on the sorted output of such a massively parallel operation. But +we are not much interested in this capability. Though something +resembling that is going to be needed when we have to shard. + +# doing node.js in C++ + +Dumb idea. We already have the node.js solution in a Rust library. + +Actix and Tokio are the (somewhat Cish) solutions. + +## Use Go + +Throw up hands in despair, and provide an interface linking Go to secure +Zooko ids, similar to the existing interface linking it to Quic and SSL. + +This solution has the substantial advantage that it would then be relatively +easy to drop in the existing social networking software written in Go, such +as Gitea. + +We probably don’t want Go to start managing C++ spawned threads, but +the Go documentation seems to claim that when a Go heavyweight thread +gets stuck at a C mutex while executing C code, Go just spawns another to +deal with the lightweight threads when the lightweight threads start piling +up. + +When a C++ thread wants to despatch an event to Go, it calls a Go routine +with a select and a default, so that the Go routine will never attempt to +pause the C++ spawned thread on the assumption that it is a Go spawned +thread. But it would likely be safer to call Goroutines on a thread that was +originally spawned by Go. + +## doing it in C the C way + +Processes represented as threads. Channels have a mutex. A thread grabs +total exclusive ownership of a channel whenever it takes something out or +puts something in. If a channel is empty or full, it then waits on a +condition on the mutex, and when the other thread grabs the mutex and +makes the channel ready, it notices that the other process or processes are +waiting on condition, the condition is now fulfilled, and sends a +notify_one. + +Or, when the channel is neither empty nor full, we have an atomic spin lock, +and when sleeping might become necessary, then we go to full mutex resolution. + +Which implies a whole pile of data global to all threads, which will have +to be atomically changed. + +This can be done by giving each thread two buffers for this global data +subject to atomic operations, and single pointer or index that points to the +currently ruling global data set. (The mutex is also of course global, but +the flag saying whether to use atomics or mutex is located in a data +structure managed by atomics.) + +When a thread wants to atomically update a large object (which should be +sixty four byte aligned) it constructs a copy of the current object, and +atomically updates the pointer to the copy, if the pointer was not changed +while it was constructing. The object is immutable while being pointed at. + +Or we could have two such objects, with the thread spinning if one is in +use and the other already grabbed, or momentarily sleeping if an atomic +count indicates other threads are spinning on a switch awaiting +completion. + +The read thread, having read, stores its read pointer atomically with +`memory_order_release`, ored with the flag saying if it is going to full +mutex resolution. It then reads the write pointer with +`memory_order_acquire`, that the write thread atomically wrote with +`memory_order_release`, and if all is well, keeps on reading, and if it is +blocked, or the write thread has gone to mutex resolution, sets its mutex +resolution flag and proceeds to mutex resolution. When it is coming out of +mutex resolution, about to release the mutex, it clears its mutex resolution +flag. The mutex is near the flags by memory location, all part of one object +that contains a mutex and atomic variables. + +So the mutex flag is atomically set when the mutex has not yet been +acquired, but the thread is unconditionally going to acquire it, but non +atomically cleared when the mutex still belongs to the thread, but is +unconditionally going to release it. + +If many read threads reading from one channel, then each thread has to +`memory_order_acquire` the read pointer, and then, instead of +`memory_order_release`ing it, has to do an +`atomic_compare_exchange_weak_explicit`, and if it changed while it was +reading abort its reads and start over. + +Similarly if many write threads writing to one channel, each write thread +will have first spin lock acquire the privilege of being the sole write thread +writing, or spin lock acquire a range to write to. Thus in the most general +case, we have a spin locked atomic write state that specifies an area that +has been written to, an area that is being written to, and an area that is +available to be acquired for writing, a spin locked atomic read state, and +mutex that holds both the write state and the read state. In the case of a +vector buffer with multiple writers, the atomic states are three wrapping +atomic pointers that go through the buffer in the same direction, + +We would like to use direct memory addresses, rather than vector or deque +addresses, which might require us to write our own vector or deque. See +the [thread safe deque](https://codereview.stackexchange.com/questions/238347/a-simple-thread-safe-deque-in-c "A simple thread-safe Deque in C++"), which however relies entirely on locks and mutexes, +and whose extension to atomic locks is not obvious. + +Suppose you are doing atomic operations, but some operations might be +expensive and lengthy. You really only want to spin lock on amending data +that is small and all in close together in memory, so on your second spin, +the lock has likely been released. + +Well, if you might need to sleep a thread, you need a regular mutex, but +how are you going to interface spin locks and regular mutexes? + +You could cleverly do it with notifies, but I suspect it is costly compared +to just using a plain old vanilla mutex. Instead you have some data +protected by atomic locks, and some data protected by regular old +mutexes, and any time the data protected by the regular old mutex might +change, you atomically flag a change coming up, and every thread then +grabs the mutex in order to look amend or even look at the data, until on +coming out of the mutex with the data, they see the flag saying the mutex +protected data might change is now clear. + +After one has flagged the change coming up, and grabbed the mutex, wha +happens if another thread is cheerfully amending the data in a fast +operation, having started before you grabbed the mutex? The other thread +has to be able to back out of that, and then try again, this try likely to be +with mutex resolution. But what if the other thread wants to write into a +great big vector, and reallocations of the vector are mutex protected. And +we want atomic operations so that not everyone has to grab the mutex every +time. + +Well, any time you want to do something to the vector, it fits or it does not. +And if it does not fit, then mutex time. You want all threads to switch +to mutex resolution, before any thread actually goes to work reallocating +the vector. So you are going to have to use the costly notify pattern. “I am +out of space, so going to sleep until I can use the mutex to amend the +vector. Wake me up when last thread using atomics has stopped using +atomics that directly reference memory, and has switched to reading the +mutex protected data, so that I can change the mutex protected data.” + +The std::vector documentation says that vector access is just as efficient as +array access, but I am a little puzzled by this claim, as a vector can be +moved, and specifically requests that you have a no throw move operation for +optimization, and having a no copy is standard where it contains things that +might have ownership. (Which leads to complications when one has containers +of containers, since C++ is apt to helpfully generate a broken copy +implementation.) + +Which would suggest that vector access is through indirection, and +indirects with threading create problems. + +## lightweight threads in C + +A lightweight thread is just a thread where, whenever a lightweight thread +needs to be paused by its heavyweight thread, the heavyweight thread +stores the current stack state in the heap, and move on to deal with other +lightweight threads that need to be taken care of. Which collection of +preserved lightweight thread stack states amount to a pile of event +handlers that are awaiting events, and having received events, are then +waiting for a heavyweight thread to process that event handler. + +Thus one winds up with what suspect it the Tokio solution, a stack that +is a tree, rather than a stack. + +Hence the equivalence between node.js and nginx event oriented +programming, and Go concurrent programming. + +# costs + +Windows 10 is limited to sixty four threads total. If you attempt to create +more threads than that, it still works, but performance is apt to bite, with +arbitrary and artificial thread blocking. Hence goroutines, that implement +unofficial threads inside the official threads. + +Thread creation and destruction is fast, five to twenty microseconds, so +thread pools do not buy you much, except that your memory is already +going to be cached. Another source says 40 microseconds on windows, +and fifty kilobytes per thread. So, a gigabyte of ram could have twenty +thousand threads hanging around. Except that the windows thread +scheduler dies on its ass. + +There is a reasonable discussion of thread costs [here](https://news.ycombinator.com/item?id=22456642) + +General message is that lots of languages have done it better, often +immensely better, Go among them. + +Checking the C++ threading libraries, they all single mindedly focus on +the particular goal of parallelizing computationally intensive work. Which +is not in fact terribly useful for anything you are interested in doing. + +# Atomics + +```C++ +typedef enum memory_order { + memory_order_relaxed, // relaxed + memory_order_consume, // consume + /* No one, least of all compiler writers, understands what + "consume" does. + It has consequences which are difficult to understand or predict, + and which are apt to be inconsistent between architectures, + libraries, and compilers. */ + memory_order_acquire, // acquire + memory_order_release, // release + memory_order_acq_rel, // acquire/release + memory_order_seq_cst // sequentially consistent + /* "sequentially consistent" interacts with the more commonly\ + used acquire and release in ways difficult to understand or + predict, and in ways that compiler and library writers + disagree on. */ +} memory_order; +``` + +I don’t think I understand how to use atomics correctly. + +`Atomic_compare_exchange_weak_explicit` inside a while loop is +a spin lock, and spin locks are complicated, apt to be inefficient, +potentially catastrophic, and avoiding catastrophe is subtle and complex. + +To cleanly express a concurrent algorithm you need a thousand +communicating processes, as goroutines or node.js continuations, nearly +all of which are sitting around waiting for the another thing to send them +a message or be ready to receive their message, while atomics give you a +fixed small number of threads all barreling full speed ahead. Whereupon +you find yourself using spin locks. + +Rather than moving data between threads, you need to move threads between +data, between one continuation and the next. + +Well, if you have a process that interacts with Sqlite, each thread has to +have its own database connection, in which case it needs to be a pool of +threads maybe you have a pool of database threads that do work received +from a bunch of asynch tasks through a single fixed sized fifo queue, and +send the results back through another fifo queue, with threads waking up +when the queue gets more stuff in it, and going to sleep when the queue +empties, with the last thread signalling “wake me up when there is +something to do”, and pushback happening when buffer is full. + +Go demonstrates that you can cleanly express algorithms as concurrent +communicating processes using fixed size channels. An unbuffered +channel is just a coprocess, with a single thread of execution switching +between the two coprocesses, without any need for locks or atomics, but +with a need for stack fixups. But Node.js seems to get by fine with code +continuations instead of Go’s stack fixups. + +A buffered channel is just a fixed size block of memory with alignment, +size, and atomic wrapping read and write pointers. + +Why do they need to be atomic? + +So that the read thread can acquire the write pointer to see how much data +is available, and release the read pointer so that the write thread can +acquire the read pointer to see how much space is available, and +conversely the write thread acquires the read pointer and releases the write +pointer.And when write thread updates the write pointer it updates it *after* +writing the data and does a release on the write pointer atomic, so that +when the read thread does an acquire on the write pointer, all the data that +the write pointer says was written will actually be there in the memory that +read thread is looking at. + +Multiple routines can send data into a single channel, and, with select, a +single channel can receive data from any channels. + +But, with go style programming, you are apt to have far more routines +than actual hardware threads servicing them, so you are still going to need +to sleep your threads, making atomic channels an optimization of limited +value. + +Your input buffer is empty. If you have one thread handling the one +process for that input stream, going to have to sleep it. But this is costly. +Better to have continuations that get executed when data is available in the +channel, which means your channels are all piping to one thread, that then +calls the appropriate code continuation. So how is one thread going to do a +select on a thousand channels? + +Well, we have a channel full of channels that need to be serviced. And +when that channel empties, mutex. + +Trouble is, I have not figured out how to have a thread wait on multiple +channels. The C++ wait function does not implement a select. Well, it +does, but you need a condition statement that looks over all the possible +wake conditions. And it looks like all those wake conditions have to be on +a single mutex, on which there is likely to be a lot of contention. + +It seems that every thread grabs the lock, modifies the data protected by +the lock, performs waits on potentially many condition variables all using +the same lock and protected by the same lock, condition variables that +look at conditions protected by the lock, then releases the lock +immediately after firing the notify. + +But it could happen that if we try to avoid unnecessarily grabbing the +mutex, one thread sees the other thread awake, just when it is going to +sleep, so I fear I have missed a spin lock somewhere in this story. + +If we want to avoid unnecessary resort to mutex, we have to spin lock on a +state machine that governs entry into mutex resolution. Each thread makes +its decision based on the current state of channel and state machine, an +does a `Atomic_compare_exchange_weak_explicit` to amend the state of the +state machine. If the state machine has not changed, the decision goes +through. If the state machine was changed, presumably by the other thread, +it re-evaluates its decision and tries again. + +Condition variables are designed to support the case where you have one +thread or a potentially vast pool of threads waiting for work, but are not +really designed to address the case where one thread is waiting for work +from a potentially vast pool of threads, and I rather think I will have to +handcraft a handler for this case from atomics and, ugh, dangerous spin +loops implemented in atomics. + +A zero capacity Go channel sort of corresponds to a C++ binary +semaphore. A finite and small Go channel sort of corresponds to C++ +finite and small semaphore. Maybe the solution is semaphores, rather than +atomic variables. But I am just not seeing a match. + +I notice that notifications seems to be built out of a critical section, with +lots of grabbing a mutex and releasing a mutex, with far too much +grabbing a mutex and releasing a mutex. Under the hood, likely a too-clever +and complicated use of threads piling up on the same critical +section. So maybe we need some spin state atomic state machine system +that drops spinning threads to wait on a semaphore. Each thread on a +channel drops the most recent state channel after reading, and most recent +state after writing, onto an atomic variable. + +But the most general case is many to many, with many processes doing a +select on many channels. We want a thread to sleep if all the channels on +which it is doing a select are blocked on the operation it wants to do, and +we want processes waiting on a channel to keep being woken up, one at a +time, as long a channel has stuff that processes are waiting on. + +# C++ Multithreading + +`std:aysnc` is designed to support the case where threads spawn more +threads if there is more work to do, and the pool of threads is not too large, +and threads terminate when they are out of work, or do the work +sequentially if doing it in parallel seems unlikely do yield benefits. C++ by +default manages the decision for you. + +Maybe the solution is to use threads where we need stack state, and +continuations serviced by a single thread where we expect to handle one +and only one reply. Node.js gets by fine on one thread and one database +connection. + +```C++ + #include &t;thread> +static_assert(__STDCPP_THREADS__==1, "Needs threads"); +// As thread resources have to be managed, need to be wrapped in +// RAII +class ThreadRAII { + std::thread & m_thread; +public: +// As a thread object is moveable but not copyable, the thread obj +// needs to be constructed inside the invocation of the ThreadRAII +// constructor. */ + ThreadRAII(std::thread & threadObj) : m_thread(threadObj){} + ~ThreadRAII(){ + // Check if thread is joinable then detach the thread + if(m_thread.joinable()){ + m_thread.detach(); + } + } + }; +``` + +Examples of thread construction + +```C++ + void foo(char *){ + … + } + + class foo_functor + { + public: + void operator()(char *){ + … + } + }; + + + int main(){ + ThreadRAII thread_one(std::thread (foo, "one")); + ThreadRAII thread_two( + std::thread ( + (foo_functor()), + "two" + ) + ); + const char three[]{"three"}; + ThreadRAII thread_lambda( + std::thread( + [three](){ + … + } + ) + ); + } +``` + +C++ has a bunch of threading facilities that are designed for the case that +a normal procedural program forks a bunch of tasks to do stuff in parallel, +and then when they are all done, merges the results with join or promise +and future, and then the main program does its thing. + +This is not so useful when the main program is a event oriented, rather +than procedural. + +If the main program is event oriented, then each thread has to stick around +for the duration, and has to have its own event queue, which C++ does not +directly provide. + +In this case threads communicate by posting events, and primitives that do +thread synchronization (promise, future, join) are not terribly useful. + +A thread grabs its event queue, using the mutex, pops out the next event, +releases the mutex, and does its thing. + +If the event queue is empty, then, without releasing it, the thread +processing events waits on a [condition variable](https://thispointer.com//c11-multithreading-part-7-condition-variables-explained/). (which wait releases the +mutex). When another thread grabs the event queue mutex and stuffs +something into into the event queue, it fires the [condition variable](https://thispointer.com//c11-multithreading-part-7-condition-variables-explained/), which +wakes up and restores the mutex of the thread that will process the event +queue. + +Mutexes need to construct RAII objects, one of which we will use in +constructing the condition object. diff --git a/docs/libraries/git_bash_undocumented.md b/docs/libraries/git_bash_undocumented.md new file mode 100644 index 0000000..50e7134 --- /dev/null +++ b/docs/libraries/git_bash_undocumented.md @@ -0,0 +1,52 @@ +--- +title: Git Bash undocumented command line +--- + +git-bash is a `mintty.exe` wrapper and bash wrapper – it winds up invoking +other processes that do the actual work. While git-bash.exe is undocumented, `mintty.exe` and [`bash.exe`](https://www.gnu.org/software/bash/manual/bash.html) [are documented](http://www.gnu.org/gethelp/). + +`git-bash.exe` sets up the environment in windows for `bash.exe`, then launches the bash shell + +Example Windows shortcut to bash script: `/x/src/wallet/docs/mkdocs.sh` + + "C:\Program Files\Git\git-bash.exe" --cd=X:\src\wallet --needs-console --no-hide --command=usr\bin\bash.exe --login -i docs/mkdocs.sh + +Notice that the paths to the left of the invocation of `bash` are in Windows +format, and the paths to the right of the invocation of bash are in gnu +format. + +Albeit this way of executing a bash script in windows is too clever by half, +since you should be able to execute it just by clicking on it. + +`--cd=D:\src` +Sets the initial working directory to `/d/src` (windows path, launches bash +with the corresponding gnu path) + +`--no-cd` +does not set working directory. + +`--cd-to-home` +Sets the working directory to home. + +`--command=`command-line +Executes `` instead of the embedded string resource. + +`--minimal-search-path` +Ensures that only `/cmd/` is added to the `PATH` instead of `/mingw??/bin` and `/usr/bin/` + +`--no-minimal-search-path` +Normal search path + +`--needs-console` +Ensures that there is a Win32 console associated with the spawned process + +`--no-needs-console` +Fails to ensure that there is a Win32 console + +`--hide` +Hides the console window. This makes sense if you are launching a script and +not expecting any feedback. But it means that the script has no means to +give you an error message. + +`--no-hide` +Does not hide the console window. diff --git a/docs/libraries/pandoc_templates/after.pandoc b/docs/libraries/pandoc_templates/after.pandoc new file mode 100644 index 0000000..c902203 --- /dev/null +++ b/docs/libraries/pandoc_templates/after.pandoc @@ -0,0 +1 @@ +

Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 International License.

diff --git a/docs/libraries/pandoc_templates/before.pandoc b/docs/libraries/pandoc_templates/before.pandoc new file mode 100644 index 0000000..ccec396 --- /dev/null +++ b/docs/libraries/pandoc_templates/before.pandoc @@ -0,0 +1 @@ +

To Home page

diff --git a/docs/libraries/pandoc_templates/header.pandoc b/docs/libraries/pandoc_templates/header.pandoc new file mode 100644 index 0000000..b386f15 --- /dev/null +++ b/docs/libraries/pandoc_templates/header.pandoc @@ -0,0 +1,21 @@ + + + \ No newline at end of file diff --git a/docs/libraries/pandoc_templates/style.css b/docs/libraries/pandoc_templates/style.css new file mode 100644 index 0000000..d66feba --- /dev/null +++ b/docs/libraries/pandoc_templates/style.css @@ -0,0 +1,31 @@ +body { + max-width: 30em; + margin-left: 1em; + } +p.center {text-align:center; + } + +table { + border-collapse: collapse; + } +td, th { + border: 1px solid #999; + padding: 0.5rem; + text-align: left; + } +code{white-space: pre-wrap; + } +span.smallcaps{font-variant: small-caps; + } +span.underline{text-decoration: underline; + } +div.column{display: inline-block; vertical-align: top; width: 50%; + } +div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em; + } +ul.task-list{list-style: none; + } +.display.math{display: block; text-align: center; margin: 0.5rem auto; + } +h1.title{text-align: center; font-size: xxx-large; + } \ No newline at end of file diff --git a/docs/libraries/review_of_crypto_libraries.md b/docs/libraries/review_of_crypto_libraries.md new file mode 100644 index 0000000..066955d --- /dev/null +++ b/docs/libraries/review_of_crypto_libraries.md @@ -0,0 +1,74 @@ +--- +title: Review of Cryptographic libraries +--- + +# Noise Protocol Framework + +[Noise](https://noiseprotocol.org/) is an architecture and a design document, + not source code. Example source code exists for it, though the [C example] +(https://github.com/rweather/noise-c) uses a build architecture that may not +fit with what you want, and uses protobuf, while you want to use Cap’n +Proto or roll your own serialization. It also is designed to use several +different implementations of the core crypto protocols, one of them being +libsodium, while you want a pure libsodium only version. It might be easier +to implement your own version, using the existing version as a guide. +Probably have to walk through the existing version. + +# [Libsodium](./building_and_using_libraries.html#instructions-for-libsodium) + +# I2P + +The [Invisible Internet Project](https://geti2p.net/en/about/intro) does a great deal of the chat capability that you want. You need to interface with their stuff, rather than duplicate it. In particular, your wallet identifiers need to be I2P identifiers, or have corresponding I2P identifiers, and your anonymized transactions should use the I2P network. + +They have a substitute for UDP, and a substitute for TCP, and your anonymized transactions are going to use that. + +# Amber + +[Amber](https://github.com/bernedogit/amber) + +Not as fast and efficient as libsodium, and further from Bernstein. Supports base 58, but [base58check](https://en.bitcoin.it/wiki/Base58Check_encoding#Base58_symbol_chart) is specifically bitcoin protocol, supporting run time typed checksummed cryptographically strong values. Note that any value you are displaying in base 58 form might as well be bitstreamed, for the nearest match between base 58 and base two is that 58^7^ is only very slightly larger than 2^41^, so you might as well use your prefix free encoding for the prefix. + +[Curve25519](https://github.com/msotoodeh/curve25519) + +Thirty two byte public key, thirty two byte private key. + +Key agreement is X25519 + +Signing is ED25519. Sixty four byte signature. + +Trouble is that amber does not include Bernstein’s assembly language optimizations. + +[ED25519/Donna](https://github.com/floodyberry/ed25519-donna) does include Bernstein’s assembly language optimizations, but is designed to compile against OpenSSL. Probably needs some customization to compile against Amber. Libsodium is designed to be uncontaminated by NSA. + +ED25519 does not directly support [Schnorr signatures](schnorr-signatures.pdf), being nonprime. Schnorr signatures can do multisig, useful for atomic exchanges between blockchains, which are multisig, or indeed arbitary algorithm sig. With some cleverness and care, they support atomic exchanges between independent block chains. + +explanation of how to do [Schnorr multisignatures](https://www.ietf.org/archive/id/draft-ford-cfrg-cosi-00.txt) [using ED25519](https://crypto.stackexchange.com/questions/50448/schnorr-signatures-multisignature-support#50450) + +Amber library packages all these in what is allegedly easy to incorporate form, but does not have Schnorr multisignatures.  + +[Bernstein paper](https://ed25519.cr.yp.to/software.html).  + +The fastest library I can find for pairing based crypto is [herumi](https://github.com/herumi/mcl).  + +How does this compare to [Curve25519](https://github.com/bernedogit/amber)?  + +There is a good discussion of the performance tradeoff for crypto and IOT in [this Internet Draft](https://datatracker.ietf.org/doc/draft-ietf-lwig-crypto-sensors/), currently in IETF last call:  + +From the abstract:.  + +> This memo describes challenges associated with securing resource- +> constrained smart object devices. The memo describes a possible +> deployment model where resource-constrained devices sign message +> objects, discusses the availability of cryptographic libraries for +> small devices and presents some preliminary experiences with those +> libraries for message signing on small devices. Lastly, the memo +> discusses trade-offs involving different types of security +> approaches. + +The draft contains measurement and evaluations of libraries, allegedly +including herumi.  But I don’t see any references to the Herumi library in +that document, nor any evaluations of the time required for pairing based +cryptography in that document. Relic-Toolkit is not Herumi and is supposedly +markedly slower than Herumi.  + +Looks like I will have to compile the libraries myself and run tests on them.  \ No newline at end of file diff --git a/docs/libraries/schnorr-signatures.pdf b/docs/libraries/schnorr-signatures.pdf new file mode 100644 index 0000000..7d967b0 Binary files /dev/null and b/docs/libraries/schnorr-signatures.pdf differ diff --git a/docs/libraries/stdafx.cpp b/docs/libraries/stdafx.cpp new file mode 100644 index 0000000..94cbf6b --- /dev/null +++ b/docs/libraries/stdafx.cpp @@ -0,0 +1,8 @@ +// stdafx.cpp : source file that includes just the standard includes +// wxHello.pch will be the pre-compiled header +// stdafx.obj will contain the pre-compiled type information + +#include "stdafx.h" + +// TODO: reference any additional headers you need in STDAFX.H +// and not in this file diff --git a/docs/libraries/stdafx.h b/docs/libraries/stdafx.h new file mode 100644 index 0000000..07488b1 --- /dev/null +++ b/docs/libraries/stdafx.h @@ -0,0 +1,20 @@ +// stdafx.h : include file for standard system include files, +// or project specific include files that are used frequently, but +// are changed infrequently +// + +#pragma once +#define wxUSE_UNICODE 1 +#define wxUSE_UNICODE_WCHAR 1 +#include +#include +#include "app.h" + +#ifdef _DEBUG +#pragma comment(lib, "wxbase31ud.lib") +#else +#pragma comment(lib, "wxbase31u.lib") +#endif + + +// TODO: reference additional headers your program requires here diff --git a/docs/lightning_layer.md b/docs/lightning_layer.md new file mode 100644 index 0000000..a98896c --- /dev/null +++ b/docs/lightning_layer.md @@ -0,0 +1,595 @@ +--- +title: + Lightning Layer +--- +# This discussion of the lightning layer may well be obsoleted + +by the elegant cryptography of [Scriptless Scripts] using adaptive Schnorr +signatures and of [Anonymous Multi-Hop Locks]. + +Contingent payments can reveal a key to an alien blockchain on the bitcoin blockchain, and [zero knowledge contingent payments on the bitcoin chain] can reveal any arbitrarily complex secret that fulfils any arbitrarily complex condition. + +[Scriptless Scripts]:https://tlu.tarilabs.com/cryptography/scriptless-scripts/introduction-to-scriptless-scripts.html +"Introduction to Scriptless Scripts – Tari Labs University" + +[Anonymous Multi-Hop Locks]: anonymous_multihop_locks_lightning_network.pdf +"Anonymous Multi-Hop Locks for Blockchain Scalability and Interoperability" + +[zero knowledge contingent payments on the bitcoin chain]:https://bitcoincore.org/en/2016/02/26/zero-knowledge-contingent-payments-announcement/ +"The first successful Zero-Knowledge Contingent Bitcoin Payment" + +I need to understand the [Anonymous Multi-Hop Locks] primitive, and +rewrite this accordingly. + +Scriptless scripts have the huge advantage of being self enforcing – it is +impossible to defect from the script, because there is no script – there are +just actions that are cryptographically possible or impossible. + +But scriptless scripts cannot in themselves solve the hard problem, that all +participants in a multilateral transaction need to know _in a short time_ that +the whole multilateral transaction has definitely succeeded or definitely +failed. This inherently requires a reliable broadcast channel, though if +everyone is cooperating, they don’t have to actually put anything on that +channel. But they need the capability to resort to that channel if something +funny happens, and that capability has to be used or lost within a time limit. + +So, the adapter secret not only has to become known to the participants, it has to become known to the participants within timing limits. + +[Anonymous Multi-Hop Locks] can ensure that the lightning network is +always in a definite state, and that those parts of it that are undergoing +state change are locked, but need to be embedded in a protocol that +ensures that those locks always go away in a short time with commitment +of the total transaction, or rollback of the total transaction, and that the +participants in the transaction always know whether the transaction was +committed or rolled back within a short time. Scriptless scripts are +timeless, and need to be embedded in scripted scripts that have timing +constraints, and which require information to be broadcast over a reliable +broadcast channel if the information is available, yet certain time limits are +nonetheless exceeded. + +My conclusion was that full circle unbreakability of lightning network +transactions within time limits needs a reliable broadcast, and I envisaged + a hierarchy of reliable broadcasters, (sidechains, with some sidechains + representing a group of bilateral lightning network gateways that act as + one multilateral lightning network gateway) But this conclusion may be + wrong or overly simple – though we are still going to need sidechains and + hierarchical reliable broadcasting, because it can do no end of things that + are very difficult otherwise. + + But reliable broadcast mechanism both supplies and requires a solution to + distributed Byzantine fault tolerant consensus, so the problem of getting a + lock up and bringing it down is a general distributed Byzantine fault + tolerant consensus problem, and perhaps viewing it as a reliable broadcast + problem is a misperception and misanalysis. + + Rather, the blockdag requires a mechanism to establish a total order of + blocks, and the distributed multihop lock requires a mechanism to + establish the state of the lock, the classic distributed state machine + problem addressed by Practical Byzantine Fault Tolerant distributed + consensus. Albeit implementing Practical Byzantine Fault Tolerant + distributed consensus as a state machine over a blockdag may well be a + simpler and more humanly intelligible form of this algorithm. But a state + machine for BEGIN … COMMIT is going to be a fundamentally different + state machine to the one that constructs a total order of transactions. + +# Lightning layer concept + +The lightning layer is a design that implements something like full reserve +correspondence banking on top of the blockchain layer, thereby solving +the problem of the blockchain exposing too much information about +everyone’s transactions to everyone, allowing a blockchain solution to +scale to completely replacing fiat currency, and allowing instant on the +spot transactions. + +A lightning gateway allows instant trustless unilateral transactions (Alice +pays Bob) between two or more people, but you have to lock up a +significant amount of money in each gateway, and if you have more than +two participants, and one of the participants goes off the internet, then +the protocols necessary for the gateway to keep going between the remaining +participants become a bit more complicated. Multiparty gateways will be +(eventually) implemented as sidechains. + +A lightning gateway consists of coin (unspent transaction output) on the +blockchain that need a joint signature to be spent. The parties to coin +construct a transaction splitting it up between them before they create that +coin on the blockchain, and then do not drop that transaction to the +blockchain layer, keeping the transaction for when one them decides to break +up the gateway. To change ownership of the coin, to increase the amount of the +coin owned by one of them and decrease the amount owned by the other, they +generate a new transaction, and then do not drop that transaction to the +blockchain layer either. + +And the problem is that we don’t want unilateral transactions. We want Alice +pays Bob, and Bob gets a proof that he paid Alice, completing the circle. + +And, to work like correspondence banking, we want Alice pays Bob, Bob pays +Carol, Carol pays Dan, Dan pays Erin, Erin pays Frank, and Frank gets proof +that he paid Ann, completing the circle. + +And we want every unilateral transaction in the circle to go through, or none +of them to go through. We want the whole circle of unilateral transactions +making up a complete multilateral transaction to succeed or every unilateral +transaction in the whole circle to fail. + +And we want it to normally and usually succeed fast, and if it fails, to take +a reasonably short time to fail. And if it fails, we want all participants to +know it has definitely failed after a reasonably short timeout. + +# The Bitcoin lightning layer is not + +It is not correspondence banking with predictable behavior enforced by +cryptography. + +It is correspondence banking that uses cryptography, where the participants +trustworthy behaviour is enforced from behind the scenes by unclear authority, +which is likely to lead to the existing problem where authority is supposed to +protect you from the bankers, but who protects you from authority? + +For the lightning network to change from a single consistent state of who owns +what part of each gateway unspent transaction output to another single +consistent state is equivalent to the well known hard problems of a reliable +broadcast channel, the Byzantine generals problem, the two generals problem, +and acid transactions on a distributed database. + +So when I looked at the lightning documents for a lightning layer on top of +bitcoin I expected to see discussion of smart contracts on the bitcoin layer +or lightning network protocols on the lightning layer for resolving these hard +problems and eventually reaching a resolution, so that the network would +always eventually reach a consistent and intended state. Not seeing them. + +Instead what I do see is that the lightning network is safe because +“misconduct” will result in you losing your bitcoins. + +It will? Who do you lose them to? How is *misconduct* defined? Who decides +*misconduct*? The guy who gets your bitcoins? Do we have a central banker of +the lightning network somewhere? + +We should not be talking about “misconduct”. We should be talking about the +lighting network entering an inconsistent state due to lost messages, nodes +going down at inopportune moments, and nodes deviating from the protocol, or +entering a state that some of the participants in the transaction did not +intend or expect. + +The cryptographic term for misconduct is “Byzantine failure”, which indeed +normally results from wickedness, but can result from bugs or data corruption. + +While the usual and archetypal cause of Byzantine failure is that someone +wrote or modified software for the purposes of betrayal, lying, cheating, and +stealing, it happens often enough as a result of data corruption or running a +program that was compiled under a compiler and in an environment different +from that it was tested and developed on. + +A gateway does a unilateral transaction between the parties on the lightning +layer who control controlling the gateway unspent transaction output by +generating a new transaction on the bitcoin layer breaking up the gateway +between the parties, and the parties refrain from committing that transaction +to the blockchain, and instead endless generate new transactions, which do not +get played either, thereby changing what part of the gateway unspent +transaction output is owned by each party to the gateway. + +*What happens if someone commits an out of date bitcoin layer transaction to +the blockchain?* + +Does the central banker confiscate the bitcoin of the party who committed the +transaction. How does he do that? + +Suppose Alice’s gateway to Bob is blocked because she now owns the whole of +that gateway unspent transaction output, and her gateway to Frankie is blocked +because Frankie owns all of that gateway. + +So she organizes a transaction that moves bitcoin from the Ann/Bob gateway to +the Ann/Frankie gateway. So what Alice intends is that Alice pays Bob, Bob +pays Carol, Carol pays Dan, Dan pays Erin, Erin pays Frank, and Frank pays +Alice. Except that in the middle the transaction Carol and Erin ungracefully +disconnect from the network, so that either Ann generated a bitcoin layer +transaction giving bitcoin to Bob, but Frankie did not generate a bitcoin +layer transaction to Ann, or the other way around. + +*What happens when a transaction fails and leaves the parties in an inconsistent +state?* + +Does the central banker decide that Carol and Erin were engaged in +misconduct and confiscate their bitcoin?* + +# Trustless Unilateral transactions + +If you wanted to buy or sell cryptocurrency for cash or gold, you could +arrange over the internet to meet someone in person, set up a lightning +gateway with him, and then, when you met him in person, pay him instantly, on +the spot, in a trustless unilateral transaction without need to wait some +considerable time for the payment to clear. + +Suppose you wanted to sell bitcoin for gold, in person, both parties are going +to meet, probably with a weapon in their pocket. You could create a jointly +signed unspent transaction output, after you and the other party jointly sign +a bitcoin layer transaction giving the bitcoin back to you. And then, when +meeting, create a jointly signed bitcoin layer transaction giving the bitcoin +to the man who is giving you the gold. Except what is going to stop you from +committing the earlier transaction to the blockchain a few seconds before the +meeting? + +OK, let us suppose we have supersedeable transactions. They have to be +committed to the blockchain for a time before they take effect, and if someone +submits a transaction with the same signatures but a higher priority, the one +with lower priority fails to take effect. Then you can endlessly generate +fresh transactions, each with higher priority than the previous one, and never + commit them to the blockchain unless the gateway between the parties is +abandoned. + +It would take a little while for the gateway to become available, but once it +was available, instant irrevocable payments become possible. + +And if one has an account with a service provider over the internet, one could + set up a gate way with that service provider, and after each session or each + service, make a small instant payment, without the cost and delay of making +transactions on the blockchain. + +It would be possible to do such instant two party transactions with bitcoin +today, although the wallets are generally not set up to support it, nor is the +way the main blockchain processes transactions, but with BTC blockchain +working as it today, such transactions are not trustless. If you want to do a + BTC transaction you are trusting the lightning network, which means you + are trusting you do not whom. + +At the time this is written, the lightning network for bitcoin has not been +adequately implemented, or even fully designed. Existing implementations on +top of the bitcoin blockchain still require some trust in intermediaries, and +thus require trust in some mysterious authority with the mysterious +capability to punish the intermediaries, and to have a real lighting network on +bitcoin requires changes in bitcoin which the miners are not going along with, +and which perhaps are underscoped and not fully thought out, ad hoc changes +to fit with what already existed, and what was politically feasible. And it +seems that rather less was politically feasible than one might hope. + +# Cryptographic implementation of trustless unilateral transactions + +Ann and Bob create on the blockchain a coin (unspent transaction output, +utxo) whose corresponding key is the sum of secret key known only to +Ann, and a secret key known only to Bob. Before they create this coin, +they create a transaction on it, signed by their joint key, that creates two +coins, one with a secret key known only to Bob, and one with a secret key +known only to Ann. They keep this transaction to themselves and do not +place it on the blockchain. + +# Multilateral (circle) transactions + +The lightning layer will function like correspondence banking, only with +good behavior cryptographically enforced on the lightning vertices, rather +than by state supervision of the “banks”. This will require a blockchain +layer designed to support it. + +Correspondence banking merges large numbers of small transactions into a small +number of large pooled transactions which are eventually settled on the +blockchain in one big transaction, with several parties to the transaction +representing a very large number of parties engaged in a very large number of +transactions. + +But correspondence banking works by trust, thus the intermediaries have to +be well known and subject to pressure, which is apt to mean subject to +government pressure – and government has interests that are in conflict with +those of people attempting to use a medium of exchange and a store of +value. + + lightning network correspondence banking + ------------------------------------------------------------------------ ----------------------------------------------------------------------------------- + merges many small two party transactions into a few large transactions merges many small two party transactions into a few large transactions + lightning vertex bank + lightning gateway bank account + trustless need to trust the banks + instant Slow in that you can never be sure if a transaction will be honored for some time + Payer id visible to payer vertex, payee id visible to payee vertex Government issued id of payer and payee visible to all intermediaries. + +We have a separate blockchain of supersedeable transactions. A +transaction gets committed to the primary blockchain after it has been +sitting on the supersedable chain for a while, if, at the time it is +evaluated for commitment, all of its inputs are valid unspent outputs on +the primary blockchain, and none of them are inputs to a currently valid +higher priority transaction on the blockchain of supersedable +transactions, the priority being an arbitrary length binary value. + +If one party tries to break up a gateway with an out of date +distribution, the other party notices it sitting on the supersedable +blockchain, and issues a more up to date transaction. Normally this +should never happen, since when one of the parties wants to break up a +gateway, the other party should agree to a regular transaction. However, +one party may go offline and stay offline, especially if the last +transaction reduced the the value of their interest in the gateway to +zero. + +A gateway in the lightning network layer is a jointly signed coin on the +blockchain on the blockchain layer. Bob’s account with BigCrypto is a coin +on the blockchain layer, for which there exists a a jointly signed +transaction distributing that block between two blocks, one wholly owned by +Bob, and one wholly owned by BigCrypto, thus giving BigCrypto effective +offchain control of one part of that block, and Bob effective control of the +other part of the block, and by generating new superseding transaction, they +can effectively transfer ownership of part of the block instantly without +anything going on the blockchain. + +But … + +# Acid + +We have to make sure that transactions are acid on the lightning network as a +whole, that transactions are atomic, consistent, isolated, and durable. + +## Atomic and consistent + +Ann has an account with Bob, Bob has an account with Carol. + +To change what is Ann’s account requires a transaction signed by +Bob and Ann, and similarly for Carol. + +Ann wants to pay Carol, but does not want to sign a reduction in her account +with Bob, unless she is sure that Carol gets the corresponding increase. Bob +does not want to sign an increase in Carol’s account, unless it gets Ann’s +signature on a decrease in her account. Not to mention that Ann probably +does not want to sign a decrease on her account without getting a receipt +from Carol. Full circle transaction. We need to guarantee that either the +full circle goes through, or none of the separate unilateral transactions in +the circle go through. + +## Reliable broadcast channel + +The solution to atomicity and maintaining consistency between different +entities on the lightning network is the reliable broadcast channel. + +Such as the blockchain itself. Create a special zero value transaction that +has no outputs and carries its own signature, but can be a required input to +other transactions, and whose construction requires the cooperation of all +the parties. Each gateway constructs a transaction only be valid if a code +is placed on the blockchain that requires the cooperation of all the gateways +within a short time. Once the code exists, and everyone knows it exists, +they proceed with bilateral transactions that do not require the code and +only potentially go on the blockchain. If not everyone knows it exists, and +it does not appear on the blockchain within a short time, then the +transaction fails. If everyone knows it exists, the transaction succeeds. If +not everyone knows it exists, but it appears on the blockchain within the +time limit, the transaction succeeds, and each party could potentially play +the transaction, and thus effectively owns the corresponding part of the +gateway coin, regardless of whether they play it or not. + +A reliable broadcast channel is something that somehow works like a +classified ad did back in the days of ink on paper newspapers. The physical +process of producing the newspaper guaranteed that every single copy had the +exact same classified ad in it, and that ad must have been made public on a +certain date. Easy to do this with a printing press that puts ink on +paper. Very hard to do this, with electronic point to point communications. + +But let us assume we somehow have a reliable broadcast channel: + +All the parties agree on a Merkle tree, which binds them if the joint +signature to that Merkle tree appears on the reliable broadcast channel +within a certain short time period. + +And, if some of them have the joint signature, then knowing that they could +upload it to the reliable broadcast channel, they each agree to superseding +unilateral transactions. If Bob expects payment from Ann and expects to +make payment to Carol, and he has the joint signature, and knows Carol has a +copy of the authenticated joint signature, because Carol sent him the +signature and he sent Ann the signature, of it, then he knows Carol can +*make* him pay her, and knows he can *make* Ann pay him. So he just goes +right ahead with unilateral transactions that supersede the transaction that +relies on the reliable broadcast channel. And if every party to the +transaction does that, none of them actually broadcast the signature the +reliable broadcast channel. Which in consequence, by merely being available +enforces correct behaviour, and is seldom likely to need to actually +broadcast anything. And when something is actually broadcast on that +channel, chances are that all the transactions that that broadcast enables +will have been superseded. + +Each party, when receives a copy of the joint signature that he *could* upload +to the reliable broadcast channel, sends a copy to the counter party that he +expects to pay him, and each party, when he receives a copy from the party he +expects to pay, performs the unilateral payment to that party that supersedes +and the transaction using the reliable broadcast network. + +And if a party has a copy of the joint signature and the document that it +signs for the full circle transaction, but finds himself unable to perform +the superseding unilateral transactions with his counterparties, (perhaps +their internet connection or their computer went down) then he uploads the +signature to the reliable broadcast channel. + +When the signature is uploaded to reliable broadcast channel, this does not +give the reliable broadcast channel any substantial information about the +amount of the transaction, and who the parties to the transaction are, but the +node of the channel sees IP addresses, and this could frequently be used to +reconstruct a pretty good guess about who is transacting with whom and why. +As we see with Monaro, a partial information leak can be put together with +lots of other sources of information to reconstruct a very large information +leak. + +But most of the time, the channel is not likely to be used, which means it +will have only small fragments of data, not enough to put together to form +a meaningful picture, hence the privacy leak is unlikely to be very useful +to those snooping on other people’s business. + +### Other use cases for a reliable broadcast channel + +The use case of joint signatures implies an immutable data structure of the +tuple oid, hash, public key, and two scalars. + +But another use case is to publicly root private immutable data. + +If you continually upload the latest version, you wind up uploading most of +tree, or all of it, which does not add significantly to the cost of each +interaction recorded. The simplest sql friendly data structure is (oid of +this item, public key, hash, your index of hash, oids of two child hashes) +with the reliable broadcast channel checking that the child hashes do in fact +generate the hash, and that the tuple (public key, index of hash) is unique. + +If the data is aged out after say, three months, cannot directly check +uniqueness and correctness for the nodes that are the roots of big trees. How +do you know someone has not made up several different and mutually +inconsistent pasts for immutable append only data structure associated with +this key? + +To work around this problem, allow unbalanced Merkle trees, consisting of +(oid of this item, public key, hash, your tree node index, index of the +highest index leaf governed by this hash, oids of two child hashes) If an +unbalanced node referencing an old tree root is uploaded at intervals of less +than three months, it can be used to prove the contents and uniqueness of +the old balanced binary tree root, since the most recent unbalanced node must +have also proved contents and uniqueness, using a less recent unbalanced +node, and unbalanced nodes must also be unique on the tuple (public key, +index of node, index of highest leaf node governed) Someone can forget his +old past, and, after three months, start making up a new past, but the +witness for the new past can only begin on the day he starts the upload. He +cannot construct several realities, and six months later, choose which +reality he finds convenient. Or rather he can, but he cannot provide a six +month earlier witness to it. + +You upload several nodes that constitute the unbalanced tree right side path +that points at the balanced rest of the old tree every two months, superseding +the previous right hand side of the unbalanced tree, and thus maintaining a +chain of proofs stretching into the past that proves that throughout this +period, there is one and only one immutable data structure associated with +this public key. + +A common case is that they key certifying the state of the immutable data may +change, with the torch being passed from key to key. In which case the +upload of total state needs to reference the used key, and an earlier, +normally the earliest, signing key, with links in the chain of keys +authorizing keys being renewed at less than the timeout interval for data to +be immutable, but unavailable from the reliable broadcast network. If the +witness asserts that key is authorized by a chain of keys going back to an +earlier or the earliest keys, then it relies on its previous witness, rather +than re-evaluating the entire, possibly very long, chain of keys every time. + +But, if the witness cannot do that, then the entire, possibly very very long, +chain of keys and signatures has be uploaded for the witness to record the +earlier, or earliest key, as authorizing the current ever changing key. +Similarly if the client starts uploading to a new witness. But such very +long proofs will only be have to done once in a very long while, and done +once for everyone. + +Because of Byzantine failure or network failure, such a chain may fork. The +protocol has to be such that if a fork develops by network failure, +it will be fixed, with one of the forks dying when the network functions +better, and if it fails by Byzantine failure, +we get two sets of reliable broadcast channels, +each testifying that the other reliable broadcast channel is unreliable, +and each testifying that a different fork is the valid fork, +and which fork you follow depends on which reliable broadcast channel you +subscribe to. + +Another use case is for wallet recovery, with mutable data structures +encrypted by the private key whose primary key is the public key. + +## implementing and funding a reliable broadcast channel + +Tardigrade has a somewhat similar architecture to the proposed Reliable +Broadcast network charges $120 per year for per TB of storage, $45 per +terabyte of download. So for uploading a single signature, and downloading +it six times, which one hash, one elliptic point, and two scalars, one +hundred and twenty eight bytes, so the cost of doing what tardigrade does +with reliable broadcast network operated by a single operator would be +$4 × 10^{-7}$ dollars. Which might as well be free, except we have to charge some tiny amount to prevent DDoS. + +But, when the system is operating at scale, will want the reliable broadcast +network to have many operators, who synchronize with each other so that the +data is available to each of them and all of them and from each of them and +all of them, and can testify when the +data became available, so the cost will be +many times that. Which is still insignificant. If the network is composed +of a hundred operators, and there are substantial overheads to maintain +synchrony and truth, we are still only looking a cost of $0.0001 per +transaction. Maybe we should charge for opening the account, and then +every hundredth transaction. + +We also want the operators to be genuinely independent and separate from each +other. We don’t want a single inherently authorized reliable broadcast channel, +because it is inherently a low cost target for the fifty one percent attack. +I have been thinking about implementing a reliable broadcast channel as +byzantine Paxos protocol, but this gives us a massive low cost fifty one +percent attack vulnerability. If the reliable broadcast channel is cheap +enough to be useful, it is cheap enough for the fifty one percent attack. +We want cheap testimony of valuable facts, which makes consensus mechanisms +unlikely to work. + +A better way reliably implementing a reliable +broadcast channel is as a network of trusted witnesses, each of which keeps an +eye on the reliability of other witnesses, because each periodically uploads +the unbalanced tree testifying to its total state on several of the others, +and each makes it data reciprocally available to several of the others, and +each monitors the availability of several of the others, and each provides a +signed chain of its total state with each witness, or some witnesses. Because +the data is reciprocally available, each can testify to the uptime and +reliability of each of the others, and none can hide its state. Each makes +each of the nodes in each of its balanced trees available by index, the top +node in each of its unbalanced right hand tree is signed, and nodes in the +unbalanced tree constitute a separate sql like table, which times out +considerably faster than the nodes with a single index. + +Downtime, failure to provide oids on request, and losing its recent state, +will be detected and will whack its reputation. + +And it is up to the individual human to decide which trusted witness to follow, +which human decision roots all the automatic computer decisions. + +## isolated + +Isolated means that one transaction cannot mangle another transaction. +Easiest way to do this is that one gateway of the lightning network cannot +handle a transaction while another is pending. Which is going to be a problem +when we have nodes handling a lot of traffic, but a more efficient fix can +wait till we have traffic problems. + +For a more efficient fix, we will need relative as well as absolute +transactions. A relative transaction, on being appended to an absolute +transaction, subtracts something from one of the outputs, and adds it to +another outputs. If this would produce a negative output, the compound +transaction is invalid. But this can wait until we have busy gateways. + +## durable + +Durable means that if something bad happens to your node, you can probably +recover, and if you don’t recover, no one else has problems. + +If one party to a gateway goes down and does not come up, possibly because he +has lost the secret key to the gateway, the other party drops the most recent +transaction of the lightning layer to the blockchain layer. + +If one party to a gateway goes down, but eventually comes up again, possibly +with lost data, they resynchronize. + +Suppose you lose your wallet? + +At present the standard way of recovering a wallet is that you create a wallet +using your BIPS master password, and it scans the blockchain for coins whose +public key corresponds to the secret keys it might have generated. + +Which works to recover your money, but does not recover your metadata, and +will not recover a lightning gateway, since the your public key is not on the +blockchain, rather the sum of your public key, and the public key of other +parties to the gateway. + +Metadata is becoming more important. If you lose your Wasabi wallet metadata, +you have a big problem, and if we use the wallet as the basis for private end +to end encrypted messages that can carry money, and for uncensorable social +networking, the wallet will have a lot of very valuable metadata. + +We really need cloud backup of wallets, with the wallet encrypted by a secret +key derived from its master key, and cloud backup paid for though a lightning +gateway. And this needs to be built into the wallet, the way recovery by +scanning the blockchain is built into wallets today. + +# trustless, fast, and free from blockchain analysis + +That full circle transactions are atomic and consistent means that the +lightning network can operate without trust or reputation – completely +pseudonymous parties with no reputation can become nodes in the network, +making it very hard for the state to apply know your customer pressure. + +If the full content of each gateway’s ownership never becomes known +except to the parties to the gateway, then the effect is to mingle and +merge coins on the blockchain, creating fungibility and untraceability, +as well as instant transactions. + +If, on the other hand, we had a rather small number of trusted nodes that +have a special, central, and privileged position on the +lightning network, this would recreate the traceability problem of +correspondence banking. The lightning network has to trust the reliable +broadcast channels with reliability and broadcasting, but the information on +that channel is meaningless to anyone other than the parties to the +transaction. diff --git a/docs/merkle_patricia-dac.md b/docs/merkle_patricia-dac.md new file mode 100644 index 0000000..7201c5e --- /dev/null +++ b/docs/merkle_patricia-dac.md @@ -0,0 +1,1461 @@ +--- +title: + Merkle-patricia Dac +# katex +--- + +# Definition + +## Merkle-patricia Trees + +A Merkle-patricia tree is a way of hashing a map, an associative array, +such that the map can have stuff added to it, or removed from it, +without having to rehash the entire map, and such that one can prove a +subset of the map, such as a single mapping, is part of the whole map, +without needing to have to have the whole map present to construct the +hash. + +Its practical application is constructing a global consensus on what +public keys have the right to control what digital assets (such as +crypto currencies, and globally defined human readable names) and +proving that everyone who matters agrees on ownership. + +If a large group of peers, each peer acting on behalf of a large group +of clients each of whom have rights to a large number of digital assets, +agree on what public keys are entitled to control what digitial assets, +then presumably their clients also agree, or they would not use that +peer. + +Thus, for example, we don’t want the Certificate Authority to be able +to tell Bob that his public key is a public key whose corresponding +secret key is on his server, while at the same telling Carol that Bob’s +public key is a public key whose corresponding secret key is in fact +controlled by the secret police. + +The Merkle-patricia tree not only allows peers to form a consensus on an +enormous body of data, it allows clients to efficiently verify that any +quite small piece of data, any datum, is in accord with that consensus. + +## Patricia trees + +A patricia tree is a way of structuring a potentially very large list of +bitstrings sorted by bitstring, with a binary point implied somewhere in the +bistring, such that bitstrings can be added or deleted without resorting or +shifting the whole list of bitstrings. + +A patricia tree defines a set of keys, the keys of its leaf nodes. Or perhaps +the keys _are_ its leaf nodes, all the information may well be in the strings +defined. The keys may be bounded on the left, but unbounded on the right, +for example strings, unbounded on the left but bounded on the right, for example +arbitrary precision integers, bounded both left and right, for example sixty +four bit, integers, or unbounded on either size, for example binary arbitrary +precision floating point. + +If unbounded left, then the edge pointing at the root node had to give its +height above the binary point. The bitstring of the root node is always the +empty string, but where is that empty string positioned in height with respect +to the binary point? + +We cannot reference nodes by bitstring in the canonical form, because the +number of leading zeroes in bitstring will change over time as the tree gets +deeper – we have to represent nodes by their height plus the bitstring +starting at the first non zero bit, or by the key, which is the bitstring +with a one bit and several zero bits appended, to align the significance of +bits in different bitstrings, in which case again only the first non zero bit +matters. + +When we give a chain of vertexes, starting at the root vertexes, the compact +representation of the location of the vertex in the tree is its vertical +position in the tree, plus the bitstring starting at the first nonzero bit. If +it its a tree of items with fixed right bound, items identified by their +integer sequence, then we give the height above the leaves, since this will +not change as the tree grows. If fixed left bound, for example names as utf8 +strings, the depth from the root. If neither bound is fixed, a case we are +unlikely to have to deal with, the signed height or depth from some arbitrary +starting point which will never change as the tree grows. Thus we always +need to implicitly or explicitly define the bit alignment in the bitstring. + +If a leaf in a patricia tree representing values with fixed right boud, for +example oids, the usual case, then the bitstring of a leaf is its oid, or its +oid minus one, which does not need leading zeroes. + +We need to be able to represent a bitstring containing all zeroes, thus if a +bitstring contains any ones, we have to represent that one. The height, +therefore, is the alignment of the right edge of the bitstring, hence we can +leave out leading zeroes in the bitstring, and indeed must leave them out of +the canonical form of a tree representing values with no fixed left bound, +such as indefinite precision integers, so that the canonical form does not +change when a parent node is placed on top of a former root node + +The key to a node, whether a vertex or a leaf, is the bitstring aligned by +padding it with a one bit, followed by as many zero bits as needed. + +The total number of vertexes equals the twice the number of leaves +minus one. Each parent node has as its key the bit string, a sequence of +bits not necessarily aligned on bit boundaries, that both its children +have in common. This creates the substring problem for patricia trees +mapping keys that have no right bound, mapping variable length keys. We +cannot permit one key of the map to be the prefix of another key. If, +however, the key is is self delimiting, as with null terminated strings, +no key can be the prefix of another key, and this tends to be the usual +way that variable length values are used as map keys. There are wide +variety of too clever by half ways of dealing with prefix keys, but they +all involve messing up a rather elegant algorithm with considerable +complexity and surprising code paths in special cases and fencepost +cases. It is better just to not allow prefix keys, as for example by +having strings null terminated. If we really wanted to define arbitrary +bit strings as leaf keys of a patricia tree, which I doubt we will, +would be better to encode them in self delimiting format. No string in +self delimiting format can be the prefix of another string. + +A Merkle-patricia dac is a patricia tree with binary radix (which is +the usual way patricia trees are implemented) where the hash of each +node depends on the hash and the skip of its two children; Which means +that each node contains proof of the entire state of all its descendant +nodes. + +The skip of a branch is the bit string that differentiates its bit +string from its parent, with the first such bit excluded as it is +implied by being a left or right branch. This is often the empty +bitstring, which when mapped to a byte string for hashing purposes, maps +to the empty byte string. + +It would often be considerably faster and more efficient to hash the +full bitstring, rather than the skip, and that may sometimes be not +merely OK, but required, but often we want the hash to depend only on +the data, and be independent of the metadata, as when the leaf index is +an arbitrary precision integer representing the global order of a +transaction, that is going to be constructed at some later time and +determined by a different authority. + +Most of the time we will be using the tree to synchronize two ’t of +pending transactions, so though a count of the number of children of a vertex +or an edge is not logically part of a Merkle-patricia tree, it will make +synchronization considerably more efficient, since the peer that has the node +with fewer children wants information from the peer that has the node with +more children. + +# Representation + +The canonical form will not directly reflect the disk organization. + +The canonical form of a sparse tree is that each vertex is represented by the +hash of its two children, and the bitstring of the offset of each child from +its parent, minus the leading bit of that bitstring. The root node, of +course, has an empty bitstring. + +Often it will be more compact to transmit the child itself rather than the +hash, in reverse polish notation order, from which the hash can be generated. + +To form the hash of a node, we need the hashes and relative bitstrings of its +children, but if we already have the children, identified by reverse polish +position in the stream or by their bitstrings relative to a common ancestor, +we don’t need and should not represent the hashes, giving them implicitly +rather than explicitly. + +Since we cannot hash a bitstring, only a bytestring, the bitstring will be +hashed in its representation as a bytecount represented by a variable +precision integer, followed by that many bytes, with the bitstring being +padded if needed to an integer number of bytes by adding a one bit followed +by as many zero bits as needed. Thus a twofiftysix bit byte string requires +a count of thirty three, plus thirty three bytes, the last byte being 0x80. + +In memory as a ceeplusplus object, the bitstring may conveniently be +represented by an integer of at least sixty four bits, with the bitstring bit +aligned so that the significan bits in one bitstring line up with bits of the +same significance in another bitstring, and padded right with a one bit +followed by as many zero bits as needed. In the canonical form, however, the +left edge of the bitstring vertex identifier is the left edge of the +bitstring, and the length of the bitstring is the depth of the vertex from +the root. The left edge of a relative bitstring, identifying a child is one +bit to the right of the bitstring identifier of its parent. The child’s +bitstring vertex identifier is the parent bitstring vertex identifier, plus a +zero bit for the left child and a one bit for the right child, plus the bits +of the relative bitstring. The canonical hash of the parent is the hash of +its left child, plus the relative bitstring of its left child, and similarly +for its right child. + +Conceptually and canonically, it is equivalent to a patricia tree where +the children of a node are identified by hashes rather than pointers. +The hashes are taken over the canonical form, and are unaffected by +location in memory and the representation, which is not necessarily +canonical. If the actual representation is in a database, it is likely +to be represented in a way that makes recursive sql statements work. + +Since we in practice cannot find the thing referred to by its hash, any actual +representation of the canonical form must contain additional information +telling us where to find the data referred to, but this additional information +is likely to vary from one situation to the next, and is not canonical. + +For this canonical form to work as a direct representation, we would need a +universal way of finding the pre-image of any hash, which would be costly, and +would deny us some useful cryptographic capabilities where a party reveals a +pre-image and a hidden part of the Merkle-patricia tree – typically when a +transaction goes bad, he would then make public the convesation leading to it. +Also, the tree will eventually grow enormous, and have numerous side chains +attached to it, in which case only the party or parties operating the side +chain can reverse their pre-images.  + +But the forms actually used should be a representation of a Merkle +patricia tree with hashes and skip fields in place of pointers. + +## Balanced binary trees of fixed height. + +We will represent an immutable and ever growing data structure as a collection +of balanced binary trees, and a balanced binary tree of fixed height makes +much of the information in this representation redundant, which suggests that +it may be desirable to use a more efficient and direct canonical form – to +ensure that the immutable append only data structure is canonically immutable +and append only. + +The schere pointing at a balanced binary tree will say that it is a balanced +binary tree whose leaves are objects of a certain type (have a certain schema) +and give the height of the root, assuming that they all have the same schema. +If they have different schemas, the then the leaves will be of type schere. + +The patricia bit string for each vertex of the balanced binary tree is +implicitly given by its position within the tree, so we do not represent it in +the canonical form, though we may well represent it in the actual +representation. + +## Hashing + +Hashing depends on the schema – to hash the bitstream, has to parse +it into fields and records by the schema, and distinguish between index +nodes and record nodes, which are hashed and represented differently, a +record node being self contained, and index node depending on its +relationships. + +In a dac, rather than a tree, an index node might be referenced by +multiple different entities, so in that case we want the hash to only +depend on the part of the key field that it governs, independent of the +parent part of its key field. + +Further, a transaction is a group of records, and we want to represent a +transaction locally, so that its records are physically close together +in storage. + +Which implies a transformation, that the canonical form, which knows +nothing about storage location, can have portions represented as a +position relative form, in which a group of records is kept in depth +first tree order, with the boundaries of the group having hashes linking +them to the outside, but internally, when converting back into canonical +form, we recalculate the hashes. For a given schema we might do this one +way in one context, or another way in another context, or have +subschemas. + +Of course, tree order assumes we have a tree. In general, we have a dac, +not a tree, the most important case here being the tree of names, where +we are continually issuing new roots for the tree, but we don’t want to +continually issue new leaves. + +In the canonical form of the Merkle-patricia tree we act as if hashes +were reversible. Of course they are not, nor do we have a general +universal look up table for reversing them. Rather, you have to hit up a +server that can reverse the hashes you care about, which it may do by +looking up a ginormous hash table, or more likely do by having the +Merkle-patricia tree on disk or in memory in the ordinary patricial form +of links pointing at file relative or absolute locations on disk or in +memory. + +# Blockchain + +Of course we want more than this – a Merkle-patricia block chain, +meaning *an immutable past and a constantly changing present*. + +Which represents an immutable and ever growing sequence of transactions, +and also a large and mutable present state of the present database that +is the result of those transactions, the database of unspent transaction +outputs. + +When we are assembling a new block, the records live in memory as native +format C++ objects. Upon a new block being finalized, they get written +to disk in key order, with implementation dependent offsets between +records and implementation dependent compression, which compression +likely reflects canonical form. Once written to disk, they are accessed +by native format records in memory, which access by bringing disk +records into memory in native format, but the least recently loaded +entry, or least recetly used entry, gets discarded. Even when we are +operating at larger scale than visa, a block representing five minutes +of transactions fits easily in memory. + +Further, a patricia tree is a tree. But we want, when we have the Merkle +patricia tree representing registered names organized by names or the +Merkle-patricia tree represenging as yet unspent transaction outputs its +Merkle characteristic to represent a directed acyclic graph. If two +branches have the same hash, despite being at different positions and +depths in the tree, all their children will be identical. And we want to +take advantage of this in that block chain will be directed acyclic +graph, each block being a tree representing the state of the system at +that block commitment, but that tree points back into previous block +commitments for those parts of the state of the system that have not +changed. So the hash of the node in such a tree will identify, probably +through an OID, a record of the block it was a originally constructed +for, and its index in that tree. + +A Merkle-patricia directed acyclic graph, Merkle-patricia dac, is a +Merkle dac, like a git repository or the block chain, with the patricia +key representing the path of hashes, and acting as index through that +chain of hashes to find the data that you want.  + +The key will thread through different computers under the control of +different people, thus providing a system of witness that the current +global consensus hash accurately reflects past global consensus hashes, +and that each entities version of the past agree with the version it +previously espoused. + +This introduces some complications when a portion of the tree represents +a database table with more than one index.  + +[Ethereum has a discussion and +definition](https://github.com/ethereum/wiki/wiki/Patricia-Tree) of this +data structure. + +Suppose, when the system is at scale, we have thousand trillion entries +in the public, readily accessible, and massively replicated part of the +blockchain. (I intend that every man and his dog will also have a +sidechain, every individual, every business. The individual will +normally not have his side chain publicly available, but in the event of +a dispute, may make a portion of it visible, so that certain of his +payments, an the invoice they were payments for, become visible to +others.) + +In that case, a new transaction output is typically going to require +forty thirty two byte hashes, taking up about two kilobytes in total on +any one peer. And a single person to person payment is typicaly going to +take ten transaction outputs or so, taking twenty kilobytes in total on +any one peer. And this is going to be massively replicated by a few +hundred peers, taking about four megabytes in total. + +(A single transaction will typically be much larger than this, because +it will mingle several person to person payments. + +Right now you can get system with sixty four terabytes of hard disk, +thirty two gigabytes of ram, under six thousand, for south of a hundred +dollars per terabyte, so storing everything forever is going to cost +about a twentieth of a cent per person to person payment.  And a single +such machine will be good to hold the whole blockchain for the first few +trillion person to person payments, good enough to handle paypal volumes +for a year. + +“OK”, I hear you say. “And after the first few trillion transactions?”. + +Well then, if we have a few trillion transactions a year, and only a few +hundred peers, then the clients of any one peer will be doing about ten +billion transactions a year. If he profits half a cent per transaction, +he is making about fifty million a year. He can buy a few more sixty +four terabyte computers every year. + +The target peer machine we will write for will have thirty two gigabytes +of ram and sixty four terabytes of hard disk, but our software should +run fine on a small peer machine, four gigabytes of ram and two +terabytes of hard disk, until the crypto currency surpasses bitcoin. + +Because we will employ fixed size transaction units – larger currency +amounts will be broken into tens, twenties, fifties, hundreds, two +hundreds, five hundreds, thousands, two thousands and so forth, and +because we will be using a blockchain in the form of a Merkle-patricia +dac, our transactions will tak up several times as much space a similar +bitcoin transaction, and currently bitcoin transactions take up several +hundred megabytes. But this is OK, because the Merkle-patricia dac gives +client wallets far more power than on the bitcoin system, so we can get +by with far fewer peer wallets and far more client wallets. + +------------------------------------------------------------------------ + +# Packed Form + +Assume we have an ordered sequence of records, as if the result of a +database query with the index fields first. + +Suppose we have a potentially very large sequence of records, and we +want to generate the Merkle-patricia hash, so that we can generate an +efficient proof that one and only one record with a certain prefix +appears in this pile. + +We want to convert from sequence of records form to patricia form. + +If we look at the difference between each pair of records, we get an +index node, which is the position of the first bit at which they differ. +Which is a patricia tree expressed in infix order. We want to convert +infix order to postfix order, a sequence of records interleaved with a +sequence of bit positions at which two records differ. + +Or equivalently, hash them as if we already had them in postfix order.  + +Suppose we want to output self delimiting records, interleaved with +postfix indicators of the difference position. We want the least index +output last, so that when we do a second sequential pass, hashing each +record when we encounter a record, and putting that has on the stack, +and when we encounter and index, hashing two hashes from the stack with +the index, and put the resulting hash on the stack. + +So, we find the difference position between the current record and the +next, then we output the current record, and make the next record the +current record. If the difference position is less than the difference +position on the stack, we output difference positions from the stack +until the difference position on the stack is greater that the current +difference position, meaning the node represented by the difference +position on the stack is a child node of the current node. We then put +the current difference position on the stack, and repeat for the next +record. + +A similar algorithm simply generates the hash as if we had the literal +nodes. + +A full path to a leaf node, proving that the leaf node is represented in +the tree, contains not the hashes of the things in the path, but the off +path hashes and off path keys. + +An incomplete tree has the same data structure as full tree, but with +missing nodes. A full path is a more compact representation of an +incomplete tree, and is treated as a compressed form of an incomplete +tree with a single leaf node present. You can regenerate an incomplete +tree from a full path, and a full path can be generated for any non +missing leaf node in an incomplete tree. + +Rather than a chain of blocks, we have a Merkle-patricia dac of blocks, +where the index is the block number. This means that the state of any +block can be proved to be part of the global consensus with a proof of +length logarithmic in the total block number. Thus peers can provide +clients with short proofs, so that clients do not have to take +assertions by peers on trust. + +We also have a small number of Merkle-patricia dacs representing the +state of the system at any given block time, for example a Merkle +patricia dac of unspent transaction outputs, and a trie linking globally +unique human readable names to probabilistically unique public keys. +These trees change each new block, though their state in past blocks is +immutable.  Each new block contains not the entire new Merkle-patricia +dac, which is apt to be enormous, but only those nodes that have +changed.  A new block contains the new roots of the new Merkle-patricia +dacs and their new descendants, which link to unchanged descendants in +past blocks. + +Peers synchronize their state by sharing new information to form a new +block. They efficiently discover what they have in common, and what is +new, by sharing the root of the Merkle-patricia dac describing the new +block, and then give each other the new information, after the fashion +of usenet news.  + +A new block is not just a list of new events, but it is generated from a +list of new events, which are themselves listed in a Merkle-patricia +dac. To compare nodes in the tree of new events, a peer sends its +neighbours an offset into the hash, the leading part of the key of the +node, a one bit flag to indicate if the node has children, a hundred and +twenty eight bits of the hash of the node, and for each of its two +children, the leading parts of their keys,a one bit flag to indicate if +the node has children, and sixty bits of their hash, and for each of the +four grandchilden, the leading parts of their keys, a one bit flag to +indicate if the node has children, twenty eight bits of hash, for each +of the eight great grandchildren, the leading parts of their keys, a one +bit flag to indicate if the node has children, and thirteen bits of +hash, for each of the sixteen great great grandchildren, the leading +parts of their keys, a one bit flag to indicate if the node has +children, and six bits of hash, for each of the thirty two great great +great grandchildren, the leading parts of their keys, a one bit flag to +indicate if the node has children, and three bits of hash, and for each +of the sixty four great great great great grandchildren, the leading +parts of their keys, a one bit flag to indicate if the node has +children, and two bits of hash. + +This tells it which subtrees are definitely different, and which are +definitely new, and for each subtree definitely different, it sends more +comparison data for that node, and for each subtree definitely missing, +it sends that subtree.  Once there are no more subtrees to be sent, +repeats the process starting at the root once again. + +When it has some information about which subtrees are definitely +missing, which are probably the same, and which are definitely +different, it then sends much the same, but for each one missing, sends +the full subtree, for each probably the same, skips, for each definitely +different, doubles the level of detail – repeats with a different part +of the hash, and twice as many bits. + +So now we need more than a one bit flag. Need to distinguish between the +cases: + +1. no children +2. Sending a full item – a leaf node. +3. skipping a child subtree because we are likely in agreement +4. dropping down to a lower level of detail, for example from thirteen + bits of hash to six bits of hash. If down to two bits of hash, + always going to leave out the children. +5. not dropping down to a lower level of detail, keeping the same + number of bits in the hash. + +The intent is to discover what parts of the tree we have agreement on, +and send an image of the tree with those parts skipped over.  Rinse and +repeat. We annotate our model of the tree with the probability that a +subtree is identical. If the other guy sent us a leaf node, we know the +node is identical, and every hash fragment that agrees creates +exponential probability that a subtree is identical.  If we recently got +a leaf node from source, not from sharing, we know for sure that the +other guy does not have it, and send it to him. + +Ok, that covers information gathering, but what about the final stages, +when we are going to throw out data that was late in coming, for the +sake of consensus? + +A proposed final hash of all items to be in a block for a certain period +is announced. And now, the job is to get those items that are missing, +and tag those items that the are not in the proposed final hash, and +exclude them from a version of the tree. So instead of “definitely not +present in the other guy’s hash” means you send the guy your item, it +now means you exclude it, and see if that gets your root hash to agree. + +A peer in good standing endorses the proposed final hash, the root of +the block if it can get its trees to agree + +When building a block, the peers share these new events. When coming to +a consensus, the peers attempt go get agreement on the new events in the +block. But the block will also contain diffs on the Merkle-patricia dac +of unspent transaction outputs, and the Merkle-patricia dac of spent +transaction inputs. The peers need to maintain these trees so that +clients can see proof of consensus on the tries, so that a peer cannot +mislead a client, and a peer should only vote for a consensus if it can +generate the same root hash for the the block thus has the same tries +describing the entire block chain and providing access to a block chain +for clients. + +After agreement on Merkle-patricia dac of all new items for the new +block, a peer generates the other revisions of the other Merkle-patricia +dacs, and add them to the block, the generated items that go into the +block, and you should get final agreement. If a peer gets final +agreement – all block items are valid, then it votes for the consensus.  + +The block contains the root of a Merkle-patricia dac that contains all +previous blocks (but not the current block) – thus not so much a block +chain, as a block trie, which means that the proof of any fact about the +state of the block trie is reasonably short, of order log N hashes, +where N is the number of items in the block trie.  + + +# Node identifiers + +We will call the thing that identifies a node a node infix order, and +the member of the subset that the partricia tree identifies a key - +because we are generally using it as a map key. Part of the map key is +part of the node infix order. + +But the canonical form of the map key is a bitstring, which is what we are going to hash. + +The node infix order is the representation of the bitstring with significant +bits in the different bitstrings aligned, padded on the right with a one bit +and as many zero bits as needed, and for a tree of quantities unbounded on the + left, padded with a one bit. and as many zero bits as needed on the left. + +In practice all integers have some finite bound, but one does not want +the computer word length to affect the data on the wire or the result of +the hash, so it usually preferable to structure the data and algorithms +so that the actual left bound has no effect provided it is sufficiently +large, as if the integers had unbounded precision. We will not actually +need numbers larger than 2^64^ until Anno Domini ten thousand or so, but +the when we do, it will have no effect on blockchain format or previous +hash results, hence for integers we will use a tree with right bound but +no left bound. + +But a patricia tree is bit string oriented. So for integer indexes, in +order that we can ascertain which nodes correspond to the leaf nodes, +need to have, associated with the root node, the the length of the bit +string for the leaf node. The length needs to be run time value, rather +than a compile time value. But for hashes, can be a compile time value. + +To map from bit strings to byte strings, and to have a bit string index +for leaf nodes and tree nodes, we either append 10000\... or 011111\... +to bitstring, to get a one to one map from byte strings to bit strings, +and from bit strings to integers. + +Compiler intrinsics are generally ffs. For Microsoft compilers use +`_BitScanForward` & `_BitScanReverse`. For GCC use `__builtin_ffs`, +`__builtin_clz`, `__builtin_ctz`.([find first +set](https://infogalactic.com/info/Find_first_set)) or ctz (count +trailing zeros) so is probably faster to represent a bit string as a +word string by appending 100000\... than 0111111\.... See [Microsoft +`__lzcnt()`](https://docs.microsoft.com/en-us/cpp/intrinsics/lzcnt16-lzcnt-lzcnt64?view=vs-2019) +and [gcc +`__builtin_clz()`.](https://gcc.gnu.org/onlinedocs/gcc/x86-Built-in-Functions.html) +. `BitScanReverse` is portable between processers, and lzcnt is not, so +you need a runtime check at the start of program to see if your code can +run. + +If the leaf nodes correspond to the integers 1 to N, where N is at most +m bits long, then the bit strings of the leaf nodes are m+1 bits, +becaust we have 2N-1 vertexes in the tree, and 2N edges. + +# Sparse or sequential, complete or partial, Merkle-patricia trees. + +## Sparse and complete + +As, for example, a Merkle-patricia tree of a map that maps globally +unique human readable and writeable names to public keys. + +A patricia tree with no right bound on its keys is necessarily sparse. +Thus, for example, in a patricia tree of strings terminated by a null +byte, the null byte ensures a gap between the address spaces governned +by two successive keys. + +If a node in the tree has direct children (no skips) or the leaf nodes +are sequential and contiguous its has its hash is the hash of the hashes +of children. If the tree is sparse part of what we know is the gaps, +hence the parent node has to hash the skip links. If the tree is +sequential and contiguous, then the root node has to directly hash its +direct descendants, and also the size of the tree. + +A block is a sparse tree, while a blockchain is a sequential tree of +sparse trees. The sparse tree in a block will itself contain sequential +trees, and the root hashes of numerous sequential trees. + +In a sequential Merkle-patricia tree, the hash of the parent node is +simply the hash of the hashes of its two child nodes, but in a sparse +tree, we want two maps from keys to objects with the same sequence of +the same objects, but different key values, to have different hashes, so +hash of a parent node has to be hash of the hashes of its two child +nodes, plus the hash of the portions of each child’s key that govern +the skip links to those two children, that portion of each child’s key. +The portion of the childs key that govern the skip link is (Level +difference -1) bits long. + +But computers do not handle bit fields easily, and databases do not +handle them at all, plus, how do you hash a bit string, such as the leaf +indicator? + +Any variable length field creates ambiguity in the hash, so that two +values could be hashed as the same stream of bytes. To avoid this +outcome, the canonical format for a bit string will be an integer in +stream format specifying the number of bytes, followed by the bit +string, followed by a zero bit, followed by enough one bits to fill to +the next byte boundary. If the representation of a bit string as an +integral number of bytes has some 0x00 bytes at the end, it is not in +canonical format and gets truncated before hashing till it no longer has +any 0xff trailing bytes. + +If a bit string has a start position that is not necessarily byte +aligned, and is known from context, we left pad it with zeroes. If its +start position is not known from context, we provide the starting bit +position as an integer in stream format. At this point in the code we +are deep in the bitbashing weeds, and are no longer worried about +passing the bit string around as a regular byte string. + +Assuming no prefix problem, one way or another way, then the index of a +node can be two fields, the bit string, and the number of bits in the +bit string. But since we already have to represent the number of bytes +in the field representing the bit string, we might use the 01111.. +canonical format trick, so that we can use the more familiar, standard, +and convenient infix order of byte string, in which case we will have to +pad the map key to form the node infix order. + +However we do this, it is an implementation detail that should not +affect the canonical form or the root hash, and the appended 0111\... +form simplifies fencepost problems on interpreting what is on the wire. +Otherwise we are always going to be bothered by distinguishing the bit +string 0101000 from the bit string 0101. One less `if` to screw up. + +Since a sequential Merkle-patricia tree always maps the integers from +zero to n-1 to the objects, hashing link information is redundant, we do +not actually need any link information. It is sufficient that the root +hash defines the objects and the sequence. + +## Sequential and complete + +The tree bears a simple and natural relationship to a linear vector of +leaves and a linear vector, one smaller, of nodes. + +Each node appears with a position in the linear vector one less than the +position of the leaf that required it to be added to the tree. + +Size eight Merkle-patricia tree by: + +- patricia bitstring +- right padded patricia bitstring (key) + +::: {style="text-align: center;"} +“”\ +1000\ + +::: {.clear} +::: + +::: {style="float: left; text-align: center; width: 45%;"} +“0”\ +0100\ + +::: {.clear} +::: + +::: {style="float: left; text-align: center; width: 45%;"} +“00”\ +0010\ + +::: {.clear} +::: + +::: {style="float: left; text-align: center; width: 45%;"} +"000"\ +0001\ +::: + +::: {style="float: right; text-align: center; width: 45%;"} +"001"\ +0001\ +::: +::: + +::: {style="float: right; text-align: center; width: 45%;"} +"01"\ +0110\ + +::: {.clear} +::: + +::: {style="float: left; text-align: center; width: 45%;"} +"010"\ +0101\ +::: + +::: {style="float: right; text-align: center; width: 45%;"} +"011"\ +0111\ +::: +::: +::: + +::: {style="float: right; text-align: center; width: 45%;"} +"1"\ +1100\ + +::: {.clear} +::: + +::: {style="float: left; text-align: center; width: 45%;"} +"10"\ +1010\ + +::: {.clear} +::: + +::: {style="float: left; text-align: center; width: 45%;"} +"100"\ +1001\ +::: + +::: {style="float: right; text-align: center; width: 45%;"} +"101"\ +1011\ +::: +::: + +::: {style="float: right; text-align: center; width: 45%;"} +"11"\ +1110\ + +::: {.clear} +::: + +::: {style="float: left; text-align: center; width: 45%;"} +"110"\ +1101\ +::: + +::: {style="float: right; text-align: center; width: 45%;"} +"111"\ +1111\ +::: +::: +::: +::: + +::: {.clear} +::: + +\ +\ +\ +\ +\ + +\ +\ +\ +\ +\ + +We want to collapse the tree into a linear +list, so that we can find the correct node without walking one bit at a +time through a binary tree, both in order to represent relatively small +blocks of hashes, and also to represent the top of very large trees. + +We extend each bit string to a fixed and uniform size by appending a one +bit, followed by as many one zero bits as are necessary fill to the standard +size, so that we can use uniform length bit strings, instead of variable sized +bit strings, so that we can use them directly to access an array in memory, +or OIDs in a database. + +The resulting padded bit strings (keys) are in infix order. But for an immutable append only file or database, we want postfix order, which is +harder. And we don't want to be restricted to only having a power of two objects. We want to be have an arbitrary number of objects, and add an arbitrary number of objects without changing the existing tree. + +A binary postfix tree, power of two, is going to look like this: + +For an append only structure, the position of a leaf node is number of prior +leaf vertexes, plus number of prior internal vertexes, and for an sql append +only database, the oid of a leaf node is number or prior leaf nodes, plus one. + +So a leaf oid is: +$$\displaystyle\frac{key}{2}+1$$ +where $key$ is the patricia bitstring right padded to a fixed size by a one +bit, followed by as many zero bits as needed. + +Let $\displaystyle{C_{key}}$ be the count of bits in the key. (which in C++ +is `bitset.count()`, +But C++ provides no access to cool intrinsic assembly instructions.) + +The number of internal vertices prior to a leaf is +$$\displaystyle{\frac{key}{2}+1-C_{key}}$$ + +So the leaf position in postfix order is: +$$\displaystyle{size_{leaf}*\frac{key}{2}+size_{vertex}*({\frac{key}{2}+1-C_{key}})}$$ + +Let the height of a vertex be $h_{key}$, the number of trailing zeroes in the key. + +So the postfix vertex position, supposing vertexes are in a separate data structure, is +$\displaystyle{size_{vertex}*\frac{key}{2}+2^{h_{key}}-C_{key}}$\ +Is that right?\ +Need to check it. + +I think we could also get the postfix vertex oid by +$\displaystyle{\frac{{(key-1)} | {key}}{2}+1-C_{key}}$, but again, needs checking. + +Given the vertex oid and the leaf oid, the absolute position in the direct immutable append only file is easy to calculate. + +Everyone seems to wind up using regular [C bit twiddling hacks], because +hardware intrinsics are erratically available, and because the efficiency +improvement of hardware intrinsics is seldom worth the thought. + +[C bit twiddling hacks]: +http://graphics.stanford.edu/~seander/bithacks.html#CountBitsSet64 +"Bit Twiddling Hacks" + +[uint64_t bitcount(uint64_t c)](../bit_hacks.h) + +To do the reverse operation, finding the key (the padded patricia index) +from the postfix position make the starting guess that the $C_{key}$ adjustment +was zero, find the corresponding patricia key, and then walk the tree from +you where guessed that you were are, to where you should be. You find +the predicted postfix position of your guess, find the order of the highest +order bit where they differ, and walk the postfix position and padded +patricia key (infix position) in parallel. + +## Adding to a sequential and complete Merkle-patricia tree + +Well that solves the problem of a postfix tree, but how do we apply this to +solve the problem of the number of items not being a power of two? + +### A sequential append only collection of postfix binary trees + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + id + + + + + Immutable append only file as a Merkle chain + + + + + + + + + + + + + + + + + + Immutable append only file as a collection of + balanced binary Merkle trees + in postfix order + + + + +The superstructure of balanced binary Merkle trees allows us to verify any +part of it with only $O(log)$ hashes, and thus to verify that one version of +this data structure that one party is using is a later version of the same data +structure that another party is using. + +This reduces the amount of trust that clients have to place in peers. When +the blockchain gets very large there will be rather few peers and a great +many peers, thus there will be a risk that the peers will plot together to +bugger the clients. This structure enables a client to verify that any part of +the blockchain is what his peer say it is, and thus avoids the risk that peer +may tell different clients different accounts of the consensus. Two clients +can quickly verify that they are on the same total order and total set of +transactions. + +Edges of the graph are represented by hashes, and thus can only travelled +from right to left, Vertices are represented by their hash, and in their +canonical form contain child hashes and their full padded Merkle patricia +key within a tree as an arbitrary precision integer, and thus their implicit +postfix position within a tree, which identities provide implicit edges that +can be traversed in any direction with respect to a given total order, while +hash edges are agnostic of total order, and are what we construct the +consensus about total order from. + +The bottom most part of the structure consists of data structures that do +not have a canonical order. But when we are figuring out how to order +them, we have to construct vertices on top of them that do have a +canonical order, where each vertex contains a hash commitment to a total +past in total order and a patricia key representing its position in the total +order. + +When the chain becomes very big, sectors and disks will be failing all the +time, and we don't want such failures to bring everything to a screaming halt. + +And, when the chain becomes very big, most people will be operating +clients, not peers, and they need to be able to ensure that the peers are not +lying to them. + +If our initial tree has has a size of zero, this is the same as creating +a sequential and complete Merkle-patricia tree. + +Since the whole point of a Merkle tree is immutable entities, we seldom +want to insert, delete, or update anything but the right hand edge of a +sequential Merkle-patricia tree, and normally only want to insert on the +right hand edge. + +Thus a sequential Merkle-patricia tree is not exactly a block chain, +since each block does not contain the hash of the previous block. If it +did, you would potentially have to receive and calculate a lot of hashes +of hashes. to ascertain that block one thousand did indeed chain to +block five hundred. This structure means that clients can calculate the +validity of the block chain for those parts of it that contain +transactions that concern them, and know that everyone else doing +similar calculations are getting results that show the same consensus as +they are getting, thus calculating the validity of block chain is +distributed to all clients without all clients needing to deal with the +entire block chain. + +If all the peers get together to screw over one client or few clients, +that client is going to have cryptographic proof of misconduct. On +publishing that proof, large numbers of people are likely to blacklist +those peers, resulting in a fork in the blockchain. We could automate +this process, with everyone automatically disregarding the signatures of +peers for which a proof existed that they had changed the rules in a way +inconsistent with the rules implemented in a client, so that if nine +tenths of the peers change their software, and nine tenths of the +clients do not, we automatically get a fork with nine tenths of the +clients on one block chain with a tenth of the peers, and one tenth of +the clients on the blockchain with nine tenths of the peers. + +This architecture allows a client peer arrangement where to pervert the +blockchain, you have to synchronously pervert everyone’s software +everywhere, or most people’s software, whereas with bitcoin, a few big +miners can pervert the blockchain. + +Bitcoin will fail because power over the blockchain lies with a few big +miners, and governments will eventually twist their arms, or themselves +get into the business of mining. This is inherent in the scaling +problems of bitcoin. Back when everyone was miner, and everyone had the +whole blockchain on their machine, power was distributed in a way that +ensured that the blockchain was conducted according to consensus rules, +but as the blockchain gets bigger and bigger, fewer and fewer people +host the complete blockchain, and fewer and fewer people mine. So we are +back to the situation we wanted to escape, where a rather small number +of people have power over other people’s money. + +A client has partial trees for all his transactions in the blockchain, +and if all clients check their own particular part of the block chain, +the entire blockchain is checked. + +Well this is a great layout if we have data structure that fits entirely +in memory, and it is a great layout if we have an enormous mutable file +and are appending some bits to it, and have log n chunks mapped into +memory leading to the part where we are appending stuff. + +But if we are frequently constructing partial hash trees here and there, +the trouble is we log n non localbits things to look up. If we are +putting a big Merkle-patricia tree in a database, better to infix order +by the size of the bit string, then the content of the bit string filled +out to the long word size with infix order bits, or if it does not fit +in a long word, as a blob. (sqlite sorts nulls first, then integers, +then strings, then blobs, and if you write something as an integer and +read it as a blob, gets converted a string representing a decimal) + +Suppose we want a literally immutable append only file, representing a +sequential patricia Mekle tree, where we append each completed binary +tree? + +Well in that case the high nodes have to follow each completed subtree, +and our postfix order would be: + +| | | | | | | | | +|---:|---:|---:|---:|---:|---:|---:|---:| +|000|001|010|011|100|101|110|111| | +||00||01||10||11| +||||0||||1| +||||||||“”| + +Thus, to construct the postfix order for the leaf position in the tree +using right side nodes, leaf `n` has postfix order +`(n<<1)-std::bitset<64>(n).count()`. Which is considerably more +complicated than the infix order for leaves with nodes uniformly +intermixed with them. + +Further, the not a power of two case is more complicated. To figure out +which nodes at the right side are skipped, have to have the infix order +of the node. Then we pack all the non skipped right hand nodes following +the last leaf node. There is no simple way to identify a skipped node +from the postfix order. + +But the postfix order has the huge advantage for enormous data +structures that when you are constructing, access is sequential and +forwards only, and if you are making an ever larger file, the additions +are append only. + +But this is essentially a plan to build our own database, presumably +with a second file containing a copy of only higher nodes, and a third +file containing a copy of higher nodes still, until one gets down to +copy of the high nodes that fits entirely within memory. X64 has 64 byte +blocks in its top level cache, and can virtual seventy terabyte files +into its copious address space. 4kilobyte blocks for disk access tend to +be fastest but 64 kilobyte blocks are only marginally slower, though in +some algorithms they waste more memory. Microsoft recommends 64 kilobyte +disk blocks for servers. This suggests a structure in which every time +we have a new two byte word in the bitstring, we have a different offset +corresponding to a different area on disk and memory, a hierarchy of +files each 256 or 65536 times smaller than the other. + +A radix 256 Merkle-patricia tree on top of a radix 2 Merkle-patricia +tree would be referencing sixteen kilobyte blocks, which sounds like it +is near the sweet spot, in which case the cache file for the equivalent +of the bitcoin blockchain will fit into a gigabyte of ram. + +It is premature to think of designing this. After two years at bitcoin +volumes, our blockchain will be two hundred gigabytes, at which point we +might want to think of a custom format for sequential patricia trees as +a collection of immutable files, append only files which grow into +immutable files, and wal files that get idempotently accumulated into +the append only files. + +For the first level nodes, the ones directly above the leaves, the +postfix order is 000🡒00010, 001🡒00101, 010🡒01001, 011🡒01100, 100🡒01001, +101🡒10100, 110🡒11000, 111🡒11011. We append two zeroes, subtract the +count of the bits of the bitstring, and add two. + +Or equivalently, we take the infix order of first level node, subtract +the count of the bits of the infix order, and add two. + +For the second level nodes 00🡒00110, 01🡒01101, 10🡒10101, 11🡒11100, we +append three zeroes, subtract the count of the bits, and add six to get +the postfix order. + +For the third level nodes 0🡒01110,1🡒11101. We append four zeroes, add +fourteen, and substract the count of the bits to get the postfix order. + +For the fourth level node, null🡒11110, we append five zeroes, add +thirty, and presumably subtract the number of bits to get the postfix +order. + +The general formula appears to be add 2\^(level+1)-2-count) + +Reversing the postfix order to get the bitstring and the level seems +rather hard. You have to truncate, then find `x: (a+x).count()=x`. And +since there is no clean and elegant way of finding `count()`) it is not +likely that there is a clean and elegant way of finding `x.` But it is +very easy, given the level, to find the parent, the children, and the +sibling, since these are at fixed offsets. Looks like when iterating +through structures in postfix order, you have to keep the level and the +bitstring implicitly or explicitly around, whereas with infix order +there is a clean and simple relationship between the infix order and +level and bitstring. + +To reconstruct the bitstring from the postfix order, the fastest way is +probably to construct it one bit at a time by conceptually walking the +tree from the root until we match the postfix order, not recalculating +count every time, but incrementing it every time we move right in the +tree, and representing the level by the power of two that we will add to +the bitstring when we move right. + +## incomplete. + +We take a list of hashes and their node infix orders, the offset for +sequential trees, and the bit strings for left trees, and stuff them +into map mapping node ids to hashes, using a map that allows random and +sequential access. The representation of sparse and incomplete tree is +similar to the representation of a sequential and incomplete vector. We +provide the nodes that allow the construction of the chain of hashes +from the object to be authenticated, but we do not provide their +children. + +## Sparse. + +In a sparse Merkle-patricia tree, we are not going to do bit bashing on +the key, because it is likely inconveniently large, and because we are +likely to be counting the height from from the left. But we might do it +for code re-use reasons, where the keys are integers of moderate size. + +It is customary to define the least significant bit as bit zero, which +is the convention I have followed in the description of sequential +patricia Markle Trees. So in a patricia tree with node height measured +from the right hand side, nodes have a height, which is zero for leaf +nodes, and variable for the root node. In a sparse tree, one is apt to +measure the from the left, so nodes have a depth, not a height. The +depth of the root node is always zero, and leaves have a variable depth. +To make the code templatable, will need a patricia tree type to have a +member `static constexpr bool left_edge = true;` for trees with node +height measured from the left edge, and `false` for trees measured from +the right edge. The depth of bit three of a byte in position seven in an +array is `8*7 − 3` The node depth is equal to the position of the bit +that selected that node. If a node is the parent of all keys with the +most signficant bit set in the start byte of the key it has depth one. + +The standard algorithm for entering a new item assumes you are using the +patricia tree as the infix order, but databases and map files do not do +this. When we are big and successful, will write a version of sqlite +that supports sharding and Merkle-patricia infix orderes. + +And because the tree is sparse, it is probably coming in somewhat non +sequentially, with updates in random locations, though block updates are +likely to be sequential. + +Nonetheless, the analysis of sequential blocks suggests we will get more +locality of reference on disk and memory, if we put all the patricia +nodes, leaf and vertex, in one big table, with the key of a vertex node +locating it close to the leaf node that necessitated its creation, +though we will need the bit offset to be part of this table. The height +means we don’t need keys of different levels be guaranteed unique, +because when we are after a particular level, we use the height, likely +measured from the right hand side, to select the correct candidate in +the improbable event of a collision. + +The bitstring of a node infix order is going, in practice to sorted on +whole bytes or whole words, rather than its exact bit length, resulting +in collisions, therefore we have sort on the bitstring and height, or, +as in a sequential patricia tree, pad the bitstring with a +representation of the height. Which for long keys is going to be a bit +verbose, though hardly a show stopper. We could pad the bitstring with +the difference between the height, and height rounded to the nearest +number of whole words, but this would have the irritating result that +256 bit hashes become 257 bit hashes, unless we truncated to 255 bits, +which is harmless but nonstandard and apt to result in complications all +over the place. And we would still be looking at the height, if only to +know the number of bytes or words. + +The analysis of sequential hash trees suggests that it is going to be +more efficient to have all nodes of all levels in the same table, +possibly including leaf nodes. We could store the complete keys of +parent and its children in a node, but this is rather redundant. If we +take the bit field offset from the left, rather than from the right as +in sequential hash trees, then we truncate the key, conceptually at the +bit indicated by the bit offset, and only have the parts of the keys of +the children that are beyond the bit offset in the node. + +In a sparse tree, we are just looking up the key to find the thing +referred to, which is probably referenced by an oid, so it may well be a +pointless indirection to have leaf nodes in the same table as their +parent node, which already has both the key to the child and an oid to +the child. Maybe we just have a leaf flag that says “this oid is not an +oid to yet another internal node, but an oid to something else, +somewhere else”. + +Two maps, one of them being the map that you are making a Merkle +patricia hash of, and one containing the nodes. The nodes The nodes +point to a other nodes, or they themselves contain the leaf value, the +leaf value being the hash of the map key and object being hashed. + +On the wire, we don’t have a representation of the sparse tree. The +nodes get generated on the receiving end when we sent the objects +themselves. But we do have representations of incomplete sparse trees, +which are just a list of nodes with their hash values and keys, but not +their child links, that you will need to construct the root node from a +subset of the objects. + +In rare and exceptional cases, the key of the root node by not be empty +bit string, so when specifying the root node of a sparse tree, we have +to specify its bit string. But when we send a partial tree, we assume +the recipient does not have, and does not intend to build, the complete +tree, and that he receives the root node by some special path that +authenticates it. + +A node with children will be inherently different to a node without, +because it is going to need the data to construct the infix orderes of +its children, which is a good argument for putting them in distinct +tables. If we put leaf hashes and vertex hashes in the same table, we +are massively violating normal form. On the other hand, Sqlite3 supports +this by having variant typing. For sparse trees, not usually a good idea +to put the children in the same table. For sequential trees, usually the +way to go. + +## Merkle-patricia Tree of 256 bit values. + +Typically these are going to be a tree of public keys and or hashes. + +Of course a database cannot handle infix order values that are not a +multiple of eight bits, and is likely to prefer a multiple of sixteen +bits, so we will make the key field of the infix order value fields the +prefix bits that all the leaves of a node have in common, with the bit +string 0 appended, rounded up to whole number of bytes with additional +one bits. + +The depth of a node may be determined by taking the number of bytes +times eight, and substracting the bits from the last zero leftwards. +leaf nodes will be special cased. + +If leaf nodes are part of the node map, we need 257 bits – but we are +also going to need special case handing of leaf nodes, which is +effectively the 257th bit. + +Since our infix order has two fields, it is efficient to reference nodes +by OID, rather than by key, though you frequently have to find a node by +its key field and its depth field. So each internal node will have its +depth, (or rather a value reversibly constructed from its depth) and its +key (or rather a value reversibly constructed from its key and its +depth), and the OIDs of its two children. For a leaf node, the child OID +fields have a different meaning. + +To reconstruct the hash of a node, we look up the two children by OID, +and hash their hashes and the bit strings that differentiate them from +their parent and each other, or rather the the bytes containing the bit +strings that differentiate them from their parent and each other. We +hash the hash and bit string of one node first, then the hash and bit +string of the next node, to avoid the possibility that one concatenated +pair of bit strings might equal another pair of concatenated bit +strings. + +If we are inserting a bunch of new map entries with the intent of +recalculating the new value of the root node once the insertions are +done, we sort them into a temporary table in order of their map key, and +after each insertion calculate the new values of the parent nodes up to +but excluding the parent node that is the parent of both the item +recently inserted, and the next item to be inserted. If no next item to +be inserted, up to and including the top. + +Pretty sure that this guarantees all nodes get recalculated as necessary +and only as necessary, but we will need debug check to make sure we have +not missed a fence post problem. Have a forced re-evaluation of the +entire tree to check for internal consistency. + +When we are generating blocks on the block chain on an ordinary computer +with an ordinary internet connection. (16mbps down, 3mbps up), and +blocks are typically 300 seconds per block, that implies that blocks are +smaller than a gigabyte, so we can prepare blocks entirely in memory, +either as an in memory sqlite database, or in a custom format using +links, rather than OIDs, and key values of the infix order in sixty four +bit integers, rather than bytes. Once a block is committed to global +blockchain, it however has to acquire global oids, hence goes in a +different database, probably in a slightly different database format, +since all the oids for a given kind item are sequential. But the hash of +a block is calculated in a way entirely independent of the oids that +will be assigned to it. Perhaps when it is committed, it gets capped by +additional data telling us what oids were assigned, but including the +oids when calculating the root hash of a block under construction, and +therefore suffering frequent inserts in random positions, would make it +too costly to insert new data. + +Such information about the type and number of objects in the block, and +the block’s position on the block chain, should not, on the dry +principle, be part of the block or the block root hash, though it might +well be useful to include such information in the authentication block +(authentication blocks alternate with blocks containing actual updates +on the block chain) When combining two proposed blocks, it is good to +start by merging the larger block into the smaller block, and as a check +that everything is working as it should, make sure that the counts are +consistent, this being a low cost check against glitches in calculating +the hash. We don’t need to incorporate such a consistency check into +the blockchain itself until the authentication block, which attests to +the previous root of the block chain, which attests to the previous root +and previous sequence number of the block chain, thus necessarily +contains its own sequence number, unlike a data block. + +The hash of a block should be independent of its oids, but we do not +want a glitch to result in the oids of a peer getting out of sync with +its peers and that peer attempting to sail on unawares that it is out of +consensus, therefore the hash of an authentication block should contain +a hash that depends on the hash of the final authoritative oids of all +blocks in the preceding block chain, or all oids affected by the final +block. + +On the dry principle, objects inside a block should not contain +information as to what block they are in, and a block should not contain +information as to what blockchain it is in nor what block number it is. +But a block should be accompanied by context information as to what +block number it is and what block chain it is in, and the root of the +blockchain should be accompanied by context information saying what +block chain it is and h1ow many blocks are in it. + +An authentication block attests to the recent previous block of the +block chain in isolation from context, and therefore links to its own +context, but in general, the meaning of a hash is given by the object +that hashes to it. And we somehow have to find or construct the object +that hashes to it. The object that hashes to it therefore needs to +contain information that implies its size and how it is to be hashed, +what it means. With a hash, we ad hoc provide context information to +hint where to find the data that generates the hash. But the chain to +more and more global meanings is on the inside of the information being +hashed not on the outside. Every transaction provids the oids of the +unspent transaction outputs that it refers to and the signatures +authenticaded by their public keys. When an object contains an unhashed +reference to data outside itself, it must hash the broadest context in +which that data occurs. Thus a transaction must contain or imply the oid +of a previous root hash of the blockchain containing all these +transaction outputs, and the hash of the transaction must hash that +root, thus the each transaction authenticates and is authenticated by +previous blockchain. + +Repeating. Where an object contains a reference to an object outside +itself which reference is not itself a hash, which is to say, when the +reference is an oid, then its hash has to chain to that object, +preferably by chaining to the largest object in which that object +occurs. The oid is merely a hint to finding data that chains into the +hash of the object, and the derivation of the correct hash depends on +correctly identifying the thing that the oid refers to, the data +structure into which the oid is a pointer. The oid could merely be the +eleventh hash in a sequential map of seventeen hashes, and to know what +the object means, you have to know, or correctly guess from context, +what it is. So for the hash of the object to uniquely determine the +meaning of the object, it would have to hash the root of the Merkle +patricia tree of that map that defines the oid. Objects containing oids +have to chain to the context that gives meaning to their oids, which +context chains to the larger context, the largest context of them all +being a recent past root of the global blockchain, which provides unique +oids global to the blockchain, for each type of oid that is directly in +the blockchain. And to the extent that the context providing oids +provides separate oid sequences for each type of object within it, and +thus an oid can refer to multiple things of multiple types within it, +then the type of the object or other data within the object has to imply +the type of the oid. + +In short, if the object mapped by the map a Merkle-patricia tree +references another object, it should it should either reference it by +hash, by the hash of a patricia tree containing that object and the key +within the referenced patricia Mekle tree containing the referenced +object, or the hash of the object should depend on the hash of the +objects that it references, or the hash of the Merkle-patricia tree +containing that object, in order that the hash of the referencing +object, and thus the hash of the patricia Markle tree containing the +object, uniquely identifies everything that the object references. + +# Implementation of the specific Merkle-patricia trees we will need. + +We have a sequential tree of attestation blocks containing proof of +consensus, each of which chains to the previous block, and each of which +contains the root hash of the tree as it was before this block was +added. + +We have a sequential tree of transactions, with each attestation block +attesting to the state of the sequential tree of transactions as it was +at that time, so that each participant knows that they agree with the +consensus on transactions, and they agree on the specific sequence +number of each transactions. They also agree on the current state after +all transactions were applied, they agree on the sparse tree of unspent +transaction outputs. + +In order to generate that order, each full peer have to construct a +sparse tree ordered by hash code. When they reach agreement on the root +of that hash tree, they then proceed to agreement on the sequential list +of transactions, and the current state of unspent transaction outputs. + +## Sparse tree of hashes + +Because the bit string is going to be generally quite short, or else +exactly two hundred and thirty six bits, we treat this as a list of +bitstrings, with each list composed of lists of bitstrings, with each +list and sublist prefigured by a bitstring corresponding to their common +prefix. We don’t try to save space by omitting the (typically short) +common part of the prefix, and we pad the bit string to a whole word +with the padding sequence 1000000\.... + +Each node contains a bit count and a count of its descendants. We +special case bit strings corresponding to whole words, and we special +case that special case for the case that bitstring is exactly two fifty +six bits. + +We don’t have a representation for the empty list, but we do have a +representation for a list consisting of a single item. So we build the +tree by starting with a single item, and then adding items to the tree. +The case where we have no items is handled separately and specially. We +never have an empty tree. A single item is itself a tree, and you can +add a tree to a tree. A tree is, of course composed of trees. + +Because we are not prefix compressing the items in the tree of hashes, +our skip links do not contain the skip data, and the hash of a tree with +more than one item it does not hash the common prefix, but simply hashes +the hash of the subtrees. The hash of a leaf node, is, of course, +itself. + +The root of this tree is recorded in the blockchain, but we do not need +to store the tree itself in the blockchain. It lives only in memory. The +sequential tree of transactions that it gives rise to lives on disk. diff --git a/docs/mkdocs.sh b/docs/mkdocs.sh new file mode 100644 index 0000000..a03d151 --- /dev/null +++ b/docs/mkdocs.sh @@ -0,0 +1,72 @@ +#!/bin/bash +set -e +cd `dirname $0` + +if [[ "$OSTYPE" == "linux-gnu"* ]]; then + osoptions="" +elif [[ "$OSTYPE" == "darwin"* ]]; then + osoptions="" +elif [[ "$OSTYPE" == "cygwin" ]]; then + osoptions="--fail-if-warnings --eol=lf " +elif [[ "$OSTYPE" == "msys" ]]; then + osoptions="--fail-if-warnings --eol=lf " +fi +templates="./pandoc_templates/" +options=$osoptions"--toc -N --toc-depth=5 --wrap=preserve --metadata=lang:en --include-in-header=$templates/header.pandoc --include-before-body=$templates/before.pandoc --include-after-body=$templates/after.pandoc --css=$templates/style.css -o" +for f in *.md +do + len=${#f} + base=${f:0:($len-3)} + if [ $f -nt $base.html ]; + then + katex="" + for i in 1 2 3 4 + do + read line + if [[ $line =~ katex$ ]]; + then + katex=" --katex=./" + fi + done <$f + pandoc $katex $options $base.html $base.md + echo "$base.html from $f" + #else + # echo " $base.html up to date" + fi +done +cd libraries +for f in *.md +do + len=${#f} + base=${f:0:($len-3)} + if [ $f -nt $base.html ]; + then + katex="" + for i in 1 2 3 4 + do + read line + if [[ $line =~ katex ]]; + then + katex=" --katex=./" + fi + done <$f + pandoc $katex $options $base.html $base.md + echo "$base.html from $f" + #else + # echo " $base.html up to date" + fi +done +cd ../.. +templates=docs/pandoc_templates/ +for f in *.md +do + len=${#f} + base=${f:0:($len-3)} + if [ $f -nt $base.html ]; + then + pandoc $osoptions --wrap=preserve --from markdown --to html --metadata=lang:en --include-in-header=$templates/header.pandoc --css=$templates/style.css -o $base.html $base.md + echo "$base.html from $f" + #else + # echo " $base.html up to date" + fi +done diff --git a/docs/multischnorr-20151012.pdf b/docs/multischnorr-20151012.pdf new file mode 100644 index 0000000..b8107fb Binary files /dev/null and b/docs/multischnorr-20151012.pdf differ diff --git a/docs/multisignature.md b/docs/multisignature.md new file mode 100644 index 0000000..3af5389 --- /dev/null +++ b/docs/multisignature.md @@ -0,0 +1,600 @@ +--- +title: + Multisignature +# katex +--- + +To do a Schnorr multisignature, you just list all the signatures that +went into it, and the test key is just adding the all the public keys +together. Which gives you pretty fast, though totally non anonymous, +voting, with no need for an elaborate, ingenious, and difficult to +understand key distribution system. + +Supposedly [Schnorr multisignature can be done on +ED25519](https://datatracker.ietf.org/meeting/99/materials/slides-99-cfrg-collective-edwards-curve-digital-signature-algorithm-draft-ford-cfrg-cosi-00-00) +, but I do not see any useful information on how to do it, and Libsodium +fails to mention Schnorr. + +However we can easily do it with ristretto25519. + +[Bernstein explains how to do Schnorr, and briefly visits Schnorr +multisignature.](./multischnorr-20151012.pdf) + +# Notation for elliptic curve cryptography + +The group operation where one obtains a third elliptic point from two elliptic +points is represented by addition, $A=B+C$, capitals represent members of the +group, (elliptic points, public keys), lower case represents integers modulo +the order of the group (scalars, private keys), and $h()$ is a hash function +that maps an arbitrary stream of arguments of arbitrary type to an integer +modulo the order of the group. + +An upper case letter represents the result of multiplying the base point of +the group by the value represented by the lower case letter. Thus if\ +$B_{base}$ is the base point then $K=kB_{base}\,$ + +$h(\cdots)$ is hash function that generates an integer modulo the order of the group (hence our use of $h$ rather than $H$) from an arbitrary stream of arguments of arbitrary type. It is hard to reverse, the only efficient way of reversing it being to guess all possible pre images from which it might be constructed.. + +# Single Schnorr Signature + +Signer secret key is $k$, his public key is $K$ + +## Sign message with $k$ + +- Generate an unpredictable $r$ from the message and the secret key. If we ever use the same $r$ with two +different values of $c$, we reveal +our public key, which is why we +make $r$ depend on the same message as we are signing. + + we could use some randomness to generate the konce (key used once, single use secret key), but this would mean that we would sign the same thing twice with two different signatures, which in a much signed item, such as a key or a Zooko id, is undesirable. + +- Compute $r = h(k,M)$, a per message secret scalar. Or $r$ can be an +unpredictable secret scalar, randomly generated, but we have to make sure it is truly random and never repeated, which is apt to fail because of computer determinism. +- Compute $R$, a per message elliptic point, a Konce, meaning an +elliptic point that can be used only once, never to be reused because +made public. +- Compute $c = h(R,$ Message$)$ +- Compute $s = r+c*k\,$\ +Note that:\ +$S= R+c*K$ + +signature is $c, s$ + +## Verify signature + +- Check that $c, s$ are valid scalars, which is trivial. +- Check that $K$, the signing public key, is a valid member of the +prime order group, which is not always trivial. +- Compute $R = S − c*K$\ +The reason the signer makes $s$ public rather than $S$, is that to prevent a signature from being faked by someone who does not know the secret underlying $S$. +- Check that $c = h(R,$ Message$)$ + +# Schnorr multisignature + +Obvious problem with multisig – you need proof that each putative signer +is able to produce a singlesig, hence no space advantage for konces, which is the major crypto currency use. but for voting on +the Paxos protocol, major advantage, since the singlesig is apt to be +durable and cached, so the signature signifying approval of the +blockchain is single Schnorr signature connected to a possibly large +collection of individual signing keys. To prove the multisig on a new +block is valid, you need to have and have checked a single sig for each +of the signatories, but it does not have to be a new singlesig. It can +be an old cached singlesig, so you don’t need to download and check +each of the singlesigs with each new block. You only need to check a +single signature, assuming you already have old singlesigs. + +So, Schnorr multisig, at least in the simple version, presupposes a +large existing data structure with good cryptographic properties, +presumably a Merkle-patricia tree, listing each party entitled to sign, +the self signed signature of that party, and the voting weight and +voting rights of that party, and the hash at the root of that tree is a +Schnorr multisig. + +This, of course, occupies slightly more space than the simple tuple of +signatures, but has the advantage that it does not need to change very +often, and usually only a small part of it needs to change, while the +multisig by itself is relatively small. + +But it is not that small. For each possible multisig, you have to list +the keys of all the signers, and then look up their rights in the Merkle +patricia tree, which is OK if the members of the board are hiring the +CEO, but is going to suck mightily if the shareholders are electing the +board. + +However, chances are in practice, it will commonly be the case that the +same group votes over and over again, so, if space and verification time +is a concern, you can record previous multisigs and their +verification in the Merkle tree, and record the signatures of the latest +signing as a delta on some past signature, and do the verification as a +delta on some past verification. + +Suppose you are doing the Paxos protocol, and every consensus block has +to be signed by one hundred peers, every few minutes. Well, you are +still going to have to verify the single sigs of the peers signing the +group signature, which gives you no savings in time or space over simple +tuple of singlesigs. You are still going to have to verify one hundred +peers. But if it is usually the same clique signing each each time, it +will take you very little time or space to prove that this consensus +hash was signed by the same clique as signed the other hundred consensus +hashes, which you verified a long time ago, and do not need to repeat +the verification. Of course their voting rights might have changed, but +if the cliques do not change much, their voting rights probably don’t +change much, so you don’t have to do full verification every single +time. + +If the latest sig is compatible with the current and older voting rights, +the sig should reference the older voting rights Merkle-patricia tree of +voting rights, and the peers should validate it on the basis of the +current voting rights, thus validating that the older voting right +suffices. This could produce a false signature and a fork if the voting +rights radically change, and old voters don’t like the change, but in +such a large scale public event, it will be highly visible and detected. +Each client wallet when talking a client wallet on another fork will +discover that its root is out of date. We implement default wallet +behavior to accept the root with the most recent voting rights list, and +of two roots with the same voting rights list, the root most recently +signed. + +## Generate Signature + +Kevin and Josephine have private keys $k_{Kevin}$ and $k_{Josephine}$, and public +keys $K_{Kevin}$ and $K_{Josephine}$, +and want to generate a multisig to verify to a verifier that both of them signed + +- Kevin generates an unpredictable scalar $r_{Kevin}$ which is never + twice the same, is random rather than deterministic as in single sig + (otherwise two multisig signings of the same message will reveal his + private key, since he cannot predict or control $c$)\ + Josephine similarly computes $r_{Josephine}\,$ +- Kevin shares $R_{Kevin}$ with each of the other +signatories, and they with him. +- Each of the signatories computes $R = R_{Kevin} +R_{Josephine} +\dots$ +- Each of the signatories computes $c = h(R,$ Message$)$ +- Kevin computes $s_{Kevin}=r_{Kevin} + c*k_{Kevin}$\ + Josephine computes $s_{Josephine}=r_{Josephine} + c*k_{Josephine}$, and + similarly each of the other signatories.\ + $S_{Kevin} = s_{Kevin}*B_{base} = R_{Kevin} + c*K_{Kevin}^{}$ +- Kevin shares $s_{Kevin}$ with each of the other signatories, and they with him + Each of the signatories computes\ + $s = s_{Kevin} + s_{Josephine} +\dots$\ + (thus Kevin cannot predict or control $s$, thus has to use a random + rather than deterministic konce) +- signature is $c, s$ + +## Verify Multsignature + +- Check that $c, s$ are valid scalars, which is trivial. +- Check that each of the signing keys is a valid member of the prime +order group, and a singlesig exists for each signing key of the multisig. +- $S=s*B_{base}\,$ +- Compute $R = S − c*(K_{Kevin} + K_{Josephine} +\dots )$ +- Check that $c = h(R,$ Message$)$, proving that $S = s*B = R + c*(K_{Kevin}+K_{Josephine}+\dots)$ + +Often the checker does not know nor care whether there are multiple +co-signers, because the co-signers have already generate an assertion that +$(K_{Kevin}+K_{Josephine}+\dots)$ is the public key, and hence does not +have to check the singlesig. + +If there are quite a few signers, this should be implemented as +server-client. + +Each of the clients submits the signing key and a konce (key used once), +server replies with\ +$K=\sum\limits_{client}K_{client}$, the sum of all signing keys\ +$R=\sum\limits_{client}R_{client}$, the sum of all konces\ +and $c=h(R,$ Message$)$, the value to be signed.\ +Each of the clients checks $c$ and returns $s_{client}$ +Server replies with\ +$s=\sum\limits_{client}s_{client}\,$ + +Each client checks the signature $c, s$ + +For a two party signature, this collapses to a simpler algorithm. Client +sends his konce. Server returns $s_{server}$, the sum of all signing keys, +and the sum of all konces. +Then client computes the signing key, returns it, and the server checks to +see if it works. Should be a three packet handshake. + +It is commonly the case that one party is the payee or beneficiary of +the thing being signed, an the other party does not particularly need or +want the signature, other than he needs to know it exists for his +records, and may need to resend the signing secret, accompanied by the +signing key and the identifier of the thing being signed. + +In this case, the payee sends the konce (key used once, the single use +secret), and the payee sends the konce, and does not need +the beneficiary’s konce, nor the completed signature, +and treats the transaction as completed, because he has seen the thing +being signed, and signed it. Doing it the other way around adds an extra +point of failure. + +For a large number of parties, you are going to have hold the protocol +open for a considerable time. + +For a vast number of parties, you are never going to get everyone to +complete the protocol all the way through, so have to use threshold +cryptography, where if everyone gets a enough information, even if they +do not get all of it, they can calculate the signature. + +# Threshold Signatures + +A threshold signature has the interesting feature that it is a randomness +beacon. If there is one honest participant party to the signing it generates a +random value unpredictable to all participants, This has obvious utility in +selecting the witness set and leader in blockdag algorithms equivalent to +Practical Byzantine Fault Tolerant Consensus, and in distributing discrete +shares in a way that is fair and on average over time approximates +continuous stake as closely as possible + +The participants sign off on assertion that their stake is such and such, and +the signature itself controls the random distribution of fractional voting +shares in the next signature as whole shares, so that voting shares, as +nearly as possible, approximate ownership shares. + +This is not in fact a useful discussion of Threshold signatures, so much as a +list of links. I don’t entirely trust myself to implement threshold +signatures. + +Suredbits describes the [FROST algorithm for Schnorr distributed key generation and signing]. +This algorithm is not robust. If any one of the +participants goes down in the middle you have to start all over, but it is +clean, simple, and easy to understand. + +[FROST algorithm for Schnorr distributed key generation and signing]:https://suredbits.com/schnorr-applications-frost/ +"Schnorr Applications: FROST" + +Each of the participants acts as the trusted dealer for his own share, hands +out shares in it to everyone else, and the final key is the sum of everyone's +share. + +This description is easily the most intelligible description of distributed +key generation and signature generation that I have read, because it is +adapted to the special case of Schnorr signatures, which have the huge +advantage of linearity. But, not robust. + +The practical limit for non robust schemes is about fifty participants. If +you have a hundred participants it keeps starting over and over again, and +eventually this causes more and more people to drop out in the middle as +they lose patience. + +[Practical Large Scale Distributed Key Generation]: +./PracticalLargeScaleDistributedKeyGeneration.pdf + +[Revisiting the Distributed Key Generation for Discrete-Log Based Cryptosystems]: +./threshold_shnorr.pdf + +[Practical Large Scale Distributed Key Generation] references +[Revisiting the Distributed Key Generation for Discrete-Log Based Cryptosystems] +which references Schnorr signatures. + +It is a follow up to the (unscalable and arguably inherently centralized) +[Secure Distributed Key Generation for Discrete-Log Based +Cryptosystems](./SecureDistributedKeyGeneration.pdf) + +The basic algorithm is that you generate a distributed key, from which +a subset of the key recipients can generate a Schnorr signature. + +[Revisiting the Distributed Key Generation for Discrete-Log Based +Cryptosystems](./threshold_shnorr.pdf) gives reasonably detailed instructions +for implementing threshold Schnorr, without any trusted central authority but +assumes various cryptographic primitives that we do not in fact have, among +them a reliable broadcast channel. + +## Reliable Broadcast Channel + +A key cryptographic primitive in threshold signatures, and indeed in almost +every group cryptographic protocol, is the reliable broadcast channel – that +any participant can reliably send a message that is available to all +participants. + +In actual practice we have unreliable point to point two party communications, +from which we have to construct a broadcast channel. + +Practical applications of these cryptographic protocols seem to be relying on +a trusted broadcaster, who is apt to be untrustworthy when there is money, +power, or valuable secrets lying on the table. + +Trouble is that in practice, certain messages are likely to be hidden from +certain participants, and other participants will be unaware that they are +hidden, or they will receive a discrepant message and incorrectly believe they +are receiving the same message as others. + +In a large group, one can assume that more than half the participants are +honest, so one can construct a broadcast channel using the Paxos protocol. + +Every distributed cryptographic protocol needs a secure broadcast +channel, and every blockchain is a secure broadcast channel. + +One of the requirements of secure reliable broadcast channel is that _it + stays up_. But a secure broadcast channel for a lightning type +transaction is going to be created and shut down. And if it can be +legitimately shut down, it can be shut down at exactly the wrong moment +for some of the participants and exactly the right time for some of the + participants. Hence the use of a “trusted” broadcast authority, who +stays up. + +We could attain the same effect by a hierarchy of secure reliable broadcast +channels, in which a narrow subchannel involving a narrower set of +participants can be set up on the broader channel, and shutdown, with its +final shutdown signature available in the broader channel, such that someone +who has the right code can find on the durable broadcast channel, the signature he needs. + +But interesting protocols are likely to involve small groups for which we want +the transaction to fail if any of the participants are defecting. + +For example, the lightning protocol is cryptographically enforced +correspondence banking, and an eternal problem in correspondence banking is +insider check kiting. A shill sends a check to another shill, so that one +correspondence banker can scam another correspondence banker, so the group +attempting to organize the transaction is going to consist of two shills, one +scammer, one pigeon, and one innocent third party roped in to obscure who is +doing the scamming and who is being scammed, giving a majority of three evil +participants against two good and trusting participants. + +By and large, the more money and power that is on the table, the smaller the +group engaging in the cryptographic protocol is apt to be. + +We want the transaction to fail in such cases. Generalizing to all +group cryptographic protocols, we want the broadcast channel to fail and +to be seen to fail in such cases. + +The Byzantine Paxos protocol is designed for a large group and is intended to +keep going permanently in the face of the hardware or software failure of some +participants, and Byzantine defection by a small conspiracy of participants. + +For a reliable broadcast channel to be reliable, you are relying on it to +stay up, because if it goes down and stays down, its state for transactions + near the time it went down cannot be clearly defined. + +For a reliable broadcast channel to be in a well defined state on shutdown, +it has to have continued broadcasting its final state to anyone interested +for some considerable time after it reached its final state. So you are +trusting someone to keep it going and available. In this sense, no group +cryptographic transaction can be entirely trustless. + +I intend that the rho blockchain will primarily a notarization system that +just happens to special case notarizing rhocoin transactions. The notaries +will be a collection of durable broadcast channels, each one typically +maintained by a single host, albeit a notarization will not be evidence usable +on the blockchain until its notary block is Merkle chained by the blockchain. +If the blockchain automatically trusts notary signatures, they will rapidly +cease to be trustworthy. The chain, not the signature, makes it officially +official. The notary signature and oid is merely a promise to make it +official. The blockchain will treat notaries as untrusted, so that everyone +else can treat them as trusted at low risk. + +## scaling + +According to [Practical Large Scale Distributed Key Generation](./PracticalLargeScaleDistributedKeyGeneration.pdf) their algorithm is of +order ${[log (n)]}^3$, meaning it should produce a Schnorr signature for +thousands of voters with hundreds of thousands of shares, in a a potentially +decentralized manner, without a trusted dealer, making it useful for digital +corporations, and capable of electing producing a chain of signatures +(shareholders sign the board, board signs the CEO, CEO signs every corporate +officer identity, CEO organizes the election of the board), capable of being +evaluated by everyone interacting with the business over the net. + +Obviously the reliable broadcast protocol of such a very large scale key +generation will look more like a regular blockchain, since many entities will +drop out or fail to complete directly. + +# [Blind SchnorrSignature.](https://www.math.uni-frankfurt.de/~dmst/teaching/WS2013/Vorlesung/Pointcheval,Stern.pdf) + +[See also](https://eprint.iacr.org/2019/877.pdf). + +[and](https://suredbits.com/schnorr-applications-blind-signatures/) + +Blind Schnorr signatures are vulnerable to the [Wagner attack], which +can be defeated by refusing to do large numbers of blind signatures in +parallel, and/or randomly failing to complete some blind signatures. + +[Wagner attack]:https://www.iacr.org/archive/crypto2002/24420288/24420288.pdf + +# Regular Elliptic Signature + +Signer secret scalar $k$. Signer public point $K=k*B$. $B$ is base point. + +Signature is scalar $s$ and point $R$, such that +$S = s*B = h(R,$ Message$)*K+ R$ + +## Signing + +Generate random secret scalar $r$, public point $R=r*B$ + +calculate public scalar $s = h(R$, Message)*k + r$ + +Reveal s and R for message. + +## [Blind signing using a regular elliptic signature](https://pdfs.semanticscholar.org/e58a/1713858a5b9355a9e18adfe3abfc05de244e.pdf) + +# Pairing based cryptography + +In pairing based cryptography the computational Diffie--Hellman problem +is believed to be infeasible while the simpler decisional +Diffie--Hellman problem can be easily solved using the pairing function. + +In a cryptographic group, given $G$, $xG$, and $yG$ it is hard to find +$xyG$, unless you know $x$ or $y$. + +In a pairing based group, given $G$, $xG$, $yG$ and $xyG$ you can +easily *verify* that $xyG$ is correct, even though you are unable to +calculate the correct value. + +Pairing based cryptography can do all sorts of really cool things, and I +am not sure how fast it can do them, but we do not in fact have any +obvious need for it. + +## Proposed uses for pairing based crypto + +We want money payments to operate in private invitation only messaging +and blog groups, without causing the group to be exposed. + +We also need pairing based crypto to interface between crypto currency, +and the corporate form, though threshold signatures. So we have a white +face, a grey face, and a black face. The white face is the interface +between crypto cash and to the corporate form that exists both +cryptographically and on government registries, the grey face is the +interface to the corporate form that exists only cryptographically, not +registered with the state (with the result that scams will abound) and +the black face is secret invitation only blogs and messaging groups, +within which it is possible to make secret payments. + +These invitation only blogs and messaging groups will exist with and +within open searchable blogs and messaging groups, hidden by the secret +handshake protocol. + +The structure will be ultimately rooted in [Zooko’s triangle](./zookos_triangle.html), but normal +people will most of the time sign in by zero knowledge password +protocol, your identity will be derivative from someone else’s Zooko +based identity. + +Useful links on this topic are “XDH assumption”, “pairing based +cryptography”, “Bilinear Diffie-Hellman”, and “gap Diffie—Hellman +(GDH) groups”. + +A list of libraries now +[available](https://gist.github.com/artjomb/f2d720010506569d3a39) PBC +looks like the best. MIRACL uses a strong copyleft license. AGPL: all +the software linked against free software (free in GNU/FSF sense) is +also free software and freely available. GNU Public License is the most +famous of such “strong copyleft” FOSS licenses. The GPL copyleft +clause triggers when an application is distributed outside of company +boundaries. And servicing customers from a company server running the +code constitutes distribution. + +MIRACL is written in C, and can be used from C, but is designed for C++. +Comes with inline C++ wrappers. + +But hell, I like C++. But C++ not going to fly on android. Scala does +not support Visual Studio, and visual studio does not really support +android, though it confidently believes that it does. + +Useful threshold signatures requires pairing based cryptography. And +pairing based cryptography also needed for useful implementation of the +secret handshake (green beard, Masonic Lodge) problem. Allegedly good +for zero knowledge password protocol, though I am pretty sure that +problem has been solved without using pairing based cryptography. + +Pairing based cryptography is usually described using the notation that +the group operation is multiplication, and the one way function +combining an integer with a member of the group to produce another +member of the group is exponentiation. But pairing based cryptography is +easier to understand if we call the group operation addition, whereupon +we call the application of an integer modulo the order of the group +multiplication, instead of calling it exponentiation. + +In pairing based cryptography, the group supports addition (it is +commutative and associative), and *also supports something like multiplication* +(it is also commutative and associative), albeit the result of multiplication +is not a member of the original group, but a member of another group, +the pair group. + +That it supports something like multiplication is described as “Bilinear +Diffie-Hellman”, and if you call it “something like multiplication” +people who are experts in the field will conclude you are an idiot. + +So, let us, as usual, use the notation that members of the group are capital +letters, and integers modulo the order of the group are lower case +italics. Let us denote members of the pair group, the result of +multiplying members of the original group with each other, with Greek letters. + +Which notation allows us to leave out all the academic stuff about $\forall P \in +G$. If it is a regular capital, it stands for any member of $G$, unless +otherwise specified. In proper academic language (apart from the fact +that I am leaving out all the $\forall P \in G$ stuff): + +Let G be an cyclic group of prime order written additively and $ϓ$ +another cyclic group of the same order written multiplicatively. A pairing is +a map: $e : G × G → ϓ$ , which satisfies the following properties: + +- Bilinearity: $\forall a, b, P, Q : e(aP, bQ) = e(P, Q)^{ab}$ +(notice that the left hand side of the equals sign is written +additively, and the right hand side written multiplicatively) +- Non-degeneracy: e ≠ 1 +- Computability: there exists an efficient algorithm to compute e + +Whereupon in this notation: +$B_{base}$, the base point, is widely known, Ann’s private key is $a$, her public key is $A = aB_{base}$ + +In C++ is is useful to represent both groups additively, allowing the +operation of the addition of any member of either group to another +member of the same group, the multiplication of any member of either +group by an integer, producing another member of the same group, and the +operation e(P,Q), where P and Q are members of the first group, as infix +multiplication producing a member of the second group. + +In this notation the magic equality becomes + +Bilinearity: $\forall a, b, P, Q :$ +$(a*P)*(b*Q) == (a*b)*(P*Q)$ + +Requiring us, in C++, to create an infix multiplication operator for the +mapping, an infix addition operator for each group, and an infix +multiplication by integer operator for each group. + +To sign a document with the secret key $a$, publish $M = $ah($Message$)B$ + +To test the signature, check that + +$A*(h(Message)B_{base})=B_{base}*M$ + +Which it should because $(a*B_{base})*(h($Message$)*B_{base}) = +B_{base}*(a*h($Message$)*B_{base})$ by bilinearity. + +The threshold variant of this scheme is called GDH threshold signature +scheme, Gap Diffie Hellman threshold signature scheme. + +This scheme is also good for blind signatures, because if you sign an +arbitrary point of the group, which the person asking for the signature +knows is sum of an already signed point of the group, multiplied by a +random secret, plus the thing he actually wants signed, he can then +subtract the two signed quantities to find a third signed quantity, +corresponding to a well formed token, that token unknown to the signer. + +## Secret Handshakes + +[Paraphrasing](./secret_handshakes.pdf) + +The Secret society of evil has a frequently changing secret key +$k_{evil}$ Ann has a secret key $k_{Ann}$ and public key $K_{Ann} = +k_{Ann}B$, Bob has a secret key $k_{Bob}$ and public key $K_{Bob} = +b_{Bob}B$ + +Let $h(…)$ represent a hash of the serialized arguments of H, which +hash is an integer modulo the order of the group. Let $H(…) = h(…)B$. +Streams are concatenated with a boundary marker, and accidental +occurrences of the boundary marker within a stream are escaped out. + +The evil overlord of the evil society of evil issues Ann and Bob a +signed hash of their public keys. For Ann, the signature is +$k_{evil}H($“Ann”, $K_{Ann},$ “evil secret society of evil”$)$, similarly for Bob + +The problem is that Ann does not know whether Bob is also a member of +the secret society of evil, and wants to send him a message that is only + going to be intelligible to him if he secretly has a key signed by the +secret society of evil, but is not going to be recognizable to third +parties as signed by the secret society of evil. + +So, they use as part of their shared secret whereby they encrypt +messages, the secret that Ann can calculate:\ $[k_{evil}H($“Ann”, +$K_{Ann},$ “evil secret society of evil”$)]*H($“Bob”, $K_{Bob},$ “evil +secret society of evil”$)$ + +Bob calculates: +$H($“Ann”, $K_{Ann},$ “evil secret society of evil”$) +*[k_{evil}H($“Bob”, $K_{Bob},$ “evil secret society of evil”$)]$ + +Thus proving possession of their evil keys to each other without +revealing them to each other, and without revealing that possession to +someone who does not have the expected evil key. + +Overly complicated and excessively clever. In practice, if you have a +secret society, you have a secret chatroom, in which case the routing +metadata on messages going to and from the chatroom are traceable if +they pass through enemy networks. + +But suppose you want battlespace iff (Identify Friend or Foe). You want to +send a message directly to an unknown target that will identify as friendly +if the other guy is a friendly, but not necessarily let him know enemy if he +is enemy. If he already knows you are enemy, you don’t want to give him +the means to identify enemy iff from your friends. diff --git a/docs/name_system.md b/docs/name_system.md new file mode 100644 index 0000000..863493e --- /dev/null +++ b/docs/name_system.md @@ -0,0 +1,350 @@ +--- +title: Name System +--- +We intend to establish a system of globally unique wallet names, to resolve +the security hole that is the domain name systm, though not all wallets will +have globally unique names, and many wallets will have many names. + +Associated with each globally unique name is set of name servers. When one’s +wallet starts up, then if your wallet has globally unique name, it logs in +to its name server, which will henceforth direct people to that wallet. If +the wallet has a network accessible tcp and/or UDP address it directs people +to that address (one port only, protocol negotiation will occur once the +connection is established, rather than protocols being defined by the port +number). If not, will direct them to a UDT4 rendevous server, probably itself. + +We probably need to support [uTP for the background download of bulk data]. +This also supports rendevous routing, though perhaps in a different and +incompatible way, excessively married to the bittorrent protocol.We might +find it easier to construct our own throttling mechanism in QUIC, +accumulating the round trip time and square of the round trip time excluding +outliers, to form a short term and long term average and variance of the +round trip time, and throttling lower priority bulk downloads and big +downloads when the short term average rises above the long term average by +more than the long term variance. The long term data is zeroed when the IP +address of the default gateway(router) is acquired, and is timed out over a +few days. It is also ceilinged at a couple of seconds. + +[uTP for the background download of bulk data]: https://github.com/bittorrent/libutp + +In this day and age, a program that lives only on one machine really is not +much of a program, and the typical user interaction is a user driving a gui +on one machine which is a gui to program that lives on a machine a thousand +miles away. + +We have a problem with the name system, the system for obtaining network +addresses, in that the name system is subject to centralized state control, +and the TCP-SSL system is screwed by the state, which is currently seizing +crimethink domain names, and will eventually seize untraceable crypto +currency domain names. + +In today’s environment, it is impossible to speak the truth under one’s true +name, and dangerous to speak the truth even under any durable and widely used +identity. Therefore, people who post under names tend to be unreliable. +Hence the term “namefag”. If someone posts under his true name, he is a +“namefag” – probably unreliable and lying. Even someone who posts under a +durable pseudonym is apt show excessive restraint on many topics. + +The aids virus does not itself kill you. The aids virus “wants” to stick +around to give itself lots of opportunities to infect other people, so wants +to disable the immune system for obvious reasons. Then, without a immune +system, something else is likely to kill you. + +When I say “wants”, of course the aids virus is not conscious, does not +literally want anything at all. Rather, natural selection means that a virus +that disables the immune system will have opportunities to spread, while a +virus that fails to disable the immune system only has a short window of +opportunity to spread before the immune system kills it, unless it is so +virulent that it likely kills its host before it has the opportunity to +spread. + +Similarly, a successful memetic disease that spreads through state power, +through the state system for propagation of official truth “wants” to disable +truth speaking and truth telling – hence the replication crisis, peer +review, and the death of science. We are now in the peculiar situation that +truth is best obtained from anonymous sources, which is seriously suboptimal. +Namefags always lie. The drug companies are abandoning drug development, +because science just does not work any more. No one believes their research, +and they do not believe anyone else’s research. + +It used to be that there were a small number of sensitive topics, and if you +stayed away from those, you could speak the truth on everything else, but now +it is near enough to all of them that it might as well be all of them, hence +the replication crisis. Similarly, the aids virus tends to wind up totally +suppressing the immune system, even though more selective shutdown would +serve its interests more effectively, and indeed the aids virus starts by +shutting down the immune system in a more selective fashion, but in the end +cannot help itself from shutting down the immune system totally. + +The memetic disease, the demon, does not “want” to shut down truth telling +wholesale. It “wants” to shut down truth telling selectively, but inevitably, +there is collateral damage, so it winds up shutting down truth telling +wholesale. + +To exorcise the demon, we need a prophet, and since the demon occupies the +role of the official state church, we need a true king. Since there is a +persistent shortage of true Kings, I here speaking as engineer rather than a +prophet, so here I am discussing the anarcho agorist solution to anarcho +tyranny, the technological solution, not the true king solution. + +Because of the namefag problem and the state snatching domain names, we need, +in order to operate an untraceable blockchain based currency not only a +decentralized system capable of generating consensus on who owns what cash, +we need a system capable of generating consensus on who who owns which human +readable globally unique names, and the mapping between human readable names, +Zooko triangle names (which correspond to encryption public keys), and +network addresses, a name system resistant to the state’s attempts to link +names to jobs, careers, and warm bodies that can be beaten up or imprisoned, +to link names to property, to property that can be confiscated or destroyed. + +A transaction output can hold an amount of currency, or a minimum amount of +currency and a name. Part of the current state, which every block contains, +is unused transaction outputs sorted by name. + +If we make unused transaction outputs sorted by name available, might as well +make them available sorted by key. + +In the hello world system, we will have a local database mapping names to +keys and to network addresses. In the minimum viable product, a global +consensus database. We will, however, urgently need a rendezvous system that +allows people to set up wallets and peers without opening ports on stable +network address to the internet. Arguably, the minimum viable product will +have a global database mapping between keys and names, but also a nameserver +system, wherein a host without a stable network address can login to a host +with a stable network address, enabling rendezvous. When one identity has its +name servers registered in the global consensus database, it always tries to +login to those and keep the connection alive with a ping that starts out +frequent, and then slows down on the Fibonacci sequence, to one ping every +1024 secondsplus a random number modulo 1024 seconds. At each ping, tells the +server when the next ping coming, and if the server does not get the +expected ping, server sends a nack. If the server gets no ack, logs the +client out. If the client gets no ack, retries, if still no ack, tries to +login to the next server. + +In the minimum viable product, we will require everyone operating a peer +wallet to have a static IP address and port forwarding for most functionality +to work, which will be unacceptable or impossible for the vast majority of +users, though necessarily we will need them to be able to receive money +without port forwarding, a static IP, or a globally identified human readable +name, by hosting their client wallet on a particular peer. Otherwise no one +could get crypto currency they would need to set up a peer. + +Because static IP is a pain, we should also support nameserver on the state +run domain name system, as well as nameserver on our peer network, but that +can wait a while. And in the end, when we grow so big that every peer is +itself a huge server farm, when we have millions of users and a thousand or +so peers, the natural state of affairs is for each peer to have a static IP. + +Eventually we want people to be able to do without static IPs and +portforwarding, which is going to require a UDP layer. One the other hand, we +only intend to have a thousand or so full peers, even if we take over and +replace the US dollar as the world monetary system. Our client wallets are +going to be the primary beneficiaries of rendevous UDT4.8 routing over UDP. + +We also need names that you can send money to, and name under which you can +receives. The current cryptocash system involves sending money to +cryptographic identifiers, which is a pain. We would like to be able to send +and receive money without relying on identifiers that look like line noise. + +So we need a system similar to namecoin, but namecoin relies on proof of +work, rather than proof of stake, and the state’s computers can easily mount +a fifty one percent attack on proof of work. We need a namecoin like system +but based on proof of stake, rather than proof of work, so that for the state +to take it over, it would need to pay off fifty one percent of the +stakeholders – and thus pay off the people who are hiding behind the name +system to perform untraceable crypto currency transactions and to speak the +unspeakable. + +For anyone to get started, we are going to have to enable them to operate a +client wallet without IP and port forwarding, by logging on to a peer wallet. +The minimum viable product will not be viable without a client wallet that +you can use like any networked program. A client wallet logged onto a peer +wallet automatically gets the name `username.peername`. The peer could give +the name to someone else though error, malice or equipment failure, but the +money will remain in the client’s wallet, and will be spendable when he +creates another username with another peer. Money is connected to wallet +master secret, which should never be revealed to anyone, not with the +username. So you can receive money with a name associated an evil nazi +identity as one username on one peer, and spend it with a username associated +with a social justice warrior on another peer. No one can tell that both +names are controlled by the same master secret. You send money to a username, +but it is held by the wallet, in effect by the master secret, not by the +user name. That people have usernames, that money goes from one username to +another, makes transferring money easy, but by default the money goes through +the username to the master secret behind the quite discardable username, +thus becomes anonymous, not merely pseudonymous after being received. Once +you have received the money, you can lose the username, throw it away, or +suffer it being confiscated by the peer, and you, not the username, still +have the money. You only lose the money if someone else gets the master +secret. + +You can leave the money in the username, in which case the peer hosting your +username can steal it, but for a hacker to steal it he needs to get your +master secret and logon password, or you transfer it to the master secret on +your computer, in which case a hacker can steal it, but the peer cannot, and +also you can spend it from a completely different username. Since most people +using this system are likely to be keen on privacy, and have no good reason +to trust the peer, the default will be for the money to go from the username +to the master secret. + +Transfers of money go from one username to another username, and this is +visible to the person who sent it and the person who received it, but if the +transfer is to the wallet and the master secret behind the username, rather +than to the username, this is not visible to the hosts. Money is associated +with a host and this association is visible, but it does not need to be the +same host as your username. By default, money is associated with the host +hosting the username that receives it, which is apt to give a hint to which +username received it, but you can change this default. If you are receiving +crypto currency under one username, and spending it under another username on +another host, it is apt to be a good idea to change this default to the host +that is hosting the username that you use for spending, because then spends +will clear more quickly. Or if both the usernames and both the hosts might +get investigated by hostile people, change the default to a host that is +hosting your respectable username that you do not use much. + +We also need a state religion that makes pretty lies low status, but that is +another post. + +# Mapping between globally unique human readable names and public keys + +The blockchain provides a Merkle-patricia dac of human readable names. Each +human readable name links to a list of signatures transferring ownership form +one public key to the next, terminating in an initial assignment of the name +by a previous block chain consensus. A client typically keeps a few leaves +of this tree. A host keeps the entire tree, and provides portions of the tree +to each client. + +When two clients link up by human readable name, they make sure that they are +working off the same early consensus, the same initial grant of user name by +an old blockchain consensus, and also off the same more recent consensus, +for possible changes in the public key that has rightful ownership of that +name. If they see different Merkle hashes at the root of their trees, the +connection fails. Thus the blockchain they are working from has to be the +same originally, and also the same more recently. + +This system ensures we know and agree what the public key associated with a +name is, but how do we find the network address? + +# Mapping between public keys and nework addresses + +## The Nameserver System + +Typically someone is logged in to a host with an identity that looks like an +email address, `paf.foo.bar`, where`bar` is the name of a host that is +reliably up, and reliably on the network, and relatively easy to find + +You can ask the host `bar` for the public key and *the network address* of +`foo.bar`, or conversely the login name and network address associated with +this public key. Of course these values are completely subject to the caprice +of the owner of `bar`. And, having obtained the network address of `foo.bar`, +you can then get the network address of `paf.foo.bar` + +Suppose someone owns the name `paf`, and you can find the global consensus as +to what public key controls `paf`, but he does not have a stable network +address. He can instead provide a nameserver – another entity that will +provide a rendevous. If `paf` is generally logged in to `foo`, you can +contact `foo`, to get rendevous data for `bar.foo`, which is, supposing `foo` +to be well behaved, rendevous data for `bar` + +Starting from a local list of commonly used name server names, keys, and +network addresses, you eventually get a live connection to the owner of that +public key, who tells you that at the time he received your message, the +information is up to date, and, for any globally unique human readable names +involved in setting up the connection, he is using the same blockchain as you +are using. + +Your local list of network addresses may well rapidly become out of date. +Information about network addresses flood fills through the system in the +form of signed assertions about network addresses by owners of public keys, +with timeouts on those assertions, and where to find more up to date +information if the assertion has timed out, but we do not attempt to create a +global consensus on network addresses. Rather, the authoritative source of +information about a network address of a public key comes from successfully +performing a live connection to the owner of that public key. You can, and +probably should, choose some host as the decider on the current tree of +network addresses, but we don’t need to agree on the host. People can work +off slightly different mappings about network addresses with no global and +complete consensus. Mappings are always incomplete, out of date, and usually +incomplete and out of date in a multitude of slightly different ways. + +We need a global consensus, a single hash of the entire blockchain, on what +public keys own what crypto currency and what human readable names. We do not +need a global consensus on the mapping between public keys and network +addresses. + +What you would like to get is an assertion that `paf.foo.bar` has public key +such and such, and whatever you need to make network connection to +`paf.foo.bar`, but likely `paf.foo.bar` has transient public key, because his +identity is merely a username and login at `foo.bar`, and transient network +address, because he is behind nat translation. So you ask `bar` about +`foo.bar`, and `foo.bar` about `paf.foo.bar`, and when you actually contact +`paf.foo.bar`, then, and only then, you know you have reliable information. +But you don’t know how long it is likely to remain reliable, though +`paf.foo.bar` will tell you (and no other source of information is +authoritative, or as likely to be accurate). + +Information about the mapping between public keys and network addresses that +is likely to be durable flood fills through the network of nameservers. + +# logon identity + +Often, indeed typically, `ann.foo` contacts `bob.bar`, and `bob.bar` needs +continuity information, needs to know that this is truly the same `ann.foo` +as contacted him last time – which is what we currently do with usernames and +passwords. + +The name `foo` is rooted in a chain of signatures of public keys and requires +a global consensus on that chain. But the name `ann.foo` is rooted in logon +on `foo`. So `bob.bar` needs to know that `ann.foo` can log on with `foo`, +which `ann.foo` does by providing `bob.bar` with a public key signed by `foo`, +which might be a transient public key generated the last time she logged +on, which will disappear the moment her session on her computer shuts down, +or might be a durable public key. But if it is a durable public key, this +does not give her any added security, since `foo` can always make up a new +public key for anyone he decides to call `ann.foo` and sign it, so he might +as well put a timeout on the key, and `ann.foo` might as well discard it when +her computer turns off or goes into sleep mode. So, it is in everyone’s +interests (except that of attackers) that only root keys are durable. + +`foo`’s key is durable, and information about it is published.`ann.foo`’s +key is transient, and information about it always obtained directly from +`ann.foo` as a result of `ann.foo` logging in with someone, or as a result of +someone contacting `foo` with the intent of logging in to `ann.foo`. + +But suppose, as is likely, the network address of `foo` is not actually all +that durable, is perhaps behind a NAT. In that case, it may well be that to +contact `foo`, you need to contact `bar`. + +So, `foo!bar` is `foo` logged in on `bar`, but not by a username and +password, but rather logged on by his durable public key, attested by the +blockchain consensus. So, you get an assertion, flood filled through the +nameservers, that the network address of the public key that the blockchain +asserts is the rightful controller of `foo`, is likely to be found at `foo!` +(public key of `bar`), or likely to be found at `foo!bar`. + +Logons by durable public key will work exactly like logons by username and +password, or logons by derived name. It is just that the name of the entity +logged on has a different form.. + +Just as openssh has logons by durable public key, logons by public key +continuity, and logons by username and password, but once you are logged on, +it is all the same, you will be able to logon to `bob.bar` as `ann.bob.bar`, +meaning a username and password at `bob.bar`, as `ann.foo`, meaning `ann` has +a single signon at `foo`, a username and password at `foo`, or as `ann`, +meaning `ann` logs on to `bob.bar` with a public key attested by the +blockchain consensus as belonging to `ann`. + +And if `ann` is currently logged on to `bob.bar` with a public key attested +by the blockchain consensus as belonging to `ann`, you can find the current +network address of `ann` by asking `bob.bar` for the network address of +`ann!bob.bar` + +`ann.bob.bar` is whosoever `bob.bar` decides to call `ann.bob.bar`, but +`ann!bob.bar` is an entity that controls the secret key of `ann`, who is at +this moment logged onto `bob.bar`. + +If `ann` asserts her current network address is likely to last a long time, +and is accessible without going through + +`bob.bar` then that network address information will flood fill through the +network. Less useful network address information, however will not get far. diff --git a/docs/number_encoding.md b/docs/number_encoding.md new file mode 100644 index 0000000..5f8e23a --- /dev/null +++ b/docs/number_encoding.md @@ -0,0 +1,266 @@ +--- +lang: en +title: Number encoding +--- + +# The problem to be solved + +As computers and networks grow, any fixed length fields +in protocols tend to become obsolete. Therefore, for future +upwards compatibility, we want to have variable precision +numbers. + +## Use case + +QR codes and prefix free number encoding is useful in cases where we want data to be self describing – this bunch of bits is to be interpreted in a certain way, used in a certain action, means one thing, and not another thing. At present there is no standard for self description. QR codes are given meanings by the application, and could carry completely arbitrary data whose meaning and purpose comes from outside, from the context. + +Ideally, it should make a connection, and that connection should then launch an interactive environment – the url case, where the url downloads a javascript app to address a particular database entry on a particular host. + +A fixed length field is always in danger of +running out, so one needs a committee to allocate numbers. +With an arbitrary length field there is always plenty of +headroom, we can just let people use what numbers seem good +to them, and if there is a collision, well, one or both of +the colliders can move to another number. + +For example, the hash of a public key structure has to contain an algorithm +identifier as to the hashing algorithm, to accommodate the possibility that +in future the existing algorithm becomes too weak, and we must introduce +new algorithms while retaining compatibility with the old. But there could +potentially be quite a lot of algorithms, though in practice initially there +will only be one, and it will be a long time before there are two. + +When I say "arbitrarily large" I do not mean arbitrarily large, since this creates the possibility that someone could break something by sending a number bigger than the software can handle. There needs to be an absolute limit, such as sixty four bits, on representable numbers. But the limit should be larger than is ever likely to have a legitimate use. + +# Solutions + +## Zero byte encoding + + Capt' Proto zero compresses out zero bytes, and uses an encoding such that uninformative and predictable fields are zero. + +## 62 bit compressed numbers + + QUIC expresses a sixty two bit number as one to four sixteen bit numbers. This is the fastest to encode and decode. + +## Leading bit as number boundary + +But it seems to me that the most efficient reasonably fast and elegant +solution is a variant on utf8 encoding, though not quite as fast as the +encoding used by QUIC: + +Split the number into seven bit fields. For the leading fields, a one bit is +prepended making an eight bit byte. For the last field, a zero bit is prepended. + +This has the capability to represent very large values, which is potentially +dangerous. The implementation has to impose a limit, but the limit can +be very large, and can be increased without breaking compatibility, and +without all implementations needing to changing their limit in the same +way at the same time. + +## Prefix Free Number Encoding + +In this class of solutions, numbers are embedded as variable sized groups of bits within a bitstream, in a way that makes it possible to find the boundary between one number and the next. It is used in data compression, but seldom used in compressed data transmission, because far too slow. + +This class of problem is that of a +[universal code for integers](http://en.wikipedia.org/wiki/Universal_code_%28data_compression%29). + +The particular coding I propose here is a variation on +Elias encoding, though I did not realize it when I +invented it. + +On reflection, my proposed encoding is too clever by half, +better to use Elias δ coding, with large arbitrary +limits on the represented numbers, rather than +clever custom coding for each field. For the intended purpose of wrapping packets, of collecting UDP packets into messages, and messages into channels, limit the range of representable values to the range j: 0 \< j \< 2\^64, and pack all the fields representing the place of this UDP package in a bunch of messages in a bunch of channels into a single bitstream header that is then rounded into an integral number of bytes.. + +We have two bitstream headers, one of which contains always starts with the number 5 to identify the protocol. (Unknown protocols immediately ignored), and then another number to identify the encryption stream and the position in the encryption stream (no windowing). Then we decrypt the rest of the packet starting on a byte boundary. The decrypted packet then has additional bitstream headers. + +For unsigned integers, we restrict the range to less than 2\^64-9. We then add 8 before encoding, and subtract 8 after encoding, so that our Elias δ encoded value always starts with two zero bits, which we always throw away. Thus the common values 0 to 7 inclusive are represented by a six bit value – I want to avoid wasting too much implied probability on the relatively low probability value of zero. + +The restriction on the range is apt to produce unexpected errors, so I suppose we special case the additional 8 values, so that we can represent every signed integer. + +For signed integers, we convert to an unsigned integer\ +`uint_fast64_t y; y= 2*((uint_fast64_t)(-x)+1) : 2*(uint_fast64_t)x;`\ +And then represent as a positive integer. The decoding algorithm has to know whether to call the routine for signed or unsigned. By using unsigned maths where values must always be positive, we save a bit. Which is a lot of farting around to save on one bit. + +We would like a way to represent an arbitrarily large +number, a Huffman style representation of the +numbers.  This is not strictly Huffman encoding, +since we want to be able to efficiently encode and decode +large numbers, without using a table, and we do not have +precise knowledge of what the probabilities of numbers are +likely to be, other than that small numbers are +substantially more probable than large numbers.  In +the example above, we would like to be able to represent +numbers up to O(2^32^), but efficiently represent +the numbers one, and two, and reasonably efficiently +represent the numbers three and four.  So to be +strictly correct, “prefix free number encoding”. As we +shall see at the end, prefix free number encoding always +corresponds to Huffman encoding for some reasonable weights +– but we are not worrying too much about weights, so are +not Huffman encoding. + +###Converting to and from the representation + +Assume X is a prefix free sequence of bit strings – that is to say, if we +are expecting a member of this sequence, we can tell where the member +ends.  + +Let \[m…n\] represent a sequence of integers m to n-1.  + +Then the function X→\[m…n\] is the function that converts a bit string of X +to the corresponding integer of \[m…n\], and similarly for \[m…n\]→X.  + +Thus X→\[m…n\] and \[m…n}→X provide us with a prefix free representation of +numbers greater than or equal to m, and less than n.  + +Assume the sequence X has n elements, and we can generate and recognize +each element.  + +Let ℓ(X,k) be a new sequence, constructed by taking the first element of +X, and appending to it the 2^k^ bit patterns of length i, the +next element of X and appending to it the 2^k+1^ bit patterns of +length k+1, and so on and so forth.  + +ℓ is a function that gives us this new sequence from an existing sequence +and an integer.  + +The new sequence ℓ(X,k) will be a sequence of prefix free bit patterns +that has 2^n+k+1^ - 2^k^ elements.  + +We can proceed iteratively, and define a sequence ℓ(ℓ(X,j),k), which class +of sequences is useful and efficient for numbers that are typically quite +small, but could often be very large. We will more precisely +prescribe what sequences are useful and efficient for what purposes when +we relate our encoding to Huffman coding. + +To generate the m+1[th]{.small} element of ℓ(X,k), where X is a +sequence that has n elements: + +Let j = m + 2^k^ + +Let p = floor(log~2~(j)) that is to say, p is the position of +the high order bit of j, zero if j is one, one if j is two +or three, two if j is four, five, six, or seven, and so on and so forth. + +We encode p into its representation using the encoding \[k…n+k\]→X, and +append to that the low order p bits of j. + +To do the reverse operation, decode from the prefix free representation to +the zero based sequence position, to perform the function ℓ(X,k)→\[0…2^n+k+1^-2^k^\], +we extract p from the bit stream using the decoding of X→\[j…n+j\], then +extract the next p bits of the bit stream, construct k from 2^p^-2^j^ +plus the number represented by those bits. + +Now all we need is an efficient sequence X for small numbers.  + +Let ℒ(n) be a such a sequence with n values. \ +The first bit pattern of ℒ(n) is 0\ +The next bit pattern of ℒ(n) is 10\ +The next bit pattern of ℒ(n) is 110\ +The next bit pattern of ℒ(n) is 1110\ +…\ +The next to last bit pattern of ℒ(n) is 11…110, containing n-2 one bits +and one zero bit.\ +The last bit pattern of ℒ(n) breaks the sequence, for it is 11…11, +containing n-1 one bits and no zero bit. + +The reason why we break the sequence, not permitting the +representation of unboundedly large numbers, is that +computers cannot handle unboundedly large numbers – one +must always specify a bound, or else some attacker will +cause our code to crash, producing results that we did not +anticipate, that the attacker may well be able to make use +of. + +Perhaps a better solution is to waste a bit, thereby +allowing future expansion. We use a representation +that can represent arbitrarily large numbers, but clients +and servers can put some arbitrary maximum on the size of +the number. If that maximum proves too low, future clients +can just expand it without breaking backward compatibility. +This is similar to the fact that different file systems +have different arbitrary maxima for the nesting of +directories, the length of paths, and the length of +directory names. Provided the maxima are generous +it does not matter that they are not the same. + +Thus the numbers 1 to 2 are represented by \[1…3\] → +ℒ(2), 1 being the pattern “0”, and 2 being the +pattern “1” + +The numbers 0 to 5 are represented by \[0…6\] → ℒ(6), being the patterns\ +“0”, “10”, “110”, “1110”, “11110”, “11111” + +Thus \[0…6\] → ℒ(6)(3) is a bit pattern that represents the number +3, and it is “1110” + +This representation is only useful if we expect our numbers +to be quite small. + +\[0…6\] → ℓ(ℒ(2),1) is the sequence “00”, “01”, +“100”, “101”, “110”, “111” representing the +numbers zero to five, representing the numbers 0 to +less than 2^2+1^ – 2^1^ + +\[1…15\] → ℓ(ℒ(3),1) is similarly the sequence\ +“00”, “01”,\ +“1000”, “1001”, “1010 1011”,\ +“11000”, “11001”, “11010”, “11011”,“11100”, “11101”, “11110”, “11111”,\ +representing the numbers one to fourteen, representing the +numbers 1 to less than 1 + 2^3+1^ – 2^1^ + +We notice that ℓ(ℒ(n),k) has 2^n+k^ – 2^k^ +patterns, and the shortest patterns are length 1+k, and the +largest patterns of length 2n+k-2 + +This representation in general requires twice as many bits +as to represent large numbers as the usual, non self +terminating representation does (assuming k to be small) + +We can iterate this process again, to get the bit string sequence:\ +ℓ(ℓ(ℒ(n),j),k)\ +which sequence has 2\^(2^n+j^ - 2^j^ + k) - 2^k^ +elements.  + +This representation is asymptotically efficient for very +large numbers, making further iterations pointless. + +ℓ(ℒ(5),1) has 62 elements, starting with a two bit pattern, and ending +with a nine bit pattern. Thus ℓ(ℓ(ℒ(5),1),2) has +2^64^-4 elements, starting with a four bit pattern, and finishing +with a 72 bit pattern.  + +### prefix free encoding as Huffman coding + +Now let us consider a Huffman representation of the +numbers when we assign the number `n` the +weight `1/(n*(n+1)) = 1/n – 1/(n+1)` + +In this case the weight of the numbers in the range `n ... m` is `1/n – 1/(m+1)` + +So our bit patterns are:\ +0 (representing 1)\ +100 101 representing 2 to 3\ +11000 11001 11010 11011 representing 4 to 7\ +1110000 1110001 1110010  1110011 1110100 1110101 +1110110 1110111 representing 8 to 15 + +We see that the Huffman coding of the numbers that are +weighted as having probability `1/(n*(n+1))` + +Is our friend \[1…\] → ℓ(ℒ(n),0), where n is very large. + +Thus this is good in a situation where we are quite unlikely to encounter +a big number.  However a very common situation, perhaps the most +common situation, is that we are quite likely to encounter numbers smaller +than a given small amount, but also quite likely to encounter numbers +larger than a given huge amount – that the probability of encountering a +number in the range 0…5 is somewhat comparable to the probability of +encountering a number in the range 5000…50000000. + +We want an encoding that corresponds to a Huffman encoding where numbers are logarithmically distributed up to some enormous limit, corresponding to an encoding where for all n, n bit numbers are represented with an only slightly larger number of bits, n+O(log(n)) bits. + +In such case, we should we should represent such values by members of a +prefix free sequence `ℓ(ℓ(ℒ,j),k)` diff --git a/docs/pandoc_templates/after.pandoc b/docs/pandoc_templates/after.pandoc new file mode 100644 index 0000000..c902203 --- /dev/null +++ b/docs/pandoc_templates/after.pandoc @@ -0,0 +1 @@ +

Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 International License.

diff --git a/docs/pandoc_templates/before.pandoc b/docs/pandoc_templates/before.pandoc new file mode 100644 index 0000000..ccec396 --- /dev/null +++ b/docs/pandoc_templates/before.pandoc @@ -0,0 +1 @@ +

To Home page

diff --git a/docs/pandoc_templates/header.pandoc b/docs/pandoc_templates/header.pandoc new file mode 100644 index 0000000..6350b9a --- /dev/null +++ b/docs/pandoc_templates/header.pandoc @@ -0,0 +1,21 @@ + + + \ No newline at end of file diff --git a/docs/pandoc_templates/style.css b/docs/pandoc_templates/style.css new file mode 100644 index 0000000..ce7d496 --- /dev/null +++ b/docs/pandoc_templates/style.css @@ -0,0 +1,31 @@ +body { + max-width: 30em; + margin-left: 1em; + } +p.center {text-align:center; + } + +table { + border-collapse: collapse; + } +td, th { + border: 1px solid #999; + padding: 0.5rem; + text-align: left; + } +code{white-space: pre-wrap; + } +span.smallcaps{font-variant: small-caps; + } +span.underline{text-decoration: underline; + } +div.column{display: inline-block; vertical-align: top; width: 50%; + } +div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em; + } +ul.task-list{list-style: none; + } +.display.math{display: block; text-align: center; margin: 0.5rem auto; + } +h1.title{text-align: center; font-size: xxx-large; +} \ No newline at end of file diff --git a/docs/parsers.md b/docs/parsers.md new file mode 100644 index 0000000..0200eb9 --- /dev/null +++ b/docs/parsers.md @@ -0,0 +1,695 @@ +--- +title: Parsers +--- +This rambles a lot. Thoughts in progress: Summarizing my thoughts here at the top. + +Linux scripts started off using lexing for parsing, resulting in complex and +incomprehensible semantics, producing unexpected results. (Try naming a +file `-r`, or a directory with spaces in the name.) + +They are rapidly converging in actual usage to the operator precedence +syntax and semantics\ +`command1 subcommand arg1 … argn infixoperator command2 subcommand …` + +Which is parsed as\ +`((staticclass1.staticmethod( arg1 … argn)) infixoperator ((staticclass2.staticmethod(…)))` + +With line feed acting as `}{` operator, start of file acting as a `{` operator, end +of file acting as a `}` operator, suggesting that in a sane language, indent +increase should act as `{` operator, indent decrease should act as a `}` +operator. + +Command line syntax sucks, because programs interpret their command +lines using a simple lexer, which lexes on spaces. Universal resource +identifier syntax sucks, because it was originally constructed so that it +could be a command line argument, hence no spaces, and because it was +designed to be parsed by a lexer. + +But EBNF parsers also suck, because they do not parse the same way +humans do. Most actual programs can be parsed by a simple parser, even +though the language in principle requires a more powerful parser, becaus +humans do not use the nightmarish full power of a grammer that an EBNF +definition winds up defining. + +Note that [LLVM language creation tools](https://llvm.org/docs/tutorial/MyFirstLanguageFrontend/) +tutorial does not user an EBNF +parser. These tools also make creating a new language with JIT semantics +very easy. + +We are programming in languages that are not parsed the way the +programmer is parsing them. + +Programming languages ignore whitespace, because programmers tend to +express their meaning with whitespace for the human reader, and +whitespace grammer is not altogether parallel to the EBNF grammer. +There is a mismatch in grammers. + +Seems to me that human parsing is combination of low level lexing, Pratt +parsing on operator right and left binding power, and a higher level of +grouping that works like lexing. Words are lexed by spaces and +punctuation, grouped by operator binding power, with operator +recognition taking into account the types on the stack, groups of parsed +words are bounded by statement separators, which can be lexed out, +groups of statements are grouped and bounded by indenting. + +Some levels in the hierarchy are lexed out, others are operator binding +power parsed out. There are some “operators” that mean group separator +for a given hierarchical level, which is a tell that reveals lex style parsing, +for example semi colon in C++, full stop and paragraph break in text. + +The never ending problems from mixing tab and spaces indenting can be +detected by making a increase or decrease of ident by a space a bracket +operator, and an increase or decrease by a tab a non matching bracket +operator. + +Pratt parsing parsers operators by their left and right binding power – +which is a superset of operator precedence parsing. EBNF does not +directly express this concept, and programming this concept into EBNF is +complicated, indirect, and imperfect – because it is too powerful a +superset, that can express anything, including things that do not make +sense to the human writing the stuff to be parsed. + +Pratt parsing finalizes an expression by visiting the operators in reverse +polish order, thus implicitly executing a stack of run time typed operands, +which eventually get compiled and eventually executed as just-in-time typed +or statically typed operands and operators. + +For [identity](identity.html), we need Cryptographic Resource Identifiers, +which cannot conform the “Universal” Resource Identifier syntax and semantics. + +Lexers are not powerful enough, and the fact that they are still used +for uniform resource identifiers, relative resource identifiers, and command +line arguments is a disgrace. + +Advanced parsers, however, are too powerful, resulting in syntax that is +counter intuitive. That ninety percent of the time a program file can be +parsed by a simple parser incapable of recognizing the full set of +syntactically correct expressions that the language allows indicates that the +programmer’s mental model of the language has a more simple structure. + +# Pratt Parsing + +I really love the Pratt Parser, because it is short and simple, because if you +add to the symbol table you can add new syntax during compilation, +because what it recognizes corresponds to human intuition and human +reading. + +But it is just not actually a parser. Given a source with invalid expressions +such as unary multiplication and unbalanced parentheses, it will cheerfully +generate a parse. It also lacks the concept out of which all the standard +parsers are constructed, that expressions are of different kinds, different +nonterminals. + +To fix Pratt parsing, it would have to recognize operators as bracketing, as +prefix undery, postfix unary, or infix, and that some operators do not have an +infix kinds, and it would have to recognize that operands have types, and that +an operator produces a type from its inputs. It would have to attribute a +nonterminal to a subtree. It would have to recognize ternary operators as +operators. + +And that is a major rewrite and reinvention. + +Lalr parsers appear to be closer to the programmer mental model, but looking at +Pratt Parsing, there is a striking resemblance between C and what falls out +Pratt’s model: + +The kind of “lexing” the Pratt parser does seems to have a natural +correspondence to the kind of parsing the programmer does as his eye rolls +over the code. Pratt’s deviations from what would be correct behavior in +simple arithmetic expressions composed of numerals and single character +symbols seem to strikingly resemble expressions that engineers find +comfortable. + +When `expr` is called, it is provided the right binding power of the +token that called it. It consumes tokens until it meets a token whose left +binding power is equal or lower than the right binding power of the operator +that called it. It collects all tokens that bind together into a tree before +returning to the operator that called it. + +The Pratt `peek` peeks ahead to see if what is coming up is an +operator, therefore needs to check what is coming up against a symbol table, +which existing implementations fail to explicitly implement. + +The Pratt algorithm, as implemented by Pratt and followers, assumes that all +operators can be unary prefix or infix (hence the nud/led distinction). It +should get the nature of the upcoming operator from the symbol table (infix, +unary, or both, and if unary, prefix or postfix. + +Although implementers have not realized it, they are treating all “non +operator” tokens as unary posfix operators. Instead of, or as well as, they +need to treat all tokens (where items recognized from a symbol table are +pre-aggregated) as operators, with ordinary characters as postfix unary, +spaces as postfix unary with weaker binding power, and a token consisting of +a utf8 iterator plus a byte count as equivalent to a left tree with right +single character leaves and a terminal left leaf. + +Pratt parsing is like lexing, breaking a stream of characters into groups, +but the grouping is hierarchical. The algorithm annotates a linear text with +hierarchy. + +Operators are characterized by a global order of left precedence, a global +order of right precedence (the difference giving us left associativity and +right associativity) + +If we extend the Pratt algorithm with the concept of unitary postfix +operators, we see it is treating each ordinary unrecognized character as a +unitary postfix operator, and each whitespace character as a unitary postfix +operator of weaker binding power. + +[Apodaca]:https://dev.to/jrop/pratt-parsing + +Pratt and [Apodaca] are primarily interested in the case of unary minus, so +they handle the case of a tree with a potentially null token by +distinguishing between nud (no left context) and led (the right hand side of +an operator with left context). + +Pratt assumes that in correct source text, `nud` is only going to encounter an +atomic token, in which case it consumes the token, constructs a leaf vertex +which points into the source, and returns, or a unary prefixoperator, or an +opening bracket. If it encounters an operator, it calls `expr` with the right +binding power of that operator, and when `expr`has finished parsing, returns +a corresponding vertex. + +Not at all clear to me how it handles brackets. Pratt gets by without the +concept of matching tokens, or hides it implicitly. Seems to me that correct +parsing is that a correct vertex has to contain all matching tokens, and the +expressions cotained therein, so a vertex corresponding to a bracketed +expression has to point to the open and closing bracket terminals, and the +contained expression. I would guess that his algorithm winds up with a +tree that just happens to contain matching tokens in related positions in the tree. + +Suppose the typical case, a tree of binary operators inside a tree of binary +operators: In that case, when `expr` is called, the source pointer is pointing +to the start of an expression. `expr` calls `nud` to parse the expression, and if +that is all she wrote (because ` peek` reveals an operator with lower left +binding power than the right binding power that `expr` was called with) +returns the edge to the vertext constructed by `nud`. Otherise, it parses out +the operator, and calls `led` with the right binding power of the operator it has encountered, to get the right hand argument of the binary operator. It +then constructs a vertex containing the operator, whose left edge points to +the node constructed by `nud` and whose right hand edge points to the node +constructed by `led`. If that is all she wrote, returns, otherwise iterates +its while loop, constructing the ever higher root of a right leaning tree of +all previous roots, whose ultimate left most leaf is the vertex constructed by +`nud`, and whose right hand vertexes were constructed by `led`. + +The nud/led distinction is not sufficiently general. They did not realize +that they were treating ordinary characters as postfix unitary operators. + +Trouble is, I want to use the parser as the lexer, which ensures that as the +human eye slides over the text, the text reads the way it is in fact +structured. But if we do Pratt parsing on single characters to group them +into larger aggregates, `p*--q*s` is going to be misaggregated by the parser +to `(( (p*)−) − (q*s)`, which is meaningless. + +And, if we employ Pratt’s trick of nud/led distinction, will evaluate as +`p*(-(-q*s))` which gives us a meaningful but wrong result ` p*q*s` + +If we allow multicharacter operators then they have to be lexed out at the +earliest stage of the process – the Pratt algorithm has to be augmented by +aggregate tokens, found by attempting to the following text against a symbol +table. Existing Pratt algorithms tend to have an implicit symbol table of +one character symbols, everything in the symbol table being assumed to be +potentially either infix or unary prefix, and everything else outside the +implicit symbol table unary postfix. + +If we extend the Pratt algorithm with the concept of unitary postfix +operators, we see it is treating each ordinary unrecognized character as a +unitary postfix operator, and each whitespace character as a unitary postfix +operator of weaker binding power. + +Suppose a token consists of a utf8 iterator and a byte count. + +So, all the entities we work with are trees, but recursion terminates because +some nodes of the tree have been collapsed to variables that consist of a +utf8 iterator and a byte count, *and some parts of the tree have been +partially collapsed to vertexes that consist of a ut8 iterator, a byte count, +and an array of trees*. + +C++ forbids `“foo bar()”` to match `“foobar()”`, but +allows `“foobar ()”` to match, which is arguably an error. + +`“foobar(”` has to lex out as a prefix operator. But is not +really a prefix unitary operator. It is a set of matching operators, like +brackets and the tenary operatro bool?value:value The commas and the closing +bracket are also part of it. Which brings us to recognizing ternary +operators. The naive single character Pratt algorithm handles ternary +operators correctly (assuming that the input text is valid) which is +surprising. So it should simply also match the commas and right bracket as a +particular case of ternary and higher operators in the initial symbol search, +albeit doing that so that it is simple and correct and naturally falls out of +the algorithm is not necessarily obvious. + +Operator precedence gets you a long way, but it messed up because it did not +recognize the distinction between right binding power and left binding +power. Pratt gets you a long way further. + +But Pratt messes up because it does not explicitly recognize the difference +between unitary prefix and unitary postfix, nor does it explicitly recognize +operator matching – that a group of operators are one big multi argument +operator. It does not recognize that brackets are expressions of the form +symbol-expression-match, let alone that ternary operators are expressions of +the form expression-symbol-match-expression. + +Needs to be able to recognize that expressions of the form +expression-symbol-expression-match-expression-match\...expression are +expressions, and convert the tree into prefix form (polish notation with +arguments bracketed) and into postfix form (reverse polish) with a count of +the stack size. + +Needs to have a stack of symbols that need left matches. + +# Lalr + +Bison and yacc are +[joined at the hip](https://tomassetti.me/why-you-should-not-use-flex-yacc-and-bison/) to seven bit ascii and BNF, (through flex and lex) +whereas [ANTLR](https://tomassetti.me/ebnf/) +recognizes unicode and the far more concise and intelligible EBNF. ANTLR +generates ALL parsers, which allow syntax that allows statements that are ugly +and humanly unintelligible, while Bison when restricted to LALR parsers allows +only grammars that forbid certain excesses, but generates unintelligible error +messages when you specify a grammar that allows such excesses. + +You could hand write your own lexer, and use it with BisonC++. Which seemingly +everyone does. + +ANTLR allows expressions that take long time to parse, but only polynomially +long, fifth power, and prays that humans seldom use such expressions, which in +practice they seldom do. But sometimes they do, resulting in hideously bad +parser performance, where the parser runs out of memory or time. Because +he parser allows non LALR syntax, it may find many potential meanings +halfway through a straightforward lengthy expression that is entirely clear +to humans because the non LALR syntax would never occur to the human. In +ninety percent of files, there is not a single expression that cannot be +parsed by very short lookahead, because even if the language allows it, +people just do not use it, finding it unintelligible. Thus, a language that +allows non LALR syntax locks you in against subsequent syntax extension, +because the extension you would like to make already has some strange and non +obvious meaning in the existing syntax. + +This makes it advisable to use a parser that can enforce a syntax definition +that does not permit non LALR expressions. + +On the other hand, LALR parsers walk the tree in Reverse Polish +order, from the bottom up. This makes it hard to debug your grammar, and +hard to report syntax errors intelligibly. And sometimes you just cannot +express the grammar you want as LALR, and you wind up writing a superset of +the grammar you want, and then ad-hoc forbidding otherwise legitimate +constructions, in which case you have abandoned the simplicity and +directness of LALR, and the fact that it naturally tends to restrict you to +humanly intelligible syntax. + +Top down makes debugging your syntax easier, and issuing useful error +messages a great deal easier. It is hard to provide any LALR handling of +syntax errors other than just stop at the first error, but top down makes it +a lot harder to implement semantics, because Reverse Polish order directly +expresses the actions you want to take in the order that you need to take +them. + +LALR allows left recursion, so that you can naturally make minus and divide +associate in the correct and expected order, while with LL, you wind up +doing something weird and complicated – you build the tree, then you have +another pass to get it into the correct order. + +Most top down parsers, such as ANTLR, have a workaround to allow left +recursion. They internally turn it into right recursion by the standard +transformation, and then optimize out the ensuing tail recursion. But that +is a hack, which increases the distance between your expression tree and +your abstract syntax tree, still increases the distance between your grammar +and your semantics during parser execution. You are walking the hack, +instead of walking your own grammar’s syntax tree in Reverse Polish order. +Implementing semantics becomes more complex. You still wind up with added +complexity when doing left recursion, just moved around a bit. + +LALR allows you to more directly express the grammar you want to express. With +top down parsers, you can accomplish the same thing, but you have to take a +more roundabout route to express the same grammar, and again you are likely +to find you have allowed expressions that you do not want and which do not +naturally have reasonable and expected semantics. + +ANTLR performs top down generation of the expression tree. Your code called by +ANTLR converts the expression tree into the Abstract Syntax tree, and the +abstract syntax tree into the High Level Intermediate Representation. + +The ANTLR algorithm can be slow as a week of sundays, or wind up eating +polynomially large amounts of memory till it crashes. To protect against +this problem, [he +suggests using the fast SLL algorithm first, and should it fail, then use +the full on potentially slow and memory hungry LL\* algorithm.](https://github.com/antlr/antlr4/issues/374) Ninety +percent of language files can be parsed by the fast algorithm, because people +just do not use too clever by half constructions. But it appears to me that +anything that cannot be parsed by SLL, but can be parsed by LL\*, is not good +code – that what confuses an SLL parser also confuses a human, that the +alternate readings permitted by the larger syntax are never code that people +want to use. + +Antlr does not know or care if your grammar makes any sense until it tries to +analyze particular texts. But you would like to know up front if your +grammar is valid. + +LALR parsers are bottom up, so have terrible error messages when they analyze +a particular example of the text, but they have the enormous advantage that +they will analyze your grammar up front and guarantee that any grammatically +correct statement is LALR. If a LALR parser can analyze it, chances are that +a human can also. ANTLR permits grammars that permit unintelligible statements. + +The [LRX parser](http://lrxpg.com/downloads.html) looks the most +suitable for your purpose. It has a restrictive license and only runs in the +visual studio environment, but you only need to distribute the source code it +builds the compiler from as open source, not the compiler compiler. It halts +at the first error message, since incapable of building intelligible multiple +error messages. The compiler it generates builds a syntax tree and a symbol +table. + +The generically named [lalr](https://github.com/cwbaker/lalr) +looks elegantly simple, and not joined at the hip to all sorts of strange +environment. Unlike Bison C++, should be able to handle unicode strings, +with its regular expressionsrx pa. It only handles BNF, not EBNF, but that +is a relatively minor detail. Its regular expressions are under documented, +but regular expression syntax is pretty standard. It does not build a symbol +table. + +And for full generality, you really need a symbol table where the symbols get +syntax, which is a major extension to any existing parser. That starts to +look like hard work. The lalr algorithm does not add syntax on the fly. The +lrxpg parser does build a symbol tree one on the fly, but not syntax on the +fly – but its website just went down. No one has attempted to write a +language that can add syntax on the fly. They build a syntax capable of +expressing an arbitrary graph with symbolic links, and then give the graph +extensible semantics. The declaration/definition semantic is not full +parsing on the definition, but rather operates on the tree. + +In practice, LALR parsers need to be extended beyond LALR with operator +precedence. Expressing operator precedence within strict LALR is apt to be +messy. And, because LALR walks the tree in reverse polish order, you want +the action that gets executed at parse time to return a value that the +generated parser puts on a stack managed by the parser, which stack is +available when the action of the operator that consumes it is called. In +which case the definition/declaration semantic declares a symbol that has a +directed graph associated with it, which graph is then walked to interpret +what is on the parse stack. The data of the declaration defines metacode +that is executed when the symbol is invoked, the directed graph associated +with the symbol definition being metacode executed by the action that parser +performs when the symbol is used. The definition/declaration semantic allows +arbitrary graphs containing cycles (full recursion) to be defined, by the +declaration adding indirections to a previously constructed directed graph. + +The operator-precedence parser can parse all LR(1) grammars where two +consecutive nonterminals and epsilon never appear in the right-hand side of any +rule. They are simple enough to write by hand, which is not generally the case +with more sophisticated right shift-reduce parsers. Second, they can be written +to consult an operator table at run time. Considering that “universal” resource +locators and command lines are parsed with mere lexers, perhaps a hand written +operator-precedence parser is good enough. After all, Forth and Lisp have less. + +C++ variadic templates are a purely functional metalanguage operating on the +that stack. Purely functional languages suck, as demonstrated by the fact +that we are now retroactively shoehorning procedural code (if constexpr) into +C++ template meta language. Really, you need the parse stack of previously +encountered arguments to potentially contain arbitrary objects. + +When a lalr parser parses an if-then-else statement, then if the parser +grammer defines “if” as the nonterminal, which may contain an “else” +clause, it is going to execute the associated actions in the reverse order. +But if you define “else” as the nonterminal, which must be preceded by an +“if” clause, then the parser will execute the associated actions in the +expected order. But suppose you have an else clause in curly brackets +inside an if-then-else. Then the parse action order is necessarily going to +be different from the procedural. Further, the very definition of an if-then-else clause implies a parse time in which all actions are performed, and a procedural time in which only one action is performed. + +Definition code metacode must operate on the parser stack, but declaration +metacode may operate on a different stack, implying a coroutine relationship +between declaration metacode and definition metacode. The parser, to be +intelligible, has to perform actions in as close to left to right order as +possible hence my comment that the “else” nonterminal must contain the “if” +nonterminal, not the other way around – but what if the else nonterminal +contains an “if then else” inside curly braces? The parser actions can and +will happen in different order to the run time actions. Every term of the +if-then-else structure is going to have its action performed in syntax order, +but the syntax order has to be capable of implying a different procedural +order, during which not all actions of an if-then-else structure will be +performed. And similarly with loops, where every term of the loop causes a +parse time action to be performed once in parse time order, but procedural +time actions in a different order, and performed many times. + +This implies that any fragment of source code in a language that uses the +declaration/definition syntax and semantic gets to do stuff in three phases +(Hence in C, you can define a variable or a function without declaring it, +resulting in link time errors, and in C++ define a class without declaring +its methods and data, resulting in compilation errors at a stage of +compilation that is ill defined and inexplicit) + +The parser action of the declaration statement constructs a declaration data +structure, which is metacode, possibly invoking the metacode generated by +previous declarations and definitions. When the term declared is then used, +then the metacode of the definition is executed. And the usage may well +invoke the metacode generated by the action associated at parse time with the +declaration statement, but attempting to do so causes an error in the parser +action if the declaration action has not yet been encountered in parse action +order. + +So, we get parser actions which construct definition and declaration metacode +and subsequent parser actions, performed later during the parse of subsequent +source code that invoke that metacode by name to construct metacode. But, as +we see in the case of the if-then-else and do-while constructions, there must +be a third execution phase, in which the explicitly procedural code +constructed, but not executed, by the metacode, is actually executed +procedurally. Which, of course, in C++ is performed after the link and load +phase. But we want procedural metacode. And since procedural metacode must +contain conditional and loops, there has to be a third phase during parsing, +executed as a result of parse time actions, that procedurally performs ifs +and loops in metacode. So a declaration can invoke the metacode constructed +by previous declarations – meaning that a parse time action executes metacode +constructed by previous parse time actions. But, to invoke procedural +metacode from a parse time action, a previous parse time action has to have +invoked metacode constructed by an even earlier parse time action to +construct procedural metacode. + +Of course all three phases can be collapsed into one, as a definition can act +as both a declaration and a definition, two phases in one, but there have to +be three phases, that can be the result parser actions widely separated in +time, triggered by code widely separated in the source, and thinking of the +common and normal case is going to result in mental confusion, collapsing +things that are distinct, because the distinction is commonly uniportant and +elided. Hence the thick syntactic soup with which I have struggling when I +write C++ templates defining classes that define operators and then attempt +to use the operators. + +In the language of C we have parse time actions, link time actions, and +execution time actions, and only at execution time is procedural code +constructed as a result of earlier actions actually performed procedurally. + +We want procedural metacode that can construct procedural metacode. So we +want execution time actions performed during parsing. So let us call the +actions definitional actions, linking actions, and execution actions. And if +we ware going to have procedural actions during parsing, we are going to have +linking actions during parsing. (Of course, in actually existent C++, second +stage compilation does a whole lot of linker actions, resulting in +excessively tight coupling between linker and compiler, and the inability of +other languages to link to C++, and the syntax soup that ensues when I define +a template class containing inline operators. + +# Forth the model + +We assume the equivalent of Forth, where the interpreter directly interprets +and executes human readable and writeable text, by looking the symbols in the +text and performing the actions they comand, which commands may command the +interpreter to generate compiled and linked code, including compiled code that +generates compiled and linked code, commands the interpreter to add names for +what it has compiled to the name table, and then commands the interpreter to +execute those routines by name. + +Except that Forth is absolutely typeless, or has only one type, fixed +precision integers that are also pointers, while we want a language in which +types are first class values, as manipulable as integers, except that they +are immutable, a language where a pointer to a pointer to an integer cannot +be added to a pointer, and subtraction of one pointer from another pointer of +the same type pointing into the same object produces an integer, where you +cannot point a pointer out of the range of the object it refers to, nor +increment a reference, only the referenced value. + +Lexing merely needs symbols to be listed. Parsing merely needs them to be, in C++ terminology, declared but not defined. Pratt parsing puts operators in forth order, but knows and cares nothing about types, so is naturally adapted to a Forth like language which has only one type, or values have run time types, or generating an intermediate language which undergoes a second state compilation that produces statically typed code. + +In forth, symbols pointed to memory addresses, and it was up to the command whether it would load an integer from an address, stored an integer at that address, execute a subroutine at that address, or go to that address, the ultimate in unsafe typelessness. + +Pratt parsing is an outstandingly elegant solution to parsing, and allows compile time extension to the parser, though it needs a lexer driven by the symbol table if you have multi character operators, but I am still lost in the problem of type safety. + +Metaprogramming in C++ is done a lazily evaluated purely functional language +where a template is usually used to construct a type from type arguments. I +want to construct types procedurally, and generate code procedurally, rather +than by invoking pure functions. + +In Pratt parsing, the the language is parserd sequentially in parser order, but +the parser maintains a tree of recursive calls, and builds a tree of pointers +into the source, such that it enters each operator in polish order, and +finishes up each operator in reverse polish order. + +On entering in polish order, this may be an operand with a variable number of +arguments (unary minus or infix minus) so it cannot know the number of operands +coming up, but on exiting in reverse polish order, it knows the number and +something about the type of the arguments, so it has to look for an +interpretation of the operator that can handle that many arguments of those +type. Which may not necessarily be a concrete type. + +Operators that change the behavior of the lexer or the parser are typically +acted upon in polish order. Compilation to byte code that does not yet have +concrete types is done in reverse polish order, so operators that alter the +compilation to byte code are executed at that point. Operators that manipulate +that byte code during the linking to concrete types act at link time, when the +typeless byte code is invoked with concrete types. + +Naming puts a symbol in the lexer symbol table. + +Declaring puts a symbol in the parser symbol table + +Defining compiles, and possibly links, the definition, and attaches that data +to the symbol where it may be used or executed in subsequent compilation and +linking steps when that symbol is subsequently invoked. If the definition +contains procedural code, it is not going to be executed procedurally until +compiled and linked, which will likely occur when the symbol is invoked later. + +An ordinary procedure definition without concrete types is the equivalent of an +ordinary C++ template. When it is used with concrete types, the linker will +interet to the operations it invokes in terms of those concrete types, and fail +if they don’t support those operations. + +A metacode procedure gets put into the lexer symbol table when it is named, +into the parser symbol table when it is defined. When it is declared, its +definition may be used when its symbol is encountered in polish order by the +parser, and may be executed at that time to modify the behavior of parser and +linker. When a named, declared, and defined symbol is encountered by the +parser in reverse polish order, its compiled code may be used to generate +linked code, and its linked and compiled code may manipulate the compiled code +preparator to linking. + +When a symbol is declared, it gets added to the parser and lexer symbol table. When it is defined, it gets added to the linker symbol table. When defined with a concrete type, also gets added to the linker symbol table with those concrete types, as an optimization. + +If an operation could produce an output of variant type, then it is an additive +algebraic type, which then has to handled by a switch statement. + +There are five steps: Lexing, parsing, compiling, linking, and running, and +any fragment of source code may experience some or all of these steps, with the +resulting entries in the symbol table then being available to the next code +fragment, Forth style. Thus `77+9`gets lexed into `77, +, `, parsed into `+(77, 9)`, compiled into `77 9 +`, +linked into `77, 9 +` and executed into `int(86` and the rest of the source code proceeds to parse, compile, link, and +run as if you had written `86`. + +Further the source code can create run time code, code that gets declared, +defined, and linked during the compile that is executed during the compile, +modifying the behavior of the lexer, the parser, the compiler, and the linker +over the course of a single compile and link. This enables a forth style +bootstrapping, where the lexer, parser, compiler and linker lexes, compiles, +and links, most of its own potentially modifiable source code in every compile, +much as every c++ compile includes the header files for the standard template +library, so that much of your program is rewritten by template metacode that +you included at the start of the program. + +Compiled but not linked code could potentially operate on variables of any +type, though if the variables did not have a type required by an operator, you +would get a link time error, not get a compile time error. This is OK because +linking of a fragment of source is not a separate step, but usually happens +before the lexer has gotten much further through the source code, happens as +soon as the code fragment is invoked with variables of defined type, though +usually of as yet undefined value. + +A console program is an operator whose values are of the type iostream, it gets +linked as soon as the variable type is defined, and executed when you assign +defined values to iostream. + +Because C++ metacode is purely functional, it gets lazily evaluated, so the +syntax and compiler can cheerfully leave it undefined when, or even if, it +gets executed. Purely functional languages only terminate by laziness. But +if we want to do the same things with procedural metacode, no option but to +explicitly define what get executed when. In which case pure lalr syntax is +going to impact the semantics, since lalr syntax defines the order of parse +time actions, and order of execution impacts the semantics. I am not +altogether certain as to whether the result is going to be intellibile and +predictable. Pratt syntax, however, is going to result in predictzble execution order. + +The declaration, obviously, defines code that can be executed by a subsequent +parse action after the declaration parse action has been performed, and the +definition code that can be compiled after the definition parse action +performed. + +The compiled code can be linked when when invoked with variables of defined +type and undefined value, and executed when invoked with variables of defined +type and an value. + +Consider what happens when the definition defines an overload for an infix +operator. The definition of the infix operator can only be procedurally +executed when the parser calls the infix action with the arguments on the parse +stack, which happens long after the infix operator is overloaded. + +The definition has to be parsed when the parser encounters it. But it is +procedural code, which cannot be procedurally executed until later, much +later. So the definition has to compile, not execute, procedural code, then +cause the data structure created by the declaration to point to that compiled +code. And then later when the parser encounters an actual use of the infix +operator, the compiled procedural code of the infix definition is actually +executed to generate linked procedural code with explicit and defined types, +which is part of the definition of the function or method in whose source code +the infix operator was used. + +One profoundly irritating feature of C++ code, probably caused by LL parsing, +is that if the left hand side of an infix expression has an appropriate +overloaded operator, it works, but if the right hand side, it fails. Here we +see parsing having an incomprehensible and arbitrary influence on semantics. + +C++ is a strongly typed language. With types, any procedure is has typed +inputs and outputs, and should only do safe and sensible things for that type. +C++ metacode manipulates types as first class objects, which implies that if we +were to do the same thing procedurally, types need a representation, and +procedural commands to make new types from old, and to garbage collect, or +memory manage, operations on these data objects, as if they were strings, +floating point numbers, or integers of known precision. So you could construct +or destruct an object of type type, generate new types by doing type operations +on old types, for example add two types or multiply two types to produce an +algebraic type, or create a type that is a const type or pointer type to an +existing type, which type actually lives in memory somewhere, in a variable +like any other variable. And, after constructing an algebraic type by +procedurally multiply two types, and perhaps storing in a variable of type +type, or invoking a function (aka C++ template type) that returns a type +dynamically, create an object of that type – or an array of objects of that +type. For every actual object, the language interpreter knows the type, +meaning the object of type X that you just constructed is somehow linked to the +continuing existence of the object of type type that has the value type X that +you used to construct it, and cannot be destroyed until all the obects created +using it are destroyed. Since the interpreter knows the type of every object, +including objects of type type, and since every command to do something with an +object is type aware, this can prevent the interpreter from being commanded to +do something stupid. Obviously type data has to be stored somewhere, and has +to be immutable, at least until garbage collected because no longer referenced. + +Can circular type references exist? Well, not if they are immutable, because +if a type references a type, that type must already exist, and so cannot +reference a type that does not yet exist. It could reference a function that +generates types, but that reference is not circular. It could have values that +are constexpr, and values that reference static variables. If no circular +references possible, garbage collection by reference counting works + +Types are algebraic types, sums and products of existing types, plus modifiers +such as `const, *,` and `&`. + +Type information is potentially massive, and if we are executing a routine that +refers to a type by the function that generates it, we don’t want that +equivalent of a C++ template invoked every time, generating a new immutable +object every time that is an exact copy of what it produced the last time it +went through the loop. Rather, the interpreter needs short circuit the +construction by looking up a hash of that type constructing template call, to +check it it has been called with those function inputs, to already produced an +object of that type. And when a function that generates a type is executed, +needs to look for duplications of existing types. A great many template +invocations simply choose the right type out of a small set of possible types. +It is frequently the case that the same template may be invoked with an +enormous variety of variables, and come up with very few different concrete +results. + +When the interpreter compiles a loop or a recursive call, the type information +is likely to be an invariant, which should get optimized out of the loops. But +when it is directly executing source code commands which command it to compile +source code. such optimization is impossible + +But, as in forth, you can tell the interpreter to store the commands in a +routine somewhere, and when they are stored, the types have already been +resolved. Typically the interpreter is going to finish interpreting the source +code, producing stored programs each containing a limited amount of type +information. diff --git a/docs/paxos-simple.pdf b/docs/paxos-simple.pdf new file mode 100644 index 0000000..9bce480 Binary files /dev/null and b/docs/paxos-simple.pdf differ diff --git a/docs/paxos_protocol.md b/docs/paxos_protocol.md new file mode 100644 index 0000000..b6f27ca --- /dev/null +++ b/docs/paxos_protocol.md @@ -0,0 +1,362 @@ +--- +title: Paxos +--- +Paxos addresses the arrow theorem, and the difficulty of having a reliable +broadcast channel. + +You want a fifty percent vote, so you don’t want anyone voting for two +candidates, at leas not until one vote has been invalidated by timeout, +and you want to somehow have a single arbitrarily selected agenda +to vote up or down. + +And, having been voted up, you don’t want anything else to be voted up, +so that you can definitely know when an agenda has been selected. + +But Paxos assumes that many of these problems such as who is eligible to vote +and what their vote is worth, have solutions that have been somewhat +arbitrarily predetermined by the engineer setting things up, and that we don’t +have the problem of the Roman Senate and popular assembly, where only about a +third of the Senate actually showed up to vote, and an insignificant number of +those eligible to vote in popular assembly showed up, most of them clients of +senators, so Paxos is not worried about people gaming the system to exclude +voters they do not want, nor worried about people gaming the somewhat +arbitrary preselection of the agenda to be voted up and down. + +# Analysing [Paxos Made Simple] in terms of Arrow and Reliable Broadcast + +[Paxos Made Simple]:./paxos-simple.pdf + +The trouble with Lamport’s proposal, described in [Paxos Made Simple] is that +it assumes no byzantine failure, and that therefore the reliable broadcast +channel is trivial, and it assumes that any proposal will be acceptable, that +all anyone cares about is converging on one proposal, therefore it always +converges on the first proposal accepted by one acceptor. + +> 1. A proposer chooses a new proposal number n and sends a `prepare n` +> request to each member of some set of acceptors, asking it to respond +> with: +> a. A promise never again to accept a proposal numbered less than +> `n`, and +> b. The proposal with the highest number less than `n` that it has +> accepted, if any. +> 2. If the proposer receives the requested responses from a majority of +> the acceptors, then it can issue a `accept n v` request with number +> `n` and value `v`, where `v` is the value of the highest-numbered +> proposal among the responses, or is any value selected by the proposer +> if the responders reported no proposals accepted. + +So the proposer either signs on with the existing possible consensus `v`, and +notifies a bunch of acceptors with the consensus, or initiates new possible +consensus `v` + +The assumption that `n` can be arbitrary seems to assume the proposers are +all agents of the same human, so do not care which proposal is accepted. But +we intend that they be agents of different humans. But let us figure out +how everything fits together before critiquing that. + +> if an acceptor ignores a `prepare` or `accept` request because it has already +> received a prepare request with a higher number, then it should probably +> inform the proposer, who should then abandon its proposal. This is a +> performance optimization that does not affect correctness. + +If a majority of acceptors accept some proposal, then we have a result. But +we do not yet have everyone, or indeed anyone, knowing the result. +Whereupon we have the reliable broadcast channel problem, which Lamport hand +waves away. The learners are going to learn it. Somehow. And once we have +a result accepted, we then happily go on to the next round. + +Well, suppose the leader’s proposal is just intolerable? Lamport assumes a +level of concord that is unlikely to exist. + +Well screw them, they have to propose the same value already accepted by an +acceptor. So we are going to get a definite result. Worst case outcome, is +that proposers keep issuing new higher numbered proposals before any proposal +is accepted. + +Lamport is assuming no byzantine failure – but, assuming no byzantine failure, +this is going to generate a definite result sooner or later. + +But because we could have overlapping proposals with no acceptances, Lamport +concludes, we need a “distinguished proposer”, a leader, +the primus inter pares, the Chief executive officer. +As an efficiency requirement, not a requirement to reach consensus. + +Trouble is that Lamport’s Paxos algorithm is that as soon as it becomes known +that one acceptor has accepted one proposal, everyone should converge to it. + +But suppose everyone becomes aware of the proposal, and 49% of acceptors +think it is great, and 51% of acceptors think it sucks intolerably? + +If a leader’s proposal could be widely heard, and widely rejected, we have a +case not addressed by Lamport’s Paxos protocol. + +Lamport does not appear to think about the case that the leader’s proposals +are rejected because they are just objectionable. + +# Analysing [Practical Byzantine Fault Tolerance] + +[Practical Byzantine Fault Tolerance]:./byzantine_paxos.pdf + +[Practical Byzantine Fault Tolerance] differs from [Paxos Made Simple] in +having “views” where the change of leader (what they call “the primary”) is +accomplished by a change of “view”, and in having three phases, pre-prepare, +prepare, and accept, instead of two phases, prepare and accept. + +Pre-prepare is the “primary” (leader, CEO, primus inter pares) notifying the +“replicas” (peers) of the total order of a client message. + +Prepare is the “replicas” (peers, reliable broadcast channel) notifying each +other of association between total order and message digest. + +Accept is the “replicas” and the client learning that $33\%+1$ of the +“replicas” (peers, reliable broadcast channel) agree on the total order of the +client’s message. + +# Analysing Raft Protocol + +The [raft protocol] is inherently insecure against Byzantine attacks, because +the leader is fully responsible for managing log replication on the other +servers of the cluster + +[raft protocol]: https://ipads.se.sjtu.edu.cn/_media/publications/wang_podc19.pdf + +We obviously want the pool to be replicated peer to peer, with the primus +inter pares (leader, what [Practical Byzantine Fault Tolerance] call +“the primary”) organizing the peers to vote for one block after they have +already come close to agreement, and the only differences are transactions +not yet widely circulated, or disagreements over which of two conflicting +spends is to be incorporated in the next block. + +I am pretty sure that this mapping of byzantine Paxos to blockchain is +garbled, confused, and incorrect. I am missing something, misunderstanding +something, there are a bunch of phases that matter which I am leaving out, +unaware I have left them out. I will have to revisit this. + +The Paxos protocol should be understood as a system wherein peers agree on a +total ordering of transactions. Each transaction happens within a block, and +each block has a sequential integer identifier. Each transaction within a +valid block must be non conflicting with every other transaction within a +block and consistent with all past transactions, so that although the the +block defines a total order on every transaction within the block, all +transactions can be applied in parallel. + +The problem is that we need authoritative agreement on what transactions are +part of block N. + +Proposed transactions flood fill through the peers. A single distinguished +entity must propose a block, the pre-prepare message, notice of this +proposal, the root hash of the proposal, flood fills through the peers, and +peers notify each other of this proposal and that they are attempting to +synchronize on it. Synchronizing on the block and validating it are likely to +require huge amounts of bandwidth and processing power, and will take +significant time. + +If a peer successfully synchronize, he issues a prepare message. If something +is wrong with the block, he issues a nack, a vote against the proposed +block, but the nack is informational only. It signals that peers should get +ready for a view change, but it is an optimization only. + +If a peer receives a voting majority of prepare messages, he issues a commit +message. + +And that is the Paxos protocol for that block of transactions. We then go +into the Paxos protocol for the block of prepare messages that proves a +majority voted “prepare”. The block of prepare messages chains to the block +of transactions, and the block of transactions chains to the previous block +of prepare messages. + +And if time goes by, and we have not managed a commit, perhaps because there +are lot of nacks due to bad transactions, perhaps because the primus inter +pares claims to have transactions that not everyone has, and then is unable +to provide them, (maybe the internet went down for it) peers become open to +another pre-prepare message from the next in line to be primus inter pares. + +In order that we can flood fill, we have to be able to simultaneously +synchronize on several different views of the pool of transactions. If +synchronization on a proposed block is stalling, perhaps because of missing +transactions, we end up synchronizing on multiple proposed blocks proposed by +multiple entities. Which is great if one runs to completion, and the others +do not, but we are likely to run into multiple proposed blocks succeeding. In +which case, we have what [Castro and Liskov](./byzantine_paxos.pdf)call a +view change. + +If things are not going anywhere, a peer issues a view change message which +flood fills around the pool, nominating the next peer in line for primus +inter pares, or the peer nominated in existing pool of view change messages. +When it has a majority for a view change in its pool, it will then pay +attention to a pre-prepare message from the new primus inter pares, (which it +may well have already received) and we the new primus inter pares restarts +the protocol for deciding on the next block, issuing a new pre-prepare +message (which is likely one that many of the peers have already synchronized +on). + +A peer has a pre-pare message. He has a recent vote that the entity that +issued the pre-pare message is primus inter pares. He has, or synchronizes +on, that block whose root hash is the one specified in that pre-prepare +message issued by that primus-inter-pares. When he has the block, and it is +valid, then away we go. He flood fills a prepare message, which prepare +message chains to the block, and to the latest vote for primus inter pares. + +Each new state of the blockchain has a final root hash at its peak, and each +peer that accepts the new state of that blockchain has a pile of commits that +add up to a majority committing to this new peak. But it is probable that +different peers will have different piles of commits. Whenever an entity +wants to change the blockchain state (issues a pre-prepare message), it will +propose an addition to the blockchain that contains one particular pile of +commits validating the previous state of the blockchain. Thus each block +contains one specific view of the consensus validating the previous root. But +it does not contain the consensus validating itself. That is in the pool, +not yet in the blockchain. + +A change of primus inter pares is itself a change in blockchain state, so +when the new primus inter pares issues a new proposed commit, which is to say +a new pre-prepare message, it is going to have, in addition to the pile of +commits that they may have already synchronized on, a pile of what [Castro +and Liskov](./byzantine_paxos.pdf)call view change messages, one specific +view of that pile of view change messages. A valid block is going to have a +valid pre-prepare message from a valid primus inter pares, and, if necessary, +the vote for that primus inter pares. But the change in primus inter pares, +for each peer, took place when that peer had a majority of votes for the +primus inter pares, and the commit to the new block took place when that peer +had the majority of votes for the block. For each peer, the consensus +depends on the votes in his pool, not the votes that subsequently get +recorded in the block chain. + +So, a peer will not normally accept a proposed transaction from a client if +it already has a conflicting transaction. There is nothing in the protocol +enforcing this, but if the double spend is coming from Joe Random Scammer, +that is just extra work for that peer and all the other peers. + +But once a peer has a accepted a double spend transaction, finding consistency +is likely to be easier in the final blockchain, where that transaction simply +has no outputs. Otherwise for the sake of premature optimization, we +complicate the algorithm for reaching consensus. + +# Fast Byzantine multi paxos + +# Generating the next block + +But in the end, we have to generate the next block, which includes some transactions and not others + +In the bitcoin blockchain, the transactions flood fill through all miners, +one miner randomly wins the right to decide on the block, forms the block, +and the block floodfills through the blockchain. + +In our chain, the primus inter pares proposes a block, peers synchronize on +it, which should be fast because they have already synchronized with each +other, and if there is nothing terribly wrong with it, send in their +signatures. When the primus inter pares has enough signatures, he then sends +out the signature block, containing all the signatures. + +If we have a primus inter pares, and his proposal is acceptable, then things +proceed straight forwardly through the same synchronization process as we use +in flood fill, except that we are focusing on flood filling older +information to make sure everyone is in agreement.. + +After a block is agreed upon, the peers focus on flood filling all the +transactions that they possess around. This corresponds to the client phase +of Fast Byzantine Collapsed MultiPaxos. When the time for the next block +arrives, they stop floodfilling, apart from floodfilling any old transactions +that they received before the previous block was agreed, but which were not +included in the previous block, and focus on achieving agreement on those old +transactions, plus a subset of their new transactions. They try to achieve +agreement by postponing new transactions, and flooding old transactions around. + +When a peer is in the Paxos phase, a synchronization event corresponds to +what [Castro and Liskov 4.2](./byzantine_paxos.pdf)call a group commit. +Synchronization events with the primary inter pares result in what [Castro +and Liskov](./byzantine_paxos.pdf) call a pre-prepare multicast, though if +the primus inter pares times out, or its proposal is rejected, we then go +into what [Castro and Liskov 4.4](./byzantine_paxos.pdf) call view changes. +In their proposal, there is a designated sequence, and if one primus inter +pares fails, then you go to the next, and the next, thus reducing the +stalling problem when two entities are trying to be primus inter pares. + +They assume an arbitrary sequence number, pre-assigned. We will instead go +through the leading signatories of the previous block, with a succession of +timeouts. “Hey, the previous primus inter pares has not responded. Let\’s +hear it from the leading signatory of the previous block. Hey. No response +from him either. Let us try the second leading signatory”. Trying for a +different consensus nominator corresponds to what Castro and Liskov call +“view change” + +The primus inter pares, Castro and Liskov\’s primary, Castro and Liskov\’s +view, issues a proposed root hash for the next block (a short message). +Everyone chats to everyone else, announcing that they are attempting to +synchronize on that root hash, also short messages, preparatory to long +messages as they attempt to synchronize. If they succeed in generating the +proposed block, and there is nothing wrong with the block, they send +approvals (short messages) to everyone else. At some point the primus inter +pares wraps up a critical mass of approvals in an approval block (a +potentially long message). When everyone has a copy of the proposed block, +and the approval block, then the block chain has added another immutable +block. + +Castro and Liskov’s pre prepare message is the primary telling everyone +“Hey, let us try for the next block having this root hash” + +Castro and Liskov’s prepare message is each peer telling all the other peers +“Trying to synchronize on this announcement”, confirming that the primus +inter pares is active, and assumed to be sane. + +Castro and Liskov’s commit message is each peer telling all the other peers +“I see a voting majority of peers trying to synchronize on this root hash”. +At this point Castro and Liskov’s protocol is complete – but of course +there is no guarantee that we will be able to synchronize on this root hash - +it might contain invalid transactions, it might reference transactions that +get lost because peers go down or internet communications are interrupted. So +our protocol is not complete. + +The peers have not committed to the block. They have committed to commit to +the block if they have it and it is valid. + +So, we have a Paxos process where they agree to try for a block of +transactions with one specific root hash. Then everyone tries to sync on the +block. Then we have a second Paxos process where they agree that they have a +block of signatories agreeing that they have the block of transactions and it +is valid. + +The block of transactions Merkle chains to previous blocks of signatories and +transactions, and the block of signatories Merkle chains to previous blocks +of transactions and signatories. Each short message of commitment contains a +short proof that it chains to previous commitments, which a peer already has +and has already committed to. + +When a peer has the block of transactions that a voting majority of peers +have agreed to commit to, and the block is valid, it announces it has the +block, and that the block has majority support, and it goes into the flood +fill transactions phase. + +In the flood fill transactions phase, it randomly synchronizes its pool data +with other random peers, where each peer synchronizes with the other peer by +each peer giving the other the pool items it does not yet possess. + +It also enters into a Paxos consensus phase where peers agree on the +authoritative block of signatures, and the time for the next block of +transactions, so that each not only has a majority of signatures, a but the +same block of signatures forming a majority. When the time for forming the +next block arrives, it switches to flood filling only old data around, the +point being to converge on common set of transactions. + +After convergence on a common set of transactions has been going for a while, +they expect an announcement of a proposed consensus on those transactions by +primus inter pares, the pre-prepare message of the Paxos protocol. + +They then proceed to Paxos consensus on the intended block, by way of the +prepare and commit messages, followed, if all goes well, by a voting majority +of peers announcing that they have the block and it is valid. + +Our implementation of Byzantine Paxos differs from that of Castro and Liskov +in that a block corresponds to what they call a stable checkpoint, and also +corresponds to what they call a group transaction. If every peer swiftly +received every transaction, and then rejected conflicting transactions, then +it would approximate their protocol, but you cannot say a transaction is +rollback proof until you reach a stable checkpoint, and if rollbacks are +possible, people are going to game the system. We want authoritative +consensus on what transactions have been accepted more than we want prompt +response, as decentralized rollback is apt to be chaotic, unpredictable, and +easily gamed. They prioritized prompt response, while allowing the +possibility of inconsistent response – which was OK in their application, +because no one wanted to manipulate the system into giving inconsistent +responses, and inconsistent responses did not leave angry clients out of +money. diff --git a/docs/pedersons_secret_sharing.pdf b/docs/pedersons_secret_sharing.pdf new file mode 100644 index 0000000..ef9e6ce Binary files /dev/null and b/docs/pedersons_secret_sharing.pdf differ diff --git a/docs/peering_through_nat.md b/docs/peering_through_nat.md new file mode 100644 index 0000000..526a638 --- /dev/null +++ b/docs/peering_through_nat.md @@ -0,0 +1,122 @@ +--- +lang: en +title: Peering through NAT +--- +A library to peer through NAT is a library to replace TCP, the domain +name system, SSL, and email. + +The NAT mapping timeout is officially 20 seconds, but I have no idea +what this means in practice. I suspect each NAT discards port mappings +according to its own idiosyncratic rules, but 20 seconds may be a widely respected minimum. + +An experiment on [hole punching] showed that most NATs had a way +longer timeout, and concluded that the way to go was to just repunch as +needed. They never bothered with keep alive. They also found that a lot of +the time, both parties were behind the same NAT, sometimes because of +NATs on top of NATs + +[hole punching]:http://www.mindcontrol.org/~hplus/nat-punch.html +"How to communicate peer-to-peer through NAT firewalls" +{target="_blank"} + +Another source says that "most NAT tables expire within 60 seconds, so +NAT keepalive allows phone ports to remain open by sending a UDP +packet every 25-50 seconds". + +The no brainer way is that each party pings the other at a mutually agreed +time every 15 seconds. Which is a significant cost in bandwidth. But if a +server has 4Mib/s of internet bandwidth, can support keepalives for couple +of million clients. On the other hand, someone on cell phone data with thirty +peers is going to make a significant dent in his bandwidth. + +With client to client keepalives, probably a client will seldom have more +than dozen peers. Suppose each keepalive is sent 15 seconds after the +counterparty's previous packet, or an expected keepalive is not received, +and each keepalive acks received packets. If not receiving expected acks +or expected keepalives, we send nack keepalives (hello-are-you-there +packets) one per second, until we give up. + +This algorithm should not be baked in stone, but rather should be an +option in the connection negotiation, so that we can do new algorithms as +the NAT problem changes, as it continually does. + +If two parties are trying to setup a connection through a third party broker, +they both fire packets at each other (at each other's IP as seen by the +broker) at the same broker time minus half the broker round trip time. If +they don't get a packet in the sum of the broker round trip times, keep +firing with slow exponential backoff until connection is achieved,or until +exponential backoff approaches the twenty second limit. + +Their initial setup packets should be steganographed as TCP startup +handshake packets. + +We assume a global map of peers that form a mesh whereby you can get +connections, but not everyone has to participate in that mesh. They can be +clients of such a peer, and only inform selected counterparties as to whom +they are a client of. + +The protocol for a program to open port forwarding is part of Universal Plug and Play, UPnP, which was invented by Microsoft but is now ISO/IEC 29341 and is implemented in most SOHO routers. + +But is it generally turned off by default, or manually. Needless to say, if relatively benign Bitcoin software can poke a hole in the +firewall and set up a port forward, so can botnet malware. + +The standard for poking a transient hole in a NAT is STUN, which only works for UDP – but generally works – not always, but most of the time. This problem everyone has dealt with, and there are standards, but not libraries, for dealing with it. There should be a library for dealing with it – but then you have to deal with names and keys, and have a reliability and bandwidth management layer on top of UDP. + +But if our messages are reasonably short and not terribly frequent, as client messages tend to be, link level buffering at the physical level will take care of bandwidth management, and reliability consists of message received, or message not received. For short messages between peers, we can probably go UDP and retry. + +STUN and ISO/IEC 29341 are incomplete, and most libraries that supply implementations are far too complete – you just want a banana, and you get the entire jungle. + +Ideally we would like a fake or alternative TCP session setup, and then you get a regular standard TCP connection on a random port, assuming that the target machine has that service running, and the default path for exporting that service results in window with a list of accessible services, and how busy they are. Real polish would be hooking the domain name resolution so that names in the peer top level domain return a true IP, and and then intercepts TCP session setup for that IP so that it will result in TCP session setup going through the NAT penetration mechanism if the peer is behind a NAT. One can always install one’s own OSI layer three or layer two, as a vpn does or the host for a virtual machine. Intercept the name lookup, and then tell the layer three to do something special when a tcp session is attempted on the recently acquired IP address, assuming the normal case where an attempt to setup a TCP session on an IP address follows very quickly after a name lookup. + +Note that the internet does not in fact use the OSI model though everyone talks as if it did. Internet layers correspond only vaguely to OSI layers, being instead: + +1. Physical +2. Data link +3. Network +4. Transport +5. Application + +And I have no idea how one would write or install one’s own network or transport layer, but something is installable, because I see no end of software that installs something, as every vpn does. + +------------------------------------------------------------------------ + +Assume an identity system that finds the entity you want to +talk to. + +If it is behind a firewall, you cannot notify it, cannot +send an interrupt, cannot ring its phone. + +Assume the identity system can notify it. Maybe it has a +permanent connection to an entity in the identity system. + +Your target agrees to take the call. Both parties are +informed of each other’s IP address and port number on which +they will be taking the call by the identity system. + +Both parties send off introduction UDP packets to the +other’s IP address and port number – thereby punching holes +in their firewall for return packets. When they get +a return packet, an introduction acknowledgement, the +connection is assumed established. + +It is that simple. + +Of course networks are necessarily non deterministic, +therefore all beliefs about the state of the network need to +be represented in a Bayesian manner, so any +assumption must be handled in such a manner that the +computer is capable of doubting it. + +We have finite, and slowly changing, probability that our +packets get into the cloud, a finite and slowly changing +probability that our messages get from the cloud to our +target. We have finite probability that our target +has opened its firewall, finite probability that our +target can open its firewall, which transitions to +extremely high probability when we get an +acknowledgement – which prior probability diminishes over +time. + +As I observe in [Estimating Frequencies from Small Samples](./estimating_frequencies_from_small_samples.html) any adequately flexible representation of the state of +the network has to be complex, a fairly large body of data, +more akin to a spam filter than a Boolean. diff --git a/docs/petname_language.md b/docs/petname_language.md new file mode 100644 index 0000000..420e2e1 --- /dev/null +++ b/docs/petname_language.md @@ -0,0 +1,26 @@ +--- +title: + Petname Language +--- +Many different cryptographic identifiers get a petname, but the primary one of thought and concern is petnames for Zooko identifiers. + +A Zooko identifier arrives as display name, nickname, public key, and signature binding the display name and nickname to the public key. + +A petname for a Zooko name is derived from the nickname: + +- remove all leading and trailing whitespace. +- If there are no alphabetic characters, prefix it with one at random. +- Move any leading characters that are not alphabetic from prefix to affix. +- Replace all whitespaces with hyphens. +- Replace all special characters with their nearest permitted equivalent.\ + \"#%&'(),.:;<=>?@[]\\^{|} ~\` are special characters that allow escape + from plain text. In displayed text, @ will signify a petname + corresponding to a Zooko name.\ + If someone's nickname is Bob, he will likely get the petname Bob, + which will be displayed in text as `@Bob`, indicating it is likely to + be translated in transmission and reception. +- If the result is a duplicate of an existing petname, append a number that renders it unique. +- The user gets the opportunity to revise the petname, but his petname has to be a valid identifier that conforms to the above rules, or else it gets revised again. +- The user may know that several petnames correspond to one entity, but he cannot assign several nicknames to one petname. + +We will also eventually have local ids for receive addresses. diff --git a/docs/proof_of_stake.md b/docs/proof_of_stake.md new file mode 100644 index 0000000..1063df2 --- /dev/null +++ b/docs/proof_of_stake.md @@ -0,0 +1,684 @@ +--- +title: + Proof of Stake +--- +::: {style="background-color : #ffdddd; font-size:120%"} +![run!](tealdeer.gif)[TL;DR Map a blockdag algorithm equivalent to the +Generalized MultiPaxos Byzantine +protocol to the corporate form:]{style="font-size:150%"} + +The proof of stake crypto currency will work like +shares. Crypto wallets, or the humans controlling the wallets, +correspond to shareholders. +Peer computers in good standing on the blockchain, or the humans +controlling them, correspond to company directors. +CEO. +::: + +We need proof of stake because our state regulated system of notaries, +bankers, accountants, and lawyers has gone off the rails, and because +proof of work means that a tiny handful of people who are [burning a +whole lot of computer power] +control your crypto money.  Apart from being wasteful, they don’t +necessarily have your best interests at heart, producing a bias towards +inflation as in Monero, and/or high transaction fees as in Bitcoin. + +[burning a whole lot of computer power]: https://news.bitcoin.com/businessman-buys-two-electric-power-stations-to-do-bitcoin-mining-in-russia/ + +The entire bitcoin network +[currently consumes just over 46 terawatt hours of energy every year]. +This is almost as much as the annual energy consumption of Portugal, +with its population of roughly 10 million. Simply settling a transaction +in the bitcoin network consumes around 427 kilowatt hours.  This amount +of energy is enough to supply an average German four-person household +with electricity for more than a month. + +[currently consumes just over 46 terawatt hours of energy every year]: https://arstechnica.com/tech-policy/2018/02/european-bankers-scoff-at-bitcoin-for-its-risk-huge-energy-inefficiency/ + +Notaries, bankers, accountants and lawyers, are professionals whom +people hire because they don’t trust each other.  They are in the trust +business.  And then there is some failure of trust, some bad behavior, +as for example Enron’s accounting, and then they turn to the government +to make trust mandatory, whereupon due to regulatory capture and the +malincentives of the state, the untrustworthy behavior becomes standard +and compulsory. + +[Sarbanes-Oxley] was the response to the misconduct of Enron’s +accountants, and was theoretically intended to make behavior similar to +that of Enron’s accountants forbidden, but the practical consequence was +that in substantial part, it made such behavior compulsory.  Which is +why Gab is now doing an Initial Coin Offering (ICO) instead of an +Initial Public Offering (IPO). + +[Sarbanes-Oxley]:sox_accounting.html +"Sarbanes-Oxley accounting" + +Because current blockchains are proof of work, rather than proof of +stake, they give coin holders no power. Thus an initial coin offering +(ICO) is not a promise of general authority over the assets of the +proposed company, but a promise of future goods or services that will be +provided by the company. A proof of stake ICO could function as a more +direct substitute for an initial public offering (IPO).  Thus we want it +to be easy to issue your own coins, and [to perform coin swaps between +chains without the need for an exchange] that would provide a potential +target for regulation. + +[to perform coin swaps between chains without the need for an exchange]: ./contracts_on_blockchain.html#atomic-swaps-on-separate-blockchains + +The block chain, an immutable append only data structure, is a +replacement for public notaries, and on that foundation of public +notarization, we can and should recreate banking, the corporate form, +accounting, and, eventually, lawyering, though the Ethereum attempt to +build lawyering is premature and wicked. We will not be able to build +lawyering until the crypto currency system is well established and has a +good reputation and record system linked to it, and attempting to build +lawyering into it prematurely will result in bad conduct. + +The blockchain governed by proof work is failing.  We need a better +solution for notarizing, based on some form of Paxos, and on that better +system of notarizing, we should build banking.  And on that banking, the +corporate form, and on that corporate form, accounting.  We will worry +about lawyering and justice at some much later date, after crypto +currency values stabilize. + +We build public notarization, then build money on top or that.  When +that is working and stable, we promptly build accounting and a +replacement of the joint stock corporation, which replacement for the +joint stock corporation will manifest as the “Initial coin offering”, +like Gab’s initial coin offering. + +The business of Enron’s accountants failed because no one trusted them +any more. + +Facing massive loss of trust in the wake of the Enron scandal, +accountants rushed to the government, and demanded the government +provide them with mandatory state supplied and state enforced trust, and +got [Sarbanes-Oxley], with the result that they failed upwards instead of +downwards.  This created a situation where it is impossible for a mid +sized company to go public.  So venture capitalists, instead of having +an initial public offering of a successful venture, attempt to sell the +successful venture to Google.  Hence Gab’s Initial Coin Offering. +Google will not buy them, for obvious reasons, and they cannot do an +initial public offering, because of [Sarbanes-Oxley], hence the initial +coin offering. + +Actually existent governments and bankers are evil and untrustworthy, +and the burdens of using trust mediated by bankers and governments to do +business are rapidly becoming intolerable.  Accounting and HR have +become vast onerous bureaucracies, burdensome tentacles of the state in +every business, making businesses larger than a family and smaller than +giant multinational corporation with a skyscraper full of Harvard +lawyers each drawing \$300 per hour, increasingly impractical. + +Proof of work is a terrible idea, and is failing disastrously, but we +need to replace it with something better than bankers and government. + +The gig economy represents the collapse of the corporate form under the +burden of HR and accounting. + +The initial coin offering (in place of the initial public offering) +represents an attempt to recreate the corporate form using crypto +currency, to which existing crypto currencies are not well adapted. + +The corporate form is collapsing in part because of [Sarbanes-Oxley], +which gives accountants far too much power, and those responsible for +deliverables and closing deals far too little, and in part because HR +provides a vector for social justice warriors to infect corporations. + +When a corporation goes social justice it abandons its original +functions, and instead dedicates itself, its resources and the wealth of +its share holders full time to infecting other corporations with social +justice, like a zombie turning those still living into zombies. + +I have been working on the problem of creating a crypto currency well +adapted for this purpose, and what I conclude always implies pre-mining, +that existing owners of a crypto currency shall be like shareholders in +an existing business, transferring shares between each other. + +Which immediately leads us to the idea of a mass of competing +currencies, with share swaps – which automatically has scalability, +particularly if one some of these crypto corporations have as their +major asset shares in the main currency, and a system of enforcement +that makes it difficult for some shareholders to steal that dangerously +liquid asset from other shareholders, in which case they are sidechains, +rather than separate shares systems. + +Western civilization is based on double entry accounting, the joint +stock corporation, and the scientific method, all of which have come +under massive attack. + +Crypto currencies need to make accounting and corporations workable +again, and are needed for this purpose. + +The joint stock corporation has always existed, and our earliest records +of them come from the time of the Roman Empire in the West, but before +Charles the Second they were grants of Kingly power to private +individuals for state purposes – they were socialism. + +Under Charles the Second, they were still theoretically socialism, but +now it was expected and high status to get rich in the course of +pursuing state purposes, socialism similar to “Socialism with Chinese +characteristics” which is not very socialist at all, or “The party +decides what communism is”, which is not very communist at all. + +Under Charles the second we first see the Randian Hero Engineer Chief +executive officer, using other people’s money and other people’s labour +to advance technology and apply technological advance to the creation of +value.  There was a sort of Silicon Valley under Charles the second, but +instead of laying down optical fibre they were laying down canals and +conquering the Indies. + +During late Victorian times, we see the corporation become wholly +private.  Corporations are now people, a situation parodied by Gilbert +and Sullivan – but at the same time, we see the regulatory state +rendering them socialist again by the back door, a problem starting in +Victorian times, and now getting out of control. + +Human Resources has long been an ever more burdensome branch of the +socialist state inserted parasitically into every business, and +[Sarbanes-Oxley] has now destroyed double entry accounting, which is +fundamental to the survival of western civilization. + +Under [Sarbanes-Oxley], the tail wags the dog.  Instead of monitoring +actual reality, accounting decides on official reality. + +Power in business has been taken out of the hands of those who provide +the capital, those responsible for closing deals, and those responsible +for deliverables, and into the hands of Accounting and Human Resources, +who have government granted power to obstruct, disrupt, delay, and +criminalize those responsible for providing capital, those responsible +for closing deals, and those responsible for deliverables. + +We need to construct crypto currency around this function.  The original +intent was for buying drugs, buying guns, violating copyright, money +laundering, and capital flight. + +These are all important and we need to support them all, especially +violating copyright, capital flight and buying guns under repressive +regimes.  But we now see big demand for crypto currencies to support a +replacement for Charles the Second’s corporate form, which is being +destroyed by HR, and to restore double entry accounting, which is being +destroyed by [Sarbanes-Oxley]. + +[Sarbanes-Oxley] is more deeply immoral than forcing doctors to perform +abortions, forcing businesses to pay for abortions, and forcing males to +pay for abortions by single women with whom they have no relationship +through “free” medical care. + +People lost their trust in accountants, so accountants went to the +government, and asked the government to force corporations to act as if +they trusted accountants – which undermines the cooperation and good +behavior on which our economy is based, by undermining the accounting +system that distinguishes positive sum behavior from negative sum +behavior, the accounting system that tracks the creation of value.  This +makes the entire crypto anarchy black market program legitimate and +necessary.  We are morally obligated to obey a government that supports +our civilization and holds it together, keeping peace and order, +defending the realm from enemies internal and external, protecting the +property of taxpayers, enforcing contracts, and honoring warriors. We +are not morally obligated to obey a hostile regime that seeks to destroy +western civilization.  If the state will not uphold trust, cooperation, +and positive sum behavior, we must find other ways.  Undermining +accounting undermines cooperation. + +Used to be that the state church was the state church. Then Education +media complex was the State Church, Harvard seamlessly transitioning +from being the headquarters of the State Church of New England, to being +the headquarters of the American Empire running the entire Education +Media complex of most of the world. + +Now the social media are the state church. + +This is a problem because the unofficially official state belief system +is more and more out of touch with reality, resulting in an increasingly +oppressive social media, that continually censors and bans an ever +increasing number of people for an ever increasing number of crime +thoughts. Thus we need for an anarchic and chaotic social media system +to oppose and resist it. We need the chans, and things like the chans. + +If we had a saner and more functional state belief system, one that did +not compel belief in so many egregious lies, and force so many +compelled affirmations of egregious point-deer-make-horse lies, then it +would be right to repress a chaotic, chan style, darknet style social +media system. If the state church was sane, then censorship to support +the state belief system would be morally justified, and resisting it +morally wrong, but because the quasi state social media are egregiously +untruthful and therefore egregiously repressive, anarchic and chaotic +social media is morally justified and necessary. + +Our system of client wallets needs to support the minimum chat system +necessary to agree to transfer of value to value, and to support +lightning networks to prevent traceability, but we need to architect it +so that wallets are potentially capable of becoming a social media +platform. This capability is not part of the minimum viable product, and +will not be part of early releases, but the minimum viable product needs +to be able to support and prove conversations comprising a contract to +exchange value. Design of the minimum viable product needs to be done in +a way that supports the future capability of wallets supporting general +conversations similar to the chans. If the state church had stuck to +commanding people to believe that God is three and God is one, no one +can prove anything different, but when the state church commands us to +believe that blacks are being unfairly targeted by police, and therefore +quotas limiting arrests of minority groups are totally justified, and +simultaneously totally nonexistent, not only can we prove something +different, but if we were to believe the official truth and official +science that are social media requires us to believe, are likely to get +killed, thus today’s state church is necessarily more repressive than it +was back when it was openly a a state church. God being both three and +one is harder to falsify than arrest quotas being both morally +obligatory and nonexistent, thus higher levels of repression are +required to enforce official belief. + +When the state religion was transliterated from the next world to this, +it became readily falsifiable, resulting in the necessity of disturbing +levels of repression, to which repression crypto anarchy is the morally +appropriate response. With a saner state religion, crypto anarchy would +be a morally objectionable attack on order and social cohesion, but the +current officially unofficial state religion is itself a morally +objectionable attack on order and social cohesion. Hence we need a +distributed system of wallet to wallet encrypted chat with no +monopolistic center. In the minimum viable product we will not attempt +to compete with chat and social media, but we need to have the ability +to discuss payments in wallets, with no monopolistic +center, and will set this up in a way compatible with future expansion +to a future challenge against current social media. In war, all things +are permitted, and the state is making war on society. A minimum system +that allows people to pay each other unsupervised +is not very different from a centreless system that allows people to +associate to discuss any hate fact, nor very different from a centerless +lightning network. + +# General structure + +The blockchain is managed by the peers.  Peers are fully accessible on +the internet, in that they have a port that one can receive UDP messages +from any peer or any client on the internet.  If they are behind a NAT, +the port is forwarded to them by port forwarding, a load balancer, or +failover redirection.  Each peer keeps a copy of the entire blockchain. +Clients only keep a copy of the data that is relevant to themselves, and +are commonly behind NATs, so that their communications with each other +must be mediated by peers.  + +The blockchain contains data linked by hashes, so that any peer can +provide any client with a short proof that the data supplied is part of +the current global consensus, the current global consensus being +represented by a short hash, which is linked to every previous global +consensus.  This chain of hashes ensures that the record of past events +is immutable.  Data can be lost, or can become temporarily or permanently +inaccessible, but it cannot be changed.  The fundamental function of the +blockchain is to function as a public notary, to ensure that everyone +sees the same immutable story about past events.  The hash dag is +structured so that a peer can also produce a short proof to a client +that it has presented all events, or the all the most recent events, +that relate to a given human readable name or a given public key.  For +the hash dag to be capable of being analysed with reasonable efficiency, +the hash dag must correspond to a very small number of approximately +balanced Merkle trees + +In order to produce short proofs that can be proven to contain all +transactions relevant to a given key or human readable name, there have +to be Merkle-patricia trees organized primarily by block number, but +also Merkle trees organized by key and by name, and the root hash of the +current consensus has to depend on the root hash of each of these three +trees, and all past versions of these trees, the root hashes of all past +consensuses.  The blockchain will contain a Merkle-patricia dag of all +unspent transaction outputs, and all spent transaction outputs, enabling +a short proof that a transaction output was spent in block it, and that +as of block n, a another transaction output was not yet spent. + +In order to produce a short proof, the consensus of the block chain has to +be organized a collection of balanced binary Merkle trees, with the state of +the consensus at any one point in time being represented as the list of roots +of the Merkle trees. + +This enables people with limited computing power and a quite small +amount of data to prove that the state of the consensus at any one time for +which they have that quite small amount of data, includes the consensus at +any earlier time for which they have that quite small amount of data. + +We need this capability because at scale, full peers are enormous and have +and handle enormous amounts of data, so we wind up with few peers and +an enormous number of clients, and the clients need to be able to monitor +the peers to resist being scammed. + +Clients need the capability to know that at any moment, the current +consensus of the peers includes that past in which value was committed to +public keys on the blockchain whose secrets he controls. + +We build a system of payments and globally unique human readable names +mapping to [Zooko’s triangle] names (Zooko’s quadrangle) on top of this +public notary functionality. + +[Zooko’s triangle]: ./zookos_triangle.html + +All wallets shall be client wallets, and all transaction outputs shall +be controlled by private keys known only to client wallets, but most +transactions or transaction outputs shall be registered with one +specific peer.  The blockchain will record a peer’s uptime, its +provision of storage and bandwidth to the blockchain, and the amount of +stake registered with a peer.  To be a peer in good standing, a peer has +to have a certain amount of uptime, supply a certain amount of bandwidth +and storage to the blockchain, and have a certain amount of stake +registered to it.  Anything it signed as being in accordance with the +rules of the blockchain must have been in accordance with the rules of +the blockchain.  Thus client wallets that control large amounts of stake +vote which peers matter, peers vote which peer is primus inter pares, +and the primus inter pares settles double spending conflicts and +suchlike. + +In the classic Byzantine Fault Resistant algorithms, the leader is decided +prospectively, and then decides the total order of all transactions. In a +blockdag algorithm, the leader is determined retrospectively rather than +prospectively. Each peer in each gossip event (each vertex of the dag) +makes a decision that implies a total order of all transactions, which +decisions are unlikely to agree, and then which vertex is the one that +actually does set the total order i decided retrospectively, equivalent to +continuous retrospective leader election in the classic Byzantine fault +resistant algorithms. + +Proof of stake works like the corporate form, or can work like the +corporate form, with the crypto currency as shares, the wallets, or the +humans controlling the wallets, as shareholders, the peers in good +standing, or the humans controlling the peers in good standing as the +board, and the primus inter pares, or the human controlling the primus +inter pares, as the CEO. + +Thus the crypto currency works, or can work, like shares in a +corporation.  Proof of stake means that the shareholders can less easily +be screwed over, since the shareholders elect the board from time to +time, and the board elects the CEO from time to time to time. + +But we need many corporations, not just one.  OK, each crypto +corporation corresponds to a sidechain, with its primus inter pares as a +named peer on the mainchain.  Buying and selling shares corresponds to +swapping shares.  The mainchain, if all goes well, has the special +privilege of being the source of root names, and its shares are the most +liquid, the most readily exchangeable, and this is the primary thing +that makes it “main”. + +# Implementation of proof of stake + +Good blockdag protocols with high consensus bandwidth rely on forming +a consensus about the total past of the blockchain during the gossip +protocol where they share transactions around. + +During gossip, they also share opinions on the total past of the blockchain. + +If each peer tries to support past consensus, tries to support the opinion of +what looks like it might be the majority of peers by stake that it sees in +past gossip events, then we get rapid convergence to a single view of the +less recent past, though each peer initially has its own view of the very +recent past. + +Blockdag algorithms of this class of consensus algorithm, if correct and +correctly implemented, are equivalent to Practical Byzantine Fault +Tolerant Consensus with continuous leader election, with the important +differences being that the the leader is elected retrospectively rather than +prospectively, with later peers deciding whom the past leader was and +adopting his opinion of the total past, rather than whom the next leader +will be, and if consensus fails to happen, we get a fork, rather than the +algorithm stalling. The one third limit is fork detection, which is not quite +the same thing as the one third limit in Practical Byzantine Fault Resistant +Consensus, though in a sense equivalent, or closely related. + +Satoshi’s design for bitcoin was that every wallet should be a peer on the +mainchain, every peer should be a miner. + +In Satoshi’s design every peer needs to be a miner, and every wallet a peer, +because that is where the power is. If you don’t have power over your money, +going to get ripped off and oppressed one way or the other way. + +This design fell apart under scaling pressure, with there being two or perhaps +three mining pools that control the blockchain, a massively centralized +system, and people are increasingly reliant on client wallets and exchanges in +a system never designed to securely support client wallets or exchanges, +leading to notorious exchange failures and ripoffs. Getting miners to validate +your transactions is slow and expensive, and getting slower and more +expensive. The lightning network is likely to wind up with all transactions +going through two big exchanges, as credit card transactions do. + +Wallets are by Satoshi’s design linked to output keys. And now every exchange +demands to see your face and evidence of your true name. What I am seeing now +is scaling failure and anonymity failure. + +Bitcoin has failed due to scaling problems, resulting in high costs of +transactions, high delay, heavy use of client wallets in a system in which +client wallets are inherently unsafe, heavy use of exchanges in a system where +exchanges are inherently unsafe. + +For scaling, we need to design a system with a limited number of peers, more +than a dozen, less than a thousand, and an enormous number of client wallets, +billions of client wallets.  And we need to give client wallets power and +prevent the peers from having too much power. + +Power to the wallets means our system has to run on proof of stake, rather +than proof of work. But since a wallet is not always on the internet, cannot +routinely exercise power moment to moment. So, we need a system where unspent +transaction outputs are hosted by particular blockchain peers, or large +transaction outputs are hosted by particular peers, but controlled by client +wallets, and the power of a peer depends on hosting sufficient value. + +The architecture will be somewhat similar to email, where you run an email +client on your computer, whose server is an email agent somewhere out in the +internet. An email agent is a peer in relation to other email agents, and a +host and server in relation to your email client. + +Wallets will rendezvous with other wallets through peers, but the peer will set +up a direct connection between wallets, where wallets send encrypted packets +directly to each other, and the peer cannot know what wallets are talking to +what wallets about what transactions.  By analogy with email agents, we will +call peers on the blockchain blockchain agents when they perform the role of +servicing wallets, but, following the tradition established by Satoshi, peers +when they perform the role of cooperating with each other to operate the +blockchain. + +The blockchain will be a directed acyclic graph indexed by Merkle trees. Being +a directed acyclic graph, the peers will have to download and process the +entire blockchain. The Merkle trees will be approximately balanced, thus of +depth of order log N, hence a blockchain agent can provide a wallet a short +chain of hashes linking any fact about the blockchain to the current root of +the blockchain. + +A wallet has a login relationship to a blockchain agent, which is a peer to +all the other blockchain hosts on that blockchain, just as an email client has +a login relationship with an email agent. A wallet is the client of its +blockchain agent. Your wallet client will be software and a user interface +that manages several wallets. Thus a wallet will be like an email account, +which has a single login relationship with a single host, and your wallet +client, which may manage several wallets, is like an email client, which is +the client of an email agent or a small number of email agents. + +What then stops the blockchain agent from lying to the wallet about the root +and contents of the blockchain?. + +Wallets perform transactions by interacting directly with other wallets +through an encrypted connection. [If both wallets are behind a firewall the +connection is set up by servers, but it does not pass through the servers] +(how_to_punch_holes_in_firewalls.html). For the interaction to succeed, for a +transaction to go through, both wallets must have the same Merkle root for the +blockchain, or one Merkle root must be a recent subsidiary of the other. The +wallet receives a short proof (log N hashes where N is the number of events on +the block chain in the last few months) that proves that the past that Bob’s +wallet received from its peer is part of the same past as the past that Ann’s +wallet received from her peer, that both wallets are on the same blockchain, +that the data about the past that one wallet has is consistent with the data +about the past that the other wallet has. + +Sometimes a wallet name will be a Zooko name, consisting of a public key, +secret key, nickname, and petname. Sometimes it will also have a globally +unique human readable name controlled by a Zooko name. Peers will usually have +globally unique human readable names controlled by a Zooko name human. +However, wallets will most commonly have login names, `wallet_name@peer_name`. +All transaction outputs and inputs have an associated secret key, and the +hash of the transaction must be signed by Schnorr multisignature of all the +inputs. + +Wallets should retain not the full block chain, but relevant transactions and +outputs, and for each transaction and output, log N hashes connecting that to +hashes close to the root. + +Peers check the entire blockchain for validity, and construct short proofs +showing to wallets that they are seeing the same view of those facts that they +care about are the same for everyone, that the past does not get rewritten, +and every wallet on the same blockchain sees the same past.  Wallets only +download short and relevant parts of the blockchain. + +Peers on the blockchain are not +behind NATS, or if they are they have port forwarding, or some similar NAT +penetration set up. One wallet sets up a connection to another wallet through +a peer, but once the connection is set up, information does not pass through +the peer, but rather information is sent directly from wallet to wallet. The +peer negotiates a port for two wallets that wish to communicate and informs +each of the other’s external IP address. It also negotiates a shared secret, +and informs each wallet of the id associated with the shared secret. Both +parties send off introduction UDP packets to the other’s IP address and port +number – thereby punching holes in their firewall for return packets. When +they get a return packet, an introduction acknowledgement, the connection is +assumed established. + +Where a wallet has a login type name, `wallet_name@peer_name`, the peer could +potentially engage in a man in the middle attack against the wallet. However, +during transactions, it is necessary to prove control of the secret key of an +unspent transaction output, which the man in the middle cannot do, with the +result that the connection will break with an error message blaming the peer +that set up the connection. + +When a group of wallets want to perform a transaction, they rendezvous on one +particular wallet, who constructs the transaction, and feeds it to his +blockchain agent.  A rendezvous on a particular wallet is a room on that +wallet. A wallet can have several rooms, and each room can have several other +wallets connected to it. One wallet connected to a room, and also connected to +another wallet by a separate connection, can invite that other wallet to that +room, without needing to go through a peer. This prevents peers from reliably +knowing who is party to a transaction. + +## Byzantine failure. + +Byzantine failure, if intentional and malicious, is lying, either explicitly - giving one party inconsistent with the information given to another party, the classic example being one Byzantine General telling one other general he is going to advance, and another general that he is going to retreat, so that the general expecting to be part of an advance finds himself alone and he is killed and his army destroyed. + +In a blockdag, this always going to become visible eventually, but the problem is, it may become visible too late. + +Mechanisms to protect against Byzantine failure look superficially like +proof of stake shareholder democracy but they are subtly different. They +protect against the ten percent attack, but assume that any one outcome +selected by any one correctly functioning peer is equally acceptable, that +the problem is selecting one of many equally possible and equally +acceptable outcomes, that the decision of any peer of the two thirds of +peers not engaged in byzantine failure is equally OK. + +We need fifty one percent rule, shareholder authority, and we need to +protect against Byzantine failure, the ten percent attack, both. + +We need computer algorithms that prevent the ten percent attack, one third +minus one, and we need social mechanisms that detect and penalize one third +plus one. + +We need computer shareholder democracy and human shareholder democracy layered +on top of and underneath byzantine fault resistance. + +Byzantine fault tolerance looks superficially like shareholder democracy, but +it is not. We need all the mechanisms, each to address its own problem space. + +If two thirds plus one voluntarily follow the +policy of half plus one, and one third plus one testify that one rather +arbitrarily selected outcome is consistent with the policy commanded by +half plus one, then that is the one verified outcome of the blockchain. + +And, to make sure we have mostly honest participants, human level social +enforcement, which means that Byzantine faults need to have some probability +of being detected and Streisanded. + +But, in the event of complete breakdown of a major connection of the +network, we need the one third plus one to reliably verify that there +is no other one third plus one, by sampling geographically distant +and network address distant groups of nodes. + +So, we have fifty percent by weight of stake plus one determining policy, +and one third of active peers on the network that have been nominated by +fifty percent plus one of weight of stake to give effect to policy +selecting particular blocks, which become official when fifty percent plus +one of active peers the network that have been nominated by fifty percent +plus one of weight of stake have acked the outcome selected by one third +plus one of active peers. + +In the rare case where half the active peers see timeouts from the other +half of the active peers, and vice versa, we could get two blocks, each +endorsed by one third of the active peers, which case would need to be +resolved by a fifty one percent vote of weight of stake voting for the +acceptable outcome that is endorsed by the largest group of active peers, +but the normal outcome is that half the weight of stake receives +notification (the host representing them receives notification) of one +final block selected by one third of the active peers on the network, +without receiving notification of a different final block. + +# Funds need to be fully traceable, and fully untraceable + +A crypto currency needs to be totally traceable and totally untraceable. +A common application is money lending in the third world. The money +lender needs to be able to prove he loaned the peasant such and such an +amount, on such and such terms, with such and such an agreement, and the +[peasant needs to be able to prove he repaid the money lender such and +such an amount] +in relation to that agreement and in fulfillment those terms. + +[peasant needs to be able to prove he repaid the money lender such and +such an amount]: http://www.scmp.com/week-asia/business/article/2148658/how-bitcoin-and-cryptocurrencies-went-wall-street-high-streets, + +The peasant’s daughter is working in Saudi Arabia. She buys some crypto +currency, perhaps for cash, and sends it to her mother. She trusts her +mother, her mother trusts her, she does not need any traceability, but +the government wants to tax transfers, and the financial system that has +a government monopoly wants to charge her and her mother a fee for getting in +her way. + +The parties to a transaction can send proof of transaction, and proof +that the transaction is in accord with an agreement, and proof of what +the agreement was, to anyone, or make it public. But if they take no +action to make it public, no one can know who the transaction took place +between, nor what the transaction was about. + +An unhappy customer can make his transaction public. A happy customer +can send a favorable acknowledgment to the supplier, that the supplier +will likely make public. But if neither of them does anything to make it +public, it remains untraceable by default, unless one of the parties +does something special to change the default. + +A company has an initial coin offering instead of an initial share +offering. It goes bust. People who claim to be owed money by the company +want to find the people who bought the initial coins. They do not want +to be found. + +# the corporate form + +To better support the corporate form, the crypto currency maintains a +name system, of globally unique human readable names on top of [Zooko’s +triangle] names. + +[Zooko’s triangle]: ./zookos_triangle.html + +Transactions between [Zooko’s triangle] identities will be untraceable, +because amounts will be in fixed sized amounts, and transactions will +mingle many people paying many recipients.  Transactions with globally +unique human readable names will usually be highly traceable, since that +is what the name is for – but you don’t have to use the name in the +payment, and by default, do not. + +A wallet name will typically look like an email address, +`human readable login name @ human readable peer name`, but the +transaction outputs it controls will be largely controlled by public +keys, which are not provably connected to the wallet, unless the wallet +operator chooses for it to be provable. + +If someone makes a traceable and provable payment to a human readable +name, he can then associate an arbitrary URL with that payment, such as +a review of the service or product provided by the entity with the human +readable name, so that people can find out what people who paid that +entity are saying. + +So if you make a provable payment to a recipient with a human readable +name, [you will have some assurance of getting what you paid +for](trust_and_privacy_on_the_blockchain.html). + +Peers may have human readable names, and wallets may have names of the +form `LoginName@PeerName`. + diff --git a/docs/recognizing_categories_and_instances.md b/docs/recognizing_categories_and_instances.md new file mode 100644 index 0000000..0a0dabe --- /dev/null +++ b/docs/recognizing_categories_and_instances.md @@ -0,0 +1,219 @@ +--- +title: Recognizing categories, and recognizing particulars as forming a category +# katex +--- +This is, of course, a deep unsolved problem in philosophy. + +However, it seems to be soluble as computer algorithm. Programs that do +this, ought to look conscious. + +There are a lot of programs solving things that I though were AI hard, for +example recognizing pornography, recognizing faces in images, predicting what +music, or what books, or what movies, a particular customer might like. + +We have clustering algorithms that work in on points in spaces of reasonably +small dimension. However, instances are sparse vectors in space of +unenumerably large dimension. + +Consider, for example, the problem of grouping like documents to like, for +spam filtering. Suppose the properties of the document are all substrings of +the document of twenty words or less and 200 characters or less. In that case, +there are as many dimensions as there are two hundred character strings. + +# Dimensional reduction + +The combinatorial explosion occurs because we have taken the wrong approach +to reducing problems that originate in the physical world of very large +dimension, large because each quality of the objects involved or potentially +involved is a dimension. + +The cool magic trick that makes this manageable is dimensional reduction. +Johnson and Lindenstrauss discovered in the early 1980s that if one has +$O(2^n)$ points in a space of very large dimension, a random projection onto a +space of dimension $O(n)$ does not much affect distances and angles. + +Achlioptas found that this is true for the not very random mapping wherein +elements of the matrix mapping the large space to the smaller space have the +form $1$, with probability $\frac{1}{6}$, $0$ with probability $\frac{4}{6}$, +$-1$ with probability $\frac{1}{6}$, though a sparse matrix is apt to +distort a sparse vector + +There exists a set of points of size $m$ that needs dimension +$$\displaystyle{O(\frac{\log(m)}{ε^2})}$$ +in order to preserve the distances +between all pairs of points within a factor of $1±ε$ + +The time to find the nearest neighbour is logarithmic in the number of points, +but exponential in the dimension of the space. So we do one pass with rather +large epsilon, and another pass, using an algorithm proportional to the small +number of candidate neighbours times the dimensionality with a small number +of candidate neighbours found in the first pass. + +So in a space of unenumerably large dimension, such as the set of substrings +of an email, or perhaps substrings of bounded length with bounds at spaces, +carriage returns, and punctuation, we deterministically hash each substring, +and use the hash to deterministically assign a mapping between the vector +corresponding to this substring, and a vector in the reduced space. + +The optimal instance recognition algorithm, for normally distributed +attributes, and for already existent, already known categories, is Mahalanobis +distance + +Is not the spam characteristic of an email just its $T.(S-G)$, where $T$ is +the vector of the email, and $S$ and $G$ are the average vectors of good +email and spam email? + +Variance works, instead of probability – Mahalanobis distance, but this is +most reasonable for things that have reasonable dimension, like attributing +skulls to races, while dimensional reduction is most useful in spaces of +unenumerably large dimension, where distributions are necessarily non +normal. + +But variance is, approximately, the log of probability, so Mahalanobis is +more or less Bayes filtering. + +So we can reasonably reduce each email into twenty questions space, or, just +to be on the safe side, forty questions space. (Will have to test how many +dimensions empirically retain angles and distances) + +We then, in the reduced space, find natural groupings, a natural grouping +being an elliptic region in high dimensional space where the density is +anomalously large, or rather a normal distribution in high dimensional space +such that assigning a particular email to a particular normal distribution +dramatically reduces the entropy. + +We label each such natural grouping with the statistically improbable phrase +that best distinguishes members of the grouping from all other such groupings. + +The end user can then issue rules that mails belonging to certain groupings +be given particular attention – or lack of attention, such as being sent +direct to spam. + +The success of face recognition, etc, suggests that this might be just a +problem of clever algorithms. Pile enough successful intelligence like +algorithms together, integrate them well, perhaps we will have sentience. +Analogously with the autonomous cars. They had no new algorithms, they just +made the old algorithms actually do something useful. + +# Robot movement + +Finding movement paths is full of singularities, looks to me that we force it +down to two and half dimensions, force the obstacles to stick figures, and +then find a path to the destination. Hence the mental limit on complex knot +problems. + +Equivalently, we want to reduce the problem space to a collection of regions +in which pathfinding algorithms that assume continuity work, and then +construct graph of such regions where nodes correspond to such convex region +within which continuity works, and edges correspond an overlap between two +such convex regions. Since the space is enormous, drastic reduction is +needed. + +In the case of robot movement we are likely to wind up with a very large +graph of such convex regions within which the assumption of singularity free +movement is correct, and because the graph is apt to be very large, finding +an efficient path through the graph is apt to be prohibitive, which is apt to +cause robot ground vehicles to crash because they cannot quickly figure out +the path to evade an unexpected object and makes it impractical for a robot +to take a can of beer from the fridge. + +We therefore use the [sybil guard algorithm] to reduce the graph by treating +groups of highly connected vertices as a single vertex. + +[sybil guard algorithm]:./sybil_attack.html + +# Artificial Intelligence + +[Gradient descent is not what makes Neural nets work] Comment by Bruce on +Echo State Networks. + +[Gradient descent is not what makes Neural nets work]:https://scottlocklin.wordpress.com/2012/08/02/30-open-questions-in-physics-and-astronomy/ + +Echo state Network is your random neural network, which mixes a great pile of +randomness into your actual data, to expand it into a much larger pile of +data that implicitly contains all the uncorrupted original information in its +very great redundancy, albeit in massively mangled form. Then “You just fit +the output layer using linear regression. You can fit it with something more +complicated, but why bother; it doesn’t help.” + +A generalization of “fitting the output layer using linear regression” is +finding groupings, recognizing categories, in the space of very large dimension +that consists of the output of the output layer. + +Fitting by linear regression assumes we already have a list of instances that +are known to be type specimens of the category, assumes that the category is +already defined and we want an efficient way of recognizing new instances as +members of this category. But living creatures can form new categories, +without having them handed to them on a platter. We want to be able to +discover that a group of instances belong together. + +So we generate a random neural network, identify those outputs that provide +useful information identifying categories, and prune those elements of the +network that do not contribute useful information identifying useful categories. + +That it does not help tells me you are doing a dimensional reduction on the +outputs of an echo state network. + +You are generating vectors in a space of uncountably large dimension, which +vectors describe probabilities, and probabilities of probabilities (Bayesian +regress, probability of a probability of a frequency, to correct priors, and +priors about priors) so that if two vectors are distant in your space, one is +uncorrelated with the other, and if two things are close, they are +correlated. + +Because the space is of uncountably large dimension, the vectors are +impossible to manipulate directly, so you are going to perform a random +dimensional reduction on a set of such vectors to a space of manageably large +dimension. + +At a higher level you eventually need to distinguish the direction of +causation in order to get an action network, a network that envisages action +to bring the external world through a causal path to an intended state, which +state has a causal correlation to *desire*, a network whose output state is +*intent*, and whose goal is desire. When the action network selects one +intended state of the external world rather than another, that selection is +*purpose*. When the action network selects one causal path rather than +another, that selection is *will*. + +The colour red is not a wavelength, a phenomenon, but is a qualia, an +estimate of the likelihood that an object has a reflectance spectrum in +visible light peaked in that wavelength, but which estimate of probability +can then be used as if it were a phenomena in forming concepts, such as +blood, which in turn can be used to form higher level concepts, as when the +Old Testament says of someone who needed killing “His blood is on his own +head”. + +Concepts are Hegelian Neural Networks: “Neurons that fire together, wire +together” + +This is related to random dimensional reduction. You have a collection of +vectors in space of uncountably large dimension. Documents, emails, what you +see when you look in a particular direction, what you experience at a +particular moment. You perform a random dimensional reduction to a space of +manageable dimension, but large enough to probabilistically preserve +distances and angles in the original space – typically twenty or a hundred +dimensions. + +By this means, you are able to calculate distances and angles in your +dimensionally reduced space which approximately reflect the distances and +angles in the original space, which was probably of dimension +$10^{100^{100^{100}}}$, the original space being phenomena that occurred +together, and collections of phenomena that occurred together that you have +some reason for bundling into a collection, and your randomly reduced space +having dimension of order that a child can count to in a reasonably short +time. + +And now you have vectors such that you can calculate the inner product and +cross product on them, and perform matrix operations on them. This gives you +qualia. Higher level qualia are *awareness* + +And, using this, you can restructure the original vectors, for example +structuring experiences into events, structuring things in the visual field +into objects, and then you can do the same process on collections of events, +and collections of objects that have something common. + +Building a flying machine was very hard, until the Wright brothers said +“three axis control, pitch, yaw, and roll” + +Now I have said the words “dimensional reduction of vectors in a space of +uncountably large dimension, desire, purpose, intent, and will” diff --git a/docs/replacing_TCP.md b/docs/replacing_TCP.md new file mode 100644 index 0000000..2232fd7 --- /dev/null +++ b/docs/replacing_TCP.md @@ -0,0 +1,1082 @@ +--- +title: Replacing TCP, SSL, DNS, CAs, and TLS +--- +# What the API should look like + +It should be a message api, not a connection api. + +On the other hand this involves re-inventing networking from scratch, as +compared to simply copying http/2, or some other reliable UDP system. + +Total rewrites, however desirable and necessary, always fail + +So on reflection this is a blue sky proposal - likely to involve immense delay: + +I need to think about the way things should be done - but I don't want to +get lost in the weeds. I have repeatedly wasted a great deal of time +re-inventing stuff from scratch, only to find that when I was finished, I had +something vastly inferior to what already existed, so I wound up tossing +my work, and using someone else's library with minimum adaptation. + +Many a time I see something is encrusted with ancient history, backward +compatibility means they cannot fix old mistakes, I design something new +and fresh, and vastly superior, and discover that there were one hundred +and one issues that old history encrusted thing had encountered and dealt +with, and I had not foreseen, that not all of that mighty pile of code is crap +to work around past mistakes which must continue to be supported, but a +lot of it is issues I had not foreseen having to deal with, and had not +planned a path to dealing with them. + +When implementing stuff from scratch, all too often one discovers there +are no end of reasons for all the stuff one thought bad and unnecessary in +existing libraries. + +But on with the vision. Though it will likely be vastly faster to just fix +someone else's library to have real security. + +Although the api represents messages, rather than connections, it will +implicitly have a very large number of connections, in that a connection is +your current state with a counterparty, expected protocols (message types) and all that. + +For an app to poll a very large number of connections over the network, +`select` does not cut the mustard. Network apis have been evolving, each in +its own idiosyncratic way, to the app making O(1) additions and deletions to +list of counterparties on the network whose messages it is listening to, +and getting notifications that are O(number of events) rather than +O(number of counterparties). + +The way this should be done is a linked list of data structures containing +events, which the app can poll locklessly, or wait on (with a timer event +guaranteed to appear in the list eventually if it is waiting on it). If the app +fails to free anything from the list after an unreasonably long time, +suggesting that the app has shut down ungracefully or crashed, and there +are rather too many things on the list, the process that is putting things on +the list will start by pushing back on the parties sending messages to the +app, and end by shutting down their connections and discarding their data. +The network events live entirely in memory and are volatile. If they +represent long lived relationships, it is up to the app to commit the +information that they represent to disk. + +Every message has a public key of sender, a public key of recipient, an +potentially an in-regards-to hash, a reply-to hash, and an in-reply-to hash. +Some or all of these hashes may be null. It seldom makes sense for all of +them to be null, and it seldom makes sense for all of them to be non null. +Usually reply-to is null, and it does not always make sense for it to be non +null. + +The reply-to field opens up a very large can of worms, in that its main use +is to reference a third party message that came from a third party server, +with its own type information and sender public key, and the how does the +sender know the recipient has or can obtain that message? + +Every hash and every public key represents a potential endpoint, and thus +represents an additive type, or rather gives the system potential clues on +how to discover a mutually known additive type. (Reflect on the slow and +chaotic semi automated complexity of how the many protocols involved in +sending and receiving an email message are discovered, every time, for +every email message.) + +Some of the time, the message type is only known from one of these +hashes – they imply the type information, without which the recipient +would not know how to parse the message, and the recipient has to be able +to recognize them before he can recognize anything else. And some of the +time, figuring out the message type from these hashes is non trivial or just +flat out fails. No general automatic one size fits all procedure can work on +every mysterious second party hash. This is a problem that has to be dealt +with ad hoc use case by use case, protocol by protocol, message type by +message type. + +Not all messages can be sent reliably, but the sender gets a notification +event – failed, succeeded, replied to, or unlikely to be known, and the +sender can immediately find out either the likely timing of such +notification, or that the likely timing of such notification is unknown – and +usually that the likely timing of such notification is unknown generates an +exception. + +The api is potentially multilayered – the message may well get translated +to a multitude of similarly structured messages, that set up the connection, +find out information about the recipient, all that stuff, and when those +messages go on the wire, they do not necessarily have any of this stuff – +commonly they just have the network, the port address, and some numbers +that uniquely identify the context, which numbers are unique to the +connection, but unlike the hashes from which they are derived, not +globally unique, are sequential identifiers, not hashes. But at the top level, +the network address, the port, and all that stuff is just not represented, +except implicitly in that the public key of the recipient may well get +looked up in a hash table that may well have the network address and the port. + +On the wire, network address and port serves the function of in-regards-to, +and will wrap stuff that provides a finer grained function of in-regards-to +and in-reply-to -- as I said, multilayered, with the hashes being internally +mapped to to data that serves equivalent functionality. Network address +and port being the outermost layer on the wire. + +On the wire, once a connection is established, the sender and recipient +public keys are implicit in the ip header, and rest is opaque payload, +maximum payload being 1kiB. Inside the payload, the representation +depends on the message type, which was established when the connection +was established – the in-reply-to of the contained message is the unique +sequential nonce of the message being replied to, rather than the hash of +that message. + +In the api, the application and api know the message type, because +otherwise the api just would not work. But on the rare occasions when the +message is represented globally, outside the api, *then* it needs a message type header. + +# TCP is broken + +TCP was designed in more trusting times, when the name system +consisted of a widely shared hosts file, and everyone trusted everyone. + +Over the years people have piled warts on top of TCP and warts on top of +warts to fix one problem after another, and every fix results in additional round trips + +Thus “Cloudfare is checking your browser, you will be redirected shortly” + +Every additional round trip before a web page comes up results in a +significant loss of viewers. Hence http2. Which fails to fix the DDOS and +cloudfare problem. + +TCP is a major problem, which is slowing down the internet. DDoS +protection and the certificate mess are warts growing on top of warts. + +Any business that resists corporate cancer is going to come under DDoS, +and if it employs a DDoS resistance service, that service is likely to place +pressure on the business to do political stuff that is counterproductive to +pursuing a profit. And even if it does not, the DDoS service slows down +people trying to view the business website. + +If the TCP replacement fixes those warts, you get more views. + +# Domain name system and SSL is broken + +Any organization that has a certificate authority in its pocket can perform +a man in the middle attack on an SSL connection, though the CAA domain +name record somewhat mitigates this problem. + +We need to also need to replace the TCP/SSL/CA/DNS system because +there is money in it. A great deal of money. + +The trouble with an ICO (initial coin offering), is that the issuer has no +obligation to do anything other than take the money and run. We are +moving to an economy where much of the value is “goodwill”, “goodwill” +being names with reputations and relationships. The blockchain (or +blockdag, since blockdags theoretically have better scaling than +blockchains) could be used to render this value liquid in IPOs by having +both names and money on the blockchain. + +Atomic transactions between blockchains, plus names on the blockchain +with money, a replacement for TCP/SSL/CAs/DNS could support sovereign +corporations on the blockchain, so that an ICO could be an IPO (Initial +Public Offering). If the blockchain is a name service as well as a money +service, it could give the investors ownership of the name. The owners of +examplecorp shares get to designate the board public key, and the board gets to +designate the public key of CEO@examplecorp from time to time, thus +rendering the value of a name potentially liquid. + +Cryptocurrency exchanges are run by crooks, and are full of crooks each +trying to scam all the other crooks. + +If you don’t know who the pigeon is, you are the pigeon. + +A healthy cryptocurrency market needs to leave the cryptocurrency +exchanges behind, replacing them with atomic blockchain transactions +between separate blockchains. They are dangerously centralized, and +linked to a corruptly regulated finance and accounting system, which +corruption we saw with Great Minority Mortgage Meltdown and the +Mortgage backed Security market from 2005 November to 2007, and saw +with MF Global. Jon Corzine did worse than embezzle client funds. He +embezzled client funds legally. + +Demand for crypto currencies is driven in substantial part by the fact that +recent regulations have cheerfully set aside laws on fiduciary duty that are +millennia old. The exchanges cheerfully adhere to such regulations as they +find dangerously convenient, while taking advantage of cryptocurrency to +avoid those regulations that they find inconvenient. + +The banks, the stock exchanges, and the big accounting firms are regulated +agencies whose regulators are in their pocket. The crypto currency exchanges +are semi regulated, taking advantage of regulations written for those who +have regulators in their pocket. + +The cryptocurrency market needs to get rid of exchanges, starting with +cryptocurrency exchanges, and proceeding to get rid of stock exchanges. + +An exchange exists to provide an escrow that faithfully observes +its fiduciary duty. And there have been a great many recent examples of such +entities getting up to no good, and in the case of the mortgage backed +security market, up to no good with enormous amounts of money. + +A cryptocurrency with a name system could eat their lunch, greatly enriching +its founders in the process. + +# Distributed Denial of Service attack + +At present, resistance to Distributed Denial of Service attacks rests on +dangerously powerful central authorities, in particular Cloudfare, whose +service in addition to being dangerously centralized, is expensive and poor. + +The TCP replacement needs an adjustable proof of work (pow) handshake +as the first part of the connection handshake, the proof of work request +being first server packet in the four packet handshake. + +First packet, client requests connection, second packet, server requests +work,and supplies a durable and a short lived public key, third packet, +client supplies work and offers transient public key, making +communication possible, plus the message it is trying to send the server, or +the first part of that message. + +The work demanded goes up as the server load increases, thus fixing the +horrors of DDoS protection. + +## Key agreement + +Key agreement needs to be part of the the TCP replacement handshake, rather +than a layer on top, to reduce round tripping. + +The name system needs to be integrated with the key system, so that you get +the key when when you get the network address associated with the name, and +the key/name pairing needs to be blockchain secured, so you don’t have one +thousand certificate authorities each with the authority to mount a man in the middle attack. + +## replacement handshake for publicly identified server + +The the TCP replacement handshake needs to be a four phase handshake. + +1. Client->Server: Give me a connection, here are my parameters, here is my +session key. + +1. Server->Client: Here is a proof of work request, my parameters, and a keyed +hash of your and my parameters. Ask again with proof of work, the same +parameters, and the keyed hash. + + Server then throws away the request, allocating no memory. + +1. Client->Server: OK, here I am again, with all that stuff you asked for. + + This includes a konce (key used once,single use elliptic point), and + assumes that the client reliably knows the server public key i + advance. This protocol is inappropriate to signons that are restricted + to identified entities, because we probably do not want everyone to + know who is + identified. + +1. Server checks the poly1305 authentication to ensure that this is a + real client reply to a real and recent server reply. Then it checks the + proof of work. + + If the proof of work passes, Server allocates memory, generates and stores a + session key, and stores connection parameters, the client and server + session keys among them. + +1. Server->Client: OK, here is my session key, authenticated but not + signed by my permanent key, and stuff, now you can start sending + actual data. + +Thus we can integrate TCP handshake and encryption hand shake and the +innumerable DDoS protection handshakes “Cloudfare is checking your browser, +oops, your browser did not pass, here is a captcha” at the cost of one single +additional trip, half a round trip. + +Instead of the person establishing the connection fuming while round trip +after round trip goes through, we get all that stuff at the cost of one +additional half round trip. + +### pow implementation + +Each sequential proof of work request contains a 64 bit sequential integer. +The integer starts at random 63 bit value, to ensure that every possible +successful proof of work ever used is unique in the universe. The +sequential integer is treated as a windowed value into a 512 bit integer, +whose high order part is an unshared secret that remains unchanged for the +duration. + +From that 512 bit value, the server generates a unique XChaCha20 512 bit +value, 256 bits of which are used to generate a Poly1305 authenticator for +the proof of work request. If it receives a completed proof of work request +containing the authentication, it knows it comes from an entity at that +network address that was able to receive the proof of work request. +Knowing it is talking to real network addresses, it can derank network +addresses that create excessive burdens, so that they cannot slow down +everyone else, only themselves. + +When it receives the completed proof of work, it first checks the sequence +number to ensure it is a recently issued request for work, then checks if +there is already a channel allocated for that pow, using a table of doubly +linked lists of recently allocated channels.indexed by the low order part of +the pow sequence number If it discovers it has already passed that proof of +work and allocated a channel, moves that proof of work to the head of list, +so that the next check will be instant, just in case it is about to receive a +million copies of that proof of work. Then it checks for revealed bits from +those generated by XChaCha20. Then it checks the work and the +Poly1305 authentication. + +Checking if there is already a channel allocated overlaps and intersects +with presence notification protocol. We want to have a very large number +of inactive presences without secrets or network addresses in the database, +a large number of long lived active presences in memory, with secrets that +are not paged to disk (`sodium_allocarray`), and considerably smaller +number of considerably shorter lived channels with flow control and +buffering. A presence can only exchange short messages that fit in one +packet, and only one message can be active in any round trip time. You +open a presence, and the presence can then open a channel. + +We probably want to do the checks in whatever order is empirically most +efficient for type of DDoS attacks that we encounter in practice, the most +common probably being garbage random values that bear no particular +resemblance to valid connection attempts. + +The next problem will valid connections that then make excessive +demands. These get deranked by the next layer, and they will then have to +make a new connection, which will face increasing pow and discrimination +against their network address. + +## replacement handshake for limited circulation server + +In this case the server is the gateway for a group, possibly many groups, +whose unique id is not widely known. It is analogous to a closely kept email address. + +The the TCP replacement handshake needs to be a four phase handshake. + +1. Client->Server: Give me a connection, here are my parameters, + here is a clue about what private group I want to connect to. + +1. Server->Client: Here is a proof of work request, my parameters, + including a use once elliptic point, and a keyed hash of your and + my parameters. Ask again with proof of work, the same parameters, + and the keyed hash. + + Server then throws away the request, allocating no memory. + +1. Client->Server: OK, here I am again, with all that stuff you asked for. + + At this point, client has given server a clue about which private + group it wants to connect to, and server has given client a clue + about which private group it expects membership of, and therefore + what public key the client should attempt to communicate with. + +1. Server checks the keyed hash to ensure that this is a real client + reply to a real and recent server reply. Then it checks the proof of + work. + + If the proof of work passes, Server allocates memory + + Then it generates a transient secret from the konces (keys used + once, single use elliptic points), and uses it to decrypt the clien + durable public key, verifying that the client does indeed know the + transient scalar. If the client durable key is OK, sign on allowed, it + constructs a shared secret from all four keys, the sum of two secrets + multiplying the sum of two elliptic points, and we now have an + encrypted stream associated with the port number and network addresses. + +# Summary of the replacement + +Thus we can integrate TCP handshake and encryption hand shake and the +innumerable DDoS protection handshakes “Cloudfare is checking your browser, +oops, your browser did not pass, here is a captcha” at the cost of one single +additional trip, half a round trip. + +Instead of the person establishing the connection fuming while round trip +after round trip goes through, we get all that stuff at the cost of one +additional half round trip. + +# messages, not streams + +TCP sockets are designed for synchronous procedural programming, on +machines with very limited memory processing limitless streams. They are +now almost always used for message processing from event oriented +asynchronous code, with a messaging layer on top of the endless stream +layer. The replacement needs to have application layer sending messages +and receiving messages in events. The application layer should not have +to deal with sockets and streams. Rather, it sends a message to destination +identified by its durable public key, and gets a reply, where the reply +might be that the socket could not be opened, or that the socket was open but +the reply timed out, among other things. When sending a message, there is a +time to wait for response before giving up, and a time for the socket that +may be created to live idle. + +# Proposed replacement + +[QUIC] is the current TCP replacement. Also known as HTTP/3 + +[QUIC]: https://github.com/private-octopus/picoquic + +We have no alternative but to interface to the vast HTTP/2 HTTP/3 +ecosystem. The wallet is going to have to talk as a client to legacy server +http/3 devices, and accept their CA certificates, preferably subject to +Zooko scrutiny, and legacy http/3 client devices are going to have to talk +to our wallet (after their wallet has downloaded a zooko based certificate +from the server wallet). + +Talking HTTP/3 means being wide open to DDOS attack, so that you are +forced to use cloudfare. When a device with our version of QUIC talks to +another device with our version of QUIC, it has to implement our DDOS +resistance, and Zooko in place of CA. But when it talks to a legacy +HTTP/3 device, it has to lay itself wide open to DDOS attack and CA +interception. + +Backwards compatibility with insecure systems always creates a massive +security hole. On the one hand, every build from scratch project dies. On +the gripping hand, every attempt to do fax over the internet failed and was +eventually replaced by pdf attachments to email. Backwards compatibility +was simply too crippling, and backwards compatibility with QUIC is +going to cripple security. + +Instead of putting the secure system transparently as an alternate protocol +within the insecure system, you non transparently put the insecure system +as a downgrade protocol within the secure system, which means our +version of QUIC simply is not going to talk to older versions of QUIC +unless you take some special measures to tell it to do so or enable it to do +so for that particular communication end point. + +The least friction interface would be that every time a new SSL name is +encountered, we get a window saying "This authority claims that this is +this entity. Trust this authority for this entity?" And if there is a change of +authority, complain. Wrap backwards compatibility in Zooko vouched +certificates, pinned certificates, and the CAA record indicating who is the +right issuer for the SSL certificate + +We have to have downgrade capability, but it has to be an afterthought, +slipped in as a special path and special case, as user friendly as possible, +but no friendlier. + +QUIC's one way streams are messages. + +Its two way streams are backwards compatibility with TCP + +It solves the long fat pipe problem with flexible window size. + +It puts multiple objects and messages in one stream, so that one message +does not have to wait for lost packets in another message to be resolved. + +TCP flow control is constructed around pushback - that the sender should +not send data faster than the receiver is able and willing to handle it. +Normally there is one thread, or pool of of threads, handling the data +received. To prevent DDoS, we should probably only have one unit of +pushback per pair of network addresses. If someone has a slow receiver +thread pool, and a fast receiver thread pool communicating with the same +machine, he needs to break the slow receiver communication into lots of +small requests and replies, hence one channel per pair of network +addresses. + +Quic implements everything you need to have one channel per pair of +network addresses, multiplexing many request-replies into a single stream, +many channels in one channel, but does not in fact implement one channel +per pair of network addresses in the sense of one unit of packet flow +control and one unit of DDoS monitoring, per pair of network addresses. + +Finer grained flow control should be implemented as request reply on +messages that may well be much larger than a packet, but much smaller than +memory + +In the request reply model, if the requests and replies are reasonably short, +pushback does not matter, and becomes a representation of flow control. It +is seldom sane to download enormous blocks of data as a single message, +and we probably just should not do it - restrict replies to what can +reasonably fit into memory, so that a very large message that the receiver +is processing one chunk at a time has to get acks of its submessages, +separate from the flow control system. + +What the LEMP stack does with request headers is dynamically allocate +8KiB buffers, stuff headers into a part or whole of at 8KiB buffer, and if a +header is bigger than 8KiB, arbitrarily truncates it, which suggests that this +is a tactic to minimize the overheads of dynamically allocating many +moderate sized buffers of variable size. Experimenting, I find that +dynamic allocation tends to be the major cost in many programs, but if +you do it LEMP style, dynamic allocation is unlikely to be a significant cost. + +QUIC has a pile of feature bloat: + ++ The push feature is married to html, and belongs in the webserver + and the browser, not in the protocol. Something sending a request + message should be aware it might have several messages in reply, + depending on the kind of the request, and simply have a message + handler that can deal with many messages. + ++ We don’t really need the unique and sequential message id if finding and + interpreting the message id is part of how to response handler handles the + messages – best to hand that as far down into the endpoints as possible. + ++ its data format, header and frames, is married to html, which is + always sending repetitious and redundant information, treating + related fragments of html as absolutely distinct. + it implements html specific compression, HPACK. + +It suffers from the SSL/TLS problem of a thousand CA authorities, NSA +friendly encryption, and, being funded in large part by Cloudfare, has no +substantial defense against DDoS. + +It fails to support rendezvous routing. + +But, it has already struggled with and solved a thousand problems whose +solutions I have been confusedly struggling with. So the obvious solution +is to adopt Quic, rip out the domain name system, add DDoS resistance, +rip out NSA friendly encryption in favour of the standard and +recommended Libsodium packet encryption. (XChaCha20-Poly1305), for +immortality rip out the 62 bit compressed integers in favour of unlimited +precision windowed integers (With a negotiated limit on precision that +will in practice always be 64 bits for the next several centuries.) + +XChaCha20 is not the fastest on a long stream, but it has key agility, can +encrypt arbitrary length values, including a single bit, and is as +fast as ChaCha20 without any limits on the nonce. + +Quic’s messaging is excessively married to HTTP. We need a generic +messaging system where every message has an short number indicating +destination handler, and you can generate a handler, code continuation, +and get number assigned to it on the fly, so that you can send a message, +and the reply goes to your code continuation. + +We need to lift as much of the [QUIC] design as possible, and also make things +act much like TCP, so that existing NATs will not notice anything has +changed. Thus packets will continue to be sent to and from a widely known +port that is usually below 1024 on the server, from a random port on the +client in the range 49152--65535. A connection will continue to require a +three phase handshake which creates a socket, albeit our sockets will be very +different. + +With a rendezvous, both peers will use the same socket in the range +1024-49151 + +The rendezvous handshake will look like the TCP handshake Syn Syn-Ack Ack, +but they will both send syn packets, both send syn-ack packets, and both +send ack packets. Their syn packets will be timed so that, if the timing +is done right, both are sent just before the other peer’s packet is +expected to be received. + +Our sockets will always have a shared secret associated, which proves +identity and enables encrypted communication, but which cannot be used to +prove identity to a third party. The initial handshake will exchange +transient secret keys, which will generate a transient durable secret, +which is used to encrypt the exchange of durable secret keys, which +establish a shared secret based on the both the durable and transient key, +establishing forward secrecy, and failing to establish identity to third +parties. + +Since setting up a shared secret is costly, this creates the opportunity to +syn flood attacks, therefore the syn-ack will always be a syn cookie, +structured rather like existing syn cookies, a cryptographic hash of the syn +based on an unshared secret known only to the server, plus it will always +have a proof of work request, which may be zero, and it will have a list of +supported protocols if the protocol proposed in the initial syn cookie is +unacceptable. The proof of work will be that the hash of the client ack +must have a certain number of zeros, and the ack +must contain the cryptographic cookie, and the data that the server checks +the cookie against. + +TCP was designed around the case of the client sending an endless stream of +characters, typed with one finger, to a program on the server. We are +going to design around message response, with responses not necessarily +returning in order. + +The client sends a message from a durable public key to a to a durable +public key. The creation and destruction of such connections is not +tightly linked to messaging. If connection exists, it is used. If it does +not exist, it is created. It may be torn down after a while of being +unused, but the tear down is not tightly linked to message completion + +In TCP a count is kept of bytes sent and bytes received, with an ack +counting as one byte. + +We need a count for each packet, since packets can arrive out of order, +repeated, or missing. The count values will be sequential nonces for the +encryption, and will start at one. As the count can potentially grow +quite large, the count value will be windowed, but, unlike TCP, the +windowed count represents a potentially much larger absolute count known +by both ends. + +Negotiating a window size is hard, since you do not really know in advance +what window size will be needed. The thirty two bit window is adequate for +all normal uses, but fails in special and important uses. + +We will specify the window size in each packet, with the high order bit of +each byte in the nonce indicating whether there is another seven bits in +the nonce window, so that we can dynamically adjust the window size. We +dynamically adjust the window size to big enough to exclude ambiguity. +Which for the first 128 packets, and on a connection that is not very busy, +all packets, will be seven windowed count bits and one window size bit. + +The window needs to be large enough to exclude the ambiguity of delayed +and duplicated packets wandering in late, so has to be several times +larger than the difference between the most recently acked value, and the +the value that will fill the reception window. Thirty two times larger +should be ample. At the start, there are no early packets capable of +wandering in late, so big enough to hold the full count always suffices. + +If `a` represents a recent nonce, `n` +represents the nonce, `w` represents the windowed nonce. and +`M` represents the window mask, communicated in each packet in +unary, then: + +`w = n&M` + +`n = (w − a)&M + a` + +We use a window large enough to give the same answer on both the most +recently acked nonce, and the most recently sent nonce. + +The nonce will serve the dual purpose of enabling the decryption of each +packet, and flow control. Each packet has a sequential nonce, we make sure +all packets are acked. Nonces on packets coming from the client refer to a +different shared secret than nonces on packets coming from + +## API + +To send a message, you will construct a response handler if you are +expecting a response, and then call the api with a network address, a +public key of the recipient, an identifying secret key and public key of +the sender, a timeout for attempting to connect, and flags permitting for +direct connection, rendezvous connection, retransmit, and store and +forward. If a response is expected for the message, give the expected +lifetime for the response handler, a nonce for the response handler and a +class identifier for the nonce. (the nonce only has to be unique within +the class). You will probably use a different nonce population for +messages that have to be handled promptly, messages that have to be +handled within a session, and non volatile nonces that survive between +sessions. Nonce populations can be windowed per class identifier, with a +window large enough to accommodate the timeout, and a different class +identifier for volatile and non volatile nonces. The nonce is used once +within a window and within a class, but can be re-used in another class +and another window. + +The application code is event oriented, like gui code. It is driven by a +message pump, with constructors creating event handlers, and the events +driving the event handler through the message pump, and event handler, on +being fired, creates new event handlers and fires old event handlers. + +When the application needs to perform a task that spans many events, it does +not call `yield` or `await,` but instead the event handler for each event +constructs or enables the next event handler. If it needs to push information +onto a stack between events, has its own explicit stack for its own multi +event task, or creates a linked list of event handlers. Non volatile event +handlers must be trivial C+ classes, therefore cannot contain an `std::stack`, + +State that would be on the stack in synchronous code is in the event +handler in asynchronous code. This potentially gets messy if you are +processing an endless stream of structured data whose structure is +orthogonal to message boundaries. Since we allow arbitrary length +messages, don’t do that. + +Notification of message failure may occur any time within the lifetime of +the response handler, but will mostly happen within the timeout for +attempting to connect. + +The usual flow of control will be create an event handler, assign a nonce +to it (fire it) and then it gets triggered when the event actually +happens, and is then usually destroyed. Events will usually create and +fire new events and trigger events that existed before they were created, +rather than changing their state. + +Below the api, additional messages, using low numbered message response +classes, may be constructed for encryption and flow control. If an +encrypted connection exists, it will use that without constructing +additional messages. If it does not exist, will construct it. + +Constructing a encrypted connection provides perfect forward secrecy +between one connection and the next by generate new random session keys +each time. + +## Reliability and flow control + +TCP achieves reliable transmission with acks and nacks. + +The original design simply acked that all bytes (not exactly bytes, because +acks and nacks are counted) had been received up to a certain byte. If the +transmitter has transmitted stuff, and not received an ack for what it +transmitted it sends a nack, after a timeout. The receiver may resend acks. + +This mechanism worked fine on short thin pipes, but if you have a million +packets in flight, and packet three hundred thousand gets lost, you then +then have to send seven hundred thousand to replace one packet. So the +duplicate ack possibility was tortured to create a half assed version of +selective acknowledgment. If the receiver receives packet 100, and 101, +but not packet 99, it sends duplicate acks for packet 98. If the receiver +receives three duplicate acks for packet 98, it retransmits packet 99. (two +duplicate acks could be just the normal randomness.) + +[QUIC], however, has fix for this built in. + +Obviously true selective acknowledgment is better. The receiver acks the +most recent received packet, and sends a list of missing packets prior to +this (acks a windowed value for the most recent packet, and the difference +between packet nonces for missing packets) The sender resends the missing +packets, except for the most recent missing packets. If they are still +missing, they will be caught on the next ack. + +In each ack, the receiver tells the sender how much more data it can +receive before it sends the next ack. This prevents the receiver from +being flooded, but a more common problem is the pipe being flooded. + +To handle pipe flooding, the sender has a timer. If it sends stuff, and +does not get an ack, it backs off, it sets the timer to a slower rate, and +retransmits with a nack. The initial value of the timer is the initial +timer value is smoothed $RTT + max(G,4*RTT variance)$ + +TCP flow control focuses on getting a segment complete and acknowledged, +so it can move on to the next segments. It may have a great many packets +in flight, but does not have too many segments in flight. The backoff +algorithm is linked with the push segments algorithm. You only push the +segment the receiver has asked for in his previous acknowledgment. So you +typically have the segment you are finalizing, the segment that is in +flight, and the segment that the receiver asked for. + +The algorithm is that the sender gets an ack that acknowledges what the +receiver has received, and tells the sender how much more the receiver can +receive. Whereupon the sender resends anything missing, and resumes pushing +new stuff up to the limit that the receiver has specified, spread out +roughly evenly over the timer period. Which implies that the receiver +should ask wisely, as well as the sender send wisely. + +Implementing our own flow control sounds like a lot of work. Need to lift +[QUIC]’s flow control, and drop our own encryption and attack resistance +into it, while letting it worry about flow control. I can hack into its library, +while I cannot hack into the TCP library. + +I have been analysing how TCP works, with a view to what needs fixing. Time to +analyse how something works for which I have a library and example code. + +Best (because smallest and least married to HTTP3) is [picoquic]. + +[picoquic]: https://github.com/private-octopus/picoquic + +The TCP state machine assumes that the server opens a connection on receiving +a syn, sends an ack-syn to the client, whereupon the client acks the +connection. But if we are using syn cookies, we are using a different state +machine, where the connection is in fact only opened on receiving the server +syn-ack cookie in the client ack. So the server has to acknowledge the +connection, which would make it a four step handshake instead of a three step +handshake. To avoid this, we have a rule that the client only opens a +connection when it has data ready to send. It then gets a server cookie, and +sends the cookie-ack with some data, which data the server acks. + +With the cookie ack, we get a round trip time and offset between server +steady time and client steady time. If we see unstable round trip times, +we suspect the pipe is overloaded, and back off our estimate of max +bandwidth. For flow control, we maintain an estimate of pipe length and +width. Sudden pipe widenings indicate an overflow condition, because pipes +may respond to overflow by massively discarding packets, or massively +backing up packets, or quite possibly both. We maintain a probability +estimate of the pipe behaviour. + +## Outline protocol + +A packet protocol that establishes an encrypted connection on top of +unreliable packets with minimal round trips without increasing fragility to +DoS. + +For servers, public keys, globally human readable names, the key owning the +name, and the temporary key signed by the key owning the name, will usually +be public and widely known, but this also supports the case of +communication where this information is only known to the parties, and the +server does not want to make the connection between a network address and a +public key widely known. + +To establish a connection, we need to set a bunch of values specific to +this particular channel, and also create a shared secret that +eavesdroppers and active attackers cannot discover. + +The client is the part that initiates the communication, the server is +the party that responds. + +I assume a mode that provides both authentication and encryption – if a +packet decrypts into a valid message, this shows it originated from an +entity possessing the shared secret. This does not provide signing – the +recipient cannot prove to a third party that he received it, rather than +making it up. + +For the moment I ignore the hard question of server key distribution, +glibly invoking Zooko’s triangle without proposing an implementation of +the other two points and three sides of the triangle or a solution to the +problem of managing distributed reputations in Zooko’s triangle.  (Be +warned that whenever people charge ahead without solving the key +distribution problem, the result is a disaster.) + +Client 🠆 Server: Equivalent to the syn of the three phase TCP +handshake. + +> Client’s network address and port on which client will receive +> packets, protocol identifier, and client steady time that the +> message was sent. + +If the requested protocol is not OK, we go into protocol negotiation, +server responds with a list of protocols and protocol versions that it will +accept, in the form of a list of lists of numbers. + +Assuming it is OK, which it probably will be, server allocates nothing, +prepares nothing, but sends the equivalent of a TCP ack-syn cookie, +containing, among other things, a cryptographic hash of the information +that was received and sent, based on a private secret known only to the +server. It sends a transient public key, which changes every few minutes +or so, plus a short windowed id for that transient public key, and a demand +for proof of work, which may be zero. The proof of work is that the +client’s ack, equivalent of the third phase of the TCP handshake, has to +hash to a value ending in `n` zero bits, where `n` +may be zero. + +This cryptographic hash based on an unshared secret will be sent to client, +and then back to server, unchanged. Its function is to avoid the necessity for + the server to allocate memory or perform asymmetric cryptographic operations +for a client that has not yet validated. Instead the state information is sent + back and forth. + +1. Server 🠆 Client: Equivalent to the syn-ack of the three phase TCP handshake. + + Cryptographic hash based on unshared secret, server steady time, + transient public key, server windowed identifier of server transient + public key, proof of work demand, and any channel parameters. + + The proof of work is trivial if the server is not under load, but is + increased as the server load approaches the maximum the server is + capable of, in order to throttle demand. + + Client computes transient handshake shared secret as its transient private + key times the server shared transient public key. It returns in the clear + a copy of the cryptographic hash that the server sent to it, the data in + the clear needed to validate the hash, performs the proof of work, and + sends its public key, which may be a per server durable public key, always + used when accessing this server on this identity, encrypted using the + transient key, and the public key it wants to talk to on the server. + + Subsequent information is not encrypted using the transient keys, but using + the sum of transient plus secret keys. + + This implies that the client has to know the public key that the server is + using, which may be a key signed by the master public key that owns the + name authorizing that new key, which key changes about as often as the + server IP changes, and is therefore distributed in the same channel as the + network address associated with global human names is distributed. If the + client gets it wrong, then the server ignores the information encrypted to + the wrong public key, and responds with the authentication of its new + public key, signed by the master public key of its globally unique name, + encrypted using the transient secret – this is usually public information, + but since by this point we have established a shared secret and allocated + memory, might as well send it securely, for sometimes it is going to be + private information. + +1. Client 🠆 Server: Equivalent to the final ack of the three phase TCP +handshake. + + Sends in the clear server hash as received, any data needed to + reconstruct the hash, and transient secret key. Then, encrypted to + transient keys, the hash of the identifier of the public key it wants to + talk to, its durable public key, and client steady time at which this was + sent, so that both sides have an estimate of the round trip time and the + offset between server steady time and client steady time. + + Server checks the proof of work, checks the cryptographic hash against the + data in the clear, *then* creates an entry in its hash table for this + connection, with the shared secret being the transient keys plus the public + keys. + +We have two protocols, one for the authenticated phase, and one for +unauthenticated phase. The client has to know one of the unauthenticated +protocols offered by the server, or else protocol negotiation will fail in +the abnormal case that protocol negotiation is needed. Normally there will +only be one protocol for secured but unauthenticated communication during +setup, but we make provision by having two protocols, trivially different, +and three protocols, trivially different for the authenticated phase. + +You will notice that the server only allocates memory and and asymmetric +encryption computation *after* the client has successfully performed proof of +work and shown that it is indeed capable of receiving data sent to the +advertised network address. + +In the normal case, the client requests one way authenticated encryption in +the syn, where the server authenticates but the server does not, and the +server may, and usually will, offer in the syn-ack only two way +authenticated encryption, where the client provides an identity unique to +that server and user’s current default name, but which cannot be used to +identify the default name, nor the same user accessing a different +website. This allows the server to see that the same user is accessing +different resources, how many uniques the server has, and what each unique +is doing, but does not enable the server’s to put their heads together and +see that the same user is doing things on one server, and also on another +server. + +Now we have a shared secret, protocol negotiated, client logged in, in +one round trip plus the third one way trip carrying the actual data – the +same number of round trips as when setting up an unencrypted +unauthenticated TCP connection. + +You will notice there is no explicit step checking that both have the +same shared secret – This is because we assume that each packet sent is +also authenticated by the shared secret, so if they do not have the same +secret, nothing will authenticate. + +# Critiques of TCP/SSL + +Does the job so badly that using a different method is just as plausible. +People fight to avoid TLS already, they’d rather send stuff in the clear if +they could.  So just solve the problems they have. + +In Web Services we frequently require message layer security in addition to +transport layer security because a Web Service transaction might involve more +than two endpoints and messages that are stored and forwarded etc. This is why +WS-\* is not TLS. (It is unfortunately horribly baroque but that was not my +doing). + +Problem that occurred with TLS was that there was an assumption that the job\ +was to secure the reliable stream connection mechanics of TCP.  False +assumption. + +Pretty much nobody uses streams by design, they use datagrams.  And they use +them in a particular fashion: request-response.  Where we went wrong with TCP +was that this was the easiest way to handle the mechanics of getting the +response back to the agent that sent the request. Without TCP, one had to deal +with the raw incoming datagrams and allocate them to the different sending +agents. + +A second problem was that the design was too intertwined with commercial PKI +so certs were hung on the side as a millstone for server authentication and +discarded as client side, leaving passwords to fill that gap.  A mess, which +is an opportunity for redesign, frequently exploited by many designs already. + +SSL came at this and built a message (record) interface on top of TCP (because +that was convenient for defining a crypto layer), and then a (mainly) stream +interface on top of its message interface – because programmers were by now +familiar with streams, not records. + +And so … here we are.  Living in a city built on top of generations of +older cities.  Dig down and see the accreted layers. + +What *is* the “right” (easiest to use correctly, hardest to use +incorrectly, with good performance, across a large number of distinct +application APIs) underlying interface for a secure network link? The fact +that the first thing pretty much all APIs do is create a message structure +on top of TCP makes it clear that “pure stream” isn’t it.  Record-oriented +designs derived from 80-column punch cards are unlikely to be the answer +either.  What a “clean slate” interface would look like is an interesting +question, and perhaps it’s finally time to explore it. + +# General and unorganized comments + +µTP, Micro Transport Protocol is a Bittorrent near drop in replacement for TCP +that provides lower priority bulk downloads in the background. The library is +not well documented, (header file plus examples) but as far as I can see, +provides a reasonably clean separation between Bittorrent and the transport +mechanism. + +Google has a TCP/SSL replacement, [QUIC], which avoids round tripping and +renegotiation by integrating the security layer with the reliability layer, +and by supporting multiple asynchronous streams within a stream + +Layering a new peer-to-peer packet network over the Internet is simply +what the Internet is designed for. UDP is broken in a few ways, but not +that can’t be fixed. It’s simply a matter of time before a new virtual +packet layer is deployed – probably one in which authentication and +encryption are inherent. + +For authentication and encryption to be inherent, needs to connect +between public keys, needs to be based on Zooko’s triangle.  Also +needs to penetrate firewalls, and do protocol negotiation with an +unlimited number of possible protocols – avoiding that internet names and +numbers authority. + +Ian Grigg “Good protocols divide into two parts, the first of which says +to the second, trust this key completely!”. + +This might well be the basis of a better problem factorization than the +layer factorization – divide the task by the way trust is embodied, rather +than the basis of layered communication. + +Trust is an application level issue, not a communication layer issue, +but neither do we want each application to roll its own trust cryptography +– which at present web servers are forced to do. (Insert my standard rant +against SSL/TLS). + +Most web servers are vulnerable to attacks akin to session cookie +fixation attack, because each web page reinvents session cookie handling, +and even experts in cryptography are apt to get it wrong. + +The correct procedure is to generate and issue a strongly unguessable +random https only cookie on successful login, representing the fact that +the possessor of this cookie has proven his association with a particular +database record, but very few people, including very few experts in +cryptography, actually do it this way. Association between a client +request and a database record needs to be part of the security system. It +should not something each web page developer is expected to build on top +of the security system. + +TCP constructs a reliable pipeline stream connection out of unreliable +packet connections. + +There are a bunch of problems with TCP.  No provision was made for +protocol negotiation and so any upgrade has to be fully backwards +compatible.  A number of fixes have been made, for example the long +fat pipe problem has been fixed by window size negotiation, which is semi +incompatible and leads to flaky behaviour with old style routers, but the +transaction problem remains intolerable.  The transaction problem has +been reduced by protocol level workarounds, such as “Keep alive” for HTTP, +but these are not entirely satisfactory.  The fix for syn flooding +works, but causes some minor unnecessary degradation of performance under +syn flood attacks, because the syn cookie is limited to 48 bits – needs to +be 128 bits both to deal with the syn flood attack, and to prevent TCP +hijacking. + +TCP is inefficient over wireless, because interference problems are +rather different to those provided for in the TCP model.  This +problem is pretty much insoluble because of the lack of protocol +negotiation. + +There are cases intermediate between TCP and UDP, which require +different balances of timeliness, reliability, streaming, and record +boundary distinction. DCCP and SCTP have been introduced to deal with +these intermediate cases, SCTP for when one has many independent +transactions running over a single connection, and DCCP for data where +time sensitivity matters more than reliability such as voice over +IP.  SCTP would have been better for HTML and HTTP than TCP is, +though it is a bit difficult to change now.  Problems such as +password-authenticated key agreement transaction to a banking site require +something that resembles encrypted SCTP, analogous to the way that TLS is +encrypted TCP, but nothing like that exists as yet. Standards exist for +encrypted DCCP, though I think the standards are unsatisfactory and +suspect that each vendor will implement his own incompatible version, each +of which will claim to conform to the standard. + +But a new threat has arrived:  TCP man in the middle forgery. + +Connection providers, such as Comcast, frequently sell more bandwidth +than they can deliver.  To curtail customer demands, they forge +connection shutdown packets (reset packets), to make it appear that the +nodes are misbehaving, when in fact it is the connection between nodes, +the connection that Comcast provides, that is misbehaving. Similarly, the +great firewall of China forges reset packets when Chinese connect to web +sites that contain information that the Chinese government does not +approve of. Not only does the Chinese government censor, but it is able to +use a mechanism that conceals the fact of censorship. + +The solution to all these problems is to have protocol negotiation, +standard encryption, and flow control inside the encryption. + +A problem with the OSI Layer model is that as one piles one layer on top +of another, one is apt to get redundant round trips. + +According to [google research] 400 +milliseconds reduces usage by 0.76%, or roughly two percent per second of delay. + +[google research]: http://googleresearch.blogspot.com/2009/06/speed-matters.html + +Redundant round trips become an ever more serious problem as bandwidths +and processor speeds increase, but round trip times reminds constant, +indeed increase as we become increasingly global and increasingly rely on +space based communications. + +Used to be that the biggest problem with encryption was the asymmetric +encryption calculations – the PKI model has lots and lots of redundant and +excessive asymmetric encryptions. It also has lots and lots of redundant +round trips. Now that we can use the NVIDIA GPU with CUDA as a very high +speed cheap massively parallel cryptographic coprocessor, excessive PKI +calculations should become less of a problem, but excess round trips are +an ever increasing problem. + +Any significant authentication and encryption overhead will result in +people being too clever by half, and only using encryption and +authentication where it is needed, with the result that they invariably +screw up and fail to use it where it is needed – for example the login on +the http page. So we have to lower the cost of encrypted authenticated +communications, so that people can simply encrypt and authenticate +everything without needing to think about it. + +To get stuff right, we have to ditch the OSI layer model, but simply +ditching it without replacement will result in problems. It exists for a +reason, and we have to replace it with something else. diff --git a/docs/scale_clients_trust.md b/docs/scale_clients_trust.md new file mode 100644 index 0000000..bba1eb7 --- /dev/null +++ b/docs/scale_clients_trust.md @@ -0,0 +1,312 @@ +--- +title: Scaling, trust and clients +--- + +# Client trust + +When there are billions of people using the blockchain, it will inevitably +only be fully verified by a few hundred or at most a few thousand major +peers, who will inevitably have [interests that do not necessarily coincide] +with those of the billions of users, who will inevitably have only client +wallets. + +[interests that do not necessarily coincide]:https://vitalik.ca/general/2021/05/23/scaling.html +"Vitalik Buterin talks blockchain scaling" + +And a few hundred seems to be the minimum size required to stop peers +with a lot of clients from doing nefarious things. At scale, we are going to +approach the limits of distributed trust. + +There are several cures for this. Well, not cures, but measures that can +alleviate the disease + +None of these are yet implemented, and we will not get around to +implementing them until we start to take over the world. But it is +necessary that what we do implement be upwards compatible with this scaling design: + +## proof of stake + +Make the stake of a peer the value of coins (unspent transaction outputs) +that were injected into the blockchain through that peer. This ensures that +the interests of the peers will be aligned with the whales, with the interests +of those that hold a whole lot of value on the blockchain. Same principle +as a well functioning company board. A company board directly represents +major shareholders, whose interests are for the most part aligned with +ordinary shareholders. (This is apt to fail horribly when an accounting or +law firm is on the board, or a converged investment fund.) This measure +gives power the whales, who do not want their hosts to do nefarious things. + +## client verification + +every single client verifies the transactions that it is directly involved in, +and a subset of the transactions that gave rise to the coins that it receives. + +If it verified the ancestry of every coin it received all the way back, it +would have to verify the entire blockchain, but it can verify the biggest +ancestor of the biggest ancestor and a random subset of ancestors, thus +invalid transactions are going immediately generate problems. If every +client unpredictably verifies a small number of transactions, the net effect +is going to be that most transactions are going to be unpredictably verified +by several clients. + +## sharding, many blockchains + +Coins in a shard are shares in [sovereign cipher corporations] whose +primary asset is a coin on the primary blockchain that vests power over +their name and assets in a frequently changing public key. Every time +money moves from the main chain to a sidechain, or from one sidechain to +another, the old coin is spent, and a new coin is created. The public key on +the mainchain coin corresponds to [a frequently changing secret that is distributed] +between the peers on the sidechain in proportion to their stake. + +The mainchain transaction is a big transaction between many sidechains, +that contains a single output or input from each side chain, with each +single input or output from each sidechain representing many single +transactions between sidechains, and each single transaction between +sidechains representing many single transactions between many clients of +each sidechain. + +The single big mainchain transaction merkle chains to the total history of +each sidechain, and each client of a sidechain can verify any state +information about his sidechain against the most recent sidechain +transaction on the mainchain, and routinely does. + +## lightning layer + +The [lightning layer] is the correct place for privacy and contracts – because +we do not want every transaction, let alone every contract, appearing on +the mainchain. Keeping as much stuff as possible *off* the blockchain helps +with both privacy and scaling. + +## zk-snarks + +Zk-snarks are not yet a solution. They have enormous potential +benefits for privacy and scaling, but as yet, no one has quite found a way. + +A zk-snark is a succinct proof that code *was* executed on an immense pile +of data, and produced the expected, succinct, result. It is a witness that +someone carried out the calculation he claims he did, and that calculation +produced the result he claimed it did. So not everyone has to verify the +blockchain from beginning to end. And not everyone has to know what +inputs justified what outputs. + +The innumerable privacy coins around based on zk-snarks are just not +doing what has to be done to make a zk-snark privacy currency that is +viable at any reasonable scale. They are intentionally scams, or by +negligence, unintentionally scams. All the zk-snark coins are doing the +step from set $N$ of valid coins, valid unspent transaction outputs, to set +$N+1$, in the old fashioned Satoshi way, and sprinkling a little bit of +zk-snark magic privacy pixie dust on top (because the task of producing a +genuine zk-snark proof of coin state for step $N$ to step $N+1$ is just too big +for them). Which is, intentionally or unintentionally, a scam. + +Not yet an effective solution for scaling the blockchain, for to scale the +blockchain, you need a concise proof that any spend in the blockchain was +only spent once, and while a zk-snark proving this is concise and +capable of being quickly evaluated by any client, generating the proof is +an enormous task. Lots of work is being done to render this task +manageable, but as yet, last time I checked, not manageable at scale. +Rendering it efficient would be a total game changer, radically changing +the problem. + +The fundamental problem is that in order to produce a compact proof that +the set of coins, unspent transaction outputs, of state $N+1$ was validly +derived from the set of coins at state $N$, you actually have to have those +sets of coins, which is not very compact at all, and generate a compact +proof about a tree lookup and cryptographic verification for each of the +changes in the set. + +This is an inherently enormous task at scale, which will have to be +factored into many, many subtasks, performed by many, many machines. +Factoring the problem up is hard, for it not only has to be factored, divided +up, it has to be divided up in a way that is incentive compatible, or else +the blockchain is going to fail at scale because of peer misconduct, +transactions are just not going to be validated. Factoring a problem is hard, +and factoring that has to be mindful of incentive compatibility is +considerably harder. I am seeing a lot of good work grappling with the +problem of factoring, dividing the problem into manageable subtasks, but +it seems to be totally oblivious to the hard problem of incentive compatibility at scale. + +Incentive compatibility was Satoshi's brilliant insight, and the client trust +problem is failure of Satoshi's solution to that problem to scale. Existing +zk-snark solutions fail at scale, though in a different way. With zk-snarks, +the client can verify the zk-snark, but producing a valid zk-snark in the +first place is going to be hard, and will rapidly get harder as the scale +increases. + +A zk-snark that succinctly proves that the set of coins (unspent transaction +outputs) at block $N+1$ was validly derived from the set of coins at +block $N$, and can also prove that any given coin is in that set or not in that +set is going to have to be a proof about many, many, zk-snarks produced +by many, many machines, a proof about a very large dag of zk-snarks, +each zk-snark a vertex in the dag proving some small part of the validity +of the step from consensus state $N$ of valid coins to consensus state +$N+1$ of valid coins, and the owners of each of those machines that produced a tree +vertex for the step from set $N$ to set $N+1$ will need a reward proportionate +to the task that they have completed, and the validity of the reward will +need to be part of the proof, and there will need to be a market in those +rewards, with each vertex in the dag preferring the cheapest source of +child vertexes. Each of the machines would only need to have a small part +of the total state $N$, and a small part of the transactions transforming state +$N$ into state $N+1$. This is hard but doable, but I am just not seeing it done yet. + +I see good [proposals for factoring the work], but I don't see them +addressing the incentive compatibility problem. It needs a whole picture +design, rather than a part of the picture design. A true zk-snark solution +has to shard the problem of producing state $N+1$, the set of unspent +transaction outputs, from state $N$, so it should also shard the problem of +producing a consensus on the total set and order of transactions. + +[proposals for factoring the work]:https://hackmd.io/@vbuterin/das +"Data Availability Sampling Phase 1 Proposal" + +### The problem with zk-snarks + +Last time I checked, [Cairo] was not ready for prime time. + +[Cairo]:https://starkware.co/cairo/ +"Cairo - StarkWare Industries Ltd." + +Maybe it is ready now. + +The two basic problems with zk-snarks is that even though a zk-snark +proving something about an enormous data set is quite small and can be +quickly verified by anyone, it requires enormous computational resources to +generate the proof, and how does the end user know that the verification +verifies what it is supposed to verify? + +To solve the first problem, need distributed generation of the proof, +constructing a zk-snark that is a proof about a dag of zk-snarks, +effectively a zk-snark implementation of the map-reduce algorithm for +massive parallelism. In general map-reduce requires trusted shards that +will not engage in Byzantine defection, but with zk-snarks they can be +untrusted, allowing the problem to be massively distributed over the +internet. + +To solve the second problem, need an [intelligible scripting language for +generating zk-snarks], a scripting language that generates serial verifiers +and massively parallel map-reduce proofs. + +[intelligible scripting language for +generating zk-snarks]:https://www.cairo-lang.org +"Welcome to Cairo +A Language For Scaling DApps Using STARKs" + +Both problems are being actively worked on. Both problems need a good deal +more work, last time I checked. For end user trust in client wallets +relying on zk-snark verification to be valid, at least some of the end +users of client wallets will need to themselves generate the verifiers from +the script. + +For trust based on zk-snarks to be valid, a very large number of people +must themselves have the source code to a large program that was +executed on an immense amount of data, and must themselves build and +run the verifier to prove that this code was run on the actual data at least +once, and produced the expected result, even though very few of them will +ever execute that program on actual data, and there is too much data for +any one computer to ever execute the program on all the data. + +Satoshi's fundamental design was that all users should verify the +blockchain, which becomes impractical when the blockchain approaches four +hundred gigabytes. A zk-snark design needs to redesign blockchains from +the beginning, with distributed generation of the proof, but the proof for +each step in the chain, from mutable state $N$ to mutable state $N+1$, from set +$N$ of coins, unspent transaction outputs, to set $N+1$ of coins only being +generated once or generated a quite small number of times, with its +generation being distributed over all peers through map-reduce, while the +proof is verified by everyone, peer and client. + +For good verifier performance, with acceptable prover performance, one +should construct a stark that can be verified quickly, and then produce +a libsnark that it was verified at least once ([libsnark proof generation +being costly], but the proofs are very small and quickly verifiable). + +At the end of the day, we still need the code generating and executing the +verification of zk-snarks to be massively replicated, in order that all +this rigmarole with zk-snarks and starks is actually worthy of producing +trust. + +[libsnark proof generation being costly]:https://eprint.iacr.org/2018/046.pdf +"Scalable computational integrity: +section 1.3.2: concrete performance" + +This is not a problem I am working on, but I would be happy to see a +solution. I am seeing a lot of scam solutions, that sprinkle zk-snarks over +existing solutions as magic pixie dust, like putting wings on a solid fuel +rocket and calling it a space plane. + +[lightning layer]:lightning_layer.html + +[sovereign cipher corporations]:social_networking.html#many-sovereign-corporations-on-the-blockchain + +[a frequently changing secret that is distributed]:multisignature.html#scaling + +# sharding within each single very large peer + +Sharding within a single peer is an easier problem than sharding the +blockchain between mutually distrustful peers capable of Byzantine +defection, and the solutions are apt to be more powerful and efficient. + +When we go to scale, when we have very large peers on the blockchain, +we are going to have to have sharding within each very large peer, which will +multiprocess in the style of Google's massively parallel multiprocessing, +where scaling and multiprocessing is embedded in interactions with the +massively distributed database, either on top of an existing distributed +database such as Rlite or Cockroach, or we will have to extend the +consensus algorithm so that the shards of each cluster form their own +distributed database, or extend the consensus algorithm so that peers can +shard. As preparation for the latter possibility, we need to have each peer +only form gossip events with a small and durable set of peers with which it +has lasting relationships, because the events, as we go to scale, tend to +have large and unequal costs and benefits for each peer. Durable +relationships make sharding possible, but we will not worry to much about +sharding until a forty terabyte blockchain comes in sight. + +When we go to scale, we are going to have to have sharding, which will +multiprocess in the style of Google’s massively parallel multiprocessing, +where scaling and multiprocessing is embedded in interactions with the +massively distributed database, either on top of an existing distributed +database such as Rlite or Cockroach, or we will have to extend the +consensus algorithm so that the shards of each cluster form their own +distributed database, or extend the consensus algorithm so that peers can +shard. As preparation for the latter possibility, we need to have each peer +only form gossip events with a small and durable set of peers with which it +has lasting relationships, because the events, as we go to scale, tend to +have large and unequal costs and benefits for each peer. Durable +relationships make sharding possible, but we will not worry to much about +sharding until a forty terabyte blockchain comes in sight. + +For sharding, each peer has a copy of a subset of the total blockchain, and +some peers have a parity set of many such subsets, each peer has a subset +of the set of unspent transaction outputs as of consensus on total order at +one time, and is working on constructing a subset of the set of unspent +transactions as of a recent consensus on total order, each peer has all the +root hashes of all the balanced binary trees of all the subsets, but not all +the subsets, each peer has durable relationships with a set of peers that +have the entire collection of subsets, and two durable relationships with +peers that have parity sets of all the subsets. + +Each subset of the append only immutable set of transactions is represented +by a balanced binary tree of hashes representing $2^n$ blocks of +the blockchain, and each subset of the mutable set of unspent transaction +outputs is a subsection of the Merkle-patricia tree of transaction outputs, +which is part of a directed acyclic graph of all consensus sets of all past +consensus states of transaction outputs, but no one keeps that entire graph +around once it gets too big, as it rapidly will, only various subsets of it. + +But they keep the hashes around that can prove that any subset of it was +part of the consensus at some time. + +Gossip vertexes immutable added to the immutable chain of blocks will +contain the total hash of the state of unspent transactions as of a previous +consensus block, thus the immutable and ever growing blockchain will contain +an immutable record of all past consensus Merkle-patricia trees of +unspent transaction outputs, and thus of the past consensus about the +dynamic and changing state resulting from the immutable set of all past +transactions + +For very old groups of blocks to be discardable, it will from time to time be +necessary to add repeat copies of old transaction outputs that are still +unspent, so that the old transactions that gave rise to them can be +discarded, and one can then re-evaluate the state of the blockchain starting +from the middle, rather than the very beginning. diff --git a/docs/scriptless_scripts.pdf b/docs/scriptless_scripts.pdf new file mode 100644 index 0000000..a189cad Binary files /dev/null and b/docs/scriptless_scripts.pdf differ diff --git a/docs/secret_handshakes.pdf b/docs/secret_handshakes.pdf new file mode 100644 index 0000000..bee6e38 Binary files /dev/null and b/docs/secret_handshakes.pdf differ diff --git a/docs/set_up_build_environments.md b/docs/set_up_build_environments.md new file mode 100644 index 0000000..31ff95e --- /dev/null +++ b/docs/set_up_build_environments.md @@ -0,0 +1,3212 @@ +--- +title: + Set up build environments +--- +# Virtual Box + +To build a cross platform application, you need to build in a cross +platform environment. + +## Setting up Ubuntu in Virtual Box + +Having a whole lot of different versions of different machines, with a +whole lot of snapshots, can suck up a remarkable amount of disk space +mighty fast. Even if your virtual disk is quite small, your snapshots +wind up eating a huge amount of space, so you really need some capacious +disk drives. And you are not going to be able to back up all this +enormous stuff, so you have to document how to recreate it. + +Each snapshot that you intend to keep around long term needs to +correspond to a documented path from install to that snapshot. + +When creating a Virtual Box machine, make sure to set the network +adapter to paravirtualization, set preferences in the file menu, the virtual +hard disk, and the snapshot directory to the desired location.  Virtual hard +disk location selection is done when creating it, snapshot directory is done +in settings/general/advanced (which also allow you to do clipboard sharing). + +```bash +apt-get -qy update && apt-get -qy upgrade +# Fetches the list of available updates and + # Strictly upgrades the current packages +``` + +To install guest additions, thus allow full communication between host +and virtual machine, update Ubuntu, hen while Ubuntu is running, +simulate placing the guest additions CD in the simulated optical drive. +Then Ubuntu will correctly activate and run the guest additions +install. + +Installing guest additions [frequently runs into +trouble](https://blog.sugoi.be/virtualbox-guest-additions-common-errors.html). +Debian especially tends to have security in place to stop random people +from sticking in CDs that get root access to the OS to run code to amend +the OS in ways the developers did not anticipate. + +## Setting up Debian in Virtual Box + +To install guest additions on Debian: + +```bash +su -l root +apt-get -qy update && apt-get -qy install build-essential module-assistant git dialog rsync && m-a -qi prepare +apt-get -qy upgrade +mount -t iso9660 /dev/sr0 /media/cdrom +cd /media/cdrom0 && sh ./VBoxLinuxAdditions.run +usermod -a -G vboxsf cherry +``` + +You will need to do another `m-a prepare` and to reinstall it after a +`apt-get -qy dist-upgrade`. Sometimes you need to do this after a mere +upgrade to Debian or to Guest Additions. Every now and then, guest +additions gets mysteriously broken on Debian, due to automatic operating +system updates in the background, the system will not shut +down correctly, and guest additions has to be reinstalled with a +`shutdown -r`. Or copy and paste mysteriously stops working. + +On Debian lightdm mate go to system/ control center/ Look and Feel/ Screensaver and turn off the screensaver screen lock + +Go to go to system / control center/ Hardware/ Power Management and turn off the computer and screen sleep. + +To set automatic login on lightdm-mate + +```bash +nano /etc/lightdm/lightdm.conf +``` + +In the `[Seat:*]` section of the configuration file (there is another section of this configuration file where these changes have no apparent effect) edit + +```ini +#autologin-guest=false +#autologin-user=user +#autologin-user-timeout=0 +``` + +to + +```ini +autologin-guest=false +autologin-user=cherry +autologin-user-timeout=0 +``` + +In the shared directory, I have a copy of /etc and ~.ssh ready to roll, so I just go into the shared directory copy them over, `chmod` .ssh and reboot. + +On the source machine + +```bash +scp -r .ssh «destination»:~ +scp -r etc «destination»:/ +``` + +On the destination machine + +```bash +chmod 700 .ssh && chmod 600 .ssh/* +``` + +I cannot do it all from within the destination machine, because linux cannot follow windows symbolic links. + +### Set the hostname + +check the hostname and dns domain name with + +```bash +hostname && domainname -s && hostnamectl status +``` + +And if need be, set them with + +```bash +domainname -b reaction.la +hostnamectl set-hostname reaction.la +``` + +Your /etc/hosts file should contain + +```text +127.0.0.1 localhost +127.0.0.1 reaction.la +# The following lines are desirable for IPv6 capable hosts +::1 ip6-localhost ip6-loopback +fe00::0 ip6-localnet +ff00::0 ip6-mcastprefix +ff02::1 ip6-allnodes +ff02::2 ip6-allrouters +ff02::3 ip6-allhosts +``` + +To change the host ssh key, so that different hosts have different +hostnames after I copied everything to a new instance: + +```bash +cd /etc/ssh +cat sshd* |grep HostKey +#Make sure that `/etc/ssh/sshd_config` has the line +# HostKey /etc/ssh/ssh_host_ed25519_key +rm -v ssh_host* +ssh-keygen -t ed25519 -f /etc/ssh/ssh_host_ed25519_key +``` + +Note that visual studio remote compile requires an `ecdsa-sha2-nistp256` key on the host machine that it is remote compiling for. If it is nist, it is +backdoored + +If the host has a domain name, the default in `/etc/bash.bashrc` will not display it in full at the prompt, which can lead to you being confused about which host on the internet you are commanding. + +```bash +nano /etc/bash.bashrc +``` + +Change the lower case `h` in ` PS1='${debian_chroot:+($debian_chroot)}\u@\h:\w\$ '` to an upper case `H` + +```text + PS1='${debian_chroot:+($debian_chroot)}\u@\H:\w\$ ' +``` + +And, similarly, in two places in `etc/skel/.bashrc` Also + +```bash +cp -rv ~/.ssh /etc/skel +``` + +# Actual server + +Setting up an actual server is similar to setting up the virtual machine +modelling it, except you have to worry about the server getting overloaded +and locking up. + +If a server is configured with an [ample swap file] an overloaded server will +lock up and have to be ungracefully powered down, which can corrupt the data +on the server. If the swap file is inadequate, the OOM killer will shut +down processes, which is also very bad, but does not risk losing data. So +by default, servers tend to be out of the box configured with a grossly +inadequate swap file, so that they will fail gracefully under overload, +rather than locking up, needing to be powered down, and then needing to +be recreated from scratch because of data corruption. + +This looks to me like a kernel defect. The kernel should detect when it is +thrashing the swap file, and respond by sleeping entire processes for +lengthy and growing periods, and logging these abnormally long sleeps +on wake. Swapping should never escalate to lockup, and if it does, bad +memory management design, though this misfeature seems common to +most operating systems. + +I prefer an ample swap file, larger than total memory, plus [thrash protect], +which will result in comparatively graceful degradation, plus the existence of +the file `/tmp/thrash-protect-frozen-pid-list` will tell you that your +overloaded server is degrading (if it is not degrading, the file exists only briefly). + +[thrash protect]:https://github.com/tobixen/thrash-protect +{target="_blank"} + +[ample swap file]:https://linuxhandbook.com/increase-swap-ubuntu/ +"How to Increase Swap Size on Ubuntu Linux" +{target="_blank"} + +## VM pretending to be cloud server + +To have it look like a cloud server, but one you can easily snapshot and +restore, set it up in bridged mode. Note the Mac address. After having +it is running as a normal system, and you can browse the web with it, +after guest additions and all that, then shut it down, go to your +router, and give it a new static IP and a new entry in hosts. + +Then configure ssh access to root. so that you can go `ssh `as +if on your real cloud system. See setting up a [server in the cloud](#setting-up-a-headless-server-in-the-cloud) + +On a system that only I have physical access to and which runs no services +that can be accessed from outside my local network my username is +always the same and the password always a short easily guessed single +word. Obviously if your system is accessible to the outside world, you +need a strong password. An easy password could be potentially really bad +if we have openssh-server installed, and ssh can be accessed from outside. +If building a headless machine with openssh-server (the typical cloud or +remote system) then need to set up public key sign in only, if the machine +should contain anything valuable.  Passwords are just not good enough – +you want your private ssh key on a machine that only you have physical +access to, and runs no services that anyone on the internet has access to, +and which you don’t use for anything that might get it infected with +malware, and you use that private key to access more exposed machines +by ssh public key corresponding to that private key. + +```bash +apt-get -qy update && apt-get -qy upgrade +# Fetches the list of available updates and +# strictly upgrades the current packages +``` + +To automatically start virtual boxes on bootup, which we will need to do +if publishing them, Open VirtualBox and right click on the VM you want +to autostart, click the option to create a shortcut on the desktop, cut +the shortcut. Open the windows 10“Run” box (Win+R) and enter +shell:startup Paste the shortcut. But all this is far too much work if +we are not publishing them. + +If a virtual machine is always running, make sure that the close default +is to save state, for otherwise shutdown might take too long, and +windows might kill it when updating. + +If we have a gui, don’t do openssh. Terminal comes up with Ctrl Alt T + +# Directory Structure + +## Linux + +`/usr` +: Secondary hierarchy for read-only user data; contains the majority +of (multi-)user utilities and applications. + +> `/usr/bin` +> : Non-essential command binaries (not needed in single user mode); +> for all users. +> +> `/usr/include` +> : Standard include files grouped in subdirectories, for example +> `/usr/include/boost` +> +> `/usr/lib` +> : Libraries for the binaries in /usr/bin and /usr/sbin. +> +> > `/usr/lib` +> > : Alternate format libraries, e.g. /usr/lib32 for 32-bit libraries +> > on a 64-bit machine (option) +> +> `/usr/local` +> : Tertiary hierarchy for local data, specific to this host. +> Typically has further subdirectories, e.g., bin, lib, share. +> +> `/usr/sbin` +> : Non-essential system binaries, e.g., daemons for various +> network-services. Blockchain daemon goes here. +> +> `/usr/share` +> : Architecture-independent (shared) data. Blockchain goes in a +> subdirectory here. +> +> `/usr/src` +> : Source code. Generally release versions of source code. Source +> code that the particular user is actively working on goes in the +> particular user’s `~/src/` directory, not this directory. +> +> `~/.` +> : Data maintained by and for specific programs for the particular +> user, for example in unix `~/.Bitcoin` is the equivalent of +> `%APPDATA%\Bitcoin` in Windows. +> +> `~/.config/` +> : Config data maintained by and for specific programs for the +> particular user, so that the users home directory does not get +> cluttered with a hundred `.` directories. +> +> `~/.local/` +> : Files maintained by and for specific programs for the particular +> user. +> +> `~/src/` +> : Source code that you, the particular user, are actively working +> on, the equivalent of `%HOMEPATH%\src\` in Windows. +> +> > `~/src/include` +> > : header files, so that they can be referenced in your source code +> > by the expected header path, thus for example this directory will +> > contain, by copying or hard linking, the `boost` directory so that +> > standard boost includes work. + +# Setting up a headless server in the cloud + +## Setting up ssh + +Login by password is second class, and there are a bunch of esoteric +special cases where it does not quite 100% work in all situations, +because stuff wants to auto log you in without asking for input. + +Putty is the windows ssh client, but you can use the Linux ssh client in +windows in the git bash shell, and the Linux remote file copy utility +`scp` is way better than the putty utility PSFTP. + +Usually a command line interface is a pain and error prone, with a +multitude of mysterious and inexplicable options and parameters, and one +typo or out of order command causing your system to unrecoverably die,but even though Putty has a windowed interface, the command line +interface of bash is easier to use. + +It is easier in practice to use the bash (or, on Windows, git-bash) to manage keys than PuTTYgen. You generate a key pair with + +```bash +ssh-keygen -t ed25519 -f keyfile +``` + +(I don't trust the other key algorithms, because I suspect the NSA has been up to cleverness with the details of the implementation.) + +On windows, your secret key should be in `%HOMEPATH%/.ssh`, on linux +in `/home/cherry/.ssh`, as is your config file for your ssh client, listing +the keys for hosts. The public keys of your authorized keys are in +`/home/cherry/.ssh/authorized_keys`, enabling you to login from afar as +that user over the internet. The linux system for remote login is a cleaner +and simpler system that the multitude of mysterious, complicated, and +failure prone facilities for remote windows login, which is a major reason +why everyone is using linux hosts in the cloud. + +In Debian, I create the directory `~/.ssh` for the user, and, using the +editor nano, the file `authorized_keys` + +```bash +mkdir ~/.ssh +nano ~/.ssh/authorized_keys +chmod 700 .ssh +chmod 600 .ssh/* +``` + +I set the ssh session host IP under /Session, the auto login username +under /Connection/data, the autologin private key under +/Connection/ssh/Auth. + +If I need KeepAlive I set that under /Connection + +I make sure auto login works, which enables me to make `ssh` do all sorts of +things, then I disable ssh password login, restrict the root login to only be +permitted via ssh keys. + +In order to do this, open up the SSHD config file (which is ssh daemon +config, not ssh_config. If you edit this into the the ssh_config file +everything goes to hell in a handbasket. ssh_config is the global +.ssh/config file): + +```bash +nano /etc/ssh/sshd_config +``` + +Your config file should have in it + +```default +HostKey /etc/ssh/ssh_host_ed25519_key +X11Forwarding yes +AllowAgentForwarding yes +AllowTcpForwarding yes +TCPKeepAlive yes +AllowStreamLocalForwarding yes +GatewayPorts yes +PermitTunnel yes +PasswordAuthentication no +ChallengeResponseAuthentication no +UsePAM no +PermitRootLogin prohibit-password +ciphers chacha20-poly1305@openssh.com +macs hmac-sha2-256-etm@openssh.com +kexalgorithms curve25519-sha256 +pubkeyacceptedkeytypes ssh-ed25519 +hostkeyalgorithms ssh-ed25519 +hostbasedacceptedkeytypes ssh-ed25519 +casignaturealgorithms ssh-ed25519 +``` + +`PermitRootLogin` defaults to prohibit-password, but best to set it +explicitly Within that file, find the line that includes +`PermitRootLogin` and if enabled modify it to ensure that users can only +connect with their ssh key. + +`ssh` out of the box by default allows every cryptographic algorithm under the sun, but we know the NSA has been industriously backdooring cryptographic code, sometimes at the level of the algorithm itself, as with their infamous elliptic curves, but more commonly at the level of implementation and api, ensuring that secure algorithms are used in a way that is insecure against someone who has the backdoor, insecurely implementing secure algorithms. On the basis of circumstantial evidence +and social connections, I believe that much of the cryptographic code used +by ssh has been backdoored by the nsa, and that this is a widely shared +secret. + +They structure the api so as to make it overwhelmingly likely that the code +will be used insecurely, and subtly tricky to use securely, and then make +sure that it is used insecurely. It is usually not that the core algorithms are +backdoored, as that the backdoor is on a more human level, gently steering +the people using core algorithms into a hidden trap. + +The backdoors are generally in the interfaces between layers, the apis, +which are subtly mismatched, and if you point at the backdoor they say +"that is not a backdoor, the code is fine, that issue is out of scope. File a +bug report against someone else's code. Out of scope, out of scope." + +And if you were to file a bug report against someone else's code, they +would tell you they are using this very secure NSA approved algorithm +with the approved and very secure api, the details of the cryptography are +someone else's problem, "out of scope, out of scope", and they have +absolutely no idea what you are talking about, because what you are +talking about is indeed very obscure, subtle, complicated, and difficult to +understand. The backdoors are usually where one api maintained by one +group is using a subtly flawed api maintained by another group. + +The more algorithms permitted, the more places for backdoors. The +certificate algorithms are particularly egregious. Why should we ever +allow more than one algorithm, the one we most trust? + +Therefore, I restrict the allowed algorithms to those that I actually use, and +only use the ones I have reason to believe are good and securely +implemented. Hence the lines: + +```default +HostKey /etc/ssh/ssh_host_ed25519_key +ciphers chacha20-poly1305@openssh.com +macs hmac-sha2-256-etm@openssh.com +kexalgorithms curve25519-sha256 +pubkeyacceptedkeytypes ssh-ed25519 +hostkeyalgorithms ssh-ed25519 +hostbasedacceptedkeytypes ssh-ed25519 +casignaturealgorithms ssh-ed25519 +``` + +Not all ssh servers recognize all these configuration options, and if you +give an unrecognized configuration option, the server dies, and then you +cannot ssh in to fix it. But they all recognize the first three, `HostKey, +ciphers, macs` which are the three that matter the most. + +To put these changes into effect: + +```bash +shutdown -r now +``` + +Now that putty can do a non interactive login, you can use `plink` to have a +script in a client window execute a program on the server, and echo the +output to the client, and psftp to transfer files, though `scp` in the Git Bash +window is better, and `rsync` (Unix to Unix only, requires `rsync` running on +both computers) is the best. `scp` and `rsync`, like `git`, get their keys from +`~/.ssh/config` + +On windows, FileZilla uses putty private keys to do scp. This is a much +more user friendly and safer interface than using scp – it is harder to +issue a catastrophic command, but rsync is more broadly capable. + +Life is simpler if you run FileZilla under linux, whereupon it uses the same +keys and config as everyone else. + +All in all, on windows, it is handier to interact with Linux machines +using the Git Bash command window, than using putty, once you have set +up `~/.ssh/config` on windows. + +Of course windows machines are insecure, and it is safer to have your +keys and your `~/.ssh/config` on Linux. + +Putty on Windows is not bad when you figure out how to use it, but ssh +in Git Bash shell is better:\ +You paste stuff into the terminal window with right click, drag stuff +out of the terminal window with the mouse, you use nano to edit stuff in +the ssh terminal window. + +Once your you can ssh into your cloud server without a password, you now need to update it and secure it with ufw. You also need rsync, to move files around + +### Remote graphical access over ssh + +```bash +ssh -cX root@reaction.la +``` + +`c` stands for compression, and `X` for X11. + +-X overrides the per host setting in `~/.ssh/config`: + +```default + ForwardX11 yes + ForwardX11Trusted yes +``` + +Which overrides the `host *` setting in `~/.ssh/config`, which overrides the settings for all users in `/etc/ssh/ssh_config` + +If ForwardX11 is set to yes, as it should be, you do not need the X. Running a gui app over ssh just works. There is a collection of useless toy +apps, `x11-apps` for test and demonstration purposes. + +I never got this working in windows, because no end of mystery + configuration issues, but it works fine on Linux. + +Then, as root on the remote machine, you issue a command to start up the +graphical program, which runs as an X11 client on the remote +machine, as a client of the X11 server on your local machine. This is a whole lot easier than setting up VNC. + +If your machine is running inside an OracleVM, and you issue the +command `startx` as root on the remote machine to start the remote +machines desktop in the X11 server on your local OracleVM, it instead +seems to start up the desktop in the OracleVM X11 server on your +OracleVM host machine. Whatever, I am confused, but the OracleVM +X11 server on Windows just works for me, and the Windows X11 server +just does not. On Linux, just works. + +Everyone uses VNC rather than SSH, but configuring login and security +on VNC is a nightmare. The only usable way to do it is to use turn off all +security on VNC, use `ufw` to shut off outside access to the VNC host's port +and access the VNC host through SSH port forwarding. + +X11 results in a vast amount of unnecessary round tripping, with the result +that things get unusable when you are separated from the other compute +by a significant ping time. VNC has less of a ping problem. + +X11 is a superior solution if your ping time is a few milliseconds or less. + +VNC is a superior solution if your ping time is humanly perceptible, fifty +milliseconds or more. In between, it depends. + +I find no solution satisfactory. Graphic software really is not designed to be used remotely. Javascript apps are. If you have a program or +functionality intended for remote use, the gui for that capability has to be +javascript/css/html. Or you design a local client or master that accesses +and displays global host or slave information. + +The best solution if you must use graphic software remotely and have a +significant ping time is to use VNC over SSH. Albeit VNC always exports +an entire desktop, while X11 exports a window. Though really, the best solution is to not use graphic software remotely, except for apps. + +## Install minimum standard software on the cloud server + +```bash +apt-get -qy update && apt-get -qy install build-essential module-assistant dialog rsync ufw +cat /etc/default/ufw | sed 's/^\#*[[:blank:]]*MANAGE_BUILTINS[[:blank:]]*=.*$/MANAGE_BUILTINS=yes/g' >tempufw +mv tempufw /etc/default/ufw +chmod 600 /etc/default/ufw +ufw status verbose +ufw disable +ufw default deny incoming && ufw default allow outgoing +ufw allow ssh && ufw limit ssh/tcp +echo "Y +" |ufw enable && ufw status verbose +``` + +## Backing up a cloud server + +`rsync` is the openssh utility to synchronize directories locally and +remotely. + +Assume rsync is installed on both machines, and you have root logon +access by openssh to the `remote_host` + +Shutdown any daemons that might cause a disk write during backup, which +would be disastrous. Login as root at both ends or else files cannot be +accessed at one end, nor permissions preserved at the other. + +```bash +rsync -aAXvzP --delete remote_host:/ --exclude={"/dev/*","/proc/*","/sys/*","/tmp/*","/run/*","/media/*","/lost+found"} local_backup +``` + +Of course, being root at both ends enables you to easily cause +catastrophe at both ends with a single typo in rsync. + +To simply logon with ssh + +```bash +ssh remote_host +``` + +To synchronize just one directory. + +```bash +rsync -aAXvzP --delete remote_host:~/name . +``` + +To make sure the files are truly identical: + +```bash +rsync -aAXvzc --delete reaction.la:~/name . +``` + +`rsync, ssh, git` and so forth know how to logon from the +`~/.ssh/config`(not to be confused with `/etc/ssh/sshd_config` or +`/etc/ssh/ssh_config` + +```default +Host remote_host +HostName remote_host +Port 22 +IdentityFile ~/.ssh/id_ed25519 +User root +ServerAliveInterval 60 +TCPKeepAlive yes +``` + +Git on windows users `%HOMEPATH/.ssh/config` and that is how it knows +what key to use + +To locally do a backup of the entire machine, excluding of course your +`/local_backup` directory which would cause an infinite loop: + +```bash +rsync -raAvX --delete / + --exclude={"/dev/*","/proc/*","/sys/*","/tmp/*","/run/*","/local_backup/*",/ + "/media/*","/lost+found"} /local_backup +``` + +The a and X options means copy the exact file structure with permission +and all that recursively, The z option is for compression of data in +motion. The data is uncompressed at the destination, so when backing up +local data locally, we don’t use it. + +To locally just copy stuff from the Linux file system to the windows +file system + +```bash +rsync -acv --del source dest/ +``` + +Which will result in the directory structure dest/source + +To merge two directories which might both have updates: + +```bash + rsync -acv source dest/ +``` + +A common error and source of confusion is that: + +```bash +rsync -a dir1/ dir2 +``` + +means make dir2 contain the same contents as dir1, while + +```bash +rsync -a dir1 dir2 +``` + +is going to put a copy of dir1 inside dir2 + +Since a copy can potentially take a very long time, you need the -v +flag. + +The -P flag (which probably should be used with the -c flag) does +incremental backups, just updating stuff that has been changed. The -z +flag does compression, which is good if your destination is far away. + +## Apache + +To bring up +[apache](https://linuxize.com/post/how-to-install-apache-on-debian-9/) +[virtual +hosting](https://linuxize.com/post/how-to-set-up-apache-virtual-hosts-on-debian-9/) + +Apache2 html files are at `/var/www//`. + +Apache’s virtual hosts are:\ +`/etc/apache2/sites-available`\ +`/etc/apache2/sites-enabled` + +The apache2 directory looks like: + +```default + apache2.conf + conf-available + conf-enabled + envvars + magic + mods-available + mods-enabled + ports.conf + sites-available + sites-enabled +``` + +The sites-available directory looks like + +```default + 000-default.conf + reaction.la.conf + default-ssl.conf +``` + +The sites enabled directory looks like + +```default + 000-default.conf -> ../sites-available/000-default.conf + reaction.la-le-ssl.conf + reaction.la.conf +``` + +And the contents of reaction.la.conf are (before the https thingly has +worked its magic) + +```default + + ServerName reaction.la + ServerAlias www.reaction.la + ServerAlias «foo.reaction.la» + ServerAlias «bar.reaction.la» + ServerAdmin «me@mysite» + DocumentRoot /var/www/reaction.la + + + Options -Indexes +FollowSymLinks + AllowOverride All + + + ErrorLog ${APACHE_LOG_DIR}/reaction.la-error.log + CustomLog ${APACHE_LOG_DIR}/reaction.la-access.log combined + RewriteEngine on + RewriteCond %{HTTP_HOST} ^www\.reaction.la\.com [NC] + RewriteRule ^(.*)$ https://reaction.la/$1 [L,R=301] + +``` + +All the other files don’t matter. The conf file gets you to the named +server. The contents of /var/www/reaction.la are the html files, the +important one being index.html. + +To get free, automatically installed and configured, ssl certificates +and configuration + +```bash +apt-get -qy install certbot python-certbot-apache +certbot register --register-unsafely-without-email --agree-tos +certbot --apache +``` + +if you have set up http virtual apache hosts for every name supported by +your nameservers, and only those names, certbot automagically converts +these from http virtual hosts to https virtual hosts and sets up +redirect from http to https. + +If you have an alias server such as www.reaction.la for reaction.la, +certbot will guess you also have the domain name www.reaction.la and get +a certificate for that. + +Thus, after certbot has worked its magic, your conf file looks like + +```default + + ServerName reaction.la + ServerAlias foo.reaction.la + ServerAlias bar.reaction.la + ServerAdmin me@mysite + DocumentRoot /var/www/reaction.la + + + Options -Indexes +FollowSymLinks + AllowOverride All + + + ErrorLog ${APACHE_LOG_DIR}/reaction.la-error.log + CustomLog ${APACHE_LOG_DIR}/reaction.la-access.log combined + RewriteEngine on + RewriteCond %{HTTP_HOST} ^www\.example\.com [NC] + RewriteRule ^(.*)$ https://reaction.la/$1 [L,R=301] + RewriteCond %{SERVER_NAME} =reaction.la [OR] + RewriteRule ^ https://%{SERVER_NAME}%{REQUEST_URI} [END,NE,R=permanent] + +``` + +## Lemp stack on Debian + +```bash +apt-get -qy update && apt-get -qy install nginx mariadb-server php php-cli php-xml php-mbstring php-mysql php7.3-fpm +nginx -t +ufw status verbose +``` + +Browse to your server, and check that nginx web page is working. Your +browser will probably give you an error page, merely because it defaults +to https, and https is not yet working. Make sure you are testing http, not +https. We will get https working shortly.. + +### Mariadb and ufw + +```bash +ufw default deny incoming && ufw default allow outgoing +ufw allow ssh && ufw allow 'Nginx Full' && ufw limit ssh/tcp +# edit /etc/default/ufw so that MANAGE_BUILTINS=yes +cat /etc/default/ufw | sed 's/^\#*[[:blank:]]*MANAGE_BUILTINS[[:blank:]]*=.*$/MANAGE_BUILTINS=yes/g' >tempufw +mv tempufw /etc/default/ufw +# "no" is bug compatibility with software long obsolete +ufw enable && ufw status verbose +# Status: active +# Logging: on (low) +# Default: deny (incoming), allow (outgoing), disabled (routed) +# New profiles: skip +# To Action From +# -- ------ ---- +# 22/tcp (SSH) ALLOW IN Anywhere +# 80,443/tcp (Nginx Full) ALLOW IN Anywhere +# 22/tcp LIMIT IN Anywhere +# 22/tcp (SSH (v6)) ALLOW IN Anywhere (v6) +# 80,443/tcp (Nginx Full (v6)) ALLOW IN Anywhere (v6) +# 22/tcp (v6) LIMIT IN Anywhere (v6) +mysql_secure_installation +#empty root password +#Don't set a root password +#remove anonymous users +#disallow remote login +#drop test database +mariadb +``` + +You should now receive a message that you are in the mariadb console + +```sql +CREATE DATABASE example_database; +GRANT ALL ON example_database.* TO 'example_user'@'localhost' +IDENTIFIED BY 'mypassword' WITH GRANT OPTION; +FLUSH PRIVILEGES; +``` + +```bash +exit +mariadb -u example_user --password=mypassword example_database +``` + +```sql +CREATE TABLE todo_list ( item_id INT +AUTO_INCREMENT, content VARCHAR(255), +PRIMARY KEY(item_id) ); +INSERT INTO todo_list (content) VALUES +("My first important item"); +INSERT INTO todo_list (content) VALUES +("My second important item"); +SELECT * FROM todo_list; +exit +``` + +OK, MariaDB is working. We will use this trivial database and easily +guessed `example_user` with the easily guessed password +`mypassword` for more testing later. Delete him and his database +when your site has your actual content on it. + +### domain names and PHP under nginx + +Check again that the default nginx web page comes up when you browse to the server. + +Create the directories `/var/www/blog.reaction.la` and `/var/www/reaction.la` and put some html files in them, substituting your actual domains for the example domains. + +```bash +mkdir /var/www/reaction.la && nano /var/www/reaction.la/index.html +mkdir /var/www/blog.reaction.la && nano /var/www/blog.reaction.la/index.html +``` + +```default + + + + + +

reaction.la index file

+ +``` + +Delete the default in `/etc/nginx/sites-enabled`, and create a file, which I +arbitrarily name `config` that specifies how your domain names are to be +handled, and how php is to be executed for each domain names. + +This config file assumes your domain is called `reaction.la` and your +service is called `php7.3-fpm.service`. Create the following config file, +substituting your actual domains for the example domains, and your actual +php fpm service for the fpm service. + +```bash +nginx -t +# find the name of your php fpm service +systemctl status php* | grep fpm.service +# substitute the actual php fpm service for +# php7.3-fpm.sock in the configuration file. +systemctl stop nginx +rm -v /etc/nginx/sites-enabled/* +nano /etc/nginx/sites-enabled/config +``` + +```default +server { + return 301 $scheme://reaction.la$request_uri; + } +server { + listen 80; + listen [::]:80; + index index.php index.html; + server_name blog.reaction.la; + root /var/www/blog.reaction.la; + index index.php index.html; + location / { + try_files $uri $uri/ =404; + } + location ~ \.php$ { + include snippets/fastcgi-php.conf; + fastcgi_pass unix:/run/php/php7.3-fpm.sock; + } + location = /favicon.ico {access_log off; } + location = /robots.txt {access_log off; allow all; } + location ~* \.(css|gif|ico|jpeg|jpg|js|png)$ { + expires max; + } + } +server { + listen 80; + listen [::]:80; + index index.php index.html; + server_name reaction.la; + root /var/www/reaction.la; + location / { + try_files $uri $uri/ =404; + } + location ~ \.php$ { + include snippets/fastcgi-php.conf; + fastcgi_pass unix:/run/php/php7.3-fpm.sock; + } + location = /favicon.ico {access_log off; } + location = /robots.txt {access_log off; allow all; } + location ~* \.(css|gif|ico|jpeg|jpg|js|png)$ { + expires max; + } + } +server { + server_name *.blog.reaction.la; + return 301 $scheme://blog.reaction.la$request_uri; + } +``` + +The first server is the default if no domain is recognized, and redirects the +request to an actual server, the next two servers are the actual domains +served, and the last server redirects to the second domain name if the +domain name looks a bit like the second domain name. Notice that this +eliminates those pesky `www`s. + +The root tells it where to find the actual files. + +The first location tells nginx that if a file name is not found, give a 404 rather than doing the disastrously clever stuff that it is apt to do, and the second location tells it that if a file name ends in `.php`, pass it to `php7.3-fpm.sock` (you did substitute your actual php fpm service for `php7.3-fpm.sock`, right?) + +Now check that your configuration is OK with `nginx -t`, and restart nginx to read your configuration. + +```bash +nginx -t +systemctl restart nginx +``` + +Browse to those domains, and check that the web pages come up, and that +www gets redirected. + +Now we will create some php files in those directories to check that php works. + +```bash +echo "" |tee /var/www/reaction.la/info.php +``` + +Then take a look at `info.php` in a browser. + +If that works, then create the file `/var/www/reaction.la/index.php` containing: + +```php +TODO
    "; + foreach($db->query("SELECT content FROM $table") as $row) { + echo "
  1. " . $row['content'] . "
    2. "; + } + echo "
"; + } + catch (PDOException $e) { + print "Error!: " . $e->getMessage() . "
"; + die(); + } + ?> +``` + +[http://reaction.la]:http://reaction.la + +Browse to [http://reaction.la] If that works, delete the `info.php` file as it reveals private information. You now have domain names being served +by lemp. Your database now is accessible over the internet through PHP +on those domain names. + +### SSL and DNSSEC + +SSL encrypts communication between your server and the client, so that +those in between cannot read it or change it. + +It also somewhat protects against malicious people fooling the client into +connecting to the wrong server. Unfortunately there are a thousand +certificate authorities, and some of them are malicious or hostile, and if +you have powerful enemies (and who cares about powerless enemies) they +will cheerfully issue a certificate your enemy for your domain name. +DNSSEC somewhat protects against this, since there is only one root of trust + +If you are reading this document, you are self hosting, in which case your +registrar is probably providing your nameservers, in which case it is easy +for them to set up DNSSEC for you. You just have to click the correct +button on their website. One click, and it is done. And now you only have +to worry about two parties that might potentially defect on you, the +DNSSEC and your registrar, instead of a thousand certificate authorities. + +If, however, someone other than your registrar is managing your +nameserver, if your DNS records live on a machine controlled by one +entity, and your nameserver is controlled by a different entity, attempting +to set up DNNSEC gets complicated, and if that someone is not you, +considerably more complicated. In this case setting up DNSSEC is like +setting up SSH, but when you are setting up SSH, you control both +machines. When you attempt to setup DNSSEC you don't. Don't even try. +If you do try, make very sure the nameserver is doing the right thing +before you submit the DNSSEC public key you generated to the registrar +and attempt to get the registrar to do the right thing. + +OK, DNSSEC was easy. (Or you just gave up because far too hard.) Now +on to SSL + +Create the necessary DNS records, an A record pointing to your IP4 +address, an AAAA record pointing to your IP6 address, a CAA record +indicating who is the right issuer for your SSL certificate, so that not every +certificate authority in the world is allowed to issue fake certificates for your enemies, and CNAME records for the www and git aliases. + +The CAA record looks like: + +```default +@ CAA 0 issue "letsencrypt.org" +``` + +[whatsmydns]:https://www.whatsmydns.net/#CAA + +Go to [whatsmydns] and check if it looks right. + +[very easy utility]:https://certbot.eff.org/docs/using.html#nginx +"Certbot Instructions" + +Certbot provides a [very easy utility] for installing ssl certificates, and if +your domain name is already publicly pointing to your new host, and your +new host is working as desired, without, however, ssl/https that is +great. + +```bash +# first make sure that your http only website is working as +# expected on your domain name and each subdomain. +# certbots many mysterious, confusing, and frequently +# changing behaviors expect a working environment. +apt-get -qy install certbot python-certbot-nginx +certbot register --register-unsafely-without-email --agree-tos +certbot --nginx +# This also, by default, sets up automatic renewal, +# and reconfigures everything to redirect to https +``` + +Not so great if you are setting up a new server, and want the old +server to keep on servicing people while you set up the new server, so here +is the hard way, where you prove that you, personally, control the DNS +records, but do not prove that the server that certbot is modifying is right +now publicly connected as that domain name. + +(Obviously on your network the domain name should map to the new +server. Meanwhile, for the rest of the world, the domain name continues to +map to the old server, until the new server works.) + +```bash +apt-get -qy install certbot python-certbot-nginx +certbot register --register-unsafely-without-email --agree-tos +certbot run -a manual --preferred-challenges dns -i nginx -d reaction.la -d blog.reaction.la +nginx -t +``` + +This does not set up automatic renewal. To get automatic renewal going, +you will need to renew with the `webroot` challenge rather than the `manual` +once DNS points to this server. + +But if you are doing this, not on your test server, but on your live server, the easy way, which will also setup automatic renewal and configure your webserver to be https only, is: + +```bash +`certbot --nginx -d mail.reaction.la,blog.reaction.la,reaction.la` +``` + +If instead you already have a certificate, because you copied over your `/etc/letsencrypt` directory + +```bash +apt-get -qy install certbot python-certbot-nginx +certbot install -i nginx +nginx -t +``` + +To renew certbot certificates, which has to be done every couple of +months:\ +If you previously did the manual challenge, then `certbot renew` will likely +fail (because no default non manual challenge exists). You need to set the +renewal parameters for renewal to take place. + +```bash +certbot renew --renew-by-default --http01 +``` + +Because certbot automatically renews using the previous defaults, you +have to have previously used a process to obtain certificate suitable for +automation, which mean you have to have given it the information\ +(`--webroot --webroot-path /var/www/reaction.la`)\ + about how to do an automatic renewal by actually obtaining a certificate that way. + +To backup and restore letsencrypt, to move your certificates from one +server to another, `rsync -HAvaX reaction.la:/etc/letsencrypt /etc`, as root +on the computer which will receive the backup. The letsencrypt directory +gets mangled by `tar`, `scp` and `sftp`. + +Again, browse to your server. You should get redirected to https, and https should work. + +Backup the directory tree `/etc/letsencrypt/`, or else you can get into +situations where renewal is a problem. Only Linux to Linux backups work, +and they do not exactly work – things go wrong. Certbot needs to fix its +backup and restore process, which is broken. Apparently you should +backup certain directories but not others. But backing up and restoring the +whole tree works well enough for `certbot install -i nginx` + +The certbot modified file for your ssl enabled domain should now look like + +```default +server { + return 301 $scheme://reaction.la$request_uri; + } +server { + index index.php index.html; + server_name blog.reaction.la; + root /var/www/blog.reaction.la; + index index.php; + location / { + try_files $uri $uri/ =404; + } + location ~ \.php$ { + include snippets/fastcgi-php.conf; + fastcgi_pass unix:/run/php/php7.3-fpm.sock; + } + location = /favicon.ico {access_log off; } + location = /robots.txt {access_log off; allow all; } + location ~* \.(css|gif|ico|jpeg|jpg|js|png)$ { + expires max; + } + listen [::]:443 ssl; # managed by Certbot + listen 443 ssl; # managed by Certbot + ssl_certificate /etc/letsencrypt/live/reaction.la/fullchain.pem; # managed by Certbot + ssl_certificate_key /etc/letsencrypt/live/reaction.la/privkey.pem; # managed by Certbot + include /etc/letsencrypt/options-ssl-nginx.conf; # managed by Certbot + ssl_dhparam /etc/letsencrypt/ssl-dhparams.pem; # managed by Certbot + } +server { + index index.html; + server_name reaction.la; + root /var/www/reaction.la; + location / { + try_files $uri $uri/ =404; + } + location ~ \.php$ { + include snippets/fastcgi-php.conf; + fastcgi_pass unix:/run/php/php7.3-fpm.sock; + } + location = /favicon.ico {access_log off; } + location = /robots.txt {access_log off; allow all; } + location ~* \.(css|gif|ico|jpeg|jpg|js|png)$ { + expires max; + } + listen [::]:443 ssl ipv6only=on; # managed by Certbot + listen 443 ssl; # managed by Certbot + ssl_certificate /etc/letsencrypt/live/reaction.la/fullchain.pem; # managed by Certbot + ssl_certificate_key /etc/letsencrypt/live/reaction.la/privkey.pem; # managed by Certbot + include /etc/letsencrypt/options-ssl-nginx.conf; # managed by Certbot + ssl_dhparam /etc/letsencrypt/ssl-dhparams.pem; # managed by Certbot + } +server { + server_name *.blog.reaction.la; + return 301 $scheme://blog.reaction.la$request_uri; + } +server { + server_name *.reaction.la; + return 301 $scheme://reaction.la$request_uri; + } +server { + if ($host = reaction.la) { + return 301 https://$host$request_uri; + } # managed by Certbot + listen 80; + listen [::]:80; + server_name reaction.la; + return 404; # managed by Certbot + } +server { + if ($host = blog.reaction.la) { + return 301 https://$host$request_uri; + } # managed by Certbot + listen 80; + listen [::]:80; + server_name blog.reaction.la; + return 404; # managed by Certbot + } +``` + +You may need to clean a few things up after certbot is done. + +The important lines that certbot created in the file being `ssl_certificate`, +the additional servers listening on port 80 which exist to redirect http to https +servers listening on port 403, and that all redirects should be `https` instead +of `$scheme` (fix them if they are not). + +nginx starts as root, but runs as unprivileged user `www-data`, who needs to +have read permissions to every relevant directory. If you want to give php +write permissions to a directory, or restrict `www-data` and `pgp`’s read +permissions to some directories and not others, you could do clever stuff with +groups and users, giving creating users that php scripts act as, and +making www-data a member of their group, but that is complicated and +easy to get wrong. + +A quick fix is to `chown -R www-data:www-data` the directories that your +web server needs to write to, and only those directories, though I can hear security gurus gritting their teeth when I say this. + +For all the directories that www-data merely needs to read: + +```bash +find /var/www -type d -exec chmod 755 {} \; +find /var/www -type f -exec chmod 644 {} \; +``` + +Now you should delete the example user and the example database: + +```sql +mariadb +REVOKE ALL PRIVILEGES, GRANT OPTION FROM +'example_user'@'localhost'; +DROP USER 'example_user'@'localhost'; +DROP DATABASE example_database; +exit +``` + +### Wordpress on Lemp + +```bash +apt-get -qy install php-curl php-gd php-intl php-mbstring php-soap php-xml php-xmlrpc zip php-zip +systemctl status php* | grep fpm.service +# restart the service indicated above +systemctl stop nginx +systemctl stop php7.3-fpm.service +mariadb +``` + +```sql +CREATE DATABASE wordpress DEFAULT CHARACTER SET +utf8mb4 COLLATE utf8mb4_unicode_ci; +GRANT ALL ON wordpress.* TO 'wordpress_user'@'localhost' +IDENTIFIED BY 'FGikkdfj3878'; +FLUSH PRIVILEGES; +exit +``` + +The lemp server block that will handle the wordpress domain needs to pass +urls to index.php instead of returning a 404. (Handle your 404s and +redirects issues with the Redirections Wordpress plugin, which is a whole +lot easier, safer, and more convenient than editing redirects into your +`/etc/nginx/sites-enabled/*` files.) + +```default +server { + . . . + location / { + #try_files $uri $uri/ =404; + try_files $uri $uri/ /index.php$is_args$args; + } + . . . + } +``` + +```bash +nginx -t +mkdir temp +cd temp +curl -LO https://wordpress.org/latest.tar.gz +tar -xzvf latest.tar.gz +cp -v wordpress/wp-config-sample.php wordpress/wp-config.php +cp -av wordpress/. /var/www/blog.reaction.la +chown -R www-data:www-data /var/www/blog.reaction.la && find /var/www -type d -exec chmod 755 {} \; && find /var/www -type f -exec chmod 644 {} \; +# so that wordpress can write to the directory +curl -s https://api.wordpress.org/secret-key/1.1/salt/ +nano /var/www/blog.reaction.la/wp-config.php +``` + +Replace the defines that are there\ +`define('LOGGED_IN_KEY', 'put your unique phrase here');`\ +with the defines you just downloaded from wordpress. + +and replace DB_NAME, DB_USER, DB_PASSWORD, and FS_METHOD + +```default +… +// ** Mariadb settings // +/** The name of the database for WordPress */ +define('DB_NAME', 'wordpress'); +/** MySQL database username */ +define('DB_USER', 'wordpress_user'); +/** MySQL database password */ +define('DB_PASSWORD', 'FGikkdfj3878'); +/** MySQL hostname */ +define( 'DB_HOST', 'localhost' ); +/** Database Charset to use in creating database tables. */ +define( 'DB_CHARSET', 'utf8mb4' ); +/** The Database Collate type. */ +define( 'DB_COLLATE', 'utf8mb4_unicode_ci' ); +… +``` + +```bash +systemctl start php7.3-fpm.service +systemctl start nginx +``` + +It should now be possible to navigate to your wordpress domain in your web browser and finish the setup there: + +### Exporting databases + +Interacting directly with your database of the MariaDB command line is apt to lead to disaster. + +Installing PhpMyAdmin has a little gotcha on Debian 9, which is covered +in [this tutorial](https://hostadvice.com/how-to/how-to-install-and-secure-phpmyadmin-on-debian-9/), but I just do not use +PhpMyAdmin even though it is easer and safer. + +#### To export by command line + +```bash +systemctl stop nginx +systemctl stop php7.3-fpm.service +mdir temp && cd temp +fn=blogdb +db=wordpress +dbuser=wordpress_user +dbpass=FGikkdfj3878 +mysqldump -u $dbuser --password=$dbpass $db > $fn.sql +head -n 30 $fn.sql +zip $fn.sql.zip $fn.sql +systemctl start php7.3-fpm.service +systemctl start nginx +``` + +### Moving a wordpress blog to new lemp server + +[Wordpress on Lemp]:#wordpress-on-lemp "installing wordpress on lemp" + +Prerequisite: you have configured [Wordpress on Lemp] + +Copy everything from the web server source directory of the previous +wordpress installation to the web server of the new wordpress installation. + +```bash +chown -R www-data:www-data /var/www/blog.reaction.la +``` + +Replace the defines for `DB_NAME`, `DB_USER`, and `DB_PASSWORD` in `wp_config.php`, as described in [Wordpress on Lemp] + +#### To import datbase by command line + +```bash +systemctl stop nginx +systemctl stop php7.3-fpm.service +# we don’t want anyone browsing the blog while we are setting it up +# nor the wordpress update service running. +mariadb +``` + +```sql +DROP DATABASE IF EXISTS wordpress; +CREATE DATABASE wordpress DEFAULT CHARACTER SET +utf8mb4 COLLATE utf8mb4_unicode_ci; +GRANT ALL ON wordpress.* TO 'wordpress_user'@'localhost' +IDENTIFIED BY 'FGikkdfj3878'; +exit +``` + +At this point, the database is still empty, so if you start nginx and browse to +the blog, you will get the wordpress five minute install, as in [Wordpress +on Lemp]. Don’t do that, or if you start nginx and do that to make sure +everything is working, then start over by deleting and recreating the +database as above. + +Now we will populate the database. + +```bash +fn=wordpress +db=wordpress +dbuser=wordpress_user +dbpass=FGikkdfj3878 +unzip $fn.sql.zip +mv *.sql $fn.sql +mariadb -u $dbuser --password=$dbpass $db < $fn.sql +mariadb -u $dbuser --password=$dbpass $db +``` + +```sql +SHOW TABLES; +SELECT COUNT(*) FROM wp_posts; +SELECT * FROM wp_posts l LIMIT 20; +exit +``` + +Adjust `$table_prefix = 'wp_';` in `wp_config.php` if necessary. + +```bash +systemctl start php7.3-fpm.service +systemctl start nginx +``` + +Inside the sql file may be references to the old directories, (search for +`'recently_edited'`), and to the old user who had the privilege to create views +(search for `DEFINER=`) Replace them with the new directories and new +database user, in this example `wordpress_user`. + +Edit the `siteurl`,`admin_email` and `new_admin_email` fields of the blog +database to the domain and new admin email. + +```bash +mariadb -u $dbuser --password=$dbpass $db < $db.sql +mariadb -u $dbuser --password=$dbpass $db +``` + +```sql +SHOW TABLES; +SELECT COUNT(*) FROM wp_comments; +SELECT * FROM wp_comments l LIMIT 10; +``` + +Adjust `$table_prefix = 'wp_';` in `wp_config.php` if necessary. + +```bash +systemctl start php7.3-fpm.service +systemctl start nginx +``` + +Your blog should now work. + +## Logging and awstats. + +### Logging + +First create, in the standard and expected location, a place for nginx to log stuff. + +```bash +mkdir /var/log/nginx +chown -R www-data:www-data /var/log/nginx +``` + +Then edit the virtual servers to be logged, which are in the directory `/etc/nginx/sites-enabled` and in this example in the file `/etc/nginx/sites-enabled/config` + +```text +server { + server_name reaction.la; + root /var/www/reaction.la; + … + access_log /var/log/nginx/reaction.la.access.log; + error_log /var/log/nginx/reaction.la.error.log; + … + } +``` + +The default log file format logs the ips, which in a server located in the cloud might be a problem. People who do not have your best interests at heart might get them. + +So you might want a custom format that does not log the remote address. On the other hand, Awstats is not going to be happy with that format. A compromise is to create a cron job that cuts the logs daily, a cron job that runs Awstats, and a cron job that then deletes the cut log when Awstats is done with it. + +There is no point to leaving a gigantic pile of data, that could hang you and your friends, sitting around wasting space. + +## Postfix and Dovecot + +[Postfix and Dovecot are a pile of matchsticks and glue] from which you are +expected to assemble a boat. + +Probably I should be using one of those email setup packages that set up +everything for you. [Mailinabox] seems to be primarily tested and +developed on ubuntu, and is explicitly not supported on debian. + +[Mailcow] however, is Debian. But [Mailcow] wants 6GiB of ram, plus one +GiB swap, plus twenty GiB disk. Ouch. [Mailinabox] can get by with one +GiB of ram, plus one GiB of swap. Says 512MiB is OK, though two GiB +of ram is strongly recommended. + +[Mailinabox] wants the domain name `box.yourdomain.com`, and, after it is +set up, wants the nameservers `ns1.box.yourdomain.com` and +`ns2.box.yourdomain.com`. They, fortunately, have a namecheap tutorial. + +[Mailcow]:https://mailcow.github.io/mailcow-dockerized-docs/prerequisite-system/ +{target="_blank"} + +[Mailinabox]:https://mailinabox.email +"Mail-in-a-Box" +{target="_blank"} + +[vps–projects]:https://vpsprojects.com/index.php/mailinabox-installation-and-configuration/ +"Mailinabox installation and configuration" +{target="_blank"} + +Linuxbabe, on the other hand, recommends [iRedMail] But [iRedMail] needs 4GiB and gets far less love from users than [Mailinabox]. + +[iRedMail]:https://www.linuxbabe.com/mail-server/debian-10-buster-iredmail-email-server +"How to Easily Set Up a Mail Server on Debian 10 Buster with iRedMail" +{target="_blank"} + +[Postfix and Dovecot are a pile of matchsticks and glue]: +https://www.digitalocean.com/community/tutorials/how-to-set-up-a-postfix-e-mail-server-with-dovecot +"How To Set Up a Postfix E-Mail Server with Dovecot | DigitalOcean" + +Postfix is a matter of installing it, but it expects an MX record in your +nameserver, and quite a few other records are almost essential. + +Postfix automagically does the right thing for users that +have accounts, using their names and passwords to set up +mailboxes. It gets complicated only when people start to pile supposedly +more advanced mail systems, databases, and webmail on top of it. + +With Postfix alone, you can receive emails at your server and have them +automatically forwarded to a more useful email address. You can also +receive, send, and reply to emails when logged in to you server using the +command line utility, which I only use to make sure Postfix is working +before I configure Dovecot on top of it. + +To receive your emails in an actually useful form, +you are going to have to forward them or set up a Dovecot service. +Dovecot provides Pop3 and IMAP. Postfix does not provide any of that. + +Postfix is not a Pop3 or IMAP server. It sends, receives, and forwards +emails. You cannot set up an email client such as Thunderbird to remotely +access your emails – they are only available to people logged in on the +server who are never going to look at them anyway, because there is no +useful UI to read them and reply to them. + +So you have to configure three or more independent things, each of which +has an endlessly complicated configuration that is intimately, arcanely, and +obscurely connected to the configuration of each of the other things. + +### Setting DNS entries for email + +An MX record for `reaction.la` will read simply `mail` (no full stop, that +is for the case that you are trying to have a totally unrelated host handle +your mail) Check that it is working by using an MX lookup service such +as [MX tools] and [Dig] + +You will need of course a corresponding CNAME record `mail`. + +You are going to need a PTR record, or else you mail is likely to be rejected as spam. + +The problem is that reverse DNS is not going the query the man who +keeps your DNS records, but the man who gave you your IP address. + +The UI for creating a PTR record is going to be in a different place, very +likely maintained by a different man, using different software, than the UI +for creating you MX record. It will have a link that is probably called +something like "reverse" rather than PTR or DNS. + +You look up your IP in [MX tools] or [Dig] + +You must create the reverse DNS zone on the authoritative DNS +nameserver for the *IP address* of your server. Which is controlled by +whoever gave you your IP address, not whoever is maintaining your DNS +records. You can find if your reverse IP is working by doing a reverse +lookup in [Dig]. + +[MX tools]: +https://mxtoolbox.com/SuperTool.aspx + +[Dig]:https://www.digwebinterface.com + + [man who provides you your IP address]: + https://support.dnsmadeeasy.com/support/solutions/articles/47001001864-how-to-setup-reverse-dns + "Contact your IP provider to request your IP’s reverse DNS zone." + + Thus the instructions for setting up a PTR record that you find on the internet are unlikely to be applicable to you, and will only confuse you. + +An MX record is useless, because apt to be massively spammed, without +a `spf` TXT record stating that mail *only* comes from your IP4 and IP6. +Otherwise your enemies will issue spam in your name, with the result that +your MX record will be blacklisted. + +An spf record is of type TXT and looks like +`v=spf1 ip4:69.64.153.131 ip6:DEAD:BEEF:DEAD:55a:0:0:0:1 -all` +indicating that all mail under this domain +name will be sent from one and only one network address. + +A DKIM record publishes a digital signature for sent mail, to prevent mail +from being modified or fake mails being injected as it goes through the +multiple intermediate servers. + +By the time the ultimate recipient sees the email, no end of intermediate +computers have had their hands on it, but pgp signing is obviously +superior, since that is controlled by the actual sender, not one more +intermediary. DKIM is the not quite good enough being the enemy of the +good enough. + +Worse, DKIM means that any email sent from your server is signed by +your server, so if you send a private message, and someone defects on you +by making it public, hard to claim that he is making it up. Sometimes you +want your emails signed with a signature verifiable by third parties, and +sometimes this is potentially dangerous. Gpg allows you to sign some +things and not other things. DKIM means that everything gets signed, +without you being aware of it. + +DKIM renders all messages non repudiable, and some messages vitally +need to be repudiable. + +Gpg is better than DKIM, but has the enormous disadvantage that it +cannot authenticate except by signing. If you send a message to a single +recipient or a quite small number of recipients, you usually want him to +know for sure it is from you, and has not been altered in transit, but not be +able to prove to the whole world that it is from you. + +A DMARK record can tell the recipient that mail from +`rhocoin.org` will always and only come senders like +`user@rhocoin.org`. This can be an inconvenient restriction on +one's ability to use a more relevant identity. + +Further, intermediate servers keep manging messages sent through them, +breaking the DKIM signatures, resulting in no end of spurious error messages + +You want to stop other people's email servers from misbehaving on the +sender addresses. You don't want to stop your server from misbehaving. + +Trouble with SPF and DKIM is that, without DMARK, they have no +impact on the sender address, thus don't stop spearphishing attacks using +your identity. They do stop spam attacks from getting your server +blacklisted by using your server identity to send junk mail. + +SPF is sufficient to stop your server from getting blacklisted, and largely +irrelevant to preventing spearphishing. But SPF with DMARK at least +does something about spearphishing, while DKIM with DMARK does a far more +thorough job on spearphishing, but produces a lot of false warnings due to +intermediate servers mangling the email in ways that invalidate the signature. + +The only useful thing that DMARK can do is ensure *address alignment* +with SPF and DKIM, so that only people who can perform an email login +to your server can send email as someone who has such a login. + +But DKIM is complicated to install, and a lot more complicated to manage +because you will be endlessly struggling to resolve the problem of +signatures being falsely invalidated. It may be far more effective against +spearphishing if the end user pays attention, but if people keep seeing +false warnings, they are being trained to ignore valid warnings. + +A solution to all these problems is to [use the value `ed25519-sha256` for `a=` in the DKIM header](https://www.mailhardener.com/kb/how-to-use-dkim-with-ed25519) +(which ensures that obsolete intermediaries will +ignore your DKIM, thus third parties will see fewer false warnings) and to +have a [cron job that regularly rotates your DKIM keys](https://rya.nc/dkim-privates.html "DKIM: Show Your Privates"), *and publishes the +old secret key on the DNS 36 hours after it has been rotated out* under the +`n=OldSecret_...` field of the DKIM record, thus rendering your emails +deniable. (RSA keys are inconveniently large for this protocol, since they +do not fit in a DNS record) + +But that is quite bit of work, so I just have not gotten around to it. No one +seems to have gotten around to it. Needs to be part of [Mailinabox] so that it +becomes a standard. + +Until you can install DKIM with a cron job that renders email repudiable, do +not install DKIM. (And disable it if [Mailinabox] enables it.]) + +### Install Postfix + +Here we will install postfix on debian so that it can be used to send emails +by local users and applications only - that is, those installed on the same +server as Postfix, such as your blog, mailutils, and Gitea, and can receive +emails to those local users. + +```bash +echo MAIL=Maildir>/etc/profile.d/maildir.sh && chmod +x /etc/profile.d/maildir.sh +``` + +Start up a new shell, so that $MAIL is correctly set + +```bash +echo $MAIL +apt -qy update +apt -qy install mailutils +apt -qy install postfix +``` + +Near the end of the installation process, you will be presented with a window that looks like the one in the image below: +![Initial Config Screen](./images/postfix_cfg1.webp){width=100%} +If `` is not highlighted, hit tab. + +Press `ENTER` to continue. +The default option is **Internet Site**, which is preselected on the following screen: +![Config Selection Screen](./images/postfix_cfg2.webp){width=100%} +Press `ENTER` to continue. + +After that, you’ll get another window to set the domain name of the site that is sending the email: +![System Mail Name Selection](./images/postfix_cfg3.webp){width=100%} +The `System mail name` should be the same as the name you assigned to the server when you were creating it. When you’ve finished, press `TAB`, then `ENTER`. + +You now have Postfix installed and are ready to modify its configuration settings. + +### configuring postfix + +```bash +postconf -e home_mailbox=Maildir/ +``` + +Which is incompatible with lots of modern mail software, but a lot more +compatible with all manner of programs trying to use Postfix, including +Dovecot + +Your forwarding file is, by default, broken. It forwards all administrative +system generated email to the nonroot local user `deb10` who probably does +not exist on your system. + +Set up forwarding, so you’ll get emails sent to `root` on the system at your +personal, external email address or to a suitable nonroot local user, or +create the local user `deb10`. + +To configure Postfix so that system-generated emails will be sent to your +email address or to some other non root local user, you need to edit the +`/etc/aliases` file. + +```bash +nano /etc/aliases +``` + +```default +mailer-daemon: postmaster +postmaster: root +nobody: root +hostmaster: root +usenet: root +news: root +webmaster: root +www: root +ftp: root +abuse: root +noc: root +security: root +root: «your_email_address» +``` + +After changing `/etc/aliases` you must issue the command `newaliases` to inform the mail system. (Rebooting does not do it.) + +`/etc/aliases` remaps mail to users on your internal mail server, but likely your mail server is also the MX host for another domain. For this, you are going to need a rather more powerful tool, which I address later. + +The `postmaster: root` setting ensures that system-generated emails are sent +to the `root` user. You want to edit these settings so these emails are rerouted +to your email address. To accomplish that, replace «your_email_address» +with your actual email address, or the name of a non root user.. Most systems do not allow email clients to +login as root, so you cannot easily access emails that wind up as `root@mail.rhocoin.org` + +Probably you should create a user `postmaster` + +If you’re hosting multiple domains on a single server, the other domains +must passed to Postfix using the `mydestination` directive if other people ware going tosend email addressed to users on those domains. But +chances are you also have other domains on another server, which declare in +their DNS this server as their MX record. `mydestination` is not the place +for the domain names of those servers, and putting them in `mydestination` +is apt to result in mysterious failures. + +Those other domains, not hosted on this physical machine, but whose MX +record points to this machine are [virtual_alias_domains](#virtual-domains-and-virtual-users) and postfix has to +handle messages addressed to such users differently + +Set the mailbox limit to an appropriate fraction of your total available disk space, and the attachment limit to an appropriate fraction of your mailbox size limit. + +Check that `myhostname` is consistent with reverse ip search. (It should already be if you setup reverse IP in advance) + +Set `mydestination` to all dns names that map to your server (it probably already does) + +```bash +postconf -e mailbox_size_limit=268435456 +postconf -e message_size_limit=67108864 +postconf +postconf myhostname +postconf mydestination +postconf smtpd_banner +# you don't want your enemies to know what OS version you are running, +# as this may make hacking easier +postconf -e smtpd_banner='$myhostname ESMTP $mail_name' +postconf -e smtpd_helo_required=yes +postconf smtpd_helo_restrictions +postconf -e smtpd_helo_restrictions='permit_mynetworks, permit_sasl_authenticated, reject_invalid_helo_hostname, reject_non_fqdn_helo_hostname, reject_unknown_helo_hostname' +postconf smtpd_sender_restrictions +postconf -e smtpd_sender_restrictions='permit_mynetworks, permit_sasl_authenticated, reject_unknown_sender_domain' +postconf smtpd_client_restrictions +postconf -e smtpd_client_restrictions=='permit_mynetworks, permit_sasl_authenticated, reject_unknown_reverse_client_hostname' +newaliases && systemctl restart postfix +# check that you have in fact set stuff as you intended +postconf -n +postfix check +ufw enable postfix +ufw status verbose +# port 25 should be open +ss -lnpt | grep master +# postfix should be listening on port 25 +su -l «nonroot-user» +mail «you@some-email» +``` + +At this point mail should just work. Check your email at you@some-email and reply to it. Then check if you have received the reply. + +```bash +cat /var/log/mail.log +su -l «nonroot-user» +mail +``` + +If mail is not working, check the logs + +Now email should be working with the command line utility. + +Send email to local users and to your external email address using the command line utility `mail`. Check that you can read it. + +Send in mail from and outside. Reply to that outside system, and check that your reply gets through. + +To read mail from the command line `mail`, but reading and writing mail +using the command line utility is too painful, except for test purposes. + +If you memorize lots of [mail](https://www.commandlinux.com/man-page/man1/mail.1.html) commands, it is usable, sort of. + +Test your mail server using [MX tools]\ +![working](./images/working_mailersever.webp){width=100%} + +Now you have a basic Postfix email server up and running. You can send plain text emails and read incoming emails using the command line. + +After sending and receiving a few emails, check for issues: + +```bash +cat /var/log/mail.log | grep -E --color "warning|error|fatal|panic" +postqueue -p +``` + +If you see a pile of warnings `warning symlink leaves directory: /etc/postfix/./makedefs.out` that is just noise. Turn it off by replacing the symbolic link with a hard link + +```bash +postfix check +rm /etc/postfix/makedefs.out; ln /usr/share/postfix/makedefs.out /etc/postfix/makedefs.out +postfix check +``` + +Or just ignore it. + +Make sure that [MX tools] thinks your mail server is working. + +### TLS + +Now you can send mail while logged in to your server, using the +command line program `mail`, programs running on your server can send +you or anyone else mail, and anyone can send mail to users that exist on +your server, which can be forwarded to actually useful email addresses. + +But your email is in the clear, and can be read, or altered, by any +unpleasant person between you and the destination that intends you harm +or harm to those that you are messaging. Of which there are likely quite a +lot. Such alterations are likely to result in your email server ending up +blacklisted as a result of other people's anti malware precautions. + +```bash +cat /var/log/mail.log | grep -E --color "warning|error|fatal|panic|TLS" +postqueue -p +``` + +You probably will not see any TLS activity. You want to configure Postfix +to always attempt SSL, but not require it. + +```bash +# TLS parameters +# +# SMTP from other servers to yours +# Make sure to substitute your certificates in for the smtp +# and smtpd certificates. +postconf -e smtpd_tls_cert_file=/etc/letsencrypt/live/rhocoin.org/fullchain.pem +postconf -e smtpd_tls_key_file=/etc/letsencrypt/live/rhocoin.org/privkey.pem +postconf -e smtpd_tls_security_level = may +postconf -e smtpd_tls_auth_only = yes +postconf -e smtpd_tls_mandatory_protocols=!SSLv2,!SSLv3,!TLSv1, !TLSv1.1 +postconf -e smtpd_tls_protocols=!SSLv2,!SSLv3, !TLSv1, !TLSv1.1 +postconf -e smtpd_tls_loglevel = 1 +postconf -e smtpd_use_tls=yes +postconf smtpd_tls_session_cache_database +# should be: +# smtpd_tls_session_cache_database = btree:${data_directory}/smtpd_scache +# +# SMTP from your server to others +postconf -e smtp_tls_cert_file=/etc/letsencrypt/live/rhocoin.org/fullchain.pem +postconf -e smtp_tls_key_file=/etc/letsencrypt/live/rhocoin.org/privkey.pem +postconf -e smtp_tls_security_level=may +postconf -e smtp_tls_note_starttls_offer=yes +postconf -e smtp_tls_mandatory_protocols='!SSLv2, !SSLv3, !TLSv1, !TLSv1.1' +postconf -e smtp_tls_protocols='!SSLv2, !SSLv3, !TLSv1, !TLSv1.1' +postconf -e smtp_tls_logleve=1 +postconf smtp_tls_session_cache_database +# should be: +# smtp_tls_session_cache_database = btree:${data_directory}/smtp_scache +# end TLS parameters +systemctl restart postfix +``` + +The excluded ciphers are weak or suspect, and SSLv2, SSLv3 have known holes. + +I have not bothered checking the list of permitted ciphers, many of which +would doubtless horrify me, because email security is incurably broken, +and trying to fix it is an endless and insoluble problem. SSH is more +readily fixable. + +Now send an email from one of your actually useful email accounts on your +client computer, and then, using the near unusable command line utility +`mail`, send a response. + +```bash +cat /var/log/mail.log |grep TLS +``` + +You should now see some TLS activity for those emails, and you should receive the emails. + +OK, now we are all done, unless you want people to send you emails at +cherry@rhocoin.org, and to be actually able to usefully read those emails +without setting up forwarding to another address. + +Well, not quite done, for now that you can receive emails, need to add your email to to your DMARC policy.\ +`v=DMARC1; p=quarantine; rua=mailto:postmaster@rhocoin.org` + +A dmarc record is a text record with the hostname `_dmarc`, and the policy is its text value. + +### SASL + +At this point any random person on the internet can send mail to +`root@rhocoin.org`, and you can automatically forward it to an actually +usable email address, but you cannot access his email account at +`root@rhocoin.org` from a laptop using thunderbird, and accessing it +through the command line using `mail` is not very useful. + +Because although Postfix by default accepts sasl authenticated mail +submissions to be relayed anywhere + +```default +smtpd_relay_restrictions = permit_mynetworks permit_sasl_authenticated defer_unauth_destination +``` + +It has yet as yet nothing configured to provide `sasl` authentication. + +We don't want random spammer on the internet to send email as +`random@rhocoin.org`, but we do want authenticated users to be able to do +as they please. + +So, need to install and configure Dovecot to provide sasl, to authenticate +cherry to Postfix. And need to tell Postfix to accept Dovecot authentication. + +However, before we do any of that, there is a very big problem, that all +email systems that allow clients to send email are a bleeding security hole, +because they do not use ssh. They instead use human memorable +passwords. And, since we are using Dovecot in its simplest configuration +they use your user login passwords. Which you did not care about until +now, because everyone logs in by ssh and you have set it up to be +impossible to login except by ssh. So you probably used short passwords +easy to guess and easy to type. + +But, once you install dovecot, every phisher, spammer and scammer on +the internet will be trying to login as one of your users using lists of +common passwords, and if he succeeds, is going to use your resources to send his spam or scam to everyone in the world. + +Big email providers have big complex systems, which require a great deal +of work to set up and manage, to monitor this. You do not. So everyone +who is going to be able to send email, which is to say all users on your +system, has to have an unguessable password, ten or more characters of +unguessable gibberish, or six or more human readable words. +`apt -y install libpam-pwquality` + +In addition to spammers, you will also, far less frequently, but far more +dangerously, have spearphishers armed with inside information. So every +user on your system has to have a strong password. + +Most of the proposals for integrating Postfix and Dovecot seem to be +endlessly complicated, endlessly different, pull in no end of additional +programs, and endlessly incompatible with Postfix working just by itself. +The basic problem is that Postfix was written assuming one actual user for +each email address, and lots of software throws away this useless vestigial +appendix, which breaks the now obsolete architecture of Postfix, and each +such re-invention has complicated workarounds for this architectural +break, each of which works in its own unique ad hoc way. + +But, since we got the old fashioned architecture of Postfix working, and our +primary interest is using it in programs that assume the old fashioned +architecture, we have to stick to the path that email users are Linux users, +to avoid getting into deep waters. We are going to map arbitrary email +addresses to arbitrary users, but for someone to send and receive email, he +has to have a user account on the server. + +PAM is just old fashioned Unix logon passwords. Pop3s is just TLS secured pop3. + +### Dovecot + +[Install and set up Dovecot, now that postfix is working](dovecot.html){target="_blank"} + +### Satellite postfix + +Suppose you have another domain name, which has no host. Then you can +make it a virtual alias domain on your actual host, or better, a virtual +mailbox domain. Virtual mailboxes rely on dovecot to actually deliver the +mail to the person who reads it on his desktop computer. Postfix abandons +the delivery problem. + +But suppose you have another domain, and another actual host, and you +don't want to go through all the grief involved in setting up email that +works. So you have its MX record point to your email host, install postfix +on it as a satellite system. + +Then nothing happens. Oops + +Since programs running on it are sending out emails in its domain name, +your primary host has to enable it to relay. The simplest way is to add +its IP address to your primary host's `mynetworks`. Now programs running +on the other domain, such as wordpress, can send emails by relaying them +through the primary host, without them being suppressed in transit as spam. + +There are rather a [lot of options and alternatives](http://www.postfix.org/VIRTUAL_README.html), and if you go down the +wrong path, you will get stuck. + +If you are the only one with a virtual mailbox, might as well administer it +by hand using only postfix, but if you have to go to virtual mailboxes, +there are probably several people involved, in which case administering it +just far too painful, both for you and those administered, better set up +[PostfixAdmin], and let those people administer themselves. + +[PostfixAdmin]:https://www.linuxbabe.com/mail-server/postfixadmin-ubuntu +"PostfixAdmin – Create Virtual Mailboxes" +{target="_blank"} + +To be allowed to relay through the primary host, the other systems have to +be listed in `mynetworks`, listed as virtual hosts, or something. There are +rather too many ways to do it, but `mynetworks` just makes the issue go away. + +### Virtual domains and virtual users + +Now that you have set this up, you don't want to set it up for several +domain name addresses corresponding to several hosts on the internet. You +want to just put an MX record pointing to this host in that host's DNS, so +that people can send and receive email using that host's domain name, regardless of whether a physical server with a network address exists for +that domain name. + +For each domain name that has an MX record pointing at this host add the + domain to the `virtual_alias_domains` in `/etc/postfix/main.cf` + +```bash +postconf virtual_alias_domains +postconf -e virtual_alias_domains=reaction.la,blog.reaction.la +postconf -e virtual_alias_maps=hash:/etc/postfix/virtual +``` + +Now create the file `/etc/postfix/virtual` which will list all the email addresses of users with email addresses ending in those domain names. + +```default +ann@reaction.la ann +bob@reaction.la bob +carol@blog.reaction.la carol +dan@blog.reaction.la dan +@reaction.la blackhole +@blog.reaction.la blackhole +# ann, bob, carol, dan, and blackhole have to be actual users +# on the actual host, or entries in its aliases file, even if there +# is no way for them to actually login except through an +# email client, and if mail to blackhole goes unread and is +# eventually automatically deleted. +# +# The addresses without username catch all emails that do not +# have an entry. +# You don't want an error message response for invalid email +# addresses, as this may reveal too much to your enemies. +``` + +Every time your `/etc/postfix/virtual` is changed, you have to recompile it +into a hash database that postfix can actually use, with the command: + +```bash +postmap /etc/postfix/virtual && postfix reload +``` + +#### set up an email client for a virtual domain + +We have setup postfix and dovecot so that clients can only use ssl/tls, and not starttls. + +On thunderbird, we go to account settings / account actions / add mail account + +We then enter the email address and password, and click on `configure manually` + +Select SSL/TLS and normal password + +For the server, thunderbird will incorrectly propose `.blog.reaction.la` + +Put in the correct value, `rhocoin.org`, then click on re-test. Thunderbird will then correctly set the port numbers itself, which are the standard port numbers. + +[tutorial](https://www.linux.com/training-tutorials/how-set-virtual-domains-and-virtual-users-postfix/) + +But the problem is, we might have an actual host running postfix, which wants to ask the host to which its MX record points, to send emails for it. + +Configuring postfix as a satellite system just works, at least for emails generated by services running on the same machine, but postfix does not provide for it logging in. Instead, postfix assumes it has been somehow authorized, typically in `mynetworks` to relay. + +Another way of setting it up, which I have not checked out, is [postfix_relaying_through_another_mailserver](https://www.howtoforge.com/postfix_relaying_through_another_mailserver){target="_blank"} + +## Your ssh client + +Your cloud server is going to keep timing you out and shutting you down, +so if using OpenSSH need to set up `~/.ssh/config` to read + +```default + ForwardX11 yes + Protocol 2 + TCPKeepAlive yes + ServerAliveInterval 10 +``` + +Putty has this stuff in the connection configuration, not in the +config file. Which makes it easier to get wrong, rather than harder. + +### A cloud server that does not shut you down + +Your cloud server is probably virtual private server, a vps running on KVM, XEN, +or OpenVZ. + +KVM is a real virtual private server, XEN is sort of almost a virtual +server, and OpenVZ is a jumped up guest account on someone else’s +server. + +KVM vps is more expensive, because when they say you get 2048 meg, +you actually do get 2048 meg. OpenVZ will allocate up to 2048 gig if it +has some to spare – which it probably does not. So if you are running +OpenVZ you can, and these guys regularly do, put far too many virtual +private servers on one physical machine. Someone can have a 32 Gigabyte +bare metal server with eight cores, and then allocate one hundred virtual +servers each supposedly with two gigabytes and one core on it, while if he +is running KVM, he can only allocate as much ram as he actually has. + +## Debian on the cloud + +Debian is significantly more lightweight than Ubuntu, harder to +configure and use, will crash and burn if you connect up to a software +repository configured for Ubuntu in order to get the latest and greatest +software. You generally cannot get the latest and greatest software, and +if you try to do so, likely your system will die on its ass, or plunge +you into expert mode, where no one sufficiently expert can be found. + +Furthermore, in the course of setting up Debian, highly likely to break +it irretrievably and have to restart from scratch. After each change, +reboot, and after each successful reboot, take a snapshot, so that you +do not have to reboot all the way from scratch. + +But, running stuff that is supposed to run, which not always the latest and +greatest, is more stable and reliable than Ubuntu. Provided it boots up +successfully after you have done configuring, will likely go on booting up +reliably and not break for strange and unforeseeable reasons. Will only +break because you try to install or reconfigure software and somehow +screw up. Which you will do with great regularity. + +On a small virtual server, Debian provides substantial advantages. + +Go Debian with ssh and no GUI for servers, and Debian with lightdm +Mate for your laptop, so that your local environment is similar to your +server environment. + +On any debian you need to run through the apt-get cycle till it stops +updating: + +```bash +apt-get -qy update && apt-get -qy upgrade +apt-get -qy install dialog dialog nano build-essential +apt-get -qy rsync linux-headers-generic +``` + +On windows, edit the command line of the startup icon for a virtual box +that you have iconized to add the command line option \--type headless, for +example + +```bat +"C:\Program Files\Oracle\VirtualBox\VirtualBoxVM.exe" --comment "vmname" --startvm --type headless "{873e0c62-acd2-4850-9faa-1aa5f0ac9c98}" +``` + +To uninstall a package + +```bash +apt-get -qy --purge remove +``` + +To uninstall a package without removing the settings + +```bash +apt-get -qy remove +``` + +On your home computer, Ubuntu has significant ease of use advantages. On +the cloud, where computing power costs and you are apt to have a quite +large number of quite small servers, Debian has significant cost +advantages, so perhaps should have Debian locally, despite its gross +pain the ass problems, in order to have the same system in the cloud and +locally. + +# installing a Wordpress blog on a new domain + +Assuming you have a backup of the files and the database. + +Create a freshly installed empty blog on the target site using one of the many +easy Wordpress setups. + +Copy over all the old files except wp_config.php + +Edit the wp_config file so that the `table_prefix` agrees with the original blogs wp_config.php `table_prefix`. + +Delete the new blog tables with new blog’s table prefix from the new blog’s database, and upload the old blog’s tables from the new blog’s database + +Upload the sql file to the new blog database. + +Should now just work. + +# Integrated Development Environments + +The cross platform open source gcc compiler produces the best object +code, but is not debugger friendly. The cross platform CLang is debugger +friendly, but this is not that useful unless you are using an ide +designed for Clang. Visual Studio has the best debugger – but you are +going to have to debug on windows. + +In Clang and gcc, use valgrind. + +In Visual Studio, [enable the CRT +debugger](https://docs.microsoft.com/en-us/visualstudio/debugger/finding-memory-leaks-using-the-crt-library). + +```C +#ifdef _MSC_VER +#ifdef _DEBUG +// Testing for memory leaks under Microsoft Visual Studio +#define _CRTDBG_MAP_ALLOC +#include +#include +#endif +#endif +``` + +At the start of program + +```C +#ifdef _MSC_VER +#ifdef _DEBUG +// Testing for memory leaks under Microsoft Visual Studio +_CrtSetDbgFlag ( _CRTDBG_ALLOC_MEM_DF | _CRTDBG_LEAK_CHECK_DF ); +#endif +#endif +``` + +At the end of program + +```C +#ifdef _MSC_VER +#ifdef _DEBUG +// Testing for memory leaks under Microsoft Visual Studio +_CrtSetReportMode( _CRT_ERROR, _CRTDBG_MODE_DEBUG ); +#endif +#endif +``` + +However, this memory leak detection is incompatible with wxWidgets, +which does its own memory leak detection. + +And since you are going to spend a lot of time in the debugger, Windows +Visual Studio recommended as the main line of development. But, for +cross compilation, wxWidgets recommended. + +Code::Blocks (wxSmith GUI developer) is one open source cross platform which +may be out of date. + +wxSQLite3 incorporates SQLite3 into wxWidgets, also provides ChaCha20 +Poly1305 encryption. There is also a wrapper that wraps SQLite3 into +modern (memory managed) C++. + +wxSQLite3 is undergoing development right now, indicating that wxWidgets +and SQLite3 are undergoing development right now. `wxSmith` is dead + +`Tk` is still live, but you get confusingly directed to the dead version. + + +## Model View Controller Architecture + +This design pattern separates the UI program from the main program, which is +thus more environment independent and easier to move between different +operating systems. + +The Model-view-controller design pattern makes sense with peers on the +server, and clients on the end user’s computer. But I am not sure it makes +sense in terms of where the power is. We want the power to be in the +client, where the secrets are. + +Model +: The central component of the pattern. It is the application’s dynamic data + structure, independent of the user interface.[5] It directly manages the +data, logic and rules of the application. + +View +: Any representation of information such as a chart, diagram or table. +Multiple views of the same information are possible, such as a bar chart for + management and a tabular view for accountants. + +Controller +: Accepts input and converts it to commands for the model or view. + +So, a common design pattern is to put as much of the code as possible into +the daemon, and as little into the gui. + +Now it makes sense that the Daemon would be assembling and checking large +numbers of transactions, but the client has to be assembling and checking +the end user’s transaction, so this model looks like massive code +duplication. + +If we follow the Model-View-Controller architecture then the Daemon provides +the model, and, on command, provides the model view to a system running on +the same hardware, the model view being a subset of the model that the view +knows how to depict to the end user. The GUI is View and Command, a +graphical program, which sends binary commands to the model. + +Store the master secret and any valuable secrets in GUI, since wxWidgets +provides a secret storing mechanism.  But the daemon needs to be able to run + on a headless server, so needs to store its own secrets – but these secrets + will be generated by and known to the master wallet, which can initialize a + new server to be identical to the first.  Since the server can likely be +accessed by lots of people, we will make its secrets lower value. + +We also write an (intentionally hard to use) command line view and command +line control, to act as prototypes for the graphical view and control, and +test beds for test instrumentation. + +## CMake + +CMake is the best cross platform build tool, but my experience is that it is +too painful to use, is not genuinely cross platform, and that useful projects +rely primarily on autotools to build on linux, and on Visual Studio to build +on Windows. + +And since I rely primarily on a pile of libraries that rely primarily on +autotools on linux and Visual Studio on windows ... + +## Windows, Git, Cmake, autotools and Mingw + +Cmake in theory provides a universal build that will build stuff on both +Windows and linux, but when I tried using it, was a pain and created +extravagantly fragile and complicated makefiles. + +Libsodium does not support CMake, but rather uses autotools on linux like +systems and visual studio project files on Windows systems. + +wxWidgets in theory supports CMake, but I could not get it working, and most people use wxWidgets with autotools on linux like systems, and visual studio project files on Windows systems. Maybe they could not get it working either. + +Far from being robustly environment agnostic and shielding you from the +unique characteristics of each environment, CMake seems to require a whole +lot of hand tuning to each particular build environment to do anything useful. + +1. Install [7zip](http://7-zip.org/download.html). + +1. Install [Notepad++](https://notepad-plus-plus.org/download/). + +1. Install [MinGW](http://tdm-gcc.tdragon.net/about) using TDM-GCC, as + the MinGW install is user hostile, and the Code::Blocks install of + MinGW broken. Also, wxWidgets tells you to use the TDM environment. + +1. Download Git from [Git for Windows](https://gitforwindows.org/) and + install it. (This is the successor to msysgit, which has a + [walkthrough](https://nathanj.github.io/gitguide/tour.html).) Select + Notepad++ as the editor. + + Note that in any command line environment where one can issue Git + commands, the commands `git gui` and `git gui citool` are available. + +1. Install [MinGW](http://tdm-gcc.tdragon.net/about) using TDM-GCC, as + the MinGW install is user hostile, and the Code::Blocks install of + MinGW broken. + +1. Download your target project using Git. + +1. Open a Windows PowerShell and navigate to the folder where you just + put your target project. + +1. Execute the following commands: + +```bat +cd build +cmake .. -G "MinGW Makefiles" +mingw32-make +``` + +## Android + +There is no satisfactory android running under Oracle Virtual Box. +However, Google supports development environments running under windows. + +Trouble is, for android clients, you will want to develop primarily in +JavaScript with a bit of Java. + +## JavaScript + +JavaScript delivers the write once, run anywhere, promised by Java, and, +unlike Java,delivers distributed computing. This, however, requires the +entire java ecology, plus html and css. And I don’t know JavaScript. +But more importantly, I don’t know the JavaScript ecology: + +The JavaScript ecology is large and getting larger, [as parodied by +Hackernoon](https://hackernoon.com/how-it-feels-to-learn-JavaScript-in-2016-d3a717dd577f). + +For an intro into JavaScript and the accompanying (large) ecology, +telling you what small parts of the forest you actually need to get an +app up: [A study plan to cure JavaScript +fatigue](https://medium.freecodecamp.org/a-study-plan-to-cure-JavaScript-fatigue-8ad3a54f2eb1). + +# Git + +To set up Git on the cloud, +[see](https://git-scm.com/book/en/v2/Git-on-the-Server-Setting-Up-the-Server) +and to use git on the cloud +[see](http://blog.davidecoppola.com/2016/12/how-to-set-up-a-git-repository-locally-and-on-a-remote-server/). + +On my system, I ssh into the remote system `reaction.la` as the user +`git` and then in the `git` home directory: + +```bash +mkdir MyProject.git +cd MyProject.git +git init --bare +``` + +and on my local system I launch the git bash shell, and go to the +`MyProject` directory. I copy a useful .gitinore and useful +.gitattributes file into that directory, then launch the bash git shell + +```bash +git init +git add * +git commit -m"this is a project to so and so" +git remote -v +git remote add origin git@reaction.la:~/MyProject +git remote -v +git push -u origin --all # pushes up the repo and its refs for the first time +git push -u origin --tags +``` + +Push,of course, requires that I have the ssh keys in putty format where +putty can find them, and another copy in openssh format where git can +find them. Git expects the ssh keys in .ssh + +If you ssh into the other system instead of puttying into it, only need +your keys in one place, which is simpler and safer + +Invoke `ssh-keygen -t ed25519 -C comment` under `git bash` to +automagically set up everything on the client side, then replace their +private key with the putty key using putty key gen’s convert key, and +their public key with the putty key gen copy and paste public key. + +Make sure the config file `~/.ssh/config` contains + + Host reaction.la + HostName reaction.la + Port 22 + IdentityFile ~/.ssh/id_ed25519 + +Host is the petname, and HostName the globally unique name.\ +An example of the use of petnames is + + Host project3 + User git + HostName github.com + IdentityFile ~/.ssh/project3-key + Host publicid + User git + HostName github.com + IdentityFile ~/.ssh/publicid-key + Host github.com + User git + IdentityFile ~/.ssh/github-key + +Putty likes its keys in a different format to git and open ssh, and +created pageant and plink so that git and openssh could handle that +format, but pageant and plink are broken. Convert format works, tplink +hangs. Just make sure that there is one copy as expected by git and +openssh, and one copy as expected by Putty. + +Save the private key in ssh format with no three letter extension, and +the corresponding public key in putty key gen’s copy and paste format +with the three letter extension.pub + +## Git Workflow + +You need a .gitignore file to stop crap from piling up in the +repository, and because not everyone is going to handle eol and locales +the same way, you need to have a .gitattributes file, which makes sure +that text files that are going to be used in both windows and Linux are +eol and utf-8, while text files that will be used only in windows are +crlf + +At github.com, create a new repository + +```bash +cd \development\MyProject +git init +git config --global user.name studi-c +git config --global core.editor 'C:/Program Files (x86)/Notepad++/notepad++.exe' -multiInst -notabbar -nosession -noPlugin +git config --global user.email studio@digsig.net +git config --global core.autocrlf false +git remote add origin https://github.com/studi-c/nameofmynewrepository.git +git add --all --dry-run +git add --all +git diff --cache +git commit -m "Initial revision" +git push origin master +``` + +After I make a change and test it and it seems to work: + +```bash + git pull origin master +``` + +Test that the application still works after pulling and merging other +developers’ changes + +```bash +git diff . +git add . +git diff --staged HEAD +git commit -m "My change" +git push origin master +``` + +For an more complete [list of commands, with useful +examples](https://www.siteground.com/tutorials/git/commands/). + +To make a git repository world readable, you need `git daemon` running, +but that a half measure, for if you publish your code to the world, you +want the world to contribute, and you will need gitlab to manage that. + +A [simpler way of making it +public](https://git-scm.com/book/en/v2/Git-on-the-Server-The-Protocols) +is to have the post-update hook turn it into a old plain dumb files, and +then put a symlink to your directory in the repository in your apache +directories, whereupon the clone command takes as its argument the +directory url (with no trailing backslash). + + +## Sharing git repositories + +### Git Daemon + +git-daemon will listen on port 9418. By default, it will allow access to any directory that looks like a git directory and contains the magic file git-daemon-export-ok. + +This is by far the simplest and most direct way of allowing the world to get at your git repository. + +### Gitweb + +Does much the same thing has git-daemon, makes your repository public with a +prettier user interface, and somewhat less efficient protocol. + +Gitweb provides a great deal of UI for viewing and interacting with your +repository, while git-daemon just allows people to clone it, and then they can +look at it. + +### [gitolite](https://gitolite.com/gitolite/) + +It seems that the lightweight way for small group cooperation on public +projects is Gitolite, git-daemon, and Gitweb. + +Gitolite allows you to easily make people identified by their ssh public key +and the filename of the file containing their public key write capability to +certain branches and not others. + +On Debian host `apt-get install gitolite3`, though someone complains this version is not up to date and you should install from github. + +It then requests your public key, and you subsequently administer it through +the cloned repository `gitolite-admin` on your local machine. + +It likes to start with a brand new empty git account, because it is going to +manage the authorized-keys file and it is going to construct the git +repositories. + +Adding existing bare git repositories (and all git repositories it manages +have to be bare) is a little bit complex. + +So, you give everyone working on the project their set of branches on your +repository,and they can do the same on their repositories. + +This seems to be a far simpler and more lightweight solution than Phabricator +or Gitlab. It also respects Git’s inherently decentralized model. Phabricator +and Gitlab provide a great big pile of special purpose collaboration tools, +which Gitolite fails to provide, but you have to use those tools and not other tools. Gitolite seems to be overtaking Phabricator. KDE seems to abandoning Phabricator: + +> The KDE project [uses](https://community.kde.org/Sysadmin/GitKdeOrgManual) +> gitolite (in combination with redmine for issue tracking and reviewboard for +> code review). Apart from the usual access control, the KDE folks are heavy +> users of the "ad hoc repo creation" features enabled by wildrepos and the +> accompanying commands. Several of the changes to the "admin defined +> commands" were also inspired by KDE’s needs. See +> [section 5](https://community.kde.org/Sysadmin/GitKdeOrgManual#Server-side_commands) and +> [section 6](https://community.kde.org/Sysadmin/GitKdeOrgManual#Personal_repositories) of the above linked page for details. + +So they are using three small tools, gitolite, redmine, and reviewboard, +instead of one big monolithic hightly integrated tool. Since we are creating a messaging system where messages can carry money and prove promises and +context, the eat-your-own dogfood principle suggests that pull requests and +code reviews should come over that messaging system. + +Gitolite is designed around giving identified users controlled read and +write access, but [can provide world read access] through gitweb and git-daemon. + +[can provide world read access]:https://gitolite.com/gitolite/gitweb-daemon#git-daemon +"gitweb and git-daemon – Gitolite" + +### [Gitea] and [Gogs] + +[Gitea]:https://gitea.io/en-us/ + +[Gogs]:https://gogs.io +"Gogs: A painless self-hosted Git service" + +Gitea is the fork, and Gogs is abandonware. Installation seems a little +scary, but far less scary than Gitlab or Phabricator. Like Gitolite, it expects +an empty git user, and, unlike Gitolite, it expects a minimal ssh setup. If +you have several users with several existing keys, pain awaits. + +It expects to run on lemp. [Install Lemp stack on Debian](#lemp-stack-on-debian) + +[Gitea Install instructions:](https://landchad.net/gitea) + +It's default identity model is username/password/email, but it supports the +username/ssh/gpg user identity model. The user can, and should, enter an +ssh and gpg key under profile and settings / ssh gpg keys, and to +prevent the use of https/certificate authority as a backdoor, require +commits to be gpg signed by people listed as collaborators. + +If email is enabled, password reset is by default enabled. Unfortunately +email password reset makes CA system the root of identity so we have to +disable it. We need to make gpg the root of identity, as a temporary +measure until our own, better, identity system is working. + +It is development model is that anyone can fork a repository, then submit a pull request, which request is then handled by someone with authority to push to that repository. + +gpg signatures should, of course, have a completely fake email address, +because the email address advertised in the gpg certificate will be +spammed and spearphished, rendering it useless. But gpg is completely +designed around being used with real email addresses. If you are using it +to sign, encrypt, and verify emails through its nice integration with the +thunderbird mail client, you have to use real email addresses. + +Uploading a repository to Gitea is problematic, because it only accepts +repositories on other Gitea, Gog, Gitlab, and Github instances. So you have +to create a fresh empty repository on Gitea, set it to accept an ssh key +from your `.ssh/config`, and merge your existing repository into it. + +Download the empty gitea repository. Make it even emptier than it already +is with `git rm`. Because the histories are unrelated, almost anything with +the same name will cause a merge conflict. + +```bash +git rm * +git commit -am "preparing to merge into empty repository" +git push +``` + +Got to the real repository, and merge the empty gitea repository on top of it: + +```bash +git pull myemptygitearepository --allow-unrelated-histories +# The histories are no longer unrelated, so push will work. +``` + +We are doing this in the real repository, so that the original history remains +unchanged - we want the new empty gitea repository on top, not underneath. + +Comes with password based membership utilities and web UI ready to +roll, unlike Gitolite which for all its power just declares all the key +management stuff that it is so deeply involved in out of scope with the result +that everyone rolls their own solution ad-hoc. + +These involve fewer hosting headaches than the great weighty offerings of +GitLab and Phabricator. They can run on a raspberry pi, and are great ads for +the capability of the Go language. + +Gitea, like Gitolite, likes to manage people’s ssh keys. You have to upload +your ssh and gpg public key as part of profile settings. Everything else, is +unfortunately, password based and email based, which is to say based on the +domain system and certificate authority system, which is an inherent massive +security hole, since the authorities can seize the website and put anything +on it they want. Therefore, need to use the [Gitea signing feature]( +https://docs.gitea.io/en-us/signing/ +"gpg Commit Signatures - Docs") + +This seems to require Gitea to do a lot of signing, which is a gaping +security hole, though it can be used with the gpg feature allowing short +lived subkeys. + +Will sign with subkey, because MasterKey should not be available on the server + +### To set up a system with gpg subkeys but without the master key + +This is horribly painful, and we need to create a better system and eat our +own dogfood. + +We ignore the Gpg Web of Trust model and instead use the Zooko identity model. + +We use Gpg signatures to verify that remote repository code is coming from an unchanging entity, not for Gpg Web of Trust. Web of Trust is too complicated and too user hostile to be workable or safe. + +Never --sign any Gpg key. --lsign it. + +Never use any public gpg key repository. + +Never use any email address on a gpg key unless it is only used for messages relating to a single group and a single purpose, or a fake email. + +Gpg fundamentally lacks the concept of one entity acting for and on behalf +of another. A subkey should be identified by the name of the master key +followed by a subname intelligible to humans, and a sequence of grant of +authority values intelligible to computers, represented by either a variable +precision integer representing a bit string, each bit corresponding to a an +authority value, which bitstring may contain a flag saying that further +authorities are represented by a null terminated Dewey decimal sequence rather +than by one bits in a sparse bitstring. + +But for the moment: + +```bash +gpg --expert --full-gen-key +``` + +Select 9 ECC and ECC, then select curve25519, then 0, key does not expire. +(This is going to be the rarely used master key, whose secret will be kept in +a safe place and seldom used) + +Much open source cryptography has been backdoored. I have no reason to +suppose that gpg is backdoored, other than that there is a great deal of +backdooring going around, but I have positive reason to suppose that +curve25519 has *not* been backdoored in gpg, and even if I did not, the +fewer cryptographic algorithms we use, the smaller the attack surface. + +Gitlab strongly recommends using only ED25519 ssh keys to interact with +git repositories, and I strongly recommend using only ED25519 ssh keys +to interact with repositories and only 25519 gpg keys to authenticate +commits and identify committers. I mention Gitlab not because I regard +them as a highly authoritative source, but to show I am not the only one +around who is paranoid about broken and corrupted cryptographic code. + +Everyone should use the same algorithm to reduce the attack surface. + +Name `«master key»>` (use a fake email address) Gpg was designed to +make email secure, but email is not secure. We will be using this as the +root of identities in Git, rather than to authenticate email. Use this root of +identity only for project related matters, only for the authentication of code +and design documents. Don't use this identity for other purposes, as this +will increase the risk that pressure or unpleasant consequences will be +applied to you through those other activities. Don't link this identity to +your broader identity and broader activities, as pressure is likely to applied +to introduce hostile code and strange design decisions that facilitate other +people's hostile code. This happens all the time in projects attempting to +implement cryptography. They get one funny feature after another whose +only utility is facilitating backdoors. + +gpg will ask you for a passphrase. If your keyfile falls into enemy hands, +your secret key is subject to offline dictionary attack, against which only a +very strong passphrase, with about 128 bits of entropy, can protect it. +Which is a problem that crypto wallets address, and gpg fails to address. +Either use a strong non human memorable passphrase or else use a trivial +passphrase or no passphrase, and instead export and hide the secret key +file and delete it from gpg when you are done. A human cannot remember +a strong passphrase. Write it down, in pencil, and hide it somewhere. +(Pencil does not fade, but some inks fade) If you can remember the +passphrase, and someone gets at your keyfile, it is unlikely to protect you +keyfile. A strong passphrase looks like a wallet master secret. + +Gpg's passphrases are merely a nuisance. They fail to serve a useful +purpose if used in the manner intended. Wallets came under real attack, +and now do things correctly. One of our objectives is to replace gpg for git +and gpg for secure messaging with something that actually works, with a +wallet, but for the moment, we use what we have. + +Now create another similar subkey. This time give it an expiry date in the +near future + +```bash +gpg --expert --edit-key «master key» +addkey +``` + +Rather than protecting your primary keys with a useless password, you +should export them to a safe place, such as a pair of thumbdrives, and then +delete them, to be re-imported when you need them to add a new subkey +when your subkey expires. + +```bash +save +gpg --list-keys --with-subkey-fingerprints --with-keygrip «master key» +gpg -a --export-keys «master key» +gpg -a --export-secret-keys «master key» +``` + +Then look in at the secret keys listed by keygrip in +`$HOME/.gnupg/private-keys-v1.d/KEYGRIP.key` and delete any +secret keys of the master key other than short lived subkey you just +added. + +```bash +gpg --list-keys --with-subkey-fingerprints --with-keygrip «master key» +# secret keys that have already been deleted will be marked with a # +``` + + You should make sure that any secret keys with no expiration or an + expiration in the distant future are marked with that sign, and if they are + not, delete them. + +On non airgapped machines, you will delete the primary key, or not +import it in the first place. + +After all this complicated, easy to get wrong, and confusing rigmarole, +you should, upon importing and trusting, have a copy of gpg that can sign +commits with your short lived subkey as an agent for your long lived +master key. + +Git can now still sign with the non expiring master key, because it still has +a subkey that will expire in the near future. + +[sovereign corporations]:social_networking.html#many-sovereign-corporations-on-the-blockchain + +We need to replace this with our own system and make life a whole lot +simpler for someone with a shortlived subkey on a cloud computer and a +master key offline or on an airgapped computer. + +Making life simpler for ourselves is a step towards making life +simpler for someone who is likely to give us money. + +This an important eat-your-own-dogfood job. But doing it right, rather +than a user hostile patch on a patch, is likely to be hard, because rewriting +Git to use our keys and our identities is likely to be an enormous job. Git +also needs fixing, not only because it uses the domain name identity model +rather than secret key but because it uses an insecure hash. + +Obviously we need to eventually replace all this complicated rigmarole +with a system whereby one identity on one wallet can send a message to +another identity on another wallet granting it an identity with authority to +act as agent for the first identity. The primary use case for such a system is +not signing code, though that will be our first use of it, but to enable +[sovereign corporations] to act through remote employees and safely use +servers in the cloud. Information wants to be free, but programmers want +to be paid. +### Phabricator + +Server Size : 2GB Ram – 1 CPU Core – 50GB SSD +If you have more than five users, this may not suffice, but you can limp +along OK with one gigabyte. + +[Installation and configuration of +Phabricator](https://www.vultr.com/docs/how-to-install-and-configure-phabricator-on-centos-7) +(which is also likely to be useful in installing PhpAdmin, because it +covers configuring your “MySQL” (Actually MariaDB) database. + +Configuring Phabricator [requires a whole lot of configuration of ssh +and +git](https://secure.phabricator.com/book/phabricator/article/diffusion_hosting/), +so that ssh and git will work with it. + +Phabricator notifications require [node.js](http://nodejs.org/), will +not run with apache. Ugh. But on the other hand, Comen needs node.js +regardless. But wordpress requires PHP, not sure that is going to play +nice with node.js. Node.js does not play well with apache PHP, and +Phabricator seems to use both of them, but likely only uses node.js for +notifications, which can wait. Usual gimmick is that to use Apache’s +ProxyPass directive to pass stuff to node.js. Running both node.js and +apache/PHP is likely to need a bigger server. Apache 2.4.6 has support +for proxying websockets mod_proxy_wstunnel + +The Phabricator website suggests nginx + php-fpm + “MySQL” (MariaDB) + +PHP. Probably this will suffice for Wordpress. nginx is the highest +performance web server, but it is not node.js + +Apache, node.js, and nginx can all coexist, by routing stuff to each +other (usually with the highest performing one, nginx, on top) but you +will need a bigger server that way + +Blender is a huge open source success, breaking into the big time, being +free and open tool for threedee drawing. It’s development platform is +of course self hosted, but not on Gitlab, on Phabricator. Maybe they +know what they are doing. + +[Phabricator](https://secure.phabricator.com/book/phabricator/article/installation_guide/) +is a development environment provided by Phacility. It is written in in +PHP, in a phabricator development environment, which is designed to +support very large development communities and giant corporations. It is +written in PHP, which makes me instantly suspicious. But it is free, +and, unlike GitLab, [genuinely open +source](https://www.phacility.com/via/pricing-download/). + +Being written in PHP, assumes a Lamp stack, apache2, php 5.2 or later, +mysql 5.5 or later + +Phabricator assumes and enforces a single truth, it throws away a lot of the +inherent decentralization of git. You always need to have one copy of +the database, which is the King – which people spontaneously do with git, + but with git it is spontaneous, and subject to change. + +KDE seems to be moving away from Phabricator to Gitolite, but they have an +enormously complex system, written in house, for managing access, of which +Gitolite is part. Perhaps, Phabricator, being too big, and doing too much was +inflexible and got in their way. + +Github clients spontaneously choose one git repository as the single truth, but then you have the problem thatGithub itself is likely to be hostile. An easy solution is to have the Github respository a clone of the remote repository, without write privileges. + +### Gitlab repository + +Git, like email, automatically works – provided that all users have ssh +login to the git user, but it is rather bare bones, better to fork out +the extra cash and support gitlab – but gitlab is far from automagic, +and expects one git address for git and one chat address for matterhorn, +and I assume expects an MX record also. + +Gitlab is a gigantic memory hog, and needs absolute minimum of one core +and four gig, and two cores and eight gig strongly recommended for +anything except testing it out and toying with it. It will absolutely +crash and burn on less than four gig. If you are running gitlab, no cost +advantage to running it on debian. But for my own private vpn, huge cost +advantage. + +[Gitlab absolutely requires postgreSQL](https://about.gitlab.com/installation/\#debian). +They make a half assed effort to stay MySQL compliant, but fall short. +Not a big disk space hog – ten gigabytes spare will do, so fine on thirty +two gigabyte system. + +Gitlab Omnibus edition contains the postfix server, thus can send and +receive email to is host address + +Setup Gitlab with Protected branch flow + +> With the protected branch flow everybody works within the same GitLab +> project. The project maintainers get Maintainer access and the regular +> developers get Developer access. The maintainers mark the +> authoritative branches as ’Protected’. The developers push feature +> branches to the project and create merge requests to have their +> feature branches reviewed and merged into one of the protected +> branches. By default, only users with Maintainer access can merge +> changes into a protected branch. Each new project requires non trivial +> manual setup. + +But, seems likely to be easier just to use the main gitlab site, as +least until my project is far enough advanced that the evil of github is +likely a threat. + +Gitlab is intended to be hosted on your own system, but to learn how to +use it in a correctly configured gitlab environment, and to learn what a +correctly configured gitlab environment looks like and how it is used, +going to need an account on gitlab. + +Gitlab requires that the Openssh port 22, the http port 80, and the +https port 443 be forwarded. Http should always get automatic redirect +to hppts governed by a lets encrypt certificate. + +[GitLab +Mattermost](https://docs.gitlab.com/omnibus/gitlab-mattermost/#getting-started) +expects to run on its own virtual host. In your DNS you would then have +two entries pointing to the same machine, e.g. gitlab.reaction.la and +mattermost.reaction.la. GitLab Mattermost is disabled by default, to +enable it just put the external url in the configuration file. + +Github, on the other hand, allows you to point [your own domain name to +your custom (static) github website and git +repository](https://pages.github.com/) as if on your own system. + +I am suspicious of placing your own website on someone else’s system, +especially a system owned by social justice warriors, and the +restriction to static web pages is likely intended to facilitate +political censorship and law enforcement, and physical attacks on +dissidents by nominally private but state department supported thugs and +Soros supported thugs. + +### Gitlab on the cloud + +[Omnibus edition of gitlab](https://about.gitlab.com/installation/), +usually available already configured to a cloud provider. + +[Instructions for installing it +yourself](https://docs.gitlab.com/omnibus/README.html). + +It is like advisable to setup apache, the apache virtual hosts, and the +apache certificates first. + +Gitlab markdown, like Github markdown, can mostly succeed in handling +html tags. + +We are not going to build on the cloud, but we will have source code, +chat, and code of conduct on the cloud, probably on `git.cpal.cw` and +`chat.cpal.cw` + +Configuring gitlab is non trival. You want anyone to be able to branch, +and anyone to be able to issue a pull request, but you only want +authorized users to be able to merge into the master. + +Since we want to be open to the world, implement +[recaptcha](https://docs.gitlab.com/ee/integration/recaptcha.html) for +signups, but allow anyone in the world to pull without signing up. To +create a branch, they have to sign up. Having created a branch, they can +issue a pull request for an authorized user to pull their branch into +the master, they have to sign up. + +Gitlab workflow is that you have a master branch with protected access, +a stable branch with protected access, and an issue tracker. + +You create the issue before you push the branch containing fixes for the +issue. There must be at most one branch for every issue. + +Developers create a branch for any issues they are trying to fix, and +their final commit should say something like "fixes \#14" or "closes +\#67." and then, when the branch is merged into the master, the issue +will be marked fixed or closed. + +[Digital Ocean offers a free entry](https://m.do.co/c/8ec94d3d2c7a), +and a quite cheap system + +But [virmach](https://github.com/Nyr/openvpn-install), +[even cheaper](https://billing.virmach.com/cart.php?gid=1). + +Eight gig, two cores, which you will need to run gitlab for everyone, is +[Debian 9 enterprise edition on virmacht.com](https://billing.virmach.com/cart.php?a=confproduct) +\$40 per month. + +Also, [vpn on the cloud](https://github.com/Nyr/openvpn-install). + +Currency project should be [hosted on digital ocean at git.reaction.la, at +\$20 per month (Four gig, two cores), using Gitlab free omnibus +edition](ww.digitalocean.com/community/tutorials/how-to-use-the-gitlab-user-interface-to-manage-projects). +They suggest configuring your own Postfix email server on the machine +also, but should this not be automatic? Probably already in the +DigitalOcean Gitlab droplet. + +[How to us the gitlab one click install image to manage git +repositories](https://www.digitalocean.com/community/tutorials/how-to-use-the-gitlab-one-click-install-image-to-manage-git-repositories). + +You will need to get your SSL certificate from cyberultra and supply it +to Gitlab (though gitlab has built in lets-encrypt automation, so maybe +not). + +Subdomains are a nameserver responsibility, so you really have to point +your domain name nameservers to cyberultra, or else move everything to +digital ocean. + +All digital ocean ips, except floating ips, are static. You will need an +A record and an AAAA record, the AAAA record for IP6 + +[Getting started with gitlab and +digitalocean](https://about.gitlab.com/2016/04/27/getting-started-with-gitlab-and-digitalocean/). + +Gitlab did not want to support fully browsable public repositories, but +they have been supported since 6.2 + +Gitlab omnibus edition comes [integrated with +Mattermost](https://docs.mattermost.com/deployment/sso-gitlab.html), but +mattermost is turned off by default. + +Before spending money and going public, you might want to [install +locally](https://about.gitlab.com/installation/) and run on your local +system. [Enable mattermost](https://forum.mattermost.org/t/where-to-find-mattermost-after-installing-gitlab-omnibus/175/5). + +Since gitlab will have the root web page on `git.jim.com`, you will need +another DNS entry pointing at the same host, something like +`chat.jim.com`. So, though both on the same machine, one is the root +http page when accessed by one domain name, [the other the root entry +when accessed by another domain +name](https://gitlab.com/gitlab-org/gitlab-ce/issues/13530). + +# Implementing Gui in linux + +## coupling to the desktop + +To couple to the desktop requires a pile of information and configuration, +which most people ignore most of the time. To the extent that they provide it, +they seem to write it for the Gnome based desktops, Cinnamon and Mate – more +for Mate because it is older and has changed less. + +Since wxWidgets is written for GDT in its linux version, it is written for Gnome. + +Gnome3, the default Debian desktop, is broken, largely because they refuse to +acknowledge that it is broken, so the most standard linux environment, +the one for which your practices are least likely to break on other linuxes, +is Debian with Lightdm and Mate. (pronounced Mahtee) + +Looks to me that KDE may be on the way out, hard to tell, Gnome3 is definitely +on the way out, and every other desktop other than Cinnamon and Mate is rather +idiosyncratic and non standard. + +Lightdm-Mate has automatic login in a rather obscure and random spot. +Linux has its command line features polished and stable, but is still +wandering around somewhat lost figuring out how desktops should work. + +Under Mate and KDE Plasma, bitcoin implements run-on-login by generating a +`bitcoin.desktop` file and writing it into `~/.config/autostart` + +It does not, however, place the `bitcoin.desktop` file in any of the expected +other places. + +Under Mate and KDE Plasma, bitcoin stores its configuration data in +`~/.config/Bitcoin/Bitcoin-Qt.conf`, rather than in `~/.bitcoin` + +wxWidgets attempts to store its configuration data in an environment +appropriate location under an environment appropriate filename. + +[desktop files]:https://developer.gnome.org/integration-guide/stable/desktop-files.html.en + +It does not, however, seem to have anything to handle or generate +[desktop files]. + +[Linux desktop standard]:https://specifications.freedesktop.org/menu-spec/latest/ + +[desktop files] are the [Linux desktop standard] for a gui program to +integrate itself into the linux desktop, used to ensure your program appears +in the main application menu, the linux equivalent of the windows startup +menu. + +Getting your desktop file into the startup menu is slightly different in KDE +to the way it is in Gnome, but there are substantial similarities. +FreeDesktop tries to maintain and promote uniformity. Gnome rather +casually changed the mechanism in a minor release, breaking all previous +desktop applications. + +## Flatpack + +Every Linux desktop is different, and programs written for one desktop +have a tendency to die, mess up, or crash the desktop when running on +another Linux desktop. + +Flatpack is a sandboxing environment designed to make every desktop +look to the program like the program’s native desktop (which for +wxWidgets is Gnome), and every program look to the desktop like a +program written for that particular desktop. + +Flatpack simulates Gnome or KDE desktops to the program, and then translates +Gnome or KDE behaviour to whatever the actual desktop expects. To do this, it +requires some additional KDE configuration for Gnome desktop programs, and some +additional Gnome information for KDE desktop programs, and some additional +information to cover the other 101 desktops. + +WxWidgets tries to make all desktops look alike to the programmer, and Flatpack +tries to make all desktops look alike to the program, but they cover different +aspects of program and desktop behaviour, so are both needed. Flatpack covers +interaction with launcher, the iconization, the install procedure, and stuff, +which wxWidgets does not cover. + +Linux installs tend to be wildly idiosyncratic, and the installed program winds +up never being integrated with the desktop, and never updated. + +Flatpack provides package management and automatic updates, all the way from the +git repository to the end user’s desktop, which wxWidgets cannot. + +This is vital, since we want every wallet to talk the same language as every +other wallet. + +Flatpack also makes all IPC look alike, so you can have your desktop program +talk to a service, and it will be talking gnome iPC on every linux. + +Unfortunately Flatpack does all this by running programs inside a virtual +machine with a virtual file system, which denies the program access to the +real machine, and denies the real machine access to the program’s +environment. So the end user cannot easily do stuff like edit the program’s +config file, or copy bitcoin’s wallet file or list of blocks. + +A program written for the real machine, but actually running in the emulated +flatpack environment, is going to not have the same behaviours. The programmer +has total control over the environment in which his program runs – which means +that the end user does not. + +# Censorship resistant internet + +## [My planned system](social_networking.html) + +## Jitsi + +Private video conferencing + +[To set up a Jitsi meet server](andchad.net/jitsi) + +## [Zeronet](https://zeronet.io/docs) + +Namecoin plus bittorrent based websites. Allows dynamic content. + +Not compatible with wordpress or phabricator. Have to write your own dynamic site in python and coffescript. + +## [Bitmessage](https://wiki.bitmessage.org/) + +Messaging system, email replacement, with proof of work and hidden servers. + +Non instant text messages. Everyone receives all messages in a stream, but only the intended recipients can read them, making tracing impossible, hence no need to hide one’s IP. Proof of work requires a lot of work, to prevent streams from being spammed. + +Not much work has been done on this project recently, though development and maintenance continues in a desultory fashion. + +# Tcl Tk + +An absolutely brilliant and ingenious language for producing cross +platform UI. Unfortunately I looked at a website that was kept around for +historical reasons, and concluded that development had stopped twenty years ago. + +In fact development continues, and it is very popular, but being absolutely +typeless (everything is a conceptually a string, including running code) +any large program becomes impossible for multiple people to work on. +Best suited for relatively small programs that hastily glue stuff together - it +is, more or less, a better bash, capable of running on any desktop, and +capable of running a gui. diff --git a/docs/sharing_the_pool.md b/docs/sharing_the_pool.md new file mode 100644 index 0000000..6741cc8 --- /dev/null +++ b/docs/sharing_the_pool.md @@ -0,0 +1,87 @@ +--- +title: Sharing the pool +#katex +--- + +Every distributed system needs a shared data pool, if only so that peers can +find out who is around. + +The peers each have connection to several others, and from time to time +each notifies the other of data he does not know the other has. Trouble is, +there is no canonical order for new data, so when forming a new +connection, each has to notify the other of all his data. Which could be a +great deal of data, most of which, though far from all of which, both of +them already has. + +A common way of dealing with this problem is Bloom filters and cuckoo filters. You send the other a body of data such that the other can detect what data the other has, that he definitely does not have. + +A bloom filter is usually described as k hashes, though in fact, quicker to use one hash, divide the bits into two halves, to get two hashes, G and H, and then construct k hashes $F_i=G+i*H$ for $i=1,k$. + +The recipient generates his own bloom filter in the corresponding way, and any time he sets a bit that is not set in the data sent to him, knows he needs to send that data. + +Each could send the other his filter, or one could receive a filter, and send a run length compressed version containing those bits corresponding to hashes he failed to generate, whereupon the sender regenerates his filter, or already has a big ass hash table corresponding to his filter generation, and sends the missing items. + +# Efficient filters + +A bloom filter is defined by $k$, the number of hashes, $m$ the number of bits, and $n$, the number of elements. + +$$k\text{, the number of hashes,} \approx \bigg\lceil\frac{m\>\ln(2)}{n}\bigg\rceil$$ + +Sending a bloom filter is equivalent to sending a list of hashes of size $\log_2(n/k)+m$, run length compressed so as to be compressed down to only about three bits bigger than $m$, regardless of how big $n$ gets. + +Which is great, but if the number of hashes gets very large, large enough for $\log_2(\text{the number of hashes})$ to matter, you are likely to be in a situation where almost all hashes match, and thus carry no useful information for performing the merge, and also in a situation where you want few or no false positives, thus $m$ has to be far larger than is required to efficiently merge the bulk of the data, so you are repeatedly sending a huge amount of redundant and useless data in order to synchronize, though vastly less than if everyone repeatedly sent everyone their entire collection of data. + +For a 1% false positive rate, you need only ten bits for every item, regardless of how many items. + +Now suppose you had a $m$ items, and you wanted a one percent false positive rate, and you sent an fragment of each hash themselves, you would need $\log_2(m)+7$ bits per item. Which is bigger, but not hugely bigger. + +Now suppose you sorted the reduced hashes in order, and sent the difference between each hash and the next, using an efficient encoding (the distribution of values would be exponentially declining). + +Suppose you want a one in sixty four false positive rate. Then you want the average difference to be around sixty four. So a difference in the range 0 to 63 would consist of a 0 bit followed by six bits representing a value from 0 to 63, a value in the range 64 to 127 would consist of a 1 bit followed by a zero bit followed by six bits, a value in the range 128 to 191 would consist of two 1 bits followed by a zero bit followed by six bits -- and what do you know, we are looking at about the same size as a bloom filter. We are sending the low order part of the difference in binary, and the high order part in unary. + +Generalizing we use a binary representation that can represent a value close to the scale height, and a unary representation of a multiple of the range covered by the binary representation. + +If the scale height is $\bigcirc(1)$ it is a bit mask. If the scale height is $\bigcirc(2)$, and if we ignore zero differences (hash collisions) so that the minimal distance is $1$ + +Indeed, one can combine both methods for a sparse bloom filter that is sent with zero runs compressed. + +## compression of exponentially distributed values + +You find the expected scale height, the amount that causes the probability of a diff to diminish by half, round it to the nearest power of two. and express quantities as a unary multiple of that amount, and fixed width binary offset to that unary quantity. It might be convenient to pack the unary values and the fixed width values separately. If the scale height is very small, representing an item where runs are rare (or never happen by definition, the item represented is being a boundary) this nothing but unary, a bit mask. + +# Minimizing filter size + +But if both sides have vast collections of identical or near identical transactions, as is highly likely because they probably just synchronized with the same people, each item in a filter is going to convey very little information. Further, you can never be sure that you are completely synchronized except by setting a lot of bits for each item. + +## Merkle Patricia tree + +So, you build a Merkle patricia tree. + +And then you want to transmit a filter that represents the upper portion of the tree where the likelihood of a discrepancy between Bob's tree and Carol's tree is around fifty percent. When you see a discrepancy, you go deeper into that part of the tree on the next sub round. A large part of the time, the discrepancy will be a single transaction. When you have isolated all the discrepancies, rinse and repeat. Eventually the root hashes will agree, so the snapshot the Bob's concurrent process took is now synchronized to Carol, and the snapshot that Carol's concurrent process took is now synchronized to Bob. But new transactions have probably arrived, so time to take the next snapshot. + +You discover how deep that is by initially sending the full filter of vertex and leaf hashes for just a portion of the address space covered by the tree. From what shows up, in the next round you will be roughly right for filter depth. + +You do want to use a cryptographically strong hash for the identifier of the each transaction, because that is global public information, and we do not want people to be able to cook up transactions that will force hash collisions, because that would enable them to engage in Byzantine defection. But you want to use Murmur for the vertices of the tree that represents transactions that Bob does not yet know whether Carol already has, since that is bilateral information maintained by concurrent process that is managing Bob's connection with Carol, so Byzantine defection is impossible. When, however, Bob's concurrent process managing the connection with Carol whips up a Merkle patricia tree, it should use Murmur3, because there will be a lot of such processes generating a lot of Merkle patricia trees, but only one cryptographic hashes representing each transaction. Lots of such trees are generated, and lots discarded. + +[SMhasher]:https://github.com/aappleby/smhasher + +The official release of Murmur3 is in the [SMhasher] test suite, and is obsolete now that C++20 defines system, machine, and compiler independent access to bit operations and to endianness. + +Since we are hashing strong hashes, probably even Murmur3 is overkill. + +Instead, if we want to hash two 128 bit hashes into one 128 bit hash: + +Let the two 128 bit hashes be four 64 bit values, $U_0 2^{64} + U_1$ and $V_0 2^{64}+V_1$, and the resulting 128 bit hash is the two 64 bit values: + +$$(U_0g^3+U_1g^2+V_0g+V_1)\%2^{64}$$ +$$(U_1g^3+V_0g^2+V_1g+U_0)\%2^{64}$$ + +where $g=11400714819323198485$, the odd number nearest to $2^{64)} divided by the golden ratio + +Which would be a disastrously weak hash if our starting values were highly ordered, but is likely to suffice because our starting values are strongly random. Needless to say, it has absolutely no resistance to cryptographic attack, but such an attack is pointless, because our starting values are cryptographically strong, our resulting values don't involve any public commitments and we intend to reveal the preimage in due course. + +Come to think of it, we can get away with 64 bit hashes, provided we subsample the underlying cryptographically strong 256 bit hashes differently each time, since we do not need to get absolutely perfect synchronization in any one synchronization event. We can live with the occasional rare Merkle patricia tree that gives the same hash for two different sets of transactions. The error will be cleaned up in the next synchronization event. + +Thus the hash of two 64 bit hashes, $U$ and $V$, is $(Ug+V)\%2^{64}$. + +But when we synchronize to the total canonical order, we do need 256 bit cryptographically strong hashes, since concocting two sets of transactions that have the same hash could be used for Byzantine defection. But we only have to construct that tree once. diff --git a/docs/social_networking.md b/docs/social_networking.md new file mode 100644 index 0000000..5d37555 --- /dev/null +++ b/docs/social_networking.md @@ -0,0 +1,936 @@ +--- +title: + Social networking +--- +# the crisis of censorship + +We have a crisis of censorship. + +Every uncensored medium of public discussion is getting the treatment. + +We need a pseudonymous social network on which it is possible to safely +discuss forbidden topics. + +We also have a crisis of shills, spamming, and scamming. So we need +moderation. But to prevent censorship, we need entirely decentralized +moderation. + +People escape censorship to unmoderated anonymous platforms, but are +driven off those platforms by shills, spammers, and scammers. + +The difference between censorship and moderation is that a moderator acts +to protect the discussion from shills, spam, and scammers, while a censor +acts to prevent the discussion of dangerous ideas. + +What the censor is suppressing is stuff that is not generally known and not +generally available. What moderators are needed for is to suppress is stuff +that is very hard to avoid, because a thousand sock puppets are robotically +posting from the same script on a thousand forums. + +# Social net architecture + +You want to be able to message everyone in the world, but you don’t want +spammers and shills to be able to message you. (The difference between a +spammer and a shill is that a shill is spearphishing. A shill’s script is +narrowly targeted to a specific target community, like Trotsky declaring +himself to be a peasant in order to kill the peasants, or the innumerable pump +and dump “investment opportunities”.) + +You want to be able to communicate to the broad public (equivalent of +web pages) as well as specific individuals (equivalent of text messages, +email), and you want people to be able to comment before the broad +public on content you have communicated to the broad public (equivalent +of blogs). You want to be notified when someone sends a message +specifically to you, and you want to be notified when someone responds to +content you have communicated to the broad public, you want to be able +to turn off the ability to respond, or turn it off for any new respondents, +want to be able to close comments, and you want, with considerably less +urgency, to be notified when something appears before the broad public by +people you are following. + +Thus the social network system of followers, followed, buddies, and feeds, +which combines the functionality of the web, the functionality of +messaging, and functionality of the blog. We can model everything as a +web, except that some parts of the web are directed acyclic graphs under +some directionality criterion. A directed acyclic graph has an implicit time +order, hence notifications and feeds. + +if we try to distinguish between posts and comments as blogs do, there are +several tricky edge cases when we have replies to replies. + +Usenet made no substantial distinction between posts and comments. +Making the distinction, as blogs do, breaks the conversation and +complicates the code and the interface. + +Usenet died of shills, as for example the science fiction and fantasy +discussion groups dying as a side effect of progressive takeover of science +fiction fandom organizations. Somehow the conversation would invariably +turn into attacks on capitalism, the market, family, fatherhood, and private +property, which did not have much to do with science fiction and fantasy. + +If fantasy came up at all, it was in a conversation that presupposed the +evils of feudalism, and the Marxist history that capitalism was the +successor of feudalism, rather than in a conversation that presupposed the +reader’s yearning for Kingship, kinship, leadership, and the hero with a +thousand faces. + +In the end the conversation in the science fiction and fantasy Usenet +discussion groups tended to be about the power of classes as defined by +Marx, rather than about power of heroes, Kings, and vigilantes. And so the +Usenet science fiction and fantasy discussion groups, and just about every +Usenet discussion group, died of shills. + +Conversations became one man with one microphone attached to a +thousand megaphones, and replying was like talking back to the national +news broadcast, because you could reply to a shill, but the man giving the +shill her script was not listening, because he was running a hundred +similar shills, and his shill would just stick to her script, the script he had +assigned to her no matter what you replied to her script. + +Her replies to your reply would be unresponsive, because they came from +a script written by a man who had never thought about or foreseen your +reply. + +Conversations came to resemble the conversations you have with a non +player character in a video game, a scripted robotic simulation of actual +conversation, or the conversations on a help line with an unhelpful third +world help line worker to whom English is a second language, and who is +reading from a script, a script written by a man whose native tongue is +English, but his script is designed to deal with certain common problems +that do not in the slightest resemble the problem you have with the +product, because the man writing her script did not foresee your problem. + +Usenet was designed for conversations, but was hijacked by Harvard and +the New York Times to give one way lectures, resembling those given in +Harvard and the pages of the New York Times. Speaking back was +pointless, no one was listening, and eventually everyone stopped +speaking back, and Usenet died. + +The conversation in the Usenet science fiction and fantasy groups died, +because the shills would stick to script no matter what, providing the +superficial simulation of a conversation about fantasy and science fiction +without the substance of a real conversation. And the same was happening +in every Usenet group, though what was being shilled varied from one +group to the next, with an immense multitude of different shilling +organizations running an immense multitude of unresponsive scripts +promoting an immense multitude of different scams, most of them about +money rather than politics. + +Two real participants in the science fiction and fantasy Usenet group would +be having a conversation about the hero’s journey, and the computer of +the man writing the economic Marxist scripts would detect a keyword or key +phrase that it had been programmed to treat as relevant to a Marxist script, +and direct a shill to interject that script into the conversation, and the shill +would do so, mixing the script up a little with random fragments from the +actual conversation to prevent it from being too readily recognizable as the +same script already posted a thousand times before in a hundred Usenet +groups, and the conversation would be derailed. Or worse, a robot +promoting class conflict based on sexual preference classes would be +triggered on seeing the post by a robot promoting class conflict based on +economic classes, and the conversation thereafter would consist of one +robot pushing scripts about economic Marxist classes, and another robot +pushing scripts about Cultural Marxist classes with the bored economic +Marxist shill industriously interweaving random fragments from Cultural +Marxist script into the economic Marxist script, and the bored Cultural +Marxist shill industriously interweaving random fragments from the +economic Marxist script into the Cultural Marxist script to create the +semblance of a conversation where no actual conversation existed, merely +random variations on scripts already robotically pushed a thousand times +in a hundred places. + +The real participants eventually left the Usenet group leaving the robots to +robotically talk to each other. + +To prevent shills, you need moderation. But moderation is apt to get +centrally controlled and become censorship, shutting down vitally needed +conversations. We need mandatory moderation, to prevent what happened +to Usenet, but we need completely decentralized moderation, to prevent +censorship. + +If you don’t have a means of stopping one way broadcasts into a medium +designed for conversations, your social net is going to be flooded with one +way broadcasts. + +The particular axe that was being ground was not the problem. There is no +end of large, wealthy, and powerful organizations with axes to grind. In +the Science Fiction and Fantasy group, the primary problem seemed to be +organizations created and funded by Harvard and the New York Times +promoting cultural Marxism and old type economic Marxism, but if it had +not been them, would have been someone else promoting something else. +One way broadcasts provide the opportunity for power and money. + +So what should happen is that when you have a feed, and are a follower of +Bob, Bob’s public posts that you have not yet seen are marked as unread +in the feed, and listed by title or first line if no title. And if you look at one +of them, the thing he is replying to is just a click away, and the replies he +has approved are just a click away. But you don’t see, and cannot click on, +the replies he has not approved, unless you are following those +commenters also. Though he might give blanket approval to Alice’s feed, +in which case you would also see Alice and everything she approved. + +Because if you could click on them, a million shills and spammers would +reply, rendering the link worthless. + +If you want two way narrowcasts, you need a means to keep out +broadcasts, or else narrowcast memes sent by individuals to individuals +will be drowned out by mass produced broadcast memes sent to everyone +indiscriminately by large bureaucratic organizations. If it had not been +Harvard and the New York Times, might have been some megachurch. +Indeed, in the early days of the internet, it was some megachurch. + +In the early days of the internet, every group, regardless of topic, +regardless of its area of interest, was full of repetitious stuff about young +earth creationism, material indistinguishable from what I had seen before +in every other group, with any attempt to debate the posters receiving +unresponsive scripted responses to the unscripted response, responses +suspiciously similar to a thousand similarly unresponsive responses in a +thousand different interest groups attempting to pursue a thousand quite +different interests. + +If young earth creationist shilling died out, it was because bigger and more +powerful organizations with more important axes to grind got in on the +business of broadcasting one way broadcasts into narrowcast two way +media. The old indiscriminate one way broadcasts were drowned out by +new indiscriminate one way broadcasts reflecting the concerns of bigger +and more powerful organizations, in the end, the biggest and most +powerful organizations of all. + +When you look at your set of feeds, you want to see which feeds have +activity on them by real people that you are following, and when you look +at a particular feed, you want to see only new activity by the person you +are following, and also which items on the feed have new replies approved +by the person you are following. + +Which, unless the person you are following is issuing indiscriminate +approvals, is going to keep out the one way broadcasts by shills, +spammers, and scammers. And if he is issuing indiscriminate approvals, +people will soon stop following his feed in the same way everyone +dropped out of Usenet groups as the shilling of one way broadcasts became +intolerable. + +If Bob replied to Carol, and you are not following Carol, and you click on +Bob’s post, you should see in his post a link to the post by Carol that Bob +replied to, and if you click on that, you see Carol’s post containing a link +that will show all the replies to that Carol approved. Which might include +Dave’s reply to Bob’s reply to Carol, a reply to Bob that Carol approved, but +Bob did not. + +In which case, if you click on Carol’s replies link, you will see Dave’s +reply to Bob in her comments page, but if you click on Bob’s replies link, +you will not see it in Bob’s comments page. + +What posts you can see will depend on the path by which you reached +them. The mesh of reply links, reply-to links, and approval links form a +graph, and when you click around the conversation you are following that +graph. So people will learn to follow the paths of good moderators, and +will ignore the paths of bad moderators. + +The web also forms a graph. What is missing from the web whose lack +makes it not a social network is the time element: feeds. You cannot have a +conversation over the web. We need to support conversations, thus need to +have different and distinct reply links, reply-to links, and approvals +while the web only has one kind of link, losing the information that makes +a conversation a conversation. A social network is a superset of the web, +email, and instant messaging, and a moderated but truly decentralized +social network is going to replace all of them. + +So everything is a post – and a reply is just a post that has a reply-to link to +the post that it replies to. And the reply-to link works both ways, in that the +post with the link will appear in the comments page of the original post +that it links to, if approved by the poster it links to. + +So we treat replies and posts the same, everything a post. If you are +following someone, you get his posts on your feed, and when you see a +post that he made public, you can click on a link to the post that he replied +to, and a link to any replies to him, or replies to his replies, that he +approved. + +So everything you can click on and read has to be by someone you are +following, a post approved by someone you are following, or approved by +the author of the post that you are now reading, supposing you clicked on +it to see the replies that he approved. Otherwise we get the Usenet problem +of a million shills, scammers, and spammers. + +So, you can navigate to whole world’s public conversation through +approved links and reply-to links – but not every spammer, scammer, and +shill in the world can fill your feed with garbage. + +The underlying protocol and mechanism is that when you are following +Bob, you get a Bob feed from a machine controlled by Bob, or controlled +by someone that Bob has chosen to act on his behalf, and that when Bob +replies to someone, the post that he replies to is copied onto his machine, +containing a link to a feed controlled by the person he replies to, and +similarly with replies that he approves. So posts and links to feeds spread +around the net Usenet style, being duplicated on every machine that +comments on them or approves them and every machine that follows a +feed that contains them. You access a feed BitTorrent style, sharing Bob’s +feed with everyone that follows Bob. Each feed is a mutable torrent, a +Merkle-patricia tree with a single authority determining the current total +state of that tree, with the continually changing root of Bob’s Merkle-patricia +tree signed by Bob using his secret key which is kept in a BIP39 +style wallet. + +The metadata in the feed sharing reveals what network addresses are +following a feed, but the keys are derived from user identity keys by a one +way hash, so are not easily linked to who is posting in the feed. + +This handles public posts. + +## Private messaging + +Private messaging is trivial. There is no end of excellent existing software +to do it, in particular Jitsi Meet, Element, and, most secure of all +Bitmessage. The hard problem is the public social net on which people +meet so that they can then engage in private messaging and form private +rooms. Our social network’s private rooms are not going to be competitive +with the innumerable excellent existing systems supporting private conversations and private rooms, except that we need to provide efficient, +convenient, and secure means to launch private rooms and private +conversations from our public social network, and, most importantly, +because programmers need to be paid, we need to support private +conversations about money and payments, which existing private +messaging systems do not provide. You cannot securely embed a private +payment in a private message in existing private messaging systems. + +The most private solution for video conferencing is Jitsi Meet *if* you run +your own Jitsi Meet server. If you are using someone else’s Jitsi Meet your +messages are still private, but the someone else owns your metadata. A +more private and far more convenient solution for text and voice is +Element, and for text alone, the most secure is Bitmessage, which leaks no +metadata. + +The highest priority for a crypto currency should be to re-open free public +discussion. (Since we are, at the moment, out of power, we are temporarily +free speech enthusiasts) But the second highest priority, and the one that +will get us money, is re-opening the path to entrepreneurship, thus we +should primarily be interested in freedom of speech about money and +transactions. The enemy is shutting down forums where people can +discuss unregulated crypto currency transactions. + +Since we are going to need a BIP39 style wallet, need to put a crypto +currency in it, and the ability to have private conversations, which will +typically be about orders, payments, and receipts. I just received over +email in the clear a payment acknowledgment containing most of my +credit card number and the three digit shared secret authorization on the +back of my credit card. This is intolerably bad. Similar things are +happening with crypto currency payments, linking crypto currency wallets +to physical locations such that the owner of the wallet can be found and +shaken down, because there is no wallet infrastructure for communicating +metadata about payments, so all the metadata, such as “deliver these goods +to this physical address” goes over public networks. + +Jitsi is great for private human conversations about human things, but we +lack an environment useful for conversations about orders, payments, and +receipts, which is going to need integration with the payments and +accounting system, and needs to be based on BIP39 identities. Jitsi Meet +XMPP identities are not really useful for the purpose. We are at present +using SSL identities based on the Domain Name System for this purpose, +hence the grossly inappropriate email I received, and these are +unsatisfactory, because there are a thousand certificate authorities, and any +organization that has a certificate authority in its pocket can intercept and +interfere with a supposedly private conversation. + +This, rather than blockchain analysis, is the big problem with crypto +currencies that needs to be fixed. + +Hence we need human readable messages that can have crypto coins, +unspent transaction outputs, and lightning network payments embedded in +them. + +A conversation between two people is an encrypted immutable +authenticated but unsigned data structure shared between two parties, + +## Private rooms + +There is plenty of excellent software supporting private messaging and +private rooms. What is lacking is a public social network where it is safe +to have conversations that would give people a reason to to move to a +private room. + +The infrastructure proposed in [Anonymous Multi-Hop Locks] for lightning +network transactions is also private room infrastructure, so we should +implement private rooms on that model. + +In order to do money over a private room, you need a `reliable broadcast channel`, +so that Bob cannot organize a private room with Ann and Carol, +and make Ann see one three party transaction, and Carol see a different +three party transaction. + +In this context, a broadcast channel is reliable if each of the participants +can know that all the other participants saw the message, or knows that the +room crashed and the conversation failed to complete. + +For most purposes, this is seldom a concern, and existing private room +software fails to address this concern, which is hard to fix without +increasing the likelihood of leaking metadata. But if we want to get paid, +need to address this concern, possibly at the expense of the usability and +security for other uses of our private rooms. + +Existing software does not implement reliable broadcast in private rooms, +and is generally used in ways and for purposes that do not need it. And +anything that does implement reliable broadcast channel is apt to leak +more metadata than software that does not implement it. So for most +existing uses, existing software is better than this software will be. But to +do business securely and privately over the internet, you need a reliable +broadcast channel + +Well then, as many dags as there are groups, and if someone is added, the +adding party adds a reply link in the data structure of the smaller group to +the larger group, and if someone is deleted, a subset of the parties to large +group get invited to the smaller group, which contains a reply-to link into +the larger group. Every conversation is a room, and if you buddy someone, +he has authority to invite you to a room. The bilateral shared secrets that +make a room possible are frequently discarded and replaced. They are not +part of the immutable data structure, but infrastructure used to +communicate what is in the room. + +Each room corresponds to a dag, a narrowly shared and transient +blockchain, so that all participants know that all participants are seeing the +same conversation. But once the shared secrets have been discarded, the +room can no longer get new messages, and its data can no longer be +decrypted. The conversation is immutable but deletable. + +Private rooms are recreated from time to time, with new transient keys, +and the old transient keys lost forever. The participants may have a record +of what was said in the old private room, but the old conversation is not +present in the new room nor accessible from the new room, whereas a feed +is around forever, and its past rendered immutable by the Merkle-patricia +tree. + +When you join a room, you submit a transient public key to each of the +participants in the room, and they respond with a transient public key, +constructing a shared secret between each pair of participants. + +You then use symmetric encryption with that shared secret. And then, over +the encrypted connection, set up a shared secret based on both the durable +public and private keys, and the transient keys. So you now have, for each +pair of participants, a shared secret that depends on both parties durable +and transient keys. Possession of the shared secret proves you know the +secret key corresponding to your public key, and you get perfect forward +secrecy because the shared secret and the transient keys are abandoned +when there is no further action in the private room. + +Assume Bob’s secret key is $b$, and his public key is $B$, and similarly +Carol’s secret key is $c$ and her public key is $C$. + +[Scriptless Scripts]:https://tlu.tarilabs.com/cryptography/scriptless-scripts/introduction-to-scriptless-scripts.html +"Introduction to Scriptless Scripts – Tari Labs University" + +[Anonymous Multi-Hop Locks]: anonymous_multihop_locks_lightning_network.pdf +"Anonymous Multi-Hop Locks for Blockchain Scalability and Interoperability" + +In order to support [Scriptless Scripts] and [Anonymous Multi-Hop Locks] +the private keys need to be scalars of an elliptic curve of prime order, such +as Ristretto25519, and the public keys need to be elliptic points on that +curve. + +$b×C=c×B$. (Lower case letters denote scalars modulo the order of +the curve, upper case denote the corresponding elliptic points on the +curve) So, to construct a shared secret that can be used for symmetric +encryption, Bob calculates $b×C$ and Carol calculates $c×B$ + +Suppose that Bob’s transient private and public keys are $b_t$ and $B_t$, and +his durable private and public keys are $b_d$ and $B_d$, and similarly for +Carol. + +Then to construct a shared secret that depends on both the transient and +the durable keys, and proves knowledge of the durable private key, Bob +calculates $(b_d + b_t)×(C_d + C_t)$, and similarly Carol calculates +$(c_d + c_t)×(B_d + B_t)$. (But this is only safe if both Bob and Carol +have first proven that they also know the shared secret $b_t×C_t = c_t×B_t$, +by first successfully decrypting and encrypting stuff symmetrically +encrypted to that secret) + +A bilateral pair conversation is a pair of feeds that should have identical +content and hashes, and though they are not signed, they are authenticated. + +A multilateral conversation is n bilateral conversations, n feeds. + +For money matters, where the point is proving to third parties, we want +signatures on the hashes and immutable messages, but for bilateral +conversations we want authentication without signing + +So the multilateral shared room is constructed through people +communicating over bilateral shared secrets constructed from unilateral +unshared secrets. The signed contents of the room are flood filled around +the participants, rather than each participant being required to communicate +his content to each of the others. + +To provide a reliable broadcast channel, a hash of the total state of the room is continually constructed + +# signing and immutability + +For money, we want signatures, and money is the highest priority. Carol +signs what she said to Bob, and she is stuck with it. Less cases to handle. +When you send money, you not only sign it, thus rendering it immutable +but deletable, you sign the message, thus signing all the previously +mutable conversations linked in through the “reply-to” and “regarding” +links in your message. The preceding conversation was authenticated but +unsigned, thus potentially mutable and deniable, but becomes signed by +and one party, and thus immutable, when he sends money, and signed by +the other party when he accepts the money. + +[Zooko’s triangle]: ./zookos_triangle.html + +A message with a single recipient is always authenticated, though possibly +only by a cheap readily discardable [Zooko’s triangle] identity. If it is about +money, normally signed, and thus immutable but discardable. + +A message with two or more recipients is always signed, regardless of +whether it has money or not, because there are too many cases to handle +otherwise, and thus immutable though deletable, and it retains perfect +forward secrecy, in that if one of the recipients loses control of his durable +secret keys, but all three participants keep their mouths shut, no one else +can know what was in it. It also renders any bilateral messages linked in +immutable, because they are linked by two fifty six bit hash, thus signing a +message acknowledges sending or receiving all messages linked to by the +signed message and renders them immutable by linking them into a +hashdag. + +# Name System + +We need an unbreakable and untraceable name system like that Jitsi or +namecoin, but Jitsi relies on the Ether blockchain, which is rather too +breakable. and in the hands of the Social Justice Warriors. Has to rely on +crypto wallet type identity keys, rather than some authority mapping names. + +You need the functional equivalent of blogs, where everything published and +every approved comment on a blog gets shared BitTorrent/Usenet style – so you +and everyone has a copy on their hard disk of everything and everyone they are +following, and shares that data with everyone who follows the same entity, +with no central server. + +And you need the functional equivalent of reposts and likes, so that if +someone you are following endorses what someone said, it eventually gets +downloaded, BitTorrent style, on your disk, and you get to see it, and comment +on it, also. + +One minor difference you will need is that the person who issued the thing +being commented on gets to moderate the comments, though obviously this will +have no effect on comments by anyone you are already following. + +A repost by someone you are following should go into your feed. His likes +should put links in your feed. Comments on the thing reposted should put +links in your feed if they survive moderation, and the comment should go into +your feed whether moderated or not if you are already following the commenter. + +You need private messages, which are easy, and we have lots of systems to +securely provide that already. + +You need private groups that no outsiders can follow, except the moderator +regularly sends them the latest keys. + +# Blockchains + +You need a cryptocurrency like Bitcoin, but Bitcoin is too readily traceable. +Wasabi wallet has a cure for bitcoin traceability, but it is not easy to use +that Wasabi capability, and most people will not, despite the very nice user +interface. The lightning network over bitcoin could fix those problems, and +also overcome Bitcoin’s scaling limit, but I don’t think much of the current +implementation of the lightning network. The latest bitcoin upgrade, +supporting Schnorr signatures and smart contracts, makes a true lightning network possible, but backwards compatibility and government pressure may prevent that from happening. + +People don’t actually understand fiat money, but are plenty comfortable +when their browser tells them they have such and such an amount in their +bank. + +And they don’t understand the difference between their wallet telling them +that secrets sitting in their wallet are worth such and such an amount, and +Coinbase telling them that secrets held by Coinbase, supposedly on their +behalf, are worth such and such an amount. They do not understand this +big important difference. But that is going to be OK if we get the user +interface done right. + +A major obstacle to crypto currency taking over the world is that making +payments in crypto currency is hard. + +[BTCPay] solves the problem of enabling your internet store to receive +payments directly to your own wallet, and integrates with your database, +automating the bookkeeping and stock keeping + +[BTCPay]:https://docs.btcpayserver.org/WooCommerce/ +"BTCPay/WooCommerce integration" + +But though it is easy on internet shopkeepers, it is hard on internet +shoppers. The fundamental user interface problem is inherent in all +existing wallets. + +You navigate through a nice easy shop page, choose some goods or +services, and proceed to a BTCPay page that is well integrated with the +shop, and the BTCPay page gives you a receive address and an amount, +presented in the context of your purchase. And then you have to launch +your wallet, and copy and paste that amount and that address into your +wallet, and tell it to pay, and when you are doing that, you are no longer +in the context of your purchase and your interaction with the internet +shopkeeper. Your wallet does not know anything about the metadata +associated with the transaction. Which makes it hard, and makes the +records kept by your wallet not very useful. The wallet of the seller is +integrated into their book keeping by BTCPay, but the buyer's wallet +cannot be integrated into the buyer's bookkeeping. + +The complexity and difficulty is because that we are using one rather +insecure channel for the metadata about the transaction, and a different +channel for the transaction. (This is also a gaping and gigantic security +hole, even if you are using Monaro.) For a truly usable crypto currency +payment mechanism, transactions and metadata have to go through the +same channel, which means human to human communication through +wallets - means that you need to be able to send human to human +messages containing requests for money, *and containing money*. + +We can make the client wallets easier to use, and third worlders are +already using payment methods that are not that easy when they have to do +transactions over distance. + +The internet tends to be unreliable in third world, and people will do +their interneting whenever it available, which is often at strange hours. + +Everyone who matters in the third world sends and receives text messages. +We want to be able to implement a wallet that everyone in third world can +use, can chat to friends and people they do business with, send money to +them, and receive money from them. + +If the wallet integrates an identity and messaging system, then making +payments and receiving payments over the wallet can be made easier than +with any existing system. + +We have to put the medium for communication about money, for +communicating metadata about transactions, inside the wallet, as in the old +days you sent a sealed envelope containing both a cheque and a letter. +Then everyone is going to use that wrapper, owning their secrets the way +they used to own their correspondence. A third party like Coinbase will +not have a handle to get inside the transaction. + +Private messages need to be able to carry the cryptocurrency embedded in +them, so that the social net is useful for business, just as you can mail a +check in an envelope with a document. So the lightning network needs to +be an application, the primary application, of the private messaging and +private room system. + +The elegant cryptography of [Scriptless Scripts] using adaptive Schnorr +signatures and of [Anonymous Multi-Hop Locks] assumes and presupposes +a lightning network that has a private room with a reliable broadcast +channel (everyone in the room can know that everyone else in the room +sees the same conversation he is seeing, and a reliable anonymous +broadcast channel within the private room. "We also assume the existence +of a functionality $F_{anon}$ which provides user with an anonymous +communication channel." + +So, before we can implement a reliable anonymizing lightning network +using the elegant cryptography of these papers, have to implement a +corresponding social media capability, even though that capability is +unlikely to be competitive with Jitsi and others for purely social purposes. +Whereupon when finally we do implement a lightning network, it will +work like paper payments sealed in envelopes, which are generally +accompanied in that envelope by metadata about the human purpose and +intent of that payment and the obligations that ensue, unlike the present +lightning networks, where such metadata goes over separate and insecure +channels. + +Business will move to the decentralized social net to escape the +increasingly disruptive social controls that are sending so many businesses +down the drain. Nasa lost the ability to build rockets because of Social +Justice (The SLS is built around rocket parts built long ago that Nasa can no +longer make). Intel lost the ability to build CPUs because of social +justice. (They now rely on a Taiwanese owned and operated chip fab), and +Disney destroyed to Star Wars franchise, turning it into a lecture on social +justice. Debian broke Gnome3 and cannot fix it because of social justice. + +Business needs a currency and [book] keeping system that enables them to +operate a business instead of a social justice crusade. + +A blockchain is just a public ledger with an immense number of columns. +Triple entry [book] keeping with immutable journal entries is a narrowly +shared ledger with a considerably smaller number of columns. Every +business needs its books on its own blockchain, to escape government +regulation of its book keeping, which public regulation tends to result in +the creation of holiness being recorded as the creation of value, as in the +Great Minority Mortgage Meltdown and MF Global. + +The disruptive interference of Human Resources in hiring and firing +shows up on the books as a profit centre instead of a cost centre, and the +legal and accounting departments work for regulators and the taxman +similarly. + +# Monetization + +Information wants to be free, but programmers want to be paid. + +An environment created to allow people to have uncensored public and +private conversations can make its creators rich if it allows businesses to +focus on creating value and making profit. + +[sox]:sox_accounting.html +"Sarbanes-Oxley accounting" + +A proof of stake currency works like a startup company used to work +before [SOX] -- the founders get shares, then they sell or issue shares to +angel investors, and then with the angel investors money they pay early +developers with both shares and fiat. + +And then in your liquidity event, these shares would become liquid (back +in those days traded on the public stock exchange), and, if the startup was +successful, everyone gets rich. + +But since [SOX] blew up that institutional and organizational structure, we +have to recreate it, before we get paid. + +The intent is to recreate the organizational form that SOX destroyed, and +recreate it for ourselves before we recreate it for anyone else. + +Its blockchain unit of value will be adopted by businesses and private +individuals if it allows them to message each other securely about +transactions, and if it allows businesses to keep their [book]s in ways that +reflect a focus on the creation of value and profit. + +At the time I write this there is an immense amount of money in play on +Taiwan Semiconductor. All the modern semiconductor fabs in the west +have died, because Human Resources forces them to hire people into +critical high tech positions on the basis of race, sex, and sexual preference, +except for one company that is privately held by the King of Dubai, who +being a foreign King with his own loyal army is in a better position than +the average boss to resist political pressure. It is easier to resist demands +from Human Resources when you have men at your back who will kill +and die for you. + +There are a lot of high tech tasks where if you have a single dud member +on the team, the whole team will fail – and if you have a woman on the +team, and exclude her from the stuff she is going to wreck, you are guilty +of mansplaining, sexual harassment, and probably rape. There will be a +big emotional scene with so much of the drama that women so love, and +even though will be absolutely obvious that the woman is disruptively +manufacturing drama, no one will admit to noticing her bad behaviour. + +Censorship resistant social media is part of the necessary infrastructure +for moron resistant high tech teams, though additional infrastructure will +be needed. + +[sovereign]:white_paper.html#sovereign-corporation + +To maintain technology, and for business to create value, the west is going +to have to create companies that can resist political pressure. Which +means that we are going to have to create software that enables companies,as +well as discussion groups, that can resist political pressure, their books +being triple entry [book]s with immutable entries, rather than [Sox], and +their shares traded on the blockchain instead of the quasi governmental +stock exchanges, sovereign corporations rather than state created +corporations. The software will be open source, so the programmers +cannot make money off it directly, but can make money from the +seigniorage and implementing the lightning network where such shares +will be traded in a way that gives the developers seigniorage. + +[triple entry accounting]:./triple_entry_accounting.html +"triple entry accounting" + +[book]:./triple_entry_accounting.html +"triple entry accounting" + +Software that enables businesses that can resist political pressure is a +superset of software than enables discussion groups that can resist political +pressure. We start by enabling discussion groups, which will be an +unprofitable exercise, but the big money is in enabling business. +Discussion groups are a necessary first step. + +# Development sequence + +First we need a replacement for Quic, TCP, and SSL. Quic and SSL encryption +is joined at the hip to the quasi governmental domain name system, and if +we put our custom encryption and name system on top of TCP, it is +inefficient, as SSL without Quic is inefficient, and will be dependent on +big quasi governmental organizations to protect against distributed denial of +service attacks, which will start up as soon as the social media has +important discussions, and redouble when the project starts to develop +serious monetary value. + +But such a communications protocol will have no value, until there is +software that uses it to communicate, no value till there is a social media +system on top of it. + +In order to communicate, it will need network addresses, and a system to +associate durable public keys with frequently changing network addresses, +though these should not have any direct one to one relationship with +durable public keys of humans, and do not always need to be very durable. + +Initially we just use the equivalent of a hosts file, though as SQLite +rather than text, but this does not scale to anything useful. For scaling, +when one computer connects to another, each learns the durable public keys +and most recent network addresses of those public keys from the other, and +if one has connections to several others, it can perform nat penetration to +allow direct connection, after the model used by Jitsi, where all +participants have a durable connection to a Jitsi meet server, which +enables them to form direct connections with each other. Though there is +no reason why every computer should not be able to perform that role +rather than just the central server as in Jitsi meet. + +If a participant does not sign the network address proposed for it by its +counterparty, perhaps because it thinks it is incorrect, perhaps because it +thinks it likely to be far from durable, or for privacy reasons, the network +address does not get propagated through network address pooling. So only +participants with stable network addresses and accessible ports get +propagated, get the relationship between durable key and network address +widely known, so if the network address is unstable, or no accessible port, +you can only contact directly if both connected to same third computer. +Which drastically limits the usefulness of this protocol in general, but it +works fine for propagating message pools, Usenet/BitTorrent style. At some +time in the future we will address nameservers, with will serve names that are +recorded on the blockchain, but at this point, we do not yet have a +blockchain, and without a domain name system for network addresses and +nameservers, we cannot do much with the protocol. And we cannot use the +quasi state domain name system, or it is going to be yanked out +underneath us as soon as we succeed. + +But we can do the important things, which are social media and blockchain. + +With social networking on top of this protocol, we can then do blockchain +and crypto currency. We then do trades between crypto currencies on the +blockchain, bypassing the regulated quasi state exchanges, which trades +are safe provided a majority of the stake of peers on the blockchain that is +held by peers holding two peer wallets, one in each crypto currency being +exchanged, are honest. + +This, a crypto currency and the capability to across blockchain exchanges +on our blockdag, gives us our liquidity event, enabling investors to fund +software. At which point we start plugging all the wonderful things we are +going to do to make the currency actually useful in the near future. Which +liquidity event will be considerably more persuasive if the system actually +is useful as a censorship resistant social media platform at the time of the +liquidity event. + +A proof of stake blockdag is a [sovereign] corporation, but in order to +actually function as a corporation it needs a human chief executive, +corporate funds that the the chief executive can dispense, a human board +of directors to keep an eye on what the chief executive is doing with those +funds, and proper [book]s so that the board can keep an eye on the chief +executive, and the shareholders keep an eye on the board. + +So, after the first liquidity event (cross blockchain trades on the blockdag) +we implement the standard corporate infrastructure, accounting, board of +directors, CEO, over proof of stake instead of under a quasi governmental +stock exchange, [triple entry accounting] with immutable journal entries +instead of [sox] accounting, for the final liquidity event, the +final liquidity event being the first [sovereign] corporation on this +blockchain. + +Information wants to be free, but programmers need to be paid. We want +the currency, the blockdag, to be able to function as a corporation so that it +can pay the developers to improve the software in ways likely to add value +to the stake. + +# Many sovereign corporations on the blockchain + +## source of corporateness + +State incorporated corporations derive their corporateness from the +authority of the sovereign, but a proof of stake currency derives its +corporateness from the cryptographically discovered consensus that gives +each stakeholder incentive to go along with the cryptographically +discovered consensus because everyone else is going with the consensus, +each stakeholder playing by the rules because all the other stakeholders +play by those rules. + +Such a corporation is sovereign. + +## shares, bookkeeping, and sidechains + +We then implement sidechains, so that many sovereign corporations will +have some place to trade their shares and to keep the narrowly shared +immutable journals for their triple entry [book]s. The shares will be on a +broadly shared blockchain, but some aspects of the books may well be +very narrowly shared and selectively revealed. A total order of a sidechain +will be reflected in a total state of the main chain, through the total state of +peers that are also managing side chain wallets, or who have clients (or +clients of clients) and that are managing side chain wallets, and each +client’s total order is Merkle linked into the client’s total state, which is +Merkle linked into the peer’s total state, so that the main block chain +contains a Merkle link that can prove that machines operating a sidechain +are in agreement about a total order for transactions that happened a little +time ago. The total order of transactions on a sidechain will be the order in +which they get linked into the mainchain. + +If a peer on a sidechain is a client of a client of a client of a peer on the +mainchain, and adds some new transactions to its sidechain wallet, the +total order of those transactions within the sidechain is the order in which +the main chain gives that total state of the mainchain peer, and within that +order the order that the peer gives the total state of the client, and within +that order … + +This does not occupy mainchain space, because a single two fifty six bit +hash on the mainchain can represent to the total state and total order of +many very large sidechains, with the very large preimage of the hash being +known to peers on the sidechain, but not to peers on the mainchain unless +they are also peers on the sidechain. + +A two fifty six bit hash gives unlimited compression, which is lossy in the +sense that it is one way compression, but lossless in that if you know the +preimage, you can always verify that the hash must have been made from +that preimage, and a hash that is the root of a Merkle tree of two fifty +six bit hashes gives the same, but with the added benefit that one can generate +a short proof that some part of the preimage, possibly a very small part of +a vast preimage, a part so small that it fits in single network packet, but +part of a preimage so vast that no one possesses all of it, that no one could +possess all of it, was part of the data from which that hash was generated. + +The mainchain is a Merkle dag, structured in a way that makes it possible +to give a single total order for the gossip vertexes of which it is composed, +and every gossip vertex in the dag is a Merkle tree that reflects, among +other things, a total state of two peers on the blockchain, each total state of +a peer being incorporated in one and only one gossip event, and every total +state of a peer contains Merkle tree that reflects the prior state of the +blockchain, and the state of its clients who want their current state +committed to the blockchain, and thus whenever one party wants to prove +to another party that he is telling the same story to all the parties involved +(has not engaged in Byzantine failure) he can generate a short proof rooted +in the mainchain. Thus shares and triple entry books will be in the +mainchain, even though much of the information in the books is going to +be private and narrowly shared. Whenever one man’s computer presents +something from the [book]s to another man’s computer, it will present a +short proof rooted in the blockchain, rooted in the same hash that everyone +can see and vast numbers of people do see, that it is telling the same story +to anyone it tells, even if only very few people should be told. + +What should happen is that every time you make a payment to a business, +you get a receipt that is connected by a Merkle chain to a Merkle tree of +the business’s narrowly shared books, which is connected by a Merkle +chain to the broadly shared main blockchain. + +[Sox] bookkeeping relies on the books being widely shared among +supposedly respectable and highly regulated people, which does not help +you much if, as in the Great Minority Mortgage Meltdown, the regulators +are engaged in evil deeds, or if, as with Enron and MF Global, the +accountants are all in the pay of powerful men engaged in evil deeds. +Triple entry [book]keeping with immutable journal entries works in a low +trust world of badly behaved elites, works in the circumstances now +prevailing, and, unlike Sox accounting, it does not require wide sharing of +the books. + +## Corporate cohesion + +The corporation exists by [book]keeping, which enables the shareholders to +keep an eye on the board, and the board to keep an eye on the CEO, and our +current system of bookkeeping is failing for lack of trust and honour. + +Corporate cohesion is hard and apt to be fragile. You are apt to have board +members that want to participate in executive decisions, want to tell the +CEO what to do instead of merely evaluating how well he is doing it and +making sure he is not stealing from the shareholders, and executives that +have a power base external to the company. And telework and video +conferencing does not make this any easier. + +But while the purchasing officer who is too pally with certain suppliers is +always a natural and inevitable problem, and your lawyer apt to be too +well connected to the judge and the other guy’s lawyers, the big problem +today with state created corporations is that the meddling of the state gives +HR, accounting, and the legal department power bases external to the +company, undermining corporate cohesion. If HR does not get its way, it is +apt to organize lawsuits against the CEO, board, and shareholders. + +Things can easily fall apart, and a cohesive group of people within the +company is apt to push on those fault lines. + +Corporate cohesion is hard, but the regulatory state is making it harder. +Sovereignty will improve the capability of corporations to cohere, because +the CEO really will be able to fire anyone, the board really able to fire the +CEO, and shareholders really able to fire the board. + +And, with the corporate [book]s better connected to reality, the shareholders +will have better information on whether to do so. A necessary precondition +for corporations to function at all is that we fix bookkeeping to work in a +world of untrustworthy elites. diff --git a/docs/sox_accounting.md b/docs/sox_accounting.md new file mode 100644 index 0000000..e0d4761 --- /dev/null +++ b/docs/sox_accounting.md @@ -0,0 +1,143 @@ +--- +title: >- + Sox Accounting +--- +Accounting and bookkeeping is failing in an increasingly low trust world +of ever less trusting and ever less untrustworthy elites, and Sarbanes-Oxley +accounting (Sox) is an evil product of this evil failure. + +Enron was a ponzi scam that pretended to be in the business of buying and +selling. They would sell stuff cheap for cash on the barrelhead, then buy it +dear on credit, and have the seller deliver the stuff to the buyer, without +themselves ever going near the stuff, or knowing anything about the stuff. + +And, as ponzi scams tend to do, need to do, they grew and grew and grew, +resulting many excited investors buying Enron shares. + +And some people wondered, “where are these mighty profits coming from. How +can you create value without ever getting anywhere near anything that actually +has value?”. To which Enron replied “financial wizardry”, which was, in a +sense, true. + +But Enron kept books. + +Enron hired the very best and most respectable accountants to go over its +books with the greatest of thoroughness, and paid them very well indeed +to do so. + +The very respectable accountants turned somersaults to avoid seeing evil. + +Some suspicious investors hired a bunch of accounting students, who were +not employed by a respectable accounting firm, nor subject to any +respectable authority of respectable accountants properly established in the +respectable accounting profession, to go over Enron’s books with a fine +tooth comb. + +It quickly became obvious to these disreputable accounting students that +Enron was a ponzi scheme, that the books were the books of ponzi +scheme, that the books were true though misleading, and truly reported a +ponzi scheme losing money hand over fist. + +The investors made the results of this examination of Enron’s books +public. + +When the whole thing blew up, a great many discussions came to light +which amounted to Enron asking the very respectable accountants how to +help the very respectable accountants refrain from seeing what they were +seeing, the very respectable accountants asking each other and other very +respectable accountants how to not see what they were seeing, and the +very respectable accountants then telling Enron how protect the very +respectable accountants from seeing what they were seeing. + +And suddenly people stopped being willing to pay Enron cash on the +barrelhead for goods, suddenly stopped being willing to sell Enron goods +on credit. Suddenly Enron could no longer pay its employees, nor its +landlord. Its employees stopped turning up, its landlord chucked their +furniture out into the street. + +Problem solved. + +If only it had stayed solved, but much bigger trouble was in store. + +Suddenly people started not respecting very respectable accountants. +Suddenly people started not trusting books kept by very respectable +accountants for very respectable businesses. + +The accountants rushed to the government, and said “Accounting is +collapsing. Business is in chaos. You must *force* people to respect us. (And +also regulate the hell out of us to ~~make our crimes legal~~ ensure that none +of us commit these crimes again)” + +Investors who had purchased Enron’s shares, and sellers who had sold Enron +stuff on credit, rushed to the government and said “Accountants are +untrustworthy, businesses are untrustworthy. We have a crisis of trust. You +must regulate the hell out of them to *force* them to be trustworthy.” + +Of course you cannot force people to be trustworthy. You can, however force +people to act as if they trusted people that they do not in fact trust. +Which prevents people from freely associating with trustworthy people +and from refusing to associate with untrustworthy people, the crisis being +in large part that people were disinclined to associate with very respectable +accountants, and were instead inclined to associate with the thoroughly +disreputable accounting students and the investors who had hired them. + +So there was in practice a great deal of overlap between very respectable +accountants asking government to force business to use their services and +naive idiots asking government to force business and accountants to +behave in a more trustworthy manner. + +And so, we got a huge bundle of accounting regulation, Sarbanes-Oxley. + +Sox required publicly traded companies to keep their books in accordance +with certain rules, and has in fact successfully prevented any publicly +traded companies from doing an Enron in the exact same way that Enron +did it, though what MF Global did was not hugely different. + +But it directly contributed the Great Minority Mortgage Meltdown. The +purpose of book keeping is to track the movement and creation of value. +Accounting is the art and science of making sense out of the books, and +making sure the books make sense. In the aftermath of the Great Minority +Mortgage Meltdown, a whole lot of entities went belly up, and when the +creditors went through the books, they found all the mortgages the entities +supposedly owned, but were frequently unable to find the people +supposedly responsible for paying these mortgages, or the properties that +were supposedly security for these mortgages. + +Sox books just tend to not track value very well, being designed for a +different, more complex, and difficult to define, task. Sox books tend to +track official reality and disregard the reality of the ground beneath your +feet, the things in your hands, and what is before your eyes, with the result +that they tend to track the creation of holiness and ritual purity, rather than +the creation of value. + +But the biggest problem with Sarbanes-Oxley is not that it has failed to +force publicly traded companies and their accountants to act in +a trustworthy manner, but that it has forced them to act in an untrustworthy +manner. + +It is not only very difficult and enormously expensive to comply with Sox. +It is also *impossible* to comply with Sox. What publicly traded companies +do instead is hire an accounting firm so well connected to the regulators +that when it blesses the books it has prepared as Sox compliant, the +regulators will pretend to believe. Which is great for the very respectable +accountants, who get paid a great deal of money, and great for the +regulators, who get paid off, but is terrible for businesses who pay a great +deal of money and do not in fact get books that accurately tell +management how the business is doing, and considerably worse for +startups trying to go public, since the potential investors know that the +books do not accurately tell the investors how the business is doing. + +What established businesses do instead is prepare one set of books for +Sox compliance, and another set of books for management that do not comply +with Sox but which actually do reflect the movement and creation of +value, but a startup is not allowed to tell potential investors about the +real books that actually reflect the movement and creation of value for the +purpose of an IPO. + +Which has killed off startups and IPOs, which used to be my bread and +butter. + +To do a startup again, we are going to have to do it on a blockdag. + +(Blockdags being the latest advance in blockchains, an unbalanced Merkle +dag instead of an unblanced Merkle tree.) diff --git a/docs/stirling_nemes_approx_to_gamma_function.xls b/docs/stirling_nemes_approx_to_gamma_function.xls new file mode 100644 index 0000000..96e5602 Binary files /dev/null and b/docs/stirling_nemes_approx_to_gamma_function.xls differ diff --git a/docs/sybil_attack.md b/docs/sybil_attack.md new file mode 100644 index 0000000..1c0f6ef --- /dev/null +++ b/docs/sybil_attack.md @@ -0,0 +1,112 @@ +--- +title: Sybil Attack +# katex +--- +We are able to pretend that bank transactions are instant, because they are +reversible. And then they get reversed for the wrong reasons, often very bad +reasons. + +But in general, with irreversible transactions, hard to pretend they go +through in an instant. Bitcoin times are pretty good, but not good enough to +pay at the cash register. + +For this reason, all rhocoin transactions are made with an identity. You can +have as many identities as you please, and these identities are not visible +to third parties unless one of the parties to the transaction chooses to make +them visible. + +To support transactions at the cash register, going to need credit rating +agencies. If you make the payment as an identity that has an adequate credit +rating, the cash register trusts that the transaction will go through. + +So, how do we stop someone from generating a credit rating by issuing a lot +of sham transactions from his right hand to his left hand, and then buying +something and not paying? The credit card problem. + +And how do we stop someone from generating a pile of fake sales, and then +making a real sale and not delivering? The Ebay problem. + +There is a solution to this problem. Each rating agency will treat payments +and reviews as a directed graph, and measure graph connectivity. each +subgraph of fake payments, and fake reviews will have good connectivity to +itself and poor connectivity to the rest of the graph. + +# SybilGuard + +The [SybilGuard paper] provides graph theory solutions against sybils. + +[SybilGuard paper]:http://www.math.cmu.edu/~adf/research/SybilGuard.pdf + +According to SybilGuard, social networks are empirically determined to be +“fast mixing” + +“social networks tend to be fast mixing (defined in the next section), which +necessarily means that subsets of honest nodes have good connectivity to the +rest of the social network” + +So the legitimate musician with legitimate fans will be connected to the rest +of the social network – his legitimate fans will have interactions with +legitimate fans of other legitimate musicians, and will purchase stuff from +other legitimate musicians, while the fake fans making fake purchases will +not. + +“in a fast mixing graph, after a small number of hops a random walk is +equally likely to be traversing any edge in a given hop.” + +The paper argues that Sybil subsets of the network are substantially less +likely to be entered after a small number of hops. + +“With a length $w$ random walk, clearly the distribution of the ending point +of the walk depends on the starting point. However, for connected and +nonbipartite graphs, the ending point distribution becomes independent of the +starting point when $w → ∞$. This distribution is called the stationary +distribution of the graph. The mixing time T of a graph quantifies how fast +the ending point of a random walk approach the stationary distribution. In +other words, after $Θ(T)$ steps, the node on the random walk becomes roughly +independent of the starting point. If $T = Θ(log(n))$, the graph is called +fast mixing.” + +$Θ(f)$ means that the space or running time is always proportional to $f$, +$O(f)$ means that the common worst case space or running time is proportional +to $f$. Engineers generally do not care about such differences, and just use +$O$ everywhere. + +If we mix the graph for a suitable time, legitimate nodes will be well mixed +with each other, and Sybil nodes will not be well mixed with legitimate nodes. + +Legitimate nodes will have positions that are close to each other in the +space of distributions of end probabilities, and Sybil nodes will be far from +legitimate nodes in that space. + +This algorithm is $O(n^2)$ in data size (the distribution for each starting +point), and $Θ(n^2{log(n)})$ in computational time, which is a lot better +than non polynomial, though still apt to be inconveniently large. It can +probably be speeded up considerably by [random dimensional reduction], +wherein a vector of very large dimension (the probability of reaching each +end point, the distribution of end points) is randomly mapped to a vector of +dimension $Θ(log(n))$ log of the original dimension. + +[random dimensional reduction]:recognizing_categories_and_instances_as_members_of_a_category.html#dimensional-reduction +"recognizing instances as members of a category" + +In this case, the vector space of large dimension is of vectors whose +elements are the probability of arriving at a given node after a random walk +of distance $w$, “the distribution of the ending point of the walk” + +Which will, I think, speed it up to $O(n×{{(log (n))}^2})$ which is +acceptable. + +Mixing for a suitable time, a suitable path distance, divides the graph into +well mixed regions, one of which is the legitimate nodes. We identify which +of the regions is non Sybil by checking the region of a small number of known +legitimate nodes. If they all belong to the same region, our mixing distance +is sufficient, and chances are the Sybil nodes are not in that region. + +During mixing, the points draw closer to each other in the space of +distributions. The fan group of a particular musician will coalesce rapidly, +and then the fan groups of all legitimate coalitions will, more slowly draw +together, while the fake fan group drifts very slowly indeed. We halt the +mixing when the known legitimate nodes are all close to each other, because +by then everyone who is a legitimate node is close to them. + +We then downrank reviews or page ranks that are distant from the coalescence. diff --git a/docs/tealdeer.gif b/docs/tealdeer.gif new file mode 100644 index 0000000..5dbce5c Binary files /dev/null and b/docs/tealdeer.gif differ diff --git a/docs/threshold_shnorr.pdf b/docs/threshold_shnorr.pdf new file mode 100644 index 0000000..a5cd95f Binary files /dev/null and b/docs/threshold_shnorr.pdf differ diff --git a/docs/triple_entry_accounting-bitcoin_mag.url b/docs/triple_entry_accounting-bitcoin_mag.url new file mode 100644 index 0000000..c43de67 --- /dev/null +++ b/docs/triple_entry_accounting-bitcoin_mag.url @@ -0,0 +1,2 @@ +[InternetShortcut] +URL=https://bitcoinmagazine.com/articles/triple-entry-bookkeeping-bitcoin-1392069656/ diff --git a/docs/triple_entry_accounting-by_Iang.mhtml b/docs/triple_entry_accounting-by_Iang.mhtml new file mode 100644 index 0000000..1be6d8c --- /dev/null +++ b/docs/triple_entry_accounting-by_Iang.mhtml @@ -0,0 +1,1506 @@ +From: +Snapshot-Content-Location: https://iang.org/papers/triple_entry.html#index +Subject: Triple Entry Accounting +Date: Sat, 3 Oct 2020 06:02:11 -0000 +MIME-Version: 1.0 +Content-Type: multipart/related; + type="text/html"; + boundary="----MultipartBoundary--ZJLbQ4mRwLhAHsHENVppcDUXKvpxlvNoq0H6MRFUAo----" + + +------MultipartBoundary--ZJLbQ4mRwLhAHsHENVppcDUXKvpxlvNoq0H6MRFUAo---- +Content-Type: text/html +Content-ID: +Content-Transfer-Encoding: quoted-printable +Content-Location: https://iang.org/papers/triple_entry.html#index + + +Triple Entry Accounting + + + +

+
+ Work - in - Progress +

+ + Triple Entry Accounting + + +

+ + Ian Grigg
+ Systemics, Inc.
+ +
+ 2005 +

+
+ + + +
+$Revision: 1.7 $
+$Date: 2005/12/25 23:04:21 $ +
+ +



+

Abstract: + The digitally signed receipt, an innovation from + financial cryptography, presents a challenge + to classical double entry bookkeeping. Rather + than compete, the two melded together + form a stronger system. Expanding the usage of + accounting into the wider domain of digital cash + gives 3 local entries for each of 3 roles, + the result of which I call + triple entry accounting. +

+

+ This system creates bullet proof accounting + systems for aggressive uses and users. + It not only lowers costs by delivering + reliable and supported accounting, + it makes much stronger governance possible + in a way that positively impacts on the future + needs of corporate and public accounting. +

+ +


+ + + + + + +

Introduction

+

+This paper brings together financial cryptography +innovations such as the Signed Receipt with the +standard accountancy techniques of double entry +bookkeeping. +

+ +

+The first section presents a brief backgrounder +to explain the importance of double entry +bookkeeping. It is aimed at the technologist, +and accountancy professionals may skip this. +The second section presents how the Signed +Receipt arises and why it challenges double +entry bookkeeping. +

+ +

+The third section integrates the two together +and the Conclusion attempts to predict wider +ramifications into Governance issues. +

+ +

Credits

+ +

+This paper benefitted from comments by +Graeme Burnett and Todd Boyle + [TB]. +

+ +

A Very Brief History of Accounting

+ +

+Accounting or accountancy is these days thought +to go back to the genesis of writing; +the earliest discovered texts have been +deciphered as simple lists of +the counts of animal and food stock. +The Sumerians of Mesopotamia, around +5000 years ago, used Cuneiform +or wedge shaped markings as a base-60 +number form, which we still remember as +seconds and minutes, and squared, as the +degrees in a circle. +Mathematics and writing themselves +may well have been derived from the +need to add, subtract and indeed +account for the basic assets and +stocks of early society. +

+ +

Single Entry

+ +

+Single entry bookkeeping is how 'everyone' +would do accounting: start a list, +and add in entries that describe each asset. +A more advanced arrangement would be +to create many lists. +Each list or 'book' would represent a +category, and each entry would record +a date, an amount, and perhaps a comment. +To move an asset around, one would cross it +off from one list and enter it onto to +another list. +

+ +

+Very simple, but it was a method that +was fraught with the potential for errors. +Worse, the errors could be either +accidental, and difficult to track down and +repair, or they could be fraudulent. +As each entry or each list stood alone, +there was nothing to stop a bad employee from +simply adding more to the list; even when +discovered there was nothing to say whether it +was an honest mistake, or a fraud. +

+ +

+Accounting based on single entry bookkeeping +places an important limitation on the trust +of the books. +Likely, only the owner's family or in times +long past, his slaves could be trusted with +the enterprise's books, leading to a supportive +influence on extended families or slavery as +economic enterprises. +

+ +

Double Entry

+ +

+ +Double Entry bookkeeping +adds an additional important +property to the accounting system; +that of a clear strategy to identify errors +and to remove them. +Even better, it has a side effect of +clearly firewalling errors as either +accident or fraud. +

+ +

+This property is enabled by means of three features, +being the separation of all books into two groups +or sides, called assets and liabilities, +the redundancy of the duplicative +double entries with each entry +having a match on the other side, +and the balance sheet equation, which says that +the sum of all entries on the asset side +must equal the sum of all entries on the +liabilities side. +

+ +

+A correct entry must refer to its counterparty, and +its counterpart entry must exist on the other side. +An entry in error +might have been created for perhaps fraudulent +reasons, but to be correct at the local level, +it must refer to its counterparty book. +If not, it can simply be eliminated as an incomplete +entry. +If it does refer, the existance of the other entry +can be easily confirmed, or indeed recreated depending +on the sense of it, and the loop is thus closed. +

+ +

+Previously, in single entry books, the fraudster +simply added his amount to a column of choice. +In double entry books, that amount has to come from +somewhere. If it comes from nowhere, it is eliminated +above as an accidental error, and if it comes from +somewhere in particular, that place is identified. +In this way, fraud leaves a trail; and its purpose +is revealed in the other book because the value taken +from that book must also have come from somewhere. +

+ +

+This then leads to an audit strategy. First, ensure +that all entries are complete, in that they refer to +their counterpart. Second, ensure that all movements +of value make sense. This simple strategy created a +record of transactions that permitted an accountancy +of a business, without easily hiding frauds in the +books themselves. +

+ +

Which Came First - Double Entry or the Enterprise?

+ +

+Double Entry bookkeeping is one of the +greatest discoveries of commerce, and its +significance is difficult to overstate. +Historians think it to have been invented +around the 1300s AD, although there are suggestions +that it existed in some form or other as far +back as the Greek empire. +The earliest strong evidence +is a 1494 treatise on mathematics +by the Venetian Friar + +Luca Pacioli + [LP]. +In his treatise, Pacioli documented many standard +techniques, including a chapter on accounting. +It was to become the basic text in double entry +bookkeeping for many a year. +

+ +

+Double Entry bookkeeping arose in concert with the +arisal of modern forms of enterprise as pioneered +by the Venetian merchants. Historians have debated whether +Double Entry was invented to support the dramatically +expanded demands of the newer ventures then taking place +surrounding the expansion of city states such as Venice +or whether Double Entry was an enabler of this expansion. +

+ +

+Our experiences weigh in on the side of enablement. +I refer to the experiences of digital money issuers. +Our own first deployment of a system was with a +single entry bookkeeping system. Its failure rate +even though coding was tight was such that it could +not sustain more than 20 accounts before errors in +accounting crept in and the system lost cohesion. +This occurred within weeks of initial testing and +was never capable of being fielded. The replacement +double entry system was fielded in early 1996 and +has never lost a transaction +(although there have been some close shaves + [IG1]). +

+ +

+Likewise, the company DigiCash BV of the Netherlands +fielded an early digital cash system into a bank in +the USA. During its testing period, the original +single entry accounting system had to be field +replaced with a double entry system for the same +reason - errors crept in and rendered the accounting +underneath the digital cash system unreliable. +

+ +

+Another major digital money system lasted for many +years on a single entry accounting system. Yet, +the company knew it was running on luck. When a +cracker managed to find a flaw in the system, an +overnight attack allowed the creation of many +millions of dollars worth of value. As this was more +than the contractual issue of value to date, it +caused dramatic contortions to the balance sheet, +including putting it in breach of its user contract +and at dire risk of a 'bank run'. Luckily, the cracker +deposited the created value into the account of an +online game that failed shortly afterwards, so the +value was able to be neutralised and monetarily +cleansed, without disclosure, and without scandal. +

+ +

+In the opinion of this author at least, single +entry bookkeeping is incapable of supporting any +enterprise more sophisticated than a household. +Given this, I suggest that evolution of complex +enterprises required double entry as an enabler. +

+ +

Computing Double Entry in Quick Time

+ +

+Double Entry has always been the foundation +of accounting systems for computers. The capability +to detect, classify and correct errors is even more +important to computers than it is to humans, +as there is no luxury of human intervention; +the distance between +the user and the bits and bytes is far greater +than the distance between the bookkeeper and +the ink marks on his ledgers. +

+ +

+How Double Entry is implemented is a +subject in and of itself. +Computer science introduces concepts such +as transactions, +which are defined as units of work that are +atomic, consistent, +isolated, and durable +(or ACID for short). +The core question for computer scientists is how to add an +entry to the assets side, then add an entry to the liabilities +side, and not crash half way through this sequence. +Or even worse, have another transaction start half way through. + +This makes more sense when considered in the context of the +millions of entries that a computer might manage, and a +very small chance that something goes wrong; eventually +something does, and computers cannot handle errors of that +nature very well. +

+ +

+For the most part, these concepts simply reduce +to "how do we implement double entry bookkeeping" ? +As this question is well answered in the literature, +we do no more than mention it here. +

+ +

A Slightly Less Brief History of the Signed Receipt

+ +

+Recent advances in financial cryptography have provided +a challenge to the concept of double entry bookkeeping. +The digital signature is capable of creating a record +with some strong degree of reliabilty, at least in the +senses expressed by ACID, above. +A digital signature can be relied +upon to keep a record safe, as +it will fail to verify if any +details in the record are changed. +

+ +

+If we can assume that the the record was originally +created correctly, then later errors are revealed, +both of an accidental nature and of fraudulent intent. +(Computers very rarely make accidental errors, +and when they do, they are most normally done +in a clumsy fashion more akin to the inkpot +being spilt than a few numbers.) +In this way, any change to a record that makes +some sort of accounting or semantic sense is +almost certainly an attempt at fraud, and a +digital signature makes this obvious. +

+ +

The Digital Signature and Digital Cash

+ +

+A digital signature gives us a particular property, +to whit: +

+ +
+"at a given point in time, this information +was seen and marked by the signing computer." +
+ +

+There are several variants, +with softer and harder claims to that property. +For example, message digests with +entanglement form one simple and +effective form of signature, and +public key cryptosystems provide +another form where signers hold a private +key and verifiers hold a public key + [MB]. + +There are also many ways to attack the +basic property. +In this essay I avoid +comparisons, and assume the basic property +as a reliable mark of having been seen by +a computer at some point in time. +

+ +

+Digital signatures then +represent a new way to create reliable +and trustworthy entries, which can be +constructed into accounting systems. +At first it was suggested that a +variant known as the +blinded signature +would enable digital cash + [DC]. +Then, certificates would +circulate as rights or contracts, in much +the same way as the share certificates +of old and thus replace centralised accounting +systems + [RAH]. + +These ideas took financial cryptography part of +the way there. Although they showed how to +strongly verify each transaction, they stopped +short of placing the the digital signature in an +overall framework of accountancy and governance. +A needed step was to add in the redundancy implied +in double entry bookkeeping in order to protect +both the transacting agents and the +system operators from fraud. +

+ +

The Initial Role of a Receipt

+ + + +
+ + +
+

1: An Interim Receipt

+
+ + + +
+ + + + + + + + + + + + + + + + + +
FromAlice
ToBob
UnitEuro
Quantity100
Date2005.12.25
+
+ + + + +
digital signature
+
+ +
+
+ +

+Designs that derived from the characteristics of +the Internet, the capabilities of cryptography +and the needs of governance +led to the development of the +signed receipt + [GH]. +In order to develop this concept, let us assume +a simple three party payment system, +wherein each party holds an authorising +key which can be used to sign their instructions. +We call these players Alice, Bob (two users) +and Ivan (the Issuer) +for convenience. +

+ +

+When Alice wishes to transfer value to Bob in some +unit or contract managed by Ivan, she writes out the +payment instruction and signs it digitally, +much like a cheque is dealt with in the +physical world. She sends this +to the server, Ivan, and he presumably agrees and +does the transfer in his internal set of books. +He then issues a receipt and signs it with his +signing key. +As an important part of the protocol, +Ivan then reliably delivers the +signed receipt to both Alice and Bob, +and they can update their internal books +accordingly. +

+ + +

The Receipt is the Transaction

+ +

+Our concept of digital value sought to eliminate as +many risks as possible. This was derived simply from +one of the high level requirements, that of being +extremely efficient at issuance of value. Efficiency +in digital issuance is primarily a function of support +costs, and a major determinant of support costs is +the costs of fraud and theft. +

+ +

+One risk that consistently blew away any +design for efficient digital value at +reasonable cost was the risk of insider fraud. +In our model of many users and a single +centralised server, the issuers of the unit +of digital value (as signatory to the contract) +and any governance partners such as the server +operators are powerful candidates for insider fraud. +Events over the last few years such as the mutual +funds and stockgate scandals are canonical cases +of risks that we decided to address. +

+ + + +
+ + +
+

2: A Signed Receipt

+
+ + +
+ + + + + + + + + + + + + + + + + + + + +
User's Cheque + + + +
+ + + + + + + + + + + + + + + + + + +
+
FromAlice
ToBob
UnitEuro
Qty100
ComPens
+
+ + + + +
Alice's sig
+
+
FromAlice
ToBob
UnitEuro
Quantity100
Date2005.04.10
+
+ + + + +
Ivan's signature
+
+ +
+
+ + +

+In order to address the risk of insider fraud, +the written receipt was historically introduced +as being a primary source of evidence. +Mostly forgotten to the buying public these +days, the purpose of a written receipt in +normal retail trade is not to permit returns +and complaints by the customer, but rather to +engage her in a protocol of documentation that +binds the shop attendant into safekeeping of +the monies. +A good customer will notice fraud by the +shop attendant and warn the owner to look +out for the monies identified by the receipt; +the same story applies to the invention of +the cash till or register, which was originally +just a box separating the owner's takings +from the monies in the shop attendant's +pockets. +We extend this primary motive into the +digital world by using a signed receipt +to bind the Issuer into a governance +protocol with the users. +

+ +

+We also go several steps further forward. +Firstly, to achieve a complete binding, +Alice's original authorisation +is also included within the record. +The receipt then includes all the +evidence of both the user's +intention and the server's action +in response, and it now becomes a +dominating record of the event. +This then means that the most efficient +record keeping strategy is to drop all +prior records and keep safe the signed +receipt. +

+ +

+This domination effects both the Issuer and the +user, and allows us to state the following principle: +

+ +
+The User and the Issuer hold the same information. +
+ +

+As the signed receipt is delivered from Issuer +to both users, all three parties hold the same +dominating record for each event. This reduces +support costs by dramatically reducing problems +caused by differences in information. +

+ +

+Secondly, we bind a signed contract +of issuance known as a +Ricardian Contract +into the receipt + [IG2]. +This invention relates a digitally signed +document securely to the signed receipt by +means of a unique identifier called a +message digest, +again provided by cryptography. +It provides strong binding for the unit +of account, the nature of the issue, the +terms, conditions and promises being made +by the Issuer, and of course the identity +of the Issuer. +

+ +

+Finally, with these enabling steps in place, +we can now introduce the principle: +

+ +
+The Receipt is the Transaction. +
+ +

+Within the full record of the signed receipt, +the user's intention is expressed, and is +fully confirmed by the server's response. +Both of these are covered by digital +signatures, locking these data down. +A reviewer such as an auditor can confirm the +two sets of data, and can verify the signatures. +

+ + + +

The Signed Receipt as a Bookkeeping system

+ +

+The principle of the Receipt as the Transaction +has become sacrosact over time. +In our client software, the principle has been +hammered into the design consistently, resulting +in a simplified accounting regime, and delivering +a high reliability. +Issues still remain, such as the +loss of receipts and the counting of balances +by the client side software, but these become +reasonably tractable once the goal of receipts +as transactions is placed paramount in the +designer's mind. +

+ +

As Single Entry

+ +

+In order to calculate balances on a related +set of receipts, or to present a transaction +history, a book would be constructed +on the fly from the set. +This amounts to using the Signed Receipt +as a basis for single entry bookkeeping. +In effect, the bookkeeping is derived from +the raw receipts, and this raises the +question as to whether to keep the books +in place. +

+ +

+The principles of Relational Databases +provide guidance here. +The fourth normal form +directs that we store the primary records, +in this case the set of receipts, and we +construct derivative records, the accounting +books, on the fly + [4NF]. +

+ +

Recovering Double Entry

+

+Similar issues arise for Ivan the Issuer. +The server has to accept each new transaction +on the basis of the available balance in the +effected books; for this reason Ivan needs +those books to be available efficiently. +Due to the greater number of receipts and +books (one for each user account), both +receipts and books will tend to exist, in +direct contrast to fourth normal form. +A meld +between relationally sound sets of receipts +and double entry books comes to assist here. +

+ +

+Alice and Bob both are granted a book +each within the server's architecture. +As is customary, we place those +books on the liabilities side. +Receipts then can be placed in a separate +single book and this could be logically +placed on the assets side. +Each transaction from Alice to Bob +now has a logical contra entry, +and is then represented in 3 places +within the accounts of the server. +Yet, the assets side remains in fourth +normal form terms as the liabilities +entries are derived, each pair from one +entry on the assets side. +

+ +

+By extension, a more sophisticated +client-side software agent, +working for Alice or Bob, +could employ the same techniques. +At this extreme, entries are now in place in +three separate locations, and each holding +potentially three records. +

+ +

Triple Entry Accounting

+ +

+The digitally signed receipt, with the entire +authorisation for a transaction, represents +a dramatic challenge to double entry bookkeeping +at least at the conceptual level. The cryptographic +invention of the digital signature gives powerful +evidentiary force to the receipt, and in practice +reduces the accounting problem to one of the +receipt's presence or its absence. This problem +is solved by sharing the records - each of the +agents has a good copy. +

+ +

+In some strict sense of relational database theory, +double entry book keeping is now redundant; +it is normalised away by the fourth normal form. +Yet this is +more a statement of theory than practice, and +in the software systems that we have built, the +two remain together, working mostly hand in hand. +

+ +

+Which leads to the pairs of double entries +connected by the central list of receipts; +three entries for each transaction. +Not only is each accounting agent led to +keep three entries, the natural roles +of a transaction are of three parties, +leading to three by three entries. +

+ +

+We term this triple entry bookkeeping. +Although the digitally signed receipt dominates +in information +terms, in processing terms it falls short. Double +entry book keeping fills in the processing gap, +and thus the two will work better together +than apart. In this sense, our term of triple +entry bookkeeping recommends an advance in +accounting, rather than a revolution. +

+ +

Software Considerations

+ +

+The precise layout of the entries in software +and data terms is not settled, +and may ultimately become one of +those ephemeral implementation issues. +The signed receipts may form a natural +asset-side contra account, or they +may be a separate non-book list underlying +the bookkeeping system and its two sides. +

+ +

+Auditing issues arise where construction +of the books derives from the receipts, +and normalisation issues arise when a +receipt is lost. These are issues for +future research. +

+ +

+Likewise, it is worth stating that +the technique of signing receipts works +both with private key signatures and also +with entanglement message digest signatures; +whether the security aspects of these +techniques is adequate to task is dependent +on the business environment. +

+ +

Roles of the Agents

+ +

+It will be noted that the above design of +triple entry bookkeeping assumed that Alice +and Bob were agents of some independence. +This was made possible, and reflected the +usage of the system as a digital cash system, +and not as a classical accounting system. +

+ +

+Far from reducing the relevance of this work +to the accounting profession, it introduces +digital cash as an alternate to corporate +bookkeeping. +If an accounting system for a corporation or +other administrative entity is recast as a +system of digital cash, or internal +money, then experience shows that +benefits accrue to the organisation. +

+ +

+Although the core of the system looks exactly +like an accounting system, each department's +books are pushed out as digital cash accounts. +Departments no longer work so much with budgets +as have control over their own corporate money. +Fundamental governance control is still held +within the accounting department by dint of their +operation of the system, and by the limited scope +of the money as only being usable within the +organisation; the accounting department might +step in as a market maker, exchanging +payments in internal money for payments in +external money to outside suppliers. +

+ +

+We have operated this system on a small scale. +Rather than be inefficient on such a small +scale, the system has generated dramatic +savings in coordination. No longer are bills +and salaries paid using conventional monies; +many transactions are dealt with by internal +money transfers and at the edges of the +corporation, formal and informal agents work +to exchange between internal money and +external money. Paperwork reduces dramatically, +as the records of the money system are reliable +enough to quickly resolve questions even years +after the event. +

+ +

+The innovations present in internal money +go beyond the present paper, but suffice +to say that they answer the obvious question +of why this design of triple entry accounting +sprung from the world of digital cash, and +has relevence back to the corporate world. +

+ +

Patterns of Commerce

+ +

+Todd Boyle looked at a similar problem from the point +of view of small business needs in an Internet age, +and reached the same conclusion - triple entry +accounting + [1]. + +His starting premises were that: +

+ +

  1. + The major need is not accounting or payments, per se, + but patterns of exchange - complex patterns of trade; +

  2. + Small businesses could not afford large complex + systems that understood these patterns; +

  3. + They would not lock themselves into proprietary + frameworks; +

+ +

+From those foundations, Boyle concluded that +therefore what is needed is a shared access repository +that provides arms-length access. Fundamentally, this +repository is akin to the classic double-entry accounting ledger +of transaction rows ("GLT" for General Ledger - Transactions), +yet its entries are dynamic and shared. +

+ +

+Simple examples will help. +When Alice forms a transaction, she enters it into her software. +Every GLT transaction requires naming her external +counterparty, Bob. When she posts the transaction, +her software stores it in her local GLT and also +submits it to the shared repository service's GLT. +

+ +

+The Shared Transaction Repository ("STR") then forwards +the transaction on to Bob. Both Bob and Alice are now +expected to store the handle to the transaction as +an index or stub, and the STR then stores the entire +transaction. +

+ +

+Boyle's ideas are logically comparable to Grigg and Howland's, +although they arive from different directions +(the STR is Grigg's Ivan, above) and are not totally +equivalent. +Where the latter limited themselves to payments, +the accuracy of amounts, +and protection with hard cryptographic shells, +Boyle looked at wider patterns of transactions, +and showed that the STR could mediate these transactions, +if the core shared data could be extracted and made into +a single shared record. +Boyle's focus was on the economic substance of the +transaction. +

+ +

Extending the Humble Invoice

+ +

+Imagine a simple invoicing procedure. +Alice creates an invoice and posts it to her software (GLT). +As she has named Bob, +the GLT automatically posts it to Ivan, +the STR, and he forwards it to Bob. +At this point Bob has a decision to make, accept or +reject. Assuming acceptance, his +software can then respond by sending +an acceptance message to Ivan. +The STR now assembles an accepted +invoice record to replace the earlier +speculative invoice record and posts that threeways. +At some related time (to do with payment policy) +Bob also posts a separate transaction to pay for the +invoice. This could operate in much the same way as +a separate transaction, linking directly to the +original invoice. +

+ +

+Now, as the payment links back, +and the invoice is a live transaction within the three +entries in the three accounting systems, it is possible +for a new updated invoice record to refer back to the +payment activity. +When the payment clears, the new record can +again replace the older unpaid copy and +promulgate to all three parties. +

+ +

Patterns of Transactions

+ +

+Software could be written to facilitate and monitor +this flow and similar flows. +If the payments system is sufficiently +flexible, and integrated with the needs of the users, +if might be possible to merge the above invoice +with the payment itself, at the Receipts level. +Seen in this light, the Signed receipt of Ricardo +is simply the smallest and simplest pattern within +the more general set of patterns. +We could then suggest that the narrow principle of +the Receipt is the Transaction +could be extended into +The Invoice is the Transaction. +

+ +

+A particular transaction in business almost never +stands alone. They come in patterns. +For example offers and acceptances form a wider +transaction but seldom encapsulate the entire +fulfillment and payment cycle. +Even if there has been a payment +accompanying a PO message, +the customer then waits for fulfillment. +

+ +

+There is a large body of science and literature +built around these patterns of transactions. +These have been adopted by the Business Process +workgroup of ebXML and other standards bodies, +where they are called "Commercial Transactions." +Where however the present work distinguishes itself +is in breaking down these transactions into the +atomic elements. It is to that we now turn. +

+ +

The Requirements of Triple Entry Accounting

+ +

+The implementation of Triple Entry Accounting +will in time evolve to support +patterns of transactions. +What has become clear is that double entry does +not sufficiently support these patterns, as it +is a framework that breaks down as soon as the +number of parties exceeds one. +Yet, even as double entry is "broken" on the net +and unable to support commercial demands, +triple entry is not widely understood, +nor are the infrastructure requirements +that it imposes well recognised. +

+ +

+Below are the list of requirements that we +believed to be important + [2] + [3]. +

+ +

+1. Strong Psuedonymity, At Least. +As there are many cycles in the patterns, the +system must support a clear relationship of +participants. At the minimum this requires +a nymous architecture of the nature of Ricardo +or AADS. (This requirement is very clear, +but space prevents any discussion of it.) +

+ +

+2. Entry Signing. +In order to neutralise the threats to and by the +parties, a mechanism that freezes and confirms +the basic data is needed. +This is signing, and we require that all entries +are capable of carrying digital signatures +(see 1, above, which suggests public key signatures). +

+ + + +

+3. Message Passing. +The system is fundamentally one of message passing, +in contrast to much of the net's connection based +architecture. +Boyle recognised early on that a critical +component was the generic message passing nature, +and Systemics proposed and built this +into Ricardo over the period 2001-2004 + [4]. +

+ +

+4. Entry Enlargement and Migration. +Each new version of a message coming in represents +an entry that is either to be updated or added. +As each message adds to a prior conversation, +the stored entry needs to enlarge and absorb the +new information, while preserving the other properties. +

+ +

+5. Local Entry Storage and Reports. +The persistent saving and responsive availability +of entries. +In practice, this is the classical accounting +general ledger, at least in storage terms. +It needs to bend somewhat +to handle much more flexible entries, +and its report capabilities become more key +as they conduct instrinsic reconciliation +on a demand or live basis. +

+ +

+6. Integrated Hard Payments. +Trade can only be as efficient as the payment. +That means that the payment must be at least as +efficient as every other part; which in practice +means that a payment system should be built-in +at the infrastructure level. C.f., Ricardo. +

+ + +

+7 Integrated Application-Level Messaging. +As distinct to the messaging at the lower protocol +levels (1 above), there is a requirement for Alice and Bob +to be able to communicate. That is because the +vast majority of the patterns turns around the +basic communications of the agents. There is no +point in establishing a better payment and invoice +mechanism than the means of communication and +negotiation. This concept is perhaps best seen +in the SWIFT system which is a messaging system, +first and foremost, to deliver instructions for +payments. +

+ + +

Conclusion

+ +

+Double Entry bookkeeping provides evidence of +intent and origin, leading to strategies for +dealing with errors of accident and fraud. +The financial cryptography invention of the +signed receipt provides the same benefits, +and thus challenges the 800 year reign of +double entry. +Indeed, in evidentiary terms, the signed receipt +is more powerful than double entry records due to +the technical qualities of its signature. +

+ +

+There remain some weaknesses in strict comparison +with double entry bookkeeping. Firstly, in +the Ricardo instantiation of triple entry +accounting, the +receipts themselves may be lost or removed, +and for this reason we stress as a principle that +the entry is the transaction. +This results in three +active agents who are charged with securing +the signed entry as their most important +record of transaction. +

+ +

+Secondly, the software ramifications of the +triple entry system that are less convenient than that +offered by double entry bookkeeping. For this +reason, we expand the information held in the +receipt into a set of double entry books; +in this way we have the best of both worlds on each node: +the evidentiary power of the signed entries and +the convenience and local crosschecking power of +the double entry concept. +

+ +

+Both of these imperitives meld signed receipts in +with double entry bookkeeping. As we end up with +a logical arrangement of three by three entries, +we feel the term triple entry bookkeeping +is useful to describe the advance on the older form. +

+ +

Drawing in the Agents

+ +

+To fully benefit from triple entry bookkeeping, +we have to expand accounting systems out to +agents and offer them direct capabilities to +do transactions. +That is, we make the agents stakeholders by +giving them internal money + [5]. +Use of digital cash to do company accounts +empowers the use of this concept as a general +replacement for accounting using books and +departmental budgets, and is an enabler for +verifying and auditing the centralised +accounts system by way of signed receipts. +

+ +

Solving Frauds

+ +

+Once there, governance receives substantial +benefits. Accounts are now much more difficult +to change, and much more transparent. It is +our opinion that various scandals and failures +of governance would have been impossible given +these techniques: the mutual funds scandal +would have shown a clear audit trail of transactions +and thus late timing and otherwise perverted or +dropped transactions would have been clearly +identified or eliminated completely + [NG]. +The emerging scandal in the USA known as +Stockgate would have been impossible +as forgery of shares and value for manipulative +trading purposes is revealed by signed receipts. +Likewise, Barings would still be a force in +investment banking if accounts had been +organised around easily transparent digital cash +with open and irreducible signed receipts that +evidence invisible accounts (88888). +Enron style scandals would have permitted more +direct "follow the money" governance lifting +the veil on various innovative but economically +meaningless swaps. +

+ + +

References

+ + +

+[TB] + A draft form of this paper credited Todd Boyle + as an author, but this was later withdrawn at + his request due to wider differences between + the views. + + +

+ +

+[LP] +Friar Luca Pacioli, + +Summa de Arithmetica, Geometria, +Proportioni et Proportionalita + +1494, Venice. + + +

+ +

+[IG1] +Ian Grigg +" +The Twilight Zone +," +Financial Cryptography blog +16th April 2005 + + +

+ +

+[MB] +Entanglement is discussed in: +Petros Maniatis and Mary Baker, +"Secure History Preservation through Timeline Entanglement," +Proc. 11th USENIX Security Symposium, +August 2002. + + +

+ +

+[DC] +David Chaum, +"Achieving Electronic Privacy," +Scientific American, +v. 267, n. 2 Aug 1992. + + +

+ +

+[RAH] +Robert A. Hettinga +" +The Book-Entry/Certificate Distinction +" +1995, Cypherpunks + + +

+ +

+[GH] +Gary Howland +" +Development of an Open and Flexible Payment System + +1996, Amsterdam, NL. + + +

+ +

+[IG2] + Ian Grigg + " + The Ricardian Contract + ," + First IEEE International + Workshop on Electronic Contracting + (WEC) 6th July 2004 + + +

+ +

+[4NF] +E.F. Codd, +" +A Relational Model of Data for Large Shared Data Banks +," +Comm. ACM 13 (6), June 1970, pp. 377-387. + + +

+ +

+[1] + Todd Boyle, + " + GLT and GLR: conceptual architecture for general ledgers," + Ledgerism.net, 1997-2005. + + +

+ +

+[2] + Todd Boyle, + " + STR software specification," + Goals, 1-5. + This section adopts that numbering convention. + + +

+ +

+[3] + Ian Grigg, + various design and requirements documents, + Systemics, unpublished. + + +

+ +

+[4] + A substantial part of the programming and design was conducted + by Edwin Woudt (first demo, SOX layers, UI) + and Jeroen van Gelderen + (message passing client architecture). + + +

+ +

+[5] + Using internal money instead of an accounting system + is not a new idea but has only been recently experienced: + Ian Grigg, + + How we raised capital at 0%, saved our + creditors from an accounting nightmare, gave our suppliers + a discount and got to bed before midnight. + Informal essay (rant), 7 Jul 2003. + + +

+ +

+[NG] +James Nesfield and Ian Grigg +" +Mutual Funds and Financial Flaws +," +U.S. Senate Finance Subcommittee +27th January, 2004 + + +

+ + + + + +------MultipartBoundary--ZJLbQ4mRwLhAHsHENVppcDUXKvpxlvNoq0H6MRFUAo------ diff --git a/docs/triple_entry_accounting-by_Iang.url b/docs/triple_entry_accounting-by_Iang.url new file mode 100644 index 0000000..b575e52 --- /dev/null +++ b/docs/triple_entry_accounting-by_Iang.url @@ -0,0 +1,2 @@ +[InternetShortcut] +URL=http://iang.org/papers/triple_entry.html diff --git a/docs/triple_entry_accounting.md b/docs/triple_entry_accounting.md new file mode 100644 index 0000000..0e7bccb --- /dev/null +++ b/docs/triple_entry_accounting.md @@ -0,0 +1,198 @@ +--- +title: >- + Triple Entry Accounting +--- +See [Sox accounting], for why we need to replace Sox accounting with triple entry accounting. + +[Sox accounting]:sox_accounting.html +"Talmudic Ritual Purity" +{target="_blank"} + +[Talmudic ritual purity]:sox_accounting.html +"Sox accounting" +{target="_blank"} + +# What is triple entry accounting + + Double entry accounting ensures that the books of the corporation are + internally consistent. Triple entry accounting ensures that the books of the + corporation are not only internally consistent, but that its account of its + obligations and transactions with clients and partners is consistent with their account. + + This requires a shared pool of data shared between several entities who +have long lasting and repeated durable business relationships with each +other, and if it is a lot of data, a blockchain. + +Blockchains are designed with considerable care against various subtle +forms of cheating, and this design is profoundly difficult, complex, and +notoriously subtle and difficult to correctly implement. + +The way of the future will be to move bookkeeping, accounting, and +various other measures against cheating to the blockchain. + +The fundamental force moving us to a blockchain based world is an +untrusted and untrustworthy elite. + +# Triple entry accounting and corporate sovereignty + +A corporation is not so much a legal fiction, as a book keeping fiction. The +first double entry accountants wanted to know how an enterprise was +doing, so they created double entry columns that made the books of the +enterprise balance, rather than the merely the books of the owner balance. +And by enough people believing and acting as though the enterprise was +itself a real thing, it became a real thing, the real thing being those people +acting as if with one will, the will of the enterprise. + +And then King Charles the second gave some of these accounting fictions +made real legal status as corporations, creating the modern joint stock publicly traded for-profit corporation, the modern corporation. + +Corporations are people, real people, because enterprises are real people, +whose will is made one through bookkeeping tracking what value they +create and cost for each other. A corporation is not so much a legal fiction, as a book keeping fiction. + +And then King Charles the second gave some of these accounting fictions +made real legal status as corporations, creating the modern joint stock +publicly traded for-profit corporation, the modern corporation. + +Corporations are people, real people, because enterprises are real people, +whose will is made one through bookkeeping tracking what value they +create and cost for each other. + +Double entry book keeping is social technology. It fundamentally shapes +our society, even though almost no one understands it. The corporation +exists through double entry book keeping. + +A small number of partners who own a business, who know and trust each +other, and understand double entry book keeping can enforce it on each +other, but the publicly traded joint stock corporation exists through state +enforcement of double entry book keeping. + +And increasingly what the state is enforcing is not double entry +bookkeeping. Instead of tracking the creation and movement of value, the +books track the creation and movement of [Talmudic ritual purity]. + +The block chain can enable a very large number of business owners who +do not know and trust each other and who do not understand double entry +bookkeeping to enforce double entry bookkeeping on each other. + +Suppose your company’s books are triple entry accounting based o +immutable journal entries, and its shares are on its blockchain, not on the +records of the government regulated stock exchange. + +Such an enterprise derives its cohesion not from a grant of corporateness +from the state, but because all the shareholders have to follow the rules +because all the other shareholders are following the rules, as every +blockchain works. They don’t have to understand how the wallet works, they +just have to understand that if they don’t have money in their wallet, +they cannot pay, and if they don’t have shares in their wallet, they cannot +sell them and cannot vote when there is a board election. + +This renders the corporation independent of the state. The state can coerce +the CEO, the board, and the shareholders, the same way as it can coerce +anyone else, assuming it can find them, and assuming it can discover what +they are doing, but the company is not longer a creation of the state, +animated by state enforcement of its corporate character and state +enforcement of its book keeping. + +This makes it vastly harder to tax and regulate, even if only 0.001% of the +population understand the cryptographic protocols employed by their +wallet. If the corporation exists as blockchain protocols, the government +cannot simply deduct money out of your paycheck – it has to send men +with guns to knock on your door and say “pay or else” – and it has to find +you, which may not be easy if you are working remotely, or you are +working in person at a small remote branch of the business. The board +probably has its meetings virtually. They probably cannot find the board, +though the CEO probably has to show up in person a lot. + +This also puts me back in business, since startups will once again be possible. Startups have been regulated out of business. + +# Differences between triple and double entry + +In triple entry accounting all parties to a transaction keep the same +digitally signed record of a transaction. + +Triple entry accounting is floodfilling digitally signed transactions around +all parties affected, and then displaying relevant totals over these digitally +signed transactions. The relevant totals correspond to the various asset and +liability subtypes of double entry accounting. It is triple entry, because the +same digitally signed record shows up in many places – and not exactly +two, nor for that matter, exactly three. + +Sums over these records preserve the invariants of double entry +accounting. Failure to preserve the double entry invariants indicates a +communication failure, update failure, or disk corruption that forces an +automatic retry until double entry invariants are restored. + +The underlying digitally signed records of transactions are flood filled +around, guaranteeing that all parties have consistent books, that not only +does one entity's books balance, but that the entries in one entity's books +are consistent with the entries in another entity's books. + +In regular double entry accounting, all totals are assets or liabilities, and +every transaction causes a change to two totals, every transaction has two +effects, such that total assets and liabilities remain equal to zero, or equal +to its initial value.  The business has zero net assets, because it owns stuff, +and owes its owners stuff. + +Every transaction has two effects. For example, if someone transacts a +purchase of a drink from a local store, he pays cash to the shopkeeper and +in return, he gets a bottle of dink. This simple transaction has two effects +from the perspective of both, the buyer as well as the seller. The buyer’s +cash balance would decrease by the amount of the cost of purchase while +on the other hand he will acquire a bottle of drink. Conversely, the seller +will be one drink short though his cash balance would increase by the +price of the drink. + +Accounting attempts to record both effects of a transaction or event on the +entity’s financial statements. This is the application of double entry +concept. Without applying double entry concept, accounting records +would only reflect a partial view of the company’s affairs. Imagine if an +entity purchased a machine during a year, but the accounting records do +not show whether the machine was purchased for cash or on credit. +Perhaps the machine was bought in exchange of another machine. Such +information can only be gained from accounting records if both effects of +a transaction are accounted for. + +Traditionally, the two effects of an accounting entry are known as Debit +(Dr) and Credit (Cr). Accounting system is based on the principal that for +every Debit entry, there will always be an equal Credit entry. This is +known as the Duality Principal. + +Debit entries are ones that account for the following effects: + +* Increase in assets +* Increase in expense +*Decrease in liability +* Decrease in equity +* Decrease in income + +Credit entries are ones that account for the following effects: + +* Decrease in assets +* Decrease in expense +* Increase in liability +* Increase in equity +* Increase in income + +Double Entry is recorded in a manner that the Accounting Equation is +always in balance. + +Assets – Liabilities = Capital + +Any increase in expense (Dr) will be offset by a decrease in assets (Cr) or +increase in liability or equity (Cr) and vice-versa. Hence, the accounting +equation will still be in equilibrium. + +Triple entry accounting is double entry accounting with each transaction +linking to signed agreement by the relevant parties, and the relevant +parties sum over these signed agreements in different ways, that result in +the assets and liabilities of each entity coming out correctly. Everyone +accumulates a pile of signed transactions, and these signed transactions +belong to categories such that the double entry invariants are preserved. +Triple entry accounting is that we have a pile of signed database records +with a rule that any complete collection of the relevant records results in +both parties seeing the accounting invariants preserved, and automatic +check and retry in the event of discrepancies. + +It ensures that not only do one company's books reflect a consistent view, +but both parties share the same consistent view of each other's obligations. diff --git a/docs/triple_entry_accounting.url b/docs/triple_entry_accounting.url new file mode 100644 index 0000000..8ea43e0 --- /dev/null +++ b/docs/triple_entry_accounting.url @@ -0,0 +1,2 @@ +[InternetShortcut] +URL=https://indiatechlaw.com/finance/blockchain-changing-finance-industry-triple-entry-accounting/ diff --git a/docs/usury.md b/docs/usury.md new file mode 100644 index 0000000..3e4045e --- /dev/null +++ b/docs/usury.md @@ -0,0 +1,136 @@ +--- +title: + The Usury problem +--- +The Christian concept of usury presupposes that capitalism was divinely +ordained in the fall, and that we are commanded to use capital productively +and wisely. + +This is quite different from the communist concept of usury, which is that +making money on a mortgage is sinful, because only labour is productive +and capital is unproductive. The communists rejects interest, and all return +on capital, full stop. Not that it matters, since communist money is +worthless, and no one has any productive property under communism. + +Christians are allowed to make money from loans made for productive +purposes, provided the loan actually turns out to be productive – +Christians are allowed to profit from capital goods, and allowed to profit +from loans made to create capital goods. If the loan turns out to be a bad +use of capital, if it turns out to be an unproductive use of capital, the +lender has to share the pain with the borrower. The lender gets back the +house in the housing slump, and the cattle in a drought. + +Credit card debt, however, is unlikely to be productive. And missed payments +are not productive either, and should not become a profit centre for someone +who lends money to short time preference people who probably should not be +borrowing. + +The old Christian concept of usury is that you can lend money at interest +against productive property, but not against the person. If you lend against +the person, the debt is erased by full repayment without interest, and if you +charge interest, not recoverable against the person, only against his credit +rating. + +So, you lend money to the peasant to buy some cattle, the cattle produce more +cattle, and you get some of the extra cattle, or, more conveniently, a fixed +amount of money per year for the peasant’s use of that cattle, or the +mortgagor’s use of that house. The loan is secured by productive property, +not secured against the person, and you are entitled to some of that +production. + +And if things don’t work out, he is free and clear if he returns the cattle or +the house. + +The Islamic ban on usury is similar to the Christian ban, but their frame is +rather than the lender shares the risk, rather than the lender is charging +rental on a productive property. The dark enlightenment frame is game +theoretic, that the lender often knows better than the borrower what is a +bad loan for the borrower, and should not have incentive to trap the +borrower in a bad loan, but all these different frames amount to the same +thing in the end -- if the loan goes well, the borrower winds up paying +back more than he borrowed, and if it goes badly, both parties suffer the +consequences. If you finance your house from Dubai Islamic Bank, and +the house appreciates, it is exactly the same as if you financed your house +from Bank of America. The difference happens if the bank lends into a +housing boom, and then there is a housing slump. + +If you lend money to buy a house in the middle of a housing boom, you +collect interest on the mortgage, but then there is a housing slump, the +mortgagor returns the house in good order and condition, but in the middle of +housing slump, the mortgagee is sol under the old Christian laws, and the mortgagor, though now houseless, is free and clear of debt. + +Well, lenders did not like that. They wanted their money even if things did +not work out, and they wanted to be able to lend money to someone to throw a +big party, someone who probably did not understand the concept of compound +interest, and then own that someone. + +And the Jews of course operated by different rules, and Kings would borrow +from the Jews, and then give the Jews exemption from the Christian laws, so +that they could lend money against the person to Christians. + +And then, things got messier with fractional reserve banking. + +People want to lend short and borrow long, borrow money with fixed schedule +for paying it back in many years, and lend money with the proviso that they +can have it back at any time, money on deposit. + +And so, the magic of term transformation. The banker takes in ten thousand +gold pieces on deposit, returnable on demand at low interest, and lends out +nine thousand gold pieces on land, on cattle, and on conspicuous consumption, +at rather higher interest. + +The banker hopes that not everyone will try to withdraw at the same time. + +And, since the banker does not want to find himself in the real estate +business when there is a slump in real estate prices, or the cattle business +when there is a drought, he lobbies against the Christian rules on usury, +and in favour of the Jewish rules. If you buy a house on a mortgage, and the +price falls below the mortgage, he wants to sell the house over your head, +and then go after you for the difference. + +Since it is rather dangerous to move gold around, and safer to move +ownership of gold around, people, instead using gold pieces as money, +start using banknotes as money, bits of paper backed by claims against real +property, if the lending is more or less Christian, and claims against real +people, if it is not all that Christian. + +And then, one day it rains on everyone, and everyone hits up the bank at +the same time for the money they had stashed away for a rainy day, and +you have a financial crisis. + +And, under the rules the bankers lobbied for, the angry depositors can take +all their stuff, and probably give the bankers a horsewhipping. + +So the bankers rush to the government, and say, “financial crisis, bailout” + +And then instead of banknotes backed by claims against property, you get +government notes backed by claims against taxpayers. + +And here we are. Jewish rules, fiat money. + +Well, how does cryptocurrency address this? It is backed by absolutely +nothing at all. + +No, not quite nothing at all, for what it is backed by is the that cryptocurrency +can be owned more securely than anything else, and moved across the +world at the speed of light, making it very useful as money. You own +crypto currency by having the secrets that control it, which are harder for +governments or bad guys to find, and a lot easier to transport through +airports. What it is backed by is that it is a form of property right that is +easy to defend, and a form of property that is easy to move around. + +Now it is probable that if cryptocurrency successfully replaces the dollar, +the same story will begin again from the beginning, only with crypto +currency in place of gold, but we are starting out from a clean slate. You +cannot lend cryptocurrency against the person, because you cannot *find* +the person. So when, in the future, people start borrowing and lending in +crypto currency, the Christian rules will be inherently in effect when the +process starts all over. + +So, clean slate. Keeping it clean may well turn out to be difficult. + +But gold was inconvenient and dangerous to move around, so people +preferred to move claims against bankers around. So, it will prove a lot +easier for people to hang onto cryptocurrency, rather than leaving their +gold with the bankers, so the pressure to repeat the story that happened +with gold will be considerably less. diff --git a/docs/white_paper.md b/docs/white_paper.md new file mode 100644 index 0000000..0b0d36a --- /dev/null +++ b/docs/white_paper.md @@ -0,0 +1,427 @@ +--- +title: >- + Rhocoin White Paper +--- +This is a preliminary draft, not the final version. + +The centre of mass of the world financial system is starting to shift from +fiat currency to crypto currency and crypto currency exchanges. Total +value of crypto currency and transaction volume is within one order of +magnitude of US\$, and when regulatory authorities cut off crypto +currency exchanges and stablecoins from the US$ banking system it has +very little impact on them. + +# Scaling + +The time urgently approaches for a cryptocurrency capable of scaling to +ten thousand transactions per second, so that eight billion people can buy a +lollipop with crypto currency. Bitcoin can currently do ten transactions per +second. + +The big problems are scaling and privacy. + +# Fungibility + +A currency has to be fungible. And a cryptocurrency needs privacy, not so +much to protect its users (there does not seem to be a lot of demand for +privacy) but to protect its fungibility. + +Lack of privacy is likely to lead to the blood diamonds attack, where +governments declare a long and ever growing list of bitcoins to be tainted, +and twist the arms of the miners to exclude transactions in those bitcoins, +with the result bitcoins cease to be fungible and thus cease to be money, as +uncut diamonds ceased to be money. + +Bitcoin is vulnerable to the one third attack. If one third of miners exclude +“tainted” bitcoins and refuse to add to a chain ending in a block containing +“tainted” bitcoins, other miners have an incentive to exclude “tainted” +bitcoins also, to improve the prospects that their block becomes part of the +longest chain. + +(This has been called the ten percent attack, but it does not actually work +until rather more than ten percent join the attack. One third, however will +suffice. Calling it the blood diamonds attack is more accurate. The blood +diamonds attack put a lot of people out of business, for they were really in +the money changer business, rather than the jewelry business.) + +[Bitcoin is already under the ten percent attack. If it approaches one third, +time to flee bitcoin.](http://reaction.la/security/bitcoin_vulnerable_to_currency_controls.html) + +Blockchain analysis is a big problem, though scaling is rapidly becoming a +bigger problem. + +Monaro and other privacy currencies solve the privacy problem by padding +the blockchain with chaff, to defeat blockchain analysis, but this +greatly worsens the scaling problem. If Bitcoin comes under the blood +diamonds attack, Monaro will be the next big crypto currency, but Monaro +has an even worse scaling problem than Bitcoin. + +The solution to privacy is not to put more data on the blockchain, but less - +considerably less. The place for clever cryptography to secure privacy is +the lightning network and sidechaining, so that most transactions happen +off the main blockchain, with primary blockchain transactions each +recording the accumulated effect of many transactions. + +A true lightning network functions as full reserve correspondence +banking with no central authority. The existing Bitcoin lightning network +functions as marginal reserve banking with good conduct enforced by central +authority. + +But even with sidechaining and the lightning network, we are going to +need the capacity to put thousands of transactions per second on the +primary blockchain. + +We need off blockchain blockchains, sidechains, but how do you transact +between one sidechain and another? It would seem that it is not much use +if it only supports half a dozen people transacting with each other. + +The answer is that such transactions reach other people’s sidechains through +the lightning network, that you are usually transacting with one +particular person in that sidechain, who is a well connected lightning node. +A sidechain transaction can only directly connect to someone who is +participating in that particular sidechain, but it can reach anyone in the +world through the lightning network if one entity of the small number of +entities in that small sidechain has half a dozen lightning gateways on +other sidechains and on the primary blockchain. + +The lightning network naturally results in lots of repeated blockchain +transactions between stable small groups of people, because you are +always sharing new cryptocoins between the same group of people when a +lighting gateway overflows. + +So every time you have a stable account with someone, your pub, your +club, your fiber connection, your bank, the shopping chain that delivers +your groceries, you form a lightning network gateway with them, whose +shared cryptocoins are likely in a sidechain much of the time. You +participate in roughly as many sidechains as you have accounts or +regular repeated transactions with people. + +The solution to all these problems is a lightning network done right, to +support both scalability and pseudonymity. Multiparty lightning network +transactions have to be trustless and full circle. Ann pays Bob to pay Carol +to pay Ed to pay Frank, and Frank sends a receipt to Ann, and all of them +go through, or none of them go through, and if someone breaks the circle, +then either none of them go through (non Byzantine failure, as with a poor +connection to the internet, or someone’s computer goes down), or, if +someone breaks the circle in a Byzantine failure suggestive of Byzantine +defection, then, and only then, it goes through on the primary blockchain. + +The problem with the existing Bitcoin lightning network is that it has a +hidden and unexplained central authority, whom you have to trust, and +which does stuff that is never explained or revealed. This is not stable, and +does not scale. Not only is it evil, it is incapable of connecting everyone +in the world to everyone in the world. The existing lightning network has the +same problem as Tether. + +Tether is not a ponzi scheme. It is an unregulated bank, but it is still doing +marginal reserve banking, and will implode sooner or later due to insider +fraud or maturity transformation, and something analogous is bound to +happen with the existing Bitcoin lightning network, because of the inherent +fragility of centralization. The moral problem of the existing lightning +network is the same as the moral problem of marginal reserve +correspondence banking. Scaling requires trustlessness. Or rather you are +trusting that if enough people see and process the transactions in full, then, +because they are not parties to that transaction and don’t have a dog in the +fight, they will process it correctly. And as soon as you have a central +authority that you have to trust, you have a party with an interest and +capability to not process it correctly. + +So we don’t want everyone in the world, or even every full peer in the +world, to process every transaction in the world. We want every full peer +in the world to process every transaction in the world where the parties +quarrel, with most other transactions never showing up directly on the +primary blockchain. And we don’t want everyone in the world to be a full +peer. We want enough full peers that the vast majority will not have a dog +in the fight, and we want anyone in the world who is reasonably affluent +and wants to be a full peer to be able to be a full peer, which is likely to be +most with substantial amounts of cryptocurrency. At scale, nearly +everyone will keep his money in his client wallet, but if it is a lot of +money, his client wallet will likely be a client of a peer that he controls. + +# The failures of bitcoin + +The pseudonymity of coins being owned by the bearer of some +cryptographic key is a failure; People have been eavesdropping and +aggressively analyzing the block chain from day 1. And the block chain +will always be there, it will always be public, and it will always be subject +to further analysis. And we are learning that analysis of that record is +sufficient to destroy any pretense of anonymity or pseudonymity. + +The scarcity of transactions has led people to re-invent every last feature +of the banks they thought they were going to be escaping. Including debt +brokering (lightning network) and fractional-reserve banking, starting with +the case of Mt. Gox and continuing to ventures today by “responsible” +business people who just don’t get, or don’t care, or both, that the entire +reason the system existed, as far as the early adopters were concerned, was +to get away from exactly that. They have made Bitcoin into a debt-based +system like any other; as long as the “exchange” holds your keys for you, +there is no obligation for them to maintain assets equal to the deposits. You +can’t prove that they are, or aren’t, maintaining sufficient assets until +after those assets are spent and the evidence appears in the block chain. + +And it’s useless for small transactions. Had it been deployed to a +market the size of, say, a college campus it could bear the load and +the bidding for block space wouldn’t exceed the value of most +transactions. But had it been deployed to a market the size of a +college campus, the small pool of miners available would make mining +bursty and unstable, and the block chain therefore not well protected +from tampering. Same could have happened to Bitcoin early on, which is +why Satoshi was mining like crazy and jumping on when needed to prop up +the block rate and back off again when the blocks were coming too fast. + +And that brings us to mining. Bitcoin mining has encouraged corruption +(Because it’s often done using electricity which is effectively stolen +from taxpayers with the help of government officials), wasted enormous +resources of energy, fostered botnets, centralized mining activity in a +country where centralization means it’s effectively owned by exactly +the kind of government most people thought they *DIDN’T* want looking +up their butts and where the people who that government allows to “own” +this whole business work together as a cartel. + +The whole idea of proof-of-work mining is broken the instant hardware +comes out which is specialized for mining and useless for general +computation because at that point the need to have compute power +for other purposes is absolutely irrelevant in having any effect on mining, +and there ceases to be any force that causes mining to be distributed +around the world. It becomes a “race to the bottom” to find where people +can get the cheapest electricity, and then mining anywhere else – anywhere +the government tries to make sure ordinary people actually get the benefit +from electricity bought for tax money, for example – becomes first pointless, +then a net loss. + +Bitcoin doesn’t scale, except by becoming the very thing it was supposed to +replace. + +Bitcoin was a Pilot system, a good first effort. It did what a Pilot system is +intended to do: show where the pitfalls lie. + +You're supposed to learn from it, then toss it out and go back to the +drawing board. + +We cannot keep pushing the prototype, we must a designing a proper production system. + +Satoshi’s main goal was to improve on DigiCash, RPOW and other similar schemes +that had a fair degree of decentralization but still relied on a central authority. Satoshi managed to solve this problem in a genius way +by combing existing technologies and understanding of human psychology. + +People had been trying to solve it for decades without any luck. People like +Wei Dai and Szabo came close but never managed to materialize their visions +(assuming they're not Satoshi). + +Bitcoin showed us where the pitfalls are, so we can focus attention on solving +them. + +Privacy, security, efficiency, and scalability are mutually opposed if if one attempts to have them all on the blockchain. For the blockchain achieves security by everyone repeating the processing of everyone else’s transactions, which is opposed to privacy, efficiency, and scalability. + +The most efficient way is obviously a single central authority deciding everything, which is not very private nor secure, and has big problems with scalability. + +If a transaction is to be processed by many people, one achieves privacy, as with Monaro, by cryptographically padding it with a lot of misinformation, which is contrary to efficiency and scalability. + +The efficient and scalable way to do privacy is not to share the +information at all. Rather we should arrange matters so that +information only goes to the blockchain to be scrutinized by +many people if the parties to the transaction have a falling out. +Which is what the Bitcoin lightning network was supposed to be, +but is not. + +Bitcoin’s pseudonymity is alarmingly weak, (though the Wasabi wallet +partially fixes this). The lightning network layer would fix this, as +well as providing instant transactions, but a true lightning network +cannot be implemented over Bitcoin as it exists today. + +A lightning network would provide instantly settled transactions and +strong fungibility. It would make bitcoins (unspent transaction outputs of +the blockchain) far less traceable, because lightning transactions happen +off chain and inherently mingle coins, thus making crypto coins fully +fungible, thus increasing their desirability as a direct substitute for cash. + +# proof of stake, Byzantine fault, and statehood + +A proof of stake currency is a corporation. Its currency is shares in that +corporation. Corporations derive their corporateness from the authority +of the sovereign, but a proof of stake currency derives its corporateness from +each stakeholder (shareholder) playing by the rules because all the other +stakeholders play by those rules. + +Which means the rules to be incentive compatible and have provide +Byzantine Fault Resistant consensus. + +This was Satoshi’s great stroke of genius. If most people follow Satoshi’s rules, everyone has an economic incentive to follow the rules. +Constructing such a set of rules is very hard. Even non Byzantine +distributed consensus is hard, because distributed consensus is very hard. + +The Byzantine Generals problem is named after Byzantium, because in the +latter days of the Byzantine empire, there were some generals who wanted +a large part of the Byzantine army defeated and annihilated so that they +could take Byzantium, overthrow the emperor, and become emperor. + +So general Malloc might send general Bob the the message: + +> facing overwhelming enemy attack, falling back. You and general Dave may soon be cut off. + +and general Dave the message: + +> enemy collapsing. In pursuit. + +With the intent that general Dave will advance and find himself cut off and isolated. + +That the messages are inconsistent is Byzantine failure, and that they are +deliberately inconsistent with malicious intent is Byzantine defection. + +The phrase “Byzantine failure” is usually used to refer to one computer in a +network sending a message to one computer that is inconsistent with the +message it sent to another computer. + +The generals need to find a consensus as to whether they are all going to +attack, or all going to retreat. They are physically separate, and messages +going between them may get lost. And some of them are traitors. The +problem of establishing a true consensus for cohesive action under these +circumstances is difficult, and the algorithms and process often hard to +understand. + +To achieve cohesive action, to act as one, all the independent actors need +to follow some process. And it can be proven that deviation from process +yields an advantage of least two to one in getting one’s way. + +This is a Byzantine fault. And if people get away with it, pretty soon no +one is following process, and the capacity to act as one collapses. Thus +process becomes bureaucracy. Hence today’s American State Department +and defense policy. Big corporations die of this, though states take longer +to die, and their deaths are messier. It is a big problem, and people, not +just computer programs, fail to solve it all the time. + +Proof of work was a brilliant and unobvious solution – but it is costing too +much, and it is slowing down the rate at which transactions can handled, +which slowness is now starting to bite hard. + +The blockdag, done right, is equivalent to the Practical Fault Tolerant +Byzantine consensus, albeit the equivalence is far from obvious, and the +blockdag is in ways simpler to understand. Practical Fault Tolerant +Byzantine consensus is arcane, but reveals a number of interesting +mathematical facts about the nature of collective action. + +## Sovereign Corporation + +[a sovereign corporation]:social_networking.html#many-sovereign-corporations-on-the-blockchain + +[that sovereign corporation]:social_networking.html#many-sovereign-corporations-on-the-blockchain + +A successful proof of stake currency would be a non state corporation, +[a sovereign corporation]. What is a sovereign corporation but a state? The +power of the US is in substantial part that it is a world currency, albeit a +major reason why it is a world currency is airsea war superiority, and as its +relative airsea war superiority power declines, its role as a world currency +declines. If the shares of a sovereign corporation took over the role of the +US dollar, that sovereign corporation would be a world power. Its power +would be in the network, as the power of the US was in the air and sea, +rather than the land. But the dollar and nukes are not the only bases of +USG power. Even more than being a financial root node, the power of +USG is a result of being the monopoly truth root node. (Via Harvard aka +the Cathedral, but including lesser official government outposts such as +the CIA.) USG establishes the world’s narratives which control what +everyone cool across the world believes – that gay marriage is justice, for +example, or that “trans” people are a real thing and not just crazy and/or +sexually deviant, or that global warming is real, human-caused, and +disastrous, or that black lives matter. A proof of stake currency is not very +functional, unless, like the Jitsi blockchain, it provides a namespace and +service, because you need to interact with peers that have authority over +the consensus – the shareholders, or their computers, need to interact with +the computer equivalent of the members of the board and CEO. A +nameservice, that unlike Domain names, cannot be seized by the +government, nor mimmed by any of a hundred organizations that have a +certificate authority in their pocket. Replacing the domain name service as +well as the US\$ would substantially undermine the US Government’s monopoly +of truth. [Yarvin analysis of bitcoin](white_paper_YarvinAppendix.html) + +## The big metadata security hole + +The necessarily cumbersome process of embedding a payment in SSL is a +huge security hole in every crypto currency, as for example when one +leases a virtual private server (cloud server) over the internet using bitcoin. +We need to replace SSL, which requires replacement of the name system +that is integrated in SSL. + +The Domain Name System means that names are ultimately owned by the +government, and the government can intercept communications to and +from such names. SSL is inherently insecure, because any entity that has +one of a thousand certificate authorities in its pocket can man in the +middle communications to and from such names. A currency cannot be +truly private, and thus is in danger of losing fungibility, if payments are +sent and received from entities with government owned names. + +Names should be owned by secrets in crypto wallets. + +# The name system is worth serious money + +Business is moving to the internet + +Increasingly, the primary assets of a business are its internet name, other people’s links to its name, and its position on other people’s pages. + +The primary asset of Amazon is the same as the primary asset of Ebay. It +silos the reputations that enable strangers to do business with each +other. You do business on Amazon, it owns your reputation. + +We need a name system that supports reviews, so that you own your own reputation. + +For the lightning network to work without central authority, we need a +cryptographic means to enforce full circle payments, so you are +guaranteed acknowledgment of your payment if it it goes through. In +which case we can have Amazon and Ebay like reviews, without a central +authority such as Amazon or Ebay. + +Amazon’s management of its primary asset is rapidly becoming worse and +worse. + +Amazon is not primarily a warehousing and delivery service, and to the +extent that is a warehousing and delivery service, it is poorly run +warehousing and delivery service. When I get something through Amazon, +it usually comes direct from the seller, not through an Amazon warehouse. +It is primarily a reputation service like Ebay suffering from the delusion +that it is a warehousing and delivery service. + +There is another option, neither FBA nor your own warehousing and +delivery service, but using some other logistics services company. +Schenker is big, precisely because they are offering to run practically +every aspect of your business, warehouse, shipping, a basic aftersales (call +center going through a script, giving replacements or credit notes for +returns), they are even offering things like billing or even building a +webshop. + +You design a product. China manufactures it for you and these logistics +services companies deal with all the details of getting the product out. So +you do not really have to have a real business with employees. Logistics is +a service, accounting is a service, marketing is a service, sales is a service, +every department can be virtualized into a service bought from another +company. + +This is probably too expensive in the long run, but very good when you +are just putting your toes into the water. Suppose you have a real actual +company in the US with employees, office, warehouse. You decide you try +to sell your stuff in France and Germany. Will you go through hiring +people and renting office and warehouse? Without even knowing if +anyone wants to buy your product there? Too risky. In 1980 it was +necessary to risk it, but not today, you can have an entirely virtual +business where every department is outsourced to local (local language +etc.) service providers, it can be started one day and liquidated the +other day if it does not work out. And if it does work out, you start +insourcing the most expensive services. + +If you take that path, you are bypassing Amazon owning your name. +Instead, the Domain Name Service owns your name. There is a lot of +money in names, and while the service is not failing as badly as the banks, +it still is mighty bad. And it is missing the capability to securely pay +money to the entity that has the name. The methods for encapsulating +payments inside SSL work, but are cumbersome and indirect, hence the +ubiquitous need to sign up, fill out a captcha, receive an email message to +confirm your sign up, click on a link in that message, before you enter +your credit card details which promptly get stolen. You should be able to +receive an invoice from `example.com` the way you can receive an email +from `name@example.com`, know for sure it is the same `example.com` you +were just clicking around in, rather than yet another scammer, and reply to +that message by clicking on a pay button. + +To accomplish this will be a great deal of work, but the foundation for +accomplishing it is that names need be on the same blockchain as cryptocoins, +and controlled by their owners secrets, rather than some central authority which +is apt to pursue its own political objectives and the financial interests of the +registrars, rather than those whose names are being registered. diff --git a/docs/white_paper_YarvinAppendix.md b/docs/white_paper_YarvinAppendix.md new file mode 100644 index 0000000..93c7f4a --- /dev/null +++ b/docs/white_paper_YarvinAppendix.md @@ -0,0 +1,1512 @@ +--- +title: >- + Bitzion: how Bitcoin becomes a state +--- +It probably won’t happen. It probably should. + +Statelike nonstates fascinate all political engineers. Can a nonstate +become a state? How are states born? What paths, if it wants to live, +should a baby state follow? + +History is full of examples. But most of them are too ancient for most +of us to parse. The vulgar and legible present is frustratingly short of +high-quality newborn states. + +One path to statehood is military. The CJNG or Jalisco New Generation +Cartel, a Mexican narco gang, recently posted a video of an impressive +military parade, with hundreds of uniformed soldiers and tens of armored +vehicles—almost ISIS tier. ISIS in its day, as Gladstone said of the +Confederacy, “had made an army, and made a nation.” + +But having made these things—a state must also keep them. Vae victis! +In the end, sovereignty is empirical. Cartels, ISIS, the Confederacy and +Rhodesia aren’t super encouraging examples. + +Yet after this list of almost-states comes very solid and palpable +nations, like Israel, Singapore and our own dear United States. +Empirically, states do get born. + +So, empirically, is Bitcoin a state? Or almost a state? Can it become a +state? Or at least, a virtual state? + +Bitcoin has not yet made a nation. It has certainly made a treasury. Its +path is not at all military—except if math is a weapon. But isn’t math +a weapon? Surely math has as good a chance as the CJNG, and you wouldn’t +want to mess with the CJNG. “Pura gente del Señor Mencho!” In fact, +Bitcoin and the CJNG even have similar market caps... + +Three attributes of sovereignty + +When we examine the early life of successful states, we see that newborn +regimes which live to become stable adults tend to focus on three +attributes: independence, legitimacy, and coherence. + +Independence means the regime does not depend on the will of any outside +power. Legitimacy means that the regime considers itself authentic and +official, and conducts itself with the dignity of a state. Coherence +means the whole state can act effectively as a single agent, with a +united regime calling on the full resources of all its citizens. + +In the ways that Bitcoin works well, it has all three. Abstractly, the +ledger itself is independent, legitimate and cohesive. But as a new +state rises, reality tests it more and more stringently. Some respond to +hard tests by becoming stronger. Others crumble. + +Bitcoin today has cracks in all three attributes of statehood. These +cracks may or may not need to be filled. Filling them may or may not +result in any kind of success. We can still describe how they should be +filled—as a theoretical lesson in political engineering. + +We can also describe how Bitcoin, with these vertical cracks filled, can +also expand horizontally. It will both increase its effectiveness, and +expand its scope. We’ll project this growth outward until Bitcoin turns +into Bitzion—a purely hypothetical utopia, which needs no fiddling +with definitions to be a clear and obvious digital state. + +Historically, Bitcoin has been extremely resistant to innovation in +governance. And, as we’ll see, the principles of effective digital +governance are extremely counterintuitive. This makes it unlikely that +anything like Bitzion will actually happen. But since it could happen, +it makes the best possible kind of thought-experiment. + +Bitcoin: from algorithm to state + +Again, while Bitcoin has made neither an army nor a nation, it has made +a treasury. This treasury is its market cap, which indeed resembles the +balance sheet of a minor state—without any of the assets. But having +made a treasury, it must keep it. + +One way to price Bitcoin is as a call option on monetary sovereignty. If +Bitcoin becomes a state, or merges with a state—becoming either a +standard hard currency, or a national or global hard currency, de facto +or de jure—we can expect the option’s price to considerably increase. +If this outcome grows less likely, the option’s price will fall. + +Are “becomes a state” and “becomes a standard” really equivalent? +Certainly not without an argument—and that argument is the crux of +this essay. + +The difference between Bitcoin as a standard, and Bitcoin as a state, is +that a state has a government. The argument: to prevail as a monetary +standard, Bitcoin must up its game. Its most effective way to up its +game is to form its own government. But once it forms a government, it +is a state—contradicting its libertarian, decentralized ideology. + +Interestingly, the 13 states faced a similar question in the mid-1780s. +The birth of a centralized national government contradicted much of the +ideology of the Revolution. Yet the Congress of the Confederation was +such a shitshow—almost an 18th-century UN—that the +pre-Constitutional United States has been almost written out of history. + +The case for Bitcoin as a state is that a monetary standard with a +government can work much better than one without one. Bitcoin already +has a government; but it is (mostly) algorithmic. A human government can +work much better than an algorithmic one. + +All Bitcoiners agree that not having a government is an advantage, since +a government is inherently a central point of failure. While this is the +truth, it is not the whole truth. + +Having a central decision point is also an advantage. As hackers, we +hate choosing between two such advantages. We would like them both. And +if we cannot have all of both, perhaps we can at least get most of both. +Such is the good old art of the tradeoff. + +Logical case for a Bitcoin government + +Why does Bitcoin need a (human) government? + +Bitcoin at present is a state without a government. If there is one +thing political scientists know, it is that there is no such thing as a +state without a government. When a state tries to have no government, it +always has something. + +A blockchain is statelike because independence is the whole point of a +blockchain. Bitcoin or any other blockchain has two layers of +governance: serialization, which is inherently algorithmic (execution of +the ledger protocol); and oversight, which is inherently human +(management of the protocol and its reference implementation). + +Government is human; it means oversight. Bitcoin needs a human central +government because centralized governments are more effective than +decentralized governments. An effective government can be trusted to +choose the best serialization algorithm; so algorithmic serialization is +just an aspect of human oversight. + +Bitcoin today has serious oversight problems. No one even agrees on who +has the responsibility of overseeing the blockchain—miners, nodes, or +developers? This is why Bitcoin is a state without a government. + +The only correct answer is hodlers—who are the only group not now +represented. They could be represented only by constructing a central +government that represents them. This is just what they should +do—though it would change the whole story of Bitcoin. + +But since Bitcoin’s independence today is secured by decentralization, +any central government would need a different security mechanism to +secure its independence. One such mechanism is pseudonymity. + +Pseudonymity has many technical limitations and is a broadly inferior +substitute for decentralization. Engineers, in pursuit of the perfect +tradeoff, must often use limited and inferior substitutes. + +Intuitive case for a Bitcoin government + +Ultimately, what is Bitcoin? It is a set of human beings with private +keys, who all use a technical tool that lets them automatically agree on +digital facts. Bitcoin is the people, their keys, and the ledger—the +public record of how much money every key owns. + +Bitcoin’s algorithm and infrastructure are a cool tool that these people +have used to act coherently for their mutual benefit—without uniting +or cooperating in any other way. Nakamoto consensus is a rope of sand +that can actually pull a load. The hodlers, a completely atomized +non-society, nonetheless act coherently when they save in coins. + +Collective coherent action for mutual benefit is a tool of essentially +unlimited power. Algorithmic, decentralized governance has unlocked a +tiny slice of this power. The normal way in which sets of human beings +undertake collective coherent action is through central organization. +This normal government can unlock the rest of the pie. + +Bitcoin, by producing governance without a government, is the exception +that proves the rule. What this exception tells us is that whatever +Bitcoin has achieved so far, these achievements are a small fraction of +what it could achieve with a government. + +While this magic rope of sand is cool and all—while Bitcoin could not +have been born without it—what about switching to, like, a regular +nylon rope? + +Must this be done? Why must it be done? Can it be done? How would it be +done? If you have a moment or two, dear reader, we’ll answer these +questions and more. + +Bitcoin’s oversight is a mess + +Bitcoin has a reputation for terrible oversight. This reputation is not +the fault of anyone involved. It is Satoshi’s fault, for leaving +oversight to anarchy. It is not even clear which group holds the power +of oversight: nodes, miners, devs, hodlers or users. (Satoshi was a +genius; but not a god.) Fact: anarchy can make anyone behave badly. + +The only proper interpretation of Bitcoin oversight is that Bitcoin’s +principals are its hodlers. Bitcoin is the ledger. The blockchain is an +agent that maintains that ledger. The Bitcoin infrastructure must be +operated in the exclusive service of the ledger, just as any agent must +act in the exclusive service of its principals. Had Satoshi made this +clear, or even built a mechanism to enforce it, much suck might have +been averted. + +Anarchy is not the end of the world. But anarchy still sucks. +Objectively minor meta-decisions, like setting the block size, turn into +Icelandic sagas. Larger changes are inconceivable. Bitcoin, the OG of +blockchains, is the couch potato of blockchains. Everyone knows this, so +there is no reason to dwell on it. + +Bitcoin’s serialization isn’t perfect, either + +Bitcoin’s serialization—maintaining the integrity of the ledger—has +a perfect record. No power has ever infringed the independence of +Bitcoin. Yet hash proof-of-work is not perfect. It is not perfectly +secure; and it is extremely expensive. + +Today, 65% of the world’s mining capacity is in China. So it would be +straightforward for the Chinese government to squeeze Bitcoin. The CCP +has not felt the need to try—not because Bitcoin is too strong to +attack; because it is not important enough to attack. This is not +independence but the illusion of independence. If Bitcoin is indeed an +call option on sovereignty, these kinds of illusions are dangerous. + +The cost of mining is the price of independence. Currently this price is +huge: perhaps 5% of market cap per year. Imagine gold had a high vapor +pressure—every year, 5% of every Krugerrand boiled off into the +atmosphere. In this alternate universe, no one would even think of using +gold as money. + +Bitcoin hodlers do not think they are paying this immense wealth tax. +They are. And it seriously hampers the competitiveness of Bitcoin as a +currency. Let’s look closely at how this works—these details matter. + +The price of Nakamoto consensus + +Mining revenue comes from fees and block rewards. Block rewards are +straight-up monetary dilution. Proper accounting must measure ownership +of any equity or cryptocurrency not as the number of units, but as the +fraction of units outstanding. So creating 1% more units is exactly the +same as a 1% wealth tax across hodlers—though easier to implement. + +Block-reward dilution is bounded, and fees are not dilutive. But fees +have the same direct impact as rewards: since they need to pay for +electricity, they are sold for fiat. Miners are generally not hodlers. +So \$5B of mining means \$5B of selling. Obviously, the seller is the +natural enemy of the hodler. + +As a currency—a stable bubble—Bitcoin acts like a battery or a +pressure tank. Its market cap is the pressure in the tank, or the energy +in the battery. To counteract \$5B in annual outflow and keep the price +flat, Bitcoin needs \$5B of annual inflow. + +This mining leakage gets heavier as the price rises—a vicious +stabilization loop which tends to cap the price. This is how the cost of +governance by Nakamoto consensus costs the hodlers money—and tends to +make hyperbitcoinization impossible. + +Imagine Bitcoin’s price if it had no mining leakage—meaning zero +automatic outflow. Everyone selling would be a defecting hodler. There +has never been a large number of these, even in the worst crashes. + +Algorithmic governance was essential in Bitcoin’s childhood. It created +a system that was owned by no one. From its genesis block it had this +essential quality of a state. Nakamoto consensus made Bitcoin +independent from day one. No other design could have achieved this +result. + +But statehood has its own ruthless laws. No startup can cling to its +teddy bear. All must change as they grow. Some of those changes will +even be in engineering. Today, the cost of Nakamoto consensus is keeping +Bitcoin from advancing as a monetary standard; and its unbalanced +decentralization leaves it under China’s thumb. + +Bitcoin imperialism + +By all except Bitcoin’s proof-of-work purists, it is generally admitted +that proof-of-stake is the path forward in both layers of +governance—serialization and oversight. + +Many alternate chains already use POS—an unfortunate acronym—as well +as smarter agreement protocols, chain sharding, and all kinds of other +fun stuff. (In fact, it would be hard to boot up a new proof-of-work +chain today—it would be attacked as soon as it was worth attacking.) +Why not just switch to a newer-generation blockchain platform? + +Alas, none of these alternate chains can solve the problems with +Bitcoin. Instead, Bitcoin will have to defeat them and erase them +utterly from the earth. + +Most Bitcoiners believe this victory is in some sense inevitable, and +will happen on its own. Is it happening on its own? Or is anyone +thinking about how to make it happen? This is the mindset of Augustulus, +not Augustus—a poor start for any empire. + +For it does need to be made to happen. It is not just a matter of +waiting for altcoins to go away. Rather, Bitcoin needs to actively clean +them up. + +This uber-hardcore flavor of Bitcoin maximalism—call it Bitcoin +imperialism—would probably shock even most Bitcoin maximalists. But it +is an inherent consequence of the right theory of why Bitcoin is worth +anything at all. + +Parallel ledgers considered harmful + +An alternate chain is a parallel ledger. All ledgers must be unified. +Any partition of monetary energy makes not only each side of the +partition, but both sides put together, a worse bet than the unified +ledger. And there is no stable equilibrium between the sides of the +partition. + +In any real economy, at least one asset must be overvalued. That asset +is money. But given one money that is both standard and perfect, the +only stable equilibrium is one in which there are no other bubbles, and +all other assets are priced by yield in that standard. Any such asset +which has no use has no value, and will sell for no price. + +There is one mathematical revolution in money. Forks and altcoins are +just bad for it. The revolution in money will be complete when there is +one standard digital money. This means one standard ledger—the +“Highlander principle.” + +This standard ledger has to begin with Bitcoin. Bitcoin is the gorilla. +In any world where Bitcoin ends up worthless, lesser cryptos end up +worthless too. So there are two ways to resolve any parallel ledger: +merge it with Bitcoin, or send it to zero. + +Generally, any alternate chain worthy of any notice at all has some +technical advance, purported or real, over Bitcoin. This may be privacy, +computation, finalization time, mining leakage, Byzantine security, or +some other useful area of improvement. + +But the real competition is monetary competition between ledgers. This, +it is hard to imagine Bitcoin losing. Muscle matters more than mind. The +chimpanzee can ask for a banana in sign language. The gorilla is three +times as big as the chimpanzee. + +The traditional Silicon Valley approach to this conundrum is for the +gorilla to just kill the chimpanzee and steal his banana. Sometimes, the +chimp can jump the gorilla from behind, and get him with a rock. + +Neither outcome seems quite right. A cryptosystem that aims at the +legitimacy of a state cannot behave like any kind of ape. We’ll return +to this issue with an actual plan. However: there can be only one. + +Proof-of-stake and political engineering + +Altcoins delenda est. But while their parallel ledgers must vanish with +the Oscans and the Samnites, as their genuine engineering innovations +are shamelessly looted like Houston lifting the Nazi space program, many +have also innovated in governance. Scholarship asks us to start by +noticing their work; and we will start by following it. + +While proof of work is one thing, proof of stake is a huge family of +things. This family is united by one general principle. The general +principle of proof-of-stake is that the ledger belongs to its +owners—who must therefore be charged with governing it. + +That principle is sound. Bitzion uses it. But it is as general as the +principle that the nation must be governed by its citizens—and as +vague. Even if everyone agrees on democracy, there are infinitely many +possible democratic constitutions. + +Proof-of-stake is not democratic, of course, because power is by wallet, +not by head. The ledger belongs to its owners, in proportion to their +ownership. The ledger is managed for the benefit of its owners, in +proportion to their ownership of it. + +The proper term for voting by wallet is plutocracy. That this is a +pejorative makes it all the more delightful. But a plutocratic republic +is still a republic—and faces the same general challenges in political +engineering that vex any democratic republic. Here we will say democracy +when we mean plutocracy. Since head-counting is impossible, there is no +ambiguity; and euphemisms are nice. + +Most proof-of-stake altcoin designers are not trying to innovate in +political theory. But their political engineering is still +awful—because it comes from the awful fashions of an awful period, in +which they happen to live. + +Why would they realize that following the fashionable ideas of their era +would produce awful results? If everyone is using the same bad ideas, +everyone’s results will be awful, and no one will notice. They will just +think the problem is very hard—which it is. + +Let’s look at how to take this principle, proof-of-stake, in an unusual +direction. Instead of building ideal governance structures, let’s +reengineer existing governance structures. + +These designs are not fashionable—since they predate the 20th century. +No one would invent them now. But everyone still inhabits them. Isn’t +that weird? + +If our fashionable ideals are realistic, this plan will just mean +inheriting ancient cruft. If today’s ideals are not realistic, it may +produce dramatically improved governance. + +From Bitcoin to Bitzion + +To turn Bitcoin, the ledger and blockchain, into Bitzion, the digital +state, we adopt the plutocratic principle of allocating authority and +responsibility by stake. We declare that Bitcoin is the ledger, not the +protocol or the infrastructure. We declare that the ledger is owned by +the hodlers; it should be governed by the hodlers, for the hodlers; and +we, the hodlers, will cultivate it into a virtual state that will +astonish the world. + +Although it uses proof-of-stake, Bitzion makes three large theoretical +adjustments to the viewpoint of most classic altcoin-style POS +governance designers. These are not technical changes; but they lead +toward completely different technical challenges. + +First, most designers focus on ledger governance (algorithmic +serialization); we’ll focus on systems governance (human oversight). + +Second, most designers consider direct democracy the ideal form of +democracy. Actually, indirect democracy creates much more effective +governance structures. + +Third, most designers think decentralization is the only tool for secure +independence. But there is another such device: pseudonymity. + +Serialization, oversight, and trust + +Serialization means recording a fair, unified, reliable and independent +ordered list of transactions. Serialization is the job of every +blockchain. Centralized, trusted serialization is easy. Decentralized, +untrusted serialization is hard. So serialization is what most +cryptosystem architects think of, when they think of governance. + +Of course, humans cannot be in the serialization loop. Serialization is +inherently algorithmic, if just because it has to be fast, secure and +scalable, as well as fair (it does not censor or reorder transactions) +and independent (no outside force can influence it). + +Here is a small philosophical crack in the thinking of most +cryptointellectuals. Why is Bitcoin like a state? Because it is fair, +reliable, and independent. Why is it fair, reliable, and independent? +Because it is decentralized. We all agree on this, but— + +It is easy to overlook the small but important fact that what users +value in Bitcoin is not its decentralization. What users value is a +consequence of its decentralization: that Bitcoin is fair, reliable and +independent. More broadly: that it is legitimate, effective, and +sovereign. Still more broadly: that it is good at being like a state. + +The Bitzion thesis is that Bitcoin has ways to become much better at +being like a state. If Bitcoin is an option on statehood, this should +make its price much higher. Any tactical path that produces this result +must be at least worth considering. + +It is essential that we understand this difference between cause and +consequence. We trust that Bitcoin is fair, reliable, and independent, +because math proves that its decentralized protocol makes it so. + +Or so the nerds tell us. Most of us cannot do the math—we probably +don’t even know what an elliptic curve is. Byzantine what? But we trust +the nerds. Trust is beautiful. + +Suppose we trust that Bitcoin is fair, reliable, and independent, +because it is run by the god Apollo. Apollo keeps the ledger on a big +SCSI RAID rack in his basement, with a nest of ribbon cables running +into a Compaq 386 PC-XT. It’s as fast as Nakamoto consensus, and even as +reliable—because he’s Apollo, and Apollo is perfect. + +Or so the priests of Apollo tell us, and we trust the priests. Trust is +beautiful. The only relevant statement in both cases is: “we trust that +Bitcoin is fair, reliable, and independent.” The cause of trust, whether +decentralization or Apollo, makes no difference. What counts is the +trust itself. (It’s better if the trust is rational, though.) + +Besides serialization, which is governance in the proper blockchain +sense, all chains have parameters; all chains have protocols; some have +full operating systems. The power to update these rules, values, and +systems is oversight. + +Oversight is an inherently human function—though the identity and +procedure of those humans may depend on the state of the ledger. Since +all blockchains require some oversight, there is no choice but to trust +the oversight nerds. Trust is beautiful. + +As previously mentioned, oversight in Bitcoin has historically been a +shitshow. In POS altcoins, extending voting by stake from serialization +to oversight is a natural idea. But that makes oversight a natural +afterthought. + +Bitzion attacks the same problem from the other end. It reasons: +effective oversight can choose the best serialization algorithm. Get +oversight right, and oversight will get serialization right. + +Again, oversight is just government. The solution to the serialization +problem is: form an effective government, and trust it to develop and +manage an efficient serialization network resilient to attacks by global +adversaries. + +Animal control is a natural function of physical government. +Serialization is a natural function of digital government. Both powers +can be extremely dangerous if misused. + +Malicious serialization cannot forge transactions, but it can censor and +revoke them. Malicious animal control can dart and capture humans, +bundle them into white vans, do some processing in an underground lab, +and peddle the organs on the dark web. Don’t a lot of people just +disappear? What do you think is happening to them? + +When citizens of a physical government think about how to keep bears and +leopards off the street, without opening the door to rogue dogcatchers +who secretly moonlight as human organleggers, they tend to focus not on +this specific state function, but on delegating their consent and +loyalty to a whole government which is stable and sane. Citizens of a +virtual government also trust that their stable and sane government will +operate a secure and efficient serialization mechanism. Again, trust is +beautiful. + +Since Apollo isn’t returning our calls, any serialization network will +almost certainly be decentralized. The government that creates it, since +it must be highly effective, is likely to be... centralized. Obviously, +to any cryptointellectual, this is a huge red flag. + +By default, any central organization is a big step away from +sovereignty. Independence is one of the three attributes of sovereignty. +A decentralized non-organization is very hard to pressure. By default, a +centralized organization is very easy to pressure— + +By default. To apply the lesson we just learned, decentralization is not +a design goal of a cryptostate. Decentralization is a mechanism. The +purpose of this mechanism is independence. The purpose of independence +is sovereignty—or at least, monetary sovereignty. So as engineers, we +are free to use other mechanisms of independence. + +But independence, in Steve Jobs’s words, is a feature—not a product. +Let’s design our new regime, then figure out how to make it independent. + +Indirect democracy: magic in plain sight + +The main assumption that makes proof-of-stake governance ineffective is +the almost wholly false assumption that direct democracy is better than +indirect democracy. + +This is a weird belief in a civilization whose realization of democracy +is, and has always been, indirect. Democracies are rare in history. But +direct democracies are extremely rare. Nor has the judgment of the past +given them great reviews. When Athens was a direct democracy, it was the +assembly who voted to execute Socrates. + +But direct democracy is more consistent with 20th-century ideology. We +feel, quite reasonably, that indirect democracy is one of our many +quaint legal anachronisms. + +Obviously not everyone had Internet in the 1780s. They had to use +dialup, or horse-drawn ballot wagons, or something. Even now, attention +is in limited supply—but, at least in theory, every citizen should +have the right to vote directly on every decision. Also, politics sucks +and politicians suck. That’s the normal perspective. + +We love democracy and hate politics. When we realize that our new +cyberstate can have democracy without politics or politicians, since it +can count zillions of ballots in zilliseconds, this seems like an easy +win. We don’t even think of doing it the old way, even though +everything in the real world works the old way—indirect democracy. + +Actually, direct democracy sucks. Virtual states will be weak and +passive at best, chaotic and ineffective at worst, until they somehow +reinvent indirect democracy. + +A jury is not a government + +All systems of direct democracy, including all proof-of-stake designs +that I know of, use their stakeholders as a jury. A jury is a group of +people who vote to ratify or reject decisions framed and proposed by +some other process. + +A legislature is a jury in which the members themselves have the right +of proposal. All actual legislatures either delegate this power to +external structures, or evolve internal structures (a cabinet in the UK, +committees in the US) which are hierarchical. + +Direct democracy just expands the jury to the whole electorate. While +nowhere in the world, except perhaps some Swiss cantons, even pretends +to be governed by direct democracy, many jurisdictions have adopted the +referendum as a decorative bauble. + +Digital governance can trivially borrow the referendum, and count itself +democratic. A proof-of-stake blockchain can use a jury, full or +sortitioned, for serialization and/or oversight. Even Bitcoin oversight +only differs from POS in that the jury is the nodes. + +Actually, juries suck. The jury is the most passive and reactive form of +government that can be devised. It is so reactive that it is not even a +complete government—it lacks any real executive arm. + +Any vacuum in sovereignty will fill itself. A sovereign jury evolves its +own informal leadership. When Parliament replaced the Crown as the locus +of English sovereignty, to replace King and Privy Council, it evolved a +Prime Minister and a Cabinet. This was not anyone’s plan, or anyone’s +theory—it happened only because it had to happen. + +Luckily the delegation from MPs to this pseudo-king, the PM, was +well-structured. Normally, the outcome of the sovereign-jury design is +an illegitimate and informal leadership agency, plus a jury which at +best is a rubber stamp and at worst a zoo. + +What the jury design is best at: making large numbers of people feel +important. As members of the jury, everyone can feel in charge of +everything. This potent narcissistic energy is the principal marketing +advantage of direct democracy—and what better moment for it to shine, +than our historic golden age of narcissism? + +If a machine was designed to minimize generated power, while maximizing +perceived power, it couldn’t be better designed than a gigantic jury. +Let’s illustrate the difference with a design which minimizes perceived +power, and maximizes generated power. + +The political amplifier + +Indirect democracy is not an anachronism. The reason that indirect +democracy is more common than direct democracy is that delegation is +amplification. Indirect democracy generates concentrated, coherent +power. Direct democracy does not. + +Delegation is amplification because delegation concentrates and focuses +trust—a laser for trust. A weak, coherent laser is stronger than a +strong, incoherent flashlight. You can look at a 40-watt bulb. A +40-milliwatt laser will rip a new asshole in your eye. + +A mob, in which every participant retains freedom of action at all +times, is much less effective than a much smaller army, whose soldiers +delegate their freedom of action upward to one commander. The difference +between a mob and an army is so strong and clear that it suggests the +analogy of turbulent versus laminar flow. + +In indirect democracy, citizens delegate power to politicians. Citizens +cannot act directly, only by voting. There are fewer politicians than +citizens; so elections must concentrate power. But voters can use power +only by deciding whom to give it to. + +Even in our indirect democracies, though their rational choice is +between politicians, voters still think more about issues. They seem to +feel they can somehow act through the politicians they elect—though +there is little evidence that this works. On paper, their democracy is +indirect. In their hearts, it is direct. + +So while they love democracy, they hate politics—or to be exact, they +hate their enemies’ politicians, while merely disliking their own. Can +we blame blockchain designers for starting from an ideal, rather than +reverse-engineering this mess? + +In direct democracy, where there is no delegation, every citizen can act +directly on every issue. There is no need for an illusion of +indirection. You actually matter. Your voice is heard. And there are no +politicians. + +But you are collectively weak—because you, the people, are a mob and +not an army—a bulb, not a laser. If an electorate as a whole does not +matter—just because it is not strong enough to keep its dominance, or +even relevance—no one’s voice matters at all. + +So the narcissistic desire for power, which is the desire to feel good +and powerful, and the ethical desire for power, which is the desire to +see power do good, are generally opposites. Almost always, the strongest +thing to do with power is not to use it, but to give it +away—concentrating it in the hands of someone who can use it more +effectively. + +So when we measure the power of an indirect democracy, we measure the +amount of power that citizens delegate to politicians. This is the +delegation efficiency of the election. If this is 1, the politicians +have the unconditional loyalty of everyone who voted. If it is 0, voters +voted for random, ironic, frivolous reasons, and didn’t actually care at +all. And again, delegation is amplification. + +Here’s an example of how constitutional designers can adjust this knob: +the length of a politician’s term. In the electronic world, terms could +last any length; they could as easily be a week. When we shorten terms, +does amplification decrease, or increase? + +It decreases, of course. When we elect a President every week, we the +citizens are saying to the election’s winner: we trust you to serve, +but just this week. We have our eyes on you! Right now, we definitely +think you’re hot stuff! See you next Sunday! + +Obviously, if you have to run for President again next week, and you +have no way to be sure anyone won’t change their mind about you, you +are on a pretty tight leash. The shorter your chain, the weaker you +are—simply because the voters who voted for you made a much smaller +commitment. + +So shorter elected terms mean weaker delegation, which mean weaker +amplification. Let’s go to the other extreme. We’ll elect a President +not for four years, but forty. Or why not\... life? + +If we mean our votes sincerely, when electing a President for life, we +are making a permanent political promise—half election, half marriage. +The candidate for whom we are voting can expect our unconditional and +lifelong loyalty and obedience. (Of course, our Supreme Court, for whom +no one ever voted at all, is in exactly this position.) + +Surely any President-for-life will be much, much more powerful than any +President-for-a-week. So the delegation is stronger. So this design +gives the voters themselves more power—since the more power their +delegates have, the more the voters have. + +By going from an election every four years to one every forty years, we +concentrate the power of ten elections into one. At this point perhaps +the awful irony is almost clear. But let’s go even farther. How would a +constitution generate maximum amplification? How would we design the +most powerful possible election? + +This election would elect not many politicians, but just one. No +organization is so big it needs multiple leaders—in fact, the bigger +it gets, the more important unity becomes. So the President would have +unconditional control of the whole government. Crazy! Also, optimal for +that attribute of sovereignty we were talking about—coherence. + +And there would be no term at all, not even life. The President would +designate a successor. This would be the most powerful election +ever—and, of course, the last. Until the next revolution, of course. + +So... we’ve just shown that absolute monarchy is the strongest form of +democracy. An interesting result, n’est ce pas? Wily intellectuals have +all seen proofs that 2+2 is actually 5. They may amuse themselves by +checking the above logic once or twice. + +It is also worth noting that the constitution this election would create +is merely that of the Roman Empire under the Antonine emperors—which +Gibbon, 1500 years later, felt was the best government humanity had ever +experienced. + +Amplification versus sensitivity + +This is not a suggestion that either Bitcoin or America should adopt the +constitution of the Antonines. Either could certainly do worse; both can +probably do better. But we are just illustrating a theoretical tradeoff +that often goes unobserved. + +This is the tradeoff, in any republic, between sensitivity and +amplification. Sensitivity is how accountable the leaders are downward +to the will of the voters. Amplification is how much energy the voters +delegate upward to the leaders. + +Our one-time election maximized amplification and minimize sensitivity. +Our weekly elections maximized sensitivity and minimized amplification. + +When designers opt for high sensitivity, their decision suggests a +powerful fear by themselves or their customers of a failure mode in +which agents betray principals. Government is a dangerous agent and must +be tightly leashed. + +But when principals micromanage their agents, the leash is so tight that +it chokes—and the agent cannot serve at all. You can walk a dog on a +leash. You can’t hunt with a dog on a leash. The optimal leash is the +leash that does not even exist, except in the unlikely case where it is +actually necessary—and in that case, the leash is inexorable. + +We are the hodlers. Government is our dog—but a dog that must hunt. +Fortunately, we have the technology to let a dog run free and wild as a +wolf, until its owners blow a magic whistle. This recall-only republic +will be low in sensitivity, high in amplification; but its sensitivity +is not zero. It is not (quite) the constitution of the Antonines. + +Pseudonymity versus decentralization + +Bitzion needs to retain Bitcoin’s independence and would ideally +enhance it. How can a centralized organization achieve this? + +Simple: by replacing open decentralization with closed pseudonymity. +Decentralized open networks are hard to kill because they are built out +of many small fish, each hard to catch. Closed pseudonymous networks can +be built out of a few big fish, each almost impossible to catch. +Satoshi’s identity is still a secret—so it can be done. + +Again: decentralization is the means; independence is the end. +Pseudonymity is trickier and more complex than decentralization. A +gasoline engine is trickier and more complex than a steam engine. It has +many more ways to go wrong. Sometimes the simplest answer ends up being +the right answer. But you don’t drive a steam car. + +Independence is inherently a security problem. In all security problems, +we need to begin with a threat model. + +The threat is a global adversary—a government or quasigovernmental +force—which wishes to involve its will in our governance—to pressure +us. Independence is boolean. Unless this pressure has absolutely no +effect, our independence is compromised. History suggests that +independence, once compromised, is quickly rent asunder. + +The power available to this adversary is not a constant. It will tend to +increase over time. It also increases as a function of the nefariousness +of its prey. Therefore, one core security measure is +legitimacy—maximizing compliance, both legal and social. Any future +state demonstrates heaven’s mandate by complying both with its own +ethical standards of tomorrow, and the bizarre whims of today’s armed +clowns. + +Suppose our closed pseudonymous system is a small committee—with +between five and fifty members. Groups this small, if well-organized, +can act with perfect cohesion. Ownership of a pseudonym is possession of +a key, ideally sealed in a physical object—like one of Tolkien’s magic +rings. Strong pseudonyms are synthetic—random syllable or word +strings, never nicknames. + +Can a global adversary find any of the keybearers—assuming all their +opsec is strong, and so is their compliance? Its best bet is to trace +their patterns of communication. So their best bet is to communicate as +little as possible. It is quite hard to trace occasional signed +statements dumped onto the Internet. + +And when keys are literally passed from hand to hand, tracing custody +chains is hard, even for a global adversary. Even if the first link in +the chain is a known individual, one link that refuses to break, cannot +be found, has randomized the transfer, etc, sends the adversary back to +looking for the physical origin of an encrypted communique. + +Can pseudonymity deliver a more secure level of independence than +decentralization? With a majority of mining capacity in China, +decentralization is not very secure at all. What if all these keybearers +were in China? Could they still communicate securely? It would be +difficult—but possibly doable. + +In today’s West it is certainly doable. For economic reasons, China is a +great place to mine Bitcoin—Satoshi never realized that the easiest +way for a state to capture Bitcoin would just be to suck in the most +miners with the cheapest electricity. China is not a great place to be +part of any kind of secret organization, however legit. So probably none +of the keybearers would be there. + +What if the Western Internet became as well-controlled as China’s? +Politically, this is not utterly implausible. Technically, it would +involve Western governments being as competent as China’s. This couldn’t +happen right away. There would be time to evade. And it is impossible to +devise a plan so perfect that it always wins without a struggle. + +Bitzion in a nutshell + +Bitzion is a design for a human government for Bitcoin. The goal of the +design is to be independent, legitimate, and coherent—as befits a +virtual state. + +(“Zion” can be interpreted in many senses—Biblical, Israeli, Mormon, +Wachowskian or, of course, Rastafarian. The analogy to the LDS may be +especially apropos.) + +The fundamental principle of Bitzion is that Bitcoin is the ledger. +Therefore, Bitcoin is the property of its hodlers. It must be governed +with their exclusive consent, and in their exclusive interest. No one +else gets a vote—“no god or government.” + +Miners, nodes, developers, and even algorithms are infrastructure that +serves the needs of the hodlers. Their opinions will always be +interesting, relevant and valuable. They are not stakeholders in any +legitimate decision process. + +All power to the hodlers! To paraphrase John Jay, a currency must be +governed by those who own it. Such is the creed of Bitzionism. Until +this creed is widely known and broadly agreed, Bitzion is not ready to +happen. + +Technically, Bitzion must begin as a fork of the Bitcoin ledger. It +cannot work as designed unless it is born as the dominant +fork—ideally, roughly the relationship between Ethereum and Ethereum +Classic. + +This requires Bitzion to achieve human reputational consensus, or +something like it, before it can even launch. At best, this would be +long, tough sledding! But afterward, the rest of the plan is actually +quite realistic. + +The hodlers declare their sovereignty. To take what is theirs, they +seize the ledger. They own and will now control it. The Bitcoin ledger, +once managed by a Mos Eisley of miners, nodes, developers and +algorithms, will now be managed by a new human regime, whose exclusive +mission is to serve the hodlers of Bitcoin. + +This new regime, Bitcoin’s new government, is Bitzion—still the one +and only Bitcoin. Samuel L. Jackson would like to know if you have any +questions about the new regime. + +To govern Bitzion effectively as possible, the hodlers delegate their +power to a board of trustees, whom they elect. These trustees inherit +all the power of the old government. + +They cannot forge transactions. They do not filter transactions. They +can and may do anything else—including creating any number of coins. +Satoshi’s constitution is dead and overthrown. The hodlers, acting +through their trustees, are now fully in the saddle. + +There are two reasons for the trustees to create coins. Regular: as a +dilutive tax, to fund the normal operations of government. Special: as +an assessment to fund some one-time event—which usually means cleaning +up some altcoin. + +Every revolution has a financial premise. The financial premise of this +revolution is that one of Bitzion’s first tasks will be a dramatic +reduction in mining leakage. Since Bitcoin is currently leaking about 5% +of its market cap per year via mining, it seems fair to assume that +grateful hodlers would be happy to approve a dilution tax up to a tenth +as high: 0.5% of market cap, more comparable to an expensive gold vault. + +This serious budget is wholly under the control of the trustees. But the +trustees take no operational role in spending it. Instead, they delegate +all their power to a single coordinator, who coordinates all the +operations of Bitzion. + +The trustees’ job is: hiring and firing the coordinator; reviewing the +coordinator’s performance; and approving regular and special tax +requests. The coordinator can request and spend regular and special +taxes. The trustees do not micromanage the coordinator. The coordinator +has no secrets from the trustees. + +And the hodlers still have a job. If they become unsatisfied with the +performance of their government, they can recall it—electing new +trustees, who will choose a new coordinator. To trigger a recall, they +need a quorum defined as a percentage of the stake that voted in the +previous election. + +The trustees are securely pseudonymous. No one knows who they are. No +one can pressure them. But given the legitimacy of Bitzion and its +commitment to compliance, the coordinator’s office must be completely +open, public, and transparent. + +But since the coordinator works for the trustees, the coordinator is a +disposable part. All kinds of powers can pressure the coordinator. The +coordinator’s job is to either perfectly resist any such pressure, or +resign—but never to bend. A coordinator who bends will just get fired +by the trustees, a very bad and shameful outcome. + +The trustees can spin up a new coordinator in another jurisdiction, or +even fall back to pseudonymous coordination. This defense works best +when the coordinator sources as much labor as possible not from +employees, but from pseudonymous contractors. + +So the government of Bitzion has three teams: accountability (hodlers +and trustees); coordination (mahogany desk, press releases, lawyers, +interviews, conferences); and production (coders, writers, designers, +etc). + +The accountability and production teams are pseudonymous (its members +never leak their identities, even to each other); the coordination team +is transparent (its staff are regular employees and have their pictures +on the website). + +The security of trustee identity is mission-critical. The production +team can be more casual. As a labor pool, it will always experience +attrition. But any producer or trustee whose identity is revealed must +quit and cannot return, even under a new pseudonym. + +The trustees’ job is: hiring and firing the coordinator; reviewing the +coordinator’s performance; and approving regular and special tax +requests. (“Tax” means dilution.) The coordinator can request and spend +regular and special taxes. The trustees do not micromanage the +coordinator. The coordinator has no secrets from the trustees. + +And the hodlers still have a job. If they become unsatisfied with the +performance of their government, they can recall it—electing new +trustees, who will choose a new coordinator. To trigger a recall, they +need a quorum defined as a function of the stake that voted in the +previous election. The trustees can call a new election, too. + +The sophisticated observer will recognize that this version of indirect +democracy, while owing little to historical Anglo-American political +constitutions, is kind of a blatant ripoff of Anglo-American corporate +governance. + +This is unsurprising in an architecture designed to get things done. No +one says our corporations aren’t sharks. Which do you want on your +side—a shark, or a hippo? As an ex-CEO, the joint-stock company has an +antiquated, paper-belt feel in many ways. If there was anything much +more powerful, some asshole would have invented it. + +Hyperdelegation + +That’s the whole constitution, except for one little detail: how the +election works. Again, Bitzion is designed to have only one election +ever, unless the hodlers or the trustees get unhappy with its results +and decide to press the reset button. There are a lot of ways to elect +trustees, but here is a clever one. + +A hyperelection is an election in which anyone can vote for anyone—so +long as they’re not creating a loop. At the start, you have one election +point for each coin you own. If you vote, you have zero points, and all +your points, both from your own coins and from whoever votes for you, +flow to whoever you vote for. The purpose of this design is to channel +trust, respect and loyalty upward to the most trustworthy, respectable +people. + +At the end, the N leaders become the N trustees. The trustees’ votes are +weighted by the number of points they got. During the election period, +the scoreboard is public, and anyone can switch their vote at any time +(or this can be quantized into rounds). After the election, any trustee +can resign their key to anyone else. + +As a total noob, the way you participate in a hyperelection is: (a) +register your wallet; (b) figure out who you know that knows the most +about Bitcoin; (c) ask that person for their pseudonym; (d) vote for +them. If you want people to be able to vote for you, also register a +pseudonym. If you have to, vote for a public candidate; but it’s always +better to just vote for a friend. + +If you are a serious candidate, campaign. Publicize your pseudonym, with +or without your real name. Speak publicly to the hodlers; speak +privately with whales. During the election, if you’re not winning, vote +for someone you admire who could win. + +The best precedent for this kind of social gameshow mayhem is an early +20th-century American political convention. In the end, most peoples’ +votes will end up going to one of the winners, through a path of mostly +ascending respect. + +Initial abnegation + +But if the election leverages existing reputations by letting pseudonyms +campaign under their real names—how pseudonymous is it? It isn’t, of +course. + +We fix this problem with another rule: after choosing the first +coordinator, all the initial trustees resign immediately—each giving +their key to someone they know, who they think will make an even better +trustee. So the first board of trustees after an election has two jobs. +They select both the first coordinator, and the second board. + +The result of initial abnegation is a political structure in which the +second board did not choose the coordinator, so feel no attachment to +that choice; the first has no lasting power, so cannot have been +attracted to power; the second did not campaign, so were not selected by +their craving for power or status; if the second board needs to ever use +any power at all, something has already gone wrong; and the second board +is securely pseudonymous, so membership cannot confer status on the +humans behind it. + +This structure is a shock absorber. Its goal is to dampen inappropriate +energy in the governance process. This bad energy could be external +pressure; it could be internal cupidity; it could even just be bad luck. +From the crooked timber of humanity nothing can be made straight, but a +good engineer can do a lot with geometry and redundancy. + +Evaluation + +Can anything that is not a physical state ever be impervious to state +pressure? Can all the devils of human cupidity ever be thoroughly +banished? These are easy questions. The answers are “no” and “no.” + +But a stronger fortress than Bitzion will be hard to build. (The main +risk is probably that the hodlers go collectively crazy—a risk we have +to accept when we choose low, but not zero, sensitivity.) + +Although adapted to the very unusual conditions of its birth and +existence, Bitzion is basically built like an early 21st-century +technology company. For all their faults and the faults of their world, +these can be very effective organizations—even at scale. Their +devotion to their shareholders is generally not in question; and their +shareholders take no actual part at all in managing them. + +Let’s assume that the hodlers and trustees make good personnel choices, +and create an organization as effective and aggressive as a top +technology company. What will it do? + +The Bitcoin space program + +Bitzion is Bitcoin—plus a trusted central government that has no +excuse not to be as effective as any startup, if only because it is +shaped like any startup. + +Let’s assume that, if the coordinator can demonstrate a use for it, the +trustees will grant the coordinator a dilution tax of up to 0.5% per +annum. At present valuations this is \$500M, which is a serious budget: +certainly enough to start a space program. Again, it is only a tenth of +the present cost of Nakamoto consensus. + +The mission of the coordinator is to advance the interests of the +ledger—meaning the weighted interests of the hodlers. There is no +other restriction on the scope or nature of Bitzion’s actions. All +prudent actions designed to benefit the hodlers are authorized. + +If the hodlers needed a space program, it would be proper for the +coordinator to start one. But other problems seem more pressing. + +For instance, it would not be hard to spend the budget of a space +program on the full-stack Bitcoin user experience. And UX is inherently +a shallow problem. Here are three of the deep architectural challenges +Bitzion has to master. + +Coordinated serialization + +Nakamoto consensus is a security measure. Like any security measure, its +effectiveness must be evaluated relative to the threat model. Like any +measure, its productivity must be evaluated relative to its price. And +its price is insane. + +With Bitzion as a trusted coordinator, the threat model assumes a +different shape. The Bitcoin ledger must be more trusted than the +coordinator, or even the trustees. In this design, we have not one but +two layers of security: government and decentralization. + +The threat model of coordinated serialization is halfway between the +threat model of Nakamoto consensus, and the threat model of a +permissioned blockchain. It is neither completely wild, nor completely +tame. + +Consider the classic threat of a 51% attack—hardly an abstract danger +when most mining is in China. 51% of all Bitcoin holders by position are +not in China or under Chinese control, so staking is more secure against +China than miner or node voting. Also, controlling miners is easier than +controlling nodes; controlling nodes is easier than controlling hodlers. + +Stake attacks also exist. But a governed ledger cannot be attacked in +any such way. Not only can Bitzion reverse any such attacks, it can +punish the attacker—by banning the staked coins from the serialization +market. + +Bitzion, like all governments, governs by combining the predictable +power of law with the unpredictable power of the prerogative or +exception. Judicious use of exceptional power can curate a reliable, +well-distributed volunteer serialization department. + +But since the ledger must be more secure than the government, there must +be a way for the ledger to keep working if the government fails or, +worse, goes crazy. It makes sense to tie this safety mechanism to the +hodler-initiated recall mechanism. + +While the serializers remain decentralized and uncoordinated volunteers, +it is hard to imagine a dying regime convincing or coercing them to +filter out a valid recall. They have no incentive to obey the regime, +and every habit and incentive to obey the recall. + +Any coordinated serializers constitute an attack against the regime, +which the regime must repel assertively by one-strike banning. In +practice, it can easily verify that most of its serializers are +independent from most others. But any hint of an attempt by the regime +to coordinate its own serializers cannot go undetected, nor can it +succeed instantly; and it justifies an immediate recall. + +While a new government is being elected, the ledger is in “safe mode.” +There is no exception. The volunteer serialization department operates +as usual, but without any supervision, and with higher safety +margins—which may be more expensive. + +The art of solving this design problem is the art of a serialization +solution which is cheap and secure, while the government is operating +properly; and less cheap, but no less secure, if the government +disappears, fails or turns evil. In other words: without a government, +this solution must degrade into conservative proof-of-stake. This is not +an easy problem, but intuitively it feels solvable. + +Bitcoin imperialism + +We return to the problem of altcoins and forks. The existence of +parallel ledgers may not be a showstopper for Bitcoin—but it also +might be a showstopper. + +The goal of Bitcoin is to become a standard. A standard must be certain. +Competition between coins creates uncertainty. Without these alt-dogs +nipping at its heels, Bitcoin has a stronger claim to be the one. As a +challenger, it needs all the strength it can get. + +Standardizing digital money is just a flex. It demonstrates strength. It +creates strength. All healthy and growing organizations are obsessed +with their own strength. And since it must be done fairly, it it is a +chance to demonstrate the grace of kings. + +So here is how Bitzion can execute a historic program of Bitcoin +imperialism, crafted in the old Silicon Valley tradition of “extend and +embrace” or “copy, acquire, kill.” + +Start by evaluating all the scarcity-based tokens currently trading (not +“utility tokens,” which are virtual stonks, or “address space,” which is +virtual real estate). Classify each as worthy or worthless. Publish this +official directory and the research behind it. + +A worthy altcoin is one which has made a significant contribution to the +cryptosphere—in technology, traction, or both. A worthless altcoin is +every other coin—or any new coin started after the official directory +was published. + +Bitzion will never recognize these worthless ledgers. It condemns them. +They will go to zero, become deadcoins, and be forgotten. If you still +have any after the directory appears, try to sell while there are +buyers. + +Bitzion recognizes the worthy ledgers and declares its intention to +merge with them—whether they like it or not. Hopefully most will like +it. The task is not easy; just doable. This makes it an excellent flex. + +First, Bitcoin will incorporate all worthy innovations of these +altchains. Second, it will emulate their protocols and +semantics—including a computational emulator for computational chains. +Third, it will build a completely smooth migration path in which anyone +working with the altchain can trivially move to the real blockchain. +Most users will not even notice the difference; developers will flip a +few switches. + +Now the engineering work is complete, and the marriage can be +consummated. Like a corporate merger, a blockchain merger is a +diplomatic and financial affair. It requires (a) the consent of the +altcoin holders to (b) accept some ratio of real coins per altcoin, +which can be stated as a premium or discount relative to the current +market ratio. + +A corporation’s board can typically force its shareholders to accept a +merger. None of these altchains has a board; and there are no +decentralized involuntary transactions. Therefore, to merge with these +anarchic collectives, Bitzion must find, assemble, and negotiate with an +informal reputation pool of potential endorsers. + +These negotiations will succeed or fail—which decides whether the +merger is friendly or hostile. Even a friendly merger cannot be +enforced, though its chances are better. If negotiations cannot set a +fair price, Bitzion will set one. Generally the hostile price is lower +than the friendly price—pour encourager les autres. + +Then, altcoin holders have a short period to make a decision. They can +either sell their altcoins for real coins at the final ratio, or grimly +hold on. If they sell, they get new but genuine coins—which the +coordinator has minted with the trustees’ approval.. + +In exchange, the coordinator gets all their altcoins. After the merger +and migration, the coordinator sells these altcoins—using the +revenues, if any, to purchase real coins. + +If even a low double-digit percentage of altcoins gets tendered, selling +them blows through the altchain’s whole buy-order book like a firehose +up a salmon’s asshole, and takes the market price of this legacy chain +to zero, where it should stay: a deadcoin. + +So grimly holding on is kind of a bad move. Are these the tactics of +Genghis Khan? Yeah, kind of. Do you know more about imperialism than +Genghis Khan? + +Bitzion starts this imperialist program on the bottom end of the worthy +list—taking down its feeblest competitors first. As these involuntary +mergers succeed, reducing any foolish holdouts to pyramids of 256-bit +skulls, the empire’s negotiating position grows stronger and stronger. +Oderint dum metuant! + +The Bitcoin intelligence agency + +The most important problem of any regime is security. For a blockchain, +security is serialization plus imperialism. The second most important +problem is intelligence. + +This isn’t James Bond stuff, of course. Intelligence in the modern sense +of the term just means collection and analysis of accurate and official +information. It doesn’t require any kind of secret volcano island +base—although that would certainly be cool. There are no martinis and +no silencers. Yes, we do like cats. + +Why do blockchains need official government intelligence? Intelligence +is data plus analysis. Analysis is great—but the first thing +blockchains need is data. They even have a word for an official data +stream: an oracle. + +The Sun has been around for four billion years. On the full power of a +computational blockchain plus an official data stream, the Sun has never +yet looked. In theory, it is possible to build prediction markets with +oracle reputation as a second moving part. In practice these weird, +wobbly markets suck. + +Without this concrete and certain connection to official facts, a +computational chain simply cannot think about the real world. A +blockchain which can compute with facts about the real world is a +different class of tool. “Prediction markets” does not cover the +potential power of this architecture. + +Of course, the difference between data and analysis is not always clear. +Once Bitzion becomes comfortable with its right to curate official data, +it will inevitably find itself considering the next step of composing +official analysis. + +Data feeds algorithms. Analysis feeds humans. The algorithms on a +blockchain can only use data. But those algorithms are ultimately +working for humans; and humans can use both data and analysis. + +The mission of Bitzion is to serve the hodlers. Authoritative +open-source real-world intelligence, also known as “news,” is often +suspect—less the data, more the analysis. If the only narrative +available to the hodlers is this “news,” and the “news” contains +systematic misjudgments, the hodlers must be likely to make systematic +mistakes. + +So Bitzion’s intelligence agency starts by collecting real-world +facts—which helps blockchain programs think about the real world. +Finding its feet with this easy work, it continues by constructing +real-world narratives—which helps blockchain users think about the +real world. + +Again, these are exactly the nominal functions of any 21st-century +intelligence agency. All these agencies still have some kind of weird +1950s James Bond historical mystique. The James Bond era was real; if it +even lasted into the ’50s, that would be surprising. + +Today the intel agencies are mostly just government think tanks, whose +wisest wisdom is not much different from the wisdom of the press and the +universities. The best test of whether they matter is what would happen +if they disappeared tomorrow, which is absolutely nothing. + +Yet their nominal function—research and analysis—remains real and +relevant. Nothing, moreover, could be more legitimate. This proposed +Bitcoin intelligence agency is, in the words of our peach-haired +President, “very legal and very cool.” + +Of course, Bitzion keeps no secrets. All this intelligence product is +public. And like the rest of Bitzion, it is independent of real-world +power. + +Needless to say, raw input from “the news” or even “science,” even from +the most “reliable” or “respected” sources, is never incorporated +directly and without a clear positive argument into the output of the +analysis directorate. + +Rather, its mission is to understand the world anew from scratch. +Everything “known” is an invaluable resource, of course. But just a +resource. How do we “know” this? Imagine trying to recreate Wikipedia +with no “reliable” sources—but instead, the conviction that all +information from the 20th century is unreliable until shown otherwise. + +So is all external information. While an intelligence agency wants to +hear everything, its mission is to think entirely for itself. Nothing in +its raw input stream is true until the agency has some clear positive +reason to consider it true. “Always trust content from Microsoft +Corporation” is not a valid form of reason. + +This question need not be asked story by story or paper by paper. It is +best asked field by field, department by department, agency by agency. +If the old 20th-century truth is a blueberry pie, we can pick out the +moldy parts slice by slice, not berry by berry. The whole pie is never +moldy—however old, there are always good parts in a pie. Pure math, +for instance, remains sweet and delicious with only a very mild +fermented bite. Then again, pure math did pretty well in the USSR. + +At this point we imagine a genuine conflict between virtual and physical +power—but this is probably a good stopping point for today. Once +Bitcoin has gone so far toward making a nation that it has made an +intelligence agency, the point seems proven. + +Both necessary and impossible + +Of course, the CJNG must feel the same way about its little panzer +parade— and that actually happened. + +Anyone who has actually read all this perhaps feels the same intuitive +sense as me: that Bitzion theoretically should happen, but actually +can’t happen. It is both necessary, and impossible; both fantastic, and +pragmatic. + +I have put a lot of words into why Bitzion is necessary. I already +warned you that it is impossible. If it becomes possible, it will not +become possible easily. + +Our theoretical interest, as political engineers, in what should be +done, remains legitimate. But a thought-experiment and a practical +conspiracy are two things. The conversation is not over without talking +about why it cannot happen. So let’s do that. + +It’s simple. Bitzion cannot happen because vanity is poison and everyone +is poisoned. + +Recently, Marc Andreessen wrote a much-cited essay which asked everyone +to build. Everyone should build. Not everyone should be an architect. + +The trouble with asking the whole world to build is that this world is +so poisoned by vanity that it can only hear one message: be an +architect. Almost all these people should be learning to put one brick +on top of another. But once a thousand architects have argued about how +to build the house, there is often not a bricklayer to be found. + +Having done both—metaphorically speaking—I will tell you +straight-up: it is more honorable to be a bricklayer. Architecture is a +gift. Masonry is a craft. Never trust anyone who has a gift but no +craft. If you have a gift, pursue your craft. Only fate can make you an +architect; but fate seems to like a man who can lay a brick. + +To lay a brick is to pursue the path laid down for you; it is to follow. +The mason is not an artist but an artisan. For an artist, the definition +of perfect is broad; for an artisan, it is narrow. The same amount of +effort and achievement can go into either. + +Yet the architect, who is an artist, leads; the mason, who is an +artisan, follows. The wall needs to go where it is supposed to go. It +needs to look like what it is supposed to look like. Yet in the hands of +an ordinary mason, it will be okay; in the hands of a fine mason, it +will be striking; anyone can tell the difference and no one can say +quite why. + +While the society that forces an artist to be an artisan has failed, so +has the society that forces an artisan to be an artist—or that exalt +the artist above the artisan. The latter is simply a society without +artisans—which is a society that cannot build. Oops. + +(In fact it is a curse to be a real artist or scientist. These people +should be derided and shunned. It’s an admission against interest—but +that way, they’ll do their best work.) + +This is where Bitzion fails. Bitzion is obvious. Bitzion would already +have happened— in a world where people thought about collective action +not in terms of leading, but in terms of joining and following—where +they knew how to be masons, not architects. + +Not that anyone knows how to be an architect, either. Being an architect +means making plans which masons follow. There are no masons; so there +are no architects. No one has any experience in the actual profession of +architecture. There are just dreamers who try to build beautiful dreams, +by themselves, sloppily. Many houses are started; few are finished; many +of these fall down. But the rest are beautiful. + +This metaphor is just a metaphor. It could describe many careers and +professions. In all such cases, there is one factor in common: vanity is +poison. + +In the case of Bitzion, vanity manifests itself as an inability to +trust. To the average hodler, the thought of delegating collective +control to the board of trustees—even through this clever mechanism, +by which no one’s vote is lost—is horrifying. + +Even the thought of trusting the hodlerate collectively would horrify +most hodlers. Literal collectivism? Why, isn’t Bitcoin supposed to be +the exact opposite of that? + +But worst of all—hyperdelegation, or any sort of delegation, or +election, or indirect democracy in general, means you are giving away +your power. Why vote, when you could rule? It’s going from weed to +meth—having ruled, it will never be enough to just vote. + +For example, when you vote, you vote every four years—for four years, +except one day, you have zero power. When you rule, you get to rule +every day of the week. And even worse: when you vote, you follow. You +literally submit to another person. You almost feel your genitals +shrinking. + +To be a mason is to say to some other human or humans: I will obey you, +I will follow your directions, I will put the bricks how and where you +tell me to put them, I will not create my own whimsical Art Nouveau +motifs. + +Of course, to join together and follow a common leader, or leadership, +or government, or constitution, is the fundamental act which creates all +regimes. This act is necessary. Without such an act of common consent, +no political structure can be created. + +But one and only one word of this recipe is utterly impossible. That +word is follow. No one wants to follow. Worse: no one even knows how to +follow. + +The practical consequence of this deficit is an inability to trust. The +project of Bitzion is impossible because it depends on taking a hodler +community dedicated, X-Files style, to trusting no one; and moving it to +trusting each other; and implementing this by trusting a roomful of +people, who then trust a single person. + +To a true Bitcoiner, this seems completely insane. It is also how every +company in the world, major or minor, works—and Bitcoin has the market +cap of a major company. And in any major company, the amount of time the +shareholders spend even caring about the governance of the company, let +alone trying to change it, is epsilon. + +If they are not delegating unconditional trust, they are delegating +something close. They have no plan at all for running the company +themselves, or even interceding in its decision loop. There are always +exceptions; but this is the normal operating loop. The shareholders seem +to trust their direct representatives, the directors, no less than they +trust the CEO selected by those directors. This is an amazing amount of +trust—which can only exist because the joint-stock company was +invented long ago. + +Everything big that works is worked by these companies. They are clearly +the most effective way to operate at scale. Yet we are incapable of +recreating this level of trust in a new context. The joint-stock company +design survives because it is grandfathered—and because it works so +well. Nonetheless, a “capitalist” government is no more than one which +allows joint-stock companies—so there remains a huge target on its +back. + +And yet, the idea of the hodlers of Bitcoin collectively agreeing to +turn themselves into the sharehodlers of Bitzion, which is the same as +Bitcoin except that it governs itself rather than being governed by +math, seems at a certain level preposterous—because it goes from +trusting math, to trusting humans. And trusting humans is following. And +following is servitude—which was exactly what we were trying to escape +from, right? + +In practice Bitcoiners have to trust humans anyway; only a government +which is both legitimate and human can be proactive, which means only +such a government can be effective; and government by hash is incredibly +expensive. And yet Bitcoin remains the leader—proving the power of +monetary incumbency. + +Bitcoin can’t do a 180 and treat decentralization not as an end, but as +a means—to the real goal, independence. This would open the door to +independence through a wholly different means, pseudonymous +centralization. It seems like a minor, obvious change. But it would +require Bitcoin to entirely change its mind—or at least, its religion. + +Bitcoin is a cult, for better or worse. Cults don’t change their +collective minds from the bottom up. Bitcoin is a cult with an +accidental and mathematical government. Bitzion is the exact same cult, +with an conscious and human government. + +An altcoin or a fork with this design could still be organized; but it +would still be an altcoin. Please don’t try this. Bitzion has to be all +of Bitcoin, or most of it; and if didn’t ask you to follow, it might +have a chance at that. But it exists to ask you to follow. + +Honor still demands honesty. In a society where everyone is too good to +lay bricks, no one can build a house. It’s a hard problem, and it won’t +get solved here. But solving the problem has to start by talking about +it. + +Consider the irony. This ancient human talent—the reflex for +unconditional loyalty—which could make Bitcoin a thousand times more +powerful—is, for a bunch of peasant narco-soldiers from Jalisco, as +easy as falling off a log. + +Yet you might as well ask the hodlers, our fellow autistic rationalist +superbrains, who whether they like it or not are citizens of this new +thing with a B, to dunk a basketball. Our 20th-century prejudice is that +intelligence is power. Sometimes it is weakness; and all we have to +fight this weakness is intelligence itself. diff --git a/docs/writing_and_editing_documentation.md b/docs/writing_and_editing_documentation.md new file mode 100644 index 0000000..e5fff97 --- /dev/null +++ b/docs/writing_and_editing_documentation.md @@ -0,0 +1,538 @@ +--- +title: >- + Writing and Editing Documentation +# katex +--- +# Pandoc Markdown + +Much documentation is in Pandoc markdown, because easier to write. But html +is easier to read, and allows superior control of appearance + +To convert Pandoc markdown to its final html form, invoke `Pandoc` by the bash +shell file `./mkdoc.sh`, which generates html. + +In the windows 10 environment, shell scripts used in this project need to be +associated with [Git Bash](libraries/git_bash_undocumented.html) or run from within Git Bash. + +If the title in the markdown file is followed by `# katex`, as in +the markdown form of this file, the shell script will tell Pandoc to display +any formulae using katex in the html file. + +More precisely, if any of the first three lines in the yaml header specifying +the title at the start of the markdown file are `# katex`, the `./mkdoc.sh` +will tell Pandoc to use katex to display maths formula. + +This vast pile of notes is out of control, and writing code and maths in +html leads to intolerable overheads. + +Hence markdown, the popular markdown conversion program +being the open source Pandoc. + +Markdown converters are apt to throw a flood of incomprehensible html code +into your final document, taking low level html control away from the writer. + +Pandoc, however, will allow you to take control. To integrate html and css +with markdown using Pandoc is a bit like rolling marbles with your +elbows through a cage. One has to work through and around the entry +points that Pandoc gives you, while if you were writing in html you could +just write what you damn well wanted directly, but having done the work, +Pandoc can then ensure it is done for every document in the same style in +the same way, and you can change the final form of every document in the +same way all at once. + +Sphinx is very popular and widely used, and written in the far more +accessible language python, but to access the power of html, css, and +JavaScript one must write a Sphinx theme, and the creation of a Sphinx theme +is less than well documented and appears to be subject to change. + +Visual Studio Code theoretically does automatic generation of the html +equivalents of markdown files, but I never was able to get it working +satisfactorily. + +Pandoc has a number of powerful extensions that allow integration of html and +markdown, among them markdown native mode divs `:::` + +```markdown +::: {style="…"} + … +::: +``` + +And native mode spans `[…]{style="…"}` + +Which extensions do not work correctly in Visual Studio Code. + +These can be used to put an anchor in text, but the easiest and most +intelligible way to insert an anchor is as a header. + +Pandoc can do a good job of rendering math markdown without invoking +katex, and in such cases, one should generate the html + +```bash +fn=filename + pandoc --toc --eol=lf --wrap=preserve --from markdown+ascii_identifiers --to html --metadata=lang:en --verbose --include-in-header=./pandoc_templates/header.pandoc --include-before-body=./pandoc_templates/before.pandoc --include-after-body=./pandoc_templates/after.Pandoc -o $fn.html $fn.md +``` + +Since markdown has no concept of a title, Pandoc expects to find the +title in a yaml inline, which is most conveniently put at the top, which +renders it somewhat legible as a title. + +Thus the markdown version of this document starts with: + +```markdown +--- +title: >- + Writing and Editing Documentation +# katex +--- +``` + +# Converting html source to markdown source + +In bash + +```bash +fn=foobar +git mv $fn.html $fn.md && cp $fn.md $fn.html && pandoc -s --to markdown-smart --eol=lf --wrap=preserve --verbose -o $fn.md $fn.html +``` + +# Math expressions and katex + +Pandoc can render most maths markdown without needing katex, for example: +$${e}^{i\pi}+1=0$$ +$$a=b+c$$ +$$f(x) = x^2$$ +$$\sin(\pi/6) = 0.5$$ +$$\int_a^b f(x) dx$$ +$$\int_a^b \tan(x) dx$$ +$$\int \sin(x) dx = \cos(x)$$ +$$\sum a_i$$ +$$\lfloor{(x+5)÷6}\rfloor = \lceil{(x÷6}\rceil$$ +$$\lfloor{(x+5)/6}\rfloor = \lceil{(x/6}\rceil$$ +Using Omicron, \bigcirc, not capital O for big \bigcirc. `\Omicron` will not always +compile correctly, but `\ln` and `\log` is more likely to compile correctly than +`ln` and `log`, which it tends to render as symbols multiplied, rather than one +symbol. +$$\ln(1+x)=x-\bigcirc(x^2)$$ +$$H(a|b|v)$$ + +though it is subtly prettier with katex, and some maths expressions will +break Pandoc unless one tells it to use katex. + +Some maths, Pandoc needs katex: + +$$\sin(\frac{\pi}{6}) = \frac12$$ +$$\displaystyle\frac{u(x)}{v(x)}$$ +Inline equation $\displaystyle\sum\limits_{i=1}^n i^2 = \frac{n(n+1)(2n+1)}{6}$ text after inline equation + +$$\displaystyle\sum\limits_{i} i^2 = \frac{i(i+1)(2i+1)}6$$ +The square root of 100 is $\sqrt{100}=10$.\ +The cubic root of 64 is $\sqrt[3]{64}=4$ +$$\bigg\lfloor\frac{x+5)}{6}\bigg\rfloor = \bigg\lceil{\frac{x}{6}}\bigg\rceil$$ + +So for documents requiring some heavy maths display, we convert from markdown +to html with, in the bash script `./mkdoc.sh`: + +```bash +fn=filename +pandoc --katex=./ --toc --eol=lf --wrap=preserve --from markdown --to html --metadata=lang:en --verbose --include-in-header=./pandoc_templates/header.pandoc --include-before-body=./pandoc_templates/before.pandoc --include-after-body=./pandoc_templates/after.pandoc -o $fn.html $fn.md +``` + +The `./` tells `pandoc` to expect to find the files + +```bash +./katex.min.css +./katex.min.js +``` + +That a file needs katex is flagged for `./mkdoc.sh` in the yaml header. + +A file that does not need katex has the header: + +```markdown +--- +title: >- + Document title +--- +``` + +But if it does need katex, it has the header + +```markdown +--- +title: >- + Document title + # katex +--- +``` + +So that the bash script file `./mkdoc.sh` will tell `Pandoc` to find the katex scripts. + +For it offends me to put unnecessary fat in html files. + +## overly clever katex tricks + +$$k \approx \frac{m\,l\!n(2)}{n}%uses\, to increase spacing, uses \! to merge letters, uses % for comments $$ + +$$k \approx\frac{m\>\ln(2)}{n}%uses\> for a marginally larger increase in spacing and uses \ln, the escape for the well known function ln $$ + +$$ \exp\bigg(\frac{a+bt}{x}\bigg)=\huge e\normalsize^{\big(\frac{a+bt}{x}\big)}%use the escape for well known functions, use text size sets$$ + +$$k\text{, the number of hashes} \approx \frac{m\ln(2)}{n}% \text{} for render as text$$ + +$$\def\mydef#1{\frac{#1}{1+#1}} \mydef{\mydef{\mydef{\mydef{y}}}}%katex macro $$ + +# diagrams + +The best way to do diagrams is Inkscape and the Visual Studio Code +scalable vector graphics extensions. + +I decided to place the data directly inline in markdown because interfacing +scalable vector graphics files (`svg` files) to html can get complicated, and +interfacing the resulting complicated html to markdown can get more +complicated. + +Which means that any time I want to edit the `svg`, I have to extract it into a +temporary `svg` file, edit it in Inkscape and Visual Studio Code, minify it in +Visual Studio Code to get rid of all confusing mystery cruft inserted by +Inkscape, edit it back into markdown compatible form, and reinsert it in +the markdown file. + +Which assumes my images, once done, are rarely or never going +to change. + +Scalable vector graphics are dimensionless, and the `` tag's height, +width, and ViewBox properties translate the dimensionless quantities into +pixels. The graphics default to fixed aspect ratio, and anything outside +the viewbox is not drawn. To adjust your image's position within the +viewbox, you put everything into a single big group, and apply a translate +to that group. + +You need to set an Inkscape tools properties from an existing element that +you have manually edited as text, as a great deal of functionality is only +available by editing vector graphics as text, and attempting to manipulate +that functionality from Inkscape, though in theory possible, just creates +an uncontrollable mess. + +Every so often, a tool's properties get set to something stupid for some +stupid reason, and there is no fix available in the Inkscape UI, other than +selecting a graphic element, double clicking on the tool, and telling it to +take its defaults from the element selected. (which can be an entire group, +or the entire high top level group, in which case it will pick up sane and +appropriate properties from the first relevant item in the group.) + +And how did you set those sane and relevant properties in the first place? + +By editing that element as text, very likely in your markdown file. + +The enormous advantage of scalable vector graphics is that it handles +repetitious items in diagrams beautifully, because you can define an item +by reference to another item, thus very large hierarchical structure can be +defined by very small source code. + +Scalable vector graphics is best edited as text, except that one needs to +draw lines and splines graphically, and in the process, all your nicely laid out text gets mangled into soup, and has to be unmangled. + +You might decide to keep it around as soup, in an `svg` file that only +Inkscape ever tries to read, but then you are going to have to edit it as text +again. And I wound up embedding my vector graphics files in markdown +rather than invoking them as separate graphics files because my last step +was apt to be editing them as text. + +It is convenient to construct and position all the graphical elements in +Inkscape, then edit the resulting `svg` file in Visual Studio Code with the +`svg` extension, watching the resulting graphic in the `svg` editor's preview +pane, then use the `svg` extension's minify command to get rid of all the +excess stuff generated by Inkscape, then edit the `svg` file to be compatible +with Markdown, then edit it into the markdown file, then edit the +markdown file in the Visual Studio Code markdown editor, watching the +graphic in the markdown preview pane. + + + + + + + + + + start animation + + + + + + + + + + A simple scalable vector graphic + + directly embedded in markdown. + + + + + + + +```svg + + + + + + + + + start animation + + + + + + + + + + A simple scalable vector graphic + + directly embedded in markdown. + + + + +``` + +```script + + +# tables + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + +
May Scale of monetary hardness
Hardness
+
Hard
1Street cash, US dollars
2Street cash, euro currencies, japan
3Major crypto currencies, such as Bitcoin and Monaro
4Street cash, other regions
5Interbank transfers of various sorts (wires etc), + bank checks
6personal checks
7 + Consumer-level electronic account transfers (eg + bPay)
8Business-account-level retail transfer systems
Soft
9Paypal and similar 'new money' entities, beenz
10Credit cards
+ +``` + +# Tables + +## Pipe table with header and alignment control + +Without counting spaces, but without multiline + +Pipe table: + +| Right | Left | Default | Center | +|------:|:-----|---------|:----------------------:| +| 12 | 12 | 12 | 12 | +| 123 | 123 | 123 | the quick brown fox jumped over the lazy dog | +| 1 | 1 | Carrian Corporation | 1 | + +## And, with less mucking about, alignments + +with alignment, without counting spaces, but without multiline + +fruit| price +:-----|-----: +apple|2.05 +pear|1.37 +orange|3.09 + +## multiline without bothering with pipes + +Counting spaces to align. Only editable in fixed font + +This allows multiline, but visual studio code does not like it. Visual Studio Code only supports tables that can be intellibly layed out in visual studio code. + +------------------------------------------------------------- + Centered Default Right Left + Header Aligned Aligned Aligned +----------- ------- --------------- ------------------------- + First row 12.0 Example of a row that + spans multiple lines. + + Second row 5.0 Here's another one. Note + the blank line between + rows. +------------------------------------------------------------- + +## The header may be omitted in multiline tables as well as simple tables + +Notice the alignment is controlled by the first item in a column + +In this table, edited in a fixed font, you are using whitespace and blank lines to lay out the table. It is unintellible in a variable width font. + +----------- ------- --------------- ------------------------- + First row 12.0 Example of a row that + spans multiple lines. + + Second row 5.0 Here's another one. Note + the blank line between + rows. +----------- ------- --------------- ------------------------- + +## Grid tabless are safer, hard to count spaces + +Allows multiline, and alignment, but visual studio does not like it, and you still have to count those spacees + ++---------------+---------------+--------------------+ +| Fruit | Price | Advantages | ++===============+==============:+====================+ +| Bananas | $1.34 | - built-in wrapper | +| | | - bright color | ++---------------+---------------+--------------------+ +| Oranges | $2.10 | - cures scurvy | +| | | - tasty | ++---------------+---------------+--------------------+ +| Durian | $22.10 | - king of fruits | ++---------------+---------------+--------------------+ + +Alignments can be specified as with pipe tables, by putting colons at the boundaries of the separator line after the header. + ++------------+---------+---------------------+ +| Left | Right | Centered | ++:===========+========:+:===================:+ +| Bananas | $1.34 | - built-in wrapper | +| | | - bright color | ++------------+---------+---------------------+ +| Durian | $22.10 | - king of fruits | ++------------+---------+---------------------+ + +## For headerless tables, the colons go on the top line instead: + ++--------------:+:--------------+:------------------:+ +| Right | Left | Centered | ++---------------+---------------+--------------------+ + + + diff --git a/docs/zookos_triangle.md b/docs/zookos_triangle.md new file mode 100644 index 0000000..2b2c3dd --- /dev/null +++ b/docs/zookos_triangle.md @@ -0,0 +1,354 @@ +--- +title: + Zooko’s Triangle +--- + +# Zooko Identity + +In a decentralized system with no trust anchors how do you figure out +which name is the right one with any certainty. If you are texting to “Bob”, +how do you know you have the right “Bob”? There are a lot of Bobs. + +Zooko’s triangle is the solution to this problem. It is explained in several places + +- [An Introduction to Petname Systems](http://www.skyhunter.com/marcs/petnames/IntroPetNames.html) +- [Lambda for Humans: The PetName Markup Language](http://www.erights.org/elib/capability/pnml.html) + +Each identity, whether a human, a server, or something else, has a globally +unique cryptographic identifier, which is not human readable or human +memorable, a title or display name, which is human readable but not +necessarily easy to type correctly or reproduce exactly from memory, a +nickname, which is not globally unique, is likely to be unique among the entities that you are dealing with, but not necessarily unique, even among them. + +The petname is the name that your system calls this entity, not necessarily +what any other system calls this entity, a name constructed by or your computer, typically constructed from, or identical to, the nickname. + +The petname is locally unique, human readable, and human typeable. It is how you know this entity and reference it. + +The computer is responsible for doing the mapping as needed, and +makes sure that when you connect to Bob, you are connecting to the right Bob. + +Typically the globally unique cryptographic identifier is a public key, and +only Bob knows the correct corresponding secret key. If the thing +identified is an immutable data structure, the globally unique +cryptographic identifier is a hash of that data, according to the hashing +rules for that data type and data structure, which type and structure you +very likely do not know and will not learn until you contact the system +that has the data. + +# Zooko’s triangle today + +We now have a lot of systems that to a greater or lesser extent implement +Zooko’s triangle, and while they are frequently incomplete and +unsatisfactory implementations, they all do a sufficient job of making sure +you are talking to the right Bob. + +The way it works as (more or less) implemented today (frequently less) is +as follows: + +In private messaging, there are separate message threads for each public +key, just as when you text on a phone to and from a particular person, +there are separate threads for each phone number, unless you have +manually associated two different phone numbers with the same identifier, +and where the name appears in text in public or private messages your +local computer should associate a unique local petname with each public key. +And mostly they do, though existing software is frequently half assed +about this. + +The petname is generated by mangling the nickname (and possibly part of +the display name) down to a valid locally unique identifier, unique on your +computer and following your computer’s local rules for valid identifiers, +which rules might well depend on the language locale in which the +computer is running. + +Existing software fouls up on the local petname issue in a variety of ways, +but does a sufficient job to ensure that one person cannot impersonate +another person + +The petnames appear in the text, and in the equivalent of reply-to and cc +fields, as @petname, but if “petname” is a local petname corresponding to +a public key that corresponds to the other party’s private key, it is colored +differently to the text and/or looks like a link in html. If you hover over the +link, you see the other party’s display name and nickname, and if you click +on that link, you go to data about that person. + +What actually gets sent, computer to computer, when Ann mentions Bob +in a message to Carol, is not the the local petname that that Ann typed into +her text, but the globally unique identifier, which should be, and usually is, +the public key or information uniquely identifying the public key, and +possibly the nickname, the display name, and information on +how to direct a message to that party. + +The display name is long, humanly meaningful, and usually +probabilistically unique to humans, but inconvenient for typing in full in +text, unlikely to be typed correctly, likely to be incorrectly formatted for +reply-to and cc fields, and often likely to disrupt the flow of text were it to +be directly typed as part of a paragraph. + +The nickname is a potential petname, or can be mangled into a locally +valid petname without too much mangling. + +Display names and nicknames are unique on any one forum If one public +key, one nickname and display name on any one forum. Different people +can have the same nickname and display name on different forums on +different forums. But if they do, if there is one Bob on one forum and a +different Bob on another forum, they are going to nonetheless have +different local petnames on your computer, and if one Bob sends you a +private message, it will appear in a different message thread to that other +Bob’s message thread. + +What appears in the address fields, and in references to particular people +in the middle of a paragraph, is always locally unique, one distinct local +petname for each globally unique public key. + +In practice, you don’t rely on petnames to distinguish people, though they +are quite adequate for distinguishing people, but on the fact that the +threading software associates messages coming from the same public key, +just as when you see text messages coming from the same phone number +on your phone, you see them displayed with previous text messages and +replies to that phone number, and pay little attention to the phone number. + +In most contexts, you don’t actually need to look at, or even remember, the +phone number, and you don’t actually need to look at, or even remember, +the petname. You click on the “reply to” button, and the software fills in +the appropriate local petnames, which become the appropriate global +identifiers when the message is sent, and when the parties receive the +message, it gets threaded according to global identifiers, and they see their +local petnames for people. Same as with phone numbers on your text +messaging application on your phone. No one needs to see phone numbers +these days, and even less do people need to see public keys, and seldom +need to pay attention to local petnames. + +# The full and correct implementation of Zooko UI + +All existing systems are half incomplete and unsatisfactory: + +> Bitmessage will never be popular, period. The common pleb +> doesn't give a shit about privacy or anonymity or security, and just +> happily uses Fecesbook. The slightly advanced computer user +> uses the "more secure" instant messengers, that offer a proper GUI and +> easy usage. The only people using bitmessage are literally some +> groups of nerds, the unfortunate few, who have coriously stumbled +> here (for a minute, until they realize what the public channels look +> like), or trolls that need an outlet for their mental diahrea. +> +> As much as I think BitMessage is an interesting concept, you +> cannot avoid realizing that it's massively unattractive for wide adoption. + +The Bitmessage UI is unbearable. What is needed for it to be popular is a +full Zooko name system. With a full Zooko name system, it is going to feel +much like one of the more popular apps. + +A Zooko name consists of + +- the Guid, the Globally unique identifier, for example the bitmessage Guid, "`BM-NB5cJuoFTkcmNEKDcBwn1RDxYDX3gGbq`", which represents + the public elliptic point of a secret scalar known only to the sender. + +- the nickname, which is global and chosen by the entity named, but + probably not unique, Many people can, and usually do, have the same + nickname. The nickname starts with a letter, and is composed of + letters, numbers, and underscore characters, for example "John_Smith". + +- The full name: The full name is likely to be unique, but only if the entity + named chooses it to be likely unique, for example + {John_Smith, Dark Lord of attack resistant cryptography}. The full + name can be any sequence of unicode characters that does not + contain unbalanced curly brackets, and the nickname is the first + sequence within the full name of three or more alphabetic, numeric, + or underscore characters starting with an alphabetic character. + +- The petname, which is the local name of someone else on your + computer. This is locally unique, and defaults to the nickname, or a + name constructed from the nickname, unless that petname already + exists and names a different guid. + + When your system locally constructs a default petname from a nickname, + it may mangle it in such a way that nicknames that look similar do + not have petnames that look similar. + +When sending a message, you can reference a guid by the petname + +For example, if Bob's petname for John_Smith is John_Smith71 (because he already had John_Smiths one to seventy as petnames on his computer), but Carol's petname for John_Smith is JohnSmith, and Bob sends a message to Carol containing the the text + +> remember when @John_Smith71 said ... + + then this gets translated on sending to + +> remember when @{John_Smith, Dark Lord of attack resistant cryptography} #`BM-NB5cJuoFTkcmNEKDcBwn1RDxYDX3gGbq` said ... + + and when displayed to Carol, gets represented as: + + > remember when @`JohnSmith` said ... + + because the guid is already defined on her computer as the local petname JohnSmith. + +If it is not yet defined, then she sees the full set of Zooko names, as sent: + +> remember when @{John_Smith, Dark Lord of attack resistant +> cryptography} `#BM-NB5cJuoFTkcmNEKDcBwn1RDxYDX3gGbq` said ... + +Petnames preceded by the '@' symbol get translated to the full Zooko +nameset on being sent from the local context where the petname is +defined, and the guid gets translated back into the corresponding locally +defined petname on entering a local context where a petname is defined +for the guid of that nameset. + +When receiving a message that references a known guid, the same guid +always translates to the same local name, even if the nickname and +fullname is different. + +The same guid can have many nicknames and full names, and the same +nickname and full name might by used by many different entities with +many different guids. But people should try to avoid that, by choosing full +names likely to be globally unique, and nicknames likely to be locally +unique on the systems of the people they communicate with. + +If someone maliciously chooses the same full name as someone else, in +order to deceive the recipient of the message, the computer will see the +guid, and assign him a different petname. To protect oneself against +against people pretending to be oneself, one should choose a nickname +likely to get a short and distinctive petname assigned on the recipient's +computer. + +Everyone's system will record each association between a full name and a +guid that is sees, the time at which it saw that association, and the guid +asserting the association. If it sees a later guid claiming a nickname that +corresponds to an existing petname, it treats this as strong reason to +suspect the entity who knows the secret key corresponding to the guid +making the assertion of scamming, spamming, and shilling. If it sees two +different guids with the same full name, also strong reason for suspicion +Two guids with the same nickname, weaker reason for suspicion. + +If you send a message to an entity defined by a guid, or add a guid to your +address book, or if you compose a text containing a full Zooko nameset, +perhaps because you are replying to a message that references an entity +that has no petname on your system, you will asked to whether you want to +store that full Zooko nameset under a local petname on your system. + +# Moving to a global Zooko system + +Zooko’s triangle is, among other things, a user interface concept for +storing, managing, and communicating, cryptographic capabilities. + +Zooko’s triangle solves half the problem: provides a trusted path. If +you trust Bob, and Bob trusts Carol, you can trust that Bob’s +introduction to Carol will in fact bring you to the correct Carol. + +This, of course, assumes a browser is somehow able to use Zooko style +links, rather than PKI links. + +The problem then becomes having large number of trusted introduction – a +map between human phrases, and documents containing information about +globally unique identifiers that are relevant to that phrase: Something +like Wikipedia, but without centralized authority forbidding “original +research”, aka any deviation from official government truth, and +something like a search engine. + +To solve the problem of making Zooko style links useful, we have to +solve the problem of providing global data that is not state dominated. + +Past experience with cryptographic capabilities is that if users are +expected to consciously and intentionally use them, they screw up, that +even experts screw up, and that end users are not only reluctant to use +them correctly, but find it profoundly difficult to use them at all that +even expert users find them a pain, as for example the regular +unpleasantness of installing a certificate on a web server so that it +can do https. + +Zooko’s triangle, correctly implemented, should hide from users that +they are using cryptographic identifiers, or indeed any globally unique +identifier. + +Because globally unique identifiers have become almost as ugly as +cryptographic identifiers, we have already implemented interfaces that +hide globally unique identifiers, for example the bookmarks folder, the +buddy list, and the email contacts list. And since those identifiers are +already hidden, they can be cryptographic, giving the user many +benefits, and no extra grief. + +We can do this with not one extra click for security, and indeed will +have to do so, for experience has proven that if we ask the user for +extra clicks for security, the user becomes frightened and confused, and +even supposedly expert users do not provide those extra clicks. + +In order that references to objects can be securely transmitted across +trust boundaries, they need to be cryptographic capabilities + +You don’t want people sending you spam pretending to be your webmaster, +or your email host, or your employer, or your bank. You want to share +files with certain people, but not always with the entire world, you +want to give some people, but not others, the ability to edit particular +files. When someone trusted recommends a bank, or a firm, you don’t want +some scammer connecting you to himself, instead of the bank you are +trying to connect to. + +IM buddies, email contacts, and important web pages should be +cryptographic capabilities. When you click on a link, it should take you +to the web page the author you are clicking on intended. When you +receive a message that purports to be from an entity you have a +relationship with, it should be from that entity, and when you receive a +message from an entity you do not have a relationship with, it should be +obvious you do not. All messages should have in the headers “Regarding +.....”, which refers to a particular capability to contact you – which +usually is the particular capability to contact you contained in some +previous outgoing message sent by you. + +Further, if we had a way of routinely and easily handling cryptographic +capabilities, lots of things that are now inconvenient and unsafe, such +as web money, could be made considerably easier and safer. + +# Monetizing a system of Zooko identity + +Information wants to be free, but programmers want to be paid. The +primary reason for centralized name systems is that people can make +money off other people's internet and corporate reputations. + +Names have value, when embedded in a network of names linking to +names, because reputation has value. We can create secure pseudonymous +payment using cryptography, but cannot create secure delivery of goods +and services using cryptography. The other side of the transaction needs +reputation. + +[sovereign corporations]:social_networking.html#many-sovereign-corporations-on-the-blockchain + +So the other side of the transaction needs to be authenticated by a secret +that *he* controls, not a secret controlled by registrars and certificate +authorities. Enabling people to own their own names is an important step +towards enabling [sovereign corporations]. + +Most of the value in the world is not in productive machinery and land. It +is in “goodwill” – the value of names linked by names. And governments +registrars, and so on and so forth keep skimming that value, and from time +to time recklessly destroy it. + +But someone has to be paid for enabling people to own their own +reputations. + +How do we make money out of enabling people and groups of people to +own their own reputations? + +The easiest and most obvious way is if we issue the money that they use to +transact on these reputations. Cryptographic transfer of funds is one half of +the problem. The other half is wondering whether the person you are +transferring it to will hold up his end of the bargain. Both parts need to be +integrated. + +Reputations have value by being embedded in a network of reputations. +Google made a lot of money by analysing the network and making the +information readily available. My fundamental plan is that the keys used +for identity will be rooted in the same wallet as the keys used for +cryptographic coins, and thus communication carrying metadata about +transactions rooted in that wallet, resulting in an intimate linkage between +the crypto currency being valuable, and the information about reputation +secured by the same wallets being readily available. + +And to make reputation embedded in the network of links from +reputations to reputations valuable, we are going to eventually need +network analysis systems similar to that provided by Google and others. + +First mover providing such analysis will have a substantial first mover advantage, as Google has and still enjoys to this day, but the first mover advantage is likely to be less valuable, since lots of search companies +will simply add analysis of secure links, links rooted in secrets that are +controlled by the owner of the name to their analysis of links rooted in +government authority, rooted in certificate authority secrets answerable to registrars. On the other hand, being the source of the wallet software is +likely to be a substantial and lasting advantage in being source of the +network analysis. diff --git a/frame.cpp b/frame.cpp new file mode 100644 index 0000000..ca6bd30 --- /dev/null +++ b/frame.cpp @@ -0,0 +1,432 @@ +#include "stdafx.h" + +// ---------------------------------------------------------------------------- +// frame +// ---------------------------------------------------------------------------- + +void Frame::RestorePositionFromConfig(const wxSize& bestSize) { + // SetPath() understands ".." but you should probably never use it. + singletonApp->pConfig->SetPath(_T("/MainFrame")); wxPoint scr{ wxSystemSettings::GetMetric(wxSYS_SCREEN_X), wxSystemSettings::GetMetric(wxSYS_SCREEN_Y) }; + // restore frame position and size + int x = singletonApp->pConfig->ReadLong(_T("x"), scr.x / 4); + int y = singletonApp->pConfig->ReadLong(_T("y"), scr.y / 4); + int w = singletonApp->pConfig->ReadLong(_T("w"), scr.x / 2); + int h = singletonApp->pConfig->ReadLong(_T("h"), scr.y / 2); + w = std::min(std::max(std::max(w, scr.x / 5), bestSize.GetWidth()), 8 * scr.x / 9); + h = std::min(std::max(std::max(h, scr.y / 9), bestSize.GetHeight()), 4 * scr.y / 5); + x = std::max(scr.x / 12, std::min(x, scr.x - w - scr.x / 12)); + y = std::max(scr.y / 10, std::min(y, scr.y - h - scr.y / 10)); + this->Move(x, y); + this->Maximize(singletonApp->pConfig->ReadBool(_T("Maximized"), false)); + this->SetSize(w, h); + singletonApp->pConfig->SetPath(_T("/")); + if (singletonApp->m_display || m_pLogWindow != nullptr) { + m_pLogWindow->GetFrame()->SetSize(w, h); + if (singletonApp->m_display_in_front) { + m_pLogWindow->GetFrame()->Move(std::min(x + 46, scr.x - w), std::min(y + 46, scr.y - h)); + } + else { + m_pLogWindow->GetFrame()->Move(std::max(x - 32, 0), std::max(y - 32, 0)); + } + m_pLogWindow->GetFrame()->SetTitle(sz_unit_test_log); + m_pLogWindow->Show(true); + } + else { + m_pLogNull = std::make_unique(); + } +} + +void Frame::StorePositionToConfig() { + if (singletonApp->pConfig) { + singletonApp->pConfig->SetPath(_T("/MainFrame")); + if (this->IsMaximized()) { + singletonApp->pConfig->Write(_T("Maximized"), true); + } + else { + // save the frame position + int x, y, w, h; + this->GetSize(&w, &h); + this->GetPosition(&x, &y); + singletonApp->pConfig->Write(_T("x"), (long)x); + singletonApp->pConfig->Write(_T("y"), (long)y); + singletonApp->pConfig->Write(_T("w"), (long)w); + singletonApp->pConfig->Write(_T("h"), (long)h); + singletonApp->pConfig->Write(_T("Maximized"), false); + } + singletonApp->pConfig->SetPath(_T("/")); + } +} + +// main frame ctor +Frame::Frame(wxString wxs) + : wxFrame(nullptr, myID_MAINFRAME, wxs, wxDefaultPosition, wxDefaultSize, wxDEFAULT_FRAME_STYLE, wxs), + m_panel(), + m_LastUsedSqlite() +{ +try { + assert(singletonFrame == nullptr); + singletonFrame = this; + SetIcon(wxICON(AAArho)); + sqlite3_init(); + if (sodium_init() == -1) { + szError = "Fatal Error: Encryption library did not init."; + errorCode = 6; + // Cannot log the error, because logging not set up yet, so logging itself causes an exception + throw FatalException(szError.c_str()); + } + wxIdleEvent::SetMode(wxIDLE_PROCESS_SPECIFIED); + if (singletonApp->m_display || singletonApp->m_log_focus_events) { + m_pLogNull.reset(nullptr); + m_pLogWindow = new wxLogWindow(this, wxs, false, false); + wxLog::EnableLogging(true); + wxLog::SetActiveTarget(m_pLogWindow); + m_pLogWindow->GetFrame()->SetName(sz_unit_test_log); + m_pLogWindow->GetFrame()->SetIcon(wxICON(AAArho)); + if (singletonApp->m_unit_test) { + wxLogMessage(_T("Command line specified unit test with%s exit on completion of unit test."), + singletonApp->m_display ? _T("out") : _T("")); + wxLogMessage(_T("If an error occurs during unit test, the program will return a non zero " + "error number on exit.")); + } + }else { + wxLog::EnableLogging(false); + wxLog::SetActiveTarget(nullptr); + m_pLogNull.reset(new wxLogNull()); + } + if (singletonApp->m_unit_test) singletonApp->Bind( + wxEVT_IDLE, + &UnitTest + ); + if (singletonApp->m_log_focus_events)wxLogMessage(_T("Logging focus events")); + if (singletonApp->m_params.empty()) { + wxLogMessage(_T("No wallet specified. Attempting to open last used wallet")); + }else { + wxString subcommands( _T("")); + for (auto& str : singletonApp->m_params) { + subcommands += str + _T(" "); + } + wxLogMessage(_T("command argument%s %s"), singletonApp->m_params.size()==1?"":"s", subcommands); + wxLogMessage(_T("attempting to open %s"), singletonApp->m_params[0]); + } + SetIcon(wxICON(AAArho)); //Does not appear to do anything. Maybe it does something in Unix. + //wxICON is a namestring on windows, and a symbol on Unix + Bind(wxEVT_CLOSE_WINDOW, &Frame::OnClose, this); + wxMenu* menuFile = new wxMenu; + menuFile->Append(wxID_NEW, _T("&New wallet..."), _T("New wallet file From new secret")); + menuFile->Bind(wxEVT_MENU, &Frame::OnSaveNew, this, wxID_NEW); + menuFile->Append(wxID_REFRESH, _T("Existing secret..."), _T("New Wallet File From the secret of an old wallet")); + menuFile->Bind(wxEVT_MENU, &Frame::RecreateWalletFromExistingSecret, this, wxID_REFRESH); + menuFile->Append(wxID_OPEN, _T("&Open existing wallet file..."), _T("")); + menuFile->Bind(wxEVT_MENU, &Frame::OnFileOpen, this, wxID_OPEN); + menuFile->Append(wxID_DELETE, _T("&Delete Wallet File"), _T("Delete ") + m_LastUsedSqlite.GetFullPath()); + menuFile->Bind(wxEVT_MENU, &Frame::OnDelete, this, wxID_DELETE); + menuFile->AppendSeparator(); + menuFile->Append(myID_DELETECONFIG, _T("&Reset defaults"), _T("Delete config file")); + menuFile->Bind(wxEVT_MENU, &Frame::OnDeleteConfiguration, this, myID_DELETECONFIG); + menuFile->AppendSeparator(); + menuFile->Append(wxID_EXIT); + menuFile->Bind(wxEVT_MENU, &Frame::OnExit, this, wxID_EXIT); + + wxMenu* menuHelp = new wxMenu; + menuHelp->Append(wxID_ABOUT); + menuHelp->Bind(wxEVT_MENU, &Frame::OnAbout, this, wxID_ABOUT); + menuHelp->Append(myID_ADD_SUBWINDOW, _T("insert new window"), _T("insert")); + menuHelp->Bind(wxEVT_MENU, &Frame::OnAddSubwindow, this, myID_ADD_SUBWINDOW); + menu_OnDeleteSubwindow.Append(menuHelp,this, _T("delete first subwindow"), _T("delete")); + menuHelp->Append(myID_DELETE_LAST_SUBWINDOW, _T("delete last subwindow"), _T("delete")); + menuHelp->Bind(wxEVT_MENU, &Frame::OnDeleteLastSubwindow, this, myID_DELETE_LAST_SUBWINDOW); +// menu_OnFirstUse.Append(menuHelp, this, _T("New wallet"), _T("add new wallet")); + wxMenuBar* menuBar = new wxMenuBar; + menuBar->Append(menuFile, "&File"); + menuBar->Append(menuHelp, "&Help"); + SetMenuBar(menuBar); + CreateStatusBar(); + // child controls + m_LastUsedSqlite.Assign(singletonApp->pConfig->Read(_T("/Wallet/LastUsed"), _T(""))); + wxString debug_data{ m_LastUsedSqlite.GetFullPath() }; + if (!m_LastUsedSqlite.IsOk() || !m_LastUsedSqlite.HasName() || !m_LastUsedSqlite.HasExt()) { + m_panel = new welcome_to_rhocoin(this); //Owner is "this", via the base class wxFrame. m_panel is a + // non owning pointer in the derived class that duplicates the owning pointer in the base class. + } + else { + display_wallet* panel = new display_wallet(this, m_LastUsedSqlite); + m_panel = panel; + } + this->RestorePositionFromConfig(ClientToWindowSize(m_panel->GetBestSize())); + SetClientSize(GetClientSize()); + } + catch (const std::exception& e) { + // cannot throw when no window is available. Construction of the base frame has to be completed, come what may. + // if an exception propagated from the constructor of the derived frame, it would destruct the base frame + // and chaos would ensue as a windowing program attempts to handle an error with no main window. + // so exceptions in the constructor of the main frame have to be caught and not rethrown. + queue_error_message(e.what()); + } +} +void Frame::OnExit(wxCommandEvent& event) { + Close(true); +} + +void Frame::OnClose(wxCloseEvent& event) { + // This event gives you the opportunity to clean up anything that needs explicit cleanup, albeit if you have done your work right nothing should need explicit cleanup, + // and to object to the closing in a "file not saved" type situation. + // https://docs.wxwidgets.org/trunk/classwx_close_event.html + if (sqlite3_shutdown())wxMessageBox(_T(R"|(Sqlite3 shutdown error)|"), _T("Error"), wxICON_ERROR); + Destroy(); //Default handler will destroy the window. This is our handler for the user calling close, replacing the default handler. +} + +void Frame::OnAbout(wxCommandEvent& event) +{ + wxMessageBox("This is a wxWidgets' Config sample", + "About sample code", wxOK | wxICON_INFORMATION); +} + +void Frame::OnDeleteConfiguration(wxCommandEvent&) +{ + std::unique_ptrpConfig{ wxConfigBase::Set(nullptr) }; + if (pConfig) + { + if (pConfig->DeleteAll()) + { + wxLogMessage(_T("Config file/registry key successfully deleted.")); + wxConfigBase::DontCreateOnDemand(); + pConfig.release(); + } + else + { + wxLogError(_T("Deleting config file/registry key failed.")); + } + } + else { + wxLogError(_T("No config to delete!")); + return; + } +} + + +void Frame::OnAddSubwindow(wxCommandEvent& WXUNUSED(event)) { + wxBoxSizer* sizer(static_cast(m_panel->GetSizer())); + sizer->Prepend(new wxStaticText(m_panel, myID_TESTWINDOW, _T("test"))); + sizer->Layout(); +} + +void Frame::OnDeleteSubwindow(wxCommandEvent& WXUNUSED(event)) { + auto sizer(m_panel->GetSizer()); + int item_number(0); + if (sizer->GetItemCount()) { + wxSizerItem* t1 = sizer->GetItem(item_number); + if (t1) { + auto t2 = t1->GetWindow(); + if (t2) { + sizer->Detach(t2); + t2->Destroy(); + sizer->Layout(); + } + } + } +} + +void Frame::OnDeleteLastSubwindow(wxCommandEvent& WXUNUSED(event)) { + auto sizer(m_panel->GetSizer()); + int item_number(sizer->GetItemCount()-1); + if (item_number >= 0) { + wxSizerItem* t1 = sizer->GetItem(item_number); + if (t1) { + auto t2 = t1->GetWindow(); + if (t2) { + sizer->Detach(t2); + t2->Destroy(); + sizer->Layout(); + } + } + } +} + +using ro::bin2hex, ro::to_base64_string; + +void Frame::NewWalletNewSecret(wxCommandEvent&) { + wxMessageBox(_T("new wallet new secret event"), _T("")); + /* If LastUsed is the empty string, check if default filename exists. If it exists, set Last Used to default file name + If LastUsed is not the empty string, or no longer the empty string, attempt to open indicated file. If open fails, or reading the opened file produces bad results, abort with exception + If LastUsed is still the empty string, attempt to create default filename. If creation fails, abort with exception. If it succeeds, set LastUsed to default filename. + The exception in unit test should simply generate an error message, but if run during initialization, should bring up the more complex UI for constructing or selecting your wallet file.*/ + ILogMessage("\tWallet file"); + /* + bool fWalletNameOk{ false }; + wxStandardPaths& StandardPaths(wxStandardPaths::Get()); + StandardPaths.UseAppInfo(3); + std::unique_ptr db; + if (fWalletNameOk) { + if (!m_LastUsedSqlite.FileExists()) + throw MyException((std::string("Expected wallet:\n") + LastUsedSqlite.GetFullPath() + "\nfile not found").c_str()); + db.reset(Sqlite3_open(m_LastUsedSqlite.GetFullPath().ToUTF8())); + sql_read_from_misc read_from_misc(db); + if (!read_from_misc(1) || read_from_misc.value() != WALLET_FILE_IDENTIFIER) + throw MyException(sz_unrecognizable_wallet_file_format); + if (!read_from_misc(2) || read_from_misc.value() != WALLET_FILE_SCHEMA_VERSION_0_0) + throw MyException(sz_unrecognized_wallet_schema); + if (!read_from_misc(4)) + throw MyException(sz_mastersecret_missing); + ristretto255::CMasterSecret MasterSecret(*read_from_misc.value()); + ro::sql read_keys(db, R"|(SELECT * FROM "Keys" LIMIT 5;)|"); + sql_read_name read_name(db); + // db.reset(nullptr);// Force error of premature destruction of Isqlite3 + while (read_keys.step() == Icompiled_sql::ROW) { + auto pubkey = read_keys.column(0); + auto id = read_keys.column(1); + auto use = read_keys.column(2); + if (use != 1)throw MyException(sz_unknown_secret_key_algorithm); + if (!read_name(id)) throw MyException(sz_no_corresponding_entry); + const char* name = read_name.name(); + if (MasterSecret(name).timesBase() != *pubkey)throw MyException(sz_name_does_not_correspond); + wxLogMessage(_T("\t\t\"%s\" has expected public key 0x%s"), name, (wxString)bin2hex(*pubkey)); + } + } + else { + // At this point in the code the filename LastUsedSqlite is a bad filename, normally the empty string, and the default wallet file does not exist in the default location. + // Construct default wallet and filename*//* + wxFileName path{ StandardPaths.GetUserLocalDataDir() }; + try { + // Disk operations to create wallet, which may throw. + // This try/catch block exists to catch disk io issues. + if (!path.DirExists())path.Mkdir(); + if (!DefaultSqlite.DirExists())DefaultSqlite.Mkdir(); + db.reset(Sqlite3_create(DefaultSqlite.GetFullPath().ToUTF8())); + db->exec(R"|( +PRAGMA journal_mode = WAL; +PRAGMA synchronous = 1; +BEGIN TRANSACTION; +CREATE TABLE "Keys"( + "pubkey" BLOB NOT NULL UNIQUE PRIMARY KEY, + "id" integer NOT NULL, + "use" INTEGER NOT NULL); + +CREATE TABLE "Names"( + "name" TEXT NOT NULL UNIQUE +); + +CREATE TABLE "Misc"( + "index" INTEGER NOT NULL UNIQUE PRIMARY KEY, + "m" BLOB +); +COMMIT;)|"); + LastUsedSqlite = DefaultSqlite; + singletonApp->pConfig->Write(_T("LastUsed"), DefaultSqlite.GetFullPath()); + wxLogMessage("\t\tConstructing default wallet %s", DefaultSqlite.GetFullPath()); + // We now have a working wallet file with no valid data. Attempting to create a strong random secret, a name, and public and private keys for that name. + + wxLogMessage("\t\tGenerating random 128 bit wallet secret"); + auto text_secret{ DeriveTextSecret(ristretto255::scalar::random(), 1) }; + ro::msec start_time{ ro::msec_since_epoch() }; + ristretto255::CMasterSecret MasterSecret(DeriveStrongSecret(&text_secret[0])); + decltype(start_time) end_time{ ro::msec_since_epoch() }; + wxLogMessage("\t\tStrong secret derivation took %d milliseconds", (end_time - start_time).count()); + sql_update_to_misc update_to_misc(db); + update_to_misc(1, WALLET_FILE_IDENTIFIER); + update_to_misc(2, WALLET_FILE_SCHEMA_VERSION_0_0); + update_to_misc(3, &text_secret[0]); + update_to_misc(4, MasterSecret); + sql_insert_name insert_name(db); + const char* const cpsz("Unit Tester"); + insert_name(cpsz, MasterSecret(cpsz).timesBase()); + } + catch (const MyException& e) { + ILogError(R"|(Failed to create or failed to properly initialize wallet)|"); + errorCode = 20; + szError = e.what(); + ILogError(szError.c_str()); + } + } // End of wallet creation branch*/ + +} + + +void Frame::OnSaveNew(wxCommandEvent& WXUNUSED(event)) +{ + wxFileDialog dialog(this, + sz_new_wallet_new_secret, + wxStandardPaths::Get().GetUserLocalDataDir(), + sz_default_wallet_name, + sz_wallet_files_title, + wxFD_SAVE | wxFD_OVERWRITE_PROMPT); + + dialog.SetFilterIndex(1); + + if (dialog.ShowModal() == wxID_OK) + { + wxLogMessage("%s, filter %d", + dialog.GetPath(), dialog.GetFilterIndex()); + } + auto wallet{ dialog.GetDirectory() + "/" + dialog.GetFilename() }; + wxLogMessage("new wallet created: %s", wallet); + wxConfigBase::Get()->Write("Wallet", wallet); +} + +void Frame::OnFileOpen(wxCommandEvent&) { + wxString directory{ _T("") }; + wxString file{ _T("") }; + if (m_LastUsedSqlite.IsOk()) { + directory = m_LastUsedSqlite.GetPath(); + file = m_LastUsedSqlite.GetFullName(); + } + + wxFileDialog + dialog(this, sz_open_wallet_file, directory, file, + sz_wallet_files_title, wxFD_OPEN | wxFD_FILE_MUST_EXIST | wxFD_CHANGE_DIR); + if (dialog.ShowModal() == wxID_CANCEL) + return; // the user changed idea... + wxLogMessage("Opening %s", dialog.GetPath()); + wxFileName walletfile(dialog.GetPath()); + display_wallet* panel = new display_wallet(this, walletfile); + if (m_panel)m_panel->Destroy(); //Destroy somehow manages to execute + // the correct derived destructor regardless of what kind of object it is. + m_panel = panel; + m_panel->Show(); +} + +void Frame::RecreateWalletFromExistingSecret(wxCommandEvent&) { + wxMessageBox(_T("new wallet existing secret event"), _T("")); + auto standardpaths = wxStandardPaths::Get(); + wxFileDialog dialog(this, + sz_new_wallet_existing_secret, + wxStandardPaths::Get().GetAppDocumentsDir(), + sz_default_wallet_name, + sz_wallet_files_title, + wxFD_SAVE | wxFD_OVERWRITE_PROMPT); + + dialog.SetFilterIndex(1); + + if (dialog.ShowModal() == wxID_OK) + { + wxLogMessage("%s, filter %d", + dialog.GetPath(), dialog.GetFilterIndex()); + } +} + +void Frame::OnDelete(wxCommandEvent& WXUNUSED(event)) +{ + singletonApp->pConfig->SetPath(_T("/Wallet")); + wxFileName LastUsedSqlite(singletonApp->pConfig->Read(_T("LastUsed"), _T(""))); + singletonApp->pConfig->DeleteEntry(_T("LastUsed")); + if (LastUsedSqlite.IsOk() && LastUsedSqlite.FileExists()) { + if (wxRemoveFile(LastUsedSqlite.GetFullPath()))wxLogMessage(_T("Deleting % s"), LastUsedSqlite.GetFullPath()); + } +} + +void Frame::OnMenuOpen(wxMenuEvent& evt) { + auto pMenu(evt.GetMenu()); + if (pMenu) { + auto label(pMenu->GetTitle()); + wxLogMessage(_T("Open menu \"%s\""), label); + } +} + +Frame::~Frame(){ + assert(singletonFrame == this); + singletonFrame = nullptr; + if (errorCode)return; + wxConfigBase *pConfig = wxConfigBase::Get(); + if (pConfig == nullptr)return; + StorePositionToConfig(); + pConfig->Write(_T("/Wallet/LastUsed"), m_LastUsedSqlite.GetFullPath()); +} \ No newline at end of file diff --git a/frame.h b/frame.h new file mode 100644 index 0000000..5059f29 --- /dev/null +++ b/frame.h @@ -0,0 +1,83 @@ +#pragma once +template +// This class exists to record the information +// needed to unbind a menu action and delete the corresponding item from the menu. +// when the handler is destroyed. +// Also avoids the need for manually creating a new windowid to link each additional bind +// to each menu item, thus avoids the likelihood of mismatching binds and menu entries. +class MenuLink { + void (handler_class::* m_method)(wxCommandEvent&); + int m_winid; + handler_class* m_handler = nullptr; + wxMenu* m_menu =nullptr; + wxMenuItem* m_menuItem = nullptr;; +public: + MenuLink( + void (handler_class::* method)(wxCommandEvent&)) : m_method{ method }, m_winid{ wxWindow::NewControlId()} {} + MenuLink( + void (handler_class::* method)(wxCommandEvent&), MyIDs winid) : m_method{ method }, m_winid{ winid } {} + ~MenuLink() { + if (m_menu != nullptr) { + m_menu->Unbind(wxEVT_MENU, m_method, m_handler, m_winid); + m_menu->Destroy(m_menuItem); + } + } + auto Append( + wxMenu* menu, + handler_class* handler, + const wxString& item = wxEmptyString, + const wxString& helpString = wxEmptyString + ) { + m_menu = menu; + m_handler = handler; + m_menuItem = menu->Append(m_winid, item, helpString); + menu->Bind(wxEVT_MENU, m_method, handler, m_winid); + return m_menuItem; + } + void queue_event() { + if (m_menu != nullptr) { + auto event = new wxCommandEvent(wxEVT_MENU, m_winid); + wxQueueEvent(m_menu, event); + } + else throw MyException("Event sent to uninitialized Menu item"); + } +}; + +class Frame : public wxFrame +{ +public: + wxLogWindow*m_pLogWindow{ nullptr }; + std::unique_ptrm_pLogNull{ nullptr }; + Frame(wxString); + ~Frame(); + wxFileName m_LastUsedSqlite; +private: + typedef MenuLink MenuLink; + wxPanel* m_panel; //The once current child panel. + void StorePositionToConfig(void); + void RestorePositionFromConfig(const wxSize&); + void OnExit(wxCommandEvent&); + void OnClose(wxCloseEvent&); + void OnAbout(wxCommandEvent&); + void OnDeleteConfiguration(wxCommandEvent&); +public: + void OnSaveNew(wxCommandEvent&); + void NewWalletNewSecret(wxCommandEvent&); + void RecreateWalletFromExistingSecret(wxCommandEvent&); + void OnFileOpen(wxCommandEvent&); +private: + void OnDelete(wxCommandEvent&); + void OnMenuOpen(wxMenuEvent&); + void OnDeleteSubwindow(wxCommandEvent&); + MenuLink menu_OnDeleteSubwindow = + MenuLink(&Frame::OnDeleteSubwindow); + void OnDeleteLastSubwindow(wxCommandEvent&); + void OnAddSubwindow(wxCommandEvent&); +public: + void OnFirstUse(wxCommandEvent&); + +public: +// MenuLink menu_OnFirstUse = +// MenuLink(&Frame::OnFirstUse, myID_WELCOME_TO_ROCOIN); +}; +inline Frame* singletonFrame(nullptr); diff --git a/introspection_of_standard_C_types.h b/introspection_of_standard_C_types.h new file mode 100644 index 0000000..a6f5498 --- /dev/null +++ b/introspection_of_standard_C_types.h @@ -0,0 +1,65 @@ +#pragma once +namespace ro { + + /* disjunction + variadic logical or metafunction + negation + logical NOT metafunction + conjunction + variadic logical AND metafunction */ + + template + inline constexpr bool _Is_any_of_v = // true if and only if T is in _Types + std::disjunction...>::value; + + template + constexpr bool is_standard_unsigned_integer = + _Is_any_of_v, unsigned char, unsigned short, unsigned int, unsigned long, unsigned long long>; + + template + constexpr bool is_standard_signed_integer = + _Is_any_of_v, signed char, signed short, signed int, signed long, signed long long>; + + // A compile time test to see if desired casts work, and make sure that + // undesired casts do not work + template struct has_constructor { + template static constexpr decltype(T(std::declval()), bool()) test() { + return true; + } + template static constexpr bool test(int = 0) { + return false; + } + static constexpr bool value = has_constructor::template test(); + }; + + template + inline constexpr bool is_constructible_from_any_of = //true if T is constructible from any of the listed types + std::disjunction...>::value; + + template + struct _Junction { // handle false trait or last trait + using type = _First; + }; + + template + struct _Junction { // first trait is false, try the next trait + using type = typename _Junction<_Next::value, _Next, _Rest...>::type; + }; + + template + struct junction : std::true_type {}; // If _Traits is empty, true_type + + template + struct junction<_First, _Rest...> : _Junction<_First::value, _First, _Rest...>::type { + // the first false trait in _Traits, or the last trait if none are false + }; + + template + inline constexpr bool is_constructible_from_all_of = //true if T is constructible from any of the listed types + junction...>::value; +} + +std::span& operator^=(std::span& lhs, byte* rhs) { + for (auto& j : lhs) { j ^= *rhs++; } + return lhs; +} \ No newline at end of file diff --git a/libsodium b/libsodium new file mode 160000 index 0000000..d30251f --- /dev/null +++ b/libsodium @@ -0,0 +1 @@ +Subproject commit d30251f03e646abd07b5399654f1f5dcea9a6b38 diff --git a/localization.cpp b/localization.cpp new file mode 100644 index 0000000..59ea323 --- /dev/null +++ b/localization.cpp @@ -0,0 +1,459 @@ +// Everything that will need translation should be in localization.cpp +// But this task is barely begun, there are no end of English language +// strings strewn at random about the program which need to be moved to +// localization.h and localization.cpp + +// Microsoft Windows, the Visual Studio Compiler, wxWidgets, +// and C++17 onwards is layer after layer of workarounds for the +// localization problem. Mutually conflicting workarounds that +// keep fouling each other up. The final solution to these problems +// is utf-8 and UTF16, and to make it work, you need wallet.manifest +// which gives you a UTF8 execution environment on program launch, +// as if we set the code page in the registry to +// Computer\HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Control\Nls\CodePage 65001 +// Instead of its normal value of 1252 +// and the /utf-8 compile option, +// Properties/C++/AdditionalOptions /utf-8 +// which gives you a utf-8 compilation and code generation environment. + +// with utf8 in effect at all three levels +// wxUSE_UNSAFE_WXSTRING_CONV == 1 +// is desirable, because all those automagical string conversions +// will no longer bite you. + +// We really should be running with wxUSE_UTF8_LOCALE_ONLY and +// wxUSE_UNICODE_UTF8, but this will require a wholesale rewrite of +// wxWidgets to use the windowsA calls rather than the windowsW calls, +// albeit this could largely be accomplished by redefining the windows +// macros to default to windowsA where at present they default to windowsW + +#include +#include "ILog.h" +#include "localization.h" + +// I intend that the download pages will in a location related to the git +// repository, and that when we support multiple languages, the end user will +// download the executable for his language, that we will build a complete +// executable and complete install for every language supported. + +// Since the network environment is utf-8, if the locale is not utf-8 +// your program has to keep track of two kinds of strings, which it +// will not do successfully. So, forcing environment to utf with +// a windows manifest and a unit test environment check. + +// Nearly all of these strings, all of them right now, are seven bit ascii, +// so unicode does not matter but it did a short time ago, and when we get +// around to localization, providing several versions of the localization +// file in several languages, it will matter again. + +// No intent to support non UTF-8 locales anywhere in the code. +// code is intentionally written to break on such locales. + +// No intent to support local date formats. +// All dates and times shall be YYYY-MM-DDtHH-mm-ss (international format) +// Local date formats are always ambiguous. Date localization is dangerous +// in a program intended to support promises and contracts. + +/* +Intended usage is that we have a different download, a different + separately compiled complete program for every locale, with a different + locale.cpp file and all the other files unchanged, rather than trying + to make a single compiled program work in a variety of locales. + +At present, there is only one locale.cpp, and if you have a better + solution for supporting multiple locales, you are probably better + qualified at supporting multiple locales than I am. + +However my experience is that all the terribly clever solutions for + supporting multiple locales with a single compiled program are fragile + and break every time they issue a new skew of the operating system. + I don't trust them, and are not interested in learning about them.*/ + +// Strings for the startup display, load existing wallet or create new +// wallet. +const char sz_NewWallet[] {R"|(New Wallet)|"}; +const char sz_new_wallet[] {R"|(new wallet)|"}; +const char sz_unit_test_log[] { "log of unit test" }; +const char sz_welcome_to_rocoin[] {R"|(Welcome to Rhocoin)|"}; +const char sz_please_select_from_three_options[]{ + R"|(Please select one of the following three options:)|" + }; +const char sz_create_a_new_wallet[]{R"|(Create a new wallet)|"}; +const char sz_recreate_wallet_from_old_secret[]{ + R"|(Recreate a wallet from an old wallet's secret)|" + }; +const char sz_existing_secret[] {R"|(existing secret)|" }; +const char sz_open_existing_wallet[]{R"|(Open an existing wallet)|"}; +const char sz_existing_wallet[] {R"|(existing wallet)|"}; +const char sz_text_buffer_overflow[]{R"|(conversion to base64 would overflow text buffer)|"}; +const char sz_unknown_error[]{ R"|(unknown error.)|"}; + +//Error message strings +const wchar_t wsz_error[] { L"Error" }; +const wchar_t wsz_operation[] { L"Operation aborted.\n" }; +const wchar_t wsz_program[]{ L"The program will terminate.\n" }; + +//Command line parameters +const wchar_t wsz_commandLineLogo[]{LR"(rho wallet +for secure communication and crypto currency transfers inside secure communication)"}; +const wchar_t wsz_usageText[]{ LR"( +Typical usage is: wallet +Subcommands are described in docs/wallet_subcommands.html)" }; + + +// Strings for the self test are not represented here, though they probably +// should be. + +FatalException::FatalException() noexcept : + MyException(R"|(unspecified fatal exception)|") {}; +HashReuseException::HashReuseException() noexcept : + MyException(R"|(finalized the same hash constructor twice or more)|") {}; +SQLexception::SQLexception() noexcept : + MyException(R"|(SQL Exception)|") {}; +BadDataException::BadDataException() noexcept : + MyException(R"|(Bad data exception)|") {}; +NonUtf8DataInDatabase::NonUtf8DataInDatabase() noexcept : + MyException(R"|(utf8 string in database is invalid utf8)|") {}; +NonRandomScalarException::NonRandomScalarException() noexcept : + MyException(R"|(Not a random scalar, something insecure is going on)|") {}; +BadScalarException::BadScalarException() noexcept : + MyException(R"|(Invalid scalar)|") {}; +BadStringRepresentationOfCryptoIdException::BadStringRepresentationOfCryptoIdException() noexcept : + MyException(R"|(Invalid string representation of a cryptographic identifier)|") {}; +NotBase58Exception::NotBase58Exception() noexcept : + MyException(R"|(Expected a base 58 character)|") {}; +OversizeBase58String::OversizeBase58String() noexcept : + MyException(R"|(Too long)|") {}; + + + +const char FailureToThrowExpectedException_sz[]{ "Exception not thrown for deliberate error in unit test" }; + + +// Random words for the passphrase. +// They should be replaced, rather than translated, +// with random short memorable words whose first characters +// are distinct. +const char* const ar_sz_bip_0039_wordlist[]{ + "abandon", "ability", "able", "about", "above", "absent", "absorb", + "abstract", "absurd", "abuse", "access", "accident", "account", + "accuse", "achieve", "acid", "acoustic", "acquire", "across", "act", + "action", "actor", "actress", "actual", "adapt", "add", "addict", + "address", "adjust", "admit", "adult", "advance", "advice", "aerobic", + "affair", "afford", "afraid", "again", "age", "agent", "agree", "ahead", + "aim", "air", "airport", "aisle", "alarm", "album", "alcohol", "alert", + "alien", "all", "alley", "allow", "almost", "alone", "alpha", "already", + "also", "alter", "always", "amateur", "amazing", "among", "amount", "amused", + "analyst", "anchor", "ancient", "anger", "angle", "angry", "animal", "ankle", + "announce", "annual", "another", "answer", "antenna", "antique", "anxiety", + "any", "apart", "apology", "appear", "apple", "approve", "april", "arch", + "arctic", "area", "arena", "argue", "arm", "armed", "armor", "army", "around", + "arrange", "arrest", "arrive", "arrow", "art", "artefact", "artist", "artwork", "ask", + "aspect", "assault", "asset", "assist", "assume", "asthma", "athlete", "atom", "attack", + "attend", "attitude", "attract", "auction", "audit", "august", "aunt", "author", "auto", + "autumn", "average", "avocado", "avoid", "awake", "aware", "away", "awesome", "awful", + "awkward", "axis", "baby", "bachelor", "bacon", "badge", "bag", "balance", "balcony", + "ball", "bamboo", "banana", "banner", "bar", "barely", "bargain", "barrel", "base", + "basic", "basket", "battle", "beach", "bean", "beauty", "because", "become", "beef", + "before", "begin", "behave", "behind", "believe", "below", "belt", "bench", "benefit", + "best", "betray", "better", "between", "beyond", "bicycle", "bid", "bike", "bind", + "biology", "bird", "birth", "bitter", "black", "blade", "blame", "blanket", "blast", + "bleak", "bless", "blind", "blood", "blossom", "blouse", "blue", "blur", "blush", + "board", "boat", "body", "boil", "bomb", "bone", "bonus", "book", "boost", + "border", "boring", "borrow", "boss", "bottom", "bounce", "box", "boy", "bracket", + "brain", "brand", "brass", "brave", "bread", "breeze", "brick", "bridge", "brief", + "bright", "bring", "brisk", "broccoli", "broken", "bronze", "broom", "brother", "brown", + "brush", "bubble", "buddy", "budget", "buffalo", "build", "bulb", "bulk", "bullet", + "bundle", "bunker", "burden", "burger", "burst", "bus", "business", "busy", "butter", + "buyer", "buzz", "cabbage", "cabin", "cable", "cactus", "cage", "cake", "call", + "calm", "camera", "camp", "can", "canal", "cancel", "candy", "cannon", "canoe", + "canvas", "canyon", "capable", "capital", "captain", "car", "carbon", "card", "cargo", + "carpet", "carry", "cart", "case", "cash", "casino", "castle", "casual", "cat", + "catalog", "catch", "category", "cattle", "caught", "cause", "caution", "cave", "ceiling", + "celery", "cement", "census", "century", "cereal", "certain", "chair", "chalk", "champion", + "change", "chaos", "chapter", "charge", "chase", "chat", "cheap", "check", "cheese", + "chef", "cherry", "chest", "chicken", "chief", "child", "chimney", "choice", "choose", + "chronic", "chuckle", "chunk", "churn", "cigar", "cinnamon", "circle", "citizen", "city", + "civil", "claim", "clap", "clarify", "claw", "clay", "clean", "clerk", "clever", + "click", "client", "cliff", "climb", "clinic", "clip", "clock", "clog", "close", + "cloth", "cloud", "clown", "club", "clump", "cluster", "clutch", "coach", "coast", + "coconut", "code", "coffee", "coil", "coin", "collect", "color", "column", "combine", + "come", "comfort", "comic", "common", "company", "concert", "conduct", "confirm", "congress", + "connect", "consider", "control", "convince", "cook", "cool", "copper", "copy", "coral", + "core", "corn", "correct", "cost", "cotton", "couch", "country", "couple", "course", + "cousin", "cover", "coyote", "crack", "cradle", "craft", "cram", "crane", "crash", + "crater", "crawl", "crazy", "cream", "credit", "creek", "crew", "cricket", "crime", + "crisp", "critic", "crop", "cross", "crouch", "crowd", "crucial", "cruel", "cruise", + "crumble", "crunch", "crush", "cry", "crystal", "cube", "culture", "cup", "cupboard", + "curious", "current", "curtain", "curve", "cushion", "custom", "cute", "cycle", "dad", + "damage", "damp", "dance", "danger", "daring", "dash", "daughter", "dawn", "day", + "deal", "debate", "debris", "decade", "december", "decide", "decline", "decorate", "decrease", + "deer", "defense", "define", "defy", "degree", "delay", "deliver", "demand", "demise", + "denial", "dentist", "deny", "depart", "depend", "deposit", "depth", "deputy", "derive", + "describe", "desert", "design", "desk", "despair", "destroy", "detail", "detect", "develop", + "device", "devote", "diagram", "dial", "diamond", "diary", "dice", "diesel", "diet", + "differ", "digital", "dignity", "dilemma", "dinner", "dinosaur", "direct", "dirt", "disagree", + "discover", "disease", "dish", "dismiss", "disorder", "display", "distance", "divert", "divide", + "divorce", "dizzy", "doctor", "document", "dog", "doll", "dolphin", "domain", "donate", + "donkey", "donor", "door", "dose", "double", "dove", "draft", "dragon", "drama", + "drastic", "draw", "dream", "dress", "drift", "drill", "drink", "drip", "drive", + "drop", "drum", "dry", "duck", "dumb", "dune", "during", "dust", "dutch", + "duty", "dwarf", "dynamic", "eager", "eagle", "early", "earn", "earth", "easily", + "east", "easy", "echo", "ecology", "economy", "edge", "edit", "educate", "effort", + "egg", "eight", "either", "elbow", "elder", "electric", "elegant", "element", "elephant", + "elevator", "elite", "else", "embark", "embody", "embrace", "emerge", "emotion", "employ", + "empower", "empty", "enable", "enact", "end", "endless", "endorse", "enemy", "energy", + "enforce", "engage", "engine", "enhance", "enjoy", "enlist", "enough", "enrich", "enroll", + "ensure", "enter", "entire", "entry", "envelope", "episode", "equal", "equip", "era", + "erase", "erode", "erosion", "error", "erupt", "escape", "essay", "essence", "estate", + "eternal", "ethics", "evidence", "evil", "evoke", "evolve", "exact", "example", "excess", + "exchange", "excite", "exclude", "excuse", "execute", "exercise", "exhaust", "exhibit", "exile", + "exist", "exit", "exotic", "expand", "expect", "expire", "explain", "expose", "express", + "extend", "extra", "eye", "eyebrow", "fabric", "face", "faculty", "fade", "faint", + "faith", "fall", "false", "fame", "family", "famous", "fan", "fancy", "fantasy", + "farm", "fashion", "fat", "fatal", "father", "fatigue", "fault", "favorite", "feature", + "february", "federal", "fee", "feed", "feel", "female", "fence", "festival", "fetch", + "fever", "few", "fiber", "fiction", "field", "figure", "file", "film", "filter", + "final", "find", "fine", "finger", "finish", "fire", "firm", "first", "fiscal", + "fish", "fit", "fitness", "fix", "flag", "flame", "flash", "flat", "flavor", + "flee", "flight", "flip", "float", "flock", "floor", "flower", "fluid", "flush", + "fly", "foam", "focus", "fog", "foil", "fold", "follow", "food", "foot", + "force", "forest", "forget", "fork", "fortune", "forum", "forward", "fossil", "foster", + "found", "fox", "fragile", "frame", "frequent", "fresh", "friend", "fringe", "frog", + "front", "frost", "frown", "frozen", "fruit", "fuel", "fun", "funny", "furnace", + "fury", "future", "gadget", "gain", "galaxy", "gallery", "game", "gap", "garage", + "garbage", "garden", "garlic", "garment", "gas", "gasp", "gate", "gather", "gauge", + "gaze", "general", "genius", "genre", "gentle", "genuine", "gesture", "ghost", "giant", + "gift", "giggle", "ginger", "giraffe", "girl", "give", "glad", "glance", "glare", + "glass", "glide", "glimpse", "globe", "gloom", "glory", "glove", "glow", "glue", + "goat", "goddess", "gold", "good", "goose", "gorilla", "gospel", "gossip", "govern", + "gown", "grab", "grace", "grain", "grant", "grape", "grass", "gravity", "great", + "green", "grid", "grief", "grit", "grocery", "group", "grow", "grunt", "guard", + "guess", "guide", "guilt", "guitar", "gun", "gym", "habit", "hair", "half", + "hammer", "hamster", "hand", "happy", "harbor", "hard", "harsh", "harvest", "hat", + "have", "hawk", "hazard", "head", "health", "heart", "heavy", "hedgehog", "height", + "hello", "helmet", "help", "hen", "hero", "hidden", "high", "hill", "hint", + "hip", "hire", "history", "hobby", "hockey", "hold", "hole", "holiday", "hollow", + "home", "honey", "hood", "hope", "horn", "horror", "horse", "hospital", "host", + "hotel", "hour", "hover", "hub", "huge", "human", "humble", "humor", "hundred", + "hungry", "hunt", "hurdle", "hurry", "hurt", "husband", "hybrid", "ice", "icon", + "idea", "identify", "idle", "ignore", "ill", "illegal", "illness", "image", "imitate", + "immense", "immune", "impact", "impose", "improve", "impulse", "inch", "include", "income", + "increase", "index", "indicate", "indoor", "industry", "infant", "inflict", "inform", "inhale", + "inherit", "initial", "inject", "injury", "inmate", "inner", "innocent", "input", "inquiry", + "insane", "insect", "inside", "inspire", "install", "intact", "interest", "into", "invest", + "invite", "involve", "iron", "island", "isolate", "issue", "item", "ivory", "jacket", + "jaguar", "jar", "jazz", "jealous", "jeans", "jelly", "jewel", "job", "join", + "joke", "journey", "joy", "judge", "juice", "jump", "jungle", "junior", "junk", + "just", "kangaroo", "keen", "keep", "ketchup", "key", "kick", "kid", "kidney", + "kind", "kingdom", "kiss", "kit", "kitchen", "kite", "kitten", "kiwi", "knee", + "knife", "knock", "know", "lab", "label", "labor", "ladder", "lady", "lake", + "lamp", "language", "laptop", "large", "later", "latin", "laugh", "laundry", "lava", + "law", "lawn", "lawsuit", "layer", "lazy", "leader", "leaf", "learn", "leave", + "lecture", "left", "leg", "legal", "legend", "leisure", "lemon", "lend", "length", + "lens", "leopard", "lesson", "letter", "level", "liar", "liberty", "library", "license", + "life", "lift", "light", "like", "limb", "limit", "link", "lion", "liquid", + "list", "little", "live", "lizard", "load", "loan", "lobster", "local", "lock", + "logic", "lonely", "long", "loop", "lottery", "loud", "lounge", "love", "loyal", + "lucky", "luggage", "lumber", "lunar", "lunch", "luxury", "lyrics", "machine", "mad", + "magic", "magnet", "maid", "mail", "main", "major", "make", "mammal", "man", + "manage", "mandate", "mango", "mansion", "manual", "maple", "marble", "march", "margin", + "marine", "market", "marriage", "mask", "mass", "master", "match", "material", "math", + "matrix", "matter", "maximum", "maze", "meadow", "mean", "measure", "meat", "mechanic", + "medal", "media", "melody", "melt", "member", "memory", "mention", "menu", "mercy", + "merge", "merit", "merry", "mesh", "message", "metal", "method", "middle", "midnight", + "milk", "million", "mimic", "mind", "minimum", "minor", "minute", "miracle", "mirror", + "misery", "miss", "mistake", "mix", "mixed", "mixture", "mobile", "model", "modify", + "mom", "moment", "monitor", "monkey", "monster", "month", "moon", "moral", "more", + "morning", "mosquito", "mother", "motion", "motor", "mountain", "mouse", "move", "movie", + "much", "muffin", "mule", "multiply", "muscle", "museum", "mushroom", "music", "must", + "mutual", "myself", "mystery", "myth", "naive", "name", "napkin", "narrow", "nasty", + "nation", "nature", "near", "neck", "need", "negative", "neglect", "neither", "nephew", + "nerve", "nest", "net", "network", "neutral", "never", "news", "next", "nice", + "night", "noble", "noise", "nominee", "noodle", "normal", "north", "nose", "notable", + "note", "nothing", "notice", "novel", "now", "nuclear", "number", "nurse", "nut", + "oak", "obey", "object", "oblige", "obscure", "observe", "obtain", "obvious", "occur", + "ocean", "october", "odor", "off", "offer", "office", "often", "oil", "okay", + "old", "olive", "olympic", "omit", "once", "one", "onion", "online", "only", + "open", "opera", "opinion", "oppose", "option", "orange", "orbit", "orchard", "order", + "ordinary", "organ", "orient", "original", "orphan", "ostrich", "other", "outdoor", "outer", + "output", "outside", "oval", "oven", "over", "own", "owner", "oxygen", "oyster", + "ozone", "pact", "paddle", "page", "pair", "palace", "palm", "panda", "panel", + "panic", "panther", "paper", "parade", "parent", "park", "parrot", "party", "pass", + "patch", "path", "patient", "patrol", "pattern", "pause", "pave", "payment", "peace", + "peanut", "pear", "peasant", "pelican", "pen", "penalty", "pencil", "people", "pepper", + "perfect", "permit", "person", "pet", "phone", "photo", "phrase", "physical", "piano", + "picnic", "picture", "piece", "pig", "pigeon", "pill", "pilot", "pink", "pioneer", + "pipe", "pistol", "pitch", "pizza", "place", "planet", "plastic", "plate", "play", + "please", "pledge", "pluck", "plug", "plunge", "poem", "poet", "point", "polar", + "pole", "police", "pond", "pony", "pool", "popular", "portion", "position", "possible", + "post", "potato", "pottery", "poverty", "powder", "power", "practice", "praise", "predict", + "prefer", "prepare", "present", "pretty", "prevent", "price", "pride", "primary", "print", + "priority", "prison", "private", "prize", "problem", "process", "produce", "profit", "program", + "project", "promote", "proof", "property", "prosper", "protect", "proud", "provide", "public", + "pudding", "pull", "pulp", "pulse", "pumpkin", "punch", "pupil", "puppy", "purchase", + "purity", "purpose", "purse", "push", "put", "puzzle", "pyramid", "quality", "quantum", + "quarter", "question", "quick", "quit", "quiz", "quote", "rabbit", "raccoon", "race", + "rack", "radar", "radio", "rail", "rain", "raise", "rally", "ramp", "ranch", + "random", "range", "rapid", "rare", "rate", "rather", "raven", "raw", "razor", + "ready", "real", "reason", "rebel", "rebuild", "recall", "receive", "recipe", "record", + "recycle", "reduce", "reflect", "reform", "refuse", "region", "regret", "regular", "reject", + "relax", "release", "relief", "rely", "remain", "remember", "remind", "remove", "render", + "renew", "rent", "reopen", "repair", "repeat", "replace", "report", "require", "rescue", + "resemble", "resist", "resource", "response", "result", "retire", "retreat", "return", "reunion", + "reveal", "review", "reward", "rhythm", "rib", "ribbon", "rice", "rich", "ride", + "ridge", "rifle", "right", "rigid", "ring", "riot", "ripple", "risk", "ritual", + "rival", "river", "road", "roast", "robot", "robust", "rocket", "romance", "roof", + "rookie", "room", "rose", "rotate", "rough", "round", "route", "royal", "rubber", + "rude", "rug", "rule", "run", "runway", "rural", "sad", "saddle", "sadness", + "safe", "sail", "salad", "salmon", "salon", "salt", "salute", "same", "sample", + "sand", "satisfy", "satoshi", "sauce", "sausage", "save", "say", "scale", "scan", + "scare", "scatter", "scene", "scheme", "school", "science", "scissors", "scorpion", "scout", + "scrap", "screen", "script", "scrub", "sea", "search", "season", "seat", "second", + "secret", "section", "security", "seed", "seek", "segment", "select", "sell", "seminar", + "senior", "sense", "sentence", "series", "service", "session", "settle", "setup", "seven", + "shadow", "shaft", "shallow", "share", "shed", "shell", "sheriff", "shield", "shift", + "shine", "ship", "shiver", "shock", "shoe", "shoot", "shop", "short", "shoulder", + "shove", "shrimp", "shrug", "shuffle", "shy", "sibling", "sick", "side", "siege", + "sight", "sign", "silent", "silk", "silly", "silver", "similar", "simple", "since", + "sing", "siren", "sister", "situate", "six", "size", "skate", "sketch", "ski", + "skill", "skin", "skirt", "skull", "slab", "slam", "sleep", "slender", "slice", + "slide", "slight", "slim", "slogan", "slot", "slow", "slush", "small", "smart", + "smile", "smoke", "smooth", "snack", "snake", "snap", "sniff", "snow", "soap", + "soccer", "social", "sock", "soda", "soft", "solar", "soldier", "solid", "solution", + "solve", "someone", "song", "soon", "sorry", "sort", "soul", "sound", "soup", + "source", "south", "space", "spare", "spatial", "spawn", "speak", "special", "speed", + "spell", "spend", "sphere", "spice", "spider", "spike", "spin", "spirit", "split", + "spoil", "sponsor", "spoon", "sport", "spot", "spray", "spread", "spring", "spy", + "square", "squeeze", "squirrel", "stable", "stadium", "staff", "stage", "stairs", "stamp", + "stand", "start", "state", "stay", "steak", "steel", "stem", "step", "stereo", + "stick", "still", "sting", "stock", "stomach", "stone", "stool", "story", "stove", + "strategy", "street", "strike", "strong", "struggle", "student", "stuff", "stumble", "style", + "subject", "submit", "subway", "success", "such", "sudden", "suffer", "sugar", "suggest", + "suit", "summer", "sun", "sunny", "sunset", "super", "supply", "supreme", "sure", + "surface", "surge", "surprise", "surround", "survey", "suspect", "sustain", "swallow", "swamp", + "swap", "swarm", "swear", "sweet", "swift", "swim", "swing", "switch", "sword", + "symbol", "symptom", "syrup", "system", "table", "tackle", "tag", "tail", "talent", + "talk", "tank", "tape", "target", "task", "taste", "tattoo", "taxi", "teach", + "team", "tell", "ten", "tenant", "tennis", "tent", "term", "test", "text", + "thank", "that", "theme", "then", "theory", "there", "they", "thing", "this", + "thought", "three", "thrive", "throw", "thumb", "thunder", "ticket", "tide", "tiger", + "tilt", "timber", "time", "tiny", "tip", "tired", "tissue", "title", "toast", + "tobacco", "today", "toddler", "toe", "together", "toilet", "token", "tomato", "tomorrow", + "tone", "tongue", "tonight", "tool", "tooth", "top", "topic", "topple", "torch", + "tornado", "tortoise", "toss", "total", "tourist", "toward", "tower", "town", "toy", + "track", "trade", "traffic", "tragic", "train", "transfer", "trap", "trash", "travel", + "tray", "treat", "tree", "trend", "trial", "tribe", "trick", "trigger", "trim", + "trip", "trophy", "trouble", "truck", "true", "truly", "trumpet", "trust", "truth", + "try", "tube", "tuition", "tumble", "tuna", "tunnel", "turkey", "turn", "turtle", + "twelve", "twenty", "twice", "twin", "twist", "two", "type", "typical", "ugly", + "umbrella", "unable", "unaware", "uncle", "uncover", "under", "undo", "unfair", "unfold", + "unhappy", "uniform", "unique", "unit", "universe", "unknown", "unlock", "until", "unusual", + "unveil", "update", "upgrade", "uphold", "upon", "upper", "upset", "urban", "urge", + "usage", "use", "used", "useful", "useless", "usual", "utility", "vacant", "vacuum", + "vague", "valid", "valley", "valve", "van", "vanish", "vapor", "various", "vast", + "vault", "vehicle", "velvet", "vendor", "venture", "venue", "verb", "verify", "version", + "very", "vessel", "veteran", "viable", "vibrant", "vicious", "victory", "video", "view", + "village", "vintage", "violin", "virtual", "virus", "visa", "visit", "visual", "vital", + "vivid", "vocal", "voice", "void", "volcano", "volume", "vote", "voyage", "wage", + "wagon", "wait", "walk", "wall", "walnut", "want", "warfare", "warm", "warrior", + "wash", "wasp", "waste", "water", "wave", "way", "wealth", "weapon", "wear", + "weasel", "weather", "web", "wedding", "weekend", "weird", "welcome", "west", "wet", + "whale", "what", "wheat", "wheel", "when", "where", "whip", "whisper", "wide", + "width", "wife", "wild", "will", "win", "window", "wine", "wing", "wink", + "winner", "winter", "wire", "wisdom", "wise", "wish", "witness", "wolf", "woman", + "wonder", "wood", "wool", "word", "work", "world", "worry", "worth", "wrap", + "wreck", "wrestle", "wrist", "write", "wrong", "yard", "year", "yellow", "you", + "young", "youth", "zebra", "zero", "zone", "zoo" +}; +static_assert(sizeof(ar_sz_bip_0039_wordlist) == sizeofwordlist * sizeof(ar_sz_bip_0039_wordlist[0]), "wrong size BIP39 word list"); + +// --------------------------------------------------------------------------- +// UTF-8 operations +// --------------------------------------------------------------------------- +// +// +// Table 3.1B from Unicode spec: Legal UTF-8 Byte Sequences +// +// Code Points | 1st Byte | 2nd Byte | 3rd Byte | 4th Byte | +// Code Points | b | c | d | e | +// -------------------+----------+----------+----------+----------+ +// U+0000..U+007F | 00..7F | | | | +// U+0080..U+07FF | C2..DF | 80..BF | | | +// U+0800..U+0FFF | E0 | A0..BF | 80..BF | | +// U+1000..U+FFFF | E1..EF | 80..BF | 80..BF | | +// U+10000..U+3FFFF | F0 | 90..BF | 80..BF | 80..BF | +// U+40000..U+FFFFF | F1..F3 | 80..BF | 80..BF | 80..BF | +// U+100000..U+10FFFF | F4 | 80..8F | 80..BF | 80..BF | +// -------------------+----------+----------+----------+----------+ +// The point and purpose of this table is to forbid non shortest +// form sequences, so that each possible unicode character has +// only one possible representation. + +// If non shortest forms are allowed, the attacker can potentially +// cause unexpected code paths to be taken, resulting in consequences +// the programmer did not plan for. + +// And, similarly, who knows what may happen if illegally large code points +// are allowed, though I think they were stupid to set a max character limit. +bool IsValidUtf8String(const char* sz){ + const uint8_t* str = static_cast(sz); + if (str==nullptr) + return true; // empty string is UTF8 string + while (*str) { + char b = *str++; + if (b < 80) {} + else if (b < 0xC2) // invalid lead bytes: 80..C1 + return false; + // two-byte sequences: + else if (b <= 0xDF) // C2..DF + { + auto c = *(str++); + if (b == 0xE0) { + if (c<0xA0 || 0xBF 0xF4)return false; + auto c = *(str++); + if (b == 0xF0) { + if (c<0x90 || 0xBF sp) { + static constexpr uint64_t two_to_the_64_divided_by_golden_ratio{ 11400714819323198485 }; + id *= two_to_the_64_divided_by_golden_ratio; + for (auto x : sp) { + id += x; + id *= two_to_the_64_divided_by_golden_ratio; //mixes and disperses the bits. + } + return id ^ (id >> 32); + } + + class charmap :public std::array { + public: + charmap() = delete; + constexpr charmap(const char* p, const char* q) { + while (unsigned int pu{ static_cast(*p++) }) { + assert((*this)[pu] == 0); + (*this)[pu] = *q++; + } + assert(*(p - 1) == '\0' && *q == '\0'); + } + }; + + //template <> class base58 : public CompileSizedString<44> {}; + static_assert(sizeof(base58) == 46, "base58 point strings unexpected size"); + + static constexpr char index2cryptoBase58[]{ "123456789ABCDEFGHJKLMNPQRSTUVWXYZabcdefghijkmnopqrstuvwxyz" }; + static constexpr char index2MpirBase58[]{ "0123456789ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuv" }; + + void map_base_from_mpir58(char* p) { + static const charmap map(index2MpirBase58, index2cryptoBase58); + while (unsigned int pu{ static_cast(*p) }) { + *p++ = map[pu]; + } + } + + void map_base_to_mpir58(const char* p, char* q, size_t count) { + static const charmap map(index2cryptoBase58, index2MpirBase58); + while (count--) { + unsigned int pu{ static_cast(*p++) }; + char c{ map[pu] }; + if (c == '\0') throw NotBase58Exception(); + *q++ = c; + } + *q = '\0'; + } + + // If the string does not fill the space, pushes it rightwards + // and pads with leading zeroes. + void right_justify_string(char* p, char* terminal_null) { + char* x = p; + for (; *x && x < terminal_null; x++) { + } + if (x != terminal_null) { + assert(x < terminal_null); + memmove(terminal_null - x + p, p, x - p); + memset(p, '0', terminal_null - x); + *terminal_null = '\0'; + } + else assert(*terminal_null == '\0'); + } +} \ No newline at end of file diff --git a/mpir_and_base58.h b/mpir_and_base58.h new file mode 100644 index 0000000..93102ac --- /dev/null +++ b/mpir_and_base58.h @@ -0,0 +1,183 @@ +#pragma once +/* This header handles the stuff related to arbitrary precision arithmetic values*/ + +namespace mpir { + class mpz { + mpz_t mpz_m; + public: + mpz() { + mpz_init2(mpz_m, 624); + } + mpz(const char* str, int base) { + mpz_init(mpz_m); + if (mpz_set_str(mpz_m, str, base))throw BadDataException(); + } + ~mpz() { + mpz_clear(mpz_m); + } + operator mpz_ptr() { + return mpz_m; + } + }; + + extern mpz ristretto25519_curve_order; +} + + + +// This global thread local object is explicitly constructed +// on the heap in the code on need once the thread starts running +// and destroyed when the thread exits. +// This explicit construction and destruction is a workaround because C++ lacks +// support for concurrent processes +// and is therefore unable to correctly handle non pod thread local objects. It will +// not correctly construct and +// destruct thread_local objects by itself, because the model machine does not have a model +// for threading. It has a pile of matchsticks and a tub of glue with which you can build +// your own model. It supports all the stuff you need for threads, but has no idea how all +// these moving parts fit together. +// So you have to construct and destruct your non pod thread local objects in code. +class thread_local__ { +public: + // These exist to avoid the high cost of repeatedly creating and destroying temporary mpz objects. + // They are expensive to create and destroy, because they use heap allocation. + // These are used all over the place as temporaries. + mpir::mpz q, r, n; +}; +extern thread_local thread_local__*thl; //Constructor in app.obj. +// Destructor does not seem to get called, hence using a pointer +// and destroying explicitly, rather than an std::unique_ptr +// needs testing to figure out what is going on + +namespace ro { + using ristretto255::scalar, ristretto255::point; + + auto fasthash(uint64_t, std::span)->uint32_t; + void right_justify_string(char*, char*); + bool is_alphanumeric_fixed_length(unsigned int, const char*); + + template typename std::enable_if< + ro::is_blob_field_type::value, + decltype(T::type_indentifier, uint32_t()) + >::type fasthash(const T& p_blob) { + static_assert(sizeof(T) % 8 == 0, "fasthash assumes a multiple of 8 bytes"); + return fasthash( + T::type_indentifier, + std::span< const uint64_t >( + reinterpret_cast(&p_blob.blob[0]), + p_blob.blob.size() / 8 + ) + ); + } + + void map_base_from_mpir58(char*); + void map_base_to_mpir58(const char*, char*, size_t); + + template class base58 : public CompileSizedString< + ((sizeof(T) * 8 + 4) * 4943ui64 + 28955ui64) / 28956ui64 - (std::is_same_v ? 1 : 0)> { + public: + // The rational number 4943 / 28956 is minisculy larger than log(2)/log(58) + // hence rounding up the nearest integer guarantees it will always be big enough. + base58() = default; + ~base58() = default; + base58(const T&); + base58(const char* p); + static const char* bin( + typename const decltype(T::type_indentifier, char())* p, + T& sc + ); + static void bin(const base58& str, T& sc); + static T bin(const char* str) { + T sc; + bin(str, sc); + return sc; + }; + T bin() const { + T sc; + bin(*this, sc); + return sc; + }; + operator T() const { + return bin(); + } + }; + + template typename const decltype(base58::length, T::type_indentifier, uint32_t()) + // cannot be consteval or constexpr, because has to be called after the mpir temp values are constructed + check_range() { + if (thl == nullptr)thl = new thread_local__(); + mpz_ui_pow_ui(thl->n, 58, base58::length); + + if constexpr (std::is_same_v) { + mpz_fdiv_q(thl->q, thl->n, mpir::ristretto25519_curve_order); + } + else { + mpz_fdiv_q_2exp(thl->q, thl->n, sizeof(T) * 8); + } + assert(mpz_cmp_ui(thl->q, UINT32_MAX) <= 0); + return static_cast(mpz_get_ui(thl->q)); + } + + template const uint32_t check_range_m{ check_range() }; + + template const char* base58::bin( + typename const decltype(T::type_indentifier, char())* p, + T& sc + ) { + const uint32_t range = check_range_m; + base58 strsc; + char* ps = strsc; + map_base_to_mpir58(p, ps, strsc.length); + if (p[base58::length] > ' ')throw OversizeBase58String(); + if (mpz_set_str(thl->n, ps, 58))throw BadStringRepresentationOfCryptoIdException(); + if constexpr (std::is_same_v) { + mpz_fdiv_qr(thl->q, thl->r, thl->n, mpir::ristretto25519_curve_order); + } + else { + mpz_fdiv_q_2exp(thl->q, thl->n, sizeof(sc.blob) * 8); + mpz_fdiv_r_2exp(thl->r, thl->n, sizeof(sc.blob) * 8); + } + size_t count; + mpz_export(&(sc.blob[0]), &count, -1, 1, -1, 0, thl->r); + if (count < sizeof(sc.blob))memset(&sc.blob[count], 0, sizeof(sc.blob) - count); + mpir_ui ck{ (static_cast(fasthash(sc)) * static_cast(range)) >> 32 }; + if (ck != mpz_get_ui(thl->q)) throw BadStringRepresentationOfCryptoIdException(); + return p + base58::length; + } + + template char* to_base58( + // does no string memory allocation, p has to point into a buffer, + // return value points to next position in the buffer, which is now null + typename decltype(check_range_m, char())* p, + const T& sc + ) { + mpir_ui ck{ (static_cast(fasthash(sc)) * static_cast(check_range_m)) >> 32 }; + mpz_import(thl->n, sizeof(sc.blob), -1, 1, -1, 0, &sc.blob[0]); + if constexpr (std::is_same_v) { + mpz_addmul_ui(thl->n, mpir::ristretto25519_curve_order, ck); + } + else { + mpz_set_ui(thl->r, ck); + mpz_mul_2exp(thl->r, thl->r, sizeof(sc.blob) * 8); + mpz_add(thl->n, thl->n, thl->r); + } + mpz_get_str(p, 58, thl->n); + char* terminal_null{ p + sizeof(base58) - 1 }; + right_justify_string(p, terminal_null); + map_base_from_mpir58(p); + return terminal_null; + // return value points to trailing null + } + + template base58::base58(const T& el) { + to_base58(static_cast(*this), el); + } + + template base58::base58(const char* p) { + memmove(this, p, (this->length)); + std::arraylength)> test; + map_base_to_mpir58(this, test, this->length); //Force an exception for bad char + if (p[this->length] > ' ')throw OversizeBase58String(); + this->operator char* [this->length] = '\0'; + } +} diff --git a/nmake/makefile b/nmake/makefile new file mode 100644 index 0000000..d6fb486 --- /dev/null +++ b/nmake/makefile @@ -0,0 +1,23 @@ +wallet.exe: stdafx.pch app.obj frame.obj Ilog.obj ISqlit3Impl.obj sqlite3.obj + link /OUT:wallet.exe /LIBPATH:%WXWIN%\lib\vc_x64_lib *.obj kernel32.lib user32.lib gdi32.lib winspool.lib comdlg32.lib advapi32.lib shell32.lib ole32.lib oleaut32.lib uuid.lib odbc32.lib odbccp32.lib Comctl32.lib Rpcrt4.lib + +stdafx.pch: ..\stdafx.cpp ..\app.h ..\frame.h ..\stdafx.h ..\ISqlite3.h + CL.exe /c /ZI /I%ROINCLUDE% /I%WXWIN%\lib\vc_x64_lib\mswud /JMC /W3 /WX- /diagnostics:classic /sdl /Od /D _DEBUG /D _WINDOWS /D _UNICODE /D UNICODE /Gm- /EHsc /RTC1 /MDd /GS /fp:precise /permissive- /Zc:wchar_t /Zc:forScope /Zc:inline /std:c++17 /Yc"stdafx.h" /Gd /TP /FC /errorReport:prompt ..\stdafx.cpp + +app.obj: ..\app.cpp stdafx.pch + CL.exe /c /ZI /JMC /W3 /WX- /diagnostics:classic /sdl /Od /D _DEBUG /D _WINDOWS /D _UNICODE /D UNICODE /Gm- /EHsc /RTC1 /MDd /GS /fp:precise /permissive- /Zc:wchar_t /Zc:forScope /Zc:inline /std:c++17 /Yu"stdafx.h" /Gd /TP /FC /errorReport:prompt ..\app.cpp +# don't need /I%ROINCLUDE% /I%WXWIN%\lib\vc_x64_lib\mswud because it is included in \Yu"stdafx.h" + +frame.obj: ..\frame.cpp stdafx.pch + CL.exe /c /ZI /JMC /W3 /WX- /diagnostics:classic /sdl /Od /D _DEBUG /D _WINDOWS /D _UNICODE /D UNICODE /Gm- /EHsc /RTC1 /MDd /GS /fp:precise /permissive- /Zc:wchar_t /Zc:forScope /Zc:inline /std:c++17 /Yu"stdafx.h" /Gd /TP /FC /errorReport:prompt ..\frame.cpp +# don't need /I%ROINCLUDE% /I%WXWIN%\lib\vc_x64_lib\mswud because it is included in \Yu"stdafx.h" + +ILog.obj: ..\ILog.cpp stdafx.pch + CL.exe /c /ZI /JMC /W3 /WX- /diagnostics:classic /sdl /Od /D _DEBUG /D _WINDOWS /D _UNICODE /D UNICODE /Gm- /EHsc /RTC1 /MDd /GS /fp:precise /permissive- /Zc:wchar_t /Zc:forScope /Zc:inline /std:c++17 /Yu"stdafx.h" /Gd /TP /FC /errorReport:prompt ..\Ilog.cpp +# don't need /I%ROINCLUDE% /I%WXWIN%\lib\vc_x64_lib\mswud because it is included in \Yu"stdafx.h" + +ISqlit3Impl.obj: ..\ISqlit3Impl.cpp ..\ISqlite3.h ..\sqlite3.h + CL.exe /c /ZI /I%ROINCLUDE% /JMC /W3 /WX- /diagnostics:classic /sdl /Od /D _DEBUG /D _WINDOWS /D _UNICODE /D UNICODE /Gm- /EHsc /RTC1 /MDd /GS /fp:precise /permissive- /Zc:wchar_t /Zc:forScope /Zc:inline /std:c++17 /Gd /TP /FC /errorReport:prompt ..\ISqlit3Impl.cpp + +sqlite3.obj: ..\sqlite3.c ..\sqlite3.h + CL.exe /c /ZI /JMC /nologo /W3 /WX- /diagnostics:classic /sdl /Od /D _DEBUG /D _WINDOWS /D _UNICODE /D UNICODE /Gm- /EHsc /RTC1 /MDd /GS /fp:precise /permissive- /Zc:wchar_t /Zc:forScope /Zc:inline /std:c++17 /Gd /TC /FC /errorReport:prompt ..\sqlite3.c diff --git a/pandoc_templates/style.css b/pandoc_templates/style.css new file mode 100644 index 0000000..8276cac --- /dev/null +++ b/pandoc_templates/style.css @@ -0,0 +1,29 @@ +body { + max-width: 30em; + margin-left: 1em; + } +p.center {text-align:center; + } + +table { + border-collapse: collapse; + } +td, th { + border: 1px solid #999; + padding: 0.5rem; + text-align: left; + } +code{white-space: pre-wrap; + } +span.smallcaps{font-variant: small-caps; + } +span.underline{text-decoration: underline; + } +div.column{display: inline-block; vertical-align: top; width: 50%; + } +div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em; + } +ul.task-list{list-style: none; + } +.display.math{display: block; text-align: center; margin: 0.5rem auto; + } \ No newline at end of file diff --git a/public_key.gpg b/public_key.gpg new file mode 100644 index 0000000..03f2f9d --- /dev/null +++ b/public_key.gpg @@ -0,0 +1,29 @@ +-----BEGIN PGP PUBLIC KEY BLOCK----- + +mDMEX1SfJRYJKwYBBAHaRw8BAQdAcXlEHKWZNw9o2Fj5ssAjCmsbthDz0DVzYKxF +pnwiDjm0JENoZW5nIFNoZW5nIDxDaGVuZ1NoZW5nQGZha2VtYWlsLmNoPoiQBBMW +CAA4AhsDBQsJCAcCBhUKCQgLAgQWAgMBAh4BAheAFiEECW6uFvuNYuddJDGZvESC +5JZzcRwFAmH4/lYACgkQvESC5JZzcRxRcAEA2uPehywNxbp9DON0CJ578vbmX7qt +OmDjsHovsQ5I1W8BAMbQyOcIRYQlPTijl3I/R4sWLZ9hEYIbP+FS3JtbDYAFiJAE +ExYIADgWIQQJbq4W+41i510kMZm8RILklnNxHAUCX1SfJQIbAwULCQgHAgYVCgkI +CwIEFgIDAQIeAQIXgAAKCRC8RILklnNxHPteAQDFs4dbz7I1a5VoKXMupmSW3MMP +uHChIilz9P2JYX6oOAEAx4+tmu7DIGwBgNQZKuxlVgWYavkuEGYkLYdGgIBpwQi4 +OARfVJ8lEgorBgEEAZdVAQUBAQdATTzhgBNNONqRkUDfj/dtps9n+BvZMcS1t6cD +tjLejT0DAQgHiHgEGBYIACAWIQQJbq4W+41i510kMZm8RILklnNxHAUCX1SfJQIb +DAAKCRC8RILklnNxHPB4AP4kKI590T81yFAEkcfKA3fFRxcDJwUqus2L4GpckyMq +ngEAv9viVp3aHvlXZMpQ4prvzC/Gpzs3TrsDXWEhg5PzDA+4MwRhhoZwFgkrBgEE +AdpHDwEBB0CuAAgmkYQetnz3BJ0ALBuVKn1aeBjMq+1sPKeyUDnjWIjvBBgWCAAg +FiEECW6uFvuNYuddJDGZvESC5JZzcRwFAmGGhnACGyIAgQkQvESC5JZzcRx2IAQZ +FggAHRYhBCuf+rsp/waNpunfl1ccOpw7nm/KBQJhhoZwAAoJEFccOpw7nm/KjykA +/2TcUdtOXUCRJsM+SvoG3xWjk7iCOuSTWsUkap9ha6daAQD2hLRL9/Z6c2mREKXz +7nR416Cs1YUACs98wTD5pBuYAk76AP9vr+uVoKbjf/q230SBNen2Hjhk8g+JD4Cu +ZpFKuVwZhQEAopykWKujX5dGU/1352QrUcoCVYbTMiXGIxL9CPrUzw+4MwRh2A+7 +FgkrBgEEAdpHDwEBB0A9rVFSCmvpkkeSCmB+PDYqnJsz7Fq0b9NkN87xQ854i4j1 +BBgWCAAmFiEECW6uFvuNYuddJDGZvESC5JZzcRwFAmHYD7sCGyIFCQHhM4AAgQkQ +vESC5JZzcRx2IAQZFggAHRYhBIDfnjfH6rZKSICJDtUTAeF2sxgoBQJh2A+7AAoJ +ENUTAeF2sxgogxsBAMAV8d12lHhnL64qbG+llKJKEXI1ZTSHLEcdXN5dOo6zAP9j +Na3cNcEwxreUlgUZhOjagxrMnGpd5l/pHVG/7ijECa/8AQDpNtaC1G/6bIwpekky +KdbZtliClpajSD7zoj2Bxe24nAD/TSwGmXzqr9+VeOctO/m6q+QKwFa3mGyOUWpp +oTNQfA0= +=Kusp +-----END PGP PUBLIC KEY BLOCK----- diff --git a/rho.ico b/rho.ico new file mode 100644 index 0000000..3847cc8 Binary files /dev/null and b/rho.ico differ diff --git a/rho.xpm b/rho.xpm new file mode 100644 index 0000000..808791c --- /dev/null +++ b/rho.xpm @@ -0,0 +1,151 @@ +/* XPM */ +static const char *const rho[] = { +"32 32 115 2 0 0", +" c #60FF60", +"! c #62FF62", +"# c #64FF64", +"$ c #63FF63", +"% c #61FF61", +"& c #5CE55C", +"' c #347A34", +"( c #1E3A1E", +") c #0E1C0E", +"* c #182818", +"+ c #2A5B2A", +", c #4CB44C", +"- c #61FA61", +". c #2E642E", +"/ c #000000", +"0 c #121E12", +"1 c #0D0D0D", +"2 c #1A311A", +"3 c #5EEB5E", +"4 c #5CEF5C", +"5 c #132713", +"6 c #203E20", +"7 c #65FF65", +"8 c #58D458", +"9 c #1E3F1E", +": c #0F1E0F", +"; c #62F262", +"< c #60FE60", +"= c #204220", +"> c #010301", +"? c #52CA52", +"@ c #386F38", +"A c #255225", +"B c #4DB84D", +"C c #060A06", +"D c #60FC60", +"E c #254E25", +"F c #040904", +"G c #5AD65A", +"H c #5FFC5F", +"I c #295C29", +"J c #111A11", +"K c #5BE35B", +"L c #020102", +"M c #030603", +"N c #378437", +"O c #5EFB5E", +"P c #121F12", +"Q c #0A1C0A", +"R c #5EF85E", +"S c #275527", +"T c #020302", +"U c #295329", +"V c #60F960", +"W c #070507", +"X c #5FFB5F", +"Y c #439D43", +"Z c #010201", +"[ c #193819", +"] c #5DF65D", +"^ c #60FD60", +"_ c #4FC14F", +"` c #010101", +"a c #173517", +"b c #59D359", +"c c #254C25", +"d c #59D159", +"e c #020502", +"f c #5FFD5F", +"g c #4ABA4A", +"h c #4FBD4F", +"i c #121912", +"j c #5FFA5F", +"k c #3A873A", +"l c #132813", +"m c #63F163", +"n c #152815", +"o c #010401", +"p c #48AD48", +"q c #409340", +"r c #429542", +"s c #2C5B2C", +"t c #050805", +"u c #51C151", +"v c #5FFF5F", +"w c #46A146", +"x c #234723", +"y c #0A150A", +"z c #316A31", +"{ c #1D391D", +"| c #367736", +"} c #346F34", +"~ c #152715", +" ! c #397C39", +"!! c #5DE25D", +"#! c #397E39", +"$! c #245424", +"%! c #295B29", +"&! c #3E903E", +"'! c #61EF61", +"(! c #0B1C0B", +")! c #63F563", +"*! c #0D1E0D", +"+! c #65ED65", +",! c #000100", +"-! c #152615", +".! c #59DE59", +"/! c #183218", +"0! c #52C452", +"1! c #214521", +"2! c #43A343", +"3! c #2F652F", +"4! c #2E672E", +"5! c #3B8B3B", +"6! c None", +"6!6!6!6!6!6!6!6!6!6!6!6! 6!6!6!6!6!6!6!6!6!6!6!6!", +"6!6!6!6!6!6!6!6!6! 6!6!6!6!6!6!6!6!6!", +"6!6!6!6!6!6!6! ! # $ % 6!6!6!6!6!6!6!", +"6!6!6!6!6!6! % & ' ( ) * + , # 6!6!6!6!6!6!", +"6!6!6!6!6! - . / 0 1 / / / / 2 3 6!6!6!6!6!", +"6!6!6!6! 4 5 6 7 $ # 8 9 / / / : ; 6!6!6!6!", +"6!6!6! < = > ? % @ / / > A # 6!6!6!", +"6!6! < B / C 7 D E / / F G 6!6!", +"6!6! H I / J ! K L / M N ! 6!6!", +"6! O P / Q ! R S / T U ! 6!", +"6! V W / Q ! X Y / Z [ $ 6!", +"6! ] / / Q ! ^ _ / ` a $ 6!", +" ] / / Q ! < b / T c ! ", +" ] / / Q ! < d / e ' ! ", +" ] / / Q ! f g / F h % ", +" ] / / i $ j k / l # ", +" ] / / M m R n o p % ", +" ] / / T q % H r / s # ", +" ] / / / t u $ v $ w / x 7 ", +" ] / / y z / { | } ~ / !# ", +"6! ] / / Q $ !!#!$!%!&!'!% 6!", +"6! ] / / (!! % % 6!", +"6! )!/ / *!! 6!", +"6!6! +!/ ,!-!! 6!6!", +"6!6! .!/ ,!/!! 6!6!", +"6!6!6! < 0!/ ` 1!% 6!6!6!", +"6!6!6!6! < 2!/ Z 3!% 6!6!6!6!", +"6!6!6!6!6! f 4!/ / 5!% 6!6!6!6!6!", +"6!6!6!6!6!6! f < < % 6!6!6!6!6!6!", +"6!6!6!6!6!6!6! 6!6!6!6!6!6!6!", +"6!6!6!6!6!6!6!6!6! 6!6!6!6!6!6!6!6!6!", +"6!6!6!6!6!6!6!6!6!6!6!6! 6!6!6!6!6!6!6!6!6!6!6!6!" +}; diff --git a/ristretto255.cpp b/ristretto255.cpp new file mode 100644 index 0000000..b311e1f --- /dev/null +++ b/ristretto255.cpp @@ -0,0 +1,131 @@ +#include "stdafx.h" +void randombytes_buf(std::span in) { randombytes_buf(&in[0], in.size_bytes()); } +void randombytes_buf(std::span< char> in) { randombytes_buf(&in[0], in.size_bytes()); } +bool operator ==(const std::span& p, const std::span& q) { + bool breturn{ true }; + for (auto xq = q.begin(); auto xp:p) { + if (xp != *xq++) { + breturn = false; + break; + } + } + return breturn; +} + +bool operator !=(const std::span& p, const std::span& q) { + bool breturn{ false }; + for (auto xq = q.begin(); auto xp:p) { + if (xp != *xq++) { + breturn = true; + break; + } + } + return breturn; +} + +namespace ristretto255 { + bool scalar::constant_time_required{ true }; + bool point::constant_time_required{ true }; + + scalar::scalar(int i) { + if (i >= 0) { + auto k{ unsigned int(i) }; + for (auto& j : blob) { j = k; k = k >> 8; } + } + else{ + std::array absdata; + auto k{ unsigned int(-i) }; + for (auto& j : absdata) { j = k; k = k >> 8; } + crypto_core_ristretto255_scalar_negate(&blob[0], &absdata[0]); + } + } + + point point::operator*(const scalar &sclr) const& noexcept { + point me; + auto i{ crypto_scalarmult_ristretto255(&me.blob[0], &sclr.blob[0], &blob[0]) }; + assert(i == 0); + return me; + } + + point point::operator*(int i) const& noexcept { + return scalar(i) * (*this); + } + + point point::ptZero({ + 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00}); + point point::ptBase({ + 0xe2, 0xf2, 0xae, 0x0a, 0x6a, 0xbc, 0x4e, 0x71, 0xa8, 0x84, 0xa9, 0x61, 0xc5, 0x00, 0x51, 0x5f, 0x58, 0xe3, 0x0b, 0x6a, 0xa5, 0x82, 0xdd, 0x8d, 0xb6, 0xa6, 0x59, 0x45, 0xe0, 0x8d, 0x2d, 0x76}); + + + signed_text::signed_text( + const scalar &k, // Signer's secret key + const point &Kin, // Signer's public key + std::span txtin // Text to be signed. + ) + : txt(txtin), K(Kin) + { + assert(k.valid()); + assert(K.valid()); + + /* • Compute r = H(k|M), a per message private key. Or r can be an unpredictable secret scalar. + • Compute R = r*B, a per message public key
+ • Compute c = H(R|M).
+ • Compute s = r+c*k. (Note that sB = r*B+c*k*B = R+c*K.)*/ + scalar r(hash<512>(k, txtin)); + point R(r.timesBase()); + c = scalar (hash<512>(R, txtin)); + s = r + c * k; + } + + bool signed_text::verify() { + /* • Check that c, s are valid scalars + • Check that K, the signing public key, is a valid member of the prime order group + • Compute R = sB − cK + • Check that c = H(R | M) */ + if (!c.valid() || !s.valid() || !K.valid())return false; + else { + point R = s.timesBase() - c * K; + return c==scalar(hash<512>(R, txt)); + } + } + +} //end ristretto255 namespace +using ristretto255::scalar, ro::CompileSizedString, ro::bin2hex, ro::to_base64_string; + +auto reverse_byte_order(std::array < uint8_t, sizeof(scalar)>const& ac) { + std::array ar; + auto p = ar.rbegin(); + for (auto x : ac) { + *p++ = x; + } + return ar; +} + +template<> CompileSizedString < 2 * sizeof(scalar)> + bin2hex(const scalar& sc) { + CompileSizedString <2 * sizeof(scalar)>sz; + auto bigendian = reverse_byte_order(sc.blob); + sodium_bin2hex(&sz[0], sizeof(sc.blob) * 2 + 1, &bigendian[0], bigendian.size()); + return sz; +} + +template<> scalar ro::hex2bin(const CompileSizedString< (2 * sizeof(scalar))>& sz) { + scalar sc; + sc.blob = reverse_byte_order(ro::hex2bin(sz)); + return sc; + } + +template<> CompileSizedString < (sizeof(scalar) * 8 + 5) / 6> + to_base64_string(const scalar& sc) { + CompileSizedString < (sizeof(sc.blob) * 8 + 4) / 6> sz; + auto bigendian = reverse_byte_order(sc.blob); + bits2base64( + &(bigendian[0]), 4, sizeof(bigendian)*8-4, + std::span(sz) + ); + return sz; +} + +namespace mpir { + mpz ristretto25519_curve_order("1000000000000000000000000000000014def9dea2f79cd65812631a5cf5d3ed", 16); +} diff --git a/ristretto255.h b/ristretto255.h new file mode 100644 index 0000000..39c67bc --- /dev/null +++ b/ristretto255.h @@ -0,0 +1,814 @@ +#pragma once +// libsodium is typeless, which I find confusing. Everything is unsigned +// bytes pointed at by naked pointers. It is well written assembly language +// written in C It is not even C written in C, let alone C++ written in C. +// +// NaCl provides a high level C++ interface to the low level C libsodium +// interface, but I just don't like the people running NaCl and suspect them +// of being in bed with my enemies, whereas I do like the people running +// LibSodium. +// +// It is nonetheless probably stupid to write my own high level interface to +// LibSodium, when one has already been written, but I am going to do so +// anyway, because elliptic curves are just way cool, and because ... +// because ... I am having trouble thinking up a good excuse ... ah yes, +// because when I read the NaCl documentation it says "X has not been +// implemented yet" when X is something that is pretty easy to implement +// if one has direct access to elliptic curve objects, and our Merkle patricia +// blockchain is going to be a great big pile of elliptic curve objects. +// NaCl suffers from the chronic disease both of open source projects, +// and also of high level interfaces to low level data structures, of getting +// fine tuned to the implementer's pet projects, which fine tuning is apt to +// get in the way of someone else's pet project. +// +// LibSodium, being assembly language written in C, rather than C written +// in C, or C++ written in C, has the problems: +// If you mix up the pointer to one kind of object with the pointer to +// another kind of object, you are sol +// If you mix up a pointer to a thirty two byte object with a pointer to a +// sixtyfour byte object, you are really sol +// if you reference something that has been de-allocated, you are sol. +// If you reference something that has not been properly initialized, +// you are sol. +// It is all raw memory, and structure exists only in the head of the +// programmer. The compiler knows nothing of these structures. +// +// That said, there is an immense amount of cryptographic knowledge +// encapsulated in this library, and I need to lift that knowledge into the +// language of C++20, from C that is almost assembly. +// There is a huge amount of knowledge embedded, and translating it +// from what is almost a machine language representation to a C++20 +// representation involves a big pile of stuff. + +// We are going to need to lift +// https://paragonie.com/blog/2017/06/libsodium-quick-reference-quick-comparison-similar-functions-and-which-one-use or their ristretto equivalents to C++ + +// Starting with ristretto points and scalars, but they are useless without a +// pile of other things, many of those other things being in +// https://download.libsodium.org/doc/helpers/ +// I went there to find out what the hell "sodium_is_zero" means, but +// found a pile of other things that I am going to need, and got distracted +// by no end of odds and ends that I am going to need to be to lift to +// C++20 in order for ristretto points and scalars to be put to any use. + + +void randombytes_buf(std::span in); +void randombytes_buf(std::span in); +bool operator !=(const std::span&, const std::span&); +bool operator ==(const std::span&, const std::span&); +namespace ro { + + // Decay to pointer is dangerously convenient, + // but in some situations it is just convenient + // This class provides an std:array one larger + // than the compile time string size, which decays + // to char*, std::string, and wxString + // In some code, this is ambiguous, so casts + // must sometimes be explicitly invoked. + template + class CompileSizedString : public std::array{ + public: + static constexpr int length{ stringlen }; + CompileSizedString() { + *(this->rbegin()) = '0'; + } + operator char* () & { + char* pc = &(static_cast*>(this)->operator[](0)); + return pc; + } + + operator const char* () const& { + const char* pc = &(static_cast*>(this)->operator[](0)); + return pc; + } + operator const char* () const&& { + const char* pc = &(static_cast*>(this)->operator[](0)); + return pc; + } + operator std::string() const& { + return std::string((const char*)*this, this->length); + } + operator std::string() const&& { + return std::string((const char*)*this, this->length); + } + operator wxString() const& { + return wxString::FromUTF8Unchecked((const char*)(*this)); + } + operator std::span() const& { + return std::span(static_cast((char*)*this), stringlen + 1); + } + operator wxString() const&& { + return wxString::FromUTF8Unchecked((const char*)(*this)); + } + operator std::span() const&& { + return std::span(static_cast((char*)*this), stringlen + 1); + } + }; + + // This template generates a span over const unsigned bytes for + // any range over bytes, such as a C array or an std::array + template < typename T> + std::enable_if_t< + sizeof(std::size(std::declval())) >= sizeof(int) && + sizeof(std::declval()[0]) == 1, + std::span + > serialize(const T& a) { + return std::span(static_cast(static_cast(&a[0])), std::size(a)); + } + + // Compile time test to see if a type has a blob array member + // This can be used in if constexpr (is_blob_field_type::value) + // By convention, blob fields are an std::array of unsigned bytes + // therefore already serializable. + template struct is_blob_field_type{ + template static constexpr decltype(std::declval().blob.size(), bool()) test() { + return sizeof(std::declval().blob[0])==1; + } + template static constexpr bool test(int = 0) { + return false; + } + static constexpr bool value = is_blob_field_type::template test(); + }; + + // At present our serial classes consist of std::span and custom classes that publicly inherit from std::span + // To handle compound objects, add custom classes inheriting from std::span[n] + + // template class that generates a std::span of bytes over the blob + // field of any object containing a blob record, which is normally sufficient + // for a machine independent representation of that object + template < + typename T, + decltype(std::size(std::declval().blob)) dummy_arg = 0 + > auto serialize(const T& pt) { + return serialize(pt.blob); + } + + // method that assumes that any char * pointer points a null terminated string + // and generates a std::span of bytes over the string including the terminating + // null. + // we assume the string is already machine independent, which is to say, we assume + // it comes from a utf8 locale. + + inline auto serialize(const char* sp) { return std::span(static_cast(static_cast(sp)), strlen(sp) + 1); } + + inline auto serialize(const decltype(std::declval().ToUTF8()) sz){ + return serialize(static_cast(sz)); + } + /* + inline auto serialize(const wxString& wxstr) { + return serialize(static_cast(wxstr.ToUTF8())); + } + If we allowed wxwidgets string to be serializable, all sorts of surprising things + would be serializable in surprising ways, because wxWidgets can convert all + sorts of things into strings that you were likely not expecting, in ways + unlikely to be machine independent, so you if you give an object to be + hashed that you have not provided some correct means for serializing, C++ is + apt to unhelpfully and unexpectedly turn it into a wxString, + + If you make wxStrings hashable, suprising things become hashable. + */ + + // data structure containing a serialized signed integer. + template, int> = 0> + class userial : public std::span { + public: + std::array::digits + 6) / 7> blob; + userial(T i) { + byte* p = &blob[0] + sizeof(blob); + *(--p) = i & 0x7f; + i >>= 7; + while (i != 0) { + *(--p) = (i & 0x7f) | 0x80; + i >>= 7; + } + assert(p >= &blob[0]); + *static_cast*>(this) = std::span(p, &blob[0] + sizeof(blob));; + } + }; + + // data structure containing a serialized signed integer. + template, int> = 0> + class iserial : public std::span { + public: + std::array::digits + 7) / 7> blob; + iserial(T i) { + // Throw away the repeated leading bits, and g + byte* p = &blob[0] + sizeof(blob); + unsigned count; + if (i < 0) { + size_t ui = i; + count = (std::numeric_limits::digits - std::countl_one(ui)) / 7; + } + else { + size_t ui = i; + count = (std::numeric_limits::digits - std::countl_zero(ui)) / 7; + } + *(--p) = i & 0x7f; + while (count-- != 0) { + i >>= 7; + *(--p) = (i & 0x7f) | 0x80; + } + assert(p >= &blob[0]); + *static_cast*>(this) = std::span(p, &blob[0] + sizeof(blob));; + } + }; + + + // converts machine dependent representation of an integer + // into a span pointin at a compact machine independent representation of an integer + // The integer is split into seven bit nibbles in big endian order, with the high + // order bit of the byte indicating that more bytes are to come. + // for an unsigned integer, all high order bytes of the form 0x80 are left out. + // for a positive signed integer, the same, except that the first byte + // of what is left must have zero at bit 6 + // for a negative signed integer, all the 0xFF bytes are left out, except + // that the first byte of what is left must have a one bit at bit six. + // + // small numbers get compressed. + // primarily used by hash and hsh so that the same numbers on different + // machines will generate the same hash + template std::enable_if_t, ro::userial > + serialize(T i) { + return userial(i); + /* we don't need all deserialize functions to have the same name, + indeed they have to be distinct because serialized data contains + no type information, but for the sake of template code we need all + things that may be serialized to be serialized by the serialize + command, so that one template can deal with any + number of serializable types */ + } + template std::enable_if_t, ro::iserial >serialize(T i) { + return iserial(i); + /* we don't need all deserialize functions to have the same name, but for the sake of template code we need all + things that may be serialized to be serialized by the serialize command, so that one template can deal with any + number of serializable types */ + } + +// Turns a compact machine independent representation of an uninteger +// into a 64 bit signed integer + template std::enable_if_t, T > + deserialize(const byte* p) { + auto oldp = p; + T i; + if (*p & 0x40)i = -64; + else i = 0; + while (*p & 0x80) { + i = (i | (*p++ & 0x7F)) << 7; + } + if (p - oldp > (std::numeric_limits::digits + 6) / 7)throw BadDataException(); + return i | *p; + } + // Turns a compact machine independent representation of an integer + // into a 64 bit unsigned integer + template std::enable_if_t, T > + deserialize(const byte * p) { + auto oldp = p; + T i{ 0 }; + while (*p & 0x80) { + i = (i | (*p++ & 0x7F)) << 7; + } + if (p - oldp > 9)throw BadDataException(); + return i | *p; + } + + // Turns a compact machine independent representation of an integer + // into a 64 bit signed integer + template std::enable_if_t || is_standard_unsigned_integer, T > + deserialize(std::span g) { + byte* p = static_cast(&g[0]); + T i{ deserialize(p) }; + if (p > &g[0]+g.size())throw BadDataException(); + return i; + } + + /* + It will be about a thousand years before numbers larger than 64 bits + appear in valid well formed input, and bad data structures have to be + dealt with a much higher level that knows what the numbers mean, + and deals with them according to their meaning + + Until then the low level code will arbitrarily truncate numbers larger + than sixty four bits, but numbers larger than sixty four bits are + permissible in input, are valid at the lowest level. + + We return uint64_t, rather than uint_fast64_t to ensure that all + implementations misinterpret garbage and malicious input in the + same way. + We cannot protect against Machiavelli perverting the input, so we + don't try very hard to prevent Murphy perverting the input, + but we do try to prevent Machiavelli from perverting the input in + ways that will induce peers to disagree. + + We use an explicit narrow_cast, rather than simply declaring th + function to be uint64_t, in order to express the intent to uniformly + force possibly garbage data being deserialized to standardized + garbage. + + We protect against malicious and ill formed data would cause the + system to go off the rails at a point of the enemy's choosing, + and we protect against malicious and ill formed data that one party + might interpret in one way, and another party might interpret in a + different way. + + Ill formed data that just gets converted into well formed, but + nonsense data can cause no harm that well formed nonsense data + could not cause. + + It suffices, therefore, to ensure that all implementations misinterpret + input containing unreasonably large numbers as the same number. + + Very large numbers are valid in themselves, but not going to be valid + as part of valid data structures for a thousand years or so. + + The largest numbers occurring in well formed valid data will be + currency amounts, and the total number of the smallest unit of + currency is fixed at 2^64-1 which will suffice for a thousand years. + Or we might allow arbitrary precision floating point with powers of + a thousand, so that sensible numbers to a human are represented by + sensible numbers in the actuall representation. + + secret keys, scalars are actually much larger numbers, modulo + 0x1000000000000000000000000000000014def9dea2f79cd65812631a5cf5d3ecU + but they are represented in a different format, their binary format + being fixed size low endian format, as 256 bit numbers, though only + 253 bits are actually needed and used, and their human readable + format being 44 digits in a base 58 representation.*/ + + // a compile time test to check if an object class has a machine independent representation + template struct is_serializable{ + template + static constexpr decltype(ro::serialize(std::declval()), bool()) test() { + if constexpr (sizeof...(Args2) > 0) { + return is_serializable::template test(); + } + else { + return true; + } + } + template static constexpr bool test(int = 0) { + return false; + } + static constexpr bool value = is_serializable::template test(); + }; + static_assert(ro::is_serializable>::value); + static_assert(ro::is_serializable>::value); + + template ro::CompileSizedString< (2 * sizeof(T))>bin2hex(const T& pt) { + ro::CompileSizedString< (2 * sizeof(T))>sz; + sodium_bin2hex(&sz[0], sizeof(pt.blob) * 2 + 1, &pt.blob[0], pt.blob.size()); + return sz; + } + + template T hex2bin(const ro::CompileSizedString< (2 * sizeof(T))>& sz){ + T pt; + size_t bin_len{ sizeof(T) }; + sodium_hex2bin( + reinterpret_cast (&pt), + sizeof(T), + &sz[0], 2 * sizeof(T), + nullptr, &bin_len, nullptr + ); + return pt; + } + + template decltype(std::declval().blob, ro::CompileSizedString < (sizeof(T) * 8 + 5) / 6>()) to_base64_string(const T& p_blob) { + ro::CompileSizedString < (sizeof(T) * 8 + 5) / 6> sz; + bits2base64( + &(p_blob.blob[0]), 0, sizeof(p_blob.blob) * 8, + std::span(sz) + ); + return sz; + } + +} //End ro namespace + +namespace ristretto255 { + using ro::to_base64_string, ro::is_serializable; + // a class representing ristretto255 elliptic points + class point; + + // a class representing ristretto255 scalars + class scalar; + + templateclass hsh; + templateclass hash; + template decltype(std::declval().blob, bool()) ConstantTimeEqualityTest(const T& x, const T& y) { + if (T::constant_time_required) { + return 0 == sodium_memcmp(&x.blob[0], &y.blob[0], sizeof(x.blob)); + } + else return x == y; + } + + template decltype(std::declval().blob, wxString()) to_wxString(const T& p_blob) { + std::array sz; + bits2base64(&(p_blob.blob[0]), 0, sizeof(p_blob.blob) * 8, std::span(sz)); + return wxString::FromUTF8Unchecked(&sz[0]); + } + + // hsh constructs a partial hash, not yet finalized + // normally never explicitly used in code, but rather a nameless rtype value on the stack. + // To which more stuff can be added with the << operator + // usage: + // hash( hsh(a, b) << c << f << g ); + // assert( hash(a, b, c, d, e) == hash(hsh(a, b, c) << d << e) ); + // hash finalizes hsh. + // Of course you would never use it that way, because you would only + // use it explicitly if you wanted to keep it around + // attempting hash more things into an hsh object that has already + // been finalized will throw an exception. + // Old type C byte array arguments after the first are vulnerable to + // array decay to pointer, so wrap them in std::span(OldTypeCarray) + // or hash them using "<<" rather than putting them in the initializer. + // It always a wise precaution to not use old type C arays, or wrap them + // in a span. + // Old type zero terminated strings work. The trailing zero is included + // in the hash + template class hsh { + public: + static_assert(hashsize > 63 && hashsize % 64 == 0 && crypto_generichash_BYTES_MIN * 8 <= hashsize && hashsize <= crypto_generichash_BYTES_MAX * 8, "Bad hash size."); + static constexpr unsigned int type_indentifier = 2 + (hashsize + 0x90 * 8) / 64; + static_assert(crypto_generichash_BYTES_MAX < 0x90, "Change in max hash has broken our type ids"); + crypto_generichash_blake2b_state st; + hsh() { + int i{ crypto_generichash_blake2b_init( + &st, + nullptr,0, + hashsize / 8) + }; + assert(i == 0); + } + template::value, int >::type dummy_arg = 0 + >explicit hsh(const T first, Args... args) { + int i{ crypto_generichash_blake2b_init( + &st, + nullptr,0, + hashsize / 8) + }; + assert(i == 0); + (*this) << first; + if constexpr (sizeof...(args) > 0) { + (*this).hashinto(args...); + } + } + + template < typename T> + decltype(ro::serialize(std::declval()), hsh())& + operator <<(const T& j) { + return *this << ro::serialize(j); + } + + + template::value, int >::type dummy_arg = 0 + > void hashinto(const T first, Args... args) { + *this << first; + if constexpr (sizeof...(args) > 0) { + (*this).hashinto(args...); + } + } + + hsh& operator <<(const std::span& j) { + int i = crypto_generichash_blake2b_update( + &st, + &j[0], + j.size() + ); + if (i) throw HashReuseException(); + return *this; + } + + hsh& operator <<(char* sz) { + int i = crypto_generichash_blake2b_update( + &st, + static_cast(sz), + strlen(sz) + 1 + ); + if (i) throw HashReuseException(); + return *this; + } + }; + + // This constructs a finalized hash. + // If it has one argument, and that argument is hsh (unfinalized hash) object, + // it finalizes the hash. (see hsh) + // Usage + // hash(a, b, c ...); + // hash and hsh serialize all their arguments, converting them into machine + // and compiler independent form. If they don't know how to serialize an + // argument type, you get a compile time error. To serialize a new type, + // create a new overload for the function "serialize" + // to hash a wxString, use its ToUTF8 member + // wxString wxsHello(L"Hello world"); + // hash hashHello(wxsHello.ToUTF8()); + // C array arguments after the first are vulnerable to array decay to pointer, so use hsh and "<<" + // for them. or wrap them in std::span(OldTypeCarray) + // It always a wise precaution to not use old type C arays, or wrap them + // in a span. + // Old type zero terminated utf8 strings work. The trailing zero is included. + // The program should by running in the UTF8 locale, attempts to set that + // locale on startup. and tests for success in the unit test. + template class hash { + static_assert(hashsize > 63 && hashsize % 64 == 0 && crypto_generichash_BYTES_MIN * 8 <= hashsize && hashsize <= crypto_generichash_BYTES_MAX * 8, "Bad hash size."); + friend point; + friend scalar; + friend hsh; + public: + static constexpr unsigned int type_indentifier = 2 + hashsize / 64; + hash() = default; + std::array blob; + ~hash() = default; + hash(hash&&) = default; // Move constructor + hash(const hash&) = default; // Copy constructor + hash& operator=(hash&&) = default; // Move assignment. + hash& operator=(const hash&) = default; // Copy assignment. + explicit hash(hsh& in) { + int i = crypto_generichash_blake2b_final( + &in.st, + &blob[0], hashsize / 8); + assert(i == 0); + if (i) throw HashReuseException(); + } + static_assert(!ro::is_serializable >::value, "Don't want to partially hash partial hashes"); + template::value, int >::type dummy_arg = 0 + >explicit hash(const T& first, Args... args) { + hsh in; + in << first; + if constexpr (sizeof...(args) > 0) { + in.hashinto(args...); + } + int i = crypto_generichash_blake2b_final( + &in.st, + &blob[0], hashsize / 8); + assert(i == 0); + } + hash& operator=(hsh&& in) { + int i = crypto_generichash_blake2b_final( + &in.st, + &blob[0], hashsize / 8); + if (i) throw HashReuseException(); + } + bool operator==(const hash& pt) const& { + return blob == pt.blob; //Do not need constant time equality test on hashes + } + bool operator!=(const hash& pt) const& { + return blob != pt.blob; //Do not need constant time equality test on hashes + } + }; + + // a class representing ristretto255 elliptic points, + // which are conveniently of prime order. + class point + { + // We will be reading points from the database, as blobs, reading them from the network as blobs, and reading them from human entered text as slash6 encoded blobs. Therefore, invalid point initialization data is all too possible. + public: + static constexpr unsigned int type_indentifier = 1; + std::array blob; + static_assert(std::is_trivially_copyable>(), "does not support memcpy init"); + static_assert(sizeof(blob) == 32, "watch for size and alignment bugs. Everyone should standarize on 256 bit public keys except for special needs"); + static point ptZero; + static point ptBase; + explicit point() = default; + // After loading a point as a blog from the network, from the database, or from user data typed as text, have to check for validity. + bool valid(void) const { return 0 != crypto_core_ristretto255_is_valid_point(&blob[0]); } + explicit point(std::array&& in) : blob{ std::forward>(in) } { }; + static_assert(crypto_core_ristretto255_BYTES == 32, "256 bit points expected"); + ~point() = default; + point(point&&) = default; // Move constructor + point(const point&) = default; // Copy constructor + point& operator=(point&&) = default; // Move assignment. + point& operator=(const point&) = default; // Copy assignment. + bool operator==(const point& pt) const& { + return ConstantTimeEqualityTest(*this, pt); + } + bool operator!=(const point& pt) const &{ + return !ConstantTimeEqualityTest(*this, pt); + } + + point operator+(const point &pt) const& { + point me; + auto i{ crypto_core_ristretto255_add(&me.blob[0], &blob[0], &pt.blob[0]) }; + assert(i == 0); + if (i != 0)throw NonRandomScalarException(); + return me; + } + + point operator-(const point& pt) const& { + point me; + auto i{ crypto_core_ristretto255_sub(&me.blob[0], &blob[0], &pt.blob[0]) }; + assert(i == 0); + if (i != 0)throw NonRandomScalarException(); + return me; + static_assert(sizeof(blob) == 32, "alignment?"); + } + + point operator*(const scalar&) const &noexcept; + point operator*(int) const& noexcept; + + explicit point(const hash<512>& x) noexcept { + static_assert(crypto_core_ristretto255_HASHBYTES * 8 == 512, "we need 512 bit randoms to ensure our points and scalars are uniformly distributed"); + // There should be scalar from hash, not point from hash + int i{ + crypto_core_ristretto255_from_hash(&blob[0], &(x.blob)[0]) }; + assert(i == 0); + } + + static point random(void) { + point me; + crypto_core_ristretto255_random(&(me.blob[0])); + return me; + } + + bool operator !() const& { + return 0 != sodium_is_zero(&blob[0], sizeof(blob)); + } + + static bool constant_time_required; + + }; + + + // a class representing ristretto255 scalars + // very large integers modulo the order of the + // ristretto255 elliptic curve, which is + // conveniently of prime order. + class scalar + { + friend point; + public: + static constexpr unsigned int type_indentifier = 2; + std::array blob; + static_assert(sizeof(blob) == 32, "watch for size and alignment bugs. Everyone should standarize on 256 bit secret keys except for special needs"); + explicit scalar() = default; + ~scalar() = default; + explicit constexpr scalar(std::array&& in) : blob{ in } {}; + explicit constexpr scalar(std::array* in) :blob(*in) {}; + + explicit scalar(int); + scalar(scalar&&) = default; // Move constructor + scalar(const scalar&) = default; // Copy constructor + scalar& operator=(scalar&&) = default; // Move assignment. + scalar& operator=(const scalar&) = default; // Copy assignment. + bool operator==(const scalar& sc) const& { + return ConstantTimeEqualityTest(*this, sc); + } + bool operator!=(const scalar& sc) const& { + return !ConstantTimeEqualityTest(*this, sc); + } + scalar operator+(const scalar sclr) const& { + scalar me; + crypto_core_ristretto255_scalar_add(&me.blob[0], &blob[0], &sclr.blob[0]); + return me; + } + static_assert(sizeof(scalar::blob) == 32, "compiled"); + + scalar multiplicative_inverse() const &{ + scalar me; + auto i = crypto_core_ristretto255_scalar_invert(&me.blob[0], &blob[0]); + assert(i == 0); + if (i != 0)throw NonRandomScalarException(); + return me; + } + + scalar operator-(const scalar& sclr) const& { + scalar me; + crypto_core_ristretto255_scalar_sub(&me.blob[0], &blob[0], &sclr.blob[0]); + return me; + } + + scalar operator*(const scalar& sclr) const& { + scalar me; + crypto_core_ristretto255_scalar_mul(&me.blob[0], &blob[0], &sclr.blob[0]); + return me; + } + + scalar operator/(const scalar sclr) const& { + return operator*(sclr.multiplicative_inverse()); + } + + scalar operator*(int i) const& { + return operator * (scalar(i)); + } + + point operator*(const point& pt) const& { + point me; + auto i{ crypto_scalarmult_ristretto255(&me.blob[0], &blob[0], &pt.blob[0]) }; + assert(i == 0); + if (i != 0)throw NonRandomScalarException(); + return me; + } + + point timesBase() const& { + point me; + auto i{ crypto_scalarmult_ristretto255_base(&me.blob[0], &blob[0]) }; + assert(i == 0); + if (i != 0)throw NonRandomScalarException(); + return me; + } + + explicit scalar(const hash<512>& x) { + static_assert(crypto_core_ristretto255_HASHBYTES == 64, "inconsistent use of magic numbers"); + crypto_core_ristretto255_scalar_reduce(&blob[0], &(x.blob)[0]); + } + + static scalar random(void) { + scalar me; + crypto_core_ristretto255_scalar_random(&me.blob[0]); + return me; + } + + bool valid(void) const& { + return _sodium_sc25519_is_canonical(&blob[0]) != 0 || + sodium_is_zero(&blob[0], crypto_core_ed25519_SCALARBYTES); + } + + bool operator !() const& { + return 0 != sodium_is_zero(&blob[0], sizeof(blob)); + } + +/* explicit operator wxString() const&; + explicit operator std::string() const&;*/ + static bool constant_time_required; + private: + void reverse(std::array < uint8_t, crypto_core_ristretto255_SCALARBYTES>const& ac) { + auto p = blob.rbegin(); + for (auto x : ac) { + *p++ = x; + } + } + }; + + static_assert(ro::is_blob_field_type::value); + static_assert(ro::is_blob_field_type::value); + static_assert(ro::is_blob_field_type::value); + static_assert(ro::is_blob_field_type >::value); + static_assert(false == ro::is_blob_field_type::value); + static_assert(ro::is_serializable::value); + static_assert(ro::is_serializable&>::value); + static_assert(ro::is_blob_field_type::value == false); + static_assert(ro::is_serializable::value); + static_assert(ro::is_serializable::value); + static_assert(ro::is_serializable::value == false); //false because uint8_t * has no inband terminator + static_assert(false == ro::is_serializable::value && !ro::has_constructor, wxString>::value, "wxStrings are apt to convert anything to anything, with surprising and unexpected results"); + static_assert(ro::is_serializable().ToUTF8())>::value == true); + static_assert(ro::is_constructible_from_all_of, std::array>); + static_assert(ro::is_constructible_from_all_of, char*, short, unsigned short, hash<512>, point, scalar>, "want to be able to hash anything serializable"); + static_assert(false == ro::is_constructible_from_any_of, hash<256>>); + static_assert(false == ro::is_constructible_from_any_of , byte*>, "do not want indiscriminate casts"); + static_assert(false == ro::is_constructible_from_any_of , byte*>, "do not want indiscriminate casts "); + static_assert(false == ro::has_constructor, float>::value); + static_assert(ro::is_serializable::value == false);//Need to convert floats to + // their machine independent representation, possibly through idexp, frexp + // and DBL_MANT_DIG + static_assert(sizeof(decltype(ro::serialize(std::declval())[0])) == 1); + static_assert (std::is_standard_layout(), "scalar for some reason is not standard layout"); + static_assert (std::is_trivial(), "scalar for some reason is not trivial"); + static_assert(sizeof(point) == 32, "funny alignment"); + static_assert(sizeof(scalar) == 32, "funny alignment"); + + class signed_text { + public: + std::span txt; + scalar c; + scalar s; + point K; + signed_text( + const scalar&, // Signer's secret key + const point&, // Signer's public key + std::span // Text to be signed. + ); + bool verify(); + }; + + class CMasterSecret :public scalar { + public: + CMasterSecret() = default; + CMasterSecret(const scalar& pt) :scalar(pt) {} + CMasterSecret(const scalar&& pt) :scalar(pt) {} + CMasterSecret(CMasterSecret&&) = default; // Move constructor + CMasterSecret(const CMasterSecret&) = default; // Copy constructor + CMasterSecret& operator=(CMasterSecret&&) = default; // Move assignment. + CMasterSecret& operator=(const CMasterSecret&) = default; // Copy assignment. + template auto operator()(T psz) { + scalar& t(*this); + return scalar(hash<512>(t, psz)); + } + + }; + + +} //End ristretto255 namespace + +// Ristretto255 scalars are defined to be little endian on all hardware +// regardless of the endianess of the underlying hardware. +// though it is entirely possible that sometime in the future, this +// definition will be changed should big endian hardware ever be +// sufficiently popular for anyone to care. +// So, because scalars are in fact integers, displaying them as +// biendian on all hardware when displayed in hex +// or base64. Everything else gets displayed in memory order. +template<> ristretto255::scalar ro::hex2bin (const ro::CompileSizedString< (2 * sizeof(ristretto255::scalar))>&); +template<> ro::CompileSizedString< (2 * sizeof(ristretto255::scalar)) > ro::bin2hex(const ristretto255::scalar&); +template<> ro::CompileSizedString< (8 * sizeof(ristretto255::scalar) + 5) / 6> ro::to_base64_string (const ristretto255::scalar&); + diff --git a/rotime.cpp b/rotime.cpp new file mode 100644 index 0000000..4c190a1 --- /dev/null +++ b/rotime.cpp @@ -0,0 +1,92 @@ +// Since I need my own time and duration types, +// that have the same binary representation on all +// operating systems and environmens + +#include "stdafx.h" + + + +/*I have spent far, far too much time trying to get overly clever layers that supposedly provide OS independence to work. +*/ + + + +#ifdef _MSC_VER +// compiling under Microsoft Visual Studio. +#include +#endif + + +namespace ro { + // need to define my own time types: + // typedef time_t time_t; // is signed time in seconds past the epoch modulo 2 ^ 64 + // because in some environments, time_t is seconds past the epoch modulo 2 ^ 32 + // and if time_t is not 64 bits, ro:time_t is a typedef of bint_fast64_t + + ro::time_t time() { return ::time(nullptr); } + // equivalent to regular old time(nullptr) except in environments where time_t + // except in environments where time_t is seconds past the epoch modulo 2 ^ 32 + + ro::sec_t system_time_now() { + ro::sec_t uu; + return uu + (ro::sec)(::time(nullptr)); + + } + + // duration past startup time modulo 2^32 Rolls over every forty eight days, making durations longer than twenty four days of unclear sign + msec msec_since_epoch(void){ + return std::chrono::duration_cast (std::chrono::steady_clock::now().time_since_epoch()); + } + + // date and time as utc time in human readable form + std::string iso_utc_time(ro::time_t t) { + std::array retval{ "0000-00-00t00:00:00" }; + std::tm tm_t; + if (0 == _gmtime64_s(&tm_t, &t)) { + strftime(retval.data(), 20, "%Y-%m-%dt%H:%M:%S", &tm_t); + } + return retval.data(); + }// which is a string expressing the utc time in human readable form + + // local date and time in human readable form + std::string iso_local_time(ro::time_t t) { + std::array retval{ "0000-00-00t00:00:00-00:00" }; + std::tm tm_t; + if (0 == _localtime64_s(&tm_t, &t)) { + strftime(retval.data(), 20, "%Y-%m-%dt%H:%M:%S", &tm_t); + long timezone; + if (0 == _get_timezone(&timezone)) { + if (timezone <= 0) { + retval.data()[19] = '+'; + timezone = -timezone; + } + sprintf_s(retval.data() + 20, 6, "%2.2d:%2.2d", (timezone / 3600) % 24, ((timezone+30) / 60) % 60); + } + } + return retval.data(); + } + + // string expressing the current utc time + std::string iso_utc_time(ro::sec_t t) { + return ro::iso_utc_time((t.time_since_epoch()).count()); + } + // string expressing the current local time + std::string iso_local_time(ro::sec_t t) { + return ro::iso_local_time((t.time_since_epoch()).count()); + } + + // String expressing steady system time interval + // Typical usage steady_clock(msec_time_now() - start_time); + // Always outputs eleven characters plus a null terminator. + std::string duration_as_string(msec t) { + std::array retval{ " .000" }; + uint_fast32_t t_as_int{ t.count() }; + uint_fast32_t milliseconds{ t_as_int % 1000 }; + uint_fast32_t seconds{ t_as_int / 1000 }; + snprintf(&retval[0], retval.size(), "%7u.%03u", seconds, milliseconds); + return retval.data(); + } + + +} + diff --git a/rotime.h b/rotime.h new file mode 100644 index 0000000..b337bc9 --- /dev/null +++ b/rotime.h @@ -0,0 +1,44 @@ +#pragma once +namespace ro { + // need to define my own time types: + typedef time_t time_t; // is always time in seconds past the epoch modulo 2 ^ 64 + // equivalent to regular old time_t + // except in environments where time_t is seconds past the epoch modulo 2 ^ 32 + ro::time_t time(); + // equivalent to regular old time(nullptr); + // except in environments where time_t is seconds past the epoch modulo 2 ^ 32 + typedef std::chrono::duration > sec; + typedef std::chrono::time_point sec_t; + ro::sec_t system_time_now(); + // Also equivalent to plain old time_t, except with modern C++14 type distinction + // between time points and time durations + std::string iso_utc_time(ro::time_t t); + std::string iso_utc_time(ro::sec_t t); + // which is a string expressing the utc time + std::string iso_local_time(ro::time_t t); + std::string iso_local_time(ro::sec_t t); + // which is a string expressing the local time + + typedef std::chrono::duration > msec; + + // duration past startup time modulo 2^32 Rolls over every forty eight days, making durations longer than twenty four days of unclear sign + msec msec_since_epoch(void); + // Obliterates chrono's elegant but pain in the ass distinction between time points and durations. + // Which type distinction in practice gets in the way off all sorts of things that just need to be done. + // In truth, it is a distinction that only matters, and is indeed only meaningful, with respect to shared global consensus time, universal coordinated time + // It is not meaningful for steady time or high precision time, which means that all sorts of stuff gets assigned to the wrong type, a duration when you need it to be a time point, or a time point when you need it to be a duration. + // A time point is just a duration relative to some globally agreed consensus time point, and if you don't have such a consensus, forget it. And when you are measuring millisecond times, you do not have such a consensus + // millisecond clocks are not expected to be synchronized between systems, nor to measure milliseconds exactly nor to be exactly the same time unit between systems, but they are expected to advance similarly on both systems for the life of any connection, making the distinction between time points and durations problematic. + // We expect to compare the change in time on one system, with the change in time on another system, in which case the time point class gets in the way. + + +/* + std::array msec::operator() const { + std::array x; + return x; + + }*/ + + + +} diff --git a/secrets.cpp b/secrets.cpp new file mode 100644 index 0000000..f1ebbd5 --- /dev/null +++ b/secrets.cpp @@ -0,0 +1,57 @@ +#include "stdafx.h" +#include "secrets.h" +//Derive a Short text Secret from a 256 bit random value and a sixty four bit integer +using ristretto255::scalar, ristretto255::hash, ristretto255::hsh; +std::array DeriveTextSecret(const scalar & blob, uint_fast64_t i){ + static constexpr std::array salt{0}; + std::array txt; + unsigned int n{ 3 }; + unsigned int m{ 3 }; + hash rand(blob,i); + uint8_t* p{ &rand.blob[0] }; + char* q = &txt[0]; + while (n || m) { + if (*p++ < ((n * 0x100) / (n + m))) { + n--; + bits2base64(p, 0, 18, std::span(q, 4)); + q+=3; + } + else { + m--; + bits2base64(p, 0, 24, std::span(q, 5)); + q+=4; + } + *q++ = ' '; + p += 3; + } + assert(q == sizeof(txt) + &txt[0]); + assert(p == &rand.blob[24]); + txt[sizeof(txt)-1]='\0'; + return txt; +} + +//Derive a strong scalar secret from a string with password strengthening. +//Net effect is convert one scalar into another by a process that is lengthy and costly. +ristretto255::scalar DeriveStrongSecret(const char* const passwd){ + static std::array salt{ 0 }; + std::array)>randb; + int i{ + crypto_pwhash( + &randb[0], + sizeof(randb), + passwd, + strlen(passwd)+1, + &salt[0], + 2, 0x10000000, + crypto_pwhash_ALG_ARGON2ID13 + ) + }; + static_assert(crypto_pwhash_OPSLIMIT_MODERATE == 0x00000003 && crypto_pwhash_MEMLIMIT_MODERATE == 0x10000000, "Argon changed, likely breaking all passphrases"); + assert(i == 0); + return scalar(hash<512>(randb)); +} + +//Derive scalar secret from another quickly. +ristretto255::scalar DeriveSecret(const scalar &sc, uint_fast64_t i) { + return ristretto255::scalar(ristretto255::hash<512>(sc, i)); +} \ No newline at end of file diff --git a/secrets.h b/secrets.h new file mode 100644 index 0000000..3df04de --- /dev/null +++ b/secrets.h @@ -0,0 +1,14 @@ +#pragma once +//Derive a Short text Secret from a 256 bit random value and a sixty four bit integer +std::array DeriveTextSecret(const ristretto255::scalar& blob, uint_fast64_t i); + +//Derive a strong scalar secret from a string with password strengthening. +//Net effect is convert one scalar into another by a process that is lengthy and costly. +ristretto255::scalar DeriveStrongSecret(const char* const passwd); + +//Derive scalar secret from another quickly. +ristretto255::scalar DeriveSecret(const ristretto255::scalar &sc, uint_fast64_t i); +template , int> = 0> +constexpr int rounded_log2(const T val) noexcept { + return std::numeric_limits::digits - std::countl_zero(val); +} \ No newline at end of file diff --git a/slash6.cpp b/slash6.cpp new file mode 100644 index 0000000..7393fe1 --- /dev/null +++ b/slash6.cpp @@ -0,0 +1,280 @@ +//#include +//#include +#include +#include +#include +//#include // for initializer_list +//#include +#include // for shared_ptr, unique_ptr +#include +#include "ILog.h" +#include "localization.h" +#include "slash6.h" +// This base64 uses the following characters as the numerals 0 to 63: +// Table to convert six bit to ascii +static constexpr uint8_t index2base64[]{ "0123456789ABCDEFGHJKLMNPQRSTUVWXYZabcdefghijkmnpqrstuvwxyz!$*+-_" }; + +// Unlike regular baseINV, and incompatible with it. Intended to encode stuff small enough that it might be human typed and human transmitted, therefore maps 'o' and 'O' to '0', 'I' and 'l' to '1' +// uses six url safe additional characters !$*+-_ to bring it up to six bits +static_assert(index2base64[63] == '_', "surprise numeral at 63"); + +// Being intended for small bits of data, assumes no whitespace within an entity +// Encode and decode are called with a bit buffer, consisting of an unsigned byte pointer, a starting bit position relative to the pointer, and a bit count. We assume that there is room enough in the object pointed to to accommodate the bytes referenced by Bit Position+BitCount. We don't change bits outside the range. +// The bit buffer does not need to be aligned, nor does it need to be a multiple of six bits, eight bits, or twenty four bits. +// If the input bit buffer to be encoded to base sixty four is not a multiple of six bits, the last base sixty four numeral output will represent the bit buffer padded with trailing zeroes. +// If there is no room in the output span for all the base sixty four digits, the encode routine will return a number of bits less than the size of the input bit buffer, less than the bitcount it was given. +// Table to convert ascii to six bit. +static constexpr std::array ascii2six_ar{ + 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, //control characters + 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, //control characters + 0xff, 0x3a, 0xff, 0xff, 0x3b, 0xff, 0xff, 0xff, 0xff, 0xff, 0x3c, 0x3d, 0xff, 0x3e, 0xff, 0xff, // ! $ * + - permitted The characters "#%&'(),. are stops + 0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07, 0x08, 0x09, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, // 0-9 permitted The characters :;<=>? are stops + 0xff, 0x0a, 0x0b, 0x0c, 0x0d, 0x0e, 0x0f, 0x10, 0x11, 0x01, 0x12, 0x13, 0x14, 0x15, 0x16, 0x00, + 0x17, 0x18, 0x19, 0x1a, 0x1b, 0x1c, 0x1d, 0x1e, 0x1f, 0x20, 0x21, 0xff, 0xff, 0xff, 0xff, 0x3f, // A-Z and _ permitted, @ and [\]^ are stops + 0xff, 0x22, 0x23, 0x24, 0x25, 0x26, 0x27, 0x28, 0x29, 0x2a, 0x2b, 0x2c, 0x01, 0x2d, 0x2e, 0x00, + 0x2f, 0x30, 0x31, 0x32, 0x33, 0x34, 0x35, 0x36, 0x37, 0x38, 0x39, 0xff, 0xff, 0xff, 0xff, 0xff, //a-z the characters {|} ~` are stops, as is the mysterious control character 0x7F (del) + 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, + 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, + 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, + 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, + 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, + 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, + 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, + 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff // utf8 multibyte characters, all stops. +}; + +// Table to convert ascii to six bit as a good old fashioned non owning naked pointer to const, whose storage is owned by a const static which exists until the program terminates. +static const uint8_t *ascii2six{ &ascii2six_ar[0] }; +// Compile time execution is C++ is a pain, because expressions are apt to unpredictably lose their constexpr character for reasons that are far from clear. +// You really have to write compile time code in templates as a language, which is the totally obscure and hard to use language apt to generate remarkably voluminous error messages will little obvious connection to the actual problem, and surprising result that are very difficult to predict in advance or understand at all. +//In general, the better solution is to have a routine that is called once and only once at the beginning of the program, which initializes a bunch of static const values, if that solution is adequate, or to have a preproces routine written in python which generates the required C files and header files. + +// After this experiment in compile time code, I swear off it. + +// Decode does not have an input span of encoded characters, but a char *, because it assumes the string is always terminated by an invalid character, such as the trailing null at the end of string or a space, or any character that is not one of our base sixty four numerals +// If the there are not enough input base sixty four numerals, it returns a size less than requested. +// if the requested bit buffer is not a multiple of six bits, and the last base 64 numeral had trailing ones that would not fit in the buffer, rather than the expected trailing zeroes, then it returns a size larger than the buffer, as if it changed stuff outside the buffer but does not actually change bits outside the buffer. This is likely an error, because obviously we want to decode something from base sixty four that was originally decoded using the same sized buffer. + +static const uint8_t INV{ UCHAR_MAX }; + +const uint8_t highBitMask[]{ 0x00, 0x80, 0xC0, 0xE0, 0xF0, 0xF8, 0xFC, 0xFE, 0xFF }; +static_assert(CHAR_BIT+1 == sizeof(highBitMask), "expecting eight bits per character"); +const uint8_t lowBitMask[]{ 0xFF, 0x7F, 0x3F, 0x1F, 0x0F, 0x07, 0x03, 0x01, 0x00 }; +static_assert(CHAR_BIT + 1 == sizeof(lowBitMask), "expecting eight bits per character"); + + +// Converts bit a bit buffer into base 64 numerals +// Start does not need to be byte aligned, nor does the length need to be a multiple of eight, six, or twenty four. +// If insufficient room is provided for the base 64 value, throws exception. +void bits2base64(const uint8_t * bitBuffer, unsigned int start, unsigned int length, std::spanbase64Numerals) { + assert(length <= 6 * base64Numerals.size() - 6); + // If you hit this assert you probably passed in a negative number, overflowing the string buffer + // which would cause you to write all over memory. + // Expects(length <= 6 * base64Numerals.size()-6); + if (length > 6 * base64Numerals.size() - 6)throw FatalException(sz_text_buffer_overflow); + auto e{ &base64Numerals[0] }; + if (length) { + bitBuffer += start / CHAR_BIT; + start -= (start / CHAR_BIT) * CHAR_BIT; + assert(start == start % 8); + unsigned int zeropadding{ ((length + 5) / 6) * 6 - length }; + assert((length + zeropadding) % 6 == 0); + assert(length + zeropadding <= 6 * (unsigned int)( base64Numerals.size())); + unsigned int bitAccumulator = (*bitBuffer++) & lowBitMask[(start)]; + unsigned int bitsInAccumulator{ 8 - start }; + assert(bitsInAccumulator < 9); + unsigned int count{ length - bitsInAccumulator }; //count is bits lifted out of the buffer. + // We need to keep track of count, because the last byte lifted out of the buffer may well have to be an incomplete byte, so when count gets down to below 8, we have to special case loading the bitAccumulator. + char * Outputsz{ &(base64Numerals[0]) }; + auto endBaseNumeralBuffer{ Outputsz + length / 6 }; + for (; e < endBaseNumeralBuffer; e++) { + assert(count + bitsInAccumulator + 6 * (unsigned int)(e - Outputsz) == length); + if (bitsInAccumulator < 6) + { + if (count > 7) { + bitAccumulator = (bitAccumulator << 8) | (*bitBuffer++); + bitsInAccumulator += 8; + count -= 8; + assert(count + bitsInAccumulator + 6 * (unsigned int)(e - Outputsz) == length); + } + else { + assert(count + bitsInAccumulator >= 6); // Should be enough bits in buffer for all numerals produced by this for loop. + bitAccumulator = (((bitAccumulator << 8) | (*bitBuffer++)) >> (8 - count)); + bitsInAccumulator += count; + count = 0; + assert(count + bitsInAccumulator + 6 * (unsigned int)(e - &base64Numerals[0]) == length); + } + } + assert(bitsInAccumulator > 5); + *e = index2base64[(bitAccumulator >> (bitsInAccumulator - 6)) & 0x3F]; + bitsInAccumulator -= 6; + } + // When we drop out of the for loop, we may have more than 0 bits left but less than six, in which case we then have to special case the last numeral by filling the bit accumulator with our zeropadding. + if (count) { + bitAccumulator = (((bitAccumulator << 8) | (*bitBuffer++)) >> (8 - count)); + bitsInAccumulator += count; + count = 0; + assert(count + bitsInAccumulator + 6 * (unsigned int)(e - &base64Numerals[0]) == length); + } + if (bitsInAccumulator) { + // Going to issue one last numeral + assert(zeropadding + bitsInAccumulator == 6); + if (bitsInAccumulator < 6) { + bitAccumulator = bitAccumulator << zeropadding; // 0 pad accumulator + bitsInAccumulator = 6; //This breaks the invariant checked by the assert. + } + assert(bitsInAccumulator == 6); + *e++ = index2base64[bitAccumulator & 0x3F]; + } + assert(count == 0); + assert(6 * (unsigned int)(e - Outputsz) == length + zeropadding); + assert((length + 5) / 6 == e - &base64Numerals[0]); //Ensures that base 64 representation is the right size to hold the bits. + } + *e = '\0'; +} + +// This may produce the runtime error that there are too few numerals to fill the bit buffer. +// Stops at the first invalid numeral - such as a space. +// This is intended for quite short bit fields, not for transmitting megabytes of data over lines that are not eight bit safe. +// Returns the actual size of the fill. +// If the bit buffer is not a multiple of six, the last numerals excess bits need to be zero +// If they are not zero will truncate the excess bits and and throw BadDataException. +unsigned int base64_to_bits(uint8_t * bitBuffer, unsigned int start, unsigned int length, const char * base64Numerals) { + bitBuffer += start / CHAR_BIT; + start -= (start / CHAR_BIT) * CHAR_BIT; + assert(start == start % 8); + unsigned int zeropadding{ ((length + 5) / 6) * 6 - length }; + assert((length + zeropadding) % 6 == 0); + unsigned int bitAccumulator = (*bitBuffer)>>(8-start); + unsigned int bitsInAccumulator{ start }; + unsigned int count{ length }; + unsigned int numeral; + uint8_t overflowBits{ '\0' }; + uint8_t * p{ bitBuffer }; + for (const char * e = base64Numerals; ((numeral = ascii2six[static_cast(*e)]), (numeral < INV)); e++) { + assert((e - base64Numerals) * 6 + count == length); + assert((p-bitBuffer)*8 + bitsInAccumulator == start + (e-base64Numerals)*6 ); + bitAccumulator = (bitAccumulator << 6) | numeral; + bitsInAccumulator += 6; + if (count < 6) { + if (count > 0) { + const uint8_t lowBitMaskIn[]{ 0x00, 0x01, 0x03, 0x07, 0x0F, 0x1F, 0x3F }; + // Bit accumulator is here numeral aligned, rather than byte aligned + overflowBits = bitAccumulator & lowBitMaskIn[6 - count]; + bitAccumulator = (bitAccumulator >> (6 - count)); + bitsInAccumulator -= 6 - count; + // Now it is bit field aligned + count = 0; + } + } + else count -= 6; + if (bitsInAccumulator > 7) { + *p++ = (bitAccumulator >> (bitsInAccumulator % 8)); + bitsInAccumulator -= 8; + } + if (count == 0) { + assert(bitsInAccumulator < 9); + if (bitsInAccumulator) { + *p = (bitAccumulator << (8 - bitsInAccumulator)) | (*p & lowBitMask[bitsInAccumulator]); + } + break; + } + } + if (overflowBits) throw BadDataException(); + return overflowBits ? length + 1 : length - count; +} + +/* Expects pointer to byte buffer and pointer to string. + Expects a string of exactly the correct number of numerals, + terminated by a non base64 character, such as null. + Throws exception if that is not what it gets. + Fills the byte buffer exactly. + Returns a uint8_t containing the excess bits of the last numeral in its low order part.*/ +uint8_t base64_to_bytes(uint8_t* byteBuffer, uint_fast32_t byteCount, const char* base64Numerals) { + auto numeralsCount{ byteCount * 8 / 6 }; + auto bitsCount{ numeralsCount * 6 }; + auto length{ base64_to_bits(byteBuffer, 0, bitsCount, base64Numerals) }; + if (length < bitsCount) throw BadDataException(); + base64Numerals += numeralsCount; + auto leftoverBitsField{ byteCount * 8 - length }; + auto leftoverBits{ 0 }; + if (leftoverBitsField) { + // we cast to unsigned character, because otherwise it is likely to be sign + // extended resulting in indexing outside the range 0-0xFF + // with an indeterminate and unpredictable number of high + // order bits set. + uint8_t numeral{ ascii2six[static_cast(*base64Numerals++)]}; + if (numeral>63) throw BadDataException(); + auto missingBitsField{8*byteCount-bitsCount}; + assert(missingBitsField + leftoverBitsField == 6); + auto missingBitsMask{ (1 << missingBitsField) - 1 }; + auto missingBits{ static_cast(numeral >> leftoverBitsField) }; + byteBuffer[byteCount - 1] = ((byteBuffer[byteCount - 1] | missingBitsMask) ^ missingBitsMask) | missingBits; + auto leftoverBits{ static_cast( numeral ^ (missingBits << missingBitsField)) }; + } + if (ascii2six[static_cast(*base64Numerals)]<64) throw BadDataException(); + return leftoverBits; +} + + +// Converts bit a bit buffer into base 2048 BIPS-39 words +// Using the array ar_sz_bip_0039_wordlist defined in localization.cpp +// The largest word in the array is eight characters, but other languages likely have longer words. +// Start does not need to be byte aligned, nor does the length need to be a multiple of eight or eleven. +void bits2base2048(const uint8_t* bitBuffer, int start, int length, std::spanszBipsWords) { + char* Outputsz{ &(szBipsWords[0]) }; + if (szBipsWords.size() == 0)throw ; + char* end_of_Outputsz{ &(szBipsWords[szBipsWords.size() - 1]) }; + if (length>0) { + bitBuffer += start / CHAR_BIT; + start -= (start / CHAR_BIT) * CHAR_BIT; + assert(start == start % 8); + uint_fast32_t bitAccumulator = (*bitBuffer++) & lowBitMask[(start)]; + int bitsInAccumulator{ 8 - start }; + assert(bitsInAccumulator < 9); + int count{ length - bitsInAccumulator }; //count is bits remaining in buffer, may go negative + // We need to keep track of count, because the last byte lifted out of the buffer may well have to be an incomplete byte, so when count gets down to below 8, we have to special case loading the bitAccumulator. + while (count) { + while (bitsInAccumulator < 11 && count >0) + { + bitAccumulator = (bitAccumulator << 8) | (*bitBuffer++); + bitsInAccumulator += 8; + count -= 8; + } + if (count < 0) { + // get rid of bad bits + bitAccumulator >>= (-count); + bitsInAccumulator += count; + if (bitsInAccumulator < 11) { + bitAccumulator <<= (11 - bitsInAccumulator); + bitsInAccumulator = 11; + } + count = 0; + } + uint_fast16_t wordnumber{ (bitAccumulator >> (bitsInAccumulator - 11)) & 2047 }; + bitsInAccumulator -= 11; + const char* psz_Word{ ar_sz_bip_0039_wordlist[wordnumber] }; + while (*psz_Word) { + if (Outputsz == end_of_Outputsz)throw FatalException("not enough room for BIPS-0039 passphrase"); + *Outputsz++ = *psz_Word++; + } + if (bitsInAccumulator + count > 0) { + if (Outputsz == end_of_Outputsz)throw FatalException("not enough room for BIPS-0039 passphrase"); + *Outputsz++ = ' '; + } + } + } + *Outputsz++ = 0; +} +/* +// Not cryptographically strong, not DoS induced collision resistant. Produces +// the same mapping on all machines, for all time. +uint32_t bernstein_hash(const uint8_t* key, unsigned int len) { + constexpr uint_fast32_t INITIAL_VALUE = 5381; + constexpr uint_fast32_t M = 33; + uint_fast32_t hash = INITIAL_VALUE; + for (uint_fast32_t i = 0; i < len; ++i) + hash = M * hash + key[i]; + return hash; +} +*/ \ No newline at end of file diff --git a/slash6.h b/slash6.h new file mode 100644 index 0000000..0ac2686 --- /dev/null +++ b/slash6.h @@ -0,0 +1,37 @@ +#pragma once + +// Converts a bit buffer, arbitrarily positioned with respect to byte boundaries, into a base sixty four numeral. +// In release mode, returns without doing anything if the output string buffer is too small. +// In debug mode, halts execution with an assert. Should throw, need to put in and unit test. +void bits2base64(const uint8_t * bitBuffer, unsigned int start, unsigned int length, std::spanbase64Numerals); + +// Converts a base64 numeral into a bit buffer. The input string is terminated by the first non base64 character, normally a space or null, or terminated when the bit buffer is full. A correct length base64 numeral does not need a terminating space or null. +// If successful, returns a length equal to the bit buffer length. If the input string is too short, returns a length shorter than the bit buffer. If the bit buffer is not a multiple of six bits, and the last base64 numeral of the bit buffer has trailing binary ones, returns a length greater than the size of the bitbuffer, without writing the trailing binary bits into the bit buffer. +unsigned int base64_to_bits(uint8_t* bitBuffer, unsigned int start, unsigned int length, const char* base64Numerals); + +/* Expects pointer to byte buffer and pointer to string. + Expects a string of exactly the correct number of numerals, + terminated by a non base64 character, such as null. + Throws exception if that is not what it gets. + Fills the byte buffer exactly. + Returns a uint8_t containing the excess bits of the last numeral in its low order part.*/ +uint8_t base64_to_bytes(uint8_t* byteBuffer, uint_fast32_t byteCount, const char* base64Numerals); + +/* Expects reference to a range object over bytes and pointer to string. + Expects a string of exactly the correct number of numerals, + terminated by a non base64 character, such as null. + Throws exception if that is not what it gets. + Fills the byte buffer exactly. + Returns a uint8_t containing the excess bits of the last numeral in its low order part.*/ +template < typename T> +std::enable_if_t< + sizeof(std::size(std::declval())) >= sizeof(int) && + sizeof(std::declval()[0]) == 1, + uint8_t +>base64_to_bytes( T& byteRange, const char* base64Numerals) { + return base64_to_bytes(static_cast(&byteRange[0]), static_cast(std::size(byteRange)), base64Numerals); +} + + + + diff --git a/sqlite3.c b/sqlite3.c new file mode 100644 index 0000000..faad19f --- /dev/null +++ b/sqlite3.c @@ -0,0 +1,225083 @@ +/****************************************************************************** +** This file is an amalgamation of many separate C source files from SQLite +** version 3.30.1. By combining all the individual C code files into this +** single large file, the entire code can be compiled as a single translation +** unit. This allows many compilers to do optimizations that would not be +** possible if the files were compiled separately. Performance improvements +** of 5% or more are commonly seen when SQLite is compiled as a single +** translation unit. +** +** This file is all you need to compile SQLite. To use SQLite in other +** programs, you need this file and the "sqlite3.h" header file that defines +** the programming interface to the SQLite library. (If you do not have +** the "sqlite3.h" header file at hand, you will find a copy embedded within +** the text of this file. Search for "Begin file sqlite3.h" to find the start +** of the embedded sqlite3.h header file.) Additional code files may be needed +** if you want a wrapper to interface SQLite with your choice of programming +** language. The code for the "sqlite3" command-line shell is also in a +** separate file. This file contains only code for the core SQLite library. +*/ +#define SQLITE_CORE 1 +#define SQLITE_AMALGAMATION 1 +#ifndef SQLITE_PRIVATE +# define SQLITE_PRIVATE static +#endif + + +//My custom compile options +#define SQLITE_DQS 0 //Doublequote names, single quote strings. This setting disables the double - quoted string literal misfeature. +#define SQLITE_THREADSAFE 2 //Sets the default mode to SQLITE_CONFIG_MULTITHREAD. One thread, one database connection. Data structures such as compiled SQL are threadlocal. But sqlite3 is empowered to do its own multithreading. Many databases per database connection. Database connection and compiled sql statements are threadlocal. last_insert_rowid() is not subject to race conditions in this mode. +#define SQLITE_DEFAULT_MEMSTATUS 0 //Don't track memory usage. Disables the ability of the program using sqlite3 to monitor its memory usage. This setting causes the sqlite3_status() interfaces that track memory usage to be disabled. This helps the sqlite3_malloc() routines run much faster, and since SQLite uses sqlite3_malloc() internally, this helps to make the entire library faster. +#define SQLITE_DEFAULT_WAL_SYNCHRONOUS 1 // in WAL mode, recent changes to the database might be rolled back by a power loss, but the database will not be corrupted. Furthermore, transaction commit is much faster in WAL mode using synchronous=NORMAL than with the default synchronous=FULL. For these reasons, it is recommended that the synchronous setting be changed from FULL to NORMAL when switching to WAL mode. This compile-time option will accomplish that. +#define SQLITE_DEFAULT_FOREIGN_KEYS 0 //Dont handle foreign key constraints. Programmer has to do it himself. +#define SQLITE_LIKE_DOESNT_MATCH_BLOBS 1 //Blobs are not strings. Historically, SQLite has allowed BLOB operands to the LIKE and GLOB operators. But having a BLOB as an operand of LIKE or GLOB complicates and slows the LIKE optimization. When this option is set, it means that the LIKE and GLOB operators always return FALSE if either operand is a BLOB. That simplifies the implementation of the LIKE optimization and allows queries that use the LIKE optimization to run faster. +#define SQLITE_MAX_EXPR_DEPTH 0 //Setting the maximum expression parse-tree depth to zero disables all checking of the expression parse-tree depth, which simplifies the code resulting in faster execution, and helps the parse tree to use less memory. +#define SQLITE_OMIT_DECLTYPE 1 // By omitting the (seldom-needed) ability to return the declared type of columns from the result set of query, prepared statements can be made to consume less memory. +#define SQLITE_OMIT_DEPRECATED 1 +#define SQLITE_DQS 0 //Don't accept double quoted string literals. +#define SQLITE_OMIT_PROGRESS_CALLBACK 1 +#define SQLITE_OMIT_SHARED_CACHE 1 +#define SQLITE_OMIT_UTF16 1 +#define SQLITE_USE_ALLOCA 1 //Make use of alloca() for dynamically allocating temporary stack space for use within a single function, on systems that support alloca(). Without this option, temporary space is allocated from the heap +#define SQLITE_OMIT_LOAD_EXTENSION 1 +#define SQLITE_TEMP_STORE 1 //Temporary files are stashed on disk when their cache overflows. +#define SQLITE_OMIT_AUTOINIT 1 //.The SQLite library needs to be initialized using a call to sqlite3_initialize() before certain interfaces are used.This initialization normally happens automatically the first time it is needed.However, with the SQLITE_OMIT_AUTOINIT option, the automatic initialization is omitted.This helps many API calls to run a little faster(since they do not have to check to see if initialization has already occurred and then run initialization if it has not previously been invoked) but it also means that the application must call sqlite3_initialize() manually.If SQLite is compiled with - DSQLITE_OMIT_AUTOINIT and a routine like sqlite3_malloc() or sqlite3_vfs_find() or sqlite3_open() is invoked without first calling sqlite3_initialize(), the likely result will be a segfault +//end my custom compile options*/ +/************** Begin file ctime.c *******************************************/ +/* +** 2010 February 23 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** +** This file implements routines used to report what compile-time options +** SQLite was built with. +*/ + +#ifndef SQLITE_OMIT_COMPILEOPTION_DIAGS /* IMP: R-16824-07538 */ + +/* +** Include the configuration header output by 'configure' if we're using the +** autoconf-based build +*/ +#if defined(_HAVE_SQLITE_CONFIG_H) && !defined(SQLITECONFIG_H) +#include "config.h" +#define SQLITECONFIG_H 1 +#endif + +/* These macros are provided to "stringify" the value of the define +** for those options in which the value is meaningful. */ +#define CTIMEOPT_VAL_(opt) #opt +#define CTIMEOPT_VAL(opt) CTIMEOPT_VAL_(opt) + +/* Like CTIMEOPT_VAL, but especially for SQLITE_DEFAULT_LOOKASIDE. This +** option requires a separate macro because legal values contain a single +** comma. e.g. (-DSQLITE_DEFAULT_LOOKASIDE="100,100") */ +#define CTIMEOPT_VAL2_(opt1,opt2) #opt1 "," #opt2 +#define CTIMEOPT_VAL2(opt) CTIMEOPT_VAL2_(opt) + +/* +** An array of names of all compile-time options. This array should +** be sorted A-Z. +** +** This array looks large, but in a typical installation actually uses +** only a handful of compile-time options, so most times this array is usually +** rather short and uses little memory space. +*/ +static const char * const sqlite3azCompileOpt[] = { + +/* +** BEGIN CODE GENERATED BY tool/mkctime.tcl +*/ +#if SQLITE_32BIT_ROWID + "32BIT_ROWID", +#endif +#if SQLITE_4_BYTE_ALIGNED_MALLOC + "4_BYTE_ALIGNED_MALLOC", +#endif +#if SQLITE_64BIT_STATS + "64BIT_STATS", +#endif +#if SQLITE_ALLOW_COVERING_INDEX_SCAN + "ALLOW_COVERING_INDEX_SCAN", +#endif +#if SQLITE_ALLOW_URI_AUTHORITY + "ALLOW_URI_AUTHORITY", +#endif +#ifdef SQLITE_BITMASK_TYPE + "BITMASK_TYPE=" CTIMEOPT_VAL(SQLITE_BITMASK_TYPE), +#endif +#if SQLITE_BUG_COMPATIBLE_20160819 + "BUG_COMPATIBLE_20160819", +#endif +#if SQLITE_CASE_SENSITIVE_LIKE + "CASE_SENSITIVE_LIKE", +#endif +#if SQLITE_CHECK_PAGES + "CHECK_PAGES", +#endif +#if defined(__clang__) && defined(__clang_major__) + "COMPILER=clang-" CTIMEOPT_VAL(__clang_major__) "." + CTIMEOPT_VAL(__clang_minor__) "." + CTIMEOPT_VAL(__clang_patchlevel__), +#elif defined(_MSC_VER) + "COMPILER=msvc-" CTIMEOPT_VAL(_MSC_VER), +#elif defined(__GNUC__) && defined(__VERSION__) + "COMPILER=gcc-" __VERSION__, +#endif +#if SQLITE_COVERAGE_TEST + "COVERAGE_TEST", +#endif +#if SQLITE_DEBUG + "DEBUG", +#endif +#if SQLITE_DEFAULT_AUTOMATIC_INDEX + "DEFAULT_AUTOMATIC_INDEX", +#endif +#if SQLITE_DEFAULT_AUTOVACUUM + "DEFAULT_AUTOVACUUM", +#endif +#ifdef SQLITE_DEFAULT_CACHE_SIZE + "DEFAULT_CACHE_SIZE=" CTIMEOPT_VAL(SQLITE_DEFAULT_CACHE_SIZE), +#endif +#if SQLITE_DEFAULT_CKPTFULLFSYNC + "DEFAULT_CKPTFULLFSYNC", +#endif +#ifdef SQLITE_DEFAULT_FILE_FORMAT + "DEFAULT_FILE_FORMAT=" CTIMEOPT_VAL(SQLITE_DEFAULT_FILE_FORMAT), +#endif +#ifdef SQLITE_DEFAULT_FILE_PERMISSIONS + "DEFAULT_FILE_PERMISSIONS=" CTIMEOPT_VAL(SQLITE_DEFAULT_FILE_PERMISSIONS), +#endif +#if SQLITE_DEFAULT_FOREIGN_KEYS + "DEFAULT_FOREIGN_KEYS", +#endif +#ifdef SQLITE_DEFAULT_JOURNAL_SIZE_LIMIT + "DEFAULT_JOURNAL_SIZE_LIMIT=" CTIMEOPT_VAL(SQLITE_DEFAULT_JOURNAL_SIZE_LIMIT), +#endif +#ifdef SQLITE_DEFAULT_LOCKING_MODE + "DEFAULT_LOCKING_MODE=" CTIMEOPT_VAL(SQLITE_DEFAULT_LOCKING_MODE), +#endif +#ifdef SQLITE_DEFAULT_LOOKASIDE + "DEFAULT_LOOKASIDE=" CTIMEOPT_VAL2(SQLITE_DEFAULT_LOOKASIDE), +#endif +#if SQLITE_DEFAULT_MEMSTATUS + "DEFAULT_MEMSTATUS", +#endif +#ifdef SQLITE_DEFAULT_MMAP_SIZE + "DEFAULT_MMAP_SIZE=" CTIMEOPT_VAL(SQLITE_DEFAULT_MMAP_SIZE), +#endif +#ifdef SQLITE_DEFAULT_PAGE_SIZE + "DEFAULT_PAGE_SIZE=" CTIMEOPT_VAL(SQLITE_DEFAULT_PAGE_SIZE), +#endif +#ifdef SQLITE_DEFAULT_PCACHE_INITSZ + "DEFAULT_PCACHE_INITSZ=" CTIMEOPT_VAL(SQLITE_DEFAULT_PCACHE_INITSZ), +#endif +#ifdef SQLITE_DEFAULT_PROXYDIR_PERMISSIONS + "DEFAULT_PROXYDIR_PERMISSIONS=" CTIMEOPT_VAL(SQLITE_DEFAULT_PROXYDIR_PERMISSIONS), +#endif +#if SQLITE_DEFAULT_RECURSIVE_TRIGGERS + "DEFAULT_RECURSIVE_TRIGGERS", +#endif +#ifdef SQLITE_DEFAULT_ROWEST + "DEFAULT_ROWEST=" CTIMEOPT_VAL(SQLITE_DEFAULT_ROWEST), +#endif +#ifdef SQLITE_DEFAULT_SECTOR_SIZE + "DEFAULT_SECTOR_SIZE=" CTIMEOPT_VAL(SQLITE_DEFAULT_SECTOR_SIZE), +#endif +#ifdef SQLITE_DEFAULT_SYNCHRONOUS + "DEFAULT_SYNCHRONOUS=" CTIMEOPT_VAL(SQLITE_DEFAULT_SYNCHRONOUS), +#endif +#ifdef SQLITE_DEFAULT_WAL_AUTOCHECKPOINT + "DEFAULT_WAL_AUTOCHECKPOINT=" CTIMEOPT_VAL(SQLITE_DEFAULT_WAL_AUTOCHECKPOINT), +#endif +#ifdef SQLITE_DEFAULT_WAL_SYNCHRONOUS + "DEFAULT_WAL_SYNCHRONOUS=" CTIMEOPT_VAL(SQLITE_DEFAULT_WAL_SYNCHRONOUS), +#endif +#ifdef SQLITE_DEFAULT_WORKER_THREADS + "DEFAULT_WORKER_THREADS=" CTIMEOPT_VAL(SQLITE_DEFAULT_WORKER_THREADS), +#endif +#if SQLITE_DIRECT_OVERFLOW_READ + "DIRECT_OVERFLOW_READ", +#endif +#if SQLITE_DISABLE_DIRSYNC + "DISABLE_DIRSYNC", +#endif +#if SQLITE_DISABLE_FTS3_UNICODE + "DISABLE_FTS3_UNICODE", +#endif +#if SQLITE_DISABLE_FTS4_DEFERRED + "DISABLE_FTS4_DEFERRED", +#endif +#if SQLITE_DISABLE_INTRINSIC + "DISABLE_INTRINSIC", +#endif +#if SQLITE_DISABLE_LFS + "DISABLE_LFS", +#endif +#if SQLITE_DISABLE_PAGECACHE_OVERFLOW_STATS + "DISABLE_PAGECACHE_OVERFLOW_STATS", +#endif +#if SQLITE_DISABLE_SKIPAHEAD_DISTINCT + "DISABLE_SKIPAHEAD_DISTINCT", +#endif +#ifdef SQLITE_ENABLE_8_3_NAMES + "ENABLE_8_3_NAMES=" CTIMEOPT_VAL(SQLITE_ENABLE_8_3_NAMES), +#endif +#if SQLITE_ENABLE_API_ARMOR + "ENABLE_API_ARMOR", +#endif +#if SQLITE_ENABLE_ATOMIC_WRITE + "ENABLE_ATOMIC_WRITE", +#endif +#if SQLITE_ENABLE_BATCH_ATOMIC_WRITE + "ENABLE_BATCH_ATOMIC_WRITE", +#endif +#if SQLITE_ENABLE_CEROD + "ENABLE_CEROD=" CTIMEOPT_VAL(SQLITE_ENABLE_CEROD), +#endif +#if SQLITE_ENABLE_COLUMN_METADATA + "ENABLE_COLUMN_METADATA", +#endif +#if SQLITE_ENABLE_COLUMN_USED_MASK + "ENABLE_COLUMN_USED_MASK", +#endif +#if SQLITE_ENABLE_COSTMULT + "ENABLE_COSTMULT", +#endif +#if SQLITE_ENABLE_CURSOR_HINTS + "ENABLE_CURSOR_HINTS", +#endif +#if SQLITE_ENABLE_DBSTAT_VTAB + "ENABLE_DBSTAT_VTAB", +#endif +#if SQLITE_ENABLE_EXPENSIVE_ASSERT + "ENABLE_EXPENSIVE_ASSERT", +#endif +#if SQLITE_ENABLE_FTS1 + "ENABLE_FTS1", +#endif +#if SQLITE_ENABLE_FTS2 + "ENABLE_FTS2", +#endif +#if SQLITE_ENABLE_FTS3 + "ENABLE_FTS3", +#endif +#if SQLITE_ENABLE_FTS3_PARENTHESIS + "ENABLE_FTS3_PARENTHESIS", +#endif +#if SQLITE_ENABLE_FTS3_TOKENIZER + "ENABLE_FTS3_TOKENIZER", +#endif +#if SQLITE_ENABLE_FTS4 + "ENABLE_FTS4", +#endif +#if SQLITE_ENABLE_FTS5 + "ENABLE_FTS5", +#endif +#if SQLITE_ENABLE_GEOPOLY + "ENABLE_GEOPOLY", +#endif +#if SQLITE_ENABLE_HIDDEN_COLUMNS + "ENABLE_HIDDEN_COLUMNS", +#endif +#if SQLITE_ENABLE_ICU + "ENABLE_ICU", +#endif +#if SQLITE_ENABLE_IOTRACE + "ENABLE_IOTRACE", +#endif +#if SQLITE_ENABLE_JSON1 + "ENABLE_JSON1", +#endif +#if SQLITE_ENABLE_LOAD_EXTENSION + "ENABLE_LOAD_EXTENSION", +#endif +#ifdef SQLITE_ENABLE_LOCKING_STYLE + "ENABLE_LOCKING_STYLE=" CTIMEOPT_VAL(SQLITE_ENABLE_LOCKING_STYLE), +#endif +#if SQLITE_ENABLE_MEMORY_MANAGEMENT + "ENABLE_MEMORY_MANAGEMENT", +#endif +#if SQLITE_ENABLE_MEMSYS3 + "ENABLE_MEMSYS3", +#endif +#if SQLITE_ENABLE_MEMSYS5 + "ENABLE_MEMSYS5", +#endif +#if SQLITE_ENABLE_MULTIPLEX + "ENABLE_MULTIPLEX", +#endif +#if SQLITE_ENABLE_NORMALIZE + "ENABLE_NORMALIZE", +#endif +#if SQLITE_ENABLE_NULL_TRIM + "ENABLE_NULL_TRIM", +#endif +#if SQLITE_ENABLE_OVERSIZE_CELL_CHECK + "ENABLE_OVERSIZE_CELL_CHECK", +#endif +#if SQLITE_ENABLE_PREUPDATE_HOOK + "ENABLE_PREUPDATE_HOOK", +#endif +#if SQLITE_ENABLE_QPSG + "ENABLE_QPSG", +#endif +#if SQLITE_ENABLE_RBU + "ENABLE_RBU", +#endif +#if SQLITE_ENABLE_RTREE + "ENABLE_RTREE", +#endif +#if SQLITE_ENABLE_SELECTTRACE + "ENABLE_SELECTTRACE", +#endif +#if SQLITE_ENABLE_SESSION + "ENABLE_SESSION", +#endif +#if SQLITE_ENABLE_SNAPSHOT + "ENABLE_SNAPSHOT", +#endif +#if SQLITE_ENABLE_SORTER_REFERENCES + "ENABLE_SORTER_REFERENCES", +#endif +#if SQLITE_ENABLE_SQLLOG + "ENABLE_SQLLOG", +#endif +#if defined(SQLITE_ENABLE_STAT4) + "ENABLE_STAT4", +#endif +#if SQLITE_ENABLE_STMTVTAB + "ENABLE_STMTVTAB", +#endif +#if SQLITE_ENABLE_STMT_SCANSTATUS + "ENABLE_STMT_SCANSTATUS", +#endif +#if SQLITE_ENABLE_UNKNOWN_SQL_FUNCTION + "ENABLE_UNKNOWN_SQL_FUNCTION", +#endif +#if SQLITE_ENABLE_UNLOCK_NOTIFY + "ENABLE_UNLOCK_NOTIFY", +#endif +#if SQLITE_ENABLE_UPDATE_DELETE_LIMIT + "ENABLE_UPDATE_DELETE_LIMIT", +#endif +#if SQLITE_ENABLE_URI_00_ERROR + "ENABLE_URI_00_ERROR", +#endif +#if SQLITE_ENABLE_VFSTRACE + "ENABLE_VFSTRACE", +#endif +#if SQLITE_ENABLE_WHERETRACE + "ENABLE_WHERETRACE", +#endif +#if SQLITE_ENABLE_ZIPVFS + "ENABLE_ZIPVFS", +#endif +#if SQLITE_EXPLAIN_ESTIMATED_ROWS + "EXPLAIN_ESTIMATED_ROWS", +#endif +#if SQLITE_EXTRA_IFNULLROW + "EXTRA_IFNULLROW", +#endif +#ifdef SQLITE_EXTRA_INIT + "EXTRA_INIT=" CTIMEOPT_VAL(SQLITE_EXTRA_INIT), +#endif +#ifdef SQLITE_EXTRA_SHUTDOWN + "EXTRA_SHUTDOWN=" CTIMEOPT_VAL(SQLITE_EXTRA_SHUTDOWN), +#endif +#ifdef SQLITE_FTS3_MAX_EXPR_DEPTH + "FTS3_MAX_EXPR_DEPTH=" CTIMEOPT_VAL(SQLITE_FTS3_MAX_EXPR_DEPTH), +#endif +#if SQLITE_FTS5_ENABLE_TEST_MI + "FTS5_ENABLE_TEST_MI", +#endif +#if SQLITE_FTS5_NO_WITHOUT_ROWID + "FTS5_NO_WITHOUT_ROWID", +#endif +#if SQLITE_HAS_CODEC + "HAS_CODEC", +#endif +#if HAVE_ISNAN || SQLITE_HAVE_ISNAN + "HAVE_ISNAN", +#endif +#if SQLITE_HOMEGROWN_RECURSIVE_MUTEX + "HOMEGROWN_RECURSIVE_MUTEX", +#endif +#if SQLITE_IGNORE_AFP_LOCK_ERRORS + "IGNORE_AFP_LOCK_ERRORS", +#endif +#if SQLITE_IGNORE_FLOCK_LOCK_ERRORS + "IGNORE_FLOCK_LOCK_ERRORS", +#endif +#if SQLITE_INLINE_MEMCPY + "INLINE_MEMCPY", +#endif +#if SQLITE_INT64_TYPE + "INT64_TYPE", +#endif +#ifdef SQLITE_INTEGRITY_CHECK_ERROR_MAX + "INTEGRITY_CHECK_ERROR_MAX=" CTIMEOPT_VAL(SQLITE_INTEGRITY_CHECK_ERROR_MAX), +#endif +#if SQLITE_LIKE_DOESNT_MATCH_BLOBS + "LIKE_DOESNT_MATCH_BLOBS", +#endif +#if SQLITE_LOCK_TRACE + "LOCK_TRACE", +#endif +#if SQLITE_LOG_CACHE_SPILL + "LOG_CACHE_SPILL", +#endif +#ifdef SQLITE_MALLOC_SOFT_LIMIT + "MALLOC_SOFT_LIMIT=" CTIMEOPT_VAL(SQLITE_MALLOC_SOFT_LIMIT), +#endif +#ifdef SQLITE_MAX_ATTACHED + "MAX_ATTACHED=" CTIMEOPT_VAL(SQLITE_MAX_ATTACHED), +#endif +#ifdef SQLITE_MAX_COLUMN + "MAX_COLUMN=" CTIMEOPT_VAL(SQLITE_MAX_COLUMN), +#endif +#ifdef SQLITE_MAX_COMPOUND_SELECT + "MAX_COMPOUND_SELECT=" CTIMEOPT_VAL(SQLITE_MAX_COMPOUND_SELECT), +#endif +#ifdef SQLITE_MAX_DEFAULT_PAGE_SIZE + "MAX_DEFAULT_PAGE_SIZE=" CTIMEOPT_VAL(SQLITE_MAX_DEFAULT_PAGE_SIZE), +#endif +#ifdef SQLITE_MAX_EXPR_DEPTH + "MAX_EXPR_DEPTH=" CTIMEOPT_VAL(SQLITE_MAX_EXPR_DEPTH), +#endif +#ifdef SQLITE_MAX_FUNCTION_ARG + "MAX_FUNCTION_ARG=" CTIMEOPT_VAL(SQLITE_MAX_FUNCTION_ARG), +#endif +#ifdef SQLITE_MAX_LENGTH + "MAX_LENGTH=" CTIMEOPT_VAL(SQLITE_MAX_LENGTH), +#endif +#ifdef SQLITE_MAX_LIKE_PATTERN_LENGTH + "MAX_LIKE_PATTERN_LENGTH=" CTIMEOPT_VAL(SQLITE_MAX_LIKE_PATTERN_LENGTH), +#endif +#ifdef SQLITE_MAX_MEMORY + "MAX_MEMORY=" CTIMEOPT_VAL(SQLITE_MAX_MEMORY), +#endif +#ifdef SQLITE_MAX_MMAP_SIZE + "MAX_MMAP_SIZE=" CTIMEOPT_VAL(SQLITE_MAX_MMAP_SIZE), +#endif +#ifdef SQLITE_MAX_MMAP_SIZE_ + "MAX_MMAP_SIZE_=" CTIMEOPT_VAL(SQLITE_MAX_MMAP_SIZE_), +#endif +#ifdef SQLITE_MAX_PAGE_COUNT + "MAX_PAGE_COUNT=" CTIMEOPT_VAL(SQLITE_MAX_PAGE_COUNT), +#endif +#ifdef SQLITE_MAX_PAGE_SIZE + "MAX_PAGE_SIZE=" CTIMEOPT_VAL(SQLITE_MAX_PAGE_SIZE), +#endif +#ifdef SQLITE_MAX_SCHEMA_RETRY + "MAX_SCHEMA_RETRY=" CTIMEOPT_VAL(SQLITE_MAX_SCHEMA_RETRY), +#endif +#ifdef SQLITE_MAX_SQL_LENGTH + "MAX_SQL_LENGTH=" CTIMEOPT_VAL(SQLITE_MAX_SQL_LENGTH), +#endif +#ifdef SQLITE_MAX_TRIGGER_DEPTH + "MAX_TRIGGER_DEPTH=" CTIMEOPT_VAL(SQLITE_MAX_TRIGGER_DEPTH), +#endif +#ifdef SQLITE_MAX_VARIABLE_NUMBER + "MAX_VARIABLE_NUMBER=" CTIMEOPT_VAL(SQLITE_MAX_VARIABLE_NUMBER), +#endif +#ifdef SQLITE_MAX_VDBE_OP + "MAX_VDBE_OP=" CTIMEOPT_VAL(SQLITE_MAX_VDBE_OP), +#endif +#ifdef SQLITE_MAX_WORKER_THREADS + "MAX_WORKER_THREADS=" CTIMEOPT_VAL(SQLITE_MAX_WORKER_THREADS), +#endif +#if SQLITE_MEMDEBUG + "MEMDEBUG", +#endif +#if SQLITE_MIXED_ENDIAN_64BIT_FLOAT + "MIXED_ENDIAN_64BIT_FLOAT", +#endif +#if SQLITE_MMAP_READWRITE + "MMAP_READWRITE", +#endif +#if SQLITE_MUTEX_NOOP + "MUTEX_NOOP", +#endif +#if SQLITE_MUTEX_NREF + "MUTEX_NREF", +#endif +#if SQLITE_MUTEX_OMIT + "MUTEX_OMIT", +#endif +#if SQLITE_MUTEX_PTHREADS + "MUTEX_PTHREADS", +#endif +#if SQLITE_MUTEX_W32 + "MUTEX_W32", +#endif +#if SQLITE_NEED_ERR_NAME + "NEED_ERR_NAME", +#endif +#if SQLITE_NOINLINE + "NOINLINE", +#endif +#if SQLITE_NO_SYNC + "NO_SYNC", +#endif +#if SQLITE_OMIT_ALTERTABLE + "OMIT_ALTERTABLE", +#endif +#if SQLITE_OMIT_ANALYZE + "OMIT_ANALYZE", +#endif +#if SQLITE_OMIT_ATTACH + "OMIT_ATTACH", +#endif +#if SQLITE_OMIT_AUTHORIZATION + "OMIT_AUTHORIZATION", +#endif +#if SQLITE_OMIT_AUTOINCREMENT + "OMIT_AUTOINCREMENT", +#endif +#if SQLITE_OMIT_AUTOINIT + "OMIT_AUTOINIT", +#endif +#if SQLITE_OMIT_AUTOMATIC_INDEX + "OMIT_AUTOMATIC_INDEX", +#endif +#if SQLITE_OMIT_AUTORESET + "OMIT_AUTORESET", +#endif +#if SQLITE_OMIT_AUTOVACUUM + "OMIT_AUTOVACUUM", +#endif +#if SQLITE_OMIT_BETWEEN_OPTIMIZATION + "OMIT_BETWEEN_OPTIMIZATION", +#endif +#if SQLITE_OMIT_BLOB_LITERAL + "OMIT_BLOB_LITERAL", +#endif +#if SQLITE_OMIT_BTREECOUNT + "OMIT_BTREECOUNT", +#endif +#if SQLITE_OMIT_CAST + "OMIT_CAST", +#endif +#if SQLITE_OMIT_CHECK + "OMIT_CHECK", +#endif +#if SQLITE_OMIT_COMPLETE + "OMIT_COMPLETE", +#endif +#if SQLITE_OMIT_COMPOUND_SELECT + "OMIT_COMPOUND_SELECT", +#endif +#if SQLITE_OMIT_CONFLICT_CLAUSE + "OMIT_CONFLICT_CLAUSE", +#endif +#if SQLITE_OMIT_CTE + "OMIT_CTE", +#endif +#if SQLITE_OMIT_DATETIME_FUNCS + "OMIT_DATETIME_FUNCS", +#endif +#if SQLITE_OMIT_DECLTYPE + "OMIT_DECLTYPE", +#endif +#if SQLITE_OMIT_DEPRECATED + "OMIT_DEPRECATED", +#endif +#if SQLITE_OMIT_DISKIO + "OMIT_DISKIO", +#endif +#if SQLITE_OMIT_EXPLAIN + "OMIT_EXPLAIN", +#endif +#if SQLITE_OMIT_FLAG_PRAGMAS + "OMIT_FLAG_PRAGMAS", +#endif +#if SQLITE_OMIT_FLOATING_POINT + "OMIT_FLOATING_POINT", +#endif +#if SQLITE_OMIT_FOREIGN_KEY + "OMIT_FOREIGN_KEY", +#endif +#if SQLITE_OMIT_GET_TABLE + "OMIT_GET_TABLE", +#endif +#if SQLITE_OMIT_HEX_INTEGER + "OMIT_HEX_INTEGER", +#endif +#if SQLITE_OMIT_INCRBLOB + "OMIT_INCRBLOB", +#endif +#if SQLITE_OMIT_INTEGRITY_CHECK + "OMIT_INTEGRITY_CHECK", +#endif +#if SQLITE_OMIT_LIKE_OPTIMIZATION + "OMIT_LIKE_OPTIMIZATION", +#endif +#if SQLITE_OMIT_LOAD_EXTENSION + "OMIT_LOAD_EXTENSION", +#endif +#if SQLITE_OMIT_LOCALTIME + "OMIT_LOCALTIME", +#endif +#if SQLITE_OMIT_LOOKASIDE + "OMIT_LOOKASIDE", +#endif +#if SQLITE_OMIT_MEMORYDB + "OMIT_MEMORYDB", +#endif +#if SQLITE_OMIT_OR_OPTIMIZATION + "OMIT_OR_OPTIMIZATION", +#endif +#if SQLITE_OMIT_PAGER_PRAGMAS + "OMIT_PAGER_PRAGMAS", +#endif +#if SQLITE_OMIT_PARSER_TRACE + "OMIT_PARSER_TRACE", +#endif +#if SQLITE_OMIT_POPEN + "OMIT_POPEN", +#endif +#if SQLITE_OMIT_PRAGMA + "OMIT_PRAGMA", +#endif +#if SQLITE_OMIT_PROGRESS_CALLBACK + "OMIT_PROGRESS_CALLBACK", +#endif +#if SQLITE_OMIT_QUICKBALANCE + "OMIT_QUICKBALANCE", +#endif +#if SQLITE_OMIT_REINDEX + "OMIT_REINDEX", +#endif +#if SQLITE_OMIT_SCHEMA_PRAGMAS + "OMIT_SCHEMA_PRAGMAS", +#endif +#if SQLITE_OMIT_SCHEMA_VERSION_PRAGMAS + "OMIT_SCHEMA_VERSION_PRAGMAS", +#endif +#if SQLITE_OMIT_SHARED_CACHE + "OMIT_SHARED_CACHE", +#endif +#if SQLITE_OMIT_SHUTDOWN_DIRECTORIES + "OMIT_SHUTDOWN_DIRECTORIES", +#endif +#if SQLITE_OMIT_SUBQUERY + "OMIT_SUBQUERY", +#endif +#if SQLITE_OMIT_TCL_VARIABLE + "OMIT_TCL_VARIABLE", +#endif +#if SQLITE_OMIT_TEMPDB + "OMIT_TEMPDB", +#endif +#if SQLITE_OMIT_TEST_CONTROL + "OMIT_TEST_CONTROL", +#endif +#if SQLITE_OMIT_TRACE + "OMIT_TRACE", +#endif +#if SQLITE_OMIT_TRIGGER + "OMIT_TRIGGER", +#endif +#if SQLITE_OMIT_TRUNCATE_OPTIMIZATION + "OMIT_TRUNCATE_OPTIMIZATION", +#endif +#if SQLITE_OMIT_UTF16 + "OMIT_UTF16", +#endif +#if SQLITE_OMIT_VACUUM + "OMIT_VACUUM", +#endif +#if SQLITE_OMIT_VIEW + "OMIT_VIEW", +#endif +#if SQLITE_OMIT_VIRTUALTABLE + "OMIT_VIRTUALTABLE", +#endif +#if SQLITE_OMIT_WAL + "OMIT_WAL", +#endif +#if SQLITE_OMIT_WSD + "OMIT_WSD", +#endif +#if SQLITE_OMIT_XFER_OPT + "OMIT_XFER_OPT", +#endif +#if SQLITE_PCACHE_SEPARATE_HEADER + "PCACHE_SEPARATE_HEADER", +#endif +#if SQLITE_PERFORMANCE_TRACE + "PERFORMANCE_TRACE", +#endif +#if SQLITE_POWERSAFE_OVERWRITE + "POWERSAFE_OVERWRITE", +#endif +#if SQLITE_PREFER_PROXY_LOCKING + "PREFER_PROXY_LOCKING", +#endif +#if SQLITE_PROXY_DEBUG + "PROXY_DEBUG", +#endif +#if SQLITE_REVERSE_UNORDERED_SELECTS + "REVERSE_UNORDERED_SELECTS", +#endif +#if SQLITE_RTREE_INT_ONLY + "RTREE_INT_ONLY", +#endif +#if SQLITE_SECURE_DELETE + "SECURE_DELETE", +#endif +#if SQLITE_SMALL_STACK + "SMALL_STACK", +#endif +#ifdef SQLITE_SORTER_PMASZ + "SORTER_PMASZ=" CTIMEOPT_VAL(SQLITE_SORTER_PMASZ), +#endif +#if SQLITE_SOUNDEX + "SOUNDEX", +#endif +#ifdef SQLITE_STAT4_SAMPLES + "STAT4_SAMPLES=" CTIMEOPT_VAL(SQLITE_STAT4_SAMPLES), +#endif +#ifdef SQLITE_STMTJRNL_SPILL + "STMTJRNL_SPILL=" CTIMEOPT_VAL(SQLITE_STMTJRNL_SPILL), +#endif +#if SQLITE_SUBSTR_COMPATIBILITY + "SUBSTR_COMPATIBILITY", +#endif +#if SQLITE_SYSTEM_MALLOC + "SYSTEM_MALLOC", +#endif +#if SQLITE_TCL + "TCL", +#endif +#ifdef SQLITE_TEMP_STORE + "TEMP_STORE=" CTIMEOPT_VAL(SQLITE_TEMP_STORE), +#endif +#if SQLITE_TEST + "TEST", +#endif +#if defined(SQLITE_THREADSAFE) + "THREADSAFE=" CTIMEOPT_VAL(SQLITE_THREADSAFE), +#elif defined(THREADSAFE) + "THREADSAFE=" CTIMEOPT_VAL(THREADSAFE), +#else + "THREADSAFE=1", +#endif +#if SQLITE_UNLINK_AFTER_CLOSE + "UNLINK_AFTER_CLOSE", +#endif +#if SQLITE_UNTESTABLE + "UNTESTABLE", +#endif +#if SQLITE_USER_AUTHENTICATION + "USER_AUTHENTICATION", +#endif +#if SQLITE_USE_ALLOCA + "USE_ALLOCA", +#endif +#if SQLITE_USE_FCNTL_TRACE + "USE_FCNTL_TRACE", +#endif +#if SQLITE_USE_URI + "USE_URI", +#endif +#if SQLITE_VDBE_COVERAGE + "VDBE_COVERAGE", +#endif +#if SQLITE_WIN32_MALLOC + "WIN32_MALLOC", +#endif +#if SQLITE_ZERO_MALLOC + "ZERO_MALLOC", +#endif +/* +** END CODE GENERATED BY tool/mkctime.tcl +*/ +}; + +SQLITE_PRIVATE const char **sqlite3CompileOptions(int *pnOpt){ + *pnOpt = sizeof(sqlite3azCompileOpt) / sizeof(sqlite3azCompileOpt[0]); + return (const char**)sqlite3azCompileOpt; +} + +#endif /* SQLITE_OMIT_COMPILEOPTION_DIAGS */ + +/************** End of ctime.c ***********************************************/ +/************** Begin file sqliteInt.h ***************************************/ +/* +** 2001 September 15 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** Internal interface definitions for SQLite. +** +*/ +#ifndef SQLITEINT_H +#define SQLITEINT_H + +/* Special Comments: +** +** Some comments have special meaning to the tools that measure test +** coverage: +** +** NO_TEST - The branches on this line are not +** measured by branch coverage. This is +** used on lines of code that actually +** implement parts of coverage testing. +** +** OPTIMIZATION-IF-TRUE - This branch is allowed to alway be false +** and the correct answer is still obtained, +** though perhaps more slowly. +** +** OPTIMIZATION-IF-FALSE - This branch is allowed to alway be true +** and the correct answer is still obtained, +** though perhaps more slowly. +** +** PREVENTS-HARMLESS-OVERREAD - This branch prevents a buffer overread +** that would be harmless and undetectable +** if it did occur. +** +** In all cases, the special comment must be enclosed in the usual +** slash-asterisk...asterisk-slash comment marks, with no spaces between the +** asterisks and the comment text. +*/ + +/* +** Make sure the Tcl calling convention macro is defined. This macro is +** only used by test code and Tcl integration code. +*/ +#ifndef SQLITE_TCLAPI +# define SQLITE_TCLAPI +#endif + +/* +** Include the header file used to customize the compiler options for MSVC. +** This should be done first so that it can successfully prevent spurious +** compiler warnings due to subsequent content in this file and other files +** that are included by this file. +*/ +/************** Include msvc.h in the middle of sqliteInt.h ******************/ +/************** Begin file msvc.h ********************************************/ +/* +** 2015 January 12 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +****************************************************************************** +** +** This file contains code that is specific to MSVC. +*/ +#ifndef SQLITE_MSVC_H +#define SQLITE_MSVC_H + +#if defined(_MSC_VER) +#pragma warning(disable : 4054) +#pragma warning(disable : 4055) +#pragma warning(disable : 4100) +#pragma warning(disable : 4127) +#pragma warning(disable : 4130) +#pragma warning(disable : 4152) +#pragma warning(disable : 4189) +#pragma warning(disable : 4206) +#pragma warning(disable : 4210) +#pragma warning(disable : 4232) +#pragma warning(disable : 4244) +#pragma warning(disable : 4305) +#pragma warning(disable : 4306) +#pragma warning(disable : 4702) +#pragma warning(disable : 4706) +#endif /* defined(_MSC_VER) */ + +#if defined(_MSC_VER) && !defined(_WIN64) +#undef SQLITE_4_BYTE_ALIGNED_MALLOC +#define SQLITE_4_BYTE_ALIGNED_MALLOC +#endif /* defined(_MSC_VER) && !defined(_WIN64) */ + +#endif /* SQLITE_MSVC_H */ + +/************** End of msvc.h ************************************************/ +/************** Continuing where we left off in sqliteInt.h ******************/ + +/* +** Special setup for VxWorks +*/ +/************** Include vxworks.h in the middle of sqliteInt.h ***************/ +/************** Begin file vxworks.h *****************************************/ +/* +** 2015-03-02 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +****************************************************************************** +** +** This file contains code that is specific to Wind River's VxWorks +*/ +#if defined(__RTP__) || defined(_WRS_KERNEL) +/* This is VxWorks. Set up things specially for that OS +*/ +#include +#include /* amalgamator: dontcache */ +#define OS_VXWORKS 1 +#define SQLITE_OS_OTHER 0 +#define SQLITE_HOMEGROWN_RECURSIVE_MUTEX 1 +#define SQLITE_OMIT_LOAD_EXTENSION 1 +#define SQLITE_ENABLE_LOCKING_STYLE 0 +#define HAVE_UTIME 1 +#else +/* This is not VxWorks. */ +#define OS_VXWORKS 0 +#define HAVE_FCHOWN 1 +#define HAVE_READLINK 1 +#define HAVE_LSTAT 1 +#endif /* defined(_WRS_KERNEL) */ + +/************** End of vxworks.h *********************************************/ +/************** Continuing where we left off in sqliteInt.h ******************/ + +/* +** These #defines should enable >2GB file support on POSIX if the +** underlying operating system supports it. If the OS lacks +** large file support, or if the OS is windows, these should be no-ops. +** +** Ticket #2739: The _LARGEFILE_SOURCE macro must appear before any +** system #includes. Hence, this block of code must be the very first +** code in all source files. +** +** Large file support can be disabled using the -DSQLITE_DISABLE_LFS switch +** on the compiler command line. This is necessary if you are compiling +** on a recent machine (ex: Red Hat 7.2) but you want your code to work +** on an older machine (ex: Red Hat 6.0). If you compile on Red Hat 7.2 +** without this option, LFS is enable. But LFS does not exist in the kernel +** in Red Hat 6.0, so the code won't work. Hence, for maximum binary +** portability you should omit LFS. +** +** The previous paragraph was written in 2005. (This paragraph is written +** on 2008-11-28.) These days, all Linux kernels support large files, so +** you should probably leave LFS enabled. But some embedded platforms might +** lack LFS in which case the SQLITE_DISABLE_LFS macro might still be useful. +** +** Similar is true for Mac OS X. LFS is only supported on Mac OS X 9 and later. +*/ +#ifndef SQLITE_DISABLE_LFS +# define _LARGE_FILE 1 +# ifndef _FILE_OFFSET_BITS +# define _FILE_OFFSET_BITS 64 +# endif +# define _LARGEFILE_SOURCE 1 +#endif + +/* The GCC_VERSION and MSVC_VERSION macros are used to +** conditionally include optimizations for each of these compilers. A +** value of 0 means that compiler is not being used. The +** SQLITE_DISABLE_INTRINSIC macro means do not use any compiler-specific +** optimizations, and hence set all compiler macros to 0 +** +** There was once also a CLANG_VERSION macro. However, we learn that the +** version numbers in clang are for "marketing" only and are inconsistent +** and unreliable. Fortunately, all versions of clang also recognize the +** gcc version numbers and have reasonable settings for gcc version numbers, +** so the GCC_VERSION macro will be set to a correct non-zero value even +** when compiling with clang. +*/ +#if defined(__GNUC__) && !defined(SQLITE_DISABLE_INTRINSIC) +# define GCC_VERSION (__GNUC__*1000000+__GNUC_MINOR__*1000+__GNUC_PATCHLEVEL__) +#else +# define GCC_VERSION 0 +#endif +#if defined(_MSC_VER) && !defined(SQLITE_DISABLE_INTRINSIC) +# define MSVC_VERSION _MSC_VER +#else +# define MSVC_VERSION 0 +#endif + +/* Needed for various definitions... */ +#if defined(__GNUC__) && !defined(_GNU_SOURCE) +# define _GNU_SOURCE +#endif + +#if defined(__OpenBSD__) && !defined(_BSD_SOURCE) +# define _BSD_SOURCE +#endif + +/* +** For MinGW, check to see if we can include the header file containing its +** version information, among other things. Normally, this internal MinGW +** header file would [only] be included automatically by other MinGW header +** files; however, the contained version information is now required by this +** header file to work around binary compatibility issues (see below) and +** this is the only known way to reliably obtain it. This entire #if block +** would be completely unnecessary if there was any other way of detecting +** MinGW via their preprocessor (e.g. if they customized their GCC to define +** some MinGW-specific macros). When compiling for MinGW, either the +** _HAVE_MINGW_H or _HAVE__MINGW_H (note the extra underscore) macro must be +** defined; otherwise, detection of conditions specific to MinGW will be +** disabled. +*/ +#if defined(_HAVE_MINGW_H) +# include "mingw.h" +#elif defined(_HAVE__MINGW_H) +# include "_mingw.h" +#endif + +/* +** For MinGW version 4.x (and higher), check to see if the _USE_32BIT_TIME_T +** define is required to maintain binary compatibility with the MSVC runtime +** library in use (e.g. for Windows XP). +*/ +#if !defined(_USE_32BIT_TIME_T) && !defined(_USE_64BIT_TIME_T) && \ + defined(_WIN32) && !defined(_WIN64) && \ + defined(__MINGW_MAJOR_VERSION) && __MINGW_MAJOR_VERSION >= 4 && \ + defined(__MSVCRT__) +# define _USE_32BIT_TIME_T +#endif + +/* The public SQLite interface. The _FILE_OFFSET_BITS macro must appear +** first in QNX. Also, the _USE_32BIT_TIME_T macro must appear first for +** MinGW. +*/ +/************** Include sqlite3.h in the middle of sqliteInt.h ***************/ +/************** Begin file sqlite3.h *****************************************/ +/* +** 2001-09-15 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** This header file defines the interface that the SQLite library +** presents to client programs. If a C-function, structure, datatype, +** or constant definition does not appear in this file, then it is +** not a published API of SQLite, is subject to change without +** notice, and should not be referenced by programs that use SQLite. +** +** Some of the definitions that are in this file are marked as +** "experimental". Experimental interfaces are normally new +** features recently added to SQLite. We do not anticipate changes +** to experimental interfaces but reserve the right to make minor changes +** if experience from use "in the wild" suggest such changes are prudent. +** +** The official C-language API documentation for SQLite is derived +** from comments in this file. This file is the authoritative source +** on how SQLite interfaces are supposed to operate. +** +** The name of this file under configuration management is "sqlite.h.in". +** The makefile makes some minor changes to this file (such as inserting +** the version number) and changes its name to "sqlite3.h" as +** part of the build process. +*/ +#ifndef SQLITE3_H +#define SQLITE3_H +#include /* Needed for the definition of va_list */ + +/* +** Make sure we can call this stuff from C++. +*/ +#if 0 +extern "C" { +#endif + + +/* +** Provide the ability to override linkage features of the interface. +*/ +#ifndef SQLITE_EXTERN +# define SQLITE_EXTERN extern +#endif +#ifndef SQLITE_API +# define SQLITE_API +#endif +#ifndef SQLITE_CDECL +# define SQLITE_CDECL +#endif +#ifndef SQLITE_APICALL +# define SQLITE_APICALL +#endif +#ifndef SQLITE_STDCALL +# define SQLITE_STDCALL SQLITE_APICALL +#endif +#ifndef SQLITE_CALLBACK +# define SQLITE_CALLBACK +#endif +#ifndef SQLITE_SYSAPI +# define SQLITE_SYSAPI +#endif + +/* +** These no-op macros are used in front of interfaces to mark those +** interfaces as either deprecated or experimental. New applications +** should not use deprecated interfaces - they are supported for backwards +** compatibility only. Application writers should be aware that +** experimental interfaces are subject to change in point releases. +** +** These macros used to resolve to various kinds of compiler magic that +** would generate warning messages when they were used. But that +** compiler magic ended up generating such a flurry of bug reports +** that we have taken it all out and gone back to using simple +** noop macros. +*/ +#define SQLITE_DEPRECATED +#define SQLITE_EXPERIMENTAL + +/* +** Ensure these symbols were not defined by some previous header file. +*/ +#ifdef SQLITE_VERSION +# undef SQLITE_VERSION +#endif +#ifdef SQLITE_VERSION_NUMBER +# undef SQLITE_VERSION_NUMBER +#endif + +/* +** CAPI3REF: Compile-Time Library Version Numbers +** +** ^(The [SQLITE_VERSION] C preprocessor macro in the sqlite3.h header +** evaluates to a string literal that is the SQLite version in the +** format "X.Y.Z" where X is the major version number (always 3 for +** SQLite3) and Y is the minor version number and Z is the release number.)^ +** ^(The [SQLITE_VERSION_NUMBER] C preprocessor macro resolves to an integer +** with the value (X*1000000 + Y*1000 + Z) where X, Y, and Z are the same +** numbers used in [SQLITE_VERSION].)^ +** The SQLITE_VERSION_NUMBER for any given release of SQLite will also +** be larger than the release from which it is derived. Either Y will +** be held constant and Z will be incremented or else Y will be incremented +** and Z will be reset to zero. +** +** Since [version 3.6.18] ([dateof:3.6.18]), +** SQLite source code has been stored in the +** Fossil configuration management +** system. ^The SQLITE_SOURCE_ID macro evaluates to +** a string which identifies a particular check-in of SQLite +** within its configuration management system. ^The SQLITE_SOURCE_ID +** string contains the date and time of the check-in (UTC) and a SHA1 +** or SHA3-256 hash of the entire source tree. If the source code has +** been edited in any way since it was last checked in, then the last +** four hexadecimal digits of the hash may be modified. +** +** See also: [sqlite3_libversion()], +** [sqlite3_libversion_number()], [sqlite3_sourceid()], +** [sqlite_version()] and [sqlite_source_id()]. +*/ +#define SQLITE_VERSION "3.30.1" +#define SQLITE_VERSION_NUMBER 3030001 +#define SQLITE_SOURCE_ID "2019-10-10 20:19:45 18db032d058f1436ce3dea84081f4ee5a0f2259ad97301d43c426bc7f3df1b0b" + +/* +** CAPI3REF: Run-Time Library Version Numbers +** KEYWORDS: sqlite3_version sqlite3_sourceid +** +** These interfaces provide the same information as the [SQLITE_VERSION], +** [SQLITE_VERSION_NUMBER], and [SQLITE_SOURCE_ID] C preprocessor macros +** but are associated with the library instead of the header file. ^(Cautious +** programmers might include assert() statements in their application to +** verify that values returned by these interfaces match the macros in +** the header, and thus ensure that the application is +** compiled with matching library and header files. +** +** 
+** assert( sqlite3_libversion_number()==SQLITE_VERSION_NUMBER );
+** assert( strncmp(sqlite3_sourceid(),SQLITE_SOURCE_ID,80)==0 );
+** assert( strcmp(sqlite3_libversion(),SQLITE_VERSION)==0 );
+**
)^ +** +** ^The sqlite3_version[] string constant contains the text of [SQLITE_VERSION] +** macro. ^The sqlite3_libversion() function returns a pointer to the +** to the sqlite3_version[] string constant. The sqlite3_libversion() +** function is provided for use in DLLs since DLL users usually do not have +** direct access to string constants within the DLL. ^The +** sqlite3_libversion_number() function returns an integer equal to +** [SQLITE_VERSION_NUMBER]. ^(The sqlite3_sourceid() function returns +** a pointer to a string constant whose value is the same as the +** [SQLITE_SOURCE_ID] C preprocessor macro. Except if SQLite is built +** using an edited copy of [the amalgamation], then the last four characters +** of the hash might be different from [SQLITE_SOURCE_ID].)^ +** +** See also: [sqlite_version()] and [sqlite_source_id()]. +*/ +SQLITE_API const char sqlite3_version[] = SQLITE_VERSION; +SQLITE_API const char *sqlite3_libversion(void); +SQLITE_API const char *sqlite3_sourceid(void); +SQLITE_API int sqlite3_libversion_number(void); + +/* +** CAPI3REF: Run-Time Library Compilation Options Diagnostics +** +** ^The sqlite3_compileoption_used() function returns 0 or 1 +** indicating whether the specified option was defined at +** compile time. ^The SQLITE_ prefix may be omitted from the +** option name passed to sqlite3_compileoption_used(). +** +** ^The sqlite3_compileoption_get() function allows iterating +** over the list of options that were defined at compile time by +** returning the N-th compile time option string. ^If N is out of range, +** sqlite3_compileoption_get() returns a NULL pointer. ^The SQLITE_ +** prefix is omitted from any strings returned by +** sqlite3_compileoption_get(). +** +** ^Support for the diagnostic functions sqlite3_compileoption_used() +** and sqlite3_compileoption_get() may be omitted by specifying the +** [SQLITE_OMIT_COMPILEOPTION_DIAGS] option at compile time. +** +** See also: SQL functions [sqlite_compileoption_used()] and +** [sqlite_compileoption_get()] and the [compile_options pragma]. +*/ +#ifndef SQLITE_OMIT_COMPILEOPTION_DIAGS +SQLITE_API int sqlite3_compileoption_used(const char *zOptName); +SQLITE_API const char *sqlite3_compileoption_get(int N); +#else +# define sqlite3_compileoption_used(X) 0 +# define sqlite3_compileoption_get(X) ((void*)0) +#endif + +/* +** CAPI3REF: Test To See If The Library Is Threadsafe +** +** ^The sqlite3_threadsafe() function returns zero if and only if +** SQLite was compiled with mutexing code omitted due to the +** [SQLITE_THREADSAFE] compile-time option being set to 0. +** +** SQLite can be compiled with or without mutexes. When +** the [SQLITE_THREADSAFE] C preprocessor macro is 1 or 2, mutexes +** are enabled and SQLite is threadsafe. When the +** [SQLITE_THREADSAFE] macro is 0, +** the mutexes are omitted. Without the mutexes, it is not safe +** to use SQLite concurrently from more than one thread. +** +** Enabling mutexes incurs a measurable performance penalty. +** So if speed is of utmost importance, it makes sense to disable +** the mutexes. But for maximum safety, mutexes should be enabled. +** ^The default behavior is for mutexes to be enabled. +** +** This interface can be used by an application to make sure that the +** version of SQLite that it is linking against was compiled with +** the desired setting of the [SQLITE_THREADSAFE] macro. +** +** This interface only reports on the compile-time mutex setting +** of the [SQLITE_THREADSAFE] flag. If SQLite is compiled with +** SQLITE_THREADSAFE=1 or =2 then mutexes are enabled by default but +** can be fully or partially disabled using a call to [sqlite3_config()] +** with the verbs [SQLITE_CONFIG_SINGLETHREAD], [SQLITE_CONFIG_MULTITHREAD], +** or [SQLITE_CONFIG_SERIALIZED]. ^(The return value of the +** sqlite3_threadsafe() function shows only the compile-time setting of +** thread safety, not any run-time changes to that setting made by +** sqlite3_config(). In other words, the return value from sqlite3_threadsafe() +** is unchanged by calls to sqlite3_config().)^ +** +** See the [threading mode] documentation for additional information. +*/ +SQLITE_API int sqlite3_threadsafe(void); + +/* +** CAPI3REF: Database Connection Handle +** KEYWORDS: {database connection} {database connections} +** +** Each open SQLite database is represented by a pointer to an instance of +** the opaque structure named "sqlite3". It is useful to think of an sqlite3 +** pointer as an object. The [sqlite3_open()], [sqlite3_open16()], and +** [sqlite3_open_v2()] interfaces are its constructors, and [sqlite3_close()] +** and [sqlite3_close_v2()] are its destructors. There are many other +** interfaces (such as +** [sqlite3_prepare_v2()], [sqlite3_create_function()], and +** [sqlite3_busy_timeout()] to name but three) that are methods on an +** sqlite3 object. +*/ +typedef struct sqlite3 sqlite3; + +/* +** CAPI3REF: 64-Bit Integer Types +** KEYWORDS: sqlite_int64 sqlite_uint64 +** +** Because there is no cross-platform way to specify 64-bit integer types +** SQLite includes typedefs for 64-bit signed and unsigned integers. +** +** The sqlite3_int64 and sqlite3_uint64 are the preferred type definitions. +** The sqlite_int64 and sqlite_uint64 types are supported for backwards +** compatibility only. +** +** ^The sqlite3_int64 and sqlite_int64 types can store integer values +** between -9223372036854775808 and +9223372036854775807 inclusive. ^The +** sqlite3_uint64 and sqlite_uint64 types can store integer values +** between 0 and +18446744073709551615 inclusive. +*/ +#ifdef SQLITE_INT64_TYPE + typedef SQLITE_INT64_TYPE sqlite_int64; +# ifdef SQLITE_UINT64_TYPE + typedef SQLITE_UINT64_TYPE sqlite_uint64; +# else + typedef unsigned SQLITE_INT64_TYPE sqlite_uint64; +# endif +#elif defined(_MSC_VER) || defined(__BORLANDC__) + typedef __int64 sqlite_int64; + typedef unsigned __int64 sqlite_uint64; +#else + typedef long long int sqlite_int64; + typedef unsigned long long int sqlite_uint64; +#endif +typedef sqlite_int64 sqlite3_int64; +typedef sqlite_uint64 sqlite3_uint64; + +/* +** If compiling for a processor that lacks floating point support, +** substitute integer for floating-point. +*/ +#ifdef SQLITE_OMIT_FLOATING_POINT +# define double sqlite3_int64 +#endif + +/* +** CAPI3REF: Closing A Database Connection +** DESTRUCTOR: sqlite3 +** +** ^The sqlite3_close() and sqlite3_close_v2() routines are destructors +** for the [sqlite3] object. +** ^Calls to sqlite3_close() and sqlite3_close_v2() return [SQLITE_OK] if +** the [sqlite3] object is successfully destroyed and all associated +** resources are deallocated. +** +** ^If the database connection is associated with unfinalized prepared +** statements or unfinished sqlite3_backup objects then sqlite3_close() +** will leave the database connection open and return [SQLITE_BUSY]. +** ^If sqlite3_close_v2() is called with unfinalized prepared statements +** and/or unfinished sqlite3_backups, then the database connection becomes +** an unusable "zombie" which will automatically be deallocated when the +** last prepared statement is finalized or the last sqlite3_backup is +** finished. The sqlite3_close_v2() interface is intended for use with +** host languages that are garbage collected, and where the order in which +** destructors are called is arbitrary. +** +** Applications should [sqlite3_finalize | finalize] all [prepared statements], +** [sqlite3_blob_close | close] all [BLOB handles], and +** [sqlite3_backup_finish | finish] all [sqlite3_backup] objects associated +** with the [sqlite3] object prior to attempting to close the object. ^If +** sqlite3_close_v2() is called on a [database connection] that still has +** outstanding [prepared statements], [BLOB handles], and/or +** [sqlite3_backup] objects then it returns [SQLITE_OK] and the deallocation +** of resources is deferred until all [prepared statements], [BLOB handles], +** and [sqlite3_backup] objects are also destroyed. +** +** ^If an [sqlite3] object is destroyed while a transaction is open, +** the transaction is automatically rolled back. +** +** The C parameter to [sqlite3_close(C)] and [sqlite3_close_v2(C)] +** must be either a NULL +** pointer or an [sqlite3] object pointer obtained +** from [sqlite3_open()], [sqlite3_open16()], or +** [sqlite3_open_v2()], and not previously closed. +** ^Calling sqlite3_close() or sqlite3_close_v2() with a NULL pointer +** argument is a harmless no-op. +*/ +SQLITE_API int sqlite3_close(sqlite3*); +SQLITE_API int sqlite3_close_v2(sqlite3*); + +/* +** The type for a callback function. +** This is legacy and deprecated. It is included for historical +** compatibility and is not documented. +*/ +typedef int (*sqlite3_callback)(void*,int,char**, char**); + +/* +** CAPI3REF: One-Step Query Execution Interface +** METHOD: sqlite3 +** +** The sqlite3_exec() interface is a convenience wrapper around +** [sqlite3_prepare_v2()], [sqlite3_step()], and [sqlite3_finalize()], +** that allows an application to run multiple statements of SQL +** without having to use a lot of C code. +** +** ^The sqlite3_exec() interface runs zero or more UTF-8 encoded, +** semicolon-separate SQL statements passed into its 2nd argument, +** in the context of the [database connection] passed in as its 1st +** argument. ^If the callback function of the 3rd argument to +** sqlite3_exec() is not NULL, then it is invoked for each result row +** coming out of the evaluated SQL statements. ^The 4th argument to +** sqlite3_exec() is relayed through to the 1st argument of each +** callback invocation. ^If the callback pointer to sqlite3_exec() +** is NULL, then no callback is ever invoked and result rows are +** ignored. +** +** ^If an error occurs while evaluating the SQL statements passed into +** sqlite3_exec(), then execution of the current statement stops and +** subsequent statements are skipped. ^If the 5th parameter to sqlite3_exec() +** is not NULL then any error message is written into memory obtained +** from [sqlite3_malloc()] and passed back through the 5th parameter. +** To avoid memory leaks, the application should invoke [sqlite3_free()] +** on error message strings returned through the 5th parameter of +** sqlite3_exec() after the error message string is no longer needed. +** ^If the 5th parameter to sqlite3_exec() is not NULL and no errors +** occur, then sqlite3_exec() sets the pointer in its 5th parameter to +** NULL before returning. +** +** ^If an sqlite3_exec() callback returns non-zero, the sqlite3_exec() +** routine returns SQLITE_ABORT without invoking the callback again and +** without running any subsequent SQL statements. +** +** ^The 2nd argument to the sqlite3_exec() callback function is the +** number of columns in the result. ^The 3rd argument to the sqlite3_exec() +** callback is an array of pointers to strings obtained as if from +** [sqlite3_column_text()], one for each column. ^If an element of a +** result row is NULL then the corresponding string pointer for the +** sqlite3_exec() callback is a NULL pointer. ^The 4th argument to the +** sqlite3_exec() callback is an array of pointers to strings where each +** entry represents the name of corresponding result column as obtained +** from [sqlite3_column_name()]. +** +** ^If the 2nd parameter to sqlite3_exec() is a NULL pointer, a pointer +** to an empty string, or a pointer that contains only whitespace and/or +** SQL comments, then no SQL statements are evaluated and the database +** is not changed. +** +** Restrictions: +** +**
    +**
  • The application must ensure that the 1st parameter to sqlite3_exec() +** is a valid and open [database connection]. +**
  • The application must not close the [database connection] specified by +** the 1st parameter to sqlite3_exec() while sqlite3_exec() is running. +**
  • The application must not modify the SQL statement text passed into +** the 2nd parameter of sqlite3_exec() while sqlite3_exec() is running. +**
+*/ +SQLITE_API int sqlite3_exec( + sqlite3*, /* An open database */ + const char *sql, /* SQL to be evaluated */ + int (*callback)(void*,int,char**,char**), /* Callback function */ + void *, /* 1st argument to callback */ + char **errmsg /* Error msg written here */ +); + +/* +** CAPI3REF: Result Codes +** KEYWORDS: {result code definitions} +** +** Many SQLite functions return an integer result code from the set shown +** here in order to indicate success or failure. +** +** New error codes may be added in future versions of SQLite. +** +** See also: [extended result code definitions] +*/ +#define SQLITE_OK 0 /* Successful result */ +/* beginning-of-error-codes */ +#define SQLITE_ERROR 1 /* Generic error */ +#define SQLITE_INTERNAL 2 /* Internal logic error in SQLite */ +#define SQLITE_PERM 3 /* Access permission denied */ +#define SQLITE_ABORT 4 /* Callback routine requested an abort */ +#define SQLITE_BUSY 5 /* The database file is locked */ +#define SQLITE_LOCKED 6 /* A table in the database is locked */ +#define SQLITE_NOMEM 7 /* A malloc() failed */ +#define SQLITE_READONLY 8 /* Attempt to write a readonly database */ +#define SQLITE_INTERRUPT 9 /* Operation terminated by sqlite3_interrupt()*/ +#define SQLITE_IOERR 10 /* Some kind of disk I/O error occurred */ +#define SQLITE_CORRUPT 11 /* The database disk image is malformed */ +#define SQLITE_NOTFOUND 12 /* Unknown opcode in sqlite3_file_control() */ +#define SQLITE_FULL 13 /* Insertion failed because database is full */ +#define SQLITE_CANTOPEN 14 /* Unable to open the database file */ +#define SQLITE_PROTOCOL 15 /* Database lock protocol error */ +#define SQLITE_EMPTY 16 /* Internal use only */ +#define SQLITE_SCHEMA 17 /* The database schema changed */ +#define SQLITE_TOOBIG 18 /* String or BLOB exceeds size limit */ +#define SQLITE_CONSTRAINT 19 /* Abort due to constraint violation */ +#define SQLITE_MISMATCH 20 /* Data type mismatch */ +#define SQLITE_MISUSE 21 /* Library used incorrectly */ +#define SQLITE_NOLFS 22 /* Uses OS features not supported on host */ +#define SQLITE_AUTH 23 /* Authorization denied */ +#define SQLITE_FORMAT 24 /* Not used */ +#define SQLITE_RANGE 25 /* 2nd parameter to sqlite3_bind out of range */ +#define SQLITE_NOTADB 26 /* File opened that is not a database file */ +#define SQLITE_NOTICE 27 /* Notifications from sqlite3_log() */ +#define SQLITE_WARNING 28 /* Warnings from sqlite3_log() */ +#define SQLITE_ROW 100 /* sqlite3_step() has another row ready */ +#define SQLITE_DONE 101 /* sqlite3_step() has finished executing */ +/* end-of-error-codes */ + +/* +** CAPI3REF: Extended Result Codes +** KEYWORDS: {extended result code definitions} +** +** In its default configuration, SQLite API routines return one of 30 integer +** [result codes]. However, experience has shown that many of +** these result codes are too coarse-grained. They do not provide as +** much information about problems as programmers might like. In an effort to +** address this, newer versions of SQLite (version 3.3.8 [dateof:3.3.8] +** and later) include +** support for additional result codes that provide more detailed information +** about errors. These [extended result codes] are enabled or disabled +** on a per database connection basis using the +** [sqlite3_extended_result_codes()] API. Or, the extended code for +** the most recent error can be obtained using +** [sqlite3_extended_errcode()]. +*/ +#define SQLITE_ERROR_MISSING_COLLSEQ (SQLITE_ERROR | (1<<8)) +#define SQLITE_ERROR_RETRY (SQLITE_ERROR | (2<<8)) +#define SQLITE_ERROR_SNAPSHOT (SQLITE_ERROR | (3<<8)) +#define SQLITE_IOERR_READ (SQLITE_IOERR | (1<<8)) +#define SQLITE_IOERR_SHORT_READ (SQLITE_IOERR | (2<<8)) +#define SQLITE_IOERR_WRITE (SQLITE_IOERR | (3<<8)) +#define SQLITE_IOERR_FSYNC (SQLITE_IOERR | (4<<8)) +#define SQLITE_IOERR_DIR_FSYNC (SQLITE_IOERR | (5<<8)) +#define SQLITE_IOERR_TRUNCATE (SQLITE_IOERR | (6<<8)) +#define SQLITE_IOERR_FSTAT (SQLITE_IOERR | (7<<8)) +#define SQLITE_IOERR_UNLOCK (SQLITE_IOERR | (8<<8)) +#define SQLITE_IOERR_RDLOCK (SQLITE_IOERR | (9<<8)) +#define SQLITE_IOERR_DELETE (SQLITE_IOERR | (10<<8)) +#define SQLITE_IOERR_BLOCKED (SQLITE_IOERR | (11<<8)) +#define SQLITE_IOERR_NOMEM (SQLITE_IOERR | (12<<8)) +#define SQLITE_IOERR_ACCESS (SQLITE_IOERR | (13<<8)) +#define SQLITE_IOERR_CHECKRESERVEDLOCK (SQLITE_IOERR | (14<<8)) +#define SQLITE_IOERR_LOCK (SQLITE_IOERR | (15<<8)) +#define SQLITE_IOERR_CLOSE (SQLITE_IOERR | (16<<8)) +#define SQLITE_IOERR_DIR_CLOSE (SQLITE_IOERR | (17<<8)) +#define SQLITE_IOERR_SHMOPEN (SQLITE_IOERR | (18<<8)) +#define SQLITE_IOERR_SHMSIZE (SQLITE_IOERR | (19<<8)) +#define SQLITE_IOERR_SHMLOCK (SQLITE_IOERR | (20<<8)) +#define SQLITE_IOERR_SHMMAP (SQLITE_IOERR | (21<<8)) +#define SQLITE_IOERR_SEEK (SQLITE_IOERR | (22<<8)) +#define SQLITE_IOERR_DELETE_NOENT (SQLITE_IOERR | (23<<8)) +#define SQLITE_IOERR_MMAP (SQLITE_IOERR | (24<<8)) +#define SQLITE_IOERR_GETTEMPPATH (SQLITE_IOERR | (25<<8)) +#define SQLITE_IOERR_CONVPATH (SQLITE_IOERR | (26<<8)) +#define SQLITE_IOERR_VNODE (SQLITE_IOERR | (27<<8)) +#define SQLITE_IOERR_AUTH (SQLITE_IOERR | (28<<8)) +#define SQLITE_IOERR_BEGIN_ATOMIC (SQLITE_IOERR | (29<<8)) +#define SQLITE_IOERR_COMMIT_ATOMIC (SQLITE_IOERR | (30<<8)) +#define SQLITE_IOERR_ROLLBACK_ATOMIC (SQLITE_IOERR | (31<<8)) +#define SQLITE_LOCKED_SHAREDCACHE (SQLITE_LOCKED | (1<<8)) +#define SQLITE_LOCKED_VTAB (SQLITE_LOCKED | (2<<8)) +#define SQLITE_BUSY_RECOVERY (SQLITE_BUSY | (1<<8)) +#define SQLITE_BUSY_SNAPSHOT (SQLITE_BUSY | (2<<8)) +#define SQLITE_CANTOPEN_NOTEMPDIR (SQLITE_CANTOPEN | (1<<8)) +#define SQLITE_CANTOPEN_ISDIR (SQLITE_CANTOPEN | (2<<8)) +#define SQLITE_CANTOPEN_FULLPATH (SQLITE_CANTOPEN | (3<<8)) +#define SQLITE_CANTOPEN_CONVPATH (SQLITE_CANTOPEN | (4<<8)) +#define SQLITE_CANTOPEN_DIRTYWAL (SQLITE_CANTOPEN | (5<<8)) /* Not Used */ +#define SQLITE_CORRUPT_VTAB (SQLITE_CORRUPT | (1<<8)) +#define SQLITE_CORRUPT_SEQUENCE (SQLITE_CORRUPT | (2<<8)) +#define SQLITE_READONLY_RECOVERY (SQLITE_READONLY | (1<<8)) +#define SQLITE_READONLY_CANTLOCK (SQLITE_READONLY | (2<<8)) +#define SQLITE_READONLY_ROLLBACK (SQLITE_READONLY | (3<<8)) +#define SQLITE_READONLY_DBMOVED (SQLITE_READONLY | (4<<8)) +#define SQLITE_READONLY_CANTINIT (SQLITE_READONLY | (5<<8)) +#define SQLITE_READONLY_DIRECTORY (SQLITE_READONLY | (6<<8)) +#define SQLITE_ABORT_ROLLBACK (SQLITE_ABORT | (2<<8)) +#define SQLITE_CONSTRAINT_CHECK (SQLITE_CONSTRAINT | (1<<8)) +#define SQLITE_CONSTRAINT_COMMITHOOK (SQLITE_CONSTRAINT | (2<<8)) +#define SQLITE_CONSTRAINT_FOREIGNKEY (SQLITE_CONSTRAINT | (3<<8)) +#define SQLITE_CONSTRAINT_FUNCTION (SQLITE_CONSTRAINT | (4<<8)) +#define SQLITE_CONSTRAINT_NOTNULL (SQLITE_CONSTRAINT | (5<<8)) +#define SQLITE_CONSTRAINT_PRIMARYKEY (SQLITE_CONSTRAINT | (6<<8)) +#define SQLITE_CONSTRAINT_TRIGGER (SQLITE_CONSTRAINT | (7<<8)) +#define SQLITE_CONSTRAINT_UNIQUE (SQLITE_CONSTRAINT | (8<<8)) +#define SQLITE_CONSTRAINT_VTAB (SQLITE_CONSTRAINT | (9<<8)) +#define SQLITE_CONSTRAINT_ROWID (SQLITE_CONSTRAINT |(10<<8)) +#define SQLITE_NOTICE_RECOVER_WAL (SQLITE_NOTICE | (1<<8)) +#define SQLITE_NOTICE_RECOVER_ROLLBACK (SQLITE_NOTICE | (2<<8)) +#define SQLITE_WARNING_AUTOINDEX (SQLITE_WARNING | (1<<8)) +#define SQLITE_AUTH_USER (SQLITE_AUTH | (1<<8)) +#define SQLITE_OK_LOAD_PERMANENTLY (SQLITE_OK | (1<<8)) + +/* +** CAPI3REF: Flags For File Open Operations +** +** These bit values are intended for use in the +** 3rd parameter to the [sqlite3_open_v2()] interface and +** in the 4th parameter to the [sqlite3_vfs.xOpen] method. +*/ +#define SQLITE_OPEN_READONLY 0x00000001 /* Ok for sqlite3_open_v2() */ +#define SQLITE_OPEN_READWRITE 0x00000002 /* Ok for sqlite3_open_v2() */ +#define SQLITE_OPEN_CREATE 0x00000004 /* Ok for sqlite3_open_v2() */ +#define SQLITE_OPEN_DELETEONCLOSE 0x00000008 /* VFS only */ +#define SQLITE_OPEN_EXCLUSIVE 0x00000010 /* VFS only */ +#define SQLITE_OPEN_AUTOPROXY 0x00000020 /* VFS only */ +#define SQLITE_OPEN_URI 0x00000040 /* Ok for sqlite3_open_v2() */ +#define SQLITE_OPEN_MEMORY 0x00000080 /* Ok for sqlite3_open_v2() */ +#define SQLITE_OPEN_MAIN_DB 0x00000100 /* VFS only */ +#define SQLITE_OPEN_TEMP_DB 0x00000200 /* VFS only */ +#define SQLITE_OPEN_TRANSIENT_DB 0x00000400 /* VFS only */ +#define SQLITE_OPEN_MAIN_JOURNAL 0x00000800 /* VFS only */ +#define SQLITE_OPEN_TEMP_JOURNAL 0x00001000 /* VFS only */ +#define SQLITE_OPEN_SUBJOURNAL 0x00002000 /* VFS only */ +#define SQLITE_OPEN_MASTER_JOURNAL 0x00004000 /* VFS only */ +#define SQLITE_OPEN_NOMUTEX 0x00008000 /* Ok for sqlite3_open_v2() */ +#define SQLITE_OPEN_FULLMUTEX 0x00010000 /* Ok for sqlite3_open_v2() */ +#define SQLITE_OPEN_SHAREDCACHE 0x00020000 /* Ok for sqlite3_open_v2() */ +#define SQLITE_OPEN_PRIVATECACHE 0x00040000 /* Ok for sqlite3_open_v2() */ +#define SQLITE_OPEN_WAL 0x00080000 /* VFS only */ + +/* Reserved: 0x00F00000 */ + +/* +** CAPI3REF: Device Characteristics +** +** The xDeviceCharacteristics method of the [sqlite3_io_methods] +** object returns an integer which is a vector of these +** bit values expressing I/O characteristics of the mass storage +** device that holds the file that the [sqlite3_io_methods] +** refers to. +** +** The SQLITE_IOCAP_ATOMIC property means that all writes of +** any size are atomic. The SQLITE_IOCAP_ATOMICnnn values +** mean that writes of blocks that are nnn bytes in size and +** are aligned to an address which is an integer multiple of +** nnn are atomic. The SQLITE_IOCAP_SAFE_APPEND value means +** that when data is appended to a file, the data is appended +** first then the size of the file is extended, never the other +** way around. The SQLITE_IOCAP_SEQUENTIAL property means that +** information is written to disk in the same order as calls +** to xWrite(). The SQLITE_IOCAP_POWERSAFE_OVERWRITE property means that +** after reboot following a crash or power loss, the only bytes in a +** file that were written at the application level might have changed +** and that adjacent bytes, even bytes within the same sector are +** guaranteed to be unchanged. The SQLITE_IOCAP_UNDELETABLE_WHEN_OPEN +** flag indicates that a file cannot be deleted when open. The +** SQLITE_IOCAP_IMMUTABLE flag indicates that the file is on +** read-only media and cannot be changed even by processes with +** elevated privileges. +** +** The SQLITE_IOCAP_BATCH_ATOMIC property means that the underlying +** filesystem supports doing multiple write operations atomically when those +** write operations are bracketed by [SQLITE_FCNTL_BEGIN_ATOMIC_WRITE] and +** [SQLITE_FCNTL_COMMIT_ATOMIC_WRITE]. +*/ +#define SQLITE_IOCAP_ATOMIC 0x00000001 +#define SQLITE_IOCAP_ATOMIC512 0x00000002 +#define SQLITE_IOCAP_ATOMIC1K 0x00000004 +#define SQLITE_IOCAP_ATOMIC2K 0x00000008 +#define SQLITE_IOCAP_ATOMIC4K 0x00000010 +#define SQLITE_IOCAP_ATOMIC8K 0x00000020 +#define SQLITE_IOCAP_ATOMIC16K 0x00000040 +#define SQLITE_IOCAP_ATOMIC32K 0x00000080 +#define SQLITE_IOCAP_ATOMIC64K 0x00000100 +#define SQLITE_IOCAP_SAFE_APPEND 0x00000200 +#define SQLITE_IOCAP_SEQUENTIAL 0x00000400 +#define SQLITE_IOCAP_UNDELETABLE_WHEN_OPEN 0x00000800 +#define SQLITE_IOCAP_POWERSAFE_OVERWRITE 0x00001000 +#define SQLITE_IOCAP_IMMUTABLE 0x00002000 +#define SQLITE_IOCAP_BATCH_ATOMIC 0x00004000 + +/* +** CAPI3REF: File Locking Levels +** +** SQLite uses one of these integer values as the second +** argument to calls it makes to the xLock() and xUnlock() methods +** of an [sqlite3_io_methods] object. +*/ +#define SQLITE_LOCK_NONE 0 +#define SQLITE_LOCK_SHARED 1 +#define SQLITE_LOCK_RESERVED 2 +#define SQLITE_LOCK_PENDING 3 +#define SQLITE_LOCK_EXCLUSIVE 4 + +/* +** CAPI3REF: Synchronization Type Flags +** +** When SQLite invokes the xSync() method of an +** [sqlite3_io_methods] object it uses a combination of +** these integer values as the second argument. +** +** When the SQLITE_SYNC_DATAONLY flag is used, it means that the +** sync operation only needs to flush data to mass storage. Inode +** information need not be flushed. If the lower four bits of the flag +** equal SQLITE_SYNC_NORMAL, that means to use normal fsync() semantics. +** If the lower four bits equal SQLITE_SYNC_FULL, that means +** to use Mac OS X style fullsync instead of fsync(). +** +** Do not confuse the SQLITE_SYNC_NORMAL and SQLITE_SYNC_FULL flags +** with the [PRAGMA synchronous]=NORMAL and [PRAGMA synchronous]=FULL +** settings. The [synchronous pragma] determines when calls to the +** xSync VFS method occur and applies uniformly across all platforms. +** The SQLITE_SYNC_NORMAL and SQLITE_SYNC_FULL flags determine how +** energetic or rigorous or forceful the sync operations are and +** only make a difference on Mac OSX for the default SQLite code. +** (Third-party VFS implementations might also make the distinction +** between SQLITE_SYNC_NORMAL and SQLITE_SYNC_FULL, but among the +** operating systems natively supported by SQLite, only Mac OSX +** cares about the difference.) +*/ +#define SQLITE_SYNC_NORMAL 0x00002 +#define SQLITE_SYNC_FULL 0x00003 +#define SQLITE_SYNC_DATAONLY 0x00010 + +/* +** CAPI3REF: OS Interface Open File Handle +** +** An [sqlite3_file] object represents an open file in the +** [sqlite3_vfs | OS interface layer]. Individual OS interface +** implementations will +** want to subclass this object by appending additional fields +** for their own use. The pMethods entry is a pointer to an +** [sqlite3_io_methods] object that defines methods for performing +** I/O operations on the open file. +*/ +typedef struct sqlite3_file sqlite3_file; +struct sqlite3_file { + const struct sqlite3_io_methods *pMethods; /* Methods for an open file */ +}; + +/* +** CAPI3REF: OS Interface File Virtual Methods Object +** +** Every file opened by the [sqlite3_vfs.xOpen] method populates an +** [sqlite3_file] object (or, more commonly, a subclass of the +** [sqlite3_file] object) with a pointer to an instance of this object. +** This object defines the methods used to perform various operations +** against the open file represented by the [sqlite3_file] object. +** +** If the [sqlite3_vfs.xOpen] method sets the sqlite3_file.pMethods element +** to a non-NULL pointer, then the sqlite3_io_methods.xClose method +** may be invoked even if the [sqlite3_vfs.xOpen] reported that it failed. The +** only way to prevent a call to xClose following a failed [sqlite3_vfs.xOpen] +** is for the [sqlite3_vfs.xOpen] to set the sqlite3_file.pMethods element +** to NULL. +** +** The flags argument to xSync may be one of [SQLITE_SYNC_NORMAL] or +** [SQLITE_SYNC_FULL]. The first choice is the normal fsync(). +** The second choice is a Mac OS X style fullsync. The [SQLITE_SYNC_DATAONLY] +** flag may be ORed in to indicate that only the data of the file +** and not its inode needs to be synced. +** +** The integer values to xLock() and xUnlock() are one of +**
    +**
  • [SQLITE_LOCK_NONE], +**
  • [SQLITE_LOCK_SHARED], +**
  • [SQLITE_LOCK_RESERVED], +**
  • [SQLITE_LOCK_PENDING], or +**
  • [SQLITE_LOCK_EXCLUSIVE]. +**
+** xLock() increases the lock. xUnlock() decreases the lock. +** The xCheckReservedLock() method checks whether any database connection, +** either in this process or in some other process, is holding a RESERVED, +** PENDING, or EXCLUSIVE lock on the file. It returns true +** if such a lock exists and false otherwise. +** +** The xFileControl() method is a generic interface that allows custom +** VFS implementations to directly control an open file using the +** [sqlite3_file_control()] interface. The second "op" argument is an +** integer opcode. The third argument is a generic pointer intended to +** point to a structure that may contain arguments or space in which to +** write return values. Potential uses for xFileControl() might be +** functions to enable blocking locks with timeouts, to change the +** locking strategy (for example to use dot-file locks), to inquire +** about the status of a lock, or to break stale locks. The SQLite +** core reserves all opcodes less than 100 for its own use. +** A [file control opcodes | list of opcodes] less than 100 is available. +** Applications that define a custom xFileControl method should use opcodes +** greater than 100 to avoid conflicts. VFS implementations should +** return [SQLITE_NOTFOUND] for file control opcodes that they do not +** recognize. +** +** The xSectorSize() method returns the sector size of the +** device that underlies the file. The sector size is the +** minimum write that can be performed without disturbing +** other bytes in the file. The xDeviceCharacteristics() +** method returns a bit vector describing behaviors of the +** underlying device: +** +**
    +**
  • [SQLITE_IOCAP_ATOMIC] +**
  • [SQLITE_IOCAP_ATOMIC512] +**
  • [SQLITE_IOCAP_ATOMIC1K] +**
  • [SQLITE_IOCAP_ATOMIC2K] +**
  • [SQLITE_IOCAP_ATOMIC4K] +**
  • [SQLITE_IOCAP_ATOMIC8K] +**
  • [SQLITE_IOCAP_ATOMIC16K] +**
  • [SQLITE_IOCAP_ATOMIC32K] +**
  • [SQLITE_IOCAP_ATOMIC64K] +**
  • [SQLITE_IOCAP_SAFE_APPEND] +**
  • [SQLITE_IOCAP_SEQUENTIAL] +**
  • [SQLITE_IOCAP_UNDELETABLE_WHEN_OPEN] +**
  • [SQLITE_IOCAP_POWERSAFE_OVERWRITE] +**
  • [SQLITE_IOCAP_IMMUTABLE] +**
  • [SQLITE_IOCAP_BATCH_ATOMIC] +**
+** +** The SQLITE_IOCAP_ATOMIC property means that all writes of +** any size are atomic. The SQLITE_IOCAP_ATOMICnnn values +** mean that writes of blocks that are nnn bytes in size and +** are aligned to an address which is an integer multiple of +** nnn are atomic. The SQLITE_IOCAP_SAFE_APPEND value means +** that when data is appended to a file, the data is appended +** first then the size of the file is extended, never the other +** way around. The SQLITE_IOCAP_SEQUENTIAL property means that +** information is written to disk in the same order as calls +** to xWrite(). +** +** If xRead() returns SQLITE_IOERR_SHORT_READ it must also fill +** in the unread portions of the buffer with zeros. A VFS that +** fails to zero-fill short reads might seem to work. However, +** failure to zero-fill short reads will eventually lead to +** database corruption. +*/ +typedef struct sqlite3_io_methods sqlite3_io_methods; +struct sqlite3_io_methods { + int iVersion; + int (*xClose)(sqlite3_file*); + int (*xRead)(sqlite3_file*, void*, int iAmt, sqlite3_int64 iOfst); + int (*xWrite)(sqlite3_file*, const void*, int iAmt, sqlite3_int64 iOfst); + int (*xTruncate)(sqlite3_file*, sqlite3_int64 size); + int (*xSync)(sqlite3_file*, int flags); + int (*xFileSize)(sqlite3_file*, sqlite3_int64 *pSize); + int (*xLock)(sqlite3_file*, int); + int (*xUnlock)(sqlite3_file*, int); + int (*xCheckReservedLock)(sqlite3_file*, int *pResOut); + int (*xFileControl)(sqlite3_file*, int op, void *pArg); + int (*xSectorSize)(sqlite3_file*); + int (*xDeviceCharacteristics)(sqlite3_file*); + /* Methods above are valid for version 1 */ + int (*xShmMap)(sqlite3_file*, int iPg, int pgsz, int, void volatile**); + int (*xShmLock)(sqlite3_file*, int offset, int n, int flags); + void (*xShmBarrier)(sqlite3_file*); + int (*xShmUnmap)(sqlite3_file*, int deleteFlag); + /* Methods above are valid for version 2 */ + int (*xFetch)(sqlite3_file*, sqlite3_int64 iOfst, int iAmt, void **pp); + int (*xUnfetch)(sqlite3_file*, sqlite3_int64 iOfst, void *p); + /* Methods above are valid for version 3 */ + /* Additional methods may be added in future releases */ +}; + +/* +** CAPI3REF: Standard File Control Opcodes +** KEYWORDS: {file control opcodes} {file control opcode} +** +** These integer constants are opcodes for the xFileControl method +** of the [sqlite3_io_methods] object and for the [sqlite3_file_control()] +** interface. +** +**
    +**
  • [[SQLITE_FCNTL_LOCKSTATE]] +** The [SQLITE_FCNTL_LOCKSTATE] opcode is used for debugging. This +** opcode causes the xFileControl method to write the current state of +** the lock (one of [SQLITE_LOCK_NONE], [SQLITE_LOCK_SHARED], +** [SQLITE_LOCK_RESERVED], [SQLITE_LOCK_PENDING], or [SQLITE_LOCK_EXCLUSIVE]) +** into an integer that the pArg argument points to. This capability +** is used during testing and is only available when the SQLITE_TEST +** compile-time option is used. +** +**
  • [[SQLITE_FCNTL_SIZE_HINT]] +** The [SQLITE_FCNTL_SIZE_HINT] opcode is used by SQLite to give the VFS +** layer a hint of how large the database file will grow to be during the +** current transaction. This hint is not guaranteed to be accurate but it +** is often close. The underlying VFS might choose to preallocate database +** file space based on this hint in order to help writes to the database +** file run faster. +** +**
  • [[SQLITE_FCNTL_SIZE_LIMIT]] +** The [SQLITE_FCNTL_SIZE_LIMIT] opcode is used by in-memory VFS that +** implements [sqlite3_deserialize()] to set an upper bound on the size +** of the in-memory database. The argument is a pointer to a [sqlite3_int64]. +** If the integer pointed to is negative, then it is filled in with the +** current limit. Otherwise the limit is set to the larger of the value +** of the integer pointed to and the current database size. The integer +** pointed to is set to the new limit. +** +**
  • [[SQLITE_FCNTL_CHUNK_SIZE]] +** The [SQLITE_FCNTL_CHUNK_SIZE] opcode is used to request that the VFS +** extends and truncates the database file in chunks of a size specified +** by the user. The fourth argument to [sqlite3_file_control()] should +** point to an integer (type int) containing the new chunk-size to use +** for the nominated database. Allocating database file space in large +** chunks (say 1MB at a time), may reduce file-system fragmentation and +** improve performance on some systems. +** +**
  • [[SQLITE_FCNTL_FILE_POINTER]] +** The [SQLITE_FCNTL_FILE_POINTER] opcode is used to obtain a pointer +** to the [sqlite3_file] object associated with a particular database +** connection. See also [SQLITE_FCNTL_JOURNAL_POINTER]. +** +**
  • [[SQLITE_FCNTL_JOURNAL_POINTER]] +** The [SQLITE_FCNTL_JOURNAL_POINTER] opcode is used to obtain a pointer +** to the [sqlite3_file] object associated with the journal file (either +** the [rollback journal] or the [write-ahead log]) for a particular database +** connection. See also [SQLITE_FCNTL_FILE_POINTER]. +** +**
  • [[SQLITE_FCNTL_SYNC_OMITTED]] +** No longer in use. +** +**
  • [[SQLITE_FCNTL_SYNC]] +** The [SQLITE_FCNTL_SYNC] opcode is generated internally by SQLite and +** sent to the VFS immediately before the xSync method is invoked on a +** database file descriptor. Or, if the xSync method is not invoked +** because the user has configured SQLite with +** [PRAGMA synchronous | PRAGMA synchronous=OFF] it is invoked in place +** of the xSync method. In most cases, the pointer argument passed with +** this file-control is NULL. However, if the database file is being synced +** as part of a multi-database commit, the argument points to a nul-terminated +** string containing the transactions master-journal file name. VFSes that +** do not need this signal should silently ignore this opcode. Applications +** should not call [sqlite3_file_control()] with this opcode as doing so may +** disrupt the operation of the specialized VFSes that do require it. +** +**
  • [[SQLITE_FCNTL_COMMIT_PHASETWO]] +** The [SQLITE_FCNTL_COMMIT_PHASETWO] opcode is generated internally by SQLite +** and sent to the VFS after a transaction has been committed immediately +** but before the database is unlocked. VFSes that do not need this signal +** should silently ignore this opcode. Applications should not call +** [sqlite3_file_control()] with this opcode as doing so may disrupt the +** operation of the specialized VFSes that do require it. +** +**
  • [[SQLITE_FCNTL_WIN32_AV_RETRY]] +** ^The [SQLITE_FCNTL_WIN32_AV_RETRY] opcode is used to configure automatic +** retry counts and intervals for certain disk I/O operations for the +** windows [VFS] in order to provide robustness in the presence of +** anti-virus programs. By default, the windows VFS will retry file read, +** file write, and file delete operations up to 10 times, with a delay +** of 25 milliseconds before the first retry and with the delay increasing +** by an additional 25 milliseconds with each subsequent retry. This +** opcode allows these two values (10 retries and 25 milliseconds of delay) +** to be adjusted. The values are changed for all database connections +** within the same process. The argument is a pointer to an array of two +** integers where the first integer is the new retry count and the second +** integer is the delay. If either integer is negative, then the setting +** is not changed but instead the prior value of that setting is written +** into the array entry, allowing the current retry settings to be +** interrogated. The zDbName parameter is ignored. +** +**
  • [[SQLITE_FCNTL_PERSIST_WAL]] +** ^The [SQLITE_FCNTL_PERSIST_WAL] opcode is used to set or query the +** persistent [WAL | Write Ahead Log] setting. By default, the auxiliary +** write ahead log ([WAL file]) and shared memory +** files used for transaction control +** are automatically deleted when the latest connection to the database +** closes. Setting persistent WAL mode causes those files to persist after +** close. Persisting the files is useful when other processes that do not +** have write permission on the directory containing the database file want +** to read the database file, as the WAL and shared memory files must exist +** in order for the database to be readable. The fourth parameter to +** [sqlite3_file_control()] for this opcode should be a pointer to an integer. +** That integer is 0 to disable persistent WAL mode or 1 to enable persistent +** WAL mode. If the integer is -1, then it is overwritten with the current +** WAL persistence setting. +** +**
  • [[SQLITE_FCNTL_POWERSAFE_OVERWRITE]] +** ^The [SQLITE_FCNTL_POWERSAFE_OVERWRITE] opcode is used to set or query the +** persistent "powersafe-overwrite" or "PSOW" setting. The PSOW setting +** determines the [SQLITE_IOCAP_POWERSAFE_OVERWRITE] bit of the +** xDeviceCharacteristics methods. The fourth parameter to +** [sqlite3_file_control()] for this opcode should be a pointer to an integer. +** That integer is 0 to disable zero-damage mode or 1 to enable zero-damage +** mode. If the integer is -1, then it is overwritten with the current +** zero-damage mode setting. +** +**
  • [[SQLITE_FCNTL_OVERWRITE]] +** ^The [SQLITE_FCNTL_OVERWRITE] opcode is invoked by SQLite after opening +** a write transaction to indicate that, unless it is rolled back for some +** reason, the entire database file will be overwritten by the current +** transaction. This is used by VACUUM operations. +** +**
  • [[SQLITE_FCNTL_VFSNAME]] +** ^The [SQLITE_FCNTL_VFSNAME] opcode can be used to obtain the names of +** all [VFSes] in the VFS stack. The names are of all VFS shims and the +** final bottom-level VFS are written into memory obtained from +** [sqlite3_malloc()] and the result is stored in the char* variable +** that the fourth parameter of [sqlite3_file_control()] points to. +** The caller is responsible for freeing the memory when done. As with +** all file-control actions, there is no guarantee that this will actually +** do anything. Callers should initialize the char* variable to a NULL +** pointer in case this file-control is not implemented. This file-control +** is intended for diagnostic use only. +** +**
  • [[SQLITE_FCNTL_VFS_POINTER]] +** ^The [SQLITE_FCNTL_VFS_POINTER] opcode finds a pointer to the top-level +** [VFSes] currently in use. ^(The argument X in +** sqlite3_file_control(db,SQLITE_FCNTL_VFS_POINTER,X) must be +** of type "[sqlite3_vfs] **". This opcodes will set *X +** to a pointer to the top-level VFS.)^ +** ^When there are multiple VFS shims in the stack, this opcode finds the +** upper-most shim only. +** +**
  • [[SQLITE_FCNTL_PRAGMA]] +** ^Whenever a [PRAGMA] statement is parsed, an [SQLITE_FCNTL_PRAGMA] +** file control is sent to the open [sqlite3_file] object corresponding +** to the database file to which the pragma statement refers. ^The argument +** to the [SQLITE_FCNTL_PRAGMA] file control is an array of +** pointers to strings (char**) in which the second element of the array +** is the name of the pragma and the third element is the argument to the +** pragma or NULL if the pragma has no argument. ^The handler for an +** [SQLITE_FCNTL_PRAGMA] file control can optionally make the first element +** of the char** argument point to a string obtained from [sqlite3_mprintf()] +** or the equivalent and that string will become the result of the pragma or +** the error message if the pragma fails. ^If the +** [SQLITE_FCNTL_PRAGMA] file control returns [SQLITE_NOTFOUND], then normal +** [PRAGMA] processing continues. ^If the [SQLITE_FCNTL_PRAGMA] +** file control returns [SQLITE_OK], then the parser assumes that the +** VFS has handled the PRAGMA itself and the parser generates a no-op +** prepared statement if result string is NULL, or that returns a copy +** of the result string if the string is non-NULL. +** ^If the [SQLITE_FCNTL_PRAGMA] file control returns +** any result code other than [SQLITE_OK] or [SQLITE_NOTFOUND], that means +** that the VFS encountered an error while handling the [PRAGMA] and the +** compilation of the PRAGMA fails with an error. ^The [SQLITE_FCNTL_PRAGMA] +** file control occurs at the beginning of pragma statement analysis and so +** it is able to override built-in [PRAGMA] statements. +** +**
  • [[SQLITE_FCNTL_BUSYHANDLER]] +** ^The [SQLITE_FCNTL_BUSYHANDLER] +** file-control may be invoked by SQLite on the database file handle +** shortly after it is opened in order to provide a custom VFS with access +** to the connections busy-handler callback. The argument is of type (void **) +** - an array of two (void *) values. The first (void *) actually points +** to a function of type (int (*)(void *)). In order to invoke the connections +** busy-handler, this function should be invoked with the second (void *) in +** the array as the only argument. If it returns non-zero, then the operation +** should be retried. If it returns zero, the custom VFS should abandon the +** current operation. +** +**
  • [[SQLITE_FCNTL_TEMPFILENAME]] +** ^Application can invoke the [SQLITE_FCNTL_TEMPFILENAME] file-control +** to have SQLite generate a +** temporary filename using the same algorithm that is followed to generate +** temporary filenames for TEMP tables and other internal uses. The +** argument should be a char** which will be filled with the filename +** written into memory obtained from [sqlite3_malloc()]. The caller should +** invoke [sqlite3_free()] on the result to avoid a memory leak. +** +**
  • [[SQLITE_FCNTL_MMAP_SIZE]] +** The [SQLITE_FCNTL_MMAP_SIZE] file control is used to query or set the +** maximum number of bytes that will be used for memory-mapped I/O. +** The argument is a pointer to a value of type sqlite3_int64 that +** is an advisory maximum number of bytes in the file to memory map. The +** pointer is overwritten with the old value. The limit is not changed if +** the value originally pointed to is negative, and so the current limit +** can be queried by passing in a pointer to a negative number. This +** file-control is used internally to implement [PRAGMA mmap_size]. +** +**
  • [[SQLITE_FCNTL_TRACE]] +** The [SQLITE_FCNTL_TRACE] file control provides advisory information +** to the VFS about what the higher layers of the SQLite stack are doing. +** This file control is used by some VFS activity tracing [shims]. +** The argument is a zero-terminated string. Higher layers in the +** SQLite stack may generate instances of this file control if +** the [SQLITE_USE_FCNTL_TRACE] compile-time option is enabled. +** +**
  • [[SQLITE_FCNTL_HAS_MOVED]] +** The [SQLITE_FCNTL_HAS_MOVED] file control interprets its argument as a +** pointer to an integer and it writes a boolean into that integer depending +** on whether or not the file has been renamed, moved, or deleted since it +** was first opened. +** +**
  • [[SQLITE_FCNTL_WIN32_GET_HANDLE]] +** The [SQLITE_FCNTL_WIN32_GET_HANDLE] opcode can be used to obtain the +** underlying native file handle associated with a file handle. This file +** control interprets its argument as a pointer to a native file handle and +** writes the resulting value there. +** +**
  • [[SQLITE_FCNTL_WIN32_SET_HANDLE]] +** The [SQLITE_FCNTL_WIN32_SET_HANDLE] opcode is used for debugging. This +** opcode causes the xFileControl method to swap the file handle with the one +** pointed to by the pArg argument. This capability is used during testing +** and only needs to be supported when SQLITE_TEST is defined. +** +**
  • [[SQLITE_FCNTL_WAL_BLOCK]] +** The [SQLITE_FCNTL_WAL_BLOCK] is a signal to the VFS layer that it might +** be advantageous to block on the next WAL lock if the lock is not immediately +** available. The WAL subsystem issues this signal during rare +** circumstances in order to fix a problem with priority inversion. +** Applications should not use this file-control. +** +**
  • [[SQLITE_FCNTL_ZIPVFS]] +** The [SQLITE_FCNTL_ZIPVFS] opcode is implemented by zipvfs only. All other +** VFS should return SQLITE_NOTFOUND for this opcode. +** +**
  • [[SQLITE_FCNTL_RBU]] +** The [SQLITE_FCNTL_RBU] opcode is implemented by the special VFS used by +** the RBU extension only. All other VFS should return SQLITE_NOTFOUND for +** this opcode. +** +**
  • [[SQLITE_FCNTL_BEGIN_ATOMIC_WRITE]] +** If the [SQLITE_FCNTL_BEGIN_ATOMIC_WRITE] opcode returns SQLITE_OK, then +** the file descriptor is placed in "batch write mode", which +** means all subsequent write operations will be deferred and done +** atomically at the next [SQLITE_FCNTL_COMMIT_ATOMIC_WRITE]. Systems +** that do not support batch atomic writes will return SQLITE_NOTFOUND. +** ^Following a successful SQLITE_FCNTL_BEGIN_ATOMIC_WRITE and prior to +** the closing [SQLITE_FCNTL_COMMIT_ATOMIC_WRITE] or +** [SQLITE_FCNTL_ROLLBACK_ATOMIC_WRITE], SQLite will make +** no VFS interface calls on the same [sqlite3_file] file descriptor +** except for calls to the xWrite method and the xFileControl method +** with [SQLITE_FCNTL_SIZE_HINT]. +** +**
  • [[SQLITE_FCNTL_COMMIT_ATOMIC_WRITE]] +** The [SQLITE_FCNTL_COMMIT_ATOMIC_WRITE] opcode causes all write +** operations since the previous successful call to +** [SQLITE_FCNTL_BEGIN_ATOMIC_WRITE] to be performed atomically. +** This file control returns [SQLITE_OK] if and only if the writes were +** all performed successfully and have been committed to persistent storage. +** ^Regardless of whether or not it is successful, this file control takes +** the file descriptor out of batch write mode so that all subsequent +** write operations are independent. +** ^SQLite will never invoke SQLITE_FCNTL_COMMIT_ATOMIC_WRITE without +** a prior successful call to [SQLITE_FCNTL_BEGIN_ATOMIC_WRITE]. +** +**
  • [[SQLITE_FCNTL_ROLLBACK_ATOMIC_WRITE]] +** The [SQLITE_FCNTL_ROLLBACK_ATOMIC_WRITE] opcode causes all write +** operations since the previous successful call to +** [SQLITE_FCNTL_BEGIN_ATOMIC_WRITE] to be rolled back. +** ^This file control takes the file descriptor out of batch write mode +** so that all subsequent write operations are independent. +** ^SQLite will never invoke SQLITE_FCNTL_ROLLBACK_ATOMIC_WRITE without +** a prior successful call to [SQLITE_FCNTL_BEGIN_ATOMIC_WRITE]. +** +**
  • [[SQLITE_FCNTL_LOCK_TIMEOUT]] +** The [SQLITE_FCNTL_LOCK_TIMEOUT] opcode causes attempts to obtain +** a file lock using the xLock or xShmLock methods of the VFS to wait +** for up to M milliseconds before failing, where M is the single +** unsigned integer parameter. +** +**
  • [[SQLITE_FCNTL_DATA_VERSION]] +** The [SQLITE_FCNTL_DATA_VERSION] opcode is used to detect changes to +** a database file. The argument is a pointer to a 32-bit unsigned integer. +** The "data version" for the pager is written into the pointer. The +** "data version" changes whenever any change occurs to the corresponding +** database file, either through SQL statements on the same database +** connection or through transactions committed by separate database +** connections possibly in other processes. The [sqlite3_total_changes()] +** interface can be used to find if any database on the connection has changed, +** but that interface responds to changes on TEMP as well as MAIN and does +** not provide a mechanism to detect changes to MAIN only. Also, the +** [sqlite3_total_changes()] interface responds to internal changes only and +** omits changes made by other database connections. The +** [PRAGMA data_version] command provide a mechanism to detect changes to +** a single attached database that occur due to other database connections, +** but omits changes implemented by the database connection on which it is +** called. This file control is the only mechanism to detect changes that +** happen either internally or externally and that are associated with +** a particular attached database. +**
+*/ +#define SQLITE_FCNTL_LOCKSTATE 1 +#define SQLITE_FCNTL_GET_LOCKPROXYFILE 2 +#define SQLITE_FCNTL_SET_LOCKPROXYFILE 3 +#define SQLITE_FCNTL_LAST_ERRNO 4 +#define SQLITE_FCNTL_SIZE_HINT 5 +#define SQLITE_FCNTL_CHUNK_SIZE 6 +#define SQLITE_FCNTL_FILE_POINTER 7 +#define SQLITE_FCNTL_SYNC_OMITTED 8 +#define SQLITE_FCNTL_WIN32_AV_RETRY 9 +#define SQLITE_FCNTL_PERSIST_WAL 10 +#define SQLITE_FCNTL_OVERWRITE 11 +#define SQLITE_FCNTL_VFSNAME 12 +#define SQLITE_FCNTL_POWERSAFE_OVERWRITE 13 +#define SQLITE_FCNTL_PRAGMA 14 +#define SQLITE_FCNTL_BUSYHANDLER 15 +#define SQLITE_FCNTL_TEMPFILENAME 16 +#define SQLITE_FCNTL_MMAP_SIZE 18 +#define SQLITE_FCNTL_TRACE 19 +#define SQLITE_FCNTL_HAS_MOVED 20 +#define SQLITE_FCNTL_SYNC 21 +#define SQLITE_FCNTL_COMMIT_PHASETWO 22 +#define SQLITE_FCNTL_WIN32_SET_HANDLE 23 +#define SQLITE_FCNTL_WAL_BLOCK 24 +#define SQLITE_FCNTL_ZIPVFS 25 +#define SQLITE_FCNTL_RBU 26 +#define SQLITE_FCNTL_VFS_POINTER 27 +#define SQLITE_FCNTL_JOURNAL_POINTER 28 +#define SQLITE_FCNTL_WIN32_GET_HANDLE 29 +#define SQLITE_FCNTL_PDB 30 +#define SQLITE_FCNTL_BEGIN_ATOMIC_WRITE 31 +#define SQLITE_FCNTL_COMMIT_ATOMIC_WRITE 32 +#define SQLITE_FCNTL_ROLLBACK_ATOMIC_WRITE 33 +#define SQLITE_FCNTL_LOCK_TIMEOUT 34 +#define SQLITE_FCNTL_DATA_VERSION 35 +#define SQLITE_FCNTL_SIZE_LIMIT 36 + +/* deprecated names */ +#define SQLITE_GET_LOCKPROXYFILE SQLITE_FCNTL_GET_LOCKPROXYFILE +#define SQLITE_SET_LOCKPROXYFILE SQLITE_FCNTL_SET_LOCKPROXYFILE +#define SQLITE_LAST_ERRNO SQLITE_FCNTL_LAST_ERRNO + + +/* +** CAPI3REF: Mutex Handle +** +** The mutex module within SQLite defines [sqlite3_mutex] to be an +** abstract type for a mutex object. The SQLite core never looks +** at the internal representation of an [sqlite3_mutex]. It only +** deals with pointers to the [sqlite3_mutex] object. +** +** Mutexes are created using [sqlite3_mutex_alloc()]. +*/ +typedef struct sqlite3_mutex sqlite3_mutex; + +/* +** CAPI3REF: Loadable Extension Thunk +** +** A pointer to the opaque sqlite3_api_routines structure is passed as +** the third parameter to entry points of [loadable extensions]. This +** structure must be typedefed in order to work around compiler warnings +** on some platforms. +*/ +typedef struct sqlite3_api_routines sqlite3_api_routines; + +/* +** CAPI3REF: OS Interface Object +** +** An instance of the sqlite3_vfs object defines the interface between +** the SQLite core and the underlying operating system. The "vfs" +** in the name of the object stands for "virtual file system". See +** the [VFS | VFS documentation] for further information. +** +** The VFS interface is sometimes extended by adding new methods onto +** the end. Each time such an extension occurs, the iVersion field +** is incremented. The iVersion value started out as 1 in +** SQLite [version 3.5.0] on [dateof:3.5.0], then increased to 2 +** with SQLite [version 3.7.0] on [dateof:3.7.0], and then increased +** to 3 with SQLite [version 3.7.6] on [dateof:3.7.6]. Additional fields +** may be appended to the sqlite3_vfs object and the iVersion value +** may increase again in future versions of SQLite. +** Note that the structure +** of the sqlite3_vfs object changes in the transition from +** SQLite [version 3.5.9] to [version 3.6.0] on [dateof:3.6.0] +** and yet the iVersion field was not modified. +** +** The szOsFile field is the size of the subclassed [sqlite3_file] +** structure used by this VFS. mxPathname is the maximum length of +** a pathname in this VFS. +** +** Registered sqlite3_vfs objects are kept on a linked list formed by +** the pNext pointer. The [sqlite3_vfs_register()] +** and [sqlite3_vfs_unregister()] interfaces manage this list +** in a thread-safe way. The [sqlite3_vfs_find()] interface +** searches the list. Neither the application code nor the VFS +** implementation should use the pNext pointer. +** +** The pNext field is the only field in the sqlite3_vfs +** structure that SQLite will ever modify. SQLite will only access +** or modify this field while holding a particular static mutex. +** The application should never modify anything within the sqlite3_vfs +** object once the object has been registered. +** +** The zName field holds the name of the VFS module. The name must +** be unique across all VFS modules. +** +** [[sqlite3_vfs.xOpen]] +** ^SQLite guarantees that the zFilename parameter to xOpen +** is either a NULL pointer or string obtained +** from xFullPathname() with an optional suffix added. +** ^If a suffix is added to the zFilename parameter, it will +** consist of a single "-" character followed by no more than +** 11 alphanumeric and/or "-" characters. +** ^SQLite further guarantees that +** the string will be valid and unchanged until xClose() is +** called. Because of the previous sentence, +** the [sqlite3_file] can safely store a pointer to the +** filename if it needs to remember the filename for some reason. +** If the zFilename parameter to xOpen is a NULL pointer then xOpen +** must invent its own temporary name for the file. ^Whenever the +** xFilename parameter is NULL it will also be the case that the +** flags parameter will include [SQLITE_OPEN_DELETEONCLOSE]. +** +** The flags argument to xOpen() includes all bits set in +** the flags argument to [sqlite3_open_v2()]. Or if [sqlite3_open()] +** or [sqlite3_open16()] is used, then flags includes at least +** [SQLITE_OPEN_READWRITE] | [SQLITE_OPEN_CREATE]. +** If xOpen() opens a file read-only then it sets *pOutFlags to +** include [SQLITE_OPEN_READONLY]. Other bits in *pOutFlags may be set. +** +** ^(SQLite will also add one of the following flags to the xOpen() +** call, depending on the object being opened: +** +**
    +**
  • [SQLITE_OPEN_MAIN_DB] +**
  • [SQLITE_OPEN_MAIN_JOURNAL] +**
  • [SQLITE_OPEN_TEMP_DB] +**
  • [SQLITE_OPEN_TEMP_JOURNAL] +**
  • [SQLITE_OPEN_TRANSIENT_DB] +**
  • [SQLITE_OPEN_SUBJOURNAL] +**
  • [SQLITE_OPEN_MASTER_JOURNAL] +**
  • [SQLITE_OPEN_WAL] +**
)^ +** +** The file I/O implementation can use the object type flags to +** change the way it deals with files. For example, an application +** that does not care about crash recovery or rollback might make +** the open of a journal file a no-op. Writes to this journal would +** also be no-ops, and any attempt to read the journal would return +** SQLITE_IOERR. Or the implementation might recognize that a database +** file will be doing page-aligned sector reads and writes in a random +** order and set up its I/O subsystem accordingly. +** +** SQLite might also add one of the following flags to the xOpen method: +** +**
    +**
  • [SQLITE_OPEN_DELETEONCLOSE] +**
  • [SQLITE_OPEN_EXCLUSIVE] +**
+** +** The [SQLITE_OPEN_DELETEONCLOSE] flag means the file should be +** deleted when it is closed. ^The [SQLITE_OPEN_DELETEONCLOSE] +** will be set for TEMP databases and their journals, transient +** databases, and subjournals. +** +** ^The [SQLITE_OPEN_EXCLUSIVE] flag is always used in conjunction +** with the [SQLITE_OPEN_CREATE] flag, which are both directly +** analogous to the O_EXCL and O_CREAT flags of the POSIX open() +** API. The SQLITE_OPEN_EXCLUSIVE flag, when paired with the +** SQLITE_OPEN_CREATE, is used to indicate that file should always +** be created, and that it is an error if it already exists. +** It is not used to indicate the file should be opened +** for exclusive access. +** +** ^At least szOsFile bytes of memory are allocated by SQLite +** to hold the [sqlite3_file] structure passed as the third +** argument to xOpen. The xOpen method does not have to +** allocate the structure; it should just fill it in. Note that +** the xOpen method must set the sqlite3_file.pMethods to either +** a valid [sqlite3_io_methods] object or to NULL. xOpen must do +** this even if the open fails. SQLite expects that the sqlite3_file.pMethods +** element will be valid after xOpen returns regardless of the success +** or failure of the xOpen call. +** +** [[sqlite3_vfs.xAccess]] +** ^The flags argument to xAccess() may be [SQLITE_ACCESS_EXISTS] +** to test for the existence of a file, or [SQLITE_ACCESS_READWRITE] to +** test whether a file is readable and writable, or [SQLITE_ACCESS_READ] +** to test whether a file is at least readable. The SQLITE_ACCESS_READ +** flag is never actually used and is not implemented in the built-in +** VFSes of SQLite. The file is named by the second argument and can be a +** directory. The xAccess method returns [SQLITE_OK] on success or some +** non-zero error code if there is an I/O error or if the name of +** the file given in the second argument is illegal. If SQLITE_OK +** is returned, then non-zero or zero is written into *pResOut to indicate +** whether or not the file is accessible. +** +** ^SQLite will always allocate at least mxPathname+1 bytes for the +** output buffer xFullPathname. The exact size of the output buffer +** is also passed as a parameter to both methods. If the output buffer +** is not large enough, [SQLITE_CANTOPEN] should be returned. Since this is +** handled as a fatal error by SQLite, vfs implementations should endeavor +** to prevent this by setting mxPathname to a sufficiently large value. +** +** The xRandomness(), xSleep(), xCurrentTime(), and xCurrentTimeInt64() +** interfaces are not strictly a part of the filesystem, but they are +** included in the VFS structure for completeness. +** The xRandomness() function attempts to return nBytes bytes +** of good-quality randomness into zOut. The return value is +** the actual number of bytes of randomness obtained. +** The xSleep() method causes the calling thread to sleep for at +** least the number of microseconds given. ^The xCurrentTime() +** method returns a Julian Day Number for the current date and time as +** a floating point value. +** ^The xCurrentTimeInt64() method returns, as an integer, the Julian +** Day Number multiplied by 86400000 (the number of milliseconds in +** a 24-hour day). +** ^SQLite will use the xCurrentTimeInt64() method to get the current +** date and time if that method is available (if iVersion is 2 or +** greater and the function pointer is not NULL) and will fall back +** to xCurrentTime() if xCurrentTimeInt64() is unavailable. +** +** ^The xSetSystemCall(), xGetSystemCall(), and xNestSystemCall() interfaces +** are not used by the SQLite core. These optional interfaces are provided +** by some VFSes to facilitate testing of the VFS code. By overriding +** system calls with functions under its control, a test program can +** simulate faults and error conditions that would otherwise be difficult +** or impossible to induce. The set of system calls that can be overridden +** varies from one VFS to another, and from one version of the same VFS to the +** next. Applications that use these interfaces must be prepared for any +** or all of these interfaces to be NULL or for their behavior to change +** from one release to the next. Applications must not attempt to access +** any of these methods if the iVersion of the VFS is less than 3. +*/ +typedef struct sqlite3_vfs sqlite3_vfs; +typedef void (*sqlite3_syscall_ptr)(void); +struct sqlite3_vfs { + int iVersion; /* Structure version number (currently 3) */ + int szOsFile; /* Size of subclassed sqlite3_file */ + int mxPathname; /* Maximum file pathname length */ + sqlite3_vfs *pNext; /* Next registered VFS */ + const char *zName; /* Name of this virtual file system */ + void *pAppData; /* Pointer to application-specific data */ + int (*xOpen)(sqlite3_vfs*, const char *zName, sqlite3_file*, + int flags, int *pOutFlags); + int (*xDelete)(sqlite3_vfs*, const char *zName, int syncDir); + int (*xAccess)(sqlite3_vfs*, const char *zName, int flags, int *pResOut); + int (*xFullPathname)(sqlite3_vfs*, const char *zName, int nOut, char *zOut); + void *(*xDlOpen)(sqlite3_vfs*, const char *zFilename); + void (*xDlError)(sqlite3_vfs*, int nByte, char *zErrMsg); + void (*(*xDlSym)(sqlite3_vfs*,void*, const char *zSymbol))(void); + void (*xDlClose)(sqlite3_vfs*, void*); + int (*xRandomness)(sqlite3_vfs*, int nByte, char *zOut); + int (*xSleep)(sqlite3_vfs*, int microseconds); + int (*xCurrentTime)(sqlite3_vfs*, double*); + int (*xGetLastError)(sqlite3_vfs*, int, char *); + /* + ** The methods above are in version 1 of the sqlite_vfs object + ** definition. Those that follow are added in version 2 or later + */ + int (*xCurrentTimeInt64)(sqlite3_vfs*, sqlite3_int64*); + /* + ** The methods above are in versions 1 and 2 of the sqlite_vfs object. + ** Those below are for version 3 and greater. + */ + int (*xSetSystemCall)(sqlite3_vfs*, const char *zName, sqlite3_syscall_ptr); + sqlite3_syscall_ptr (*xGetSystemCall)(sqlite3_vfs*, const char *zName); + const char *(*xNextSystemCall)(sqlite3_vfs*, const char *zName); + /* + ** The methods above are in versions 1 through 3 of the sqlite_vfs object. + ** New fields may be appended in future versions. The iVersion + ** value will increment whenever this happens. + */ +}; + +/* +** CAPI3REF: Flags for the xAccess VFS method +** +** These integer constants can be used as the third parameter to +** the xAccess method of an [sqlite3_vfs] object. They determine +** what kind of permissions the xAccess method is looking for. +** With SQLITE_ACCESS_EXISTS, the xAccess method +** simply checks whether the file exists. +** With SQLITE_ACCESS_READWRITE, the xAccess method +** checks whether the named directory is both readable and writable +** (in other words, if files can be added, removed, and renamed within +** the directory). +** The SQLITE_ACCESS_READWRITE constant is currently used only by the +** [temp_store_directory pragma], though this could change in a future +** release of SQLite. +** With SQLITE_ACCESS_READ, the xAccess method +** checks whether the file is readable. The SQLITE_ACCESS_READ constant is +** currently unused, though it might be used in a future release of +** SQLite. +*/ +#define SQLITE_ACCESS_EXISTS 0 +#define SQLITE_ACCESS_READWRITE 1 /* Used by PRAGMA temp_store_directory */ +#define SQLITE_ACCESS_READ 2 /* Unused */ + +/* +** CAPI3REF: Flags for the xShmLock VFS method +** +** These integer constants define the various locking operations +** allowed by the xShmLock method of [sqlite3_io_methods]. The +** following are the only legal combinations of flags to the +** xShmLock method: +** +**
    +**
  • SQLITE_SHM_LOCK | SQLITE_SHM_SHARED +**
  • SQLITE_SHM_LOCK | SQLITE_SHM_EXCLUSIVE +**
  • SQLITE_SHM_UNLOCK | SQLITE_SHM_SHARED +**
  • SQLITE_SHM_UNLOCK | SQLITE_SHM_EXCLUSIVE +**
+** +** When unlocking, the same SHARED or EXCLUSIVE flag must be supplied as +** was given on the corresponding lock. +** +** The xShmLock method can transition between unlocked and SHARED or +** between unlocked and EXCLUSIVE. It cannot transition between SHARED +** and EXCLUSIVE. +*/ +#define SQLITE_SHM_UNLOCK 1 +#define SQLITE_SHM_LOCK 2 +#define SQLITE_SHM_SHARED 4 +#define SQLITE_SHM_EXCLUSIVE 8 + +/* +** CAPI3REF: Maximum xShmLock index +** +** The xShmLock method on [sqlite3_io_methods] may use values +** between 0 and this upper bound as its "offset" argument. +** The SQLite core will never attempt to acquire or release a +** lock outside of this range +*/ +#define SQLITE_SHM_NLOCK 8 + + +/* +** CAPI3REF: Initialize The SQLite Library +** +** ^The sqlite3_initialize() routine initializes the +** SQLite library. ^The sqlite3_shutdown() routine +** deallocates any resources that were allocated by sqlite3_initialize(). +** These routines are designed to aid in process initialization and +** shutdown on embedded systems. Workstation applications using +** SQLite normally do not need to invoke either of these routines. +** +** A call to sqlite3_initialize() is an "effective" call if it is +** the first time sqlite3_initialize() is invoked during the lifetime of +** the process, or if it is the first time sqlite3_initialize() is invoked +** following a call to sqlite3_shutdown(). ^(Only an effective call +** of sqlite3_initialize() does any initialization. All other calls +** are harmless no-ops.)^ +** +** A call to sqlite3_shutdown() is an "effective" call if it is the first +** call to sqlite3_shutdown() since the last sqlite3_initialize(). ^(Only +** an effective call to sqlite3_shutdown() does any deinitialization. +** All other valid calls to sqlite3_shutdown() are harmless no-ops.)^ +** +** The sqlite3_initialize() interface is threadsafe, but sqlite3_shutdown() +** is not. The sqlite3_shutdown() interface must only be called from a +** single thread. All open [database connections] must be closed and all +** other SQLite resources must be deallocated prior to invoking +** sqlite3_shutdown(). +** +** Among other things, ^sqlite3_initialize() will invoke +** sqlite3_os_init(). Similarly, ^sqlite3_shutdown() +** will invoke sqlite3_os_end(). +** +** ^The sqlite3_initialize() routine returns [SQLITE_OK] on success. +** ^If for some reason, sqlite3_initialize() is unable to initialize +** the library (perhaps it is unable to allocate a needed resource such +** as a mutex) it returns an [error code] other than [SQLITE_OK]. +** +** ^The sqlite3_initialize() routine is called internally by many other +** SQLite interfaces so that an application usually does not need to +** invoke sqlite3_initialize() directly. For example, [sqlite3_open()] +** calls sqlite3_initialize() so the SQLite library will be automatically +** initialized when [sqlite3_open()] is called if it has not be initialized +** already. ^However, if SQLite is compiled with the [SQLITE_OMIT_AUTOINIT] +** compile-time option, then the automatic calls to sqlite3_initialize() +** are omitted and the application must call sqlite3_initialize() directly +** prior to using any other SQLite interface. For maximum portability, +** it is recommended that applications always invoke sqlite3_initialize() +** directly prior to using any other SQLite interface. Future releases +** of SQLite may require this. In other words, the behavior exhibited +** when SQLite is compiled with [SQLITE_OMIT_AUTOINIT] might become the +** default behavior in some future release of SQLite. +** +** The sqlite3_os_init() routine does operating-system specific +** initialization of the SQLite library. The sqlite3_os_end() +** routine undoes the effect of sqlite3_os_init(). Typical tasks +** performed by these routines include allocation or deallocation +** of static resources, initialization of global variables, +** setting up a default [sqlite3_vfs] module, or setting up +** a default configuration using [sqlite3_config()]. +** +** The application should never invoke either sqlite3_os_init() +** or sqlite3_os_end() directly. The application should only invoke +** sqlite3_initialize() and sqlite3_shutdown(). The sqlite3_os_init() +** interface is called automatically by sqlite3_initialize() and +** sqlite3_os_end() is called by sqlite3_shutdown(). Appropriate +** implementations for sqlite3_os_init() and sqlite3_os_end() +** are built into SQLite when it is compiled for Unix, Windows, or OS/2. +** When [custom builds | built for other platforms] +** (using the [SQLITE_OS_OTHER=1] compile-time +** option) the application must supply a suitable implementation for +** sqlite3_os_init() and sqlite3_os_end(). An application-supplied +** implementation of sqlite3_os_init() or sqlite3_os_end() +** must return [SQLITE_OK] on success and some other [error code] upon +** failure. +*/ +SQLITE_API int sqlite3_initialize(void); +SQLITE_API int sqlite3_shutdown(void); +SQLITE_API int sqlite3_os_init(void); +SQLITE_API int sqlite3_os_end(void); + +/* +** CAPI3REF: Configuring The SQLite Library +** +** The sqlite3_config() interface is used to make global configuration +** changes to SQLite in order to tune SQLite to the specific needs of +** the application. The default configuration is recommended for most +** applications and so this routine is usually not necessary. It is +** provided to support rare applications with unusual needs. +** +** The sqlite3_config() interface is not threadsafe. The application +** must ensure that no other SQLite interfaces are invoked by other +** threads while sqlite3_config() is running. +** +** The sqlite3_config() interface +** may only be invoked prior to library initialization using +** [sqlite3_initialize()] or after shutdown by [sqlite3_shutdown()]. +** ^If sqlite3_config() is called after [sqlite3_initialize()] and before +** [sqlite3_shutdown()] then it will return SQLITE_MISUSE. +** Note, however, that ^sqlite3_config() can be called as part of the +** implementation of an application-defined [sqlite3_os_init()]. +** +** The first argument to sqlite3_config() is an integer +** [configuration option] that determines +** what property of SQLite is to be configured. Subsequent arguments +** vary depending on the [configuration option] +** in the first argument. +** +** ^When a configuration option is set, sqlite3_config() returns [SQLITE_OK]. +** ^If the option is unknown or SQLite is unable to set the option +** then this routine returns a non-zero [error code]. +*/ +SQLITE_API int sqlite3_config(int, ...); + +/* +** CAPI3REF: Configure database connections +** METHOD: sqlite3 +** +** The sqlite3_db_config() interface is used to make configuration +** changes to a [database connection]. The interface is similar to +** [sqlite3_config()] except that the changes apply to a single +** [database connection] (specified in the first argument). +** +** The second argument to sqlite3_db_config(D,V,...) is the +** [SQLITE_DBCONFIG_LOOKASIDE | configuration verb] - an integer code +** that indicates what aspect of the [database connection] is being configured. +** Subsequent arguments vary depending on the configuration verb. +** +** ^Calls to sqlite3_db_config() return SQLITE_OK if and only if +** the call is considered successful. +*/ +SQLITE_API int sqlite3_db_config(sqlite3*, int op, ...); + +/* +** CAPI3REF: Memory Allocation Routines +** +** An instance of this object defines the interface between SQLite +** and low-level memory allocation routines. +** +** This object is used in only one place in the SQLite interface. +** A pointer to an instance of this object is the argument to +** [sqlite3_config()] when the configuration option is +** [SQLITE_CONFIG_MALLOC] or [SQLITE_CONFIG_GETMALLOC]. +** By creating an instance of this object +** and passing it to [sqlite3_config]([SQLITE_CONFIG_MALLOC]) +** during configuration, an application can specify an alternative +** memory allocation subsystem for SQLite to use for all of its +** dynamic memory needs. +** +** Note that SQLite comes with several [built-in memory allocators] +** that are perfectly adequate for the overwhelming majority of applications +** and that this object is only useful to a tiny minority of applications +** with specialized memory allocation requirements. This object is +** also used during testing of SQLite in order to specify an alternative +** memory allocator that simulates memory out-of-memory conditions in +** order to verify that SQLite recovers gracefully from such +** conditions. +** +** The xMalloc, xRealloc, and xFree methods must work like the +** malloc(), realloc() and free() functions from the standard C library. +** ^SQLite guarantees that the second argument to +** xRealloc is always a value returned by a prior call to xRoundup. +** +** xSize should return the allocated size of a memory allocation +** previously obtained from xMalloc or xRealloc. The allocated size +** is always at least as big as the requested size but may be larger. +** +** The xRoundup method returns what would be the allocated size of +** a memory allocation given a particular requested size. Most memory +** allocators round up memory allocations at least to the next multiple +** of 8. Some allocators round up to a larger multiple or to a power of 2. +** Every memory allocation request coming in through [sqlite3_malloc()] +** or [sqlite3_realloc()] first calls xRoundup. If xRoundup returns 0, +** that causes the corresponding memory allocation to fail. +** +** The xInit method initializes the memory allocator. For example, +** it might allocate any require mutexes or initialize internal data +** structures. The xShutdown method is invoked (indirectly) by +** [sqlite3_shutdown()] and should deallocate any resources acquired +** by xInit. The pAppData pointer is used as the only parameter to +** xInit and xShutdown. +** +** SQLite holds the [SQLITE_MUTEX_STATIC_MASTER] mutex when it invokes +** the xInit method, so the xInit method need not be threadsafe. The +** xShutdown method is only called from [sqlite3_shutdown()] so it does +** not need to be threadsafe either. For all other methods, SQLite +** holds the [SQLITE_MUTEX_STATIC_MEM] mutex as long as the +** [SQLITE_CONFIG_MEMSTATUS] configuration option is turned on (which +** it is by default) and so the methods are automatically serialized. +** However, if [SQLITE_CONFIG_MEMSTATUS] is disabled, then the other +** methods must be threadsafe or else make their own arrangements for +** serialization. +** +** SQLite will never invoke xInit() more than once without an intervening +** call to xShutdown(). +*/ +typedef struct sqlite3_mem_methods sqlite3_mem_methods; +struct sqlite3_mem_methods { + void *(*xMalloc)(int); /* Memory allocation function */ + void (*xFree)(void*); /* Free a prior allocation */ + void *(*xRealloc)(void*,int); /* Resize an allocation */ + int (*xSize)(void*); /* Return the size of an allocation */ + int (*xRoundup)(int); /* Round up request size to allocation size */ + int (*xInit)(void*); /* Initialize the memory allocator */ + void (*xShutdown)(void*); /* Deinitialize the memory allocator */ + void *pAppData; /* Argument to xInit() and xShutdown() */ +}; + +/* +** CAPI3REF: Configuration Options +** KEYWORDS: {configuration option} +** +** These constants are the available integer configuration options that +** can be passed as the first argument to the [sqlite3_config()] interface. +** +** New configuration options may be added in future releases of SQLite. +** Existing configuration options might be discontinued. Applications +** should check the return code from [sqlite3_config()] to make sure that +** the call worked. The [sqlite3_config()] interface will return a +** non-zero [error code] if a discontinued or unsupported configuration option +** is invoked. +** +**
+** [[SQLITE_CONFIG_SINGLETHREAD]]
SQLITE_CONFIG_SINGLETHREAD
+**
There are no arguments to this option. ^This option sets the +** [threading mode] to Single-thread. In other words, it disables +** all mutexing and puts SQLite into a mode where it can only be used +** by a single thread. ^If SQLite is compiled with +** the [SQLITE_THREADSAFE | SQLITE_THREADSAFE=0] compile-time option then +** it is not possible to change the [threading mode] from its default +** value of Single-thread and so [sqlite3_config()] will return +** [SQLITE_ERROR] if called with the SQLITE_CONFIG_SINGLETHREAD +** configuration option.
+** +** [[SQLITE_CONFIG_MULTITHREAD]]
SQLITE_CONFIG_MULTITHREAD
+**
There are no arguments to this option. ^This option sets the +** [threading mode] to Multi-thread. In other words, it disables +** mutexing on [database connection] and [prepared statement] objects. +** The application is responsible for serializing access to +** [database connections] and [prepared statements]. But other mutexes +** are enabled so that SQLite will be safe to use in a multi-threaded +** environment as long as no two threads attempt to use the same +** [database connection] at the same time. ^If SQLite is compiled with +** the [SQLITE_THREADSAFE | SQLITE_THREADSAFE=0] compile-time option then +** it is not possible to set the Multi-thread [threading mode] and +** [sqlite3_config()] will return [SQLITE_ERROR] if called with the +** SQLITE_CONFIG_MULTITHREAD configuration option.
+** +** [[SQLITE_CONFIG_SERIALIZED]]
SQLITE_CONFIG_SERIALIZED
+**
There are no arguments to this option. ^This option sets the +** [threading mode] to Serialized. In other words, this option enables +** all mutexes including the recursive +** mutexes on [database connection] and [prepared statement] objects. +** In this mode (which is the default when SQLite is compiled with +** [SQLITE_THREADSAFE=1]) the SQLite library will itself serialize access +** to [database connections] and [prepared statements] so that the +** application is free to use the same [database connection] or the +** same [prepared statement] in different threads at the same time. +** ^If SQLite is compiled with +** the [SQLITE_THREADSAFE | SQLITE_THREADSAFE=0] compile-time option then +** it is not possible to set the Serialized [threading mode] and +** [sqlite3_config()] will return [SQLITE_ERROR] if called with the +** SQLITE_CONFIG_SERIALIZED configuration option.
+** +** [[SQLITE_CONFIG_MALLOC]]
SQLITE_CONFIG_MALLOC
+**
^(The SQLITE_CONFIG_MALLOC option takes a single argument which is +** a pointer to an instance of the [sqlite3_mem_methods] structure. +** The argument specifies +** alternative low-level memory allocation routines to be used in place of +** the memory allocation routines built into SQLite.)^ ^SQLite makes +** its own private copy of the content of the [sqlite3_mem_methods] structure +** before the [sqlite3_config()] call returns.
+** +** [[SQLITE_CONFIG_GETMALLOC]]
SQLITE_CONFIG_GETMALLOC
+**
^(The SQLITE_CONFIG_GETMALLOC option takes a single argument which +** is a pointer to an instance of the [sqlite3_mem_methods] structure. +** The [sqlite3_mem_methods] +** structure is filled with the currently defined memory allocation routines.)^ +** This option can be used to overload the default memory allocation +** routines with a wrapper that simulations memory allocation failure or +** tracks memory usage, for example.
+** +** [[SQLITE_CONFIG_SMALL_MALLOC]]
SQLITE_CONFIG_SMALL_MALLOC
+**
^The SQLITE_CONFIG_SMALL_MALLOC option takes single argument of +** type int, interpreted as a boolean, which if true provides a hint to +** SQLite that it should avoid large memory allocations if possible. +** SQLite will run faster if it is free to make large memory allocations, +** but some application might prefer to run slower in exchange for +** guarantees about memory fragmentation that are possible if large +** allocations are avoided. This hint is normally off. +**
+** +** [[SQLITE_CONFIG_MEMSTATUS]]
SQLITE_CONFIG_MEMSTATUS
+**
^The SQLITE_CONFIG_MEMSTATUS option takes single argument of type int, +** interpreted as a boolean, which enables or disables the collection of +** memory allocation statistics. ^(When memory allocation statistics are +** disabled, the following SQLite interfaces become non-operational: +**
    +**
  • [sqlite3_memory_used()] +**
  • [sqlite3_memory_highwater()] +**
  • [sqlite3_soft_heap_limit64()] +**
  • [sqlite3_status64()] +**
)^ +** ^Memory allocation statistics are enabled by default unless SQLite is +** compiled with [SQLITE_DEFAULT_MEMSTATUS]=0 in which case memory +** allocation statistics are disabled by default. +**
+** +** [[SQLITE_CONFIG_SCRATCH]]
SQLITE_CONFIG_SCRATCH
+**
The SQLITE_CONFIG_SCRATCH option is no longer used. +**
+** +** [[SQLITE_CONFIG_PAGECACHE]]
SQLITE_CONFIG_PAGECACHE
+**
^The SQLITE_CONFIG_PAGECACHE option specifies a memory pool +** that SQLite can use for the database page cache with the default page +** cache implementation. +** This configuration option is a no-op if an application-define page +** cache implementation is loaded using the [SQLITE_CONFIG_PCACHE2]. +** ^There are three arguments to SQLITE_CONFIG_PAGECACHE: A pointer to +** 8-byte aligned memory (pMem), the size of each page cache line (sz), +** and the number of cache lines (N). +** The sz argument should be the size of the largest database page +** (a power of two between 512 and 65536) plus some extra bytes for each +** page header. ^The number of extra bytes needed by the page header +** can be determined using [SQLITE_CONFIG_PCACHE_HDRSZ]. +** ^It is harmless, apart from the wasted memory, +** for the sz parameter to be larger than necessary. The pMem +** argument must be either a NULL pointer or a pointer to an 8-byte +** aligned block of memory of at least sz*N bytes, otherwise +** subsequent behavior is undefined. +** ^When pMem is not NULL, SQLite will strive to use the memory provided +** to satisfy page cache needs, falling back to [sqlite3_malloc()] if +** a page cache line is larger than sz bytes or if all of the pMem buffer +** is exhausted. +** ^If pMem is NULL and N is non-zero, then each database connection +** does an initial bulk allocation for page cache memory +** from [sqlite3_malloc()] sufficient for N cache lines if N is positive or +** of -1024*N bytes if N is negative, . ^If additional +** page cache memory is needed beyond what is provided by the initial +** allocation, then SQLite goes to [sqlite3_malloc()] separately for each +** additional cache line.
+** +** [[SQLITE_CONFIG_HEAP]]
SQLITE_CONFIG_HEAP
+**
^The SQLITE_CONFIG_HEAP option specifies a static memory buffer +** that SQLite will use for all of its dynamic memory allocation needs +** beyond those provided for by [SQLITE_CONFIG_PAGECACHE]. +** ^The SQLITE_CONFIG_HEAP option is only available if SQLite is compiled +** with either [SQLITE_ENABLE_MEMSYS3] or [SQLITE_ENABLE_MEMSYS5] and returns +** [SQLITE_ERROR] if invoked otherwise. +** ^There are three arguments to SQLITE_CONFIG_HEAP: +** An 8-byte aligned pointer to the memory, +** the number of bytes in the memory buffer, and the minimum allocation size. +** ^If the first pointer (the memory pointer) is NULL, then SQLite reverts +** to using its default memory allocator (the system malloc() implementation), +** undoing any prior invocation of [SQLITE_CONFIG_MALLOC]. ^If the +** memory pointer is not NULL then the alternative memory +** allocator is engaged to handle all of SQLites memory allocation needs. +** The first pointer (the memory pointer) must be aligned to an 8-byte +** boundary or subsequent behavior of SQLite will be undefined. +** The minimum allocation size is capped at 2**12. Reasonable values +** for the minimum allocation size are 2**5 through 2**8.
+** +** [[SQLITE_CONFIG_MUTEX]]
SQLITE_CONFIG_MUTEX
+**
^(The SQLITE_CONFIG_MUTEX option takes a single argument which is a +** pointer to an instance of the [sqlite3_mutex_methods] structure. +** The argument specifies alternative low-level mutex routines to be used +** in place the mutex routines built into SQLite.)^ ^SQLite makes a copy of +** the content of the [sqlite3_mutex_methods] structure before the call to +** [sqlite3_config()] returns. ^If SQLite is compiled with +** the [SQLITE_THREADSAFE | SQLITE_THREADSAFE=0] compile-time option then +** the entire mutexing subsystem is omitted from the build and hence calls to +** [sqlite3_config()] with the SQLITE_CONFIG_MUTEX configuration option will +** return [SQLITE_ERROR].
+** +** [[SQLITE_CONFIG_GETMUTEX]]
SQLITE_CONFIG_GETMUTEX
+**
^(The SQLITE_CONFIG_GETMUTEX option takes a single argument which +** is a pointer to an instance of the [sqlite3_mutex_methods] structure. The +** [sqlite3_mutex_methods] +** structure is filled with the currently defined mutex routines.)^ +** This option can be used to overload the default mutex allocation +** routines with a wrapper used to track mutex usage for performance +** profiling or testing, for example. ^If SQLite is compiled with +** the [SQLITE_THREADSAFE | SQLITE_THREADSAFE=0] compile-time option then +** the entire mutexing subsystem is omitted from the build and hence calls to +** [sqlite3_config()] with the SQLITE_CONFIG_GETMUTEX configuration option will +** return [SQLITE_ERROR].
+** +** [[SQLITE_CONFIG_LOOKASIDE]]
SQLITE_CONFIG_LOOKASIDE
+**
^(The SQLITE_CONFIG_LOOKASIDE option takes two arguments that determine +** the default size of lookaside memory on each [database connection]. +** The first argument is the +** size of each lookaside buffer slot and the second is the number of +** slots allocated to each database connection.)^ ^(SQLITE_CONFIG_LOOKASIDE +** sets the default lookaside size. The [SQLITE_DBCONFIG_LOOKASIDE] +** option to [sqlite3_db_config()] can be used to change the lookaside +** configuration on individual connections.)^
+** +** [[SQLITE_CONFIG_PCACHE2]]
SQLITE_CONFIG_PCACHE2
+**
^(The SQLITE_CONFIG_PCACHE2 option takes a single argument which is +** a pointer to an [sqlite3_pcache_methods2] object. This object specifies +** the interface to a custom page cache implementation.)^ +** ^SQLite makes a copy of the [sqlite3_pcache_methods2] object.
+** +** [[SQLITE_CONFIG_GETPCACHE2]]
SQLITE_CONFIG_GETPCACHE2
+**
^(The SQLITE_CONFIG_GETPCACHE2 option takes a single argument which +** is a pointer to an [sqlite3_pcache_methods2] object. SQLite copies of +** the current page cache implementation into that object.)^
+** +** [[SQLITE_CONFIG_LOG]]
SQLITE_CONFIG_LOG
+**
The SQLITE_CONFIG_LOG option is used to configure the SQLite +** global [error log]. +** (^The SQLITE_CONFIG_LOG option takes two arguments: a pointer to a +** function with a call signature of void(*)(void*,int,const char*), +** and a pointer to void. ^If the function pointer is not NULL, it is +** invoked by [sqlite3_log()] to process each logging event. ^If the +** function pointer is NULL, the [sqlite3_log()] interface becomes a no-op. +** ^The void pointer that is the second argument to SQLITE_CONFIG_LOG is +** passed through as the first parameter to the application-defined logger +** function whenever that function is invoked. ^The second parameter to +** the logger function is a copy of the first parameter to the corresponding +** [sqlite3_log()] call and is intended to be a [result code] or an +** [extended result code]. ^The third parameter passed to the logger is +** log message after formatting via [sqlite3_snprintf()]. +** The SQLite logging interface is not reentrant; the logger function +** supplied by the application must not invoke any SQLite interface. +** In a multi-threaded application, the application-defined logger +** function must be threadsafe.
+** +** [[SQLITE_CONFIG_URI]]
SQLITE_CONFIG_URI +**
^(The SQLITE_CONFIG_URI option takes a single argument of type int. +** If non-zero, then URI handling is globally enabled. If the parameter is zero, +** then URI handling is globally disabled.)^ ^If URI handling is globally +** enabled, all filenames passed to [sqlite3_open()], [sqlite3_open_v2()], +** [sqlite3_open16()] or +** specified as part of [ATTACH] commands are interpreted as URIs, regardless +** of whether or not the [SQLITE_OPEN_URI] flag is set when the database +** connection is opened. ^If it is globally disabled, filenames are +** only interpreted as URIs if the SQLITE_OPEN_URI flag is set when the +** database connection is opened. ^(By default, URI handling is globally +** disabled. The default value may be changed by compiling with the +** [SQLITE_USE_URI] symbol defined.)^ +** +** [[SQLITE_CONFIG_COVERING_INDEX_SCAN]]
SQLITE_CONFIG_COVERING_INDEX_SCAN +**
^The SQLITE_CONFIG_COVERING_INDEX_SCAN option takes a single integer +** argument which is interpreted as a boolean in order to enable or disable +** the use of covering indices for full table scans in the query optimizer. +** ^The default setting is determined +** by the [SQLITE_ALLOW_COVERING_INDEX_SCAN] compile-time option, or is "on" +** if that compile-time option is omitted. +** The ability to disable the use of covering indices for full table scans +** is because some incorrectly coded legacy applications might malfunction +** when the optimization is enabled. Providing the ability to +** disable the optimization allows the older, buggy application code to work +** without change even with newer versions of SQLite. +** +** [[SQLITE_CONFIG_PCACHE]] [[SQLITE_CONFIG_GETPCACHE]] +**
SQLITE_CONFIG_PCACHE and SQLITE_CONFIG_GETPCACHE +**
These options are obsolete and should not be used by new code. +** They are retained for backwards compatibility but are now no-ops. +**
+** +** [[SQLITE_CONFIG_SQLLOG]] +**
SQLITE_CONFIG_SQLLOG +**
This option is only available if sqlite is compiled with the +** [SQLITE_ENABLE_SQLLOG] pre-processor macro defined. The first argument should +** be a pointer to a function of type void(*)(void*,sqlite3*,const char*, int). +** The second should be of type (void*). The callback is invoked by the library +** in three separate circumstances, identified by the value passed as the +** fourth parameter. If the fourth parameter is 0, then the database connection +** passed as the second argument has just been opened. The third argument +** points to a buffer containing the name of the main database file. If the +** fourth parameter is 1, then the SQL statement that the third parameter +** points to has just been executed. Or, if the fourth parameter is 2, then +** the connection being passed as the second parameter is being closed. The +** third parameter is passed NULL In this case. An example of using this +** configuration option can be seen in the "test_sqllog.c" source file in +** the canonical SQLite source tree.
+** +** [[SQLITE_CONFIG_MMAP_SIZE]] +**
SQLITE_CONFIG_MMAP_SIZE +**
^SQLITE_CONFIG_MMAP_SIZE takes two 64-bit integer (sqlite3_int64) values +** that are the default mmap size limit (the default setting for +** [PRAGMA mmap_size]) and the maximum allowed mmap size limit. +** ^The default setting can be overridden by each database connection using +** either the [PRAGMA mmap_size] command, or by using the +** [SQLITE_FCNTL_MMAP_SIZE] file control. ^(The maximum allowed mmap size +** will be silently truncated if necessary so that it does not exceed the +** compile-time maximum mmap size set by the +** [SQLITE_MAX_MMAP_SIZE] compile-time option.)^ +** ^If either argument to this option is negative, then that argument is +** changed to its compile-time default. +** +** [[SQLITE_CONFIG_WIN32_HEAPSIZE]] +**
SQLITE_CONFIG_WIN32_HEAPSIZE +**
^The SQLITE_CONFIG_WIN32_HEAPSIZE option is only available if SQLite is +** compiled for Windows with the [SQLITE_WIN32_MALLOC] pre-processor macro +** defined. ^SQLITE_CONFIG_WIN32_HEAPSIZE takes a 32-bit unsigned integer value +** that specifies the maximum size of the created heap. +** +** [[SQLITE_CONFIG_PCACHE_HDRSZ]] +**
SQLITE_CONFIG_PCACHE_HDRSZ +**
^The SQLITE_CONFIG_PCACHE_HDRSZ option takes a single parameter which +** is a pointer to an integer and writes into that integer the number of extra +** bytes per page required for each page in [SQLITE_CONFIG_PAGECACHE]. +** The amount of extra space required can change depending on the compiler, +** target platform, and SQLite version. +** +** [[SQLITE_CONFIG_PMASZ]] +**
SQLITE_CONFIG_PMASZ +**
^The SQLITE_CONFIG_PMASZ option takes a single parameter which +** is an unsigned integer and sets the "Minimum PMA Size" for the multithreaded +** sorter to that integer. The default minimum PMA Size is set by the +** [SQLITE_SORTER_PMASZ] compile-time option. New threads are launched +** to help with sort operations when multithreaded sorting +** is enabled (using the [PRAGMA threads] command) and the amount of content +** to be sorted exceeds the page size times the minimum of the +** [PRAGMA cache_size] setting and this value. +** +** [[SQLITE_CONFIG_STMTJRNL_SPILL]] +**
SQLITE_CONFIG_STMTJRNL_SPILL +**
^The SQLITE_CONFIG_STMTJRNL_SPILL option takes a single parameter which +** becomes the [statement journal] spill-to-disk threshold. +** [Statement journals] are held in memory until their size (in bytes) +** exceeds this threshold, at which point they are written to disk. +** Or if the threshold is -1, statement journals are always held +** exclusively in memory. +** Since many statement journals never become large, setting the spill +** threshold to a value such as 64KiB can greatly reduce the amount of +** I/O required to support statement rollback. +** The default value for this setting is controlled by the +** [SQLITE_STMTJRNL_SPILL] compile-time option. +** +** [[SQLITE_CONFIG_SORTERREF_SIZE]] +**
SQLITE_CONFIG_SORTERREF_SIZE +**
The SQLITE_CONFIG_SORTERREF_SIZE option accepts a single parameter +** of type (int) - the new value of the sorter-reference size threshold. +** Usually, when SQLite uses an external sort to order records according +** to an ORDER BY clause, all fields required by the caller are present in the +** sorted records. However, if SQLite determines based on the declared type +** of a table column that its values are likely to be very large - larger +** than the configured sorter-reference size threshold - then a reference +** is stored in each sorted record and the required column values loaded +** from the database as records are returned in sorted order. The default +** value for this option is to never use this optimization. Specifying a +** negative value for this option restores the default behaviour. +** This option is only available if SQLite is compiled with the +** [SQLITE_ENABLE_SORTER_REFERENCES] compile-time option. +** +** [[SQLITE_CONFIG_MEMDB_MAXSIZE]] +**
SQLITE_CONFIG_MEMDB_MAXSIZE +**
The SQLITE_CONFIG_MEMDB_MAXSIZE option accepts a single parameter +** [sqlite3_int64] parameter which is the default maximum size for an in-memory +** database created using [sqlite3_deserialize()]. This default maximum +** size can be adjusted up or down for individual databases using the +** [SQLITE_FCNTL_SIZE_LIMIT] [sqlite3_file_control|file-control]. If this +** configuration setting is never used, then the default maximum is determined +** by the [SQLITE_MEMDB_DEFAULT_MAXSIZE] compile-time option. If that +** compile-time option is not set, then the default maximum is 1073741824. +**
+*/ +#define SQLITE_CONFIG_SINGLETHREAD 1 /* nil */ +#define SQLITE_CONFIG_MULTITHREAD 2 /* nil */ +#define SQLITE_CONFIG_SERIALIZED 3 /* nil */ +#define SQLITE_CONFIG_MALLOC 4 /* sqlite3_mem_methods* */ +#define SQLITE_CONFIG_GETMALLOC 5 /* sqlite3_mem_methods* */ +#define SQLITE_CONFIG_SCRATCH 6 /* No longer used */ +#define SQLITE_CONFIG_PAGECACHE 7 /* void*, int sz, int N */ +#define SQLITE_CONFIG_HEAP 8 /* void*, int nByte, int min */ +#define SQLITE_CONFIG_MEMSTATUS 9 /* boolean */ +#define SQLITE_CONFIG_MUTEX 10 /* sqlite3_mutex_methods* */ +#define SQLITE_CONFIG_GETMUTEX 11 /* sqlite3_mutex_methods* */ +/* previously SQLITE_CONFIG_CHUNKALLOC 12 which is now unused. */ +#define SQLITE_CONFIG_LOOKASIDE 13 /* int int */ +#define SQLITE_CONFIG_PCACHE 14 /* no-op */ +#define SQLITE_CONFIG_GETPCACHE 15 /* no-op */ +#define SQLITE_CONFIG_LOG 16 /* xFunc, void* */ +#define SQLITE_CONFIG_URI 17 /* int */ +#define SQLITE_CONFIG_PCACHE2 18 /* sqlite3_pcache_methods2* */ +#define SQLITE_CONFIG_GETPCACHE2 19 /* sqlite3_pcache_methods2* */ +#define SQLITE_CONFIG_COVERING_INDEX_SCAN 20 /* int */ +#define SQLITE_CONFIG_SQLLOG 21 /* xSqllog, void* */ +#define SQLITE_CONFIG_MMAP_SIZE 22 /* sqlite3_int64, sqlite3_int64 */ +#define SQLITE_CONFIG_WIN32_HEAPSIZE 23 /* int nByte */ +#define SQLITE_CONFIG_PCACHE_HDRSZ 24 /* int *psz */ +#define SQLITE_CONFIG_PMASZ 25 /* unsigned int szPma */ +#define SQLITE_CONFIG_STMTJRNL_SPILL 26 /* int nByte */ +#define SQLITE_CONFIG_SMALL_MALLOC 27 /* boolean */ +#define SQLITE_CONFIG_SORTERREF_SIZE 28 /* int nByte */ +#define SQLITE_CONFIG_MEMDB_MAXSIZE 29 /* sqlite3_int64 */ + +/* +** CAPI3REF: Database Connection Configuration Options +** +** These constants are the available integer configuration options that +** can be passed as the second argument to the [sqlite3_db_config()] interface. +** +** New configuration options may be added in future releases of SQLite. +** Existing configuration options might be discontinued. Applications +** should check the return code from [sqlite3_db_config()] to make sure that +** the call worked. ^The [sqlite3_db_config()] interface will return a +** non-zero [error code] if a discontinued or unsupported configuration option +** is invoked. +** +**
+** [[SQLITE_DBCONFIG_LOOKASIDE]] +**
SQLITE_DBCONFIG_LOOKASIDE
+**
^This option takes three additional arguments that determine the +** [lookaside memory allocator] configuration for the [database connection]. +** ^The first argument (the third parameter to [sqlite3_db_config()] is a +** pointer to a memory buffer to use for lookaside memory. +** ^The first argument after the SQLITE_DBCONFIG_LOOKASIDE verb +** may be NULL in which case SQLite will allocate the +** lookaside buffer itself using [sqlite3_malloc()]. ^The second argument is the +** size of each lookaside buffer slot. ^The third argument is the number of +** slots. The size of the buffer in the first argument must be greater than +** or equal to the product of the second and third arguments. The buffer +** must be aligned to an 8-byte boundary. ^If the second argument to +** SQLITE_DBCONFIG_LOOKASIDE is not a multiple of 8, it is internally +** rounded down to the next smaller multiple of 8. ^(The lookaside memory +** configuration for a database connection can only be changed when that +** connection is not currently using lookaside memory, or in other words +** when the "current value" returned by +** [sqlite3_db_status](D,[SQLITE_CONFIG_LOOKASIDE],...) is zero. +** Any attempt to change the lookaside memory configuration when lookaside +** memory is in use leaves the configuration unchanged and returns +** [SQLITE_BUSY].)^
+** +** [[SQLITE_DBCONFIG_ENABLE_FKEY]] +**
SQLITE_DBCONFIG_ENABLE_FKEY
+**
^This option is used to enable or disable the enforcement of +** [foreign key constraints]. There should be two additional arguments. +** The first argument is an integer which is 0 to disable FK enforcement, +** positive to enable FK enforcement or negative to leave FK enforcement +** unchanged. The second parameter is a pointer to an integer into which +** is written 0 or 1 to indicate whether FK enforcement is off or on +** following this call. The second parameter may be a NULL pointer, in +** which case the FK enforcement setting is not reported back.
+** +** [[SQLITE_DBCONFIG_ENABLE_TRIGGER]] +**
SQLITE_DBCONFIG_ENABLE_TRIGGER
+**
^This option is used to enable or disable [CREATE TRIGGER | triggers]. +** There should be two additional arguments. +** The first argument is an integer which is 0 to disable triggers, +** positive to enable triggers or negative to leave the setting unchanged. +** The second parameter is a pointer to an integer into which +** is written 0 or 1 to indicate whether triggers are disabled or enabled +** following this call. The second parameter may be a NULL pointer, in +** which case the trigger setting is not reported back.
+** +** [[SQLITE_DBCONFIG_ENABLE_VIEW]] +**
SQLITE_DBCONFIG_ENABLE_VIEW
+**
^This option is used to enable or disable [CREATE VIEW | views]. +** There should be two additional arguments. +** The first argument is an integer which is 0 to disable views, +** positive to enable views or negative to leave the setting unchanged. +** The second parameter is a pointer to an integer into which +** is written 0 or 1 to indicate whether views are disabled or enabled +** following this call. The second parameter may be a NULL pointer, in +** which case the view setting is not reported back.
+** +** [[SQLITE_DBCONFIG_ENABLE_FTS3_TOKENIZER]] +**
SQLITE_DBCONFIG_ENABLE_FTS3_TOKENIZER
+**
^This option is used to enable or disable the +** [fts3_tokenizer()] function which is part of the +** [FTS3] full-text search engine extension. +** There should be two additional arguments. +** The first argument is an integer which is 0 to disable fts3_tokenizer() or +** positive to enable fts3_tokenizer() or negative to leave the setting +** unchanged. +** The second parameter is a pointer to an integer into which +** is written 0 or 1 to indicate whether fts3_tokenizer is disabled or enabled +** following this call. The second parameter may be a NULL pointer, in +** which case the new setting is not reported back.
+** +** [[SQLITE_DBCONFIG_ENABLE_LOAD_EXTENSION]] +**
SQLITE_DBCONFIG_ENABLE_LOAD_EXTENSION
+**
^This option is used to enable or disable the [sqlite3_load_extension()] +** interface independently of the [load_extension()] SQL function. +** The [sqlite3_enable_load_extension()] API enables or disables both the +** C-API [sqlite3_load_extension()] and the SQL function [load_extension()]. +** There should be two additional arguments. +** When the first argument to this interface is 1, then only the C-API is +** enabled and the SQL function remains disabled. If the first argument to +** this interface is 0, then both the C-API and the SQL function are disabled. +** If the first argument is -1, then no changes are made to state of either the +** C-API or the SQL function. +** The second parameter is a pointer to an integer into which +** is written 0 or 1 to indicate whether [sqlite3_load_extension()] interface +** is disabled or enabled following this call. The second parameter may +** be a NULL pointer, in which case the new setting is not reported back. +**
+** +** [[SQLITE_DBCONFIG_MAINDBNAME]]
SQLITE_DBCONFIG_MAINDBNAME
+**
^This option is used to change the name of the "main" database +** schema. ^The sole argument is a pointer to a constant UTF8 string +** which will become the new schema name in place of "main". ^SQLite +** does not make a copy of the new main schema name string, so the application +** must ensure that the argument passed into this DBCONFIG option is unchanged +** until after the database connection closes. +**
+** +** [[SQLITE_DBCONFIG_NO_CKPT_ON_CLOSE]] +**
SQLITE_DBCONFIG_NO_CKPT_ON_CLOSE
+**
Usually, when a database in wal mode is closed or detached from a +** database handle, SQLite checks if this will mean that there are now no +** connections at all to the database. If so, it performs a checkpoint +** operation before closing the connection. This option may be used to +** override this behaviour. The first parameter passed to this operation +** is an integer - positive to disable checkpoints-on-close, or zero (the +** default) to enable them, and negative to leave the setting unchanged. +** The second parameter is a pointer to an integer +** into which is written 0 or 1 to indicate whether checkpoints-on-close +** have been disabled - 0 if they are not disabled, 1 if they are. +**
+** +** [[SQLITE_DBCONFIG_ENABLE_QPSG]]
SQLITE_DBCONFIG_ENABLE_QPSG
+**
^(The SQLITE_DBCONFIG_ENABLE_QPSG option activates or deactivates +** the [query planner stability guarantee] (QPSG). When the QPSG is active, +** a single SQL query statement will always use the same algorithm regardless +** of values of [bound parameters].)^ The QPSG disables some query optimizations +** that look at the values of bound parameters, which can make some queries +** slower. But the QPSG has the advantage of more predictable behavior. With +** the QPSG active, SQLite will always use the same query plan in the field as +** was used during testing in the lab. +** The first argument to this setting is an integer which is 0 to disable +** the QPSG, positive to enable QPSG, or negative to leave the setting +** unchanged. The second parameter is a pointer to an integer into which +** is written 0 or 1 to indicate whether the QPSG is disabled or enabled +** following this call. +**
+** +** [[SQLITE_DBCONFIG_TRIGGER_EQP]]
SQLITE_DBCONFIG_TRIGGER_EQP
+**
By default, the output of EXPLAIN QUERY PLAN commands does not +** include output for any operations performed by trigger programs. This +** option is used to set or clear (the default) a flag that governs this +** behavior. The first parameter passed to this operation is an integer - +** positive to enable output for trigger programs, or zero to disable it, +** or negative to leave the setting unchanged. +** The second parameter is a pointer to an integer into which is written +** 0 or 1 to indicate whether output-for-triggers has been disabled - 0 if +** it is not disabled, 1 if it is. +**
+** +** [[SQLITE_DBCONFIG_RESET_DATABASE]]
SQLITE_DBCONFIG_RESET_DATABASE
+**
Set the SQLITE_DBCONFIG_RESET_DATABASE flag and then run +** [VACUUM] in order to reset a database back to an empty database +** with no schema and no content. The following process works even for +** a badly corrupted database file: +**
    +**
  1. If the database connection is newly opened, make sure it has read the +** database schema by preparing then discarding some query against the +** database, or calling sqlite3_table_column_metadata(), ignoring any +** errors. This step is only necessary if the application desires to keep +** the database in WAL mode after the reset if it was in WAL mode before +** the reset. +**
  2. sqlite3_db_config(db, SQLITE_DBCONFIG_RESET_DATABASE, 1, 0); +**
  3. [sqlite3_exec](db, "[VACUUM]", 0, 0, 0); +**
  4. sqlite3_db_config(db, SQLITE_DBCONFIG_RESET_DATABASE, 0, 0); +**
+** Because resetting a database is destructive and irreversible, the +** process requires the use of this obscure API and multiple steps to help +** ensure that it does not happen by accident. +** +** [[SQLITE_DBCONFIG_DEFENSIVE]]
SQLITE_DBCONFIG_DEFENSIVE
+**
The SQLITE_DBCONFIG_DEFENSIVE option activates or deactivates the +** "defensive" flag for a database connection. When the defensive +** flag is enabled, language features that allow ordinary SQL to +** deliberately corrupt the database file are disabled. The disabled +** features include but are not limited to the following: +**
    +**
  • The [PRAGMA writable_schema=ON] statement. +**
  • The [PRAGMA journal_mode=OFF] statement. +**
  • Writes to the [sqlite_dbpage] virtual table. +**
  • Direct writes to [shadow tables]. +**
+**
+** +** [[SQLITE_DBCONFIG_WRITABLE_SCHEMA]]
SQLITE_DBCONFIG_WRITABLE_SCHEMA
+**
The SQLITE_DBCONFIG_WRITABLE_SCHEMA option activates or deactivates the +** "writable_schema" flag. This has the same effect and is logically equivalent +** to setting [PRAGMA writable_schema=ON] or [PRAGMA writable_schema=OFF]. +** The first argument to this setting is an integer which is 0 to disable +** the writable_schema, positive to enable writable_schema, or negative to +** leave the setting unchanged. The second parameter is a pointer to an +** integer into which is written 0 or 1 to indicate whether the writable_schema +** is enabled or disabled following this call. +**
+** +** [[SQLITE_DBCONFIG_LEGACY_ALTER_TABLE]] +**
SQLITE_DBCONFIG_LEGACY_ALTER_TABLE
+**
The SQLITE_DBCONFIG_LEGACY_ALTER_TABLE option activates or deactivates +** the legacy behavior of the [ALTER TABLE RENAME] command such it +** behaves as it did prior to [version 3.24.0] (2018-06-04). See the +** "Compatibility Notice" on the [ALTER TABLE RENAME documentation] for +** additional information. This feature can also be turned on and off +** using the [PRAGMA legacy_alter_table] statement. +**
+** +** [[SQLITE_DBCONFIG_DQS_DML]] +**
SQLITE_DBCONFIG_DQS_DML +**
The SQLITE_DBCONFIG_DQS_DML option activates or deactivates +** the legacy [double-quoted string literal] misfeature for DML statement +** only, that is DELETE, INSERT, SELECT, and UPDATE statements. The +** default value of this setting is determined by the [-DSQLITE_DQS] +** compile-time option. +**
+** +** [[SQLITE_DBCONFIG_DQS_DDL]] +**
SQLITE_DBCONFIG_DQS_DDL +**
The SQLITE_DBCONFIG_DQS option activates or deactivates +** the legacy [double-quoted string literal] misfeature for DDL statements, +** such as CREATE TABLE and CREATE INDEX. The +** default value of this setting is determined by the [-DSQLITE_DQS] +** compile-time option. +**
+**
+*/ +#define SQLITE_DBCONFIG_MAINDBNAME 1000 /* const char* */ +#define SQLITE_DBCONFIG_LOOKASIDE 1001 /* void* int int */ +#define SQLITE_DBCONFIG_ENABLE_FKEY 1002 /* int int* */ +#define SQLITE_DBCONFIG_ENABLE_TRIGGER 1003 /* int int* */ +#define SQLITE_DBCONFIG_ENABLE_FTS3_TOKENIZER 1004 /* int int* */ +#define SQLITE_DBCONFIG_ENABLE_LOAD_EXTENSION 1005 /* int int* */ +#define SQLITE_DBCONFIG_NO_CKPT_ON_CLOSE 1006 /* int int* */ +#define SQLITE_DBCONFIG_ENABLE_QPSG 1007 /* int int* */ +#define SQLITE_DBCONFIG_TRIGGER_EQP 1008 /* int int* */ +#define SQLITE_DBCONFIG_RESET_DATABASE 1009 /* int int* */ +#define SQLITE_DBCONFIG_DEFENSIVE 1010 /* int int* */ +#define SQLITE_DBCONFIG_WRITABLE_SCHEMA 1011 /* int int* */ +#define SQLITE_DBCONFIG_LEGACY_ALTER_TABLE 1012 /* int int* */ +#define SQLITE_DBCONFIG_DQS_DML 1013 /* int int* */ +#define SQLITE_DBCONFIG_DQS_DDL 1014 /* int int* */ +#define SQLITE_DBCONFIG_ENABLE_VIEW 1015 /* int int* */ +#define SQLITE_DBCONFIG_MAX 1015 /* Largest DBCONFIG */ + +/* +** CAPI3REF: Enable Or Disable Extended Result Codes +** METHOD: sqlite3 +** +** ^The sqlite3_extended_result_codes() routine enables or disables the +** [extended result codes] feature of SQLite. ^The extended result +** codes are disabled by default for historical compatibility. +*/ +SQLITE_API int sqlite3_extended_result_codes(sqlite3*, int onoff); + +/* +** CAPI3REF: Last Insert Rowid +** METHOD: sqlite3 +** +** ^Each entry in most SQLite tables (except for [WITHOUT ROWID] tables) +** has a unique 64-bit signed +** integer key called the [ROWID | "rowid"]. ^The rowid is always available +** as an undeclared column named ROWID, OID, or _ROWID_ as long as those +** names are not also used by explicitly declared columns. ^If +** the table has a column of type [INTEGER PRIMARY KEY] then that column +** is another alias for the rowid. +** +** ^The sqlite3_last_insert_rowid(D) interface usually returns the [rowid] of +** the most recent successful [INSERT] into a rowid table or [virtual table] +** on database connection D. ^Inserts into [WITHOUT ROWID] tables are not +** recorded. ^If no successful [INSERT]s into rowid tables have ever occurred +** on the database connection D, then sqlite3_last_insert_rowid(D) returns +** zero. +** +** As well as being set automatically as rows are inserted into database +** tables, the value returned by this function may be set explicitly by +** [sqlite3_set_last_insert_rowid()] +** +** Some virtual table implementations may INSERT rows into rowid tables as +** part of committing a transaction (e.g. to flush data accumulated in memory +** to disk). In this case subsequent calls to this function return the rowid +** associated with these internal INSERT operations, which leads to +** unintuitive results. Virtual table implementations that do write to rowid +** tables in this way can avoid this problem by restoring the original +** rowid value using [sqlite3_set_last_insert_rowid()] before returning +** control to the user. +** +** ^(If an [INSERT] occurs within a trigger then this routine will +** return the [rowid] of the inserted row as long as the trigger is +** running. Once the trigger program ends, the value returned +** by this routine reverts to what it was before the trigger was fired.)^ +** +** ^An [INSERT] that fails due to a constraint violation is not a +** successful [INSERT] and does not change the value returned by this +** routine. ^Thus INSERT OR FAIL, INSERT OR IGNORE, INSERT OR ROLLBACK, +** and INSERT OR ABORT make no changes to the return value of this +** routine when their insertion fails. ^(When INSERT OR REPLACE +** encounters a constraint violation, it does not fail. The +** INSERT continues to completion after deleting rows that caused +** the constraint problem so INSERT OR REPLACE will always change +** the return value of this interface.)^ +** +** ^For the purposes of this routine, an [INSERT] is considered to +** be successful even if it is subsequently rolled back. +** +** This function is accessible to SQL statements via the +** [last_insert_rowid() SQL function]. +** +** If a separate thread performs a new [INSERT] on the same +** database connection while the [sqlite3_last_insert_rowid()] +** function is running and thus changes the last insert [rowid], +** then the value returned by [sqlite3_last_insert_rowid()] is +** unpredictable and might not equal either the old or the new +** last insert [rowid]. +*/ +SQLITE_API sqlite3_int64 sqlite3_last_insert_rowid(sqlite3*); + +/* +** CAPI3REF: Set the Last Insert Rowid value. +** METHOD: sqlite3 +** +** The sqlite3_set_last_insert_rowid(D, R) method allows the application to +** set the value returned by calling sqlite3_last_insert_rowid(D) to R +** without inserting a row into the database. +*/ +SQLITE_API void sqlite3_set_last_insert_rowid(sqlite3*,sqlite3_int64); + +/* +** CAPI3REF: Count The Number Of Rows Modified +** METHOD: sqlite3 +** +** ^This function returns the number of rows modified, inserted or +** deleted by the most recently completed INSERT, UPDATE or DELETE +** statement on the database connection specified by the only parameter. +** ^Executing any other type of SQL statement does not modify the value +** returned by this function. +** +** ^Only changes made directly by the INSERT, UPDATE or DELETE statement are +** considered - auxiliary changes caused by [CREATE TRIGGER | triggers], +** [foreign key actions] or [REPLACE] constraint resolution are not counted. +** +** Changes to a view that are intercepted by +** [INSTEAD OF trigger | INSTEAD OF triggers] are not counted. ^The value +** returned by sqlite3_changes() immediately after an INSERT, UPDATE or +** DELETE statement run on a view is always zero. Only changes made to real +** tables are counted. +** +** Things are more complicated if the sqlite3_changes() function is +** executed while a trigger program is running. This may happen if the +** program uses the [changes() SQL function], or if some other callback +** function invokes sqlite3_changes() directly. Essentially: +** +**
    +**
  • ^(Before entering a trigger program the value returned by +** sqlite3_changes() function is saved. After the trigger program +** has finished, the original value is restored.)^ +** +**
  • ^(Within a trigger program each INSERT, UPDATE and DELETE +** statement sets the value returned by sqlite3_changes() +** upon completion as normal. Of course, this value will not include +** any changes performed by sub-triggers, as the sqlite3_changes() +** value will be saved and restored after each sub-trigger has run.)^ +**
+** +** ^This means that if the changes() SQL function (or similar) is used +** by the first INSERT, UPDATE or DELETE statement within a trigger, it +** returns the value as set when the calling statement began executing. +** ^If it is used by the second or subsequent such statement within a trigger +** program, the value returned reflects the number of rows modified by the +** previous INSERT, UPDATE or DELETE statement within the same trigger. +** +** If a separate thread makes changes on the same database connection +** while [sqlite3_changes()] is running then the value returned +** is unpredictable and not meaningful. +** +** See also: +**
    +**
  • the [sqlite3_total_changes()] interface +**
  • the [count_changes pragma] +**
  • the [changes() SQL function] +**
  • the [data_version pragma] +**
+*/ +SQLITE_API int sqlite3_changes(sqlite3*); + +/* +** CAPI3REF: Total Number Of Rows Modified +** METHOD: sqlite3 +** +** ^This function returns the total number of rows inserted, modified or +** deleted by all [INSERT], [UPDATE] or [DELETE] statements completed +** since the database connection was opened, including those executed as +** part of trigger programs. ^Executing any other type of SQL statement +** does not affect the value returned by sqlite3_total_changes(). +** +** ^Changes made as part of [foreign key actions] are included in the +** count, but those made as part of REPLACE constraint resolution are +** not. ^Changes to a view that are intercepted by INSTEAD OF triggers +** are not counted. +** +** The [sqlite3_total_changes(D)] interface only reports the number +** of rows that changed due to SQL statement run against database +** connection D. Any changes by other database connections are ignored. +** To detect changes against a database file from other database +** connections use the [PRAGMA data_version] command or the +** [SQLITE_FCNTL_DATA_VERSION] [file control]. +** +** If a separate thread makes changes on the same database connection +** while [sqlite3_total_changes()] is running then the value +** returned is unpredictable and not meaningful. +** +** See also: +**
    +**
  • the [sqlite3_changes()] interface +**
  • the [count_changes pragma] +**
  • the [changes() SQL function] +**
  • the [data_version pragma] +**
  • the [SQLITE_FCNTL_DATA_VERSION] [file control] +**
+*/ +SQLITE_API int sqlite3_total_changes(sqlite3*); + +/* +** CAPI3REF: Interrupt A Long-Running Query +** METHOD: sqlite3 +** +** ^This function causes any pending database operation to abort and +** return at its earliest opportunity. This routine is typically +** called in response to a user action such as pressing "Cancel" +** or Ctrl-C where the user wants a long query operation to halt +** immediately. +** +** ^It is safe to call this routine from a thread different from the +** thread that is currently running the database operation. But it +** is not safe to call this routine with a [database connection] that +** is closed or might close before sqlite3_interrupt() returns. +** +** ^If an SQL operation is very nearly finished at the time when +** sqlite3_interrupt() is called, then it might not have an opportunity +** to be interrupted and might continue to completion. +** +** ^An SQL operation that is interrupted will return [SQLITE_INTERRUPT]. +** ^If the interrupted SQL operation is an INSERT, UPDATE, or DELETE +** that is inside an explicit transaction, then the entire transaction +** will be rolled back automatically. +** +** ^The sqlite3_interrupt(D) call is in effect until all currently running +** SQL statements on [database connection] D complete. ^Any new SQL statements +** that are started after the sqlite3_interrupt() call and before the +** running statements reaches zero are interrupted as if they had been +** running prior to the sqlite3_interrupt() call. ^New SQL statements +** that are started after the running statement count reaches zero are +** not effected by the sqlite3_interrupt(). +** ^A call to sqlite3_interrupt(D) that occurs when there are no running +** SQL statements is a no-op and has no effect on SQL statements +** that are started after the sqlite3_interrupt() call returns. +*/ +SQLITE_API void sqlite3_interrupt(sqlite3*); + +/* +** CAPI3REF: Determine If An SQL Statement Is Complete +** +** These routines are useful during command-line input to determine if the +** currently entered text seems to form a complete SQL statement or +** if additional input is needed before sending the text into +** SQLite for parsing. ^These routines return 1 if the input string +** appears to be a complete SQL statement. ^A statement is judged to be +** complete if it ends with a semicolon token and is not a prefix of a +** well-formed CREATE TRIGGER statement. ^Semicolons that are embedded within +** string literals or quoted identifier names or comments are not +** independent tokens (they are part of the token in which they are +** embedded) and thus do not count as a statement terminator. ^Whitespace +** and comments that follow the final semicolon are ignored. +** +** ^These routines return 0 if the statement is incomplete. ^If a +** memory allocation fails, then SQLITE_NOMEM is returned. +** +** ^These routines do not parse the SQL statements thus +** will not detect syntactically incorrect SQL. +** +** ^(If SQLite has not been initialized using [sqlite3_initialize()] prior +** to invoking sqlite3_complete16() then sqlite3_initialize() is invoked +** automatically by sqlite3_complete16(). If that initialization fails, +** then the return value from sqlite3_complete16() will be non-zero +** regardless of whether or not the input SQL is complete.)^ +** +** The input to [sqlite3_complete()] must be a zero-terminated +** UTF-8 string. +** +** The input to [sqlite3_complete16()] must be a zero-terminated +** UTF-16 string in native byte order. +*/ +SQLITE_API int sqlite3_complete(const char *sql); +SQLITE_API int sqlite3_complete16(const void *sql); + +/* +** CAPI3REF: Register A Callback To Handle SQLITE_BUSY Errors +** KEYWORDS: {busy-handler callback} {busy handler} +** METHOD: sqlite3 +** +** ^The sqlite3_busy_handler(D,X,P) routine sets a callback function X +** that might be invoked with argument P whenever +** an attempt is made to access a database table associated with +** [database connection] D when another thread +** or process has the table locked. +** The sqlite3_busy_handler() interface is used to implement +** [sqlite3_busy_timeout()] and [PRAGMA busy_timeout]. +** +** ^If the busy callback is NULL, then [SQLITE_BUSY] +** is returned immediately upon encountering the lock. ^If the busy callback +** is not NULL, then the callback might be invoked with two arguments. +** +** ^The first argument to the busy handler is a copy of the void* pointer which +** is the third argument to sqlite3_busy_handler(). ^The second argument to +** the busy handler callback is the number of times that the busy handler has +** been invoked previously for the same locking event. ^If the +** busy callback returns 0, then no additional attempts are made to +** access the database and [SQLITE_BUSY] is returned +** to the application. +** ^If the callback returns non-zero, then another attempt +** is made to access the database and the cycle repeats. +** +** The presence of a busy handler does not guarantee that it will be invoked +** when there is lock contention. ^If SQLite determines that invoking the busy +** handler could result in a deadlock, it will go ahead and return [SQLITE_BUSY] +** to the application instead of invoking the +** busy handler. +** Consider a scenario where one process is holding a read lock that +** it is trying to promote to a reserved lock and +** a second process is holding a reserved lock that it is trying +** to promote to an exclusive lock. The first process cannot proceed +** because it is blocked by the second and the second process cannot +** proceed because it is blocked by the first. If both processes +** invoke the busy handlers, neither will make any progress. Therefore, +** SQLite returns [SQLITE_BUSY] for the first process, hoping that this +** will induce the first process to release its read lock and allow +** the second process to proceed. +** +** ^The default busy callback is NULL. +** +** ^(There can only be a single busy handler defined for each +** [database connection]. Setting a new busy handler clears any +** previously set handler.)^ ^Note that calling [sqlite3_busy_timeout()] +** or evaluating [PRAGMA busy_timeout=N] will change the +** busy handler and thus clear any previously set busy handler. +** +** The busy callback should not take any actions which modify the +** database connection that invoked the busy handler. In other words, +** the busy handler is not reentrant. Any such actions +** result in undefined behavior. +** +** A busy handler must not close the database connection +** or [prepared statement] that invoked the busy handler. +*/ +SQLITE_API int sqlite3_busy_handler(sqlite3*,int(*)(void*,int),void*); + +/* +** CAPI3REF: Set A Busy Timeout +** METHOD: sqlite3 +** +** ^This routine sets a [sqlite3_busy_handler | busy handler] that sleeps +** for a specified amount of time when a table is locked. ^The handler +** will sleep multiple times until at least "ms" milliseconds of sleeping +** have accumulated. ^After at least "ms" milliseconds of sleeping, +** the handler returns 0 which causes [sqlite3_step()] to return +** [SQLITE_BUSY]. +** +** ^Calling this routine with an argument less than or equal to zero +** turns off all busy handlers. +** +** ^(There can only be a single busy handler for a particular +** [database connection] at any given moment. If another busy handler +** was defined (using [sqlite3_busy_handler()]) prior to calling +** this routine, that other busy handler is cleared.)^ +** +** See also: [PRAGMA busy_timeout] +*/ +SQLITE_API int sqlite3_busy_timeout(sqlite3*, int ms); + +/* +** CAPI3REF: Convenience Routines For Running Queries +** METHOD: sqlite3 +** +** This is a legacy interface that is preserved for backwards compatibility. +** Use of this interface is not recommended. +** +** Definition: A result table is memory data structure created by the +** [sqlite3_get_table()] interface. A result table records the +** complete query results from one or more queries. +** +** The table conceptually has a number of rows and columns. But +** these numbers are not part of the result table itself. These +** numbers are obtained separately. Let N be the number of rows +** and M be the number of columns. +** +** A result table is an array of pointers to zero-terminated UTF-8 strings. +** There are (N+1)*M elements in the array. The first M pointers point +** to zero-terminated strings that contain the names of the columns. +** The remaining entries all point to query results. NULL values result +** in NULL pointers. All other values are in their UTF-8 zero-terminated +** string representation as returned by [sqlite3_column_text()]. +** +** A result table might consist of one or more memory allocations. +** It is not safe to pass a result table directly to [sqlite3_free()]. +** A result table should be deallocated using [sqlite3_free_table()]. +** +** ^(As an example of the result table format, suppose a query result +** is as follows: +** +** 
+**        Name        | Age
+**        -----------------------
+**        Alice       | 43
+**        Bob         | 28
+**        Cindy       | 21
+**
+** +** There are two column (M==2) and three rows (N==3). Thus the +** result table has 8 entries. Suppose the result table is stored +** in an array names azResult. Then azResult holds this content: +** +** 
+**        azResult[0] = "Name";
+**        azResult[1] = "Age";
+**        azResult[2] = "Alice";
+**        azResult[3] = "43";
+**        azResult[4] = "Bob";
+**        azResult[5] = "28";
+**        azResult[6] = "Cindy";
+**        azResult[7] = "21";
+**
)^ +** +** ^The sqlite3_get_table() function evaluates one or more +** semicolon-separated SQL statements in the zero-terminated UTF-8 +** string of its 2nd parameter and returns a result table to the +** pointer given in its 3rd parameter. +** +** After the application has finished with the result from sqlite3_get_table(), +** it must pass the result table pointer to sqlite3_free_table() in order to +** release the memory that was malloced. Because of the way the +** [sqlite3_malloc()] happens within sqlite3_get_table(), the calling +** function must not try to call [sqlite3_free()] directly. Only +** [sqlite3_free_table()] is able to release the memory properly and safely. +** +** The sqlite3_get_table() interface is implemented as a wrapper around +** [sqlite3_exec()]. The sqlite3_get_table() routine does not have access +** to any internal data structures of SQLite. It uses only the public +** interface defined here. As a consequence, errors that occur in the +** wrapper layer outside of the internal [sqlite3_exec()] call are not +** reflected in subsequent calls to [sqlite3_errcode()] or +** [sqlite3_errmsg()]. +*/ +SQLITE_API int sqlite3_get_table( + sqlite3 *db, /* An open database */ + const char *zSql, /* SQL to be evaluated */ + char ***pazResult, /* Results of the query */ + int *pnRow, /* Number of result rows written here */ + int *pnColumn, /* Number of result columns written here */ + char **pzErrmsg /* Error msg written here */ +); +SQLITE_API void sqlite3_free_table(char **result); + +/* +** CAPI3REF: Formatted String Printing Functions +** +** These routines are work-alikes of the "printf()" family of functions +** from the standard C library. +** These routines understand most of the common formatting options from +** the standard library printf() +** plus some additional non-standard formats ([%q], [%Q], [%w], and [%z]). +** See the [built-in printf()] documentation for details. +** +** ^The sqlite3_mprintf() and sqlite3_vmprintf() routines write their +** results into memory obtained from [sqlite3_malloc64()]. +** The strings returned by these two routines should be +** released by [sqlite3_free()]. ^Both routines return a +** NULL pointer if [sqlite3_malloc64()] is unable to allocate enough +** memory to hold the resulting string. +** +** ^(The sqlite3_snprintf() routine is similar to "snprintf()" from +** the standard C library. The result is written into the +** buffer supplied as the second parameter whose size is given by +** the first parameter. Note that the order of the +** first two parameters is reversed from snprintf().)^ This is an +** historical accident that cannot be fixed without breaking +** backwards compatibility. ^(Note also that sqlite3_snprintf() +** returns a pointer to its buffer instead of the number of +** characters actually written into the buffer.)^ We admit that +** the number of characters written would be a more useful return +** value but we cannot change the implementation of sqlite3_snprintf() +** now without breaking compatibility. +** +** ^As long as the buffer size is greater than zero, sqlite3_snprintf() +** guarantees that the buffer is always zero-terminated. ^The first +** parameter "n" is the total size of the buffer, including space for +** the zero terminator. So the longest string that can be completely +** written will be n-1 characters. +** +** ^The sqlite3_vsnprintf() routine is a varargs version of sqlite3_snprintf(). +** +** See also: [built-in printf()], [printf() SQL function] +*/ +SQLITE_API char *sqlite3_mprintf(const char*,...); +SQLITE_API char *sqlite3_vmprintf(const char*, va_list); +SQLITE_API char *sqlite3_snprintf(int,char*,const char*, ...); +SQLITE_API char *sqlite3_vsnprintf(int,char*,const char*, va_list); + +/* +** CAPI3REF: Memory Allocation Subsystem +** +** The SQLite core uses these three routines for all of its own +** internal memory allocation needs. "Core" in the previous sentence +** does not include operating-system specific VFS implementation. The +** Windows VFS uses native malloc() and free() for some operations. +** +** ^The sqlite3_malloc() routine returns a pointer to a block +** of memory at least N bytes in length, where N is the parameter. +** ^If sqlite3_malloc() is unable to obtain sufficient free +** memory, it returns a NULL pointer. ^If the parameter N to +** sqlite3_malloc() is zero or negative then sqlite3_malloc() returns +** a NULL pointer. +** +** ^The sqlite3_malloc64(N) routine works just like +** sqlite3_malloc(N) except that N is an unsigned 64-bit integer instead +** of a signed 32-bit integer. +** +** ^Calling sqlite3_free() with a pointer previously returned +** by sqlite3_malloc() or sqlite3_realloc() releases that memory so +** that it might be reused. ^The sqlite3_free() routine is +** a no-op if is called with a NULL pointer. Passing a NULL pointer +** to sqlite3_free() is harmless. After being freed, memory +** should neither be read nor written. Even reading previously freed +** memory might result in a segmentation fault or other severe error. +** Memory corruption, a segmentation fault, or other severe error +** might result if sqlite3_free() is called with a non-NULL pointer that +** was not obtained from sqlite3_malloc() or sqlite3_realloc(). +** +** ^The sqlite3_realloc(X,N) interface attempts to resize a +** prior memory allocation X to be at least N bytes. +** ^If the X parameter to sqlite3_realloc(X,N) +** is a NULL pointer then its behavior is identical to calling +** sqlite3_malloc(N). +** ^If the N parameter to sqlite3_realloc(X,N) is zero or +** negative then the behavior is exactly the same as calling +** sqlite3_free(X). +** ^sqlite3_realloc(X,N) returns a pointer to a memory allocation +** of at least N bytes in size or NULL if insufficient memory is available. +** ^If M is the size of the prior allocation, then min(N,M) bytes +** of the prior allocation are copied into the beginning of buffer returned +** by sqlite3_realloc(X,N) and the prior allocation is freed. +** ^If sqlite3_realloc(X,N) returns NULL and N is positive, then the +** prior allocation is not freed. +** +** ^The sqlite3_realloc64(X,N) interfaces works the same as +** sqlite3_realloc(X,N) except that N is a 64-bit unsigned integer instead +** of a 32-bit signed integer. +** +** ^If X is a memory allocation previously obtained from sqlite3_malloc(), +** sqlite3_malloc64(), sqlite3_realloc(), or sqlite3_realloc64(), then +** sqlite3_msize(X) returns the size of that memory allocation in bytes. +** ^The value returned by sqlite3_msize(X) might be larger than the number +** of bytes requested when X was allocated. ^If X is a NULL pointer then +** sqlite3_msize(X) returns zero. If X points to something that is not +** the beginning of memory allocation, or if it points to a formerly +** valid memory allocation that has now been freed, then the behavior +** of sqlite3_msize(X) is undefined and possibly harmful. +** +** ^The memory returned by sqlite3_malloc(), sqlite3_realloc(), +** sqlite3_malloc64(), and sqlite3_realloc64() +** is always aligned to at least an 8 byte boundary, or to a +** 4 byte boundary if the [SQLITE_4_BYTE_ALIGNED_MALLOC] compile-time +** option is used. +** +** In SQLite version 3.5.0 and 3.5.1, it was possible to define +** the SQLITE_OMIT_MEMORY_ALLOCATION which would cause the built-in +** implementation of these routines to be omitted. That capability +** is no longer provided. Only built-in memory allocators can be used. +** +** Prior to SQLite version 3.7.10, the Windows OS interface layer called +** the system malloc() and free() directly when converting +** filenames between the UTF-8 encoding used by SQLite +** and whatever filename encoding is used by the particular Windows +** installation. Memory allocation errors were detected, but +** they were reported back as [SQLITE_CANTOPEN] or +** [SQLITE_IOERR] rather than [SQLITE_NOMEM]. +** +** The pointer arguments to [sqlite3_free()] and [sqlite3_realloc()] +** must be either NULL or else pointers obtained from a prior +** invocation of [sqlite3_malloc()] or [sqlite3_realloc()] that have +** not yet been released. +** +** The application must not read or write any part of +** a block of memory after it has been released using +** [sqlite3_free()] or [sqlite3_realloc()]. +*/ +SQLITE_API void *sqlite3_malloc(int); +SQLITE_API void *sqlite3_malloc64(sqlite3_uint64); +SQLITE_API void *sqlite3_realloc(void*, int); +SQLITE_API void *sqlite3_realloc64(void*, sqlite3_uint64); +SQLITE_API void sqlite3_free(void*); +SQLITE_API sqlite3_uint64 sqlite3_msize(void*); + +/* +** CAPI3REF: Memory Allocator Statistics +** +** SQLite provides these two interfaces for reporting on the status +** of the [sqlite3_malloc()], [sqlite3_free()], and [sqlite3_realloc()] +** routines, which form the built-in memory allocation subsystem. +** +** ^The [sqlite3_memory_used()] routine returns the number of bytes +** of memory currently outstanding (malloced but not freed). +** ^The [sqlite3_memory_highwater()] routine returns the maximum +** value of [sqlite3_memory_used()] since the high-water mark +** was last reset. ^The values returned by [sqlite3_memory_used()] and +** [sqlite3_memory_highwater()] include any overhead +** added by SQLite in its implementation of [sqlite3_malloc()], +** but not overhead added by the any underlying system library +** routines that [sqlite3_malloc()] may call. +** +** ^The memory high-water mark is reset to the current value of +** [sqlite3_memory_used()] if and only if the parameter to +** [sqlite3_memory_highwater()] is true. ^The value returned +** by [sqlite3_memory_highwater(1)] is the high-water mark +** prior to the reset. +*/ +SQLITE_API sqlite3_int64 sqlite3_memory_used(void); +SQLITE_API sqlite3_int64 sqlite3_memory_highwater(int resetFlag); + +/* +** CAPI3REF: Pseudo-Random Number Generator +** +** SQLite contains a high-quality pseudo-random number generator (PRNG) used to +** select random [ROWID | ROWIDs] when inserting new records into a table that +** already uses the largest possible [ROWID]. The PRNG is also used for +** the build-in random() and randomblob() SQL functions. This interface allows +** applications to access the same PRNG for other purposes. +** +** ^A call to this routine stores N bytes of randomness into buffer P. +** ^The P parameter can be a NULL pointer. +** +** ^If this routine has not been previously called or if the previous +** call had N less than one or a NULL pointer for P, then the PRNG is +** seeded using randomness obtained from the xRandomness method of +** the default [sqlite3_vfs] object. +** ^If the previous call to this routine had an N of 1 or more and a +** non-NULL P then the pseudo-randomness is generated +** internally and without recourse to the [sqlite3_vfs] xRandomness +** method. +*/ +SQLITE_API void sqlite3_randomness(int N, void *P); + +/* +** CAPI3REF: Compile-Time Authorization Callbacks +** METHOD: sqlite3 +** KEYWORDS: {authorizer callback} +** +** ^This routine registers an authorizer callback with a particular +** [database connection], supplied in the first argument. +** ^The authorizer callback is invoked as SQL statements are being compiled +** by [sqlite3_prepare()] or its variants [sqlite3_prepare_v2()], +** [sqlite3_prepare_v3()], [sqlite3_prepare16()], [sqlite3_prepare16_v2()], +** and [sqlite3_prepare16_v3()]. ^At various +** points during the compilation process, as logic is being created +** to perform various actions, the authorizer callback is invoked to +** see if those actions are allowed. ^The authorizer callback should +** return [SQLITE_OK] to allow the action, [SQLITE_IGNORE] to disallow the +** specific action but allow the SQL statement to continue to be +** compiled, or [SQLITE_DENY] to cause the entire SQL statement to be +** rejected with an error. ^If the authorizer callback returns +** any value other than [SQLITE_IGNORE], [SQLITE_OK], or [SQLITE_DENY] +** then the [sqlite3_prepare_v2()] or equivalent call that triggered +** the authorizer will fail with an error message. +** +** When the callback returns [SQLITE_OK], that means the operation +** requested is ok. ^When the callback returns [SQLITE_DENY], the +** [sqlite3_prepare_v2()] or equivalent call that triggered the +** authorizer will fail with an error message explaining that +** access is denied. +** +** ^The first parameter to the authorizer callback is a copy of the third +** parameter to the sqlite3_set_authorizer() interface. ^The second parameter +** to the callback is an integer [SQLITE_COPY | action code] that specifies +** the particular action to be authorized. ^The third through sixth parameters +** to the callback are either NULL pointers or zero-terminated strings +** that contain additional details about the action to be authorized. +** Applications must always be prepared to encounter a NULL pointer in any +** of the third through the sixth parameters of the authorization callback. +** +** ^If the action code is [SQLITE_READ] +** and the callback returns [SQLITE_IGNORE] then the +** [prepared statement] statement is constructed to substitute +** a NULL value in place of the table column that would have +** been read if [SQLITE_OK] had been returned. The [SQLITE_IGNORE] +** return can be used to deny an untrusted user access to individual +** columns of a table. +** ^When a table is referenced by a [SELECT] but no column values are +** extracted from that table (for example in a query like +** "SELECT count(*) FROM tab") then the [SQLITE_READ] authorizer callback +** is invoked once for that table with a column name that is an empty string. +** ^If the action code is [SQLITE_DELETE] and the callback returns +** [SQLITE_IGNORE] then the [DELETE] operation proceeds but the +** [truncate optimization] is disabled and all rows are deleted individually. +** +** An authorizer is used when [sqlite3_prepare | preparing] +** SQL statements from an untrusted source, to ensure that the SQL statements +** do not try to access data they are not allowed to see, or that they do not +** try to execute malicious statements that damage the database. For +** example, an application may allow a user to enter arbitrary +** SQL queries for evaluation by a database. But the application does +** not want the user to be able to make arbitrary changes to the +** database. An authorizer could then be put in place while the +** user-entered SQL is being [sqlite3_prepare | prepared] that +** disallows everything except [SELECT] statements. +** +** Applications that need to process SQL from untrusted sources +** might also consider lowering resource limits using [sqlite3_limit()] +** and limiting database size using the [max_page_count] [PRAGMA] +** in addition to using an authorizer. +** +** ^(Only a single authorizer can be in place on a database connection +** at a time. Each call to sqlite3_set_authorizer overrides the +** previous call.)^ ^Disable the authorizer by installing a NULL callback. +** The authorizer is disabled by default. +** +** The authorizer callback must not do anything that will modify +** the database connection that invoked the authorizer callback. +** Note that [sqlite3_prepare_v2()] and [sqlite3_step()] both modify their +** database connections for the meaning of "modify" in this paragraph. +** +** ^When [sqlite3_prepare_v2()] is used to prepare a statement, the +** statement might be re-prepared during [sqlite3_step()] due to a +** schema change. Hence, the application should ensure that the +** correct authorizer callback remains in place during the [sqlite3_step()]. +** +** ^Note that the authorizer callback is invoked only during +** [sqlite3_prepare()] or its variants. Authorization is not +** performed during statement evaluation in [sqlite3_step()], unless +** as stated in the previous paragraph, sqlite3_step() invokes +** sqlite3_prepare_v2() to reprepare a statement after a schema change. +*/ +SQLITE_API int sqlite3_set_authorizer( + sqlite3*, + int (*xAuth)(void*,int,const char*,const char*,const char*,const char*), + void *pUserData +); + +/* +** CAPI3REF: Authorizer Return Codes +** +** The [sqlite3_set_authorizer | authorizer callback function] must +** return either [SQLITE_OK] or one of these two constants in order +** to signal SQLite whether or not the action is permitted. See the +** [sqlite3_set_authorizer | authorizer documentation] for additional +** information. +** +** Note that SQLITE_IGNORE is also used as a [conflict resolution mode] +** returned from the [sqlite3_vtab_on_conflict()] interface. +*/ +#define SQLITE_DENY 1 /* Abort the SQL statement with an error */ +#define SQLITE_IGNORE 2 /* Don't allow access, but don't generate an error */ + +/* +** CAPI3REF: Authorizer Action Codes +** +** The [sqlite3_set_authorizer()] interface registers a callback function +** that is invoked to authorize certain SQL statement actions. The +** second parameter to the callback is an integer code that specifies +** what action is being authorized. These are the integer action codes that +** the authorizer callback may be passed. +** +** These action code values signify what kind of operation is to be +** authorized. The 3rd and 4th parameters to the authorization +** callback function will be parameters or NULL depending on which of these +** codes is used as the second parameter. ^(The 5th parameter to the +** authorizer callback is the name of the database ("main", "temp", +** etc.) if applicable.)^ ^The 6th parameter to the authorizer callback +** is the name of the inner-most trigger or view that is responsible for +** the access attempt or NULL if this access attempt is directly from +** top-level SQL code. +*/ +/******************************************* 3rd ************ 4th ***********/ +#define SQLITE_CREATE_INDEX 1 /* Index Name Table Name */ +#define SQLITE_CREATE_TABLE 2 /* Table Name NULL */ +#define SQLITE_CREATE_TEMP_INDEX 3 /* Index Name Table Name */ +#define SQLITE_CREATE_TEMP_TABLE 4 /* Table Name NULL */ +#define SQLITE_CREATE_TEMP_TRIGGER 5 /* Trigger Name Table Name */ +#define SQLITE_CREATE_TEMP_VIEW 6 /* View Name NULL */ +#define SQLITE_CREATE_TRIGGER 7 /* Trigger Name Table Name */ +#define SQLITE_CREATE_VIEW 8 /* View Name NULL */ +#define SQLITE_DELETE 9 /* Table Name NULL */ +#define SQLITE_DROP_INDEX 10 /* Index Name Table Name */ +#define SQLITE_DROP_TABLE 11 /* Table Name NULL */ +#define SQLITE_DROP_TEMP_INDEX 12 /* Index Name Table Name */ +#define SQLITE_DROP_TEMP_TABLE 13 /* Table Name NULL */ +#define SQLITE_DROP_TEMP_TRIGGER 14 /* Trigger Name Table Name */ +#define SQLITE_DROP_TEMP_VIEW 15 /* View Name NULL */ +#define SQLITE_DROP_TRIGGER 16 /* Trigger Name Table Name */ +#define SQLITE_DROP_VIEW 17 /* View Name NULL */ +#define SQLITE_INSERT 18 /* Table Name NULL */ +#define SQLITE_PRAGMA 19 /* Pragma Name 1st arg or NULL */ +#define SQLITE_READ 20 /* Table Name Column Name */ +#define SQLITE_SELECT 21 /* NULL NULL */ +#define SQLITE_TRANSACTION 22 /* Operation NULL */ +#define SQLITE_UPDATE 23 /* Table Name Column Name */ +#define SQLITE_ATTACH 24 /* Filename NULL */ +#define SQLITE_DETACH 25 /* Database Name NULL */ +#define SQLITE_ALTER_TABLE 26 /* Database Name Table Name */ +#define SQLITE_REINDEX 27 /* Index Name NULL */ +#define SQLITE_ANALYZE 28 /* Table Name NULL */ +#define SQLITE_CREATE_VTABLE 29 /* Table Name Module Name */ +#define SQLITE_DROP_VTABLE 30 /* Table Name Module Name */ +#define SQLITE_FUNCTION 31 /* NULL Function Name */ +#define SQLITE_SAVEPOINT 32 /* Operation Savepoint Name */ +#define SQLITE_COPY 0 /* No longer used */ +#define SQLITE_RECURSIVE 33 /* NULL NULL */ + +/* +** CAPI3REF: Tracing And Profiling Functions +** METHOD: sqlite3 +** +** These routines are deprecated. Use the [sqlite3_trace_v2()] interface +** instead of the routines described here. +** +** These routines register callback functions that can be used for +** tracing and profiling the execution of SQL statements. +** +** ^The callback function registered by sqlite3_trace() is invoked at +** various times when an SQL statement is being run by [sqlite3_step()]. +** ^The sqlite3_trace() callback is invoked with a UTF-8 rendering of the +** SQL statement text as the statement first begins executing. +** ^(Additional sqlite3_trace() callbacks might occur +** as each triggered subprogram is entered. The callbacks for triggers +** contain a UTF-8 SQL comment that identifies the trigger.)^ +** +** The [SQLITE_TRACE_SIZE_LIMIT] compile-time option can be used to limit +** the length of [bound parameter] expansion in the output of sqlite3_trace(). +** +** ^The callback function registered by sqlite3_profile() is invoked +** as each SQL statement finishes. ^The profile callback contains +** the original statement text and an estimate of wall-clock time +** of how long that statement took to run. ^The profile callback +** time is in units of nanoseconds, however the current implementation +** is only capable of millisecond resolution so the six least significant +** digits in the time are meaningless. Future versions of SQLite +** might provide greater resolution on the profiler callback. Invoking +** either [sqlite3_trace()] or [sqlite3_trace_v2()] will cancel the +** profile callback. +*/ +SQLITE_API SQLITE_DEPRECATED void *sqlite3_trace(sqlite3*, + void(*xTrace)(void*,const char*), void*); +SQLITE_API SQLITE_DEPRECATED void *sqlite3_profile(sqlite3*, + void(*xProfile)(void*,const char*,sqlite3_uint64), void*); + +/* +** CAPI3REF: SQL Trace Event Codes +** KEYWORDS: SQLITE_TRACE +** +** These constants identify classes of events that can be monitored +** using the [sqlite3_trace_v2()] tracing logic. The M argument +** to [sqlite3_trace_v2(D,M,X,P)] is an OR-ed combination of one or more of +** the following constants. ^The first argument to the trace callback +** is one of the following constants. +** +** New tracing constants may be added in future releases. +** +** ^A trace callback has four arguments: xCallback(T,C,P,X). +** ^The T argument is one of the integer type codes above. +** ^The C argument is a copy of the context pointer passed in as the +** fourth argument to [sqlite3_trace_v2()]. +** The P and X arguments are pointers whose meanings depend on T. +** +**
+** [[SQLITE_TRACE_STMT]]
SQLITE_TRACE_STMT
+**
^An SQLITE_TRACE_STMT callback is invoked when a prepared statement +** first begins running and possibly at other times during the +** execution of the prepared statement, such as at the start of each +** trigger subprogram. ^The P argument is a pointer to the +** [prepared statement]. ^The X argument is a pointer to a string which +** is the unexpanded SQL text of the prepared statement or an SQL comment +** that indicates the invocation of a trigger. ^The callback can compute +** the same text that would have been returned by the legacy [sqlite3_trace()] +** interface by using the X argument when X begins with "--" and invoking +** [sqlite3_expanded_sql(P)] otherwise. +** +** [[SQLITE_TRACE_PROFILE]]
SQLITE_TRACE_PROFILE
+**
^An SQLITE_TRACE_PROFILE callback provides approximately the same +** information as is provided by the [sqlite3_profile()] callback. +** ^The P argument is a pointer to the [prepared statement] and the +** X argument points to a 64-bit integer which is the estimated of +** the number of nanosecond that the prepared statement took to run. +** ^The SQLITE_TRACE_PROFILE callback is invoked when the statement finishes. +** +** [[SQLITE_TRACE_ROW]]
SQLITE_TRACE_ROW
+**
^An SQLITE_TRACE_ROW callback is invoked whenever a prepared +** statement generates a single row of result. +** ^The P argument is a pointer to the [prepared statement] and the +** X argument is unused. +** +** [[SQLITE_TRACE_CLOSE]]
SQLITE_TRACE_CLOSE
+**
^An SQLITE_TRACE_CLOSE callback is invoked when a database +** connection closes. +** ^The P argument is a pointer to the [database connection] object +** and the X argument is unused. +**
+*/ +#define SQLITE_TRACE_STMT 0x01 +#define SQLITE_TRACE_PROFILE 0x02 +#define SQLITE_TRACE_ROW 0x04 +#define SQLITE_TRACE_CLOSE 0x08 + +/* +** CAPI3REF: SQL Trace Hook +** METHOD: sqlite3 +** +** ^The sqlite3_trace_v2(D,M,X,P) interface registers a trace callback +** function X against [database connection] D, using property mask M +** and context pointer P. ^If the X callback is +** NULL or if the M mask is zero, then tracing is disabled. The +** M argument should be the bitwise OR-ed combination of +** zero or more [SQLITE_TRACE] constants. +** +** ^Each call to either sqlite3_trace() or sqlite3_trace_v2() overrides +** (cancels) any prior calls to sqlite3_trace() or sqlite3_trace_v2(). +** +** ^The X callback is invoked whenever any of the events identified by +** mask M occur. ^The integer return value from the callback is currently +** ignored, though this may change in future releases. Callback +** implementations should return zero to ensure future compatibility. +** +** ^A trace callback is invoked with four arguments: callback(T,C,P,X). +** ^The T argument is one of the [SQLITE_TRACE] +** constants to indicate why the callback was invoked. +** ^The C argument is a copy of the context pointer. +** The P and X arguments are pointers whose meanings depend on T. +** +** The sqlite3_trace_v2() interface is intended to replace the legacy +** interfaces [sqlite3_trace()] and [sqlite3_profile()], both of which +** are deprecated. +*/ +SQLITE_API int sqlite3_trace_v2( + sqlite3*, + unsigned uMask, + int(*xCallback)(unsigned,void*,void*,void*), + void *pCtx +); + +/* +** CAPI3REF: Query Progress Callbacks +** METHOD: sqlite3 +** +** ^The sqlite3_progress_handler(D,N,X,P) interface causes the callback +** function X to be invoked periodically during long running calls to +** [sqlite3_exec()], [sqlite3_step()] and [sqlite3_get_table()] for +** database connection D. An example use for this +** interface is to keep a GUI updated during a large query. +** +** ^The parameter P is passed through as the only parameter to the +** callback function X. ^The parameter N is the approximate number of +** [virtual machine instructions] that are evaluated between successive +** invocations of the callback X. ^If N is less than one then the progress +** handler is disabled. +** +** ^Only a single progress handler may be defined at one time per +** [database connection]; setting a new progress handler cancels the +** old one. ^Setting parameter X to NULL disables the progress handler. +** ^The progress handler is also disabled by setting N to a value less +** than 1. +** +** ^If the progress callback returns non-zero, the operation is +** interrupted. This feature can be used to implement a +** "Cancel" button on a GUI progress dialog box. +** +** The progress handler callback must not do anything that will modify +** the database connection that invoked the progress handler. +** Note that [sqlite3_prepare_v2()] and [sqlite3_step()] both modify their +** database connections for the meaning of "modify" in this paragraph. +** +*/ +SQLITE_API void sqlite3_progress_handler(sqlite3*, int, int(*)(void*), void*); + +/* +** CAPI3REF: Opening A New Database Connection +** CONSTRUCTOR: sqlite3 +** +** ^These routines open an SQLite database file as specified by the +** filename argument. ^The filename argument is interpreted as UTF-8 for +** sqlite3_open() and sqlite3_open_v2() and as UTF-16 in the native byte +** order for sqlite3_open16(). ^(A [database connection] handle is usually +** returned in *ppDb, even if an error occurs. The only exception is that +** if SQLite is unable to allocate memory to hold the [sqlite3] object, +** a NULL will be written into *ppDb instead of a pointer to the [sqlite3] +** object.)^ ^(If the database is opened (and/or created) successfully, then +** [SQLITE_OK] is returned. Otherwise an [error code] is returned.)^ ^The +** [sqlite3_errmsg()] or [sqlite3_errmsg16()] routines can be used to obtain +** an English language description of the error following a failure of any +** of the sqlite3_open() routines. +** +** ^The default encoding will be UTF-8 for databases created using +** sqlite3_open() or sqlite3_open_v2(). ^The default encoding for databases +** created using sqlite3_open16() will be UTF-16 in the native byte order. +** +** Whether or not an error occurs when it is opened, resources +** associated with the [database connection] handle should be released by +** passing it to [sqlite3_close()] when it is no longer required. +** +** The sqlite3_open_v2() interface works like sqlite3_open() +** except that it accepts two additional parameters for additional control +** over the new database connection. ^(The flags parameter to +** sqlite3_open_v2() can take one of +** the following three values, optionally combined with the +** [SQLITE_OPEN_NOMUTEX], [SQLITE_OPEN_FULLMUTEX], [SQLITE_OPEN_SHAREDCACHE], +** [SQLITE_OPEN_PRIVATECACHE], and/or [SQLITE_OPEN_URI] flags:)^ +** +**
+** ^(
[SQLITE_OPEN_READONLY]
+**
The database is opened in read-only mode. If the database does not +** already exist, an error is returned.
)^ +** +** ^(
[SQLITE_OPEN_READWRITE]
+**
The database is opened for reading and writing if possible, or reading +** only if the file is write protected by the operating system. In either +** case the database must already exist, otherwise an error is returned.
)^ +** +** ^(
[SQLITE_OPEN_READWRITE] | [SQLITE_OPEN_CREATE]
+**
The database is opened for reading and writing, and is created if +** it does not already exist. This is the behavior that is always used for +** sqlite3_open() and sqlite3_open16().
)^ +**
+** +** If the 3rd parameter to sqlite3_open_v2() is not one of the +** combinations shown above optionally combined with other +** [SQLITE_OPEN_READONLY | SQLITE_OPEN_* bits] +** then the behavior is undefined. +** +** ^If the [SQLITE_OPEN_NOMUTEX] flag is set, then the database connection +** opens in the multi-thread [threading mode] as long as the single-thread +** mode has not been set at compile-time or start-time. ^If the +** [SQLITE_OPEN_FULLMUTEX] flag is set then the database connection opens +** in the serialized [threading mode] unless single-thread was +** previously selected at compile-time or start-time. +** ^The [SQLITE_OPEN_SHAREDCACHE] flag causes the database connection to be +** eligible to use [shared cache mode], regardless of whether or not shared +** cache is enabled using [sqlite3_enable_shared_cache()]. ^The +** [SQLITE_OPEN_PRIVATECACHE] flag causes the database connection to not +** participate in [shared cache mode] even if it is enabled. +** +** ^The fourth parameter to sqlite3_open_v2() is the name of the +** [sqlite3_vfs] object that defines the operating system interface that +** the new database connection should use. ^If the fourth parameter is +** a NULL pointer then the default [sqlite3_vfs] object is used. +** +** ^If the filename is ":memory:", then a private, temporary in-memory database +** is created for the connection. ^This in-memory database will vanish when +** the database connection is closed. Future versions of SQLite might +** make use of additional special filenames that begin with the ":" character. +** It is recommended that when a database filename actually does begin with +** a ":" character you should prefix the filename with a pathname such as +** "./" to avoid ambiguity. +** +** ^If the filename is an empty string, then a private, temporary +** on-disk database will be created. ^This private database will be +** automatically deleted as soon as the database connection is closed. +** +** [[URI filenames in sqlite3_open()]]

URI Filenames

+** +** ^If [URI filename] interpretation is enabled, and the filename argument +** begins with "file:", then the filename is interpreted as a URI. ^URI +** filename interpretation is enabled if the [SQLITE_OPEN_URI] flag is +** set in the third argument to sqlite3_open_v2(), or if it has +** been enabled globally using the [SQLITE_CONFIG_URI] option with the +** [sqlite3_config()] method or by the [SQLITE_USE_URI] compile-time option. +** URI filename interpretation is turned off +** by default, but future releases of SQLite might enable URI filename +** interpretation by default. See "[URI filenames]" for additional +** information. +** +** URI filenames are parsed according to RFC 3986. ^If the URI contains an +** authority, then it must be either an empty string or the string +** "localhost". ^If the authority is not an empty string or "localhost", an +** error is returned to the caller. ^The fragment component of a URI, if +** present, is ignored. +** +** ^SQLite uses the path component of the URI as the name of the disk file +** which contains the database. ^If the path begins with a '/' character, +** then it is interpreted as an absolute path. ^If the path does not begin +** with a '/' (meaning that the authority section is omitted from the URI) +** then the path is interpreted as a relative path. +** ^(On windows, the first component of an absolute path +** is a drive specification (e.g. "C:").)^ +** +** [[core URI query parameters]] +** The query component of a URI may contain parameters that are interpreted +** either by SQLite itself, or by a [VFS | custom VFS implementation]. +** SQLite and its built-in [VFSes] interpret the +** following query parameters: +** +**
    +**
  • vfs: ^The "vfs" parameter may be used to specify the name of +** a VFS object that provides the operating system interface that should +** be used to access the database file on disk. ^If this option is set to +** an empty string the default VFS object is used. ^Specifying an unknown +** VFS is an error. ^If sqlite3_open_v2() is used and the vfs option is +** present, then the VFS specified by the option takes precedence over +** the value passed as the fourth parameter to sqlite3_open_v2(). +** +**
  • mode: ^(The mode parameter may be set to either "ro", "rw", +** "rwc", or "memory". Attempting to set it to any other value is +** an error)^. +** ^If "ro" is specified, then the database is opened for read-only +** access, just as if the [SQLITE_OPEN_READONLY] flag had been set in the +** third argument to sqlite3_open_v2(). ^If the mode option is set to +** "rw", then the database is opened for read-write (but not create) +** access, as if SQLITE_OPEN_READWRITE (but not SQLITE_OPEN_CREATE) had +** been set. ^Value "rwc" is equivalent to setting both +** SQLITE_OPEN_READWRITE and SQLITE_OPEN_CREATE. ^If the mode option is +** set to "memory" then a pure [in-memory database] that never reads +** or writes from disk is used. ^It is an error to specify a value for +** the mode parameter that is less restrictive than that specified by +** the flags passed in the third parameter to sqlite3_open_v2(). +** +**
  • cache: ^The cache parameter may be set to either "shared" or +** "private". ^Setting it to "shared" is equivalent to setting the +** SQLITE_OPEN_SHAREDCACHE bit in the flags argument passed to +** sqlite3_open_v2(). ^Setting the cache parameter to "private" is +** equivalent to setting the SQLITE_OPEN_PRIVATECACHE bit. +** ^If sqlite3_open_v2() is used and the "cache" parameter is present in +** a URI filename, its value overrides any behavior requested by setting +** SQLITE_OPEN_PRIVATECACHE or SQLITE_OPEN_SHAREDCACHE flag. +** +**
  • psow: ^The psow parameter indicates whether or not the +** [powersafe overwrite] property does or does not apply to the +** storage media on which the database file resides. +** +**
  • nolock: ^The nolock parameter is a boolean query parameter +** which if set disables file locking in rollback journal modes. This +** is useful for accessing a database on a filesystem that does not +** support locking. Caution: Database corruption might result if two +** or more processes write to the same database and any one of those +** processes uses nolock=1. +** +**
  • immutable: ^The immutable parameter is a boolean query +** parameter that indicates that the database file is stored on +** read-only media. ^When immutable is set, SQLite assumes that the +** database file cannot be changed, even by a process with higher +** privilege, and so the database is opened read-only and all locking +** and change detection is disabled. Caution: Setting the immutable +** property on a database file that does in fact change can result +** in incorrect query results and/or [SQLITE_CORRUPT] errors. +** See also: [SQLITE_IOCAP_IMMUTABLE]. +** +**
+** +** ^Specifying an unknown parameter in the query component of a URI is not an +** error. Future versions of SQLite might understand additional query +** parameters. See "[query parameters with special meaning to SQLite]" for +** additional information. +** +** [[URI filename examples]]

URI filename examples

+** +** +**
URI filenames Results +**
file:data.db +** Open the file "data.db" in the current directory. +**
file:/home/fred/data.db
+** file:///home/fred/data.db
+** file://localhost/home/fred/data.db
+** Open the database file "/home/fred/data.db". +**
file://darkstar/home/fred/data.db +** An error. "darkstar" is not a recognized authority. +**
+** file:///C:/Documents%20and%20Settings/fred/Desktop/data.db +** Windows only: Open the file "data.db" on fred's desktop on drive +** C:. Note that the %20 escaping in this example is not strictly +** necessary - space characters can be used literally +** in URI filenames. +**
file:data.db?mode=ro&cache=private +** Open file "data.db" in the current directory for read-only access. +** Regardless of whether or not shared-cache mode is enabled by +** default, use a private cache. +**
file:/home/fred/data.db?vfs=unix-dotfile +** Open file "/home/fred/data.db". Use the special VFS "unix-dotfile" +** that uses dot-files in place of posix advisory locking. +**
file:data.db?mode=readonly +** An error. "readonly" is not a valid option for the "mode" parameter. +**
+** +** ^URI hexadecimal escape sequences (%HH) are supported within the path and +** query components of a URI. A hexadecimal escape sequence consists of a +** percent sign - "%" - followed by exactly two hexadecimal digits +** specifying an octet value. ^Before the path or query components of a +** URI filename are interpreted, they are encoded using UTF-8 and all +** hexadecimal escape sequences replaced by a single byte containing the +** corresponding octet. If this process generates an invalid UTF-8 encoding, +** the results are undefined. +** +** Note to Windows users: The encoding used for the filename argument +** of sqlite3_open() and sqlite3_open_v2() must be UTF-8, not whatever +** codepage is currently defined. Filenames containing international +** characters must be converted to UTF-8 prior to passing them into +** sqlite3_open() or sqlite3_open_v2(). +** +** Note to Windows Runtime users: The temporary directory must be set +** prior to calling sqlite3_open() or sqlite3_open_v2(). Otherwise, various +** features that require the use of temporary files may fail. +** +** See also: [sqlite3_temp_directory] +*/ +SQLITE_API int sqlite3_open( + const char *filename, /* Database filename (UTF-8) */ + sqlite3 **ppDb /* OUT: SQLite db handle */ +); +SQLITE_API int sqlite3_open16( + const void *filename, /* Database filename (UTF-16) */ + sqlite3 **ppDb /* OUT: SQLite db handle */ +); +SQLITE_API int sqlite3_open_v2( + const char *filename, /* Database filename (UTF-8) */ + sqlite3 **ppDb, /* OUT: SQLite db handle */ + int flags, /* Flags */ + const char *zVfs /* Name of VFS module to use */ +); + +/* +** CAPI3REF: Obtain Values For URI Parameters +** +** These are utility routines, useful to VFS implementations, that check +** to see if a database file was a URI that contained a specific query +** parameter, and if so obtains the value of that query parameter. +** +** If F is the database filename pointer passed into the xOpen() method of +** a VFS implementation when the flags parameter to xOpen() has one or +** more of the [SQLITE_OPEN_URI] or [SQLITE_OPEN_MAIN_DB] bits set and +** P is the name of the query parameter, then +** sqlite3_uri_parameter(F,P) returns the value of the P +** parameter if it exists or a NULL pointer if P does not appear as a +** query parameter on F. If P is a query parameter of F +** has no explicit value, then sqlite3_uri_parameter(F,P) returns +** a pointer to an empty string. +** +** The sqlite3_uri_boolean(F,P,B) routine assumes that P is a boolean +** parameter and returns true (1) or false (0) according to the value +** of P. The sqlite3_uri_boolean(F,P,B) routine returns true (1) if the +** value of query parameter P is one of "yes", "true", or "on" in any +** case or if the value begins with a non-zero number. The +** sqlite3_uri_boolean(F,P,B) routines returns false (0) if the value of +** query parameter P is one of "no", "false", or "off" in any case or +** if the value begins with a numeric zero. If P is not a query +** parameter on F or if the value of P is does not match any of the +** above, then sqlite3_uri_boolean(F,P,B) returns (B!=0). +** +** The sqlite3_uri_int64(F,P,D) routine converts the value of P into a +** 64-bit signed integer and returns that integer, or D if P does not +** exist. If the value of P is something other than an integer, then +** zero is returned. +** +** If F is a NULL pointer, then sqlite3_uri_parameter(F,P) returns NULL and +** sqlite3_uri_boolean(F,P,B) returns B. If F is not a NULL pointer and +** is not a database file pathname pointer that SQLite passed into the xOpen +** VFS method, then the behavior of this routine is undefined and probably +** undesirable. +** +** See the [URI filename] documentation for additional information. +*/ +SQLITE_API const char *sqlite3_uri_parameter(const char *zFilename, const char *zParam); +SQLITE_API int sqlite3_uri_boolean(const char *zFile, const char *zParam, int bDefault); +SQLITE_API sqlite3_int64 sqlite3_uri_int64(const char*, const char*, sqlite3_int64); + + +/* +** CAPI3REF: Error Codes And Messages +** METHOD: sqlite3 +** +** ^If the most recent sqlite3_* API call associated with +** [database connection] D failed, then the sqlite3_errcode(D) interface +** returns the numeric [result code] or [extended result code] for that +** API call. +** ^The sqlite3_extended_errcode() +** interface is the same except that it always returns the +** [extended result code] even when extended result codes are +** disabled. +** +** The values returned by sqlite3_errcode() and/or +** sqlite3_extended_errcode() might change with each API call. +** Except, there are some interfaces that are guaranteed to never +** change the value of the error code. The error-code preserving +** interfaces are: +** +**
    +**
  • sqlite3_errcode() +**
  • sqlite3_extended_errcode() +**
  • sqlite3_errmsg() +**
  • sqlite3_errmsg16() +**
+** +** ^The sqlite3_errmsg() and sqlite3_errmsg16() return English-language +** text that describes the error, as either UTF-8 or UTF-16 respectively. +** ^(Memory to hold the error message string is managed internally. +** The application does not need to worry about freeing the result. +** However, the error string might be overwritten or deallocated by +** subsequent calls to other SQLite interface functions.)^ +** +** ^The sqlite3_errstr() interface returns the English-language text +** that describes the [result code], as UTF-8. +** ^(Memory to hold the error message string is managed internally +** and must not be freed by the application)^. +** +** When the serialized [threading mode] is in use, it might be the +** case that a second error occurs on a separate thread in between +** the time of the first error and the call to these interfaces. +** When that happens, the second error will be reported since these +** interfaces always report the most recent result. To avoid +** this, each thread can obtain exclusive use of the [database connection] D +** by invoking [sqlite3_mutex_enter]([sqlite3_db_mutex](D)) before beginning +** to use D and invoking [sqlite3_mutex_leave]([sqlite3_db_mutex](D)) after +** all calls to the interfaces listed here are completed. +** +** If an interface fails with SQLITE_MISUSE, that means the interface +** was invoked incorrectly by the application. In that case, the +** error code and message may or may not be set. +*/ +SQLITE_API int sqlite3_errcode(sqlite3 *db); +SQLITE_API int sqlite3_extended_errcode(sqlite3 *db); +SQLITE_API const char *sqlite3_errmsg(sqlite3*); +SQLITE_API const void *sqlite3_errmsg16(sqlite3*); +SQLITE_API const char *sqlite3_errstr(int); + +/* +** CAPI3REF: Prepared Statement Object +** KEYWORDS: {prepared statement} {prepared statements} +** +** An instance of this object represents a single SQL statement that +** has been compiled into binary form and is ready to be evaluated. +** +** Think of each SQL statement as a separate computer program. The +** original SQL text is source code. A prepared statement object +** is the compiled object code. All SQL must be converted into a +** prepared statement before it can be run. +** +** The life-cycle of a prepared statement object usually goes like this: +** +**
    +**
  1. Create the prepared statement object using [sqlite3_prepare_v2()]. +**
  2. Bind values to [parameters] using the sqlite3_bind_*() +** interfaces. +**
  3. Run the SQL by calling [sqlite3_step()] one or more times. +**
  4. Reset the prepared statement using [sqlite3_reset()] then go back +** to step 2. Do this zero or more times. +**
  5. Destroy the object using [sqlite3_finalize()]. +**
+*/ +typedef struct sqlite3_stmt sqlite3_stmt; + +/* +** CAPI3REF: Run-time Limits +** METHOD: sqlite3 +** +** ^(This interface allows the size of various constructs to be limited +** on a connection by connection basis. The first parameter is the +** [database connection] whose limit is to be set or queried. The +** second parameter is one of the [limit categories] that define a +** class of constructs to be size limited. The third parameter is the +** new limit for that construct.)^ +** +** ^If the new limit is a negative number, the limit is unchanged. +** ^(For each limit category SQLITE_LIMIT_NAME there is a +** [limits | hard upper bound] +** set at compile-time by a C preprocessor macro called +** [limits | SQLITE_MAX_NAME]. +** (The "_LIMIT_" in the name is changed to "_MAX_".))^ +** ^Attempts to increase a limit above its hard upper bound are +** silently truncated to the hard upper bound. +** +** ^Regardless of whether or not the limit was changed, the +** [sqlite3_limit()] interface returns the prior value of the limit. +** ^Hence, to find the current value of a limit without changing it, +** simply invoke this interface with the third parameter set to -1. +** +** Run-time limits are intended for use in applications that manage +** both their own internal database and also databases that are controlled +** by untrusted external sources. An example application might be a +** web browser that has its own databases for storing history and +** separate databases controlled by JavaScript applications downloaded +** off the Internet. The internal databases can be given the +** large, default limits. Databases managed by external sources can +** be given much smaller limits designed to prevent a denial of service +** attack. Developers might also want to use the [sqlite3_set_authorizer()] +** interface to further control untrusted SQL. The size of the database +** created by an untrusted script can be contained using the +** [max_page_count] [PRAGMA]. +** +** New run-time limit categories may be added in future releases. +*/ +SQLITE_API int sqlite3_limit(sqlite3*, int id, int newVal); + +/* +** CAPI3REF: Run-Time Limit Categories +** KEYWORDS: {limit category} {*limit categories} +** +** These constants define various performance limits +** that can be lowered at run-time using [sqlite3_limit()]. +** The synopsis of the meanings of the various limits is shown below. +** Additional information is available at [limits | Limits in SQLite]. +** +**
+** [[SQLITE_LIMIT_LENGTH]] ^(
SQLITE_LIMIT_LENGTH
+**
The maximum size of any string or BLOB or table row, in bytes.
)^ +** +** [[SQLITE_LIMIT_SQL_LENGTH]] ^(
SQLITE_LIMIT_SQL_LENGTH
+**
The maximum length of an SQL statement, in bytes.
)^ +** +** [[SQLITE_LIMIT_COLUMN]] ^(
SQLITE_LIMIT_COLUMN
+**
The maximum number of columns in a table definition or in the +** result set of a [SELECT] or the maximum number of columns in an index +** or in an ORDER BY or GROUP BY clause.
)^ +** +** [[SQLITE_LIMIT_EXPR_DEPTH]] ^(
SQLITE_LIMIT_EXPR_DEPTH
+**
The maximum depth of the parse tree on any expression.
)^ +** +** [[SQLITE_LIMIT_COMPOUND_SELECT]] ^(
SQLITE_LIMIT_COMPOUND_SELECT
+**
The maximum number of terms in a compound SELECT statement.
)^ +** +** [[SQLITE_LIMIT_VDBE_OP]] ^(
SQLITE_LIMIT_VDBE_OP
+**
The maximum number of instructions in a virtual machine program +** used to implement an SQL statement. If [sqlite3_prepare_v2()] or +** the equivalent tries to allocate space for more than this many opcodes +** in a single prepared statement, an SQLITE_NOMEM error is returned.
)^ +** +** [[SQLITE_LIMIT_FUNCTION_ARG]] ^(
SQLITE_LIMIT_FUNCTION_ARG
+**
The maximum number of arguments on a function.
)^ +** +** [[SQLITE_LIMIT_ATTACHED]] ^(
SQLITE_LIMIT_ATTACHED
+**
The maximum number of [ATTACH | attached databases].)^
+** +** [[SQLITE_LIMIT_LIKE_PATTERN_LENGTH]] +** ^(
SQLITE_LIMIT_LIKE_PATTERN_LENGTH
+**
The maximum length of the pattern argument to the [LIKE] or +** [GLOB] operators.
)^ +** +** [[SQLITE_LIMIT_VARIABLE_NUMBER]] +** ^(
SQLITE_LIMIT_VARIABLE_NUMBER
+**
The maximum index number of any [parameter] in an SQL statement.)^ +** +** [[SQLITE_LIMIT_TRIGGER_DEPTH]] ^(
SQLITE_LIMIT_TRIGGER_DEPTH
+**
The maximum depth of recursion for triggers.
)^ +** +** [[SQLITE_LIMIT_WORKER_THREADS]] ^(
SQLITE_LIMIT_WORKER_THREADS
+**
The maximum number of auxiliary worker threads that a single +** [prepared statement] may start.
)^ +**
+*/ +#define SQLITE_LIMIT_LENGTH 0 +#define SQLITE_LIMIT_SQL_LENGTH 1 +#define SQLITE_LIMIT_COLUMN 2 +#define SQLITE_LIMIT_EXPR_DEPTH 3 +#define SQLITE_LIMIT_COMPOUND_SELECT 4 +#define SQLITE_LIMIT_VDBE_OP 5 +#define SQLITE_LIMIT_FUNCTION_ARG 6 +#define SQLITE_LIMIT_ATTACHED 7 +#define SQLITE_LIMIT_LIKE_PATTERN_LENGTH 8 +#define SQLITE_LIMIT_VARIABLE_NUMBER 9 +#define SQLITE_LIMIT_TRIGGER_DEPTH 10 +#define SQLITE_LIMIT_WORKER_THREADS 11 + +/* +** CAPI3REF: Prepare Flags +** +** These constants define various flags that can be passed into +** "prepFlags" parameter of the [sqlite3_prepare_v3()] and +** [sqlite3_prepare16_v3()] interfaces. +** +** New flags may be added in future releases of SQLite. +** +**
+** [[SQLITE_PREPARE_PERSISTENT]] ^(
SQLITE_PREPARE_PERSISTENT
+**
The SQLITE_PREPARE_PERSISTENT flag is a hint to the query planner +** that the prepared statement will be retained for a long time and +** probably reused many times.)^ ^Without this flag, [sqlite3_prepare_v3()] +** and [sqlite3_prepare16_v3()] assume that the prepared statement will +** be used just once or at most a few times and then destroyed using +** [sqlite3_finalize()] relatively soon. The current implementation acts +** on this hint by avoiding the use of [lookaside memory] so as not to +** deplete the limited store of lookaside memory. Future versions of +** SQLite may act on this hint differently. +** +** [[SQLITE_PREPARE_NORMALIZE]]
SQLITE_PREPARE_NORMALIZE
+**
The SQLITE_PREPARE_NORMALIZE flag is a no-op. This flag used +** to be required for any prepared statement that wanted to use the +** [sqlite3_normalized_sql()] interface. However, the +** [sqlite3_normalized_sql()] interface is now available to all +** prepared statements, regardless of whether or not they use this +** flag. +** +** [[SQLITE_PREPARE_NO_VTAB]]
SQLITE_PREPARE_NO_VTAB
+**
The SQLITE_PREPARE_NO_VTAB flag causes the SQL compiler +** to return an error (error code SQLITE_ERROR) if the statement uses +** any virtual tables. +**
+*/ +#define SQLITE_PREPARE_PERSISTENT 0x01 +#define SQLITE_PREPARE_NORMALIZE 0x02 +#define SQLITE_PREPARE_NO_VTAB 0x04 + +/* +** CAPI3REF: Compiling An SQL Statement +** KEYWORDS: {SQL statement compiler} +** METHOD: sqlite3 +** CONSTRUCTOR: sqlite3_stmt +** +** To execute an SQL statement, it must first be compiled into a byte-code +** program using one of these routines. Or, in other words, these routines +** are constructors for the [prepared statement] object. +** +** The preferred routine to use is [sqlite3_prepare_v2()]. The +** [sqlite3_prepare()] interface is legacy and should be avoided. +** [sqlite3_prepare_v3()] has an extra "prepFlags" option that is used +** for special purposes. +** +** The use of the UTF-8 interfaces is preferred, as SQLite currently +** does all parsing using UTF-8. The UTF-16 interfaces are provided +** as a convenience. The UTF-16 interfaces work by converting the +** input text into UTF-8, then invoking the corresponding UTF-8 interface. +** +** The first argument, "db", is a [database connection] obtained from a +** prior successful call to [sqlite3_open()], [sqlite3_open_v2()] or +** [sqlite3_open16()]. The database connection must not have been closed. +** +** The second argument, "zSql", is the statement to be compiled, encoded +** as either UTF-8 or UTF-16. The sqlite3_prepare(), sqlite3_prepare_v2(), +** and sqlite3_prepare_v3() +** interfaces use UTF-8, and sqlite3_prepare16(), sqlite3_prepare16_v2(), +** and sqlite3_prepare16_v3() use UTF-16. +** +** ^If the nByte argument is negative, then zSql is read up to the +** first zero terminator. ^If nByte is positive, then it is the +** number of bytes read from zSql. ^If nByte is zero, then no prepared +** statement is generated. +** If the caller knows that the supplied string is nul-terminated, then +** there is a small performance advantage to passing an nByte parameter that +** is the number of bytes in the input string including +** the nul-terminator. +** +** ^If pzTail is not NULL then *pzTail is made to point to the first byte +** past the end of the first SQL statement in zSql. These routines only +** compile the first statement in zSql, so *pzTail is left pointing to +** what remains uncompiled. +** +** ^*ppStmt is left pointing to a compiled [prepared statement] that can be +** executed using [sqlite3_step()]. ^If there is an error, *ppStmt is set +** to NULL. ^If the input text contains no SQL (if the input is an empty +** string or a comment) then *ppStmt is set to NULL. +** The calling procedure is responsible for deleting the compiled +** SQL statement using [sqlite3_finalize()] after it has finished with it. +** ppStmt may not be NULL. +** +** ^On success, the sqlite3_prepare() family of routines return [SQLITE_OK]; +** otherwise an [error code] is returned. +** +** The sqlite3_prepare_v2(), sqlite3_prepare_v3(), sqlite3_prepare16_v2(), +** and sqlite3_prepare16_v3() interfaces are recommended for all new programs. +** The older interfaces (sqlite3_prepare() and sqlite3_prepare16()) +** are retained for backwards compatibility, but their use is discouraged. +** ^In the "vX" interfaces, the prepared statement +** that is returned (the [sqlite3_stmt] object) contains a copy of the +** original SQL text. This causes the [sqlite3_step()] interface to +** behave differently in three ways: +** +**
    +**
  1. +** ^If the database schema changes, instead of returning [SQLITE_SCHEMA] as it +** always used to do, [sqlite3_step()] will automatically recompile the SQL +** statement and try to run it again. As many as [SQLITE_MAX_SCHEMA_RETRY] +** retries will occur before sqlite3_step() gives up and returns an error. +**
    2. +** +**
  2. +** ^When an error occurs, [sqlite3_step()] will return one of the detailed +** [error codes] or [extended error codes]. ^The legacy behavior was that +** [sqlite3_step()] would only return a generic [SQLITE_ERROR] result code +** and the application would have to make a second call to [sqlite3_reset()] +** in order to find the underlying cause of the problem. With the "v2" prepare +** interfaces, the underlying reason for the error is returned immediately. +**
    3. +** +**
  3. +** ^If the specific value bound to [parameter | host parameter] in the +** WHERE clause might influence the choice of query plan for a statement, +** then the statement will be automatically recompiled, as if there had been +** a schema change, on the first [sqlite3_step()] call following any change +** to the [sqlite3_bind_text | bindings] of that [parameter]. +** ^The specific value of WHERE-clause [parameter] might influence the +** choice of query plan if the parameter is the left-hand side of a [LIKE] +** or [GLOB] operator or if the parameter is compared to an indexed column +** and the [SQLITE_ENABLE_STAT4] compile-time option is enabled. +**
    4. +**
+** +**

^sqlite3_prepare_v3() differs from sqlite3_prepare_v2() only in having +** the extra prepFlags parameter, which is a bit array consisting of zero or +** more of the [SQLITE_PREPARE_PERSISTENT|SQLITE_PREPARE_*] flags. ^The +** sqlite3_prepare_v2() interface works exactly the same as +** sqlite3_prepare_v3() with a zero prepFlags parameter. +*/ +SQLITE_API int sqlite3_prepare( + sqlite3 *db, /* Database handle */ + const char *zSql, /* SQL statement, UTF-8 encoded */ + int nByte, /* Maximum length of zSql in bytes. */ + sqlite3_stmt **ppStmt, /* OUT: Statement handle */ + const char **pzTail /* OUT: Pointer to unused portion of zSql */ +); +SQLITE_API int sqlite3_prepare_v2( + sqlite3 *db, /* Database handle */ + const char *zSql, /* SQL statement, UTF-8 encoded */ + int nByte, /* Maximum length of zSql in bytes. */ + sqlite3_stmt **ppStmt, /* OUT: Statement handle */ + const char **pzTail /* OUT: Pointer to unused portion of zSql */ +); +SQLITE_API int sqlite3_prepare_v3( + sqlite3 *db, /* Database handle */ + const char *zSql, /* SQL statement, UTF-8 encoded */ + int nByte, /* Maximum length of zSql in bytes. */ + unsigned int prepFlags, /* Zero or more SQLITE_PREPARE_ flags */ + sqlite3_stmt **ppStmt, /* OUT: Statement handle */ + const char **pzTail /* OUT: Pointer to unused portion of zSql */ +); +SQLITE_API int sqlite3_prepare16( + sqlite3 *db, /* Database handle */ + const void *zSql, /* SQL statement, UTF-16 encoded */ + int nByte, /* Maximum length of zSql in bytes. */ + sqlite3_stmt **ppStmt, /* OUT: Statement handle */ + const void **pzTail /* OUT: Pointer to unused portion of zSql */ +); +SQLITE_API int sqlite3_prepare16_v2( + sqlite3 *db, /* Database handle */ + const void *zSql, /* SQL statement, UTF-16 encoded */ + int nByte, /* Maximum length of zSql in bytes. */ + sqlite3_stmt **ppStmt, /* OUT: Statement handle */ + const void **pzTail /* OUT: Pointer to unused portion of zSql */ +); +SQLITE_API int sqlite3_prepare16_v3( + sqlite3 *db, /* Database handle */ + const void *zSql, /* SQL statement, UTF-16 encoded */ + int nByte, /* Maximum length of zSql in bytes. */ + unsigned int prepFlags, /* Zero or more SQLITE_PREPARE_ flags */ + sqlite3_stmt **ppStmt, /* OUT: Statement handle */ + const void **pzTail /* OUT: Pointer to unused portion of zSql */ +); + +/* +** CAPI3REF: Retrieving Statement SQL +** METHOD: sqlite3_stmt +** +** ^The sqlite3_sql(P) interface returns a pointer to a copy of the UTF-8 +** SQL text used to create [prepared statement] P if P was +** created by [sqlite3_prepare_v2()], [sqlite3_prepare_v3()], +** [sqlite3_prepare16_v2()], or [sqlite3_prepare16_v3()]. +** ^The sqlite3_expanded_sql(P) interface returns a pointer to a UTF-8 +** string containing the SQL text of prepared statement P with +** [bound parameters] expanded. +** ^The sqlite3_normalized_sql(P) interface returns a pointer to a UTF-8 +** string containing the normalized SQL text of prepared statement P. The +** semantics used to normalize a SQL statement are unspecified and subject +** to change. At a minimum, literal values will be replaced with suitable +** placeholders. +** +** ^(For example, if a prepared statement is created using the SQL +** text "SELECT $abc,:xyz" and if parameter $abc is bound to integer 2345 +** and parameter :xyz is unbound, then sqlite3_sql() will return +** the original string, "SELECT $abc,:xyz" but sqlite3_expanded_sql() +** will return "SELECT 2345,NULL".)^ +** +** ^The sqlite3_expanded_sql() interface returns NULL if insufficient memory +** is available to hold the result, or if the result would exceed the +** the maximum string length determined by the [SQLITE_LIMIT_LENGTH]. +** +** ^The [SQLITE_TRACE_SIZE_LIMIT] compile-time option limits the size of +** bound parameter expansions. ^The [SQLITE_OMIT_TRACE] compile-time +** option causes sqlite3_expanded_sql() to always return NULL. +** +** ^The strings returned by sqlite3_sql(P) and sqlite3_normalized_sql(P) +** are managed by SQLite and are automatically freed when the prepared +** statement is finalized. +** ^The string returned by sqlite3_expanded_sql(P), on the other hand, +** is obtained from [sqlite3_malloc()] and must be free by the application +** by passing it to [sqlite3_free()]. +*/ +SQLITE_API const char *sqlite3_sql(sqlite3_stmt *pStmt); +SQLITE_API char *sqlite3_expanded_sql(sqlite3_stmt *pStmt); +SQLITE_API const char *sqlite3_normalized_sql(sqlite3_stmt *pStmt); + +/* +** CAPI3REF: Determine If An SQL Statement Writes The Database +** METHOD: sqlite3_stmt +** +** ^The sqlite3_stmt_readonly(X) interface returns true (non-zero) if +** and only if the [prepared statement] X makes no direct changes to +** the content of the database file. +** +** Note that [application-defined SQL functions] or +** [virtual tables] might change the database indirectly as a side effect. +** ^(For example, if an application defines a function "eval()" that +** calls [sqlite3_exec()], then the following SQL statement would +** change the database file through side-effects: +** +** 

+**    SELECT eval('DELETE FROM t1') FROM t2;
+**
+** +** But because the [SELECT] statement does not change the database file +** directly, sqlite3_stmt_readonly() would still return true.)^ +** +** ^Transaction control statements such as [BEGIN], [COMMIT], [ROLLBACK], +** [SAVEPOINT], and [RELEASE] cause sqlite3_stmt_readonly() to return true, +** since the statements themselves do not actually modify the database but +** rather they control the timing of when other statements modify the +** database. ^The [ATTACH] and [DETACH] statements also cause +** sqlite3_stmt_readonly() to return true since, while those statements +** change the configuration of a database connection, they do not make +** changes to the content of the database files on disk. +** ^The sqlite3_stmt_readonly() interface returns true for [BEGIN] since +** [BEGIN] merely sets internal flags, but the [BEGIN|BEGIN IMMEDIATE] and +** [BEGIN|BEGIN EXCLUSIVE] commands do touch the database and so +** sqlite3_stmt_readonly() returns false for those commands. +*/ +SQLITE_API int sqlite3_stmt_readonly(sqlite3_stmt *pStmt); + +/* +** CAPI3REF: Query The EXPLAIN Setting For A Prepared Statement +** METHOD: sqlite3_stmt +** +** ^The sqlite3_stmt_isexplain(S) interface returns 1 if the +** prepared statement S is an EXPLAIN statement, or 2 if the +** statement S is an EXPLAIN QUERY PLAN. +** ^The sqlite3_stmt_isexplain(S) interface returns 0 if S is +** an ordinary statement or a NULL pointer. +*/ +SQLITE_API int sqlite3_stmt_isexplain(sqlite3_stmt *pStmt); + +/* +** CAPI3REF: Determine If A Prepared Statement Has Been Reset +** METHOD: sqlite3_stmt +** +** ^The sqlite3_stmt_busy(S) interface returns true (non-zero) if the +** [prepared statement] S has been stepped at least once using +** [sqlite3_step(S)] but has neither run to completion (returned +** [SQLITE_DONE] from [sqlite3_step(S)]) nor +** been reset using [sqlite3_reset(S)]. ^The sqlite3_stmt_busy(S) +** interface returns false if S is a NULL pointer. If S is not a +** NULL pointer and is not a pointer to a valid [prepared statement] +** object, then the behavior is undefined and probably undesirable. +** +** This interface can be used in combination [sqlite3_next_stmt()] +** to locate all prepared statements associated with a database +** connection that are in need of being reset. This can be used, +** for example, in diagnostic routines to search for prepared +** statements that are holding a transaction open. +*/ +SQLITE_API int sqlite3_stmt_busy(sqlite3_stmt*); + +/* +** CAPI3REF: Dynamically Typed Value Object +** KEYWORDS: {protected sqlite3_value} {unprotected sqlite3_value} +** +** SQLite uses the sqlite3_value object to represent all values +** that can be stored in a database table. SQLite uses dynamic typing +** for the values it stores. ^Values stored in sqlite3_value objects +** can be integers, floating point values, strings, BLOBs, or NULL. +** +** An sqlite3_value object may be either "protected" or "unprotected". +** Some interfaces require a protected sqlite3_value. Other interfaces +** will accept either a protected or an unprotected sqlite3_value. +** Every interface that accepts sqlite3_value arguments specifies +** whether or not it requires a protected sqlite3_value. The +** [sqlite3_value_dup()] interface can be used to construct a new +** protected sqlite3_value from an unprotected sqlite3_value. +** +** The terms "protected" and "unprotected" refer to whether or not +** a mutex is held. An internal mutex is held for a protected +** sqlite3_value object but no mutex is held for an unprotected +** sqlite3_value object. If SQLite is compiled to be single-threaded +** (with [SQLITE_THREADSAFE=0] and with [sqlite3_threadsafe()] returning 0) +** or if SQLite is run in one of reduced mutex modes +** [SQLITE_CONFIG_SINGLETHREAD] or [SQLITE_CONFIG_MULTITHREAD] +** then there is no distinction between protected and unprotected +** sqlite3_value objects and they can be used interchangeably. However, +** for maximum code portability it is recommended that applications +** still make the distinction between protected and unprotected +** sqlite3_value objects even when not strictly required. +** +** ^The sqlite3_value objects that are passed as parameters into the +** implementation of [application-defined SQL functions] are protected. +** ^The sqlite3_value object returned by +** [sqlite3_column_value()] is unprotected. +** Unprotected sqlite3_value objects may only be used as arguments +** to [sqlite3_result_value()], [sqlite3_bind_value()], and +** [sqlite3_value_dup()]. +** The [sqlite3_value_blob | sqlite3_value_type()] family of +** interfaces require protected sqlite3_value objects. +*/ +typedef struct sqlite3_value sqlite3_value; + +/* +** CAPI3REF: SQL Function Context Object +** +** The context in which an SQL function executes is stored in an +** sqlite3_context object. ^A pointer to an sqlite3_context object +** is always first parameter to [application-defined SQL functions]. +** The application-defined SQL function implementation will pass this +** pointer through into calls to [sqlite3_result_int | sqlite3_result()], +** [sqlite3_aggregate_context()], [sqlite3_user_data()], +** [sqlite3_context_db_handle()], [sqlite3_get_auxdata()], +** and/or [sqlite3_set_auxdata()]. +*/ +typedef struct sqlite3_context sqlite3_context; + +/* +** CAPI3REF: Binding Values To Prepared Statements +** KEYWORDS: {host parameter} {host parameters} {host parameter name} +** KEYWORDS: {SQL parameter} {SQL parameters} {parameter binding} +** METHOD: sqlite3_stmt +** +** ^(In the SQL statement text input to [sqlite3_prepare_v2()] and its variants, +** literals may be replaced by a [parameter] that matches one of following +** templates: +** +**
    +**
  • ? +**
  • ?NNN +**
  • :VVV +**
  • @VVV +**
  • $VVV +**
+** +** In the templates above, NNN represents an integer literal, +** and VVV represents an alphanumeric identifier.)^ ^The values of these +** parameters (also called "host parameter names" or "SQL parameters") +** can be set using the sqlite3_bind_*() routines defined here. +** +** ^The first argument to the sqlite3_bind_*() routines is always +** a pointer to the [sqlite3_stmt] object returned from +** [sqlite3_prepare_v2()] or its variants. +** +** ^The second argument is the index of the SQL parameter to be set. +** ^The leftmost SQL parameter has an index of 1. ^When the same named +** SQL parameter is used more than once, second and subsequent +** occurrences have the same index as the first occurrence. +** ^The index for named parameters can be looked up using the +** [sqlite3_bind_parameter_index()] API if desired. ^The index +** for "?NNN" parameters is the value of NNN. +** ^The NNN value must be between 1 and the [sqlite3_limit()] +** parameter [SQLITE_LIMIT_VARIABLE_NUMBER] (default value: 999). +** +** ^The third argument is the value to bind to the parameter. +** ^If the third parameter to sqlite3_bind_text() or sqlite3_bind_text16() +** or sqlite3_bind_blob() is a NULL pointer then the fourth parameter +** is ignored and the end result is the same as sqlite3_bind_null(). +** +** ^(In those routines that have a fourth argument, its value is the +** number of bytes in the parameter. To be clear: the value is the +** number of bytes in the value, not the number of characters.)^ +** ^If the fourth parameter to sqlite3_bind_text() or sqlite3_bind_text16() +** is negative, then the length of the string is +** the number of bytes up to the first zero terminator. +** If the fourth parameter to sqlite3_bind_blob() is negative, then +** the behavior is undefined. +** If a non-negative fourth parameter is provided to sqlite3_bind_text() +** or sqlite3_bind_text16() or sqlite3_bind_text64() then +** that parameter must be the byte offset +** where the NUL terminator would occur assuming the string were NUL +** terminated. If any NUL characters occur at byte offsets less than +** the value of the fourth parameter then the resulting string value will +** contain embedded NULs. The result of expressions involving strings +** with embedded NULs is undefined. +** +** ^The fifth argument to the BLOB and string binding interfaces +** is a destructor used to dispose of the BLOB or +** string after SQLite has finished with it. ^The destructor is called +** to dispose of the BLOB or string even if the call to the bind API fails, +** except the destructor is not called if the third parameter is a NULL +** pointer or the fourth parameter is negative. +** ^If the fifth argument is +** the special value [SQLITE_STATIC], then SQLite assumes that the +** information is in static, unmanaged space and does not need to be freed. +** ^If the fifth argument has the value [SQLITE_TRANSIENT], then +** SQLite makes its own private copy of the data immediately, before +** the sqlite3_bind_*() routine returns. +** +** ^The sixth argument to sqlite3_bind_text64() must be one of +** [SQLITE_UTF8], [SQLITE_UTF16], [SQLITE_UTF16BE], or [SQLITE_UTF16LE] +** to specify the encoding of the text in the third parameter. If +** the sixth argument to sqlite3_bind_text64() is not one of the +** allowed values shown above, or if the text encoding is different +** from the encoding specified by the sixth parameter, then the behavior +** is undefined. +** +** ^The sqlite3_bind_zeroblob() routine binds a BLOB of length N that +** is filled with zeroes. ^A zeroblob uses a fixed amount of memory +** (just an integer to hold its size) while it is being processed. +** Zeroblobs are intended to serve as placeholders for BLOBs whose +** content is later written using +** [sqlite3_blob_open | incremental BLOB I/O] routines. +** ^A negative value for the zeroblob results in a zero-length BLOB. +** +** ^The sqlite3_bind_pointer(S,I,P,T,D) routine causes the I-th parameter in +** [prepared statement] S to have an SQL value of NULL, but to also be +** associated with the pointer P of type T. ^D is either a NULL pointer or +** a pointer to a destructor function for P. ^SQLite will invoke the +** destructor D with a single argument of P when it is finished using +** P. The T parameter should be a static string, preferably a string +** literal. The sqlite3_bind_pointer() routine is part of the +** [pointer passing interface] added for SQLite 3.20.0. +** +** ^If any of the sqlite3_bind_*() routines are called with a NULL pointer +** for the [prepared statement] or with a prepared statement for which +** [sqlite3_step()] has been called more recently than [sqlite3_reset()], +** then the call will return [SQLITE_MISUSE]. If any sqlite3_bind_() +** routine is passed a [prepared statement] that has been finalized, the +** result is undefined and probably harmful. +** +** ^Bindings are not cleared by the [sqlite3_reset()] routine. +** ^Unbound parameters are interpreted as NULL. +** +** ^The sqlite3_bind_* routines return [SQLITE_OK] on success or an +** [error code] if anything goes wrong. +** ^[SQLITE_TOOBIG] might be returned if the size of a string or BLOB +** exceeds limits imposed by [sqlite3_limit]([SQLITE_LIMIT_LENGTH]) or +** [SQLITE_MAX_LENGTH]. +** ^[SQLITE_RANGE] is returned if the parameter +** index is out of range. ^[SQLITE_NOMEM] is returned if malloc() fails. +** +** See also: [sqlite3_bind_parameter_count()], +** [sqlite3_bind_parameter_name()], and [sqlite3_bind_parameter_index()]. +*/ +SQLITE_API int sqlite3_bind_blob(sqlite3_stmt*, int, const void*, int n, void(*)(void*)); +SQLITE_API int sqlite3_bind_blob64(sqlite3_stmt*, int, const void*, sqlite3_uint64, + void(*)(void*)); +SQLITE_API int sqlite3_bind_double(sqlite3_stmt*, int, double); +SQLITE_API int sqlite3_bind_int(sqlite3_stmt*, int, int); +SQLITE_API int sqlite3_bind_int64(sqlite3_stmt*, int, sqlite3_int64); +SQLITE_API int sqlite3_bind_null(sqlite3_stmt*, int); +SQLITE_API int sqlite3_bind_text(sqlite3_stmt*,int,const char*,int,void(*)(void*)); +SQLITE_API int sqlite3_bind_text16(sqlite3_stmt*, int, const void*, int, void(*)(void*)); +SQLITE_API int sqlite3_bind_text64(sqlite3_stmt*, int, const char*, sqlite3_uint64, + void(*)(void*), unsigned char encoding); +SQLITE_API int sqlite3_bind_value(sqlite3_stmt*, int, const sqlite3_value*); +SQLITE_API int sqlite3_bind_pointer(sqlite3_stmt*, int, void*, const char*,void(*)(void*)); +SQLITE_API int sqlite3_bind_zeroblob(sqlite3_stmt*, int, int n); +SQLITE_API int sqlite3_bind_zeroblob64(sqlite3_stmt*, int, sqlite3_uint64); + +/* +** CAPI3REF: Number Of SQL Parameters +** METHOD: sqlite3_stmt +** +** ^This routine can be used to find the number of [SQL parameters] +** in a [prepared statement]. SQL parameters are tokens of the +** form "?", "?NNN", ":AAA", "$AAA", or "@AAA" that serve as +** placeholders for values that are [sqlite3_bind_blob | bound] +** to the parameters at a later time. +** +** ^(This routine actually returns the index of the largest (rightmost) +** parameter. For all forms except ?NNN, this will correspond to the +** number of unique parameters. If parameters of the ?NNN form are used, +** there may be gaps in the list.)^ +** +** See also: [sqlite3_bind_blob|sqlite3_bind()], +** [sqlite3_bind_parameter_name()], and +** [sqlite3_bind_parameter_index()]. +*/ +SQLITE_API int sqlite3_bind_parameter_count(sqlite3_stmt*); + +/* +** CAPI3REF: Name Of A Host Parameter +** METHOD: sqlite3_stmt +** +** ^The sqlite3_bind_parameter_name(P,N) interface returns +** the name of the N-th [SQL parameter] in the [prepared statement] P. +** ^(SQL parameters of the form "?NNN" or ":AAA" or "@AAA" or "$AAA" +** have a name which is the string "?NNN" or ":AAA" or "@AAA" or "$AAA" +** respectively. +** In other words, the initial ":" or "$" or "@" or "?" +** is included as part of the name.)^ +** ^Parameters of the form "?" without a following integer have no name +** and are referred to as "nameless" or "anonymous parameters". +** +** ^The first host parameter has an index of 1, not 0. +** +** ^If the value N is out of range or if the N-th parameter is +** nameless, then NULL is returned. ^The returned string is +** always in UTF-8 encoding even if the named parameter was +** originally specified as UTF-16 in [sqlite3_prepare16()], +** [sqlite3_prepare16_v2()], or [sqlite3_prepare16_v3()]. +** +** See also: [sqlite3_bind_blob|sqlite3_bind()], +** [sqlite3_bind_parameter_count()], and +** [sqlite3_bind_parameter_index()]. +*/ +SQLITE_API const char *sqlite3_bind_parameter_name(sqlite3_stmt*, int); + +/* +** CAPI3REF: Index Of A Parameter With A Given Name +** METHOD: sqlite3_stmt +** +** ^Return the index of an SQL parameter given its name. ^The +** index value returned is suitable for use as the second +** parameter to [sqlite3_bind_blob|sqlite3_bind()]. ^A zero +** is returned if no matching parameter is found. ^The parameter +** name must be given in UTF-8 even if the original statement +** was prepared from UTF-16 text using [sqlite3_prepare16_v2()] or +** [sqlite3_prepare16_v3()]. +** +** See also: [sqlite3_bind_blob|sqlite3_bind()], +** [sqlite3_bind_parameter_count()], and +** [sqlite3_bind_parameter_name()]. +*/ +SQLITE_API int sqlite3_bind_parameter_index(sqlite3_stmt*, const char *zName); + +/* +** CAPI3REF: Reset All Bindings On A Prepared Statement +** METHOD: sqlite3_stmt +** +** ^Contrary to the intuition of many, [sqlite3_reset()] does not reset +** the [sqlite3_bind_blob | bindings] on a [prepared statement]. +** ^Use this routine to reset all host parameters to NULL. +*/ +SQLITE_API int sqlite3_clear_bindings(sqlite3_stmt*); + +/* +** CAPI3REF: Number Of Columns In A Result Set +** METHOD: sqlite3_stmt +** +** ^Return the number of columns in the result set returned by the +** [prepared statement]. ^If this routine returns 0, that means the +** [prepared statement] returns no data (for example an [UPDATE]). +** ^However, just because this routine returns a positive number does not +** mean that one or more rows of data will be returned. ^A SELECT statement +** will always have a positive sqlite3_column_count() but depending on the +** WHERE clause constraints and the table content, it might return no rows. +** +** See also: [sqlite3_data_count()] +*/ +SQLITE_API int sqlite3_column_count(sqlite3_stmt *pStmt); + +/* +** CAPI3REF: Column Names In A Result Set +** METHOD: sqlite3_stmt +** +** ^These routines return the name assigned to a particular column +** in the result set of a [SELECT] statement. ^The sqlite3_column_name() +** interface returns a pointer to a zero-terminated UTF-8 string +** and sqlite3_column_name16() returns a pointer to a zero-terminated +** UTF-16 string. ^The first parameter is the [prepared statement] +** that implements the [SELECT] statement. ^The second parameter is the +** column number. ^The leftmost column is number 0. +** +** ^The returned string pointer is valid until either the [prepared statement] +** is destroyed by [sqlite3_finalize()] or until the statement is automatically +** reprepared by the first call to [sqlite3_step()] for a particular run +** or until the next call to +** sqlite3_column_name() or sqlite3_column_name16() on the same column. +** +** ^If sqlite3_malloc() fails during the processing of either routine +** (for example during a conversion from UTF-8 to UTF-16) then a +** NULL pointer is returned. +** +** ^The name of a result column is the value of the "AS" clause for +** that column, if there is an AS clause. If there is no AS clause +** then the name of the column is unspecified and may change from +** one release of SQLite to the next. +*/ +SQLITE_API const char *sqlite3_column_name(sqlite3_stmt*, int N); +SQLITE_API const void *sqlite3_column_name16(sqlite3_stmt*, int N); + +/* +** CAPI3REF: Source Of Data In A Query Result +** METHOD: sqlite3_stmt +** +** ^These routines provide a means to determine the database, table, and +** table column that is the origin of a particular result column in +** [SELECT] statement. +** ^The name of the database or table or column can be returned as +** either a UTF-8 or UTF-16 string. ^The _database_ routines return +** the database name, the _table_ routines return the table name, and +** the origin_ routines return the column name. +** ^The returned string is valid until the [prepared statement] is destroyed +** using [sqlite3_finalize()] or until the statement is automatically +** reprepared by the first call to [sqlite3_step()] for a particular run +** or until the same information is requested +** again in a different encoding. +** +** ^The names returned are the original un-aliased names of the +** database, table, and column. +** +** ^The first argument to these interfaces is a [prepared statement]. +** ^These functions return information about the Nth result column returned by +** the statement, where N is the second function argument. +** ^The left-most column is column 0 for these routines. +** +** ^If the Nth column returned by the statement is an expression or +** subquery and is not a column value, then all of these functions return +** NULL. ^These routine might also return NULL if a memory allocation error +** occurs. ^Otherwise, they return the name of the attached database, table, +** or column that query result column was extracted from. +** +** ^As with all other SQLite APIs, those whose names end with "16" return +** UTF-16 encoded strings and the other functions return UTF-8. +** +** ^These APIs are only available if the library was compiled with the +** [SQLITE_ENABLE_COLUMN_METADATA] C-preprocessor symbol. +** +** If two or more threads call one or more of these routines against the same +** prepared statement and column at the same time then the results are +** undefined. +** +** If two or more threads call one or more +** [sqlite3_column_database_name | column metadata interfaces] +** for the same [prepared statement] and result column +** at the same time then the results are undefined. +*/ +SQLITE_API const char *sqlite3_column_database_name(sqlite3_stmt*,int); +SQLITE_API const void *sqlite3_column_database_name16(sqlite3_stmt*,int); +SQLITE_API const char *sqlite3_column_table_name(sqlite3_stmt*,int); +SQLITE_API const void *sqlite3_column_table_name16(sqlite3_stmt*,int); +SQLITE_API const char *sqlite3_column_origin_name(sqlite3_stmt*,int); +SQLITE_API const void *sqlite3_column_origin_name16(sqlite3_stmt*,int); + +/* +** CAPI3REF: Declared Datatype Of A Query Result +** METHOD: sqlite3_stmt +** +** ^(The first parameter is a [prepared statement]. +** If this statement is a [SELECT] statement and the Nth column of the +** returned result set of that [SELECT] is a table column (not an +** expression or subquery) then the declared type of the table +** column is returned.)^ ^If the Nth column of the result set is an +** expression or subquery, then a NULL pointer is returned. +** ^The returned string is always UTF-8 encoded. +** +** ^(For example, given the database schema: +** +** CREATE TABLE t1(c1 VARIANT); +** +** and the following statement to be compiled: +** +** SELECT c1 + 1, c1 FROM t1; +** +** this routine would return the string "VARIANT" for the second result +** column (i==1), and a NULL pointer for the first result column (i==0).)^ +** +** ^SQLite uses dynamic run-time typing. ^So just because a column +** is declared to contain a particular type does not mean that the +** data stored in that column is of the declared type. SQLite is +** strongly typed, but the typing is dynamic not static. ^Type +** is associated with individual values, not with the containers +** used to hold those values. +*/ +SQLITE_API const char *sqlite3_column_decltype(sqlite3_stmt*,int); +SQLITE_API const void *sqlite3_column_decltype16(sqlite3_stmt*,int); + +/* +** CAPI3REF: Evaluate An SQL Statement +** METHOD: sqlite3_stmt +** +** After a [prepared statement] has been prepared using any of +** [sqlite3_prepare_v2()], [sqlite3_prepare_v3()], [sqlite3_prepare16_v2()], +** or [sqlite3_prepare16_v3()] or one of the legacy +** interfaces [sqlite3_prepare()] or [sqlite3_prepare16()], this function +** must be called one or more times to evaluate the statement. +** +** The details of the behavior of the sqlite3_step() interface depend +** on whether the statement was prepared using the newer "vX" interfaces +** [sqlite3_prepare_v3()], [sqlite3_prepare_v2()], [sqlite3_prepare16_v3()], +** [sqlite3_prepare16_v2()] or the older legacy +** interfaces [sqlite3_prepare()] and [sqlite3_prepare16()]. The use of the +** new "vX" interface is recommended for new applications but the legacy +** interface will continue to be supported. +** +** ^In the legacy interface, the return value will be either [SQLITE_BUSY], +** [SQLITE_DONE], [SQLITE_ROW], [SQLITE_ERROR], or [SQLITE_MISUSE]. +** ^With the "v2" interface, any of the other [result codes] or +** [extended result codes] might be returned as well. +** +** ^[SQLITE_BUSY] means that the database engine was unable to acquire the +** database locks it needs to do its job. ^If the statement is a [COMMIT] +** or occurs outside of an explicit transaction, then you can retry the +** statement. If the statement is not a [COMMIT] and occurs within an +** explicit transaction then you should rollback the transaction before +** continuing. +** +** ^[SQLITE_DONE] means that the statement has finished executing +** successfully. sqlite3_step() should not be called again on this virtual +** machine without first calling [sqlite3_reset()] to reset the virtual +** machine back to its initial state. +** +** ^If the SQL statement being executed returns any data, then [SQLITE_ROW] +** is returned each time a new row of data is ready for processing by the +** caller. The values may be accessed using the [column access functions]. +** sqlite3_step() is called again to retrieve the next row of data. +** +** ^[SQLITE_ERROR] means that a run-time error (such as a constraint +** violation) has occurred. sqlite3_step() should not be called again on +** the VM. More information may be found by calling [sqlite3_errmsg()]. +** ^With the legacy interface, a more specific error code (for example, +** [SQLITE_INTERRUPT], [SQLITE_SCHEMA], [SQLITE_CORRUPT], and so forth) +** can be obtained by calling [sqlite3_reset()] on the +** [prepared statement]. ^In the "v2" interface, +** the more specific error code is returned directly by sqlite3_step(). +** +** [SQLITE_MISUSE] means that the this routine was called inappropriately. +** Perhaps it was called on a [prepared statement] that has +** already been [sqlite3_finalize | finalized] or on one that had +** previously returned [SQLITE_ERROR] or [SQLITE_DONE]. Or it could +** be the case that the same database connection is being used by two or +** more threads at the same moment in time. +** +** For all versions of SQLite up to and including 3.6.23.1, a call to +** [sqlite3_reset()] was required after sqlite3_step() returned anything +** other than [SQLITE_ROW] before any subsequent invocation of +** sqlite3_step(). Failure to reset the prepared statement using +** [sqlite3_reset()] would result in an [SQLITE_MISUSE] return from +** sqlite3_step(). But after [version 3.6.23.1] ([dateof:3.6.23.1], +** sqlite3_step() began +** calling [sqlite3_reset()] automatically in this circumstance rather +** than returning [SQLITE_MISUSE]. This is not considered a compatibility +** break because any application that ever receives an SQLITE_MISUSE error +** is broken by definition. The [SQLITE_OMIT_AUTORESET] compile-time option +** can be used to restore the legacy behavior. +** +** Goofy Interface Alert: In the legacy interface, the sqlite3_step() +** API always returns a generic error code, [SQLITE_ERROR], following any +** error other than [SQLITE_BUSY] and [SQLITE_MISUSE]. You must call +** [sqlite3_reset()] or [sqlite3_finalize()] in order to find one of the +** specific [error codes] that better describes the error. +** We admit that this is a goofy design. The problem has been fixed +** with the "v2" interface. If you prepare all of your SQL statements +** using [sqlite3_prepare_v3()] or [sqlite3_prepare_v2()] +** or [sqlite3_prepare16_v2()] or [sqlite3_prepare16_v3()] instead +** of the legacy [sqlite3_prepare()] and [sqlite3_prepare16()] interfaces, +** then the more specific [error codes] are returned directly +** by sqlite3_step(). The use of the "vX" interfaces is recommended. +*/ +SQLITE_API int sqlite3_step(sqlite3_stmt*); + +/* +** CAPI3REF: Number of columns in a result set +** METHOD: sqlite3_stmt +** +** ^The sqlite3_data_count(P) interface returns the number of columns in the +** current row of the result set of [prepared statement] P. +** ^If prepared statement P does not have results ready to return +** (via calls to the [sqlite3_column_int | sqlite3_column_*()] of +** interfaces) then sqlite3_data_count(P) returns 0. +** ^The sqlite3_data_count(P) routine also returns 0 if P is a NULL pointer. +** ^The sqlite3_data_count(P) routine returns 0 if the previous call to +** [sqlite3_step](P) returned [SQLITE_DONE]. ^The sqlite3_data_count(P) +** will return non-zero if previous call to [sqlite3_step](P) returned +** [SQLITE_ROW], except in the case of the [PRAGMA incremental_vacuum] +** where it always returns zero since each step of that multi-step +** pragma returns 0 columns of data. +** +** See also: [sqlite3_column_count()] +*/ +SQLITE_API int sqlite3_data_count(sqlite3_stmt *pStmt); + +/* +** CAPI3REF: Fundamental Datatypes +** KEYWORDS: SQLITE_TEXT +** +** ^(Every value in SQLite has one of five fundamental datatypes: +** +**
    +**
  • 64-bit signed integer +**
  • 64-bit IEEE floating point number +**
  • string +**
  • BLOB +**
  • NULL +**
)^ +** +** These constants are codes for each of those types. +** +** Note that the SQLITE_TEXT constant was also used in SQLite version 2 +** for a completely different meaning. Software that links against both +** SQLite version 2 and SQLite version 3 should use SQLITE3_TEXT, not +** SQLITE_TEXT. +*/ +#define SQLITE_INTEGER 1 +#define SQLITE_FLOAT 2 +#define SQLITE_BLOB 4 +#define SQLITE_NULL 5 +#ifdef SQLITE_TEXT +# undef SQLITE_TEXT +#else +# define SQLITE_TEXT 3 +#endif +#define SQLITE3_TEXT 3 + +/* +** CAPI3REF: Result Values From A Query +** KEYWORDS: {column access functions} +** METHOD: sqlite3_stmt +** +** Summary: +**
+**
sqlite3_column_blob→BLOB result +**
sqlite3_column_double→REAL result +**
sqlite3_column_int→32-bit INTEGER result +**
sqlite3_column_int64→64-bit INTEGER result +**
sqlite3_column_text→UTF-8 TEXT result +**
sqlite3_column_text16→UTF-16 TEXT result +**
sqlite3_column_value→The result as an +** [sqlite3_value|unprotected sqlite3_value] object. +**
    +**
sqlite3_column_bytes→Size of a BLOB +** or a UTF-8 TEXT result in bytes +**
sqlite3_column_bytes16   +** →  Size of UTF-16 +** TEXT in bytes +**
sqlite3_column_type→Default +** datatype of the result +**
+** +** Details: +** +** ^These routines return information about a single column of the current +** result row of a query. ^In every case the first argument is a pointer +** to the [prepared statement] that is being evaluated (the [sqlite3_stmt*] +** that was returned from [sqlite3_prepare_v2()] or one of its variants) +** and the second argument is the index of the column for which information +** should be returned. ^The leftmost column of the result set has the index 0. +** ^The number of columns in the result can be determined using +** [sqlite3_column_count()]. +** +** If the SQL statement does not currently point to a valid row, or if the +** column index is out of range, the result is undefined. +** These routines may only be called when the most recent call to +** [sqlite3_step()] has returned [SQLITE_ROW] and neither +** [sqlite3_reset()] nor [sqlite3_finalize()] have been called subsequently. +** If any of these routines are called after [sqlite3_reset()] or +** [sqlite3_finalize()] or after [sqlite3_step()] has returned +** something other than [SQLITE_ROW], the results are undefined. +** If [sqlite3_step()] or [sqlite3_reset()] or [sqlite3_finalize()] +** are called from a different thread while any of these routines +** are pending, then the results are undefined. +** +** The first six interfaces (_blob, _double, _int, _int64, _text, and _text16) +** each return the value of a result column in a specific data format. If +** the result column is not initially in the requested format (for example, +** if the query returns an integer but the sqlite3_column_text() interface +** is used to extract the value) then an automatic type conversion is performed. +** +** ^The sqlite3_column_type() routine returns the +** [SQLITE_INTEGER | datatype code] for the initial data type +** of the result column. ^The returned value is one of [SQLITE_INTEGER], +** [SQLITE_FLOAT], [SQLITE_TEXT], [SQLITE_BLOB], or [SQLITE_NULL]. +** The return value of sqlite3_column_type() can be used to decide which +** of the first six interface should be used to extract the column value. +** The value returned by sqlite3_column_type() is only meaningful if no +** automatic type conversions have occurred for the value in question. +** After a type conversion, the result of calling sqlite3_column_type() +** is undefined, though harmless. Future +** versions of SQLite may change the behavior of sqlite3_column_type() +** following a type conversion. +** +** If the result is a BLOB or a TEXT string, then the sqlite3_column_bytes() +** or sqlite3_column_bytes16() interfaces can be used to determine the size +** of that BLOB or string. +** +** ^If the result is a BLOB or UTF-8 string then the sqlite3_column_bytes() +** routine returns the number of bytes in that BLOB or string. +** ^If the result is a UTF-16 string, then sqlite3_column_bytes() converts +** the string to UTF-8 and then returns the number of bytes. +** ^If the result is a numeric value then sqlite3_column_bytes() uses +** [sqlite3_snprintf()] to convert that value to a UTF-8 string and returns +** the number of bytes in that string. +** ^If the result is NULL, then sqlite3_column_bytes() returns zero. +** +** ^If the result is a BLOB or UTF-16 string then the sqlite3_column_bytes16() +** routine returns the number of bytes in that BLOB or string. +** ^If the result is a UTF-8 string, then sqlite3_column_bytes16() converts +** the string to UTF-16 and then returns the number of bytes. +** ^If the result is a numeric value then sqlite3_column_bytes16() uses +** [sqlite3_snprintf()] to convert that value to a UTF-16 string and returns +** the number of bytes in that string. +** ^If the result is NULL, then sqlite3_column_bytes16() returns zero. +** +** ^The values returned by [sqlite3_column_bytes()] and +** [sqlite3_column_bytes16()] do not include the zero terminators at the end +** of the string. ^For clarity: the values returned by +** [sqlite3_column_bytes()] and [sqlite3_column_bytes16()] are the number of +** bytes in the string, not the number of characters. +** +** ^Strings returned by sqlite3_column_text() and sqlite3_column_text16(), +** even empty strings, are always zero-terminated. ^The return +** value from sqlite3_column_blob() for a zero-length BLOB is a NULL pointer. +** +** Warning: ^The object returned by [sqlite3_column_value()] is an +** [unprotected sqlite3_value] object. In a multithreaded environment, +** an unprotected sqlite3_value object may only be used safely with +** [sqlite3_bind_value()] and [sqlite3_result_value()]. +** If the [unprotected sqlite3_value] object returned by +** [sqlite3_column_value()] is used in any other way, including calls +** to routines like [sqlite3_value_int()], [sqlite3_value_text()], +** or [sqlite3_value_bytes()], the behavior is not threadsafe. +** Hence, the sqlite3_column_value() interface +** is normally only useful within the implementation of +** [application-defined SQL functions] or [virtual tables], not within +** top-level application code. +** +** The these routines may attempt to convert the datatype of the result. +** ^For example, if the internal representation is FLOAT and a text result +** is requested, [sqlite3_snprintf()] is used internally to perform the +** conversion automatically. ^(The following table details the conversions +** that are applied: +** +**
+** +**
Internal
Type 		Requested
Type 		Conversion +** +**
NULL INTEGER Result is 0 +**
NULL FLOAT Result is 0.0 +**
NULL TEXT Result is a NULL pointer +**
NULL BLOB Result is a NULL pointer +**
INTEGER FLOAT Convert from integer to float +**
INTEGER TEXT ASCII rendering of the integer +**
INTEGER BLOB Same as INTEGER->TEXT +**
FLOAT INTEGER [CAST] to INTEGER +**
FLOAT TEXT ASCII rendering of the float +**
FLOAT BLOB [CAST] to BLOB +**
TEXT INTEGER [CAST] to INTEGER +**
TEXT FLOAT [CAST] to REAL +**
TEXT BLOB No change +**
BLOB INTEGER [CAST] to INTEGER +**
BLOB FLOAT [CAST] to REAL +**
BLOB TEXT Add a zero terminator if needed +**
+**
)^ +** +** Note that when type conversions occur, pointers returned by prior +** calls to sqlite3_column_blob(), sqlite3_column_text(), and/or +** sqlite3_column_text16() may be invalidated. +** Type conversions and pointer invalidations might occur +** in the following cases: +** +**
    +**
  • The initial content is a BLOB and sqlite3_column_text() or +** sqlite3_column_text16() is called. A zero-terminator might +** need to be added to the string.
    • +**
  • The initial content is UTF-8 text and sqlite3_column_bytes16() or +** sqlite3_column_text16() is called. The content must be converted +** to UTF-16.
    • +**
  • The initial content is UTF-16 text and sqlite3_column_bytes() or +** sqlite3_column_text() is called. The content must be converted +** to UTF-8.
    • +**
+** +** ^Conversions between UTF-16be and UTF-16le are always done in place and do +** not invalidate a prior pointer, though of course the content of the buffer +** that the prior pointer references will have been modified. Other kinds +** of conversion are done in place when it is possible, but sometimes they +** are not possible and in those cases prior pointers are invalidated. +** +** The safest policy is to invoke these routines +** in one of the following ways: +** +**
    +**
  • sqlite3_column_text() followed by sqlite3_column_bytes()
    • +**
  • sqlite3_column_blob() followed by sqlite3_column_bytes()
    • +**
  • sqlite3_column_text16() followed by sqlite3_column_bytes16()
    • +**
+** +** In other words, you should call sqlite3_column_text(), +** sqlite3_column_blob(), or sqlite3_column_text16() first to force the result +** into the desired format, then invoke sqlite3_column_bytes() or +** sqlite3_column_bytes16() to find the size of the result. Do not mix calls +** to sqlite3_column_text() or sqlite3_column_blob() with calls to +** sqlite3_column_bytes16(), and do not mix calls to sqlite3_column_text16() +** with calls to sqlite3_column_bytes(). +** +** ^The pointers returned are valid until a type conversion occurs as +** described above, or until [sqlite3_step()] or [sqlite3_reset()] or +** [sqlite3_finalize()] is called. ^The memory space used to hold strings +** and BLOBs is freed automatically. Do not pass the pointers returned +** from [sqlite3_column_blob()], [sqlite3_column_text()], etc. into +** [sqlite3_free()]. +** +** As long as the input parameters are correct, these routines will only +** fail if an out-of-memory error occurs during a format conversion. +** Only the following subset of interfaces are subject to out-of-memory +** errors: +** +**
    +**
  • sqlite3_column_blob() +**
  • sqlite3_column_text() +**
  • sqlite3_column_text16() +**
  • sqlite3_column_bytes() +**
  • sqlite3_column_bytes16() +**
+** +** If an out-of-memory error occurs, then the return value from these +** routines is the same as if the column had contained an SQL NULL value. +** Valid SQL NULL returns can be distinguished from out-of-memory errors +** by invoking the [sqlite3_errcode()] immediately after the suspect +** return value is obtained and before any +** other SQLite interface is called on the same [database connection]. +*/ +SQLITE_API const void *sqlite3_column_blob(sqlite3_stmt*, int iCol); +SQLITE_API double sqlite3_column_double(sqlite3_stmt*, int iCol); +SQLITE_API int sqlite3_column_int(sqlite3_stmt*, int iCol); +SQLITE_API sqlite3_int64 sqlite3_column_int64(sqlite3_stmt*, int iCol); +SQLITE_API const unsigned char *sqlite3_column_text(sqlite3_stmt*, int iCol); +SQLITE_API const void *sqlite3_column_text16(sqlite3_stmt*, int iCol); +SQLITE_API sqlite3_value *sqlite3_column_value(sqlite3_stmt*, int iCol); +SQLITE_API int sqlite3_column_bytes(sqlite3_stmt*, int iCol); +SQLITE_API int sqlite3_column_bytes16(sqlite3_stmt*, int iCol); +SQLITE_API int sqlite3_column_type(sqlite3_stmt*, int iCol); + +/* +** CAPI3REF: Destroy A Prepared Statement Object +** DESTRUCTOR: sqlite3_stmt +** +** ^The sqlite3_finalize() function is called to delete a [prepared statement]. +** ^If the most recent evaluation of the statement encountered no errors +** or if the statement is never been evaluated, then sqlite3_finalize() returns +** SQLITE_OK. ^If the most recent evaluation of statement S failed, then +** sqlite3_finalize(S) returns the appropriate [error code] or +** [extended error code]. +** +** ^The sqlite3_finalize(S) routine can be called at any point during +** the life cycle of [prepared statement] S: +** before statement S is ever evaluated, after +** one or more calls to [sqlite3_reset()], or after any call +** to [sqlite3_step()] regardless of whether or not the statement has +** completed execution. +** +** ^Invoking sqlite3_finalize() on a NULL pointer is a harmless no-op. +** +** The application must finalize every [prepared statement] in order to avoid +** resource leaks. It is a grievous error for the application to try to use +** a prepared statement after it has been finalized. Any use of a prepared +** statement after it has been finalized can result in undefined and +** undesirable behavior such as segfaults and heap corruption. +*/ +SQLITE_API int sqlite3_finalize(sqlite3_stmt *pStmt); + +/* +** CAPI3REF: Reset A Prepared Statement Object +** METHOD: sqlite3_stmt +** +** The sqlite3_reset() function is called to reset a [prepared statement] +** object back to its initial state, ready to be re-executed. +** ^Any SQL statement variables that had values bound to them using +** the [sqlite3_bind_blob | sqlite3_bind_*() API] retain their values. +** Use [sqlite3_clear_bindings()] to reset the bindings. +** +** ^The [sqlite3_reset(S)] interface resets the [prepared statement] S +** back to the beginning of its program. +** +** ^If the most recent call to [sqlite3_step(S)] for the +** [prepared statement] S returned [SQLITE_ROW] or [SQLITE_DONE], +** or if [sqlite3_step(S)] has never before been called on S, +** then [sqlite3_reset(S)] returns [SQLITE_OK]. +** +** ^If the most recent call to [sqlite3_step(S)] for the +** [prepared statement] S indicated an error, then +** [sqlite3_reset(S)] returns an appropriate [error code]. +** +** ^The [sqlite3_reset(S)] interface does not change the values +** of any [sqlite3_bind_blob|bindings] on the [prepared statement] S. +*/ +SQLITE_API int sqlite3_reset(sqlite3_stmt *pStmt); + +/* +** CAPI3REF: Create Or Redefine SQL Functions +** KEYWORDS: {function creation routines} +** KEYWORDS: {application-defined SQL function} +** KEYWORDS: {application-defined SQL functions} +** METHOD: sqlite3 +** +** ^These functions (collectively known as "function creation routines") +** are used to add SQL functions or aggregates or to redefine the behavior +** of existing SQL functions or aggregates. The only differences between +** the three "sqlite3_create_function*" routines are the text encoding +** expected for the second parameter (the name of the function being +** created) and the presence or absence of a destructor callback for +** the application data pointer. Function sqlite3_create_window_function() +** is similar, but allows the user to supply the extra callback functions +** needed by [aggregate window functions]. +** +** ^The first parameter is the [database connection] to which the SQL +** function is to be added. ^If an application uses more than one database +** connection then application-defined SQL functions must be added +** to each database connection separately. +** +** ^The second parameter is the name of the SQL function to be created or +** redefined. ^The length of the name is limited to 255 bytes in a UTF-8 +** representation, exclusive of the zero-terminator. ^Note that the name +** length limit is in UTF-8 bytes, not characters nor UTF-16 bytes. +** ^Any attempt to create a function with a longer name +** will result in [SQLITE_MISUSE] being returned. +** +** ^The third parameter (nArg) +** is the number of arguments that the SQL function or +** aggregate takes. ^If this parameter is -1, then the SQL function or +** aggregate may take any number of arguments between 0 and the limit +** set by [sqlite3_limit]([SQLITE_LIMIT_FUNCTION_ARG]). If the third +** parameter is less than -1 or greater than 127 then the behavior is +** undefined. +** +** ^The fourth parameter, eTextRep, specifies what +** [SQLITE_UTF8 | text encoding] this SQL function prefers for +** its parameters. The application should set this parameter to +** [SQLITE_UTF16LE] if the function implementation invokes +** [sqlite3_value_text16le()] on an input, or [SQLITE_UTF16BE] if the +** implementation invokes [sqlite3_value_text16be()] on an input, or +** [SQLITE_UTF16] if [sqlite3_value_text16()] is used, or [SQLITE_UTF8] +** otherwise. ^The same SQL function may be registered multiple times using +** different preferred text encodings, with different implementations for +** each encoding. +** ^When multiple implementations of the same function are available, SQLite +** will pick the one that involves the least amount of data conversion. +** +** ^The fourth parameter may optionally be ORed with [SQLITE_DETERMINISTIC] +** to signal that the function will always return the same result given +** the same inputs within a single SQL statement. Most SQL functions are +** deterministic. The built-in [random()] SQL function is an example of a +** function that is not deterministic. The SQLite query planner is able to +** perform additional optimizations on deterministic functions, so use +** of the [SQLITE_DETERMINISTIC] flag is recommended where possible. +** +** ^The fourth parameter may also optionally include the [SQLITE_DIRECTONLY] +** flag, which if present prevents the function from being invoked from +** within VIEWs or TRIGGERs. For security reasons, the [SQLITE_DIRECTONLY] +** flag is recommended for any application-defined SQL function that has +** side-effects. +** +** ^(The fifth parameter is an arbitrary pointer. The implementation of the +** function can gain access to this pointer using [sqlite3_user_data()].)^ +** +** ^The sixth, seventh and eighth parameters passed to the three +** "sqlite3_create_function*" functions, xFunc, xStep and xFinal, are +** pointers to C-language functions that implement the SQL function or +** aggregate. ^A scalar SQL function requires an implementation of the xFunc +** callback only; NULL pointers must be passed as the xStep and xFinal +** parameters. ^An aggregate SQL function requires an implementation of xStep +** and xFinal and NULL pointer must be passed for xFunc. ^To delete an existing +** SQL function or aggregate, pass NULL pointers for all three function +** callbacks. +** +** ^The sixth, seventh, eighth and ninth parameters (xStep, xFinal, xValue +** and xInverse) passed to sqlite3_create_window_function are pointers to +** C-language callbacks that implement the new function. xStep and xFinal +** must both be non-NULL. xValue and xInverse may either both be NULL, in +** which case a regular aggregate function is created, or must both be +** non-NULL, in which case the new function may be used as either an aggregate +** or aggregate window function. More details regarding the implementation +** of aggregate window functions are +** [user-defined window functions|available here]. +** +** ^(If the final parameter to sqlite3_create_function_v2() or +** sqlite3_create_window_function() is not NULL, then it is destructor for +** the application data pointer. The destructor is invoked when the function +** is deleted, either by being overloaded or when the database connection +** closes.)^ ^The destructor is also invoked if the call to +** sqlite3_create_function_v2() fails. ^When the destructor callback is +** invoked, it is passed a single argument which is a copy of the application +** data pointer which was the fifth parameter to sqlite3_create_function_v2(). +** +** ^It is permitted to register multiple implementations of the same +** functions with the same name but with either differing numbers of +** arguments or differing preferred text encodings. ^SQLite will use +** the implementation that most closely matches the way in which the +** SQL function is used. ^A function implementation with a non-negative +** nArg parameter is a better match than a function implementation with +** a negative nArg. ^A function where the preferred text encoding +** matches the database encoding is a better +** match than a function where the encoding is different. +** ^A function where the encoding difference is between UTF16le and UTF16be +** is a closer match than a function where the encoding difference is +** between UTF8 and UTF16. +** +** ^Built-in functions may be overloaded by new application-defined functions. +** +** ^An application-defined function is permitted to call other +** SQLite interfaces. However, such calls must not +** close the database connection nor finalize or reset the prepared +** statement in which the function is running. +*/ +SQLITE_API int sqlite3_create_function( + sqlite3 *db, + const char *zFunctionName, + int nArg, + int eTextRep, + void *pApp, + void (*xFunc)(sqlite3_context*,int,sqlite3_value**), + void (*xStep)(sqlite3_context*,int,sqlite3_value**), + void (*xFinal)(sqlite3_context*) +); +SQLITE_API int sqlite3_create_function16( + sqlite3 *db, + const void *zFunctionName, + int nArg, + int eTextRep, + void *pApp, + void (*xFunc)(sqlite3_context*,int,sqlite3_value**), + void (*xStep)(sqlite3_context*,int,sqlite3_value**), + void (*xFinal)(sqlite3_context*) +); +SQLITE_API int sqlite3_create_function_v2( + sqlite3 *db, + const char *zFunctionName, + int nArg, + int eTextRep, + void *pApp, + void (*xFunc)(sqlite3_context*,int,sqlite3_value**), + void (*xStep)(sqlite3_context*,int,sqlite3_value**), + void (*xFinal)(sqlite3_context*), + void(*xDestroy)(void*) +); +SQLITE_API int sqlite3_create_window_function( + sqlite3 *db, + const char *zFunctionName, + int nArg, + int eTextRep, + void *pApp, + void (*xStep)(sqlite3_context*,int,sqlite3_value**), + void (*xFinal)(sqlite3_context*), + void (*xValue)(sqlite3_context*), + void (*xInverse)(sqlite3_context*,int,sqlite3_value**), + void(*xDestroy)(void*) +); + +/* +** CAPI3REF: Text Encodings +** +** These constant define integer codes that represent the various +** text encodings supported by SQLite. +*/ +#define SQLITE_UTF8 1 /* IMP: R-37514-35566 */ +#define SQLITE_UTF16LE 2 /* IMP: R-03371-37637 */ +#define SQLITE_UTF16BE 3 /* IMP: R-51971-34154 */ +#define SQLITE_UTF16 4 /* Use native byte order */ +#define SQLITE_ANY 5 /* Deprecated */ +#define SQLITE_UTF16_ALIGNED 8 /* sqlite3_create_collation only */ + +/* +** CAPI3REF: Function Flags +** +** These constants may be ORed together with the +** [SQLITE_UTF8 | preferred text encoding] as the fourth argument +** to [sqlite3_create_function()], [sqlite3_create_function16()], or +** [sqlite3_create_function_v2()]. +** +** The SQLITE_DETERMINISTIC flag means that the new function will always +** maps the same inputs into the same output. The abs() function is +** deterministic, for example, but randomblob() is not. +** +** The SQLITE_DIRECTONLY flag means that the function may only be invoked +** from top-level SQL, and cannot be used in VIEWs or TRIGGERs. This is +** a security feature which is recommended for all +** [application-defined SQL functions] that have side-effects. This flag +** prevents an attacker from adding triggers and views to a schema then +** tricking a high-privilege application into causing unintended side-effects +** while performing ordinary queries. +** +** The SQLITE_SUBTYPE flag indicates to SQLite that a function may call +** [sqlite3_value_subtype()] to inspect the sub-types of its arguments. +** Specifying this flag makes no difference for scalar or aggregate user +** functions. However, if it is not specified for a user-defined window +** function, then any sub-types belonging to arguments passed to the window +** function may be discarded before the window function is called (i.e. +** sqlite3_value_subtype() will always return 0). +*/ +#define SQLITE_DETERMINISTIC 0x000000800 +#define SQLITE_DIRECTONLY 0x000080000 +#define SQLITE_SUBTYPE 0x000100000 + +/* +** CAPI3REF: Deprecated Functions +** DEPRECATED +** +** These functions are [deprecated]. In order to maintain +** backwards compatibility with older code, these functions continue +** to be supported. However, new applications should avoid +** the use of these functions. To encourage programmers to avoid +** these functions, we will not explain what they do. +*/ +#ifndef SQLITE_OMIT_DEPRECATED +SQLITE_API SQLITE_DEPRECATED int sqlite3_aggregate_count(sqlite3_context*); +SQLITE_API SQLITE_DEPRECATED int sqlite3_expired(sqlite3_stmt*); +SQLITE_API SQLITE_DEPRECATED int sqlite3_transfer_bindings(sqlite3_stmt*, sqlite3_stmt*); +SQLITE_API SQLITE_DEPRECATED int sqlite3_global_recover(void); +SQLITE_API SQLITE_DEPRECATED void sqlite3_thread_cleanup(void); +SQLITE_API SQLITE_DEPRECATED int sqlite3_memory_alarm(void(*)(void*,sqlite3_int64,int), + void*,sqlite3_int64); +#endif + +/* +** CAPI3REF: Obtaining SQL Values +** METHOD: sqlite3_value +** +** Summary: +**
+**
sqlite3_value_blob→BLOB value +**
sqlite3_value_double→REAL value +**
sqlite3_value_int→32-bit INTEGER value +**
sqlite3_value_int64→64-bit INTEGER value +**
sqlite3_value_pointer→Pointer value +**
sqlite3_value_text→UTF-8 TEXT value +**
sqlite3_value_text16→UTF-16 TEXT value in +** the native byteorder +**
sqlite3_value_text16be→UTF-16be TEXT value +**
sqlite3_value_text16le→UTF-16le TEXT value +**
    +**
sqlite3_value_bytes→Size of a BLOB +** or a UTF-8 TEXT in bytes +**
sqlite3_value_bytes16   +** →  Size of UTF-16 +** TEXT in bytes +**
sqlite3_value_type→Default +** datatype of the value +**
sqlite3_value_numeric_type   +** →  Best numeric datatype of the value +**
sqlite3_value_nochange   +** →  True if the column is unchanged in an UPDATE +** against a virtual table. +**
sqlite3_value_frombind   +** →  True if value originated from a [bound parameter] +**
+** +** Details: +** +** These routines extract type, size, and content information from +** [protected sqlite3_value] objects. Protected sqlite3_value objects +** are used to pass parameter information into implementation of +** [application-defined SQL functions] and [virtual tables]. +** +** These routines work only with [protected sqlite3_value] objects. +** Any attempt to use these routines on an [unprotected sqlite3_value] +** is not threadsafe. +** +** ^These routines work just like the corresponding [column access functions] +** except that these routines take a single [protected sqlite3_value] object +** pointer instead of a [sqlite3_stmt*] pointer and an integer column number. +** +** ^The sqlite3_value_text16() interface extracts a UTF-16 string +** in the native byte-order of the host machine. ^The +** sqlite3_value_text16be() and sqlite3_value_text16le() interfaces +** extract UTF-16 strings as big-endian and little-endian respectively. +** +** ^If [sqlite3_value] object V was initialized +** using [sqlite3_bind_pointer(S,I,P,X,D)] or [sqlite3_result_pointer(C,P,X,D)] +** and if X and Y are strings that compare equal according to strcmp(X,Y), +** then sqlite3_value_pointer(V,Y) will return the pointer P. ^Otherwise, +** sqlite3_value_pointer(V,Y) returns a NULL. The sqlite3_bind_pointer() +** routine is part of the [pointer passing interface] added for SQLite 3.20.0. +** +** ^(The sqlite3_value_type(V) interface returns the +** [SQLITE_INTEGER | datatype code] for the initial datatype of the +** [sqlite3_value] object V. The returned value is one of [SQLITE_INTEGER], +** [SQLITE_FLOAT], [SQLITE_TEXT], [SQLITE_BLOB], or [SQLITE_NULL].)^ +** Other interfaces might change the datatype for an sqlite3_value object. +** For example, if the datatype is initially SQLITE_INTEGER and +** sqlite3_value_text(V) is called to extract a text value for that +** integer, then subsequent calls to sqlite3_value_type(V) might return +** SQLITE_TEXT. Whether or not a persistent internal datatype conversion +** occurs is undefined and may change from one release of SQLite to the next. +** +** ^(The sqlite3_value_numeric_type() interface attempts to apply +** numeric affinity to the value. This means that an attempt is +** made to convert the value to an integer or floating point. If +** such a conversion is possible without loss of information (in other +** words, if the value is a string that looks like a number) +** then the conversion is performed. Otherwise no conversion occurs. +** The [SQLITE_INTEGER | datatype] after conversion is returned.)^ +** +** ^Within the [xUpdate] method of a [virtual table], the +** sqlite3_value_nochange(X) interface returns true if and only if +** the column corresponding to X is unchanged by the UPDATE operation +** that the xUpdate method call was invoked to implement and if +** and the prior [xColumn] method call that was invoked to extracted +** the value for that column returned without setting a result (probably +** because it queried [sqlite3_vtab_nochange()] and found that the column +** was unchanging). ^Within an [xUpdate] method, any value for which +** sqlite3_value_nochange(X) is true will in all other respects appear +** to be a NULL value. If sqlite3_value_nochange(X) is invoked anywhere other +** than within an [xUpdate] method call for an UPDATE statement, then +** the return value is arbitrary and meaningless. +** +** ^The sqlite3_value_frombind(X) interface returns non-zero if the +** value X originated from one of the [sqlite3_bind_int|sqlite3_bind()] +** interfaces. ^If X comes from an SQL literal value, or a table column, +** and expression, then sqlite3_value_frombind(X) returns zero. +** +** Please pay particular attention to the fact that the pointer returned +** from [sqlite3_value_blob()], [sqlite3_value_text()], or +** [sqlite3_value_text16()] can be invalidated by a subsequent call to +** [sqlite3_value_bytes()], [sqlite3_value_bytes16()], [sqlite3_value_text()], +** or [sqlite3_value_text16()]. +** +** These routines must be called from the same thread as +** the SQL function that supplied the [sqlite3_value*] parameters. +** +** As long as the input parameter is correct, these routines can only +** fail if an out-of-memory error occurs during a format conversion. +** Only the following subset of interfaces are subject to out-of-memory +** errors: +** +**
    +**
  • sqlite3_value_blob() +**
  • sqlite3_value_text() +**
  • sqlite3_value_text16() +**
  • sqlite3_value_text16le() +**
  • sqlite3_value_text16be() +**
  • sqlite3_value_bytes() +**
  • sqlite3_value_bytes16() +**
+** +** If an out-of-memory error occurs, then the return value from these +** routines is the same as if the column had contained an SQL NULL value. +** Valid SQL NULL returns can be distinguished from out-of-memory errors +** by invoking the [sqlite3_errcode()] immediately after the suspect +** return value is obtained and before any +** other SQLite interface is called on the same [database connection]. +*/ +SQLITE_API const void *sqlite3_value_blob(sqlite3_value*); +SQLITE_API double sqlite3_value_double(sqlite3_value*); +SQLITE_API int sqlite3_value_int(sqlite3_value*); +SQLITE_API sqlite3_int64 sqlite3_value_int64(sqlite3_value*); +SQLITE_API void *sqlite3_value_pointer(sqlite3_value*, const char*); +SQLITE_API const unsigned char *sqlite3_value_text(sqlite3_value*); +SQLITE_API const void *sqlite3_value_text16(sqlite3_value*); +SQLITE_API const void *sqlite3_value_text16le(sqlite3_value*); +SQLITE_API const void *sqlite3_value_text16be(sqlite3_value*); +SQLITE_API int sqlite3_value_bytes(sqlite3_value*); +SQLITE_API int sqlite3_value_bytes16(sqlite3_value*); +SQLITE_API int sqlite3_value_type(sqlite3_value*); +SQLITE_API int sqlite3_value_numeric_type(sqlite3_value*); +SQLITE_API int sqlite3_value_nochange(sqlite3_value*); +SQLITE_API int sqlite3_value_frombind(sqlite3_value*); + +/* +** CAPI3REF: Finding The Subtype Of SQL Values +** METHOD: sqlite3_value +** +** The sqlite3_value_subtype(V) function returns the subtype for +** an [application-defined SQL function] argument V. The subtype +** information can be used to pass a limited amount of context from +** one SQL function to another. Use the [sqlite3_result_subtype()] +** routine to set the subtype for the return value of an SQL function. +*/ +SQLITE_API unsigned int sqlite3_value_subtype(sqlite3_value*); + +/* +** CAPI3REF: Copy And Free SQL Values +** METHOD: sqlite3_value +** +** ^The sqlite3_value_dup(V) interface makes a copy of the [sqlite3_value] +** object D and returns a pointer to that copy. ^The [sqlite3_value] returned +** is a [protected sqlite3_value] object even if the input is not. +** ^The sqlite3_value_dup(V) interface returns NULL if V is NULL or if a +** memory allocation fails. +** +** ^The sqlite3_value_free(V) interface frees an [sqlite3_value] object +** previously obtained from [sqlite3_value_dup()]. ^If V is a NULL pointer +** then sqlite3_value_free(V) is a harmless no-op. +*/ +SQLITE_API sqlite3_value *sqlite3_value_dup(const sqlite3_value*); +SQLITE_API void sqlite3_value_free(sqlite3_value*); + +/* +** CAPI3REF: Obtain Aggregate Function Context +** METHOD: sqlite3_context +** +** Implementations of aggregate SQL functions use this +** routine to allocate memory for storing their state. +** +** ^The first time the sqlite3_aggregate_context(C,N) routine is called +** for a particular aggregate function, SQLite +** allocates N of memory, zeroes out that memory, and returns a pointer +** to the new memory. ^On second and subsequent calls to +** sqlite3_aggregate_context() for the same aggregate function instance, +** the same buffer is returned. Sqlite3_aggregate_context() is normally +** called once for each invocation of the xStep callback and then one +** last time when the xFinal callback is invoked. ^(When no rows match +** an aggregate query, the xStep() callback of the aggregate function +** implementation is never called and xFinal() is called exactly once. +** In those cases, sqlite3_aggregate_context() might be called for the +** first time from within xFinal().)^ +** +** ^The sqlite3_aggregate_context(C,N) routine returns a NULL pointer +** when first called if N is less than or equal to zero or if a memory +** allocate error occurs. +** +** ^(The amount of space allocated by sqlite3_aggregate_context(C,N) is +** determined by the N parameter on first successful call. Changing the +** value of N in subsequent call to sqlite3_aggregate_context() within +** the same aggregate function instance will not resize the memory +** allocation.)^ Within the xFinal callback, it is customary to set +** N=0 in calls to sqlite3_aggregate_context(C,N) so that no +** pointless memory allocations occur. +** +** ^SQLite automatically frees the memory allocated by +** sqlite3_aggregate_context() when the aggregate query concludes. +** +** The first parameter must be a copy of the +** [sqlite3_context | SQL function context] that is the first parameter +** to the xStep or xFinal callback routine that implements the aggregate +** function. +** +** This routine must be called from the same thread in which +** the aggregate SQL function is running. +*/ +SQLITE_API void *sqlite3_aggregate_context(sqlite3_context*, int nBytes); + +/* +** CAPI3REF: User Data For Functions +** METHOD: sqlite3_context +** +** ^The sqlite3_user_data() interface returns a copy of +** the pointer that was the pUserData parameter (the 5th parameter) +** of the [sqlite3_create_function()] +** and [sqlite3_create_function16()] routines that originally +** registered the application defined function. +** +** This routine must be called from the same thread in which +** the application-defined function is running. +*/ +SQLITE_API void *sqlite3_user_data(sqlite3_context*); + +/* +** CAPI3REF: Database Connection For Functions +** METHOD: sqlite3_context +** +** ^The sqlite3_context_db_handle() interface returns a copy of +** the pointer to the [database connection] (the 1st parameter) +** of the [sqlite3_create_function()] +** and [sqlite3_create_function16()] routines that originally +** registered the application defined function. +*/ +SQLITE_API sqlite3 *sqlite3_context_db_handle(sqlite3_context*); + +/* +** CAPI3REF: Function Auxiliary Data +** METHOD: sqlite3_context +** +** These functions may be used by (non-aggregate) SQL functions to +** associate metadata with argument values. If the same value is passed to +** multiple invocations of the same SQL function during query execution, under +** some circumstances the associated metadata may be preserved. An example +** of where this might be useful is in a regular-expression matching +** function. The compiled version of the regular expression can be stored as +** metadata associated with the pattern string. +** Then as long as the pattern string remains the same, +** the compiled regular expression can be reused on multiple +** invocations of the same function. +** +** ^The sqlite3_get_auxdata(C,N) interface returns a pointer to the metadata +** associated by the sqlite3_set_auxdata(C,N,P,X) function with the Nth argument +** value to the application-defined function. ^N is zero for the left-most +** function argument. ^If there is no metadata +** associated with the function argument, the sqlite3_get_auxdata(C,N) interface +** returns a NULL pointer. +** +** ^The sqlite3_set_auxdata(C,N,P,X) interface saves P as metadata for the N-th +** argument of the application-defined function. ^Subsequent +** calls to sqlite3_get_auxdata(C,N) return P from the most recent +** sqlite3_set_auxdata(C,N,P,X) call if the metadata is still valid or +** NULL if the metadata has been discarded. +** ^After each call to sqlite3_set_auxdata(C,N,P,X) where X is not NULL, +** SQLite will invoke the destructor function X with parameter P exactly +** once, when the metadata is discarded. +** SQLite is free to discard the metadata at any time, including:
    +**
  • ^(when the corresponding function parameter changes)^, or +**
  • ^(when [sqlite3_reset()] or [sqlite3_finalize()] is called for the +** SQL statement)^, or +**
  • ^(when sqlite3_set_auxdata() is invoked again on the same +** parameter)^, or +**
  • ^(during the original sqlite3_set_auxdata() call when a memory +** allocation error occurs.)^
+** +** Note the last bullet in particular. The destructor X in +** sqlite3_set_auxdata(C,N,P,X) might be called immediately, before the +** sqlite3_set_auxdata() interface even returns. Hence sqlite3_set_auxdata() +** should be called near the end of the function implementation and the +** function implementation should not make any use of P after +** sqlite3_set_auxdata() has been called. +** +** ^(In practice, metadata is preserved between function calls for +** function parameters that are compile-time constants, including literal +** values and [parameters] and expressions composed from the same.)^ +** +** The value of the N parameter to these interfaces should be non-negative. +** Future enhancements may make use of negative N values to define new +** kinds of function caching behavior. +** +** These routines must be called from the same thread in which +** the SQL function is running. +*/ +SQLITE_API void *sqlite3_get_auxdata(sqlite3_context*, int N); +SQLITE_API void sqlite3_set_auxdata(sqlite3_context*, int N, void*, void (*)(void*)); + + +/* +** CAPI3REF: Constants Defining Special Destructor Behavior +** +** These are special values for the destructor that is passed in as the +** final argument to routines like [sqlite3_result_blob()]. ^If the destructor +** argument is SQLITE_STATIC, it means that the content pointer is constant +** and will never change. It does not need to be destroyed. ^The +** SQLITE_TRANSIENT value means that the content will likely change in +** the near future and that SQLite should make its own private copy of +** the content before returning. +** +** The typedef is necessary to work around problems in certain +** C++ compilers. +*/ +typedef void (*sqlite3_destructor_type)(void*); +#define SQLITE_STATIC ((sqlite3_destructor_type)0) +#define SQLITE_TRANSIENT ((sqlite3_destructor_type)-1) + +/* +** CAPI3REF: Setting The Result Of An SQL Function +** METHOD: sqlite3_context +** +** These routines are used by the xFunc or xFinal callbacks that +** implement SQL functions and aggregates. See +** [sqlite3_create_function()] and [sqlite3_create_function16()] +** for additional information. +** +** These functions work very much like the [parameter binding] family of +** functions used to bind values to host parameters in prepared statements. +** Refer to the [SQL parameter] documentation for additional information. +** +** ^The sqlite3_result_blob() interface sets the result from +** an application-defined function to be the BLOB whose content is pointed +** to by the second parameter and which is N bytes long where N is the +** third parameter. +** +** ^The sqlite3_result_zeroblob(C,N) and sqlite3_result_zeroblob64(C,N) +** interfaces set the result of the application-defined function to be +** a BLOB containing all zero bytes and N bytes in size. +** +** ^The sqlite3_result_double() interface sets the result from +** an application-defined function to be a floating point value specified +** by its 2nd argument. +** +** ^The sqlite3_result_error() and sqlite3_result_error16() functions +** cause the implemented SQL function to throw an exception. +** ^SQLite uses the string pointed to by the +** 2nd parameter of sqlite3_result_error() or sqlite3_result_error16() +** as the text of an error message. ^SQLite interprets the error +** message string from sqlite3_result_error() as UTF-8. ^SQLite +** interprets the string from sqlite3_result_error16() as UTF-16 in native +** byte order. ^If the third parameter to sqlite3_result_error() +** or sqlite3_result_error16() is negative then SQLite takes as the error +** message all text up through the first zero character. +** ^If the third parameter to sqlite3_result_error() or +** sqlite3_result_error16() is non-negative then SQLite takes that many +** bytes (not characters) from the 2nd parameter as the error message. +** ^The sqlite3_result_error() and sqlite3_result_error16() +** routines make a private copy of the error message text before +** they return. Hence, the calling function can deallocate or +** modify the text after they return without harm. +** ^The sqlite3_result_error_code() function changes the error code +** returned by SQLite as a result of an error in a function. ^By default, +** the error code is SQLITE_ERROR. ^A subsequent call to sqlite3_result_error() +** or sqlite3_result_error16() resets the error code to SQLITE_ERROR. +** +** ^The sqlite3_result_error_toobig() interface causes SQLite to throw an +** error indicating that a string or BLOB is too long to represent. +** +** ^The sqlite3_result_error_nomem() interface causes SQLite to throw an +** error indicating that a memory allocation failed. +** +** ^The sqlite3_result_int() interface sets the return value +** of the application-defined function to be the 32-bit signed integer +** value given in the 2nd argument. +** ^The sqlite3_result_int64() interface sets the return value +** of the application-defined function to be the 64-bit signed integer +** value given in the 2nd argument. +** +** ^The sqlite3_result_null() interface sets the return value +** of the application-defined function to be NULL. +** +** ^The sqlite3_result_text(), sqlite3_result_text16(), +** sqlite3_result_text16le(), and sqlite3_result_text16be() interfaces +** set the return value of the application-defined function to be +** a text string which is represented as UTF-8, UTF-16 native byte order, +** UTF-16 little endian, or UTF-16 big endian, respectively. +** ^The sqlite3_result_text64() interface sets the return value of an +** application-defined function to be a text string in an encoding +** specified by the fifth (and last) parameter, which must be one +** of [SQLITE_UTF8], [SQLITE_UTF16], [SQLITE_UTF16BE], or [SQLITE_UTF16LE]. +** ^SQLite takes the text result from the application from +** the 2nd parameter of the sqlite3_result_text* interfaces. +** ^If the 3rd parameter to the sqlite3_result_text* interfaces +** is negative, then SQLite takes result text from the 2nd parameter +** through the first zero character. +** ^If the 3rd parameter to the sqlite3_result_text* interfaces +** is non-negative, then as many bytes (not characters) of the text +** pointed to by the 2nd parameter are taken as the application-defined +** function result. If the 3rd parameter is non-negative, then it +** must be the byte offset into the string where the NUL terminator would +** appear if the string where NUL terminated. If any NUL characters occur +** in the string at a byte offset that is less than the value of the 3rd +** parameter, then the resulting string will contain embedded NULs and the +** result of expressions operating on strings with embedded NULs is undefined. +** ^If the 4th parameter to the sqlite3_result_text* interfaces +** or sqlite3_result_blob is a non-NULL pointer, then SQLite calls that +** function as the destructor on the text or BLOB result when it has +** finished using that result. +** ^If the 4th parameter to the sqlite3_result_text* interfaces or to +** sqlite3_result_blob is the special constant SQLITE_STATIC, then SQLite +** assumes that the text or BLOB result is in constant space and does not +** copy the content of the parameter nor call a destructor on the content +** when it has finished using that result. +** ^If the 4th parameter to the sqlite3_result_text* interfaces +** or sqlite3_result_blob is the special constant SQLITE_TRANSIENT +** then SQLite makes a copy of the result into space obtained +** from [sqlite3_malloc()] before it returns. +** +** ^The sqlite3_result_value() interface sets the result of +** the application-defined function to be a copy of the +** [unprotected sqlite3_value] object specified by the 2nd parameter. ^The +** sqlite3_result_value() interface makes a copy of the [sqlite3_value] +** so that the [sqlite3_value] specified in the parameter may change or +** be deallocated after sqlite3_result_value() returns without harm. +** ^A [protected sqlite3_value] object may always be used where an +** [unprotected sqlite3_value] object is required, so either +** kind of [sqlite3_value] object can be used with this interface. +** +** ^The sqlite3_result_pointer(C,P,T,D) interface sets the result to an +** SQL NULL value, just like [sqlite3_result_null(C)], except that it +** also associates the host-language pointer P or type T with that +** NULL value such that the pointer can be retrieved within an +** [application-defined SQL function] using [sqlite3_value_pointer()]. +** ^If the D parameter is not NULL, then it is a pointer to a destructor +** for the P parameter. ^SQLite invokes D with P as its only argument +** when SQLite is finished with P. The T parameter should be a static +** string and preferably a string literal. The sqlite3_result_pointer() +** routine is part of the [pointer passing interface] added for SQLite 3.20.0. +** +** If these routines are called from within the different thread +** than the one containing the application-defined function that received +** the [sqlite3_context] pointer, the results are undefined. +*/ +SQLITE_API void sqlite3_result_blob(sqlite3_context*, const void*, int, void(*)(void*)); +SQLITE_API void sqlite3_result_blob64(sqlite3_context*,const void*, + sqlite3_uint64,void(*)(void*)); +SQLITE_API void sqlite3_result_double(sqlite3_context*, double); +SQLITE_API void sqlite3_result_error(sqlite3_context*, const char*, int); +SQLITE_API void sqlite3_result_error16(sqlite3_context*, const void*, int); +SQLITE_API void sqlite3_result_error_toobig(sqlite3_context*); +SQLITE_API void sqlite3_result_error_nomem(sqlite3_context*); +SQLITE_API void sqlite3_result_error_code(sqlite3_context*, int); +SQLITE_API void sqlite3_result_int(sqlite3_context*, int); +SQLITE_API void sqlite3_result_int64(sqlite3_context*, sqlite3_int64); +SQLITE_API void sqlite3_result_null(sqlite3_context*); +SQLITE_API void sqlite3_result_text(sqlite3_context*, const char*, int, void(*)(void*)); +SQLITE_API void sqlite3_result_text64(sqlite3_context*, const char*,sqlite3_uint64, + void(*)(void*), unsigned char encoding); +SQLITE_API void sqlite3_result_text16(sqlite3_context*, const void*, int, void(*)(void*)); +SQLITE_API void sqlite3_result_text16le(sqlite3_context*, const void*, int,void(*)(void*)); +SQLITE_API void sqlite3_result_text16be(sqlite3_context*, const void*, int,void(*)(void*)); +SQLITE_API void sqlite3_result_value(sqlite3_context*, sqlite3_value*); +SQLITE_API void sqlite3_result_pointer(sqlite3_context*, void*,const char*,void(*)(void*)); +SQLITE_API void sqlite3_result_zeroblob(sqlite3_context*, int n); +SQLITE_API int sqlite3_result_zeroblob64(sqlite3_context*, sqlite3_uint64 n); + + +/* +** CAPI3REF: Setting The Subtype Of An SQL Function +** METHOD: sqlite3_context +** +** The sqlite3_result_subtype(C,T) function causes the subtype of +** the result from the [application-defined SQL function] with +** [sqlite3_context] C to be the value T. Only the lower 8 bits +** of the subtype T are preserved in current versions of SQLite; +** higher order bits are discarded. +** The number of subtype bytes preserved by SQLite might increase +** in future releases of SQLite. +*/ +SQLITE_API void sqlite3_result_subtype(sqlite3_context*,unsigned int); + +/* +** CAPI3REF: Define New Collating Sequences +** METHOD: sqlite3 +** +** ^These functions add, remove, or modify a [collation] associated +** with the [database connection] specified as the first argument. +** +** ^The name of the collation is a UTF-8 string +** for sqlite3_create_collation() and sqlite3_create_collation_v2() +** and a UTF-16 string in native byte order for sqlite3_create_collation16(). +** ^Collation names that compare equal according to [sqlite3_strnicmp()] are +** considered to be the same name. +** +** ^(The third argument (eTextRep) must be one of the constants: +**
    +**
  • [SQLITE_UTF8], +**
  • [SQLITE_UTF16LE], +**
  • [SQLITE_UTF16BE], +**
  • [SQLITE_UTF16], or +**
  • [SQLITE_UTF16_ALIGNED]. +**
)^ +** ^The eTextRep argument determines the encoding of strings passed +** to the collating function callback, xCallback. +** ^The [SQLITE_UTF16] and [SQLITE_UTF16_ALIGNED] values for eTextRep +** force strings to be UTF16 with native byte order. +** ^The [SQLITE_UTF16_ALIGNED] value for eTextRep forces strings to begin +** on an even byte address. +** +** ^The fourth argument, pArg, is an application data pointer that is passed +** through as the first argument to the collating function callback. +** +** ^The fifth argument, xCallback, is a pointer to the collating function. +** ^Multiple collating functions can be registered using the same name but +** with different eTextRep parameters and SQLite will use whichever +** function requires the least amount of data transformation. +** ^If the xCallback argument is NULL then the collating function is +** deleted. ^When all collating functions having the same name are deleted, +** that collation is no longer usable. +** +** ^The collating function callback is invoked with a copy of the pArg +** application data pointer and with two strings in the encoding specified +** by the eTextRep argument. The collating function must return an +** integer that is negative, zero, or positive +** if the first string is less than, equal to, or greater than the second, +** respectively. A collating function must always return the same answer +** given the same inputs. If two or more collating functions are registered +** to the same collation name (using different eTextRep values) then all +** must give an equivalent answer when invoked with equivalent strings. +** The collating function must obey the following properties for all +** strings A, B, and C: +** +**
    +**
  1. If A==B then B==A. +**
  2. If A==B and B==C then A==C. +**
  3. If A<B THEN B>A. +**
  4. If A<B and B<C then A<C. +**
+** +** If a collating function fails any of the above constraints and that +** collating function is registered and used, then the behavior of SQLite +** is undefined. +** +** ^The sqlite3_create_collation_v2() works like sqlite3_create_collation() +** with the addition that the xDestroy callback is invoked on pArg when +** the collating function is deleted. +** ^Collating functions are deleted when they are overridden by later +** calls to the collation creation functions or when the +** [database connection] is closed using [sqlite3_close()]. +** +** ^The xDestroy callback is not called if the +** sqlite3_create_collation_v2() function fails. Applications that invoke +** sqlite3_create_collation_v2() with a non-NULL xDestroy argument should +** check the return code and dispose of the application data pointer +** themselves rather than expecting SQLite to deal with it for them. +** This is different from every other SQLite interface. The inconsistency +** is unfortunate but cannot be changed without breaking backwards +** compatibility. +** +** See also: [sqlite3_collation_needed()] and [sqlite3_collation_needed16()]. +*/ +SQLITE_API int sqlite3_create_collation( + sqlite3*, + const char *zName, + int eTextRep, + void *pArg, + int(*xCompare)(void*,int,const void*,int,const void*) +); +SQLITE_API int sqlite3_create_collation_v2( + sqlite3*, + const char *zName, + int eTextRep, + void *pArg, + int(*xCompare)(void*,int,const void*,int,const void*), + void(*xDestroy)(void*) +); +SQLITE_API int sqlite3_create_collation16( + sqlite3*, + const void *zName, + int eTextRep, + void *pArg, + int(*xCompare)(void*,int,const void*,int,const void*) +); + +/* +** CAPI3REF: Collation Needed Callbacks +** METHOD: sqlite3 +** +** ^To avoid having to register all collation sequences before a database +** can be used, a single callback function may be registered with the +** [database connection] to be invoked whenever an undefined collation +** sequence is required. +** +** ^If the function is registered using the sqlite3_collation_needed() API, +** then it is passed the names of undefined collation sequences as strings +** encoded in UTF-8. ^If sqlite3_collation_needed16() is used, +** the names are passed as UTF-16 in machine native byte order. +** ^A call to either function replaces the existing collation-needed callback. +** +** ^(When the callback is invoked, the first argument passed is a copy +** of the second argument to sqlite3_collation_needed() or +** sqlite3_collation_needed16(). The second argument is the database +** connection. The third argument is one of [SQLITE_UTF8], [SQLITE_UTF16BE], +** or [SQLITE_UTF16LE], indicating the most desirable form of the collation +** sequence function required. The fourth parameter is the name of the +** required collation sequence.)^ +** +** The callback function should register the desired collation using +** [sqlite3_create_collation()], [sqlite3_create_collation16()], or +** [sqlite3_create_collation_v2()]. +*/ +SQLITE_API int sqlite3_collation_needed( + sqlite3*, + void*, + void(*)(void*,sqlite3*,int eTextRep,const char*) +); +SQLITE_API int sqlite3_collation_needed16( + sqlite3*, + void*, + void(*)(void*,sqlite3*,int eTextRep,const void*) +); + +#ifdef SQLITE_HAS_CODEC +/* +** Specify the key for an encrypted database. This routine should be +** called right after sqlite3_open(). +** +** The code to implement this API is not available in the public release +** of SQLite. +*/ +SQLITE_API int sqlite3_key( + sqlite3 *db, /* Database to be rekeyed */ + const void *pKey, int nKey /* The key */ +); +SQLITE_API int sqlite3_key_v2( + sqlite3 *db, /* Database to be rekeyed */ + const char *zDbName, /* Name of the database */ + const void *pKey, int nKey /* The key */ +); + +/* +** Change the key on an open database. If the current database is not +** encrypted, this routine will encrypt it. If pNew==0 or nNew==0, the +** database is decrypted. +** +** The code to implement this API is not available in the public release +** of SQLite. +*/ +SQLITE_API int sqlite3_rekey( + sqlite3 *db, /* Database to be rekeyed */ + const void *pKey, int nKey /* The new key */ +); +SQLITE_API int sqlite3_rekey_v2( + sqlite3 *db, /* Database to be rekeyed */ + const char *zDbName, /* Name of the database */ + const void *pKey, int nKey /* The new key */ +); + +/* +** Specify the activation key for a SEE database. Unless +** activated, none of the SEE routines will work. +*/ +SQLITE_API void sqlite3_activate_see( + const char *zPassPhrase /* Activation phrase */ +); +#endif + +#ifdef SQLITE_ENABLE_CEROD +/* +** Specify the activation key for a CEROD database. Unless +** activated, none of the CEROD routines will work. +*/ +SQLITE_API void sqlite3_activate_cerod( + const char *zPassPhrase /* Activation phrase */ +); +#endif + +/* +** CAPI3REF: Suspend Execution For A Short Time +** +** The sqlite3_sleep() function causes the current thread to suspend execution +** for at least a number of milliseconds specified in its parameter. +** +** If the operating system does not support sleep requests with +** millisecond time resolution, then the time will be rounded up to +** the nearest second. The number of milliseconds of sleep actually +** requested from the operating system is returned. +** +** ^SQLite implements this interface by calling the xSleep() +** method of the default [sqlite3_vfs] object. If the xSleep() method +** of the default VFS is not implemented correctly, or not implemented at +** all, then the behavior of sqlite3_sleep() may deviate from the description +** in the previous paragraphs. +*/ +SQLITE_API int sqlite3_sleep(int); + +/* +** CAPI3REF: Name Of The Folder Holding Temporary Files +** +** ^(If this global variable is made to point to a string which is +** the name of a folder (a.k.a. directory), then all temporary files +** created by SQLite when using a built-in [sqlite3_vfs | VFS] +** will be placed in that directory.)^ ^If this variable +** is a NULL pointer, then SQLite performs a search for an appropriate +** temporary file directory. +** +** Applications are strongly discouraged from using this global variable. +** It is required to set a temporary folder on Windows Runtime (WinRT). +** But for all other platforms, it is highly recommended that applications +** neither read nor write this variable. This global variable is a relic +** that exists for backwards compatibility of legacy applications and should +** be avoided in new projects. +** +** It is not safe to read or modify this variable in more than one +** thread at a time. It is not safe to read or modify this variable +** if a [database connection] is being used at the same time in a separate +** thread. +** It is intended that this variable be set once +** as part of process initialization and before any SQLite interface +** routines have been called and that this variable remain unchanged +** thereafter. +** +** ^The [temp_store_directory pragma] may modify this variable and cause +** it to point to memory obtained from [sqlite3_malloc]. ^Furthermore, +** the [temp_store_directory pragma] always assumes that any string +** that this variable points to is held in memory obtained from +** [sqlite3_malloc] and the pragma may attempt to free that memory +** using [sqlite3_free]. +** Hence, if this variable is modified directly, either it should be +** made NULL or made to point to memory obtained from [sqlite3_malloc] +** or else the use of the [temp_store_directory pragma] should be avoided. +** Except when requested by the [temp_store_directory pragma], SQLite +** does not free the memory that sqlite3_temp_directory points to. If +** the application wants that memory to be freed, it must do +** so itself, taking care to only do so after all [database connection] +** objects have been destroyed. +** +** Note to Windows Runtime users: The temporary directory must be set +** prior to calling [sqlite3_open] or [sqlite3_open_v2]. Otherwise, various +** features that require the use of temporary files may fail. Here is an +** example of how to do this using C++ with the Windows Runtime: +** +** 
+** LPCWSTR zPath = Windows::Storage::ApplicationData::Current->
+**       TemporaryFolder->Path->Data();
+** char zPathBuf[MAX_PATH + 1];
+** memset(zPathBuf, 0, sizeof(zPathBuf));
+** WideCharToMultiByte(CP_UTF8, 0, zPath, -1, zPathBuf, sizeof(zPathBuf),
+**       NULL, NULL);
+** sqlite3_temp_directory = sqlite3_mprintf("%s", zPathBuf);
+**
+*/ +SQLITE_API char *sqlite3_temp_directory; + +/* +** CAPI3REF: Name Of The Folder Holding Database Files +** +** ^(If this global variable is made to point to a string which is +** the name of a folder (a.k.a. directory), then all database files +** specified with a relative pathname and created or accessed by +** SQLite when using a built-in windows [sqlite3_vfs | VFS] will be assumed +** to be relative to that directory.)^ ^If this variable is a NULL +** pointer, then SQLite assumes that all database files specified +** with a relative pathname are relative to the current directory +** for the process. Only the windows VFS makes use of this global +** variable; it is ignored by the unix VFS. +** +** Changing the value of this variable while a database connection is +** open can result in a corrupt database. +** +** It is not safe to read or modify this variable in more than one +** thread at a time. It is not safe to read or modify this variable +** if a [database connection] is being used at the same time in a separate +** thread. +** It is intended that this variable be set once +** as part of process initialization and before any SQLite interface +** routines have been called and that this variable remain unchanged +** thereafter. +** +** ^The [data_store_directory pragma] may modify this variable and cause +** it to point to memory obtained from [sqlite3_malloc]. ^Furthermore, +** the [data_store_directory pragma] always assumes that any string +** that this variable points to is held in memory obtained from +** [sqlite3_malloc] and the pragma may attempt to free that memory +** using [sqlite3_free]. +** Hence, if this variable is modified directly, either it should be +** made NULL or made to point to memory obtained from [sqlite3_malloc] +** or else the use of the [data_store_directory pragma] should be avoided. +*/ +SQLITE_API char *sqlite3_data_directory; + +/* +** CAPI3REF: Win32 Specific Interface +** +** These interfaces are available only on Windows. The +** [sqlite3_win32_set_directory] interface is used to set the value associated +** with the [sqlite3_temp_directory] or [sqlite3_data_directory] variable, to +** zValue, depending on the value of the type parameter. The zValue parameter +** should be NULL to cause the previous value to be freed via [sqlite3_free]; +** a non-NULL value will be copied into memory obtained from [sqlite3_malloc] +** prior to being used. The [sqlite3_win32_set_directory] interface returns +** [SQLITE_OK] to indicate success, [SQLITE_ERROR] if the type is unsupported, +** or [SQLITE_NOMEM] if memory could not be allocated. The value of the +** [sqlite3_data_directory] variable is intended to act as a replacement for +** the current directory on the sub-platforms of Win32 where that concept is +** not present, e.g. WinRT and UWP. The [sqlite3_win32_set_directory8] and +** [sqlite3_win32_set_directory16] interfaces behave exactly the same as the +** sqlite3_win32_set_directory interface except the string parameter must be +** UTF-8 or UTF-16, respectively. +*/ +SQLITE_API int sqlite3_win32_set_directory( + unsigned long type, /* Identifier for directory being set or reset */ + void *zValue /* New value for directory being set or reset */ +); +SQLITE_API int sqlite3_win32_set_directory8(unsigned long type, const char *zValue); +SQLITE_API int sqlite3_win32_set_directory16(unsigned long type, const void *zValue); + +/* +** CAPI3REF: Win32 Directory Types +** +** These macros are only available on Windows. They define the allowed values +** for the type argument to the [sqlite3_win32_set_directory] interface. +*/ +#define SQLITE_WIN32_DATA_DIRECTORY_TYPE 1 +#define SQLITE_WIN32_TEMP_DIRECTORY_TYPE 2 + +/* +** CAPI3REF: Test For Auto-Commit Mode +** KEYWORDS: {autocommit mode} +** METHOD: sqlite3 +** +** ^The sqlite3_get_autocommit() interface returns non-zero or +** zero if the given database connection is or is not in autocommit mode, +** respectively. ^Autocommit mode is on by default. +** ^Autocommit mode is disabled by a [BEGIN] statement. +** ^Autocommit mode is re-enabled by a [COMMIT] or [ROLLBACK]. +** +** If certain kinds of errors occur on a statement within a multi-statement +** transaction (errors including [SQLITE_FULL], [SQLITE_IOERR], +** [SQLITE_NOMEM], [SQLITE_BUSY], and [SQLITE_INTERRUPT]) then the +** transaction might be rolled back automatically. The only way to +** find out whether SQLite automatically rolled back the transaction after +** an error is to use this function. +** +** If another thread changes the autocommit status of the database +** connection while this routine is running, then the return value +** is undefined. +*/ +SQLITE_API int sqlite3_get_autocommit(sqlite3*); + +/* +** CAPI3REF: Find The Database Handle Of A Prepared Statement +** METHOD: sqlite3_stmt +** +** ^The sqlite3_db_handle interface returns the [database connection] handle +** to which a [prepared statement] belongs. ^The [database connection] +** returned by sqlite3_db_handle is the same [database connection] +** that was the first argument +** to the [sqlite3_prepare_v2()] call (or its variants) that was used to +** create the statement in the first place. +*/ +SQLITE_API sqlite3 *sqlite3_db_handle(sqlite3_stmt*); + +/* +** CAPI3REF: Return The Filename For A Database Connection +** METHOD: sqlite3 +** +** ^The sqlite3_db_filename(D,N) interface returns a pointer to a filename +** associated with database N of connection D. ^The main database file +** has the name "main". If there is no attached database N on the database +** connection D, or if database N is a temporary or in-memory database, then +** this function will return either a NULL pointer or an empty string. +** +** ^The filename returned by this function is the output of the +** xFullPathname method of the [VFS]. ^In other words, the filename +** will be an absolute pathname, even if the filename used +** to open the database originally was a URI or relative pathname. +*/ +SQLITE_API const char *sqlite3_db_filename(sqlite3 *db, const char *zDbName); + +/* +** CAPI3REF: Determine if a database is read-only +** METHOD: sqlite3 +** +** ^The sqlite3_db_readonly(D,N) interface returns 1 if the database N +** of connection D is read-only, 0 if it is read/write, or -1 if N is not +** the name of a database on connection D. +*/ +SQLITE_API int sqlite3_db_readonly(sqlite3 *db, const char *zDbName); + +/* +** CAPI3REF: Find the next prepared statement +** METHOD: sqlite3 +** +** ^This interface returns a pointer to the next [prepared statement] after +** pStmt associated with the [database connection] pDb. ^If pStmt is NULL +** then this interface returns a pointer to the first prepared statement +** associated with the database connection pDb. ^If no prepared statement +** satisfies the conditions of this routine, it returns NULL. +** +** The [database connection] pointer D in a call to +** [sqlite3_next_stmt(D,S)] must refer to an open database +** connection and in particular must not be a NULL pointer. +*/ +SQLITE_API sqlite3_stmt *sqlite3_next_stmt(sqlite3 *pDb, sqlite3_stmt *pStmt); + +/* +** CAPI3REF: Commit And Rollback Notification Callbacks +** METHOD: sqlite3 +** +** ^The sqlite3_commit_hook() interface registers a callback +** function to be invoked whenever a transaction is [COMMIT | committed]. +** ^Any callback set by a previous call to sqlite3_commit_hook() +** for the same database connection is overridden. +** ^The sqlite3_rollback_hook() interface registers a callback +** function to be invoked whenever a transaction is [ROLLBACK | rolled back]. +** ^Any callback set by a previous call to sqlite3_rollback_hook() +** for the same database connection is overridden. +** ^The pArg argument is passed through to the callback. +** ^If the callback on a commit hook function returns non-zero, +** then the commit is converted into a rollback. +** +** ^The sqlite3_commit_hook(D,C,P) and sqlite3_rollback_hook(D,C,P) functions +** return the P argument from the previous call of the same function +** on the same [database connection] D, or NULL for +** the first call for each function on D. +** +** The commit and rollback hook callbacks are not reentrant. +** The callback implementation must not do anything that will modify +** the database connection that invoked the callback. Any actions +** to modify the database connection must be deferred until after the +** completion of the [sqlite3_step()] call that triggered the commit +** or rollback hook in the first place. +** Note that running any other SQL statements, including SELECT statements, +** or merely calling [sqlite3_prepare_v2()] and [sqlite3_step()] will modify +** the database connections for the meaning of "modify" in this paragraph. +** +** ^Registering a NULL function disables the callback. +** +** ^When the commit hook callback routine returns zero, the [COMMIT] +** operation is allowed to continue normally. ^If the commit hook +** returns non-zero, then the [COMMIT] is converted into a [ROLLBACK]. +** ^The rollback hook is invoked on a rollback that results from a commit +** hook returning non-zero, just as it would be with any other rollback. +** +** ^For the purposes of this API, a transaction is said to have been +** rolled back if an explicit "ROLLBACK" statement is executed, or +** an error or constraint causes an implicit rollback to occur. +** ^The rollback callback is not invoked if a transaction is +** automatically rolled back because the database connection is closed. +** +** See also the [sqlite3_update_hook()] interface. +*/ +SQLITE_API void *sqlite3_commit_hook(sqlite3*, int(*)(void*), void*); +SQLITE_API void *sqlite3_rollback_hook(sqlite3*, void(*)(void *), void*); + +/* +** CAPI3REF: Data Change Notification Callbacks +** METHOD: sqlite3 +** +** ^The sqlite3_update_hook() interface registers a callback function +** with the [database connection] identified by the first argument +** to be invoked whenever a row is updated, inserted or deleted in +** a [rowid table]. +** ^Any callback set by a previous call to this function +** for the same database connection is overridden. +** +** ^The second argument is a pointer to the function to invoke when a +** row is updated, inserted or deleted in a rowid table. +** ^The first argument to the callback is a copy of the third argument +** to sqlite3_update_hook(). +** ^The second callback argument is one of [SQLITE_INSERT], [SQLITE_DELETE], +** or [SQLITE_UPDATE], depending on the operation that caused the callback +** to be invoked. +** ^The third and fourth arguments to the callback contain pointers to the +** database and table name containing the affected row. +** ^The final callback parameter is the [rowid] of the row. +** ^In the case of an update, this is the [rowid] after the update takes place. +** +** ^(The update hook is not invoked when internal system tables are +** modified (i.e. sqlite_master and sqlite_sequence).)^ +** ^The update hook is not invoked when [WITHOUT ROWID] tables are modified. +** +** ^In the current implementation, the update hook +** is not invoked when conflicting rows are deleted because of an +** [ON CONFLICT | ON CONFLICT REPLACE] clause. ^Nor is the update hook +** invoked when rows are deleted using the [truncate optimization]. +** The exceptions defined in this paragraph might change in a future +** release of SQLite. +** +** The update hook implementation must not do anything that will modify +** the database connection that invoked the update hook. Any actions +** to modify the database connection must be deferred until after the +** completion of the [sqlite3_step()] call that triggered the update hook. +** Note that [sqlite3_prepare_v2()] and [sqlite3_step()] both modify their +** database connections for the meaning of "modify" in this paragraph. +** +** ^The sqlite3_update_hook(D,C,P) function +** returns the P argument from the previous call +** on the same [database connection] D, or NULL for +** the first call on D. +** +** See also the [sqlite3_commit_hook()], [sqlite3_rollback_hook()], +** and [sqlite3_preupdate_hook()] interfaces. +*/ +SQLITE_API void *sqlite3_update_hook( + sqlite3*, + void(*)(void *,int ,char const *,char const *,sqlite3_int64), + void* +); + +/* +** CAPI3REF: Enable Or Disable Shared Pager Cache +** +** ^(This routine enables or disables the sharing of the database cache +** and schema data structures between [database connection | connections] +** to the same database. Sharing is enabled if the argument is true +** and disabled if the argument is false.)^ +** +** ^Cache sharing is enabled and disabled for an entire process. +** This is a change as of SQLite [version 3.5.0] ([dateof:3.5.0]). +** In prior versions of SQLite, +** sharing was enabled or disabled for each thread separately. +** +** ^(The cache sharing mode set by this interface effects all subsequent +** calls to [sqlite3_open()], [sqlite3_open_v2()], and [sqlite3_open16()]. +** Existing database connections continue use the sharing mode +** that was in effect at the time they were opened.)^ +** +** ^(This routine returns [SQLITE_OK] if shared cache was enabled or disabled +** successfully. An [error code] is returned otherwise.)^ +** +** ^Shared cache is disabled by default. But this might change in +** future releases of SQLite. Applications that care about shared +** cache setting should set it explicitly. +** +** Note: This method is disabled on MacOS X 10.7 and iOS version 5.0 +** and will always return SQLITE_MISUSE. On those systems, +** shared cache mode should be enabled per-database connection via +** [sqlite3_open_v2()] with [SQLITE_OPEN_SHAREDCACHE]. +** +** This interface is threadsafe on processors where writing a +** 32-bit integer is atomic. +** +** See Also: [SQLite Shared-Cache Mode] +*/ +SQLITE_API int sqlite3_enable_shared_cache(int); + +/* +** CAPI3REF: Attempt To Free Heap Memory +** +** ^The sqlite3_release_memory() interface attempts to free N bytes +** of heap memory by deallocating non-essential memory allocations +** held by the database library. Memory used to cache database +** pages to improve performance is an example of non-essential memory. +** ^sqlite3_release_memory() returns the number of bytes actually freed, +** which might be more or less than the amount requested. +** ^The sqlite3_release_memory() routine is a no-op returning zero +** if SQLite is not compiled with [SQLITE_ENABLE_MEMORY_MANAGEMENT]. +** +** See also: [sqlite3_db_release_memory()] +*/ +SQLITE_API int sqlite3_release_memory(int); + +/* +** CAPI3REF: Free Memory Used By A Database Connection +** METHOD: sqlite3 +** +** ^The sqlite3_db_release_memory(D) interface attempts to free as much heap +** memory as possible from database connection D. Unlike the +** [sqlite3_release_memory()] interface, this interface is in effect even +** when the [SQLITE_ENABLE_MEMORY_MANAGEMENT] compile-time option is +** omitted. +** +** See also: [sqlite3_release_memory()] +*/ +SQLITE_API int sqlite3_db_release_memory(sqlite3*); + +/* +** CAPI3REF: Impose A Limit On Heap Size +** +** ^The sqlite3_soft_heap_limit64() interface sets and/or queries the +** soft limit on the amount of heap memory that may be allocated by SQLite. +** ^SQLite strives to keep heap memory utilization below the soft heap +** limit by reducing the number of pages held in the page cache +** as heap memory usages approaches the limit. +** ^The soft heap limit is "soft" because even though SQLite strives to stay +** below the limit, it will exceed the limit rather than generate +** an [SQLITE_NOMEM] error. In other words, the soft heap limit +** is advisory only. +** +** ^The return value from sqlite3_soft_heap_limit64() is the size of +** the soft heap limit prior to the call, or negative in the case of an +** error. ^If the argument N is negative +** then no change is made to the soft heap limit. Hence, the current +** size of the soft heap limit can be determined by invoking +** sqlite3_soft_heap_limit64() with a negative argument. +** +** ^If the argument N is zero then the soft heap limit is disabled. +** +** ^(The soft heap limit is not enforced in the current implementation +** if one or more of following conditions are true: +** +**
    +**
  • The soft heap limit is set to zero. +**
  • Memory accounting is disabled using a combination of the +** [sqlite3_config]([SQLITE_CONFIG_MEMSTATUS],...) start-time option and +** the [SQLITE_DEFAULT_MEMSTATUS] compile-time option. +**
  • An alternative page cache implementation is specified using +** [sqlite3_config]([SQLITE_CONFIG_PCACHE2],...). +**
  • The page cache allocates from its own memory pool supplied +** by [sqlite3_config]([SQLITE_CONFIG_PAGECACHE],...) rather than +** from the heap. +**
)^ +** +** Beginning with SQLite [version 3.7.3] ([dateof:3.7.3]), +** the soft heap limit is enforced +** regardless of whether or not the [SQLITE_ENABLE_MEMORY_MANAGEMENT] +** compile-time option is invoked. With [SQLITE_ENABLE_MEMORY_MANAGEMENT], +** the soft heap limit is enforced on every memory allocation. Without +** [SQLITE_ENABLE_MEMORY_MANAGEMENT], the soft heap limit is only enforced +** when memory is allocated by the page cache. Testing suggests that because +** the page cache is the predominate memory user in SQLite, most +** applications will achieve adequate soft heap limit enforcement without +** the use of [SQLITE_ENABLE_MEMORY_MANAGEMENT]. +** +** The circumstances under which SQLite will enforce the soft heap limit may +** changes in future releases of SQLite. +*/ +SQLITE_API sqlite3_int64 sqlite3_soft_heap_limit64(sqlite3_int64 N); + +/* +** CAPI3REF: Deprecated Soft Heap Limit Interface +** DEPRECATED +** +** This is a deprecated version of the [sqlite3_soft_heap_limit64()] +** interface. This routine is provided for historical compatibility +** only. All new applications should use the +** [sqlite3_soft_heap_limit64()] interface rather than this one. +*/ +SQLITE_API SQLITE_DEPRECATED void sqlite3_soft_heap_limit(int N); + + +/* +** CAPI3REF: Extract Metadata About A Column Of A Table +** METHOD: sqlite3 +** +** ^(The sqlite3_table_column_metadata(X,D,T,C,....) routine returns +** information about column C of table T in database D +** on [database connection] X.)^ ^The sqlite3_table_column_metadata() +** interface returns SQLITE_OK and fills in the non-NULL pointers in +** the final five arguments with appropriate values if the specified +** column exists. ^The sqlite3_table_column_metadata() interface returns +** SQLITE_ERROR and if the specified column does not exist. +** ^If the column-name parameter to sqlite3_table_column_metadata() is a +** NULL pointer, then this routine simply checks for the existence of the +** table and returns SQLITE_OK if the table exists and SQLITE_ERROR if it +** does not. If the table name parameter T in a call to +** sqlite3_table_column_metadata(X,D,T,C,...) is NULL then the result is +** undefined behavior. +** +** ^The column is identified by the second, third and fourth parameters to +** this function. ^(The second parameter is either the name of the database +** (i.e. "main", "temp", or an attached database) containing the specified +** table or NULL.)^ ^If it is NULL, then all attached databases are searched +** for the table using the same algorithm used by the database engine to +** resolve unqualified table references. +** +** ^The third and fourth parameters to this function are the table and column +** name of the desired column, respectively. +** +** ^Metadata is returned by writing to the memory locations passed as the 5th +** and subsequent parameters to this function. ^Any of these arguments may be +** NULL, in which case the corresponding element of metadata is omitted. +** +** ^(
+** +**
Parameter Output
Type 		Description +** +**
5th const char* Data type +**
6th const char* Name of default collation sequence +**
7th int True if column has a NOT NULL constraint +**
8th int True if column is part of the PRIMARY KEY +**
9th int True if column is [AUTOINCREMENT] +**
+**
)^ +** +** ^The memory pointed to by the character pointers returned for the +** declaration type and collation sequence is valid until the next +** call to any SQLite API function. +** +** ^If the specified table is actually a view, an [error code] is returned. +** +** ^If the specified column is "rowid", "oid" or "_rowid_" and the table +** is not a [WITHOUT ROWID] table and an +** [INTEGER PRIMARY KEY] column has been explicitly declared, then the output +** parameters are set for the explicitly declared column. ^(If there is no +** [INTEGER PRIMARY KEY] column, then the outputs +** for the [rowid] are set as follows: +** +** 
+**     data type: "INTEGER"
+**     collation sequence: "BINARY"
+**     not null: 0
+**     primary key: 1
+**     auto increment: 0
+**
)^ +** +** ^This function causes all database schemas to be read from disk and +** parsed, if that has not already been done, and returns an error if +** any errors are encountered while loading the schema. +*/ +SQLITE_API int sqlite3_table_column_metadata( + sqlite3 *db, /* Connection handle */ + const char *zDbName, /* Database name or NULL */ + const char *zTableName, /* Table name */ + const char *zColumnName, /* Column name */ + char const **pzDataType, /* OUTPUT: Declared data type */ + char const **pzCollSeq, /* OUTPUT: Collation sequence name */ + int *pNotNull, /* OUTPUT: True if NOT NULL constraint exists */ + int *pPrimaryKey, /* OUTPUT: True if column part of PK */ + int *pAutoinc /* OUTPUT: True if column is auto-increment */ +); + +/* +** CAPI3REF: Load An Extension +** METHOD: sqlite3 +** +** ^This interface loads an SQLite extension library from the named file. +** +** ^The sqlite3_load_extension() interface attempts to load an +** [SQLite extension] library contained in the file zFile. If +** the file cannot be loaded directly, attempts are made to load +** with various operating-system specific extensions added. +** So for example, if "samplelib" cannot be loaded, then names like +** "samplelib.so" or "samplelib.dylib" or "samplelib.dll" might +** be tried also. +** +** ^The entry point is zProc. +** ^(zProc may be 0, in which case SQLite will try to come up with an +** entry point name on its own. It first tries "sqlite3_extension_init". +** If that does not work, it constructs a name "sqlite3_X_init" where the +** X is consists of the lower-case equivalent of all ASCII alphabetic +** characters in the filename from the last "/" to the first following +** "." and omitting any initial "lib".)^ +** ^The sqlite3_load_extension() interface returns +** [SQLITE_OK] on success and [SQLITE_ERROR] if something goes wrong. +** ^If an error occurs and pzErrMsg is not 0, then the +** [sqlite3_load_extension()] interface shall attempt to +** fill *pzErrMsg with error message text stored in memory +** obtained from [sqlite3_malloc()]. The calling function +** should free this memory by calling [sqlite3_free()]. +** +** ^Extension loading must be enabled using +** [sqlite3_enable_load_extension()] or +** [sqlite3_db_config](db,[SQLITE_DBCONFIG_ENABLE_LOAD_EXTENSION],1,NULL) +** prior to calling this API, +** otherwise an error will be returned. +** +** Security warning: It is recommended that the +** [SQLITE_DBCONFIG_ENABLE_LOAD_EXTENSION] method be used to enable only this +** interface. The use of the [sqlite3_enable_load_extension()] interface +** should be avoided. This will keep the SQL function [load_extension()] +** disabled and prevent SQL injections from giving attackers +** access to extension loading capabilities. +** +** See also the [load_extension() SQL function]. +*/ +SQLITE_API int sqlite3_load_extension( + sqlite3 *db, /* Load the extension into this database connection */ + const char *zFile, /* Name of the shared library containing extension */ + const char *zProc, /* Entry point. Derived from zFile if 0 */ + char **pzErrMsg /* Put error message here if not 0 */ +); + +/* +** CAPI3REF: Enable Or Disable Extension Loading +** METHOD: sqlite3 +** +** ^So as not to open security holes in older applications that are +** unprepared to deal with [extension loading], and as a means of disabling +** [extension loading] while evaluating user-entered SQL, the following API +** is provided to turn the [sqlite3_load_extension()] mechanism on and off. +** +** ^Extension loading is off by default. +** ^Call the sqlite3_enable_load_extension() routine with onoff==1 +** to turn extension loading on and call it with onoff==0 to turn +** it back off again. +** +** ^This interface enables or disables both the C-API +** [sqlite3_load_extension()] and the SQL function [load_extension()]. +** ^(Use [sqlite3_db_config](db,[SQLITE_DBCONFIG_ENABLE_LOAD_EXTENSION],..) +** to enable or disable only the C-API.)^ +** +** Security warning: It is recommended that extension loading +** be disabled using the [SQLITE_DBCONFIG_ENABLE_LOAD_EXTENSION] method +** rather than this interface, so the [load_extension()] SQL function +** remains disabled. This will prevent SQL injections from giving attackers +** access to extension loading capabilities. +*/ +SQLITE_API int sqlite3_enable_load_extension(sqlite3 *db, int onoff); + +/* +** CAPI3REF: Automatically Load Statically Linked Extensions +** +** ^This interface causes the xEntryPoint() function to be invoked for +** each new [database connection] that is created. The idea here is that +** xEntryPoint() is the entry point for a statically linked [SQLite extension] +** that is to be automatically loaded into all new database connections. +** +** ^(Even though the function prototype shows that xEntryPoint() takes +** no arguments and returns void, SQLite invokes xEntryPoint() with three +** arguments and expects an integer result as if the signature of the +** entry point where as follows: +** +** 
+**    int xEntryPoint(
+**      sqlite3 *db,
+**      const char **pzErrMsg,
+**      const struct sqlite3_api_routines *pThunk
+**    );
+**
)^ +** +** If the xEntryPoint routine encounters an error, it should make *pzErrMsg +** point to an appropriate error message (obtained from [sqlite3_mprintf()]) +** and return an appropriate [error code]. ^SQLite ensures that *pzErrMsg +** is NULL before calling the xEntryPoint(). ^SQLite will invoke +** [sqlite3_free()] on *pzErrMsg after xEntryPoint() returns. ^If any +** xEntryPoint() returns an error, the [sqlite3_open()], [sqlite3_open16()], +** or [sqlite3_open_v2()] call that provoked the xEntryPoint() will fail. +** +** ^Calling sqlite3_auto_extension(X) with an entry point X that is already +** on the list of automatic extensions is a harmless no-op. ^No entry point +** will be called more than once for each database connection that is opened. +** +** See also: [sqlite3_reset_auto_extension()] +** and [sqlite3_cancel_auto_extension()] +*/ +SQLITE_API int sqlite3_auto_extension(void(*xEntryPoint)(void)); + +/* +** CAPI3REF: Cancel Automatic Extension Loading +** +** ^The [sqlite3_cancel_auto_extension(X)] interface unregisters the +** initialization routine X that was registered using a prior call to +** [sqlite3_auto_extension(X)]. ^The [sqlite3_cancel_auto_extension(X)] +** routine returns 1 if initialization routine X was successfully +** unregistered and it returns 0 if X was not on the list of initialization +** routines. +*/ +SQLITE_API int sqlite3_cancel_auto_extension(void(*xEntryPoint)(void)); + +/* +** CAPI3REF: Reset Automatic Extension Loading +** +** ^This interface disables all automatic extensions previously +** registered using [sqlite3_auto_extension()]. +*/ +SQLITE_API void sqlite3_reset_auto_extension(void); + +/* +** The interface to the virtual-table mechanism is currently considered +** to be experimental. The interface might change in incompatible ways. +** If this is a problem for you, do not use the interface at this time. +** +** When the virtual-table mechanism stabilizes, we will declare the +** interface fixed, support it indefinitely, and remove this comment. +*/ + +/* +** Structures used by the virtual table interface +*/ +typedef struct sqlite3_vtab sqlite3_vtab; +typedef struct sqlite3_index_info sqlite3_index_info; +typedef struct sqlite3_vtab_cursor sqlite3_vtab_cursor; +typedef struct sqlite3_module sqlite3_module; + +/* +** CAPI3REF: Virtual Table Object +** KEYWORDS: sqlite3_module {virtual table module} +** +** This structure, sometimes called a "virtual table module", +** defines the implementation of a [virtual tables]. +** This structure consists mostly of methods for the module. +** +** ^A virtual table module is created by filling in a persistent +** instance of this structure and passing a pointer to that instance +** to [sqlite3_create_module()] or [sqlite3_create_module_v2()]. +** ^The registration remains valid until it is replaced by a different +** module or until the [database connection] closes. The content +** of this structure must not change while it is registered with +** any database connection. +*/ +struct sqlite3_module { + int iVersion; + int (*xCreate)(sqlite3*, void *pAux, + int argc, const char *const*argv, + sqlite3_vtab **ppVTab, char**); + int (*xConnect)(sqlite3*, void *pAux, + int argc, const char *const*argv, + sqlite3_vtab **ppVTab, char**); + int (*xBestIndex)(sqlite3_vtab *pVTab, sqlite3_index_info*); + int (*xDisconnect)(sqlite3_vtab *pVTab); + int (*xDestroy)(sqlite3_vtab *pVTab); + int (*xOpen)(sqlite3_vtab *pVTab, sqlite3_vtab_cursor **ppCursor); + int (*xClose)(sqlite3_vtab_cursor*); + int (*xFilter)(sqlite3_vtab_cursor*, int idxNum, const char *idxStr, + int argc, sqlite3_value **argv); + int (*xNext)(sqlite3_vtab_cursor*); + int (*xEof)(sqlite3_vtab_cursor*); + int (*xColumn)(sqlite3_vtab_cursor*, sqlite3_context*, int); + int (*xRowid)(sqlite3_vtab_cursor*, sqlite3_int64 *pRowid); + int (*xUpdate)(sqlite3_vtab *, int, sqlite3_value **, sqlite3_int64 *); + int (*xBegin)(sqlite3_vtab *pVTab); + int (*xSync)(sqlite3_vtab *pVTab); + int (*xCommit)(sqlite3_vtab *pVTab); + int (*xRollback)(sqlite3_vtab *pVTab); + int (*xFindFunction)(sqlite3_vtab *pVtab, int nArg, const char *zName, + void (**pxFunc)(sqlite3_context*,int,sqlite3_value**), + void **ppArg); + int (*xRename)(sqlite3_vtab *pVtab, const char *zNew); + /* The methods above are in version 1 of the sqlite_module object. Those + ** below are for version 2 and greater. */ + int (*xSavepoint)(sqlite3_vtab *pVTab, int); + int (*xRelease)(sqlite3_vtab *pVTab, int); + int (*xRollbackTo)(sqlite3_vtab *pVTab, int); + /* The methods above are in versions 1 and 2 of the sqlite_module object. + ** Those below are for version 3 and greater. */ + int (*xShadowName)(const char*); +}; + +/* +** CAPI3REF: Virtual Table Indexing Information +** KEYWORDS: sqlite3_index_info +** +** The sqlite3_index_info structure and its substructures is used as part +** of the [virtual table] interface to +** pass information into and receive the reply from the [xBestIndex] +** method of a [virtual table module]. The fields under **Inputs** are the +** inputs to xBestIndex and are read-only. xBestIndex inserts its +** results into the **Outputs** fields. +** +** ^(The aConstraint[] array records WHERE clause constraints of the form: +** +**
column OP expr
+** +** where OP is =, <, <=, >, or >=.)^ ^(The particular operator is +** stored in aConstraint[].op using one of the +** [SQLITE_INDEX_CONSTRAINT_EQ | SQLITE_INDEX_CONSTRAINT_ values].)^ +** ^(The index of the column is stored in +** aConstraint[].iColumn.)^ ^(aConstraint[].usable is TRUE if the +** expr on the right-hand side can be evaluated (and thus the constraint +** is usable) and false if it cannot.)^ +** +** ^The optimizer automatically inverts terms of the form "expr OP column" +** and makes other simplifications to the WHERE clause in an attempt to +** get as many WHERE clause terms into the form shown above as possible. +** ^The aConstraint[] array only reports WHERE clause terms that are +** relevant to the particular virtual table being queried. +** +** ^Information about the ORDER BY clause is stored in aOrderBy[]. +** ^Each term of aOrderBy records a column of the ORDER BY clause. +** +** The colUsed field indicates which columns of the virtual table may be +** required by the current scan. Virtual table columns are numbered from +** zero in the order in which they appear within the CREATE TABLE statement +** passed to sqlite3_declare_vtab(). For the first 63 columns (columns 0-62), +** the corresponding bit is set within the colUsed mask if the column may be +** required by SQLite. If the table has at least 64 columns and any column +** to the right of the first 63 is required, then bit 63 of colUsed is also +** set. In other words, column iCol may be required if the expression +** (colUsed & ((sqlite3_uint64)1 << (iCol>=63 ? 63 : iCol))) evaluates to +** non-zero. +** +** The [xBestIndex] method must fill aConstraintUsage[] with information +** about what parameters to pass to xFilter. ^If argvIndex>0 then +** the right-hand side of the corresponding aConstraint[] is evaluated +** and becomes the argvIndex-th entry in argv. ^(If aConstraintUsage[].omit +** is true, then the constraint is assumed to be fully handled by the +** virtual table and is not checked again by SQLite.)^ +** +** ^The idxNum and idxPtr values are recorded and passed into the +** [xFilter] method. +** ^[sqlite3_free()] is used to free idxPtr if and only if +** needToFreeIdxPtr is true. +** +** ^The orderByConsumed means that output from [xFilter]/[xNext] will occur in +** the correct order to satisfy the ORDER BY clause so that no separate +** sorting step is required. +** +** ^The estimatedCost value is an estimate of the cost of a particular +** strategy. A cost of N indicates that the cost of the strategy is similar +** to a linear scan of an SQLite table with N rows. A cost of log(N) +** indicates that the expense of the operation is similar to that of a +** binary search on a unique indexed field of an SQLite table with N rows. +** +** ^The estimatedRows value is an estimate of the number of rows that +** will be returned by the strategy. +** +** The xBestIndex method may optionally populate the idxFlags field with a +** mask of SQLITE_INDEX_SCAN_* flags. Currently there is only one such flag - +** SQLITE_INDEX_SCAN_UNIQUE. If the xBestIndex method sets this flag, SQLite +** assumes that the strategy may visit at most one row. +** +** Additionally, if xBestIndex sets the SQLITE_INDEX_SCAN_UNIQUE flag, then +** SQLite also assumes that if a call to the xUpdate() method is made as +** part of the same statement to delete or update a virtual table row and the +** implementation returns SQLITE_CONSTRAINT, then there is no need to rollback +** any database changes. In other words, if the xUpdate() returns +** SQLITE_CONSTRAINT, the database contents must be exactly as they were +** before xUpdate was called. By contrast, if SQLITE_INDEX_SCAN_UNIQUE is not +** set and xUpdate returns SQLITE_CONSTRAINT, any database changes made by +** the xUpdate method are automatically rolled back by SQLite. +** +** IMPORTANT: The estimatedRows field was added to the sqlite3_index_info +** structure for SQLite [version 3.8.2] ([dateof:3.8.2]). +** If a virtual table extension is +** used with an SQLite version earlier than 3.8.2, the results of attempting +** to read or write the estimatedRows field are undefined (but are likely +** to included crashing the application). The estimatedRows field should +** therefore only be used if [sqlite3_libversion_number()] returns a +** value greater than or equal to 3008002. Similarly, the idxFlags field +** was added for [version 3.9.0] ([dateof:3.9.0]). +** It may therefore only be used if +** sqlite3_libversion_number() returns a value greater than or equal to +** 3009000. +*/ +struct sqlite3_index_info { + /* Inputs */ + int nConstraint; /* Number of entries in aConstraint */ + struct sqlite3_index_constraint { + int iColumn; /* Column constrained. -1 for ROWID */ + unsigned char op; /* Constraint operator */ + unsigned char usable; /* True if this constraint is usable */ + int iTermOffset; /* Used internally - xBestIndex should ignore */ + } *aConstraint; /* Table of WHERE clause constraints */ + int nOrderBy; /* Number of terms in the ORDER BY clause */ + struct sqlite3_index_orderby { + int iColumn; /* Column number */ + unsigned char desc; /* True for DESC. False for ASC. */ + } *aOrderBy; /* The ORDER BY clause */ + /* Outputs */ + struct sqlite3_index_constraint_usage { + int argvIndex; /* if >0, constraint is part of argv to xFilter */ + unsigned char omit; /* Do not code a test for this constraint */ + } *aConstraintUsage; + int idxNum; /* Number used to identify the index */ + char *idxStr; /* String, possibly obtained from sqlite3_malloc */ + int needToFreeIdxStr; /* Free idxStr using sqlite3_free() if true */ + int orderByConsumed; /* True if output is already ordered */ + double estimatedCost; /* Estimated cost of using this index */ + /* Fields below are only available in SQLite 3.8.2 and later */ + sqlite3_int64 estimatedRows; /* Estimated number of rows returned */ + /* Fields below are only available in SQLite 3.9.0 and later */ + int idxFlags; /* Mask of SQLITE_INDEX_SCAN_* flags */ + /* Fields below are only available in SQLite 3.10.0 and later */ + sqlite3_uint64 colUsed; /* Input: Mask of columns used by statement */ +}; + +/* +** CAPI3REF: Virtual Table Scan Flags +** +** Virtual table implementations are allowed to set the +** [sqlite3_index_info].idxFlags field to some combination of +** these bits. +*/ +#define SQLITE_INDEX_SCAN_UNIQUE 1 /* Scan visits at most 1 row */ + +/* +** CAPI3REF: Virtual Table Constraint Operator Codes +** +** These macros defined the allowed values for the +** [sqlite3_index_info].aConstraint[].op field. Each value represents +** an operator that is part of a constraint term in the wHERE clause of +** a query that uses a [virtual table]. +*/ +#define SQLITE_INDEX_CONSTRAINT_EQ 2 +#define SQLITE_INDEX_CONSTRAINT_GT 4 +#define SQLITE_INDEX_CONSTRAINT_LE 8 +#define SQLITE_INDEX_CONSTRAINT_LT 16 +#define SQLITE_INDEX_CONSTRAINT_GE 32 +#define SQLITE_INDEX_CONSTRAINT_MATCH 64 +#define SQLITE_INDEX_CONSTRAINT_LIKE 65 +#define SQLITE_INDEX_CONSTRAINT_GLOB 66 +#define SQLITE_INDEX_CONSTRAINT_REGEXP 67 +#define SQLITE_INDEX_CONSTRAINT_NE 68 +#define SQLITE_INDEX_CONSTRAINT_ISNOT 69 +#define SQLITE_INDEX_CONSTRAINT_ISNOTNULL 70 +#define SQLITE_INDEX_CONSTRAINT_ISNULL 71 +#define SQLITE_INDEX_CONSTRAINT_IS 72 +#define SQLITE_INDEX_CONSTRAINT_FUNCTION 150 + +/* +** CAPI3REF: Register A Virtual Table Implementation +** METHOD: sqlite3 +** +** ^These routines are used to register a new [virtual table module] name. +** ^Module names must be registered before +** creating a new [virtual table] using the module and before using a +** preexisting [virtual table] for the module. +** +** ^The module name is registered on the [database connection] specified +** by the first parameter. ^The name of the module is given by the +** second parameter. ^The third parameter is a pointer to +** the implementation of the [virtual table module]. ^The fourth +** parameter is an arbitrary client data pointer that is passed through +** into the [xCreate] and [xConnect] methods of the virtual table module +** when a new virtual table is be being created or reinitialized. +** +** ^The sqlite3_create_module_v2() interface has a fifth parameter which +** is a pointer to a destructor for the pClientData. ^SQLite will +** invoke the destructor function (if it is not NULL) when SQLite +** no longer needs the pClientData pointer. ^The destructor will also +** be invoked if the call to sqlite3_create_module_v2() fails. +** ^The sqlite3_create_module() +** interface is equivalent to sqlite3_create_module_v2() with a NULL +** destructor. +** +** ^If the third parameter (the pointer to the sqlite3_module object) is +** NULL then no new module is create and any existing modules with the +** same name are dropped. +** +** See also: [sqlite3_drop_modules()] +*/ +SQLITE_API int sqlite3_create_module( + sqlite3 *db, /* SQLite connection to register module with */ + const char *zName, /* Name of the module */ + const sqlite3_module *p, /* Methods for the module */ + void *pClientData /* Client data for xCreate/xConnect */ +); +SQLITE_API int sqlite3_create_module_v2( + sqlite3 *db, /* SQLite connection to register module with */ + const char *zName, /* Name of the module */ + const sqlite3_module *p, /* Methods for the module */ + void *pClientData, /* Client data for xCreate/xConnect */ + void(*xDestroy)(void*) /* Module destructor function */ +); + +/* +** CAPI3REF: Remove Unnecessary Virtual Table Implementations +** METHOD: sqlite3 +** +** ^The sqlite3_drop_modules(D,L) interface removes all virtual +** table modules from database connection D except those named on list L. +** The L parameter must be either NULL or a pointer to an array of pointers +** to strings where the array is terminated by a single NULL pointer. +** ^If the L parameter is NULL, then all virtual table modules are removed. +** +** See also: [sqlite3_create_module()] +*/ +SQLITE_API int sqlite3_drop_modules( + sqlite3 *db, /* Remove modules from this connection */ + const char **azKeep /* Except, do not remove the ones named here */ +); + +/* +** CAPI3REF: Virtual Table Instance Object +** KEYWORDS: sqlite3_vtab +** +** Every [virtual table module] implementation uses a subclass +** of this object to describe a particular instance +** of the [virtual table]. Each subclass will +** be tailored to the specific needs of the module implementation. +** The purpose of this superclass is to define certain fields that are +** common to all module implementations. +** +** ^Virtual tables methods can set an error message by assigning a +** string obtained from [sqlite3_mprintf()] to zErrMsg. The method should +** take care that any prior string is freed by a call to [sqlite3_free()] +** prior to assigning a new string to zErrMsg. ^After the error message +** is delivered up to the client application, the string will be automatically +** freed by sqlite3_free() and the zErrMsg field will be zeroed. +*/ +struct sqlite3_vtab { + const sqlite3_module *pModule; /* The module for this virtual table */ + int nRef; /* Number of open cursors */ + char *zErrMsg; /* Error message from sqlite3_mprintf() */ + /* Virtual table implementations will typically add additional fields */ +}; + +/* +** CAPI3REF: Virtual Table Cursor Object +** KEYWORDS: sqlite3_vtab_cursor {virtual table cursor} +** +** Every [virtual table module] implementation uses a subclass of the +** following structure to describe cursors that point into the +** [virtual table] and are used +** to loop through the virtual table. Cursors are created using the +** [sqlite3_module.xOpen | xOpen] method of the module and are destroyed +** by the [sqlite3_module.xClose | xClose] method. Cursors are used +** by the [xFilter], [xNext], [xEof], [xColumn], and [xRowid] methods +** of the module. Each module implementation will define +** the content of a cursor structure to suit its own needs. +** +** This superclass exists in order to define fields of the cursor that +** are common to all implementations. +*/ +struct sqlite3_vtab_cursor { + sqlite3_vtab *pVtab; /* Virtual table of this cursor */ + /* Virtual table implementations will typically add additional fields */ +}; + +/* +** CAPI3REF: Declare The Schema Of A Virtual Table +** +** ^The [xCreate] and [xConnect] methods of a +** [virtual table module] call this interface +** to declare the format (the names and datatypes of the columns) of +** the virtual tables they implement. +*/ +SQLITE_API int sqlite3_declare_vtab(sqlite3*, const char *zSQL); + +/* +** CAPI3REF: Overload A Function For A Virtual Table +** METHOD: sqlite3 +** +** ^(Virtual tables can provide alternative implementations of functions +** using the [xFindFunction] method of the [virtual table module]. +** But global versions of those functions +** must exist in order to be overloaded.)^ +** +** ^(This API makes sure a global version of a function with a particular +** name and number of parameters exists. If no such function exists +** before this API is called, a new function is created.)^ ^The implementation +** of the new function always causes an exception to be thrown. So +** the new function is not good for anything by itself. Its only +** purpose is to be a placeholder function that can be overloaded +** by a [virtual table]. +*/ +SQLITE_API int sqlite3_overload_function(sqlite3*, const char *zFuncName, int nArg); + +/* +** The interface to the virtual-table mechanism defined above (back up +** to a comment remarkably similar to this one) is currently considered +** to be experimental. The interface might change in incompatible ways. +** If this is a problem for you, do not use the interface at this time. +** +** When the virtual-table mechanism stabilizes, we will declare the +** interface fixed, support it indefinitely, and remove this comment. +*/ + +/* +** CAPI3REF: A Handle To An Open BLOB +** KEYWORDS: {BLOB handle} {BLOB handles} +** +** An instance of this object represents an open BLOB on which +** [sqlite3_blob_open | incremental BLOB I/O] can be performed. +** ^Objects of this type are created by [sqlite3_blob_open()] +** and destroyed by [sqlite3_blob_close()]. +** ^The [sqlite3_blob_read()] and [sqlite3_blob_write()] interfaces +** can be used to read or write small subsections of the BLOB. +** ^The [sqlite3_blob_bytes()] interface returns the size of the BLOB in bytes. +*/ +typedef struct sqlite3_blob sqlite3_blob; + +/* +** CAPI3REF: Open A BLOB For Incremental I/O +** METHOD: sqlite3 +** CONSTRUCTOR: sqlite3_blob +** +** ^(This interfaces opens a [BLOB handle | handle] to the BLOB located +** in row iRow, column zColumn, table zTable in database zDb; +** in other words, the same BLOB that would be selected by: +** +** 
+**     SELECT zColumn FROM zDb.zTable WHERE [rowid] = iRow;
+**
)^ +** +** ^(Parameter zDb is not the filename that contains the database, but +** rather the symbolic name of the database. For attached databases, this is +** the name that appears after the AS keyword in the [ATTACH] statement. +** For the main database file, the database name is "main". For TEMP +** tables, the database name is "temp".)^ +** +** ^If the flags parameter is non-zero, then the BLOB is opened for read +** and write access. ^If the flags parameter is zero, the BLOB is opened for +** read-only access. +** +** ^(On success, [SQLITE_OK] is returned and the new [BLOB handle] is stored +** in *ppBlob. Otherwise an [error code] is returned and, unless the error +** code is SQLITE_MISUSE, *ppBlob is set to NULL.)^ ^This means that, provided +** the API is not misused, it is always safe to call [sqlite3_blob_close()] +** on *ppBlob after this function it returns. +** +** This function fails with SQLITE_ERROR if any of the following are true: +**
    +**
  • ^(Database zDb does not exist)^, +**
  • ^(Table zTable does not exist within database zDb)^, +**
  • ^(Table zTable is a WITHOUT ROWID table)^, +**
  • ^(Column zColumn does not exist)^, +**
  • ^(Row iRow is not present in the table)^, +**
  • ^(The specified column of row iRow contains a value that is not +** a TEXT or BLOB value)^, +**
  • ^(Column zColumn is part of an index, PRIMARY KEY or UNIQUE +** constraint and the blob is being opened for read/write access)^, +**
  • ^([foreign key constraints | Foreign key constraints] are enabled, +** column zColumn is part of a [child key] definition and the blob is +** being opened for read/write access)^. +**
+** +** ^Unless it returns SQLITE_MISUSE, this function sets the +** [database connection] error code and message accessible via +** [sqlite3_errcode()] and [sqlite3_errmsg()] and related functions. +** +** A BLOB referenced by sqlite3_blob_open() may be read using the +** [sqlite3_blob_read()] interface and modified by using +** [sqlite3_blob_write()]. The [BLOB handle] can be moved to a +** different row of the same table using the [sqlite3_blob_reopen()] +** interface. However, the column, table, or database of a [BLOB handle] +** cannot be changed after the [BLOB handle] is opened. +** +** ^(If the row that a BLOB handle points to is modified by an +** [UPDATE], [DELETE], or by [ON CONFLICT] side-effects +** then the BLOB handle is marked as "expired". +** This is true if any column of the row is changed, even a column +** other than the one the BLOB handle is open on.)^ +** ^Calls to [sqlite3_blob_read()] and [sqlite3_blob_write()] for +** an expired BLOB handle fail with a return code of [SQLITE_ABORT]. +** ^(Changes written into a BLOB prior to the BLOB expiring are not +** rolled back by the expiration of the BLOB. Such changes will eventually +** commit if the transaction continues to completion.)^ +** +** ^Use the [sqlite3_blob_bytes()] interface to determine the size of +** the opened blob. ^The size of a blob may not be changed by this +** interface. Use the [UPDATE] SQL command to change the size of a +** blob. +** +** ^The [sqlite3_bind_zeroblob()] and [sqlite3_result_zeroblob()] interfaces +** and the built-in [zeroblob] SQL function may be used to create a +** zero-filled blob to read or write using the incremental-blob interface. +** +** To avoid a resource leak, every open [BLOB handle] should eventually +** be released by a call to [sqlite3_blob_close()]. +** +** See also: [sqlite3_blob_close()], +** [sqlite3_blob_reopen()], [sqlite3_blob_read()], +** [sqlite3_blob_bytes()], [sqlite3_blob_write()]. +*/ +SQLITE_API int sqlite3_blob_open( + sqlite3*, + const char *zDb, + const char *zTable, + const char *zColumn, + sqlite3_int64 iRow, + int flags, + sqlite3_blob **ppBlob +); + +/* +** CAPI3REF: Move a BLOB Handle to a New Row +** METHOD: sqlite3_blob +** +** ^This function is used to move an existing [BLOB handle] so that it points +** to a different row of the same database table. ^The new row is identified +** by the rowid value passed as the second argument. Only the row can be +** changed. ^The database, table and column on which the blob handle is open +** remain the same. Moving an existing [BLOB handle] to a new row is +** faster than closing the existing handle and opening a new one. +** +** ^(The new row must meet the same criteria as for [sqlite3_blob_open()] - +** it must exist and there must be either a blob or text value stored in +** the nominated column.)^ ^If the new row is not present in the table, or if +** it does not contain a blob or text value, or if another error occurs, an +** SQLite error code is returned and the blob handle is considered aborted. +** ^All subsequent calls to [sqlite3_blob_read()], [sqlite3_blob_write()] or +** [sqlite3_blob_reopen()] on an aborted blob handle immediately return +** SQLITE_ABORT. ^Calling [sqlite3_blob_bytes()] on an aborted blob handle +** always returns zero. +** +** ^This function sets the database handle error code and message. +*/ +SQLITE_API int sqlite3_blob_reopen(sqlite3_blob *, sqlite3_int64); + +/* +** CAPI3REF: Close A BLOB Handle +** DESTRUCTOR: sqlite3_blob +** +** ^This function closes an open [BLOB handle]. ^(The BLOB handle is closed +** unconditionally. Even if this routine returns an error code, the +** handle is still closed.)^ +** +** ^If the blob handle being closed was opened for read-write access, and if +** the database is in auto-commit mode and there are no other open read-write +** blob handles or active write statements, the current transaction is +** committed. ^If an error occurs while committing the transaction, an error +** code is returned and the transaction rolled back. +** +** Calling this function with an argument that is not a NULL pointer or an +** open blob handle results in undefined behaviour. ^Calling this routine +** with a null pointer (such as would be returned by a failed call to +** [sqlite3_blob_open()]) is a harmless no-op. ^Otherwise, if this function +** is passed a valid open blob handle, the values returned by the +** sqlite3_errcode() and sqlite3_errmsg() functions are set before returning. +*/ +SQLITE_API int sqlite3_blob_close(sqlite3_blob *); + +/* +** CAPI3REF: Return The Size Of An Open BLOB +** METHOD: sqlite3_blob +** +** ^Returns the size in bytes of the BLOB accessible via the +** successfully opened [BLOB handle] in its only argument. ^The +** incremental blob I/O routines can only read or overwriting existing +** blob content; they cannot change the size of a blob. +** +** This routine only works on a [BLOB handle] which has been created +** by a prior successful call to [sqlite3_blob_open()] and which has not +** been closed by [sqlite3_blob_close()]. Passing any other pointer in +** to this routine results in undefined and probably undesirable behavior. +*/ +SQLITE_API int sqlite3_blob_bytes(sqlite3_blob *); + +/* +** CAPI3REF: Read Data From A BLOB Incrementally +** METHOD: sqlite3_blob +** +** ^(This function is used to read data from an open [BLOB handle] into a +** caller-supplied buffer. N bytes of data are copied into buffer Z +** from the open BLOB, starting at offset iOffset.)^ +** +** ^If offset iOffset is less than N bytes from the end of the BLOB, +** [SQLITE_ERROR] is returned and no data is read. ^If N or iOffset is +** less than zero, [SQLITE_ERROR] is returned and no data is read. +** ^The size of the blob (and hence the maximum value of N+iOffset) +** can be determined using the [sqlite3_blob_bytes()] interface. +** +** ^An attempt to read from an expired [BLOB handle] fails with an +** error code of [SQLITE_ABORT]. +** +** ^(On success, sqlite3_blob_read() returns SQLITE_OK. +** Otherwise, an [error code] or an [extended error code] is returned.)^ +** +** This routine only works on a [BLOB handle] which has been created +** by a prior successful call to [sqlite3_blob_open()] and which has not +** been closed by [sqlite3_blob_close()]. Passing any other pointer in +** to this routine results in undefined and probably undesirable behavior. +** +** See also: [sqlite3_blob_write()]. +*/ +SQLITE_API int sqlite3_blob_read(sqlite3_blob *, void *Z, int N, int iOffset); + +/* +** CAPI3REF: Write Data Into A BLOB Incrementally +** METHOD: sqlite3_blob +** +** ^(This function is used to write data into an open [BLOB handle] from a +** caller-supplied buffer. N bytes of data are copied from the buffer Z +** into the open BLOB, starting at offset iOffset.)^ +** +** ^(On success, sqlite3_blob_write() returns SQLITE_OK. +** Otherwise, an [error code] or an [extended error code] is returned.)^ +** ^Unless SQLITE_MISUSE is returned, this function sets the +** [database connection] error code and message accessible via +** [sqlite3_errcode()] and [sqlite3_errmsg()] and related functions. +** +** ^If the [BLOB handle] passed as the first argument was not opened for +** writing (the flags parameter to [sqlite3_blob_open()] was zero), +** this function returns [SQLITE_READONLY]. +** +** This function may only modify the contents of the BLOB; it is +** not possible to increase the size of a BLOB using this API. +** ^If offset iOffset is less than N bytes from the end of the BLOB, +** [SQLITE_ERROR] is returned and no data is written. The size of the +** BLOB (and hence the maximum value of N+iOffset) can be determined +** using the [sqlite3_blob_bytes()] interface. ^If N or iOffset are less +** than zero [SQLITE_ERROR] is returned and no data is written. +** +** ^An attempt to write to an expired [BLOB handle] fails with an +** error code of [SQLITE_ABORT]. ^Writes to the BLOB that occurred +** before the [BLOB handle] expired are not rolled back by the +** expiration of the handle, though of course those changes might +** have been overwritten by the statement that expired the BLOB handle +** or by other independent statements. +** +** This routine only works on a [BLOB handle] which has been created +** by a prior successful call to [sqlite3_blob_open()] and which has not +** been closed by [sqlite3_blob_close()]. Passing any other pointer in +** to this routine results in undefined and probably undesirable behavior. +** +** See also: [sqlite3_blob_read()]. +*/ +SQLITE_API int sqlite3_blob_write(sqlite3_blob *, const void *z, int n, int iOffset); + +/* +** CAPI3REF: Virtual File System Objects +** +** A virtual filesystem (VFS) is an [sqlite3_vfs] object +** that SQLite uses to interact +** with the underlying operating system. Most SQLite builds come with a +** single default VFS that is appropriate for the host computer. +** New VFSes can be registered and existing VFSes can be unregistered. +** The following interfaces are provided. +** +** ^The sqlite3_vfs_find() interface returns a pointer to a VFS given its name. +** ^Names are case sensitive. +** ^Names are zero-terminated UTF-8 strings. +** ^If there is no match, a NULL pointer is returned. +** ^If zVfsName is NULL then the default VFS is returned. +** +** ^New VFSes are registered with sqlite3_vfs_register(). +** ^Each new VFS becomes the default VFS if the makeDflt flag is set. +** ^The same VFS can be registered multiple times without injury. +** ^To make an existing VFS into the default VFS, register it again +** with the makeDflt flag set. If two different VFSes with the +** same name are registered, the behavior is undefined. If a +** VFS is registered with a name that is NULL or an empty string, +** then the behavior is undefined. +** +** ^Unregister a VFS with the sqlite3_vfs_unregister() interface. +** ^(If the default VFS is unregistered, another VFS is chosen as +** the default. The choice for the new VFS is arbitrary.)^ +*/ +SQLITE_API sqlite3_vfs *sqlite3_vfs_find(const char *zVfsName); +SQLITE_API int sqlite3_vfs_register(sqlite3_vfs*, int makeDflt); +SQLITE_API int sqlite3_vfs_unregister(sqlite3_vfs*); + +/* +** CAPI3REF: Mutexes +** +** The SQLite core uses these routines for thread +** synchronization. Though they are intended for internal +** use by SQLite, code that links against SQLite is +** permitted to use any of these routines. +** +** The SQLite source code contains multiple implementations +** of these mutex routines. An appropriate implementation +** is selected automatically at compile-time. The following +** implementations are available in the SQLite core: +** +**
    +**
  • SQLITE_MUTEX_PTHREADS +**
  • SQLITE_MUTEX_W32 +**
  • SQLITE_MUTEX_NOOP +**
+** +** The SQLITE_MUTEX_NOOP implementation is a set of routines +** that does no real locking and is appropriate for use in +** a single-threaded application. The SQLITE_MUTEX_PTHREADS and +** SQLITE_MUTEX_W32 implementations are appropriate for use on Unix +** and Windows. +** +** If SQLite is compiled with the SQLITE_MUTEX_APPDEF preprocessor +** macro defined (with "-DSQLITE_MUTEX_APPDEF=1"), then no mutex +** implementation is included with the library. In this case the +** application must supply a custom mutex implementation using the +** [SQLITE_CONFIG_MUTEX] option of the sqlite3_config() function +** before calling sqlite3_initialize() or any other public sqlite3_ +** function that calls sqlite3_initialize(). +** +** ^The sqlite3_mutex_alloc() routine allocates a new +** mutex and returns a pointer to it. ^The sqlite3_mutex_alloc() +** routine returns NULL if it is unable to allocate the requested +** mutex. The argument to sqlite3_mutex_alloc() must one of these +** integer constants: +** +**
    +**
  • SQLITE_MUTEX_FAST +**
  • SQLITE_MUTEX_RECURSIVE +**
  • SQLITE_MUTEX_STATIC_MASTER +**
  • SQLITE_MUTEX_STATIC_MEM +**
  • SQLITE_MUTEX_STATIC_OPEN +**
  • SQLITE_MUTEX_STATIC_PRNG +**
  • SQLITE_MUTEX_STATIC_LRU +**
  • SQLITE_MUTEX_STATIC_PMEM +**
  • SQLITE_MUTEX_STATIC_APP1 +**
  • SQLITE_MUTEX_STATIC_APP2 +**
  • SQLITE_MUTEX_STATIC_APP3 +**
  • SQLITE_MUTEX_STATIC_VFS1 +**
  • SQLITE_MUTEX_STATIC_VFS2 +**
  • SQLITE_MUTEX_STATIC_VFS3 +**
+** +** ^The first two constants (SQLITE_MUTEX_FAST and SQLITE_MUTEX_RECURSIVE) +** cause sqlite3_mutex_alloc() to create +** a new mutex. ^The new mutex is recursive when SQLITE_MUTEX_RECURSIVE +** is used but not necessarily so when SQLITE_MUTEX_FAST is used. +** The mutex implementation does not need to make a distinction +** between SQLITE_MUTEX_RECURSIVE and SQLITE_MUTEX_FAST if it does +** not want to. SQLite will only request a recursive mutex in +** cases where it really needs one. If a faster non-recursive mutex +** implementation is available on the host platform, the mutex subsystem +** might return such a mutex in response to SQLITE_MUTEX_FAST. +** +** ^The other allowed parameters to sqlite3_mutex_alloc() (anything other +** than SQLITE_MUTEX_FAST and SQLITE_MUTEX_RECURSIVE) each return +** a pointer to a static preexisting mutex. ^Nine static mutexes are +** used by the current version of SQLite. Future versions of SQLite +** may add additional static mutexes. Static mutexes are for internal +** use by SQLite only. Applications that use SQLite mutexes should +** use only the dynamic mutexes returned by SQLITE_MUTEX_FAST or +** SQLITE_MUTEX_RECURSIVE. +** +** ^Note that if one of the dynamic mutex parameters (SQLITE_MUTEX_FAST +** or SQLITE_MUTEX_RECURSIVE) is used then sqlite3_mutex_alloc() +** returns a different mutex on every call. ^For the static +** mutex types, the same mutex is returned on every call that has +** the same type number. +** +** ^The sqlite3_mutex_free() routine deallocates a previously +** allocated dynamic mutex. Attempting to deallocate a static +** mutex results in undefined behavior. +** +** ^The sqlite3_mutex_enter() and sqlite3_mutex_try() routines attempt +** to enter a mutex. ^If another thread is already within the mutex, +** sqlite3_mutex_enter() will block and sqlite3_mutex_try() will return +** SQLITE_BUSY. ^The sqlite3_mutex_try() interface returns [SQLITE_OK] +** upon successful entry. ^(Mutexes created using +** SQLITE_MUTEX_RECURSIVE can be entered multiple times by the same thread. +** In such cases, the +** mutex must be exited an equal number of times before another thread +** can enter.)^ If the same thread tries to enter any mutex other +** than an SQLITE_MUTEX_RECURSIVE more than once, the behavior is undefined. +** +** ^(Some systems (for example, Windows 95) do not support the operation +** implemented by sqlite3_mutex_try(). On those systems, sqlite3_mutex_try() +** will always return SQLITE_BUSY. The SQLite core only ever uses +** sqlite3_mutex_try() as an optimization so this is acceptable +** behavior.)^ +** +** ^The sqlite3_mutex_leave() routine exits a mutex that was +** previously entered by the same thread. The behavior +** is undefined if the mutex is not currently entered by the +** calling thread or is not currently allocated. +** +** ^If the argument to sqlite3_mutex_enter(), sqlite3_mutex_try(), or +** sqlite3_mutex_leave() is a NULL pointer, then all three routines +** behave as no-ops. +** +** See also: [sqlite3_mutex_held()] and [sqlite3_mutex_notheld()]. +*/ +SQLITE_API sqlite3_mutex *sqlite3_mutex_alloc(int); +SQLITE_API void sqlite3_mutex_free(sqlite3_mutex*); +SQLITE_API void sqlite3_mutex_enter(sqlite3_mutex*); +SQLITE_API int sqlite3_mutex_try(sqlite3_mutex*); +SQLITE_API void sqlite3_mutex_leave(sqlite3_mutex*); + +/* +** CAPI3REF: Mutex Methods Object +** +** An instance of this structure defines the low-level routines +** used to allocate and use mutexes. +** +** Usually, the default mutex implementations provided by SQLite are +** sufficient, however the application has the option of substituting a custom +** implementation for specialized deployments or systems for which SQLite +** does not provide a suitable implementation. In this case, the application +** creates and populates an instance of this structure to pass +** to sqlite3_config() along with the [SQLITE_CONFIG_MUTEX] option. +** Additionally, an instance of this structure can be used as an +** output variable when querying the system for the current mutex +** implementation, using the [SQLITE_CONFIG_GETMUTEX] option. +** +** ^The xMutexInit method defined by this structure is invoked as +** part of system initialization by the sqlite3_initialize() function. +** ^The xMutexInit routine is called by SQLite exactly once for each +** effective call to [sqlite3_initialize()]. +** +** ^The xMutexEnd method defined by this structure is invoked as +** part of system shutdown by the sqlite3_shutdown() function. The +** implementation of this method is expected to release all outstanding +** resources obtained by the mutex methods implementation, especially +** those obtained by the xMutexInit method. ^The xMutexEnd() +** interface is invoked exactly once for each call to [sqlite3_shutdown()]. +** +** ^(The remaining seven methods defined by this structure (xMutexAlloc, +** xMutexFree, xMutexEnter, xMutexTry, xMutexLeave, xMutexHeld and +** xMutexNotheld) implement the following interfaces (respectively): +** +**
    +**
  • [sqlite3_mutex_alloc()]
    • +**
  • [sqlite3_mutex_free()]
    • +**
  • [sqlite3_mutex_enter()]
    • +**
  • [sqlite3_mutex_try()]
    • +**
  • [sqlite3_mutex_leave()]
    • +**
  • [sqlite3_mutex_held()]
    • +**
  • [sqlite3_mutex_notheld()]
    • +**
)^ +** +** The only difference is that the public sqlite3_XXX functions enumerated +** above silently ignore any invocations that pass a NULL pointer instead +** of a valid mutex handle. The implementations of the methods defined +** by this structure are not required to handle this case, the results +** of passing a NULL pointer instead of a valid mutex handle are undefined +** (i.e. it is acceptable to provide an implementation that segfaults if +** it is passed a NULL pointer). +** +** The xMutexInit() method must be threadsafe. It must be harmless to +** invoke xMutexInit() multiple times within the same process and without +** intervening calls to xMutexEnd(). Second and subsequent calls to +** xMutexInit() must be no-ops. +** +** xMutexInit() must not use SQLite memory allocation ([sqlite3_malloc()] +** and its associates). Similarly, xMutexAlloc() must not use SQLite memory +** allocation for a static mutex. ^However xMutexAlloc() may use SQLite +** memory allocation for a fast or recursive mutex. +** +** ^SQLite will invoke the xMutexEnd() method when [sqlite3_shutdown()] is +** called, but only if the prior call to xMutexInit returned SQLITE_OK. +** If xMutexInit fails in any way, it is expected to clean up after itself +** prior to returning. +*/ +typedef struct sqlite3_mutex_methods sqlite3_mutex_methods; +struct sqlite3_mutex_methods { + int (*xMutexInit)(void); + int (*xMutexEnd)(void); + sqlite3_mutex *(*xMutexAlloc)(int); + void (*xMutexFree)(sqlite3_mutex *); + void (*xMutexEnter)(sqlite3_mutex *); + int (*xMutexTry)(sqlite3_mutex *); + void (*xMutexLeave)(sqlite3_mutex *); + int (*xMutexHeld)(sqlite3_mutex *); + int (*xMutexNotheld)(sqlite3_mutex *); +}; + +/* +** CAPI3REF: Mutex Verification Routines +** +** The sqlite3_mutex_held() and sqlite3_mutex_notheld() routines +** are intended for use inside assert() statements. The SQLite core +** never uses these routines except inside an assert() and applications +** are advised to follow the lead of the core. The SQLite core only +** provides implementations for these routines when it is compiled +** with the SQLITE_DEBUG flag. External mutex implementations +** are only required to provide these routines if SQLITE_DEBUG is +** defined and if NDEBUG is not defined. +** +** These routines should return true if the mutex in their argument +** is held or not held, respectively, by the calling thread. +** +** The implementation is not required to provide versions of these +** routines that actually work. If the implementation does not provide working +** versions of these routines, it should at least provide stubs that always +** return true so that one does not get spurious assertion failures. +** +** If the argument to sqlite3_mutex_held() is a NULL pointer then +** the routine should return 1. This seems counter-intuitive since +** clearly the mutex cannot be held if it does not exist. But +** the reason the mutex does not exist is because the build is not +** using mutexes. And we do not want the assert() containing the +** call to sqlite3_mutex_held() to fail, so a non-zero return is +** the appropriate thing to do. The sqlite3_mutex_notheld() +** interface should also return 1 when given a NULL pointer. +*/ +#ifndef NDEBUG +SQLITE_API int sqlite3_mutex_held(sqlite3_mutex*); +SQLITE_API int sqlite3_mutex_notheld(sqlite3_mutex*); +#endif + +/* +** CAPI3REF: Mutex Types +** +** The [sqlite3_mutex_alloc()] interface takes a single argument +** which is one of these integer constants. +** +** The set of static mutexes may change from one SQLite release to the +** next. Applications that override the built-in mutex logic must be +** prepared to accommodate additional static mutexes. +*/ +#define SQLITE_MUTEX_FAST 0 +#define SQLITE_MUTEX_RECURSIVE 1 +#define SQLITE_MUTEX_STATIC_MASTER 2 +#define SQLITE_MUTEX_STATIC_MEM 3 /* sqlite3_malloc() */ +#define SQLITE_MUTEX_STATIC_MEM2 4 /* NOT USED */ +#define SQLITE_MUTEX_STATIC_OPEN 4 /* sqlite3BtreeOpen() */ +#define SQLITE_MUTEX_STATIC_PRNG 5 /* sqlite3_randomness() */ +#define SQLITE_MUTEX_STATIC_LRU 6 /* lru page list */ +#define SQLITE_MUTEX_STATIC_LRU2 7 /* NOT USED */ +#define SQLITE_MUTEX_STATIC_PMEM 7 /* sqlite3PageMalloc() */ +#define SQLITE_MUTEX_STATIC_APP1 8 /* For use by application */ +#define SQLITE_MUTEX_STATIC_APP2 9 /* For use by application */ +#define SQLITE_MUTEX_STATIC_APP3 10 /* For use by application */ +#define SQLITE_MUTEX_STATIC_VFS1 11 /* For use by built-in VFS */ +#define SQLITE_MUTEX_STATIC_VFS2 12 /* For use by extension VFS */ +#define SQLITE_MUTEX_STATIC_VFS3 13 /* For use by application VFS */ + +/* +** CAPI3REF: Retrieve the mutex for a database connection +** METHOD: sqlite3 +** +** ^This interface returns a pointer the [sqlite3_mutex] object that +** serializes access to the [database connection] given in the argument +** when the [threading mode] is Serialized. +** ^If the [threading mode] is Single-thread or Multi-thread then this +** routine returns a NULL pointer. +*/ +SQLITE_API sqlite3_mutex *sqlite3_db_mutex(sqlite3*); + +/* +** CAPI3REF: Low-Level Control Of Database Files +** METHOD: sqlite3 +** KEYWORDS: {file control} +** +** ^The [sqlite3_file_control()] interface makes a direct call to the +** xFileControl method for the [sqlite3_io_methods] object associated +** with a particular database identified by the second argument. ^The +** name of the database is "main" for the main database or "temp" for the +** TEMP database, or the name that appears after the AS keyword for +** databases that are added using the [ATTACH] SQL command. +** ^A NULL pointer can be used in place of "main" to refer to the +** main database file. +** ^The third and fourth parameters to this routine +** are passed directly through to the second and third parameters of +** the xFileControl method. ^The return value of the xFileControl +** method becomes the return value of this routine. +** +** A few opcodes for [sqlite3_file_control()] are handled directly +** by the SQLite core and never invoke the +** sqlite3_io_methods.xFileControl method. +** ^The [SQLITE_FCNTL_FILE_POINTER] value for the op parameter causes +** a pointer to the underlying [sqlite3_file] object to be written into +** the space pointed to by the 4th parameter. The +** [SQLITE_FCNTL_JOURNAL_POINTER] works similarly except that it returns +** the [sqlite3_file] object associated with the journal file instead of +** the main database. The [SQLITE_FCNTL_VFS_POINTER] opcode returns +** a pointer to the underlying [sqlite3_vfs] object for the file. +** The [SQLITE_FCNTL_DATA_VERSION] returns the data version counter +** from the pager. +** +** ^If the second parameter (zDbName) does not match the name of any +** open database file, then SQLITE_ERROR is returned. ^This error +** code is not remembered and will not be recalled by [sqlite3_errcode()] +** or [sqlite3_errmsg()]. The underlying xFileControl method might +** also return SQLITE_ERROR. There is no way to distinguish between +** an incorrect zDbName and an SQLITE_ERROR return from the underlying +** xFileControl method. +** +** See also: [file control opcodes] +*/ +SQLITE_API int sqlite3_file_control(sqlite3*, const char *zDbName, int op, void*); + +/* +** CAPI3REF: Testing Interface +** +** ^The sqlite3_test_control() interface is used to read out internal +** state of SQLite and to inject faults into SQLite for testing +** purposes. ^The first parameter is an operation code that determines +** the number, meaning, and operation of all subsequent parameters. +** +** This interface is not for use by applications. It exists solely +** for verifying the correct operation of the SQLite library. Depending +** on how the SQLite library is compiled, this interface might not exist. +** +** The details of the operation codes, their meanings, the parameters +** they take, and what they do are all subject to change without notice. +** Unlike most of the SQLite API, this function is not guaranteed to +** operate consistently from one release to the next. +*/ +SQLITE_API int sqlite3_test_control(int op, ...); + +/* +** CAPI3REF: Testing Interface Operation Codes +** +** These constants are the valid operation code parameters used +** as the first argument to [sqlite3_test_control()]. +** +** These parameters and their meanings are subject to change +** without notice. These values are for testing purposes only. +** Applications should not use any of these parameters or the +** [sqlite3_test_control()] interface. +*/ +#define SQLITE_TESTCTRL_FIRST 5 +#define SQLITE_TESTCTRL_PRNG_SAVE 5 +#define SQLITE_TESTCTRL_PRNG_RESTORE 6 +#define SQLITE_TESTCTRL_PRNG_RESET 7 /* NOT USED */ +#define SQLITE_TESTCTRL_BITVEC_TEST 8 +#define SQLITE_TESTCTRL_FAULT_INSTALL 9 +#define SQLITE_TESTCTRL_BENIGN_MALLOC_HOOKS 10 +#define SQLITE_TESTCTRL_PENDING_BYTE 11 +#define SQLITE_TESTCTRL_ASSERT 12 +#define SQLITE_TESTCTRL_ALWAYS 13 +#define SQLITE_TESTCTRL_RESERVE 14 +#define SQLITE_TESTCTRL_OPTIMIZATIONS 15 +#define SQLITE_TESTCTRL_ISKEYWORD 16 /* NOT USED */ +#define SQLITE_TESTCTRL_SCRATCHMALLOC 17 /* NOT USED */ +#define SQLITE_TESTCTRL_INTERNAL_FUNCTIONS 17 +#define SQLITE_TESTCTRL_LOCALTIME_FAULT 18 +#define SQLITE_TESTCTRL_EXPLAIN_STMT 19 /* NOT USED */ +#define SQLITE_TESTCTRL_ONCE_RESET_THRESHOLD 19 +#define SQLITE_TESTCTRL_NEVER_CORRUPT 20 +#define SQLITE_TESTCTRL_VDBE_COVERAGE 21 +#define SQLITE_TESTCTRL_BYTEORDER 22 +#define SQLITE_TESTCTRL_ISINIT 23 +#define SQLITE_TESTCTRL_SORTER_MMAP 24 +#define SQLITE_TESTCTRL_IMPOSTER 25 +#define SQLITE_TESTCTRL_PARSER_COVERAGE 26 +#define SQLITE_TESTCTRL_RESULT_INTREAL 27 +#define SQLITE_TESTCTRL_PRNG_SEED 28 +#define SQLITE_TESTCTRL_EXTRA_SCHEMA_CHECKS 29 +#define SQLITE_TESTCTRL_LAST 29 /* Largest TESTCTRL */ + +/* +** CAPI3REF: SQL Keyword Checking +** +** These routines provide access to the set of SQL language keywords +** recognized by SQLite. Applications can uses these routines to determine +** whether or not a specific identifier needs to be escaped (for example, +** by enclosing in double-quotes) so as not to confuse the parser. +** +** The sqlite3_keyword_count() interface returns the number of distinct +** keywords understood by SQLite. +** +** The sqlite3_keyword_name(N,Z,L) interface finds the N-th keyword and +** makes *Z point to that keyword expressed as UTF8 and writes the number +** of bytes in the keyword into *L. The string that *Z points to is not +** zero-terminated. The sqlite3_keyword_name(N,Z,L) routine returns +** SQLITE_OK if N is within bounds and SQLITE_ERROR if not. If either Z +** or L are NULL or invalid pointers then calls to +** sqlite3_keyword_name(N,Z,L) result in undefined behavior. +** +** The sqlite3_keyword_check(Z,L) interface checks to see whether or not +** the L-byte UTF8 identifier that Z points to is a keyword, returning non-zero +** if it is and zero if not. +** +** The parser used by SQLite is forgiving. It is often possible to use +** a keyword as an identifier as long as such use does not result in a +** parsing ambiguity. For example, the statement +** "CREATE TABLE BEGIN(REPLACE,PRAGMA,END);" is accepted by SQLite, and +** creates a new table named "BEGIN" with three columns named +** "REPLACE", "PRAGMA", and "END". Nevertheless, best practice is to avoid +** using keywords as identifiers. Common techniques used to avoid keyword +** name collisions include: +**
    +**
  • Put all identifier names inside double-quotes. This is the official +** SQL way to escape identifier names. +**
  • Put identifier names inside [...]. This is not standard SQL, +** but it is what SQL Server does and so lots of programmers use this +** technique. +**
  • Begin every identifier with the letter "Z" as no SQL keywords start +** with "Z". +**
  • Include a digit somewhere in every identifier name. +**
+** +** Note that the number of keywords understood by SQLite can depend on +** compile-time options. For example, "VACUUM" is not a keyword if +** SQLite is compiled with the [-DSQLITE_OMIT_VACUUM] option. Also, +** new keywords may be added to future releases of SQLite. +*/ +SQLITE_API int sqlite3_keyword_count(void); +SQLITE_API int sqlite3_keyword_name(int,const char**,int*); +SQLITE_API int sqlite3_keyword_check(const char*,int); + +/* +** CAPI3REF: Dynamic String Object +** KEYWORDS: {dynamic string} +** +** An instance of the sqlite3_str object contains a dynamically-sized +** string under construction. +** +** The lifecycle of an sqlite3_str object is as follows: +**
    +**
  1. ^The sqlite3_str object is created using [sqlite3_str_new()]. +**
  2. ^Text is appended to the sqlite3_str object using various +** methods, such as [sqlite3_str_appendf()]. +**
  3. ^The sqlite3_str object is destroyed and the string it created +** is returned using the [sqlite3_str_finish()] interface. +**
+*/ +typedef struct sqlite3_str sqlite3_str; + +/* +** CAPI3REF: Create A New Dynamic String Object +** CONSTRUCTOR: sqlite3_str +** +** ^The [sqlite3_str_new(D)] interface allocates and initializes +** a new [sqlite3_str] object. To avoid memory leaks, the object returned by +** [sqlite3_str_new()] must be freed by a subsequent call to +** [sqlite3_str_finish(X)]. +** +** ^The [sqlite3_str_new(D)] interface always returns a pointer to a +** valid [sqlite3_str] object, though in the event of an out-of-memory +** error the returned object might be a special singleton that will +** silently reject new text, always return SQLITE_NOMEM from +** [sqlite3_str_errcode()], always return 0 for +** [sqlite3_str_length()], and always return NULL from +** [sqlite3_str_finish(X)]. It is always safe to use the value +** returned by [sqlite3_str_new(D)] as the sqlite3_str parameter +** to any of the other [sqlite3_str] methods. +** +** The D parameter to [sqlite3_str_new(D)] may be NULL. If the +** D parameter in [sqlite3_str_new(D)] is not NULL, then the maximum +** length of the string contained in the [sqlite3_str] object will be +** the value set for [sqlite3_limit](D,[SQLITE_LIMIT_LENGTH]) instead +** of [SQLITE_MAX_LENGTH]. +*/ +SQLITE_API sqlite3_str *sqlite3_str_new(sqlite3*); + +/* +** CAPI3REF: Finalize A Dynamic String +** DESTRUCTOR: sqlite3_str +** +** ^The [sqlite3_str_finish(X)] interface destroys the sqlite3_str object X +** and returns a pointer to a memory buffer obtained from [sqlite3_malloc64()] +** that contains the constructed string. The calling application should +** pass the returned value to [sqlite3_free()] to avoid a memory leak. +** ^The [sqlite3_str_finish(X)] interface may return a NULL pointer if any +** errors were encountered during construction of the string. ^The +** [sqlite3_str_finish(X)] interface will also return a NULL pointer if the +** string in [sqlite3_str] object X is zero bytes long. +*/ +SQLITE_API char *sqlite3_str_finish(sqlite3_str*); + +/* +** CAPI3REF: Add Content To A Dynamic String +** METHOD: sqlite3_str +** +** These interfaces add content to an sqlite3_str object previously obtained +** from [sqlite3_str_new()]. +** +** ^The [sqlite3_str_appendf(X,F,...)] and +** [sqlite3_str_vappendf(X,F,V)] interfaces uses the [built-in printf] +** functionality of SQLite to append formatted text onto the end of +** [sqlite3_str] object X. +** +** ^The [sqlite3_str_append(X,S,N)] method appends exactly N bytes from string S +** onto the end of the [sqlite3_str] object X. N must be non-negative. +** S must contain at least N non-zero bytes of content. To append a +** zero-terminated string in its entirety, use the [sqlite3_str_appendall()] +** method instead. +** +** ^The [sqlite3_str_appendall(X,S)] method appends the complete content of +** zero-terminated string S onto the end of [sqlite3_str] object X. +** +** ^The [sqlite3_str_appendchar(X,N,C)] method appends N copies of the +** single-byte character C onto the end of [sqlite3_str] object X. +** ^This method can be used, for example, to add whitespace indentation. +** +** ^The [sqlite3_str_reset(X)] method resets the string under construction +** inside [sqlite3_str] object X back to zero bytes in length. +** +** These methods do not return a result code. ^If an error occurs, that fact +** is recorded in the [sqlite3_str] object and can be recovered by a +** subsequent call to [sqlite3_str_errcode(X)]. +*/ +SQLITE_API void sqlite3_str_appendf(sqlite3_str*, const char *zFormat, ...); +SQLITE_API void sqlite3_str_vappendf(sqlite3_str*, const char *zFormat, va_list); +SQLITE_API void sqlite3_str_append(sqlite3_str*, const char *zIn, int N); +SQLITE_API void sqlite3_str_appendall(sqlite3_str*, const char *zIn); +SQLITE_API void sqlite3_str_appendchar(sqlite3_str*, int N, char C); +SQLITE_API void sqlite3_str_reset(sqlite3_str*); + +/* +** CAPI3REF: Status Of A Dynamic String +** METHOD: sqlite3_str +** +** These interfaces return the current status of an [sqlite3_str] object. +** +** ^If any prior errors have occurred while constructing the dynamic string +** in sqlite3_str X, then the [sqlite3_str_errcode(X)] method will return +** an appropriate error code. ^The [sqlite3_str_errcode(X)] method returns +** [SQLITE_NOMEM] following any out-of-memory error, or +** [SQLITE_TOOBIG] if the size of the dynamic string exceeds +** [SQLITE_MAX_LENGTH], or [SQLITE_OK] if there have been no errors. +** +** ^The [sqlite3_str_length(X)] method returns the current length, in bytes, +** of the dynamic string under construction in [sqlite3_str] object X. +** ^The length returned by [sqlite3_str_length(X)] does not include the +** zero-termination byte. +** +** ^The [sqlite3_str_value(X)] method returns a pointer to the current +** content of the dynamic string under construction in X. The value +** returned by [sqlite3_str_value(X)] is managed by the sqlite3_str object X +** and might be freed or altered by any subsequent method on the same +** [sqlite3_str] object. Applications must not used the pointer returned +** [sqlite3_str_value(X)] after any subsequent method call on the same +** object. ^Applications may change the content of the string returned +** by [sqlite3_str_value(X)] as long as they do not write into any bytes +** outside the range of 0 to [sqlite3_str_length(X)] and do not read or +** write any byte after any subsequent sqlite3_str method call. +*/ +SQLITE_API int sqlite3_str_errcode(sqlite3_str*); +SQLITE_API int sqlite3_str_length(sqlite3_str*); +SQLITE_API char *sqlite3_str_value(sqlite3_str*); + +/* +** CAPI3REF: SQLite Runtime Status +** +** ^These interfaces are used to retrieve runtime status information +** about the performance of SQLite, and optionally to reset various +** highwater marks. ^The first argument is an integer code for +** the specific parameter to measure. ^(Recognized integer codes +** are of the form [status parameters | SQLITE_STATUS_...].)^ +** ^The current value of the parameter is returned into *pCurrent. +** ^The highest recorded value is returned in *pHighwater. ^If the +** resetFlag is true, then the highest record value is reset after +** *pHighwater is written. ^(Some parameters do not record the highest +** value. For those parameters +** nothing is written into *pHighwater and the resetFlag is ignored.)^ +** ^(Other parameters record only the highwater mark and not the current +** value. For these latter parameters nothing is written into *pCurrent.)^ +** +** ^The sqlite3_status() and sqlite3_status64() routines return +** SQLITE_OK on success and a non-zero [error code] on failure. +** +** If either the current value or the highwater mark is too large to +** be represented by a 32-bit integer, then the values returned by +** sqlite3_status() are undefined. +** +** See also: [sqlite3_db_status()] +*/ +SQLITE_API int sqlite3_status(int op, int *pCurrent, int *pHighwater, int resetFlag); +SQLITE_API int sqlite3_status64( + int op, + sqlite3_int64 *pCurrent, + sqlite3_int64 *pHighwater, + int resetFlag +); + + +/* +** CAPI3REF: Status Parameters +** KEYWORDS: {status parameters} +** +** These integer constants designate various run-time status parameters +** that can be returned by [sqlite3_status()]. +** +**
+** [[SQLITE_STATUS_MEMORY_USED]] ^(
SQLITE_STATUS_MEMORY_USED
+**
This parameter is the current amount of memory checked out +** using [sqlite3_malloc()], either directly or indirectly. The +** figure includes calls made to [sqlite3_malloc()] by the application +** and internal memory usage by the SQLite library. Auxiliary page-cache +** memory controlled by [SQLITE_CONFIG_PAGECACHE] is not included in +** this parameter. The amount returned is the sum of the allocation +** sizes as reported by the xSize method in [sqlite3_mem_methods].
)^ +** +** [[SQLITE_STATUS_MALLOC_SIZE]] ^(
SQLITE_STATUS_MALLOC_SIZE
+**
This parameter records the largest memory allocation request +** handed to [sqlite3_malloc()] or [sqlite3_realloc()] (or their +** internal equivalents). Only the value returned in the +** *pHighwater parameter to [sqlite3_status()] is of interest. +** The value written into the *pCurrent parameter is undefined.
)^ +** +** [[SQLITE_STATUS_MALLOC_COUNT]] ^(
SQLITE_STATUS_MALLOC_COUNT
+**
This parameter records the number of separate memory allocations +** currently checked out.
)^ +** +** [[SQLITE_STATUS_PAGECACHE_USED]] ^(
SQLITE_STATUS_PAGECACHE_USED
+**
This parameter returns the number of pages used out of the +** [pagecache memory allocator] that was configured using +** [SQLITE_CONFIG_PAGECACHE]. The +** value returned is in pages, not in bytes.
)^ +** +** [[SQLITE_STATUS_PAGECACHE_OVERFLOW]] +** ^(
SQLITE_STATUS_PAGECACHE_OVERFLOW
+**
This parameter returns the number of bytes of page cache +** allocation which could not be satisfied by the [SQLITE_CONFIG_PAGECACHE] +** buffer and where forced to overflow to [sqlite3_malloc()]. The +** returned value includes allocations that overflowed because they +** where too large (they were larger than the "sz" parameter to +** [SQLITE_CONFIG_PAGECACHE]) and allocations that overflowed because +** no space was left in the page cache.
)^ +** +** [[SQLITE_STATUS_PAGECACHE_SIZE]] ^(
SQLITE_STATUS_PAGECACHE_SIZE
+**
This parameter records the largest memory allocation request +** handed to [pagecache memory allocator]. Only the value returned in the +** *pHighwater parameter to [sqlite3_status()] is of interest. +** The value written into the *pCurrent parameter is undefined.
)^ +** +** [[SQLITE_STATUS_SCRATCH_USED]]
SQLITE_STATUS_SCRATCH_USED
+**
No longer used.
+** +** [[SQLITE_STATUS_SCRATCH_OVERFLOW]] ^(
SQLITE_STATUS_SCRATCH_OVERFLOW
+**
No longer used.
+** +** [[SQLITE_STATUS_SCRATCH_SIZE]]
SQLITE_STATUS_SCRATCH_SIZE
+**
No longer used.
+** +** [[SQLITE_STATUS_PARSER_STACK]] ^(
SQLITE_STATUS_PARSER_STACK
+**
The *pHighwater parameter records the deepest parser stack. +** The *pCurrent value is undefined. The *pHighwater value is only +** meaningful if SQLite is compiled with [YYTRACKMAXSTACKDEPTH].
)^ +**
+** +** New status parameters may be added from time to time. +*/ +#define SQLITE_STATUS_MEMORY_USED 0 +#define SQLITE_STATUS_PAGECACHE_USED 1 +#define SQLITE_STATUS_PAGECACHE_OVERFLOW 2 +#define SQLITE_STATUS_SCRATCH_USED 3 /* NOT USED */ +#define SQLITE_STATUS_SCRATCH_OVERFLOW 4 /* NOT USED */ +#define SQLITE_STATUS_MALLOC_SIZE 5 +#define SQLITE_STATUS_PARSER_STACK 6 +#define SQLITE_STATUS_PAGECACHE_SIZE 7 +#define SQLITE_STATUS_SCRATCH_SIZE 8 /* NOT USED */ +#define SQLITE_STATUS_MALLOC_COUNT 9 + +/* +** CAPI3REF: Database Connection Status +** METHOD: sqlite3 +** +** ^This interface is used to retrieve runtime status information +** about a single [database connection]. ^The first argument is the +** database connection object to be interrogated. ^The second argument +** is an integer constant, taken from the set of +** [SQLITE_DBSTATUS options], that +** determines the parameter to interrogate. The set of +** [SQLITE_DBSTATUS options] is likely +** to grow in future releases of SQLite. +** +** ^The current value of the requested parameter is written into *pCur +** and the highest instantaneous value is written into *pHiwtr. ^If +** the resetFlg is true, then the highest instantaneous value is +** reset back down to the current value. +** +** ^The sqlite3_db_status() routine returns SQLITE_OK on success and a +** non-zero [error code] on failure. +** +** See also: [sqlite3_status()] and [sqlite3_stmt_status()]. +*/ +SQLITE_API int sqlite3_db_status(sqlite3*, int op, int *pCur, int *pHiwtr, int resetFlg); + +/* +** CAPI3REF: Status Parameters for database connections +** KEYWORDS: {SQLITE_DBSTATUS options} +** +** These constants are the available integer "verbs" that can be passed as +** the second argument to the [sqlite3_db_status()] interface. +** +** New verbs may be added in future releases of SQLite. Existing verbs +** might be discontinued. Applications should check the return code from +** [sqlite3_db_status()] to make sure that the call worked. +** The [sqlite3_db_status()] interface will return a non-zero error code +** if a discontinued or unsupported verb is invoked. +** +**
+** [[SQLITE_DBSTATUS_LOOKASIDE_USED]] ^(
SQLITE_DBSTATUS_LOOKASIDE_USED
+**
This parameter returns the number of lookaside memory slots currently +** checked out.
)^ +** +** [[SQLITE_DBSTATUS_LOOKASIDE_HIT]] ^(
SQLITE_DBSTATUS_LOOKASIDE_HIT
+**
This parameter returns the number malloc attempts that were +** satisfied using lookaside memory. Only the high-water value is meaningful; +** the current value is always zero.)^ +** +** [[SQLITE_DBSTATUS_LOOKASIDE_MISS_SIZE]] +** ^(
SQLITE_DBSTATUS_LOOKASIDE_MISS_SIZE
+**
This parameter returns the number malloc attempts that might have +** been satisfied using lookaside memory but failed due to the amount of +** memory requested being larger than the lookaside slot size. +** Only the high-water value is meaningful; +** the current value is always zero.)^ +** +** [[SQLITE_DBSTATUS_LOOKASIDE_MISS_FULL]] +** ^(
SQLITE_DBSTATUS_LOOKASIDE_MISS_FULL
+**
This parameter returns the number malloc attempts that might have +** been satisfied using lookaside memory but failed due to all lookaside +** memory already being in use. +** Only the high-water value is meaningful; +** the current value is always zero.)^ +** +** [[SQLITE_DBSTATUS_CACHE_USED]] ^(
SQLITE_DBSTATUS_CACHE_USED
+**
This parameter returns the approximate number of bytes of heap +** memory used by all pager caches associated with the database connection.)^ +** ^The highwater mark associated with SQLITE_DBSTATUS_CACHE_USED is always 0. +** +** [[SQLITE_DBSTATUS_CACHE_USED_SHARED]] +** ^(
SQLITE_DBSTATUS_CACHE_USED_SHARED
+**
This parameter is similar to DBSTATUS_CACHE_USED, except that if a +** pager cache is shared between two or more connections the bytes of heap +** memory used by that pager cache is divided evenly between the attached +** connections.)^ In other words, if none of the pager caches associated +** with the database connection are shared, this request returns the same +** value as DBSTATUS_CACHE_USED. Or, if one or more or the pager caches are +** shared, the value returned by this call will be smaller than that returned +** by DBSTATUS_CACHE_USED. ^The highwater mark associated with +** SQLITE_DBSTATUS_CACHE_USED_SHARED is always 0. +** +** [[SQLITE_DBSTATUS_SCHEMA_USED]] ^(
SQLITE_DBSTATUS_SCHEMA_USED
+**
This parameter returns the approximate number of bytes of heap +** memory used to store the schema for all databases associated +** with the connection - main, temp, and any [ATTACH]-ed databases.)^ +** ^The full amount of memory used by the schemas is reported, even if the +** schema memory is shared with other database connections due to +** [shared cache mode] being enabled. +** ^The highwater mark associated with SQLITE_DBSTATUS_SCHEMA_USED is always 0. +** +** [[SQLITE_DBSTATUS_STMT_USED]] ^(
SQLITE_DBSTATUS_STMT_USED
+**
This parameter returns the approximate number of bytes of heap +** and lookaside memory used by all prepared statements associated with +** the database connection.)^ +** ^The highwater mark associated with SQLITE_DBSTATUS_STMT_USED is always 0. +**
+** +** [[SQLITE_DBSTATUS_CACHE_HIT]] ^(
SQLITE_DBSTATUS_CACHE_HIT
+**
This parameter returns the number of pager cache hits that have +** occurred.)^ ^The highwater mark associated with SQLITE_DBSTATUS_CACHE_HIT +** is always 0. +**
+** +** [[SQLITE_DBSTATUS_CACHE_MISS]] ^(
SQLITE_DBSTATUS_CACHE_MISS
+**
This parameter returns the number of pager cache misses that have +** occurred.)^ ^The highwater mark associated with SQLITE_DBSTATUS_CACHE_MISS +** is always 0. +**
+** +** [[SQLITE_DBSTATUS_CACHE_WRITE]] ^(
SQLITE_DBSTATUS_CACHE_WRITE
+**
This parameter returns the number of dirty cache entries that have +** been written to disk. Specifically, the number of pages written to the +** wal file in wal mode databases, or the number of pages written to the +** database file in rollback mode databases. Any pages written as part of +** transaction rollback or database recovery operations are not included. +** If an IO or other error occurs while writing a page to disk, the effect +** on subsequent SQLITE_DBSTATUS_CACHE_WRITE requests is undefined.)^ ^The +** highwater mark associated with SQLITE_DBSTATUS_CACHE_WRITE is always 0. +**
+** +** [[SQLITE_DBSTATUS_CACHE_SPILL]] ^(
SQLITE_DBSTATUS_CACHE_SPILL
+**
This parameter returns the number of dirty cache entries that have +** been written to disk in the middle of a transaction due to the page +** cache overflowing. Transactions are more efficient if they are written +** to disk all at once. When pages spill mid-transaction, that introduces +** additional overhead. This parameter can be used help identify +** inefficiencies that can be resolve by increasing the cache size. +**
+** +** [[SQLITE_DBSTATUS_DEFERRED_FKS]] ^(
SQLITE_DBSTATUS_DEFERRED_FKS
+**
This parameter returns zero for the current value if and only if +** all foreign key constraints (deferred or immediate) have been +** resolved.)^ ^The highwater mark is always 0. +**
+**
+*/ +#define SQLITE_DBSTATUS_LOOKASIDE_USED 0 +#define SQLITE_DBSTATUS_CACHE_USED 1 +#define SQLITE_DBSTATUS_SCHEMA_USED 2 +#define SQLITE_DBSTATUS_STMT_USED 3 +#define SQLITE_DBSTATUS_LOOKASIDE_HIT 4 +#define SQLITE_DBSTATUS_LOOKASIDE_MISS_SIZE 5 +#define SQLITE_DBSTATUS_LOOKASIDE_MISS_FULL 6 +#define SQLITE_DBSTATUS_CACHE_HIT 7 +#define SQLITE_DBSTATUS_CACHE_MISS 8 +#define SQLITE_DBSTATUS_CACHE_WRITE 9 +#define SQLITE_DBSTATUS_DEFERRED_FKS 10 +#define SQLITE_DBSTATUS_CACHE_USED_SHARED 11 +#define SQLITE_DBSTATUS_CACHE_SPILL 12 +#define SQLITE_DBSTATUS_MAX 12 /* Largest defined DBSTATUS */ + + +/* +** CAPI3REF: Prepared Statement Status +** METHOD: sqlite3_stmt +** +** ^(Each prepared statement maintains various +** [SQLITE_STMTSTATUS counters] that measure the number +** of times it has performed specific operations.)^ These counters can +** be used to monitor the performance characteristics of the prepared +** statements. For example, if the number of table steps greatly exceeds +** the number of table searches or result rows, that would tend to indicate +** that the prepared statement is using a full table scan rather than +** an index. +** +** ^(This interface is used to retrieve and reset counter values from +** a [prepared statement]. The first argument is the prepared statement +** object to be interrogated. The second argument +** is an integer code for a specific [SQLITE_STMTSTATUS counter] +** to be interrogated.)^ +** ^The current value of the requested counter is returned. +** ^If the resetFlg is true, then the counter is reset to zero after this +** interface call returns. +** +** See also: [sqlite3_status()] and [sqlite3_db_status()]. +*/ +SQLITE_API int sqlite3_stmt_status(sqlite3_stmt*, int op,int resetFlg); + +/* +** CAPI3REF: Status Parameters for prepared statements +** KEYWORDS: {SQLITE_STMTSTATUS counter} {SQLITE_STMTSTATUS counters} +** +** These preprocessor macros define integer codes that name counter +** values associated with the [sqlite3_stmt_status()] interface. +** The meanings of the various counters are as follows: +** +**
+** [[SQLITE_STMTSTATUS_FULLSCAN_STEP]]
SQLITE_STMTSTATUS_FULLSCAN_STEP
+**
^This is the number of times that SQLite has stepped forward in +** a table as part of a full table scan. Large numbers for this counter +** may indicate opportunities for performance improvement through +** careful use of indices.
+** +** [[SQLITE_STMTSTATUS_SORT]]
SQLITE_STMTSTATUS_SORT
+**
^This is the number of sort operations that have occurred. +** A non-zero value in this counter may indicate an opportunity to +** improvement performance through careful use of indices.
+** +** [[SQLITE_STMTSTATUS_AUTOINDEX]]
SQLITE_STMTSTATUS_AUTOINDEX
+**
^This is the number of rows inserted into transient indices that +** were created automatically in order to help joins run faster. +** A non-zero value in this counter may indicate an opportunity to +** improvement performance by adding permanent indices that do not +** need to be reinitialized each time the statement is run.
+** +** [[SQLITE_STMTSTATUS_VM_STEP]]
SQLITE_STMTSTATUS_VM_STEP
+**
^This is the number of virtual machine operations executed +** by the prepared statement if that number is less than or equal +** to 2147483647. The number of virtual machine operations can be +** used as a proxy for the total work done by the prepared statement. +** If the number of virtual machine operations exceeds 2147483647 +** then the value returned by this statement status code is undefined. +** +** [[SQLITE_STMTSTATUS_REPREPARE]]
SQLITE_STMTSTATUS_REPREPARE
+**
^This is the number of times that the prepare statement has been +** automatically regenerated due to schema changes or change to +** [bound parameters] that might affect the query plan. +** +** [[SQLITE_STMTSTATUS_RUN]]
SQLITE_STMTSTATUS_RUN
+**
^This is the number of times that the prepared statement has +** been run. A single "run" for the purposes of this counter is one +** or more calls to [sqlite3_step()] followed by a call to [sqlite3_reset()]. +** The counter is incremented on the first [sqlite3_step()] call of each +** cycle. +** +** [[SQLITE_STMTSTATUS_MEMUSED]]
SQLITE_STMTSTATUS_MEMUSED
+**
^This is the approximate number of bytes of heap memory +** used to store the prepared statement. ^This value is not actually +** a counter, and so the resetFlg parameter to sqlite3_stmt_status() +** is ignored when the opcode is SQLITE_STMTSTATUS_MEMUSED. +**
+**
+*/ +#define SQLITE_STMTSTATUS_FULLSCAN_STEP 1 +#define SQLITE_STMTSTATUS_SORT 2 +#define SQLITE_STMTSTATUS_AUTOINDEX 3 +#define SQLITE_STMTSTATUS_VM_STEP 4 +#define SQLITE_STMTSTATUS_REPREPARE 5 +#define SQLITE_STMTSTATUS_RUN 6 +#define SQLITE_STMTSTATUS_MEMUSED 99 + +/* +** CAPI3REF: Custom Page Cache Object +** +** The sqlite3_pcache type is opaque. It is implemented by +** the pluggable module. The SQLite core has no knowledge of +** its size or internal structure and never deals with the +** sqlite3_pcache object except by holding and passing pointers +** to the object. +** +** See [sqlite3_pcache_methods2] for additional information. +*/ +typedef struct sqlite3_pcache sqlite3_pcache; + +/* +** CAPI3REF: Custom Page Cache Object +** +** The sqlite3_pcache_page object represents a single page in the +** page cache. The page cache will allocate instances of this +** object. Various methods of the page cache use pointers to instances +** of this object as parameters or as their return value. +** +** See [sqlite3_pcache_methods2] for additional information. +*/ +typedef struct sqlite3_pcache_page sqlite3_pcache_page; +struct sqlite3_pcache_page { + void *pBuf; /* The content of the page */ + void *pExtra; /* Extra information associated with the page */ +}; + +/* +** CAPI3REF: Application Defined Page Cache. +** KEYWORDS: {page cache} +** +** ^(The [sqlite3_config]([SQLITE_CONFIG_PCACHE2], ...) interface can +** register an alternative page cache implementation by passing in an +** instance of the sqlite3_pcache_methods2 structure.)^ +** In many applications, most of the heap memory allocated by +** SQLite is used for the page cache. +** By implementing a +** custom page cache using this API, an application can better control +** the amount of memory consumed by SQLite, the way in which +** that memory is allocated and released, and the policies used to +** determine exactly which parts of a database file are cached and for +** how long. +** +** The alternative page cache mechanism is an +** extreme measure that is only needed by the most demanding applications. +** The built-in page cache is recommended for most uses. +** +** ^(The contents of the sqlite3_pcache_methods2 structure are copied to an +** internal buffer by SQLite within the call to [sqlite3_config]. Hence +** the application may discard the parameter after the call to +** [sqlite3_config()] returns.)^ +** +** [[the xInit() page cache method]] +** ^(The xInit() method is called once for each effective +** call to [sqlite3_initialize()])^ +** (usually only once during the lifetime of the process). ^(The xInit() +** method is passed a copy of the sqlite3_pcache_methods2.pArg value.)^ +** The intent of the xInit() method is to set up global data structures +** required by the custom page cache implementation. +** ^(If the xInit() method is NULL, then the +** built-in default page cache is used instead of the application defined +** page cache.)^ +** +** [[the xShutdown() page cache method]] +** ^The xShutdown() method is called by [sqlite3_shutdown()]. +** It can be used to clean up +** any outstanding resources before process shutdown, if required. +** ^The xShutdown() method may be NULL. +** +** ^SQLite automatically serializes calls to the xInit method, +** so the xInit method need not be threadsafe. ^The +** xShutdown method is only called from [sqlite3_shutdown()] so it does +** not need to be threadsafe either. All other methods must be threadsafe +** in multithreaded applications. +** +** ^SQLite will never invoke xInit() more than once without an intervening +** call to xShutdown(). +** +** [[the xCreate() page cache methods]] +** ^SQLite invokes the xCreate() method to construct a new cache instance. +** SQLite will typically create one cache instance for each open database file, +** though this is not guaranteed. ^The +** first parameter, szPage, is the size in bytes of the pages that must +** be allocated by the cache. ^szPage will always a power of two. ^The +** second parameter szExtra is a number of bytes of extra storage +** associated with each page cache entry. ^The szExtra parameter will +** a number less than 250. SQLite will use the +** extra szExtra bytes on each page to store metadata about the underlying +** database page on disk. The value passed into szExtra depends +** on the SQLite version, the target platform, and how SQLite was compiled. +** ^The third argument to xCreate(), bPurgeable, is true if the cache being +** created will be used to cache database pages of a file stored on disk, or +** false if it is used for an in-memory database. The cache implementation +** does not have to do anything special based with the value of bPurgeable; +** it is purely advisory. ^On a cache where bPurgeable is false, SQLite will +** never invoke xUnpin() except to deliberately delete a page. +** ^In other words, calls to xUnpin() on a cache with bPurgeable set to +** false will always have the "discard" flag set to true. +** ^Hence, a cache created with bPurgeable false will +** never contain any unpinned pages. +** +** [[the xCachesize() page cache method]] +** ^(The xCachesize() method may be called at any time by SQLite to set the +** suggested maximum cache-size (number of pages stored by) the cache +** instance passed as the first argument. This is the value configured using +** the SQLite "[PRAGMA cache_size]" command.)^ As with the bPurgeable +** parameter, the implementation is not required to do anything with this +** value; it is advisory only. +** +** [[the xPagecount() page cache methods]] +** The xPagecount() method must return the number of pages currently +** stored in the cache, both pinned and unpinned. +** +** [[the xFetch() page cache methods]] +** The xFetch() method locates a page in the cache and returns a pointer to +** an sqlite3_pcache_page object associated with that page, or a NULL pointer. +** The pBuf element of the returned sqlite3_pcache_page object will be a +** pointer to a buffer of szPage bytes used to store the content of a +** single database page. The pExtra element of sqlite3_pcache_page will be +** a pointer to the szExtra bytes of extra storage that SQLite has requested +** for each entry in the page cache. +** +** The page to be fetched is determined by the key. ^The minimum key value +** is 1. After it has been retrieved using xFetch, the page is considered +** to be "pinned". +** +** If the requested page is already in the page cache, then the page cache +** implementation must return a pointer to the page buffer with its content +** intact. If the requested page is not already in the cache, then the +** cache implementation should use the value of the createFlag +** parameter to help it determined what action to take: +** +** +**
createFlag Behavior when page is not already in cache +**
0 Do not allocate a new page. Return NULL. +**
1 Allocate a new page if it easy and convenient to do so. +** Otherwise return NULL. +**
2 Make every effort to allocate a new page. Only return +** NULL if allocating a new page is effectively impossible. +**
+** +** ^(SQLite will normally invoke xFetch() with a createFlag of 0 or 1. SQLite +** will only use a createFlag of 2 after a prior call with a createFlag of 1 +** failed.)^ In between the to xFetch() calls, SQLite may +** attempt to unpin one or more cache pages by spilling the content of +** pinned pages to disk and synching the operating system disk cache. +** +** [[the xUnpin() page cache method]] +** ^xUnpin() is called by SQLite with a pointer to a currently pinned page +** as its second argument. If the third parameter, discard, is non-zero, +** then the page must be evicted from the cache. +** ^If the discard parameter is +** zero, then the page may be discarded or retained at the discretion of +** page cache implementation. ^The page cache implementation +** may choose to evict unpinned pages at any time. +** +** The cache must not perform any reference counting. A single +** call to xUnpin() unpins the page regardless of the number of prior calls +** to xFetch(). +** +** [[the xRekey() page cache methods]] +** The xRekey() method is used to change the key value associated with the +** page passed as the second argument. If the cache +** previously contains an entry associated with newKey, it must be +** discarded. ^Any prior cache entry associated with newKey is guaranteed not +** to be pinned. +** +** When SQLite calls the xTruncate() method, the cache must discard all +** existing cache entries with page numbers (keys) greater than or equal +** to the value of the iLimit parameter passed to xTruncate(). If any +** of these pages are pinned, they are implicitly unpinned, meaning that +** they can be safely discarded. +** +** [[the xDestroy() page cache method]] +** ^The xDestroy() method is used to delete a cache allocated by xCreate(). +** All resources associated with the specified cache should be freed. ^After +** calling the xDestroy() method, SQLite considers the [sqlite3_pcache*] +** handle invalid, and will not use it with any other sqlite3_pcache_methods2 +** functions. +** +** [[the xShrink() page cache method]] +** ^SQLite invokes the xShrink() method when it wants the page cache to +** free up as much of heap memory as possible. The page cache implementation +** is not obligated to free any memory, but well-behaved implementations should +** do their best. +*/ +typedef struct sqlite3_pcache_methods2 sqlite3_pcache_methods2; +struct sqlite3_pcache_methods2 { + int iVersion; + void *pArg; + int (*xInit)(void*); + void (*xShutdown)(void*); + sqlite3_pcache *(*xCreate)(int szPage, int szExtra, int bPurgeable); + void (*xCachesize)(sqlite3_pcache*, int nCachesize); + int (*xPagecount)(sqlite3_pcache*); + sqlite3_pcache_page *(*xFetch)(sqlite3_pcache*, unsigned key, int createFlag); + void (*xUnpin)(sqlite3_pcache*, sqlite3_pcache_page*, int discard); + void (*xRekey)(sqlite3_pcache*, sqlite3_pcache_page*, + unsigned oldKey, unsigned newKey); + void (*xTruncate)(sqlite3_pcache*, unsigned iLimit); + void (*xDestroy)(sqlite3_pcache*); + void (*xShrink)(sqlite3_pcache*); +}; + +/* +** This is the obsolete pcache_methods object that has now been replaced +** by sqlite3_pcache_methods2. This object is not used by SQLite. It is +** retained in the header file for backwards compatibility only. +*/ +typedef struct sqlite3_pcache_methods sqlite3_pcache_methods; +struct sqlite3_pcache_methods { + void *pArg; + int (*xInit)(void*); + void (*xShutdown)(void*); + sqlite3_pcache *(*xCreate)(int szPage, int bPurgeable); + void (*xCachesize)(sqlite3_pcache*, int nCachesize); + int (*xPagecount)(sqlite3_pcache*); + void *(*xFetch)(sqlite3_pcache*, unsigned key, int createFlag); + void (*xUnpin)(sqlite3_pcache*, void*, int discard); + void (*xRekey)(sqlite3_pcache*, void*, unsigned oldKey, unsigned newKey); + void (*xTruncate)(sqlite3_pcache*, unsigned iLimit); + void (*xDestroy)(sqlite3_pcache*); +}; + + +/* +** CAPI3REF: Online Backup Object +** +** The sqlite3_backup object records state information about an ongoing +** online backup operation. ^The sqlite3_backup object is created by +** a call to [sqlite3_backup_init()] and is destroyed by a call to +** [sqlite3_backup_finish()]. +** +** See Also: [Using the SQLite Online Backup API] +*/ +typedef struct sqlite3_backup sqlite3_backup; + +/* +** CAPI3REF: Online Backup API. +** +** The backup API copies the content of one database into another. +** It is useful either for creating backups of databases or +** for copying in-memory databases to or from persistent files. +** +** See Also: [Using the SQLite Online Backup API] +** +** ^SQLite holds a write transaction open on the destination database file +** for the duration of the backup operation. +** ^The source database is read-locked only while it is being read; +** it is not locked continuously for the entire backup operation. +** ^Thus, the backup may be performed on a live source database without +** preventing other database connections from +** reading or writing to the source database while the backup is underway. +** +** ^(To perform a backup operation: +**
    +**
  1. sqlite3_backup_init() is called once to initialize the +** backup, +**
  2. sqlite3_backup_step() is called one or more times to transfer +** the data between the two databases, and finally +**
  3. sqlite3_backup_finish() is called to release all resources +** associated with the backup operation. +**
)^ +** There should be exactly one call to sqlite3_backup_finish() for each +** successful call to sqlite3_backup_init(). +** +** [[sqlite3_backup_init()]] sqlite3_backup_init() +** +** ^The D and N arguments to sqlite3_backup_init(D,N,S,M) are the +** [database connection] associated with the destination database +** and the database name, respectively. +** ^The database name is "main" for the main database, "temp" for the +** temporary database, or the name specified after the AS keyword in +** an [ATTACH] statement for an attached database. +** ^The S and M arguments passed to +** sqlite3_backup_init(D,N,S,M) identify the [database connection] +** and database name of the source database, respectively. +** ^The source and destination [database connections] (parameters S and D) +** must be different or else sqlite3_backup_init(D,N,S,M) will fail with +** an error. +** +** ^A call to sqlite3_backup_init() will fail, returning NULL, if +** there is already a read or read-write transaction open on the +** destination database. +** +** ^If an error occurs within sqlite3_backup_init(D,N,S,M), then NULL is +** returned and an error code and error message are stored in the +** destination [database connection] D. +** ^The error code and message for the failed call to sqlite3_backup_init() +** can be retrieved using the [sqlite3_errcode()], [sqlite3_errmsg()], and/or +** [sqlite3_errmsg16()] functions. +** ^A successful call to sqlite3_backup_init() returns a pointer to an +** [sqlite3_backup] object. +** ^The [sqlite3_backup] object may be used with the sqlite3_backup_step() and +** sqlite3_backup_finish() functions to perform the specified backup +** operation. +** +** [[sqlite3_backup_step()]] sqlite3_backup_step() +** +** ^Function sqlite3_backup_step(B,N) will copy up to N pages between +** the source and destination databases specified by [sqlite3_backup] object B. +** ^If N is negative, all remaining source pages are copied. +** ^If sqlite3_backup_step(B,N) successfully copies N pages and there +** are still more pages to be copied, then the function returns [SQLITE_OK]. +** ^If sqlite3_backup_step(B,N) successfully finishes copying all pages +** from source to destination, then it returns [SQLITE_DONE]. +** ^If an error occurs while running sqlite3_backup_step(B,N), +** then an [error code] is returned. ^As well as [SQLITE_OK] and +** [SQLITE_DONE], a call to sqlite3_backup_step() may return [SQLITE_READONLY], +** [SQLITE_NOMEM], [SQLITE_BUSY], [SQLITE_LOCKED], or an +** [SQLITE_IOERR_ACCESS | SQLITE_IOERR_XXX] extended error code. +** +** ^(The sqlite3_backup_step() might return [SQLITE_READONLY] if +**
    +**
  1. the destination database was opened read-only, or +**
  2. the destination database is using write-ahead-log journaling +** and the destination and source page sizes differ, or +**
  3. the destination database is an in-memory database and the +** destination and source page sizes differ. +**
)^ +** +** ^If sqlite3_backup_step() cannot obtain a required file-system lock, then +** the [sqlite3_busy_handler | busy-handler function] +** is invoked (if one is specified). ^If the +** busy-handler returns non-zero before the lock is available, then +** [SQLITE_BUSY] is returned to the caller. ^In this case the call to +** sqlite3_backup_step() can be retried later. ^If the source +** [database connection] +** is being used to write to the source database when sqlite3_backup_step() +** is called, then [SQLITE_LOCKED] is returned immediately. ^Again, in this +** case the call to sqlite3_backup_step() can be retried later on. ^(If +** [SQLITE_IOERR_ACCESS | SQLITE_IOERR_XXX], [SQLITE_NOMEM], or +** [SQLITE_READONLY] is returned, then +** there is no point in retrying the call to sqlite3_backup_step(). These +** errors are considered fatal.)^ The application must accept +** that the backup operation has failed and pass the backup operation handle +** to the sqlite3_backup_finish() to release associated resources. +** +** ^The first call to sqlite3_backup_step() obtains an exclusive lock +** on the destination file. ^The exclusive lock is not released until either +** sqlite3_backup_finish() is called or the backup operation is complete +** and sqlite3_backup_step() returns [SQLITE_DONE]. ^Every call to +** sqlite3_backup_step() obtains a [shared lock] on the source database that +** lasts for the duration of the sqlite3_backup_step() call. +** ^Because the source database is not locked between calls to +** sqlite3_backup_step(), the source database may be modified mid-way +** through the backup process. ^If the source database is modified by an +** external process or via a database connection other than the one being +** used by the backup operation, then the backup will be automatically +** restarted by the next call to sqlite3_backup_step(). ^If the source +** database is modified by the using the same database connection as is used +** by the backup operation, then the backup database is automatically +** updated at the same time. +** +** [[sqlite3_backup_finish()]] sqlite3_backup_finish() +** +** When sqlite3_backup_step() has returned [SQLITE_DONE], or when the +** application wishes to abandon the backup operation, the application +** should destroy the [sqlite3_backup] by passing it to sqlite3_backup_finish(). +** ^The sqlite3_backup_finish() interfaces releases all +** resources associated with the [sqlite3_backup] object. +** ^If sqlite3_backup_step() has not yet returned [SQLITE_DONE], then any +** active write-transaction on the destination database is rolled back. +** The [sqlite3_backup] object is invalid +** and may not be used following a call to sqlite3_backup_finish(). +** +** ^The value returned by sqlite3_backup_finish is [SQLITE_OK] if no +** sqlite3_backup_step() errors occurred, regardless or whether or not +** sqlite3_backup_step() completed. +** ^If an out-of-memory condition or IO error occurred during any prior +** sqlite3_backup_step() call on the same [sqlite3_backup] object, then +** sqlite3_backup_finish() returns the corresponding [error code]. +** +** ^A return of [SQLITE_BUSY] or [SQLITE_LOCKED] from sqlite3_backup_step() +** is not a permanent error and does not affect the return value of +** sqlite3_backup_finish(). +** +** [[sqlite3_backup_remaining()]] [[sqlite3_backup_pagecount()]] +** sqlite3_backup_remaining() and sqlite3_backup_pagecount() +** +** ^The sqlite3_backup_remaining() routine returns the number of pages still +** to be backed up at the conclusion of the most recent sqlite3_backup_step(). +** ^The sqlite3_backup_pagecount() routine returns the total number of pages +** in the source database at the conclusion of the most recent +** sqlite3_backup_step(). +** ^(The values returned by these functions are only updated by +** sqlite3_backup_step(). If the source database is modified in a way that +** changes the size of the source database or the number of pages remaining, +** those changes are not reflected in the output of sqlite3_backup_pagecount() +** and sqlite3_backup_remaining() until after the next +** sqlite3_backup_step().)^ +** +** Concurrent Usage of Database Handles +** +** ^The source [database connection] may be used by the application for other +** purposes while a backup operation is underway or being initialized. +** ^If SQLite is compiled and configured to support threadsafe database +** connections, then the source database connection may be used concurrently +** from within other threads. +** +** However, the application must guarantee that the destination +** [database connection] is not passed to any other API (by any thread) after +** sqlite3_backup_init() is called and before the corresponding call to +** sqlite3_backup_finish(). SQLite does not currently check to see +** if the application incorrectly accesses the destination [database connection] +** and so no error code is reported, but the operations may malfunction +** nevertheless. Use of the destination database connection while a +** backup is in progress might also also cause a mutex deadlock. +** +** If running in [shared cache mode], the application must +** guarantee that the shared cache used by the destination database +** is not accessed while the backup is running. In practice this means +** that the application must guarantee that the disk file being +** backed up to is not accessed by any connection within the process, +** not just the specific connection that was passed to sqlite3_backup_init(). +** +** The [sqlite3_backup] object itself is partially threadsafe. Multiple +** threads may safely make multiple concurrent calls to sqlite3_backup_step(). +** However, the sqlite3_backup_remaining() and sqlite3_backup_pagecount() +** APIs are not strictly speaking threadsafe. If they are invoked at the +** same time as another thread is invoking sqlite3_backup_step() it is +** possible that they return invalid values. +*/ +SQLITE_API sqlite3_backup *sqlite3_backup_init( + sqlite3 *pDest, /* Destination database handle */ + const char *zDestName, /* Destination database name */ + sqlite3 *pSource, /* Source database handle */ + const char *zSourceName /* Source database name */ +); +SQLITE_API int sqlite3_backup_step(sqlite3_backup *p, int nPage); +SQLITE_API int sqlite3_backup_finish(sqlite3_backup *p); +SQLITE_API int sqlite3_backup_remaining(sqlite3_backup *p); +SQLITE_API int sqlite3_backup_pagecount(sqlite3_backup *p); + +/* +** CAPI3REF: Unlock Notification +** METHOD: sqlite3 +** +** ^When running in shared-cache mode, a database operation may fail with +** an [SQLITE_LOCKED] error if the required locks on the shared-cache or +** individual tables within the shared-cache cannot be obtained. See +** [SQLite Shared-Cache Mode] for a description of shared-cache locking. +** ^This API may be used to register a callback that SQLite will invoke +** when the connection currently holding the required lock relinquishes it. +** ^This API is only available if the library was compiled with the +** [SQLITE_ENABLE_UNLOCK_NOTIFY] C-preprocessor symbol defined. +** +** See Also: [Using the SQLite Unlock Notification Feature]. +** +** ^Shared-cache locks are released when a database connection concludes +** its current transaction, either by committing it or rolling it back. +** +** ^When a connection (known as the blocked connection) fails to obtain a +** shared-cache lock and SQLITE_LOCKED is returned to the caller, the +** identity of the database connection (the blocking connection) that +** has locked the required resource is stored internally. ^After an +** application receives an SQLITE_LOCKED error, it may call the +** sqlite3_unlock_notify() method with the blocked connection handle as +** the first argument to register for a callback that will be invoked +** when the blocking connections current transaction is concluded. ^The +** callback is invoked from within the [sqlite3_step] or [sqlite3_close] +** call that concludes the blocking connections transaction. +** +** ^(If sqlite3_unlock_notify() is called in a multi-threaded application, +** there is a chance that the blocking connection will have already +** concluded its transaction by the time sqlite3_unlock_notify() is invoked. +** If this happens, then the specified callback is invoked immediately, +** from within the call to sqlite3_unlock_notify().)^ +** +** ^If the blocked connection is attempting to obtain a write-lock on a +** shared-cache table, and more than one other connection currently holds +** a read-lock on the same table, then SQLite arbitrarily selects one of +** the other connections to use as the blocking connection. +** +** ^(There may be at most one unlock-notify callback registered by a +** blocked connection. If sqlite3_unlock_notify() is called when the +** blocked connection already has a registered unlock-notify callback, +** then the new callback replaces the old.)^ ^If sqlite3_unlock_notify() is +** called with a NULL pointer as its second argument, then any existing +** unlock-notify callback is canceled. ^The blocked connections +** unlock-notify callback may also be canceled by closing the blocked +** connection using [sqlite3_close()]. +** +** The unlock-notify callback is not reentrant. If an application invokes +** any sqlite3_xxx API functions from within an unlock-notify callback, a +** crash or deadlock may be the result. +** +** ^Unless deadlock is detected (see below), sqlite3_unlock_notify() always +** returns SQLITE_OK. +** +** Callback Invocation Details +** +** When an unlock-notify callback is registered, the application provides a +** single void* pointer that is passed to the callback when it is invoked. +** However, the signature of the callback function allows SQLite to pass +** it an array of void* context pointers. The first argument passed to +** an unlock-notify callback is a pointer to an array of void* pointers, +** and the second is the number of entries in the array. +** +** When a blocking connections transaction is concluded, there may be +** more than one blocked connection that has registered for an unlock-notify +** callback. ^If two or more such blocked connections have specified the +** same callback function, then instead of invoking the callback function +** multiple times, it is invoked once with the set of void* context pointers +** specified by the blocked connections bundled together into an array. +** This gives the application an opportunity to prioritize any actions +** related to the set of unblocked database connections. +** +** Deadlock Detection +** +** Assuming that after registering for an unlock-notify callback a +** database waits for the callback to be issued before taking any further +** action (a reasonable assumption), then using this API may cause the +** application to deadlock. For example, if connection X is waiting for +** connection Y's transaction to be concluded, and similarly connection +** Y is waiting on connection X's transaction, then neither connection +** will proceed and the system may remain deadlocked indefinitely. +** +** To avoid this scenario, the sqlite3_unlock_notify() performs deadlock +** detection. ^If a given call to sqlite3_unlock_notify() would put the +** system in a deadlocked state, then SQLITE_LOCKED is returned and no +** unlock-notify callback is registered. The system is said to be in +** a deadlocked state if connection A has registered for an unlock-notify +** callback on the conclusion of connection B's transaction, and connection +** B has itself registered for an unlock-notify callback when connection +** A's transaction is concluded. ^Indirect deadlock is also detected, so +** the system is also considered to be deadlocked if connection B has +** registered for an unlock-notify callback on the conclusion of connection +** C's transaction, where connection C is waiting on connection A. ^Any +** number of levels of indirection are allowed. +** +** The "DROP TABLE" Exception +** +** When a call to [sqlite3_step()] returns SQLITE_LOCKED, it is almost +** always appropriate to call sqlite3_unlock_notify(). There is however, +** one exception. When executing a "DROP TABLE" or "DROP INDEX" statement, +** SQLite checks if there are any currently executing SELECT statements +** that belong to the same connection. If there are, SQLITE_LOCKED is +** returned. In this case there is no "blocking connection", so invoking +** sqlite3_unlock_notify() results in the unlock-notify callback being +** invoked immediately. If the application then re-attempts the "DROP TABLE" +** or "DROP INDEX" query, an infinite loop might be the result. +** +** One way around this problem is to check the extended error code returned +** by an sqlite3_step() call. ^(If there is a blocking connection, then the +** extended error code is set to SQLITE_LOCKED_SHAREDCACHE. Otherwise, in +** the special "DROP TABLE/INDEX" case, the extended error code is just +** SQLITE_LOCKED.)^ +*/ +SQLITE_API int sqlite3_unlock_notify( + sqlite3 *pBlocked, /* Waiting connection */ + void (*xNotify)(void **apArg, int nArg), /* Callback function to invoke */ + void *pNotifyArg /* Argument to pass to xNotify */ +); + + +/* +** CAPI3REF: String Comparison +** +** ^The [sqlite3_stricmp()] and [sqlite3_strnicmp()] APIs allow applications +** and extensions to compare the contents of two buffers containing UTF-8 +** strings in a case-independent fashion, using the same definition of "case +** independence" that SQLite uses internally when comparing identifiers. +*/ +SQLITE_API int sqlite3_stricmp(const char *, const char *); +SQLITE_API int sqlite3_strnicmp(const char *, const char *, int); + +/* +** CAPI3REF: String Globbing +* +** ^The [sqlite3_strglob(P,X)] interface returns zero if and only if +** string X matches the [GLOB] pattern P. +** ^The definition of [GLOB] pattern matching used in +** [sqlite3_strglob(P,X)] is the same as for the "X GLOB P" operator in the +** SQL dialect understood by SQLite. ^The [sqlite3_strglob(P,X)] function +** is case sensitive. +** +** Note that this routine returns zero on a match and non-zero if the strings +** do not match, the same as [sqlite3_stricmp()] and [sqlite3_strnicmp()]. +** +** See also: [sqlite3_strlike()]. +*/ +SQLITE_API int sqlite3_strglob(const char *zGlob, const char *zStr); + +/* +** CAPI3REF: String LIKE Matching +* +** ^The [sqlite3_strlike(P,X,E)] interface returns zero if and only if +** string X matches the [LIKE] pattern P with escape character E. +** ^The definition of [LIKE] pattern matching used in +** [sqlite3_strlike(P,X,E)] is the same as for the "X LIKE P ESCAPE E" +** operator in the SQL dialect understood by SQLite. ^For "X LIKE P" without +** the ESCAPE clause, set the E parameter of [sqlite3_strlike(P,X,E)] to 0. +** ^As with the LIKE operator, the [sqlite3_strlike(P,X,E)] function is case +** insensitive - equivalent upper and lower case ASCII characters match +** one another. +** +** ^The [sqlite3_strlike(P,X,E)] function matches Unicode characters, though +** only ASCII characters are case folded. +** +** Note that this routine returns zero on a match and non-zero if the strings +** do not match, the same as [sqlite3_stricmp()] and [sqlite3_strnicmp()]. +** +** See also: [sqlite3_strglob()]. +*/ +SQLITE_API int sqlite3_strlike(const char *zGlob, const char *zStr, unsigned int cEsc); + +/* +** CAPI3REF: Error Logging Interface +** +** ^The [sqlite3_log()] interface writes a message into the [error log] +** established by the [SQLITE_CONFIG_LOG] option to [sqlite3_config()]. +** ^If logging is enabled, the zFormat string and subsequent arguments are +** used with [sqlite3_snprintf()] to generate the final output string. +** +** The sqlite3_log() interface is intended for use by extensions such as +** virtual tables, collating functions, and SQL functions. While there is +** nothing to prevent an application from calling sqlite3_log(), doing so +** is considered bad form. +** +** The zFormat string must not be NULL. +** +** To avoid deadlocks and other threading problems, the sqlite3_log() routine +** will not use dynamically allocated memory. The log message is stored in +** a fixed-length buffer on the stack. If the log message is longer than +** a few hundred characters, it will be truncated to the length of the +** buffer. +*/ +SQLITE_API void sqlite3_log(int iErrCode, const char *zFormat, ...); + +/* +** CAPI3REF: Write-Ahead Log Commit Hook +** METHOD: sqlite3 +** +** ^The [sqlite3_wal_hook()] function is used to register a callback that +** is invoked each time data is committed to a database in wal mode. +** +** ^(The callback is invoked by SQLite after the commit has taken place and +** the associated write-lock on the database released)^, so the implementation +** may read, write or [checkpoint] the database as required. +** +** ^The first parameter passed to the callback function when it is invoked +** is a copy of the third parameter passed to sqlite3_wal_hook() when +** registering the callback. ^The second is a copy of the database handle. +** ^The third parameter is the name of the database that was written to - +** either "main" or the name of an [ATTACH]-ed database. ^The fourth parameter +** is the number of pages currently in the write-ahead log file, +** including those that were just committed. +** +** The callback function should normally return [SQLITE_OK]. ^If an error +** code is returned, that error will propagate back up through the +** SQLite code base to cause the statement that provoked the callback +** to report an error, though the commit will have still occurred. If the +** callback returns [SQLITE_ROW] or [SQLITE_DONE], or if it returns a value +** that does not correspond to any valid SQLite error code, the results +** are undefined. +** +** A single database handle may have at most a single write-ahead log callback +** registered at one time. ^Calling [sqlite3_wal_hook()] replaces any +** previously registered write-ahead log callback. ^Note that the +** [sqlite3_wal_autocheckpoint()] interface and the +** [wal_autocheckpoint pragma] both invoke [sqlite3_wal_hook()] and will +** overwrite any prior [sqlite3_wal_hook()] settings. +*/ +SQLITE_API void *sqlite3_wal_hook( + sqlite3*, + int(*)(void *,sqlite3*,const char*,int), + void* +); + +/* +** CAPI3REF: Configure an auto-checkpoint +** METHOD: sqlite3 +** +** ^The [sqlite3_wal_autocheckpoint(D,N)] is a wrapper around +** [sqlite3_wal_hook()] that causes any database on [database connection] D +** to automatically [checkpoint] +** after committing a transaction if there are N or +** more frames in the [write-ahead log] file. ^Passing zero or +** a negative value as the nFrame parameter disables automatic +** checkpoints entirely. +** +** ^The callback registered by this function replaces any existing callback +** registered using [sqlite3_wal_hook()]. ^Likewise, registering a callback +** using [sqlite3_wal_hook()] disables the automatic checkpoint mechanism +** configured by this function. +** +** ^The [wal_autocheckpoint pragma] can be used to invoke this interface +** from SQL. +** +** ^Checkpoints initiated by this mechanism are +** [sqlite3_wal_checkpoint_v2|PASSIVE]. +** +** ^Every new [database connection] defaults to having the auto-checkpoint +** enabled with a threshold of 1000 or [SQLITE_DEFAULT_WAL_AUTOCHECKPOINT] +** pages. The use of this interface +** is only necessary if the default setting is found to be suboptimal +** for a particular application. +*/ +SQLITE_API int sqlite3_wal_autocheckpoint(sqlite3 *db, int N); + +/* +** CAPI3REF: Checkpoint a database +** METHOD: sqlite3 +** +** ^(The sqlite3_wal_checkpoint(D,X) is equivalent to +** [sqlite3_wal_checkpoint_v2](D,X,[SQLITE_CHECKPOINT_PASSIVE],0,0).)^ +** +** In brief, sqlite3_wal_checkpoint(D,X) causes the content in the +** [write-ahead log] for database X on [database connection] D to be +** transferred into the database file and for the write-ahead log to +** be reset. See the [checkpointing] documentation for addition +** information. +** +** This interface used to be the only way to cause a checkpoint to +** occur. But then the newer and more powerful [sqlite3_wal_checkpoint_v2()] +** interface was added. This interface is retained for backwards +** compatibility and as a convenience for applications that need to manually +** start a callback but which do not need the full power (and corresponding +** complication) of [sqlite3_wal_checkpoint_v2()]. +*/ +SQLITE_API int sqlite3_wal_checkpoint(sqlite3 *db, const char *zDb); + +/* +** CAPI3REF: Checkpoint a database +** METHOD: sqlite3 +** +** ^(The sqlite3_wal_checkpoint_v2(D,X,M,L,C) interface runs a checkpoint +** operation on database X of [database connection] D in mode M. Status +** information is written back into integers pointed to by L and C.)^ +** ^(The M parameter must be a valid [checkpoint mode]:)^ +** +**
+**
SQLITE_CHECKPOINT_PASSIVE
+** ^Checkpoint as many frames as possible without waiting for any database +** readers or writers to finish, then sync the database file if all frames +** in the log were checkpointed. ^The [busy-handler callback] +** is never invoked in the SQLITE_CHECKPOINT_PASSIVE mode. +** ^On the other hand, passive mode might leave the checkpoint unfinished +** if there are concurrent readers or writers. +** +**
SQLITE_CHECKPOINT_FULL
+** ^This mode blocks (it invokes the +** [sqlite3_busy_handler|busy-handler callback]) until there is no +** database writer and all readers are reading from the most recent database +** snapshot. ^It then checkpoints all frames in the log file and syncs the +** database file. ^This mode blocks new database writers while it is pending, +** but new database readers are allowed to continue unimpeded. +** +**
SQLITE_CHECKPOINT_RESTART
+** ^This mode works the same way as SQLITE_CHECKPOINT_FULL with the addition +** that after checkpointing the log file it blocks (calls the +** [busy-handler callback]) +** until all readers are reading from the database file only. ^This ensures +** that the next writer will restart the log file from the beginning. +** ^Like SQLITE_CHECKPOINT_FULL, this mode blocks new +** database writer attempts while it is pending, but does not impede readers. +** +**
SQLITE_CHECKPOINT_TRUNCATE
+** ^This mode works the same way as SQLITE_CHECKPOINT_RESTART with the +** addition that it also truncates the log file to zero bytes just prior +** to a successful return. +**
+** +** ^If pnLog is not NULL, then *pnLog is set to the total number of frames in +** the log file or to -1 if the checkpoint could not run because +** of an error or because the database is not in [WAL mode]. ^If pnCkpt is not +** NULL,then *pnCkpt is set to the total number of checkpointed frames in the +** log file (including any that were already checkpointed before the function +** was called) or to -1 if the checkpoint could not run due to an error or +** because the database is not in WAL mode. ^Note that upon successful +** completion of an SQLITE_CHECKPOINT_TRUNCATE, the log file will have been +** truncated to zero bytes and so both *pnLog and *pnCkpt will be set to zero. +** +** ^All calls obtain an exclusive "checkpoint" lock on the database file. ^If +** any other process is running a checkpoint operation at the same time, the +** lock cannot be obtained and SQLITE_BUSY is returned. ^Even if there is a +** busy-handler configured, it will not be invoked in this case. +** +** ^The SQLITE_CHECKPOINT_FULL, RESTART and TRUNCATE modes also obtain the +** exclusive "writer" lock on the database file. ^If the writer lock cannot be +** obtained immediately, and a busy-handler is configured, it is invoked and +** the writer lock retried until either the busy-handler returns 0 or the lock +** is successfully obtained. ^The busy-handler is also invoked while waiting for +** database readers as described above. ^If the busy-handler returns 0 before +** the writer lock is obtained or while waiting for database readers, the +** checkpoint operation proceeds from that point in the same way as +** SQLITE_CHECKPOINT_PASSIVE - checkpointing as many frames as possible +** without blocking any further. ^SQLITE_BUSY is returned in this case. +** +** ^If parameter zDb is NULL or points to a zero length string, then the +** specified operation is attempted on all WAL databases [attached] to +** [database connection] db. In this case the +** values written to output parameters *pnLog and *pnCkpt are undefined. ^If +** an SQLITE_BUSY error is encountered when processing one or more of the +** attached WAL databases, the operation is still attempted on any remaining +** attached databases and SQLITE_BUSY is returned at the end. ^If any other +** error occurs while processing an attached database, processing is abandoned +** and the error code is returned to the caller immediately. ^If no error +** (SQLITE_BUSY or otherwise) is encountered while processing the attached +** databases, SQLITE_OK is returned. +** +** ^If database zDb is the name of an attached database that is not in WAL +** mode, SQLITE_OK is returned and both *pnLog and *pnCkpt set to -1. ^If +** zDb is not NULL (or a zero length string) and is not the name of any +** attached database, SQLITE_ERROR is returned to the caller. +** +** ^Unless it returns SQLITE_MISUSE, +** the sqlite3_wal_checkpoint_v2() interface +** sets the error information that is queried by +** [sqlite3_errcode()] and [sqlite3_errmsg()]. +** +** ^The [PRAGMA wal_checkpoint] command can be used to invoke this interface +** from SQL. +*/ +SQLITE_API int sqlite3_wal_checkpoint_v2( + sqlite3 *db, /* Database handle */ + const char *zDb, /* Name of attached database (or NULL) */ + int eMode, /* SQLITE_CHECKPOINT_* value */ + int *pnLog, /* OUT: Size of WAL log in frames */ + int *pnCkpt /* OUT: Total number of frames checkpointed */ +); + +/* +** CAPI3REF: Checkpoint Mode Values +** KEYWORDS: {checkpoint mode} +** +** These constants define all valid values for the "checkpoint mode" passed +** as the third parameter to the [sqlite3_wal_checkpoint_v2()] interface. +** See the [sqlite3_wal_checkpoint_v2()] documentation for details on the +** meaning of each of these checkpoint modes. +*/ +#define SQLITE_CHECKPOINT_PASSIVE 0 /* Do as much as possible w/o blocking */ +#define SQLITE_CHECKPOINT_FULL 1 /* Wait for writers, then checkpoint */ +#define SQLITE_CHECKPOINT_RESTART 2 /* Like FULL but wait for for readers */ +#define SQLITE_CHECKPOINT_TRUNCATE 3 /* Like RESTART but also truncate WAL */ + +/* +** CAPI3REF: Virtual Table Interface Configuration +** +** This function may be called by either the [xConnect] or [xCreate] method +** of a [virtual table] implementation to configure +** various facets of the virtual table interface. +** +** If this interface is invoked outside the context of an xConnect or +** xCreate virtual table method then the behavior is undefined. +** +** At present, there is only one option that may be configured using +** this function. (See [SQLITE_VTAB_CONSTRAINT_SUPPORT].) Further options +** may be added in the future. +*/ +SQLITE_API int sqlite3_vtab_config(sqlite3*, int op, ...); + +/* +** CAPI3REF: Virtual Table Configuration Options +** +** These macros define the various options to the +** [sqlite3_vtab_config()] interface that [virtual table] implementations +** can use to customize and optimize their behavior. +** +**
+** [[SQLITE_VTAB_CONSTRAINT_SUPPORT]] +**
SQLITE_VTAB_CONSTRAINT_SUPPORT +**
Calls of the form +** [sqlite3_vtab_config](db,SQLITE_VTAB_CONSTRAINT_SUPPORT,X) are supported, +** where X is an integer. If X is zero, then the [virtual table] whose +** [xCreate] or [xConnect] method invoked [sqlite3_vtab_config()] does not +** support constraints. In this configuration (which is the default) if +** a call to the [xUpdate] method returns [SQLITE_CONSTRAINT], then the entire +** statement is rolled back as if [ON CONFLICT | OR ABORT] had been +** specified as part of the users SQL statement, regardless of the actual +** ON CONFLICT mode specified. +** +** If X is non-zero, then the virtual table implementation guarantees +** that if [xUpdate] returns [SQLITE_CONSTRAINT], it will do so before +** any modifications to internal or persistent data structures have been made. +** If the [ON CONFLICT] mode is ABORT, FAIL, IGNORE or ROLLBACK, SQLite +** is able to roll back a statement or database transaction, and abandon +** or continue processing the current SQL statement as appropriate. +** If the ON CONFLICT mode is REPLACE and the [xUpdate] method returns +** [SQLITE_CONSTRAINT], SQLite handles this as if the ON CONFLICT mode +** had been ABORT. +** +** Virtual table implementations that are required to handle OR REPLACE +** must do so within the [xUpdate] method. If a call to the +** [sqlite3_vtab_on_conflict()] function indicates that the current ON +** CONFLICT policy is REPLACE, the virtual table implementation should +** silently replace the appropriate rows within the xUpdate callback and +** return SQLITE_OK. Or, if this is not possible, it may return +** SQLITE_CONSTRAINT, in which case SQLite falls back to OR ABORT +** constraint handling. +**
+*/ +#define SQLITE_VTAB_CONSTRAINT_SUPPORT 1 + +/* +** CAPI3REF: Determine The Virtual Table Conflict Policy +** +** This function may only be called from within a call to the [xUpdate] method +** of a [virtual table] implementation for an INSERT or UPDATE operation. ^The +** value returned is one of [SQLITE_ROLLBACK], [SQLITE_IGNORE], [SQLITE_FAIL], +** [SQLITE_ABORT], or [SQLITE_REPLACE], according to the [ON CONFLICT] mode +** of the SQL statement that triggered the call to the [xUpdate] method of the +** [virtual table]. +*/ +SQLITE_API int sqlite3_vtab_on_conflict(sqlite3 *); + +/* +** CAPI3REF: Determine If Virtual Table Column Access Is For UPDATE +** +** If the sqlite3_vtab_nochange(X) routine is called within the [xColumn] +** method of a [virtual table], then it returns true if and only if the +** column is being fetched as part of an UPDATE operation during which the +** column value will not change. Applications might use this to substitute +** a return value that is less expensive to compute and that the corresponding +** [xUpdate] method understands as a "no-change" value. +** +** If the [xColumn] method calls sqlite3_vtab_nochange() and finds that +** the column is not changed by the UPDATE statement, then the xColumn +** method can optionally return without setting a result, without calling +** any of the [sqlite3_result_int|sqlite3_result_xxxxx() interfaces]. +** In that case, [sqlite3_value_nochange(X)] will return true for the +** same column in the [xUpdate] method. +*/ +SQLITE_API int sqlite3_vtab_nochange(sqlite3_context*); + +/* +** CAPI3REF: Determine The Collation For a Virtual Table Constraint +** +** This function may only be called from within a call to the [xBestIndex] +** method of a [virtual table]. +** +** The first argument must be the sqlite3_index_info object that is the +** first parameter to the xBestIndex() method. The second argument must be +** an index into the aConstraint[] array belonging to the sqlite3_index_info +** structure passed to xBestIndex. This function returns a pointer to a buffer +** containing the name of the collation sequence for the corresponding +** constraint. +*/ +SQLITE_API SQLITE_EXPERIMENTAL const char *sqlite3_vtab_collation(sqlite3_index_info*,int); + +/* +** CAPI3REF: Conflict resolution modes +** KEYWORDS: {conflict resolution mode} +** +** These constants are returned by [sqlite3_vtab_on_conflict()] to +** inform a [virtual table] implementation what the [ON CONFLICT] mode +** is for the SQL statement being evaluated. +** +** Note that the [SQLITE_IGNORE] constant is also used as a potential +** return value from the [sqlite3_set_authorizer()] callback and that +** [SQLITE_ABORT] is also a [result code]. +*/ +#define SQLITE_ROLLBACK 1 +/* #define SQLITE_IGNORE 2 // Also used by sqlite3_authorizer() callback */ +#define SQLITE_FAIL 3 +/* #define SQLITE_ABORT 4 // Also an error code */ +#define SQLITE_REPLACE 5 + +/* +** CAPI3REF: Prepared Statement Scan Status Opcodes +** KEYWORDS: {scanstatus options} +** +** The following constants can be used for the T parameter to the +** [sqlite3_stmt_scanstatus(S,X,T,V)] interface. Each constant designates a +** different metric for sqlite3_stmt_scanstatus() to return. +** +** When the value returned to V is a string, space to hold that string is +** managed by the prepared statement S and will be automatically freed when +** S is finalized. +** +**
+** [[SQLITE_SCANSTAT_NLOOP]]
SQLITE_SCANSTAT_NLOOP
+**
^The [sqlite3_int64] variable pointed to by the T parameter will be +** set to the total number of times that the X-th loop has run.
+** +** [[SQLITE_SCANSTAT_NVISIT]]
SQLITE_SCANSTAT_NVISIT
+**
^The [sqlite3_int64] variable pointed to by the T parameter will be set +** to the total number of rows examined by all iterations of the X-th loop.
+** +** [[SQLITE_SCANSTAT_EST]]
SQLITE_SCANSTAT_EST
+**
^The "double" variable pointed to by the T parameter will be set to the +** query planner's estimate for the average number of rows output from each +** iteration of the X-th loop. If the query planner's estimates was accurate, +** then this value will approximate the quotient NVISIT/NLOOP and the +** product of this value for all prior loops with the same SELECTID will +** be the NLOOP value for the current loop. +** +** [[SQLITE_SCANSTAT_NAME]]
SQLITE_SCANSTAT_NAME
+**
^The "const char *" variable pointed to by the T parameter will be set +** to a zero-terminated UTF-8 string containing the name of the index or table +** used for the X-th loop. +** +** [[SQLITE_SCANSTAT_EXPLAIN]]
SQLITE_SCANSTAT_EXPLAIN
+**
^The "const char *" variable pointed to by the T parameter will be set +** to a zero-terminated UTF-8 string containing the [EXPLAIN QUERY PLAN] +** description for the X-th loop. +** +** [[SQLITE_SCANSTAT_SELECTID]]
SQLITE_SCANSTAT_SELECT
+**
^The "int" variable pointed to by the T parameter will be set to the +** "select-id" for the X-th loop. The select-id identifies which query or +** subquery the loop is part of. The main query has a select-id of zero. +** The select-id is the same value as is output in the first column +** of an [EXPLAIN QUERY PLAN] query. +**
+*/ +#define SQLITE_SCANSTAT_NLOOP 0 +#define SQLITE_SCANSTAT_NVISIT 1 +#define SQLITE_SCANSTAT_EST 2 +#define SQLITE_SCANSTAT_NAME 3 +#define SQLITE_SCANSTAT_EXPLAIN 4 +#define SQLITE_SCANSTAT_SELECTID 5 + +/* +** CAPI3REF: Prepared Statement Scan Status +** METHOD: sqlite3_stmt +** +** This interface returns information about the predicted and measured +** performance for pStmt. Advanced applications can use this +** interface to compare the predicted and the measured performance and +** issue warnings and/or rerun [ANALYZE] if discrepancies are found. +** +** Since this interface is expected to be rarely used, it is only +** available if SQLite is compiled using the [SQLITE_ENABLE_STMT_SCANSTATUS] +** compile-time option. +** +** The "iScanStatusOp" parameter determines which status information to return. +** The "iScanStatusOp" must be one of the [scanstatus options] or the behavior +** of this interface is undefined. +** ^The requested measurement is written into a variable pointed to by +** the "pOut" parameter. +** Parameter "idx" identifies the specific loop to retrieve statistics for. +** Loops are numbered starting from zero. ^If idx is out of range - less than +** zero or greater than or equal to the total number of loops used to implement +** the statement - a non-zero value is returned and the variable that pOut +** points to is unchanged. +** +** ^Statistics might not be available for all loops in all statements. ^In cases +** where there exist loops with no available statistics, this function behaves +** as if the loop did not exist - it returns non-zero and leave the variable +** that pOut points to unchanged. +** +** See also: [sqlite3_stmt_scanstatus_reset()] +*/ +SQLITE_API int sqlite3_stmt_scanstatus( + sqlite3_stmt *pStmt, /* Prepared statement for which info desired */ + int idx, /* Index of loop to report on */ + int iScanStatusOp, /* Information desired. SQLITE_SCANSTAT_* */ + void *pOut /* Result written here */ +); + +/* +** CAPI3REF: Zero Scan-Status Counters +** METHOD: sqlite3_stmt +** +** ^Zero all [sqlite3_stmt_scanstatus()] related event counters. +** +** This API is only available if the library is built with pre-processor +** symbol [SQLITE_ENABLE_STMT_SCANSTATUS] defined. +*/ +SQLITE_API void sqlite3_stmt_scanstatus_reset(sqlite3_stmt*); + +/* +** CAPI3REF: Flush caches to disk mid-transaction +** +** ^If a write-transaction is open on [database connection] D when the +** [sqlite3_db_cacheflush(D)] interface invoked, any dirty +** pages in the pager-cache that are not currently in use are written out +** to disk. A dirty page may be in use if a database cursor created by an +** active SQL statement is reading from it, or if it is page 1 of a database +** file (page 1 is always "in use"). ^The [sqlite3_db_cacheflush(D)] +** interface flushes caches for all schemas - "main", "temp", and +** any [attached] databases. +** +** ^If this function needs to obtain extra database locks before dirty pages +** can be flushed to disk, it does so. ^If those locks cannot be obtained +** immediately and there is a busy-handler callback configured, it is invoked +** in the usual manner. ^If the required lock still cannot be obtained, then +** the database is skipped and an attempt made to flush any dirty pages +** belonging to the next (if any) database. ^If any databases are skipped +** because locks cannot be obtained, but no other error occurs, this +** function returns SQLITE_BUSY. +** +** ^If any other error occurs while flushing dirty pages to disk (for +** example an IO error or out-of-memory condition), then processing is +** abandoned and an SQLite [error code] is returned to the caller immediately. +** +** ^Otherwise, if no error occurs, [sqlite3_db_cacheflush()] returns SQLITE_OK. +** +** ^This function does not set the database handle error code or message +** returned by the [sqlite3_errcode()] and [sqlite3_errmsg()] functions. +*/ +SQLITE_API int sqlite3_db_cacheflush(sqlite3*); + +/* +** CAPI3REF: The pre-update hook. +** +** ^These interfaces are only available if SQLite is compiled using the +** [SQLITE_ENABLE_PREUPDATE_HOOK] compile-time option. +** +** ^The [sqlite3_preupdate_hook()] interface registers a callback function +** that is invoked prior to each [INSERT], [UPDATE], and [DELETE] operation +** on a database table. +** ^At most one preupdate hook may be registered at a time on a single +** [database connection]; each call to [sqlite3_preupdate_hook()] overrides +** the previous setting. +** ^The preupdate hook is disabled by invoking [sqlite3_preupdate_hook()] +** with a NULL pointer as the second parameter. +** ^The third parameter to [sqlite3_preupdate_hook()] is passed through as +** the first parameter to callbacks. +** +** ^The preupdate hook only fires for changes to real database tables; the +** preupdate hook is not invoked for changes to [virtual tables] or to +** system tables like sqlite_master or sqlite_stat1. +** +** ^The second parameter to the preupdate callback is a pointer to +** the [database connection] that registered the preupdate hook. +** ^The third parameter to the preupdate callback is one of the constants +** [SQLITE_INSERT], [SQLITE_DELETE], or [SQLITE_UPDATE] to identify the +** kind of update operation that is about to occur. +** ^(The fourth parameter to the preupdate callback is the name of the +** database within the database connection that is being modified. This +** will be "main" for the main database or "temp" for TEMP tables or +** the name given after the AS keyword in the [ATTACH] statement for attached +** databases.)^ +** ^The fifth parameter to the preupdate callback is the name of the +** table that is being modified. +** +** For an UPDATE or DELETE operation on a [rowid table], the sixth +** parameter passed to the preupdate callback is the initial [rowid] of the +** row being modified or deleted. For an INSERT operation on a rowid table, +** or any operation on a WITHOUT ROWID table, the value of the sixth +** parameter is undefined. For an INSERT or UPDATE on a rowid table the +** seventh parameter is the final rowid value of the row being inserted +** or updated. The value of the seventh parameter passed to the callback +** function is not defined for operations on WITHOUT ROWID tables, or for +** INSERT operations on rowid tables. +** +** The [sqlite3_preupdate_old()], [sqlite3_preupdate_new()], +** [sqlite3_preupdate_count()], and [sqlite3_preupdate_depth()] interfaces +** provide additional information about a preupdate event. These routines +** may only be called from within a preupdate callback. Invoking any of +** these routines from outside of a preupdate callback or with a +** [database connection] pointer that is different from the one supplied +** to the preupdate callback results in undefined and probably undesirable +** behavior. +** +** ^The [sqlite3_preupdate_count(D)] interface returns the number of columns +** in the row that is being inserted, updated, or deleted. +** +** ^The [sqlite3_preupdate_old(D,N,P)] interface writes into P a pointer to +** a [protected sqlite3_value] that contains the value of the Nth column of +** the table row before it is updated. The N parameter must be between 0 +** and one less than the number of columns or the behavior will be +** undefined. This must only be used within SQLITE_UPDATE and SQLITE_DELETE +** preupdate callbacks; if it is used by an SQLITE_INSERT callback then the +** behavior is undefined. The [sqlite3_value] that P points to +** will be destroyed when the preupdate callback returns. +** +** ^The [sqlite3_preupdate_new(D,N,P)] interface writes into P a pointer to +** a [protected sqlite3_value] that contains the value of the Nth column of +** the table row after it is updated. The N parameter must be between 0 +** and one less than the number of columns or the behavior will be +** undefined. This must only be used within SQLITE_INSERT and SQLITE_UPDATE +** preupdate callbacks; if it is used by an SQLITE_DELETE callback then the +** behavior is undefined. The [sqlite3_value] that P points to +** will be destroyed when the preupdate callback returns. +** +** ^The [sqlite3_preupdate_depth(D)] interface returns 0 if the preupdate +** callback was invoked as a result of a direct insert, update, or delete +** operation; or 1 for inserts, updates, or deletes invoked by top-level +** triggers; or 2 for changes resulting from triggers called by top-level +** triggers; and so forth. +** +** See also: [sqlite3_update_hook()] +*/ +#if defined(SQLITE_ENABLE_PREUPDATE_HOOK) +SQLITE_API void *sqlite3_preupdate_hook( + sqlite3 *db, + void(*xPreUpdate)( + void *pCtx, /* Copy of third arg to preupdate_hook() */ + sqlite3 *db, /* Database handle */ + int op, /* SQLITE_UPDATE, DELETE or INSERT */ + char const *zDb, /* Database name */ + char const *zName, /* Table name */ + sqlite3_int64 iKey1, /* Rowid of row about to be deleted/updated */ + sqlite3_int64 iKey2 /* New rowid value (for a rowid UPDATE) */ + ), + void* +); +SQLITE_API int sqlite3_preupdate_old(sqlite3 *, int, sqlite3_value **); +SQLITE_API int sqlite3_preupdate_count(sqlite3 *); +SQLITE_API int sqlite3_preupdate_depth(sqlite3 *); +SQLITE_API int sqlite3_preupdate_new(sqlite3 *, int, sqlite3_value **); +#endif + +/* +** CAPI3REF: Low-level system error code +** +** ^Attempt to return the underlying operating system error code or error +** number that caused the most recent I/O error or failure to open a file. +** The return value is OS-dependent. For example, on unix systems, after +** [sqlite3_open_v2()] returns [SQLITE_CANTOPEN], this interface could be +** called to get back the underlying "errno" that caused the problem, such +** as ENOSPC, EAUTH, EISDIR, and so forth. +*/ +SQLITE_API int sqlite3_system_errno(sqlite3*); + +/* +** CAPI3REF: Database Snapshot +** KEYWORDS: {snapshot} {sqlite3_snapshot} +** +** An instance of the snapshot object records the state of a [WAL mode] +** database for some specific point in history. +** +** In [WAL mode], multiple [database connections] that are open on the +** same database file can each be reading a different historical version +** of the database file. When a [database connection] begins a read +** transaction, that connection sees an unchanging copy of the database +** as it existed for the point in time when the transaction first started. +** Subsequent changes to the database from other connections are not seen +** by the reader until a new read transaction is started. +** +** The sqlite3_snapshot object records state information about an historical +** version of the database file so that it is possible to later open a new read +** transaction that sees that historical version of the database rather than +** the most recent version. +*/ +typedef struct sqlite3_snapshot { + unsigned char hidden[48]; +} sqlite3_snapshot; + +/* +** CAPI3REF: Record A Database Snapshot +** CONSTRUCTOR: sqlite3_snapshot +** +** ^The [sqlite3_snapshot_get(D,S,P)] interface attempts to make a +** new [sqlite3_snapshot] object that records the current state of +** schema S in database connection D. ^On success, the +** [sqlite3_snapshot_get(D,S,P)] interface writes a pointer to the newly +** created [sqlite3_snapshot] object into *P and returns SQLITE_OK. +** If there is not already a read-transaction open on schema S when +** this function is called, one is opened automatically. +** +** The following must be true for this function to succeed. If any of +** the following statements are false when sqlite3_snapshot_get() is +** called, SQLITE_ERROR is returned. The final value of *P is undefined +** in this case. +** +**
    +**
  • The database handle must not be in [autocommit mode]. +** +**
  • Schema S of [database connection] D must be a [WAL mode] database. +** +**
  • There must not be a write transaction open on schema S of database +** connection D. +** +**
  • One or more transactions must have been written to the current wal +** file since it was created on disk (by any connection). This means +** that a snapshot cannot be taken on a wal mode database with no wal +** file immediately after it is first opened. At least one transaction +** must be written to it first. +**
+** +** This function may also return SQLITE_NOMEM. If it is called with the +** database handle in autocommit mode but fails for some other reason, +** whether or not a read transaction is opened on schema S is undefined. +** +** The [sqlite3_snapshot] object returned from a successful call to +** [sqlite3_snapshot_get()] must be freed using [sqlite3_snapshot_free()] +** to avoid a memory leak. +** +** The [sqlite3_snapshot_get()] interface is only available when the +** [SQLITE_ENABLE_SNAPSHOT] compile-time option is used. +*/ +SQLITE_API SQLITE_EXPERIMENTAL int sqlite3_snapshot_get( + sqlite3 *db, + const char *zSchema, + sqlite3_snapshot **ppSnapshot +); + +/* +** CAPI3REF: Start a read transaction on an historical snapshot +** METHOD: sqlite3_snapshot +** +** ^The [sqlite3_snapshot_open(D,S,P)] interface either starts a new read +** transaction or upgrades an existing one for schema S of +** [database connection] D such that the read transaction refers to +** historical [snapshot] P, rather than the most recent change to the +** database. ^The [sqlite3_snapshot_open()] interface returns SQLITE_OK +** on success or an appropriate [error code] if it fails. +** +** ^In order to succeed, the database connection must not be in +** [autocommit mode] when [sqlite3_snapshot_open(D,S,P)] is called. If there +** is already a read transaction open on schema S, then the database handle +** must have no active statements (SELECT statements that have been passed +** to sqlite3_step() but not sqlite3_reset() or sqlite3_finalize()). +** SQLITE_ERROR is returned if either of these conditions is violated, or +** if schema S does not exist, or if the snapshot object is invalid. +** +** ^A call to sqlite3_snapshot_open() will fail to open if the specified +** snapshot has been overwritten by a [checkpoint]. In this case +** SQLITE_ERROR_SNAPSHOT is returned. +** +** If there is already a read transaction open when this function is +** invoked, then the same read transaction remains open (on the same +** database snapshot) if SQLITE_ERROR, SQLITE_BUSY or SQLITE_ERROR_SNAPSHOT +** is returned. If another error code - for example SQLITE_PROTOCOL or an +** SQLITE_IOERR error code - is returned, then the final state of the +** read transaction is undefined. If SQLITE_OK is returned, then the +** read transaction is now open on database snapshot P. +** +** ^(A call to [sqlite3_snapshot_open(D,S,P)] will fail if the +** database connection D does not know that the database file for +** schema S is in [WAL mode]. A database connection might not know +** that the database file is in [WAL mode] if there has been no prior +** I/O on that database connection, or if the database entered [WAL mode] +** after the most recent I/O on the database connection.)^ +** (Hint: Run "[PRAGMA application_id]" against a newly opened +** database connection in order to make it ready to use snapshots.) +** +** The [sqlite3_snapshot_open()] interface is only available when the +** [SQLITE_ENABLE_SNAPSHOT] compile-time option is used. +*/ +SQLITE_API SQLITE_EXPERIMENTAL int sqlite3_snapshot_open( + sqlite3 *db, + const char *zSchema, + sqlite3_snapshot *pSnapshot +); + +/* +** CAPI3REF: Destroy a snapshot +** DESTRUCTOR: sqlite3_snapshot +** +** ^The [sqlite3_snapshot_free(P)] interface destroys [sqlite3_snapshot] P. +** The application must eventually free every [sqlite3_snapshot] object +** using this routine to avoid a memory leak. +** +** The [sqlite3_snapshot_free()] interface is only available when the +** [SQLITE_ENABLE_SNAPSHOT] compile-time option is used. +*/ +SQLITE_API SQLITE_EXPERIMENTAL void sqlite3_snapshot_free(sqlite3_snapshot*); + +/* +** CAPI3REF: Compare the ages of two snapshot handles. +** METHOD: sqlite3_snapshot +** +** The sqlite3_snapshot_cmp(P1, P2) interface is used to compare the ages +** of two valid snapshot handles. +** +** If the two snapshot handles are not associated with the same database +** file, the result of the comparison is undefined. +** +** Additionally, the result of the comparison is only valid if both of the +** snapshot handles were obtained by calling sqlite3_snapshot_get() since the +** last time the wal file was deleted. The wal file is deleted when the +** database is changed back to rollback mode or when the number of database +** clients drops to zero. If either snapshot handle was obtained before the +** wal file was last deleted, the value returned by this function +** is undefined. +** +** Otherwise, this API returns a negative value if P1 refers to an older +** snapshot than P2, zero if the two handles refer to the same database +** snapshot, and a positive value if P1 is a newer snapshot than P2. +** +** This interface is only available if SQLite is compiled with the +** [SQLITE_ENABLE_SNAPSHOT] option. +*/ +SQLITE_API SQLITE_EXPERIMENTAL int sqlite3_snapshot_cmp( + sqlite3_snapshot *p1, + sqlite3_snapshot *p2 +); + +/* +** CAPI3REF: Recover snapshots from a wal file +** METHOD: sqlite3_snapshot +** +** If a [WAL file] remains on disk after all database connections close +** (either through the use of the [SQLITE_FCNTL_PERSIST_WAL] [file control] +** or because the last process to have the database opened exited without +** calling [sqlite3_close()]) and a new connection is subsequently opened +** on that database and [WAL file], the [sqlite3_snapshot_open()] interface +** will only be able to open the last transaction added to the WAL file +** even though the WAL file contains other valid transactions. +** +** This function attempts to scan the WAL file associated with database zDb +** of database handle db and make all valid snapshots available to +** sqlite3_snapshot_open(). It is an error if there is already a read +** transaction open on the database, or if the database is not a WAL mode +** database. +** +** SQLITE_OK is returned if successful, or an SQLite error code otherwise. +** +** This interface is only available if SQLite is compiled with the +** [SQLITE_ENABLE_SNAPSHOT] option. +*/ +SQLITE_API SQLITE_EXPERIMENTAL int sqlite3_snapshot_recover(sqlite3 *db, const char *zDb); + +/* +** CAPI3REF: Serialize a database +** +** The sqlite3_serialize(D,S,P,F) interface returns a pointer to memory +** that is a serialization of the S database on [database connection] D. +** If P is not a NULL pointer, then the size of the database in bytes +** is written into *P. +** +** For an ordinary on-disk database file, the serialization is just a +** copy of the disk file. For an in-memory database or a "TEMP" database, +** the serialization is the same sequence of bytes which would be written +** to disk if that database where backed up to disk. +** +** The usual case is that sqlite3_serialize() copies the serialization of +** the database into memory obtained from [sqlite3_malloc64()] and returns +** a pointer to that memory. The caller is responsible for freeing the +** returned value to avoid a memory leak. However, if the F argument +** contains the SQLITE_SERIALIZE_NOCOPY bit, then no memory allocations +** are made, and the sqlite3_serialize() function will return a pointer +** to the contiguous memory representation of the database that SQLite +** is currently using for that database, or NULL if the no such contiguous +** memory representation of the database exists. A contiguous memory +** representation of the database will usually only exist if there has +** been a prior call to [sqlite3_deserialize(D,S,...)] with the same +** values of D and S. +** The size of the database is written into *P even if the +** SQLITE_SERIALIZE_NOCOPY bit is set but no contiguous copy +** of the database exists. +** +** A call to sqlite3_serialize(D,S,P,F) might return NULL even if the +** SQLITE_SERIALIZE_NOCOPY bit is omitted from argument F if a memory +** allocation error occurs. +** +** This interface is only available if SQLite is compiled with the +** [SQLITE_ENABLE_DESERIALIZE] option. +*/ +SQLITE_API unsigned char *sqlite3_serialize( + sqlite3 *db, /* The database connection */ + const char *zSchema, /* Which DB to serialize. ex: "main", "temp", ... */ + sqlite3_int64 *piSize, /* Write size of the DB here, if not NULL */ + unsigned int mFlags /* Zero or more SQLITE_SERIALIZE_* flags */ +); + +/* +** CAPI3REF: Flags for sqlite3_serialize +** +** Zero or more of the following constants can be OR-ed together for +** the F argument to [sqlite3_serialize(D,S,P,F)]. +** +** SQLITE_SERIALIZE_NOCOPY means that [sqlite3_serialize()] will return +** a pointer to contiguous in-memory database that it is currently using, +** without making a copy of the database. If SQLite is not currently using +** a contiguous in-memory database, then this option causes +** [sqlite3_serialize()] to return a NULL pointer. SQLite will only be +** using a contiguous in-memory database if it has been initialized by a +** prior call to [sqlite3_deserialize()]. +*/ +#define SQLITE_SERIALIZE_NOCOPY 0x001 /* Do no memory allocations */ + +/* +** CAPI3REF: Deserialize a database +** +** The sqlite3_deserialize(D,S,P,N,M,F) interface causes the +** [database connection] D to disconnect from database S and then +** reopen S as an in-memory database based on the serialization contained +** in P. The serialized database P is N bytes in size. M is the size of +** the buffer P, which might be larger than N. If M is larger than N, and +** the SQLITE_DESERIALIZE_READONLY bit is not set in F, then SQLite is +** permitted to add content to the in-memory database as long as the total +** size does not exceed M bytes. +** +** If the SQLITE_DESERIALIZE_FREEONCLOSE bit is set in F, then SQLite will +** invoke sqlite3_free() on the serialization buffer when the database +** connection closes. If the SQLITE_DESERIALIZE_RESIZEABLE bit is set, then +** SQLite will try to increase the buffer size using sqlite3_realloc64() +** if writes on the database cause it to grow larger than M bytes. +** +** The sqlite3_deserialize() interface will fail with SQLITE_BUSY if the +** database is currently in a read transaction or is involved in a backup +** operation. +** +** If sqlite3_deserialize(D,S,P,N,M,F) fails for any reason and if the +** SQLITE_DESERIALIZE_FREEONCLOSE bit is set in argument F, then +** [sqlite3_free()] is invoked on argument P prior to returning. +** +** This interface is only available if SQLite is compiled with the +** [SQLITE_ENABLE_DESERIALIZE] option. +*/ +SQLITE_API int sqlite3_deserialize( + sqlite3 *db, /* The database connection */ + const char *zSchema, /* Which DB to reopen with the deserialization */ + unsigned char *pData, /* The serialized database content */ + sqlite3_int64 szDb, /* Number bytes in the deserialization */ + sqlite3_int64 szBuf, /* Total size of buffer pData[] */ + unsigned mFlags /* Zero or more SQLITE_DESERIALIZE_* flags */ +); + +/* +** CAPI3REF: Flags for sqlite3_deserialize() +** +** The following are allowed values for 6th argument (the F argument) to +** the [sqlite3_deserialize(D,S,P,N,M,F)] interface. +** +** The SQLITE_DESERIALIZE_FREEONCLOSE means that the database serialization +** in the P argument is held in memory obtained from [sqlite3_malloc64()] +** and that SQLite should take ownership of this memory and automatically +** free it when it has finished using it. Without this flag, the caller +** is responsible for freeing any dynamically allocated memory. +** +** The SQLITE_DESERIALIZE_RESIZEABLE flag means that SQLite is allowed to +** grow the size of the database using calls to [sqlite3_realloc64()]. This +** flag should only be used if SQLITE_DESERIALIZE_FREEONCLOSE is also used. +** Without this flag, the deserialized database cannot increase in size beyond +** the number of bytes specified by the M parameter. +** +** The SQLITE_DESERIALIZE_READONLY flag means that the deserialized database +** should be treated as read-only. +*/ +#define SQLITE_DESERIALIZE_FREEONCLOSE 1 /* Call sqlite3_free() on close */ +#define SQLITE_DESERIALIZE_RESIZEABLE 2 /* Resize using sqlite3_realloc64() */ +#define SQLITE_DESERIALIZE_READONLY 4 /* Database is read-only */ + +/* +** Undo the hack that converts floating point types to integer for +** builds on processors without floating point support. +*/ +#ifdef SQLITE_OMIT_FLOATING_POINT +# undef double +#endif + +#if 0 +} /* End of the 'extern "C"' block */ +#endif +#endif /* SQLITE3_H */ + +/******** Begin file sqlite3rtree.h *********/ +/* +** 2010 August 30 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +*/ + +#ifndef _SQLITE3RTREE_H_ +#define _SQLITE3RTREE_H_ + + +#if 0 +extern "C" { +#endif + +typedef struct sqlite3_rtree_geometry sqlite3_rtree_geometry; +typedef struct sqlite3_rtree_query_info sqlite3_rtree_query_info; + +/* The double-precision datatype used by RTree depends on the +** SQLITE_RTREE_INT_ONLY compile-time option. +*/ +#ifdef SQLITE_RTREE_INT_ONLY + typedef sqlite3_int64 sqlite3_rtree_dbl; +#else + typedef double sqlite3_rtree_dbl; +#endif + +/* +** Register a geometry callback named zGeom that can be used as part of an +** R-Tree geometry query as follows: +** +** SELECT ... FROM WHERE MATCH $zGeom(... params ...) +*/ +SQLITE_API int sqlite3_rtree_geometry_callback( + sqlite3 *db, + const char *zGeom, + int (*xGeom)(sqlite3_rtree_geometry*, int, sqlite3_rtree_dbl*,int*), + void *pContext +); + + +/* +** A pointer to a structure of the following type is passed as the first +** argument to callbacks registered using rtree_geometry_callback(). +*/ +struct sqlite3_rtree_geometry { + void *pContext; /* Copy of pContext passed to s_r_g_c() */ + int nParam; /* Size of array aParam[] */ + sqlite3_rtree_dbl *aParam; /* Parameters passed to SQL geom function */ + void *pUser; /* Callback implementation user data */ + void (*xDelUser)(void *); /* Called by SQLite to clean up pUser */ +}; + +/* +** Register a 2nd-generation geometry callback named zScore that can be +** used as part of an R-Tree geometry query as follows: +** +** SELECT ... FROM WHERE MATCH $zQueryFunc(... params ...) +*/ +SQLITE_API int sqlite3_rtree_query_callback( + sqlite3 *db, + const char *zQueryFunc, + int (*xQueryFunc)(sqlite3_rtree_query_info*), + void *pContext, + void (*xDestructor)(void*) +); + + +/* +** A pointer to a structure of the following type is passed as the +** argument to scored geometry callback registered using +** sqlite3_rtree_query_callback(). +** +** Note that the first 5 fields of this structure are identical to +** sqlite3_rtree_geometry. This structure is a subclass of +** sqlite3_rtree_geometry. +*/ +struct sqlite3_rtree_query_info { + void *pContext; /* pContext from when function registered */ + int nParam; /* Number of function parameters */ + sqlite3_rtree_dbl *aParam; /* value of function parameters */ + void *pUser; /* callback can use this, if desired */ + void (*xDelUser)(void*); /* function to free pUser */ + sqlite3_rtree_dbl *aCoord; /* Coordinates of node or entry to check */ + unsigned int *anQueue; /* Number of pending entries in the queue */ + int nCoord; /* Number of coordinates */ + int iLevel; /* Level of current node or entry */ + int mxLevel; /* The largest iLevel value in the tree */ + sqlite3_int64 iRowid; /* Rowid for current entry */ + sqlite3_rtree_dbl rParentScore; /* Score of parent node */ + int eParentWithin; /* Visibility of parent node */ + int eWithin; /* OUT: Visibility */ + sqlite3_rtree_dbl rScore; /* OUT: Write the score here */ + /* The following fields are only available in 3.8.11 and later */ + sqlite3_value **apSqlParam; /* Original SQL values of parameters */ +}; + +/* +** Allowed values for sqlite3_rtree_query.eWithin and .eParentWithin. +*/ +#define NOT_WITHIN 0 /* Object completely outside of query region */ +#define PARTLY_WITHIN 1 /* Object partially overlaps query region */ +#define FULLY_WITHIN 2 /* Object fully contained within query region */ + + +#if 0 +} /* end of the 'extern "C"' block */ +#endif + +#endif /* ifndef _SQLITE3RTREE_H_ */ + +/******** End of sqlite3rtree.h *********/ +/******** Begin file sqlite3session.h *********/ + +#if !defined(__SQLITESESSION_H_) && defined(SQLITE_ENABLE_SESSION) +#define __SQLITESESSION_H_ 1 + +/* +** Make sure we can call this stuff from C++. +*/ +#if 0 +extern "C" { +#endif + + +/* +** CAPI3REF: Session Object Handle +** +** An instance of this object is a [session] that can be used to +** record changes to a database. +*/ +typedef struct sqlite3_session sqlite3_session; + +/* +** CAPI3REF: Changeset Iterator Handle +** +** An instance of this object acts as a cursor for iterating +** over the elements of a [changeset] or [patchset]. +*/ +typedef struct sqlite3_changeset_iter sqlite3_changeset_iter; + +/* +** CAPI3REF: Create A New Session Object +** CONSTRUCTOR: sqlite3_session +** +** Create a new session object attached to database handle db. If successful, +** a pointer to the new object is written to *ppSession and SQLITE_OK is +** returned. If an error occurs, *ppSession is set to NULL and an SQLite +** error code (e.g. SQLITE_NOMEM) is returned. +** +** It is possible to create multiple session objects attached to a single +** database handle. +** +** Session objects created using this function should be deleted using the +** [sqlite3session_delete()] function before the database handle that they +** are attached to is itself closed. If the database handle is closed before +** the session object is deleted, then the results of calling any session +** module function, including [sqlite3session_delete()] on the session object +** are undefined. +** +** Because the session module uses the [sqlite3_preupdate_hook()] API, it +** is not possible for an application to register a pre-update hook on a +** database handle that has one or more session objects attached. Nor is +** it possible to create a session object attached to a database handle for +** which a pre-update hook is already defined. The results of attempting +** either of these things are undefined. +** +** The session object will be used to create changesets for tables in +** database zDb, where zDb is either "main", or "temp", or the name of an +** attached database. It is not an error if database zDb is not attached +** to the database when the session object is created. +*/ +SQLITE_API int sqlite3session_create( + sqlite3 *db, /* Database handle */ + const char *zDb, /* Name of db (e.g. "main") */ + sqlite3_session **ppSession /* OUT: New session object */ +); + +/* +** CAPI3REF: Delete A Session Object +** DESTRUCTOR: sqlite3_session +** +** Delete a session object previously allocated using +** [sqlite3session_create()]. Once a session object has been deleted, the +** results of attempting to use pSession with any other session module +** function are undefined. +** +** Session objects must be deleted before the database handle to which they +** are attached is closed. Refer to the documentation for +** [sqlite3session_create()] for details. +*/ +SQLITE_API void sqlite3session_delete(sqlite3_session *pSession); + + +/* +** CAPI3REF: Enable Or Disable A Session Object +** METHOD: sqlite3_session +** +** Enable or disable the recording of changes by a session object. When +** enabled, a session object records changes made to the database. When +** disabled - it does not. A newly created session object is enabled. +** Refer to the documentation for [sqlite3session_changeset()] for further +** details regarding how enabling and disabling a session object affects +** the eventual changesets. +** +** Passing zero to this function disables the session. Passing a value +** greater than zero enables it. Passing a value less than zero is a +** no-op, and may be used to query the current state of the session. +** +** The return value indicates the final state of the session object: 0 if +** the session is disabled, or 1 if it is enabled. +*/ +SQLITE_API int sqlite3session_enable(sqlite3_session *pSession, int bEnable); + +/* +** CAPI3REF: Set Or Clear the Indirect Change Flag +** METHOD: sqlite3_session +** +** Each change recorded by a session object is marked as either direct or +** indirect. A change is marked as indirect if either: +** +**
    +**
  • The session object "indirect" flag is set when the change is +** made, or +**
  • The change is made by an SQL trigger or foreign key action +** instead of directly as a result of a users SQL statement. +**
+** +** If a single row is affected by more than one operation within a session, +** then the change is considered indirect if all operations meet the criteria +** for an indirect change above, or direct otherwise. +** +** This function is used to set, clear or query the session object indirect +** flag. If the second argument passed to this function is zero, then the +** indirect flag is cleared. If it is greater than zero, the indirect flag +** is set. Passing a value less than zero does not modify the current value +** of the indirect flag, and may be used to query the current state of the +** indirect flag for the specified session object. +** +** The return value indicates the final state of the indirect flag: 0 if +** it is clear, or 1 if it is set. +*/ +SQLITE_API int sqlite3session_indirect(sqlite3_session *pSession, int bIndirect); + +/* +** CAPI3REF: Attach A Table To A Session Object +** METHOD: sqlite3_session +** +** If argument zTab is not NULL, then it is the name of a table to attach +** to the session object passed as the first argument. All subsequent changes +** made to the table while the session object is enabled will be recorded. See +** documentation for [sqlite3session_changeset()] for further details. +** +** Or, if argument zTab is NULL, then changes are recorded for all tables +** in the database. If additional tables are added to the database (by +** executing "CREATE TABLE" statements) after this call is made, changes for +** the new tables are also recorded. +** +** Changes can only be recorded for tables that have a PRIMARY KEY explicitly +** defined as part of their CREATE TABLE statement. It does not matter if the +** PRIMARY KEY is an "INTEGER PRIMARY KEY" (rowid alias) or not. The PRIMARY +** KEY may consist of a single column, or may be a composite key. +** +** It is not an error if the named table does not exist in the database. Nor +** is it an error if the named table does not have a PRIMARY KEY. However, +** no changes will be recorded in either of these scenarios. +** +** Changes are not recorded for individual rows that have NULL values stored +** in one or more of their PRIMARY KEY columns. +** +** SQLITE_OK is returned if the call completes without error. Or, if an error +** occurs, an SQLite error code (e.g. SQLITE_NOMEM) is returned. +** +**

Special sqlite_stat1 Handling

+** +** As of SQLite version 3.22.0, the "sqlite_stat1" table is an exception to +** some of the rules above. In SQLite, the schema of sqlite_stat1 is: +** 
+**        CREATE TABLE sqlite_stat1(tbl,idx,stat)  
+**
+** +** Even though sqlite_stat1 does not have a PRIMARY KEY, changes are +** recorded for it as if the PRIMARY KEY is (tbl,idx). Additionally, changes +** are recorded for rows for which (idx IS NULL) is true. However, for such +** rows a zero-length blob (SQL value X'') is stored in the changeset or +** patchset instead of a NULL value. This allows such changesets to be +** manipulated by legacy implementations of sqlite3changeset_invert(), +** concat() and similar. +** +** The sqlite3changeset_apply() function automatically converts the +** zero-length blob back to a NULL value when updating the sqlite_stat1 +** table. However, if the application calls sqlite3changeset_new(), +** sqlite3changeset_old() or sqlite3changeset_conflict on a changeset +** iterator directly (including on a changeset iterator passed to a +** conflict-handler callback) then the X'' value is returned. The application +** must translate X'' to NULL itself if required. +** +** Legacy (older than 3.22.0) versions of the sessions module cannot capture +** changes made to the sqlite_stat1 table. Legacy versions of the +** sqlite3changeset_apply() function silently ignore any modifications to the +** sqlite_stat1 table that are part of a changeset or patchset. +*/ +SQLITE_API int sqlite3session_attach( + sqlite3_session *pSession, /* Session object */ + const char *zTab /* Table name */ +); + +/* +** CAPI3REF: Set a table filter on a Session Object. +** METHOD: sqlite3_session +** +** The second argument (xFilter) is the "filter callback". For changes to rows +** in tables that are not attached to the Session object, the filter is called +** to determine whether changes to the table's rows should be tracked or not. +** If xFilter returns 0, changes is not tracked. Note that once a table is +** attached, xFilter will not be called again. +*/ +SQLITE_API void sqlite3session_table_filter( + sqlite3_session *pSession, /* Session object */ + int(*xFilter)( + void *pCtx, /* Copy of third arg to _filter_table() */ + const char *zTab /* Table name */ + ), + void *pCtx /* First argument passed to xFilter */ +); + +/* +** CAPI3REF: Generate A Changeset From A Session Object +** METHOD: sqlite3_session +** +** Obtain a changeset containing changes to the tables attached to the +** session object passed as the first argument. If successful, +** set *ppChangeset to point to a buffer containing the changeset +** and *pnChangeset to the size of the changeset in bytes before returning +** SQLITE_OK. If an error occurs, set both *ppChangeset and *pnChangeset to +** zero and return an SQLite error code. +** +** A changeset consists of zero or more INSERT, UPDATE and/or DELETE changes, +** each representing a change to a single row of an attached table. An INSERT +** change contains the values of each field of a new database row. A DELETE +** contains the original values of each field of a deleted database row. An +** UPDATE change contains the original values of each field of an updated +** database row along with the updated values for each updated non-primary-key +** column. It is not possible for an UPDATE change to represent a change that +** modifies the values of primary key columns. If such a change is made, it +** is represented in a changeset as a DELETE followed by an INSERT. +** +** Changes are not recorded for rows that have NULL values stored in one or +** more of their PRIMARY KEY columns. If such a row is inserted or deleted, +** no corresponding change is present in the changesets returned by this +** function. If an existing row with one or more NULL values stored in +** PRIMARY KEY columns is updated so that all PRIMARY KEY columns are non-NULL, +** only an INSERT is appears in the changeset. Similarly, if an existing row +** with non-NULL PRIMARY KEY values is updated so that one or more of its +** PRIMARY KEY columns are set to NULL, the resulting changeset contains a +** DELETE change only. +** +** The contents of a changeset may be traversed using an iterator created +** using the [sqlite3changeset_start()] API. A changeset may be applied to +** a database with a compatible schema using the [sqlite3changeset_apply()] +** API. +** +** Within a changeset generated by this function, all changes related to a +** single table are grouped together. In other words, when iterating through +** a changeset or when applying a changeset to a database, all changes related +** to a single table are processed before moving on to the next table. Tables +** are sorted in the same order in which they were attached (or auto-attached) +** to the sqlite3_session object. The order in which the changes related to +** a single table are stored is undefined. +** +** Following a successful call to this function, it is the responsibility of +** the caller to eventually free the buffer that *ppChangeset points to using +** [sqlite3_free()]. +** +**

Changeset Generation

+** +** Once a table has been attached to a session object, the session object +** records the primary key values of all new rows inserted into the table. +** It also records the original primary key and other column values of any +** deleted or updated rows. For each unique primary key value, data is only +** recorded once - the first time a row with said primary key is inserted, +** updated or deleted in the lifetime of the session. +** +** There is one exception to the previous paragraph: when a row is inserted, +** updated or deleted, if one or more of its primary key columns contain a +** NULL value, no record of the change is made. +** +** The session object therefore accumulates two types of records - those +** that consist of primary key values only (created when the user inserts +** a new record) and those that consist of the primary key values and the +** original values of other table columns (created when the users deletes +** or updates a record). +** +** When this function is called, the requested changeset is created using +** both the accumulated records and the current contents of the database +** file. Specifically: +** +**
    +**
  • For each record generated by an insert, the database is queried +** for a row with a matching primary key. If one is found, an INSERT +** change is added to the changeset. If no such row is found, no change +** is added to the changeset. +** +**
  • For each record generated by an update or delete, the database is +** queried for a row with a matching primary key. If such a row is +** found and one or more of the non-primary key fields have been +** modified from their original values, an UPDATE change is added to +** the changeset. Or, if no such row is found in the table, a DELETE +** change is added to the changeset. If there is a row with a matching +** primary key in the database, but all fields contain their original +** values, no change is added to the changeset. +**
+** +** This means, amongst other things, that if a row is inserted and then later +** deleted while a session object is active, neither the insert nor the delete +** will be present in the changeset. Or if a row is deleted and then later a +** row with the same primary key values inserted while a session object is +** active, the resulting changeset will contain an UPDATE change instead of +** a DELETE and an INSERT. +** +** When a session object is disabled (see the [sqlite3session_enable()] API), +** it does not accumulate records when rows are inserted, updated or deleted. +** This may appear to have some counter-intuitive effects if a single row +** is written to more than once during a session. For example, if a row +** is inserted while a session object is enabled, then later deleted while +** the same session object is disabled, no INSERT record will appear in the +** changeset, even though the delete took place while the session was disabled. +** Or, if one field of a row is updated while a session is disabled, and +** another field of the same row is updated while the session is enabled, the +** resulting changeset will contain an UPDATE change that updates both fields. +*/ +SQLITE_API int sqlite3session_changeset( + sqlite3_session *pSession, /* Session object */ + int *pnChangeset, /* OUT: Size of buffer at *ppChangeset */ + void **ppChangeset /* OUT: Buffer containing changeset */ +); + +/* +** CAPI3REF: Load The Difference Between Tables Into A Session +** METHOD: sqlite3_session +** +** If it is not already attached to the session object passed as the first +** argument, this function attaches table zTbl in the same manner as the +** [sqlite3session_attach()] function. If zTbl does not exist, or if it +** does not have a primary key, this function is a no-op (but does not return +** an error). +** +** Argument zFromDb must be the name of a database ("main", "temp" etc.) +** attached to the same database handle as the session object that contains +** a table compatible with the table attached to the session by this function. +** A table is considered compatible if it: +** +**
    +**
  • Has the same name, +**
  • Has the same set of columns declared in the same order, and +**
  • Has the same PRIMARY KEY definition. +**
+** +** If the tables are not compatible, SQLITE_SCHEMA is returned. If the tables +** are compatible but do not have any PRIMARY KEY columns, it is not an error +** but no changes are added to the session object. As with other session +** APIs, tables without PRIMARY KEYs are simply ignored. +** +** This function adds a set of changes to the session object that could be +** used to update the table in database zFrom (call this the "from-table") +** so that its content is the same as the table attached to the session +** object (call this the "to-table"). Specifically: +** +**
    +**
  • For each row (primary key) that exists in the to-table but not in +** the from-table, an INSERT record is added to the session object. +** +**
  • For each row (primary key) that exists in the to-table but not in +** the from-table, a DELETE record is added to the session object. +** +**
  • For each row (primary key) that exists in both tables, but features +** different non-PK values in each, an UPDATE record is added to the +** session. +**
+** +** To clarify, if this function is called and then a changeset constructed +** using [sqlite3session_changeset()], then after applying that changeset to +** database zFrom the contents of the two compatible tables would be +** identical. +** +** It an error if database zFrom does not exist or does not contain the +** required compatible table. +** +** If the operation successful, SQLITE_OK is returned. Otherwise, an SQLite +** error code. In this case, if argument pzErrMsg is not NULL, *pzErrMsg +** may be set to point to a buffer containing an English language error +** message. It is the responsibility of the caller to free this buffer using +** sqlite3_free(). +*/ +SQLITE_API int sqlite3session_diff( + sqlite3_session *pSession, + const char *zFromDb, + const char *zTbl, + char **pzErrMsg +); + + +/* +** CAPI3REF: Generate A Patchset From A Session Object +** METHOD: sqlite3_session +** +** The differences between a patchset and a changeset are that: +** +**
    +**
  • DELETE records consist of the primary key fields only. The +** original values of other fields are omitted. +**
  • The original values of any modified fields are omitted from +** UPDATE records. +**
+** +** A patchset blob may be used with up to date versions of all +** sqlite3changeset_xxx API functions except for sqlite3changeset_invert(), +** which returns SQLITE_CORRUPT if it is passed a patchset. Similarly, +** attempting to use a patchset blob with old versions of the +** sqlite3changeset_xxx APIs also provokes an SQLITE_CORRUPT error. +** +** Because the non-primary key "old.*" fields are omitted, no +** SQLITE_CHANGESET_DATA conflicts can be detected or reported if a patchset +** is passed to the sqlite3changeset_apply() API. Other conflict types work +** in the same way as for changesets. +** +** Changes within a patchset are ordered in the same way as for changesets +** generated by the sqlite3session_changeset() function (i.e. all changes for +** a single table are grouped together, tables appear in the order in which +** they were attached to the session object). +*/ +SQLITE_API int sqlite3session_patchset( + sqlite3_session *pSession, /* Session object */ + int *pnPatchset, /* OUT: Size of buffer at *ppPatchset */ + void **ppPatchset /* OUT: Buffer containing patchset */ +); + +/* +** CAPI3REF: Test if a changeset has recorded any changes. +** +** Return non-zero if no changes to attached tables have been recorded by +** the session object passed as the first argument. Otherwise, if one or +** more changes have been recorded, return zero. +** +** Even if this function returns zero, it is possible that calling +** [sqlite3session_changeset()] on the session handle may still return a +** changeset that contains no changes. This can happen when a row in +** an attached table is modified and then later on the original values +** are restored. However, if this function returns non-zero, then it is +** guaranteed that a call to sqlite3session_changeset() will return a +** changeset containing zero changes. +*/ +SQLITE_API int sqlite3session_isempty(sqlite3_session *pSession); + +/* +** CAPI3REF: Create An Iterator To Traverse A Changeset +** CONSTRUCTOR: sqlite3_changeset_iter +** +** Create an iterator used to iterate through the contents of a changeset. +** If successful, *pp is set to point to the iterator handle and SQLITE_OK +** is returned. Otherwise, if an error occurs, *pp is set to zero and an +** SQLite error code is returned. +** +** The following functions can be used to advance and query a changeset +** iterator created by this function: +** +**
    +**
  • [sqlite3changeset_next()] +**
  • [sqlite3changeset_op()] +**
  • [sqlite3changeset_new()] +**
  • [sqlite3changeset_old()] +**
+** +** It is the responsibility of the caller to eventually destroy the iterator +** by passing it to [sqlite3changeset_finalize()]. The buffer containing the +** changeset (pChangeset) must remain valid until after the iterator is +** destroyed. +** +** Assuming the changeset blob was created by one of the +** [sqlite3session_changeset()], [sqlite3changeset_concat()] or +** [sqlite3changeset_invert()] functions, all changes within the changeset +** that apply to a single table are grouped together. This means that when +** an application iterates through a changeset using an iterator created by +** this function, all changes that relate to a single table are visited +** consecutively. There is no chance that the iterator will visit a change +** the applies to table X, then one for table Y, and then later on visit +** another change for table X. +** +** The behavior of sqlite3changeset_start_v2() and its streaming equivalent +** may be modified by passing a combination of +** [SQLITE_CHANGESETSTART_INVERT | supported flags] as the 4th parameter. +** +** Note that the sqlite3changeset_start_v2() API is still experimental +** and therefore subject to change. +*/ +SQLITE_API int sqlite3changeset_start( + sqlite3_changeset_iter **pp, /* OUT: New changeset iterator handle */ + int nChangeset, /* Size of changeset blob in bytes */ + void *pChangeset /* Pointer to blob containing changeset */ +); +SQLITE_API int sqlite3changeset_start_v2( + sqlite3_changeset_iter **pp, /* OUT: New changeset iterator handle */ + int nChangeset, /* Size of changeset blob in bytes */ + void *pChangeset, /* Pointer to blob containing changeset */ + int flags /* SESSION_CHANGESETSTART_* flags */ +); + +/* +** CAPI3REF: Flags for sqlite3changeset_start_v2 +** +** The following flags may passed via the 4th parameter to +** [sqlite3changeset_start_v2] and [sqlite3changeset_start_v2_strm]: +** +**
SQLITE_CHANGESETAPPLY_INVERT
+** Invert the changeset while iterating through it. This is equivalent to +** inverting a changeset using sqlite3changeset_invert() before applying it. +** It is an error to specify this flag with a patchset. +*/ +#define SQLITE_CHANGESETSTART_INVERT 0x0002 + + +/* +** CAPI3REF: Advance A Changeset Iterator +** METHOD: sqlite3_changeset_iter +** +** This function may only be used with iterators created by function +** [sqlite3changeset_start()]. If it is called on an iterator passed to +** a conflict-handler callback by [sqlite3changeset_apply()], SQLITE_MISUSE +** is returned and the call has no effect. +** +** Immediately after an iterator is created by sqlite3changeset_start(), it +** does not point to any change in the changeset. Assuming the changeset +** is not empty, the first call to this function advances the iterator to +** point to the first change in the changeset. Each subsequent call advances +** the iterator to point to the next change in the changeset (if any). If +** no error occurs and the iterator points to a valid change after a call +** to sqlite3changeset_next() has advanced it, SQLITE_ROW is returned. +** Otherwise, if all changes in the changeset have already been visited, +** SQLITE_DONE is returned. +** +** If an error occurs, an SQLite error code is returned. Possible error +** codes include SQLITE_CORRUPT (if the changeset buffer is corrupt) or +** SQLITE_NOMEM. +*/ +SQLITE_API int sqlite3changeset_next(sqlite3_changeset_iter *pIter); + +/* +** CAPI3REF: Obtain The Current Operation From A Changeset Iterator +** METHOD: sqlite3_changeset_iter +** +** The pIter argument passed to this function may either be an iterator +** passed to a conflict-handler by [sqlite3changeset_apply()], or an iterator +** created by [sqlite3changeset_start()]. In the latter case, the most recent +** call to [sqlite3changeset_next()] must have returned [SQLITE_ROW]. If this +** is not the case, this function returns [SQLITE_MISUSE]. +** +** If argument pzTab is not NULL, then *pzTab is set to point to a +** nul-terminated utf-8 encoded string containing the name of the table +** affected by the current change. The buffer remains valid until either +** sqlite3changeset_next() is called on the iterator or until the +** conflict-handler function returns. If pnCol is not NULL, then *pnCol is +** set to the number of columns in the table affected by the change. If +** pbIndirect is not NULL, then *pbIndirect is set to true (1) if the change +** is an indirect change, or false (0) otherwise. See the documentation for +** [sqlite3session_indirect()] for a description of direct and indirect +** changes. Finally, if pOp is not NULL, then *pOp is set to one of +** [SQLITE_INSERT], [SQLITE_DELETE] or [SQLITE_UPDATE], depending on the +** type of change that the iterator currently points to. +** +** If no error occurs, SQLITE_OK is returned. If an error does occur, an +** SQLite error code is returned. The values of the output variables may not +** be trusted in this case. +*/ +SQLITE_API int sqlite3changeset_op( + sqlite3_changeset_iter *pIter, /* Iterator object */ + const char **pzTab, /* OUT: Pointer to table name */ + int *pnCol, /* OUT: Number of columns in table */ + int *pOp, /* OUT: SQLITE_INSERT, DELETE or UPDATE */ + int *pbIndirect /* OUT: True for an 'indirect' change */ +); + +/* +** CAPI3REF: Obtain The Primary Key Definition Of A Table +** METHOD: sqlite3_changeset_iter +** +** For each modified table, a changeset includes the following: +** +**
    +**
  • The number of columns in the table, and +**
  • Which of those columns make up the tables PRIMARY KEY. +**
+** +** This function is used to find which columns comprise the PRIMARY KEY of +** the table modified by the change that iterator pIter currently points to. +** If successful, *pabPK is set to point to an array of nCol entries, where +** nCol is the number of columns in the table. Elements of *pabPK are set to +** 0x01 if the corresponding column is part of the tables primary key, or +** 0x00 if it is not. +** +** If argument pnCol is not NULL, then *pnCol is set to the number of columns +** in the table. +** +** If this function is called when the iterator does not point to a valid +** entry, SQLITE_MISUSE is returned and the output variables zeroed. Otherwise, +** SQLITE_OK is returned and the output variables populated as described +** above. +*/ +SQLITE_API int sqlite3changeset_pk( + sqlite3_changeset_iter *pIter, /* Iterator object */ + unsigned char **pabPK, /* OUT: Array of boolean - true for PK cols */ + int *pnCol /* OUT: Number of entries in output array */ +); + +/* +** CAPI3REF: Obtain old.* Values From A Changeset Iterator +** METHOD: sqlite3_changeset_iter +** +** The pIter argument passed to this function may either be an iterator +** passed to a conflict-handler by [sqlite3changeset_apply()], or an iterator +** created by [sqlite3changeset_start()]. In the latter case, the most recent +** call to [sqlite3changeset_next()] must have returned SQLITE_ROW. +** Furthermore, it may only be called if the type of change that the iterator +** currently points to is either [SQLITE_DELETE] or [SQLITE_UPDATE]. Otherwise, +** this function returns [SQLITE_MISUSE] and sets *ppValue to NULL. +** +** Argument iVal must be greater than or equal to 0, and less than the number +** of columns in the table affected by the current change. Otherwise, +** [SQLITE_RANGE] is returned and *ppValue is set to NULL. +** +** If successful, this function sets *ppValue to point to a protected +** sqlite3_value object containing the iVal'th value from the vector of +** original row values stored as part of the UPDATE or DELETE change and +** returns SQLITE_OK. The name of the function comes from the fact that this +** is similar to the "old.*" columns available to update or delete triggers. +** +** If some other error occurs (e.g. an OOM condition), an SQLite error code +** is returned and *ppValue is set to NULL. +*/ +SQLITE_API int sqlite3changeset_old( + sqlite3_changeset_iter *pIter, /* Changeset iterator */ + int iVal, /* Column number */ + sqlite3_value **ppValue /* OUT: Old value (or NULL pointer) */ +); + +/* +** CAPI3REF: Obtain new.* Values From A Changeset Iterator +** METHOD: sqlite3_changeset_iter +** +** The pIter argument passed to this function may either be an iterator +** passed to a conflict-handler by [sqlite3changeset_apply()], or an iterator +** created by [sqlite3changeset_start()]. In the latter case, the most recent +** call to [sqlite3changeset_next()] must have returned SQLITE_ROW. +** Furthermore, it may only be called if the type of change that the iterator +** currently points to is either [SQLITE_UPDATE] or [SQLITE_INSERT]. Otherwise, +** this function returns [SQLITE_MISUSE] and sets *ppValue to NULL. +** +** Argument iVal must be greater than or equal to 0, and less than the number +** of columns in the table affected by the current change. Otherwise, +** [SQLITE_RANGE] is returned and *ppValue is set to NULL. +** +** If successful, this function sets *ppValue to point to a protected +** sqlite3_value object containing the iVal'th value from the vector of +** new row values stored as part of the UPDATE or INSERT change and +** returns SQLITE_OK. If the change is an UPDATE and does not include +** a new value for the requested column, *ppValue is set to NULL and +** SQLITE_OK returned. The name of the function comes from the fact that +** this is similar to the "new.*" columns available to update or delete +** triggers. +** +** If some other error occurs (e.g. an OOM condition), an SQLite error code +** is returned and *ppValue is set to NULL. +*/ +SQLITE_API int sqlite3changeset_new( + sqlite3_changeset_iter *pIter, /* Changeset iterator */ + int iVal, /* Column number */ + sqlite3_value **ppValue /* OUT: New value (or NULL pointer) */ +); + +/* +** CAPI3REF: Obtain Conflicting Row Values From A Changeset Iterator +** METHOD: sqlite3_changeset_iter +** +** This function should only be used with iterator objects passed to a +** conflict-handler callback by [sqlite3changeset_apply()] with either +** [SQLITE_CHANGESET_DATA] or [SQLITE_CHANGESET_CONFLICT]. If this function +** is called on any other iterator, [SQLITE_MISUSE] is returned and *ppValue +** is set to NULL. +** +** Argument iVal must be greater than or equal to 0, and less than the number +** of columns in the table affected by the current change. Otherwise, +** [SQLITE_RANGE] is returned and *ppValue is set to NULL. +** +** If successful, this function sets *ppValue to point to a protected +** sqlite3_value object containing the iVal'th value from the +** "conflicting row" associated with the current conflict-handler callback +** and returns SQLITE_OK. +** +** If some other error occurs (e.g. an OOM condition), an SQLite error code +** is returned and *ppValue is set to NULL. +*/ +SQLITE_API int sqlite3changeset_conflict( + sqlite3_changeset_iter *pIter, /* Changeset iterator */ + int iVal, /* Column number */ + sqlite3_value **ppValue /* OUT: Value from conflicting row */ +); + +/* +** CAPI3REF: Determine The Number Of Foreign Key Constraint Violations +** METHOD: sqlite3_changeset_iter +** +** This function may only be called with an iterator passed to an +** SQLITE_CHANGESET_FOREIGN_KEY conflict handler callback. In this case +** it sets the output variable to the total number of known foreign key +** violations in the destination database and returns SQLITE_OK. +** +** In all other cases this function returns SQLITE_MISUSE. +*/ +SQLITE_API int sqlite3changeset_fk_conflicts( + sqlite3_changeset_iter *pIter, /* Changeset iterator */ + int *pnOut /* OUT: Number of FK violations */ +); + + +/* +** CAPI3REF: Finalize A Changeset Iterator +** METHOD: sqlite3_changeset_iter +** +** This function is used to finalize an iterator allocated with +** [sqlite3changeset_start()]. +** +** This function should only be called on iterators created using the +** [sqlite3changeset_start()] function. If an application calls this +** function with an iterator passed to a conflict-handler by +** [sqlite3changeset_apply()], [SQLITE_MISUSE] is immediately returned and the +** call has no effect. +** +** If an error was encountered within a call to an sqlite3changeset_xxx() +** function (for example an [SQLITE_CORRUPT] in [sqlite3changeset_next()] or an +** [SQLITE_NOMEM] in [sqlite3changeset_new()]) then an error code corresponding +** to that error is returned by this function. Otherwise, SQLITE_OK is +** returned. This is to allow the following pattern (pseudo-code): +** +** 
+**   sqlite3changeset_start();
+**   while( SQLITE_ROW==sqlite3changeset_next() ){
+**     // Do something with change.
+**   }
+**   rc = sqlite3changeset_finalize();
+**   if( rc!=SQLITE_OK ){
+**     // An error has occurred 
+**   }
+**
+*/ +SQLITE_API int sqlite3changeset_finalize(sqlite3_changeset_iter *pIter); + +/* +** CAPI3REF: Invert A Changeset +** +** This function is used to "invert" a changeset object. Applying an inverted +** changeset to a database reverses the effects of applying the uninverted +** changeset. Specifically: +** +**
    +**
  • Each DELETE change is changed to an INSERT, and +**
  • Each INSERT change is changed to a DELETE, and +**
  • For each UPDATE change, the old.* and new.* values are exchanged. +**
+** +** This function does not change the order in which changes appear within +** the changeset. It merely reverses the sense of each individual change. +** +** If successful, a pointer to a buffer containing the inverted changeset +** is stored in *ppOut, the size of the same buffer is stored in *pnOut, and +** SQLITE_OK is returned. If an error occurs, both *pnOut and *ppOut are +** zeroed and an SQLite error code returned. +** +** It is the responsibility of the caller to eventually call sqlite3_free() +** on the *ppOut pointer to free the buffer allocation following a successful +** call to this function. +** +** WARNING/TODO: This function currently assumes that the input is a valid +** changeset. If it is not, the results are undefined. +*/ +SQLITE_API int sqlite3changeset_invert( + int nIn, const void *pIn, /* Input changeset */ + int *pnOut, void **ppOut /* OUT: Inverse of input */ +); + +/* +** CAPI3REF: Concatenate Two Changeset Objects +** +** This function is used to concatenate two changesets, A and B, into a +** single changeset. The result is a changeset equivalent to applying +** changeset A followed by changeset B. +** +** This function combines the two input changesets using an +** sqlite3_changegroup object. Calling it produces similar results as the +** following code fragment: +** +** 
+**   sqlite3_changegroup *pGrp;
+**   rc = sqlite3_changegroup_new(&pGrp);
+**   if( rc==SQLITE_OK ) rc = sqlite3changegroup_add(pGrp, nA, pA);
+**   if( rc==SQLITE_OK ) rc = sqlite3changegroup_add(pGrp, nB, pB);
+**   if( rc==SQLITE_OK ){
+**     rc = sqlite3changegroup_output(pGrp, pnOut, ppOut);
+**   }else{
+**     *ppOut = 0;
+**     *pnOut = 0;
+**   }
+**
+** +** Refer to the sqlite3_changegroup documentation below for details. +*/ +SQLITE_API int sqlite3changeset_concat( + int nA, /* Number of bytes in buffer pA */ + void *pA, /* Pointer to buffer containing changeset A */ + int nB, /* Number of bytes in buffer pB */ + void *pB, /* Pointer to buffer containing changeset B */ + int *pnOut, /* OUT: Number of bytes in output changeset */ + void **ppOut /* OUT: Buffer containing output changeset */ +); + + +/* +** CAPI3REF: Changegroup Handle +** +** A changegroup is an object used to combine two or more +** [changesets] or [patchsets] +*/ +typedef struct sqlite3_changegroup sqlite3_changegroup; + +/* +** CAPI3REF: Create A New Changegroup Object +** CONSTRUCTOR: sqlite3_changegroup +** +** An sqlite3_changegroup object is used to combine two or more changesets +** (or patchsets) into a single changeset (or patchset). A single changegroup +** object may combine changesets or patchsets, but not both. The output is +** always in the same format as the input. +** +** If successful, this function returns SQLITE_OK and populates (*pp) with +** a pointer to a new sqlite3_changegroup object before returning. The caller +** should eventually free the returned object using a call to +** sqlite3changegroup_delete(). If an error occurs, an SQLite error code +** (i.e. SQLITE_NOMEM) is returned and *pp is set to NULL. +** +** The usual usage pattern for an sqlite3_changegroup object is as follows: +** +**
    +**
  • It is created using a call to sqlite3changegroup_new(). +** +**
  • Zero or more changesets (or patchsets) are added to the object +** by calling sqlite3changegroup_add(). +** +**
  • The result of combining all input changesets together is obtained +** by the application via a call to sqlite3changegroup_output(). +** +**
  • The object is deleted using a call to sqlite3changegroup_delete(). +**
+** +** Any number of calls to add() and output() may be made between the calls to +** new() and delete(), and in any order. +** +** As well as the regular sqlite3changegroup_add() and +** sqlite3changegroup_output() functions, also available are the streaming +** versions sqlite3changegroup_add_strm() and sqlite3changegroup_output_strm(). +*/ +SQLITE_API int sqlite3changegroup_new(sqlite3_changegroup **pp); + +/* +** CAPI3REF: Add A Changeset To A Changegroup +** METHOD: sqlite3_changegroup +** +** Add all changes within the changeset (or patchset) in buffer pData (size +** nData bytes) to the changegroup. +** +** If the buffer contains a patchset, then all prior calls to this function +** on the same changegroup object must also have specified patchsets. Or, if +** the buffer contains a changeset, so must have the earlier calls to this +** function. Otherwise, SQLITE_ERROR is returned and no changes are added +** to the changegroup. +** +** Rows within the changeset and changegroup are identified by the values in +** their PRIMARY KEY columns. A change in the changeset is considered to +** apply to the same row as a change already present in the changegroup if +** the two rows have the same primary key. +** +** Changes to rows that do not already appear in the changegroup are +** simply copied into it. Or, if both the new changeset and the changegroup +** contain changes that apply to a single row, the final contents of the +** changegroup depends on the type of each change, as follows: +** +** +** +** +**
Existing Change New Change Output Change +**
INSERT INSERT +** The new change is ignored. This case does not occur if the new +** changeset was recorded immediately after the changesets already +** added to the changegroup. +**
INSERT UPDATE +** The INSERT change remains in the changegroup. The values in the +** INSERT change are modified as if the row was inserted by the +** existing change and then updated according to the new change. +**
INSERT DELETE +** The existing INSERT is removed from the changegroup. The DELETE is +** not added. +**
UPDATE INSERT +** The new change is ignored. This case does not occur if the new +** changeset was recorded immediately after the changesets already +** added to the changegroup. +**
UPDATE UPDATE +** The existing UPDATE remains within the changegroup. It is amended +** so that the accompanying values are as if the row was updated once +** by the existing change and then again by the new change. +**
UPDATE DELETE +** The existing UPDATE is replaced by the new DELETE within the +** changegroup. +**
DELETE INSERT +** If one or more of the column values in the row inserted by the +** new change differ from those in the row deleted by the existing +** change, the existing DELETE is replaced by an UPDATE within the +** changegroup. Otherwise, if the inserted row is exactly the same +** as the deleted row, the existing DELETE is simply discarded. +**
DELETE UPDATE +** The new change is ignored. This case does not occur if the new +** changeset was recorded immediately after the changesets already +** added to the changegroup. +**
DELETE DELETE +** The new change is ignored. This case does not occur if the new +** changeset was recorded immediately after the changesets already +** added to the changegroup. +**
+** +** If the new changeset contains changes to a table that is already present +** in the changegroup, then the number of columns and the position of the +** primary key columns for the table must be consistent. If this is not the +** case, this function fails with SQLITE_SCHEMA. If the input changeset +** appears to be corrupt and the corruption is detected, SQLITE_CORRUPT is +** returned. Or, if an out-of-memory condition occurs during processing, this +** function returns SQLITE_NOMEM. In all cases, if an error occurs the +** final contents of the changegroup is undefined. +** +** If no error occurs, SQLITE_OK is returned. +*/ +SQLITE_API int sqlite3changegroup_add(sqlite3_changegroup*, int nData, void *pData); + +/* +** CAPI3REF: Obtain A Composite Changeset From A Changegroup +** METHOD: sqlite3_changegroup +** +** Obtain a buffer containing a changeset (or patchset) representing the +** current contents of the changegroup. If the inputs to the changegroup +** were themselves changesets, the output is a changeset. Or, if the +** inputs were patchsets, the output is also a patchset. +** +** As with the output of the sqlite3session_changeset() and +** sqlite3session_patchset() functions, all changes related to a single +** table are grouped together in the output of this function. Tables appear +** in the same order as for the very first changeset added to the changegroup. +** If the second or subsequent changesets added to the changegroup contain +** changes for tables that do not appear in the first changeset, they are +** appended onto the end of the output changeset, again in the order in +** which they are first encountered. +** +** If an error occurs, an SQLite error code is returned and the output +** variables (*pnData) and (*ppData) are set to 0. Otherwise, SQLITE_OK +** is returned and the output variables are set to the size of and a +** pointer to the output buffer, respectively. In this case it is the +** responsibility of the caller to eventually free the buffer using a +** call to sqlite3_free(). +*/ +SQLITE_API int sqlite3changegroup_output( + sqlite3_changegroup*, + int *pnData, /* OUT: Size of output buffer in bytes */ + void **ppData /* OUT: Pointer to output buffer */ +); + +/* +** CAPI3REF: Delete A Changegroup Object +** DESTRUCTOR: sqlite3_changegroup +*/ +SQLITE_API void sqlite3changegroup_delete(sqlite3_changegroup*); + +/* +** CAPI3REF: Apply A Changeset To A Database +** +** Apply a changeset or patchset to a database. These functions attempt to +** update the "main" database attached to handle db with the changes found in +** the changeset passed via the second and third arguments. +** +** The fourth argument (xFilter) passed to these functions is the "filter +** callback". If it is not NULL, then for each table affected by at least one +** change in the changeset, the filter callback is invoked with +** the table name as the second argument, and a copy of the context pointer +** passed as the sixth argument as the first. If the "filter callback" +** returns zero, then no attempt is made to apply any changes to the table. +** Otherwise, if the return value is non-zero or the xFilter argument to +** is NULL, all changes related to the table are attempted. +** +** For each table that is not excluded by the filter callback, this function +** tests that the target database contains a compatible table. A table is +** considered compatible if all of the following are true: +** +**
    +**
  • The table has the same name as the name recorded in the +** changeset, and +**
  • The table has at least as many columns as recorded in the +** changeset, and +**
  • The table has primary key columns in the same position as +** recorded in the changeset. +**
+** +** If there is no compatible table, it is not an error, but none of the +** changes associated with the table are applied. A warning message is issued +** via the sqlite3_log() mechanism with the error code SQLITE_SCHEMA. At most +** one such warning is issued for each table in the changeset. +** +** For each change for which there is a compatible table, an attempt is made +** to modify the table contents according to the UPDATE, INSERT or DELETE +** change. If a change cannot be applied cleanly, the conflict handler +** function passed as the fifth argument to sqlite3changeset_apply() may be +** invoked. A description of exactly when the conflict handler is invoked for +** each type of change is below. +** +** Unlike the xFilter argument, xConflict may not be passed NULL. The results +** of passing anything other than a valid function pointer as the xConflict +** argument are undefined. +** +** Each time the conflict handler function is invoked, it must return one +** of [SQLITE_CHANGESET_OMIT], [SQLITE_CHANGESET_ABORT] or +** [SQLITE_CHANGESET_REPLACE]. SQLITE_CHANGESET_REPLACE may only be returned +** if the second argument passed to the conflict handler is either +** SQLITE_CHANGESET_DATA or SQLITE_CHANGESET_CONFLICT. If the conflict-handler +** returns an illegal value, any changes already made are rolled back and +** the call to sqlite3changeset_apply() returns SQLITE_MISUSE. Different +** actions are taken by sqlite3changeset_apply() depending on the value +** returned by each invocation of the conflict-handler function. Refer to +** the documentation for the three +** [SQLITE_CHANGESET_OMIT|available return values] for details. +** +**
+**
DELETE Changes
+** For each DELETE change, the function checks if the target database +** contains a row with the same primary key value (or values) as the +** original row values stored in the changeset. If it does, and the values +** stored in all non-primary key columns also match the values stored in +** the changeset the row is deleted from the target database. +** +** If a row with matching primary key values is found, but one or more of +** the non-primary key fields contains a value different from the original +** row value stored in the changeset, the conflict-handler function is +** invoked with [SQLITE_CHANGESET_DATA] as the second argument. If the +** database table has more columns than are recorded in the changeset, +** only the values of those non-primary key fields are compared against +** the current database contents - any trailing database table columns +** are ignored. +** +** If no row with matching primary key values is found in the database, +** the conflict-handler function is invoked with [SQLITE_CHANGESET_NOTFOUND] +** passed as the second argument. +** +** If the DELETE operation is attempted, but SQLite returns SQLITE_CONSTRAINT +** (which can only happen if a foreign key constraint is violated), the +** conflict-handler function is invoked with [SQLITE_CHANGESET_CONSTRAINT] +** passed as the second argument. This includes the case where the DELETE +** operation is attempted because an earlier call to the conflict handler +** function returned [SQLITE_CHANGESET_REPLACE]. +** +**
INSERT Changes
+** For each INSERT change, an attempt is made to insert the new row into +** the database. If the changeset row contains fewer fields than the +** database table, the trailing fields are populated with their default +** values. +** +** If the attempt to insert the row fails because the database already +** contains a row with the same primary key values, the conflict handler +** function is invoked with the second argument set to +** [SQLITE_CHANGESET_CONFLICT]. +** +** If the attempt to insert the row fails because of some other constraint +** violation (e.g. NOT NULL or UNIQUE), the conflict handler function is +** invoked with the second argument set to [SQLITE_CHANGESET_CONSTRAINT]. +** This includes the case where the INSERT operation is re-attempted because +** an earlier call to the conflict handler function returned +** [SQLITE_CHANGESET_REPLACE]. +** +**
UPDATE Changes
+** For each UPDATE change, the function checks if the target database +** contains a row with the same primary key value (or values) as the +** original row values stored in the changeset. If it does, and the values +** stored in all modified non-primary key columns also match the values +** stored in the changeset the row is updated within the target database. +** +** If a row with matching primary key values is found, but one or more of +** the modified non-primary key fields contains a value different from an +** original row value stored in the changeset, the conflict-handler function +** is invoked with [SQLITE_CHANGESET_DATA] as the second argument. Since +** UPDATE changes only contain values for non-primary key fields that are +** to be modified, only those fields need to match the original values to +** avoid the SQLITE_CHANGESET_DATA conflict-handler callback. +** +** If no row with matching primary key values is found in the database, +** the conflict-handler function is invoked with [SQLITE_CHANGESET_NOTFOUND] +** passed as the second argument. +** +** If the UPDATE operation is attempted, but SQLite returns +** SQLITE_CONSTRAINT, the conflict-handler function is invoked with +** [SQLITE_CHANGESET_CONSTRAINT] passed as the second argument. +** This includes the case where the UPDATE operation is attempted after +** an earlier call to the conflict handler function returned +** [SQLITE_CHANGESET_REPLACE]. +**
+** +** It is safe to execute SQL statements, including those that write to the +** table that the callback related to, from within the xConflict callback. +** This can be used to further customize the applications conflict +** resolution strategy. +** +** All changes made by these functions are enclosed in a savepoint transaction. +** If any other error (aside from a constraint failure when attempting to +** write to the target database) occurs, then the savepoint transaction is +** rolled back, restoring the target database to its original state, and an +** SQLite error code returned. +** +** If the output parameters (ppRebase) and (pnRebase) are non-NULL and +** the input is a changeset (not a patchset), then sqlite3changeset_apply_v2() +** may set (*ppRebase) to point to a "rebase" that may be used with the +** sqlite3_rebaser APIs buffer before returning. In this case (*pnRebase) +** is set to the size of the buffer in bytes. It is the responsibility of the +** caller to eventually free any such buffer using sqlite3_free(). The buffer +** is only allocated and populated if one or more conflicts were encountered +** while applying the patchset. See comments surrounding the sqlite3_rebaser +** APIs for further details. +** +** The behavior of sqlite3changeset_apply_v2() and its streaming equivalent +** may be modified by passing a combination of +** [SQLITE_CHANGESETAPPLY_NOSAVEPOINT | supported flags] as the 9th parameter. +** +** Note that the sqlite3changeset_apply_v2() API is still experimental +** and therefore subject to change. +*/ +SQLITE_API int sqlite3changeset_apply( + sqlite3 *db, /* Apply change to "main" db of this handle */ + int nChangeset, /* Size of changeset in bytes */ + void *pChangeset, /* Changeset blob */ + int(*xFilter)( + void *pCtx, /* Copy of sixth arg to _apply() */ + const char *zTab /* Table name */ + ), + int(*xConflict)( + void *pCtx, /* Copy of sixth arg to _apply() */ + int eConflict, /* DATA, MISSING, CONFLICT, CONSTRAINT */ + sqlite3_changeset_iter *p /* Handle describing change and conflict */ + ), + void *pCtx /* First argument passed to xConflict */ +); +SQLITE_API int sqlite3changeset_apply_v2( + sqlite3 *db, /* Apply change to "main" db of this handle */ + int nChangeset, /* Size of changeset in bytes */ + void *pChangeset, /* Changeset blob */ + int(*xFilter)( + void *pCtx, /* Copy of sixth arg to _apply() */ + const char *zTab /* Table name */ + ), + int(*xConflict)( + void *pCtx, /* Copy of sixth arg to _apply() */ + int eConflict, /* DATA, MISSING, CONFLICT, CONSTRAINT */ + sqlite3_changeset_iter *p /* Handle describing change and conflict */ + ), + void *pCtx, /* First argument passed to xConflict */ + void **ppRebase, int *pnRebase, /* OUT: Rebase data */ + int flags /* SESSION_CHANGESETAPPLY_* flags */ +); + +/* +** CAPI3REF: Flags for sqlite3changeset_apply_v2 +** +** The following flags may passed via the 9th parameter to +** [sqlite3changeset_apply_v2] and [sqlite3changeset_apply_v2_strm]: +** +**
+**
SQLITE_CHANGESETAPPLY_NOSAVEPOINT
+** Usually, the sessions module encloses all operations performed by +** a single call to apply_v2() or apply_v2_strm() in a [SAVEPOINT]. The +** SAVEPOINT is committed if the changeset or patchset is successfully +** applied, or rolled back if an error occurs. Specifying this flag +** causes the sessions module to omit this savepoint. In this case, if the +** caller has an open transaction or savepoint when apply_v2() is called, +** it may revert the partially applied changeset by rolling it back. +** +**
SQLITE_CHANGESETAPPLY_INVERT
+** Invert the changeset before applying it. This is equivalent to inverting +** a changeset using sqlite3changeset_invert() before applying it. It is +** an error to specify this flag with a patchset. +*/ +#define SQLITE_CHANGESETAPPLY_NOSAVEPOINT 0x0001 +#define SQLITE_CHANGESETAPPLY_INVERT 0x0002 + +/* +** CAPI3REF: Constants Passed To The Conflict Handler +** +** Values that may be passed as the second argument to a conflict-handler. +** +**
+**
SQLITE_CHANGESET_DATA
+** The conflict handler is invoked with CHANGESET_DATA as the second argument +** when processing a DELETE or UPDATE change if a row with the required +** PRIMARY KEY fields is present in the database, but one or more other +** (non primary-key) fields modified by the update do not contain the +** expected "before" values. +** +** The conflicting row, in this case, is the database row with the matching +** primary key. +** +**
SQLITE_CHANGESET_NOTFOUND
+** The conflict handler is invoked with CHANGESET_NOTFOUND as the second +** argument when processing a DELETE or UPDATE change if a row with the +** required PRIMARY KEY fields is not present in the database. +** +** There is no conflicting row in this case. The results of invoking the +** sqlite3changeset_conflict() API are undefined. +** +**
SQLITE_CHANGESET_CONFLICT
+** CHANGESET_CONFLICT is passed as the second argument to the conflict +** handler while processing an INSERT change if the operation would result +** in duplicate primary key values. +** +** The conflicting row in this case is the database row with the matching +** primary key. +** +**
SQLITE_CHANGESET_FOREIGN_KEY
+** If foreign key handling is enabled, and applying a changeset leaves the +** database in a state containing foreign key violations, the conflict +** handler is invoked with CHANGESET_FOREIGN_KEY as the second argument +** exactly once before the changeset is committed. If the conflict handler +** returns CHANGESET_OMIT, the changes, including those that caused the +** foreign key constraint violation, are committed. Or, if it returns +** CHANGESET_ABORT, the changeset is rolled back. +** +** No current or conflicting row information is provided. The only function +** it is possible to call on the supplied sqlite3_changeset_iter handle +** is sqlite3changeset_fk_conflicts(). +** +**
SQLITE_CHANGESET_CONSTRAINT
+** If any other constraint violation occurs while applying a change (i.e. +** a UNIQUE, CHECK or NOT NULL constraint), the conflict handler is +** invoked with CHANGESET_CONSTRAINT as the second argument. +** +** There is no conflicting row in this case. The results of invoking the +** sqlite3changeset_conflict() API are undefined. +** +**
+*/ +#define SQLITE_CHANGESET_DATA 1 +#define SQLITE_CHANGESET_NOTFOUND 2 +#define SQLITE_CHANGESET_CONFLICT 3 +#define SQLITE_CHANGESET_CONSTRAINT 4 +#define SQLITE_CHANGESET_FOREIGN_KEY 5 + +/* +** CAPI3REF: Constants Returned By The Conflict Handler +** +** A conflict handler callback must return one of the following three values. +** +**
+**
SQLITE_CHANGESET_OMIT
+** If a conflict handler returns this value no special action is taken. The +** change that caused the conflict is not applied. The session module +** continues to the next change in the changeset. +** +**
SQLITE_CHANGESET_REPLACE
+** This value may only be returned if the second argument to the conflict +** handler was SQLITE_CHANGESET_DATA or SQLITE_CHANGESET_CONFLICT. If this +** is not the case, any changes applied so far are rolled back and the +** call to sqlite3changeset_apply() returns SQLITE_MISUSE. +** +** If CHANGESET_REPLACE is returned by an SQLITE_CHANGESET_DATA conflict +** handler, then the conflicting row is either updated or deleted, depending +** on the type of change. +** +** If CHANGESET_REPLACE is returned by an SQLITE_CHANGESET_CONFLICT conflict +** handler, then the conflicting row is removed from the database and a +** second attempt to apply the change is made. If this second attempt fails, +** the original row is restored to the database before continuing. +** +**
SQLITE_CHANGESET_ABORT
+** If this value is returned, any changes applied so far are rolled back +** and the call to sqlite3changeset_apply() returns SQLITE_ABORT. +**
+*/ +#define SQLITE_CHANGESET_OMIT 0 +#define SQLITE_CHANGESET_REPLACE 1 +#define SQLITE_CHANGESET_ABORT 2 + +/* +** CAPI3REF: Rebasing changesets +** EXPERIMENTAL +** +** Suppose there is a site hosting a database in state S0. And that +** modifications are made that move that database to state S1 and a +** changeset recorded (the "local" changeset). Then, a changeset based +** on S0 is received from another site (the "remote" changeset) and +** applied to the database. The database is then in state +** (S1+"remote"), where the exact state depends on any conflict +** resolution decisions (OMIT or REPLACE) made while applying "remote". +** Rebasing a changeset is to update it to take those conflict +** resolution decisions into account, so that the same conflicts +** do not have to be resolved elsewhere in the network. +** +** For example, if both the local and remote changesets contain an +** INSERT of the same key on "CREATE TABLE t1(a PRIMARY KEY, b)": +** +** local: INSERT INTO t1 VALUES(1, 'v1'); +** remote: INSERT INTO t1 VALUES(1, 'v2'); +** +** and the conflict resolution is REPLACE, then the INSERT change is +** removed from the local changeset (it was overridden). Or, if the +** conflict resolution was "OMIT", then the local changeset is modified +** to instead contain: +** +** UPDATE t1 SET b = 'v2' WHERE a=1; +** +** Changes within the local changeset are rebased as follows: +** +**
+**
Local INSERT
+** This may only conflict with a remote INSERT. If the conflict +** resolution was OMIT, then add an UPDATE change to the rebased +** changeset. Or, if the conflict resolution was REPLACE, add +** nothing to the rebased changeset. +** +**
Local DELETE
+** This may conflict with a remote UPDATE or DELETE. In both cases the +** only possible resolution is OMIT. If the remote operation was a +** DELETE, then add no change to the rebased changeset. If the remote +** operation was an UPDATE, then the old.* fields of change are updated +** to reflect the new.* values in the UPDATE. +** +**
Local UPDATE
+** This may conflict with a remote UPDATE or DELETE. If it conflicts +** with a DELETE, and the conflict resolution was OMIT, then the update +** is changed into an INSERT. Any undefined values in the new.* record +** from the update change are filled in using the old.* values from +** the conflicting DELETE. Or, if the conflict resolution was REPLACE, +** the UPDATE change is simply omitted from the rebased changeset. +** +** If conflict is with a remote UPDATE and the resolution is OMIT, then +** the old.* values are rebased using the new.* values in the remote +** change. Or, if the resolution is REPLACE, then the change is copied +** into the rebased changeset with updates to columns also updated by +** the conflicting remote UPDATE removed. If this means no columns would +** be updated, the change is omitted. +**
+** +** A local change may be rebased against multiple remote changes +** simultaneously. If a single key is modified by multiple remote +** changesets, they are combined as follows before the local changeset +** is rebased: +** +**
    +**
  • If there has been one or more REPLACE resolutions on a +** key, it is rebased according to a REPLACE. +** +**
  • If there have been no REPLACE resolutions on a key, then +** the local changeset is rebased according to the most recent +** of the OMIT resolutions. +**
+** +** Note that conflict resolutions from multiple remote changesets are +** combined on a per-field basis, not per-row. This means that in the +** case of multiple remote UPDATE operations, some fields of a single +** local change may be rebased for REPLACE while others are rebased for +** OMIT. +** +** In order to rebase a local changeset, the remote changeset must first +** be applied to the local database using sqlite3changeset_apply_v2() and +** the buffer of rebase information captured. Then: +** +**
    +**
  1. An sqlite3_rebaser object is created by calling +** sqlite3rebaser_create(). +**
  2. The new object is configured with the rebase buffer obtained from +** sqlite3changeset_apply_v2() by calling sqlite3rebaser_configure(). +** If the local changeset is to be rebased against multiple remote +** changesets, then sqlite3rebaser_configure() should be called +** multiple times, in the same order that the multiple +** sqlite3changeset_apply_v2() calls were made. +**
  3. Each local changeset is rebased by calling sqlite3rebaser_rebase(). +**
  4. The sqlite3_rebaser object is deleted by calling +** sqlite3rebaser_delete(). +**
+*/ +typedef struct sqlite3_rebaser sqlite3_rebaser; + +/* +** CAPI3REF: Create a changeset rebaser object. +** EXPERIMENTAL +** +** Allocate a new changeset rebaser object. If successful, set (*ppNew) to +** point to the new object and return SQLITE_OK. Otherwise, if an error +** occurs, return an SQLite error code (e.g. SQLITE_NOMEM) and set (*ppNew) +** to NULL. +*/ +SQLITE_API int sqlite3rebaser_create(sqlite3_rebaser **ppNew); + +/* +** CAPI3REF: Configure a changeset rebaser object. +** EXPERIMENTAL +** +** Configure the changeset rebaser object to rebase changesets according +** to the conflict resolutions described by buffer pRebase (size nRebase +** bytes), which must have been obtained from a previous call to +** sqlite3changeset_apply_v2(). +*/ +SQLITE_API int sqlite3rebaser_configure( + sqlite3_rebaser*, + int nRebase, const void *pRebase +); + +/* +** CAPI3REF: Rebase a changeset +** EXPERIMENTAL +** +** Argument pIn must point to a buffer containing a changeset nIn bytes +** in size. This function allocates and populates a buffer with a copy +** of the changeset rebased rebased according to the configuration of the +** rebaser object passed as the first argument. If successful, (*ppOut) +** is set to point to the new buffer containing the rebased changeset and +** (*pnOut) to its size in bytes and SQLITE_OK returned. It is the +** responsibility of the caller to eventually free the new buffer using +** sqlite3_free(). Otherwise, if an error occurs, (*ppOut) and (*pnOut) +** are set to zero and an SQLite error code returned. +*/ +SQLITE_API int sqlite3rebaser_rebase( + sqlite3_rebaser*, + int nIn, const void *pIn, + int *pnOut, void **ppOut +); + +/* +** CAPI3REF: Delete a changeset rebaser object. +** EXPERIMENTAL +** +** Delete the changeset rebaser object and all associated resources. There +** should be one call to this function for each successful invocation +** of sqlite3rebaser_create(). +*/ +SQLITE_API void sqlite3rebaser_delete(sqlite3_rebaser *p); + +/* +** CAPI3REF: Streaming Versions of API functions. +** +** The six streaming API xxx_strm() functions serve similar purposes to the +** corresponding non-streaming API functions: +** +** +** +**
Streaming functionNon-streaming equivalent
sqlite3changeset_apply_strm[sqlite3changeset_apply] +**
sqlite3changeset_apply_strm_v2[sqlite3changeset_apply_v2] +**
sqlite3changeset_concat_strm[sqlite3changeset_concat] +**
sqlite3changeset_invert_strm[sqlite3changeset_invert] +**
sqlite3changeset_start_strm[sqlite3changeset_start] +**
sqlite3session_changeset_strm[sqlite3session_changeset] +**
sqlite3session_patchset_strm[sqlite3session_patchset] +**
+** +** Non-streaming functions that accept changesets (or patchsets) as input +** require that the entire changeset be stored in a single buffer in memory. +** Similarly, those that return a changeset or patchset do so by returning +** a pointer to a single large buffer allocated using sqlite3_malloc(). +** Normally this is convenient. However, if an application running in a +** low-memory environment is required to handle very large changesets, the +** large contiguous memory allocations required can become onerous. +** +** In order to avoid this problem, instead of a single large buffer, input +** is passed to a streaming API functions by way of a callback function that +** the sessions module invokes to incrementally request input data as it is +** required. In all cases, a pair of API function parameters such as +** +** 
+**        int nChangeset,
+**        void *pChangeset,
+**
+** +** Is replaced by: +** +** 
+**        int (*xInput)(void *pIn, void *pData, int *pnData),
+**        void *pIn,
+**
+** +** Each time the xInput callback is invoked by the sessions module, the first +** argument passed is a copy of the supplied pIn context pointer. The second +** argument, pData, points to a buffer (*pnData) bytes in size. Assuming no +** error occurs the xInput method should copy up to (*pnData) bytes of data +** into the buffer and set (*pnData) to the actual number of bytes copied +** before returning SQLITE_OK. If the input is completely exhausted, (*pnData) +** should be set to zero to indicate this. Or, if an error occurs, an SQLite +** error code should be returned. In all cases, if an xInput callback returns +** an error, all processing is abandoned and the streaming API function +** returns a copy of the error code to the caller. +** +** In the case of sqlite3changeset_start_strm(), the xInput callback may be +** invoked by the sessions module at any point during the lifetime of the +** iterator. If such an xInput callback returns an error, the iterator enters +** an error state, whereby all subsequent calls to iterator functions +** immediately fail with the same error code as returned by xInput. +** +** Similarly, streaming API functions that return changesets (or patchsets) +** return them in chunks by way of a callback function instead of via a +** pointer to a single large buffer. In this case, a pair of parameters such +** as: +** +** 
+**        int *pnChangeset,
+**        void **ppChangeset,
+**
+** +** Is replaced by: +** +** 
+**        int (*xOutput)(void *pOut, const void *pData, int nData),
+**        void *pOut
+**
+** +** The xOutput callback is invoked zero or more times to return data to +** the application. The first parameter passed to each call is a copy of the +** pOut pointer supplied by the application. The second parameter, pData, +** points to a buffer nData bytes in size containing the chunk of output +** data being returned. If the xOutput callback successfully processes the +** supplied data, it should return SQLITE_OK to indicate success. Otherwise, +** it should return some other SQLite error code. In this case processing +** is immediately abandoned and the streaming API function returns a copy +** of the xOutput error code to the application. +** +** The sessions module never invokes an xOutput callback with the third +** parameter set to a value less than or equal to zero. Other than this, +** no guarantees are made as to the size of the chunks of data returned. +*/ +SQLITE_API int sqlite3changeset_apply_strm( + sqlite3 *db, /* Apply change to "main" db of this handle */ + int (*xInput)(void *pIn, void *pData, int *pnData), /* Input function */ + void *pIn, /* First arg for xInput */ + int(*xFilter)( + void *pCtx, /* Copy of sixth arg to _apply() */ + const char *zTab /* Table name */ + ), + int(*xConflict)( + void *pCtx, /* Copy of sixth arg to _apply() */ + int eConflict, /* DATA, MISSING, CONFLICT, CONSTRAINT */ + sqlite3_changeset_iter *p /* Handle describing change and conflict */ + ), + void *pCtx /* First argument passed to xConflict */ +); +SQLITE_API int sqlite3changeset_apply_v2_strm( + sqlite3 *db, /* Apply change to "main" db of this handle */ + int (*xInput)(void *pIn, void *pData, int *pnData), /* Input function */ + void *pIn, /* First arg for xInput */ + int(*xFilter)( + void *pCtx, /* Copy of sixth arg to _apply() */ + const char *zTab /* Table name */ + ), + int(*xConflict)( + void *pCtx, /* Copy of sixth arg to _apply() */ + int eConflict, /* DATA, MISSING, CONFLICT, CONSTRAINT */ + sqlite3_changeset_iter *p /* Handle describing change and conflict */ + ), + void *pCtx, /* First argument passed to xConflict */ + void **ppRebase, int *pnRebase, + int flags +); +SQLITE_API int sqlite3changeset_concat_strm( + int (*xInputA)(void *pIn, void *pData, int *pnData), + void *pInA, + int (*xInputB)(void *pIn, void *pData, int *pnData), + void *pInB, + int (*xOutput)(void *pOut, const void *pData, int nData), + void *pOut +); +SQLITE_API int sqlite3changeset_invert_strm( + int (*xInput)(void *pIn, void *pData, int *pnData), + void *pIn, + int (*xOutput)(void *pOut, const void *pData, int nData), + void *pOut +); +SQLITE_API int sqlite3changeset_start_strm( + sqlite3_changeset_iter **pp, + int (*xInput)(void *pIn, void *pData, int *pnData), + void *pIn +); +SQLITE_API int sqlite3changeset_start_v2_strm( + sqlite3_changeset_iter **pp, + int (*xInput)(void *pIn, void *pData, int *pnData), + void *pIn, + int flags +); +SQLITE_API int sqlite3session_changeset_strm( + sqlite3_session *pSession, + int (*xOutput)(void *pOut, const void *pData, int nData), + void *pOut +); +SQLITE_API int sqlite3session_patchset_strm( + sqlite3_session *pSession, + int (*xOutput)(void *pOut, const void *pData, int nData), + void *pOut +); +SQLITE_API int sqlite3changegroup_add_strm(sqlite3_changegroup*, + int (*xInput)(void *pIn, void *pData, int *pnData), + void *pIn +); +SQLITE_API int sqlite3changegroup_output_strm(sqlite3_changegroup*, + int (*xOutput)(void *pOut, const void *pData, int nData), + void *pOut +); +SQLITE_API int sqlite3rebaser_rebase_strm( + sqlite3_rebaser *pRebaser, + int (*xInput)(void *pIn, void *pData, int *pnData), + void *pIn, + int (*xOutput)(void *pOut, const void *pData, int nData), + void *pOut +); + +/* +** CAPI3REF: Configure global parameters +** +** The sqlite3session_config() interface is used to make global configuration +** changes to the sessions module in order to tune it to the specific needs +** of the application. +** +** The sqlite3session_config() interface is not threadsafe. If it is invoked +** while any other thread is inside any other sessions method then the +** results are undefined. Furthermore, if it is invoked after any sessions +** related objects have been created, the results are also undefined. +** +** The first argument to the sqlite3session_config() function must be one +** of the SQLITE_SESSION_CONFIG_XXX constants defined below. The +** interpretation of the (void*) value passed as the second parameter and +** the effect of calling this function depends on the value of the first +** parameter. +** +**
+**
SQLITE_SESSION_CONFIG_STRMSIZE
+** By default, the sessions module streaming interfaces attempt to input +** and output data in approximately 1 KiB chunks. This operand may be used +** to set and query the value of this configuration setting. The pointer +** passed as the second argument must point to a value of type (int). +** If this value is greater than 0, it is used as the new streaming data +** chunk size for both input and output. Before returning, the (int) value +** pointed to by pArg is set to the final value of the streaming interface +** chunk size. +**
+** +** This function returns SQLITE_OK if successful, or an SQLite error code +** otherwise. +*/ +SQLITE_API int sqlite3session_config(int op, void *pArg); + +/* +** CAPI3REF: Values for sqlite3session_config(). +*/ +#define SQLITE_SESSION_CONFIG_STRMSIZE 1 + +/* +** Make sure we can call this stuff from C++. +*/ +#if 0 +} +#endif + +#endif /* !defined(__SQLITESESSION_H_) && defined(SQLITE_ENABLE_SESSION) */ + +/******** End of sqlite3session.h *********/ +/******** Begin file fts5.h *********/ +/* +** 2014 May 31 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +****************************************************************************** +** +** Interfaces to extend FTS5. Using the interfaces defined in this file, +** FTS5 may be extended with: +** +** * custom tokenizers, and +** * custom auxiliary functions. +*/ + + +#ifndef _FTS5_H +#define _FTS5_H + + +#if 0 +extern "C" { +#endif + +/************************************************************************* +** CUSTOM AUXILIARY FUNCTIONS +** +** Virtual table implementations may overload SQL functions by implementing +** the sqlite3_module.xFindFunction() method. +*/ + +typedef struct Fts5ExtensionApi Fts5ExtensionApi; +typedef struct Fts5Context Fts5Context; +typedef struct Fts5PhraseIter Fts5PhraseIter; + +typedef void (*fts5_extension_function)( + const Fts5ExtensionApi *pApi, /* API offered by current FTS version */ + Fts5Context *pFts, /* First arg to pass to pApi functions */ + sqlite3_context *pCtx, /* Context for returning result/error */ + int nVal, /* Number of values in apVal[] array */ + sqlite3_value **apVal /* Array of trailing arguments */ +); + +struct Fts5PhraseIter { + const unsigned char *a; + const unsigned char *b; +}; + +/* +** EXTENSION API FUNCTIONS +** +** xUserData(pFts): +** Return a copy of the context pointer the extension function was +** registered with. +** +** xColumnTotalSize(pFts, iCol, pnToken): +** If parameter iCol is less than zero, set output variable *pnToken +** to the total number of tokens in the FTS5 table. Or, if iCol is +** non-negative but less than the number of columns in the table, return +** the total number of tokens in column iCol, considering all rows in +** the FTS5 table. +** +** If parameter iCol is greater than or equal to the number of columns +** in the table, SQLITE_RANGE is returned. Or, if an error occurs (e.g. +** an OOM condition or IO error), an appropriate SQLite error code is +** returned. +** +** xColumnCount(pFts): +** Return the number of columns in the table. +** +** xColumnSize(pFts, iCol, pnToken): +** If parameter iCol is less than zero, set output variable *pnToken +** to the total number of tokens in the current row. Or, if iCol is +** non-negative but less than the number of columns in the table, set +** *pnToken to the number of tokens in column iCol of the current row. +** +** If parameter iCol is greater than or equal to the number of columns +** in the table, SQLITE_RANGE is returned. Or, if an error occurs (e.g. +** an OOM condition or IO error), an appropriate SQLite error code is +** returned. +** +** This function may be quite inefficient if used with an FTS5 table +** created with the "columnsize=0" option. +** +** xColumnText: +** This function attempts to retrieve the text of column iCol of the +** current document. If successful, (*pz) is set to point to a buffer +** containing the text in utf-8 encoding, (*pn) is set to the size in bytes +** (not characters) of the buffer and SQLITE_OK is returned. Otherwise, +** if an error occurs, an SQLite error code is returned and the final values +** of (*pz) and (*pn) are undefined. +** +** xPhraseCount: +** Returns the number of phrases in the current query expression. +** +** xPhraseSize: +** Returns the number of tokens in phrase iPhrase of the query. Phrases +** are numbered starting from zero. +** +** xInstCount: +** Set *pnInst to the total number of occurrences of all phrases within +** the query within the current row. Return SQLITE_OK if successful, or +** an error code (i.e. SQLITE_NOMEM) if an error occurs. +** +** This API can be quite slow if used with an FTS5 table created with the +** "detail=none" or "detail=column" option. If the FTS5 table is created +** with either "detail=none" or "detail=column" and "content=" option +** (i.e. if it is a contentless table), then this API always returns 0. +** +** xInst: +** Query for the details of phrase match iIdx within the current row. +** Phrase matches are numbered starting from zero, so the iIdx argument +** should be greater than or equal to zero and smaller than the value +** output by xInstCount(). +** +** Usually, output parameter *piPhrase is set to the phrase number, *piCol +** to the column in which it occurs and *piOff the token offset of the +** first token of the phrase. Returns SQLITE_OK if successful, or an error +** code (i.e. SQLITE_NOMEM) if an error occurs. +** +** This API can be quite slow if used with an FTS5 table created with the +** "detail=none" or "detail=column" option. +** +** xRowid: +** Returns the rowid of the current row. +** +** xTokenize: +** Tokenize text using the tokenizer belonging to the FTS5 table. +** +** xQueryPhrase(pFts5, iPhrase, pUserData, xCallback): +** This API function is used to query the FTS table for phrase iPhrase +** of the current query. Specifically, a query equivalent to: +** +** ... FROM ftstable WHERE ftstable MATCH $p ORDER BY rowid +** +** with $p set to a phrase equivalent to the phrase iPhrase of the +** current query is executed. Any column filter that applies to +** phrase iPhrase of the current query is included in $p. For each +** row visited, the callback function passed as the fourth argument +** is invoked. The context and API objects passed to the callback +** function may be used to access the properties of each matched row. +** Invoking Api.xUserData() returns a copy of the pointer passed as +** the third argument to pUserData. +** +** If the callback function returns any value other than SQLITE_OK, the +** query is abandoned and the xQueryPhrase function returns immediately. +** If the returned value is SQLITE_DONE, xQueryPhrase returns SQLITE_OK. +** Otherwise, the error code is propagated upwards. +** +** If the query runs to completion without incident, SQLITE_OK is returned. +** Or, if some error occurs before the query completes or is aborted by +** the callback, an SQLite error code is returned. +** +** +** xSetAuxdata(pFts5, pAux, xDelete) +** +** Save the pointer passed as the second argument as the extension functions +** "auxiliary data". The pointer may then be retrieved by the current or any +** future invocation of the same fts5 extension function made as part of +** the same MATCH query using the xGetAuxdata() API. +** +** Each extension function is allocated a single auxiliary data slot for +** each FTS query (MATCH expression). If the extension function is invoked +** more than once for a single FTS query, then all invocations share a +** single auxiliary data context. +** +** If there is already an auxiliary data pointer when this function is +** invoked, then it is replaced by the new pointer. If an xDelete callback +** was specified along with the original pointer, it is invoked at this +** point. +** +** The xDelete callback, if one is specified, is also invoked on the +** auxiliary data pointer after the FTS5 query has finished. +** +** If an error (e.g. an OOM condition) occurs within this function, +** the auxiliary data is set to NULL and an error code returned. If the +** xDelete parameter was not NULL, it is invoked on the auxiliary data +** pointer before returning. +** +** +** xGetAuxdata(pFts5, bClear) +** +** Returns the current auxiliary data pointer for the fts5 extension +** function. See the xSetAuxdata() method for details. +** +** If the bClear argument is non-zero, then the auxiliary data is cleared +** (set to NULL) before this function returns. In this case the xDelete, +** if any, is not invoked. +** +** +** xRowCount(pFts5, pnRow) +** +** This function is used to retrieve the total number of rows in the table. +** In other words, the same value that would be returned by: +** +** SELECT count(*) FROM ftstable; +** +** xPhraseFirst() +** This function is used, along with type Fts5PhraseIter and the xPhraseNext +** method, to iterate through all instances of a single query phrase within +** the current row. This is the same information as is accessible via the +** xInstCount/xInst APIs. While the xInstCount/xInst APIs are more convenient +** to use, this API may be faster under some circumstances. To iterate +** through instances of phrase iPhrase, use the following code: +** +** Fts5PhraseIter iter; +** int iCol, iOff; +** for(pApi->xPhraseFirst(pFts, iPhrase, &iter, &iCol, &iOff); +** iCol>=0; +** pApi->xPhraseNext(pFts, &iter, &iCol, &iOff) +** ){ +** // An instance of phrase iPhrase at offset iOff of column iCol +** } +** +** The Fts5PhraseIter structure is defined above. Applications should not +** modify this structure directly - it should only be used as shown above +** with the xPhraseFirst() and xPhraseNext() API methods (and by +** xPhraseFirstColumn() and xPhraseNextColumn() as illustrated below). +** +** This API can be quite slow if used with an FTS5 table created with the +** "detail=none" or "detail=column" option. If the FTS5 table is created +** with either "detail=none" or "detail=column" and "content=" option +** (i.e. if it is a contentless table), then this API always iterates +** through an empty set (all calls to xPhraseFirst() set iCol to -1). +** +** xPhraseNext() +** See xPhraseFirst above. +** +** xPhraseFirstColumn() +** This function and xPhraseNextColumn() are similar to the xPhraseFirst() +** and xPhraseNext() APIs described above. The difference is that instead +** of iterating through all instances of a phrase in the current row, these +** APIs are used to iterate through the set of columns in the current row +** that contain one or more instances of a specified phrase. For example: +** +** Fts5PhraseIter iter; +** int iCol; +** for(pApi->xPhraseFirstColumn(pFts, iPhrase, &iter, &iCol); +** iCol>=0; +** pApi->xPhraseNextColumn(pFts, &iter, &iCol) +** ){ +** // Column iCol contains at least one instance of phrase iPhrase +** } +** +** This API can be quite slow if used with an FTS5 table created with the +** "detail=none" option. If the FTS5 table is created with either +** "detail=none" "content=" option (i.e. if it is a contentless table), +** then this API always iterates through an empty set (all calls to +** xPhraseFirstColumn() set iCol to -1). +** +** The information accessed using this API and its companion +** xPhraseFirstColumn() may also be obtained using xPhraseFirst/xPhraseNext +** (or xInst/xInstCount). The chief advantage of this API is that it is +** significantly more efficient than those alternatives when used with +** "detail=column" tables. +** +** xPhraseNextColumn() +** See xPhraseFirstColumn above. +*/ +struct Fts5ExtensionApi { + int iVersion; /* Currently always set to 3 */ + + void *(*xUserData)(Fts5Context*); + + int (*xColumnCount)(Fts5Context*); + int (*xRowCount)(Fts5Context*, sqlite3_int64 *pnRow); + int (*xColumnTotalSize)(Fts5Context*, int iCol, sqlite3_int64 *pnToken); + + int (*xTokenize)(Fts5Context*, + const char *pText, int nText, /* Text to tokenize */ + void *pCtx, /* Context passed to xToken() */ + int (*xToken)(void*, int, const char*, int, int, int) /* Callback */ + ); + + int (*xPhraseCount)(Fts5Context*); + int (*xPhraseSize)(Fts5Context*, int iPhrase); + + int (*xInstCount)(Fts5Context*, int *pnInst); + int (*xInst)(Fts5Context*, int iIdx, int *piPhrase, int *piCol, int *piOff); + + sqlite3_int64 (*xRowid)(Fts5Context*); + int (*xColumnText)(Fts5Context*, int iCol, const char **pz, int *pn); + int (*xColumnSize)(Fts5Context*, int iCol, int *pnToken); + + int (*xQueryPhrase)(Fts5Context*, int iPhrase, void *pUserData, + int(*)(const Fts5ExtensionApi*,Fts5Context*,void*) + ); + int (*xSetAuxdata)(Fts5Context*, void *pAux, void(*xDelete)(void*)); + void *(*xGetAuxdata)(Fts5Context*, int bClear); + + int (*xPhraseFirst)(Fts5Context*, int iPhrase, Fts5PhraseIter*, int*, int*); + void (*xPhraseNext)(Fts5Context*, Fts5PhraseIter*, int *piCol, int *piOff); + + int (*xPhraseFirstColumn)(Fts5Context*, int iPhrase, Fts5PhraseIter*, int*); + void (*xPhraseNextColumn)(Fts5Context*, Fts5PhraseIter*, int *piCol); +}; + +/* +** CUSTOM AUXILIARY FUNCTIONS +*************************************************************************/ + +/************************************************************************* +** CUSTOM TOKENIZERS +** +** Applications may also register custom tokenizer types. A tokenizer +** is registered by providing fts5 with a populated instance of the +** following structure. All structure methods must be defined, setting +** any member of the fts5_tokenizer struct to NULL leads to undefined +** behaviour. The structure methods are expected to function as follows: +** +** xCreate: +** This function is used to allocate and initialize a tokenizer instance. +** A tokenizer instance is required to actually tokenize text. +** +** The first argument passed to this function is a copy of the (void*) +** pointer provided by the application when the fts5_tokenizer object +** was registered with FTS5 (the third argument to xCreateTokenizer()). +** The second and third arguments are an array of nul-terminated strings +** containing the tokenizer arguments, if any, specified following the +** tokenizer name as part of the CREATE VIRTUAL TABLE statement used +** to create the FTS5 table. +** +** The final argument is an output variable. If successful, (*ppOut) +** should be set to point to the new tokenizer handle and SQLITE_OK +** returned. If an error occurs, some value other than SQLITE_OK should +** be returned. In this case, fts5 assumes that the final value of *ppOut +** is undefined. +** +** xDelete: +** This function is invoked to delete a tokenizer handle previously +** allocated using xCreate(). Fts5 guarantees that this function will +** be invoked exactly once for each successful call to xCreate(). +** +** xTokenize: +** This function is expected to tokenize the nText byte string indicated +** by argument pText. pText may or may not be nul-terminated. The first +** argument passed to this function is a pointer to an Fts5Tokenizer object +** returned by an earlier call to xCreate(). +** +** The second argument indicates the reason that FTS5 is requesting +** tokenization of the supplied text. This is always one of the following +** four values: +** +**
  • FTS5_TOKENIZE_DOCUMENT - A document is being inserted into +** or removed from the FTS table. The tokenizer is being invoked to +** determine the set of tokens to add to (or delete from) the +** FTS index. +** +**
  • FTS5_TOKENIZE_QUERY - A MATCH query is being executed +** against the FTS index. The tokenizer is being called to tokenize +** a bareword or quoted string specified as part of the query. +** +**
  • (FTS5_TOKENIZE_QUERY | FTS5_TOKENIZE_PREFIX) - Same as +** FTS5_TOKENIZE_QUERY, except that the bareword or quoted string is +** followed by a "*" character, indicating that the last token +** returned by the tokenizer will be treated as a token prefix. +** +**
  • FTS5_TOKENIZE_AUX - The tokenizer is being invoked to +** satisfy an fts5_api.xTokenize() request made by an auxiliary +** function. Or an fts5_api.xColumnSize() request made by the same +** on a columnsize=0 database. +**
+** +** For each token in the input string, the supplied callback xToken() must +** be invoked. The first argument to it should be a copy of the pointer +** passed as the second argument to xTokenize(). The third and fourth +** arguments are a pointer to a buffer containing the token text, and the +** size of the token in bytes. The 4th and 5th arguments are the byte offsets +** of the first byte of and first byte immediately following the text from +** which the token is derived within the input. +** +** The second argument passed to the xToken() callback ("tflags") should +** normally be set to 0. The exception is if the tokenizer supports +** synonyms. In this case see the discussion below for details. +** +** FTS5 assumes the xToken() callback is invoked for each token in the +** order that they occur within the input text. +** +** If an xToken() callback returns any value other than SQLITE_OK, then +** the tokenization should be abandoned and the xTokenize() method should +** immediately return a copy of the xToken() return value. Or, if the +** input buffer is exhausted, xTokenize() should return SQLITE_OK. Finally, +** if an error occurs with the xTokenize() implementation itself, it +** may abandon the tokenization and return any error code other than +** SQLITE_OK or SQLITE_DONE. +** +** SYNONYM SUPPORT +** +** Custom tokenizers may also support synonyms. Consider a case in which a +** user wishes to query for a phrase such as "first place". Using the +** built-in tokenizers, the FTS5 query 'first + place' will match instances +** of "first place" within the document set, but not alternative forms +** such as "1st place". In some applications, it would be better to match +** all instances of "first place" or "1st place" regardless of which form +** the user specified in the MATCH query text. +** +** There are several ways to approach this in FTS5: +** +**
  1. By mapping all synonyms to a single token. In this case, the +** In the above example, this means that the tokenizer returns the +** same token for inputs "first" and "1st". Say that token is in +** fact "first", so that when the user inserts the document "I won +** 1st place" entries are added to the index for tokens "i", "won", +** "first" and "place". If the user then queries for '1st + place', +** the tokenizer substitutes "first" for "1st" and the query works +** as expected. +** +**
  2. By querying the index for all synonyms of each query term +** separately. In this case, when tokenizing query text, the +** tokenizer may provide multiple synonyms for a single term +** within the document. FTS5 then queries the index for each +** synonym individually. For example, faced with the query: +** +** +** ... MATCH 'first place' +** +** the tokenizer offers both "1st" and "first" as synonyms for the +** first token in the MATCH query and FTS5 effectively runs a query +** similar to: +** +** +** ... MATCH '(first OR 1st) place' +** +** except that, for the purposes of auxiliary functions, the query +** still appears to contain just two phrases - "(first OR 1st)" +** being treated as a single phrase. +** +**
  3. By adding multiple synonyms for a single term to the FTS index. +** Using this method, when tokenizing document text, the tokenizer +** provides multiple synonyms for each token. So that when a +** document such as "I won first place" is tokenized, entries are +** added to the FTS index for "i", "won", "first", "1st" and +** "place". +** +** This way, even if the tokenizer does not provide synonyms +** when tokenizing query text (it should not - to do so would be +** inefficient), it doesn't matter if the user queries for +** 'first + place' or '1st + place', as there are entries in the +** FTS index corresponding to both forms of the first token. +**
+** +** Whether it is parsing document or query text, any call to xToken that +** specifies a tflags argument with the FTS5_TOKEN_COLOCATED bit +** is considered to supply a synonym for the previous token. For example, +** when parsing the document "I won first place", a tokenizer that supports +** synonyms would call xToken() 5 times, as follows: +** +** +** xToken(pCtx, 0, "i", 1, 0, 1); +** xToken(pCtx, 0, "won", 3, 2, 5); +** xToken(pCtx, 0, "first", 5, 6, 11); +** xToken(pCtx, FTS5_TOKEN_COLOCATED, "1st", 3, 6, 11); +** xToken(pCtx, 0, "place", 5, 12, 17); +** +** +** It is an error to specify the FTS5_TOKEN_COLOCATED flag the first time +** xToken() is called. Multiple synonyms may be specified for a single token +** by making multiple calls to xToken(FTS5_TOKEN_COLOCATED) in sequence. +** There is no limit to the number of synonyms that may be provided for a +** single token. +** +** In many cases, method (1) above is the best approach. It does not add +** extra data to the FTS index or require FTS5 to query for multiple terms, +** so it is efficient in terms of disk space and query speed. However, it +** does not support prefix queries very well. If, as suggested above, the +** token "first" is substituted for "1st" by the tokenizer, then the query: +** +** +** ... MATCH '1s*' +** +** will not match documents that contain the token "1st" (as the tokenizer +** will probably not map "1s" to any prefix of "first"). +** +** For full prefix support, method (3) may be preferred. In this case, +** because the index contains entries for both "first" and "1st", prefix +** queries such as 'fi*' or '1s*' will match correctly. However, because +** extra entries are added to the FTS index, this method uses more space +** within the database. +** +** Method (2) offers a midpoint between (1) and (3). Using this method, +** a query such as '1s*' will match documents that contain the literal +** token "1st", but not "first" (assuming the tokenizer is not able to +** provide synonyms for prefixes). However, a non-prefix query like '1st' +** will match against "1st" and "first". This method does not require +** extra disk space, as no extra entries are added to the FTS index. +** On the other hand, it may require more CPU cycles to run MATCH queries, +** as separate queries of the FTS index are required for each synonym. +** +** When using methods (2) or (3), it is important that the tokenizer only +** provide synonyms when tokenizing document text (method (2)) or query +** text (method (3)), not both. Doing so will not cause any errors, but is +** inefficient. +*/ +typedef struct Fts5Tokenizer Fts5Tokenizer; +typedef struct fts5_tokenizer fts5_tokenizer; +struct fts5_tokenizer { + int (*xCreate)(void*, const char **azArg, int nArg, Fts5Tokenizer **ppOut); + void (*xDelete)(Fts5Tokenizer*); + int (*xTokenize)(Fts5Tokenizer*, + void *pCtx, + int flags, /* Mask of FTS5_TOKENIZE_* flags */ + const char *pText, int nText, + int (*xToken)( + void *pCtx, /* Copy of 2nd argument to xTokenize() */ + int tflags, /* Mask of FTS5_TOKEN_* flags */ + const char *pToken, /* Pointer to buffer containing token */ + int nToken, /* Size of token in bytes */ + int iStart, /* Byte offset of token within input text */ + int iEnd /* Byte offset of end of token within input text */ + ) + ); +}; + +/* Flags that may be passed as the third argument to xTokenize() */ +#define FTS5_TOKENIZE_QUERY 0x0001 +#define FTS5_TOKENIZE_PREFIX 0x0002 +#define FTS5_TOKENIZE_DOCUMENT 0x0004 +#define FTS5_TOKENIZE_AUX 0x0008 + +/* Flags that may be passed by the tokenizer implementation back to FTS5 +** as the third argument to the supplied xToken callback. */ +#define FTS5_TOKEN_COLOCATED 0x0001 /* Same position as prev. token */ + +/* +** END OF CUSTOM TOKENIZERS +*************************************************************************/ + +/************************************************************************* +** FTS5 EXTENSION REGISTRATION API +*/ +typedef struct fts5_api fts5_api; +struct fts5_api { + int iVersion; /* Currently always set to 2 */ + + /* Create a new tokenizer */ + int (*xCreateTokenizer)( + fts5_api *pApi, + const char *zName, + void *pContext, + fts5_tokenizer *pTokenizer, + void (*xDestroy)(void*) + ); + + /* Find an existing tokenizer */ + int (*xFindTokenizer)( + fts5_api *pApi, + const char *zName, + void **ppContext, + fts5_tokenizer *pTokenizer + ); + + /* Create a new auxiliary function */ + int (*xCreateFunction)( + fts5_api *pApi, + const char *zName, + void *pContext, + fts5_extension_function xFunction, + void (*xDestroy)(void*) + ); +}; + +/* +** END OF REGISTRATION API +*************************************************************************/ + +#if 0 +} /* end of the 'extern "C"' block */ +#endif + +#endif /* _FTS5_H */ + +/******** End of fts5.h *********/ + +/************** End of sqlite3.h *********************************************/ +/************** Continuing where we left off in sqliteInt.h ******************/ + +/* +** Include the configuration header output by 'configure' if we're using the +** autoconf-based build +*/ +#if defined(_HAVE_SQLITE_CONFIG_H) && !defined(SQLITECONFIG_H) +/* #include "config.h" */ +#define SQLITECONFIG_H 1 +#endif + +/************** Include sqliteLimit.h in the middle of sqliteInt.h ***********/ +/************** Begin file sqliteLimit.h *************************************/ +/* +** 2007 May 7 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** +** This file defines various limits of what SQLite can process. +*/ + +/* +** The maximum length of a TEXT or BLOB in bytes. This also +** limits the size of a row in a table or index. +** +** The hard limit is the ability of a 32-bit signed integer +** to count the size: 2^31-1 or 2147483647. +*/ +#ifndef SQLITE_MAX_LENGTH +# define SQLITE_MAX_LENGTH 1000000000 +#endif + +/* +** This is the maximum number of +** +** * Columns in a table +** * Columns in an index +** * Columns in a view +** * Terms in the SET clause of an UPDATE statement +** * Terms in the result set of a SELECT statement +** * Terms in the GROUP BY or ORDER BY clauses of a SELECT statement. +** * Terms in the VALUES clause of an INSERT statement +** +** The hard upper limit here is 32676. Most database people will +** tell you that in a well-normalized database, you usually should +** not have more than a dozen or so columns in any table. And if +** that is the case, there is no point in having more than a few +** dozen values in any of the other situations described above. +*/ +#ifndef SQLITE_MAX_COLUMN +# define SQLITE_MAX_COLUMN 2000 +#endif + +/* +** The maximum length of a single SQL statement in bytes. +** +** It used to be the case that setting this value to zero would +** turn the limit off. That is no longer true. It is not possible +** to turn this limit off. +*/ +#ifndef SQLITE_MAX_SQL_LENGTH +# define SQLITE_MAX_SQL_LENGTH 1000000000 +#endif + +/* +** The maximum depth of an expression tree. This is limited to +** some extent by SQLITE_MAX_SQL_LENGTH. But sometime you might +** want to place more severe limits on the complexity of an +** expression. +** +** A value of 0 used to mean that the limit was not enforced. +** But that is no longer true. The limit is now strictly enforced +** at all times. +*/ +#ifndef SQLITE_MAX_EXPR_DEPTH +# define SQLITE_MAX_EXPR_DEPTH 1000 +#endif + +/* +** The maximum number of terms in a compound SELECT statement. +** The code generator for compound SELECT statements does one +** level of recursion for each term. A stack overflow can result +** if the number of terms is too large. In practice, most SQL +** never has more than 3 or 4 terms. Use a value of 0 to disable +** any limit on the number of terms in a compount SELECT. +*/ +#ifndef SQLITE_MAX_COMPOUND_SELECT +# define SQLITE_MAX_COMPOUND_SELECT 500 +#endif + +/* +** The maximum number of opcodes in a VDBE program. +** Not currently enforced. +*/ +#ifndef SQLITE_MAX_VDBE_OP +# define SQLITE_MAX_VDBE_OP 250000000 +#endif + +/* +** The maximum number of arguments to an SQL function. +*/ +#ifndef SQLITE_MAX_FUNCTION_ARG +# define SQLITE_MAX_FUNCTION_ARG 127 +#endif + +/* +** The suggested maximum number of in-memory pages to use for +** the main database table and for temporary tables. +** +** IMPLEMENTATION-OF: R-30185-15359 The default suggested cache size is -2000, +** which means the cache size is limited to 2048000 bytes of memory. +** IMPLEMENTATION-OF: R-48205-43578 The default suggested cache size can be +** altered using the SQLITE_DEFAULT_CACHE_SIZE compile-time options. +*/ +#ifndef SQLITE_DEFAULT_CACHE_SIZE +# define SQLITE_DEFAULT_CACHE_SIZE -2000 +#endif + +/* +** The default number of frames to accumulate in the log file before +** checkpointing the database in WAL mode. +*/ +#ifndef SQLITE_DEFAULT_WAL_AUTOCHECKPOINT +# define SQLITE_DEFAULT_WAL_AUTOCHECKPOINT 1000 +#endif + +/* +** The maximum number of attached databases. This must be between 0 +** and 125. The upper bound of 125 is because the attached databases are +** counted using a signed 8-bit integer which has a maximum value of 127 +** and we have to allow 2 extra counts for the "main" and "temp" databases. +*/ +#ifndef SQLITE_MAX_ATTACHED +# define SQLITE_MAX_ATTACHED 10 +#endif + + +/* +** The maximum value of a ?nnn wildcard that the parser will accept. +*/ +#ifndef SQLITE_MAX_VARIABLE_NUMBER +# define SQLITE_MAX_VARIABLE_NUMBER 999 +#endif + +/* Maximum page size. The upper bound on this value is 65536. This a limit +** imposed by the use of 16-bit offsets within each page. +** +** Earlier versions of SQLite allowed the user to change this value at +** compile time. This is no longer permitted, on the grounds that it creates +** a library that is technically incompatible with an SQLite library +** compiled with a different limit. If a process operating on a database +** with a page-size of 65536 bytes crashes, then an instance of SQLite +** compiled with the default page-size limit will not be able to rollback +** the aborted transaction. This could lead to database corruption. +*/ +#ifdef SQLITE_MAX_PAGE_SIZE +# undef SQLITE_MAX_PAGE_SIZE +#endif +#define SQLITE_MAX_PAGE_SIZE 65536 + + +/* +** The default size of a database page. +*/ +#ifndef SQLITE_DEFAULT_PAGE_SIZE +# define SQLITE_DEFAULT_PAGE_SIZE 4096 +#endif +#if SQLITE_DEFAULT_PAGE_SIZE>SQLITE_MAX_PAGE_SIZE +# undef SQLITE_DEFAULT_PAGE_SIZE +# define SQLITE_DEFAULT_PAGE_SIZE SQLITE_MAX_PAGE_SIZE +#endif + +/* +** Ordinarily, if no value is explicitly provided, SQLite creates databases +** with page size SQLITE_DEFAULT_PAGE_SIZE. However, based on certain +** device characteristics (sector-size and atomic write() support), +** SQLite may choose a larger value. This constant is the maximum value +** SQLite will choose on its own. +*/ +#ifndef SQLITE_MAX_DEFAULT_PAGE_SIZE +# define SQLITE_MAX_DEFAULT_PAGE_SIZE 8192 +#endif +#if SQLITE_MAX_DEFAULT_PAGE_SIZE>SQLITE_MAX_PAGE_SIZE +# undef SQLITE_MAX_DEFAULT_PAGE_SIZE +# define SQLITE_MAX_DEFAULT_PAGE_SIZE SQLITE_MAX_PAGE_SIZE +#endif + + +/* +** Maximum number of pages in one database file. +** +** This is really just the default value for the max_page_count pragma. +** This value can be lowered (or raised) at run-time using that the +** max_page_count macro. +*/ +#ifndef SQLITE_MAX_PAGE_COUNT +# define SQLITE_MAX_PAGE_COUNT 1073741823 +#endif + +/* +** Maximum length (in bytes) of the pattern in a LIKE or GLOB +** operator. +*/ +#ifndef SQLITE_MAX_LIKE_PATTERN_LENGTH +# define SQLITE_MAX_LIKE_PATTERN_LENGTH 50000 +#endif + +/* +** Maximum depth of recursion for triggers. +** +** A value of 1 means that a trigger program will not be able to itself +** fire any triggers. A value of 0 means that no trigger programs at all +** may be executed. +*/ +#ifndef SQLITE_MAX_TRIGGER_DEPTH +# define SQLITE_MAX_TRIGGER_DEPTH 1000 +#endif + +/************** End of sqliteLimit.h *****************************************/ +/************** Continuing where we left off in sqliteInt.h ******************/ + +/* Disable nuisance warnings on Borland compilers */ +#if defined(__BORLANDC__) +#pragma warn -rch /* unreachable code */ +#pragma warn -ccc /* Condition is always true or false */ +#pragma warn -aus /* Assigned value is never used */ +#pragma warn -csu /* Comparing signed and unsigned */ +#pragma warn -spa /* Suspicious pointer arithmetic */ +#endif + +/* +** Include standard header files as necessary +*/ +#ifdef HAVE_STDINT_H +#include +#endif +#ifdef HAVE_INTTYPES_H +#include +#endif + +/* +** The following macros are used to cast pointers to integers and +** integers to pointers. The way you do this varies from one compiler +** to the next, so we have developed the following set of #if statements +** to generate appropriate macros for a wide range of compilers. +** +** The correct "ANSI" way to do this is to use the intptr_t type. +** Unfortunately, that typedef is not available on all compilers, or +** if it is available, it requires an #include of specific headers +** that vary from one machine to the next. +** +** Ticket #3860: The llvm-gcc-4.2 compiler from Apple chokes on +** the ((void*)&((char*)0)[X]) construct. But MSVC chokes on ((void*)(X)). +** So we have to define the macros in different ways depending on the +** compiler. +*/ +#if defined(HAVE_STDINT_H) /* Use this case if we have ANSI headers */ +# define SQLITE_INT_TO_PTR(X) ((void*)(intptr_t)(X)) +# define SQLITE_PTR_TO_INT(X) ((int)(intptr_t)(X)) +#elif defined(__PTRDIFF_TYPE__) /* This case should work for GCC */ +# define SQLITE_INT_TO_PTR(X) ((void*)(__PTRDIFF_TYPE__)(X)) +# define SQLITE_PTR_TO_INT(X) ((int)(__PTRDIFF_TYPE__)(X)) +#elif !defined(__GNUC__) /* Works for compilers other than LLVM */ +# define SQLITE_INT_TO_PTR(X) ((void*)&((char*)0)[X]) +# define SQLITE_PTR_TO_INT(X) ((int)(((char*)X)-(char*)0)) +#else /* Generates a warning - but it always works */ +# define SQLITE_INT_TO_PTR(X) ((void*)(X)) +# define SQLITE_PTR_TO_INT(X) ((int)(X)) +#endif + +/* +** A macro to hint to the compiler that a function should not be +** inlined. +*/ +#if defined(__GNUC__) +# define SQLITE_NOINLINE __attribute__((noinline)) +#elif defined(_MSC_VER) && _MSC_VER>=1310 +# define SQLITE_NOINLINE __declspec(noinline) +#else +# define SQLITE_NOINLINE +#endif + +/* +** Make sure that the compiler intrinsics we desire are enabled when +** compiling with an appropriate version of MSVC unless prevented by +** the SQLITE_DISABLE_INTRINSIC define. +*/ +#if !defined(SQLITE_DISABLE_INTRINSIC) +# if defined(_MSC_VER) && _MSC_VER>=1400 +# if !defined(_WIN32_WCE) +# include +# pragma intrinsic(_byteswap_ushort) +# pragma intrinsic(_byteswap_ulong) +# pragma intrinsic(_byteswap_uint64) +# pragma intrinsic(_ReadWriteBarrier) +# else +# include +# endif +# endif +#endif + +/* +** The SQLITE_THREADSAFE macro must be defined as 0, 1, or 2. +** 0 means mutexes are permanently disable and the library is never +** threadsafe. 1 means the library is serialized which is the highest +** level of threadsafety. 2 means the library is multithreaded - multiple +** threads can use SQLite as long as no two threads try to use the same +** database connection at the same time. +** +** Older versions of SQLite used an optional THREADSAFE macro. +** We support that for legacy. +** +** To ensure that the correct value of "THREADSAFE" is reported when querying +** for compile-time options at runtime (e.g. "PRAGMA compile_options"), this +** logic is partially replicated in ctime.c. If it is updated here, it should +** also be updated there. +*/ +#if !defined(SQLITE_THREADSAFE) +# if defined(THREADSAFE) +# define SQLITE_THREADSAFE THREADSAFE +# else +# define SQLITE_THREADSAFE 1 /* IMP: R-07272-22309 */ +# endif +#endif + +/* +** Powersafe overwrite is on by default. But can be turned off using +** the -DSQLITE_POWERSAFE_OVERWRITE=0 command-line option. +*/ +#ifndef SQLITE_POWERSAFE_OVERWRITE +# define SQLITE_POWERSAFE_OVERWRITE 1 +#endif + +/* +** EVIDENCE-OF: R-25715-37072 Memory allocation statistics are enabled by +** default unless SQLite is compiled with SQLITE_DEFAULT_MEMSTATUS=0 in +** which case memory allocation statistics are disabled by default. +*/ +#if !defined(SQLITE_DEFAULT_MEMSTATUS) +# define SQLITE_DEFAULT_MEMSTATUS 1 +#endif + +/* +** Exactly one of the following macros must be defined in order to +** specify which memory allocation subsystem to use. +** +** SQLITE_SYSTEM_MALLOC // Use normal system malloc() +** SQLITE_WIN32_MALLOC // Use Win32 native heap API +** SQLITE_ZERO_MALLOC // Use a stub allocator that always fails +** SQLITE_MEMDEBUG // Debugging version of system malloc() +** +** On Windows, if the SQLITE_WIN32_MALLOC_VALIDATE macro is defined and the +** assert() macro is enabled, each call into the Win32 native heap subsystem +** will cause HeapValidate to be called. If heap validation should fail, an +** assertion will be triggered. +** +** If none of the above are defined, then set SQLITE_SYSTEM_MALLOC as +** the default. +*/ +#if defined(SQLITE_SYSTEM_MALLOC) \ + + defined(SQLITE_WIN32_MALLOC) \ + + defined(SQLITE_ZERO_MALLOC) \ + + defined(SQLITE_MEMDEBUG)>1 +# error "Two or more of the following compile-time configuration options\ + are defined but at most one is allowed:\ + SQLITE_SYSTEM_MALLOC, SQLITE_WIN32_MALLOC, SQLITE_MEMDEBUG,\ + SQLITE_ZERO_MALLOC" +#endif +#if defined(SQLITE_SYSTEM_MALLOC) \ + + defined(SQLITE_WIN32_MALLOC) \ + + defined(SQLITE_ZERO_MALLOC) \ + + defined(SQLITE_MEMDEBUG)==0 +# define SQLITE_SYSTEM_MALLOC 1 +#endif + +/* +** If SQLITE_MALLOC_SOFT_LIMIT is not zero, then try to keep the +** sizes of memory allocations below this value where possible. +*/ +#if !defined(SQLITE_MALLOC_SOFT_LIMIT) +# define SQLITE_MALLOC_SOFT_LIMIT 1024 +#endif + +/* +** We need to define _XOPEN_SOURCE as follows in order to enable +** recursive mutexes on most Unix systems and fchmod() on OpenBSD. +** But _XOPEN_SOURCE define causes problems for Mac OS X, so omit +** it. +*/ +#if !defined(_XOPEN_SOURCE) && !defined(__DARWIN__) && !defined(__APPLE__) +# define _XOPEN_SOURCE 600 +#endif + +/* +** NDEBUG and SQLITE_DEBUG are opposites. It should always be true that +** defined(NDEBUG)==!defined(SQLITE_DEBUG). If this is not currently true, +** make it true by defining or undefining NDEBUG. +** +** Setting NDEBUG makes the code smaller and faster by disabling the +** assert() statements in the code. So we want the default action +** to be for NDEBUG to be set and NDEBUG to be undefined only if SQLITE_DEBUG +** is set. Thus NDEBUG becomes an opt-in rather than an opt-out +** feature. +*/ +#if !defined(NDEBUG) && !defined(SQLITE_DEBUG) +# define NDEBUG 1 +#endif +#if defined(NDEBUG) && defined(SQLITE_DEBUG) +# undef NDEBUG +#endif + +/* +** Enable SQLITE_ENABLE_EXPLAIN_COMMENTS if SQLITE_DEBUG is turned on. +*/ +#if !defined(SQLITE_ENABLE_EXPLAIN_COMMENTS) && defined(SQLITE_DEBUG) +# define SQLITE_ENABLE_EXPLAIN_COMMENTS 1 +#endif + +/* +** The testcase() macro is used to aid in coverage testing. When +** doing coverage testing, the condition inside the argument to +** testcase() must be evaluated both true and false in order to +** get full branch coverage. The testcase() macro is inserted +** to help ensure adequate test coverage in places where simple +** condition/decision coverage is inadequate. For example, testcase() +** can be used to make sure boundary values are tested. For +** bitmask tests, testcase() can be used to make sure each bit +** is significant and used at least once. On switch statements +** where multiple cases go to the same block of code, testcase() +** can insure that all cases are evaluated. +** +*/ +#ifdef SQLITE_COVERAGE_TEST +SQLITE_PRIVATE void sqlite3Coverage(int); +# define testcase(X) if( X ){ sqlite3Coverage(__LINE__); } +#else +# define testcase(X) +#endif + +/* +** The TESTONLY macro is used to enclose variable declarations or +** other bits of code that are needed to support the arguments +** within testcase() and assert() macros. +*/ +#if !defined(NDEBUG) || defined(SQLITE_COVERAGE_TEST) +# define TESTONLY(X) X +#else +# define TESTONLY(X) +#endif + +/* +** Sometimes we need a small amount of code such as a variable initialization +** to setup for a later assert() statement. We do not want this code to +** appear when assert() is disabled. The following macro is therefore +** used to contain that setup code. The "VVA" acronym stands for +** "Verification, Validation, and Accreditation". In other words, the +** code within VVA_ONLY() will only run during verification processes. +*/ +#ifndef NDEBUG +# define VVA_ONLY(X) X +#else +# define VVA_ONLY(X) +#endif + +/* +** The ALWAYS and NEVER macros surround boolean expressions which +** are intended to always be true or false, respectively. Such +** expressions could be omitted from the code completely. But they +** are included in a few cases in order to enhance the resilience +** of SQLite to unexpected behavior - to make the code "self-healing" +** or "ductile" rather than being "brittle" and crashing at the first +** hint of unplanned behavior. +** +** In other words, ALWAYS and NEVER are added for defensive code. +** +** When doing coverage testing ALWAYS and NEVER are hard-coded to +** be true and false so that the unreachable code they specify will +** not be counted as untested code. +*/ +#if defined(SQLITE_COVERAGE_TEST) || defined(SQLITE_MUTATION_TEST) +# define ALWAYS(X) (1) +# define NEVER(X) (0) +#elif !defined(NDEBUG) +# define ALWAYS(X) ((X)?1:(assert(0),0)) +# define NEVER(X) ((X)?(assert(0),1):0) +#else +# define ALWAYS(X) (X) +# define NEVER(X) (X) +#endif + +/* +** Some conditionals are optimizations only. In other words, if the +** conditionals are replaced with a constant 1 (true) or 0 (false) then +** the correct answer is still obtained, though perhaps not as quickly. +** +** The following macros mark these optimizations conditionals. +*/ +#if defined(SQLITE_MUTATION_TEST) +# define OK_IF_ALWAYS_TRUE(X) (1) +# define OK_IF_ALWAYS_FALSE(X) (0) +#else +# define OK_IF_ALWAYS_TRUE(X) (X) +# define OK_IF_ALWAYS_FALSE(X) (X) +#endif + +/* +** Some malloc failures are only possible if SQLITE_TEST_REALLOC_STRESS is +** defined. We need to defend against those failures when testing with +** SQLITE_TEST_REALLOC_STRESS, but we don't want the unreachable branches +** during a normal build. The following macro can be used to disable tests +** that are always false except when SQLITE_TEST_REALLOC_STRESS is set. +*/ +#if defined(SQLITE_TEST_REALLOC_STRESS) +# define ONLY_IF_REALLOC_STRESS(X) (X) +#elif !defined(NDEBUG) +# define ONLY_IF_REALLOC_STRESS(X) ((X)?(assert(0),1):0) +#else +# define ONLY_IF_REALLOC_STRESS(X) (0) +#endif + +/* +** Declarations used for tracing the operating system interfaces. +*/ +#if defined(SQLITE_FORCE_OS_TRACE) || defined(SQLITE_TEST) || \ + (defined(SQLITE_DEBUG) && SQLITE_OS_WIN) + extern int sqlite3OSTrace; +# define OSTRACE(X) if( sqlite3OSTrace ) sqlite3DebugPrintf X +# define SQLITE_HAVE_OS_TRACE +#else +# define OSTRACE(X) +# undef SQLITE_HAVE_OS_TRACE +#endif + +/* +** Is the sqlite3ErrName() function needed in the build? Currently, +** it is needed by "mutex_w32.c" (when debugging), "os_win.c" (when +** OSTRACE is enabled), and by several "test*.c" files (which are +** compiled using SQLITE_TEST). +*/ +#if defined(SQLITE_HAVE_OS_TRACE) || defined(SQLITE_TEST) || \ + (defined(SQLITE_DEBUG) && SQLITE_OS_WIN) +# define SQLITE_NEED_ERR_NAME +#else +# undef SQLITE_NEED_ERR_NAME +#endif + +/* +** SQLITE_ENABLE_EXPLAIN_COMMENTS is incompatible with SQLITE_OMIT_EXPLAIN +*/ +#ifdef SQLITE_OMIT_EXPLAIN +# undef SQLITE_ENABLE_EXPLAIN_COMMENTS +#endif + +/* +** Return true (non-zero) if the input is an integer that is too large +** to fit in 32-bits. This macro is used inside of various testcase() +** macros to verify that we have tested SQLite for large-file support. +*/ +#define IS_BIG_INT(X) (((X)&~(i64)0xffffffff)!=0) + +/* +** The macro unlikely() is a hint that surrounds a boolean +** expression that is usually false. Macro likely() surrounds +** a boolean expression that is usually true. These hints could, +** in theory, be used by the compiler to generate better code, but +** currently they are just comments for human readers. +*/ +#define likely(X) (X) +#define unlikely(X) (X) + +/************** Include hash.h in the middle of sqliteInt.h ******************/ +/************** Begin file hash.h ********************************************/ +/* +** 2001 September 22 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** This is the header file for the generic hash-table implementation +** used in SQLite. +*/ +#ifndef SQLITE_HASH_H +#define SQLITE_HASH_H + +/* Forward declarations of structures. */ +typedef struct Hash Hash; +typedef struct HashElem HashElem; + +/* A complete hash table is an instance of the following structure. +** The internals of this structure are intended to be opaque -- client +** code should not attempt to access or modify the fields of this structure +** directly. Change this structure only by using the routines below. +** However, some of the "procedures" and "functions" for modifying and +** accessing this structure are really macros, so we can't really make +** this structure opaque. +** +** All elements of the hash table are on a single doubly-linked list. +** Hash.first points to the head of this list. +** +** There are Hash.htsize buckets. Each bucket points to a spot in +** the global doubly-linked list. The contents of the bucket are the +** element pointed to plus the next _ht.count-1 elements in the list. +** +** Hash.htsize and Hash.ht may be zero. In that case lookup is done +** by a linear search of the global list. For small tables, the +** Hash.ht table is never allocated because if there are few elements +** in the table, it is faster to do a linear search than to manage +** the hash table. +*/ +struct Hash { + unsigned int htsize; /* Number of buckets in the hash table */ + unsigned int count; /* Number of entries in this table */ + HashElem *first; /* The first element of the array */ + struct _ht { /* the hash table */ + unsigned int count; /* Number of entries with this hash */ + HashElem *chain; /* Pointer to first entry with this hash */ + } *ht; +}; + +/* Each element in the hash table is an instance of the following +** structure. All elements are stored on a single doubly-linked list. +** +** Again, this structure is intended to be opaque, but it can't really +** be opaque because it is used by macros. +*/ +struct HashElem { + HashElem *next, *prev; /* Next and previous elements in the table */ + void *data; /* Data associated with this element */ + const char *pKey; /* Key associated with this element */ +}; + +/* +** Access routines. To delete, insert a NULL pointer. +*/ +SQLITE_PRIVATE void sqlite3HashInit(Hash*); +SQLITE_PRIVATE void *sqlite3HashInsert(Hash*, const char *pKey, void *pData); +SQLITE_PRIVATE void *sqlite3HashFind(const Hash*, const char *pKey); +SQLITE_PRIVATE void sqlite3HashClear(Hash*); + +/* +** Macros for looping over all elements of a hash table. The idiom is +** like this: +** +** Hash h; +** HashElem *p; +** ... +** for(p=sqliteHashFirst(&h); p; p=sqliteHashNext(p)){ +** SomeStructure *pData = sqliteHashData(p); +** // do something with pData +** } +*/ +#define sqliteHashFirst(H) ((H)->first) +#define sqliteHashNext(E) ((E)->next) +#define sqliteHashData(E) ((E)->data) +/* #define sqliteHashKey(E) ((E)->pKey) // NOT USED */ +/* #define sqliteHashKeysize(E) ((E)->nKey) // NOT USED */ + +/* +** Number of entries in a hash table +*/ +/* #define sqliteHashCount(H) ((H)->count) // NOT USED */ + +#endif /* SQLITE_HASH_H */ + +/************** End of hash.h ************************************************/ +/************** Continuing where we left off in sqliteInt.h ******************/ +/************** Include parse.h in the middle of sqliteInt.h *****************/ +/************** Begin file parse.h *******************************************/ +#define TK_SEMI 1 +#define TK_EXPLAIN 2 +#define TK_QUERY 3 +#define TK_PLAN 4 +#define TK_BEGIN 5 +#define TK_TRANSACTION 6 +#define TK_DEFERRED 7 +#define TK_IMMEDIATE 8 +#define TK_EXCLUSIVE 9 +#define TK_COMMIT 10 +#define TK_END 11 +#define TK_ROLLBACK 12 +#define TK_SAVEPOINT 13 +#define TK_RELEASE 14 +#define TK_TO 15 +#define TK_TABLE 16 +#define TK_CREATE 17 +#define TK_IF 18 +#define TK_NOT 19 +#define TK_EXISTS 20 +#define TK_TEMP 21 +#define TK_LP 22 +#define TK_RP 23 +#define TK_AS 24 +#define TK_WITHOUT 25 +#define TK_COMMA 26 +#define TK_ABORT 27 +#define TK_ACTION 28 +#define TK_AFTER 29 +#define TK_ANALYZE 30 +#define TK_ASC 31 +#define TK_ATTACH 32 +#define TK_BEFORE 33 +#define TK_BY 34 +#define TK_CASCADE 35 +#define TK_CAST 36 +#define TK_CONFLICT 37 +#define TK_DATABASE 38 +#define TK_DESC 39 +#define TK_DETACH 40 +#define TK_EACH 41 +#define TK_FAIL 42 +#define TK_OR 43 +#define TK_AND 44 +#define TK_IS 45 +#define TK_MATCH 46 +#define TK_LIKE_KW 47 +#define TK_BETWEEN 48 +#define TK_IN 49 +#define TK_ISNULL 50 +#define TK_NOTNULL 51 +#define TK_NE 52 +#define TK_EQ 53 +#define TK_GT 54 +#define TK_LE 55 +#define TK_LT 56 +#define TK_GE 57 +#define TK_ESCAPE 58 +#define TK_ID 59 +#define TK_COLUMNKW 60 +#define TK_DO 61 +#define TK_FOR 62 +#define TK_IGNORE 63 +#define TK_INITIALLY 64 +#define TK_INSTEAD 65 +#define TK_NO 66 +#define TK_KEY 67 +#define TK_OF 68 +#define TK_OFFSET 69 +#define TK_PRAGMA 70 +#define TK_RAISE 71 +#define TK_RECURSIVE 72 +#define TK_REPLACE 73 +#define TK_RESTRICT 74 +#define TK_ROW 75 +#define TK_ROWS 76 +#define TK_TRIGGER 77 +#define TK_VACUUM 78 +#define TK_VIEW 79 +#define TK_VIRTUAL 80 +#define TK_WITH 81 +#define TK_NULLS 82 +#define TK_FIRST 83 +#define TK_LAST 84 +#define TK_CURRENT 85 +#define TK_FOLLOWING 86 +#define TK_PARTITION 87 +#define TK_PRECEDING 88 +#define TK_RANGE 89 +#define TK_UNBOUNDED 90 +#define TK_EXCLUDE 91 +#define TK_GROUPS 92 +#define TK_OTHERS 93 +#define TK_TIES 94 +#define TK_REINDEX 95 +#define TK_RENAME 96 +#define TK_CTIME_KW 97 +#define TK_ANY 98 +#define TK_BITAND 99 +#define TK_BITOR 100 +#define TK_LSHIFT 101 +#define TK_RSHIFT 102 +#define TK_PLUS 103 +#define TK_MINUS 104 +#define TK_STAR 105 +#define TK_SLASH 106 +#define TK_REM 107 +#define TK_CONCAT 108 +#define TK_COLLATE 109 +#define TK_BITNOT 110 +#define TK_ON 111 +#define TK_INDEXED 112 +#define TK_STRING 113 +#define TK_JOIN_KW 114 +#define TK_CONSTRAINT 115 +#define TK_DEFAULT 116 +#define TK_NULL 117 +#define TK_PRIMARY 118 +#define TK_UNIQUE 119 +#define TK_CHECK 120 +#define TK_REFERENCES 121 +#define TK_AUTOINCR 122 +#define TK_INSERT 123 +#define TK_DELETE 124 +#define TK_UPDATE 125 +#define TK_SET 126 +#define TK_DEFERRABLE 127 +#define TK_FOREIGN 128 +#define TK_DROP 129 +#define TK_UNION 130 +#define TK_ALL 131 +#define TK_EXCEPT 132 +#define TK_INTERSECT 133 +#define TK_SELECT 134 +#define TK_VALUES 135 +#define TK_DISTINCT 136 +#define TK_DOT 137 +#define TK_FROM 138 +#define TK_JOIN 139 +#define TK_USING 140 +#define TK_ORDER 141 +#define TK_GROUP 142 +#define TK_HAVING 143 +#define TK_LIMIT 144 +#define TK_WHERE 145 +#define TK_INTO 146 +#define TK_NOTHING 147 +#define TK_FLOAT 148 +#define TK_BLOB 149 +#define TK_INTEGER 150 +#define TK_VARIABLE 151 +#define TK_CASE 152 +#define TK_WHEN 153 +#define TK_THEN 154 +#define TK_ELSE 155 +#define TK_INDEX 156 +#define TK_ALTER 157 +#define TK_ADD 158 +#define TK_WINDOW 159 +#define TK_OVER 160 +#define TK_FILTER 161 +#define TK_COLUMN 162 +#define TK_AGG_FUNCTION 163 +#define TK_AGG_COLUMN 164 +#define TK_TRUEFALSE 165 +#define TK_ISNOT 166 +#define TK_FUNCTION 167 +#define TK_UMINUS 168 +#define TK_UPLUS 169 +#define TK_TRUTH 170 +#define TK_REGISTER 171 +#define TK_VECTOR 172 +#define TK_SELECT_COLUMN 173 +#define TK_IF_NULL_ROW 174 +#define TK_ASTERISK 175 +#define TK_SPAN 176 +#define TK_SPACE 177 +#define TK_ILLEGAL 178 + +/************** End of parse.h ***********************************************/ +/************** Continuing where we left off in sqliteInt.h ******************/ +#include +#include +#include +#include +#include + +/* +** Use a macro to replace memcpy() if compiled with SQLITE_INLINE_MEMCPY. +** This allows better measurements of where memcpy() is used when running +** cachegrind. But this macro version of memcpy() is very slow so it +** should not be used in production. This is a performance measurement +** hack only. +*/ +#ifdef SQLITE_INLINE_MEMCPY +# define memcpy(D,S,N) {char*xxd=(char*)(D);const char*xxs=(const char*)(S);\ + int xxn=(N);while(xxn-->0)*(xxd++)=*(xxs++);} +#endif + +/* +** If compiling for a processor that lacks floating point support, +** substitute integer for floating-point +*/ +#ifdef SQLITE_OMIT_FLOATING_POINT +# define double sqlite_int64 +# define float sqlite_int64 +# define LONGDOUBLE_TYPE sqlite_int64 +# ifndef SQLITE_BIG_DBL +# define SQLITE_BIG_DBL (((sqlite3_int64)1)<<50) +# endif +# define SQLITE_OMIT_DATETIME_FUNCS 1 +# define SQLITE_OMIT_TRACE 1 +# undef SQLITE_MIXED_ENDIAN_64BIT_FLOAT +# undef SQLITE_HAVE_ISNAN +#endif +#ifndef SQLITE_BIG_DBL +# define SQLITE_BIG_DBL (1e99) +#endif + +/* +** OMIT_TEMPDB is set to 1 if SQLITE_OMIT_TEMPDB is defined, or 0 +** afterward. Having this macro allows us to cause the C compiler +** to omit code used by TEMP tables without messy #ifndef statements. +*/ +#ifdef SQLITE_OMIT_TEMPDB +#define OMIT_TEMPDB 1 +#else +#define OMIT_TEMPDB 0 +#endif + +/* +** The "file format" number is an integer that is incremented whenever +** the VDBE-level file format changes. The following macros define the +** the default file format for new databases and the maximum file format +** that the library can read. +*/ +#define SQLITE_MAX_FILE_FORMAT 4 +#ifndef SQLITE_DEFAULT_FILE_FORMAT +# define SQLITE_DEFAULT_FILE_FORMAT 4 +#endif + +/* +** Determine whether triggers are recursive by default. This can be +** changed at run-time using a pragma. +*/ +#ifndef SQLITE_DEFAULT_RECURSIVE_TRIGGERS +# define SQLITE_DEFAULT_RECURSIVE_TRIGGERS 0 +#endif + +/* +** Provide a default value for SQLITE_TEMP_STORE in case it is not specified +** on the command-line +*/ +#ifndef SQLITE_TEMP_STORE +# define SQLITE_TEMP_STORE 1 +#endif + +/* +** If no value has been provided for SQLITE_MAX_WORKER_THREADS, or if +** SQLITE_TEMP_STORE is set to 3 (never use temporary files), set it +** to zero. +*/ +#if SQLITE_TEMP_STORE==3 || SQLITE_THREADSAFE==0 +# undef SQLITE_MAX_WORKER_THREADS +# define SQLITE_MAX_WORKER_THREADS 0 +#endif +#ifndef SQLITE_MAX_WORKER_THREADS +# define SQLITE_MAX_WORKER_THREADS 8 +#endif +#ifndef SQLITE_DEFAULT_WORKER_THREADS +# define SQLITE_DEFAULT_WORKER_THREADS 0 +#endif +#if SQLITE_DEFAULT_WORKER_THREADS>SQLITE_MAX_WORKER_THREADS +# undef SQLITE_MAX_WORKER_THREADS +# define SQLITE_MAX_WORKER_THREADS SQLITE_DEFAULT_WORKER_THREADS +#endif + +/* +** The default initial allocation for the pagecache when using separate +** pagecaches for each database connection. A positive number is the +** number of pages. A negative number N translations means that a buffer +** of -1024*N bytes is allocated and used for as many pages as it will hold. +** +** The default value of "20" was choosen to minimize the run-time of the +** speedtest1 test program with options: --shrink-memory --reprepare +*/ +#ifndef SQLITE_DEFAULT_PCACHE_INITSZ +# define SQLITE_DEFAULT_PCACHE_INITSZ 20 +#endif + +/* +** Default value for the SQLITE_CONFIG_SORTERREF_SIZE option. +*/ +#ifndef SQLITE_DEFAULT_SORTERREF_SIZE +# define SQLITE_DEFAULT_SORTERREF_SIZE 0x7fffffff +#endif + +/* +** The compile-time options SQLITE_MMAP_READWRITE and +** SQLITE_ENABLE_BATCH_ATOMIC_WRITE are not compatible with one another. +** You must choose one or the other (or neither) but not both. +*/ +#if defined(SQLITE_MMAP_READWRITE) && defined(SQLITE_ENABLE_BATCH_ATOMIC_WRITE) +#error Cannot use both SQLITE_MMAP_READWRITE and SQLITE_ENABLE_BATCH_ATOMIC_WRITE +#endif + +/* +** GCC does not define the offsetof() macro so we'll have to do it +** ourselves. +*/ +#ifndef offsetof +#define offsetof(STRUCTURE,FIELD) ((int)((char*)&((STRUCTURE*)0)->FIELD)) +#endif + +/* +** Macros to compute minimum and maximum of two numbers. +*/ +#ifndef MIN +# define MIN(A,B) ((A)<(B)?(A):(B)) +#endif +#ifndef MAX +# define MAX(A,B) ((A)>(B)?(A):(B)) +#endif + +/* +** Swap two objects of type TYPE. +*/ +#define SWAP(TYPE,A,B) {TYPE t=A; A=B; B=t;} + +/* +** Check to see if this machine uses EBCDIC. (Yes, believe it or +** not, there are still machines out there that use EBCDIC.) +*/ +#if 'A' == '\301' +# define SQLITE_EBCDIC 1 +#else +# define SQLITE_ASCII 1 +#endif + +/* +** Integers of known sizes. These typedefs might change for architectures +** where the sizes very. Preprocessor macros are available so that the +** types can be conveniently redefined at compile-type. Like this: +** +** cc '-DUINTPTR_TYPE=long long int' ... +*/ +#ifndef UINT32_TYPE +# ifdef HAVE_UINT32_T +# define UINT32_TYPE uint32_t +# else +# define UINT32_TYPE unsigned int +# endif +#endif +#ifndef UINT16_TYPE +# ifdef HAVE_UINT16_T +# define UINT16_TYPE uint16_t +# else +# define UINT16_TYPE unsigned short int +# endif +#endif +#ifndef INT16_TYPE +# ifdef HAVE_INT16_T +# define INT16_TYPE int16_t +# else +# define INT16_TYPE short int +# endif +#endif +#ifndef UINT8_TYPE +# ifdef HAVE_UINT8_T +# define UINT8_TYPE uint8_t +# else +# define UINT8_TYPE unsigned char +# endif +#endif +#ifndef INT8_TYPE +# ifdef HAVE_INT8_T +# define INT8_TYPE int8_t +# else +# define INT8_TYPE signed char +# endif +#endif +#ifndef LONGDOUBLE_TYPE +# define LONGDOUBLE_TYPE long double +#endif +typedef sqlite_int64 i64; /* 8-byte signed integer */ +typedef sqlite_uint64 u64; /* 8-byte unsigned integer */ +typedef UINT32_TYPE u32; /* 4-byte unsigned integer */ +typedef UINT16_TYPE u16; /* 2-byte unsigned integer */ +typedef INT16_TYPE i16; /* 2-byte signed integer */ +typedef UINT8_TYPE u8; /* 1-byte unsigned integer */ +typedef INT8_TYPE i8; /* 1-byte signed integer */ + +/* +** SQLITE_MAX_U32 is a u64 constant that is the maximum u64 value +** that can be stored in a u32 without loss of data. The value +** is 0x00000000ffffffff. But because of quirks of some compilers, we +** have to specify the value in the less intuitive manner shown: +*/ +#define SQLITE_MAX_U32 ((((u64)1)<<32)-1) + +/* +** The datatype used to store estimates of the number of rows in a +** table or index. This is an unsigned integer type. For 99.9% of +** the world, a 32-bit integer is sufficient. But a 64-bit integer +** can be used at compile-time if desired. +*/ +#ifdef SQLITE_64BIT_STATS + typedef u64 tRowcnt; /* 64-bit only if requested at compile-time */ +#else + typedef u32 tRowcnt; /* 32-bit is the default */ +#endif + +/* +** Estimated quantities used for query planning are stored as 16-bit +** logarithms. For quantity X, the value stored is 10*log2(X). This +** gives a possible range of values of approximately 1.0e986 to 1e-986. +** But the allowed values are "grainy". Not every value is representable. +** For example, quantities 16 and 17 are both represented by a LogEst +** of 40. However, since LogEst quantities are suppose to be estimates, +** not exact values, this imprecision is not a problem. +** +** "LogEst" is short for "Logarithmic Estimate". +** +** Examples: +** 1 -> 0 20 -> 43 10000 -> 132 +** 2 -> 10 25 -> 46 25000 -> 146 +** 3 -> 16 100 -> 66 1000000 -> 199 +** 4 -> 20 1000 -> 99 1048576 -> 200 +** 10 -> 33 1024 -> 100 4294967296 -> 320 +** +** The LogEst can be negative to indicate fractional values. +** Examples: +** +** 0.5 -> -10 0.1 -> -33 0.0625 -> -40 +*/ +typedef INT16_TYPE LogEst; + +/* +** Set the SQLITE_PTRSIZE macro to the number of bytes in a pointer +*/ +#ifndef SQLITE_PTRSIZE +# if defined(__SIZEOF_POINTER__) +# define SQLITE_PTRSIZE __SIZEOF_POINTER__ +# elif defined(i386) || defined(__i386__) || defined(_M_IX86) || \ + defined(_M_ARM) || defined(__arm__) || defined(__x86) || \ + (defined(__TOS_AIX__) && !defined(__64BIT__)) +# define SQLITE_PTRSIZE 4 +# else +# define SQLITE_PTRSIZE 8 +# endif +#endif + +/* The uptr type is an unsigned integer large enough to hold a pointer +*/ +#if defined(HAVE_STDINT_H) + typedef uintptr_t uptr; +#elif SQLITE_PTRSIZE==4 + typedef u32 uptr; +#else + typedef u64 uptr; +#endif + +/* +** The SQLITE_WITHIN(P,S,E) macro checks to see if pointer P points to +** something between S (inclusive) and E (exclusive). +** +** In other words, S is a buffer and E is a pointer to the first byte after +** the end of buffer S. This macro returns true if P points to something +** contained within the buffer S. +*/ +#define SQLITE_WITHIN(P,S,E) (((uptr)(P)>=(uptr)(S))&&((uptr)(P)<(uptr)(E))) + + +/* +** Macros to determine whether the machine is big or little endian, +** and whether or not that determination is run-time or compile-time. +** +** For best performance, an attempt is made to guess at the byte-order +** using C-preprocessor macros. If that is unsuccessful, or if +** -DSQLITE_BYTEORDER=0 is set, then byte-order is determined +** at run-time. +*/ +#ifndef SQLITE_BYTEORDER +# if defined(i386) || defined(__i386__) || defined(_M_IX86) || \ + defined(__x86_64) || defined(__x86_64__) || defined(_M_X64) || \ + defined(_M_AMD64) || defined(_M_ARM) || defined(__x86) || \ + defined(__ARMEL__) || defined(__AARCH64EL__) || defined(_M_ARM64) +# define SQLITE_BYTEORDER 1234 +# elif defined(sparc) || defined(__ppc__) || \ + defined(__ARMEB__) || defined(__AARCH64EB__) +# define SQLITE_BYTEORDER 4321 +# else +# define SQLITE_BYTEORDER 0 +# endif +#endif +#if SQLITE_BYTEORDER==4321 +# define SQLITE_BIGENDIAN 1 +# define SQLITE_LITTLEENDIAN 0 +# define SQLITE_UTF16NATIVE SQLITE_UTF16BE +#elif SQLITE_BYTEORDER==1234 +# define SQLITE_BIGENDIAN 0 +# define SQLITE_LITTLEENDIAN 1 +# define SQLITE_UTF16NATIVE SQLITE_UTF16LE +#else +# ifdef SQLITE_AMALGAMATION + const int sqlite3one = 1; +# else + extern const int sqlite3one; +# endif +# define SQLITE_BIGENDIAN (*(char *)(&sqlite3one)==0) +# define SQLITE_LITTLEENDIAN (*(char *)(&sqlite3one)==1) +# define SQLITE_UTF16NATIVE (SQLITE_BIGENDIAN?SQLITE_UTF16BE:SQLITE_UTF16LE) +#endif + +/* +** Constants for the largest and smallest possible 64-bit signed integers. +** These macros are designed to work correctly on both 32-bit and 64-bit +** compilers. +*/ +#define LARGEST_INT64 (0xffffffff|(((i64)0x7fffffff)<<32)) +#define SMALLEST_INT64 (((i64)-1) - LARGEST_INT64) + +/* +** Round up a number to the next larger multiple of 8. This is used +** to force 8-byte alignment on 64-bit architectures. +*/ +#define ROUND8(x) (((x)+7)&~7) + +/* +** Round down to the nearest multiple of 8 +*/ +#define ROUNDDOWN8(x) ((x)&~7) + +/* +** Assert that the pointer X is aligned to an 8-byte boundary. This +** macro is used only within assert() to verify that the code gets +** all alignment restrictions correct. +** +** Except, if SQLITE_4_BYTE_ALIGNED_MALLOC is defined, then the +** underlying malloc() implementation might return us 4-byte aligned +** pointers. In that case, only verify 4-byte alignment. +*/ +#ifdef SQLITE_4_BYTE_ALIGNED_MALLOC +# define EIGHT_BYTE_ALIGNMENT(X) ((((char*)(X) - (char*)0)&3)==0) +#else +# define EIGHT_BYTE_ALIGNMENT(X) ((((char*)(X) - (char*)0)&7)==0) +#endif + +/* +** Disable MMAP on platforms where it is known to not work +*/ +#if defined(__OpenBSD__) || defined(__QNXNTO__) +# undef SQLITE_MAX_MMAP_SIZE +# define SQLITE_MAX_MMAP_SIZE 0 +#endif + +/* +** Default maximum size of memory used by memory-mapped I/O in the VFS +*/ +#ifdef __APPLE__ +# include +#endif +#ifndef SQLITE_MAX_MMAP_SIZE +# if defined(__linux__) \ + || defined(_WIN32) \ + || (defined(__APPLE__) && defined(__MACH__)) \ + || defined(__sun) \ + || defined(__FreeBSD__) \ + || defined(__DragonFly__) +# define SQLITE_MAX_MMAP_SIZE 0x7fff0000 /* 2147418112 */ +# else +# define SQLITE_MAX_MMAP_SIZE 0 +# endif +#endif + +/* +** The default MMAP_SIZE is zero on all platforms. Or, even if a larger +** default MMAP_SIZE is specified at compile-time, make sure that it does +** not exceed the maximum mmap size. +*/ +#ifndef SQLITE_DEFAULT_MMAP_SIZE +# define SQLITE_DEFAULT_MMAP_SIZE 0 +#endif +#if SQLITE_DEFAULT_MMAP_SIZE>SQLITE_MAX_MMAP_SIZE +# undef SQLITE_DEFAULT_MMAP_SIZE +# define SQLITE_DEFAULT_MMAP_SIZE SQLITE_MAX_MMAP_SIZE +#endif + +/* +** SELECTTRACE_ENABLED will be either 1 or 0 depending on whether or not +** the Select query generator tracing logic is turned on. +*/ +#if defined(SQLITE_ENABLE_SELECTTRACE) +# define SELECTTRACE_ENABLED 1 +#else +# define SELECTTRACE_ENABLED 0 +#endif + +/* +** An instance of the following structure is used to store the busy-handler +** callback for a given sqlite handle. +** +** The sqlite.busyHandler member of the sqlite struct contains the busy +** callback for the database handle. Each pager opened via the sqlite +** handle is passed a pointer to sqlite.busyHandler. The busy-handler +** callback is currently invoked only from within pager.c. +*/ +typedef struct BusyHandler BusyHandler; +struct BusyHandler { + int (*xBusyHandler)(void *,int); /* The busy callback */ + void *pBusyArg; /* First arg to busy callback */ + int nBusy; /* Incremented with each busy call */ + u8 bExtraFileArg; /* Include sqlite3_file as callback arg */ +}; + +/* +** Name of the master database table. The master database table +** is a special table that holds the names and attributes of all +** user tables and indices. +*/ +#define MASTER_NAME "sqlite_master" +#define TEMP_MASTER_NAME "sqlite_temp_master" + +/* +** The root-page of the master database table. +*/ +#define MASTER_ROOT 1 + +/* +** The name of the schema table. +*/ +#define SCHEMA_TABLE(x) ((!OMIT_TEMPDB)&&(x==1)?TEMP_MASTER_NAME:MASTER_NAME) + +/* +** A convenience macro that returns the number of elements in +** an array. +*/ +#define ArraySize(X) ((int)(sizeof(X)/sizeof(X[0]))) + +/* +** Determine if the argument is a power of two +*/ +#define IsPowerOfTwo(X) (((X)&((X)-1))==0) + +/* +** The following value as a destructor means to use sqlite3DbFree(). +** The sqlite3DbFree() routine requires two parameters instead of the +** one parameter that destructors normally want. So we have to introduce +** this magic value that the code knows to handle differently. Any +** pointer will work here as long as it is distinct from SQLITE_STATIC +** and SQLITE_TRANSIENT. +*/ +#define SQLITE_DYNAMIC ((sqlite3_destructor_type)sqlite3MallocSize) + +/* +** When SQLITE_OMIT_WSD is defined, it means that the target platform does +** not support Writable Static Data (WSD) such as global and static variables. +** All variables must either be on the stack or dynamically allocated from +** the heap. When WSD is unsupported, the variable declarations scattered +** throughout the SQLite code must become constants instead. The SQLITE_WSD +** macro is used for this purpose. And instead of referencing the variable +** directly, we use its constant as a key to lookup the run-time allocated +** buffer that holds real variable. The constant is also the initializer +** for the run-time allocated buffer. +** +** In the usual case where WSD is supported, the SQLITE_WSD and GLOBAL +** macros become no-ops and have zero performance impact. +*/ +#ifdef SQLITE_OMIT_WSD + #define SQLITE_WSD const + #define GLOBAL(t,v) (*(t*)sqlite3_wsd_find((void*)&(v), sizeof(v))) + #define sqlite3GlobalConfig GLOBAL(struct Sqlite3Config, sqlite3Config) +SQLITE_API int sqlite3_wsd_init(int N, int J); +SQLITE_API void *sqlite3_wsd_find(void *K, int L); +#else + #define SQLITE_WSD + #define GLOBAL(t,v) v + #define sqlite3GlobalConfig sqlite3Config +#endif + +/* +** The following macros are used to suppress compiler warnings and to +** make it clear to human readers when a function parameter is deliberately +** left unused within the body of a function. This usually happens when +** a function is called via a function pointer. For example the +** implementation of an SQL aggregate step callback may not use the +** parameter indicating the number of arguments passed to the aggregate, +** if it knows that this is enforced elsewhere. +** +** When a function parameter is not used at all within the body of a function, +** it is generally named "NotUsed" or "NotUsed2" to make things even clearer. +** However, these macros may also be used to suppress warnings related to +** parameters that may or may not be used depending on compilation options. +** For example those parameters only used in assert() statements. In these +** cases the parameters are named as per the usual conventions. +*/ +#define UNUSED_PARAMETER(x) (void)(x) +#define UNUSED_PARAMETER2(x,y) UNUSED_PARAMETER(x),UNUSED_PARAMETER(y) + +/* +** Forward references to structures +*/ +typedef struct AggInfo AggInfo; +typedef struct AuthContext AuthContext; +typedef struct AutoincInfo AutoincInfo; +typedef struct Bitvec Bitvec; +typedef struct CollSeq CollSeq; +typedef struct Column Column; +typedef struct Db Db; +typedef struct Schema Schema; +typedef struct Expr Expr; +typedef struct ExprList ExprList; +typedef struct FKey FKey; +typedef struct FuncDestructor FuncDestructor; +typedef struct FuncDef FuncDef; +typedef struct FuncDefHash FuncDefHash; +typedef struct IdList IdList; +typedef struct Index Index; +typedef struct IndexSample IndexSample; +typedef struct KeyClass KeyClass; +typedef struct KeyInfo KeyInfo; +typedef struct Lookaside Lookaside; +typedef struct LookasideSlot LookasideSlot; +typedef struct Module Module; +typedef struct NameContext NameContext; +typedef struct Parse Parse; +typedef struct PreUpdate PreUpdate; +typedef struct PrintfArguments PrintfArguments; +typedef struct RenameToken RenameToken; +typedef struct RowSet RowSet; +typedef struct Savepoint Savepoint; +typedef struct Select Select; +typedef struct SQLiteThread SQLiteThread; +typedef struct SelectDest SelectDest; +typedef struct SrcList SrcList; +typedef struct sqlite3_str StrAccum; /* Internal alias for sqlite3_str */ +typedef struct Table Table; +typedef struct TableLock TableLock; +typedef struct Token Token; +typedef struct TreeView TreeView; +typedef struct Trigger Trigger; +typedef struct TriggerPrg TriggerPrg; +typedef struct TriggerStep TriggerStep; +typedef struct UnpackedRecord UnpackedRecord; +typedef struct Upsert Upsert; +typedef struct VTable VTable; +typedef struct VtabCtx VtabCtx; +typedef struct Walker Walker; +typedef struct WhereInfo WhereInfo; +typedef struct Window Window; +typedef struct With With; + + +/* +** The bitmask datatype defined below is used for various optimizations. +** +** Changing this from a 64-bit to a 32-bit type limits the number of +** tables in a join to 32 instead of 64. But it also reduces the size +** of the library by 738 bytes on ix86. +*/ +#ifdef SQLITE_BITMASK_TYPE + typedef SQLITE_BITMASK_TYPE Bitmask; +#else + typedef u64 Bitmask; +#endif + +/* +** The number of bits in a Bitmask. "BMS" means "BitMask Size". +*/ +#define BMS ((int)(sizeof(Bitmask)*8)) + +/* +** A bit in a Bitmask +*/ +#define MASKBIT(n) (((Bitmask)1)<<(n)) +#define MASKBIT32(n) (((unsigned int)1)<<(n)) +#define ALLBITS ((Bitmask)-1) + +/* A VList object records a mapping between parameters/variables/wildcards +** in the SQL statement (such as $abc, @pqr, or :xyz) and the integer +** variable number associated with that parameter. See the format description +** on the sqlite3VListAdd() routine for more information. A VList is really +** just an array of integers. +*/ +typedef int VList; + +/* +** Defer sourcing vdbe.h and btree.h until after the "u8" and +** "BusyHandler" typedefs. vdbe.h also requires a few of the opaque +** pointer types (i.e. FuncDef) defined above. +*/ +/************** Include btree.h in the middle of sqliteInt.h *****************/ +/************** Begin file btree.h *******************************************/ +/* +** 2001 September 15 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** This header file defines the interface that the sqlite B-Tree file +** subsystem. See comments in the source code for a detailed description +** of what each interface routine does. +*/ +#ifndef SQLITE_BTREE_H +#define SQLITE_BTREE_H + +/* TODO: This definition is just included so other modules compile. It +** needs to be revisited. +*/ +#define SQLITE_N_BTREE_META 16 + +/* +** If defined as non-zero, auto-vacuum is enabled by default. Otherwise +** it must be turned on for each database using "PRAGMA auto_vacuum = 1". +*/ +#ifndef SQLITE_DEFAULT_AUTOVACUUM + #define SQLITE_DEFAULT_AUTOVACUUM 0 +#endif + +#define BTREE_AUTOVACUUM_NONE 0 /* Do not do auto-vacuum */ +#define BTREE_AUTOVACUUM_FULL 1 /* Do full auto-vacuum */ +#define BTREE_AUTOVACUUM_INCR 2 /* Incremental vacuum */ + +/* +** Forward declarations of structure +*/ +typedef struct Btree Btree; +typedef struct BtCursor BtCursor; +typedef struct BtShared BtShared; +typedef struct BtreePayload BtreePayload; + + +SQLITE_PRIVATE int sqlite3BtreeOpen( + sqlite3_vfs *pVfs, /* VFS to use with this b-tree */ + const char *zFilename, /* Name of database file to open */ + sqlite3 *db, /* Associated database connection */ + Btree **ppBtree, /* Return open Btree* here */ + int flags, /* Flags */ + int vfsFlags /* Flags passed through to VFS open */ +); + +/* The flags parameter to sqlite3BtreeOpen can be the bitwise or of the +** following values. +** +** NOTE: These values must match the corresponding PAGER_ values in +** pager.h. +*/ +#define BTREE_OMIT_JOURNAL 1 /* Do not create or use a rollback journal */ +#define BTREE_MEMORY 2 /* This is an in-memory DB */ +#define BTREE_SINGLE 4 /* The file contains at most 1 b-tree */ +#define BTREE_UNORDERED 8 /* Use of a hash implementation is OK */ + +SQLITE_PRIVATE int sqlite3BtreeClose(Btree*); +SQLITE_PRIVATE int sqlite3BtreeSetCacheSize(Btree*,int); +SQLITE_PRIVATE int sqlite3BtreeSetSpillSize(Btree*,int); +#if SQLITE_MAX_MMAP_SIZE>0 +SQLITE_PRIVATE int sqlite3BtreeSetMmapLimit(Btree*,sqlite3_int64); +#endif +SQLITE_PRIVATE int sqlite3BtreeSetPagerFlags(Btree*,unsigned); +SQLITE_PRIVATE int sqlite3BtreeSetPageSize(Btree *p, int nPagesize, int nReserve, int eFix); +SQLITE_PRIVATE int sqlite3BtreeGetPageSize(Btree*); +SQLITE_PRIVATE int sqlite3BtreeMaxPageCount(Btree*,int); +SQLITE_PRIVATE u32 sqlite3BtreeLastPage(Btree*); +SQLITE_PRIVATE int sqlite3BtreeSecureDelete(Btree*,int); +SQLITE_PRIVATE int sqlite3BtreeGetOptimalReserve(Btree*); +SQLITE_PRIVATE int sqlite3BtreeGetReserveNoMutex(Btree *p); +SQLITE_PRIVATE int sqlite3BtreeSetAutoVacuum(Btree *, int); +SQLITE_PRIVATE int sqlite3BtreeGetAutoVacuum(Btree *); +SQLITE_PRIVATE int sqlite3BtreeBeginTrans(Btree*,int,int*); +SQLITE_PRIVATE int sqlite3BtreeCommitPhaseOne(Btree*, const char *zMaster); +SQLITE_PRIVATE int sqlite3BtreeCommitPhaseTwo(Btree*, int); +SQLITE_PRIVATE int sqlite3BtreeCommit(Btree*); +SQLITE_PRIVATE int sqlite3BtreeRollback(Btree*,int,int); +SQLITE_PRIVATE int sqlite3BtreeBeginStmt(Btree*,int); +SQLITE_PRIVATE int sqlite3BtreeCreateTable(Btree*, int*, int flags); +SQLITE_PRIVATE int sqlite3BtreeIsInTrans(Btree*); +SQLITE_PRIVATE int sqlite3BtreeIsInReadTrans(Btree*); +SQLITE_PRIVATE int sqlite3BtreeIsInBackup(Btree*); +SQLITE_PRIVATE void *sqlite3BtreeSchema(Btree *, int, void(*)(void *)); +SQLITE_PRIVATE int sqlite3BtreeSchemaLocked(Btree *pBtree); +#ifndef SQLITE_OMIT_SHARED_CACHE +SQLITE_PRIVATE int sqlite3BtreeLockTable(Btree *pBtree, int iTab, u8 isWriteLock); +#endif +SQLITE_PRIVATE int sqlite3BtreeSavepoint(Btree *, int, int); + +SQLITE_PRIVATE const char *sqlite3BtreeGetFilename(Btree *); +SQLITE_PRIVATE const char *sqlite3BtreeGetJournalname(Btree *); +SQLITE_PRIVATE int sqlite3BtreeCopyFile(Btree *, Btree *); + +SQLITE_PRIVATE int sqlite3BtreeIncrVacuum(Btree *); + +/* The flags parameter to sqlite3BtreeCreateTable can be the bitwise OR +** of the flags shown below. +** +** Every SQLite table must have either BTREE_INTKEY or BTREE_BLOBKEY set. +** With BTREE_INTKEY, the table key is a 64-bit integer and arbitrary data +** is stored in the leaves. (BTREE_INTKEY is used for SQL tables.) With +** BTREE_BLOBKEY, the key is an arbitrary BLOB and no content is stored +** anywhere - the key is the content. (BTREE_BLOBKEY is used for SQL +** indices.) +*/ +#define BTREE_INTKEY 1 /* Table has only 64-bit signed integer keys */ +#define BTREE_BLOBKEY 2 /* Table has keys only - no data */ + +SQLITE_PRIVATE int sqlite3BtreeDropTable(Btree*, int, int*); +SQLITE_PRIVATE int sqlite3BtreeClearTable(Btree*, int, int*); +SQLITE_PRIVATE int sqlite3BtreeClearTableOfCursor(BtCursor*); +SQLITE_PRIVATE int sqlite3BtreeTripAllCursors(Btree*, int, int); + +SQLITE_PRIVATE void sqlite3BtreeGetMeta(Btree *pBtree, int idx, u32 *pValue); +SQLITE_PRIVATE int sqlite3BtreeUpdateMeta(Btree*, int idx, u32 value); + +SQLITE_PRIVATE int sqlite3BtreeNewDb(Btree *p); + +/* +** The second parameter to sqlite3BtreeGetMeta or sqlite3BtreeUpdateMeta +** should be one of the following values. The integer values are assigned +** to constants so that the offset of the corresponding field in an +** SQLite database header may be found using the following formula: +** +** offset = 36 + (idx * 4) +** +** For example, the free-page-count field is located at byte offset 36 of +** the database file header. The incr-vacuum-flag field is located at +** byte offset 64 (== 36+4*7). +** +** The BTREE_DATA_VERSION value is not really a value stored in the header. +** It is a read-only number computed by the pager. But we merge it with +** the header value access routines since its access pattern is the same. +** Call it a "virtual meta value". +*/ +#define BTREE_FREE_PAGE_COUNT 0 +#define BTREE_SCHEMA_VERSION 1 +#define BTREE_FILE_FORMAT 2 +#define BTREE_DEFAULT_CACHE_SIZE 3 +#define BTREE_LARGEST_ROOT_PAGE 4 +#define BTREE_TEXT_ENCODING 5 +#define BTREE_USER_VERSION 6 +#define BTREE_INCR_VACUUM 7 +#define BTREE_APPLICATION_ID 8 +#define BTREE_DATA_VERSION 15 /* A virtual meta-value */ + +/* +** Kinds of hints that can be passed into the sqlite3BtreeCursorHint() +** interface. +** +** BTREE_HINT_RANGE (arguments: Expr*, Mem*) +** +** The first argument is an Expr* (which is guaranteed to be constant for +** the lifetime of the cursor) that defines constraints on which rows +** might be fetched with this cursor. The Expr* tree may contain +** TK_REGISTER nodes that refer to values stored in the array of registers +** passed as the second parameter. In other words, if Expr.op==TK_REGISTER +** then the value of the node is the value in Mem[pExpr.iTable]. Any +** TK_COLUMN node in the expression tree refers to the Expr.iColumn-th +** column of the b-tree of the cursor. The Expr tree will not contain +** any function calls nor subqueries nor references to b-trees other than +** the cursor being hinted. +** +** The design of the _RANGE hint is aid b-tree implementations that try +** to prefetch content from remote machines - to provide those +** implementations with limits on what needs to be prefetched and thereby +** reduce network bandwidth. +** +** Note that BTREE_HINT_FLAGS with BTREE_BULKLOAD is the only hint used by +** standard SQLite. The other hints are provided for extentions that use +** the SQLite parser and code generator but substitute their own storage +** engine. +*/ +#define BTREE_HINT_RANGE 0 /* Range constraints on queries */ + +/* +** Values that may be OR'd together to form the argument to the +** BTREE_HINT_FLAGS hint for sqlite3BtreeCursorHint(): +** +** The BTREE_BULKLOAD flag is set on index cursors when the index is going +** to be filled with content that is already in sorted order. +** +** The BTREE_SEEK_EQ flag is set on cursors that will get OP_SeekGE or +** OP_SeekLE opcodes for a range search, but where the range of entries +** selected will all have the same key. In other words, the cursor will +** be used only for equality key searches. +** +*/ +#define BTREE_BULKLOAD 0x00000001 /* Used to full index in sorted order */ +#define BTREE_SEEK_EQ 0x00000002 /* EQ seeks only - no range seeks */ + +/* +** Flags passed as the third argument to sqlite3BtreeCursor(). +** +** For read-only cursors the wrFlag argument is always zero. For read-write +** cursors it may be set to either (BTREE_WRCSR|BTREE_FORDELETE) or just +** (BTREE_WRCSR). If the BTREE_FORDELETE bit is set, then the cursor will +** only be used by SQLite for the following: +** +** * to seek to and then delete specific entries, and/or +** +** * to read values that will be used to create keys that other +** BTREE_FORDELETE cursors will seek to and delete. +** +** The BTREE_FORDELETE flag is an optimization hint. It is not used by +** by this, the native b-tree engine of SQLite, but it is available to +** alternative storage engines that might be substituted in place of this +** b-tree system. For alternative storage engines in which a delete of +** the main table row automatically deletes corresponding index rows, +** the FORDELETE flag hint allows those alternative storage engines to +** skip a lot of work. Namely: FORDELETE cursors may treat all SEEK +** and DELETE operations as no-ops, and any READ operation against a +** FORDELETE cursor may return a null row: 0x01 0x00. +*/ +#define BTREE_WRCSR 0x00000004 /* read-write cursor */ +#define BTREE_FORDELETE 0x00000008 /* Cursor is for seek/delete only */ + +SQLITE_PRIVATE int sqlite3BtreeCursor( + Btree*, /* BTree containing table to open */ + int iTable, /* Index of root page */ + int wrFlag, /* 1 for writing. 0 for read-only */ + struct KeyInfo*, /* First argument to compare function */ + BtCursor *pCursor /* Space to write cursor structure */ +); +SQLITE_PRIVATE BtCursor *sqlite3BtreeFakeValidCursor(void); +SQLITE_PRIVATE int sqlite3BtreeCursorSize(void); +SQLITE_PRIVATE void sqlite3BtreeCursorZero(BtCursor*); +SQLITE_PRIVATE void sqlite3BtreeCursorHintFlags(BtCursor*, unsigned); +#ifdef SQLITE_ENABLE_CURSOR_HINTS +SQLITE_PRIVATE void sqlite3BtreeCursorHint(BtCursor*, int, ...); +#endif + +SQLITE_PRIVATE int sqlite3BtreeCloseCursor(BtCursor*); +SQLITE_PRIVATE int sqlite3BtreeMovetoUnpacked( + BtCursor*, + UnpackedRecord *pUnKey, + i64 intKey, + int bias, + int *pRes +); +SQLITE_PRIVATE int sqlite3BtreeCursorHasMoved(BtCursor*); +SQLITE_PRIVATE int sqlite3BtreeCursorRestore(BtCursor*, int*); +SQLITE_PRIVATE int sqlite3BtreeDelete(BtCursor*, u8 flags); + +/* Allowed flags for sqlite3BtreeDelete() and sqlite3BtreeInsert() */ +#define BTREE_SAVEPOSITION 0x02 /* Leave cursor pointing at NEXT or PREV */ +#define BTREE_AUXDELETE 0x04 /* not the primary delete operation */ +#define BTREE_APPEND 0x08 /* Insert is likely an append */ + +/* An instance of the BtreePayload object describes the content of a single +** entry in either an index or table btree. +** +** Index btrees (used for indexes and also WITHOUT ROWID tables) contain +** an arbitrary key and no data. These btrees have pKey,nKey set to the +** key and the pData,nData,nZero fields are uninitialized. The aMem,nMem +** fields give an array of Mem objects that are a decomposition of the key. +** The nMem field might be zero, indicating that no decomposition is available. +** +** Table btrees (used for rowid tables) contain an integer rowid used as +** the key and passed in the nKey field. The pKey field is zero. +** pData,nData hold the content of the new entry. nZero extra zero bytes +** are appended to the end of the content when constructing the entry. +** The aMem,nMem fields are uninitialized for table btrees. +** +** Field usage summary: +** +** Table BTrees Index Btrees +** +** pKey always NULL encoded key +** nKey the ROWID length of pKey +** pData data not used +** aMem not used decomposed key value +** nMem not used entries in aMem +** nData length of pData not used +** nZero extra zeros after pData not used +** +** This object is used to pass information into sqlite3BtreeInsert(). The +** same information used to be passed as five separate parameters. But placing +** the information into this object helps to keep the interface more +** organized and understandable, and it also helps the resulting code to +** run a little faster by using fewer registers for parameter passing. +*/ +struct BtreePayload { + const void *pKey; /* Key content for indexes. NULL for tables */ + sqlite3_int64 nKey; /* Size of pKey for indexes. PRIMARY KEY for tabs */ + const void *pData; /* Data for tables. */ + sqlite3_value *aMem; /* First of nMem value in the unpacked pKey */ + u16 nMem; /* Number of aMem[] value. Might be zero */ + int nData; /* Size of pData. 0 if none. */ + int nZero; /* Extra zero data appended after pData,nData */ +}; + +SQLITE_PRIVATE int sqlite3BtreeInsert(BtCursor*, const BtreePayload *pPayload, + int flags, int seekResult); +SQLITE_PRIVATE int sqlite3BtreeFirst(BtCursor*, int *pRes); +SQLITE_PRIVATE int sqlite3BtreeLast(BtCursor*, int *pRes); +SQLITE_PRIVATE int sqlite3BtreeNext(BtCursor*, int flags); +SQLITE_PRIVATE int sqlite3BtreeEof(BtCursor*); +SQLITE_PRIVATE int sqlite3BtreePrevious(BtCursor*, int flags); +SQLITE_PRIVATE i64 sqlite3BtreeIntegerKey(BtCursor*); +#ifdef SQLITE_ENABLE_OFFSET_SQL_FUNC +SQLITE_PRIVATE i64 sqlite3BtreeOffset(BtCursor*); +#endif +SQLITE_PRIVATE int sqlite3BtreePayload(BtCursor*, u32 offset, u32 amt, void*); +SQLITE_PRIVATE const void *sqlite3BtreePayloadFetch(BtCursor*, u32 *pAmt); +SQLITE_PRIVATE u32 sqlite3BtreePayloadSize(BtCursor*); +SQLITE_PRIVATE sqlite3_int64 sqlite3BtreeMaxRecordSize(BtCursor*); + +SQLITE_PRIVATE char *sqlite3BtreeIntegrityCheck(Btree*, int *aRoot, int nRoot, int, int*); +SQLITE_PRIVATE struct Pager *sqlite3BtreePager(Btree*); +SQLITE_PRIVATE i64 sqlite3BtreeRowCountEst(BtCursor*); + +#ifndef SQLITE_OMIT_INCRBLOB +SQLITE_PRIVATE int sqlite3BtreePayloadChecked(BtCursor*, u32 offset, u32 amt, void*); +SQLITE_PRIVATE int sqlite3BtreePutData(BtCursor*, u32 offset, u32 amt, void*); +SQLITE_PRIVATE void sqlite3BtreeIncrblobCursor(BtCursor *); +#endif +SQLITE_PRIVATE void sqlite3BtreeClearCursor(BtCursor *); +SQLITE_PRIVATE int sqlite3BtreeSetVersion(Btree *pBt, int iVersion); +SQLITE_PRIVATE int sqlite3BtreeCursorHasHint(BtCursor*, unsigned int mask); +SQLITE_PRIVATE int sqlite3BtreeIsReadonly(Btree *pBt); +SQLITE_PRIVATE int sqlite3HeaderSizeBtree(void); + +#ifndef NDEBUG +SQLITE_PRIVATE int sqlite3BtreeCursorIsValid(BtCursor*); +#endif +SQLITE_PRIVATE int sqlite3BtreeCursorIsValidNN(BtCursor*); + +#ifndef SQLITE_OMIT_BTREECOUNT +SQLITE_PRIVATE int sqlite3BtreeCount(BtCursor *, i64 *); +#endif + +#ifdef SQLITE_TEST +SQLITE_PRIVATE int sqlite3BtreeCursorInfo(BtCursor*, int*, int); +SQLITE_PRIVATE void sqlite3BtreeCursorList(Btree*); +#endif + +#ifndef SQLITE_OMIT_WAL +SQLITE_PRIVATE int sqlite3BtreeCheckpoint(Btree*, int, int *, int *); +#endif + +/* +** If we are not using shared cache, then there is no need to +** use mutexes to access the BtShared structures. So make the +** Enter and Leave procedures no-ops. +*/ +#ifndef SQLITE_OMIT_SHARED_CACHE +SQLITE_PRIVATE void sqlite3BtreeEnter(Btree*); +SQLITE_PRIVATE void sqlite3BtreeEnterAll(sqlite3*); +SQLITE_PRIVATE int sqlite3BtreeSharable(Btree*); +SQLITE_PRIVATE void sqlite3BtreeEnterCursor(BtCursor*); +SQLITE_PRIVATE int sqlite3BtreeConnectionCount(Btree*); +#else +# define sqlite3BtreeEnter(X) +# define sqlite3BtreeEnterAll(X) +# define sqlite3BtreeSharable(X) 0 +# define sqlite3BtreeEnterCursor(X) +# define sqlite3BtreeConnectionCount(X) 1 +#endif + +#if !defined(SQLITE_OMIT_SHARED_CACHE) && SQLITE_THREADSAFE +SQLITE_PRIVATE void sqlite3BtreeLeave(Btree*); +SQLITE_PRIVATE void sqlite3BtreeLeaveCursor(BtCursor*); +SQLITE_PRIVATE void sqlite3BtreeLeaveAll(sqlite3*); +#ifndef NDEBUG + /* These routines are used inside assert() statements only. */ +SQLITE_PRIVATE int sqlite3BtreeHoldsMutex(Btree*); +SQLITE_PRIVATE int sqlite3BtreeHoldsAllMutexes(sqlite3*); +SQLITE_PRIVATE int sqlite3SchemaMutexHeld(sqlite3*,int,Schema*); +#endif +#else + +# define sqlite3BtreeLeave(X) +# define sqlite3BtreeLeaveCursor(X) +# define sqlite3BtreeLeaveAll(X) + +# define sqlite3BtreeHoldsMutex(X) 1 +# define sqlite3BtreeHoldsAllMutexes(X) 1 +# define sqlite3SchemaMutexHeld(X,Y,Z) 1 +#endif + + +#endif /* SQLITE_BTREE_H */ + +/************** End of btree.h ***********************************************/ +/************** Continuing where we left off in sqliteInt.h ******************/ +/************** Include vdbe.h in the middle of sqliteInt.h ******************/ +/************** Begin file vdbe.h ********************************************/ +/* +** 2001 September 15 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** Header file for the Virtual DataBase Engine (VDBE) +** +** This header defines the interface to the virtual database engine +** or VDBE. The VDBE implements an abstract machine that runs a +** simple program to access and modify the underlying database. +*/ +#ifndef SQLITE_VDBE_H +#define SQLITE_VDBE_H +/* #include */ + +/* +** A single VDBE is an opaque structure named "Vdbe". Only routines +** in the source file sqliteVdbe.c are allowed to see the insides +** of this structure. +*/ +typedef struct Vdbe Vdbe; + +/* +** The names of the following types declared in vdbeInt.h are required +** for the VdbeOp definition. +*/ +typedef struct sqlite3_value Mem; +typedef struct SubProgram SubProgram; + +/* +** A single instruction of the virtual machine has an opcode +** and as many as three operands. The instruction is recorded +** as an instance of the following structure: +*/ +struct VdbeOp { + u8 opcode; /* What operation to perform */ + signed char p4type; /* One of the P4_xxx constants for p4 */ + u16 p5; /* Fifth parameter is an unsigned 16-bit integer */ + int p1; /* First operand */ + int p2; /* Second parameter (often the jump destination) */ + int p3; /* The third parameter */ + union p4union { /* fourth parameter */ + int i; /* Integer value if p4type==P4_INT32 */ + void *p; /* Generic pointer */ + char *z; /* Pointer to data for string (char array) types */ + i64 *pI64; /* Used when p4type is P4_INT64 */ + double *pReal; /* Used when p4type is P4_REAL */ + FuncDef *pFunc; /* Used when p4type is P4_FUNCDEF */ + sqlite3_context *pCtx; /* Used when p4type is P4_FUNCCTX */ + CollSeq *pColl; /* Used when p4type is P4_COLLSEQ */ + Mem *pMem; /* Used when p4type is P4_MEM */ + VTable *pVtab; /* Used when p4type is P4_VTAB */ + KeyInfo *pKeyInfo; /* Used when p4type is P4_KEYINFO */ + int *ai; /* Used when p4type is P4_INTARRAY */ + SubProgram *pProgram; /* Used when p4type is P4_SUBPROGRAM */ + Table *pTab; /* Used when p4type is P4_TABLE */ +#ifdef SQLITE_ENABLE_CURSOR_HINTS + Expr *pExpr; /* Used when p4type is P4_EXPR */ +#endif + int (*xAdvance)(BtCursor *, int); + } p4; +#ifdef SQLITE_ENABLE_EXPLAIN_COMMENTS + char *zComment; /* Comment to improve readability */ +#endif +#ifdef VDBE_PROFILE + u32 cnt; /* Number of times this instruction was executed */ + u64 cycles; /* Total time spent executing this instruction */ +#endif +#ifdef SQLITE_VDBE_COVERAGE + u32 iSrcLine; /* Source-code line that generated this opcode + ** with flags in the upper 8 bits */ +#endif +}; +typedef struct VdbeOp VdbeOp; + + +/* +** A sub-routine used to implement a trigger program. +*/ +struct SubProgram { + VdbeOp *aOp; /* Array of opcodes for sub-program */ + int nOp; /* Elements in aOp[] */ + int nMem; /* Number of memory cells required */ + int nCsr; /* Number of cursors required */ + u8 *aOnce; /* Array of OP_Once flags */ + void *token; /* id that may be used to recursive triggers */ + SubProgram *pNext; /* Next sub-program already visited */ +}; + +/* +** A smaller version of VdbeOp used for the VdbeAddOpList() function because +** it takes up less space. +*/ +struct VdbeOpList { + u8 opcode; /* What operation to perform */ + signed char p1; /* First operand */ + signed char p2; /* Second parameter (often the jump destination) */ + signed char p3; /* Third parameter */ +}; +typedef struct VdbeOpList VdbeOpList; + +/* +** Allowed values of VdbeOp.p4type +*/ +#define P4_NOTUSED 0 /* The P4 parameter is not used */ +#define P4_TRANSIENT 0 /* P4 is a pointer to a transient string */ +#define P4_STATIC (-1) /* Pointer to a static string */ +#define P4_COLLSEQ (-2) /* P4 is a pointer to a CollSeq structure */ +#define P4_INT32 (-3) /* P4 is a 32-bit signed integer */ +#define P4_SUBPROGRAM (-4) /* P4 is a pointer to a SubProgram structure */ +#define P4_ADVANCE (-5) /* P4 is a pointer to BtreeNext() or BtreePrev() */ +#define P4_TABLE (-6) /* P4 is a pointer to a Table structure */ +/* Above do not own any resources. Must free those below */ +#define P4_FREE_IF_LE (-7) +#define P4_DYNAMIC (-7) /* Pointer to memory from sqliteMalloc() */ +#define P4_FUNCDEF (-8) /* P4 is a pointer to a FuncDef structure */ +#define P4_KEYINFO (-9) /* P4 is a pointer to a KeyInfo structure */ +#define P4_EXPR (-10) /* P4 is a pointer to an Expr tree */ +#define P4_MEM (-11) /* P4 is a pointer to a Mem* structure */ +#define P4_VTAB (-12) /* P4 is a pointer to an sqlite3_vtab structure */ +#define P4_REAL (-13) /* P4 is a 64-bit floating point value */ +#define P4_INT64 (-14) /* P4 is a 64-bit signed integer */ +#define P4_INTARRAY (-15) /* P4 is a vector of 32-bit integers */ +#define P4_FUNCCTX (-16) /* P4 is a pointer to an sqlite3_context object */ +#define P4_DYNBLOB (-17) /* Pointer to memory from sqliteMalloc() */ + +/* Error message codes for OP_Halt */ +#define P5_ConstraintNotNull 1 +#define P5_ConstraintUnique 2 +#define P5_ConstraintCheck 3 +#define P5_ConstraintFK 4 + +/* +** The Vdbe.aColName array contains 5n Mem structures, where n is the +** number of columns of data returned by the statement. +*/ +#define COLNAME_NAME 0 +#define COLNAME_DECLTYPE 1 +#define COLNAME_DATABASE 2 +#define COLNAME_TABLE 3 +#define COLNAME_COLUMN 4 +#ifdef SQLITE_ENABLE_COLUMN_METADATA +# define COLNAME_N 5 /* Number of COLNAME_xxx symbols */ +#else +# ifdef SQLITE_OMIT_DECLTYPE +# define COLNAME_N 1 /* Store only the name */ +# else +# define COLNAME_N 2 /* Store the name and decltype */ +# endif +#endif + +/* +** The following macro converts a label returned by sqlite3VdbeMakeLabel() +** into an index into the Parse.aLabel[] array that contains the resolved +** address of that label. +*/ +#define ADDR(X) (~(X)) + +/* +** The makefile scans the vdbe.c source file and creates the "opcodes.h" +** header file that defines a number for each opcode used by the VDBE. +*/ +/************** Include opcodes.h in the middle of vdbe.h ********************/ +/************** Begin file opcodes.h *****************************************/ +/* Automatically generated. Do not edit */ +/* See the tool/mkopcodeh.tcl script for details */ +#define OP_Savepoint 0 +#define OP_AutoCommit 1 +#define OP_Transaction 2 +#define OP_SorterNext 3 /* jump */ +#define OP_Prev 4 /* jump */ +#define OP_Next 5 /* jump */ +#define OP_Checkpoint 6 +#define OP_JournalMode 7 +#define OP_Vacuum 8 +#define OP_VFilter 9 /* jump, synopsis: iplan=r[P3] zplan='P4' */ +#define OP_VUpdate 10 /* synopsis: data=r[P3@P2] */ +#define OP_Goto 11 /* jump */ +#define OP_Gosub 12 /* jump */ +#define OP_InitCoroutine 13 /* jump */ +#define OP_Yield 14 /* jump */ +#define OP_MustBeInt 15 /* jump */ +#define OP_Jump 16 /* jump */ +#define OP_Once 17 /* jump */ +#define OP_If 18 /* jump */ +#define OP_Not 19 /* same as TK_NOT, synopsis: r[P2]= !r[P1] */ +#define OP_IfNot 20 /* jump */ +#define OP_IfNullRow 21 /* jump, synopsis: if P1.nullRow then r[P3]=NULL, goto P2 */ +#define OP_SeekLT 22 /* jump, synopsis: key=r[P3@P4] */ +#define OP_SeekLE 23 /* jump, synopsis: key=r[P3@P4] */ +#define OP_SeekGE 24 /* jump, synopsis: key=r[P3@P4] */ +#define OP_SeekGT 25 /* jump, synopsis: key=r[P3@P4] */ +#define OP_IfNoHope 26 /* jump, synopsis: key=r[P3@P4] */ +#define OP_NoConflict 27 /* jump, synopsis: key=r[P3@P4] */ +#define OP_NotFound 28 /* jump, synopsis: key=r[P3@P4] */ +#define OP_Found 29 /* jump, synopsis: key=r[P3@P4] */ +#define OP_SeekRowid 30 /* jump, synopsis: intkey=r[P3] */ +#define OP_NotExists 31 /* jump, synopsis: intkey=r[P3] */ +#define OP_Last 32 /* jump */ +#define OP_IfSmaller 33 /* jump */ +#define OP_SorterSort 34 /* jump */ +#define OP_Sort 35 /* jump */ +#define OP_Rewind 36 /* jump */ +#define OP_IdxLE 37 /* jump, synopsis: key=r[P3@P4] */ +#define OP_IdxGT 38 /* jump, synopsis: key=r[P3@P4] */ +#define OP_IdxLT 39 /* jump, synopsis: key=r[P3@P4] */ +#define OP_IdxGE 40 /* jump, synopsis: key=r[P3@P4] */ +#define OP_RowSetRead 41 /* jump, synopsis: r[P3]=rowset(P1) */ +#define OP_RowSetTest 42 /* jump, synopsis: if r[P3] in rowset(P1) goto P2 */ +#define OP_Or 43 /* same as TK_OR, synopsis: r[P3]=(r[P1] || r[P2]) */ +#define OP_And 44 /* same as TK_AND, synopsis: r[P3]=(r[P1] && r[P2]) */ +#define OP_Program 45 /* jump */ +#define OP_FkIfZero 46 /* jump, synopsis: if fkctr[P1]==0 goto P2 */ +#define OP_IfPos 47 /* jump, synopsis: if r[P1]>0 then r[P1]-=P3, goto P2 */ +#define OP_IfNotZero 48 /* jump, synopsis: if r[P1]!=0 then r[P1]--, goto P2 */ +#define OP_DecrJumpZero 49 /* jump, synopsis: if (--r[P1])==0 goto P2 */ +#define OP_IsNull 50 /* jump, same as TK_ISNULL, synopsis: if r[P1]==NULL goto P2 */ +#define OP_NotNull 51 /* jump, same as TK_NOTNULL, synopsis: if r[P1]!=NULL goto P2 */ +#define OP_Ne 52 /* jump, same as TK_NE, synopsis: IF r[P3]!=r[P1] */ +#define OP_Eq 53 /* jump, same as TK_EQ, synopsis: IF r[P3]==r[P1] */ +#define OP_Gt 54 /* jump, same as TK_GT, synopsis: IF r[P3]>r[P1] */ +#define OP_Le 55 /* jump, same as TK_LE, synopsis: IF r[P3]<=r[P1] */ +#define OP_Lt 56 /* jump, same as TK_LT, synopsis: IF r[P3]=r[P1] */ +#define OP_ElseNotEq 58 /* jump, same as TK_ESCAPE */ +#define OP_IncrVacuum 59 /* jump */ +#define OP_VNext 60 /* jump */ +#define OP_Init 61 /* jump, synopsis: Start at P2 */ +#define OP_PureFunc0 62 +#define OP_Function0 63 /* synopsis: r[P3]=func(r[P2@P5]) */ +#define OP_PureFunc 64 +#define OP_Function 65 /* synopsis: r[P3]=func(r[P2@P5]) */ +#define OP_Return 66 +#define OP_EndCoroutine 67 +#define OP_HaltIfNull 68 /* synopsis: if r[P3]=null halt */ +#define OP_Halt 69 +#define OP_Integer 70 /* synopsis: r[P2]=P1 */ +#define OP_Int64 71 /* synopsis: r[P2]=P4 */ +#define OP_String 72 /* synopsis: r[P2]='P4' (len=P1) */ +#define OP_Null 73 /* synopsis: r[P2..P3]=NULL */ +#define OP_SoftNull 74 /* synopsis: r[P1]=NULL */ +#define OP_Blob 75 /* synopsis: r[P2]=P4 (len=P1) */ +#define OP_Variable 76 /* synopsis: r[P2]=parameter(P1,P4) */ +#define OP_Move 77 /* synopsis: r[P2@P3]=r[P1@P3] */ +#define OP_Copy 78 /* synopsis: r[P2@P3+1]=r[P1@P3+1] */ +#define OP_SCopy 79 /* synopsis: r[P2]=r[P1] */ +#define OP_IntCopy 80 /* synopsis: r[P2]=r[P1] */ +#define OP_ResultRow 81 /* synopsis: output=r[P1@P2] */ +#define OP_CollSeq 82 +#define OP_AddImm 83 /* synopsis: r[P1]=r[P1]+P2 */ +#define OP_RealAffinity 84 +#define OP_Cast 85 /* synopsis: affinity(r[P1]) */ +#define OP_Permutation 86 +#define OP_Compare 87 /* synopsis: r[P1@P3] <-> r[P2@P3] */ +#define OP_IsTrue 88 /* synopsis: r[P2] = coalesce(r[P1]==TRUE,P3) ^ P4 */ +#define OP_Offset 89 /* synopsis: r[P3] = sqlite_offset(P1) */ +#define OP_Column 90 /* synopsis: r[P3]=PX */ +#define OP_Affinity 91 /* synopsis: affinity(r[P1@P2]) */ +#define OP_MakeRecord 92 /* synopsis: r[P3]=mkrec(r[P1@P2]) */ +#define OP_Count 93 /* synopsis: r[P2]=count() */ +#define OP_ReadCookie 94 +#define OP_SetCookie 95 +#define OP_ReopenIdx 96 /* synopsis: root=P2 iDb=P3 */ +#define OP_OpenRead 97 /* synopsis: root=P2 iDb=P3 */ +#define OP_OpenWrite 98 /* synopsis: root=P2 iDb=P3 */ +#define OP_BitAnd 99 /* same as TK_BITAND, synopsis: r[P3]=r[P1]&r[P2] */ +#define OP_BitOr 100 /* same as TK_BITOR, synopsis: r[P3]=r[P1]|r[P2] */ +#define OP_ShiftLeft 101 /* same as TK_LSHIFT, synopsis: r[P3]=r[P2]<>r[P1] */ +#define OP_Add 103 /* same as TK_PLUS, synopsis: r[P3]=r[P1]+r[P2] */ +#define OP_Subtract 104 /* same as TK_MINUS, synopsis: r[P3]=r[P2]-r[P1] */ +#define OP_Multiply 105 /* same as TK_STAR, synopsis: r[P3]=r[P1]*r[P2] */ +#define OP_Divide 106 /* same as TK_SLASH, synopsis: r[P3]=r[P2]/r[P1] */ +#define OP_Remainder 107 /* same as TK_REM, synopsis: r[P3]=r[P2]%r[P1] */ +#define OP_Concat 108 /* same as TK_CONCAT, synopsis: r[P3]=r[P2]+r[P1] */ +#define OP_OpenDup 109 +#define OP_BitNot 110 /* same as TK_BITNOT, synopsis: r[P2]= ~r[P1] */ +#define OP_OpenAutoindex 111 /* synopsis: nColumn=P2 */ +#define OP_OpenEphemeral 112 /* synopsis: nColumn=P2 */ +#define OP_String8 113 /* same as TK_STRING, synopsis: r[P2]='P4' */ +#define OP_SorterOpen 114 +#define OP_SequenceTest 115 /* synopsis: if( cursor[P1].ctr++ ) pc = P2 */ +#define OP_OpenPseudo 116 /* synopsis: P3 columns in r[P2] */ +#define OP_Close 117 +#define OP_ColumnsUsed 118 +#define OP_SeekHit 119 /* synopsis: seekHit=P2 */ +#define OP_Sequence 120 /* synopsis: r[P2]=cursor[P1].ctr++ */ +#define OP_NewRowid 121 /* synopsis: r[P2]=rowid */ +#define OP_Insert 122 /* synopsis: intkey=r[P3] data=r[P2] */ +#define OP_Delete 123 +#define OP_ResetCount 124 +#define OP_SorterCompare 125 /* synopsis: if key(P1)!=trim(r[P3],P4) goto P2 */ +#define OP_SorterData 126 /* synopsis: r[P2]=data */ +#define OP_RowData 127 /* synopsis: r[P2]=data */ +#define OP_Rowid 128 /* synopsis: r[P2]=rowid */ +#define OP_NullRow 129 +#define OP_SeekEnd 130 +#define OP_SorterInsert 131 /* synopsis: key=r[P2] */ +#define OP_IdxInsert 132 /* synopsis: key=r[P2] */ +#define OP_IdxDelete 133 /* synopsis: key=r[P2@P3] */ +#define OP_DeferredSeek 134 /* synopsis: Move P3 to P1.rowid if needed */ +#define OP_IdxRowid 135 /* synopsis: r[P2]=rowid */ +#define OP_Destroy 136 +#define OP_Clear 137 +#define OP_ResetSorter 138 +#define OP_CreateBtree 139 /* synopsis: r[P2]=root iDb=P1 flags=P3 */ +#define OP_SqlExec 140 +#define OP_ParseSchema 141 +#define OP_LoadAnalysis 142 +#define OP_DropTable 143 +#define OP_DropIndex 144 +#define OP_DropTrigger 145 +#define OP_IntegrityCk 146 +#define OP_RowSetAdd 147 /* synopsis: rowset(P1)=r[P2] */ +#define OP_Real 148 /* same as TK_FLOAT, synopsis: r[P2]=P4 */ +#define OP_Param 149 +#define OP_FkCounter 150 /* synopsis: fkctr[P1]+=P2 */ +#define OP_MemMax 151 /* synopsis: r[P1]=max(r[P1],r[P2]) */ +#define OP_OffsetLimit 152 /* synopsis: if r[P1]>0 then r[P2]=r[P1]+max(0,r[P3]) else r[P2]=(-1) */ +#define OP_AggInverse 153 /* synopsis: accum=r[P3] inverse(r[P2@P5]) */ +#define OP_AggStep 154 /* synopsis: accum=r[P3] step(r[P2@P5]) */ +#define OP_AggStep1 155 /* synopsis: accum=r[P3] step(r[P2@P5]) */ +#define OP_AggValue 156 /* synopsis: r[P3]=value N=P2 */ +#define OP_AggFinal 157 /* synopsis: accum=r[P1] N=P2 */ +#define OP_Expire 158 +#define OP_TableLock 159 /* synopsis: iDb=P1 root=P2 write=P3 */ +#define OP_VBegin 160 +#define OP_VCreate 161 +#define OP_VDestroy 162 +#define OP_VOpen 163 +#define OP_VColumn 164 /* synopsis: r[P3]=vcolumn(P2) */ +#define OP_VRename 165 +#define OP_Pagecount 166 +#define OP_MaxPgcnt 167 +#define OP_Trace 168 +#define OP_CursorHint 169 +#define OP_Noop 170 +#define OP_Explain 171 +#define OP_Abortable 172 + +/* Properties such as "out2" or "jump" that are specified in +** comments following the "case" for each opcode in the vdbe.c +** are encoded into bitvectors as follows: +*/ +#define OPFLG_JUMP 0x01 /* jump: P2 holds jmp target */ +#define OPFLG_IN1 0x02 /* in1: P1 is an input */ +#define OPFLG_IN2 0x04 /* in2: P2 is an input */ +#define OPFLG_IN3 0x08 /* in3: P3 is an input */ +#define OPFLG_OUT2 0x10 /* out2: P2 is an output */ +#define OPFLG_OUT3 0x20 /* out3: P3 is an output */ +#define OPFLG_INITIALIZER {\ +/* 0 */ 0x00, 0x00, 0x00, 0x01, 0x01, 0x01, 0x00, 0x10,\ +/* 8 */ 0x00, 0x01, 0x00, 0x01, 0x01, 0x01, 0x03, 0x03,\ +/* 16 */ 0x01, 0x01, 0x03, 0x12, 0x03, 0x01, 0x09, 0x09,\ +/* 24 */ 0x09, 0x09, 0x09, 0x09, 0x09, 0x09, 0x09, 0x09,\ +/* 32 */ 0x01, 0x01, 0x01, 0x01, 0x01, 0x01, 0x01, 0x01,\ +/* 40 */ 0x01, 0x23, 0x0b, 0x26, 0x26, 0x01, 0x01, 0x03,\ +/* 48 */ 0x03, 0x03, 0x03, 0x03, 0x0b, 0x0b, 0x0b, 0x0b,\ +/* 56 */ 0x0b, 0x0b, 0x01, 0x01, 0x01, 0x01, 0x00, 0x00,\ +/* 64 */ 0x00, 0x00, 0x02, 0x02, 0x08, 0x00, 0x10, 0x10,\ +/* 72 */ 0x10, 0x10, 0x00, 0x10, 0x10, 0x00, 0x00, 0x10,\ +/* 80 */ 0x10, 0x00, 0x00, 0x02, 0x02, 0x02, 0x00, 0x00,\ +/* 88 */ 0x12, 0x20, 0x00, 0x00, 0x00, 0x10, 0x10, 0x00,\ +/* 96 */ 0x00, 0x00, 0x00, 0x26, 0x26, 0x26, 0x26, 0x26,\ +/* 104 */ 0x26, 0x26, 0x26, 0x26, 0x26, 0x00, 0x12, 0x00,\ +/* 112 */ 0x00, 0x10, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,\ +/* 120 */ 0x10, 0x10, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,\ +/* 128 */ 0x10, 0x00, 0x00, 0x04, 0x04, 0x00, 0x00, 0x10,\ +/* 136 */ 0x10, 0x00, 0x00, 0x10, 0x00, 0x00, 0x00, 0x00,\ +/* 144 */ 0x00, 0x00, 0x00, 0x06, 0x10, 0x10, 0x00, 0x04,\ +/* 152 */ 0x1a, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,\ +/* 160 */ 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x10, 0x10,\ +/* 168 */ 0x00, 0x00, 0x00, 0x00, 0x00,} + +/* The sqlite3P2Values() routine is able to run faster if it knows +** the value of the largest JUMP opcode. The smaller the maximum +** JUMP opcode the better, so the mkopcodeh.tcl script that +** generated this include file strives to group all JUMP opcodes +** together near the beginning of the list. +*/ +#define SQLITE_MX_JUMP_OPCODE 61 /* Maximum JUMP opcode */ + +/************** End of opcodes.h *********************************************/ +/************** Continuing where we left off in vdbe.h ***********************/ + +/* +** Additional non-public SQLITE_PREPARE_* flags +*/ +#define SQLITE_PREPARE_SAVESQL 0x80 /* Preserve SQL text */ +#define SQLITE_PREPARE_MASK 0x0f /* Mask of public flags */ + +/* +** Prototypes for the VDBE interface. See comments on the implementation +** for a description of what each of these routines does. +*/ +SQLITE_PRIVATE Vdbe *sqlite3VdbeCreate(Parse*); +SQLITE_PRIVATE int sqlite3VdbeAddOp0(Vdbe*,int); +SQLITE_PRIVATE int sqlite3VdbeAddOp1(Vdbe*,int,int); +SQLITE_PRIVATE int sqlite3VdbeAddOp2(Vdbe*,int,int,int); +SQLITE_PRIVATE int sqlite3VdbeGoto(Vdbe*,int); +SQLITE_PRIVATE int sqlite3VdbeLoadString(Vdbe*,int,const char*); +SQLITE_PRIVATE void sqlite3VdbeMultiLoad(Vdbe*,int,const char*,...); +SQLITE_PRIVATE int sqlite3VdbeAddOp3(Vdbe*,int,int,int,int); +SQLITE_PRIVATE int sqlite3VdbeAddOp4(Vdbe*,int,int,int,int,const char *zP4,int); +SQLITE_PRIVATE int sqlite3VdbeAddOp4Dup8(Vdbe*,int,int,int,int,const u8*,int); +SQLITE_PRIVATE int sqlite3VdbeAddOp4Int(Vdbe*,int,int,int,int,int); +SQLITE_PRIVATE void sqlite3VdbeEndCoroutine(Vdbe*,int); +#if defined(SQLITE_DEBUG) && !defined(SQLITE_TEST_REALLOC_STRESS) +SQLITE_PRIVATE void sqlite3VdbeVerifyNoMallocRequired(Vdbe *p, int N); +SQLITE_PRIVATE void sqlite3VdbeVerifyNoResultRow(Vdbe *p); +#else +# define sqlite3VdbeVerifyNoMallocRequired(A,B) +# define sqlite3VdbeVerifyNoResultRow(A) +#endif +#if defined(SQLITE_DEBUG) +SQLITE_PRIVATE void sqlite3VdbeVerifyAbortable(Vdbe *p, int); +#else +# define sqlite3VdbeVerifyAbortable(A,B) +#endif +SQLITE_PRIVATE VdbeOp *sqlite3VdbeAddOpList(Vdbe*, int nOp, VdbeOpList const *aOp,int iLineno); +#ifndef SQLITE_OMIT_EXPLAIN +SQLITE_PRIVATE void sqlite3VdbeExplain(Parse*,u8,const char*,...); +SQLITE_PRIVATE void sqlite3VdbeExplainPop(Parse*); +SQLITE_PRIVATE int sqlite3VdbeExplainParent(Parse*); +# define ExplainQueryPlan(P) sqlite3VdbeExplain P +# define ExplainQueryPlanPop(P) sqlite3VdbeExplainPop(P) +# define ExplainQueryPlanParent(P) sqlite3VdbeExplainParent(P) +#else +# define ExplainQueryPlan(P) +# define ExplainQueryPlanPop(P) +# define ExplainQueryPlanParent(P) 0 +# define sqlite3ExplainBreakpoint(A,B) /*no-op*/ +#endif +#if defined(SQLITE_DEBUG) && !defined(SQLITE_OMIT_EXPLAIN) +SQLITE_PRIVATE void sqlite3ExplainBreakpoint(const char*,const char*); +#else +# define sqlite3ExplainBreakpoint(A,B) /*no-op*/ +#endif +SQLITE_PRIVATE void sqlite3VdbeAddParseSchemaOp(Vdbe*,int,char*); +SQLITE_PRIVATE void sqlite3VdbeChangeOpcode(Vdbe*, int addr, u8); +SQLITE_PRIVATE void sqlite3VdbeChangeP1(Vdbe*, int addr, int P1); +SQLITE_PRIVATE void sqlite3VdbeChangeP2(Vdbe*, int addr, int P2); +SQLITE_PRIVATE void sqlite3VdbeChangeP3(Vdbe*, int addr, int P3); +SQLITE_PRIVATE void sqlite3VdbeChangeP5(Vdbe*, u16 P5); +SQLITE_PRIVATE void sqlite3VdbeJumpHere(Vdbe*, int addr); +SQLITE_PRIVATE int sqlite3VdbeChangeToNoop(Vdbe*, int addr); +SQLITE_PRIVATE int sqlite3VdbeDeletePriorOpcode(Vdbe*, u8 op); +SQLITE_PRIVATE void sqlite3VdbeChangeP4(Vdbe*, int addr, const char *zP4, int N); +SQLITE_PRIVATE void sqlite3VdbeAppendP4(Vdbe*, void *pP4, int p4type); +SQLITE_PRIVATE void sqlite3VdbeSetP4KeyInfo(Parse*, Index*); +SQLITE_PRIVATE void sqlite3VdbeUsesBtree(Vdbe*, int); +SQLITE_PRIVATE VdbeOp *sqlite3VdbeGetOp(Vdbe*, int); +SQLITE_PRIVATE int sqlite3VdbeMakeLabel(Parse*); +SQLITE_PRIVATE void sqlite3VdbeRunOnlyOnce(Vdbe*); +SQLITE_PRIVATE void sqlite3VdbeReusable(Vdbe*); +SQLITE_PRIVATE void sqlite3VdbeDelete(Vdbe*); +SQLITE_PRIVATE void sqlite3VdbeClearObject(sqlite3*,Vdbe*); +SQLITE_PRIVATE void sqlite3VdbeMakeReady(Vdbe*,Parse*); +SQLITE_PRIVATE int sqlite3VdbeFinalize(Vdbe*); +SQLITE_PRIVATE void sqlite3VdbeResolveLabel(Vdbe*, int); +SQLITE_PRIVATE int sqlite3VdbeCurrentAddr(Vdbe*); +#ifdef SQLITE_DEBUG +SQLITE_PRIVATE int sqlite3VdbeAssertMayAbort(Vdbe *, int); +#endif +SQLITE_PRIVATE void sqlite3VdbeResetStepResult(Vdbe*); +SQLITE_PRIVATE void sqlite3VdbeRewind(Vdbe*); +SQLITE_PRIVATE int sqlite3VdbeReset(Vdbe*); +SQLITE_PRIVATE void sqlite3VdbeSetNumCols(Vdbe*,int); +SQLITE_PRIVATE int sqlite3VdbeSetColName(Vdbe*, int, int, const char *, void(*)(void*)); +SQLITE_PRIVATE void sqlite3VdbeCountChanges(Vdbe*); +SQLITE_PRIVATE sqlite3 *sqlite3VdbeDb(Vdbe*); +SQLITE_PRIVATE u8 sqlite3VdbePrepareFlags(Vdbe*); +SQLITE_PRIVATE void sqlite3VdbeSetSql(Vdbe*, const char *z, int n, u8); +#ifdef SQLITE_ENABLE_NORMALIZE +SQLITE_PRIVATE void sqlite3VdbeAddDblquoteStr(sqlite3*,Vdbe*,const char*); +SQLITE_PRIVATE int sqlite3VdbeUsesDoubleQuotedString(Vdbe*,const char*); +#endif +SQLITE_PRIVATE void sqlite3VdbeSwap(Vdbe*,Vdbe*); +SQLITE_PRIVATE VdbeOp *sqlite3VdbeTakeOpArray(Vdbe*, int*, int*); +SQLITE_PRIVATE sqlite3_value *sqlite3VdbeGetBoundValue(Vdbe*, int, u8); +SQLITE_PRIVATE void sqlite3VdbeSetVarmask(Vdbe*, int); +#ifndef SQLITE_OMIT_TRACE +SQLITE_PRIVATE char *sqlite3VdbeExpandSql(Vdbe*, const char*); +#endif +SQLITE_PRIVATE int sqlite3MemCompare(const Mem*, const Mem*, const CollSeq*); +SQLITE_PRIVATE int sqlite3BlobCompare(const Mem*, const Mem*); + +SQLITE_PRIVATE void sqlite3VdbeRecordUnpack(KeyInfo*,int,const void*,UnpackedRecord*); +SQLITE_PRIVATE int sqlite3VdbeRecordCompare(int,const void*,UnpackedRecord*); +SQLITE_PRIVATE int sqlite3VdbeRecordCompareWithSkip(int, const void *, UnpackedRecord *, int); +SQLITE_PRIVATE UnpackedRecord *sqlite3VdbeAllocUnpackedRecord(KeyInfo*); + +typedef int (*RecordCompare)(int,const void*,UnpackedRecord*); +SQLITE_PRIVATE RecordCompare sqlite3VdbeFindCompare(UnpackedRecord*); + +#ifndef SQLITE_OMIT_TRIGGER +SQLITE_PRIVATE void sqlite3VdbeLinkSubProgram(Vdbe *, SubProgram *); +#endif + +SQLITE_PRIVATE int sqlite3NotPureFunc(sqlite3_context*); + +/* Use SQLITE_ENABLE_COMMENTS to enable generation of extra comments on +** each VDBE opcode. +** +** Use the SQLITE_ENABLE_MODULE_COMMENTS macro to see some extra no-op +** comments in VDBE programs that show key decision points in the code +** generator. +*/ +#ifdef SQLITE_ENABLE_EXPLAIN_COMMENTS +SQLITE_PRIVATE void sqlite3VdbeComment(Vdbe*, const char*, ...); +# define VdbeComment(X) sqlite3VdbeComment X +SQLITE_PRIVATE void sqlite3VdbeNoopComment(Vdbe*, const char*, ...); +# define VdbeNoopComment(X) sqlite3VdbeNoopComment X +# ifdef SQLITE_ENABLE_MODULE_COMMENTS +# define VdbeModuleComment(X) sqlite3VdbeNoopComment X +# else +# define VdbeModuleComment(X) +# endif +#else +# define VdbeComment(X) +# define VdbeNoopComment(X) +# define VdbeModuleComment(X) +#endif + +/* +** The VdbeCoverage macros are used to set a coverage testing point +** for VDBE branch instructions. The coverage testing points are line +** numbers in the sqlite3.c source file. VDBE branch coverage testing +** only works with an amalagmation build. That's ok since a VDBE branch +** coverage build designed for testing the test suite only. No application +** should ever ship with VDBE branch coverage measuring turned on. +** +** VdbeCoverage(v) // Mark the previously coded instruction +** // as a branch +** +** VdbeCoverageIf(v, conditional) // Mark previous if conditional true +** +** VdbeCoverageAlwaysTaken(v) // Previous branch is always taken +** +** VdbeCoverageNeverTaken(v) // Previous branch is never taken +** +** VdbeCoverageNeverNull(v) // Previous three-way branch is only +** // taken on the first two ways. The +** // NULL option is not possible +** +** VdbeCoverageEqNe(v) // Previous OP_Jump is only interested +** // in distingishing equal and not-equal. +** +** Every VDBE branch operation must be tagged with one of the macros above. +** If not, then when "make test" is run with -DSQLITE_VDBE_COVERAGE and +** -DSQLITE_DEBUG then an ALWAYS() will fail in the vdbeTakeBranch() +** routine in vdbe.c, alerting the developer to the missed tag. +** +** During testing, the test application will invoke +** sqlite3_test_control(SQLITE_TESTCTRL_VDBE_COVERAGE,...) to set a callback +** routine that is invoked as each bytecode branch is taken. The callback +** contains the sqlite3.c source line number ov the VdbeCoverage macro and +** flags to indicate whether or not the branch was taken. The test application +** is responsible for keeping track of this and reporting byte-code branches +** that are never taken. +** +** See the VdbeBranchTaken() macro and vdbeTakeBranch() function in the +** vdbe.c source file for additional information. +*/ +#ifdef SQLITE_VDBE_COVERAGE +SQLITE_PRIVATE void sqlite3VdbeSetLineNumber(Vdbe*,int); +# define VdbeCoverage(v) sqlite3VdbeSetLineNumber(v,__LINE__) +# define VdbeCoverageIf(v,x) if(x)sqlite3VdbeSetLineNumber(v,__LINE__) +# define VdbeCoverageAlwaysTaken(v) \ + sqlite3VdbeSetLineNumber(v,__LINE__|0x5000000); +# define VdbeCoverageNeverTaken(v) \ + sqlite3VdbeSetLineNumber(v,__LINE__|0x6000000); +# define VdbeCoverageNeverNull(v) \ + sqlite3VdbeSetLineNumber(v,__LINE__|0x4000000); +# define VdbeCoverageNeverNullIf(v,x) \ + if(x)sqlite3VdbeSetLineNumber(v,__LINE__|0x4000000); +# define VdbeCoverageEqNe(v) \ + sqlite3VdbeSetLineNumber(v,__LINE__|0x8000000); +# define VDBE_OFFSET_LINENO(x) (__LINE__+x) +#else +# define VdbeCoverage(v) +# define VdbeCoverageIf(v,x) +# define VdbeCoverageAlwaysTaken(v) +# define VdbeCoverageNeverTaken(v) +# define VdbeCoverageNeverNull(v) +# define VdbeCoverageNeverNullIf(v,x) +# define VdbeCoverageEqNe(v) +# define VDBE_OFFSET_LINENO(x) 0 +#endif + +#ifdef SQLITE_ENABLE_STMT_SCANSTATUS +SQLITE_PRIVATE void sqlite3VdbeScanStatus(Vdbe*, int, int, int, LogEst, const char*); +#else +# define sqlite3VdbeScanStatus(a,b,c,d,e) +#endif + +#if defined(SQLITE_DEBUG) || defined(VDBE_PROFILE) +SQLITE_PRIVATE void sqlite3VdbePrintOp(FILE*, int, VdbeOp*); +#endif + +#endif /* SQLITE_VDBE_H */ + +/************** End of vdbe.h ************************************************/ +/************** Continuing where we left off in sqliteInt.h ******************/ +/************** Include pager.h in the middle of sqliteInt.h *****************/ +/************** Begin file pager.h *******************************************/ +/* +** 2001 September 15 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** This header file defines the interface that the sqlite page cache +** subsystem. The page cache subsystem reads and writes a file a page +** at a time and provides a journal for rollback. +*/ + +#ifndef SQLITE_PAGER_H +#define SQLITE_PAGER_H + +/* +** Default maximum size for persistent journal files. A negative +** value means no limit. This value may be overridden using the +** sqlite3PagerJournalSizeLimit() API. See also "PRAGMA journal_size_limit". +*/ +#ifndef SQLITE_DEFAULT_JOURNAL_SIZE_LIMIT + #define SQLITE_DEFAULT_JOURNAL_SIZE_LIMIT -1 +#endif + +/* +** The type used to represent a page number. The first page in a file +** is called page 1. 0 is used to represent "not a page". +*/ +typedef u32 Pgno; + +/* +** Each open file is managed by a separate instance of the "Pager" structure. +*/ +typedef struct Pager Pager; + +/* +** Handle type for pages. +*/ +typedef struct PgHdr DbPage; + +/* +** Page number PAGER_MJ_PGNO is never used in an SQLite database (it is +** reserved for working around a windows/posix incompatibility). It is +** used in the journal to signify that the remainder of the journal file +** is devoted to storing a master journal name - there are no more pages to +** roll back. See comments for function writeMasterJournal() in pager.c +** for details. +*/ +#define PAGER_MJ_PGNO(x) ((Pgno)((PENDING_BYTE/((x)->pageSize))+1)) + +/* +** Allowed values for the flags parameter to sqlite3PagerOpen(). +** +** NOTE: These values must match the corresponding BTREE_ values in btree.h. +*/ +#define PAGER_OMIT_JOURNAL 0x0001 /* Do not use a rollback journal */ +#define PAGER_MEMORY 0x0002 /* In-memory database */ + +/* +** Valid values for the second argument to sqlite3PagerLockingMode(). +*/ +#define PAGER_LOCKINGMODE_QUERY -1 +#define PAGER_LOCKINGMODE_NORMAL 0 +#define PAGER_LOCKINGMODE_EXCLUSIVE 1 + +/* +** Numeric constants that encode the journalmode. +** +** The numeric values encoded here (other than PAGER_JOURNALMODE_QUERY) +** are exposed in the API via the "PRAGMA journal_mode" command and +** therefore cannot be changed without a compatibility break. +*/ +#define PAGER_JOURNALMODE_QUERY (-1) /* Query the value of journalmode */ +#define PAGER_JOURNALMODE_DELETE 0 /* Commit by deleting journal file */ +#define PAGER_JOURNALMODE_PERSIST 1 /* Commit by zeroing journal header */ +#define PAGER_JOURNALMODE_OFF 2 /* Journal omitted. */ +#define PAGER_JOURNALMODE_TRUNCATE 3 /* Commit by truncating journal */ +#define PAGER_JOURNALMODE_MEMORY 4 /* In-memory journal file */ +#define PAGER_JOURNALMODE_WAL 5 /* Use write-ahead logging */ + +/* +** Flags that make up the mask passed to sqlite3PagerGet(). +*/ +#define PAGER_GET_NOCONTENT 0x01 /* Do not load data from disk */ +#define PAGER_GET_READONLY 0x02 /* Read-only page is acceptable */ + +/* +** Flags for sqlite3PagerSetFlags() +** +** Value constraints (enforced via assert()): +** PAGER_FULLFSYNC == SQLITE_FullFSync +** PAGER_CKPT_FULLFSYNC == SQLITE_CkptFullFSync +** PAGER_CACHE_SPILL == SQLITE_CacheSpill +*/ +#define PAGER_SYNCHRONOUS_OFF 0x01 /* PRAGMA synchronous=OFF */ +#define PAGER_SYNCHRONOUS_NORMAL 0x02 /* PRAGMA synchronous=NORMAL */ +#define PAGER_SYNCHRONOUS_FULL 0x03 /* PRAGMA synchronous=FULL */ +#define PAGER_SYNCHRONOUS_EXTRA 0x04 /* PRAGMA synchronous=EXTRA */ +#define PAGER_SYNCHRONOUS_MASK 0x07 /* Mask for four values above */ +#define PAGER_FULLFSYNC 0x08 /* PRAGMA fullfsync=ON */ +#define PAGER_CKPT_FULLFSYNC 0x10 /* PRAGMA checkpoint_fullfsync=ON */ +#define PAGER_CACHESPILL 0x20 /* PRAGMA cache_spill=ON */ +#define PAGER_FLAGS_MASK 0x38 /* All above except SYNCHRONOUS */ + +/* +** The remainder of this file contains the declarations of the functions +** that make up the Pager sub-system API. See source code comments for +** a detailed description of each routine. +*/ + +/* Open and close a Pager connection. */ +SQLITE_PRIVATE int sqlite3PagerOpen( + sqlite3_vfs*, + Pager **ppPager, + const char*, + int, + int, + int, + void(*)(DbPage*) +); +SQLITE_PRIVATE int sqlite3PagerClose(Pager *pPager, sqlite3*); +SQLITE_PRIVATE int sqlite3PagerReadFileheader(Pager*, int, unsigned char*); + +/* Functions used to configure a Pager object. */ +SQLITE_PRIVATE void sqlite3PagerSetBusyHandler(Pager*, int(*)(void *), void *); +SQLITE_PRIVATE int sqlite3PagerSetPagesize(Pager*, u32*, int); +#ifdef SQLITE_HAS_CODEC +SQLITE_PRIVATE void sqlite3PagerAlignReserve(Pager*,Pager*); +#endif +SQLITE_PRIVATE int sqlite3PagerMaxPageCount(Pager*, int); +SQLITE_PRIVATE void sqlite3PagerSetCachesize(Pager*, int); +SQLITE_PRIVATE int sqlite3PagerSetSpillsize(Pager*, int); +SQLITE_PRIVATE void sqlite3PagerSetMmapLimit(Pager *, sqlite3_int64); +SQLITE_PRIVATE void sqlite3PagerShrink(Pager*); +SQLITE_PRIVATE void sqlite3PagerSetFlags(Pager*,unsigned); +SQLITE_PRIVATE int sqlite3PagerLockingMode(Pager *, int); +SQLITE_PRIVATE int sqlite3PagerSetJournalMode(Pager *, int); +SQLITE_PRIVATE int sqlite3PagerGetJournalMode(Pager*); +SQLITE_PRIVATE int sqlite3PagerOkToChangeJournalMode(Pager*); +SQLITE_PRIVATE i64 sqlite3PagerJournalSizeLimit(Pager *, i64); +SQLITE_PRIVATE sqlite3_backup **sqlite3PagerBackupPtr(Pager*); +SQLITE_PRIVATE int sqlite3PagerFlush(Pager*); + +/* Functions used to obtain and release page references. */ +SQLITE_PRIVATE int sqlite3PagerGet(Pager *pPager, Pgno pgno, DbPage **ppPage, int clrFlag); +SQLITE_PRIVATE DbPage *sqlite3PagerLookup(Pager *pPager, Pgno pgno); +SQLITE_PRIVATE void sqlite3PagerRef(DbPage*); +SQLITE_PRIVATE void sqlite3PagerUnref(DbPage*); +SQLITE_PRIVATE void sqlite3PagerUnrefNotNull(DbPage*); +SQLITE_PRIVATE void sqlite3PagerUnrefPageOne(DbPage*); + +/* Operations on page references. */ +SQLITE_PRIVATE int sqlite3PagerWrite(DbPage*); +SQLITE_PRIVATE void sqlite3PagerDontWrite(DbPage*); +SQLITE_PRIVATE int sqlite3PagerMovepage(Pager*,DbPage*,Pgno,int); +SQLITE_PRIVATE int sqlite3PagerPageRefcount(DbPage*); +SQLITE_PRIVATE void *sqlite3PagerGetData(DbPage *); +SQLITE_PRIVATE void *sqlite3PagerGetExtra(DbPage *); + +/* Functions used to manage pager transactions and savepoints. */ +SQLITE_PRIVATE void sqlite3PagerPagecount(Pager*, int*); +SQLITE_PRIVATE int sqlite3PagerBegin(Pager*, int exFlag, int); +SQLITE_PRIVATE int sqlite3PagerCommitPhaseOne(Pager*,const char *zMaster, int); +SQLITE_PRIVATE int sqlite3PagerExclusiveLock(Pager*); +SQLITE_PRIVATE int sqlite3PagerSync(Pager *pPager, const char *zMaster); +SQLITE_PRIVATE int sqlite3PagerCommitPhaseTwo(Pager*); +SQLITE_PRIVATE int sqlite3PagerRollback(Pager*); +SQLITE_PRIVATE int sqlite3PagerOpenSavepoint(Pager *pPager, int n); +SQLITE_PRIVATE int sqlite3PagerSavepoint(Pager *pPager, int op, int iSavepoint); +SQLITE_PRIVATE int sqlite3PagerSharedLock(Pager *pPager); + +#ifndef SQLITE_OMIT_WAL +SQLITE_PRIVATE int sqlite3PagerCheckpoint(Pager *pPager, sqlite3*, int, int*, int*); +SQLITE_PRIVATE int sqlite3PagerWalSupported(Pager *pPager); +SQLITE_PRIVATE int sqlite3PagerWalCallback(Pager *pPager); +SQLITE_PRIVATE int sqlite3PagerOpenWal(Pager *pPager, int *pisOpen); +SQLITE_PRIVATE int sqlite3PagerCloseWal(Pager *pPager, sqlite3*); +# ifdef SQLITE_ENABLE_SNAPSHOT +SQLITE_PRIVATE int sqlite3PagerSnapshotGet(Pager *pPager, sqlite3_snapshot **ppSnapshot); +SQLITE_PRIVATE int sqlite3PagerSnapshotOpen(Pager *pPager, sqlite3_snapshot *pSnapshot); +SQLITE_PRIVATE int sqlite3PagerSnapshotRecover(Pager *pPager); +SQLITE_PRIVATE int sqlite3PagerSnapshotCheck(Pager *pPager, sqlite3_snapshot *pSnapshot); +SQLITE_PRIVATE void sqlite3PagerSnapshotUnlock(Pager *pPager); +# endif +#endif + +#ifdef SQLITE_DIRECT_OVERFLOW_READ +SQLITE_PRIVATE int sqlite3PagerDirectReadOk(Pager *pPager, Pgno pgno); +#endif + +#ifdef SQLITE_ENABLE_ZIPVFS +SQLITE_PRIVATE int sqlite3PagerWalFramesize(Pager *pPager); +#endif + +/* Functions used to query pager state and configuration. */ +SQLITE_PRIVATE u8 sqlite3PagerIsreadonly(Pager*); +SQLITE_PRIVATE u32 sqlite3PagerDataVersion(Pager*); +#ifdef SQLITE_DEBUG +SQLITE_PRIVATE int sqlite3PagerRefcount(Pager*); +#endif +SQLITE_PRIVATE int sqlite3PagerMemUsed(Pager*); +SQLITE_PRIVATE const char *sqlite3PagerFilename(Pager*, int); +SQLITE_PRIVATE sqlite3_vfs *sqlite3PagerVfs(Pager*); +SQLITE_PRIVATE sqlite3_file *sqlite3PagerFile(Pager*); +SQLITE_PRIVATE sqlite3_file *sqlite3PagerJrnlFile(Pager*); +SQLITE_PRIVATE const char *sqlite3PagerJournalname(Pager*); +SQLITE_PRIVATE void *sqlite3PagerTempSpace(Pager*); +SQLITE_PRIVATE int sqlite3PagerIsMemdb(Pager*); +SQLITE_PRIVATE void sqlite3PagerCacheStat(Pager *, int, int, int *); +SQLITE_PRIVATE void sqlite3PagerClearCache(Pager*); +SQLITE_PRIVATE int sqlite3SectorSize(sqlite3_file *); +#ifdef SQLITE_ENABLE_SETLK_TIMEOUT +SQLITE_PRIVATE void sqlite3PagerResetLockTimeout(Pager *pPager); +#else +# define sqlite3PagerResetLockTimeout(X) +#endif + +/* Functions used to truncate the database file. */ +SQLITE_PRIVATE void sqlite3PagerTruncateImage(Pager*,Pgno); + +SQLITE_PRIVATE void sqlite3PagerRekey(DbPage*, Pgno, u16); + +#if defined(SQLITE_HAS_CODEC) && !defined(SQLITE_OMIT_WAL) +SQLITE_PRIVATE void *sqlite3PagerCodec(DbPage *); +#endif + +/* Functions to support testing and debugging. */ +#if !defined(NDEBUG) || defined(SQLITE_TEST) +SQLITE_PRIVATE Pgno sqlite3PagerPagenumber(DbPage*); +SQLITE_PRIVATE int sqlite3PagerIswriteable(DbPage*); +#endif +#ifdef SQLITE_TEST +SQLITE_PRIVATE int *sqlite3PagerStats(Pager*); +SQLITE_PRIVATE void sqlite3PagerRefdump(Pager*); + void disable_simulated_io_errors(void); + void enable_simulated_io_errors(void); +#else +# define disable_simulated_io_errors() +# define enable_simulated_io_errors() +#endif + +#endif /* SQLITE_PAGER_H */ + +/************** End of pager.h ***********************************************/ +/************** Continuing where we left off in sqliteInt.h ******************/ +/************** Include pcache.h in the middle of sqliteInt.h ****************/ +/************** Begin file pcache.h ******************************************/ +/* +** 2008 August 05 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** This header file defines the interface that the sqlite page cache +** subsystem. +*/ + +#ifndef _PCACHE_H_ + +typedef struct PgHdr PgHdr; +typedef struct PCache PCache; + +/* +** Every page in the cache is controlled by an instance of the following +** structure. +*/ +struct PgHdr { + sqlite3_pcache_page *pPage; /* Pcache object page handle */ + void *pData; /* Page data */ + void *pExtra; /* Extra content */ + PCache *pCache; /* PRIVATE: Cache that owns this page */ + PgHdr *pDirty; /* Transient list of dirty sorted by pgno */ + Pager *pPager; /* The pager this page is part of */ + Pgno pgno; /* Page number for this page */ +#ifdef SQLITE_CHECK_PAGES + u32 pageHash; /* Hash of page content */ +#endif + u16 flags; /* PGHDR flags defined below */ + + /********************************************************************** + ** Elements above, except pCache, are public. All that follow are + ** private to pcache.c and should not be accessed by other modules. + ** pCache is grouped with the public elements for efficiency. + */ + i16 nRef; /* Number of users of this page */ + PgHdr *pDirtyNext; /* Next element in list of dirty pages */ + PgHdr *pDirtyPrev; /* Previous element in list of dirty pages */ + /* NB: pDirtyNext and pDirtyPrev are undefined if the + ** PgHdr object is not dirty */ +}; + +/* Bit values for PgHdr.flags */ +#define PGHDR_CLEAN 0x001 /* Page not on the PCache.pDirty list */ +#define PGHDR_DIRTY 0x002 /* Page is on the PCache.pDirty list */ +#define PGHDR_WRITEABLE 0x004 /* Journaled and ready to modify */ +#define PGHDR_NEED_SYNC 0x008 /* Fsync the rollback journal before + ** writing this page to the database */ +#define PGHDR_DONT_WRITE 0x010 /* Do not write content to disk */ +#define PGHDR_MMAP 0x020 /* This is an mmap page object */ + +#define PGHDR_WAL_APPEND 0x040 /* Appended to wal file */ + +/* Initialize and shutdown the page cache subsystem */ +SQLITE_PRIVATE int sqlite3PcacheInitialize(void); +SQLITE_PRIVATE void sqlite3PcacheShutdown(void); + +/* Page cache buffer management: +** These routines implement SQLITE_CONFIG_PAGECACHE. +*/ +SQLITE_PRIVATE void sqlite3PCacheBufferSetup(void *, int sz, int n); + +/* Create a new pager cache. +** Under memory stress, invoke xStress to try to make pages clean. +** Only clean and unpinned pages can be reclaimed. +*/ +SQLITE_PRIVATE int sqlite3PcacheOpen( + int szPage, /* Size of every page */ + int szExtra, /* Extra space associated with each page */ + int bPurgeable, /* True if pages are on backing store */ + int (*xStress)(void*, PgHdr*), /* Call to try to make pages clean */ + void *pStress, /* Argument to xStress */ + PCache *pToInit /* Preallocated space for the PCache */ +); + +/* Modify the page-size after the cache has been created. */ +SQLITE_PRIVATE int sqlite3PcacheSetPageSize(PCache *, int); + +/* Return the size in bytes of a PCache object. Used to preallocate +** storage space. +*/ +SQLITE_PRIVATE int sqlite3PcacheSize(void); + +/* One release per successful fetch. Page is pinned until released. +** Reference counted. +*/ +SQLITE_PRIVATE sqlite3_pcache_page *sqlite3PcacheFetch(PCache*, Pgno, int createFlag); +SQLITE_PRIVATE int sqlite3PcacheFetchStress(PCache*, Pgno, sqlite3_pcache_page**); +SQLITE_PRIVATE PgHdr *sqlite3PcacheFetchFinish(PCache*, Pgno, sqlite3_pcache_page *pPage); +SQLITE_PRIVATE void sqlite3PcacheRelease(PgHdr*); + +SQLITE_PRIVATE void sqlite3PcacheDrop(PgHdr*); /* Remove page from cache */ +SQLITE_PRIVATE void sqlite3PcacheMakeDirty(PgHdr*); /* Make sure page is marked dirty */ +SQLITE_PRIVATE void sqlite3PcacheMakeClean(PgHdr*); /* Mark a single page as clean */ +SQLITE_PRIVATE void sqlite3PcacheCleanAll(PCache*); /* Mark all dirty list pages as clean */ +SQLITE_PRIVATE void sqlite3PcacheClearWritable(PCache*); + +/* Change a page number. Used by incr-vacuum. */ +SQLITE_PRIVATE void sqlite3PcacheMove(PgHdr*, Pgno); + +/* Remove all pages with pgno>x. Reset the cache if x==0 */ +SQLITE_PRIVATE void sqlite3PcacheTruncate(PCache*, Pgno x); + +/* Get a list of all dirty pages in the cache, sorted by page number */ +SQLITE_PRIVATE PgHdr *sqlite3PcacheDirtyList(PCache*); + +/* Reset and close the cache object */ +SQLITE_PRIVATE void sqlite3PcacheClose(PCache*); + +/* Clear flags from pages of the page cache */ +SQLITE_PRIVATE void sqlite3PcacheClearSyncFlags(PCache *); + +/* Discard the contents of the cache */ +SQLITE_PRIVATE void sqlite3PcacheClear(PCache*); + +/* Return the total number of outstanding page references */ +SQLITE_PRIVATE int sqlite3PcacheRefCount(PCache*); + +/* Increment the reference count of an existing page */ +SQLITE_PRIVATE void sqlite3PcacheRef(PgHdr*); + +SQLITE_PRIVATE int sqlite3PcachePageRefcount(PgHdr*); + +/* Return the total number of pages stored in the cache */ +SQLITE_PRIVATE int sqlite3PcachePagecount(PCache*); + +#if defined(SQLITE_CHECK_PAGES) || defined(SQLITE_DEBUG) +/* Iterate through all dirty pages currently stored in the cache. This +** interface is only available if SQLITE_CHECK_PAGES is defined when the +** library is built. +*/ +SQLITE_PRIVATE void sqlite3PcacheIterateDirty(PCache *pCache, void (*xIter)(PgHdr *)); +#endif + +#if defined(SQLITE_DEBUG) +/* Check invariants on a PgHdr object */ +SQLITE_PRIVATE int sqlite3PcachePageSanity(PgHdr*); +#endif + +/* Set and get the suggested cache-size for the specified pager-cache. +** +** If no global maximum is configured, then the system attempts to limit +** the total number of pages cached by purgeable pager-caches to the sum +** of the suggested cache-sizes. +*/ +SQLITE_PRIVATE void sqlite3PcacheSetCachesize(PCache *, int); +#ifdef SQLITE_TEST +SQLITE_PRIVATE int sqlite3PcacheGetCachesize(PCache *); +#endif + +/* Set or get the suggested spill-size for the specified pager-cache. +** +** The spill-size is the minimum number of pages in cache before the cache +** will attempt to spill dirty pages by calling xStress. +*/ +SQLITE_PRIVATE int sqlite3PcacheSetSpillsize(PCache *, int); + +/* Free up as much memory as possible from the page cache */ +SQLITE_PRIVATE void sqlite3PcacheShrink(PCache*); + +#ifdef SQLITE_ENABLE_MEMORY_MANAGEMENT +/* Try to return memory used by the pcache module to the main memory heap */ +SQLITE_PRIVATE int sqlite3PcacheReleaseMemory(int); +#endif + +#ifdef SQLITE_TEST +SQLITE_PRIVATE void sqlite3PcacheStats(int*,int*,int*,int*); +#endif + +SQLITE_PRIVATE void sqlite3PCacheSetDefault(void); + +/* Return the header size */ +SQLITE_PRIVATE int sqlite3HeaderSizePcache(void); +SQLITE_PRIVATE int sqlite3HeaderSizePcache1(void); + +/* Number of dirty pages as a percentage of the configured cache size */ +SQLITE_PRIVATE int sqlite3PCachePercentDirty(PCache*); + +#ifdef SQLITE_DIRECT_OVERFLOW_READ +SQLITE_PRIVATE int sqlite3PCacheIsDirty(PCache *pCache); +#endif + +#endif /* _PCACHE_H_ */ + +/************** End of pcache.h **********************************************/ +/************** Continuing where we left off in sqliteInt.h ******************/ +/************** Include os.h in the middle of sqliteInt.h ********************/ +/************** Begin file os.h **********************************************/ +/* +** 2001 September 16 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +****************************************************************************** +** +** This header file (together with is companion C source-code file +** "os.c") attempt to abstract the underlying operating system so that +** the SQLite library will work on both POSIX and windows systems. +** +** This header file is #include-ed by sqliteInt.h and thus ends up +** being included by every source file. +*/ +#ifndef _SQLITE_OS_H_ +#define _SQLITE_OS_H_ + +/* +** Attempt to automatically detect the operating system and setup the +** necessary pre-processor macros for it. +*/ +/************** Include os_setup.h in the middle of os.h *********************/ +/************** Begin file os_setup.h ****************************************/ +/* +** 2013 November 25 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +****************************************************************************** +** +** This file contains pre-processor directives related to operating system +** detection and/or setup. +*/ +#ifndef SQLITE_OS_SETUP_H +#define SQLITE_OS_SETUP_H + +/* +** Figure out if we are dealing with Unix, Windows, or some other operating +** system. +** +** After the following block of preprocess macros, all of SQLITE_OS_UNIX, +** SQLITE_OS_WIN, and SQLITE_OS_OTHER will defined to either 1 or 0. One of +** the three will be 1. The other two will be 0. +*/ +#if defined(SQLITE_OS_OTHER) +# if SQLITE_OS_OTHER==1 +# undef SQLITE_OS_UNIX +# define SQLITE_OS_UNIX 0 +# undef SQLITE_OS_WIN +# define SQLITE_OS_WIN 0 +# else +# undef SQLITE_OS_OTHER +# endif +#endif +#if !defined(SQLITE_OS_UNIX) && !defined(SQLITE_OS_OTHER) +# define SQLITE_OS_OTHER 0 +# ifndef SQLITE_OS_WIN +# if defined(_WIN32) || defined(WIN32) || defined(__CYGWIN__) || \ + defined(__MINGW32__) || defined(__BORLANDC__) +# define SQLITE_OS_WIN 1 +# define SQLITE_OS_UNIX 0 +# else +# define SQLITE_OS_WIN 0 +# define SQLITE_OS_UNIX 1 +# endif +# else +# define SQLITE_OS_UNIX 0 +# endif +#else +# ifndef SQLITE_OS_WIN +# define SQLITE_OS_WIN 0 +# endif +#endif + +#endif /* SQLITE_OS_SETUP_H */ + +/************** End of os_setup.h ********************************************/ +/************** Continuing where we left off in os.h *************************/ + +/* If the SET_FULLSYNC macro is not defined above, then make it +** a no-op +*/ +#ifndef SET_FULLSYNC +# define SET_FULLSYNC(x,y) +#endif + +/* +** The default size of a disk sector +*/ +#ifndef SQLITE_DEFAULT_SECTOR_SIZE +# define SQLITE_DEFAULT_SECTOR_SIZE 4096 +#endif + +/* +** Temporary files are named starting with this prefix followed by 16 random +** alphanumeric characters, and no file extension. They are stored in the +** OS's standard temporary file directory, and are deleted prior to exit. +** If sqlite is being embedded in another program, you may wish to change the +** prefix to reflect your program's name, so that if your program exits +** prematurely, old temporary files can be easily identified. This can be done +** using -DSQLITE_TEMP_FILE_PREFIX=myprefix_ on the compiler command line. +** +** 2006-10-31: The default prefix used to be "sqlite_". But then +** Mcafee started using SQLite in their anti-virus product and it +** started putting files with the "sqlite" name in the c:/temp folder. +** This annoyed many windows users. Those users would then do a +** Google search for "sqlite", find the telephone numbers of the +** developers and call to wake them up at night and complain. +** For this reason, the default name prefix is changed to be "sqlite" +** spelled backwards. So the temp files are still identified, but +** anybody smart enough to figure out the code is also likely smart +** enough to know that calling the developer will not help get rid +** of the file. +*/ +#ifndef SQLITE_TEMP_FILE_PREFIX +# define SQLITE_TEMP_FILE_PREFIX "etilqs_" +#endif + +/* +** The following values may be passed as the second argument to +** sqlite3OsLock(). The various locks exhibit the following semantics: +** +** SHARED: Any number of processes may hold a SHARED lock simultaneously. +** RESERVED: A single process may hold a RESERVED lock on a file at +** any time. Other processes may hold and obtain new SHARED locks. +** PENDING: A single process may hold a PENDING lock on a file at +** any one time. Existing SHARED locks may persist, but no new +** SHARED locks may be obtained by other processes. +** EXCLUSIVE: An EXCLUSIVE lock precludes all other locks. +** +** PENDING_LOCK may not be passed directly to sqlite3OsLock(). Instead, a +** process that requests an EXCLUSIVE lock may actually obtain a PENDING +** lock. This can be upgraded to an EXCLUSIVE lock by a subsequent call to +** sqlite3OsLock(). +*/ +#define NO_LOCK 0 +#define SHARED_LOCK 1 +#define RESERVED_LOCK 2 +#define PENDING_LOCK 3 +#define EXCLUSIVE_LOCK 4 + +/* +** File Locking Notes: (Mostly about windows but also some info for Unix) +** +** We cannot use LockFileEx() or UnlockFileEx() on Win95/98/ME because +** those functions are not available. So we use only LockFile() and +** UnlockFile(). +** +** LockFile() prevents not just writing but also reading by other processes. +** A SHARED_LOCK is obtained by locking a single randomly-chosen +** byte out of a specific range of bytes. The lock byte is obtained at +** random so two separate readers can probably access the file at the +** same time, unless they are unlucky and choose the same lock byte. +** An EXCLUSIVE_LOCK is obtained by locking all bytes in the range. +** There can only be one writer. A RESERVED_LOCK is obtained by locking +** a single byte of the file that is designated as the reserved lock byte. +** A PENDING_LOCK is obtained by locking a designated byte different from +** the RESERVED_LOCK byte. +** +** On WinNT/2K/XP systems, LockFileEx() and UnlockFileEx() are available, +** which means we can use reader/writer locks. When reader/writer locks +** are used, the lock is placed on the same range of bytes that is used +** for probabilistic locking in Win95/98/ME. Hence, the locking scheme +** will support two or more Win95 readers or two or more WinNT readers. +** But a single Win95 reader will lock out all WinNT readers and a single +** WinNT reader will lock out all other Win95 readers. +** +** The following #defines specify the range of bytes used for locking. +** SHARED_SIZE is the number of bytes available in the pool from which +** a random byte is selected for a shared lock. The pool of bytes for +** shared locks begins at SHARED_FIRST. +** +** The same locking strategy and +** byte ranges are used for Unix. This leaves open the possibility of having +** clients on win95, winNT, and unix all talking to the same shared file +** and all locking correctly. To do so would require that samba (or whatever +** tool is being used for file sharing) implements locks correctly between +** windows and unix. I'm guessing that isn't likely to happen, but by +** using the same locking range we are at least open to the possibility. +** +** Locking in windows is manditory. For this reason, we cannot store +** actual data in the bytes used for locking. The pager never allocates +** the pages involved in locking therefore. SHARED_SIZE is selected so +** that all locks will fit on a single page even at the minimum page size. +** PENDING_BYTE defines the beginning of the locks. By default PENDING_BYTE +** is set high so that we don't have to allocate an unused page except +** for very large databases. But one should test the page skipping logic +** by setting PENDING_BYTE low and running the entire regression suite. +** +** Changing the value of PENDING_BYTE results in a subtly incompatible +** file format. Depending on how it is changed, you might not notice +** the incompatibility right away, even running a full regression test. +** The default location of PENDING_BYTE is the first byte past the +** 1GB boundary. +** +*/ +#ifdef SQLITE_OMIT_WSD +# define PENDING_BYTE (0x40000000) +#else +# define PENDING_BYTE sqlite3PendingByte +#endif +#define RESERVED_BYTE (PENDING_BYTE+1) +#define SHARED_FIRST (PENDING_BYTE+2) +#define SHARED_SIZE 510 + +/* +** Wrapper around OS specific sqlite3_os_init() function. +*/ +SQLITE_PRIVATE int sqlite3OsInit(void); + +/* +** Functions for accessing sqlite3_file methods +*/ +SQLITE_PRIVATE void sqlite3OsClose(sqlite3_file*); +SQLITE_PRIVATE int sqlite3OsRead(sqlite3_file*, void*, int amt, i64 offset); +SQLITE_PRIVATE int sqlite3OsWrite(sqlite3_file*, const void*, int amt, i64 offset); +SQLITE_PRIVATE int sqlite3OsTruncate(sqlite3_file*, i64 size); +SQLITE_PRIVATE int sqlite3OsSync(sqlite3_file*, int); +SQLITE_PRIVATE int sqlite3OsFileSize(sqlite3_file*, i64 *pSize); +SQLITE_PRIVATE int sqlite3OsLock(sqlite3_file*, int); +SQLITE_PRIVATE int sqlite3OsUnlock(sqlite3_file*, int); +SQLITE_PRIVATE int sqlite3OsCheckReservedLock(sqlite3_file *id, int *pResOut); +SQLITE_PRIVATE int sqlite3OsFileControl(sqlite3_file*,int,void*); +SQLITE_PRIVATE void sqlite3OsFileControlHint(sqlite3_file*,int,void*); +#define SQLITE_FCNTL_DB_UNCHANGED 0xca093fa0 +SQLITE_PRIVATE int sqlite3OsSectorSize(sqlite3_file *id); +SQLITE_PRIVATE int sqlite3OsDeviceCharacteristics(sqlite3_file *id); +#ifndef SQLITE_OMIT_WAL +SQLITE_PRIVATE int sqlite3OsShmMap(sqlite3_file *,int,int,int,void volatile **); +SQLITE_PRIVATE int sqlite3OsShmLock(sqlite3_file *id, int, int, int); +SQLITE_PRIVATE void sqlite3OsShmBarrier(sqlite3_file *id); +SQLITE_PRIVATE int sqlite3OsShmUnmap(sqlite3_file *id, int); +#endif /* SQLITE_OMIT_WAL */ +SQLITE_PRIVATE int sqlite3OsFetch(sqlite3_file *id, i64, int, void **); +SQLITE_PRIVATE int sqlite3OsUnfetch(sqlite3_file *, i64, void *); + + +/* +** Functions for accessing sqlite3_vfs methods +*/ +SQLITE_PRIVATE int sqlite3OsOpen(sqlite3_vfs *, const char *, sqlite3_file*, int, int *); +SQLITE_PRIVATE int sqlite3OsDelete(sqlite3_vfs *, const char *, int); +SQLITE_PRIVATE int sqlite3OsAccess(sqlite3_vfs *, const char *, int, int *pResOut); +SQLITE_PRIVATE int sqlite3OsFullPathname(sqlite3_vfs *, const char *, int, char *); +#ifndef SQLITE_OMIT_LOAD_EXTENSION +SQLITE_PRIVATE void *sqlite3OsDlOpen(sqlite3_vfs *, const char *); +SQLITE_PRIVATE void sqlite3OsDlError(sqlite3_vfs *, int, char *); +SQLITE_PRIVATE void (*sqlite3OsDlSym(sqlite3_vfs *, void *, const char *))(void); +SQLITE_PRIVATE void sqlite3OsDlClose(sqlite3_vfs *, void *); +#endif /* SQLITE_OMIT_LOAD_EXTENSION */ +SQLITE_PRIVATE int sqlite3OsRandomness(sqlite3_vfs *, int, char *); +SQLITE_PRIVATE int sqlite3OsSleep(sqlite3_vfs *, int); +SQLITE_PRIVATE int sqlite3OsGetLastError(sqlite3_vfs*); +SQLITE_PRIVATE int sqlite3OsCurrentTimeInt64(sqlite3_vfs *, sqlite3_int64*); + +/* +** Convenience functions for opening and closing files using +** sqlite3_malloc() to obtain space for the file-handle structure. +*/ +SQLITE_PRIVATE int sqlite3OsOpenMalloc(sqlite3_vfs *, const char *, sqlite3_file **, int,int*); +SQLITE_PRIVATE void sqlite3OsCloseFree(sqlite3_file *); + +#endif /* _SQLITE_OS_H_ */ + +/************** End of os.h **************************************************/ +/************** Continuing where we left off in sqliteInt.h ******************/ +/************** Include mutex.h in the middle of sqliteInt.h *****************/ +/************** Begin file mutex.h *******************************************/ +/* +** 2007 August 28 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** +** This file contains the common header for all mutex implementations. +** The sqliteInt.h header #includes this file so that it is available +** to all source files. We break it out in an effort to keep the code +** better organized. +** +** NOTE: source files should *not* #include this header file directly. +** Source files should #include the sqliteInt.h file and let that file +** include this one indirectly. +*/ + + +/* +** Figure out what version of the code to use. The choices are +** +** SQLITE_MUTEX_OMIT No mutex logic. Not even stubs. The +** mutexes implementation cannot be overridden +** at start-time. +** +** SQLITE_MUTEX_NOOP For single-threaded applications. No +** mutual exclusion is provided. But this +** implementation can be overridden at +** start-time. +** +** SQLITE_MUTEX_PTHREADS For multi-threaded applications on Unix. +** +** SQLITE_MUTEX_W32 For multi-threaded applications on Win32. +*/ +#if !SQLITE_THREADSAFE +# define SQLITE_MUTEX_OMIT +#endif +#if SQLITE_THREADSAFE && !defined(SQLITE_MUTEX_NOOP) +# if SQLITE_OS_UNIX +# define SQLITE_MUTEX_PTHREADS +# elif SQLITE_OS_WIN +# define SQLITE_MUTEX_W32 +# else +# define SQLITE_MUTEX_NOOP +# endif +#endif + +#ifdef SQLITE_MUTEX_OMIT +/* +** If this is a no-op implementation, implement everything as macros. +*/ +#define sqlite3_mutex_alloc(X) ((sqlite3_mutex*)8) +#define sqlite3_mutex_free(X) +#define sqlite3_mutex_enter(X) +#define sqlite3_mutex_try(X) SQLITE_OK +#define sqlite3_mutex_leave(X) +#define sqlite3_mutex_held(X) ((void)(X),1) +#define sqlite3_mutex_notheld(X) ((void)(X),1) +#define sqlite3MutexAlloc(X) ((sqlite3_mutex*)8) +#define sqlite3MutexInit() SQLITE_OK +#define sqlite3MutexEnd() +#define MUTEX_LOGIC(X) +#else +#define MUTEX_LOGIC(X) X +SQLITE_API int sqlite3_mutex_held(sqlite3_mutex*); +#endif /* defined(SQLITE_MUTEX_OMIT) */ + +/************** End of mutex.h ***********************************************/ +/************** Continuing where we left off in sqliteInt.h ******************/ + +/* The SQLITE_EXTRA_DURABLE compile-time option used to set the default +** synchronous setting to EXTRA. It is no longer supported. +*/ +#ifdef SQLITE_EXTRA_DURABLE +# warning Use SQLITE_DEFAULT_SYNCHRONOUS=3 instead of SQLITE_EXTRA_DURABLE +# define SQLITE_DEFAULT_SYNCHRONOUS 3 +#endif + +/* +** Default synchronous levels. +** +** Note that (for historcal reasons) the PAGER_SYNCHRONOUS_* macros differ +** from the SQLITE_DEFAULT_SYNCHRONOUS value by 1. +** +** PAGER_SYNCHRONOUS DEFAULT_SYNCHRONOUS +** OFF 1 0 +** NORMAL 2 1 +** FULL 3 2 +** EXTRA 4 3 +** +** The "PRAGMA synchronous" statement also uses the zero-based numbers. +** In other words, the zero-based numbers are used for all external interfaces +** and the one-based values are used internally. +*/ +#ifndef SQLITE_DEFAULT_SYNCHRONOUS +# define SQLITE_DEFAULT_SYNCHRONOUS 2 +#endif +#ifndef SQLITE_DEFAULT_WAL_SYNCHRONOUS +# define SQLITE_DEFAULT_WAL_SYNCHRONOUS SQLITE_DEFAULT_SYNCHRONOUS +#endif + +/* +** Each database file to be accessed by the system is an instance +** of the following structure. There are normally two of these structures +** in the sqlite.aDb[] array. aDb[0] is the main database file and +** aDb[1] is the database file used to hold temporary tables. Additional +** databases may be attached. +*/ +struct Db { + char *zDbSName; /* Name of this database. (schema name, not filename) */ + Btree *pBt; /* The B*Tree structure for this database file */ + u8 safety_level; /* How aggressive at syncing data to disk */ + u8 bSyncSet; /* True if "PRAGMA synchronous=N" has been run */ + Schema *pSchema; /* Pointer to database schema (possibly shared) */ +}; + +/* +** An instance of the following structure stores a database schema. +** +** Most Schema objects are associated with a Btree. The exception is +** the Schema for the TEMP databaes (sqlite3.aDb[1]) which is free-standing. +** In shared cache mode, a single Schema object can be shared by multiple +** Btrees that refer to the same underlying BtShared object. +** +** Schema objects are automatically deallocated when the last Btree that +** references them is destroyed. The TEMP Schema is manually freed by +** sqlite3_close(). +* +** A thread must be holding a mutex on the corresponding Btree in order +** to access Schema content. This implies that the thread must also be +** holding a mutex on the sqlite3 connection pointer that owns the Btree. +** For a TEMP Schema, only the connection mutex is required. +*/ +struct Schema { + int schema_cookie; /* Database schema version number for this file */ + int iGeneration; /* Generation counter. Incremented with each change */ + Hash tblHash; /* All tables indexed by name */ + Hash idxHash; /* All (named) indices indexed by name */ + Hash trigHash; /* All triggers indexed by name */ + Hash fkeyHash; /* All foreign keys by referenced table name */ + Table *pSeqTab; /* The sqlite_sequence table used by AUTOINCREMENT */ + u8 file_format; /* Schema format version for this file */ + u8 enc; /* Text encoding used by this database */ + u16 schemaFlags; /* Flags associated with this schema */ + int cache_size; /* Number of pages to use in the cache */ +}; + +/* +** These macros can be used to test, set, or clear bits in the +** Db.pSchema->flags field. +*/ +#define DbHasProperty(D,I,P) (((D)->aDb[I].pSchema->schemaFlags&(P))==(P)) +#define DbHasAnyProperty(D,I,P) (((D)->aDb[I].pSchema->schemaFlags&(P))!=0) +#define DbSetProperty(D,I,P) (D)->aDb[I].pSchema->schemaFlags|=(P) +#define DbClearProperty(D,I,P) (D)->aDb[I].pSchema->schemaFlags&=~(P) + +/* +** Allowed values for the DB.pSchema->flags field. +** +** The DB_SchemaLoaded flag is set after the database schema has been +** read into internal hash tables. +** +** DB_UnresetViews means that one or more views have column names that +** have been filled out. If the schema changes, these column names might +** changes and so the view will need to be reset. +*/ +#define DB_SchemaLoaded 0x0001 /* The schema has been loaded */ +#define DB_UnresetViews 0x0002 /* Some views have defined column names */ +#define DB_Empty 0x0004 /* The file is empty (length 0 bytes) */ +#define DB_ResetWanted 0x0008 /* Reset the schema when nSchemaLock==0 */ + +/* +** The number of different kinds of things that can be limited +** using the sqlite3_limit() interface. +*/ +#define SQLITE_N_LIMIT (SQLITE_LIMIT_WORKER_THREADS+1) + +/* +** Lookaside malloc is a set of fixed-size buffers that can be used +** to satisfy small transient memory allocation requests for objects +** associated with a particular database connection. The use of +** lookaside malloc provides a significant performance enhancement +** (approx 10%) by avoiding numerous malloc/free requests while parsing +** SQL statements. +** +** The Lookaside structure holds configuration information about the +** lookaside malloc subsystem. Each available memory allocation in +** the lookaside subsystem is stored on a linked list of LookasideSlot +** objects. +** +** Lookaside allocations are only allowed for objects that are associated +** with a particular database connection. Hence, schema information cannot +** be stored in lookaside because in shared cache mode the schema information +** is shared by multiple database connections. Therefore, while parsing +** schema information, the Lookaside.bEnabled flag is cleared so that +** lookaside allocations are not used to construct the schema objects. +*/ +struct Lookaside { + u32 bDisable; /* Only operate the lookaside when zero */ + u16 sz; /* Size of each buffer in bytes */ + u8 bMalloced; /* True if pStart obtained from sqlite3_malloc() */ + u32 nSlot; /* Number of lookaside slots allocated */ + u32 anStat[3]; /* 0: hits. 1: size misses. 2: full misses */ + LookasideSlot *pInit; /* List of buffers not previously used */ + LookasideSlot *pFree; /* List of available buffers */ + void *pStart; /* First byte of available memory space */ + void *pEnd; /* First byte past end of available space */ +}; +struct LookasideSlot { + LookasideSlot *pNext; /* Next buffer in the list of free buffers */ +}; + +/* +** A hash table for built-in function definitions. (Application-defined +** functions use a regular table table from hash.h.) +** +** Hash each FuncDef structure into one of the FuncDefHash.a[] slots. +** Collisions are on the FuncDef.u.pHash chain. Use the SQLITE_FUNC_HASH() +** macro to compute a hash on the function name. +*/ +#define SQLITE_FUNC_HASH_SZ 23 +struct FuncDefHash { + FuncDef *a[SQLITE_FUNC_HASH_SZ]; /* Hash table for functions */ +}; +#define SQLITE_FUNC_HASH(C,L) (((C)+(L))%SQLITE_FUNC_HASH_SZ) + +#ifdef SQLITE_USER_AUTHENTICATION +/* +** Information held in the "sqlite3" database connection object and used +** to manage user authentication. +*/ +typedef struct sqlite3_userauth sqlite3_userauth; +struct sqlite3_userauth { + u8 authLevel; /* Current authentication level */ + int nAuthPW; /* Size of the zAuthPW in bytes */ + char *zAuthPW; /* Password used to authenticate */ + char *zAuthUser; /* User name used to authenticate */ +}; + +/* Allowed values for sqlite3_userauth.authLevel */ +#define UAUTH_Unknown 0 /* Authentication not yet checked */ +#define UAUTH_Fail 1 /* User authentication failed */ +#define UAUTH_User 2 /* Authenticated as a normal user */ +#define UAUTH_Admin 3 /* Authenticated as an administrator */ + +/* Functions used only by user authorization logic */ +SQLITE_PRIVATE int sqlite3UserAuthTable(const char*); +SQLITE_PRIVATE int sqlite3UserAuthCheckLogin(sqlite3*,const char*,u8*); +SQLITE_PRIVATE void sqlite3UserAuthInit(sqlite3*); +SQLITE_PRIVATE void sqlite3CryptFunc(sqlite3_context*,int,sqlite3_value**); + +#endif /* SQLITE_USER_AUTHENTICATION */ + +/* +** typedef for the authorization callback function. +*/ +#ifdef SQLITE_USER_AUTHENTICATION + typedef int (*sqlite3_xauth)(void*,int,const char*,const char*,const char*, + const char*, const char*); +#else + typedef int (*sqlite3_xauth)(void*,int,const char*,const char*,const char*, + const char*); +#endif + +#ifndef SQLITE_OMIT_DEPRECATED +/* This is an extra SQLITE_TRACE macro that indicates "legacy" tracing +** in the style of sqlite3_trace() +*/ +#define SQLITE_TRACE_LEGACY 0x40 /* Use the legacy xTrace */ +#define SQLITE_TRACE_XPROFILE 0x80 /* Use the legacy xProfile */ +#else +#define SQLITE_TRACE_LEGACY 0 +#define SQLITE_TRACE_XPROFILE 0 +#endif /* SQLITE_OMIT_DEPRECATED */ +#define SQLITE_TRACE_NONLEGACY_MASK 0x0f /* Normal flags */ + + +/* +** Each database connection is an instance of the following structure. +*/ +struct sqlite3 { + sqlite3_vfs *pVfs; /* OS Interface */ + struct Vdbe *pVdbe; /* List of active virtual machines */ + CollSeq *pDfltColl; /* The default collating sequence (BINARY) */ + sqlite3_mutex *mutex; /* Connection mutex */ + Db *aDb; /* All backends */ + int nDb; /* Number of backends currently in use */ + u32 mDbFlags; /* flags recording internal state */ + u64 flags; /* flags settable by pragmas. See below */ + i64 lastRowid; /* ROWID of most recent insert (see above) */ + i64 szMmap; /* Default mmap_size setting */ + u32 nSchemaLock; /* Do not reset the schema when non-zero */ + unsigned int openFlags; /* Flags passed to sqlite3_vfs.xOpen() */ + int errCode; /* Most recent error code (SQLITE_*) */ + int errMask; /* & result codes with this before returning */ + int iSysErrno; /* Errno value from last system error */ + u16 dbOptFlags; /* Flags to enable/disable optimizations */ + u8 enc; /* Text encoding */ + u8 autoCommit; /* The auto-commit flag. */ + u8 temp_store; /* 1: file 2: memory 0: default */ + u8 mallocFailed; /* True if we have seen a malloc failure */ + u8 bBenignMalloc; /* Do not require OOMs if true */ + u8 dfltLockMode; /* Default locking-mode for attached dbs */ + signed char nextAutovac; /* Autovac setting after VACUUM if >=0 */ + u8 suppressErr; /* Do not issue error messages if true */ + u8 vtabOnConflict; /* Value to return for s3_vtab_on_conflict() */ + u8 isTransactionSavepoint; /* True if the outermost savepoint is a TS */ + u8 mTrace; /* zero or more SQLITE_TRACE flags */ + u8 noSharedCache; /* True if no shared-cache backends */ + u8 nSqlExec; /* Number of pending OP_SqlExec opcodes */ + int nextPagesize; /* Pagesize after VACUUM if >0 */ + u32 magic; /* Magic number for detect library misuse */ + int nChange; /* Value returned by sqlite3_changes() */ + int nTotalChange; /* Value returned by sqlite3_total_changes() */ + int aLimit[SQLITE_N_LIMIT]; /* Limits */ + int nMaxSorterMmap; /* Maximum size of regions mapped by sorter */ + struct sqlite3InitInfo { /* Information used during initialization */ + int newTnum; /* Rootpage of table being initialized */ + u8 iDb; /* Which db file is being initialized */ + u8 busy; /* TRUE if currently initializing */ + unsigned orphanTrigger : 1; /* Last statement is orphaned TEMP trigger */ + unsigned imposterTable : 1; /* Building an imposter table */ + unsigned reopenMemdb : 1; /* ATTACH is really a reopen using MemDB */ + char **azInit; /* "type", "name", and "tbl_name" columns */ + } init; + int nVdbeActive; /* Number of VDBEs currently running */ + int nVdbeRead; /* Number of active VDBEs that read or write */ + int nVdbeWrite; /* Number of active VDBEs that read and write */ + int nVdbeExec; /* Number of nested calls to VdbeExec() */ + int nVDestroy; /* Number of active OP_VDestroy operations */ + int nExtension; /* Number of loaded extensions */ + void **aExtension; /* Array of shared library handles */ + int (*xTrace)(u32,void*,void*,void*); /* Trace function */ + void *pTraceArg; /* Argument to the trace function */ +#ifndef SQLITE_OMIT_DEPRECATED + void (*xProfile)(void*,const char*,u64); /* Profiling function */ + void *pProfileArg; /* Argument to profile function */ +#endif + void *pCommitArg; /* Argument to xCommitCallback() */ + int (*xCommitCallback)(void*); /* Invoked at every commit. */ + void *pRollbackArg; /* Argument to xRollbackCallback() */ + void (*xRollbackCallback)(void*); /* Invoked at every commit. */ + void *pUpdateArg; + void (*xUpdateCallback)(void*,int, const char*,const char*,sqlite_int64); + Parse *pParse; /* Current parse */ +#ifdef SQLITE_ENABLE_PREUPDATE_HOOK + void *pPreUpdateArg; /* First argument to xPreUpdateCallback */ + void (*xPreUpdateCallback)( /* Registered using sqlite3_preupdate_hook() */ + void*,sqlite3*,int,char const*,char const*,sqlite3_int64,sqlite3_int64 + ); + PreUpdate *pPreUpdate; /* Context for active pre-update callback */ +#endif /* SQLITE_ENABLE_PREUPDATE_HOOK */ +#ifndef SQLITE_OMIT_WAL + int (*xWalCallback)(void *, sqlite3 *, const char *, int); + void *pWalArg; +#endif + void(*xCollNeeded)(void*,sqlite3*,int eTextRep,const char*); + void(*xCollNeeded16)(void*,sqlite3*,int eTextRep,const void*); + void *pCollNeededArg; + sqlite3_value *pErr; /* Most recent error message */ + union { + volatile int isInterrupted; /* True if sqlite3_interrupt has been called */ + double notUsed1; /* Spacer */ + } u1; + Lookaside lookaside; /* Lookaside malloc configuration */ +#ifndef SQLITE_OMIT_AUTHORIZATION + sqlite3_xauth xAuth; /* Access authorization function */ + void *pAuthArg; /* 1st argument to the access auth function */ +#endif +#ifndef SQLITE_OMIT_PROGRESS_CALLBACK + int (*xProgress)(void *); /* The progress callback */ + void *pProgressArg; /* Argument to the progress callback */ + unsigned nProgressOps; /* Number of opcodes for progress callback */ +#endif +#ifndef SQLITE_OMIT_VIRTUALTABLE + int nVTrans; /* Allocated size of aVTrans */ + Hash aModule; /* populated by sqlite3_create_module() */ + VtabCtx *pVtabCtx; /* Context for active vtab connect/create */ + VTable **aVTrans; /* Virtual tables with open transactions */ + VTable *pDisconnect; /* Disconnect these in next sqlite3_prepare() */ +#endif + Hash aFunc; /* Hash table of connection functions */ + Hash aCollSeq; /* All collating sequences */ + BusyHandler busyHandler; /* Busy callback */ + Db aDbStatic[2]; /* Static space for the 2 default backends */ + Savepoint *pSavepoint; /* List of active savepoints */ + int busyTimeout; /* Busy handler timeout, in msec */ + int nSavepoint; /* Number of non-transaction savepoints */ + int nStatement; /* Number of nested statement-transactions */ + i64 nDeferredCons; /* Net deferred constraints this transaction. */ + i64 nDeferredImmCons; /* Net deferred immediate constraints */ + int *pnBytesFreed; /* If not NULL, increment this in DbFree() */ +#ifdef SQLITE_ENABLE_UNLOCK_NOTIFY + /* The following variables are all protected by the STATIC_MASTER + ** mutex, not by sqlite3.mutex. They are used by code in notify.c. + ** + ** When X.pUnlockConnection==Y, that means that X is waiting for Y to + ** unlock so that it can proceed. + ** + ** When X.pBlockingConnection==Y, that means that something that X tried + ** tried to do recently failed with an SQLITE_LOCKED error due to locks + ** held by Y. + */ + sqlite3 *pBlockingConnection; /* Connection that caused SQLITE_LOCKED */ + sqlite3 *pUnlockConnection; /* Connection to watch for unlock */ + void *pUnlockArg; /* Argument to xUnlockNotify */ + void (*xUnlockNotify)(void **, int); /* Unlock notify callback */ + sqlite3 *pNextBlocked; /* Next in list of all blocked connections */ +#endif +#ifdef SQLITE_USER_AUTHENTICATION + sqlite3_userauth auth; /* User authentication information */ +#endif +}; + +/* +** A macro to discover the encoding of a database. +*/ +#define SCHEMA_ENC(db) ((db)->aDb[0].pSchema->enc) +#define ENC(db) ((db)->enc) + +/* +** Possible values for the sqlite3.flags. +** +** Value constraints (enforced via assert()): +** SQLITE_FullFSync == PAGER_FULLFSYNC +** SQLITE_CkptFullFSync == PAGER_CKPT_FULLFSYNC +** SQLITE_CacheSpill == PAGER_CACHE_SPILL +*/ +#define SQLITE_WriteSchema 0x00000001 /* OK to update SQLITE_MASTER */ +#define SQLITE_LegacyFileFmt 0x00000002 /* Create new databases in format 1 */ +#define SQLITE_FullColNames 0x00000004 /* Show full column names on SELECT */ +#define SQLITE_FullFSync 0x00000008 /* Use full fsync on the backend */ +#define SQLITE_CkptFullFSync 0x00000010 /* Use full fsync for checkpoint */ +#define SQLITE_CacheSpill 0x00000020 /* OK to spill pager cache */ +#define SQLITE_ShortColNames 0x00000040 /* Show short columns names */ +#define SQLITE_CountRows 0x00000080 /* Count rows changed by INSERT, */ + /* DELETE, or UPDATE and return */ + /* the count using a callback. */ +#define SQLITE_NullCallback 0x00000100 /* Invoke the callback once if the */ + /* result set is empty */ +#define SQLITE_IgnoreChecks 0x00000200 /* Do not enforce check constraints */ +#define SQLITE_ReadUncommit 0x00000400 /* READ UNCOMMITTED in shared-cache */ +#define SQLITE_NoCkptOnClose 0x00000800 /* No checkpoint on close()/DETACH */ +#define SQLITE_ReverseOrder 0x00001000 /* Reverse unordered SELECTs */ +#define SQLITE_RecTriggers 0x00002000 /* Enable recursive triggers */ +#define SQLITE_ForeignKeys 0x00004000 /* Enforce foreign key constraints */ +#define SQLITE_AutoIndex 0x00008000 /* Enable automatic indexes */ +#define SQLITE_LoadExtension 0x00010000 /* Enable load_extension */ +#define SQLITE_LoadExtFunc 0x00020000 /* Enable load_extension() SQL func */ +#define SQLITE_EnableTrigger 0x00040000 /* True to enable triggers */ +#define SQLITE_DeferFKs 0x00080000 /* Defer all FK constraints */ +#define SQLITE_QueryOnly 0x00100000 /* Disable database changes */ +#define SQLITE_CellSizeCk 0x00200000 /* Check btree cell sizes on load */ +#define SQLITE_Fts3Tokenizer 0x00400000 /* Enable fts3_tokenizer(2) */ +#define SQLITE_EnableQPSG 0x00800000 /* Query Planner Stability Guarantee*/ +#define SQLITE_TriggerEQP 0x01000000 /* Show trigger EXPLAIN QUERY PLAN */ +#define SQLITE_ResetDatabase 0x02000000 /* Reset the database */ +#define SQLITE_LegacyAlter 0x04000000 /* Legacy ALTER TABLE behaviour */ +#define SQLITE_NoSchemaError 0x08000000 /* Do not report schema parse errors*/ +#define SQLITE_Defensive 0x10000000 /* Input SQL is likely hostile */ +#define SQLITE_DqsDDL 0x20000000 /* dbl-quoted strings allowed in DDL*/ +#define SQLITE_DqsDML 0x40000000 /* dbl-quoted strings allowed in DML*/ +#define SQLITE_EnableView 0x80000000 /* Enable the use of views */ + +/* Flags used only if debugging */ +#define HI(X) ((u64)(X)<<32) +#ifdef SQLITE_DEBUG +#define SQLITE_SqlTrace HI(0x0100000) /* Debug print SQL as it executes */ +#define SQLITE_VdbeListing HI(0x0200000) /* Debug listings of VDBE progs */ +#define SQLITE_VdbeTrace HI(0x0400000) /* True to trace VDBE execution */ +#define SQLITE_VdbeAddopTrace HI(0x0800000) /* Trace sqlite3VdbeAddOp() calls */ +#define SQLITE_VdbeEQP HI(0x1000000) /* Debug EXPLAIN QUERY PLAN */ +#define SQLITE_ParserTrace HI(0x2000000) /* PRAGMA parser_trace=ON */ +#endif + +/* +** Allowed values for sqlite3.mDbFlags +*/ +#define DBFLAG_SchemaChange 0x0001 /* Uncommitted Hash table changes */ +#define DBFLAG_PreferBuiltin 0x0002 /* Preference to built-in funcs */ +#define DBFLAG_Vacuum 0x0004 /* Currently in a VACUUM */ +#define DBFLAG_VacuumInto 0x0008 /* Currently running VACUUM INTO */ +#define DBFLAG_SchemaKnownOk 0x0010 /* Schema is known to be valid */ + +/* +** Bits of the sqlite3.dbOptFlags field that are used by the +** sqlite3_test_control(SQLITE_TESTCTRL_OPTIMIZATIONS,...) interface to +** selectively disable various optimizations. +*/ +#define SQLITE_QueryFlattener 0x0001 /* Query flattening */ +#define SQLITE_WindowFunc 0x0002 /* Use xInverse for window functions */ +#define SQLITE_GroupByOrder 0x0004 /* GROUPBY cover of ORDERBY */ +#define SQLITE_FactorOutConst 0x0008 /* Constant factoring */ +#define SQLITE_DistinctOpt 0x0010 /* DISTINCT using indexes */ +#define SQLITE_CoverIdxScan 0x0020 /* Covering index scans */ +#define SQLITE_OrderByIdxJoin 0x0040 /* ORDER BY of joins via index */ +#define SQLITE_Transitive 0x0080 /* Transitive constraints */ +#define SQLITE_OmitNoopJoin 0x0100 /* Omit unused tables in joins */ +#define SQLITE_CountOfView 0x0200 /* The count-of-view optimization */ +#define SQLITE_CursorHints 0x0400 /* Add OP_CursorHint opcodes */ +#define SQLITE_Stat4 0x0800 /* Use STAT4 data */ + /* TH3 expects the Stat4 ^^^^^^ value to be 0x0800. Don't change it */ +#define SQLITE_PushDown 0x1000 /* The push-down optimization */ +#define SQLITE_SimplifyJoin 0x2000 /* Convert LEFT JOIN to JOIN */ +#define SQLITE_SkipScan 0x4000 /* Skip-scans */ +#define SQLITE_PropagateConst 0x8000 /* The constant propagation opt */ +#define SQLITE_AllOpts 0xffff /* All optimizations */ + +/* +** Macros for testing whether or not optimizations are enabled or disabled. +*/ +#define OptimizationDisabled(db, mask) (((db)->dbOptFlags&(mask))!=0) +#define OptimizationEnabled(db, mask) (((db)->dbOptFlags&(mask))==0) + +/* +** Return true if it OK to factor constant expressions into the initialization +** code. The argument is a Parse object for the code generator. +*/ +#define ConstFactorOk(P) ((P)->okConstFactor) + +/* +** Possible values for the sqlite.magic field. +** The numbers are obtained at random and have no special meaning, other +** than being distinct from one another. +*/ +#define SQLITE_MAGIC_OPEN 0xa029a697 /* Database is open */ +#define SQLITE_MAGIC_CLOSED 0x9f3c2d33 /* Database is closed */ +#define SQLITE_MAGIC_SICK 0x4b771290 /* Error and awaiting close */ +#define SQLITE_MAGIC_BUSY 0xf03b7906 /* Database currently in use */ +#define SQLITE_MAGIC_ERROR 0xb5357930 /* An SQLITE_MISUSE error occurred */ +#define SQLITE_MAGIC_ZOMBIE 0x64cffc7f /* Close with last statement close */ + +/* +** Each SQL function is defined by an instance of the following +** structure. For global built-in functions (ex: substr(), max(), count()) +** a pointer to this structure is held in the sqlite3BuiltinFunctions object. +** For per-connection application-defined functions, a pointer to this +** structure is held in the db->aHash hash table. +** +** The u.pHash field is used by the global built-ins. The u.pDestructor +** field is used by per-connection app-def functions. +*/ +struct FuncDef { + i8 nArg; /* Number of arguments. -1 means unlimited */ + u32 funcFlags; /* Some combination of SQLITE_FUNC_* */ + void *pUserData; /* User data parameter */ + FuncDef *pNext; /* Next function with same name */ + void (*xSFunc)(sqlite3_context*,int,sqlite3_value**); /* func or agg-step */ + void (*xFinalize)(sqlite3_context*); /* Agg finalizer */ + void (*xValue)(sqlite3_context*); /* Current agg value */ + void (*xInverse)(sqlite3_context*,int,sqlite3_value**); /* inverse agg-step */ + const char *zName; /* SQL name of the function. */ + union { + FuncDef *pHash; /* Next with a different name but the same hash */ + FuncDestructor *pDestructor; /* Reference counted destructor function */ + } u; +}; + +/* +** This structure encapsulates a user-function destructor callback (as +** configured using create_function_v2()) and a reference counter. When +** create_function_v2() is called to create a function with a destructor, +** a single object of this type is allocated. FuncDestructor.nRef is set to +** the number of FuncDef objects created (either 1 or 3, depending on whether +** or not the specified encoding is SQLITE_ANY). The FuncDef.pDestructor +** member of each of the new FuncDef objects is set to point to the allocated +** FuncDestructor. +** +** Thereafter, when one of the FuncDef objects is deleted, the reference +** count on this object is decremented. When it reaches 0, the destructor +** is invoked and the FuncDestructor structure freed. +*/ +struct FuncDestructor { + int nRef; + void (*xDestroy)(void *); + void *pUserData; +}; + +/* +** Possible values for FuncDef.flags. Note that the _LENGTH and _TYPEOF +** values must correspond to OPFLAG_LENGTHARG and OPFLAG_TYPEOFARG. And +** SQLITE_FUNC_CONSTANT must be the same as SQLITE_DETERMINISTIC. There +** are assert() statements in the code to verify this. +** +** Value constraints (enforced via assert()): +** SQLITE_FUNC_MINMAX == NC_MinMaxAgg == SF_MinMaxAgg +** SQLITE_FUNC_LENGTH == OPFLAG_LENGTHARG +** SQLITE_FUNC_TYPEOF == OPFLAG_TYPEOFARG +** SQLITE_FUNC_CONSTANT == SQLITE_DETERMINISTIC from the API +** SQLITE_FUNC_DIRECT == SQLITE_DIRECTONLY from the API +** SQLITE_FUNC_ENCMASK depends on SQLITE_UTF* macros in the API +*/ +#define SQLITE_FUNC_ENCMASK 0x0003 /* SQLITE_UTF8, SQLITE_UTF16BE or UTF16LE */ +#define SQLITE_FUNC_LIKE 0x0004 /* Candidate for the LIKE optimization */ +#define SQLITE_FUNC_CASE 0x0008 /* Case-sensitive LIKE-type function */ +#define SQLITE_FUNC_EPHEM 0x0010 /* Ephemeral. Delete with VDBE */ +#define SQLITE_FUNC_NEEDCOLL 0x0020 /* sqlite3GetFuncCollSeq() might be called*/ +#define SQLITE_FUNC_LENGTH 0x0040 /* Built-in length() function */ +#define SQLITE_FUNC_TYPEOF 0x0080 /* Built-in typeof() function */ +#define SQLITE_FUNC_COUNT 0x0100 /* Built-in count(*) aggregate */ +#define SQLITE_FUNC_COALESCE 0x0200 /* Built-in coalesce() or ifnull() */ +#define SQLITE_FUNC_UNLIKELY 0x0400 /* Built-in unlikely() function */ +#define SQLITE_FUNC_CONSTANT 0x0800 /* Constant inputs give a constant output */ +#define SQLITE_FUNC_MINMAX 0x1000 /* True for min() and max() aggregates */ +#define SQLITE_FUNC_SLOCHNG 0x2000 /* "Slow Change". Value constant during a + ** single query - might change over time */ +#define SQLITE_FUNC_AFFINITY 0x4000 /* Built-in affinity() function */ +#define SQLITE_FUNC_OFFSET 0x8000 /* Built-in sqlite_offset() function */ +#define SQLITE_FUNC_WINDOW 0x00010000 /* Built-in window-only function */ +#define SQLITE_FUNC_INTERNAL 0x00040000 /* For use by NestedParse() only */ +#define SQLITE_FUNC_DIRECT 0x00080000 /* Not for use in TRIGGERs or VIEWs */ +#define SQLITE_FUNC_SUBTYPE 0x00100000 /* Result likely to have sub-type */ + +/* +** The following three macros, FUNCTION(), LIKEFUNC() and AGGREGATE() are +** used to create the initializers for the FuncDef structures. +** +** FUNCTION(zName, nArg, iArg, bNC, xFunc) +** Used to create a scalar function definition of a function zName +** implemented by C function xFunc that accepts nArg arguments. The +** value passed as iArg is cast to a (void*) and made available +** as the user-data (sqlite3_user_data()) for the function. If +** argument bNC is true, then the SQLITE_FUNC_NEEDCOLL flag is set. +** +** VFUNCTION(zName, nArg, iArg, bNC, xFunc) +** Like FUNCTION except it omits the SQLITE_FUNC_CONSTANT flag. +** +** DFUNCTION(zName, nArg, iArg, bNC, xFunc) +** Like FUNCTION except it omits the SQLITE_FUNC_CONSTANT flag and +** adds the SQLITE_FUNC_SLOCHNG flag. Used for date & time functions +** and functions like sqlite_version() that can change, but not during +** a single query. The iArg is ignored. The user-data is always set +** to a NULL pointer. The bNC parameter is not used. +** +** PURE_DATE(zName, nArg, iArg, bNC, xFunc) +** Used for "pure" date/time functions, this macro is like DFUNCTION +** except that it does set the SQLITE_FUNC_CONSTANT flags. iArg is +** ignored and the user-data for these functions is set to an +** arbitrary non-NULL pointer. The bNC parameter is not used. +** +** AGGREGATE(zName, nArg, iArg, bNC, xStep, xFinal) +** Used to create an aggregate function definition implemented by +** the C functions xStep and xFinal. The first four parameters +** are interpreted in the same way as the first 4 parameters to +** FUNCTION(). +** +** WFUNCTION(zName, nArg, iArg, xStep, xFinal, xValue, xInverse) +** Used to create an aggregate function definition implemented by +** the C functions xStep and xFinal. The first four parameters +** are interpreted in the same way as the first 4 parameters to +** FUNCTION(). +** +** LIKEFUNC(zName, nArg, pArg, flags) +** Used to create a scalar function definition of a function zName +** that accepts nArg arguments and is implemented by a call to C +** function likeFunc. Argument pArg is cast to a (void *) and made +** available as the function user-data (sqlite3_user_data()). The +** FuncDef.flags variable is set to the value passed as the flags +** parameter. +*/ +#define FUNCTION(zName, nArg, iArg, bNC, xFunc) \ + {nArg, SQLITE_FUNC_CONSTANT|SQLITE_UTF8|(bNC*SQLITE_FUNC_NEEDCOLL), \ + SQLITE_INT_TO_PTR(iArg), 0, xFunc, 0, 0, 0, #zName, {0} } +#define VFUNCTION(zName, nArg, iArg, bNC, xFunc) \ + {nArg, SQLITE_UTF8|(bNC*SQLITE_FUNC_NEEDCOLL), \ + SQLITE_INT_TO_PTR(iArg), 0, xFunc, 0, 0, 0, #zName, {0} } +#define DFUNCTION(zName, nArg, iArg, bNC, xFunc) \ + {nArg, SQLITE_FUNC_SLOCHNG|SQLITE_UTF8, \ + 0, 0, xFunc, 0, 0, 0, #zName, {0} } +#define PURE_DATE(zName, nArg, iArg, bNC, xFunc) \ + {nArg, SQLITE_FUNC_SLOCHNG|SQLITE_UTF8|SQLITE_FUNC_CONSTANT, \ + (void*)&sqlite3Config, 0, xFunc, 0, 0, 0, #zName, {0} } +#define FUNCTION2(zName, nArg, iArg, bNC, xFunc, extraFlags) \ + {nArg,SQLITE_FUNC_CONSTANT|SQLITE_UTF8|(bNC*SQLITE_FUNC_NEEDCOLL)|extraFlags,\ + SQLITE_INT_TO_PTR(iArg), 0, xFunc, 0, 0, 0, #zName, {0} } +#define STR_FUNCTION(zName, nArg, pArg, bNC, xFunc) \ + {nArg, SQLITE_FUNC_SLOCHNG|SQLITE_UTF8|(bNC*SQLITE_FUNC_NEEDCOLL), \ + pArg, 0, xFunc, 0, 0, 0, #zName, } +#define LIKEFUNC(zName, nArg, arg, flags) \ + {nArg, SQLITE_FUNC_CONSTANT|SQLITE_UTF8|flags, \ + (void *)arg, 0, likeFunc, 0, 0, 0, #zName, {0} } +#define AGGREGATE(zName, nArg, arg, nc, xStep, xFinal, xValue) \ + {nArg, SQLITE_UTF8|(nc*SQLITE_FUNC_NEEDCOLL), \ + SQLITE_INT_TO_PTR(arg), 0, xStep,xFinal,xValue,0,#zName, {0}} +#define AGGREGATE2(zName, nArg, arg, nc, xStep, xFinal, extraFlags) \ + {nArg, SQLITE_UTF8|(nc*SQLITE_FUNC_NEEDCOLL)|extraFlags, \ + SQLITE_INT_TO_PTR(arg), 0, xStep,xFinal,xFinal,0,#zName, {0}} +#define WAGGREGATE(zName, nArg, arg, nc, xStep, xFinal, xValue, xInverse, f) \ + {nArg, SQLITE_UTF8|(nc*SQLITE_FUNC_NEEDCOLL)|f, \ + SQLITE_INT_TO_PTR(arg), 0, xStep,xFinal,xValue,xInverse,#zName, {0}} +#define INTERNAL_FUNCTION(zName, nArg, xFunc) \ + {nArg, SQLITE_FUNC_INTERNAL|SQLITE_UTF8|SQLITE_FUNC_CONSTANT, \ + 0, 0, xFunc, 0, 0, 0, #zName, {0} } + + +/* +** All current savepoints are stored in a linked list starting at +** sqlite3.pSavepoint. The first element in the list is the most recently +** opened savepoint. Savepoints are added to the list by the vdbe +** OP_Savepoint instruction. +*/ +struct Savepoint { + char *zName; /* Savepoint name (nul-terminated) */ + i64 nDeferredCons; /* Number of deferred fk violations */ + i64 nDeferredImmCons; /* Number of deferred imm fk. */ + Savepoint *pNext; /* Parent savepoint (if any) */ +}; + +/* +** The following are used as the second parameter to sqlite3Savepoint(), +** and as the P1 argument to the OP_Savepoint instruction. +*/ +#define SAVEPOINT_BEGIN 0 +#define SAVEPOINT_RELEASE 1 +#define SAVEPOINT_ROLLBACK 2 + + +/* +** Each SQLite module (virtual table definition) is defined by an +** instance of the following structure, stored in the sqlite3.aModule +** hash table. +*/ +struct Module { + const sqlite3_module *pModule; /* Callback pointers */ + const char *zName; /* Name passed to create_module() */ + int nRefModule; /* Number of pointers to this object */ + void *pAux; /* pAux passed to create_module() */ + void (*xDestroy)(void *); /* Module destructor function */ + Table *pEpoTab; /* Eponymous table for this module */ +}; + +/* +** information about each column of an SQL table is held in an instance +** of this structure. +*/ +struct Column { + char *zName; /* Name of this column, \000, then the type */ + Expr *pDflt; /* Default value of this column */ + char *zColl; /* Collating sequence. If NULL, use the default */ + u8 notNull; /* An OE_ code for handling a NOT NULL constraint */ + char affinity; /* One of the SQLITE_AFF_... values */ + u8 szEst; /* Estimated size of value in this column. sizeof(INT)==1 */ + u8 colFlags; /* Boolean properties. See COLFLAG_ defines below */ +}; + +/* Allowed values for Column.colFlags: +*/ +#define COLFLAG_PRIMKEY 0x0001 /* Column is part of the primary key */ +#define COLFLAG_HIDDEN 0x0002 /* A hidden column in a virtual table */ +#define COLFLAG_HASTYPE 0x0004 /* Type name follows column name */ +#define COLFLAG_UNIQUE 0x0008 /* Column def contains "UNIQUE" or "PK" */ +#define COLFLAG_SORTERREF 0x0010 /* Use sorter-refs with this column */ + +/* +** A "Collating Sequence" is defined by an instance of the following +** structure. Conceptually, a collating sequence consists of a name and +** a comparison routine that defines the order of that sequence. +** +** If CollSeq.xCmp is NULL, it means that the +** collating sequence is undefined. Indices built on an undefined +** collating sequence may not be read or written. +*/ +struct CollSeq { + char *zName; /* Name of the collating sequence, UTF-8 encoded */ + u8 enc; /* Text encoding handled by xCmp() */ + void *pUser; /* First argument to xCmp() */ + int (*xCmp)(void*,int, const void*, int, const void*); + void (*xDel)(void*); /* Destructor for pUser */ +}; + +/* +** A sort order can be either ASC or DESC. +*/ +#define SQLITE_SO_ASC 0 /* Sort in ascending order */ +#define SQLITE_SO_DESC 1 /* Sort in ascending order */ +#define SQLITE_SO_UNDEFINED -1 /* No sort order specified */ + +/* +** Column affinity types. +** +** These used to have mnemonic name like 'i' for SQLITE_AFF_INTEGER and +** 't' for SQLITE_AFF_TEXT. But we can save a little space and improve +** the speed a little by numbering the values consecutively. +** +** But rather than start with 0 or 1, we begin with 'A'. That way, +** when multiple affinity types are concatenated into a string and +** used as the P4 operand, they will be more readable. +** +** Note also that the numeric types are grouped together so that testing +** for a numeric type is a single comparison. And the BLOB type is first. +*/ +#define SQLITE_AFF_NONE 0x40 /* '@' */ +#define SQLITE_AFF_BLOB 0x41 /* 'A' */ +#define SQLITE_AFF_TEXT 0x42 /* 'B' */ +#define SQLITE_AFF_NUMERIC 0x43 /* 'C' */ +#define SQLITE_AFF_INTEGER 0x44 /* 'D' */ +#define SQLITE_AFF_REAL 0x45 /* 'E' */ + +#define sqlite3IsNumericAffinity(X) ((X)>=SQLITE_AFF_NUMERIC) + +/* +** The SQLITE_AFF_MASK values masks off the significant bits of an +** affinity value. +*/ +#define SQLITE_AFF_MASK 0x47 + +/* +** Additional bit values that can be ORed with an affinity without +** changing the affinity. +** +** The SQLITE_NOTNULL flag is a combination of NULLEQ and JUMPIFNULL. +** It causes an assert() to fire if either operand to a comparison +** operator is NULL. It is added to certain comparison operators to +** prove that the operands are always NOT NULL. +*/ +#define SQLITE_KEEPNULL 0x08 /* Used by vector == or <> */ +#define SQLITE_JUMPIFNULL 0x10 /* jumps if either operand is NULL */ +#define SQLITE_STOREP2 0x20 /* Store result in reg[P2] rather than jump */ +#define SQLITE_NULLEQ 0x80 /* NULL=NULL */ +#define SQLITE_NOTNULL 0x90 /* Assert that operands are never NULL */ + +/* +** An object of this type is created for each virtual table present in +** the database schema. +** +** If the database schema is shared, then there is one instance of this +** structure for each database connection (sqlite3*) that uses the shared +** schema. This is because each database connection requires its own unique +** instance of the sqlite3_vtab* handle used to access the virtual table +** implementation. sqlite3_vtab* handles can not be shared between +** database connections, even when the rest of the in-memory database +** schema is shared, as the implementation often stores the database +** connection handle passed to it via the xConnect() or xCreate() method +** during initialization internally. This database connection handle may +** then be used by the virtual table implementation to access real tables +** within the database. So that they appear as part of the callers +** transaction, these accesses need to be made via the same database +** connection as that used to execute SQL operations on the virtual table. +** +** All VTable objects that correspond to a single table in a shared +** database schema are initially stored in a linked-list pointed to by +** the Table.pVTable member variable of the corresponding Table object. +** When an sqlite3_prepare() operation is required to access the virtual +** table, it searches the list for the VTable that corresponds to the +** database connection doing the preparing so as to use the correct +** sqlite3_vtab* handle in the compiled query. +** +** When an in-memory Table object is deleted (for example when the +** schema is being reloaded for some reason), the VTable objects are not +** deleted and the sqlite3_vtab* handles are not xDisconnect()ed +** immediately. Instead, they are moved from the Table.pVTable list to +** another linked list headed by the sqlite3.pDisconnect member of the +** corresponding sqlite3 structure. They are then deleted/xDisconnected +** next time a statement is prepared using said sqlite3*. This is done +** to avoid deadlock issues involving multiple sqlite3.mutex mutexes. +** Refer to comments above function sqlite3VtabUnlockList() for an +** explanation as to why it is safe to add an entry to an sqlite3.pDisconnect +** list without holding the corresponding sqlite3.mutex mutex. +** +** The memory for objects of this type is always allocated by +** sqlite3DbMalloc(), using the connection handle stored in VTable.db as +** the first argument. +*/ +struct VTable { + sqlite3 *db; /* Database connection associated with this table */ + Module *pMod; /* Pointer to module implementation */ + sqlite3_vtab *pVtab; /* Pointer to vtab instance */ + int nRef; /* Number of pointers to this structure */ + u8 bConstraint; /* True if constraints are supported */ + int iSavepoint; /* Depth of the SAVEPOINT stack */ + VTable *pNext; /* Next in linked list (see above) */ +}; + +/* +** The schema for each SQL table and view is represented in memory +** by an instance of the following structure. +*/ +struct Table { + char *zName; /* Name of the table or view */ + Column *aCol; /* Information about each column */ + Index *pIndex; /* List of SQL indexes on this table. */ + Select *pSelect; /* NULL for tables. Points to definition if a view. */ + FKey *pFKey; /* Linked list of all foreign keys in this table */ + char *zColAff; /* String defining the affinity of each column */ + ExprList *pCheck; /* All CHECK constraints */ + /* ... also used as column name list in a VIEW */ + int tnum; /* Root BTree page for this table */ + u32 nTabRef; /* Number of pointers to this Table */ + u32 tabFlags; /* Mask of TF_* values */ + i16 iPKey; /* If not negative, use aCol[iPKey] as the rowid */ + i16 nCol; /* Number of columns in this table */ + LogEst nRowLogEst; /* Estimated rows in table - from sqlite_stat1 table */ + LogEst szTabRow; /* Estimated size of each table row in bytes */ +#ifdef SQLITE_ENABLE_COSTMULT + LogEst costMult; /* Cost multiplier for using this table */ +#endif + u8 keyConf; /* What to do in case of uniqueness conflict on iPKey */ +#ifndef SQLITE_OMIT_ALTERTABLE + int addColOffset; /* Offset in CREATE TABLE stmt to add a new column */ +#endif +#ifndef SQLITE_OMIT_VIRTUALTABLE + int nModuleArg; /* Number of arguments to the module */ + char **azModuleArg; /* 0: module 1: schema 2: vtab name 3...: args */ + VTable *pVTable; /* List of VTable objects. */ +#endif + Trigger *pTrigger; /* List of triggers stored in pSchema */ + Schema *pSchema; /* Schema that contains this table */ + Table *pNextZombie; /* Next on the Parse.pZombieTab list */ +}; + +/* +** Allowed values for Table.tabFlags. +** +** TF_OOOHidden applies to tables or view that have hidden columns that are +** followed by non-hidden columns. Example: "CREATE VIRTUAL TABLE x USING +** vtab1(a HIDDEN, b);". Since "b" is a non-hidden column but "a" is hidden, +** the TF_OOOHidden attribute would apply in this case. Such tables require +** special handling during INSERT processing. +*/ +#define TF_Readonly 0x0001 /* Read-only system table */ +#define TF_Ephemeral 0x0002 /* An ephemeral table */ +#define TF_HasPrimaryKey 0x0004 /* Table has a primary key */ +#define TF_Autoincrement 0x0008 /* Integer primary key is autoincrement */ +#define TF_HasStat1 0x0010 /* nRowLogEst set from sqlite_stat1 */ +#define TF_WithoutRowid 0x0020 /* No rowid. PRIMARY KEY is the key */ +#define TF_NoVisibleRowid 0x0040 /* No user-visible "rowid" column */ +#define TF_OOOHidden 0x0080 /* Out-of-Order hidden columns */ +#define TF_StatsUsed 0x0100 /* Query planner decisions affected by + ** Index.aiRowLogEst[] values */ +#define TF_HasNotNull 0x0200 /* Contains NOT NULL constraints */ +#define TF_Shadow 0x0400 /* True for a shadow table */ + +/* +** Test to see whether or not a table is a virtual table. This is +** done as a macro so that it will be optimized out when virtual +** table support is omitted from the build. +*/ +#ifndef SQLITE_OMIT_VIRTUALTABLE +# define IsVirtual(X) ((X)->nModuleArg) +#else +# define IsVirtual(X) 0 +#endif + +/* +** Macros to determine if a column is hidden. IsOrdinaryHiddenColumn() +** only works for non-virtual tables (ordinary tables and views) and is +** always false unless SQLITE_ENABLE_HIDDEN_COLUMNS is defined. The +** IsHiddenColumn() macro is general purpose. +*/ +#if defined(SQLITE_ENABLE_HIDDEN_COLUMNS) +# define IsHiddenColumn(X) (((X)->colFlags & COLFLAG_HIDDEN)!=0) +# define IsOrdinaryHiddenColumn(X) (((X)->colFlags & COLFLAG_HIDDEN)!=0) +#elif !defined(SQLITE_OMIT_VIRTUALTABLE) +# define IsHiddenColumn(X) (((X)->colFlags & COLFLAG_HIDDEN)!=0) +# define IsOrdinaryHiddenColumn(X) 0 +#else +# define IsHiddenColumn(X) 0 +# define IsOrdinaryHiddenColumn(X) 0 +#endif + + +/* Does the table have a rowid */ +#define HasRowid(X) (((X)->tabFlags & TF_WithoutRowid)==0) +#define VisibleRowid(X) (((X)->tabFlags & TF_NoVisibleRowid)==0) + +/* +** Each foreign key constraint is an instance of the following structure. +** +** A foreign key is associated with two tables. The "from" table is +** the table that contains the REFERENCES clause that creates the foreign +** key. The "to" table is the table that is named in the REFERENCES clause. +** Consider this example: +** +** CREATE TABLE ex1( +** a INTEGER PRIMARY KEY, +** b INTEGER CONSTRAINT fk1 REFERENCES ex2(x) +** ); +** +** For foreign key "fk1", the from-table is "ex1" and the to-table is "ex2". +** Equivalent names: +** +** from-table == child-table +** to-table == parent-table +** +** Each REFERENCES clause generates an instance of the following structure +** which is attached to the from-table. The to-table need not exist when +** the from-table is created. The existence of the to-table is not checked. +** +** The list of all parents for child Table X is held at X.pFKey. +** +** A list of all children for a table named Z (which might not even exist) +** is held in Schema.fkeyHash with a hash key of Z. +*/ +struct FKey { + Table *pFrom; /* Table containing the REFERENCES clause (aka: Child) */ + FKey *pNextFrom; /* Next FKey with the same in pFrom. Next parent of pFrom */ + char *zTo; /* Name of table that the key points to (aka: Parent) */ + FKey *pNextTo; /* Next with the same zTo. Next child of zTo. */ + FKey *pPrevTo; /* Previous with the same zTo */ + int nCol; /* Number of columns in this key */ + /* EV: R-30323-21917 */ + u8 isDeferred; /* True if constraint checking is deferred till COMMIT */ + u8 aAction[2]; /* ON DELETE and ON UPDATE actions, respectively */ + Trigger *apTrigger[2];/* Triggers for aAction[] actions */ + struct sColMap { /* Mapping of columns in pFrom to columns in zTo */ + int iFrom; /* Index of column in pFrom */ + char *zCol; /* Name of column in zTo. If NULL use PRIMARY KEY */ + } aCol[1]; /* One entry for each of nCol columns */ +}; + +/* +** SQLite supports many different ways to resolve a constraint +** error. ROLLBACK processing means that a constraint violation +** causes the operation in process to fail and for the current transaction +** to be rolled back. ABORT processing means the operation in process +** fails and any prior changes from that one operation are backed out, +** but the transaction is not rolled back. FAIL processing means that +** the operation in progress stops and returns an error code. But prior +** changes due to the same operation are not backed out and no rollback +** occurs. IGNORE means that the particular row that caused the constraint +** error is not inserted or updated. Processing continues and no error +** is returned. REPLACE means that preexisting database rows that caused +** a UNIQUE constraint violation are removed so that the new insert or +** update can proceed. Processing continues and no error is reported. +** +** RESTRICT, SETNULL, and CASCADE actions apply only to foreign keys. +** RESTRICT is the same as ABORT for IMMEDIATE foreign keys and the +** same as ROLLBACK for DEFERRED keys. SETNULL means that the foreign +** key is set to NULL. CASCADE means that a DELETE or UPDATE of the +** referenced table row is propagated into the row that holds the +** foreign key. +** +** The following symbolic values are used to record which type +** of action to take. +*/ +#define OE_None 0 /* There is no constraint to check */ +#define OE_Rollback 1 /* Fail the operation and rollback the transaction */ +#define OE_Abort 2 /* Back out changes but do no rollback transaction */ +#define OE_Fail 3 /* Stop the operation but leave all prior changes */ +#define OE_Ignore 4 /* Ignore the error. Do not do the INSERT or UPDATE */ +#define OE_Replace 5 /* Delete existing record, then do INSERT or UPDATE */ +#define OE_Update 6 /* Process as a DO UPDATE in an upsert */ +#define OE_Restrict 7 /* OE_Abort for IMMEDIATE, OE_Rollback for DEFERRED */ +#define OE_SetNull 8 /* Set the foreign key value to NULL */ +#define OE_SetDflt 9 /* Set the foreign key value to its default */ +#define OE_Cascade 10 /* Cascade the changes */ +#define OE_Default 11 /* Do whatever the default action is */ + + +/* +** An instance of the following structure is passed as the first +** argument to sqlite3VdbeKeyCompare and is used to control the +** comparison of the two index keys. +** +** Note that aSortOrder[] and aColl[] have nField+1 slots. There +** are nField slots for the columns of an index then one extra slot +** for the rowid at the end. +*/ +struct KeyInfo { + u32 nRef; /* Number of references to this KeyInfo object */ + u8 enc; /* Text encoding - one of the SQLITE_UTF* values */ + u16 nKeyField; /* Number of key columns in the index */ + u16 nAllField; /* Total columns, including key plus others */ + sqlite3 *db; /* The database connection */ + u8 *aSortFlags; /* Sort order for each column. */ + CollSeq *aColl[1]; /* Collating sequence for each term of the key */ +}; + +/* +** Allowed bit values for entries in the KeyInfo.aSortFlags[] array. +*/ +#define KEYINFO_ORDER_DESC 0x01 /* DESC sort order */ +#define KEYINFO_ORDER_BIGNULL 0x02 /* NULL is larger than any other value */ + +/* +** This object holds a record which has been parsed out into individual +** fields, for the purposes of doing a comparison. +** +** A record is an object that contains one or more fields of data. +** Records are used to store the content of a table row and to store +** the key of an index. A blob encoding of a record is created by +** the OP_MakeRecord opcode of the VDBE and is disassembled by the +** OP_Column opcode. +** +** An instance of this object serves as a "key" for doing a search on +** an index b+tree. The goal of the search is to find the entry that +** is closed to the key described by this object. This object might hold +** just a prefix of the key. The number of fields is given by +** pKeyInfo->nField. +** +** The r1 and r2 fields are the values to return if this key is less than +** or greater than a key in the btree, respectively. These are normally +** -1 and +1 respectively, but might be inverted to +1 and -1 if the b-tree +** is in DESC order. +** +** The key comparison functions actually return default_rc when they find +** an equals comparison. default_rc can be -1, 0, or +1. If there are +** multiple entries in the b-tree with the same key (when only looking +** at the first pKeyInfo->nFields,) then default_rc can be set to -1 to +** cause the search to find the last match, or +1 to cause the search to +** find the first match. +** +** The key comparison functions will set eqSeen to true if they ever +** get and equal results when comparing this structure to a b-tree record. +** When default_rc!=0, the search might end up on the record immediately +** before the first match or immediately after the last match. The +** eqSeen field will indicate whether or not an exact match exists in the +** b-tree. +*/ +struct UnpackedRecord { + KeyInfo *pKeyInfo; /* Collation and sort-order information */ + Mem *aMem; /* Values */ + u16 nField; /* Number of entries in apMem[] */ + i8 default_rc; /* Comparison result if keys are equal */ + u8 errCode; /* Error detected by xRecordCompare (CORRUPT or NOMEM) */ + i8 r1; /* Value to return if (lhs < rhs) */ + i8 r2; /* Value to return if (lhs > rhs) */ + u8 eqSeen; /* True if an equality comparison has been seen */ +}; + + +/* +** Each SQL index is represented in memory by an +** instance of the following structure. +** +** The columns of the table that are to be indexed are described +** by the aiColumn[] field of this structure. For example, suppose +** we have the following table and index: +** +** CREATE TABLE Ex1(c1 int, c2 int, c3 text); +** CREATE INDEX Ex2 ON Ex1(c3,c1); +** +** In the Table structure describing Ex1, nCol==3 because there are +** three columns in the table. In the Index structure describing +** Ex2, nColumn==2 since 2 of the 3 columns of Ex1 are indexed. +** The value of aiColumn is {2, 0}. aiColumn[0]==2 because the +** first column to be indexed (c3) has an index of 2 in Ex1.aCol[]. +** The second column to be indexed (c1) has an index of 0 in +** Ex1.aCol[], hence Ex2.aiColumn[1]==0. +** +** The Index.onError field determines whether or not the indexed columns +** must be unique and what to do if they are not. When Index.onError=OE_None, +** it means this is not a unique index. Otherwise it is a unique index +** and the value of Index.onError indicate the which conflict resolution +** algorithm to employ whenever an attempt is made to insert a non-unique +** element. +** +** While parsing a CREATE TABLE or CREATE INDEX statement in order to +** generate VDBE code (as opposed to parsing one read from an sqlite_master +** table as part of parsing an existing database schema), transient instances +** of this structure may be created. In this case the Index.tnum variable is +** used to store the address of a VDBE instruction, not a database page +** number (it cannot - the database page is not allocated until the VDBE +** program is executed). See convertToWithoutRowidTable() for details. +*/ +struct Index { + char *zName; /* Name of this index */ + i16 *aiColumn; /* Which columns are used by this index. 1st is 0 */ + LogEst *aiRowLogEst; /* From ANALYZE: Est. rows selected by each column */ + Table *pTable; /* The SQL table being indexed */ + char *zColAff; /* String defining the affinity of each column */ + Index *pNext; /* The next index associated with the same table */ + Schema *pSchema; /* Schema containing this index */ + u8 *aSortOrder; /* for each column: True==DESC, False==ASC */ + const char **azColl; /* Array of collation sequence names for index */ + Expr *pPartIdxWhere; /* WHERE clause for partial indices */ + ExprList *aColExpr; /* Column expressions */ + int tnum; /* DB Page containing root of this index */ + LogEst szIdxRow; /* Estimated average row size in bytes */ + u16 nKeyCol; /* Number of columns forming the key */ + u16 nColumn; /* Number of columns stored in the index */ + u8 onError; /* OE_Abort, OE_Ignore, OE_Replace, or OE_None */ + unsigned idxType:2; /* 0:Normal 1:UNIQUE, 2:PRIMARY KEY, 3:IPK */ + unsigned bUnordered:1; /* Use this index for == or IN queries only */ + unsigned uniqNotNull:1; /* True if UNIQUE and NOT NULL for all columns */ + unsigned isResized:1; /* True if resizeIndexObject() has been called */ + unsigned isCovering:1; /* True if this is a covering index */ + unsigned noSkipScan:1; /* Do not try to use skip-scan if true */ + unsigned hasStat1:1; /* aiRowLogEst values come from sqlite_stat1 */ + unsigned bNoQuery:1; /* Do not use this index to optimize queries */ + unsigned bAscKeyBug:1; /* True if the bba7b69f9849b5bf bug applies */ +#ifdef SQLITE_ENABLE_STAT4 + int nSample; /* Number of elements in aSample[] */ + int nSampleCol; /* Size of IndexSample.anEq[] and so on */ + tRowcnt *aAvgEq; /* Average nEq values for keys not in aSample */ + IndexSample *aSample; /* Samples of the left-most key */ + tRowcnt *aiRowEst; /* Non-logarithmic stat1 data for this index */ + tRowcnt nRowEst0; /* Non-logarithmic number of rows in the index */ +#endif + Bitmask colNotIdxed; /* 0 for unindexed columns in pTab */ +}; + +/* +** Allowed values for Index.idxType +*/ +#define SQLITE_IDXTYPE_APPDEF 0 /* Created using CREATE INDEX */ +#define SQLITE_IDXTYPE_UNIQUE 1 /* Implements a UNIQUE constraint */ +#define SQLITE_IDXTYPE_PRIMARYKEY 2 /* Is the PRIMARY KEY for the table */ +#define SQLITE_IDXTYPE_IPK 3 /* INTEGER PRIMARY KEY index */ + +/* Return true if index X is a PRIMARY KEY index */ +#define IsPrimaryKeyIndex(X) ((X)->idxType==SQLITE_IDXTYPE_PRIMARYKEY) + +/* Return true if index X is a UNIQUE index */ +#define IsUniqueIndex(X) ((X)->onError!=OE_None) + +/* The Index.aiColumn[] values are normally positive integer. But +** there are some negative values that have special meaning: +*/ +#define XN_ROWID (-1) /* Indexed column is the rowid */ +#define XN_EXPR (-2) /* Indexed column is an expression */ + +/* +** Each sample stored in the sqlite_stat4 table is represented in memory +** using a structure of this type. See documentation at the top of the +** analyze.c source file for additional information. +*/ +struct IndexSample { + void *p; /* Pointer to sampled record */ + int n; /* Size of record in bytes */ + tRowcnt *anEq; /* Est. number of rows where the key equals this sample */ + tRowcnt *anLt; /* Est. number of rows where key is less than this sample */ + tRowcnt *anDLt; /* Est. number of distinct keys less than this sample */ +}; + +/* +** Possible values to use within the flags argument to sqlite3GetToken(). +*/ +#define SQLITE_TOKEN_QUOTED 0x1 /* Token is a quoted identifier. */ +#define SQLITE_TOKEN_KEYWORD 0x2 /* Token is a keyword. */ + +/* +** Each token coming out of the lexer is an instance of +** this structure. Tokens are also used as part of an expression. +** +** The memory that "z" points to is owned by other objects. Take care +** that the owner of the "z" string does not deallocate the string before +** the Token goes out of scope! Very often, the "z" points to some place +** in the middle of the Parse.zSql text. But it might also point to a +** static string. +*/ +struct Token { + const char *z; /* Text of the token. Not NULL-terminated! */ + unsigned int n; /* Number of characters in this token */ +}; + +/* +** An instance of this structure contains information needed to generate +** code for a SELECT that contains aggregate functions. +** +** If Expr.op==TK_AGG_COLUMN or TK_AGG_FUNCTION then Expr.pAggInfo is a +** pointer to this structure. The Expr.iColumn field is the index in +** AggInfo.aCol[] or AggInfo.aFunc[] of information needed to generate +** code for that node. +** +** AggInfo.pGroupBy and AggInfo.aFunc.pExpr point to fields within the +** original Select structure that describes the SELECT statement. These +** fields do not need to be freed when deallocating the AggInfo structure. +*/ +struct AggInfo { + u8 directMode; /* Direct rendering mode means take data directly + ** from source tables rather than from accumulators */ + u8 useSortingIdx; /* In direct mode, reference the sorting index rather + ** than the source table */ + int sortingIdx; /* Cursor number of the sorting index */ + int sortingIdxPTab; /* Cursor number of pseudo-table */ + int nSortingColumn; /* Number of columns in the sorting index */ + int mnReg, mxReg; /* Range of registers allocated for aCol and aFunc */ + ExprList *pGroupBy; /* The group by clause */ + struct AggInfo_col { /* For each column used in source tables */ + Table *pTab; /* Source table */ + int iTable; /* Cursor number of the source table */ + int iColumn; /* Column number within the source table */ + int iSorterColumn; /* Column number in the sorting index */ + int iMem; /* Memory location that acts as accumulator */ + Expr *pExpr; /* The original expression */ + } *aCol; + int nColumn; /* Number of used entries in aCol[] */ + int nAccumulator; /* Number of columns that show through to the output. + ** Additional columns are used only as parameters to + ** aggregate functions */ + struct AggInfo_func { /* For each aggregate function */ + Expr *pExpr; /* Expression encoding the function */ + FuncDef *pFunc; /* The aggregate function implementation */ + int iMem; /* Memory location that acts as accumulator */ + int iDistinct; /* Ephemeral table used to enforce DISTINCT */ + } *aFunc; + int nFunc; /* Number of entries in aFunc[] */ +}; + +/* +** The datatype ynVar is a signed integer, either 16-bit or 32-bit. +** Usually it is 16-bits. But if SQLITE_MAX_VARIABLE_NUMBER is greater +** than 32767 we have to make it 32-bit. 16-bit is preferred because +** it uses less memory in the Expr object, which is a big memory user +** in systems with lots of prepared statements. And few applications +** need more than about 10 or 20 variables. But some extreme users want +** to have prepared statements with over 32767 variables, and for them +** the option is available (at compile-time). +*/ +#if SQLITE_MAX_VARIABLE_NUMBER<=32767 +typedef i16 ynVar; +#else +typedef int ynVar; +#endif + +/* +** Each node of an expression in the parse tree is an instance +** of this structure. +** +** Expr.op is the opcode. The integer parser token codes are reused +** as opcodes here. For example, the parser defines TK_GE to be an integer +** code representing the ">=" operator. This same integer code is reused +** to represent the greater-than-or-equal-to operator in the expression +** tree. +** +** If the expression is an SQL literal (TK_INTEGER, TK_FLOAT, TK_BLOB, +** or TK_STRING), then Expr.token contains the text of the SQL literal. If +** the expression is a variable (TK_VARIABLE), then Expr.token contains the +** variable name. Finally, if the expression is an SQL function (TK_FUNCTION), +** then Expr.token contains the name of the function. +** +** Expr.pRight and Expr.pLeft are the left and right subexpressions of a +** binary operator. Either or both may be NULL. +** +** Expr.x.pList is a list of arguments if the expression is an SQL function, +** a CASE expression or an IN expression of the form " IN (, ...)". +** Expr.x.pSelect is used if the expression is a sub-select or an expression of +** the form " IN (SELECT ...)". If the EP_xIsSelect bit is set in the +** Expr.flags mask, then Expr.x.pSelect is valid. Otherwise, Expr.x.pList is +** valid. +** +** An expression of the form ID or ID.ID refers to a column in a table. +** For such expressions, Expr.op is set to TK_COLUMN and Expr.iTable is +** the integer cursor number of a VDBE cursor pointing to that table and +** Expr.iColumn is the column number for the specific column. If the +** expression is used as a result in an aggregate SELECT, then the +** value is also stored in the Expr.iAgg column in the aggregate so that +** it can be accessed after all aggregates are computed. +** +** If the expression is an unbound variable marker (a question mark +** character '?' in the original SQL) then the Expr.iTable holds the index +** number for that variable. +** +** If the expression is a subquery then Expr.iColumn holds an integer +** register number containing the result of the subquery. If the +** subquery gives a constant result, then iTable is -1. If the subquery +** gives a different answer at different times during statement processing +** then iTable is the address of a subroutine that computes the subquery. +** +** If the Expr is of type OP_Column, and the table it is selecting from +** is a disk table or the "old.*" pseudo-table, then pTab points to the +** corresponding table definition. +** +** ALLOCATION NOTES: +** +** Expr objects can use a lot of memory space in database schema. To +** help reduce memory requirements, sometimes an Expr object will be +** truncated. And to reduce the number of memory allocations, sometimes +** two or more Expr objects will be stored in a single memory allocation, +** together with Expr.zToken strings. +** +** If the EP_Reduced and EP_TokenOnly flags are set when +** an Expr object is truncated. When EP_Reduced is set, then all +** the child Expr objects in the Expr.pLeft and Expr.pRight subtrees +** are contained within the same memory allocation. Note, however, that +** the subtrees in Expr.x.pList or Expr.x.pSelect are always separately +** allocated, regardless of whether or not EP_Reduced is set. +*/ +struct Expr { + u8 op; /* Operation performed by this node */ + char affExpr; /* affinity, or RAISE type */ + u32 flags; /* Various flags. EP_* See below */ + union { + char *zToken; /* Token value. Zero terminated and dequoted */ + int iValue; /* Non-negative integer value if EP_IntValue */ + } u; + + /* If the EP_TokenOnly flag is set in the Expr.flags mask, then no + ** space is allocated for the fields below this point. An attempt to + ** access them will result in a segfault or malfunction. + *********************************************************************/ + + Expr *pLeft; /* Left subnode */ + Expr *pRight; /* Right subnode */ + union { + ExprList *pList; /* op = IN, EXISTS, SELECT, CASE, FUNCTION, BETWEEN */ + Select *pSelect; /* EP_xIsSelect and op = IN, EXISTS, SELECT */ + } x; + + /* If the EP_Reduced flag is set in the Expr.flags mask, then no + ** space is allocated for the fields below this point. An attempt to + ** access them will result in a segfault or malfunction. + *********************************************************************/ + +#if SQLITE_MAX_EXPR_DEPTH>0 + int nHeight; /* Height of the tree headed by this node */ +#endif + int iTable; /* TK_COLUMN: cursor number of table holding column + ** TK_REGISTER: register number + ** TK_TRIGGER: 1 -> new, 0 -> old + ** EP_Unlikely: 134217728 times likelihood + ** TK_IN: ephemerial table holding RHS + ** TK_SELECT_COLUMN: Number of columns on the LHS + ** TK_SELECT: 1st register of result vector */ + ynVar iColumn; /* TK_COLUMN: column index. -1 for rowid. + ** TK_VARIABLE: variable number (always >= 1). + ** TK_SELECT_COLUMN: column of the result vector */ + i16 iAgg; /* Which entry in pAggInfo->aCol[] or ->aFunc[] */ + i16 iRightJoinTable; /* If EP_FromJoin, the right table of the join */ + u8 op2; /* TK_REGISTER/TK_TRUTH: original value of Expr.op + ** TK_COLUMN: the value of p5 for OP_Column + ** TK_AGG_FUNCTION: nesting depth */ + AggInfo *pAggInfo; /* Used by TK_AGG_COLUMN and TK_AGG_FUNCTION */ + union { + Table *pTab; /* TK_COLUMN: Table containing column. Can be NULL + ** for a column of an index on an expression */ + Window *pWin; /* EP_WinFunc: Window/Filter defn for a function */ + struct { /* TK_IN, TK_SELECT, and TK_EXISTS */ + int iAddr; /* Subroutine entry address */ + int regReturn; /* Register used to hold return address */ + } sub; + } y; +}; + +/* +** The following are the meanings of bits in the Expr.flags field. +** Value restrictions: +** +** EP_Agg == NC_HasAgg == SF_HasAgg +** EP_Win == NC_HasWin +*/ +#define EP_FromJoin 0x000001 /* Originates in ON/USING clause of outer join */ +#define EP_Distinct 0x000002 /* Aggregate function with DISTINCT keyword */ +#define EP_HasFunc 0x000004 /* Contains one or more functions of any kind */ +#define EP_FixedCol 0x000008 /* TK_Column with a known fixed value */ +#define EP_Agg 0x000010 /* Contains one or more aggregate functions */ +#define EP_VarSelect 0x000020 /* pSelect is correlated, not constant */ +#define EP_DblQuoted 0x000040 /* token.z was originally in "..." */ +#define EP_InfixFunc 0x000080 /* True for an infix function: LIKE, GLOB, etc */ +#define EP_Collate 0x000100 /* Tree contains a TK_COLLATE operator */ + /* 0x000200 Available for reuse */ +#define EP_IntValue 0x000400 /* Integer value contained in u.iValue */ +#define EP_xIsSelect 0x000800 /* x.pSelect is valid (otherwise x.pList is) */ +#define EP_Skip 0x001000 /* Operator does not contribute to affinity */ +#define EP_Reduced 0x002000 /* Expr struct EXPR_REDUCEDSIZE bytes only */ +#define EP_TokenOnly 0x004000 /* Expr struct EXPR_TOKENONLYSIZE bytes only */ +#define EP_Win 0x008000 /* Contains window functions */ +#define EP_MemToken 0x010000 /* Need to sqlite3DbFree() Expr.zToken */ +#define EP_NoReduce 0x020000 /* Cannot EXPRDUP_REDUCE this Expr */ +#define EP_Unlikely 0x040000 /* unlikely() or likelihood() function */ +#define EP_ConstFunc 0x080000 /* A SQLITE_FUNC_CONSTANT or _SLOCHNG function */ +#define EP_CanBeNull 0x100000 /* Can be null despite NOT NULL constraint */ +#define EP_Subquery 0x200000 /* Tree contains a TK_SELECT operator */ +#define EP_Alias 0x400000 /* Is an alias for a result set column */ +#define EP_Leaf 0x800000 /* Expr.pLeft, .pRight, .u.pSelect all NULL */ +#define EP_WinFunc 0x1000000 /* TK_FUNCTION with Expr.y.pWin set */ +#define EP_Subrtn 0x2000000 /* Uses Expr.y.sub. TK_IN, _SELECT, or _EXISTS */ +#define EP_Quoted 0x4000000 /* TK_ID was originally quoted */ +#define EP_Static 0x8000000 /* Held in memory not obtained from malloc() */ +#define EP_IsTrue 0x10000000 /* Always has boolean value of TRUE */ +#define EP_IsFalse 0x20000000 /* Always has boolean value of FALSE */ +#define EP_Indirect 0x40000000 /* Contained within a TRIGGER or a VIEW */ + +/* +** The EP_Propagate mask is a set of properties that automatically propagate +** upwards into parent nodes. +*/ +#define EP_Propagate (EP_Collate|EP_Subquery|EP_HasFunc) + +/* +** These macros can be used to test, set, or clear bits in the +** Expr.flags field. +*/ +#define ExprHasProperty(E,P) (((E)->flags&(P))!=0) +#define ExprHasAllProperty(E,P) (((E)->flags&(P))==(P)) +#define ExprSetProperty(E,P) (E)->flags|=(P) +#define ExprClearProperty(E,P) (E)->flags&=~(P) +#define ExprAlwaysTrue(E) (((E)->flags&(EP_FromJoin|EP_IsTrue))==EP_IsTrue) +#define ExprAlwaysFalse(E) (((E)->flags&(EP_FromJoin|EP_IsFalse))==EP_IsFalse) + +/* The ExprSetVVAProperty() macro is used for Verification, Validation, +** and Accreditation only. It works like ExprSetProperty() during VVA +** processes but is a no-op for delivery. +*/ +#ifdef SQLITE_DEBUG +# define ExprSetVVAProperty(E,P) (E)->flags|=(P) +#else +# define ExprSetVVAProperty(E,P) +#endif + +/* +** Macros to determine the number of bytes required by a normal Expr +** struct, an Expr struct with the EP_Reduced flag set in Expr.flags +** and an Expr struct with the EP_TokenOnly flag set. +*/ +#define EXPR_FULLSIZE sizeof(Expr) /* Full size */ +#define EXPR_REDUCEDSIZE offsetof(Expr,iTable) /* Common features */ +#define EXPR_TOKENONLYSIZE offsetof(Expr,pLeft) /* Fewer features */ + +/* +** Flags passed to the sqlite3ExprDup() function. See the header comment +** above sqlite3ExprDup() for details. +*/ +#define EXPRDUP_REDUCE 0x0001 /* Used reduced-size Expr nodes */ + +/* +** True if the expression passed as an argument was a function with +** an OVER() clause (a window function). +*/ +#ifdef SQLITE_OMIT_WINDOWFUNC +# define IsWindowFunc(p) 0 +#else +# define IsWindowFunc(p) ( \ + ExprHasProperty((p), EP_WinFunc) && p->y.pWin->eFrmType!=TK_FILTER \ + ) +#endif + +/* +** A list of expressions. Each expression may optionally have a +** name. An expr/name combination can be used in several ways, such +** as the list of "expr AS ID" fields following a "SELECT" or in the +** list of "ID = expr" items in an UPDATE. A list of expressions can +** also be used as the argument to a function, in which case the a.zName +** field is not used. +** +** By default the Expr.zSpan field holds a human-readable description of +** the expression that is used in the generation of error messages and +** column labels. In this case, Expr.zSpan is typically the text of a +** column expression as it exists in a SELECT statement. However, if +** the bSpanIsTab flag is set, then zSpan is overloaded to mean the name +** of the result column in the form: DATABASE.TABLE.COLUMN. This later +** form is used for name resolution with nested FROM clauses. +*/ +struct ExprList { + int nExpr; /* Number of expressions on the list */ + struct ExprList_item { /* For each expression in the list */ + Expr *pExpr; /* The parse tree for this expression */ + char *zName; /* Token associated with this expression */ + char *zSpan; /* Original text of the expression */ + u8 sortFlags; /* Mask of KEYINFO_ORDER_* flags */ + unsigned done :1; /* A flag to indicate when processing is finished */ + unsigned bSpanIsTab :1; /* zSpan holds DB.TABLE.COLUMN */ + unsigned reusable :1; /* Constant expression is reusable */ + unsigned bSorterRef :1; /* Defer evaluation until after sorting */ + unsigned bNulls: 1; /* True if explicit "NULLS FIRST/LAST" */ + union { + struct { + u16 iOrderByCol; /* For ORDER BY, column number in result set */ + u16 iAlias; /* Index into Parse.aAlias[] for zName */ + } x; + int iConstExprReg; /* Register in which Expr value is cached */ + } u; + } a[1]; /* One slot for each expression in the list */ +}; + +/* +** An instance of this structure can hold a simple list of identifiers, +** such as the list "a,b,c" in the following statements: +** +** INSERT INTO t(a,b,c) VALUES ...; +** CREATE INDEX idx ON t(a,b,c); +** CREATE TRIGGER trig BEFORE UPDATE ON t(a,b,c) ...; +** +** The IdList.a.idx field is used when the IdList represents the list of +** column names after a table name in an INSERT statement. In the statement +** +** INSERT INTO t(a,b,c) ... +** +** If "a" is the k-th column of table "t", then IdList.a[0].idx==k. +*/ +struct IdList { + struct IdList_item { + char *zName; /* Name of the identifier */ + int idx; /* Index in some Table.aCol[] of a column named zName */ + } *a; + int nId; /* Number of identifiers on the list */ +}; + +/* +** The following structure describes the FROM clause of a SELECT statement. +** Each table or subquery in the FROM clause is a separate element of +** the SrcList.a[] array. +** +** With the addition of multiple database support, the following structure +** can also be used to describe a particular table such as the table that +** is modified by an INSERT, DELETE, or UPDATE statement. In standard SQL, +** such a table must be a simple name: ID. But in SQLite, the table can +** now be identified by a database name, a dot, then the table name: ID.ID. +** +** The jointype starts out showing the join type between the current table +** and the next table on the list. The parser builds the list this way. +** But sqlite3SrcListShiftJoinType() later shifts the jointypes so that each +** jointype expresses the join between the table and the previous table. +** +** In the colUsed field, the high-order bit (bit 63) is set if the table +** contains more than 63 columns and the 64-th or later column is used. +*/ +struct SrcList { + int nSrc; /* Number of tables or subqueries in the FROM clause */ + u32 nAlloc; /* Number of entries allocated in a[] below */ + struct SrcList_item { + Schema *pSchema; /* Schema to which this item is fixed */ + char *zDatabase; /* Name of database holding this table */ + char *zName; /* Name of the table */ + char *zAlias; /* The "B" part of a "A AS B" phrase. zName is the "A" */ + Table *pTab; /* An SQL table corresponding to zName */ + Select *pSelect; /* A SELECT statement used in place of a table name */ + int addrFillSub; /* Address of subroutine to manifest a subquery */ + int regReturn; /* Register holding return address of addrFillSub */ + int regResult; /* Registers holding results of a co-routine */ + struct { + u8 jointype; /* Type of join between this table and the previous */ + unsigned notIndexed :1; /* True if there is a NOT INDEXED clause */ + unsigned isIndexedBy :1; /* True if there is an INDEXED BY clause */ + unsigned isTabFunc :1; /* True if table-valued-function syntax */ + unsigned isCorrelated :1; /* True if sub-query is correlated */ + unsigned viaCoroutine :1; /* Implemented as a co-routine */ + unsigned isRecursive :1; /* True for recursive reference in WITH */ + } fg; + int iCursor; /* The VDBE cursor number used to access this table */ + Expr *pOn; /* The ON clause of a join */ + IdList *pUsing; /* The USING clause of a join */ + Bitmask colUsed; /* Bit N (1<" clause */ + ExprList *pFuncArg; /* Arguments to table-valued-function */ + } u1; + Index *pIBIndex; /* Index structure corresponding to u1.zIndexedBy */ + } a[1]; /* One entry for each identifier on the list */ +}; + +/* +** Permitted values of the SrcList.a.jointype field +*/ +#define JT_INNER 0x0001 /* Any kind of inner or cross join */ +#define JT_CROSS 0x0002 /* Explicit use of the CROSS keyword */ +#define JT_NATURAL 0x0004 /* True for a "natural" join */ +#define JT_LEFT 0x0008 /* Left outer join */ +#define JT_RIGHT 0x0010 /* Right outer join */ +#define JT_OUTER 0x0020 /* The "OUTER" keyword is present */ +#define JT_ERROR 0x0040 /* unknown or unsupported join type */ + + +/* +** Flags appropriate for the wctrlFlags parameter of sqlite3WhereBegin() +** and the WhereInfo.wctrlFlags member. +** +** Value constraints (enforced via assert()): +** WHERE_USE_LIMIT == SF_FixedLimit +*/ +#define WHERE_ORDERBY_NORMAL 0x0000 /* No-op */ +#define WHERE_ORDERBY_MIN 0x0001 /* ORDER BY processing for min() func */ +#define WHERE_ORDERBY_MAX 0x0002 /* ORDER BY processing for max() func */ +#define WHERE_ONEPASS_DESIRED 0x0004 /* Want to do one-pass UPDATE/DELETE */ +#define WHERE_ONEPASS_MULTIROW 0x0008 /* ONEPASS is ok with multiple rows */ +#define WHERE_DUPLICATES_OK 0x0010 /* Ok to return a row more than once */ +#define WHERE_OR_SUBCLAUSE 0x0020 /* Processing a sub-WHERE as part of + ** the OR optimization */ +#define WHERE_GROUPBY 0x0040 /* pOrderBy is really a GROUP BY */ +#define WHERE_DISTINCTBY 0x0080 /* pOrderby is really a DISTINCT clause */ +#define WHERE_WANT_DISTINCT 0x0100 /* All output needs to be distinct */ +#define WHERE_SORTBYGROUP 0x0200 /* Support sqlite3WhereIsSorted() */ +#define WHERE_SEEK_TABLE 0x0400 /* Do not defer seeks on main table */ +#define WHERE_ORDERBY_LIMIT 0x0800 /* ORDERBY+LIMIT on the inner loop */ +#define WHERE_SEEK_UNIQ_TABLE 0x1000 /* Do not defer seeks if unique */ + /* 0x2000 not currently used */ +#define WHERE_USE_LIMIT 0x4000 /* Use the LIMIT in cost estimates */ + /* 0x8000 not currently used */ + +/* Allowed return values from sqlite3WhereIsDistinct() +*/ +#define WHERE_DISTINCT_NOOP 0 /* DISTINCT keyword not used */ +#define WHERE_DISTINCT_UNIQUE 1 /* No duplicates */ +#define WHERE_DISTINCT_ORDERED 2 /* All duplicates are adjacent */ +#define WHERE_DISTINCT_UNORDERED 3 /* Duplicates are scattered */ + +/* +** A NameContext defines a context in which to resolve table and column +** names. The context consists of a list of tables (the pSrcList) field and +** a list of named expression (pEList). The named expression list may +** be NULL. The pSrc corresponds to the FROM clause of a SELECT or +** to the table being operated on by INSERT, UPDATE, or DELETE. The +** pEList corresponds to the result set of a SELECT and is NULL for +** other statements. +** +** NameContexts can be nested. When resolving names, the inner-most +** context is searched first. If no match is found, the next outer +** context is checked. If there is still no match, the next context +** is checked. This process continues until either a match is found +** or all contexts are check. When a match is found, the nRef member of +** the context containing the match is incremented. +** +** Each subquery gets a new NameContext. The pNext field points to the +** NameContext in the parent query. Thus the process of scanning the +** NameContext list corresponds to searching through successively outer +** subqueries looking for a match. +*/ +struct NameContext { + Parse *pParse; /* The parser */ + SrcList *pSrcList; /* One or more tables used to resolve names */ + union { + ExprList *pEList; /* Optional list of result-set columns */ + AggInfo *pAggInfo; /* Information about aggregates at this level */ + Upsert *pUpsert; /* ON CONFLICT clause information from an upsert */ + } uNC; + NameContext *pNext; /* Next outer name context. NULL for outermost */ + int nRef; /* Number of names resolved by this context */ + int nErr; /* Number of errors encountered while resolving names */ + int ncFlags; /* Zero or more NC_* flags defined below */ + Select *pWinSelect; /* SELECT statement for any window functions */ +}; + +/* +** Allowed values for the NameContext, ncFlags field. +** +** Value constraints (all checked via assert()): +** NC_HasAgg == SF_HasAgg == EP_Agg +** NC_MinMaxAgg == SF_MinMaxAgg == SQLITE_FUNC_MINMAX +** NC_HasWin == EP_Win +** +*/ +#define NC_AllowAgg 0x0001 /* Aggregate functions are allowed here */ +#define NC_PartIdx 0x0002 /* True if resolving a partial index WHERE */ +#define NC_IsCheck 0x0004 /* True if resolving names in a CHECK constraint */ +#define NC_InAggFunc 0x0008 /* True if analyzing arguments to an agg func */ +#define NC_HasAgg 0x0010 /* One or more aggregate functions seen */ +#define NC_IdxExpr 0x0020 /* True if resolving columns of CREATE INDEX */ +#define NC_VarSelect 0x0040 /* A correlated subquery has been seen */ +#define NC_UEList 0x0080 /* True if uNC.pEList is used */ +#define NC_UAggInfo 0x0100 /* True if uNC.pAggInfo is used */ +#define NC_UUpsert 0x0200 /* True if uNC.pUpsert is used */ +#define NC_MinMaxAgg 0x1000 /* min/max aggregates seen. See note above */ +#define NC_Complex 0x2000 /* True if a function or subquery seen */ +#define NC_AllowWin 0x4000 /* Window functions are allowed here */ +#define NC_HasWin 0x8000 /* One or more window functions seen */ +#define NC_IsDDL 0x10000 /* Resolving names in a CREATE statement */ + +/* +** An instance of the following object describes a single ON CONFLICT +** clause in an upsert. +** +** The pUpsertTarget field is only set if the ON CONFLICT clause includes +** conflict-target clause. (In "ON CONFLICT(a,b)" the "(a,b)" is the +** conflict-target clause.) The pUpsertTargetWhere is the optional +** WHERE clause used to identify partial unique indexes. +** +** pUpsertSet is the list of column=expr terms of the UPDATE statement. +** The pUpsertSet field is NULL for a ON CONFLICT DO NOTHING. The +** pUpsertWhere is the WHERE clause for the UPDATE and is NULL if the +** WHERE clause is omitted. +*/ +struct Upsert { + ExprList *pUpsertTarget; /* Optional description of conflicting index */ + Expr *pUpsertTargetWhere; /* WHERE clause for partial index targets */ + ExprList *pUpsertSet; /* The SET clause from an ON CONFLICT UPDATE */ + Expr *pUpsertWhere; /* WHERE clause for the ON CONFLICT UPDATE */ + /* The fields above comprise the parse tree for the upsert clause. + ** The fields below are used to transfer information from the INSERT + ** processing down into the UPDATE processing while generating code. + ** Upsert owns the memory allocated above, but not the memory below. */ + Index *pUpsertIdx; /* Constraint that pUpsertTarget identifies */ + SrcList *pUpsertSrc; /* Table to be updated */ + int regData; /* First register holding array of VALUES */ + int iDataCur; /* Index of the data cursor */ + int iIdxCur; /* Index of the first index cursor */ +}; + +/* +** An instance of the following structure contains all information +** needed to generate code for a single SELECT statement. +** +** See the header comment on the computeLimitRegisters() routine for a +** detailed description of the meaning of the iLimit and iOffset fields. +** +** addrOpenEphm[] entries contain the address of OP_OpenEphemeral opcodes. +** These addresses must be stored so that we can go back and fill in +** the P4_KEYINFO and P2 parameters later. Neither the KeyInfo nor +** the number of columns in P2 can be computed at the same time +** as the OP_OpenEphm instruction is coded because not +** enough information about the compound query is known at that point. +** The KeyInfo for addrOpenTran[0] and [1] contains collating sequences +** for the result set. The KeyInfo for addrOpenEphm[2] contains collating +** sequences for the ORDER BY clause. +*/ +struct Select { + ExprList *pEList; /* The fields of the result */ + u8 op; /* One of: TK_UNION TK_ALL TK_INTERSECT TK_EXCEPT */ + LogEst nSelectRow; /* Estimated number of result rows */ + u32 selFlags; /* Various SF_* values */ + int iLimit, iOffset; /* Memory registers holding LIMIT & OFFSET counters */ + u32 selId; /* Unique identifier number for this SELECT */ + int addrOpenEphm[2]; /* OP_OpenEphem opcodes related to this select */ + SrcList *pSrc; /* The FROM clause */ + Expr *pWhere; /* The WHERE clause */ + ExprList *pGroupBy; /* The GROUP BY clause */ + Expr *pHaving; /* The HAVING clause */ + ExprList *pOrderBy; /* The ORDER BY clause */ + Select *pPrior; /* Prior select in a compound select statement */ + Select *pNext; /* Next select to the left in a compound */ + Expr *pLimit; /* LIMIT expression. NULL means not used. */ + With *pWith; /* WITH clause attached to this select. Or NULL. */ +#ifndef SQLITE_OMIT_WINDOWFUNC + Window *pWin; /* List of window functions */ + Window *pWinDefn; /* List of named window definitions */ +#endif +}; + +/* +** Allowed values for Select.selFlags. The "SF" prefix stands for +** "Select Flag". +** +** Value constraints (all checked via assert()) +** SF_HasAgg == NC_HasAgg +** SF_MinMaxAgg == NC_MinMaxAgg == SQLITE_FUNC_MINMAX +** SF_FixedLimit == WHERE_USE_LIMIT +*/ +#define SF_Distinct 0x00001 /* Output should be DISTINCT */ +#define SF_All 0x00002 /* Includes the ALL keyword */ +#define SF_Resolved 0x00004 /* Identifiers have been resolved */ +#define SF_Aggregate 0x00008 /* Contains agg functions or a GROUP BY */ +#define SF_HasAgg 0x00010 /* Contains aggregate functions */ +#define SF_UsesEphemeral 0x00020 /* Uses the OpenEphemeral opcode */ +#define SF_Expanded 0x00040 /* sqlite3SelectExpand() called on this */ +#define SF_HasTypeInfo 0x00080 /* FROM subqueries have Table metadata */ +#define SF_Compound 0x00100 /* Part of a compound query */ +#define SF_Values 0x00200 /* Synthesized from VALUES clause */ +#define SF_MultiValue 0x00400 /* Single VALUES term with multiple rows */ +#define SF_NestedFrom 0x00800 /* Part of a parenthesized FROM clause */ +#define SF_MinMaxAgg 0x01000 /* Aggregate containing min() or max() */ +#define SF_Recursive 0x02000 /* The recursive part of a recursive CTE */ +#define SF_FixedLimit 0x04000 /* nSelectRow set by a constant LIMIT */ +#define SF_MaybeConvert 0x08000 /* Need convertCompoundSelectToSubquery() */ +#define SF_Converted 0x10000 /* By convertCompoundSelectToSubquery() */ +#define SF_IncludeHidden 0x20000 /* Include hidden columns in output */ +#define SF_ComplexResult 0x40000 /* Result contains subquery or function */ +#define SF_WhereBegin 0x80000 /* Really a WhereBegin() call. Debug Only */ + +/* +** The results of a SELECT can be distributed in several ways, as defined +** by one of the following macros. The "SRT" prefix means "SELECT Result +** Type". +** +** SRT_Union Store results as a key in a temporary index +** identified by pDest->iSDParm. +** +** SRT_Except Remove results from the temporary index pDest->iSDParm. +** +** SRT_Exists Store a 1 in memory cell pDest->iSDParm if the result +** set is not empty. +** +** SRT_Discard Throw the results away. This is used by SELECT +** statements within triggers whose only purpose is +** the side-effects of functions. +** +** All of the above are free to ignore their ORDER BY clause. Those that +** follow must honor the ORDER BY clause. +** +** SRT_Output Generate a row of output (using the OP_ResultRow +** opcode) for each row in the result set. +** +** SRT_Mem Only valid if the result is a single column. +** Store the first column of the first result row +** in register pDest->iSDParm then abandon the rest +** of the query. This destination implies "LIMIT 1". +** +** SRT_Set The result must be a single column. Store each +** row of result as the key in table pDest->iSDParm. +** Apply the affinity pDest->affSdst before storing +** results. Used to implement "IN (SELECT ...)". +** +** SRT_EphemTab Create an temporary table pDest->iSDParm and store +** the result there. The cursor is left open after +** returning. This is like SRT_Table except that +** this destination uses OP_OpenEphemeral to create +** the table first. +** +** SRT_Coroutine Generate a co-routine that returns a new row of +** results each time it is invoked. The entry point +** of the co-routine is stored in register pDest->iSDParm +** and the result row is stored in pDest->nDest registers +** starting with pDest->iSdst. +** +** SRT_Table Store results in temporary table pDest->iSDParm. +** SRT_Fifo This is like SRT_EphemTab except that the table +** is assumed to already be open. SRT_Fifo has +** the additional property of being able to ignore +** the ORDER BY clause. +** +** SRT_DistFifo Store results in a temporary table pDest->iSDParm. +** But also use temporary table pDest->iSDParm+1 as +** a record of all prior results and ignore any duplicate +** rows. Name means: "Distinct Fifo". +** +** SRT_Queue Store results in priority queue pDest->iSDParm (really +** an index). Append a sequence number so that all entries +** are distinct. +** +** SRT_DistQueue Store results in priority queue pDest->iSDParm only if +** the same record has never been stored before. The +** index at pDest->iSDParm+1 hold all prior stores. +*/ +#define SRT_Union 1 /* Store result as keys in an index */ +#define SRT_Except 2 /* Remove result from a UNION index */ +#define SRT_Exists 3 /* Store 1 if the result is not empty */ +#define SRT_Discard 4 /* Do not save the results anywhere */ +#define SRT_Fifo 5 /* Store result as data with an automatic rowid */ +#define SRT_DistFifo 6 /* Like SRT_Fifo, but unique results only */ +#define SRT_Queue 7 /* Store result in an queue */ +#define SRT_DistQueue 8 /* Like SRT_Queue, but unique results only */ + +/* The ORDER BY clause is ignored for all of the above */ +#define IgnorableOrderby(X) ((X->eDest)<=SRT_DistQueue) + +#define SRT_Output 9 /* Output each row of result */ +#define SRT_Mem 10 /* Store result in a memory cell */ +#define SRT_Set 11 /* Store results as keys in an index */ +#define SRT_EphemTab 12 /* Create transient tab and store like SRT_Table */ +#define SRT_Coroutine 13 /* Generate a single row of result */ +#define SRT_Table 14 /* Store result as data with an automatic rowid */ + +/* +** An instance of this object describes where to put of the results of +** a SELECT statement. +*/ +struct SelectDest { + u8 eDest; /* How to dispose of the results. On of SRT_* above. */ + int iSDParm; /* A parameter used by the eDest disposal method */ + int iSdst; /* Base register where results are written */ + int nSdst; /* Number of registers allocated */ + char *zAffSdst; /* Affinity used when eDest==SRT_Set */ + ExprList *pOrderBy; /* Key columns for SRT_Queue and SRT_DistQueue */ +}; + +/* +** During code generation of statements that do inserts into AUTOINCREMENT +** tables, the following information is attached to the Table.u.autoInc.p +** pointer of each autoincrement table to record some side information that +** the code generator needs. We have to keep per-table autoincrement +** information in case inserts are done within triggers. Triggers do not +** normally coordinate their activities, but we do need to coordinate the +** loading and saving of autoincrement information. +*/ +struct AutoincInfo { + AutoincInfo *pNext; /* Next info block in a list of them all */ + Table *pTab; /* Table this info block refers to */ + int iDb; /* Index in sqlite3.aDb[] of database holding pTab */ + int regCtr; /* Memory register holding the rowid counter */ +}; + +/* +** At least one instance of the following structure is created for each +** trigger that may be fired while parsing an INSERT, UPDATE or DELETE +** statement. All such objects are stored in the linked list headed at +** Parse.pTriggerPrg and deleted once statement compilation has been +** completed. +** +** A Vdbe sub-program that implements the body and WHEN clause of trigger +** TriggerPrg.pTrigger, assuming a default ON CONFLICT clause of +** TriggerPrg.orconf, is stored in the TriggerPrg.pProgram variable. +** The Parse.pTriggerPrg list never contains two entries with the same +** values for both pTrigger and orconf. +** +** The TriggerPrg.aColmask[0] variable is set to a mask of old.* columns +** accessed (or set to 0 for triggers fired as a result of INSERT +** statements). Similarly, the TriggerPrg.aColmask[1] variable is set to +** a mask of new.* columns used by the program. +*/ +struct TriggerPrg { + Trigger *pTrigger; /* Trigger this program was coded from */ + TriggerPrg *pNext; /* Next entry in Parse.pTriggerPrg list */ + SubProgram *pProgram; /* Program implementing pTrigger/orconf */ + int orconf; /* Default ON CONFLICT policy */ + u32 aColmask[2]; /* Masks of old.*, new.* columns accessed */ +}; + +/* +** The yDbMask datatype for the bitmask of all attached databases. +*/ +#if SQLITE_MAX_ATTACHED>30 + typedef unsigned char yDbMask[(SQLITE_MAX_ATTACHED+9)/8]; +# define DbMaskTest(M,I) (((M)[(I)/8]&(1<<((I)&7)))!=0) +# define DbMaskZero(M) memset((M),0,sizeof(M)) +# define DbMaskSet(M,I) (M)[(I)/8]|=(1<<((I)&7)) +# define DbMaskAllZero(M) sqlite3DbMaskAllZero(M) +# define DbMaskNonZero(M) (sqlite3DbMaskAllZero(M)==0) +#else + typedef unsigned int yDbMask; +# define DbMaskTest(M,I) (((M)&(((yDbMask)1)<<(I)))!=0) +# define DbMaskZero(M) (M)=0 +# define DbMaskSet(M,I) (M)|=(((yDbMask)1)<<(I)) +# define DbMaskAllZero(M) (M)==0 +# define DbMaskNonZero(M) (M)!=0 +#endif + +/* +** An SQL parser context. A copy of this structure is passed through +** the parser and down into all the parser action routine in order to +** carry around information that is global to the entire parse. +** +** The structure is divided into two parts. When the parser and code +** generate call themselves recursively, the first part of the structure +** is constant but the second part is reset at the beginning and end of +** each recursion. +** +** The nTableLock and aTableLock variables are only used if the shared-cache +** feature is enabled (if sqlite3Tsd()->useSharedData is true). They are +** used to store the set of table-locks required by the statement being +** compiled. Function sqlite3TableLock() is used to add entries to the +** list. +*/ +struct Parse { + sqlite3 *db; /* The main database structure */ + char *zErrMsg; /* An error message */ + Vdbe *pVdbe; /* An engine for executing database bytecode */ + int rc; /* Return code from execution */ + u8 colNamesSet; /* TRUE after OP_ColumnName has been issued to pVdbe */ + u8 checkSchema; /* Causes schema cookie check after an error */ + u8 nested; /* Number of nested calls to the parser/code generator */ + u8 nTempReg; /* Number of temporary registers in aTempReg[] */ + u8 isMultiWrite; /* True if statement may modify/insert multiple rows */ + u8 mayAbort; /* True if statement may throw an ABORT exception */ + u8 hasCompound; /* Need to invoke convertCompoundSelectToSubquery() */ + u8 okConstFactor; /* OK to factor out constants */ + u8 disableLookaside; /* Number of times lookaside has been disabled */ + u8 disableVtab; /* Disable all virtual tables for this parse */ + int nRangeReg; /* Size of the temporary register block */ + int iRangeReg; /* First register in temporary register block */ + int nErr; /* Number of errors seen */ + int nTab; /* Number of previously allocated VDBE cursors */ + int nMem; /* Number of memory cells used so far */ + int szOpAlloc; /* Bytes of memory space allocated for Vdbe.aOp[] */ + int iSelfTab; /* Table associated with an index on expr, or negative + ** of the base register during check-constraint eval */ + int nLabel; /* The *negative* of the number of labels used */ + int nLabelAlloc; /* Number of slots in aLabel */ + int *aLabel; /* Space to hold the labels */ + ExprList *pConstExpr;/* Constant expressions */ + Token constraintName;/* Name of the constraint currently being parsed */ + yDbMask writeMask; /* Start a write transaction on these databases */ + yDbMask cookieMask; /* Bitmask of schema verified databases */ + int regRowid; /* Register holding rowid of CREATE TABLE entry */ + int regRoot; /* Register holding root page number for new objects */ + int nMaxArg; /* Max args passed to user function by sub-program */ + int nSelect; /* Number of SELECT stmts. Counter for Select.selId */ +#ifndef SQLITE_OMIT_SHARED_CACHE + int nTableLock; /* Number of locks in aTableLock */ + TableLock *aTableLock; /* Required table locks for shared-cache mode */ +#endif + AutoincInfo *pAinc; /* Information about AUTOINCREMENT counters */ + Parse *pToplevel; /* Parse structure for main program (or NULL) */ + Table *pTriggerTab; /* Table triggers are being coded for */ + Parse *pParentParse; /* Parent parser if this parser is nested */ + int addrCrTab; /* Address of OP_CreateBtree opcode on CREATE TABLE */ + u32 nQueryLoop; /* Est number of iterations of a query (10*log2(N)) */ + u32 oldmask; /* Mask of old.* columns referenced */ + u32 newmask; /* Mask of new.* columns referenced */ + u8 eTriggerOp; /* TK_UPDATE, TK_INSERT or TK_DELETE */ + u8 eOrconf; /* Default ON CONFLICT policy for trigger steps */ + u8 disableTriggers; /* True to disable triggers */ + + /************************************************************************** + ** Fields above must be initialized to zero. The fields that follow, + ** down to the beginning of the recursive section, do not need to be + ** initialized as they will be set before being used. The boundary is + ** determined by offsetof(Parse,aTempReg). + **************************************************************************/ + + int aTempReg[8]; /* Holding area for temporary registers */ + Token sNameToken; /* Token with unqualified schema object name */ + + /************************************************************************ + ** Above is constant between recursions. Below is reset before and after + ** each recursion. The boundary between these two regions is determined + ** using offsetof(Parse,sLastToken) so the sLastToken field must be the + ** first field in the recursive region. + ************************************************************************/ + + Token sLastToken; /* The last token parsed */ + ynVar nVar; /* Number of '?' variables seen in the SQL so far */ + u8 iPkSortOrder; /* ASC or DESC for INTEGER PRIMARY KEY */ + u8 explain; /* True if the EXPLAIN flag is found on the query */ +#if !(defined(SQLITE_OMIT_VIRTUALTABLE) && defined(SQLITE_OMIT_ALTERTABLE)) + u8 eParseMode; /* PARSE_MODE_XXX constant */ +#endif +#ifndef SQLITE_OMIT_VIRTUALTABLE + int nVtabLock; /* Number of virtual tables to lock */ +#endif + int nHeight; /* Expression tree height of current sub-select */ +#ifndef SQLITE_OMIT_EXPLAIN + int addrExplain; /* Address of current OP_Explain opcode */ +#endif + VList *pVList; /* Mapping between variable names and numbers */ + Vdbe *pReprepare; /* VM being reprepared (sqlite3Reprepare()) */ + const char *zTail; /* All SQL text past the last semicolon parsed */ + Table *pNewTable; /* A table being constructed by CREATE TABLE */ + Index *pNewIndex; /* An index being constructed by CREATE INDEX. + ** Also used to hold redundant UNIQUE constraints + ** during a RENAME COLUMN */ + Trigger *pNewTrigger; /* Trigger under construct by a CREATE TRIGGER */ + const char *zAuthContext; /* The 6th parameter to db->xAuth callbacks */ +#ifndef SQLITE_OMIT_VIRTUALTABLE + Token sArg; /* Complete text of a module argument */ + Table **apVtabLock; /* Pointer to virtual tables needing locking */ +#endif + Table *pZombieTab; /* List of Table objects to delete after code gen */ + TriggerPrg *pTriggerPrg; /* Linked list of coded triggers */ + With *pWith; /* Current WITH clause, or NULL */ + With *pWithToFree; /* Free this WITH object at the end of the parse */ +#ifndef SQLITE_OMIT_ALTERTABLE + RenameToken *pRename; /* Tokens subject to renaming by ALTER TABLE */ +#endif +}; + +#define PARSE_MODE_NORMAL 0 +#define PARSE_MODE_DECLARE_VTAB 1 +#define PARSE_MODE_RENAME_COLUMN 2 +#define PARSE_MODE_RENAME_TABLE 3 + +/* +** Sizes and pointers of various parts of the Parse object. +*/ +#define PARSE_HDR_SZ offsetof(Parse,aTempReg) /* Recursive part w/o aColCache*/ +#define PARSE_RECURSE_SZ offsetof(Parse,sLastToken) /* Recursive part */ +#define PARSE_TAIL_SZ (sizeof(Parse)-PARSE_RECURSE_SZ) /* Non-recursive part */ +#define PARSE_TAIL(X) (((char*)(X))+PARSE_RECURSE_SZ) /* Pointer to tail */ + +/* +** Return true if currently inside an sqlite3_declare_vtab() call. +*/ +#ifdef SQLITE_OMIT_VIRTUALTABLE + #define IN_DECLARE_VTAB 0 +#else + #define IN_DECLARE_VTAB (pParse->eParseMode==PARSE_MODE_DECLARE_VTAB) +#endif + +#if defined(SQLITE_OMIT_ALTERTABLE) + #define IN_RENAME_OBJECT 0 +#else + #define IN_RENAME_OBJECT (pParse->eParseMode>=PARSE_MODE_RENAME_COLUMN) +#endif + +#if defined(SQLITE_OMIT_VIRTUALTABLE) && defined(SQLITE_OMIT_ALTERTABLE) + #define IN_SPECIAL_PARSE 0 +#else + #define IN_SPECIAL_PARSE (pParse->eParseMode!=PARSE_MODE_NORMAL) +#endif + +/* +** An instance of the following structure can be declared on a stack and used +** to save the Parse.zAuthContext value so that it can be restored later. +*/ +struct AuthContext { + const char *zAuthContext; /* Put saved Parse.zAuthContext here */ + Parse *pParse; /* The Parse structure */ +}; + +/* +** Bitfield flags for P5 value in various opcodes. +** +** Value constraints (enforced via assert()): +** OPFLAG_LENGTHARG == SQLITE_FUNC_LENGTH +** OPFLAG_TYPEOFARG == SQLITE_FUNC_TYPEOF +** OPFLAG_BULKCSR == BTREE_BULKLOAD +** OPFLAG_SEEKEQ == BTREE_SEEK_EQ +** OPFLAG_FORDELETE == BTREE_FORDELETE +** OPFLAG_SAVEPOSITION == BTREE_SAVEPOSITION +** OPFLAG_AUXDELETE == BTREE_AUXDELETE +*/ +#define OPFLAG_NCHANGE 0x01 /* OP_Insert: Set to update db->nChange */ + /* Also used in P2 (not P5) of OP_Delete */ +#define OPFLAG_NOCHNG 0x01 /* OP_VColumn nochange for UPDATE */ +#define OPFLAG_EPHEM 0x01 /* OP_Column: Ephemeral output is ok */ +#define OPFLAG_LASTROWID 0x20 /* Set to update db->lastRowid */ +#define OPFLAG_ISUPDATE 0x04 /* This OP_Insert is an sql UPDATE */ +#define OPFLAG_APPEND 0x08 /* This is likely to be an append */ +#define OPFLAG_USESEEKRESULT 0x10 /* Try to avoid a seek in BtreeInsert() */ +#define OPFLAG_ISNOOP 0x40 /* OP_Delete does pre-update-hook only */ +#define OPFLAG_LENGTHARG 0x40 /* OP_Column only used for length() */ +#define OPFLAG_TYPEOFARG 0x80 /* OP_Column only used for typeof() */ +#define OPFLAG_BULKCSR 0x01 /* OP_Open** used to open bulk cursor */ +#define OPFLAG_SEEKEQ 0x02 /* OP_Open** cursor uses EQ seek only */ +#define OPFLAG_FORDELETE 0x08 /* OP_Open should use BTREE_FORDELETE */ +#define OPFLAG_P2ISREG 0x10 /* P2 to OP_Open** is a register number */ +#define OPFLAG_PERMUTE 0x01 /* OP_Compare: use the permutation */ +#define OPFLAG_SAVEPOSITION 0x02 /* OP_Delete/Insert: save cursor pos */ +#define OPFLAG_AUXDELETE 0x04 /* OP_Delete: index in a DELETE op */ +#define OPFLAG_NOCHNG_MAGIC 0x6d /* OP_MakeRecord: serialtype 10 is ok */ + +/* + * Each trigger present in the database schema is stored as an instance of + * struct Trigger. + * + * Pointers to instances of struct Trigger are stored in two ways. + * 1. In the "trigHash" hash table (part of the sqlite3* that represents the + * database). This allows Trigger structures to be retrieved by name. + * 2. All triggers associated with a single table form a linked list, using the + * pNext member of struct Trigger. A pointer to the first element of the + * linked list is stored as the "pTrigger" member of the associated + * struct Table. + * + * The "step_list" member points to the first element of a linked list + * containing the SQL statements specified as the trigger program. + */ +struct Trigger { + char *zName; /* The name of the trigger */ + char *table; /* The table or view to which the trigger applies */ + u8 op; /* One of TK_DELETE, TK_UPDATE, TK_INSERT */ + u8 tr_tm; /* One of TRIGGER_BEFORE, TRIGGER_AFTER */ + Expr *pWhen; /* The WHEN clause of the expression (may be NULL) */ + IdList *pColumns; /* If this is an UPDATE OF trigger, + the is stored here */ + Schema *pSchema; /* Schema containing the trigger */ + Schema *pTabSchema; /* Schema containing the table */ + TriggerStep *step_list; /* Link list of trigger program steps */ + Trigger *pNext; /* Next trigger associated with the table */ +}; + +/* +** A trigger is either a BEFORE or an AFTER trigger. The following constants +** determine which. +** +** If there are multiple triggers, you might of some BEFORE and some AFTER. +** In that cases, the constants below can be ORed together. +*/ +#define TRIGGER_BEFORE 1 +#define TRIGGER_AFTER 2 + +/* + * An instance of struct TriggerStep is used to store a single SQL statement + * that is a part of a trigger-program. + * + * Instances of struct TriggerStep are stored in a singly linked list (linked + * using the "pNext" member) referenced by the "step_list" member of the + * associated struct Trigger instance. The first element of the linked list is + * the first step of the trigger-program. + * + * The "op" member indicates whether this is a "DELETE", "INSERT", "UPDATE" or + * "SELECT" statement. The meanings of the other members is determined by the + * value of "op" as follows: + * + * (op == TK_INSERT) + * orconf -> stores the ON CONFLICT algorithm + * pSelect -> If this is an INSERT INTO ... SELECT ... statement, then + * this stores a pointer to the SELECT statement. Otherwise NULL. + * zTarget -> Dequoted name of the table to insert into. + * pExprList -> If this is an INSERT INTO ... VALUES ... statement, then + * this stores values to be inserted. Otherwise NULL. + * pIdList -> If this is an INSERT INTO ... () VALUES ... + * statement, then this stores the column-names to be + * inserted into. + * + * (op == TK_DELETE) + * zTarget -> Dequoted name of the table to delete from. + * pWhere -> The WHERE clause of the DELETE statement if one is specified. + * Otherwise NULL. + * + * (op == TK_UPDATE) + * zTarget -> Dequoted name of the table to update. + * pWhere -> The WHERE clause of the UPDATE statement if one is specified. + * Otherwise NULL. + * pExprList -> A list of the columns to update and the expressions to update + * them to. See sqlite3Update() documentation of "pChanges" + * argument. + * + */ +struct TriggerStep { + u8 op; /* One of TK_DELETE, TK_UPDATE, TK_INSERT, TK_SELECT */ + u8 orconf; /* OE_Rollback etc. */ + Trigger *pTrig; /* The trigger that this step is a part of */ + Select *pSelect; /* SELECT statement or RHS of INSERT INTO SELECT ... */ + char *zTarget; /* Target table for DELETE, UPDATE, INSERT */ + Expr *pWhere; /* The WHERE clause for DELETE or UPDATE steps */ + ExprList *pExprList; /* SET clause for UPDATE */ + IdList *pIdList; /* Column names for INSERT */ + Upsert *pUpsert; /* Upsert clauses on an INSERT */ + char *zSpan; /* Original SQL text of this command */ + TriggerStep *pNext; /* Next in the link-list */ + TriggerStep *pLast; /* Last element in link-list. Valid for 1st elem only */ +}; + +/* +** The following structure contains information used by the sqliteFix... +** routines as they walk the parse tree to make database references +** explicit. +*/ +typedef struct DbFixer DbFixer; +struct DbFixer { + Parse *pParse; /* The parsing context. Error messages written here */ + Schema *pSchema; /* Fix items to this schema */ + int bVarOnly; /* Check for variable references only */ + const char *zDb; /* Make sure all objects are contained in this database */ + const char *zType; /* Type of the container - used for error messages */ + const Token *pName; /* Name of the container - used for error messages */ +}; + +/* +** An objected used to accumulate the text of a string where we +** do not necessarily know how big the string will be in the end. +*/ +struct sqlite3_str { + sqlite3 *db; /* Optional database for lookaside. Can be NULL */ + char *zText; /* The string collected so far */ + u32 nAlloc; /* Amount of space allocated in zText */ + u32 mxAlloc; /* Maximum allowed allocation. 0 for no malloc usage */ + u32 nChar; /* Length of the string so far */ + u8 accError; /* SQLITE_NOMEM or SQLITE_TOOBIG */ + u8 printfFlags; /* SQLITE_PRINTF flags below */ +}; +#define SQLITE_PRINTF_INTERNAL 0x01 /* Internal-use-only converters allowed */ +#define SQLITE_PRINTF_SQLFUNC 0x02 /* SQL function arguments to VXPrintf */ +#define SQLITE_PRINTF_MALLOCED 0x04 /* True if xText is allocated space */ + +#define isMalloced(X) (((X)->printfFlags & SQLITE_PRINTF_MALLOCED)!=0) + + +/* +** A pointer to this structure is used to communicate information +** from sqlite3Init and OP_ParseSchema into the sqlite3InitCallback. +*/ +typedef struct { + sqlite3 *db; /* The database being initialized */ + char **pzErrMsg; /* Error message stored here */ + int iDb; /* 0 for main database. 1 for TEMP, 2.. for ATTACHed */ + int rc; /* Result code stored here */ + u32 mInitFlags; /* Flags controlling error messages */ + u32 nInitRow; /* Number of rows processed */ +} InitData; + +/* +** Allowed values for mInitFlags +*/ +#define INITFLAG_AlterTable 0x0001 /* This is a reparse after ALTER TABLE */ + +/* +** Structure containing global configuration data for the SQLite library. +** +** This structure also contains some state information. +*/ +struct Sqlite3Config { + int bMemstat; /* True to enable memory status */ + u8 bCoreMutex; /* True to enable core mutexing */ + u8 bFullMutex; /* True to enable full mutexing */ + u8 bOpenUri; /* True to interpret filenames as URIs */ + u8 bUseCis; /* Use covering indices for full-scans */ + u8 bSmallMalloc; /* Avoid large memory allocations if true */ + u8 bExtraSchemaChecks; /* Verify type,name,tbl_name in schema */ + int mxStrlen; /* Maximum string length */ + int neverCorrupt; /* Database is always well-formed */ + int szLookaside; /* Default lookaside buffer size */ + int nLookaside; /* Default lookaside buffer count */ + int nStmtSpill; /* Stmt-journal spill-to-disk threshold */ + sqlite3_mem_methods m; /* Low-level memory allocation interface */ + sqlite3_mutex_methods mutex; /* Low-level mutex interface */ + sqlite3_pcache_methods2 pcache2; /* Low-level page-cache interface */ + void *pHeap; /* Heap storage space */ + int nHeap; /* Size of pHeap[] */ + int mnReq, mxReq; /* Min and max heap requests sizes */ + sqlite3_int64 szMmap; /* mmap() space per open file */ + sqlite3_int64 mxMmap; /* Maximum value for szMmap */ + void *pPage; /* Page cache memory */ + int szPage; /* Size of each page in pPage[] */ + int nPage; /* Number of pages in pPage[] */ + int mxParserStack; /* maximum depth of the parser stack */ + int sharedCacheEnabled; /* true if shared-cache mode enabled */ + u32 szPma; /* Maximum Sorter PMA size */ + /* The above might be initialized to non-zero. The following need to always + ** initially be zero, however. */ + int isInit; /* True after initialization has finished */ + int inProgress; /* True while initialization in progress */ + int isMutexInit; /* True after mutexes are initialized */ + int isMallocInit; /* True after malloc is initialized */ + int isPCacheInit; /* True after malloc is initialized */ + int nRefInitMutex; /* Number of users of pInitMutex */ + sqlite3_mutex *pInitMutex; /* Mutex used by sqlite3_initialize() */ + void (*xLog)(void*,int,const char*); /* Function for logging */ + void *pLogArg; /* First argument to xLog() */ +#ifdef SQLITE_ENABLE_SQLLOG + void(*xSqllog)(void*,sqlite3*,const char*, int); + void *pSqllogArg; +#endif +#ifdef SQLITE_VDBE_COVERAGE + /* The following callback (if not NULL) is invoked on every VDBE branch + ** operation. Set the callback using SQLITE_TESTCTRL_VDBE_COVERAGE. + */ + void (*xVdbeBranch)(void*,unsigned iSrcLine,u8 eThis,u8 eMx); /* Callback */ + void *pVdbeBranchArg; /* 1st argument */ +#endif +#ifdef SQLITE_ENABLE_DESERIALIZE + sqlite3_int64 mxMemdbSize; /* Default max memdb size */ +#endif +#ifndef SQLITE_UNTESTABLE + int (*xTestCallback)(int); /* Invoked by sqlite3FaultSim() */ +#endif + int bLocaltimeFault; /* True to fail localtime() calls */ + int bInternalFunctions; /* Internal SQL functions are visible */ + int iOnceResetThreshold; /* When to reset OP_Once counters */ + u32 szSorterRef; /* Min size in bytes to use sorter-refs */ + unsigned int iPrngSeed; /* Alternative fixed seed for the PRNG */ +}; + +/* +** This macro is used inside of assert() statements to indicate that +** the assert is only valid on a well-formed database. Instead of: +** +** assert( X ); +** +** One writes: +** +** assert( X || CORRUPT_DB ); +** +** CORRUPT_DB is true during normal operation. CORRUPT_DB does not indicate +** that the database is definitely corrupt, only that it might be corrupt. +** For most test cases, CORRUPT_DB is set to false using a special +** sqlite3_test_control(). This enables assert() statements to prove +** things that are always true for well-formed databases. +*/ +#define CORRUPT_DB (sqlite3Config.neverCorrupt==0) + +/* +** Context pointer passed down through the tree-walk. +*/ +struct Walker { + Parse *pParse; /* Parser context. */ + int (*xExprCallback)(Walker*, Expr*); /* Callback for expressions */ + int (*xSelectCallback)(Walker*,Select*); /* Callback for SELECTs */ + void (*xSelectCallback2)(Walker*,Select*);/* Second callback for SELECTs */ + int walkerDepth; /* Number of subqueries */ + u8 eCode; /* A small processing code */ + union { /* Extra data for callback */ + NameContext *pNC; /* Naming context */ + int n; /* A counter */ + int iCur; /* A cursor number */ + SrcList *pSrcList; /* FROM clause */ + struct SrcCount *pSrcCount; /* Counting column references */ + struct CCurHint *pCCurHint; /* Used by codeCursorHint() */ + int *aiCol; /* array of column indexes */ + struct IdxCover *pIdxCover; /* Check for index coverage */ + struct IdxExprTrans *pIdxTrans; /* Convert idxed expr to column */ + ExprList *pGroupBy; /* GROUP BY clause */ + Select *pSelect; /* HAVING to WHERE clause ctx */ + struct WindowRewrite *pRewrite; /* Window rewrite context */ + struct WhereConst *pConst; /* WHERE clause constants */ + struct RenameCtx *pRename; /* RENAME COLUMN context */ + } u; +}; + +/* Forward declarations */ +SQLITE_PRIVATE int sqlite3WalkExpr(Walker*, Expr*); +SQLITE_PRIVATE int sqlite3WalkExprList(Walker*, ExprList*); +SQLITE_PRIVATE int sqlite3WalkSelect(Walker*, Select*); +SQLITE_PRIVATE int sqlite3WalkSelectExpr(Walker*, Select*); +SQLITE_PRIVATE int sqlite3WalkSelectFrom(Walker*, Select*); +SQLITE_PRIVATE int sqlite3ExprWalkNoop(Walker*, Expr*); +SQLITE_PRIVATE int sqlite3SelectWalkNoop(Walker*, Select*); +SQLITE_PRIVATE int sqlite3SelectWalkFail(Walker*, Select*); +#ifdef SQLITE_DEBUG +SQLITE_PRIVATE void sqlite3SelectWalkAssert2(Walker*, Select*); +#endif + +/* +** Return code from the parse-tree walking primitives and their +** callbacks. +*/ +#define WRC_Continue 0 /* Continue down into children */ +#define WRC_Prune 1 /* Omit children but continue walking siblings */ +#define WRC_Abort 2 /* Abandon the tree walk */ + +/* +** An instance of this structure represents a set of one or more CTEs +** (common table expressions) created by a single WITH clause. +*/ +struct With { + int nCte; /* Number of CTEs in the WITH clause */ + With *pOuter; /* Containing WITH clause, or NULL */ + struct Cte { /* For each CTE in the WITH clause.... */ + char *zName; /* Name of this CTE */ + ExprList *pCols; /* List of explicit column names, or NULL */ + Select *pSelect; /* The definition of this CTE */ + const char *zCteErr; /* Error message for circular references */ + } a[1]; +}; + +#ifdef SQLITE_DEBUG +/* +** An instance of the TreeView object is used for printing the content of +** data structures on sqlite3DebugPrintf() using a tree-like view. +*/ +struct TreeView { + int iLevel; /* Which level of the tree we are on */ + u8 bLine[100]; /* Draw vertical in column i if bLine[i] is true */ +}; +#endif /* SQLITE_DEBUG */ + +/* +** This object is used in various ways, most (but not all) related to window +** functions. +** +** (1) A single instance of this structure is attached to the +** the Expr.y.pWin field for each window function in an expression tree. +** This object holds the information contained in the OVER clause, +** plus additional fields used during code generation. +** +** (2) All window functions in a single SELECT form a linked-list +** attached to Select.pWin. The Window.pFunc and Window.pExpr +** fields point back to the expression that is the window function. +** +** (3) The terms of the WINDOW clause of a SELECT are instances of this +** object on a linked list attached to Select.pWinDefn. +** +** (4) For an aggregate function with a FILTER clause, an instance +** of this object is stored in Expr.y.pWin with eFrmType set to +** TK_FILTER. In this case the only field used is Window.pFilter. +** +** The uses (1) and (2) are really the same Window object that just happens +** to be accessible in two different ways. Use case (3) are separate objects. +*/ +struct Window { + char *zName; /* Name of window (may be NULL) */ + char *zBase; /* Name of base window for chaining (may be NULL) */ + ExprList *pPartition; /* PARTITION BY clause */ + ExprList *pOrderBy; /* ORDER BY clause */ + u8 eFrmType; /* TK_RANGE, TK_GROUPS, TK_ROWS, or 0 */ + u8 eStart; /* UNBOUNDED, CURRENT, PRECEDING or FOLLOWING */ + u8 eEnd; /* UNBOUNDED, CURRENT, PRECEDING or FOLLOWING */ + u8 bImplicitFrame; /* True if frame was implicitly specified */ + u8 eExclude; /* TK_NO, TK_CURRENT, TK_TIES, TK_GROUP, or 0 */ + Expr *pStart; /* Expression for " PRECEDING" */ + Expr *pEnd; /* Expression for " FOLLOWING" */ + Window **ppThis; /* Pointer to this object in Select.pWin list */ + Window *pNextWin; /* Next window function belonging to this SELECT */ + Expr *pFilter; /* The FILTER expression */ + FuncDef *pFunc; /* The function */ + int iEphCsr; /* Partition buffer or Peer buffer */ + int regAccum; /* Accumulator */ + int regResult; /* Interim result */ + int csrApp; /* Function cursor (used by min/max) */ + int regApp; /* Function register (also used by min/max) */ + int regPart; /* Array of registers for PARTITION BY values */ + Expr *pOwner; /* Expression object this window is attached to */ + int nBufferCol; /* Number of columns in buffer table */ + int iArgCol; /* Offset of first argument for this function */ + int regOne; /* Register containing constant value 1 */ + int regStartRowid; + int regEndRowid; + u8 bExprArgs; /* Defer evaluation of window function arguments + ** due to the SQLITE_SUBTYPE flag */ +}; + +#ifndef SQLITE_OMIT_WINDOWFUNC +SQLITE_PRIVATE void sqlite3WindowDelete(sqlite3*, Window*); +SQLITE_PRIVATE void sqlite3WindowUnlinkFromSelect(Window*); +SQLITE_PRIVATE void sqlite3WindowListDelete(sqlite3 *db, Window *p); +SQLITE_PRIVATE Window *sqlite3WindowAlloc(Parse*, int, int, Expr*, int , Expr*, u8); +SQLITE_PRIVATE void sqlite3WindowAttach(Parse*, Expr*, Window*); +SQLITE_PRIVATE void sqlite3WindowLink(Select *pSel, Window *pWin); +SQLITE_PRIVATE int sqlite3WindowCompare(Parse*, Window*, Window*, int); +SQLITE_PRIVATE void sqlite3WindowCodeInit(Parse*, Window*); +SQLITE_PRIVATE void sqlite3WindowCodeStep(Parse*, Select*, WhereInfo*, int, int); +SQLITE_PRIVATE int sqlite3WindowRewrite(Parse*, Select*); +SQLITE_PRIVATE int sqlite3ExpandSubquery(Parse*, struct SrcList_item*); +SQLITE_PRIVATE void sqlite3WindowUpdate(Parse*, Window*, Window*, FuncDef*); +SQLITE_PRIVATE Window *sqlite3WindowDup(sqlite3 *db, Expr *pOwner, Window *p); +SQLITE_PRIVATE Window *sqlite3WindowListDup(sqlite3 *db, Window *p); +SQLITE_PRIVATE void sqlite3WindowFunctions(void); +SQLITE_PRIVATE void sqlite3WindowChain(Parse*, Window*, Window*); +SQLITE_PRIVATE Window *sqlite3WindowAssemble(Parse*, Window*, ExprList*, ExprList*, Token*); +#else +# define sqlite3WindowDelete(a,b) +# define sqlite3WindowFunctions() +# define sqlite3WindowAttach(a,b,c) +#endif + +/* +** Assuming zIn points to the first byte of a UTF-8 character, +** advance zIn to point to the first byte of the next UTF-8 character. +*/ +#define SQLITE_SKIP_UTF8(zIn) { \ + if( (*(zIn++))>=0xc0 ){ \ + while( (*zIn & 0xc0)==0x80 ){ zIn++; } \ + } \ +} + +/* +** The SQLITE_*_BKPT macros are substitutes for the error codes with +** the same name but without the _BKPT suffix. These macros invoke +** routines that report the line-number on which the error originated +** using sqlite3_log(). The routines also provide a convenient place +** to set a debugger breakpoint. +*/ +SQLITE_PRIVATE int sqlite3ReportError(int iErr, int lineno, const char *zType); +SQLITE_PRIVATE int sqlite3CorruptError(int); +SQLITE_PRIVATE int sqlite3MisuseError(int); +SQLITE_PRIVATE int sqlite3CantopenError(int); +#define SQLITE_CORRUPT_BKPT sqlite3CorruptError(__LINE__) +#define SQLITE_MISUSE_BKPT sqlite3MisuseError(__LINE__) +#define SQLITE_CANTOPEN_BKPT sqlite3CantopenError(__LINE__) +#ifdef SQLITE_DEBUG +SQLITE_PRIVATE int sqlite3NomemError(int); +SQLITE_PRIVATE int sqlite3IoerrnomemError(int); +SQLITE_PRIVATE int sqlite3CorruptPgnoError(int,Pgno); +# define SQLITE_NOMEM_BKPT sqlite3NomemError(__LINE__) +# define SQLITE_IOERR_NOMEM_BKPT sqlite3IoerrnomemError(__LINE__) +# define SQLITE_CORRUPT_PGNO(P) sqlite3CorruptPgnoError(__LINE__,(P)) +#else +# define SQLITE_NOMEM_BKPT SQLITE_NOMEM +# define SQLITE_IOERR_NOMEM_BKPT SQLITE_IOERR_NOMEM +# define SQLITE_CORRUPT_PGNO(P) sqlite3CorruptError(__LINE__) +#endif + +/* +** FTS3 and FTS4 both require virtual table support +*/ +#if defined(SQLITE_OMIT_VIRTUALTABLE) +# undef SQLITE_ENABLE_FTS3 +# undef SQLITE_ENABLE_FTS4 +#endif + +/* +** FTS4 is really an extension for FTS3. It is enabled using the +** SQLITE_ENABLE_FTS3 macro. But to avoid confusion we also call +** the SQLITE_ENABLE_FTS4 macro to serve as an alias for SQLITE_ENABLE_FTS3. +*/ +#if defined(SQLITE_ENABLE_FTS4) && !defined(SQLITE_ENABLE_FTS3) +# define SQLITE_ENABLE_FTS3 1 +#endif + +/* +** The ctype.h header is needed for non-ASCII systems. It is also +** needed by FTS3 when FTS3 is included in the amalgamation. +*/ +#if !defined(SQLITE_ASCII) || \ + (defined(SQLITE_ENABLE_FTS3) && defined(SQLITE_AMALGAMATION)) +# include +#endif + +/* +** The following macros mimic the standard library functions toupper(), +** isspace(), isalnum(), isdigit() and isxdigit(), respectively. The +** sqlite versions only work for ASCII characters, regardless of locale. +*/ +#ifdef SQLITE_ASCII +# define sqlite3Toupper(x) ((x)&~(sqlite3CtypeMap[(unsigned char)(x)]&0x20)) +# define sqlite3Isspace(x) (sqlite3CtypeMap[(unsigned char)(x)]&0x01) +# define sqlite3Isalnum(x) (sqlite3CtypeMap[(unsigned char)(x)]&0x06) +# define sqlite3Isalpha(x) (sqlite3CtypeMap[(unsigned char)(x)]&0x02) +# define sqlite3Isdigit(x) (sqlite3CtypeMap[(unsigned char)(x)]&0x04) +# define sqlite3Isxdigit(x) (sqlite3CtypeMap[(unsigned char)(x)]&0x08) +# define sqlite3Tolower(x) (sqlite3UpperToLower[(unsigned char)(x)]) +# define sqlite3Isquote(x) (sqlite3CtypeMap[(unsigned char)(x)]&0x80) +#else +# define sqlite3Toupper(x) toupper((unsigned char)(x)) +# define sqlite3Isspace(x) isspace((unsigned char)(x)) +# define sqlite3Isalnum(x) isalnum((unsigned char)(x)) +# define sqlite3Isalpha(x) isalpha((unsigned char)(x)) +# define sqlite3Isdigit(x) isdigit((unsigned char)(x)) +# define sqlite3Isxdigit(x) isxdigit((unsigned char)(x)) +# define sqlite3Tolower(x) tolower((unsigned char)(x)) +# define sqlite3Isquote(x) ((x)=='"'||(x)=='\''||(x)=='['||(x)=='`') +#endif +SQLITE_PRIVATE int sqlite3IsIdChar(u8); + +/* +** Internal function prototypes +*/ +SQLITE_PRIVATE int sqlite3StrICmp(const char*,const char*); +SQLITE_PRIVATE int sqlite3Strlen30(const char*); +#define sqlite3Strlen30NN(C) (strlen(C)&0x3fffffff) +SQLITE_PRIVATE char *sqlite3ColumnType(Column*,char*); +#define sqlite3StrNICmp sqlite3_strnicmp + +SQLITE_PRIVATE int sqlite3MallocInit(void); +SQLITE_PRIVATE void sqlite3MallocEnd(void); +SQLITE_PRIVATE void *sqlite3Malloc(u64); +SQLITE_PRIVATE void *sqlite3MallocZero(u64); +SQLITE_PRIVATE void *sqlite3DbMallocZero(sqlite3*, u64); +SQLITE_PRIVATE void *sqlite3DbMallocRaw(sqlite3*, u64); +SQLITE_PRIVATE void *sqlite3DbMallocRawNN(sqlite3*, u64); +SQLITE_PRIVATE char *sqlite3DbStrDup(sqlite3*,const char*); +SQLITE_PRIVATE char *sqlite3DbStrNDup(sqlite3*,const char*, u64); +SQLITE_PRIVATE char *sqlite3DbSpanDup(sqlite3*,const char*,const char*); +SQLITE_PRIVATE void *sqlite3Realloc(void*, u64); +SQLITE_PRIVATE void *sqlite3DbReallocOrFree(sqlite3 *, void *, u64); +SQLITE_PRIVATE void *sqlite3DbRealloc(sqlite3 *, void *, u64); +SQLITE_PRIVATE void sqlite3DbFree(sqlite3*, void*); +SQLITE_PRIVATE void sqlite3DbFreeNN(sqlite3*, void*); +SQLITE_PRIVATE int sqlite3MallocSize(void*); +SQLITE_PRIVATE int sqlite3DbMallocSize(sqlite3*, void*); +SQLITE_PRIVATE void *sqlite3PageMalloc(int); +SQLITE_PRIVATE void sqlite3PageFree(void*); +SQLITE_PRIVATE void sqlite3MemSetDefault(void); +#ifndef SQLITE_UNTESTABLE +SQLITE_PRIVATE void sqlite3BenignMallocHooks(void (*)(void), void (*)(void)); +#endif +SQLITE_PRIVATE int sqlite3HeapNearlyFull(void); + +/* +** On systems with ample stack space and that support alloca(), make +** use of alloca() to obtain space for large automatic objects. By default, +** obtain space from malloc(). +** +** The alloca() routine never returns NULL. This will cause code paths +** that deal with sqlite3StackAlloc() failures to be unreachable. +*/ +#ifdef SQLITE_USE_ALLOCA +# define sqlite3StackAllocRaw(D,N) alloca(N) +# define sqlite3StackAllocZero(D,N) memset(alloca(N), 0, N) +# define sqlite3StackFree(D,P) +#else +# define sqlite3StackAllocRaw(D,N) sqlite3DbMallocRaw(D,N) +# define sqlite3StackAllocZero(D,N) sqlite3DbMallocZero(D,N) +# define sqlite3StackFree(D,P) sqlite3DbFree(D,P) +#endif + +/* Do not allow both MEMSYS5 and MEMSYS3 to be defined together. If they +** are, disable MEMSYS3 +*/ +#ifdef SQLITE_ENABLE_MEMSYS5 +SQLITE_PRIVATE const sqlite3_mem_methods *sqlite3MemGetMemsys5(void); +#undef SQLITE_ENABLE_MEMSYS3 +#endif +#ifdef SQLITE_ENABLE_MEMSYS3 +SQLITE_PRIVATE const sqlite3_mem_methods *sqlite3MemGetMemsys3(void); +#endif + + +#ifndef SQLITE_MUTEX_OMIT +SQLITE_PRIVATE sqlite3_mutex_methods const *sqlite3DefaultMutex(void); +SQLITE_PRIVATE sqlite3_mutex_methods const *sqlite3NoopMutex(void); +SQLITE_PRIVATE sqlite3_mutex *sqlite3MutexAlloc(int); +SQLITE_PRIVATE int sqlite3MutexInit(void); +SQLITE_PRIVATE int sqlite3MutexEnd(void); +#endif +#if !defined(SQLITE_MUTEX_OMIT) && !defined(SQLITE_MUTEX_NOOP) +SQLITE_PRIVATE void sqlite3MemoryBarrier(void); +#else +# define sqlite3MemoryBarrier() +#endif + +SQLITE_PRIVATE sqlite3_int64 sqlite3StatusValue(int); +SQLITE_PRIVATE void sqlite3StatusUp(int, int); +SQLITE_PRIVATE void sqlite3StatusDown(int, int); +SQLITE_PRIVATE void sqlite3StatusHighwater(int, int); +SQLITE_PRIVATE int sqlite3LookasideUsed(sqlite3*,int*); + +/* Access to mutexes used by sqlite3_status() */ +SQLITE_PRIVATE sqlite3_mutex *sqlite3Pcache1Mutex(void); +SQLITE_PRIVATE sqlite3_mutex *sqlite3MallocMutex(void); + +#if defined(SQLITE_ENABLE_MULTITHREADED_CHECKS) && !defined(SQLITE_MUTEX_OMIT) +SQLITE_PRIVATE void sqlite3MutexWarnOnContention(sqlite3_mutex*); +#else +# define sqlite3MutexWarnOnContention(x) +#endif + +#ifndef SQLITE_OMIT_FLOATING_POINT +# define EXP754 (((u64)0x7ff)<<52) +# define MAN754 ((((u64)1)<<52)-1) +# define IsNaN(X) (((X)&EXP754)==EXP754 && ((X)&MAN754)!=0) +SQLITE_PRIVATE int sqlite3IsNaN(double); +#else +# define IsNaN(X) 0 +# define sqlite3IsNaN(X) 0 +#endif + +/* +** An instance of the following structure holds information about SQL +** functions arguments that are the parameters to the printf() function. +*/ +struct PrintfArguments { + int nArg; /* Total number of arguments */ + int nUsed; /* Number of arguments used so far */ + sqlite3_value **apArg; /* The argument values */ +}; + +SQLITE_PRIVATE char *sqlite3MPrintf(sqlite3*,const char*, ...); +SQLITE_PRIVATE char *sqlite3VMPrintf(sqlite3*,const char*, va_list); +#if defined(SQLITE_DEBUG) || defined(SQLITE_HAVE_OS_TRACE) +SQLITE_PRIVATE void sqlite3DebugPrintf(const char*, ...); +#endif +#if defined(SQLITE_TEST) +SQLITE_PRIVATE void *sqlite3TestTextToPtr(const char*); +#endif + +#if defined(SQLITE_DEBUG) +SQLITE_PRIVATE void sqlite3TreeViewExpr(TreeView*, const Expr*, u8); +SQLITE_PRIVATE void sqlite3TreeViewBareExprList(TreeView*, const ExprList*, const char*); +SQLITE_PRIVATE void sqlite3TreeViewExprList(TreeView*, const ExprList*, u8, const char*); +SQLITE_PRIVATE void sqlite3TreeViewSrcList(TreeView*, const SrcList*); +SQLITE_PRIVATE void sqlite3TreeViewSelect(TreeView*, const Select*, u8); +SQLITE_PRIVATE void sqlite3TreeViewWith(TreeView*, const With*, u8); +#ifndef SQLITE_OMIT_WINDOWFUNC +SQLITE_PRIVATE void sqlite3TreeViewWindow(TreeView*, const Window*, u8); +SQLITE_PRIVATE void sqlite3TreeViewWinFunc(TreeView*, const Window*, u8); +#endif +#endif + + +SQLITE_PRIVATE void sqlite3SetString(char **, sqlite3*, const char*); +SQLITE_PRIVATE void sqlite3ErrorMsg(Parse*, const char*, ...); +SQLITE_PRIVATE int sqlite3ErrorToParser(sqlite3*,int); +SQLITE_PRIVATE void sqlite3Dequote(char*); +SQLITE_PRIVATE void sqlite3DequoteExpr(Expr*); +SQLITE_PRIVATE void sqlite3TokenInit(Token*,char*); +SQLITE_PRIVATE int sqlite3KeywordCode(const unsigned char*, int); +SQLITE_PRIVATE int sqlite3RunParser(Parse*, const char*, char **); +SQLITE_PRIVATE void sqlite3FinishCoding(Parse*); +SQLITE_PRIVATE int sqlite3GetTempReg(Parse*); +SQLITE_PRIVATE void sqlite3ReleaseTempReg(Parse*,int); +SQLITE_PRIVATE int sqlite3GetTempRange(Parse*,int); +SQLITE_PRIVATE void sqlite3ReleaseTempRange(Parse*,int,int); +SQLITE_PRIVATE void sqlite3ClearTempRegCache(Parse*); +#ifdef SQLITE_DEBUG +SQLITE_PRIVATE int sqlite3NoTempsInRange(Parse*,int,int); +#endif +SQLITE_PRIVATE Expr *sqlite3ExprAlloc(sqlite3*,int,const Token*,int); +SQLITE_PRIVATE Expr *sqlite3Expr(sqlite3*,int,const char*); +SQLITE_PRIVATE void sqlite3ExprAttachSubtrees(sqlite3*,Expr*,Expr*,Expr*); +SQLITE_PRIVATE Expr *sqlite3PExpr(Parse*, int, Expr*, Expr*); +SQLITE_PRIVATE void sqlite3PExprAddSelect(Parse*, Expr*, Select*); +SQLITE_PRIVATE Expr *sqlite3ExprAnd(Parse*,Expr*, Expr*); +SQLITE_PRIVATE Expr *sqlite3ExprSimplifiedAndOr(Expr*); +SQLITE_PRIVATE Expr *sqlite3ExprFunction(Parse*,ExprList*, Token*, int); +SQLITE_PRIVATE void sqlite3ExprAssignVarNumber(Parse*, Expr*, u32); +SQLITE_PRIVATE void sqlite3ExprDelete(sqlite3*, Expr*); +SQLITE_PRIVATE void sqlite3ExprUnmapAndDelete(Parse*, Expr*); +SQLITE_PRIVATE ExprList *sqlite3ExprListAppend(Parse*,ExprList*,Expr*); +SQLITE_PRIVATE ExprList *sqlite3ExprListAppendVector(Parse*,ExprList*,IdList*,Expr*); +SQLITE_PRIVATE void sqlite3ExprListSetSortOrder(ExprList*,int,int); +SQLITE_PRIVATE void sqlite3ExprListSetName(Parse*,ExprList*,Token*,int); +SQLITE_PRIVATE void sqlite3ExprListSetSpan(Parse*,ExprList*,const char*,const char*); +SQLITE_PRIVATE void sqlite3ExprListDelete(sqlite3*, ExprList*); +SQLITE_PRIVATE u32 sqlite3ExprListFlags(const ExprList*); +SQLITE_PRIVATE int sqlite3IndexHasDuplicateRootPage(Index*); +SQLITE_PRIVATE int sqlite3Init(sqlite3*, char**); +SQLITE_PRIVATE int sqlite3InitCallback(void*, int, char**, char**); +SQLITE_PRIVATE int sqlite3InitOne(sqlite3*, int, char**, u32); +SQLITE_PRIVATE void sqlite3Pragma(Parse*,Token*,Token*,Token*,int); +#ifndef SQLITE_OMIT_VIRTUALTABLE +SQLITE_PRIVATE Module *sqlite3PragmaVtabRegister(sqlite3*,const char *zName); +#endif +SQLITE_PRIVATE void sqlite3ResetAllSchemasOfConnection(sqlite3*); +SQLITE_PRIVATE void sqlite3ResetOneSchema(sqlite3*,int); +SQLITE_PRIVATE void sqlite3CollapseDatabaseArray(sqlite3*); +SQLITE_PRIVATE void sqlite3CommitInternalChanges(sqlite3*); +SQLITE_PRIVATE void sqlite3DeleteColumnNames(sqlite3*,Table*); +SQLITE_PRIVATE int sqlite3ColumnsFromExprList(Parse*,ExprList*,i16*,Column**); +SQLITE_PRIVATE void sqlite3SelectAddColumnTypeAndCollation(Parse*,Table*,Select*,char); +SQLITE_PRIVATE Table *sqlite3ResultSetOfSelect(Parse*,Select*,char); +SQLITE_PRIVATE void sqlite3OpenMasterTable(Parse *, int); +SQLITE_PRIVATE Index *sqlite3PrimaryKeyIndex(Table*); +SQLITE_PRIVATE i16 sqlite3ColumnOfIndex(Index*, i16); +SQLITE_PRIVATE void sqlite3StartTable(Parse*,Token*,Token*,int,int,int,int); +#if SQLITE_ENABLE_HIDDEN_COLUMNS +SQLITE_PRIVATE void sqlite3ColumnPropertiesFromName(Table*, Column*); +#else +# define sqlite3ColumnPropertiesFromName(T,C) /* no-op */ +#endif +SQLITE_PRIVATE void sqlite3AddColumn(Parse*,Token*,Token*); +SQLITE_PRIVATE void sqlite3AddNotNull(Parse*, int); +SQLITE_PRIVATE void sqlite3AddPrimaryKey(Parse*, ExprList*, int, int, int); +SQLITE_PRIVATE void sqlite3AddCheckConstraint(Parse*, Expr*); +SQLITE_PRIVATE void sqlite3AddDefaultValue(Parse*,Expr*,const char*,const char*); +SQLITE_PRIVATE void sqlite3AddCollateType(Parse*, Token*); +SQLITE_PRIVATE void sqlite3EndTable(Parse*,Token*,Token*,u8,Select*); +SQLITE_PRIVATE int sqlite3ParseUri(const char*,const char*,unsigned int*, + sqlite3_vfs**,char**,char **); +#ifdef SQLITE_HAS_CODEC +SQLITE_PRIVATE int sqlite3CodecQueryParameters(sqlite3*,const char*,const char*); +#else +# define sqlite3CodecQueryParameters(A,B,C) 0 +#endif +SQLITE_PRIVATE Btree *sqlite3DbNameToBtree(sqlite3*,const char*); + +#ifdef SQLITE_UNTESTABLE +# define sqlite3FaultSim(X) SQLITE_OK +#else +SQLITE_PRIVATE int sqlite3FaultSim(int); +#endif + +SQLITE_PRIVATE Bitvec *sqlite3BitvecCreate(u32); +SQLITE_PRIVATE int sqlite3BitvecTest(Bitvec*, u32); +SQLITE_PRIVATE int sqlite3BitvecTestNotNull(Bitvec*, u32); +SQLITE_PRIVATE int sqlite3BitvecSet(Bitvec*, u32); +SQLITE_PRIVATE void sqlite3BitvecClear(Bitvec*, u32, void*); +SQLITE_PRIVATE void sqlite3BitvecDestroy(Bitvec*); +SQLITE_PRIVATE u32 sqlite3BitvecSize(Bitvec*); +#ifndef SQLITE_UNTESTABLE +SQLITE_PRIVATE int sqlite3BitvecBuiltinTest(int,int*); +#endif + +SQLITE_PRIVATE RowSet *sqlite3RowSetInit(sqlite3*); +SQLITE_PRIVATE void sqlite3RowSetDelete(void*); +SQLITE_PRIVATE void sqlite3RowSetClear(void*); +SQLITE_PRIVATE void sqlite3RowSetInsert(RowSet*, i64); +SQLITE_PRIVATE int sqlite3RowSetTest(RowSet*, int iBatch, i64); +SQLITE_PRIVATE int sqlite3RowSetNext(RowSet*, i64*); + +SQLITE_PRIVATE void sqlite3CreateView(Parse*,Token*,Token*,Token*,ExprList*,Select*,int,int); + +#if !defined(SQLITE_OMIT_VIEW) || !defined(SQLITE_OMIT_VIRTUALTABLE) +SQLITE_PRIVATE int sqlite3ViewGetColumnNames(Parse*,Table*); +#else +# define sqlite3ViewGetColumnNames(A,B) 0 +#endif + +#if SQLITE_MAX_ATTACHED>30 +SQLITE_PRIVATE int sqlite3DbMaskAllZero(yDbMask); +#endif +SQLITE_PRIVATE void sqlite3DropTable(Parse*, SrcList*, int, int); +SQLITE_PRIVATE void sqlite3CodeDropTable(Parse*, Table*, int, int); +SQLITE_PRIVATE void sqlite3DeleteTable(sqlite3*, Table*); +SQLITE_PRIVATE void sqlite3FreeIndex(sqlite3*, Index*); +#ifndef SQLITE_OMIT_AUTOINCREMENT +SQLITE_PRIVATE void sqlite3AutoincrementBegin(Parse *pParse); +SQLITE_PRIVATE void sqlite3AutoincrementEnd(Parse *pParse); +#else +# define sqlite3AutoincrementBegin(X) +# define sqlite3AutoincrementEnd(X) +#endif +SQLITE_PRIVATE void sqlite3Insert(Parse*, SrcList*, Select*, IdList*, int, Upsert*); +SQLITE_PRIVATE void *sqlite3ArrayAllocate(sqlite3*,void*,int,int*,int*); +SQLITE_PRIVATE IdList *sqlite3IdListAppend(Parse*, IdList*, Token*); +SQLITE_PRIVATE int sqlite3IdListIndex(IdList*,const char*); +SQLITE_PRIVATE SrcList *sqlite3SrcListEnlarge(Parse*, SrcList*, int, int); +SQLITE_PRIVATE SrcList *sqlite3SrcListAppend(Parse*, SrcList*, Token*, Token*); +SQLITE_PRIVATE SrcList *sqlite3SrcListAppendFromTerm(Parse*, SrcList*, Token*, Token*, + Token*, Select*, Expr*, IdList*); +SQLITE_PRIVATE void sqlite3SrcListIndexedBy(Parse *, SrcList *, Token *); +SQLITE_PRIVATE void sqlite3SrcListFuncArgs(Parse*, SrcList*, ExprList*); +SQLITE_PRIVATE int sqlite3IndexedByLookup(Parse *, struct SrcList_item *); +SQLITE_PRIVATE void sqlite3SrcListShiftJoinType(SrcList*); +SQLITE_PRIVATE void sqlite3SrcListAssignCursors(Parse*, SrcList*); +SQLITE_PRIVATE void sqlite3IdListDelete(sqlite3*, IdList*); +SQLITE_PRIVATE void sqlite3SrcListDelete(sqlite3*, SrcList*); +SQLITE_PRIVATE Index *sqlite3AllocateIndexObject(sqlite3*,i16,int,char**); +SQLITE_PRIVATE void sqlite3CreateIndex(Parse*,Token*,Token*,SrcList*,ExprList*,int,Token*, + Expr*, int, int, u8); +SQLITE_PRIVATE void sqlite3DropIndex(Parse*, SrcList*, int); +SQLITE_PRIVATE int sqlite3Select(Parse*, Select*, SelectDest*); +SQLITE_PRIVATE Select *sqlite3SelectNew(Parse*,ExprList*,SrcList*,Expr*,ExprList*, + Expr*,ExprList*,u32,Expr*); +SQLITE_PRIVATE void sqlite3SelectDelete(sqlite3*, Select*); +SQLITE_PRIVATE Table *sqlite3SrcListLookup(Parse*, SrcList*); +SQLITE_PRIVATE int sqlite3IsReadOnly(Parse*, Table*, int); +SQLITE_PRIVATE void sqlite3OpenTable(Parse*, int iCur, int iDb, Table*, int); +#if defined(SQLITE_ENABLE_UPDATE_DELETE_LIMIT) && !defined(SQLITE_OMIT_SUBQUERY) +SQLITE_PRIVATE Expr *sqlite3LimitWhere(Parse*,SrcList*,Expr*,ExprList*,Expr*,char*); +#endif +SQLITE_PRIVATE void sqlite3DeleteFrom(Parse*, SrcList*, Expr*, ExprList*, Expr*); +SQLITE_PRIVATE void sqlite3Update(Parse*, SrcList*, ExprList*,Expr*,int,ExprList*,Expr*, + Upsert*); +SQLITE_PRIVATE WhereInfo *sqlite3WhereBegin(Parse*,SrcList*,Expr*,ExprList*,ExprList*,u16,int); +SQLITE_PRIVATE void sqlite3WhereEnd(WhereInfo*); +SQLITE_PRIVATE LogEst sqlite3WhereOutputRowCount(WhereInfo*); +SQLITE_PRIVATE int sqlite3WhereIsDistinct(WhereInfo*); +SQLITE_PRIVATE int sqlite3WhereIsOrdered(WhereInfo*); +SQLITE_PRIVATE int sqlite3WhereOrderByLimitOptLabel(WhereInfo*); +SQLITE_PRIVATE int sqlite3WhereIsSorted(WhereInfo*); +SQLITE_PRIVATE int sqlite3WhereContinueLabel(WhereInfo*); +SQLITE_PRIVATE int sqlite3WhereBreakLabel(WhereInfo*); +SQLITE_PRIVATE int sqlite3WhereOkOnePass(WhereInfo*, int*); +#define ONEPASS_OFF 0 /* Use of ONEPASS not allowed */ +#define ONEPASS_SINGLE 1 /* ONEPASS valid for a single row update */ +#define ONEPASS_MULTI 2 /* ONEPASS is valid for multiple rows */ +SQLITE_PRIVATE void sqlite3ExprCodeLoadIndexColumn(Parse*, Index*, int, int, int); +SQLITE_PRIVATE int sqlite3ExprCodeGetColumn(Parse*, Table*, int, int, int, u8); +SQLITE_PRIVATE void sqlite3ExprCodeGetColumnOfTable(Vdbe*, Table*, int, int, int); +SQLITE_PRIVATE void sqlite3ExprCodeMove(Parse*, int, int, int); +SQLITE_PRIVATE void sqlite3ExprCode(Parse*, Expr*, int); +SQLITE_PRIVATE void sqlite3ExprCodeCopy(Parse*, Expr*, int); +SQLITE_PRIVATE void sqlite3ExprCodeFactorable(Parse*, Expr*, int); +SQLITE_PRIVATE int sqlite3ExprCodeAtInit(Parse*, Expr*, int); +SQLITE_PRIVATE int sqlite3ExprCodeTemp(Parse*, Expr*, int*); +SQLITE_PRIVATE int sqlite3ExprCodeTarget(Parse*, Expr*, int); +SQLITE_PRIVATE void sqlite3ExprCodeAndCache(Parse*, Expr*, int); +SQLITE_PRIVATE int sqlite3ExprCodeExprList(Parse*, ExprList*, int, int, u8); +#define SQLITE_ECEL_DUP 0x01 /* Deep, not shallow copies */ +#define SQLITE_ECEL_FACTOR 0x02 /* Factor out constant terms */ +#define SQLITE_ECEL_REF 0x04 /* Use ExprList.u.x.iOrderByCol */ +#define SQLITE_ECEL_OMITREF 0x08 /* Omit if ExprList.u.x.iOrderByCol */ +SQLITE_PRIVATE void sqlite3ExprIfTrue(Parse*, Expr*, int, int); +SQLITE_PRIVATE void sqlite3ExprIfFalse(Parse*, Expr*, int, int); +SQLITE_PRIVATE void sqlite3ExprIfFalseDup(Parse*, Expr*, int, int); +SQLITE_PRIVATE Table *sqlite3FindTable(sqlite3*,const char*, const char*); +#define LOCATE_VIEW 0x01 +#define LOCATE_NOERR 0x02 +SQLITE_PRIVATE Table *sqlite3LocateTable(Parse*,u32 flags,const char*, const char*); +SQLITE_PRIVATE Table *sqlite3LocateTableItem(Parse*,u32 flags,struct SrcList_item *); +SQLITE_PRIVATE Index *sqlite3FindIndex(sqlite3*,const char*, const char*); +SQLITE_PRIVATE void sqlite3UnlinkAndDeleteTable(sqlite3*,int,const char*); +SQLITE_PRIVATE void sqlite3UnlinkAndDeleteIndex(sqlite3*,int,const char*); +SQLITE_PRIVATE void sqlite3Vacuum(Parse*,Token*,Expr*); +SQLITE_PRIVATE int sqlite3RunVacuum(char**, sqlite3*, int, sqlite3_value*); +SQLITE_PRIVATE char *sqlite3NameFromToken(sqlite3*, Token*); +SQLITE_PRIVATE int sqlite3ExprCompare(Parse*,Expr*, Expr*, int); +SQLITE_PRIVATE int sqlite3ExprCompareSkip(Expr*, Expr*, int); +SQLITE_PRIVATE int sqlite3ExprListCompare(ExprList*, ExprList*, int); +SQLITE_PRIVATE int sqlite3ExprImpliesExpr(Parse*,Expr*, Expr*, int); +SQLITE_PRIVATE int sqlite3ExprImpliesNonNullRow(Expr*,int); +SQLITE_PRIVATE void sqlite3ExprAnalyzeAggregates(NameContext*, Expr*); +SQLITE_PRIVATE void sqlite3ExprAnalyzeAggList(NameContext*,ExprList*); +SQLITE_PRIVATE int sqlite3ExprCoveredByIndex(Expr*, int iCur, Index *pIdx); +SQLITE_PRIVATE int sqlite3FunctionUsesThisSrc(Expr*, SrcList*); +SQLITE_PRIVATE Vdbe *sqlite3GetVdbe(Parse*); +#ifndef SQLITE_UNTESTABLE +SQLITE_PRIVATE void sqlite3PrngSaveState(void); +SQLITE_PRIVATE void sqlite3PrngRestoreState(void); +#endif +SQLITE_PRIVATE void sqlite3RollbackAll(sqlite3*,int); +SQLITE_PRIVATE void sqlite3CodeVerifySchema(Parse*, int); +SQLITE_PRIVATE void sqlite3CodeVerifyNamedSchema(Parse*, const char *zDb); +SQLITE_PRIVATE void sqlite3BeginTransaction(Parse*, int); +SQLITE_PRIVATE void sqlite3EndTransaction(Parse*,int); +SQLITE_PRIVATE void sqlite3Savepoint(Parse*, int, Token*); +SQLITE_PRIVATE void sqlite3CloseSavepoints(sqlite3 *); +SQLITE_PRIVATE void sqlite3LeaveMutexAndCloseZombie(sqlite3*); +SQLITE_PRIVATE int sqlite3ExprIdToTrueFalse(Expr*); +SQLITE_PRIVATE int sqlite3ExprTruthValue(const Expr*); +SQLITE_PRIVATE int sqlite3ExprIsConstant(Expr*); +SQLITE_PRIVATE int sqlite3ExprIsConstantNotJoin(Expr*); +SQLITE_PRIVATE int sqlite3ExprIsConstantOrFunction(Expr*, u8); +SQLITE_PRIVATE int sqlite3ExprIsConstantOrGroupBy(Parse*, Expr*, ExprList*); +SQLITE_PRIVATE int sqlite3ExprIsTableConstant(Expr*,int); +#ifdef SQLITE_ENABLE_CURSOR_HINTS +SQLITE_PRIVATE int sqlite3ExprContainsSubquery(Expr*); +#endif +SQLITE_PRIVATE int sqlite3ExprIsInteger(Expr*, int*); +SQLITE_PRIVATE int sqlite3ExprCanBeNull(const Expr*); +SQLITE_PRIVATE int sqlite3ExprNeedsNoAffinityChange(const Expr*, char); +SQLITE_PRIVATE int sqlite3IsRowid(const char*); +SQLITE_PRIVATE void sqlite3GenerateRowDelete( + Parse*,Table*,Trigger*,int,int,int,i16,u8,u8,u8,int); +SQLITE_PRIVATE void sqlite3GenerateRowIndexDelete(Parse*, Table*, int, int, int*, int); +SQLITE_PRIVATE int sqlite3GenerateIndexKey(Parse*, Index*, int, int, int, int*,Index*,int); +SQLITE_PRIVATE void sqlite3ResolvePartIdxLabel(Parse*,int); +SQLITE_PRIVATE int sqlite3ExprReferencesUpdatedColumn(Expr*,int*,int); +SQLITE_PRIVATE void sqlite3GenerateConstraintChecks(Parse*,Table*,int*,int,int,int,int, + u8,u8,int,int*,int*,Upsert*); +#ifdef SQLITE_ENABLE_NULL_TRIM +SQLITE_PRIVATE void sqlite3SetMakeRecordP5(Vdbe*,Table*); +#else +# define sqlite3SetMakeRecordP5(A,B) +#endif +SQLITE_PRIVATE void sqlite3CompleteInsertion(Parse*,Table*,int,int,int,int*,int,int,int); +SQLITE_PRIVATE int sqlite3OpenTableAndIndices(Parse*, Table*, int, u8, int, u8*, int*, int*); +SQLITE_PRIVATE void sqlite3BeginWriteOperation(Parse*, int, int); +SQLITE_PRIVATE void sqlite3MultiWrite(Parse*); +SQLITE_PRIVATE void sqlite3MayAbort(Parse*); +SQLITE_PRIVATE void sqlite3HaltConstraint(Parse*, int, int, char*, i8, u8); +SQLITE_PRIVATE void sqlite3UniqueConstraint(Parse*, int, Index*); +SQLITE_PRIVATE void sqlite3RowidConstraint(Parse*, int, Table*); +SQLITE_PRIVATE Expr *sqlite3ExprDup(sqlite3*,Expr*,int); +SQLITE_PRIVATE ExprList *sqlite3ExprListDup(sqlite3*,ExprList*,int); +SQLITE_PRIVATE SrcList *sqlite3SrcListDup(sqlite3*,SrcList*,int); +SQLITE_PRIVATE IdList *sqlite3IdListDup(sqlite3*,IdList*); +SQLITE_PRIVATE Select *sqlite3SelectDup(sqlite3*,Select*,int); +SQLITE_PRIVATE FuncDef *sqlite3FunctionSearch(int,const char*); +SQLITE_PRIVATE void sqlite3InsertBuiltinFuncs(FuncDef*,int); +SQLITE_PRIVATE FuncDef *sqlite3FindFunction(sqlite3*,const char*,int,u8,u8); +SQLITE_PRIVATE void sqlite3RegisterBuiltinFunctions(void); +SQLITE_PRIVATE void sqlite3RegisterDateTimeFunctions(void); +SQLITE_PRIVATE void sqlite3RegisterPerConnectionBuiltinFunctions(sqlite3*); +SQLITE_PRIVATE int sqlite3SafetyCheckOk(sqlite3*); +SQLITE_PRIVATE int sqlite3SafetyCheckSickOrOk(sqlite3*); +SQLITE_PRIVATE void sqlite3ChangeCookie(Parse*, int); + +#if !defined(SQLITE_OMIT_VIEW) && !defined(SQLITE_OMIT_TRIGGER) +SQLITE_PRIVATE void sqlite3MaterializeView(Parse*, Table*, Expr*, ExprList*,Expr*,int); +#endif + +#ifndef SQLITE_OMIT_TRIGGER +SQLITE_PRIVATE void sqlite3BeginTrigger(Parse*, Token*,Token*,int,int,IdList*,SrcList*, + Expr*,int, int); +SQLITE_PRIVATE void sqlite3FinishTrigger(Parse*, TriggerStep*, Token*); +SQLITE_PRIVATE void sqlite3DropTrigger(Parse*, SrcList*, int); +SQLITE_PRIVATE void sqlite3DropTriggerPtr(Parse*, Trigger*); +SQLITE_PRIVATE Trigger *sqlite3TriggersExist(Parse *, Table*, int, ExprList*, int *pMask); +SQLITE_PRIVATE Trigger *sqlite3TriggerList(Parse *, Table *); +SQLITE_PRIVATE void sqlite3CodeRowTrigger(Parse*, Trigger *, int, ExprList*, int, Table *, + int, int, int); +SQLITE_PRIVATE void sqlite3CodeRowTriggerDirect(Parse *, Trigger *, Table *, int, int, int); + void sqliteViewTriggers(Parse*, Table*, Expr*, int, ExprList*); +SQLITE_PRIVATE void sqlite3DeleteTriggerStep(sqlite3*, TriggerStep*); +SQLITE_PRIVATE TriggerStep *sqlite3TriggerSelectStep(sqlite3*,Select*, + const char*,const char*); +SQLITE_PRIVATE TriggerStep *sqlite3TriggerInsertStep(Parse*,Token*, IdList*, + Select*,u8,Upsert*, + const char*,const char*); +SQLITE_PRIVATE TriggerStep *sqlite3TriggerUpdateStep(Parse*,Token*,ExprList*, Expr*, u8, + const char*,const char*); +SQLITE_PRIVATE TriggerStep *sqlite3TriggerDeleteStep(Parse*,Token*, Expr*, + const char*,const char*); +SQLITE_PRIVATE void sqlite3DeleteTrigger(sqlite3*, Trigger*); +SQLITE_PRIVATE void sqlite3UnlinkAndDeleteTrigger(sqlite3*,int,const char*); +SQLITE_PRIVATE u32 sqlite3TriggerColmask(Parse*,Trigger*,ExprList*,int,int,Table*,int); +# define sqlite3ParseToplevel(p) ((p)->pToplevel ? (p)->pToplevel : (p)) +# define sqlite3IsToplevel(p) ((p)->pToplevel==0) +#else +# define sqlite3TriggersExist(B,C,D,E,F) 0 +# define sqlite3DeleteTrigger(A,B) +# define sqlite3DropTriggerPtr(A,B) +# define sqlite3UnlinkAndDeleteTrigger(A,B,C) +# define sqlite3CodeRowTrigger(A,B,C,D,E,F,G,H,I) +# define sqlite3CodeRowTriggerDirect(A,B,C,D,E,F) +# define sqlite3TriggerList(X, Y) 0 +# define sqlite3ParseToplevel(p) p +# define sqlite3IsToplevel(p) 1 +# define sqlite3TriggerColmask(A,B,C,D,E,F,G) 0 +#endif + +SQLITE_PRIVATE int sqlite3JoinType(Parse*, Token*, Token*, Token*); +SQLITE_PRIVATE void sqlite3CreateForeignKey(Parse*, ExprList*, Token*, ExprList*, int); +SQLITE_PRIVATE void sqlite3DeferForeignKey(Parse*, int); +#ifndef SQLITE_OMIT_AUTHORIZATION +SQLITE_PRIVATE void sqlite3AuthRead(Parse*,Expr*,Schema*,SrcList*); +SQLITE_PRIVATE int sqlite3AuthCheck(Parse*,int, const char*, const char*, const char*); +SQLITE_PRIVATE void sqlite3AuthContextPush(Parse*, AuthContext*, const char*); +SQLITE_PRIVATE void sqlite3AuthContextPop(AuthContext*); +SQLITE_PRIVATE int sqlite3AuthReadCol(Parse*, const char *, const char *, int); +#else +# define sqlite3AuthRead(a,b,c,d) +# define sqlite3AuthCheck(a,b,c,d,e) SQLITE_OK +# define sqlite3AuthContextPush(a,b,c) +# define sqlite3AuthContextPop(a) ((void)(a)) +#endif +SQLITE_PRIVATE void sqlite3Attach(Parse*, Expr*, Expr*, Expr*); +SQLITE_PRIVATE void sqlite3Detach(Parse*, Expr*); +SQLITE_PRIVATE void sqlite3FixInit(DbFixer*, Parse*, int, const char*, const Token*); +SQLITE_PRIVATE int sqlite3FixSrcList(DbFixer*, SrcList*); +SQLITE_PRIVATE int sqlite3FixSelect(DbFixer*, Select*); +SQLITE_PRIVATE int sqlite3FixExpr(DbFixer*, Expr*); +SQLITE_PRIVATE int sqlite3FixExprList(DbFixer*, ExprList*); +SQLITE_PRIVATE int sqlite3FixTriggerStep(DbFixer*, TriggerStep*); +SQLITE_PRIVATE int sqlite3RealSameAsInt(double,sqlite3_int64); +SQLITE_PRIVATE int sqlite3AtoF(const char *z, double*, int, u8); +SQLITE_PRIVATE int sqlite3GetInt32(const char *, int*); +SQLITE_PRIVATE int sqlite3Atoi(const char*); +#ifndef SQLITE_OMIT_UTF16 +SQLITE_PRIVATE int sqlite3Utf16ByteLen(const void *pData, int nChar); +#endif +SQLITE_PRIVATE int sqlite3Utf8CharLen(const char *pData, int nByte); +SQLITE_PRIVATE u32 sqlite3Utf8Read(const u8**); +SQLITE_PRIVATE LogEst sqlite3LogEst(u64); +SQLITE_PRIVATE LogEst sqlite3LogEstAdd(LogEst,LogEst); +#ifndef SQLITE_OMIT_VIRTUALTABLE +SQLITE_PRIVATE LogEst sqlite3LogEstFromDouble(double); +#endif +#if defined(SQLITE_ENABLE_STMT_SCANSTATUS) || \ + defined(SQLITE_ENABLE_STAT4) || \ + defined(SQLITE_EXPLAIN_ESTIMATED_ROWS) +SQLITE_PRIVATE u64 sqlite3LogEstToInt(LogEst); +#endif +SQLITE_PRIVATE VList *sqlite3VListAdd(sqlite3*,VList*,const char*,int,int); +SQLITE_PRIVATE const char *sqlite3VListNumToName(VList*,int); +SQLITE_PRIVATE int sqlite3VListNameToNum(VList*,const char*,int); + +/* +** Routines to read and write variable-length integers. These used to +** be defined locally, but now we use the varint routines in the util.c +** file. +*/ +SQLITE_PRIVATE int sqlite3PutVarint(unsigned char*, u64); +SQLITE_PRIVATE u8 sqlite3GetVarint(const unsigned char *, u64 *); +SQLITE_PRIVATE u8 sqlite3GetVarint32(const unsigned char *, u32 *); +SQLITE_PRIVATE int sqlite3VarintLen(u64 v); + +/* +** The common case is for a varint to be a single byte. They following +** macros handle the common case without a procedure call, but then call +** the procedure for larger varints. +*/ +#define getVarint32(A,B) \ + (u8)((*(A)<(u8)0x80)?((B)=(u32)*(A)),1:sqlite3GetVarint32((A),(u32 *)&(B))) +#define putVarint32(A,B) \ + (u8)(((u32)(B)<(u32)0x80)?(*(A)=(unsigned char)(B)),1:\ + sqlite3PutVarint((A),(B))) +#define getVarint sqlite3GetVarint +#define putVarint sqlite3PutVarint + + +SQLITE_PRIVATE const char *sqlite3IndexAffinityStr(sqlite3*, Index*); +SQLITE_PRIVATE void sqlite3TableAffinity(Vdbe*, Table*, int); +SQLITE_PRIVATE char sqlite3CompareAffinity(Expr *pExpr, char aff2); +SQLITE_PRIVATE int sqlite3IndexAffinityOk(Expr *pExpr, char idx_affinity); +SQLITE_PRIVATE char sqlite3TableColumnAffinity(Table*,int); +SQLITE_PRIVATE char sqlite3ExprAffinity(Expr *pExpr); +SQLITE_PRIVATE int sqlite3Atoi64(const char*, i64*, int, u8); +SQLITE_PRIVATE int sqlite3DecOrHexToI64(const char*, i64*); +SQLITE_PRIVATE void sqlite3ErrorWithMsg(sqlite3*, int, const char*,...); +SQLITE_PRIVATE void sqlite3Error(sqlite3*,int); +SQLITE_PRIVATE void sqlite3SystemError(sqlite3*,int); +SQLITE_PRIVATE void *sqlite3HexToBlob(sqlite3*, const char *z, int n); +SQLITE_PRIVATE u8 sqlite3HexToInt(int h); +SQLITE_PRIVATE int sqlite3TwoPartName(Parse *, Token *, Token *, Token **); + +#if defined(SQLITE_NEED_ERR_NAME) +SQLITE_PRIVATE const char *sqlite3ErrName(int); +#endif + +#ifdef SQLITE_ENABLE_DESERIALIZE +SQLITE_PRIVATE int sqlite3MemdbInit(void); +#endif + +SQLITE_PRIVATE const char *sqlite3ErrStr(int); +SQLITE_PRIVATE int sqlite3ReadSchema(Parse *pParse); +SQLITE_PRIVATE CollSeq *sqlite3FindCollSeq(sqlite3*,u8 enc, const char*,int); +SQLITE_PRIVATE int sqlite3IsBinary(const CollSeq*); +SQLITE_PRIVATE CollSeq *sqlite3LocateCollSeq(Parse *pParse, const char*zName); +SQLITE_PRIVATE CollSeq *sqlite3ExprCollSeq(Parse *pParse, Expr *pExpr); +SQLITE_PRIVATE CollSeq *sqlite3ExprNNCollSeq(Parse *pParse, Expr *pExpr); +SQLITE_PRIVATE int sqlite3ExprCollSeqMatch(Parse*,Expr*,Expr*); +SQLITE_PRIVATE Expr *sqlite3ExprAddCollateToken(Parse *pParse, Expr*, const Token*, int); +SQLITE_PRIVATE Expr *sqlite3ExprAddCollateString(Parse*,Expr*,const char*); +SQLITE_PRIVATE Expr *sqlite3ExprSkipCollate(Expr*); +SQLITE_PRIVATE Expr *sqlite3ExprSkipCollateAndLikely(Expr*); +SQLITE_PRIVATE int sqlite3CheckCollSeq(Parse *, CollSeq *); +SQLITE_PRIVATE int sqlite3WritableSchema(sqlite3*); +SQLITE_PRIVATE int sqlite3CheckObjectName(Parse*, const char*,const char*,const char*); +SQLITE_PRIVATE void sqlite3VdbeSetChanges(sqlite3 *, int); +SQLITE_PRIVATE int sqlite3AddInt64(i64*,i64); +SQLITE_PRIVATE int sqlite3SubInt64(i64*,i64); +SQLITE_PRIVATE int sqlite3MulInt64(i64*,i64); +SQLITE_PRIVATE int sqlite3AbsInt32(int); +#ifdef SQLITE_ENABLE_8_3_NAMES +SQLITE_PRIVATE void sqlite3FileSuffix3(const char*, char*); +#else +# define sqlite3FileSuffix3(X,Y) +#endif +SQLITE_PRIVATE u8 sqlite3GetBoolean(const char *z,u8); + +SQLITE_PRIVATE const void *sqlite3ValueText(sqlite3_value*, u8); +SQLITE_PRIVATE int sqlite3ValueBytes(sqlite3_value*, u8); +SQLITE_PRIVATE void sqlite3ValueSetStr(sqlite3_value*, int, const void *,u8, + void(*)(void*)); +SQLITE_PRIVATE void sqlite3ValueSetNull(sqlite3_value*); +SQLITE_PRIVATE void sqlite3ValueFree(sqlite3_value*); +#ifndef SQLITE_UNTESTABLE +SQLITE_PRIVATE void sqlite3ResultIntReal(sqlite3_context*); +#endif +SQLITE_PRIVATE sqlite3_value *sqlite3ValueNew(sqlite3 *); +#ifndef SQLITE_OMIT_UTF16 +SQLITE_PRIVATE char *sqlite3Utf16to8(sqlite3 *, const void*, int, u8); +#endif +SQLITE_PRIVATE int sqlite3ValueFromExpr(sqlite3 *, Expr *, u8, u8, sqlite3_value **); +SQLITE_PRIVATE void sqlite3ValueApplyAffinity(sqlite3_value *, u8, u8); +#ifndef SQLITE_AMALGAMATION +SQLITE_PRIVATE const unsigned char sqlite3OpcodeProperty[]; +SQLITE_PRIVATE const char sqlite3StrBINARY[]; +SQLITE_PRIVATE const unsigned char sqlite3UpperToLower[]; +SQLITE_PRIVATE const unsigned char sqlite3CtypeMap[]; +SQLITE_PRIVATE SQLITE_WSD struct Sqlite3Config sqlite3Config; +SQLITE_PRIVATE FuncDefHash sqlite3BuiltinFunctions; +#ifndef SQLITE_OMIT_WSD +SQLITE_PRIVATE int sqlite3PendingByte; +#endif +#endif +#ifdef VDBE_PROFILE +SQLITE_PRIVATE sqlite3_uint64 sqlite3NProfileCnt; +#endif +SQLITE_PRIVATE void sqlite3RootPageMoved(sqlite3*, int, int, int); +SQLITE_PRIVATE void sqlite3Reindex(Parse*, Token*, Token*); +SQLITE_PRIVATE void sqlite3AlterFunctions(void); +SQLITE_PRIVATE void sqlite3AlterRenameTable(Parse*, SrcList*, Token*); +SQLITE_PRIVATE void sqlite3AlterRenameColumn(Parse*, SrcList*, Token*, Token*); +SQLITE_PRIVATE int sqlite3GetToken(const unsigned char *, int *); +SQLITE_PRIVATE void sqlite3NestedParse(Parse*, const char*, ...); +SQLITE_PRIVATE void sqlite3ExpirePreparedStatements(sqlite3*, int); +SQLITE_PRIVATE void sqlite3CodeRhsOfIN(Parse*, Expr*, int); +SQLITE_PRIVATE int sqlite3CodeSubselect(Parse*, Expr*); +SQLITE_PRIVATE void sqlite3SelectPrep(Parse*, Select*, NameContext*); +SQLITE_PRIVATE void sqlite3SelectWrongNumTermsError(Parse *pParse, Select *p); +SQLITE_PRIVATE int sqlite3MatchSpanName(const char*, const char*, const char*, const char*); +SQLITE_PRIVATE int sqlite3ResolveExprNames(NameContext*, Expr*); +SQLITE_PRIVATE int sqlite3ResolveExprListNames(NameContext*, ExprList*); +SQLITE_PRIVATE void sqlite3ResolveSelectNames(Parse*, Select*, NameContext*); +SQLITE_PRIVATE int sqlite3ResolveSelfReference(Parse*,Table*,int,Expr*,ExprList*); +SQLITE_PRIVATE int sqlite3ResolveOrderGroupBy(Parse*, Select*, ExprList*, const char*); +SQLITE_PRIVATE void sqlite3ColumnDefault(Vdbe *, Table *, int, int); +SQLITE_PRIVATE void sqlite3AlterFinishAddColumn(Parse *, Token *); +SQLITE_PRIVATE void sqlite3AlterBeginAddColumn(Parse *, SrcList *); +SQLITE_PRIVATE void *sqlite3RenameTokenMap(Parse*, void*, Token*); +SQLITE_PRIVATE void sqlite3RenameTokenRemap(Parse*, void *pTo, void *pFrom); +SQLITE_PRIVATE void sqlite3RenameExprUnmap(Parse*, Expr*); +SQLITE_PRIVATE void sqlite3RenameExprlistUnmap(Parse*, ExprList*); +SQLITE_PRIVATE CollSeq *sqlite3GetCollSeq(Parse*, u8, CollSeq *, const char*); +SQLITE_PRIVATE char sqlite3AffinityType(const char*, Column*); +SQLITE_PRIVATE void sqlite3Analyze(Parse*, Token*, Token*); +SQLITE_PRIVATE int sqlite3InvokeBusyHandler(BusyHandler*, sqlite3_file*); +SQLITE_PRIVATE int sqlite3FindDb(sqlite3*, Token*); +SQLITE_PRIVATE int sqlite3FindDbName(sqlite3 *, const char *); +SQLITE_PRIVATE int sqlite3AnalysisLoad(sqlite3*,int iDB); +SQLITE_PRIVATE void sqlite3DeleteIndexSamples(sqlite3*,Index*); +SQLITE_PRIVATE void sqlite3DefaultRowEst(Index*); +SQLITE_PRIVATE void sqlite3RegisterLikeFunctions(sqlite3*, int); +SQLITE_PRIVATE int sqlite3IsLikeFunction(sqlite3*,Expr*,int*,char*); +SQLITE_PRIVATE void sqlite3SchemaClear(void *); +SQLITE_PRIVATE Schema *sqlite3SchemaGet(sqlite3 *, Btree *); +SQLITE_PRIVATE int sqlite3SchemaToIndex(sqlite3 *db, Schema *); +SQLITE_PRIVATE KeyInfo *sqlite3KeyInfoAlloc(sqlite3*,int,int); +SQLITE_PRIVATE void sqlite3KeyInfoUnref(KeyInfo*); +SQLITE_PRIVATE KeyInfo *sqlite3KeyInfoRef(KeyInfo*); +SQLITE_PRIVATE KeyInfo *sqlite3KeyInfoOfIndex(Parse*, Index*); +SQLITE_PRIVATE KeyInfo *sqlite3KeyInfoFromExprList(Parse*, ExprList*, int, int); +SQLITE_PRIVATE int sqlite3HasExplicitNulls(Parse*, ExprList*); + +#ifdef SQLITE_DEBUG +SQLITE_PRIVATE int sqlite3KeyInfoIsWriteable(KeyInfo*); +#endif +SQLITE_PRIVATE int sqlite3CreateFunc(sqlite3 *, const char *, int, int, void *, + void (*)(sqlite3_context*,int,sqlite3_value **), + void (*)(sqlite3_context*,int,sqlite3_value **), + void (*)(sqlite3_context*), + void (*)(sqlite3_context*), + void (*)(sqlite3_context*,int,sqlite3_value **), + FuncDestructor *pDestructor +); +SQLITE_PRIVATE void sqlite3NoopDestructor(void*); +SQLITE_PRIVATE void sqlite3OomFault(sqlite3*); +SQLITE_PRIVATE void sqlite3OomClear(sqlite3*); +SQLITE_PRIVATE int sqlite3ApiExit(sqlite3 *db, int); +SQLITE_PRIVATE int sqlite3OpenTempDatabase(Parse *); + +SQLITE_PRIVATE void sqlite3StrAccumInit(StrAccum*, sqlite3*, char*, int, int); +SQLITE_PRIVATE char *sqlite3StrAccumFinish(StrAccum*); +SQLITE_PRIVATE void sqlite3SelectDestInit(SelectDest*,int,int); +SQLITE_PRIVATE Expr *sqlite3CreateColumnExpr(sqlite3 *, SrcList *, int, int); + +SQLITE_PRIVATE void sqlite3BackupRestart(sqlite3_backup *); +SQLITE_PRIVATE void sqlite3BackupUpdate(sqlite3_backup *, Pgno, const u8 *); + +#ifndef SQLITE_OMIT_SUBQUERY +SQLITE_PRIVATE int sqlite3ExprCheckIN(Parse*, Expr*); +#else +# define sqlite3ExprCheckIN(x,y) SQLITE_OK +#endif + +#ifdef SQLITE_ENABLE_STAT4 +SQLITE_PRIVATE int sqlite3Stat4ProbeSetValue( + Parse*,Index*,UnpackedRecord**,Expr*,int,int,int*); +SQLITE_PRIVATE int sqlite3Stat4ValueFromExpr(Parse*, Expr*, u8, sqlite3_value**); +SQLITE_PRIVATE void sqlite3Stat4ProbeFree(UnpackedRecord*); +SQLITE_PRIVATE int sqlite3Stat4Column(sqlite3*, const void*, int, int, sqlite3_value**); +SQLITE_PRIVATE char sqlite3IndexColumnAffinity(sqlite3*, Index*, int); +#endif + +/* +** The interface to the LEMON-generated parser +*/ +#ifndef SQLITE_AMALGAMATION +SQLITE_PRIVATE void *sqlite3ParserAlloc(void*(*)(u64), Parse*); +SQLITE_PRIVATE void sqlite3ParserFree(void*, void(*)(void*)); +#endif +SQLITE_PRIVATE void sqlite3Parser(void*, int, Token); +SQLITE_PRIVATE int sqlite3ParserFallback(int); +#ifdef YYTRACKMAXSTACKDEPTH +SQLITE_PRIVATE int sqlite3ParserStackPeak(void*); +#endif + +SQLITE_PRIVATE void sqlite3AutoLoadExtensions(sqlite3*); +#ifndef SQLITE_OMIT_LOAD_EXTENSION +SQLITE_PRIVATE void sqlite3CloseExtensions(sqlite3*); +#else +# define sqlite3CloseExtensions(X) +#endif + +#ifndef SQLITE_OMIT_SHARED_CACHE +SQLITE_PRIVATE void sqlite3TableLock(Parse *, int, int, u8, const char *); +#else + #define sqlite3TableLock(v,w,x,y,z) +#endif + +#ifdef SQLITE_TEST +SQLITE_PRIVATE int sqlite3Utf8To8(unsigned char*); +#endif + +#ifdef SQLITE_OMIT_VIRTUALTABLE +# define sqlite3VtabClear(Y) +# define sqlite3VtabSync(X,Y) SQLITE_OK +# define sqlite3VtabRollback(X) +# define sqlite3VtabCommit(X) +# define sqlite3VtabInSync(db) 0 +# define sqlite3VtabLock(X) +# define sqlite3VtabUnlock(X) +# define sqlite3VtabModuleUnref(D,X) +# define sqlite3VtabUnlockList(X) +# define sqlite3VtabSavepoint(X, Y, Z) SQLITE_OK +# define sqlite3GetVTable(X,Y) ((VTable*)0) +#else +SQLITE_PRIVATE void sqlite3VtabClear(sqlite3 *db, Table*); +SQLITE_PRIVATE void sqlite3VtabDisconnect(sqlite3 *db, Table *p); +SQLITE_PRIVATE int sqlite3VtabSync(sqlite3 *db, Vdbe*); +SQLITE_PRIVATE int sqlite3VtabRollback(sqlite3 *db); +SQLITE_PRIVATE int sqlite3VtabCommit(sqlite3 *db); +SQLITE_PRIVATE void sqlite3VtabLock(VTable *); +SQLITE_PRIVATE void sqlite3VtabUnlock(VTable *); +SQLITE_PRIVATE void sqlite3VtabModuleUnref(sqlite3*,Module*); +SQLITE_PRIVATE void sqlite3VtabUnlockList(sqlite3*); +SQLITE_PRIVATE int sqlite3VtabSavepoint(sqlite3 *, int, int); +SQLITE_PRIVATE void sqlite3VtabImportErrmsg(Vdbe*, sqlite3_vtab*); +SQLITE_PRIVATE VTable *sqlite3GetVTable(sqlite3*, Table*); +SQLITE_PRIVATE Module *sqlite3VtabCreateModule( + sqlite3*, + const char*, + const sqlite3_module*, + void*, + void(*)(void*) + ); +# define sqlite3VtabInSync(db) ((db)->nVTrans>0 && (db)->aVTrans==0) +#endif +SQLITE_PRIVATE int sqlite3VtabEponymousTableInit(Parse*,Module*); +SQLITE_PRIVATE void sqlite3VtabEponymousTableClear(sqlite3*,Module*); +SQLITE_PRIVATE void sqlite3VtabMakeWritable(Parse*,Table*); +SQLITE_PRIVATE void sqlite3VtabBeginParse(Parse*, Token*, Token*, Token*, int); +SQLITE_PRIVATE void sqlite3VtabFinishParse(Parse*, Token*); +SQLITE_PRIVATE void sqlite3VtabArgInit(Parse*); +SQLITE_PRIVATE void sqlite3VtabArgExtend(Parse*, Token*); +SQLITE_PRIVATE int sqlite3VtabCallCreate(sqlite3*, int, const char *, char **); +SQLITE_PRIVATE int sqlite3VtabCallConnect(Parse*, Table*); +SQLITE_PRIVATE int sqlite3VtabCallDestroy(sqlite3*, int, const char *); +SQLITE_PRIVATE int sqlite3VtabBegin(sqlite3 *, VTable *); +SQLITE_PRIVATE FuncDef *sqlite3VtabOverloadFunction(sqlite3 *,FuncDef*, int nArg, Expr*); +SQLITE_PRIVATE sqlite3_int64 sqlite3StmtCurrentTime(sqlite3_context*); +SQLITE_PRIVATE int sqlite3VdbeParameterIndex(Vdbe*, const char*, int); +SQLITE_PRIVATE int sqlite3TransferBindings(sqlite3_stmt *, sqlite3_stmt *); +SQLITE_PRIVATE void sqlite3ParserReset(Parse*); +#ifdef SQLITE_ENABLE_NORMALIZE +SQLITE_PRIVATE char *sqlite3Normalize(Vdbe*, const char*); +#endif +SQLITE_PRIVATE int sqlite3Reprepare(Vdbe*); +SQLITE_PRIVATE void sqlite3ExprListCheckLength(Parse*, ExprList*, const char*); +SQLITE_PRIVATE CollSeq *sqlite3BinaryCompareCollSeq(Parse *, Expr *, Expr *); +SQLITE_PRIVATE int sqlite3TempInMemory(const sqlite3*); +SQLITE_PRIVATE const char *sqlite3JournalModename(int); +#ifndef SQLITE_OMIT_WAL +SQLITE_PRIVATE int sqlite3Checkpoint(sqlite3*, int, int, int*, int*); +SQLITE_PRIVATE int sqlite3WalDefaultHook(void*,sqlite3*,const char*,int); +#endif +#ifndef SQLITE_OMIT_CTE +SQLITE_PRIVATE With *sqlite3WithAdd(Parse*,With*,Token*,ExprList*,Select*); +SQLITE_PRIVATE void sqlite3WithDelete(sqlite3*,With*); +SQLITE_PRIVATE void sqlite3WithPush(Parse*, With*, u8); +#else +#define sqlite3WithPush(x,y,z) +#define sqlite3WithDelete(x,y) +#endif +#ifndef SQLITE_OMIT_UPSERT +SQLITE_PRIVATE Upsert *sqlite3UpsertNew(sqlite3*,ExprList*,Expr*,ExprList*,Expr*); +SQLITE_PRIVATE void sqlite3UpsertDelete(sqlite3*,Upsert*); +SQLITE_PRIVATE Upsert *sqlite3UpsertDup(sqlite3*,Upsert*); +SQLITE_PRIVATE int sqlite3UpsertAnalyzeTarget(Parse*,SrcList*,Upsert*); +SQLITE_PRIVATE void sqlite3UpsertDoUpdate(Parse*,Upsert*,Table*,Index*,int); +#else +#define sqlite3UpsertNew(v,w,x,y,z) ((Upsert*)0) +#define sqlite3UpsertDelete(x,y) +#define sqlite3UpsertDup(x,y) ((Upsert*)0) +#endif + + +/* Declarations for functions in fkey.c. All of these are replaced by +** no-op macros if OMIT_FOREIGN_KEY is defined. In this case no foreign +** key functionality is available. If OMIT_TRIGGER is defined but +** OMIT_FOREIGN_KEY is not, only some of the functions are no-oped. In +** this case foreign keys are parsed, but no other functionality is +** provided (enforcement of FK constraints requires the triggers sub-system). +*/ +#if !defined(SQLITE_OMIT_FOREIGN_KEY) && !defined(SQLITE_OMIT_TRIGGER) +SQLITE_PRIVATE void sqlite3FkCheck(Parse*, Table*, int, int, int*, int); +SQLITE_PRIVATE void sqlite3FkDropTable(Parse*, SrcList *, Table*); +SQLITE_PRIVATE void sqlite3FkActions(Parse*, Table*, ExprList*, int, int*, int); +SQLITE_PRIVATE int sqlite3FkRequired(Parse*, Table*, int*, int); +SQLITE_PRIVATE u32 sqlite3FkOldmask(Parse*, Table*); +SQLITE_PRIVATE FKey *sqlite3FkReferences(Table *); +#else + #define sqlite3FkActions(a,b,c,d,e,f) + #define sqlite3FkCheck(a,b,c,d,e,f) + #define sqlite3FkDropTable(a,b,c) + #define sqlite3FkOldmask(a,b) 0 + #define sqlite3FkRequired(a,b,c,d) 0 + #define sqlite3FkReferences(a) 0 +#endif +#ifndef SQLITE_OMIT_FOREIGN_KEY +SQLITE_PRIVATE void sqlite3FkDelete(sqlite3 *, Table*); +SQLITE_PRIVATE int sqlite3FkLocateIndex(Parse*,Table*,FKey*,Index**,int**); +#else + #define sqlite3FkDelete(a,b) + #define sqlite3FkLocateIndex(a,b,c,d,e) +#endif + + +/* +** Available fault injectors. Should be numbered beginning with 0. +*/ +#define SQLITE_FAULTINJECTOR_MALLOC 0 +#define SQLITE_FAULTINJECTOR_COUNT 1 + +/* +** The interface to the code in fault.c used for identifying "benign" +** malloc failures. This is only present if SQLITE_UNTESTABLE +** is not defined. +*/ +#ifndef SQLITE_UNTESTABLE +SQLITE_PRIVATE void sqlite3BeginBenignMalloc(void); +SQLITE_PRIVATE void sqlite3EndBenignMalloc(void); +#else + #define sqlite3BeginBenignMalloc() + #define sqlite3EndBenignMalloc() +#endif + +/* +** Allowed return values from sqlite3FindInIndex() +*/ +#define IN_INDEX_ROWID 1 /* Search the rowid of the table */ +#define IN_INDEX_EPH 2 /* Search an ephemeral b-tree */ +#define IN_INDEX_INDEX_ASC 3 /* Existing index ASCENDING */ +#define IN_INDEX_INDEX_DESC 4 /* Existing index DESCENDING */ +#define IN_INDEX_NOOP 5 /* No table available. Use comparisons */ +/* +** Allowed flags for the 3rd parameter to sqlite3FindInIndex(). +*/ +#define IN_INDEX_NOOP_OK 0x0001 /* OK to return IN_INDEX_NOOP */ +#define IN_INDEX_MEMBERSHIP 0x0002 /* IN operator used for membership test */ +#define IN_INDEX_LOOP 0x0004 /* IN operator used as a loop */ +SQLITE_PRIVATE int sqlite3FindInIndex(Parse *, Expr *, u32, int*, int*, int*); + +SQLITE_PRIVATE int sqlite3JournalOpen(sqlite3_vfs *, const char *, sqlite3_file *, int, int); +SQLITE_PRIVATE int sqlite3JournalSize(sqlite3_vfs *); +#if defined(SQLITE_ENABLE_ATOMIC_WRITE) \ + || defined(SQLITE_ENABLE_BATCH_ATOMIC_WRITE) +SQLITE_PRIVATE int sqlite3JournalCreate(sqlite3_file *); +#endif + +SQLITE_PRIVATE int sqlite3JournalIsInMemory(sqlite3_file *p); +SQLITE_PRIVATE void sqlite3MemJournalOpen(sqlite3_file *); + +SQLITE_PRIVATE void sqlite3ExprSetHeightAndFlags(Parse *pParse, Expr *p); +#if SQLITE_MAX_EXPR_DEPTH>0 +SQLITE_PRIVATE int sqlite3SelectExprHeight(Select *); +SQLITE_PRIVATE int sqlite3ExprCheckHeight(Parse*, int); +#else + #define sqlite3SelectExprHeight(x) 0 + #define sqlite3ExprCheckHeight(x,y) +#endif + +SQLITE_PRIVATE u32 sqlite3Get4byte(const u8*); +SQLITE_PRIVATE void sqlite3Put4byte(u8*, u32); + +#ifdef SQLITE_ENABLE_UNLOCK_NOTIFY +SQLITE_PRIVATE void sqlite3ConnectionBlocked(sqlite3 *, sqlite3 *); +SQLITE_PRIVATE void sqlite3ConnectionUnlocked(sqlite3 *db); +SQLITE_PRIVATE void sqlite3ConnectionClosed(sqlite3 *db); +#else + #define sqlite3ConnectionBlocked(x,y) + #define sqlite3ConnectionUnlocked(x) + #define sqlite3ConnectionClosed(x) +#endif + +#ifdef SQLITE_DEBUG +SQLITE_PRIVATE void sqlite3ParserTrace(FILE*, char *); +#endif +#if defined(YYCOVERAGE) +SQLITE_PRIVATE int sqlite3ParserCoverage(FILE*); +#endif + +/* +** If the SQLITE_ENABLE IOTRACE exists then the global variable +** sqlite3IoTrace is a pointer to a printf-like routine used to +** print I/O tracing messages. +*/ +#ifdef SQLITE_ENABLE_IOTRACE +# define IOTRACE(A) if( sqlite3IoTrace ){ sqlite3IoTrace A; } +SQLITE_PRIVATE void sqlite3VdbeIOTraceSql(Vdbe*); +SQLITE_API SQLITE_EXTERN void (SQLITE_CDECL *sqlite3IoTrace)(const char*,...); +#else +# define IOTRACE(A) +# define sqlite3VdbeIOTraceSql(X) +#endif + +/* +** These routines are available for the mem2.c debugging memory allocator +** only. They are used to verify that different "types" of memory +** allocations are properly tracked by the system. +** +** sqlite3MemdebugSetType() sets the "type" of an allocation to one of +** the MEMTYPE_* macros defined below. The type must be a bitmask with +** a single bit set. +** +** sqlite3MemdebugHasType() returns true if any of the bits in its second +** argument match the type set by the previous sqlite3MemdebugSetType(). +** sqlite3MemdebugHasType() is intended for use inside assert() statements. +** +** sqlite3MemdebugNoType() returns true if none of the bits in its second +** argument match the type set by the previous sqlite3MemdebugSetType(). +** +** Perhaps the most important point is the difference between MEMTYPE_HEAP +** and MEMTYPE_LOOKASIDE. If an allocation is MEMTYPE_LOOKASIDE, that means +** it might have been allocated by lookaside, except the allocation was +** too large or lookaside was already full. It is important to verify +** that allocations that might have been satisfied by lookaside are not +** passed back to non-lookaside free() routines. Asserts such as the +** example above are placed on the non-lookaside free() routines to verify +** this constraint. +** +** All of this is no-op for a production build. It only comes into +** play when the SQLITE_MEMDEBUG compile-time option is used. +*/ +#ifdef SQLITE_MEMDEBUG +SQLITE_PRIVATE void sqlite3MemdebugSetType(void*,u8); +SQLITE_PRIVATE int sqlite3MemdebugHasType(void*,u8); +SQLITE_PRIVATE int sqlite3MemdebugNoType(void*,u8); +#else +# define sqlite3MemdebugSetType(X,Y) /* no-op */ +# define sqlite3MemdebugHasType(X,Y) 1 +# define sqlite3MemdebugNoType(X,Y) 1 +#endif +#define MEMTYPE_HEAP 0x01 /* General heap allocations */ +#define MEMTYPE_LOOKASIDE 0x02 /* Heap that might have been lookaside */ +#define MEMTYPE_PCACHE 0x04 /* Page cache allocations */ + +/* +** Threading interface +*/ +#if SQLITE_MAX_WORKER_THREADS>0 +SQLITE_PRIVATE int sqlite3ThreadCreate(SQLiteThread**,void*(*)(void*),void*); +SQLITE_PRIVATE int sqlite3ThreadJoin(SQLiteThread*, void**); +#endif + +#if defined(SQLITE_ENABLE_DBPAGE_VTAB) || defined(SQLITE_TEST) +SQLITE_PRIVATE int sqlite3DbpageRegister(sqlite3*); +#endif +#if defined(SQLITE_ENABLE_DBSTAT_VTAB) || defined(SQLITE_TEST) +SQLITE_PRIVATE int sqlite3DbstatRegister(sqlite3*); +#endif + +SQLITE_PRIVATE int sqlite3ExprVectorSize(Expr *pExpr); +SQLITE_PRIVATE int sqlite3ExprIsVector(Expr *pExpr); +SQLITE_PRIVATE Expr *sqlite3VectorFieldSubexpr(Expr*, int); +SQLITE_PRIVATE Expr *sqlite3ExprForVectorField(Parse*,Expr*,int); +SQLITE_PRIVATE void sqlite3VectorErrorMsg(Parse*, Expr*); + +#ifndef SQLITE_OMIT_COMPILEOPTION_DIAGS +SQLITE_PRIVATE const char **sqlite3CompileOptions(int *pnOpt); +#endif + +#endif /* SQLITEINT_H */ + +/************** End of sqliteInt.h *******************************************/ +/************** Begin file global.c ******************************************/ +/* +** 2008 June 13 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** +** This file contains definitions of global variables and constants. +*/ +/* #include "sqliteInt.h" */ + +/* An array to map all upper-case characters into their corresponding +** lower-case character. +** +** SQLite only considers US-ASCII (or EBCDIC) characters. We do not +** handle case conversions for the UTF character set since the tables +** involved are nearly as big or bigger than SQLite itself. +*/ +SQLITE_PRIVATE const unsigned char sqlite3UpperToLower[] = { +#ifdef SQLITE_ASCII + 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, + 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, + 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 52, 53, + 54, 55, 56, 57, 58, 59, 60, 61, 62, 63, 64, 97, 98, 99,100,101,102,103, + 104,105,106,107,108,109,110,111,112,113,114,115,116,117,118,119,120,121, + 122, 91, 92, 93, 94, 95, 96, 97, 98, 99,100,101,102,103,104,105,106,107, + 108,109,110,111,112,113,114,115,116,117,118,119,120,121,122,123,124,125, + 126,127,128,129,130,131,132,133,134,135,136,137,138,139,140,141,142,143, + 144,145,146,147,148,149,150,151,152,153,154,155,156,157,158,159,160,161, + 162,163,164,165,166,167,168,169,170,171,172,173,174,175,176,177,178,179, + 180,181,182,183,184,185,186,187,188,189,190,191,192,193,194,195,196,197, + 198,199,200,201,202,203,204,205,206,207,208,209,210,211,212,213,214,215, + 216,217,218,219,220,221,222,223,224,225,226,227,228,229,230,231,232,233, + 234,235,236,237,238,239,240,241,242,243,244,245,246,247,248,249,250,251, + 252,253,254,255 +#endif +#ifdef SQLITE_EBCDIC + 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, /* 0x */ + 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, /* 1x */ + 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, /* 2x */ + 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 62, 63, /* 3x */ + 64, 65, 66, 67, 68, 69, 70, 71, 72, 73, 74, 75, 76, 77, 78, 79, /* 4x */ + 80, 81, 82, 83, 84, 85, 86, 87, 88, 89, 90, 91, 92, 93, 94, 95, /* 5x */ + 96, 97, 98, 99,100,101,102,103,104,105,106,107,108,109,110,111, /* 6x */ + 112,113,114,115,116,117,118,119,120,121,122,123,124,125,126,127, /* 7x */ + 128,129,130,131,132,133,134,135,136,137,138,139,140,141,142,143, /* 8x */ + 144,145,146,147,148,149,150,151,152,153,154,155,156,157,158,159, /* 9x */ + 160,161,162,163,164,165,166,167,168,169,170,171,140,141,142,175, /* Ax */ + 176,177,178,179,180,181,182,183,184,185,186,187,188,189,190,191, /* Bx */ + 192,129,130,131,132,133,134,135,136,137,202,203,204,205,206,207, /* Cx */ + 208,145,146,147,148,149,150,151,152,153,218,219,220,221,222,223, /* Dx */ + 224,225,162,163,164,165,166,167,168,169,234,235,236,237,238,239, /* Ex */ + 240,241,242,243,244,245,246,247,248,249,250,251,252,253,254,255, /* Fx */ +#endif +}; + +/* +** The following 256 byte lookup table is used to support SQLites built-in +** equivalents to the following standard library functions: +** +** isspace() 0x01 +** isalpha() 0x02 +** isdigit() 0x04 +** isalnum() 0x06 +** isxdigit() 0x08 +** toupper() 0x20 +** SQLite identifier character 0x40 +** Quote character 0x80 +** +** Bit 0x20 is set if the mapped character requires translation to upper +** case. i.e. if the character is a lower-case ASCII character. +** If x is a lower-case ASCII character, then its upper-case equivalent +** is (x - 0x20). Therefore toupper() can be implemented as: +** +** (x & ~(map[x]&0x20)) +** +** The equivalent of tolower() is implemented using the sqlite3UpperToLower[] +** array. tolower() is used more often than toupper() by SQLite. +** +** Bit 0x40 is set if the character is non-alphanumeric and can be used in an +** SQLite identifier. Identifiers are alphanumerics, "_", "$", and any +** non-ASCII UTF character. Hence the test for whether or not a character is +** part of an identifier is 0x46. +*/ +#ifdef SQLITE_ASCII +SQLITE_PRIVATE const unsigned char sqlite3CtypeMap[256] = { + 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, /* 00..07 ........ */ + 0x00, 0x01, 0x01, 0x01, 0x01, 0x01, 0x00, 0x00, /* 08..0f ........ */ + 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, /* 10..17 ........ */ + 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, /* 18..1f ........ */ + 0x01, 0x00, 0x80, 0x00, 0x40, 0x00, 0x00, 0x80, /* 20..27 !"#$%&' */ + 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, /* 28..2f ()*+,-./ */ + 0x0c, 0x0c, 0x0c, 0x0c, 0x0c, 0x0c, 0x0c, 0x0c, /* 30..37 01234567 */ + 0x0c, 0x0c, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, /* 38..3f 89:;<=>? */ + + 0x00, 0x0a, 0x0a, 0x0a, 0x0a, 0x0a, 0x0a, 0x02, /* 40..47 @ABCDEFG */ + 0x02, 0x02, 0x02, 0x02, 0x02, 0x02, 0x02, 0x02, /* 48..4f HIJKLMNO */ + 0x02, 0x02, 0x02, 0x02, 0x02, 0x02, 0x02, 0x02, /* 50..57 PQRSTUVW */ + 0x02, 0x02, 0x02, 0x80, 0x00, 0x00, 0x00, 0x40, /* 58..5f XYZ[\]^_ */ + 0x80, 0x2a, 0x2a, 0x2a, 0x2a, 0x2a, 0x2a, 0x22, /* 60..67 `abcdefg */ + 0x22, 0x22, 0x22, 0x22, 0x22, 0x22, 0x22, 0x22, /* 68..6f hijklmno */ + 0x22, 0x22, 0x22, 0x22, 0x22, 0x22, 0x22, 0x22, /* 70..77 pqrstuvw */ + 0x22, 0x22, 0x22, 0x00, 0x00, 0x00, 0x00, 0x00, /* 78..7f xyz{|}~. */ + + 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, /* 80..87 ........ */ + 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, /* 88..8f ........ */ + 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, /* 90..97 ........ */ + 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, /* 98..9f ........ */ + 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, /* a0..a7 ........ */ + 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, /* a8..af ........ */ + 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, /* b0..b7 ........ */ + 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, /* b8..bf ........ */ + + 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, /* c0..c7 ........ */ + 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, /* c8..cf ........ */ + 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, /* d0..d7 ........ */ + 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, /* d8..df ........ */ + 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, /* e0..e7 ........ */ + 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, /* e8..ef ........ */ + 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, /* f0..f7 ........ */ + 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40 /* f8..ff ........ */ +}; +#endif + +/* EVIDENCE-OF: R-02982-34736 In order to maintain full backwards +** compatibility for legacy applications, the URI filename capability is +** disabled by default. +** +** EVIDENCE-OF: R-38799-08373 URI filenames can be enabled or disabled +** using the SQLITE_USE_URI=1 or SQLITE_USE_URI=0 compile-time options. +** +** EVIDENCE-OF: R-43642-56306 By default, URI handling is globally +** disabled. The default value may be changed by compiling with the +** SQLITE_USE_URI symbol defined. +** +** URI filenames are enabled by default if SQLITE_HAS_CODEC is +** enabled. +*/ +#ifndef SQLITE_USE_URI +# ifdef SQLITE_HAS_CODEC +# define SQLITE_USE_URI 1 +# else +# define SQLITE_USE_URI 0 +# endif +#endif + +/* EVIDENCE-OF: R-38720-18127 The default setting is determined by the +** SQLITE_ALLOW_COVERING_INDEX_SCAN compile-time option, or is "on" if +** that compile-time option is omitted. +*/ +#if !defined(SQLITE_ALLOW_COVERING_INDEX_SCAN) +# define SQLITE_ALLOW_COVERING_INDEX_SCAN 1 +#else +# if !SQLITE_ALLOW_COVERING_INDEX_SCAN +# error "Compile-time disabling of covering index scan using the\ + -DSQLITE_ALLOW_COVERING_INDEX_SCAN=0 option is deprecated.\ + Contact SQLite developers if this is a problem for you, and\ + delete this #error macro to continue with your build." +# endif +#endif + +/* The minimum PMA size is set to this value multiplied by the database +** page size in bytes. +*/ +#ifndef SQLITE_SORTER_PMASZ +# define SQLITE_SORTER_PMASZ 250 +#endif + +/* Statement journals spill to disk when their size exceeds the following +** threshold (in bytes). 0 means that statement journals are created and +** written to disk immediately (the default behavior for SQLite versions +** before 3.12.0). -1 means always keep the entire statement journal in +** memory. (The statement journal is also always held entirely in memory +** if journal_mode=MEMORY or if temp_store=MEMORY, regardless of this +** setting.) +*/ +#ifndef SQLITE_STMTJRNL_SPILL +# define SQLITE_STMTJRNL_SPILL (64*1024) +#endif + +/* +** The default lookaside-configuration, the format "SZ,N". SZ is the +** number of bytes in each lookaside slot (should be a multiple of 8) +** and N is the number of slots. The lookaside-configuration can be +** changed as start-time using sqlite3_config(SQLITE_CONFIG_LOOKASIDE) +** or at run-time for an individual database connection using +** sqlite3_db_config(db, SQLITE_DBCONFIG_LOOKASIDE); +*/ +#ifndef SQLITE_DEFAULT_LOOKASIDE +# define SQLITE_DEFAULT_LOOKASIDE 1200,100 +#endif + + +/* The default maximum size of an in-memory database created using +** sqlite3_deserialize() +*/ +#ifndef SQLITE_MEMDB_DEFAULT_MAXSIZE +# define SQLITE_MEMDB_DEFAULT_MAXSIZE 1073741824 +#endif + +/* +** The following singleton contains the global configuration for +** the SQLite library. +*/ +SQLITE_PRIVATE SQLITE_WSD struct Sqlite3Config sqlite3Config = { + SQLITE_DEFAULT_MEMSTATUS, /* bMemstat */ + 1, /* bCoreMutex */ + SQLITE_THREADSAFE==1, /* bFullMutex */ + SQLITE_USE_URI, /* bOpenUri */ + SQLITE_ALLOW_COVERING_INDEX_SCAN, /* bUseCis */ + 0, /* bSmallMalloc */ + 1, /* bExtraSchemaChecks */ + 0x7ffffffe, /* mxStrlen */ + 0, /* neverCorrupt */ + SQLITE_DEFAULT_LOOKASIDE, /* szLookaside, nLookaside */ + SQLITE_STMTJRNL_SPILL, /* nStmtSpill */ + {0,0,0,0,0,0,0,0}, /* m */ + {0,0,0,0,0,0,0,0,0}, /* mutex */ + {0,0,0,0,0,0,0,0,0,0,0,0,0},/* pcache2 */ + (void*)0, /* pHeap */ + 0, /* nHeap */ + 0, 0, /* mnHeap, mxHeap */ + SQLITE_DEFAULT_MMAP_SIZE, /* szMmap */ + SQLITE_MAX_MMAP_SIZE, /* mxMmap */ + (void*)0, /* pPage */ + 0, /* szPage */ + SQLITE_DEFAULT_PCACHE_INITSZ, /* nPage */ + 0, /* mxParserStack */ + 0, /* sharedCacheEnabled */ + SQLITE_SORTER_PMASZ, /* szPma */ + /* All the rest should always be initialized to zero */ + 0, /* isInit */ + 0, /* inProgress */ + 0, /* isMutexInit */ + 0, /* isMallocInit */ + 0, /* isPCacheInit */ + 0, /* nRefInitMutex */ + 0, /* pInitMutex */ + 0, /* xLog */ + 0, /* pLogArg */ +#ifdef SQLITE_ENABLE_SQLLOG + 0, /* xSqllog */ + 0, /* pSqllogArg */ +#endif +#ifdef SQLITE_VDBE_COVERAGE + 0, /* xVdbeBranch */ + 0, /* pVbeBranchArg */ +#endif +#ifdef SQLITE_ENABLE_DESERIALIZE + SQLITE_MEMDB_DEFAULT_MAXSIZE, /* mxMemdbSize */ +#endif +#ifndef SQLITE_UNTESTABLE + 0, /* xTestCallback */ +#endif + 0, /* bLocaltimeFault */ + 0, /* bInternalFunctions */ + 0x7ffffffe, /* iOnceResetThreshold */ + SQLITE_DEFAULT_SORTERREF_SIZE, /* szSorterRef */ + 0, /* iPrngSeed */ +}; + +/* +** Hash table for global functions - functions common to all +** database connections. After initialization, this table is +** read-only. +*/ +SQLITE_PRIVATE FuncDefHash sqlite3BuiltinFunctions; + +#ifdef VDBE_PROFILE +/* +** The following performance counter can be used in place of +** sqlite3Hwtime() for profiling. This is a no-op on standard builds. +*/ +SQLITE_PRIVATE sqlite3_uint64 sqlite3NProfileCnt = 0; +#endif + +/* +** The value of the "pending" byte must be 0x40000000 (1 byte past the +** 1-gibabyte boundary) in a compatible database. SQLite never uses +** the database page that contains the pending byte. It never attempts +** to read or write that page. The pending byte page is set aside +** for use by the VFS layers as space for managing file locks. +** +** During testing, it is often desirable to move the pending byte to +** a different position in the file. This allows code that has to +** deal with the pending byte to run on files that are much smaller +** than 1 GiB. The sqlite3_test_control() interface can be used to +** move the pending byte. +** +** IMPORTANT: Changing the pending byte to any value other than +** 0x40000000 results in an incompatible database file format! +** Changing the pending byte during operation will result in undefined +** and incorrect behavior. +*/ +#ifndef SQLITE_OMIT_WSD +SQLITE_PRIVATE int sqlite3PendingByte = 0x40000000; +#endif + +/* #include "opcodes.h" */ +/* +** Properties of opcodes. The OPFLG_INITIALIZER macro is +** created by mkopcodeh.awk during compilation. Data is obtained +** from the comments following the "case OP_xxxx:" statements in +** the vdbe.c file. +*/ +SQLITE_PRIVATE const unsigned char sqlite3OpcodeProperty[] = OPFLG_INITIALIZER; + +/* +** Name of the default collating sequence +*/ +SQLITE_PRIVATE const char sqlite3StrBINARY[] = "BINARY"; + +/************** End of global.c **********************************************/ +/************** Begin file status.c ******************************************/ +/* +** 2008 June 18 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** +** This module implements the sqlite3_status() interface and related +** functionality. +*/ +/* #include "sqliteInt.h" */ +/************** Include vdbeInt.h in the middle of status.c ******************/ +/************** Begin file vdbeInt.h *****************************************/ +/* +** 2003 September 6 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** This is the header file for information that is private to the +** VDBE. This information used to all be at the top of the single +** source code file "vdbe.c". When that file became too big (over +** 6000 lines long) it was split up into several smaller files and +** this header information was factored out. +*/ +#ifndef SQLITE_VDBEINT_H +#define SQLITE_VDBEINT_H + +/* +** The maximum number of times that a statement will try to reparse +** itself before giving up and returning SQLITE_SCHEMA. +*/ +#ifndef SQLITE_MAX_SCHEMA_RETRY +# define SQLITE_MAX_SCHEMA_RETRY 50 +#endif + +/* +** VDBE_DISPLAY_P4 is true or false depending on whether or not the +** "explain" P4 display logic is enabled. +*/ +#if !defined(SQLITE_OMIT_EXPLAIN) || !defined(NDEBUG) \ + || defined(VDBE_PROFILE) || defined(SQLITE_DEBUG) +# define VDBE_DISPLAY_P4 1 +#else +# define VDBE_DISPLAY_P4 0 +#endif + +/* +** SQL is translated into a sequence of instructions to be +** executed by a virtual machine. Each instruction is an instance +** of the following structure. +*/ +typedef struct VdbeOp Op; + +/* +** Boolean values +*/ +typedef unsigned Bool; + +/* Opaque type used by code in vdbesort.c */ +typedef struct VdbeSorter VdbeSorter; + +/* Elements of the linked list at Vdbe.pAuxData */ +typedef struct AuxData AuxData; + +/* Types of VDBE cursors */ +#define CURTYPE_BTREE 0 +#define CURTYPE_SORTER 1 +#define CURTYPE_VTAB 2 +#define CURTYPE_PSEUDO 3 + +/* +** A VdbeCursor is an superclass (a wrapper) for various cursor objects: +** +** * A b-tree cursor +** - In the main database or in an ephemeral database +** - On either an index or a table +** * A sorter +** * A virtual table +** * A one-row "pseudotable" stored in a single register +*/ +typedef struct VdbeCursor VdbeCursor; +struct VdbeCursor { + u8 eCurType; /* One of the CURTYPE_* values above */ + i8 iDb; /* Index of cursor database in db->aDb[] (or -1) */ + u8 nullRow; /* True if pointing to a row with no data */ + u8 deferredMoveto; /* A call to sqlite3BtreeMoveto() is needed */ + u8 isTable; /* True for rowid tables. False for indexes */ +#ifdef SQLITE_DEBUG + u8 seekOp; /* Most recent seek operation on this cursor */ + u8 wrFlag; /* The wrFlag argument to sqlite3BtreeCursor() */ +#endif + Bool isEphemeral:1; /* True for an ephemeral table */ + Bool useRandomRowid:1; /* Generate new record numbers semi-randomly */ + Bool isOrdered:1; /* True if the table is not BTREE_UNORDERED */ + Bool seekHit:1; /* See the OP_SeekHit and OP_IfNoHope opcodes */ + Btree *pBtx; /* Separate file holding temporary table */ + i64 seqCount; /* Sequence counter */ + int *aAltMap; /* Mapping from table to index column numbers */ + + /* Cached OP_Column parse information is only valid if cacheStatus matches + ** Vdbe.cacheCtr. Vdbe.cacheCtr will never take on the value of + ** CACHE_STALE (0) and so setting cacheStatus=CACHE_STALE guarantees that + ** the cache is out of date. */ + u32 cacheStatus; /* Cache is valid if this matches Vdbe.cacheCtr */ + int seekResult; /* Result of previous sqlite3BtreeMoveto() or 0 + ** if there have been no prior seeks on the cursor. */ + /* seekResult does not distinguish between "no seeks have ever occurred + ** on this cursor" and "the most recent seek was an exact match". + ** For CURTYPE_PSEUDO, seekResult is the register holding the record */ + + /* When a new VdbeCursor is allocated, only the fields above are zeroed. + ** The fields that follow are uninitialized, and must be individually + ** initialized prior to first use. */ + VdbeCursor *pAltCursor; /* Associated index cursor from which to read */ + union { + BtCursor *pCursor; /* CURTYPE_BTREE or _PSEUDO. Btree cursor */ + sqlite3_vtab_cursor *pVCur; /* CURTYPE_VTAB. Vtab cursor */ + VdbeSorter *pSorter; /* CURTYPE_SORTER. Sorter object */ + } uc; + KeyInfo *pKeyInfo; /* Info about index keys needed by index cursors */ + u32 iHdrOffset; /* Offset to next unparsed byte of the header */ + Pgno pgnoRoot; /* Root page of the open btree cursor */ + i16 nField; /* Number of fields in the header */ + u16 nHdrParsed; /* Number of header fields parsed so far */ + i64 movetoTarget; /* Argument to the deferred sqlite3BtreeMoveto() */ + u32 *aOffset; /* Pointer to aType[nField] */ + const u8 *aRow; /* Data for the current row, if all on one page */ + u32 payloadSize; /* Total number of bytes in the record */ + u32 szRow; /* Byte available in aRow */ +#ifdef SQLITE_ENABLE_COLUMN_USED_MASK + u64 maskUsed; /* Mask of columns used by this cursor */ +#endif + + /* 2*nField extra array elements allocated for aType[], beyond the one + ** static element declared in the structure. nField total array slots for + ** aType[] and nField+1 array slots for aOffset[] */ + u32 aType[1]; /* Type values record decode. MUST BE LAST */ +}; + + +/* +** A value for VdbeCursor.cacheStatus that means the cache is always invalid. +*/ +#define CACHE_STALE 0 + +/* +** When a sub-program is executed (OP_Program), a structure of this type +** is allocated to store the current value of the program counter, as +** well as the current memory cell array and various other frame specific +** values stored in the Vdbe struct. When the sub-program is finished, +** these values are copied back to the Vdbe from the VdbeFrame structure, +** restoring the state of the VM to as it was before the sub-program +** began executing. +** +** The memory for a VdbeFrame object is allocated and managed by a memory +** cell in the parent (calling) frame. When the memory cell is deleted or +** overwritten, the VdbeFrame object is not freed immediately. Instead, it +** is linked into the Vdbe.pDelFrame list. The contents of the Vdbe.pDelFrame +** list is deleted when the VM is reset in VdbeHalt(). The reason for doing +** this instead of deleting the VdbeFrame immediately is to avoid recursive +** calls to sqlite3VdbeMemRelease() when the memory cells belonging to the +** child frame are released. +** +** The currently executing frame is stored in Vdbe.pFrame. Vdbe.pFrame is +** set to NULL if the currently executing frame is the main program. +*/ +typedef struct VdbeFrame VdbeFrame; +struct VdbeFrame { + Vdbe *v; /* VM this frame belongs to */ + VdbeFrame *pParent; /* Parent of this frame, or NULL if parent is main */ + Op *aOp; /* Program instructions for parent frame */ + i64 *anExec; /* Event counters from parent frame */ + Mem *aMem; /* Array of memory cells for parent frame */ + VdbeCursor **apCsr; /* Array of Vdbe cursors for parent frame */ + u8 *aOnce; /* Bitmask used by OP_Once */ + void *token; /* Copy of SubProgram.token */ + i64 lastRowid; /* Last insert rowid (sqlite3.lastRowid) */ + AuxData *pAuxData; /* Linked list of auxdata allocations */ +#if SQLITE_DEBUG + u32 iFrameMagic; /* magic number for sanity checking */ +#endif + int nCursor; /* Number of entries in apCsr */ + int pc; /* Program Counter in parent (calling) frame */ + int nOp; /* Size of aOp array */ + int nMem; /* Number of entries in aMem */ + int nChildMem; /* Number of memory cells for child frame */ + int nChildCsr; /* Number of cursors for child frame */ + int nChange; /* Statement changes (Vdbe.nChange) */ + int nDbChange; /* Value of db->nChange */ +}; + +/* Magic number for sanity checking on VdbeFrame objects */ +#define SQLITE_FRAME_MAGIC 0x879fb71e + +/* +** Return a pointer to the array of registers allocated for use +** by a VdbeFrame. +*/ +#define VdbeFrameMem(p) ((Mem *)&((u8 *)p)[ROUND8(sizeof(VdbeFrame))]) + +/* +** Internally, the vdbe manipulates nearly all SQL values as Mem +** structures. Each Mem struct may cache multiple representations (string, +** integer etc.) of the same value. +*/ +struct sqlite3_value { + union MemValue { + double r; /* Real value used when MEM_Real is set in flags */ + i64 i; /* Integer value used when MEM_Int is set in flags */ + int nZero; /* Extra zero bytes when MEM_Zero and MEM_Blob set */ + const char *zPType; /* Pointer type when MEM_Term|MEM_Subtype|MEM_Null */ + FuncDef *pDef; /* Used only when flags==MEM_Agg */ + } u; + u16 flags; /* Some combination of MEM_Null, MEM_Str, MEM_Dyn, etc. */ + u8 enc; /* SQLITE_UTF8, SQLITE_UTF16BE, SQLITE_UTF16LE */ + u8 eSubtype; /* Subtype for this value */ + int n; /* Number of characters in string value, excluding '\0' */ + char *z; /* String or BLOB value */ + /* ShallowCopy only needs to copy the information above */ + char *zMalloc; /* Space to hold MEM_Str or MEM_Blob if szMalloc>0 */ + int szMalloc; /* Size of the zMalloc allocation */ + u32 uTemp; /* Transient storage for serial_type in OP_MakeRecord */ + sqlite3 *db; /* The associated database connection */ + void (*xDel)(void*);/* Destructor for Mem.z - only valid if MEM_Dyn */ +#ifdef SQLITE_DEBUG + Mem *pScopyFrom; /* This Mem is a shallow copy of pScopyFrom */ + u16 mScopyFlags; /* flags value immediately after the shallow copy */ +#endif +}; + +/* +** Size of struct Mem not including the Mem.zMalloc member or anything that +** follows. +*/ +#define MEMCELLSIZE offsetof(Mem,zMalloc) + +/* One or more of the following flags are set to indicate the validOK +** representations of the value stored in the Mem struct. +** +** If the MEM_Null flag is set, then the value is an SQL NULL value. +** For a pointer type created using sqlite3_bind_pointer() or +** sqlite3_result_pointer() the MEM_Term and MEM_Subtype flags are also set. +** +** If the MEM_Str flag is set then Mem.z points at a string representation. +** Usually this is encoded in the same unicode encoding as the main +** database (see below for exceptions). If the MEM_Term flag is also +** set, then the string is nul terminated. The MEM_Int and MEM_Real +** flags may coexist with the MEM_Str flag. +*/ +#define MEM_Null 0x0001 /* Value is NULL (or a pointer) */ +#define MEM_Str 0x0002 /* Value is a string */ +#define MEM_Int 0x0004 /* Value is an integer */ +#define MEM_Real 0x0008 /* Value is a real number */ +#define MEM_Blob 0x0010 /* Value is a BLOB */ +#define MEM_IntReal 0x0020 /* MEM_Int that stringifies like MEM_Real */ +#define MEM_AffMask 0x003f /* Mask of affinity bits */ +#define MEM_FromBind 0x0040 /* Value originates from sqlite3_bind() */ +#define MEM_Undefined 0x0080 /* Value is undefined */ +#define MEM_Cleared 0x0100 /* NULL set by OP_Null, not from data */ +#define MEM_TypeMask 0xc1bf /* Mask of type bits */ + + +/* Whenever Mem contains a valid string or blob representation, one of +** the following flags must be set to determine the memory management +** policy for Mem.z. The MEM_Term flag tells us whether or not the +** string is \000 or \u0000 terminated +*/ +#define MEM_Term 0x0200 /* String in Mem.z is zero terminated */ +#define MEM_Dyn 0x0400 /* Need to call Mem.xDel() on Mem.z */ +#define MEM_Static 0x0800 /* Mem.z points to a static string */ +#define MEM_Ephem 0x1000 /* Mem.z points to an ephemeral string */ +#define MEM_Agg 0x2000 /* Mem.z points to an agg function context */ +#define MEM_Zero 0x4000 /* Mem.i contains count of 0s appended to blob */ +#define MEM_Subtype 0x8000 /* Mem.eSubtype is valid */ +#ifdef SQLITE_OMIT_INCRBLOB + #undef MEM_Zero + #define MEM_Zero 0x0000 +#endif + +/* Return TRUE if Mem X contains dynamically allocated content - anything +** that needs to be deallocated to avoid a leak. +*/ +#define VdbeMemDynamic(X) \ + (((X)->flags&(MEM_Agg|MEM_Dyn))!=0) + +/* +** Clear any existing type flags from a Mem and replace them with f +*/ +#define MemSetTypeFlag(p, f) \ + ((p)->flags = ((p)->flags&~(MEM_TypeMask|MEM_Zero))|f) + +/* +** True if Mem X is a NULL-nochng type. +*/ +#define MemNullNochng(X) \ + ((X)->flags==(MEM_Null|MEM_Zero) && (X)->n==0 && (X)->u.nZero==0) + +/* +** Return true if a memory cell is not marked as invalid. This macro +** is for use inside assert() statements only. +*/ +#ifdef SQLITE_DEBUG +#define memIsValid(M) ((M)->flags & MEM_Undefined)==0 +#endif + +/* +** Each auxiliary data pointer stored by a user defined function +** implementation calling sqlite3_set_auxdata() is stored in an instance +** of this structure. All such structures associated with a single VM +** are stored in a linked list headed at Vdbe.pAuxData. All are destroyed +** when the VM is halted (if not before). +*/ +struct AuxData { + int iAuxOp; /* Instruction number of OP_Function opcode */ + int iAuxArg; /* Index of function argument. */ + void *pAux; /* Aux data pointer */ + void (*xDeleteAux)(void*); /* Destructor for the aux data */ + AuxData *pNextAux; /* Next element in list */ +}; + +/* +** The "context" argument for an installable function. A pointer to an +** instance of this structure is the first argument to the routines used +** implement the SQL functions. +** +** There is a typedef for this structure in sqlite.h. So all routines, +** even the public interface to SQLite, can use a pointer to this structure. +** But this file is the only place where the internal details of this +** structure are known. +** +** This structure is defined inside of vdbeInt.h because it uses substructures +** (Mem) which are only defined there. +*/ +struct sqlite3_context { + Mem *pOut; /* The return value is stored here */ + FuncDef *pFunc; /* Pointer to function information */ + Mem *pMem; /* Memory cell used to store aggregate context */ + Vdbe *pVdbe; /* The VM that owns this context */ + int iOp; /* Instruction number of OP_Function */ + int isError; /* Error code returned by the function. */ + u8 skipFlag; /* Skip accumulator loading if true */ + u8 argc; /* Number of arguments */ + sqlite3_value *argv[1]; /* Argument set */ +}; + +/* A bitfield type for use inside of structures. Always follow with :N where +** N is the number of bits. +*/ +typedef unsigned bft; /* Bit Field Type */ + +/* The ScanStatus object holds a single value for the +** sqlite3_stmt_scanstatus() interface. +*/ +typedef struct ScanStatus ScanStatus; +struct ScanStatus { + int addrExplain; /* OP_Explain for loop */ + int addrLoop; /* Address of "loops" counter */ + int addrVisit; /* Address of "rows visited" counter */ + int iSelectID; /* The "Select-ID" for this loop */ + LogEst nEst; /* Estimated output rows per loop */ + char *zName; /* Name of table or index */ +}; + +/* The DblquoteStr object holds the text of a double-quoted +** string for a prepared statement. A linked list of these objects +** is constructed during statement parsing and is held on Vdbe.pDblStr. +** When computing a normalized SQL statement for an SQL statement, that +** list is consulted for each double-quoted identifier to see if the +** identifier should really be a string literal. +*/ +typedef struct DblquoteStr DblquoteStr; +struct DblquoteStr { + DblquoteStr *pNextStr; /* Next string literal in the list */ + char z[8]; /* Dequoted value for the string */ +}; + +/* +** An instance of the virtual machine. This structure contains the complete +** state of the virtual machine. +** +** The "sqlite3_stmt" structure pointer that is returned by sqlite3_prepare() +** is really a pointer to an instance of this structure. +*/ +struct Vdbe { + sqlite3 *db; /* The database connection that owns this statement */ + Vdbe *pPrev,*pNext; /* Linked list of VDBEs with the same Vdbe.db */ + Parse *pParse; /* Parsing context used to create this Vdbe */ + ynVar nVar; /* Number of entries in aVar[] */ + u32 magic; /* Magic number for sanity checking */ + int nMem; /* Number of memory locations currently allocated */ + int nCursor; /* Number of slots in apCsr[] */ + u32 cacheCtr; /* VdbeCursor row cache generation counter */ + int pc; /* The program counter */ + int rc; /* Value to return */ + int nChange; /* Number of db changes made since last reset */ + int iStatement; /* Statement number (or 0 if has no opened stmt) */ + i64 iCurrentTime; /* Value of julianday('now') for this statement */ + i64 nFkConstraint; /* Number of imm. FK constraints this VM */ + i64 nStmtDefCons; /* Number of def. constraints when stmt started */ + i64 nStmtDefImmCons; /* Number of def. imm constraints when stmt started */ + Mem *aMem; /* The memory locations */ + Mem **apArg; /* Arguments to currently executing user function */ + VdbeCursor **apCsr; /* One element of this array for each open cursor */ + Mem *aVar; /* Values for the OP_Variable opcode. */ + + /* When allocating a new Vdbe object, all of the fields below should be + ** initialized to zero or NULL */ + + Op *aOp; /* Space to hold the virtual machine's program */ + int nOp; /* Number of instructions in the program */ + int nOpAlloc; /* Slots allocated for aOp[] */ + Mem *aColName; /* Column names to return */ + Mem *pResultSet; /* Pointer to an array of results */ + char *zErrMsg; /* Error message written here */ + VList *pVList; /* Name of variables */ +#ifndef SQLITE_OMIT_TRACE + i64 startTime; /* Time when query started - used for profiling */ +#endif +#ifdef SQLITE_DEBUG + int rcApp; /* errcode set by sqlite3_result_error_code() */ + u32 nWrite; /* Number of write operations that have occurred */ +#endif + u16 nResColumn; /* Number of columns in one row of the result set */ + u8 errorAction; /* Recovery action to do in case of an error */ + u8 minWriteFileFormat; /* Minimum file format for writable database files */ + u8 prepFlags; /* SQLITE_PREPARE_* flags */ + bft expired:2; /* 1: recompile VM immediately 2: when convenient */ + bft explain:2; /* True if EXPLAIN present on SQL command */ + bft doingRerun:1; /* True if rerunning after an auto-reprepare */ + bft changeCntOn:1; /* True to update the change-counter */ + bft runOnlyOnce:1; /* Automatically expire on reset */ + bft usesStmtJournal:1; /* True if uses a statement journal */ + bft readOnly:1; /* True for statements that do not write */ + bft bIsReader:1; /* True for statements that read */ + yDbMask btreeMask; /* Bitmask of db->aDb[] entries referenced */ + yDbMask lockMask; /* Subset of btreeMask that requires a lock */ + u32 aCounter[7]; /* Counters used by sqlite3_stmt_status() */ + char *zSql; /* Text of the SQL statement that generated this */ +#ifdef SQLITE_ENABLE_NORMALIZE + char *zNormSql; /* Normalization of the associated SQL statement */ + DblquoteStr *pDblStr; /* List of double-quoted string literals */ +#endif + void *pFree; /* Free this when deleting the vdbe */ + VdbeFrame *pFrame; /* Parent frame */ + VdbeFrame *pDelFrame; /* List of frame objects to free on VM reset */ + int nFrame; /* Number of frames in pFrame list */ + u32 expmask; /* Binding to these vars invalidates VM */ + SubProgram *pProgram; /* Linked list of all sub-programs used by VM */ + AuxData *pAuxData; /* Linked list of auxdata allocations */ +#ifdef SQLITE_ENABLE_STMT_SCANSTATUS + i64 *anExec; /* Number of times each op has been executed */ + int nScan; /* Entries in aScan[] */ + ScanStatus *aScan; /* Scan definitions for sqlite3_stmt_scanstatus() */ +#endif +}; + +/* +** The following are allowed values for Vdbe.magic +*/ +#define VDBE_MAGIC_INIT 0x16bceaa5 /* Building a VDBE program */ +#define VDBE_MAGIC_RUN 0x2df20da3 /* VDBE is ready to execute */ +#define VDBE_MAGIC_HALT 0x319c2973 /* VDBE has completed execution */ +#define VDBE_MAGIC_RESET 0x48fa9f76 /* Reset and ready to run again */ +#define VDBE_MAGIC_DEAD 0x5606c3c8 /* The VDBE has been deallocated */ + +/* +** Structure used to store the context required by the +** sqlite3_preupdate_*() API functions. +*/ +struct PreUpdate { + Vdbe *v; + VdbeCursor *pCsr; /* Cursor to read old values from */ + int op; /* One of SQLITE_INSERT, UPDATE, DELETE */ + u8 *aRecord; /* old.* database record */ + KeyInfo keyinfo; + UnpackedRecord *pUnpacked; /* Unpacked version of aRecord[] */ + UnpackedRecord *pNewUnpacked; /* Unpacked version of new.* record */ + int iNewReg; /* Register for new.* values */ + i64 iKey1; /* First key value passed to hook */ + i64 iKey2; /* Second key value passed to hook */ + Mem *aNew; /* Array of new.* values */ + Table *pTab; /* Schema object being upated */ + Index *pPk; /* PK index if pTab is WITHOUT ROWID */ +}; + +/* +** Function prototypes +*/ +SQLITE_PRIVATE void sqlite3VdbeError(Vdbe*, const char *, ...); +SQLITE_PRIVATE void sqlite3VdbeFreeCursor(Vdbe *, VdbeCursor*); +void sqliteVdbePopStack(Vdbe*,int); +SQLITE_PRIVATE int sqlite3VdbeCursorMoveto(VdbeCursor**, int*); +SQLITE_PRIVATE int sqlite3VdbeCursorRestore(VdbeCursor*); +SQLITE_PRIVATE u32 sqlite3VdbeSerialTypeLen(u32); +SQLITE_PRIVATE u8 sqlite3VdbeOneByteSerialTypeLen(u8); +SQLITE_PRIVATE u32 sqlite3VdbeSerialPut(unsigned char*, Mem*, u32); +SQLITE_PRIVATE u32 sqlite3VdbeSerialGet(const unsigned char*, u32, Mem*); +SQLITE_PRIVATE void sqlite3VdbeDeleteAuxData(sqlite3*, AuxData**, int, int); + +int sqlite2BtreeKeyCompare(BtCursor *, const void *, int, int, int *); +SQLITE_PRIVATE int sqlite3VdbeIdxKeyCompare(sqlite3*,VdbeCursor*,UnpackedRecord*,int*); +SQLITE_PRIVATE int sqlite3VdbeIdxRowid(sqlite3*, BtCursor*, i64*); +SQLITE_PRIVATE int sqlite3VdbeExec(Vdbe*); +#ifndef SQLITE_OMIT_EXPLAIN +SQLITE_PRIVATE int sqlite3VdbeList(Vdbe*); +#endif +SQLITE_PRIVATE int sqlite3VdbeHalt(Vdbe*); +SQLITE_PRIVATE int sqlite3VdbeChangeEncoding(Mem *, int); +SQLITE_PRIVATE int sqlite3VdbeMemTooBig(Mem*); +SQLITE_PRIVATE int sqlite3VdbeMemCopy(Mem*, const Mem*); +SQLITE_PRIVATE void sqlite3VdbeMemShallowCopy(Mem*, const Mem*, int); +SQLITE_PRIVATE void sqlite3VdbeMemMove(Mem*, Mem*); +SQLITE_PRIVATE int sqlite3VdbeMemNulTerminate(Mem*); +SQLITE_PRIVATE int sqlite3VdbeMemSetStr(Mem*, const char*, int, u8, void(*)(void*)); +SQLITE_PRIVATE void sqlite3VdbeMemSetInt64(Mem*, i64); +#ifdef SQLITE_OMIT_FLOATING_POINT +# define sqlite3VdbeMemSetDouble sqlite3VdbeMemSetInt64 +#else +SQLITE_PRIVATE void sqlite3VdbeMemSetDouble(Mem*, double); +#endif +SQLITE_PRIVATE void sqlite3VdbeMemSetPointer(Mem*, void*, const char*, void(*)(void*)); +SQLITE_PRIVATE void sqlite3VdbeMemInit(Mem*,sqlite3*,u16); +SQLITE_PRIVATE void sqlite3VdbeMemSetNull(Mem*); +SQLITE_PRIVATE void sqlite3VdbeMemSetZeroBlob(Mem*,int); +#ifdef SQLITE_DEBUG +SQLITE_PRIVATE int sqlite3VdbeMemIsRowSet(const Mem*); +#endif +SQLITE_PRIVATE int sqlite3VdbeMemSetRowSet(Mem*); +SQLITE_PRIVATE int sqlite3VdbeMemMakeWriteable(Mem*); +SQLITE_PRIVATE int sqlite3VdbeMemStringify(Mem*, u8, u8); +SQLITE_PRIVATE i64 sqlite3VdbeIntValue(Mem*); +SQLITE_PRIVATE int sqlite3VdbeMemIntegerify(Mem*); +SQLITE_PRIVATE double sqlite3VdbeRealValue(Mem*); +SQLITE_PRIVATE int sqlite3VdbeBooleanValue(Mem*, int ifNull); +SQLITE_PRIVATE void sqlite3VdbeIntegerAffinity(Mem*); +SQLITE_PRIVATE int sqlite3VdbeMemRealify(Mem*); +SQLITE_PRIVATE int sqlite3VdbeMemNumerify(Mem*); +SQLITE_PRIVATE void sqlite3VdbeMemCast(Mem*,u8,u8); +SQLITE_PRIVATE int sqlite3VdbeMemFromBtree(BtCursor*,u32,u32,Mem*); +SQLITE_PRIVATE void sqlite3VdbeMemRelease(Mem *p); +SQLITE_PRIVATE int sqlite3VdbeMemFinalize(Mem*, FuncDef*); +#ifndef SQLITE_OMIT_WINDOWFUNC +SQLITE_PRIVATE int sqlite3VdbeMemAggValue(Mem*, Mem*, FuncDef*); +#endif +#ifndef SQLITE_OMIT_EXPLAIN +SQLITE_PRIVATE const char *sqlite3OpcodeName(int); +#endif +SQLITE_PRIVATE int sqlite3VdbeMemGrow(Mem *pMem, int n, int preserve); +SQLITE_PRIVATE int sqlite3VdbeMemClearAndResize(Mem *pMem, int n); +SQLITE_PRIVATE int sqlite3VdbeCloseStatement(Vdbe *, int); +#ifdef SQLITE_DEBUG +SQLITE_PRIVATE int sqlite3VdbeFrameIsValid(VdbeFrame*); +#endif +SQLITE_PRIVATE void sqlite3VdbeFrameMemDel(void*); /* Destructor on Mem */ +SQLITE_PRIVATE void sqlite3VdbeFrameDelete(VdbeFrame*); /* Actually deletes the Frame */ +SQLITE_PRIVATE int sqlite3VdbeFrameRestore(VdbeFrame *); +#ifdef SQLITE_ENABLE_PREUPDATE_HOOK +SQLITE_PRIVATE void sqlite3VdbePreUpdateHook(Vdbe*,VdbeCursor*,int,const char*,Table*,i64,int); +#endif +SQLITE_PRIVATE int sqlite3VdbeTransferError(Vdbe *p); + +SQLITE_PRIVATE int sqlite3VdbeSorterInit(sqlite3 *, int, VdbeCursor *); +SQLITE_PRIVATE void sqlite3VdbeSorterReset(sqlite3 *, VdbeSorter *); +SQLITE_PRIVATE void sqlite3VdbeSorterClose(sqlite3 *, VdbeCursor *); +SQLITE_PRIVATE int sqlite3VdbeSorterRowkey(const VdbeCursor *, Mem *); +SQLITE_PRIVATE int sqlite3VdbeSorterNext(sqlite3 *, const VdbeCursor *); +SQLITE_PRIVATE int sqlite3VdbeSorterRewind(const VdbeCursor *, int *); +SQLITE_PRIVATE int sqlite3VdbeSorterWrite(const VdbeCursor *, Mem *); +SQLITE_PRIVATE int sqlite3VdbeSorterCompare(const VdbeCursor *, Mem *, int, int *); + +#ifdef SQLITE_DEBUG +SQLITE_PRIVATE void sqlite3VdbeIncrWriteCounter(Vdbe*, VdbeCursor*); +SQLITE_PRIVATE void sqlite3VdbeAssertAbortable(Vdbe*); +#else +# define sqlite3VdbeIncrWriteCounter(V,C) +# define sqlite3VdbeAssertAbortable(V) +#endif + +#if !defined(SQLITE_OMIT_SHARED_CACHE) +SQLITE_PRIVATE void sqlite3VdbeEnter(Vdbe*); +#else +# define sqlite3VdbeEnter(X) +#endif + +#if !defined(SQLITE_OMIT_SHARED_CACHE) && SQLITE_THREADSAFE>0 +SQLITE_PRIVATE void sqlite3VdbeLeave(Vdbe*); +#else +# define sqlite3VdbeLeave(X) +#endif + +#ifdef SQLITE_DEBUG +SQLITE_PRIVATE void sqlite3VdbeMemAboutToChange(Vdbe*,Mem*); +SQLITE_PRIVATE int sqlite3VdbeCheckMemInvariants(Mem*); +#endif + +#ifndef SQLITE_OMIT_FOREIGN_KEY +SQLITE_PRIVATE int sqlite3VdbeCheckFk(Vdbe *, int); +#else +# define sqlite3VdbeCheckFk(p,i) 0 +#endif + +#ifdef SQLITE_DEBUG +SQLITE_PRIVATE void sqlite3VdbePrintSql(Vdbe*); +SQLITE_PRIVATE void sqlite3VdbeMemPrettyPrint(Mem *pMem, char *zBuf); +#endif +#ifndef SQLITE_OMIT_UTF16 +SQLITE_PRIVATE int sqlite3VdbeMemTranslate(Mem*, u8); +SQLITE_PRIVATE int sqlite3VdbeMemHandleBom(Mem *pMem); +#endif + +#ifndef SQLITE_OMIT_INCRBLOB +SQLITE_PRIVATE int sqlite3VdbeMemExpandBlob(Mem *); + #define ExpandBlob(P) (((P)->flags&MEM_Zero)?sqlite3VdbeMemExpandBlob(P):0) +#else + #define sqlite3VdbeMemExpandBlob(x) SQLITE_OK + #define ExpandBlob(P) SQLITE_OK +#endif + +#endif /* !defined(SQLITE_VDBEINT_H) */ + +/************** End of vdbeInt.h *********************************************/ +/************** Continuing where we left off in status.c *********************/ + +/* +** Variables in which to record status information. +*/ +#if SQLITE_PTRSIZE>4 +typedef sqlite3_int64 sqlite3StatValueType; +#else +typedef u32 sqlite3StatValueType; +#endif +typedef struct sqlite3StatType sqlite3StatType; +static SQLITE_WSD struct sqlite3StatType { + sqlite3StatValueType nowValue[10]; /* Current value */ + sqlite3StatValueType mxValue[10]; /* Maximum value */ +} sqlite3Stat = { {0,}, {0,} }; + +/* +** Elements of sqlite3Stat[] are protected by either the memory allocator +** mutex, or by the pcache1 mutex. The following array determines which. +*/ +static const char statMutex[] = { + 0, /* SQLITE_STATUS_MEMORY_USED */ + 1, /* SQLITE_STATUS_PAGECACHE_USED */ + 1, /* SQLITE_STATUS_PAGECACHE_OVERFLOW */ + 0, /* SQLITE_STATUS_SCRATCH_USED */ + 0, /* SQLITE_STATUS_SCRATCH_OVERFLOW */ + 0, /* SQLITE_STATUS_MALLOC_SIZE */ + 0, /* SQLITE_STATUS_PARSER_STACK */ + 1, /* SQLITE_STATUS_PAGECACHE_SIZE */ + 0, /* SQLITE_STATUS_SCRATCH_SIZE */ + 0, /* SQLITE_STATUS_MALLOC_COUNT */ +}; + + +/* The "wsdStat" macro will resolve to the status information +** state vector. If writable static data is unsupported on the target, +** we have to locate the state vector at run-time. In the more common +** case where writable static data is supported, wsdStat can refer directly +** to the "sqlite3Stat" state vector declared above. +*/ +#ifdef SQLITE_OMIT_WSD +# define wsdStatInit sqlite3StatType *x = &GLOBAL(sqlite3StatType,sqlite3Stat) +# define wsdStat x[0] +#else +# define wsdStatInit +# define wsdStat sqlite3Stat +#endif + +/* +** Return the current value of a status parameter. The caller must +** be holding the appropriate mutex. +*/ +SQLITE_PRIVATE sqlite3_int64 sqlite3StatusValue(int op){ + wsdStatInit; + assert( op>=0 && op=0 && op=0 && op=0 && opwsdStat.mxValue[op] ){ + wsdStat.mxValue[op] = wsdStat.nowValue[op]; + } +} +SQLITE_PRIVATE void sqlite3StatusDown(int op, int N){ + wsdStatInit; + assert( N>=0 ); + assert( op>=0 && op=0 && op=0 ); + newValue = (sqlite3StatValueType)X; + assert( op>=0 && op=0 && opwsdStat.mxValue[op] ){ + wsdStat.mxValue[op] = newValue; + } +} + +/* +** Query status information. +*/ +SQLITE_API int sqlite3_status64( + int op, + sqlite3_int64 *pCurrent, + sqlite3_int64 *pHighwater, + int resetFlag +){ + sqlite3_mutex *pMutex; + wsdStatInit; + if( op<0 || op>=ArraySize(wsdStat.nowValue) ){ + return SQLITE_MISUSE_BKPT; + } +#ifdef SQLITE_ENABLE_API_ARMOR + if( pCurrent==0 || pHighwater==0 ) return SQLITE_MISUSE_BKPT; +#endif + pMutex = statMutex[op] ? sqlite3Pcache1Mutex() : sqlite3MallocMutex(); + sqlite3_mutex_enter(pMutex); + *pCurrent = wsdStat.nowValue[op]; + *pHighwater = wsdStat.mxValue[op]; + if( resetFlag ){ + wsdStat.mxValue[op] = wsdStat.nowValue[op]; + } + sqlite3_mutex_leave(pMutex); + (void)pMutex; /* Prevent warning when SQLITE_THREADSAFE=0 */ + return SQLITE_OK; +} +SQLITE_API int sqlite3_status(int op, int *pCurrent, int *pHighwater, int resetFlag){ + sqlite3_int64 iCur = 0, iHwtr = 0; + int rc; +#ifdef SQLITE_ENABLE_API_ARMOR + if( pCurrent==0 || pHighwater==0 ) return SQLITE_MISUSE_BKPT; +#endif + rc = sqlite3_status64(op, &iCur, &iHwtr, resetFlag); + if( rc==0 ){ + *pCurrent = (int)iCur; + *pHighwater = (int)iHwtr; + } + return rc; +} + +/* +** Return the number of LookasideSlot elements on the linked list +*/ +static u32 countLookasideSlots(LookasideSlot *p){ + u32 cnt = 0; + while( p ){ + p = p->pNext; + cnt++; + } + return cnt; +} + +/* +** Count the number of slots of lookaside memory that are outstanding +*/ +SQLITE_PRIVATE int sqlite3LookasideUsed(sqlite3 *db, int *pHighwater){ + u32 nInit = countLookasideSlots(db->lookaside.pInit); + u32 nFree = countLookasideSlots(db->lookaside.pFree); + if( pHighwater ) *pHighwater = db->lookaside.nSlot - nInit; + return db->lookaside.nSlot - (nInit+nFree); +} + +/* +** Query status information for a single database connection +*/ +SQLITE_API int sqlite3_db_status( + sqlite3 *db, /* The database connection whose status is desired */ + int op, /* Status verb */ + int *pCurrent, /* Write current value here */ + int *pHighwater, /* Write high-water mark here */ + int resetFlag /* Reset high-water mark if true */ +){ + int rc = SQLITE_OK; /* Return code */ +#ifdef SQLITE_ENABLE_API_ARMOR + if( !sqlite3SafetyCheckOk(db) || pCurrent==0|| pHighwater==0 ){ + return SQLITE_MISUSE_BKPT; + } +#endif + sqlite3_mutex_enter(db->mutex); + switch( op ){ + case SQLITE_DBSTATUS_LOOKASIDE_USED: { + *pCurrent = sqlite3LookasideUsed(db, pHighwater); + if( resetFlag ){ + LookasideSlot *p = db->lookaside.pFree; + if( p ){ + while( p->pNext ) p = p->pNext; + p->pNext = db->lookaside.pInit; + db->lookaside.pInit = db->lookaside.pFree; + db->lookaside.pFree = 0; + } + } + break; + } + + case SQLITE_DBSTATUS_LOOKASIDE_HIT: + case SQLITE_DBSTATUS_LOOKASIDE_MISS_SIZE: + case SQLITE_DBSTATUS_LOOKASIDE_MISS_FULL: { + testcase( op==SQLITE_DBSTATUS_LOOKASIDE_HIT ); + testcase( op==SQLITE_DBSTATUS_LOOKASIDE_MISS_SIZE ); + testcase( op==SQLITE_DBSTATUS_LOOKASIDE_MISS_FULL ); + assert( (op-SQLITE_DBSTATUS_LOOKASIDE_HIT)>=0 ); + assert( (op-SQLITE_DBSTATUS_LOOKASIDE_HIT)<3 ); + *pCurrent = 0; + *pHighwater = db->lookaside.anStat[op - SQLITE_DBSTATUS_LOOKASIDE_HIT]; + if( resetFlag ){ + db->lookaside.anStat[op - SQLITE_DBSTATUS_LOOKASIDE_HIT] = 0; + } + break; + } + + /* + ** Return an approximation for the amount of memory currently used + ** by all pagers associated with the given database connection. The + ** highwater mark is meaningless and is returned as zero. + */ + case SQLITE_DBSTATUS_CACHE_USED_SHARED: + case SQLITE_DBSTATUS_CACHE_USED: { + int totalUsed = 0; + int i; + sqlite3BtreeEnterAll(db); + for(i=0; inDb; i++){ + Btree *pBt = db->aDb[i].pBt; + if( pBt ){ + Pager *pPager = sqlite3BtreePager(pBt); + int nByte = sqlite3PagerMemUsed(pPager); + if( op==SQLITE_DBSTATUS_CACHE_USED_SHARED ){ + nByte = nByte / sqlite3BtreeConnectionCount(pBt); + } + totalUsed += nByte; + } + } + sqlite3BtreeLeaveAll(db); + *pCurrent = totalUsed; + *pHighwater = 0; + break; + } + + /* + ** *pCurrent gets an accurate estimate of the amount of memory used + ** to store the schema for all databases (main, temp, and any ATTACHed + ** databases. *pHighwater is set to zero. + */ + case SQLITE_DBSTATUS_SCHEMA_USED: { + int i; /* Used to iterate through schemas */ + int nByte = 0; /* Used to accumulate return value */ + + sqlite3BtreeEnterAll(db); + db->pnBytesFreed = &nByte; + for(i=0; inDb; i++){ + Schema *pSchema = db->aDb[i].pSchema; + if( ALWAYS(pSchema!=0) ){ + HashElem *p; + + nByte += sqlite3GlobalConfig.m.xRoundup(sizeof(HashElem)) * ( + pSchema->tblHash.count + + pSchema->trigHash.count + + pSchema->idxHash.count + + pSchema->fkeyHash.count + ); + nByte += sqlite3_msize(pSchema->tblHash.ht); + nByte += sqlite3_msize(pSchema->trigHash.ht); + nByte += sqlite3_msize(pSchema->idxHash.ht); + nByte += sqlite3_msize(pSchema->fkeyHash.ht); + + for(p=sqliteHashFirst(&pSchema->trigHash); p; p=sqliteHashNext(p)){ + sqlite3DeleteTrigger(db, (Trigger*)sqliteHashData(p)); + } + for(p=sqliteHashFirst(&pSchema->tblHash); p; p=sqliteHashNext(p)){ + sqlite3DeleteTable(db, (Table *)sqliteHashData(p)); + } + } + } + db->pnBytesFreed = 0; + sqlite3BtreeLeaveAll(db); + + *pHighwater = 0; + *pCurrent = nByte; + break; + } + + /* + ** *pCurrent gets an accurate estimate of the amount of memory used + ** to store all prepared statements. + ** *pHighwater is set to zero. + */ + case SQLITE_DBSTATUS_STMT_USED: { + struct Vdbe *pVdbe; /* Used to iterate through VMs */ + int nByte = 0; /* Used to accumulate return value */ + + db->pnBytesFreed = &nByte; + for(pVdbe=db->pVdbe; pVdbe; pVdbe=pVdbe->pNext){ + sqlite3VdbeClearObject(db, pVdbe); + sqlite3DbFree(db, pVdbe); + } + db->pnBytesFreed = 0; + + *pHighwater = 0; /* IMP: R-64479-57858 */ + *pCurrent = nByte; + + break; + } + + /* + ** Set *pCurrent to the total cache hits or misses encountered by all + ** pagers the database handle is connected to. *pHighwater is always set + ** to zero. + */ + case SQLITE_DBSTATUS_CACHE_SPILL: + op = SQLITE_DBSTATUS_CACHE_WRITE+1; + /* Fall through into the next case */ + case SQLITE_DBSTATUS_CACHE_HIT: + case SQLITE_DBSTATUS_CACHE_MISS: + case SQLITE_DBSTATUS_CACHE_WRITE:{ + int i; + int nRet = 0; + assert( SQLITE_DBSTATUS_CACHE_MISS==SQLITE_DBSTATUS_CACHE_HIT+1 ); + assert( SQLITE_DBSTATUS_CACHE_WRITE==SQLITE_DBSTATUS_CACHE_HIT+2 ); + + for(i=0; inDb; i++){ + if( db->aDb[i].pBt ){ + Pager *pPager = sqlite3BtreePager(db->aDb[i].pBt); + sqlite3PagerCacheStat(pPager, op, resetFlag, &nRet); + } + } + *pHighwater = 0; /* IMP: R-42420-56072 */ + /* IMP: R-54100-20147 */ + /* IMP: R-29431-39229 */ + *pCurrent = nRet; + break; + } + + /* Set *pCurrent to non-zero if there are unresolved deferred foreign + ** key constraints. Set *pCurrent to zero if all foreign key constraints + ** have been satisfied. The *pHighwater is always set to zero. + */ + case SQLITE_DBSTATUS_DEFERRED_FKS: { + *pHighwater = 0; /* IMP: R-11967-56545 */ + *pCurrent = db->nDeferredImmCons>0 || db->nDeferredCons>0; + break; + } + + default: { + rc = SQLITE_ERROR; + } + } + sqlite3_mutex_leave(db->mutex); + return rc; +} + +/************** End of status.c **********************************************/ +/************** Begin file date.c ********************************************/ +/* +** 2003 October 31 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** This file contains the C functions that implement date and time +** functions for SQLite. +** +** There is only one exported symbol in this file - the function +** sqlite3RegisterDateTimeFunctions() found at the bottom of the file. +** All other code has file scope. +** +** SQLite processes all times and dates as julian day numbers. The +** dates and times are stored as the number of days since noon +** in Greenwich on November 24, 4714 B.C. according to the Gregorian +** calendar system. +** +** 1970-01-01 00:00:00 is JD 2440587.5 +** 2000-01-01 00:00:00 is JD 2451544.5 +** +** This implementation requires years to be expressed as a 4-digit number +** which means that only dates between 0000-01-01 and 9999-12-31 can +** be represented, even though julian day numbers allow a much wider +** range of dates. +** +** The Gregorian calendar system is used for all dates and times, +** even those that predate the Gregorian calendar. Historians usually +** use the julian calendar for dates prior to 1582-10-15 and for some +** dates afterwards, depending on locale. Beware of this difference. +** +** The conversion algorithms are implemented based on descriptions +** in the following text: +** +** Jean Meeus +** Astronomical Algorithms, 2nd Edition, 1998 +** ISBN 0-943396-61-1 +** Willmann-Bell, Inc +** Richmond, Virginia (USA) +*/ +/* #include "sqliteInt.h" */ +/* #include */ +/* #include */ +#include + +#ifndef SQLITE_OMIT_DATETIME_FUNCS + +/* +** The MSVC CRT on Windows CE may not have a localtime() function. +** So declare a substitute. The substitute function itself is +** defined in "os_win.c". +*/ +#if !defined(SQLITE_OMIT_LOCALTIME) && defined(_WIN32_WCE) && \ + (!defined(SQLITE_MSVC_LOCALTIME_API) || !SQLITE_MSVC_LOCALTIME_API) +struct tm *__cdecl localtime(const time_t *); +#endif + +/* +** A structure for holding a single date and time. +*/ +typedef struct DateTime DateTime; +struct DateTime { + sqlite3_int64 iJD; /* The julian day number times 86400000 */ + int Y, M, D; /* Year, month, and day */ + int h, m; /* Hour and minutes */ + int tz; /* Timezone offset in minutes */ + double s; /* Seconds */ + char validJD; /* True (1) if iJD is valid */ + char rawS; /* Raw numeric value stored in s */ + char validYMD; /* True (1) if Y,M,D are valid */ + char validHMS; /* True (1) if h,m,s are valid */ + char validTZ; /* True (1) if tz is valid */ + char tzSet; /* Timezone was set explicitly */ + char isError; /* An overflow has occurred */ +}; + + +/* +** Convert zDate into one or more integers according to the conversion +** specifier zFormat. +** +** zFormat[] contains 4 characters for each integer converted, except for +** the last integer which is specified by three characters. The meaning +** of a four-character format specifiers ABCD is: +** +** A: number of digits to convert. Always "2" or "4". +** B: minimum value. Always "0" or "1". +** C: maximum value, decoded as: +** a: 12 +** b: 14 +** c: 24 +** d: 31 +** e: 59 +** f: 9999 +** D: the separator character, or \000 to indicate this is the +** last number to convert. +** +** Example: To translate an ISO-8601 date YYYY-MM-DD, the format would +** be "40f-21a-20c". The "40f-" indicates the 4-digit year followed by "-". +** The "21a-" indicates the 2-digit month followed by "-". The "20c" indicates +** the 2-digit day which is the last integer in the set. +** +** The function returns the number of successful conversions. +*/ +static int getDigits(const char *zDate, const char *zFormat, ...){ + /* The aMx[] array translates the 3rd character of each format + ** spec into a max size: a b c d e f */ + static const u16 aMx[] = { 12, 14, 24, 31, 59, 9999 }; + va_list ap; + int cnt = 0; + char nextC; + va_start(ap, zFormat); + do{ + char N = zFormat[0] - '0'; + char min = zFormat[1] - '0'; + int val = 0; + u16 max; + + assert( zFormat[2]>='a' && zFormat[2]<='f' ); + max = aMx[zFormat[2] - 'a']; + nextC = zFormat[3]; + val = 0; + while( N-- ){ + if( !sqlite3Isdigit(*zDate) ){ + goto end_getDigits; + } + val = val*10 + *zDate - '0'; + zDate++; + } + if( val<(int)min || val>(int)max || (nextC!=0 && nextC!=*zDate) ){ + goto end_getDigits; + } + *va_arg(ap,int*) = val; + zDate++; + cnt++; + zFormat += 4; + }while( nextC ); +end_getDigits: + va_end(ap); + return cnt; +} + +/* +** Parse a timezone extension on the end of a date-time. +** The extension is of the form: +** +** (+/-)HH:MM +** +** Or the "zulu" notation: +** +** Z +** +** If the parse is successful, write the number of minutes +** of change in p->tz and return 0. If a parser error occurs, +** return non-zero. +** +** A missing specifier is not considered an error. +*/ +static int parseTimezone(const char *zDate, DateTime *p){ + int sgn = 0; + int nHr, nMn; + int c; + while( sqlite3Isspace(*zDate) ){ zDate++; } + p->tz = 0; + c = *zDate; + if( c=='-' ){ + sgn = -1; + }else if( c=='+' ){ + sgn = +1; + }else if( c=='Z' || c=='z' ){ + zDate++; + goto zulu_time; + }else{ + return c!=0; + } + zDate++; + if( getDigits(zDate, "20b:20e", &nHr, &nMn)!=2 ){ + return 1; + } + zDate += 5; + p->tz = sgn*(nMn + nHr*60); +zulu_time: + while( sqlite3Isspace(*zDate) ){ zDate++; } + p->tzSet = 1; + return *zDate!=0; +} + +/* +** Parse times of the form HH:MM or HH:MM:SS or HH:MM:SS.FFFF. +** The HH, MM, and SS must each be exactly 2 digits. The +** fractional seconds FFFF can be one or more digits. +** +** Return 1 if there is a parsing error and 0 on success. +*/ +static int parseHhMmSs(const char *zDate, DateTime *p){ + int h, m, s; + double ms = 0.0; + if( getDigits(zDate, "20c:20e", &h, &m)!=2 ){ + return 1; + } + zDate += 5; + if( *zDate==':' ){ + zDate++; + if( getDigits(zDate, "20e", &s)!=1 ){ + return 1; + } + zDate += 2; + if( *zDate=='.' && sqlite3Isdigit(zDate[1]) ){ + double rScale = 1.0; + zDate++; + while( sqlite3Isdigit(*zDate) ){ + ms = ms*10.0 + *zDate - '0'; + rScale *= 10.0; + zDate++; + } + ms /= rScale; + } + }else{ + s = 0; + } + p->validJD = 0; + p->rawS = 0; + p->validHMS = 1; + p->h = h; + p->m = m; + p->s = s + ms; + if( parseTimezone(zDate, p) ) return 1; + p->validTZ = (p->tz!=0)?1:0; + return 0; +} + +/* +** Put the DateTime object into its error state. +*/ +static void datetimeError(DateTime *p){ + memset(p, 0, sizeof(*p)); + p->isError = 1; +} + +/* +** Convert from YYYY-MM-DD HH:MM:SS to julian day. We always assume +** that the YYYY-MM-DD is according to the Gregorian calendar. +** +** Reference: Meeus page 61 +*/ +static void computeJD(DateTime *p){ + int Y, M, D, A, B, X1, X2; + + if( p->validJD ) return; + if( p->validYMD ){ + Y = p->Y; + M = p->M; + D = p->D; + }else{ + Y = 2000; /* If no YMD specified, assume 2000-Jan-01 */ + M = 1; + D = 1; + } + if( Y<-4713 || Y>9999 || p->rawS ){ + datetimeError(p); + return; + } + if( M<=2 ){ + Y--; + M += 12; + } + A = Y/100; + B = 2 - A + (A/4); + X1 = 36525*(Y+4716)/100; + X2 = 306001*(M+1)/10000; + p->iJD = (sqlite3_int64)((X1 + X2 + D + B - 1524.5 ) * 86400000); + p->validJD = 1; + if( p->validHMS ){ + p->iJD += p->h*3600000 + p->m*60000 + (sqlite3_int64)(p->s*1000); + if( p->validTZ ){ + p->iJD -= p->tz*60000; + p->validYMD = 0; + p->validHMS = 0; + p->validTZ = 0; + } + } +} + +/* +** Parse dates of the form +** +** YYYY-MM-DD HH:MM:SS.FFF +** YYYY-MM-DD HH:MM:SS +** YYYY-MM-DD HH:MM +** YYYY-MM-DD +** +** Write the result into the DateTime structure and return 0 +** on success and 1 if the input string is not a well-formed +** date. +*/ +static int parseYyyyMmDd(const char *zDate, DateTime *p){ + int Y, M, D, neg; + + if( zDate[0]=='-' ){ + zDate++; + neg = 1; + }else{ + neg = 0; + } + if( getDigits(zDate, "40f-21a-21d", &Y, &M, &D)!=3 ){ + return 1; + } + zDate += 10; + while( sqlite3Isspace(*zDate) || 'T'==*(u8*)zDate ){ zDate++; } + if( parseHhMmSs(zDate, p)==0 ){ + /* We got the time */ + }else if( *zDate==0 ){ + p->validHMS = 0; + }else{ + return 1; + } + p->validJD = 0; + p->validYMD = 1; + p->Y = neg ? -Y : Y; + p->M = M; + p->D = D; + if( p->validTZ ){ + computeJD(p); + } + return 0; +} + +/* +** Set the time to the current time reported by the VFS. +** +** Return the number of errors. +*/ +static int setDateTimeToCurrent(sqlite3_context *context, DateTime *p){ + p->iJD = sqlite3StmtCurrentTime(context); + if( p->iJD>0 ){ + p->validJD = 1; + return 0; + }else{ + return 1; + } +} + +/* +** Input "r" is a numeric quantity which might be a julian day number, +** or the number of seconds since 1970. If the value if r is within +** range of a julian day number, install it as such and set validJD. +** If the value is a valid unix timestamp, put it in p->s and set p->rawS. +*/ +static void setRawDateNumber(DateTime *p, double r){ + p->s = r; + p->rawS = 1; + if( r>=0.0 && r<5373484.5 ){ + p->iJD = (sqlite3_int64)(r*86400000.0 + 0.5); + p->validJD = 1; + } +} + +/* +** Attempt to parse the given string into a julian day number. Return +** the number of errors. +** +** The following are acceptable forms for the input string: +** +** YYYY-MM-DD HH:MM:SS.FFF +/-HH:MM +** DDDD.DD +** now +** +** In the first form, the +/-HH:MM is always optional. The fractional +** seconds extension (the ".FFF") is optional. The seconds portion +** (":SS.FFF") is option. The year and date can be omitted as long +** as there is a time string. The time string can be omitted as long +** as there is a year and date. +*/ +static int parseDateOrTime( + sqlite3_context *context, + const char *zDate, + DateTime *p +){ + double r; + if( parseYyyyMmDd(zDate,p)==0 ){ + return 0; + }else if( parseHhMmSs(zDate, p)==0 ){ + return 0; + }else if( sqlite3StrICmp(zDate,"now")==0 && sqlite3NotPureFunc(context) ){ + return setDateTimeToCurrent(context, p); + }else if( sqlite3AtoF(zDate, &r, sqlite3Strlen30(zDate), SQLITE_UTF8)>0 ){ + setRawDateNumber(p, r); + return 0; + } + return 1; +} + +/* The julian day number for 9999-12-31 23:59:59.999 is 5373484.4999999. +** Multiplying this by 86400000 gives 464269060799999 as the maximum value +** for DateTime.iJD. +** +** But some older compilers (ex: gcc 4.2.1 on older Macs) cannot deal with +** such a large integer literal, so we have to encode it. +*/ +#define INT_464269060799999 ((((i64)0x1a640)<<32)|0x1072fdff) + +/* +** Return TRUE if the given julian day number is within range. +** +** The input is the JulianDay times 86400000. +*/ +static int validJulianDay(sqlite3_int64 iJD){ + return iJD>=0 && iJD<=INT_464269060799999; +} + +/* +** Compute the Year, Month, and Day from the julian day number. +*/ +static void computeYMD(DateTime *p){ + int Z, A, B, C, D, E, X1; + if( p->validYMD ) return; + if( !p->validJD ){ + p->Y = 2000; + p->M = 1; + p->D = 1; + }else if( !validJulianDay(p->iJD) ){ + datetimeError(p); + return; + }else{ + Z = (int)((p->iJD + 43200000)/86400000); + A = (int)((Z - 1867216.25)/36524.25); + A = Z + 1 + A - (A/4); + B = A + 1524; + C = (int)((B - 122.1)/365.25); + D = (36525*(C&32767))/100; + E = (int)((B-D)/30.6001); + X1 = (int)(30.6001*E); + p->D = B - D - X1; + p->M = E<14 ? E-1 : E-13; + p->Y = p->M>2 ? C - 4716 : C - 4715; + } + p->validYMD = 1; +} + +/* +** Compute the Hour, Minute, and Seconds from the julian day number. +*/ +static void computeHMS(DateTime *p){ + int s; + if( p->validHMS ) return; + computeJD(p); + s = (int)((p->iJD + 43200000) % 86400000); + p->s = s/1000.0; + s = (int)p->s; + p->s -= s; + p->h = s/3600; + s -= p->h*3600; + p->m = s/60; + p->s += s - p->m*60; + p->rawS = 0; + p->validHMS = 1; +} + +/* +** Compute both YMD and HMS +*/ +static void computeYMD_HMS(DateTime *p){ + computeYMD(p); + computeHMS(p); +} + +/* +** Clear the YMD and HMS and the TZ +*/ +static void clearYMD_HMS_TZ(DateTime *p){ + p->validYMD = 0; + p->validHMS = 0; + p->validTZ = 0; +} + +#ifndef SQLITE_OMIT_LOCALTIME +/* +** On recent Windows platforms, the localtime_s() function is available +** as part of the "Secure CRT". It is essentially equivalent to +** localtime_r() available under most POSIX platforms, except that the +** order of the parameters is reversed. +** +** See http://msdn.microsoft.com/en-us/library/a442x3ye(VS.80).aspx. +** +** If the user has not indicated to use localtime_r() or localtime_s() +** already, check for an MSVC build environment that provides +** localtime_s(). +*/ +#if !HAVE_LOCALTIME_R && !HAVE_LOCALTIME_S \ + && defined(_MSC_VER) && defined(_CRT_INSECURE_DEPRECATE) +#undef HAVE_LOCALTIME_S +#define HAVE_LOCALTIME_S 1 +#endif + +/* +** The following routine implements the rough equivalent of localtime_r() +** using whatever operating-system specific localtime facility that +** is available. This routine returns 0 on success and +** non-zero on any kind of error. +** +** If the sqlite3GlobalConfig.bLocaltimeFault variable is true then this +** routine will always fail. +** +** EVIDENCE-OF: R-62172-00036 In this implementation, the standard C +** library function localtime_r() is used to assist in the calculation of +** local time. +*/ +static int osLocaltime(time_t *t, struct tm *pTm){ + int rc; +#if !HAVE_LOCALTIME_R && !HAVE_LOCALTIME_S + struct tm *pX; +#if SQLITE_THREADSAFE>0 + sqlite3_mutex *mutex = sqlite3MutexAlloc(SQLITE_MUTEX_STATIC_MASTER); +#endif + sqlite3_mutex_enter(mutex); + pX = localtime(t); +#ifndef SQLITE_UNTESTABLE + if( sqlite3GlobalConfig.bLocaltimeFault ) pX = 0; +#endif + if( pX ) *pTm = *pX; + sqlite3_mutex_leave(mutex); + rc = pX==0; +#else +#ifndef SQLITE_UNTESTABLE + if( sqlite3GlobalConfig.bLocaltimeFault ) return 1; +#endif +#if HAVE_LOCALTIME_R + rc = localtime_r(t, pTm)==0; +#else + rc = localtime_s(pTm, t); +#endif /* HAVE_LOCALTIME_R */ +#endif /* HAVE_LOCALTIME_R || HAVE_LOCALTIME_S */ + return rc; +} +#endif /* SQLITE_OMIT_LOCALTIME */ + + +#ifndef SQLITE_OMIT_LOCALTIME +/* +** Compute the difference (in milliseconds) between localtime and UTC +** (a.k.a. GMT) for the time value p where p is in UTC. If no error occurs, +** return this value and set *pRc to SQLITE_OK. +** +** Or, if an error does occur, set *pRc to SQLITE_ERROR. The returned value +** is undefined in this case. +*/ +static sqlite3_int64 localtimeOffset( + DateTime *p, /* Date at which to calculate offset */ + sqlite3_context *pCtx, /* Write error here if one occurs */ + int *pRc /* OUT: Error code. SQLITE_OK or ERROR */ +){ + DateTime x, y; + time_t t; + struct tm sLocal; + + /* Initialize the contents of sLocal to avoid a compiler warning. */ + memset(&sLocal, 0, sizeof(sLocal)); + + x = *p; + computeYMD_HMS(&x); + if( x.Y<1971 || x.Y>=2038 ){ + /* EVIDENCE-OF: R-55269-29598 The localtime_r() C function normally only + ** works for years between 1970 and 2037. For dates outside this range, + ** SQLite attempts to map the year into an equivalent year within this + ** range, do the calculation, then map the year back. + */ + x.Y = 2000; + x.M = 1; + x.D = 1; + x.h = 0; + x.m = 0; + x.s = 0.0; + } else { + int s = (int)(x.s + 0.5); + x.s = s; + } + x.tz = 0; + x.validJD = 0; + computeJD(&x); + t = (time_t)(x.iJD/1000 - 21086676*(i64)10000); + if( osLocaltime(&t, &sLocal) ){ + sqlite3_result_error(pCtx, "local time unavailable", -1); + *pRc = SQLITE_ERROR; + return 0; + } + y.Y = sLocal.tm_year + 1900; + y.M = sLocal.tm_mon + 1; + y.D = sLocal.tm_mday; + y.h = sLocal.tm_hour; + y.m = sLocal.tm_min; + y.s = sLocal.tm_sec; + y.validYMD = 1; + y.validHMS = 1; + y.validJD = 0; + y.rawS = 0; + y.validTZ = 0; + y.isError = 0; + computeJD(&y); + *pRc = SQLITE_OK; + return y.iJD - x.iJD; +} +#endif /* SQLITE_OMIT_LOCALTIME */ + +/* +** The following table defines various date transformations of the form +** +** 'NNN days' +** +** Where NNN is an arbitrary floating-point number and "days" can be one +** of several units of time. +*/ +static const struct { + u8 eType; /* Transformation type code */ + u8 nName; /* Length of th name */ + char *zName; /* Name of the transformation */ + double rLimit; /* Maximum NNN value for this transform */ + double rXform; /* Constant used for this transform */ +} aXformType[] = { + { 0, 6, "second", 464269060800.0, 86400000.0/(24.0*60.0*60.0) }, + { 0, 6, "minute", 7737817680.0, 86400000.0/(24.0*60.0) }, + { 0, 4, "hour", 128963628.0, 86400000.0/24.0 }, + { 0, 3, "day", 5373485.0, 86400000.0 }, + { 1, 5, "month", 176546.0, 30.0*86400000.0 }, + { 2, 4, "year", 14713.0, 365.0*86400000.0 }, +}; + +/* +** Process a modifier to a date-time stamp. The modifiers are +** as follows: +** +** NNN days +** NNN hours +** NNN minutes +** NNN.NNNN seconds +** NNN months +** NNN years +** start of month +** start of year +** start of week +** start of day +** weekday N +** unixepoch +** localtime +** utc +** +** Return 0 on success and 1 if there is any kind of error. If the error +** is in a system call (i.e. localtime()), then an error message is written +** to context pCtx. If the error is an unrecognized modifier, no error is +** written to pCtx. +*/ +static int parseModifier( + sqlite3_context *pCtx, /* Function context */ + const char *z, /* The text of the modifier */ + int n, /* Length of zMod in bytes */ + DateTime *p /* The date/time value to be modified */ +){ + int rc = 1; + double r; + switch(sqlite3UpperToLower[(u8)z[0]] ){ +#ifndef SQLITE_OMIT_LOCALTIME + case 'l': { + /* localtime + ** + ** Assuming the current time value is UTC (a.k.a. GMT), shift it to + ** show local time. + */ + if( sqlite3_stricmp(z, "localtime")==0 && sqlite3NotPureFunc(pCtx) ){ + computeJD(p); + p->iJD += localtimeOffset(p, pCtx, &rc); + clearYMD_HMS_TZ(p); + } + break; + } +#endif + case 'u': { + /* + ** unixepoch + ** + ** Treat the current value of p->s as the number of + ** seconds since 1970. Convert to a real julian day number. + */ + if( sqlite3_stricmp(z, "unixepoch")==0 && p->rawS ){ + r = p->s*1000.0 + 210866760000000.0; + if( r>=0.0 && r<464269060800000.0 ){ + clearYMD_HMS_TZ(p); + p->iJD = (sqlite3_int64)r; + p->validJD = 1; + p->rawS = 0; + rc = 0; + } + } +#ifndef SQLITE_OMIT_LOCALTIME + else if( sqlite3_stricmp(z, "utc")==0 && sqlite3NotPureFunc(pCtx) ){ + if( p->tzSet==0 ){ + sqlite3_int64 c1; + computeJD(p); + c1 = localtimeOffset(p, pCtx, &rc); + if( rc==SQLITE_OK ){ + p->iJD -= c1; + clearYMD_HMS_TZ(p); + p->iJD += c1 - localtimeOffset(p, pCtx, &rc); + } + p->tzSet = 1; + }else{ + rc = SQLITE_OK; + } + } +#endif + break; + } + case 'w': { + /* + ** weekday N + ** + ** Move the date to the same time on the next occurrence of + ** weekday N where 0==Sunday, 1==Monday, and so forth. If the + ** date is already on the appropriate weekday, this is a no-op. + */ + if( sqlite3_strnicmp(z, "weekday ", 8)==0 + && sqlite3AtoF(&z[8], &r, sqlite3Strlen30(&z[8]), SQLITE_UTF8)>0 + && (n=(int)r)==r && n>=0 && r<7 ){ + sqlite3_int64 Z; + computeYMD_HMS(p); + p->validTZ = 0; + p->validJD = 0; + computeJD(p); + Z = ((p->iJD + 129600000)/86400000) % 7; + if( Z>n ) Z -= 7; + p->iJD += (n - Z)*86400000; + clearYMD_HMS_TZ(p); + rc = 0; + } + break; + } + case 's': { + /* + ** start of TTTTT + ** + ** Move the date backwards to the beginning of the current day, + ** or month or year. + */ + if( sqlite3_strnicmp(z, "start of ", 9)!=0 ) break; + if( !p->validJD && !p->validYMD && !p->validHMS ) break; + z += 9; + computeYMD(p); + p->validHMS = 1; + p->h = p->m = 0; + p->s = 0.0; + p->rawS = 0; + p->validTZ = 0; + p->validJD = 0; + if( sqlite3_stricmp(z,"month")==0 ){ + p->D = 1; + rc = 0; + }else if( sqlite3_stricmp(z,"year")==0 ){ + p->M = 1; + p->D = 1; + rc = 0; + }else if( sqlite3_stricmp(z,"day")==0 ){ + rc = 0; + } + break; + } + case '+': + case '-': + case '0': + case '1': + case '2': + case '3': + case '4': + case '5': + case '6': + case '7': + case '8': + case '9': { + double rRounder; + int i; + for(n=1; z[n] && z[n]!=':' && !sqlite3Isspace(z[n]); n++){} + if( sqlite3AtoF(z, &r, n, SQLITE_UTF8)<=0 ){ + rc = 1; + break; + } + if( z[n]==':' ){ + /* A modifier of the form (+|-)HH:MM:SS.FFF adds (or subtracts) the + ** specified number of hours, minutes, seconds, and fractional seconds + ** to the time. The ".FFF" may be omitted. The ":SS.FFF" may be + ** omitted. + */ + const char *z2 = z; + DateTime tx; + sqlite3_int64 day; + if( !sqlite3Isdigit(*z2) ) z2++; + memset(&tx, 0, sizeof(tx)); + if( parseHhMmSs(z2, &tx) ) break; + computeJD(&tx); + tx.iJD -= 43200000; + day = tx.iJD/86400000; + tx.iJD -= day*86400000; + if( z[0]=='-' ) tx.iJD = -tx.iJD; + computeJD(p); + clearYMD_HMS_TZ(p); + p->iJD += tx.iJD; + rc = 0; + break; + } + + /* If control reaches this point, it means the transformation is + ** one of the forms like "+NNN days". */ + z += n; + while( sqlite3Isspace(*z) ) z++; + n = sqlite3Strlen30(z); + if( n>10 || n<3 ) break; + if( sqlite3UpperToLower[(u8)z[n-1]]=='s' ) n--; + computeJD(p); + rc = 1; + rRounder = r<0 ? -0.5 : +0.5; + for(i=0; i-aXformType[i].rLimit && rM += (int)r; + x = p->M>0 ? (p->M-1)/12 : (p->M-12)/12; + p->Y += x; + p->M -= x*12; + p->validJD = 0; + r -= (int)r; + break; + } + case 2: { /* Special processing to add years */ + int y = (int)r; + computeYMD_HMS(p); + p->Y += y; + p->validJD = 0; + r -= (int)r; + break; + } + } + computeJD(p); + p->iJD += (sqlite3_int64)(r*aXformType[i].rXform + rRounder); + rc = 0; + break; + } + } + clearYMD_HMS_TZ(p); + break; + } + default: { + break; + } + } + return rc; +} + +/* +** Process time function arguments. argv[0] is a date-time stamp. +** argv[1] and following are modifiers. Parse them all and write +** the resulting time into the DateTime structure p. Return 0 +** on success and 1 if there are any errors. +** +** If there are zero parameters (if even argv[0] is undefined) +** then assume a default value of "now" for argv[0]. +*/ +static int isDate( + sqlite3_context *context, + int argc, + sqlite3_value **argv, + DateTime *p +){ + int i, n; + const unsigned char *z; + int eType; + memset(p, 0, sizeof(*p)); + if( argc==0 ){ + return setDateTimeToCurrent(context, p); + } + if( (eType = sqlite3_value_type(argv[0]))==SQLITE_FLOAT + || eType==SQLITE_INTEGER ){ + setRawDateNumber(p, sqlite3_value_double(argv[0])); + }else{ + z = sqlite3_value_text(argv[0]); + if( !z || parseDateOrTime(context, (char*)z, p) ){ + return 1; + } + } + for(i=1; iisError || !validJulianDay(p->iJD) ) return 1; + return 0; +} + + +/* +** The following routines implement the various date and time functions +** of SQLite. +*/ + +/* +** julianday( TIMESTRING, MOD, MOD, ...) +** +** Return the julian day number of the date specified in the arguments +*/ +static void juliandayFunc( + sqlite3_context *context, + int argc, + sqlite3_value **argv +){ + DateTime x; + if( isDate(context, argc, argv, &x)==0 ){ + computeJD(&x); + sqlite3_result_double(context, x.iJD/86400000.0); + } +} + +/* +** datetime( TIMESTRING, MOD, MOD, ...) +** +** Return YYYY-MM-DD HH:MM:SS +*/ +static void datetimeFunc( + sqlite3_context *context, + int argc, + sqlite3_value **argv +){ + DateTime x; + if( isDate(context, argc, argv, &x)==0 ){ + char zBuf[100]; + computeYMD_HMS(&x); + sqlite3_snprintf(sizeof(zBuf), zBuf, "%04d-%02d-%02d %02d:%02d:%02d", + x.Y, x.M, x.D, x.h, x.m, (int)(x.s)); + sqlite3_result_text(context, zBuf, -1, SQLITE_TRANSIENT); + } +} + +/* +** time( TIMESTRING, MOD, MOD, ...) +** +** Return HH:MM:SS +*/ +static void timeFunc( + sqlite3_context *context, + int argc, + sqlite3_value **argv +){ + DateTime x; + if( isDate(context, argc, argv, &x)==0 ){ + char zBuf[100]; + computeHMS(&x); + sqlite3_snprintf(sizeof(zBuf), zBuf, "%02d:%02d:%02d", x.h, x.m, (int)x.s); + sqlite3_result_text(context, zBuf, -1, SQLITE_TRANSIENT); + } +} + +/* +** date( TIMESTRING, MOD, MOD, ...) +** +** Return YYYY-MM-DD +*/ +static void dateFunc( + sqlite3_context *context, + int argc, + sqlite3_value **argv +){ + DateTime x; + if( isDate(context, argc, argv, &x)==0 ){ + char zBuf[100]; + computeYMD(&x); + sqlite3_snprintf(sizeof(zBuf), zBuf, "%04d-%02d-%02d", x.Y, x.M, x.D); + sqlite3_result_text(context, zBuf, -1, SQLITE_TRANSIENT); + } +} + +/* +** strftime( FORMAT, TIMESTRING, MOD, MOD, ...) +** +** Return a string described by FORMAT. Conversions as follows: +** +** %d day of month +** %f ** fractional seconds SS.SSS +** %H hour 00-24 +** %j day of year 000-366 +** %J ** julian day number +** %m month 01-12 +** %M minute 00-59 +** %s seconds since 1970-01-01 +** %S seconds 00-59 +** %w day of week 0-6 sunday==0 +** %W week of year 00-53 +** %Y year 0000-9999 +** %% % +*/ +static void strftimeFunc( + sqlite3_context *context, + int argc, + sqlite3_value **argv +){ + DateTime x; + u64 n; + size_t i,j; + char *z; + sqlite3 *db; + const char *zFmt; + char zBuf[100]; + if( argc==0 ) return; + zFmt = (const char*)sqlite3_value_text(argv[0]); + if( zFmt==0 || isDate(context, argc-1, argv+1, &x) ) return; + db = sqlite3_context_db_handle(context); + for(i=0, n=1; zFmt[i]; i++, n++){ + if( zFmt[i]=='%' ){ + switch( zFmt[i+1] ){ + case 'd': + case 'H': + case 'm': + case 'M': + case 'S': + case 'W': + n++; + /* fall thru */ + case 'w': + case '%': + break; + case 'f': + n += 8; + break; + case 'j': + n += 3; + break; + case 'Y': + n += 8; + break; + case 's': + case 'J': + n += 50; + break; + default: + return; /* ERROR. return a NULL */ + } + i++; + } + } + testcase( n==sizeof(zBuf)-1 ); + testcase( n==sizeof(zBuf) ); + testcase( n==(u64)db->aLimit[SQLITE_LIMIT_LENGTH]+1 ); + testcase( n==(u64)db->aLimit[SQLITE_LIMIT_LENGTH] ); + if( n(u64)db->aLimit[SQLITE_LIMIT_LENGTH] ){ + sqlite3_result_error_toobig(context); + return; + }else{ + z = sqlite3DbMallocRawNN(db, (int)n); + if( z==0 ){ + sqlite3_result_error_nomem(context); + return; + } + } + computeJD(&x); + computeYMD_HMS(&x); + for(i=j=0; zFmt[i]; i++){ + if( zFmt[i]!='%' ){ + z[j++] = zFmt[i]; + }else{ + i++; + switch( zFmt[i] ){ + case 'd': sqlite3_snprintf(3, &z[j],"%02d",x.D); j+=2; break; + case 'f': { + double s = x.s; + if( s>59.999 ) s = 59.999; + sqlite3_snprintf(7, &z[j],"%06.3f", s); + j += sqlite3Strlen30(&z[j]); + break; + } + case 'H': sqlite3_snprintf(3, &z[j],"%02d",x.h); j+=2; break; + case 'W': /* Fall thru */ + case 'j': { + int nDay; /* Number of days since 1st day of year */ + DateTime y = x; + y.validJD = 0; + y.M = 1; + y.D = 1; + computeJD(&y); + nDay = (int)((x.iJD-y.iJD+43200000)/86400000); + if( zFmt[i]=='W' ){ + int wd; /* 0=Monday, 1=Tuesday, ... 6=Sunday */ + wd = (int)(((x.iJD+43200000)/86400000)%7); + sqlite3_snprintf(3, &z[j],"%02d",(nDay+7-wd)/7); + j += 2; + }else{ + sqlite3_snprintf(4, &z[j],"%03d",nDay+1); + j += 3; + } + break; + } + case 'J': { + sqlite3_snprintf(20, &z[j],"%.16g",x.iJD/86400000.0); + j+=sqlite3Strlen30(&z[j]); + break; + } + case 'm': sqlite3_snprintf(3, &z[j],"%02d",x.M); j+=2; break; + case 'M': sqlite3_snprintf(3, &z[j],"%02d",x.m); j+=2; break; + case 's': { + sqlite3_snprintf(30,&z[j],"%lld", + (i64)(x.iJD/1000 - 21086676*(i64)10000)); + j += sqlite3Strlen30(&z[j]); + break; + } + case 'S': sqlite3_snprintf(3,&z[j],"%02d",(int)x.s); j+=2; break; + case 'w': { + z[j++] = (char)(((x.iJD+129600000)/86400000) % 7) + '0'; + break; + } + case 'Y': { + sqlite3_snprintf(5,&z[j],"%04d",x.Y); j+=sqlite3Strlen30(&z[j]); + break; + } + default: z[j++] = '%'; break; + } + } + } + z[j] = 0; + sqlite3_result_text(context, z, -1, + z==zBuf ? SQLITE_TRANSIENT : SQLITE_DYNAMIC); +} + +/* +** current_time() +** +** This function returns the same value as time('now'). +*/ +static void ctimeFunc( + sqlite3_context *context, + int NotUsed, + sqlite3_value **NotUsed2 +){ + UNUSED_PARAMETER2(NotUsed, NotUsed2); + timeFunc(context, 0, 0); +} + +/* +** current_date() +** +** This function returns the same value as date('now'). +*/ +static void cdateFunc( + sqlite3_context *context, + int NotUsed, + sqlite3_value **NotUsed2 +){ + UNUSED_PARAMETER2(NotUsed, NotUsed2); + dateFunc(context, 0, 0); +} + +/* +** current_timestamp() +** +** This function returns the same value as datetime('now'). +*/ +static void ctimestampFunc( + sqlite3_context *context, + int NotUsed, + sqlite3_value **NotUsed2 +){ + UNUSED_PARAMETER2(NotUsed, NotUsed2); + datetimeFunc(context, 0, 0); +} +#endif /* !defined(SQLITE_OMIT_DATETIME_FUNCS) */ + +#ifdef SQLITE_OMIT_DATETIME_FUNCS +/* +** If the library is compiled to omit the full-scale date and time +** handling (to get a smaller binary), the following minimal version +** of the functions current_time(), current_date() and current_timestamp() +** are included instead. This is to support column declarations that +** include "DEFAULT CURRENT_TIME" etc. +** +** This function uses the C-library functions time(), gmtime() +** and strftime(). The format string to pass to strftime() is supplied +** as the user-data for the function. +*/ +static void currentTimeFunc( + sqlite3_context *context, + int argc, + sqlite3_value **argv +){ + time_t t; + char *zFormat = (char *)sqlite3_user_data(context); + sqlite3_int64 iT; + struct tm *pTm; + struct tm sNow; + char zBuf[20]; + + UNUSED_PARAMETER(argc); + UNUSED_PARAMETER(argv); + + iT = sqlite3StmtCurrentTime(context); + if( iT<=0 ) return; + t = iT/1000 - 10000*(sqlite3_int64)21086676; +#if HAVE_GMTIME_R + pTm = gmtime_r(&t, &sNow); +#else + sqlite3_mutex_enter(sqlite3MutexAlloc(SQLITE_MUTEX_STATIC_MASTER)); + pTm = gmtime(&t); + if( pTm ) memcpy(&sNow, pTm, sizeof(sNow)); + sqlite3_mutex_leave(sqlite3MutexAlloc(SQLITE_MUTEX_STATIC_MASTER)); +#endif + if( pTm ){ + strftime(zBuf, 20, zFormat, &sNow); + sqlite3_result_text(context, zBuf, -1, SQLITE_TRANSIENT); + } +} +#endif + +/* +** This function registered all of the above C functions as SQL +** functions. This should be the only routine in this file with +** external linkage. +*/ +SQLITE_PRIVATE void sqlite3RegisterDateTimeFunctions(void){ + static FuncDef aDateTimeFuncs[] = { +#ifndef SQLITE_OMIT_DATETIME_FUNCS + PURE_DATE(julianday, -1, 0, 0, juliandayFunc ), + PURE_DATE(date, -1, 0, 0, dateFunc ), + PURE_DATE(time, -1, 0, 0, timeFunc ), + PURE_DATE(datetime, -1, 0, 0, datetimeFunc ), + PURE_DATE(strftime, -1, 0, 0, strftimeFunc ), + DFUNCTION(current_time, 0, 0, 0, ctimeFunc ), + DFUNCTION(current_timestamp, 0, 0, 0, ctimestampFunc), + DFUNCTION(current_date, 0, 0, 0, cdateFunc ), +#else + STR_FUNCTION(current_time, 0, "%H:%M:%S", 0, currentTimeFunc), + STR_FUNCTION(current_date, 0, "%Y-%m-%d", 0, currentTimeFunc), + STR_FUNCTION(current_timestamp, 0, "%Y-%m-%d %H:%M:%S", 0, currentTimeFunc), +#endif + }; + sqlite3InsertBuiltinFuncs(aDateTimeFuncs, ArraySize(aDateTimeFuncs)); +} + +/************** End of date.c ************************************************/ +/************** Begin file os.c **********************************************/ +/* +** 2005 November 29 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +****************************************************************************** +** +** This file contains OS interface code that is common to all +** architectures. +*/ +/* #include "sqliteInt.h" */ + +/* +** If we compile with the SQLITE_TEST macro set, then the following block +** of code will give us the ability to simulate a disk I/O error. This +** is used for testing the I/O recovery logic. +*/ +#if defined(SQLITE_TEST) +SQLITE_API int sqlite3_io_error_hit = 0; /* Total number of I/O Errors */ +SQLITE_API int sqlite3_io_error_hardhit = 0; /* Number of non-benign errors */ +SQLITE_API int sqlite3_io_error_pending = 0; /* Count down to first I/O error */ +SQLITE_API int sqlite3_io_error_persist = 0; /* True if I/O errors persist */ +SQLITE_API int sqlite3_io_error_benign = 0; /* True if errors are benign */ +SQLITE_API int sqlite3_diskfull_pending = 0; +SQLITE_API int sqlite3_diskfull = 0; +#endif /* defined(SQLITE_TEST) */ + +/* +** When testing, also keep a count of the number of open files. +*/ +#if defined(SQLITE_TEST) +SQLITE_API int sqlite3_open_file_count = 0; +#endif /* defined(SQLITE_TEST) */ + +/* +** The default SQLite sqlite3_vfs implementations do not allocate +** memory (actually, os_unix.c allocates a small amount of memory +** from within OsOpen()), but some third-party implementations may. +** So we test the effects of a malloc() failing and the sqlite3OsXXX() +** function returning SQLITE_IOERR_NOMEM using the DO_OS_MALLOC_TEST macro. +** +** The following functions are instrumented for malloc() failure +** testing: +** +** sqlite3OsRead() +** sqlite3OsWrite() +** sqlite3OsSync() +** sqlite3OsFileSize() +** sqlite3OsLock() +** sqlite3OsCheckReservedLock() +** sqlite3OsFileControl() +** sqlite3OsShmMap() +** sqlite3OsOpen() +** sqlite3OsDelete() +** sqlite3OsAccess() +** sqlite3OsFullPathname() +** +*/ +#if defined(SQLITE_TEST) +SQLITE_API int sqlite3_memdebug_vfs_oom_test = 1; + #define DO_OS_MALLOC_TEST(x) \ + if (sqlite3_memdebug_vfs_oom_test && (!x || !sqlite3JournalIsInMemory(x))) { \ + void *pTstAlloc = sqlite3Malloc(10); \ + if (!pTstAlloc) return SQLITE_IOERR_NOMEM_BKPT; \ + sqlite3_free(pTstAlloc); \ + } +#else + #define DO_OS_MALLOC_TEST(x) +#endif + +/* +** The following routines are convenience wrappers around methods +** of the sqlite3_file object. This is mostly just syntactic sugar. All +** of this would be completely automatic if SQLite were coded using +** C++ instead of plain old C. +*/ +SQLITE_PRIVATE void sqlite3OsClose(sqlite3_file *pId){ + if( pId->pMethods ){ + pId->pMethods->xClose(pId); + pId->pMethods = 0; + } +} +SQLITE_PRIVATE int sqlite3OsRead(sqlite3_file *id, void *pBuf, int amt, i64 offset){ + DO_OS_MALLOC_TEST(id); + return id->pMethods->xRead(id, pBuf, amt, offset); +} +SQLITE_PRIVATE int sqlite3OsWrite(sqlite3_file *id, const void *pBuf, int amt, i64 offset){ + DO_OS_MALLOC_TEST(id); + return id->pMethods->xWrite(id, pBuf, amt, offset); +} +SQLITE_PRIVATE int sqlite3OsTruncate(sqlite3_file *id, i64 size){ + return id->pMethods->xTruncate(id, size); +} +SQLITE_PRIVATE int sqlite3OsSync(sqlite3_file *id, int flags){ + DO_OS_MALLOC_TEST(id); + return flags ? id->pMethods->xSync(id, flags) : SQLITE_OK; +} +SQLITE_PRIVATE int sqlite3OsFileSize(sqlite3_file *id, i64 *pSize){ + DO_OS_MALLOC_TEST(id); + return id->pMethods->xFileSize(id, pSize); +} +SQLITE_PRIVATE int sqlite3OsLock(sqlite3_file *id, int lockType){ + DO_OS_MALLOC_TEST(id); + return id->pMethods->xLock(id, lockType); +} +SQLITE_PRIVATE int sqlite3OsUnlock(sqlite3_file *id, int lockType){ + return id->pMethods->xUnlock(id, lockType); +} +SQLITE_PRIVATE int sqlite3OsCheckReservedLock(sqlite3_file *id, int *pResOut){ + DO_OS_MALLOC_TEST(id); + return id->pMethods->xCheckReservedLock(id, pResOut); +} + +/* +** Use sqlite3OsFileControl() when we are doing something that might fail +** and we need to know about the failures. Use sqlite3OsFileControlHint() +** when simply tossing information over the wall to the VFS and we do not +** really care if the VFS receives and understands the information since it +** is only a hint and can be safely ignored. The sqlite3OsFileControlHint() +** routine has no return value since the return value would be meaningless. +*/ +SQLITE_PRIVATE int sqlite3OsFileControl(sqlite3_file *id, int op, void *pArg){ + if( id->pMethods==0 ) return SQLITE_NOTFOUND; +#ifdef SQLITE_TEST + if( op!=SQLITE_FCNTL_COMMIT_PHASETWO + && op!=SQLITE_FCNTL_LOCK_TIMEOUT + ){ + /* Faults are not injected into COMMIT_PHASETWO because, assuming SQLite + ** is using a regular VFS, it is called after the corresponding + ** transaction has been committed. Injecting a fault at this point + ** confuses the test scripts - the COMMIT comand returns SQLITE_NOMEM + ** but the transaction is committed anyway. + ** + ** The core must call OsFileControl() though, not OsFileControlHint(), + ** as if a custom VFS (e.g. zipvfs) returns an error here, it probably + ** means the commit really has failed and an error should be returned + ** to the user. */ + DO_OS_MALLOC_TEST(id); + } +#endif + return id->pMethods->xFileControl(id, op, pArg); +} +SQLITE_PRIVATE void sqlite3OsFileControlHint(sqlite3_file *id, int op, void *pArg){ + if( id->pMethods ) (void)id->pMethods->xFileControl(id, op, pArg); +} + +SQLITE_PRIVATE int sqlite3OsSectorSize(sqlite3_file *id){ + int (*xSectorSize)(sqlite3_file*) = id->pMethods->xSectorSize; + return (xSectorSize ? xSectorSize(id) : SQLITE_DEFAULT_SECTOR_SIZE); +} +SQLITE_PRIVATE int sqlite3OsDeviceCharacteristics(sqlite3_file *id){ + return id->pMethods->xDeviceCharacteristics(id); +} +#ifndef SQLITE_OMIT_WAL +SQLITE_PRIVATE int sqlite3OsShmLock(sqlite3_file *id, int offset, int n, int flags){ + return id->pMethods->xShmLock(id, offset, n, flags); +} +SQLITE_PRIVATE void sqlite3OsShmBarrier(sqlite3_file *id){ + id->pMethods->xShmBarrier(id); +} +SQLITE_PRIVATE int sqlite3OsShmUnmap(sqlite3_file *id, int deleteFlag){ + return id->pMethods->xShmUnmap(id, deleteFlag); +} +SQLITE_PRIVATE int sqlite3OsShmMap( + sqlite3_file *id, /* Database file handle */ + int iPage, + int pgsz, + int bExtend, /* True to extend file if necessary */ + void volatile **pp /* OUT: Pointer to mapping */ +){ + DO_OS_MALLOC_TEST(id); + return id->pMethods->xShmMap(id, iPage, pgsz, bExtend, pp); +} +#endif /* SQLITE_OMIT_WAL */ + +#if SQLITE_MAX_MMAP_SIZE>0 +/* The real implementation of xFetch and xUnfetch */ +SQLITE_PRIVATE int sqlite3OsFetch(sqlite3_file *id, i64 iOff, int iAmt, void **pp){ + DO_OS_MALLOC_TEST(id); + return id->pMethods->xFetch(id, iOff, iAmt, pp); +} +SQLITE_PRIVATE int sqlite3OsUnfetch(sqlite3_file *id, i64 iOff, void *p){ + return id->pMethods->xUnfetch(id, iOff, p); +} +#else +/* No-op stubs to use when memory-mapped I/O is disabled */ +SQLITE_PRIVATE int sqlite3OsFetch(sqlite3_file *id, i64 iOff, int iAmt, void **pp){ + *pp = 0; + return SQLITE_OK; +} +SQLITE_PRIVATE int sqlite3OsUnfetch(sqlite3_file *id, i64 iOff, void *p){ + return SQLITE_OK; +} +#endif + +/* +** The next group of routines are convenience wrappers around the +** VFS methods. +*/ +SQLITE_PRIVATE int sqlite3OsOpen( + sqlite3_vfs *pVfs, + const char *zPath, + sqlite3_file *pFile, + int flags, + int *pFlagsOut +){ + int rc; + DO_OS_MALLOC_TEST(0); + /* 0x87f7f is a mask of SQLITE_OPEN_ flags that are valid to be passed + ** down into the VFS layer. Some SQLITE_OPEN_ flags (for example, + ** SQLITE_OPEN_FULLMUTEX or SQLITE_OPEN_SHAREDCACHE) are blocked before + ** reaching the VFS. */ + rc = pVfs->xOpen(pVfs, zPath, pFile, flags & 0x87f7f, pFlagsOut); + assert( rc==SQLITE_OK || pFile->pMethods==0 ); + return rc; +} +SQLITE_PRIVATE int sqlite3OsDelete(sqlite3_vfs *pVfs, const char *zPath, int dirSync){ + DO_OS_MALLOC_TEST(0); + assert( dirSync==0 || dirSync==1 ); + return pVfs->xDelete(pVfs, zPath, dirSync); +} +SQLITE_PRIVATE int sqlite3OsAccess( + sqlite3_vfs *pVfs, + const char *zPath, + int flags, + int *pResOut +){ + DO_OS_MALLOC_TEST(0); + return pVfs->xAccess(pVfs, zPath, flags, pResOut); +} +SQLITE_PRIVATE int sqlite3OsFullPathname( + sqlite3_vfs *pVfs, + const char *zPath, + int nPathOut, + char *zPathOut +){ + DO_OS_MALLOC_TEST(0); + zPathOut[0] = 0; + return pVfs->xFullPathname(pVfs, zPath, nPathOut, zPathOut); +} +#ifndef SQLITE_OMIT_LOAD_EXTENSION +SQLITE_PRIVATE void *sqlite3OsDlOpen(sqlite3_vfs *pVfs, const char *zPath){ + return pVfs->xDlOpen(pVfs, zPath); +} +SQLITE_PRIVATE void sqlite3OsDlError(sqlite3_vfs *pVfs, int nByte, char *zBufOut){ + pVfs->xDlError(pVfs, nByte, zBufOut); +} +SQLITE_PRIVATE void (*sqlite3OsDlSym(sqlite3_vfs *pVfs, void *pHdle, const char *zSym))(void){ + return pVfs->xDlSym(pVfs, pHdle, zSym); +} +SQLITE_PRIVATE void sqlite3OsDlClose(sqlite3_vfs *pVfs, void *pHandle){ + pVfs->xDlClose(pVfs, pHandle); +} +#endif /* SQLITE_OMIT_LOAD_EXTENSION */ +SQLITE_PRIVATE int sqlite3OsRandomness(sqlite3_vfs *pVfs, int nByte, char *zBufOut){ + if( sqlite3Config.iPrngSeed ){ + memset(zBufOut, 0, nByte); + if( ALWAYS(nByte>(signed)sizeof(unsigned)) ) nByte = sizeof(unsigned int); + memcpy(zBufOut, &sqlite3Config.iPrngSeed, nByte); + return SQLITE_OK; + }else{ + return pVfs->xRandomness(pVfs, nByte, zBufOut); + } + +} +SQLITE_PRIVATE int sqlite3OsSleep(sqlite3_vfs *pVfs, int nMicro){ + return pVfs->xSleep(pVfs, nMicro); +} +SQLITE_PRIVATE int sqlite3OsGetLastError(sqlite3_vfs *pVfs){ + return pVfs->xGetLastError ? pVfs->xGetLastError(pVfs, 0, 0) : 0; +} +SQLITE_PRIVATE int sqlite3OsCurrentTimeInt64(sqlite3_vfs *pVfs, sqlite3_int64 *pTimeOut){ + int rc; + /* IMPLEMENTATION-OF: R-49045-42493 SQLite will use the xCurrentTimeInt64() + ** method to get the current date and time if that method is available + ** (if iVersion is 2 or greater and the function pointer is not NULL) and + ** will fall back to xCurrentTime() if xCurrentTimeInt64() is + ** unavailable. + */ + if( pVfs->iVersion>=2 && pVfs->xCurrentTimeInt64 ){ + rc = pVfs->xCurrentTimeInt64(pVfs, pTimeOut); + }else{ + double r; + rc = pVfs->xCurrentTime(pVfs, &r); + *pTimeOut = (sqlite3_int64)(r*86400000.0); + } + return rc; +} + +SQLITE_PRIVATE int sqlite3OsOpenMalloc( + sqlite3_vfs *pVfs, + const char *zFile, + sqlite3_file **ppFile, + int flags, + int *pOutFlags +){ + int rc; + sqlite3_file *pFile; + pFile = (sqlite3_file *)sqlite3MallocZero(pVfs->szOsFile); + if( pFile ){ + rc = sqlite3OsOpen(pVfs, zFile, pFile, flags, pOutFlags); + if( rc!=SQLITE_OK ){ + sqlite3_free(pFile); + }else{ + *ppFile = pFile; + } + }else{ + rc = SQLITE_NOMEM_BKPT; + } + return rc; +} +SQLITE_PRIVATE void sqlite3OsCloseFree(sqlite3_file *pFile){ + assert( pFile ); + sqlite3OsClose(pFile); + sqlite3_free(pFile); +} + +/* +** This function is a wrapper around the OS specific implementation of +** sqlite3_os_init(). The purpose of the wrapper is to provide the +** ability to simulate a malloc failure, so that the handling of an +** error in sqlite3_os_init() by the upper layers can be tested. +*/ +SQLITE_PRIVATE int sqlite3OsInit(void){ + void *p = sqlite3_malloc(10); + if( p==0 ) return SQLITE_NOMEM_BKPT; + sqlite3_free(p); + return sqlite3_os_init(); +} + +/* +** The list of all registered VFS implementations. +*/ +static sqlite3_vfs * SQLITE_WSD vfsList = 0; +#define vfsList GLOBAL(sqlite3_vfs *, vfsList) + +/* +** Locate a VFS by name. If no name is given, simply return the +** first VFS on the list. +*/ +SQLITE_API sqlite3_vfs *sqlite3_vfs_find(const char *zVfs){ + sqlite3_vfs *pVfs = 0; +#if SQLITE_THREADSAFE + sqlite3_mutex *mutex; +#endif +#ifndef SQLITE_OMIT_AUTOINIT + int rc = sqlite3_initialize(); + if( rc ) return 0; +#endif +#if SQLITE_THREADSAFE + mutex = sqlite3MutexAlloc(SQLITE_MUTEX_STATIC_MASTER); +#endif + sqlite3_mutex_enter(mutex); + for(pVfs = vfsList; pVfs; pVfs=pVfs->pNext){ + if( zVfs==0 ) break; + if( strcmp(zVfs, pVfs->zName)==0 ) break; + } + sqlite3_mutex_leave(mutex); + return pVfs; +} + +/* +** Unlink a VFS from the linked list +*/ +static void vfsUnlink(sqlite3_vfs *pVfs){ + assert( sqlite3_mutex_held(sqlite3MutexAlloc(SQLITE_MUTEX_STATIC_MASTER)) ); + if( pVfs==0 ){ + /* No-op */ + }else if( vfsList==pVfs ){ + vfsList = pVfs->pNext; + }else if( vfsList ){ + sqlite3_vfs *p = vfsList; + while( p->pNext && p->pNext!=pVfs ){ + p = p->pNext; + } + if( p->pNext==pVfs ){ + p->pNext = pVfs->pNext; + } + } +} + +/* +** Register a VFS with the system. It is harmless to register the same +** VFS multiple times. The new VFS becomes the default if makeDflt is +** true. +*/ +SQLITE_API int sqlite3_vfs_register(sqlite3_vfs *pVfs, int makeDflt){ + MUTEX_LOGIC(sqlite3_mutex *mutex;) +#ifndef SQLITE_OMIT_AUTOINIT + int rc = sqlite3_initialize(); + if( rc ) return rc; +#endif +#ifdef SQLITE_ENABLE_API_ARMOR + if( pVfs==0 ) return SQLITE_MISUSE_BKPT; +#endif + + MUTEX_LOGIC( mutex = sqlite3MutexAlloc(SQLITE_MUTEX_STATIC_MASTER); ) + sqlite3_mutex_enter(mutex); + vfsUnlink(pVfs); + if( makeDflt || vfsList==0 ){ + pVfs->pNext = vfsList; + vfsList = pVfs; + }else{ + pVfs->pNext = vfsList->pNext; + vfsList->pNext = pVfs; + } + assert(vfsList); + sqlite3_mutex_leave(mutex); + return SQLITE_OK; +} + +/* +** Unregister a VFS so that it is no longer accessible. +*/ +SQLITE_API int sqlite3_vfs_unregister(sqlite3_vfs *pVfs){ + MUTEX_LOGIC(sqlite3_mutex *mutex;) +#ifndef SQLITE_OMIT_AUTOINIT + int rc = sqlite3_initialize(); + if( rc ) return rc; +#endif + MUTEX_LOGIC( mutex = sqlite3MutexAlloc(SQLITE_MUTEX_STATIC_MASTER); ) + sqlite3_mutex_enter(mutex); + vfsUnlink(pVfs); + sqlite3_mutex_leave(mutex); + return SQLITE_OK; +} + +/************** End of os.c **************************************************/ +/************** Begin file fault.c *******************************************/ +/* +** 2008 Jan 22 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** +** This file contains code to support the concept of "benign" +** malloc failures (when the xMalloc() or xRealloc() method of the +** sqlite3_mem_methods structure fails to allocate a block of memory +** and returns 0). +** +** Most malloc failures are non-benign. After they occur, SQLite +** abandons the current operation and returns an error code (usually +** SQLITE_NOMEM) to the user. However, sometimes a fault is not necessarily +** fatal. For example, if a malloc fails while resizing a hash table, this +** is completely recoverable simply by not carrying out the resize. The +** hash table will continue to function normally. So a malloc failure +** during a hash table resize is a benign fault. +*/ + +/* #include "sqliteInt.h" */ + +#ifndef SQLITE_UNTESTABLE + +/* +** Global variables. +*/ +typedef struct BenignMallocHooks BenignMallocHooks; +static SQLITE_WSD struct BenignMallocHooks { + void (*xBenignBegin)(void); + void (*xBenignEnd)(void); +} sqlite3Hooks = { 0, 0 }; + +/* The "wsdHooks" macro will resolve to the appropriate BenignMallocHooks +** structure. If writable static data is unsupported on the target, +** we have to locate the state vector at run-time. In the more common +** case where writable static data is supported, wsdHooks can refer directly +** to the "sqlite3Hooks" state vector declared above. +*/ +#ifdef SQLITE_OMIT_WSD +# define wsdHooksInit \ + BenignMallocHooks *x = &GLOBAL(BenignMallocHooks,sqlite3Hooks) +# define wsdHooks x[0] +#else +# define wsdHooksInit +# define wsdHooks sqlite3Hooks +#endif + + +/* +** Register hooks to call when sqlite3BeginBenignMalloc() and +** sqlite3EndBenignMalloc() are called, respectively. +*/ +SQLITE_PRIVATE void sqlite3BenignMallocHooks( + void (*xBenignBegin)(void), + void (*xBenignEnd)(void) +){ + wsdHooksInit; + wsdHooks.xBenignBegin = xBenignBegin; + wsdHooks.xBenignEnd = xBenignEnd; +} + +/* +** This (sqlite3EndBenignMalloc()) is called by SQLite code to indicate that +** subsequent malloc failures are benign. A call to sqlite3EndBenignMalloc() +** indicates that subsequent malloc failures are non-benign. +*/ +SQLITE_PRIVATE void sqlite3BeginBenignMalloc(void){ + wsdHooksInit; + if( wsdHooks.xBenignBegin ){ + wsdHooks.xBenignBegin(); + } +} +SQLITE_PRIVATE void sqlite3EndBenignMalloc(void){ + wsdHooksInit; + if( wsdHooks.xBenignEnd ){ + wsdHooks.xBenignEnd(); + } +} + +#endif /* #ifndef SQLITE_UNTESTABLE */ + +/************** End of fault.c ***********************************************/ +/************** Begin file mem0.c ********************************************/ +/* +** 2008 October 28 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** +** This file contains a no-op memory allocation drivers for use when +** SQLITE_ZERO_MALLOC is defined. The allocation drivers implemented +** here always fail. SQLite will not operate with these drivers. These +** are merely placeholders. Real drivers must be substituted using +** sqlite3_config() before SQLite will operate. +*/ +/* #include "sqliteInt.h" */ + +/* +** This version of the memory allocator is the default. It is +** used when no other memory allocator is specified using compile-time +** macros. +*/ +#ifdef SQLITE_ZERO_MALLOC + +/* +** No-op versions of all memory allocation routines +*/ +static void *sqlite3MemMalloc(int nByte){ return 0; } +static void sqlite3MemFree(void *pPrior){ return; } +static void *sqlite3MemRealloc(void *pPrior, int nByte){ return 0; } +static int sqlite3MemSize(void *pPrior){ return 0; } +static int sqlite3MemRoundup(int n){ return n; } +static int sqlite3MemInit(void *NotUsed){ return SQLITE_OK; } +static void sqlite3MemShutdown(void *NotUsed){ return; } + +/* +** This routine is the only routine in this file with external linkage. +** +** Populate the low-level memory allocation function pointers in +** sqlite3GlobalConfig.m with pointers to the routines in this file. +*/ +SQLITE_PRIVATE void sqlite3MemSetDefault(void){ + static const sqlite3_mem_methods defaultMethods = { + sqlite3MemMalloc, + sqlite3MemFree, + sqlite3MemRealloc, + sqlite3MemSize, + sqlite3MemRoundup, + sqlite3MemInit, + sqlite3MemShutdown, + 0 + }; + sqlite3_config(SQLITE_CONFIG_MALLOC, &defaultMethods); +} + +#endif /* SQLITE_ZERO_MALLOC */ + +/************** End of mem0.c ************************************************/ +/************** Begin file mem1.c ********************************************/ +/* +** 2007 August 14 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** +** This file contains low-level memory allocation drivers for when +** SQLite will use the standard C-library malloc/realloc/free interface +** to obtain the memory it needs. +** +** This file contains implementations of the low-level memory allocation +** routines specified in the sqlite3_mem_methods object. The content of +** this file is only used if SQLITE_SYSTEM_MALLOC is defined. The +** SQLITE_SYSTEM_MALLOC macro is defined automatically if neither the +** SQLITE_MEMDEBUG nor the SQLITE_WIN32_MALLOC macros are defined. The +** default configuration is to use memory allocation routines in this +** file. +** +** C-preprocessor macro summary: +** +** HAVE_MALLOC_USABLE_SIZE The configure script sets this symbol if +** the malloc_usable_size() interface exists +** on the target platform. Or, this symbol +** can be set manually, if desired. +** If an equivalent interface exists by +** a different name, using a separate -D +** option to rename it. +** +** SQLITE_WITHOUT_ZONEMALLOC Some older macs lack support for the zone +** memory allocator. Set this symbol to enable +** building on older macs. +** +** SQLITE_WITHOUT_MSIZE Set this symbol to disable the use of +** _msize() on windows systems. This might +** be necessary when compiling for Delphi, +** for example. +*/ +/* #include "sqliteInt.h" */ + +/* +** This version of the memory allocator is the default. It is +** used when no other memory allocator is specified using compile-time +** macros. +*/ +#ifdef SQLITE_SYSTEM_MALLOC +#if defined(__APPLE__) && !defined(SQLITE_WITHOUT_ZONEMALLOC) + +/* +** Use the zone allocator available on apple products unless the +** SQLITE_WITHOUT_ZONEMALLOC symbol is defined. +*/ +#include +#include +#ifdef SQLITE_MIGHT_BE_SINGLE_CORE +#include +#endif /* SQLITE_MIGHT_BE_SINGLE_CORE */ +static malloc_zone_t* _sqliteZone_; +#define SQLITE_MALLOC(x) malloc_zone_malloc(_sqliteZone_, (x)) +#define SQLITE_FREE(x) malloc_zone_free(_sqliteZone_, (x)); +#define SQLITE_REALLOC(x,y) malloc_zone_realloc(_sqliteZone_, (x), (y)) +#define SQLITE_MALLOCSIZE(x) \ + (_sqliteZone_ ? _sqliteZone_->size(_sqliteZone_,x) : malloc_size(x)) + +#else /* if not __APPLE__ */ + +/* +** Use standard C library malloc and free on non-Apple systems. +** Also used by Apple systems if SQLITE_WITHOUT_ZONEMALLOC is defined. +*/ +#define SQLITE_MALLOC(x) malloc(x) +#define SQLITE_FREE(x) free(x) +#define SQLITE_REALLOC(x,y) realloc((x),(y)) + +/* +** The malloc.h header file is needed for malloc_usable_size() function +** on some systems (e.g. Linux). +*/ +#if HAVE_MALLOC_H && HAVE_MALLOC_USABLE_SIZE +# define SQLITE_USE_MALLOC_H 1 +# define SQLITE_USE_MALLOC_USABLE_SIZE 1 +/* +** The MSVCRT has malloc_usable_size(), but it is called _msize(). The +** use of _msize() is automatic, but can be disabled by compiling with +** -DSQLITE_WITHOUT_MSIZE. Using the _msize() function also requires +** the malloc.h header file. +*/ +#elif defined(_MSC_VER) && !defined(SQLITE_WITHOUT_MSIZE) +# define SQLITE_USE_MALLOC_H +# define SQLITE_USE_MSIZE +#endif + +/* +** Include the malloc.h header file, if necessary. Also set define macro +** SQLITE_MALLOCSIZE to the appropriate function name, which is _msize() +** for MSVC and malloc_usable_size() for most other systems (e.g. Linux). +** The memory size function can always be overridden manually by defining +** the macro SQLITE_MALLOCSIZE to the desired function name. +*/ +#if defined(SQLITE_USE_MALLOC_H) +# include +# if defined(SQLITE_USE_MALLOC_USABLE_SIZE) +# if !defined(SQLITE_MALLOCSIZE) +# define SQLITE_MALLOCSIZE(x) malloc_usable_size(x) +# endif +# elif defined(SQLITE_USE_MSIZE) +# if !defined(SQLITE_MALLOCSIZE) +# define SQLITE_MALLOCSIZE _msize +# endif +# endif +#endif /* defined(SQLITE_USE_MALLOC_H) */ + +#endif /* __APPLE__ or not __APPLE__ */ + +/* +** Like malloc(), but remember the size of the allocation +** so that we can find it later using sqlite3MemSize(). +** +** For this low-level routine, we are guaranteed that nByte>0 because +** cases of nByte<=0 will be intercepted and dealt with by higher level +** routines. +*/ +static void *sqlite3MemMalloc(int nByte){ +#ifdef SQLITE_MALLOCSIZE + void *p; + testcase( ROUND8(nByte)==nByte ); + p = SQLITE_MALLOC( nByte ); + if( p==0 ){ + testcase( sqlite3GlobalConfig.xLog!=0 ); + sqlite3_log(SQLITE_NOMEM, "failed to allocate %u bytes of memory", nByte); + } + return p; +#else + sqlite3_int64 *p; + assert( nByte>0 ); + testcase( ROUND8(nByte)!=nByte ); + p = SQLITE_MALLOC( nByte+8 ); + if( p ){ + p[0] = nByte; + p++; + }else{ + testcase( sqlite3GlobalConfig.xLog!=0 ); + sqlite3_log(SQLITE_NOMEM, "failed to allocate %u bytes of memory", nByte); + } + return (void *)p; +#endif +} + +/* +** Like free() but works for allocations obtained from sqlite3MemMalloc() +** or sqlite3MemRealloc(). +** +** For this low-level routine, we already know that pPrior!=0 since +** cases where pPrior==0 will have been intecepted and dealt with +** by higher-level routines. +*/ +static void sqlite3MemFree(void *pPrior){ +#ifdef SQLITE_MALLOCSIZE + SQLITE_FREE(pPrior); +#else + sqlite3_int64 *p = (sqlite3_int64*)pPrior; + assert( pPrior!=0 ); + p--; + SQLITE_FREE(p); +#endif +} + +/* +** Report the allocated size of a prior return from xMalloc() +** or xRealloc(). +*/ +static int sqlite3MemSize(void *pPrior){ +#ifdef SQLITE_MALLOCSIZE + assert( pPrior!=0 ); + return (int)SQLITE_MALLOCSIZE(pPrior); +#else + sqlite3_int64 *p; + assert( pPrior!=0 ); + p = (sqlite3_int64*)pPrior; + p--; + return (int)p[0]; +#endif +} + +/* +** Like realloc(). Resize an allocation previously obtained from +** sqlite3MemMalloc(). +** +** For this low-level interface, we know that pPrior!=0. Cases where +** pPrior==0 while have been intercepted by higher-level routine and +** redirected to xMalloc. Similarly, we know that nByte>0 because +** cases where nByte<=0 will have been intercepted by higher-level +** routines and redirected to xFree. +*/ +static void *sqlite3MemRealloc(void *pPrior, int nByte){ +#ifdef SQLITE_MALLOCSIZE + void *p = SQLITE_REALLOC(pPrior, nByte); + if( p==0 ){ + testcase( sqlite3GlobalConfig.xLog!=0 ); + sqlite3_log(SQLITE_NOMEM, + "failed memory resize %u to %u bytes", + SQLITE_MALLOCSIZE(pPrior), nByte); + } + return p; +#else + sqlite3_int64 *p = (sqlite3_int64*)pPrior; + assert( pPrior!=0 && nByte>0 ); + assert( nByte==ROUND8(nByte) ); /* EV: R-46199-30249 */ + p--; + p = SQLITE_REALLOC(p, nByte+8 ); + if( p ){ + p[0] = nByte; + p++; + }else{ + testcase( sqlite3GlobalConfig.xLog!=0 ); + sqlite3_log(SQLITE_NOMEM, + "failed memory resize %u to %u bytes", + sqlite3MemSize(pPrior), nByte); + } + return (void*)p; +#endif +} + +/* +** Round up a request size to the next valid allocation size. +*/ +static int sqlite3MemRoundup(int n){ + return ROUND8(n); +} + +/* +** Initialize this module. +*/ +static int sqlite3MemInit(void *NotUsed){ +#if defined(__APPLE__) && !defined(SQLITE_WITHOUT_ZONEMALLOC) + int cpuCount; + size_t len; + if( _sqliteZone_ ){ + return SQLITE_OK; + } + len = sizeof(cpuCount); + /* One usually wants to use hw.acctivecpu for MT decisions, but not here */ + sysctlbyname("hw.ncpu", &cpuCount, &len, NULL, 0); + if( cpuCount>1 ){ + /* defer MT decisions to system malloc */ + _sqliteZone_ = malloc_default_zone(); + }else{ + /* only 1 core, use our own zone to contention over global locks, + ** e.g. we have our own dedicated locks */ + _sqliteZone_ = malloc_create_zone(4096, 0); + malloc_set_zone_name(_sqliteZone_, "Sqlite_Heap"); + } +#endif /* defined(__APPLE__) && !defined(SQLITE_WITHOUT_ZONEMALLOC) */ + UNUSED_PARAMETER(NotUsed); + return SQLITE_OK; +} + +/* +** Deinitialize this module. +*/ +static void sqlite3MemShutdown(void *NotUsed){ + UNUSED_PARAMETER(NotUsed); + return; +} + +/* +** This routine is the only routine in this file with external linkage. +** +** Populate the low-level memory allocation function pointers in +** sqlite3GlobalConfig.m with pointers to the routines in this file. +*/ +SQLITE_PRIVATE void sqlite3MemSetDefault(void){ + static const sqlite3_mem_methods defaultMethods = { + sqlite3MemMalloc, + sqlite3MemFree, + sqlite3MemRealloc, + sqlite3MemSize, + sqlite3MemRoundup, + sqlite3MemInit, + sqlite3MemShutdown, + 0 + }; + sqlite3_config(SQLITE_CONFIG_MALLOC, &defaultMethods); +} + +#endif /* SQLITE_SYSTEM_MALLOC */ + +/************** End of mem1.c ************************************************/ +/************** Begin file mem2.c ********************************************/ +/* +** 2007 August 15 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** +** This file contains low-level memory allocation drivers for when +** SQLite will use the standard C-library malloc/realloc/free interface +** to obtain the memory it needs while adding lots of additional debugging +** information to each allocation in order to help detect and fix memory +** leaks and memory usage errors. +** +** This file contains implementations of the low-level memory allocation +** routines specified in the sqlite3_mem_methods object. +*/ +/* #include "sqliteInt.h" */ + +/* +** This version of the memory allocator is used only if the +** SQLITE_MEMDEBUG macro is defined +*/ +#ifdef SQLITE_MEMDEBUG + +/* +** The backtrace functionality is only available with GLIBC +*/ +#ifdef __GLIBC__ + extern int backtrace(void**,int); + extern void backtrace_symbols_fd(void*const*,int,int); +#else +# define backtrace(A,B) 1 +# define backtrace_symbols_fd(A,B,C) +#endif +/* #include */ + +/* +** Each memory allocation looks like this: +** +** ------------------------------------------------------------------------ +** | Title | backtrace pointers | MemBlockHdr | allocation | EndGuard | +** ------------------------------------------------------------------------ +** +** The application code sees only a pointer to the allocation. We have +** to back up from the allocation pointer to find the MemBlockHdr. The +** MemBlockHdr tells us the size of the allocation and the number of +** backtrace pointers. There is also a guard word at the end of the +** MemBlockHdr. +*/ +struct MemBlockHdr { + i64 iSize; /* Size of this allocation */ + struct MemBlockHdr *pNext, *pPrev; /* Linked list of all unfreed memory */ + char nBacktrace; /* Number of backtraces on this alloc */ + char nBacktraceSlots; /* Available backtrace slots */ + u8 nTitle; /* Bytes of title; includes '\0' */ + u8 eType; /* Allocation type code */ + int iForeGuard; /* Guard word for sanity */ +}; + +/* +** Guard words +*/ +#define FOREGUARD 0x80F5E153 +#define REARGUARD 0xE4676B53 + +/* +** Number of malloc size increments to track. +*/ +#define NCSIZE 1000 + +/* +** All of the static variables used by this module are collected +** into a single structure named "mem". This is to keep the +** static variables organized and to reduce namespace pollution +** when this module is combined with other in the amalgamation. +*/ +static struct { + + /* + ** Mutex to control access to the memory allocation subsystem. + */ + sqlite3_mutex *mutex; + + /* + ** Head and tail of a linked list of all outstanding allocations + */ + struct MemBlockHdr *pFirst; + struct MemBlockHdr *pLast; + + /* + ** The number of levels of backtrace to save in new allocations. + */ + int nBacktrace; + void (*xBacktrace)(int, int, void **); + + /* + ** Title text to insert in front of each block + */ + int nTitle; /* Bytes of zTitle to save. Includes '\0' and padding */ + char zTitle[100]; /* The title text */ + + /* + ** sqlite3MallocDisallow() increments the following counter. + ** sqlite3MallocAllow() decrements it. + */ + int disallow; /* Do not allow memory allocation */ + + /* + ** Gather statistics on the sizes of memory allocations. + ** nAlloc[i] is the number of allocation attempts of i*8 + ** bytes. i==NCSIZE is the number of allocation attempts for + ** sizes more than NCSIZE*8 bytes. + */ + int nAlloc[NCSIZE]; /* Total number of allocations */ + int nCurrent[NCSIZE]; /* Current number of allocations */ + int mxCurrent[NCSIZE]; /* Highwater mark for nCurrent */ + +} mem; + + +/* +** Adjust memory usage statistics +*/ +static void adjustStats(int iSize, int increment){ + int i = ROUND8(iSize)/8; + if( i>NCSIZE-1 ){ + i = NCSIZE - 1; + } + if( increment>0 ){ + mem.nAlloc[i]++; + mem.nCurrent[i]++; + if( mem.nCurrent[i]>mem.mxCurrent[i] ){ + mem.mxCurrent[i] = mem.nCurrent[i]; + } + }else{ + mem.nCurrent[i]--; + assert( mem.nCurrent[i]>=0 ); + } +} + +/* +** Given an allocation, find the MemBlockHdr for that allocation. +** +** This routine checks the guards at either end of the allocation and +** if they are incorrect it asserts. +*/ +static struct MemBlockHdr *sqlite3MemsysGetHeader(void *pAllocation){ + struct MemBlockHdr *p; + int *pInt; + u8 *pU8; + int nReserve; + + p = (struct MemBlockHdr*)pAllocation; + p--; + assert( p->iForeGuard==(int)FOREGUARD ); + nReserve = ROUND8(p->iSize); + pInt = (int*)pAllocation; + pU8 = (u8*)pAllocation; + assert( pInt[nReserve/sizeof(int)]==(int)REARGUARD ); + /* This checks any of the "extra" bytes allocated due + ** to rounding up to an 8 byte boundary to ensure + ** they haven't been overwritten. + */ + while( nReserve-- > p->iSize ) assert( pU8[nReserve]==0x65 ); + return p; +} + +/* +** Return the number of bytes currently allocated at address p. +*/ +static int sqlite3MemSize(void *p){ + struct MemBlockHdr *pHdr; + if( !p ){ + return 0; + } + pHdr = sqlite3MemsysGetHeader(p); + return (int)pHdr->iSize; +} + +/* +** Initialize the memory allocation subsystem. +*/ +static int sqlite3MemInit(void *NotUsed){ + UNUSED_PARAMETER(NotUsed); + assert( (sizeof(struct MemBlockHdr)&7) == 0 ); + if( !sqlite3GlobalConfig.bMemstat ){ + /* If memory status is enabled, then the malloc.c wrapper will already + ** hold the STATIC_MEM mutex when the routines here are invoked. */ + mem.mutex = sqlite3MutexAlloc(SQLITE_MUTEX_STATIC_MEM); + } + return SQLITE_OK; +} + +/* +** Deinitialize the memory allocation subsystem. +*/ +static void sqlite3MemShutdown(void *NotUsed){ + UNUSED_PARAMETER(NotUsed); + mem.mutex = 0; +} + +/* +** Round up a request size to the next valid allocation size. +*/ +static int sqlite3MemRoundup(int n){ + return ROUND8(n); +} + +/* +** Fill a buffer with pseudo-random bytes. This is used to preset +** the content of a new memory allocation to unpredictable values and +** to clear the content of a freed allocation to unpredictable values. +*/ +static void randomFill(char *pBuf, int nByte){ + unsigned int x, y, r; + x = SQLITE_PTR_TO_INT(pBuf); + y = nByte | 1; + while( nByte >= 4 ){ + x = (x>>1) ^ (-(int)(x&1) & 0xd0000001); + y = y*1103515245 + 12345; + r = x ^ y; + *(int*)pBuf = r; + pBuf += 4; + nByte -= 4; + } + while( nByte-- > 0 ){ + x = (x>>1) ^ (-(int)(x&1) & 0xd0000001); + y = y*1103515245 + 12345; + r = x ^ y; + *(pBuf++) = r & 0xff; + } +} + +/* +** Allocate nByte bytes of memory. +*/ +static void *sqlite3MemMalloc(int nByte){ + struct MemBlockHdr *pHdr; + void **pBt; + char *z; + int *pInt; + void *p = 0; + int totalSize; + int nReserve; + sqlite3_mutex_enter(mem.mutex); + assert( mem.disallow==0 ); + nReserve = ROUND8(nByte); + totalSize = nReserve + sizeof(*pHdr) + sizeof(int) + + mem.nBacktrace*sizeof(void*) + mem.nTitle; + p = malloc(totalSize); + if( p ){ + z = p; + pBt = (void**)&z[mem.nTitle]; + pHdr = (struct MemBlockHdr*)&pBt[mem.nBacktrace]; + pHdr->pNext = 0; + pHdr->pPrev = mem.pLast; + if( mem.pLast ){ + mem.pLast->pNext = pHdr; + }else{ + mem.pFirst = pHdr; + } + mem.pLast = pHdr; + pHdr->iForeGuard = FOREGUARD; + pHdr->eType = MEMTYPE_HEAP; + pHdr->nBacktraceSlots = mem.nBacktrace; + pHdr->nTitle = mem.nTitle; + if( mem.nBacktrace ){ + void *aAddr[40]; + pHdr->nBacktrace = backtrace(aAddr, mem.nBacktrace+1)-1; + memcpy(pBt, &aAddr[1], pHdr->nBacktrace*sizeof(void*)); + assert(pBt[0]); + if( mem.xBacktrace ){ + mem.xBacktrace(nByte, pHdr->nBacktrace-1, &aAddr[1]); + } + }else{ + pHdr->nBacktrace = 0; + } + if( mem.nTitle ){ + memcpy(z, mem.zTitle, mem.nTitle); + } + pHdr->iSize = nByte; + adjustStats(nByte, +1); + pInt = (int*)&pHdr[1]; + pInt[nReserve/sizeof(int)] = REARGUARD; + randomFill((char*)pInt, nByte); + memset(((char*)pInt)+nByte, 0x65, nReserve-nByte); + p = (void*)pInt; + } + sqlite3_mutex_leave(mem.mutex); + return p; +} + +/* +** Free memory. +*/ +static void sqlite3MemFree(void *pPrior){ + struct MemBlockHdr *pHdr; + void **pBt; + char *z; + assert( sqlite3GlobalConfig.bMemstat || sqlite3GlobalConfig.bCoreMutex==0 + || mem.mutex!=0 ); + pHdr = sqlite3MemsysGetHeader(pPrior); + pBt = (void**)pHdr; + pBt -= pHdr->nBacktraceSlots; + sqlite3_mutex_enter(mem.mutex); + if( pHdr->pPrev ){ + assert( pHdr->pPrev->pNext==pHdr ); + pHdr->pPrev->pNext = pHdr->pNext; + }else{ + assert( mem.pFirst==pHdr ); + mem.pFirst = pHdr->pNext; + } + if( pHdr->pNext ){ + assert( pHdr->pNext->pPrev==pHdr ); + pHdr->pNext->pPrev = pHdr->pPrev; + }else{ + assert( mem.pLast==pHdr ); + mem.pLast = pHdr->pPrev; + } + z = (char*)pBt; + z -= pHdr->nTitle; + adjustStats((int)pHdr->iSize, -1); + randomFill(z, sizeof(void*)*pHdr->nBacktraceSlots + sizeof(*pHdr) + + (int)pHdr->iSize + sizeof(int) + pHdr->nTitle); + free(z); + sqlite3_mutex_leave(mem.mutex); +} + +/* +** Change the size of an existing memory allocation. +** +** For this debugging implementation, we *always* make a copy of the +** allocation into a new place in memory. In this way, if the +** higher level code is using pointer to the old allocation, it is +** much more likely to break and we are much more liking to find +** the error. +*/ +static void *sqlite3MemRealloc(void *pPrior, int nByte){ + struct MemBlockHdr *pOldHdr; + void *pNew; + assert( mem.disallow==0 ); + assert( (nByte & 7)==0 ); /* EV: R-46199-30249 */ + pOldHdr = sqlite3MemsysGetHeader(pPrior); + pNew = sqlite3MemMalloc(nByte); + if( pNew ){ + memcpy(pNew, pPrior, (int)(nByteiSize ? nByte : pOldHdr->iSize)); + if( nByte>pOldHdr->iSize ){ + randomFill(&((char*)pNew)[pOldHdr->iSize], nByte - (int)pOldHdr->iSize); + } + sqlite3MemFree(pPrior); + } + return pNew; +} + +/* +** Populate the low-level memory allocation function pointers in +** sqlite3GlobalConfig.m with pointers to the routines in this file. +*/ +SQLITE_PRIVATE void sqlite3MemSetDefault(void){ + static const sqlite3_mem_methods defaultMethods = { + sqlite3MemMalloc, + sqlite3MemFree, + sqlite3MemRealloc, + sqlite3MemSize, + sqlite3MemRoundup, + sqlite3MemInit, + sqlite3MemShutdown, + 0 + }; + sqlite3_config(SQLITE_CONFIG_MALLOC, &defaultMethods); +} + +/* +** Set the "type" of an allocation. +*/ +SQLITE_PRIVATE void sqlite3MemdebugSetType(void *p, u8 eType){ + if( p && sqlite3GlobalConfig.m.xMalloc==sqlite3MemMalloc ){ + struct MemBlockHdr *pHdr; + pHdr = sqlite3MemsysGetHeader(p); + assert( pHdr->iForeGuard==FOREGUARD ); + pHdr->eType = eType; + } +} + +/* +** Return TRUE if the mask of type in eType matches the type of the +** allocation p. Also return true if p==NULL. +** +** This routine is designed for use within an assert() statement, to +** verify the type of an allocation. For example: +** +** assert( sqlite3MemdebugHasType(p, MEMTYPE_HEAP) ); +*/ +SQLITE_PRIVATE int sqlite3MemdebugHasType(void *p, u8 eType){ + int rc = 1; + if( p && sqlite3GlobalConfig.m.xMalloc==sqlite3MemMalloc ){ + struct MemBlockHdr *pHdr; + pHdr = sqlite3MemsysGetHeader(p); + assert( pHdr->iForeGuard==FOREGUARD ); /* Allocation is valid */ + if( (pHdr->eType&eType)==0 ){ + rc = 0; + } + } + return rc; +} + +/* +** Return TRUE if the mask of type in eType matches no bits of the type of the +** allocation p. Also return true if p==NULL. +** +** This routine is designed for use within an assert() statement, to +** verify the type of an allocation. For example: +** +** assert( sqlite3MemdebugNoType(p, MEMTYPE_LOOKASIDE) ); +*/ +SQLITE_PRIVATE int sqlite3MemdebugNoType(void *p, u8 eType){ + int rc = 1; + if( p && sqlite3GlobalConfig.m.xMalloc==sqlite3MemMalloc ){ + struct MemBlockHdr *pHdr; + pHdr = sqlite3MemsysGetHeader(p); + assert( pHdr->iForeGuard==FOREGUARD ); /* Allocation is valid */ + if( (pHdr->eType&eType)!=0 ){ + rc = 0; + } + } + return rc; +} + +/* +** Set the number of backtrace levels kept for each allocation. +** A value of zero turns off backtracing. The number is always rounded +** up to a multiple of 2. +*/ +SQLITE_PRIVATE void sqlite3MemdebugBacktrace(int depth){ + if( depth<0 ){ depth = 0; } + if( depth>20 ){ depth = 20; } + depth = (depth+1)&0xfe; + mem.nBacktrace = depth; +} + +SQLITE_PRIVATE void sqlite3MemdebugBacktraceCallback(void (*xBacktrace)(int, int, void **)){ + mem.xBacktrace = xBacktrace; +} + +/* +** Set the title string for subsequent allocations. +*/ +SQLITE_PRIVATE void sqlite3MemdebugSettitle(const char *zTitle){ + unsigned int n = sqlite3Strlen30(zTitle) + 1; + sqlite3_mutex_enter(mem.mutex); + if( n>=sizeof(mem.zTitle) ) n = sizeof(mem.zTitle)-1; + memcpy(mem.zTitle, zTitle, n); + mem.zTitle[n] = 0; + mem.nTitle = ROUND8(n); + sqlite3_mutex_leave(mem.mutex); +} + +SQLITE_PRIVATE void sqlite3MemdebugSync(){ + struct MemBlockHdr *pHdr; + for(pHdr=mem.pFirst; pHdr; pHdr=pHdr->pNext){ + void **pBt = (void**)pHdr; + pBt -= pHdr->nBacktraceSlots; + mem.xBacktrace((int)pHdr->iSize, pHdr->nBacktrace-1, &pBt[1]); + } +} + +/* +** Open the file indicated and write a log of all unfreed memory +** allocations into that log. +*/ +SQLITE_PRIVATE void sqlite3MemdebugDump(const char *zFilename){ + FILE *out; + struct MemBlockHdr *pHdr; + void **pBt; + int i; + out = fopen(zFilename, "w"); + if( out==0 ){ + fprintf(stderr, "** Unable to output memory debug output log: %s **\n", + zFilename); + return; + } + for(pHdr=mem.pFirst; pHdr; pHdr=pHdr->pNext){ + char *z = (char*)pHdr; + z -= pHdr->nBacktraceSlots*sizeof(void*) + pHdr->nTitle; + fprintf(out, "**** %lld bytes at %p from %s ****\n", + pHdr->iSize, &pHdr[1], pHdr->nTitle ? z : "???"); + if( pHdr->nBacktrace ){ + fflush(out); + pBt = (void**)pHdr; + pBt -= pHdr->nBacktraceSlots; + backtrace_symbols_fd(pBt, pHdr->nBacktrace, fileno(out)); + fprintf(out, "\n"); + } + } + fprintf(out, "COUNTS:\n"); + for(i=0; i=1 ); + size = mem3.aPool[i-1].u.hdr.size4x/4; + assert( size==mem3.aPool[i+size-1].u.hdr.prevSize ); + assert( size>=2 ); + if( size <= MX_SMALL ){ + memsys3UnlinkFromList(i, &mem3.aiSmall[size-2]); + }else{ + hash = size % N_HASH; + memsys3UnlinkFromList(i, &mem3.aiHash[hash]); + } +} + +/* +** Link the chunk at mem3.aPool[i] so that is on the list rooted +** at *pRoot. +*/ +static void memsys3LinkIntoList(u32 i, u32 *pRoot){ + assert( sqlite3_mutex_held(mem3.mutex) ); + mem3.aPool[i].u.list.next = *pRoot; + mem3.aPool[i].u.list.prev = 0; + if( *pRoot ){ + mem3.aPool[*pRoot].u.list.prev = i; + } + *pRoot = i; +} + +/* +** Link the chunk at index i into either the appropriate +** small chunk list, or into the large chunk hash table. +*/ +static void memsys3Link(u32 i){ + u32 size, hash; + assert( sqlite3_mutex_held(mem3.mutex) ); + assert( i>=1 ); + assert( (mem3.aPool[i-1].u.hdr.size4x & 1)==0 ); + size = mem3.aPool[i-1].u.hdr.size4x/4; + assert( size==mem3.aPool[i+size-1].u.hdr.prevSize ); + assert( size>=2 ); + if( size <= MX_SMALL ){ + memsys3LinkIntoList(i, &mem3.aiSmall[size-2]); + }else{ + hash = size % N_HASH; + memsys3LinkIntoList(i, &mem3.aiHash[hash]); + } +} + +/* +** If the STATIC_MEM mutex is not already held, obtain it now. The mutex +** will already be held (obtained by code in malloc.c) if +** sqlite3GlobalConfig.bMemStat is true. +*/ +static void memsys3Enter(void){ + if( sqlite3GlobalConfig.bMemstat==0 && mem3.mutex==0 ){ + mem3.mutex = sqlite3MutexAlloc(SQLITE_MUTEX_STATIC_MEM); + } + sqlite3_mutex_enter(mem3.mutex); +} +static void memsys3Leave(void){ + sqlite3_mutex_leave(mem3.mutex); +} + +/* +** Called when we are unable to satisfy an allocation of nBytes. +*/ +static void memsys3OutOfMemory(int nByte){ + if( !mem3.alarmBusy ){ + mem3.alarmBusy = 1; + assert( sqlite3_mutex_held(mem3.mutex) ); + sqlite3_mutex_leave(mem3.mutex); + sqlite3_release_memory(nByte); + sqlite3_mutex_enter(mem3.mutex); + mem3.alarmBusy = 0; + } +} + + +/* +** Chunk i is a free chunk that has been unlinked. Adjust its +** size parameters for check-out and return a pointer to the +** user portion of the chunk. +*/ +static void *memsys3Checkout(u32 i, u32 nBlock){ + u32 x; + assert( sqlite3_mutex_held(mem3.mutex) ); + assert( i>=1 ); + assert( mem3.aPool[i-1].u.hdr.size4x/4==nBlock ); + assert( mem3.aPool[i+nBlock-1].u.hdr.prevSize==nBlock ); + x = mem3.aPool[i-1].u.hdr.size4x; + mem3.aPool[i-1].u.hdr.size4x = nBlock*4 | 1 | (x&2); + mem3.aPool[i+nBlock-1].u.hdr.prevSize = nBlock; + mem3.aPool[i+nBlock-1].u.hdr.size4x |= 2; + return &mem3.aPool[i]; +} + +/* +** Carve a piece off of the end of the mem3.iMaster free chunk. +** Return a pointer to the new allocation. Or, if the master chunk +** is not large enough, return 0. +*/ +static void *memsys3FromMaster(u32 nBlock){ + assert( sqlite3_mutex_held(mem3.mutex) ); + assert( mem3.szMaster>=nBlock ); + if( nBlock>=mem3.szMaster-1 ){ + /* Use the entire master */ + void *p = memsys3Checkout(mem3.iMaster, mem3.szMaster); + mem3.iMaster = 0; + mem3.szMaster = 0; + mem3.mnMaster = 0; + return p; + }else{ + /* Split the master block. Return the tail. */ + u32 newi, x; + newi = mem3.iMaster + mem3.szMaster - nBlock; + assert( newi > mem3.iMaster+1 ); + mem3.aPool[mem3.iMaster+mem3.szMaster-1].u.hdr.prevSize = nBlock; + mem3.aPool[mem3.iMaster+mem3.szMaster-1].u.hdr.size4x |= 2; + mem3.aPool[newi-1].u.hdr.size4x = nBlock*4 + 1; + mem3.szMaster -= nBlock; + mem3.aPool[newi-1].u.hdr.prevSize = mem3.szMaster; + x = mem3.aPool[mem3.iMaster-1].u.hdr.size4x & 2; + mem3.aPool[mem3.iMaster-1].u.hdr.size4x = mem3.szMaster*4 | x; + if( mem3.szMaster < mem3.mnMaster ){ + mem3.mnMaster = mem3.szMaster; + } + return (void*)&mem3.aPool[newi]; + } +} + +/* +** *pRoot is the head of a list of free chunks of the same size +** or same size hash. In other words, *pRoot is an entry in either +** mem3.aiSmall[] or mem3.aiHash[]. +** +** This routine examines all entries on the given list and tries +** to coalesce each entries with adjacent free chunks. +** +** If it sees a chunk that is larger than mem3.iMaster, it replaces +** the current mem3.iMaster with the new larger chunk. In order for +** this mem3.iMaster replacement to work, the master chunk must be +** linked into the hash tables. That is not the normal state of +** affairs, of course. The calling routine must link the master +** chunk before invoking this routine, then must unlink the (possibly +** changed) master chunk once this routine has finished. +*/ +static void memsys3Merge(u32 *pRoot){ + u32 iNext, prev, size, i, x; + + assert( sqlite3_mutex_held(mem3.mutex) ); + for(i=*pRoot; i>0; i=iNext){ + iNext = mem3.aPool[i].u.list.next; + size = mem3.aPool[i-1].u.hdr.size4x; + assert( (size&1)==0 ); + if( (size&2)==0 ){ + memsys3UnlinkFromList(i, pRoot); + assert( i > mem3.aPool[i-1].u.hdr.prevSize ); + prev = i - mem3.aPool[i-1].u.hdr.prevSize; + if( prev==iNext ){ + iNext = mem3.aPool[prev].u.list.next; + } + memsys3Unlink(prev); + size = i + size/4 - prev; + x = mem3.aPool[prev-1].u.hdr.size4x & 2; + mem3.aPool[prev-1].u.hdr.size4x = size*4 | x; + mem3.aPool[prev+size-1].u.hdr.prevSize = size; + memsys3Link(prev); + i = prev; + }else{ + size /= 4; + } + if( size>mem3.szMaster ){ + mem3.iMaster = i; + mem3.szMaster = size; + } + } +} + +/* +** Return a block of memory of at least nBytes in size. +** Return NULL if unable. +** +** This function assumes that the necessary mutexes, if any, are +** already held by the caller. Hence "Unsafe". +*/ +static void *memsys3MallocUnsafe(int nByte){ + u32 i; + u32 nBlock; + u32 toFree; + + assert( sqlite3_mutex_held(mem3.mutex) ); + assert( sizeof(Mem3Block)==8 ); + if( nByte<=12 ){ + nBlock = 2; + }else{ + nBlock = (nByte + 11)/8; + } + assert( nBlock>=2 ); + + /* STEP 1: + ** Look for an entry of the correct size in either the small + ** chunk table or in the large chunk hash table. This is + ** successful most of the time (about 9 times out of 10). + */ + if( nBlock <= MX_SMALL ){ + i = mem3.aiSmall[nBlock-2]; + if( i>0 ){ + memsys3UnlinkFromList(i, &mem3.aiSmall[nBlock-2]); + return memsys3Checkout(i, nBlock); + } + }else{ + int hash = nBlock % N_HASH; + for(i=mem3.aiHash[hash]; i>0; i=mem3.aPool[i].u.list.next){ + if( mem3.aPool[i-1].u.hdr.size4x/4==nBlock ){ + memsys3UnlinkFromList(i, &mem3.aiHash[hash]); + return memsys3Checkout(i, nBlock); + } + } + } + + /* STEP 2: + ** Try to satisfy the allocation by carving a piece off of the end + ** of the master chunk. This step usually works if step 1 fails. + */ + if( mem3.szMaster>=nBlock ){ + return memsys3FromMaster(nBlock); + } + + + /* STEP 3: + ** Loop through the entire memory pool. Coalesce adjacent free + ** chunks. Recompute the master chunk as the largest free chunk. + ** Then try again to satisfy the allocation by carving a piece off + ** of the end of the master chunk. This step happens very + ** rarely (we hope!) + */ + for(toFree=nBlock*16; toFree<(mem3.nPool*16); toFree *= 2){ + memsys3OutOfMemory(toFree); + if( mem3.iMaster ){ + memsys3Link(mem3.iMaster); + mem3.iMaster = 0; + mem3.szMaster = 0; + } + for(i=0; i=nBlock ){ + return memsys3FromMaster(nBlock); + } + } + } + + /* If none of the above worked, then we fail. */ + return 0; +} + +/* +** Free an outstanding memory allocation. +** +** This function assumes that the necessary mutexes, if any, are +** already held by the caller. Hence "Unsafe". +*/ +static void memsys3FreeUnsafe(void *pOld){ + Mem3Block *p = (Mem3Block*)pOld; + int i; + u32 size, x; + assert( sqlite3_mutex_held(mem3.mutex) ); + assert( p>mem3.aPool && p<&mem3.aPool[mem3.nPool] ); + i = p - mem3.aPool; + assert( (mem3.aPool[i-1].u.hdr.size4x&1)==1 ); + size = mem3.aPool[i-1].u.hdr.size4x/4; + assert( i+size<=mem3.nPool+1 ); + mem3.aPool[i-1].u.hdr.size4x &= ~1; + mem3.aPool[i+size-1].u.hdr.prevSize = size; + mem3.aPool[i+size-1].u.hdr.size4x &= ~2; + memsys3Link(i); + + /* Try to expand the master using the newly freed chunk */ + if( mem3.iMaster ){ + while( (mem3.aPool[mem3.iMaster-1].u.hdr.size4x&2)==0 ){ + size = mem3.aPool[mem3.iMaster-1].u.hdr.prevSize; + mem3.iMaster -= size; + mem3.szMaster += size; + memsys3Unlink(mem3.iMaster); + x = mem3.aPool[mem3.iMaster-1].u.hdr.size4x & 2; + mem3.aPool[mem3.iMaster-1].u.hdr.size4x = mem3.szMaster*4 | x; + mem3.aPool[mem3.iMaster+mem3.szMaster-1].u.hdr.prevSize = mem3.szMaster; + } + x = mem3.aPool[mem3.iMaster-1].u.hdr.size4x & 2; + while( (mem3.aPool[mem3.iMaster+mem3.szMaster-1].u.hdr.size4x&1)==0 ){ + memsys3Unlink(mem3.iMaster+mem3.szMaster); + mem3.szMaster += mem3.aPool[mem3.iMaster+mem3.szMaster-1].u.hdr.size4x/4; + mem3.aPool[mem3.iMaster-1].u.hdr.size4x = mem3.szMaster*4 | x; + mem3.aPool[mem3.iMaster+mem3.szMaster-1].u.hdr.prevSize = mem3.szMaster; + } + } +} + +/* +** Return the size of an outstanding allocation, in bytes. The +** size returned omits the 8-byte header overhead. This only +** works for chunks that are currently checked out. +*/ +static int memsys3Size(void *p){ + Mem3Block *pBlock; + assert( p!=0 ); + pBlock = (Mem3Block*)p; + assert( (pBlock[-1].u.hdr.size4x&1)!=0 ); + return (pBlock[-1].u.hdr.size4x&~3)*2 - 4; +} + +/* +** Round up a request size to the next valid allocation size. +*/ +static int memsys3Roundup(int n){ + if( n<=12 ){ + return 12; + }else{ + return ((n+11)&~7) - 4; + } +} + +/* +** Allocate nBytes of memory. +*/ +static void *memsys3Malloc(int nBytes){ + sqlite3_int64 *p; + assert( nBytes>0 ); /* malloc.c filters out 0 byte requests */ + memsys3Enter(); + p = memsys3MallocUnsafe(nBytes); + memsys3Leave(); + return (void*)p; +} + +/* +** Free memory. +*/ +static void memsys3Free(void *pPrior){ + assert( pPrior ); + memsys3Enter(); + memsys3FreeUnsafe(pPrior); + memsys3Leave(); +} + +/* +** Change the size of an existing memory allocation +*/ +static void *memsys3Realloc(void *pPrior, int nBytes){ + int nOld; + void *p; + if( pPrior==0 ){ + return sqlite3_malloc(nBytes); + } + if( nBytes<=0 ){ + sqlite3_free(pPrior); + return 0; + } + nOld = memsys3Size(pPrior); + if( nBytes<=nOld && nBytes>=nOld-128 ){ + return pPrior; + } + memsys3Enter(); + p = memsys3MallocUnsafe(nBytes); + if( p ){ + if( nOld>1)!=(size&1) ){ + fprintf(out, "%p tail checkout bit is incorrect\n", &mem3.aPool[i]); + assert( 0 ); + break; + } + if( size&1 ){ + fprintf(out, "%p %6d bytes checked out\n", &mem3.aPool[i], (size/4)*8-8); + }else{ + fprintf(out, "%p %6d bytes free%s\n", &mem3.aPool[i], (size/4)*8-8, + i==mem3.iMaster ? " **master**" : ""); + } + } + for(i=0; i0; j=mem3.aPool[j].u.list.next){ + fprintf(out, " %p(%d)", &mem3.aPool[j], + (mem3.aPool[j-1].u.hdr.size4x/4)*8-8); + } + fprintf(out, "\n"); + } + for(i=0; i0; j=mem3.aPool[j].u.list.next){ + fprintf(out, " %p(%d)", &mem3.aPool[j], + (mem3.aPool[j-1].u.hdr.size4x/4)*8-8); + } + fprintf(out, "\n"); + } + fprintf(out, "master=%d\n", mem3.iMaster); + fprintf(out, "nowUsed=%d\n", mem3.nPool*8 - mem3.szMaster*8); + fprintf(out, "mxUsed=%d\n", mem3.nPool*8 - mem3.mnMaster*8); + sqlite3_mutex_leave(mem3.mutex); + if( out==stdout ){ + fflush(stdout); + }else{ + fclose(out); + } +#else + UNUSED_PARAMETER(zFilename); +#endif +} + +/* +** This routine is the only routine in this file with external +** linkage. +** +** Populate the low-level memory allocation function pointers in +** sqlite3GlobalConfig.m with pointers to the routines in this file. The +** arguments specify the block of memory to manage. +** +** This routine is only called by sqlite3_config(), and therefore +** is not required to be threadsafe (it is not). +*/ +SQLITE_PRIVATE const sqlite3_mem_methods *sqlite3MemGetMemsys3(void){ + static const sqlite3_mem_methods mempoolMethods = { + memsys3Malloc, + memsys3Free, + memsys3Realloc, + memsys3Size, + memsys3Roundup, + memsys3Init, + memsys3Shutdown, + 0 + }; + return &mempoolMethods; +} + +#endif /* SQLITE_ENABLE_MEMSYS3 */ + +/************** End of mem3.c ************************************************/ +/************** Begin file mem5.c ********************************************/ +/* +** 2007 October 14 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** This file contains the C functions that implement a memory +** allocation subsystem for use by SQLite. +** +** This version of the memory allocation subsystem omits all +** use of malloc(). The application gives SQLite a block of memory +** before calling sqlite3_initialize() from which allocations +** are made and returned by the xMalloc() and xRealloc() +** implementations. Once sqlite3_initialize() has been called, +** the amount of memory available to SQLite is fixed and cannot +** be changed. +** +** This version of the memory allocation subsystem is included +** in the build only if SQLITE_ENABLE_MEMSYS5 is defined. +** +** This memory allocator uses the following algorithm: +** +** 1. All memory allocation sizes are rounded up to a power of 2. +** +** 2. If two adjacent free blocks are the halves of a larger block, +** then the two blocks are coalesced into the single larger block. +** +** 3. New memory is allocated from the first available free block. +** +** This algorithm is described in: J. M. Robson. "Bounds for Some Functions +** Concerning Dynamic Storage Allocation". Journal of the Association for +** Computing Machinery, Volume 21, Number 8, July 1974, pages 491-499. +** +** Let n be the size of the largest allocation divided by the minimum +** allocation size (after rounding all sizes up to a power of 2.) Let M +** be the maximum amount of memory ever outstanding at one time. Let +** N be the total amount of memory available for allocation. Robson +** proved that this memory allocator will never breakdown due to +** fragmentation as long as the following constraint holds: +** +** N >= M*(1 + log2(n)/2) - n + 1 +** +** The sqlite3_status() logic tracks the maximum values of n and M so +** that an application can, at any time, verify this constraint. +*/ +/* #include "sqliteInt.h" */ + +/* +** This version of the memory allocator is used only when +** SQLITE_ENABLE_MEMSYS5 is defined. +*/ +#ifdef SQLITE_ENABLE_MEMSYS5 + +/* +** A minimum allocation is an instance of the following structure. +** Larger allocations are an array of these structures where the +** size of the array is a power of 2. +** +** The size of this object must be a power of two. That fact is +** verified in memsys5Init(). +*/ +typedef struct Mem5Link Mem5Link; +struct Mem5Link { + int next; /* Index of next free chunk */ + int prev; /* Index of previous free chunk */ +}; + +/* +** Maximum size of any allocation is ((1<=0 && i=0 && iLogsize<=LOGMAX ); + assert( (mem5.aCtrl[i] & CTRL_LOGSIZE)==iLogsize ); + + next = MEM5LINK(i)->next; + prev = MEM5LINK(i)->prev; + if( prev<0 ){ + mem5.aiFreelist[iLogsize] = next; + }else{ + MEM5LINK(prev)->next = next; + } + if( next>=0 ){ + MEM5LINK(next)->prev = prev; + } +} + +/* +** Link the chunk at mem5.aPool[i] so that is on the iLogsize +** free list. +*/ +static void memsys5Link(int i, int iLogsize){ + int x; + assert( sqlite3_mutex_held(mem5.mutex) ); + assert( i>=0 && i=0 && iLogsize<=LOGMAX ); + assert( (mem5.aCtrl[i] & CTRL_LOGSIZE)==iLogsize ); + + x = MEM5LINK(i)->next = mem5.aiFreelist[iLogsize]; + MEM5LINK(i)->prev = -1; + if( x>=0 ){ + assert( xprev = i; + } + mem5.aiFreelist[iLogsize] = i; +} + +/* +** Obtain or release the mutex needed to access global data structures. +*/ +static void memsys5Enter(void){ + sqlite3_mutex_enter(mem5.mutex); +} +static void memsys5Leave(void){ + sqlite3_mutex_leave(mem5.mutex); +} + +/* +** Return the size of an outstanding allocation, in bytes. +** This only works for chunks that are currently checked out. +*/ +static int memsys5Size(void *p){ + int iSize, i; + assert( p!=0 ); + i = (int)(((u8 *)p-mem5.zPool)/mem5.szAtom); + assert( i>=0 && i0 ); + + /* No more than 1GiB per allocation */ + if( nByte > 0x40000000 ) return 0; + +#if defined(SQLITE_DEBUG) || defined(SQLITE_TEST) + /* Keep track of the maximum allocation request. Even unfulfilled + ** requests are counted */ + if( (u32)nByte>mem5.maxRequest ){ + mem5.maxRequest = nByte; + } +#endif + + + /* Round nByte up to the next valid power of two */ + for(iFullSz=mem5.szAtom,iLogsize=0; iFullSzLOGMAX ){ + testcase( sqlite3GlobalConfig.xLog!=0 ); + sqlite3_log(SQLITE_NOMEM, "failed to allocate %u bytes", nByte); + return 0; + } + i = mem5.aiFreelist[iBin]; + memsys5Unlink(i, iBin); + while( iBin>iLogsize ){ + int newSize; + + iBin--; + newSize = 1 << iBin; + mem5.aCtrl[i+newSize] = CTRL_FREE | iBin; + memsys5Link(i+newSize, iBin); + } + mem5.aCtrl[i] = iLogsize; + +#if defined(SQLITE_DEBUG) || defined(SQLITE_TEST) + /* Update allocator performance statistics. */ + mem5.nAlloc++; + mem5.totalAlloc += iFullSz; + mem5.totalExcess += iFullSz - nByte; + mem5.currentCount++; + mem5.currentOut += iFullSz; + if( mem5.maxCount=0 && iBlock0 ); + assert( mem5.currentOut>=(size*mem5.szAtom) ); + mem5.currentCount--; + mem5.currentOut -= size*mem5.szAtom; + assert( mem5.currentOut>0 || mem5.currentCount==0 ); + assert( mem5.currentCount>0 || mem5.currentOut==0 ); +#endif + + mem5.aCtrl[iBlock] = CTRL_FREE | iLogsize; + while( ALWAYS(iLogsize>iLogsize) & 1 ){ + iBuddy = iBlock - size; + assert( iBuddy>=0 ); + }else{ + iBuddy = iBlock + size; + if( iBuddy>=mem5.nBlock ) break; + } + if( mem5.aCtrl[iBuddy]!=(CTRL_FREE | iLogsize) ) break; + memsys5Unlink(iBuddy, iLogsize); + iLogsize++; + if( iBuddy0 ){ + memsys5Enter(); + p = memsys5MallocUnsafe(nBytes); + memsys5Leave(); + } + return (void*)p; +} + +/* +** Free memory. +** +** The outer layer memory allocator prevents this routine from +** being called with pPrior==0. +*/ +static void memsys5Free(void *pPrior){ + assert( pPrior!=0 ); + memsys5Enter(); + memsys5FreeUnsafe(pPrior); + memsys5Leave(); +} + +/* +** Change the size of an existing memory allocation. +** +** The outer layer memory allocator prevents this routine from +** being called with pPrior==0. +** +** nBytes is always a value obtained from a prior call to +** memsys5Round(). Hence nBytes is always a non-negative power +** of two. If nBytes==0 that means that an oversize allocation +** (an allocation larger than 0x40000000) was requested and this +** routine should return 0 without freeing pPrior. +*/ +static void *memsys5Realloc(void *pPrior, int nBytes){ + int nOld; + void *p; + assert( pPrior!=0 ); + assert( (nBytes&(nBytes-1))==0 ); /* EV: R-46199-30249 */ + assert( nBytes>=0 ); + if( nBytes==0 ){ + return 0; + } + nOld = memsys5Size(pPrior); + if( nBytes<=nOld ){ + return pPrior; + } + p = memsys5Malloc(nBytes); + if( p ){ + memcpy(p, pPrior, nOld); + memsys5Free(pPrior); + } + return p; +} + +/* +** Round up a request size to the next valid allocation size. If +** the allocation is too large to be handled by this allocation system, +** return 0. +** +** All allocations must be a power of two and must be expressed by a +** 32-bit signed integer. Hence the largest allocation is 0x40000000 +** or 1073741824 bytes. +*/ +static int memsys5Roundup(int n){ + int iFullSz; + if( n > 0x40000000 ) return 0; + for(iFullSz=mem5.szAtom; iFullSz 0 +** memsys5Log(2) -> 1 +** memsys5Log(4) -> 2 +** memsys5Log(5) -> 3 +** memsys5Log(8) -> 3 +** memsys5Log(9) -> 4 +*/ +static int memsys5Log(int iValue){ + int iLog; + for(iLog=0; (iLog<(int)((sizeof(int)*8)-1)) && (1<mem5.szAtom ){ + mem5.szAtom = mem5.szAtom << 1; + } + + mem5.nBlock = (nByte / (mem5.szAtom+sizeof(u8))); + mem5.zPool = zByte; + mem5.aCtrl = (u8 *)&mem5.zPool[mem5.nBlock*mem5.szAtom]; + + for(ii=0; ii<=LOGMAX; ii++){ + mem5.aiFreelist[ii] = -1; + } + + iOffset = 0; + for(ii=LOGMAX; ii>=0; ii--){ + int nAlloc = (1<mem5.nBlock); + } + + /* If a mutex is required for normal operation, allocate one */ + if( sqlite3GlobalConfig.bMemstat==0 ){ + mem5.mutex = sqlite3MutexAlloc(SQLITE_MUTEX_STATIC_MEM); + } + + return SQLITE_OK; +} + +/* +** Deinitialize this module. +*/ +static void memsys5Shutdown(void *NotUsed){ + UNUSED_PARAMETER(NotUsed); + mem5.mutex = 0; + return; +} + +#ifdef SQLITE_TEST +/* +** Open the file indicated and write a log of all unfreed memory +** allocations into that log. +*/ +SQLITE_PRIVATE void sqlite3Memsys5Dump(const char *zFilename){ + FILE *out; + int i, j, n; + int nMinLog; + + if( zFilename==0 || zFilename[0]==0 ){ + out = stdout; + }else{ + out = fopen(zFilename, "w"); + if( out==0 ){ + fprintf(stderr, "** Unable to output memory debug output log: %s **\n", + zFilename); + return; + } + } + memsys5Enter(); + nMinLog = memsys5Log(mem5.szAtom); + for(i=0; i<=LOGMAX && i+nMinLog<32; i++){ + for(n=0, j=mem5.aiFreelist[i]; j>=0; j = MEM5LINK(j)->next, n++){} + fprintf(out, "freelist items of size %d: %d\n", mem5.szAtom << i, n); + } + fprintf(out, "mem5.nAlloc = %llu\n", mem5.nAlloc); + fprintf(out, "mem5.totalAlloc = %llu\n", mem5.totalAlloc); + fprintf(out, "mem5.totalExcess = %llu\n", mem5.totalExcess); + fprintf(out, "mem5.currentOut = %u\n", mem5.currentOut); + fprintf(out, "mem5.currentCount = %u\n", mem5.currentCount); + fprintf(out, "mem5.maxOut = %u\n", mem5.maxOut); + fprintf(out, "mem5.maxCount = %u\n", mem5.maxCount); + fprintf(out, "mem5.maxRequest = %u\n", mem5.maxRequest); + memsys5Leave(); + if( out==stdout ){ + fflush(stdout); + }else{ + fclose(out); + } +} +#endif + +/* +** This routine is the only routine in this file with external +** linkage. It returns a pointer to a static sqlite3_mem_methods +** struct populated with the memsys5 methods. +*/ +SQLITE_PRIVATE const sqlite3_mem_methods *sqlite3MemGetMemsys5(void){ + static const sqlite3_mem_methods memsys5Methods = { + memsys5Malloc, + memsys5Free, + memsys5Realloc, + memsys5Size, + memsys5Roundup, + memsys5Init, + memsys5Shutdown, + 0 + }; + return &memsys5Methods; +} + +#endif /* SQLITE_ENABLE_MEMSYS5 */ + +/************** End of mem5.c ************************************************/ +/************** Begin file mutex.c *******************************************/ +/* +** 2007 August 14 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** This file contains the C functions that implement mutexes. +** +** This file contains code that is common across all mutex implementations. +*/ +/* #include "sqliteInt.h" */ + +#if defined(SQLITE_DEBUG) && !defined(SQLITE_MUTEX_OMIT) +/* +** For debugging purposes, record when the mutex subsystem is initialized +** and uninitialized so that we can assert() if there is an attempt to +** allocate a mutex while the system is uninitialized. +*/ +static SQLITE_WSD int mutexIsInit = 0; +#endif /* SQLITE_DEBUG && !defined(SQLITE_MUTEX_OMIT) */ + + +#ifndef SQLITE_MUTEX_OMIT + +#ifdef SQLITE_ENABLE_MULTITHREADED_CHECKS +/* +** This block (enclosed by SQLITE_ENABLE_MULTITHREADED_CHECKS) contains +** the implementation of a wrapper around the system default mutex +** implementation (sqlite3DefaultMutex()). +** +** Most calls are passed directly through to the underlying default +** mutex implementation. Except, if a mutex is configured by calling +** sqlite3MutexWarnOnContention() on it, then if contention is ever +** encountered within xMutexEnter() a warning is emitted via sqlite3_log(). +** +** This type of mutex is used as the database handle mutex when testing +** apps that usually use SQLITE_CONFIG_MULTITHREAD mode. +*/ + +/* +** Type for all mutexes used when SQLITE_ENABLE_MULTITHREADED_CHECKS +** is defined. Variable CheckMutex.mutex is a pointer to the real mutex +** allocated by the system mutex implementation. Variable iType is usually set +** to the type of mutex requested - SQLITE_MUTEX_RECURSIVE, SQLITE_MUTEX_FAST +** or one of the static mutex identifiers. Or, if this is a recursive mutex +** that has been configured using sqlite3MutexWarnOnContention(), it is +** set to SQLITE_MUTEX_WARNONCONTENTION. +*/ +typedef struct CheckMutex CheckMutex; +struct CheckMutex { + int iType; + sqlite3_mutex *mutex; +}; + +#define SQLITE_MUTEX_WARNONCONTENTION (-1) + +/* +** Pointer to real mutex methods object used by the CheckMutex +** implementation. Set by checkMutexInit(). +*/ +static SQLITE_WSD const sqlite3_mutex_methods *pGlobalMutexMethods; + +#ifdef SQLITE_DEBUG +static int checkMutexHeld(sqlite3_mutex *p){ + return pGlobalMutexMethods->xMutexHeld(((CheckMutex*)p)->mutex); +} +static int checkMutexNotheld(sqlite3_mutex *p){ + return pGlobalMutexMethods->xMutexNotheld(((CheckMutex*)p)->mutex); +} +#endif + +/* +** Initialize and deinitialize the mutex subsystem. +*/ +static int checkMutexInit(void){ + pGlobalMutexMethods = sqlite3DefaultMutex(); + return SQLITE_OK; +} +static int checkMutexEnd(void){ + pGlobalMutexMethods = 0; + return SQLITE_OK; +} + +/* +** Allocate a mutex. +*/ +static sqlite3_mutex *checkMutexAlloc(int iType){ + static CheckMutex staticMutexes[] = { + {2, 0}, {3, 0}, {4, 0}, {5, 0}, + {6, 0}, {7, 0}, {8, 0}, {9, 0}, + {10, 0}, {11, 0}, {12, 0}, {13, 0} + }; + CheckMutex *p = 0; + + assert( SQLITE_MUTEX_RECURSIVE==1 && SQLITE_MUTEX_FAST==0 ); + if( iType<2 ){ + p = sqlite3MallocZero(sizeof(CheckMutex)); + if( p==0 ) return 0; + p->iType = iType; + }else{ +#ifdef SQLITE_ENABLE_API_ARMOR + if( iType-2>=ArraySize(staticMutexes) ){ + (void)SQLITE_MISUSE_BKPT; + return 0; + } +#endif + p = &staticMutexes[iType-2]; + } + + if( p->mutex==0 ){ + p->mutex = pGlobalMutexMethods->xMutexAlloc(iType); + if( p->mutex==0 ){ + if( iType<2 ){ + sqlite3_free(p); + } + p = 0; + } + } + + return (sqlite3_mutex*)p; +} + +/* +** Free a mutex. +*/ +static void checkMutexFree(sqlite3_mutex *p){ + assert( SQLITE_MUTEX_RECURSIVE<2 ); + assert( SQLITE_MUTEX_FAST<2 ); + assert( SQLITE_MUTEX_WARNONCONTENTION<2 ); + +#if SQLITE_ENABLE_API_ARMOR + if( ((CheckMutex*)p)->iType<2 ) +#endif + { + CheckMutex *pCheck = (CheckMutex*)p; + pGlobalMutexMethods->xMutexFree(pCheck->mutex); + sqlite3_free(pCheck); + } +#ifdef SQLITE_ENABLE_API_ARMOR + else{ + (void)SQLITE_MISUSE_BKPT; + } +#endif +} + +/* +** Enter the mutex. +*/ +static void checkMutexEnter(sqlite3_mutex *p){ + CheckMutex *pCheck = (CheckMutex*)p; + if( pCheck->iType==SQLITE_MUTEX_WARNONCONTENTION ){ + if( SQLITE_OK==pGlobalMutexMethods->xMutexTry(pCheck->mutex) ){ + return; + } + sqlite3_log(SQLITE_MISUSE, + "illegal multi-threaded access to database connection" + ); + } + pGlobalMutexMethods->xMutexEnter(pCheck->mutex); +} + +/* +** Enter the mutex (do not block). +*/ +static int checkMutexTry(sqlite3_mutex *p){ + CheckMutex *pCheck = (CheckMutex*)p; + return pGlobalMutexMethods->xMutexTry(pCheck->mutex); +} + +/* +** Leave the mutex. +*/ +static void checkMutexLeave(sqlite3_mutex *p){ + CheckMutex *pCheck = (CheckMutex*)p; + pGlobalMutexMethods->xMutexLeave(pCheck->mutex); +} + +sqlite3_mutex_methods const *multiThreadedCheckMutex(void){ + static const sqlite3_mutex_methods sMutex = { + checkMutexInit, + checkMutexEnd, + checkMutexAlloc, + checkMutexFree, + checkMutexEnter, + checkMutexTry, + checkMutexLeave, +#ifdef SQLITE_DEBUG + checkMutexHeld, + checkMutexNotheld +#else + 0, + 0 +#endif + }; + return &sMutex; +} + +/* +** Mark the SQLITE_MUTEX_RECURSIVE mutex passed as the only argument as +** one on which there should be no contention. +*/ +SQLITE_PRIVATE void sqlite3MutexWarnOnContention(sqlite3_mutex *p){ + if( sqlite3GlobalConfig.mutex.xMutexAlloc==checkMutexAlloc ){ + CheckMutex *pCheck = (CheckMutex*)p; + assert( pCheck->iType==SQLITE_MUTEX_RECURSIVE ); + pCheck->iType = SQLITE_MUTEX_WARNONCONTENTION; + } +} +#endif /* ifdef SQLITE_ENABLE_MULTITHREADED_CHECKS */ + +/* +** Initialize the mutex system. +*/ +SQLITE_PRIVATE int sqlite3MutexInit(void){ + int rc = SQLITE_OK; + if( !sqlite3GlobalConfig.mutex.xMutexAlloc ){ + /* If the xMutexAlloc method has not been set, then the user did not + ** install a mutex implementation via sqlite3_config() prior to + ** sqlite3_initialize() being called. This block copies pointers to + ** the default implementation into the sqlite3GlobalConfig structure. + */ + sqlite3_mutex_methods const *pFrom; + sqlite3_mutex_methods *pTo = &sqlite3GlobalConfig.mutex; + + if( sqlite3GlobalConfig.bCoreMutex ){ +#ifdef SQLITE_ENABLE_MULTITHREADED_CHECKS + pFrom = multiThreadedCheckMutex(); +#else + pFrom = sqlite3DefaultMutex(); +#endif + }else{ + pFrom = sqlite3NoopMutex(); + } + pTo->xMutexInit = pFrom->xMutexInit; + pTo->xMutexEnd = pFrom->xMutexEnd; + pTo->xMutexFree = pFrom->xMutexFree; + pTo->xMutexEnter = pFrom->xMutexEnter; + pTo->xMutexTry = pFrom->xMutexTry; + pTo->xMutexLeave = pFrom->xMutexLeave; + pTo->xMutexHeld = pFrom->xMutexHeld; + pTo->xMutexNotheld = pFrom->xMutexNotheld; + sqlite3MemoryBarrier(); + pTo->xMutexAlloc = pFrom->xMutexAlloc; + } + assert( sqlite3GlobalConfig.mutex.xMutexInit ); + rc = sqlite3GlobalConfig.mutex.xMutexInit(); + +#ifdef SQLITE_DEBUG + GLOBAL(int, mutexIsInit) = 1; +#endif + + return rc; +} + +/* +** Shutdown the mutex system. This call frees resources allocated by +** sqlite3MutexInit(). +*/ +SQLITE_PRIVATE int sqlite3MutexEnd(void){ + int rc = SQLITE_OK; + if( sqlite3GlobalConfig.mutex.xMutexEnd ){ + rc = sqlite3GlobalConfig.mutex.xMutexEnd(); + } + +#ifdef SQLITE_DEBUG + GLOBAL(int, mutexIsInit) = 0; +#endif + + return rc; +} + +/* +** Retrieve a pointer to a static mutex or allocate a new dynamic one. +*/ +SQLITE_API sqlite3_mutex *sqlite3_mutex_alloc(int id){ +#ifndef SQLITE_OMIT_AUTOINIT + if( id<=SQLITE_MUTEX_RECURSIVE && sqlite3_initialize() ) return 0; + if( id>SQLITE_MUTEX_RECURSIVE && sqlite3MutexInit() ) return 0; +#endif + assert( sqlite3GlobalConfig.mutex.xMutexAlloc ); + return sqlite3GlobalConfig.mutex.xMutexAlloc(id); +} + +SQLITE_PRIVATE sqlite3_mutex *sqlite3MutexAlloc(int id){ + if( !sqlite3GlobalConfig.bCoreMutex ){ + return 0; + } + assert( GLOBAL(int, mutexIsInit) ); + assert( sqlite3GlobalConfig.mutex.xMutexAlloc ); + return sqlite3GlobalConfig.mutex.xMutexAlloc(id); +} + +/* +** Free a dynamic mutex. +*/ +SQLITE_API void sqlite3_mutex_free(sqlite3_mutex *p){ + if( p ){ + assert( sqlite3GlobalConfig.mutex.xMutexFree ); + sqlite3GlobalConfig.mutex.xMutexFree(p); + } +} + +/* +** Obtain the mutex p. If some other thread already has the mutex, block +** until it can be obtained. +*/ +SQLITE_API void sqlite3_mutex_enter(sqlite3_mutex *p){ + if( p ){ + assert( sqlite3GlobalConfig.mutex.xMutexEnter ); + sqlite3GlobalConfig.mutex.xMutexEnter(p); + } +} + +/* +** Obtain the mutex p. If successful, return SQLITE_OK. Otherwise, if another +** thread holds the mutex and it cannot be obtained, return SQLITE_BUSY. +*/ +SQLITE_API int sqlite3_mutex_try(sqlite3_mutex *p){ + int rc = SQLITE_OK; + if( p ){ + assert( sqlite3GlobalConfig.mutex.xMutexTry ); + return sqlite3GlobalConfig.mutex.xMutexTry(p); + } + return rc; +} + +/* +** The sqlite3_mutex_leave() routine exits a mutex that was previously +** entered by the same thread. The behavior is undefined if the mutex +** is not currently entered. If a NULL pointer is passed as an argument +** this function is a no-op. +*/ +SQLITE_API void sqlite3_mutex_leave(sqlite3_mutex *p){ + if( p ){ + assert( sqlite3GlobalConfig.mutex.xMutexLeave ); + sqlite3GlobalConfig.mutex.xMutexLeave(p); + } +} + +#ifndef NDEBUG +/* +** The sqlite3_mutex_held() and sqlite3_mutex_notheld() routine are +** intended for use inside assert() statements. +*/ +SQLITE_API int sqlite3_mutex_held(sqlite3_mutex *p){ + assert( p==0 || sqlite3GlobalConfig.mutex.xMutexHeld ); + return p==0 || sqlite3GlobalConfig.mutex.xMutexHeld(p); +} +SQLITE_API int sqlite3_mutex_notheld(sqlite3_mutex *p){ + assert( p==0 || sqlite3GlobalConfig.mutex.xMutexNotheld ); + return p==0 || sqlite3GlobalConfig.mutex.xMutexNotheld(p); +} +#endif + +#endif /* !defined(SQLITE_MUTEX_OMIT) */ + +/************** End of mutex.c ***********************************************/ +/************** Begin file mutex_noop.c **************************************/ +/* +** 2008 October 07 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** This file contains the C functions that implement mutexes. +** +** This implementation in this file does not provide any mutual +** exclusion and is thus suitable for use only in applications +** that use SQLite in a single thread. The routines defined +** here are place-holders. Applications can substitute working +** mutex routines at start-time using the +** +** sqlite3_config(SQLITE_CONFIG_MUTEX,...) +** +** interface. +** +** If compiled with SQLITE_DEBUG, then additional logic is inserted +** that does error checking on mutexes to make sure they are being +** called correctly. +*/ +/* #include "sqliteInt.h" */ + +#ifndef SQLITE_MUTEX_OMIT + +#ifndef SQLITE_DEBUG +/* +** Stub routines for all mutex methods. +** +** This routines provide no mutual exclusion or error checking. +*/ +static int noopMutexInit(void){ return SQLITE_OK; } +static int noopMutexEnd(void){ return SQLITE_OK; } +static sqlite3_mutex *noopMutexAlloc(int id){ + UNUSED_PARAMETER(id); + return (sqlite3_mutex*)8; +} +static void noopMutexFree(sqlite3_mutex *p){ UNUSED_PARAMETER(p); return; } +static void noopMutexEnter(sqlite3_mutex *p){ UNUSED_PARAMETER(p); return; } +static int noopMutexTry(sqlite3_mutex *p){ + UNUSED_PARAMETER(p); + return SQLITE_OK; +} +static void noopMutexLeave(sqlite3_mutex *p){ UNUSED_PARAMETER(p); return; } + +SQLITE_PRIVATE sqlite3_mutex_methods const *sqlite3NoopMutex(void){ + static const sqlite3_mutex_methods sMutex = { + noopMutexInit, + noopMutexEnd, + noopMutexAlloc, + noopMutexFree, + noopMutexEnter, + noopMutexTry, + noopMutexLeave, + + 0, + 0, + }; + + return &sMutex; +} +#endif /* !SQLITE_DEBUG */ + +#ifdef SQLITE_DEBUG +/* +** In this implementation, error checking is provided for testing +** and debugging purposes. The mutexes still do not provide any +** mutual exclusion. +*/ + +/* +** The mutex object +*/ +typedef struct sqlite3_debug_mutex { + int id; /* The mutex type */ + int cnt; /* Number of entries without a matching leave */ +} sqlite3_debug_mutex; + +/* +** The sqlite3_mutex_held() and sqlite3_mutex_notheld() routine are +** intended for use inside assert() statements. +*/ +static int debugMutexHeld(sqlite3_mutex *pX){ + sqlite3_debug_mutex *p = (sqlite3_debug_mutex*)pX; + return p==0 || p->cnt>0; +} +static int debugMutexNotheld(sqlite3_mutex *pX){ + sqlite3_debug_mutex *p = (sqlite3_debug_mutex*)pX; + return p==0 || p->cnt==0; +} + +/* +** Initialize and deinitialize the mutex subsystem. +*/ +static int debugMutexInit(void){ return SQLITE_OK; } +static int debugMutexEnd(void){ return SQLITE_OK; } + +/* +** The sqlite3_mutex_alloc() routine allocates a new +** mutex and returns a pointer to it. If it returns NULL +** that means that a mutex could not be allocated. +*/ +static sqlite3_mutex *debugMutexAlloc(int id){ + static sqlite3_debug_mutex aStatic[SQLITE_MUTEX_STATIC_VFS3 - 1]; + sqlite3_debug_mutex *pNew = 0; + switch( id ){ + case SQLITE_MUTEX_FAST: + case SQLITE_MUTEX_RECURSIVE: { + pNew = sqlite3Malloc(sizeof(*pNew)); + if( pNew ){ + pNew->id = id; + pNew->cnt = 0; + } + break; + } + default: { +#ifdef SQLITE_ENABLE_API_ARMOR + if( id-2<0 || id-2>=ArraySize(aStatic) ){ + (void)SQLITE_MISUSE_BKPT; + return 0; + } +#endif + pNew = &aStatic[id-2]; + pNew->id = id; + break; + } + } + return (sqlite3_mutex*)pNew; +} + +/* +** This routine deallocates a previously allocated mutex. +*/ +static void debugMutexFree(sqlite3_mutex *pX){ + sqlite3_debug_mutex *p = (sqlite3_debug_mutex*)pX; + assert( p->cnt==0 ); + if( p->id==SQLITE_MUTEX_RECURSIVE || p->id==SQLITE_MUTEX_FAST ){ + sqlite3_free(p); + }else{ +#ifdef SQLITE_ENABLE_API_ARMOR + (void)SQLITE_MISUSE_BKPT; +#endif + } +} + +/* +** The sqlite3_mutex_enter() and sqlite3_mutex_try() routines attempt +** to enter a mutex. If another thread is already within the mutex, +** sqlite3_mutex_enter() will block and sqlite3_mutex_try() will return +** SQLITE_BUSY. The sqlite3_mutex_try() interface returns SQLITE_OK +** upon successful entry. Mutexes created using SQLITE_MUTEX_RECURSIVE can +** be entered multiple times by the same thread. In such cases the, +** mutex must be exited an equal number of times before another thread +** can enter. If the same thread tries to enter any other kind of mutex +** more than once, the behavior is undefined. +*/ +static void debugMutexEnter(sqlite3_mutex *pX){ + sqlite3_debug_mutex *p = (sqlite3_debug_mutex*)pX; + assert( p->id==SQLITE_MUTEX_RECURSIVE || debugMutexNotheld(pX) ); + p->cnt++; +} +static int debugMutexTry(sqlite3_mutex *pX){ + sqlite3_debug_mutex *p = (sqlite3_debug_mutex*)pX; + assert( p->id==SQLITE_MUTEX_RECURSIVE || debugMutexNotheld(pX) ); + p->cnt++; + return SQLITE_OK; +} + +/* +** The sqlite3_mutex_leave() routine exits a mutex that was +** previously entered by the same thread. The behavior +** is undefined if the mutex is not currently entered or +** is not currently allocated. SQLite will never do either. +*/ +static void debugMutexLeave(sqlite3_mutex *pX){ + sqlite3_debug_mutex *p = (sqlite3_debug_mutex*)pX; + assert( debugMutexHeld(pX) ); + p->cnt--; + assert( p->id==SQLITE_MUTEX_RECURSIVE || debugMutexNotheld(pX) ); +} + +SQLITE_PRIVATE sqlite3_mutex_methods const *sqlite3NoopMutex(void){ + static const sqlite3_mutex_methods sMutex = { + debugMutexInit, + debugMutexEnd, + debugMutexAlloc, + debugMutexFree, + debugMutexEnter, + debugMutexTry, + debugMutexLeave, + + debugMutexHeld, + debugMutexNotheld + }; + + return &sMutex; +} +#endif /* SQLITE_DEBUG */ + +/* +** If compiled with SQLITE_MUTEX_NOOP, then the no-op mutex implementation +** is used regardless of the run-time threadsafety setting. +*/ +#ifdef SQLITE_MUTEX_NOOP +SQLITE_PRIVATE sqlite3_mutex_methods const *sqlite3DefaultMutex(void){ + return sqlite3NoopMutex(); +} +#endif /* defined(SQLITE_MUTEX_NOOP) */ +#endif /* !defined(SQLITE_MUTEX_OMIT) */ + +/************** End of mutex_noop.c ******************************************/ +/************** Begin file mutex_unix.c **************************************/ +/* +** 2007 August 28 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** This file contains the C functions that implement mutexes for pthreads +*/ +/* #include "sqliteInt.h" */ + +/* +** The code in this file is only used if we are compiling threadsafe +** under unix with pthreads. +** +** Note that this implementation requires a version of pthreads that +** supports recursive mutexes. +*/ +#ifdef SQLITE_MUTEX_PTHREADS + +#include + +/* +** The sqlite3_mutex.id, sqlite3_mutex.nRef, and sqlite3_mutex.owner fields +** are necessary under two condidtions: (1) Debug builds and (2) using +** home-grown mutexes. Encapsulate these conditions into a single #define. +*/ +#if defined(SQLITE_DEBUG) || defined(SQLITE_HOMEGROWN_RECURSIVE_MUTEX) +# define SQLITE_MUTEX_NREF 1 +#else +# define SQLITE_MUTEX_NREF 0 +#endif + +/* +** Each recursive mutex is an instance of the following structure. +*/ +struct sqlite3_mutex { + pthread_mutex_t mutex; /* Mutex controlling the lock */ +#if SQLITE_MUTEX_NREF || defined(SQLITE_ENABLE_API_ARMOR) + int id; /* Mutex type */ +#endif +#if SQLITE_MUTEX_NREF + volatile int nRef; /* Number of entrances */ + volatile pthread_t owner; /* Thread that is within this mutex */ + int trace; /* True to trace changes */ +#endif +}; +#if SQLITE_MUTEX_NREF +# define SQLITE3_MUTEX_INITIALIZER(id) \ + {PTHREAD_MUTEX_INITIALIZER,id,0,(pthread_t)0,0} +#elif defined(SQLITE_ENABLE_API_ARMOR) +# define SQLITE3_MUTEX_INITIALIZER(id) { PTHREAD_MUTEX_INITIALIZER, id } +#else +#define SQLITE3_MUTEX_INITIALIZER(id) { PTHREAD_MUTEX_INITIALIZER } +#endif + +/* +** The sqlite3_mutex_held() and sqlite3_mutex_notheld() routine are +** intended for use only inside assert() statements. On some platforms, +** there might be race conditions that can cause these routines to +** deliver incorrect results. In particular, if pthread_equal() is +** not an atomic operation, then these routines might delivery +** incorrect results. On most platforms, pthread_equal() is a +** comparison of two integers and is therefore atomic. But we are +** told that HPUX is not such a platform. If so, then these routines +** will not always work correctly on HPUX. +** +** On those platforms where pthread_equal() is not atomic, SQLite +** should be compiled without -DSQLITE_DEBUG and with -DNDEBUG to +** make sure no assert() statements are evaluated and hence these +** routines are never called. +*/ +#if !defined(NDEBUG) || defined(SQLITE_DEBUG) +static int pthreadMutexHeld(sqlite3_mutex *p){ + return (p->nRef!=0 && pthread_equal(p->owner, pthread_self())); +} +static int pthreadMutexNotheld(sqlite3_mutex *p){ + return p->nRef==0 || pthread_equal(p->owner, pthread_self())==0; +} +#endif + +/* +** Try to provide a memory barrier operation, needed for initialization +** and also for the implementation of xShmBarrier in the VFS in cases +** where SQLite is compiled without mutexes. +*/ +SQLITE_PRIVATE void sqlite3MemoryBarrier(void){ +#if defined(SQLITE_MEMORY_BARRIER) + SQLITE_MEMORY_BARRIER; +#elif defined(__GNUC__) && GCC_VERSION>=4001000 + __sync_synchronize(); +#endif +} + +/* +** Initialize and deinitialize the mutex subsystem. +*/ +static int pthreadMutexInit(void){ return SQLITE_OK; } +static int pthreadMutexEnd(void){ return SQLITE_OK; } + +/* +** The sqlite3_mutex_alloc() routine allocates a new +** mutex and returns a pointer to it. If it returns NULL +** that means that a mutex could not be allocated. SQLite +** will unwind its stack and return an error. The argument +** to sqlite3_mutex_alloc() is one of these integer constants: +** +**
    +**
  • SQLITE_MUTEX_FAST +**
  • SQLITE_MUTEX_RECURSIVE +**
  • SQLITE_MUTEX_STATIC_MASTER +**
  • SQLITE_MUTEX_STATIC_MEM +**
  • SQLITE_MUTEX_STATIC_OPEN +**
  • SQLITE_MUTEX_STATIC_PRNG +**
  • SQLITE_MUTEX_STATIC_LRU +**
  • SQLITE_MUTEX_STATIC_PMEM +**
  • SQLITE_MUTEX_STATIC_APP1 +**
  • SQLITE_MUTEX_STATIC_APP2 +**
  • SQLITE_MUTEX_STATIC_APP3 +**
  • SQLITE_MUTEX_STATIC_VFS1 +**
  • SQLITE_MUTEX_STATIC_VFS2 +**
  • SQLITE_MUTEX_STATIC_VFS3 +**
+** +** The first two constants cause sqlite3_mutex_alloc() to create +** a new mutex. The new mutex is recursive when SQLITE_MUTEX_RECURSIVE +** is used but not necessarily so when SQLITE_MUTEX_FAST is used. +** The mutex implementation does not need to make a distinction +** between SQLITE_MUTEX_RECURSIVE and SQLITE_MUTEX_FAST if it does +** not want to. But SQLite will only request a recursive mutex in +** cases where it really needs one. If a faster non-recursive mutex +** implementation is available on the host platform, the mutex subsystem +** might return such a mutex in response to SQLITE_MUTEX_FAST. +** +** The other allowed parameters to sqlite3_mutex_alloc() each return +** a pointer to a static preexisting mutex. Six static mutexes are +** used by the current version of SQLite. Future versions of SQLite +** may add additional static mutexes. Static mutexes are for internal +** use by SQLite only. Applications that use SQLite mutexes should +** use only the dynamic mutexes returned by SQLITE_MUTEX_FAST or +** SQLITE_MUTEX_RECURSIVE. +** +** Note that if one of the dynamic mutex parameters (SQLITE_MUTEX_FAST +** or SQLITE_MUTEX_RECURSIVE) is used then sqlite3_mutex_alloc() +** returns a different mutex on every call. But for the static +** mutex types, the same mutex is returned on every call that has +** the same type number. +*/ +static sqlite3_mutex *pthreadMutexAlloc(int iType){ + static sqlite3_mutex staticMutexes[] = { + SQLITE3_MUTEX_INITIALIZER(2), + SQLITE3_MUTEX_INITIALIZER(3), + SQLITE3_MUTEX_INITIALIZER(4), + SQLITE3_MUTEX_INITIALIZER(5), + SQLITE3_MUTEX_INITIALIZER(6), + SQLITE3_MUTEX_INITIALIZER(7), + SQLITE3_MUTEX_INITIALIZER(8), + SQLITE3_MUTEX_INITIALIZER(9), + SQLITE3_MUTEX_INITIALIZER(10), + SQLITE3_MUTEX_INITIALIZER(11), + SQLITE3_MUTEX_INITIALIZER(12), + SQLITE3_MUTEX_INITIALIZER(13) + }; + sqlite3_mutex *p; + switch( iType ){ + case SQLITE_MUTEX_RECURSIVE: { + p = sqlite3MallocZero( sizeof(*p) ); + if( p ){ +#ifdef SQLITE_HOMEGROWN_RECURSIVE_MUTEX + /* If recursive mutexes are not available, we will have to + ** build our own. See below. */ + pthread_mutex_init(&p->mutex, 0); +#else + /* Use a recursive mutex if it is available */ + pthread_mutexattr_t recursiveAttr; + pthread_mutexattr_init(&recursiveAttr); + pthread_mutexattr_settype(&recursiveAttr, PTHREAD_MUTEX_RECURSIVE); + pthread_mutex_init(&p->mutex, &recursiveAttr); + pthread_mutexattr_destroy(&recursiveAttr); +#endif +#if SQLITE_MUTEX_NREF || defined(SQLITE_ENABLE_API_ARMOR) + p->id = SQLITE_MUTEX_RECURSIVE; +#endif + } + break; + } + case SQLITE_MUTEX_FAST: { + p = sqlite3MallocZero( sizeof(*p) ); + if( p ){ + pthread_mutex_init(&p->mutex, 0); +#if SQLITE_MUTEX_NREF || defined(SQLITE_ENABLE_API_ARMOR) + p->id = SQLITE_MUTEX_FAST; +#endif + } + break; + } + default: { +#ifdef SQLITE_ENABLE_API_ARMOR + if( iType-2<0 || iType-2>=ArraySize(staticMutexes) ){ + (void)SQLITE_MISUSE_BKPT; + return 0; + } +#endif + p = &staticMutexes[iType-2]; + break; + } + } +#if SQLITE_MUTEX_NREF || defined(SQLITE_ENABLE_API_ARMOR) + assert( p==0 || p->id==iType ); +#endif + return p; +} + + +/* +** This routine deallocates a previously +** allocated mutex. SQLite is careful to deallocate every +** mutex that it allocates. +*/ +static void pthreadMutexFree(sqlite3_mutex *p){ + assert( p->nRef==0 ); +#if SQLITE_ENABLE_API_ARMOR + if( p->id==SQLITE_MUTEX_FAST || p->id==SQLITE_MUTEX_RECURSIVE ) +#endif + { + pthread_mutex_destroy(&p->mutex); + sqlite3_free(p); + } +#ifdef SQLITE_ENABLE_API_ARMOR + else{ + (void)SQLITE_MISUSE_BKPT; + } +#endif +} + +/* +** The sqlite3_mutex_enter() and sqlite3_mutex_try() routines attempt +** to enter a mutex. If another thread is already within the mutex, +** sqlite3_mutex_enter() will block and sqlite3_mutex_try() will return +** SQLITE_BUSY. The sqlite3_mutex_try() interface returns SQLITE_OK +** upon successful entry. Mutexes created using SQLITE_MUTEX_RECURSIVE can +** be entered multiple times by the same thread. In such cases the, +** mutex must be exited an equal number of times before another thread +** can enter. If the same thread tries to enter any other kind of mutex +** more than once, the behavior is undefined. +*/ +static void pthreadMutexEnter(sqlite3_mutex *p){ + assert( p->id==SQLITE_MUTEX_RECURSIVE || pthreadMutexNotheld(p) ); + +#ifdef SQLITE_HOMEGROWN_RECURSIVE_MUTEX + /* If recursive mutexes are not available, then we have to grow + ** our own. This implementation assumes that pthread_equal() + ** is atomic - that it cannot be deceived into thinking self + ** and p->owner are equal if p->owner changes between two values + ** that are not equal to self while the comparison is taking place. + ** This implementation also assumes a coherent cache - that + ** separate processes cannot read different values from the same + ** address at the same time. If either of these two conditions + ** are not met, then the mutexes will fail and problems will result. + */ + { + pthread_t self = pthread_self(); + if( p->nRef>0 && pthread_equal(p->owner, self) ){ + p->nRef++; + }else{ + pthread_mutex_lock(&p->mutex); + assert( p->nRef==0 ); + p->owner = self; + p->nRef = 1; + } + } +#else + /* Use the built-in recursive mutexes if they are available. + */ + pthread_mutex_lock(&p->mutex); +#if SQLITE_MUTEX_NREF + assert( p->nRef>0 || p->owner==0 ); + p->owner = pthread_self(); + p->nRef++; +#endif +#endif + +#ifdef SQLITE_DEBUG + if( p->trace ){ + printf("enter mutex %p (%d) with nRef=%d\n", p, p->trace, p->nRef); + } +#endif +} +static int pthreadMutexTry(sqlite3_mutex *p){ + int rc; + assert( p->id==SQLITE_MUTEX_RECURSIVE || pthreadMutexNotheld(p) ); + +#ifdef SQLITE_HOMEGROWN_RECURSIVE_MUTEX + /* If recursive mutexes are not available, then we have to grow + ** our own. This implementation assumes that pthread_equal() + ** is atomic - that it cannot be deceived into thinking self + ** and p->owner are equal if p->owner changes between two values + ** that are not equal to self while the comparison is taking place. + ** This implementation also assumes a coherent cache - that + ** separate processes cannot read different values from the same + ** address at the same time. If either of these two conditions + ** are not met, then the mutexes will fail and problems will result. + */ + { + pthread_t self = pthread_self(); + if( p->nRef>0 && pthread_equal(p->owner, self) ){ + p->nRef++; + rc = SQLITE_OK; + }else if( pthread_mutex_trylock(&p->mutex)==0 ){ + assert( p->nRef==0 ); + p->owner = self; + p->nRef = 1; + rc = SQLITE_OK; + }else{ + rc = SQLITE_BUSY; + } + } +#else + /* Use the built-in recursive mutexes if they are available. + */ + if( pthread_mutex_trylock(&p->mutex)==0 ){ +#if SQLITE_MUTEX_NREF + p->owner = pthread_self(); + p->nRef++; +#endif + rc = SQLITE_OK; + }else{ + rc = SQLITE_BUSY; + } +#endif + +#ifdef SQLITE_DEBUG + if( rc==SQLITE_OK && p->trace ){ + printf("enter mutex %p (%d) with nRef=%d\n", p, p->trace, p->nRef); + } +#endif + return rc; +} + +/* +** The sqlite3_mutex_leave() routine exits a mutex that was +** previously entered by the same thread. The behavior +** is undefined if the mutex is not currently entered or +** is not currently allocated. SQLite will never do either. +*/ +static void pthreadMutexLeave(sqlite3_mutex *p){ + assert( pthreadMutexHeld(p) ); +#if SQLITE_MUTEX_NREF + p->nRef--; + if( p->nRef==0 ) p->owner = 0; +#endif + assert( p->nRef==0 || p->id==SQLITE_MUTEX_RECURSIVE ); + +#ifdef SQLITE_HOMEGROWN_RECURSIVE_MUTEX + if( p->nRef==0 ){ + pthread_mutex_unlock(&p->mutex); + } +#else + pthread_mutex_unlock(&p->mutex); +#endif + +#ifdef SQLITE_DEBUG + if( p->trace ){ + printf("leave mutex %p (%d) with nRef=%d\n", p, p->trace, p->nRef); + } +#endif +} + +SQLITE_PRIVATE sqlite3_mutex_methods const *sqlite3DefaultMutex(void){ + static const sqlite3_mutex_methods sMutex = { + pthreadMutexInit, + pthreadMutexEnd, + pthreadMutexAlloc, + pthreadMutexFree, + pthreadMutexEnter, + pthreadMutexTry, + pthreadMutexLeave, +#ifdef SQLITE_DEBUG + pthreadMutexHeld, + pthreadMutexNotheld +#else + 0, + 0 +#endif + }; + + return &sMutex; +} + +#endif /* SQLITE_MUTEX_PTHREADS */ + +/************** End of mutex_unix.c ******************************************/ +/************** Begin file mutex_w32.c ***************************************/ +/* +** 2007 August 14 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** This file contains the C functions that implement mutexes for Win32. +*/ +/* #include "sqliteInt.h" */ + +#if SQLITE_OS_WIN +/* +** Include code that is common to all os_*.c files +*/ +/************** Include os_common.h in the middle of mutex_w32.c *************/ +/************** Begin file os_common.h ***************************************/ +/* +** 2004 May 22 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +****************************************************************************** +** +** This file contains macros and a little bit of code that is common to +** all of the platform-specific files (os_*.c) and is #included into those +** files. +** +** This file should be #included by the os_*.c files only. It is not a +** general purpose header file. +*/ +#ifndef _OS_COMMON_H_ +#define _OS_COMMON_H_ + +/* +** At least two bugs have slipped in because we changed the MEMORY_DEBUG +** macro to SQLITE_DEBUG and some older makefiles have not yet made the +** switch. The following code should catch this problem at compile-time. +*/ +#ifdef MEMORY_DEBUG +# error "The MEMORY_DEBUG macro is obsolete. Use SQLITE_DEBUG instead." +#endif + +/* +** Macros for performance tracing. Normally turned off. Only works +** on i486 hardware. +*/ +#ifdef SQLITE_PERFORMANCE_TRACE + +/* +** hwtime.h contains inline assembler code for implementing +** high-performance timing routines. +*/ +/************** Include hwtime.h in the middle of os_common.h ****************/ +/************** Begin file hwtime.h ******************************************/ +/* +** 2008 May 27 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +****************************************************************************** +** +** This file contains inline asm code for retrieving "high-performance" +** counters for x86 class CPUs. +*/ +#ifndef SQLITE_HWTIME_H +#define SQLITE_HWTIME_H + +/* +** The following routine only works on pentium-class (or newer) processors. +** It uses the RDTSC opcode to read the cycle count value out of the +** processor and returns that value. This can be used for high-res +** profiling. +*/ +#if (defined(__GNUC__) || defined(_MSC_VER)) && \ + (defined(i386) || defined(__i386__) || defined(_M_IX86)) + + #if defined(__GNUC__) + + __inline__ sqlite_uint64 sqlite3Hwtime(void){ + unsigned int lo, hi; + __asm__ __volatile__ ("rdtsc" : "=a" (lo), "=d" (hi)); + return (sqlite_uint64)hi << 32 | lo; + } + + #elif defined(_MSC_VER) + + __declspec(naked) __inline sqlite_uint64 __cdecl sqlite3Hwtime(void){ + __asm { + rdtsc + ret ; return value at EDX:EAX + } + } + + #endif + +#elif (defined(__GNUC__) && defined(__x86_64__)) + + __inline__ sqlite_uint64 sqlite3Hwtime(void){ + unsigned long val; + __asm__ __volatile__ ("rdtsc" : "=A" (val)); + return val; + } + +#elif (defined(__GNUC__) && defined(__ppc__)) + + __inline__ sqlite_uint64 sqlite3Hwtime(void){ + unsigned long long retval; + unsigned long junk; + __asm__ __volatile__ ("\n\ + 1: mftbu %1\n\ + mftb %L0\n\ + mftbu %0\n\ + cmpw %0,%1\n\ + bne 1b" + : "=r" (retval), "=r" (junk)); + return retval; + } + +#else + + #error Need implementation of sqlite3Hwtime() for your platform. + + /* + ** To compile without implementing sqlite3Hwtime() for your platform, + ** you can remove the above #error and use the following + ** stub function. You will lose timing support for many + ** of the debugging and testing utilities, but it should at + ** least compile and run. + */ +SQLITE_PRIVATE sqlite_uint64 sqlite3Hwtime(void){ return ((sqlite_uint64)0); } + +#endif + +#endif /* !defined(SQLITE_HWTIME_H) */ + +/************** End of hwtime.h **********************************************/ +/************** Continuing where we left off in os_common.h ******************/ + +static sqlite_uint64 g_start; +static sqlite_uint64 g_elapsed; +#define TIMER_START g_start=sqlite3Hwtime() +#define TIMER_END g_elapsed=sqlite3Hwtime()-g_start +#define TIMER_ELAPSED g_elapsed +#else +#define TIMER_START +#define TIMER_END +#define TIMER_ELAPSED ((sqlite_uint64)0) +#endif + +/* +** If we compile with the SQLITE_TEST macro set, then the following block +** of code will give us the ability to simulate a disk I/O error. This +** is used for testing the I/O recovery logic. +*/ +#if defined(SQLITE_TEST) +SQLITE_API extern int sqlite3_io_error_hit; +SQLITE_API extern int sqlite3_io_error_hardhit; +SQLITE_API extern int sqlite3_io_error_pending; +SQLITE_API extern int sqlite3_io_error_persist; +SQLITE_API extern int sqlite3_io_error_benign; +SQLITE_API extern int sqlite3_diskfull_pending; +SQLITE_API extern int sqlite3_diskfull; +#define SimulateIOErrorBenign(X) sqlite3_io_error_benign=(X) +#define SimulateIOError(CODE) \ + if( (sqlite3_io_error_persist && sqlite3_io_error_hit) \ + || sqlite3_io_error_pending-- == 1 ) \ + { local_ioerr(); CODE; } +static void local_ioerr(){ + IOTRACE(("IOERR\n")); + sqlite3_io_error_hit++; + if( !sqlite3_io_error_benign ) sqlite3_io_error_hardhit++; +} +#define SimulateDiskfullError(CODE) \ + if( sqlite3_diskfull_pending ){ \ + if( sqlite3_diskfull_pending == 1 ){ \ + local_ioerr(); \ + sqlite3_diskfull = 1; \ + sqlite3_io_error_hit = 1; \ + CODE; \ + }else{ \ + sqlite3_diskfull_pending--; \ + } \ + } +#else +#define SimulateIOErrorBenign(X) +#define SimulateIOError(A) +#define SimulateDiskfullError(A) +#endif /* defined(SQLITE_TEST) */ + +/* +** When testing, keep a count of the number of open files. +*/ +#if defined(SQLITE_TEST) +SQLITE_API extern int sqlite3_open_file_count; +#define OpenCounter(X) sqlite3_open_file_count+=(X) +#else +#define OpenCounter(X) +#endif /* defined(SQLITE_TEST) */ + +#endif /* !defined(_OS_COMMON_H_) */ + +/************** End of os_common.h *******************************************/ +/************** Continuing where we left off in mutex_w32.c ******************/ + +/* +** Include the header file for the Windows VFS. +*/ +/************** Include os_win.h in the middle of mutex_w32.c ****************/ +/************** Begin file os_win.h ******************************************/ +/* +** 2013 November 25 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +****************************************************************************** +** +** This file contains code that is specific to Windows. +*/ +#ifndef SQLITE_OS_WIN_H +#define SQLITE_OS_WIN_H + +/* +** Include the primary Windows SDK header file. +*/ +#include "windows.h" + +#ifdef __CYGWIN__ +# include +# include /* amalgamator: dontcache */ +#endif + +/* +** Determine if we are dealing with Windows NT. +** +** We ought to be able to determine if we are compiling for Windows 9x or +** Windows NT using the _WIN32_WINNT macro as follows: +** +** #if defined(_WIN32_WINNT) +** # define SQLITE_OS_WINNT 1 +** #else +** # define SQLITE_OS_WINNT 0 +** #endif +** +** However, Visual Studio 2005 does not set _WIN32_WINNT by default, as +** it ought to, so the above test does not work. We'll just assume that +** everything is Windows NT unless the programmer explicitly says otherwise +** by setting SQLITE_OS_WINNT to 0. +*/ +#if SQLITE_OS_WIN && !defined(SQLITE_OS_WINNT) +# define SQLITE_OS_WINNT 1 +#endif + +/* +** Determine if we are dealing with Windows CE - which has a much reduced +** API. +*/ +#if defined(_WIN32_WCE) +# define SQLITE_OS_WINCE 1 +#else +# define SQLITE_OS_WINCE 0 +#endif + +/* +** Determine if we are dealing with WinRT, which provides only a subset of +** the full Win32 API. +*/ +#if !defined(SQLITE_OS_WINRT) +# define SQLITE_OS_WINRT 0 +#endif + +/* +** For WinCE, some API function parameters do not appear to be declared as +** volatile. +*/ +#if SQLITE_OS_WINCE +# define SQLITE_WIN32_VOLATILE +#else +# define SQLITE_WIN32_VOLATILE volatile +#endif + +/* +** For some Windows sub-platforms, the _beginthreadex() / _endthreadex() +** functions are not available (e.g. those not using MSVC, Cygwin, etc). +*/ +#if SQLITE_OS_WIN && !SQLITE_OS_WINCE && !SQLITE_OS_WINRT && \ + SQLITE_THREADSAFE>0 && !defined(__CYGWIN__) +# define SQLITE_OS_WIN_THREADS 1 +#else +# define SQLITE_OS_WIN_THREADS 0 +#endif + +#endif /* SQLITE_OS_WIN_H */ + +/************** End of os_win.h **********************************************/ +/************** Continuing where we left off in mutex_w32.c ******************/ +#endif + +/* +** The code in this file is only used if we are compiling multithreaded +** on a Win32 system. +*/ +#ifdef SQLITE_MUTEX_W32 + +/* +** Each recursive mutex is an instance of the following structure. +*/ +struct sqlite3_mutex { + CRITICAL_SECTION mutex; /* Mutex controlling the lock */ + int id; /* Mutex type */ +#ifdef SQLITE_DEBUG + volatile int nRef; /* Number of enterances */ + volatile DWORD owner; /* Thread holding this mutex */ + volatile LONG trace; /* True to trace changes */ +#endif +}; + +/* +** These are the initializer values used when declaring a "static" mutex +** on Win32. It should be noted that all mutexes require initialization +** on the Win32 platform. +*/ +#define SQLITE_W32_MUTEX_INITIALIZER { 0 } + +#ifdef SQLITE_DEBUG +#define SQLITE3_MUTEX_INITIALIZER(id) { SQLITE_W32_MUTEX_INITIALIZER, id, \ + 0L, (DWORD)0, 0 } +#else +#define SQLITE3_MUTEX_INITIALIZER(id) { SQLITE_W32_MUTEX_INITIALIZER, id } +#endif + +#ifdef SQLITE_DEBUG +/* +** The sqlite3_mutex_held() and sqlite3_mutex_notheld() routine are +** intended for use only inside assert() statements. +*/ +static int winMutexHeld(sqlite3_mutex *p){ + return p->nRef!=0 && p->owner==GetCurrentThreadId(); +} + +static int winMutexNotheld2(sqlite3_mutex *p, DWORD tid){ + return p->nRef==0 || p->owner!=tid; +} + +static int winMutexNotheld(sqlite3_mutex *p){ + DWORD tid = GetCurrentThreadId(); + return winMutexNotheld2(p, tid); +} +#endif + +/* +** Try to provide a memory barrier operation, needed for initialization +** and also for the xShmBarrier method of the VFS in cases when SQLite is +** compiled without mutexes (SQLITE_THREADSAFE=0). +*/ +SQLITE_PRIVATE void sqlite3MemoryBarrier(void){ +#if defined(SQLITE_MEMORY_BARRIER) + SQLITE_MEMORY_BARRIER; +#elif defined(__GNUC__) + __sync_synchronize(); +#elif MSVC_VERSION>=1300 + _ReadWriteBarrier(); +#elif defined(MemoryBarrier) + MemoryBarrier(); +#endif +} + +/* +** Initialize and deinitialize the mutex subsystem. +*/ +static sqlite3_mutex winMutex_staticMutexes[] = { + SQLITE3_MUTEX_INITIALIZER(2), + SQLITE3_MUTEX_INITIALIZER(3), + SQLITE3_MUTEX_INITIALIZER(4), + SQLITE3_MUTEX_INITIALIZER(5), + SQLITE3_MUTEX_INITIALIZER(6), + SQLITE3_MUTEX_INITIALIZER(7), + SQLITE3_MUTEX_INITIALIZER(8), + SQLITE3_MUTEX_INITIALIZER(9), + SQLITE3_MUTEX_INITIALIZER(10), + SQLITE3_MUTEX_INITIALIZER(11), + SQLITE3_MUTEX_INITIALIZER(12), + SQLITE3_MUTEX_INITIALIZER(13) +}; + +static int winMutex_isInit = 0; +static int winMutex_isNt = -1; /* <0 means "need to query" */ + +/* As the winMutexInit() and winMutexEnd() functions are called as part +** of the sqlite3_initialize() and sqlite3_shutdown() processing, the +** "interlocked" magic used here is probably not strictly necessary. +*/ +static LONG SQLITE_WIN32_VOLATILE winMutex_lock = 0; + +SQLITE_API int sqlite3_win32_is_nt(void); /* os_win.c */ +SQLITE_API void sqlite3_win32_sleep(DWORD milliseconds); /* os_win.c */ + +static int winMutexInit(void){ + /* The first to increment to 1 does actual initialization */ + if( InterlockedCompareExchange(&winMutex_lock, 1, 0)==0 ){ + int i; + for(i=0; i +**
  • SQLITE_MUTEX_FAST +**
  • SQLITE_MUTEX_RECURSIVE +**
  • SQLITE_MUTEX_STATIC_MASTER +**
  • SQLITE_MUTEX_STATIC_MEM +**
  • SQLITE_MUTEX_STATIC_OPEN +**
  • SQLITE_MUTEX_STATIC_PRNG +**
  • SQLITE_MUTEX_STATIC_LRU +**
  • SQLITE_MUTEX_STATIC_PMEM +**
  • SQLITE_MUTEX_STATIC_APP1 +**
  • SQLITE_MUTEX_STATIC_APP2 +**
  • SQLITE_MUTEX_STATIC_APP3 +**
  • SQLITE_MUTEX_STATIC_VFS1 +**
  • SQLITE_MUTEX_STATIC_VFS2 +**
  • SQLITE_MUTEX_STATIC_VFS3 +** +** +** The first two constants cause sqlite3_mutex_alloc() to create +** a new mutex. The new mutex is recursive when SQLITE_MUTEX_RECURSIVE +** is used but not necessarily so when SQLITE_MUTEX_FAST is used. +** The mutex implementation does not need to make a distinction +** between SQLITE_MUTEX_RECURSIVE and SQLITE_MUTEX_FAST if it does +** not want to. But SQLite will only request a recursive mutex in +** cases where it really needs one. If a faster non-recursive mutex +** implementation is available on the host platform, the mutex subsystem +** might return such a mutex in response to SQLITE_MUTEX_FAST. +** +** The other allowed parameters to sqlite3_mutex_alloc() each return +** a pointer to a static preexisting mutex. Six static mutexes are +** used by the current version of SQLite. Future versions of SQLite +** may add additional static mutexes. Static mutexes are for internal +** use by SQLite only. Applications that use SQLite mutexes should +** use only the dynamic mutexes returned by SQLITE_MUTEX_FAST or +** SQLITE_MUTEX_RECURSIVE. +** +** Note that if one of the dynamic mutex parameters (SQLITE_MUTEX_FAST +** or SQLITE_MUTEX_RECURSIVE) is used then sqlite3_mutex_alloc() +** returns a different mutex on every call. But for the static +** mutex types, the same mutex is returned on every call that has +** the same type number. +*/ +static sqlite3_mutex *winMutexAlloc(int iType){ + sqlite3_mutex *p; + + switch( iType ){ + case SQLITE_MUTEX_FAST: + case SQLITE_MUTEX_RECURSIVE: { + p = sqlite3MallocZero( sizeof(*p) ); + if( p ){ + p->id = iType; +#ifdef SQLITE_DEBUG +#ifdef SQLITE_WIN32_MUTEX_TRACE_DYNAMIC + p->trace = 1; +#endif +#endif +#if SQLITE_OS_WINRT + InitializeCriticalSectionEx(&p->mutex, 0, 0); +#else + InitializeCriticalSection(&p->mutex); +#endif + } + break; + } + default: { +#ifdef SQLITE_ENABLE_API_ARMOR + if( iType-2<0 || iType-2>=ArraySize(winMutex_staticMutexes) ){ + (void)SQLITE_MISUSE_BKPT; + return 0; + } +#endif + p = &winMutex_staticMutexes[iType-2]; +#ifdef SQLITE_DEBUG +#ifdef SQLITE_WIN32_MUTEX_TRACE_STATIC + InterlockedCompareExchange(&p->trace, 1, 0); +#endif +#endif + break; + } + } + assert( p==0 || p->id==iType ); + return p; +} + + +/* +** This routine deallocates a previously +** allocated mutex. SQLite is careful to deallocate every +** mutex that it allocates. +*/ +static void winMutexFree(sqlite3_mutex *p){ + assert( p ); + assert( p->nRef==0 && p->owner==0 ); + if( p->id==SQLITE_MUTEX_FAST || p->id==SQLITE_MUTEX_RECURSIVE ){ + DeleteCriticalSection(&p->mutex); + sqlite3_free(p); + }else{ +#ifdef SQLITE_ENABLE_API_ARMOR + (void)SQLITE_MISUSE_BKPT; +#endif + } +} + +/* +** The sqlite3_mutex_enter() and sqlite3_mutex_try() routines attempt +** to enter a mutex. If another thread is already within the mutex, +** sqlite3_mutex_enter() will block and sqlite3_mutex_try() will return +** SQLITE_BUSY. The sqlite3_mutex_try() interface returns SQLITE_OK +** upon successful entry. Mutexes created using SQLITE_MUTEX_RECURSIVE can +** be entered multiple times by the same thread. In such cases the, +** mutex must be exited an equal number of times before another thread +** can enter. If the same thread tries to enter any other kind of mutex +** more than once, the behavior is undefined. +*/ +static void winMutexEnter(sqlite3_mutex *p){ +#if defined(SQLITE_DEBUG) || defined(SQLITE_TEST) + DWORD tid = GetCurrentThreadId(); +#endif +#ifdef SQLITE_DEBUG + assert( p ); + assert( p->id==SQLITE_MUTEX_RECURSIVE || winMutexNotheld2(p, tid) ); +#else + assert( p ); +#endif + assert( winMutex_isInit==1 ); + EnterCriticalSection(&p->mutex); +#ifdef SQLITE_DEBUG + assert( p->nRef>0 || p->owner==0 ); + p->owner = tid; + p->nRef++; + if( p->trace ){ + OSTRACE(("ENTER-MUTEX tid=%lu, mutex(%d)=%p (%d), nRef=%d\n", + tid, p->id, p, p->trace, p->nRef)); + } +#endif +} + +static int winMutexTry(sqlite3_mutex *p){ +#if defined(SQLITE_DEBUG) || defined(SQLITE_TEST) + DWORD tid = GetCurrentThreadId(); +#endif + int rc = SQLITE_BUSY; + assert( p ); + assert( p->id==SQLITE_MUTEX_RECURSIVE || winMutexNotheld2(p, tid) ); + /* + ** The sqlite3_mutex_try() routine is very rarely used, and when it + ** is used it is merely an optimization. So it is OK for it to always + ** fail. + ** + ** The TryEnterCriticalSection() interface is only available on WinNT. + ** And some windows compilers complain if you try to use it without + ** first doing some #defines that prevent SQLite from building on Win98. + ** For that reason, we will omit this optimization for now. See + ** ticket #2685. + */ +#if defined(_WIN32_WINNT) && _WIN32_WINNT >= 0x0400 + assert( winMutex_isInit==1 ); + assert( winMutex_isNt>=-1 && winMutex_isNt<=1 ); + if( winMutex_isNt<0 ){ + winMutex_isNt = sqlite3_win32_is_nt(); + } + assert( winMutex_isNt==0 || winMutex_isNt==1 ); + if( winMutex_isNt && TryEnterCriticalSection(&p->mutex) ){ +#ifdef SQLITE_DEBUG + p->owner = tid; + p->nRef++; +#endif + rc = SQLITE_OK; + } +#else + UNUSED_PARAMETER(p); +#endif +#ifdef SQLITE_DEBUG + if( p->trace ){ + OSTRACE(("TRY-MUTEX tid=%lu, mutex(%d)=%p (%d), owner=%lu, nRef=%d, rc=%s\n", + tid, p->id, p, p->trace, p->owner, p->nRef, sqlite3ErrName(rc))); + } +#endif + return rc; +} + +/* +** The sqlite3_mutex_leave() routine exits a mutex that was +** previously entered by the same thread. The behavior +** is undefined if the mutex is not currently entered or +** is not currently allocated. SQLite will never do either. +*/ +static void winMutexLeave(sqlite3_mutex *p){ +#if defined(SQLITE_DEBUG) || defined(SQLITE_TEST) + DWORD tid = GetCurrentThreadId(); +#endif + assert( p ); +#ifdef SQLITE_DEBUG + assert( p->nRef>0 ); + assert( p->owner==tid ); + p->nRef--; + if( p->nRef==0 ) p->owner = 0; + assert( p->nRef==0 || p->id==SQLITE_MUTEX_RECURSIVE ); +#endif + assert( winMutex_isInit==1 ); + LeaveCriticalSection(&p->mutex); +#ifdef SQLITE_DEBUG + if( p->trace ){ + OSTRACE(("LEAVE-MUTEX tid=%lu, mutex(%d)=%p (%d), nRef=%d\n", + tid, p->id, p, p->trace, p->nRef)); + } +#endif +} + +SQLITE_PRIVATE sqlite3_mutex_methods const *sqlite3DefaultMutex(void){ + static const sqlite3_mutex_methods sMutex = { + winMutexInit, + winMutexEnd, + winMutexAlloc, + winMutexFree, + winMutexEnter, + winMutexTry, + winMutexLeave, +#ifdef SQLITE_DEBUG + winMutexHeld, + winMutexNotheld +#else + 0, + 0 +#endif + }; + return &sMutex; +} + +#endif /* SQLITE_MUTEX_W32 */ + +/************** End of mutex_w32.c *******************************************/ +/************** Begin file malloc.c ******************************************/ +/* +** 2001 September 15 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** +** Memory allocation functions used throughout sqlite. +*/ +/* #include "sqliteInt.h" */ +/* #include */ + +/* +** Attempt to release up to n bytes of non-essential memory currently +** held by SQLite. An example of non-essential memory is memory used to +** cache database pages that are not currently in use. +*/ +SQLITE_API int sqlite3_release_memory(int n){ +#ifdef SQLITE_ENABLE_MEMORY_MANAGEMENT + return sqlite3PcacheReleaseMemory(n); +#else + /* IMPLEMENTATION-OF: R-34391-24921 The sqlite3_release_memory() routine + ** is a no-op returning zero if SQLite is not compiled with + ** SQLITE_ENABLE_MEMORY_MANAGEMENT. */ + UNUSED_PARAMETER(n); + return 0; +#endif +} + +/* +** State information local to the memory allocation subsystem. +*/ +static SQLITE_WSD struct Mem0Global { + sqlite3_mutex *mutex; /* Mutex to serialize access */ + sqlite3_int64 alarmThreshold; /* The soft heap limit */ + + /* + ** True if heap is nearly "full" where "full" is defined by the + ** sqlite3_soft_heap_limit() setting. + */ + int nearlyFull; +} mem0 = { 0, 0, 0 }; + +#define mem0 GLOBAL(struct Mem0Global, mem0) + +/* +** Return the memory allocator mutex. sqlite3_status() needs it. +*/ +SQLITE_PRIVATE sqlite3_mutex *sqlite3MallocMutex(void){ + return mem0.mutex; +} + +#ifndef SQLITE_OMIT_DEPRECATED +/* +** Deprecated external interface. It used to set an alarm callback +** that was invoked when memory usage grew too large. Now it is a +** no-op. +*/ +SQLITE_API int sqlite3_memory_alarm( + void(*xCallback)(void *pArg, sqlite3_int64 used,int N), + void *pArg, + sqlite3_int64 iThreshold +){ + (void)xCallback; + (void)pArg; + (void)iThreshold; + return SQLITE_OK; +} +#endif + +/* +** Set the soft heap-size limit for the library. Passing a zero or +** negative value indicates no limit. +*/ +SQLITE_API sqlite3_int64 sqlite3_soft_heap_limit64(sqlite3_int64 n){ + sqlite3_int64 priorLimit; + sqlite3_int64 excess; + sqlite3_int64 nUsed; +#ifndef SQLITE_OMIT_AUTOINIT + int rc = sqlite3_initialize(); + if( rc ) return -1; +#endif + sqlite3_mutex_enter(mem0.mutex); + priorLimit = mem0.alarmThreshold; + if( n<0 ){ + sqlite3_mutex_leave(mem0.mutex); + return priorLimit; + } + mem0.alarmThreshold = n; + nUsed = sqlite3StatusValue(SQLITE_STATUS_MEMORY_USED); + mem0.nearlyFull = (n>0 && n<=nUsed); + sqlite3_mutex_leave(mem0.mutex); + excess = sqlite3_memory_used() - n; + if( excess>0 ) sqlite3_release_memory((int)(excess & 0x7fffffff)); + return priorLimit; +} +SQLITE_API void sqlite3_soft_heap_limit(int n){ + if( n<0 ) n = 0; + sqlite3_soft_heap_limit64(n); +} + +/* +** Initialize the memory allocation subsystem. +*/ +SQLITE_PRIVATE int sqlite3MallocInit(void){ + int rc; + if( sqlite3GlobalConfig.m.xMalloc==0 ){ + sqlite3MemSetDefault(); + } + memset(&mem0, 0, sizeof(mem0)); + mem0.mutex = sqlite3MutexAlloc(SQLITE_MUTEX_STATIC_MEM); + if( sqlite3GlobalConfig.pPage==0 || sqlite3GlobalConfig.szPage<512 + || sqlite3GlobalConfig.nPage<=0 ){ + sqlite3GlobalConfig.pPage = 0; + sqlite3GlobalConfig.szPage = 0; + } + rc = sqlite3GlobalConfig.m.xInit(sqlite3GlobalConfig.m.pAppData); + if( rc!=SQLITE_OK ) memset(&mem0, 0, sizeof(mem0)); + return rc; +} + +/* +** Return true if the heap is currently under memory pressure - in other +** words if the amount of heap used is close to the limit set by +** sqlite3_soft_heap_limit(). +*/ +SQLITE_PRIVATE int sqlite3HeapNearlyFull(void){ + return mem0.nearlyFull; +} + +/* +** Deinitialize the memory allocation subsystem. +*/ +SQLITE_PRIVATE void sqlite3MallocEnd(void){ + if( sqlite3GlobalConfig.m.xShutdown ){ + sqlite3GlobalConfig.m.xShutdown(sqlite3GlobalConfig.m.pAppData); + } + memset(&mem0, 0, sizeof(mem0)); +} + +/* +** Return the amount of memory currently checked out. +*/ +SQLITE_API sqlite3_int64 sqlite3_memory_used(void){ + sqlite3_int64 res, mx; + sqlite3_status64(SQLITE_STATUS_MEMORY_USED, &res, &mx, 0); + return res; +} + +/* +** Return the maximum amount of memory that has ever been +** checked out since either the beginning of this process +** or since the most recent reset. +*/ +SQLITE_API sqlite3_int64 sqlite3_memory_highwater(int resetFlag){ + sqlite3_int64 res, mx; + sqlite3_status64(SQLITE_STATUS_MEMORY_USED, &res, &mx, resetFlag); + return mx; +} + +/* +** Trigger the alarm +*/ +static void sqlite3MallocAlarm(int nByte){ + if( mem0.alarmThreshold<=0 ) return; + sqlite3_mutex_leave(mem0.mutex); + sqlite3_release_memory(nByte); + sqlite3_mutex_enter(mem0.mutex); +} + +/* +** Do a memory allocation with statistics and alarms. Assume the +** lock is already held. +*/ +static void mallocWithAlarm(int n, void **pp){ + void *p; + int nFull; + assert( sqlite3_mutex_held(mem0.mutex) ); + assert( n>0 ); + + /* In Firefox (circa 2017-02-08), xRoundup() is remapped to an internal + ** implementation of malloc_good_size(), which must be called in debug + ** mode and specifically when the DMD "Dark Matter Detector" is enabled + ** or else a crash results. Hence, do not attempt to optimize out the + ** following xRoundup() call. */ + nFull = sqlite3GlobalConfig.m.xRoundup(n); + +#ifdef SQLITE_MAX_MEMORY + if( sqlite3StatusValue(SQLITE_STATUS_MEMORY_USED)+nFull>SQLITE_MAX_MEMORY ){ + *pp = 0; + return; + } +#endif + + sqlite3StatusHighwater(SQLITE_STATUS_MALLOC_SIZE, n); + if( mem0.alarmThreshold>0 ){ + sqlite3_int64 nUsed = sqlite3StatusValue(SQLITE_STATUS_MEMORY_USED); + if( nUsed >= mem0.alarmThreshold - nFull ){ + mem0.nearlyFull = 1; + sqlite3MallocAlarm(nFull); + }else{ + mem0.nearlyFull = 0; + } + } + p = sqlite3GlobalConfig.m.xMalloc(nFull); +#ifdef SQLITE_ENABLE_MEMORY_MANAGEMENT + if( p==0 && mem0.alarmThreshold>0 ){ + sqlite3MallocAlarm(nFull); + p = sqlite3GlobalConfig.m.xMalloc(nFull); + } +#endif + if( p ){ + nFull = sqlite3MallocSize(p); + sqlite3StatusUp(SQLITE_STATUS_MEMORY_USED, nFull); + sqlite3StatusUp(SQLITE_STATUS_MALLOC_COUNT, 1); + } + *pp = p; +} + +/* +** Allocate memory. This routine is like sqlite3_malloc() except that it +** assumes the memory subsystem has already been initialized. +*/ +SQLITE_PRIVATE void *sqlite3Malloc(u64 n){ + void *p; + if( n==0 || n>=0x7fffff00 ){ + /* A memory allocation of a number of bytes which is near the maximum + ** signed integer value might cause an integer overflow inside of the + ** xMalloc(). Hence we limit the maximum size to 0x7fffff00, giving + ** 255 bytes of overhead. SQLite itself will never use anything near + ** this amount. The only way to reach the limit is with sqlite3_malloc() */ + p = 0; + }else if( sqlite3GlobalConfig.bMemstat ){ + sqlite3_mutex_enter(mem0.mutex); + mallocWithAlarm((int)n, &p); + sqlite3_mutex_leave(mem0.mutex); + }else{ + p = sqlite3GlobalConfig.m.xMalloc((int)n); + } + assert( EIGHT_BYTE_ALIGNMENT(p) ); /* IMP: R-11148-40995 */ + return p; +} + +/* +** This version of the memory allocation is for use by the application. +** First make sure the memory subsystem is initialized, then do the +** allocation. +*/ +SQLITE_API void *sqlite3_malloc(int n){ +#ifndef SQLITE_OMIT_AUTOINIT + if( sqlite3_initialize() ) return 0; +#endif + return n<=0 ? 0 : sqlite3Malloc(n); +} +SQLITE_API void *sqlite3_malloc64(sqlite3_uint64 n){ +#ifndef SQLITE_OMIT_AUTOINIT + if( sqlite3_initialize() ) return 0; +#endif + return sqlite3Malloc(n); +} + +/* +** TRUE if p is a lookaside memory allocation from db +*/ +#ifndef SQLITE_OMIT_LOOKASIDE +static int isLookaside(sqlite3 *db, void *p){ + return SQLITE_WITHIN(p, db->lookaside.pStart, db->lookaside.pEnd); +} +#else +#define isLookaside(A,B) 0 +#endif + +/* +** Return the size of a memory allocation previously obtained from +** sqlite3Malloc() or sqlite3_malloc(). +*/ +SQLITE_PRIVATE int sqlite3MallocSize(void *p){ + assert( sqlite3MemdebugHasType(p, MEMTYPE_HEAP) ); + return sqlite3GlobalConfig.m.xSize(p); +} +SQLITE_PRIVATE int sqlite3DbMallocSize(sqlite3 *db, void *p){ + assert( p!=0 ); + if( db==0 || !isLookaside(db,p) ){ +#ifdef SQLITE_DEBUG + if( db==0 ){ + assert( sqlite3MemdebugNoType(p, (u8)~MEMTYPE_HEAP) ); + assert( sqlite3MemdebugHasType(p, MEMTYPE_HEAP) ); + }else{ + assert( sqlite3MemdebugHasType(p, (MEMTYPE_LOOKASIDE|MEMTYPE_HEAP)) ); + assert( sqlite3MemdebugNoType(p, (u8)~(MEMTYPE_LOOKASIDE|MEMTYPE_HEAP)) ); + } +#endif + return sqlite3GlobalConfig.m.xSize(p); + }else{ + assert( sqlite3_mutex_held(db->mutex) ); + return db->lookaside.sz; + } +} +SQLITE_API sqlite3_uint64 sqlite3_msize(void *p){ + assert( sqlite3MemdebugNoType(p, (u8)~MEMTYPE_HEAP) ); + assert( sqlite3MemdebugHasType(p, MEMTYPE_HEAP) ); + return p ? sqlite3GlobalConfig.m.xSize(p) : 0; +} + +/* +** Free memory previously obtained from sqlite3Malloc(). +*/ +SQLITE_API void sqlite3_free(void *p){ + if( p==0 ) return; /* IMP: R-49053-54554 */ + assert( sqlite3MemdebugHasType(p, MEMTYPE_HEAP) ); + assert( sqlite3MemdebugNoType(p, (u8)~MEMTYPE_HEAP) ); + if( sqlite3GlobalConfig.bMemstat ){ + sqlite3_mutex_enter(mem0.mutex); + sqlite3StatusDown(SQLITE_STATUS_MEMORY_USED, sqlite3MallocSize(p)); + sqlite3StatusDown(SQLITE_STATUS_MALLOC_COUNT, 1); + sqlite3GlobalConfig.m.xFree(p); + sqlite3_mutex_leave(mem0.mutex); + }else{ + sqlite3GlobalConfig.m.xFree(p); + } +} + +/* +** Add the size of memory allocation "p" to the count in +** *db->pnBytesFreed. +*/ +static SQLITE_NOINLINE void measureAllocationSize(sqlite3 *db, void *p){ + *db->pnBytesFreed += sqlite3DbMallocSize(db,p); +} + +/* +** Free memory that might be associated with a particular database +** connection. Calling sqlite3DbFree(D,X) for X==0 is a harmless no-op. +** The sqlite3DbFreeNN(D,X) version requires that X be non-NULL. +*/ +SQLITE_PRIVATE void sqlite3DbFreeNN(sqlite3 *db, void *p){ + assert( db==0 || sqlite3_mutex_held(db->mutex) ); + assert( p!=0 ); + if( db ){ + if( db->pnBytesFreed ){ + measureAllocationSize(db, p); + return; + } + if( isLookaside(db, p) ){ + LookasideSlot *pBuf = (LookasideSlot*)p; +#ifdef SQLITE_DEBUG + /* Trash all content in the buffer being freed */ + memset(p, 0xaa, db->lookaside.sz); +#endif + pBuf->pNext = db->lookaside.pFree; + db->lookaside.pFree = pBuf; + return; + } + } + assert( sqlite3MemdebugHasType(p, (MEMTYPE_LOOKASIDE|MEMTYPE_HEAP)) ); + assert( sqlite3MemdebugNoType(p, (u8)~(MEMTYPE_LOOKASIDE|MEMTYPE_HEAP)) ); + assert( db!=0 || sqlite3MemdebugNoType(p, MEMTYPE_LOOKASIDE) ); + sqlite3MemdebugSetType(p, MEMTYPE_HEAP); + sqlite3_free(p); +} +SQLITE_PRIVATE void sqlite3DbFree(sqlite3 *db, void *p){ + assert( db==0 || sqlite3_mutex_held(db->mutex) ); + if( p ) sqlite3DbFreeNN(db, p); +} + +/* +** Change the size of an existing memory allocation +*/ +SQLITE_PRIVATE void *sqlite3Realloc(void *pOld, u64 nBytes){ + int nOld, nNew, nDiff; + void *pNew; + assert( sqlite3MemdebugHasType(pOld, MEMTYPE_HEAP) ); + assert( sqlite3MemdebugNoType(pOld, (u8)~MEMTYPE_HEAP) ); + if( pOld==0 ){ + return sqlite3Malloc(nBytes); /* IMP: R-04300-56712 */ + } + if( nBytes==0 ){ + sqlite3_free(pOld); /* IMP: R-26507-47431 */ + return 0; + } + if( nBytes>=0x7fffff00 ){ + /* The 0x7ffff00 limit term is explained in comments on sqlite3Malloc() */ + return 0; + } + nOld = sqlite3MallocSize(pOld); + /* IMPLEMENTATION-OF: R-46199-30249 SQLite guarantees that the second + ** argument to xRealloc is always a value returned by a prior call to + ** xRoundup. */ + nNew = sqlite3GlobalConfig.m.xRoundup((int)nBytes); + if( nOld==nNew ){ + pNew = pOld; + }else if( sqlite3GlobalConfig.bMemstat ){ + sqlite3_mutex_enter(mem0.mutex); + sqlite3StatusHighwater(SQLITE_STATUS_MALLOC_SIZE, (int)nBytes); + nDiff = nNew - nOld; + if( nDiff>0 && sqlite3StatusValue(SQLITE_STATUS_MEMORY_USED) >= + mem0.alarmThreshold-nDiff ){ + sqlite3MallocAlarm(nDiff); + } + pNew = sqlite3GlobalConfig.m.xRealloc(pOld, nNew); + if( pNew==0 && mem0.alarmThreshold>0 ){ + sqlite3MallocAlarm((int)nBytes); + pNew = sqlite3GlobalConfig.m.xRealloc(pOld, nNew); + } + if( pNew ){ + nNew = sqlite3MallocSize(pNew); + sqlite3StatusUp(SQLITE_STATUS_MEMORY_USED, nNew-nOld); + } + sqlite3_mutex_leave(mem0.mutex); + }else{ + pNew = sqlite3GlobalConfig.m.xRealloc(pOld, nNew); + } + assert( EIGHT_BYTE_ALIGNMENT(pNew) ); /* IMP: R-11148-40995 */ + return pNew; +} + +/* +** The public interface to sqlite3Realloc. Make sure that the memory +** subsystem is initialized prior to invoking sqliteRealloc. +*/ +SQLITE_API void *sqlite3_realloc(void *pOld, int n){ +#ifndef SQLITE_OMIT_AUTOINIT + if( sqlite3_initialize() ) return 0; +#endif + if( n<0 ) n = 0; /* IMP: R-26507-47431 */ + return sqlite3Realloc(pOld, n); +} +SQLITE_API void *sqlite3_realloc64(void *pOld, sqlite3_uint64 n){ +#ifndef SQLITE_OMIT_AUTOINIT + if( sqlite3_initialize() ) return 0; +#endif + return sqlite3Realloc(pOld, n); +} + + +/* +** Allocate and zero memory. +*/ +SQLITE_PRIVATE void *sqlite3MallocZero(u64 n){ + void *p = sqlite3Malloc(n); + if( p ){ + memset(p, 0, (size_t)n); + } + return p; +} + +/* +** Allocate and zero memory. If the allocation fails, make +** the mallocFailed flag in the connection pointer. +*/ +SQLITE_PRIVATE void *sqlite3DbMallocZero(sqlite3 *db, u64 n){ + void *p; + testcase( db==0 ); + p = sqlite3DbMallocRaw(db, n); + if( p ) memset(p, 0, (size_t)n); + return p; +} + + +/* Finish the work of sqlite3DbMallocRawNN for the unusual and +** slower case when the allocation cannot be fulfilled using lookaside. +*/ +static SQLITE_NOINLINE void *dbMallocRawFinish(sqlite3 *db, u64 n){ + void *p; + assert( db!=0 ); + p = sqlite3Malloc(n); + if( !p ) sqlite3OomFault(db); + sqlite3MemdebugSetType(p, + (db->lookaside.bDisable==0) ? MEMTYPE_LOOKASIDE : MEMTYPE_HEAP); + return p; +} + +/* +** Allocate memory, either lookaside (if possible) or heap. +** If the allocation fails, set the mallocFailed flag in +** the connection pointer. +** +** If db!=0 and db->mallocFailed is true (indicating a prior malloc +** failure on the same database connection) then always return 0. +** Hence for a particular database connection, once malloc starts +** failing, it fails consistently until mallocFailed is reset. +** This is an important assumption. There are many places in the +** code that do things like this: +** +** int *a = (int*)sqlite3DbMallocRaw(db, 100); +** int *b = (int*)sqlite3DbMallocRaw(db, 200); +** if( b ) a[10] = 9; +** +** In other words, if a subsequent malloc (ex: "b") worked, it is assumed +** that all prior mallocs (ex: "a") worked too. +** +** The sqlite3MallocRawNN() variant guarantees that the "db" parameter is +** not a NULL pointer. +*/ +SQLITE_PRIVATE void *sqlite3DbMallocRaw(sqlite3 *db, u64 n){ + void *p; + if( db ) return sqlite3DbMallocRawNN(db, n); + p = sqlite3Malloc(n); + sqlite3MemdebugSetType(p, MEMTYPE_HEAP); + return p; +} +SQLITE_PRIVATE void *sqlite3DbMallocRawNN(sqlite3 *db, u64 n){ +#ifndef SQLITE_OMIT_LOOKASIDE + LookasideSlot *pBuf; + assert( db!=0 ); + assert( sqlite3_mutex_held(db->mutex) ); + assert( db->pnBytesFreed==0 ); + if( db->lookaside.bDisable==0 ){ + assert( db->mallocFailed==0 ); + if( n>db->lookaside.sz ){ + db->lookaside.anStat[1]++; + }else if( (pBuf = db->lookaside.pFree)!=0 ){ + db->lookaside.pFree = pBuf->pNext; + db->lookaside.anStat[0]++; + return (void*)pBuf; + }else if( (pBuf = db->lookaside.pInit)!=0 ){ + db->lookaside.pInit = pBuf->pNext; + db->lookaside.anStat[0]++; + return (void*)pBuf; + }else{ + db->lookaside.anStat[2]++; + } + }else if( db->mallocFailed ){ + return 0; + } +#else + assert( db!=0 ); + assert( sqlite3_mutex_held(db->mutex) ); + assert( db->pnBytesFreed==0 ); + if( db->mallocFailed ){ + return 0; + } +#endif + return dbMallocRawFinish(db, n); +} + +/* Forward declaration */ +static SQLITE_NOINLINE void *dbReallocFinish(sqlite3 *db, void *p, u64 n); + +/* +** Resize the block of memory pointed to by p to n bytes. If the +** resize fails, set the mallocFailed flag in the connection object. +*/ +SQLITE_PRIVATE void *sqlite3DbRealloc(sqlite3 *db, void *p, u64 n){ + assert( db!=0 ); + if( p==0 ) return sqlite3DbMallocRawNN(db, n); + assert( sqlite3_mutex_held(db->mutex) ); + if( isLookaside(db,p) && n<=db->lookaside.sz ) return p; + return dbReallocFinish(db, p, n); +} +static SQLITE_NOINLINE void *dbReallocFinish(sqlite3 *db, void *p, u64 n){ + void *pNew = 0; + assert( db!=0 ); + assert( p!=0 ); + if( db->mallocFailed==0 ){ + if( isLookaside(db, p) ){ + pNew = sqlite3DbMallocRawNN(db, n); + if( pNew ){ + memcpy(pNew, p, db->lookaside.sz); + sqlite3DbFree(db, p); + } + }else{ + assert( sqlite3MemdebugHasType(p, (MEMTYPE_LOOKASIDE|MEMTYPE_HEAP)) ); + assert( sqlite3MemdebugNoType(p, (u8)~(MEMTYPE_LOOKASIDE|MEMTYPE_HEAP)) ); + sqlite3MemdebugSetType(p, MEMTYPE_HEAP); + pNew = sqlite3_realloc64(p, n); + if( !pNew ){ + sqlite3OomFault(db); + } + sqlite3MemdebugSetType(pNew, + (db->lookaside.bDisable==0 ? MEMTYPE_LOOKASIDE : MEMTYPE_HEAP)); + } + } + return pNew; +} + +/* +** Attempt to reallocate p. If the reallocation fails, then free p +** and set the mallocFailed flag in the database connection. +*/ +SQLITE_PRIVATE void *sqlite3DbReallocOrFree(sqlite3 *db, void *p, u64 n){ + void *pNew; + pNew = sqlite3DbRealloc(db, p, n); + if( !pNew ){ + sqlite3DbFree(db, p); + } + return pNew; +} + +/* +** Make a copy of a string in memory obtained from sqliteMalloc(). These +** functions call sqlite3MallocRaw() directly instead of sqliteMalloc(). This +** is because when memory debugging is turned on, these two functions are +** called via macros that record the current file and line number in the +** ThreadData structure. +*/ +SQLITE_PRIVATE char *sqlite3DbStrDup(sqlite3 *db, const char *z){ + char *zNew; + size_t n; + if( z==0 ){ + return 0; + } + n = strlen(z) + 1; + zNew = sqlite3DbMallocRaw(db, n); + if( zNew ){ + memcpy(zNew, z, n); + } + return zNew; +} +SQLITE_PRIVATE char *sqlite3DbStrNDup(sqlite3 *db, const char *z, u64 n){ + char *zNew; + assert( db!=0 ); + if( z==0 ){ + return 0; + } + assert( (n&0x7fffffff)==n ); + zNew = sqlite3DbMallocRawNN(db, n+1); + if( zNew ){ + memcpy(zNew, z, (size_t)n); + zNew[n] = 0; + } + return zNew; +} + +/* +** The text between zStart and zEnd represents a phrase within a larger +** SQL statement. Make a copy of this phrase in space obtained form +** sqlite3DbMalloc(). Omit leading and trailing whitespace. +*/ +SQLITE_PRIVATE char *sqlite3DbSpanDup(sqlite3 *db, const char *zStart, const char *zEnd){ + int n; + while( sqlite3Isspace(zStart[0]) ) zStart++; + n = (int)(zEnd - zStart); + while( ALWAYS(n>0) && sqlite3Isspace(zStart[n-1]) ) n--; + return sqlite3DbStrNDup(db, zStart, n); +} + +/* +** Free any prior content in *pz and replace it with a copy of zNew. +*/ +SQLITE_PRIVATE void sqlite3SetString(char **pz, sqlite3 *db, const char *zNew){ + sqlite3DbFree(db, *pz); + *pz = sqlite3DbStrDup(db, zNew); +} + +/* +** Call this routine to record the fact that an OOM (out-of-memory) error +** has happened. This routine will set db->mallocFailed, and also +** temporarily disable the lookaside memory allocator and interrupt +** any running VDBEs. +*/ +SQLITE_PRIVATE void sqlite3OomFault(sqlite3 *db){ + if( db->mallocFailed==0 && db->bBenignMalloc==0 ){ + db->mallocFailed = 1; + if( db->nVdbeExec>0 ){ + db->u1.isInterrupted = 1; + } + db->lookaside.bDisable++; + if( db->pParse ){ + db->pParse->rc = SQLITE_NOMEM_BKPT; + } + } +} + +/* +** This routine reactivates the memory allocator and clears the +** db->mallocFailed flag as necessary. +** +** The memory allocator is not restarted if there are running +** VDBEs. +*/ +SQLITE_PRIVATE void sqlite3OomClear(sqlite3 *db){ + if( db->mallocFailed && db->nVdbeExec==0 ){ + db->mallocFailed = 0; + db->u1.isInterrupted = 0; + assert( db->lookaside.bDisable>0 ); + db->lookaside.bDisable--; + } +} + +/* +** Take actions at the end of an API call to indicate an OOM error +*/ +static SQLITE_NOINLINE int apiOomError(sqlite3 *db){ + sqlite3OomClear(db); + sqlite3Error(db, SQLITE_NOMEM); + return SQLITE_NOMEM_BKPT; +} + +/* +** This function must be called before exiting any API function (i.e. +** returning control to the user) that has called sqlite3_malloc or +** sqlite3_realloc. +** +** The returned value is normally a copy of the second argument to this +** function. However, if a malloc() failure has occurred since the previous +** invocation SQLITE_NOMEM is returned instead. +** +** If an OOM as occurred, then the connection error-code (the value +** returned by sqlite3_errcode()) is set to SQLITE_NOMEM. +*/ +SQLITE_PRIVATE int sqlite3ApiExit(sqlite3* db, int rc){ + /* If the db handle must hold the connection handle mutex here. + ** Otherwise the read (and possible write) of db->mallocFailed + ** is unsafe, as is the call to sqlite3Error(). + */ + assert( db!=0 ); + assert( sqlite3_mutex_held(db->mutex) ); + if( db->mallocFailed || rc==SQLITE_IOERR_NOMEM ){ + return apiOomError(db); + } + return rc & db->errMask; +} + +/************** End of malloc.c **********************************************/ +/************** Begin file printf.c ******************************************/ +/* +** The "printf" code that follows dates from the 1980's. It is in +** the public domain. +** +************************************************************************** +** +** This file contains code for a set of "printf"-like routines. These +** routines format strings much like the printf() from the standard C +** library, though the implementation here has enhancements to support +** SQLite. +*/ +/* #include "sqliteInt.h" */ + +/* +** Conversion types fall into various categories as defined by the +** following enumeration. +*/ +#define etRADIX 0 /* non-decimal integer types. %x %o */ +#define etFLOAT 1 /* Floating point. %f */ +#define etEXP 2 /* Exponentional notation. %e and %E */ +#define etGENERIC 3 /* Floating or exponential, depending on exponent. %g */ +#define etSIZE 4 /* Return number of characters processed so far. %n */ +#define etSTRING 5 /* Strings. %s */ +#define etDYNSTRING 6 /* Dynamically allocated strings. %z */ +#define etPERCENT 7 /* Percent symbol. %% */ +#define etCHARX 8 /* Characters. %c */ +/* The rest are extensions, not normally found in printf() */ +#define etSQLESCAPE 9 /* Strings with '\'' doubled. %q */ +#define etSQLESCAPE2 10 /* Strings with '\'' doubled and enclosed in '', + NULL pointers replaced by SQL NULL. %Q */ +#define etTOKEN 11 /* a pointer to a Token structure */ +#define etSRCLIST 12 /* a pointer to a SrcList */ +#define etPOINTER 13 /* The %p conversion */ +#define etSQLESCAPE3 14 /* %w -> Strings with '\"' doubled */ +#define etORDINAL 15 /* %r -> 1st, 2nd, 3rd, 4th, etc. English only */ +#define etDECIMAL 16 /* %d or %u, but not %x, %o */ + +#define etINVALID 17 /* Any unrecognized conversion type */ + + +/* +** An "etByte" is an 8-bit unsigned value. +*/ +typedef unsigned char etByte; + +/* +** Each builtin conversion character (ex: the 'd' in "%d") is described +** by an instance of the following structure +*/ +typedef struct et_info { /* Information about each format field */ + char fmttype; /* The format field code letter */ + etByte base; /* The base for radix conversion */ + etByte flags; /* One or more of FLAG_ constants below */ + etByte type; /* Conversion paradigm */ + etByte charset; /* Offset into aDigits[] of the digits string */ + etByte prefix; /* Offset into aPrefix[] of the prefix string */ +} et_info; + +/* +** Allowed values for et_info.flags +*/ +#define FLAG_SIGNED 1 /* True if the value to convert is signed */ +#define FLAG_STRING 4 /* Allow infinite precision */ + + +/* +** The following table is searched linearly, so it is good to put the +** most frequently used conversion types first. +*/ +static const char aDigits[] = "0123456789ABCDEF0123456789abcdef"; +static const char aPrefix[] = "-x0\000X0"; +static const et_info fmtinfo[] = { + { 'd', 10, 1, etDECIMAL, 0, 0 }, + { 's', 0, 4, etSTRING, 0, 0 }, + { 'g', 0, 1, etGENERIC, 30, 0 }, + { 'z', 0, 4, etDYNSTRING, 0, 0 }, + { 'q', 0, 4, etSQLESCAPE, 0, 0 }, + { 'Q', 0, 4, etSQLESCAPE2, 0, 0 }, + { 'w', 0, 4, etSQLESCAPE3, 0, 0 }, + { 'c', 0, 0, etCHARX, 0, 0 }, + { 'o', 8, 0, etRADIX, 0, 2 }, + { 'u', 10, 0, etDECIMAL, 0, 0 }, + { 'x', 16, 0, etRADIX, 16, 1 }, + { 'X', 16, 0, etRADIX, 0, 4 }, +#ifndef SQLITE_OMIT_FLOATING_POINT + { 'f', 0, 1, etFLOAT, 0, 0 }, + { 'e', 0, 1, etEXP, 30, 0 }, + { 'E', 0, 1, etEXP, 14, 0 }, + { 'G', 0, 1, etGENERIC, 14, 0 }, +#endif + { 'i', 10, 1, etDECIMAL, 0, 0 }, + { 'n', 0, 0, etSIZE, 0, 0 }, + { '%', 0, 0, etPERCENT, 0, 0 }, + { 'p', 16, 0, etPOINTER, 0, 1 }, + + /* All the rest are undocumented and are for internal use only */ + { 'T', 0, 0, etTOKEN, 0, 0 }, + { 'S', 0, 0, etSRCLIST, 0, 0 }, + { 'r', 10, 1, etORDINAL, 0, 0 }, +}; + +/* Floating point constants used for rounding */ +static const double arRound[] = { + 5.0e-01, 5.0e-02, 5.0e-03, 5.0e-04, 5.0e-05, + 5.0e-06, 5.0e-07, 5.0e-08, 5.0e-09, 5.0e-10, +}; + +/* +** If SQLITE_OMIT_FLOATING_POINT is defined, then none of the floating point +** conversions will work. +*/ +#ifndef SQLITE_OMIT_FLOATING_POINT +/* +** "*val" is a double such that 0.1 <= *val < 10.0 +** Return the ascii code for the leading digit of *val, then +** multiply "*val" by 10.0 to renormalize. +** +** Example: +** input: *val = 3.14159 +** output: *val = 1.4159 function return = '3' +** +** The counter *cnt is incremented each time. After counter exceeds +** 16 (the number of significant digits in a 64-bit float) '0' is +** always returned. +*/ +static char et_getdigit(LONGDOUBLE_TYPE *val, int *cnt){ + int digit; + LONGDOUBLE_TYPE d; + if( (*cnt)<=0 ) return '0'; + (*cnt)--; + digit = (int)*val; + d = digit; + digit += '0'; + *val = (*val - d)*10.0; + return (char)digit; +} +#endif /* SQLITE_OMIT_FLOATING_POINT */ + +/* +** Set the StrAccum object to an error mode. +*/ +static void setStrAccumError(StrAccum *p, u8 eError){ + assert( eError==SQLITE_NOMEM || eError==SQLITE_TOOBIG ); + p->accError = eError; + if( p->mxAlloc ) sqlite3_str_reset(p); + if( eError==SQLITE_TOOBIG ) sqlite3ErrorToParser(p->db, eError); +} + +/* +** Extra argument values from a PrintfArguments object +*/ +static sqlite3_int64 getIntArg(PrintfArguments *p){ + if( p->nArg<=p->nUsed ) return 0; + return sqlite3_value_int64(p->apArg[p->nUsed++]); +} +static double getDoubleArg(PrintfArguments *p){ + if( p->nArg<=p->nUsed ) return 0.0; + return sqlite3_value_double(p->apArg[p->nUsed++]); +} +static char *getTextArg(PrintfArguments *p){ + if( p->nArg<=p->nUsed ) return 0; + return (char*)sqlite3_value_text(p->apArg[p->nUsed++]); +} + +/* +** Allocate memory for a temporary buffer needed for printf rendering. +** +** If the requested size of the temp buffer is larger than the size +** of the output buffer in pAccum, then cause an SQLITE_TOOBIG error. +** Do the size check before the memory allocation to prevent rogue +** SQL from requesting large allocations using the precision or width +** field of the printf() function. +*/ +static char *printfTempBuf(sqlite3_str *pAccum, sqlite3_int64 n){ + char *z; + if( pAccum->accError ) return 0; + if( n>pAccum->nAlloc && n>pAccum->mxAlloc ){ + setStrAccumError(pAccum, SQLITE_TOOBIG); + return 0; + } + z = sqlite3DbMallocRaw(pAccum->db, n); + if( z==0 ){ + setStrAccumError(pAccum, SQLITE_NOMEM); + } + return z; +} + +/* +** On machines with a small stack size, you can redefine the +** SQLITE_PRINT_BUF_SIZE to be something smaller, if desired. +*/ +#ifndef SQLITE_PRINT_BUF_SIZE +# define SQLITE_PRINT_BUF_SIZE 70 +#endif +#define etBUFSIZE SQLITE_PRINT_BUF_SIZE /* Size of the output buffer */ + +/* +** Render a string given by "fmt" into the StrAccum object. +*/ +SQLITE_API void sqlite3_str_vappendf( + sqlite3_str *pAccum, /* Accumulate results here */ + const char *fmt, /* Format string */ + va_list ap /* arguments */ +){ + int c; /* Next character in the format string */ + char *bufpt; /* Pointer to the conversion buffer */ + int precision; /* Precision of the current field */ + int length; /* Length of the field */ + int idx; /* A general purpose loop counter */ + int width; /* Width of the current field */ + etByte flag_leftjustify; /* True if "-" flag is present */ + etByte flag_prefix; /* '+' or ' ' or 0 for prefix */ + etByte flag_alternateform; /* True if "#" flag is present */ + etByte flag_altform2; /* True if "!" flag is present */ + etByte flag_zeropad; /* True if field width constant starts with zero */ + etByte flag_long; /* 1 for the "l" flag, 2 for "ll", 0 by default */ + etByte done; /* Loop termination flag */ + etByte cThousand; /* Thousands separator for %d and %u */ + etByte xtype = etINVALID; /* Conversion paradigm */ + u8 bArgList; /* True for SQLITE_PRINTF_SQLFUNC */ + char prefix; /* Prefix character. "+" or "-" or " " or '\0'. */ + sqlite_uint64 longvalue; /* Value for integer types */ + LONGDOUBLE_TYPE realvalue; /* Value for real types */ + const et_info *infop; /* Pointer to the appropriate info structure */ + char *zOut; /* Rendering buffer */ + int nOut; /* Size of the rendering buffer */ + char *zExtra = 0; /* Malloced memory used by some conversion */ +#ifndef SQLITE_OMIT_FLOATING_POINT + int exp, e2; /* exponent of real numbers */ + int nsd; /* Number of significant digits returned */ + double rounder; /* Used for rounding floating point values */ + etByte flag_dp; /* True if decimal point should be shown */ + etByte flag_rtz; /* True if trailing zeros should be removed */ +#endif + PrintfArguments *pArgList = 0; /* Arguments for SQLITE_PRINTF_SQLFUNC */ + char buf[etBUFSIZE]; /* Conversion buffer */ + + /* pAccum never starts out with an empty buffer that was obtained from + ** malloc(). This precondition is required by the mprintf("%z...") + ** optimization. */ + assert( pAccum->nChar>0 || (pAccum->printfFlags&SQLITE_PRINTF_MALLOCED)==0 ); + + bufpt = 0; + if( (pAccum->printfFlags & SQLITE_PRINTF_SQLFUNC)!=0 ){ + pArgList = va_arg(ap, PrintfArguments*); + bArgList = 1; + }else{ + bArgList = 0; + } + for(; (c=(*fmt))!=0; ++fmt){ + if( c!='%' ){ + bufpt = (char *)fmt; +#if HAVE_STRCHRNUL + fmt = strchrnul(fmt, '%'); +#else + do{ fmt++; }while( *fmt && *fmt != '%' ); +#endif + sqlite3_str_append(pAccum, bufpt, (int)(fmt - bufpt)); + if( *fmt==0 ) break; + } + if( (c=(*++fmt))==0 ){ + sqlite3_str_append(pAccum, "%", 1); + break; + } + /* Find out what flags are present */ + flag_leftjustify = flag_prefix = cThousand = + flag_alternateform = flag_altform2 = flag_zeropad = 0; + done = 0; + width = 0; + flag_long = 0; + precision = -1; + do{ + switch( c ){ + case '-': flag_leftjustify = 1; break; + case '+': flag_prefix = '+'; break; + case ' ': flag_prefix = ' '; break; + case '#': flag_alternateform = 1; break; + case '!': flag_altform2 = 1; break; + case '0': flag_zeropad = 1; break; + case ',': cThousand = ','; break; + default: done = 1; break; + case 'l': { + flag_long = 1; + c = *++fmt; + if( c=='l' ){ + c = *++fmt; + flag_long = 2; + } + done = 1; + break; + } + case '1': case '2': case '3': case '4': case '5': + case '6': case '7': case '8': case '9': { + unsigned wx = c - '0'; + while( (c = *++fmt)>='0' && c<='9' ){ + wx = wx*10 + c - '0'; + } + testcase( wx>0x7fffffff ); + width = wx & 0x7fffffff; +#ifdef SQLITE_PRINTF_PRECISION_LIMIT + if( width>SQLITE_PRINTF_PRECISION_LIMIT ){ + width = SQLITE_PRINTF_PRECISION_LIMIT; + } +#endif + if( c!='.' && c!='l' ){ + done = 1; + }else{ + fmt--; + } + break; + } + case '*': { + if( bArgList ){ + width = (int)getIntArg(pArgList); + }else{ + width = va_arg(ap,int); + } + if( width<0 ){ + flag_leftjustify = 1; + width = width >= -2147483647 ? -width : 0; + } +#ifdef SQLITE_PRINTF_PRECISION_LIMIT + if( width>SQLITE_PRINTF_PRECISION_LIMIT ){ + width = SQLITE_PRINTF_PRECISION_LIMIT; + } +#endif + if( (c = fmt[1])!='.' && c!='l' ){ + c = *++fmt; + done = 1; + } + break; + } + case '.': { + c = *++fmt; + if( c=='*' ){ + if( bArgList ){ + precision = (int)getIntArg(pArgList); + }else{ + precision = va_arg(ap,int); + } + if( precision<0 ){ + precision = precision >= -2147483647 ? -precision : -1; + } + c = *++fmt; + }else{ + unsigned px = 0; + while( c>='0' && c<='9' ){ + px = px*10 + c - '0'; + c = *++fmt; + } + testcase( px>0x7fffffff ); + precision = px & 0x7fffffff; + } +#ifdef SQLITE_PRINTF_PRECISION_LIMIT + if( precision>SQLITE_PRINTF_PRECISION_LIMIT ){ + precision = SQLITE_PRINTF_PRECISION_LIMIT; + } +#endif + if( c=='l' ){ + --fmt; + }else{ + done = 1; + } + break; + } + } + }while( !done && (c=(*++fmt))!=0 ); + + /* Fetch the info entry for the field */ + infop = &fmtinfo[0]; + xtype = etINVALID; + for(idx=0; idxtype; + break; + } + } + + /* + ** At this point, variables are initialized as follows: + ** + ** flag_alternateform TRUE if a '#' is present. + ** flag_altform2 TRUE if a '!' is present. + ** flag_prefix '+' or ' ' or zero + ** flag_leftjustify TRUE if a '-' is present or if the + ** field width was negative. + ** flag_zeropad TRUE if the width began with 0. + ** flag_long 1 for "l", 2 for "ll" + ** width The specified field width. This is + ** always non-negative. Zero is the default. + ** precision The specified precision. The default + ** is -1. + ** xtype The class of the conversion. + ** infop Pointer to the appropriate info struct. + */ + switch( xtype ){ + case etPOINTER: + flag_long = sizeof(char*)==sizeof(i64) ? 2 : + sizeof(char*)==sizeof(long int) ? 1 : 0; + /* Fall through into the next case */ + case etORDINAL: + case etRADIX: + cThousand = 0; + /* Fall through into the next case */ + case etDECIMAL: + if( infop->flags & FLAG_SIGNED ){ + i64 v; + if( bArgList ){ + v = getIntArg(pArgList); + }else if( flag_long ){ + if( flag_long==2 ){ + v = va_arg(ap,i64) ; + }else{ + v = va_arg(ap,long int); + } + }else{ + v = va_arg(ap,int); + } + if( v<0 ){ + if( v==SMALLEST_INT64 ){ + longvalue = ((u64)1)<<63; + }else{ + longvalue = -v; + } + prefix = '-'; + }else{ + longvalue = v; + prefix = flag_prefix; + } + }else{ + if( bArgList ){ + longvalue = (u64)getIntArg(pArgList); + }else if( flag_long ){ + if( flag_long==2 ){ + longvalue = va_arg(ap,u64); + }else{ + longvalue = va_arg(ap,unsigned long int); + } + }else{ + longvalue = va_arg(ap,unsigned int); + } + prefix = 0; + } + if( longvalue==0 ) flag_alternateform = 0; + if( flag_zeropad && precision=4 || (longvalue/10)%10==1 ){ + x = 0; + } + *(--bufpt) = zOrd[x*2+1]; + *(--bufpt) = zOrd[x*2]; + } + { + const char *cset = &aDigits[infop->charset]; + u8 base = infop->base; + do{ /* Convert to ascii */ + *(--bufpt) = cset[longvalue%base]; + longvalue = longvalue/base; + }while( longvalue>0 ); + } + length = (int)(&zOut[nOut-1]-bufpt); + while( precision>length ){ + *(--bufpt) = '0'; /* Zero pad */ + length++; + } + if( cThousand ){ + int nn = (length - 1)/3; /* Number of "," to insert */ + int ix = (length - 1)%3 + 1; + bufpt -= nn; + for(idx=0; nn>0; idx++){ + bufpt[idx] = bufpt[idx+nn]; + ix--; + if( ix==0 ){ + bufpt[++idx] = cThousand; + nn--; + ix = 3; + } + } + } + if( prefix ) *(--bufpt) = prefix; /* Add sign */ + if( flag_alternateform && infop->prefix ){ /* Add "0" or "0x" */ + const char *pre; + char x; + pre = &aPrefix[infop->prefix]; + for(; (x=(*pre))!=0; pre++) *(--bufpt) = x; + } + length = (int)(&zOut[nOut-1]-bufpt); + break; + case etFLOAT: + case etEXP: + case etGENERIC: + if( bArgList ){ + realvalue = getDoubleArg(pArgList); + }else{ + realvalue = va_arg(ap,double); + } +#ifdef SQLITE_OMIT_FLOATING_POINT + length = 0; +#else + if( precision<0 ) precision = 6; /* Set default precision */ + if( realvalue<0.0 ){ + realvalue = -realvalue; + prefix = '-'; + }else{ + prefix = flag_prefix; + } + if( xtype==etGENERIC && precision>0 ) precision--; + testcase( precision>0xfff ); + idx = precision & 0xfff; + rounder = arRound[idx%10]; + while( idx>=10 ){ rounder *= 1.0e-10; idx -= 10; } + if( xtype==etFLOAT ){ + double rx = (double)realvalue; + sqlite3_uint64 u; + int ex; + memcpy(&u, &rx, sizeof(u)); + ex = -1023 + (int)((u>>52)&0x7ff); + if( precision+(ex/3) < 15 ) rounder += realvalue*3e-16; + realvalue += rounder; + } + /* Normalize realvalue to within 10.0 > realvalue >= 1.0 */ + exp = 0; + if( sqlite3IsNaN((double)realvalue) ){ + bufpt = "NaN"; + length = 3; + break; + } + if( realvalue>0.0 ){ + LONGDOUBLE_TYPE scale = 1.0; + while( realvalue>=1e100*scale && exp<=350 ){ scale *= 1e100;exp+=100;} + while( realvalue>=1e10*scale && exp<=350 ){ scale *= 1e10; exp+=10; } + while( realvalue>=10.0*scale && exp<=350 ){ scale *= 10.0; exp++; } + realvalue /= scale; + while( realvalue<1e-8 ){ realvalue *= 1e8; exp-=8; } + while( realvalue<1.0 ){ realvalue *= 10.0; exp--; } + if( exp>350 ){ + bufpt = buf; + buf[0] = prefix; + memcpy(buf+(prefix!=0),"Inf",4); + length = 3+(prefix!=0); + break; + } + } + bufpt = buf; + /* + ** If the field type is etGENERIC, then convert to either etEXP + ** or etFLOAT, as appropriate. + */ + if( xtype!=etFLOAT ){ + realvalue += rounder; + if( realvalue>=10.0 ){ realvalue *= 0.1; exp++; } + } + if( xtype==etGENERIC ){ + flag_rtz = !flag_alternateform; + if( exp<-4 || exp>precision ){ + xtype = etEXP; + }else{ + precision = precision - exp; + xtype = etFLOAT; + } + }else{ + flag_rtz = flag_altform2; + } + if( xtype==etEXP ){ + e2 = 0; + }else{ + e2 = exp; + } + { + i64 szBufNeeded; /* Size of a temporary buffer needed */ + szBufNeeded = MAX(e2,0)+(i64)precision+(i64)width+15; + if( szBufNeeded > etBUFSIZE ){ + bufpt = zExtra = printfTempBuf(pAccum, szBufNeeded); + if( bufpt==0 ) return; + } + } + zOut = bufpt; + nsd = 16 + flag_altform2*10; + flag_dp = (precision>0 ?1:0) | flag_alternateform | flag_altform2; + /* The sign in front of the number */ + if( prefix ){ + *(bufpt++) = prefix; + } + /* Digits prior to the decimal point */ + if( e2<0 ){ + *(bufpt++) = '0'; + }else{ + for(; e2>=0; e2--){ + *(bufpt++) = et_getdigit(&realvalue,&nsd); + } + } + /* The decimal point */ + if( flag_dp ){ + *(bufpt++) = '.'; + } + /* "0" digits after the decimal point but before the first + ** significant digit of the number */ + for(e2++; e2<0; precision--, e2++){ + assert( precision>0 ); + *(bufpt++) = '0'; + } + /* Significant digits after the decimal point */ + while( (precision--)>0 ){ + *(bufpt++) = et_getdigit(&realvalue,&nsd); + } + /* Remove trailing zeros and the "." if no digits follow the "." */ + if( flag_rtz && flag_dp ){ + while( bufpt[-1]=='0' ) *(--bufpt) = 0; + assert( bufpt>zOut ); + if( bufpt[-1]=='.' ){ + if( flag_altform2 ){ + *(bufpt++) = '0'; + }else{ + *(--bufpt) = 0; + } + } + } + /* Add the "eNNN" suffix */ + if( xtype==etEXP ){ + *(bufpt++) = aDigits[infop->charset]; + if( exp<0 ){ + *(bufpt++) = '-'; exp = -exp; + }else{ + *(bufpt++) = '+'; + } + if( exp>=100 ){ + *(bufpt++) = (char)((exp/100)+'0'); /* 100's digit */ + exp %= 100; + } + *(bufpt++) = (char)(exp/10+'0'); /* 10's digit */ + *(bufpt++) = (char)(exp%10+'0'); /* 1's digit */ + } + *bufpt = 0; + + /* The converted number is in buf[] and zero terminated. Output it. + ** Note that the number is in the usual order, not reversed as with + ** integer conversions. */ + length = (int)(bufpt-zOut); + bufpt = zOut; + + /* Special case: Add leading zeros if the flag_zeropad flag is + ** set and we are not left justified */ + if( flag_zeropad && !flag_leftjustify && length < width){ + int i; + int nPad = width - length; + for(i=width; i>=nPad; i--){ + bufpt[i] = bufpt[i-nPad]; + } + i = prefix!=0; + while( nPad-- ) bufpt[i++] = '0'; + length = width; + } +#endif /* !defined(SQLITE_OMIT_FLOATING_POINT) */ + break; + case etSIZE: + if( !bArgList ){ + *(va_arg(ap,int*)) = pAccum->nChar; + } + length = width = 0; + break; + case etPERCENT: + buf[0] = '%'; + bufpt = buf; + length = 1; + break; + case etCHARX: + if( bArgList ){ + bufpt = getTextArg(pArgList); + length = 1; + if( bufpt ){ + buf[0] = c = *(bufpt++); + if( (c&0xc0)==0xc0 ){ + while( length<4 && (bufpt[0]&0xc0)==0x80 ){ + buf[length++] = *(bufpt++); + } + } + }else{ + buf[0] = 0; + } + }else{ + unsigned int ch = va_arg(ap,unsigned int); + if( ch<0x00080 ){ + buf[0] = ch & 0xff; + length = 1; + }else if( ch<0x00800 ){ + buf[0] = 0xc0 + (u8)((ch>>6)&0x1f); + buf[1] = 0x80 + (u8)(ch & 0x3f); + length = 2; + }else if( ch<0x10000 ){ + buf[0] = 0xe0 + (u8)((ch>>12)&0x0f); + buf[1] = 0x80 + (u8)((ch>>6) & 0x3f); + buf[2] = 0x80 + (u8)(ch & 0x3f); + length = 3; + }else{ + buf[0] = 0xf0 + (u8)((ch>>18) & 0x07); + buf[1] = 0x80 + (u8)((ch>>12) & 0x3f); + buf[2] = 0x80 + (u8)((ch>>6) & 0x3f); + buf[3] = 0x80 + (u8)(ch & 0x3f); + length = 4; + } + } + if( precision>1 ){ + width -= precision-1; + if( width>1 && !flag_leftjustify ){ + sqlite3_str_appendchar(pAccum, width-1, ' '); + width = 0; + } + while( precision-- > 1 ){ + sqlite3_str_append(pAccum, buf, length); + } + } + bufpt = buf; + flag_altform2 = 1; + goto adjust_width_for_utf8; + case etSTRING: + case etDYNSTRING: + if( bArgList ){ + bufpt = getTextArg(pArgList); + xtype = etSTRING; + }else{ + bufpt = va_arg(ap,char*); + } + if( bufpt==0 ){ + bufpt = ""; + }else if( xtype==etDYNSTRING ){ + if( pAccum->nChar==0 + && pAccum->mxAlloc + && width==0 + && precision<0 + && pAccum->accError==0 + ){ + /* Special optimization for sqlite3_mprintf("%z..."): + ** Extend an existing memory allocation rather than creating + ** a new one. */ + assert( (pAccum->printfFlags&SQLITE_PRINTF_MALLOCED)==0 ); + pAccum->zText = bufpt; + pAccum->nAlloc = sqlite3DbMallocSize(pAccum->db, bufpt); + pAccum->nChar = 0x7fffffff & (int)strlen(bufpt); + pAccum->printfFlags |= SQLITE_PRINTF_MALLOCED; + length = 0; + break; + } + zExtra = bufpt; + } + if( precision>=0 ){ + if( flag_altform2 ){ + /* Set length to the number of bytes needed in order to display + ** precision characters */ + unsigned char *z = (unsigned char*)bufpt; + while( precision-- > 0 && z[0] ){ + SQLITE_SKIP_UTF8(z); + } + length = (int)(z - (unsigned char*)bufpt); + }else{ + for(length=0; length0 ){ + /* Adjust width to account for extra bytes in UTF-8 characters */ + int ii = length - 1; + while( ii>=0 ) if( (bufpt[ii--] & 0xc0)==0x80 ) width++; + } + break; + case etSQLESCAPE: /* %q: Escape ' characters */ + case etSQLESCAPE2: /* %Q: Escape ' and enclose in '...' */ + case etSQLESCAPE3: { /* %w: Escape " characters */ + int i, j, k, n, isnull; + int needQuote; + char ch; + char q = ((xtype==etSQLESCAPE3)?'"':'\''); /* Quote character */ + char *escarg; + + if( bArgList ){ + escarg = getTextArg(pArgList); + }else{ + escarg = va_arg(ap,char*); + } + isnull = escarg==0; + if( isnull ) escarg = (xtype==etSQLESCAPE2 ? "NULL" : "(NULL)"); + /* For %q, %Q, and %w, the precision is the number of byte (or + ** characters if the ! flags is present) to use from the input. + ** Because of the extra quoting characters inserted, the number + ** of output characters may be larger than the precision. + */ + k = precision; + for(i=n=0; k!=0 && (ch=escarg[i])!=0; i++, k--){ + if( ch==q ) n++; + if( flag_altform2 && (ch&0xc0)==0xc0 ){ + while( (escarg[i+1]&0xc0)==0x80 ){ i++; } + } + } + needQuote = !isnull && xtype==etSQLESCAPE2; + n += i + 3; + if( n>etBUFSIZE ){ + bufpt = zExtra = printfTempBuf(pAccum, n); + if( bufpt==0 ) return; + }else{ + bufpt = buf; + } + j = 0; + if( needQuote ) bufpt[j++] = q; + k = i; + for(i=0; iprintfFlags & SQLITE_PRINTF_INTERNAL)==0 ) return; + pToken = va_arg(ap, Token*); + assert( bArgList==0 ); + if( pToken && pToken->n ){ + sqlite3_str_append(pAccum, (const char*)pToken->z, pToken->n); + } + length = width = 0; + break; + } + case etSRCLIST: { + SrcList *pSrc; + int k; + struct SrcList_item *pItem; + if( (pAccum->printfFlags & SQLITE_PRINTF_INTERNAL)==0 ) return; + pSrc = va_arg(ap, SrcList*); + k = va_arg(ap, int); + pItem = &pSrc->a[k]; + assert( bArgList==0 ); + assert( k>=0 && knSrc ); + if( pItem->zDatabase ){ + sqlite3_str_appendall(pAccum, pItem->zDatabase); + sqlite3_str_append(pAccum, ".", 1); + } + sqlite3_str_appendall(pAccum, pItem->zName); + length = width = 0; + break; + } + default: { + assert( xtype==etINVALID ); + return; + } + }/* End switch over the format type */ + /* + ** The text of the conversion is pointed to by "bufpt" and is + ** "length" characters long. The field width is "width". Do + ** the output. Both length and width are in bytes, not characters, + ** at this point. If the "!" flag was present on string conversions + ** indicating that width and precision should be expressed in characters, + ** then the values have been translated prior to reaching this point. + */ + width -= length; + if( width>0 ){ + if( !flag_leftjustify ) sqlite3_str_appendchar(pAccum, width, ' '); + sqlite3_str_append(pAccum, bufpt, length); + if( flag_leftjustify ) sqlite3_str_appendchar(pAccum, width, ' '); + }else{ + sqlite3_str_append(pAccum, bufpt, length); + } + + if( zExtra ){ + sqlite3DbFree(pAccum->db, zExtra); + zExtra = 0; + } + }/* End for loop over the format string */ +} /* End of function */ + +/* +** Enlarge the memory allocation on a StrAccum object so that it is +** able to accept at least N more bytes of text. +** +** Return the number of bytes of text that StrAccum is able to accept +** after the attempted enlargement. The value returned might be zero. +*/ +static int sqlite3StrAccumEnlarge(StrAccum *p, int N){ + char *zNew; + assert( p->nChar+(i64)N >= p->nAlloc ); /* Only called if really needed */ + if( p->accError ){ + testcase(p->accError==SQLITE_TOOBIG); + testcase(p->accError==SQLITE_NOMEM); + return 0; + } + if( p->mxAlloc==0 ){ + setStrAccumError(p, SQLITE_TOOBIG); + return p->nAlloc - p->nChar - 1; + }else{ + char *zOld = isMalloced(p) ? p->zText : 0; + i64 szNew = p->nChar; + szNew += N + 1; + if( szNew+p->nChar<=p->mxAlloc ){ + /* Force exponential buffer size growth as long as it does not overflow, + ** to avoid having to call this routine too often */ + szNew += p->nChar; + } + if( szNew > p->mxAlloc ){ + sqlite3_str_reset(p); + setStrAccumError(p, SQLITE_TOOBIG); + return 0; + }else{ + p->nAlloc = (int)szNew; + } + if( p->db ){ + zNew = sqlite3DbRealloc(p->db, zOld, p->nAlloc); + }else{ + zNew = sqlite3_realloc64(zOld, p->nAlloc); + } + if( zNew ){ + assert( p->zText!=0 || p->nChar==0 ); + if( !isMalloced(p) && p->nChar>0 ) memcpy(zNew, p->zText, p->nChar); + p->zText = zNew; + p->nAlloc = sqlite3DbMallocSize(p->db, zNew); + p->printfFlags |= SQLITE_PRINTF_MALLOCED; + }else{ + sqlite3_str_reset(p); + setStrAccumError(p, SQLITE_NOMEM); + return 0; + } + } + return N; +} + +/* +** Append N copies of character c to the given string buffer. +*/ +SQLITE_API void sqlite3_str_appendchar(sqlite3_str *p, int N, char c){ + testcase( p->nChar + (i64)N > 0x7fffffff ); + if( p->nChar+(i64)N >= p->nAlloc && (N = sqlite3StrAccumEnlarge(p, N))<=0 ){ + return; + } + while( (N--)>0 ) p->zText[p->nChar++] = c; +} + +/* +** The StrAccum "p" is not large enough to accept N new bytes of z[]. +** So enlarge if first, then do the append. +** +** This is a helper routine to sqlite3_str_append() that does special-case +** work (enlarging the buffer) using tail recursion, so that the +** sqlite3_str_append() routine can use fast calling semantics. +*/ +static void SQLITE_NOINLINE enlargeAndAppend(StrAccum *p, const char *z, int N){ + N = sqlite3StrAccumEnlarge(p, N); + if( N>0 ){ + memcpy(&p->zText[p->nChar], z, N); + p->nChar += N; + } +} + +/* +** Append N bytes of text from z to the StrAccum object. Increase the +** size of the memory allocation for StrAccum if necessary. +*/ +SQLITE_API void sqlite3_str_append(sqlite3_str *p, const char *z, int N){ + assert( z!=0 || N==0 ); + assert( p->zText!=0 || p->nChar==0 || p->accError ); + assert( N>=0 ); + assert( p->accError==0 || p->nAlloc==0 || p->mxAlloc==0 ); + if( p->nChar+N >= p->nAlloc ){ + enlargeAndAppend(p,z,N); + }else if( N ){ + assert( p->zText ); + p->nChar += N; + memcpy(&p->zText[p->nChar-N], z, N); + } +} + +/* +** Append the complete text of zero-terminated string z[] to the p string. +*/ +SQLITE_API void sqlite3_str_appendall(sqlite3_str *p, const char *z){ + sqlite3_str_append(p, z, sqlite3Strlen30(z)); +} + + +/* +** Finish off a string by making sure it is zero-terminated. +** Return a pointer to the resulting string. Return a NULL +** pointer if any kind of error was encountered. +*/ +static SQLITE_NOINLINE char *strAccumFinishRealloc(StrAccum *p){ + char *zText; + assert( p->mxAlloc>0 && !isMalloced(p) ); + zText = sqlite3DbMallocRaw(p->db, p->nChar+1 ); + if( zText ){ + memcpy(zText, p->zText, p->nChar+1); + p->printfFlags |= SQLITE_PRINTF_MALLOCED; + }else{ + setStrAccumError(p, SQLITE_NOMEM); + } + p->zText = zText; + return zText; +} +SQLITE_PRIVATE char *sqlite3StrAccumFinish(StrAccum *p){ + if( p->zText ){ + p->zText[p->nChar] = 0; + if( p->mxAlloc>0 && !isMalloced(p) ){ + return strAccumFinishRealloc(p); + } + } + return p->zText; +} + +/* +** This singleton is an sqlite3_str object that is returned if +** sqlite3_malloc() fails to provide space for a real one. This +** sqlite3_str object accepts no new text and always returns +** an SQLITE_NOMEM error. +*/ +static sqlite3_str sqlite3OomStr = { + 0, 0, 0, 0, 0, SQLITE_NOMEM, 0 +}; + +/* Finalize a string created using sqlite3_str_new(). +*/ +SQLITE_API char *sqlite3_str_finish(sqlite3_str *p){ + char *z; + if( p!=0 && p!=&sqlite3OomStr ){ + z = sqlite3StrAccumFinish(p); + sqlite3_free(p); + }else{ + z = 0; + } + return z; +} + +/* Return any error code associated with p */ +SQLITE_API int sqlite3_str_errcode(sqlite3_str *p){ + return p ? p->accError : SQLITE_NOMEM; +} + +/* Return the current length of p in bytes */ +SQLITE_API int sqlite3_str_length(sqlite3_str *p){ + return p ? p->nChar : 0; +} + +/* Return the current value for p */ +SQLITE_API char *sqlite3_str_value(sqlite3_str *p){ + if( p==0 || p->nChar==0 ) return 0; + p->zText[p->nChar] = 0; + return p->zText; +} + +/* +** Reset an StrAccum string. Reclaim all malloced memory. +*/ +SQLITE_API void sqlite3_str_reset(StrAccum *p){ + if( isMalloced(p) ){ + sqlite3DbFree(p->db, p->zText); + p->printfFlags &= ~SQLITE_PRINTF_MALLOCED; + } + p->nAlloc = 0; + p->nChar = 0; + p->zText = 0; +} + +/* +** Initialize a string accumulator. +** +** p: The accumulator to be initialized. +** db: Pointer to a database connection. May be NULL. Lookaside +** memory is used if not NULL. db->mallocFailed is set appropriately +** when not NULL. +** zBase: An initial buffer. May be NULL in which case the initial buffer +** is malloced. +** n: Size of zBase in bytes. If total space requirements never exceed +** n then no memory allocations ever occur. +** mx: Maximum number of bytes to accumulate. If mx==0 then no memory +** allocations will ever occur. +*/ +SQLITE_PRIVATE void sqlite3StrAccumInit(StrAccum *p, sqlite3 *db, char *zBase, int n, int mx){ + p->zText = zBase; + p->db = db; + p->nAlloc = n; + p->mxAlloc = mx; + p->nChar = 0; + p->accError = 0; + p->printfFlags = 0; +} + +/* Allocate and initialize a new dynamic string object */ +SQLITE_API sqlite3_str *sqlite3_str_new(sqlite3 *db){ + sqlite3_str *p = sqlite3_malloc64(sizeof(*p)); + if( p ){ + sqlite3StrAccumInit(p, 0, 0, 0, + db ? db->aLimit[SQLITE_LIMIT_LENGTH] : SQLITE_MAX_LENGTH); + }else{ + p = &sqlite3OomStr; + } + return p; +} + +/* +** Print into memory obtained from sqliteMalloc(). Use the internal +** %-conversion extensions. +*/ +SQLITE_PRIVATE char *sqlite3VMPrintf(sqlite3 *db, const char *zFormat, va_list ap){ + char *z; + char zBase[SQLITE_PRINT_BUF_SIZE]; + StrAccum acc; + assert( db!=0 ); + sqlite3StrAccumInit(&acc, db, zBase, sizeof(zBase), + db->aLimit[SQLITE_LIMIT_LENGTH]); + acc.printfFlags = SQLITE_PRINTF_INTERNAL; + sqlite3_str_vappendf(&acc, zFormat, ap); + z = sqlite3StrAccumFinish(&acc); + if( acc.accError==SQLITE_NOMEM ){ + sqlite3OomFault(db); + } + return z; +} + +/* +** Print into memory obtained from sqliteMalloc(). Use the internal +** %-conversion extensions. +*/ +SQLITE_PRIVATE char *sqlite3MPrintf(sqlite3 *db, const char *zFormat, ...){ + va_list ap; + char *z; + va_start(ap, zFormat); + z = sqlite3VMPrintf(db, zFormat, ap); + va_end(ap); + return z; +} + +/* +** Print into memory obtained from sqlite3_malloc(). Omit the internal +** %-conversion extensions. +*/ +SQLITE_API char *sqlite3_vmprintf(const char *zFormat, va_list ap){ + char *z; + char zBase[SQLITE_PRINT_BUF_SIZE]; + StrAccum acc; + +#ifdef SQLITE_ENABLE_API_ARMOR + if( zFormat==0 ){ + (void)SQLITE_MISUSE_BKPT; + return 0; + } +#endif +#ifndef SQLITE_OMIT_AUTOINIT + if( sqlite3_initialize() ) return 0; +#endif + sqlite3StrAccumInit(&acc, 0, zBase, sizeof(zBase), SQLITE_MAX_LENGTH); + sqlite3_str_vappendf(&acc, zFormat, ap); + z = sqlite3StrAccumFinish(&acc); + return z; +} + +/* +** Print into memory obtained from sqlite3_malloc()(). Omit the internal +** %-conversion extensions. +*/ +SQLITE_API char *sqlite3_mprintf(const char *zFormat, ...){ + va_list ap; + char *z; +#ifndef SQLITE_OMIT_AUTOINIT + if( sqlite3_initialize() ) return 0; +#endif + va_start(ap, zFormat); + z = sqlite3_vmprintf(zFormat, ap); + va_end(ap); + return z; +} + +/* +** sqlite3_snprintf() works like snprintf() except that it ignores the +** current locale settings. This is important for SQLite because we +** are not able to use a "," as the decimal point in place of "." as +** specified by some locales. +** +** Oops: The first two arguments of sqlite3_snprintf() are backwards +** from the snprintf() standard. Unfortunately, it is too late to change +** this without breaking compatibility, so we just have to live with the +** mistake. +** +** sqlite3_vsnprintf() is the varargs version. +*/ +SQLITE_API char *sqlite3_vsnprintf(int n, char *zBuf, const char *zFormat, va_list ap){ + StrAccum acc; + if( n<=0 ) return zBuf; +#ifdef SQLITE_ENABLE_API_ARMOR + if( zBuf==0 || zFormat==0 ) { + (void)SQLITE_MISUSE_BKPT; + if( zBuf ) zBuf[0] = 0; + return zBuf; + } +#endif + sqlite3StrAccumInit(&acc, 0, zBuf, n, 0); + sqlite3_str_vappendf(&acc, zFormat, ap); + zBuf[acc.nChar] = 0; + return zBuf; +} +SQLITE_API char *sqlite3_snprintf(int n, char *zBuf, const char *zFormat, ...){ + char *z; + va_list ap; + va_start(ap,zFormat); + z = sqlite3_vsnprintf(n, zBuf, zFormat, ap); + va_end(ap); + return z; +} + +/* +** This is the routine that actually formats the sqlite3_log() message. +** We house it in a separate routine from sqlite3_log() to avoid using +** stack space on small-stack systems when logging is disabled. +** +** sqlite3_log() must render into a static buffer. It cannot dynamically +** allocate memory because it might be called while the memory allocator +** mutex is held. +** +** sqlite3_str_vappendf() might ask for *temporary* memory allocations for +** certain format characters (%q) or for very large precisions or widths. +** Care must be taken that any sqlite3_log() calls that occur while the +** memory mutex is held do not use these mechanisms. +*/ +static void renderLogMsg(int iErrCode, const char *zFormat, va_list ap){ + StrAccum acc; /* String accumulator */ + char zMsg[SQLITE_PRINT_BUF_SIZE*3]; /* Complete log message */ + + sqlite3StrAccumInit(&acc, 0, zMsg, sizeof(zMsg), 0); + sqlite3_str_vappendf(&acc, zFormat, ap); + sqlite3GlobalConfig.xLog(sqlite3GlobalConfig.pLogArg, iErrCode, + sqlite3StrAccumFinish(&acc)); +} + +/* +** Format and write a message to the log if logging is enabled. +*/ +SQLITE_API void sqlite3_log(int iErrCode, const char *zFormat, ...){ + va_list ap; /* Vararg list */ + if( sqlite3GlobalConfig.xLog ){ + va_start(ap, zFormat); + renderLogMsg(iErrCode, zFormat, ap); + va_end(ap); + } +} + +#if defined(SQLITE_DEBUG) || defined(SQLITE_HAVE_OS_TRACE) +/* +** A version of printf() that understands %lld. Used for debugging. +** The printf() built into some versions of windows does not understand %lld +** and segfaults if you give it a long long int. +*/ +SQLITE_PRIVATE void sqlite3DebugPrintf(const char *zFormat, ...){ + va_list ap; + StrAccum acc; + char zBuf[500]; + sqlite3StrAccumInit(&acc, 0, zBuf, sizeof(zBuf), 0); + va_start(ap,zFormat); + sqlite3_str_vappendf(&acc, zFormat, ap); + va_end(ap); + sqlite3StrAccumFinish(&acc); +#ifdef SQLITE_OS_TRACE_PROC + { + extern void SQLITE_OS_TRACE_PROC(const char *zBuf, int nBuf); + SQLITE_OS_TRACE_PROC(zBuf, sizeof(zBuf)); + } +#else + fprintf(stdout,"%s", zBuf); + fflush(stdout); +#endif +} +#endif + + +/* +** variable-argument wrapper around sqlite3_str_vappendf(). The bFlags argument +** can contain the bit SQLITE_PRINTF_INTERNAL enable internal formats. +*/ +SQLITE_API void sqlite3_str_appendf(StrAccum *p, const char *zFormat, ...){ + va_list ap; + va_start(ap,zFormat); + sqlite3_str_vappendf(p, zFormat, ap); + va_end(ap); +} + +/************** End of printf.c **********************************************/ +/************** Begin file treeview.c ****************************************/ +/* +** 2015-06-08 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** +** This file contains C code to implement the TreeView debugging routines. +** These routines print a parse tree to standard output for debugging and +** analysis. +** +** The interfaces in this file is only available when compiling +** with SQLITE_DEBUG. +*/ +/* #include "sqliteInt.h" */ +#ifdef SQLITE_DEBUG + +/* +** Add a new subitem to the tree. The moreToFollow flag indicates that this +** is not the last item in the tree. +*/ +static TreeView *sqlite3TreeViewPush(TreeView *p, u8 moreToFollow){ + if( p==0 ){ + p = sqlite3_malloc64( sizeof(*p) ); + if( p==0 ) return 0; + memset(p, 0, sizeof(*p)); + }else{ + p->iLevel++; + } + assert( moreToFollow==0 || moreToFollow==1 ); + if( p->iLevelbLine) ) p->bLine[p->iLevel] = moreToFollow; + return p; +} + +/* +** Finished with one layer of the tree +*/ +static void sqlite3TreeViewPop(TreeView *p){ + if( p==0 ) return; + p->iLevel--; + if( p->iLevel<0 ) sqlite3_free(p); +} + +/* +** Generate a single line of output for the tree, with a prefix that contains +** all the appropriate tree lines +*/ +static void sqlite3TreeViewLine(TreeView *p, const char *zFormat, ...){ + va_list ap; + int i; + StrAccum acc; + char zBuf[500]; + sqlite3StrAccumInit(&acc, 0, zBuf, sizeof(zBuf), 0); + if( p ){ + for(i=0; iiLevel && ibLine)-1; i++){ + sqlite3_str_append(&acc, p->bLine[i] ? "| " : " ", 4); + } + sqlite3_str_append(&acc, p->bLine[i] ? "|-- " : "'-- ", 4); + } + if( zFormat!=0 ){ + va_start(ap, zFormat); + sqlite3_str_vappendf(&acc, zFormat, ap); + va_end(ap); + assert( acc.nChar>0 ); + sqlite3_str_append(&acc, "\n", 1); + } + sqlite3StrAccumFinish(&acc); + fprintf(stdout,"%s", zBuf); + fflush(stdout); +} + +/* +** Shorthand for starting a new tree item that consists of a single label +*/ +static void sqlite3TreeViewItem(TreeView *p, const char *zLabel,u8 moreFollows){ + p = sqlite3TreeViewPush(p, moreFollows); + sqlite3TreeViewLine(p, "%s", zLabel); +} + +/* +** Generate a human-readable description of a WITH clause. +*/ +SQLITE_PRIVATE void sqlite3TreeViewWith(TreeView *pView, const With *pWith, u8 moreToFollow){ + int i; + if( pWith==0 ) return; + if( pWith->nCte==0 ) return; + if( pWith->pOuter ){ + sqlite3TreeViewLine(pView, "WITH (0x%p, pOuter=0x%p)",pWith,pWith->pOuter); + }else{ + sqlite3TreeViewLine(pView, "WITH (0x%p)", pWith); + } + if( pWith->nCte>0 ){ + pView = sqlite3TreeViewPush(pView, 1); + for(i=0; inCte; i++){ + StrAccum x; + char zLine[1000]; + const struct Cte *pCte = &pWith->a[i]; + sqlite3StrAccumInit(&x, 0, zLine, sizeof(zLine), 0); + sqlite3_str_appendf(&x, "%s", pCte->zName); + if( pCte->pCols && pCte->pCols->nExpr>0 ){ + char cSep = '('; + int j; + for(j=0; jpCols->nExpr; j++){ + sqlite3_str_appendf(&x, "%c%s", cSep, pCte->pCols->a[j].zName); + cSep = ','; + } + sqlite3_str_appendf(&x, ")"); + } + sqlite3_str_appendf(&x, " AS"); + sqlite3StrAccumFinish(&x); + sqlite3TreeViewItem(pView, zLine, inCte-1); + sqlite3TreeViewSelect(pView, pCte->pSelect, 0); + sqlite3TreeViewPop(pView); + } + sqlite3TreeViewPop(pView); + } +} + +/* +** Generate a human-readable description of a SrcList object. +*/ +SQLITE_PRIVATE void sqlite3TreeViewSrcList(TreeView *pView, const SrcList *pSrc){ + int i; + for(i=0; inSrc; i++){ + const struct SrcList_item *pItem = &pSrc->a[i]; + StrAccum x; + char zLine[100]; + sqlite3StrAccumInit(&x, 0, zLine, sizeof(zLine), 0); + sqlite3_str_appendf(&x, "{%d,*}", pItem->iCursor); + if( pItem->zDatabase ){ + sqlite3_str_appendf(&x, " %s.%s", pItem->zDatabase, pItem->zName); + }else if( pItem->zName ){ + sqlite3_str_appendf(&x, " %s", pItem->zName); + } + if( pItem->pTab ){ + sqlite3_str_appendf(&x, " tab=%Q nCol=%d ptr=%p", + pItem->pTab->zName, pItem->pTab->nCol, pItem->pTab); + } + if( pItem->zAlias ){ + sqlite3_str_appendf(&x, " (AS %s)", pItem->zAlias); + } + if( pItem->fg.jointype & JT_LEFT ){ + sqlite3_str_appendf(&x, " LEFT-JOIN"); + } + sqlite3StrAccumFinish(&x); + sqlite3TreeViewItem(pView, zLine, inSrc-1); + if( pItem->pSelect ){ + sqlite3TreeViewSelect(pView, pItem->pSelect, 0); + } + if( pItem->fg.isTabFunc ){ + sqlite3TreeViewExprList(pView, pItem->u1.pFuncArg, 0, "func-args:"); + } + sqlite3TreeViewPop(pView); + } +} + +/* +** Generate a human-readable description of a Select object. +*/ +SQLITE_PRIVATE void sqlite3TreeViewSelect(TreeView *pView, const Select *p, u8 moreToFollow){ + int n = 0; + int cnt = 0; + if( p==0 ){ + sqlite3TreeViewLine(pView, "nil-SELECT"); + return; + } + pView = sqlite3TreeViewPush(pView, moreToFollow); + if( p->pWith ){ + sqlite3TreeViewWith(pView, p->pWith, 1); + cnt = 1; + sqlite3TreeViewPush(pView, 1); + } + do{ + if( p->selFlags & SF_WhereBegin ){ + sqlite3TreeViewLine(pView, "sqlite3WhereBegin()"); + }else{ + sqlite3TreeViewLine(pView, + "SELECT%s%s (%u/%p) selFlags=0x%x nSelectRow=%d", + ((p->selFlags & SF_Distinct) ? " DISTINCT" : ""), + ((p->selFlags & SF_Aggregate) ? " agg_flag" : ""), + p->selId, p, p->selFlags, + (int)p->nSelectRow + ); + } + if( cnt++ ) sqlite3TreeViewPop(pView); + if( p->pPrior ){ + n = 1000; + }else{ + n = 0; + if( p->pSrc && p->pSrc->nSrc ) n++; + if( p->pWhere ) n++; + if( p->pGroupBy ) n++; + if( p->pHaving ) n++; + if( p->pOrderBy ) n++; + if( p->pLimit ) n++; +#ifndef SQLITE_OMIT_WINDOWFUNC + if( p->pWin ) n++; + if( p->pWinDefn ) n++; +#endif + } + if( p->pEList ){ + sqlite3TreeViewExprList(pView, p->pEList, n>0, "result-set"); + } + n--; +#ifndef SQLITE_OMIT_WINDOWFUNC + if( p->pWin ){ + Window *pX; + pView = sqlite3TreeViewPush(pView, (n--)>0); + sqlite3TreeViewLine(pView, "window-functions"); + for(pX=p->pWin; pX; pX=pX->pNextWin){ + sqlite3TreeViewWinFunc(pView, pX, pX->pNextWin!=0); + } + sqlite3TreeViewPop(pView); + } +#endif + if( p->pSrc && p->pSrc->nSrc ){ + pView = sqlite3TreeViewPush(pView, (n--)>0); + sqlite3TreeViewLine(pView, "FROM"); + sqlite3TreeViewSrcList(pView, p->pSrc); + sqlite3TreeViewPop(pView); + } + if( p->pWhere ){ + sqlite3TreeViewItem(pView, "WHERE", (n--)>0); + sqlite3TreeViewExpr(pView, p->pWhere, 0); + sqlite3TreeViewPop(pView); + } + if( p->pGroupBy ){ + sqlite3TreeViewExprList(pView, p->pGroupBy, (n--)>0, "GROUPBY"); + } + if( p->pHaving ){ + sqlite3TreeViewItem(pView, "HAVING", (n--)>0); + sqlite3TreeViewExpr(pView, p->pHaving, 0); + sqlite3TreeViewPop(pView); + } +#ifndef SQLITE_OMIT_WINDOWFUNC + if( p->pWinDefn ){ + Window *pX; + sqlite3TreeViewItem(pView, "WINDOW", (n--)>0); + for(pX=p->pWinDefn; pX; pX=pX->pNextWin){ + sqlite3TreeViewWindow(pView, pX, pX->pNextWin!=0); + } + sqlite3TreeViewPop(pView); + } +#endif + if( p->pOrderBy ){ + sqlite3TreeViewExprList(pView, p->pOrderBy, (n--)>0, "ORDERBY"); + } + if( p->pLimit ){ + sqlite3TreeViewItem(pView, "LIMIT", (n--)>0); + sqlite3TreeViewExpr(pView, p->pLimit->pLeft, p->pLimit->pRight!=0); + if( p->pLimit->pRight ){ + sqlite3TreeViewItem(pView, "OFFSET", (n--)>0); + sqlite3TreeViewExpr(pView, p->pLimit->pRight, 0); + sqlite3TreeViewPop(pView); + } + sqlite3TreeViewPop(pView); + } + if( p->pPrior ){ + const char *zOp = "UNION"; + switch( p->op ){ + case TK_ALL: zOp = "UNION ALL"; break; + case TK_INTERSECT: zOp = "INTERSECT"; break; + case TK_EXCEPT: zOp = "EXCEPT"; break; + } + sqlite3TreeViewItem(pView, zOp, 1); + } + p = p->pPrior; + }while( p!=0 ); + sqlite3TreeViewPop(pView); +} + +#ifndef SQLITE_OMIT_WINDOWFUNC +/* +** Generate a description of starting or stopping bounds +*/ +SQLITE_PRIVATE void sqlite3TreeViewBound( + TreeView *pView, /* View context */ + u8 eBound, /* UNBOUNDED, CURRENT, PRECEDING, FOLLOWING */ + Expr *pExpr, /* Value for PRECEDING or FOLLOWING */ + u8 moreToFollow /* True if more to follow */ +){ + switch( eBound ){ + case TK_UNBOUNDED: { + sqlite3TreeViewItem(pView, "UNBOUNDED", moreToFollow); + sqlite3TreeViewPop(pView); + break; + } + case TK_CURRENT: { + sqlite3TreeViewItem(pView, "CURRENT", moreToFollow); + sqlite3TreeViewPop(pView); + break; + } + case TK_PRECEDING: { + sqlite3TreeViewItem(pView, "PRECEDING", moreToFollow); + sqlite3TreeViewExpr(pView, pExpr, 0); + sqlite3TreeViewPop(pView); + break; + } + case TK_FOLLOWING: { + sqlite3TreeViewItem(pView, "FOLLOWING", moreToFollow); + sqlite3TreeViewExpr(pView, pExpr, 0); + sqlite3TreeViewPop(pView); + break; + } + } +} +#endif /* SQLITE_OMIT_WINDOWFUNC */ + +#ifndef SQLITE_OMIT_WINDOWFUNC +/* +** Generate a human-readable explanation for a Window object +*/ +SQLITE_PRIVATE void sqlite3TreeViewWindow(TreeView *pView, const Window *pWin, u8 more){ + int nElement = 0; + if( pWin->pFilter ){ + sqlite3TreeViewItem(pView, "FILTER", 1); + sqlite3TreeViewExpr(pView, pWin->pFilter, 0); + sqlite3TreeViewPop(pView); + } + pView = sqlite3TreeViewPush(pView, more); + if( pWin->zName ){ + sqlite3TreeViewLine(pView, "OVER %s (%p)", pWin->zName, pWin); + }else{ + sqlite3TreeViewLine(pView, "OVER (%p)", pWin); + } + if( pWin->zBase ) nElement++; + if( pWin->pOrderBy ) nElement++; + if( pWin->eFrmType ) nElement++; + if( pWin->eExclude ) nElement++; + if( pWin->zBase ){ + sqlite3TreeViewPush(pView, (--nElement)>0); + sqlite3TreeViewLine(pView, "window: %s", pWin->zBase); + sqlite3TreeViewPop(pView); + } + if( pWin->pPartition ){ + sqlite3TreeViewExprList(pView, pWin->pPartition, nElement>0,"PARTITION-BY"); + } + if( pWin->pOrderBy ){ + sqlite3TreeViewExprList(pView, pWin->pOrderBy, (--nElement)>0, "ORDER-BY"); + } + if( pWin->eFrmType ){ + char zBuf[30]; + const char *zFrmType = "ROWS"; + if( pWin->eFrmType==TK_RANGE ) zFrmType = "RANGE"; + if( pWin->eFrmType==TK_GROUPS ) zFrmType = "GROUPS"; + sqlite3_snprintf(sizeof(zBuf),zBuf,"%s%s",zFrmType, + pWin->bImplicitFrame ? " (implied)" : ""); + sqlite3TreeViewItem(pView, zBuf, (--nElement)>0); + sqlite3TreeViewBound(pView, pWin->eStart, pWin->pStart, 1); + sqlite3TreeViewBound(pView, pWin->eEnd, pWin->pEnd, 0); + sqlite3TreeViewPop(pView); + } + if( pWin->eExclude ){ + char zBuf[30]; + const char *zExclude; + switch( pWin->eExclude ){ + case TK_NO: zExclude = "NO OTHERS"; break; + case TK_CURRENT: zExclude = "CURRENT ROW"; break; + case TK_GROUP: zExclude = "GROUP"; break; + case TK_TIES: zExclude = "TIES"; break; + default: + sqlite3_snprintf(sizeof(zBuf),zBuf,"invalid(%d)", pWin->eExclude); + zExclude = zBuf; + break; + } + sqlite3TreeViewPush(pView, 0); + sqlite3TreeViewLine(pView, "EXCLUDE %s", zExclude); + sqlite3TreeViewPop(pView); + } + sqlite3TreeViewPop(pView); +} +#endif /* SQLITE_OMIT_WINDOWFUNC */ + +#ifndef SQLITE_OMIT_WINDOWFUNC +/* +** Generate a human-readable explanation for a Window Function object +*/ +SQLITE_PRIVATE void sqlite3TreeViewWinFunc(TreeView *pView, const Window *pWin, u8 more){ + pView = sqlite3TreeViewPush(pView, more); + sqlite3TreeViewLine(pView, "WINFUNC %s(%d)", + pWin->pFunc->zName, pWin->pFunc->nArg); + sqlite3TreeViewWindow(pView, pWin, 0); + sqlite3TreeViewPop(pView); +} +#endif /* SQLITE_OMIT_WINDOWFUNC */ + +/* +** Generate a human-readable explanation of an expression tree. +*/ +SQLITE_PRIVATE void sqlite3TreeViewExpr(TreeView *pView, const Expr *pExpr, u8 moreToFollow){ + const char *zBinOp = 0; /* Binary operator */ + const char *zUniOp = 0; /* Unary operator */ + char zFlgs[60]; + pView = sqlite3TreeViewPush(pView, moreToFollow); + if( pExpr==0 ){ + sqlite3TreeViewLine(pView, "nil"); + sqlite3TreeViewPop(pView); + return; + } + if( pExpr->flags || pExpr->affExpr ){ + if( ExprHasProperty(pExpr, EP_FromJoin) ){ + sqlite3_snprintf(sizeof(zFlgs),zFlgs," fg.af=%x.%c iRJT=%d", + pExpr->flags, pExpr->affExpr ? pExpr->affExpr : 'n', + pExpr->iRightJoinTable); + }else{ + sqlite3_snprintf(sizeof(zFlgs),zFlgs," fg.af=%x.%c", + pExpr->flags, pExpr->affExpr ? pExpr->affExpr : 'n'); + } + }else{ + zFlgs[0] = 0; + } + switch( pExpr->op ){ + case TK_AGG_COLUMN: { + sqlite3TreeViewLine(pView, "AGG{%d:%d}%s", + pExpr->iTable, pExpr->iColumn, zFlgs); + break; + } + case TK_COLUMN: { + if( pExpr->iTable<0 ){ + /* This only happens when coding check constraints */ + sqlite3TreeViewLine(pView, "COLUMN(%d)%s", pExpr->iColumn, zFlgs); + }else{ + sqlite3TreeViewLine(pView, "{%d:%d}%s", + pExpr->iTable, pExpr->iColumn, zFlgs); + } + if( ExprHasProperty(pExpr, EP_FixedCol) ){ + sqlite3TreeViewExpr(pView, pExpr->pLeft, 0); + } + break; + } + case TK_INTEGER: { + if( pExpr->flags & EP_IntValue ){ + sqlite3TreeViewLine(pView, "%d", pExpr->u.iValue); + }else{ + sqlite3TreeViewLine(pView, "%s", pExpr->u.zToken); + } + break; + } +#ifndef SQLITE_OMIT_FLOATING_POINT + case TK_FLOAT: { + sqlite3TreeViewLine(pView,"%s", pExpr->u.zToken); + break; + } +#endif + case TK_STRING: { + sqlite3TreeViewLine(pView,"%Q", pExpr->u.zToken); + break; + } + case TK_NULL: { + sqlite3TreeViewLine(pView,"NULL"); + break; + } + case TK_TRUEFALSE: { + sqlite3TreeViewLine(pView, + sqlite3ExprTruthValue(pExpr) ? "TRUE" : "FALSE"); + break; + } +#ifndef SQLITE_OMIT_BLOB_LITERAL + case TK_BLOB: { + sqlite3TreeViewLine(pView,"%s", pExpr->u.zToken); + break; + } +#endif + case TK_VARIABLE: { + sqlite3TreeViewLine(pView,"VARIABLE(%s,%d)", + pExpr->u.zToken, pExpr->iColumn); + break; + } + case TK_REGISTER: { + sqlite3TreeViewLine(pView,"REGISTER(%d)", pExpr->iTable); + break; + } + case TK_ID: { + sqlite3TreeViewLine(pView,"ID \"%w\"", pExpr->u.zToken); + break; + } +#ifndef SQLITE_OMIT_CAST + case TK_CAST: { + /* Expressions of the form: CAST(pLeft AS token) */ + sqlite3TreeViewLine(pView,"CAST %Q", pExpr->u.zToken); + sqlite3TreeViewExpr(pView, pExpr->pLeft, 0); + break; + } +#endif /* SQLITE_OMIT_CAST */ + case TK_LT: zBinOp = "LT"; break; + case TK_LE: zBinOp = "LE"; break; + case TK_GT: zBinOp = "GT"; break; + case TK_GE: zBinOp = "GE"; break; + case TK_NE: zBinOp = "NE"; break; + case TK_EQ: zBinOp = "EQ"; break; + case TK_IS: zBinOp = "IS"; break; + case TK_ISNOT: zBinOp = "ISNOT"; break; + case TK_AND: zBinOp = "AND"; break; + case TK_OR: zBinOp = "OR"; break; + case TK_PLUS: zBinOp = "ADD"; break; + case TK_STAR: zBinOp = "MUL"; break; + case TK_MINUS: zBinOp = "SUB"; break; + case TK_REM: zBinOp = "REM"; break; + case TK_BITAND: zBinOp = "BITAND"; break; + case TK_BITOR: zBinOp = "BITOR"; break; + case TK_SLASH: zBinOp = "DIV"; break; + case TK_LSHIFT: zBinOp = "LSHIFT"; break; + case TK_RSHIFT: zBinOp = "RSHIFT"; break; + case TK_CONCAT: zBinOp = "CONCAT"; break; + case TK_DOT: zBinOp = "DOT"; break; + + case TK_UMINUS: zUniOp = "UMINUS"; break; + case TK_UPLUS: zUniOp = "UPLUS"; break; + case TK_BITNOT: zUniOp = "BITNOT"; break; + case TK_NOT: zUniOp = "NOT"; break; + case TK_ISNULL: zUniOp = "ISNULL"; break; + case TK_NOTNULL: zUniOp = "NOTNULL"; break; + + case TK_TRUTH: { + int x; + const char *azOp[] = { + "IS-FALSE", "IS-TRUE", "IS-NOT-FALSE", "IS-NOT-TRUE" + }; + assert( pExpr->op2==TK_IS || pExpr->op2==TK_ISNOT ); + assert( pExpr->pRight ); + assert( sqlite3ExprSkipCollate(pExpr->pRight)->op==TK_TRUEFALSE ); + x = (pExpr->op2==TK_ISNOT)*2 + sqlite3ExprTruthValue(pExpr->pRight); + zUniOp = azOp[x]; + break; + } + + case TK_SPAN: { + sqlite3TreeViewLine(pView, "SPAN %Q", pExpr->u.zToken); + sqlite3TreeViewExpr(pView, pExpr->pLeft, 0); + break; + } + + case TK_COLLATE: { + /* COLLATE operators without the EP_Collate flag are intended to + ** emulate collation associated with a table column. These show + ** up in the treeview output as "SOFT-COLLATE". Explicit COLLATE + ** operators that appear in the original SQL always have the + ** EP_Collate bit set and appear in treeview output as just "COLLATE" */ + sqlite3TreeViewLine(pView, "%sCOLLATE %Q%s", + !ExprHasProperty(pExpr, EP_Collate) ? "SOFT-" : "", + pExpr->u.zToken, zFlgs); + sqlite3TreeViewExpr(pView, pExpr->pLeft, 0); + break; + } + + case TK_AGG_FUNCTION: + case TK_FUNCTION: { + ExprList *pFarg; /* List of function arguments */ + Window *pWin; + if( ExprHasProperty(pExpr, EP_TokenOnly) ){ + pFarg = 0; + pWin = 0; + }else{ + pFarg = pExpr->x.pList; +#ifndef SQLITE_OMIT_WINDOWFUNC + pWin = pExpr->y.pWin; +#else + pWin = 0; +#endif + } + if( pExpr->op==TK_AGG_FUNCTION ){ + sqlite3TreeViewLine(pView, "AGG_FUNCTION%d %Q%s", + pExpr->op2, pExpr->u.zToken, zFlgs); + }else{ + sqlite3TreeViewLine(pView, "FUNCTION %Q%s", pExpr->u.zToken, zFlgs); + } + if( pFarg ){ + sqlite3TreeViewExprList(pView, pFarg, pWin!=0, 0); + } +#ifndef SQLITE_OMIT_WINDOWFUNC + if( pWin ){ + sqlite3TreeViewWindow(pView, pWin, 0); + } +#endif + break; + } +#ifndef SQLITE_OMIT_SUBQUERY + case TK_EXISTS: { + sqlite3TreeViewLine(pView, "EXISTS-expr flags=0x%x", pExpr->flags); + sqlite3TreeViewSelect(pView, pExpr->x.pSelect, 0); + break; + } + case TK_SELECT: { + sqlite3TreeViewLine(pView, "SELECT-expr flags=0x%x", pExpr->flags); + sqlite3TreeViewSelect(pView, pExpr->x.pSelect, 0); + break; + } + case TK_IN: { + sqlite3TreeViewLine(pView, "IN flags=0x%x", pExpr->flags); + sqlite3TreeViewExpr(pView, pExpr->pLeft, 1); + if( ExprHasProperty(pExpr, EP_xIsSelect) ){ + sqlite3TreeViewSelect(pView, pExpr->x.pSelect, 0); + }else{ + sqlite3TreeViewExprList(pView, pExpr->x.pList, 0, 0); + } + break; + } +#endif /* SQLITE_OMIT_SUBQUERY */ + + /* + ** x BETWEEN y AND z + ** + ** This is equivalent to + ** + ** x>=y AND x<=z + ** + ** X is stored in pExpr->pLeft. + ** Y is stored in pExpr->pList->a[0].pExpr. + ** Z is stored in pExpr->pList->a[1].pExpr. + */ + case TK_BETWEEN: { + Expr *pX = pExpr->pLeft; + Expr *pY = pExpr->x.pList->a[0].pExpr; + Expr *pZ = pExpr->x.pList->a[1].pExpr; + sqlite3TreeViewLine(pView, "BETWEEN"); + sqlite3TreeViewExpr(pView, pX, 1); + sqlite3TreeViewExpr(pView, pY, 1); + sqlite3TreeViewExpr(pView, pZ, 0); + break; + } + case TK_TRIGGER: { + /* If the opcode is TK_TRIGGER, then the expression is a reference + ** to a column in the new.* or old.* pseudo-tables available to + ** trigger programs. In this case Expr.iTable is set to 1 for the + ** new.* pseudo-table, or 0 for the old.* pseudo-table. Expr.iColumn + ** is set to the column of the pseudo-table to read, or to -1 to + ** read the rowid field. + */ + sqlite3TreeViewLine(pView, "%s(%d)", + pExpr->iTable ? "NEW" : "OLD", pExpr->iColumn); + break; + } + case TK_CASE: { + sqlite3TreeViewLine(pView, "CASE"); + sqlite3TreeViewExpr(pView, pExpr->pLeft, 1); + sqlite3TreeViewExprList(pView, pExpr->x.pList, 0, 0); + break; + } +#ifndef SQLITE_OMIT_TRIGGER + case TK_RAISE: { + const char *zType = "unk"; + switch( pExpr->affExpr ){ + case OE_Rollback: zType = "rollback"; break; + case OE_Abort: zType = "abort"; break; + case OE_Fail: zType = "fail"; break; + case OE_Ignore: zType = "ignore"; break; + } + sqlite3TreeViewLine(pView, "RAISE %s(%Q)", zType, pExpr->u.zToken); + break; + } +#endif + case TK_MATCH: { + sqlite3TreeViewLine(pView, "MATCH {%d:%d}%s", + pExpr->iTable, pExpr->iColumn, zFlgs); + sqlite3TreeViewExpr(pView, pExpr->pRight, 0); + break; + } + case TK_VECTOR: { + sqlite3TreeViewBareExprList(pView, pExpr->x.pList, "VECTOR"); + break; + } + case TK_SELECT_COLUMN: { + sqlite3TreeViewLine(pView, "SELECT-COLUMN %d", pExpr->iColumn); + sqlite3TreeViewSelect(pView, pExpr->pLeft->x.pSelect, 0); + break; + } + case TK_IF_NULL_ROW: { + sqlite3TreeViewLine(pView, "IF-NULL-ROW %d", pExpr->iTable); + sqlite3TreeViewExpr(pView, pExpr->pLeft, 0); + break; + } + default: { + sqlite3TreeViewLine(pView, "op=%d", pExpr->op); + break; + } + } + if( zBinOp ){ + sqlite3TreeViewLine(pView, "%s%s", zBinOp, zFlgs); + sqlite3TreeViewExpr(pView, pExpr->pLeft, 1); + sqlite3TreeViewExpr(pView, pExpr->pRight, 0); + }else if( zUniOp ){ + sqlite3TreeViewLine(pView, "%s%s", zUniOp, zFlgs); + sqlite3TreeViewExpr(pView, pExpr->pLeft, 0); + } + sqlite3TreeViewPop(pView); +} + + +/* +** Generate a human-readable explanation of an expression list. +*/ +SQLITE_PRIVATE void sqlite3TreeViewBareExprList( + TreeView *pView, + const ExprList *pList, + const char *zLabel +){ + if( zLabel==0 || zLabel[0]==0 ) zLabel = "LIST"; + if( pList==0 ){ + sqlite3TreeViewLine(pView, "%s (empty)", zLabel); + }else{ + int i; + sqlite3TreeViewLine(pView, "%s", zLabel); + for(i=0; inExpr; i++){ + int j = pList->a[i].u.x.iOrderByCol; + char *zName = pList->a[i].zName; + int moreToFollow = inExpr - 1; + if( j || zName ){ + sqlite3TreeViewPush(pView, moreToFollow); + moreToFollow = 0; + sqlite3TreeViewLine(pView, 0); + if( zName ){ + fprintf(stdout, "AS %s ", zName); + } + if( j ){ + fprintf(stdout, "iOrderByCol=%d", j); + } + fprintf(stdout, "\n"); + fflush(stdout); + } + sqlite3TreeViewExpr(pView, pList->a[i].pExpr, moreToFollow); + if( j || zName ){ + sqlite3TreeViewPop(pView); + } + } + } +} +SQLITE_PRIVATE void sqlite3TreeViewExprList( + TreeView *pView, + const ExprList *pList, + u8 moreToFollow, + const char *zLabel +){ + pView = sqlite3TreeViewPush(pView, moreToFollow); + sqlite3TreeViewBareExprList(pView, pList, zLabel); + sqlite3TreeViewPop(pView); +} + +#endif /* SQLITE_DEBUG */ + +/************** End of treeview.c ********************************************/ +/************** Begin file random.c ******************************************/ +/* +** 2001 September 15 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** This file contains code to implement a pseudo-random number +** generator (PRNG) for SQLite. +** +** Random numbers are used by some of the database backends in order +** to generate random integer keys for tables or random filenames. +*/ +/* #include "sqliteInt.h" */ + + +/* All threads share a single random number generator. +** This structure is the current state of the generator. +*/ +static SQLITE_WSD struct sqlite3PrngType { + unsigned char isInit; /* True if initialized */ + unsigned char i, j; /* State variables */ + unsigned char s[256]; /* State variables */ +} sqlite3Prng; + +/* +** Return N random bytes. +*/ +SQLITE_API void sqlite3_randomness(int N, void *pBuf){ + unsigned char t; + unsigned char *zBuf = pBuf; + + /* The "wsdPrng" macro will resolve to the pseudo-random number generator + ** state vector. If writable static data is unsupported on the target, + ** we have to locate the state vector at run-time. In the more common + ** case where writable static data is supported, wsdPrng can refer directly + ** to the "sqlite3Prng" state vector declared above. + */ +#ifdef SQLITE_OMIT_WSD + struct sqlite3PrngType *p = &GLOBAL(struct sqlite3PrngType, sqlite3Prng); +# define wsdPrng p[0] +#else +# define wsdPrng sqlite3Prng +#endif + +#if SQLITE_THREADSAFE + sqlite3_mutex *mutex; +#endif + +#ifndef SQLITE_OMIT_AUTOINIT + if( sqlite3_initialize() ) return; +#endif + +#if SQLITE_THREADSAFE + mutex = sqlite3MutexAlloc(SQLITE_MUTEX_STATIC_PRNG); +#endif + + sqlite3_mutex_enter(mutex); + if( N<=0 || pBuf==0 ){ + wsdPrng.isInit = 0; + sqlite3_mutex_leave(mutex); + return; + } + + /* Initialize the state of the random number generator once, + ** the first time this routine is called. The seed value does + ** not need to contain a lot of randomness since we are not + ** trying to do secure encryption or anything like that... + ** + ** Nothing in this file or anywhere else in SQLite does any kind of + ** encryption. The RC4 algorithm is being used as a PRNG (pseudo-random + ** number generator) not as an encryption device. + */ + if( !wsdPrng.isInit ){ + int i; + char k[256]; + wsdPrng.j = 0; + wsdPrng.i = 0; + sqlite3OsRandomness(sqlite3_vfs_find(0), 256, k); + for(i=0; i<256; i++){ + wsdPrng.s[i] = (u8)i; + } + for(i=0; i<256; i++){ + wsdPrng.j += wsdPrng.s[i] + k[i]; + t = wsdPrng.s[wsdPrng.j]; + wsdPrng.s[wsdPrng.j] = wsdPrng.s[i]; + wsdPrng.s[i] = t; + } + wsdPrng.isInit = 1; + } + + assert( N>0 ); + do{ + wsdPrng.i++; + t = wsdPrng.s[wsdPrng.i]; + wsdPrng.j += t; + wsdPrng.s[wsdPrng.i] = wsdPrng.s[wsdPrng.j]; + wsdPrng.s[wsdPrng.j] = t; + t += wsdPrng.s[wsdPrng.i]; + *(zBuf++) = wsdPrng.s[t]; + }while( --N ); + sqlite3_mutex_leave(mutex); +} + +#ifndef SQLITE_UNTESTABLE +/* +** For testing purposes, we sometimes want to preserve the state of +** PRNG and restore the PRNG to its saved state at a later time, or +** to reset the PRNG to its initial state. These routines accomplish +** those tasks. +** +** The sqlite3_test_control() interface calls these routines to +** control the PRNG. +*/ +static SQLITE_WSD struct sqlite3PrngType sqlite3SavedPrng; +SQLITE_PRIVATE void sqlite3PrngSaveState(void){ + memcpy( + &GLOBAL(struct sqlite3PrngType, sqlite3SavedPrng), + &GLOBAL(struct sqlite3PrngType, sqlite3Prng), + sizeof(sqlite3Prng) + ); +} +SQLITE_PRIVATE void sqlite3PrngRestoreState(void){ + memcpy( + &GLOBAL(struct sqlite3PrngType, sqlite3Prng), + &GLOBAL(struct sqlite3PrngType, sqlite3SavedPrng), + sizeof(sqlite3Prng) + ); +} +#endif /* SQLITE_UNTESTABLE */ + +/************** End of random.c **********************************************/ +/************** Begin file threads.c *****************************************/ +/* +** 2012 July 21 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +****************************************************************************** +** +** This file presents a simple cross-platform threading interface for +** use internally by SQLite. +** +** A "thread" can be created using sqlite3ThreadCreate(). This thread +** runs independently of its creator until it is joined using +** sqlite3ThreadJoin(), at which point it terminates. +** +** Threads do not have to be real. It could be that the work of the +** "thread" is done by the main thread at either the sqlite3ThreadCreate() +** or sqlite3ThreadJoin() call. This is, in fact, what happens in +** single threaded systems. Nothing in SQLite requires multiple threads. +** This interface exists so that applications that want to take advantage +** of multiple cores can do so, while also allowing applications to stay +** single-threaded if desired. +*/ +/* #include "sqliteInt.h" */ +#if SQLITE_OS_WIN +/* # include "os_win.h" */ +#endif + +#if SQLITE_MAX_WORKER_THREADS>0 + +/********************************* Unix Pthreads ****************************/ +#if SQLITE_OS_UNIX && defined(SQLITE_MUTEX_PTHREADS) && SQLITE_THREADSAFE>0 + +#define SQLITE_THREADS_IMPLEMENTED 1 /* Prevent the single-thread code below */ +/* #include */ + +/* A running thread */ +struct SQLiteThread { + pthread_t tid; /* Thread ID */ + int done; /* Set to true when thread finishes */ + void *pOut; /* Result returned by the thread */ + void *(*xTask)(void*); /* The thread routine */ + void *pIn; /* Argument to the thread */ +}; + +/* Create a new thread */ +SQLITE_PRIVATE int sqlite3ThreadCreate( + SQLiteThread **ppThread, /* OUT: Write the thread object here */ + void *(*xTask)(void*), /* Routine to run in a separate thread */ + void *pIn /* Argument passed into xTask() */ +){ + SQLiteThread *p; + int rc; + + assert( ppThread!=0 ); + assert( xTask!=0 ); + /* This routine is never used in single-threaded mode */ + assert( sqlite3GlobalConfig.bCoreMutex!=0 ); + + *ppThread = 0; + p = sqlite3Malloc(sizeof(*p)); + if( p==0 ) return SQLITE_NOMEM_BKPT; + memset(p, 0, sizeof(*p)); + p->xTask = xTask; + p->pIn = pIn; + /* If the SQLITE_TESTCTRL_FAULT_INSTALL callback is registered to a + ** function that returns SQLITE_ERROR when passed the argument 200, that + ** forces worker threads to run sequentially and deterministically + ** for testing purposes. */ + if( sqlite3FaultSim(200) ){ + rc = 1; + }else{ + rc = pthread_create(&p->tid, 0, xTask, pIn); + } + if( rc ){ + p->done = 1; + p->pOut = xTask(pIn); + } + *ppThread = p; + return SQLITE_OK; +} + +/* Get the results of the thread */ +SQLITE_PRIVATE int sqlite3ThreadJoin(SQLiteThread *p, void **ppOut){ + int rc; + + assert( ppOut!=0 ); + if( NEVER(p==0) ) return SQLITE_NOMEM_BKPT; + if( p->done ){ + *ppOut = p->pOut; + rc = SQLITE_OK; + }else{ + rc = pthread_join(p->tid, ppOut) ? SQLITE_ERROR : SQLITE_OK; + } + sqlite3_free(p); + return rc; +} + +#endif /* SQLITE_OS_UNIX && defined(SQLITE_MUTEX_PTHREADS) */ +/******************************** End Unix Pthreads *************************/ + + +/********************************* Win32 Threads ****************************/ +#if SQLITE_OS_WIN_THREADS + +#define SQLITE_THREADS_IMPLEMENTED 1 /* Prevent the single-thread code below */ +#include + +/* A running thread */ +struct SQLiteThread { + void *tid; /* The thread handle */ + unsigned id; /* The thread identifier */ + void *(*xTask)(void*); /* The routine to run as a thread */ + void *pIn; /* Argument to xTask */ + void *pResult; /* Result of xTask */ +}; + +/* Thread procedure Win32 compatibility shim */ +static unsigned __stdcall sqlite3ThreadProc( + void *pArg /* IN: Pointer to the SQLiteThread structure */ +){ + SQLiteThread *p = (SQLiteThread *)pArg; + + assert( p!=0 ); +#if 0 + /* + ** This assert appears to trigger spuriously on certain + ** versions of Windows, possibly due to _beginthreadex() + ** and/or CreateThread() not fully setting their thread + ** ID parameter before starting the thread. + */ + assert( p->id==GetCurrentThreadId() ); +#endif + assert( p->xTask!=0 ); + p->pResult = p->xTask(p->pIn); + + _endthreadex(0); + return 0; /* NOT REACHED */ +} + +/* Create a new thread */ +SQLITE_PRIVATE int sqlite3ThreadCreate( + SQLiteThread **ppThread, /* OUT: Write the thread object here */ + void *(*xTask)(void*), /* Routine to run in a separate thread */ + void *pIn /* Argument passed into xTask() */ +){ + SQLiteThread *p; + + assert( ppThread!=0 ); + assert( xTask!=0 ); + *ppThread = 0; + p = sqlite3Malloc(sizeof(*p)); + if( p==0 ) return SQLITE_NOMEM_BKPT; + /* If the SQLITE_TESTCTRL_FAULT_INSTALL callback is registered to a + ** function that returns SQLITE_ERROR when passed the argument 200, that + ** forces worker threads to run sequentially and deterministically + ** (via the sqlite3FaultSim() term of the conditional) for testing + ** purposes. */ + if( sqlite3GlobalConfig.bCoreMutex==0 || sqlite3FaultSim(200) ){ + memset(p, 0, sizeof(*p)); + }else{ + p->xTask = xTask; + p->pIn = pIn; + p->tid = (void*)_beginthreadex(0, 0, sqlite3ThreadProc, p, 0, &p->id); + if( p->tid==0 ){ + memset(p, 0, sizeof(*p)); + } + } + if( p->xTask==0 ){ + p->id = GetCurrentThreadId(); + p->pResult = xTask(pIn); + } + *ppThread = p; + return SQLITE_OK; +} + +SQLITE_PRIVATE DWORD sqlite3Win32Wait(HANDLE hObject); /* os_win.c */ + +/* Get the results of the thread */ +SQLITE_PRIVATE int sqlite3ThreadJoin(SQLiteThread *p, void **ppOut){ + DWORD rc; + BOOL bRc; + + assert( ppOut!=0 ); + if( NEVER(p==0) ) return SQLITE_NOMEM_BKPT; + if( p->xTask==0 ){ + /* assert( p->id==GetCurrentThreadId() ); */ + rc = WAIT_OBJECT_0; + assert( p->tid==0 ); + }else{ + assert( p->id!=0 && p->id!=GetCurrentThreadId() ); + rc = sqlite3Win32Wait((HANDLE)p->tid); + assert( rc!=WAIT_IO_COMPLETION ); + bRc = CloseHandle((HANDLE)p->tid); + assert( bRc ); + } + if( rc==WAIT_OBJECT_0 ) *ppOut = p->pResult; + sqlite3_free(p); + return (rc==WAIT_OBJECT_0) ? SQLITE_OK : SQLITE_ERROR; +} + +#endif /* SQLITE_OS_WIN_THREADS */ +/******************************** End Win32 Threads *************************/ + + +/********************************* Single-Threaded **************************/ +#ifndef SQLITE_THREADS_IMPLEMENTED +/* +** This implementation does not actually create a new thread. It does the +** work of the thread in the main thread, when either the thread is created +** or when it is joined +*/ + +/* A running thread */ +struct SQLiteThread { + void *(*xTask)(void*); /* The routine to run as a thread */ + void *pIn; /* Argument to xTask */ + void *pResult; /* Result of xTask */ +}; + +/* Create a new thread */ +SQLITE_PRIVATE int sqlite3ThreadCreate( + SQLiteThread **ppThread, /* OUT: Write the thread object here */ + void *(*xTask)(void*), /* Routine to run in a separate thread */ + void *pIn /* Argument passed into xTask() */ +){ + SQLiteThread *p; + + assert( ppThread!=0 ); + assert( xTask!=0 ); + *ppThread = 0; + p = sqlite3Malloc(sizeof(*p)); + if( p==0 ) return SQLITE_NOMEM_BKPT; + if( (SQLITE_PTR_TO_INT(p)/17)&1 ){ + p->xTask = xTask; + p->pIn = pIn; + }else{ + p->xTask = 0; + p->pResult = xTask(pIn); + } + *ppThread = p; + return SQLITE_OK; +} + +/* Get the results of the thread */ +SQLITE_PRIVATE int sqlite3ThreadJoin(SQLiteThread *p, void **ppOut){ + + assert( ppOut!=0 ); + if( NEVER(p==0) ) return SQLITE_NOMEM_BKPT; + if( p->xTask ){ + *ppOut = p->xTask(p->pIn); + }else{ + *ppOut = p->pResult; + } + sqlite3_free(p); + +#if defined(SQLITE_TEST) + { + void *pTstAlloc = sqlite3Malloc(10); + if (!pTstAlloc) return SQLITE_NOMEM_BKPT; + sqlite3_free(pTstAlloc); + } +#endif + + return SQLITE_OK; +} + +#endif /* !defined(SQLITE_THREADS_IMPLEMENTED) */ +/****************************** End Single-Threaded *************************/ +#endif /* SQLITE_MAX_WORKER_THREADS>0 */ + +/************** End of threads.c *********************************************/ +/************** Begin file utf.c *********************************************/ +/* +** 2004 April 13 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** This file contains routines used to translate between UTF-8, +** UTF-16, UTF-16BE, and UTF-16LE. +** +** Notes on UTF-8: +** +** Byte-0 Byte-1 Byte-2 Byte-3 Value +** 0xxxxxxx 00000000 00000000 0xxxxxxx +** 110yyyyy 10xxxxxx 00000000 00000yyy yyxxxxxx +** 1110zzzz 10yyyyyy 10xxxxxx 00000000 zzzzyyyy yyxxxxxx +** 11110uuu 10uuzzzz 10yyyyyy 10xxxxxx 000uuuuu zzzzyyyy yyxxxxxx +** +** +** Notes on UTF-16: (with wwww+1==uuuuu) +** +** Word-0 Word-1 Value +** 110110ww wwzzzzyy 110111yy yyxxxxxx 000uuuuu zzzzyyyy yyxxxxxx +** zzzzyyyy yyxxxxxx 00000000 zzzzyyyy yyxxxxxx +** +** +** BOM or Byte Order Mark: +** 0xff 0xfe little-endian utf-16 follows +** 0xfe 0xff big-endian utf-16 follows +** +*/ +/* #include "sqliteInt.h" */ +/* #include */ +/* #include "vdbeInt.h" */ + +#if !defined(SQLITE_AMALGAMATION) && SQLITE_BYTEORDER==0 +/* +** The following constant value is used by the SQLITE_BIGENDIAN and +** SQLITE_LITTLEENDIAN macros. +*/ +SQLITE_PRIVATE const int sqlite3one = 1; +#endif /* SQLITE_AMALGAMATION && SQLITE_BYTEORDER==0 */ + +/* +** This lookup table is used to help decode the first byte of +** a multi-byte UTF8 character. +*/ +static const unsigned char sqlite3Utf8Trans1[] = { + 0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07, + 0x08, 0x09, 0x0a, 0x0b, 0x0c, 0x0d, 0x0e, 0x0f, + 0x10, 0x11, 0x12, 0x13, 0x14, 0x15, 0x16, 0x17, + 0x18, 0x19, 0x1a, 0x1b, 0x1c, 0x1d, 0x1e, 0x1f, + 0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07, + 0x08, 0x09, 0x0a, 0x0b, 0x0c, 0x0d, 0x0e, 0x0f, + 0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07, + 0x00, 0x01, 0x02, 0x03, 0x00, 0x01, 0x00, 0x00, +}; + + +#define WRITE_UTF8(zOut, c) { \ + if( c<0x00080 ){ \ + *zOut++ = (u8)(c&0xFF); \ + } \ + else if( c<0x00800 ){ \ + *zOut++ = 0xC0 + (u8)((c>>6)&0x1F); \ + *zOut++ = 0x80 + (u8)(c & 0x3F); \ + } \ + else if( c<0x10000 ){ \ + *zOut++ = 0xE0 + (u8)((c>>12)&0x0F); \ + *zOut++ = 0x80 + (u8)((c>>6) & 0x3F); \ + *zOut++ = 0x80 + (u8)(c & 0x3F); \ + }else{ \ + *zOut++ = 0xF0 + (u8)((c>>18) & 0x07); \ + *zOut++ = 0x80 + (u8)((c>>12) & 0x3F); \ + *zOut++ = 0x80 + (u8)((c>>6) & 0x3F); \ + *zOut++ = 0x80 + (u8)(c & 0x3F); \ + } \ +} + +#define WRITE_UTF16LE(zOut, c) { \ + if( c<=0xFFFF ){ \ + *zOut++ = (u8)(c&0x00FF); \ + *zOut++ = (u8)((c>>8)&0x00FF); \ + }else{ \ + *zOut++ = (u8)(((c>>10)&0x003F) + (((c-0x10000)>>10)&0x00C0)); \ + *zOut++ = (u8)(0x00D8 + (((c-0x10000)>>18)&0x03)); \ + *zOut++ = (u8)(c&0x00FF); \ + *zOut++ = (u8)(0x00DC + ((c>>8)&0x03)); \ + } \ +} + +#define WRITE_UTF16BE(zOut, c) { \ + if( c<=0xFFFF ){ \ + *zOut++ = (u8)((c>>8)&0x00FF); \ + *zOut++ = (u8)(c&0x00FF); \ + }else{ \ + *zOut++ = (u8)(0x00D8 + (((c-0x10000)>>18)&0x03)); \ + *zOut++ = (u8)(((c>>10)&0x003F) + (((c-0x10000)>>10)&0x00C0)); \ + *zOut++ = (u8)(0x00DC + ((c>>8)&0x03)); \ + *zOut++ = (u8)(c&0x00FF); \ + } \ +} + +#define READ_UTF16LE(zIn, TERM, c){ \ + c = (*zIn++); \ + c += ((*zIn++)<<8); \ + if( c>=0xD800 && c<0xE000 && TERM ){ \ + int c2 = (*zIn++); \ + c2 += ((*zIn++)<<8); \ + c = (c2&0x03FF) + ((c&0x003F)<<10) + (((c&0x03C0)+0x0040)<<10); \ + } \ +} + +#define READ_UTF16BE(zIn, TERM, c){ \ + c = ((*zIn++)<<8); \ + c += (*zIn++); \ + if( c>=0xD800 && c<0xE000 && TERM ){ \ + int c2 = ((*zIn++)<<8); \ + c2 += (*zIn++); \ + c = (c2&0x03FF) + ((c&0x003F)<<10) + (((c&0x03C0)+0x0040)<<10); \ + } \ +} + +/* +** Translate a single UTF-8 character. Return the unicode value. +** +** During translation, assume that the byte that zTerm points +** is a 0x00. +** +** Write a pointer to the next unread byte back into *pzNext. +** +** Notes On Invalid UTF-8: +** +** * This routine never allows a 7-bit character (0x00 through 0x7f) to +** be encoded as a multi-byte character. Any multi-byte character that +** attempts to encode a value between 0x00 and 0x7f is rendered as 0xfffd. +** +** * This routine never allows a UTF16 surrogate value to be encoded. +** If a multi-byte character attempts to encode a value between +** 0xd800 and 0xe000 then it is rendered as 0xfffd. +** +** * Bytes in the range of 0x80 through 0xbf which occur as the first +** byte of a character are interpreted as single-byte characters +** and rendered as themselves even though they are technically +** invalid characters. +** +** * This routine accepts over-length UTF8 encodings +** for unicode values 0x80 and greater. It does not change over-length +** encodings to 0xfffd as some systems recommend. +*/ +#define READ_UTF8(zIn, zTerm, c) \ + c = *(zIn++); \ + if( c>=0xc0 ){ \ + c = sqlite3Utf8Trans1[c-0xc0]; \ + while( zIn!=zTerm && (*zIn & 0xc0)==0x80 ){ \ + c = (c<<6) + (0x3f & *(zIn++)); \ + } \ + if( c<0x80 \ + || (c&0xFFFFF800)==0xD800 \ + || (c&0xFFFFFFFE)==0xFFFE ){ c = 0xFFFD; } \ + } +SQLITE_PRIVATE u32 sqlite3Utf8Read( + const unsigned char **pz /* Pointer to string from which to read char */ +){ + unsigned int c; + + /* Same as READ_UTF8() above but without the zTerm parameter. + ** For this routine, we assume the UTF8 string is always zero-terminated. + */ + c = *((*pz)++); + if( c>=0xc0 ){ + c = sqlite3Utf8Trans1[c-0xc0]; + while( (*(*pz) & 0xc0)==0x80 ){ + c = (c<<6) + (0x3f & *((*pz)++)); + } + if( c<0x80 + || (c&0xFFFFF800)==0xD800 + || (c&0xFFFFFFFE)==0xFFFE ){ c = 0xFFFD; } + } + return c; +} + + + + +/* +** If the TRANSLATE_TRACE macro is defined, the value of each Mem is +** printed on stderr on the way into and out of sqlite3VdbeMemTranslate(). +*/ +/* #define TRANSLATE_TRACE 1 */ + +#ifndef SQLITE_OMIT_UTF16 +/* +** This routine transforms the internal text encoding used by pMem to +** desiredEnc. It is an error if the string is already of the desired +** encoding, or if *pMem does not contain a string value. +*/ +SQLITE_PRIVATE SQLITE_NOINLINE int sqlite3VdbeMemTranslate(Mem *pMem, u8 desiredEnc){ + sqlite3_int64 len; /* Maximum length of output string in bytes */ + unsigned char *zOut; /* Output buffer */ + unsigned char *zIn; /* Input iterator */ + unsigned char *zTerm; /* End of input */ + unsigned char *z; /* Output iterator */ + unsigned int c; + + assert( pMem->db==0 || sqlite3_mutex_held(pMem->db->mutex) ); + assert( pMem->flags&MEM_Str ); + assert( pMem->enc!=desiredEnc ); + assert( pMem->enc!=0 ); + assert( pMem->n>=0 ); + +#if defined(TRANSLATE_TRACE) && defined(SQLITE_DEBUG) + { + char zBuf[100]; + sqlite3VdbeMemPrettyPrint(pMem, zBuf); + fprintf(stderr, "INPUT: %s\n", zBuf); + } +#endif + + /* If the translation is between UTF-16 little and big endian, then + ** all that is required is to swap the byte order. This case is handled + ** differently from the others. + */ + if( pMem->enc!=SQLITE_UTF8 && desiredEnc!=SQLITE_UTF8 ){ + u8 temp; + int rc; + rc = sqlite3VdbeMemMakeWriteable(pMem); + if( rc!=SQLITE_OK ){ + assert( rc==SQLITE_NOMEM ); + return SQLITE_NOMEM_BKPT; + } + zIn = (u8*)pMem->z; + zTerm = &zIn[pMem->n&~1]; + while( zInenc = desiredEnc; + goto translate_out; + } + + /* Set len to the maximum number of bytes required in the output buffer. */ + if( desiredEnc==SQLITE_UTF8 ){ + /* When converting from UTF-16, the maximum growth results from + ** translating a 2-byte character to a 4-byte UTF-8 character. + ** A single byte is required for the output string + ** nul-terminator. + */ + pMem->n &= ~1; + len = 2 * (sqlite3_int64)pMem->n + 1; + }else{ + /* When converting from UTF-8 to UTF-16 the maximum growth is caused + ** when a 1-byte UTF-8 character is translated into a 2-byte UTF-16 + ** character. Two bytes are required in the output buffer for the + ** nul-terminator. + */ + len = 2 * (sqlite3_int64)pMem->n + 2; + } + + /* Set zIn to point at the start of the input buffer and zTerm to point 1 + ** byte past the end. + ** + ** Variable zOut is set to point at the output buffer, space obtained + ** from sqlite3_malloc(). + */ + zIn = (u8*)pMem->z; + zTerm = &zIn[pMem->n]; + zOut = sqlite3DbMallocRaw(pMem->db, len); + if( !zOut ){ + return SQLITE_NOMEM_BKPT; + } + z = zOut; + + if( pMem->enc==SQLITE_UTF8 ){ + if( desiredEnc==SQLITE_UTF16LE ){ + /* UTF-8 -> UTF-16 Little-endian */ + while( zIn UTF-16 Big-endian */ + while( zInn = (int)(z - zOut); + *z++ = 0; + }else{ + assert( desiredEnc==SQLITE_UTF8 ); + if( pMem->enc==SQLITE_UTF16LE ){ + /* UTF-16 Little-endian -> UTF-8 */ + while( zIn UTF-8 */ + while( zInn = (int)(z - zOut); + } + *z = 0; + assert( (pMem->n+(desiredEnc==SQLITE_UTF8?1:2))<=len ); + + c = pMem->flags; + sqlite3VdbeMemRelease(pMem); + pMem->flags = MEM_Str|MEM_Term|(c&(MEM_AffMask|MEM_Subtype)); + pMem->enc = desiredEnc; + pMem->z = (char*)zOut; + pMem->zMalloc = pMem->z; + pMem->szMalloc = sqlite3DbMallocSize(pMem->db, pMem->z); + +translate_out: +#if defined(TRANSLATE_TRACE) && defined(SQLITE_DEBUG) + { + char zBuf[100]; + sqlite3VdbeMemPrettyPrint(pMem, zBuf); + fprintf(stderr, "OUTPUT: %s\n", zBuf); + } +#endif + return SQLITE_OK; +} +#endif /* SQLITE_OMIT_UTF16 */ + +#ifndef SQLITE_OMIT_UTF16 +/* +** This routine checks for a byte-order mark at the beginning of the +** UTF-16 string stored in *pMem. If one is present, it is removed and +** the encoding of the Mem adjusted. This routine does not do any +** byte-swapping, it just sets Mem.enc appropriately. +** +** The allocation (static, dynamic etc.) and encoding of the Mem may be +** changed by this function. +*/ +SQLITE_PRIVATE int sqlite3VdbeMemHandleBom(Mem *pMem){ + int rc = SQLITE_OK; + u8 bom = 0; + + assert( pMem->n>=0 ); + if( pMem->n>1 ){ + u8 b1 = *(u8 *)pMem->z; + u8 b2 = *(((u8 *)pMem->z) + 1); + if( b1==0xFE && b2==0xFF ){ + bom = SQLITE_UTF16BE; + } + if( b1==0xFF && b2==0xFE ){ + bom = SQLITE_UTF16LE; + } + } + + if( bom ){ + rc = sqlite3VdbeMemMakeWriteable(pMem); + if( rc==SQLITE_OK ){ + pMem->n -= 2; + memmove(pMem->z, &pMem->z[2], pMem->n); + pMem->z[pMem->n] = '\0'; + pMem->z[pMem->n+1] = '\0'; + pMem->flags |= MEM_Term; + pMem->enc = bom; + } + } + return rc; +} +#endif /* SQLITE_OMIT_UTF16 */ + +/* +** pZ is a UTF-8 encoded unicode string. If nByte is less than zero, +** return the number of unicode characters in pZ up to (but not including) +** the first 0x00 byte. If nByte is not less than zero, return the +** number of unicode characters in the first nByte of pZ (or up to +** the first 0x00, whichever comes first). +*/ +SQLITE_PRIVATE int sqlite3Utf8CharLen(const char *zIn, int nByte){ + int r = 0; + const u8 *z = (const u8*)zIn; + const u8 *zTerm; + if( nByte>=0 ){ + zTerm = &z[nByte]; + }else{ + zTerm = (const u8*)(-1); + } + assert( z<=zTerm ); + while( *z!=0 && zmallocFailed ){ + sqlite3VdbeMemRelease(&m); + m.z = 0; + } + assert( (m.flags & MEM_Term)!=0 || db->mallocFailed ); + assert( (m.flags & MEM_Str)!=0 || db->mallocFailed ); + assert( m.z || db->mallocFailed ); + return m.z; +} + +/* +** zIn is a UTF-16 encoded unicode string at least nChar characters long. +** Return the number of bytes in the first nChar unicode characters +** in pZ. nChar must be non-negative. +*/ +SQLITE_PRIVATE int sqlite3Utf16ByteLen(const void *zIn, int nChar){ + int c; + unsigned char const *z = zIn; + int n = 0; + + if( SQLITE_UTF16NATIVE==SQLITE_UTF16BE ){ + while( n0 && n<=4 ); + z[0] = 0; + z = zBuf; + c = sqlite3Utf8Read((const u8**)&z); + t = i; + if( i>=0xD800 && i<=0xDFFF ) t = 0xFFFD; + if( (i&0xFFFFFFFE)==0xFFFE ) t = 0xFFFD; + assert( c==t ); + assert( (z-zBuf)==n ); + } + for(i=0; i<0x00110000; i++){ + if( i>=0xD800 && i<0xE000 ) continue; + z = zBuf; + WRITE_UTF16LE(z, i); + n = (int)(z-zBuf); + assert( n>0 && n<=4 ); + z[0] = 0; + z = zBuf; + READ_UTF16LE(z, 1, c); + assert( c==i ); + assert( (z-zBuf)==n ); + } + for(i=0; i<0x00110000; i++){ + if( i>=0xD800 && i<0xE000 ) continue; + z = zBuf; + WRITE_UTF16BE(z, i); + n = (int)(z-zBuf); + assert( n>0 && n<=4 ); + z[0] = 0; + z = zBuf; + READ_UTF16BE(z, 1, c); + assert( c==i ); + assert( (z-zBuf)==n ); + } +} +#endif /* SQLITE_TEST */ +#endif /* SQLITE_OMIT_UTF16 */ + +/************** End of utf.c *************************************************/ +/************** Begin file util.c ********************************************/ +/* +** 2001 September 15 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** Utility functions used throughout sqlite. +** +** This file contains functions for allocating memory, comparing +** strings, and stuff like that. +** +*/ +/* #include "sqliteInt.h" */ +/* #include */ +#include + +/* +** Routine needed to support the testcase() macro. +*/ +#ifdef SQLITE_COVERAGE_TEST +SQLITE_PRIVATE void sqlite3Coverage(int x){ + static unsigned dummy = 0; + dummy += (unsigned)x; +} +#endif + +/* +** Calls to sqlite3FaultSim() are used to simulate a failure during testing, +** or to bypass normal error detection during testing in order to let +** execute proceed futher downstream. +** +** In deployment, sqlite3FaultSim() *always* return SQLITE_OK (0). The +** sqlite3FaultSim() function only returns non-zero during testing. +** +** During testing, if the test harness has set a fault-sim callback using +** a call to sqlite3_test_control(SQLITE_TESTCTRL_FAULT_INSTALL), then +** each call to sqlite3FaultSim() is relayed to that application-supplied +** callback and the integer return value form the application-supplied +** callback is returned by sqlite3FaultSim(). +** +** The integer argument to sqlite3FaultSim() is a code to identify which +** sqlite3FaultSim() instance is being invoked. Each call to sqlite3FaultSim() +** should have a unique code. To prevent legacy testing applications from +** breaking, the codes should not be changed or reused. +*/ +#ifndef SQLITE_UNTESTABLE +SQLITE_PRIVATE int sqlite3FaultSim(int iTest){ + int (*xCallback)(int) = sqlite3GlobalConfig.xTestCallback; + return xCallback ? xCallback(iTest) : SQLITE_OK; +} +#endif + +#ifndef SQLITE_OMIT_FLOATING_POINT +/* +** Return true if the floating point value is Not a Number (NaN). +*/ +SQLITE_PRIVATE int sqlite3IsNaN(double x){ + u64 y; + memcpy(&y,&x,sizeof(y)); + return IsNaN(y); +} +#endif /* SQLITE_OMIT_FLOATING_POINT */ + +/* +** Compute a string length that is limited to what can be stored in +** lower 30 bits of a 32-bit signed integer. +** +** The value returned will never be negative. Nor will it ever be greater +** than the actual length of the string. For very long strings (greater +** than 1GiB) the value returned might be less than the true string length. +*/ +SQLITE_PRIVATE int sqlite3Strlen30(const char *z){ + if( z==0 ) return 0; + return 0x3fffffff & (int)strlen(z); +} + +/* +** Return the declared type of a column. Or return zDflt if the column +** has no declared type. +** +** The column type is an extra string stored after the zero-terminator on +** the column name if and only if the COLFLAG_HASTYPE flag is set. +*/ +SQLITE_PRIVATE char *sqlite3ColumnType(Column *pCol, char *zDflt){ + if( (pCol->colFlags & COLFLAG_HASTYPE)==0 ) return zDflt; + return pCol->zName + strlen(pCol->zName) + 1; +} + +/* +** Helper function for sqlite3Error() - called rarely. Broken out into +** a separate routine to avoid unnecessary register saves on entry to +** sqlite3Error(). +*/ +static SQLITE_NOINLINE void sqlite3ErrorFinish(sqlite3 *db, int err_code){ + if( db->pErr ) sqlite3ValueSetNull(db->pErr); + sqlite3SystemError(db, err_code); +} + +/* +** Set the current error code to err_code and clear any prior error message. +** Also set iSysErrno (by calling sqlite3System) if the err_code indicates +** that would be appropriate. +*/ +SQLITE_PRIVATE void sqlite3Error(sqlite3 *db, int err_code){ + assert( db!=0 ); + db->errCode = err_code; + if( err_code || db->pErr ) sqlite3ErrorFinish(db, err_code); +} + +/* +** Load the sqlite3.iSysErrno field if that is an appropriate thing +** to do based on the SQLite error code in rc. +*/ +SQLITE_PRIVATE void sqlite3SystemError(sqlite3 *db, int rc){ + if( rc==SQLITE_IOERR_NOMEM ) return; + rc &= 0xff; + if( rc==SQLITE_CANTOPEN || rc==SQLITE_IOERR ){ + db->iSysErrno = sqlite3OsGetLastError(db->pVfs); + } +} + +/* +** Set the most recent error code and error string for the sqlite +** handle "db". The error code is set to "err_code". +** +** If it is not NULL, string zFormat specifies the format of the +** error string in the style of the printf functions: The following +** format characters are allowed: +** +** %s Insert a string +** %z A string that should be freed after use +** %d Insert an integer +** %T Insert a token +** %S Insert the first element of a SrcList +** +** zFormat and any string tokens that follow it are assumed to be +** encoded in UTF-8. +** +** To clear the most recent error for sqlite handle "db", sqlite3Error +** should be called with err_code set to SQLITE_OK and zFormat set +** to NULL. +*/ +SQLITE_PRIVATE void sqlite3ErrorWithMsg(sqlite3 *db, int err_code, const char *zFormat, ...){ + assert( db!=0 ); + db->errCode = err_code; + sqlite3SystemError(db, err_code); + if( zFormat==0 ){ + sqlite3Error(db, err_code); + }else if( db->pErr || (db->pErr = sqlite3ValueNew(db))!=0 ){ + char *z; + va_list ap; + va_start(ap, zFormat); + z = sqlite3VMPrintf(db, zFormat, ap); + va_end(ap); + sqlite3ValueSetStr(db->pErr, -1, z, SQLITE_UTF8, SQLITE_DYNAMIC); + } +} + +/* +** Add an error message to pParse->zErrMsg and increment pParse->nErr. +** The following formatting characters are allowed: +** +** %s Insert a string +** %z A string that should be freed after use +** %d Insert an integer +** %T Insert a token +** %S Insert the first element of a SrcList +** +** This function should be used to report any error that occurs while +** compiling an SQL statement (i.e. within sqlite3_prepare()). The +** last thing the sqlite3_prepare() function does is copy the error +** stored by this function into the database handle using sqlite3Error(). +** Functions sqlite3Error() or sqlite3ErrorWithMsg() should be used +** during statement execution (sqlite3_step() etc.). +*/ +SQLITE_PRIVATE void sqlite3ErrorMsg(Parse *pParse, const char *zFormat, ...){ + char *zMsg; + va_list ap; + sqlite3 *db = pParse->db; + va_start(ap, zFormat); + zMsg = sqlite3VMPrintf(db, zFormat, ap); + va_end(ap); + if( db->suppressErr ){ + sqlite3DbFree(db, zMsg); + }else{ + pParse->nErr++; + sqlite3DbFree(db, pParse->zErrMsg); + pParse->zErrMsg = zMsg; + pParse->rc = SQLITE_ERROR; + } +} + +/* +** If database connection db is currently parsing SQL, then transfer +** error code errCode to that parser if the parser has not already +** encountered some other kind of error. +*/ +SQLITE_PRIVATE int sqlite3ErrorToParser(sqlite3 *db, int errCode){ + Parse *pParse; + if( db==0 || (pParse = db->pParse)==0 ) return errCode; + pParse->rc = errCode; + pParse->nErr++; + return errCode; +} + +/* +** Convert an SQL-style quoted string into a normal string by removing +** the quote characters. The conversion is done in-place. If the +** input does not begin with a quote character, then this routine +** is a no-op. +** +** The input string must be zero-terminated. A new zero-terminator +** is added to the dequoted string. +** +** The return value is -1 if no dequoting occurs or the length of the +** dequoted string, exclusive of the zero terminator, if dequoting does +** occur. +** +** 2002-02-14: This routine is extended to remove MS-Access style +** brackets from around identifiers. For example: "[a-b-c]" becomes +** "a-b-c". +*/ +SQLITE_PRIVATE void sqlite3Dequote(char *z){ + char quote; + int i, j; + if( z==0 ) return; + quote = z[0]; + if( !sqlite3Isquote(quote) ) return; + if( quote=='[' ) quote = ']'; + for(i=1, j=0;; i++){ + assert( z[i] ); + if( z[i]==quote ){ + if( z[i+1]==quote ){ + z[j++] = quote; + i++; + }else{ + break; + } + }else{ + z[j++] = z[i]; + } + } + z[j] = 0; +} +SQLITE_PRIVATE void sqlite3DequoteExpr(Expr *p){ + assert( sqlite3Isquote(p->u.zToken[0]) ); + p->flags |= p->u.zToken[0]=='"' ? EP_Quoted|EP_DblQuoted : EP_Quoted; + sqlite3Dequote(p->u.zToken); +} + +/* +** Generate a Token object from a string +*/ +SQLITE_PRIVATE void sqlite3TokenInit(Token *p, char *z){ + p->z = z; + p->n = sqlite3Strlen30(z); +} + +/* Convenient short-hand */ +#define UpperToLower sqlite3UpperToLower + +/* +** Some systems have stricmp(). Others have strcasecmp(). Because +** there is no consistency, we will define our own. +** +** IMPLEMENTATION-OF: R-30243-02494 The sqlite3_stricmp() and +** sqlite3_strnicmp() APIs allow applications and extensions to compare +** the contents of two buffers containing UTF-8 strings in a +** case-independent fashion, using the same definition of "case +** independence" that SQLite uses internally when comparing identifiers. +*/ +SQLITE_API int sqlite3_stricmp(const char *zLeft, const char *zRight){ + if( zLeft==0 ){ + return zRight ? -1 : 0; + }else if( zRight==0 ){ + return 1; + } + return sqlite3StrICmp(zLeft, zRight); +} +SQLITE_PRIVATE int sqlite3StrICmp(const char *zLeft, const char *zRight){ + unsigned char *a, *b; + int c, x; + a = (unsigned char *)zLeft; + b = (unsigned char *)zRight; + for(;;){ + c = *a; + x = *b; + if( c==x ){ + if( c==0 ) break; + }else{ + c = (int)UpperToLower[c] - (int)UpperToLower[x]; + if( c ) break; + } + a++; + b++; + } + return c; +} +SQLITE_API int sqlite3_strnicmp(const char *zLeft, const char *zRight, int N){ + register unsigned char *a, *b; + if( zLeft==0 ){ + return zRight ? -1 : 0; + }else if( zRight==0 ){ + return 1; + } + a = (unsigned char *)zLeft; + b = (unsigned char *)zRight; + while( N-- > 0 && *a!=0 && UpperToLower[*a]==UpperToLower[*b]){ a++; b++; } + return N<0 ? 0 : UpperToLower[*a] - UpperToLower[*b]; +} + +/* +** Compute 10 to the E-th power. Examples: E==1 results in 10. +** E==2 results in 100. E==50 results in 1.0e50. +** +** This routine only works for values of E between 1 and 341. +*/ +static LONGDOUBLE_TYPE sqlite3Pow10(int E){ +#if defined(_MSC_VER) + static const LONGDOUBLE_TYPE x[] = { + 1.0e+001L, + 1.0e+002L, + 1.0e+004L, + 1.0e+008L, + 1.0e+016L, + 1.0e+032L, + 1.0e+064L, + 1.0e+128L, + 1.0e+256L + }; + LONGDOUBLE_TYPE r = 1.0; + int i; + assert( E>=0 && E<=307 ); + for(i=0; E!=0; i++, E >>=1){ + if( E & 1 ) r *= x[i]; + } + return r; +#else + LONGDOUBLE_TYPE x = 10.0; + LONGDOUBLE_TYPE r = 1.0; + while(1){ + if( E & 1 ) r *= x; + E >>= 1; + if( E==0 ) break; + x *= x; + } + return r; +#endif +} + +/* +** The string z[] is an text representation of a real number. +** Convert this string to a double and write it into *pResult. +** +** The string z[] is length bytes in length (bytes, not characters) and +** uses the encoding enc. The string is not necessarily zero-terminated. +** +** Return TRUE if the result is a valid real number (or integer) and FALSE +** if the string is empty or contains extraneous text. More specifically +** return +** 1 => The input string is a pure integer +** 2 or more => The input has a decimal point or eNNN clause +** 0 or less => The input string is not a valid number +** -1 => Not a valid number, but has a valid prefix which +** includes a decimal point and/or an eNNN clause +** +** Valid numbers are in one of these formats: +** +** [+-]digits[E[+-]digits] +** [+-]digits.[digits][E[+-]digits] +** [+-].digits[E[+-]digits] +** +** Leading and trailing whitespace is ignored for the purpose of determining +** validity. +** +** If some prefix of the input string is a valid number, this routine +** returns FALSE but it still converts the prefix and writes the result +** into *pResult. +*/ +SQLITE_PRIVATE int sqlite3AtoF(const char *z, double *pResult, int length, u8 enc){ +#ifndef SQLITE_OMIT_FLOATING_POINT + int incr; + const char *zEnd = z + length; + /* sign * significand * (10 ^ (esign * exponent)) */ + int sign = 1; /* sign of significand */ + i64 s = 0; /* significand */ + int d = 0; /* adjust exponent for shifting decimal point */ + int esign = 1; /* sign of exponent */ + int e = 0; /* exponent */ + int eValid = 1; /* True exponent is either not used or is well-formed */ + double result; + int nDigit = 0; /* Number of digits processed */ + int eType = 1; /* 1: pure integer, 2+: fractional -1 or less: bad UTF16 */ + + assert( enc==SQLITE_UTF8 || enc==SQLITE_UTF16LE || enc==SQLITE_UTF16BE ); + *pResult = 0.0; /* Default return value, in case of an error */ + + if( enc==SQLITE_UTF8 ){ + incr = 1; + }else{ + int i; + incr = 2; + assert( SQLITE_UTF16LE==2 && SQLITE_UTF16BE==3 ); + testcase( enc==SQLITE_UTF16LE ); + testcase( enc==SQLITE_UTF16BE ); + for(i=3-enc; i=zEnd ) return 0; + + /* get sign of significand */ + if( *z=='-' ){ + sign = -1; + z+=incr; + }else if( *z=='+' ){ + z+=incr; + } + + /* copy max significant digits to significand */ + while( z=((LARGEST_INT64-9)/10) ){ + /* skip non-significant significand digits + ** (increase exponent by d to shift decimal left) */ + while( z=zEnd ) goto do_atof_calc; + + /* if decimal point is present */ + if( *z=='.' ){ + z+=incr; + eType++; + /* copy digits from after decimal to significand + ** (decrease exponent by d to shift decimal right) */ + while( z=zEnd ) goto do_atof_calc; + + /* if exponent is present */ + if( *z=='e' || *z=='E' ){ + z+=incr; + eValid = 0; + eType++; + + /* This branch is needed to avoid a (harmless) buffer overread. The + ** special comment alerts the mutation tester that the correct answer + ** is obtained even if the branch is omitted */ + if( z>=zEnd ) goto do_atof_calc; /*PREVENTS-HARMLESS-OVERREAD*/ + + /* get sign of exponent */ + if( *z=='-' ){ + esign = -1; + z+=incr; + }else if( *z=='+' ){ + z+=incr; + } + /* copy digits to exponent */ + while( z0 ){ /*OPTIMIZATION-IF-TRUE*/ + if( esign>0 ){ + if( s>=(LARGEST_INT64/10) ) break; /*OPTIMIZATION-IF-FALSE*/ + s *= 10; + }else{ + if( s%10!=0 ) break; /*OPTIMIZATION-IF-FALSE*/ + s /= 10; + } + e--; + } + + /* adjust the sign of significand */ + s = sign<0 ? -s : s; + + if( e==0 ){ /*OPTIMIZATION-IF-TRUE*/ + result = (double)s; + }else{ + /* attempt to handle extremely small/large numbers better */ + if( e>307 ){ /*OPTIMIZATION-IF-TRUE*/ + if( e<342 ){ /*OPTIMIZATION-IF-TRUE*/ + LONGDOUBLE_TYPE scale = sqlite3Pow10(e-308); + if( esign<0 ){ + result = s / scale; + result /= 1.0e+308; + }else{ + result = s * scale; + result *= 1.0e+308; + } + }else{ assert( e>=342 ); + if( esign<0 ){ + result = 0.0*s; + }else{ +#ifdef INFINITY + result = INFINITY*s; +#else + result = 1e308*1e308*s; /* Infinity */ +#endif + } + } + }else{ + LONGDOUBLE_TYPE scale = sqlite3Pow10(e); + if( esign<0 ){ + result = s / scale; + }else{ + result = s * scale; + } + } + } + } + + /* store the result */ + *pResult = result; + + /* return true if number and no extra non-whitespace chracters after */ + if( z==zEnd && nDigit>0 && eValid && eType>0 ){ + return eType; + }else if( eType>=2 && (eType==3 || eValid) && nDigit>0 ){ + return -1; + }else{ + return 0; + } +#else + return !sqlite3Atoi64(z, pResult, length, enc); +#endif /* SQLITE_OMIT_FLOATING_POINT */ +} + +/* +** Compare the 19-character string zNum against the text representation +** value 2^63: 9223372036854775808. Return negative, zero, or positive +** if zNum is less than, equal to, or greater than the string. +** Note that zNum must contain exactly 19 characters. +** +** Unlike memcmp() this routine is guaranteed to return the difference +** in the values of the last digit if the only difference is in the +** last digit. So, for example, +** +** compare2pow63("9223372036854775800", 1) +** +** will return -8. +*/ +static int compare2pow63(const char *zNum, int incr){ + int c = 0; + int i; + /* 012345678901234567 */ + const char *pow63 = "922337203685477580"; + for(i=0; c==0 && i<18; i++){ + c = (zNum[i*incr]-pow63[i])*10; + } + if( c==0 ){ + c = zNum[18*incr] - '8'; + testcase( c==(-1) ); + testcase( c==0 ); + testcase( c==(+1) ); + } + return c; +} + +/* +** Convert zNum to a 64-bit signed integer. zNum must be decimal. This +** routine does *not* accept hexadecimal notation. +** +** Returns: +** +** -1 Not even a prefix of the input text looks like an integer +** 0 Successful transformation. Fits in a 64-bit signed integer. +** 1 Excess non-space text after the integer value +** 2 Integer too large for a 64-bit signed integer or is malformed +** 3 Special case of 9223372036854775808 +** +** length is the number of bytes in the string (bytes, not characters). +** The string is not necessarily zero-terminated. The encoding is +** given by enc. +*/ +SQLITE_PRIVATE int sqlite3Atoi64(const char *zNum, i64 *pNum, int length, u8 enc){ + int incr; + u64 u = 0; + int neg = 0; /* assume positive */ + int i; + int c = 0; + int nonNum = 0; /* True if input contains UTF16 with high byte non-zero */ + int rc; /* Baseline return code */ + const char *zStart; + const char *zEnd = zNum + length; + assert( enc==SQLITE_UTF8 || enc==SQLITE_UTF16LE || enc==SQLITE_UTF16BE ); + if( enc==SQLITE_UTF8 ){ + incr = 1; + }else{ + incr = 2; + assert( SQLITE_UTF16LE==2 && SQLITE_UTF16BE==3 ); + for(i=3-enc; i='0' && c<='9'; i+=incr){ + u = u*10 + c - '0'; + } + testcase( i==18*incr ); + testcase( i==19*incr ); + testcase( i==20*incr ); + if( u>LARGEST_INT64 ){ + /* This test and assignment is needed only to suppress UB warnings + ** from clang and -fsanitize=undefined. This test and assignment make + ** the code a little larger and slower, and no harm comes from omitting + ** them, but we must appaise the undefined-behavior pharisees. */ + *pNum = neg ? SMALLEST_INT64 : LARGEST_INT64; + }else if( neg ){ + *pNum = -(i64)u; + }else{ + *pNum = (i64)u; + } + rc = 0; + if( i==0 && zStart==zNum ){ /* No digits */ + rc = -1; + }else if( nonNum ){ /* UTF16 with high-order bytes non-zero */ + rc = 1; + }else if( &zNum[i]19*incr ? 1 : compare2pow63(zNum, incr); + if( c<0 ){ + /* zNum is less than 9223372036854775808 so it fits */ + assert( u<=LARGEST_INT64 ); + return rc; + }else{ + *pNum = neg ? SMALLEST_INT64 : LARGEST_INT64; + if( c>0 ){ + /* zNum is greater than 9223372036854775808 so it overflows */ + return 2; + }else{ + /* zNum is exactly 9223372036854775808. Fits if negative. The + ** special case 2 overflow if positive */ + assert( u-1==LARGEST_INT64 ); + return neg ? rc : 3; + } + } + } +} + +/* +** Transform a UTF-8 integer literal, in either decimal or hexadecimal, +** into a 64-bit signed integer. This routine accepts hexadecimal literals, +** whereas sqlite3Atoi64() does not. +** +** Returns: +** +** 0 Successful transformation. Fits in a 64-bit signed integer. +** 1 Excess text after the integer value +** 2 Integer too large for a 64-bit signed integer or is malformed +** 3 Special case of 9223372036854775808 +*/ +SQLITE_PRIVATE int sqlite3DecOrHexToI64(const char *z, i64 *pOut){ +#ifndef SQLITE_OMIT_HEX_INTEGER + if( z[0]=='0' + && (z[1]=='x' || z[1]=='X') + ){ + u64 u = 0; + int i, k; + for(i=2; z[i]=='0'; i++){} + for(k=i; sqlite3Isxdigit(z[k]); k++){ + u = u*16 + sqlite3HexToInt(z[k]); + } + memcpy(pOut, &u, 8); + return (z[k]==0 && k-i<=16) ? 0 : 2; + }else +#endif /* SQLITE_OMIT_HEX_INTEGER */ + { + return sqlite3Atoi64(z, pOut, sqlite3Strlen30(z), SQLITE_UTF8); + } +} + +/* +** If zNum represents an integer that will fit in 32-bits, then set +** *pValue to that integer and return true. Otherwise return false. +** +** This routine accepts both decimal and hexadecimal notation for integers. +** +** Any non-numeric characters that following zNum are ignored. +** This is different from sqlite3Atoi64() which requires the +** input number to be zero-terminated. +*/ +SQLITE_PRIVATE int sqlite3GetInt32(const char *zNum, int *pValue){ + sqlite_int64 v = 0; + int i, c; + int neg = 0; + if( zNum[0]=='-' ){ + neg = 1; + zNum++; + }else if( zNum[0]=='+' ){ + zNum++; + } +#ifndef SQLITE_OMIT_HEX_INTEGER + else if( zNum[0]=='0' + && (zNum[1]=='x' || zNum[1]=='X') + && sqlite3Isxdigit(zNum[2]) + ){ + u32 u = 0; + zNum += 2; + while( zNum[0]=='0' ) zNum++; + for(i=0; sqlite3Isxdigit(zNum[i]) && i<8; i++){ + u = u*16 + sqlite3HexToInt(zNum[i]); + } + if( (u&0x80000000)==0 && sqlite3Isxdigit(zNum[i])==0 ){ + memcpy(pValue, &u, 4); + return 1; + }else{ + return 0; + } + } +#endif + if( !sqlite3Isdigit(zNum[0]) ) return 0; + while( zNum[0]=='0' ) zNum++; + for(i=0; i<11 && (c = zNum[i] - '0')>=0 && c<=9; i++){ + v = v*10 + c; + } + + /* The longest decimal representation of a 32 bit integer is 10 digits: + ** + ** 1234567890 + ** 2^31 -> 2147483648 + */ + testcase( i==10 ); + if( i>10 ){ + return 0; + } + testcase( v-neg==2147483647 ); + if( v-neg>2147483647 ){ + return 0; + } + if( neg ){ + v = -v; + } + *pValue = (int)v; + return 1; +} + +/* +** Return a 32-bit integer value extracted from a string. If the +** string is not an integer, just return 0. +*/ +SQLITE_PRIVATE int sqlite3Atoi(const char *z){ + int x = 0; + if( z ) sqlite3GetInt32(z, &x); + return x; +} + +/* +** The variable-length integer encoding is as follows: +** +** KEY: +** A = 0xxxxxxx 7 bits of data and one flag bit +** B = 1xxxxxxx 7 bits of data and one flag bit +** C = xxxxxxxx 8 bits of data +** +** 7 bits - A +** 14 bits - BA +** 21 bits - BBA +** 28 bits - BBBA +** 35 bits - BBBBA +** 42 bits - BBBBBA +** 49 bits - BBBBBBA +** 56 bits - BBBBBBBA +** 64 bits - BBBBBBBBC +*/ + +/* +** Write a 64-bit variable-length integer to memory starting at p[0]. +** The length of data write will be between 1 and 9 bytes. The number +** of bytes written is returned. +** +** A variable-length integer consists of the lower 7 bits of each byte +** for all bytes that have the 8th bit set and one byte with the 8th +** bit clear. Except, if we get to the 9th byte, it stores the full +** 8 bits and is the last byte. +*/ +static int SQLITE_NOINLINE putVarint64(unsigned char *p, u64 v){ + int i, j, n; + u8 buf[10]; + if( v & (((u64)0xff000000)<<32) ){ + p[8] = (u8)v; + v >>= 8; + for(i=7; i>=0; i--){ + p[i] = (u8)((v & 0x7f) | 0x80); + v >>= 7; + } + return 9; + } + n = 0; + do{ + buf[n++] = (u8)((v & 0x7f) | 0x80); + v >>= 7; + }while( v!=0 ); + buf[0] &= 0x7f; + assert( n<=9 ); + for(i=0, j=n-1; j>=0; j--, i++){ + p[i] = buf[j]; + } + return n; +} +SQLITE_PRIVATE int sqlite3PutVarint(unsigned char *p, u64 v){ + if( v<=0x7f ){ + p[0] = v&0x7f; + return 1; + } + if( v<=0x3fff ){ + p[0] = ((v>>7)&0x7f)|0x80; + p[1] = v&0x7f; + return 2; + } + return putVarint64(p,v); +} + +/* +** Bitmasks used by sqlite3GetVarint(). These precomputed constants +** are defined here rather than simply putting the constant expressions +** inline in order to work around bugs in the RVT compiler. +** +** SLOT_2_0 A mask for (0x7f<<14) | 0x7f +** +** SLOT_4_2_0 A mask for (0x7f<<28) | SLOT_2_0 +*/ +#define SLOT_2_0 0x001fc07f +#define SLOT_4_2_0 0xf01fc07f + + +/* +** Read a 64-bit variable-length integer from memory starting at p[0]. +** Return the number of bytes read. The value is stored in *v. +*/ +SQLITE_PRIVATE u8 sqlite3GetVarint(const unsigned char *p, u64 *v){ + u32 a,b,s; + + if( ((signed char*)p)[0]>=0 ){ + *v = *p; + return 1; + } + if( ((signed char*)p)[1]>=0 ){ + *v = ((u32)(p[0]&0x7f)<<7) | p[1]; + return 2; + } + + /* Verify that constants are precomputed correctly */ + assert( SLOT_2_0 == ((0x7f<<14) | (0x7f)) ); + assert( SLOT_4_2_0 == ((0xfU<<28) | (0x7f<<14) | (0x7f)) ); + + a = ((u32)p[0])<<14; + b = p[1]; + p += 2; + a |= *p; + /* a: p0<<14 | p2 (unmasked) */ + if (!(a&0x80)) + { + a &= SLOT_2_0; + b &= 0x7f; + b = b<<7; + a |= b; + *v = a; + return 3; + } + + /* CSE1 from below */ + a &= SLOT_2_0; + p++; + b = b<<14; + b |= *p; + /* b: p1<<14 | p3 (unmasked) */ + if (!(b&0x80)) + { + b &= SLOT_2_0; + /* moved CSE1 up */ + /* a &= (0x7f<<14)|(0x7f); */ + a = a<<7; + a |= b; + *v = a; + return 4; + } + + /* a: p0<<14 | p2 (masked) */ + /* b: p1<<14 | p3 (unmasked) */ + /* 1:save off p0<<21 | p1<<14 | p2<<7 | p3 (masked) */ + /* moved CSE1 up */ + /* a &= (0x7f<<14)|(0x7f); */ + b &= SLOT_2_0; + s = a; + /* s: p0<<14 | p2 (masked) */ + + p++; + a = a<<14; + a |= *p; + /* a: p0<<28 | p2<<14 | p4 (unmasked) */ + if (!(a&0x80)) + { + /* we can skip these cause they were (effectively) done above + ** while calculating s */ + /* a &= (0x7f<<28)|(0x7f<<14)|(0x7f); */ + /* b &= (0x7f<<14)|(0x7f); */ + b = b<<7; + a |= b; + s = s>>18; + *v = ((u64)s)<<32 | a; + return 5; + } + + /* 2:save off p0<<21 | p1<<14 | p2<<7 | p3 (masked) */ + s = s<<7; + s |= b; + /* s: p0<<21 | p1<<14 | p2<<7 | p3 (masked) */ + + p++; + b = b<<14; + b |= *p; + /* b: p1<<28 | p3<<14 | p5 (unmasked) */ + if (!(b&0x80)) + { + /* we can skip this cause it was (effectively) done above in calc'ing s */ + /* b &= (0x7f<<28)|(0x7f<<14)|(0x7f); */ + a &= SLOT_2_0; + a = a<<7; + a |= b; + s = s>>18; + *v = ((u64)s)<<32 | a; + return 6; + } + + p++; + a = a<<14; + a |= *p; + /* a: p2<<28 | p4<<14 | p6 (unmasked) */ + if (!(a&0x80)) + { + a &= SLOT_4_2_0; + b &= SLOT_2_0; + b = b<<7; + a |= b; + s = s>>11; + *v = ((u64)s)<<32 | a; + return 7; + } + + /* CSE2 from below */ + a &= SLOT_2_0; + p++; + b = b<<14; + b |= *p; + /* b: p3<<28 | p5<<14 | p7 (unmasked) */ + if (!(b&0x80)) + { + b &= SLOT_4_2_0; + /* moved CSE2 up */ + /* a &= (0x7f<<14)|(0x7f); */ + a = a<<7; + a |= b; + s = s>>4; + *v = ((u64)s)<<32 | a; + return 8; + } + + p++; + a = a<<15; + a |= *p; + /* a: p4<<29 | p6<<15 | p8 (unmasked) */ + + /* moved CSE2 up */ + /* a &= (0x7f<<29)|(0x7f<<15)|(0xff); */ + b &= SLOT_2_0; + b = b<<8; + a |= b; + + s = s<<4; + b = p[-4]; + b &= 0x7f; + b = b>>3; + s |= b; + + *v = ((u64)s)<<32 | a; + + return 9; +} + +/* +** Read a 32-bit variable-length integer from memory starting at p[0]. +** Return the number of bytes read. The value is stored in *v. +** +** If the varint stored in p[0] is larger than can fit in a 32-bit unsigned +** integer, then set *v to 0xffffffff. +** +** A MACRO version, getVarint32, is provided which inlines the +** single-byte case. All code should use the MACRO version as +** this function assumes the single-byte case has already been handled. +*/ +SQLITE_PRIVATE u8 sqlite3GetVarint32(const unsigned char *p, u32 *v){ + u32 a,b; + + /* The 1-byte case. Overwhelmingly the most common. Handled inline + ** by the getVarin32() macro */ + a = *p; + /* a: p0 (unmasked) */ +#ifndef getVarint32 + if (!(a&0x80)) + { + /* Values between 0 and 127 */ + *v = a; + return 1; + } +#endif + + /* The 2-byte case */ + p++; + b = *p; + /* b: p1 (unmasked) */ + if (!(b&0x80)) + { + /* Values between 128 and 16383 */ + a &= 0x7f; + a = a<<7; + *v = a | b; + return 2; + } + + /* The 3-byte case */ + p++; + a = a<<14; + a |= *p; + /* a: p0<<14 | p2 (unmasked) */ + if (!(a&0x80)) + { + /* Values between 16384 and 2097151 */ + a &= (0x7f<<14)|(0x7f); + b &= 0x7f; + b = b<<7; + *v = a | b; + return 3; + } + + /* A 32-bit varint is used to store size information in btrees. + ** Objects are rarely larger than 2MiB limit of a 3-byte varint. + ** A 3-byte varint is sufficient, for example, to record the size + ** of a 1048569-byte BLOB or string. + ** + ** We only unroll the first 1-, 2-, and 3- byte cases. The very + ** rare larger cases can be handled by the slower 64-bit varint + ** routine. + */ +#if 1 + { + u64 v64; + u8 n; + + p -= 2; + n = sqlite3GetVarint(p, &v64); + assert( n>3 && n<=9 ); + if( (v64 & SQLITE_MAX_U32)!=v64 ){ + *v = 0xffffffff; + }else{ + *v = (u32)v64; + } + return n; + } + +#else + /* For following code (kept for historical record only) shows an + ** unrolling for the 3- and 4-byte varint cases. This code is + ** slightly faster, but it is also larger and much harder to test. + */ + p++; + b = b<<14; + b |= *p; + /* b: p1<<14 | p3 (unmasked) */ + if (!(b&0x80)) + { + /* Values between 2097152 and 268435455 */ + b &= (0x7f<<14)|(0x7f); + a &= (0x7f<<14)|(0x7f); + a = a<<7; + *v = a | b; + return 4; + } + + p++; + a = a<<14; + a |= *p; + /* a: p0<<28 | p2<<14 | p4 (unmasked) */ + if (!(a&0x80)) + { + /* Values between 268435456 and 34359738367 */ + a &= SLOT_4_2_0; + b &= SLOT_4_2_0; + b = b<<7; + *v = a | b; + return 5; + } + + /* We can only reach this point when reading a corrupt database + ** file. In that case we are not in any hurry. Use the (relatively + ** slow) general-purpose sqlite3GetVarint() routine to extract the + ** value. */ + { + u64 v64; + u8 n; + + p -= 4; + n = sqlite3GetVarint(p, &v64); + assert( n>5 && n<=9 ); + *v = (u32)v64; + return n; + } +#endif +} + +/* +** Return the number of bytes that will be needed to store the given +** 64-bit integer. +*/ +SQLITE_PRIVATE int sqlite3VarintLen(u64 v){ + int i; + for(i=1; (v >>= 7)!=0; i++){ assert( i<10 ); } + return i; +} + + +/* +** Read or write a four-byte big-endian integer value. +*/ +SQLITE_PRIVATE u32 sqlite3Get4byte(const u8 *p){ +#if SQLITE_BYTEORDER==4321 + u32 x; + memcpy(&x,p,4); + return x; +#elif SQLITE_BYTEORDER==1234 && GCC_VERSION>=4003000 + u32 x; + memcpy(&x,p,4); + return __builtin_bswap32(x); +#elif SQLITE_BYTEORDER==1234 && MSVC_VERSION>=1300 + u32 x; + memcpy(&x,p,4); + return _byteswap_ulong(x); +#else + testcase( p[0]&0x80 ); + return ((unsigned)p[0]<<24) | (p[1]<<16) | (p[2]<<8) | p[3]; +#endif +} +SQLITE_PRIVATE void sqlite3Put4byte(unsigned char *p, u32 v){ +#if SQLITE_BYTEORDER==4321 + memcpy(p,&v,4); +#elif SQLITE_BYTEORDER==1234 && GCC_VERSION>=4003000 + u32 x = __builtin_bswap32(v); + memcpy(p,&x,4); +#elif SQLITE_BYTEORDER==1234 && MSVC_VERSION>=1300 + u32 x = _byteswap_ulong(v); + memcpy(p,&x,4); +#else + p[0] = (u8)(v>>24); + p[1] = (u8)(v>>16); + p[2] = (u8)(v>>8); + p[3] = (u8)v; +#endif +} + + + +/* +** Translate a single byte of Hex into an integer. +** This routine only works if h really is a valid hexadecimal +** character: 0..9a..fA..F +*/ +SQLITE_PRIVATE u8 sqlite3HexToInt(int h){ + assert( (h>='0' && h<='9') || (h>='a' && h<='f') || (h>='A' && h<='F') ); +#ifdef SQLITE_ASCII + h += 9*(1&(h>>6)); +#endif +#ifdef SQLITE_EBCDIC + h += 9*(1&~(h>>4)); +#endif + return (u8)(h & 0xf); +} + +#if !defined(SQLITE_OMIT_BLOB_LITERAL) || defined(SQLITE_HAS_CODEC) +/* +** Convert a BLOB literal of the form "x'hhhhhh'" into its binary +** value. Return a pointer to its binary value. Space to hold the +** binary value has been obtained from malloc and must be freed by +** the calling routine. +*/ +SQLITE_PRIVATE void *sqlite3HexToBlob(sqlite3 *db, const char *z, int n){ + char *zBlob; + int i; + + zBlob = (char *)sqlite3DbMallocRawNN(db, n/2 + 1); + n--; + if( zBlob ){ + for(i=0; imagic; + if( magic!=SQLITE_MAGIC_OPEN ){ + if( sqlite3SafetyCheckSickOrOk(db) ){ + testcase( sqlite3GlobalConfig.xLog!=0 ); + logBadConnection("unopened"); + } + return 0; + }else{ + return 1; + } +} +SQLITE_PRIVATE int sqlite3SafetyCheckSickOrOk(sqlite3 *db){ + u32 magic; + magic = db->magic; + if( magic!=SQLITE_MAGIC_SICK && + magic!=SQLITE_MAGIC_OPEN && + magic!=SQLITE_MAGIC_BUSY ){ + testcase( sqlite3GlobalConfig.xLog!=0 ); + logBadConnection("invalid"); + return 0; + }else{ + return 1; + } +} + +/* +** Attempt to add, substract, or multiply the 64-bit signed value iB against +** the other 64-bit signed integer at *pA and store the result in *pA. +** Return 0 on success. Or if the operation would have resulted in an +** overflow, leave *pA unchanged and return 1. +*/ +SQLITE_PRIVATE int sqlite3AddInt64(i64 *pA, i64 iB){ +#if GCC_VERSION>=5004000 && !defined(__INTEL_COMPILER) + return __builtin_add_overflow(*pA, iB, pA); +#else + i64 iA = *pA; + testcase( iA==0 ); testcase( iA==1 ); + testcase( iB==-1 ); testcase( iB==0 ); + if( iB>=0 ){ + testcase( iA>0 && LARGEST_INT64 - iA == iB ); + testcase( iA>0 && LARGEST_INT64 - iA == iB - 1 ); + if( iA>0 && LARGEST_INT64 - iA < iB ) return 1; + }else{ + testcase( iA<0 && -(iA + LARGEST_INT64) == iB + 1 ); + testcase( iA<0 && -(iA + LARGEST_INT64) == iB + 2 ); + if( iA<0 && -(iA + LARGEST_INT64) > iB + 1 ) return 1; + } + *pA += iB; + return 0; +#endif +} +SQLITE_PRIVATE int sqlite3SubInt64(i64 *pA, i64 iB){ +#if GCC_VERSION>=5004000 && !defined(__INTEL_COMPILER) + return __builtin_sub_overflow(*pA, iB, pA); +#else + testcase( iB==SMALLEST_INT64+1 ); + if( iB==SMALLEST_INT64 ){ + testcase( (*pA)==(-1) ); testcase( (*pA)==0 ); + if( (*pA)>=0 ) return 1; + *pA -= iB; + return 0; + }else{ + return sqlite3AddInt64(pA, -iB); + } +#endif +} +SQLITE_PRIVATE int sqlite3MulInt64(i64 *pA, i64 iB){ +#if GCC_VERSION>=5004000 && !defined(__INTEL_COMPILER) + return __builtin_mul_overflow(*pA, iB, pA); +#else + i64 iA = *pA; + if( iB>0 ){ + if( iA>LARGEST_INT64/iB ) return 1; + if( iA0 ){ + if( iBLARGEST_INT64/-iB ) return 1; + } + } + *pA = iA*iB; + return 0; +#endif +} + +/* +** Compute the absolute value of a 32-bit signed integer, of possible. Or +** if the integer has a value of -2147483648, return +2147483647 +*/ +SQLITE_PRIVATE int sqlite3AbsInt32(int x){ + if( x>=0 ) return x; + if( x==(int)0x80000000 ) return 0x7fffffff; + return -x; +} + +#ifdef SQLITE_ENABLE_8_3_NAMES +/* +** If SQLITE_ENABLE_8_3_NAMES is set at compile-time and if the database +** filename in zBaseFilename is a URI with the "8_3_names=1" parameter and +** if filename in z[] has a suffix (a.k.a. "extension") that is longer than +** three characters, then shorten the suffix on z[] to be the last three +** characters of the original suffix. +** +** If SQLITE_ENABLE_8_3_NAMES is set to 2 at compile-time, then always +** do the suffix shortening regardless of URI parameter. +** +** Examples: +** +** test.db-journal => test.nal +** test.db-wal => test.wal +** test.db-shm => test.shm +** test.db-mj7f3319fa => test.9fa +*/ +SQLITE_PRIVATE void sqlite3FileSuffix3(const char *zBaseFilename, char *z){ +#if SQLITE_ENABLE_8_3_NAMES<2 + if( sqlite3_uri_boolean(zBaseFilename, "8_3_names", 0) ) +#endif + { + int i, sz; + sz = sqlite3Strlen30(z); + for(i=sz-1; i>0 && z[i]!='/' && z[i]!='.'; i--){} + if( z[i]=='.' && ALWAYS(sz>i+4) ) memmove(&z[i+1], &z[sz-3], 4); + } +} +#endif + +/* +** Find (an approximate) sum of two LogEst values. This computation is +** not a simple "+" operator because LogEst is stored as a logarithmic +** value. +** +*/ +SQLITE_PRIVATE LogEst sqlite3LogEstAdd(LogEst a, LogEst b){ + static const unsigned char x[] = { + 10, 10, /* 0,1 */ + 9, 9, /* 2,3 */ + 8, 8, /* 4,5 */ + 7, 7, 7, /* 6,7,8 */ + 6, 6, 6, /* 9,10,11 */ + 5, 5, 5, /* 12-14 */ + 4, 4, 4, 4, /* 15-18 */ + 3, 3, 3, 3, 3, 3, /* 19-24 */ + 2, 2, 2, 2, 2, 2, 2, /* 25-31 */ + }; + if( a>=b ){ + if( a>b+49 ) return a; + if( a>b+31 ) return a+1; + return a+x[a-b]; + }else{ + if( b>a+49 ) return b; + if( b>a+31 ) return b+1; + return b+x[b-a]; + } +} + +/* +** Convert an integer into a LogEst. In other words, compute an +** approximation for 10*log2(x). +*/ +SQLITE_PRIVATE LogEst sqlite3LogEst(u64 x){ + static LogEst a[] = { 0, 2, 3, 5, 6, 7, 8, 9 }; + LogEst y = 40; + if( x<8 ){ + if( x<2 ) return 0; + while( x<8 ){ y -= 10; x <<= 1; } + }else{ +#if GCC_VERSION>=5004000 + int i = 60 - __builtin_clzll(x); + y += i*10; + x >>= i; +#else + while( x>255 ){ y += 40; x >>= 4; } /*OPTIMIZATION-IF-TRUE*/ + while( x>15 ){ y += 10; x >>= 1; } +#endif + } + return a[x&7] + y - 10; +} + +#ifndef SQLITE_OMIT_VIRTUALTABLE +/* +** Convert a double into a LogEst +** In other words, compute an approximation for 10*log2(x). +*/ +SQLITE_PRIVATE LogEst sqlite3LogEstFromDouble(double x){ + u64 a; + LogEst e; + assert( sizeof(x)==8 && sizeof(a)==8 ); + if( x<=1 ) return 0; + if( x<=2000000000 ) return sqlite3LogEst((u64)x); + memcpy(&a, &x, 8); + e = (a>>52) - 1022; + return e*10; +} +#endif /* SQLITE_OMIT_VIRTUALTABLE */ + +#if defined(SQLITE_ENABLE_STMT_SCANSTATUS) || \ + defined(SQLITE_ENABLE_STAT4) || \ + defined(SQLITE_EXPLAIN_ESTIMATED_ROWS) +/* +** Convert a LogEst into an integer. +** +** Note that this routine is only used when one or more of various +** non-standard compile-time options is enabled. +*/ +SQLITE_PRIVATE u64 sqlite3LogEstToInt(LogEst x){ + u64 n; + n = x%10; + x /= 10; + if( n>=5 ) n -= 2; + else if( n>=1 ) n -= 1; +#if defined(SQLITE_ENABLE_STMT_SCANSTATUS) || \ + defined(SQLITE_EXPLAIN_ESTIMATED_ROWS) + if( x>60 ) return (u64)LARGEST_INT64; +#else + /* If only SQLITE_ENABLE_STAT4 is on, then the largest input + ** possible to this routine is 310, resulting in a maximum x of 31 */ + assert( x<=60 ); +#endif + return x>=3 ? (n+8)<<(x-3) : (n+8)>>(3-x); +} +#endif /* defined SCANSTAT or STAT4 or ESTIMATED_ROWS */ + +/* +** Add a new name/number pair to a VList. This might require that the +** VList object be reallocated, so return the new VList. If an OOM +** error occurs, the original VList returned and the +** db->mallocFailed flag is set. +** +** A VList is really just an array of integers. To destroy a VList, +** simply pass it to sqlite3DbFree(). +** +** The first integer is the number of integers allocated for the whole +** VList. The second integer is the number of integers actually used. +** Each name/number pair is encoded by subsequent groups of 3 or more +** integers. +** +** Each name/number pair starts with two integers which are the numeric +** value for the pair and the size of the name/number pair, respectively. +** The text name overlays one or more following integers. The text name +** is always zero-terminated. +** +** Conceptually: +** +** struct VList { +** int nAlloc; // Number of allocated slots +** int nUsed; // Number of used slots +** struct VListEntry { +** int iValue; // Value for this entry +** int nSlot; // Slots used by this entry +** // ... variable name goes here +** } a[0]; +** } +** +** During code generation, pointers to the variable names within the +** VList are taken. When that happens, nAlloc is set to zero as an +** indication that the VList may never again be enlarged, since the +** accompanying realloc() would invalidate the pointers. +*/ +SQLITE_PRIVATE VList *sqlite3VListAdd( + sqlite3 *db, /* The database connection used for malloc() */ + VList *pIn, /* The input VList. Might be NULL */ + const char *zName, /* Name of symbol to add */ + int nName, /* Bytes of text in zName */ + int iVal /* Value to associate with zName */ +){ + int nInt; /* number of sizeof(int) objects needed for zName */ + char *z; /* Pointer to where zName will be stored */ + int i; /* Index in pIn[] where zName is stored */ + + nInt = nName/4 + 3; + assert( pIn==0 || pIn[0]>=3 ); /* Verify ok to add new elements */ + if( pIn==0 || pIn[1]+nInt > pIn[0] ){ + /* Enlarge the allocation */ + sqlite3_int64 nAlloc = (pIn ? 2*(sqlite3_int64)pIn[0] : 10) + nInt; + VList *pOut = sqlite3DbRealloc(db, pIn, nAlloc*sizeof(int)); + if( pOut==0 ) return pIn; + if( pIn==0 ) pOut[1] = 2; + pIn = pOut; + pIn[0] = nAlloc; + } + i = pIn[1]; + pIn[i] = iVal; + pIn[i+1] = nInt; + z = (char*)&pIn[i+2]; + pIn[1] = i+nInt; + assert( pIn[1]<=pIn[0] ); + memcpy(z, zName, nName); + z[nName] = 0; + return pIn; +} + +/* +** Return a pointer to the name of a variable in the given VList that +** has the value iVal. Or return a NULL if there is no such variable in +** the list +*/ +SQLITE_PRIVATE const char *sqlite3VListNumToName(VList *pIn, int iVal){ + int i, mx; + if( pIn==0 ) return 0; + mx = pIn[1]; + i = 2; + do{ + if( pIn[i]==iVal ) return (char*)&pIn[i+2]; + i += pIn[i+1]; + }while( i */ + +/* Turn bulk memory into a hash table object by initializing the +** fields of the Hash structure. +** +** "pNew" is a pointer to the hash table that is to be initialized. +*/ +SQLITE_PRIVATE void sqlite3HashInit(Hash *pNew){ + assert( pNew!=0 ); + pNew->first = 0; + pNew->count = 0; + pNew->htsize = 0; + pNew->ht = 0; +} + +/* Remove all entries from a hash table. Reclaim all memory. +** Call this routine to delete a hash table or to reset a hash table +** to the empty state. +*/ +SQLITE_PRIVATE void sqlite3HashClear(Hash *pH){ + HashElem *elem; /* For looping over all elements of the table */ + + assert( pH!=0 ); + elem = pH->first; + pH->first = 0; + sqlite3_free(pH->ht); + pH->ht = 0; + pH->htsize = 0; + while( elem ){ + HashElem *next_elem = elem->next; + sqlite3_free(elem); + elem = next_elem; + } + pH->count = 0; +} + +/* +** The hashing function. +*/ +static unsigned int strHash(const char *z){ + unsigned int h = 0; + unsigned char c; + while( (c = (unsigned char)*z++)!=0 ){ /*OPTIMIZATION-IF-TRUE*/ + /* Knuth multiplicative hashing. (Sorting & Searching, p. 510). + ** 0x9e3779b1 is 2654435761 which is the closest prime number to + ** (2**32)*golden_ratio, where golden_ratio = (sqrt(5) - 1)/2. */ + h += sqlite3UpperToLower[c]; + h *= 0x9e3779b1; + } + return h; +} + + +/* Link pNew element into the hash table pH. If pEntry!=0 then also +** insert pNew into the pEntry hash bucket. +*/ +static void insertElement( + Hash *pH, /* The complete hash table */ + struct _ht *pEntry, /* The entry into which pNew is inserted */ + HashElem *pNew /* The element to be inserted */ +){ + HashElem *pHead; /* First element already in pEntry */ + if( pEntry ){ + pHead = pEntry->count ? pEntry->chain : 0; + pEntry->count++; + pEntry->chain = pNew; + }else{ + pHead = 0; + } + if( pHead ){ + pNew->next = pHead; + pNew->prev = pHead->prev; + if( pHead->prev ){ pHead->prev->next = pNew; } + else { pH->first = pNew; } + pHead->prev = pNew; + }else{ + pNew->next = pH->first; + if( pH->first ){ pH->first->prev = pNew; } + pNew->prev = 0; + pH->first = pNew; + } +} + + +/* Resize the hash table so that it cantains "new_size" buckets. +** +** The hash table might fail to resize if sqlite3_malloc() fails or +** if the new size is the same as the prior size. +** Return TRUE if the resize occurs and false if not. +*/ +static int rehash(Hash *pH, unsigned int new_size){ + struct _ht *new_ht; /* The new hash table */ + HashElem *elem, *next_elem; /* For looping over existing elements */ + +#if SQLITE_MALLOC_SOFT_LIMIT>0 + if( new_size*sizeof(struct _ht)>SQLITE_MALLOC_SOFT_LIMIT ){ + new_size = SQLITE_MALLOC_SOFT_LIMIT/sizeof(struct _ht); + } + if( new_size==pH->htsize ) return 0; +#endif + + /* The inability to allocates space for a larger hash table is + ** a performance hit but it is not a fatal error. So mark the + ** allocation as a benign. Use sqlite3Malloc()/memset(0) instead of + ** sqlite3MallocZero() to make the allocation, as sqlite3MallocZero() + ** only zeroes the requested number of bytes whereas this module will + ** use the actual amount of space allocated for the hash table (which + ** may be larger than the requested amount). + */ + sqlite3BeginBenignMalloc(); + new_ht = (struct _ht *)sqlite3Malloc( new_size*sizeof(struct _ht) ); + sqlite3EndBenignMalloc(); + + if( new_ht==0 ) return 0; + sqlite3_free(pH->ht); + pH->ht = new_ht; + pH->htsize = new_size = sqlite3MallocSize(new_ht)/sizeof(struct _ht); + memset(new_ht, 0, new_size*sizeof(struct _ht)); + for(elem=pH->first, pH->first=0; elem; elem = next_elem){ + unsigned int h = strHash(elem->pKey) % new_size; + next_elem = elem->next; + insertElement(pH, &new_ht[h], elem); + } + return 1; +} + +/* This function (for internal use only) locates an element in an +** hash table that matches the given key. If no element is found, +** a pointer to a static null element with HashElem.data==0 is returned. +** If pH is not NULL, then the hash for this key is written to *pH. +*/ +static HashElem *findElementWithHash( + const Hash *pH, /* The pH to be searched */ + const char *pKey, /* The key we are searching for */ + unsigned int *pHash /* Write the hash value here */ +){ + HashElem *elem; /* Used to loop thru the element list */ + unsigned int count; /* Number of elements left to test */ + unsigned int h; /* The computed hash */ + static HashElem nullElement = { 0, 0, 0, 0 }; + + if( pH->ht ){ /*OPTIMIZATION-IF-TRUE*/ + struct _ht *pEntry; + h = strHash(pKey) % pH->htsize; + pEntry = &pH->ht[h]; + elem = pEntry->chain; + count = pEntry->count; + }else{ + h = 0; + elem = pH->first; + count = pH->count; + } + if( pHash ) *pHash = h; + while( count-- ){ + assert( elem!=0 ); + if( sqlite3StrICmp(elem->pKey,pKey)==0 ){ + return elem; + } + elem = elem->next; + } + return &nullElement; +} + +/* Remove a single entry from the hash table given a pointer to that +** element and a hash on the element's key. +*/ +static void removeElementGivenHash( + Hash *pH, /* The pH containing "elem" */ + HashElem* elem, /* The element to be removed from the pH */ + unsigned int h /* Hash value for the element */ +){ + struct _ht *pEntry; + if( elem->prev ){ + elem->prev->next = elem->next; + }else{ + pH->first = elem->next; + } + if( elem->next ){ + elem->next->prev = elem->prev; + } + if( pH->ht ){ + pEntry = &pH->ht[h]; + if( pEntry->chain==elem ){ + pEntry->chain = elem->next; + } + assert( pEntry->count>0 ); + pEntry->count--; + } + sqlite3_free( elem ); + pH->count--; + if( pH->count==0 ){ + assert( pH->first==0 ); + assert( pH->count==0 ); + sqlite3HashClear(pH); + } +} + +/* Attempt to locate an element of the hash table pH with a key +** that matches pKey. Return the data for this element if it is +** found, or NULL if there is no match. +*/ +SQLITE_PRIVATE void *sqlite3HashFind(const Hash *pH, const char *pKey){ + assert( pH!=0 ); + assert( pKey!=0 ); + return findElementWithHash(pH, pKey, 0)->data; +} + +/* Insert an element into the hash table pH. The key is pKey +** and the data is "data". +** +** If no element exists with a matching key, then a new +** element is created and NULL is returned. +** +** If another element already exists with the same key, then the +** new data replaces the old data and the old data is returned. +** The key is not copied in this instance. If a malloc fails, then +** the new data is returned and the hash table is unchanged. +** +** If the "data" parameter to this function is NULL, then the +** element corresponding to "key" is removed from the hash table. +*/ +SQLITE_PRIVATE void *sqlite3HashInsert(Hash *pH, const char *pKey, void *data){ + unsigned int h; /* the hash of the key modulo hash table size */ + HashElem *elem; /* Used to loop thru the element list */ + HashElem *new_elem; /* New element added to the pH */ + + assert( pH!=0 ); + assert( pKey!=0 ); + elem = findElementWithHash(pH,pKey,&h); + if( elem->data ){ + void *old_data = elem->data; + if( data==0 ){ + removeElementGivenHash(pH,elem,h); + }else{ + elem->data = data; + elem->pKey = pKey; + } + return old_data; + } + if( data==0 ) return 0; + new_elem = (HashElem*)sqlite3Malloc( sizeof(HashElem) ); + if( new_elem==0 ) return data; + new_elem->pKey = pKey; + new_elem->data = data; + pH->count++; + if( pH->count>=10 && pH->count > 2*pH->htsize ){ + if( rehash(pH, pH->count*2) ){ + assert( pH->htsize>0 ); + h = strHash(pKey) % pH->htsize; + } + } + insertElement(pH, pH->ht ? &pH->ht[h] : 0, new_elem); + return 0; +} + +/************** End of hash.c ************************************************/ +/************** Begin file opcodes.c *****************************************/ +/* Automatically generated. Do not edit */ +/* See the tool/mkopcodec.tcl script for details. */ +#if !defined(SQLITE_OMIT_EXPLAIN) \ + || defined(VDBE_PROFILE) \ + || defined(SQLITE_DEBUG) +#if defined(SQLITE_ENABLE_EXPLAIN_COMMENTS) || defined(SQLITE_DEBUG) +# define OpHelp(X) "\0" X +#else +# define OpHelp(X) +#endif +SQLITE_PRIVATE const char *sqlite3OpcodeName(int i){ + static const char *const azName[] = { + /* 0 */ "Savepoint" OpHelp(""), + /* 1 */ "AutoCommit" OpHelp(""), + /* 2 */ "Transaction" OpHelp(""), + /* 3 */ "SorterNext" OpHelp(""), + /* 4 */ "Prev" OpHelp(""), + /* 5 */ "Next" OpHelp(""), + /* 6 */ "Checkpoint" OpHelp(""), + /* 7 */ "JournalMode" OpHelp(""), + /* 8 */ "Vacuum" OpHelp(""), + /* 9 */ "VFilter" OpHelp("iplan=r[P3] zplan='P4'"), + /* 10 */ "VUpdate" OpHelp("data=r[P3@P2]"), + /* 11 */ "Goto" OpHelp(""), + /* 12 */ "Gosub" OpHelp(""), + /* 13 */ "InitCoroutine" OpHelp(""), + /* 14 */ "Yield" OpHelp(""), + /* 15 */ "MustBeInt" OpHelp(""), + /* 16 */ "Jump" OpHelp(""), + /* 17 */ "Once" OpHelp(""), + /* 18 */ "If" OpHelp(""), + /* 19 */ "Not" OpHelp("r[P2]= !r[P1]"), + /* 20 */ "IfNot" OpHelp(""), + /* 21 */ "IfNullRow" OpHelp("if P1.nullRow then r[P3]=NULL, goto P2"), + /* 22 */ "SeekLT" OpHelp("key=r[P3@P4]"), + /* 23 */ "SeekLE" OpHelp("key=r[P3@P4]"), + /* 24 */ "SeekGE" OpHelp("key=r[P3@P4]"), + /* 25 */ "SeekGT" OpHelp("key=r[P3@P4]"), + /* 26 */ "IfNoHope" OpHelp("key=r[P3@P4]"), + /* 27 */ "NoConflict" OpHelp("key=r[P3@P4]"), + /* 28 */ "NotFound" OpHelp("key=r[P3@P4]"), + /* 29 */ "Found" OpHelp("key=r[P3@P4]"), + /* 30 */ "SeekRowid" OpHelp("intkey=r[P3]"), + /* 31 */ "NotExists" OpHelp("intkey=r[P3]"), + /* 32 */ "Last" OpHelp(""), + /* 33 */ "IfSmaller" OpHelp(""), + /* 34 */ "SorterSort" OpHelp(""), + /* 35 */ "Sort" OpHelp(""), + /* 36 */ "Rewind" OpHelp(""), + /* 37 */ "IdxLE" OpHelp("key=r[P3@P4]"), + /* 38 */ "IdxGT" OpHelp("key=r[P3@P4]"), + /* 39 */ "IdxLT" OpHelp("key=r[P3@P4]"), + /* 40 */ "IdxGE" OpHelp("key=r[P3@P4]"), + /* 41 */ "RowSetRead" OpHelp("r[P3]=rowset(P1)"), + /* 42 */ "RowSetTest" OpHelp("if r[P3] in rowset(P1) goto P2"), + /* 43 */ "Or" OpHelp("r[P3]=(r[P1] || r[P2])"), + /* 44 */ "And" OpHelp("r[P3]=(r[P1] && r[P2])"), + /* 45 */ "Program" OpHelp(""), + /* 46 */ "FkIfZero" OpHelp("if fkctr[P1]==0 goto P2"), + /* 47 */ "IfPos" OpHelp("if r[P1]>0 then r[P1]-=P3, goto P2"), + /* 48 */ "IfNotZero" OpHelp("if r[P1]!=0 then r[P1]--, goto P2"), + /* 49 */ "DecrJumpZero" OpHelp("if (--r[P1])==0 goto P2"), + /* 50 */ "IsNull" OpHelp("if r[P1]==NULL goto P2"), + /* 51 */ "NotNull" OpHelp("if r[P1]!=NULL goto P2"), + /* 52 */ "Ne" OpHelp("IF r[P3]!=r[P1]"), + /* 53 */ "Eq" OpHelp("IF r[P3]==r[P1]"), + /* 54 */ "Gt" OpHelp("IF r[P3]>r[P1]"), + /* 55 */ "Le" OpHelp("IF r[P3]<=r[P1]"), + /* 56 */ "Lt" OpHelp("IF r[P3]=r[P1]"), + /* 58 */ "ElseNotEq" OpHelp(""), + /* 59 */ "IncrVacuum" OpHelp(""), + /* 60 */ "VNext" OpHelp(""), + /* 61 */ "Init" OpHelp("Start at P2"), + /* 62 */ "PureFunc0" OpHelp(""), + /* 63 */ "Function0" OpHelp("r[P3]=func(r[P2@P5])"), + /* 64 */ "PureFunc" OpHelp(""), + /* 65 */ "Function" OpHelp("r[P3]=func(r[P2@P5])"), + /* 66 */ "Return" OpHelp(""), + /* 67 */ "EndCoroutine" OpHelp(""), + /* 68 */ "HaltIfNull" OpHelp("if r[P3]=null halt"), + /* 69 */ "Halt" OpHelp(""), + /* 70 */ "Integer" OpHelp("r[P2]=P1"), + /* 71 */ "Int64" OpHelp("r[P2]=P4"), + /* 72 */ "String" OpHelp("r[P2]='P4' (len=P1)"), + /* 73 */ "Null" OpHelp("r[P2..P3]=NULL"), + /* 74 */ "SoftNull" OpHelp("r[P1]=NULL"), + /* 75 */ "Blob" OpHelp("r[P2]=P4 (len=P1)"), + /* 76 */ "Variable" OpHelp("r[P2]=parameter(P1,P4)"), + /* 77 */ "Move" OpHelp("r[P2@P3]=r[P1@P3]"), + /* 78 */ "Copy" OpHelp("r[P2@P3+1]=r[P1@P3+1]"), + /* 79 */ "SCopy" OpHelp("r[P2]=r[P1]"), + /* 80 */ "IntCopy" OpHelp("r[P2]=r[P1]"), + /* 81 */ "ResultRow" OpHelp("output=r[P1@P2]"), + /* 82 */ "CollSeq" OpHelp(""), + /* 83 */ "AddImm" OpHelp("r[P1]=r[P1]+P2"), + /* 84 */ "RealAffinity" OpHelp(""), + /* 85 */ "Cast" OpHelp("affinity(r[P1])"), + /* 86 */ "Permutation" OpHelp(""), + /* 87 */ "Compare" OpHelp("r[P1@P3] <-> r[P2@P3]"), + /* 88 */ "IsTrue" OpHelp("r[P2] = coalesce(r[P1]==TRUE,P3) ^ P4"), + /* 89 */ "Offset" OpHelp("r[P3] = sqlite_offset(P1)"), + /* 90 */ "Column" OpHelp("r[P3]=PX"), + /* 91 */ "Affinity" OpHelp("affinity(r[P1@P2])"), + /* 92 */ "MakeRecord" OpHelp("r[P3]=mkrec(r[P1@P2])"), + /* 93 */ "Count" OpHelp("r[P2]=count()"), + /* 94 */ "ReadCookie" OpHelp(""), + /* 95 */ "SetCookie" OpHelp(""), + /* 96 */ "ReopenIdx" OpHelp("root=P2 iDb=P3"), + /* 97 */ "OpenRead" OpHelp("root=P2 iDb=P3"), + /* 98 */ "OpenWrite" OpHelp("root=P2 iDb=P3"), + /* 99 */ "BitAnd" OpHelp("r[P3]=r[P1]&r[P2]"), + /* 100 */ "BitOr" OpHelp("r[P3]=r[P1]|r[P2]"), + /* 101 */ "ShiftLeft" OpHelp("r[P3]=r[P2]<>r[P1]"), + /* 103 */ "Add" OpHelp("r[P3]=r[P1]+r[P2]"), + /* 104 */ "Subtract" OpHelp("r[P3]=r[P2]-r[P1]"), + /* 105 */ "Multiply" OpHelp("r[P3]=r[P1]*r[P2]"), + /* 106 */ "Divide" OpHelp("r[P3]=r[P2]/r[P1]"), + /* 107 */ "Remainder" OpHelp("r[P3]=r[P2]%r[P1]"), + /* 108 */ "Concat" OpHelp("r[P3]=r[P2]+r[P1]"), + /* 109 */ "OpenDup" OpHelp(""), + /* 110 */ "BitNot" OpHelp("r[P2]= ~r[P1]"), + /* 111 */ "OpenAutoindex" OpHelp("nColumn=P2"), + /* 112 */ "OpenEphemeral" OpHelp("nColumn=P2"), + /* 113 */ "String8" OpHelp("r[P2]='P4'"), + /* 114 */ "SorterOpen" OpHelp(""), + /* 115 */ "SequenceTest" OpHelp("if( cursor[P1].ctr++ ) pc = P2"), + /* 116 */ "OpenPseudo" OpHelp("P3 columns in r[P2]"), + /* 117 */ "Close" OpHelp(""), + /* 118 */ "ColumnsUsed" OpHelp(""), + /* 119 */ "SeekHit" OpHelp("seekHit=P2"), + /* 120 */ "Sequence" OpHelp("r[P2]=cursor[P1].ctr++"), + /* 121 */ "NewRowid" OpHelp("r[P2]=rowid"), + /* 122 */ "Insert" OpHelp("intkey=r[P3] data=r[P2]"), + /* 123 */ "Delete" OpHelp(""), + /* 124 */ "ResetCount" OpHelp(""), + /* 125 */ "SorterCompare" OpHelp("if key(P1)!=trim(r[P3],P4) goto P2"), + /* 126 */ "SorterData" OpHelp("r[P2]=data"), + /* 127 */ "RowData" OpHelp("r[P2]=data"), + /* 128 */ "Rowid" OpHelp("r[P2]=rowid"), + /* 129 */ "NullRow" OpHelp(""), + /* 130 */ "SeekEnd" OpHelp(""), + /* 131 */ "SorterInsert" OpHelp("key=r[P2]"), + /* 132 */ "IdxInsert" OpHelp("key=r[P2]"), + /* 133 */ "IdxDelete" OpHelp("key=r[P2@P3]"), + /* 134 */ "DeferredSeek" OpHelp("Move P3 to P1.rowid if needed"), + /* 135 */ "IdxRowid" OpHelp("r[P2]=rowid"), + /* 136 */ "Destroy" OpHelp(""), + /* 137 */ "Clear" OpHelp(""), + /* 138 */ "ResetSorter" OpHelp(""), + /* 139 */ "CreateBtree" OpHelp("r[P2]=root iDb=P1 flags=P3"), + /* 140 */ "SqlExec" OpHelp(""), + /* 141 */ "ParseSchema" OpHelp(""), + /* 142 */ "LoadAnalysis" OpHelp(""), + /* 143 */ "DropTable" OpHelp(""), + /* 144 */ "DropIndex" OpHelp(""), + /* 145 */ "DropTrigger" OpHelp(""), + /* 146 */ "IntegrityCk" OpHelp(""), + /* 147 */ "RowSetAdd" OpHelp("rowset(P1)=r[P2]"), + /* 148 */ "Real" OpHelp("r[P2]=P4"), + /* 149 */ "Param" OpHelp(""), + /* 150 */ "FkCounter" OpHelp("fkctr[P1]+=P2"), + /* 151 */ "MemMax" OpHelp("r[P1]=max(r[P1],r[P2])"), + /* 152 */ "OffsetLimit" OpHelp("if r[P1]>0 then r[P2]=r[P1]+max(0,r[P3]) else r[P2]=(-1)"), + /* 153 */ "AggInverse" OpHelp("accum=r[P3] inverse(r[P2@P5])"), + /* 154 */ "AggStep" OpHelp("accum=r[P3] step(r[P2@P5])"), + /* 155 */ "AggStep1" OpHelp("accum=r[P3] step(r[P2@P5])"), + /* 156 */ "AggValue" OpHelp("r[P3]=value N=P2"), + /* 157 */ "AggFinal" OpHelp("accum=r[P1] N=P2"), + /* 158 */ "Expire" OpHelp(""), + /* 159 */ "TableLock" OpHelp("iDb=P1 root=P2 write=P3"), + /* 160 */ "VBegin" OpHelp(""), + /* 161 */ "VCreate" OpHelp(""), + /* 162 */ "VDestroy" OpHelp(""), + /* 163 */ "VOpen" OpHelp(""), + /* 164 */ "VColumn" OpHelp("r[P3]=vcolumn(P2)"), + /* 165 */ "VRename" OpHelp(""), + /* 166 */ "Pagecount" OpHelp(""), + /* 167 */ "MaxPgcnt" OpHelp(""), + /* 168 */ "Trace" OpHelp(""), + /* 169 */ "CursorHint" OpHelp(""), + /* 170 */ "Noop" OpHelp(""), + /* 171 */ "Explain" OpHelp(""), + /* 172 */ "Abortable" OpHelp(""), + }; + return azName[i]; +} +#endif + +/************** End of opcodes.c *********************************************/ +/************** Begin file os_unix.c *****************************************/ +/* +** 2004 May 22 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +****************************************************************************** +** +** This file contains the VFS implementation for unix-like operating systems +** include Linux, MacOSX, *BSD, QNX, VxWorks, AIX, HPUX, and others. +** +** There are actually several different VFS implementations in this file. +** The differences are in the way that file locking is done. The default +** implementation uses Posix Advisory Locks. Alternative implementations +** use flock(), dot-files, various proprietary locking schemas, or simply +** skip locking all together. +** +** This source file is organized into divisions where the logic for various +** subfunctions is contained within the appropriate division. PLEASE +** KEEP THE STRUCTURE OF THIS FILE INTACT. New code should be placed +** in the correct division and should be clearly labeled. +** +** The layout of divisions is as follows: +** +** * General-purpose declarations and utility functions. +** * Unique file ID logic used by VxWorks. +** * Various locking primitive implementations (all except proxy locking): +** + for Posix Advisory Locks +** + for no-op locks +** + for dot-file locks +** + for flock() locking +** + for named semaphore locks (VxWorks only) +** + for AFP filesystem locks (MacOSX only) +** * sqlite3_file methods not associated with locking. +** * Definitions of sqlite3_io_methods objects for all locking +** methods plus "finder" functions for each locking method. +** * sqlite3_vfs method implementations. +** * Locking primitives for the proxy uber-locking-method. (MacOSX only) +** * Definitions of sqlite3_vfs objects for all locking methods +** plus implementations of sqlite3_os_init() and sqlite3_os_end(). +*/ +/* #include "sqliteInt.h" */ +#if SQLITE_OS_UNIX /* This file is used on unix only */ + +/* +** There are various methods for file locking used for concurrency +** control: +** +** 1. POSIX locking (the default), +** 2. No locking, +** 3. Dot-file locking, +** 4. flock() locking, +** 5. AFP locking (OSX only), +** 6. Named POSIX semaphores (VXWorks only), +** 7. proxy locking. (OSX only) +** +** Styles 4, 5, and 7 are only available of SQLITE_ENABLE_LOCKING_STYLE +** is defined to 1. The SQLITE_ENABLE_LOCKING_STYLE also enables automatic +** selection of the appropriate locking style based on the filesystem +** where the database is located. +*/ +#if !defined(SQLITE_ENABLE_LOCKING_STYLE) +# if defined(__APPLE__) +# define SQLITE_ENABLE_LOCKING_STYLE 1 +# else +# define SQLITE_ENABLE_LOCKING_STYLE 0 +# endif +#endif + +/* Use pread() and pwrite() if they are available */ +#if defined(__APPLE__) +# define HAVE_PREAD 1 +# define HAVE_PWRITE 1 +#endif +#if defined(HAVE_PREAD64) && defined(HAVE_PWRITE64) +# undef USE_PREAD +# define USE_PREAD64 1 +#elif defined(HAVE_PREAD) && defined(HAVE_PWRITE) +# undef USE_PREAD64 +# define USE_PREAD 1 +#endif + +/* +** standard include files. +*/ +#include +#include +#include +#include +#include +/* #include */ +#include +#include +#if !defined(SQLITE_OMIT_WAL) || SQLITE_MAX_MMAP_SIZE>0 +# include +#endif + +#if SQLITE_ENABLE_LOCKING_STYLE +/* # include */ +# include +# include +#endif /* SQLITE_ENABLE_LOCKING_STYLE */ + +/* +** Try to determine if gethostuuid() is available based on standard +** macros. This might sometimes compute the wrong value for some +** obscure platforms. For those cases, simply compile with one of +** the following: +** +** -DHAVE_GETHOSTUUID=0 +** -DHAVE_GETHOSTUUID=1 +** +** None if this matters except when building on Apple products with +** -DSQLITE_ENABLE_LOCKING_STYLE. +*/ +#ifndef HAVE_GETHOSTUUID +# define HAVE_GETHOSTUUID 0 +# if defined(__APPLE__) && ((__MAC_OS_X_VERSION_MIN_REQUIRED > 1050) || \ + (__IPHONE_OS_VERSION_MIN_REQUIRED > 2000)) +# if (!defined(TARGET_OS_EMBEDDED) || (TARGET_OS_EMBEDDED==0)) \ + && (!defined(TARGET_IPHONE_SIMULATOR) || (TARGET_IPHONE_SIMULATOR==0)) +# undef HAVE_GETHOSTUUID +# define HAVE_GETHOSTUUID 1 +# else +# warning "gethostuuid() is disabled." +# endif +# endif +#endif + + +#if OS_VXWORKS +/* # include */ +# include +# include +#endif /* OS_VXWORKS */ + +#if defined(__APPLE__) || SQLITE_ENABLE_LOCKING_STYLE +# include +#endif + +#ifdef HAVE_UTIME +# include +#endif + +/* +** Allowed values of unixFile.fsFlags +*/ +#define SQLITE_FSFLAGS_IS_MSDOS 0x1 + +/* +** If we are to be thread-safe, include the pthreads header. +*/ +#if SQLITE_THREADSAFE +/* # include */ +#endif + +/* +** Default permissions when creating a new file +*/ +#ifndef SQLITE_DEFAULT_FILE_PERMISSIONS +# define SQLITE_DEFAULT_FILE_PERMISSIONS 0644 +#endif + +/* +** Default permissions when creating auto proxy dir +*/ +#ifndef SQLITE_DEFAULT_PROXYDIR_PERMISSIONS +# define SQLITE_DEFAULT_PROXYDIR_PERMISSIONS 0755 +#endif + +/* +** Maximum supported path-length. +*/ +#define MAX_PATHNAME 512 + +/* +** Maximum supported symbolic links +*/ +#define SQLITE_MAX_SYMLINKS 100 + +/* Always cast the getpid() return type for compatibility with +** kernel modules in VxWorks. */ +#define osGetpid(X) (pid_t)getpid() + +/* +** Only set the lastErrno if the error code is a real error and not +** a normal expected return code of SQLITE_BUSY or SQLITE_OK +*/ +#define IS_LOCK_ERROR(x) ((x != SQLITE_OK) && (x != SQLITE_BUSY)) + +/* Forward references */ +typedef struct unixShm unixShm; /* Connection shared memory */ +typedef struct unixShmNode unixShmNode; /* Shared memory instance */ +typedef struct unixInodeInfo unixInodeInfo; /* An i-node */ +typedef struct UnixUnusedFd UnixUnusedFd; /* An unused file descriptor */ + +/* +** Sometimes, after a file handle is closed by SQLite, the file descriptor +** cannot be closed immediately. In these cases, instances of the following +** structure are used to store the file descriptor while waiting for an +** opportunity to either close or reuse it. +*/ +struct UnixUnusedFd { + int fd; /* File descriptor to close */ + int flags; /* Flags this file descriptor was opened with */ + UnixUnusedFd *pNext; /* Next unused file descriptor on same file */ +}; + +/* +** The unixFile structure is subclass of sqlite3_file specific to the unix +** VFS implementations. +*/ +typedef struct unixFile unixFile; +struct unixFile { + sqlite3_io_methods const *pMethod; /* Always the first entry */ + sqlite3_vfs *pVfs; /* The VFS that created this unixFile */ + unixInodeInfo *pInode; /* Info about locks on this inode */ + int h; /* The file descriptor */ + unsigned char eFileLock; /* The type of lock held on this fd */ + unsigned short int ctrlFlags; /* Behavioral bits. UNIXFILE_* flags */ + int lastErrno; /* The unix errno from last I/O error */ + void *lockingContext; /* Locking style specific state */ + UnixUnusedFd *pPreallocatedUnused; /* Pre-allocated UnixUnusedFd */ + const char *zPath; /* Name of the file */ + unixShm *pShm; /* Shared memory segment information */ + int szChunk; /* Configured by FCNTL_CHUNK_SIZE */ +#if SQLITE_MAX_MMAP_SIZE>0 + int nFetchOut; /* Number of outstanding xFetch refs */ + sqlite3_int64 mmapSize; /* Usable size of mapping at pMapRegion */ + sqlite3_int64 mmapSizeActual; /* Actual size of mapping at pMapRegion */ + sqlite3_int64 mmapSizeMax; /* Configured FCNTL_MMAP_SIZE value */ + void *pMapRegion; /* Memory mapped region */ +#endif + int sectorSize; /* Device sector size */ + int deviceCharacteristics; /* Precomputed device characteristics */ +#if SQLITE_ENABLE_LOCKING_STYLE + int openFlags; /* The flags specified at open() */ +#endif +#if SQLITE_ENABLE_LOCKING_STYLE || defined(__APPLE__) + unsigned fsFlags; /* cached details from statfs() */ +#endif +#ifdef SQLITE_ENABLE_SETLK_TIMEOUT + unsigned iBusyTimeout; /* Wait this many millisec on locks */ +#endif +#if OS_VXWORKS + struct vxworksFileId *pId; /* Unique file ID */ +#endif +#ifdef SQLITE_DEBUG + /* The next group of variables are used to track whether or not the + ** transaction counter in bytes 24-27 of database files are updated + ** whenever any part of the database changes. An assertion fault will + ** occur if a file is updated without also updating the transaction + ** counter. This test is made to avoid new problems similar to the + ** one described by ticket #3584. + */ + unsigned char transCntrChng; /* True if the transaction counter changed */ + unsigned char dbUpdate; /* True if any part of database file changed */ + unsigned char inNormalWrite; /* True if in a normal write operation */ + +#endif + +#ifdef SQLITE_TEST + /* In test mode, increase the size of this structure a bit so that + ** it is larger than the struct CrashFile defined in test6.c. + */ + char aPadding[32]; +#endif +}; + +/* This variable holds the process id (pid) from when the xRandomness() +** method was called. If xOpen() is called from a different process id, +** indicating that a fork() has occurred, the PRNG will be reset. +*/ +static pid_t randomnessPid = 0; + +/* +** Allowed values for the unixFile.ctrlFlags bitmask: +*/ +#define UNIXFILE_EXCL 0x01 /* Connections from one process only */ +#define UNIXFILE_RDONLY 0x02 /* Connection is read only */ +#define UNIXFILE_PERSIST_WAL 0x04 /* Persistent WAL mode */ +#ifndef SQLITE_DISABLE_DIRSYNC +# define UNIXFILE_DIRSYNC 0x08 /* Directory sync needed */ +#else +# define UNIXFILE_DIRSYNC 0x00 +#endif +#define UNIXFILE_PSOW 0x10 /* SQLITE_IOCAP_POWERSAFE_OVERWRITE */ +#define UNIXFILE_DELETE 0x20 /* Delete on close */ +#define UNIXFILE_URI 0x40 /* Filename might have query parameters */ +#define UNIXFILE_NOLOCK 0x80 /* Do no file locking */ + +/* +** Include code that is common to all os_*.c files +*/ +/************** Include os_common.h in the middle of os_unix.c ***************/ +/************** Begin file os_common.h ***************************************/ +/* +** 2004 May 22 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +****************************************************************************** +** +** This file contains macros and a little bit of code that is common to +** all of the platform-specific files (os_*.c) and is #included into those +** files. +** +** This file should be #included by the os_*.c files only. It is not a +** general purpose header file. +*/ +#ifndef _OS_COMMON_H_ +#define _OS_COMMON_H_ + +/* +** At least two bugs have slipped in because we changed the MEMORY_DEBUG +** macro to SQLITE_DEBUG and some older makefiles have not yet made the +** switch. The following code should catch this problem at compile-time. +*/ +#ifdef MEMORY_DEBUG +# error "The MEMORY_DEBUG macro is obsolete. Use SQLITE_DEBUG instead." +#endif + +/* +** Macros for performance tracing. Normally turned off. Only works +** on i486 hardware. +*/ +#ifdef SQLITE_PERFORMANCE_TRACE + +/* +** hwtime.h contains inline assembler code for implementing +** high-performance timing routines. +*/ +/************** Include hwtime.h in the middle of os_common.h ****************/ +/************** Begin file hwtime.h ******************************************/ +/* +** 2008 May 27 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +****************************************************************************** +** +** This file contains inline asm code for retrieving "high-performance" +** counters for x86 class CPUs. +*/ +#ifndef SQLITE_HWTIME_H +#define SQLITE_HWTIME_H + +/* +** The following routine only works on pentium-class (or newer) processors. +** It uses the RDTSC opcode to read the cycle count value out of the +** processor and returns that value. This can be used for high-res +** profiling. +*/ +#if (defined(__GNUC__) || defined(_MSC_VER)) && \ + (defined(i386) || defined(__i386__) || defined(_M_IX86)) + + #if defined(__GNUC__) + + __inline__ sqlite_uint64 sqlite3Hwtime(void){ + unsigned int lo, hi; + __asm__ __volatile__ ("rdtsc" : "=a" (lo), "=d" (hi)); + return (sqlite_uint64)hi << 32 | lo; + } + + #elif defined(_MSC_VER) + + __declspec(naked) __inline sqlite_uint64 __cdecl sqlite3Hwtime(void){ + __asm { + rdtsc + ret ; return value at EDX:EAX + } + } + + #endif + +#elif (defined(__GNUC__) && defined(__x86_64__)) + + __inline__ sqlite_uint64 sqlite3Hwtime(void){ + unsigned long val; + __asm__ __volatile__ ("rdtsc" : "=A" (val)); + return val; + } + +#elif (defined(__GNUC__) && defined(__ppc__)) + + __inline__ sqlite_uint64 sqlite3Hwtime(void){ + unsigned long long retval; + unsigned long junk; + __asm__ __volatile__ ("\n\ + 1: mftbu %1\n\ + mftb %L0\n\ + mftbu %0\n\ + cmpw %0,%1\n\ + bne 1b" + : "=r" (retval), "=r" (junk)); + return retval; + } + +#else + + #error Need implementation of sqlite3Hwtime() for your platform. + + /* + ** To compile without implementing sqlite3Hwtime() for your platform, + ** you can remove the above #error and use the following + ** stub function. You will lose timing support for many + ** of the debugging and testing utilities, but it should at + ** least compile and run. + */ +SQLITE_PRIVATE sqlite_uint64 sqlite3Hwtime(void){ return ((sqlite_uint64)0); } + +#endif + +#endif /* !defined(SQLITE_HWTIME_H) */ + +/************** End of hwtime.h **********************************************/ +/************** Continuing where we left off in os_common.h ******************/ + +static sqlite_uint64 g_start; +static sqlite_uint64 g_elapsed; +#define TIMER_START g_start=sqlite3Hwtime() +#define TIMER_END g_elapsed=sqlite3Hwtime()-g_start +#define TIMER_ELAPSED g_elapsed +#else +#define TIMER_START +#define TIMER_END +#define TIMER_ELAPSED ((sqlite_uint64)0) +#endif + +/* +** If we compile with the SQLITE_TEST macro set, then the following block +** of code will give us the ability to simulate a disk I/O error. This +** is used for testing the I/O recovery logic. +*/ +#if defined(SQLITE_TEST) +SQLITE_API extern int sqlite3_io_error_hit; +SQLITE_API extern int sqlite3_io_error_hardhit; +SQLITE_API extern int sqlite3_io_error_pending; +SQLITE_API extern int sqlite3_io_error_persist; +SQLITE_API extern int sqlite3_io_error_benign; +SQLITE_API extern int sqlite3_diskfull_pending; +SQLITE_API extern int sqlite3_diskfull; +#define SimulateIOErrorBenign(X) sqlite3_io_error_benign=(X) +#define SimulateIOError(CODE) \ + if( (sqlite3_io_error_persist && sqlite3_io_error_hit) \ + || sqlite3_io_error_pending-- == 1 ) \ + { local_ioerr(); CODE; } +static void local_ioerr(){ + IOTRACE(("IOERR\n")); + sqlite3_io_error_hit++; + if( !sqlite3_io_error_benign ) sqlite3_io_error_hardhit++; +} +#define SimulateDiskfullError(CODE) \ + if( sqlite3_diskfull_pending ){ \ + if( sqlite3_diskfull_pending == 1 ){ \ + local_ioerr(); \ + sqlite3_diskfull = 1; \ + sqlite3_io_error_hit = 1; \ + CODE; \ + }else{ \ + sqlite3_diskfull_pending--; \ + } \ + } +#else +#define SimulateIOErrorBenign(X) +#define SimulateIOError(A) +#define SimulateDiskfullError(A) +#endif /* defined(SQLITE_TEST) */ + +/* +** When testing, keep a count of the number of open files. +*/ +#if defined(SQLITE_TEST) +SQLITE_API extern int sqlite3_open_file_count; +#define OpenCounter(X) sqlite3_open_file_count+=(X) +#else +#define OpenCounter(X) +#endif /* defined(SQLITE_TEST) */ + +#endif /* !defined(_OS_COMMON_H_) */ + +/************** End of os_common.h *******************************************/ +/************** Continuing where we left off in os_unix.c ********************/ + +/* +** Define various macros that are missing from some systems. +*/ +#ifndef O_LARGEFILE +# define O_LARGEFILE 0 +#endif +#ifdef SQLITE_DISABLE_LFS +# undef O_LARGEFILE +# define O_LARGEFILE 0 +#endif +#ifndef O_NOFOLLOW +# define O_NOFOLLOW 0 +#endif +#ifndef O_BINARY +# define O_BINARY 0 +#endif + +/* +** The threadid macro resolves to the thread-id or to 0. Used for +** testing and debugging only. +*/ +#if SQLITE_THREADSAFE +#define threadid pthread_self() +#else +#define threadid 0 +#endif + +/* +** HAVE_MREMAP defaults to true on Linux and false everywhere else. +*/ +#if !defined(HAVE_MREMAP) +# if defined(__linux__) && defined(_GNU_SOURCE) +# define HAVE_MREMAP 1 +# else +# define HAVE_MREMAP 0 +# endif +#endif + +/* +** Explicitly call the 64-bit version of lseek() on Android. Otherwise, lseek() +** is the 32-bit version, even if _FILE_OFFSET_BITS=64 is defined. +*/ +#ifdef __ANDROID__ +# define lseek lseek64 +#endif + +#ifdef __linux__ +/* +** Linux-specific IOCTL magic numbers used for controlling F2FS +*/ +#define F2FS_IOCTL_MAGIC 0xf5 +#define F2FS_IOC_START_ATOMIC_WRITE _IO(F2FS_IOCTL_MAGIC, 1) +#define F2FS_IOC_COMMIT_ATOMIC_WRITE _IO(F2FS_IOCTL_MAGIC, 2) +#define F2FS_IOC_START_VOLATILE_WRITE _IO(F2FS_IOCTL_MAGIC, 3) +#define F2FS_IOC_ABORT_VOLATILE_WRITE _IO(F2FS_IOCTL_MAGIC, 5) +#define F2FS_IOC_GET_FEATURES _IOR(F2FS_IOCTL_MAGIC, 12, u32) +#define F2FS_FEATURE_ATOMIC_WRITE 0x0004 +#endif /* __linux__ */ + + +/* +** Different Unix systems declare open() in different ways. Same use +** open(const char*,int,mode_t). Others use open(const char*,int,...). +** The difference is important when using a pointer to the function. +** +** The safest way to deal with the problem is to always use this wrapper +** which always has the same well-defined interface. +*/ +static int posixOpen(const char *zFile, int flags, int mode){ + return open(zFile, flags, mode); +} + +/* Forward reference */ +static int openDirectory(const char*, int*); +static int unixGetpagesize(void); + +/* +** Many system calls are accessed through pointer-to-functions so that +** they may be overridden at runtime to facilitate fault injection during +** testing and sandboxing. The following array holds the names and pointers +** to all overrideable system calls. +*/ +static struct unix_syscall { + const char *zName; /* Name of the system call */ + sqlite3_syscall_ptr pCurrent; /* Current value of the system call */ + sqlite3_syscall_ptr pDefault; /* Default value */ +} aSyscall[] = { + { "open", (sqlite3_syscall_ptr)posixOpen, 0 }, +#define osOpen ((int(*)(const char*,int,int))aSyscall[0].pCurrent) + + { "close", (sqlite3_syscall_ptr)close, 0 }, +#define osClose ((int(*)(int))aSyscall[1].pCurrent) + + { "access", (sqlite3_syscall_ptr)access, 0 }, +#define osAccess ((int(*)(const char*,int))aSyscall[2].pCurrent) + + { "getcwd", (sqlite3_syscall_ptr)getcwd, 0 }, +#define osGetcwd ((char*(*)(char*,size_t))aSyscall[3].pCurrent) + + { "stat", (sqlite3_syscall_ptr)stat, 0 }, +#define osStat ((int(*)(const char*,struct stat*))aSyscall[4].pCurrent) + +/* +** The DJGPP compiler environment looks mostly like Unix, but it +** lacks the fcntl() system call. So redefine fcntl() to be something +** that always succeeds. This means that locking does not occur under +** DJGPP. But it is DOS - what did you expect? +*/ +#ifdef __DJGPP__ + { "fstat", 0, 0 }, +#define osFstat(a,b,c) 0 +#else + { "fstat", (sqlite3_syscall_ptr)fstat, 0 }, +#define osFstat ((int(*)(int,struct stat*))aSyscall[5].pCurrent) +#endif + + { "ftruncate", (sqlite3_syscall_ptr)ftruncate, 0 }, +#define osFtruncate ((int(*)(int,off_t))aSyscall[6].pCurrent) + + { "fcntl", (sqlite3_syscall_ptr)fcntl, 0 }, +#define osFcntl ((int(*)(int,int,...))aSyscall[7].pCurrent) + + { "read", (sqlite3_syscall_ptr)read, 0 }, +#define osRead ((ssize_t(*)(int,void*,size_t))aSyscall[8].pCurrent) + +#if defined(USE_PREAD) || SQLITE_ENABLE_LOCKING_STYLE + { "pread", (sqlite3_syscall_ptr)pread, 0 }, +#else + { "pread", (sqlite3_syscall_ptr)0, 0 }, +#endif +#define osPread ((ssize_t(*)(int,void*,size_t,off_t))aSyscall[9].pCurrent) + +#if defined(USE_PREAD64) + { "pread64", (sqlite3_syscall_ptr)pread64, 0 }, +#else + { "pread64", (sqlite3_syscall_ptr)0, 0 }, +#endif +#define osPread64 ((ssize_t(*)(int,void*,size_t,off64_t))aSyscall[10].pCurrent) + + { "write", (sqlite3_syscall_ptr)write, 0 }, +#define osWrite ((ssize_t(*)(int,const void*,size_t))aSyscall[11].pCurrent) + +#if defined(USE_PREAD) || SQLITE_ENABLE_LOCKING_STYLE + { "pwrite", (sqlite3_syscall_ptr)pwrite, 0 }, +#else + { "pwrite", (sqlite3_syscall_ptr)0, 0 }, +#endif +#define osPwrite ((ssize_t(*)(int,const void*,size_t,off_t))\ + aSyscall[12].pCurrent) + +#if defined(USE_PREAD64) + { "pwrite64", (sqlite3_syscall_ptr)pwrite64, 0 }, +#else + { "pwrite64", (sqlite3_syscall_ptr)0, 0 }, +#endif +#define osPwrite64 ((ssize_t(*)(int,const void*,size_t,off64_t))\ + aSyscall[13].pCurrent) + + { "fchmod", (sqlite3_syscall_ptr)fchmod, 0 }, +#define osFchmod ((int(*)(int,mode_t))aSyscall[14].pCurrent) + +#if defined(HAVE_POSIX_FALLOCATE) && HAVE_POSIX_FALLOCATE + { "fallocate", (sqlite3_syscall_ptr)posix_fallocate, 0 }, +#else + { "fallocate", (sqlite3_syscall_ptr)0, 0 }, +#endif +#define osFallocate ((int(*)(int,off_t,off_t))aSyscall[15].pCurrent) + + { "unlink", (sqlite3_syscall_ptr)unlink, 0 }, +#define osUnlink ((int(*)(const char*))aSyscall[16].pCurrent) + + { "openDirectory", (sqlite3_syscall_ptr)openDirectory, 0 }, +#define osOpenDirectory ((int(*)(const char*,int*))aSyscall[17].pCurrent) + + { "mkdir", (sqlite3_syscall_ptr)mkdir, 0 }, +#define osMkdir ((int(*)(const char*,mode_t))aSyscall[18].pCurrent) + + { "rmdir", (sqlite3_syscall_ptr)rmdir, 0 }, +#define osRmdir ((int(*)(const char*))aSyscall[19].pCurrent) + +#if defined(HAVE_FCHOWN) + { "fchown", (sqlite3_syscall_ptr)fchown, 0 }, +#else + { "fchown", (sqlite3_syscall_ptr)0, 0 }, +#endif +#define osFchown ((int(*)(int,uid_t,gid_t))aSyscall[20].pCurrent) + +#if defined(HAVE_FCHOWN) + { "geteuid", (sqlite3_syscall_ptr)geteuid, 0 }, +#else + { "geteuid", (sqlite3_syscall_ptr)0, 0 }, +#endif +#define osGeteuid ((uid_t(*)(void))aSyscall[21].pCurrent) + +#if !defined(SQLITE_OMIT_WAL) || SQLITE_MAX_MMAP_SIZE>0 + { "mmap", (sqlite3_syscall_ptr)mmap, 0 }, +#else + { "mmap", (sqlite3_syscall_ptr)0, 0 }, +#endif +#define osMmap ((void*(*)(void*,size_t,int,int,int,off_t))aSyscall[22].pCurrent) + +#if !defined(SQLITE_OMIT_WAL) || SQLITE_MAX_MMAP_SIZE>0 + { "munmap", (sqlite3_syscall_ptr)munmap, 0 }, +#else + { "munmap", (sqlite3_syscall_ptr)0, 0 }, +#endif +#define osMunmap ((int(*)(void*,size_t))aSyscall[23].pCurrent) + +#if HAVE_MREMAP && (!defined(SQLITE_OMIT_WAL) || SQLITE_MAX_MMAP_SIZE>0) + { "mremap", (sqlite3_syscall_ptr)mremap, 0 }, +#else + { "mremap", (sqlite3_syscall_ptr)0, 0 }, +#endif +#define osMremap ((void*(*)(void*,size_t,size_t,int,...))aSyscall[24].pCurrent) + +#if !defined(SQLITE_OMIT_WAL) || SQLITE_MAX_MMAP_SIZE>0 + { "getpagesize", (sqlite3_syscall_ptr)unixGetpagesize, 0 }, +#else + { "getpagesize", (sqlite3_syscall_ptr)0, 0 }, +#endif +#define osGetpagesize ((int(*)(void))aSyscall[25].pCurrent) + +#if defined(HAVE_READLINK) + { "readlink", (sqlite3_syscall_ptr)readlink, 0 }, +#else + { "readlink", (sqlite3_syscall_ptr)0, 0 }, +#endif +#define osReadlink ((ssize_t(*)(const char*,char*,size_t))aSyscall[26].pCurrent) + +#if defined(HAVE_LSTAT) + { "lstat", (sqlite3_syscall_ptr)lstat, 0 }, +#else + { "lstat", (sqlite3_syscall_ptr)0, 0 }, +#endif +#define osLstat ((int(*)(const char*,struct stat*))aSyscall[27].pCurrent) + +#if defined(__linux__) && defined(SQLITE_ENABLE_BATCH_ATOMIC_WRITE) +# ifdef __ANDROID__ + { "ioctl", (sqlite3_syscall_ptr)(int(*)(int, int, ...))ioctl, 0 }, +#define osIoctl ((int(*)(int,int,...))aSyscall[28].pCurrent) +# else + { "ioctl", (sqlite3_syscall_ptr)ioctl, 0 }, +#define osIoctl ((int(*)(int,unsigned long,...))aSyscall[28].pCurrent) +# endif +#else + { "ioctl", (sqlite3_syscall_ptr)0, 0 }, +#endif + +}; /* End of the overrideable system calls */ + + +/* +** On some systems, calls to fchown() will trigger a message in a security +** log if they come from non-root processes. So avoid calling fchown() if +** we are not running as root. +*/ +static int robustFchown(int fd, uid_t uid, gid_t gid){ +#if defined(HAVE_FCHOWN) + return osGeteuid() ? 0 : osFchown(fd,uid,gid); +#else + return 0; +#endif +} + +/* +** This is the xSetSystemCall() method of sqlite3_vfs for all of the +** "unix" VFSes. Return SQLITE_OK opon successfully updating the +** system call pointer, or SQLITE_NOTFOUND if there is no configurable +** system call named zName. +*/ +static int unixSetSystemCall( + sqlite3_vfs *pNotUsed, /* The VFS pointer. Not used */ + const char *zName, /* Name of system call to override */ + sqlite3_syscall_ptr pNewFunc /* Pointer to new system call value */ +){ + unsigned int i; + int rc = SQLITE_NOTFOUND; + + UNUSED_PARAMETER(pNotUsed); + if( zName==0 ){ + /* If no zName is given, restore all system calls to their default + ** settings and return NULL + */ + rc = SQLITE_OK; + for(i=0; i=SQLITE_MINIMUM_FILE_DESCRIPTOR ) break; + osClose(fd); + sqlite3_log(SQLITE_WARNING, + "attempt to open \"%s\" as file descriptor %d", z, fd); + fd = -1; + if( osOpen("/dev/null", f, m)<0 ) break; + } + if( fd>=0 ){ + if( m!=0 ){ + struct stat statbuf; + if( osFstat(fd, &statbuf)==0 + && statbuf.st_size==0 + && (statbuf.st_mode&0777)!=m + ){ + osFchmod(fd, m); + } + } +#if defined(FD_CLOEXEC) && (!defined(O_CLOEXEC) || O_CLOEXEC==0) + osFcntl(fd, F_SETFD, osFcntl(fd, F_GETFD, 0) | FD_CLOEXEC); +#endif + } + return fd; +} + +/* +** Helper functions to obtain and relinquish the global mutex. The +** global mutex is used to protect the unixInodeInfo and +** vxworksFileId objects used by this file, all of which may be +** shared by multiple threads. +** +** Function unixMutexHeld() is used to assert() that the global mutex +** is held when required. This function is only used as part of assert() +** statements. e.g. +** +** unixEnterMutex() +** assert( unixMutexHeld() ); +** unixEnterLeave() +** +** To prevent deadlock, the global unixBigLock must must be acquired +** before the unixInodeInfo.pLockMutex mutex, if both are held. It is +** OK to get the pLockMutex without holding unixBigLock first, but if +** that happens, the unixBigLock mutex must not be acquired until after +** pLockMutex is released. +** +** OK: enter(unixBigLock), enter(pLockInfo) +** OK: enter(unixBigLock) +** OK: enter(pLockInfo) +** ERROR: enter(pLockInfo), enter(unixBigLock) +*/ +static sqlite3_mutex *unixBigLock = 0; +static void unixEnterMutex(void){ + assert( sqlite3_mutex_notheld(unixBigLock) ); /* Not a recursive mutex */ + sqlite3_mutex_enter(unixBigLock); +} +static void unixLeaveMutex(void){ + assert( sqlite3_mutex_held(unixBigLock) ); + sqlite3_mutex_leave(unixBigLock); +} +#ifdef SQLITE_DEBUG +static int unixMutexHeld(void) { + return sqlite3_mutex_held(unixBigLock); +} +#endif + + +#ifdef SQLITE_HAVE_OS_TRACE +/* +** Helper function for printing out trace information from debugging +** binaries. This returns the string representation of the supplied +** integer lock-type. +*/ +static const char *azFileLock(int eFileLock){ + switch( eFileLock ){ + case NO_LOCK: return "NONE"; + case SHARED_LOCK: return "SHARED"; + case RESERVED_LOCK: return "RESERVED"; + case PENDING_LOCK: return "PENDING"; + case EXCLUSIVE_LOCK: return "EXCLUSIVE"; + } + return "ERROR"; +} +#endif + +#ifdef SQLITE_LOCK_TRACE +/* +** Print out information about all locking operations. +** +** This routine is used for troubleshooting locks on multithreaded +** platforms. Enable by compiling with the -DSQLITE_LOCK_TRACE +** command-line option on the compiler. This code is normally +** turned off. +*/ +static int lockTrace(int fd, int op, struct flock *p){ + char *zOpName, *zType; + int s; + int savedErrno; + if( op==F_GETLK ){ + zOpName = "GETLK"; + }else if( op==F_SETLK ){ + zOpName = "SETLK"; + }else{ + s = osFcntl(fd, op, p); + sqlite3DebugPrintf("fcntl unknown %d %d %d\n", fd, op, s); + return s; + } + if( p->l_type==F_RDLCK ){ + zType = "RDLCK"; + }else if( p->l_type==F_WRLCK ){ + zType = "WRLCK"; + }else if( p->l_type==F_UNLCK ){ + zType = "UNLCK"; + }else{ + assert( 0 ); + } + assert( p->l_whence==SEEK_SET ); + s = osFcntl(fd, op, p); + savedErrno = errno; + sqlite3DebugPrintf("fcntl %d %d %s %s %d %d %d %d\n", + threadid, fd, zOpName, zType, (int)p->l_start, (int)p->l_len, + (int)p->l_pid, s); + if( s==(-1) && op==F_SETLK && (p->l_type==F_RDLCK || p->l_type==F_WRLCK) ){ + struct flock l2; + l2 = *p; + osFcntl(fd, F_GETLK, &l2); + if( l2.l_type==F_RDLCK ){ + zType = "RDLCK"; + }else if( l2.l_type==F_WRLCK ){ + zType = "WRLCK"; + }else if( l2.l_type==F_UNLCK ){ + zType = "UNLCK"; + }else{ + assert( 0 ); + } + sqlite3DebugPrintf("fcntl-failure-reason: %s %d %d %d\n", + zType, (int)l2.l_start, (int)l2.l_len, (int)l2.l_pid); + } + errno = savedErrno; + return s; +} +#undef osFcntl +#define osFcntl lockTrace +#endif /* SQLITE_LOCK_TRACE */ + +/* +** Retry ftruncate() calls that fail due to EINTR +** +** All calls to ftruncate() within this file should be made through +** this wrapper. On the Android platform, bypassing the logic below +** could lead to a corrupt database. +*/ +static int robust_ftruncate(int h, sqlite3_int64 sz){ + int rc; +#ifdef __ANDROID__ + /* On Android, ftruncate() always uses 32-bit offsets, even if + ** _FILE_OFFSET_BITS=64 is defined. This means it is unsafe to attempt to + ** truncate a file to any size larger than 2GiB. Silently ignore any + ** such attempts. */ + if( sz>(sqlite3_int64)0x7FFFFFFF ){ + rc = SQLITE_OK; + }else +#endif + do{ rc = osFtruncate(h,sz); }while( rc<0 && errno==EINTR ); + return rc; +} + +/* +** This routine translates a standard POSIX errno code into something +** useful to the clients of the sqlite3 functions. Specifically, it is +** intended to translate a variety of "try again" errors into SQLITE_BUSY +** and a variety of "please close the file descriptor NOW" errors into +** SQLITE_IOERR +** +** Errors during initialization of locks, or file system support for locks, +** should handle ENOLCK, ENOTSUP, EOPNOTSUPP separately. +*/ +static int sqliteErrorFromPosixError(int posixError, int sqliteIOErr) { + assert( (sqliteIOErr == SQLITE_IOERR_LOCK) || + (sqliteIOErr == SQLITE_IOERR_UNLOCK) || + (sqliteIOErr == SQLITE_IOERR_RDLOCK) || + (sqliteIOErr == SQLITE_IOERR_CHECKRESERVEDLOCK) ); + switch (posixError) { + case EACCES: + case EAGAIN: + case ETIMEDOUT: + case EBUSY: + case EINTR: + case ENOLCK: + /* random NFS retry error, unless during file system support + * introspection, in which it actually means what it says */ + return SQLITE_BUSY; + + case EPERM: + return SQLITE_PERM; + + default: + return sqliteIOErr; + } +} + + +/****************************************************************************** +****************** Begin Unique File ID Utility Used By VxWorks *************** +** +** On most versions of unix, we can get a unique ID for a file by concatenating +** the device number and the inode number. But this does not work on VxWorks. +** On VxWorks, a unique file id must be based on the canonical filename. +** +** A pointer to an instance of the following structure can be used as a +** unique file ID in VxWorks. Each instance of this structure contains +** a copy of the canonical filename. There is also a reference count. +** The structure is reclaimed when the number of pointers to it drops to +** zero. +** +** There are never very many files open at one time and lookups are not +** a performance-critical path, so it is sufficient to put these +** structures on a linked list. +*/ +struct vxworksFileId { + struct vxworksFileId *pNext; /* Next in a list of them all */ + int nRef; /* Number of references to this one */ + int nName; /* Length of the zCanonicalName[] string */ + char *zCanonicalName; /* Canonical filename */ +}; + +#if OS_VXWORKS +/* +** All unique filenames are held on a linked list headed by this +** variable: +*/ +static struct vxworksFileId *vxworksFileList = 0; + +/* +** Simplify a filename into its canonical form +** by making the following changes: +** +** * removing any trailing and duplicate / +** * convert /./ into just / +** * convert /A/../ where A is any simple name into just / +** +** Changes are made in-place. Return the new name length. +** +** The original filename is in z[0..n-1]. Return the number of +** characters in the simplified name. +*/ +static int vxworksSimplifyName(char *z, int n){ + int i, j; + while( n>1 && z[n-1]=='/' ){ n--; } + for(i=j=0; i0 && z[j-1]!='/' ){ j--; } + if( j>0 ){ j--; } + i += 2; + continue; + } + } + z[j++] = z[i]; + } + z[j] = 0; + return j; +} + +/* +** Find a unique file ID for the given absolute pathname. Return +** a pointer to the vxworksFileId object. This pointer is the unique +** file ID. +** +** The nRef field of the vxworksFileId object is incremented before +** the object is returned. A new vxworksFileId object is created +** and added to the global list if necessary. +** +** If a memory allocation error occurs, return NULL. +*/ +static struct vxworksFileId *vxworksFindFileId(const char *zAbsoluteName){ + struct vxworksFileId *pNew; /* search key and new file ID */ + struct vxworksFileId *pCandidate; /* For looping over existing file IDs */ + int n; /* Length of zAbsoluteName string */ + + assert( zAbsoluteName[0]=='/' ); + n = (int)strlen(zAbsoluteName); + pNew = sqlite3_malloc64( sizeof(*pNew) + (n+1) ); + if( pNew==0 ) return 0; + pNew->zCanonicalName = (char*)&pNew[1]; + memcpy(pNew->zCanonicalName, zAbsoluteName, n+1); + n = vxworksSimplifyName(pNew->zCanonicalName, n); + + /* Search for an existing entry that matching the canonical name. + ** If found, increment the reference count and return a pointer to + ** the existing file ID. + */ + unixEnterMutex(); + for(pCandidate=vxworksFileList; pCandidate; pCandidate=pCandidate->pNext){ + if( pCandidate->nName==n + && memcmp(pCandidate->zCanonicalName, pNew->zCanonicalName, n)==0 + ){ + sqlite3_free(pNew); + pCandidate->nRef++; + unixLeaveMutex(); + return pCandidate; + } + } + + /* No match was found. We will make a new file ID */ + pNew->nRef = 1; + pNew->nName = n; + pNew->pNext = vxworksFileList; + vxworksFileList = pNew; + unixLeaveMutex(); + return pNew; +} + +/* +** Decrement the reference count on a vxworksFileId object. Free +** the object when the reference count reaches zero. +*/ +static void vxworksReleaseFileId(struct vxworksFileId *pId){ + unixEnterMutex(); + assert( pId->nRef>0 ); + pId->nRef--; + if( pId->nRef==0 ){ + struct vxworksFileId **pp; + for(pp=&vxworksFileList; *pp && *pp!=pId; pp = &((*pp)->pNext)){} + assert( *pp==pId ); + *pp = pId->pNext; + sqlite3_free(pId); + } + unixLeaveMutex(); +} +#endif /* OS_VXWORKS */ +/*************** End of Unique File ID Utility Used By VxWorks **************** +******************************************************************************/ + + +/****************************************************************************** +*************************** Posix Advisory Locking **************************** +** +** POSIX advisory locks are broken by design. ANSI STD 1003.1 (1996) +** section 6.5.2.2 lines 483 through 490 specify that when a process +** sets or clears a lock, that operation overrides any prior locks set +** by the same process. It does not explicitly say so, but this implies +** that it overrides locks set by the same process using a different +** file descriptor. Consider this test case: +** +** int fd1 = open("./file1", O_RDWR|O_CREAT, 0644); +** int fd2 = open("./file2", O_RDWR|O_CREAT, 0644); +** +** Suppose ./file1 and ./file2 are really the same file (because +** one is a hard or symbolic link to the other) then if you set +** an exclusive lock on fd1, then try to get an exclusive lock +** on fd2, it works. I would have expected the second lock to +** fail since there was already a lock on the file due to fd1. +** But not so. Since both locks came from the same process, the +** second overrides the first, even though they were on different +** file descriptors opened on different file names. +** +** This means that we cannot use POSIX locks to synchronize file access +** among competing threads of the same process. POSIX locks will work fine +** to synchronize access for threads in separate processes, but not +** threads within the same process. +** +** To work around the problem, SQLite has to manage file locks internally +** on its own. Whenever a new database is opened, we have to find the +** specific inode of the database file (the inode is determined by the +** st_dev and st_ino fields of the stat structure that fstat() fills in) +** and check for locks already existing on that inode. When locks are +** created or removed, we have to look at our own internal record of the +** locks to see if another thread has previously set a lock on that same +** inode. +** +** (Aside: The use of inode numbers as unique IDs does not work on VxWorks. +** For VxWorks, we have to use the alternative unique ID system based on +** canonical filename and implemented in the previous division.) +** +** The sqlite3_file structure for POSIX is no longer just an integer file +** descriptor. It is now a structure that holds the integer file +** descriptor and a pointer to a structure that describes the internal +** locks on the corresponding inode. There is one locking structure +** per inode, so if the same inode is opened twice, both unixFile structures +** point to the same locking structure. The locking structure keeps +** a reference count (so we will know when to delete it) and a "cnt" +** field that tells us its internal lock status. cnt==0 means the +** file is unlocked. cnt==-1 means the file has an exclusive lock. +** cnt>0 means there are cnt shared locks on the file. +** +** Any attempt to lock or unlock a file first checks the locking +** structure. The fcntl() system call is only invoked to set a +** POSIX lock if the internal lock structure transitions between +** a locked and an unlocked state. +** +** But wait: there are yet more problems with POSIX advisory locks. +** +** If you close a file descriptor that points to a file that has locks, +** all locks on that file that are owned by the current process are +** released. To work around this problem, each unixInodeInfo object +** maintains a count of the number of pending locks on tha inode. +** When an attempt is made to close an unixFile, if there are +** other unixFile open on the same inode that are holding locks, the call +** to close() the file descriptor is deferred until all of the locks clear. +** The unixInodeInfo structure keeps a list of file descriptors that need to +** be closed and that list is walked (and cleared) when the last lock +** clears. +** +** Yet another problem: LinuxThreads do not play well with posix locks. +** +** Many older versions of linux use the LinuxThreads library which is +** not posix compliant. Under LinuxThreads, a lock created by thread +** A cannot be modified or overridden by a different thread B. +** Only thread A can modify the lock. Locking behavior is correct +** if the appliation uses the newer Native Posix Thread Library (NPTL) +** on linux - with NPTL a lock created by thread A can override locks +** in thread B. But there is no way to know at compile-time which +** threading library is being used. So there is no way to know at +** compile-time whether or not thread A can override locks on thread B. +** One has to do a run-time check to discover the behavior of the +** current process. +** +** SQLite used to support LinuxThreads. But support for LinuxThreads +** was dropped beginning with version 3.7.0. SQLite will still work with +** LinuxThreads provided that (1) there is no more than one connection +** per database file in the same process and (2) database connections +** do not move across threads. +*/ + +/* +** An instance of the following structure serves as the key used +** to locate a particular unixInodeInfo object. +*/ +struct unixFileId { + dev_t dev; /* Device number */ +#if OS_VXWORKS + struct vxworksFileId *pId; /* Unique file ID for vxworks. */ +#else + /* We are told that some versions of Android contain a bug that + ** sizes ino_t at only 32-bits instead of 64-bits. (See + ** https://android-review.googlesource.com/#/c/115351/3/dist/sqlite3.c) + ** To work around this, always allocate 64-bits for the inode number. + ** On small machines that only have 32-bit inodes, this wastes 4 bytes, + ** but that should not be a big deal. */ + /* WAS: ino_t ino; */ + u64 ino; /* Inode number */ +#endif +}; + +/* +** An instance of the following structure is allocated for each open +** inode. +** +** A single inode can have multiple file descriptors, so each unixFile +** structure contains a pointer to an instance of this object and this +** object keeps a count of the number of unixFile pointing to it. +** +** Mutex rules: +** +** (1) Only the pLockMutex mutex must be held in order to read or write +** any of the locking fields: +** nShared, nLock, eFileLock, bProcessLock, pUnused +** +** (2) When nRef>0, then the following fields are unchanging and can +** be read (but not written) without holding any mutex: +** fileId, pLockMutex +** +** (3) With the exceptions above, all the fields may only be read +** or written while holding the global unixBigLock mutex. +** +** Deadlock prevention: The global unixBigLock mutex may not +** be acquired while holding the pLockMutex mutex. If both unixBigLock +** and pLockMutex are needed, then unixBigLock must be acquired first. +*/ +struct unixInodeInfo { + struct unixFileId fileId; /* The lookup key */ + sqlite3_mutex *pLockMutex; /* Hold this mutex for... */ + int nShared; /* Number of SHARED locks held */ + int nLock; /* Number of outstanding file locks */ + unsigned char eFileLock; /* One of SHARED_LOCK, RESERVED_LOCK etc. */ + unsigned char bProcessLock; /* An exclusive process lock is held */ + UnixUnusedFd *pUnused; /* Unused file descriptors to close */ + int nRef; /* Number of pointers to this structure */ + unixShmNode *pShmNode; /* Shared memory associated with this inode */ + unixInodeInfo *pNext; /* List of all unixInodeInfo objects */ + unixInodeInfo *pPrev; /* .... doubly linked */ +#if SQLITE_ENABLE_LOCKING_STYLE + unsigned long long sharedByte; /* for AFP simulated shared lock */ +#endif +#if OS_VXWORKS + sem_t *pSem; /* Named POSIX semaphore */ + char aSemName[MAX_PATHNAME+2]; /* Name of that semaphore */ +#endif +}; + +/* +** A lists of all unixInodeInfo objects. +** +** Must hold unixBigLock in order to read or write this variable. +*/ +static unixInodeInfo *inodeList = 0; /* All unixInodeInfo objects */ + +#ifdef SQLITE_DEBUG +/* +** True if the inode mutex (on the unixFile.pFileMutex field) is held, or not. +** This routine is used only within assert() to help verify correct mutex +** usage. +*/ +int unixFileMutexHeld(unixFile *pFile){ + assert( pFile->pInode ); + return sqlite3_mutex_held(pFile->pInode->pLockMutex); +} +int unixFileMutexNotheld(unixFile *pFile){ + assert( pFile->pInode ); + return sqlite3_mutex_notheld(pFile->pInode->pLockMutex); +} +#endif + +/* +** +** This function - unixLogErrorAtLine(), is only ever called via the macro +** unixLogError(). +** +** It is invoked after an error occurs in an OS function and errno has been +** set. It logs a message using sqlite3_log() containing the current value of +** errno and, if possible, the human-readable equivalent from strerror() or +** strerror_r(). +** +** The first argument passed to the macro should be the error code that +** will be returned to SQLite (e.g. SQLITE_IOERR_DELETE, SQLITE_CANTOPEN). +** The two subsequent arguments should be the name of the OS function that +** failed (e.g. "unlink", "open") and the associated file-system path, +** if any. +*/ +#define unixLogError(a,b,c) unixLogErrorAtLine(a,b,c,__LINE__) +static int unixLogErrorAtLine( + int errcode, /* SQLite error code */ + const char *zFunc, /* Name of OS function that failed */ + const char *zPath, /* File path associated with error */ + int iLine /* Source line number where error occurred */ +){ + char *zErr; /* Message from strerror() or equivalent */ + int iErrno = errno; /* Saved syscall error number */ + + /* If this is not a threadsafe build (SQLITE_THREADSAFE==0), then use + ** the strerror() function to obtain the human-readable error message + ** equivalent to errno. Otherwise, use strerror_r(). + */ +#if SQLITE_THREADSAFE && defined(HAVE_STRERROR_R) + char aErr[80]; + memset(aErr, 0, sizeof(aErr)); + zErr = aErr; + + /* If STRERROR_R_CHAR_P (set by autoconf scripts) or __USE_GNU is defined, + ** assume that the system provides the GNU version of strerror_r() that + ** returns a pointer to a buffer containing the error message. That pointer + ** may point to aErr[], or it may point to some static storage somewhere. + ** Otherwise, assume that the system provides the POSIX version of + ** strerror_r(), which always writes an error message into aErr[]. + ** + ** If the code incorrectly assumes that it is the POSIX version that is + ** available, the error message will often be an empty string. Not a + ** huge problem. Incorrectly concluding that the GNU version is available + ** could lead to a segfault though. + */ +#if defined(STRERROR_R_CHAR_P) || defined(__USE_GNU) + zErr = +# endif + strerror_r(iErrno, aErr, sizeof(aErr)-1); + +#elif SQLITE_THREADSAFE + /* This is a threadsafe build, but strerror_r() is not available. */ + zErr = ""; +#else + /* Non-threadsafe build, use strerror(). */ + zErr = strerror(iErrno); +#endif + + if( zPath==0 ) zPath = ""; + sqlite3_log(errcode, + "os_unix.c:%d: (%d) %s(%s) - %s", + iLine, iErrno, zFunc, zPath, zErr + ); + + return errcode; +} + +/* +** Close a file descriptor. +** +** We assume that close() almost always works, since it is only in a +** very sick application or on a very sick platform that it might fail. +** If it does fail, simply leak the file descriptor, but do log the +** error. +** +** Note that it is not safe to retry close() after EINTR since the +** file descriptor might have already been reused by another thread. +** So we don't even try to recover from an EINTR. Just log the error +** and move on. +*/ +static void robust_close(unixFile *pFile, int h, int lineno){ + if( osClose(h) ){ + unixLogErrorAtLine(SQLITE_IOERR_CLOSE, "close", + pFile ? pFile->zPath : 0, lineno); + } +} + +/* +** Set the pFile->lastErrno. Do this in a subroutine as that provides +** a convenient place to set a breakpoint. +*/ +static void storeLastErrno(unixFile *pFile, int error){ + pFile->lastErrno = error; +} + +/* +** Close all file descriptors accumuated in the unixInodeInfo->pUnused list. +*/ +static void closePendingFds(unixFile *pFile){ + unixInodeInfo *pInode = pFile->pInode; + UnixUnusedFd *p; + UnixUnusedFd *pNext; + assert( unixFileMutexHeld(pFile) ); + for(p=pInode->pUnused; p; p=pNext){ + pNext = p->pNext; + robust_close(pFile, p->fd, __LINE__); + sqlite3_free(p); + } + pInode->pUnused = 0; +} + +/* +** Release a unixInodeInfo structure previously allocated by findInodeInfo(). +** +** The global mutex must be held when this routine is called, but the mutex +** on the inode being deleted must NOT be held. +*/ +static void releaseInodeInfo(unixFile *pFile){ + unixInodeInfo *pInode = pFile->pInode; + assert( unixMutexHeld() ); + assert( unixFileMutexNotheld(pFile) ); + if( ALWAYS(pInode) ){ + pInode->nRef--; + if( pInode->nRef==0 ){ + assert( pInode->pShmNode==0 ); + sqlite3_mutex_enter(pInode->pLockMutex); + closePendingFds(pFile); + sqlite3_mutex_leave(pInode->pLockMutex); + if( pInode->pPrev ){ + assert( pInode->pPrev->pNext==pInode ); + pInode->pPrev->pNext = pInode->pNext; + }else{ + assert( inodeList==pInode ); + inodeList = pInode->pNext; + } + if( pInode->pNext ){ + assert( pInode->pNext->pPrev==pInode ); + pInode->pNext->pPrev = pInode->pPrev; + } + sqlite3_mutex_free(pInode->pLockMutex); + sqlite3_free(pInode); + } + } +} + +/* +** Given a file descriptor, locate the unixInodeInfo object that +** describes that file descriptor. Create a new one if necessary. The +** return value might be uninitialized if an error occurs. +** +** The global mutex must held when calling this routine. +** +** Return an appropriate error code. +*/ +static int findInodeInfo( + unixFile *pFile, /* Unix file with file desc used in the key */ + unixInodeInfo **ppInode /* Return the unixInodeInfo object here */ +){ + int rc; /* System call return code */ + int fd; /* The file descriptor for pFile */ + struct unixFileId fileId; /* Lookup key for the unixInodeInfo */ + struct stat statbuf; /* Low-level file information */ + unixInodeInfo *pInode = 0; /* Candidate unixInodeInfo object */ + + assert( unixMutexHeld() ); + + /* Get low-level information about the file that we can used to + ** create a unique name for the file. + */ + fd = pFile->h; + rc = osFstat(fd, &statbuf); + if( rc!=0 ){ + storeLastErrno(pFile, errno); +#if defined(EOVERFLOW) && defined(SQLITE_DISABLE_LFS) + if( pFile->lastErrno==EOVERFLOW ) return SQLITE_NOLFS; +#endif + return SQLITE_IOERR; + } + +#ifdef __APPLE__ + /* On OS X on an msdos filesystem, the inode number is reported + ** incorrectly for zero-size files. See ticket #3260. To work + ** around this problem (we consider it a bug in OS X, not SQLite) + ** we always increase the file size to 1 by writing a single byte + ** prior to accessing the inode number. The one byte written is + ** an ASCII 'S' character which also happens to be the first byte + ** in the header of every SQLite database. In this way, if there + ** is a race condition such that another thread has already populated + ** the first page of the database, no damage is done. + */ + if( statbuf.st_size==0 && (pFile->fsFlags & SQLITE_FSFLAGS_IS_MSDOS)!=0 ){ + do{ rc = osWrite(fd, "S", 1); }while( rc<0 && errno==EINTR ); + if( rc!=1 ){ + storeLastErrno(pFile, errno); + return SQLITE_IOERR; + } + rc = osFstat(fd, &statbuf); + if( rc!=0 ){ + storeLastErrno(pFile, errno); + return SQLITE_IOERR; + } + } +#endif + + memset(&fileId, 0, sizeof(fileId)); + fileId.dev = statbuf.st_dev; +#if OS_VXWORKS + fileId.pId = pFile->pId; +#else + fileId.ino = (u64)statbuf.st_ino; +#endif + assert( unixMutexHeld() ); + pInode = inodeList; + while( pInode && memcmp(&fileId, &pInode->fileId, sizeof(fileId)) ){ + pInode = pInode->pNext; + } + if( pInode==0 ){ + pInode = sqlite3_malloc64( sizeof(*pInode) ); + if( pInode==0 ){ + return SQLITE_NOMEM_BKPT; + } + memset(pInode, 0, sizeof(*pInode)); + memcpy(&pInode->fileId, &fileId, sizeof(fileId)); + if( sqlite3GlobalConfig.bCoreMutex ){ + pInode->pLockMutex = sqlite3_mutex_alloc(SQLITE_MUTEX_FAST); + if( pInode->pLockMutex==0 ){ + sqlite3_free(pInode); + return SQLITE_NOMEM_BKPT; + } + } + pInode->nRef = 1; + assert( unixMutexHeld() ); + pInode->pNext = inodeList; + pInode->pPrev = 0; + if( inodeList ) inodeList->pPrev = pInode; + inodeList = pInode; + }else{ + pInode->nRef++; + } + *ppInode = pInode; + return SQLITE_OK; +} + +/* +** Return TRUE if pFile has been renamed or unlinked since it was first opened. +*/ +static int fileHasMoved(unixFile *pFile){ +#if OS_VXWORKS + return pFile->pInode!=0 && pFile->pId!=pFile->pInode->fileId.pId; +#else + struct stat buf; + return pFile->pInode!=0 && + (osStat(pFile->zPath, &buf)!=0 + || (u64)buf.st_ino!=pFile->pInode->fileId.ino); +#endif +} + + +/* +** Check a unixFile that is a database. Verify the following: +** +** (1) There is exactly one hard link on the file +** (2) The file is not a symbolic link +** (3) The file has not been renamed or unlinked +** +** Issue sqlite3_log(SQLITE_WARNING,...) messages if anything is not right. +*/ +static void verifyDbFile(unixFile *pFile){ + struct stat buf; + int rc; + + /* These verifications occurs for the main database only */ + if( pFile->ctrlFlags & UNIXFILE_NOLOCK ) return; + + rc = osFstat(pFile->h, &buf); + if( rc!=0 ){ + sqlite3_log(SQLITE_WARNING, "cannot fstat db file %s", pFile->zPath); + return; + } + if( buf.st_nlink==0 ){ + sqlite3_log(SQLITE_WARNING, "file unlinked while open: %s", pFile->zPath); + return; + } + if( buf.st_nlink>1 ){ + sqlite3_log(SQLITE_WARNING, "multiple links to file: %s", pFile->zPath); + return; + } + if( fileHasMoved(pFile) ){ + sqlite3_log(SQLITE_WARNING, "file renamed while open: %s", pFile->zPath); + return; + } +} + + +/* +** This routine checks if there is a RESERVED lock held on the specified +** file by this or any other process. If such a lock is held, set *pResOut +** to a non-zero value otherwise *pResOut is set to zero. The return value +** is set to SQLITE_OK unless an I/O error occurs during lock checking. +*/ +static int unixCheckReservedLock(sqlite3_file *id, int *pResOut){ + int rc = SQLITE_OK; + int reserved = 0; + unixFile *pFile = (unixFile*)id; + + SimulateIOError( return SQLITE_IOERR_CHECKRESERVEDLOCK; ); + + assert( pFile ); + assert( pFile->eFileLock<=SHARED_LOCK ); + sqlite3_mutex_enter(pFile->pInode->pLockMutex); + + /* Check if a thread in this process holds such a lock */ + if( pFile->pInode->eFileLock>SHARED_LOCK ){ + reserved = 1; + } + + /* Otherwise see if some other process holds it. + */ +#ifndef __DJGPP__ + if( !reserved && !pFile->pInode->bProcessLock ){ + struct flock lock; + lock.l_whence = SEEK_SET; + lock.l_start = RESERVED_BYTE; + lock.l_len = 1; + lock.l_type = F_WRLCK; + if( osFcntl(pFile->h, F_GETLK, &lock) ){ + rc = SQLITE_IOERR_CHECKRESERVEDLOCK; + storeLastErrno(pFile, errno); + } else if( lock.l_type!=F_UNLCK ){ + reserved = 1; + } + } +#endif + + sqlite3_mutex_leave(pFile->pInode->pLockMutex); + OSTRACE(("TEST WR-LOCK %d %d %d (unix)\n", pFile->h, rc, reserved)); + + *pResOut = reserved; + return rc; +} + +/* +** Set a posix-advisory-lock. +** +** There are two versions of this routine. If compiled with +** SQLITE_ENABLE_SETLK_TIMEOUT then the routine has an extra parameter +** which is a pointer to a unixFile. If the unixFile->iBusyTimeout +** value is set, then it is the number of milliseconds to wait before +** failing the lock. The iBusyTimeout value is always reset back to +** zero on each call. +** +** If SQLITE_ENABLE_SETLK_TIMEOUT is not defined, then do a non-blocking +** attempt to set the lock. +*/ +#ifndef SQLITE_ENABLE_SETLK_TIMEOUT +# define osSetPosixAdvisoryLock(h,x,t) osFcntl(h,F_SETLK,x) +#else +static int osSetPosixAdvisoryLock( + int h, /* The file descriptor on which to take the lock */ + struct flock *pLock, /* The description of the lock */ + unixFile *pFile /* Structure holding timeout value */ +){ + int rc = osFcntl(h,F_SETLK,pLock); + while( rc<0 && pFile->iBusyTimeout>0 ){ + /* On systems that support some kind of blocking file lock with a timeout, + ** make appropriate changes here to invoke that blocking file lock. On + ** generic posix, however, there is no such API. So we simply try the + ** lock once every millisecond until either the timeout expires, or until + ** the lock is obtained. */ + usleep(1000); + rc = osFcntl(h,F_SETLK,pLock); + pFile->iBusyTimeout--; + } + return rc; +} +#endif /* SQLITE_ENABLE_SETLK_TIMEOUT */ + + +/* +** Attempt to set a system-lock on the file pFile. The lock is +** described by pLock. +** +** If the pFile was opened read/write from unix-excl, then the only lock +** ever obtained is an exclusive lock, and it is obtained exactly once +** the first time any lock is attempted. All subsequent system locking +** operations become no-ops. Locking operations still happen internally, +** in order to coordinate access between separate database connections +** within this process, but all of that is handled in memory and the +** operating system does not participate. +** +** This function is a pass-through to fcntl(F_SETLK) if pFile is using +** any VFS other than "unix-excl" or if pFile is opened on "unix-excl" +** and is read-only. +** +** Zero is returned if the call completes successfully, or -1 if a call +** to fcntl() fails. In this case, errno is set appropriately (by fcntl()). +*/ +static int unixFileLock(unixFile *pFile, struct flock *pLock){ + int rc; + unixInodeInfo *pInode = pFile->pInode; + assert( pInode!=0 ); + assert( sqlite3_mutex_held(pInode->pLockMutex) ); + if( (pFile->ctrlFlags & (UNIXFILE_EXCL|UNIXFILE_RDONLY))==UNIXFILE_EXCL ){ + if( pInode->bProcessLock==0 ){ + struct flock lock; + assert( pInode->nLock==0 ); + lock.l_whence = SEEK_SET; + lock.l_start = SHARED_FIRST; + lock.l_len = SHARED_SIZE; + lock.l_type = F_WRLCK; + rc = osSetPosixAdvisoryLock(pFile->h, &lock, pFile); + if( rc<0 ) return rc; + pInode->bProcessLock = 1; + pInode->nLock++; + }else{ + rc = 0; + } + }else{ + rc = osSetPosixAdvisoryLock(pFile->h, pLock, pFile); + } + return rc; +} + +/* +** Lock the file with the lock specified by parameter eFileLock - one +** of the following: +** +** (1) SHARED_LOCK +** (2) RESERVED_LOCK +** (3) PENDING_LOCK +** (4) EXCLUSIVE_LOCK +** +** Sometimes when requesting one lock state, additional lock states +** are inserted in between. The locking might fail on one of the later +** transitions leaving the lock state different from what it started but +** still short of its goal. The following chart shows the allowed +** transitions and the inserted intermediate states: +** +** UNLOCKED -> SHARED +** SHARED -> RESERVED +** SHARED -> (PENDING) -> EXCLUSIVE +** RESERVED -> (PENDING) -> EXCLUSIVE +** PENDING -> EXCLUSIVE +** +** This routine will only increase a lock. Use the sqlite3OsUnlock() +** routine to lower a locking level. +*/ +static int unixLock(sqlite3_file *id, int eFileLock){ + /* The following describes the implementation of the various locks and + ** lock transitions in terms of the POSIX advisory shared and exclusive + ** lock primitives (called read-locks and write-locks below, to avoid + ** confusion with SQLite lock names). The algorithms are complicated + ** slightly in order to be compatible with Windows95 systems simultaneously + ** accessing the same database file, in case that is ever required. + ** + ** Symbols defined in os.h indentify the 'pending byte' and the 'reserved + ** byte', each single bytes at well known offsets, and the 'shared byte + ** range', a range of 510 bytes at a well known offset. + ** + ** To obtain a SHARED lock, a read-lock is obtained on the 'pending + ** byte'. If this is successful, 'shared byte range' is read-locked + ** and the lock on the 'pending byte' released. (Legacy note: When + ** SQLite was first developed, Windows95 systems were still very common, + ** and Widnows95 lacks a shared-lock capability. So on Windows95, a + ** single randomly selected by from the 'shared byte range' is locked. + ** Windows95 is now pretty much extinct, but this work-around for the + ** lack of shared-locks on Windows95 lives on, for backwards + ** compatibility.) + ** + ** A process may only obtain a RESERVED lock after it has a SHARED lock. + ** A RESERVED lock is implemented by grabbing a write-lock on the + ** 'reserved byte'. + ** + ** A process may only obtain a PENDING lock after it has obtained a + ** SHARED lock. A PENDING lock is implemented by obtaining a write-lock + ** on the 'pending byte'. This ensures that no new SHARED locks can be + ** obtained, but existing SHARED locks are allowed to persist. A process + ** does not have to obtain a RESERVED lock on the way to a PENDING lock. + ** This property is used by the algorithm for rolling back a journal file + ** after a crash. + ** + ** An EXCLUSIVE lock, obtained after a PENDING lock is held, is + ** implemented by obtaining a write-lock on the entire 'shared byte + ** range'. Since all other locks require a read-lock on one of the bytes + ** within this range, this ensures that no other locks are held on the + ** database. + */ + int rc = SQLITE_OK; + unixFile *pFile = (unixFile*)id; + unixInodeInfo *pInode; + struct flock lock; + int tErrno = 0; + + assert( pFile ); + OSTRACE(("LOCK %d %s was %s(%s,%d) pid=%d (unix)\n", pFile->h, + azFileLock(eFileLock), azFileLock(pFile->eFileLock), + azFileLock(pFile->pInode->eFileLock), pFile->pInode->nShared, + osGetpid(0))); + + /* If there is already a lock of this type or more restrictive on the + ** unixFile, do nothing. Don't use the end_lock: exit path, as + ** unixEnterMutex() hasn't been called yet. + */ + if( pFile->eFileLock>=eFileLock ){ + OSTRACE(("LOCK %d %s ok (already held) (unix)\n", pFile->h, + azFileLock(eFileLock))); + return SQLITE_OK; + } + + /* Make sure the locking sequence is correct. + ** (1) We never move from unlocked to anything higher than shared lock. + ** (2) SQLite never explicitly requests a pendig lock. + ** (3) A shared lock is always held when a reserve lock is requested. + */ + assert( pFile->eFileLock!=NO_LOCK || eFileLock==SHARED_LOCK ); + assert( eFileLock!=PENDING_LOCK ); + assert( eFileLock!=RESERVED_LOCK || pFile->eFileLock==SHARED_LOCK ); + + /* This mutex is needed because pFile->pInode is shared across threads + */ + pInode = pFile->pInode; + sqlite3_mutex_enter(pInode->pLockMutex); + + /* If some thread using this PID has a lock via a different unixFile* + ** handle that precludes the requested lock, return BUSY. + */ + if( (pFile->eFileLock!=pInode->eFileLock && + (pInode->eFileLock>=PENDING_LOCK || eFileLock>SHARED_LOCK)) + ){ + rc = SQLITE_BUSY; + goto end_lock; + } + + /* If a SHARED lock is requested, and some thread using this PID already + ** has a SHARED or RESERVED lock, then increment reference counts and + ** return SQLITE_OK. + */ + if( eFileLock==SHARED_LOCK && + (pInode->eFileLock==SHARED_LOCK || pInode->eFileLock==RESERVED_LOCK) ){ + assert( eFileLock==SHARED_LOCK ); + assert( pFile->eFileLock==0 ); + assert( pInode->nShared>0 ); + pFile->eFileLock = SHARED_LOCK; + pInode->nShared++; + pInode->nLock++; + goto end_lock; + } + + + /* A PENDING lock is needed before acquiring a SHARED lock and before + ** acquiring an EXCLUSIVE lock. For the SHARED lock, the PENDING will + ** be released. + */ + lock.l_len = 1L; + lock.l_whence = SEEK_SET; + if( eFileLock==SHARED_LOCK + || (eFileLock==EXCLUSIVE_LOCK && pFile->eFileLocknShared==0 ); + assert( pInode->eFileLock==0 ); + assert( rc==SQLITE_OK ); + + /* Now get the read-lock */ + lock.l_start = SHARED_FIRST; + lock.l_len = SHARED_SIZE; + if( unixFileLock(pFile, &lock) ){ + tErrno = errno; + rc = sqliteErrorFromPosixError(tErrno, SQLITE_IOERR_LOCK); + } + + /* Drop the temporary PENDING lock */ + lock.l_start = PENDING_BYTE; + lock.l_len = 1L; + lock.l_type = F_UNLCK; + if( unixFileLock(pFile, &lock) && rc==SQLITE_OK ){ + /* This could happen with a network mount */ + tErrno = errno; + rc = SQLITE_IOERR_UNLOCK; + } + + if( rc ){ + if( rc!=SQLITE_BUSY ){ + storeLastErrno(pFile, tErrno); + } + goto end_lock; + }else{ + pFile->eFileLock = SHARED_LOCK; + pInode->nLock++; + pInode->nShared = 1; + } + }else if( eFileLock==EXCLUSIVE_LOCK && pInode->nShared>1 ){ + /* We are trying for an exclusive lock but another thread in this + ** same process is still holding a shared lock. */ + rc = SQLITE_BUSY; + }else{ + /* The request was for a RESERVED or EXCLUSIVE lock. It is + ** assumed that there is a SHARED or greater lock on the file + ** already. + */ + assert( 0!=pFile->eFileLock ); + lock.l_type = F_WRLCK; + + assert( eFileLock==RESERVED_LOCK || eFileLock==EXCLUSIVE_LOCK ); + if( eFileLock==RESERVED_LOCK ){ + lock.l_start = RESERVED_BYTE; + lock.l_len = 1L; + }else{ + lock.l_start = SHARED_FIRST; + lock.l_len = SHARED_SIZE; + } + + if( unixFileLock(pFile, &lock) ){ + tErrno = errno; + rc = sqliteErrorFromPosixError(tErrno, SQLITE_IOERR_LOCK); + if( rc!=SQLITE_BUSY ){ + storeLastErrno(pFile, tErrno); + } + } + } + + +#ifdef SQLITE_DEBUG + /* Set up the transaction-counter change checking flags when + ** transitioning from a SHARED to a RESERVED lock. The change + ** from SHARED to RESERVED marks the beginning of a normal + ** write operation (not a hot journal rollback). + */ + if( rc==SQLITE_OK + && pFile->eFileLock<=SHARED_LOCK + && eFileLock==RESERVED_LOCK + ){ + pFile->transCntrChng = 0; + pFile->dbUpdate = 0; + pFile->inNormalWrite = 1; + } +#endif + + + if( rc==SQLITE_OK ){ + pFile->eFileLock = eFileLock; + pInode->eFileLock = eFileLock; + }else if( eFileLock==EXCLUSIVE_LOCK ){ + pFile->eFileLock = PENDING_LOCK; + pInode->eFileLock = PENDING_LOCK; + } + +end_lock: + sqlite3_mutex_leave(pInode->pLockMutex); + OSTRACE(("LOCK %d %s %s (unix)\n", pFile->h, azFileLock(eFileLock), + rc==SQLITE_OK ? "ok" : "failed")); + return rc; +} + +/* +** Add the file descriptor used by file handle pFile to the corresponding +** pUnused list. +*/ +static void setPendingFd(unixFile *pFile){ + unixInodeInfo *pInode = pFile->pInode; + UnixUnusedFd *p = pFile->pPreallocatedUnused; + assert( unixFileMutexHeld(pFile) ); + p->pNext = pInode->pUnused; + pInode->pUnused = p; + pFile->h = -1; + pFile->pPreallocatedUnused = 0; +} + +/* +** Lower the locking level on file descriptor pFile to eFileLock. eFileLock +** must be either NO_LOCK or SHARED_LOCK. +** +** If the locking level of the file descriptor is already at or below +** the requested locking level, this routine is a no-op. +** +** If handleNFSUnlock is true, then on downgrading an EXCLUSIVE_LOCK to SHARED +** the byte range is divided into 2 parts and the first part is unlocked then +** set to a read lock, then the other part is simply unlocked. This works +** around a bug in BSD NFS lockd (also seen on MacOSX 10.3+) that fails to +** remove the write lock on a region when a read lock is set. +*/ +static int posixUnlock(sqlite3_file *id, int eFileLock, int handleNFSUnlock){ + unixFile *pFile = (unixFile*)id; + unixInodeInfo *pInode; + struct flock lock; + int rc = SQLITE_OK; + + assert( pFile ); + OSTRACE(("UNLOCK %d %d was %d(%d,%d) pid=%d (unix)\n", pFile->h, eFileLock, + pFile->eFileLock, pFile->pInode->eFileLock, pFile->pInode->nShared, + osGetpid(0))); + + assert( eFileLock<=SHARED_LOCK ); + if( pFile->eFileLock<=eFileLock ){ + return SQLITE_OK; + } + pInode = pFile->pInode; + sqlite3_mutex_enter(pInode->pLockMutex); + assert( pInode->nShared!=0 ); + if( pFile->eFileLock>SHARED_LOCK ){ + assert( pInode->eFileLock==pFile->eFileLock ); + +#ifdef SQLITE_DEBUG + /* When reducing a lock such that other processes can start + ** reading the database file again, make sure that the + ** transaction counter was updated if any part of the database + ** file changed. If the transaction counter is not updated, + ** other connections to the same file might not realize that + ** the file has changed and hence might not know to flush their + ** cache. The use of a stale cache can lead to database corruption. + */ + pFile->inNormalWrite = 0; +#endif + + /* downgrading to a shared lock on NFS involves clearing the write lock + ** before establishing the readlock - to avoid a race condition we downgrade + ** the lock in 2 blocks, so that part of the range will be covered by a + ** write lock until the rest is covered by a read lock: + ** 1: [WWWWW] + ** 2: [....W] + ** 3: [RRRRW] + ** 4: [RRRR.] + */ + if( eFileLock==SHARED_LOCK ){ +#if !defined(__APPLE__) || !SQLITE_ENABLE_LOCKING_STYLE + (void)handleNFSUnlock; + assert( handleNFSUnlock==0 ); +#endif +#if defined(__APPLE__) && SQLITE_ENABLE_LOCKING_STYLE + if( handleNFSUnlock ){ + int tErrno; /* Error code from system call errors */ + off_t divSize = SHARED_SIZE - 1; + + lock.l_type = F_UNLCK; + lock.l_whence = SEEK_SET; + lock.l_start = SHARED_FIRST; + lock.l_len = divSize; + if( unixFileLock(pFile, &lock)==(-1) ){ + tErrno = errno; + rc = SQLITE_IOERR_UNLOCK; + storeLastErrno(pFile, tErrno); + goto end_unlock; + } + lock.l_type = F_RDLCK; + lock.l_whence = SEEK_SET; + lock.l_start = SHARED_FIRST; + lock.l_len = divSize; + if( unixFileLock(pFile, &lock)==(-1) ){ + tErrno = errno; + rc = sqliteErrorFromPosixError(tErrno, SQLITE_IOERR_RDLOCK); + if( IS_LOCK_ERROR(rc) ){ + storeLastErrno(pFile, tErrno); + } + goto end_unlock; + } + lock.l_type = F_UNLCK; + lock.l_whence = SEEK_SET; + lock.l_start = SHARED_FIRST+divSize; + lock.l_len = SHARED_SIZE-divSize; + if( unixFileLock(pFile, &lock)==(-1) ){ + tErrno = errno; + rc = SQLITE_IOERR_UNLOCK; + storeLastErrno(pFile, tErrno); + goto end_unlock; + } + }else +#endif /* defined(__APPLE__) && SQLITE_ENABLE_LOCKING_STYLE */ + { + lock.l_type = F_RDLCK; + lock.l_whence = SEEK_SET; + lock.l_start = SHARED_FIRST; + lock.l_len = SHARED_SIZE; + if( unixFileLock(pFile, &lock) ){ + /* In theory, the call to unixFileLock() cannot fail because another + ** process is holding an incompatible lock. If it does, this + ** indicates that the other process is not following the locking + ** protocol. If this happens, return SQLITE_IOERR_RDLOCK. Returning + ** SQLITE_BUSY would confuse the upper layer (in practice it causes + ** an assert to fail). */ + rc = SQLITE_IOERR_RDLOCK; + storeLastErrno(pFile, errno); + goto end_unlock; + } + } + } + lock.l_type = F_UNLCK; + lock.l_whence = SEEK_SET; + lock.l_start = PENDING_BYTE; + lock.l_len = 2L; assert( PENDING_BYTE+1==RESERVED_BYTE ); + if( unixFileLock(pFile, &lock)==0 ){ + pInode->eFileLock = SHARED_LOCK; + }else{ + rc = SQLITE_IOERR_UNLOCK; + storeLastErrno(pFile, errno); + goto end_unlock; + } + } + if( eFileLock==NO_LOCK ){ + /* Decrement the shared lock counter. Release the lock using an + ** OS call only when all threads in this same process have released + ** the lock. + */ + pInode->nShared--; + if( pInode->nShared==0 ){ + lock.l_type = F_UNLCK; + lock.l_whence = SEEK_SET; + lock.l_start = lock.l_len = 0L; + if( unixFileLock(pFile, &lock)==0 ){ + pInode->eFileLock = NO_LOCK; + }else{ + rc = SQLITE_IOERR_UNLOCK; + storeLastErrno(pFile, errno); + pInode->eFileLock = NO_LOCK; + pFile->eFileLock = NO_LOCK; + } + } + + /* Decrement the count of locks against this same file. When the + ** count reaches zero, close any other file descriptors whose close + ** was deferred because of outstanding locks. + */ + pInode->nLock--; + assert( pInode->nLock>=0 ); + if( pInode->nLock==0 ) closePendingFds(pFile); + } + +end_unlock: + sqlite3_mutex_leave(pInode->pLockMutex); + if( rc==SQLITE_OK ){ + pFile->eFileLock = eFileLock; + } + return rc; +} + +/* +** Lower the locking level on file descriptor pFile to eFileLock. eFileLock +** must be either NO_LOCK or SHARED_LOCK. +** +** If the locking level of the file descriptor is already at or below +** the requested locking level, this routine is a no-op. +*/ +static int unixUnlock(sqlite3_file *id, int eFileLock){ +#if SQLITE_MAX_MMAP_SIZE>0 + assert( eFileLock==SHARED_LOCK || ((unixFile *)id)->nFetchOut==0 ); +#endif + return posixUnlock(id, eFileLock, 0); +} + +#if SQLITE_MAX_MMAP_SIZE>0 +static int unixMapfile(unixFile *pFd, i64 nByte); +static void unixUnmapfile(unixFile *pFd); +#endif + +/* +** This function performs the parts of the "close file" operation +** common to all locking schemes. It closes the directory and file +** handles, if they are valid, and sets all fields of the unixFile +** structure to 0. +** +** It is *not* necessary to hold the mutex when this routine is called, +** even on VxWorks. A mutex will be acquired on VxWorks by the +** vxworksReleaseFileId() routine. +*/ +static int closeUnixFile(sqlite3_file *id){ + unixFile *pFile = (unixFile*)id; +#if SQLITE_MAX_MMAP_SIZE>0 + unixUnmapfile(pFile); +#endif + if( pFile->h>=0 ){ + robust_close(pFile, pFile->h, __LINE__); + pFile->h = -1; + } +#if OS_VXWORKS + if( pFile->pId ){ + if( pFile->ctrlFlags & UNIXFILE_DELETE ){ + osUnlink(pFile->pId->zCanonicalName); + } + vxworksReleaseFileId(pFile->pId); + pFile->pId = 0; + } +#endif +#ifdef SQLITE_UNLINK_AFTER_CLOSE + if( pFile->ctrlFlags & UNIXFILE_DELETE ){ + osUnlink(pFile->zPath); + sqlite3_free(*(char**)&pFile->zPath); + pFile->zPath = 0; + } +#endif + OSTRACE(("CLOSE %-3d\n", pFile->h)); + OpenCounter(-1); + sqlite3_free(pFile->pPreallocatedUnused); + memset(pFile, 0, sizeof(unixFile)); + return SQLITE_OK; +} + +/* +** Close a file. +*/ +static int unixClose(sqlite3_file *id){ + int rc = SQLITE_OK; + unixFile *pFile = (unixFile *)id; + unixInodeInfo *pInode = pFile->pInode; + + assert( pInode!=0 ); + verifyDbFile(pFile); + unixUnlock(id, NO_LOCK); + assert( unixFileMutexNotheld(pFile) ); + unixEnterMutex(); + + /* unixFile.pInode is always valid here. Otherwise, a different close + ** routine (e.g. nolockClose()) would be called instead. + */ + assert( pFile->pInode->nLock>0 || pFile->pInode->bProcessLock==0 ); + sqlite3_mutex_enter(pInode->pLockMutex); + if( pInode->nLock ){ + /* If there are outstanding locks, do not actually close the file just + ** yet because that would clear those locks. Instead, add the file + ** descriptor to pInode->pUnused list. It will be automatically closed + ** when the last lock is cleared. + */ + setPendingFd(pFile); + } + sqlite3_mutex_leave(pInode->pLockMutex); + releaseInodeInfo(pFile); + rc = closeUnixFile(id); + unixLeaveMutex(); + return rc; +} + +/************** End of the posix advisory lock implementation ***************** +******************************************************************************/ + +/****************************************************************************** +****************************** No-op Locking ********************************** +** +** Of the various locking implementations available, this is by far the +** simplest: locking is ignored. No attempt is made to lock the database +** file for reading or writing. +** +** This locking mode is appropriate for use on read-only databases +** (ex: databases that are burned into CD-ROM, for example.) It can +** also be used if the application employs some external mechanism to +** prevent simultaneous access of the same database by two or more +** database connections. But there is a serious risk of database +** corruption if this locking mode is used in situations where multiple +** database connections are accessing the same database file at the same +** time and one or more of those connections are writing. +*/ + +static int nolockCheckReservedLock(sqlite3_file *NotUsed, int *pResOut){ + UNUSED_PARAMETER(NotUsed); + *pResOut = 0; + return SQLITE_OK; +} +static int nolockLock(sqlite3_file *NotUsed, int NotUsed2){ + UNUSED_PARAMETER2(NotUsed, NotUsed2); + return SQLITE_OK; +} +static int nolockUnlock(sqlite3_file *NotUsed, int NotUsed2){ + UNUSED_PARAMETER2(NotUsed, NotUsed2); + return SQLITE_OK; +} + +/* +** Close the file. +*/ +static int nolockClose(sqlite3_file *id) { + return closeUnixFile(id); +} + +/******************* End of the no-op lock implementation ********************* +******************************************************************************/ + +/****************************************************************************** +************************* Begin dot-file Locking ****************************** +** +** The dotfile locking implementation uses the existence of separate lock +** files (really a directory) to control access to the database. This works +** on just about every filesystem imaginable. But there are serious downsides: +** +** (1) There is zero concurrency. A single reader blocks all other +** connections from reading or writing the database. +** +** (2) An application crash or power loss can leave stale lock files +** sitting around that need to be cleared manually. +** +** Nevertheless, a dotlock is an appropriate locking mode for use if no +** other locking strategy is available. +** +** Dotfile locking works by creating a subdirectory in the same directory as +** the database and with the same name but with a ".lock" extension added. +** The existence of a lock directory implies an EXCLUSIVE lock. All other +** lock types (SHARED, RESERVED, PENDING) are mapped into EXCLUSIVE. +*/ + +/* +** The file suffix added to the data base filename in order to create the +** lock directory. +*/ +#define DOTLOCK_SUFFIX ".lock" + +/* +** This routine checks if there is a RESERVED lock held on the specified +** file by this or any other process. If such a lock is held, set *pResOut +** to a non-zero value otherwise *pResOut is set to zero. The return value +** is set to SQLITE_OK unless an I/O error occurs during lock checking. +** +** In dotfile locking, either a lock exists or it does not. So in this +** variation of CheckReservedLock(), *pResOut is set to true if any lock +** is held on the file and false if the file is unlocked. +*/ +static int dotlockCheckReservedLock(sqlite3_file *id, int *pResOut) { + int rc = SQLITE_OK; + int reserved = 0; + unixFile *pFile = (unixFile*)id; + + SimulateIOError( return SQLITE_IOERR_CHECKRESERVEDLOCK; ); + + assert( pFile ); + reserved = osAccess((const char*)pFile->lockingContext, 0)==0; + OSTRACE(("TEST WR-LOCK %d %d %d (dotlock)\n", pFile->h, rc, reserved)); + *pResOut = reserved; + return rc; +} + +/* +** Lock the file with the lock specified by parameter eFileLock - one +** of the following: +** +** (1) SHARED_LOCK +** (2) RESERVED_LOCK +** (3) PENDING_LOCK +** (4) EXCLUSIVE_LOCK +** +** Sometimes when requesting one lock state, additional lock states +** are inserted in between. The locking might fail on one of the later +** transitions leaving the lock state different from what it started but +** still short of its goal. The following chart shows the allowed +** transitions and the inserted intermediate states: +** +** UNLOCKED -> SHARED +** SHARED -> RESERVED +** SHARED -> (PENDING) -> EXCLUSIVE +** RESERVED -> (PENDING) -> EXCLUSIVE +** PENDING -> EXCLUSIVE +** +** This routine will only increase a lock. Use the sqlite3OsUnlock() +** routine to lower a locking level. +** +** With dotfile locking, we really only support state (4): EXCLUSIVE. +** But we track the other locking levels internally. +*/ +static int dotlockLock(sqlite3_file *id, int eFileLock) { + unixFile *pFile = (unixFile*)id; + char *zLockFile = (char *)pFile->lockingContext; + int rc = SQLITE_OK; + + + /* If we have any lock, then the lock file already exists. All we have + ** to do is adjust our internal record of the lock level. + */ + if( pFile->eFileLock > NO_LOCK ){ + pFile->eFileLock = eFileLock; + /* Always update the timestamp on the old file */ +#ifdef HAVE_UTIME + utime(zLockFile, NULL); +#else + utimes(zLockFile, NULL); +#endif + return SQLITE_OK; + } + + /* grab an exclusive lock */ + rc = osMkdir(zLockFile, 0777); + if( rc<0 ){ + /* failed to open/create the lock directory */ + int tErrno = errno; + if( EEXIST == tErrno ){ + rc = SQLITE_BUSY; + } else { + rc = sqliteErrorFromPosixError(tErrno, SQLITE_IOERR_LOCK); + if( rc!=SQLITE_BUSY ){ + storeLastErrno(pFile, tErrno); + } + } + return rc; + } + + /* got it, set the type and return ok */ + pFile->eFileLock = eFileLock; + return rc; +} + +/* +** Lower the locking level on file descriptor pFile to eFileLock. eFileLock +** must be either NO_LOCK or SHARED_LOCK. +** +** If the locking level of the file descriptor is already at or below +** the requested locking level, this routine is a no-op. +** +** When the locking level reaches NO_LOCK, delete the lock file. +*/ +static int dotlockUnlock(sqlite3_file *id, int eFileLock) { + unixFile *pFile = (unixFile*)id; + char *zLockFile = (char *)pFile->lockingContext; + int rc; + + assert( pFile ); + OSTRACE(("UNLOCK %d %d was %d pid=%d (dotlock)\n", pFile->h, eFileLock, + pFile->eFileLock, osGetpid(0))); + assert( eFileLock<=SHARED_LOCK ); + + /* no-op if possible */ + if( pFile->eFileLock==eFileLock ){ + return SQLITE_OK; + } + + /* To downgrade to shared, simply update our internal notion of the + ** lock state. No need to mess with the file on disk. + */ + if( eFileLock==SHARED_LOCK ){ + pFile->eFileLock = SHARED_LOCK; + return SQLITE_OK; + } + + /* To fully unlock the database, delete the lock file */ + assert( eFileLock==NO_LOCK ); + rc = osRmdir(zLockFile); + if( rc<0 ){ + int tErrno = errno; + if( tErrno==ENOENT ){ + rc = SQLITE_OK; + }else{ + rc = SQLITE_IOERR_UNLOCK; + storeLastErrno(pFile, tErrno); + } + return rc; + } + pFile->eFileLock = NO_LOCK; + return SQLITE_OK; +} + +/* +** Close a file. Make sure the lock has been released before closing. +*/ +static int dotlockClose(sqlite3_file *id) { + unixFile *pFile = (unixFile*)id; + assert( id!=0 ); + dotlockUnlock(id, NO_LOCK); + sqlite3_free(pFile->lockingContext); + return closeUnixFile(id); +} +/****************** End of the dot-file lock implementation ******************* +******************************************************************************/ + +/****************************************************************************** +************************** Begin flock Locking ******************************** +** +** Use the flock() system call to do file locking. +** +** flock() locking is like dot-file locking in that the various +** fine-grain locking levels supported by SQLite are collapsed into +** a single exclusive lock. In other words, SHARED, RESERVED, and +** PENDING locks are the same thing as an EXCLUSIVE lock. SQLite +** still works when you do this, but concurrency is reduced since +** only a single process can be reading the database at a time. +** +** Omit this section if SQLITE_ENABLE_LOCKING_STYLE is turned off +*/ +#if SQLITE_ENABLE_LOCKING_STYLE + +/* +** Retry flock() calls that fail with EINTR +*/ +#ifdef EINTR +static int robust_flock(int fd, int op){ + int rc; + do{ rc = flock(fd,op); }while( rc<0 && errno==EINTR ); + return rc; +} +#else +# define robust_flock(a,b) flock(a,b) +#endif + + +/* +** This routine checks if there is a RESERVED lock held on the specified +** file by this or any other process. If such a lock is held, set *pResOut +** to a non-zero value otherwise *pResOut is set to zero. The return value +** is set to SQLITE_OK unless an I/O error occurs during lock checking. +*/ +static int flockCheckReservedLock(sqlite3_file *id, int *pResOut){ + int rc = SQLITE_OK; + int reserved = 0; + unixFile *pFile = (unixFile*)id; + + SimulateIOError( return SQLITE_IOERR_CHECKRESERVEDLOCK; ); + + assert( pFile ); + + /* Check if a thread in this process holds such a lock */ + if( pFile->eFileLock>SHARED_LOCK ){ + reserved = 1; + } + + /* Otherwise see if some other process holds it. */ + if( !reserved ){ + /* attempt to get the lock */ + int lrc = robust_flock(pFile->h, LOCK_EX | LOCK_NB); + if( !lrc ){ + /* got the lock, unlock it */ + lrc = robust_flock(pFile->h, LOCK_UN); + if ( lrc ) { + int tErrno = errno; + /* unlock failed with an error */ + lrc = SQLITE_IOERR_UNLOCK; + storeLastErrno(pFile, tErrno); + rc = lrc; + } + } else { + int tErrno = errno; + reserved = 1; + /* someone else might have it reserved */ + lrc = sqliteErrorFromPosixError(tErrno, SQLITE_IOERR_LOCK); + if( IS_LOCK_ERROR(lrc) ){ + storeLastErrno(pFile, tErrno); + rc = lrc; + } + } + } + OSTRACE(("TEST WR-LOCK %d %d %d (flock)\n", pFile->h, rc, reserved)); + +#ifdef SQLITE_IGNORE_FLOCK_LOCK_ERRORS + if( (rc & 0xff) == SQLITE_IOERR ){ + rc = SQLITE_OK; + reserved=1; + } +#endif /* SQLITE_IGNORE_FLOCK_LOCK_ERRORS */ + *pResOut = reserved; + return rc; +} + +/* +** Lock the file with the lock specified by parameter eFileLock - one +** of the following: +** +** (1) SHARED_LOCK +** (2) RESERVED_LOCK +** (3) PENDING_LOCK +** (4) EXCLUSIVE_LOCK +** +** Sometimes when requesting one lock state, additional lock states +** are inserted in between. The locking might fail on one of the later +** transitions leaving the lock state different from what it started but +** still short of its goal. The following chart shows the allowed +** transitions and the inserted intermediate states: +** +** UNLOCKED -> SHARED +** SHARED -> RESERVED +** SHARED -> (PENDING) -> EXCLUSIVE +** RESERVED -> (PENDING) -> EXCLUSIVE +** PENDING -> EXCLUSIVE +** +** flock() only really support EXCLUSIVE locks. We track intermediate +** lock states in the sqlite3_file structure, but all locks SHARED or +** above are really EXCLUSIVE locks and exclude all other processes from +** access the file. +** +** This routine will only increase a lock. Use the sqlite3OsUnlock() +** routine to lower a locking level. +*/ +static int flockLock(sqlite3_file *id, int eFileLock) { + int rc = SQLITE_OK; + unixFile *pFile = (unixFile*)id; + + assert( pFile ); + + /* if we already have a lock, it is exclusive. + ** Just adjust level and punt on outta here. */ + if (pFile->eFileLock > NO_LOCK) { + pFile->eFileLock = eFileLock; + return SQLITE_OK; + } + + /* grab an exclusive lock */ + + if (robust_flock(pFile->h, LOCK_EX | LOCK_NB)) { + int tErrno = errno; + /* didn't get, must be busy */ + rc = sqliteErrorFromPosixError(tErrno, SQLITE_IOERR_LOCK); + if( IS_LOCK_ERROR(rc) ){ + storeLastErrno(pFile, tErrno); + } + } else { + /* got it, set the type and return ok */ + pFile->eFileLock = eFileLock; + } + OSTRACE(("LOCK %d %s %s (flock)\n", pFile->h, azFileLock(eFileLock), + rc==SQLITE_OK ? "ok" : "failed")); +#ifdef SQLITE_IGNORE_FLOCK_LOCK_ERRORS + if( (rc & 0xff) == SQLITE_IOERR ){ + rc = SQLITE_BUSY; + } +#endif /* SQLITE_IGNORE_FLOCK_LOCK_ERRORS */ + return rc; +} + + +/* +** Lower the locking level on file descriptor pFile to eFileLock. eFileLock +** must be either NO_LOCK or SHARED_LOCK. +** +** If the locking level of the file descriptor is already at or below +** the requested locking level, this routine is a no-op. +*/ +static int flockUnlock(sqlite3_file *id, int eFileLock) { + unixFile *pFile = (unixFile*)id; + + assert( pFile ); + OSTRACE(("UNLOCK %d %d was %d pid=%d (flock)\n", pFile->h, eFileLock, + pFile->eFileLock, osGetpid(0))); + assert( eFileLock<=SHARED_LOCK ); + + /* no-op if possible */ + if( pFile->eFileLock==eFileLock ){ + return SQLITE_OK; + } + + /* shared can just be set because we always have an exclusive */ + if (eFileLock==SHARED_LOCK) { + pFile->eFileLock = eFileLock; + return SQLITE_OK; + } + + /* no, really, unlock. */ + if( robust_flock(pFile->h, LOCK_UN) ){ +#ifdef SQLITE_IGNORE_FLOCK_LOCK_ERRORS + return SQLITE_OK; +#endif /* SQLITE_IGNORE_FLOCK_LOCK_ERRORS */ + return SQLITE_IOERR_UNLOCK; + }else{ + pFile->eFileLock = NO_LOCK; + return SQLITE_OK; + } +} + +/* +** Close a file. +*/ +static int flockClose(sqlite3_file *id) { + assert( id!=0 ); + flockUnlock(id, NO_LOCK); + return closeUnixFile(id); +} + +#endif /* SQLITE_ENABLE_LOCKING_STYLE && !OS_VXWORK */ + +/******************* End of the flock lock implementation ********************* +******************************************************************************/ + +/****************************************************************************** +************************ Begin Named Semaphore Locking ************************ +** +** Named semaphore locking is only supported on VxWorks. +** +** Semaphore locking is like dot-lock and flock in that it really only +** supports EXCLUSIVE locking. Only a single process can read or write +** the database file at a time. This reduces potential concurrency, but +** makes the lock implementation much easier. +*/ +#if OS_VXWORKS + +/* +** This routine checks if there is a RESERVED lock held on the specified +** file by this or any other process. If such a lock is held, set *pResOut +** to a non-zero value otherwise *pResOut is set to zero. The return value +** is set to SQLITE_OK unless an I/O error occurs during lock checking. +*/ +static int semXCheckReservedLock(sqlite3_file *id, int *pResOut) { + int rc = SQLITE_OK; + int reserved = 0; + unixFile *pFile = (unixFile*)id; + + SimulateIOError( return SQLITE_IOERR_CHECKRESERVEDLOCK; ); + + assert( pFile ); + + /* Check if a thread in this process holds such a lock */ + if( pFile->eFileLock>SHARED_LOCK ){ + reserved = 1; + } + + /* Otherwise see if some other process holds it. */ + if( !reserved ){ + sem_t *pSem = pFile->pInode->pSem; + + if( sem_trywait(pSem)==-1 ){ + int tErrno = errno; + if( EAGAIN != tErrno ){ + rc = sqliteErrorFromPosixError(tErrno, SQLITE_IOERR_CHECKRESERVEDLOCK); + storeLastErrno(pFile, tErrno); + } else { + /* someone else has the lock when we are in NO_LOCK */ + reserved = (pFile->eFileLock < SHARED_LOCK); + } + }else{ + /* we could have it if we want it */ + sem_post(pSem); + } + } + OSTRACE(("TEST WR-LOCK %d %d %d (sem)\n", pFile->h, rc, reserved)); + + *pResOut = reserved; + return rc; +} + +/* +** Lock the file with the lock specified by parameter eFileLock - one +** of the following: +** +** (1) SHARED_LOCK +** (2) RESERVED_LOCK +** (3) PENDING_LOCK +** (4) EXCLUSIVE_LOCK +** +** Sometimes when requesting one lock state, additional lock states +** are inserted in between. The locking might fail on one of the later +** transitions leaving the lock state different from what it started but +** still short of its goal. The following chart shows the allowed +** transitions and the inserted intermediate states: +** +** UNLOCKED -> SHARED +** SHARED -> RESERVED +** SHARED -> (PENDING) -> EXCLUSIVE +** RESERVED -> (PENDING) -> EXCLUSIVE +** PENDING -> EXCLUSIVE +** +** Semaphore locks only really support EXCLUSIVE locks. We track intermediate +** lock states in the sqlite3_file structure, but all locks SHARED or +** above are really EXCLUSIVE locks and exclude all other processes from +** access the file. +** +** This routine will only increase a lock. Use the sqlite3OsUnlock() +** routine to lower a locking level. +*/ +static int semXLock(sqlite3_file *id, int eFileLock) { + unixFile *pFile = (unixFile*)id; + sem_t *pSem = pFile->pInode->pSem; + int rc = SQLITE_OK; + + /* if we already have a lock, it is exclusive. + ** Just adjust level and punt on outta here. */ + if (pFile->eFileLock > NO_LOCK) { + pFile->eFileLock = eFileLock; + rc = SQLITE_OK; + goto sem_end_lock; + } + + /* lock semaphore now but bail out when already locked. */ + if( sem_trywait(pSem)==-1 ){ + rc = SQLITE_BUSY; + goto sem_end_lock; + } + + /* got it, set the type and return ok */ + pFile->eFileLock = eFileLock; + + sem_end_lock: + return rc; +} + +/* +** Lower the locking level on file descriptor pFile to eFileLock. eFileLock +** must be either NO_LOCK or SHARED_LOCK. +** +** If the locking level of the file descriptor is already at or below +** the requested locking level, this routine is a no-op. +*/ +static int semXUnlock(sqlite3_file *id, int eFileLock) { + unixFile *pFile = (unixFile*)id; + sem_t *pSem = pFile->pInode->pSem; + + assert( pFile ); + assert( pSem ); + OSTRACE(("UNLOCK %d %d was %d pid=%d (sem)\n", pFile->h, eFileLock, + pFile->eFileLock, osGetpid(0))); + assert( eFileLock<=SHARED_LOCK ); + + /* no-op if possible */ + if( pFile->eFileLock==eFileLock ){ + return SQLITE_OK; + } + + /* shared can just be set because we always have an exclusive */ + if (eFileLock==SHARED_LOCK) { + pFile->eFileLock = eFileLock; + return SQLITE_OK; + } + + /* no, really unlock. */ + if ( sem_post(pSem)==-1 ) { + int rc, tErrno = errno; + rc = sqliteErrorFromPosixError(tErrno, SQLITE_IOERR_UNLOCK); + if( IS_LOCK_ERROR(rc) ){ + storeLastErrno(pFile, tErrno); + } + return rc; + } + pFile->eFileLock = NO_LOCK; + return SQLITE_OK; +} + +/* + ** Close a file. + */ +static int semXClose(sqlite3_file *id) { + if( id ){ + unixFile *pFile = (unixFile*)id; + semXUnlock(id, NO_LOCK); + assert( pFile ); + assert( unixFileMutexNotheld(pFile) ); + unixEnterMutex(); + releaseInodeInfo(pFile); + unixLeaveMutex(); + closeUnixFile(id); + } + return SQLITE_OK; +} + +#endif /* OS_VXWORKS */ +/* +** Named semaphore locking is only available on VxWorks. +** +*************** End of the named semaphore lock implementation **************** +******************************************************************************/ + + +/****************************************************************************** +*************************** Begin AFP Locking ********************************* +** +** AFP is the Apple Filing Protocol. AFP is a network filesystem found +** on Apple Macintosh computers - both OS9 and OSX. +** +** Third-party implementations of AFP are available. But this code here +** only works on OSX. +*/ + +#if defined(__APPLE__) && SQLITE_ENABLE_LOCKING_STYLE +/* +** The afpLockingContext structure contains all afp lock specific state +*/ +typedef struct afpLockingContext afpLockingContext; +struct afpLockingContext { + int reserved; + const char *dbPath; /* Name of the open file */ +}; + +struct ByteRangeLockPB2 +{ + unsigned long long offset; /* offset to first byte to lock */ + unsigned long long length; /* nbr of bytes to lock */ + unsigned long long retRangeStart; /* nbr of 1st byte locked if successful */ + unsigned char unLockFlag; /* 1 = unlock, 0 = lock */ + unsigned char startEndFlag; /* 1=rel to end of fork, 0=rel to start */ + int fd; /* file desc to assoc this lock with */ +}; + +#define afpfsByteRangeLock2FSCTL _IOWR('z', 23, struct ByteRangeLockPB2) + +/* +** This is a utility for setting or clearing a bit-range lock on an +** AFP filesystem. +** +** Return SQLITE_OK on success, SQLITE_BUSY on failure. +*/ +static int afpSetLock( + const char *path, /* Name of the file to be locked or unlocked */ + unixFile *pFile, /* Open file descriptor on path */ + unsigned long long offset, /* First byte to be locked */ + unsigned long long length, /* Number of bytes to lock */ + int setLockFlag /* True to set lock. False to clear lock */ +){ + struct ByteRangeLockPB2 pb; + int err; + + pb.unLockFlag = setLockFlag ? 0 : 1; + pb.startEndFlag = 0; + pb.offset = offset; + pb.length = length; + pb.fd = pFile->h; + + OSTRACE(("AFPSETLOCK [%s] for %d%s in range %llx:%llx\n", + (setLockFlag?"ON":"OFF"), pFile->h, (pb.fd==-1?"[testval-1]":""), + offset, length)); + err = fsctl(path, afpfsByteRangeLock2FSCTL, &pb, 0); + if ( err==-1 ) { + int rc; + int tErrno = errno; + OSTRACE(("AFPSETLOCK failed to fsctl() '%s' %d %s\n", + path, tErrno, strerror(tErrno))); +#ifdef SQLITE_IGNORE_AFP_LOCK_ERRORS + rc = SQLITE_BUSY; +#else + rc = sqliteErrorFromPosixError(tErrno, + setLockFlag ? SQLITE_IOERR_LOCK : SQLITE_IOERR_UNLOCK); +#endif /* SQLITE_IGNORE_AFP_LOCK_ERRORS */ + if( IS_LOCK_ERROR(rc) ){ + storeLastErrno(pFile, tErrno); + } + return rc; + } else { + return SQLITE_OK; + } +} + +/* +** This routine checks if there is a RESERVED lock held on the specified +** file by this or any other process. If such a lock is held, set *pResOut +** to a non-zero value otherwise *pResOut is set to zero. The return value +** is set to SQLITE_OK unless an I/O error occurs during lock checking. +*/ +static int afpCheckReservedLock(sqlite3_file *id, int *pResOut){ + int rc = SQLITE_OK; + int reserved = 0; + unixFile *pFile = (unixFile*)id; + afpLockingContext *context; + + SimulateIOError( return SQLITE_IOERR_CHECKRESERVEDLOCK; ); + + assert( pFile ); + context = (afpLockingContext *) pFile->lockingContext; + if( context->reserved ){ + *pResOut = 1; + return SQLITE_OK; + } + sqlite3_mutex_enter(pFile->pInode->pLockMutex); + /* Check if a thread in this process holds such a lock */ + if( pFile->pInode->eFileLock>SHARED_LOCK ){ + reserved = 1; + } + + /* Otherwise see if some other process holds it. + */ + if( !reserved ){ + /* lock the RESERVED byte */ + int lrc = afpSetLock(context->dbPath, pFile, RESERVED_BYTE, 1,1); + if( SQLITE_OK==lrc ){ + /* if we succeeded in taking the reserved lock, unlock it to restore + ** the original state */ + lrc = afpSetLock(context->dbPath, pFile, RESERVED_BYTE, 1, 0); + } else { + /* if we failed to get the lock then someone else must have it */ + reserved = 1; + } + if( IS_LOCK_ERROR(lrc) ){ + rc=lrc; + } + } + + sqlite3_mutex_leave(pFile->pInode->pLockMutex); + OSTRACE(("TEST WR-LOCK %d %d %d (afp)\n", pFile->h, rc, reserved)); + + *pResOut = reserved; + return rc; +} + +/* +** Lock the file with the lock specified by parameter eFileLock - one +** of the following: +** +** (1) SHARED_LOCK +** (2) RESERVED_LOCK +** (3) PENDING_LOCK +** (4) EXCLUSIVE_LOCK +** +** Sometimes when requesting one lock state, additional lock states +** are inserted in between. The locking might fail on one of the later +** transitions leaving the lock state different from what it started but +** still short of its goal. The following chart shows the allowed +** transitions and the inserted intermediate states: +** +** UNLOCKED -> SHARED +** SHARED -> RESERVED +** SHARED -> (PENDING) -> EXCLUSIVE +** RESERVED -> (PENDING) -> EXCLUSIVE +** PENDING -> EXCLUSIVE +** +** This routine will only increase a lock. Use the sqlite3OsUnlock() +** routine to lower a locking level. +*/ +static int afpLock(sqlite3_file *id, int eFileLock){ + int rc = SQLITE_OK; + unixFile *pFile = (unixFile*)id; + unixInodeInfo *pInode = pFile->pInode; + afpLockingContext *context = (afpLockingContext *) pFile->lockingContext; + + assert( pFile ); + OSTRACE(("LOCK %d %s was %s(%s,%d) pid=%d (afp)\n", pFile->h, + azFileLock(eFileLock), azFileLock(pFile->eFileLock), + azFileLock(pInode->eFileLock), pInode->nShared , osGetpid(0))); + + /* If there is already a lock of this type or more restrictive on the + ** unixFile, do nothing. Don't use the afp_end_lock: exit path, as + ** unixEnterMutex() hasn't been called yet. + */ + if( pFile->eFileLock>=eFileLock ){ + OSTRACE(("LOCK %d %s ok (already held) (afp)\n", pFile->h, + azFileLock(eFileLock))); + return SQLITE_OK; + } + + /* Make sure the locking sequence is correct + ** (1) We never move from unlocked to anything higher than shared lock. + ** (2) SQLite never explicitly requests a pendig lock. + ** (3) A shared lock is always held when a reserve lock is requested. + */ + assert( pFile->eFileLock!=NO_LOCK || eFileLock==SHARED_LOCK ); + assert( eFileLock!=PENDING_LOCK ); + assert( eFileLock!=RESERVED_LOCK || pFile->eFileLock==SHARED_LOCK ); + + /* This mutex is needed because pFile->pInode is shared across threads + */ + pInode = pFile->pInode; + sqlite3_mutex_enter(pInode->pLockMutex); + + /* If some thread using this PID has a lock via a different unixFile* + ** handle that precludes the requested lock, return BUSY. + */ + if( (pFile->eFileLock!=pInode->eFileLock && + (pInode->eFileLock>=PENDING_LOCK || eFileLock>SHARED_LOCK)) + ){ + rc = SQLITE_BUSY; + goto afp_end_lock; + } + + /* If a SHARED lock is requested, and some thread using this PID already + ** has a SHARED or RESERVED lock, then increment reference counts and + ** return SQLITE_OK. + */ + if( eFileLock==SHARED_LOCK && + (pInode->eFileLock==SHARED_LOCK || pInode->eFileLock==RESERVED_LOCK) ){ + assert( eFileLock==SHARED_LOCK ); + assert( pFile->eFileLock==0 ); + assert( pInode->nShared>0 ); + pFile->eFileLock = SHARED_LOCK; + pInode->nShared++; + pInode->nLock++; + goto afp_end_lock; + } + + /* A PENDING lock is needed before acquiring a SHARED lock and before + ** acquiring an EXCLUSIVE lock. For the SHARED lock, the PENDING will + ** be released. + */ + if( eFileLock==SHARED_LOCK + || (eFileLock==EXCLUSIVE_LOCK && pFile->eFileLockdbPath, pFile, PENDING_BYTE, 1, 1); + if (failed) { + rc = failed; + goto afp_end_lock; + } + } + + /* If control gets to this point, then actually go ahead and make + ** operating system calls for the specified lock. + */ + if( eFileLock==SHARED_LOCK ){ + int lrc1, lrc2, lrc1Errno = 0; + long lk, mask; + + assert( pInode->nShared==0 ); + assert( pInode->eFileLock==0 ); + + mask = (sizeof(long)==8) ? LARGEST_INT64 : 0x7fffffff; + /* Now get the read-lock SHARED_LOCK */ + /* note that the quality of the randomness doesn't matter that much */ + lk = random(); + pInode->sharedByte = (lk & mask)%(SHARED_SIZE - 1); + lrc1 = afpSetLock(context->dbPath, pFile, + SHARED_FIRST+pInode->sharedByte, 1, 1); + if( IS_LOCK_ERROR(lrc1) ){ + lrc1Errno = pFile->lastErrno; + } + /* Drop the temporary PENDING lock */ + lrc2 = afpSetLock(context->dbPath, pFile, PENDING_BYTE, 1, 0); + + if( IS_LOCK_ERROR(lrc1) ) { + storeLastErrno(pFile, lrc1Errno); + rc = lrc1; + goto afp_end_lock; + } else if( IS_LOCK_ERROR(lrc2) ){ + rc = lrc2; + goto afp_end_lock; + } else if( lrc1 != SQLITE_OK ) { + rc = lrc1; + } else { + pFile->eFileLock = SHARED_LOCK; + pInode->nLock++; + pInode->nShared = 1; + } + }else if( eFileLock==EXCLUSIVE_LOCK && pInode->nShared>1 ){ + /* We are trying for an exclusive lock but another thread in this + ** same process is still holding a shared lock. */ + rc = SQLITE_BUSY; + }else{ + /* The request was for a RESERVED or EXCLUSIVE lock. It is + ** assumed that there is a SHARED or greater lock on the file + ** already. + */ + int failed = 0; + assert( 0!=pFile->eFileLock ); + if (eFileLock >= RESERVED_LOCK && pFile->eFileLock < RESERVED_LOCK) { + /* Acquire a RESERVED lock */ + failed = afpSetLock(context->dbPath, pFile, RESERVED_BYTE, 1,1); + if( !failed ){ + context->reserved = 1; + } + } + if (!failed && eFileLock == EXCLUSIVE_LOCK) { + /* Acquire an EXCLUSIVE lock */ + + /* Remove the shared lock before trying the range. we'll need to + ** reestablish the shared lock if we can't get the afpUnlock + */ + if( !(failed = afpSetLock(context->dbPath, pFile, SHARED_FIRST + + pInode->sharedByte, 1, 0)) ){ + int failed2 = SQLITE_OK; + /* now attemmpt to get the exclusive lock range */ + failed = afpSetLock(context->dbPath, pFile, SHARED_FIRST, + SHARED_SIZE, 1); + if( failed && (failed2 = afpSetLock(context->dbPath, pFile, + SHARED_FIRST + pInode->sharedByte, 1, 1)) ){ + /* Can't reestablish the shared lock. Sqlite can't deal, this is + ** a critical I/O error + */ + rc = ((failed & 0xff) == SQLITE_IOERR) ? failed2 : + SQLITE_IOERR_LOCK; + goto afp_end_lock; + } + }else{ + rc = failed; + } + } + if( failed ){ + rc = failed; + } + } + + if( rc==SQLITE_OK ){ + pFile->eFileLock = eFileLock; + pInode->eFileLock = eFileLock; + }else if( eFileLock==EXCLUSIVE_LOCK ){ + pFile->eFileLock = PENDING_LOCK; + pInode->eFileLock = PENDING_LOCK; + } + +afp_end_lock: + sqlite3_mutex_leave(pInode->pLockMutex); + OSTRACE(("LOCK %d %s %s (afp)\n", pFile->h, azFileLock(eFileLock), + rc==SQLITE_OK ? "ok" : "failed")); + return rc; +} + +/* +** Lower the locking level on file descriptor pFile to eFileLock. eFileLock +** must be either NO_LOCK or SHARED_LOCK. +** +** If the locking level of the file descriptor is already at or below +** the requested locking level, this routine is a no-op. +*/ +static int afpUnlock(sqlite3_file *id, int eFileLock) { + int rc = SQLITE_OK; + unixFile *pFile = (unixFile*)id; + unixInodeInfo *pInode; + afpLockingContext *context = (afpLockingContext *) pFile->lockingContext; + int skipShared = 0; +#ifdef SQLITE_TEST + int h = pFile->h; +#endif + + assert( pFile ); + OSTRACE(("UNLOCK %d %d was %d(%d,%d) pid=%d (afp)\n", pFile->h, eFileLock, + pFile->eFileLock, pFile->pInode->eFileLock, pFile->pInode->nShared, + osGetpid(0))); + + assert( eFileLock<=SHARED_LOCK ); + if( pFile->eFileLock<=eFileLock ){ + return SQLITE_OK; + } + pInode = pFile->pInode; + sqlite3_mutex_enter(pInode->pLockMutex); + assert( pInode->nShared!=0 ); + if( pFile->eFileLock>SHARED_LOCK ){ + assert( pInode->eFileLock==pFile->eFileLock ); + SimulateIOErrorBenign(1); + SimulateIOError( h=(-1) ) + SimulateIOErrorBenign(0); + +#ifdef SQLITE_DEBUG + /* When reducing a lock such that other processes can start + ** reading the database file again, make sure that the + ** transaction counter was updated if any part of the database + ** file changed. If the transaction counter is not updated, + ** other connections to the same file might not realize that + ** the file has changed and hence might not know to flush their + ** cache. The use of a stale cache can lead to database corruption. + */ + assert( pFile->inNormalWrite==0 + || pFile->dbUpdate==0 + || pFile->transCntrChng==1 ); + pFile->inNormalWrite = 0; +#endif + + if( pFile->eFileLock==EXCLUSIVE_LOCK ){ + rc = afpSetLock(context->dbPath, pFile, SHARED_FIRST, SHARED_SIZE, 0); + if( rc==SQLITE_OK && (eFileLock==SHARED_LOCK || pInode->nShared>1) ){ + /* only re-establish the shared lock if necessary */ + int sharedLockByte = SHARED_FIRST+pInode->sharedByte; + rc = afpSetLock(context->dbPath, pFile, sharedLockByte, 1, 1); + } else { + skipShared = 1; + } + } + if( rc==SQLITE_OK && pFile->eFileLock>=PENDING_LOCK ){ + rc = afpSetLock(context->dbPath, pFile, PENDING_BYTE, 1, 0); + } + if( rc==SQLITE_OK && pFile->eFileLock>=RESERVED_LOCK && context->reserved ){ + rc = afpSetLock(context->dbPath, pFile, RESERVED_BYTE, 1, 0); + if( !rc ){ + context->reserved = 0; + } + } + if( rc==SQLITE_OK && (eFileLock==SHARED_LOCK || pInode->nShared>1)){ + pInode->eFileLock = SHARED_LOCK; + } + } + if( rc==SQLITE_OK && eFileLock==NO_LOCK ){ + + /* Decrement the shared lock counter. Release the lock using an + ** OS call only when all threads in this same process have released + ** the lock. + */ + unsigned long long sharedLockByte = SHARED_FIRST+pInode->sharedByte; + pInode->nShared--; + if( pInode->nShared==0 ){ + SimulateIOErrorBenign(1); + SimulateIOError( h=(-1) ) + SimulateIOErrorBenign(0); + if( !skipShared ){ + rc = afpSetLock(context->dbPath, pFile, sharedLockByte, 1, 0); + } + if( !rc ){ + pInode->eFileLock = NO_LOCK; + pFile->eFileLock = NO_LOCK; + } + } + if( rc==SQLITE_OK ){ + pInode->nLock--; + assert( pInode->nLock>=0 ); + if( pInode->nLock==0 ) closePendingFds(pFile); + } + } + + sqlite3_mutex_leave(pInode->pLockMutex); + if( rc==SQLITE_OK ){ + pFile->eFileLock = eFileLock; + } + return rc; +} + +/* +** Close a file & cleanup AFP specific locking context +*/ +static int afpClose(sqlite3_file *id) { + int rc = SQLITE_OK; + unixFile *pFile = (unixFile*)id; + assert( id!=0 ); + afpUnlock(id, NO_LOCK); + assert( unixFileMutexNotheld(pFile) ); + unixEnterMutex(); + if( pFile->pInode ){ + unixInodeInfo *pInode = pFile->pInode; + sqlite3_mutex_enter(pInode->pLockMutex); + if( pInode->nLock ){ + /* If there are outstanding locks, do not actually close the file just + ** yet because that would clear those locks. Instead, add the file + ** descriptor to pInode->aPending. It will be automatically closed when + ** the last lock is cleared. + */ + setPendingFd(pFile); + } + sqlite3_mutex_leave(pInode->pLockMutex); + } + releaseInodeInfo(pFile); + sqlite3_free(pFile->lockingContext); + rc = closeUnixFile(id); + unixLeaveMutex(); + return rc; +} + +#endif /* defined(__APPLE__) && SQLITE_ENABLE_LOCKING_STYLE */ +/* +** The code above is the AFP lock implementation. The code is specific +** to MacOSX and does not work on other unix platforms. No alternative +** is available. If you don't compile for a mac, then the "unix-afp" +** VFS is not available. +** +********************* End of the AFP lock implementation ********************** +******************************************************************************/ + +/****************************************************************************** +*************************** Begin NFS Locking ********************************/ + +#if defined(__APPLE__) && SQLITE_ENABLE_LOCKING_STYLE +/* + ** Lower the locking level on file descriptor pFile to eFileLock. eFileLock + ** must be either NO_LOCK or SHARED_LOCK. + ** + ** If the locking level of the file descriptor is already at or below + ** the requested locking level, this routine is a no-op. + */ +static int nfsUnlock(sqlite3_file *id, int eFileLock){ + return posixUnlock(id, eFileLock, 1); +} + +#endif /* defined(__APPLE__) && SQLITE_ENABLE_LOCKING_STYLE */ +/* +** The code above is the NFS lock implementation. The code is specific +** to MacOSX and does not work on other unix platforms. No alternative +** is available. +** +********************* End of the NFS lock implementation ********************** +******************************************************************************/ + +/****************************************************************************** +**************** Non-locking sqlite3_file methods ***************************** +** +** The next division contains implementations for all methods of the +** sqlite3_file object other than the locking methods. The locking +** methods were defined in divisions above (one locking method per +** division). Those methods that are common to all locking modes +** are gather together into this division. +*/ + +/* +** Seek to the offset passed as the second argument, then read cnt +** bytes into pBuf. Return the number of bytes actually read. +** +** NB: If you define USE_PREAD or USE_PREAD64, then it might also +** be necessary to define _XOPEN_SOURCE to be 500. This varies from +** one system to another. Since SQLite does not define USE_PREAD +** in any form by default, we will not attempt to define _XOPEN_SOURCE. +** See tickets #2741 and #2681. +** +** To avoid stomping the errno value on a failed read the lastErrno value +** is set before returning. +*/ +static int seekAndRead(unixFile *id, sqlite3_int64 offset, void *pBuf, int cnt){ + int got; + int prior = 0; +#if (!defined(USE_PREAD) && !defined(USE_PREAD64)) + i64 newOffset; +#endif + TIMER_START; + assert( cnt==(cnt&0x1ffff) ); + assert( id->h>2 ); + do{ +#if defined(USE_PREAD) + got = osPread(id->h, pBuf, cnt, offset); + SimulateIOError( got = -1 ); +#elif defined(USE_PREAD64) + got = osPread64(id->h, pBuf, cnt, offset); + SimulateIOError( got = -1 ); +#else + newOffset = lseek(id->h, offset, SEEK_SET); + SimulateIOError( newOffset = -1 ); + if( newOffset<0 ){ + storeLastErrno((unixFile*)id, errno); + return -1; + } + got = osRead(id->h, pBuf, cnt); +#endif + if( got==cnt ) break; + if( got<0 ){ + if( errno==EINTR ){ got = 1; continue; } + prior = 0; + storeLastErrno((unixFile*)id, errno); + break; + }else if( got>0 ){ + cnt -= got; + offset += got; + prior += got; + pBuf = (void*)(got + (char*)pBuf); + } + }while( got>0 ); + TIMER_END; + OSTRACE(("READ %-3d %5d %7lld %llu\n", + id->h, got+prior, offset-prior, TIMER_ELAPSED)); + return got+prior; +} + +/* +** Read data from a file into a buffer. Return SQLITE_OK if all +** bytes were read successfully and SQLITE_IOERR if anything goes +** wrong. +*/ +static int unixRead( + sqlite3_file *id, + void *pBuf, + int amt, + sqlite3_int64 offset +){ + unixFile *pFile = (unixFile *)id; + int got; + assert( id ); + assert( offset>=0 ); + assert( amt>0 ); + + /* If this is a database file (not a journal, master-journal or temp + ** file), the bytes in the locking range should never be read or written. */ +#if 0 + assert( pFile->pPreallocatedUnused==0 + || offset>=PENDING_BYTE+512 + || offset+amt<=PENDING_BYTE + ); +#endif + +#if SQLITE_MAX_MMAP_SIZE>0 + /* Deal with as much of this read request as possible by transfering + ** data from the memory mapping using memcpy(). */ + if( offsetmmapSize ){ + if( offset+amt <= pFile->mmapSize ){ + memcpy(pBuf, &((u8 *)(pFile->pMapRegion))[offset], amt); + return SQLITE_OK; + }else{ + int nCopy = pFile->mmapSize - offset; + memcpy(pBuf, &((u8 *)(pFile->pMapRegion))[offset], nCopy); + pBuf = &((u8 *)pBuf)[nCopy]; + amt -= nCopy; + offset += nCopy; + } + } +#endif + + got = seekAndRead(pFile, offset, pBuf, amt); + if( got==amt ){ + return SQLITE_OK; + }else if( got<0 ){ + /* lastErrno set by seekAndRead */ + return SQLITE_IOERR_READ; + }else{ + storeLastErrno(pFile, 0); /* not a system error */ + /* Unread parts of the buffer must be zero-filled */ + memset(&((char*)pBuf)[got], 0, amt-got); + return SQLITE_IOERR_SHORT_READ; + } +} + +/* +** Attempt to seek the file-descriptor passed as the first argument to +** absolute offset iOff, then attempt to write nBuf bytes of data from +** pBuf to it. If an error occurs, return -1 and set *piErrno. Otherwise, +** return the actual number of bytes written (which may be less than +** nBuf). +*/ +static int seekAndWriteFd( + int fd, /* File descriptor to write to */ + i64 iOff, /* File offset to begin writing at */ + const void *pBuf, /* Copy data from this buffer to the file */ + int nBuf, /* Size of buffer pBuf in bytes */ + int *piErrno /* OUT: Error number if error occurs */ +){ + int rc = 0; /* Value returned by system call */ + + assert( nBuf==(nBuf&0x1ffff) ); + assert( fd>2 ); + assert( piErrno!=0 ); + nBuf &= 0x1ffff; + TIMER_START; + +#if defined(USE_PREAD) + do{ rc = (int)osPwrite(fd, pBuf, nBuf, iOff); }while( rc<0 && errno==EINTR ); +#elif defined(USE_PREAD64) + do{ rc = (int)osPwrite64(fd, pBuf, nBuf, iOff);}while( rc<0 && errno==EINTR); +#else + do{ + i64 iSeek = lseek(fd, iOff, SEEK_SET); + SimulateIOError( iSeek = -1 ); + if( iSeek<0 ){ + rc = -1; + break; + } + rc = osWrite(fd, pBuf, nBuf); + }while( rc<0 && errno==EINTR ); +#endif + + TIMER_END; + OSTRACE(("WRITE %-3d %5d %7lld %llu\n", fd, rc, iOff, TIMER_ELAPSED)); + + if( rc<0 ) *piErrno = errno; + return rc; +} + + +/* +** Seek to the offset in id->offset then read cnt bytes into pBuf. +** Return the number of bytes actually read. Update the offset. +** +** To avoid stomping the errno value on a failed write the lastErrno value +** is set before returning. +*/ +static int seekAndWrite(unixFile *id, i64 offset, const void *pBuf, int cnt){ + return seekAndWriteFd(id->h, offset, pBuf, cnt, &id->lastErrno); +} + + +/* +** Write data from a buffer into a file. Return SQLITE_OK on success +** or some other error code on failure. +*/ +static int unixWrite( + sqlite3_file *id, + const void *pBuf, + int amt, + sqlite3_int64 offset +){ + unixFile *pFile = (unixFile*)id; + int wrote = 0; + assert( id ); + assert( amt>0 ); + + /* If this is a database file (not a journal, master-journal or temp + ** file), the bytes in the locking range should never be read or written. */ +#if 0 + assert( pFile->pPreallocatedUnused==0 + || offset>=PENDING_BYTE+512 + || offset+amt<=PENDING_BYTE + ); +#endif + +#ifdef SQLITE_DEBUG + /* If we are doing a normal write to a database file (as opposed to + ** doing a hot-journal rollback or a write to some file other than a + ** normal database file) then record the fact that the database + ** has changed. If the transaction counter is modified, record that + ** fact too. + */ + if( pFile->inNormalWrite ){ + pFile->dbUpdate = 1; /* The database has been modified */ + if( offset<=24 && offset+amt>=27 ){ + int rc; + char oldCntr[4]; + SimulateIOErrorBenign(1); + rc = seekAndRead(pFile, 24, oldCntr, 4); + SimulateIOErrorBenign(0); + if( rc!=4 || memcmp(oldCntr, &((char*)pBuf)[24-offset], 4)!=0 ){ + pFile->transCntrChng = 1; /* The transaction counter has changed */ + } + } + } +#endif + +#if defined(SQLITE_MMAP_READWRITE) && SQLITE_MAX_MMAP_SIZE>0 + /* Deal with as much of this write request as possible by transfering + ** data from the memory mapping using memcpy(). */ + if( offsetmmapSize ){ + if( offset+amt <= pFile->mmapSize ){ + memcpy(&((u8 *)(pFile->pMapRegion))[offset], pBuf, amt); + return SQLITE_OK; + }else{ + int nCopy = pFile->mmapSize - offset; + memcpy(&((u8 *)(pFile->pMapRegion))[offset], pBuf, nCopy); + pBuf = &((u8 *)pBuf)[nCopy]; + amt -= nCopy; + offset += nCopy; + } + } +#endif + + while( (wrote = seekAndWrite(pFile, offset, pBuf, amt))0 ){ + amt -= wrote; + offset += wrote; + pBuf = &((char*)pBuf)[wrote]; + } + SimulateIOError(( wrote=(-1), amt=1 )); + SimulateDiskfullError(( wrote=0, amt=1 )); + + if( amt>wrote ){ + if( wrote<0 && pFile->lastErrno!=ENOSPC ){ + /* lastErrno set by seekAndWrite */ + return SQLITE_IOERR_WRITE; + }else{ + storeLastErrno(pFile, 0); /* not a system error */ + return SQLITE_FULL; + } + } + + return SQLITE_OK; +} + +#ifdef SQLITE_TEST +/* +** Count the number of fullsyncs and normal syncs. This is used to test +** that syncs and fullsyncs are occurring at the right times. +*/ +SQLITE_API int sqlite3_sync_count = 0; +SQLITE_API int sqlite3_fullsync_count = 0; +#endif + +/* +** We do not trust systems to provide a working fdatasync(). Some do. +** Others do no. To be safe, we will stick with the (slightly slower) +** fsync(). If you know that your system does support fdatasync() correctly, +** then simply compile with -Dfdatasync=fdatasync or -DHAVE_FDATASYNC +*/ +#if !defined(fdatasync) && !HAVE_FDATASYNC +# define fdatasync fsync +#endif + +/* +** Define HAVE_FULLFSYNC to 0 or 1 depending on whether or not +** the F_FULLFSYNC macro is defined. F_FULLFSYNC is currently +** only available on Mac OS X. But that could change. +*/ +#ifdef F_FULLFSYNC +# define HAVE_FULLFSYNC 1 +#else +# define HAVE_FULLFSYNC 0 +#endif + + +/* +** The fsync() system call does not work as advertised on many +** unix systems. The following procedure is an attempt to make +** it work better. +** +** The SQLITE_NO_SYNC macro disables all fsync()s. This is useful +** for testing when we want to run through the test suite quickly. +** You are strongly advised *not* to deploy with SQLITE_NO_SYNC +** enabled, however, since with SQLITE_NO_SYNC enabled, an OS crash +** or power failure will likely corrupt the database file. +** +** SQLite sets the dataOnly flag if the size of the file is unchanged. +** The idea behind dataOnly is that it should only write the file content +** to disk, not the inode. We only set dataOnly if the file size is +** unchanged since the file size is part of the inode. However, +** Ted Ts'o tells us that fdatasync() will also write the inode if the +** file size has changed. The only real difference between fdatasync() +** and fsync(), Ted tells us, is that fdatasync() will not flush the +** inode if the mtime or owner or other inode attributes have changed. +** We only care about the file size, not the other file attributes, so +** as far as SQLite is concerned, an fdatasync() is always adequate. +** So, we always use fdatasync() if it is available, regardless of +** the value of the dataOnly flag. +*/ +static int full_fsync(int fd, int fullSync, int dataOnly){ + int rc; + + /* The following "ifdef/elif/else/" block has the same structure as + ** the one below. It is replicated here solely to avoid cluttering + ** up the real code with the UNUSED_PARAMETER() macros. + */ +#ifdef SQLITE_NO_SYNC + UNUSED_PARAMETER(fd); + UNUSED_PARAMETER(fullSync); + UNUSED_PARAMETER(dataOnly); +#elif HAVE_FULLFSYNC + UNUSED_PARAMETER(dataOnly); +#else + UNUSED_PARAMETER(fullSync); + UNUSED_PARAMETER(dataOnly); +#endif + + /* Record the number of times that we do a normal fsync() and + ** FULLSYNC. This is used during testing to verify that this procedure + ** gets called with the correct arguments. + */ +#ifdef SQLITE_TEST + if( fullSync ) sqlite3_fullsync_count++; + sqlite3_sync_count++; +#endif + + /* If we compiled with the SQLITE_NO_SYNC flag, then syncing is a + ** no-op. But go ahead and call fstat() to validate the file + ** descriptor as we need a method to provoke a failure during + ** coverate testing. + */ +#ifdef SQLITE_NO_SYNC + { + struct stat buf; + rc = osFstat(fd, &buf); + } +#elif HAVE_FULLFSYNC + if( fullSync ){ + rc = osFcntl(fd, F_FULLFSYNC, 0); + }else{ + rc = 1; + } + /* If the FULLFSYNC failed, fall back to attempting an fsync(). + ** It shouldn't be possible for fullfsync to fail on the local + ** file system (on OSX), so failure indicates that FULLFSYNC + ** isn't supported for this file system. So, attempt an fsync + ** and (for now) ignore the overhead of a superfluous fcntl call. + ** It'd be better to detect fullfsync support once and avoid + ** the fcntl call every time sync is called. + */ + if( rc ) rc = fsync(fd); + +#elif defined(__APPLE__) + /* fdatasync() on HFS+ doesn't yet flush the file size if it changed correctly + ** so currently we default to the macro that redefines fdatasync to fsync + */ + rc = fsync(fd); +#else + rc = fdatasync(fd); +#if OS_VXWORKS + if( rc==-1 && errno==ENOTSUP ){ + rc = fsync(fd); + } +#endif /* OS_VXWORKS */ +#endif /* ifdef SQLITE_NO_SYNC elif HAVE_FULLFSYNC */ + + if( OS_VXWORKS && rc!= -1 ){ + rc = 0; + } + return rc; +} + +/* +** Open a file descriptor to the directory containing file zFilename. +** If successful, *pFd is set to the opened file descriptor and +** SQLITE_OK is returned. If an error occurs, either SQLITE_NOMEM +** or SQLITE_CANTOPEN is returned and *pFd is set to an undefined +** value. +** +** The directory file descriptor is used for only one thing - to +** fsync() a directory to make sure file creation and deletion events +** are flushed to disk. Such fsyncs are not needed on newer +** journaling filesystems, but are required on older filesystems. +** +** This routine can be overridden using the xSetSysCall interface. +** The ability to override this routine was added in support of the +** chromium sandbox. Opening a directory is a security risk (we are +** told) so making it overrideable allows the chromium sandbox to +** replace this routine with a harmless no-op. To make this routine +** a no-op, replace it with a stub that returns SQLITE_OK but leaves +** *pFd set to a negative number. +** +** If SQLITE_OK is returned, the caller is responsible for closing +** the file descriptor *pFd using close(). +*/ +static int openDirectory(const char *zFilename, int *pFd){ + int ii; + int fd = -1; + char zDirname[MAX_PATHNAME+1]; + + sqlite3_snprintf(MAX_PATHNAME, zDirname, "%s", zFilename); + for(ii=(int)strlen(zDirname); ii>0 && zDirname[ii]!='/'; ii--); + if( ii>0 ){ + zDirname[ii] = '\0'; + }else{ + if( zDirname[0]!='/' ) zDirname[0] = '.'; + zDirname[1] = 0; + } + fd = robust_open(zDirname, O_RDONLY|O_BINARY, 0); + if( fd>=0 ){ + OSTRACE(("OPENDIR %-3d %s\n", fd, zDirname)); + } + *pFd = fd; + if( fd>=0 ) return SQLITE_OK; + return unixLogError(SQLITE_CANTOPEN_BKPT, "openDirectory", zDirname); +} + +/* +** Make sure all writes to a particular file are committed to disk. +** +** If dataOnly==0 then both the file itself and its metadata (file +** size, access time, etc) are synced. If dataOnly!=0 then only the +** file data is synced. +** +** Under Unix, also make sure that the directory entry for the file +** has been created by fsync-ing the directory that contains the file. +** If we do not do this and we encounter a power failure, the directory +** entry for the journal might not exist after we reboot. The next +** SQLite to access the file will not know that the journal exists (because +** the directory entry for the journal was never created) and the transaction +** will not roll back - possibly leading to database corruption. +*/ +static int unixSync(sqlite3_file *id, int flags){ + int rc; + unixFile *pFile = (unixFile*)id; + + int isDataOnly = (flags&SQLITE_SYNC_DATAONLY); + int isFullsync = (flags&0x0F)==SQLITE_SYNC_FULL; + + /* Check that one of SQLITE_SYNC_NORMAL or FULL was passed */ + assert((flags&0x0F)==SQLITE_SYNC_NORMAL + || (flags&0x0F)==SQLITE_SYNC_FULL + ); + + /* Unix cannot, but some systems may return SQLITE_FULL from here. This + ** line is to test that doing so does not cause any problems. + */ + SimulateDiskfullError( return SQLITE_FULL ); + + assert( pFile ); + OSTRACE(("SYNC %-3d\n", pFile->h)); + rc = full_fsync(pFile->h, isFullsync, isDataOnly); + SimulateIOError( rc=1 ); + if( rc ){ + storeLastErrno(pFile, errno); + return unixLogError(SQLITE_IOERR_FSYNC, "full_fsync", pFile->zPath); + } + + /* Also fsync the directory containing the file if the DIRSYNC flag + ** is set. This is a one-time occurrence. Many systems (examples: AIX) + ** are unable to fsync a directory, so ignore errors on the fsync. + */ + if( pFile->ctrlFlags & UNIXFILE_DIRSYNC ){ + int dirfd; + OSTRACE(("DIRSYNC %s (have_fullfsync=%d fullsync=%d)\n", pFile->zPath, + HAVE_FULLFSYNC, isFullsync)); + rc = osOpenDirectory(pFile->zPath, &dirfd); + if( rc==SQLITE_OK ){ + full_fsync(dirfd, 0, 0); + robust_close(pFile, dirfd, __LINE__); + }else{ + assert( rc==SQLITE_CANTOPEN ); + rc = SQLITE_OK; + } + pFile->ctrlFlags &= ~UNIXFILE_DIRSYNC; + } + return rc; +} + +/* +** Truncate an open file to a specified size +*/ +static int unixTruncate(sqlite3_file *id, i64 nByte){ + unixFile *pFile = (unixFile *)id; + int rc; + assert( pFile ); + SimulateIOError( return SQLITE_IOERR_TRUNCATE ); + + /* If the user has configured a chunk-size for this file, truncate the + ** file so that it consists of an integer number of chunks (i.e. the + ** actual file size after the operation may be larger than the requested + ** size). + */ + if( pFile->szChunk>0 ){ + nByte = ((nByte + pFile->szChunk - 1)/pFile->szChunk) * pFile->szChunk; + } + + rc = robust_ftruncate(pFile->h, nByte); + if( rc ){ + storeLastErrno(pFile, errno); + return unixLogError(SQLITE_IOERR_TRUNCATE, "ftruncate", pFile->zPath); + }else{ +#ifdef SQLITE_DEBUG + /* If we are doing a normal write to a database file (as opposed to + ** doing a hot-journal rollback or a write to some file other than a + ** normal database file) and we truncate the file to zero length, + ** that effectively updates the change counter. This might happen + ** when restoring a database using the backup API from a zero-length + ** source. + */ + if( pFile->inNormalWrite && nByte==0 ){ + pFile->transCntrChng = 1; + } +#endif + +#if SQLITE_MAX_MMAP_SIZE>0 + /* If the file was just truncated to a size smaller than the currently + ** mapped region, reduce the effective mapping size as well. SQLite will + ** use read() and write() to access data beyond this point from now on. + */ + if( nBytemmapSize ){ + pFile->mmapSize = nByte; + } +#endif + + return SQLITE_OK; + } +} + +/* +** Determine the current size of a file in bytes +*/ +static int unixFileSize(sqlite3_file *id, i64 *pSize){ + int rc; + struct stat buf; + assert( id ); + rc = osFstat(((unixFile*)id)->h, &buf); + SimulateIOError( rc=1 ); + if( rc!=0 ){ + storeLastErrno((unixFile*)id, errno); + return SQLITE_IOERR_FSTAT; + } + *pSize = buf.st_size; + + /* When opening a zero-size database, the findInodeInfo() procedure + ** writes a single byte into that file in order to work around a bug + ** in the OS-X msdos filesystem. In order to avoid problems with upper + ** layers, we need to report this file size as zero even though it is + ** really 1. Ticket #3260. + */ + if( *pSize==1 ) *pSize = 0; + + + return SQLITE_OK; +} + +#if SQLITE_ENABLE_LOCKING_STYLE && defined(__APPLE__) +/* +** Handler for proxy-locking file-control verbs. Defined below in the +** proxying locking division. +*/ +static int proxyFileControl(sqlite3_file*,int,void*); +#endif + +/* +** This function is called to handle the SQLITE_FCNTL_SIZE_HINT +** file-control operation. Enlarge the database to nBytes in size +** (rounded up to the next chunk-size). If the database is already +** nBytes or larger, this routine is a no-op. +*/ +static int fcntlSizeHint(unixFile *pFile, i64 nByte){ + if( pFile->szChunk>0 ){ + i64 nSize; /* Required file size */ + struct stat buf; /* Used to hold return values of fstat() */ + + if( osFstat(pFile->h, &buf) ){ + return SQLITE_IOERR_FSTAT; + } + + nSize = ((nByte+pFile->szChunk-1) / pFile->szChunk) * pFile->szChunk; + if( nSize>(i64)buf.st_size ){ + +#if defined(HAVE_POSIX_FALLOCATE) && HAVE_POSIX_FALLOCATE + /* The code below is handling the return value of osFallocate() + ** correctly. posix_fallocate() is defined to "returns zero on success, + ** or an error number on failure". See the manpage for details. */ + int err; + do{ + err = osFallocate(pFile->h, buf.st_size, nSize-buf.st_size); + }while( err==EINTR ); + if( err && err!=EINVAL ) return SQLITE_IOERR_WRITE; +#else + /* If the OS does not have posix_fallocate(), fake it. Write a + ** single byte to the last byte in each block that falls entirely + ** within the extended region. Then, if required, a single byte + ** at offset (nSize-1), to set the size of the file correctly. + ** This is a similar technique to that used by glibc on systems + ** that do not have a real fallocate() call. + */ + int nBlk = buf.st_blksize; /* File-system block size */ + int nWrite = 0; /* Number of bytes written by seekAndWrite */ + i64 iWrite; /* Next offset to write to */ + + iWrite = (buf.st_size/nBlk)*nBlk + nBlk - 1; + assert( iWrite>=buf.st_size ); + assert( ((iWrite+1)%nBlk)==0 ); + for(/*no-op*/; iWrite=nSize ) iWrite = nSize - 1; + nWrite = seekAndWrite(pFile, iWrite, "", 1); + if( nWrite!=1 ) return SQLITE_IOERR_WRITE; + } +#endif + } + } + +#if SQLITE_MAX_MMAP_SIZE>0 + if( pFile->mmapSizeMax>0 && nByte>pFile->mmapSize ){ + int rc; + if( pFile->szChunk<=0 ){ + if( robust_ftruncate(pFile->h, nByte) ){ + storeLastErrno(pFile, errno); + return unixLogError(SQLITE_IOERR_TRUNCATE, "ftruncate", pFile->zPath); + } + } + + rc = unixMapfile(pFile, nByte); + return rc; + } +#endif + + return SQLITE_OK; +} + +/* +** If *pArg is initially negative then this is a query. Set *pArg to +** 1 or 0 depending on whether or not bit mask of pFile->ctrlFlags is set. +** +** If *pArg is 0 or 1, then clear or set the mask bit of pFile->ctrlFlags. +*/ +static void unixModeBit(unixFile *pFile, unsigned char mask, int *pArg){ + if( *pArg<0 ){ + *pArg = (pFile->ctrlFlags & mask)!=0; + }else if( (*pArg)==0 ){ + pFile->ctrlFlags &= ~mask; + }else{ + pFile->ctrlFlags |= mask; + } +} + +/* Forward declaration */ +static int unixGetTempname(int nBuf, char *zBuf); + +/* +** Information and control of an open file handle. +*/ +static int unixFileControl(sqlite3_file *id, int op, void *pArg){ + unixFile *pFile = (unixFile*)id; + switch( op ){ +#if defined(__linux__) && defined(SQLITE_ENABLE_BATCH_ATOMIC_WRITE) + case SQLITE_FCNTL_BEGIN_ATOMIC_WRITE: { + int rc = osIoctl(pFile->h, F2FS_IOC_START_ATOMIC_WRITE); + return rc ? SQLITE_IOERR_BEGIN_ATOMIC : SQLITE_OK; + } + case SQLITE_FCNTL_COMMIT_ATOMIC_WRITE: { + int rc = osIoctl(pFile->h, F2FS_IOC_COMMIT_ATOMIC_WRITE); + return rc ? SQLITE_IOERR_COMMIT_ATOMIC : SQLITE_OK; + } + case SQLITE_FCNTL_ROLLBACK_ATOMIC_WRITE: { + int rc = osIoctl(pFile->h, F2FS_IOC_ABORT_VOLATILE_WRITE); + return rc ? SQLITE_IOERR_ROLLBACK_ATOMIC : SQLITE_OK; + } +#endif /* __linux__ && SQLITE_ENABLE_BATCH_ATOMIC_WRITE */ + + case SQLITE_FCNTL_LOCKSTATE: { + *(int*)pArg = pFile->eFileLock; + return SQLITE_OK; + } + case SQLITE_FCNTL_LAST_ERRNO: { + *(int*)pArg = pFile->lastErrno; + return SQLITE_OK; + } + case SQLITE_FCNTL_CHUNK_SIZE: { + pFile->szChunk = *(int *)pArg; + return SQLITE_OK; + } + case SQLITE_FCNTL_SIZE_HINT: { + int rc; + SimulateIOErrorBenign(1); + rc = fcntlSizeHint(pFile, *(i64 *)pArg); + SimulateIOErrorBenign(0); + return rc; + } + case SQLITE_FCNTL_PERSIST_WAL: { + unixModeBit(pFile, UNIXFILE_PERSIST_WAL, (int*)pArg); + return SQLITE_OK; + } + case SQLITE_FCNTL_POWERSAFE_OVERWRITE: { + unixModeBit(pFile, UNIXFILE_PSOW, (int*)pArg); + return SQLITE_OK; + } + case SQLITE_FCNTL_VFSNAME: { + *(char**)pArg = sqlite3_mprintf("%s", pFile->pVfs->zName); + return SQLITE_OK; + } + case SQLITE_FCNTL_TEMPFILENAME: { + char *zTFile = sqlite3_malloc64( pFile->pVfs->mxPathname ); + if( zTFile ){ + unixGetTempname(pFile->pVfs->mxPathname, zTFile); + *(char**)pArg = zTFile; + } + return SQLITE_OK; + } + case SQLITE_FCNTL_HAS_MOVED: { + *(int*)pArg = fileHasMoved(pFile); + return SQLITE_OK; + } +#ifdef SQLITE_ENABLE_SETLK_TIMEOUT + case SQLITE_FCNTL_LOCK_TIMEOUT: { + pFile->iBusyTimeout = *(int*)pArg; + return SQLITE_OK; + } +#endif +#if SQLITE_MAX_MMAP_SIZE>0 + case SQLITE_FCNTL_MMAP_SIZE: { + i64 newLimit = *(i64*)pArg; + int rc = SQLITE_OK; + if( newLimit>sqlite3GlobalConfig.mxMmap ){ + newLimit = sqlite3GlobalConfig.mxMmap; + } + + /* The value of newLimit may be eventually cast to (size_t) and passed + ** to mmap(). Restrict its value to 2GB if (size_t) is not at least a + ** 64-bit type. */ + if( newLimit>0 && sizeof(size_t)<8 ){ + newLimit = (newLimit & 0x7FFFFFFF); + } + + *(i64*)pArg = pFile->mmapSizeMax; + if( newLimit>=0 && newLimit!=pFile->mmapSizeMax && pFile->nFetchOut==0 ){ + pFile->mmapSizeMax = newLimit; + if( pFile->mmapSize>0 ){ + unixUnmapfile(pFile); + rc = unixMapfile(pFile, -1); + } + } + return rc; + } +#endif +#ifdef SQLITE_DEBUG + /* The pager calls this method to signal that it has done + ** a rollback and that the database is therefore unchanged and + ** it hence it is OK for the transaction change counter to be + ** unchanged. + */ + case SQLITE_FCNTL_DB_UNCHANGED: { + ((unixFile*)id)->dbUpdate = 0; + return SQLITE_OK; + } +#endif +#if SQLITE_ENABLE_LOCKING_STYLE && defined(__APPLE__) + case SQLITE_FCNTL_SET_LOCKPROXYFILE: + case SQLITE_FCNTL_GET_LOCKPROXYFILE: { + return proxyFileControl(id,op,pArg); + } +#endif /* SQLITE_ENABLE_LOCKING_STYLE && defined(__APPLE__) */ + } + return SQLITE_NOTFOUND; +} + +/* +** If pFd->sectorSize is non-zero when this function is called, it is a +** no-op. Otherwise, the values of pFd->sectorSize and +** pFd->deviceCharacteristics are set according to the file-system +** characteristics. +** +** There are two versions of this function. One for QNX and one for all +** other systems. +*/ +#ifndef __QNXNTO__ +static void setDeviceCharacteristics(unixFile *pFd){ + assert( pFd->deviceCharacteristics==0 || pFd->sectorSize!=0 ); + if( pFd->sectorSize==0 ){ +#if defined(__linux__) && defined(SQLITE_ENABLE_BATCH_ATOMIC_WRITE) + int res; + u32 f = 0; + + /* Check for support for F2FS atomic batch writes. */ + res = osIoctl(pFd->h, F2FS_IOC_GET_FEATURES, &f); + if( res==0 && (f & F2FS_FEATURE_ATOMIC_WRITE) ){ + pFd->deviceCharacteristics = SQLITE_IOCAP_BATCH_ATOMIC; + } +#endif /* __linux__ && SQLITE_ENABLE_BATCH_ATOMIC_WRITE */ + + /* Set the POWERSAFE_OVERWRITE flag if requested. */ + if( pFd->ctrlFlags & UNIXFILE_PSOW ){ + pFd->deviceCharacteristics |= SQLITE_IOCAP_POWERSAFE_OVERWRITE; + } + + pFd->sectorSize = SQLITE_DEFAULT_SECTOR_SIZE; + } +} +#else +#include +#include +static void setDeviceCharacteristics(unixFile *pFile){ + if( pFile->sectorSize == 0 ){ + struct statvfs fsInfo; + + /* Set defaults for non-supported filesystems */ + pFile->sectorSize = SQLITE_DEFAULT_SECTOR_SIZE; + pFile->deviceCharacteristics = 0; + if( fstatvfs(pFile->h, &fsInfo) == -1 ) { + return; + } + + if( !strcmp(fsInfo.f_basetype, "tmp") ) { + pFile->sectorSize = fsInfo.f_bsize; + pFile->deviceCharacteristics = + SQLITE_IOCAP_ATOMIC4K | /* All ram filesystem writes are atomic */ + SQLITE_IOCAP_SAFE_APPEND | /* growing the file does not occur until + ** the write succeeds */ + SQLITE_IOCAP_SEQUENTIAL | /* The ram filesystem has no write behind + ** so it is ordered */ + 0; + }else if( strstr(fsInfo.f_basetype, "etfs") ){ + pFile->sectorSize = fsInfo.f_bsize; + pFile->deviceCharacteristics = + /* etfs cluster size writes are atomic */ + (pFile->sectorSize / 512 * SQLITE_IOCAP_ATOMIC512) | + SQLITE_IOCAP_SAFE_APPEND | /* growing the file does not occur until + ** the write succeeds */ + SQLITE_IOCAP_SEQUENTIAL | /* The ram filesystem has no write behind + ** so it is ordered */ + 0; + }else if( !strcmp(fsInfo.f_basetype, "qnx6") ){ + pFile->sectorSize = fsInfo.f_bsize; + pFile->deviceCharacteristics = + SQLITE_IOCAP_ATOMIC | /* All filesystem writes are atomic */ + SQLITE_IOCAP_SAFE_APPEND | /* growing the file does not occur until + ** the write succeeds */ + SQLITE_IOCAP_SEQUENTIAL | /* The ram filesystem has no write behind + ** so it is ordered */ + 0; + }else if( !strcmp(fsInfo.f_basetype, "qnx4") ){ + pFile->sectorSize = fsInfo.f_bsize; + pFile->deviceCharacteristics = + /* full bitset of atomics from max sector size and smaller */ + ((pFile->sectorSize / 512 * SQLITE_IOCAP_ATOMIC512) << 1) - 2 | + SQLITE_IOCAP_SEQUENTIAL | /* The ram filesystem has no write behind + ** so it is ordered */ + 0; + }else if( strstr(fsInfo.f_basetype, "dos") ){ + pFile->sectorSize = fsInfo.f_bsize; + pFile->deviceCharacteristics = + /* full bitset of atomics from max sector size and smaller */ + ((pFile->sectorSize / 512 * SQLITE_IOCAP_ATOMIC512) << 1) - 2 | + SQLITE_IOCAP_SEQUENTIAL | /* The ram filesystem has no write behind + ** so it is ordered */ + 0; + }else{ + pFile->deviceCharacteristics = + SQLITE_IOCAP_ATOMIC512 | /* blocks are atomic */ + SQLITE_IOCAP_SAFE_APPEND | /* growing the file does not occur until + ** the write succeeds */ + 0; + } + } + /* Last chance verification. If the sector size isn't a multiple of 512 + ** then it isn't valid.*/ + if( pFile->sectorSize % 512 != 0 ){ + pFile->deviceCharacteristics = 0; + pFile->sectorSize = SQLITE_DEFAULT_SECTOR_SIZE; + } +} +#endif + +/* +** Return the sector size in bytes of the underlying block device for +** the specified file. This is almost always 512 bytes, but may be +** larger for some devices. +** +** SQLite code assumes this function cannot fail. It also assumes that +** if two files are created in the same file-system directory (i.e. +** a database and its journal file) that the sector size will be the +** same for both. +*/ +static int unixSectorSize(sqlite3_file *id){ + unixFile *pFd = (unixFile*)id; + setDeviceCharacteristics(pFd); + return pFd->sectorSize; +} + +/* +** Return the device characteristics for the file. +** +** This VFS is set up to return SQLITE_IOCAP_POWERSAFE_OVERWRITE by default. +** However, that choice is controversial since technically the underlying +** file system does not always provide powersafe overwrites. (In other +** words, after a power-loss event, parts of the file that were never +** written might end up being altered.) However, non-PSOW behavior is very, +** very rare. And asserting PSOW makes a large reduction in the amount +** of required I/O for journaling, since a lot of padding is eliminated. +** Hence, while POWERSAFE_OVERWRITE is on by default, there is a file-control +** available to turn it off and URI query parameter available to turn it off. +*/ +static int unixDeviceCharacteristics(sqlite3_file *id){ + unixFile *pFd = (unixFile*)id; + setDeviceCharacteristics(pFd); + return pFd->deviceCharacteristics; +} + +#if !defined(SQLITE_OMIT_WAL) || SQLITE_MAX_MMAP_SIZE>0 + +/* +** Return the system page size. +** +** This function should not be called directly by other code in this file. +** Instead, it should be called via macro osGetpagesize(). +*/ +static int unixGetpagesize(void){ +#if OS_VXWORKS + return 1024; +#elif defined(_BSD_SOURCE) + return getpagesize(); +#else + return (int)sysconf(_SC_PAGESIZE); +#endif +} + +#endif /* !defined(SQLITE_OMIT_WAL) || SQLITE_MAX_MMAP_SIZE>0 */ + +#ifndef SQLITE_OMIT_WAL + +/* +** Object used to represent an shared memory buffer. +** +** When multiple threads all reference the same wal-index, each thread +** has its own unixShm object, but they all point to a single instance +** of this unixShmNode object. In other words, each wal-index is opened +** only once per process. +** +** Each unixShmNode object is connected to a single unixInodeInfo object. +** We could coalesce this object into unixInodeInfo, but that would mean +** every open file that does not use shared memory (in other words, most +** open files) would have to carry around this extra information. So +** the unixInodeInfo object contains a pointer to this unixShmNode object +** and the unixShmNode object is created only when needed. +** +** unixMutexHeld() must be true when creating or destroying +** this object or while reading or writing the following fields: +** +** nRef +** +** The following fields are read-only after the object is created: +** +** hShm +** zFilename +** +** Either unixShmNode.pShmMutex must be held or unixShmNode.nRef==0 and +** unixMutexHeld() is true when reading or writing any other field +** in this structure. +*/ +struct unixShmNode { + unixInodeInfo *pInode; /* unixInodeInfo that owns this SHM node */ + sqlite3_mutex *pShmMutex; /* Mutex to access this object */ + char *zFilename; /* Name of the mmapped file */ + int hShm; /* Open file descriptor */ + int szRegion; /* Size of shared-memory regions */ + u16 nRegion; /* Size of array apRegion */ + u8 isReadonly; /* True if read-only */ + u8 isUnlocked; /* True if no DMS lock held */ + char **apRegion; /* Array of mapped shared-memory regions */ + int nRef; /* Number of unixShm objects pointing to this */ + unixShm *pFirst; /* All unixShm objects pointing to this */ +#ifdef SQLITE_DEBUG + u8 exclMask; /* Mask of exclusive locks held */ + u8 sharedMask; /* Mask of shared locks held */ + u8 nextShmId; /* Next available unixShm.id value */ +#endif +}; + +/* +** Structure used internally by this VFS to record the state of an +** open shared memory connection. +** +** The following fields are initialized when this object is created and +** are read-only thereafter: +** +** unixShm.pShmNode +** unixShm.id +** +** All other fields are read/write. The unixShm.pShmNode->pShmMutex must +** be held while accessing any read/write fields. +*/ +struct unixShm { + unixShmNode *pShmNode; /* The underlying unixShmNode object */ + unixShm *pNext; /* Next unixShm with the same unixShmNode */ + u8 hasMutex; /* True if holding the unixShmNode->pShmMutex */ + u8 id; /* Id of this connection within its unixShmNode */ + u16 sharedMask; /* Mask of shared locks held */ + u16 exclMask; /* Mask of exclusive locks held */ +}; + +/* +** Constants used for locking +*/ +#define UNIX_SHM_BASE ((22+SQLITE_SHM_NLOCK)*4) /* first lock byte */ +#define UNIX_SHM_DMS (UNIX_SHM_BASE+SQLITE_SHM_NLOCK) /* deadman switch */ + +/* +** Apply posix advisory locks for all bytes from ofst through ofst+n-1. +** +** Locks block if the mask is exactly UNIX_SHM_C and are non-blocking +** otherwise. +*/ +static int unixShmSystemLock( + unixFile *pFile, /* Open connection to the WAL file */ + int lockType, /* F_UNLCK, F_RDLCK, or F_WRLCK */ + int ofst, /* First byte of the locking range */ + int n /* Number of bytes to lock */ +){ + unixShmNode *pShmNode; /* Apply locks to this open shared-memory segment */ + struct flock f; /* The posix advisory locking structure */ + int rc = SQLITE_OK; /* Result code form fcntl() */ + + /* Access to the unixShmNode object is serialized by the caller */ + pShmNode = pFile->pInode->pShmNode; + assert( pShmNode->nRef==0 || sqlite3_mutex_held(pShmNode->pShmMutex) ); + assert( pShmNode->nRef>0 || unixMutexHeld() ); + + /* Shared locks never span more than one byte */ + assert( n==1 || lockType!=F_RDLCK ); + + /* Locks are within range */ + assert( n>=1 && n<=SQLITE_SHM_NLOCK ); + + if( pShmNode->hShm>=0 ){ + /* Initialize the locking parameters */ + f.l_type = lockType; + f.l_whence = SEEK_SET; + f.l_start = ofst; + f.l_len = n; + rc = osSetPosixAdvisoryLock(pShmNode->hShm, &f, pFile); + rc = (rc!=(-1)) ? SQLITE_OK : SQLITE_BUSY; + } + + /* Update the global lock state and do debug tracing */ +#ifdef SQLITE_DEBUG + { u16 mask; + OSTRACE(("SHM-LOCK ")); + mask = ofst>31 ? 0xffff : (1<<(ofst+n)) - (1<exclMask &= ~mask; + pShmNode->sharedMask &= ~mask; + }else if( lockType==F_RDLCK ){ + OSTRACE(("read-lock %d ok", ofst)); + pShmNode->exclMask &= ~mask; + pShmNode->sharedMask |= mask; + }else{ + assert( lockType==F_WRLCK ); + OSTRACE(("write-lock %d ok", ofst)); + pShmNode->exclMask |= mask; + pShmNode->sharedMask &= ~mask; + } + }else{ + if( lockType==F_UNLCK ){ + OSTRACE(("unlock %d failed", ofst)); + }else if( lockType==F_RDLCK ){ + OSTRACE(("read-lock failed")); + }else{ + assert( lockType==F_WRLCK ); + OSTRACE(("write-lock %d failed", ofst)); + } + } + OSTRACE((" - afterwards %03x,%03x\n", + pShmNode->sharedMask, pShmNode->exclMask)); + } +#endif + + return rc; +} + +/* +** Return the minimum number of 32KB shm regions that should be mapped at +** a time, assuming that each mapping must be an integer multiple of the +** current system page-size. +** +** Usually, this is 1. The exception seems to be systems that are configured +** to use 64KB pages - in this case each mapping must cover at least two +** shm regions. +*/ +static int unixShmRegionPerMap(void){ + int shmsz = 32*1024; /* SHM region size */ + int pgsz = osGetpagesize(); /* System page size */ + assert( ((pgsz-1)&pgsz)==0 ); /* Page size must be a power of 2 */ + if( pgszpInode->pShmNode; + assert( unixMutexHeld() ); + if( p && ALWAYS(p->nRef==0) ){ + int nShmPerMap = unixShmRegionPerMap(); + int i; + assert( p->pInode==pFd->pInode ); + sqlite3_mutex_free(p->pShmMutex); + for(i=0; inRegion; i+=nShmPerMap){ + if( p->hShm>=0 ){ + osMunmap(p->apRegion[i], p->szRegion); + }else{ + sqlite3_free(p->apRegion[i]); + } + } + sqlite3_free(p->apRegion); + if( p->hShm>=0 ){ + robust_close(pFd, p->hShm, __LINE__); + p->hShm = -1; + } + p->pInode->pShmNode = 0; + sqlite3_free(p); + } +} + +/* +** The DMS lock has not yet been taken on shm file pShmNode. Attempt to +** take it now. Return SQLITE_OK if successful, or an SQLite error +** code otherwise. +** +** If the DMS cannot be locked because this is a readonly_shm=1 +** connection and no other process already holds a lock, return +** SQLITE_READONLY_CANTINIT and set pShmNode->isUnlocked=1. +*/ +static int unixLockSharedMemory(unixFile *pDbFd, unixShmNode *pShmNode){ + struct flock lock; + int rc = SQLITE_OK; + + /* Use F_GETLK to determine the locks other processes are holding + ** on the DMS byte. If it indicates that another process is holding + ** a SHARED lock, then this process may also take a SHARED lock + ** and proceed with opening the *-shm file. + ** + ** Or, if no other process is holding any lock, then this process + ** is the first to open it. In this case take an EXCLUSIVE lock on the + ** DMS byte and truncate the *-shm file to zero bytes in size. Then + ** downgrade to a SHARED lock on the DMS byte. + ** + ** If another process is holding an EXCLUSIVE lock on the DMS byte, + ** return SQLITE_BUSY to the caller (it will try again). An earlier + ** version of this code attempted the SHARED lock at this point. But + ** this introduced a subtle race condition: if the process holding + ** EXCLUSIVE failed just before truncating the *-shm file, then this + ** process might open and use the *-shm file without truncating it. + ** And if the *-shm file has been corrupted by a power failure or + ** system crash, the database itself may also become corrupt. */ + lock.l_whence = SEEK_SET; + lock.l_start = UNIX_SHM_DMS; + lock.l_len = 1; + lock.l_type = F_WRLCK; + if( osFcntl(pShmNode->hShm, F_GETLK, &lock)!=0 ) { + rc = SQLITE_IOERR_LOCK; + }else if( lock.l_type==F_UNLCK ){ + if( pShmNode->isReadonly ){ + pShmNode->isUnlocked = 1; + rc = SQLITE_READONLY_CANTINIT; + }else{ + rc = unixShmSystemLock(pDbFd, F_WRLCK, UNIX_SHM_DMS, 1); + /* The first connection to attach must truncate the -shm file. We + ** truncate to 3 bytes (an arbitrary small number, less than the + ** -shm header size) rather than 0 as a system debugging aid, to + ** help detect if a -shm file truncation is legitimate or is the work + ** or a rogue process. */ + if( rc==SQLITE_OK && robust_ftruncate(pShmNode->hShm, 3) ){ + rc = unixLogError(SQLITE_IOERR_SHMOPEN,"ftruncate",pShmNode->zFilename); + } + } + }else if( lock.l_type==F_WRLCK ){ + rc = SQLITE_BUSY; + } + + if( rc==SQLITE_OK ){ + assert( lock.l_type==F_UNLCK || lock.l_type==F_RDLCK ); + rc = unixShmSystemLock(pDbFd, F_RDLCK, UNIX_SHM_DMS, 1); + } + return rc; +} + +/* +** Open a shared-memory area associated with open database file pDbFd. +** This particular implementation uses mmapped files. +** +** The file used to implement shared-memory is in the same directory +** as the open database file and has the same name as the open database +** file with the "-shm" suffix added. For example, if the database file +** is "/home/user1/config.db" then the file that is created and mmapped +** for shared memory will be called "/home/user1/config.db-shm". +** +** Another approach to is to use files in /dev/shm or /dev/tmp or an +** some other tmpfs mount. But if a file in a different directory +** from the database file is used, then differing access permissions +** or a chroot() might cause two different processes on the same +** database to end up using different files for shared memory - +** meaning that their memory would not really be shared - resulting +** in database corruption. Nevertheless, this tmpfs file usage +** can be enabled at compile-time using -DSQLITE_SHM_DIRECTORY="/dev/shm" +** or the equivalent. The use of the SQLITE_SHM_DIRECTORY compile-time +** option results in an incompatible build of SQLite; builds of SQLite +** that with differing SQLITE_SHM_DIRECTORY settings attempt to use the +** same database file at the same time, database corruption will likely +** result. The SQLITE_SHM_DIRECTORY compile-time option is considered +** "unsupported" and may go away in a future SQLite release. +** +** When opening a new shared-memory file, if no other instances of that +** file are currently open, in this process or in other processes, then +** the file must be truncated to zero length or have its header cleared. +** +** If the original database file (pDbFd) is using the "unix-excl" VFS +** that means that an exclusive lock is held on the database file and +** that no other processes are able to read or write the database. In +** that case, we do not really need shared memory. No shared memory +** file is created. The shared memory will be simulated with heap memory. +*/ +static int unixOpenSharedMemory(unixFile *pDbFd){ + struct unixShm *p = 0; /* The connection to be opened */ + struct unixShmNode *pShmNode; /* The underlying mmapped file */ + int rc = SQLITE_OK; /* Result code */ + unixInodeInfo *pInode; /* The inode of fd */ + char *zShm; /* Name of the file used for SHM */ + int nShmFilename; /* Size of the SHM filename in bytes */ + + /* Allocate space for the new unixShm object. */ + p = sqlite3_malloc64( sizeof(*p) ); + if( p==0 ) return SQLITE_NOMEM_BKPT; + memset(p, 0, sizeof(*p)); + assert( pDbFd->pShm==0 ); + + /* Check to see if a unixShmNode object already exists. Reuse an existing + ** one if present. Create a new one if necessary. + */ + assert( unixFileMutexNotheld(pDbFd) ); + unixEnterMutex(); + pInode = pDbFd->pInode; + pShmNode = pInode->pShmNode; + if( pShmNode==0 ){ + struct stat sStat; /* fstat() info for database file */ +#ifndef SQLITE_SHM_DIRECTORY + const char *zBasePath = pDbFd->zPath; +#endif + + /* Call fstat() to figure out the permissions on the database file. If + ** a new *-shm file is created, an attempt will be made to create it + ** with the same permissions. + */ + if( osFstat(pDbFd->h, &sStat) ){ + rc = SQLITE_IOERR_FSTAT; + goto shm_open_err; + } + +#ifdef SQLITE_SHM_DIRECTORY + nShmFilename = sizeof(SQLITE_SHM_DIRECTORY) + 31; +#else + nShmFilename = 6 + (int)strlen(zBasePath); +#endif + pShmNode = sqlite3_malloc64( sizeof(*pShmNode) + nShmFilename ); + if( pShmNode==0 ){ + rc = SQLITE_NOMEM_BKPT; + goto shm_open_err; + } + memset(pShmNode, 0, sizeof(*pShmNode)+nShmFilename); + zShm = pShmNode->zFilename = (char*)&pShmNode[1]; +#ifdef SQLITE_SHM_DIRECTORY + sqlite3_snprintf(nShmFilename, zShm, + SQLITE_SHM_DIRECTORY "/sqlite-shm-%x-%x", + (u32)sStat.st_ino, (u32)sStat.st_dev); +#else + sqlite3_snprintf(nShmFilename, zShm, "%s-shm", zBasePath); + sqlite3FileSuffix3(pDbFd->zPath, zShm); +#endif + pShmNode->hShm = -1; + pDbFd->pInode->pShmNode = pShmNode; + pShmNode->pInode = pDbFd->pInode; + if( sqlite3GlobalConfig.bCoreMutex ){ + pShmNode->pShmMutex = sqlite3_mutex_alloc(SQLITE_MUTEX_FAST); + if( pShmNode->pShmMutex==0 ){ + rc = SQLITE_NOMEM_BKPT; + goto shm_open_err; + } + } + + if( pInode->bProcessLock==0 ){ + if( 0==sqlite3_uri_boolean(pDbFd->zPath, "readonly_shm", 0) ){ + pShmNode->hShm = robust_open(zShm, O_RDWR|O_CREAT,(sStat.st_mode&0777)); + } + if( pShmNode->hShm<0 ){ + pShmNode->hShm = robust_open(zShm, O_RDONLY, (sStat.st_mode&0777)); + if( pShmNode->hShm<0 ){ + rc = unixLogError(SQLITE_CANTOPEN_BKPT, "open", zShm); + goto shm_open_err; + } + pShmNode->isReadonly = 1; + } + + /* If this process is running as root, make sure that the SHM file + ** is owned by the same user that owns the original database. Otherwise, + ** the original owner will not be able to connect. + */ + robustFchown(pShmNode->hShm, sStat.st_uid, sStat.st_gid); + + rc = unixLockSharedMemory(pDbFd, pShmNode); + if( rc!=SQLITE_OK && rc!=SQLITE_READONLY_CANTINIT ) goto shm_open_err; + } + } + + /* Make the new connection a child of the unixShmNode */ + p->pShmNode = pShmNode; +#ifdef SQLITE_DEBUG + p->id = pShmNode->nextShmId++; +#endif + pShmNode->nRef++; + pDbFd->pShm = p; + unixLeaveMutex(); + + /* The reference count on pShmNode has already been incremented under + ** the cover of the unixEnterMutex() mutex and the pointer from the + ** new (struct unixShm) object to the pShmNode has been set. All that is + ** left to do is to link the new object into the linked list starting + ** at pShmNode->pFirst. This must be done while holding the + ** pShmNode->pShmMutex. + */ + sqlite3_mutex_enter(pShmNode->pShmMutex); + p->pNext = pShmNode->pFirst; + pShmNode->pFirst = p; + sqlite3_mutex_leave(pShmNode->pShmMutex); + return rc; + + /* Jump here on any error */ +shm_open_err: + unixShmPurge(pDbFd); /* This call frees pShmNode if required */ + sqlite3_free(p); + unixLeaveMutex(); + return rc; +} + +/* +** This function is called to obtain a pointer to region iRegion of the +** shared-memory associated with the database file fd. Shared-memory regions +** are numbered starting from zero. Each shared-memory region is szRegion +** bytes in size. +** +** If an error occurs, an error code is returned and *pp is set to NULL. +** +** Otherwise, if the bExtend parameter is 0 and the requested shared-memory +** region has not been allocated (by any client, including one running in a +** separate process), then *pp is set to NULL and SQLITE_OK returned. If +** bExtend is non-zero and the requested shared-memory region has not yet +** been allocated, it is allocated by this function. +** +** If the shared-memory region has already been allocated or is allocated by +** this call as described above, then it is mapped into this processes +** address space (if it is not already), *pp is set to point to the mapped +** memory and SQLITE_OK returned. +*/ +static int unixShmMap( + sqlite3_file *fd, /* Handle open on database file */ + int iRegion, /* Region to retrieve */ + int szRegion, /* Size of regions */ + int bExtend, /* True to extend file if necessary */ + void volatile **pp /* OUT: Mapped memory */ +){ + unixFile *pDbFd = (unixFile*)fd; + unixShm *p; + unixShmNode *pShmNode; + int rc = SQLITE_OK; + int nShmPerMap = unixShmRegionPerMap(); + int nReqRegion; + + /* If the shared-memory file has not yet been opened, open it now. */ + if( pDbFd->pShm==0 ){ + rc = unixOpenSharedMemory(pDbFd); + if( rc!=SQLITE_OK ) return rc; + } + + p = pDbFd->pShm; + pShmNode = p->pShmNode; + sqlite3_mutex_enter(pShmNode->pShmMutex); + if( pShmNode->isUnlocked ){ + rc = unixLockSharedMemory(pDbFd, pShmNode); + if( rc!=SQLITE_OK ) goto shmpage_out; + pShmNode->isUnlocked = 0; + } + assert( szRegion==pShmNode->szRegion || pShmNode->nRegion==0 ); + assert( pShmNode->pInode==pDbFd->pInode ); + assert( pShmNode->hShm>=0 || pDbFd->pInode->bProcessLock==1 ); + assert( pShmNode->hShm<0 || pDbFd->pInode->bProcessLock==0 ); + + /* Minimum number of regions required to be mapped. */ + nReqRegion = ((iRegion+nShmPerMap) / nShmPerMap) * nShmPerMap; + + if( pShmNode->nRegionszRegion = szRegion; + + if( pShmNode->hShm>=0 ){ + /* The requested region is not mapped into this processes address space. + ** Check to see if it has been allocated (i.e. if the wal-index file is + ** large enough to contain the requested region). + */ + if( osFstat(pShmNode->hShm, &sStat) ){ + rc = SQLITE_IOERR_SHMSIZE; + goto shmpage_out; + } + + if( sStat.st_sizehShm, iPg*pgsz + pgsz-1,"",1,&x)!=1 ){ + const char *zFile = pShmNode->zFilename; + rc = unixLogError(SQLITE_IOERR_SHMSIZE, "write", zFile); + goto shmpage_out; + } + } + } + } + } + + /* Map the requested memory region into this processes address space. */ + apNew = (char **)sqlite3_realloc( + pShmNode->apRegion, nReqRegion*sizeof(char *) + ); + if( !apNew ){ + rc = SQLITE_IOERR_NOMEM_BKPT; + goto shmpage_out; + } + pShmNode->apRegion = apNew; + while( pShmNode->nRegionhShm>=0 ){ + pMem = osMmap(0, nMap, + pShmNode->isReadonly ? PROT_READ : PROT_READ|PROT_WRITE, + MAP_SHARED, pShmNode->hShm, szRegion*(i64)pShmNode->nRegion + ); + if( pMem==MAP_FAILED ){ + rc = unixLogError(SQLITE_IOERR_SHMMAP, "mmap", pShmNode->zFilename); + goto shmpage_out; + } + }else{ + pMem = sqlite3_malloc64(nMap); + if( pMem==0 ){ + rc = SQLITE_NOMEM_BKPT; + goto shmpage_out; + } + memset(pMem, 0, nMap); + } + + for(i=0; iapRegion[pShmNode->nRegion+i] = &((char*)pMem)[szRegion*i]; + } + pShmNode->nRegion += nShmPerMap; + } + } + +shmpage_out: + if( pShmNode->nRegion>iRegion ){ + *pp = pShmNode->apRegion[iRegion]; + }else{ + *pp = 0; + } + if( pShmNode->isReadonly && rc==SQLITE_OK ) rc = SQLITE_READONLY; + sqlite3_mutex_leave(pShmNode->pShmMutex); + return rc; +} + +/* +** Change the lock state for a shared-memory segment. +** +** Note that the relationship between SHAREd and EXCLUSIVE locks is a little +** different here than in posix. In xShmLock(), one can go from unlocked +** to shared and back or from unlocked to exclusive and back. But one may +** not go from shared to exclusive or from exclusive to shared. +*/ +static int unixShmLock( + sqlite3_file *fd, /* Database file holding the shared memory */ + int ofst, /* First lock to acquire or release */ + int n, /* Number of locks to acquire or release */ + int flags /* What to do with the lock */ +){ + unixFile *pDbFd = (unixFile*)fd; /* Connection holding shared memory */ + unixShm *p = pDbFd->pShm; /* The shared memory being locked */ + unixShm *pX; /* For looping over all siblings */ + unixShmNode *pShmNode = p->pShmNode; /* The underlying file iNode */ + int rc = SQLITE_OK; /* Result code */ + u16 mask; /* Mask of locks to take or release */ + + assert( pShmNode==pDbFd->pInode->pShmNode ); + assert( pShmNode->pInode==pDbFd->pInode ); + assert( ofst>=0 && ofst+n<=SQLITE_SHM_NLOCK ); + assert( n>=1 ); + assert( flags==(SQLITE_SHM_LOCK | SQLITE_SHM_SHARED) + || flags==(SQLITE_SHM_LOCK | SQLITE_SHM_EXCLUSIVE) + || flags==(SQLITE_SHM_UNLOCK | SQLITE_SHM_SHARED) + || flags==(SQLITE_SHM_UNLOCK | SQLITE_SHM_EXCLUSIVE) ); + assert( n==1 || (flags & SQLITE_SHM_EXCLUSIVE)!=0 ); + assert( pShmNode->hShm>=0 || pDbFd->pInode->bProcessLock==1 ); + assert( pShmNode->hShm<0 || pDbFd->pInode->bProcessLock==0 ); + + mask = (1<<(ofst+n)) - (1<1 || mask==(1<pShmMutex); + if( flags & SQLITE_SHM_UNLOCK ){ + u16 allMask = 0; /* Mask of locks held by siblings */ + + /* See if any siblings hold this same lock */ + for(pX=pShmNode->pFirst; pX; pX=pX->pNext){ + if( pX==p ) continue; + assert( (pX->exclMask & (p->exclMask|p->sharedMask))==0 ); + allMask |= pX->sharedMask; + } + + /* Unlock the system-level locks */ + if( (mask & allMask)==0 ){ + rc = unixShmSystemLock(pDbFd, F_UNLCK, ofst+UNIX_SHM_BASE, n); + }else{ + rc = SQLITE_OK; + } + + /* Undo the local locks */ + if( rc==SQLITE_OK ){ + p->exclMask &= ~mask; + p->sharedMask &= ~mask; + } + }else if( flags & SQLITE_SHM_SHARED ){ + u16 allShared = 0; /* Union of locks held by connections other than "p" */ + + /* Find out which shared locks are already held by sibling connections. + ** If any sibling already holds an exclusive lock, go ahead and return + ** SQLITE_BUSY. + */ + for(pX=pShmNode->pFirst; pX; pX=pX->pNext){ + if( (pX->exclMask & mask)!=0 ){ + rc = SQLITE_BUSY; + break; + } + allShared |= pX->sharedMask; + } + + /* Get shared locks at the system level, if necessary */ + if( rc==SQLITE_OK ){ + if( (allShared & mask)==0 ){ + rc = unixShmSystemLock(pDbFd, F_RDLCK, ofst+UNIX_SHM_BASE, n); + }else{ + rc = SQLITE_OK; + } + } + + /* Get the local shared locks */ + if( rc==SQLITE_OK ){ + p->sharedMask |= mask; + } + }else{ + /* Make sure no sibling connections hold locks that will block this + ** lock. If any do, return SQLITE_BUSY right away. + */ + for(pX=pShmNode->pFirst; pX; pX=pX->pNext){ + if( (pX->exclMask & mask)!=0 || (pX->sharedMask & mask)!=0 ){ + rc = SQLITE_BUSY; + break; + } + } + + /* Get the exclusive locks at the system level. Then if successful + ** also mark the local connection as being locked. + */ + if( rc==SQLITE_OK ){ + rc = unixShmSystemLock(pDbFd, F_WRLCK, ofst+UNIX_SHM_BASE, n); + if( rc==SQLITE_OK ){ + assert( (p->sharedMask & mask)==0 ); + p->exclMask |= mask; + } + } + } + sqlite3_mutex_leave(pShmNode->pShmMutex); + OSTRACE(("SHM-LOCK shmid-%d, pid-%d got %03x,%03x\n", + p->id, osGetpid(0), p->sharedMask, p->exclMask)); + return rc; +} + +/* +** Implement a memory barrier or memory fence on shared memory. +** +** All loads and stores begun before the barrier must complete before +** any load or store begun after the barrier. +*/ +static void unixShmBarrier( + sqlite3_file *fd /* Database file holding the shared memory */ +){ + UNUSED_PARAMETER(fd); + sqlite3MemoryBarrier(); /* compiler-defined memory barrier */ + assert( fd->pMethods->xLock==nolockLock + || unixFileMutexNotheld((unixFile*)fd) + ); + unixEnterMutex(); /* Also mutex, for redundancy */ + unixLeaveMutex(); +} + +/* +** Close a connection to shared-memory. Delete the underlying +** storage if deleteFlag is true. +** +** If there is no shared memory associated with the connection then this +** routine is a harmless no-op. +*/ +static int unixShmUnmap( + sqlite3_file *fd, /* The underlying database file */ + int deleteFlag /* Delete shared-memory if true */ +){ + unixShm *p; /* The connection to be closed */ + unixShmNode *pShmNode; /* The underlying shared-memory file */ + unixShm **pp; /* For looping over sibling connections */ + unixFile *pDbFd; /* The underlying database file */ + + pDbFd = (unixFile*)fd; + p = pDbFd->pShm; + if( p==0 ) return SQLITE_OK; + pShmNode = p->pShmNode; + + assert( pShmNode==pDbFd->pInode->pShmNode ); + assert( pShmNode->pInode==pDbFd->pInode ); + + /* Remove connection p from the set of connections associated + ** with pShmNode */ + sqlite3_mutex_enter(pShmNode->pShmMutex); + for(pp=&pShmNode->pFirst; (*pp)!=p; pp = &(*pp)->pNext){} + *pp = p->pNext; + + /* Free the connection p */ + sqlite3_free(p); + pDbFd->pShm = 0; + sqlite3_mutex_leave(pShmNode->pShmMutex); + + /* If pShmNode->nRef has reached 0, then close the underlying + ** shared-memory file, too */ + assert( unixFileMutexNotheld(pDbFd) ); + unixEnterMutex(); + assert( pShmNode->nRef>0 ); + pShmNode->nRef--; + if( pShmNode->nRef==0 ){ + if( deleteFlag && pShmNode->hShm>=0 ){ + osUnlink(pShmNode->zFilename); + } + unixShmPurge(pDbFd); + } + unixLeaveMutex(); + + return SQLITE_OK; +} + + +#else +# define unixShmMap 0 +# define unixShmLock 0 +# define unixShmBarrier 0 +# define unixShmUnmap 0 +#endif /* #ifndef SQLITE_OMIT_WAL */ + +#if SQLITE_MAX_MMAP_SIZE>0 +/* +** If it is currently memory mapped, unmap file pFd. +*/ +static void unixUnmapfile(unixFile *pFd){ + assert( pFd->nFetchOut==0 ); + if( pFd->pMapRegion ){ + osMunmap(pFd->pMapRegion, pFd->mmapSizeActual); + pFd->pMapRegion = 0; + pFd->mmapSize = 0; + pFd->mmapSizeActual = 0; + } +} + +/* +** Attempt to set the size of the memory mapping maintained by file +** descriptor pFd to nNew bytes. Any existing mapping is discarded. +** +** If successful, this function sets the following variables: +** +** unixFile.pMapRegion +** unixFile.mmapSize +** unixFile.mmapSizeActual +** +** If unsuccessful, an error message is logged via sqlite3_log() and +** the three variables above are zeroed. In this case SQLite should +** continue accessing the database using the xRead() and xWrite() +** methods. +*/ +static void unixRemapfile( + unixFile *pFd, /* File descriptor object */ + i64 nNew /* Required mapping size */ +){ + const char *zErr = "mmap"; + int h = pFd->h; /* File descriptor open on db file */ + u8 *pOrig = (u8 *)pFd->pMapRegion; /* Pointer to current file mapping */ + i64 nOrig = pFd->mmapSizeActual; /* Size of pOrig region in bytes */ + u8 *pNew = 0; /* Location of new mapping */ + int flags = PROT_READ; /* Flags to pass to mmap() */ + + assert( pFd->nFetchOut==0 ); + assert( nNew>pFd->mmapSize ); + assert( nNew<=pFd->mmapSizeMax ); + assert( nNew>0 ); + assert( pFd->mmapSizeActual>=pFd->mmapSize ); + assert( MAP_FAILED!=0 ); + +#ifdef SQLITE_MMAP_READWRITE + if( (pFd->ctrlFlags & UNIXFILE_RDONLY)==0 ) flags |= PROT_WRITE; +#endif + + if( pOrig ){ +#if HAVE_MREMAP + i64 nReuse = pFd->mmapSize; +#else + const int szSyspage = osGetpagesize(); + i64 nReuse = (pFd->mmapSize & ~(szSyspage-1)); +#endif + u8 *pReq = &pOrig[nReuse]; + + /* Unmap any pages of the existing mapping that cannot be reused. */ + if( nReuse!=nOrig ){ + osMunmap(pReq, nOrig-nReuse); + } + +#if HAVE_MREMAP + pNew = osMremap(pOrig, nReuse, nNew, MREMAP_MAYMOVE); + zErr = "mremap"; +#else + pNew = osMmap(pReq, nNew-nReuse, flags, MAP_SHARED, h, nReuse); + if( pNew!=MAP_FAILED ){ + if( pNew!=pReq ){ + osMunmap(pNew, nNew - nReuse); + pNew = 0; + }else{ + pNew = pOrig; + } + } +#endif + + /* The attempt to extend the existing mapping failed. Free it. */ + if( pNew==MAP_FAILED || pNew==0 ){ + osMunmap(pOrig, nReuse); + } + } + + /* If pNew is still NULL, try to create an entirely new mapping. */ + if( pNew==0 ){ + pNew = osMmap(0, nNew, flags, MAP_SHARED, h, 0); + } + + if( pNew==MAP_FAILED ){ + pNew = 0; + nNew = 0; + unixLogError(SQLITE_OK, zErr, pFd->zPath); + + /* If the mmap() above failed, assume that all subsequent mmap() calls + ** will probably fail too. Fall back to using xRead/xWrite exclusively + ** in this case. */ + pFd->mmapSizeMax = 0; + } + pFd->pMapRegion = (void *)pNew; + pFd->mmapSize = pFd->mmapSizeActual = nNew; +} + +/* +** Memory map or remap the file opened by file-descriptor pFd (if the file +** is already mapped, the existing mapping is replaced by the new). Or, if +** there already exists a mapping for this file, and there are still +** outstanding xFetch() references to it, this function is a no-op. +** +** If parameter nByte is non-negative, then it is the requested size of +** the mapping to create. Otherwise, if nByte is less than zero, then the +** requested size is the size of the file on disk. The actual size of the +** created mapping is either the requested size or the value configured +** using SQLITE_FCNTL_MMAP_LIMIT, whichever is smaller. +** +** SQLITE_OK is returned if no error occurs (even if the mapping is not +** recreated as a result of outstanding references) or an SQLite error +** code otherwise. +*/ +static int unixMapfile(unixFile *pFd, i64 nMap){ + assert( nMap>=0 || pFd->nFetchOut==0 ); + assert( nMap>0 || (pFd->mmapSize==0 && pFd->pMapRegion==0) ); + if( pFd->nFetchOut>0 ) return SQLITE_OK; + + if( nMap<0 ){ + struct stat statbuf; /* Low-level file information */ + if( osFstat(pFd->h, &statbuf) ){ + return SQLITE_IOERR_FSTAT; + } + nMap = statbuf.st_size; + } + if( nMap>pFd->mmapSizeMax ){ + nMap = pFd->mmapSizeMax; + } + + assert( nMap>0 || (pFd->mmapSize==0 && pFd->pMapRegion==0) ); + if( nMap!=pFd->mmapSize ){ + unixRemapfile(pFd, nMap); + } + + return SQLITE_OK; +} +#endif /* SQLITE_MAX_MMAP_SIZE>0 */ + +/* +** If possible, return a pointer to a mapping of file fd starting at offset +** iOff. The mapping must be valid for at least nAmt bytes. +** +** If such a pointer can be obtained, store it in *pp and return SQLITE_OK. +** Or, if one cannot but no error occurs, set *pp to 0 and return SQLITE_OK. +** Finally, if an error does occur, return an SQLite error code. The final +** value of *pp is undefined in this case. +** +** If this function does return a pointer, the caller must eventually +** release the reference by calling unixUnfetch(). +*/ +static int unixFetch(sqlite3_file *fd, i64 iOff, int nAmt, void **pp){ +#if SQLITE_MAX_MMAP_SIZE>0 + unixFile *pFd = (unixFile *)fd; /* The underlying database file */ +#endif + *pp = 0; + +#if SQLITE_MAX_MMAP_SIZE>0 + if( pFd->mmapSizeMax>0 ){ + if( pFd->pMapRegion==0 ){ + int rc = unixMapfile(pFd, -1); + if( rc!=SQLITE_OK ) return rc; + } + if( pFd->mmapSize >= iOff+nAmt ){ + *pp = &((u8 *)pFd->pMapRegion)[iOff]; + pFd->nFetchOut++; + } + } +#endif + return SQLITE_OK; +} + +/* +** If the third argument is non-NULL, then this function releases a +** reference obtained by an earlier call to unixFetch(). The second +** argument passed to this function must be the same as the corresponding +** argument that was passed to the unixFetch() invocation. +** +** Or, if the third argument is NULL, then this function is being called +** to inform the VFS layer that, according to POSIX, any existing mapping +** may now be invalid and should be unmapped. +*/ +static int unixUnfetch(sqlite3_file *fd, i64 iOff, void *p){ +#if SQLITE_MAX_MMAP_SIZE>0 + unixFile *pFd = (unixFile *)fd; /* The underlying database file */ + UNUSED_PARAMETER(iOff); + + /* If p==0 (unmap the entire file) then there must be no outstanding + ** xFetch references. Or, if p!=0 (meaning it is an xFetch reference), + ** then there must be at least one outstanding. */ + assert( (p==0)==(pFd->nFetchOut==0) ); + + /* If p!=0, it must match the iOff value. */ + assert( p==0 || p==&((u8 *)pFd->pMapRegion)[iOff] ); + + if( p ){ + pFd->nFetchOut--; + }else{ + unixUnmapfile(pFd); + } + + assert( pFd->nFetchOut>=0 ); +#else + UNUSED_PARAMETER(fd); + UNUSED_PARAMETER(p); + UNUSED_PARAMETER(iOff); +#endif + return SQLITE_OK; +} + +/* +** Here ends the implementation of all sqlite3_file methods. +** +********************** End sqlite3_file Methods ******************************* +******************************************************************************/ + +/* +** This division contains definitions of sqlite3_io_methods objects that +** implement various file locking strategies. It also contains definitions +** of "finder" functions. A finder-function is used to locate the appropriate +** sqlite3_io_methods object for a particular database file. The pAppData +** field of the sqlite3_vfs VFS objects are initialized to be pointers to +** the correct finder-function for that VFS. +** +** Most finder functions return a pointer to a fixed sqlite3_io_methods +** object. The only interesting finder-function is autolockIoFinder, which +** looks at the filesystem type and tries to guess the best locking +** strategy from that. +** +** For finder-function F, two objects are created: +** +** (1) The real finder-function named "FImpt()". +** +** (2) A constant pointer to this function named just "F". +** +** +** A pointer to the F pointer is used as the pAppData value for VFS +** objects. We have to do this instead of letting pAppData point +** directly at the finder-function since C90 rules prevent a void* +** from be cast into a function pointer. +** +** +** Each instance of this macro generates two objects: +** +** * A constant sqlite3_io_methods object call METHOD that has locking +** methods CLOSE, LOCK, UNLOCK, CKRESLOCK. +** +** * An I/O method finder function called FINDER that returns a pointer +** to the METHOD object in the previous bullet. +*/ +#define IOMETHODS(FINDER,METHOD,VERSION,CLOSE,LOCK,UNLOCK,CKLOCK,SHMMAP) \ +static const sqlite3_io_methods METHOD = { \ + VERSION, /* iVersion */ \ + CLOSE, /* xClose */ \ + unixRead, /* xRead */ \ + unixWrite, /* xWrite */ \ + unixTruncate, /* xTruncate */ \ + unixSync, /* xSync */ \ + unixFileSize, /* xFileSize */ \ + LOCK, /* xLock */ \ + UNLOCK, /* xUnlock */ \ + CKLOCK, /* xCheckReservedLock */ \ + unixFileControl, /* xFileControl */ \ + unixSectorSize, /* xSectorSize */ \ + unixDeviceCharacteristics, /* xDeviceCapabilities */ \ + SHMMAP, /* xShmMap */ \ + unixShmLock, /* xShmLock */ \ + unixShmBarrier, /* xShmBarrier */ \ + unixShmUnmap, /* xShmUnmap */ \ + unixFetch, /* xFetch */ \ + unixUnfetch, /* xUnfetch */ \ +}; \ +static const sqlite3_io_methods *FINDER##Impl(const char *z, unixFile *p){ \ + UNUSED_PARAMETER(z); UNUSED_PARAMETER(p); \ + return &METHOD; \ +} \ +static const sqlite3_io_methods *(*const FINDER)(const char*,unixFile *p) \ + = FINDER##Impl; + +/* +** Here are all of the sqlite3_io_methods objects for each of the +** locking strategies. Functions that return pointers to these methods +** are also created. +*/ +IOMETHODS( + posixIoFinder, /* Finder function name */ + posixIoMethods, /* sqlite3_io_methods object name */ + 3, /* shared memory and mmap are enabled */ + unixClose, /* xClose method */ + unixLock, /* xLock method */ + unixUnlock, /* xUnlock method */ + unixCheckReservedLock, /* xCheckReservedLock method */ + unixShmMap /* xShmMap method */ +) +IOMETHODS( + nolockIoFinder, /* Finder function name */ + nolockIoMethods, /* sqlite3_io_methods object name */ + 3, /* shared memory and mmap are enabled */ + nolockClose, /* xClose method */ + nolockLock, /* xLock method */ + nolockUnlock, /* xUnlock method */ + nolockCheckReservedLock, /* xCheckReservedLock method */ + 0 /* xShmMap method */ +) +IOMETHODS( + dotlockIoFinder, /* Finder function name */ + dotlockIoMethods, /* sqlite3_io_methods object name */ + 1, /* shared memory is disabled */ + dotlockClose, /* xClose method */ + dotlockLock, /* xLock method */ + dotlockUnlock, /* xUnlock method */ + dotlockCheckReservedLock, /* xCheckReservedLock method */ + 0 /* xShmMap method */ +) + +#if SQLITE_ENABLE_LOCKING_STYLE +IOMETHODS( + flockIoFinder, /* Finder function name */ + flockIoMethods, /* sqlite3_io_methods object name */ + 1, /* shared memory is disabled */ + flockClose, /* xClose method */ + flockLock, /* xLock method */ + flockUnlock, /* xUnlock method */ + flockCheckReservedLock, /* xCheckReservedLock method */ + 0 /* xShmMap method */ +) +#endif + +#if OS_VXWORKS +IOMETHODS( + semIoFinder, /* Finder function name */ + semIoMethods, /* sqlite3_io_methods object name */ + 1, /* shared memory is disabled */ + semXClose, /* xClose method */ + semXLock, /* xLock method */ + semXUnlock, /* xUnlock method */ + semXCheckReservedLock, /* xCheckReservedLock method */ + 0 /* xShmMap method */ +) +#endif + +#if defined(__APPLE__) && SQLITE_ENABLE_LOCKING_STYLE +IOMETHODS( + afpIoFinder, /* Finder function name */ + afpIoMethods, /* sqlite3_io_methods object name */ + 1, /* shared memory is disabled */ + afpClose, /* xClose method */ + afpLock, /* xLock method */ + afpUnlock, /* xUnlock method */ + afpCheckReservedLock, /* xCheckReservedLock method */ + 0 /* xShmMap method */ +) +#endif + +/* +** The proxy locking method is a "super-method" in the sense that it +** opens secondary file descriptors for the conch and lock files and +** it uses proxy, dot-file, AFP, and flock() locking methods on those +** secondary files. For this reason, the division that implements +** proxy locking is located much further down in the file. But we need +** to go ahead and define the sqlite3_io_methods and finder function +** for proxy locking here. So we forward declare the I/O methods. +*/ +#if defined(__APPLE__) && SQLITE_ENABLE_LOCKING_STYLE +static int proxyClose(sqlite3_file*); +static int proxyLock(sqlite3_file*, int); +static int proxyUnlock(sqlite3_file*, int); +static int proxyCheckReservedLock(sqlite3_file*, int*); +IOMETHODS( + proxyIoFinder, /* Finder function name */ + proxyIoMethods, /* sqlite3_io_methods object name */ + 1, /* shared memory is disabled */ + proxyClose, /* xClose method */ + proxyLock, /* xLock method */ + proxyUnlock, /* xUnlock method */ + proxyCheckReservedLock, /* xCheckReservedLock method */ + 0 /* xShmMap method */ +) +#endif + +/* nfs lockd on OSX 10.3+ doesn't clear write locks when a read lock is set */ +#if defined(__APPLE__) && SQLITE_ENABLE_LOCKING_STYLE +IOMETHODS( + nfsIoFinder, /* Finder function name */ + nfsIoMethods, /* sqlite3_io_methods object name */ + 1, /* shared memory is disabled */ + unixClose, /* xClose method */ + unixLock, /* xLock method */ + nfsUnlock, /* xUnlock method */ + unixCheckReservedLock, /* xCheckReservedLock method */ + 0 /* xShmMap method */ +) +#endif + +#if defined(__APPLE__) && SQLITE_ENABLE_LOCKING_STYLE +/* +** This "finder" function attempts to determine the best locking strategy +** for the database file "filePath". It then returns the sqlite3_io_methods +** object that implements that strategy. +** +** This is for MacOSX only. +*/ +static const sqlite3_io_methods *autolockIoFinderImpl( + const char *filePath, /* name of the database file */ + unixFile *pNew /* open file object for the database file */ +){ + static const struct Mapping { + const char *zFilesystem; /* Filesystem type name */ + const sqlite3_io_methods *pMethods; /* Appropriate locking method */ + } aMap[] = { + { "hfs", &posixIoMethods }, + { "ufs", &posixIoMethods }, + { "afpfs", &afpIoMethods }, + { "smbfs", &afpIoMethods }, + { "webdav", &nolockIoMethods }, + { 0, 0 } + }; + int i; + struct statfs fsInfo; + struct flock lockInfo; + + if( !filePath ){ + /* If filePath==NULL that means we are dealing with a transient file + ** that does not need to be locked. */ + return &nolockIoMethods; + } + if( statfs(filePath, &fsInfo) != -1 ){ + if( fsInfo.f_flags & MNT_RDONLY ){ + return &nolockIoMethods; + } + for(i=0; aMap[i].zFilesystem; i++){ + if( strcmp(fsInfo.f_fstypename, aMap[i].zFilesystem)==0 ){ + return aMap[i].pMethods; + } + } + } + + /* Default case. Handles, amongst others, "nfs". + ** Test byte-range lock using fcntl(). If the call succeeds, + ** assume that the file-system supports POSIX style locks. + */ + lockInfo.l_len = 1; + lockInfo.l_start = 0; + lockInfo.l_whence = SEEK_SET; + lockInfo.l_type = F_RDLCK; + if( osFcntl(pNew->h, F_GETLK, &lockInfo)!=-1 ) { + if( strcmp(fsInfo.f_fstypename, "nfs")==0 ){ + return &nfsIoMethods; + } else { + return &posixIoMethods; + } + }else{ + return &dotlockIoMethods; + } +} +static const sqlite3_io_methods + *(*const autolockIoFinder)(const char*,unixFile*) = autolockIoFinderImpl; + +#endif /* defined(__APPLE__) && SQLITE_ENABLE_LOCKING_STYLE */ + +#if OS_VXWORKS +/* +** This "finder" function for VxWorks checks to see if posix advisory +** locking works. If it does, then that is what is used. If it does not +** work, then fallback to named semaphore locking. +*/ +static const sqlite3_io_methods *vxworksIoFinderImpl( + const char *filePath, /* name of the database file */ + unixFile *pNew /* the open file object */ +){ + struct flock lockInfo; + + if( !filePath ){ + /* If filePath==NULL that means we are dealing with a transient file + ** that does not need to be locked. */ + return &nolockIoMethods; + } + + /* Test if fcntl() is supported and use POSIX style locks. + ** Otherwise fall back to the named semaphore method. + */ + lockInfo.l_len = 1; + lockInfo.l_start = 0; + lockInfo.l_whence = SEEK_SET; + lockInfo.l_type = F_RDLCK; + if( osFcntl(pNew->h, F_GETLK, &lockInfo)!=-1 ) { + return &posixIoMethods; + }else{ + return &semIoMethods; + } +} +static const sqlite3_io_methods + *(*const vxworksIoFinder)(const char*,unixFile*) = vxworksIoFinderImpl; + +#endif /* OS_VXWORKS */ + +/* +** An abstract type for a pointer to an IO method finder function: +*/ +typedef const sqlite3_io_methods *(*finder_type)(const char*,unixFile*); + + +/**************************************************************************** +**************************** sqlite3_vfs methods **************************** +** +** This division contains the implementation of methods on the +** sqlite3_vfs object. +*/ + +/* +** Initialize the contents of the unixFile structure pointed to by pId. +*/ +static int fillInUnixFile( + sqlite3_vfs *pVfs, /* Pointer to vfs object */ + int h, /* Open file descriptor of file being opened */ + sqlite3_file *pId, /* Write to the unixFile structure here */ + const char *zFilename, /* Name of the file being opened */ + int ctrlFlags /* Zero or more UNIXFILE_* values */ +){ + const sqlite3_io_methods *pLockingStyle; + unixFile *pNew = (unixFile *)pId; + int rc = SQLITE_OK; + + assert( pNew->pInode==NULL ); + + /* No locking occurs in temporary files */ + assert( zFilename!=0 || (ctrlFlags & UNIXFILE_NOLOCK)!=0 ); + + OSTRACE(("OPEN %-3d %s\n", h, zFilename)); + pNew->h = h; + pNew->pVfs = pVfs; + pNew->zPath = zFilename; + pNew->ctrlFlags = (u8)ctrlFlags; +#if SQLITE_MAX_MMAP_SIZE>0 + pNew->mmapSizeMax = sqlite3GlobalConfig.szMmap; +#endif + if( sqlite3_uri_boolean(((ctrlFlags & UNIXFILE_URI) ? zFilename : 0), + "psow", SQLITE_POWERSAFE_OVERWRITE) ){ + pNew->ctrlFlags |= UNIXFILE_PSOW; + } + if( strcmp(pVfs->zName,"unix-excl")==0 ){ + pNew->ctrlFlags |= UNIXFILE_EXCL; + } + +#if OS_VXWORKS + pNew->pId = vxworksFindFileId(zFilename); + if( pNew->pId==0 ){ + ctrlFlags |= UNIXFILE_NOLOCK; + rc = SQLITE_NOMEM_BKPT; + } +#endif + + if( ctrlFlags & UNIXFILE_NOLOCK ){ + pLockingStyle = &nolockIoMethods; + }else{ + pLockingStyle = (**(finder_type*)pVfs->pAppData)(zFilename, pNew); +#if SQLITE_ENABLE_LOCKING_STYLE + /* Cache zFilename in the locking context (AFP and dotlock override) for + ** proxyLock activation is possible (remote proxy is based on db name) + ** zFilename remains valid until file is closed, to support */ + pNew->lockingContext = (void*)zFilename; +#endif + } + + if( pLockingStyle == &posixIoMethods +#if defined(__APPLE__) && SQLITE_ENABLE_LOCKING_STYLE + || pLockingStyle == &nfsIoMethods +#endif + ){ + unixEnterMutex(); + rc = findInodeInfo(pNew, &pNew->pInode); + if( rc!=SQLITE_OK ){ + /* If an error occurred in findInodeInfo(), close the file descriptor + ** immediately, before releasing the mutex. findInodeInfo() may fail + ** in two scenarios: + ** + ** (a) A call to fstat() failed. + ** (b) A malloc failed. + ** + ** Scenario (b) may only occur if the process is holding no other + ** file descriptors open on the same file. If there were other file + ** descriptors on this file, then no malloc would be required by + ** findInodeInfo(). If this is the case, it is quite safe to close + ** handle h - as it is guaranteed that no posix locks will be released + ** by doing so. + ** + ** If scenario (a) caused the error then things are not so safe. The + ** implicit assumption here is that if fstat() fails, things are in + ** such bad shape that dropping a lock or two doesn't matter much. + */ + robust_close(pNew, h, __LINE__); + h = -1; + } + unixLeaveMutex(); + } + +#if SQLITE_ENABLE_LOCKING_STYLE && defined(__APPLE__) + else if( pLockingStyle == &afpIoMethods ){ + /* AFP locking uses the file path so it needs to be included in + ** the afpLockingContext. + */ + afpLockingContext *pCtx; + pNew->lockingContext = pCtx = sqlite3_malloc64( sizeof(*pCtx) ); + if( pCtx==0 ){ + rc = SQLITE_NOMEM_BKPT; + }else{ + /* NB: zFilename exists and remains valid until the file is closed + ** according to requirement F11141. So we do not need to make a + ** copy of the filename. */ + pCtx->dbPath = zFilename; + pCtx->reserved = 0; + srandomdev(); + unixEnterMutex(); + rc = findInodeInfo(pNew, &pNew->pInode); + if( rc!=SQLITE_OK ){ + sqlite3_free(pNew->lockingContext); + robust_close(pNew, h, __LINE__); + h = -1; + } + unixLeaveMutex(); + } + } +#endif + + else if( pLockingStyle == &dotlockIoMethods ){ + /* Dotfile locking uses the file path so it needs to be included in + ** the dotlockLockingContext + */ + char *zLockFile; + int nFilename; + assert( zFilename!=0 ); + nFilename = (int)strlen(zFilename) + 6; + zLockFile = (char *)sqlite3_malloc64(nFilename); + if( zLockFile==0 ){ + rc = SQLITE_NOMEM_BKPT; + }else{ + sqlite3_snprintf(nFilename, zLockFile, "%s" DOTLOCK_SUFFIX, zFilename); + } + pNew->lockingContext = zLockFile; + } + +#if OS_VXWORKS + else if( pLockingStyle == &semIoMethods ){ + /* Named semaphore locking uses the file path so it needs to be + ** included in the semLockingContext + */ + unixEnterMutex(); + rc = findInodeInfo(pNew, &pNew->pInode); + if( (rc==SQLITE_OK) && (pNew->pInode->pSem==NULL) ){ + char *zSemName = pNew->pInode->aSemName; + int n; + sqlite3_snprintf(MAX_PATHNAME, zSemName, "/%s.sem", + pNew->pId->zCanonicalName); + for( n=1; zSemName[n]; n++ ) + if( zSemName[n]=='/' ) zSemName[n] = '_'; + pNew->pInode->pSem = sem_open(zSemName, O_CREAT, 0666, 1); + if( pNew->pInode->pSem == SEM_FAILED ){ + rc = SQLITE_NOMEM_BKPT; + pNew->pInode->aSemName[0] = '\0'; + } + } + unixLeaveMutex(); + } +#endif + + storeLastErrno(pNew, 0); +#if OS_VXWORKS + if( rc!=SQLITE_OK ){ + if( h>=0 ) robust_close(pNew, h, __LINE__); + h = -1; + osUnlink(zFilename); + pNew->ctrlFlags |= UNIXFILE_DELETE; + } +#endif + if( rc!=SQLITE_OK ){ + if( h>=0 ) robust_close(pNew, h, __LINE__); + }else{ + pNew->pMethod = pLockingStyle; + OpenCounter(+1); + verifyDbFile(pNew); + } + return rc; +} + +/* +** Return the name of a directory in which to put temporary files. +** If no suitable temporary file directory can be found, return NULL. +*/ +static const char *unixTempFileDir(void){ + static const char *azDirs[] = { + 0, + 0, + "/var/tmp", + "/usr/tmp", + "/tmp", + "." + }; + unsigned int i = 0; + struct stat buf; + const char *zDir = sqlite3_temp_directory; + + if( !azDirs[0] ) azDirs[0] = getenv("SQLITE_TMPDIR"); + if( !azDirs[1] ) azDirs[1] = getenv("TMPDIR"); + while(1){ + if( zDir!=0 + && osStat(zDir, &buf)==0 + && S_ISDIR(buf.st_mode) + && osAccess(zDir, 03)==0 + ){ + return zDir; + } + if( i>=sizeof(azDirs)/sizeof(azDirs[0]) ) break; + zDir = azDirs[i++]; + } + return 0; +} + +/* +** Create a temporary file name in zBuf. zBuf must be allocated +** by the calling process and must be big enough to hold at least +** pVfs->mxPathname bytes. +*/ +static int unixGetTempname(int nBuf, char *zBuf){ + const char *zDir; + int iLimit = 0; + + /* It's odd to simulate an io-error here, but really this is just + ** using the io-error infrastructure to test that SQLite handles this + ** function failing. + */ + zBuf[0] = 0; + SimulateIOError( return SQLITE_IOERR ); + + zDir = unixTempFileDir(); + if( zDir==0 ) return SQLITE_IOERR_GETTEMPPATH; + do{ + u64 r; + sqlite3_randomness(sizeof(r), &r); + assert( nBuf>2 ); + zBuf[nBuf-2] = 0; + sqlite3_snprintf(nBuf, zBuf, "%s/"SQLITE_TEMP_FILE_PREFIX"%llx%c", + zDir, r, 0); + if( zBuf[nBuf-2]!=0 || (iLimit++)>10 ) return SQLITE_ERROR; + }while( osAccess(zBuf,0)==0 ); + return SQLITE_OK; +} + +#if SQLITE_ENABLE_LOCKING_STYLE && defined(__APPLE__) +/* +** Routine to transform a unixFile into a proxy-locking unixFile. +** Implementation in the proxy-lock division, but used by unixOpen() +** if SQLITE_PREFER_PROXY_LOCKING is defined. +*/ +static int proxyTransformUnixFile(unixFile*, const char*); +#endif + +/* +** Search for an unused file descriptor that was opened on the database +** file (not a journal or master-journal file) identified by pathname +** zPath with SQLITE_OPEN_XXX flags matching those passed as the second +** argument to this function. +** +** Such a file descriptor may exist if a database connection was closed +** but the associated file descriptor could not be closed because some +** other file descriptor open on the same file is holding a file-lock. +** Refer to comments in the unixClose() function and the lengthy comment +** describing "Posix Advisory Locking" at the start of this file for +** further details. Also, ticket #4018. +** +** If a suitable file descriptor is found, then it is returned. If no +** such file descriptor is located, -1 is returned. +*/ +static UnixUnusedFd *findReusableFd(const char *zPath, int flags){ + UnixUnusedFd *pUnused = 0; + + /* Do not search for an unused file descriptor on vxworks. Not because + ** vxworks would not benefit from the change (it might, we're not sure), + ** but because no way to test it is currently available. It is better + ** not to risk breaking vxworks support for the sake of such an obscure + ** feature. */ +#if !OS_VXWORKS + struct stat sStat; /* Results of stat() call */ + + unixEnterMutex(); + + /* A stat() call may fail for various reasons. If this happens, it is + ** almost certain that an open() call on the same path will also fail. + ** For this reason, if an error occurs in the stat() call here, it is + ** ignored and -1 is returned. The caller will try to open a new file + ** descriptor on the same path, fail, and return an error to SQLite. + ** + ** Even if a subsequent open() call does succeed, the consequences of + ** not searching for a reusable file descriptor are not dire. */ + if( inodeList!=0 && 0==osStat(zPath, &sStat) ){ + unixInodeInfo *pInode; + + pInode = inodeList; + while( pInode && (pInode->fileId.dev!=sStat.st_dev + || pInode->fileId.ino!=(u64)sStat.st_ino) ){ + pInode = pInode->pNext; + } + if( pInode ){ + UnixUnusedFd **pp; + assert( sqlite3_mutex_notheld(pInode->pLockMutex) ); + sqlite3_mutex_enter(pInode->pLockMutex); + flags &= (SQLITE_OPEN_READONLY|SQLITE_OPEN_READWRITE); + for(pp=&pInode->pUnused; *pp && (*pp)->flags!=flags; pp=&((*pp)->pNext)); + pUnused = *pp; + if( pUnused ){ + *pp = pUnused->pNext; + } + sqlite3_mutex_leave(pInode->pLockMutex); + } + } + unixLeaveMutex(); +#endif /* if !OS_VXWORKS */ + return pUnused; +} + +/* +** Find the mode, uid and gid of file zFile. +*/ +static int getFileMode( + const char *zFile, /* File name */ + mode_t *pMode, /* OUT: Permissions of zFile */ + uid_t *pUid, /* OUT: uid of zFile. */ + gid_t *pGid /* OUT: gid of zFile. */ +){ + struct stat sStat; /* Output of stat() on database file */ + int rc = SQLITE_OK; + if( 0==osStat(zFile, &sStat) ){ + *pMode = sStat.st_mode & 0777; + *pUid = sStat.st_uid; + *pGid = sStat.st_gid; + }else{ + rc = SQLITE_IOERR_FSTAT; + } + return rc; +} + +/* +** This function is called by unixOpen() to determine the unix permissions +** to create new files with. If no error occurs, then SQLITE_OK is returned +** and a value suitable for passing as the third argument to open(2) is +** written to *pMode. If an IO error occurs, an SQLite error code is +** returned and the value of *pMode is not modified. +** +** In most cases, this routine sets *pMode to 0, which will become +** an indication to robust_open() to create the file using +** SQLITE_DEFAULT_FILE_PERMISSIONS adjusted by the umask. +** But if the file being opened is a WAL or regular journal file, then +** this function queries the file-system for the permissions on the +** corresponding database file and sets *pMode to this value. Whenever +** possible, WAL and journal files are created using the same permissions +** as the associated database file. +** +** If the SQLITE_ENABLE_8_3_NAMES option is enabled, then the +** original filename is unavailable. But 8_3_NAMES is only used for +** FAT filesystems and permissions do not matter there, so just use +** the default permissions. In 8_3_NAMES mode, leave *pMode set to zero. +*/ +static int findCreateFileMode( + const char *zPath, /* Path of file (possibly) being created */ + int flags, /* Flags passed as 4th argument to xOpen() */ + mode_t *pMode, /* OUT: Permissions to open file with */ + uid_t *pUid, /* OUT: uid to set on the file */ + gid_t *pGid /* OUT: gid to set on the file */ +){ + int rc = SQLITE_OK; /* Return Code */ + *pMode = 0; + *pUid = 0; + *pGid = 0; + if( flags & (SQLITE_OPEN_WAL|SQLITE_OPEN_MAIN_JOURNAL) ){ + char zDb[MAX_PATHNAME+1]; /* Database file path */ + int nDb; /* Number of valid bytes in zDb */ + + /* zPath is a path to a WAL or journal file. The following block derives + ** the path to the associated database file from zPath. This block handles + ** the following naming conventions: + ** + ** "-journal" + ** "-wal" + ** "-journalNN" + ** "-walNN" + ** + ** where NN is a decimal number. The NN naming schemes are + ** used by the test_multiplex.c module. + */ + nDb = sqlite3Strlen30(zPath) - 1; + while( zPath[nDb]!='-' ){ + /* In normal operation, the journal file name will always contain + ** a '-' character. However in 8+3 filename mode, or if a corrupt + ** rollback journal specifies a master journal with a goofy name, then + ** the '-' might be missing. */ + if( nDb==0 || zPath[nDb]=='.' ) return SQLITE_OK; + nDb--; + } + memcpy(zDb, zPath, nDb); + zDb[nDb] = '\0'; + + rc = getFileMode(zDb, pMode, pUid, pGid); + }else if( flags & SQLITE_OPEN_DELETEONCLOSE ){ + *pMode = 0600; + }else if( flags & SQLITE_OPEN_URI ){ + /* If this is a main database file and the file was opened using a URI + ** filename, check for the "modeof" parameter. If present, interpret + ** its value as a filename and try to copy the mode, uid and gid from + ** that file. */ + const char *z = sqlite3_uri_parameter(zPath, "modeof"); + if( z ){ + rc = getFileMode(z, pMode, pUid, pGid); + } + } + return rc; +} + +/* +** Open the file zPath. +** +** Previously, the SQLite OS layer used three functions in place of this +** one: +** +** sqlite3OsOpenReadWrite(); +** sqlite3OsOpenReadOnly(); +** sqlite3OsOpenExclusive(); +** +** These calls correspond to the following combinations of flags: +** +** ReadWrite() -> (READWRITE | CREATE) +** ReadOnly() -> (READONLY) +** OpenExclusive() -> (READWRITE | CREATE | EXCLUSIVE) +** +** The old OpenExclusive() accepted a boolean argument - "delFlag". If +** true, the file was configured to be automatically deleted when the +** file handle closed. To achieve the same effect using this new +** interface, add the DELETEONCLOSE flag to those specified above for +** OpenExclusive(). +*/ +static int unixOpen( + sqlite3_vfs *pVfs, /* The VFS for which this is the xOpen method */ + const char *zPath, /* Pathname of file to be opened */ + sqlite3_file *pFile, /* The file descriptor to be filled in */ + int flags, /* Input flags to control the opening */ + int *pOutFlags /* Output flags returned to SQLite core */ +){ + unixFile *p = (unixFile *)pFile; + int fd = -1; /* File descriptor returned by open() */ + int openFlags = 0; /* Flags to pass to open() */ + int eType = flags&0xFFFFFF00; /* Type of file to open */ + int noLock; /* True to omit locking primitives */ + int rc = SQLITE_OK; /* Function Return Code */ + int ctrlFlags = 0; /* UNIXFILE_* flags */ + + int isExclusive = (flags & SQLITE_OPEN_EXCLUSIVE); + int isDelete = (flags & SQLITE_OPEN_DELETEONCLOSE); + int isCreate = (flags & SQLITE_OPEN_CREATE); + int isReadonly = (flags & SQLITE_OPEN_READONLY); + int isReadWrite = (flags & SQLITE_OPEN_READWRITE); +#if SQLITE_ENABLE_LOCKING_STYLE + int isAutoProxy = (flags & SQLITE_OPEN_AUTOPROXY); +#endif +#if defined(__APPLE__) || SQLITE_ENABLE_LOCKING_STYLE + struct statfs fsInfo; +#endif + + /* If creating a master or main-file journal, this function will open + ** a file-descriptor on the directory too. The first time unixSync() + ** is called the directory file descriptor will be fsync()ed and close()d. + */ + int isNewJrnl = (isCreate && ( + eType==SQLITE_OPEN_MASTER_JOURNAL + || eType==SQLITE_OPEN_MAIN_JOURNAL + || eType==SQLITE_OPEN_WAL + )); + + /* If argument zPath is a NULL pointer, this function is required to open + ** a temporary file. Use this buffer to store the file name in. + */ + char zTmpname[MAX_PATHNAME+2]; + const char *zName = zPath; + + /* Check the following statements are true: + ** + ** (a) Exactly one of the READWRITE and READONLY flags must be set, and + ** (b) if CREATE is set, then READWRITE must also be set, and + ** (c) if EXCLUSIVE is set, then CREATE must also be set. + ** (d) if DELETEONCLOSE is set, then CREATE must also be set. + */ + assert((isReadonly==0 || isReadWrite==0) && (isReadWrite || isReadonly)); + assert(isCreate==0 || isReadWrite); + assert(isExclusive==0 || isCreate); + assert(isDelete==0 || isCreate); + + /* The main DB, main journal, WAL file and master journal are never + ** automatically deleted. Nor are they ever temporary files. */ + assert( (!isDelete && zName) || eType!=SQLITE_OPEN_MAIN_DB ); + assert( (!isDelete && zName) || eType!=SQLITE_OPEN_MAIN_JOURNAL ); + assert( (!isDelete && zName) || eType!=SQLITE_OPEN_MASTER_JOURNAL ); + assert( (!isDelete && zName) || eType!=SQLITE_OPEN_WAL ); + + /* Assert that the upper layer has set one of the "file-type" flags. */ + assert( eType==SQLITE_OPEN_MAIN_DB || eType==SQLITE_OPEN_TEMP_DB + || eType==SQLITE_OPEN_MAIN_JOURNAL || eType==SQLITE_OPEN_TEMP_JOURNAL + || eType==SQLITE_OPEN_SUBJOURNAL || eType==SQLITE_OPEN_MASTER_JOURNAL + || eType==SQLITE_OPEN_TRANSIENT_DB || eType==SQLITE_OPEN_WAL + ); + + /* Detect a pid change and reset the PRNG. There is a race condition + ** here such that two or more threads all trying to open databases at + ** the same instant might all reset the PRNG. But multiple resets + ** are harmless. + */ + if( randomnessPid!=osGetpid(0) ){ + randomnessPid = osGetpid(0); + sqlite3_randomness(0,0); + } + memset(p, 0, sizeof(unixFile)); + + if( eType==SQLITE_OPEN_MAIN_DB ){ + UnixUnusedFd *pUnused; + pUnused = findReusableFd(zName, flags); + if( pUnused ){ + fd = pUnused->fd; + }else{ + pUnused = sqlite3_malloc64(sizeof(*pUnused)); + if( !pUnused ){ + return SQLITE_NOMEM_BKPT; + } + } + p->pPreallocatedUnused = pUnused; + + /* Database filenames are double-zero terminated if they are not + ** URIs with parameters. Hence, they can always be passed into + ** sqlite3_uri_parameter(). */ + assert( (flags & SQLITE_OPEN_URI) || zName[strlen(zName)+1]==0 ); + + }else if( !zName ){ + /* If zName is NULL, the upper layer is requesting a temp file. */ + assert(isDelete && !isNewJrnl); + rc = unixGetTempname(pVfs->mxPathname, zTmpname); + if( rc!=SQLITE_OK ){ + return rc; + } + zName = zTmpname; + + /* Generated temporary filenames are always double-zero terminated + ** for use by sqlite3_uri_parameter(). */ + assert( zName[strlen(zName)+1]==0 ); + } + + /* Determine the value of the flags parameter passed to POSIX function + ** open(). These must be calculated even if open() is not called, as + ** they may be stored as part of the file handle and used by the + ** 'conch file' locking functions later on. */ + if( isReadonly ) openFlags |= O_RDONLY; + if( isReadWrite ) openFlags |= O_RDWR; + if( isCreate ) openFlags |= O_CREAT; + if( isExclusive ) openFlags |= (O_EXCL|O_NOFOLLOW); + openFlags |= (O_LARGEFILE|O_BINARY); + + if( fd<0 ){ + mode_t openMode; /* Permissions to create file with */ + uid_t uid; /* Userid for the file */ + gid_t gid; /* Groupid for the file */ + rc = findCreateFileMode(zName, flags, &openMode, &uid, &gid); + if( rc!=SQLITE_OK ){ + assert( !p->pPreallocatedUnused ); + assert( eType==SQLITE_OPEN_WAL || eType==SQLITE_OPEN_MAIN_JOURNAL ); + return rc; + } + fd = robust_open(zName, openFlags, openMode); + OSTRACE(("OPENX %-3d %s 0%o\n", fd, zName, openFlags)); + assert( !isExclusive || (openFlags & O_CREAT)!=0 ); + if( fd<0 ){ + if( isNewJrnl && errno==EACCES && osAccess(zName, F_OK) ){ + /* If unable to create a journal because the directory is not + ** writable, change the error code to indicate that. */ + rc = SQLITE_READONLY_DIRECTORY; + }else if( errno!=EISDIR && isReadWrite ){ + /* Failed to open the file for read/write access. Try read-only. */ + flags &= ~(SQLITE_OPEN_READWRITE|SQLITE_OPEN_CREATE); + openFlags &= ~(O_RDWR|O_CREAT); + flags |= SQLITE_OPEN_READONLY; + openFlags |= O_RDONLY; + isReadonly = 1; + fd = robust_open(zName, openFlags, openMode); + } + } + if( fd<0 ){ + int rc2 = unixLogError(SQLITE_CANTOPEN_BKPT, "open", zName); + if( rc==SQLITE_OK ) rc = rc2; + goto open_finished; + } + + /* The owner of the rollback journal or WAL file should always be the + ** same as the owner of the database file. Try to ensure that this is + ** the case. The chown() system call will be a no-op if the current + ** process lacks root privileges, be we should at least try. Without + ** this step, if a root process opens a database file, it can leave + ** behinds a journal/WAL that is owned by root and hence make the + ** database inaccessible to unprivileged processes. + ** + ** If openMode==0, then that means uid and gid are not set correctly + ** (probably because SQLite is configured to use 8+3 filename mode) and + ** in that case we do not want to attempt the chown(). + */ + if( openMode && (flags & (SQLITE_OPEN_WAL|SQLITE_OPEN_MAIN_JOURNAL))!=0 ){ + robustFchown(fd, uid, gid); + } + } + assert( fd>=0 ); + if( pOutFlags ){ + *pOutFlags = flags; + } + + if( p->pPreallocatedUnused ){ + p->pPreallocatedUnused->fd = fd; + p->pPreallocatedUnused->flags = + flags & (SQLITE_OPEN_READONLY|SQLITE_OPEN_READWRITE); + } + + if( isDelete ){ +#if OS_VXWORKS + zPath = zName; +#elif defined(SQLITE_UNLINK_AFTER_CLOSE) + zPath = sqlite3_mprintf("%s", zName); + if( zPath==0 ){ + robust_close(p, fd, __LINE__); + return SQLITE_NOMEM_BKPT; + } +#else + osUnlink(zName); +#endif + } +#if SQLITE_ENABLE_LOCKING_STYLE + else{ + p->openFlags = openFlags; + } +#endif + +#if defined(__APPLE__) || SQLITE_ENABLE_LOCKING_STYLE + if( fstatfs(fd, &fsInfo) == -1 ){ + storeLastErrno(p, errno); + robust_close(p, fd, __LINE__); + return SQLITE_IOERR_ACCESS; + } + if (0 == strncmp("msdos", fsInfo.f_fstypename, 5)) { + ((unixFile*)pFile)->fsFlags |= SQLITE_FSFLAGS_IS_MSDOS; + } + if (0 == strncmp("exfat", fsInfo.f_fstypename, 5)) { + ((unixFile*)pFile)->fsFlags |= SQLITE_FSFLAGS_IS_MSDOS; + } +#endif + + /* Set up appropriate ctrlFlags */ + if( isDelete ) ctrlFlags |= UNIXFILE_DELETE; + if( isReadonly ) ctrlFlags |= UNIXFILE_RDONLY; + noLock = eType!=SQLITE_OPEN_MAIN_DB; + if( noLock ) ctrlFlags |= UNIXFILE_NOLOCK; + if( isNewJrnl ) ctrlFlags |= UNIXFILE_DIRSYNC; + if( flags & SQLITE_OPEN_URI ) ctrlFlags |= UNIXFILE_URI; + +#if SQLITE_ENABLE_LOCKING_STYLE +#if SQLITE_PREFER_PROXY_LOCKING + isAutoProxy = 1; +#endif + if( isAutoProxy && (zPath!=NULL) && (!noLock) && pVfs->xOpen ){ + char *envforce = getenv("SQLITE_FORCE_PROXY_LOCKING"); + int useProxy = 0; + + /* SQLITE_FORCE_PROXY_LOCKING==1 means force always use proxy, 0 means + ** never use proxy, NULL means use proxy for non-local files only. */ + if( envforce!=NULL ){ + useProxy = atoi(envforce)>0; + }else{ + useProxy = !(fsInfo.f_flags&MNT_LOCAL); + } + if( useProxy ){ + rc = fillInUnixFile(pVfs, fd, pFile, zPath, ctrlFlags); + if( rc==SQLITE_OK ){ + rc = proxyTransformUnixFile((unixFile*)pFile, ":auto:"); + if( rc!=SQLITE_OK ){ + /* Use unixClose to clean up the resources added in fillInUnixFile + ** and clear all the structure's references. Specifically, + ** pFile->pMethods will be NULL so sqlite3OsClose will be a no-op + */ + unixClose(pFile); + return rc; + } + } + goto open_finished; + } + } +#endif + + assert( zPath==0 || zPath[0]=='/' + || eType==SQLITE_OPEN_MASTER_JOURNAL || eType==SQLITE_OPEN_MAIN_JOURNAL + ); + rc = fillInUnixFile(pVfs, fd, pFile, zPath, ctrlFlags); + +open_finished: + if( rc!=SQLITE_OK ){ + sqlite3_free(p->pPreallocatedUnused); + } + return rc; +} + + +/* +** Delete the file at zPath. If the dirSync argument is true, fsync() +** the directory after deleting the file. +*/ +static int unixDelete( + sqlite3_vfs *NotUsed, /* VFS containing this as the xDelete method */ + const char *zPath, /* Name of file to be deleted */ + int dirSync /* If true, fsync() directory after deleting file */ +){ + int rc = SQLITE_OK; + UNUSED_PARAMETER(NotUsed); + SimulateIOError(return SQLITE_IOERR_DELETE); + if( osUnlink(zPath)==(-1) ){ + if( errno==ENOENT +#if OS_VXWORKS + || osAccess(zPath,0)!=0 +#endif + ){ + rc = SQLITE_IOERR_DELETE_NOENT; + }else{ + rc = unixLogError(SQLITE_IOERR_DELETE, "unlink", zPath); + } + return rc; + } +#ifndef SQLITE_DISABLE_DIRSYNC + if( (dirSync & 1)!=0 ){ + int fd; + rc = osOpenDirectory(zPath, &fd); + if( rc==SQLITE_OK ){ + if( full_fsync(fd,0,0) ){ + rc = unixLogError(SQLITE_IOERR_DIR_FSYNC, "fsync", zPath); + } + robust_close(0, fd, __LINE__); + }else{ + assert( rc==SQLITE_CANTOPEN ); + rc = SQLITE_OK; + } + } +#endif + return rc; +} + +/* +** Test the existence of or access permissions of file zPath. The +** test performed depends on the value of flags: +** +** SQLITE_ACCESS_EXISTS: Return 1 if the file exists +** SQLITE_ACCESS_READWRITE: Return 1 if the file is read and writable. +** SQLITE_ACCESS_READONLY: Return 1 if the file is readable. +** +** Otherwise return 0. +*/ +static int unixAccess( + sqlite3_vfs *NotUsed, /* The VFS containing this xAccess method */ + const char *zPath, /* Path of the file to examine */ + int flags, /* What do we want to learn about the zPath file? */ + int *pResOut /* Write result boolean here */ +){ + UNUSED_PARAMETER(NotUsed); + SimulateIOError( return SQLITE_IOERR_ACCESS; ); + assert( pResOut!=0 ); + + /* The spec says there are three possible values for flags. But only + ** two of them are actually used */ + assert( flags==SQLITE_ACCESS_EXISTS || flags==SQLITE_ACCESS_READWRITE ); + + if( flags==SQLITE_ACCESS_EXISTS ){ + struct stat buf; + *pResOut = (0==osStat(zPath, &buf) && buf.st_size>0); + }else{ + *pResOut = osAccess(zPath, W_OK|R_OK)==0; + } + return SQLITE_OK; +} + +/* +** +*/ +static int mkFullPathname( + const char *zPath, /* Input path */ + char *zOut, /* Output buffer */ + int nOut /* Allocated size of buffer zOut */ +){ + int nPath = sqlite3Strlen30(zPath); + int iOff = 0; + if( zPath[0]!='/' ){ + if( osGetcwd(zOut, nOut-2)==0 ){ + return unixLogError(SQLITE_CANTOPEN_BKPT, "getcwd", zPath); + } + iOff = sqlite3Strlen30(zOut); + zOut[iOff++] = '/'; + } + if( (iOff+nPath+1)>nOut ){ + /* SQLite assumes that xFullPathname() nul-terminates the output buffer + ** even if it returns an error. */ + zOut[iOff] = '\0'; + return SQLITE_CANTOPEN_BKPT; + } + sqlite3_snprintf(nOut-iOff, &zOut[iOff], "%s", zPath); + return SQLITE_OK; +} + +/* +** Turn a relative pathname into a full pathname. The relative path +** is stored as a nul-terminated string in the buffer pointed to by +** zPath. +** +** zOut points to a buffer of at least sqlite3_vfs.mxPathname bytes +** (in this case, MAX_PATHNAME bytes). The full-path is written to +** this buffer before returning. +*/ +static int unixFullPathname( + sqlite3_vfs *pVfs, /* Pointer to vfs object */ + const char *zPath, /* Possibly relative input path */ + int nOut, /* Size of output buffer in bytes */ + char *zOut /* Output buffer */ +){ +#if !defined(HAVE_READLINK) || !defined(HAVE_LSTAT) + return mkFullPathname(zPath, zOut, nOut); +#else + int rc = SQLITE_OK; + int nByte; + int nLink = 1; /* Number of symbolic links followed so far */ + const char *zIn = zPath; /* Input path for each iteration of loop */ + char *zDel = 0; + + assert( pVfs->mxPathname==MAX_PATHNAME ); + UNUSED_PARAMETER(pVfs); + + /* It's odd to simulate an io-error here, but really this is just + ** using the io-error infrastructure to test that SQLite handles this + ** function failing. This function could fail if, for example, the + ** current working directory has been unlinked. + */ + SimulateIOError( return SQLITE_ERROR ); + + do { + + /* Call stat() on path zIn. Set bLink to true if the path is a symbolic + ** link, or false otherwise. */ + int bLink = 0; + struct stat buf; + if( osLstat(zIn, &buf)!=0 ){ + if( errno!=ENOENT ){ + rc = unixLogError(SQLITE_CANTOPEN_BKPT, "lstat", zIn); + } + }else{ + bLink = S_ISLNK(buf.st_mode); + } + + if( bLink ){ + if( zDel==0 ){ + zDel = sqlite3_malloc(nOut); + if( zDel==0 ) rc = SQLITE_NOMEM_BKPT; + }else if( ++nLink>SQLITE_MAX_SYMLINKS ){ + rc = SQLITE_CANTOPEN_BKPT; + } + + if( rc==SQLITE_OK ){ + nByte = osReadlink(zIn, zDel, nOut-1); + if( nByte<0 ){ + rc = unixLogError(SQLITE_CANTOPEN_BKPT, "readlink", zIn); + }else{ + if( zDel[0]!='/' ){ + int n; + for(n = sqlite3Strlen30(zIn); n>0 && zIn[n-1]!='/'; n--); + if( nByte+n+1>nOut ){ + rc = SQLITE_CANTOPEN_BKPT; + }else{ + memmove(&zDel[n], zDel, nByte+1); + memcpy(zDel, zIn, n); + nByte += n; + } + } + zDel[nByte] = '\0'; + } + } + + zIn = zDel; + } + + assert( rc!=SQLITE_OK || zIn!=zOut || zIn[0]=='/' ); + if( rc==SQLITE_OK && zIn!=zOut ){ + rc = mkFullPathname(zIn, zOut, nOut); + } + if( bLink==0 ) break; + zIn = zOut; + }while( rc==SQLITE_OK ); + + sqlite3_free(zDel); + return rc; +#endif /* HAVE_READLINK && HAVE_LSTAT */ +} + + +#ifndef SQLITE_OMIT_LOAD_EXTENSION +/* +** Interfaces for opening a shared library, finding entry points +** within the shared library, and closing the shared library. +*/ +#include +static void *unixDlOpen(sqlite3_vfs *NotUsed, const char *zFilename){ + UNUSED_PARAMETER(NotUsed); + return dlopen(zFilename, RTLD_NOW | RTLD_GLOBAL); +} + +/* +** SQLite calls this function immediately after a call to unixDlSym() or +** unixDlOpen() fails (returns a null pointer). If a more detailed error +** message is available, it is written to zBufOut. If no error message +** is available, zBufOut is left unmodified and SQLite uses a default +** error message. +*/ +static void unixDlError(sqlite3_vfs *NotUsed, int nBuf, char *zBufOut){ + const char *zErr; + UNUSED_PARAMETER(NotUsed); + unixEnterMutex(); + zErr = dlerror(); + if( zErr ){ + sqlite3_snprintf(nBuf, zBufOut, "%s", zErr); + } + unixLeaveMutex(); +} +static void (*unixDlSym(sqlite3_vfs *NotUsed, void *p, const char*zSym))(void){ + /* + ** GCC with -pedantic-errors says that C90 does not allow a void* to be + ** cast into a pointer to a function. And yet the library dlsym() routine + ** returns a void* which is really a pointer to a function. So how do we + ** use dlsym() with -pedantic-errors? + ** + ** Variable x below is defined to be a pointer to a function taking + ** parameters void* and const char* and returning a pointer to a function. + ** We initialize x by assigning it a pointer to the dlsym() function. + ** (That assignment requires a cast.) Then we call the function that + ** x points to. + ** + ** This work-around is unlikely to work correctly on any system where + ** you really cannot cast a function pointer into void*. But then, on the + ** other hand, dlsym() will not work on such a system either, so we have + ** not really lost anything. + */ + void (*(*x)(void*,const char*))(void); + UNUSED_PARAMETER(NotUsed); + x = (void(*(*)(void*,const char*))(void))dlsym; + return (*x)(p, zSym); +} +static void unixDlClose(sqlite3_vfs *NotUsed, void *pHandle){ + UNUSED_PARAMETER(NotUsed); + dlclose(pHandle); +} +#else /* if SQLITE_OMIT_LOAD_EXTENSION is defined: */ + #define unixDlOpen 0 + #define unixDlError 0 + #define unixDlSym 0 + #define unixDlClose 0 +#endif + +/* +** Write nBuf bytes of random data to the supplied buffer zBuf. +*/ +static int unixRandomness(sqlite3_vfs *NotUsed, int nBuf, char *zBuf){ + UNUSED_PARAMETER(NotUsed); + assert((size_t)nBuf>=(sizeof(time_t)+sizeof(int))); + + /* We have to initialize zBuf to prevent valgrind from reporting + ** errors. The reports issued by valgrind are incorrect - we would + ** prefer that the randomness be increased by making use of the + ** uninitialized space in zBuf - but valgrind errors tend to worry + ** some users. Rather than argue, it seems easier just to initialize + ** the whole array and silence valgrind, even if that means less randomness + ** in the random seed. + ** + ** When testing, initializing zBuf[] to zero is all we do. That means + ** that we always use the same random number sequence. This makes the + ** tests repeatable. + */ + memset(zBuf, 0, nBuf); + randomnessPid = osGetpid(0); +#if !defined(SQLITE_TEST) && !defined(SQLITE_OMIT_RANDOMNESS) + { + int fd, got; + fd = robust_open("/dev/urandom", O_RDONLY, 0); + if( fd<0 ){ + time_t t; + time(&t); + memcpy(zBuf, &t, sizeof(t)); + memcpy(&zBuf[sizeof(t)], &randomnessPid, sizeof(randomnessPid)); + assert( sizeof(t)+sizeof(randomnessPid)<=(size_t)nBuf ); + nBuf = sizeof(t) + sizeof(randomnessPid); + }else{ + do{ got = osRead(fd, zBuf, nBuf); }while( got<0 && errno==EINTR ); + robust_close(0, fd, __LINE__); + } + } +#endif + return nBuf; +} + + +/* +** Sleep for a little while. Return the amount of time slept. +** The argument is the number of microseconds we want to sleep. +** The return value is the number of microseconds of sleep actually +** requested from the underlying operating system, a number which +** might be greater than or equal to the argument, but not less +** than the argument. +*/ +static int unixSleep(sqlite3_vfs *NotUsed, int microseconds){ +#if OS_VXWORKS + struct timespec sp; + + sp.tv_sec = microseconds / 1000000; + sp.tv_nsec = (microseconds % 1000000) * 1000; + nanosleep(&sp, NULL); + UNUSED_PARAMETER(NotUsed); + return microseconds; +#elif defined(HAVE_USLEEP) && HAVE_USLEEP + usleep(microseconds); + UNUSED_PARAMETER(NotUsed); + return microseconds; +#else + int seconds = (microseconds+999999)/1000000; + sleep(seconds); + UNUSED_PARAMETER(NotUsed); + return seconds*1000000; +#endif +} + +/* +** The following variable, if set to a non-zero value, is interpreted as +** the number of seconds since 1970 and is used to set the result of +** sqlite3OsCurrentTime() during testing. +*/ +#ifdef SQLITE_TEST +SQLITE_API int sqlite3_current_time = 0; /* Fake system time in seconds since 1970. */ +#endif + +/* +** Find the current time (in Universal Coordinated Time). Write into *piNow +** the current time and date as a Julian Day number times 86_400_000. In +** other words, write into *piNow the number of milliseconds since the Julian +** epoch of noon in Greenwich on November 24, 4714 B.C according to the +** proleptic Gregorian calendar. +** +** On success, return SQLITE_OK. Return SQLITE_ERROR if the time and date +** cannot be found. +*/ +static int unixCurrentTimeInt64(sqlite3_vfs *NotUsed, sqlite3_int64 *piNow){ + static const sqlite3_int64 unixEpoch = 24405875*(sqlite3_int64)8640000; + int rc = SQLITE_OK; +#if defined(NO_GETTOD) + time_t t; + time(&t); + *piNow = ((sqlite3_int64)t)*1000 + unixEpoch; +#elif OS_VXWORKS + struct timespec sNow; + clock_gettime(CLOCK_REALTIME, &sNow); + *piNow = unixEpoch + 1000*(sqlite3_int64)sNow.tv_sec + sNow.tv_nsec/1000000; +#else + struct timeval sNow; + (void)gettimeofday(&sNow, 0); /* Cannot fail given valid arguments */ + *piNow = unixEpoch + 1000*(sqlite3_int64)sNow.tv_sec + sNow.tv_usec/1000; +#endif + +#ifdef SQLITE_TEST + if( sqlite3_current_time ){ + *piNow = 1000*(sqlite3_int64)sqlite3_current_time + unixEpoch; + } +#endif + UNUSED_PARAMETER(NotUsed); + return rc; +} + +#ifndef SQLITE_OMIT_DEPRECATED +/* +** Find the current time (in Universal Coordinated Time). Write the +** current time and date as a Julian Day number into *prNow and +** return 0. Return 1 if the time and date cannot be found. +*/ +static int unixCurrentTime(sqlite3_vfs *NotUsed, double *prNow){ + sqlite3_int64 i = 0; + int rc; + UNUSED_PARAMETER(NotUsed); + rc = unixCurrentTimeInt64(0, &i); + *prNow = i/86400000.0; + return rc; +} +#else +# define unixCurrentTime 0 +#endif + +/* +** The xGetLastError() method is designed to return a better +** low-level error message when operating-system problems come up +** during SQLite operation. Only the integer return code is currently +** used. +*/ +static int unixGetLastError(sqlite3_vfs *NotUsed, int NotUsed2, char *NotUsed3){ + UNUSED_PARAMETER(NotUsed); + UNUSED_PARAMETER(NotUsed2); + UNUSED_PARAMETER(NotUsed3); + return errno; +} + + +/* +************************ End of sqlite3_vfs methods *************************** +******************************************************************************/ + +/****************************************************************************** +************************** Begin Proxy Locking ******************************** +** +** Proxy locking is a "uber-locking-method" in this sense: It uses the +** other locking methods on secondary lock files. Proxy locking is a +** meta-layer over top of the primitive locking implemented above. For +** this reason, the division that implements of proxy locking is deferred +** until late in the file (here) after all of the other I/O methods have +** been defined - so that the primitive locking methods are available +** as services to help with the implementation of proxy locking. +** +**** +** +** The default locking schemes in SQLite use byte-range locks on the +** database file to coordinate safe, concurrent access by multiple readers +** and writers [http://sqlite.org/lockingv3.html]. The five file locking +** states (UNLOCKED, PENDING, SHARED, RESERVED, EXCLUSIVE) are implemented +** as POSIX read & write locks over fixed set of locations (via fsctl), +** on AFP and SMB only exclusive byte-range locks are available via fsctl +** with _IOWR('z', 23, struct ByteRangeLockPB2) to track the same 5 states. +** To simulate a F_RDLCK on the shared range, on AFP a randomly selected +** address in the shared range is taken for a SHARED lock, the entire +** shared range is taken for an EXCLUSIVE lock): +** +** PENDING_BYTE 0x40000000 +** RESERVED_BYTE 0x40000001 +** SHARED_RANGE 0x40000002 -> 0x40000200 +** +** This works well on the local file system, but shows a nearly 100x +** slowdown in read performance on AFP because the AFP client disables +** the read cache when byte-range locks are present. Enabling the read +** cache exposes a cache coherency problem that is present on all OS X +** supported network file systems. NFS and AFP both observe the +** close-to-open semantics for ensuring cache coherency +** [http://nfs.sourceforge.net/#faq_a8], which does not effectively +** address the requirements for concurrent database access by multiple +** readers and writers +** [http://www.nabble.com/SQLite-on-NFS-cache-coherency-td15655701.html]. +** +** To address the performance and cache coherency issues, proxy file locking +** changes the way database access is controlled by limiting access to a +** single host at a time and moving file locks off of the database file +** and onto a proxy file on the local file system. +** +** +** Using proxy locks +** ----------------- +** +** C APIs +** +** sqlite3_file_control(db, dbname, SQLITE_FCNTL_SET_LOCKPROXYFILE, +** | ":auto:"); +** sqlite3_file_control(db, dbname, SQLITE_FCNTL_GET_LOCKPROXYFILE, +** &); +** +** +** SQL pragmas +** +** PRAGMA [database.]lock_proxy_file= | :auto: +** PRAGMA [database.]lock_proxy_file +** +** Specifying ":auto:" means that if there is a conch file with a matching +** host ID in it, the proxy path in the conch file will be used, otherwise +** a proxy path based on the user's temp dir +** (via confstr(_CS_DARWIN_USER_TEMP_DIR,...)) will be used and the +** actual proxy file name is generated from the name and path of the +** database file. For example: +** +** For database path "/Users/me/foo.db" +** The lock path will be "/sqliteplocks/_Users_me_foo.db:auto:") +** +** Once a lock proxy is configured for a database connection, it can not +** be removed, however it may be switched to a different proxy path via +** the above APIs (assuming the conch file is not being held by another +** connection or process). +** +** +** How proxy locking works +** ----------------------- +** +** Proxy file locking relies primarily on two new supporting files: +** +** * conch file to limit access to the database file to a single host +** at a time +** +** * proxy file to act as a proxy for the advisory locks normally +** taken on the database +** +** The conch file - to use a proxy file, sqlite must first "hold the conch" +** by taking an sqlite-style shared lock on the conch file, reading the +** contents and comparing the host's unique host ID (see below) and lock +** proxy path against the values stored in the conch. The conch file is +** stored in the same directory as the database file and the file name +** is patterned after the database file name as ".-conch". +** If the conch file does not exist, or its contents do not match the +** host ID and/or proxy path, then the lock is escalated to an exclusive +** lock and the conch file contents is updated with the host ID and proxy +** path and the lock is downgraded to a shared lock again. If the conch +** is held by another process (with a shared lock), the exclusive lock +** will fail and SQLITE_BUSY is returned. +** +** The proxy file - a single-byte file used for all advisory file locks +** normally taken on the database file. This allows for safe sharing +** of the database file for multiple readers and writers on the same +** host (the conch ensures that they all use the same local lock file). +** +** Requesting the lock proxy does not immediately take the conch, it is +** only taken when the first request to lock database file is made. +** This matches the semantics of the traditional locking behavior, where +** opening a connection to a database file does not take a lock on it. +** The shared lock and an open file descriptor are maintained until +** the connection to the database is closed. +** +** The proxy file and the lock file are never deleted so they only need +** to be created the first time they are used. +** +** Configuration options +** --------------------- +** +** SQLITE_PREFER_PROXY_LOCKING +** +** Database files accessed on non-local file systems are +** automatically configured for proxy locking, lock files are +** named automatically using the same logic as +** PRAGMA lock_proxy_file=":auto:" +** +** SQLITE_PROXY_DEBUG +** +** Enables the logging of error messages during host id file +** retrieval and creation +** +** LOCKPROXYDIR +** +** Overrides the default directory used for lock proxy files that +** are named automatically via the ":auto:" setting +** +** SQLITE_DEFAULT_PROXYDIR_PERMISSIONS +** +** Permissions to use when creating a directory for storing the +** lock proxy files, only used when LOCKPROXYDIR is not set. +** +** +** As mentioned above, when compiled with SQLITE_PREFER_PROXY_LOCKING, +** setting the environment variable SQLITE_FORCE_PROXY_LOCKING to 1 will +** force proxy locking to be used for every database file opened, and 0 +** will force automatic proxy locking to be disabled for all database +** files (explicitly calling the SQLITE_FCNTL_SET_LOCKPROXYFILE pragma or +** sqlite_file_control API is not affected by SQLITE_FORCE_PROXY_LOCKING). +*/ + +/* +** Proxy locking is only available on MacOSX +*/ +#if defined(__APPLE__) && SQLITE_ENABLE_LOCKING_STYLE + +/* +** The proxyLockingContext has the path and file structures for the remote +** and local proxy files in it +*/ +typedef struct proxyLockingContext proxyLockingContext; +struct proxyLockingContext { + unixFile *conchFile; /* Open conch file */ + char *conchFilePath; /* Name of the conch file */ + unixFile *lockProxy; /* Open proxy lock file */ + char *lockProxyPath; /* Name of the proxy lock file */ + char *dbPath; /* Name of the open file */ + int conchHeld; /* 1 if the conch is held, -1 if lockless */ + int nFails; /* Number of conch taking failures */ + void *oldLockingContext; /* Original lockingcontext to restore on close */ + sqlite3_io_methods const *pOldMethod; /* Original I/O methods for close */ +}; + +/* +** The proxy lock file path for the database at dbPath is written into lPath, +** which must point to valid, writable memory large enough for a maxLen length +** file path. +*/ +static int proxyGetLockPath(const char *dbPath, char *lPath, size_t maxLen){ + int len; + int dbLen; + int i; + +#ifdef LOCKPROXYDIR + len = strlcpy(lPath, LOCKPROXYDIR, maxLen); +#else +# ifdef _CS_DARWIN_USER_TEMP_DIR + { + if( !confstr(_CS_DARWIN_USER_TEMP_DIR, lPath, maxLen) ){ + OSTRACE(("GETLOCKPATH failed %s errno=%d pid=%d\n", + lPath, errno, osGetpid(0))); + return SQLITE_IOERR_LOCK; + } + len = strlcat(lPath, "sqliteplocks", maxLen); + } +# else + len = strlcpy(lPath, "/tmp/", maxLen); +# endif +#endif + + if( lPath[len-1]!='/' ){ + len = strlcat(lPath, "/", maxLen); + } + + /* transform the db path to a unique cache name */ + dbLen = (int)strlen(dbPath); + for( i=0; i 0) ){ + /* only mkdir if leaf dir != "." or "/" or ".." */ + if( i-start>2 || (i-start==1 && buf[start] != '.' && buf[start] != '/') + || (i-start==2 && buf[start] != '.' && buf[start+1] != '.') ){ + buf[i]='\0'; + if( osMkdir(buf, SQLITE_DEFAULT_PROXYDIR_PERMISSIONS) ){ + int err=errno; + if( err!=EEXIST ) { + OSTRACE(("CREATELOCKPATH FAILED creating %s, " + "'%s' proxy lock path=%s pid=%d\n", + buf, strerror(err), lockPath, osGetpid(0))); + return err; + } + } + } + start=i+1; + } + buf[i] = lockPath[i]; + } + OSTRACE(("CREATELOCKPATH proxy lock path=%s pid=%d\n",lockPath,osGetpid(0))); + return 0; +} + +/* +** Create a new VFS file descriptor (stored in memory obtained from +** sqlite3_malloc) and open the file named "path" in the file descriptor. +** +** The caller is responsible not only for closing the file descriptor +** but also for freeing the memory associated with the file descriptor. +*/ +static int proxyCreateUnixFile( + const char *path, /* path for the new unixFile */ + unixFile **ppFile, /* unixFile created and returned by ref */ + int islockfile /* if non zero missing dirs will be created */ +) { + int fd = -1; + unixFile *pNew; + int rc = SQLITE_OK; + int openFlags = O_RDWR | O_CREAT; + sqlite3_vfs dummyVfs; + int terrno = 0; + UnixUnusedFd *pUnused = NULL; + + /* 1. first try to open/create the file + ** 2. if that fails, and this is a lock file (not-conch), try creating + ** the parent directories and then try again. + ** 3. if that fails, try to open the file read-only + ** otherwise return BUSY (if lock file) or CANTOPEN for the conch file + */ + pUnused = findReusableFd(path, openFlags); + if( pUnused ){ + fd = pUnused->fd; + }else{ + pUnused = sqlite3_malloc64(sizeof(*pUnused)); + if( !pUnused ){ + return SQLITE_NOMEM_BKPT; + } + } + if( fd<0 ){ + fd = robust_open(path, openFlags, 0); + terrno = errno; + if( fd<0 && errno==ENOENT && islockfile ){ + if( proxyCreateLockPath(path) == SQLITE_OK ){ + fd = robust_open(path, openFlags, 0); + } + } + } + if( fd<0 ){ + openFlags = O_RDONLY; + fd = robust_open(path, openFlags, 0); + terrno = errno; + } + if( fd<0 ){ + if( islockfile ){ + return SQLITE_BUSY; + } + switch (terrno) { + case EACCES: + return SQLITE_PERM; + case EIO: + return SQLITE_IOERR_LOCK; /* even though it is the conch */ + default: + return SQLITE_CANTOPEN_BKPT; + } + } + + pNew = (unixFile *)sqlite3_malloc64(sizeof(*pNew)); + if( pNew==NULL ){ + rc = SQLITE_NOMEM_BKPT; + goto end_create_proxy; + } + memset(pNew, 0, sizeof(unixFile)); + pNew->openFlags = openFlags; + memset(&dummyVfs, 0, sizeof(dummyVfs)); + dummyVfs.pAppData = (void*)&autolockIoFinder; + dummyVfs.zName = "dummy"; + pUnused->fd = fd; + pUnused->flags = openFlags; + pNew->pPreallocatedUnused = pUnused; + + rc = fillInUnixFile(&dummyVfs, fd, (sqlite3_file*)pNew, path, 0); + if( rc==SQLITE_OK ){ + *ppFile = pNew; + return SQLITE_OK; + } +end_create_proxy: + robust_close(pNew, fd, __LINE__); + sqlite3_free(pNew); + sqlite3_free(pUnused); + return rc; +} + +#ifdef SQLITE_TEST +/* simulate multiple hosts by creating unique hostid file paths */ +SQLITE_API int sqlite3_hostid_num = 0; +#endif + +#define PROXY_HOSTIDLEN 16 /* conch file host id length */ + +#if HAVE_GETHOSTUUID +/* Not always defined in the headers as it ought to be */ +extern int gethostuuid(uuid_t id, const struct timespec *wait); +#endif + +/* get the host ID via gethostuuid(), pHostID must point to PROXY_HOSTIDLEN +** bytes of writable memory. +*/ +static int proxyGetHostID(unsigned char *pHostID, int *pError){ + assert(PROXY_HOSTIDLEN == sizeof(uuid_t)); + memset(pHostID, 0, PROXY_HOSTIDLEN); +#if HAVE_GETHOSTUUID + { + struct timespec timeout = {1, 0}; /* 1 sec timeout */ + if( gethostuuid(pHostID, &timeout) ){ + int err = errno; + if( pError ){ + *pError = err; + } + return SQLITE_IOERR; + } + } +#else + UNUSED_PARAMETER(pError); +#endif +#ifdef SQLITE_TEST + /* simulate multiple hosts by creating unique hostid file paths */ + if( sqlite3_hostid_num != 0){ + pHostID[0] = (char)(pHostID[0] + (char)(sqlite3_hostid_num & 0xFF)); + } +#endif + + return SQLITE_OK; +} + +/* The conch file contains the header, host id and lock file path + */ +#define PROXY_CONCHVERSION 2 /* 1-byte header, 16-byte host id, path */ +#define PROXY_HEADERLEN 1 /* conch file header length */ +#define PROXY_PATHINDEX (PROXY_HEADERLEN+PROXY_HOSTIDLEN) +#define PROXY_MAXCONCHLEN (PROXY_HEADERLEN+PROXY_HOSTIDLEN+MAXPATHLEN) + +/* +** Takes an open conch file, copies the contents to a new path and then moves +** it back. The newly created file's file descriptor is assigned to the +** conch file structure and finally the original conch file descriptor is +** closed. Returns zero if successful. +*/ +static int proxyBreakConchLock(unixFile *pFile, uuid_t myHostID){ + proxyLockingContext *pCtx = (proxyLockingContext *)pFile->lockingContext; + unixFile *conchFile = pCtx->conchFile; + char tPath[MAXPATHLEN]; + char buf[PROXY_MAXCONCHLEN]; + char *cPath = pCtx->conchFilePath; + size_t readLen = 0; + size_t pathLen = 0; + char errmsg[64] = ""; + int fd = -1; + int rc = -1; + UNUSED_PARAMETER(myHostID); + + /* create a new path by replace the trailing '-conch' with '-break' */ + pathLen = strlcpy(tPath, cPath, MAXPATHLEN); + if( pathLen>MAXPATHLEN || pathLen<6 || + (strlcpy(&tPath[pathLen-5], "break", 6) != 5) ){ + sqlite3_snprintf(sizeof(errmsg),errmsg,"path error (len %d)",(int)pathLen); + goto end_breaklock; + } + /* read the conch content */ + readLen = osPread(conchFile->h, buf, PROXY_MAXCONCHLEN, 0); + if( readLenh, __LINE__); + conchFile->h = fd; + conchFile->openFlags = O_RDWR | O_CREAT; + +end_breaklock: + if( rc ){ + if( fd>=0 ){ + osUnlink(tPath); + robust_close(pFile, fd, __LINE__); + } + fprintf(stderr, "failed to break stale lock on %s, %s\n", cPath, errmsg); + } + return rc; +} + +/* Take the requested lock on the conch file and break a stale lock if the +** host id matches. +*/ +static int proxyConchLock(unixFile *pFile, uuid_t myHostID, int lockType){ + proxyLockingContext *pCtx = (proxyLockingContext *)pFile->lockingContext; + unixFile *conchFile = pCtx->conchFile; + int rc = SQLITE_OK; + int nTries = 0; + struct timespec conchModTime; + + memset(&conchModTime, 0, sizeof(conchModTime)); + do { + rc = conchFile->pMethod->xLock((sqlite3_file*)conchFile, lockType); + nTries ++; + if( rc==SQLITE_BUSY ){ + /* If the lock failed (busy): + * 1st try: get the mod time of the conch, wait 0.5s and try again. + * 2nd try: fail if the mod time changed or host id is different, wait + * 10 sec and try again + * 3rd try: break the lock unless the mod time has changed. + */ + struct stat buf; + if( osFstat(conchFile->h, &buf) ){ + storeLastErrno(pFile, errno); + return SQLITE_IOERR_LOCK; + } + + if( nTries==1 ){ + conchModTime = buf.st_mtimespec; + usleep(500000); /* wait 0.5 sec and try the lock again*/ + continue; + } + + assert( nTries>1 ); + if( conchModTime.tv_sec != buf.st_mtimespec.tv_sec || + conchModTime.tv_nsec != buf.st_mtimespec.tv_nsec ){ + return SQLITE_BUSY; + } + + if( nTries==2 ){ + char tBuf[PROXY_MAXCONCHLEN]; + int len = osPread(conchFile->h, tBuf, PROXY_MAXCONCHLEN, 0); + if( len<0 ){ + storeLastErrno(pFile, errno); + return SQLITE_IOERR_LOCK; + } + if( len>PROXY_PATHINDEX && tBuf[0]==(char)PROXY_CONCHVERSION){ + /* don't break the lock if the host id doesn't match */ + if( 0!=memcmp(&tBuf[PROXY_HEADERLEN], myHostID, PROXY_HOSTIDLEN) ){ + return SQLITE_BUSY; + } + }else{ + /* don't break the lock on short read or a version mismatch */ + return SQLITE_BUSY; + } + usleep(10000000); /* wait 10 sec and try the lock again */ + continue; + } + + assert( nTries==3 ); + if( 0==proxyBreakConchLock(pFile, myHostID) ){ + rc = SQLITE_OK; + if( lockType==EXCLUSIVE_LOCK ){ + rc = conchFile->pMethod->xLock((sqlite3_file*)conchFile, SHARED_LOCK); + } + if( !rc ){ + rc = conchFile->pMethod->xLock((sqlite3_file*)conchFile, lockType); + } + } + } + } while( rc==SQLITE_BUSY && nTries<3 ); + + return rc; +} + +/* Takes the conch by taking a shared lock and read the contents conch, if +** lockPath is non-NULL, the host ID and lock file path must match. A NULL +** lockPath means that the lockPath in the conch file will be used if the +** host IDs match, or a new lock path will be generated automatically +** and written to the conch file. +*/ +static int proxyTakeConch(unixFile *pFile){ + proxyLockingContext *pCtx = (proxyLockingContext *)pFile->lockingContext; + + if( pCtx->conchHeld!=0 ){ + return SQLITE_OK; + }else{ + unixFile *conchFile = pCtx->conchFile; + uuid_t myHostID; + int pError = 0; + char readBuf[PROXY_MAXCONCHLEN]; + char lockPath[MAXPATHLEN]; + char *tempLockPath = NULL; + int rc = SQLITE_OK; + int createConch = 0; + int hostIdMatch = 0; + int readLen = 0; + int tryOldLockPath = 0; + int forceNewLockPath = 0; + + OSTRACE(("TAKECONCH %d for %s pid=%d\n", conchFile->h, + (pCtx->lockProxyPath ? pCtx->lockProxyPath : ":auto:"), + osGetpid(0))); + + rc = proxyGetHostID(myHostID, &pError); + if( (rc&0xff)==SQLITE_IOERR ){ + storeLastErrno(pFile, pError); + goto end_takeconch; + } + rc = proxyConchLock(pFile, myHostID, SHARED_LOCK); + if( rc!=SQLITE_OK ){ + goto end_takeconch; + } + /* read the existing conch file */ + readLen = seekAndRead((unixFile*)conchFile, 0, readBuf, PROXY_MAXCONCHLEN); + if( readLen<0 ){ + /* I/O error: lastErrno set by seekAndRead */ + storeLastErrno(pFile, conchFile->lastErrno); + rc = SQLITE_IOERR_READ; + goto end_takeconch; + }else if( readLen<=(PROXY_HEADERLEN+PROXY_HOSTIDLEN) || + readBuf[0]!=(char)PROXY_CONCHVERSION ){ + /* a short read or version format mismatch means we need to create a new + ** conch file. + */ + createConch = 1; + } + /* if the host id matches and the lock path already exists in the conch + ** we'll try to use the path there, if we can't open that path, we'll + ** retry with a new auto-generated path + */ + do { /* in case we need to try again for an :auto: named lock file */ + + if( !createConch && !forceNewLockPath ){ + hostIdMatch = !memcmp(&readBuf[PROXY_HEADERLEN], myHostID, + PROXY_HOSTIDLEN); + /* if the conch has data compare the contents */ + if( !pCtx->lockProxyPath ){ + /* for auto-named local lock file, just check the host ID and we'll + ** use the local lock file path that's already in there + */ + if( hostIdMatch ){ + size_t pathLen = (readLen - PROXY_PATHINDEX); + + if( pathLen>=MAXPATHLEN ){ + pathLen=MAXPATHLEN-1; + } + memcpy(lockPath, &readBuf[PROXY_PATHINDEX], pathLen); + lockPath[pathLen] = 0; + tempLockPath = lockPath; + tryOldLockPath = 1; + /* create a copy of the lock path if the conch is taken */ + goto end_takeconch; + } + }else if( hostIdMatch + && !strncmp(pCtx->lockProxyPath, &readBuf[PROXY_PATHINDEX], + readLen-PROXY_PATHINDEX) + ){ + /* conch host and lock path match */ + goto end_takeconch; + } + } + + /* if the conch isn't writable and doesn't match, we can't take it */ + if( (conchFile->openFlags&O_RDWR) == 0 ){ + rc = SQLITE_BUSY; + goto end_takeconch; + } + + /* either the conch didn't match or we need to create a new one */ + if( !pCtx->lockProxyPath ){ + proxyGetLockPath(pCtx->dbPath, lockPath, MAXPATHLEN); + tempLockPath = lockPath; + /* create a copy of the lock path _only_ if the conch is taken */ + } + + /* update conch with host and path (this will fail if other process + ** has a shared lock already), if the host id matches, use the big + ** stick. + */ + futimes(conchFile->h, NULL); + if( hostIdMatch && !createConch ){ + if( conchFile->pInode && conchFile->pInode->nShared>1 ){ + /* We are trying for an exclusive lock but another thread in this + ** same process is still holding a shared lock. */ + rc = SQLITE_BUSY; + } else { + rc = proxyConchLock(pFile, myHostID, EXCLUSIVE_LOCK); + } + }else{ + rc = proxyConchLock(pFile, myHostID, EXCLUSIVE_LOCK); + } + if( rc==SQLITE_OK ){ + char writeBuffer[PROXY_MAXCONCHLEN]; + int writeSize = 0; + + writeBuffer[0] = (char)PROXY_CONCHVERSION; + memcpy(&writeBuffer[PROXY_HEADERLEN], myHostID, PROXY_HOSTIDLEN); + if( pCtx->lockProxyPath!=NULL ){ + strlcpy(&writeBuffer[PROXY_PATHINDEX], pCtx->lockProxyPath, + MAXPATHLEN); + }else{ + strlcpy(&writeBuffer[PROXY_PATHINDEX], tempLockPath, MAXPATHLEN); + } + writeSize = PROXY_PATHINDEX + strlen(&writeBuffer[PROXY_PATHINDEX]); + robust_ftruncate(conchFile->h, writeSize); + rc = unixWrite((sqlite3_file *)conchFile, writeBuffer, writeSize, 0); + full_fsync(conchFile->h,0,0); + /* If we created a new conch file (not just updated the contents of a + ** valid conch file), try to match the permissions of the database + */ + if( rc==SQLITE_OK && createConch ){ + struct stat buf; + int err = osFstat(pFile->h, &buf); + if( err==0 ){ + mode_t cmode = buf.st_mode&(S_IRUSR|S_IWUSR | S_IRGRP|S_IWGRP | + S_IROTH|S_IWOTH); + /* try to match the database file R/W permissions, ignore failure */ +#ifndef SQLITE_PROXY_DEBUG + osFchmod(conchFile->h, cmode); +#else + do{ + rc = osFchmod(conchFile->h, cmode); + }while( rc==(-1) && errno==EINTR ); + if( rc!=0 ){ + int code = errno; + fprintf(stderr, "fchmod %o FAILED with %d %s\n", + cmode, code, strerror(code)); + } else { + fprintf(stderr, "fchmod %o SUCCEDED\n",cmode); + } + }else{ + int code = errno; + fprintf(stderr, "STAT FAILED[%d] with %d %s\n", + err, code, strerror(code)); +#endif + } + } + } + conchFile->pMethod->xUnlock((sqlite3_file*)conchFile, SHARED_LOCK); + + end_takeconch: + OSTRACE(("TRANSPROXY: CLOSE %d\n", pFile->h)); + if( rc==SQLITE_OK && pFile->openFlags ){ + int fd; + if( pFile->h>=0 ){ + robust_close(pFile, pFile->h, __LINE__); + } + pFile->h = -1; + fd = robust_open(pCtx->dbPath, pFile->openFlags, 0); + OSTRACE(("TRANSPROXY: OPEN %d\n", fd)); + if( fd>=0 ){ + pFile->h = fd; + }else{ + rc=SQLITE_CANTOPEN_BKPT; /* SQLITE_BUSY? proxyTakeConch called + during locking */ + } + } + if( rc==SQLITE_OK && !pCtx->lockProxy ){ + char *path = tempLockPath ? tempLockPath : pCtx->lockProxyPath; + rc = proxyCreateUnixFile(path, &pCtx->lockProxy, 1); + if( rc!=SQLITE_OK && rc!=SQLITE_NOMEM && tryOldLockPath ){ + /* we couldn't create the proxy lock file with the old lock file path + ** so try again via auto-naming + */ + forceNewLockPath = 1; + tryOldLockPath = 0; + continue; /* go back to the do {} while start point, try again */ + } + } + if( rc==SQLITE_OK ){ + /* Need to make a copy of path if we extracted the value + ** from the conch file or the path was allocated on the stack + */ + if( tempLockPath ){ + pCtx->lockProxyPath = sqlite3DbStrDup(0, tempLockPath); + if( !pCtx->lockProxyPath ){ + rc = SQLITE_NOMEM_BKPT; + } + } + } + if( rc==SQLITE_OK ){ + pCtx->conchHeld = 1; + + if( pCtx->lockProxy->pMethod == &afpIoMethods ){ + afpLockingContext *afpCtx; + afpCtx = (afpLockingContext *)pCtx->lockProxy->lockingContext; + afpCtx->dbPath = pCtx->lockProxyPath; + } + } else { + conchFile->pMethod->xUnlock((sqlite3_file*)conchFile, NO_LOCK); + } + OSTRACE(("TAKECONCH %d %s\n", conchFile->h, + rc==SQLITE_OK?"ok":"failed")); + return rc; + } while (1); /* in case we need to retry the :auto: lock file - + ** we should never get here except via the 'continue' call. */ + } +} + +/* +** If pFile holds a lock on a conch file, then release that lock. +*/ +static int proxyReleaseConch(unixFile *pFile){ + int rc = SQLITE_OK; /* Subroutine return code */ + proxyLockingContext *pCtx; /* The locking context for the proxy lock */ + unixFile *conchFile; /* Name of the conch file */ + + pCtx = (proxyLockingContext *)pFile->lockingContext; + conchFile = pCtx->conchFile; + OSTRACE(("RELEASECONCH %d for %s pid=%d\n", conchFile->h, + (pCtx->lockProxyPath ? pCtx->lockProxyPath : ":auto:"), + osGetpid(0))); + if( pCtx->conchHeld>0 ){ + rc = conchFile->pMethod->xUnlock((sqlite3_file*)conchFile, NO_LOCK); + } + pCtx->conchHeld = 0; + OSTRACE(("RELEASECONCH %d %s\n", conchFile->h, + (rc==SQLITE_OK ? "ok" : "failed"))); + return rc; +} + +/* +** Given the name of a database file, compute the name of its conch file. +** Store the conch filename in memory obtained from sqlite3_malloc64(). +** Make *pConchPath point to the new name. Return SQLITE_OK on success +** or SQLITE_NOMEM if unable to obtain memory. +** +** The caller is responsible for ensuring that the allocated memory +** space is eventually freed. +** +** *pConchPath is set to NULL if a memory allocation error occurs. +*/ +static int proxyCreateConchPathname(char *dbPath, char **pConchPath){ + int i; /* Loop counter */ + int len = (int)strlen(dbPath); /* Length of database filename - dbPath */ + char *conchPath; /* buffer in which to construct conch name */ + + /* Allocate space for the conch filename and initialize the name to + ** the name of the original database file. */ + *pConchPath = conchPath = (char *)sqlite3_malloc64(len + 8); + if( conchPath==0 ){ + return SQLITE_NOMEM_BKPT; + } + memcpy(conchPath, dbPath, len+1); + + /* now insert a "." before the last / character */ + for( i=(len-1); i>=0; i-- ){ + if( conchPath[i]=='/' ){ + i++; + break; + } + } + conchPath[i]='.'; + while ( ilockingContext; + char *oldPath = pCtx->lockProxyPath; + int rc = SQLITE_OK; + + if( pFile->eFileLock!=NO_LOCK ){ + return SQLITE_BUSY; + } + + /* nothing to do if the path is NULL, :auto: or matches the existing path */ + if( !path || path[0]=='\0' || !strcmp(path, ":auto:") || + (oldPath && !strncmp(oldPath, path, MAXPATHLEN)) ){ + return SQLITE_OK; + }else{ + unixFile *lockProxy = pCtx->lockProxy; + pCtx->lockProxy=NULL; + pCtx->conchHeld = 0; + if( lockProxy!=NULL ){ + rc=lockProxy->pMethod->xClose((sqlite3_file *)lockProxy); + if( rc ) return rc; + sqlite3_free(lockProxy); + } + sqlite3_free(oldPath); + pCtx->lockProxyPath = sqlite3DbStrDup(0, path); + } + + return rc; +} + +/* +** pFile is a file that has been opened by a prior xOpen call. dbPath +** is a string buffer at least MAXPATHLEN+1 characters in size. +** +** This routine find the filename associated with pFile and writes it +** int dbPath. +*/ +static int proxyGetDbPathForUnixFile(unixFile *pFile, char *dbPath){ +#if defined(__APPLE__) + if( pFile->pMethod == &afpIoMethods ){ + /* afp style keeps a reference to the db path in the filePath field + ** of the struct */ + assert( (int)strlen((char*)pFile->lockingContext)<=MAXPATHLEN ); + strlcpy(dbPath, ((afpLockingContext *)pFile->lockingContext)->dbPath, + MAXPATHLEN); + } else +#endif + if( pFile->pMethod == &dotlockIoMethods ){ + /* dot lock style uses the locking context to store the dot lock + ** file path */ + int len = strlen((char *)pFile->lockingContext) - strlen(DOTLOCK_SUFFIX); + memcpy(dbPath, (char *)pFile->lockingContext, len + 1); + }else{ + /* all other styles use the locking context to store the db file path */ + assert( strlen((char*)pFile->lockingContext)<=MAXPATHLEN ); + strlcpy(dbPath, (char *)pFile->lockingContext, MAXPATHLEN); + } + return SQLITE_OK; +} + +/* +** Takes an already filled in unix file and alters it so all file locking +** will be performed on the local proxy lock file. The following fields +** are preserved in the locking context so that they can be restored and +** the unix structure properly cleaned up at close time: +** ->lockingContext +** ->pMethod +*/ +static int proxyTransformUnixFile(unixFile *pFile, const char *path) { + proxyLockingContext *pCtx; + char dbPath[MAXPATHLEN+1]; /* Name of the database file */ + char *lockPath=NULL; + int rc = SQLITE_OK; + + if( pFile->eFileLock!=NO_LOCK ){ + return SQLITE_BUSY; + } + proxyGetDbPathForUnixFile(pFile, dbPath); + if( !path || path[0]=='\0' || !strcmp(path, ":auto:") ){ + lockPath=NULL; + }else{ + lockPath=(char *)path; + } + + OSTRACE(("TRANSPROXY %d for %s pid=%d\n", pFile->h, + (lockPath ? lockPath : ":auto:"), osGetpid(0))); + + pCtx = sqlite3_malloc64( sizeof(*pCtx) ); + if( pCtx==0 ){ + return SQLITE_NOMEM_BKPT; + } + memset(pCtx, 0, sizeof(*pCtx)); + + rc = proxyCreateConchPathname(dbPath, &pCtx->conchFilePath); + if( rc==SQLITE_OK ){ + rc = proxyCreateUnixFile(pCtx->conchFilePath, &pCtx->conchFile, 0); + if( rc==SQLITE_CANTOPEN && ((pFile->openFlags&O_RDWR) == 0) ){ + /* if (a) the open flags are not O_RDWR, (b) the conch isn't there, and + ** (c) the file system is read-only, then enable no-locking access. + ** Ugh, since O_RDONLY==0x0000 we test for !O_RDWR since unixOpen asserts + ** that openFlags will have only one of O_RDONLY or O_RDWR. + */ + struct statfs fsInfo; + struct stat conchInfo; + int goLockless = 0; + + if( osStat(pCtx->conchFilePath, &conchInfo) == -1 ) { + int err = errno; + if( (err==ENOENT) && (statfs(dbPath, &fsInfo) != -1) ){ + goLockless = (fsInfo.f_flags&MNT_RDONLY) == MNT_RDONLY; + } + } + if( goLockless ){ + pCtx->conchHeld = -1; /* read only FS/ lockless */ + rc = SQLITE_OK; + } + } + } + if( rc==SQLITE_OK && lockPath ){ + pCtx->lockProxyPath = sqlite3DbStrDup(0, lockPath); + } + + if( rc==SQLITE_OK ){ + pCtx->dbPath = sqlite3DbStrDup(0, dbPath); + if( pCtx->dbPath==NULL ){ + rc = SQLITE_NOMEM_BKPT; + } + } + if( rc==SQLITE_OK ){ + /* all memory is allocated, proxys are created and assigned, + ** switch the locking context and pMethod then return. + */ + pCtx->oldLockingContext = pFile->lockingContext; + pFile->lockingContext = pCtx; + pCtx->pOldMethod = pFile->pMethod; + pFile->pMethod = &proxyIoMethods; + }else{ + if( pCtx->conchFile ){ + pCtx->conchFile->pMethod->xClose((sqlite3_file *)pCtx->conchFile); + sqlite3_free(pCtx->conchFile); + } + sqlite3DbFree(0, pCtx->lockProxyPath); + sqlite3_free(pCtx->conchFilePath); + sqlite3_free(pCtx); + } + OSTRACE(("TRANSPROXY %d %s\n", pFile->h, + (rc==SQLITE_OK ? "ok" : "failed"))); + return rc; +} + + +/* +** This routine handles sqlite3_file_control() calls that are specific +** to proxy locking. +*/ +static int proxyFileControl(sqlite3_file *id, int op, void *pArg){ + switch( op ){ + case SQLITE_FCNTL_GET_LOCKPROXYFILE: { + unixFile *pFile = (unixFile*)id; + if( pFile->pMethod == &proxyIoMethods ){ + proxyLockingContext *pCtx = (proxyLockingContext*)pFile->lockingContext; + proxyTakeConch(pFile); + if( pCtx->lockProxyPath ){ + *(const char **)pArg = pCtx->lockProxyPath; + }else{ + *(const char **)pArg = ":auto: (not held)"; + } + } else { + *(const char **)pArg = NULL; + } + return SQLITE_OK; + } + case SQLITE_FCNTL_SET_LOCKPROXYFILE: { + unixFile *pFile = (unixFile*)id; + int rc = SQLITE_OK; + int isProxyStyle = (pFile->pMethod == &proxyIoMethods); + if( pArg==NULL || (const char *)pArg==0 ){ + if( isProxyStyle ){ + /* turn off proxy locking - not supported. If support is added for + ** switching proxy locking mode off then it will need to fail if + ** the journal mode is WAL mode. + */ + rc = SQLITE_ERROR /*SQLITE_PROTOCOL? SQLITE_MISUSE?*/; + }else{ + /* turn off proxy locking - already off - NOOP */ + rc = SQLITE_OK; + } + }else{ + const char *proxyPath = (const char *)pArg; + if( isProxyStyle ){ + proxyLockingContext *pCtx = + (proxyLockingContext*)pFile->lockingContext; + if( !strcmp(pArg, ":auto:") + || (pCtx->lockProxyPath && + !strncmp(pCtx->lockProxyPath, proxyPath, MAXPATHLEN)) + ){ + rc = SQLITE_OK; + }else{ + rc = switchLockProxyPath(pFile, proxyPath); + } + }else{ + /* turn on proxy file locking */ + rc = proxyTransformUnixFile(pFile, proxyPath); + } + } + return rc; + } + default: { + assert( 0 ); /* The call assures that only valid opcodes are sent */ + } + } + /*NOTREACHED*/ assert(0); + return SQLITE_ERROR; +} + +/* +** Within this division (the proxying locking implementation) the procedures +** above this point are all utilities. The lock-related methods of the +** proxy-locking sqlite3_io_method object follow. +*/ + + +/* +** This routine checks if there is a RESERVED lock held on the specified +** file by this or any other process. If such a lock is held, set *pResOut +** to a non-zero value otherwise *pResOut is set to zero. The return value +** is set to SQLITE_OK unless an I/O error occurs during lock checking. +*/ +static int proxyCheckReservedLock(sqlite3_file *id, int *pResOut) { + unixFile *pFile = (unixFile*)id; + int rc = proxyTakeConch(pFile); + if( rc==SQLITE_OK ){ + proxyLockingContext *pCtx = (proxyLockingContext *)pFile->lockingContext; + if( pCtx->conchHeld>0 ){ + unixFile *proxy = pCtx->lockProxy; + return proxy->pMethod->xCheckReservedLock((sqlite3_file*)proxy, pResOut); + }else{ /* conchHeld < 0 is lockless */ + pResOut=0; + } + } + return rc; +} + +/* +** Lock the file with the lock specified by parameter eFileLock - one +** of the following: +** +** (1) SHARED_LOCK +** (2) RESERVED_LOCK +** (3) PENDING_LOCK +** (4) EXCLUSIVE_LOCK +** +** Sometimes when requesting one lock state, additional lock states +** are inserted in between. The locking might fail on one of the later +** transitions leaving the lock state different from what it started but +** still short of its goal. The following chart shows the allowed +** transitions and the inserted intermediate states: +** +** UNLOCKED -> SHARED +** SHARED -> RESERVED +** SHARED -> (PENDING) -> EXCLUSIVE +** RESERVED -> (PENDING) -> EXCLUSIVE +** PENDING -> EXCLUSIVE +** +** This routine will only increase a lock. Use the sqlite3OsUnlock() +** routine to lower a locking level. +*/ +static int proxyLock(sqlite3_file *id, int eFileLock) { + unixFile *pFile = (unixFile*)id; + int rc = proxyTakeConch(pFile); + if( rc==SQLITE_OK ){ + proxyLockingContext *pCtx = (proxyLockingContext *)pFile->lockingContext; + if( pCtx->conchHeld>0 ){ + unixFile *proxy = pCtx->lockProxy; + rc = proxy->pMethod->xLock((sqlite3_file*)proxy, eFileLock); + pFile->eFileLock = proxy->eFileLock; + }else{ + /* conchHeld < 0 is lockless */ + } + } + return rc; +} + + +/* +** Lower the locking level on file descriptor pFile to eFileLock. eFileLock +** must be either NO_LOCK or SHARED_LOCK. +** +** If the locking level of the file descriptor is already at or below +** the requested locking level, this routine is a no-op. +*/ +static int proxyUnlock(sqlite3_file *id, int eFileLock) { + unixFile *pFile = (unixFile*)id; + int rc = proxyTakeConch(pFile); + if( rc==SQLITE_OK ){ + proxyLockingContext *pCtx = (proxyLockingContext *)pFile->lockingContext; + if( pCtx->conchHeld>0 ){ + unixFile *proxy = pCtx->lockProxy; + rc = proxy->pMethod->xUnlock((sqlite3_file*)proxy, eFileLock); + pFile->eFileLock = proxy->eFileLock; + }else{ + /* conchHeld < 0 is lockless */ + } + } + return rc; +} + +/* +** Close a file that uses proxy locks. +*/ +static int proxyClose(sqlite3_file *id) { + if( ALWAYS(id) ){ + unixFile *pFile = (unixFile*)id; + proxyLockingContext *pCtx = (proxyLockingContext *)pFile->lockingContext; + unixFile *lockProxy = pCtx->lockProxy; + unixFile *conchFile = pCtx->conchFile; + int rc = SQLITE_OK; + + if( lockProxy ){ + rc = lockProxy->pMethod->xUnlock((sqlite3_file*)lockProxy, NO_LOCK); + if( rc ) return rc; + rc = lockProxy->pMethod->xClose((sqlite3_file*)lockProxy); + if( rc ) return rc; + sqlite3_free(lockProxy); + pCtx->lockProxy = 0; + } + if( conchFile ){ + if( pCtx->conchHeld ){ + rc = proxyReleaseConch(pFile); + if( rc ) return rc; + } + rc = conchFile->pMethod->xClose((sqlite3_file*)conchFile); + if( rc ) return rc; + sqlite3_free(conchFile); + } + sqlite3DbFree(0, pCtx->lockProxyPath); + sqlite3_free(pCtx->conchFilePath); + sqlite3DbFree(0, pCtx->dbPath); + /* restore the original locking context and pMethod then close it */ + pFile->lockingContext = pCtx->oldLockingContext; + pFile->pMethod = pCtx->pOldMethod; + sqlite3_free(pCtx); + return pFile->pMethod->xClose(id); + } + return SQLITE_OK; +} + + + +#endif /* defined(__APPLE__) && SQLITE_ENABLE_LOCKING_STYLE */ +/* +** The proxy locking style is intended for use with AFP filesystems. +** And since AFP is only supported on MacOSX, the proxy locking is also +** restricted to MacOSX. +** +** +******************* End of the proxy lock implementation ********************** +******************************************************************************/ + +/* +** Initialize the operating system interface. +** +** This routine registers all VFS implementations for unix-like operating +** systems. This routine, and the sqlite3_os_end() routine that follows, +** should be the only routines in this file that are visible from other +** files. +** +** This routine is called once during SQLite initialization and by a +** single thread. The memory allocation and mutex subsystems have not +** necessarily been initialized when this routine is called, and so they +** should not be used. +*/ +SQLITE_API int sqlite3_os_init(void){ + /* + ** The following macro defines an initializer for an sqlite3_vfs object. + ** The name of the VFS is NAME. The pAppData is a pointer to a pointer + ** to the "finder" function. (pAppData is a pointer to a pointer because + ** silly C90 rules prohibit a void* from being cast to a function pointer + ** and so we have to go through the intermediate pointer to avoid problems + ** when compiling with -pedantic-errors on GCC.) + ** + ** The FINDER parameter to this macro is the name of the pointer to the + ** finder-function. The finder-function returns a pointer to the + ** sqlite_io_methods object that implements the desired locking + ** behaviors. See the division above that contains the IOMETHODS + ** macro for addition information on finder-functions. + ** + ** Most finders simply return a pointer to a fixed sqlite3_io_methods + ** object. But the "autolockIoFinder" available on MacOSX does a little + ** more than that; it looks at the filesystem type that hosts the + ** database file and tries to choose an locking method appropriate for + ** that filesystem time. + */ + #define UNIXVFS(VFSNAME, FINDER) { \ + 3, /* iVersion */ \ + sizeof(unixFile), /* szOsFile */ \ + MAX_PATHNAME, /* mxPathname */ \ + 0, /* pNext */ \ + VFSNAME, /* zName */ \ + (void*)&FINDER, /* pAppData */ \ + unixOpen, /* xOpen */ \ + unixDelete, /* xDelete */ \ + unixAccess, /* xAccess */ \ + unixFullPathname, /* xFullPathname */ \ + unixDlOpen, /* xDlOpen */ \ + unixDlError, /* xDlError */ \ + unixDlSym, /* xDlSym */ \ + unixDlClose, /* xDlClose */ \ + unixRandomness, /* xRandomness */ \ + unixSleep, /* xSleep */ \ + unixCurrentTime, /* xCurrentTime */ \ + unixGetLastError, /* xGetLastError */ \ + unixCurrentTimeInt64, /* xCurrentTimeInt64 */ \ + unixSetSystemCall, /* xSetSystemCall */ \ + unixGetSystemCall, /* xGetSystemCall */ \ + unixNextSystemCall, /* xNextSystemCall */ \ + } + + /* + ** All default VFSes for unix are contained in the following array. + ** + ** Note that the sqlite3_vfs.pNext field of the VFS object is modified + ** by the SQLite core when the VFS is registered. So the following + ** array cannot be const. + */ + static sqlite3_vfs aVfs[] = { +#if SQLITE_ENABLE_LOCKING_STYLE && defined(__APPLE__) + UNIXVFS("unix", autolockIoFinder ), +#elif OS_VXWORKS + UNIXVFS("unix", vxworksIoFinder ), +#else + UNIXVFS("unix", posixIoFinder ), +#endif + UNIXVFS("unix-none", nolockIoFinder ), + UNIXVFS("unix-dotfile", dotlockIoFinder ), + UNIXVFS("unix-excl", posixIoFinder ), +#if OS_VXWORKS + UNIXVFS("unix-namedsem", semIoFinder ), +#endif +#if SQLITE_ENABLE_LOCKING_STYLE || OS_VXWORKS + UNIXVFS("unix-posix", posixIoFinder ), +#endif +#if SQLITE_ENABLE_LOCKING_STYLE + UNIXVFS("unix-flock", flockIoFinder ), +#endif +#if SQLITE_ENABLE_LOCKING_STYLE && defined(__APPLE__) + UNIXVFS("unix-afp", afpIoFinder ), + UNIXVFS("unix-nfs", nfsIoFinder ), + UNIXVFS("unix-proxy", proxyIoFinder ), +#endif + }; + unsigned int i; /* Loop counter */ + + /* Double-check that the aSyscall[] array has been constructed + ** correctly. See ticket [bb3a86e890c8e96ab] */ + assert( ArraySize(aSyscall)==29 ); + + /* Register all VFSes defined in the aVfs[] array */ + for(i=0; i<(sizeof(aVfs)/sizeof(sqlite3_vfs)); i++){ + sqlite3_vfs_register(&aVfs[i], i==0); + } + unixBigLock = sqlite3MutexAlloc(SQLITE_MUTEX_STATIC_VFS1); + return SQLITE_OK; +} + +/* +** Shutdown the operating system interface. +** +** Some operating systems might need to do some cleanup in this routine, +** to release dynamically allocated objects. But not on unix. +** This routine is a no-op for unix. +*/ +SQLITE_API int sqlite3_os_end(void){ + unixBigLock = 0; + return SQLITE_OK; +} + +#endif /* SQLITE_OS_UNIX */ + +/************** End of os_unix.c *********************************************/ +/************** Begin file os_win.c ******************************************/ +/* +** 2004 May 22 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +****************************************************************************** +** +** This file contains code that is specific to Windows. +*/ +/* #include "sqliteInt.h" */ +#if SQLITE_OS_WIN /* This file is used for Windows only */ + +/* +** Include code that is common to all os_*.c files +*/ +/************** Include os_common.h in the middle of os_win.c ****************/ +/************** Begin file os_common.h ***************************************/ +/* +** 2004 May 22 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +****************************************************************************** +** +** This file contains macros and a little bit of code that is common to +** all of the platform-specific files (os_*.c) and is #included into those +** files. +** +** This file should be #included by the os_*.c files only. It is not a +** general purpose header file. +*/ +#ifndef _OS_COMMON_H_ +#define _OS_COMMON_H_ + +/* +** At least two bugs have slipped in because we changed the MEMORY_DEBUG +** macro to SQLITE_DEBUG and some older makefiles have not yet made the +** switch. The following code should catch this problem at compile-time. +*/ +#ifdef MEMORY_DEBUG +# error "The MEMORY_DEBUG macro is obsolete. Use SQLITE_DEBUG instead." +#endif + +/* +** Macros for performance tracing. Normally turned off. Only works +** on i486 hardware. +*/ +#ifdef SQLITE_PERFORMANCE_TRACE + +/* +** hwtime.h contains inline assembler code for implementing +** high-performance timing routines. +*/ +/************** Include hwtime.h in the middle of os_common.h ****************/ +/************** Begin file hwtime.h ******************************************/ +/* +** 2008 May 27 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +****************************************************************************** +** +** This file contains inline asm code for retrieving "high-performance" +** counters for x86 class CPUs. +*/ +#ifndef SQLITE_HWTIME_H +#define SQLITE_HWTIME_H + +/* +** The following routine only works on pentium-class (or newer) processors. +** It uses the RDTSC opcode to read the cycle count value out of the +** processor and returns that value. This can be used for high-res +** profiling. +*/ +#if (defined(__GNUC__) || defined(_MSC_VER)) && \ + (defined(i386) || defined(__i386__) || defined(_M_IX86)) + + #if defined(__GNUC__) + + __inline__ sqlite_uint64 sqlite3Hwtime(void){ + unsigned int lo, hi; + __asm__ __volatile__ ("rdtsc" : "=a" (lo), "=d" (hi)); + return (sqlite_uint64)hi << 32 | lo; + } + + #elif defined(_MSC_VER) + + __declspec(naked) __inline sqlite_uint64 __cdecl sqlite3Hwtime(void){ + __asm { + rdtsc + ret ; return value at EDX:EAX + } + } + + #endif + +#elif (defined(__GNUC__) && defined(__x86_64__)) + + __inline__ sqlite_uint64 sqlite3Hwtime(void){ + unsigned long val; + __asm__ __volatile__ ("rdtsc" : "=A" (val)); + return val; + } + +#elif (defined(__GNUC__) && defined(__ppc__)) + + __inline__ sqlite_uint64 sqlite3Hwtime(void){ + unsigned long long retval; + unsigned long junk; + __asm__ __volatile__ ("\n\ + 1: mftbu %1\n\ + mftb %L0\n\ + mftbu %0\n\ + cmpw %0,%1\n\ + bne 1b" + : "=r" (retval), "=r" (junk)); + return retval; + } + +#else + + #error Need implementation of sqlite3Hwtime() for your platform. + + /* + ** To compile without implementing sqlite3Hwtime() for your platform, + ** you can remove the above #error and use the following + ** stub function. You will lose timing support for many + ** of the debugging and testing utilities, but it should at + ** least compile and run. + */ +SQLITE_PRIVATE sqlite_uint64 sqlite3Hwtime(void){ return ((sqlite_uint64)0); } + +#endif + +#endif /* !defined(SQLITE_HWTIME_H) */ + +/************** End of hwtime.h **********************************************/ +/************** Continuing where we left off in os_common.h ******************/ + +static sqlite_uint64 g_start; +static sqlite_uint64 g_elapsed; +#define TIMER_START g_start=sqlite3Hwtime() +#define TIMER_END g_elapsed=sqlite3Hwtime()-g_start +#define TIMER_ELAPSED g_elapsed +#else +#define TIMER_START +#define TIMER_END +#define TIMER_ELAPSED ((sqlite_uint64)0) +#endif + +/* +** If we compile with the SQLITE_TEST macro set, then the following block +** of code will give us the ability to simulate a disk I/O error. This +** is used for testing the I/O recovery logic. +*/ +#if defined(SQLITE_TEST) +SQLITE_API extern int sqlite3_io_error_hit; +SQLITE_API extern int sqlite3_io_error_hardhit; +SQLITE_API extern int sqlite3_io_error_pending; +SQLITE_API extern int sqlite3_io_error_persist; +SQLITE_API extern int sqlite3_io_error_benign; +SQLITE_API extern int sqlite3_diskfull_pending; +SQLITE_API extern int sqlite3_diskfull; +#define SimulateIOErrorBenign(X) sqlite3_io_error_benign=(X) +#define SimulateIOError(CODE) \ + if( (sqlite3_io_error_persist && sqlite3_io_error_hit) \ + || sqlite3_io_error_pending-- == 1 ) \ + { local_ioerr(); CODE; } +static void local_ioerr(){ + IOTRACE(("IOERR\n")); + sqlite3_io_error_hit++; + if( !sqlite3_io_error_benign ) sqlite3_io_error_hardhit++; +} +#define SimulateDiskfullError(CODE) \ + if( sqlite3_diskfull_pending ){ \ + if( sqlite3_diskfull_pending == 1 ){ \ + local_ioerr(); \ + sqlite3_diskfull = 1; \ + sqlite3_io_error_hit = 1; \ + CODE; \ + }else{ \ + sqlite3_diskfull_pending--; \ + } \ + } +#else +#define SimulateIOErrorBenign(X) +#define SimulateIOError(A) +#define SimulateDiskfullError(A) +#endif /* defined(SQLITE_TEST) */ + +/* +** When testing, keep a count of the number of open files. +*/ +#if defined(SQLITE_TEST) +SQLITE_API extern int sqlite3_open_file_count; +#define OpenCounter(X) sqlite3_open_file_count+=(X) +#else +#define OpenCounter(X) +#endif /* defined(SQLITE_TEST) */ + +#endif /* !defined(_OS_COMMON_H_) */ + +/************** End of os_common.h *******************************************/ +/************** Continuing where we left off in os_win.c *********************/ + +/* +** Include the header file for the Windows VFS. +*/ +/* #include "os_win.h" */ + +/* +** Compiling and using WAL mode requires several APIs that are only +** available in Windows platforms based on the NT kernel. +*/ +#if !SQLITE_OS_WINNT && !defined(SQLITE_OMIT_WAL) +# error "WAL mode requires support from the Windows NT kernel, compile\ + with SQLITE_OMIT_WAL." +#endif + +#if !SQLITE_OS_WINNT && SQLITE_MAX_MMAP_SIZE>0 +# error "Memory mapped files require support from the Windows NT kernel,\ + compile with SQLITE_MAX_MMAP_SIZE=0." +#endif + +/* +** Are most of the Win32 ANSI APIs available (i.e. with certain exceptions +** based on the sub-platform)? +*/ +#if !SQLITE_OS_WINCE && !SQLITE_OS_WINRT && !defined(SQLITE_WIN32_NO_ANSI) +# define SQLITE_WIN32_HAS_ANSI +#endif + +/* +** Are most of the Win32 Unicode APIs available (i.e. with certain exceptions +** based on the sub-platform)? +*/ +#if (SQLITE_OS_WINCE || SQLITE_OS_WINNT || SQLITE_OS_WINRT) && \ + !defined(SQLITE_WIN32_NO_WIDE) +# define SQLITE_WIN32_HAS_WIDE +#endif + +/* +** Make sure at least one set of Win32 APIs is available. +*/ +#if !defined(SQLITE_WIN32_HAS_ANSI) && !defined(SQLITE_WIN32_HAS_WIDE) +# error "At least one of SQLITE_WIN32_HAS_ANSI and SQLITE_WIN32_HAS_WIDE\ + must be defined." +#endif + +/* +** Define the required Windows SDK version constants if they are not +** already available. +*/ +#ifndef NTDDI_WIN8 +# define NTDDI_WIN8 0x06020000 +#endif + +#ifndef NTDDI_WINBLUE +# define NTDDI_WINBLUE 0x06030000 +#endif + +#ifndef NTDDI_WINTHRESHOLD +# define NTDDI_WINTHRESHOLD 0x06040000 +#endif + +/* +** Check to see if the GetVersionEx[AW] functions are deprecated on the +** target system. GetVersionEx was first deprecated in Win8.1. +*/ +#ifndef SQLITE_WIN32_GETVERSIONEX +# if defined(NTDDI_VERSION) && NTDDI_VERSION >= NTDDI_WINBLUE +# define SQLITE_WIN32_GETVERSIONEX 0 /* GetVersionEx() is deprecated */ +# else +# define SQLITE_WIN32_GETVERSIONEX 1 /* GetVersionEx() is current */ +# endif +#endif + +/* +** Check to see if the CreateFileMappingA function is supported on the +** target system. It is unavailable when using "mincore.lib" on Win10. +** When compiling for Windows 10, always assume "mincore.lib" is in use. +*/ +#ifndef SQLITE_WIN32_CREATEFILEMAPPINGA +# if defined(NTDDI_VERSION) && NTDDI_VERSION >= NTDDI_WINTHRESHOLD +# define SQLITE_WIN32_CREATEFILEMAPPINGA 0 +# else +# define SQLITE_WIN32_CREATEFILEMAPPINGA 1 +# endif +#endif + +/* +** This constant should already be defined (in the "WinDef.h" SDK file). +*/ +#ifndef MAX_PATH +# define MAX_PATH (260) +#endif + +/* +** Maximum pathname length (in chars) for Win32. This should normally be +** MAX_PATH. +*/ +#ifndef SQLITE_WIN32_MAX_PATH_CHARS +# define SQLITE_WIN32_MAX_PATH_CHARS (MAX_PATH) +#endif + +/* +** This constant should already be defined (in the "WinNT.h" SDK file). +*/ +#ifndef UNICODE_STRING_MAX_CHARS +# define UNICODE_STRING_MAX_CHARS (32767) +#endif + +/* +** Maximum pathname length (in chars) for WinNT. This should normally be +** UNICODE_STRING_MAX_CHARS. +*/ +#ifndef SQLITE_WINNT_MAX_PATH_CHARS +# define SQLITE_WINNT_MAX_PATH_CHARS (UNICODE_STRING_MAX_CHARS) +#endif + +/* +** Maximum pathname length (in bytes) for Win32. The MAX_PATH macro is in +** characters, so we allocate 4 bytes per character assuming worst-case of +** 4-bytes-per-character for UTF8. +*/ +#ifndef SQLITE_WIN32_MAX_PATH_BYTES +# define SQLITE_WIN32_MAX_PATH_BYTES (SQLITE_WIN32_MAX_PATH_CHARS*4) +#endif + +/* +** Maximum pathname length (in bytes) for WinNT. This should normally be +** UNICODE_STRING_MAX_CHARS * sizeof(WCHAR). +*/ +#ifndef SQLITE_WINNT_MAX_PATH_BYTES +# define SQLITE_WINNT_MAX_PATH_BYTES \ + (sizeof(WCHAR) * SQLITE_WINNT_MAX_PATH_CHARS) +#endif + +/* +** Maximum error message length (in chars) for WinRT. +*/ +#ifndef SQLITE_WIN32_MAX_ERRMSG_CHARS +# define SQLITE_WIN32_MAX_ERRMSG_CHARS (1024) +#endif + +/* +** Returns non-zero if the character should be treated as a directory +** separator. +*/ +#ifndef winIsDirSep +# define winIsDirSep(a) (((a) == '/') || ((a) == '\\')) +#endif + +/* +** This macro is used when a local variable is set to a value that is +** [sometimes] not used by the code (e.g. via conditional compilation). +*/ +#ifndef UNUSED_VARIABLE_VALUE +# define UNUSED_VARIABLE_VALUE(x) (void)(x) +#endif + +/* +** Returns the character that should be used as the directory separator. +*/ +#ifndef winGetDirSep +# define winGetDirSep() '\\' +#endif + +/* +** Do we need to manually define the Win32 file mapping APIs for use with WAL +** mode or memory mapped files (e.g. these APIs are available in the Windows +** CE SDK; however, they are not present in the header file)? +*/ +#if SQLITE_WIN32_FILEMAPPING_API && \ + (!defined(SQLITE_OMIT_WAL) || SQLITE_MAX_MMAP_SIZE>0) +/* +** Two of the file mapping APIs are different under WinRT. Figure out which +** set we need. +*/ +#if SQLITE_OS_WINRT +WINBASEAPI HANDLE WINAPI CreateFileMappingFromApp(HANDLE, \ + LPSECURITY_ATTRIBUTES, ULONG, ULONG64, LPCWSTR); + +WINBASEAPI LPVOID WINAPI MapViewOfFileFromApp(HANDLE, ULONG, ULONG64, SIZE_T); +#else +#if defined(SQLITE_WIN32_HAS_ANSI) +WINBASEAPI HANDLE WINAPI CreateFileMappingA(HANDLE, LPSECURITY_ATTRIBUTES, \ + DWORD, DWORD, DWORD, LPCSTR); +#endif /* defined(SQLITE_WIN32_HAS_ANSI) */ + +#if defined(SQLITE_WIN32_HAS_WIDE) +WINBASEAPI HANDLE WINAPI CreateFileMappingW(HANDLE, LPSECURITY_ATTRIBUTES, \ + DWORD, DWORD, DWORD, LPCWSTR); +#endif /* defined(SQLITE_WIN32_HAS_WIDE) */ + +WINBASEAPI LPVOID WINAPI MapViewOfFile(HANDLE, DWORD, DWORD, DWORD, SIZE_T); +#endif /* SQLITE_OS_WINRT */ + +/* +** These file mapping APIs are common to both Win32 and WinRT. +*/ + +WINBASEAPI BOOL WINAPI FlushViewOfFile(LPCVOID, SIZE_T); +WINBASEAPI BOOL WINAPI UnmapViewOfFile(LPCVOID); +#endif /* SQLITE_WIN32_FILEMAPPING_API */ + +/* +** Some Microsoft compilers lack this definition. +*/ +#ifndef INVALID_FILE_ATTRIBUTES +# define INVALID_FILE_ATTRIBUTES ((DWORD)-1) +#endif + +#ifndef FILE_FLAG_MASK +# define FILE_FLAG_MASK (0xFF3C0000) +#endif + +#ifndef FILE_ATTRIBUTE_MASK +# define FILE_ATTRIBUTE_MASK (0x0003FFF7) +#endif + +#ifndef SQLITE_OMIT_WAL +/* Forward references to structures used for WAL */ +typedef struct winShm winShm; /* A connection to shared-memory */ +typedef struct winShmNode winShmNode; /* A region of shared-memory */ +#endif + +/* +** WinCE lacks native support for file locking so we have to fake it +** with some code of our own. +*/ +#if SQLITE_OS_WINCE +typedef struct winceLock { + int nReaders; /* Number of reader locks obtained */ + BOOL bPending; /* Indicates a pending lock has been obtained */ + BOOL bReserved; /* Indicates a reserved lock has been obtained */ + BOOL bExclusive; /* Indicates an exclusive lock has been obtained */ +} winceLock; +#endif + +/* +** The winFile structure is a subclass of sqlite3_file* specific to the win32 +** portability layer. +*/ +typedef struct winFile winFile; +struct winFile { + const sqlite3_io_methods *pMethod; /*** Must be first ***/ + sqlite3_vfs *pVfs; /* The VFS used to open this file */ + HANDLE h; /* Handle for accessing the file */ + u8 locktype; /* Type of lock currently held on this file */ + short sharedLockByte; /* Randomly chosen byte used as a shared lock */ + u8 ctrlFlags; /* Flags. See WINFILE_* below */ + DWORD lastErrno; /* The Windows errno from the last I/O error */ +#ifndef SQLITE_OMIT_WAL + winShm *pShm; /* Instance of shared memory on this file */ +#endif + const char *zPath; /* Full pathname of this file */ + int szChunk; /* Chunk size configured by FCNTL_CHUNK_SIZE */ +#if SQLITE_OS_WINCE + LPWSTR zDeleteOnClose; /* Name of file to delete when closing */ + HANDLE hMutex; /* Mutex used to control access to shared lock */ + HANDLE hShared; /* Shared memory segment used for locking */ + winceLock local; /* Locks obtained by this instance of winFile */ + winceLock *shared; /* Global shared lock memory for the file */ +#endif +#if SQLITE_MAX_MMAP_SIZE>0 + int nFetchOut; /* Number of outstanding xFetch references */ + HANDLE hMap; /* Handle for accessing memory mapping */ + void *pMapRegion; /* Area memory mapped */ + sqlite3_int64 mmapSize; /* Size of mapped region */ + sqlite3_int64 mmapSizeMax; /* Configured FCNTL_MMAP_SIZE value */ +#endif +}; + +/* +** The winVfsAppData structure is used for the pAppData member for all of the +** Win32 VFS variants. +*/ +typedef struct winVfsAppData winVfsAppData; +struct winVfsAppData { + const sqlite3_io_methods *pMethod; /* The file I/O methods to use. */ + void *pAppData; /* The extra pAppData, if any. */ + BOOL bNoLock; /* Non-zero if locking is disabled. */ +}; + +/* +** Allowed values for winFile.ctrlFlags +*/ +#define WINFILE_RDONLY 0x02 /* Connection is read only */ +#define WINFILE_PERSIST_WAL 0x04 /* Persistent WAL mode */ +#define WINFILE_PSOW 0x10 /* SQLITE_IOCAP_POWERSAFE_OVERWRITE */ + +/* + * The size of the buffer used by sqlite3_win32_write_debug(). + */ +#ifndef SQLITE_WIN32_DBG_BUF_SIZE +# define SQLITE_WIN32_DBG_BUF_SIZE ((int)(4096-sizeof(DWORD))) +#endif + +/* + * If compiled with SQLITE_WIN32_MALLOC on Windows, we will use the + * various Win32 API heap functions instead of our own. + */ +#ifdef SQLITE_WIN32_MALLOC + +/* + * If this is non-zero, an isolated heap will be created by the native Win32 + * allocator subsystem; otherwise, the default process heap will be used. This + * setting has no effect when compiling for WinRT. By default, this is enabled + * and an isolated heap will be created to store all allocated data. + * + ****************************************************************************** + * WARNING: It is important to note that when this setting is non-zero and the + * winMemShutdown function is called (e.g. by the sqlite3_shutdown + * function), all data that was allocated using the isolated heap will + * be freed immediately and any attempt to access any of that freed + * data will almost certainly result in an immediate access violation. + ****************************************************************************** + */ +#ifndef SQLITE_WIN32_HEAP_CREATE +# define SQLITE_WIN32_HEAP_CREATE (TRUE) +#endif + +/* + * This is the maximum possible initial size of the Win32-specific heap, in + * bytes. + */ +#ifndef SQLITE_WIN32_HEAP_MAX_INIT_SIZE +# define SQLITE_WIN32_HEAP_MAX_INIT_SIZE (4294967295U) +#endif + +/* + * This is the extra space for the initial size of the Win32-specific heap, + * in bytes. This value may be zero. + */ +#ifndef SQLITE_WIN32_HEAP_INIT_EXTRA +# define SQLITE_WIN32_HEAP_INIT_EXTRA (4194304) +#endif + +/* + * Calculate the maximum legal cache size, in pages, based on the maximum + * possible initial heap size and the default page size, setting aside the + * needed extra space. + */ +#ifndef SQLITE_WIN32_MAX_CACHE_SIZE +# define SQLITE_WIN32_MAX_CACHE_SIZE (((SQLITE_WIN32_HEAP_MAX_INIT_SIZE) - \ + (SQLITE_WIN32_HEAP_INIT_EXTRA)) / \ + (SQLITE_DEFAULT_PAGE_SIZE)) +#endif + +/* + * This is cache size used in the calculation of the initial size of the + * Win32-specific heap. It cannot be negative. + */ +#ifndef SQLITE_WIN32_CACHE_SIZE +# if SQLITE_DEFAULT_CACHE_SIZE>=0 +# define SQLITE_WIN32_CACHE_SIZE (SQLITE_DEFAULT_CACHE_SIZE) +# else +# define SQLITE_WIN32_CACHE_SIZE (-(SQLITE_DEFAULT_CACHE_SIZE)) +# endif +#endif + +/* + * Make sure that the calculated cache size, in pages, cannot cause the + * initial size of the Win32-specific heap to exceed the maximum amount + * of memory that can be specified in the call to HeapCreate. + */ +#if SQLITE_WIN32_CACHE_SIZE>SQLITE_WIN32_MAX_CACHE_SIZE +# undef SQLITE_WIN32_CACHE_SIZE +# define SQLITE_WIN32_CACHE_SIZE (2000) +#endif + +/* + * The initial size of the Win32-specific heap. This value may be zero. + */ +#ifndef SQLITE_WIN32_HEAP_INIT_SIZE +# define SQLITE_WIN32_HEAP_INIT_SIZE ((SQLITE_WIN32_CACHE_SIZE) * \ + (SQLITE_DEFAULT_PAGE_SIZE) + \ + (SQLITE_WIN32_HEAP_INIT_EXTRA)) +#endif + +/* + * The maximum size of the Win32-specific heap. This value may be zero. + */ +#ifndef SQLITE_WIN32_HEAP_MAX_SIZE +# define SQLITE_WIN32_HEAP_MAX_SIZE (0) +#endif + +/* + * The extra flags to use in calls to the Win32 heap APIs. This value may be + * zero for the default behavior. + */ +#ifndef SQLITE_WIN32_HEAP_FLAGS +# define SQLITE_WIN32_HEAP_FLAGS (0) +#endif + + +/* +** The winMemData structure stores information required by the Win32-specific +** sqlite3_mem_methods implementation. +*/ +typedef struct winMemData winMemData; +struct winMemData { +#ifndef NDEBUG + u32 magic1; /* Magic number to detect structure corruption. */ +#endif + HANDLE hHeap; /* The handle to our heap. */ + BOOL bOwned; /* Do we own the heap (i.e. destroy it on shutdown)? */ +#ifndef NDEBUG + u32 magic2; /* Magic number to detect structure corruption. */ +#endif +}; + +#ifndef NDEBUG +#define WINMEM_MAGIC1 0x42b2830b +#define WINMEM_MAGIC2 0xbd4d7cf4 +#endif + +static struct winMemData win_mem_data = { +#ifndef NDEBUG + WINMEM_MAGIC1, +#endif + NULL, FALSE +#ifndef NDEBUG + ,WINMEM_MAGIC2 +#endif +}; + +#ifndef NDEBUG +#define winMemAssertMagic1() assert( win_mem_data.magic1==WINMEM_MAGIC1 ) +#define winMemAssertMagic2() assert( win_mem_data.magic2==WINMEM_MAGIC2 ) +#define winMemAssertMagic() winMemAssertMagic1(); winMemAssertMagic2(); +#else +#define winMemAssertMagic() +#endif + +#define winMemGetDataPtr() &win_mem_data +#define winMemGetHeap() win_mem_data.hHeap +#define winMemGetOwned() win_mem_data.bOwned + +static void *winMemMalloc(int nBytes); +static void winMemFree(void *pPrior); +static void *winMemRealloc(void *pPrior, int nBytes); +static int winMemSize(void *p); +static int winMemRoundup(int n); +static int winMemInit(void *pAppData); +static void winMemShutdown(void *pAppData); + +SQLITE_PRIVATE const sqlite3_mem_methods *sqlite3MemGetWin32(void); +#endif /* SQLITE_WIN32_MALLOC */ + +/* +** The following variable is (normally) set once and never changes +** thereafter. It records whether the operating system is Win9x +** or WinNT. +** +** 0: Operating system unknown. +** 1: Operating system is Win9x. +** 2: Operating system is WinNT. +** +** In order to facilitate testing on a WinNT system, the test fixture +** can manually set this value to 1 to emulate Win98 behavior. +*/ +#ifdef SQLITE_TEST +SQLITE_API LONG SQLITE_WIN32_VOLATILE sqlite3_os_type = 0; +#else +static LONG SQLITE_WIN32_VOLATILE sqlite3_os_type = 0; +#endif + +#ifndef SYSCALL +# define SYSCALL sqlite3_syscall_ptr +#endif + +/* +** This function is not available on Windows CE or WinRT. + */ + +#if SQLITE_OS_WINCE || SQLITE_OS_WINRT +# define osAreFileApisANSI() 1 +#endif + +/* +** Many system calls are accessed through pointer-to-functions so that +** they may be overridden at runtime to facilitate fault injection during +** testing and sandboxing. The following array holds the names and pointers +** to all overrideable system calls. +*/ +static struct win_syscall { + const char *zName; /* Name of the system call */ + sqlite3_syscall_ptr pCurrent; /* Current value of the system call */ + sqlite3_syscall_ptr pDefault; /* Default value */ +} aSyscall[] = { +#if !SQLITE_OS_WINCE && !SQLITE_OS_WINRT + { "AreFileApisANSI", (SYSCALL)AreFileApisANSI, 0 }, +#else + { "AreFileApisANSI", (SYSCALL)0, 0 }, +#endif + +#ifndef osAreFileApisANSI +#define osAreFileApisANSI ((BOOL(WINAPI*)(VOID))aSyscall[0].pCurrent) +#endif + +#if SQLITE_OS_WINCE && defined(SQLITE_WIN32_HAS_WIDE) + { "CharLowerW", (SYSCALL)CharLowerW, 0 }, +#else + { "CharLowerW", (SYSCALL)0, 0 }, +#endif + +#define osCharLowerW ((LPWSTR(WINAPI*)(LPWSTR))aSyscall[1].pCurrent) + +#if SQLITE_OS_WINCE && defined(SQLITE_WIN32_HAS_WIDE) + { "CharUpperW", (SYSCALL)CharUpperW, 0 }, +#else + { "CharUpperW", (SYSCALL)0, 0 }, +#endif + +#define osCharUpperW ((LPWSTR(WINAPI*)(LPWSTR))aSyscall[2].pCurrent) + + { "CloseHandle", (SYSCALL)CloseHandle, 0 }, + +#define osCloseHandle ((BOOL(WINAPI*)(HANDLE))aSyscall[3].pCurrent) + +#if defined(SQLITE_WIN32_HAS_ANSI) + { "CreateFileA", (SYSCALL)CreateFileA, 0 }, +#else + { "CreateFileA", (SYSCALL)0, 0 }, +#endif + +#define osCreateFileA ((HANDLE(WINAPI*)(LPCSTR,DWORD,DWORD, \ + LPSECURITY_ATTRIBUTES,DWORD,DWORD,HANDLE))aSyscall[4].pCurrent) + +#if !SQLITE_OS_WINRT && defined(SQLITE_WIN32_HAS_WIDE) + { "CreateFileW", (SYSCALL)CreateFileW, 0 }, +#else + { "CreateFileW", (SYSCALL)0, 0 }, +#endif + +#define osCreateFileW ((HANDLE(WINAPI*)(LPCWSTR,DWORD,DWORD, \ + LPSECURITY_ATTRIBUTES,DWORD,DWORD,HANDLE))aSyscall[5].pCurrent) + +#if !SQLITE_OS_WINRT && defined(SQLITE_WIN32_HAS_ANSI) && \ + (!defined(SQLITE_OMIT_WAL) || SQLITE_MAX_MMAP_SIZE>0) && \ + SQLITE_WIN32_CREATEFILEMAPPINGA + { "CreateFileMappingA", (SYSCALL)CreateFileMappingA, 0 }, +#else + { "CreateFileMappingA", (SYSCALL)0, 0 }, +#endif + +#define osCreateFileMappingA ((HANDLE(WINAPI*)(HANDLE,LPSECURITY_ATTRIBUTES, \ + DWORD,DWORD,DWORD,LPCSTR))aSyscall[6].pCurrent) + +#if SQLITE_OS_WINCE || (!SQLITE_OS_WINRT && defined(SQLITE_WIN32_HAS_WIDE) && \ + (!defined(SQLITE_OMIT_WAL) || SQLITE_MAX_MMAP_SIZE>0)) + { "CreateFileMappingW", (SYSCALL)CreateFileMappingW, 0 }, +#else + { "CreateFileMappingW", (SYSCALL)0, 0 }, +#endif + +#define osCreateFileMappingW ((HANDLE(WINAPI*)(HANDLE,LPSECURITY_ATTRIBUTES, \ + DWORD,DWORD,DWORD,LPCWSTR))aSyscall[7].pCurrent) + +#if !SQLITE_OS_WINRT && defined(SQLITE_WIN32_HAS_WIDE) + { "CreateMutexW", (SYSCALL)CreateMutexW, 0 }, +#else + { "CreateMutexW", (SYSCALL)0, 0 }, +#endif + +#define osCreateMutexW ((HANDLE(WINAPI*)(LPSECURITY_ATTRIBUTES,BOOL, \ + LPCWSTR))aSyscall[8].pCurrent) + +#if defined(SQLITE_WIN32_HAS_ANSI) + { "DeleteFileA", (SYSCALL)DeleteFileA, 0 }, +#else + { "DeleteFileA", (SYSCALL)0, 0 }, +#endif + +#define osDeleteFileA ((BOOL(WINAPI*)(LPCSTR))aSyscall[9].pCurrent) + +#if defined(SQLITE_WIN32_HAS_WIDE) + { "DeleteFileW", (SYSCALL)DeleteFileW, 0 }, +#else + { "DeleteFileW", (SYSCALL)0, 0 }, +#endif + +#define osDeleteFileW ((BOOL(WINAPI*)(LPCWSTR))aSyscall[10].pCurrent) + +#if SQLITE_OS_WINCE + { "FileTimeToLocalFileTime", (SYSCALL)FileTimeToLocalFileTime, 0 }, +#else + { "FileTimeToLocalFileTime", (SYSCALL)0, 0 }, +#endif + +#define osFileTimeToLocalFileTime ((BOOL(WINAPI*)(CONST FILETIME*, \ + LPFILETIME))aSyscall[11].pCurrent) + +#if SQLITE_OS_WINCE + { "FileTimeToSystemTime", (SYSCALL)FileTimeToSystemTime, 0 }, +#else + { "FileTimeToSystemTime", (SYSCALL)0, 0 }, +#endif + +#define osFileTimeToSystemTime ((BOOL(WINAPI*)(CONST FILETIME*, \ + LPSYSTEMTIME))aSyscall[12].pCurrent) + + { "FlushFileBuffers", (SYSCALL)FlushFileBuffers, 0 }, + +#define osFlushFileBuffers ((BOOL(WINAPI*)(HANDLE))aSyscall[13].pCurrent) + +#if defined(SQLITE_WIN32_HAS_ANSI) + { "FormatMessageA", (SYSCALL)FormatMessageA, 0 }, +#else + { "FormatMessageA", (SYSCALL)0, 0 }, +#endif + +#define osFormatMessageA ((DWORD(WINAPI*)(DWORD,LPCVOID,DWORD,DWORD,LPSTR, \ + DWORD,va_list*))aSyscall[14].pCurrent) + +#if defined(SQLITE_WIN32_HAS_WIDE) + { "FormatMessageW", (SYSCALL)FormatMessageW, 0 }, +#else + { "FormatMessageW", (SYSCALL)0, 0 }, +#endif + +#define osFormatMessageW ((DWORD(WINAPI*)(DWORD,LPCVOID,DWORD,DWORD,LPWSTR, \ + DWORD,va_list*))aSyscall[15].pCurrent) + +#if !defined(SQLITE_OMIT_LOAD_EXTENSION) + { "FreeLibrary", (SYSCALL)FreeLibrary, 0 }, +#else + { "FreeLibrary", (SYSCALL)0, 0 }, +#endif + +#define osFreeLibrary ((BOOL(WINAPI*)(HMODULE))aSyscall[16].pCurrent) + + { "GetCurrentProcessId", (SYSCALL)GetCurrentProcessId, 0 }, + +#define osGetCurrentProcessId ((DWORD(WINAPI*)(VOID))aSyscall[17].pCurrent) + +#if !SQLITE_OS_WINCE && defined(SQLITE_WIN32_HAS_ANSI) + { "GetDiskFreeSpaceA", (SYSCALL)GetDiskFreeSpaceA, 0 }, +#else + { "GetDiskFreeSpaceA", (SYSCALL)0, 0 }, +#endif + +#define osGetDiskFreeSpaceA ((BOOL(WINAPI*)(LPCSTR,LPDWORD,LPDWORD,LPDWORD, \ + LPDWORD))aSyscall[18].pCurrent) + +#if !SQLITE_OS_WINCE && !SQLITE_OS_WINRT && defined(SQLITE_WIN32_HAS_WIDE) + { "GetDiskFreeSpaceW", (SYSCALL)GetDiskFreeSpaceW, 0 }, +#else + { "GetDiskFreeSpaceW", (SYSCALL)0, 0 }, +#endif + +#define osGetDiskFreeSpaceW ((BOOL(WINAPI*)(LPCWSTR,LPDWORD,LPDWORD,LPDWORD, \ + LPDWORD))aSyscall[19].pCurrent) + +#if defined(SQLITE_WIN32_HAS_ANSI) + { "GetFileAttributesA", (SYSCALL)GetFileAttributesA, 0 }, +#else + { "GetFileAttributesA", (SYSCALL)0, 0 }, +#endif + +#define osGetFileAttributesA ((DWORD(WINAPI*)(LPCSTR))aSyscall[20].pCurrent) + +#if !SQLITE_OS_WINRT && defined(SQLITE_WIN32_HAS_WIDE) + { "GetFileAttributesW", (SYSCALL)GetFileAttributesW, 0 }, +#else + { "GetFileAttributesW", (SYSCALL)0, 0 }, +#endif + +#define osGetFileAttributesW ((DWORD(WINAPI*)(LPCWSTR))aSyscall[21].pCurrent) + +#if defined(SQLITE_WIN32_HAS_WIDE) + { "GetFileAttributesExW", (SYSCALL)GetFileAttributesExW, 0 }, +#else + { "GetFileAttributesExW", (SYSCALL)0, 0 }, +#endif + +#define osGetFileAttributesExW ((BOOL(WINAPI*)(LPCWSTR,GET_FILEEX_INFO_LEVELS, \ + LPVOID))aSyscall[22].pCurrent) + +#if !SQLITE_OS_WINRT + { "GetFileSize", (SYSCALL)GetFileSize, 0 }, +#else + { "GetFileSize", (SYSCALL)0, 0 }, +#endif + +#define osGetFileSize ((DWORD(WINAPI*)(HANDLE,LPDWORD))aSyscall[23].pCurrent) + +#if !SQLITE_OS_WINCE && defined(SQLITE_WIN32_HAS_ANSI) + { "GetFullPathNameA", (SYSCALL)GetFullPathNameA, 0 }, +#else + { "GetFullPathNameA", (SYSCALL)0, 0 }, +#endif + +#define osGetFullPathNameA ((DWORD(WINAPI*)(LPCSTR,DWORD,LPSTR, \ + LPSTR*))aSyscall[24].pCurrent) + +#if !SQLITE_OS_WINCE && !SQLITE_OS_WINRT && defined(SQLITE_WIN32_HAS_WIDE) + { "GetFullPathNameW", (SYSCALL)GetFullPathNameW, 0 }, +#else + { "GetFullPathNameW", (SYSCALL)0, 0 }, +#endif + +#define osGetFullPathNameW ((DWORD(WINAPI*)(LPCWSTR,DWORD,LPWSTR, \ + LPWSTR*))aSyscall[25].pCurrent) + + { "GetLastError", (SYSCALL)GetLastError, 0 }, + +#define osGetLastError ((DWORD(WINAPI*)(VOID))aSyscall[26].pCurrent) + +#if !defined(SQLITE_OMIT_LOAD_EXTENSION) +#if SQLITE_OS_WINCE + /* The GetProcAddressA() routine is only available on Windows CE. */ + { "GetProcAddressA", (SYSCALL)GetProcAddressA, 0 }, +#else + /* All other Windows platforms expect GetProcAddress() to take + ** an ANSI string regardless of the _UNICODE setting */ + { "GetProcAddressA", (SYSCALL)GetProcAddress, 0 }, +#endif +#else + { "GetProcAddressA", (SYSCALL)0, 0 }, +#endif + +#define osGetProcAddressA ((FARPROC(WINAPI*)(HMODULE, \ + LPCSTR))aSyscall[27].pCurrent) + +#if !SQLITE_OS_WINRT + { "GetSystemInfo", (SYSCALL)GetSystemInfo, 0 }, +#else + { "GetSystemInfo", (SYSCALL)0, 0 }, +#endif + +#define osGetSystemInfo ((VOID(WINAPI*)(LPSYSTEM_INFO))aSyscall[28].pCurrent) + + { "GetSystemTime", (SYSCALL)GetSystemTime, 0 }, + +#define osGetSystemTime ((VOID(WINAPI*)(LPSYSTEMTIME))aSyscall[29].pCurrent) + +#if !SQLITE_OS_WINCE + { "GetSystemTimeAsFileTime", (SYSCALL)GetSystemTimeAsFileTime, 0 }, +#else + { "GetSystemTimeAsFileTime", (SYSCALL)0, 0 }, +#endif + +#define osGetSystemTimeAsFileTime ((VOID(WINAPI*)( \ + LPFILETIME))aSyscall[30].pCurrent) + +#if defined(SQLITE_WIN32_HAS_ANSI) + { "GetTempPathA", (SYSCALL)GetTempPathA, 0 }, +#else + { "GetTempPathA", (SYSCALL)0, 0 }, +#endif + +#define osGetTempPathA ((DWORD(WINAPI*)(DWORD,LPSTR))aSyscall[31].pCurrent) + +#if !SQLITE_OS_WINRT && defined(SQLITE_WIN32_HAS_WIDE) + { "GetTempPathW", (SYSCALL)GetTempPathW, 0 }, +#else + { "GetTempPathW", (SYSCALL)0, 0 }, +#endif + +#define osGetTempPathW ((DWORD(WINAPI*)(DWORD,LPWSTR))aSyscall[32].pCurrent) + +#if !SQLITE_OS_WINRT + { "GetTickCount", (SYSCALL)GetTickCount, 0 }, +#else + { "GetTickCount", (SYSCALL)0, 0 }, +#endif + +#define osGetTickCount ((DWORD(WINAPI*)(VOID))aSyscall[33].pCurrent) + +#if defined(SQLITE_WIN32_HAS_ANSI) && SQLITE_WIN32_GETVERSIONEX + { "GetVersionExA", (SYSCALL)GetVersionExA, 0 }, +#else + { "GetVersionExA", (SYSCALL)0, 0 }, +#endif + +#define osGetVersionExA ((BOOL(WINAPI*)( \ + LPOSVERSIONINFOA))aSyscall[34].pCurrent) + +#if !SQLITE_OS_WINRT && defined(SQLITE_WIN32_HAS_WIDE) && \ + SQLITE_WIN32_GETVERSIONEX + { "GetVersionExW", (SYSCALL)GetVersionExW, 0 }, +#else + { "GetVersionExW", (SYSCALL)0, 0 }, +#endif + +#define osGetVersionExW ((BOOL(WINAPI*)( \ + LPOSVERSIONINFOW))aSyscall[35].pCurrent) + + { "HeapAlloc", (SYSCALL)HeapAlloc, 0 }, + +#define osHeapAlloc ((LPVOID(WINAPI*)(HANDLE,DWORD, \ + SIZE_T))aSyscall[36].pCurrent) + +#if !SQLITE_OS_WINRT + { "HeapCreate", (SYSCALL)HeapCreate, 0 }, +#else + { "HeapCreate", (SYSCALL)0, 0 }, +#endif + +#define osHeapCreate ((HANDLE(WINAPI*)(DWORD,SIZE_T, \ + SIZE_T))aSyscall[37].pCurrent) + +#if !SQLITE_OS_WINRT + { "HeapDestroy", (SYSCALL)HeapDestroy, 0 }, +#else + { "HeapDestroy", (SYSCALL)0, 0 }, +#endif + +#define osHeapDestroy ((BOOL(WINAPI*)(HANDLE))aSyscall[38].pCurrent) + + { "HeapFree", (SYSCALL)HeapFree, 0 }, + +#define osHeapFree ((BOOL(WINAPI*)(HANDLE,DWORD,LPVOID))aSyscall[39].pCurrent) + + { "HeapReAlloc", (SYSCALL)HeapReAlloc, 0 }, + +#define osHeapReAlloc ((LPVOID(WINAPI*)(HANDLE,DWORD,LPVOID, \ + SIZE_T))aSyscall[40].pCurrent) + + { "HeapSize", (SYSCALL)HeapSize, 0 }, + +#define osHeapSize ((SIZE_T(WINAPI*)(HANDLE,DWORD, \ + LPCVOID))aSyscall[41].pCurrent) + +#if !SQLITE_OS_WINRT + { "HeapValidate", (SYSCALL)HeapValidate, 0 }, +#else + { "HeapValidate", (SYSCALL)0, 0 }, +#endif + +#define osHeapValidate ((BOOL(WINAPI*)(HANDLE,DWORD, \ + LPCVOID))aSyscall[42].pCurrent) + +#if !SQLITE_OS_WINCE && !SQLITE_OS_WINRT + { "HeapCompact", (SYSCALL)HeapCompact, 0 }, +#else + { "HeapCompact", (SYSCALL)0, 0 }, +#endif + +#define osHeapCompact ((UINT(WINAPI*)(HANDLE,DWORD))aSyscall[43].pCurrent) + +#if defined(SQLITE_WIN32_HAS_ANSI) && !defined(SQLITE_OMIT_LOAD_EXTENSION) + { "LoadLibraryA", (SYSCALL)LoadLibraryA, 0 }, +#else + { "LoadLibraryA", (SYSCALL)0, 0 }, +#endif + +#define osLoadLibraryA ((HMODULE(WINAPI*)(LPCSTR))aSyscall[44].pCurrent) + +#if !SQLITE_OS_WINRT && defined(SQLITE_WIN32_HAS_WIDE) && \ + !defined(SQLITE_OMIT_LOAD_EXTENSION) + { "LoadLibraryW", (SYSCALL)LoadLibraryW, 0 }, +#else + { "LoadLibraryW", (SYSCALL)0, 0 }, +#endif + +#define osLoadLibraryW ((HMODULE(WINAPI*)(LPCWSTR))aSyscall[45].pCurrent) + +#if !SQLITE_OS_WINRT + { "LocalFree", (SYSCALL)LocalFree, 0 }, +#else + { "LocalFree", (SYSCALL)0, 0 }, +#endif + +#define osLocalFree ((HLOCAL(WINAPI*)(HLOCAL))aSyscall[46].pCurrent) + +#if !SQLITE_OS_WINCE && !SQLITE_OS_WINRT + { "LockFile", (SYSCALL)LockFile, 0 }, +#else + { "LockFile", (SYSCALL)0, 0 }, +#endif + +#ifndef osLockFile +#define osLockFile ((BOOL(WINAPI*)(HANDLE,DWORD,DWORD,DWORD, \ + DWORD))aSyscall[47].pCurrent) +#endif + +#if !SQLITE_OS_WINCE + { "LockFileEx", (SYSCALL)LockFileEx, 0 }, +#else + { "LockFileEx", (SYSCALL)0, 0 }, +#endif + +#ifndef osLockFileEx +#define osLockFileEx ((BOOL(WINAPI*)(HANDLE,DWORD,DWORD,DWORD,DWORD, \ + LPOVERLAPPED))aSyscall[48].pCurrent) +#endif + +#if SQLITE_OS_WINCE || (!SQLITE_OS_WINRT && \ + (!defined(SQLITE_OMIT_WAL) || SQLITE_MAX_MMAP_SIZE>0)) + { "MapViewOfFile", (SYSCALL)MapViewOfFile, 0 }, +#else + { "MapViewOfFile", (SYSCALL)0, 0 }, +#endif + +#define osMapViewOfFile ((LPVOID(WINAPI*)(HANDLE,DWORD,DWORD,DWORD, \ + SIZE_T))aSyscall[49].pCurrent) + + { "MultiByteToWideChar", (SYSCALL)MultiByteToWideChar, 0 }, + +#define osMultiByteToWideChar ((int(WINAPI*)(UINT,DWORD,LPCSTR,int,LPWSTR, \ + int))aSyscall[50].pCurrent) + + { "QueryPerformanceCounter", (SYSCALL)QueryPerformanceCounter, 0 }, + +#define osQueryPerformanceCounter ((BOOL(WINAPI*)( \ + LARGE_INTEGER*))aSyscall[51].pCurrent) + + { "ReadFile", (SYSCALL)ReadFile, 0 }, + +#define osReadFile ((BOOL(WINAPI*)(HANDLE,LPVOID,DWORD,LPDWORD, \ + LPOVERLAPPED))aSyscall[52].pCurrent) + + { "SetEndOfFile", (SYSCALL)SetEndOfFile, 0 }, + +#define osSetEndOfFile ((BOOL(WINAPI*)(HANDLE))aSyscall[53].pCurrent) + +#if !SQLITE_OS_WINRT + { "SetFilePointer", (SYSCALL)SetFilePointer, 0 }, +#else + { "SetFilePointer", (SYSCALL)0, 0 }, +#endif + +#define osSetFilePointer ((DWORD(WINAPI*)(HANDLE,LONG,PLONG, \ + DWORD))aSyscall[54].pCurrent) + +#if !SQLITE_OS_WINRT + { "Sleep", (SYSCALL)Sleep, 0 }, +#else + { "Sleep", (SYSCALL)0, 0 }, +#endif + +#define osSleep ((VOID(WINAPI*)(DWORD))aSyscall[55].pCurrent) + + { "SystemTimeToFileTime", (SYSCALL)SystemTimeToFileTime, 0 }, + +#define osSystemTimeToFileTime ((BOOL(WINAPI*)(CONST SYSTEMTIME*, \ + LPFILETIME))aSyscall[56].pCurrent) + +#if !SQLITE_OS_WINCE && !SQLITE_OS_WINRT + { "UnlockFile", (SYSCALL)UnlockFile, 0 }, +#else + { "UnlockFile", (SYSCALL)0, 0 }, +#endif + +#ifndef osUnlockFile +#define osUnlockFile ((BOOL(WINAPI*)(HANDLE,DWORD,DWORD,DWORD, \ + DWORD))aSyscall[57].pCurrent) +#endif + +#if !SQLITE_OS_WINCE + { "UnlockFileEx", (SYSCALL)UnlockFileEx, 0 }, +#else + { "UnlockFileEx", (SYSCALL)0, 0 }, +#endif + +#define osUnlockFileEx ((BOOL(WINAPI*)(HANDLE,DWORD,DWORD,DWORD, \ + LPOVERLAPPED))aSyscall[58].pCurrent) + +#if SQLITE_OS_WINCE || !defined(SQLITE_OMIT_WAL) || SQLITE_MAX_MMAP_SIZE>0 + { "UnmapViewOfFile", (SYSCALL)UnmapViewOfFile, 0 }, +#else + { "UnmapViewOfFile", (SYSCALL)0, 0 }, +#endif + +#define osUnmapViewOfFile ((BOOL(WINAPI*)(LPCVOID))aSyscall[59].pCurrent) + + { "WideCharToMultiByte", (SYSCALL)WideCharToMultiByte, 0 }, + +#define osWideCharToMultiByte ((int(WINAPI*)(UINT,DWORD,LPCWSTR,int,LPSTR,int, \ + LPCSTR,LPBOOL))aSyscall[60].pCurrent) + + { "WriteFile", (SYSCALL)WriteFile, 0 }, + +#define osWriteFile ((BOOL(WINAPI*)(HANDLE,LPCVOID,DWORD,LPDWORD, \ + LPOVERLAPPED))aSyscall[61].pCurrent) + +#if SQLITE_OS_WINRT + { "CreateEventExW", (SYSCALL)CreateEventExW, 0 }, +#else + { "CreateEventExW", (SYSCALL)0, 0 }, +#endif + +#define osCreateEventExW ((HANDLE(WINAPI*)(LPSECURITY_ATTRIBUTES,LPCWSTR, \ + DWORD,DWORD))aSyscall[62].pCurrent) + +#if !SQLITE_OS_WINRT + { "WaitForSingleObject", (SYSCALL)WaitForSingleObject, 0 }, +#else + { "WaitForSingleObject", (SYSCALL)0, 0 }, +#endif + +#define osWaitForSingleObject ((DWORD(WINAPI*)(HANDLE, \ + DWORD))aSyscall[63].pCurrent) + +#if !SQLITE_OS_WINCE + { "WaitForSingleObjectEx", (SYSCALL)WaitForSingleObjectEx, 0 }, +#else + { "WaitForSingleObjectEx", (SYSCALL)0, 0 }, +#endif + +#define osWaitForSingleObjectEx ((DWORD(WINAPI*)(HANDLE,DWORD, \ + BOOL))aSyscall[64].pCurrent) + +#if SQLITE_OS_WINRT + { "SetFilePointerEx", (SYSCALL)SetFilePointerEx, 0 }, +#else + { "SetFilePointerEx", (SYSCALL)0, 0 }, +#endif + +#define osSetFilePointerEx ((BOOL(WINAPI*)(HANDLE,LARGE_INTEGER, \ + PLARGE_INTEGER,DWORD))aSyscall[65].pCurrent) + +#if SQLITE_OS_WINRT + { "GetFileInformationByHandleEx", (SYSCALL)GetFileInformationByHandleEx, 0 }, +#else + { "GetFileInformationByHandleEx", (SYSCALL)0, 0 }, +#endif + +#define osGetFileInformationByHandleEx ((BOOL(WINAPI*)(HANDLE, \ + FILE_INFO_BY_HANDLE_CLASS,LPVOID,DWORD))aSyscall[66].pCurrent) + +#if SQLITE_OS_WINRT && (!defined(SQLITE_OMIT_WAL) || SQLITE_MAX_MMAP_SIZE>0) + { "MapViewOfFileFromApp", (SYSCALL)MapViewOfFileFromApp, 0 }, +#else + { "MapViewOfFileFromApp", (SYSCALL)0, 0 }, +#endif + +#define osMapViewOfFileFromApp ((LPVOID(WINAPI*)(HANDLE,ULONG,ULONG64, \ + SIZE_T))aSyscall[67].pCurrent) + +#if SQLITE_OS_WINRT + { "CreateFile2", (SYSCALL)CreateFile2, 0 }, +#else + { "CreateFile2", (SYSCALL)0, 0 }, +#endif + +#define osCreateFile2 ((HANDLE(WINAPI*)(LPCWSTR,DWORD,DWORD,DWORD, \ + LPCREATEFILE2_EXTENDED_PARAMETERS))aSyscall[68].pCurrent) + +#if SQLITE_OS_WINRT && !defined(SQLITE_OMIT_LOAD_EXTENSION) + { "LoadPackagedLibrary", (SYSCALL)LoadPackagedLibrary, 0 }, +#else + { "LoadPackagedLibrary", (SYSCALL)0, 0 }, +#endif + +#define osLoadPackagedLibrary ((HMODULE(WINAPI*)(LPCWSTR, \ + DWORD))aSyscall[69].pCurrent) + +#if SQLITE_OS_WINRT + { "GetTickCount64", (SYSCALL)GetTickCount64, 0 }, +#else + { "GetTickCount64", (SYSCALL)0, 0 }, +#endif + +#define osGetTickCount64 ((ULONGLONG(WINAPI*)(VOID))aSyscall[70].pCurrent) + +#if SQLITE_OS_WINRT + { "GetNativeSystemInfo", (SYSCALL)GetNativeSystemInfo, 0 }, +#else + { "GetNativeSystemInfo", (SYSCALL)0, 0 }, +#endif + +#define osGetNativeSystemInfo ((VOID(WINAPI*)( \ + LPSYSTEM_INFO))aSyscall[71].pCurrent) + +#if defined(SQLITE_WIN32_HAS_ANSI) + { "OutputDebugStringA", (SYSCALL)OutputDebugStringA, 0 }, +#else + { "OutputDebugStringA", (SYSCALL)0, 0 }, +#endif + +#define osOutputDebugStringA ((VOID(WINAPI*)(LPCSTR))aSyscall[72].pCurrent) + +#if defined(SQLITE_WIN32_HAS_WIDE) + { "OutputDebugStringW", (SYSCALL)OutputDebugStringW, 0 }, +#else + { "OutputDebugStringW", (SYSCALL)0, 0 }, +#endif + +#define osOutputDebugStringW ((VOID(WINAPI*)(LPCWSTR))aSyscall[73].pCurrent) + + { "GetProcessHeap", (SYSCALL)GetProcessHeap, 0 }, + +#define osGetProcessHeap ((HANDLE(WINAPI*)(VOID))aSyscall[74].pCurrent) + +#if SQLITE_OS_WINRT && (!defined(SQLITE_OMIT_WAL) || SQLITE_MAX_MMAP_SIZE>0) + { "CreateFileMappingFromApp", (SYSCALL)CreateFileMappingFromApp, 0 }, +#else + { "CreateFileMappingFromApp", (SYSCALL)0, 0 }, +#endif + +#define osCreateFileMappingFromApp ((HANDLE(WINAPI*)(HANDLE, \ + LPSECURITY_ATTRIBUTES,ULONG,ULONG64,LPCWSTR))aSyscall[75].pCurrent) + +/* +** NOTE: On some sub-platforms, the InterlockedCompareExchange "function" +** is really just a macro that uses a compiler intrinsic (e.g. x64). +** So do not try to make this is into a redefinable interface. +*/ +#if defined(InterlockedCompareExchange) + { "InterlockedCompareExchange", (SYSCALL)0, 0 }, + +#define osInterlockedCompareExchange InterlockedCompareExchange +#else + { "InterlockedCompareExchange", (SYSCALL)InterlockedCompareExchange, 0 }, + +#define osInterlockedCompareExchange ((LONG(WINAPI*)(LONG \ + SQLITE_WIN32_VOLATILE*, LONG,LONG))aSyscall[76].pCurrent) +#endif /* defined(InterlockedCompareExchange) */ + +#if !SQLITE_OS_WINCE && !SQLITE_OS_WINRT && SQLITE_WIN32_USE_UUID + { "UuidCreate", (SYSCALL)UuidCreate, 0 }, +#else + { "UuidCreate", (SYSCALL)0, 0 }, +#endif + +#define osUuidCreate ((RPC_STATUS(RPC_ENTRY*)(UUID*))aSyscall[77].pCurrent) + +#if !SQLITE_OS_WINCE && !SQLITE_OS_WINRT && SQLITE_WIN32_USE_UUID + { "UuidCreateSequential", (SYSCALL)UuidCreateSequential, 0 }, +#else + { "UuidCreateSequential", (SYSCALL)0, 0 }, +#endif + +#define osUuidCreateSequential \ + ((RPC_STATUS(RPC_ENTRY*)(UUID*))aSyscall[78].pCurrent) + +#if !defined(SQLITE_NO_SYNC) && SQLITE_MAX_MMAP_SIZE>0 + { "FlushViewOfFile", (SYSCALL)FlushViewOfFile, 0 }, +#else + { "FlushViewOfFile", (SYSCALL)0, 0 }, +#endif + +#define osFlushViewOfFile \ + ((BOOL(WINAPI*)(LPCVOID,SIZE_T))aSyscall[79].pCurrent) + +}; /* End of the overrideable system calls */ + +/* +** This is the xSetSystemCall() method of sqlite3_vfs for all of the +** "win32" VFSes. Return SQLITE_OK opon successfully updating the +** system call pointer, or SQLITE_NOTFOUND if there is no configurable +** system call named zName. +*/ +static int winSetSystemCall( + sqlite3_vfs *pNotUsed, /* The VFS pointer. Not used */ + const char *zName, /* Name of system call to override */ + sqlite3_syscall_ptr pNewFunc /* Pointer to new system call value */ +){ + unsigned int i; + int rc = SQLITE_NOTFOUND; + + UNUSED_PARAMETER(pNotUsed); + if( zName==0 ){ + /* If no zName is given, restore all system calls to their default + ** settings and return NULL + */ + rc = SQLITE_OK; + for(i=0; i0 ){ + memset(zDbgBuf, 0, SQLITE_WIN32_DBG_BUF_SIZE); + memcpy(zDbgBuf, zBuf, nMin); + osOutputDebugStringA(zDbgBuf); + }else{ + osOutputDebugStringA(zBuf); + } +#elif defined(SQLITE_WIN32_HAS_WIDE) + memset(zDbgBuf, 0, SQLITE_WIN32_DBG_BUF_SIZE); + if ( osMultiByteToWideChar( + osAreFileApisANSI() ? CP_ACP : CP_OEMCP, 0, zBuf, + nMin, (LPWSTR)zDbgBuf, SQLITE_WIN32_DBG_BUF_SIZE/sizeof(WCHAR))<=0 ){ + return; + } + osOutputDebugStringW((LPCWSTR)zDbgBuf); +#else + if( nMin>0 ){ + memset(zDbgBuf, 0, SQLITE_WIN32_DBG_BUF_SIZE); + memcpy(zDbgBuf, zBuf, nMin); + fprintf(stderr, "%s", zDbgBuf); + }else{ + fprintf(stderr, "%s", zBuf); + } +#endif +} + +/* +** The following routine suspends the current thread for at least ms +** milliseconds. This is equivalent to the Win32 Sleep() interface. +*/ +#if SQLITE_OS_WINRT +static HANDLE sleepObj = NULL; +#endif + +SQLITE_API void sqlite3_win32_sleep(DWORD milliseconds){ +#if SQLITE_OS_WINRT + if ( sleepObj==NULL ){ + sleepObj = osCreateEventExW(NULL, NULL, CREATE_EVENT_MANUAL_RESET, + SYNCHRONIZE); + } + assert( sleepObj!=NULL ); + osWaitForSingleObjectEx(sleepObj, milliseconds, FALSE); +#else + osSleep(milliseconds); +#endif +} + +#if SQLITE_MAX_WORKER_THREADS>0 && !SQLITE_OS_WINCE && !SQLITE_OS_WINRT && \ + SQLITE_THREADSAFE>0 +SQLITE_PRIVATE DWORD sqlite3Win32Wait(HANDLE hObject){ + DWORD rc; + while( (rc = osWaitForSingleObjectEx(hObject, INFINITE, + TRUE))==WAIT_IO_COMPLETION ){} + return rc; +} +#endif + +/* +** Return true (non-zero) if we are running under WinNT, Win2K, WinXP, +** or WinCE. Return false (zero) for Win95, Win98, or WinME. +** +** Here is an interesting observation: Win95, Win98, and WinME lack +** the LockFileEx() API. But we can still statically link against that +** API as long as we don't call it when running Win95/98/ME. A call to +** this routine is used to determine if the host is Win95/98/ME or +** WinNT/2K/XP so that we will know whether or not we can safely call +** the LockFileEx() API. +*/ + +#if !SQLITE_WIN32_GETVERSIONEX +# define osIsNT() (1) +#elif SQLITE_OS_WINCE || SQLITE_OS_WINRT || !defined(SQLITE_WIN32_HAS_ANSI) +# define osIsNT() (1) +#elif !defined(SQLITE_WIN32_HAS_WIDE) +# define osIsNT() (0) +#else +# define osIsNT() ((sqlite3_os_type==2) || sqlite3_win32_is_nt()) +#endif + +/* +** This function determines if the machine is running a version of Windows +** based on the NT kernel. +*/ +SQLITE_API int sqlite3_win32_is_nt(void){ +#if SQLITE_OS_WINRT + /* + ** NOTE: The WinRT sub-platform is always assumed to be based on the NT + ** kernel. + */ + return 1; +#elif SQLITE_WIN32_GETVERSIONEX + if( osInterlockedCompareExchange(&sqlite3_os_type, 0, 0)==0 ){ +#if defined(SQLITE_WIN32_HAS_ANSI) + OSVERSIONINFOA sInfo; + sInfo.dwOSVersionInfoSize = sizeof(sInfo); + osGetVersionExA(&sInfo); + osInterlockedCompareExchange(&sqlite3_os_type, + (sInfo.dwPlatformId == VER_PLATFORM_WIN32_NT) ? 2 : 1, 0); +#elif defined(SQLITE_WIN32_HAS_WIDE) + OSVERSIONINFOW sInfo; + sInfo.dwOSVersionInfoSize = sizeof(sInfo); + osGetVersionExW(&sInfo); + osInterlockedCompareExchange(&sqlite3_os_type, + (sInfo.dwPlatformId == VER_PLATFORM_WIN32_NT) ? 2 : 1, 0); +#endif + } + return osInterlockedCompareExchange(&sqlite3_os_type, 2, 2)==2; +#elif SQLITE_TEST + return osInterlockedCompareExchange(&sqlite3_os_type, 2, 2)==2; +#else + /* + ** NOTE: All sub-platforms where the GetVersionEx[AW] functions are + ** deprecated are always assumed to be based on the NT kernel. + */ + return 1; +#endif +} + +#ifdef SQLITE_WIN32_MALLOC +/* +** Allocate nBytes of memory. +*/ +static void *winMemMalloc(int nBytes){ + HANDLE hHeap; + void *p; + + winMemAssertMagic(); + hHeap = winMemGetHeap(); + assert( hHeap!=0 ); + assert( hHeap!=INVALID_HANDLE_VALUE ); +#if !SQLITE_OS_WINRT && defined(SQLITE_WIN32_MALLOC_VALIDATE) + assert( osHeapValidate(hHeap, SQLITE_WIN32_HEAP_FLAGS, NULL) ); +#endif + assert( nBytes>=0 ); + p = osHeapAlloc(hHeap, SQLITE_WIN32_HEAP_FLAGS, (SIZE_T)nBytes); + if( !p ){ + sqlite3_log(SQLITE_NOMEM, "failed to HeapAlloc %u bytes (%lu), heap=%p", + nBytes, osGetLastError(), (void*)hHeap); + } + return p; +} + +/* +** Free memory. +*/ +static void winMemFree(void *pPrior){ + HANDLE hHeap; + + winMemAssertMagic(); + hHeap = winMemGetHeap(); + assert( hHeap!=0 ); + assert( hHeap!=INVALID_HANDLE_VALUE ); +#if !SQLITE_OS_WINRT && defined(SQLITE_WIN32_MALLOC_VALIDATE) + assert( osHeapValidate(hHeap, SQLITE_WIN32_HEAP_FLAGS, pPrior) ); +#endif + if( !pPrior ) return; /* Passing NULL to HeapFree is undefined. */ + if( !osHeapFree(hHeap, SQLITE_WIN32_HEAP_FLAGS, pPrior) ){ + sqlite3_log(SQLITE_NOMEM, "failed to HeapFree block %p (%lu), heap=%p", + pPrior, osGetLastError(), (void*)hHeap); + } +} + +/* +** Change the size of an existing memory allocation +*/ +static void *winMemRealloc(void *pPrior, int nBytes){ + HANDLE hHeap; + void *p; + + winMemAssertMagic(); + hHeap = winMemGetHeap(); + assert( hHeap!=0 ); + assert( hHeap!=INVALID_HANDLE_VALUE ); +#if !SQLITE_OS_WINRT && defined(SQLITE_WIN32_MALLOC_VALIDATE) + assert( osHeapValidate(hHeap, SQLITE_WIN32_HEAP_FLAGS, pPrior) ); +#endif + assert( nBytes>=0 ); + if( !pPrior ){ + p = osHeapAlloc(hHeap, SQLITE_WIN32_HEAP_FLAGS, (SIZE_T)nBytes); + }else{ + p = osHeapReAlloc(hHeap, SQLITE_WIN32_HEAP_FLAGS, pPrior, (SIZE_T)nBytes); + } + if( !p ){ + sqlite3_log(SQLITE_NOMEM, "failed to %s %u bytes (%lu), heap=%p", + pPrior ? "HeapReAlloc" : "HeapAlloc", nBytes, osGetLastError(), + (void*)hHeap); + } + return p; +} + +/* +** Return the size of an outstanding allocation, in bytes. +*/ +static int winMemSize(void *p){ + HANDLE hHeap; + SIZE_T n; + + winMemAssertMagic(); + hHeap = winMemGetHeap(); + assert( hHeap!=0 ); + assert( hHeap!=INVALID_HANDLE_VALUE ); +#if !SQLITE_OS_WINRT && defined(SQLITE_WIN32_MALLOC_VALIDATE) + assert( osHeapValidate(hHeap, SQLITE_WIN32_HEAP_FLAGS, p) ); +#endif + if( !p ) return 0; + n = osHeapSize(hHeap, SQLITE_WIN32_HEAP_FLAGS, p); + if( n==(SIZE_T)-1 ){ + sqlite3_log(SQLITE_NOMEM, "failed to HeapSize block %p (%lu), heap=%p", + p, osGetLastError(), (void*)hHeap); + return 0; + } + return (int)n; +} + +/* +** Round up a request size to the next valid allocation size. +*/ +static int winMemRoundup(int n){ + return n; +} + +/* +** Initialize this module. +*/ +static int winMemInit(void *pAppData){ + winMemData *pWinMemData = (winMemData *)pAppData; + + if( !pWinMemData ) return SQLITE_ERROR; + assert( pWinMemData->magic1==WINMEM_MAGIC1 ); + assert( pWinMemData->magic2==WINMEM_MAGIC2 ); + +#if !SQLITE_OS_WINRT && SQLITE_WIN32_HEAP_CREATE + if( !pWinMemData->hHeap ){ + DWORD dwInitialSize = SQLITE_WIN32_HEAP_INIT_SIZE; + DWORD dwMaximumSize = (DWORD)sqlite3GlobalConfig.nHeap; + if( dwMaximumSize==0 ){ + dwMaximumSize = SQLITE_WIN32_HEAP_MAX_SIZE; + }else if( dwInitialSize>dwMaximumSize ){ + dwInitialSize = dwMaximumSize; + } + pWinMemData->hHeap = osHeapCreate(SQLITE_WIN32_HEAP_FLAGS, + dwInitialSize, dwMaximumSize); + if( !pWinMemData->hHeap ){ + sqlite3_log(SQLITE_NOMEM, + "failed to HeapCreate (%lu), flags=%u, initSize=%lu, maxSize=%lu", + osGetLastError(), SQLITE_WIN32_HEAP_FLAGS, dwInitialSize, + dwMaximumSize); + return SQLITE_NOMEM_BKPT; + } + pWinMemData->bOwned = TRUE; + assert( pWinMemData->bOwned ); + } +#else + pWinMemData->hHeap = osGetProcessHeap(); + if( !pWinMemData->hHeap ){ + sqlite3_log(SQLITE_NOMEM, + "failed to GetProcessHeap (%lu)", osGetLastError()); + return SQLITE_NOMEM_BKPT; + } + pWinMemData->bOwned = FALSE; + assert( !pWinMemData->bOwned ); +#endif + assert( pWinMemData->hHeap!=0 ); + assert( pWinMemData->hHeap!=INVALID_HANDLE_VALUE ); +#if !SQLITE_OS_WINRT && defined(SQLITE_WIN32_MALLOC_VALIDATE) + assert( osHeapValidate(pWinMemData->hHeap, SQLITE_WIN32_HEAP_FLAGS, NULL) ); +#endif + return SQLITE_OK; +} + +/* +** Deinitialize this module. +*/ +static void winMemShutdown(void *pAppData){ + winMemData *pWinMemData = (winMemData *)pAppData; + + if( !pWinMemData ) return; + assert( pWinMemData->magic1==WINMEM_MAGIC1 ); + assert( pWinMemData->magic2==WINMEM_MAGIC2 ); + + if( pWinMemData->hHeap ){ + assert( pWinMemData->hHeap!=INVALID_HANDLE_VALUE ); +#if !SQLITE_OS_WINRT && defined(SQLITE_WIN32_MALLOC_VALIDATE) + assert( osHeapValidate(pWinMemData->hHeap, SQLITE_WIN32_HEAP_FLAGS, NULL) ); +#endif + if( pWinMemData->bOwned ){ + if( !osHeapDestroy(pWinMemData->hHeap) ){ + sqlite3_log(SQLITE_NOMEM, "failed to HeapDestroy (%lu), heap=%p", + osGetLastError(), (void*)pWinMemData->hHeap); + } + pWinMemData->bOwned = FALSE; + } + pWinMemData->hHeap = NULL; + } +} + +/* +** Populate the low-level memory allocation function pointers in +** sqlite3GlobalConfig.m with pointers to the routines in this file. The +** arguments specify the block of memory to manage. +** +** This routine is only called by sqlite3_config(), and therefore +** is not required to be threadsafe (it is not). +*/ +SQLITE_PRIVATE const sqlite3_mem_methods *sqlite3MemGetWin32(void){ + static const sqlite3_mem_methods winMemMethods = { + winMemMalloc, + winMemFree, + winMemRealloc, + winMemSize, + winMemRoundup, + winMemInit, + winMemShutdown, + &win_mem_data + }; + return &winMemMethods; +} + +SQLITE_PRIVATE void sqlite3MemSetDefault(void){ + sqlite3_config(SQLITE_CONFIG_MALLOC, sqlite3MemGetWin32()); +} +#endif /* SQLITE_WIN32_MALLOC */ + +/* +** Convert a UTF-8 string to Microsoft Unicode. +** +** Space to hold the returned string is obtained from sqlite3_malloc(). +*/ +static LPWSTR winUtf8ToUnicode(const char *zText){ + int nChar; + LPWSTR zWideText; + + nChar = osMultiByteToWideChar(CP_UTF8, 0, zText, -1, NULL, 0); + if( nChar==0 ){ + return 0; + } + zWideText = sqlite3MallocZero( nChar*sizeof(WCHAR) ); + if( zWideText==0 ){ + return 0; + } + nChar = osMultiByteToWideChar(CP_UTF8, 0, zText, -1, zWideText, + nChar); + if( nChar==0 ){ + sqlite3_free(zWideText); + zWideText = 0; + } + return zWideText; +} + +/* +** Convert a Microsoft Unicode string to UTF-8. +** +** Space to hold the returned string is obtained from sqlite3_malloc(). +*/ +static char *winUnicodeToUtf8(LPCWSTR zWideText){ + int nByte; + char *zText; + + nByte = osWideCharToMultiByte(CP_UTF8, 0, zWideText, -1, 0, 0, 0, 0); + if( nByte == 0 ){ + return 0; + } + zText = sqlite3MallocZero( nByte ); + if( zText==0 ){ + return 0; + } + nByte = osWideCharToMultiByte(CP_UTF8, 0, zWideText, -1, zText, nByte, + 0, 0); + if( nByte == 0 ){ + sqlite3_free(zText); + zText = 0; + } + return zText; +} + +/* +** Convert an ANSI string to Microsoft Unicode, using the ANSI or OEM +** code page. +** +** Space to hold the returned string is obtained from sqlite3_malloc(). +*/ +static LPWSTR winMbcsToUnicode(const char *zText, int useAnsi){ + int nByte; + LPWSTR zMbcsText; + int codepage = useAnsi ? CP_ACP : CP_OEMCP; + + nByte = osMultiByteToWideChar(codepage, 0, zText, -1, NULL, + 0)*sizeof(WCHAR); + if( nByte==0 ){ + return 0; + } + zMbcsText = sqlite3MallocZero( nByte*sizeof(WCHAR) ); + if( zMbcsText==0 ){ + return 0; + } + nByte = osMultiByteToWideChar(codepage, 0, zText, -1, zMbcsText, + nByte); + if( nByte==0 ){ + sqlite3_free(zMbcsText); + zMbcsText = 0; + } + return zMbcsText; +} + +/* +** Convert a Microsoft Unicode string to a multi-byte character string, +** using the ANSI or OEM code page. +** +** Space to hold the returned string is obtained from sqlite3_malloc(). +*/ +static char *winUnicodeToMbcs(LPCWSTR zWideText, int useAnsi){ + int nByte; + char *zText; + int codepage = useAnsi ? CP_ACP : CP_OEMCP; + + nByte = osWideCharToMultiByte(codepage, 0, zWideText, -1, 0, 0, 0, 0); + if( nByte == 0 ){ + return 0; + } + zText = sqlite3MallocZero( nByte ); + if( zText==0 ){ + return 0; + } + nByte = osWideCharToMultiByte(codepage, 0, zWideText, -1, zText, + nByte, 0, 0); + if( nByte == 0 ){ + sqlite3_free(zText); + zText = 0; + } + return zText; +} + +/* +** Convert a multi-byte character string to UTF-8. +** +** Space to hold the returned string is obtained from sqlite3_malloc(). +*/ +static char *winMbcsToUtf8(const char *zText, int useAnsi){ + char *zTextUtf8; + LPWSTR zTmpWide; + + zTmpWide = winMbcsToUnicode(zText, useAnsi); + if( zTmpWide==0 ){ + return 0; + } + zTextUtf8 = winUnicodeToUtf8(zTmpWide); + sqlite3_free(zTmpWide); + return zTextUtf8; +} + +/* +** Convert a UTF-8 string to a multi-byte character string. +** +** Space to hold the returned string is obtained from sqlite3_malloc(). +*/ +static char *winUtf8ToMbcs(const char *zText, int useAnsi){ + char *zTextMbcs; + LPWSTR zTmpWide; + + zTmpWide = winUtf8ToUnicode(zText); + if( zTmpWide==0 ){ + return 0; + } + zTextMbcs = winUnicodeToMbcs(zTmpWide, useAnsi); + sqlite3_free(zTmpWide); + return zTextMbcs; +} + +/* +** This is a public wrapper for the winUtf8ToUnicode() function. +*/ +SQLITE_API LPWSTR sqlite3_win32_utf8_to_unicode(const char *zText){ +#ifdef SQLITE_ENABLE_API_ARMOR + if( !zText ){ + (void)SQLITE_MISUSE_BKPT; + return 0; + } +#endif +#ifndef SQLITE_OMIT_AUTOINIT + if( sqlite3_initialize() ) return 0; +#endif + return winUtf8ToUnicode(zText); +} + +/* +** This is a public wrapper for the winUnicodeToUtf8() function. +*/ +SQLITE_API char *sqlite3_win32_unicode_to_utf8(LPCWSTR zWideText){ +#ifdef SQLITE_ENABLE_API_ARMOR + if( !zWideText ){ + (void)SQLITE_MISUSE_BKPT; + return 0; + } +#endif +#ifndef SQLITE_OMIT_AUTOINIT + if( sqlite3_initialize() ) return 0; +#endif + return winUnicodeToUtf8(zWideText); +} + +/* +** This is a public wrapper for the winMbcsToUtf8() function. +*/ +SQLITE_API char *sqlite3_win32_mbcs_to_utf8(const char *zText){ +#ifdef SQLITE_ENABLE_API_ARMOR + if( !zText ){ + (void)SQLITE_MISUSE_BKPT; + return 0; + } +#endif +#ifndef SQLITE_OMIT_AUTOINIT + if( sqlite3_initialize() ) return 0; +#endif + return winMbcsToUtf8(zText, osAreFileApisANSI()); +} + +/* +** This is a public wrapper for the winMbcsToUtf8() function. +*/ +SQLITE_API char *sqlite3_win32_mbcs_to_utf8_v2(const char *zText, int useAnsi){ +#ifdef SQLITE_ENABLE_API_ARMOR + if( !zText ){ + (void)SQLITE_MISUSE_BKPT; + return 0; + } +#endif +#ifndef SQLITE_OMIT_AUTOINIT + if( sqlite3_initialize() ) return 0; +#endif + return winMbcsToUtf8(zText, useAnsi); +} + +/* +** This is a public wrapper for the winUtf8ToMbcs() function. +*/ +SQLITE_API char *sqlite3_win32_utf8_to_mbcs(const char *zText){ +#ifdef SQLITE_ENABLE_API_ARMOR + if( !zText ){ + (void)SQLITE_MISUSE_BKPT; + return 0; + } +#endif +#ifndef SQLITE_OMIT_AUTOINIT + if( sqlite3_initialize() ) return 0; +#endif + return winUtf8ToMbcs(zText, osAreFileApisANSI()); +} + +/* +** This is a public wrapper for the winUtf8ToMbcs() function. +*/ +SQLITE_API char *sqlite3_win32_utf8_to_mbcs_v2(const char *zText, int useAnsi){ +#ifdef SQLITE_ENABLE_API_ARMOR + if( !zText ){ + (void)SQLITE_MISUSE_BKPT; + return 0; + } +#endif +#ifndef SQLITE_OMIT_AUTOINIT + if( sqlite3_initialize() ) return 0; +#endif + return winUtf8ToMbcs(zText, useAnsi); +} + +/* +** This function is the same as sqlite3_win32_set_directory (below); however, +** it accepts a UTF-8 string. +*/ +SQLITE_API int sqlite3_win32_set_directory8( + unsigned long type, /* Identifier for directory being set or reset */ + const char *zValue /* New value for directory being set or reset */ +){ + char **ppDirectory = 0; +#ifndef SQLITE_OMIT_AUTOINIT + int rc = sqlite3_initialize(); + if( rc ) return rc; +#endif + if( type==SQLITE_WIN32_DATA_DIRECTORY_TYPE ){ + ppDirectory = &sqlite3_data_directory; + }else if( type==SQLITE_WIN32_TEMP_DIRECTORY_TYPE ){ + ppDirectory = &sqlite3_temp_directory; + } + assert( !ppDirectory || type==SQLITE_WIN32_DATA_DIRECTORY_TYPE + || type==SQLITE_WIN32_TEMP_DIRECTORY_TYPE + ); + assert( !ppDirectory || sqlite3MemdebugHasType(*ppDirectory, MEMTYPE_HEAP) ); + if( ppDirectory ){ + char *zCopy = 0; + if( zValue && zValue[0] ){ + zCopy = sqlite3_mprintf("%s", zValue); + if ( zCopy==0 ){ + return SQLITE_NOMEM_BKPT; + } + } + sqlite3_free(*ppDirectory); + *ppDirectory = zCopy; + return SQLITE_OK; + } + return SQLITE_ERROR; +} + +/* +** This function is the same as sqlite3_win32_set_directory (below); however, +** it accepts a UTF-16 string. +*/ +SQLITE_API int sqlite3_win32_set_directory16( + unsigned long type, /* Identifier for directory being set or reset */ + const void *zValue /* New value for directory being set or reset */ +){ + int rc; + char *zUtf8 = 0; + if( zValue ){ + zUtf8 = sqlite3_win32_unicode_to_utf8(zValue); + if( zUtf8==0 ) return SQLITE_NOMEM_BKPT; + } + rc = sqlite3_win32_set_directory8(type, zUtf8); + if( zUtf8 ) sqlite3_free(zUtf8); + return rc; +} + +/* +** This function sets the data directory or the temporary directory based on +** the provided arguments. The type argument must be 1 in order to set the +** data directory or 2 in order to set the temporary directory. The zValue +** argument is the name of the directory to use. The return value will be +** SQLITE_OK if successful. +*/ +SQLITE_API int sqlite3_win32_set_directory( + unsigned long type, /* Identifier for directory being set or reset */ + void *zValue /* New value for directory being set or reset */ +){ + return sqlite3_win32_set_directory16(type, zValue); +} + +/* +** The return value of winGetLastErrorMsg +** is zero if the error message fits in the buffer, or non-zero +** otherwise (if the message was truncated). +*/ +static int winGetLastErrorMsg(DWORD lastErrno, int nBuf, char *zBuf){ + /* FormatMessage returns 0 on failure. Otherwise it + ** returns the number of TCHARs written to the output + ** buffer, excluding the terminating null char. + */ + DWORD dwLen = 0; + char *zOut = 0; + + if( osIsNT() ){ +#if SQLITE_OS_WINRT + WCHAR zTempWide[SQLITE_WIN32_MAX_ERRMSG_CHARS+1]; + dwLen = osFormatMessageW(FORMAT_MESSAGE_FROM_SYSTEM | + FORMAT_MESSAGE_IGNORE_INSERTS, + NULL, + lastErrno, + 0, + zTempWide, + SQLITE_WIN32_MAX_ERRMSG_CHARS, + 0); +#else + LPWSTR zTempWide = NULL; + dwLen = osFormatMessageW(FORMAT_MESSAGE_ALLOCATE_BUFFER | + FORMAT_MESSAGE_FROM_SYSTEM | + FORMAT_MESSAGE_IGNORE_INSERTS, + NULL, + lastErrno, + 0, + (LPWSTR) &zTempWide, + 0, + 0); +#endif + if( dwLen > 0 ){ + /* allocate a buffer and convert to UTF8 */ + sqlite3BeginBenignMalloc(); + zOut = winUnicodeToUtf8(zTempWide); + sqlite3EndBenignMalloc(); +#if !SQLITE_OS_WINRT + /* free the system buffer allocated by FormatMessage */ + osLocalFree(zTempWide); +#endif + } + } +#ifdef SQLITE_WIN32_HAS_ANSI + else{ + char *zTemp = NULL; + dwLen = osFormatMessageA(FORMAT_MESSAGE_ALLOCATE_BUFFER | + FORMAT_MESSAGE_FROM_SYSTEM | + FORMAT_MESSAGE_IGNORE_INSERTS, + NULL, + lastErrno, + 0, + (LPSTR) &zTemp, + 0, + 0); + if( dwLen > 0 ){ + /* allocate a buffer and convert to UTF8 */ + sqlite3BeginBenignMalloc(); + zOut = winMbcsToUtf8(zTemp, osAreFileApisANSI()); + sqlite3EndBenignMalloc(); + /* free the system buffer allocated by FormatMessage */ + osLocalFree(zTemp); + } + } +#endif + if( 0 == dwLen ){ + sqlite3_snprintf(nBuf, zBuf, "OsError 0x%lx (%lu)", lastErrno, lastErrno); + }else{ + /* copy a maximum of nBuf chars to output buffer */ + sqlite3_snprintf(nBuf, zBuf, "%s", zOut); + /* free the UTF8 buffer */ + sqlite3_free(zOut); + } + return 0; +} + +/* +** +** This function - winLogErrorAtLine() - is only ever called via the macro +** winLogError(). +** +** This routine is invoked after an error occurs in an OS function. +** It logs a message using sqlite3_log() containing the current value of +** error code and, if possible, the human-readable equivalent from +** FormatMessage. +** +** The first argument passed to the macro should be the error code that +** will be returned to SQLite (e.g. SQLITE_IOERR_DELETE, SQLITE_CANTOPEN). +** The two subsequent arguments should be the name of the OS function that +** failed and the associated file-system path, if any. +*/ +#define winLogError(a,b,c,d) winLogErrorAtLine(a,b,c,d,__LINE__) +static int winLogErrorAtLine( + int errcode, /* SQLite error code */ + DWORD lastErrno, /* Win32 last error */ + const char *zFunc, /* Name of OS function that failed */ + const char *zPath, /* File path associated with error */ + int iLine /* Source line number where error occurred */ +){ + char zMsg[500]; /* Human readable error text */ + int i; /* Loop counter */ + + zMsg[0] = 0; + winGetLastErrorMsg(lastErrno, sizeof(zMsg), zMsg); + assert( errcode!=SQLITE_OK ); + if( zPath==0 ) zPath = ""; + for(i=0; zMsg[i] && zMsg[i]!='\r' && zMsg[i]!='\n'; i++){} + zMsg[i] = 0; + sqlite3_log(errcode, + "os_win.c:%d: (%lu) %s(%s) - %s", + iLine, lastErrno, zFunc, zPath, zMsg + ); + + return errcode; +} + +/* +** The number of times that a ReadFile(), WriteFile(), and DeleteFile() +** will be retried following a locking error - probably caused by +** antivirus software. Also the initial delay before the first retry. +** The delay increases linearly with each retry. +*/ +#ifndef SQLITE_WIN32_IOERR_RETRY +# define SQLITE_WIN32_IOERR_RETRY 10 +#endif +#ifndef SQLITE_WIN32_IOERR_RETRY_DELAY +# define SQLITE_WIN32_IOERR_RETRY_DELAY 25 +#endif +static int winIoerrRetry = SQLITE_WIN32_IOERR_RETRY; +static int winIoerrRetryDelay = SQLITE_WIN32_IOERR_RETRY_DELAY; + +/* +** The "winIoerrCanRetry1" macro is used to determine if a particular I/O +** error code obtained via GetLastError() is eligible to be retried. It +** must accept the error code DWORD as its only argument and should return +** non-zero if the error code is transient in nature and the operation +** responsible for generating the original error might succeed upon being +** retried. The argument to this macro should be a variable. +** +** Additionally, a macro named "winIoerrCanRetry2" may be defined. If it +** is defined, it will be consulted only when the macro "winIoerrCanRetry1" +** returns zero. The "winIoerrCanRetry2" macro is completely optional and +** may be used to include additional error codes in the set that should +** result in the failing I/O operation being retried by the caller. If +** defined, the "winIoerrCanRetry2" macro must exhibit external semantics +** identical to those of the "winIoerrCanRetry1" macro. +*/ +#if !defined(winIoerrCanRetry1) +#define winIoerrCanRetry1(a) (((a)==ERROR_ACCESS_DENIED) || \ + ((a)==ERROR_SHARING_VIOLATION) || \ + ((a)==ERROR_LOCK_VIOLATION) || \ + ((a)==ERROR_DEV_NOT_EXIST) || \ + ((a)==ERROR_NETNAME_DELETED) || \ + ((a)==ERROR_SEM_TIMEOUT) || \ + ((a)==ERROR_NETWORK_UNREACHABLE)) +#endif + +/* +** If a ReadFile() or WriteFile() error occurs, invoke this routine +** to see if it should be retried. Return TRUE to retry. Return FALSE +** to give up with an error. +*/ +static int winRetryIoerr(int *pnRetry, DWORD *pError){ + DWORD e = osGetLastError(); + if( *pnRetry>=winIoerrRetry ){ + if( pError ){ + *pError = e; + } + return 0; + } + if( winIoerrCanRetry1(e) ){ + sqlite3_win32_sleep(winIoerrRetryDelay*(1+*pnRetry)); + ++*pnRetry; + return 1; + } +#if defined(winIoerrCanRetry2) + else if( winIoerrCanRetry2(e) ){ + sqlite3_win32_sleep(winIoerrRetryDelay*(1+*pnRetry)); + ++*pnRetry; + return 1; + } +#endif + if( pError ){ + *pError = e; + } + return 0; +} + +/* +** Log a I/O error retry episode. +*/ +static void winLogIoerr(int nRetry, int lineno){ + if( nRetry ){ + sqlite3_log(SQLITE_NOTICE, + "delayed %dms for lock/sharing conflict at line %d", + winIoerrRetryDelay*nRetry*(nRetry+1)/2, lineno + ); + } +} + +/* +** This #if does not rely on the SQLITE_OS_WINCE define because the +** corresponding section in "date.c" cannot use it. +*/ +#if !defined(SQLITE_OMIT_LOCALTIME) && defined(_WIN32_WCE) && \ + (!defined(SQLITE_MSVC_LOCALTIME_API) || !SQLITE_MSVC_LOCALTIME_API) +/* +** The MSVC CRT on Windows CE may not have a localtime() function. +** So define a substitute. +*/ +/* # include */ +struct tm *__cdecl localtime(const time_t *t) +{ + static struct tm y; + FILETIME uTm, lTm; + SYSTEMTIME pTm; + sqlite3_int64 t64; + t64 = *t; + t64 = (t64 + 11644473600)*10000000; + uTm.dwLowDateTime = (DWORD)(t64 & 0xFFFFFFFF); + uTm.dwHighDateTime= (DWORD)(t64 >> 32); + osFileTimeToLocalFileTime(&uTm,&lTm); + osFileTimeToSystemTime(&lTm,&pTm); + y.tm_year = pTm.wYear - 1900; + y.tm_mon = pTm.wMonth - 1; + y.tm_wday = pTm.wDayOfWeek; + y.tm_mday = pTm.wDay; + y.tm_hour = pTm.wHour; + y.tm_min = pTm.wMinute; + y.tm_sec = pTm.wSecond; + return &y; +} +#endif + +#if SQLITE_OS_WINCE +/************************************************************************* +** This section contains code for WinCE only. +*/ +#define HANDLE_TO_WINFILE(a) (winFile*)&((char*)a)[-(int)offsetof(winFile,h)] + +/* +** Acquire a lock on the handle h +*/ +static void winceMutexAcquire(HANDLE h){ + DWORD dwErr; + do { + dwErr = osWaitForSingleObject(h, INFINITE); + } while (dwErr != WAIT_OBJECT_0 && dwErr != WAIT_ABANDONED); +} +/* +** Release a lock acquired by winceMutexAcquire() +*/ +#define winceMutexRelease(h) ReleaseMutex(h) + +/* +** Create the mutex and shared memory used for locking in the file +** descriptor pFile +*/ +static int winceCreateLock(const char *zFilename, winFile *pFile){ + LPWSTR zTok; + LPWSTR zName; + DWORD lastErrno; + BOOL bLogged = FALSE; + BOOL bInit = TRUE; + + zName = winUtf8ToUnicode(zFilename); + if( zName==0 ){ + /* out of memory */ + return SQLITE_IOERR_NOMEM_BKPT; + } + + /* Initialize the local lockdata */ + memset(&pFile->local, 0, sizeof(pFile->local)); + + /* Replace the backslashes from the filename and lowercase it + ** to derive a mutex name. */ + zTok = osCharLowerW(zName); + for (;*zTok;zTok++){ + if (*zTok == '\\') *zTok = '_'; + } + + /* Create/open the named mutex */ + pFile->hMutex = osCreateMutexW(NULL, FALSE, zName); + if (!pFile->hMutex){ + pFile->lastErrno = osGetLastError(); + sqlite3_free(zName); + return winLogError(SQLITE_IOERR, pFile->lastErrno, + "winceCreateLock1", zFilename); + } + + /* Acquire the mutex before continuing */ + winceMutexAcquire(pFile->hMutex); + + /* Since the names of named mutexes, semaphores, file mappings etc are + ** case-sensitive, take advantage of that by uppercasing the mutex name + ** and using that as the shared filemapping name. + */ + osCharUpperW(zName); + pFile->hShared = osCreateFileMappingW(INVALID_HANDLE_VALUE, NULL, + PAGE_READWRITE, 0, sizeof(winceLock), + zName); + + /* Set a flag that indicates we're the first to create the memory so it + ** must be zero-initialized */ + lastErrno = osGetLastError(); + if (lastErrno == ERROR_ALREADY_EXISTS){ + bInit = FALSE; + } + + sqlite3_free(zName); + + /* If we succeeded in making the shared memory handle, map it. */ + if( pFile->hShared ){ + pFile->shared = (winceLock*)osMapViewOfFile(pFile->hShared, + FILE_MAP_READ|FILE_MAP_WRITE, 0, 0, sizeof(winceLock)); + /* If mapping failed, close the shared memory handle and erase it */ + if( !pFile->shared ){ + pFile->lastErrno = osGetLastError(); + winLogError(SQLITE_IOERR, pFile->lastErrno, + "winceCreateLock2", zFilename); + bLogged = TRUE; + osCloseHandle(pFile->hShared); + pFile->hShared = NULL; + } + } + + /* If shared memory could not be created, then close the mutex and fail */ + if( pFile->hShared==NULL ){ + if( !bLogged ){ + pFile->lastErrno = lastErrno; + winLogError(SQLITE_IOERR, pFile->lastErrno, + "winceCreateLock3", zFilename); + bLogged = TRUE; + } + winceMutexRelease(pFile->hMutex); + osCloseHandle(pFile->hMutex); + pFile->hMutex = NULL; + return SQLITE_IOERR; + } + + /* Initialize the shared memory if we're supposed to */ + if( bInit ){ + memset(pFile->shared, 0, sizeof(winceLock)); + } + + winceMutexRelease(pFile->hMutex); + return SQLITE_OK; +} + +/* +** Destroy the part of winFile that deals with wince locks +*/ +static void winceDestroyLock(winFile *pFile){ + if (pFile->hMutex){ + /* Acquire the mutex */ + winceMutexAcquire(pFile->hMutex); + + /* The following blocks should probably assert in debug mode, but they + are to cleanup in case any locks remained open */ + if (pFile->local.nReaders){ + pFile->shared->nReaders --; + } + if (pFile->local.bReserved){ + pFile->shared->bReserved = FALSE; + } + if (pFile->local.bPending){ + pFile->shared->bPending = FALSE; + } + if (pFile->local.bExclusive){ + pFile->shared->bExclusive = FALSE; + } + + /* De-reference and close our copy of the shared memory handle */ + osUnmapViewOfFile(pFile->shared); + osCloseHandle(pFile->hShared); + + /* Done with the mutex */ + winceMutexRelease(pFile->hMutex); + osCloseHandle(pFile->hMutex); + pFile->hMutex = NULL; + } +} + +/* +** An implementation of the LockFile() API of Windows for CE +*/ +static BOOL winceLockFile( + LPHANDLE phFile, + DWORD dwFileOffsetLow, + DWORD dwFileOffsetHigh, + DWORD nNumberOfBytesToLockLow, + DWORD nNumberOfBytesToLockHigh +){ + winFile *pFile = HANDLE_TO_WINFILE(phFile); + BOOL bReturn = FALSE; + + UNUSED_PARAMETER(dwFileOffsetHigh); + UNUSED_PARAMETER(nNumberOfBytesToLockHigh); + + if (!pFile->hMutex) return TRUE; + winceMutexAcquire(pFile->hMutex); + + /* Wanting an exclusive lock? */ + if (dwFileOffsetLow == (DWORD)SHARED_FIRST + && nNumberOfBytesToLockLow == (DWORD)SHARED_SIZE){ + if (pFile->shared->nReaders == 0 && pFile->shared->bExclusive == 0){ + pFile->shared->bExclusive = TRUE; + pFile->local.bExclusive = TRUE; + bReturn = TRUE; + } + } + + /* Want a read-only lock? */ + else if (dwFileOffsetLow == (DWORD)SHARED_FIRST && + nNumberOfBytesToLockLow == 1){ + if (pFile->shared->bExclusive == 0){ + pFile->local.nReaders ++; + if (pFile->local.nReaders == 1){ + pFile->shared->nReaders ++; + } + bReturn = TRUE; + } + } + + /* Want a pending lock? */ + else if (dwFileOffsetLow == (DWORD)PENDING_BYTE + && nNumberOfBytesToLockLow == 1){ + /* If no pending lock has been acquired, then acquire it */ + if (pFile->shared->bPending == 0) { + pFile->shared->bPending = TRUE; + pFile->local.bPending = TRUE; + bReturn = TRUE; + } + } + + /* Want a reserved lock? */ + else if (dwFileOffsetLow == (DWORD)RESERVED_BYTE + && nNumberOfBytesToLockLow == 1){ + if (pFile->shared->bReserved == 0) { + pFile->shared->bReserved = TRUE; + pFile->local.bReserved = TRUE; + bReturn = TRUE; + } + } + + winceMutexRelease(pFile->hMutex); + return bReturn; +} + +/* +** An implementation of the UnlockFile API of Windows for CE +*/ +static BOOL winceUnlockFile( + LPHANDLE phFile, + DWORD dwFileOffsetLow, + DWORD dwFileOffsetHigh, + DWORD nNumberOfBytesToUnlockLow, + DWORD nNumberOfBytesToUnlockHigh +){ + winFile *pFile = HANDLE_TO_WINFILE(phFile); + BOOL bReturn = FALSE; + + UNUSED_PARAMETER(dwFileOffsetHigh); + UNUSED_PARAMETER(nNumberOfBytesToUnlockHigh); + + if (!pFile->hMutex) return TRUE; + winceMutexAcquire(pFile->hMutex); + + /* Releasing a reader lock or an exclusive lock */ + if (dwFileOffsetLow == (DWORD)SHARED_FIRST){ + /* Did we have an exclusive lock? */ + if (pFile->local.bExclusive){ + assert(nNumberOfBytesToUnlockLow == (DWORD)SHARED_SIZE); + pFile->local.bExclusive = FALSE; + pFile->shared->bExclusive = FALSE; + bReturn = TRUE; + } + + /* Did we just have a reader lock? */ + else if (pFile->local.nReaders){ + assert(nNumberOfBytesToUnlockLow == (DWORD)SHARED_SIZE + || nNumberOfBytesToUnlockLow == 1); + pFile->local.nReaders --; + if (pFile->local.nReaders == 0) + { + pFile->shared->nReaders --; + } + bReturn = TRUE; + } + } + + /* Releasing a pending lock */ + else if (dwFileOffsetLow == (DWORD)PENDING_BYTE + && nNumberOfBytesToUnlockLow == 1){ + if (pFile->local.bPending){ + pFile->local.bPending = FALSE; + pFile->shared->bPending = FALSE; + bReturn = TRUE; + } + } + /* Releasing a reserved lock */ + else if (dwFileOffsetLow == (DWORD)RESERVED_BYTE + && nNumberOfBytesToUnlockLow == 1){ + if (pFile->local.bReserved) { + pFile->local.bReserved = FALSE; + pFile->shared->bReserved = FALSE; + bReturn = TRUE; + } + } + + winceMutexRelease(pFile->hMutex); + return bReturn; +} +/* +** End of the special code for wince +*****************************************************************************/ +#endif /* SQLITE_OS_WINCE */ + +/* +** Lock a file region. +*/ +static BOOL winLockFile( + LPHANDLE phFile, + DWORD flags, + DWORD offsetLow, + DWORD offsetHigh, + DWORD numBytesLow, + DWORD numBytesHigh +){ +#if SQLITE_OS_WINCE + /* + ** NOTE: Windows CE is handled differently here due its lack of the Win32 + ** API LockFile. + */ + return winceLockFile(phFile, offsetLow, offsetHigh, + numBytesLow, numBytesHigh); +#else + if( osIsNT() ){ + OVERLAPPED ovlp; + memset(&ovlp, 0, sizeof(OVERLAPPED)); + ovlp.Offset = offsetLow; + ovlp.OffsetHigh = offsetHigh; + return osLockFileEx(*phFile, flags, 0, numBytesLow, numBytesHigh, &ovlp); + }else{ + return osLockFile(*phFile, offsetLow, offsetHigh, numBytesLow, + numBytesHigh); + } +#endif +} + +/* +** Unlock a file region. + */ +static BOOL winUnlockFile( + LPHANDLE phFile, + DWORD offsetLow, + DWORD offsetHigh, + DWORD numBytesLow, + DWORD numBytesHigh +){ +#if SQLITE_OS_WINCE + /* + ** NOTE: Windows CE is handled differently here due its lack of the Win32 + ** API UnlockFile. + */ + return winceUnlockFile(phFile, offsetLow, offsetHigh, + numBytesLow, numBytesHigh); +#else + if( osIsNT() ){ + OVERLAPPED ovlp; + memset(&ovlp, 0, sizeof(OVERLAPPED)); + ovlp.Offset = offsetLow; + ovlp.OffsetHigh = offsetHigh; + return osUnlockFileEx(*phFile, 0, numBytesLow, numBytesHigh, &ovlp); + }else{ + return osUnlockFile(*phFile, offsetLow, offsetHigh, numBytesLow, + numBytesHigh); + } +#endif +} + +/***************************************************************************** +** The next group of routines implement the I/O methods specified +** by the sqlite3_io_methods object. +******************************************************************************/ + +/* +** Some Microsoft compilers lack this definition. +*/ +#ifndef INVALID_SET_FILE_POINTER +# define INVALID_SET_FILE_POINTER ((DWORD)-1) +#endif + +/* +** Move the current position of the file handle passed as the first +** argument to offset iOffset within the file. If successful, return 0. +** Otherwise, set pFile->lastErrno and return non-zero. +*/ +static int winSeekFile(winFile *pFile, sqlite3_int64 iOffset){ +#if !SQLITE_OS_WINRT + LONG upperBits; /* Most sig. 32 bits of new offset */ + LONG lowerBits; /* Least sig. 32 bits of new offset */ + DWORD dwRet; /* Value returned by SetFilePointer() */ + DWORD lastErrno; /* Value returned by GetLastError() */ + + OSTRACE(("SEEK file=%p, offset=%lld\n", pFile->h, iOffset)); + + upperBits = (LONG)((iOffset>>32) & 0x7fffffff); + lowerBits = (LONG)(iOffset & 0xffffffff); + + /* API oddity: If successful, SetFilePointer() returns a dword + ** containing the lower 32-bits of the new file-offset. Or, if it fails, + ** it returns INVALID_SET_FILE_POINTER. However according to MSDN, + ** INVALID_SET_FILE_POINTER may also be a valid new offset. So to determine + ** whether an error has actually occurred, it is also necessary to call + ** GetLastError(). + */ + dwRet = osSetFilePointer(pFile->h, lowerBits, &upperBits, FILE_BEGIN); + + if( (dwRet==INVALID_SET_FILE_POINTER + && ((lastErrno = osGetLastError())!=NO_ERROR)) ){ + pFile->lastErrno = lastErrno; + winLogError(SQLITE_IOERR_SEEK, pFile->lastErrno, + "winSeekFile", pFile->zPath); + OSTRACE(("SEEK file=%p, rc=SQLITE_IOERR_SEEK\n", pFile->h)); + return 1; + } + + OSTRACE(("SEEK file=%p, rc=SQLITE_OK\n", pFile->h)); + return 0; +#else + /* + ** Same as above, except that this implementation works for WinRT. + */ + + LARGE_INTEGER x; /* The new offset */ + BOOL bRet; /* Value returned by SetFilePointerEx() */ + + x.QuadPart = iOffset; + bRet = osSetFilePointerEx(pFile->h, x, 0, FILE_BEGIN); + + if(!bRet){ + pFile->lastErrno = osGetLastError(); + winLogError(SQLITE_IOERR_SEEK, pFile->lastErrno, + "winSeekFile", pFile->zPath); + OSTRACE(("SEEK file=%p, rc=SQLITE_IOERR_SEEK\n", pFile->h)); + return 1; + } + + OSTRACE(("SEEK file=%p, rc=SQLITE_OK\n", pFile->h)); + return 0; +#endif +} + +#if SQLITE_MAX_MMAP_SIZE>0 +/* Forward references to VFS helper methods used for memory mapped files */ +static int winMapfile(winFile*, sqlite3_int64); +static int winUnmapfile(winFile*); +#endif + +/* +** Close a file. +** +** It is reported that an attempt to close a handle might sometimes +** fail. This is a very unreasonable result, but Windows is notorious +** for being unreasonable so I do not doubt that it might happen. If +** the close fails, we pause for 100 milliseconds and try again. As +** many as MX_CLOSE_ATTEMPT attempts to close the handle are made before +** giving up and returning an error. +*/ +#define MX_CLOSE_ATTEMPT 3 +static int winClose(sqlite3_file *id){ + int rc, cnt = 0; + winFile *pFile = (winFile*)id; + + assert( id!=0 ); +#ifndef SQLITE_OMIT_WAL + assert( pFile->pShm==0 ); +#endif + assert( pFile->h!=NULL && pFile->h!=INVALID_HANDLE_VALUE ); + OSTRACE(("CLOSE pid=%lu, pFile=%p, file=%p\n", + osGetCurrentProcessId(), pFile, pFile->h)); + +#if SQLITE_MAX_MMAP_SIZE>0 + winUnmapfile(pFile); +#endif + + do{ + rc = osCloseHandle(pFile->h); + /* SimulateIOError( rc=0; cnt=MX_CLOSE_ATTEMPT; ); */ + }while( rc==0 && ++cnt < MX_CLOSE_ATTEMPT && (sqlite3_win32_sleep(100), 1) ); +#if SQLITE_OS_WINCE +#define WINCE_DELETION_ATTEMPTS 3 + { + winVfsAppData *pAppData = (winVfsAppData*)pFile->pVfs->pAppData; + if( pAppData==NULL || !pAppData->bNoLock ){ + winceDestroyLock(pFile); + } + } + if( pFile->zDeleteOnClose ){ + int cnt = 0; + while( + osDeleteFileW(pFile->zDeleteOnClose)==0 + && osGetFileAttributesW(pFile->zDeleteOnClose)!=0xffffffff + && cnt++ < WINCE_DELETION_ATTEMPTS + ){ + sqlite3_win32_sleep(100); /* Wait a little before trying again */ + } + sqlite3_free(pFile->zDeleteOnClose); + } +#endif + if( rc ){ + pFile->h = NULL; + } + OpenCounter(-1); + OSTRACE(("CLOSE pid=%lu, pFile=%p, file=%p, rc=%s\n", + osGetCurrentProcessId(), pFile, pFile->h, rc ? "ok" : "failed")); + return rc ? SQLITE_OK + : winLogError(SQLITE_IOERR_CLOSE, osGetLastError(), + "winClose", pFile->zPath); +} + +/* +** Read data from a file into a buffer. Return SQLITE_OK if all +** bytes were read successfully and SQLITE_IOERR if anything goes +** wrong. +*/ +static int winRead( + sqlite3_file *id, /* File to read from */ + void *pBuf, /* Write content into this buffer */ + int amt, /* Number of bytes to read */ + sqlite3_int64 offset /* Begin reading at this offset */ +){ +#if !SQLITE_OS_WINCE && !defined(SQLITE_WIN32_NO_OVERLAPPED) + OVERLAPPED overlapped; /* The offset for ReadFile. */ +#endif + winFile *pFile = (winFile*)id; /* file handle */ + DWORD nRead; /* Number of bytes actually read from file */ + int nRetry = 0; /* Number of retrys */ + + assert( id!=0 ); + assert( amt>0 ); + assert( offset>=0 ); + SimulateIOError(return SQLITE_IOERR_READ); + OSTRACE(("READ pid=%lu, pFile=%p, file=%p, buffer=%p, amount=%d, " + "offset=%lld, lock=%d\n", osGetCurrentProcessId(), pFile, + pFile->h, pBuf, amt, offset, pFile->locktype)); + +#if SQLITE_MAX_MMAP_SIZE>0 + /* Deal with as much of this read request as possible by transfering + ** data from the memory mapping using memcpy(). */ + if( offsetmmapSize ){ + if( offset+amt <= pFile->mmapSize ){ + memcpy(pBuf, &((u8 *)(pFile->pMapRegion))[offset], amt); + OSTRACE(("READ-MMAP pid=%lu, pFile=%p, file=%p, rc=SQLITE_OK\n", + osGetCurrentProcessId(), pFile, pFile->h)); + return SQLITE_OK; + }else{ + int nCopy = (int)(pFile->mmapSize - offset); + memcpy(pBuf, &((u8 *)(pFile->pMapRegion))[offset], nCopy); + pBuf = &((u8 *)pBuf)[nCopy]; + amt -= nCopy; + offset += nCopy; + } + } +#endif + +#if SQLITE_OS_WINCE || defined(SQLITE_WIN32_NO_OVERLAPPED) + if( winSeekFile(pFile, offset) ){ + OSTRACE(("READ pid=%lu, pFile=%p, file=%p, rc=SQLITE_FULL\n", + osGetCurrentProcessId(), pFile, pFile->h)); + return SQLITE_FULL; + } + while( !osReadFile(pFile->h, pBuf, amt, &nRead, 0) ){ +#else + memset(&overlapped, 0, sizeof(OVERLAPPED)); + overlapped.Offset = (LONG)(offset & 0xffffffff); + overlapped.OffsetHigh = (LONG)((offset>>32) & 0x7fffffff); + while( !osReadFile(pFile->h, pBuf, amt, &nRead, &overlapped) && + osGetLastError()!=ERROR_HANDLE_EOF ){ +#endif + DWORD lastErrno; + if( winRetryIoerr(&nRetry, &lastErrno) ) continue; + pFile->lastErrno = lastErrno; + OSTRACE(("READ pid=%lu, pFile=%p, file=%p, rc=SQLITE_IOERR_READ\n", + osGetCurrentProcessId(), pFile, pFile->h)); + return winLogError(SQLITE_IOERR_READ, pFile->lastErrno, + "winRead", pFile->zPath); + } + winLogIoerr(nRetry, __LINE__); + if( nRead<(DWORD)amt ){ + /* Unread parts of the buffer must be zero-filled */ + memset(&((char*)pBuf)[nRead], 0, amt-nRead); + OSTRACE(("READ pid=%lu, pFile=%p, file=%p, rc=SQLITE_IOERR_SHORT_READ\n", + osGetCurrentProcessId(), pFile, pFile->h)); + return SQLITE_IOERR_SHORT_READ; + } + + OSTRACE(("READ pid=%lu, pFile=%p, file=%p, rc=SQLITE_OK\n", + osGetCurrentProcessId(), pFile, pFile->h)); + return SQLITE_OK; +} + +/* +** Write data from a buffer into a file. Return SQLITE_OK on success +** or some other error code on failure. +*/ +static int winWrite( + sqlite3_file *id, /* File to write into */ + const void *pBuf, /* The bytes to be written */ + int amt, /* Number of bytes to write */ + sqlite3_int64 offset /* Offset into the file to begin writing at */ +){ + int rc = 0; /* True if error has occurred, else false */ + winFile *pFile = (winFile*)id; /* File handle */ + int nRetry = 0; /* Number of retries */ + + assert( amt>0 ); + assert( pFile ); + SimulateIOError(return SQLITE_IOERR_WRITE); + SimulateDiskfullError(return SQLITE_FULL); + + OSTRACE(("WRITE pid=%lu, pFile=%p, file=%p, buffer=%p, amount=%d, " + "offset=%lld, lock=%d\n", osGetCurrentProcessId(), pFile, + pFile->h, pBuf, amt, offset, pFile->locktype)); + +#if defined(SQLITE_MMAP_READWRITE) && SQLITE_MAX_MMAP_SIZE>0 + /* Deal with as much of this write request as possible by transfering + ** data from the memory mapping using memcpy(). */ + if( offsetmmapSize ){ + if( offset+amt <= pFile->mmapSize ){ + memcpy(&((u8 *)(pFile->pMapRegion))[offset], pBuf, amt); + OSTRACE(("WRITE-MMAP pid=%lu, pFile=%p, file=%p, rc=SQLITE_OK\n", + osGetCurrentProcessId(), pFile, pFile->h)); + return SQLITE_OK; + }else{ + int nCopy = (int)(pFile->mmapSize - offset); + memcpy(&((u8 *)(pFile->pMapRegion))[offset], pBuf, nCopy); + pBuf = &((u8 *)pBuf)[nCopy]; + amt -= nCopy; + offset += nCopy; + } + } +#endif + +#if SQLITE_OS_WINCE || defined(SQLITE_WIN32_NO_OVERLAPPED) + rc = winSeekFile(pFile, offset); + if( rc==0 ){ +#else + { +#endif +#if !SQLITE_OS_WINCE && !defined(SQLITE_WIN32_NO_OVERLAPPED) + OVERLAPPED overlapped; /* The offset for WriteFile. */ +#endif + u8 *aRem = (u8 *)pBuf; /* Data yet to be written */ + int nRem = amt; /* Number of bytes yet to be written */ + DWORD nWrite; /* Bytes written by each WriteFile() call */ + DWORD lastErrno = NO_ERROR; /* Value returned by GetLastError() */ + +#if !SQLITE_OS_WINCE && !defined(SQLITE_WIN32_NO_OVERLAPPED) + memset(&overlapped, 0, sizeof(OVERLAPPED)); + overlapped.Offset = (LONG)(offset & 0xffffffff); + overlapped.OffsetHigh = (LONG)((offset>>32) & 0x7fffffff); +#endif + + while( nRem>0 ){ +#if SQLITE_OS_WINCE || defined(SQLITE_WIN32_NO_OVERLAPPED) + if( !osWriteFile(pFile->h, aRem, nRem, &nWrite, 0) ){ +#else + if( !osWriteFile(pFile->h, aRem, nRem, &nWrite, &overlapped) ){ +#endif + if( winRetryIoerr(&nRetry, &lastErrno) ) continue; + break; + } + assert( nWrite==0 || nWrite<=(DWORD)nRem ); + if( nWrite==0 || nWrite>(DWORD)nRem ){ + lastErrno = osGetLastError(); + break; + } +#if !SQLITE_OS_WINCE && !defined(SQLITE_WIN32_NO_OVERLAPPED) + offset += nWrite; + overlapped.Offset = (LONG)(offset & 0xffffffff); + overlapped.OffsetHigh = (LONG)((offset>>32) & 0x7fffffff); +#endif + aRem += nWrite; + nRem -= nWrite; + } + if( nRem>0 ){ + pFile->lastErrno = lastErrno; + rc = 1; + } + } + + if( rc ){ + if( ( pFile->lastErrno==ERROR_HANDLE_DISK_FULL ) + || ( pFile->lastErrno==ERROR_DISK_FULL )){ + OSTRACE(("WRITE pid=%lu, pFile=%p, file=%p, rc=SQLITE_FULL\n", + osGetCurrentProcessId(), pFile, pFile->h)); + return winLogError(SQLITE_FULL, pFile->lastErrno, + "winWrite1", pFile->zPath); + } + OSTRACE(("WRITE pid=%lu, pFile=%p, file=%p, rc=SQLITE_IOERR_WRITE\n", + osGetCurrentProcessId(), pFile, pFile->h)); + return winLogError(SQLITE_IOERR_WRITE, pFile->lastErrno, + "winWrite2", pFile->zPath); + }else{ + winLogIoerr(nRetry, __LINE__); + } + OSTRACE(("WRITE pid=%lu, pFile=%p, file=%p, rc=SQLITE_OK\n", + osGetCurrentProcessId(), pFile, pFile->h)); + return SQLITE_OK; +} + +/* +** Truncate an open file to a specified size +*/ +static int winTruncate(sqlite3_file *id, sqlite3_int64 nByte){ + winFile *pFile = (winFile*)id; /* File handle object */ + int rc = SQLITE_OK; /* Return code for this function */ + DWORD lastErrno; +#if SQLITE_MAX_MMAP_SIZE>0 + sqlite3_int64 oldMmapSize; + if( pFile->nFetchOut>0 ){ + /* File truncation is a no-op if there are outstanding memory mapped + ** pages. This is because truncating the file means temporarily unmapping + ** the file, and that might delete memory out from under existing cursors. + ** + ** This can result in incremental vacuum not truncating the file, + ** if there is an active read cursor when the incremental vacuum occurs. + ** No real harm comes of this - the database file is not corrupted, + ** though some folks might complain that the file is bigger than it + ** needs to be. + ** + ** The only feasible work-around is to defer the truncation until after + ** all references to memory-mapped content are closed. That is doable, + ** but involves adding a few branches in the common write code path which + ** could slow down normal operations slightly. Hence, we have decided for + ** now to simply make trancations a no-op if there are pending reads. We + ** can maybe revisit this decision in the future. + */ + return SQLITE_OK; + } +#endif + + assert( pFile ); + SimulateIOError(return SQLITE_IOERR_TRUNCATE); + OSTRACE(("TRUNCATE pid=%lu, pFile=%p, file=%p, size=%lld, lock=%d\n", + osGetCurrentProcessId(), pFile, pFile->h, nByte, pFile->locktype)); + + /* If the user has configured a chunk-size for this file, truncate the + ** file so that it consists of an integer number of chunks (i.e. the + ** actual file size after the operation may be larger than the requested + ** size). + */ + if( pFile->szChunk>0 ){ + nByte = ((nByte + pFile->szChunk - 1)/pFile->szChunk) * pFile->szChunk; + } + +#if SQLITE_MAX_MMAP_SIZE>0 + if( pFile->pMapRegion ){ + oldMmapSize = pFile->mmapSize; + }else{ + oldMmapSize = 0; + } + winUnmapfile(pFile); +#endif + + /* SetEndOfFile() returns non-zero when successful, or zero when it fails. */ + if( winSeekFile(pFile, nByte) ){ + rc = winLogError(SQLITE_IOERR_TRUNCATE, pFile->lastErrno, + "winTruncate1", pFile->zPath); + }else if( 0==osSetEndOfFile(pFile->h) && + ((lastErrno = osGetLastError())!=ERROR_USER_MAPPED_FILE) ){ + pFile->lastErrno = lastErrno; + rc = winLogError(SQLITE_IOERR_TRUNCATE, pFile->lastErrno, + "winTruncate2", pFile->zPath); + } + +#if SQLITE_MAX_MMAP_SIZE>0 + if( rc==SQLITE_OK && oldMmapSize>0 ){ + if( oldMmapSize>nByte ){ + winMapfile(pFile, -1); + }else{ + winMapfile(pFile, oldMmapSize); + } + } +#endif + + OSTRACE(("TRUNCATE pid=%lu, pFile=%p, file=%p, rc=%s\n", + osGetCurrentProcessId(), pFile, pFile->h, sqlite3ErrName(rc))); + return rc; +} + +#ifdef SQLITE_TEST +/* +** Count the number of fullsyncs and normal syncs. This is used to test +** that syncs and fullsyncs are occuring at the right times. +*/ +SQLITE_API int sqlite3_sync_count = 0; +SQLITE_API int sqlite3_fullsync_count = 0; +#endif + +/* +** Make sure all writes to a particular file are committed to disk. +*/ +static int winSync(sqlite3_file *id, int flags){ +#ifndef SQLITE_NO_SYNC + /* + ** Used only when SQLITE_NO_SYNC is not defined. + */ + BOOL rc; +#endif +#if !defined(NDEBUG) || !defined(SQLITE_NO_SYNC) || \ + defined(SQLITE_HAVE_OS_TRACE) + /* + ** Used when SQLITE_NO_SYNC is not defined and by the assert() and/or + ** OSTRACE() macros. + */ + winFile *pFile = (winFile*)id; +#else + UNUSED_PARAMETER(id); +#endif + + assert( pFile ); + /* Check that one of SQLITE_SYNC_NORMAL or FULL was passed */ + assert((flags&0x0F)==SQLITE_SYNC_NORMAL + || (flags&0x0F)==SQLITE_SYNC_FULL + ); + + /* Unix cannot, but some systems may return SQLITE_FULL from here. This + ** line is to test that doing so does not cause any problems. + */ + SimulateDiskfullError( return SQLITE_FULL ); + + OSTRACE(("SYNC pid=%lu, pFile=%p, file=%p, flags=%x, lock=%d\n", + osGetCurrentProcessId(), pFile, pFile->h, flags, + pFile->locktype)); + +#ifndef SQLITE_TEST + UNUSED_PARAMETER(flags); +#else + if( (flags&0x0F)==SQLITE_SYNC_FULL ){ + sqlite3_fullsync_count++; + } + sqlite3_sync_count++; +#endif + + /* If we compiled with the SQLITE_NO_SYNC flag, then syncing is a + ** no-op + */ +#ifdef SQLITE_NO_SYNC + OSTRACE(("SYNC-NOP pid=%lu, pFile=%p, file=%p, rc=SQLITE_OK\n", + osGetCurrentProcessId(), pFile, pFile->h)); + return SQLITE_OK; +#else +#if SQLITE_MAX_MMAP_SIZE>0 + if( pFile->pMapRegion ){ + if( osFlushViewOfFile(pFile->pMapRegion, 0) ){ + OSTRACE(("SYNC-MMAP pid=%lu, pFile=%p, pMapRegion=%p, " + "rc=SQLITE_OK\n", osGetCurrentProcessId(), + pFile, pFile->pMapRegion)); + }else{ + pFile->lastErrno = osGetLastError(); + OSTRACE(("SYNC-MMAP pid=%lu, pFile=%p, pMapRegion=%p, " + "rc=SQLITE_IOERR_MMAP\n", osGetCurrentProcessId(), + pFile, pFile->pMapRegion)); + return winLogError(SQLITE_IOERR_MMAP, pFile->lastErrno, + "winSync1", pFile->zPath); + } + } +#endif + rc = osFlushFileBuffers(pFile->h); + SimulateIOError( rc=FALSE ); + if( rc ){ + OSTRACE(("SYNC pid=%lu, pFile=%p, file=%p, rc=SQLITE_OK\n", + osGetCurrentProcessId(), pFile, pFile->h)); + return SQLITE_OK; + }else{ + pFile->lastErrno = osGetLastError(); + OSTRACE(("SYNC pid=%lu, pFile=%p, file=%p, rc=SQLITE_IOERR_FSYNC\n", + osGetCurrentProcessId(), pFile, pFile->h)); + return winLogError(SQLITE_IOERR_FSYNC, pFile->lastErrno, + "winSync2", pFile->zPath); + } +#endif +} + +/* +** Determine the current size of a file in bytes +*/ +static int winFileSize(sqlite3_file *id, sqlite3_int64 *pSize){ + winFile *pFile = (winFile*)id; + int rc = SQLITE_OK; + + assert( id!=0 ); + assert( pSize!=0 ); + SimulateIOError(return SQLITE_IOERR_FSTAT); + OSTRACE(("SIZE file=%p, pSize=%p\n", pFile->h, pSize)); + +#if SQLITE_OS_WINRT + { + FILE_STANDARD_INFO info; + if( osGetFileInformationByHandleEx(pFile->h, FileStandardInfo, + &info, sizeof(info)) ){ + *pSize = info.EndOfFile.QuadPart; + }else{ + pFile->lastErrno = osGetLastError(); + rc = winLogError(SQLITE_IOERR_FSTAT, pFile->lastErrno, + "winFileSize", pFile->zPath); + } + } +#else + { + DWORD upperBits; + DWORD lowerBits; + DWORD lastErrno; + + lowerBits = osGetFileSize(pFile->h, &upperBits); + *pSize = (((sqlite3_int64)upperBits)<<32) + lowerBits; + if( (lowerBits == INVALID_FILE_SIZE) + && ((lastErrno = osGetLastError())!=NO_ERROR) ){ + pFile->lastErrno = lastErrno; + rc = winLogError(SQLITE_IOERR_FSTAT, pFile->lastErrno, + "winFileSize", pFile->zPath); + } + } +#endif + OSTRACE(("SIZE file=%p, pSize=%p, *pSize=%lld, rc=%s\n", + pFile->h, pSize, *pSize, sqlite3ErrName(rc))); + return rc; +} + +/* +** LOCKFILE_FAIL_IMMEDIATELY is undefined on some Windows systems. +*/ +#ifndef LOCKFILE_FAIL_IMMEDIATELY +# define LOCKFILE_FAIL_IMMEDIATELY 1 +#endif + +#ifndef LOCKFILE_EXCLUSIVE_LOCK +# define LOCKFILE_EXCLUSIVE_LOCK 2 +#endif + +/* +** Historically, SQLite has used both the LockFile and LockFileEx functions. +** When the LockFile function was used, it was always expected to fail +** immediately if the lock could not be obtained. Also, it always expected to +** obtain an exclusive lock. These flags are used with the LockFileEx function +** and reflect those expectations; therefore, they should not be changed. +*/ +#ifndef SQLITE_LOCKFILE_FLAGS +# define SQLITE_LOCKFILE_FLAGS (LOCKFILE_FAIL_IMMEDIATELY | \ + LOCKFILE_EXCLUSIVE_LOCK) +#endif + +/* +** Currently, SQLite never calls the LockFileEx function without wanting the +** call to fail immediately if the lock cannot be obtained. +*/ +#ifndef SQLITE_LOCKFILEEX_FLAGS +# define SQLITE_LOCKFILEEX_FLAGS (LOCKFILE_FAIL_IMMEDIATELY) +#endif + +/* +** Acquire a reader lock. +** Different API routines are called depending on whether or not this +** is Win9x or WinNT. +*/ +static int winGetReadLock(winFile *pFile){ + int res; + OSTRACE(("READ-LOCK file=%p, lock=%d\n", pFile->h, pFile->locktype)); + if( osIsNT() ){ +#if SQLITE_OS_WINCE + /* + ** NOTE: Windows CE is handled differently here due its lack of the Win32 + ** API LockFileEx. + */ + res = winceLockFile(&pFile->h, SHARED_FIRST, 0, 1, 0); +#else + res = winLockFile(&pFile->h, SQLITE_LOCKFILEEX_FLAGS, SHARED_FIRST, 0, + SHARED_SIZE, 0); +#endif + } +#ifdef SQLITE_WIN32_HAS_ANSI + else{ + int lk; + sqlite3_randomness(sizeof(lk), &lk); + pFile->sharedLockByte = (short)((lk & 0x7fffffff)%(SHARED_SIZE - 1)); + res = winLockFile(&pFile->h, SQLITE_LOCKFILE_FLAGS, + SHARED_FIRST+pFile->sharedLockByte, 0, 1, 0); + } +#endif + if( res == 0 ){ + pFile->lastErrno = osGetLastError(); + /* No need to log a failure to lock */ + } + OSTRACE(("READ-LOCK file=%p, result=%d\n", pFile->h, res)); + return res; +} + +/* +** Undo a readlock +*/ +static int winUnlockReadLock(winFile *pFile){ + int res; + DWORD lastErrno; + OSTRACE(("READ-UNLOCK file=%p, lock=%d\n", pFile->h, pFile->locktype)); + if( osIsNT() ){ + res = winUnlockFile(&pFile->h, SHARED_FIRST, 0, SHARED_SIZE, 0); + } +#ifdef SQLITE_WIN32_HAS_ANSI + else{ + res = winUnlockFile(&pFile->h, SHARED_FIRST+pFile->sharedLockByte, 0, 1, 0); + } +#endif + if( res==0 && ((lastErrno = osGetLastError())!=ERROR_NOT_LOCKED) ){ + pFile->lastErrno = lastErrno; + winLogError(SQLITE_IOERR_UNLOCK, pFile->lastErrno, + "winUnlockReadLock", pFile->zPath); + } + OSTRACE(("READ-UNLOCK file=%p, result=%d\n", pFile->h, res)); + return res; +} + +/* +** Lock the file with the lock specified by parameter locktype - one +** of the following: +** +** (1) SHARED_LOCK +** (2) RESERVED_LOCK +** (3) PENDING_LOCK +** (4) EXCLUSIVE_LOCK +** +** Sometimes when requesting one lock state, additional lock states +** are inserted in between. The locking might fail on one of the later +** transitions leaving the lock state different from what it started but +** still short of its goal. The following chart shows the allowed +** transitions and the inserted intermediate states: +** +** UNLOCKED -> SHARED +** SHARED -> RESERVED +** SHARED -> (PENDING) -> EXCLUSIVE +** RESERVED -> (PENDING) -> EXCLUSIVE +** PENDING -> EXCLUSIVE +** +** This routine will only increase a lock. The winUnlock() routine +** erases all locks at once and returns us immediately to locking level 0. +** It is not possible to lower the locking level one step at a time. You +** must go straight to locking level 0. +*/ +static int winLock(sqlite3_file *id, int locktype){ + int rc = SQLITE_OK; /* Return code from subroutines */ + int res = 1; /* Result of a Windows lock call */ + int newLocktype; /* Set pFile->locktype to this value before exiting */ + int gotPendingLock = 0;/* True if we acquired a PENDING lock this time */ + winFile *pFile = (winFile*)id; + DWORD lastErrno = NO_ERROR; + + assert( id!=0 ); + OSTRACE(("LOCK file=%p, oldLock=%d(%d), newLock=%d\n", + pFile->h, pFile->locktype, pFile->sharedLockByte, locktype)); + + /* If there is already a lock of this type or more restrictive on the + ** OsFile, do nothing. Don't use the end_lock: exit path, as + ** sqlite3OsEnterMutex() hasn't been called yet. + */ + if( pFile->locktype>=locktype ){ + OSTRACE(("LOCK-HELD file=%p, rc=SQLITE_OK\n", pFile->h)); + return SQLITE_OK; + } + + /* Do not allow any kind of write-lock on a read-only database + */ + if( (pFile->ctrlFlags & WINFILE_RDONLY)!=0 && locktype>=RESERVED_LOCK ){ + return SQLITE_IOERR_LOCK; + } + + /* Make sure the locking sequence is correct + */ + assert( pFile->locktype!=NO_LOCK || locktype==SHARED_LOCK ); + assert( locktype!=PENDING_LOCK ); + assert( locktype!=RESERVED_LOCK || pFile->locktype==SHARED_LOCK ); + + /* Lock the PENDING_LOCK byte if we need to acquire a PENDING lock or + ** a SHARED lock. If we are acquiring a SHARED lock, the acquisition of + ** the PENDING_LOCK byte is temporary. + */ + newLocktype = pFile->locktype; + if( pFile->locktype==NO_LOCK + || (locktype==EXCLUSIVE_LOCK && pFile->locktype<=RESERVED_LOCK) + ){ + int cnt = 3; + while( cnt-->0 && (res = winLockFile(&pFile->h, SQLITE_LOCKFILE_FLAGS, + PENDING_BYTE, 0, 1, 0))==0 ){ + /* Try 3 times to get the pending lock. This is needed to work + ** around problems caused by indexing and/or anti-virus software on + ** Windows systems. + ** If you are using this code as a model for alternative VFSes, do not + ** copy this retry logic. It is a hack intended for Windows only. + */ + lastErrno = osGetLastError(); + OSTRACE(("LOCK-PENDING-FAIL file=%p, count=%d, result=%d\n", + pFile->h, cnt, res)); + if( lastErrno==ERROR_INVALID_HANDLE ){ + pFile->lastErrno = lastErrno; + rc = SQLITE_IOERR_LOCK; + OSTRACE(("LOCK-FAIL file=%p, count=%d, rc=%s\n", + pFile->h, cnt, sqlite3ErrName(rc))); + return rc; + } + if( cnt ) sqlite3_win32_sleep(1); + } + gotPendingLock = res; + if( !res ){ + lastErrno = osGetLastError(); + } + } + + /* Acquire a shared lock + */ + if( locktype==SHARED_LOCK && res ){ + assert( pFile->locktype==NO_LOCK ); + res = winGetReadLock(pFile); + if( res ){ + newLocktype = SHARED_LOCK; + }else{ + lastErrno = osGetLastError(); + } + } + + /* Acquire a RESERVED lock + */ + if( locktype==RESERVED_LOCK && res ){ + assert( pFile->locktype==SHARED_LOCK ); + res = winLockFile(&pFile->h, SQLITE_LOCKFILE_FLAGS, RESERVED_BYTE, 0, 1, 0); + if( res ){ + newLocktype = RESERVED_LOCK; + }else{ + lastErrno = osGetLastError(); + } + } + + /* Acquire a PENDING lock + */ + if( locktype==EXCLUSIVE_LOCK && res ){ + newLocktype = PENDING_LOCK; + gotPendingLock = 0; + } + + /* Acquire an EXCLUSIVE lock + */ + if( locktype==EXCLUSIVE_LOCK && res ){ + assert( pFile->locktype>=SHARED_LOCK ); + res = winUnlockReadLock(pFile); + res = winLockFile(&pFile->h, SQLITE_LOCKFILE_FLAGS, SHARED_FIRST, 0, + SHARED_SIZE, 0); + if( res ){ + newLocktype = EXCLUSIVE_LOCK; + }else{ + lastErrno = osGetLastError(); + winGetReadLock(pFile); + } + } + + /* If we are holding a PENDING lock that ought to be released, then + ** release it now. + */ + if( gotPendingLock && locktype==SHARED_LOCK ){ + winUnlockFile(&pFile->h, PENDING_BYTE, 0, 1, 0); + } + + /* Update the state of the lock has held in the file descriptor then + ** return the appropriate result code. + */ + if( res ){ + rc = SQLITE_OK; + }else{ + pFile->lastErrno = lastErrno; + rc = SQLITE_BUSY; + OSTRACE(("LOCK-FAIL file=%p, wanted=%d, got=%d\n", + pFile->h, locktype, newLocktype)); + } + pFile->locktype = (u8)newLocktype; + OSTRACE(("LOCK file=%p, lock=%d, rc=%s\n", + pFile->h, pFile->locktype, sqlite3ErrName(rc))); + return rc; +} + +/* +** This routine checks if there is a RESERVED lock held on the specified +** file by this or any other process. If such a lock is held, return +** non-zero, otherwise zero. +*/ +static int winCheckReservedLock(sqlite3_file *id, int *pResOut){ + int res; + winFile *pFile = (winFile*)id; + + SimulateIOError( return SQLITE_IOERR_CHECKRESERVEDLOCK; ); + OSTRACE(("TEST-WR-LOCK file=%p, pResOut=%p\n", pFile->h, pResOut)); + + assert( id!=0 ); + if( pFile->locktype>=RESERVED_LOCK ){ + res = 1; + OSTRACE(("TEST-WR-LOCK file=%p, result=%d (local)\n", pFile->h, res)); + }else{ + res = winLockFile(&pFile->h, SQLITE_LOCKFILEEX_FLAGS,RESERVED_BYTE,0,1,0); + if( res ){ + winUnlockFile(&pFile->h, RESERVED_BYTE, 0, 1, 0); + } + res = !res; + OSTRACE(("TEST-WR-LOCK file=%p, result=%d (remote)\n", pFile->h, res)); + } + *pResOut = res; + OSTRACE(("TEST-WR-LOCK file=%p, pResOut=%p, *pResOut=%d, rc=SQLITE_OK\n", + pFile->h, pResOut, *pResOut)); + return SQLITE_OK; +} + +/* +** Lower the locking level on file descriptor id to locktype. locktype +** must be either NO_LOCK or SHARED_LOCK. +** +** If the locking level of the file descriptor is already at or below +** the requested locking level, this routine is a no-op. +** +** It is not possible for this routine to fail if the second argument +** is NO_LOCK. If the second argument is SHARED_LOCK then this routine +** might return SQLITE_IOERR; +*/ +static int winUnlock(sqlite3_file *id, int locktype){ + int type; + winFile *pFile = (winFile*)id; + int rc = SQLITE_OK; + assert( pFile!=0 ); + assert( locktype<=SHARED_LOCK ); + OSTRACE(("UNLOCK file=%p, oldLock=%d(%d), newLock=%d\n", + pFile->h, pFile->locktype, pFile->sharedLockByte, locktype)); + type = pFile->locktype; + if( type>=EXCLUSIVE_LOCK ){ + winUnlockFile(&pFile->h, SHARED_FIRST, 0, SHARED_SIZE, 0); + if( locktype==SHARED_LOCK && !winGetReadLock(pFile) ){ + /* This should never happen. We should always be able to + ** reacquire the read lock */ + rc = winLogError(SQLITE_IOERR_UNLOCK, osGetLastError(), + "winUnlock", pFile->zPath); + } + } + if( type>=RESERVED_LOCK ){ + winUnlockFile(&pFile->h, RESERVED_BYTE, 0, 1, 0); + } + if( locktype==NO_LOCK && type>=SHARED_LOCK ){ + winUnlockReadLock(pFile); + } + if( type>=PENDING_LOCK ){ + winUnlockFile(&pFile->h, PENDING_BYTE, 0, 1, 0); + } + pFile->locktype = (u8)locktype; + OSTRACE(("UNLOCK file=%p, lock=%d, rc=%s\n", + pFile->h, pFile->locktype, sqlite3ErrName(rc))); + return rc; +} + +/****************************************************************************** +****************************** No-op Locking ********************************** +** +** Of the various locking implementations available, this is by far the +** simplest: locking is ignored. No attempt is made to lock the database +** file for reading or writing. +** +** This locking mode is appropriate for use on read-only databases +** (ex: databases that are burned into CD-ROM, for example.) It can +** also be used if the application employs some external mechanism to +** prevent simultaneous access of the same database by two or more +** database connections. But there is a serious risk of database +** corruption if this locking mode is used in situations where multiple +** database connections are accessing the same database file at the same +** time and one or more of those connections are writing. +*/ + +static int winNolockLock(sqlite3_file *id, int locktype){ + UNUSED_PARAMETER(id); + UNUSED_PARAMETER(locktype); + return SQLITE_OK; +} + +static int winNolockCheckReservedLock(sqlite3_file *id, int *pResOut){ + UNUSED_PARAMETER(id); + UNUSED_PARAMETER(pResOut); + return SQLITE_OK; +} + +static int winNolockUnlock(sqlite3_file *id, int locktype){ + UNUSED_PARAMETER(id); + UNUSED_PARAMETER(locktype); + return SQLITE_OK; +} + +/******************* End of the no-op lock implementation ********************* +******************************************************************************/ + +/* +** If *pArg is initially negative then this is a query. Set *pArg to +** 1 or 0 depending on whether or not bit mask of pFile->ctrlFlags is set. +** +** If *pArg is 0 or 1, then clear or set the mask bit of pFile->ctrlFlags. +*/ +static void winModeBit(winFile *pFile, unsigned char mask, int *pArg){ + if( *pArg<0 ){ + *pArg = (pFile->ctrlFlags & mask)!=0; + }else if( (*pArg)==0 ){ + pFile->ctrlFlags &= ~mask; + }else{ + pFile->ctrlFlags |= mask; + } +} + +/* Forward references to VFS helper methods used for temporary files */ +static int winGetTempname(sqlite3_vfs *, char **); +static int winIsDir(const void *); +static BOOL winIsDriveLetterAndColon(const char *); + +/* +** Control and query of the open file handle. +*/ +static int winFileControl(sqlite3_file *id, int op, void *pArg){ + winFile *pFile = (winFile*)id; + OSTRACE(("FCNTL file=%p, op=%d, pArg=%p\n", pFile->h, op, pArg)); + switch( op ){ + case SQLITE_FCNTL_LOCKSTATE: { + *(int*)pArg = pFile->locktype; + OSTRACE(("FCNTL file=%p, rc=SQLITE_OK\n", pFile->h)); + return SQLITE_OK; + } + case SQLITE_FCNTL_LAST_ERRNO: { + *(int*)pArg = (int)pFile->lastErrno; + OSTRACE(("FCNTL file=%p, rc=SQLITE_OK\n", pFile->h)); + return SQLITE_OK; + } + case SQLITE_FCNTL_CHUNK_SIZE: { + pFile->szChunk = *(int *)pArg; + OSTRACE(("FCNTL file=%p, rc=SQLITE_OK\n", pFile->h)); + return SQLITE_OK; + } + case SQLITE_FCNTL_SIZE_HINT: { + if( pFile->szChunk>0 ){ + sqlite3_int64 oldSz; + int rc = winFileSize(id, &oldSz); + if( rc==SQLITE_OK ){ + sqlite3_int64 newSz = *(sqlite3_int64*)pArg; + if( newSz>oldSz ){ + SimulateIOErrorBenign(1); + rc = winTruncate(id, newSz); + SimulateIOErrorBenign(0); + } + } + OSTRACE(("FCNTL file=%p, rc=%s\n", pFile->h, sqlite3ErrName(rc))); + return rc; + } + OSTRACE(("FCNTL file=%p, rc=SQLITE_OK\n", pFile->h)); + return SQLITE_OK; + } + case SQLITE_FCNTL_PERSIST_WAL: { + winModeBit(pFile, WINFILE_PERSIST_WAL, (int*)pArg); + OSTRACE(("FCNTL file=%p, rc=SQLITE_OK\n", pFile->h)); + return SQLITE_OK; + } + case SQLITE_FCNTL_POWERSAFE_OVERWRITE: { + winModeBit(pFile, WINFILE_PSOW, (int*)pArg); + OSTRACE(("FCNTL file=%p, rc=SQLITE_OK\n", pFile->h)); + return SQLITE_OK; + } + case SQLITE_FCNTL_VFSNAME: { + *(char**)pArg = sqlite3_mprintf("%s", pFile->pVfs->zName); + OSTRACE(("FCNTL file=%p, rc=SQLITE_OK\n", pFile->h)); + return SQLITE_OK; + } + case SQLITE_FCNTL_WIN32_AV_RETRY: { + int *a = (int*)pArg; + if( a[0]>0 ){ + winIoerrRetry = a[0]; + }else{ + a[0] = winIoerrRetry; + } + if( a[1]>0 ){ + winIoerrRetryDelay = a[1]; + }else{ + a[1] = winIoerrRetryDelay; + } + OSTRACE(("FCNTL file=%p, rc=SQLITE_OK\n", pFile->h)); + return SQLITE_OK; + } + case SQLITE_FCNTL_WIN32_GET_HANDLE: { + LPHANDLE phFile = (LPHANDLE)pArg; + *phFile = pFile->h; + OSTRACE(("FCNTL file=%p, rc=SQLITE_OK\n", pFile->h)); + return SQLITE_OK; + } +#ifdef SQLITE_TEST + case SQLITE_FCNTL_WIN32_SET_HANDLE: { + LPHANDLE phFile = (LPHANDLE)pArg; + HANDLE hOldFile = pFile->h; + pFile->h = *phFile; + *phFile = hOldFile; + OSTRACE(("FCNTL oldFile=%p, newFile=%p, rc=SQLITE_OK\n", + hOldFile, pFile->h)); + return SQLITE_OK; + } +#endif + case SQLITE_FCNTL_TEMPFILENAME: { + char *zTFile = 0; + int rc = winGetTempname(pFile->pVfs, &zTFile); + if( rc==SQLITE_OK ){ + *(char**)pArg = zTFile; + } + OSTRACE(("FCNTL file=%p, rc=%s\n", pFile->h, sqlite3ErrName(rc))); + return rc; + } +#if SQLITE_MAX_MMAP_SIZE>0 + case SQLITE_FCNTL_MMAP_SIZE: { + i64 newLimit = *(i64*)pArg; + int rc = SQLITE_OK; + if( newLimit>sqlite3GlobalConfig.mxMmap ){ + newLimit = sqlite3GlobalConfig.mxMmap; + } + + /* The value of newLimit may be eventually cast to (SIZE_T) and passed + ** to MapViewOfFile(). Restrict its value to 2GB if (SIZE_T) is not at + ** least a 64-bit type. */ + if( newLimit>0 && sizeof(SIZE_T)<8 ){ + newLimit = (newLimit & 0x7FFFFFFF); + } + + *(i64*)pArg = pFile->mmapSizeMax; + if( newLimit>=0 && newLimit!=pFile->mmapSizeMax && pFile->nFetchOut==0 ){ + pFile->mmapSizeMax = newLimit; + if( pFile->mmapSize>0 ){ + winUnmapfile(pFile); + rc = winMapfile(pFile, -1); + } + } + OSTRACE(("FCNTL file=%p, rc=%s\n", pFile->h, sqlite3ErrName(rc))); + return rc; + } +#endif + } + OSTRACE(("FCNTL file=%p, rc=SQLITE_NOTFOUND\n", pFile->h)); + return SQLITE_NOTFOUND; +} + +/* +** Return the sector size in bytes of the underlying block device for +** the specified file. This is almost always 512 bytes, but may be +** larger for some devices. +** +** SQLite code assumes this function cannot fail. It also assumes that +** if two files are created in the same file-system directory (i.e. +** a database and its journal file) that the sector size will be the +** same for both. +*/ +static int winSectorSize(sqlite3_file *id){ + (void)id; + return SQLITE_DEFAULT_SECTOR_SIZE; +} + +/* +** Return a vector of device characteristics. +*/ +static int winDeviceCharacteristics(sqlite3_file *id){ + winFile *p = (winFile*)id; + return SQLITE_IOCAP_UNDELETABLE_WHEN_OPEN | + ((p->ctrlFlags & WINFILE_PSOW)?SQLITE_IOCAP_POWERSAFE_OVERWRITE:0); +} + +/* +** Windows will only let you create file view mappings +** on allocation size granularity boundaries. +** During sqlite3_os_init() we do a GetSystemInfo() +** to get the granularity size. +*/ +static SYSTEM_INFO winSysInfo; + +#ifndef SQLITE_OMIT_WAL + +/* +** Helper functions to obtain and relinquish the global mutex. The +** global mutex is used to protect the winLockInfo objects used by +** this file, all of which may be shared by multiple threads. +** +** Function winShmMutexHeld() is used to assert() that the global mutex +** is held when required. This function is only used as part of assert() +** statements. e.g. +** +** winShmEnterMutex() +** assert( winShmMutexHeld() ); +** winShmLeaveMutex() +*/ +static sqlite3_mutex *winBigLock = 0; +static void winShmEnterMutex(void){ + sqlite3_mutex_enter(winBigLock); +} +static void winShmLeaveMutex(void){ + sqlite3_mutex_leave(winBigLock); +} +#ifndef NDEBUG +static int winShmMutexHeld(void) { + return sqlite3_mutex_held(winBigLock); +} +#endif + +/* +** Object used to represent a single file opened and mmapped to provide +** shared memory. When multiple threads all reference the same +** log-summary, each thread has its own winFile object, but they all +** point to a single instance of this object. In other words, each +** log-summary is opened only once per process. +** +** winShmMutexHeld() must be true when creating or destroying +** this object or while reading or writing the following fields: +** +** nRef +** pNext +** +** The following fields are read-only after the object is created: +** +** fid +** zFilename +** +** Either winShmNode.mutex must be held or winShmNode.nRef==0 and +** winShmMutexHeld() is true when reading or writing any other field +** in this structure. +** +*/ +struct winShmNode { + sqlite3_mutex *mutex; /* Mutex to access this object */ + char *zFilename; /* Name of the file */ + winFile hFile; /* File handle from winOpen */ + + int szRegion; /* Size of shared-memory regions */ + int nRegion; /* Size of array apRegion */ + u8 isReadonly; /* True if read-only */ + u8 isUnlocked; /* True if no DMS lock held */ + + struct ShmRegion { + HANDLE hMap; /* File handle from CreateFileMapping */ + void *pMap; + } *aRegion; + DWORD lastErrno; /* The Windows errno from the last I/O error */ + + int nRef; /* Number of winShm objects pointing to this */ + winShm *pFirst; /* All winShm objects pointing to this */ + winShmNode *pNext; /* Next in list of all winShmNode objects */ +#if defined(SQLITE_DEBUG) || defined(SQLITE_HAVE_OS_TRACE) + u8 nextShmId; /* Next available winShm.id value */ +#endif +}; + +/* +** A global array of all winShmNode objects. +** +** The winShmMutexHeld() must be true while reading or writing this list. +*/ +static winShmNode *winShmNodeList = 0; + +/* +** Structure used internally by this VFS to record the state of an +** open shared memory connection. +** +** The following fields are initialized when this object is created and +** are read-only thereafter: +** +** winShm.pShmNode +** winShm.id +** +** All other fields are read/write. The winShm.pShmNode->mutex must be held +** while accessing any read/write fields. +*/ +struct winShm { + winShmNode *pShmNode; /* The underlying winShmNode object */ + winShm *pNext; /* Next winShm with the same winShmNode */ + u8 hasMutex; /* True if holding the winShmNode mutex */ + u16 sharedMask; /* Mask of shared locks held */ + u16 exclMask; /* Mask of exclusive locks held */ +#if defined(SQLITE_DEBUG) || defined(SQLITE_HAVE_OS_TRACE) + u8 id; /* Id of this connection with its winShmNode */ +#endif +}; + +/* +** Constants used for locking +*/ +#define WIN_SHM_BASE ((22+SQLITE_SHM_NLOCK)*4) /* first lock byte */ +#define WIN_SHM_DMS (WIN_SHM_BASE+SQLITE_SHM_NLOCK) /* deadman switch */ + +/* +** Apply advisory locks for all n bytes beginning at ofst. +*/ +#define WINSHM_UNLCK 1 +#define WINSHM_RDLCK 2 +#define WINSHM_WRLCK 3 +static int winShmSystemLock( + winShmNode *pFile, /* Apply locks to this open shared-memory segment */ + int lockType, /* WINSHM_UNLCK, WINSHM_RDLCK, or WINSHM_WRLCK */ + int ofst, /* Offset to first byte to be locked/unlocked */ + int nByte /* Number of bytes to lock or unlock */ +){ + int rc = 0; /* Result code form Lock/UnlockFileEx() */ + + /* Access to the winShmNode object is serialized by the caller */ + assert( pFile->nRef==0 || sqlite3_mutex_held(pFile->mutex) ); + + OSTRACE(("SHM-LOCK file=%p, lock=%d, offset=%d, size=%d\n", + pFile->hFile.h, lockType, ofst, nByte)); + + /* Release/Acquire the system-level lock */ + if( lockType==WINSHM_UNLCK ){ + rc = winUnlockFile(&pFile->hFile.h, ofst, 0, nByte, 0); + }else{ + /* Initialize the locking parameters */ + DWORD dwFlags = LOCKFILE_FAIL_IMMEDIATELY; + if( lockType == WINSHM_WRLCK ) dwFlags |= LOCKFILE_EXCLUSIVE_LOCK; + rc = winLockFile(&pFile->hFile.h, dwFlags, ofst, 0, nByte, 0); + } + + if( rc!= 0 ){ + rc = SQLITE_OK; + }else{ + pFile->lastErrno = osGetLastError(); + rc = SQLITE_BUSY; + } + + OSTRACE(("SHM-LOCK file=%p, func=%s, errno=%lu, rc=%s\n", + pFile->hFile.h, (lockType == WINSHM_UNLCK) ? "winUnlockFile" : + "winLockFile", pFile->lastErrno, sqlite3ErrName(rc))); + + return rc; +} + +/* Forward references to VFS methods */ +static int winOpen(sqlite3_vfs*,const char*,sqlite3_file*,int,int*); +static int winDelete(sqlite3_vfs *,const char*,int); + +/* +** Purge the winShmNodeList list of all entries with winShmNode.nRef==0. +** +** This is not a VFS shared-memory method; it is a utility function called +** by VFS shared-memory methods. +*/ +static void winShmPurge(sqlite3_vfs *pVfs, int deleteFlag){ + winShmNode **pp; + winShmNode *p; + assert( winShmMutexHeld() ); + OSTRACE(("SHM-PURGE pid=%lu, deleteFlag=%d\n", + osGetCurrentProcessId(), deleteFlag)); + pp = &winShmNodeList; + while( (p = *pp)!=0 ){ + if( p->nRef==0 ){ + int i; + if( p->mutex ){ sqlite3_mutex_free(p->mutex); } + for(i=0; inRegion; i++){ + BOOL bRc = osUnmapViewOfFile(p->aRegion[i].pMap); + OSTRACE(("SHM-PURGE-UNMAP pid=%lu, region=%d, rc=%s\n", + osGetCurrentProcessId(), i, bRc ? "ok" : "failed")); + UNUSED_VARIABLE_VALUE(bRc); + bRc = osCloseHandle(p->aRegion[i].hMap); + OSTRACE(("SHM-PURGE-CLOSE pid=%lu, region=%d, rc=%s\n", + osGetCurrentProcessId(), i, bRc ? "ok" : "failed")); + UNUSED_VARIABLE_VALUE(bRc); + } + if( p->hFile.h!=NULL && p->hFile.h!=INVALID_HANDLE_VALUE ){ + SimulateIOErrorBenign(1); + winClose((sqlite3_file *)&p->hFile); + SimulateIOErrorBenign(0); + } + if( deleteFlag ){ + SimulateIOErrorBenign(1); + sqlite3BeginBenignMalloc(); + winDelete(pVfs, p->zFilename, 0); + sqlite3EndBenignMalloc(); + SimulateIOErrorBenign(0); + } + *pp = p->pNext; + sqlite3_free(p->aRegion); + sqlite3_free(p); + }else{ + pp = &p->pNext; + } + } +} + +/* +** The DMS lock has not yet been taken on shm file pShmNode. Attempt to +** take it now. Return SQLITE_OK if successful, or an SQLite error +** code otherwise. +** +** If the DMS cannot be locked because this is a readonly_shm=1 +** connection and no other process already holds a lock, return +** SQLITE_READONLY_CANTINIT and set pShmNode->isUnlocked=1. +*/ +static int winLockSharedMemory(winShmNode *pShmNode){ + int rc = winShmSystemLock(pShmNode, WINSHM_WRLCK, WIN_SHM_DMS, 1); + + if( rc==SQLITE_OK ){ + if( pShmNode->isReadonly ){ + pShmNode->isUnlocked = 1; + winShmSystemLock(pShmNode, WINSHM_UNLCK, WIN_SHM_DMS, 1); + return SQLITE_READONLY_CANTINIT; + }else if( winTruncate((sqlite3_file*)&pShmNode->hFile, 0) ){ + winShmSystemLock(pShmNode, WINSHM_UNLCK, WIN_SHM_DMS, 1); + return winLogError(SQLITE_IOERR_SHMOPEN, osGetLastError(), + "winLockSharedMemory", pShmNode->zFilename); + } + } + + if( rc==SQLITE_OK ){ + winShmSystemLock(pShmNode, WINSHM_UNLCK, WIN_SHM_DMS, 1); + } + + return winShmSystemLock(pShmNode, WINSHM_RDLCK, WIN_SHM_DMS, 1); +} + +/* +** Open the shared-memory area associated with database file pDbFd. +** +** When opening a new shared-memory file, if no other instances of that +** file are currently open, in this process or in other processes, then +** the file must be truncated to zero length or have its header cleared. +*/ +static int winOpenSharedMemory(winFile *pDbFd){ + struct winShm *p; /* The connection to be opened */ + winShmNode *pShmNode = 0; /* The underlying mmapped file */ + int rc = SQLITE_OK; /* Result code */ + winShmNode *pNew; /* Newly allocated winShmNode */ + int nName; /* Size of zName in bytes */ + + assert( pDbFd->pShm==0 ); /* Not previously opened */ + + /* Allocate space for the new sqlite3_shm object. Also speculatively + ** allocate space for a new winShmNode and filename. + */ + p = sqlite3MallocZero( sizeof(*p) ); + if( p==0 ) return SQLITE_IOERR_NOMEM_BKPT; + nName = sqlite3Strlen30(pDbFd->zPath); + pNew = sqlite3MallocZero( sizeof(*pShmNode) + nName + 17 ); + if( pNew==0 ){ + sqlite3_free(p); + return SQLITE_IOERR_NOMEM_BKPT; + } + pNew->zFilename = (char*)&pNew[1]; + sqlite3_snprintf(nName+15, pNew->zFilename, "%s-shm", pDbFd->zPath); + sqlite3FileSuffix3(pDbFd->zPath, pNew->zFilename); + + /* Look to see if there is an existing winShmNode that can be used. + ** If no matching winShmNode currently exists, create a new one. + */ + winShmEnterMutex(); + for(pShmNode = winShmNodeList; pShmNode; pShmNode=pShmNode->pNext){ + /* TBD need to come up with better match here. Perhaps + ** use FILE_ID_BOTH_DIR_INFO Structure. + */ + if( sqlite3StrICmp(pShmNode->zFilename, pNew->zFilename)==0 ) break; + } + if( pShmNode ){ + sqlite3_free(pNew); + }else{ + int inFlags = SQLITE_OPEN_WAL; + int outFlags = 0; + + pShmNode = pNew; + pNew = 0; + ((winFile*)(&pShmNode->hFile))->h = INVALID_HANDLE_VALUE; + pShmNode->pNext = winShmNodeList; + winShmNodeList = pShmNode; + + if( sqlite3GlobalConfig.bCoreMutex ){ + pShmNode->mutex = sqlite3_mutex_alloc(SQLITE_MUTEX_FAST); + if( pShmNode->mutex==0 ){ + rc = SQLITE_IOERR_NOMEM_BKPT; + goto shm_open_err; + } + } + + if( 0==sqlite3_uri_boolean(pDbFd->zPath, "readonly_shm", 0) ){ + inFlags |= SQLITE_OPEN_READWRITE | SQLITE_OPEN_CREATE; + }else{ + inFlags |= SQLITE_OPEN_READONLY; + } + rc = winOpen(pDbFd->pVfs, pShmNode->zFilename, + (sqlite3_file*)&pShmNode->hFile, + inFlags, &outFlags); + if( rc!=SQLITE_OK ){ + rc = winLogError(rc, osGetLastError(), "winOpenShm", + pShmNode->zFilename); + goto shm_open_err; + } + if( outFlags==SQLITE_OPEN_READONLY ) pShmNode->isReadonly = 1; + + rc = winLockSharedMemory(pShmNode); + if( rc!=SQLITE_OK && rc!=SQLITE_READONLY_CANTINIT ) goto shm_open_err; + } + + /* Make the new connection a child of the winShmNode */ + p->pShmNode = pShmNode; +#if defined(SQLITE_DEBUG) || defined(SQLITE_HAVE_OS_TRACE) + p->id = pShmNode->nextShmId++; +#endif + pShmNode->nRef++; + pDbFd->pShm = p; + winShmLeaveMutex(); + + /* The reference count on pShmNode has already been incremented under + ** the cover of the winShmEnterMutex() mutex and the pointer from the + ** new (struct winShm) object to the pShmNode has been set. All that is + ** left to do is to link the new object into the linked list starting + ** at pShmNode->pFirst. This must be done while holding the pShmNode->mutex + ** mutex. + */ + sqlite3_mutex_enter(pShmNode->mutex); + p->pNext = pShmNode->pFirst; + pShmNode->pFirst = p; + sqlite3_mutex_leave(pShmNode->mutex); + return rc; + + /* Jump here on any error */ +shm_open_err: + winShmSystemLock(pShmNode, WINSHM_UNLCK, WIN_SHM_DMS, 1); + winShmPurge(pDbFd->pVfs, 0); /* This call frees pShmNode if required */ + sqlite3_free(p); + sqlite3_free(pNew); + winShmLeaveMutex(); + return rc; +} + +/* +** Close a connection to shared-memory. Delete the underlying +** storage if deleteFlag is true. +*/ +static int winShmUnmap( + sqlite3_file *fd, /* Database holding shared memory */ + int deleteFlag /* Delete after closing if true */ +){ + winFile *pDbFd; /* Database holding shared-memory */ + winShm *p; /* The connection to be closed */ + winShmNode *pShmNode; /* The underlying shared-memory file */ + winShm **pp; /* For looping over sibling connections */ + + pDbFd = (winFile*)fd; + p = pDbFd->pShm; + if( p==0 ) return SQLITE_OK; + pShmNode = p->pShmNode; + + /* Remove connection p from the set of connections associated + ** with pShmNode */ + sqlite3_mutex_enter(pShmNode->mutex); + for(pp=&pShmNode->pFirst; (*pp)!=p; pp = &(*pp)->pNext){} + *pp = p->pNext; + + /* Free the connection p */ + sqlite3_free(p); + pDbFd->pShm = 0; + sqlite3_mutex_leave(pShmNode->mutex); + + /* If pShmNode->nRef has reached 0, then close the underlying + ** shared-memory file, too */ + winShmEnterMutex(); + assert( pShmNode->nRef>0 ); + pShmNode->nRef--; + if( pShmNode->nRef==0 ){ + winShmPurge(pDbFd->pVfs, deleteFlag); + } + winShmLeaveMutex(); + + return SQLITE_OK; +} + +/* +** Change the lock state for a shared-memory segment. +*/ +static int winShmLock( + sqlite3_file *fd, /* Database file holding the shared memory */ + int ofst, /* First lock to acquire or release */ + int n, /* Number of locks to acquire or release */ + int flags /* What to do with the lock */ +){ + winFile *pDbFd = (winFile*)fd; /* Connection holding shared memory */ + winShm *p = pDbFd->pShm; /* The shared memory being locked */ + winShm *pX; /* For looping over all siblings */ + winShmNode *pShmNode = p->pShmNode; + int rc = SQLITE_OK; /* Result code */ + u16 mask; /* Mask of locks to take or release */ + + assert( ofst>=0 && ofst+n<=SQLITE_SHM_NLOCK ); + assert( n>=1 ); + assert( flags==(SQLITE_SHM_LOCK | SQLITE_SHM_SHARED) + || flags==(SQLITE_SHM_LOCK | SQLITE_SHM_EXCLUSIVE) + || flags==(SQLITE_SHM_UNLOCK | SQLITE_SHM_SHARED) + || flags==(SQLITE_SHM_UNLOCK | SQLITE_SHM_EXCLUSIVE) ); + assert( n==1 || (flags & SQLITE_SHM_EXCLUSIVE)!=0 ); + + mask = (u16)((1U<<(ofst+n)) - (1U<1 || mask==(1<mutex); + if( flags & SQLITE_SHM_UNLOCK ){ + u16 allMask = 0; /* Mask of locks held by siblings */ + + /* See if any siblings hold this same lock */ + for(pX=pShmNode->pFirst; pX; pX=pX->pNext){ + if( pX==p ) continue; + assert( (pX->exclMask & (p->exclMask|p->sharedMask))==0 ); + allMask |= pX->sharedMask; + } + + /* Unlock the system-level locks */ + if( (mask & allMask)==0 ){ + rc = winShmSystemLock(pShmNode, WINSHM_UNLCK, ofst+WIN_SHM_BASE, n); + }else{ + rc = SQLITE_OK; + } + + /* Undo the local locks */ + if( rc==SQLITE_OK ){ + p->exclMask &= ~mask; + p->sharedMask &= ~mask; + } + }else if( flags & SQLITE_SHM_SHARED ){ + u16 allShared = 0; /* Union of locks held by connections other than "p" */ + + /* Find out which shared locks are already held by sibling connections. + ** If any sibling already holds an exclusive lock, go ahead and return + ** SQLITE_BUSY. + */ + for(pX=pShmNode->pFirst; pX; pX=pX->pNext){ + if( (pX->exclMask & mask)!=0 ){ + rc = SQLITE_BUSY; + break; + } + allShared |= pX->sharedMask; + } + + /* Get shared locks at the system level, if necessary */ + if( rc==SQLITE_OK ){ + if( (allShared & mask)==0 ){ + rc = winShmSystemLock(pShmNode, WINSHM_RDLCK, ofst+WIN_SHM_BASE, n); + }else{ + rc = SQLITE_OK; + } + } + + /* Get the local shared locks */ + if( rc==SQLITE_OK ){ + p->sharedMask |= mask; + } + }else{ + /* Make sure no sibling connections hold locks that will block this + ** lock. If any do, return SQLITE_BUSY right away. + */ + for(pX=pShmNode->pFirst; pX; pX=pX->pNext){ + if( (pX->exclMask & mask)!=0 || (pX->sharedMask & mask)!=0 ){ + rc = SQLITE_BUSY; + break; + } + } + + /* Get the exclusive locks at the system level. Then if successful + ** also mark the local connection as being locked. + */ + if( rc==SQLITE_OK ){ + rc = winShmSystemLock(pShmNode, WINSHM_WRLCK, ofst+WIN_SHM_BASE, n); + if( rc==SQLITE_OK ){ + assert( (p->sharedMask & mask)==0 ); + p->exclMask |= mask; + } + } + } + sqlite3_mutex_leave(pShmNode->mutex); + OSTRACE(("SHM-LOCK pid=%lu, id=%d, sharedMask=%03x, exclMask=%03x, rc=%s\n", + osGetCurrentProcessId(), p->id, p->sharedMask, p->exclMask, + sqlite3ErrName(rc))); + return rc; +} + +/* +** Implement a memory barrier or memory fence on shared memory. +** +** All loads and stores begun before the barrier must complete before +** any load or store begun after the barrier. +*/ +static void winShmBarrier( + sqlite3_file *fd /* Database holding the shared memory */ +){ + UNUSED_PARAMETER(fd); + sqlite3MemoryBarrier(); /* compiler-defined memory barrier */ + winShmEnterMutex(); /* Also mutex, for redundancy */ + winShmLeaveMutex(); +} + +/* +** This function is called to obtain a pointer to region iRegion of the +** shared-memory associated with the database file fd. Shared-memory regions +** are numbered starting from zero. Each shared-memory region is szRegion +** bytes in size. +** +** If an error occurs, an error code is returned and *pp is set to NULL. +** +** Otherwise, if the isWrite parameter is 0 and the requested shared-memory +** region has not been allocated (by any client, including one running in a +** separate process), then *pp is set to NULL and SQLITE_OK returned. If +** isWrite is non-zero and the requested shared-memory region has not yet +** been allocated, it is allocated by this function. +** +** If the shared-memory region has already been allocated or is allocated by +** this call as described above, then it is mapped into this processes +** address space (if it is not already), *pp is set to point to the mapped +** memory and SQLITE_OK returned. +*/ +static int winShmMap( + sqlite3_file *fd, /* Handle open on database file */ + int iRegion, /* Region to retrieve */ + int szRegion, /* Size of regions */ + int isWrite, /* True to extend file if necessary */ + void volatile **pp /* OUT: Mapped memory */ +){ + winFile *pDbFd = (winFile*)fd; + winShm *pShm = pDbFd->pShm; + winShmNode *pShmNode; + DWORD protect = PAGE_READWRITE; + DWORD flags = FILE_MAP_WRITE | FILE_MAP_READ; + int rc = SQLITE_OK; + + if( !pShm ){ + rc = winOpenSharedMemory(pDbFd); + if( rc!=SQLITE_OK ) return rc; + pShm = pDbFd->pShm; + assert( pShm!=0 ); + } + pShmNode = pShm->pShmNode; + + sqlite3_mutex_enter(pShmNode->mutex); + if( pShmNode->isUnlocked ){ + rc = winLockSharedMemory(pShmNode); + if( rc!=SQLITE_OK ) goto shmpage_out; + pShmNode->isUnlocked = 0; + } + assert( szRegion==pShmNode->szRegion || pShmNode->nRegion==0 ); + + if( pShmNode->nRegion<=iRegion ){ + struct ShmRegion *apNew; /* New aRegion[] array */ + int nByte = (iRegion+1)*szRegion; /* Minimum required file size */ + sqlite3_int64 sz; /* Current size of wal-index file */ + + pShmNode->szRegion = szRegion; + + /* The requested region is not mapped into this processes address space. + ** Check to see if it has been allocated (i.e. if the wal-index file is + ** large enough to contain the requested region). + */ + rc = winFileSize((sqlite3_file *)&pShmNode->hFile, &sz); + if( rc!=SQLITE_OK ){ + rc = winLogError(SQLITE_IOERR_SHMSIZE, osGetLastError(), + "winShmMap1", pDbFd->zPath); + goto shmpage_out; + } + + if( szhFile, nByte); + if( rc!=SQLITE_OK ){ + rc = winLogError(SQLITE_IOERR_SHMSIZE, osGetLastError(), + "winShmMap2", pDbFd->zPath); + goto shmpage_out; + } + } + + /* Map the requested memory region into this processes address space. */ + apNew = (struct ShmRegion *)sqlite3_realloc64( + pShmNode->aRegion, (iRegion+1)*sizeof(apNew[0]) + ); + if( !apNew ){ + rc = SQLITE_IOERR_NOMEM_BKPT; + goto shmpage_out; + } + pShmNode->aRegion = apNew; + + if( pShmNode->isReadonly ){ + protect = PAGE_READONLY; + flags = FILE_MAP_READ; + } + + while( pShmNode->nRegion<=iRegion ){ + HANDLE hMap = NULL; /* file-mapping handle */ + void *pMap = 0; /* Mapped memory region */ + +#if SQLITE_OS_WINRT + hMap = osCreateFileMappingFromApp(pShmNode->hFile.h, + NULL, protect, nByte, NULL + ); +#elif defined(SQLITE_WIN32_HAS_WIDE) + hMap = osCreateFileMappingW(pShmNode->hFile.h, + NULL, protect, 0, nByte, NULL + ); +#elif defined(SQLITE_WIN32_HAS_ANSI) && SQLITE_WIN32_CREATEFILEMAPPINGA + hMap = osCreateFileMappingA(pShmNode->hFile.h, + NULL, protect, 0, nByte, NULL + ); +#endif + OSTRACE(("SHM-MAP-CREATE pid=%lu, region=%d, size=%d, rc=%s\n", + osGetCurrentProcessId(), pShmNode->nRegion, nByte, + hMap ? "ok" : "failed")); + if( hMap ){ + int iOffset = pShmNode->nRegion*szRegion; + int iOffsetShift = iOffset % winSysInfo.dwAllocationGranularity; +#if SQLITE_OS_WINRT + pMap = osMapViewOfFileFromApp(hMap, flags, + iOffset - iOffsetShift, szRegion + iOffsetShift + ); +#else + pMap = osMapViewOfFile(hMap, flags, + 0, iOffset - iOffsetShift, szRegion + iOffsetShift + ); +#endif + OSTRACE(("SHM-MAP-MAP pid=%lu, region=%d, offset=%d, size=%d, rc=%s\n", + osGetCurrentProcessId(), pShmNode->nRegion, iOffset, + szRegion, pMap ? "ok" : "failed")); + } + if( !pMap ){ + pShmNode->lastErrno = osGetLastError(); + rc = winLogError(SQLITE_IOERR_SHMMAP, pShmNode->lastErrno, + "winShmMap3", pDbFd->zPath); + if( hMap ) osCloseHandle(hMap); + goto shmpage_out; + } + + pShmNode->aRegion[pShmNode->nRegion].pMap = pMap; + pShmNode->aRegion[pShmNode->nRegion].hMap = hMap; + pShmNode->nRegion++; + } + } + +shmpage_out: + if( pShmNode->nRegion>iRegion ){ + int iOffset = iRegion*szRegion; + int iOffsetShift = iOffset % winSysInfo.dwAllocationGranularity; + char *p = (char *)pShmNode->aRegion[iRegion].pMap; + *pp = (void *)&p[iOffsetShift]; + }else{ + *pp = 0; + } + if( pShmNode->isReadonly && rc==SQLITE_OK ) rc = SQLITE_READONLY; + sqlite3_mutex_leave(pShmNode->mutex); + return rc; +} + +#else +# define winShmMap 0 +# define winShmLock 0 +# define winShmBarrier 0 +# define winShmUnmap 0 +#endif /* #ifndef SQLITE_OMIT_WAL */ + +/* +** Cleans up the mapped region of the specified file, if any. +*/ +#if SQLITE_MAX_MMAP_SIZE>0 +static int winUnmapfile(winFile *pFile){ + assert( pFile!=0 ); + OSTRACE(("UNMAP-FILE pid=%lu, pFile=%p, hMap=%p, pMapRegion=%p, " + "mmapSize=%lld, mmapSizeMax=%lld\n", + osGetCurrentProcessId(), pFile, pFile->hMap, pFile->pMapRegion, + pFile->mmapSize, pFile->mmapSizeMax)); + if( pFile->pMapRegion ){ + if( !osUnmapViewOfFile(pFile->pMapRegion) ){ + pFile->lastErrno = osGetLastError(); + OSTRACE(("UNMAP-FILE pid=%lu, pFile=%p, pMapRegion=%p, " + "rc=SQLITE_IOERR_MMAP\n", osGetCurrentProcessId(), pFile, + pFile->pMapRegion)); + return winLogError(SQLITE_IOERR_MMAP, pFile->lastErrno, + "winUnmapfile1", pFile->zPath); + } + pFile->pMapRegion = 0; + pFile->mmapSize = 0; + } + if( pFile->hMap!=NULL ){ + if( !osCloseHandle(pFile->hMap) ){ + pFile->lastErrno = osGetLastError(); + OSTRACE(("UNMAP-FILE pid=%lu, pFile=%p, hMap=%p, rc=SQLITE_IOERR_MMAP\n", + osGetCurrentProcessId(), pFile, pFile->hMap)); + return winLogError(SQLITE_IOERR_MMAP, pFile->lastErrno, + "winUnmapfile2", pFile->zPath); + } + pFile->hMap = NULL; + } + OSTRACE(("UNMAP-FILE pid=%lu, pFile=%p, rc=SQLITE_OK\n", + osGetCurrentProcessId(), pFile)); + return SQLITE_OK; +} + +/* +** Memory map or remap the file opened by file-descriptor pFd (if the file +** is already mapped, the existing mapping is replaced by the new). Or, if +** there already exists a mapping for this file, and there are still +** outstanding xFetch() references to it, this function is a no-op. +** +** If parameter nByte is non-negative, then it is the requested size of +** the mapping to create. Otherwise, if nByte is less than zero, then the +** requested size is the size of the file on disk. The actual size of the +** created mapping is either the requested size or the value configured +** using SQLITE_FCNTL_MMAP_SIZE, whichever is smaller. +** +** SQLITE_OK is returned if no error occurs (even if the mapping is not +** recreated as a result of outstanding references) or an SQLite error +** code otherwise. +*/ +static int winMapfile(winFile *pFd, sqlite3_int64 nByte){ + sqlite3_int64 nMap = nByte; + int rc; + + assert( nMap>=0 || pFd->nFetchOut==0 ); + OSTRACE(("MAP-FILE pid=%lu, pFile=%p, size=%lld\n", + osGetCurrentProcessId(), pFd, nByte)); + + if( pFd->nFetchOut>0 ) return SQLITE_OK; + + if( nMap<0 ){ + rc = winFileSize((sqlite3_file*)pFd, &nMap); + if( rc ){ + OSTRACE(("MAP-FILE pid=%lu, pFile=%p, rc=SQLITE_IOERR_FSTAT\n", + osGetCurrentProcessId(), pFd)); + return SQLITE_IOERR_FSTAT; + } + } + if( nMap>pFd->mmapSizeMax ){ + nMap = pFd->mmapSizeMax; + } + nMap &= ~(sqlite3_int64)(winSysInfo.dwPageSize - 1); + + if( nMap==0 && pFd->mmapSize>0 ){ + winUnmapfile(pFd); + } + if( nMap!=pFd->mmapSize ){ + void *pNew = 0; + DWORD protect = PAGE_READONLY; + DWORD flags = FILE_MAP_READ; + + winUnmapfile(pFd); +#ifdef SQLITE_MMAP_READWRITE + if( (pFd->ctrlFlags & WINFILE_RDONLY)==0 ){ + protect = PAGE_READWRITE; + flags |= FILE_MAP_WRITE; + } +#endif +#if SQLITE_OS_WINRT + pFd->hMap = osCreateFileMappingFromApp(pFd->h, NULL, protect, nMap, NULL); +#elif defined(SQLITE_WIN32_HAS_WIDE) + pFd->hMap = osCreateFileMappingW(pFd->h, NULL, protect, + (DWORD)((nMap>>32) & 0xffffffff), + (DWORD)(nMap & 0xffffffff), NULL); +#elif defined(SQLITE_WIN32_HAS_ANSI) && SQLITE_WIN32_CREATEFILEMAPPINGA + pFd->hMap = osCreateFileMappingA(pFd->h, NULL, protect, + (DWORD)((nMap>>32) & 0xffffffff), + (DWORD)(nMap & 0xffffffff), NULL); +#endif + if( pFd->hMap==NULL ){ + pFd->lastErrno = osGetLastError(); + rc = winLogError(SQLITE_IOERR_MMAP, pFd->lastErrno, + "winMapfile1", pFd->zPath); + /* Log the error, but continue normal operation using xRead/xWrite */ + OSTRACE(("MAP-FILE-CREATE pid=%lu, pFile=%p, rc=%s\n", + osGetCurrentProcessId(), pFd, sqlite3ErrName(rc))); + return SQLITE_OK; + } + assert( (nMap % winSysInfo.dwPageSize)==0 ); + assert( sizeof(SIZE_T)==sizeof(sqlite3_int64) || nMap<=0xffffffff ); +#if SQLITE_OS_WINRT + pNew = osMapViewOfFileFromApp(pFd->hMap, flags, 0, (SIZE_T)nMap); +#else + pNew = osMapViewOfFile(pFd->hMap, flags, 0, 0, (SIZE_T)nMap); +#endif + if( pNew==NULL ){ + osCloseHandle(pFd->hMap); + pFd->hMap = NULL; + pFd->lastErrno = osGetLastError(); + rc = winLogError(SQLITE_IOERR_MMAP, pFd->lastErrno, + "winMapfile2", pFd->zPath); + /* Log the error, but continue normal operation using xRead/xWrite */ + OSTRACE(("MAP-FILE-MAP pid=%lu, pFile=%p, rc=%s\n", + osGetCurrentProcessId(), pFd, sqlite3ErrName(rc))); + return SQLITE_OK; + } + pFd->pMapRegion = pNew; + pFd->mmapSize = nMap; + } + + OSTRACE(("MAP-FILE pid=%lu, pFile=%p, rc=SQLITE_OK\n", + osGetCurrentProcessId(), pFd)); + return SQLITE_OK; +} +#endif /* SQLITE_MAX_MMAP_SIZE>0 */ + +/* +** If possible, return a pointer to a mapping of file fd starting at offset +** iOff. The mapping must be valid for at least nAmt bytes. +** +** If such a pointer can be obtained, store it in *pp and return SQLITE_OK. +** Or, if one cannot but no error occurs, set *pp to 0 and return SQLITE_OK. +** Finally, if an error does occur, return an SQLite error code. The final +** value of *pp is undefined in this case. +** +** If this function does return a pointer, the caller must eventually +** release the reference by calling winUnfetch(). +*/ +static int winFetch(sqlite3_file *fd, i64 iOff, int nAmt, void **pp){ +#if SQLITE_MAX_MMAP_SIZE>0 + winFile *pFd = (winFile*)fd; /* The underlying database file */ +#endif + *pp = 0; + + OSTRACE(("FETCH pid=%lu, pFile=%p, offset=%lld, amount=%d, pp=%p\n", + osGetCurrentProcessId(), fd, iOff, nAmt, pp)); + +#if SQLITE_MAX_MMAP_SIZE>0 + if( pFd->mmapSizeMax>0 ){ + if( pFd->pMapRegion==0 ){ + int rc = winMapfile(pFd, -1); + if( rc!=SQLITE_OK ){ + OSTRACE(("FETCH pid=%lu, pFile=%p, rc=%s\n", + osGetCurrentProcessId(), pFd, sqlite3ErrName(rc))); + return rc; + } + } + if( pFd->mmapSize >= iOff+nAmt ){ + assert( pFd->pMapRegion!=0 ); + *pp = &((u8 *)pFd->pMapRegion)[iOff]; + pFd->nFetchOut++; + } + } +#endif + + OSTRACE(("FETCH pid=%lu, pFile=%p, pp=%p, *pp=%p, rc=SQLITE_OK\n", + osGetCurrentProcessId(), fd, pp, *pp)); + return SQLITE_OK; +} + +/* +** If the third argument is non-NULL, then this function releases a +** reference obtained by an earlier call to winFetch(). The second +** argument passed to this function must be the same as the corresponding +** argument that was passed to the winFetch() invocation. +** +** Or, if the third argument is NULL, then this function is being called +** to inform the VFS layer that, according to POSIX, any existing mapping +** may now be invalid and should be unmapped. +*/ +static int winUnfetch(sqlite3_file *fd, i64 iOff, void *p){ +#if SQLITE_MAX_MMAP_SIZE>0 + winFile *pFd = (winFile*)fd; /* The underlying database file */ + + /* If p==0 (unmap the entire file) then there must be no outstanding + ** xFetch references. Or, if p!=0 (meaning it is an xFetch reference), + ** then there must be at least one outstanding. */ + assert( (p==0)==(pFd->nFetchOut==0) ); + + /* If p!=0, it must match the iOff value. */ + assert( p==0 || p==&((u8 *)pFd->pMapRegion)[iOff] ); + + OSTRACE(("UNFETCH pid=%lu, pFile=%p, offset=%lld, p=%p\n", + osGetCurrentProcessId(), pFd, iOff, p)); + + if( p ){ + pFd->nFetchOut--; + }else{ + /* FIXME: If Windows truly always prevents truncating or deleting a + ** file while a mapping is held, then the following winUnmapfile() call + ** is unnecessary can be omitted - potentially improving + ** performance. */ + winUnmapfile(pFd); + } + + assert( pFd->nFetchOut>=0 ); +#endif + + OSTRACE(("UNFETCH pid=%lu, pFile=%p, rc=SQLITE_OK\n", + osGetCurrentProcessId(), fd)); + return SQLITE_OK; +} + +/* +** Here ends the implementation of all sqlite3_file methods. +** +********************** End sqlite3_file Methods ******************************* +******************************************************************************/ + +/* +** This vector defines all the methods that can operate on an +** sqlite3_file for win32. +*/ +static const sqlite3_io_methods winIoMethod = { + 3, /* iVersion */ + winClose, /* xClose */ + winRead, /* xRead */ + winWrite, /* xWrite */ + winTruncate, /* xTruncate */ + winSync, /* xSync */ + winFileSize, /* xFileSize */ + winLock, /* xLock */ + winUnlock, /* xUnlock */ + winCheckReservedLock, /* xCheckReservedLock */ + winFileControl, /* xFileControl */ + winSectorSize, /* xSectorSize */ + winDeviceCharacteristics, /* xDeviceCharacteristics */ + winShmMap, /* xShmMap */ + winShmLock, /* xShmLock */ + winShmBarrier, /* xShmBarrier */ + winShmUnmap, /* xShmUnmap */ + winFetch, /* xFetch */ + winUnfetch /* xUnfetch */ +}; + +/* +** This vector defines all the methods that can operate on an +** sqlite3_file for win32 without performing any locking. +*/ +static const sqlite3_io_methods winIoNolockMethod = { + 3, /* iVersion */ + winClose, /* xClose */ + winRead, /* xRead */ + winWrite, /* xWrite */ + winTruncate, /* xTruncate */ + winSync, /* xSync */ + winFileSize, /* xFileSize */ + winNolockLock, /* xLock */ + winNolockUnlock, /* xUnlock */ + winNolockCheckReservedLock, /* xCheckReservedLock */ + winFileControl, /* xFileControl */ + winSectorSize, /* xSectorSize */ + winDeviceCharacteristics, /* xDeviceCharacteristics */ + winShmMap, /* xShmMap */ + winShmLock, /* xShmLock */ + winShmBarrier, /* xShmBarrier */ + winShmUnmap, /* xShmUnmap */ + winFetch, /* xFetch */ + winUnfetch /* xUnfetch */ +}; + +static winVfsAppData winAppData = { + &winIoMethod, /* pMethod */ + 0, /* pAppData */ + 0 /* bNoLock */ +}; + +static winVfsAppData winNolockAppData = { + &winIoNolockMethod, /* pMethod */ + 0, /* pAppData */ + 1 /* bNoLock */ +}; + +/**************************************************************************** +**************************** sqlite3_vfs methods **************************** +** +** This division contains the implementation of methods on the +** sqlite3_vfs object. +*/ + +#if defined(__CYGWIN__) +/* +** Convert a filename from whatever the underlying operating system +** supports for filenames into UTF-8. Space to hold the result is +** obtained from malloc and must be freed by the calling function. +*/ +static char *winConvertToUtf8Filename(const void *zFilename){ + char *zConverted = 0; + if( osIsNT() ){ + zConverted = winUnicodeToUtf8(zFilename); + } +#ifdef SQLITE_WIN32_HAS_ANSI + else{ + zConverted = winMbcsToUtf8(zFilename, osAreFileApisANSI()); + } +#endif + /* caller will handle out of memory */ + return zConverted; +} +#endif + +/* +** Convert a UTF-8 filename into whatever form the underlying +** operating system wants filenames in. Space to hold the result +** is obtained from malloc and must be freed by the calling +** function. +*/ +static void *winConvertFromUtf8Filename(const char *zFilename){ + void *zConverted = 0; + if( osIsNT() ){ + zConverted = winUtf8ToUnicode(zFilename); + } +#ifdef SQLITE_WIN32_HAS_ANSI + else{ + zConverted = winUtf8ToMbcs(zFilename, osAreFileApisANSI()); + } +#endif + /* caller will handle out of memory */ + return zConverted; +} + +/* +** This function returns non-zero if the specified UTF-8 string buffer +** ends with a directory separator character or one was successfully +** added to it. +*/ +static int winMakeEndInDirSep(int nBuf, char *zBuf){ + if( zBuf ){ + int nLen = sqlite3Strlen30(zBuf); + if( nLen>0 ){ + if( winIsDirSep(zBuf[nLen-1]) ){ + return 1; + }else if( nLen+1mxPathname; nBuf = nMax + 2; + zBuf = sqlite3MallocZero( nBuf ); + if( !zBuf ){ + OSTRACE(("TEMP-FILENAME rc=SQLITE_IOERR_NOMEM\n")); + return SQLITE_IOERR_NOMEM_BKPT; + } + + /* Figure out the effective temporary directory. First, check if one + ** has been explicitly set by the application; otherwise, use the one + ** configured by the operating system. + */ + nDir = nMax - (nPre + 15); + assert( nDir>0 ); + if( sqlite3_temp_directory ){ + int nDirLen = sqlite3Strlen30(sqlite3_temp_directory); + if( nDirLen>0 ){ + if( !winIsDirSep(sqlite3_temp_directory[nDirLen-1]) ){ + nDirLen++; + } + if( nDirLen>nDir ){ + sqlite3_free(zBuf); + OSTRACE(("TEMP-FILENAME rc=SQLITE_ERROR\n")); + return winLogError(SQLITE_ERROR, 0, "winGetTempname1", 0); + } + sqlite3_snprintf(nMax, zBuf, "%s", sqlite3_temp_directory); + } + } +#if defined(__CYGWIN__) + else{ + static const char *azDirs[] = { + 0, /* getenv("SQLITE_TMPDIR") */ + 0, /* getenv("TMPDIR") */ + 0, /* getenv("TMP") */ + 0, /* getenv("TEMP") */ + 0, /* getenv("USERPROFILE") */ + "/var/tmp", + "/usr/tmp", + "/tmp", + ".", + 0 /* List terminator */ + }; + unsigned int i; + const char *zDir = 0; + + if( !azDirs[0] ) azDirs[0] = getenv("SQLITE_TMPDIR"); + if( !azDirs[1] ) azDirs[1] = getenv("TMPDIR"); + if( !azDirs[2] ) azDirs[2] = getenv("TMP"); + if( !azDirs[3] ) azDirs[3] = getenv("TEMP"); + if( !azDirs[4] ) azDirs[4] = getenv("USERPROFILE"); + for(i=0; i/etilqs_XXXXXXXXXXXXXXX\0\0" + ** + ** If not, return SQLITE_ERROR. The number 17 is used here in order to + ** account for the space used by the 15 character random suffix and the + ** two trailing NUL characters. The final directory separator character + ** has already added if it was not already present. + */ + nLen = sqlite3Strlen30(zBuf); + if( (nLen + nPre + 17) > nBuf ){ + sqlite3_free(zBuf); + OSTRACE(("TEMP-FILENAME rc=SQLITE_ERROR\n")); + return winLogError(SQLITE_ERROR, 0, "winGetTempname5", 0); + } + + sqlite3_snprintf(nBuf-16-nLen, zBuf+nLen, SQLITE_TEMP_FILE_PREFIX); + + j = sqlite3Strlen30(zBuf); + sqlite3_randomness(15, &zBuf[j]); + for(i=0; i<15; i++, j++){ + zBuf[j] = (char)zChars[ ((unsigned char)zBuf[j])%(sizeof(zChars)-1) ]; + } + zBuf[j] = 0; + zBuf[j+1] = 0; + *pzBuf = zBuf; + + OSTRACE(("TEMP-FILENAME name=%s, rc=SQLITE_OK\n", zBuf)); + return SQLITE_OK; +} + +/* +** Return TRUE if the named file is really a directory. Return false if +** it is something other than a directory, or if there is any kind of memory +** allocation failure. +*/ +static int winIsDir(const void *zConverted){ + DWORD attr; + int rc = 0; + DWORD lastErrno; + + if( osIsNT() ){ + int cnt = 0; + WIN32_FILE_ATTRIBUTE_DATA sAttrData; + memset(&sAttrData, 0, sizeof(sAttrData)); + while( !(rc = osGetFileAttributesExW((LPCWSTR)zConverted, + GetFileExInfoStandard, + &sAttrData)) && winRetryIoerr(&cnt, &lastErrno) ){} + if( !rc ){ + return 0; /* Invalid name? */ + } + attr = sAttrData.dwFileAttributes; +#if SQLITE_OS_WINCE==0 + }else{ + attr = osGetFileAttributesA((char*)zConverted); +#endif + } + return (attr!=INVALID_FILE_ATTRIBUTES) && (attr&FILE_ATTRIBUTE_DIRECTORY); +} + +/* forward reference */ +static int winAccess( + sqlite3_vfs *pVfs, /* Not used on win32 */ + const char *zFilename, /* Name of file to check */ + int flags, /* Type of test to make on this file */ + int *pResOut /* OUT: Result */ +); + +/* +** Open a file. +*/ +static int winOpen( + sqlite3_vfs *pVfs, /* Used to get maximum path length and AppData */ + const char *zName, /* Name of the file (UTF-8) */ + sqlite3_file *id, /* Write the SQLite file handle here */ + int flags, /* Open mode flags */ + int *pOutFlags /* Status return flags */ +){ + HANDLE h; + DWORD lastErrno = 0; + DWORD dwDesiredAccess; + DWORD dwShareMode; + DWORD dwCreationDisposition; + DWORD dwFlagsAndAttributes = 0; +#if SQLITE_OS_WINCE + int isTemp = 0; +#endif + winVfsAppData *pAppData; + winFile *pFile = (winFile*)id; + void *zConverted; /* Filename in OS encoding */ + const char *zUtf8Name = zName; /* Filename in UTF-8 encoding */ + int cnt = 0; + + /* If argument zPath is a NULL pointer, this function is required to open + ** a temporary file. Use this buffer to store the file name in. + */ + char *zTmpname = 0; /* For temporary filename, if necessary. */ + + int rc = SQLITE_OK; /* Function Return Code */ +#if !defined(NDEBUG) || SQLITE_OS_WINCE + int eType = flags&0xFFFFFF00; /* Type of file to open */ +#endif + + int isExclusive = (flags & SQLITE_OPEN_EXCLUSIVE); + int isDelete = (flags & SQLITE_OPEN_DELETEONCLOSE); + int isCreate = (flags & SQLITE_OPEN_CREATE); + int isReadonly = (flags & SQLITE_OPEN_READONLY); + int isReadWrite = (flags & SQLITE_OPEN_READWRITE); + +#ifndef NDEBUG + int isOpenJournal = (isCreate && ( + eType==SQLITE_OPEN_MASTER_JOURNAL + || eType==SQLITE_OPEN_MAIN_JOURNAL + || eType==SQLITE_OPEN_WAL + )); +#endif + + OSTRACE(("OPEN name=%s, pFile=%p, flags=%x, pOutFlags=%p\n", + zUtf8Name, id, flags, pOutFlags)); + + /* Check the following statements are true: + ** + ** (a) Exactly one of the READWRITE and READONLY flags must be set, and + ** (b) if CREATE is set, then READWRITE must also be set, and + ** (c) if EXCLUSIVE is set, then CREATE must also be set. + ** (d) if DELETEONCLOSE is set, then CREATE must also be set. + */ + assert((isReadonly==0 || isReadWrite==0) && (isReadWrite || isReadonly)); + assert(isCreate==0 || isReadWrite); + assert(isExclusive==0 || isCreate); + assert(isDelete==0 || isCreate); + + /* The main DB, main journal, WAL file and master journal are never + ** automatically deleted. Nor are they ever temporary files. */ + assert( (!isDelete && zName) || eType!=SQLITE_OPEN_MAIN_DB ); + assert( (!isDelete && zName) || eType!=SQLITE_OPEN_MAIN_JOURNAL ); + assert( (!isDelete && zName) || eType!=SQLITE_OPEN_MASTER_JOURNAL ); + assert( (!isDelete && zName) || eType!=SQLITE_OPEN_WAL ); + + /* Assert that the upper layer has set one of the "file-type" flags. */ + assert( eType==SQLITE_OPEN_MAIN_DB || eType==SQLITE_OPEN_TEMP_DB + || eType==SQLITE_OPEN_MAIN_JOURNAL || eType==SQLITE_OPEN_TEMP_JOURNAL + || eType==SQLITE_OPEN_SUBJOURNAL || eType==SQLITE_OPEN_MASTER_JOURNAL + || eType==SQLITE_OPEN_TRANSIENT_DB || eType==SQLITE_OPEN_WAL + ); + + assert( pFile!=0 ); + memset(pFile, 0, sizeof(winFile)); + pFile->h = INVALID_HANDLE_VALUE; + +#if SQLITE_OS_WINRT + if( !zUtf8Name && !sqlite3_temp_directory ){ + sqlite3_log(SQLITE_ERROR, + "sqlite3_temp_directory variable should be set for WinRT"); + } +#endif + + /* If the second argument to this function is NULL, generate a + ** temporary file name to use + */ + if( !zUtf8Name ){ + assert( isDelete && !isOpenJournal ); + rc = winGetTempname(pVfs, &zTmpname); + if( rc!=SQLITE_OK ){ + OSTRACE(("OPEN name=%s, rc=%s", zUtf8Name, sqlite3ErrName(rc))); + return rc; + } + zUtf8Name = zTmpname; + } + + /* Database filenames are double-zero terminated if they are not + ** URIs with parameters. Hence, they can always be passed into + ** sqlite3_uri_parameter(). + */ + assert( (eType!=SQLITE_OPEN_MAIN_DB) || (flags & SQLITE_OPEN_URI) || + zUtf8Name[sqlite3Strlen30(zUtf8Name)+1]==0 ); + + /* Convert the filename to the system encoding. */ + zConverted = winConvertFromUtf8Filename(zUtf8Name); + if( zConverted==0 ){ + sqlite3_free(zTmpname); + OSTRACE(("OPEN name=%s, rc=SQLITE_IOERR_NOMEM", zUtf8Name)); + return SQLITE_IOERR_NOMEM_BKPT; + } + + if( winIsDir(zConverted) ){ + sqlite3_free(zConverted); + sqlite3_free(zTmpname); + OSTRACE(("OPEN name=%s, rc=SQLITE_CANTOPEN_ISDIR", zUtf8Name)); + return SQLITE_CANTOPEN_ISDIR; + } + + if( isReadWrite ){ + dwDesiredAccess = GENERIC_READ | GENERIC_WRITE; + }else{ + dwDesiredAccess = GENERIC_READ; + } + + /* SQLITE_OPEN_EXCLUSIVE is used to make sure that a new file is + ** created. SQLite doesn't use it to indicate "exclusive access" + ** as it is usually understood. + */ + if( isExclusive ){ + /* Creates a new file, only if it does not already exist. */ + /* If the file exists, it fails. */ + dwCreationDisposition = CREATE_NEW; + }else if( isCreate ){ + /* Open existing file, or create if it doesn't exist */ + dwCreationDisposition = OPEN_ALWAYS; + }else{ + /* Opens a file, only if it exists. */ + dwCreationDisposition = OPEN_EXISTING; + } + + dwShareMode = FILE_SHARE_READ | FILE_SHARE_WRITE; + + if( isDelete ){ +#if SQLITE_OS_WINCE + dwFlagsAndAttributes = FILE_ATTRIBUTE_HIDDEN; + isTemp = 1; +#else + dwFlagsAndAttributes = FILE_ATTRIBUTE_TEMPORARY + | FILE_ATTRIBUTE_HIDDEN + | FILE_FLAG_DELETE_ON_CLOSE; +#endif + }else{ + dwFlagsAndAttributes = FILE_ATTRIBUTE_NORMAL; + } + /* Reports from the internet are that performance is always + ** better if FILE_FLAG_RANDOM_ACCESS is used. Ticket #2699. */ +#if SQLITE_OS_WINCE + dwFlagsAndAttributes |= FILE_FLAG_RANDOM_ACCESS; +#endif + + if( osIsNT() ){ +#if SQLITE_OS_WINRT + CREATEFILE2_EXTENDED_PARAMETERS extendedParameters; + extendedParameters.dwSize = sizeof(CREATEFILE2_EXTENDED_PARAMETERS); + extendedParameters.dwFileAttributes = + dwFlagsAndAttributes & FILE_ATTRIBUTE_MASK; + extendedParameters.dwFileFlags = dwFlagsAndAttributes & FILE_FLAG_MASK; + extendedParameters.dwSecurityQosFlags = SECURITY_ANONYMOUS; + extendedParameters.lpSecurityAttributes = NULL; + extendedParameters.hTemplateFile = NULL; + do{ + h = osCreateFile2((LPCWSTR)zConverted, + dwDesiredAccess, + dwShareMode, + dwCreationDisposition, + &extendedParameters); + if( h!=INVALID_HANDLE_VALUE ) break; + if( isReadWrite ){ + int rc2, isRO = 0; + sqlite3BeginBenignMalloc(); + rc2 = winAccess(pVfs, zName, SQLITE_ACCESS_READ, &isRO); + sqlite3EndBenignMalloc(); + if( rc2==SQLITE_OK && isRO ) break; + } + }while( winRetryIoerr(&cnt, &lastErrno) ); +#else + do{ + h = osCreateFileW((LPCWSTR)zConverted, + dwDesiredAccess, + dwShareMode, NULL, + dwCreationDisposition, + dwFlagsAndAttributes, + NULL); + if( h!=INVALID_HANDLE_VALUE ) break; + if( isReadWrite ){ + int rc2, isRO = 0; + sqlite3BeginBenignMalloc(); + rc2 = winAccess(pVfs, zName, SQLITE_ACCESS_READ, &isRO); + sqlite3EndBenignMalloc(); + if( rc2==SQLITE_OK && isRO ) break; + } + }while( winRetryIoerr(&cnt, &lastErrno) ); +#endif + } +#ifdef SQLITE_WIN32_HAS_ANSI + else{ + do{ + h = osCreateFileA((LPCSTR)zConverted, + dwDesiredAccess, + dwShareMode, NULL, + dwCreationDisposition, + dwFlagsAndAttributes, + NULL); + if( h!=INVALID_HANDLE_VALUE ) break; + if( isReadWrite ){ + int rc2, isRO = 0; + sqlite3BeginBenignMalloc(); + rc2 = winAccess(pVfs, zName, SQLITE_ACCESS_READ, &isRO); + sqlite3EndBenignMalloc(); + if( rc2==SQLITE_OK && isRO ) break; + } + }while( winRetryIoerr(&cnt, &lastErrno) ); + } +#endif + winLogIoerr(cnt, __LINE__); + + OSTRACE(("OPEN file=%p, name=%s, access=%lx, rc=%s\n", h, zUtf8Name, + dwDesiredAccess, (h==INVALID_HANDLE_VALUE) ? "failed" : "ok")); + + if( h==INVALID_HANDLE_VALUE ){ + sqlite3_free(zConverted); + sqlite3_free(zTmpname); + if( isReadWrite && !isExclusive ){ + return winOpen(pVfs, zName, id, + ((flags|SQLITE_OPEN_READONLY) & + ~(SQLITE_OPEN_CREATE|SQLITE_OPEN_READWRITE)), + pOutFlags); + }else{ + pFile->lastErrno = lastErrno; + winLogError(SQLITE_CANTOPEN, pFile->lastErrno, "winOpen", zUtf8Name); + return SQLITE_CANTOPEN_BKPT; + } + } + + if( pOutFlags ){ + if( isReadWrite ){ + *pOutFlags = SQLITE_OPEN_READWRITE; + }else{ + *pOutFlags = SQLITE_OPEN_READONLY; + } + } + + OSTRACE(("OPEN file=%p, name=%s, access=%lx, pOutFlags=%p, *pOutFlags=%d, " + "rc=%s\n", h, zUtf8Name, dwDesiredAccess, pOutFlags, pOutFlags ? + *pOutFlags : 0, (h==INVALID_HANDLE_VALUE) ? "failed" : "ok")); + + pAppData = (winVfsAppData*)pVfs->pAppData; + +#if SQLITE_OS_WINCE + { + if( isReadWrite && eType==SQLITE_OPEN_MAIN_DB + && ((pAppData==NULL) || !pAppData->bNoLock) + && (rc = winceCreateLock(zName, pFile))!=SQLITE_OK + ){ + osCloseHandle(h); + sqlite3_free(zConverted); + sqlite3_free(zTmpname); + OSTRACE(("OPEN-CE-LOCK name=%s, rc=%s\n", zName, sqlite3ErrName(rc))); + return rc; + } + } + if( isTemp ){ + pFile->zDeleteOnClose = zConverted; + }else +#endif + { + sqlite3_free(zConverted); + } + + sqlite3_free(zTmpname); + pFile->pMethod = pAppData ? pAppData->pMethod : &winIoMethod; + pFile->pVfs = pVfs; + pFile->h = h; + if( isReadonly ){ + pFile->ctrlFlags |= WINFILE_RDONLY; + } + if( sqlite3_uri_boolean(zName, "psow", SQLITE_POWERSAFE_OVERWRITE) ){ + pFile->ctrlFlags |= WINFILE_PSOW; + } + pFile->lastErrno = NO_ERROR; + pFile->zPath = zName; +#if SQLITE_MAX_MMAP_SIZE>0 + pFile->hMap = NULL; + pFile->pMapRegion = 0; + pFile->mmapSize = 0; + pFile->mmapSizeMax = sqlite3GlobalConfig.szMmap; +#endif + + OpenCounter(+1); + return rc; +} + +/* +** Delete the named file. +** +** Note that Windows does not allow a file to be deleted if some other +** process has it open. Sometimes a virus scanner or indexing program +** will open a journal file shortly after it is created in order to do +** whatever it does. While this other process is holding the +** file open, we will be unable to delete it. To work around this +** problem, we delay 100 milliseconds and try to delete again. Up +** to MX_DELETION_ATTEMPTs deletion attempts are run before giving +** up and returning an error. +*/ +static int winDelete( + sqlite3_vfs *pVfs, /* Not used on win32 */ + const char *zFilename, /* Name of file to delete */ + int syncDir /* Not used on win32 */ +){ + int cnt = 0; + int rc; + DWORD attr; + DWORD lastErrno = 0; + void *zConverted; + UNUSED_PARAMETER(pVfs); + UNUSED_PARAMETER(syncDir); + + SimulateIOError(return SQLITE_IOERR_DELETE); + OSTRACE(("DELETE name=%s, syncDir=%d\n", zFilename, syncDir)); + + zConverted = winConvertFromUtf8Filename(zFilename); + if( zConverted==0 ){ + OSTRACE(("DELETE name=%s, rc=SQLITE_IOERR_NOMEM\n", zFilename)); + return SQLITE_IOERR_NOMEM_BKPT; + } + if( osIsNT() ){ + do { +#if SQLITE_OS_WINRT + WIN32_FILE_ATTRIBUTE_DATA sAttrData; + memset(&sAttrData, 0, sizeof(sAttrData)); + if ( osGetFileAttributesExW(zConverted, GetFileExInfoStandard, + &sAttrData) ){ + attr = sAttrData.dwFileAttributes; + }else{ + lastErrno = osGetLastError(); + if( lastErrno==ERROR_FILE_NOT_FOUND + || lastErrno==ERROR_PATH_NOT_FOUND ){ + rc = SQLITE_IOERR_DELETE_NOENT; /* Already gone? */ + }else{ + rc = SQLITE_ERROR; + } + break; + } +#else + attr = osGetFileAttributesW(zConverted); +#endif + if ( attr==INVALID_FILE_ATTRIBUTES ){ + lastErrno = osGetLastError(); + if( lastErrno==ERROR_FILE_NOT_FOUND + || lastErrno==ERROR_PATH_NOT_FOUND ){ + rc = SQLITE_IOERR_DELETE_NOENT; /* Already gone? */ + }else{ + rc = SQLITE_ERROR; + } + break; + } + if ( attr&FILE_ATTRIBUTE_DIRECTORY ){ + rc = SQLITE_ERROR; /* Files only. */ + break; + } + if ( osDeleteFileW(zConverted) ){ + rc = SQLITE_OK; /* Deleted OK. */ + break; + } + if ( !winRetryIoerr(&cnt, &lastErrno) ){ + rc = SQLITE_ERROR; /* No more retries. */ + break; + } + } while(1); + } +#ifdef SQLITE_WIN32_HAS_ANSI + else{ + do { + attr = osGetFileAttributesA(zConverted); + if ( attr==INVALID_FILE_ATTRIBUTES ){ + lastErrno = osGetLastError(); + if( lastErrno==ERROR_FILE_NOT_FOUND + || lastErrno==ERROR_PATH_NOT_FOUND ){ + rc = SQLITE_IOERR_DELETE_NOENT; /* Already gone? */ + }else{ + rc = SQLITE_ERROR; + } + break; + } + if ( attr&FILE_ATTRIBUTE_DIRECTORY ){ + rc = SQLITE_ERROR; /* Files only. */ + break; + } + if ( osDeleteFileA(zConverted) ){ + rc = SQLITE_OK; /* Deleted OK. */ + break; + } + if ( !winRetryIoerr(&cnt, &lastErrno) ){ + rc = SQLITE_ERROR; /* No more retries. */ + break; + } + } while(1); + } +#endif + if( rc && rc!=SQLITE_IOERR_DELETE_NOENT ){ + rc = winLogError(SQLITE_IOERR_DELETE, lastErrno, "winDelete", zFilename); + }else{ + winLogIoerr(cnt, __LINE__); + } + sqlite3_free(zConverted); + OSTRACE(("DELETE name=%s, rc=%s\n", zFilename, sqlite3ErrName(rc))); + return rc; +} + +/* +** Check the existence and status of a file. +*/ +static int winAccess( + sqlite3_vfs *pVfs, /* Not used on win32 */ + const char *zFilename, /* Name of file to check */ + int flags, /* Type of test to make on this file */ + int *pResOut /* OUT: Result */ +){ + DWORD attr; + int rc = 0; + DWORD lastErrno = 0; + void *zConverted; + UNUSED_PARAMETER(pVfs); + + SimulateIOError( return SQLITE_IOERR_ACCESS; ); + OSTRACE(("ACCESS name=%s, flags=%x, pResOut=%p\n", + zFilename, flags, pResOut)); + + zConverted = winConvertFromUtf8Filename(zFilename); + if( zConverted==0 ){ + OSTRACE(("ACCESS name=%s, rc=SQLITE_IOERR_NOMEM\n", zFilename)); + return SQLITE_IOERR_NOMEM_BKPT; + } + if( osIsNT() ){ + int cnt = 0; + WIN32_FILE_ATTRIBUTE_DATA sAttrData; + memset(&sAttrData, 0, sizeof(sAttrData)); + while( !(rc = osGetFileAttributesExW((LPCWSTR)zConverted, + GetFileExInfoStandard, + &sAttrData)) && winRetryIoerr(&cnt, &lastErrno) ){} + if( rc ){ + /* For an SQLITE_ACCESS_EXISTS query, treat a zero-length file + ** as if it does not exist. + */ + if( flags==SQLITE_ACCESS_EXISTS + && sAttrData.nFileSizeHigh==0 + && sAttrData.nFileSizeLow==0 ){ + attr = INVALID_FILE_ATTRIBUTES; + }else{ + attr = sAttrData.dwFileAttributes; + } + }else{ + winLogIoerr(cnt, __LINE__); + if( lastErrno!=ERROR_FILE_NOT_FOUND && lastErrno!=ERROR_PATH_NOT_FOUND ){ + sqlite3_free(zConverted); + return winLogError(SQLITE_IOERR_ACCESS, lastErrno, "winAccess", + zFilename); + }else{ + attr = INVALID_FILE_ATTRIBUTES; + } + } + } +#ifdef SQLITE_WIN32_HAS_ANSI + else{ + attr = osGetFileAttributesA((char*)zConverted); + } +#endif + sqlite3_free(zConverted); + switch( flags ){ + case SQLITE_ACCESS_READ: + case SQLITE_ACCESS_EXISTS: + rc = attr!=INVALID_FILE_ATTRIBUTES; + break; + case SQLITE_ACCESS_READWRITE: + rc = attr!=INVALID_FILE_ATTRIBUTES && + (attr & FILE_ATTRIBUTE_READONLY)==0; + break; + default: + assert(!"Invalid flags argument"); + } + *pResOut = rc; + OSTRACE(("ACCESS name=%s, pResOut=%p, *pResOut=%d, rc=SQLITE_OK\n", + zFilename, pResOut, *pResOut)); + return SQLITE_OK; +} + +/* +** Returns non-zero if the specified path name starts with a drive letter +** followed by a colon character. +*/ +static BOOL winIsDriveLetterAndColon( + const char *zPathname +){ + return ( sqlite3Isalpha(zPathname[0]) && zPathname[1]==':' ); +} + +/* +** Returns non-zero if the specified path name should be used verbatim. If +** non-zero is returned from this function, the calling function must simply +** use the provided path name verbatim -OR- resolve it into a full path name +** using the GetFullPathName Win32 API function (if available). +*/ +static BOOL winIsVerbatimPathname( + const char *zPathname +){ + /* + ** If the path name starts with a forward slash or a backslash, it is either + ** a legal UNC name, a volume relative path, or an absolute path name in the + ** "Unix" format on Windows. There is no easy way to differentiate between + ** the final two cases; therefore, we return the safer return value of TRUE + ** so that callers of this function will simply use it verbatim. + */ + if ( winIsDirSep(zPathname[0]) ){ + return TRUE; + } + + /* + ** If the path name starts with a letter and a colon it is either a volume + ** relative path or an absolute path. Callers of this function must not + ** attempt to treat it as a relative path name (i.e. they should simply use + ** it verbatim). + */ + if ( winIsDriveLetterAndColon(zPathname) ){ + return TRUE; + } + + /* + ** If we get to this point, the path name should almost certainly be a purely + ** relative one (i.e. not a UNC name, not absolute, and not volume relative). + */ + return FALSE; +} + +/* +** Turn a relative pathname into a full pathname. Write the full +** pathname into zOut[]. zOut[] will be at least pVfs->mxPathname +** bytes in size. +*/ +static int winFullPathname( + sqlite3_vfs *pVfs, /* Pointer to vfs object */ + const char *zRelative, /* Possibly relative input path */ + int nFull, /* Size of output buffer in bytes */ + char *zFull /* Output buffer */ +){ +#if !SQLITE_OS_WINCE && !SQLITE_OS_WINRT && !defined(__CYGWIN__) + DWORD nByte; + void *zConverted; + char *zOut; +#endif + + /* If this path name begins with "/X:", where "X" is any alphabetic + ** character, discard the initial "/" from the pathname. + */ + if( zRelative[0]=='/' && winIsDriveLetterAndColon(zRelative+1) ){ + zRelative++; + } + +#if defined(__CYGWIN__) + SimulateIOError( return SQLITE_ERROR ); + UNUSED_PARAMETER(nFull); + assert( nFull>=pVfs->mxPathname ); + if ( sqlite3_data_directory && !winIsVerbatimPathname(zRelative) ){ + /* + ** NOTE: We are dealing with a relative path name and the data + ** directory has been set. Therefore, use it as the basis + ** for converting the relative path name to an absolute + ** one by prepending the data directory and a slash. + */ + char *zOut = sqlite3MallocZero( pVfs->mxPathname+1 ); + if( !zOut ){ + return SQLITE_IOERR_NOMEM_BKPT; + } + if( cygwin_conv_path( + (osIsNT() ? CCP_POSIX_TO_WIN_W : CCP_POSIX_TO_WIN_A) | + CCP_RELATIVE, zRelative, zOut, pVfs->mxPathname+1)<0 ){ + sqlite3_free(zOut); + return winLogError(SQLITE_CANTOPEN_CONVPATH, (DWORD)errno, + "winFullPathname1", zRelative); + }else{ + char *zUtf8 = winConvertToUtf8Filename(zOut); + if( !zUtf8 ){ + sqlite3_free(zOut); + return SQLITE_IOERR_NOMEM_BKPT; + } + sqlite3_snprintf(MIN(nFull, pVfs->mxPathname), zFull, "%s%c%s", + sqlite3_data_directory, winGetDirSep(), zUtf8); + sqlite3_free(zUtf8); + sqlite3_free(zOut); + } + }else{ + char *zOut = sqlite3MallocZero( pVfs->mxPathname+1 ); + if( !zOut ){ + return SQLITE_IOERR_NOMEM_BKPT; + } + if( cygwin_conv_path( + (osIsNT() ? CCP_POSIX_TO_WIN_W : CCP_POSIX_TO_WIN_A), + zRelative, zOut, pVfs->mxPathname+1)<0 ){ + sqlite3_free(zOut); + return winLogError(SQLITE_CANTOPEN_CONVPATH, (DWORD)errno, + "winFullPathname2", zRelative); + }else{ + char *zUtf8 = winConvertToUtf8Filename(zOut); + if( !zUtf8 ){ + sqlite3_free(zOut); + return SQLITE_IOERR_NOMEM_BKPT; + } + sqlite3_snprintf(MIN(nFull, pVfs->mxPathname), zFull, "%s", zUtf8); + sqlite3_free(zUtf8); + sqlite3_free(zOut); + } + } + return SQLITE_OK; +#endif + +#if (SQLITE_OS_WINCE || SQLITE_OS_WINRT) && !defined(__CYGWIN__) + SimulateIOError( return SQLITE_ERROR ); + /* WinCE has no concept of a relative pathname, or so I am told. */ + /* WinRT has no way to convert a relative path to an absolute one. */ + if ( sqlite3_data_directory && !winIsVerbatimPathname(zRelative) ){ + /* + ** NOTE: We are dealing with a relative path name and the data + ** directory has been set. Therefore, use it as the basis + ** for converting the relative path name to an absolute + ** one by prepending the data directory and a backslash. + */ + sqlite3_snprintf(MIN(nFull, pVfs->mxPathname), zFull, "%s%c%s", + sqlite3_data_directory, winGetDirSep(), zRelative); + }else{ + sqlite3_snprintf(MIN(nFull, pVfs->mxPathname), zFull, "%s", zRelative); + } + return SQLITE_OK; +#endif + +#if !SQLITE_OS_WINCE && !SQLITE_OS_WINRT && !defined(__CYGWIN__) + /* It's odd to simulate an io-error here, but really this is just + ** using the io-error infrastructure to test that SQLite handles this + ** function failing. This function could fail if, for example, the + ** current working directory has been unlinked. + */ + SimulateIOError( return SQLITE_ERROR ); + if ( sqlite3_data_directory && !winIsVerbatimPathname(zRelative) ){ + /* + ** NOTE: We are dealing with a relative path name and the data + ** directory has been set. Therefore, use it as the basis + ** for converting the relative path name to an absolute + ** one by prepending the data directory and a backslash. + */ + sqlite3_snprintf(MIN(nFull, pVfs->mxPathname), zFull, "%s%c%s", + sqlite3_data_directory, winGetDirSep(), zRelative); + return SQLITE_OK; + } + zConverted = winConvertFromUtf8Filename(zRelative); + if( zConverted==0 ){ + return SQLITE_IOERR_NOMEM_BKPT; + } + if( osIsNT() ){ + LPWSTR zTemp; + nByte = osGetFullPathNameW((LPCWSTR)zConverted, 0, 0, 0); + if( nByte==0 ){ + sqlite3_free(zConverted); + return winLogError(SQLITE_CANTOPEN_FULLPATH, osGetLastError(), + "winFullPathname1", zRelative); + } + nByte += 3; + zTemp = sqlite3MallocZero( nByte*sizeof(zTemp[0]) ); + if( zTemp==0 ){ + sqlite3_free(zConverted); + return SQLITE_IOERR_NOMEM_BKPT; + } + nByte = osGetFullPathNameW((LPCWSTR)zConverted, nByte, zTemp, 0); + if( nByte==0 ){ + sqlite3_free(zConverted); + sqlite3_free(zTemp); + return winLogError(SQLITE_CANTOPEN_FULLPATH, osGetLastError(), + "winFullPathname2", zRelative); + } + sqlite3_free(zConverted); + zOut = winUnicodeToUtf8(zTemp); + sqlite3_free(zTemp); + } +#ifdef SQLITE_WIN32_HAS_ANSI + else{ + char *zTemp; + nByte = osGetFullPathNameA((char*)zConverted, 0, 0, 0); + if( nByte==0 ){ + sqlite3_free(zConverted); + return winLogError(SQLITE_CANTOPEN_FULLPATH, osGetLastError(), + "winFullPathname3", zRelative); + } + nByte += 3; + zTemp = sqlite3MallocZero( nByte*sizeof(zTemp[0]) ); + if( zTemp==0 ){ + sqlite3_free(zConverted); + return SQLITE_IOERR_NOMEM_BKPT; + } + nByte = osGetFullPathNameA((char*)zConverted, nByte, zTemp, 0); + if( nByte==0 ){ + sqlite3_free(zConverted); + sqlite3_free(zTemp); + return winLogError(SQLITE_CANTOPEN_FULLPATH, osGetLastError(), + "winFullPathname4", zRelative); + } + sqlite3_free(zConverted); + zOut = winMbcsToUtf8(zTemp, osAreFileApisANSI()); + sqlite3_free(zTemp); + } +#endif + if( zOut ){ + sqlite3_snprintf(MIN(nFull, pVfs->mxPathname), zFull, "%s", zOut); + sqlite3_free(zOut); + return SQLITE_OK; + }else{ + return SQLITE_IOERR_NOMEM_BKPT; + } +#endif +} + +#ifndef SQLITE_OMIT_LOAD_EXTENSION +/* +** Interfaces for opening a shared library, finding entry points +** within the shared library, and closing the shared library. +*/ +static void *winDlOpen(sqlite3_vfs *pVfs, const char *zFilename){ + HANDLE h; +#if defined(__CYGWIN__) + int nFull = pVfs->mxPathname+1; + char *zFull = sqlite3MallocZero( nFull ); + void *zConverted = 0; + if( zFull==0 ){ + OSTRACE(("DLOPEN name=%s, handle=%p\n", zFilename, (void*)0)); + return 0; + } + if( winFullPathname(pVfs, zFilename, nFull, zFull)!=SQLITE_OK ){ + sqlite3_free(zFull); + OSTRACE(("DLOPEN name=%s, handle=%p\n", zFilename, (void*)0)); + return 0; + } + zConverted = winConvertFromUtf8Filename(zFull); + sqlite3_free(zFull); +#else + void *zConverted = winConvertFromUtf8Filename(zFilename); + UNUSED_PARAMETER(pVfs); +#endif + if( zConverted==0 ){ + OSTRACE(("DLOPEN name=%s, handle=%p\n", zFilename, (void*)0)); + return 0; + } + if( osIsNT() ){ +#if SQLITE_OS_WINRT + h = osLoadPackagedLibrary((LPCWSTR)zConverted, 0); +#else + h = osLoadLibraryW((LPCWSTR)zConverted); +#endif + } +#ifdef SQLITE_WIN32_HAS_ANSI + else{ + h = osLoadLibraryA((char*)zConverted); + } +#endif + OSTRACE(("DLOPEN name=%s, handle=%p\n", zFilename, (void*)h)); + sqlite3_free(zConverted); + return (void*)h; +} +static void winDlError(sqlite3_vfs *pVfs, int nBuf, char *zBufOut){ + UNUSED_PARAMETER(pVfs); + winGetLastErrorMsg(osGetLastError(), nBuf, zBufOut); +} +static void (*winDlSym(sqlite3_vfs *pVfs,void *pH,const char *zSym))(void){ + FARPROC proc; + UNUSED_PARAMETER(pVfs); + proc = osGetProcAddressA((HANDLE)pH, zSym); + OSTRACE(("DLSYM handle=%p, symbol=%s, address=%p\n", + (void*)pH, zSym, (void*)proc)); + return (void(*)(void))proc; +} +static void winDlClose(sqlite3_vfs *pVfs, void *pHandle){ + UNUSED_PARAMETER(pVfs); + osFreeLibrary((HANDLE)pHandle); + OSTRACE(("DLCLOSE handle=%p\n", (void*)pHandle)); +} +#else /* if SQLITE_OMIT_LOAD_EXTENSION is defined: */ + #define winDlOpen 0 + #define winDlError 0 + #define winDlSym 0 + #define winDlClose 0 +#endif + +/* State information for the randomness gatherer. */ +typedef struct EntropyGatherer EntropyGatherer; +struct EntropyGatherer { + unsigned char *a; /* Gather entropy into this buffer */ + int na; /* Size of a[] in bytes */ + int i; /* XOR next input into a[i] */ + int nXor; /* Number of XOR operations done */ +}; + +#if !defined(SQLITE_TEST) && !defined(SQLITE_OMIT_RANDOMNESS) +/* Mix sz bytes of entropy into p. */ +static void xorMemory(EntropyGatherer *p, unsigned char *x, int sz){ + int j, k; + for(j=0, k=p->i; ja[k++] ^= x[j]; + if( k>=p->na ) k = 0; + } + p->i = k; + p->nXor += sz; +} +#endif /* !defined(SQLITE_TEST) && !defined(SQLITE_OMIT_RANDOMNESS) */ + +/* +** Write up to nBuf bytes of randomness into zBuf. +*/ +static int winRandomness(sqlite3_vfs *pVfs, int nBuf, char *zBuf){ +#if defined(SQLITE_TEST) || defined(SQLITE_OMIT_RANDOMNESS) + UNUSED_PARAMETER(pVfs); + memset(zBuf, 0, nBuf); + return nBuf; +#else + EntropyGatherer e; + UNUSED_PARAMETER(pVfs); + memset(zBuf, 0, nBuf); + e.a = (unsigned char*)zBuf; + e.na = nBuf; + e.nXor = 0; + e.i = 0; + { + SYSTEMTIME x; + osGetSystemTime(&x); + xorMemory(&e, (unsigned char*)&x, sizeof(SYSTEMTIME)); + } + { + DWORD pid = osGetCurrentProcessId(); + xorMemory(&e, (unsigned char*)&pid, sizeof(DWORD)); + } +#if SQLITE_OS_WINRT + { + ULONGLONG cnt = osGetTickCount64(); + xorMemory(&e, (unsigned char*)&cnt, sizeof(ULONGLONG)); + } +#else + { + DWORD cnt = osGetTickCount(); + xorMemory(&e, (unsigned char*)&cnt, sizeof(DWORD)); + } +#endif /* SQLITE_OS_WINRT */ + { + LARGE_INTEGER i; + osQueryPerformanceCounter(&i); + xorMemory(&e, (unsigned char*)&i, sizeof(LARGE_INTEGER)); + } +#if !SQLITE_OS_WINCE && !SQLITE_OS_WINRT && SQLITE_WIN32_USE_UUID + { + UUID id; + memset(&id, 0, sizeof(UUID)); + osUuidCreate(&id); + xorMemory(&e, (unsigned char*)&id, sizeof(UUID)); + memset(&id, 0, sizeof(UUID)); + osUuidCreateSequential(&id); + xorMemory(&e, (unsigned char*)&id, sizeof(UUID)); + } +#endif /* !SQLITE_OS_WINCE && !SQLITE_OS_WINRT && SQLITE_WIN32_USE_UUID */ + return e.nXor>nBuf ? nBuf : e.nXor; +#endif /* defined(SQLITE_TEST) || defined(SQLITE_OMIT_RANDOMNESS) */ +} + + +/* +** Sleep for a little while. Return the amount of time slept. +*/ +static int winSleep(sqlite3_vfs *pVfs, int microsec){ + sqlite3_win32_sleep((microsec+999)/1000); + UNUSED_PARAMETER(pVfs); + return ((microsec+999)/1000)*1000; +} + +/* +** The following variable, if set to a non-zero value, is interpreted as +** the number of seconds since 1970 and is used to set the result of +** sqlite3OsCurrentTime() during testing. +*/ +#ifdef SQLITE_TEST +SQLITE_API int sqlite3_current_time = 0; /* Fake system time in seconds since 1970. */ +#endif + +/* +** Find the current time (in Universal Coordinated Time). Write into *piNow +** the current time and date as a Julian Day number times 86_400_000. In +** other words, write into *piNow the number of milliseconds since the Julian +** epoch of noon in Greenwich on November 24, 4714 B.C according to the +** proleptic Gregorian calendar. +** +** On success, return SQLITE_OK. Return SQLITE_ERROR if the time and date +** cannot be found. +*/ +static int winCurrentTimeInt64(sqlite3_vfs *pVfs, sqlite3_int64 *piNow){ + /* FILETIME structure is a 64-bit value representing the number of + 100-nanosecond intervals since January 1, 1601 (= JD 2305813.5). + */ + FILETIME ft; + static const sqlite3_int64 winFiletimeEpoch = 23058135*(sqlite3_int64)8640000; +#ifdef SQLITE_TEST + static const sqlite3_int64 unixEpoch = 24405875*(sqlite3_int64)8640000; +#endif + /* 2^32 - to avoid use of LL and warnings in gcc */ + static const sqlite3_int64 max32BitValue = + (sqlite3_int64)2000000000 + (sqlite3_int64)2000000000 + + (sqlite3_int64)294967296; + +#if SQLITE_OS_WINCE + SYSTEMTIME time; + osGetSystemTime(&time); + /* if SystemTimeToFileTime() fails, it returns zero. */ + if (!osSystemTimeToFileTime(&time,&ft)){ + return SQLITE_ERROR; + } +#else + osGetSystemTimeAsFileTime( &ft ); +#endif + + *piNow = winFiletimeEpoch + + ((((sqlite3_int64)ft.dwHighDateTime)*max32BitValue) + + (sqlite3_int64)ft.dwLowDateTime)/(sqlite3_int64)10000; + +#ifdef SQLITE_TEST + if( sqlite3_current_time ){ + *piNow = 1000*(sqlite3_int64)sqlite3_current_time + unixEpoch; + } +#endif + UNUSED_PARAMETER(pVfs); + return SQLITE_OK; +} + +/* +** Find the current time (in Universal Coordinated Time). Write the +** current time and date as a Julian Day number into *prNow and +** return 0. Return 1 if the time and date cannot be found. +*/ +static int winCurrentTime(sqlite3_vfs *pVfs, double *prNow){ + int rc; + sqlite3_int64 i; + rc = winCurrentTimeInt64(pVfs, &i); + if( !rc ){ + *prNow = i/86400000.0; + } + return rc; +} + +/* +** The idea is that this function works like a combination of +** GetLastError() and FormatMessage() on Windows (or errno and +** strerror_r() on Unix). After an error is returned by an OS +** function, SQLite calls this function with zBuf pointing to +** a buffer of nBuf bytes. The OS layer should populate the +** buffer with a nul-terminated UTF-8 encoded error message +** describing the last IO error to have occurred within the calling +** thread. +** +** If the error message is too large for the supplied buffer, +** it should be truncated. The return value of xGetLastError +** is zero if the error message fits in the buffer, or non-zero +** otherwise (if the message was truncated). If non-zero is returned, +** then it is not necessary to include the nul-terminator character +** in the output buffer. +** +** Not supplying an error message will have no adverse effect +** on SQLite. It is fine to have an implementation that never +** returns an error message: +** +** int xGetLastError(sqlite3_vfs *pVfs, int nBuf, char *zBuf){ +** assert(zBuf[0]=='\0'); +** return 0; +** } +** +** However if an error message is supplied, it will be incorporated +** by sqlite into the error message available to the user using +** sqlite3_errmsg(), possibly making IO errors easier to debug. +*/ +static int winGetLastError(sqlite3_vfs *pVfs, int nBuf, char *zBuf){ + DWORD e = osGetLastError(); + UNUSED_PARAMETER(pVfs); + if( nBuf>0 ) winGetLastErrorMsg(e, nBuf, zBuf); + return e; +} + +/* +** Initialize and deinitialize the operating system interface. +*/ +SQLITE_API int sqlite3_os_init(void){ + static sqlite3_vfs winVfs = { + 3, /* iVersion */ + sizeof(winFile), /* szOsFile */ + SQLITE_WIN32_MAX_PATH_BYTES, /* mxPathname */ + 0, /* pNext */ + "win32", /* zName */ + &winAppData, /* pAppData */ + winOpen, /* xOpen */ + winDelete, /* xDelete */ + winAccess, /* xAccess */ + winFullPathname, /* xFullPathname */ + winDlOpen, /* xDlOpen */ + winDlError, /* xDlError */ + winDlSym, /* xDlSym */ + winDlClose, /* xDlClose */ + winRandomness, /* xRandomness */ + winSleep, /* xSleep */ + winCurrentTime, /* xCurrentTime */ + winGetLastError, /* xGetLastError */ + winCurrentTimeInt64, /* xCurrentTimeInt64 */ + winSetSystemCall, /* xSetSystemCall */ + winGetSystemCall, /* xGetSystemCall */ + winNextSystemCall, /* xNextSystemCall */ + }; +#if defined(SQLITE_WIN32_HAS_WIDE) + static sqlite3_vfs winLongPathVfs = { + 3, /* iVersion */ + sizeof(winFile), /* szOsFile */ + SQLITE_WINNT_MAX_PATH_BYTES, /* mxPathname */ + 0, /* pNext */ + "win32-longpath", /* zName */ + &winAppData, /* pAppData */ + winOpen, /* xOpen */ + winDelete, /* xDelete */ + winAccess, /* xAccess */ + winFullPathname, /* xFullPathname */ + winDlOpen, /* xDlOpen */ + winDlError, /* xDlError */ + winDlSym, /* xDlSym */ + winDlClose, /* xDlClose */ + winRandomness, /* xRandomness */ + winSleep, /* xSleep */ + winCurrentTime, /* xCurrentTime */ + winGetLastError, /* xGetLastError */ + winCurrentTimeInt64, /* xCurrentTimeInt64 */ + winSetSystemCall, /* xSetSystemCall */ + winGetSystemCall, /* xGetSystemCall */ + winNextSystemCall, /* xNextSystemCall */ + }; +#endif + static sqlite3_vfs winNolockVfs = { + 3, /* iVersion */ + sizeof(winFile), /* szOsFile */ + SQLITE_WIN32_MAX_PATH_BYTES, /* mxPathname */ + 0, /* pNext */ + "win32-none", /* zName */ + &winNolockAppData, /* pAppData */ + winOpen, /* xOpen */ + winDelete, /* xDelete */ + winAccess, /* xAccess */ + winFullPathname, /* xFullPathname */ + winDlOpen, /* xDlOpen */ + winDlError, /* xDlError */ + winDlSym, /* xDlSym */ + winDlClose, /* xDlClose */ + winRandomness, /* xRandomness */ + winSleep, /* xSleep */ + winCurrentTime, /* xCurrentTime */ + winGetLastError, /* xGetLastError */ + winCurrentTimeInt64, /* xCurrentTimeInt64 */ + winSetSystemCall, /* xSetSystemCall */ + winGetSystemCall, /* xGetSystemCall */ + winNextSystemCall, /* xNextSystemCall */ + }; +#if defined(SQLITE_WIN32_HAS_WIDE) + static sqlite3_vfs winLongPathNolockVfs = { + 3, /* iVersion */ + sizeof(winFile), /* szOsFile */ + SQLITE_WINNT_MAX_PATH_BYTES, /* mxPathname */ + 0, /* pNext */ + "win32-longpath-none", /* zName */ + &winNolockAppData, /* pAppData */ + winOpen, /* xOpen */ + winDelete, /* xDelete */ + winAccess, /* xAccess */ + winFullPathname, /* xFullPathname */ + winDlOpen, /* xDlOpen */ + winDlError, /* xDlError */ + winDlSym, /* xDlSym */ + winDlClose, /* xDlClose */ + winRandomness, /* xRandomness */ + winSleep, /* xSleep */ + winCurrentTime, /* xCurrentTime */ + winGetLastError, /* xGetLastError */ + winCurrentTimeInt64, /* xCurrentTimeInt64 */ + winSetSystemCall, /* xSetSystemCall */ + winGetSystemCall, /* xGetSystemCall */ + winNextSystemCall, /* xNextSystemCall */ + }; +#endif + + /* Double-check that the aSyscall[] array has been constructed + ** correctly. See ticket [bb3a86e890c8e96ab] */ + assert( ArraySize(aSyscall)==80 ); + + /* get memory map allocation granularity */ + memset(&winSysInfo, 0, sizeof(SYSTEM_INFO)); +#if SQLITE_OS_WINRT + osGetNativeSystemInfo(&winSysInfo); +#else + osGetSystemInfo(&winSysInfo); +#endif + assert( winSysInfo.dwAllocationGranularity>0 ); + assert( winSysInfo.dwPageSize>0 ); + + sqlite3_vfs_register(&winVfs, 1); + +#if defined(SQLITE_WIN32_HAS_WIDE) + sqlite3_vfs_register(&winLongPathVfs, 0); +#endif + + sqlite3_vfs_register(&winNolockVfs, 0); + +#if defined(SQLITE_WIN32_HAS_WIDE) + sqlite3_vfs_register(&winLongPathNolockVfs, 0); +#endif + +#ifndef SQLITE_OMIT_WAL + winBigLock = sqlite3MutexAlloc(SQLITE_MUTEX_STATIC_VFS1); +#endif + + return SQLITE_OK; +} + +SQLITE_API int sqlite3_os_end(void){ +#if SQLITE_OS_WINRT + if( sleepObj!=NULL ){ + osCloseHandle(sleepObj); + sleepObj = NULL; + } +#endif + +#ifndef SQLITE_OMIT_WAL + winBigLock = 0; +#endif + + return SQLITE_OK; +} + +#endif /* SQLITE_OS_WIN */ + +/************** End of os_win.c **********************************************/ +/************** Begin file memdb.c *******************************************/ +/* +** 2016-09-07 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +****************************************************************************** +** +** This file implements an in-memory VFS. A database is held as a contiguous +** block of memory. +** +** This file also implements interface sqlite3_serialize() and +** sqlite3_deserialize(). +*/ +/* #include "sqliteInt.h" */ +#ifdef SQLITE_ENABLE_DESERIALIZE + +/* +** Forward declaration of objects used by this utility +*/ +typedef struct sqlite3_vfs MemVfs; +typedef struct MemFile MemFile; + +/* Access to a lower-level VFS that (might) implement dynamic loading, +** access to randomness, etc. +*/ +#define ORIGVFS(p) ((sqlite3_vfs*)((p)->pAppData)) + +/* An open file */ +struct MemFile { + sqlite3_file base; /* IO methods */ + sqlite3_int64 sz; /* Size of the file */ + sqlite3_int64 szAlloc; /* Space allocated to aData */ + sqlite3_int64 szMax; /* Maximum allowed size of the file */ + unsigned char *aData; /* content of the file */ + int nMmap; /* Number of memory mapped pages */ + unsigned mFlags; /* Flags */ + int eLock; /* Most recent lock against this file */ +}; + +/* +** Methods for MemFile +*/ +static int memdbClose(sqlite3_file*); +static int memdbRead(sqlite3_file*, void*, int iAmt, sqlite3_int64 iOfst); +static int memdbWrite(sqlite3_file*,const void*,int iAmt, sqlite3_int64 iOfst); +static int memdbTruncate(sqlite3_file*, sqlite3_int64 size); +static int memdbSync(sqlite3_file*, int flags); +static int memdbFileSize(sqlite3_file*, sqlite3_int64 *pSize); +static int memdbLock(sqlite3_file*, int); +/* static int memdbCheckReservedLock(sqlite3_file*, int *pResOut);// not used */ +static int memdbFileControl(sqlite3_file*, int op, void *pArg); +/* static int memdbSectorSize(sqlite3_file*); // not used */ +static int memdbDeviceCharacteristics(sqlite3_file*); +static int memdbFetch(sqlite3_file*, sqlite3_int64 iOfst, int iAmt, void **pp); +static int memdbUnfetch(sqlite3_file*, sqlite3_int64 iOfst, void *p); + +/* +** Methods for MemVfs +*/ +static int memdbOpen(sqlite3_vfs*, const char *, sqlite3_file*, int , int *); +/* static int memdbDelete(sqlite3_vfs*, const char *zName, int syncDir); */ +static int memdbAccess(sqlite3_vfs*, const char *zName, int flags, int *); +static int memdbFullPathname(sqlite3_vfs*, const char *zName, int, char *zOut); +static void *memdbDlOpen(sqlite3_vfs*, const char *zFilename); +static void memdbDlError(sqlite3_vfs*, int nByte, char *zErrMsg); +static void (*memdbDlSym(sqlite3_vfs *pVfs, void *p, const char*zSym))(void); +static void memdbDlClose(sqlite3_vfs*, void*); +static int memdbRandomness(sqlite3_vfs*, int nByte, char *zOut); +static int memdbSleep(sqlite3_vfs*, int microseconds); +/* static int memdbCurrentTime(sqlite3_vfs*, double*); */ +static int memdbGetLastError(sqlite3_vfs*, int, char *); +static int memdbCurrentTimeInt64(sqlite3_vfs*, sqlite3_int64*); + +static sqlite3_vfs memdb_vfs = { + 2, /* iVersion */ + 0, /* szOsFile (set when registered) */ + 1024, /* mxPathname */ + 0, /* pNext */ + "memdb", /* zName */ + 0, /* pAppData (set when registered) */ + memdbOpen, /* xOpen */ + 0, /* memdbDelete, */ /* xDelete */ + memdbAccess, /* xAccess */ + memdbFullPathname, /* xFullPathname */ + memdbDlOpen, /* xDlOpen */ + memdbDlError, /* xDlError */ + memdbDlSym, /* xDlSym */ + memdbDlClose, /* xDlClose */ + memdbRandomness, /* xRandomness */ + memdbSleep, /* xSleep */ + 0, /* memdbCurrentTime, */ /* xCurrentTime */ + memdbGetLastError, /* xGetLastError */ + memdbCurrentTimeInt64 /* xCurrentTimeInt64 */ +}; + +static const sqlite3_io_methods memdb_io_methods = { + 3, /* iVersion */ + memdbClose, /* xClose */ + memdbRead, /* xRead */ + memdbWrite, /* xWrite */ + memdbTruncate, /* xTruncate */ + memdbSync, /* xSync */ + memdbFileSize, /* xFileSize */ + memdbLock, /* xLock */ + memdbLock, /* xUnlock - same as xLock in this case */ + 0, /* memdbCheckReservedLock, */ /* xCheckReservedLock */ + memdbFileControl, /* xFileControl */ + 0, /* memdbSectorSize,*/ /* xSectorSize */ + memdbDeviceCharacteristics, /* xDeviceCharacteristics */ + 0, /* xShmMap */ + 0, /* xShmLock */ + 0, /* xShmBarrier */ + 0, /* xShmUnmap */ + memdbFetch, /* xFetch */ + memdbUnfetch /* xUnfetch */ +}; + + + +/* +** Close an memdb-file. +** +** The pData pointer is owned by the application, so there is nothing +** to free. +*/ +static int memdbClose(sqlite3_file *pFile){ + MemFile *p = (MemFile *)pFile; + if( p->mFlags & SQLITE_DESERIALIZE_FREEONCLOSE ) sqlite3_free(p->aData); + return SQLITE_OK; +} + +/* +** Read data from an memdb-file. +*/ +static int memdbRead( + sqlite3_file *pFile, + void *zBuf, + int iAmt, + sqlite_int64 iOfst +){ + MemFile *p = (MemFile *)pFile; + if( iOfst+iAmt>p->sz ){ + memset(zBuf, 0, iAmt); + if( iOfstsz ) memcpy(zBuf, p->aData+iOfst, p->sz - iOfst); + return SQLITE_IOERR_SHORT_READ; + } + memcpy(zBuf, p->aData+iOfst, iAmt); + return SQLITE_OK; +} + +/* +** Try to enlarge the memory allocation to hold at least sz bytes +*/ +static int memdbEnlarge(MemFile *p, sqlite3_int64 newSz){ + unsigned char *pNew; + if( (p->mFlags & SQLITE_DESERIALIZE_RESIZEABLE)==0 || p->nMmap>0 ){ + return SQLITE_FULL; + } + if( newSz>p->szMax ){ + return SQLITE_FULL; + } + newSz *= 2; + if( newSz>p->szMax ) newSz = p->szMax; + pNew = sqlite3_realloc64(p->aData, newSz); + if( pNew==0 ) return SQLITE_NOMEM; + p->aData = pNew; + p->szAlloc = newSz; + return SQLITE_OK; +} + +/* +** Write data to an memdb-file. +*/ +static int memdbWrite( + sqlite3_file *pFile, + const void *z, + int iAmt, + sqlite_int64 iOfst +){ + MemFile *p = (MemFile *)pFile; + if( NEVER(p->mFlags & SQLITE_DESERIALIZE_READONLY) ) return SQLITE_READONLY; + if( iOfst+iAmt>p->sz ){ + int rc; + if( iOfst+iAmt>p->szAlloc + && (rc = memdbEnlarge(p, iOfst+iAmt))!=SQLITE_OK + ){ + return rc; + } + if( iOfst>p->sz ) memset(p->aData+p->sz, 0, iOfst-p->sz); + p->sz = iOfst+iAmt; + } + memcpy(p->aData+iOfst, z, iAmt); + return SQLITE_OK; +} + +/* +** Truncate an memdb-file. +** +** In rollback mode (which is always the case for memdb, as it does not +** support WAL mode) the truncate() method is only used to reduce +** the size of a file, never to increase the size. +*/ +static int memdbTruncate(sqlite3_file *pFile, sqlite_int64 size){ + MemFile *p = (MemFile *)pFile; + if( NEVER(size>p->sz) ) return SQLITE_FULL; + p->sz = size; + return SQLITE_OK; +} + +/* +** Sync an memdb-file. +*/ +static int memdbSync(sqlite3_file *pFile, int flags){ + return SQLITE_OK; +} + +/* +** Return the current file-size of an memdb-file. +*/ +static int memdbFileSize(sqlite3_file *pFile, sqlite_int64 *pSize){ + MemFile *p = (MemFile *)pFile; + *pSize = p->sz; + return SQLITE_OK; +} + +/* +** Lock an memdb-file. +*/ +static int memdbLock(sqlite3_file *pFile, int eLock){ + MemFile *p = (MemFile *)pFile; + if( eLock>SQLITE_LOCK_SHARED + && (p->mFlags & SQLITE_DESERIALIZE_READONLY)!=0 + ){ + return SQLITE_READONLY; + } + p->eLock = eLock; + return SQLITE_OK; +} + +#if 0 /* Never used because memdbAccess() always returns false */ +/* +** Check if another file-handle holds a RESERVED lock on an memdb-file. +*/ +static int memdbCheckReservedLock(sqlite3_file *pFile, int *pResOut){ + *pResOut = 0; + return SQLITE_OK; +} +#endif + +/* +** File control method. For custom operations on an memdb-file. +*/ +static int memdbFileControl(sqlite3_file *pFile, int op, void *pArg){ + MemFile *p = (MemFile *)pFile; + int rc = SQLITE_NOTFOUND; + if( op==SQLITE_FCNTL_VFSNAME ){ + *(char**)pArg = sqlite3_mprintf("memdb(%p,%lld)", p->aData, p->sz); + rc = SQLITE_OK; + } + if( op==SQLITE_FCNTL_SIZE_LIMIT ){ + sqlite3_int64 iLimit = *(sqlite3_int64*)pArg; + if( iLimitsz ){ + if( iLimit<0 ){ + iLimit = p->szMax; + }else{ + iLimit = p->sz; + } + } + p->szMax = iLimit; + *(sqlite3_int64*)pArg = iLimit; + rc = SQLITE_OK; + } + return rc; +} + +#if 0 /* Not used because of SQLITE_IOCAP_POWERSAFE_OVERWRITE */ +/* +** Return the sector-size in bytes for an memdb-file. +*/ +static int memdbSectorSize(sqlite3_file *pFile){ + return 1024; +} +#endif + +/* +** Return the device characteristic flags supported by an memdb-file. +*/ +static int memdbDeviceCharacteristics(sqlite3_file *pFile){ + return SQLITE_IOCAP_ATOMIC | + SQLITE_IOCAP_POWERSAFE_OVERWRITE | + SQLITE_IOCAP_SAFE_APPEND | + SQLITE_IOCAP_SEQUENTIAL; +} + +/* Fetch a page of a memory-mapped file */ +static int memdbFetch( + sqlite3_file *pFile, + sqlite3_int64 iOfst, + int iAmt, + void **pp +){ + MemFile *p = (MemFile *)pFile; + if( iOfst+iAmt>p->sz ){ + *pp = 0; + }else{ + p->nMmap++; + *pp = (void*)(p->aData + iOfst); + } + return SQLITE_OK; +} + +/* Release a memory-mapped page */ +static int memdbUnfetch(sqlite3_file *pFile, sqlite3_int64 iOfst, void *pPage){ + MemFile *p = (MemFile *)pFile; + p->nMmap--; + return SQLITE_OK; +} + +/* +** Open an mem file handle. +*/ +static int memdbOpen( + sqlite3_vfs *pVfs, + const char *zName, + sqlite3_file *pFile, + int flags, + int *pOutFlags +){ + MemFile *p = (MemFile*)pFile; + if( (flags & SQLITE_OPEN_MAIN_DB)==0 ){ + return ORIGVFS(pVfs)->xOpen(ORIGVFS(pVfs), zName, pFile, flags, pOutFlags); + } + memset(p, 0, sizeof(*p)); + p->mFlags = SQLITE_DESERIALIZE_RESIZEABLE | SQLITE_DESERIALIZE_FREEONCLOSE; + assert( pOutFlags!=0 ); /* True because flags==SQLITE_OPEN_MAIN_DB */ + *pOutFlags = flags | SQLITE_OPEN_MEMORY; + p->base.pMethods = &memdb_io_methods; + p->szMax = sqlite3GlobalConfig.mxMemdbSize; + return SQLITE_OK; +} + +#if 0 /* Only used to delete rollback journals, master journals, and WAL + ** files, none of which exist in memdb. So this routine is never used */ +/* +** Delete the file located at zPath. If the dirSync argument is true, +** ensure the file-system modifications are synced to disk before +** returning. +*/ +static int memdbDelete(sqlite3_vfs *pVfs, const char *zPath, int dirSync){ + return SQLITE_IOERR_DELETE; +} +#endif + +/* +** Test for access permissions. Return true if the requested permission +** is available, or false otherwise. +** +** With memdb, no files ever exist on disk. So always return false. +*/ +static int memdbAccess( + sqlite3_vfs *pVfs, + const char *zPath, + int flags, + int *pResOut +){ + *pResOut = 0; + return SQLITE_OK; +} + +/* +** Populate buffer zOut with the full canonical pathname corresponding +** to the pathname in zPath. zOut is guaranteed to point to a buffer +** of at least (INST_MAX_PATHNAME+1) bytes. +*/ +static int memdbFullPathname( + sqlite3_vfs *pVfs, + const char *zPath, + int nOut, + char *zOut +){ + sqlite3_snprintf(nOut, zOut, "%s", zPath); + return SQLITE_OK; +} + +/* +** Open the dynamic library located at zPath and return a handle. +*/ +static void *memdbDlOpen(sqlite3_vfs *pVfs, const char *zPath){ + return ORIGVFS(pVfs)->xDlOpen(ORIGVFS(pVfs), zPath); +} + +/* +** Populate the buffer zErrMsg (size nByte bytes) with a human readable +** utf-8 string describing the most recent error encountered associated +** with dynamic libraries. +*/ +static void memdbDlError(sqlite3_vfs *pVfs, int nByte, char *zErrMsg){ + ORIGVFS(pVfs)->xDlError(ORIGVFS(pVfs), nByte, zErrMsg); +} + +/* +** Return a pointer to the symbol zSymbol in the dynamic library pHandle. +*/ +static void (*memdbDlSym(sqlite3_vfs *pVfs, void *p, const char *zSym))(void){ + return ORIGVFS(pVfs)->xDlSym(ORIGVFS(pVfs), p, zSym); +} + +/* +** Close the dynamic library handle pHandle. +*/ +static void memdbDlClose(sqlite3_vfs *pVfs, void *pHandle){ + ORIGVFS(pVfs)->xDlClose(ORIGVFS(pVfs), pHandle); +} + +/* +** Populate the buffer pointed to by zBufOut with nByte bytes of +** random data. +*/ +static int memdbRandomness(sqlite3_vfs *pVfs, int nByte, char *zBufOut){ + return ORIGVFS(pVfs)->xRandomness(ORIGVFS(pVfs), nByte, zBufOut); +} + +/* +** Sleep for nMicro microseconds. Return the number of microseconds +** actually slept. +*/ +static int memdbSleep(sqlite3_vfs *pVfs, int nMicro){ + return ORIGVFS(pVfs)->xSleep(ORIGVFS(pVfs), nMicro); +} + +#if 0 /* Never used. Modern cores only call xCurrentTimeInt64() */ +/* +** Return the current time as a Julian Day number in *pTimeOut. +*/ +static int memdbCurrentTime(sqlite3_vfs *pVfs, double *pTimeOut){ + return ORIGVFS(pVfs)->xCurrentTime(ORIGVFS(pVfs), pTimeOut); +} +#endif + +static int memdbGetLastError(sqlite3_vfs *pVfs, int a, char *b){ + return ORIGVFS(pVfs)->xGetLastError(ORIGVFS(pVfs), a, b); +} +static int memdbCurrentTimeInt64(sqlite3_vfs *pVfs, sqlite3_int64 *p){ + return ORIGVFS(pVfs)->xCurrentTimeInt64(ORIGVFS(pVfs), p); +} + +/* +** Translate a database connection pointer and schema name into a +** MemFile pointer. +*/ +static MemFile *memdbFromDbSchema(sqlite3 *db, const char *zSchema){ + MemFile *p = 0; + int rc = sqlite3_file_control(db, zSchema, SQLITE_FCNTL_FILE_POINTER, &p); + if( rc ) return 0; + if( p->base.pMethods!=&memdb_io_methods ) return 0; + return p; +} + +/* +** Return the serialization of a database +*/ +SQLITE_API unsigned char *sqlite3_serialize( + sqlite3 *db, /* The database connection */ + const char *zSchema, /* Which database within the connection */ + sqlite3_int64 *piSize, /* Write size here, if not NULL */ + unsigned int mFlags /* Maybe SQLITE_SERIALIZE_NOCOPY */ +){ + MemFile *p; + int iDb; + Btree *pBt; + sqlite3_int64 sz; + int szPage = 0; + sqlite3_stmt *pStmt = 0; + unsigned char *pOut; + char *zSql; + int rc; + +#ifdef SQLITE_ENABLE_API_ARMOR + if( !sqlite3SafetyCheckOk(db) ){ + (void)SQLITE_MISUSE_BKPT; + return 0; + } +#endif + + if( zSchema==0 ) zSchema = db->aDb[0].zDbSName; + p = memdbFromDbSchema(db, zSchema); + iDb = sqlite3FindDbName(db, zSchema); + if( piSize ) *piSize = -1; + if( iDb<0 ) return 0; + if( p ){ + if( piSize ) *piSize = p->sz; + if( mFlags & SQLITE_SERIALIZE_NOCOPY ){ + pOut = p->aData; + }else{ + pOut = sqlite3_malloc64( p->sz ); + if( pOut ) memcpy(pOut, p->aData, p->sz); + } + return pOut; + } + pBt = db->aDb[iDb].pBt; + if( pBt==0 ) return 0; + szPage = sqlite3BtreeGetPageSize(pBt); + zSql = sqlite3_mprintf("PRAGMA \"%w\".page_count", zSchema); + rc = zSql ? sqlite3_prepare_v2(db, zSql, -1, &pStmt, 0) : SQLITE_NOMEM; + sqlite3_free(zSql); + if( rc ) return 0; + rc = sqlite3_step(pStmt); + if( rc!=SQLITE_ROW ){ + pOut = 0; + }else{ + sz = sqlite3_column_int64(pStmt, 0)*szPage; + if( piSize ) *piSize = sz; + if( mFlags & SQLITE_SERIALIZE_NOCOPY ){ + pOut = 0; + }else{ + pOut = sqlite3_malloc64( sz ); + if( pOut ){ + int nPage = sqlite3_column_int(pStmt, 0); + Pager *pPager = sqlite3BtreePager(pBt); + int pgno; + for(pgno=1; pgno<=nPage; pgno++){ + DbPage *pPage = 0; + unsigned char *pTo = pOut + szPage*(sqlite3_int64)(pgno-1); + rc = sqlite3PagerGet(pPager, pgno, (DbPage**)&pPage, 0); + if( rc==SQLITE_OK ){ + memcpy(pTo, sqlite3PagerGetData(pPage), szPage); + }else{ + memset(pTo, 0, szPage); + } + sqlite3PagerUnref(pPage); + } + } + } + } + sqlite3_finalize(pStmt); + return pOut; +} + +/* Convert zSchema to a MemDB and initialize its content. +*/ +SQLITE_API int sqlite3_deserialize( + sqlite3 *db, /* The database connection */ + const char *zSchema, /* Which DB to reopen with the deserialization */ + unsigned char *pData, /* The serialized database content */ + sqlite3_int64 szDb, /* Number bytes in the deserialization */ + sqlite3_int64 szBuf, /* Total size of buffer pData[] */ + unsigned mFlags /* Zero or more SQLITE_DESERIALIZE_* flags */ +){ + MemFile *p; + char *zSql; + sqlite3_stmt *pStmt = 0; + int rc; + int iDb; + +#ifdef SQLITE_ENABLE_API_ARMOR + if( !sqlite3SafetyCheckOk(db) ){ + return SQLITE_MISUSE_BKPT; + } + if( szDb<0 ) return SQLITE_MISUSE_BKPT; + if( szBuf<0 ) return SQLITE_MISUSE_BKPT; +#endif + + sqlite3_mutex_enter(db->mutex); + if( zSchema==0 ) zSchema = db->aDb[0].zDbSName; + iDb = sqlite3FindDbName(db, zSchema); + if( iDb<0 ){ + rc = SQLITE_ERROR; + goto end_deserialize; + } + zSql = sqlite3_mprintf("ATTACH x AS %Q", zSchema); + rc = sqlite3_prepare_v2(db, zSql, -1, &pStmt, 0); + sqlite3_free(zSql); + if( rc ) goto end_deserialize; + db->init.iDb = (u8)iDb; + db->init.reopenMemdb = 1; + rc = sqlite3_step(pStmt); + db->init.reopenMemdb = 0; + if( rc!=SQLITE_DONE ){ + rc = SQLITE_ERROR; + goto end_deserialize; + } + p = memdbFromDbSchema(db, zSchema); + if( p==0 ){ + rc = SQLITE_ERROR; + }else{ + p->aData = pData; + p->sz = szDb; + p->szAlloc = szBuf; + p->szMax = szBuf; + if( p->szMaxszMax = sqlite3GlobalConfig.mxMemdbSize; + } + p->mFlags = mFlags; + rc = SQLITE_OK; + } + +end_deserialize: + sqlite3_finalize(pStmt); + sqlite3_mutex_leave(db->mutex); + return rc; +} + +/* +** This routine is called when the extension is loaded. +** Register the new VFS. +*/ +SQLITE_PRIVATE int sqlite3MemdbInit(void){ + sqlite3_vfs *pLower = sqlite3_vfs_find(0); + int sz = pLower->szOsFile; + memdb_vfs.pAppData = pLower; + /* In all known configurations of SQLite, the size of a default + ** sqlite3_file is greater than the size of a memdb sqlite3_file. + ** Should that ever change, remove the following NEVER() */ + if( NEVER(szBITVEC_NBIT and iDivisor==0 then Bitvec.u.aHash[] is +** a hash table that will hold up to BITVEC_MXHASH distinct values. +** +** Otherwise, the value i is redirected into one of BITVEC_NPTR +** sub-bitmaps pointed to by Bitvec.u.apSub[]. Each subbitmap +** handles up to iDivisor separate values of i. apSub[0] holds +** values between 1 and iDivisor. apSub[1] holds values between +** iDivisor+1 and 2*iDivisor. apSub[N] holds values between +** N*iDivisor+1 and (N+1)*iDivisor. Each subbitmap is normalized +** to hold deal with values between 1 and iDivisor. +*/ +struct Bitvec { + u32 iSize; /* Maximum bit index. Max iSize is 4,294,967,296. */ + u32 nSet; /* Number of bits that are set - only valid for aHash + ** element. Max is BITVEC_NINT. For BITVEC_SZ of 512, + ** this would be 125. */ + u32 iDivisor; /* Number of bits handled by each apSub[] entry. */ + /* Should >=0 for apSub element. */ + /* Max iDivisor is max(u32) / BITVEC_NPTR + 1. */ + /* For a BITVEC_SZ of 512, this would be 34,359,739. */ + union { + BITVEC_TELEM aBitmap[BITVEC_NELEM]; /* Bitmap representation */ + u32 aHash[BITVEC_NINT]; /* Hash table representation */ + Bitvec *apSub[BITVEC_NPTR]; /* Recursive representation */ + } u; +}; + +/* +** Create a new bitmap object able to handle bits between 0 and iSize, +** inclusive. Return a pointer to the new object. Return NULL if +** malloc fails. +*/ +SQLITE_PRIVATE Bitvec *sqlite3BitvecCreate(u32 iSize){ + Bitvec *p; + assert( sizeof(*p)==BITVEC_SZ ); + p = sqlite3MallocZero( sizeof(*p) ); + if( p ){ + p->iSize = iSize; + } + return p; +} + +/* +** Check to see if the i-th bit is set. Return true or false. +** If p is NULL (if the bitmap has not been created) or if +** i is out of range, then return false. +*/ +SQLITE_PRIVATE int sqlite3BitvecTestNotNull(Bitvec *p, u32 i){ + assert( p!=0 ); + i--; + if( i>=p->iSize ) return 0; + while( p->iDivisor ){ + u32 bin = i/p->iDivisor; + i = i%p->iDivisor; + p = p->u.apSub[bin]; + if (!p) { + return 0; + } + } + if( p->iSize<=BITVEC_NBIT ){ + return (p->u.aBitmap[i/BITVEC_SZELEM] & (1<<(i&(BITVEC_SZELEM-1))))!=0; + } else{ + u32 h = BITVEC_HASH(i++); + while( p->u.aHash[h] ){ + if( p->u.aHash[h]==i ) return 1; + h = (h+1) % BITVEC_NINT; + } + return 0; + } +} +SQLITE_PRIVATE int sqlite3BitvecTest(Bitvec *p, u32 i){ + return p!=0 && sqlite3BitvecTestNotNull(p,i); +} + +/* +** Set the i-th bit. Return 0 on success and an error code if +** anything goes wrong. +** +** This routine might cause sub-bitmaps to be allocated. Failing +** to get the memory needed to hold the sub-bitmap is the only +** that can go wrong with an insert, assuming p and i are valid. +** +** The calling function must ensure that p is a valid Bitvec object +** and that the value for "i" is within range of the Bitvec object. +** Otherwise the behavior is undefined. +*/ +SQLITE_PRIVATE int sqlite3BitvecSet(Bitvec *p, u32 i){ + u32 h; + if( p==0 ) return SQLITE_OK; + assert( i>0 ); + assert( i<=p->iSize ); + i--; + while((p->iSize > BITVEC_NBIT) && p->iDivisor) { + u32 bin = i/p->iDivisor; + i = i%p->iDivisor; + if( p->u.apSub[bin]==0 ){ + p->u.apSub[bin] = sqlite3BitvecCreate( p->iDivisor ); + if( p->u.apSub[bin]==0 ) return SQLITE_NOMEM_BKPT; + } + p = p->u.apSub[bin]; + } + if( p->iSize<=BITVEC_NBIT ){ + p->u.aBitmap[i/BITVEC_SZELEM] |= 1 << (i&(BITVEC_SZELEM-1)); + return SQLITE_OK; + } + h = BITVEC_HASH(i++); + /* if there wasn't a hash collision, and this doesn't */ + /* completely fill the hash, then just add it without */ + /* worring about sub-dividing and re-hashing. */ + if( !p->u.aHash[h] ){ + if (p->nSet<(BITVEC_NINT-1)) { + goto bitvec_set_end; + } else { + goto bitvec_set_rehash; + } + } + /* there was a collision, check to see if it's already */ + /* in hash, if not, try to find a spot for it */ + do { + if( p->u.aHash[h]==i ) return SQLITE_OK; + h++; + if( h>=BITVEC_NINT ) h = 0; + } while( p->u.aHash[h] ); + /* we didn't find it in the hash. h points to the first */ + /* available free spot. check to see if this is going to */ + /* make our hash too "full". */ +bitvec_set_rehash: + if( p->nSet>=BITVEC_MXHASH ){ + unsigned int j; + int rc; + u32 *aiValues = sqlite3StackAllocRaw(0, sizeof(p->u.aHash)); + if( aiValues==0 ){ + return SQLITE_NOMEM_BKPT; + }else{ + memcpy(aiValues, p->u.aHash, sizeof(p->u.aHash)); + memset(p->u.apSub, 0, sizeof(p->u.apSub)); + p->iDivisor = (p->iSize + BITVEC_NPTR - 1)/BITVEC_NPTR; + rc = sqlite3BitvecSet(p, i); + for(j=0; jnSet++; + p->u.aHash[h] = i; + return SQLITE_OK; +} + +/* +** Clear the i-th bit. +** +** pBuf must be a pointer to at least BITVEC_SZ bytes of temporary storage +** that BitvecClear can use to rebuilt its hash table. +*/ +SQLITE_PRIVATE void sqlite3BitvecClear(Bitvec *p, u32 i, void *pBuf){ + if( p==0 ) return; + assert( i>0 ); + i--; + while( p->iDivisor ){ + u32 bin = i/p->iDivisor; + i = i%p->iDivisor; + p = p->u.apSub[bin]; + if (!p) { + return; + } + } + if( p->iSize<=BITVEC_NBIT ){ + p->u.aBitmap[i/BITVEC_SZELEM] &= ~(1 << (i&(BITVEC_SZELEM-1))); + }else{ + unsigned int j; + u32 *aiValues = pBuf; + memcpy(aiValues, p->u.aHash, sizeof(p->u.aHash)); + memset(p->u.aHash, 0, sizeof(p->u.aHash)); + p->nSet = 0; + for(j=0; jnSet++; + while( p->u.aHash[h] ){ + h++; + if( h>=BITVEC_NINT ) h = 0; + } + p->u.aHash[h] = aiValues[j]; + } + } + } +} + +/* +** Destroy a bitmap object. Reclaim all memory used. +*/ +SQLITE_PRIVATE void sqlite3BitvecDestroy(Bitvec *p){ + if( p==0 ) return; + if( p->iDivisor ){ + unsigned int i; + for(i=0; iu.apSub[i]); + } + } + sqlite3_free(p); +} + +/* +** Return the value of the iSize parameter specified when Bitvec *p +** was created. +*/ +SQLITE_PRIVATE u32 sqlite3BitvecSize(Bitvec *p){ + return p->iSize; +} + +#ifndef SQLITE_UNTESTABLE +/* +** Let V[] be an array of unsigned characters sufficient to hold +** up to N bits. Let I be an integer between 0 and N. 0<=I>3] |= (1<<(I&7)) +#define CLEARBIT(V,I) V[I>>3] &= ~(1<<(I&7)) +#define TESTBIT(V,I) (V[I>>3]&(1<<(I&7)))!=0 + +/* +** This routine runs an extensive test of the Bitvec code. +** +** The input is an array of integers that acts as a program +** to test the Bitvec. The integers are opcodes followed +** by 0, 1, or 3 operands, depending on the opcode. Another +** opcode follows immediately after the last operand. +** +** There are 6 opcodes numbered from 0 through 5. 0 is the +** "halt" opcode and causes the test to end. +** +** 0 Halt and return the number of errors +** 1 N S X Set N bits beginning with S and incrementing by X +** 2 N S X Clear N bits beginning with S and incrementing by X +** 3 N Set N randomly chosen bits +** 4 N Clear N randomly chosen bits +** 5 N S X Set N bits from S increment X in array only, not in bitvec +** +** The opcodes 1 through 4 perform set and clear operations are performed +** on both a Bitvec object and on a linear array of bits obtained from malloc. +** Opcode 5 works on the linear array only, not on the Bitvec. +** Opcode 5 is used to deliberately induce a fault in order to +** confirm that error detection works. +** +** At the conclusion of the test the linear array is compared +** against the Bitvec object. If there are any differences, +** an error is returned. If they are the same, zero is returned. +** +** If a memory allocation error occurs, return -1. +*/ +SQLITE_PRIVATE int sqlite3BitvecBuiltinTest(int sz, int *aOp){ + Bitvec *pBitvec = 0; + unsigned char *pV = 0; + int rc = -1; + int i, nx, pc, op; + void *pTmpSpace; + + /* Allocate the Bitvec to be tested and a linear array of + ** bits to act as the reference */ + pBitvec = sqlite3BitvecCreate( sz ); + pV = sqlite3MallocZero( (sz+7)/8 + 1 ); + pTmpSpace = sqlite3_malloc64(BITVEC_SZ); + if( pBitvec==0 || pV==0 || pTmpSpace==0 ) goto bitvec_end; + + /* NULL pBitvec tests */ + sqlite3BitvecSet(0, 1); + sqlite3BitvecClear(0, 1, pTmpSpace); + + /* Run the program */ + pc = 0; + while( (op = aOp[pc])!=0 ){ + switch( op ){ + case 1: + case 2: + case 5: { + nx = 4; + i = aOp[pc+2] - 1; + aOp[pc+2] += aOp[pc+3]; + break; + } + case 3: + case 4: + default: { + nx = 2; + sqlite3_randomness(sizeof(i), &i); + break; + } + } + if( (--aOp[pc+1]) > 0 ) nx = 0; + pc += nx; + i = (i & 0x7fffffff)%sz; + if( (op & 1)!=0 ){ + SETBIT(pV, (i+1)); + if( op!=5 ){ + if( sqlite3BitvecSet(pBitvec, i+1) ) goto bitvec_end; + } + }else{ + CLEARBIT(pV, (i+1)); + sqlite3BitvecClear(pBitvec, i+1, pTmpSpace); + } + } + + /* Test to make sure the linear array exactly matches the + ** Bitvec object. Start with the assumption that they do + ** match (rc==0). Change rc to non-zero if a discrepancy + ** is found. + */ + rc = sqlite3BitvecTest(0,0) + sqlite3BitvecTest(pBitvec, sz+1) + + sqlite3BitvecTest(pBitvec, 0) + + (sqlite3BitvecSize(pBitvec) - sz); + for(i=1; i<=sz; i++){ + if( (TESTBIT(pV,i))!=sqlite3BitvecTest(pBitvec,i) ){ + rc = i; + break; + } + } + + /* Free allocated structure */ +bitvec_end: + sqlite3_free(pTmpSpace); + sqlite3_free(pV); + sqlite3BitvecDestroy(pBitvec); + return rc; +} +#endif /* SQLITE_UNTESTABLE */ + +/************** End of bitvec.c **********************************************/ +/************** Begin file pcache.c ******************************************/ +/* +** 2008 August 05 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** This file implements that page cache. +*/ +/* #include "sqliteInt.h" */ + +/* +** A complete page cache is an instance of this structure. Every +** entry in the cache holds a single page of the database file. The +** btree layer only operates on the cached copy of the database pages. +** +** A page cache entry is "clean" if it exactly matches what is currently +** on disk. A page is "dirty" if it has been modified and needs to be +** persisted to disk. +** +** pDirty, pDirtyTail, pSynced: +** All dirty pages are linked into the doubly linked list using +** PgHdr.pDirtyNext and pDirtyPrev. The list is maintained in LRU order +** such that p was added to the list more recently than p->pDirtyNext. +** PCache.pDirty points to the first (newest) element in the list and +** pDirtyTail to the last (oldest). +** +** The PCache.pSynced variable is used to optimize searching for a dirty +** page to eject from the cache mid-transaction. It is better to eject +** a page that does not require a journal sync than one that does. +** Therefore, pSynced is maintained so that it *almost* always points +** to either the oldest page in the pDirty/pDirtyTail list that has a +** clear PGHDR_NEED_SYNC flag or to a page that is older than this one +** (so that the right page to eject can be found by following pDirtyPrev +** pointers). +*/ +struct PCache { + PgHdr *pDirty, *pDirtyTail; /* List of dirty pages in LRU order */ + PgHdr *pSynced; /* Last synced page in dirty page list */ + int nRefSum; /* Sum of ref counts over all pages */ + int szCache; /* Configured cache size */ + int szSpill; /* Size before spilling occurs */ + int szPage; /* Size of every page in this cache */ + int szExtra; /* Size of extra space for each page */ + u8 bPurgeable; /* True if pages are on backing store */ + u8 eCreate; /* eCreate value for for xFetch() */ + int (*xStress)(void*,PgHdr*); /* Call to try make a page clean */ + void *pStress; /* Argument to xStress */ + sqlite3_pcache *pCache; /* Pluggable cache module */ +}; + +/********************************** Test and Debug Logic **********************/ +/* +** Debug tracing macros. Enable by by changing the "0" to "1" and +** recompiling. +** +** When sqlite3PcacheTrace is 1, single line trace messages are issued. +** When sqlite3PcacheTrace is 2, a dump of the pcache showing all cache entries +** is displayed for many operations, resulting in a lot of output. +*/ +#if defined(SQLITE_DEBUG) && 0 + int sqlite3PcacheTrace = 2; /* 0: off 1: simple 2: cache dumps */ + int sqlite3PcacheMxDump = 9999; /* Max cache entries for pcacheDump() */ +# define pcacheTrace(X) if(sqlite3PcacheTrace){sqlite3DebugPrintf X;} + void pcacheDump(PCache *pCache){ + int N; + int i, j; + sqlite3_pcache_page *pLower; + PgHdr *pPg; + unsigned char *a; + + if( sqlite3PcacheTrace<2 ) return; + if( pCache->pCache==0 ) return; + N = sqlite3PcachePagecount(pCache); + if( N>sqlite3PcacheMxDump ) N = sqlite3PcacheMxDump; + for(i=1; i<=N; i++){ + pLower = sqlite3GlobalConfig.pcache2.xFetch(pCache->pCache, i, 0); + if( pLower==0 ) continue; + pPg = (PgHdr*)pLower->pExtra; + printf("%3d: nRef %2d flgs %02x data ", i, pPg->nRef, pPg->flags); + a = (unsigned char *)pLower->pBuf; + for(j=0; j<12; j++) printf("%02x", a[j]); + printf("\n"); + if( pPg->pPage==0 ){ + sqlite3GlobalConfig.pcache2.xUnpin(pCache->pCache, pLower, 0); + } + } + } + #else +# define pcacheTrace(X) +# define pcacheDump(X) +#endif + +/* +** Check invariants on a PgHdr entry. Return true if everything is OK. +** Return false if any invariant is violated. +** +** This routine is for use inside of assert() statements only. For +** example: +** +** assert( sqlite3PcachePageSanity(pPg) ); +*/ +#ifdef SQLITE_DEBUG +SQLITE_PRIVATE int sqlite3PcachePageSanity(PgHdr *pPg){ + PCache *pCache; + assert( pPg!=0 ); + assert( pPg->pgno>0 || pPg->pPager==0 ); /* Page number is 1 or more */ + pCache = pPg->pCache; + assert( pCache!=0 ); /* Every page has an associated PCache */ + if( pPg->flags & PGHDR_CLEAN ){ + assert( (pPg->flags & PGHDR_DIRTY)==0 );/* Cannot be both CLEAN and DIRTY */ + assert( pCache->pDirty!=pPg ); /* CLEAN pages not on dirty list */ + assert( pCache->pDirtyTail!=pPg ); + } + /* WRITEABLE pages must also be DIRTY */ + if( pPg->flags & PGHDR_WRITEABLE ){ + assert( pPg->flags & PGHDR_DIRTY ); /* WRITEABLE implies DIRTY */ + } + /* NEED_SYNC can be set independently of WRITEABLE. This can happen, + ** for example, when using the sqlite3PagerDontWrite() optimization: + ** (1) Page X is journalled, and gets WRITEABLE and NEED_SEEK. + ** (2) Page X moved to freelist, WRITEABLE is cleared + ** (3) Page X reused, WRITEABLE is set again + ** If NEED_SYNC had been cleared in step 2, then it would not be reset + ** in step 3, and page might be written into the database without first + ** syncing the rollback journal, which might cause corruption on a power + ** loss. + ** + ** Another example is when the database page size is smaller than the + ** disk sector size. When any page of a sector is journalled, all pages + ** in that sector are marked NEED_SYNC even if they are still CLEAN, just + ** in case they are later modified, since all pages in the same sector + ** must be journalled and synced before any of those pages can be safely + ** written. + */ + return 1; +} +#endif /* SQLITE_DEBUG */ + + +/********************************** Linked List Management ********************/ + +/* Allowed values for second argument to pcacheManageDirtyList() */ +#define PCACHE_DIRTYLIST_REMOVE 1 /* Remove pPage from dirty list */ +#define PCACHE_DIRTYLIST_ADD 2 /* Add pPage to the dirty list */ +#define PCACHE_DIRTYLIST_FRONT 3 /* Move pPage to the front of the list */ + +/* +** Manage pPage's participation on the dirty list. Bits of the addRemove +** argument determines what operation to do. The 0x01 bit means first +** remove pPage from the dirty list. The 0x02 means add pPage back to +** the dirty list. Doing both moves pPage to the front of the dirty list. +*/ +static void pcacheManageDirtyList(PgHdr *pPage, u8 addRemove){ + PCache *p = pPage->pCache; + + pcacheTrace(("%p.DIRTYLIST.%s %d\n", p, + addRemove==1 ? "REMOVE" : addRemove==2 ? "ADD" : "FRONT", + pPage->pgno)); + if( addRemove & PCACHE_DIRTYLIST_REMOVE ){ + assert( pPage->pDirtyNext || pPage==p->pDirtyTail ); + assert( pPage->pDirtyPrev || pPage==p->pDirty ); + + /* Update the PCache1.pSynced variable if necessary. */ + if( p->pSynced==pPage ){ + p->pSynced = pPage->pDirtyPrev; + } + + if( pPage->pDirtyNext ){ + pPage->pDirtyNext->pDirtyPrev = pPage->pDirtyPrev; + }else{ + assert( pPage==p->pDirtyTail ); + p->pDirtyTail = pPage->pDirtyPrev; + } + if( pPage->pDirtyPrev ){ + pPage->pDirtyPrev->pDirtyNext = pPage->pDirtyNext; + }else{ + /* If there are now no dirty pages in the cache, set eCreate to 2. + ** This is an optimization that allows sqlite3PcacheFetch() to skip + ** searching for a dirty page to eject from the cache when it might + ** otherwise have to. */ + assert( pPage==p->pDirty ); + p->pDirty = pPage->pDirtyNext; + assert( p->bPurgeable || p->eCreate==2 ); + if( p->pDirty==0 ){ /*OPTIMIZATION-IF-TRUE*/ + assert( p->bPurgeable==0 || p->eCreate==1 ); + p->eCreate = 2; + } + } + } + if( addRemove & PCACHE_DIRTYLIST_ADD ){ + pPage->pDirtyPrev = 0; + pPage->pDirtyNext = p->pDirty; + if( pPage->pDirtyNext ){ + assert( pPage->pDirtyNext->pDirtyPrev==0 ); + pPage->pDirtyNext->pDirtyPrev = pPage; + }else{ + p->pDirtyTail = pPage; + if( p->bPurgeable ){ + assert( p->eCreate==2 ); + p->eCreate = 1; + } + } + p->pDirty = pPage; + + /* If pSynced is NULL and this page has a clear NEED_SYNC flag, set + ** pSynced to point to it. Checking the NEED_SYNC flag is an + ** optimization, as if pSynced points to a page with the NEED_SYNC + ** flag set sqlite3PcacheFetchStress() searches through all newer + ** entries of the dirty-list for a page with NEED_SYNC clear anyway. */ + if( !p->pSynced + && 0==(pPage->flags&PGHDR_NEED_SYNC) /*OPTIMIZATION-IF-FALSE*/ + ){ + p->pSynced = pPage; + } + } + pcacheDump(p); +} + +/* +** Wrapper around the pluggable caches xUnpin method. If the cache is +** being used for an in-memory database, this function is a no-op. +*/ +static void pcacheUnpin(PgHdr *p){ + if( p->pCache->bPurgeable ){ + pcacheTrace(("%p.UNPIN %d\n", p->pCache, p->pgno)); + sqlite3GlobalConfig.pcache2.xUnpin(p->pCache->pCache, p->pPage, 0); + pcacheDump(p->pCache); + } +} + +/* +** Compute the number of pages of cache requested. p->szCache is the +** cache size requested by the "PRAGMA cache_size" statement. +*/ +static int numberOfCachePages(PCache *p){ + if( p->szCache>=0 ){ + /* IMPLEMENTATION-OF: R-42059-47211 If the argument N is positive then the + ** suggested cache size is set to N. */ + return p->szCache; + }else{ + /* IMPLEMANTATION-OF: R-59858-46238 If the argument N is negative, then the + ** number of cache pages is adjusted to be a number of pages that would + ** use approximately abs(N*1024) bytes of memory based on the current + ** page size. */ + return (int)((-1024*(i64)p->szCache)/(p->szPage+p->szExtra)); + } +} + +/*************************************************** General Interfaces ****** +** +** Initialize and shutdown the page cache subsystem. Neither of these +** functions are threadsafe. +*/ +SQLITE_PRIVATE int sqlite3PcacheInitialize(void){ + if( sqlite3GlobalConfig.pcache2.xInit==0 ){ + /* IMPLEMENTATION-OF: R-26801-64137 If the xInit() method is NULL, then the + ** built-in default page cache is used instead of the application defined + ** page cache. */ + sqlite3PCacheSetDefault(); + assert( sqlite3GlobalConfig.pcache2.xInit!=0 ); + } + return sqlite3GlobalConfig.pcache2.xInit(sqlite3GlobalConfig.pcache2.pArg); +} +SQLITE_PRIVATE void sqlite3PcacheShutdown(void){ + if( sqlite3GlobalConfig.pcache2.xShutdown ){ + /* IMPLEMENTATION-OF: R-26000-56589 The xShutdown() method may be NULL. */ + sqlite3GlobalConfig.pcache2.xShutdown(sqlite3GlobalConfig.pcache2.pArg); + } +} + +/* +** Return the size in bytes of a PCache object. +*/ +SQLITE_PRIVATE int sqlite3PcacheSize(void){ return sizeof(PCache); } + +/* +** Create a new PCache object. Storage space to hold the object +** has already been allocated and is passed in as the p pointer. +** The caller discovers how much space needs to be allocated by +** calling sqlite3PcacheSize(). +** +** szExtra is some extra space allocated for each page. The first +** 8 bytes of the extra space will be zeroed as the page is allocated, +** but remaining content will be uninitialized. Though it is opaque +** to this module, the extra space really ends up being the MemPage +** structure in the pager. +*/ +SQLITE_PRIVATE int sqlite3PcacheOpen( + int szPage, /* Size of every page */ + int szExtra, /* Extra space associated with each page */ + int bPurgeable, /* True if pages are on backing store */ + int (*xStress)(void*,PgHdr*),/* Call to try to make pages clean */ + void *pStress, /* Argument to xStress */ + PCache *p /* Preallocated space for the PCache */ +){ + memset(p, 0, sizeof(PCache)); + p->szPage = 1; + p->szExtra = szExtra; + assert( szExtra>=8 ); /* First 8 bytes will be zeroed */ + p->bPurgeable = bPurgeable; + p->eCreate = 2; + p->xStress = xStress; + p->pStress = pStress; + p->szCache = 100; + p->szSpill = 1; + pcacheTrace(("%p.OPEN szPage %d bPurgeable %d\n",p,szPage,bPurgeable)); + return sqlite3PcacheSetPageSize(p, szPage); +} + +/* +** Change the page size for PCache object. The caller must ensure that there +** are no outstanding page references when this function is called. +*/ +SQLITE_PRIVATE int sqlite3PcacheSetPageSize(PCache *pCache, int szPage){ + assert( pCache->nRefSum==0 && pCache->pDirty==0 ); + if( pCache->szPage ){ + sqlite3_pcache *pNew; + pNew = sqlite3GlobalConfig.pcache2.xCreate( + szPage, pCache->szExtra + ROUND8(sizeof(PgHdr)), + pCache->bPurgeable + ); + if( pNew==0 ) return SQLITE_NOMEM_BKPT; + sqlite3GlobalConfig.pcache2.xCachesize(pNew, numberOfCachePages(pCache)); + if( pCache->pCache ){ + sqlite3GlobalConfig.pcache2.xDestroy(pCache->pCache); + } + pCache->pCache = pNew; + pCache->szPage = szPage; + pcacheTrace(("%p.PAGESIZE %d\n",pCache,szPage)); + } + return SQLITE_OK; +} + +/* +** Try to obtain a page from the cache. +** +** This routine returns a pointer to an sqlite3_pcache_page object if +** such an object is already in cache, or if a new one is created. +** This routine returns a NULL pointer if the object was not in cache +** and could not be created. +** +** The createFlags should be 0 to check for existing pages and should +** be 3 (not 1, but 3) to try to create a new page. +** +** If the createFlag is 0, then NULL is always returned if the page +** is not already in the cache. If createFlag is 1, then a new page +** is created only if that can be done without spilling dirty pages +** and without exceeding the cache size limit. +** +** The caller needs to invoke sqlite3PcacheFetchFinish() to properly +** initialize the sqlite3_pcache_page object and convert it into a +** PgHdr object. The sqlite3PcacheFetch() and sqlite3PcacheFetchFinish() +** routines are split this way for performance reasons. When separated +** they can both (usually) operate without having to push values to +** the stack on entry and pop them back off on exit, which saves a +** lot of pushing and popping. +*/ +SQLITE_PRIVATE sqlite3_pcache_page *sqlite3PcacheFetch( + PCache *pCache, /* Obtain the page from this cache */ + Pgno pgno, /* Page number to obtain */ + int createFlag /* If true, create page if it does not exist already */ +){ + int eCreate; + sqlite3_pcache_page *pRes; + + assert( pCache!=0 ); + assert( pCache->pCache!=0 ); + assert( createFlag==3 || createFlag==0 ); + assert( pCache->eCreate==((pCache->bPurgeable && pCache->pDirty) ? 1 : 2) ); + + /* eCreate defines what to do if the page does not exist. + ** 0 Do not allocate a new page. (createFlag==0) + ** 1 Allocate a new page if doing so is inexpensive. + ** (createFlag==1 AND bPurgeable AND pDirty) + ** 2 Allocate a new page even it doing so is difficult. + ** (createFlag==1 AND !(bPurgeable AND pDirty) + */ + eCreate = createFlag & pCache->eCreate; + assert( eCreate==0 || eCreate==1 || eCreate==2 ); + assert( createFlag==0 || pCache->eCreate==eCreate ); + assert( createFlag==0 || eCreate==1+(!pCache->bPurgeable||!pCache->pDirty) ); + pRes = sqlite3GlobalConfig.pcache2.xFetch(pCache->pCache, pgno, eCreate); + pcacheTrace(("%p.FETCH %d%s (result: %p)\n",pCache,pgno, + createFlag?" create":"",pRes)); + return pRes; +} + +/* +** If the sqlite3PcacheFetch() routine is unable to allocate a new +** page because no clean pages are available for reuse and the cache +** size limit has been reached, then this routine can be invoked to +** try harder to allocate a page. This routine might invoke the stress +** callback to spill dirty pages to the journal. It will then try to +** allocate the new page and will only fail to allocate a new page on +** an OOM error. +** +** This routine should be invoked only after sqlite3PcacheFetch() fails. +*/ +SQLITE_PRIVATE int sqlite3PcacheFetchStress( + PCache *pCache, /* Obtain the page from this cache */ + Pgno pgno, /* Page number to obtain */ + sqlite3_pcache_page **ppPage /* Write result here */ +){ + PgHdr *pPg; + if( pCache->eCreate==2 ) return 0; + + if( sqlite3PcachePagecount(pCache)>pCache->szSpill ){ + /* Find a dirty page to write-out and recycle. First try to find a + ** page that does not require a journal-sync (one with PGHDR_NEED_SYNC + ** cleared), but if that is not possible settle for any other + ** unreferenced dirty page. + ** + ** If the LRU page in the dirty list that has a clear PGHDR_NEED_SYNC + ** flag is currently referenced, then the following may leave pSynced + ** set incorrectly (pointing to other than the LRU page with NEED_SYNC + ** cleared). This is Ok, as pSynced is just an optimization. */ + for(pPg=pCache->pSynced; + pPg && (pPg->nRef || (pPg->flags&PGHDR_NEED_SYNC)); + pPg=pPg->pDirtyPrev + ); + pCache->pSynced = pPg; + if( !pPg ){ + for(pPg=pCache->pDirtyTail; pPg && pPg->nRef; pPg=pPg->pDirtyPrev); + } + if( pPg ){ + int rc; +#ifdef SQLITE_LOG_CACHE_SPILL + sqlite3_log(SQLITE_FULL, + "spill page %d making room for %d - cache used: %d/%d", + pPg->pgno, pgno, + sqlite3GlobalConfig.pcache2.xPagecount(pCache->pCache), + numberOfCachePages(pCache)); +#endif + pcacheTrace(("%p.SPILL %d\n",pCache,pPg->pgno)); + rc = pCache->xStress(pCache->pStress, pPg); + pcacheDump(pCache); + if( rc!=SQLITE_OK && rc!=SQLITE_BUSY ){ + return rc; + } + } + } + *ppPage = sqlite3GlobalConfig.pcache2.xFetch(pCache->pCache, pgno, 2); + return *ppPage==0 ? SQLITE_NOMEM_BKPT : SQLITE_OK; +} + +/* +** This is a helper routine for sqlite3PcacheFetchFinish() +** +** In the uncommon case where the page being fetched has not been +** initialized, this routine is invoked to do the initialization. +** This routine is broken out into a separate function since it +** requires extra stack manipulation that can be avoided in the common +** case. +*/ +static SQLITE_NOINLINE PgHdr *pcacheFetchFinishWithInit( + PCache *pCache, /* Obtain the page from this cache */ + Pgno pgno, /* Page number obtained */ + sqlite3_pcache_page *pPage /* Page obtained by prior PcacheFetch() call */ +){ + PgHdr *pPgHdr; + assert( pPage!=0 ); + pPgHdr = (PgHdr*)pPage->pExtra; + assert( pPgHdr->pPage==0 ); + memset(&pPgHdr->pDirty, 0, sizeof(PgHdr) - offsetof(PgHdr,pDirty)); + pPgHdr->pPage = pPage; + pPgHdr->pData = pPage->pBuf; + pPgHdr->pExtra = (void *)&pPgHdr[1]; + memset(pPgHdr->pExtra, 0, 8); + pPgHdr->pCache = pCache; + pPgHdr->pgno = pgno; + pPgHdr->flags = PGHDR_CLEAN; + return sqlite3PcacheFetchFinish(pCache,pgno,pPage); +} + +/* +** This routine converts the sqlite3_pcache_page object returned by +** sqlite3PcacheFetch() into an initialized PgHdr object. This routine +** must be called after sqlite3PcacheFetch() in order to get a usable +** result. +*/ +SQLITE_PRIVATE PgHdr *sqlite3PcacheFetchFinish( + PCache *pCache, /* Obtain the page from this cache */ + Pgno pgno, /* Page number obtained */ + sqlite3_pcache_page *pPage /* Page obtained by prior PcacheFetch() call */ +){ + PgHdr *pPgHdr; + + assert( pPage!=0 ); + pPgHdr = (PgHdr *)pPage->pExtra; + + if( !pPgHdr->pPage ){ + return pcacheFetchFinishWithInit(pCache, pgno, pPage); + } + pCache->nRefSum++; + pPgHdr->nRef++; + assert( sqlite3PcachePageSanity(pPgHdr) ); + return pPgHdr; +} + +/* +** Decrement the reference count on a page. If the page is clean and the +** reference count drops to 0, then it is made eligible for recycling. +*/ +SQLITE_PRIVATE void SQLITE_NOINLINE sqlite3PcacheRelease(PgHdr *p){ + assert( p->nRef>0 ); + p->pCache->nRefSum--; + if( (--p->nRef)==0 ){ + if( p->flags&PGHDR_CLEAN ){ + pcacheUnpin(p); + }else{ + pcacheManageDirtyList(p, PCACHE_DIRTYLIST_FRONT); + } + } +} + +/* +** Increase the reference count of a supplied page by 1. +*/ +SQLITE_PRIVATE void sqlite3PcacheRef(PgHdr *p){ + assert(p->nRef>0); + assert( sqlite3PcachePageSanity(p) ); + p->nRef++; + p->pCache->nRefSum++; +} + +/* +** Drop a page from the cache. There must be exactly one reference to the +** page. This function deletes that reference, so after it returns the +** page pointed to by p is invalid. +*/ +SQLITE_PRIVATE void sqlite3PcacheDrop(PgHdr *p){ + assert( p->nRef==1 ); + assert( sqlite3PcachePageSanity(p) ); + if( p->flags&PGHDR_DIRTY ){ + pcacheManageDirtyList(p, PCACHE_DIRTYLIST_REMOVE); + } + p->pCache->nRefSum--; + sqlite3GlobalConfig.pcache2.xUnpin(p->pCache->pCache, p->pPage, 1); +} + +/* +** Make sure the page is marked as dirty. If it isn't dirty already, +** make it so. +*/ +SQLITE_PRIVATE void sqlite3PcacheMakeDirty(PgHdr *p){ + assert( p->nRef>0 ); + assert( sqlite3PcachePageSanity(p) ); + if( p->flags & (PGHDR_CLEAN|PGHDR_DONT_WRITE) ){ /*OPTIMIZATION-IF-FALSE*/ + p->flags &= ~PGHDR_DONT_WRITE; + if( p->flags & PGHDR_CLEAN ){ + p->flags ^= (PGHDR_DIRTY|PGHDR_CLEAN); + pcacheTrace(("%p.DIRTY %d\n",p->pCache,p->pgno)); + assert( (p->flags & (PGHDR_DIRTY|PGHDR_CLEAN))==PGHDR_DIRTY ); + pcacheManageDirtyList(p, PCACHE_DIRTYLIST_ADD); + } + assert( sqlite3PcachePageSanity(p) ); + } +} + +/* +** Make sure the page is marked as clean. If it isn't clean already, +** make it so. +*/ +SQLITE_PRIVATE void sqlite3PcacheMakeClean(PgHdr *p){ + assert( sqlite3PcachePageSanity(p) ); + assert( (p->flags & PGHDR_DIRTY)!=0 ); + assert( (p->flags & PGHDR_CLEAN)==0 ); + pcacheManageDirtyList(p, PCACHE_DIRTYLIST_REMOVE); + p->flags &= ~(PGHDR_DIRTY|PGHDR_NEED_SYNC|PGHDR_WRITEABLE); + p->flags |= PGHDR_CLEAN; + pcacheTrace(("%p.CLEAN %d\n",p->pCache,p->pgno)); + assert( sqlite3PcachePageSanity(p) ); + if( p->nRef==0 ){ + pcacheUnpin(p); + } +} + +/* +** Make every page in the cache clean. +*/ +SQLITE_PRIVATE void sqlite3PcacheCleanAll(PCache *pCache){ + PgHdr *p; + pcacheTrace(("%p.CLEAN-ALL\n",pCache)); + while( (p = pCache->pDirty)!=0 ){ + sqlite3PcacheMakeClean(p); + } +} + +/* +** Clear the PGHDR_NEED_SYNC and PGHDR_WRITEABLE flag from all dirty pages. +*/ +SQLITE_PRIVATE void sqlite3PcacheClearWritable(PCache *pCache){ + PgHdr *p; + pcacheTrace(("%p.CLEAR-WRITEABLE\n",pCache)); + for(p=pCache->pDirty; p; p=p->pDirtyNext){ + p->flags &= ~(PGHDR_NEED_SYNC|PGHDR_WRITEABLE); + } + pCache->pSynced = pCache->pDirtyTail; +} + +/* +** Clear the PGHDR_NEED_SYNC flag from all dirty pages. +*/ +SQLITE_PRIVATE void sqlite3PcacheClearSyncFlags(PCache *pCache){ + PgHdr *p; + for(p=pCache->pDirty; p; p=p->pDirtyNext){ + p->flags &= ~PGHDR_NEED_SYNC; + } + pCache->pSynced = pCache->pDirtyTail; +} + +/* +** Change the page number of page p to newPgno. +*/ +SQLITE_PRIVATE void sqlite3PcacheMove(PgHdr *p, Pgno newPgno){ + PCache *pCache = p->pCache; + assert( p->nRef>0 ); + assert( newPgno>0 ); + assert( sqlite3PcachePageSanity(p) ); + pcacheTrace(("%p.MOVE %d -> %d\n",pCache,p->pgno,newPgno)); + sqlite3GlobalConfig.pcache2.xRekey(pCache->pCache, p->pPage, p->pgno,newPgno); + p->pgno = newPgno; + if( (p->flags&PGHDR_DIRTY) && (p->flags&PGHDR_NEED_SYNC) ){ + pcacheManageDirtyList(p, PCACHE_DIRTYLIST_FRONT); + } +} + +/* +** Drop every cache entry whose page number is greater than "pgno". The +** caller must ensure that there are no outstanding references to any pages +** other than page 1 with a page number greater than pgno. +** +** If there is a reference to page 1 and the pgno parameter passed to this +** function is 0, then the data area associated with page 1 is zeroed, but +** the page object is not dropped. +*/ +SQLITE_PRIVATE void sqlite3PcacheTruncate(PCache *pCache, Pgno pgno){ + if( pCache->pCache ){ + PgHdr *p; + PgHdr *pNext; + pcacheTrace(("%p.TRUNCATE %d\n",pCache,pgno)); + for(p=pCache->pDirty; p; p=pNext){ + pNext = p->pDirtyNext; + /* This routine never gets call with a positive pgno except right + ** after sqlite3PcacheCleanAll(). So if there are dirty pages, + ** it must be that pgno==0. + */ + assert( p->pgno>0 ); + if( p->pgno>pgno ){ + assert( p->flags&PGHDR_DIRTY ); + sqlite3PcacheMakeClean(p); + } + } + if( pgno==0 && pCache->nRefSum ){ + sqlite3_pcache_page *pPage1; + pPage1 = sqlite3GlobalConfig.pcache2.xFetch(pCache->pCache,1,0); + if( ALWAYS(pPage1) ){ /* Page 1 is always available in cache, because + ** pCache->nRefSum>0 */ + memset(pPage1->pBuf, 0, pCache->szPage); + pgno = 1; + } + } + sqlite3GlobalConfig.pcache2.xTruncate(pCache->pCache, pgno+1); + } +} + +/* +** Close a cache. +*/ +SQLITE_PRIVATE void sqlite3PcacheClose(PCache *pCache){ + assert( pCache->pCache!=0 ); + pcacheTrace(("%p.CLOSE\n",pCache)); + sqlite3GlobalConfig.pcache2.xDestroy(pCache->pCache); +} + +/* +** Discard the contents of the cache. +*/ +SQLITE_PRIVATE void sqlite3PcacheClear(PCache *pCache){ + sqlite3PcacheTruncate(pCache, 0); +} + +/* +** Merge two lists of pages connected by pDirty and in pgno order. +** Do not bother fixing the pDirtyPrev pointers. +*/ +static PgHdr *pcacheMergeDirtyList(PgHdr *pA, PgHdr *pB){ + PgHdr result, *pTail; + pTail = &result; + assert( pA!=0 && pB!=0 ); + for(;;){ + if( pA->pgnopgno ){ + pTail->pDirty = pA; + pTail = pA; + pA = pA->pDirty; + if( pA==0 ){ + pTail->pDirty = pB; + break; + } + }else{ + pTail->pDirty = pB; + pTail = pB; + pB = pB->pDirty; + if( pB==0 ){ + pTail->pDirty = pA; + break; + } + } + } + return result.pDirty; +} + +/* +** Sort the list of pages in accending order by pgno. Pages are +** connected by pDirty pointers. The pDirtyPrev pointers are +** corrupted by this sort. +** +** Since there cannot be more than 2^31 distinct pages in a database, +** there cannot be more than 31 buckets required by the merge sorter. +** One extra bucket is added to catch overflow in case something +** ever changes to make the previous sentence incorrect. +*/ +#define N_SORT_BUCKET 32 +static PgHdr *pcacheSortDirtyList(PgHdr *pIn){ + PgHdr *a[N_SORT_BUCKET], *p; + int i; + memset(a, 0, sizeof(a)); + while( pIn ){ + p = pIn; + pIn = p->pDirty; + p->pDirty = 0; + for(i=0; ALWAYS(ipDirty; p; p=p->pDirtyNext){ + p->pDirty = p->pDirtyNext; + } + return pcacheSortDirtyList(pCache->pDirty); +} + +/* +** Return the total number of references to all pages held by the cache. +** +** This is not the total number of pages referenced, but the sum of the +** reference count for all pages. +*/ +SQLITE_PRIVATE int sqlite3PcacheRefCount(PCache *pCache){ + return pCache->nRefSum; +} + +/* +** Return the number of references to the page supplied as an argument. +*/ +SQLITE_PRIVATE int sqlite3PcachePageRefcount(PgHdr *p){ + return p->nRef; +} + +/* +** Return the total number of pages in the cache. +*/ +SQLITE_PRIVATE int sqlite3PcachePagecount(PCache *pCache){ + assert( pCache->pCache!=0 ); + return sqlite3GlobalConfig.pcache2.xPagecount(pCache->pCache); +} + +#ifdef SQLITE_TEST +/* +** Get the suggested cache-size value. +*/ +SQLITE_PRIVATE int sqlite3PcacheGetCachesize(PCache *pCache){ + return numberOfCachePages(pCache); +} +#endif + +/* +** Set the suggested cache-size value. +*/ +SQLITE_PRIVATE void sqlite3PcacheSetCachesize(PCache *pCache, int mxPage){ + assert( pCache->pCache!=0 ); + pCache->szCache = mxPage; + sqlite3GlobalConfig.pcache2.xCachesize(pCache->pCache, + numberOfCachePages(pCache)); +} + +/* +** Set the suggested cache-spill value. Make no changes if if the +** argument is zero. Return the effective cache-spill size, which will +** be the larger of the szSpill and szCache. +*/ +SQLITE_PRIVATE int sqlite3PcacheSetSpillsize(PCache *p, int mxPage){ + int res; + assert( p->pCache!=0 ); + if( mxPage ){ + if( mxPage<0 ){ + mxPage = (int)((-1024*(i64)mxPage)/(p->szPage+p->szExtra)); + } + p->szSpill = mxPage; + } + res = numberOfCachePages(p); + if( resszSpill ) res = p->szSpill; + return res; +} + +/* +** Free up as much memory as possible from the page cache. +*/ +SQLITE_PRIVATE void sqlite3PcacheShrink(PCache *pCache){ + assert( pCache->pCache!=0 ); + sqlite3GlobalConfig.pcache2.xShrink(pCache->pCache); +} + +/* +** Return the size of the header added by this middleware layer +** in the page-cache hierarchy. +*/ +SQLITE_PRIVATE int sqlite3HeaderSizePcache(void){ return ROUND8(sizeof(PgHdr)); } + +/* +** Return the number of dirty pages currently in the cache, as a percentage +** of the configured cache size. +*/ +SQLITE_PRIVATE int sqlite3PCachePercentDirty(PCache *pCache){ + PgHdr *pDirty; + int nDirty = 0; + int nCache = numberOfCachePages(pCache); + for(pDirty=pCache->pDirty; pDirty; pDirty=pDirty->pDirtyNext) nDirty++; + return nCache ? (int)(((i64)nDirty * 100) / nCache) : 0; +} + +#ifdef SQLITE_DIRECT_OVERFLOW_READ +/* +** Return true if there are one or more dirty pages in the cache. Else false. +*/ +SQLITE_PRIVATE int sqlite3PCacheIsDirty(PCache *pCache){ + return (pCache->pDirty!=0); +} +#endif + +#if defined(SQLITE_CHECK_PAGES) || defined(SQLITE_DEBUG) +/* +** For all dirty pages currently in the cache, invoke the specified +** callback. This is only used if the SQLITE_CHECK_PAGES macro is +** defined. +*/ +SQLITE_PRIVATE void sqlite3PcacheIterateDirty(PCache *pCache, void (*xIter)(PgHdr *)){ + PgHdr *pDirty; + for(pDirty=pCache->pDirty; pDirty; pDirty=pDirty->pDirtyNext){ + xIter(pDirty); + } +} +#endif + +/************** End of pcache.c **********************************************/ +/************** Begin file pcache1.c *****************************************/ +/* +** 2008 November 05 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** +** This file implements the default page cache implementation (the +** sqlite3_pcache interface). It also contains part of the implementation +** of the SQLITE_CONFIG_PAGECACHE and sqlite3_release_memory() features. +** If the default page cache implementation is overridden, then neither of +** these two features are available. +** +** A Page cache line looks like this: +** +** ------------------------------------------------------------- +** | database page content | PgHdr1 | MemPage | PgHdr | +** ------------------------------------------------------------- +** +** The database page content is up front (so that buffer overreads tend to +** flow harmlessly into the PgHdr1, MemPage, and PgHdr extensions). MemPage +** is the extension added by the btree.c module containing information such +** as the database page number and how that database page is used. PgHdr +** is added by the pcache.c layer and contains information used to keep track +** of which pages are "dirty". PgHdr1 is an extension added by this +** module (pcache1.c). The PgHdr1 header is a subclass of sqlite3_pcache_page. +** PgHdr1 contains information needed to look up a page by its page number. +** The superclass sqlite3_pcache_page.pBuf points to the start of the +** database page content and sqlite3_pcache_page.pExtra points to PgHdr. +** +** The size of the extension (MemPage+PgHdr+PgHdr1) can be determined at +** runtime using sqlite3_config(SQLITE_CONFIG_PCACHE_HDRSZ, &size). The +** sizes of the extensions sum to 272 bytes on x64 for 3.8.10, but this +** size can vary according to architecture, compile-time options, and +** SQLite library version number. +** +** If SQLITE_PCACHE_SEPARATE_HEADER is defined, then the extension is obtained +** using a separate memory allocation from the database page content. This +** seeks to overcome the "clownshoe" problem (also called "internal +** fragmentation" in academic literature) of allocating a few bytes more +** than a power of two with the memory allocator rounding up to the next +** power of two, and leaving the rounded-up space unused. +** +** This module tracks pointers to PgHdr1 objects. Only pcache.c communicates +** with this module. Information is passed back and forth as PgHdr1 pointers. +** +** The pcache.c and pager.c modules deal pointers to PgHdr objects. +** The btree.c module deals with pointers to MemPage objects. +** +** SOURCE OF PAGE CACHE MEMORY: +** +** Memory for a page might come from any of three sources: +** +** (1) The general-purpose memory allocator - sqlite3Malloc() +** (2) Global page-cache memory provided using sqlite3_config() with +** SQLITE_CONFIG_PAGECACHE. +** (3) PCache-local bulk allocation. +** +** The third case is a chunk of heap memory (defaulting to 100 pages worth) +** that is allocated when the page cache is created. The size of the local +** bulk allocation can be adjusted using +** +** sqlite3_config(SQLITE_CONFIG_PAGECACHE, (void*)0, 0, N). +** +** If N is positive, then N pages worth of memory are allocated using a single +** sqlite3Malloc() call and that memory is used for the first N pages allocated. +** Or if N is negative, then -1024*N bytes of memory are allocated and used +** for as many pages as can be accomodated. +** +** Only one of (2) or (3) can be used. Once the memory available to (2) or +** (3) is exhausted, subsequent allocations fail over to the general-purpose +** memory allocator (1). +** +** Earlier versions of SQLite used only methods (1) and (2). But experiments +** show that method (3) with N==100 provides about a 5% performance boost for +** common workloads. +*/ +/* #include "sqliteInt.h" */ + +typedef struct PCache1 PCache1; +typedef struct PgHdr1 PgHdr1; +typedef struct PgFreeslot PgFreeslot; +typedef struct PGroup PGroup; + +/* +** Each cache entry is represented by an instance of the following +** structure. Unless SQLITE_PCACHE_SEPARATE_HEADER is defined, a buffer of +** PgHdr1.pCache->szPage bytes is allocated directly before this structure +** in memory. +** +** Note: Variables isBulkLocal and isAnchor were once type "u8". That works, +** but causes a 2-byte gap in the structure for most architectures (since +** pointers must be either 4 or 8-byte aligned). As this structure is located +** in memory directly after the associated page data, if the database is +** corrupt, code at the b-tree layer may overread the page buffer and +** read part of this structure before the corruption is detected. This +** can cause a valgrind error if the unitialized gap is accessed. Using u16 +** ensures there is no such gap, and therefore no bytes of unitialized memory +** in the structure. +*/ +struct PgHdr1 { + sqlite3_pcache_page page; /* Base class. Must be first. pBuf & pExtra */ + unsigned int iKey; /* Key value (page number) */ + u16 isBulkLocal; /* This page from bulk local storage */ + u16 isAnchor; /* This is the PGroup.lru element */ + PgHdr1 *pNext; /* Next in hash table chain */ + PCache1 *pCache; /* Cache that currently owns this page */ + PgHdr1 *pLruNext; /* Next in LRU list of unpinned pages */ + PgHdr1 *pLruPrev; /* Previous in LRU list of unpinned pages */ + /* NB: pLruPrev is only valid if pLruNext!=0 */ +}; + +/* +** A page is pinned if it is not on the LRU list. To be "pinned" means +** that the page is in active use and must not be deallocated. +*/ +#define PAGE_IS_PINNED(p) ((p)->pLruNext==0) +#define PAGE_IS_UNPINNED(p) ((p)->pLruNext!=0) + +/* Each page cache (or PCache) belongs to a PGroup. A PGroup is a set +** of one or more PCaches that are able to recycle each other's unpinned +** pages when they are under memory pressure. A PGroup is an instance of +** the following object. +** +** This page cache implementation works in one of two modes: +** +** (1) Every PCache is the sole member of its own PGroup. There is +** one PGroup per PCache. +** +** (2) There is a single global PGroup that all PCaches are a member +** of. +** +** Mode 1 uses more memory (since PCache instances are not able to rob +** unused pages from other PCaches) but it also operates without a mutex, +** and is therefore often faster. Mode 2 requires a mutex in order to be +** threadsafe, but recycles pages more efficiently. +** +** For mode (1), PGroup.mutex is NULL. For mode (2) there is only a single +** PGroup which is the pcache1.grp global variable and its mutex is +** SQLITE_MUTEX_STATIC_LRU. +*/ +struct PGroup { + sqlite3_mutex *mutex; /* MUTEX_STATIC_LRU or NULL */ + unsigned int nMaxPage; /* Sum of nMax for purgeable caches */ + unsigned int nMinPage; /* Sum of nMin for purgeable caches */ + unsigned int mxPinned; /* nMaxpage + 10 - nMinPage */ + unsigned int nPurgeable; /* Number of purgeable pages allocated */ + PgHdr1 lru; /* The beginning and end of the LRU list */ +}; + +/* Each page cache is an instance of the following object. Every +** open database file (including each in-memory database and each +** temporary or transient database) has a single page cache which +** is an instance of this object. +** +** Pointers to structures of this type are cast and returned as +** opaque sqlite3_pcache* handles. +*/ +struct PCache1 { + /* Cache configuration parameters. Page size (szPage) and the purgeable + ** flag (bPurgeable) and the pnPurgeable pointer are all set when the + ** cache is created and are never changed thereafter. nMax may be + ** modified at any time by a call to the pcache1Cachesize() method. + ** The PGroup mutex must be held when accessing nMax. + */ + PGroup *pGroup; /* PGroup this cache belongs to */ + unsigned int *pnPurgeable; /* Pointer to pGroup->nPurgeable */ + int szPage; /* Size of database content section */ + int szExtra; /* sizeof(MemPage)+sizeof(PgHdr) */ + int szAlloc; /* Total size of one pcache line */ + int bPurgeable; /* True if cache is purgeable */ + unsigned int nMin; /* Minimum number of pages reserved */ + unsigned int nMax; /* Configured "cache_size" value */ + unsigned int n90pct; /* nMax*9/10 */ + unsigned int iMaxKey; /* Largest key seen since xTruncate() */ + unsigned int nPurgeableDummy; /* pnPurgeable points here when not used*/ + + /* Hash table of all pages. The following variables may only be accessed + ** when the accessor is holding the PGroup mutex. + */ + unsigned int nRecyclable; /* Number of pages in the LRU list */ + unsigned int nPage; /* Total number of pages in apHash */ + unsigned int nHash; /* Number of slots in apHash[] */ + PgHdr1 **apHash; /* Hash table for fast lookup by key */ + PgHdr1 *pFree; /* List of unused pcache-local pages */ + void *pBulk; /* Bulk memory used by pcache-local */ +}; + +/* +** Free slots in the allocator used to divide up the global page cache +** buffer provided using the SQLITE_CONFIG_PAGECACHE mechanism. +*/ +struct PgFreeslot { + PgFreeslot *pNext; /* Next free slot */ +}; + +/* +** Global data used by this cache. +*/ +static SQLITE_WSD struct PCacheGlobal { + PGroup grp; /* The global PGroup for mode (2) */ + + /* Variables related to SQLITE_CONFIG_PAGECACHE settings. The + ** szSlot, nSlot, pStart, pEnd, nReserve, and isInit values are all + ** fixed at sqlite3_initialize() time and do not require mutex protection. + ** The nFreeSlot and pFree values do require mutex protection. + */ + int isInit; /* True if initialized */ + int separateCache; /* Use a new PGroup for each PCache */ + int nInitPage; /* Initial bulk allocation size */ + int szSlot; /* Size of each free slot */ + int nSlot; /* The number of pcache slots */ + int nReserve; /* Try to keep nFreeSlot above this */ + void *pStart, *pEnd; /* Bounds of global page cache memory */ + /* Above requires no mutex. Use mutex below for variable that follow. */ + sqlite3_mutex *mutex; /* Mutex for accessing the following: */ + PgFreeslot *pFree; /* Free page blocks */ + int nFreeSlot; /* Number of unused pcache slots */ + /* The following value requires a mutex to change. We skip the mutex on + ** reading because (1) most platforms read a 32-bit integer atomically and + ** (2) even if an incorrect value is read, no great harm is done since this + ** is really just an optimization. */ + int bUnderPressure; /* True if low on PAGECACHE memory */ +} pcache1_g; + +/* +** All code in this file should access the global structure above via the +** alias "pcache1". This ensures that the WSD emulation is used when +** compiling for systems that do not support real WSD. +*/ +#define pcache1 (GLOBAL(struct PCacheGlobal, pcache1_g)) + +/* +** Macros to enter and leave the PCache LRU mutex. +*/ +#if !defined(SQLITE_ENABLE_MEMORY_MANAGEMENT) || SQLITE_THREADSAFE==0 +# define pcache1EnterMutex(X) assert((X)->mutex==0) +# define pcache1LeaveMutex(X) assert((X)->mutex==0) +# define PCACHE1_MIGHT_USE_GROUP_MUTEX 0 +#else +# define pcache1EnterMutex(X) sqlite3_mutex_enter((X)->mutex) +# define pcache1LeaveMutex(X) sqlite3_mutex_leave((X)->mutex) +# define PCACHE1_MIGHT_USE_GROUP_MUTEX 1 +#endif + +/******************************************************************************/ +/******** Page Allocation/SQLITE_CONFIG_PCACHE Related Functions **************/ + + +/* +** This function is called during initialization if a static buffer is +** supplied to use for the page-cache by passing the SQLITE_CONFIG_PAGECACHE +** verb to sqlite3_config(). Parameter pBuf points to an allocation large +** enough to contain 'n' buffers of 'sz' bytes each. +** +** This routine is called from sqlite3_initialize() and so it is guaranteed +** to be serialized already. There is no need for further mutexing. +*/ +SQLITE_PRIVATE void sqlite3PCacheBufferSetup(void *pBuf, int sz, int n){ + if( pcache1.isInit ){ + PgFreeslot *p; + if( pBuf==0 ) sz = n = 0; + if( n==0 ) sz = 0; + sz = ROUNDDOWN8(sz); + pcache1.szSlot = sz; + pcache1.nSlot = pcache1.nFreeSlot = n; + pcache1.nReserve = n>90 ? 10 : (n/10 + 1); + pcache1.pStart = pBuf; + pcache1.pFree = 0; + pcache1.bUnderPressure = 0; + while( n-- ){ + p = (PgFreeslot*)pBuf; + p->pNext = pcache1.pFree; + pcache1.pFree = p; + pBuf = (void*)&((char*)pBuf)[sz]; + } + pcache1.pEnd = pBuf; + } +} + +/* +** Try to initialize the pCache->pFree and pCache->pBulk fields. Return +** true if pCache->pFree ends up containing one or more free pages. +*/ +static int pcache1InitBulk(PCache1 *pCache){ + i64 szBulk; + char *zBulk; + if( pcache1.nInitPage==0 ) return 0; + /* Do not bother with a bulk allocation if the cache size very small */ + if( pCache->nMax<3 ) return 0; + sqlite3BeginBenignMalloc(); + if( pcache1.nInitPage>0 ){ + szBulk = pCache->szAlloc * (i64)pcache1.nInitPage; + }else{ + szBulk = -1024 * (i64)pcache1.nInitPage; + } + if( szBulk > pCache->szAlloc*(i64)pCache->nMax ){ + szBulk = pCache->szAlloc*(i64)pCache->nMax; + } + zBulk = pCache->pBulk = sqlite3Malloc( szBulk ); + sqlite3EndBenignMalloc(); + if( zBulk ){ + int nBulk = sqlite3MallocSize(zBulk)/pCache->szAlloc; + do{ + PgHdr1 *pX = (PgHdr1*)&zBulk[pCache->szPage]; + pX->page.pBuf = zBulk; + pX->page.pExtra = &pX[1]; + pX->isBulkLocal = 1; + pX->isAnchor = 0; + pX->pNext = pCache->pFree; + pX->pLruPrev = 0; /* Initializing this saves a valgrind error */ + pCache->pFree = pX; + zBulk += pCache->szAlloc; + }while( --nBulk ); + } + return pCache->pFree!=0; +} + +/* +** Malloc function used within this file to allocate space from the buffer +** configured using sqlite3_config(SQLITE_CONFIG_PAGECACHE) option. If no +** such buffer exists or there is no space left in it, this function falls +** back to sqlite3Malloc(). +** +** Multiple threads can run this routine at the same time. Global variables +** in pcache1 need to be protected via mutex. +*/ +static void *pcache1Alloc(int nByte){ + void *p = 0; + assert( sqlite3_mutex_notheld(pcache1.grp.mutex) ); + if( nByte<=pcache1.szSlot ){ + sqlite3_mutex_enter(pcache1.mutex); + p = (PgHdr1 *)pcache1.pFree; + if( p ){ + pcache1.pFree = pcache1.pFree->pNext; + pcache1.nFreeSlot--; + pcache1.bUnderPressure = pcache1.nFreeSlot=0 ); + sqlite3StatusHighwater(SQLITE_STATUS_PAGECACHE_SIZE, nByte); + sqlite3StatusUp(SQLITE_STATUS_PAGECACHE_USED, 1); + } + sqlite3_mutex_leave(pcache1.mutex); + } + if( p==0 ){ + /* Memory is not available in the SQLITE_CONFIG_PAGECACHE pool. Get + ** it from sqlite3Malloc instead. + */ + p = sqlite3Malloc(nByte); +#ifndef SQLITE_DISABLE_PAGECACHE_OVERFLOW_STATS + if( p ){ + int sz = sqlite3MallocSize(p); + sqlite3_mutex_enter(pcache1.mutex); + sqlite3StatusHighwater(SQLITE_STATUS_PAGECACHE_SIZE, nByte); + sqlite3StatusUp(SQLITE_STATUS_PAGECACHE_OVERFLOW, sz); + sqlite3_mutex_leave(pcache1.mutex); + } +#endif + sqlite3MemdebugSetType(p, MEMTYPE_PCACHE); + } + return p; +} + +/* +** Free an allocated buffer obtained from pcache1Alloc(). +*/ +static void pcache1Free(void *p){ + if( p==0 ) return; + if( SQLITE_WITHIN(p, pcache1.pStart, pcache1.pEnd) ){ + PgFreeslot *pSlot; + sqlite3_mutex_enter(pcache1.mutex); + sqlite3StatusDown(SQLITE_STATUS_PAGECACHE_USED, 1); + pSlot = (PgFreeslot*)p; + pSlot->pNext = pcache1.pFree; + pcache1.pFree = pSlot; + pcache1.nFreeSlot++; + pcache1.bUnderPressure = pcache1.nFreeSlot=pcache1.pStart && ppGroup->mutex) ); + if( pCache->pFree || (pCache->nPage==0 && pcache1InitBulk(pCache)) ){ + assert( pCache->pFree!=0 ); + p = pCache->pFree; + pCache->pFree = p->pNext; + p->pNext = 0; + }else{ +#ifdef SQLITE_ENABLE_MEMORY_MANAGEMENT + /* The group mutex must be released before pcache1Alloc() is called. This + ** is because it might call sqlite3_release_memory(), which assumes that + ** this mutex is not held. */ + assert( pcache1.separateCache==0 ); + assert( pCache->pGroup==&pcache1.grp ); + pcache1LeaveMutex(pCache->pGroup); +#endif + if( benignMalloc ){ sqlite3BeginBenignMalloc(); } +#ifdef SQLITE_PCACHE_SEPARATE_HEADER + pPg = pcache1Alloc(pCache->szPage); + p = sqlite3Malloc(sizeof(PgHdr1) + pCache->szExtra); + if( !pPg || !p ){ + pcache1Free(pPg); + sqlite3_free(p); + pPg = 0; + } +#else + pPg = pcache1Alloc(pCache->szAlloc); + p = (PgHdr1 *)&((u8 *)pPg)[pCache->szPage]; +#endif + if( benignMalloc ){ sqlite3EndBenignMalloc(); } +#ifdef SQLITE_ENABLE_MEMORY_MANAGEMENT + pcache1EnterMutex(pCache->pGroup); +#endif + if( pPg==0 ) return 0; + p->page.pBuf = pPg; + p->page.pExtra = &p[1]; + p->isBulkLocal = 0; + p->isAnchor = 0; + } + (*pCache->pnPurgeable)++; + return p; +} + +/* +** Free a page object allocated by pcache1AllocPage(). +*/ +static void pcache1FreePage(PgHdr1 *p){ + PCache1 *pCache; + assert( p!=0 ); + pCache = p->pCache; + assert( sqlite3_mutex_held(p->pCache->pGroup->mutex) ); + if( p->isBulkLocal ){ + p->pNext = pCache->pFree; + pCache->pFree = p; + }else{ + pcache1Free(p->page.pBuf); +#ifdef SQLITE_PCACHE_SEPARATE_HEADER + sqlite3_free(p); +#endif + } + (*pCache->pnPurgeable)--; +} + +/* +** Malloc function used by SQLite to obtain space from the buffer configured +** using sqlite3_config(SQLITE_CONFIG_PAGECACHE) option. If no such buffer +** exists, this function falls back to sqlite3Malloc(). +*/ +SQLITE_PRIVATE void *sqlite3PageMalloc(int sz){ + assert( sz<=65536+8 ); /* These allocations are never very large */ + return pcache1Alloc(sz); +} + +/* +** Free an allocated buffer obtained from sqlite3PageMalloc(). +*/ +SQLITE_PRIVATE void sqlite3PageFree(void *p){ + pcache1Free(p); +} + + +/* +** Return true if it desirable to avoid allocating a new page cache +** entry. +** +** If memory was allocated specifically to the page cache using +** SQLITE_CONFIG_PAGECACHE but that memory has all been used, then +** it is desirable to avoid allocating a new page cache entry because +** presumably SQLITE_CONFIG_PAGECACHE was suppose to be sufficient +** for all page cache needs and we should not need to spill the +** allocation onto the heap. +** +** Or, the heap is used for all page cache memory but the heap is +** under memory pressure, then again it is desirable to avoid +** allocating a new page cache entry in order to avoid stressing +** the heap even further. +*/ +static int pcache1UnderMemoryPressure(PCache1 *pCache){ + if( pcache1.nSlot && (pCache->szPage+pCache->szExtra)<=pcache1.szSlot ){ + return pcache1.bUnderPressure; + }else{ + return sqlite3HeapNearlyFull(); + } +} + +/******************************************************************************/ +/******** General Implementation Functions ************************************/ + +/* +** This function is used to resize the hash table used by the cache passed +** as the first argument. +** +** The PCache mutex must be held when this function is called. +*/ +static void pcache1ResizeHash(PCache1 *p){ + PgHdr1 **apNew; + unsigned int nNew; + unsigned int i; + + assert( sqlite3_mutex_held(p->pGroup->mutex) ); + + nNew = p->nHash*2; + if( nNew<256 ){ + nNew = 256; + } + + pcache1LeaveMutex(p->pGroup); + if( p->nHash ){ sqlite3BeginBenignMalloc(); } + apNew = (PgHdr1 **)sqlite3MallocZero(sizeof(PgHdr1 *)*nNew); + if( p->nHash ){ sqlite3EndBenignMalloc(); } + pcache1EnterMutex(p->pGroup); + if( apNew ){ + for(i=0; inHash; i++){ + PgHdr1 *pPage; + PgHdr1 *pNext = p->apHash[i]; + while( (pPage = pNext)!=0 ){ + unsigned int h = pPage->iKey % nNew; + pNext = pPage->pNext; + pPage->pNext = apNew[h]; + apNew[h] = pPage; + } + } + sqlite3_free(p->apHash); + p->apHash = apNew; + p->nHash = nNew; + } +} + +/* +** This function is used internally to remove the page pPage from the +** PGroup LRU list, if is part of it. If pPage is not part of the PGroup +** LRU list, then this function is a no-op. +** +** The PGroup mutex must be held when this function is called. +*/ +static PgHdr1 *pcache1PinPage(PgHdr1 *pPage){ + assert( pPage!=0 ); + assert( PAGE_IS_UNPINNED(pPage) ); + assert( pPage->pLruNext ); + assert( pPage->pLruPrev ); + assert( sqlite3_mutex_held(pPage->pCache->pGroup->mutex) ); + pPage->pLruPrev->pLruNext = pPage->pLruNext; + pPage->pLruNext->pLruPrev = pPage->pLruPrev; + pPage->pLruNext = 0; + /* pPage->pLruPrev = 0; + ** No need to clear pLruPrev as it is never accessed if pLruNext is 0 */ + assert( pPage->isAnchor==0 ); + assert( pPage->pCache->pGroup->lru.isAnchor==1 ); + pPage->pCache->nRecyclable--; + return pPage; +} + + +/* +** Remove the page supplied as an argument from the hash table +** (PCache1.apHash structure) that it is currently stored in. +** Also free the page if freePage is true. +** +** The PGroup mutex must be held when this function is called. +*/ +static void pcache1RemoveFromHash(PgHdr1 *pPage, int freeFlag){ + unsigned int h; + PCache1 *pCache = pPage->pCache; + PgHdr1 **pp; + + assert( sqlite3_mutex_held(pCache->pGroup->mutex) ); + h = pPage->iKey % pCache->nHash; + for(pp=&pCache->apHash[h]; (*pp)!=pPage; pp=&(*pp)->pNext); + *pp = (*pp)->pNext; + + pCache->nPage--; + if( freeFlag ) pcache1FreePage(pPage); +} + +/* +** If there are currently more than nMaxPage pages allocated, try +** to recycle pages to reduce the number allocated to nMaxPage. +*/ +static void pcache1EnforceMaxPage(PCache1 *pCache){ + PGroup *pGroup = pCache->pGroup; + PgHdr1 *p; + assert( sqlite3_mutex_held(pGroup->mutex) ); + while( pGroup->nPurgeable>pGroup->nMaxPage + && (p=pGroup->lru.pLruPrev)->isAnchor==0 + ){ + assert( p->pCache->pGroup==pGroup ); + assert( PAGE_IS_UNPINNED(p) ); + pcache1PinPage(p); + pcache1RemoveFromHash(p, 1); + } + if( pCache->nPage==0 && pCache->pBulk ){ + sqlite3_free(pCache->pBulk); + pCache->pBulk = pCache->pFree = 0; + } +} + +/* +** Discard all pages from cache pCache with a page number (key value) +** greater than or equal to iLimit. Any pinned pages that meet this +** criteria are unpinned before they are discarded. +** +** The PCache mutex must be held when this function is called. +*/ +static void pcache1TruncateUnsafe( + PCache1 *pCache, /* The cache to truncate */ + unsigned int iLimit /* Drop pages with this pgno or larger */ +){ + TESTONLY( int nPage = 0; ) /* To assert pCache->nPage is correct */ + unsigned int h, iStop; + assert( sqlite3_mutex_held(pCache->pGroup->mutex) ); + assert( pCache->iMaxKey >= iLimit ); + assert( pCache->nHash > 0 ); + if( pCache->iMaxKey - iLimit < pCache->nHash ){ + /* If we are just shaving the last few pages off the end of the + ** cache, then there is no point in scanning the entire hash table. + ** Only scan those hash slots that might contain pages that need to + ** be removed. */ + h = iLimit % pCache->nHash; + iStop = pCache->iMaxKey % pCache->nHash; + TESTONLY( nPage = -10; ) /* Disable the pCache->nPage validity check */ + }else{ + /* This is the general case where many pages are being removed. + ** It is necessary to scan the entire hash table */ + h = pCache->nHash/2; + iStop = h - 1; + } + for(;;){ + PgHdr1 **pp; + PgHdr1 *pPage; + assert( hnHash ); + pp = &pCache->apHash[h]; + while( (pPage = *pp)!=0 ){ + if( pPage->iKey>=iLimit ){ + pCache->nPage--; + *pp = pPage->pNext; + if( PAGE_IS_UNPINNED(pPage) ) pcache1PinPage(pPage); + pcache1FreePage(pPage); + }else{ + pp = &pPage->pNext; + TESTONLY( if( nPage>=0 ) nPage++; ) + } + } + if( h==iStop ) break; + h = (h+1) % pCache->nHash; + } + assert( nPage<0 || pCache->nPage==(unsigned)nPage ); +} + +/******************************************************************************/ +/******** sqlite3_pcache Methods **********************************************/ + +/* +** Implementation of the sqlite3_pcache.xInit method. +*/ +static int pcache1Init(void *NotUsed){ + UNUSED_PARAMETER(NotUsed); + assert( pcache1.isInit==0 ); + memset(&pcache1, 0, sizeof(pcache1)); + + + /* + ** The pcache1.separateCache variable is true if each PCache has its own + ** private PGroup (mode-1). pcache1.separateCache is false if the single + ** PGroup in pcache1.grp is used for all page caches (mode-2). + ** + ** * Always use a unified cache (mode-2) if ENABLE_MEMORY_MANAGEMENT + ** + ** * Use a unified cache in single-threaded applications that have + ** configured a start-time buffer for use as page-cache memory using + ** sqlite3_config(SQLITE_CONFIG_PAGECACHE, pBuf, sz, N) with non-NULL + ** pBuf argument. + ** + ** * Otherwise use separate caches (mode-1) + */ +#if defined(SQLITE_ENABLE_MEMORY_MANAGEMENT) + pcache1.separateCache = 0; +#elif SQLITE_THREADSAFE + pcache1.separateCache = sqlite3GlobalConfig.pPage==0 + || sqlite3GlobalConfig.bCoreMutex>0; +#else + pcache1.separateCache = sqlite3GlobalConfig.pPage==0; +#endif + +#if SQLITE_THREADSAFE + if( sqlite3GlobalConfig.bCoreMutex ){ + pcache1.grp.mutex = sqlite3MutexAlloc(SQLITE_MUTEX_STATIC_LRU); + pcache1.mutex = sqlite3MutexAlloc(SQLITE_MUTEX_STATIC_PMEM); + } +#endif + if( pcache1.separateCache + && sqlite3GlobalConfig.nPage!=0 + && sqlite3GlobalConfig.pPage==0 + ){ + pcache1.nInitPage = sqlite3GlobalConfig.nPage; + }else{ + pcache1.nInitPage = 0; + } + pcache1.grp.mxPinned = 10; + pcache1.isInit = 1; + return SQLITE_OK; +} + +/* +** Implementation of the sqlite3_pcache.xShutdown method. +** Note that the static mutex allocated in xInit does +** not need to be freed. +*/ +static void pcache1Shutdown(void *NotUsed){ + UNUSED_PARAMETER(NotUsed); + assert( pcache1.isInit!=0 ); + memset(&pcache1, 0, sizeof(pcache1)); +} + +/* forward declaration */ +static void pcache1Destroy(sqlite3_pcache *p); + +/* +** Implementation of the sqlite3_pcache.xCreate method. +** +** Allocate a new cache. +*/ +static sqlite3_pcache *pcache1Create(int szPage, int szExtra, int bPurgeable){ + PCache1 *pCache; /* The newly created page cache */ + PGroup *pGroup; /* The group the new page cache will belong to */ + int sz; /* Bytes of memory required to allocate the new cache */ + + assert( (szPage & (szPage-1))==0 && szPage>=512 && szPage<=65536 ); + assert( szExtra < 300 ); + + sz = sizeof(PCache1) + sizeof(PGroup)*pcache1.separateCache; + pCache = (PCache1 *)sqlite3MallocZero(sz); + if( pCache ){ + if( pcache1.separateCache ){ + pGroup = (PGroup*)&pCache[1]; + pGroup->mxPinned = 10; + }else{ + pGroup = &pcache1.grp; + } + pcache1EnterMutex(pGroup); + if( pGroup->lru.isAnchor==0 ){ + pGroup->lru.isAnchor = 1; + pGroup->lru.pLruPrev = pGroup->lru.pLruNext = &pGroup->lru; + } + pCache->pGroup = pGroup; + pCache->szPage = szPage; + pCache->szExtra = szExtra; + pCache->szAlloc = szPage + szExtra + ROUND8(sizeof(PgHdr1)); + pCache->bPurgeable = (bPurgeable ? 1 : 0); + pcache1ResizeHash(pCache); + if( bPurgeable ){ + pCache->nMin = 10; + pGroup->nMinPage += pCache->nMin; + pGroup->mxPinned = pGroup->nMaxPage + 10 - pGroup->nMinPage; + pCache->pnPurgeable = &pGroup->nPurgeable; + }else{ + pCache->pnPurgeable = &pCache->nPurgeableDummy; + } + pcache1LeaveMutex(pGroup); + if( pCache->nHash==0 ){ + pcache1Destroy((sqlite3_pcache*)pCache); + pCache = 0; + } + } + return (sqlite3_pcache *)pCache; +} + +/* +** Implementation of the sqlite3_pcache.xCachesize method. +** +** Configure the cache_size limit for a cache. +*/ +static void pcache1Cachesize(sqlite3_pcache *p, int nMax){ + PCache1 *pCache = (PCache1 *)p; + if( pCache->bPurgeable ){ + PGroup *pGroup = pCache->pGroup; + pcache1EnterMutex(pGroup); + pGroup->nMaxPage += (nMax - pCache->nMax); + pGroup->mxPinned = pGroup->nMaxPage + 10 - pGroup->nMinPage; + pCache->nMax = nMax; + pCache->n90pct = pCache->nMax*9/10; + pcache1EnforceMaxPage(pCache); + pcache1LeaveMutex(pGroup); + } +} + +/* +** Implementation of the sqlite3_pcache.xShrink method. +** +** Free up as much memory as possible. +*/ +static void pcache1Shrink(sqlite3_pcache *p){ + PCache1 *pCache = (PCache1*)p; + if( pCache->bPurgeable ){ + PGroup *pGroup = pCache->pGroup; + int savedMaxPage; + pcache1EnterMutex(pGroup); + savedMaxPage = pGroup->nMaxPage; + pGroup->nMaxPage = 0; + pcache1EnforceMaxPage(pCache); + pGroup->nMaxPage = savedMaxPage; + pcache1LeaveMutex(pGroup); + } +} + +/* +** Implementation of the sqlite3_pcache.xPagecount method. +*/ +static int pcache1Pagecount(sqlite3_pcache *p){ + int n; + PCache1 *pCache = (PCache1*)p; + pcache1EnterMutex(pCache->pGroup); + n = pCache->nPage; + pcache1LeaveMutex(pCache->pGroup); + return n; +} + + +/* +** Implement steps 3, 4, and 5 of the pcache1Fetch() algorithm described +** in the header of the pcache1Fetch() procedure. +** +** This steps are broken out into a separate procedure because they are +** usually not needed, and by avoiding the stack initialization required +** for these steps, the main pcache1Fetch() procedure can run faster. +*/ +static SQLITE_NOINLINE PgHdr1 *pcache1FetchStage2( + PCache1 *pCache, + unsigned int iKey, + int createFlag +){ + unsigned int nPinned; + PGroup *pGroup = pCache->pGroup; + PgHdr1 *pPage = 0; + + /* Step 3: Abort if createFlag is 1 but the cache is nearly full */ + assert( pCache->nPage >= pCache->nRecyclable ); + nPinned = pCache->nPage - pCache->nRecyclable; + assert( pGroup->mxPinned == pGroup->nMaxPage + 10 - pGroup->nMinPage ); + assert( pCache->n90pct == pCache->nMax*9/10 ); + if( createFlag==1 && ( + nPinned>=pGroup->mxPinned + || nPinned>=pCache->n90pct + || (pcache1UnderMemoryPressure(pCache) && pCache->nRecyclablenPage>=pCache->nHash ) pcache1ResizeHash(pCache); + assert( pCache->nHash>0 && pCache->apHash ); + + /* Step 4. Try to recycle a page. */ + if( pCache->bPurgeable + && !pGroup->lru.pLruPrev->isAnchor + && ((pCache->nPage+1>=pCache->nMax) || pcache1UnderMemoryPressure(pCache)) + ){ + PCache1 *pOther; + pPage = pGroup->lru.pLruPrev; + assert( PAGE_IS_UNPINNED(pPage) ); + pcache1RemoveFromHash(pPage, 0); + pcache1PinPage(pPage); + pOther = pPage->pCache; + if( pOther->szAlloc != pCache->szAlloc ){ + pcache1FreePage(pPage); + pPage = 0; + }else{ + pGroup->nPurgeable -= (pOther->bPurgeable - pCache->bPurgeable); + } + } + + /* Step 5. If a usable page buffer has still not been found, + ** attempt to allocate a new one. + */ + if( !pPage ){ + pPage = pcache1AllocPage(pCache, createFlag==1); + } + + if( pPage ){ + unsigned int h = iKey % pCache->nHash; + pCache->nPage++; + pPage->iKey = iKey; + pPage->pNext = pCache->apHash[h]; + pPage->pCache = pCache; + pPage->pLruNext = 0; + /* pPage->pLruPrev = 0; + ** No need to clear pLruPrev since it is not accessed when pLruNext==0 */ + *(void **)pPage->page.pExtra = 0; + pCache->apHash[h] = pPage; + if( iKey>pCache->iMaxKey ){ + pCache->iMaxKey = iKey; + } + } + return pPage; +} + +/* +** Implementation of the sqlite3_pcache.xFetch method. +** +** Fetch a page by key value. +** +** Whether or not a new page may be allocated by this function depends on +** the value of the createFlag argument. 0 means do not allocate a new +** page. 1 means allocate a new page if space is easily available. 2 +** means to try really hard to allocate a new page. +** +** For a non-purgeable cache (a cache used as the storage for an in-memory +** database) there is really no difference between createFlag 1 and 2. So +** the calling function (pcache.c) will never have a createFlag of 1 on +** a non-purgeable cache. +** +** There are three different approaches to obtaining space for a page, +** depending on the value of parameter createFlag (which may be 0, 1 or 2). +** +** 1. Regardless of the value of createFlag, the cache is searched for a +** copy of the requested page. If one is found, it is returned. +** +** 2. If createFlag==0 and the page is not already in the cache, NULL is +** returned. +** +** 3. If createFlag is 1, and the page is not already in the cache, then +** return NULL (do not allocate a new page) if any of the following +** conditions are true: +** +** (a) the number of pages pinned by the cache is greater than +** PCache1.nMax, or +** +** (b) the number of pages pinned by the cache is greater than +** the sum of nMax for all purgeable caches, less the sum of +** nMin for all other purgeable caches, or +** +** 4. If none of the first three conditions apply and the cache is marked +** as purgeable, and if one of the following is true: +** +** (a) The number of pages allocated for the cache is already +** PCache1.nMax, or +** +** (b) The number of pages allocated for all purgeable caches is +** already equal to or greater than the sum of nMax for all +** purgeable caches, +** +** (c) The system is under memory pressure and wants to avoid +** unnecessary pages cache entry allocations +** +** then attempt to recycle a page from the LRU list. If it is the right +** size, return the recycled buffer. Otherwise, free the buffer and +** proceed to step 5. +** +** 5. Otherwise, allocate and return a new page buffer. +** +** There are two versions of this routine. pcache1FetchWithMutex() is +** the general case. pcache1FetchNoMutex() is a faster implementation for +** the common case where pGroup->mutex is NULL. The pcache1Fetch() wrapper +** invokes the appropriate routine. +*/ +static PgHdr1 *pcache1FetchNoMutex( + sqlite3_pcache *p, + unsigned int iKey, + int createFlag +){ + PCache1 *pCache = (PCache1 *)p; + PgHdr1 *pPage = 0; + + /* Step 1: Search the hash table for an existing entry. */ + pPage = pCache->apHash[iKey % pCache->nHash]; + while( pPage && pPage->iKey!=iKey ){ pPage = pPage->pNext; } + + /* Step 2: If the page was found in the hash table, then return it. + ** If the page was not in the hash table and createFlag is 0, abort. + ** Otherwise (page not in hash and createFlag!=0) continue with + ** subsequent steps to try to create the page. */ + if( pPage ){ + if( PAGE_IS_UNPINNED(pPage) ){ + return pcache1PinPage(pPage); + }else{ + return pPage; + } + }else if( createFlag ){ + /* Steps 3, 4, and 5 implemented by this subroutine */ + return pcache1FetchStage2(pCache, iKey, createFlag); + }else{ + return 0; + } +} +#if PCACHE1_MIGHT_USE_GROUP_MUTEX +static PgHdr1 *pcache1FetchWithMutex( + sqlite3_pcache *p, + unsigned int iKey, + int createFlag +){ + PCache1 *pCache = (PCache1 *)p; + PgHdr1 *pPage; + + pcache1EnterMutex(pCache->pGroup); + pPage = pcache1FetchNoMutex(p, iKey, createFlag); + assert( pPage==0 || pCache->iMaxKey>=iKey ); + pcache1LeaveMutex(pCache->pGroup); + return pPage; +} +#endif +static sqlite3_pcache_page *pcache1Fetch( + sqlite3_pcache *p, + unsigned int iKey, + int createFlag +){ +#if PCACHE1_MIGHT_USE_GROUP_MUTEX || defined(SQLITE_DEBUG) + PCache1 *pCache = (PCache1 *)p; +#endif + + assert( offsetof(PgHdr1,page)==0 ); + assert( pCache->bPurgeable || createFlag!=1 ); + assert( pCache->bPurgeable || pCache->nMin==0 ); + assert( pCache->bPurgeable==0 || pCache->nMin==10 ); + assert( pCache->nMin==0 || pCache->bPurgeable ); + assert( pCache->nHash>0 ); +#if PCACHE1_MIGHT_USE_GROUP_MUTEX + if( pCache->pGroup->mutex ){ + return (sqlite3_pcache_page*)pcache1FetchWithMutex(p, iKey, createFlag); + }else +#endif + { + return (sqlite3_pcache_page*)pcache1FetchNoMutex(p, iKey, createFlag); + } +} + + +/* +** Implementation of the sqlite3_pcache.xUnpin method. +** +** Mark a page as unpinned (eligible for asynchronous recycling). +*/ +static void pcache1Unpin( + sqlite3_pcache *p, + sqlite3_pcache_page *pPg, + int reuseUnlikely +){ + PCache1 *pCache = (PCache1 *)p; + PgHdr1 *pPage = (PgHdr1 *)pPg; + PGroup *pGroup = pCache->pGroup; + + assert( pPage->pCache==pCache ); + pcache1EnterMutex(pGroup); + + /* It is an error to call this function if the page is already + ** part of the PGroup LRU list. + */ + assert( pPage->pLruNext==0 ); + assert( PAGE_IS_PINNED(pPage) ); + + if( reuseUnlikely || pGroup->nPurgeable>pGroup->nMaxPage ){ + pcache1RemoveFromHash(pPage, 1); + }else{ + /* Add the page to the PGroup LRU list. */ + PgHdr1 **ppFirst = &pGroup->lru.pLruNext; + pPage->pLruPrev = &pGroup->lru; + (pPage->pLruNext = *ppFirst)->pLruPrev = pPage; + *ppFirst = pPage; + pCache->nRecyclable++; + } + + pcache1LeaveMutex(pCache->pGroup); +} + +/* +** Implementation of the sqlite3_pcache.xRekey method. +*/ +static void pcache1Rekey( + sqlite3_pcache *p, + sqlite3_pcache_page *pPg, + unsigned int iOld, + unsigned int iNew +){ + PCache1 *pCache = (PCache1 *)p; + PgHdr1 *pPage = (PgHdr1 *)pPg; + PgHdr1 **pp; + unsigned int h; + assert( pPage->iKey==iOld ); + assert( pPage->pCache==pCache ); + + pcache1EnterMutex(pCache->pGroup); + + h = iOld%pCache->nHash; + pp = &pCache->apHash[h]; + while( (*pp)!=pPage ){ + pp = &(*pp)->pNext; + } + *pp = pPage->pNext; + + h = iNew%pCache->nHash; + pPage->iKey = iNew; + pPage->pNext = pCache->apHash[h]; + pCache->apHash[h] = pPage; + if( iNew>pCache->iMaxKey ){ + pCache->iMaxKey = iNew; + } + + pcache1LeaveMutex(pCache->pGroup); +} + +/* +** Implementation of the sqlite3_pcache.xTruncate method. +** +** Discard all unpinned pages in the cache with a page number equal to +** or greater than parameter iLimit. Any pinned pages with a page number +** equal to or greater than iLimit are implicitly unpinned. +*/ +static void pcache1Truncate(sqlite3_pcache *p, unsigned int iLimit){ + PCache1 *pCache = (PCache1 *)p; + pcache1EnterMutex(pCache->pGroup); + if( iLimit<=pCache->iMaxKey ){ + pcache1TruncateUnsafe(pCache, iLimit); + pCache->iMaxKey = iLimit-1; + } + pcache1LeaveMutex(pCache->pGroup); +} + +/* +** Implementation of the sqlite3_pcache.xDestroy method. +** +** Destroy a cache allocated using pcache1Create(). +*/ +static void pcache1Destroy(sqlite3_pcache *p){ + PCache1 *pCache = (PCache1 *)p; + PGroup *pGroup = pCache->pGroup; + assert( pCache->bPurgeable || (pCache->nMax==0 && pCache->nMin==0) ); + pcache1EnterMutex(pGroup); + if( pCache->nPage ) pcache1TruncateUnsafe(pCache, 0); + assert( pGroup->nMaxPage >= pCache->nMax ); + pGroup->nMaxPage -= pCache->nMax; + assert( pGroup->nMinPage >= pCache->nMin ); + pGroup->nMinPage -= pCache->nMin; + pGroup->mxPinned = pGroup->nMaxPage + 10 - pGroup->nMinPage; + pcache1EnforceMaxPage(pCache); + pcache1LeaveMutex(pGroup); + sqlite3_free(pCache->pBulk); + sqlite3_free(pCache->apHash); + sqlite3_free(pCache); +} + +/* +** This function is called during initialization (sqlite3_initialize()) to +** install the default pluggable cache module, assuming the user has not +** already provided an alternative. +*/ +SQLITE_PRIVATE void sqlite3PCacheSetDefault(void){ + static const sqlite3_pcache_methods2 defaultMethods = { + 1, /* iVersion */ + 0, /* pArg */ + pcache1Init, /* xInit */ + pcache1Shutdown, /* xShutdown */ + pcache1Create, /* xCreate */ + pcache1Cachesize, /* xCachesize */ + pcache1Pagecount, /* xPagecount */ + pcache1Fetch, /* xFetch */ + pcache1Unpin, /* xUnpin */ + pcache1Rekey, /* xRekey */ + pcache1Truncate, /* xTruncate */ + pcache1Destroy, /* xDestroy */ + pcache1Shrink /* xShrink */ + }; + sqlite3_config(SQLITE_CONFIG_PCACHE2, &defaultMethods); +} + +/* +** Return the size of the header on each page of this PCACHE implementation. +*/ +SQLITE_PRIVATE int sqlite3HeaderSizePcache1(void){ return ROUND8(sizeof(PgHdr1)); } + +/* +** Return the global mutex used by this PCACHE implementation. The +** sqlite3_status() routine needs access to this mutex. +*/ +SQLITE_PRIVATE sqlite3_mutex *sqlite3Pcache1Mutex(void){ + return pcache1.mutex; +} + +#ifdef SQLITE_ENABLE_MEMORY_MANAGEMENT +/* +** This function is called to free superfluous dynamically allocated memory +** held by the pager system. Memory in use by any SQLite pager allocated +** by the current thread may be sqlite3_free()ed. +** +** nReq is the number of bytes of memory required. Once this much has +** been released, the function returns. The return value is the total number +** of bytes of memory released. +*/ +SQLITE_PRIVATE int sqlite3PcacheReleaseMemory(int nReq){ + int nFree = 0; + assert( sqlite3_mutex_notheld(pcache1.grp.mutex) ); + assert( sqlite3_mutex_notheld(pcache1.mutex) ); + if( sqlite3GlobalConfig.pPage==0 ){ + PgHdr1 *p; + pcache1EnterMutex(&pcache1.grp); + while( (nReq<0 || nFreeisAnchor==0 + ){ + nFree += pcache1MemSize(p->page.pBuf); +#ifdef SQLITE_PCACHE_SEPARATE_HEADER + nFree += sqlite3MemSize(p); +#endif + assert( PAGE_IS_UNPINNED(p) ); + pcache1PinPage(p); + pcache1RemoveFromHash(p, 1); + } + pcache1LeaveMutex(&pcache1.grp); + } + return nFree; +} +#endif /* SQLITE_ENABLE_MEMORY_MANAGEMENT */ + +#ifdef SQLITE_TEST +/* +** This function is used by test procedures to inspect the internal state +** of the global cache. +*/ +SQLITE_PRIVATE void sqlite3PcacheStats( + int *pnCurrent, /* OUT: Total number of pages cached */ + int *pnMax, /* OUT: Global maximum cache size */ + int *pnMin, /* OUT: Sum of PCache1.nMin for purgeable caches */ + int *pnRecyclable /* OUT: Total number of pages available for recycling */ +){ + PgHdr1 *p; + int nRecyclable = 0; + for(p=pcache1.grp.lru.pLruNext; p && !p->isAnchor; p=p->pLruNext){ + assert( PAGE_IS_UNPINNED(p) ); + nRecyclable++; + } + *pnCurrent = pcache1.grp.nPurgeable; + *pnMax = (int)pcache1.grp.nMaxPage; + *pnMin = (int)pcache1.grp.nMinPage; + *pnRecyclable = nRecyclable; +} +#endif + +/************** End of pcache1.c *********************************************/ +/************** Begin file rowset.c ******************************************/ +/* +** 2008 December 3 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** +** This module implements an object we call a "RowSet". +** +** The RowSet object is a collection of rowids. Rowids +** are inserted into the RowSet in an arbitrary order. Inserts +** can be intermixed with tests to see if a given rowid has been +** previously inserted into the RowSet. +** +** After all inserts are finished, it is possible to extract the +** elements of the RowSet in sorted order. Once this extraction +** process has started, no new elements may be inserted. +** +** Hence, the primitive operations for a RowSet are: +** +** CREATE +** INSERT +** TEST +** SMALLEST +** DESTROY +** +** The CREATE and DESTROY primitives are the constructor and destructor, +** obviously. The INSERT primitive adds a new element to the RowSet. +** TEST checks to see if an element is already in the RowSet. SMALLEST +** extracts the least value from the RowSet. +** +** The INSERT primitive might allocate additional memory. Memory is +** allocated in chunks so most INSERTs do no allocation. There is an +** upper bound on the size of allocated memory. No memory is freed +** until DESTROY. +** +** The TEST primitive includes a "batch" number. The TEST primitive +** will only see elements that were inserted before the last change +** in the batch number. In other words, if an INSERT occurs between +** two TESTs where the TESTs have the same batch nubmer, then the +** value added by the INSERT will not be visible to the second TEST. +** The initial batch number is zero, so if the very first TEST contains +** a non-zero batch number, it will see all prior INSERTs. +** +** No INSERTs may occurs after a SMALLEST. An assertion will fail if +** that is attempted. +** +** The cost of an INSERT is roughly constant. (Sometimes new memory +** has to be allocated on an INSERT.) The cost of a TEST with a new +** batch number is O(NlogN) where N is the number of elements in the RowSet. +** The cost of a TEST using the same batch number is O(logN). The cost +** of the first SMALLEST is O(NlogN). Second and subsequent SMALLEST +** primitives are constant time. The cost of DESTROY is O(N). +** +** TEST and SMALLEST may not be used by the same RowSet. This used to +** be possible, but the feature was not used, so it was removed in order +** to simplify the code. +*/ +/* #include "sqliteInt.h" */ + + +/* +** Target size for allocation chunks. +*/ +#define ROWSET_ALLOCATION_SIZE 1024 + +/* +** The number of rowset entries per allocation chunk. +*/ +#define ROWSET_ENTRY_PER_CHUNK \ + ((ROWSET_ALLOCATION_SIZE-8)/sizeof(struct RowSetEntry)) + +/* +** Each entry in a RowSet is an instance of the following object. +** +** This same object is reused to store a linked list of trees of RowSetEntry +** objects. In that alternative use, pRight points to the next entry +** in the list, pLeft points to the tree, and v is unused. The +** RowSet.pForest value points to the head of this forest list. +*/ +struct RowSetEntry { + i64 v; /* ROWID value for this entry */ + struct RowSetEntry *pRight; /* Right subtree (larger entries) or list */ + struct RowSetEntry *pLeft; /* Left subtree (smaller entries) */ +}; + +/* +** RowSetEntry objects are allocated in large chunks (instances of the +** following structure) to reduce memory allocation overhead. The +** chunks are kept on a linked list so that they can be deallocated +** when the RowSet is destroyed. +*/ +struct RowSetChunk { + struct RowSetChunk *pNextChunk; /* Next chunk on list of them all */ + struct RowSetEntry aEntry[ROWSET_ENTRY_PER_CHUNK]; /* Allocated entries */ +}; + +/* +** A RowSet in an instance of the following structure. +** +** A typedef of this structure if found in sqliteInt.h. +*/ +struct RowSet { + struct RowSetChunk *pChunk; /* List of all chunk allocations */ + sqlite3 *db; /* The database connection */ + struct RowSetEntry *pEntry; /* List of entries using pRight */ + struct RowSetEntry *pLast; /* Last entry on the pEntry list */ + struct RowSetEntry *pFresh; /* Source of new entry objects */ + struct RowSetEntry *pForest; /* List of binary trees of entries */ + u16 nFresh; /* Number of objects on pFresh */ + u16 rsFlags; /* Various flags */ + int iBatch; /* Current insert batch */ +}; + +/* +** Allowed values for RowSet.rsFlags +*/ +#define ROWSET_SORTED 0x01 /* True if RowSet.pEntry is sorted */ +#define ROWSET_NEXT 0x02 /* True if sqlite3RowSetNext() has been called */ + +/* +** Allocate a RowSet object. Return NULL if a memory allocation +** error occurs. +*/ +SQLITE_PRIVATE RowSet *sqlite3RowSetInit(sqlite3 *db){ + RowSet *p = sqlite3DbMallocRawNN(db, sizeof(*p)); + if( p ){ + int N = sqlite3DbMallocSize(db, p); + p->pChunk = 0; + p->db = db; + p->pEntry = 0; + p->pLast = 0; + p->pForest = 0; + p->pFresh = (struct RowSetEntry*)(ROUND8(sizeof(*p)) + (char*)p); + p->nFresh = (u16)((N - ROUND8(sizeof(*p)))/sizeof(struct RowSetEntry)); + p->rsFlags = ROWSET_SORTED; + p->iBatch = 0; + } + return p; +} + +/* +** Deallocate all chunks from a RowSet. This frees all memory that +** the RowSet has allocated over its lifetime. This routine is +** the destructor for the RowSet. +*/ +SQLITE_PRIVATE void sqlite3RowSetClear(void *pArg){ + RowSet *p = (RowSet*)pArg; + struct RowSetChunk *pChunk, *pNextChunk; + for(pChunk=p->pChunk; pChunk; pChunk = pNextChunk){ + pNextChunk = pChunk->pNextChunk; + sqlite3DbFree(p->db, pChunk); + } + p->pChunk = 0; + p->nFresh = 0; + p->pEntry = 0; + p->pLast = 0; + p->pForest = 0; + p->rsFlags = ROWSET_SORTED; +} + +/* +** Deallocate all chunks from a RowSet. This frees all memory that +** the RowSet has allocated over its lifetime. This routine is +** the destructor for the RowSet. +*/ +SQLITE_PRIVATE void sqlite3RowSetDelete(void *pArg){ + sqlite3RowSetClear(pArg); + sqlite3DbFree(((RowSet*)pArg)->db, pArg); +} + +/* +** Allocate a new RowSetEntry object that is associated with the +** given RowSet. Return a pointer to the new and completely uninitialized +** objected. +** +** In an OOM situation, the RowSet.db->mallocFailed flag is set and this +** routine returns NULL. +*/ +static struct RowSetEntry *rowSetEntryAlloc(RowSet *p){ + assert( p!=0 ); + if( p->nFresh==0 ){ /*OPTIMIZATION-IF-FALSE*/ + /* We could allocate a fresh RowSetEntry each time one is needed, but it + ** is more efficient to pull a preallocated entry from the pool */ + struct RowSetChunk *pNew; + pNew = sqlite3DbMallocRawNN(p->db, sizeof(*pNew)); + if( pNew==0 ){ + return 0; + } + pNew->pNextChunk = p->pChunk; + p->pChunk = pNew; + p->pFresh = pNew->aEntry; + p->nFresh = ROWSET_ENTRY_PER_CHUNK; + } + p->nFresh--; + return p->pFresh++; +} + +/* +** Insert a new value into a RowSet. +** +** The mallocFailed flag of the database connection is set if a +** memory allocation fails. +*/ +SQLITE_PRIVATE void sqlite3RowSetInsert(RowSet *p, i64 rowid){ + struct RowSetEntry *pEntry; /* The new entry */ + struct RowSetEntry *pLast; /* The last prior entry */ + + /* This routine is never called after sqlite3RowSetNext() */ + assert( p!=0 && (p->rsFlags & ROWSET_NEXT)==0 ); + + pEntry = rowSetEntryAlloc(p); + if( pEntry==0 ) return; + pEntry->v = rowid; + pEntry->pRight = 0; + pLast = p->pLast; + if( pLast ){ + if( rowid<=pLast->v ){ /*OPTIMIZATION-IF-FALSE*/ + /* Avoid unnecessary sorts by preserving the ROWSET_SORTED flags + ** where possible */ + p->rsFlags &= ~ROWSET_SORTED; + } + pLast->pRight = pEntry; + }else{ + p->pEntry = pEntry; + } + p->pLast = pEntry; +} + +/* +** Merge two lists of RowSetEntry objects. Remove duplicates. +** +** The input lists are connected via pRight pointers and are +** assumed to each already be in sorted order. +*/ +static struct RowSetEntry *rowSetEntryMerge( + struct RowSetEntry *pA, /* First sorted list to be merged */ + struct RowSetEntry *pB /* Second sorted list to be merged */ +){ + struct RowSetEntry head; + struct RowSetEntry *pTail; + + pTail = &head; + assert( pA!=0 && pB!=0 ); + for(;;){ + assert( pA->pRight==0 || pA->v<=pA->pRight->v ); + assert( pB->pRight==0 || pB->v<=pB->pRight->v ); + if( pA->v<=pB->v ){ + if( pA->vv ) pTail = pTail->pRight = pA; + pA = pA->pRight; + if( pA==0 ){ + pTail->pRight = pB; + break; + } + }else{ + pTail = pTail->pRight = pB; + pB = pB->pRight; + if( pB==0 ){ + pTail->pRight = pA; + break; + } + } + } + return head.pRight; +} + +/* +** Sort all elements on the list of RowSetEntry objects into order of +** increasing v. +*/ +static struct RowSetEntry *rowSetEntrySort(struct RowSetEntry *pIn){ + unsigned int i; + struct RowSetEntry *pNext, *aBucket[40]; + + memset(aBucket, 0, sizeof(aBucket)); + while( pIn ){ + pNext = pIn->pRight; + pIn->pRight = 0; + for(i=0; aBucket[i]; i++){ + pIn = rowSetEntryMerge(aBucket[i], pIn); + aBucket[i] = 0; + } + aBucket[i] = pIn; + pIn = pNext; + } + pIn = aBucket[0]; + for(i=1; ipLeft ){ + struct RowSetEntry *p; + rowSetTreeToList(pIn->pLeft, ppFirst, &p); + p->pRight = pIn; + }else{ + *ppFirst = pIn; + } + if( pIn->pRight ){ + rowSetTreeToList(pIn->pRight, &pIn->pRight, ppLast); + }else{ + *ppLast = pIn; + } + assert( (*ppLast)->pRight==0 ); +} + + +/* +** Convert a sorted list of elements (connected by pRight) into a binary +** tree with depth of iDepth. A depth of 1 means the tree contains a single +** node taken from the head of *ppList. A depth of 2 means a tree with +** three nodes. And so forth. +** +** Use as many entries from the input list as required and update the +** *ppList to point to the unused elements of the list. If the input +** list contains too few elements, then construct an incomplete tree +** and leave *ppList set to NULL. +** +** Return a pointer to the root of the constructed binary tree. +*/ +static struct RowSetEntry *rowSetNDeepTree( + struct RowSetEntry **ppList, + int iDepth +){ + struct RowSetEntry *p; /* Root of the new tree */ + struct RowSetEntry *pLeft; /* Left subtree */ + if( *ppList==0 ){ /*OPTIMIZATION-IF-TRUE*/ + /* Prevent unnecessary deep recursion when we run out of entries */ + return 0; + } + if( iDepth>1 ){ /*OPTIMIZATION-IF-TRUE*/ + /* This branch causes a *balanced* tree to be generated. A valid tree + ** is still generated without this branch, but the tree is wildly + ** unbalanced and inefficient. */ + pLeft = rowSetNDeepTree(ppList, iDepth-1); + p = *ppList; + if( p==0 ){ /*OPTIMIZATION-IF-FALSE*/ + /* It is safe to always return here, but the resulting tree + ** would be unbalanced */ + return pLeft; + } + p->pLeft = pLeft; + *ppList = p->pRight; + p->pRight = rowSetNDeepTree(ppList, iDepth-1); + }else{ + p = *ppList; + *ppList = p->pRight; + p->pLeft = p->pRight = 0; + } + return p; +} + +/* +** Convert a sorted list of elements into a binary tree. Make the tree +** as deep as it needs to be in order to contain the entire list. +*/ +static struct RowSetEntry *rowSetListToTree(struct RowSetEntry *pList){ + int iDepth; /* Depth of the tree so far */ + struct RowSetEntry *p; /* Current tree root */ + struct RowSetEntry *pLeft; /* Left subtree */ + + assert( pList!=0 ); + p = pList; + pList = p->pRight; + p->pLeft = p->pRight = 0; + for(iDepth=1; pList; iDepth++){ + pLeft = p; + p = pList; + pList = p->pRight; + p->pLeft = pLeft; + p->pRight = rowSetNDeepTree(&pList, iDepth); + } + return p; +} + +/* +** Extract the smallest element from the RowSet. +** Write the element into *pRowid. Return 1 on success. Return +** 0 if the RowSet is already empty. +** +** After this routine has been called, the sqlite3RowSetInsert() +** routine may not be called again. +** +** This routine may not be called after sqlite3RowSetTest() has +** been used. Older versions of RowSet allowed that, but as the +** capability was not used by the code generator, it was removed +** for code economy. +*/ +SQLITE_PRIVATE int sqlite3RowSetNext(RowSet *p, i64 *pRowid){ + assert( p!=0 ); + assert( p->pForest==0 ); /* Cannot be used with sqlite3RowSetText() */ + + /* Merge the forest into a single sorted list on first call */ + if( (p->rsFlags & ROWSET_NEXT)==0 ){ /*OPTIMIZATION-IF-FALSE*/ + if( (p->rsFlags & ROWSET_SORTED)==0 ){ /*OPTIMIZATION-IF-FALSE*/ + p->pEntry = rowSetEntrySort(p->pEntry); + } + p->rsFlags |= ROWSET_SORTED|ROWSET_NEXT; + } + + /* Return the next entry on the list */ + if( p->pEntry ){ + *pRowid = p->pEntry->v; + p->pEntry = p->pEntry->pRight; + if( p->pEntry==0 ){ /*OPTIMIZATION-IF-TRUE*/ + /* Free memory immediately, rather than waiting on sqlite3_finalize() */ + sqlite3RowSetClear(p); + } + return 1; + }else{ + return 0; + } +} + +/* +** Check to see if element iRowid was inserted into the rowset as +** part of any insert batch prior to iBatch. Return 1 or 0. +** +** If this is the first test of a new batch and if there exist entries +** on pRowSet->pEntry, then sort those entries into the forest at +** pRowSet->pForest so that they can be tested. +*/ +SQLITE_PRIVATE int sqlite3RowSetTest(RowSet *pRowSet, int iBatch, sqlite3_int64 iRowid){ + struct RowSetEntry *p, *pTree; + + /* This routine is never called after sqlite3RowSetNext() */ + assert( pRowSet!=0 && (pRowSet->rsFlags & ROWSET_NEXT)==0 ); + + /* Sort entries into the forest on the first test of a new batch. + ** To save unnecessary work, only do this when the batch number changes. + */ + if( iBatch!=pRowSet->iBatch ){ /*OPTIMIZATION-IF-FALSE*/ + p = pRowSet->pEntry; + if( p ){ + struct RowSetEntry **ppPrevTree = &pRowSet->pForest; + if( (pRowSet->rsFlags & ROWSET_SORTED)==0 ){ /*OPTIMIZATION-IF-FALSE*/ + /* Only sort the current set of entiries if they need it */ + p = rowSetEntrySort(p); + } + for(pTree = pRowSet->pForest; pTree; pTree=pTree->pRight){ + ppPrevTree = &pTree->pRight; + if( pTree->pLeft==0 ){ + pTree->pLeft = rowSetListToTree(p); + break; + }else{ + struct RowSetEntry *pAux, *pTail; + rowSetTreeToList(pTree->pLeft, &pAux, &pTail); + pTree->pLeft = 0; + p = rowSetEntryMerge(pAux, p); + } + } + if( pTree==0 ){ + *ppPrevTree = pTree = rowSetEntryAlloc(pRowSet); + if( pTree ){ + pTree->v = 0; + pTree->pRight = 0; + pTree->pLeft = rowSetListToTree(p); + } + } + pRowSet->pEntry = 0; + pRowSet->pLast = 0; + pRowSet->rsFlags |= ROWSET_SORTED; + } + pRowSet->iBatch = iBatch; + } + + /* Test to see if the iRowid value appears anywhere in the forest. + ** Return 1 if it does and 0 if not. + */ + for(pTree = pRowSet->pForest; pTree; pTree=pTree->pRight){ + p = pTree->pLeft; + while( p ){ + if( p->vpRight; + }else if( p->v>iRowid ){ + p = p->pLeft; + }else{ + return 1; + } + } + } + return 0; +} + +/************** End of rowset.c **********************************************/ +/************** Begin file pager.c *******************************************/ +/* +** 2001 September 15 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** This is the implementation of the page cache subsystem or "pager". +** +** The pager is used to access a database disk file. It implements +** atomic commit and rollback through the use of a journal file that +** is separate from the database file. The pager also implements file +** locking to prevent two processes from writing the same database +** file simultaneously, or one process from reading the database while +** another is writing. +*/ +#ifndef SQLITE_OMIT_DISKIO +/* #include "sqliteInt.h" */ +/************** Include wal.h in the middle of pager.c ***********************/ +/************** Begin file wal.h *********************************************/ +/* +** 2010 February 1 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** This header file defines the interface to the write-ahead logging +** system. Refer to the comments below and the header comment attached to +** the implementation of each function in log.c for further details. +*/ + +#ifndef SQLITE_WAL_H +#define SQLITE_WAL_H + +/* #include "sqliteInt.h" */ + +/* Macros for extracting appropriate sync flags for either transaction +** commits (WAL_SYNC_FLAGS(X)) or for checkpoint ops (CKPT_SYNC_FLAGS(X)): +*/ +#define WAL_SYNC_FLAGS(X) ((X)&0x03) +#define CKPT_SYNC_FLAGS(X) (((X)>>2)&0x03) + +#ifdef SQLITE_OMIT_WAL +# define sqlite3WalOpen(x,y,z) 0 +# define sqlite3WalLimit(x,y) +# define sqlite3WalClose(v,w,x,y,z) 0 +# define sqlite3WalBeginReadTransaction(y,z) 0 +# define sqlite3WalEndReadTransaction(z) +# define sqlite3WalDbsize(y) 0 +# define sqlite3WalBeginWriteTransaction(y) 0 +# define sqlite3WalEndWriteTransaction(x) 0 +# define sqlite3WalUndo(x,y,z) 0 +# define sqlite3WalSavepoint(y,z) +# define sqlite3WalSavepointUndo(y,z) 0 +# define sqlite3WalFrames(u,v,w,x,y,z) 0 +# define sqlite3WalCheckpoint(q,r,s,t,u,v,w,x,y,z) 0 +# define sqlite3WalCallback(z) 0 +# define sqlite3WalExclusiveMode(y,z) 0 +# define sqlite3WalHeapMemory(z) 0 +# define sqlite3WalFramesize(z) 0 +# define sqlite3WalFindFrame(x,y,z) 0 +# define sqlite3WalFile(x) 0 +#else + +#define WAL_SAVEPOINT_NDATA 4 + +/* Connection to a write-ahead log (WAL) file. +** There is one object of this type for each pager. +*/ +typedef struct Wal Wal; + +/* Open and close a connection to a write-ahead log. */ +SQLITE_PRIVATE int sqlite3WalOpen(sqlite3_vfs*, sqlite3_file*, const char *, int, i64, Wal**); +SQLITE_PRIVATE int sqlite3WalClose(Wal *pWal, sqlite3*, int sync_flags, int, u8 *); + +/* Set the limiting size of a WAL file. */ +SQLITE_PRIVATE void sqlite3WalLimit(Wal*, i64); + +/* Used by readers to open (lock) and close (unlock) a snapshot. A +** snapshot is like a read-transaction. It is the state of the database +** at an instant in time. sqlite3WalOpenSnapshot gets a read lock and +** preserves the current state even if the other threads or processes +** write to or checkpoint the WAL. sqlite3WalCloseSnapshot() closes the +** transaction and releases the lock. +*/ +SQLITE_PRIVATE int sqlite3WalBeginReadTransaction(Wal *pWal, int *); +SQLITE_PRIVATE void sqlite3WalEndReadTransaction(Wal *pWal); + +/* Read a page from the write-ahead log, if it is present. */ +SQLITE_PRIVATE int sqlite3WalFindFrame(Wal *, Pgno, u32 *); +SQLITE_PRIVATE int sqlite3WalReadFrame(Wal *, u32, int, u8 *); + +/* If the WAL is not empty, return the size of the database. */ +SQLITE_PRIVATE Pgno sqlite3WalDbsize(Wal *pWal); + +/* Obtain or release the WRITER lock. */ +SQLITE_PRIVATE int sqlite3WalBeginWriteTransaction(Wal *pWal); +SQLITE_PRIVATE int sqlite3WalEndWriteTransaction(Wal *pWal); + +/* Undo any frames written (but not committed) to the log */ +SQLITE_PRIVATE int sqlite3WalUndo(Wal *pWal, int (*xUndo)(void *, Pgno), void *pUndoCtx); + +/* Return an integer that records the current (uncommitted) write +** position in the WAL */ +SQLITE_PRIVATE void sqlite3WalSavepoint(Wal *pWal, u32 *aWalData); + +/* Move the write position of the WAL back to iFrame. Called in +** response to a ROLLBACK TO command. */ +SQLITE_PRIVATE int sqlite3WalSavepointUndo(Wal *pWal, u32 *aWalData); + +/* Write a frame or frames to the log. */ +SQLITE_PRIVATE int sqlite3WalFrames(Wal *pWal, int, PgHdr *, Pgno, int, int); + +/* Copy pages from the log to the database file */ +SQLITE_PRIVATE int sqlite3WalCheckpoint( + Wal *pWal, /* Write-ahead log connection */ + sqlite3 *db, /* Check this handle's interrupt flag */ + int eMode, /* One of PASSIVE, FULL and RESTART */ + int (*xBusy)(void*), /* Function to call when busy */ + void *pBusyArg, /* Context argument for xBusyHandler */ + int sync_flags, /* Flags to sync db file with (or 0) */ + int nBuf, /* Size of buffer nBuf */ + u8 *zBuf, /* Temporary buffer to use */ + int *pnLog, /* OUT: Number of frames in WAL */ + int *pnCkpt /* OUT: Number of backfilled frames in WAL */ +); + +/* Return the value to pass to a sqlite3_wal_hook callback, the +** number of frames in the WAL at the point of the last commit since +** sqlite3WalCallback() was called. If no commits have occurred since +** the last call, then return 0. +*/ +SQLITE_PRIVATE int sqlite3WalCallback(Wal *pWal); + +/* Tell the wal layer that an EXCLUSIVE lock has been obtained (or released) +** by the pager layer on the database file. +*/ +SQLITE_PRIVATE int sqlite3WalExclusiveMode(Wal *pWal, int op); + +/* Return true if the argument is non-NULL and the WAL module is using +** heap-memory for the wal-index. Otherwise, if the argument is NULL or the +** WAL module is using shared-memory, return false. +*/ +SQLITE_PRIVATE int sqlite3WalHeapMemory(Wal *pWal); + +#ifdef SQLITE_ENABLE_SNAPSHOT +SQLITE_PRIVATE int sqlite3WalSnapshotGet(Wal *pWal, sqlite3_snapshot **ppSnapshot); +SQLITE_PRIVATE void sqlite3WalSnapshotOpen(Wal *pWal, sqlite3_snapshot *pSnapshot); +SQLITE_PRIVATE int sqlite3WalSnapshotRecover(Wal *pWal); +SQLITE_PRIVATE int sqlite3WalSnapshotCheck(Wal *pWal, sqlite3_snapshot *pSnapshot); +SQLITE_PRIVATE void sqlite3WalSnapshotUnlock(Wal *pWal); +#endif + +#ifdef SQLITE_ENABLE_ZIPVFS +/* If the WAL file is not empty, return the number of bytes of content +** stored in each frame (i.e. the db page-size when the WAL was created). +*/ +SQLITE_PRIVATE int sqlite3WalFramesize(Wal *pWal); +#endif + +/* Return the sqlite3_file object for the WAL file */ +SQLITE_PRIVATE sqlite3_file *sqlite3WalFile(Wal *pWal); + +#endif /* ifndef SQLITE_OMIT_WAL */ +#endif /* SQLITE_WAL_H */ + +/************** End of wal.h *************************************************/ +/************** Continuing where we left off in pager.c **********************/ + + +/******************* NOTES ON THE DESIGN OF THE PAGER ************************ +** +** This comment block describes invariants that hold when using a rollback +** journal. These invariants do not apply for journal_mode=WAL, +** journal_mode=MEMORY, or journal_mode=OFF. +** +** Within this comment block, a page is deemed to have been synced +** automatically as soon as it is written when PRAGMA synchronous=OFF. +** Otherwise, the page is not synced until the xSync method of the VFS +** is called successfully on the file containing the page. +** +** Definition: A page of the database file is said to be "overwriteable" if +** one or more of the following are true about the page: +** +** (a) The original content of the page as it was at the beginning of +** the transaction has been written into the rollback journal and +** synced. +** +** (b) The page was a freelist leaf page at the start of the transaction. +** +** (c) The page number is greater than the largest page that existed in +** the database file at the start of the transaction. +** +** (1) A page of the database file is never overwritten unless one of the +** following are true: +** +** (a) The page and all other pages on the same sector are overwriteable. +** +** (b) The atomic page write optimization is enabled, and the entire +** transaction other than the update of the transaction sequence +** number consists of a single page change. +** +** (2) The content of a page written into the rollback journal exactly matches +** both the content in the database when the rollback journal was written +** and the content in the database at the beginning of the current +** transaction. +** +** (3) Writes to the database file are an integer multiple of the page size +** in length and are aligned on a page boundary. +** +** (4) Reads from the database file are either aligned on a page boundary and +** an integer multiple of the page size in length or are taken from the +** first 100 bytes of the database file. +** +** (5) All writes to the database file are synced prior to the rollback journal +** being deleted, truncated, or zeroed. +** +** (6) If a master journal file is used, then all writes to the database file +** are synced prior to the master journal being deleted. +** +** Definition: Two databases (or the same database at two points it time) +** are said to be "logically equivalent" if they give the same answer to +** all queries. Note in particular the content of freelist leaf +** pages can be changed arbitrarily without affecting the logical equivalence +** of the database. +** +** (7) At any time, if any subset, including the empty set and the total set, +** of the unsynced changes to a rollback journal are removed and the +** journal is rolled back, the resulting database file will be logically +** equivalent to the database file at the beginning of the transaction. +** +** (8) When a transaction is rolled back, the xTruncate method of the VFS +** is called to restore the database file to the same size it was at +** the beginning of the transaction. (In some VFSes, the xTruncate +** method is a no-op, but that does not change the fact the SQLite will +** invoke it.) +** +** (9) Whenever the database file is modified, at least one bit in the range +** of bytes from 24 through 39 inclusive will be changed prior to releasing +** the EXCLUSIVE lock, thus signaling other connections on the same +** database to flush their caches. +** +** (10) The pattern of bits in bytes 24 through 39 shall not repeat in less +** than one billion transactions. +** +** (11) A database file is well-formed at the beginning and at the conclusion +** of every transaction. +** +** (12) An EXCLUSIVE lock is held on the database file when writing to +** the database file. +** +** (13) A SHARED lock is held on the database file while reading any +** content out of the database file. +** +******************************************************************************/ + +/* +** Macros for troubleshooting. Normally turned off +*/ +#if 0 +int sqlite3PagerTrace=1; /* True to enable tracing */ +#define sqlite3DebugPrintf printf +#define PAGERTRACE(X) if( sqlite3PagerTrace ){ sqlite3DebugPrintf X; } +#else +#define PAGERTRACE(X) +#endif + +/* +** The following two macros are used within the PAGERTRACE() macros above +** to print out file-descriptors. +** +** PAGERID() takes a pointer to a Pager struct as its argument. The +** associated file-descriptor is returned. FILEHANDLEID() takes an sqlite3_file +** struct as its argument. +*/ +#define PAGERID(p) (SQLITE_PTR_TO_INT(p->fd)) +#define FILEHANDLEID(fd) (SQLITE_PTR_TO_INT(fd)) + +/* +** The Pager.eState variable stores the current 'state' of a pager. A +** pager may be in any one of the seven states shown in the following +** state diagram. +** +** OPEN <------+------+ +** | | | +** V | | +** +---------> READER-------+ | +** | | | +** | V | +** |<-------WRITER_LOCKED------> ERROR +** | | ^ +** | V | +** |<------WRITER_CACHEMOD-------->| +** | | | +** | V | +** |<-------WRITER_DBMOD---------->| +** | | | +** | V | +** +<------WRITER_FINISHED-------->+ +** +** +** List of state transitions and the C [function] that performs each: +** +** OPEN -> READER [sqlite3PagerSharedLock] +** READER -> OPEN [pager_unlock] +** +** READER -> WRITER_LOCKED [sqlite3PagerBegin] +** WRITER_LOCKED -> WRITER_CACHEMOD [pager_open_journal] +** WRITER_CACHEMOD -> WRITER_DBMOD [syncJournal] +** WRITER_DBMOD -> WRITER_FINISHED [sqlite3PagerCommitPhaseOne] +** WRITER_*** -> READER [pager_end_transaction] +** +** WRITER_*** -> ERROR [pager_error] +** ERROR -> OPEN [pager_unlock] +** +** +** OPEN: +** +** The pager starts up in this state. Nothing is guaranteed in this +** state - the file may or may not be locked and the database size is +** unknown. The database may not be read or written. +** +** * No read or write transaction is active. +** * Any lock, or no lock at all, may be held on the database file. +** * The dbSize, dbOrigSize and dbFileSize variables may not be trusted. +** +** READER: +** +** In this state all the requirements for reading the database in +** rollback (non-WAL) mode are met. Unless the pager is (or recently +** was) in exclusive-locking mode, a user-level read transaction is +** open. The database size is known in this state. +** +** A connection running with locking_mode=normal enters this state when +** it opens a read-transaction on the database and returns to state +** OPEN after the read-transaction is completed. However a connection +** running in locking_mode=exclusive (including temp databases) remains in +** this state even after the read-transaction is closed. The only way +** a locking_mode=exclusive connection can transition from READER to OPEN +** is via the ERROR state (see below). +** +** * A read transaction may be active (but a write-transaction cannot). +** * A SHARED or greater lock is held on the database file. +** * The dbSize variable may be trusted (even if a user-level read +** transaction is not active). The dbOrigSize and dbFileSize variables +** may not be trusted at this point. +** * If the database is a WAL database, then the WAL connection is open. +** * Even if a read-transaction is not open, it is guaranteed that +** there is no hot-journal in the file-system. +** +** WRITER_LOCKED: +** +** The pager moves to this state from READER when a write-transaction +** is first opened on the database. In WRITER_LOCKED state, all locks +** required to start a write-transaction are held, but no actual +** modifications to the cache or database have taken place. +** +** In rollback mode, a RESERVED or (if the transaction was opened with +** BEGIN EXCLUSIVE) EXCLUSIVE lock is obtained on the database file when +** moving to this state, but the journal file is not written to or opened +** to in this state. If the transaction is committed or rolled back while +** in WRITER_LOCKED state, all that is required is to unlock the database +** file. +** +** IN WAL mode, WalBeginWriteTransaction() is called to lock the log file. +** If the connection is running with locking_mode=exclusive, an attempt +** is made to obtain an EXCLUSIVE lock on the database file. +** +** * A write transaction is active. +** * If the connection is open in rollback-mode, a RESERVED or greater +** lock is held on the database file. +** * If the connection is open in WAL-mode, a WAL write transaction +** is open (i.e. sqlite3WalBeginWriteTransaction() has been successfully +** called). +** * The dbSize, dbOrigSize and dbFileSize variables are all valid. +** * The contents of the pager cache have not been modified. +** * The journal file may or may not be open. +** * Nothing (not even the first header) has been written to the journal. +** +** WRITER_CACHEMOD: +** +** A pager moves from WRITER_LOCKED state to this state when a page is +** first modified by the upper layer. In rollback mode the journal file +** is opened (if it is not already open) and a header written to the +** start of it. The database file on disk has not been modified. +** +** * A write transaction is active. +** * A RESERVED or greater lock is held on the database file. +** * The journal file is open and the first header has been written +** to it, but the header has not been synced to disk. +** * The contents of the page cache have been modified. +** +** WRITER_DBMOD: +** +** The pager transitions from WRITER_CACHEMOD into WRITER_DBMOD state +** when it modifies the contents of the database file. WAL connections +** never enter this state (since they do not modify the database file, +** just the log file). +** +** * A write transaction is active. +** * An EXCLUSIVE or greater lock is held on the database file. +** * The journal file is open and the first header has been written +** and synced to disk. +** * The contents of the page cache have been modified (and possibly +** written to disk). +** +** WRITER_FINISHED: +** +** It is not possible for a WAL connection to enter this state. +** +** A rollback-mode pager changes to WRITER_FINISHED state from WRITER_DBMOD +** state after the entire transaction has been successfully written into the +** database file. In this state the transaction may be committed simply +** by finalizing the journal file. Once in WRITER_FINISHED state, it is +** not possible to modify the database further. At this point, the upper +** layer must either commit or rollback the transaction. +** +** * A write transaction is active. +** * An EXCLUSIVE or greater lock is held on the database file. +** * All writing and syncing of journal and database data has finished. +** If no error occurred, all that remains is to finalize the journal to +** commit the transaction. If an error did occur, the caller will need +** to rollback the transaction. +** +** ERROR: +** +** The ERROR state is entered when an IO or disk-full error (including +** SQLITE_IOERR_NOMEM) occurs at a point in the code that makes it +** difficult to be sure that the in-memory pager state (cache contents, +** db size etc.) are consistent with the contents of the file-system. +** +** Temporary pager files may enter the ERROR state, but in-memory pagers +** cannot. +** +** For example, if an IO error occurs while performing a rollback, +** the contents of the page-cache may be left in an inconsistent state. +** At this point it would be dangerous to change back to READER state +** (as usually happens after a rollback). Any subsequent readers might +** report database corruption (due to the inconsistent cache), and if +** they upgrade to writers, they may inadvertently corrupt the database +** file. To avoid this hazard, the pager switches into the ERROR state +** instead of READER following such an error. +** +** Once it has entered the ERROR state, any attempt to use the pager +** to read or write data returns an error. Eventually, once all +** outstanding transactions have been abandoned, the pager is able to +** transition back to OPEN state, discarding the contents of the +** page-cache and any other in-memory state at the same time. Everything +** is reloaded from disk (and, if necessary, hot-journal rollback peformed) +** when a read-transaction is next opened on the pager (transitioning +** the pager into READER state). At that point the system has recovered +** from the error. +** +** Specifically, the pager jumps into the ERROR state if: +** +** 1. An error occurs while attempting a rollback. This happens in +** function sqlite3PagerRollback(). +** +** 2. An error occurs while attempting to finalize a journal file +** following a commit in function sqlite3PagerCommitPhaseTwo(). +** +** 3. An error occurs while attempting to write to the journal or +** database file in function pagerStress() in order to free up +** memory. +** +** In other cases, the error is returned to the b-tree layer. The b-tree +** layer then attempts a rollback operation. If the error condition +** persists, the pager enters the ERROR state via condition (1) above. +** +** Condition (3) is necessary because it can be triggered by a read-only +** statement executed within a transaction. In this case, if the error +** code were simply returned to the user, the b-tree layer would not +** automatically attempt a rollback, as it assumes that an error in a +** read-only statement cannot leave the pager in an internally inconsistent +** state. +** +** * The Pager.errCode variable is set to something other than SQLITE_OK. +** * There are one or more outstanding references to pages (after the +** last reference is dropped the pager should move back to OPEN state). +** * The pager is not an in-memory pager. +** +** +** Notes: +** +** * A pager is never in WRITER_DBMOD or WRITER_FINISHED state if the +** connection is open in WAL mode. A WAL connection is always in one +** of the first four states. +** +** * Normally, a connection open in exclusive mode is never in PAGER_OPEN +** state. There are two exceptions: immediately after exclusive-mode has +** been turned on (and before any read or write transactions are +** executed), and when the pager is leaving the "error state". +** +** * See also: assert_pager_state(). +*/ +#define PAGER_OPEN 0 +#define PAGER_READER 1 +#define PAGER_WRITER_LOCKED 2 +#define PAGER_WRITER_CACHEMOD 3 +#define PAGER_WRITER_DBMOD 4 +#define PAGER_WRITER_FINISHED 5 +#define PAGER_ERROR 6 + +/* +** The Pager.eLock variable is almost always set to one of the +** following locking-states, according to the lock currently held on +** the database file: NO_LOCK, SHARED_LOCK, RESERVED_LOCK or EXCLUSIVE_LOCK. +** This variable is kept up to date as locks are taken and released by +** the pagerLockDb() and pagerUnlockDb() wrappers. +** +** If the VFS xLock() or xUnlock() returns an error other than SQLITE_BUSY +** (i.e. one of the SQLITE_IOERR subtypes), it is not clear whether or not +** the operation was successful. In these circumstances pagerLockDb() and +** pagerUnlockDb() take a conservative approach - eLock is always updated +** when unlocking the file, and only updated when locking the file if the +** VFS call is successful. This way, the Pager.eLock variable may be set +** to a less exclusive (lower) value than the lock that is actually held +** at the system level, but it is never set to a more exclusive value. +** +** This is usually safe. If an xUnlock fails or appears to fail, there may +** be a few redundant xLock() calls or a lock may be held for longer than +** required, but nothing really goes wrong. +** +** The exception is when the database file is unlocked as the pager moves +** from ERROR to OPEN state. At this point there may be a hot-journal file +** in the file-system that needs to be rolled back (as part of an OPEN->SHARED +** transition, by the same pager or any other). If the call to xUnlock() +** fails at this point and the pager is left holding an EXCLUSIVE lock, this +** can confuse the call to xCheckReservedLock() call made later as part +** of hot-journal detection. +** +** xCheckReservedLock() is defined as returning true "if there is a RESERVED +** lock held by this process or any others". So xCheckReservedLock may +** return true because the caller itself is holding an EXCLUSIVE lock (but +** doesn't know it because of a previous error in xUnlock). If this happens +** a hot-journal may be mistaken for a journal being created by an active +** transaction in another process, causing SQLite to read from the database +** without rolling it back. +** +** To work around this, if a call to xUnlock() fails when unlocking the +** database in the ERROR state, Pager.eLock is set to UNKNOWN_LOCK. It +** is only changed back to a real locking state after a successful call +** to xLock(EXCLUSIVE). Also, the code to do the OPEN->SHARED state transition +** omits the check for a hot-journal if Pager.eLock is set to UNKNOWN_LOCK +** lock. Instead, it assumes a hot-journal exists and obtains an EXCLUSIVE +** lock on the database file before attempting to roll it back. See function +** PagerSharedLock() for more detail. +** +** Pager.eLock may only be set to UNKNOWN_LOCK when the pager is in +** PAGER_OPEN state. +*/ +#define UNKNOWN_LOCK (EXCLUSIVE_LOCK+1) + +/* +** A macro used for invoking the codec if there is one +*/ +#ifdef SQLITE_HAS_CODEC +# define CODEC1(P,D,N,X,E) \ + if( P->xCodec && P->xCodec(P->pCodec,D,N,X)==0 ){ E; } +# define CODEC2(P,D,N,X,E,O) \ + if( P->xCodec==0 ){ O=(char*)D; }else \ + if( (O=(char*)(P->xCodec(P->pCodec,D,N,X)))==0 ){ E; } +#else +# define CODEC1(P,D,N,X,E) /* NO-OP */ +# define CODEC2(P,D,N,X,E,O) O=(char*)D +#endif + +/* +** The maximum allowed sector size. 64KiB. If the xSectorsize() method +** returns a value larger than this, then MAX_SECTOR_SIZE is used instead. +** This could conceivably cause corruption following a power failure on +** such a system. This is currently an undocumented limit. +*/ +#define MAX_SECTOR_SIZE 0x10000 + + +/* +** An instance of the following structure is allocated for each active +** savepoint and statement transaction in the system. All such structures +** are stored in the Pager.aSavepoint[] array, which is allocated and +** resized using sqlite3Realloc(). +** +** When a savepoint is created, the PagerSavepoint.iHdrOffset field is +** set to 0. If a journal-header is written into the main journal while +** the savepoint is active, then iHdrOffset is set to the byte offset +** immediately following the last journal record written into the main +** journal before the journal-header. This is required during savepoint +** rollback (see pagerPlaybackSavepoint()). +*/ +typedef struct PagerSavepoint PagerSavepoint; +struct PagerSavepoint { + i64 iOffset; /* Starting offset in main journal */ + i64 iHdrOffset; /* See above */ + Bitvec *pInSavepoint; /* Set of pages in this savepoint */ + Pgno nOrig; /* Original number of pages in file */ + Pgno iSubRec; /* Index of first record in sub-journal */ +#ifndef SQLITE_OMIT_WAL + u32 aWalData[WAL_SAVEPOINT_NDATA]; /* WAL savepoint context */ +#endif +}; + +/* +** Bits of the Pager.doNotSpill flag. See further description below. +*/ +#define SPILLFLAG_OFF 0x01 /* Never spill cache. Set via pragma */ +#define SPILLFLAG_ROLLBACK 0x02 /* Current rolling back, so do not spill */ +#define SPILLFLAG_NOSYNC 0x04 /* Spill is ok, but do not sync */ + +/* +** An open page cache is an instance of struct Pager. A description of +** some of the more important member variables follows: +** +** eState +** +** The current 'state' of the pager object. See the comment and state +** diagram above for a description of the pager state. +** +** eLock +** +** For a real on-disk database, the current lock held on the database file - +** NO_LOCK, SHARED_LOCK, RESERVED_LOCK or EXCLUSIVE_LOCK. +** +** For a temporary or in-memory database (neither of which require any +** locks), this variable is always set to EXCLUSIVE_LOCK. Since such +** databases always have Pager.exclusiveMode==1, this tricks the pager +** logic into thinking that it already has all the locks it will ever +** need (and no reason to release them). +** +** In some (obscure) circumstances, this variable may also be set to +** UNKNOWN_LOCK. See the comment above the #define of UNKNOWN_LOCK for +** details. +** +** changeCountDone +** +** This boolean variable is used to make sure that the change-counter +** (the 4-byte header field at byte offset 24 of the database file) is +** not updated more often than necessary. +** +** It is set to true when the change-counter field is updated, which +** can only happen if an exclusive lock is held on the database file. +** It is cleared (set to false) whenever an exclusive lock is +** relinquished on the database file. Each time a transaction is committed, +** The changeCountDone flag is inspected. If it is true, the work of +** updating the change-counter is omitted for the current transaction. +** +** This mechanism means that when running in exclusive mode, a connection +** need only update the change-counter once, for the first transaction +** committed. +** +** setMaster +** +** When PagerCommitPhaseOne() is called to commit a transaction, it may +** (or may not) specify a master-journal name to be written into the +** journal file before it is synced to disk. +** +** Whether or not a journal file contains a master-journal pointer affects +** the way in which the journal file is finalized after the transaction is +** committed or rolled back when running in "journal_mode=PERSIST" mode. +** If a journal file does not contain a master-journal pointer, it is +** finalized by overwriting the first journal header with zeroes. If +** it does contain a master-journal pointer the journal file is finalized +** by truncating it to zero bytes, just as if the connection were +** running in "journal_mode=truncate" mode. +** +** Journal files that contain master journal pointers cannot be finalized +** simply by overwriting the first journal-header with zeroes, as the +** master journal pointer could interfere with hot-journal rollback of any +** subsequently interrupted transaction that reuses the journal file. +** +** The flag is cleared as soon as the journal file is finalized (either +** by PagerCommitPhaseTwo or PagerRollback). If an IO error prevents the +** journal file from being successfully finalized, the setMaster flag +** is cleared anyway (and the pager will move to ERROR state). +** +** doNotSpill +** +** This variables control the behavior of cache-spills (calls made by +** the pcache module to the pagerStress() routine to write cached data +** to the file-system in order to free up memory). +** +** When bits SPILLFLAG_OFF or SPILLFLAG_ROLLBACK of doNotSpill are set, +** writing to the database from pagerStress() is disabled altogether. +** The SPILLFLAG_ROLLBACK case is done in a very obscure case that +** comes up during savepoint rollback that requires the pcache module +** to allocate a new page to prevent the journal file from being written +** while it is being traversed by code in pager_playback(). The SPILLFLAG_OFF +** case is a user preference. +** +** If the SPILLFLAG_NOSYNC bit is set, writing to the database from +** pagerStress() is permitted, but syncing the journal file is not. +** This flag is set by sqlite3PagerWrite() when the file-system sector-size +** is larger than the database page-size in order to prevent a journal sync +** from happening in between the journalling of two pages on the same sector. +** +** subjInMemory +** +** This is a boolean variable. If true, then any required sub-journal +** is opened as an in-memory journal file. If false, then in-memory +** sub-journals are only used for in-memory pager files. +** +** This variable is updated by the upper layer each time a new +** write-transaction is opened. +** +** dbSize, dbOrigSize, dbFileSize +** +** Variable dbSize is set to the number of pages in the database file. +** It is valid in PAGER_READER and higher states (all states except for +** OPEN and ERROR). +** +** dbSize is set based on the size of the database file, which may be +** larger than the size of the database (the value stored at offset +** 28 of the database header by the btree). If the size of the file +** is not an integer multiple of the page-size, the value stored in +** dbSize is rounded down (i.e. a 5KB file with 2K page-size has dbSize==2). +** Except, any file that is greater than 0 bytes in size is considered +** to have at least one page. (i.e. a 1KB file with 2K page-size leads +** to dbSize==1). +** +** During a write-transaction, if pages with page-numbers greater than +** dbSize are modified in the cache, dbSize is updated accordingly. +** Similarly, if the database is truncated using PagerTruncateImage(), +** dbSize is updated. +** +** Variables dbOrigSize and dbFileSize are valid in states +** PAGER_WRITER_LOCKED and higher. dbOrigSize is a copy of the dbSize +** variable at the start of the transaction. It is used during rollback, +** and to determine whether or not pages need to be journalled before +** being modified. +** +** Throughout a write-transaction, dbFileSize contains the size of +** the file on disk in pages. It is set to a copy of dbSize when the +** write-transaction is first opened, and updated when VFS calls are made +** to write or truncate the database file on disk. +** +** The only reason the dbFileSize variable is required is to suppress +** unnecessary calls to xTruncate() after committing a transaction. If, +** when a transaction is committed, the dbFileSize variable indicates +** that the database file is larger than the database image (Pager.dbSize), +** pager_truncate() is called. The pager_truncate() call uses xFilesize() +** to measure the database file on disk, and then truncates it if required. +** dbFileSize is not used when rolling back a transaction. In this case +** pager_truncate() is called unconditionally (which means there may be +** a call to xFilesize() that is not strictly required). In either case, +** pager_truncate() may cause the file to become smaller or larger. +** +** dbHintSize +** +** The dbHintSize variable is used to limit the number of calls made to +** the VFS xFileControl(FCNTL_SIZE_HINT) method. +** +** dbHintSize is set to a copy of the dbSize variable when a +** write-transaction is opened (at the same time as dbFileSize and +** dbOrigSize). If the xFileControl(FCNTL_SIZE_HINT) method is called, +** dbHintSize is increased to the number of pages that correspond to the +** size-hint passed to the method call. See pager_write_pagelist() for +** details. +** +** errCode +** +** The Pager.errCode variable is only ever used in PAGER_ERROR state. It +** is set to zero in all other states. In PAGER_ERROR state, Pager.errCode +** is always set to SQLITE_FULL, SQLITE_IOERR or one of the SQLITE_IOERR_XXX +** sub-codes. +** +** syncFlags, walSyncFlags +** +** syncFlags is either SQLITE_SYNC_NORMAL (0x02) or SQLITE_SYNC_FULL (0x03). +** syncFlags is used for rollback mode. walSyncFlags is used for WAL mode +** and contains the flags used to sync the checkpoint operations in the +** lower two bits, and sync flags used for transaction commits in the WAL +** file in bits 0x04 and 0x08. In other words, to get the correct sync flags +** for checkpoint operations, use (walSyncFlags&0x03) and to get the correct +** sync flags for transaction commit, use ((walSyncFlags>>2)&0x03). Note +** that with synchronous=NORMAL in WAL mode, transaction commit is not synced +** meaning that the 0x04 and 0x08 bits are both zero. +*/ +struct Pager { + sqlite3_vfs *pVfs; /* OS functions to use for IO */ + u8 exclusiveMode; /* Boolean. True if locking_mode==EXCLUSIVE */ + u8 journalMode; /* One of the PAGER_JOURNALMODE_* values */ + u8 useJournal; /* Use a rollback journal on this file */ + u8 noSync; /* Do not sync the journal if true */ + u8 fullSync; /* Do extra syncs of the journal for robustness */ + u8 extraSync; /* sync directory after journal delete */ + u8 syncFlags; /* SYNC_NORMAL or SYNC_FULL otherwise */ + u8 walSyncFlags; /* See description above */ + u8 tempFile; /* zFilename is a temporary or immutable file */ + u8 noLock; /* Do not lock (except in WAL mode) */ + u8 readOnly; /* True for a read-only database */ + u8 memDb; /* True to inhibit all file I/O */ + + /************************************************************************** + ** The following block contains those class members that change during + ** routine operation. Class members not in this block are either fixed + ** when the pager is first created or else only change when there is a + ** significant mode change (such as changing the page_size, locking_mode, + ** or the journal_mode). From another view, these class members describe + ** the "state" of the pager, while other class members describe the + ** "configuration" of the pager. + */ + u8 eState; /* Pager state (OPEN, READER, WRITER_LOCKED..) */ + u8 eLock; /* Current lock held on database file */ + u8 changeCountDone; /* Set after incrementing the change-counter */ + u8 setMaster; /* True if a m-j name has been written to jrnl */ + u8 doNotSpill; /* Do not spill the cache when non-zero */ + u8 subjInMemory; /* True to use in-memory sub-journals */ + u8 bUseFetch; /* True to use xFetch() */ + u8 hasHeldSharedLock; /* True if a shared lock has ever been held */ + Pgno dbSize; /* Number of pages in the database */ + Pgno dbOrigSize; /* dbSize before the current transaction */ + Pgno dbFileSize; /* Number of pages in the database file */ + Pgno dbHintSize; /* Value passed to FCNTL_SIZE_HINT call */ + int errCode; /* One of several kinds of errors */ + int nRec; /* Pages journalled since last j-header written */ + u32 cksumInit; /* Quasi-random value added to every checksum */ + u32 nSubRec; /* Number of records written to sub-journal */ + Bitvec *pInJournal; /* One bit for each page in the database file */ + sqlite3_file *fd; /* File descriptor for database */ + sqlite3_file *jfd; /* File descriptor for main journal */ + sqlite3_file *sjfd; /* File descriptor for sub-journal */ + i64 journalOff; /* Current write offset in the journal file */ + i64 journalHdr; /* Byte offset to previous journal header */ + sqlite3_backup *pBackup; /* Pointer to list of ongoing backup processes */ + PagerSavepoint *aSavepoint; /* Array of active savepoints */ + int nSavepoint; /* Number of elements in aSavepoint[] */ + u32 iDataVersion; /* Changes whenever database content changes */ + char dbFileVers[16]; /* Changes whenever database file changes */ + + int nMmapOut; /* Number of mmap pages currently outstanding */ + sqlite3_int64 szMmap; /* Desired maximum mmap size */ + PgHdr *pMmapFreelist; /* List of free mmap page headers (pDirty) */ + /* + ** End of the routinely-changing class members + ***************************************************************************/ + + u16 nExtra; /* Add this many bytes to each in-memory page */ + i16 nReserve; /* Number of unused bytes at end of each page */ + u32 vfsFlags; /* Flags for sqlite3_vfs.xOpen() */ + u32 sectorSize; /* Assumed sector size during rollback */ + int pageSize; /* Number of bytes in a page */ + Pgno mxPgno; /* Maximum allowed size of the database */ + i64 journalSizeLimit; /* Size limit for persistent journal files */ + char *zFilename; /* Name of the database file */ + char *zJournal; /* Name of the journal file */ + int (*xBusyHandler)(void*); /* Function to call when busy */ + void *pBusyHandlerArg; /* Context argument for xBusyHandler */ + int aStat[4]; /* Total cache hits, misses, writes, spills */ +#ifdef SQLITE_TEST + int nRead; /* Database pages read */ +#endif + void (*xReiniter)(DbPage*); /* Call this routine when reloading pages */ + int (*xGet)(Pager*,Pgno,DbPage**,int); /* Routine to fetch a patch */ +#ifdef SQLITE_HAS_CODEC + void *(*xCodec)(void*,void*,Pgno,int); /* Routine for en/decoding data */ + void (*xCodecSizeChng)(void*,int,int); /* Notify of page size changes */ + void (*xCodecFree)(void*); /* Destructor for the codec */ + void *pCodec; /* First argument to xCodec... methods */ +#endif + char *pTmpSpace; /* Pager.pageSize bytes of space for tmp use */ + PCache *pPCache; /* Pointer to page cache object */ +#ifndef SQLITE_OMIT_WAL + Wal *pWal; /* Write-ahead log used by "journal_mode=wal" */ + char *zWal; /* File name for write-ahead log */ +#endif +}; + +/* +** Indexes for use with Pager.aStat[]. The Pager.aStat[] array contains +** the values accessed by passing SQLITE_DBSTATUS_CACHE_HIT, CACHE_MISS +** or CACHE_WRITE to sqlite3_db_status(). +*/ +#define PAGER_STAT_HIT 0 +#define PAGER_STAT_MISS 1 +#define PAGER_STAT_WRITE 2 +#define PAGER_STAT_SPILL 3 + +/* +** The following global variables hold counters used for +** testing purposes only. These variables do not exist in +** a non-testing build. These variables are not thread-safe. +*/ +#ifdef SQLITE_TEST +SQLITE_API int sqlite3_pager_readdb_count = 0; /* Number of full pages read from DB */ +SQLITE_API int sqlite3_pager_writedb_count = 0; /* Number of full pages written to DB */ +SQLITE_API int sqlite3_pager_writej_count = 0; /* Number of pages written to journal */ +# define PAGER_INCR(v) v++ +#else +# define PAGER_INCR(v) +#endif + + + +/* +** Journal files begin with the following magic string. The data +** was obtained from /dev/random. It is used only as a sanity check. +** +** Since version 2.8.0, the journal format contains additional sanity +** checking information. If the power fails while the journal is being +** written, semi-random garbage data might appear in the journal +** file after power is restored. If an attempt is then made +** to roll the journal back, the database could be corrupted. The additional +** sanity checking data is an attempt to discover the garbage in the +** journal and ignore it. +** +** The sanity checking information for the new journal format consists +** of a 32-bit checksum on each page of data. The checksum covers both +** the page number and the pPager->pageSize bytes of data for the page. +** This cksum is initialized to a 32-bit random value that appears in the +** journal file right after the header. The random initializer is important, +** because garbage data that appears at the end of a journal is likely +** data that was once in other files that have now been deleted. If the +** garbage data came from an obsolete journal file, the checksums might +** be correct. But by initializing the checksum to random value which +** is different for every journal, we minimize that risk. +*/ +static const unsigned char aJournalMagic[] = { + 0xd9, 0xd5, 0x05, 0xf9, 0x20, 0xa1, 0x63, 0xd7, +}; + +/* +** The size of the of each page record in the journal is given by +** the following macro. +*/ +#define JOURNAL_PG_SZ(pPager) ((pPager->pageSize) + 8) + +/* +** The journal header size for this pager. This is usually the same +** size as a single disk sector. See also setSectorSize(). +*/ +#define JOURNAL_HDR_SZ(pPager) (pPager->sectorSize) + +/* +** The macro MEMDB is true if we are dealing with an in-memory database. +** We do this as a macro so that if the SQLITE_OMIT_MEMORYDB macro is set, +** the value of MEMDB will be a constant and the compiler will optimize +** out code that would never execute. +*/ +#ifdef SQLITE_OMIT_MEMORYDB +# define MEMDB 0 +#else +# define MEMDB pPager->memDb +#endif + +/* +** The macro USEFETCH is true if we are allowed to use the xFetch and xUnfetch +** interfaces to access the database using memory-mapped I/O. +*/ +#if SQLITE_MAX_MMAP_SIZE>0 +# define USEFETCH(x) ((x)->bUseFetch) +#else +# define USEFETCH(x) 0 +#endif + +/* +** The maximum legal page number is (2^31 - 1). +*/ +#define PAGER_MAX_PGNO 2147483647 + +/* +** The argument to this macro is a file descriptor (type sqlite3_file*). +** Return 0 if it is not open, or non-zero (but not 1) if it is. +** +** This is so that expressions can be written as: +** +** if( isOpen(pPager->jfd) ){ ... +** +** instead of +** +** if( pPager->jfd->pMethods ){ ... +*/ +#define isOpen(pFd) ((pFd)->pMethods!=0) + +#ifdef SQLITE_DIRECT_OVERFLOW_READ +/* +** Return true if page pgno can be read directly from the database file +** by the b-tree layer. This is the case if: +** +** * the database file is open, +** * there are no dirty pages in the cache, and +** * the desired page is not currently in the wal file. +*/ +SQLITE_PRIVATE int sqlite3PagerDirectReadOk(Pager *pPager, Pgno pgno){ + if( pPager->fd->pMethods==0 ) return 0; + if( sqlite3PCacheIsDirty(pPager->pPCache) ) return 0; +#ifdef SQLITE_HAS_CODEC + if( pPager->xCodec!=0 ) return 0; +#endif +#ifndef SQLITE_OMIT_WAL + if( pPager->pWal ){ + u32 iRead = 0; + int rc; + rc = sqlite3WalFindFrame(pPager->pWal, pgno, &iRead); + return (rc==SQLITE_OK && iRead==0); + } +#endif + return 1; +} +#endif + +#ifndef SQLITE_OMIT_WAL +# define pagerUseWal(x) ((x)->pWal!=0) +#else +# define pagerUseWal(x) 0 +# define pagerRollbackWal(x) 0 +# define pagerWalFrames(v,w,x,y) 0 +# define pagerOpenWalIfPresent(z) SQLITE_OK +# define pagerBeginReadTransaction(z) SQLITE_OK +#endif + +#ifndef NDEBUG +/* +** Usage: +** +** assert( assert_pager_state(pPager) ); +** +** This function runs many asserts to try to find inconsistencies in +** the internal state of the Pager object. +*/ +static int assert_pager_state(Pager *p){ + Pager *pPager = p; + + /* State must be valid. */ + assert( p->eState==PAGER_OPEN + || p->eState==PAGER_READER + || p->eState==PAGER_WRITER_LOCKED + || p->eState==PAGER_WRITER_CACHEMOD + || p->eState==PAGER_WRITER_DBMOD + || p->eState==PAGER_WRITER_FINISHED + || p->eState==PAGER_ERROR + ); + + /* Regardless of the current state, a temp-file connection always behaves + ** as if it has an exclusive lock on the database file. It never updates + ** the change-counter field, so the changeCountDone flag is always set. + */ + assert( p->tempFile==0 || p->eLock==EXCLUSIVE_LOCK ); + assert( p->tempFile==0 || pPager->changeCountDone ); + + /* If the useJournal flag is clear, the journal-mode must be "OFF". + ** And if the journal-mode is "OFF", the journal file must not be open. + */ + assert( p->journalMode==PAGER_JOURNALMODE_OFF || p->useJournal ); + assert( p->journalMode!=PAGER_JOURNALMODE_OFF || !isOpen(p->jfd) ); + + /* Check that MEMDB implies noSync. And an in-memory journal. Since + ** this means an in-memory pager performs no IO at all, it cannot encounter + ** either SQLITE_IOERR or SQLITE_FULL during rollback or while finalizing + ** a journal file. (although the in-memory journal implementation may + ** return SQLITE_IOERR_NOMEM while the journal file is being written). It + ** is therefore not possible for an in-memory pager to enter the ERROR + ** state. + */ + if( MEMDB ){ + assert( !isOpen(p->fd) ); + assert( p->noSync ); + assert( p->journalMode==PAGER_JOURNALMODE_OFF + || p->journalMode==PAGER_JOURNALMODE_MEMORY + ); + assert( p->eState!=PAGER_ERROR && p->eState!=PAGER_OPEN ); + assert( pagerUseWal(p)==0 ); + } + + /* If changeCountDone is set, a RESERVED lock or greater must be held + ** on the file. + */ + assert( pPager->changeCountDone==0 || pPager->eLock>=RESERVED_LOCK ); + assert( p->eLock!=PENDING_LOCK ); + + switch( p->eState ){ + case PAGER_OPEN: + assert( !MEMDB ); + assert( pPager->errCode==SQLITE_OK ); + assert( sqlite3PcacheRefCount(pPager->pPCache)==0 || pPager->tempFile ); + break; + + case PAGER_READER: + assert( pPager->errCode==SQLITE_OK ); + assert( p->eLock!=UNKNOWN_LOCK ); + assert( p->eLock>=SHARED_LOCK ); + break; + + case PAGER_WRITER_LOCKED: + assert( p->eLock!=UNKNOWN_LOCK ); + assert( pPager->errCode==SQLITE_OK ); + if( !pagerUseWal(pPager) ){ + assert( p->eLock>=RESERVED_LOCK ); + } + assert( pPager->dbSize==pPager->dbOrigSize ); + assert( pPager->dbOrigSize==pPager->dbFileSize ); + assert( pPager->dbOrigSize==pPager->dbHintSize ); + assert( pPager->setMaster==0 ); + break; + + case PAGER_WRITER_CACHEMOD: + assert( p->eLock!=UNKNOWN_LOCK ); + assert( pPager->errCode==SQLITE_OK ); + if( !pagerUseWal(pPager) ){ + /* It is possible that if journal_mode=wal here that neither the + ** journal file nor the WAL file are open. This happens during + ** a rollback transaction that switches from journal_mode=off + ** to journal_mode=wal. + */ + assert( p->eLock>=RESERVED_LOCK ); + assert( isOpen(p->jfd) + || p->journalMode==PAGER_JOURNALMODE_OFF + || p->journalMode==PAGER_JOURNALMODE_WAL + ); + } + assert( pPager->dbOrigSize==pPager->dbFileSize ); + assert( pPager->dbOrigSize==pPager->dbHintSize ); + break; + + case PAGER_WRITER_DBMOD: + assert( p->eLock==EXCLUSIVE_LOCK ); + assert( pPager->errCode==SQLITE_OK ); + assert( !pagerUseWal(pPager) ); + assert( p->eLock>=EXCLUSIVE_LOCK ); + assert( isOpen(p->jfd) + || p->journalMode==PAGER_JOURNALMODE_OFF + || p->journalMode==PAGER_JOURNALMODE_WAL + || (sqlite3OsDeviceCharacteristics(p->fd)&SQLITE_IOCAP_BATCH_ATOMIC) + ); + assert( pPager->dbOrigSize<=pPager->dbHintSize ); + break; + + case PAGER_WRITER_FINISHED: + assert( p->eLock==EXCLUSIVE_LOCK ); + assert( pPager->errCode==SQLITE_OK ); + assert( !pagerUseWal(pPager) ); + assert( isOpen(p->jfd) + || p->journalMode==PAGER_JOURNALMODE_OFF + || p->journalMode==PAGER_JOURNALMODE_WAL + || (sqlite3OsDeviceCharacteristics(p->fd)&SQLITE_IOCAP_BATCH_ATOMIC) + ); + break; + + case PAGER_ERROR: + /* There must be at least one outstanding reference to the pager if + ** in ERROR state. Otherwise the pager should have already dropped + ** back to OPEN state. + */ + assert( pPager->errCode!=SQLITE_OK ); + assert( sqlite3PcacheRefCount(pPager->pPCache)>0 || pPager->tempFile ); + break; + } + + return 1; +} +#endif /* ifndef NDEBUG */ + +#ifdef SQLITE_DEBUG +/* +** Return a pointer to a human readable string in a static buffer +** containing the state of the Pager object passed as an argument. This +** is intended to be used within debuggers. For example, as an alternative +** to "print *pPager" in gdb: +** +** (gdb) printf "%s", print_pager_state(pPager) +** +** This routine has external linkage in order to suppress compiler warnings +** about an unused function. It is enclosed within SQLITE_DEBUG and so does +** not appear in normal builds. +*/ +char *print_pager_state(Pager *p){ + static char zRet[1024]; + + sqlite3_snprintf(1024, zRet, + "Filename: %s\n" + "State: %s errCode=%d\n" + "Lock: %s\n" + "Locking mode: locking_mode=%s\n" + "Journal mode: journal_mode=%s\n" + "Backing store: tempFile=%d memDb=%d useJournal=%d\n" + "Journal: journalOff=%lld journalHdr=%lld\n" + "Size: dbsize=%d dbOrigSize=%d dbFileSize=%d\n" + , p->zFilename + , p->eState==PAGER_OPEN ? "OPEN" : + p->eState==PAGER_READER ? "READER" : + p->eState==PAGER_WRITER_LOCKED ? "WRITER_LOCKED" : + p->eState==PAGER_WRITER_CACHEMOD ? "WRITER_CACHEMOD" : + p->eState==PAGER_WRITER_DBMOD ? "WRITER_DBMOD" : + p->eState==PAGER_WRITER_FINISHED ? "WRITER_FINISHED" : + p->eState==PAGER_ERROR ? "ERROR" : "?error?" + , (int)p->errCode + , p->eLock==NO_LOCK ? "NO_LOCK" : + p->eLock==RESERVED_LOCK ? "RESERVED" : + p->eLock==EXCLUSIVE_LOCK ? "EXCLUSIVE" : + p->eLock==SHARED_LOCK ? "SHARED" : + p->eLock==UNKNOWN_LOCK ? "UNKNOWN" : "?error?" + , p->exclusiveMode ? "exclusive" : "normal" + , p->journalMode==PAGER_JOURNALMODE_MEMORY ? "memory" : + p->journalMode==PAGER_JOURNALMODE_OFF ? "off" : + p->journalMode==PAGER_JOURNALMODE_DELETE ? "delete" : + p->journalMode==PAGER_JOURNALMODE_PERSIST ? "persist" : + p->journalMode==PAGER_JOURNALMODE_TRUNCATE ? "truncate" : + p->journalMode==PAGER_JOURNALMODE_WAL ? "wal" : "?error?" + , (int)p->tempFile, (int)p->memDb, (int)p->useJournal + , p->journalOff, p->journalHdr + , (int)p->dbSize, (int)p->dbOrigSize, (int)p->dbFileSize + ); + + return zRet; +} +#endif + +/* Forward references to the various page getters */ +static int getPageNormal(Pager*,Pgno,DbPage**,int); +static int getPageError(Pager*,Pgno,DbPage**,int); +#if SQLITE_MAX_MMAP_SIZE>0 +static int getPageMMap(Pager*,Pgno,DbPage**,int); +#endif + +/* +** Set the Pager.xGet method for the appropriate routine used to fetch +** content from the pager. +*/ +static void setGetterMethod(Pager *pPager){ + if( pPager->errCode ){ + pPager->xGet = getPageError; +#if SQLITE_MAX_MMAP_SIZE>0 + }else if( USEFETCH(pPager) +#ifdef SQLITE_HAS_CODEC + && pPager->xCodec==0 +#endif + ){ + pPager->xGet = getPageMMap; +#endif /* SQLITE_MAX_MMAP_SIZE>0 */ + }else{ + pPager->xGet = getPageNormal; + } +} + +/* +** Return true if it is necessary to write page *pPg into the sub-journal. +** A page needs to be written into the sub-journal if there exists one +** or more open savepoints for which: +** +** * The page-number is less than or equal to PagerSavepoint.nOrig, and +** * The bit corresponding to the page-number is not set in +** PagerSavepoint.pInSavepoint. +*/ +static int subjRequiresPage(PgHdr *pPg){ + Pager *pPager = pPg->pPager; + PagerSavepoint *p; + Pgno pgno = pPg->pgno; + int i; + for(i=0; inSavepoint; i++){ + p = &pPager->aSavepoint[i]; + if( p->nOrig>=pgno && 0==sqlite3BitvecTestNotNull(p->pInSavepoint, pgno) ){ + return 1; + } + } + return 0; +} + +#ifdef SQLITE_DEBUG +/* +** Return true if the page is already in the journal file. +*/ +static int pageInJournal(Pager *pPager, PgHdr *pPg){ + return sqlite3BitvecTest(pPager->pInJournal, pPg->pgno); +} +#endif + +/* +** Read a 32-bit integer from the given file descriptor. Store the integer +** that is read in *pRes. Return SQLITE_OK if everything worked, or an +** error code is something goes wrong. +** +** All values are stored on disk as big-endian. +*/ +static int read32bits(sqlite3_file *fd, i64 offset, u32 *pRes){ + unsigned char ac[4]; + int rc = sqlite3OsRead(fd, ac, sizeof(ac), offset); + if( rc==SQLITE_OK ){ + *pRes = sqlite3Get4byte(ac); + } + return rc; +} + +/* +** Write a 32-bit integer into a string buffer in big-endian byte order. +*/ +#define put32bits(A,B) sqlite3Put4byte((u8*)A,B) + + +/* +** Write a 32-bit integer into the given file descriptor. Return SQLITE_OK +** on success or an error code is something goes wrong. +*/ +static int write32bits(sqlite3_file *fd, i64 offset, u32 val){ + char ac[4]; + put32bits(ac, val); + return sqlite3OsWrite(fd, ac, 4, offset); +} + +/* +** Unlock the database file to level eLock, which must be either NO_LOCK +** or SHARED_LOCK. Regardless of whether or not the call to xUnlock() +** succeeds, set the Pager.eLock variable to match the (attempted) new lock. +** +** Except, if Pager.eLock is set to UNKNOWN_LOCK when this function is +** called, do not modify it. See the comment above the #define of +** UNKNOWN_LOCK for an explanation of this. +*/ +static int pagerUnlockDb(Pager *pPager, int eLock){ + int rc = SQLITE_OK; + + assert( !pPager->exclusiveMode || pPager->eLock==eLock ); + assert( eLock==NO_LOCK || eLock==SHARED_LOCK ); + assert( eLock!=NO_LOCK || pagerUseWal(pPager)==0 ); + if( isOpen(pPager->fd) ){ + assert( pPager->eLock>=eLock ); + rc = pPager->noLock ? SQLITE_OK : sqlite3OsUnlock(pPager->fd, eLock); + if( pPager->eLock!=UNKNOWN_LOCK ){ + pPager->eLock = (u8)eLock; + } + IOTRACE(("UNLOCK %p %d\n", pPager, eLock)) + } + return rc; +} + +/* +** Lock the database file to level eLock, which must be either SHARED_LOCK, +** RESERVED_LOCK or EXCLUSIVE_LOCK. If the caller is successful, set the +** Pager.eLock variable to the new locking state. +** +** Except, if Pager.eLock is set to UNKNOWN_LOCK when this function is +** called, do not modify it unless the new locking state is EXCLUSIVE_LOCK. +** See the comment above the #define of UNKNOWN_LOCK for an explanation +** of this. +*/ +static int pagerLockDb(Pager *pPager, int eLock){ + int rc = SQLITE_OK; + + assert( eLock==SHARED_LOCK || eLock==RESERVED_LOCK || eLock==EXCLUSIVE_LOCK ); + if( pPager->eLockeLock==UNKNOWN_LOCK ){ + rc = pPager->noLock ? SQLITE_OK : sqlite3OsLock(pPager->fd, eLock); + if( rc==SQLITE_OK && (pPager->eLock!=UNKNOWN_LOCK||eLock==EXCLUSIVE_LOCK) ){ + pPager->eLock = (u8)eLock; + IOTRACE(("LOCK %p %d\n", pPager, eLock)) + } + } + return rc; +} + +/* +** This function determines whether or not the atomic-write or +** atomic-batch-write optimizations can be used with this pager. The +** atomic-write optimization can be used if: +** +** (a) the value returned by OsDeviceCharacteristics() indicates that +** a database page may be written atomically, and +** (b) the value returned by OsSectorSize() is less than or equal +** to the page size. +** +** If it can be used, then the value returned is the size of the journal +** file when it contains rollback data for exactly one page. +** +** The atomic-batch-write optimization can be used if OsDeviceCharacteristics() +** returns a value with the SQLITE_IOCAP_BATCH_ATOMIC bit set. -1 is +** returned in this case. +** +** If neither optimization can be used, 0 is returned. +*/ +static int jrnlBufferSize(Pager *pPager){ + assert( !MEMDB ); + +#if defined(SQLITE_ENABLE_ATOMIC_WRITE) \ + || defined(SQLITE_ENABLE_BATCH_ATOMIC_WRITE) + int dc; /* Device characteristics */ + + assert( isOpen(pPager->fd) ); + dc = sqlite3OsDeviceCharacteristics(pPager->fd); +#else + UNUSED_PARAMETER(pPager); +#endif + +#ifdef SQLITE_ENABLE_BATCH_ATOMIC_WRITE + if( pPager->dbSize>0 && (dc&SQLITE_IOCAP_BATCH_ATOMIC) ){ + return -1; + } +#endif + +#ifdef SQLITE_ENABLE_ATOMIC_WRITE + { + int nSector = pPager->sectorSize; + int szPage = pPager->pageSize; + + assert(SQLITE_IOCAP_ATOMIC512==(512>>8)); + assert(SQLITE_IOCAP_ATOMIC64K==(65536>>8)); + if( 0==(dc&(SQLITE_IOCAP_ATOMIC|(szPage>>8)) || nSector>szPage) ){ + return 0; + } + } + + return JOURNAL_HDR_SZ(pPager) + JOURNAL_PG_SZ(pPager); +#endif + + return 0; +} + +/* +** If SQLITE_CHECK_PAGES is defined then we do some sanity checking +** on the cache using a hash function. This is used for testing +** and debugging only. +*/ +#ifdef SQLITE_CHECK_PAGES +/* +** Return a 32-bit hash of the page data for pPage. +*/ +static u32 pager_datahash(int nByte, unsigned char *pData){ + u32 hash = 0; + int i; + for(i=0; ipPager->pageSize, (unsigned char *)pPage->pData); +} +static void pager_set_pagehash(PgHdr *pPage){ + pPage->pageHash = pager_pagehash(pPage); +} + +/* +** The CHECK_PAGE macro takes a PgHdr* as an argument. If SQLITE_CHECK_PAGES +** is defined, and NDEBUG is not defined, an assert() statement checks +** that the page is either dirty or still matches the calculated page-hash. +*/ +#define CHECK_PAGE(x) checkPage(x) +static void checkPage(PgHdr *pPg){ + Pager *pPager = pPg->pPager; + assert( pPager->eState!=PAGER_ERROR ); + assert( (pPg->flags&PGHDR_DIRTY) || pPg->pageHash==pager_pagehash(pPg) ); +} + +#else +#define pager_datahash(X,Y) 0 +#define pager_pagehash(X) 0 +#define pager_set_pagehash(X) +#define CHECK_PAGE(x) +#endif /* SQLITE_CHECK_PAGES */ + +/* +** When this is called the journal file for pager pPager must be open. +** This function attempts to read a master journal file name from the +** end of the file and, if successful, copies it into memory supplied +** by the caller. See comments above writeMasterJournal() for the format +** used to store a master journal file name at the end of a journal file. +** +** zMaster must point to a buffer of at least nMaster bytes allocated by +** the caller. This should be sqlite3_vfs.mxPathname+1 (to ensure there is +** enough space to write the master journal name). If the master journal +** name in the journal is longer than nMaster bytes (including a +** nul-terminator), then this is handled as if no master journal name +** were present in the journal. +** +** If a master journal file name is present at the end of the journal +** file, then it is copied into the buffer pointed to by zMaster. A +** nul-terminator byte is appended to the buffer following the master +** journal file name. +** +** If it is determined that no master journal file name is present +** zMaster[0] is set to 0 and SQLITE_OK returned. +** +** If an error occurs while reading from the journal file, an SQLite +** error code is returned. +*/ +static int readMasterJournal(sqlite3_file *pJrnl, char *zMaster, u32 nMaster){ + int rc; /* Return code */ + u32 len; /* Length in bytes of master journal name */ + i64 szJ; /* Total size in bytes of journal file pJrnl */ + u32 cksum; /* MJ checksum value read from journal */ + u32 u; /* Unsigned loop counter */ + unsigned char aMagic[8]; /* A buffer to hold the magic header */ + zMaster[0] = '\0'; + + if( SQLITE_OK!=(rc = sqlite3OsFileSize(pJrnl, &szJ)) + || szJ<16 + || SQLITE_OK!=(rc = read32bits(pJrnl, szJ-16, &len)) + || len>=nMaster + || len>szJ-16 + || len==0 + || SQLITE_OK!=(rc = read32bits(pJrnl, szJ-12, &cksum)) + || SQLITE_OK!=(rc = sqlite3OsRead(pJrnl, aMagic, 8, szJ-8)) + || memcmp(aMagic, aJournalMagic, 8) + || SQLITE_OK!=(rc = sqlite3OsRead(pJrnl, zMaster, len, szJ-16-len)) + ){ + return rc; + } + + /* See if the checksum matches the master journal name */ + for(u=0; ujournalOff, assuming a sector +** size of pPager->sectorSize bytes. +** +** i.e for a sector size of 512: +** +** Pager.journalOff Return value +** --------------------------------------- +** 0 0 +** 512 512 +** 100 512 +** 2000 2048 +** +*/ +static i64 journalHdrOffset(Pager *pPager){ + i64 offset = 0; + i64 c = pPager->journalOff; + if( c ){ + offset = ((c-1)/JOURNAL_HDR_SZ(pPager) + 1) * JOURNAL_HDR_SZ(pPager); + } + assert( offset%JOURNAL_HDR_SZ(pPager)==0 ); + assert( offset>=c ); + assert( (offset-c)jfd) ); + assert( !sqlite3JournalIsInMemory(pPager->jfd) ); + if( pPager->journalOff ){ + const i64 iLimit = pPager->journalSizeLimit; /* Local cache of jsl */ + + IOTRACE(("JZEROHDR %p\n", pPager)) + if( doTruncate || iLimit==0 ){ + rc = sqlite3OsTruncate(pPager->jfd, 0); + }else{ + static const char zeroHdr[28] = {0}; + rc = sqlite3OsWrite(pPager->jfd, zeroHdr, sizeof(zeroHdr), 0); + } + if( rc==SQLITE_OK && !pPager->noSync ){ + rc = sqlite3OsSync(pPager->jfd, SQLITE_SYNC_DATAONLY|pPager->syncFlags); + } + + /* At this point the transaction is committed but the write lock + ** is still held on the file. If there is a size limit configured for + ** the persistent journal and the journal file currently consumes more + ** space than that limit allows for, truncate it now. There is no need + ** to sync the file following this operation. + */ + if( rc==SQLITE_OK && iLimit>0 ){ + i64 sz; + rc = sqlite3OsFileSize(pPager->jfd, &sz); + if( rc==SQLITE_OK && sz>iLimit ){ + rc = sqlite3OsTruncate(pPager->jfd, iLimit); + } + } + } + return rc; +} + +/* +** The journal file must be open when this routine is called. A journal +** header (JOURNAL_HDR_SZ bytes) is written into the journal file at the +** current location. +** +** The format for the journal header is as follows: +** - 8 bytes: Magic identifying journal format. +** - 4 bytes: Number of records in journal, or -1 no-sync mode is on. +** - 4 bytes: Random number used for page hash. +** - 4 bytes: Initial database page count. +** - 4 bytes: Sector size used by the process that wrote this journal. +** - 4 bytes: Database page size. +** +** Followed by (JOURNAL_HDR_SZ - 28) bytes of unused space. +*/ +static int writeJournalHdr(Pager *pPager){ + int rc = SQLITE_OK; /* Return code */ + char *zHeader = pPager->pTmpSpace; /* Temporary space used to build header */ + u32 nHeader = (u32)pPager->pageSize;/* Size of buffer pointed to by zHeader */ + u32 nWrite; /* Bytes of header sector written */ + int ii; /* Loop counter */ + + assert( isOpen(pPager->jfd) ); /* Journal file must be open. */ + + if( nHeader>JOURNAL_HDR_SZ(pPager) ){ + nHeader = JOURNAL_HDR_SZ(pPager); + } + + /* If there are active savepoints and any of them were created + ** since the most recent journal header was written, update the + ** PagerSavepoint.iHdrOffset fields now. + */ + for(ii=0; iinSavepoint; ii++){ + if( pPager->aSavepoint[ii].iHdrOffset==0 ){ + pPager->aSavepoint[ii].iHdrOffset = pPager->journalOff; + } + } + + pPager->journalHdr = pPager->journalOff = journalHdrOffset(pPager); + + /* + ** Write the nRec Field - the number of page records that follow this + ** journal header. Normally, zero is written to this value at this time. + ** After the records are added to the journal (and the journal synced, + ** if in full-sync mode), the zero is overwritten with the true number + ** of records (see syncJournal()). + ** + ** A faster alternative is to write 0xFFFFFFFF to the nRec field. When + ** reading the journal this value tells SQLite to assume that the + ** rest of the journal file contains valid page records. This assumption + ** is dangerous, as if a failure occurred whilst writing to the journal + ** file it may contain some garbage data. There are two scenarios + ** where this risk can be ignored: + ** + ** * When the pager is in no-sync mode. Corruption can follow a + ** power failure in this case anyway. + ** + ** * When the SQLITE_IOCAP_SAFE_APPEND flag is set. This guarantees + ** that garbage data is never appended to the journal file. + */ + assert( isOpen(pPager->fd) || pPager->noSync ); + if( pPager->noSync || (pPager->journalMode==PAGER_JOURNALMODE_MEMORY) + || (sqlite3OsDeviceCharacteristics(pPager->fd)&SQLITE_IOCAP_SAFE_APPEND) + ){ + memcpy(zHeader, aJournalMagic, sizeof(aJournalMagic)); + put32bits(&zHeader[sizeof(aJournalMagic)], 0xffffffff); + }else{ + memset(zHeader, 0, sizeof(aJournalMagic)+4); + } + + /* The random check-hash initializer */ + sqlite3_randomness(sizeof(pPager->cksumInit), &pPager->cksumInit); + put32bits(&zHeader[sizeof(aJournalMagic)+4], pPager->cksumInit); + /* The initial database size */ + put32bits(&zHeader[sizeof(aJournalMagic)+8], pPager->dbOrigSize); + /* The assumed sector size for this process */ + put32bits(&zHeader[sizeof(aJournalMagic)+12], pPager->sectorSize); + + /* The page size */ + put32bits(&zHeader[sizeof(aJournalMagic)+16], pPager->pageSize); + + /* Initializing the tail of the buffer is not necessary. Everything + ** works find if the following memset() is omitted. But initializing + ** the memory prevents valgrind from complaining, so we are willing to + ** take the performance hit. + */ + memset(&zHeader[sizeof(aJournalMagic)+20], 0, + nHeader-(sizeof(aJournalMagic)+20)); + + /* In theory, it is only necessary to write the 28 bytes that the + ** journal header consumes to the journal file here. Then increment the + ** Pager.journalOff variable by JOURNAL_HDR_SZ so that the next + ** record is written to the following sector (leaving a gap in the file + ** that will be implicitly filled in by the OS). + ** + ** However it has been discovered that on some systems this pattern can + ** be significantly slower than contiguously writing data to the file, + ** even if that means explicitly writing data to the block of + ** (JOURNAL_HDR_SZ - 28) bytes that will not be used. So that is what + ** is done. + ** + ** The loop is required here in case the sector-size is larger than the + ** database page size. Since the zHeader buffer is only Pager.pageSize + ** bytes in size, more than one call to sqlite3OsWrite() may be required + ** to populate the entire journal header sector. + */ + for(nWrite=0; rc==SQLITE_OK&&nWritejournalHdr, nHeader)) + rc = sqlite3OsWrite(pPager->jfd, zHeader, nHeader, pPager->journalOff); + assert( pPager->journalHdr <= pPager->journalOff ); + pPager->journalOff += nHeader; + } + + return rc; +} + +/* +** The journal file must be open when this is called. A journal header file +** (JOURNAL_HDR_SZ bytes) is read from the current location in the journal +** file. The current location in the journal file is given by +** pPager->journalOff. See comments above function writeJournalHdr() for +** a description of the journal header format. +** +** If the header is read successfully, *pNRec is set to the number of +** page records following this header and *pDbSize is set to the size of the +** database before the transaction began, in pages. Also, pPager->cksumInit +** is set to the value read from the journal header. SQLITE_OK is returned +** in this case. +** +** If the journal header file appears to be corrupted, SQLITE_DONE is +** returned and *pNRec and *PDbSize are undefined. If JOURNAL_HDR_SZ bytes +** cannot be read from the journal file an error code is returned. +*/ +static int readJournalHdr( + Pager *pPager, /* Pager object */ + int isHot, + i64 journalSize, /* Size of the open journal file in bytes */ + u32 *pNRec, /* OUT: Value read from the nRec field */ + u32 *pDbSize /* OUT: Value of original database size field */ +){ + int rc; /* Return code */ + unsigned char aMagic[8]; /* A buffer to hold the magic header */ + i64 iHdrOff; /* Offset of journal header being read */ + + assert( isOpen(pPager->jfd) ); /* Journal file must be open. */ + + /* Advance Pager.journalOff to the start of the next sector. If the + ** journal file is too small for there to be a header stored at this + ** point, return SQLITE_DONE. + */ + pPager->journalOff = journalHdrOffset(pPager); + if( pPager->journalOff+JOURNAL_HDR_SZ(pPager) > journalSize ){ + return SQLITE_DONE; + } + iHdrOff = pPager->journalOff; + + /* Read in the first 8 bytes of the journal header. If they do not match + ** the magic string found at the start of each journal header, return + ** SQLITE_DONE. If an IO error occurs, return an error code. Otherwise, + ** proceed. + */ + if( isHot || iHdrOff!=pPager->journalHdr ){ + rc = sqlite3OsRead(pPager->jfd, aMagic, sizeof(aMagic), iHdrOff); + if( rc ){ + return rc; + } + if( memcmp(aMagic, aJournalMagic, sizeof(aMagic))!=0 ){ + return SQLITE_DONE; + } + } + + /* Read the first three 32-bit fields of the journal header: The nRec + ** field, the checksum-initializer and the database size at the start + ** of the transaction. Return an error code if anything goes wrong. + */ + if( SQLITE_OK!=(rc = read32bits(pPager->jfd, iHdrOff+8, pNRec)) + || SQLITE_OK!=(rc = read32bits(pPager->jfd, iHdrOff+12, &pPager->cksumInit)) + || SQLITE_OK!=(rc = read32bits(pPager->jfd, iHdrOff+16, pDbSize)) + ){ + return rc; + } + + if( pPager->journalOff==0 ){ + u32 iPageSize; /* Page-size field of journal header */ + u32 iSectorSize; /* Sector-size field of journal header */ + + /* Read the page-size and sector-size journal header fields. */ + if( SQLITE_OK!=(rc = read32bits(pPager->jfd, iHdrOff+20, &iSectorSize)) + || SQLITE_OK!=(rc = read32bits(pPager->jfd, iHdrOff+24, &iPageSize)) + ){ + return rc; + } + + /* Versions of SQLite prior to 3.5.8 set the page-size field of the + ** journal header to zero. In this case, assume that the Pager.pageSize + ** variable is already set to the correct page size. + */ + if( iPageSize==0 ){ + iPageSize = pPager->pageSize; + } + + /* Check that the values read from the page-size and sector-size fields + ** are within range. To be 'in range', both values need to be a power + ** of two greater than or equal to 512 or 32, and not greater than their + ** respective compile time maximum limits. + */ + if( iPageSize<512 || iSectorSize<32 + || iPageSize>SQLITE_MAX_PAGE_SIZE || iSectorSize>MAX_SECTOR_SIZE + || ((iPageSize-1)&iPageSize)!=0 || ((iSectorSize-1)&iSectorSize)!=0 + ){ + /* If the either the page-size or sector-size in the journal-header is + ** invalid, then the process that wrote the journal-header must have + ** crashed before the header was synced. In this case stop reading + ** the journal file here. + */ + return SQLITE_DONE; + } + + /* Update the page-size to match the value read from the journal. + ** Use a testcase() macro to make sure that malloc failure within + ** PagerSetPagesize() is tested. + */ + rc = sqlite3PagerSetPagesize(pPager, &iPageSize, -1); + testcase( rc!=SQLITE_OK ); + + /* Update the assumed sector-size to match the value used by + ** the process that created this journal. If this journal was + ** created by a process other than this one, then this routine + ** is being called from within pager_playback(). The local value + ** of Pager.sectorSize is restored at the end of that routine. + */ + pPager->sectorSize = iSectorSize; + } + + pPager->journalOff += JOURNAL_HDR_SZ(pPager); + return rc; +} + + +/* +** Write the supplied master journal name into the journal file for pager +** pPager at the current location. The master journal name must be the last +** thing written to a journal file. If the pager is in full-sync mode, the +** journal file descriptor is advanced to the next sector boundary before +** anything is written. The format is: +** +** + 4 bytes: PAGER_MJ_PGNO. +** + N bytes: Master journal filename in utf-8. +** + 4 bytes: N (length of master journal name in bytes, no nul-terminator). +** + 4 bytes: Master journal name checksum. +** + 8 bytes: aJournalMagic[]. +** +** The master journal page checksum is the sum of the bytes in the master +** journal name, where each byte is interpreted as a signed 8-bit integer. +** +** If zMaster is a NULL pointer (occurs for a single database transaction), +** this call is a no-op. +*/ +static int writeMasterJournal(Pager *pPager, const char *zMaster){ + int rc; /* Return code */ + int nMaster; /* Length of string zMaster */ + i64 iHdrOff; /* Offset of header in journal file */ + i64 jrnlSize; /* Size of journal file on disk */ + u32 cksum = 0; /* Checksum of string zMaster */ + + assert( pPager->setMaster==0 ); + assert( !pagerUseWal(pPager) ); + + if( !zMaster + || pPager->journalMode==PAGER_JOURNALMODE_MEMORY + || !isOpen(pPager->jfd) + ){ + return SQLITE_OK; + } + pPager->setMaster = 1; + assert( pPager->journalHdr <= pPager->journalOff ); + + /* Calculate the length in bytes and the checksum of zMaster */ + for(nMaster=0; zMaster[nMaster]; nMaster++){ + cksum += zMaster[nMaster]; + } + + /* If in full-sync mode, advance to the next disk sector before writing + ** the master journal name. This is in case the previous page written to + ** the journal has already been synced. + */ + if( pPager->fullSync ){ + pPager->journalOff = journalHdrOffset(pPager); + } + iHdrOff = pPager->journalOff; + + /* Write the master journal data to the end of the journal file. If + ** an error occurs, return the error code to the caller. + */ + if( (0 != (rc = write32bits(pPager->jfd, iHdrOff, PAGER_MJ_PGNO(pPager)))) + || (0 != (rc = sqlite3OsWrite(pPager->jfd, zMaster, nMaster, iHdrOff+4))) + || (0 != (rc = write32bits(pPager->jfd, iHdrOff+4+nMaster, nMaster))) + || (0 != (rc = write32bits(pPager->jfd, iHdrOff+4+nMaster+4, cksum))) + || (0 != (rc = sqlite3OsWrite(pPager->jfd, aJournalMagic, 8, + iHdrOff+4+nMaster+8))) + ){ + return rc; + } + pPager->journalOff += (nMaster+20); + + /* If the pager is in peristent-journal mode, then the physical + ** journal-file may extend past the end of the master-journal name + ** and 8 bytes of magic data just written to the file. This is + ** dangerous because the code to rollback a hot-journal file + ** will not be able to find the master-journal name to determine + ** whether or not the journal is hot. + ** + ** Easiest thing to do in this scenario is to truncate the journal + ** file to the required size. + */ + if( SQLITE_OK==(rc = sqlite3OsFileSize(pPager->jfd, &jrnlSize)) + && jrnlSize>pPager->journalOff + ){ + rc = sqlite3OsTruncate(pPager->jfd, pPager->journalOff); + } + return rc; +} + +/* +** Discard the entire contents of the in-memory page-cache. +*/ +static void pager_reset(Pager *pPager){ + pPager->iDataVersion++; + sqlite3BackupRestart(pPager->pBackup); + sqlite3PcacheClear(pPager->pPCache); +} + +/* +** Return the pPager->iDataVersion value +*/ +SQLITE_PRIVATE u32 sqlite3PagerDataVersion(Pager *pPager){ + return pPager->iDataVersion; +} + +/* +** Free all structures in the Pager.aSavepoint[] array and set both +** Pager.aSavepoint and Pager.nSavepoint to zero. Close the sub-journal +** if it is open and the pager is not in exclusive mode. +*/ +static void releaseAllSavepoints(Pager *pPager){ + int ii; /* Iterator for looping through Pager.aSavepoint */ + for(ii=0; iinSavepoint; ii++){ + sqlite3BitvecDestroy(pPager->aSavepoint[ii].pInSavepoint); + } + if( !pPager->exclusiveMode || sqlite3JournalIsInMemory(pPager->sjfd) ){ + sqlite3OsClose(pPager->sjfd); + } + sqlite3_free(pPager->aSavepoint); + pPager->aSavepoint = 0; + pPager->nSavepoint = 0; + pPager->nSubRec = 0; +} + +/* +** Set the bit number pgno in the PagerSavepoint.pInSavepoint +** bitvecs of all open savepoints. Return SQLITE_OK if successful +** or SQLITE_NOMEM if a malloc failure occurs. +*/ +static int addToSavepointBitvecs(Pager *pPager, Pgno pgno){ + int ii; /* Loop counter */ + int rc = SQLITE_OK; /* Result code */ + + for(ii=0; iinSavepoint; ii++){ + PagerSavepoint *p = &pPager->aSavepoint[ii]; + if( pgno<=p->nOrig ){ + rc |= sqlite3BitvecSet(p->pInSavepoint, pgno); + testcase( rc==SQLITE_NOMEM ); + assert( rc==SQLITE_OK || rc==SQLITE_NOMEM ); + } + } + return rc; +} + +/* +** This function is a no-op if the pager is in exclusive mode and not +** in the ERROR state. Otherwise, it switches the pager to PAGER_OPEN +** state. +** +** If the pager is not in exclusive-access mode, the database file is +** completely unlocked. If the file is unlocked and the file-system does +** not exhibit the UNDELETABLE_WHEN_OPEN property, the journal file is +** closed (if it is open). +** +** If the pager is in ERROR state when this function is called, the +** contents of the pager cache are discarded before switching back to +** the OPEN state. Regardless of whether the pager is in exclusive-mode +** or not, any journal file left in the file-system will be treated +** as a hot-journal and rolled back the next time a read-transaction +** is opened (by this or by any other connection). +*/ +static void pager_unlock(Pager *pPager){ + + assert( pPager->eState==PAGER_READER + || pPager->eState==PAGER_OPEN + || pPager->eState==PAGER_ERROR + ); + + sqlite3BitvecDestroy(pPager->pInJournal); + pPager->pInJournal = 0; + releaseAllSavepoints(pPager); + + if( pagerUseWal(pPager) ){ + assert( !isOpen(pPager->jfd) ); + sqlite3WalEndReadTransaction(pPager->pWal); + pPager->eState = PAGER_OPEN; + }else if( !pPager->exclusiveMode ){ + int rc; /* Error code returned by pagerUnlockDb() */ + int iDc = isOpen(pPager->fd)?sqlite3OsDeviceCharacteristics(pPager->fd):0; + + /* If the operating system support deletion of open files, then + ** close the journal file when dropping the database lock. Otherwise + ** another connection with journal_mode=delete might delete the file + ** out from under us. + */ + assert( (PAGER_JOURNALMODE_MEMORY & 5)!=1 ); + assert( (PAGER_JOURNALMODE_OFF & 5)!=1 ); + assert( (PAGER_JOURNALMODE_WAL & 5)!=1 ); + assert( (PAGER_JOURNALMODE_DELETE & 5)!=1 ); + assert( (PAGER_JOURNALMODE_TRUNCATE & 5)==1 ); + assert( (PAGER_JOURNALMODE_PERSIST & 5)==1 ); + if( 0==(iDc & SQLITE_IOCAP_UNDELETABLE_WHEN_OPEN) + || 1!=(pPager->journalMode & 5) + ){ + sqlite3OsClose(pPager->jfd); + } + + /* If the pager is in the ERROR state and the call to unlock the database + ** file fails, set the current lock to UNKNOWN_LOCK. See the comment + ** above the #define for UNKNOWN_LOCK for an explanation of why this + ** is necessary. + */ + rc = pagerUnlockDb(pPager, NO_LOCK); + if( rc!=SQLITE_OK && pPager->eState==PAGER_ERROR ){ + pPager->eLock = UNKNOWN_LOCK; + } + + /* The pager state may be changed from PAGER_ERROR to PAGER_OPEN here + ** without clearing the error code. This is intentional - the error + ** code is cleared and the cache reset in the block below. + */ + assert( pPager->errCode || pPager->eState!=PAGER_ERROR ); + pPager->changeCountDone = 0; + pPager->eState = PAGER_OPEN; + } + + /* If Pager.errCode is set, the contents of the pager cache cannot be + ** trusted. Now that there are no outstanding references to the pager, + ** it can safely move back to PAGER_OPEN state. This happens in both + ** normal and exclusive-locking mode. + */ + assert( pPager->errCode==SQLITE_OK || !MEMDB ); + if( pPager->errCode ){ + if( pPager->tempFile==0 ){ + pager_reset(pPager); + pPager->changeCountDone = 0; + pPager->eState = PAGER_OPEN; + }else{ + pPager->eState = (isOpen(pPager->jfd) ? PAGER_OPEN : PAGER_READER); + } + if( USEFETCH(pPager) ) sqlite3OsUnfetch(pPager->fd, 0, 0); + pPager->errCode = SQLITE_OK; + setGetterMethod(pPager); + } + + pPager->journalOff = 0; + pPager->journalHdr = 0; + pPager->setMaster = 0; +} + +/* +** This function is called whenever an IOERR or FULL error that requires +** the pager to transition into the ERROR state may ahve occurred. +** The first argument is a pointer to the pager structure, the second +** the error-code about to be returned by a pager API function. The +** value returned is a copy of the second argument to this function. +** +** If the second argument is SQLITE_FULL, SQLITE_IOERR or one of the +** IOERR sub-codes, the pager enters the ERROR state and the error code +** is stored in Pager.errCode. While the pager remains in the ERROR state, +** all major API calls on the Pager will immediately return Pager.errCode. +** +** The ERROR state indicates that the contents of the pager-cache +** cannot be trusted. This state can be cleared by completely discarding +** the contents of the pager-cache. If a transaction was active when +** the persistent error occurred, then the rollback journal may need +** to be replayed to restore the contents of the database file (as if +** it were a hot-journal). +*/ +static int pager_error(Pager *pPager, int rc){ + int rc2 = rc & 0xff; + assert( rc==SQLITE_OK || !MEMDB ); + assert( + pPager->errCode==SQLITE_FULL || + pPager->errCode==SQLITE_OK || + (pPager->errCode & 0xff)==SQLITE_IOERR + ); + if( rc2==SQLITE_FULL || rc2==SQLITE_IOERR ){ + pPager->errCode = rc; + pPager->eState = PAGER_ERROR; + setGetterMethod(pPager); + } + return rc; +} + +static int pager_truncate(Pager *pPager, Pgno nPage); + +/* +** The write transaction open on pPager is being committed (bCommit==1) +** or rolled back (bCommit==0). +** +** Return TRUE if and only if all dirty pages should be flushed to disk. +** +** Rules: +** +** * For non-TEMP databases, always sync to disk. This is necessary +** for transactions to be durable. +** +** * Sync TEMP database only on a COMMIT (not a ROLLBACK) when the backing +** file has been created already (via a spill on pagerStress()) and +** when the number of dirty pages in memory exceeds 25% of the total +** cache size. +*/ +static int pagerFlushOnCommit(Pager *pPager, int bCommit){ + if( pPager->tempFile==0 ) return 1; + if( !bCommit ) return 0; + if( !isOpen(pPager->fd) ) return 0; + return (sqlite3PCachePercentDirty(pPager->pPCache)>=25); +} + +/* +** This routine ends a transaction. A transaction is usually ended by +** either a COMMIT or a ROLLBACK operation. This routine may be called +** after rollback of a hot-journal, or if an error occurs while opening +** the journal file or writing the very first journal-header of a +** database transaction. +** +** This routine is never called in PAGER_ERROR state. If it is called +** in PAGER_NONE or PAGER_SHARED state and the lock held is less +** exclusive than a RESERVED lock, it is a no-op. +** +** Otherwise, any active savepoints are released. +** +** If the journal file is open, then it is "finalized". Once a journal +** file has been finalized it is not possible to use it to roll back a +** transaction. Nor will it be considered to be a hot-journal by this +** or any other database connection. Exactly how a journal is finalized +** depends on whether or not the pager is running in exclusive mode and +** the current journal-mode (Pager.journalMode value), as follows: +** +** journalMode==MEMORY +** Journal file descriptor is simply closed. This destroys an +** in-memory journal. +** +** journalMode==TRUNCATE +** Journal file is truncated to zero bytes in size. +** +** journalMode==PERSIST +** The first 28 bytes of the journal file are zeroed. This invalidates +** the first journal header in the file, and hence the entire journal +** file. An invalid journal file cannot be rolled back. +** +** journalMode==DELETE +** The journal file is closed and deleted using sqlite3OsDelete(). +** +** If the pager is running in exclusive mode, this method of finalizing +** the journal file is never used. Instead, if the journalMode is +** DELETE and the pager is in exclusive mode, the method described under +** journalMode==PERSIST is used instead. +** +** After the journal is finalized, the pager moves to PAGER_READER state. +** If running in non-exclusive rollback mode, the lock on the file is +** downgraded to a SHARED_LOCK. +** +** SQLITE_OK is returned if no error occurs. If an error occurs during +** any of the IO operations to finalize the journal file or unlock the +** database then the IO error code is returned to the user. If the +** operation to finalize the journal file fails, then the code still +** tries to unlock the database file if not in exclusive mode. If the +** unlock operation fails as well, then the first error code related +** to the first error encountered (the journal finalization one) is +** returned. +*/ +static int pager_end_transaction(Pager *pPager, int hasMaster, int bCommit){ + int rc = SQLITE_OK; /* Error code from journal finalization operation */ + int rc2 = SQLITE_OK; /* Error code from db file unlock operation */ + + /* Do nothing if the pager does not have an open write transaction + ** or at least a RESERVED lock. This function may be called when there + ** is no write-transaction active but a RESERVED or greater lock is + ** held under two circumstances: + ** + ** 1. After a successful hot-journal rollback, it is called with + ** eState==PAGER_NONE and eLock==EXCLUSIVE_LOCK. + ** + ** 2. If a connection with locking_mode=exclusive holding an EXCLUSIVE + ** lock switches back to locking_mode=normal and then executes a + ** read-transaction, this function is called with eState==PAGER_READER + ** and eLock==EXCLUSIVE_LOCK when the read-transaction is closed. + */ + assert( assert_pager_state(pPager) ); + assert( pPager->eState!=PAGER_ERROR ); + if( pPager->eStateeLockjfd) || pPager->pInJournal==0 + || (sqlite3OsDeviceCharacteristics(pPager->fd)&SQLITE_IOCAP_BATCH_ATOMIC) + ); + if( isOpen(pPager->jfd) ){ + assert( !pagerUseWal(pPager) ); + + /* Finalize the journal file. */ + if( sqlite3JournalIsInMemory(pPager->jfd) ){ + /* assert( pPager->journalMode==PAGER_JOURNALMODE_MEMORY ); */ + sqlite3OsClose(pPager->jfd); + }else if( pPager->journalMode==PAGER_JOURNALMODE_TRUNCATE ){ + if( pPager->journalOff==0 ){ + rc = SQLITE_OK; + }else{ + rc = sqlite3OsTruncate(pPager->jfd, 0); + if( rc==SQLITE_OK && pPager->fullSync ){ + /* Make sure the new file size is written into the inode right away. + ** Otherwise the journal might resurrect following a power loss and + ** cause the last transaction to roll back. See + ** https://bugzilla.mozilla.org/show_bug.cgi?id=1072773 + */ + rc = sqlite3OsSync(pPager->jfd, pPager->syncFlags); + } + } + pPager->journalOff = 0; + }else if( pPager->journalMode==PAGER_JOURNALMODE_PERSIST + || (pPager->exclusiveMode && pPager->journalMode!=PAGER_JOURNALMODE_WAL) + ){ + rc = zeroJournalHdr(pPager, hasMaster||pPager->tempFile); + pPager->journalOff = 0; + }else{ + /* This branch may be executed with Pager.journalMode==MEMORY if + ** a hot-journal was just rolled back. In this case the journal + ** file should be closed and deleted. If this connection writes to + ** the database file, it will do so using an in-memory journal. + */ + int bDelete = !pPager->tempFile; + assert( sqlite3JournalIsInMemory(pPager->jfd)==0 ); + assert( pPager->journalMode==PAGER_JOURNALMODE_DELETE + || pPager->journalMode==PAGER_JOURNALMODE_MEMORY + || pPager->journalMode==PAGER_JOURNALMODE_WAL + ); + sqlite3OsClose(pPager->jfd); + if( bDelete ){ + rc = sqlite3OsDelete(pPager->pVfs, pPager->zJournal, pPager->extraSync); + } + } + } + +#ifdef SQLITE_CHECK_PAGES + sqlite3PcacheIterateDirty(pPager->pPCache, pager_set_pagehash); + if( pPager->dbSize==0 && sqlite3PcacheRefCount(pPager->pPCache)>0 ){ + PgHdr *p = sqlite3PagerLookup(pPager, 1); + if( p ){ + p->pageHash = 0; + sqlite3PagerUnrefNotNull(p); + } + } +#endif + + sqlite3BitvecDestroy(pPager->pInJournal); + pPager->pInJournal = 0; + pPager->nRec = 0; + if( rc==SQLITE_OK ){ + if( MEMDB || pagerFlushOnCommit(pPager, bCommit) ){ + sqlite3PcacheCleanAll(pPager->pPCache); + }else{ + sqlite3PcacheClearWritable(pPager->pPCache); + } + sqlite3PcacheTruncate(pPager->pPCache, pPager->dbSize); + } + + if( pagerUseWal(pPager) ){ + /* Drop the WAL write-lock, if any. Also, if the connection was in + ** locking_mode=exclusive mode but is no longer, drop the EXCLUSIVE + ** lock held on the database file. + */ + rc2 = sqlite3WalEndWriteTransaction(pPager->pWal); + assert( rc2==SQLITE_OK ); + }else if( rc==SQLITE_OK && bCommit && pPager->dbFileSize>pPager->dbSize ){ + /* This branch is taken when committing a transaction in rollback-journal + ** mode if the database file on disk is larger than the database image. + ** At this point the journal has been finalized and the transaction + ** successfully committed, but the EXCLUSIVE lock is still held on the + ** file. So it is safe to truncate the database file to its minimum + ** required size. */ + assert( pPager->eLock==EXCLUSIVE_LOCK ); + rc = pager_truncate(pPager, pPager->dbSize); + } + + if( rc==SQLITE_OK && bCommit ){ + rc = sqlite3OsFileControl(pPager->fd, SQLITE_FCNTL_COMMIT_PHASETWO, 0); + if( rc==SQLITE_NOTFOUND ) rc = SQLITE_OK; + } + + if( !pPager->exclusiveMode + && (!pagerUseWal(pPager) || sqlite3WalExclusiveMode(pPager->pWal, 0)) + ){ + rc2 = pagerUnlockDb(pPager, SHARED_LOCK); + pPager->changeCountDone = 0; + } + pPager->eState = PAGER_READER; + pPager->setMaster = 0; + + return (rc==SQLITE_OK?rc2:rc); +} + +/* +** Execute a rollback if a transaction is active and unlock the +** database file. +** +** If the pager has already entered the ERROR state, do not attempt +** the rollback at this time. Instead, pager_unlock() is called. The +** call to pager_unlock() will discard all in-memory pages, unlock +** the database file and move the pager back to OPEN state. If this +** means that there is a hot-journal left in the file-system, the next +** connection to obtain a shared lock on the pager (which may be this one) +** will roll it back. +** +** If the pager has not already entered the ERROR state, but an IO or +** malloc error occurs during a rollback, then this will itself cause +** the pager to enter the ERROR state. Which will be cleared by the +** call to pager_unlock(), as described above. +*/ +static void pagerUnlockAndRollback(Pager *pPager){ + if( pPager->eState!=PAGER_ERROR && pPager->eState!=PAGER_OPEN ){ + assert( assert_pager_state(pPager) ); + if( pPager->eState>=PAGER_WRITER_LOCKED ){ + sqlite3BeginBenignMalloc(); + sqlite3PagerRollback(pPager); + sqlite3EndBenignMalloc(); + }else if( !pPager->exclusiveMode ){ + assert( pPager->eState==PAGER_READER ); + pager_end_transaction(pPager, 0, 0); + } + } + pager_unlock(pPager); +} + +/* +** Parameter aData must point to a buffer of pPager->pageSize bytes +** of data. Compute and return a checksum based ont the contents of the +** page of data and the current value of pPager->cksumInit. +** +** This is not a real checksum. It is really just the sum of the +** random initial value (pPager->cksumInit) and every 200th byte +** of the page data, starting with byte offset (pPager->pageSize%200). +** Each byte is interpreted as an 8-bit unsigned integer. +** +** Changing the formula used to compute this checksum results in an +** incompatible journal file format. +** +** If journal corruption occurs due to a power failure, the most likely +** scenario is that one end or the other of the record will be changed. +** It is much less likely that the two ends of the journal record will be +** correct and the middle be corrupt. Thus, this "checksum" scheme, +** though fast and simple, catches the mostly likely kind of corruption. +*/ +static u32 pager_cksum(Pager *pPager, const u8 *aData){ + u32 cksum = pPager->cksumInit; /* Checksum value to return */ + int i = pPager->pageSize-200; /* Loop counter */ + while( i>0 ){ + cksum += aData[i]; + i -= 200; + } + return cksum; +} + +/* +** Report the current page size and number of reserved bytes back +** to the codec. +*/ +#ifdef SQLITE_HAS_CODEC +static void pagerReportSize(Pager *pPager){ + if( pPager->xCodecSizeChng ){ + pPager->xCodecSizeChng(pPager->pCodec, pPager->pageSize, + (int)pPager->nReserve); + } +} +#else +# define pagerReportSize(X) /* No-op if we do not support a codec */ +#endif + +#ifdef SQLITE_HAS_CODEC +/* +** Make sure the number of reserved bits is the same in the destination +** pager as it is in the source. This comes up when a VACUUM changes the +** number of reserved bits to the "optimal" amount. +*/ +SQLITE_PRIVATE void sqlite3PagerAlignReserve(Pager *pDest, Pager *pSrc){ + if( pDest->nReserve!=pSrc->nReserve ){ + pDest->nReserve = pSrc->nReserve; + pagerReportSize(pDest); + } +} +#endif + +/* +** Read a single page from either the journal file (if isMainJrnl==1) or +** from the sub-journal (if isMainJrnl==0) and playback that page. +** The page begins at offset *pOffset into the file. The *pOffset +** value is increased to the start of the next page in the journal. +** +** The main rollback journal uses checksums - the statement journal does +** not. +** +** If the page number of the page record read from the (sub-)journal file +** is greater than the current value of Pager.dbSize, then playback is +** skipped and SQLITE_OK is returned. +** +** If pDone is not NULL, then it is a record of pages that have already +** been played back. If the page at *pOffset has already been played back +** (if the corresponding pDone bit is set) then skip the playback. +** Make sure the pDone bit corresponding to the *pOffset page is set +** prior to returning. +** +** If the page record is successfully read from the (sub-)journal file +** and played back, then SQLITE_OK is returned. If an IO error occurs +** while reading the record from the (sub-)journal file or while writing +** to the database file, then the IO error code is returned. If data +** is successfully read from the (sub-)journal file but appears to be +** corrupted, SQLITE_DONE is returned. Data is considered corrupted in +** two circumstances: +** +** * If the record page-number is illegal (0 or PAGER_MJ_PGNO), or +** * If the record is being rolled back from the main journal file +** and the checksum field does not match the record content. +** +** Neither of these two scenarios are possible during a savepoint rollback. +** +** If this is a savepoint rollback, then memory may have to be dynamically +** allocated by this function. If this is the case and an allocation fails, +** SQLITE_NOMEM is returned. +*/ +static int pager_playback_one_page( + Pager *pPager, /* The pager being played back */ + i64 *pOffset, /* Offset of record to playback */ + Bitvec *pDone, /* Bitvec of pages already played back */ + int isMainJrnl, /* 1 -> main journal. 0 -> sub-journal. */ + int isSavepnt /* True for a savepoint rollback */ +){ + int rc; + PgHdr *pPg; /* An existing page in the cache */ + Pgno pgno; /* The page number of a page in journal */ + u32 cksum; /* Checksum used for sanity checking */ + char *aData; /* Temporary storage for the page */ + sqlite3_file *jfd; /* The file descriptor for the journal file */ + int isSynced; /* True if journal page is synced */ +#ifdef SQLITE_HAS_CODEC + /* The jrnlEnc flag is true if Journal pages should be passed through + ** the codec. It is false for pure in-memory journals. */ + const int jrnlEnc = (isMainJrnl || pPager->subjInMemory==0); +#endif + + assert( (isMainJrnl&~1)==0 ); /* isMainJrnl is 0 or 1 */ + assert( (isSavepnt&~1)==0 ); /* isSavepnt is 0 or 1 */ + assert( isMainJrnl || pDone ); /* pDone always used on sub-journals */ + assert( isSavepnt || pDone==0 ); /* pDone never used on non-savepoint */ + + aData = pPager->pTmpSpace; + assert( aData ); /* Temp storage must have already been allocated */ + assert( pagerUseWal(pPager)==0 || (!isMainJrnl && isSavepnt) ); + + /* Either the state is greater than PAGER_WRITER_CACHEMOD (a transaction + ** or savepoint rollback done at the request of the caller) or this is + ** a hot-journal rollback. If it is a hot-journal rollback, the pager + ** is in state OPEN and holds an EXCLUSIVE lock. Hot-journal rollback + ** only reads from the main journal, not the sub-journal. + */ + assert( pPager->eState>=PAGER_WRITER_CACHEMOD + || (pPager->eState==PAGER_OPEN && pPager->eLock==EXCLUSIVE_LOCK) + ); + assert( pPager->eState>=PAGER_WRITER_CACHEMOD || isMainJrnl ); + + /* Read the page number and page data from the journal or sub-journal + ** file. Return an error code to the caller if an IO error occurs. + */ + jfd = isMainJrnl ? pPager->jfd : pPager->sjfd; + rc = read32bits(jfd, *pOffset, &pgno); + if( rc!=SQLITE_OK ) return rc; + rc = sqlite3OsRead(jfd, (u8*)aData, pPager->pageSize, (*pOffset)+4); + if( rc!=SQLITE_OK ) return rc; + *pOffset += pPager->pageSize + 4 + isMainJrnl*4; + + /* Sanity checking on the page. This is more important that I originally + ** thought. If a power failure occurs while the journal is being written, + ** it could cause invalid data to be written into the journal. We need to + ** detect this invalid data (with high probability) and ignore it. + */ + if( pgno==0 || pgno==PAGER_MJ_PGNO(pPager) ){ + assert( !isSavepnt ); + return SQLITE_DONE; + } + if( pgno>(Pgno)pPager->dbSize || sqlite3BitvecTest(pDone, pgno) ){ + return SQLITE_OK; + } + if( isMainJrnl ){ + rc = read32bits(jfd, (*pOffset)-4, &cksum); + if( rc ) return rc; + if( !isSavepnt && pager_cksum(pPager, (u8*)aData)!=cksum ){ + return SQLITE_DONE; + } + } + + /* If this page has already been played back before during the current + ** rollback, then don't bother to play it back again. + */ + if( pDone && (rc = sqlite3BitvecSet(pDone, pgno))!=SQLITE_OK ){ + return rc; + } + + /* When playing back page 1, restore the nReserve setting + */ + if( pgno==1 && pPager->nReserve!=((u8*)aData)[20] ){ + pPager->nReserve = ((u8*)aData)[20]; + pagerReportSize(pPager); + } + + /* If the pager is in CACHEMOD state, then there must be a copy of this + ** page in the pager cache. In this case just update the pager cache, + ** not the database file. The page is left marked dirty in this case. + ** + ** An exception to the above rule: If the database is in no-sync mode + ** and a page is moved during an incremental vacuum then the page may + ** not be in the pager cache. Later: if a malloc() or IO error occurs + ** during a Movepage() call, then the page may not be in the cache + ** either. So the condition described in the above paragraph is not + ** assert()able. + ** + ** If in WRITER_DBMOD, WRITER_FINISHED or OPEN state, then we update the + ** pager cache if it exists and the main file. The page is then marked + ** not dirty. Since this code is only executed in PAGER_OPEN state for + ** a hot-journal rollback, it is guaranteed that the page-cache is empty + ** if the pager is in OPEN state. + ** + ** Ticket #1171: The statement journal might contain page content that is + ** different from the page content at the start of the transaction. + ** This occurs when a page is changed prior to the start of a statement + ** then changed again within the statement. When rolling back such a + ** statement we must not write to the original database unless we know + ** for certain that original page contents are synced into the main rollback + ** journal. Otherwise, a power loss might leave modified data in the + ** database file without an entry in the rollback journal that can + ** restore the database to its original form. Two conditions must be + ** met before writing to the database files. (1) the database must be + ** locked. (2) we know that the original page content is fully synced + ** in the main journal either because the page is not in cache or else + ** the page is marked as needSync==0. + ** + ** 2008-04-14: When attempting to vacuum a corrupt database file, it + ** is possible to fail a statement on a database that does not yet exist. + ** Do not attempt to write if database file has never been opened. + */ + if( pagerUseWal(pPager) ){ + pPg = 0; + }else{ + pPg = sqlite3PagerLookup(pPager, pgno); + } + assert( pPg || !MEMDB ); + assert( pPager->eState!=PAGER_OPEN || pPg==0 || pPager->tempFile ); + PAGERTRACE(("PLAYBACK %d page %d hash(%08x) %s\n", + PAGERID(pPager), pgno, pager_datahash(pPager->pageSize, (u8*)aData), + (isMainJrnl?"main-journal":"sub-journal") + )); + if( isMainJrnl ){ + isSynced = pPager->noSync || (*pOffset <= pPager->journalHdr); + }else{ + isSynced = (pPg==0 || 0==(pPg->flags & PGHDR_NEED_SYNC)); + } + if( isOpen(pPager->fd) + && (pPager->eState>=PAGER_WRITER_DBMOD || pPager->eState==PAGER_OPEN) + && isSynced + ){ + i64 ofst = (pgno-1)*(i64)pPager->pageSize; + testcase( !isSavepnt && pPg!=0 && (pPg->flags&PGHDR_NEED_SYNC)!=0 ); + assert( !pagerUseWal(pPager) ); + + /* Write the data read from the journal back into the database file. + ** This is usually safe even for an encrypted database - as the data + ** was encrypted before it was written to the journal file. The exception + ** is if the data was just read from an in-memory sub-journal. In that + ** case it must be encrypted here before it is copied into the database + ** file. */ +#ifdef SQLITE_HAS_CODEC + if( !jrnlEnc ){ + CODEC2(pPager, aData, pgno, 7, rc=SQLITE_NOMEM_BKPT, aData); + rc = sqlite3OsWrite(pPager->fd, (u8 *)aData, pPager->pageSize, ofst); + CODEC1(pPager, aData, pgno, 3, rc=SQLITE_NOMEM_BKPT); + }else +#endif + rc = sqlite3OsWrite(pPager->fd, (u8 *)aData, pPager->pageSize, ofst); + + if( pgno>pPager->dbFileSize ){ + pPager->dbFileSize = pgno; + } + if( pPager->pBackup ){ +#ifdef SQLITE_HAS_CODEC + if( jrnlEnc ){ + CODEC1(pPager, aData, pgno, 3, rc=SQLITE_NOMEM_BKPT); + sqlite3BackupUpdate(pPager->pBackup, pgno, (u8*)aData); + CODEC2(pPager, aData, pgno, 7, rc=SQLITE_NOMEM_BKPT,aData); + }else +#endif + sqlite3BackupUpdate(pPager->pBackup, pgno, (u8*)aData); + } + }else if( !isMainJrnl && pPg==0 ){ + /* If this is a rollback of a savepoint and data was not written to + ** the database and the page is not in-memory, there is a potential + ** problem. When the page is next fetched by the b-tree layer, it + ** will be read from the database file, which may or may not be + ** current. + ** + ** There are a couple of different ways this can happen. All are quite + ** obscure. When running in synchronous mode, this can only happen + ** if the page is on the free-list at the start of the transaction, then + ** populated, then moved using sqlite3PagerMovepage(). + ** + ** The solution is to add an in-memory page to the cache containing + ** the data just read from the sub-journal. Mark the page as dirty + ** and if the pager requires a journal-sync, then mark the page as + ** requiring a journal-sync before it is written. + */ + assert( isSavepnt ); + assert( (pPager->doNotSpill & SPILLFLAG_ROLLBACK)==0 ); + pPager->doNotSpill |= SPILLFLAG_ROLLBACK; + rc = sqlite3PagerGet(pPager, pgno, &pPg, 1); + assert( (pPager->doNotSpill & SPILLFLAG_ROLLBACK)!=0 ); + pPager->doNotSpill &= ~SPILLFLAG_ROLLBACK; + if( rc!=SQLITE_OK ) return rc; + sqlite3PcacheMakeDirty(pPg); + } + if( pPg ){ + /* No page should ever be explicitly rolled back that is in use, except + ** for page 1 which is held in use in order to keep the lock on the + ** database active. However such a page may be rolled back as a result + ** of an internal error resulting in an automatic call to + ** sqlite3PagerRollback(). + */ + void *pData; + pData = pPg->pData; + memcpy(pData, (u8*)aData, pPager->pageSize); + pPager->xReiniter(pPg); + /* It used to be that sqlite3PcacheMakeClean(pPg) was called here. But + ** that call was dangerous and had no detectable benefit since the cache + ** is normally cleaned by sqlite3PcacheCleanAll() after rollback and so + ** has been removed. */ + pager_set_pagehash(pPg); + + /* If this was page 1, then restore the value of Pager.dbFileVers. + ** Do this before any decoding. */ + if( pgno==1 ){ + memcpy(&pPager->dbFileVers, &((u8*)pData)[24],sizeof(pPager->dbFileVers)); + } + + /* Decode the page just read from disk */ +#if SQLITE_HAS_CODEC + if( jrnlEnc ){ CODEC1(pPager, pData, pPg->pgno, 3, rc=SQLITE_NOMEM_BKPT); } +#endif + sqlite3PcacheRelease(pPg); + } + return rc; +} + +/* +** Parameter zMaster is the name of a master journal file. A single journal +** file that referred to the master journal file has just been rolled back. +** This routine checks if it is possible to delete the master journal file, +** and does so if it is. +** +** Argument zMaster may point to Pager.pTmpSpace. So that buffer is not +** available for use within this function. +** +** When a master journal file is created, it is populated with the names +** of all of its child journals, one after another, formatted as utf-8 +** encoded text. The end of each child journal file is marked with a +** nul-terminator byte (0x00). i.e. the entire contents of a master journal +** file for a transaction involving two databases might be: +** +** "/home/bill/a.db-journal\x00/home/bill/b.db-journal\x00" +** +** A master journal file may only be deleted once all of its child +** journals have been rolled back. +** +** This function reads the contents of the master-journal file into +** memory and loops through each of the child journal names. For +** each child journal, it checks if: +** +** * if the child journal exists, and if so +** * if the child journal contains a reference to master journal +** file zMaster +** +** If a child journal can be found that matches both of the criteria +** above, this function returns without doing anything. Otherwise, if +** no such child journal can be found, file zMaster is deleted from +** the file-system using sqlite3OsDelete(). +** +** If an IO error within this function, an error code is returned. This +** function allocates memory by calling sqlite3Malloc(). If an allocation +** fails, SQLITE_NOMEM is returned. Otherwise, if no IO or malloc errors +** occur, SQLITE_OK is returned. +** +** TODO: This function allocates a single block of memory to load +** the entire contents of the master journal file. This could be +** a couple of kilobytes or so - potentially larger than the page +** size. +*/ +static int pager_delmaster(Pager *pPager, const char *zMaster){ + sqlite3_vfs *pVfs = pPager->pVfs; + int rc; /* Return code */ + sqlite3_file *pMaster; /* Malloc'd master-journal file descriptor */ + sqlite3_file *pJournal; /* Malloc'd child-journal file descriptor */ + char *zMasterJournal = 0; /* Contents of master journal file */ + i64 nMasterJournal; /* Size of master journal file */ + char *zJournal; /* Pointer to one journal within MJ file */ + char *zMasterPtr; /* Space to hold MJ filename from a journal file */ + int nMasterPtr; /* Amount of space allocated to zMasterPtr[] */ + + /* Allocate space for both the pJournal and pMaster file descriptors. + ** If successful, open the master journal file for reading. + */ + pMaster = (sqlite3_file *)sqlite3MallocZero(pVfs->szOsFile * 2); + pJournal = (sqlite3_file *)(((u8 *)pMaster) + pVfs->szOsFile); + if( !pMaster ){ + rc = SQLITE_NOMEM_BKPT; + }else{ + const int flags = (SQLITE_OPEN_READONLY|SQLITE_OPEN_MASTER_JOURNAL); + rc = sqlite3OsOpen(pVfs, zMaster, pMaster, flags, 0); + } + if( rc!=SQLITE_OK ) goto delmaster_out; + + /* Load the entire master journal file into space obtained from + ** sqlite3_malloc() and pointed to by zMasterJournal. Also obtain + ** sufficient space (in zMasterPtr) to hold the names of master + ** journal files extracted from regular rollback-journals. + */ + rc = sqlite3OsFileSize(pMaster, &nMasterJournal); + if( rc!=SQLITE_OK ) goto delmaster_out; + nMasterPtr = pVfs->mxPathname+1; + zMasterJournal = sqlite3Malloc(nMasterJournal + nMasterPtr + 1); + if( !zMasterJournal ){ + rc = SQLITE_NOMEM_BKPT; + goto delmaster_out; + } + zMasterPtr = &zMasterJournal[nMasterJournal+1]; + rc = sqlite3OsRead(pMaster, zMasterJournal, (int)nMasterJournal, 0); + if( rc!=SQLITE_OK ) goto delmaster_out; + zMasterJournal[nMasterJournal] = 0; + + zJournal = zMasterJournal; + while( (zJournal-zMasterJournal)pageSize bytes). +** If the file on disk is currently larger than nPage pages, then use the VFS +** xTruncate() method to truncate it. +** +** Or, it might be the case that the file on disk is smaller than +** nPage pages. Some operating system implementations can get confused if +** you try to truncate a file to some size that is larger than it +** currently is, so detect this case and write a single zero byte to +** the end of the new file instead. +** +** If successful, return SQLITE_OK. If an IO error occurs while modifying +** the database file, return the error code to the caller. +*/ +static int pager_truncate(Pager *pPager, Pgno nPage){ + int rc = SQLITE_OK; + assert( pPager->eState!=PAGER_ERROR ); + assert( pPager->eState!=PAGER_READER ); + + if( isOpen(pPager->fd) + && (pPager->eState>=PAGER_WRITER_DBMOD || pPager->eState==PAGER_OPEN) + ){ + i64 currentSize, newSize; + int szPage = pPager->pageSize; + assert( pPager->eLock==EXCLUSIVE_LOCK ); + /* TODO: Is it safe to use Pager.dbFileSize here? */ + rc = sqlite3OsFileSize(pPager->fd, ¤tSize); + newSize = szPage*(i64)nPage; + if( rc==SQLITE_OK && currentSize!=newSize ){ + if( currentSize>newSize ){ + rc = sqlite3OsTruncate(pPager->fd, newSize); + }else if( (currentSize+szPage)<=newSize ){ + char *pTmp = pPager->pTmpSpace; + memset(pTmp, 0, szPage); + testcase( (newSize-szPage) == currentSize ); + testcase( (newSize-szPage) > currentSize ); + rc = sqlite3OsWrite(pPager->fd, pTmp, szPage, newSize-szPage); + } + if( rc==SQLITE_OK ){ + pPager->dbFileSize = nPage; + } + } + } + return rc; +} + +/* +** Return a sanitized version of the sector-size of OS file pFile. The +** return value is guaranteed to lie between 32 and MAX_SECTOR_SIZE. +*/ +SQLITE_PRIVATE int sqlite3SectorSize(sqlite3_file *pFile){ + int iRet = sqlite3OsSectorSize(pFile); + if( iRet<32 ){ + iRet = 512; + }else if( iRet>MAX_SECTOR_SIZE ){ + assert( MAX_SECTOR_SIZE>=512 ); + iRet = MAX_SECTOR_SIZE; + } + return iRet; +} + +/* +** Set the value of the Pager.sectorSize variable for the given +** pager based on the value returned by the xSectorSize method +** of the open database file. The sector size will be used +** to determine the size and alignment of journal header and +** master journal pointers within created journal files. +** +** For temporary files the effective sector size is always 512 bytes. +** +** Otherwise, for non-temporary files, the effective sector size is +** the value returned by the xSectorSize() method rounded up to 32 if +** it is less than 32, or rounded down to MAX_SECTOR_SIZE if it +** is greater than MAX_SECTOR_SIZE. +** +** If the file has the SQLITE_IOCAP_POWERSAFE_OVERWRITE property, then set +** the effective sector size to its minimum value (512). The purpose of +** pPager->sectorSize is to define the "blast radius" of bytes that +** might change if a crash occurs while writing to a single byte in +** that range. But with POWERSAFE_OVERWRITE, the blast radius is zero +** (that is what POWERSAFE_OVERWRITE means), so we minimize the sector +** size. For backwards compatibility of the rollback journal file format, +** we cannot reduce the effective sector size below 512. +*/ +static void setSectorSize(Pager *pPager){ + assert( isOpen(pPager->fd) || pPager->tempFile ); + + if( pPager->tempFile + || (sqlite3OsDeviceCharacteristics(pPager->fd) & + SQLITE_IOCAP_POWERSAFE_OVERWRITE)!=0 + ){ + /* Sector size doesn't matter for temporary files. Also, the file + ** may not have been opened yet, in which case the OsSectorSize() + ** call will segfault. */ + pPager->sectorSize = 512; + }else{ + pPager->sectorSize = sqlite3SectorSize(pPager->fd); + } +} + +/* +** Playback the journal and thus restore the database file to +** the state it was in before we started making changes. +** +** The journal file format is as follows: +** +** (1) 8 byte prefix. A copy of aJournalMagic[]. +** (2) 4 byte big-endian integer which is the number of valid page records +** in the journal. If this value is 0xffffffff, then compute the +** number of page records from the journal size. +** (3) 4 byte big-endian integer which is the initial value for the +** sanity checksum. +** (4) 4 byte integer which is the number of pages to truncate the +** database to during a rollback. +** (5) 4 byte big-endian integer which is the sector size. The header +** is this many bytes in size. +** (6) 4 byte big-endian integer which is the page size. +** (7) zero padding out to the next sector size. +** (8) Zero or more pages instances, each as follows: +** + 4 byte page number. +** + pPager->pageSize bytes of data. +** + 4 byte checksum +** +** When we speak of the journal header, we mean the first 7 items above. +** Each entry in the journal is an instance of the 8th item. +** +** Call the value from the second bullet "nRec". nRec is the number of +** valid page entries in the journal. In most cases, you can compute the +** value of nRec from the size of the journal file. But if a power +** failure occurred while the journal was being written, it could be the +** case that the size of the journal file had already been increased but +** the extra entries had not yet made it safely to disk. In such a case, +** the value of nRec computed from the file size would be too large. For +** that reason, we always use the nRec value in the header. +** +** If the nRec value is 0xffffffff it means that nRec should be computed +** from the file size. This value is used when the user selects the +** no-sync option for the journal. A power failure could lead to corruption +** in this case. But for things like temporary table (which will be +** deleted when the power is restored) we don't care. +** +** If the file opened as the journal file is not a well-formed +** journal file then all pages up to the first corrupted page are rolled +** back (or no pages if the journal header is corrupted). The journal file +** is then deleted and SQLITE_OK returned, just as if no corruption had +** been encountered. +** +** If an I/O or malloc() error occurs, the journal-file is not deleted +** and an error code is returned. +** +** The isHot parameter indicates that we are trying to rollback a journal +** that might be a hot journal. Or, it could be that the journal is +** preserved because of JOURNALMODE_PERSIST or JOURNALMODE_TRUNCATE. +** If the journal really is hot, reset the pager cache prior rolling +** back any content. If the journal is merely persistent, no reset is +** needed. +*/ +static int pager_playback(Pager *pPager, int isHot){ + sqlite3_vfs *pVfs = pPager->pVfs; + i64 szJ; /* Size of the journal file in bytes */ + u32 nRec; /* Number of Records in the journal */ + u32 u; /* Unsigned loop counter */ + Pgno mxPg = 0; /* Size of the original file in pages */ + int rc; /* Result code of a subroutine */ + int res = 1; /* Value returned by sqlite3OsAccess() */ + char *zMaster = 0; /* Name of master journal file if any */ + int needPagerReset; /* True to reset page prior to first page rollback */ + int nPlayback = 0; /* Total number of pages restored from journal */ + u32 savedPageSize = pPager->pageSize; + + /* Figure out how many records are in the journal. Abort early if + ** the journal is empty. + */ + assert( isOpen(pPager->jfd) ); + rc = sqlite3OsFileSize(pPager->jfd, &szJ); + if( rc!=SQLITE_OK ){ + goto end_playback; + } + + /* Read the master journal name from the journal, if it is present. + ** If a master journal file name is specified, but the file is not + ** present on disk, then the journal is not hot and does not need to be + ** played back. + ** + ** TODO: Technically the following is an error because it assumes that + ** buffer Pager.pTmpSpace is (mxPathname+1) bytes or larger. i.e. that + ** (pPager->pageSize >= pPager->pVfs->mxPathname+1). Using os_unix.c, + ** mxPathname is 512, which is the same as the minimum allowable value + ** for pageSize. + */ + zMaster = pPager->pTmpSpace; + rc = readMasterJournal(pPager->jfd, zMaster, pPager->pVfs->mxPathname+1); + if( rc==SQLITE_OK && zMaster[0] ){ + rc = sqlite3OsAccess(pVfs, zMaster, SQLITE_ACCESS_EXISTS, &res); + } + zMaster = 0; + if( rc!=SQLITE_OK || !res ){ + goto end_playback; + } + pPager->journalOff = 0; + needPagerReset = isHot; + + /* This loop terminates either when a readJournalHdr() or + ** pager_playback_one_page() call returns SQLITE_DONE or an IO error + ** occurs. + */ + while( 1 ){ + /* Read the next journal header from the journal file. If there are + ** not enough bytes left in the journal file for a complete header, or + ** it is corrupted, then a process must have failed while writing it. + ** This indicates nothing more needs to be rolled back. + */ + rc = readJournalHdr(pPager, isHot, szJ, &nRec, &mxPg); + if( rc!=SQLITE_OK ){ + if( rc==SQLITE_DONE ){ + rc = SQLITE_OK; + } + goto end_playback; + } + + /* If nRec is 0xffffffff, then this journal was created by a process + ** working in no-sync mode. This means that the rest of the journal + ** file consists of pages, there are no more journal headers. Compute + ** the value of nRec based on this assumption. + */ + if( nRec==0xffffffff ){ + assert( pPager->journalOff==JOURNAL_HDR_SZ(pPager) ); + nRec = (int)((szJ - JOURNAL_HDR_SZ(pPager))/JOURNAL_PG_SZ(pPager)); + } + + /* If nRec is 0 and this rollback is of a transaction created by this + ** process and if this is the final header in the journal, then it means + ** that this part of the journal was being filled but has not yet been + ** synced to disk. Compute the number of pages based on the remaining + ** size of the file. + ** + ** The third term of the test was added to fix ticket #2565. + ** When rolling back a hot journal, nRec==0 always means that the next + ** chunk of the journal contains zero pages to be rolled back. But + ** when doing a ROLLBACK and the nRec==0 chunk is the last chunk in + ** the journal, it means that the journal might contain additional + ** pages that need to be rolled back and that the number of pages + ** should be computed based on the journal file size. + */ + if( nRec==0 && !isHot && + pPager->journalHdr+JOURNAL_HDR_SZ(pPager)==pPager->journalOff ){ + nRec = (int)((szJ - pPager->journalOff) / JOURNAL_PG_SZ(pPager)); + } + + /* If this is the first header read from the journal, truncate the + ** database file back to its original size. + */ + if( pPager->journalOff==JOURNAL_HDR_SZ(pPager) ){ + rc = pager_truncate(pPager, mxPg); + if( rc!=SQLITE_OK ){ + goto end_playback; + } + pPager->dbSize = mxPg; + } + + /* Copy original pages out of the journal and back into the + ** database file and/or page cache. + */ + for(u=0; ujournalOff,0,1,0); + if( rc==SQLITE_OK ){ + nPlayback++; + }else{ + if( rc==SQLITE_DONE ){ + pPager->journalOff = szJ; + break; + }else if( rc==SQLITE_IOERR_SHORT_READ ){ + /* If the journal has been truncated, simply stop reading and + ** processing the journal. This might happen if the journal was + ** not completely written and synced prior to a crash. In that + ** case, the database should have never been written in the + ** first place so it is OK to simply abandon the rollback. */ + rc = SQLITE_OK; + goto end_playback; + }else{ + /* If we are unable to rollback, quit and return the error + ** code. This will cause the pager to enter the error state + ** so that no further harm will be done. Perhaps the next + ** process to come along will be able to rollback the database. + */ + goto end_playback; + } + } + } + } + /*NOTREACHED*/ + assert( 0 ); + +end_playback: + if( rc==SQLITE_OK ){ + rc = sqlite3PagerSetPagesize(pPager, &savedPageSize, -1); + } + /* Following a rollback, the database file should be back in its original + ** state prior to the start of the transaction, so invoke the + ** SQLITE_FCNTL_DB_UNCHANGED file-control method to disable the + ** assertion that the transaction counter was modified. + */ +#ifdef SQLITE_DEBUG + sqlite3OsFileControlHint(pPager->fd,SQLITE_FCNTL_DB_UNCHANGED,0); +#endif + + /* If this playback is happening automatically as a result of an IO or + ** malloc error that occurred after the change-counter was updated but + ** before the transaction was committed, then the change-counter + ** modification may just have been reverted. If this happens in exclusive + ** mode, then subsequent transactions performed by the connection will not + ** update the change-counter at all. This may lead to cache inconsistency + ** problems for other processes at some point in the future. So, just + ** in case this has happened, clear the changeCountDone flag now. + */ + pPager->changeCountDone = pPager->tempFile; + + if( rc==SQLITE_OK ){ + zMaster = pPager->pTmpSpace; + rc = readMasterJournal(pPager->jfd, zMaster, pPager->pVfs->mxPathname+1); + testcase( rc!=SQLITE_OK ); + } + if( rc==SQLITE_OK + && (pPager->eState>=PAGER_WRITER_DBMOD || pPager->eState==PAGER_OPEN) + ){ + rc = sqlite3PagerSync(pPager, 0); + } + if( rc==SQLITE_OK ){ + rc = pager_end_transaction(pPager, zMaster[0]!='\0', 0); + testcase( rc!=SQLITE_OK ); + } + if( rc==SQLITE_OK && zMaster[0] && res ){ + /* If there was a master journal and this routine will return success, + ** see if it is possible to delete the master journal. + */ + rc = pager_delmaster(pPager, zMaster); + testcase( rc!=SQLITE_OK ); + } + if( isHot && nPlayback ){ + sqlite3_log(SQLITE_NOTICE_RECOVER_ROLLBACK, "recovered %d pages from %s", + nPlayback, pPager->zJournal); + } + + /* The Pager.sectorSize variable may have been updated while rolling + ** back a journal created by a process with a different sector size + ** value. Reset it to the correct value for this process. + */ + setSectorSize(pPager); + return rc; +} + + +/* +** Read the content for page pPg out of the database file (or out of +** the WAL if that is where the most recent copy if found) into +** pPg->pData. A shared lock or greater must be held on the database +** file before this function is called. +** +** If page 1 is read, then the value of Pager.dbFileVers[] is set to +** the value read from the database file. +** +** If an IO error occurs, then the IO error is returned to the caller. +** Otherwise, SQLITE_OK is returned. +*/ +static int readDbPage(PgHdr *pPg){ + Pager *pPager = pPg->pPager; /* Pager object associated with page pPg */ + int rc = SQLITE_OK; /* Return code */ + +#ifndef SQLITE_OMIT_WAL + u32 iFrame = 0; /* Frame of WAL containing pgno */ + + assert( pPager->eState>=PAGER_READER && !MEMDB ); + assert( isOpen(pPager->fd) ); + + if( pagerUseWal(pPager) ){ + rc = sqlite3WalFindFrame(pPager->pWal, pPg->pgno, &iFrame); + if( rc ) return rc; + } + if( iFrame ){ + rc = sqlite3WalReadFrame(pPager->pWal, iFrame,pPager->pageSize,pPg->pData); + }else +#endif + { + i64 iOffset = (pPg->pgno-1)*(i64)pPager->pageSize; + rc = sqlite3OsRead(pPager->fd, pPg->pData, pPager->pageSize, iOffset); + if( rc==SQLITE_IOERR_SHORT_READ ){ + rc = SQLITE_OK; + } + } + + if( pPg->pgno==1 ){ + if( rc ){ + /* If the read is unsuccessful, set the dbFileVers[] to something + ** that will never be a valid file version. dbFileVers[] is a copy + ** of bytes 24..39 of the database. Bytes 28..31 should always be + ** zero or the size of the database in page. Bytes 32..35 and 35..39 + ** should be page numbers which are never 0xffffffff. So filling + ** pPager->dbFileVers[] with all 0xff bytes should suffice. + ** + ** For an encrypted database, the situation is more complex: bytes + ** 24..39 of the database are white noise. But the probability of + ** white noise equaling 16 bytes of 0xff is vanishingly small so + ** we should still be ok. + */ + memset(pPager->dbFileVers, 0xff, sizeof(pPager->dbFileVers)); + }else{ + u8 *dbFileVers = &((u8*)pPg->pData)[24]; + memcpy(&pPager->dbFileVers, dbFileVers, sizeof(pPager->dbFileVers)); + } + } + CODEC1(pPager, pPg->pData, pPg->pgno, 3, rc = SQLITE_NOMEM_BKPT); + + PAGER_INCR(sqlite3_pager_readdb_count); + PAGER_INCR(pPager->nRead); + IOTRACE(("PGIN %p %d\n", pPager, pPg->pgno)); + PAGERTRACE(("FETCH %d page %d hash(%08x)\n", + PAGERID(pPager), pPg->pgno, pager_pagehash(pPg))); + + return rc; +} + +/* +** Update the value of the change-counter at offsets 24 and 92 in +** the header and the sqlite version number at offset 96. +** +** This is an unconditional update. See also the pager_incr_changecounter() +** routine which only updates the change-counter if the update is actually +** needed, as determined by the pPager->changeCountDone state variable. +*/ +static void pager_write_changecounter(PgHdr *pPg){ + u32 change_counter; + + /* Increment the value just read and write it back to byte 24. */ + change_counter = sqlite3Get4byte((u8*)pPg->pPager->dbFileVers)+1; + put32bits(((char*)pPg->pData)+24, change_counter); + + /* Also store the SQLite version number in bytes 96..99 and in + ** bytes 92..95 store the change counter for which the version number + ** is valid. */ + put32bits(((char*)pPg->pData)+92, change_counter); + put32bits(((char*)pPg->pData)+96, SQLITE_VERSION_NUMBER); +} + +#ifndef SQLITE_OMIT_WAL +/* +** This function is invoked once for each page that has already been +** written into the log file when a WAL transaction is rolled back. +** Parameter iPg is the page number of said page. The pCtx argument +** is actually a pointer to the Pager structure. +** +** If page iPg is present in the cache, and has no outstanding references, +** it is discarded. Otherwise, if there are one or more outstanding +** references, the page content is reloaded from the database. If the +** attempt to reload content from the database is required and fails, +** return an SQLite error code. Otherwise, SQLITE_OK. +*/ +static int pagerUndoCallback(void *pCtx, Pgno iPg){ + int rc = SQLITE_OK; + Pager *pPager = (Pager *)pCtx; + PgHdr *pPg; + + assert( pagerUseWal(pPager) ); + pPg = sqlite3PagerLookup(pPager, iPg); + if( pPg ){ + if( sqlite3PcachePageRefcount(pPg)==1 ){ + sqlite3PcacheDrop(pPg); + }else{ + rc = readDbPage(pPg); + if( rc==SQLITE_OK ){ + pPager->xReiniter(pPg); + } + sqlite3PagerUnrefNotNull(pPg); + } + } + + /* Normally, if a transaction is rolled back, any backup processes are + ** updated as data is copied out of the rollback journal and into the + ** database. This is not generally possible with a WAL database, as + ** rollback involves simply truncating the log file. Therefore, if one + ** or more frames have already been written to the log (and therefore + ** also copied into the backup databases) as part of this transaction, + ** the backups must be restarted. + */ + sqlite3BackupRestart(pPager->pBackup); + + return rc; +} + +/* +** This function is called to rollback a transaction on a WAL database. +*/ +static int pagerRollbackWal(Pager *pPager){ + int rc; /* Return Code */ + PgHdr *pList; /* List of dirty pages to revert */ + + /* For all pages in the cache that are currently dirty or have already + ** been written (but not committed) to the log file, do one of the + ** following: + ** + ** + Discard the cached page (if refcount==0), or + ** + Reload page content from the database (if refcount>0). + */ + pPager->dbSize = pPager->dbOrigSize; + rc = sqlite3WalUndo(pPager->pWal, pagerUndoCallback, (void *)pPager); + pList = sqlite3PcacheDirtyList(pPager->pPCache); + while( pList && rc==SQLITE_OK ){ + PgHdr *pNext = pList->pDirty; + rc = pagerUndoCallback((void *)pPager, pList->pgno); + pList = pNext; + } + + return rc; +} + +/* +** This function is a wrapper around sqlite3WalFrames(). As well as logging +** the contents of the list of pages headed by pList (connected by pDirty), +** this function notifies any active backup processes that the pages have +** changed. +** +** The list of pages passed into this routine is always sorted by page number. +** Hence, if page 1 appears anywhere on the list, it will be the first page. +*/ +static int pagerWalFrames( + Pager *pPager, /* Pager object */ + PgHdr *pList, /* List of frames to log */ + Pgno nTruncate, /* Database size after this commit */ + int isCommit /* True if this is a commit */ +){ + int rc; /* Return code */ + int nList; /* Number of pages in pList */ + PgHdr *p; /* For looping over pages */ + + assert( pPager->pWal ); + assert( pList ); +#ifdef SQLITE_DEBUG + /* Verify that the page list is in accending order */ + for(p=pList; p && p->pDirty; p=p->pDirty){ + assert( p->pgno < p->pDirty->pgno ); + } +#endif + + assert( pList->pDirty==0 || isCommit ); + if( isCommit ){ + /* If a WAL transaction is being committed, there is no point in writing + ** any pages with page numbers greater than nTruncate into the WAL file. + ** They will never be read by any client. So remove them from the pDirty + ** list here. */ + PgHdr **ppNext = &pList; + nList = 0; + for(p=pList; (*ppNext = p)!=0; p=p->pDirty){ + if( p->pgno<=nTruncate ){ + ppNext = &p->pDirty; + nList++; + } + } + assert( pList ); + }else{ + nList = 1; + } + pPager->aStat[PAGER_STAT_WRITE] += nList; + + if( pList->pgno==1 ) pager_write_changecounter(pList); + rc = sqlite3WalFrames(pPager->pWal, + pPager->pageSize, pList, nTruncate, isCommit, pPager->walSyncFlags + ); + if( rc==SQLITE_OK && pPager->pBackup ){ + for(p=pList; p; p=p->pDirty){ + sqlite3BackupUpdate(pPager->pBackup, p->pgno, (u8 *)p->pData); + } + } + +#ifdef SQLITE_CHECK_PAGES + pList = sqlite3PcacheDirtyList(pPager->pPCache); + for(p=pList; p; p=p->pDirty){ + pager_set_pagehash(p); + } +#endif + + return rc; +} + +/* +** Begin a read transaction on the WAL. +** +** This routine used to be called "pagerOpenSnapshot()" because it essentially +** makes a snapshot of the database at the current point in time and preserves +** that snapshot for use by the reader in spite of concurrently changes by +** other writers or checkpointers. +*/ +static int pagerBeginReadTransaction(Pager *pPager){ + int rc; /* Return code */ + int changed = 0; /* True if cache must be reset */ + + assert( pagerUseWal(pPager) ); + assert( pPager->eState==PAGER_OPEN || pPager->eState==PAGER_READER ); + + /* sqlite3WalEndReadTransaction() was not called for the previous + ** transaction in locking_mode=EXCLUSIVE. So call it now. If we + ** are in locking_mode=NORMAL and EndRead() was previously called, + ** the duplicate call is harmless. + */ + sqlite3WalEndReadTransaction(pPager->pWal); + + rc = sqlite3WalBeginReadTransaction(pPager->pWal, &changed); + if( rc!=SQLITE_OK || changed ){ + pager_reset(pPager); + if( USEFETCH(pPager) ) sqlite3OsUnfetch(pPager->fd, 0, 0); + } + + return rc; +} +#endif + +/* +** This function is called as part of the transition from PAGER_OPEN +** to PAGER_READER state to determine the size of the database file +** in pages (assuming the page size currently stored in Pager.pageSize). +** +** If no error occurs, SQLITE_OK is returned and the size of the database +** in pages is stored in *pnPage. Otherwise, an error code (perhaps +** SQLITE_IOERR_FSTAT) is returned and *pnPage is left unmodified. +*/ +static int pagerPagecount(Pager *pPager, Pgno *pnPage){ + Pgno nPage; /* Value to return via *pnPage */ + + /* Query the WAL sub-system for the database size. The WalDbsize() + ** function returns zero if the WAL is not open (i.e. Pager.pWal==0), or + ** if the database size is not available. The database size is not + ** available from the WAL sub-system if the log file is empty or + ** contains no valid committed transactions. + */ + assert( pPager->eState==PAGER_OPEN ); + assert( pPager->eLock>=SHARED_LOCK ); + assert( isOpen(pPager->fd) ); + assert( pPager->tempFile==0 ); + nPage = sqlite3WalDbsize(pPager->pWal); + + /* If the number of pages in the database is not available from the + ** WAL sub-system, determine the page count based on the size of + ** the database file. If the size of the database file is not an + ** integer multiple of the page-size, round up the result. + */ + if( nPage==0 && ALWAYS(isOpen(pPager->fd)) ){ + i64 n = 0; /* Size of db file in bytes */ + int rc = sqlite3OsFileSize(pPager->fd, &n); + if( rc!=SQLITE_OK ){ + return rc; + } + nPage = (Pgno)((n+pPager->pageSize-1) / pPager->pageSize); + } + + /* If the current number of pages in the file is greater than the + ** configured maximum pager number, increase the allowed limit so + ** that the file can be read. + */ + if( nPage>pPager->mxPgno ){ + pPager->mxPgno = (Pgno)nPage; + } + + *pnPage = nPage; + return SQLITE_OK; +} + +#ifndef SQLITE_OMIT_WAL +/* +** Check if the *-wal file that corresponds to the database opened by pPager +** exists if the database is not empy, or verify that the *-wal file does +** not exist (by deleting it) if the database file is empty. +** +** If the database is not empty and the *-wal file exists, open the pager +** in WAL mode. If the database is empty or if no *-wal file exists and +** if no error occurs, make sure Pager.journalMode is not set to +** PAGER_JOURNALMODE_WAL. +** +** Return SQLITE_OK or an error code. +** +** The caller must hold a SHARED lock on the database file to call this +** function. Because an EXCLUSIVE lock on the db file is required to delete +** a WAL on a none-empty database, this ensures there is no race condition +** between the xAccess() below and an xDelete() being executed by some +** other connection. +*/ +static int pagerOpenWalIfPresent(Pager *pPager){ + int rc = SQLITE_OK; + assert( pPager->eState==PAGER_OPEN ); + assert( pPager->eLock>=SHARED_LOCK ); + + if( !pPager->tempFile ){ + int isWal; /* True if WAL file exists */ + rc = sqlite3OsAccess( + pPager->pVfs, pPager->zWal, SQLITE_ACCESS_EXISTS, &isWal + ); + if( rc==SQLITE_OK ){ + if( isWal ){ + Pgno nPage; /* Size of the database file */ + + rc = pagerPagecount(pPager, &nPage); + if( rc ) return rc; + if( nPage==0 ){ + rc = sqlite3OsDelete(pPager->pVfs, pPager->zWal, 0); + }else{ + testcase( sqlite3PcachePagecount(pPager->pPCache)==0 ); + rc = sqlite3PagerOpenWal(pPager, 0); + } + }else if( pPager->journalMode==PAGER_JOURNALMODE_WAL ){ + pPager->journalMode = PAGER_JOURNALMODE_DELETE; + } + } + } + return rc; +} +#endif + +/* +** Playback savepoint pSavepoint. Or, if pSavepoint==NULL, then playback +** the entire master journal file. The case pSavepoint==NULL occurs when +** a ROLLBACK TO command is invoked on a SAVEPOINT that is a transaction +** savepoint. +** +** When pSavepoint is not NULL (meaning a non-transaction savepoint is +** being rolled back), then the rollback consists of up to three stages, +** performed in the order specified: +** +** * Pages are played back from the main journal starting at byte +** offset PagerSavepoint.iOffset and continuing to +** PagerSavepoint.iHdrOffset, or to the end of the main journal +** file if PagerSavepoint.iHdrOffset is zero. +** +** * If PagerSavepoint.iHdrOffset is not zero, then pages are played +** back starting from the journal header immediately following +** PagerSavepoint.iHdrOffset to the end of the main journal file. +** +** * Pages are then played back from the sub-journal file, starting +** with the PagerSavepoint.iSubRec and continuing to the end of +** the journal file. +** +** Throughout the rollback process, each time a page is rolled back, the +** corresponding bit is set in a bitvec structure (variable pDone in the +** implementation below). This is used to ensure that a page is only +** rolled back the first time it is encountered in either journal. +** +** If pSavepoint is NULL, then pages are only played back from the main +** journal file. There is no need for a bitvec in this case. +** +** In either case, before playback commences the Pager.dbSize variable +** is reset to the value that it held at the start of the savepoint +** (or transaction). No page with a page-number greater than this value +** is played back. If one is encountered it is simply skipped. +*/ +static int pagerPlaybackSavepoint(Pager *pPager, PagerSavepoint *pSavepoint){ + i64 szJ; /* Effective size of the main journal */ + i64 iHdrOff; /* End of first segment of main-journal records */ + int rc = SQLITE_OK; /* Return code */ + Bitvec *pDone = 0; /* Bitvec to ensure pages played back only once */ + + assert( pPager->eState!=PAGER_ERROR ); + assert( pPager->eState>=PAGER_WRITER_LOCKED ); + + /* Allocate a bitvec to use to store the set of pages rolled back */ + if( pSavepoint ){ + pDone = sqlite3BitvecCreate(pSavepoint->nOrig); + if( !pDone ){ + return SQLITE_NOMEM_BKPT; + } + } + + /* Set the database size back to the value it was before the savepoint + ** being reverted was opened. + */ + pPager->dbSize = pSavepoint ? pSavepoint->nOrig : pPager->dbOrigSize; + pPager->changeCountDone = pPager->tempFile; + + if( !pSavepoint && pagerUseWal(pPager) ){ + return pagerRollbackWal(pPager); + } + + /* Use pPager->journalOff as the effective size of the main rollback + ** journal. The actual file might be larger than this in + ** PAGER_JOURNALMODE_TRUNCATE or PAGER_JOURNALMODE_PERSIST. But anything + ** past pPager->journalOff is off-limits to us. + */ + szJ = pPager->journalOff; + assert( pagerUseWal(pPager)==0 || szJ==0 ); + + /* Begin by rolling back records from the main journal starting at + ** PagerSavepoint.iOffset and continuing to the next journal header. + ** There might be records in the main journal that have a page number + ** greater than the current database size (pPager->dbSize) but those + ** will be skipped automatically. Pages are added to pDone as they + ** are played back. + */ + if( pSavepoint && !pagerUseWal(pPager) ){ + iHdrOff = pSavepoint->iHdrOffset ? pSavepoint->iHdrOffset : szJ; + pPager->journalOff = pSavepoint->iOffset; + while( rc==SQLITE_OK && pPager->journalOffjournalOff, pDone, 1, 1); + } + assert( rc!=SQLITE_DONE ); + }else{ + pPager->journalOff = 0; + } + + /* Continue rolling back records out of the main journal starting at + ** the first journal header seen and continuing until the effective end + ** of the main journal file. Continue to skip out-of-range pages and + ** continue adding pages rolled back to pDone. + */ + while( rc==SQLITE_OK && pPager->journalOffjournalHdr+JOURNAL_HDR_SZ(pPager)==pPager->journalOff" + ** test is related to ticket #2565. See the discussion in the + ** pager_playback() function for additional information. + */ + if( nJRec==0 + && pPager->journalHdr+JOURNAL_HDR_SZ(pPager)==pPager->journalOff + ){ + nJRec = (u32)((szJ - pPager->journalOff)/JOURNAL_PG_SZ(pPager)); + } + for(ii=0; rc==SQLITE_OK && iijournalOffjournalOff, pDone, 1, 1); + } + assert( rc!=SQLITE_DONE ); + } + assert( rc!=SQLITE_OK || pPager->journalOff>=szJ ); + + /* Finally, rollback pages from the sub-journal. Page that were + ** previously rolled back out of the main journal (and are hence in pDone) + ** will be skipped. Out-of-range pages are also skipped. + */ + if( pSavepoint ){ + u32 ii; /* Loop counter */ + i64 offset = (i64)pSavepoint->iSubRec*(4+pPager->pageSize); + + if( pagerUseWal(pPager) ){ + rc = sqlite3WalSavepointUndo(pPager->pWal, pSavepoint->aWalData); + } + for(ii=pSavepoint->iSubRec; rc==SQLITE_OK && iinSubRec; ii++){ + assert( offset==(i64)ii*(4+pPager->pageSize) ); + rc = pager_playback_one_page(pPager, &offset, pDone, 0, 1); + } + assert( rc!=SQLITE_DONE ); + } + + sqlite3BitvecDestroy(pDone); + if( rc==SQLITE_OK ){ + pPager->journalOff = szJ; + } + + return rc; +} + +/* +** Change the maximum number of in-memory pages that are allowed +** before attempting to recycle clean and unused pages. +*/ +SQLITE_PRIVATE void sqlite3PagerSetCachesize(Pager *pPager, int mxPage){ + sqlite3PcacheSetCachesize(pPager->pPCache, mxPage); +} + +/* +** Change the maximum number of in-memory pages that are allowed +** before attempting to spill pages to journal. +*/ +SQLITE_PRIVATE int sqlite3PagerSetSpillsize(Pager *pPager, int mxPage){ + return sqlite3PcacheSetSpillsize(pPager->pPCache, mxPage); +} + +/* +** Invoke SQLITE_FCNTL_MMAP_SIZE based on the current value of szMmap. +*/ +static void pagerFixMaplimit(Pager *pPager){ +#if SQLITE_MAX_MMAP_SIZE>0 + sqlite3_file *fd = pPager->fd; + if( isOpen(fd) && fd->pMethods->iVersion>=3 ){ + sqlite3_int64 sz; + sz = pPager->szMmap; + pPager->bUseFetch = (sz>0); + setGetterMethod(pPager); + sqlite3OsFileControlHint(pPager->fd, SQLITE_FCNTL_MMAP_SIZE, &sz); + } +#endif +} + +/* +** Change the maximum size of any memory mapping made of the database file. +*/ +SQLITE_PRIVATE void sqlite3PagerSetMmapLimit(Pager *pPager, sqlite3_int64 szMmap){ + pPager->szMmap = szMmap; + pagerFixMaplimit(pPager); +} + +/* +** Free as much memory as possible from the pager. +*/ +SQLITE_PRIVATE void sqlite3PagerShrink(Pager *pPager){ + sqlite3PcacheShrink(pPager->pPCache); +} + +/* +** Adjust settings of the pager to those specified in the pgFlags parameter. +** +** The "level" in pgFlags & PAGER_SYNCHRONOUS_MASK sets the robustness +** of the database to damage due to OS crashes or power failures by +** changing the number of syncs()s when writing the journals. +** There are four levels: +** +** OFF sqlite3OsSync() is never called. This is the default +** for temporary and transient files. +** +** NORMAL The journal is synced once before writes begin on the +** database. This is normally adequate protection, but +** it is theoretically possible, though very unlikely, +** that an inopertune power failure could leave the journal +** in a state which would cause damage to the database +** when it is rolled back. +** +** FULL The journal is synced twice before writes begin on the +** database (with some additional information - the nRec field +** of the journal header - being written in between the two +** syncs). If we assume that writing a +** single disk sector is atomic, then this mode provides +** assurance that the journal will not be corrupted to the +** point of causing damage to the database during rollback. +** +** EXTRA This is like FULL except that is also syncs the directory +** that contains the rollback journal after the rollback +** journal is unlinked. +** +** The above is for a rollback-journal mode. For WAL mode, OFF continues +** to mean that no syncs ever occur. NORMAL means that the WAL is synced +** prior to the start of checkpoint and that the database file is synced +** at the conclusion of the checkpoint if the entire content of the WAL +** was written back into the database. But no sync operations occur for +** an ordinary commit in NORMAL mode with WAL. FULL means that the WAL +** file is synced following each commit operation, in addition to the +** syncs associated with NORMAL. There is no difference between FULL +** and EXTRA for WAL mode. +** +** Do not confuse synchronous=FULL with SQLITE_SYNC_FULL. The +** SQLITE_SYNC_FULL macro means to use the MacOSX-style full-fsync +** using fcntl(F_FULLFSYNC). SQLITE_SYNC_NORMAL means to do an +** ordinary fsync() call. There is no difference between SQLITE_SYNC_FULL +** and SQLITE_SYNC_NORMAL on platforms other than MacOSX. But the +** synchronous=FULL versus synchronous=NORMAL setting determines when +** the xSync primitive is called and is relevant to all platforms. +** +** Numeric values associated with these states are OFF==1, NORMAL=2, +** and FULL=3. +*/ +#ifndef SQLITE_OMIT_PAGER_PRAGMAS +SQLITE_PRIVATE void sqlite3PagerSetFlags( + Pager *pPager, /* The pager to set safety level for */ + unsigned pgFlags /* Various flags */ +){ + unsigned level = pgFlags & PAGER_SYNCHRONOUS_MASK; + if( pPager->tempFile ){ + pPager->noSync = 1; + pPager->fullSync = 0; + pPager->extraSync = 0; + }else{ + pPager->noSync = level==PAGER_SYNCHRONOUS_OFF ?1:0; + pPager->fullSync = level>=PAGER_SYNCHRONOUS_FULL ?1:0; + pPager->extraSync = level==PAGER_SYNCHRONOUS_EXTRA ?1:0; + } + if( pPager->noSync ){ + pPager->syncFlags = 0; + }else if( pgFlags & PAGER_FULLFSYNC ){ + pPager->syncFlags = SQLITE_SYNC_FULL; + }else{ + pPager->syncFlags = SQLITE_SYNC_NORMAL; + } + pPager->walSyncFlags = (pPager->syncFlags<<2); + if( pPager->fullSync ){ + pPager->walSyncFlags |= pPager->syncFlags; + } + if( (pgFlags & PAGER_CKPT_FULLFSYNC) && !pPager->noSync ){ + pPager->walSyncFlags |= (SQLITE_SYNC_FULL<<2); + } + if( pgFlags & PAGER_CACHESPILL ){ + pPager->doNotSpill &= ~SPILLFLAG_OFF; + }else{ + pPager->doNotSpill |= SPILLFLAG_OFF; + } +} +#endif + +/* +** The following global variable is incremented whenever the library +** attempts to open a temporary file. This information is used for +** testing and analysis only. +*/ +#ifdef SQLITE_TEST +SQLITE_API int sqlite3_opentemp_count = 0; +#endif + +/* +** Open a temporary file. +** +** Write the file descriptor into *pFile. Return SQLITE_OK on success +** or some other error code if we fail. The OS will automatically +** delete the temporary file when it is closed. +** +** The flags passed to the VFS layer xOpen() call are those specified +** by parameter vfsFlags ORed with the following: +** +** SQLITE_OPEN_READWRITE +** SQLITE_OPEN_CREATE +** SQLITE_OPEN_EXCLUSIVE +** SQLITE_OPEN_DELETEONCLOSE +*/ +static int pagerOpentemp( + Pager *pPager, /* The pager object */ + sqlite3_file *pFile, /* Write the file descriptor here */ + int vfsFlags /* Flags passed through to the VFS */ +){ + int rc; /* Return code */ + +#ifdef SQLITE_TEST + sqlite3_opentemp_count++; /* Used for testing and analysis only */ +#endif + + vfsFlags |= SQLITE_OPEN_READWRITE | SQLITE_OPEN_CREATE | + SQLITE_OPEN_EXCLUSIVE | SQLITE_OPEN_DELETEONCLOSE; + rc = sqlite3OsOpen(pPager->pVfs, 0, pFile, vfsFlags, 0); + assert( rc!=SQLITE_OK || isOpen(pFile) ); + return rc; +} + +/* +** Set the busy handler function. +** +** The pager invokes the busy-handler if sqlite3OsLock() returns +** SQLITE_BUSY when trying to upgrade from no-lock to a SHARED lock, +** or when trying to upgrade from a RESERVED lock to an EXCLUSIVE +** lock. It does *not* invoke the busy handler when upgrading from +** SHARED to RESERVED, or when upgrading from SHARED to EXCLUSIVE +** (which occurs during hot-journal rollback). Summary: +** +** Transition | Invokes xBusyHandler +** -------------------------------------------------------- +** NO_LOCK -> SHARED_LOCK | Yes +** SHARED_LOCK -> RESERVED_LOCK | No +** SHARED_LOCK -> EXCLUSIVE_LOCK | No +** RESERVED_LOCK -> EXCLUSIVE_LOCK | Yes +** +** If the busy-handler callback returns non-zero, the lock is +** retried. If it returns zero, then the SQLITE_BUSY error is +** returned to the caller of the pager API function. +*/ +SQLITE_PRIVATE void sqlite3PagerSetBusyHandler( + Pager *pPager, /* Pager object */ + int (*xBusyHandler)(void *), /* Pointer to busy-handler function */ + void *pBusyHandlerArg /* Argument to pass to xBusyHandler */ +){ + void **ap; + pPager->xBusyHandler = xBusyHandler; + pPager->pBusyHandlerArg = pBusyHandlerArg; + ap = (void **)&pPager->xBusyHandler; + assert( ((int(*)(void *))(ap[0]))==xBusyHandler ); + assert( ap[1]==pBusyHandlerArg ); + sqlite3OsFileControlHint(pPager->fd, SQLITE_FCNTL_BUSYHANDLER, (void *)ap); +} + +/* +** Change the page size used by the Pager object. The new page size +** is passed in *pPageSize. +** +** If the pager is in the error state when this function is called, it +** is a no-op. The value returned is the error state error code (i.e. +** one of SQLITE_IOERR, an SQLITE_IOERR_xxx sub-code or SQLITE_FULL). +** +** Otherwise, if all of the following are true: +** +** * the new page size (value of *pPageSize) is valid (a power +** of two between 512 and SQLITE_MAX_PAGE_SIZE, inclusive), and +** +** * there are no outstanding page references, and +** +** * the database is either not an in-memory database or it is +** an in-memory database that currently consists of zero pages. +** +** then the pager object page size is set to *pPageSize. +** +** If the page size is changed, then this function uses sqlite3PagerMalloc() +** to obtain a new Pager.pTmpSpace buffer. If this allocation attempt +** fails, SQLITE_NOMEM is returned and the page size remains unchanged. +** In all other cases, SQLITE_OK is returned. +** +** If the page size is not changed, either because one of the enumerated +** conditions above is not true, the pager was in error state when this +** function was called, or because the memory allocation attempt failed, +** then *pPageSize is set to the old, retained page size before returning. +*/ +SQLITE_PRIVATE int sqlite3PagerSetPagesize(Pager *pPager, u32 *pPageSize, int nReserve){ + int rc = SQLITE_OK; + + /* It is not possible to do a full assert_pager_state() here, as this + ** function may be called from within PagerOpen(), before the state + ** of the Pager object is internally consistent. + ** + ** At one point this function returned an error if the pager was in + ** PAGER_ERROR state. But since PAGER_ERROR state guarantees that + ** there is at least one outstanding page reference, this function + ** is a no-op for that case anyhow. + */ + + u32 pageSize = *pPageSize; + assert( pageSize==0 || (pageSize>=512 && pageSize<=SQLITE_MAX_PAGE_SIZE) ); + if( (pPager->memDb==0 || pPager->dbSize==0) + && sqlite3PcacheRefCount(pPager->pPCache)==0 + && pageSize && pageSize!=(u32)pPager->pageSize + ){ + char *pNew = NULL; /* New temp space */ + i64 nByte = 0; + + if( pPager->eState>PAGER_OPEN && isOpen(pPager->fd) ){ + rc = sqlite3OsFileSize(pPager->fd, &nByte); + } + if( rc==SQLITE_OK ){ + /* 8 bytes of zeroed overrun space is sufficient so that the b-tree + * cell header parser will never run off the end of the allocation */ + pNew = (char *)sqlite3PageMalloc(pageSize+8); + if( !pNew ){ + rc = SQLITE_NOMEM_BKPT; + }else{ + memset(pNew+pageSize, 0, 8); + } + } + + if( rc==SQLITE_OK ){ + pager_reset(pPager); + rc = sqlite3PcacheSetPageSize(pPager->pPCache, pageSize); + } + if( rc==SQLITE_OK ){ + sqlite3PageFree(pPager->pTmpSpace); + pPager->pTmpSpace = pNew; + pPager->dbSize = (Pgno)((nByte+pageSize-1)/pageSize); + pPager->pageSize = pageSize; + }else{ + sqlite3PageFree(pNew); + } + } + + *pPageSize = pPager->pageSize; + if( rc==SQLITE_OK ){ + if( nReserve<0 ) nReserve = pPager->nReserve; + assert( nReserve>=0 && nReserve<1000 ); + pPager->nReserve = (i16)nReserve; + pagerReportSize(pPager); + pagerFixMaplimit(pPager); + } + return rc; +} + +/* +** Return a pointer to the "temporary page" buffer held internally +** by the pager. This is a buffer that is big enough to hold the +** entire content of a database page. This buffer is used internally +** during rollback and will be overwritten whenever a rollback +** occurs. But other modules are free to use it too, as long as +** no rollbacks are happening. +*/ +SQLITE_PRIVATE void *sqlite3PagerTempSpace(Pager *pPager){ + return pPager->pTmpSpace; +} + +/* +** Attempt to set the maximum database page count if mxPage is positive. +** Make no changes if mxPage is zero or negative. And never reduce the +** maximum page count below the current size of the database. +** +** Regardless of mxPage, return the current maximum page count. +*/ +SQLITE_PRIVATE int sqlite3PagerMaxPageCount(Pager *pPager, int mxPage){ + if( mxPage>0 ){ + pPager->mxPgno = mxPage; + } + assert( pPager->eState!=PAGER_OPEN ); /* Called only by OP_MaxPgcnt */ + /* assert( pPager->mxPgno>=pPager->dbSize ); */ + /* OP_MaxPgcnt ensures that the parameter passed to this function is not + ** less than the total number of valid pages in the database. But this + ** may be less than Pager.dbSize, and so the assert() above is not valid */ + return pPager->mxPgno; +} + +/* +** The following set of routines are used to disable the simulated +** I/O error mechanism. These routines are used to avoid simulated +** errors in places where we do not care about errors. +** +** Unless -DSQLITE_TEST=1 is used, these routines are all no-ops +** and generate no code. +*/ +#ifdef SQLITE_TEST +SQLITE_API extern int sqlite3_io_error_pending; +SQLITE_API extern int sqlite3_io_error_hit; +static int saved_cnt; +void disable_simulated_io_errors(void){ + saved_cnt = sqlite3_io_error_pending; + sqlite3_io_error_pending = -1; +} +void enable_simulated_io_errors(void){ + sqlite3_io_error_pending = saved_cnt; +} +#else +# define disable_simulated_io_errors() +# define enable_simulated_io_errors() +#endif + +/* +** Read the first N bytes from the beginning of the file into memory +** that pDest points to. +** +** If the pager was opened on a transient file (zFilename==""), or +** opened on a file less than N bytes in size, the output buffer is +** zeroed and SQLITE_OK returned. The rationale for this is that this +** function is used to read database headers, and a new transient or +** zero sized database has a header than consists entirely of zeroes. +** +** If any IO error apart from SQLITE_IOERR_SHORT_READ is encountered, +** the error code is returned to the caller and the contents of the +** output buffer undefined. +*/ +SQLITE_PRIVATE int sqlite3PagerReadFileheader(Pager *pPager, int N, unsigned char *pDest){ + int rc = SQLITE_OK; + memset(pDest, 0, N); + assert( isOpen(pPager->fd) || pPager->tempFile ); + + /* This routine is only called by btree immediately after creating + ** the Pager object. There has not been an opportunity to transition + ** to WAL mode yet. + */ + assert( !pagerUseWal(pPager) ); + + if( isOpen(pPager->fd) ){ + IOTRACE(("DBHDR %p 0 %d\n", pPager, N)) + rc = sqlite3OsRead(pPager->fd, pDest, N, 0); + if( rc==SQLITE_IOERR_SHORT_READ ){ + rc = SQLITE_OK; + } + } + return rc; +} + +/* +** This function may only be called when a read-transaction is open on +** the pager. It returns the total number of pages in the database. +** +** However, if the file is between 1 and bytes in size, then +** this is considered a 1 page file. +*/ +SQLITE_PRIVATE void sqlite3PagerPagecount(Pager *pPager, int *pnPage){ + assert( pPager->eState>=PAGER_READER ); + assert( pPager->eState!=PAGER_WRITER_FINISHED ); + *pnPage = (int)pPager->dbSize; +} + + +/* +** Try to obtain a lock of type locktype on the database file. If +** a similar or greater lock is already held, this function is a no-op +** (returning SQLITE_OK immediately). +** +** Otherwise, attempt to obtain the lock using sqlite3OsLock(). Invoke +** the busy callback if the lock is currently not available. Repeat +** until the busy callback returns false or until the attempt to +** obtain the lock succeeds. +** +** Return SQLITE_OK on success and an error code if we cannot obtain +** the lock. If the lock is obtained successfully, set the Pager.state +** variable to locktype before returning. +*/ +static int pager_wait_on_lock(Pager *pPager, int locktype){ + int rc; /* Return code */ + + /* Check that this is either a no-op (because the requested lock is + ** already held), or one of the transitions that the busy-handler + ** may be invoked during, according to the comment above + ** sqlite3PagerSetBusyhandler(). + */ + assert( (pPager->eLock>=locktype) + || (pPager->eLock==NO_LOCK && locktype==SHARED_LOCK) + || (pPager->eLock==RESERVED_LOCK && locktype==EXCLUSIVE_LOCK) + ); + + do { + rc = pagerLockDb(pPager, locktype); + }while( rc==SQLITE_BUSY && pPager->xBusyHandler(pPager->pBusyHandlerArg) ); + return rc; +} + +/* +** Function assertTruncateConstraint(pPager) checks that one of the +** following is true for all dirty pages currently in the page-cache: +** +** a) The page number is less than or equal to the size of the +** current database image, in pages, OR +** +** b) if the page content were written at this time, it would not +** be necessary to write the current content out to the sub-journal +** (as determined by function subjRequiresPage()). +** +** If the condition asserted by this function were not true, and the +** dirty page were to be discarded from the cache via the pagerStress() +** routine, pagerStress() would not write the current page content to +** the database file. If a savepoint transaction were rolled back after +** this happened, the correct behavior would be to restore the current +** content of the page. However, since this content is not present in either +** the database file or the portion of the rollback journal and +** sub-journal rolled back the content could not be restored and the +** database image would become corrupt. It is therefore fortunate that +** this circumstance cannot arise. +*/ +#if defined(SQLITE_DEBUG) +static void assertTruncateConstraintCb(PgHdr *pPg){ + assert( pPg->flags&PGHDR_DIRTY ); + assert( !subjRequiresPage(pPg) || pPg->pgno<=pPg->pPager->dbSize ); +} +static void assertTruncateConstraint(Pager *pPager){ + sqlite3PcacheIterateDirty(pPager->pPCache, assertTruncateConstraintCb); +} +#else +# define assertTruncateConstraint(pPager) +#endif + +/* +** Truncate the in-memory database file image to nPage pages. This +** function does not actually modify the database file on disk. It +** just sets the internal state of the pager object so that the +** truncation will be done when the current transaction is committed. +** +** This function is only called right before committing a transaction. +** Once this function has been called, the transaction must either be +** rolled back or committed. It is not safe to call this function and +** then continue writing to the database. +*/ +SQLITE_PRIVATE void sqlite3PagerTruncateImage(Pager *pPager, Pgno nPage){ + assert( pPager->dbSize>=nPage ); + assert( pPager->eState>=PAGER_WRITER_CACHEMOD ); + pPager->dbSize = nPage; + + /* At one point the code here called assertTruncateConstraint() to + ** ensure that all pages being truncated away by this operation are, + ** if one or more savepoints are open, present in the savepoint + ** journal so that they can be restored if the savepoint is rolled + ** back. This is no longer necessary as this function is now only + ** called right before committing a transaction. So although the + ** Pager object may still have open savepoints (Pager.nSavepoint!=0), + ** they cannot be rolled back. So the assertTruncateConstraint() call + ** is no longer correct. */ +} + + +/* +** This function is called before attempting a hot-journal rollback. It +** syncs the journal file to disk, then sets pPager->journalHdr to the +** size of the journal file so that the pager_playback() routine knows +** that the entire journal file has been synced. +** +** Syncing a hot-journal to disk before attempting to roll it back ensures +** that if a power-failure occurs during the rollback, the process that +** attempts rollback following system recovery sees the same journal +** content as this process. +** +** If everything goes as planned, SQLITE_OK is returned. Otherwise, +** an SQLite error code. +*/ +static int pagerSyncHotJournal(Pager *pPager){ + int rc = SQLITE_OK; + if( !pPager->noSync ){ + rc = sqlite3OsSync(pPager->jfd, SQLITE_SYNC_NORMAL); + } + if( rc==SQLITE_OK ){ + rc = sqlite3OsFileSize(pPager->jfd, &pPager->journalHdr); + } + return rc; +} + +#if SQLITE_MAX_MMAP_SIZE>0 +/* +** Obtain a reference to a memory mapped page object for page number pgno. +** The new object will use the pointer pData, obtained from xFetch(). +** If successful, set *ppPage to point to the new page reference +** and return SQLITE_OK. Otherwise, return an SQLite error code and set +** *ppPage to zero. +** +** Page references obtained by calling this function should be released +** by calling pagerReleaseMapPage(). +*/ +static int pagerAcquireMapPage( + Pager *pPager, /* Pager object */ + Pgno pgno, /* Page number */ + void *pData, /* xFetch()'d data for this page */ + PgHdr **ppPage /* OUT: Acquired page object */ +){ + PgHdr *p; /* Memory mapped page to return */ + + if( pPager->pMmapFreelist ){ + *ppPage = p = pPager->pMmapFreelist; + pPager->pMmapFreelist = p->pDirty; + p->pDirty = 0; + assert( pPager->nExtra>=8 ); + memset(p->pExtra, 0, 8); + }else{ + *ppPage = p = (PgHdr *)sqlite3MallocZero(sizeof(PgHdr) + pPager->nExtra); + if( p==0 ){ + sqlite3OsUnfetch(pPager->fd, (i64)(pgno-1) * pPager->pageSize, pData); + return SQLITE_NOMEM_BKPT; + } + p->pExtra = (void *)&p[1]; + p->flags = PGHDR_MMAP; + p->nRef = 1; + p->pPager = pPager; + } + + assert( p->pExtra==(void *)&p[1] ); + assert( p->pPage==0 ); + assert( p->flags==PGHDR_MMAP ); + assert( p->pPager==pPager ); + assert( p->nRef==1 ); + + p->pgno = pgno; + p->pData = pData; + pPager->nMmapOut++; + + return SQLITE_OK; +} +#endif + +/* +** Release a reference to page pPg. pPg must have been returned by an +** earlier call to pagerAcquireMapPage(). +*/ +static void pagerReleaseMapPage(PgHdr *pPg){ + Pager *pPager = pPg->pPager; + pPager->nMmapOut--; + pPg->pDirty = pPager->pMmapFreelist; + pPager->pMmapFreelist = pPg; + + assert( pPager->fd->pMethods->iVersion>=3 ); + sqlite3OsUnfetch(pPager->fd, (i64)(pPg->pgno-1)*pPager->pageSize, pPg->pData); +} + +/* +** Free all PgHdr objects stored in the Pager.pMmapFreelist list. +*/ +static void pagerFreeMapHdrs(Pager *pPager){ + PgHdr *p; + PgHdr *pNext; + for(p=pPager->pMmapFreelist; p; p=pNext){ + pNext = p->pDirty; + sqlite3_free(p); + } +} + +/* Verify that the database file has not be deleted or renamed out from +** under the pager. Return SQLITE_OK if the database is still where it ought +** to be on disk. Return non-zero (SQLITE_READONLY_DBMOVED or some other error +** code from sqlite3OsAccess()) if the database has gone missing. +*/ +static int databaseIsUnmoved(Pager *pPager){ + int bHasMoved = 0; + int rc; + + if( pPager->tempFile ) return SQLITE_OK; + if( pPager->dbSize==0 ) return SQLITE_OK; + assert( pPager->zFilename && pPager->zFilename[0] ); + rc = sqlite3OsFileControl(pPager->fd, SQLITE_FCNTL_HAS_MOVED, &bHasMoved); + if( rc==SQLITE_NOTFOUND ){ + /* If the HAS_MOVED file-control is unimplemented, assume that the file + ** has not been moved. That is the historical behavior of SQLite: prior to + ** version 3.8.3, it never checked */ + rc = SQLITE_OK; + }else if( rc==SQLITE_OK && bHasMoved ){ + rc = SQLITE_READONLY_DBMOVED; + } + return rc; +} + + +/* +** Shutdown the page cache. Free all memory and close all files. +** +** If a transaction was in progress when this routine is called, that +** transaction is rolled back. All outstanding pages are invalidated +** and their memory is freed. Any attempt to use a page associated +** with this page cache after this function returns will likely +** result in a coredump. +** +** This function always succeeds. If a transaction is active an attempt +** is made to roll it back. If an error occurs during the rollback +** a hot journal may be left in the filesystem but no error is returned +** to the caller. +*/ +SQLITE_PRIVATE int sqlite3PagerClose(Pager *pPager, sqlite3 *db){ + u8 *pTmp = (u8*)pPager->pTmpSpace; + assert( db || pagerUseWal(pPager)==0 ); + assert( assert_pager_state(pPager) ); + disable_simulated_io_errors(); + sqlite3BeginBenignMalloc(); + pagerFreeMapHdrs(pPager); + /* pPager->errCode = 0; */ + pPager->exclusiveMode = 0; +#ifndef SQLITE_OMIT_WAL + { + u8 *a = 0; + assert( db || pPager->pWal==0 ); + if( db && 0==(db->flags & SQLITE_NoCkptOnClose) + && SQLITE_OK==databaseIsUnmoved(pPager) + ){ + a = pTmp; + } + sqlite3WalClose(pPager->pWal, db, pPager->walSyncFlags, pPager->pageSize,a); + pPager->pWal = 0; + } +#endif + pager_reset(pPager); + if( MEMDB ){ + pager_unlock(pPager); + }else{ + /* If it is open, sync the journal file before calling UnlockAndRollback. + ** If this is not done, then an unsynced portion of the open journal + ** file may be played back into the database. If a power failure occurs + ** while this is happening, the database could become corrupt. + ** + ** If an error occurs while trying to sync the journal, shift the pager + ** into the ERROR state. This causes UnlockAndRollback to unlock the + ** database and close the journal file without attempting to roll it + ** back or finalize it. The next database user will have to do hot-journal + ** rollback before accessing the database file. + */ + if( isOpen(pPager->jfd) ){ + pager_error(pPager, pagerSyncHotJournal(pPager)); + } + pagerUnlockAndRollback(pPager); + } + sqlite3EndBenignMalloc(); + enable_simulated_io_errors(); + PAGERTRACE(("CLOSE %d\n", PAGERID(pPager))); + IOTRACE(("CLOSE %p\n", pPager)) + sqlite3OsClose(pPager->jfd); + sqlite3OsClose(pPager->fd); + sqlite3PageFree(pTmp); + sqlite3PcacheClose(pPager->pPCache); + +#ifdef SQLITE_HAS_CODEC + if( pPager->xCodecFree ) pPager->xCodecFree(pPager->pCodec); +#endif + + assert( !pPager->aSavepoint && !pPager->pInJournal ); + assert( !isOpen(pPager->jfd) && !isOpen(pPager->sjfd) ); + + sqlite3_free(pPager); + return SQLITE_OK; +} + +#if !defined(NDEBUG) || defined(SQLITE_TEST) +/* +** Return the page number for page pPg. +*/ +SQLITE_PRIVATE Pgno sqlite3PagerPagenumber(DbPage *pPg){ + return pPg->pgno; +} +#endif + +/* +** Increment the reference count for page pPg. +*/ +SQLITE_PRIVATE void sqlite3PagerRef(DbPage *pPg){ + sqlite3PcacheRef(pPg); +} + +/* +** Sync the journal. In other words, make sure all the pages that have +** been written to the journal have actually reached the surface of the +** disk and can be restored in the event of a hot-journal rollback. +** +** If the Pager.noSync flag is set, then this function is a no-op. +** Otherwise, the actions required depend on the journal-mode and the +** device characteristics of the file-system, as follows: +** +** * If the journal file is an in-memory journal file, no action need +** be taken. +** +** * Otherwise, if the device does not support the SAFE_APPEND property, +** then the nRec field of the most recently written journal header +** is updated to contain the number of journal records that have +** been written following it. If the pager is operating in full-sync +** mode, then the journal file is synced before this field is updated. +** +** * If the device does not support the SEQUENTIAL property, then +** journal file is synced. +** +** Or, in pseudo-code: +** +** if( NOT ){ +** if( NOT SAFE_APPEND ){ +** if( ) xSync(); +** +** } +** if( NOT SEQUENTIAL ) xSync(); +** } +** +** If successful, this routine clears the PGHDR_NEED_SYNC flag of every +** page currently held in memory before returning SQLITE_OK. If an IO +** error is encountered, then the IO error code is returned to the caller. +*/ +static int syncJournal(Pager *pPager, int newHdr){ + int rc; /* Return code */ + + assert( pPager->eState==PAGER_WRITER_CACHEMOD + || pPager->eState==PAGER_WRITER_DBMOD + ); + assert( assert_pager_state(pPager) ); + assert( !pagerUseWal(pPager) ); + + rc = sqlite3PagerExclusiveLock(pPager); + if( rc!=SQLITE_OK ) return rc; + + if( !pPager->noSync ){ + assert( !pPager->tempFile ); + if( isOpen(pPager->jfd) && pPager->journalMode!=PAGER_JOURNALMODE_MEMORY ){ + const int iDc = sqlite3OsDeviceCharacteristics(pPager->fd); + assert( isOpen(pPager->jfd) ); + + if( 0==(iDc&SQLITE_IOCAP_SAFE_APPEND) ){ + /* This block deals with an obscure problem. If the last connection + ** that wrote to this database was operating in persistent-journal + ** mode, then the journal file may at this point actually be larger + ** than Pager.journalOff bytes. If the next thing in the journal + ** file happens to be a journal-header (written as part of the + ** previous connection's transaction), and a crash or power-failure + ** occurs after nRec is updated but before this connection writes + ** anything else to the journal file (or commits/rolls back its + ** transaction), then SQLite may become confused when doing the + ** hot-journal rollback following recovery. It may roll back all + ** of this connections data, then proceed to rolling back the old, + ** out-of-date data that follows it. Database corruption. + ** + ** To work around this, if the journal file does appear to contain + ** a valid header following Pager.journalOff, then write a 0x00 + ** byte to the start of it to prevent it from being recognized. + ** + ** Variable iNextHdrOffset is set to the offset at which this + ** problematic header will occur, if it exists. aMagic is used + ** as a temporary buffer to inspect the first couple of bytes of + ** the potential journal header. + */ + i64 iNextHdrOffset; + u8 aMagic[8]; + u8 zHeader[sizeof(aJournalMagic)+4]; + + memcpy(zHeader, aJournalMagic, sizeof(aJournalMagic)); + put32bits(&zHeader[sizeof(aJournalMagic)], pPager->nRec); + + iNextHdrOffset = journalHdrOffset(pPager); + rc = sqlite3OsRead(pPager->jfd, aMagic, 8, iNextHdrOffset); + if( rc==SQLITE_OK && 0==memcmp(aMagic, aJournalMagic, 8) ){ + static const u8 zerobyte = 0; + rc = sqlite3OsWrite(pPager->jfd, &zerobyte, 1, iNextHdrOffset); + } + if( rc!=SQLITE_OK && rc!=SQLITE_IOERR_SHORT_READ ){ + return rc; + } + + /* Write the nRec value into the journal file header. If in + ** full-synchronous mode, sync the journal first. This ensures that + ** all data has really hit the disk before nRec is updated to mark + ** it as a candidate for rollback. + ** + ** This is not required if the persistent media supports the + ** SAFE_APPEND property. Because in this case it is not possible + ** for garbage data to be appended to the file, the nRec field + ** is populated with 0xFFFFFFFF when the journal header is written + ** and never needs to be updated. + */ + if( pPager->fullSync && 0==(iDc&SQLITE_IOCAP_SEQUENTIAL) ){ + PAGERTRACE(("SYNC journal of %d\n", PAGERID(pPager))); + IOTRACE(("JSYNC %p\n", pPager)) + rc = sqlite3OsSync(pPager->jfd, pPager->syncFlags); + if( rc!=SQLITE_OK ) return rc; + } + IOTRACE(("JHDR %p %lld\n", pPager, pPager->journalHdr)); + rc = sqlite3OsWrite( + pPager->jfd, zHeader, sizeof(zHeader), pPager->journalHdr + ); + if( rc!=SQLITE_OK ) return rc; + } + if( 0==(iDc&SQLITE_IOCAP_SEQUENTIAL) ){ + PAGERTRACE(("SYNC journal of %d\n", PAGERID(pPager))); + IOTRACE(("JSYNC %p\n", pPager)) + rc = sqlite3OsSync(pPager->jfd, pPager->syncFlags| + (pPager->syncFlags==SQLITE_SYNC_FULL?SQLITE_SYNC_DATAONLY:0) + ); + if( rc!=SQLITE_OK ) return rc; + } + + pPager->journalHdr = pPager->journalOff; + if( newHdr && 0==(iDc&SQLITE_IOCAP_SAFE_APPEND) ){ + pPager->nRec = 0; + rc = writeJournalHdr(pPager); + if( rc!=SQLITE_OK ) return rc; + } + }else{ + pPager->journalHdr = pPager->journalOff; + } + } + + /* Unless the pager is in noSync mode, the journal file was just + ** successfully synced. Either way, clear the PGHDR_NEED_SYNC flag on + ** all pages. + */ + sqlite3PcacheClearSyncFlags(pPager->pPCache); + pPager->eState = PAGER_WRITER_DBMOD; + assert( assert_pager_state(pPager) ); + return SQLITE_OK; +} + +/* +** The argument is the first in a linked list of dirty pages connected +** by the PgHdr.pDirty pointer. This function writes each one of the +** in-memory pages in the list to the database file. The argument may +** be NULL, representing an empty list. In this case this function is +** a no-op. +** +** The pager must hold at least a RESERVED lock when this function +** is called. Before writing anything to the database file, this lock +** is upgraded to an EXCLUSIVE lock. If the lock cannot be obtained, +** SQLITE_BUSY is returned and no data is written to the database file. +** +** If the pager is a temp-file pager and the actual file-system file +** is not yet open, it is created and opened before any data is +** written out. +** +** Once the lock has been upgraded and, if necessary, the file opened, +** the pages are written out to the database file in list order. Writing +** a page is skipped if it meets either of the following criteria: +** +** * The page number is greater than Pager.dbSize, or +** * The PGHDR_DONT_WRITE flag is set on the page. +** +** If writing out a page causes the database file to grow, Pager.dbFileSize +** is updated accordingly. If page 1 is written out, then the value cached +** in Pager.dbFileVers[] is updated to match the new value stored in +** the database file. +** +** If everything is successful, SQLITE_OK is returned. If an IO error +** occurs, an IO error code is returned. Or, if the EXCLUSIVE lock cannot +** be obtained, SQLITE_BUSY is returned. +*/ +static int pager_write_pagelist(Pager *pPager, PgHdr *pList){ + int rc = SQLITE_OK; /* Return code */ + + /* This function is only called for rollback pagers in WRITER_DBMOD state. */ + assert( !pagerUseWal(pPager) ); + assert( pPager->tempFile || pPager->eState==PAGER_WRITER_DBMOD ); + assert( pPager->eLock==EXCLUSIVE_LOCK ); + assert( isOpen(pPager->fd) || pList->pDirty==0 ); + + /* If the file is a temp-file has not yet been opened, open it now. It + ** is not possible for rc to be other than SQLITE_OK if this branch + ** is taken, as pager_wait_on_lock() is a no-op for temp-files. + */ + if( !isOpen(pPager->fd) ){ + assert( pPager->tempFile && rc==SQLITE_OK ); + rc = pagerOpentemp(pPager, pPager->fd, pPager->vfsFlags); + } + + /* Before the first write, give the VFS a hint of what the final + ** file size will be. + */ + assert( rc!=SQLITE_OK || isOpen(pPager->fd) ); + if( rc==SQLITE_OK + && pPager->dbHintSizedbSize + && (pList->pDirty || pList->pgno>pPager->dbHintSize) + ){ + sqlite3_int64 szFile = pPager->pageSize * (sqlite3_int64)pPager->dbSize; + sqlite3OsFileControlHint(pPager->fd, SQLITE_FCNTL_SIZE_HINT, &szFile); + pPager->dbHintSize = pPager->dbSize; + } + + while( rc==SQLITE_OK && pList ){ + Pgno pgno = pList->pgno; + + /* If there are dirty pages in the page cache with page numbers greater + ** than Pager.dbSize, this means sqlite3PagerTruncateImage() was called to + ** make the file smaller (presumably by auto-vacuum code). Do not write + ** any such pages to the file. + ** + ** Also, do not write out any page that has the PGHDR_DONT_WRITE flag + ** set (set by sqlite3PagerDontWrite()). + */ + if( pgno<=pPager->dbSize && 0==(pList->flags&PGHDR_DONT_WRITE) ){ + i64 offset = (pgno-1)*(i64)pPager->pageSize; /* Offset to write */ + char *pData; /* Data to write */ + + assert( (pList->flags&PGHDR_NEED_SYNC)==0 ); + if( pList->pgno==1 ) pager_write_changecounter(pList); + + /* Encode the database */ + CODEC2(pPager, pList->pData, pgno, 6, return SQLITE_NOMEM_BKPT, pData); + + /* Write out the page data. */ + rc = sqlite3OsWrite(pPager->fd, pData, pPager->pageSize, offset); + + /* If page 1 was just written, update Pager.dbFileVers to match + ** the value now stored in the database file. If writing this + ** page caused the database file to grow, update dbFileSize. + */ + if( pgno==1 ){ + memcpy(&pPager->dbFileVers, &pData[24], sizeof(pPager->dbFileVers)); + } + if( pgno>pPager->dbFileSize ){ + pPager->dbFileSize = pgno; + } + pPager->aStat[PAGER_STAT_WRITE]++; + + /* Update any backup objects copying the contents of this pager. */ + sqlite3BackupUpdate(pPager->pBackup, pgno, (u8*)pList->pData); + + PAGERTRACE(("STORE %d page %d hash(%08x)\n", + PAGERID(pPager), pgno, pager_pagehash(pList))); + IOTRACE(("PGOUT %p %d\n", pPager, pgno)); + PAGER_INCR(sqlite3_pager_writedb_count); + }else{ + PAGERTRACE(("NOSTORE %d page %d\n", PAGERID(pPager), pgno)); + } + pager_set_pagehash(pList); + pList = pList->pDirty; + } + + return rc; +} + +/* +** Ensure that the sub-journal file is open. If it is already open, this +** function is a no-op. +** +** SQLITE_OK is returned if everything goes according to plan. An +** SQLITE_IOERR_XXX error code is returned if a call to sqlite3OsOpen() +** fails. +*/ +static int openSubJournal(Pager *pPager){ + int rc = SQLITE_OK; + if( !isOpen(pPager->sjfd) ){ + const int flags = SQLITE_OPEN_SUBJOURNAL | SQLITE_OPEN_READWRITE + | SQLITE_OPEN_CREATE | SQLITE_OPEN_EXCLUSIVE + | SQLITE_OPEN_DELETEONCLOSE; + int nStmtSpill = sqlite3Config.nStmtSpill; + if( pPager->journalMode==PAGER_JOURNALMODE_MEMORY || pPager->subjInMemory ){ + nStmtSpill = -1; + } + rc = sqlite3JournalOpen(pPager->pVfs, 0, pPager->sjfd, flags, nStmtSpill); + } + return rc; +} + +/* +** Append a record of the current state of page pPg to the sub-journal. +** +** If successful, set the bit corresponding to pPg->pgno in the bitvecs +** for all open savepoints before returning. +** +** This function returns SQLITE_OK if everything is successful, an IO +** error code if the attempt to write to the sub-journal fails, or +** SQLITE_NOMEM if a malloc fails while setting a bit in a savepoint +** bitvec. +*/ +static int subjournalPage(PgHdr *pPg){ + int rc = SQLITE_OK; + Pager *pPager = pPg->pPager; + if( pPager->journalMode!=PAGER_JOURNALMODE_OFF ){ + + /* Open the sub-journal, if it has not already been opened */ + assert( pPager->useJournal ); + assert( isOpen(pPager->jfd) || pagerUseWal(pPager) ); + assert( isOpen(pPager->sjfd) || pPager->nSubRec==0 ); + assert( pagerUseWal(pPager) + || pageInJournal(pPager, pPg) + || pPg->pgno>pPager->dbOrigSize + ); + rc = openSubJournal(pPager); + + /* If the sub-journal was opened successfully (or was already open), + ** write the journal record into the file. */ + if( rc==SQLITE_OK ){ + void *pData = pPg->pData; + i64 offset = (i64)pPager->nSubRec*(4+pPager->pageSize); + char *pData2; + +#if SQLITE_HAS_CODEC + if( !pPager->subjInMemory ){ + CODEC2(pPager, pData, pPg->pgno, 7, return SQLITE_NOMEM_BKPT, pData2); + }else +#endif + pData2 = pData; + PAGERTRACE(("STMT-JOURNAL %d page %d\n", PAGERID(pPager), pPg->pgno)); + rc = write32bits(pPager->sjfd, offset, pPg->pgno); + if( rc==SQLITE_OK ){ + rc = sqlite3OsWrite(pPager->sjfd, pData2, pPager->pageSize, offset+4); + } + } + } + if( rc==SQLITE_OK ){ + pPager->nSubRec++; + assert( pPager->nSavepoint>0 ); + rc = addToSavepointBitvecs(pPager, pPg->pgno); + } + return rc; +} +static int subjournalPageIfRequired(PgHdr *pPg){ + if( subjRequiresPage(pPg) ){ + return subjournalPage(pPg); + }else{ + return SQLITE_OK; + } +} + +/* +** This function is called by the pcache layer when it has reached some +** soft memory limit. The first argument is a pointer to a Pager object +** (cast as a void*). The pager is always 'purgeable' (not an in-memory +** database). The second argument is a reference to a page that is +** currently dirty but has no outstanding references. The page +** is always associated with the Pager object passed as the first +** argument. +** +** The job of this function is to make pPg clean by writing its contents +** out to the database file, if possible. This may involve syncing the +** journal file. +** +** If successful, sqlite3PcacheMakeClean() is called on the page and +** SQLITE_OK returned. If an IO error occurs while trying to make the +** page clean, the IO error code is returned. If the page cannot be +** made clean for some other reason, but no error occurs, then SQLITE_OK +** is returned by sqlite3PcacheMakeClean() is not called. +*/ +static int pagerStress(void *p, PgHdr *pPg){ + Pager *pPager = (Pager *)p; + int rc = SQLITE_OK; + + assert( pPg->pPager==pPager ); + assert( pPg->flags&PGHDR_DIRTY ); + + /* The doNotSpill NOSYNC bit is set during times when doing a sync of + ** journal (and adding a new header) is not allowed. This occurs + ** during calls to sqlite3PagerWrite() while trying to journal multiple + ** pages belonging to the same sector. + ** + ** The doNotSpill ROLLBACK and OFF bits inhibits all cache spilling + ** regardless of whether or not a sync is required. This is set during + ** a rollback or by user request, respectively. + ** + ** Spilling is also prohibited when in an error state since that could + ** lead to database corruption. In the current implementation it + ** is impossible for sqlite3PcacheFetch() to be called with createFlag==3 + ** while in the error state, hence it is impossible for this routine to + ** be called in the error state. Nevertheless, we include a NEVER() + ** test for the error state as a safeguard against future changes. + */ + if( NEVER(pPager->errCode) ) return SQLITE_OK; + testcase( pPager->doNotSpill & SPILLFLAG_ROLLBACK ); + testcase( pPager->doNotSpill & SPILLFLAG_OFF ); + testcase( pPager->doNotSpill & SPILLFLAG_NOSYNC ); + if( pPager->doNotSpill + && ((pPager->doNotSpill & (SPILLFLAG_ROLLBACK|SPILLFLAG_OFF))!=0 + || (pPg->flags & PGHDR_NEED_SYNC)!=0) + ){ + return SQLITE_OK; + } + + pPager->aStat[PAGER_STAT_SPILL]++; + pPg->pDirty = 0; + if( pagerUseWal(pPager) ){ + /* Write a single frame for this page to the log. */ + rc = subjournalPageIfRequired(pPg); + if( rc==SQLITE_OK ){ + rc = pagerWalFrames(pPager, pPg, 0, 0); + } + }else{ + +#ifdef SQLITE_ENABLE_BATCH_ATOMIC_WRITE + if( pPager->tempFile==0 ){ + rc = sqlite3JournalCreate(pPager->jfd); + if( rc!=SQLITE_OK ) return pager_error(pPager, rc); + } +#endif + + /* Sync the journal file if required. */ + if( pPg->flags&PGHDR_NEED_SYNC + || pPager->eState==PAGER_WRITER_CACHEMOD + ){ + rc = syncJournal(pPager, 1); + } + + /* Write the contents of the page out to the database file. */ + if( rc==SQLITE_OK ){ + assert( (pPg->flags&PGHDR_NEED_SYNC)==0 ); + rc = pager_write_pagelist(pPager, pPg); + } + } + + /* Mark the page as clean. */ + if( rc==SQLITE_OK ){ + PAGERTRACE(("STRESS %d page %d\n", PAGERID(pPager), pPg->pgno)); + sqlite3PcacheMakeClean(pPg); + } + + return pager_error(pPager, rc); +} + +/* +** Flush all unreferenced dirty pages to disk. +*/ +SQLITE_PRIVATE int sqlite3PagerFlush(Pager *pPager){ + int rc = pPager->errCode; + if( !MEMDB ){ + PgHdr *pList = sqlite3PcacheDirtyList(pPager->pPCache); + assert( assert_pager_state(pPager) ); + while( rc==SQLITE_OK && pList ){ + PgHdr *pNext = pList->pDirty; + if( pList->nRef==0 ){ + rc = pagerStress((void*)pPager, pList); + } + pList = pNext; + } + } + + return rc; +} + +/* +** Allocate and initialize a new Pager object and put a pointer to it +** in *ppPager. The pager should eventually be freed by passing it +** to sqlite3PagerClose(). +** +** The zFilename argument is the path to the database file to open. +** If zFilename is NULL then a randomly-named temporary file is created +** and used as the file to be cached. Temporary files are be deleted +** automatically when they are closed. If zFilename is ":memory:" then +** all information is held in cache. It is never written to disk. +** This can be used to implement an in-memory database. +** +** The nExtra parameter specifies the number of bytes of space allocated +** along with each page reference. This space is available to the user +** via the sqlite3PagerGetExtra() API. When a new page is allocated, the +** first 8 bytes of this space are zeroed but the remainder is uninitialized. +** (The extra space is used by btree as the MemPage object.) +** +** The flags argument is used to specify properties that affect the +** operation of the pager. It should be passed some bitwise combination +** of the PAGER_* flags. +** +** The vfsFlags parameter is a bitmask to pass to the flags parameter +** of the xOpen() method of the supplied VFS when opening files. +** +** If the pager object is allocated and the specified file opened +** successfully, SQLITE_OK is returned and *ppPager set to point to +** the new pager object. If an error occurs, *ppPager is set to NULL +** and error code returned. This function may return SQLITE_NOMEM +** (sqlite3Malloc() is used to allocate memory), SQLITE_CANTOPEN or +** various SQLITE_IO_XXX errors. +*/ +SQLITE_PRIVATE int sqlite3PagerOpen( + sqlite3_vfs *pVfs, /* The virtual file system to use */ + Pager **ppPager, /* OUT: Return the Pager structure here */ + const char *zFilename, /* Name of the database file to open */ + int nExtra, /* Extra bytes append to each in-memory page */ + int flags, /* flags controlling this file */ + int vfsFlags, /* flags passed through to sqlite3_vfs.xOpen() */ + void (*xReinit)(DbPage*) /* Function to reinitialize pages */ +){ + u8 *pPtr; + Pager *pPager = 0; /* Pager object to allocate and return */ + int rc = SQLITE_OK; /* Return code */ + int tempFile = 0; /* True for temp files (incl. in-memory files) */ + int memDb = 0; /* True if this is an in-memory file */ +#ifdef SQLITE_ENABLE_DESERIALIZE + int memJM = 0; /* Memory journal mode */ +#else +# define memJM 0 +#endif + int readOnly = 0; /* True if this is a read-only file */ + int journalFileSize; /* Bytes to allocate for each journal fd */ + char *zPathname = 0; /* Full path to database file */ + int nPathname = 0; /* Number of bytes in zPathname */ + int useJournal = (flags & PAGER_OMIT_JOURNAL)==0; /* False to omit journal */ + int pcacheSize = sqlite3PcacheSize(); /* Bytes to allocate for PCache */ + u32 szPageDflt = SQLITE_DEFAULT_PAGE_SIZE; /* Default page size */ + const char *zUri = 0; /* URI args to copy */ + int nUri = 0; /* Number of bytes of URI args at *zUri */ + + /* Figure out how much space is required for each journal file-handle + ** (there are two of them, the main journal and the sub-journal). */ + journalFileSize = ROUND8(sqlite3JournalSize(pVfs)); + + /* Set the output variable to NULL in case an error occurs. */ + *ppPager = 0; + +#ifndef SQLITE_OMIT_MEMORYDB + if( flags & PAGER_MEMORY ){ + memDb = 1; + if( zFilename && zFilename[0] ){ + zPathname = sqlite3DbStrDup(0, zFilename); + if( zPathname==0 ) return SQLITE_NOMEM_BKPT; + nPathname = sqlite3Strlen30(zPathname); + zFilename = 0; + } + } +#endif + + /* Compute and store the full pathname in an allocated buffer pointed + ** to by zPathname, length nPathname. Or, if this is a temporary file, + ** leave both nPathname and zPathname set to 0. + */ + if( zFilename && zFilename[0] ){ + const char *z; + nPathname = pVfs->mxPathname+1; + zPathname = sqlite3DbMallocRaw(0, nPathname*2); + if( zPathname==0 ){ + return SQLITE_NOMEM_BKPT; + } + zPathname[0] = 0; /* Make sure initialized even if FullPathname() fails */ + rc = sqlite3OsFullPathname(pVfs, zFilename, nPathname, zPathname); + nPathname = sqlite3Strlen30(zPathname); + z = zUri = &zFilename[sqlite3Strlen30(zFilename)+1]; + while( *z ){ + z += sqlite3Strlen30(z)+1; + z += sqlite3Strlen30(z)+1; + } + nUri = (int)(&z[1] - zUri); + assert( nUri>=0 ); + if( rc==SQLITE_OK && nPathname+8>pVfs->mxPathname ){ + /* This branch is taken when the journal path required by + ** the database being opened will be more than pVfs->mxPathname + ** bytes in length. This means the database cannot be opened, + ** as it will not be possible to open the journal file or even + ** check for a hot-journal before reading. + */ + rc = SQLITE_CANTOPEN_BKPT; + } + if( rc!=SQLITE_OK ){ + sqlite3DbFree(0, zPathname); + return rc; + } + } + + /* Allocate memory for the Pager structure, PCache object, the + ** three file descriptors, the database file name and the journal + ** file name. The layout in memory is as follows: + ** + ** Pager object (sizeof(Pager) bytes) + ** PCache object (sqlite3PcacheSize() bytes) + ** Database file handle (pVfs->szOsFile bytes) + ** Sub-journal file handle (journalFileSize bytes) + ** Main journal file handle (journalFileSize bytes) + ** Database file name (nPathname+1 bytes) + ** Journal file name (nPathname+8+1 bytes) + */ + pPtr = (u8 *)sqlite3MallocZero( + ROUND8(sizeof(*pPager)) + /* Pager structure */ + ROUND8(pcacheSize) + /* PCache object */ + ROUND8(pVfs->szOsFile) + /* The main db file */ + journalFileSize * 2 + /* The two journal files */ + nPathname + 1 + nUri + /* zFilename */ + nPathname + 8 + 2 /* zJournal */ +#ifndef SQLITE_OMIT_WAL + + nPathname + 4 + 2 /* zWal */ +#endif + ); + assert( EIGHT_BYTE_ALIGNMENT(SQLITE_INT_TO_PTR(journalFileSize)) ); + if( !pPtr ){ + sqlite3DbFree(0, zPathname); + return SQLITE_NOMEM_BKPT; + } + pPager = (Pager*)(pPtr); + pPager->pPCache = (PCache*)(pPtr += ROUND8(sizeof(*pPager))); + pPager->fd = (sqlite3_file*)(pPtr += ROUND8(pcacheSize)); + pPager->sjfd = (sqlite3_file*)(pPtr += ROUND8(pVfs->szOsFile)); + pPager->jfd = (sqlite3_file*)(pPtr += journalFileSize); + pPager->zFilename = (char*)(pPtr += journalFileSize); + assert( EIGHT_BYTE_ALIGNMENT(pPager->jfd) ); + + /* Fill in the Pager.zFilename and Pager.zJournal buffers, if required. */ + if( zPathname ){ + assert( nPathname>0 ); + pPager->zJournal = (char*)(pPtr += nPathname + 1 + nUri); + memcpy(pPager->zFilename, zPathname, nPathname); + if( nUri ) memcpy(&pPager->zFilename[nPathname+1], zUri, nUri); + memcpy(pPager->zJournal, zPathname, nPathname); + memcpy(&pPager->zJournal[nPathname], "-journal\000", 8+2); + sqlite3FileSuffix3(pPager->zFilename, pPager->zJournal); +#ifndef SQLITE_OMIT_WAL + pPager->zWal = &pPager->zJournal[nPathname+8+1]; + memcpy(pPager->zWal, zPathname, nPathname); + memcpy(&pPager->zWal[nPathname], "-wal\000", 4+1); + sqlite3FileSuffix3(pPager->zFilename, pPager->zWal); +#endif + sqlite3DbFree(0, zPathname); + } + pPager->pVfs = pVfs; + pPager->vfsFlags = vfsFlags; + + /* Open the pager file. + */ + if( zFilename && zFilename[0] ){ + int fout = 0; /* VFS flags returned by xOpen() */ + rc = sqlite3OsOpen(pVfs, pPager->zFilename, pPager->fd, vfsFlags, &fout); + assert( !memDb ); +#ifdef SQLITE_ENABLE_DESERIALIZE + memJM = (fout&SQLITE_OPEN_MEMORY)!=0; +#endif + readOnly = (fout&SQLITE_OPEN_READONLY)!=0; + + /* If the file was successfully opened for read/write access, + ** choose a default page size in case we have to create the + ** database file. The default page size is the maximum of: + ** + ** + SQLITE_DEFAULT_PAGE_SIZE, + ** + The value returned by sqlite3OsSectorSize() + ** + The largest page size that can be written atomically. + */ + if( rc==SQLITE_OK ){ + int iDc = sqlite3OsDeviceCharacteristics(pPager->fd); + if( !readOnly ){ + setSectorSize(pPager); + assert(SQLITE_DEFAULT_PAGE_SIZE<=SQLITE_MAX_DEFAULT_PAGE_SIZE); + if( szPageDfltsectorSize ){ + if( pPager->sectorSize>SQLITE_MAX_DEFAULT_PAGE_SIZE ){ + szPageDflt = SQLITE_MAX_DEFAULT_PAGE_SIZE; + }else{ + szPageDflt = (u32)pPager->sectorSize; + } + } +#ifdef SQLITE_ENABLE_ATOMIC_WRITE + { + int ii; + assert(SQLITE_IOCAP_ATOMIC512==(512>>8)); + assert(SQLITE_IOCAP_ATOMIC64K==(65536>>8)); + assert(SQLITE_MAX_DEFAULT_PAGE_SIZE<=65536); + for(ii=szPageDflt; ii<=SQLITE_MAX_DEFAULT_PAGE_SIZE; ii=ii*2){ + if( iDc&(SQLITE_IOCAP_ATOMIC|(ii>>8)) ){ + szPageDflt = ii; + } + } + } +#endif + } + pPager->noLock = sqlite3_uri_boolean(zFilename, "nolock", 0); + if( (iDc & SQLITE_IOCAP_IMMUTABLE)!=0 + || sqlite3_uri_boolean(zFilename, "immutable", 0) ){ + vfsFlags |= SQLITE_OPEN_READONLY; + goto act_like_temp_file; + } + } + }else{ + /* If a temporary file is requested, it is not opened immediately. + ** In this case we accept the default page size and delay actually + ** opening the file until the first call to OsWrite(). + ** + ** This branch is also run for an in-memory database. An in-memory + ** database is the same as a temp-file that is never written out to + ** disk and uses an in-memory rollback journal. + ** + ** This branch also runs for files marked as immutable. + */ +act_like_temp_file: + tempFile = 1; + pPager->eState = PAGER_READER; /* Pretend we already have a lock */ + pPager->eLock = EXCLUSIVE_LOCK; /* Pretend we are in EXCLUSIVE mode */ + pPager->noLock = 1; /* Do no locking */ + readOnly = (vfsFlags&SQLITE_OPEN_READONLY); + } + + /* The following call to PagerSetPagesize() serves to set the value of + ** Pager.pageSize and to allocate the Pager.pTmpSpace buffer. + */ + if( rc==SQLITE_OK ){ + assert( pPager->memDb==0 ); + rc = sqlite3PagerSetPagesize(pPager, &szPageDflt, -1); + testcase( rc!=SQLITE_OK ); + } + + /* Initialize the PCache object. */ + if( rc==SQLITE_OK ){ + nExtra = ROUND8(nExtra); + assert( nExtra>=8 && nExtra<1000 ); + rc = sqlite3PcacheOpen(szPageDflt, nExtra, !memDb, + !memDb?pagerStress:0, (void *)pPager, pPager->pPCache); + } + + /* If an error occurred above, free the Pager structure and close the file. + */ + if( rc!=SQLITE_OK ){ + sqlite3OsClose(pPager->fd); + sqlite3PageFree(pPager->pTmpSpace); + sqlite3_free(pPager); + return rc; + } + + PAGERTRACE(("OPEN %d %s\n", FILEHANDLEID(pPager->fd), pPager->zFilename)); + IOTRACE(("OPEN %p %s\n", pPager, pPager->zFilename)) + + pPager->useJournal = (u8)useJournal; + /* pPager->stmtOpen = 0; */ + /* pPager->stmtInUse = 0; */ + /* pPager->nRef = 0; */ + /* pPager->stmtSize = 0; */ + /* pPager->stmtJSize = 0; */ + /* pPager->nPage = 0; */ + pPager->mxPgno = SQLITE_MAX_PAGE_COUNT; + /* pPager->state = PAGER_UNLOCK; */ + /* pPager->errMask = 0; */ + pPager->tempFile = (u8)tempFile; + assert( tempFile==PAGER_LOCKINGMODE_NORMAL + || tempFile==PAGER_LOCKINGMODE_EXCLUSIVE ); + assert( PAGER_LOCKINGMODE_EXCLUSIVE==1 ); + pPager->exclusiveMode = (u8)tempFile; + pPager->changeCountDone = pPager->tempFile; + pPager->memDb = (u8)memDb; + pPager->readOnly = (u8)readOnly; + assert( useJournal || pPager->tempFile ); + pPager->noSync = pPager->tempFile; + if( pPager->noSync ){ + assert( pPager->fullSync==0 ); + assert( pPager->extraSync==0 ); + assert( pPager->syncFlags==0 ); + assert( pPager->walSyncFlags==0 ); + }else{ + pPager->fullSync = 1; + pPager->extraSync = 0; + pPager->syncFlags = SQLITE_SYNC_NORMAL; + pPager->walSyncFlags = SQLITE_SYNC_NORMAL | (SQLITE_SYNC_NORMAL<<2); + } + /* pPager->pFirst = 0; */ + /* pPager->pFirstSynced = 0; */ + /* pPager->pLast = 0; */ + pPager->nExtra = (u16)nExtra; + pPager->journalSizeLimit = SQLITE_DEFAULT_JOURNAL_SIZE_LIMIT; + assert( isOpen(pPager->fd) || tempFile ); + setSectorSize(pPager); + if( !useJournal ){ + pPager->journalMode = PAGER_JOURNALMODE_OFF; + }else if( memDb || memJM ){ + pPager->journalMode = PAGER_JOURNALMODE_MEMORY; + } + /* pPager->xBusyHandler = 0; */ + /* pPager->pBusyHandlerArg = 0; */ + pPager->xReiniter = xReinit; + setGetterMethod(pPager); + /* memset(pPager->aHash, 0, sizeof(pPager->aHash)); */ + /* pPager->szMmap = SQLITE_DEFAULT_MMAP_SIZE // will be set by btree.c */ + + *ppPager = pPager; + return SQLITE_OK; +} + + + +/* +** This function is called after transitioning from PAGER_UNLOCK to +** PAGER_SHARED state. It tests if there is a hot journal present in +** the file-system for the given pager. A hot journal is one that +** needs to be played back. According to this function, a hot-journal +** file exists if the following criteria are met: +** +** * The journal file exists in the file system, and +** * No process holds a RESERVED or greater lock on the database file, and +** * The database file itself is greater than 0 bytes in size, and +** * The first byte of the journal file exists and is not 0x00. +** +** If the current size of the database file is 0 but a journal file +** exists, that is probably an old journal left over from a prior +** database with the same name. In this case the journal file is +** just deleted using OsDelete, *pExists is set to 0 and SQLITE_OK +** is returned. +** +** This routine does not check if there is a master journal filename +** at the end of the file. If there is, and that master journal file +** does not exist, then the journal file is not really hot. In this +** case this routine will return a false-positive. The pager_playback() +** routine will discover that the journal file is not really hot and +** will not roll it back. +** +** If a hot-journal file is found to exist, *pExists is set to 1 and +** SQLITE_OK returned. If no hot-journal file is present, *pExists is +** set to 0 and SQLITE_OK returned. If an IO error occurs while trying +** to determine whether or not a hot-journal file exists, the IO error +** code is returned and the value of *pExists is undefined. +*/ +static int hasHotJournal(Pager *pPager, int *pExists){ + sqlite3_vfs * const pVfs = pPager->pVfs; + int rc = SQLITE_OK; /* Return code */ + int exists = 1; /* True if a journal file is present */ + int jrnlOpen = !!isOpen(pPager->jfd); + + assert( pPager->useJournal ); + assert( isOpen(pPager->fd) ); + assert( pPager->eState==PAGER_OPEN ); + + assert( jrnlOpen==0 || ( sqlite3OsDeviceCharacteristics(pPager->jfd) & + SQLITE_IOCAP_UNDELETABLE_WHEN_OPEN + )); + + *pExists = 0; + if( !jrnlOpen ){ + rc = sqlite3OsAccess(pVfs, pPager->zJournal, SQLITE_ACCESS_EXISTS, &exists); + } + if( rc==SQLITE_OK && exists ){ + int locked = 0; /* True if some process holds a RESERVED lock */ + + /* Race condition here: Another process might have been holding the + ** the RESERVED lock and have a journal open at the sqlite3OsAccess() + ** call above, but then delete the journal and drop the lock before + ** we get to the following sqlite3OsCheckReservedLock() call. If that + ** is the case, this routine might think there is a hot journal when + ** in fact there is none. This results in a false-positive which will + ** be dealt with by the playback routine. Ticket #3883. + */ + rc = sqlite3OsCheckReservedLock(pPager->fd, &locked); + if( rc==SQLITE_OK && !locked ){ + Pgno nPage; /* Number of pages in database file */ + + assert( pPager->tempFile==0 ); + rc = pagerPagecount(pPager, &nPage); + if( rc==SQLITE_OK ){ + /* If the database is zero pages in size, that means that either (1) the + ** journal is a remnant from a prior database with the same name where + ** the database file but not the journal was deleted, or (2) the initial + ** transaction that populates a new database is being rolled back. + ** In either case, the journal file can be deleted. However, take care + ** not to delete the journal file if it is already open due to + ** journal_mode=PERSIST. + */ + if( nPage==0 && !jrnlOpen ){ + sqlite3BeginBenignMalloc(); + if( pagerLockDb(pPager, RESERVED_LOCK)==SQLITE_OK ){ + sqlite3OsDelete(pVfs, pPager->zJournal, 0); + if( !pPager->exclusiveMode ) pagerUnlockDb(pPager, SHARED_LOCK); + } + sqlite3EndBenignMalloc(); + }else{ + /* The journal file exists and no other connection has a reserved + ** or greater lock on the database file. Now check that there is + ** at least one non-zero bytes at the start of the journal file. + ** If there is, then we consider this journal to be hot. If not, + ** it can be ignored. + */ + if( !jrnlOpen ){ + int f = SQLITE_OPEN_READONLY|SQLITE_OPEN_MAIN_JOURNAL; + rc = sqlite3OsOpen(pVfs, pPager->zJournal, pPager->jfd, f, &f); + } + if( rc==SQLITE_OK ){ + u8 first = 0; + rc = sqlite3OsRead(pPager->jfd, (void *)&first, 1, 0); + if( rc==SQLITE_IOERR_SHORT_READ ){ + rc = SQLITE_OK; + } + if( !jrnlOpen ){ + sqlite3OsClose(pPager->jfd); + } + *pExists = (first!=0); + }else if( rc==SQLITE_CANTOPEN ){ + /* If we cannot open the rollback journal file in order to see if + ** it has a zero header, that might be due to an I/O error, or + ** it might be due to the race condition described above and in + ** ticket #3883. Either way, assume that the journal is hot. + ** This might be a false positive. But if it is, then the + ** automatic journal playback and recovery mechanism will deal + ** with it under an EXCLUSIVE lock where we do not need to + ** worry so much with race conditions. + */ + *pExists = 1; + rc = SQLITE_OK; + } + } + } + } + } + + return rc; +} + +/* +** This function is called to obtain a shared lock on the database file. +** It is illegal to call sqlite3PagerGet() until after this function +** has been successfully called. If a shared-lock is already held when +** this function is called, it is a no-op. +** +** The following operations are also performed by this function. +** +** 1) If the pager is currently in PAGER_OPEN state (no lock held +** on the database file), then an attempt is made to obtain a +** SHARED lock on the database file. Immediately after obtaining +** the SHARED lock, the file-system is checked for a hot-journal, +** which is played back if present. Following any hot-journal +** rollback, the contents of the cache are validated by checking +** the 'change-counter' field of the database file header and +** discarded if they are found to be invalid. +** +** 2) If the pager is running in exclusive-mode, and there are currently +** no outstanding references to any pages, and is in the error state, +** then an attempt is made to clear the error state by discarding +** the contents of the page cache and rolling back any open journal +** file. +** +** If everything is successful, SQLITE_OK is returned. If an IO error +** occurs while locking the database, checking for a hot-journal file or +** rolling back a journal file, the IO error code is returned. +*/ +SQLITE_PRIVATE int sqlite3PagerSharedLock(Pager *pPager){ + int rc = SQLITE_OK; /* Return code */ + + /* This routine is only called from b-tree and only when there are no + ** outstanding pages. This implies that the pager state should either + ** be OPEN or READER. READER is only possible if the pager is or was in + ** exclusive access mode. */ + assert( sqlite3PcacheRefCount(pPager->pPCache)==0 ); + assert( assert_pager_state(pPager) ); + assert( pPager->eState==PAGER_OPEN || pPager->eState==PAGER_READER ); + assert( pPager->errCode==SQLITE_OK ); + + if( !pagerUseWal(pPager) && pPager->eState==PAGER_OPEN ){ + int bHotJournal = 1; /* True if there exists a hot journal-file */ + + assert( !MEMDB ); + assert( pPager->tempFile==0 || pPager->eLock==EXCLUSIVE_LOCK ); + + rc = pager_wait_on_lock(pPager, SHARED_LOCK); + if( rc!=SQLITE_OK ){ + assert( pPager->eLock==NO_LOCK || pPager->eLock==UNKNOWN_LOCK ); + goto failed; + } + + /* If a journal file exists, and there is no RESERVED lock on the + ** database file, then it either needs to be played back or deleted. + */ + if( pPager->eLock<=SHARED_LOCK ){ + rc = hasHotJournal(pPager, &bHotJournal); + } + if( rc!=SQLITE_OK ){ + goto failed; + } + if( bHotJournal ){ + if( pPager->readOnly ){ + rc = SQLITE_READONLY_ROLLBACK; + goto failed; + } + + /* Get an EXCLUSIVE lock on the database file. At this point it is + ** important that a RESERVED lock is not obtained on the way to the + ** EXCLUSIVE lock. If it were, another process might open the + ** database file, detect the RESERVED lock, and conclude that the + ** database is safe to read while this process is still rolling the + ** hot-journal back. + ** + ** Because the intermediate RESERVED lock is not requested, any + ** other process attempting to access the database file will get to + ** this point in the code and fail to obtain its own EXCLUSIVE lock + ** on the database file. + ** + ** Unless the pager is in locking_mode=exclusive mode, the lock is + ** downgraded to SHARED_LOCK before this function returns. + */ + rc = pagerLockDb(pPager, EXCLUSIVE_LOCK); + if( rc!=SQLITE_OK ){ + goto failed; + } + + /* If it is not already open and the file exists on disk, open the + ** journal for read/write access. Write access is required because + ** in exclusive-access mode the file descriptor will be kept open + ** and possibly used for a transaction later on. Also, write-access + ** is usually required to finalize the journal in journal_mode=persist + ** mode (and also for journal_mode=truncate on some systems). + ** + ** If the journal does not exist, it usually means that some + ** other connection managed to get in and roll it back before + ** this connection obtained the exclusive lock above. Or, it + ** may mean that the pager was in the error-state when this + ** function was called and the journal file does not exist. + */ + if( !isOpen(pPager->jfd) ){ + sqlite3_vfs * const pVfs = pPager->pVfs; + int bExists; /* True if journal file exists */ + rc = sqlite3OsAccess( + pVfs, pPager->zJournal, SQLITE_ACCESS_EXISTS, &bExists); + if( rc==SQLITE_OK && bExists ){ + int fout = 0; + int f = SQLITE_OPEN_READWRITE|SQLITE_OPEN_MAIN_JOURNAL; + assert( !pPager->tempFile ); + rc = sqlite3OsOpen(pVfs, pPager->zJournal, pPager->jfd, f, &fout); + assert( rc!=SQLITE_OK || isOpen(pPager->jfd) ); + if( rc==SQLITE_OK && fout&SQLITE_OPEN_READONLY ){ + rc = SQLITE_CANTOPEN_BKPT; + sqlite3OsClose(pPager->jfd); + } + } + } + + /* Playback and delete the journal. Drop the database write + ** lock and reacquire the read lock. Purge the cache before + ** playing back the hot-journal so that we don't end up with + ** an inconsistent cache. Sync the hot journal before playing + ** it back since the process that crashed and left the hot journal + ** probably did not sync it and we are required to always sync + ** the journal before playing it back. + */ + if( isOpen(pPager->jfd) ){ + assert( rc==SQLITE_OK ); + rc = pagerSyncHotJournal(pPager); + if( rc==SQLITE_OK ){ + rc = pager_playback(pPager, !pPager->tempFile); + pPager->eState = PAGER_OPEN; + } + }else if( !pPager->exclusiveMode ){ + pagerUnlockDb(pPager, SHARED_LOCK); + } + + if( rc!=SQLITE_OK ){ + /* This branch is taken if an error occurs while trying to open + ** or roll back a hot-journal while holding an EXCLUSIVE lock. The + ** pager_unlock() routine will be called before returning to unlock + ** the file. If the unlock attempt fails, then Pager.eLock must be + ** set to UNKNOWN_LOCK (see the comment above the #define for + ** UNKNOWN_LOCK above for an explanation). + ** + ** In order to get pager_unlock() to do this, set Pager.eState to + ** PAGER_ERROR now. This is not actually counted as a transition + ** to ERROR state in the state diagram at the top of this file, + ** since we know that the same call to pager_unlock() will very + ** shortly transition the pager object to the OPEN state. Calling + ** assert_pager_state() would fail now, as it should not be possible + ** to be in ERROR state when there are zero outstanding page + ** references. + */ + pager_error(pPager, rc); + goto failed; + } + + assert( pPager->eState==PAGER_OPEN ); + assert( (pPager->eLock==SHARED_LOCK) + || (pPager->exclusiveMode && pPager->eLock>SHARED_LOCK) + ); + } + + if( !pPager->tempFile && pPager->hasHeldSharedLock ){ + /* The shared-lock has just been acquired then check to + ** see if the database has been modified. If the database has changed, + ** flush the cache. The hasHeldSharedLock flag prevents this from + ** occurring on the very first access to a file, in order to save a + ** single unnecessary sqlite3OsRead() call at the start-up. + ** + ** Database changes are detected by looking at 15 bytes beginning + ** at offset 24 into the file. The first 4 of these 16 bytes are + ** a 32-bit counter that is incremented with each change. The + ** other bytes change randomly with each file change when + ** a codec is in use. + ** + ** There is a vanishingly small chance that a change will not be + ** detected. The chance of an undetected change is so small that + ** it can be neglected. + */ + char dbFileVers[sizeof(pPager->dbFileVers)]; + + IOTRACE(("CKVERS %p %d\n", pPager, sizeof(dbFileVers))); + rc = sqlite3OsRead(pPager->fd, &dbFileVers, sizeof(dbFileVers), 24); + if( rc!=SQLITE_OK ){ + if( rc!=SQLITE_IOERR_SHORT_READ ){ + goto failed; + } + memset(dbFileVers, 0, sizeof(dbFileVers)); + } + + if( memcmp(pPager->dbFileVers, dbFileVers, sizeof(dbFileVers))!=0 ){ + pager_reset(pPager); + + /* Unmap the database file. It is possible that external processes + ** may have truncated the database file and then extended it back + ** to its original size while this process was not holding a lock. + ** In this case there may exist a Pager.pMap mapping that appears + ** to be the right size but is not actually valid. Avoid this + ** possibility by unmapping the db here. */ + if( USEFETCH(pPager) ){ + sqlite3OsUnfetch(pPager->fd, 0, 0); + } + } + } + + /* If there is a WAL file in the file-system, open this database in WAL + ** mode. Otherwise, the following function call is a no-op. + */ + rc = pagerOpenWalIfPresent(pPager); +#ifndef SQLITE_OMIT_WAL + assert( pPager->pWal==0 || rc==SQLITE_OK ); +#endif + } + + if( pagerUseWal(pPager) ){ + assert( rc==SQLITE_OK ); + rc = pagerBeginReadTransaction(pPager); + } + + if( pPager->tempFile==0 && pPager->eState==PAGER_OPEN && rc==SQLITE_OK ){ + rc = pagerPagecount(pPager, &pPager->dbSize); + } + + failed: + if( rc!=SQLITE_OK ){ + assert( !MEMDB ); + pager_unlock(pPager); + assert( pPager->eState==PAGER_OPEN ); + }else{ + pPager->eState = PAGER_READER; + pPager->hasHeldSharedLock = 1; + } + return rc; +} + +/* +** If the reference count has reached zero, rollback any active +** transaction and unlock the pager. +** +** Except, in locking_mode=EXCLUSIVE when there is nothing to in +** the rollback journal, the unlock is not performed and there is +** nothing to rollback, so this routine is a no-op. +*/ +static void pagerUnlockIfUnused(Pager *pPager){ + if( sqlite3PcacheRefCount(pPager->pPCache)==0 ){ + assert( pPager->nMmapOut==0 ); /* because page1 is never memory mapped */ + pagerUnlockAndRollback(pPager); + } +} + +/* +** The page getter methods each try to acquire a reference to a +** page with page number pgno. If the requested reference is +** successfully obtained, it is copied to *ppPage and SQLITE_OK returned. +** +** There are different implementations of the getter method depending +** on the current state of the pager. +** +** getPageNormal() -- The normal getter +** getPageError() -- Used if the pager is in an error state +** getPageMmap() -- Used if memory-mapped I/O is enabled +** +** If the requested page is already in the cache, it is returned. +** Otherwise, a new page object is allocated and populated with data +** read from the database file. In some cases, the pcache module may +** choose not to allocate a new page object and may reuse an existing +** object with no outstanding references. +** +** The extra data appended to a page is always initialized to zeros the +** first time a page is loaded into memory. If the page requested is +** already in the cache when this function is called, then the extra +** data is left as it was when the page object was last used. +** +** If the database image is smaller than the requested page or if +** the flags parameter contains the PAGER_GET_NOCONTENT bit and the +** requested page is not already stored in the cache, then no +** actual disk read occurs. In this case the memory image of the +** page is initialized to all zeros. +** +** If PAGER_GET_NOCONTENT is true, it means that we do not care about +** the contents of the page. This occurs in two scenarios: +** +** a) When reading a free-list leaf page from the database, and +** +** b) When a savepoint is being rolled back and we need to load +** a new page into the cache to be filled with the data read +** from the savepoint journal. +** +** If PAGER_GET_NOCONTENT is true, then the data returned is zeroed instead +** of being read from the database. Additionally, the bits corresponding +** to pgno in Pager.pInJournal (bitvec of pages already written to the +** journal file) and the PagerSavepoint.pInSavepoint bitvecs of any open +** savepoints are set. This means if the page is made writable at any +** point in the future, using a call to sqlite3PagerWrite(), its contents +** will not be journaled. This saves IO. +** +** The acquisition might fail for several reasons. In all cases, +** an appropriate error code is returned and *ppPage is set to NULL. +** +** See also sqlite3PagerLookup(). Both this routine and Lookup() attempt +** to find a page in the in-memory cache first. If the page is not already +** in memory, this routine goes to disk to read it in whereas Lookup() +** just returns 0. This routine acquires a read-lock the first time it +** has to go to disk, and could also playback an old journal if necessary. +** Since Lookup() never goes to disk, it never has to deal with locks +** or journal files. +*/ +static int getPageNormal( + Pager *pPager, /* The pager open on the database file */ + Pgno pgno, /* Page number to fetch */ + DbPage **ppPage, /* Write a pointer to the page here */ + int flags /* PAGER_GET_XXX flags */ +){ + int rc = SQLITE_OK; + PgHdr *pPg; + u8 noContent; /* True if PAGER_GET_NOCONTENT is set */ + sqlite3_pcache_page *pBase; + + assert( pPager->errCode==SQLITE_OK ); + assert( pPager->eState>=PAGER_READER ); + assert( assert_pager_state(pPager) ); + assert( pPager->hasHeldSharedLock==1 ); + + if( pgno==0 ) return SQLITE_CORRUPT_BKPT; + pBase = sqlite3PcacheFetch(pPager->pPCache, pgno, 3); + if( pBase==0 ){ + pPg = 0; + rc = sqlite3PcacheFetchStress(pPager->pPCache, pgno, &pBase); + if( rc!=SQLITE_OK ) goto pager_acquire_err; + if( pBase==0 ){ + rc = SQLITE_NOMEM_BKPT; + goto pager_acquire_err; + } + } + pPg = *ppPage = sqlite3PcacheFetchFinish(pPager->pPCache, pgno, pBase); + assert( pPg==(*ppPage) ); + assert( pPg->pgno==pgno ); + assert( pPg->pPager==pPager || pPg->pPager==0 ); + + noContent = (flags & PAGER_GET_NOCONTENT)!=0; + if( pPg->pPager && !noContent ){ + /* In this case the pcache already contains an initialized copy of + ** the page. Return without further ado. */ + assert( pgno<=PAGER_MAX_PGNO && pgno!=PAGER_MJ_PGNO(pPager) ); + pPager->aStat[PAGER_STAT_HIT]++; + return SQLITE_OK; + + }else{ + /* The pager cache has created a new page. Its content needs to + ** be initialized. But first some error checks: + ** + ** (1) The maximum page number is 2^31 + ** (2) Never try to fetch the locking page + */ + if( pgno>PAGER_MAX_PGNO || pgno==PAGER_MJ_PGNO(pPager) ){ + rc = SQLITE_CORRUPT_BKPT; + goto pager_acquire_err; + } + + pPg->pPager = pPager; + + assert( !isOpen(pPager->fd) || !MEMDB ); + if( !isOpen(pPager->fd) || pPager->dbSizepPager->mxPgno ){ + rc = SQLITE_FULL; + goto pager_acquire_err; + } + if( noContent ){ + /* Failure to set the bits in the InJournal bit-vectors is benign. + ** It merely means that we might do some extra work to journal a + ** page that does not need to be journaled. Nevertheless, be sure + ** to test the case where a malloc error occurs while trying to set + ** a bit in a bit vector. + */ + sqlite3BeginBenignMalloc(); + if( pgno<=pPager->dbOrigSize ){ + TESTONLY( rc = ) sqlite3BitvecSet(pPager->pInJournal, pgno); + testcase( rc==SQLITE_NOMEM ); + } + TESTONLY( rc = ) addToSavepointBitvecs(pPager, pgno); + testcase( rc==SQLITE_NOMEM ); + sqlite3EndBenignMalloc(); + } + memset(pPg->pData, 0, pPager->pageSize); + IOTRACE(("ZERO %p %d\n", pPager, pgno)); + }else{ + assert( pPg->pPager==pPager ); + pPager->aStat[PAGER_STAT_MISS]++; + rc = readDbPage(pPg); + if( rc!=SQLITE_OK ){ + goto pager_acquire_err; + } + } + pager_set_pagehash(pPg); + } + return SQLITE_OK; + +pager_acquire_err: + assert( rc!=SQLITE_OK ); + if( pPg ){ + sqlite3PcacheDrop(pPg); + } + pagerUnlockIfUnused(pPager); + *ppPage = 0; + return rc; +} + +#if SQLITE_MAX_MMAP_SIZE>0 +/* The page getter for when memory-mapped I/O is enabled */ +static int getPageMMap( + Pager *pPager, /* The pager open on the database file */ + Pgno pgno, /* Page number to fetch */ + DbPage **ppPage, /* Write a pointer to the page here */ + int flags /* PAGER_GET_XXX flags */ +){ + int rc = SQLITE_OK; + PgHdr *pPg = 0; + u32 iFrame = 0; /* Frame to read from WAL file */ + + /* It is acceptable to use a read-only (mmap) page for any page except + ** page 1 if there is no write-transaction open or the ACQUIRE_READONLY + ** flag was specified by the caller. And so long as the db is not a + ** temporary or in-memory database. */ + const int bMmapOk = (pgno>1 + && (pPager->eState==PAGER_READER || (flags & PAGER_GET_READONLY)) + ); + + assert( USEFETCH(pPager) ); +#ifdef SQLITE_HAS_CODEC + assert( pPager->xCodec==0 ); +#endif + + /* Optimization note: Adding the "pgno<=1" term before "pgno==0" here + ** allows the compiler optimizer to reuse the results of the "pgno>1" + ** test in the previous statement, and avoid testing pgno==0 in the + ** common case where pgno is large. */ + if( pgno<=1 && pgno==0 ){ + return SQLITE_CORRUPT_BKPT; + } + assert( pPager->eState>=PAGER_READER ); + assert( assert_pager_state(pPager) ); + assert( pPager->hasHeldSharedLock==1 ); + assert( pPager->errCode==SQLITE_OK ); + + if( bMmapOk && pagerUseWal(pPager) ){ + rc = sqlite3WalFindFrame(pPager->pWal, pgno, &iFrame); + if( rc!=SQLITE_OK ){ + *ppPage = 0; + return rc; + } + } + if( bMmapOk && iFrame==0 ){ + void *pData = 0; + rc = sqlite3OsFetch(pPager->fd, + (i64)(pgno-1) * pPager->pageSize, pPager->pageSize, &pData + ); + if( rc==SQLITE_OK && pData ){ + if( pPager->eState>PAGER_READER || pPager->tempFile ){ + pPg = sqlite3PagerLookup(pPager, pgno); + } + if( pPg==0 ){ + rc = pagerAcquireMapPage(pPager, pgno, pData, &pPg); + }else{ + sqlite3OsUnfetch(pPager->fd, (i64)(pgno-1)*pPager->pageSize, pData); + } + if( pPg ){ + assert( rc==SQLITE_OK ); + *ppPage = pPg; + return SQLITE_OK; + } + } + if( rc!=SQLITE_OK ){ + *ppPage = 0; + return rc; + } + } + return getPageNormal(pPager, pgno, ppPage, flags); +} +#endif /* SQLITE_MAX_MMAP_SIZE>0 */ + +/* The page getter method for when the pager is an error state */ +static int getPageError( + Pager *pPager, /* The pager open on the database file */ + Pgno pgno, /* Page number to fetch */ + DbPage **ppPage, /* Write a pointer to the page here */ + int flags /* PAGER_GET_XXX flags */ +){ + UNUSED_PARAMETER(pgno); + UNUSED_PARAMETER(flags); + assert( pPager->errCode!=SQLITE_OK ); + *ppPage = 0; + return pPager->errCode; +} + + +/* Dispatch all page fetch requests to the appropriate getter method. +*/ +SQLITE_PRIVATE int sqlite3PagerGet( + Pager *pPager, /* The pager open on the database file */ + Pgno pgno, /* Page number to fetch */ + DbPage **ppPage, /* Write a pointer to the page here */ + int flags /* PAGER_GET_XXX flags */ +){ + return pPager->xGet(pPager, pgno, ppPage, flags); +} + +/* +** Acquire a page if it is already in the in-memory cache. Do +** not read the page from disk. Return a pointer to the page, +** or 0 if the page is not in cache. +** +** See also sqlite3PagerGet(). The difference between this routine +** and sqlite3PagerGet() is that _get() will go to the disk and read +** in the page if the page is not already in cache. This routine +** returns NULL if the page is not in cache or if a disk I/O error +** has ever happened. +*/ +SQLITE_PRIVATE DbPage *sqlite3PagerLookup(Pager *pPager, Pgno pgno){ + sqlite3_pcache_page *pPage; + assert( pPager!=0 ); + assert( pgno!=0 ); + assert( pPager->pPCache!=0 ); + pPage = sqlite3PcacheFetch(pPager->pPCache, pgno, 0); + assert( pPage==0 || pPager->hasHeldSharedLock ); + if( pPage==0 ) return 0; + return sqlite3PcacheFetchFinish(pPager->pPCache, pgno, pPage); +} + +/* +** Release a page reference. +** +** The sqlite3PagerUnref() and sqlite3PagerUnrefNotNull() may only be +** used if we know that the page being released is not the last page. +** The btree layer always holds page1 open until the end, so these first +** to routines can be used to release any page other than BtShared.pPage1. +** +** Use sqlite3PagerUnrefPageOne() to release page1. This latter routine +** checks the total number of outstanding pages and if the number of +** pages reaches zero it drops the database lock. +*/ +SQLITE_PRIVATE void sqlite3PagerUnrefNotNull(DbPage *pPg){ + TESTONLY( Pager *pPager = pPg->pPager; ) + assert( pPg!=0 ); + if( pPg->flags & PGHDR_MMAP ){ + assert( pPg->pgno!=1 ); /* Page1 is never memory mapped */ + pagerReleaseMapPage(pPg); + }else{ + sqlite3PcacheRelease(pPg); + } + /* Do not use this routine to release the last reference to page1 */ + assert( sqlite3PcacheRefCount(pPager->pPCache)>0 ); +} +SQLITE_PRIVATE void sqlite3PagerUnref(DbPage *pPg){ + if( pPg ) sqlite3PagerUnrefNotNull(pPg); +} +SQLITE_PRIVATE void sqlite3PagerUnrefPageOne(DbPage *pPg){ + Pager *pPager; + assert( pPg!=0 ); + assert( pPg->pgno==1 ); + assert( (pPg->flags & PGHDR_MMAP)==0 ); /* Page1 is never memory mapped */ + pPager = pPg->pPager; + sqlite3PagerResetLockTimeout(pPager); + sqlite3PcacheRelease(pPg); + pagerUnlockIfUnused(pPager); +} + +/* +** This function is called at the start of every write transaction. +** There must already be a RESERVED or EXCLUSIVE lock on the database +** file when this routine is called. +** +** Open the journal file for pager pPager and write a journal header +** to the start of it. If there are active savepoints, open the sub-journal +** as well. This function is only used when the journal file is being +** opened to write a rollback log for a transaction. It is not used +** when opening a hot journal file to roll it back. +** +** If the journal file is already open (as it may be in exclusive mode), +** then this function just writes a journal header to the start of the +** already open file. +** +** Whether or not the journal file is opened by this function, the +** Pager.pInJournal bitvec structure is allocated. +** +** Return SQLITE_OK if everything is successful. Otherwise, return +** SQLITE_NOMEM if the attempt to allocate Pager.pInJournal fails, or +** an IO error code if opening or writing the journal file fails. +*/ +static int pager_open_journal(Pager *pPager){ + int rc = SQLITE_OK; /* Return code */ + sqlite3_vfs * const pVfs = pPager->pVfs; /* Local cache of vfs pointer */ + + assert( pPager->eState==PAGER_WRITER_LOCKED ); + assert( assert_pager_state(pPager) ); + assert( pPager->pInJournal==0 ); + + /* If already in the error state, this function is a no-op. But on + ** the other hand, this routine is never called if we are already in + ** an error state. */ + if( NEVER(pPager->errCode) ) return pPager->errCode; + + if( !pagerUseWal(pPager) && pPager->journalMode!=PAGER_JOURNALMODE_OFF ){ + pPager->pInJournal = sqlite3BitvecCreate(pPager->dbSize); + if( pPager->pInJournal==0 ){ + return SQLITE_NOMEM_BKPT; + } + + /* Open the journal file if it is not already open. */ + if( !isOpen(pPager->jfd) ){ + if( pPager->journalMode==PAGER_JOURNALMODE_MEMORY ){ + sqlite3MemJournalOpen(pPager->jfd); + }else{ + int flags = SQLITE_OPEN_READWRITE|SQLITE_OPEN_CREATE; + int nSpill; + + if( pPager->tempFile ){ + flags |= (SQLITE_OPEN_DELETEONCLOSE|SQLITE_OPEN_TEMP_JOURNAL); + nSpill = sqlite3Config.nStmtSpill; + }else{ + flags |= SQLITE_OPEN_MAIN_JOURNAL; + nSpill = jrnlBufferSize(pPager); + } + + /* Verify that the database still has the same name as it did when + ** it was originally opened. */ + rc = databaseIsUnmoved(pPager); + if( rc==SQLITE_OK ){ + rc = sqlite3JournalOpen ( + pVfs, pPager->zJournal, pPager->jfd, flags, nSpill + ); + } + } + assert( rc!=SQLITE_OK || isOpen(pPager->jfd) ); + } + + + /* Write the first journal header to the journal file and open + ** the sub-journal if necessary. + */ + if( rc==SQLITE_OK ){ + /* TODO: Check if all of these are really required. */ + pPager->nRec = 0; + pPager->journalOff = 0; + pPager->setMaster = 0; + pPager->journalHdr = 0; + rc = writeJournalHdr(pPager); + } + } + + if( rc!=SQLITE_OK ){ + sqlite3BitvecDestroy(pPager->pInJournal); + pPager->pInJournal = 0; + }else{ + assert( pPager->eState==PAGER_WRITER_LOCKED ); + pPager->eState = PAGER_WRITER_CACHEMOD; + } + + return rc; +} + +/* +** Begin a write-transaction on the specified pager object. If a +** write-transaction has already been opened, this function is a no-op. +** +** If the exFlag argument is false, then acquire at least a RESERVED +** lock on the database file. If exFlag is true, then acquire at least +** an EXCLUSIVE lock. If such a lock is already held, no locking +** functions need be called. +** +** If the subjInMemory argument is non-zero, then any sub-journal opened +** within this transaction will be opened as an in-memory file. This +** has no effect if the sub-journal is already opened (as it may be when +** running in exclusive mode) or if the transaction does not require a +** sub-journal. If the subjInMemory argument is zero, then any required +** sub-journal is implemented in-memory if pPager is an in-memory database, +** or using a temporary file otherwise. +*/ +SQLITE_PRIVATE int sqlite3PagerBegin(Pager *pPager, int exFlag, int subjInMemory){ + int rc = SQLITE_OK; + + if( pPager->errCode ) return pPager->errCode; + assert( pPager->eState>=PAGER_READER && pPager->eStatesubjInMemory = (u8)subjInMemory; + + if( ALWAYS(pPager->eState==PAGER_READER) ){ + assert( pPager->pInJournal==0 ); + + if( pagerUseWal(pPager) ){ + /* If the pager is configured to use locking_mode=exclusive, and an + ** exclusive lock on the database is not already held, obtain it now. + */ + if( pPager->exclusiveMode && sqlite3WalExclusiveMode(pPager->pWal, -1) ){ + rc = pagerLockDb(pPager, EXCLUSIVE_LOCK); + if( rc!=SQLITE_OK ){ + return rc; + } + (void)sqlite3WalExclusiveMode(pPager->pWal, 1); + } + + /* Grab the write lock on the log file. If successful, upgrade to + ** PAGER_RESERVED state. Otherwise, return an error code to the caller. + ** The busy-handler is not invoked if another connection already + ** holds the write-lock. If possible, the upper layer will call it. + */ + rc = sqlite3WalBeginWriteTransaction(pPager->pWal); + }else{ + /* Obtain a RESERVED lock on the database file. If the exFlag parameter + ** is true, then immediately upgrade this to an EXCLUSIVE lock. The + ** busy-handler callback can be used when upgrading to the EXCLUSIVE + ** lock, but not when obtaining the RESERVED lock. + */ + rc = pagerLockDb(pPager, RESERVED_LOCK); + if( rc==SQLITE_OK && exFlag ){ + rc = pager_wait_on_lock(pPager, EXCLUSIVE_LOCK); + } + } + + if( rc==SQLITE_OK ){ + /* Change to WRITER_LOCKED state. + ** + ** WAL mode sets Pager.eState to PAGER_WRITER_LOCKED or CACHEMOD + ** when it has an open transaction, but never to DBMOD or FINISHED. + ** This is because in those states the code to roll back savepoint + ** transactions may copy data from the sub-journal into the database + ** file as well as into the page cache. Which would be incorrect in + ** WAL mode. + */ + pPager->eState = PAGER_WRITER_LOCKED; + pPager->dbHintSize = pPager->dbSize; + pPager->dbFileSize = pPager->dbSize; + pPager->dbOrigSize = pPager->dbSize; + pPager->journalOff = 0; + } + + assert( rc==SQLITE_OK || pPager->eState==PAGER_READER ); + assert( rc!=SQLITE_OK || pPager->eState==PAGER_WRITER_LOCKED ); + assert( assert_pager_state(pPager) ); + } + + PAGERTRACE(("TRANSACTION %d\n", PAGERID(pPager))); + return rc; +} + +/* +** Write page pPg onto the end of the rollback journal. +*/ +static SQLITE_NOINLINE int pagerAddPageToRollbackJournal(PgHdr *pPg){ + Pager *pPager = pPg->pPager; + int rc; + u32 cksum; + char *pData2; + i64 iOff = pPager->journalOff; + + /* We should never write to the journal file the page that + ** contains the database locks. The following assert verifies + ** that we do not. */ + assert( pPg->pgno!=PAGER_MJ_PGNO(pPager) ); + + assert( pPager->journalHdr<=pPager->journalOff ); + CODEC2(pPager, pPg->pData, pPg->pgno, 7, return SQLITE_NOMEM_BKPT, pData2); + cksum = pager_cksum(pPager, (u8*)pData2); + + /* Even if an IO or diskfull error occurs while journalling the + ** page in the block above, set the need-sync flag for the page. + ** Otherwise, when the transaction is rolled back, the logic in + ** playback_one_page() will think that the page needs to be restored + ** in the database file. And if an IO error occurs while doing so, + ** then corruption may follow. + */ + pPg->flags |= PGHDR_NEED_SYNC; + + rc = write32bits(pPager->jfd, iOff, pPg->pgno); + if( rc!=SQLITE_OK ) return rc; + rc = sqlite3OsWrite(pPager->jfd, pData2, pPager->pageSize, iOff+4); + if( rc!=SQLITE_OK ) return rc; + rc = write32bits(pPager->jfd, iOff+pPager->pageSize+4, cksum); + if( rc!=SQLITE_OK ) return rc; + + IOTRACE(("JOUT %p %d %lld %d\n", pPager, pPg->pgno, + pPager->journalOff, pPager->pageSize)); + PAGER_INCR(sqlite3_pager_writej_count); + PAGERTRACE(("JOURNAL %d page %d needSync=%d hash(%08x)\n", + PAGERID(pPager), pPg->pgno, + ((pPg->flags&PGHDR_NEED_SYNC)?1:0), pager_pagehash(pPg))); + + pPager->journalOff += 8 + pPager->pageSize; + pPager->nRec++; + assert( pPager->pInJournal!=0 ); + rc = sqlite3BitvecSet(pPager->pInJournal, pPg->pgno); + testcase( rc==SQLITE_NOMEM ); + assert( rc==SQLITE_OK || rc==SQLITE_NOMEM ); + rc |= addToSavepointBitvecs(pPager, pPg->pgno); + assert( rc==SQLITE_OK || rc==SQLITE_NOMEM ); + return rc; +} + +/* +** Mark a single data page as writeable. The page is written into the +** main journal or sub-journal as required. If the page is written into +** one of the journals, the corresponding bit is set in the +** Pager.pInJournal bitvec and the PagerSavepoint.pInSavepoint bitvecs +** of any open savepoints as appropriate. +*/ +static int pager_write(PgHdr *pPg){ + Pager *pPager = pPg->pPager; + int rc = SQLITE_OK; + + /* This routine is not called unless a write-transaction has already + ** been started. The journal file may or may not be open at this point. + ** It is never called in the ERROR state. + */ + assert( pPager->eState==PAGER_WRITER_LOCKED + || pPager->eState==PAGER_WRITER_CACHEMOD + || pPager->eState==PAGER_WRITER_DBMOD + ); + assert( assert_pager_state(pPager) ); + assert( pPager->errCode==0 ); + assert( pPager->readOnly==0 ); + CHECK_PAGE(pPg); + + /* The journal file needs to be opened. Higher level routines have already + ** obtained the necessary locks to begin the write-transaction, but the + ** rollback journal might not yet be open. Open it now if this is the case. + ** + ** This is done before calling sqlite3PcacheMakeDirty() on the page. + ** Otherwise, if it were done after calling sqlite3PcacheMakeDirty(), then + ** an error might occur and the pager would end up in WRITER_LOCKED state + ** with pages marked as dirty in the cache. + */ + if( pPager->eState==PAGER_WRITER_LOCKED ){ + rc = pager_open_journal(pPager); + if( rc!=SQLITE_OK ) return rc; + } + assert( pPager->eState>=PAGER_WRITER_CACHEMOD ); + assert( assert_pager_state(pPager) ); + + /* Mark the page that is about to be modified as dirty. */ + sqlite3PcacheMakeDirty(pPg); + + /* If a rollback journal is in use, them make sure the page that is about + ** to change is in the rollback journal, or if the page is a new page off + ** then end of the file, make sure it is marked as PGHDR_NEED_SYNC. + */ + assert( (pPager->pInJournal!=0) == isOpen(pPager->jfd) ); + if( pPager->pInJournal!=0 + && sqlite3BitvecTestNotNull(pPager->pInJournal, pPg->pgno)==0 + ){ + assert( pagerUseWal(pPager)==0 ); + if( pPg->pgno<=pPager->dbOrigSize ){ + rc = pagerAddPageToRollbackJournal(pPg); + if( rc!=SQLITE_OK ){ + return rc; + } + }else{ + if( pPager->eState!=PAGER_WRITER_DBMOD ){ + pPg->flags |= PGHDR_NEED_SYNC; + } + PAGERTRACE(("APPEND %d page %d needSync=%d\n", + PAGERID(pPager), pPg->pgno, + ((pPg->flags&PGHDR_NEED_SYNC)?1:0))); + } + } + + /* The PGHDR_DIRTY bit is set above when the page was added to the dirty-list + ** and before writing the page into the rollback journal. Wait until now, + ** after the page has been successfully journalled, before setting the + ** PGHDR_WRITEABLE bit that indicates that the page can be safely modified. + */ + pPg->flags |= PGHDR_WRITEABLE; + + /* If the statement journal is open and the page is not in it, + ** then write the page into the statement journal. + */ + if( pPager->nSavepoint>0 ){ + rc = subjournalPageIfRequired(pPg); + } + + /* Update the database size and return. */ + if( pPager->dbSizepgno ){ + pPager->dbSize = pPg->pgno; + } + return rc; +} + +/* +** This is a variant of sqlite3PagerWrite() that runs when the sector size +** is larger than the page size. SQLite makes the (reasonable) assumption that +** all bytes of a sector are written together by hardware. Hence, all bytes of +** a sector need to be journalled in case of a power loss in the middle of +** a write. +** +** Usually, the sector size is less than or equal to the page size, in which +** case pages can be individually written. This routine only runs in the +** exceptional case where the page size is smaller than the sector size. +*/ +static SQLITE_NOINLINE int pagerWriteLargeSector(PgHdr *pPg){ + int rc = SQLITE_OK; /* Return code */ + Pgno nPageCount; /* Total number of pages in database file */ + Pgno pg1; /* First page of the sector pPg is located on. */ + int nPage = 0; /* Number of pages starting at pg1 to journal */ + int ii; /* Loop counter */ + int needSync = 0; /* True if any page has PGHDR_NEED_SYNC */ + Pager *pPager = pPg->pPager; /* The pager that owns pPg */ + Pgno nPagePerSector = (pPager->sectorSize/pPager->pageSize); + + /* Set the doNotSpill NOSYNC bit to 1. This is because we cannot allow + ** a journal header to be written between the pages journaled by + ** this function. + */ + assert( !MEMDB ); + assert( (pPager->doNotSpill & SPILLFLAG_NOSYNC)==0 ); + pPager->doNotSpill |= SPILLFLAG_NOSYNC; + + /* This trick assumes that both the page-size and sector-size are + ** an integer power of 2. It sets variable pg1 to the identifier + ** of the first page of the sector pPg is located on. + */ + pg1 = ((pPg->pgno-1) & ~(nPagePerSector-1)) + 1; + + nPageCount = pPager->dbSize; + if( pPg->pgno>nPageCount ){ + nPage = (pPg->pgno - pg1)+1; + }else if( (pg1+nPagePerSector-1)>nPageCount ){ + nPage = nPageCount+1-pg1; + }else{ + nPage = nPagePerSector; + } + assert(nPage>0); + assert(pg1<=pPg->pgno); + assert((pg1+nPage)>pPg->pgno); + + for(ii=0; iipgno || !sqlite3BitvecTest(pPager->pInJournal, pg) ){ + if( pg!=PAGER_MJ_PGNO(pPager) ){ + rc = sqlite3PagerGet(pPager, pg, &pPage, 0); + if( rc==SQLITE_OK ){ + rc = pager_write(pPage); + if( pPage->flags&PGHDR_NEED_SYNC ){ + needSync = 1; + } + sqlite3PagerUnrefNotNull(pPage); + } + } + }else if( (pPage = sqlite3PagerLookup(pPager, pg))!=0 ){ + if( pPage->flags&PGHDR_NEED_SYNC ){ + needSync = 1; + } + sqlite3PagerUnrefNotNull(pPage); + } + } + + /* If the PGHDR_NEED_SYNC flag is set for any of the nPage pages + ** starting at pg1, then it needs to be set for all of them. Because + ** writing to any of these nPage pages may damage the others, the + ** journal file must contain sync()ed copies of all of them + ** before any of them can be written out to the database file. + */ + if( rc==SQLITE_OK && needSync ){ + assert( !MEMDB ); + for(ii=0; iiflags |= PGHDR_NEED_SYNC; + sqlite3PagerUnrefNotNull(pPage); + } + } + } + + assert( (pPager->doNotSpill & SPILLFLAG_NOSYNC)!=0 ); + pPager->doNotSpill &= ~SPILLFLAG_NOSYNC; + return rc; +} + +/* +** Mark a data page as writeable. This routine must be called before +** making changes to a page. The caller must check the return value +** of this function and be careful not to change any page data unless +** this routine returns SQLITE_OK. +** +** The difference between this function and pager_write() is that this +** function also deals with the special case where 2 or more pages +** fit on a single disk sector. In this case all co-resident pages +** must have been written to the journal file before returning. +** +** If an error occurs, SQLITE_NOMEM or an IO error code is returned +** as appropriate. Otherwise, SQLITE_OK. +*/ +SQLITE_PRIVATE int sqlite3PagerWrite(PgHdr *pPg){ + Pager *pPager = pPg->pPager; + assert( (pPg->flags & PGHDR_MMAP)==0 ); + assert( pPager->eState>=PAGER_WRITER_LOCKED ); + assert( assert_pager_state(pPager) ); + if( (pPg->flags & PGHDR_WRITEABLE)!=0 && pPager->dbSize>=pPg->pgno ){ + if( pPager->nSavepoint ) return subjournalPageIfRequired(pPg); + return SQLITE_OK; + }else if( pPager->errCode ){ + return pPager->errCode; + }else if( pPager->sectorSize > (u32)pPager->pageSize ){ + assert( pPager->tempFile==0 ); + return pagerWriteLargeSector(pPg); + }else{ + return pager_write(pPg); + } +} + +/* +** Return TRUE if the page given in the argument was previously passed +** to sqlite3PagerWrite(). In other words, return TRUE if it is ok +** to change the content of the page. +*/ +#ifndef NDEBUG +SQLITE_PRIVATE int sqlite3PagerIswriteable(DbPage *pPg){ + return pPg->flags & PGHDR_WRITEABLE; +} +#endif + +/* +** A call to this routine tells the pager that it is not necessary to +** write the information on page pPg back to the disk, even though +** that page might be marked as dirty. This happens, for example, when +** the page has been added as a leaf of the freelist and so its +** content no longer matters. +** +** The overlying software layer calls this routine when all of the data +** on the given page is unused. The pager marks the page as clean so +** that it does not get written to disk. +** +** Tests show that this optimization can quadruple the speed of large +** DELETE operations. +** +** This optimization cannot be used with a temp-file, as the page may +** have been dirty at the start of the transaction. In that case, if +** memory pressure forces page pPg out of the cache, the data does need +** to be written out to disk so that it may be read back in if the +** current transaction is rolled back. +*/ +SQLITE_PRIVATE void sqlite3PagerDontWrite(PgHdr *pPg){ + Pager *pPager = pPg->pPager; + if( !pPager->tempFile && (pPg->flags&PGHDR_DIRTY) && pPager->nSavepoint==0 ){ + PAGERTRACE(("DONT_WRITE page %d of %d\n", pPg->pgno, PAGERID(pPager))); + IOTRACE(("CLEAN %p %d\n", pPager, pPg->pgno)) + pPg->flags |= PGHDR_DONT_WRITE; + pPg->flags &= ~PGHDR_WRITEABLE; + testcase( pPg->flags & PGHDR_NEED_SYNC ); + pager_set_pagehash(pPg); + } +} + +/* +** This routine is called to increment the value of the database file +** change-counter, stored as a 4-byte big-endian integer starting at +** byte offset 24 of the pager file. The secondary change counter at +** 92 is also updated, as is the SQLite version number at offset 96. +** +** But this only happens if the pPager->changeCountDone flag is false. +** To avoid excess churning of page 1, the update only happens once. +** See also the pager_write_changecounter() routine that does an +** unconditional update of the change counters. +** +** If the isDirectMode flag is zero, then this is done by calling +** sqlite3PagerWrite() on page 1, then modifying the contents of the +** page data. In this case the file will be updated when the current +** transaction is committed. +** +** The isDirectMode flag may only be non-zero if the library was compiled +** with the SQLITE_ENABLE_ATOMIC_WRITE macro defined. In this case, +** if isDirect is non-zero, then the database file is updated directly +** by writing an updated version of page 1 using a call to the +** sqlite3OsWrite() function. +*/ +static int pager_incr_changecounter(Pager *pPager, int isDirectMode){ + int rc = SQLITE_OK; + + assert( pPager->eState==PAGER_WRITER_CACHEMOD + || pPager->eState==PAGER_WRITER_DBMOD + ); + assert( assert_pager_state(pPager) ); + + /* Declare and initialize constant integer 'isDirect'. If the + ** atomic-write optimization is enabled in this build, then isDirect + ** is initialized to the value passed as the isDirectMode parameter + ** to this function. Otherwise, it is always set to zero. + ** + ** The idea is that if the atomic-write optimization is not + ** enabled at compile time, the compiler can omit the tests of + ** 'isDirect' below, as well as the block enclosed in the + ** "if( isDirect )" condition. + */ +#ifndef SQLITE_ENABLE_ATOMIC_WRITE +# define DIRECT_MODE 0 + assert( isDirectMode==0 ); + UNUSED_PARAMETER(isDirectMode); +#else +# define DIRECT_MODE isDirectMode +#endif + + if( !pPager->changeCountDone && ALWAYS(pPager->dbSize>0) ){ + PgHdr *pPgHdr; /* Reference to page 1 */ + + assert( !pPager->tempFile && isOpen(pPager->fd) ); + + /* Open page 1 of the file for writing. */ + rc = sqlite3PagerGet(pPager, 1, &pPgHdr, 0); + assert( pPgHdr==0 || rc==SQLITE_OK ); + + /* If page one was fetched successfully, and this function is not + ** operating in direct-mode, make page 1 writable. When not in + ** direct mode, page 1 is always held in cache and hence the PagerGet() + ** above is always successful - hence the ALWAYS on rc==SQLITE_OK. + */ + if( !DIRECT_MODE && ALWAYS(rc==SQLITE_OK) ){ + rc = sqlite3PagerWrite(pPgHdr); + } + + if( rc==SQLITE_OK ){ + /* Actually do the update of the change counter */ + pager_write_changecounter(pPgHdr); + + /* If running in direct mode, write the contents of page 1 to the file. */ + if( DIRECT_MODE ){ + const void *zBuf; + assert( pPager->dbFileSize>0 ); + CODEC2(pPager, pPgHdr->pData, 1, 6, rc=SQLITE_NOMEM_BKPT, zBuf); + if( rc==SQLITE_OK ){ + rc = sqlite3OsWrite(pPager->fd, zBuf, pPager->pageSize, 0); + pPager->aStat[PAGER_STAT_WRITE]++; + } + if( rc==SQLITE_OK ){ + /* Update the pager's copy of the change-counter. Otherwise, the + ** next time a read transaction is opened the cache will be + ** flushed (as the change-counter values will not match). */ + const void *pCopy = (const void *)&((const char *)zBuf)[24]; + memcpy(&pPager->dbFileVers, pCopy, sizeof(pPager->dbFileVers)); + pPager->changeCountDone = 1; + } + }else{ + pPager->changeCountDone = 1; + } + } + + /* Release the page reference. */ + sqlite3PagerUnref(pPgHdr); + } + return rc; +} + +/* +** Sync the database file to disk. This is a no-op for in-memory databases +** or pages with the Pager.noSync flag set. +** +** If successful, or if called on a pager for which it is a no-op, this +** function returns SQLITE_OK. Otherwise, an IO error code is returned. +*/ +SQLITE_PRIVATE int sqlite3PagerSync(Pager *pPager, const char *zMaster){ + int rc = SQLITE_OK; + void *pArg = (void*)zMaster; + rc = sqlite3OsFileControl(pPager->fd, SQLITE_FCNTL_SYNC, pArg); + if( rc==SQLITE_NOTFOUND ) rc = SQLITE_OK; + if( rc==SQLITE_OK && !pPager->noSync ){ + assert( !MEMDB ); + rc = sqlite3OsSync(pPager->fd, pPager->syncFlags); + } + return rc; +} + +/* +** This function may only be called while a write-transaction is active in +** rollback. If the connection is in WAL mode, this call is a no-op. +** Otherwise, if the connection does not already have an EXCLUSIVE lock on +** the database file, an attempt is made to obtain one. +** +** If the EXCLUSIVE lock is already held or the attempt to obtain it is +** successful, or the connection is in WAL mode, SQLITE_OK is returned. +** Otherwise, either SQLITE_BUSY or an SQLITE_IOERR_XXX error code is +** returned. +*/ +SQLITE_PRIVATE int sqlite3PagerExclusiveLock(Pager *pPager){ + int rc = pPager->errCode; + assert( assert_pager_state(pPager) ); + if( rc==SQLITE_OK ){ + assert( pPager->eState==PAGER_WRITER_CACHEMOD + || pPager->eState==PAGER_WRITER_DBMOD + || pPager->eState==PAGER_WRITER_LOCKED + ); + assert( assert_pager_state(pPager) ); + if( 0==pagerUseWal(pPager) ){ + rc = pager_wait_on_lock(pPager, EXCLUSIVE_LOCK); + } + } + return rc; +} + +/* +** Sync the database file for the pager pPager. zMaster points to the name +** of a master journal file that should be written into the individual +** journal file. zMaster may be NULL, which is interpreted as no master +** journal (a single database transaction). +** +** This routine ensures that: +** +** * The database file change-counter is updated, +** * the journal is synced (unless the atomic-write optimization is used), +** * all dirty pages are written to the database file, +** * the database file is truncated (if required), and +** * the database file synced. +** +** The only thing that remains to commit the transaction is to finalize +** (delete, truncate or zero the first part of) the journal file (or +** delete the master journal file if specified). +** +** Note that if zMaster==NULL, this does not overwrite a previous value +** passed to an sqlite3PagerCommitPhaseOne() call. +** +** If the final parameter - noSync - is true, then the database file itself +** is not synced. The caller must call sqlite3PagerSync() directly to +** sync the database file before calling CommitPhaseTwo() to delete the +** journal file in this case. +*/ +SQLITE_PRIVATE int sqlite3PagerCommitPhaseOne( + Pager *pPager, /* Pager object */ + const char *zMaster, /* If not NULL, the master journal name */ + int noSync /* True to omit the xSync on the db file */ +){ + int rc = SQLITE_OK; /* Return code */ + + assert( pPager->eState==PAGER_WRITER_LOCKED + || pPager->eState==PAGER_WRITER_CACHEMOD + || pPager->eState==PAGER_WRITER_DBMOD + || pPager->eState==PAGER_ERROR + ); + assert( assert_pager_state(pPager) ); + + /* If a prior error occurred, report that error again. */ + if( NEVER(pPager->errCode) ) return pPager->errCode; + + /* Provide the ability to easily simulate an I/O error during testing */ + if( sqlite3FaultSim(400) ) return SQLITE_IOERR; + + PAGERTRACE(("DATABASE SYNC: File=%s zMaster=%s nSize=%d\n", + pPager->zFilename, zMaster, pPager->dbSize)); + + /* If no database changes have been made, return early. */ + if( pPager->eStatetempFile ); + assert( isOpen(pPager->fd) || pPager->tempFile ); + if( 0==pagerFlushOnCommit(pPager, 1) ){ + /* If this is an in-memory db, or no pages have been written to, or this + ** function has already been called, it is mostly a no-op. However, any + ** backup in progress needs to be restarted. */ + sqlite3BackupRestart(pPager->pBackup); + }else{ + PgHdr *pList; + if( pagerUseWal(pPager) ){ + PgHdr *pPageOne = 0; + pList = sqlite3PcacheDirtyList(pPager->pPCache); + if( pList==0 ){ + /* Must have at least one page for the WAL commit flag. + ** Ticket [2d1a5c67dfc2363e44f29d9bbd57f] 2011-05-18 */ + rc = sqlite3PagerGet(pPager, 1, &pPageOne, 0); + pList = pPageOne; + pList->pDirty = 0; + } + assert( rc==SQLITE_OK ); + if( ALWAYS(pList) ){ + rc = pagerWalFrames(pPager, pList, pPager->dbSize, 1); + } + sqlite3PagerUnref(pPageOne); + if( rc==SQLITE_OK ){ + sqlite3PcacheCleanAll(pPager->pPCache); + } + }else{ + /* The bBatch boolean is true if the batch-atomic-write commit method + ** should be used. No rollback journal is created if batch-atomic-write + ** is enabled. + */ +#ifdef SQLITE_ENABLE_BATCH_ATOMIC_WRITE + sqlite3_file *fd = pPager->fd; + int bBatch = zMaster==0 /* An SQLITE_IOCAP_BATCH_ATOMIC commit */ + && (sqlite3OsDeviceCharacteristics(fd) & SQLITE_IOCAP_BATCH_ATOMIC) + && !pPager->noSync + && sqlite3JournalIsInMemory(pPager->jfd); +#else +# define bBatch 0 +#endif + +#ifdef SQLITE_ENABLE_ATOMIC_WRITE + /* The following block updates the change-counter. Exactly how it + ** does this depends on whether or not the atomic-update optimization + ** was enabled at compile time, and if this transaction meets the + ** runtime criteria to use the operation: + ** + ** * The file-system supports the atomic-write property for + ** blocks of size page-size, and + ** * This commit is not part of a multi-file transaction, and + ** * Exactly one page has been modified and store in the journal file. + ** + ** If the optimization was not enabled at compile time, then the + ** pager_incr_changecounter() function is called to update the change + ** counter in 'indirect-mode'. If the optimization is compiled in but + ** is not applicable to this transaction, call sqlite3JournalCreate() + ** to make sure the journal file has actually been created, then call + ** pager_incr_changecounter() to update the change-counter in indirect + ** mode. + ** + ** Otherwise, if the optimization is both enabled and applicable, + ** then call pager_incr_changecounter() to update the change-counter + ** in 'direct' mode. In this case the journal file will never be + ** created for this transaction. + */ + if( bBatch==0 ){ + PgHdr *pPg; + assert( isOpen(pPager->jfd) + || pPager->journalMode==PAGER_JOURNALMODE_OFF + || pPager->journalMode==PAGER_JOURNALMODE_WAL + ); + if( !zMaster && isOpen(pPager->jfd) + && pPager->journalOff==jrnlBufferSize(pPager) + && pPager->dbSize>=pPager->dbOrigSize + && (!(pPg = sqlite3PcacheDirtyList(pPager->pPCache)) || 0==pPg->pDirty) + ){ + /* Update the db file change counter via the direct-write method. The + ** following call will modify the in-memory representation of page 1 + ** to include the updated change counter and then write page 1 + ** directly to the database file. Because of the atomic-write + ** property of the host file-system, this is safe. + */ + rc = pager_incr_changecounter(pPager, 1); + }else{ + rc = sqlite3JournalCreate(pPager->jfd); + if( rc==SQLITE_OK ){ + rc = pager_incr_changecounter(pPager, 0); + } + } + } +#else /* SQLITE_ENABLE_ATOMIC_WRITE */ +#ifdef SQLITE_ENABLE_BATCH_ATOMIC_WRITE + if( zMaster ){ + rc = sqlite3JournalCreate(pPager->jfd); + if( rc!=SQLITE_OK ) goto commit_phase_one_exit; + assert( bBatch==0 ); + } +#endif + rc = pager_incr_changecounter(pPager, 0); +#endif /* !SQLITE_ENABLE_ATOMIC_WRITE */ + if( rc!=SQLITE_OK ) goto commit_phase_one_exit; + + /* Write the master journal name into the journal file. If a master + ** journal file name has already been written to the journal file, + ** or if zMaster is NULL (no master journal), then this call is a no-op. + */ + rc = writeMasterJournal(pPager, zMaster); + if( rc!=SQLITE_OK ) goto commit_phase_one_exit; + + /* Sync the journal file and write all dirty pages to the database. + ** If the atomic-update optimization is being used, this sync will not + ** create the journal file or perform any real IO. + ** + ** Because the change-counter page was just modified, unless the + ** atomic-update optimization is used it is almost certain that the + ** journal requires a sync here. However, in locking_mode=exclusive + ** on a system under memory pressure it is just possible that this is + ** not the case. In this case it is likely enough that the redundant + ** xSync() call will be changed to a no-op by the OS anyhow. + */ + rc = syncJournal(pPager, 0); + if( rc!=SQLITE_OK ) goto commit_phase_one_exit; + + pList = sqlite3PcacheDirtyList(pPager->pPCache); +#ifdef SQLITE_ENABLE_BATCH_ATOMIC_WRITE + if( bBatch ){ + rc = sqlite3OsFileControl(fd, SQLITE_FCNTL_BEGIN_ATOMIC_WRITE, 0); + if( rc==SQLITE_OK ){ + rc = pager_write_pagelist(pPager, pList); + if( rc==SQLITE_OK ){ + rc = sqlite3OsFileControl(fd, SQLITE_FCNTL_COMMIT_ATOMIC_WRITE, 0); + } + if( rc!=SQLITE_OK ){ + sqlite3OsFileControlHint(fd, SQLITE_FCNTL_ROLLBACK_ATOMIC_WRITE, 0); + } + } + + if( (rc&0xFF)==SQLITE_IOERR && rc!=SQLITE_IOERR_NOMEM ){ + rc = sqlite3JournalCreate(pPager->jfd); + if( rc!=SQLITE_OK ){ + sqlite3OsClose(pPager->jfd); + goto commit_phase_one_exit; + } + bBatch = 0; + }else{ + sqlite3OsClose(pPager->jfd); + } + } +#endif /* SQLITE_ENABLE_BATCH_ATOMIC_WRITE */ + + if( bBatch==0 ){ + rc = pager_write_pagelist(pPager, pList); + } + if( rc!=SQLITE_OK ){ + assert( rc!=SQLITE_IOERR_BLOCKED ); + goto commit_phase_one_exit; + } + sqlite3PcacheCleanAll(pPager->pPCache); + + /* If the file on disk is smaller than the database image, use + ** pager_truncate to grow the file here. This can happen if the database + ** image was extended as part of the current transaction and then the + ** last page in the db image moved to the free-list. In this case the + ** last page is never written out to disk, leaving the database file + ** undersized. Fix this now if it is the case. */ + if( pPager->dbSize>pPager->dbFileSize ){ + Pgno nNew = pPager->dbSize - (pPager->dbSize==PAGER_MJ_PGNO(pPager)); + assert( pPager->eState==PAGER_WRITER_DBMOD ); + rc = pager_truncate(pPager, nNew); + if( rc!=SQLITE_OK ) goto commit_phase_one_exit; + } + + /* Finally, sync the database file. */ + if( !noSync ){ + rc = sqlite3PagerSync(pPager, zMaster); + } + IOTRACE(("DBSYNC %p\n", pPager)) + } + } + +commit_phase_one_exit: + if( rc==SQLITE_OK && !pagerUseWal(pPager) ){ + pPager->eState = PAGER_WRITER_FINISHED; + } + return rc; +} + + +/* +** When this function is called, the database file has been completely +** updated to reflect the changes made by the current transaction and +** synced to disk. The journal file still exists in the file-system +** though, and if a failure occurs at this point it will eventually +** be used as a hot-journal and the current transaction rolled back. +** +** This function finalizes the journal file, either by deleting, +** truncating or partially zeroing it, so that it cannot be used +** for hot-journal rollback. Once this is done the transaction is +** irrevocably committed. +** +** If an error occurs, an IO error code is returned and the pager +** moves into the error state. Otherwise, SQLITE_OK is returned. +*/ +SQLITE_PRIVATE int sqlite3PagerCommitPhaseTwo(Pager *pPager){ + int rc = SQLITE_OK; /* Return code */ + + /* This routine should not be called if a prior error has occurred. + ** But if (due to a coding error elsewhere in the system) it does get + ** called, just return the same error code without doing anything. */ + if( NEVER(pPager->errCode) ) return pPager->errCode; + + assert( pPager->eState==PAGER_WRITER_LOCKED + || pPager->eState==PAGER_WRITER_FINISHED + || (pagerUseWal(pPager) && pPager->eState==PAGER_WRITER_CACHEMOD) + ); + assert( assert_pager_state(pPager) ); + + /* An optimization. If the database was not actually modified during + ** this transaction, the pager is running in exclusive-mode and is + ** using persistent journals, then this function is a no-op. + ** + ** The start of the journal file currently contains a single journal + ** header with the nRec field set to 0. If such a journal is used as + ** a hot-journal during hot-journal rollback, 0 changes will be made + ** to the database file. So there is no need to zero the journal + ** header. Since the pager is in exclusive mode, there is no need + ** to drop any locks either. + */ + if( pPager->eState==PAGER_WRITER_LOCKED + && pPager->exclusiveMode + && pPager->journalMode==PAGER_JOURNALMODE_PERSIST + ){ + assert( pPager->journalOff==JOURNAL_HDR_SZ(pPager) || !pPager->journalOff ); + pPager->eState = PAGER_READER; + return SQLITE_OK; + } + + PAGERTRACE(("COMMIT %d\n", PAGERID(pPager))); + pPager->iDataVersion++; + rc = pager_end_transaction(pPager, pPager->setMaster, 1); + return pager_error(pPager, rc); +} + +/* +** If a write transaction is open, then all changes made within the +** transaction are reverted and the current write-transaction is closed. +** The pager falls back to PAGER_READER state if successful, or PAGER_ERROR +** state if an error occurs. +** +** If the pager is already in PAGER_ERROR state when this function is called, +** it returns Pager.errCode immediately. No work is performed in this case. +** +** Otherwise, in rollback mode, this function performs two functions: +** +** 1) It rolls back the journal file, restoring all database file and +** in-memory cache pages to the state they were in when the transaction +** was opened, and +** +** 2) It finalizes the journal file, so that it is not used for hot +** rollback at any point in the future. +** +** Finalization of the journal file (task 2) is only performed if the +** rollback is successful. +** +** In WAL mode, all cache-entries containing data modified within the +** current transaction are either expelled from the cache or reverted to +** their pre-transaction state by re-reading data from the database or +** WAL files. The WAL transaction is then closed. +*/ +SQLITE_PRIVATE int sqlite3PagerRollback(Pager *pPager){ + int rc = SQLITE_OK; /* Return code */ + PAGERTRACE(("ROLLBACK %d\n", PAGERID(pPager))); + + /* PagerRollback() is a no-op if called in READER or OPEN state. If + ** the pager is already in the ERROR state, the rollback is not + ** attempted here. Instead, the error code is returned to the caller. + */ + assert( assert_pager_state(pPager) ); + if( pPager->eState==PAGER_ERROR ) return pPager->errCode; + if( pPager->eState<=PAGER_READER ) return SQLITE_OK; + + if( pagerUseWal(pPager) ){ + int rc2; + rc = sqlite3PagerSavepoint(pPager, SAVEPOINT_ROLLBACK, -1); + rc2 = pager_end_transaction(pPager, pPager->setMaster, 0); + if( rc==SQLITE_OK ) rc = rc2; + }else if( !isOpen(pPager->jfd) || pPager->eState==PAGER_WRITER_LOCKED ){ + int eState = pPager->eState; + rc = pager_end_transaction(pPager, 0, 0); + if( !MEMDB && eState>PAGER_WRITER_LOCKED ){ + /* This can happen using journal_mode=off. Move the pager to the error + ** state to indicate that the contents of the cache may not be trusted. + ** Any active readers will get SQLITE_ABORT. + */ + pPager->errCode = SQLITE_ABORT; + pPager->eState = PAGER_ERROR; + setGetterMethod(pPager); + return rc; + } + }else{ + rc = pager_playback(pPager, 0); + } + + assert( pPager->eState==PAGER_READER || rc!=SQLITE_OK ); + assert( rc==SQLITE_OK || rc==SQLITE_FULL || rc==SQLITE_CORRUPT + || rc==SQLITE_NOMEM || (rc&0xFF)==SQLITE_IOERR + || rc==SQLITE_CANTOPEN + ); + + /* If an error occurs during a ROLLBACK, we can no longer trust the pager + ** cache. So call pager_error() on the way out to make any error persistent. + */ + return pager_error(pPager, rc); +} + +/* +** Return TRUE if the database file is opened read-only. Return FALSE +** if the database is (in theory) writable. +*/ +SQLITE_PRIVATE u8 sqlite3PagerIsreadonly(Pager *pPager){ + return pPager->readOnly; +} + +#ifdef SQLITE_DEBUG +/* +** Return the sum of the reference counts for all pages held by pPager. +*/ +SQLITE_PRIVATE int sqlite3PagerRefcount(Pager *pPager){ + return sqlite3PcacheRefCount(pPager->pPCache); +} +#endif + +/* +** Return the approximate number of bytes of memory currently +** used by the pager and its associated cache. +*/ +SQLITE_PRIVATE int sqlite3PagerMemUsed(Pager *pPager){ + int perPageSize = pPager->pageSize + pPager->nExtra + sizeof(PgHdr) + + 5*sizeof(void*); + return perPageSize*sqlite3PcachePagecount(pPager->pPCache) + + sqlite3MallocSize(pPager) + + pPager->pageSize; +} + +/* +** Return the number of references to the specified page. +*/ +SQLITE_PRIVATE int sqlite3PagerPageRefcount(DbPage *pPage){ + return sqlite3PcachePageRefcount(pPage); +} + +#ifdef SQLITE_TEST +/* +** This routine is used for testing and analysis only. +*/ +SQLITE_PRIVATE int *sqlite3PagerStats(Pager *pPager){ + static int a[11]; + a[0] = sqlite3PcacheRefCount(pPager->pPCache); + a[1] = sqlite3PcachePagecount(pPager->pPCache); + a[2] = sqlite3PcacheGetCachesize(pPager->pPCache); + a[3] = pPager->eState==PAGER_OPEN ? -1 : (int) pPager->dbSize; + a[4] = pPager->eState; + a[5] = pPager->errCode; + a[6] = pPager->aStat[PAGER_STAT_HIT]; + a[7] = pPager->aStat[PAGER_STAT_MISS]; + a[8] = 0; /* Used to be pPager->nOvfl */ + a[9] = pPager->nRead; + a[10] = pPager->aStat[PAGER_STAT_WRITE]; + return a; +} +#endif + +/* +** Parameter eStat must be one of SQLITE_DBSTATUS_CACHE_HIT, _MISS, _WRITE, +** or _WRITE+1. The SQLITE_DBSTATUS_CACHE_WRITE+1 case is a translation +** of SQLITE_DBSTATUS_CACHE_SPILL. The _SPILL case is not contiguous because +** it was added later. +** +** Before returning, *pnVal is incremented by the +** current cache hit or miss count, according to the value of eStat. If the +** reset parameter is non-zero, the cache hit or miss count is zeroed before +** returning. +*/ +SQLITE_PRIVATE void sqlite3PagerCacheStat(Pager *pPager, int eStat, int reset, int *pnVal){ + + assert( eStat==SQLITE_DBSTATUS_CACHE_HIT + || eStat==SQLITE_DBSTATUS_CACHE_MISS + || eStat==SQLITE_DBSTATUS_CACHE_WRITE + || eStat==SQLITE_DBSTATUS_CACHE_WRITE+1 + ); + + assert( SQLITE_DBSTATUS_CACHE_HIT+1==SQLITE_DBSTATUS_CACHE_MISS ); + assert( SQLITE_DBSTATUS_CACHE_HIT+2==SQLITE_DBSTATUS_CACHE_WRITE ); + assert( PAGER_STAT_HIT==0 && PAGER_STAT_MISS==1 + && PAGER_STAT_WRITE==2 && PAGER_STAT_SPILL==3 ); + + eStat -= SQLITE_DBSTATUS_CACHE_HIT; + *pnVal += pPager->aStat[eStat]; + if( reset ){ + pPager->aStat[eStat] = 0; + } +} + +/* +** Return true if this is an in-memory or temp-file backed pager. +*/ +SQLITE_PRIVATE int sqlite3PagerIsMemdb(Pager *pPager){ + return pPager->tempFile; +} + +/* +** Check that there are at least nSavepoint savepoints open. If there are +** currently less than nSavepoints open, then open one or more savepoints +** to make up the difference. If the number of savepoints is already +** equal to nSavepoint, then this function is a no-op. +** +** If a memory allocation fails, SQLITE_NOMEM is returned. If an error +** occurs while opening the sub-journal file, then an IO error code is +** returned. Otherwise, SQLITE_OK. +*/ +static SQLITE_NOINLINE int pagerOpenSavepoint(Pager *pPager, int nSavepoint){ + int rc = SQLITE_OK; /* Return code */ + int nCurrent = pPager->nSavepoint; /* Current number of savepoints */ + int ii; /* Iterator variable */ + PagerSavepoint *aNew; /* New Pager.aSavepoint array */ + + assert( pPager->eState>=PAGER_WRITER_LOCKED ); + assert( assert_pager_state(pPager) ); + assert( nSavepoint>nCurrent && pPager->useJournal ); + + /* Grow the Pager.aSavepoint array using realloc(). Return SQLITE_NOMEM + ** if the allocation fails. Otherwise, zero the new portion in case a + ** malloc failure occurs while populating it in the for(...) loop below. + */ + aNew = (PagerSavepoint *)sqlite3Realloc( + pPager->aSavepoint, sizeof(PagerSavepoint)*nSavepoint + ); + if( !aNew ){ + return SQLITE_NOMEM_BKPT; + } + memset(&aNew[nCurrent], 0, (nSavepoint-nCurrent) * sizeof(PagerSavepoint)); + pPager->aSavepoint = aNew; + + /* Populate the PagerSavepoint structures just allocated. */ + for(ii=nCurrent; iidbSize; + if( isOpen(pPager->jfd) && pPager->journalOff>0 ){ + aNew[ii].iOffset = pPager->journalOff; + }else{ + aNew[ii].iOffset = JOURNAL_HDR_SZ(pPager); + } + aNew[ii].iSubRec = pPager->nSubRec; + aNew[ii].pInSavepoint = sqlite3BitvecCreate(pPager->dbSize); + if( !aNew[ii].pInSavepoint ){ + return SQLITE_NOMEM_BKPT; + } + if( pagerUseWal(pPager) ){ + sqlite3WalSavepoint(pPager->pWal, aNew[ii].aWalData); + } + pPager->nSavepoint = ii+1; + } + assert( pPager->nSavepoint==nSavepoint ); + assertTruncateConstraint(pPager); + return rc; +} +SQLITE_PRIVATE int sqlite3PagerOpenSavepoint(Pager *pPager, int nSavepoint){ + assert( pPager->eState>=PAGER_WRITER_LOCKED ); + assert( assert_pager_state(pPager) ); + + if( nSavepoint>pPager->nSavepoint && pPager->useJournal ){ + return pagerOpenSavepoint(pPager, nSavepoint); + }else{ + return SQLITE_OK; + } +} + + +/* +** This function is called to rollback or release (commit) a savepoint. +** The savepoint to release or rollback need not be the most recently +** created savepoint. +** +** Parameter op is always either SAVEPOINT_ROLLBACK or SAVEPOINT_RELEASE. +** If it is SAVEPOINT_RELEASE, then release and destroy the savepoint with +** index iSavepoint. If it is SAVEPOINT_ROLLBACK, then rollback all changes +** that have occurred since the specified savepoint was created. +** +** The savepoint to rollback or release is identified by parameter +** iSavepoint. A value of 0 means to operate on the outermost savepoint +** (the first created). A value of (Pager.nSavepoint-1) means operate +** on the most recently created savepoint. If iSavepoint is greater than +** (Pager.nSavepoint-1), then this function is a no-op. +** +** If a negative value is passed to this function, then the current +** transaction is rolled back. This is different to calling +** sqlite3PagerRollback() because this function does not terminate +** the transaction or unlock the database, it just restores the +** contents of the database to its original state. +** +** In any case, all savepoints with an index greater than iSavepoint +** are destroyed. If this is a release operation (op==SAVEPOINT_RELEASE), +** then savepoint iSavepoint is also destroyed. +** +** This function may return SQLITE_NOMEM if a memory allocation fails, +** or an IO error code if an IO error occurs while rolling back a +** savepoint. If no errors occur, SQLITE_OK is returned. +*/ +SQLITE_PRIVATE int sqlite3PagerSavepoint(Pager *pPager, int op, int iSavepoint){ + int rc = pPager->errCode; + +#ifdef SQLITE_ENABLE_ZIPVFS + if( op==SAVEPOINT_RELEASE ) rc = SQLITE_OK; +#endif + + assert( op==SAVEPOINT_RELEASE || op==SAVEPOINT_ROLLBACK ); + assert( iSavepoint>=0 || op==SAVEPOINT_ROLLBACK ); + + if( rc==SQLITE_OK && iSavepointnSavepoint ){ + int ii; /* Iterator variable */ + int nNew; /* Number of remaining savepoints after this op. */ + + /* Figure out how many savepoints will still be active after this + ** operation. Store this value in nNew. Then free resources associated + ** with any savepoints that are destroyed by this operation. + */ + nNew = iSavepoint + (( op==SAVEPOINT_RELEASE ) ? 0 : 1); + for(ii=nNew; iinSavepoint; ii++){ + sqlite3BitvecDestroy(pPager->aSavepoint[ii].pInSavepoint); + } + pPager->nSavepoint = nNew; + + /* If this is a release of the outermost savepoint, truncate + ** the sub-journal to zero bytes in size. */ + if( op==SAVEPOINT_RELEASE ){ + if( nNew==0 && isOpen(pPager->sjfd) ){ + /* Only truncate if it is an in-memory sub-journal. */ + if( sqlite3JournalIsInMemory(pPager->sjfd) ){ + rc = sqlite3OsTruncate(pPager->sjfd, 0); + assert( rc==SQLITE_OK ); + } + pPager->nSubRec = 0; + } + } + /* Else this is a rollback operation, playback the specified savepoint. + ** If this is a temp-file, it is possible that the journal file has + ** not yet been opened. In this case there have been no changes to + ** the database file, so the playback operation can be skipped. + */ + else if( pagerUseWal(pPager) || isOpen(pPager->jfd) ){ + PagerSavepoint *pSavepoint = (nNew==0)?0:&pPager->aSavepoint[nNew-1]; + rc = pagerPlaybackSavepoint(pPager, pSavepoint); + assert(rc!=SQLITE_DONE); + } + +#ifdef SQLITE_ENABLE_ZIPVFS + /* If the cache has been modified but the savepoint cannot be rolled + ** back journal_mode=off, put the pager in the error state. This way, + ** if the VFS used by this pager includes ZipVFS, the entire transaction + ** can be rolled back at the ZipVFS level. */ + else if( + pPager->journalMode==PAGER_JOURNALMODE_OFF + && pPager->eState>=PAGER_WRITER_CACHEMOD + ){ + pPager->errCode = SQLITE_ABORT; + pPager->eState = PAGER_ERROR; + setGetterMethod(pPager); + } +#endif + } + + return rc; +} + +/* +** Return the full pathname of the database file. +** +** Except, if the pager is in-memory only, then return an empty string if +** nullIfMemDb is true. This routine is called with nullIfMemDb==1 when +** used to report the filename to the user, for compatibility with legacy +** behavior. But when the Btree needs to know the filename for matching to +** shared cache, it uses nullIfMemDb==0 so that in-memory databases can +** participate in shared-cache. +*/ +SQLITE_PRIVATE const char *sqlite3PagerFilename(Pager *pPager, int nullIfMemDb){ + return (nullIfMemDb && pPager->memDb) ? "" : pPager->zFilename; +} + +/* +** Return the VFS structure for the pager. +*/ +SQLITE_PRIVATE sqlite3_vfs *sqlite3PagerVfs(Pager *pPager){ + return pPager->pVfs; +} + +/* +** Return the file handle for the database file associated +** with the pager. This might return NULL if the file has +** not yet been opened. +*/ +SQLITE_PRIVATE sqlite3_file *sqlite3PagerFile(Pager *pPager){ + return pPager->fd; +} + +#ifdef SQLITE_ENABLE_SETLK_TIMEOUT +/* +** Reset the lock timeout for pager. +*/ +SQLITE_PRIVATE void sqlite3PagerResetLockTimeout(Pager *pPager){ + int x = 0; + sqlite3OsFileControl(pPager->fd, SQLITE_FCNTL_LOCK_TIMEOUT, &x); +} +#endif + +/* +** Return the file handle for the journal file (if it exists). +** This will be either the rollback journal or the WAL file. +*/ +SQLITE_PRIVATE sqlite3_file *sqlite3PagerJrnlFile(Pager *pPager){ +#if SQLITE_OMIT_WAL + return pPager->jfd; +#else + return pPager->pWal ? sqlite3WalFile(pPager->pWal) : pPager->jfd; +#endif +} + +/* +** Return the full pathname of the journal file. +*/ +SQLITE_PRIVATE const char *sqlite3PagerJournalname(Pager *pPager){ + return pPager->zJournal; +} + +#ifdef SQLITE_HAS_CODEC +/* +** Set or retrieve the codec for this pager +*/ +SQLITE_PRIVATE void sqlite3PagerSetCodec( + Pager *pPager, + void *(*xCodec)(void*,void*,Pgno,int), + void (*xCodecSizeChng)(void*,int,int), + void (*xCodecFree)(void*), + void *pCodec +){ + if( pPager->xCodecFree ){ + pPager->xCodecFree(pPager->pCodec); + }else{ + pager_reset(pPager); + } + pPager->xCodec = pPager->memDb ? 0 : xCodec; + pPager->xCodecSizeChng = xCodecSizeChng; + pPager->xCodecFree = xCodecFree; + pPager->pCodec = pCodec; + setGetterMethod(pPager); + pagerReportSize(pPager); +} +SQLITE_PRIVATE void *sqlite3PagerGetCodec(Pager *pPager){ + return pPager->pCodec; +} + +/* +** This function is called by the wal module when writing page content +** into the log file. +** +** This function returns a pointer to a buffer containing the encrypted +** page content. If a malloc fails, this function may return NULL. +*/ +SQLITE_PRIVATE void *sqlite3PagerCodec(PgHdr *pPg){ + void *aData = 0; + CODEC2(pPg->pPager, pPg->pData, pPg->pgno, 6, return 0, aData); + return aData; +} + +/* +** Return the current pager state +*/ +SQLITE_PRIVATE int sqlite3PagerState(Pager *pPager){ + return pPager->eState; +} +#endif /* SQLITE_HAS_CODEC */ + +#ifndef SQLITE_OMIT_AUTOVACUUM +/* +** Move the page pPg to location pgno in the file. +** +** There must be no references to the page previously located at +** pgno (which we call pPgOld) though that page is allowed to be +** in cache. If the page previously located at pgno is not already +** in the rollback journal, it is not put there by by this routine. +** +** References to the page pPg remain valid. Updating any +** meta-data associated with pPg (i.e. data stored in the nExtra bytes +** allocated along with the page) is the responsibility of the caller. +** +** A transaction must be active when this routine is called. It used to be +** required that a statement transaction was not active, but this restriction +** has been removed (CREATE INDEX needs to move a page when a statement +** transaction is active). +** +** If the fourth argument, isCommit, is non-zero, then this page is being +** moved as part of a database reorganization just before the transaction +** is being committed. In this case, it is guaranteed that the database page +** pPg refers to will not be written to again within this transaction. +** +** This function may return SQLITE_NOMEM or an IO error code if an error +** occurs. Otherwise, it returns SQLITE_OK. +*/ +SQLITE_PRIVATE int sqlite3PagerMovepage(Pager *pPager, DbPage *pPg, Pgno pgno, int isCommit){ + PgHdr *pPgOld; /* The page being overwritten. */ + Pgno needSyncPgno = 0; /* Old value of pPg->pgno, if sync is required */ + int rc; /* Return code */ + Pgno origPgno; /* The original page number */ + + assert( pPg->nRef>0 ); + assert( pPager->eState==PAGER_WRITER_CACHEMOD + || pPager->eState==PAGER_WRITER_DBMOD + ); + assert( assert_pager_state(pPager) ); + + /* In order to be able to rollback, an in-memory database must journal + ** the page we are moving from. + */ + assert( pPager->tempFile || !MEMDB ); + if( pPager->tempFile ){ + rc = sqlite3PagerWrite(pPg); + if( rc ) return rc; + } + + /* If the page being moved is dirty and has not been saved by the latest + ** savepoint, then save the current contents of the page into the + ** sub-journal now. This is required to handle the following scenario: + ** + ** BEGIN; + ** + ** SAVEPOINT one; + ** + ** ROLLBACK TO one; + ** + ** If page X were not written to the sub-journal here, it would not + ** be possible to restore its contents when the "ROLLBACK TO one" + ** statement were is processed. + ** + ** subjournalPage() may need to allocate space to store pPg->pgno into + ** one or more savepoint bitvecs. This is the reason this function + ** may return SQLITE_NOMEM. + */ + if( (pPg->flags & PGHDR_DIRTY)!=0 + && SQLITE_OK!=(rc = subjournalPageIfRequired(pPg)) + ){ + return rc; + } + + PAGERTRACE(("MOVE %d page %d (needSync=%d) moves to %d\n", + PAGERID(pPager), pPg->pgno, (pPg->flags&PGHDR_NEED_SYNC)?1:0, pgno)); + IOTRACE(("MOVE %p %d %d\n", pPager, pPg->pgno, pgno)) + + /* If the journal needs to be sync()ed before page pPg->pgno can + ** be written to, store pPg->pgno in local variable needSyncPgno. + ** + ** If the isCommit flag is set, there is no need to remember that + ** the journal needs to be sync()ed before database page pPg->pgno + ** can be written to. The caller has already promised not to write to it. + */ + if( (pPg->flags&PGHDR_NEED_SYNC) && !isCommit ){ + needSyncPgno = pPg->pgno; + assert( pPager->journalMode==PAGER_JOURNALMODE_OFF || + pageInJournal(pPager, pPg) || pPg->pgno>pPager->dbOrigSize ); + assert( pPg->flags&PGHDR_DIRTY ); + } + + /* If the cache contains a page with page-number pgno, remove it + ** from its hash chain. Also, if the PGHDR_NEED_SYNC flag was set for + ** page pgno before the 'move' operation, it needs to be retained + ** for the page moved there. + */ + pPg->flags &= ~PGHDR_NEED_SYNC; + pPgOld = sqlite3PagerLookup(pPager, pgno); + assert( !pPgOld || pPgOld->nRef==1 || CORRUPT_DB ); + if( pPgOld ){ + if( pPgOld->nRef>1 ){ + sqlite3PagerUnrefNotNull(pPgOld); + return SQLITE_CORRUPT_BKPT; + } + pPg->flags |= (pPgOld->flags&PGHDR_NEED_SYNC); + if( pPager->tempFile ){ + /* Do not discard pages from an in-memory database since we might + ** need to rollback later. Just move the page out of the way. */ + sqlite3PcacheMove(pPgOld, pPager->dbSize+1); + }else{ + sqlite3PcacheDrop(pPgOld); + } + } + + origPgno = pPg->pgno; + sqlite3PcacheMove(pPg, pgno); + sqlite3PcacheMakeDirty(pPg); + + /* For an in-memory database, make sure the original page continues + ** to exist, in case the transaction needs to roll back. Use pPgOld + ** as the original page since it has already been allocated. + */ + if( pPager->tempFile && pPgOld ){ + sqlite3PcacheMove(pPgOld, origPgno); + sqlite3PagerUnrefNotNull(pPgOld); + } + + if( needSyncPgno ){ + /* If needSyncPgno is non-zero, then the journal file needs to be + ** sync()ed before any data is written to database file page needSyncPgno. + ** Currently, no such page exists in the page-cache and the + ** "is journaled" bitvec flag has been set. This needs to be remedied by + ** loading the page into the pager-cache and setting the PGHDR_NEED_SYNC + ** flag. + ** + ** If the attempt to load the page into the page-cache fails, (due + ** to a malloc() or IO failure), clear the bit in the pInJournal[] + ** array. Otherwise, if the page is loaded and written again in + ** this transaction, it may be written to the database file before + ** it is synced into the journal file. This way, it may end up in + ** the journal file twice, but that is not a problem. + */ + PgHdr *pPgHdr; + rc = sqlite3PagerGet(pPager, needSyncPgno, &pPgHdr, 0); + if( rc!=SQLITE_OK ){ + if( needSyncPgno<=pPager->dbOrigSize ){ + assert( pPager->pTmpSpace!=0 ); + sqlite3BitvecClear(pPager->pInJournal, needSyncPgno, pPager->pTmpSpace); + } + return rc; + } + pPgHdr->flags |= PGHDR_NEED_SYNC; + sqlite3PcacheMakeDirty(pPgHdr); + sqlite3PagerUnrefNotNull(pPgHdr); + } + + return SQLITE_OK; +} +#endif + +/* +** The page handle passed as the first argument refers to a dirty page +** with a page number other than iNew. This function changes the page's +** page number to iNew and sets the value of the PgHdr.flags field to +** the value passed as the third parameter. +*/ +SQLITE_PRIVATE void sqlite3PagerRekey(DbPage *pPg, Pgno iNew, u16 flags){ + assert( pPg->pgno!=iNew ); + pPg->flags = flags; + sqlite3PcacheMove(pPg, iNew); +} + +/* +** Return a pointer to the data for the specified page. +*/ +SQLITE_PRIVATE void *sqlite3PagerGetData(DbPage *pPg){ + assert( pPg->nRef>0 || pPg->pPager->memDb ); + return pPg->pData; +} + +/* +** Return a pointer to the Pager.nExtra bytes of "extra" space +** allocated along with the specified page. +*/ +SQLITE_PRIVATE void *sqlite3PagerGetExtra(DbPage *pPg){ + return pPg->pExtra; +} + +/* +** Get/set the locking-mode for this pager. Parameter eMode must be one +** of PAGER_LOCKINGMODE_QUERY, PAGER_LOCKINGMODE_NORMAL or +** PAGER_LOCKINGMODE_EXCLUSIVE. If the parameter is not _QUERY, then +** the locking-mode is set to the value specified. +** +** The returned value is either PAGER_LOCKINGMODE_NORMAL or +** PAGER_LOCKINGMODE_EXCLUSIVE, indicating the current (possibly updated) +** locking-mode. +*/ +SQLITE_PRIVATE int sqlite3PagerLockingMode(Pager *pPager, int eMode){ + assert( eMode==PAGER_LOCKINGMODE_QUERY + || eMode==PAGER_LOCKINGMODE_NORMAL + || eMode==PAGER_LOCKINGMODE_EXCLUSIVE ); + assert( PAGER_LOCKINGMODE_QUERY<0 ); + assert( PAGER_LOCKINGMODE_NORMAL>=0 && PAGER_LOCKINGMODE_EXCLUSIVE>=0 ); + assert( pPager->exclusiveMode || 0==sqlite3WalHeapMemory(pPager->pWal) ); + if( eMode>=0 && !pPager->tempFile && !sqlite3WalHeapMemory(pPager->pWal) ){ + pPager->exclusiveMode = (u8)eMode; + } + return (int)pPager->exclusiveMode; +} + +/* +** Set the journal-mode for this pager. Parameter eMode must be one of: +** +** PAGER_JOURNALMODE_DELETE +** PAGER_JOURNALMODE_TRUNCATE +** PAGER_JOURNALMODE_PERSIST +** PAGER_JOURNALMODE_OFF +** PAGER_JOURNALMODE_MEMORY +** PAGER_JOURNALMODE_WAL +** +** The journalmode is set to the value specified if the change is allowed. +** The change may be disallowed for the following reasons: +** +** * An in-memory database can only have its journal_mode set to _OFF +** or _MEMORY. +** +** * Temporary databases cannot have _WAL journalmode. +** +** The returned indicate the current (possibly updated) journal-mode. +*/ +SQLITE_PRIVATE int sqlite3PagerSetJournalMode(Pager *pPager, int eMode){ + u8 eOld = pPager->journalMode; /* Prior journalmode */ + + /* The eMode parameter is always valid */ + assert( eMode==PAGER_JOURNALMODE_DELETE + || eMode==PAGER_JOURNALMODE_TRUNCATE + || eMode==PAGER_JOURNALMODE_PERSIST + || eMode==PAGER_JOURNALMODE_OFF + || eMode==PAGER_JOURNALMODE_WAL + || eMode==PAGER_JOURNALMODE_MEMORY ); + + /* This routine is only called from the OP_JournalMode opcode, and + ** the logic there will never allow a temporary file to be changed + ** to WAL mode. + */ + assert( pPager->tempFile==0 || eMode!=PAGER_JOURNALMODE_WAL ); + + /* Do allow the journalmode of an in-memory database to be set to + ** anything other than MEMORY or OFF + */ + if( MEMDB ){ + assert( eOld==PAGER_JOURNALMODE_MEMORY || eOld==PAGER_JOURNALMODE_OFF ); + if( eMode!=PAGER_JOURNALMODE_MEMORY && eMode!=PAGER_JOURNALMODE_OFF ){ + eMode = eOld; + } + } + + if( eMode!=eOld ){ + + /* Change the journal mode. */ + assert( pPager->eState!=PAGER_ERROR ); + pPager->journalMode = (u8)eMode; + + /* When transistioning from TRUNCATE or PERSIST to any other journal + ** mode except WAL, unless the pager is in locking_mode=exclusive mode, + ** delete the journal file. + */ + assert( (PAGER_JOURNALMODE_TRUNCATE & 5)==1 ); + assert( (PAGER_JOURNALMODE_PERSIST & 5)==1 ); + assert( (PAGER_JOURNALMODE_DELETE & 5)==0 ); + assert( (PAGER_JOURNALMODE_MEMORY & 5)==4 ); + assert( (PAGER_JOURNALMODE_OFF & 5)==0 ); + assert( (PAGER_JOURNALMODE_WAL & 5)==5 ); + + assert( isOpen(pPager->fd) || pPager->exclusiveMode ); + if( !pPager->exclusiveMode && (eOld & 5)==1 && (eMode & 1)==0 ){ + + /* In this case we would like to delete the journal file. If it is + ** not possible, then that is not a problem. Deleting the journal file + ** here is an optimization only. + ** + ** Before deleting the journal file, obtain a RESERVED lock on the + ** database file. This ensures that the journal file is not deleted + ** while it is in use by some other client. + */ + sqlite3OsClose(pPager->jfd); + if( pPager->eLock>=RESERVED_LOCK ){ + sqlite3OsDelete(pPager->pVfs, pPager->zJournal, 0); + }else{ + int rc = SQLITE_OK; + int state = pPager->eState; + assert( state==PAGER_OPEN || state==PAGER_READER ); + if( state==PAGER_OPEN ){ + rc = sqlite3PagerSharedLock(pPager); + } + if( pPager->eState==PAGER_READER ){ + assert( rc==SQLITE_OK ); + rc = pagerLockDb(pPager, RESERVED_LOCK); + } + if( rc==SQLITE_OK ){ + sqlite3OsDelete(pPager->pVfs, pPager->zJournal, 0); + } + if( rc==SQLITE_OK && state==PAGER_READER ){ + pagerUnlockDb(pPager, SHARED_LOCK); + }else if( state==PAGER_OPEN ){ + pager_unlock(pPager); + } + assert( state==pPager->eState ); + } + }else if( eMode==PAGER_JOURNALMODE_OFF ){ + sqlite3OsClose(pPager->jfd); + } + } + + /* Return the new journal mode */ + return (int)pPager->journalMode; +} + +/* +** Return the current journal mode. +*/ +SQLITE_PRIVATE int sqlite3PagerGetJournalMode(Pager *pPager){ + return (int)pPager->journalMode; +} + +/* +** Return TRUE if the pager is in a state where it is OK to change the +** journalmode. Journalmode changes can only happen when the database +** is unmodified. +*/ +SQLITE_PRIVATE int sqlite3PagerOkToChangeJournalMode(Pager *pPager){ + assert( assert_pager_state(pPager) ); + if( pPager->eState>=PAGER_WRITER_CACHEMOD ) return 0; + if( NEVER(isOpen(pPager->jfd) && pPager->journalOff>0) ) return 0; + return 1; +} + +/* +** Get/set the size-limit used for persistent journal files. +** +** Setting the size limit to -1 means no limit is enforced. +** An attempt to set a limit smaller than -1 is a no-op. +*/ +SQLITE_PRIVATE i64 sqlite3PagerJournalSizeLimit(Pager *pPager, i64 iLimit){ + if( iLimit>=-1 ){ + pPager->journalSizeLimit = iLimit; + sqlite3WalLimit(pPager->pWal, iLimit); + } + return pPager->journalSizeLimit; +} + +/* +** Return a pointer to the pPager->pBackup variable. The backup module +** in backup.c maintains the content of this variable. This module +** uses it opaquely as an argument to sqlite3BackupRestart() and +** sqlite3BackupUpdate() only. +*/ +SQLITE_PRIVATE sqlite3_backup **sqlite3PagerBackupPtr(Pager *pPager){ + return &pPager->pBackup; +} + +#ifndef SQLITE_OMIT_VACUUM +/* +** Unless this is an in-memory or temporary database, clear the pager cache. +*/ +SQLITE_PRIVATE void sqlite3PagerClearCache(Pager *pPager){ + assert( MEMDB==0 || pPager->tempFile ); + if( pPager->tempFile==0 ) pager_reset(pPager); +} +#endif + + +#ifndef SQLITE_OMIT_WAL +/* +** This function is called when the user invokes "PRAGMA wal_checkpoint", +** "PRAGMA wal_blocking_checkpoint" or calls the sqlite3_wal_checkpoint() +** or wal_blocking_checkpoint() API functions. +** +** Parameter eMode is one of SQLITE_CHECKPOINT_PASSIVE, FULL or RESTART. +*/ +SQLITE_PRIVATE int sqlite3PagerCheckpoint( + Pager *pPager, /* Checkpoint on this pager */ + sqlite3 *db, /* Db handle used to check for interrupts */ + int eMode, /* Type of checkpoint */ + int *pnLog, /* OUT: Final number of frames in log */ + int *pnCkpt /* OUT: Final number of checkpointed frames */ +){ + int rc = SQLITE_OK; + if( pPager->pWal ){ + rc = sqlite3WalCheckpoint(pPager->pWal, db, eMode, + (eMode==SQLITE_CHECKPOINT_PASSIVE ? 0 : pPager->xBusyHandler), + pPager->pBusyHandlerArg, + pPager->walSyncFlags, pPager->pageSize, (u8 *)pPager->pTmpSpace, + pnLog, pnCkpt + ); + sqlite3PagerResetLockTimeout(pPager); + } + return rc; +} + +SQLITE_PRIVATE int sqlite3PagerWalCallback(Pager *pPager){ + return sqlite3WalCallback(pPager->pWal); +} + +/* +** Return true if the underlying VFS for the given pager supports the +** primitives necessary for write-ahead logging. +*/ +SQLITE_PRIVATE int sqlite3PagerWalSupported(Pager *pPager){ + const sqlite3_io_methods *pMethods = pPager->fd->pMethods; + if( pPager->noLock ) return 0; + return pPager->exclusiveMode || (pMethods->iVersion>=2 && pMethods->xShmMap); +} + +/* +** Attempt to take an exclusive lock on the database file. If a PENDING lock +** is obtained instead, immediately release it. +*/ +static int pagerExclusiveLock(Pager *pPager){ + int rc; /* Return code */ + + assert( pPager->eLock==SHARED_LOCK || pPager->eLock==EXCLUSIVE_LOCK ); + rc = pagerLockDb(pPager, EXCLUSIVE_LOCK); + if( rc!=SQLITE_OK ){ + /* If the attempt to grab the exclusive lock failed, release the + ** pending lock that may have been obtained instead. */ + pagerUnlockDb(pPager, SHARED_LOCK); + } + + return rc; +} + +/* +** Call sqlite3WalOpen() to open the WAL handle. If the pager is in +** exclusive-locking mode when this function is called, take an EXCLUSIVE +** lock on the database file and use heap-memory to store the wal-index +** in. Otherwise, use the normal shared-memory. +*/ +static int pagerOpenWal(Pager *pPager){ + int rc = SQLITE_OK; + + assert( pPager->pWal==0 && pPager->tempFile==0 ); + assert( pPager->eLock==SHARED_LOCK || pPager->eLock==EXCLUSIVE_LOCK ); + + /* If the pager is already in exclusive-mode, the WAL module will use + ** heap-memory for the wal-index instead of the VFS shared-memory + ** implementation. Take the exclusive lock now, before opening the WAL + ** file, to make sure this is safe. + */ + if( pPager->exclusiveMode ){ + rc = pagerExclusiveLock(pPager); + } + + /* Open the connection to the log file. If this operation fails, + ** (e.g. due to malloc() failure), return an error code. + */ + if( rc==SQLITE_OK ){ + rc = sqlite3WalOpen(pPager->pVfs, + pPager->fd, pPager->zWal, pPager->exclusiveMode, + pPager->journalSizeLimit, &pPager->pWal + ); + } + pagerFixMaplimit(pPager); + + return rc; +} + + +/* +** The caller must be holding a SHARED lock on the database file to call +** this function. +** +** If the pager passed as the first argument is open on a real database +** file (not a temp file or an in-memory database), and the WAL file +** is not already open, make an attempt to open it now. If successful, +** return SQLITE_OK. If an error occurs or the VFS used by the pager does +** not support the xShmXXX() methods, return an error code. *pbOpen is +** not modified in either case. +** +** If the pager is open on a temp-file (or in-memory database), or if +** the WAL file is already open, set *pbOpen to 1 and return SQLITE_OK +** without doing anything. +*/ +SQLITE_PRIVATE int sqlite3PagerOpenWal( + Pager *pPager, /* Pager object */ + int *pbOpen /* OUT: Set to true if call is a no-op */ +){ + int rc = SQLITE_OK; /* Return code */ + + assert( assert_pager_state(pPager) ); + assert( pPager->eState==PAGER_OPEN || pbOpen ); + assert( pPager->eState==PAGER_READER || !pbOpen ); + assert( pbOpen==0 || *pbOpen==0 ); + assert( pbOpen!=0 || (!pPager->tempFile && !pPager->pWal) ); + + if( !pPager->tempFile && !pPager->pWal ){ + if( !sqlite3PagerWalSupported(pPager) ) return SQLITE_CANTOPEN; + + /* Close any rollback journal previously open */ + sqlite3OsClose(pPager->jfd); + + rc = pagerOpenWal(pPager); + if( rc==SQLITE_OK ){ + pPager->journalMode = PAGER_JOURNALMODE_WAL; + pPager->eState = PAGER_OPEN; + } + }else{ + *pbOpen = 1; + } + + return rc; +} + +/* +** This function is called to close the connection to the log file prior +** to switching from WAL to rollback mode. +** +** Before closing the log file, this function attempts to take an +** EXCLUSIVE lock on the database file. If this cannot be obtained, an +** error (SQLITE_BUSY) is returned and the log connection is not closed. +** If successful, the EXCLUSIVE lock is not released before returning. +*/ +SQLITE_PRIVATE int sqlite3PagerCloseWal(Pager *pPager, sqlite3 *db){ + int rc = SQLITE_OK; + + assert( pPager->journalMode==PAGER_JOURNALMODE_WAL ); + + /* If the log file is not already open, but does exist in the file-system, + ** it may need to be checkpointed before the connection can switch to + ** rollback mode. Open it now so this can happen. + */ + if( !pPager->pWal ){ + int logexists = 0; + rc = pagerLockDb(pPager, SHARED_LOCK); + if( rc==SQLITE_OK ){ + rc = sqlite3OsAccess( + pPager->pVfs, pPager->zWal, SQLITE_ACCESS_EXISTS, &logexists + ); + } + if( rc==SQLITE_OK && logexists ){ + rc = pagerOpenWal(pPager); + } + } + + /* Checkpoint and close the log. Because an EXCLUSIVE lock is held on + ** the database file, the log and log-summary files will be deleted. + */ + if( rc==SQLITE_OK && pPager->pWal ){ + rc = pagerExclusiveLock(pPager); + if( rc==SQLITE_OK ){ + rc = sqlite3WalClose(pPager->pWal, db, pPager->walSyncFlags, + pPager->pageSize, (u8*)pPager->pTmpSpace); + pPager->pWal = 0; + pagerFixMaplimit(pPager); + if( rc && !pPager->exclusiveMode ) pagerUnlockDb(pPager, SHARED_LOCK); + } + } + return rc; +} + +#ifdef SQLITE_ENABLE_SNAPSHOT +/* +** If this is a WAL database, obtain a snapshot handle for the snapshot +** currently open. Otherwise, return an error. +*/ +SQLITE_PRIVATE int sqlite3PagerSnapshotGet(Pager *pPager, sqlite3_snapshot **ppSnapshot){ + int rc = SQLITE_ERROR; + if( pPager->pWal ){ + rc = sqlite3WalSnapshotGet(pPager->pWal, ppSnapshot); + } + return rc; +} + +/* +** If this is a WAL database, store a pointer to pSnapshot. Next time a +** read transaction is opened, attempt to read from the snapshot it +** identifies. If this is not a WAL database, return an error. +*/ +SQLITE_PRIVATE int sqlite3PagerSnapshotOpen(Pager *pPager, sqlite3_snapshot *pSnapshot){ + int rc = SQLITE_OK; + if( pPager->pWal ){ + sqlite3WalSnapshotOpen(pPager->pWal, pSnapshot); + }else{ + rc = SQLITE_ERROR; + } + return rc; +} + +/* +** If this is a WAL database, call sqlite3WalSnapshotRecover(). If this +** is not a WAL database, return an error. +*/ +SQLITE_PRIVATE int sqlite3PagerSnapshotRecover(Pager *pPager){ + int rc; + if( pPager->pWal ){ + rc = sqlite3WalSnapshotRecover(pPager->pWal); + }else{ + rc = SQLITE_ERROR; + } + return rc; +} + +/* +** The caller currently has a read transaction open on the database. +** If this is not a WAL database, SQLITE_ERROR is returned. Otherwise, +** this function takes a SHARED lock on the CHECKPOINTER slot and then +** checks if the snapshot passed as the second argument is still +** available. If so, SQLITE_OK is returned. +** +** If the snapshot is not available, SQLITE_ERROR is returned. Or, if +** the CHECKPOINTER lock cannot be obtained, SQLITE_BUSY. If any error +** occurs (any value other than SQLITE_OK is returned), the CHECKPOINTER +** lock is released before returning. +*/ +SQLITE_PRIVATE int sqlite3PagerSnapshotCheck(Pager *pPager, sqlite3_snapshot *pSnapshot){ + int rc; + if( pPager->pWal ){ + rc = sqlite3WalSnapshotCheck(pPager->pWal, pSnapshot); + }else{ + rc = SQLITE_ERROR; + } + return rc; +} + +/* +** Release a lock obtained by an earlier successful call to +** sqlite3PagerSnapshotCheck(). +*/ +SQLITE_PRIVATE void sqlite3PagerSnapshotUnlock(Pager *pPager){ + assert( pPager->pWal ); + sqlite3WalSnapshotUnlock(pPager->pWal); +} + +#endif /* SQLITE_ENABLE_SNAPSHOT */ +#endif /* !SQLITE_OMIT_WAL */ + +#ifdef SQLITE_ENABLE_ZIPVFS +/* +** A read-lock must be held on the pager when this function is called. If +** the pager is in WAL mode and the WAL file currently contains one or more +** frames, return the size in bytes of the page images stored within the +** WAL frames. Otherwise, if this is not a WAL database or the WAL file +** is empty, return 0. +*/ +SQLITE_PRIVATE int sqlite3PagerWalFramesize(Pager *pPager){ + assert( pPager->eState>=PAGER_READER ); + return sqlite3WalFramesize(pPager->pWal); +} +#endif + +#endif /* SQLITE_OMIT_DISKIO */ + +/************** End of pager.c ***********************************************/ +/************** Begin file wal.c *********************************************/ +/* +** 2010 February 1 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** +** This file contains the implementation of a write-ahead log (WAL) used in +** "journal_mode=WAL" mode. +** +** WRITE-AHEAD LOG (WAL) FILE FORMAT +** +** A WAL file consists of a header followed by zero or more "frames". +** Each frame records the revised content of a single page from the +** database file. All changes to the database are recorded by writing +** frames into the WAL. Transactions commit when a frame is written that +** contains a commit marker. A single WAL can and usually does record +** multiple transactions. Periodically, the content of the WAL is +** transferred back into the database file in an operation called a +** "checkpoint". +** +** A single WAL file can be used multiple times. In other words, the +** WAL can fill up with frames and then be checkpointed and then new +** frames can overwrite the old ones. A WAL always grows from beginning +** toward the end. Checksums and counters attached to each frame are +** used to determine which frames within the WAL are valid and which +** are leftovers from prior checkpoints. +** +** The WAL header is 32 bytes in size and consists of the following eight +** big-endian 32-bit unsigned integer values: +** +** 0: Magic number. 0x377f0682 or 0x377f0683 +** 4: File format version. Currently 3007000 +** 8: Database page size. Example: 1024 +** 12: Checkpoint sequence number +** 16: Salt-1, random integer incremented with each checkpoint +** 20: Salt-2, a different random integer changing with each ckpt +** 24: Checksum-1 (first part of checksum for first 24 bytes of header). +** 28: Checksum-2 (second part of checksum for first 24 bytes of header). +** +** Immediately following the wal-header are zero or more frames. Each +** frame consists of a 24-byte frame-header followed by a bytes +** of page data. The frame-header is six big-endian 32-bit unsigned +** integer values, as follows: +** +** 0: Page number. +** 4: For commit records, the size of the database image in pages +** after the commit. For all other records, zero. +** 8: Salt-1 (copied from the header) +** 12: Salt-2 (copied from the header) +** 16: Checksum-1. +** 20: Checksum-2. +** +** A frame is considered valid if and only if the following conditions are +** true: +** +** (1) The salt-1 and salt-2 values in the frame-header match +** salt values in the wal-header +** +** (2) The checksum values in the final 8 bytes of the frame-header +** exactly match the checksum computed consecutively on the +** WAL header and the first 8 bytes and the content of all frames +** up to and including the current frame. +** +** The checksum is computed using 32-bit big-endian integers if the +** magic number in the first 4 bytes of the WAL is 0x377f0683 and it +** is computed using little-endian if the magic number is 0x377f0682. +** The checksum values are always stored in the frame header in a +** big-endian format regardless of which byte order is used to compute +** the checksum. The checksum is computed by interpreting the input as +** an even number of unsigned 32-bit integers: x[0] through x[N]. The +** algorithm used for the checksum is as follows: +** +** for i from 0 to n-1 step 2: +** s0 += x[i] + s1; +** s1 += x[i+1] + s0; +** endfor +** +** Note that s0 and s1 are both weighted checksums using fibonacci weights +** in reverse order (the largest fibonacci weight occurs on the first element +** of the sequence being summed.) The s1 value spans all 32-bit +** terms of the sequence whereas s0 omits the final term. +** +** On a checkpoint, the WAL is first VFS.xSync-ed, then valid content of the +** WAL is transferred into the database, then the database is VFS.xSync-ed. +** The VFS.xSync operations serve as write barriers - all writes launched +** before the xSync must complete before any write that launches after the +** xSync begins. +** +** After each checkpoint, the salt-1 value is incremented and the salt-2 +** value is randomized. This prevents old and new frames in the WAL from +** being considered valid at the same time and being checkpointing together +** following a crash. +** +** READER ALGORITHM +** +** To read a page from the database (call it page number P), a reader +** first checks the WAL to see if it contains page P. If so, then the +** last valid instance of page P that is a followed by a commit frame +** or is a commit frame itself becomes the value read. If the WAL +** contains no copies of page P that are valid and which are a commit +** frame or are followed by a commit frame, then page P is read from +** the database file. +** +** To start a read transaction, the reader records the index of the last +** valid frame in the WAL. The reader uses this recorded "mxFrame" value +** for all subsequent read operations. New transactions can be appended +** to the WAL, but as long as the reader uses its original mxFrame value +** and ignores the newly appended content, it will see a consistent snapshot +** of the database from a single point in time. This technique allows +** multiple concurrent readers to view different versions of the database +** content simultaneously. +** +** The reader algorithm in the previous paragraphs works correctly, but +** because frames for page P can appear anywhere within the WAL, the +** reader has to scan the entire WAL looking for page P frames. If the +** WAL is large (multiple megabytes is typical) that scan can be slow, +** and read performance suffers. To overcome this problem, a separate +** data structure called the wal-index is maintained to expedite the +** search for frames of a particular page. +** +** WAL-INDEX FORMAT +** +** Conceptually, the wal-index is shared memory, though VFS implementations +** might choose to implement the wal-index using a mmapped file. Because +** the wal-index is shared memory, SQLite does not support journal_mode=WAL +** on a network filesystem. All users of the database must be able to +** share memory. +** +** In the default unix and windows implementation, the wal-index is a mmapped +** file whose name is the database name with a "-shm" suffix added. For that +** reason, the wal-index is sometimes called the "shm" file. +** +** The wal-index is transient. After a crash, the wal-index can (and should +** be) reconstructed from the original WAL file. In fact, the VFS is required +** to either truncate or zero the header of the wal-index when the last +** connection to it closes. Because the wal-index is transient, it can +** use an architecture-specific format; it does not have to be cross-platform. +** Hence, unlike the database and WAL file formats which store all values +** as big endian, the wal-index can store multi-byte values in the native +** byte order of the host computer. +** +** The purpose of the wal-index is to answer this question quickly: Given +** a page number P and a maximum frame index M, return the index of the +** last frame in the wal before frame M for page P in the WAL, or return +** NULL if there are no frames for page P in the WAL prior to M. +** +** The wal-index consists of a header region, followed by an one or +** more index blocks. +** +** The wal-index header contains the total number of frames within the WAL +** in the mxFrame field. +** +** Each index block except for the first contains information on +** HASHTABLE_NPAGE frames. The first index block contains information on +** HASHTABLE_NPAGE_ONE frames. The values of HASHTABLE_NPAGE_ONE and +** HASHTABLE_NPAGE are selected so that together the wal-index header and +** first index block are the same size as all other index blocks in the +** wal-index. +** +** Each index block contains two sections, a page-mapping that contains the +** database page number associated with each wal frame, and a hash-table +** that allows readers to query an index block for a specific page number. +** The page-mapping is an array of HASHTABLE_NPAGE (or HASHTABLE_NPAGE_ONE +** for the first index block) 32-bit page numbers. The first entry in the +** first index-block contains the database page number corresponding to the +** first frame in the WAL file. The first entry in the second index block +** in the WAL file corresponds to the (HASHTABLE_NPAGE_ONE+1)th frame in +** the log, and so on. +** +** The last index block in a wal-index usually contains less than the full +** complement of HASHTABLE_NPAGE (or HASHTABLE_NPAGE_ONE) page-numbers, +** depending on the contents of the WAL file. This does not change the +** allocated size of the page-mapping array - the page-mapping array merely +** contains unused entries. +** +** Even without using the hash table, the last frame for page P +** can be found by scanning the page-mapping sections of each index block +** starting with the last index block and moving toward the first, and +** within each index block, starting at the end and moving toward the +** beginning. The first entry that equals P corresponds to the frame +** holding the content for that page. +** +** The hash table consists of HASHTABLE_NSLOT 16-bit unsigned integers. +** HASHTABLE_NSLOT = 2*HASHTABLE_NPAGE, and there is one entry in the +** hash table for each page number in the mapping section, so the hash +** table is never more than half full. The expected number of collisions +** prior to finding a match is 1. Each entry of the hash table is an +** 1-based index of an entry in the mapping section of the same +** index block. Let K be the 1-based index of the largest entry in +** the mapping section. (For index blocks other than the last, K will +** always be exactly HASHTABLE_NPAGE (4096) and for the last index block +** K will be (mxFrame%HASHTABLE_NPAGE).) Unused slots of the hash table +** contain a value of 0. +** +** To look for page P in the hash table, first compute a hash iKey on +** P as follows: +** +** iKey = (P * 383) % HASHTABLE_NSLOT +** +** Then start scanning entries of the hash table, starting with iKey +** (wrapping around to the beginning when the end of the hash table is +** reached) until an unused hash slot is found. Let the first unused slot +** be at index iUnused. (iUnused might be less than iKey if there was +** wrap-around.) Because the hash table is never more than half full, +** the search is guaranteed to eventually hit an unused entry. Let +** iMax be the value between iKey and iUnused, closest to iUnused, +** where aHash[iMax]==P. If there is no iMax entry (if there exists +** no hash slot such that aHash[i]==p) then page P is not in the +** current index block. Otherwise the iMax-th mapping entry of the +** current index block corresponds to the last entry that references +** page P. +** +** A hash search begins with the last index block and moves toward the +** first index block, looking for entries corresponding to page P. On +** average, only two or three slots in each index block need to be +** examined in order to either find the last entry for page P, or to +** establish that no such entry exists in the block. Each index block +** holds over 4000 entries. So two or three index blocks are sufficient +** to cover a typical 10 megabyte WAL file, assuming 1K pages. 8 or 10 +** comparisons (on average) suffice to either locate a frame in the +** WAL or to establish that the frame does not exist in the WAL. This +** is much faster than scanning the entire 10MB WAL. +** +** Note that entries are added in order of increasing K. Hence, one +** reader might be using some value K0 and a second reader that started +** at a later time (after additional transactions were added to the WAL +** and to the wal-index) might be using a different value K1, where K1>K0. +** Both readers can use the same hash table and mapping section to get +** the correct result. There may be entries in the hash table with +** K>K0 but to the first reader, those entries will appear to be unused +** slots in the hash table and so the first reader will get an answer as +** if no values greater than K0 had ever been inserted into the hash table +** in the first place - which is what reader one wants. Meanwhile, the +** second reader using K1 will see additional values that were inserted +** later, which is exactly what reader two wants. +** +** When a rollback occurs, the value of K is decreased. Hash table entries +** that correspond to frames greater than the new K value are removed +** from the hash table at this point. +*/ +#ifndef SQLITE_OMIT_WAL + +/* #include "wal.h" */ + +/* +** Trace output macros +*/ +#if defined(SQLITE_TEST) && defined(SQLITE_DEBUG) +SQLITE_PRIVATE int sqlite3WalTrace = 0; +# define WALTRACE(X) if(sqlite3WalTrace) sqlite3DebugPrintf X +#else +# define WALTRACE(X) +#endif + +/* +** WAL mode depends on atomic aligned 32-bit loads and stores in a few +** places. The following macros try to make this explicit. +*/ +#if GCC_VESRION>=5004000 +# define AtomicLoad(PTR) __atomic_load_n((PTR),__ATOMIC_RELAXED) +# define AtomicStore(PTR,VAL) __atomic_store_n((PTR),(VAL),__ATOMIC_RELAXED) +#else +# define AtomicLoad(PTR) (*(PTR)) +# define AtomicStore(PTR,VAL) (*(PTR) = (VAL)) +#endif + +/* +** The maximum (and only) versions of the wal and wal-index formats +** that may be interpreted by this version of SQLite. +** +** If a client begins recovering a WAL file and finds that (a) the checksum +** values in the wal-header are correct and (b) the version field is not +** WAL_MAX_VERSION, recovery fails and SQLite returns SQLITE_CANTOPEN. +** +** Similarly, if a client successfully reads a wal-index header (i.e. the +** checksum test is successful) and finds that the version field is not +** WALINDEX_MAX_VERSION, then no read-transaction is opened and SQLite +** returns SQLITE_CANTOPEN. +*/ +#define WAL_MAX_VERSION 3007000 +#define WALINDEX_MAX_VERSION 3007000 + +/* +** Index numbers for various locking bytes. WAL_NREADER is the number +** of available reader locks and should be at least 3. The default +** is SQLITE_SHM_NLOCK==8 and WAL_NREADER==5. +** +** Technically, the various VFSes are free to implement these locks however +** they see fit. However, compatibility is encouraged so that VFSes can +** interoperate. The standard implemention used on both unix and windows +** is for the index number to indicate a byte offset into the +** WalCkptInfo.aLock[] array in the wal-index header. In other words, all +** locks are on the shm file. The WALINDEX_LOCK_OFFSET constant (which +** should be 120) is the location in the shm file for the first locking +** byte. +*/ +#define WAL_WRITE_LOCK 0 +#define WAL_ALL_BUT_WRITE 1 +#define WAL_CKPT_LOCK 1 +#define WAL_RECOVER_LOCK 2 +#define WAL_READ_LOCK(I) (3+(I)) +#define WAL_NREADER (SQLITE_SHM_NLOCK-3) + + +/* Object declarations */ +typedef struct WalIndexHdr WalIndexHdr; +typedef struct WalIterator WalIterator; +typedef struct WalCkptInfo WalCkptInfo; + + +/* +** The following object holds a copy of the wal-index header content. +** +** The actual header in the wal-index consists of two copies of this +** object followed by one instance of the WalCkptInfo object. +** For all versions of SQLite through 3.10.0 and probably beyond, +** the locking bytes (WalCkptInfo.aLock) start at offset 120 and +** the total header size is 136 bytes. +** +** The szPage value can be any power of 2 between 512 and 32768, inclusive. +** Or it can be 1 to represent a 65536-byte page. The latter case was +** added in 3.7.1 when support for 64K pages was added. +*/ +struct WalIndexHdr { + u32 iVersion; /* Wal-index version */ + u32 unused; /* Unused (padding) field */ + u32 iChange; /* Counter incremented each transaction */ + u8 isInit; /* 1 when initialized */ + u8 bigEndCksum; /* True if checksums in WAL are big-endian */ + u16 szPage; /* Database page size in bytes. 1==64K */ + u32 mxFrame; /* Index of last valid frame in the WAL */ + u32 nPage; /* Size of database in pages */ + u32 aFrameCksum[2]; /* Checksum of last frame in log */ + u32 aSalt[2]; /* Two salt values copied from WAL header */ + u32 aCksum[2]; /* Checksum over all prior fields */ +}; + +/* +** A copy of the following object occurs in the wal-index immediately +** following the second copy of the WalIndexHdr. This object stores +** information used by checkpoint. +** +** nBackfill is the number of frames in the WAL that have been written +** back into the database. (We call the act of moving content from WAL to +** database "backfilling".) The nBackfill number is never greater than +** WalIndexHdr.mxFrame. nBackfill can only be increased by threads +** holding the WAL_CKPT_LOCK lock (which includes a recovery thread). +** However, a WAL_WRITE_LOCK thread can move the value of nBackfill from +** mxFrame back to zero when the WAL is reset. +** +** nBackfillAttempted is the largest value of nBackfill that a checkpoint +** has attempted to achieve. Normally nBackfill==nBackfillAtempted, however +** the nBackfillAttempted is set before any backfilling is done and the +** nBackfill is only set after all backfilling completes. So if a checkpoint +** crashes, nBackfillAttempted might be larger than nBackfill. The +** WalIndexHdr.mxFrame must never be less than nBackfillAttempted. +** +** The aLock[] field is a set of bytes used for locking. These bytes should +** never be read or written. +** +** There is one entry in aReadMark[] for each reader lock. If a reader +** holds read-lock K, then the value in aReadMark[K] is no greater than +** the mxFrame for that reader. The value READMARK_NOT_USED (0xffffffff) +** for any aReadMark[] means that entry is unused. aReadMark[0] is +** a special case; its value is never used and it exists as a place-holder +** to avoid having to offset aReadMark[] indexs by one. Readers holding +** WAL_READ_LOCK(0) always ignore the entire WAL and read all content +** directly from the database. +** +** The value of aReadMark[K] may only be changed by a thread that +** is holding an exclusive lock on WAL_READ_LOCK(K). Thus, the value of +** aReadMark[K] cannot changed while there is a reader is using that mark +** since the reader will be holding a shared lock on WAL_READ_LOCK(K). +** +** The checkpointer may only transfer frames from WAL to database where +** the frame numbers are less than or equal to every aReadMark[] that is +** in use (that is, every aReadMark[j] for which there is a corresponding +** WAL_READ_LOCK(j)). New readers (usually) pick the aReadMark[] with the +** largest value and will increase an unused aReadMark[] to mxFrame if there +** is not already an aReadMark[] equal to mxFrame. The exception to the +** previous sentence is when nBackfill equals mxFrame (meaning that everything +** in the WAL has been backfilled into the database) then new readers +** will choose aReadMark[0] which has value 0 and hence such reader will +** get all their all content directly from the database file and ignore +** the WAL. +** +** Writers normally append new frames to the end of the WAL. However, +** if nBackfill equals mxFrame (meaning that all WAL content has been +** written back into the database) and if no readers are using the WAL +** (in other words, if there are no WAL_READ_LOCK(i) where i>0) then +** the writer will first "reset" the WAL back to the beginning and start +** writing new content beginning at frame 1. +** +** We assume that 32-bit loads are atomic and so no locks are needed in +** order to read from any aReadMark[] entries. +*/ +struct WalCkptInfo { + u32 nBackfill; /* Number of WAL frames backfilled into DB */ + u32 aReadMark[WAL_NREADER]; /* Reader marks */ + u8 aLock[SQLITE_SHM_NLOCK]; /* Reserved space for locks */ + u32 nBackfillAttempted; /* WAL frames perhaps written, or maybe not */ + u32 notUsed0; /* Available for future enhancements */ +}; +#define READMARK_NOT_USED 0xffffffff + + +/* A block of WALINDEX_LOCK_RESERVED bytes beginning at +** WALINDEX_LOCK_OFFSET is reserved for locks. Since some systems +** only support mandatory file-locks, we do not read or write data +** from the region of the file on which locks are applied. +*/ +#define WALINDEX_LOCK_OFFSET (sizeof(WalIndexHdr)*2+offsetof(WalCkptInfo,aLock)) +#define WALINDEX_HDR_SIZE (sizeof(WalIndexHdr)*2+sizeof(WalCkptInfo)) + +/* Size of header before each frame in wal */ +#define WAL_FRAME_HDRSIZE 24 + +/* Size of write ahead log header, including checksum. */ +#define WAL_HDRSIZE 32 + +/* WAL magic value. Either this value, or the same value with the least +** significant bit also set (WAL_MAGIC | 0x00000001) is stored in 32-bit +** big-endian format in the first 4 bytes of a WAL file. +** +** If the LSB is set, then the checksums for each frame within the WAL +** file are calculated by treating all data as an array of 32-bit +** big-endian words. Otherwise, they are calculated by interpreting +** all data as 32-bit little-endian words. +*/ +#define WAL_MAGIC 0x377f0682 + +/* +** Return the offset of frame iFrame in the write-ahead log file, +** assuming a database page size of szPage bytes. The offset returned +** is to the start of the write-ahead log frame-header. +*/ +#define walFrameOffset(iFrame, szPage) ( \ + WAL_HDRSIZE + ((iFrame)-1)*(i64)((szPage)+WAL_FRAME_HDRSIZE) \ +) + +/* +** An open write-ahead log file is represented by an instance of the +** following object. +*/ +struct Wal { + sqlite3_vfs *pVfs; /* The VFS used to create pDbFd */ + sqlite3_file *pDbFd; /* File handle for the database file */ + sqlite3_file *pWalFd; /* File handle for WAL file */ + u32 iCallback; /* Value to pass to log callback (or 0) */ + i64 mxWalSize; /* Truncate WAL to this size upon reset */ + int nWiData; /* Size of array apWiData */ + int szFirstBlock; /* Size of first block written to WAL file */ + volatile u32 **apWiData; /* Pointer to wal-index content in memory */ + u32 szPage; /* Database page size */ + i16 readLock; /* Which read lock is being held. -1 for none */ + u8 syncFlags; /* Flags to use to sync header writes */ + u8 exclusiveMode; /* Non-zero if connection is in exclusive mode */ + u8 writeLock; /* True if in a write transaction */ + u8 ckptLock; /* True if holding a checkpoint lock */ + u8 readOnly; /* WAL_RDWR, WAL_RDONLY, or WAL_SHM_RDONLY */ + u8 truncateOnCommit; /* True to truncate WAL file on commit */ + u8 syncHeader; /* Fsync the WAL header if true */ + u8 padToSectorBoundary; /* Pad transactions out to the next sector */ + u8 bShmUnreliable; /* SHM content is read-only and unreliable */ + WalIndexHdr hdr; /* Wal-index header for current transaction */ + u32 minFrame; /* Ignore wal frames before this one */ + u32 iReCksum; /* On commit, recalculate checksums from here */ + const char *zWalName; /* Name of WAL file */ + u32 nCkpt; /* Checkpoint sequence counter in the wal-header */ +#ifdef SQLITE_DEBUG + u8 lockError; /* True if a locking error has occurred */ +#endif +#ifdef SQLITE_ENABLE_SNAPSHOT + WalIndexHdr *pSnapshot; /* Start transaction here if not NULL */ +#endif +}; + +/* +** Candidate values for Wal.exclusiveMode. +*/ +#define WAL_NORMAL_MODE 0 +#define WAL_EXCLUSIVE_MODE 1 +#define WAL_HEAPMEMORY_MODE 2 + +/* +** Possible values for WAL.readOnly +*/ +#define WAL_RDWR 0 /* Normal read/write connection */ +#define WAL_RDONLY 1 /* The WAL file is readonly */ +#define WAL_SHM_RDONLY 2 /* The SHM file is readonly */ + +/* +** Each page of the wal-index mapping contains a hash-table made up of +** an array of HASHTABLE_NSLOT elements of the following type. +*/ +typedef u16 ht_slot; + +/* +** This structure is used to implement an iterator that loops through +** all frames in the WAL in database page order. Where two or more frames +** correspond to the same database page, the iterator visits only the +** frame most recently written to the WAL (in other words, the frame with +** the largest index). +** +** The internals of this structure are only accessed by: +** +** walIteratorInit() - Create a new iterator, +** walIteratorNext() - Step an iterator, +** walIteratorFree() - Free an iterator. +** +** This functionality is used by the checkpoint code (see walCheckpoint()). +*/ +struct WalIterator { + int iPrior; /* Last result returned from the iterator */ + int nSegment; /* Number of entries in aSegment[] */ + struct WalSegment { + int iNext; /* Next slot in aIndex[] not yet returned */ + ht_slot *aIndex; /* i0, i1, i2... such that aPgno[iN] ascend */ + u32 *aPgno; /* Array of page numbers. */ + int nEntry; /* Nr. of entries in aPgno[] and aIndex[] */ + int iZero; /* Frame number associated with aPgno[0] */ + } aSegment[1]; /* One for every 32KB page in the wal-index */ +}; + +/* +** Define the parameters of the hash tables in the wal-index file. There +** is a hash-table following every HASHTABLE_NPAGE page numbers in the +** wal-index. +** +** Changing any of these constants will alter the wal-index format and +** create incompatibilities. +*/ +#define HASHTABLE_NPAGE 4096 /* Must be power of 2 */ +#define HASHTABLE_HASH_1 383 /* Should be prime */ +#define HASHTABLE_NSLOT (HASHTABLE_NPAGE*2) /* Must be a power of 2 */ + +/* +** The block of page numbers associated with the first hash-table in a +** wal-index is smaller than usual. This is so that there is a complete +** hash-table on each aligned 32KB page of the wal-index. +*/ +#define HASHTABLE_NPAGE_ONE (HASHTABLE_NPAGE - (WALINDEX_HDR_SIZE/sizeof(u32))) + +/* The wal-index is divided into pages of WALINDEX_PGSZ bytes each. */ +#define WALINDEX_PGSZ ( \ + sizeof(ht_slot)*HASHTABLE_NSLOT + HASHTABLE_NPAGE*sizeof(u32) \ +) + +/* +** Obtain a pointer to the iPage'th page of the wal-index. The wal-index +** is broken into pages of WALINDEX_PGSZ bytes. Wal-index pages are +** numbered from zero. +** +** If the wal-index is currently smaller the iPage pages then the size +** of the wal-index might be increased, but only if it is safe to do +** so. It is safe to enlarge the wal-index if pWal->writeLock is true +** or pWal->exclusiveMode==WAL_HEAPMEMORY_MODE. +** +** If this call is successful, *ppPage is set to point to the wal-index +** page and SQLITE_OK is returned. If an error (an OOM or VFS error) occurs, +** then an SQLite error code is returned and *ppPage is set to 0. +*/ +static SQLITE_NOINLINE int walIndexPageRealloc( + Wal *pWal, /* The WAL context */ + int iPage, /* The page we seek */ + volatile u32 **ppPage /* Write the page pointer here */ +){ + int rc = SQLITE_OK; + + /* Enlarge the pWal->apWiData[] array if required */ + if( pWal->nWiData<=iPage ){ + sqlite3_int64 nByte = sizeof(u32*)*(iPage+1); + volatile u32 **apNew; + apNew = (volatile u32 **)sqlite3_realloc64((void *)pWal->apWiData, nByte); + if( !apNew ){ + *ppPage = 0; + return SQLITE_NOMEM_BKPT; + } + memset((void*)&apNew[pWal->nWiData], 0, + sizeof(u32*)*(iPage+1-pWal->nWiData)); + pWal->apWiData = apNew; + pWal->nWiData = iPage+1; + } + + /* Request a pointer to the required page from the VFS */ + assert( pWal->apWiData[iPage]==0 ); + if( pWal->exclusiveMode==WAL_HEAPMEMORY_MODE ){ + pWal->apWiData[iPage] = (u32 volatile *)sqlite3MallocZero(WALINDEX_PGSZ); + if( !pWal->apWiData[iPage] ) rc = SQLITE_NOMEM_BKPT; + }else{ + rc = sqlite3OsShmMap(pWal->pDbFd, iPage, WALINDEX_PGSZ, + pWal->writeLock, (void volatile **)&pWal->apWiData[iPage] + ); + assert( pWal->apWiData[iPage]!=0 || rc!=SQLITE_OK || pWal->writeLock==0 ); + testcase( pWal->apWiData[iPage]==0 && rc==SQLITE_OK ); + if( (rc&0xff)==SQLITE_READONLY ){ + pWal->readOnly |= WAL_SHM_RDONLY; + if( rc==SQLITE_READONLY ){ + rc = SQLITE_OK; + } + } + } + + *ppPage = pWal->apWiData[iPage]; + assert( iPage==0 || *ppPage || rc!=SQLITE_OK ); + return rc; +} +static int walIndexPage( + Wal *pWal, /* The WAL context */ + int iPage, /* The page we seek */ + volatile u32 **ppPage /* Write the page pointer here */ +){ + if( pWal->nWiData<=iPage || (*ppPage = pWal->apWiData[iPage])==0 ){ + return walIndexPageRealloc(pWal, iPage, ppPage); + } + return SQLITE_OK; +} + +/* +** Return a pointer to the WalCkptInfo structure in the wal-index. +*/ +static volatile WalCkptInfo *walCkptInfo(Wal *pWal){ + assert( pWal->nWiData>0 && pWal->apWiData[0] ); + return (volatile WalCkptInfo*)&(pWal->apWiData[0][sizeof(WalIndexHdr)/2]); +} + +/* +** Return a pointer to the WalIndexHdr structure in the wal-index. +*/ +static volatile WalIndexHdr *walIndexHdr(Wal *pWal){ + assert( pWal->nWiData>0 && pWal->apWiData[0] ); + return (volatile WalIndexHdr*)pWal->apWiData[0]; +} + +/* +** The argument to this macro must be of type u32. On a little-endian +** architecture, it returns the u32 value that results from interpreting +** the 4 bytes as a big-endian value. On a big-endian architecture, it +** returns the value that would be produced by interpreting the 4 bytes +** of the input value as a little-endian integer. +*/ +#define BYTESWAP32(x) ( \ + (((x)&0x000000FF)<<24) + (((x)&0x0000FF00)<<8) \ + + (((x)&0x00FF0000)>>8) + (((x)&0xFF000000)>>24) \ +) + +/* +** Generate or extend an 8 byte checksum based on the data in +** array aByte[] and the initial values of aIn[0] and aIn[1] (or +** initial values of 0 and 0 if aIn==NULL). +** +** The checksum is written back into aOut[] before returning. +** +** nByte must be a positive multiple of 8. +*/ +static void walChecksumBytes( + int nativeCksum, /* True for native byte-order, false for non-native */ + u8 *a, /* Content to be checksummed */ + int nByte, /* Bytes of content in a[]. Must be a multiple of 8. */ + const u32 *aIn, /* Initial checksum value input */ + u32 *aOut /* OUT: Final checksum value output */ +){ + u32 s1, s2; + u32 *aData = (u32 *)a; + u32 *aEnd = (u32 *)&a[nByte]; + + if( aIn ){ + s1 = aIn[0]; + s2 = aIn[1]; + }else{ + s1 = s2 = 0; + } + + assert( nByte>=8 ); + assert( (nByte&0x00000007)==0 ); + assert( nByte<=65536 ); + + if( nativeCksum ){ + do { + s1 += *aData++ + s2; + s2 += *aData++ + s1; + }while( aDataexclusiveMode!=WAL_HEAPMEMORY_MODE ){ + sqlite3OsShmBarrier(pWal->pDbFd); + } +} + +/* +** Write the header information in pWal->hdr into the wal-index. +** +** The checksum on pWal->hdr is updated before it is written. +*/ +static void walIndexWriteHdr(Wal *pWal){ + volatile WalIndexHdr *aHdr = walIndexHdr(pWal); + const int nCksum = offsetof(WalIndexHdr, aCksum); + + assert( pWal->writeLock ); + pWal->hdr.isInit = 1; + pWal->hdr.iVersion = WALINDEX_MAX_VERSION; + walChecksumBytes(1, (u8*)&pWal->hdr, nCksum, 0, pWal->hdr.aCksum); + memcpy((void*)&aHdr[1], (const void*)&pWal->hdr, sizeof(WalIndexHdr)); + walShmBarrier(pWal); + memcpy((void*)&aHdr[0], (const void*)&pWal->hdr, sizeof(WalIndexHdr)); +} + +/* +** This function encodes a single frame header and writes it to a buffer +** supplied by the caller. A frame-header is made up of a series of +** 4-byte big-endian integers, as follows: +** +** 0: Page number. +** 4: For commit records, the size of the database image in pages +** after the commit. For all other records, zero. +** 8: Salt-1 (copied from the wal-header) +** 12: Salt-2 (copied from the wal-header) +** 16: Checksum-1. +** 20: Checksum-2. +*/ +static void walEncodeFrame( + Wal *pWal, /* The write-ahead log */ + u32 iPage, /* Database page number for frame */ + u32 nTruncate, /* New db size (or 0 for non-commit frames) */ + u8 *aData, /* Pointer to page data */ + u8 *aFrame /* OUT: Write encoded frame here */ +){ + int nativeCksum; /* True for native byte-order checksums */ + u32 *aCksum = pWal->hdr.aFrameCksum; + assert( WAL_FRAME_HDRSIZE==24 ); + sqlite3Put4byte(&aFrame[0], iPage); + sqlite3Put4byte(&aFrame[4], nTruncate); + if( pWal->iReCksum==0 ){ + memcpy(&aFrame[8], pWal->hdr.aSalt, 8); + + nativeCksum = (pWal->hdr.bigEndCksum==SQLITE_BIGENDIAN); + walChecksumBytes(nativeCksum, aFrame, 8, aCksum, aCksum); + walChecksumBytes(nativeCksum, aData, pWal->szPage, aCksum, aCksum); + + sqlite3Put4byte(&aFrame[16], aCksum[0]); + sqlite3Put4byte(&aFrame[20], aCksum[1]); + }else{ + memset(&aFrame[8], 0, 16); + } +} + +/* +** Check to see if the frame with header in aFrame[] and content +** in aData[] is valid. If it is a valid frame, fill *piPage and +** *pnTruncate and return true. Return if the frame is not valid. +*/ +static int walDecodeFrame( + Wal *pWal, /* The write-ahead log */ + u32 *piPage, /* OUT: Database page number for frame */ + u32 *pnTruncate, /* OUT: New db size (or 0 if not commit) */ + u8 *aData, /* Pointer to page data (for checksum) */ + u8 *aFrame /* Frame data */ +){ + int nativeCksum; /* True for native byte-order checksums */ + u32 *aCksum = pWal->hdr.aFrameCksum; + u32 pgno; /* Page number of the frame */ + assert( WAL_FRAME_HDRSIZE==24 ); + + /* A frame is only valid if the salt values in the frame-header + ** match the salt values in the wal-header. + */ + if( memcmp(&pWal->hdr.aSalt, &aFrame[8], 8)!=0 ){ + return 0; + } + + /* A frame is only valid if the page number is creater than zero. + */ + pgno = sqlite3Get4byte(&aFrame[0]); + if( pgno==0 ){ + return 0; + } + + /* A frame is only valid if a checksum of the WAL header, + ** all prior frams, the first 16 bytes of this frame-header, + ** and the frame-data matches the checksum in the last 8 + ** bytes of this frame-header. + */ + nativeCksum = (pWal->hdr.bigEndCksum==SQLITE_BIGENDIAN); + walChecksumBytes(nativeCksum, aFrame, 8, aCksum, aCksum); + walChecksumBytes(nativeCksum, aData, pWal->szPage, aCksum, aCksum); + if( aCksum[0]!=sqlite3Get4byte(&aFrame[16]) + || aCksum[1]!=sqlite3Get4byte(&aFrame[20]) + ){ + /* Checksum failed. */ + return 0; + } + + /* If we reach this point, the frame is valid. Return the page number + ** and the new database size. + */ + *piPage = pgno; + *pnTruncate = sqlite3Get4byte(&aFrame[4]); + return 1; +} + + +#if defined(SQLITE_TEST) && defined(SQLITE_DEBUG) +/* +** Names of locks. This routine is used to provide debugging output and is not +** a part of an ordinary build. +*/ +static const char *walLockName(int lockIdx){ + if( lockIdx==WAL_WRITE_LOCK ){ + return "WRITE-LOCK"; + }else if( lockIdx==WAL_CKPT_LOCK ){ + return "CKPT-LOCK"; + }else if( lockIdx==WAL_RECOVER_LOCK ){ + return "RECOVER-LOCK"; + }else{ + static char zName[15]; + sqlite3_snprintf(sizeof(zName), zName, "READ-LOCK[%d]", + lockIdx-WAL_READ_LOCK(0)); + return zName; + } +} +#endif /*defined(SQLITE_TEST) || defined(SQLITE_DEBUG) */ + + +/* +** Set or release locks on the WAL. Locks are either shared or exclusive. +** A lock cannot be moved directly between shared and exclusive - it must go +** through the unlocked state first. +** +** In locking_mode=EXCLUSIVE, all of these routines become no-ops. +*/ +static int walLockShared(Wal *pWal, int lockIdx){ + int rc; + if( pWal->exclusiveMode ) return SQLITE_OK; + rc = sqlite3OsShmLock(pWal->pDbFd, lockIdx, 1, + SQLITE_SHM_LOCK | SQLITE_SHM_SHARED); + WALTRACE(("WAL%p: acquire SHARED-%s %s\n", pWal, + walLockName(lockIdx), rc ? "failed" : "ok")); + VVA_ONLY( pWal->lockError = (u8)(rc!=SQLITE_OK && rc!=SQLITE_BUSY); ) + return rc; +} +static void walUnlockShared(Wal *pWal, int lockIdx){ + if( pWal->exclusiveMode ) return; + (void)sqlite3OsShmLock(pWal->pDbFd, lockIdx, 1, + SQLITE_SHM_UNLOCK | SQLITE_SHM_SHARED); + WALTRACE(("WAL%p: release SHARED-%s\n", pWal, walLockName(lockIdx))); +} +static int walLockExclusive(Wal *pWal, int lockIdx, int n){ + int rc; + if( pWal->exclusiveMode ) return SQLITE_OK; + rc = sqlite3OsShmLock(pWal->pDbFd, lockIdx, n, + SQLITE_SHM_LOCK | SQLITE_SHM_EXCLUSIVE); + WALTRACE(("WAL%p: acquire EXCLUSIVE-%s cnt=%d %s\n", pWal, + walLockName(lockIdx), n, rc ? "failed" : "ok")); + VVA_ONLY( pWal->lockError = (u8)(rc!=SQLITE_OK && rc!=SQLITE_BUSY); ) + return rc; +} +static void walUnlockExclusive(Wal *pWal, int lockIdx, int n){ + if( pWal->exclusiveMode ) return; + (void)sqlite3OsShmLock(pWal->pDbFd, lockIdx, n, + SQLITE_SHM_UNLOCK | SQLITE_SHM_EXCLUSIVE); + WALTRACE(("WAL%p: release EXCLUSIVE-%s cnt=%d\n", pWal, + walLockName(lockIdx), n)); +} + +/* +** Compute a hash on a page number. The resulting hash value must land +** between 0 and (HASHTABLE_NSLOT-1). The walHashNext() function advances +** the hash to the next value in the event of a collision. +*/ +static int walHash(u32 iPage){ + assert( iPage>0 ); + assert( (HASHTABLE_NSLOT & (HASHTABLE_NSLOT-1))==0 ); + return (iPage*HASHTABLE_HASH_1) & (HASHTABLE_NSLOT-1); +} +static int walNextHash(int iPriorHash){ + return (iPriorHash+1)&(HASHTABLE_NSLOT-1); +} + +/* +** An instance of the WalHashLoc object is used to describe the location +** of a page hash table in the wal-index. This becomes the return value +** from walHashGet(). +*/ +typedef struct WalHashLoc WalHashLoc; +struct WalHashLoc { + volatile ht_slot *aHash; /* Start of the wal-index hash table */ + volatile u32 *aPgno; /* aPgno[1] is the page of first frame indexed */ + u32 iZero; /* One less than the frame number of first indexed*/ +}; + +/* +** Return pointers to the hash table and page number array stored on +** page iHash of the wal-index. The wal-index is broken into 32KB pages +** numbered starting from 0. +** +** Set output variable pLoc->aHash to point to the start of the hash table +** in the wal-index file. Set pLoc->iZero to one less than the frame +** number of the first frame indexed by this hash table. If a +** slot in the hash table is set to N, it refers to frame number +** (pLoc->iZero+N) in the log. +** +** Finally, set pLoc->aPgno so that pLoc->aPgno[1] is the page number of the +** first frame indexed by the hash table, frame (pLoc->iZero+1). +*/ +static int walHashGet( + Wal *pWal, /* WAL handle */ + int iHash, /* Find the iHash'th table */ + WalHashLoc *pLoc /* OUT: Hash table location */ +){ + int rc; /* Return code */ + + rc = walIndexPage(pWal, iHash, &pLoc->aPgno); + assert( rc==SQLITE_OK || iHash>0 ); + + if( rc==SQLITE_OK ){ + pLoc->aHash = (volatile ht_slot *)&pLoc->aPgno[HASHTABLE_NPAGE]; + if( iHash==0 ){ + pLoc->aPgno = &pLoc->aPgno[WALINDEX_HDR_SIZE/sizeof(u32)]; + pLoc->iZero = 0; + }else{ + pLoc->iZero = HASHTABLE_NPAGE_ONE + (iHash-1)*HASHTABLE_NPAGE; + } + pLoc->aPgno = &pLoc->aPgno[-1]; + } + return rc; +} + +/* +** Return the number of the wal-index page that contains the hash-table +** and page-number array that contain entries corresponding to WAL frame +** iFrame. The wal-index is broken up into 32KB pages. Wal-index pages +** are numbered starting from 0. +*/ +static int walFramePage(u32 iFrame){ + int iHash = (iFrame+HASHTABLE_NPAGE-HASHTABLE_NPAGE_ONE-1) / HASHTABLE_NPAGE; + assert( (iHash==0 || iFrame>HASHTABLE_NPAGE_ONE) + && (iHash>=1 || iFrame<=HASHTABLE_NPAGE_ONE) + && (iHash<=1 || iFrame>(HASHTABLE_NPAGE_ONE+HASHTABLE_NPAGE)) + && (iHash>=2 || iFrame<=HASHTABLE_NPAGE_ONE+HASHTABLE_NPAGE) + && (iHash<=2 || iFrame>(HASHTABLE_NPAGE_ONE+2*HASHTABLE_NPAGE)) + ); + return iHash; +} + +/* +** Return the page number associated with frame iFrame in this WAL. +*/ +static u32 walFramePgno(Wal *pWal, u32 iFrame){ + int iHash = walFramePage(iFrame); + if( iHash==0 ){ + return pWal->apWiData[0][WALINDEX_HDR_SIZE/sizeof(u32) + iFrame - 1]; + } + return pWal->apWiData[iHash][(iFrame-1-HASHTABLE_NPAGE_ONE)%HASHTABLE_NPAGE]; +} + +/* +** Remove entries from the hash table that point to WAL slots greater +** than pWal->hdr.mxFrame. +** +** This function is called whenever pWal->hdr.mxFrame is decreased due +** to a rollback or savepoint. +** +** At most only the hash table containing pWal->hdr.mxFrame needs to be +** updated. Any later hash tables will be automatically cleared when +** pWal->hdr.mxFrame advances to the point where those hash tables are +** actually needed. +*/ +static void walCleanupHash(Wal *pWal){ + WalHashLoc sLoc; /* Hash table location */ + int iLimit = 0; /* Zero values greater than this */ + int nByte; /* Number of bytes to zero in aPgno[] */ + int i; /* Used to iterate through aHash[] */ + int rc; /* Return code form walHashGet() */ + + assert( pWal->writeLock ); + testcase( pWal->hdr.mxFrame==HASHTABLE_NPAGE_ONE-1 ); + testcase( pWal->hdr.mxFrame==HASHTABLE_NPAGE_ONE ); + testcase( pWal->hdr.mxFrame==HASHTABLE_NPAGE_ONE+1 ); + + if( pWal->hdr.mxFrame==0 ) return; + + /* Obtain pointers to the hash-table and page-number array containing + ** the entry that corresponds to frame pWal->hdr.mxFrame. It is guaranteed + ** that the page said hash-table and array reside on is already mapped.(1) + */ + assert( pWal->nWiData>walFramePage(pWal->hdr.mxFrame) ); + assert( pWal->apWiData[walFramePage(pWal->hdr.mxFrame)] ); + rc = walHashGet(pWal, walFramePage(pWal->hdr.mxFrame), &sLoc); + if( NEVER(rc) ) return; /* Defense-in-depth, in case (1) above is wrong */ + + /* Zero all hash-table entries that correspond to frame numbers greater + ** than pWal->hdr.mxFrame. + */ + iLimit = pWal->hdr.mxFrame - sLoc.iZero; + assert( iLimit>0 ); + for(i=0; iiLimit ){ + sLoc.aHash[i] = 0; + } + } + + /* Zero the entries in the aPgno array that correspond to frames with + ** frame numbers greater than pWal->hdr.mxFrame. + */ + nByte = (int)((char *)sLoc.aHash - (char *)&sLoc.aPgno[iLimit+1]); + memset((void *)&sLoc.aPgno[iLimit+1], 0, nByte); + +#ifdef SQLITE_ENABLE_EXPENSIVE_ASSERT + /* Verify that the every entry in the mapping region is still reachable + ** via the hash table even after the cleanup. + */ + if( iLimit ){ + int j; /* Loop counter */ + int iKey; /* Hash key */ + for(j=1; j<=iLimit; j++){ + for(iKey=walHash(sLoc.aPgno[j]);sLoc.aHash[iKey];iKey=walNextHash(iKey)){ + if( sLoc.aHash[iKey]==j ) break; + } + assert( sLoc.aHash[iKey]==j ); + } + } +#endif /* SQLITE_ENABLE_EXPENSIVE_ASSERT */ +} + + +/* +** Set an entry in the wal-index that will map database page number +** pPage into WAL frame iFrame. +*/ +static int walIndexAppend(Wal *pWal, u32 iFrame, u32 iPage){ + int rc; /* Return code */ + WalHashLoc sLoc; /* Wal-index hash table location */ + + rc = walHashGet(pWal, walFramePage(iFrame), &sLoc); + + /* Assuming the wal-index file was successfully mapped, populate the + ** page number array and hash table entry. + */ + if( rc==SQLITE_OK ){ + int iKey; /* Hash table key */ + int idx; /* Value to write to hash-table slot */ + int nCollide; /* Number of hash collisions */ + + idx = iFrame - sLoc.iZero; + assert( idx <= HASHTABLE_NSLOT/2 + 1 ); + + /* If this is the first entry to be added to this hash-table, zero the + ** entire hash table and aPgno[] array before proceeding. + */ + if( idx==1 ){ + int nByte = (int)((u8 *)&sLoc.aHash[HASHTABLE_NSLOT] + - (u8 *)&sLoc.aPgno[1]); + memset((void*)&sLoc.aPgno[1], 0, nByte); + } + + /* If the entry in aPgno[] is already set, then the previous writer + ** must have exited unexpectedly in the middle of a transaction (after + ** writing one or more dirty pages to the WAL to free up memory). + ** Remove the remnants of that writers uncommitted transaction from + ** the hash-table before writing any new entries. + */ + if( sLoc.aPgno[idx] ){ + walCleanupHash(pWal); + assert( !sLoc.aPgno[idx] ); + } + + /* Write the aPgno[] array entry and the hash-table slot. */ + nCollide = idx; + for(iKey=walHash(iPage); sLoc.aHash[iKey]; iKey=walNextHash(iKey)){ + if( (nCollide--)==0 ) return SQLITE_CORRUPT_BKPT; + } + sLoc.aPgno[idx] = iPage; + sLoc.aHash[iKey] = (ht_slot)idx; + +#ifdef SQLITE_ENABLE_EXPENSIVE_ASSERT + /* Verify that the number of entries in the hash table exactly equals + ** the number of entries in the mapping region. + */ + { + int i; /* Loop counter */ + int nEntry = 0; /* Number of entries in the hash table */ + for(i=0; ickptLock==1 || pWal->ckptLock==0 ); + assert( WAL_ALL_BUT_WRITE==WAL_WRITE_LOCK+1 ); + assert( WAL_CKPT_LOCK==WAL_ALL_BUT_WRITE ); + assert( pWal->writeLock ); + iLock = WAL_ALL_BUT_WRITE + pWal->ckptLock; + rc = walLockExclusive(pWal, iLock, WAL_READ_LOCK(0)-iLock); + if( rc==SQLITE_OK ){ + rc = walLockExclusive(pWal, WAL_READ_LOCK(1), WAL_NREADER-1); + if( rc!=SQLITE_OK ){ + walUnlockExclusive(pWal, iLock, WAL_READ_LOCK(0)-iLock); + } + } + if( rc ){ + return rc; + } + + WALTRACE(("WAL%p: recovery begin...\n", pWal)); + + memset(&pWal->hdr, 0, sizeof(WalIndexHdr)); + + rc = sqlite3OsFileSize(pWal->pWalFd, &nSize); + if( rc!=SQLITE_OK ){ + goto recovery_error; + } + + if( nSize>WAL_HDRSIZE ){ + u8 aBuf[WAL_HDRSIZE]; /* Buffer to load WAL header into */ + u8 *aFrame = 0; /* Malloc'd buffer to load entire frame */ + int szFrame; /* Number of bytes in buffer aFrame[] */ + u8 *aData; /* Pointer to data part of aFrame buffer */ + int iFrame; /* Index of last frame read */ + i64 iOffset; /* Next offset to read from log file */ + int szPage; /* Page size according to the log */ + u32 magic; /* Magic value read from WAL header */ + u32 version; /* Magic value read from WAL header */ + int isValid; /* True if this frame is valid */ + + /* Read in the WAL header. */ + rc = sqlite3OsRead(pWal->pWalFd, aBuf, WAL_HDRSIZE, 0); + if( rc!=SQLITE_OK ){ + goto recovery_error; + } + + /* If the database page size is not a power of two, or is greater than + ** SQLITE_MAX_PAGE_SIZE, conclude that the WAL file contains no valid + ** data. Similarly, if the 'magic' value is invalid, ignore the whole + ** WAL file. + */ + magic = sqlite3Get4byte(&aBuf[0]); + szPage = sqlite3Get4byte(&aBuf[8]); + if( (magic&0xFFFFFFFE)!=WAL_MAGIC + || szPage&(szPage-1) + || szPage>SQLITE_MAX_PAGE_SIZE + || szPage<512 + ){ + goto finished; + } + pWal->hdr.bigEndCksum = (u8)(magic&0x00000001); + pWal->szPage = szPage; + pWal->nCkpt = sqlite3Get4byte(&aBuf[12]); + memcpy(&pWal->hdr.aSalt, &aBuf[16], 8); + + /* Verify that the WAL header checksum is correct */ + walChecksumBytes(pWal->hdr.bigEndCksum==SQLITE_BIGENDIAN, + aBuf, WAL_HDRSIZE-2*4, 0, pWal->hdr.aFrameCksum + ); + if( pWal->hdr.aFrameCksum[0]!=sqlite3Get4byte(&aBuf[24]) + || pWal->hdr.aFrameCksum[1]!=sqlite3Get4byte(&aBuf[28]) + ){ + goto finished; + } + + /* Verify that the version number on the WAL format is one that + ** are able to understand */ + version = sqlite3Get4byte(&aBuf[4]); + if( version!=WAL_MAX_VERSION ){ + rc = SQLITE_CANTOPEN_BKPT; + goto finished; + } + + /* Malloc a buffer to read frames into. */ + szFrame = szPage + WAL_FRAME_HDRSIZE; + aFrame = (u8 *)sqlite3_malloc64(szFrame); + if( !aFrame ){ + rc = SQLITE_NOMEM_BKPT; + goto recovery_error; + } + aData = &aFrame[WAL_FRAME_HDRSIZE]; + + /* Read all frames from the log file. */ + iFrame = 0; + for(iOffset=WAL_HDRSIZE; (iOffset+szFrame)<=nSize; iOffset+=szFrame){ + u32 pgno; /* Database page number for frame */ + u32 nTruncate; /* dbsize field from frame header */ + + /* Read and decode the next log frame. */ + iFrame++; + rc = sqlite3OsRead(pWal->pWalFd, aFrame, szFrame, iOffset); + if( rc!=SQLITE_OK ) break; + isValid = walDecodeFrame(pWal, &pgno, &nTruncate, aData, aFrame); + if( !isValid ) break; + rc = walIndexAppend(pWal, iFrame, pgno); + if( rc!=SQLITE_OK ) break; + + /* If nTruncate is non-zero, this is a commit record. */ + if( nTruncate ){ + pWal->hdr.mxFrame = iFrame; + pWal->hdr.nPage = nTruncate; + pWal->hdr.szPage = (u16)((szPage&0xff00) | (szPage>>16)); + testcase( szPage<=32768 ); + testcase( szPage>=65536 ); + aFrameCksum[0] = pWal->hdr.aFrameCksum[0]; + aFrameCksum[1] = pWal->hdr.aFrameCksum[1]; + } + } + + sqlite3_free(aFrame); + } + +finished: + if( rc==SQLITE_OK ){ + volatile WalCkptInfo *pInfo; + int i; + pWal->hdr.aFrameCksum[0] = aFrameCksum[0]; + pWal->hdr.aFrameCksum[1] = aFrameCksum[1]; + walIndexWriteHdr(pWal); + + /* Reset the checkpoint-header. This is safe because this thread is + ** currently holding locks that exclude all other readers, writers and + ** checkpointers. + */ + pInfo = walCkptInfo(pWal); + pInfo->nBackfill = 0; + pInfo->nBackfillAttempted = pWal->hdr.mxFrame; + pInfo->aReadMark[0] = 0; + for(i=1; iaReadMark[i] = READMARK_NOT_USED; + if( pWal->hdr.mxFrame ) pInfo->aReadMark[1] = pWal->hdr.mxFrame; + + /* If more than one frame was recovered from the log file, report an + ** event via sqlite3_log(). This is to help with identifying performance + ** problems caused by applications routinely shutting down without + ** checkpointing the log file. + */ + if( pWal->hdr.nPage ){ + sqlite3_log(SQLITE_NOTICE_RECOVER_WAL, + "recovered %d frames from WAL file %s", + pWal->hdr.mxFrame, pWal->zWalName + ); + } + } + +recovery_error: + WALTRACE(("WAL%p: recovery %s\n", pWal, rc ? "failed" : "ok")); + walUnlockExclusive(pWal, iLock, WAL_READ_LOCK(0)-iLock); + walUnlockExclusive(pWal, WAL_READ_LOCK(1), WAL_NREADER-1); + return rc; +} + +/* +** Close an open wal-index. +*/ +static void walIndexClose(Wal *pWal, int isDelete){ + if( pWal->exclusiveMode==WAL_HEAPMEMORY_MODE || pWal->bShmUnreliable ){ + int i; + for(i=0; inWiData; i++){ + sqlite3_free((void *)pWal->apWiData[i]); + pWal->apWiData[i] = 0; + } + } + if( pWal->exclusiveMode!=WAL_HEAPMEMORY_MODE ){ + sqlite3OsShmUnmap(pWal->pDbFd, isDelete); + } +} + +/* +** Open a connection to the WAL file zWalName. The database file must +** already be opened on connection pDbFd. The buffer that zWalName points +** to must remain valid for the lifetime of the returned Wal* handle. +** +** A SHARED lock should be held on the database file when this function +** is called. The purpose of this SHARED lock is to prevent any other +** client from unlinking the WAL or wal-index file. If another process +** were to do this just after this client opened one of these files, the +** system would be badly broken. +** +** If the log file is successfully opened, SQLITE_OK is returned and +** *ppWal is set to point to a new WAL handle. If an error occurs, +** an SQLite error code is returned and *ppWal is left unmodified. +*/ +SQLITE_PRIVATE int sqlite3WalOpen( + sqlite3_vfs *pVfs, /* vfs module to open wal and wal-index */ + sqlite3_file *pDbFd, /* The open database file */ + const char *zWalName, /* Name of the WAL file */ + int bNoShm, /* True to run in heap-memory mode */ + i64 mxWalSize, /* Truncate WAL to this size on reset */ + Wal **ppWal /* OUT: Allocated Wal handle */ +){ + int rc; /* Return Code */ + Wal *pRet; /* Object to allocate and return */ + int flags; /* Flags passed to OsOpen() */ + + assert( zWalName && zWalName[0] ); + assert( pDbFd ); + + /* In the amalgamation, the os_unix.c and os_win.c source files come before + ** this source file. Verify that the #defines of the locking byte offsets + ** in os_unix.c and os_win.c agree with the WALINDEX_LOCK_OFFSET value. + ** For that matter, if the lock offset ever changes from its initial design + ** value of 120, we need to know that so there is an assert() to check it. + */ + assert( 120==WALINDEX_LOCK_OFFSET ); + assert( 136==WALINDEX_HDR_SIZE ); +#ifdef WIN_SHM_BASE + assert( WIN_SHM_BASE==WALINDEX_LOCK_OFFSET ); +#endif +#ifdef UNIX_SHM_BASE + assert( UNIX_SHM_BASE==WALINDEX_LOCK_OFFSET ); +#endif + + + /* Allocate an instance of struct Wal to return. */ + *ppWal = 0; + pRet = (Wal*)sqlite3MallocZero(sizeof(Wal) + pVfs->szOsFile); + if( !pRet ){ + return SQLITE_NOMEM_BKPT; + } + + pRet->pVfs = pVfs; + pRet->pWalFd = (sqlite3_file *)&pRet[1]; + pRet->pDbFd = pDbFd; + pRet->readLock = -1; + pRet->mxWalSize = mxWalSize; + pRet->zWalName = zWalName; + pRet->syncHeader = 1; + pRet->padToSectorBoundary = 1; + pRet->exclusiveMode = (bNoShm ? WAL_HEAPMEMORY_MODE: WAL_NORMAL_MODE); + + /* Open file handle on the write-ahead log file. */ + flags = (SQLITE_OPEN_READWRITE|SQLITE_OPEN_CREATE|SQLITE_OPEN_WAL); + rc = sqlite3OsOpen(pVfs, zWalName, pRet->pWalFd, flags, &flags); + if( rc==SQLITE_OK && flags&SQLITE_OPEN_READONLY ){ + pRet->readOnly = WAL_RDONLY; + } + + if( rc!=SQLITE_OK ){ + walIndexClose(pRet, 0); + sqlite3OsClose(pRet->pWalFd); + sqlite3_free(pRet); + }else{ + int iDC = sqlite3OsDeviceCharacteristics(pDbFd); + if( iDC & SQLITE_IOCAP_SEQUENTIAL ){ pRet->syncHeader = 0; } + if( iDC & SQLITE_IOCAP_POWERSAFE_OVERWRITE ){ + pRet->padToSectorBoundary = 0; + } + *ppWal = pRet; + WALTRACE(("WAL%d: opened\n", pRet)); + } + return rc; +} + +/* +** Change the size to which the WAL file is trucated on each reset. +*/ +SQLITE_PRIVATE void sqlite3WalLimit(Wal *pWal, i64 iLimit){ + if( pWal ) pWal->mxWalSize = iLimit; +} + +/* +** Find the smallest page number out of all pages held in the WAL that +** has not been returned by any prior invocation of this method on the +** same WalIterator object. Write into *piFrame the frame index where +** that page was last written into the WAL. Write into *piPage the page +** number. +** +** Return 0 on success. If there are no pages in the WAL with a page +** number larger than *piPage, then return 1. +*/ +static int walIteratorNext( + WalIterator *p, /* Iterator */ + u32 *piPage, /* OUT: The page number of the next page */ + u32 *piFrame /* OUT: Wal frame index of next page */ +){ + u32 iMin; /* Result pgno must be greater than iMin */ + u32 iRet = 0xFFFFFFFF; /* 0xffffffff is never a valid page number */ + int i; /* For looping through segments */ + + iMin = p->iPrior; + assert( iMin<0xffffffff ); + for(i=p->nSegment-1; i>=0; i--){ + struct WalSegment *pSegment = &p->aSegment[i]; + while( pSegment->iNextnEntry ){ + u32 iPg = pSegment->aPgno[pSegment->aIndex[pSegment->iNext]]; + if( iPg>iMin ){ + if( iPgiZero + pSegment->aIndex[pSegment->iNext]; + } + break; + } + pSegment->iNext++; + } + } + + *piPage = p->iPrior = iRet; + return (iRet==0xFFFFFFFF); +} + +/* +** This function merges two sorted lists into a single sorted list. +** +** aLeft[] and aRight[] are arrays of indices. The sort key is +** aContent[aLeft[]] and aContent[aRight[]]. Upon entry, the following +** is guaranteed for all J0 && nRight>0 ); + while( iRight=nRight || aContent[aLeft[iLeft]]=nLeft || aContent[aLeft[iLeft]]>dbpage ); + assert( iRight>=nRight || aContent[aRight[iRight]]>dbpage ); + } + + *paRight = aLeft; + *pnRight = iOut; + memcpy(aLeft, aTmp, sizeof(aTmp[0])*iOut); +} + +/* +** Sort the elements in list aList using aContent[] as the sort key. +** Remove elements with duplicate keys, preferring to keep the +** larger aList[] values. +** +** The aList[] entries are indices into aContent[]. The values in +** aList[] are to be sorted so that for all J0 ); + assert( HASHTABLE_NPAGE==(1<<(ArraySize(aSub)-1)) ); + + for(iList=0; iListaList && p->nList<=(1<aList==&aList[iList&~((2<aList, p->nList, &aMerge, &nMerge, aBuffer); + } + aSub[iSub].aList = aMerge; + aSub[iSub].nList = nMerge; + } + + for(iSub++; iSubnList<=(1<aList==&aList[nList&~((2<aList, p->nList, &aMerge, &nMerge, aBuffer); + } + } + assert( aMerge==aList ); + *pnList = nMerge; + +#ifdef SQLITE_DEBUG + { + int i; + for(i=1; i<*pnList; i++){ + assert( aContent[aList[i]] > aContent[aList[i-1]] ); + } + } +#endif +} + +/* +** Free an iterator allocated by walIteratorInit(). +*/ +static void walIteratorFree(WalIterator *p){ + sqlite3_free(p); +} + +/* +** Construct a WalInterator object that can be used to loop over all +** pages in the WAL following frame nBackfill in ascending order. Frames +** nBackfill or earlier may be included - excluding them is an optimization +** only. The caller must hold the checkpoint lock. +** +** On success, make *pp point to the newly allocated WalInterator object +** return SQLITE_OK. Otherwise, return an error code. If this routine +** returns an error, the value of *pp is undefined. +** +** The calling routine should invoke walIteratorFree() to destroy the +** WalIterator object when it has finished with it. +*/ +static int walIteratorInit(Wal *pWal, u32 nBackfill, WalIterator **pp){ + WalIterator *p; /* Return value */ + int nSegment; /* Number of segments to merge */ + u32 iLast; /* Last frame in log */ + sqlite3_int64 nByte; /* Number of bytes to allocate */ + int i; /* Iterator variable */ + ht_slot *aTmp; /* Temp space used by merge-sort */ + int rc = SQLITE_OK; /* Return Code */ + + /* This routine only runs while holding the checkpoint lock. And + ** it only runs if there is actually content in the log (mxFrame>0). + */ + assert( pWal->ckptLock && pWal->hdr.mxFrame>0 ); + iLast = pWal->hdr.mxFrame; + + /* Allocate space for the WalIterator object. */ + nSegment = walFramePage(iLast) + 1; + nByte = sizeof(WalIterator) + + (nSegment-1)*sizeof(struct WalSegment) + + iLast*sizeof(ht_slot); + p = (WalIterator *)sqlite3_malloc64(nByte); + if( !p ){ + return SQLITE_NOMEM_BKPT; + } + memset(p, 0, nByte); + p->nSegment = nSegment; + + /* Allocate temporary space used by the merge-sort routine. This block + ** of memory will be freed before this function returns. + */ + aTmp = (ht_slot *)sqlite3_malloc64( + sizeof(ht_slot) * (iLast>HASHTABLE_NPAGE?HASHTABLE_NPAGE:iLast) + ); + if( !aTmp ){ + rc = SQLITE_NOMEM_BKPT; + } + + for(i=walFramePage(nBackfill+1); rc==SQLITE_OK && iaSegment[p->nSegment])[sLoc.iZero]; + sLoc.iZero++; + + for(j=0; jaSegment[i].iZero = sLoc.iZero; + p->aSegment[i].nEntry = nEntry; + p->aSegment[i].aIndex = aIndex; + p->aSegment[i].aPgno = (u32 *)sLoc.aPgno; + } + } + sqlite3_free(aTmp); + + if( rc!=SQLITE_OK ){ + walIteratorFree(p); + p = 0; + } + *pp = p; + return rc; +} + +/* +** Attempt to obtain the exclusive WAL lock defined by parameters lockIdx and +** n. If the attempt fails and parameter xBusy is not NULL, then it is a +** busy-handler function. Invoke it and retry the lock until either the +** lock is successfully obtained or the busy-handler returns 0. +*/ +static int walBusyLock( + Wal *pWal, /* WAL connection */ + int (*xBusy)(void*), /* Function to call when busy */ + void *pBusyArg, /* Context argument for xBusyHandler */ + int lockIdx, /* Offset of first byte to lock */ + int n /* Number of bytes to lock */ +){ + int rc; + do { + rc = walLockExclusive(pWal, lockIdx, n); + }while( xBusy && rc==SQLITE_BUSY && xBusy(pBusyArg) ); + return rc; +} + +/* +** The cache of the wal-index header must be valid to call this function. +** Return the page-size in bytes used by the database. +*/ +static int walPagesize(Wal *pWal){ + return (pWal->hdr.szPage&0xfe00) + ((pWal->hdr.szPage&0x0001)<<16); +} + +/* +** The following is guaranteed when this function is called: +** +** a) the WRITER lock is held, +** b) the entire log file has been checkpointed, and +** c) any existing readers are reading exclusively from the database +** file - there are no readers that may attempt to read a frame from +** the log file. +** +** This function updates the shared-memory structures so that the next +** client to write to the database (which may be this one) does so by +** writing frames into the start of the log file. +** +** The value of parameter salt1 is used as the aSalt[1] value in the +** new wal-index header. It should be passed a pseudo-random value (i.e. +** one obtained from sqlite3_randomness()). +*/ +static void walRestartHdr(Wal *pWal, u32 salt1){ + volatile WalCkptInfo *pInfo = walCkptInfo(pWal); + int i; /* Loop counter */ + u32 *aSalt = pWal->hdr.aSalt; /* Big-endian salt values */ + pWal->nCkpt++; + pWal->hdr.mxFrame = 0; + sqlite3Put4byte((u8*)&aSalt[0], 1 + sqlite3Get4byte((u8*)&aSalt[0])); + memcpy(&pWal->hdr.aSalt[1], &salt1, 4); + walIndexWriteHdr(pWal); + pInfo->nBackfill = 0; + pInfo->nBackfillAttempted = 0; + pInfo->aReadMark[1] = 0; + for(i=2; iaReadMark[i] = READMARK_NOT_USED; + assert( pInfo->aReadMark[0]==0 ); +} + +/* +** Copy as much content as we can from the WAL back into the database file +** in response to an sqlite3_wal_checkpoint() request or the equivalent. +** +** The amount of information copies from WAL to database might be limited +** by active readers. This routine will never overwrite a database page +** that a concurrent reader might be using. +** +** All I/O barrier operations (a.k.a fsyncs) occur in this routine when +** SQLite is in WAL-mode in synchronous=NORMAL. That means that if +** checkpoints are always run by a background thread or background +** process, foreground threads will never block on a lengthy fsync call. +** +** Fsync is called on the WAL before writing content out of the WAL and +** into the database. This ensures that if the new content is persistent +** in the WAL and can be recovered following a power-loss or hard reset. +** +** Fsync is also called on the database file if (and only if) the entire +** WAL content is copied into the database file. This second fsync makes +** it safe to delete the WAL since the new content will persist in the +** database file. +** +** This routine uses and updates the nBackfill field of the wal-index header. +** This is the only routine that will increase the value of nBackfill. +** (A WAL reset or recovery will revert nBackfill to zero, but not increase +** its value.) +** +** The caller must be holding sufficient locks to ensure that no other +** checkpoint is running (in any other thread or process) at the same +** time. +*/ +static int walCheckpoint( + Wal *pWal, /* Wal connection */ + sqlite3 *db, /* Check for interrupts on this handle */ + int eMode, /* One of PASSIVE, FULL or RESTART */ + int (*xBusy)(void*), /* Function to call when busy */ + void *pBusyArg, /* Context argument for xBusyHandler */ + int sync_flags, /* Flags for OsSync() (or 0) */ + u8 *zBuf /* Temporary buffer to use */ +){ + int rc = SQLITE_OK; /* Return code */ + int szPage; /* Database page-size */ + WalIterator *pIter = 0; /* Wal iterator context */ + u32 iDbpage = 0; /* Next database page to write */ + u32 iFrame = 0; /* Wal frame containing data for iDbpage */ + u32 mxSafeFrame; /* Max frame that can be backfilled */ + u32 mxPage; /* Max database page to write */ + int i; /* Loop counter */ + volatile WalCkptInfo *pInfo; /* The checkpoint status information */ + + szPage = walPagesize(pWal); + testcase( szPage<=32768 ); + testcase( szPage>=65536 ); + pInfo = walCkptInfo(pWal); + if( pInfo->nBackfillhdr.mxFrame ){ + + /* EVIDENCE-OF: R-62920-47450 The busy-handler callback is never invoked + ** in the SQLITE_CHECKPOINT_PASSIVE mode. */ + assert( eMode!=SQLITE_CHECKPOINT_PASSIVE || xBusy==0 ); + + /* Compute in mxSafeFrame the index of the last frame of the WAL that is + ** safe to write into the database. Frames beyond mxSafeFrame might + ** overwrite database pages that are in use by active readers and thus + ** cannot be backfilled from the WAL. + */ + mxSafeFrame = pWal->hdr.mxFrame; + mxPage = pWal->hdr.nPage; + for(i=1; iaReadMark[i]; + if( mxSafeFrame>y ){ + assert( y<=pWal->hdr.mxFrame ); + rc = walBusyLock(pWal, xBusy, pBusyArg, WAL_READ_LOCK(i), 1); + if( rc==SQLITE_OK ){ + pInfo->aReadMark[i] = (i==1 ? mxSafeFrame : READMARK_NOT_USED); + walUnlockExclusive(pWal, WAL_READ_LOCK(i), 1); + }else if( rc==SQLITE_BUSY ){ + mxSafeFrame = y; + xBusy = 0; + }else{ + goto walcheckpoint_out; + } + } + } + + /* Allocate the iterator */ + if( pInfo->nBackfillnBackfill, &pIter); + assert( rc==SQLITE_OK || pIter==0 ); + } + + if( pIter + && (rc = walBusyLock(pWal, xBusy, pBusyArg, WAL_READ_LOCK(0),1))==SQLITE_OK + ){ + u32 nBackfill = pInfo->nBackfill; + + pInfo->nBackfillAttempted = mxSafeFrame; + + /* Sync the WAL to disk */ + rc = sqlite3OsSync(pWal->pWalFd, CKPT_SYNC_FLAGS(sync_flags)); + + /* If the database may grow as a result of this checkpoint, hint + ** about the eventual size of the db file to the VFS layer. + */ + if( rc==SQLITE_OK ){ + i64 nReq = ((i64)mxPage * szPage); + i64 nSize; /* Current size of database file */ + rc = sqlite3OsFileSize(pWal->pDbFd, &nSize); + if( rc==SQLITE_OK && nSizepDbFd, SQLITE_FCNTL_SIZE_HINT, &nReq); + } + } + + + /* Iterate through the contents of the WAL, copying data to the db file */ + while( rc==SQLITE_OK && 0==walIteratorNext(pIter, &iDbpage, &iFrame) ){ + i64 iOffset; + assert( walFramePgno(pWal, iFrame)==iDbpage ); + if( db->u1.isInterrupted ){ + rc = db->mallocFailed ? SQLITE_NOMEM_BKPT : SQLITE_INTERRUPT; + break; + } + if( iFrame<=nBackfill || iFrame>mxSafeFrame || iDbpage>mxPage ){ + continue; + } + iOffset = walFrameOffset(iFrame, szPage) + WAL_FRAME_HDRSIZE; + /* testcase( IS_BIG_INT(iOffset) ); // requires a 4GiB WAL file */ + rc = sqlite3OsRead(pWal->pWalFd, zBuf, szPage, iOffset); + if( rc!=SQLITE_OK ) break; + iOffset = (iDbpage-1)*(i64)szPage; + testcase( IS_BIG_INT(iOffset) ); + rc = sqlite3OsWrite(pWal->pDbFd, zBuf, szPage, iOffset); + if( rc!=SQLITE_OK ) break; + } + + /* If work was actually accomplished... */ + if( rc==SQLITE_OK ){ + if( mxSafeFrame==walIndexHdr(pWal)->mxFrame ){ + i64 szDb = pWal->hdr.nPage*(i64)szPage; + testcase( IS_BIG_INT(szDb) ); + rc = sqlite3OsTruncate(pWal->pDbFd, szDb); + if( rc==SQLITE_OK ){ + rc = sqlite3OsSync(pWal->pDbFd, CKPT_SYNC_FLAGS(sync_flags)); + } + } + if( rc==SQLITE_OK ){ + pInfo->nBackfill = mxSafeFrame; + } + } + + /* Release the reader lock held while backfilling */ + walUnlockExclusive(pWal, WAL_READ_LOCK(0), 1); + } + + if( rc==SQLITE_BUSY ){ + /* Reset the return code so as not to report a checkpoint failure + ** just because there are active readers. */ + rc = SQLITE_OK; + } + } + + /* If this is an SQLITE_CHECKPOINT_RESTART or TRUNCATE operation, and the + ** entire wal file has been copied into the database file, then block + ** until all readers have finished using the wal file. This ensures that + ** the next process to write to the database restarts the wal file. + */ + if( rc==SQLITE_OK && eMode!=SQLITE_CHECKPOINT_PASSIVE ){ + assert( pWal->writeLock ); + if( pInfo->nBackfillhdr.mxFrame ){ + rc = SQLITE_BUSY; + }else if( eMode>=SQLITE_CHECKPOINT_RESTART ){ + u32 salt1; + sqlite3_randomness(4, &salt1); + assert( pInfo->nBackfill==pWal->hdr.mxFrame ); + rc = walBusyLock(pWal, xBusy, pBusyArg, WAL_READ_LOCK(1), WAL_NREADER-1); + if( rc==SQLITE_OK ){ + if( eMode==SQLITE_CHECKPOINT_TRUNCATE ){ + /* IMPLEMENTATION-OF: R-44699-57140 This mode works the same way as + ** SQLITE_CHECKPOINT_RESTART with the addition that it also + ** truncates the log file to zero bytes just prior to a + ** successful return. + ** + ** In theory, it might be safe to do this without updating the + ** wal-index header in shared memory, as all subsequent reader or + ** writer clients should see that the entire log file has been + ** checkpointed and behave accordingly. This seems unsafe though, + ** as it would leave the system in a state where the contents of + ** the wal-index header do not match the contents of the + ** file-system. To avoid this, update the wal-index header to + ** indicate that the log file contains zero valid frames. */ + walRestartHdr(pWal, salt1); + rc = sqlite3OsTruncate(pWal->pWalFd, 0); + } + walUnlockExclusive(pWal, WAL_READ_LOCK(1), WAL_NREADER-1); + } + } + } + + walcheckpoint_out: + walIteratorFree(pIter); + return rc; +} + +/* +** If the WAL file is currently larger than nMax bytes in size, truncate +** it to exactly nMax bytes. If an error occurs while doing so, ignore it. +*/ +static void walLimitSize(Wal *pWal, i64 nMax){ + i64 sz; + int rx; + sqlite3BeginBenignMalloc(); + rx = sqlite3OsFileSize(pWal->pWalFd, &sz); + if( rx==SQLITE_OK && (sz > nMax ) ){ + rx = sqlite3OsTruncate(pWal->pWalFd, nMax); + } + sqlite3EndBenignMalloc(); + if( rx ){ + sqlite3_log(rx, "cannot limit WAL size: %s", pWal->zWalName); + } +} + +/* +** Close a connection to a log file. +*/ +SQLITE_PRIVATE int sqlite3WalClose( + Wal *pWal, /* Wal to close */ + sqlite3 *db, /* For interrupt flag */ + int sync_flags, /* Flags to pass to OsSync() (or 0) */ + int nBuf, + u8 *zBuf /* Buffer of at least nBuf bytes */ +){ + int rc = SQLITE_OK; + if( pWal ){ + int isDelete = 0; /* True to unlink wal and wal-index files */ + + /* If an EXCLUSIVE lock can be obtained on the database file (using the + ** ordinary, rollback-mode locking methods, this guarantees that the + ** connection associated with this log file is the only connection to + ** the database. In this case checkpoint the database and unlink both + ** the wal and wal-index files. + ** + ** The EXCLUSIVE lock is not released before returning. + */ + if( zBuf!=0 + && SQLITE_OK==(rc = sqlite3OsLock(pWal->pDbFd, SQLITE_LOCK_EXCLUSIVE)) + ){ + if( pWal->exclusiveMode==WAL_NORMAL_MODE ){ + pWal->exclusiveMode = WAL_EXCLUSIVE_MODE; + } + rc = sqlite3WalCheckpoint(pWal, db, + SQLITE_CHECKPOINT_PASSIVE, 0, 0, sync_flags, nBuf, zBuf, 0, 0 + ); + if( rc==SQLITE_OK ){ + int bPersist = -1; + sqlite3OsFileControlHint( + pWal->pDbFd, SQLITE_FCNTL_PERSIST_WAL, &bPersist + ); + if( bPersist!=1 ){ + /* Try to delete the WAL file if the checkpoint completed and + ** fsyned (rc==SQLITE_OK) and if we are not in persistent-wal + ** mode (!bPersist) */ + isDelete = 1; + }else if( pWal->mxWalSize>=0 ){ + /* Try to truncate the WAL file to zero bytes if the checkpoint + ** completed and fsynced (rc==SQLITE_OK) and we are in persistent + ** WAL mode (bPersist) and if the PRAGMA journal_size_limit is a + ** non-negative value (pWal->mxWalSize>=0). Note that we truncate + ** to zero bytes as truncating to the journal_size_limit might + ** leave a corrupt WAL file on disk. */ + walLimitSize(pWal, 0); + } + } + } + + walIndexClose(pWal, isDelete); + sqlite3OsClose(pWal->pWalFd); + if( isDelete ){ + sqlite3BeginBenignMalloc(); + sqlite3OsDelete(pWal->pVfs, pWal->zWalName, 0); + sqlite3EndBenignMalloc(); + } + WALTRACE(("WAL%p: closed\n", pWal)); + sqlite3_free((void *)pWal->apWiData); + sqlite3_free(pWal); + } + return rc; +} + +/* +** Try to read the wal-index header. Return 0 on success and 1 if +** there is a problem. +** +** The wal-index is in shared memory. Another thread or process might +** be writing the header at the same time this procedure is trying to +** read it, which might result in inconsistency. A dirty read is detected +** by verifying that both copies of the header are the same and also by +** a checksum on the header. +** +** If and only if the read is consistent and the header is different from +** pWal->hdr, then pWal->hdr is updated to the content of the new header +** and *pChanged is set to 1. +** +** If the checksum cannot be verified return non-zero. If the header +** is read successfully and the checksum verified, return zero. +*/ +static int walIndexTryHdr(Wal *pWal, int *pChanged){ + u32 aCksum[2]; /* Checksum on the header content */ + WalIndexHdr h1, h2; /* Two copies of the header content */ + WalIndexHdr volatile *aHdr; /* Header in shared memory */ + + /* The first page of the wal-index must be mapped at this point. */ + assert( pWal->nWiData>0 && pWal->apWiData[0] ); + + /* Read the header. This might happen concurrently with a write to the + ** same area of shared memory on a different CPU in a SMP, + ** meaning it is possible that an inconsistent snapshot is read + ** from the file. If this happens, return non-zero. + ** + ** There are two copies of the header at the beginning of the wal-index. + ** When reading, read [0] first then [1]. Writes are in the reverse order. + ** Memory barriers are used to prevent the compiler or the hardware from + ** reordering the reads and writes. + */ + aHdr = walIndexHdr(pWal); + memcpy(&h1, (void *)&aHdr[0], sizeof(h1)); + walShmBarrier(pWal); + memcpy(&h2, (void *)&aHdr[1], sizeof(h2)); + + if( memcmp(&h1, &h2, sizeof(h1))!=0 ){ + return 1; /* Dirty read */ + } + if( h1.isInit==0 ){ + return 1; /* Malformed header - probably all zeros */ + } + walChecksumBytes(1, (u8*)&h1, sizeof(h1)-sizeof(h1.aCksum), 0, aCksum); + if( aCksum[0]!=h1.aCksum[0] || aCksum[1]!=h1.aCksum[1] ){ + return 1; /* Checksum does not match */ + } + + if( memcmp(&pWal->hdr, &h1, sizeof(WalIndexHdr)) ){ + *pChanged = 1; + memcpy(&pWal->hdr, &h1, sizeof(WalIndexHdr)); + pWal->szPage = (pWal->hdr.szPage&0xfe00) + ((pWal->hdr.szPage&0x0001)<<16); + testcase( pWal->szPage<=32768 ); + testcase( pWal->szPage>=65536 ); + } + + /* The header was successfully read. Return zero. */ + return 0; +} + +/* +** This is the value that walTryBeginRead returns when it needs to +** be retried. +*/ +#define WAL_RETRY (-1) + +/* +** Read the wal-index header from the wal-index and into pWal->hdr. +** If the wal-header appears to be corrupt, try to reconstruct the +** wal-index from the WAL before returning. +** +** Set *pChanged to 1 if the wal-index header value in pWal->hdr is +** changed by this operation. If pWal->hdr is unchanged, set *pChanged +** to 0. +** +** If the wal-index header is successfully read, return SQLITE_OK. +** Otherwise an SQLite error code. +*/ +static int walIndexReadHdr(Wal *pWal, int *pChanged){ + int rc; /* Return code */ + int badHdr; /* True if a header read failed */ + volatile u32 *page0; /* Chunk of wal-index containing header */ + + /* Ensure that page 0 of the wal-index (the page that contains the + ** wal-index header) is mapped. Return early if an error occurs here. + */ + assert( pChanged ); + rc = walIndexPage(pWal, 0, &page0); + if( rc!=SQLITE_OK ){ + assert( rc!=SQLITE_READONLY ); /* READONLY changed to OK in walIndexPage */ + if( rc==SQLITE_READONLY_CANTINIT ){ + /* The SQLITE_READONLY_CANTINIT return means that the shared-memory + ** was openable but is not writable, and this thread is unable to + ** confirm that another write-capable connection has the shared-memory + ** open, and hence the content of the shared-memory is unreliable, + ** since the shared-memory might be inconsistent with the WAL file + ** and there is no writer on hand to fix it. */ + assert( page0==0 ); + assert( pWal->writeLock==0 ); + assert( pWal->readOnly & WAL_SHM_RDONLY ); + pWal->bShmUnreliable = 1; + pWal->exclusiveMode = WAL_HEAPMEMORY_MODE; + *pChanged = 1; + }else{ + return rc; /* Any other non-OK return is just an error */ + } + }else{ + /* page0 can be NULL if the SHM is zero bytes in size and pWal->writeLock + ** is zero, which prevents the SHM from growing */ + testcase( page0!=0 ); + } + assert( page0!=0 || pWal->writeLock==0 ); + + /* If the first page of the wal-index has been mapped, try to read the + ** wal-index header immediately, without holding any lock. This usually + ** works, but may fail if the wal-index header is corrupt or currently + ** being modified by another thread or process. + */ + badHdr = (page0 ? walIndexTryHdr(pWal, pChanged) : 1); + + /* If the first attempt failed, it might have been due to a race + ** with a writer. So get a WRITE lock and try again. + */ + assert( badHdr==0 || pWal->writeLock==0 ); + if( badHdr ){ + if( pWal->bShmUnreliable==0 && (pWal->readOnly & WAL_SHM_RDONLY) ){ + if( SQLITE_OK==(rc = walLockShared(pWal, WAL_WRITE_LOCK)) ){ + walUnlockShared(pWal, WAL_WRITE_LOCK); + rc = SQLITE_READONLY_RECOVERY; + } + }else if( SQLITE_OK==(rc = walLockExclusive(pWal, WAL_WRITE_LOCK, 1)) ){ + pWal->writeLock = 1; + if( SQLITE_OK==(rc = walIndexPage(pWal, 0, &page0)) ){ + badHdr = walIndexTryHdr(pWal, pChanged); + if( badHdr ){ + /* If the wal-index header is still malformed even while holding + ** a WRITE lock, it can only mean that the header is corrupted and + ** needs to be reconstructed. So run recovery to do exactly that. + */ + rc = walIndexRecover(pWal); + *pChanged = 1; + } + } + pWal->writeLock = 0; + walUnlockExclusive(pWal, WAL_WRITE_LOCK, 1); + } + } + + /* If the header is read successfully, check the version number to make + ** sure the wal-index was not constructed with some future format that + ** this version of SQLite cannot understand. + */ + if( badHdr==0 && pWal->hdr.iVersion!=WALINDEX_MAX_VERSION ){ + rc = SQLITE_CANTOPEN_BKPT; + } + if( pWal->bShmUnreliable ){ + if( rc!=SQLITE_OK ){ + walIndexClose(pWal, 0); + pWal->bShmUnreliable = 0; + assert( pWal->nWiData>0 && pWal->apWiData[0]==0 ); + /* walIndexRecover() might have returned SHORT_READ if a concurrent + ** writer truncated the WAL out from under it. If that happens, it + ** indicates that a writer has fixed the SHM file for us, so retry */ + if( rc==SQLITE_IOERR_SHORT_READ ) rc = WAL_RETRY; + } + pWal->exclusiveMode = WAL_NORMAL_MODE; + } + + return rc; +} + +/* +** Open a transaction in a connection where the shared-memory is read-only +** and where we cannot verify that there is a separate write-capable connection +** on hand to keep the shared-memory up-to-date with the WAL file. +** +** This can happen, for example, when the shared-memory is implemented by +** memory-mapping a *-shm file, where a prior writer has shut down and +** left the *-shm file on disk, and now the present connection is trying +** to use that database but lacks write permission on the *-shm file. +** Other scenarios are also possible, depending on the VFS implementation. +** +** Precondition: +** +** The *-wal file has been read and an appropriate wal-index has been +** constructed in pWal->apWiData[] using heap memory instead of shared +** memory. +** +** If this function returns SQLITE_OK, then the read transaction has +** been successfully opened. In this case output variable (*pChanged) +** is set to true before returning if the caller should discard the +** contents of the page cache before proceeding. Or, if it returns +** WAL_RETRY, then the heap memory wal-index has been discarded and +** the caller should retry opening the read transaction from the +** beginning (including attempting to map the *-shm file). +** +** If an error occurs, an SQLite error code is returned. +*/ +static int walBeginShmUnreliable(Wal *pWal, int *pChanged){ + i64 szWal; /* Size of wal file on disk in bytes */ + i64 iOffset; /* Current offset when reading wal file */ + u8 aBuf[WAL_HDRSIZE]; /* Buffer to load WAL header into */ + u8 *aFrame = 0; /* Malloc'd buffer to load entire frame */ + int szFrame; /* Number of bytes in buffer aFrame[] */ + u8 *aData; /* Pointer to data part of aFrame buffer */ + volatile void *pDummy; /* Dummy argument for xShmMap */ + int rc; /* Return code */ + u32 aSaveCksum[2]; /* Saved copy of pWal->hdr.aFrameCksum */ + + assert( pWal->bShmUnreliable ); + assert( pWal->readOnly & WAL_SHM_RDONLY ); + assert( pWal->nWiData>0 && pWal->apWiData[0] ); + + /* Take WAL_READ_LOCK(0). This has the effect of preventing any + ** writers from running a checkpoint, but does not stop them + ** from running recovery. */ + rc = walLockShared(pWal, WAL_READ_LOCK(0)); + if( rc!=SQLITE_OK ){ + if( rc==SQLITE_BUSY ) rc = WAL_RETRY; + goto begin_unreliable_shm_out; + } + pWal->readLock = 0; + + /* Check to see if a separate writer has attached to the shared-memory area, + ** thus making the shared-memory "reliable" again. Do this by invoking + ** the xShmMap() routine of the VFS and looking to see if the return + ** is SQLITE_READONLY instead of SQLITE_READONLY_CANTINIT. + ** + ** If the shared-memory is now "reliable" return WAL_RETRY, which will + ** cause the heap-memory WAL-index to be discarded and the actual + ** shared memory to be used in its place. + ** + ** This step is important because, even though this connection is holding + ** the WAL_READ_LOCK(0) which prevents a checkpoint, a writer might + ** have already checkpointed the WAL file and, while the current + ** is active, wrap the WAL and start overwriting frames that this + ** process wants to use. + ** + ** Once sqlite3OsShmMap() has been called for an sqlite3_file and has + ** returned any SQLITE_READONLY value, it must return only SQLITE_READONLY + ** or SQLITE_READONLY_CANTINIT or some error for all subsequent invocations, + ** even if some external agent does a "chmod" to make the shared-memory + ** writable by us, until sqlite3OsShmUnmap() has been called. + ** This is a requirement on the VFS implementation. + */ + rc = sqlite3OsShmMap(pWal->pDbFd, 0, WALINDEX_PGSZ, 0, &pDummy); + assert( rc!=SQLITE_OK ); /* SQLITE_OK not possible for read-only connection */ + if( rc!=SQLITE_READONLY_CANTINIT ){ + rc = (rc==SQLITE_READONLY ? WAL_RETRY : rc); + goto begin_unreliable_shm_out; + } + + /* We reach this point only if the real shared-memory is still unreliable. + ** Assume the in-memory WAL-index substitute is correct and load it + ** into pWal->hdr. + */ + memcpy(&pWal->hdr, (void*)walIndexHdr(pWal), sizeof(WalIndexHdr)); + + /* Make sure some writer hasn't come in and changed the WAL file out + ** from under us, then disconnected, while we were not looking. + */ + rc = sqlite3OsFileSize(pWal->pWalFd, &szWal); + if( rc!=SQLITE_OK ){ + goto begin_unreliable_shm_out; + } + if( szWalhdr.mxFrame==0 ? SQLITE_OK : WAL_RETRY); + goto begin_unreliable_shm_out; + } + + /* Check the salt keys at the start of the wal file still match. */ + rc = sqlite3OsRead(pWal->pWalFd, aBuf, WAL_HDRSIZE, 0); + if( rc!=SQLITE_OK ){ + goto begin_unreliable_shm_out; + } + if( memcmp(&pWal->hdr.aSalt, &aBuf[16], 8) ){ + /* Some writer has wrapped the WAL file while we were not looking. + ** Return WAL_RETRY which will cause the in-memory WAL-index to be + ** rebuilt. */ + rc = WAL_RETRY; + goto begin_unreliable_shm_out; + } + + /* Allocate a buffer to read frames into */ + szFrame = pWal->hdr.szPage + WAL_FRAME_HDRSIZE; + aFrame = (u8 *)sqlite3_malloc64(szFrame); + if( aFrame==0 ){ + rc = SQLITE_NOMEM_BKPT; + goto begin_unreliable_shm_out; + } + aData = &aFrame[WAL_FRAME_HDRSIZE]; + + /* Check to see if a complete transaction has been appended to the + ** wal file since the heap-memory wal-index was created. If so, the + ** heap-memory wal-index is discarded and WAL_RETRY returned to + ** the caller. */ + aSaveCksum[0] = pWal->hdr.aFrameCksum[0]; + aSaveCksum[1] = pWal->hdr.aFrameCksum[1]; + for(iOffset=walFrameOffset(pWal->hdr.mxFrame+1, pWal->hdr.szPage); + iOffset+szFrame<=szWal; + iOffset+=szFrame + ){ + u32 pgno; /* Database page number for frame */ + u32 nTruncate; /* dbsize field from frame header */ + + /* Read and decode the next log frame. */ + rc = sqlite3OsRead(pWal->pWalFd, aFrame, szFrame, iOffset); + if( rc!=SQLITE_OK ) break; + if( !walDecodeFrame(pWal, &pgno, &nTruncate, aData, aFrame) ) break; + + /* If nTruncate is non-zero, then a complete transaction has been + ** appended to this wal file. Set rc to WAL_RETRY and break out of + ** the loop. */ + if( nTruncate ){ + rc = WAL_RETRY; + break; + } + } + pWal->hdr.aFrameCksum[0] = aSaveCksum[0]; + pWal->hdr.aFrameCksum[1] = aSaveCksum[1]; + + begin_unreliable_shm_out: + sqlite3_free(aFrame); + if( rc!=SQLITE_OK ){ + int i; + for(i=0; inWiData; i++){ + sqlite3_free((void*)pWal->apWiData[i]); + pWal->apWiData[i] = 0; + } + pWal->bShmUnreliable = 0; + sqlite3WalEndReadTransaction(pWal); + *pChanged = 1; + } + return rc; +} + +/* +** Attempt to start a read transaction. This might fail due to a race or +** other transient condition. When that happens, it returns WAL_RETRY to +** indicate to the caller that it is safe to retry immediately. +** +** On success return SQLITE_OK. On a permanent failure (such an +** I/O error or an SQLITE_BUSY because another process is running +** recovery) return a positive error code. +** +** The useWal parameter is true to force the use of the WAL and disable +** the case where the WAL is bypassed because it has been completely +** checkpointed. If useWal==0 then this routine calls walIndexReadHdr() +** to make a copy of the wal-index header into pWal->hdr. If the +** wal-index header has changed, *pChanged is set to 1 (as an indication +** to the caller that the local page cache is obsolete and needs to be +** flushed.) When useWal==1, the wal-index header is assumed to already +** be loaded and the pChanged parameter is unused. +** +** The caller must set the cnt parameter to the number of prior calls to +** this routine during the current read attempt that returned WAL_RETRY. +** This routine will start taking more aggressive measures to clear the +** race conditions after multiple WAL_RETRY returns, and after an excessive +** number of errors will ultimately return SQLITE_PROTOCOL. The +** SQLITE_PROTOCOL return indicates that some other process has gone rogue +** and is not honoring the locking protocol. There is a vanishingly small +** chance that SQLITE_PROTOCOL could be returned because of a run of really +** bad luck when there is lots of contention for the wal-index, but that +** possibility is so small that it can be safely neglected, we believe. +** +** On success, this routine obtains a read lock on +** WAL_READ_LOCK(pWal->readLock). The pWal->readLock integer is +** in the range 0 <= pWal->readLock < WAL_NREADER. If pWal->readLock==(-1) +** that means the Wal does not hold any read lock. The reader must not +** access any database page that is modified by a WAL frame up to and +** including frame number aReadMark[pWal->readLock]. The reader will +** use WAL frames up to and including pWal->hdr.mxFrame if pWal->readLock>0 +** Or if pWal->readLock==0, then the reader will ignore the WAL +** completely and get all content directly from the database file. +** If the useWal parameter is 1 then the WAL will never be ignored and +** this routine will always set pWal->readLock>0 on success. +** When the read transaction is completed, the caller must release the +** lock on WAL_READ_LOCK(pWal->readLock) and set pWal->readLock to -1. +** +** This routine uses the nBackfill and aReadMark[] fields of the header +** to select a particular WAL_READ_LOCK() that strives to let the +** checkpoint process do as much work as possible. This routine might +** update values of the aReadMark[] array in the header, but if it does +** so it takes care to hold an exclusive lock on the corresponding +** WAL_READ_LOCK() while changing values. +*/ +static int walTryBeginRead(Wal *pWal, int *pChanged, int useWal, int cnt){ + volatile WalCkptInfo *pInfo; /* Checkpoint information in wal-index */ + u32 mxReadMark; /* Largest aReadMark[] value */ + int mxI; /* Index of largest aReadMark[] value */ + int i; /* Loop counter */ + int rc = SQLITE_OK; /* Return code */ + u32 mxFrame; /* Wal frame to lock to */ + + assert( pWal->readLock<0 ); /* Not currently locked */ + + /* useWal may only be set for read/write connections */ + assert( (pWal->readOnly & WAL_SHM_RDONLY)==0 || useWal==0 ); + + /* Take steps to avoid spinning forever if there is a protocol error. + ** + ** Circumstances that cause a RETRY should only last for the briefest + ** instances of time. No I/O or other system calls are done while the + ** locks are held, so the locks should not be held for very long. But + ** if we are unlucky, another process that is holding a lock might get + ** paged out or take a page-fault that is time-consuming to resolve, + ** during the few nanoseconds that it is holding the lock. In that case, + ** it might take longer than normal for the lock to free. + ** + ** After 5 RETRYs, we begin calling sqlite3OsSleep(). The first few + ** calls to sqlite3OsSleep() have a delay of 1 microsecond. Really this + ** is more of a scheduler yield than an actual delay. But on the 10th + ** an subsequent retries, the delays start becoming longer and longer, + ** so that on the 100th (and last) RETRY we delay for 323 milliseconds. + ** The total delay time before giving up is less than 10 seconds. + */ + if( cnt>5 ){ + int nDelay = 1; /* Pause time in microseconds */ + if( cnt>100 ){ + VVA_ONLY( pWal->lockError = 1; ) + return SQLITE_PROTOCOL; + } + if( cnt>=10 ) nDelay = (cnt-9)*(cnt-9)*39; + sqlite3OsSleep(pWal->pVfs, nDelay); + } + + if( !useWal ){ + assert( rc==SQLITE_OK ); + if( pWal->bShmUnreliable==0 ){ + rc = walIndexReadHdr(pWal, pChanged); + } + if( rc==SQLITE_BUSY ){ + /* If there is not a recovery running in another thread or process + ** then convert BUSY errors to WAL_RETRY. If recovery is known to + ** be running, convert BUSY to BUSY_RECOVERY. There is a race here + ** which might cause WAL_RETRY to be returned even if BUSY_RECOVERY + ** would be technically correct. But the race is benign since with + ** WAL_RETRY this routine will be called again and will probably be + ** right on the second iteration. + */ + if( pWal->apWiData[0]==0 ){ + /* This branch is taken when the xShmMap() method returns SQLITE_BUSY. + ** We assume this is a transient condition, so return WAL_RETRY. The + ** xShmMap() implementation used by the default unix and win32 VFS + ** modules may return SQLITE_BUSY due to a race condition in the + ** code that determines whether or not the shared-memory region + ** must be zeroed before the requested page is returned. + */ + rc = WAL_RETRY; + }else if( SQLITE_OK==(rc = walLockShared(pWal, WAL_RECOVER_LOCK)) ){ + walUnlockShared(pWal, WAL_RECOVER_LOCK); + rc = WAL_RETRY; + }else if( rc==SQLITE_BUSY ){ + rc = SQLITE_BUSY_RECOVERY; + } + } + if( rc!=SQLITE_OK ){ + return rc; + } + else if( pWal->bShmUnreliable ){ + return walBeginShmUnreliable(pWal, pChanged); + } + } + + assert( pWal->nWiData>0 ); + assert( pWal->apWiData[0]!=0 ); + pInfo = walCkptInfo(pWal); + if( !useWal && pInfo->nBackfill==pWal->hdr.mxFrame +#ifdef SQLITE_ENABLE_SNAPSHOT + && (pWal->pSnapshot==0 || pWal->hdr.mxFrame==0) +#endif + ){ + /* The WAL has been completely backfilled (or it is empty). + ** and can be safely ignored. + */ + rc = walLockShared(pWal, WAL_READ_LOCK(0)); + walShmBarrier(pWal); + if( rc==SQLITE_OK ){ + if( memcmp((void *)walIndexHdr(pWal), &pWal->hdr, sizeof(WalIndexHdr)) ){ + /* It is not safe to allow the reader to continue here if frames + ** may have been appended to the log before READ_LOCK(0) was obtained. + ** When holding READ_LOCK(0), the reader ignores the entire log file, + ** which implies that the database file contains a trustworthy + ** snapshot. Since holding READ_LOCK(0) prevents a checkpoint from + ** happening, this is usually correct. + ** + ** However, if frames have been appended to the log (or if the log + ** is wrapped and written for that matter) before the READ_LOCK(0) + ** is obtained, that is not necessarily true. A checkpointer may + ** have started to backfill the appended frames but crashed before + ** it finished. Leaving a corrupt image in the database file. + */ + walUnlockShared(pWal, WAL_READ_LOCK(0)); + return WAL_RETRY; + } + pWal->readLock = 0; + return SQLITE_OK; + }else if( rc!=SQLITE_BUSY ){ + return rc; + } + } + + /* If we get this far, it means that the reader will want to use + ** the WAL to get at content from recent commits. The job now is + ** to select one of the aReadMark[] entries that is closest to + ** but not exceeding pWal->hdr.mxFrame and lock that entry. + */ + mxReadMark = 0; + mxI = 0; + mxFrame = pWal->hdr.mxFrame; +#ifdef SQLITE_ENABLE_SNAPSHOT + if( pWal->pSnapshot && pWal->pSnapshot->mxFramepSnapshot->mxFrame; + } +#endif + for(i=1; iaReadMark+i); + if( mxReadMark<=thisMark && thisMark<=mxFrame ){ + assert( thisMark!=READMARK_NOT_USED ); + mxReadMark = thisMark; + mxI = i; + } + } + if( (pWal->readOnly & WAL_SHM_RDONLY)==0 + && (mxReadMarkaReadMark+i,mxFrame); + mxI = i; + walUnlockExclusive(pWal, WAL_READ_LOCK(i), 1); + break; + }else if( rc!=SQLITE_BUSY ){ + return rc; + } + } + } + if( mxI==0 ){ + assert( rc==SQLITE_BUSY || (pWal->readOnly & WAL_SHM_RDONLY)!=0 ); + return rc==SQLITE_BUSY ? WAL_RETRY : SQLITE_READONLY_CANTINIT; + } + + rc = walLockShared(pWal, WAL_READ_LOCK(mxI)); + if( rc ){ + return rc==SQLITE_BUSY ? WAL_RETRY : rc; + } + /* Now that the read-lock has been obtained, check that neither the + ** value in the aReadMark[] array or the contents of the wal-index + ** header have changed. + ** + ** It is necessary to check that the wal-index header did not change + ** between the time it was read and when the shared-lock was obtained + ** on WAL_READ_LOCK(mxI) was obtained to account for the possibility + ** that the log file may have been wrapped by a writer, or that frames + ** that occur later in the log than pWal->hdr.mxFrame may have been + ** copied into the database by a checkpointer. If either of these things + ** happened, then reading the database with the current value of + ** pWal->hdr.mxFrame risks reading a corrupted snapshot. So, retry + ** instead. + ** + ** Before checking that the live wal-index header has not changed + ** since it was read, set Wal.minFrame to the first frame in the wal + ** file that has not yet been checkpointed. This client will not need + ** to read any frames earlier than minFrame from the wal file - they + ** can be safely read directly from the database file. + ** + ** Because a ShmBarrier() call is made between taking the copy of + ** nBackfill and checking that the wal-header in shared-memory still + ** matches the one cached in pWal->hdr, it is guaranteed that the + ** checkpointer that set nBackfill was not working with a wal-index + ** header newer than that cached in pWal->hdr. If it were, that could + ** cause a problem. The checkpointer could omit to checkpoint + ** a version of page X that lies before pWal->minFrame (call that version + ** A) on the basis that there is a newer version (version B) of the same + ** page later in the wal file. But if version B happens to like past + ** frame pWal->hdr.mxFrame - then the client would incorrectly assume + ** that it can read version A from the database file. However, since + ** we can guarantee that the checkpointer that set nBackfill could not + ** see any pages past pWal->hdr.mxFrame, this problem does not come up. + */ + pWal->minFrame = AtomicLoad(&pInfo->nBackfill)+1; + walShmBarrier(pWal); + if( AtomicLoad(pInfo->aReadMark+mxI)!=mxReadMark + || memcmp((void *)walIndexHdr(pWal), &pWal->hdr, sizeof(WalIndexHdr)) + ){ + walUnlockShared(pWal, WAL_READ_LOCK(mxI)); + return WAL_RETRY; + }else{ + assert( mxReadMark<=pWal->hdr.mxFrame ); + pWal->readLock = (i16)mxI; + } + return rc; +} + +#ifdef SQLITE_ENABLE_SNAPSHOT +/* +** Attempt to reduce the value of the WalCkptInfo.nBackfillAttempted +** variable so that older snapshots can be accessed. To do this, loop +** through all wal frames from nBackfillAttempted to (nBackfill+1), +** comparing their content to the corresponding page with the database +** file, if any. Set nBackfillAttempted to the frame number of the +** first frame for which the wal file content matches the db file. +** +** This is only really safe if the file-system is such that any page +** writes made by earlier checkpointers were atomic operations, which +** is not always true. It is also possible that nBackfillAttempted +** may be left set to a value larger than expected, if a wal frame +** contains content that duplicate of an earlier version of the same +** page. +** +** SQLITE_OK is returned if successful, or an SQLite error code if an +** error occurs. It is not an error if nBackfillAttempted cannot be +** decreased at all. +*/ +SQLITE_PRIVATE int sqlite3WalSnapshotRecover(Wal *pWal){ + int rc; + + assert( pWal->readLock>=0 ); + rc = walLockExclusive(pWal, WAL_CKPT_LOCK, 1); + if( rc==SQLITE_OK ){ + volatile WalCkptInfo *pInfo = walCkptInfo(pWal); + int szPage = (int)pWal->szPage; + i64 szDb; /* Size of db file in bytes */ + + rc = sqlite3OsFileSize(pWal->pDbFd, &szDb); + if( rc==SQLITE_OK ){ + void *pBuf1 = sqlite3_malloc(szPage); + void *pBuf2 = sqlite3_malloc(szPage); + if( pBuf1==0 || pBuf2==0 ){ + rc = SQLITE_NOMEM; + }else{ + u32 i = pInfo->nBackfillAttempted; + for(i=pInfo->nBackfillAttempted; i>pInfo->nBackfill; i--){ + WalHashLoc sLoc; /* Hash table location */ + u32 pgno; /* Page number in db file */ + i64 iDbOff; /* Offset of db file entry */ + i64 iWalOff; /* Offset of wal file entry */ + + rc = walHashGet(pWal, walFramePage(i), &sLoc); + if( rc!=SQLITE_OK ) break; + pgno = sLoc.aPgno[i-sLoc.iZero]; + iDbOff = (i64)(pgno-1) * szPage; + + if( iDbOff+szPage<=szDb ){ + iWalOff = walFrameOffset(i, szPage) + WAL_FRAME_HDRSIZE; + rc = sqlite3OsRead(pWal->pWalFd, pBuf1, szPage, iWalOff); + + if( rc==SQLITE_OK ){ + rc = sqlite3OsRead(pWal->pDbFd, pBuf2, szPage, iDbOff); + } + + if( rc!=SQLITE_OK || 0==memcmp(pBuf1, pBuf2, szPage) ){ + break; + } + } + + pInfo->nBackfillAttempted = i-1; + } + } + + sqlite3_free(pBuf1); + sqlite3_free(pBuf2); + } + walUnlockExclusive(pWal, WAL_CKPT_LOCK, 1); + } + + return rc; +} +#endif /* SQLITE_ENABLE_SNAPSHOT */ + +/* +** Begin a read transaction on the database. +** +** This routine used to be called sqlite3OpenSnapshot() and with good reason: +** it takes a snapshot of the state of the WAL and wal-index for the current +** instant in time. The current thread will continue to use this snapshot. +** Other threads might append new content to the WAL and wal-index but +** that extra content is ignored by the current thread. +** +** If the database contents have changes since the previous read +** transaction, then *pChanged is set to 1 before returning. The +** Pager layer will use this to know that its cache is stale and +** needs to be flushed. +*/ +SQLITE_PRIVATE int sqlite3WalBeginReadTransaction(Wal *pWal, int *pChanged){ + int rc; /* Return code */ + int cnt = 0; /* Number of TryBeginRead attempts */ + +#ifdef SQLITE_ENABLE_SNAPSHOT + int bChanged = 0; + WalIndexHdr *pSnapshot = pWal->pSnapshot; + if( pSnapshot && memcmp(pSnapshot, &pWal->hdr, sizeof(WalIndexHdr))!=0 ){ + bChanged = 1; + } +#endif + + do{ + rc = walTryBeginRead(pWal, pChanged, 0, ++cnt); + }while( rc==WAL_RETRY ); + testcase( (rc&0xff)==SQLITE_BUSY ); + testcase( (rc&0xff)==SQLITE_IOERR ); + testcase( rc==SQLITE_PROTOCOL ); + testcase( rc==SQLITE_OK ); + +#ifdef SQLITE_ENABLE_SNAPSHOT + if( rc==SQLITE_OK ){ + if( pSnapshot && memcmp(pSnapshot, &pWal->hdr, sizeof(WalIndexHdr))!=0 ){ + /* At this point the client has a lock on an aReadMark[] slot holding + ** a value equal to or smaller than pSnapshot->mxFrame, but pWal->hdr + ** is populated with the wal-index header corresponding to the head + ** of the wal file. Verify that pSnapshot is still valid before + ** continuing. Reasons why pSnapshot might no longer be valid: + ** + ** (1) The WAL file has been reset since the snapshot was taken. + ** In this case, the salt will have changed. + ** + ** (2) A checkpoint as been attempted that wrote frames past + ** pSnapshot->mxFrame into the database file. Note that the + ** checkpoint need not have completed for this to cause problems. + */ + volatile WalCkptInfo *pInfo = walCkptInfo(pWal); + + assert( pWal->readLock>0 || pWal->hdr.mxFrame==0 ); + assert( pInfo->aReadMark[pWal->readLock]<=pSnapshot->mxFrame ); + + /* It is possible that there is a checkpointer thread running + ** concurrent with this code. If this is the case, it may be that the + ** checkpointer has already determined that it will checkpoint + ** snapshot X, where X is later in the wal file than pSnapshot, but + ** has not yet set the pInfo->nBackfillAttempted variable to indicate + ** its intent. To avoid the race condition this leads to, ensure that + ** there is no checkpointer process by taking a shared CKPT lock + ** before checking pInfo->nBackfillAttempted. + ** + ** TODO: Does the aReadMark[] lock prevent a checkpointer from doing + ** this already? + */ + rc = walLockShared(pWal, WAL_CKPT_LOCK); + + if( rc==SQLITE_OK ){ + /* Check that the wal file has not been wrapped. Assuming that it has + ** not, also check that no checkpointer has attempted to checkpoint any + ** frames beyond pSnapshot->mxFrame. If either of these conditions are + ** true, return SQLITE_ERROR_SNAPSHOT. Otherwise, overwrite pWal->hdr + ** with *pSnapshot and set *pChanged as appropriate for opening the + ** snapshot. */ + if( !memcmp(pSnapshot->aSalt, pWal->hdr.aSalt, sizeof(pWal->hdr.aSalt)) + && pSnapshot->mxFrame>=pInfo->nBackfillAttempted + ){ + assert( pWal->readLock>0 ); + memcpy(&pWal->hdr, pSnapshot, sizeof(WalIndexHdr)); + *pChanged = bChanged; + }else{ + rc = SQLITE_ERROR_SNAPSHOT; + } + + /* Release the shared CKPT lock obtained above. */ + walUnlockShared(pWal, WAL_CKPT_LOCK); + pWal->minFrame = 1; + } + + + if( rc!=SQLITE_OK ){ + sqlite3WalEndReadTransaction(pWal); + } + } + } +#endif + return rc; +} + +/* +** Finish with a read transaction. All this does is release the +** read-lock. +*/ +SQLITE_PRIVATE void sqlite3WalEndReadTransaction(Wal *pWal){ + sqlite3WalEndWriteTransaction(pWal); + if( pWal->readLock>=0 ){ + walUnlockShared(pWal, WAL_READ_LOCK(pWal->readLock)); + pWal->readLock = -1; + } +} + +/* +** Search the wal file for page pgno. If found, set *piRead to the frame that +** contains the page. Otherwise, if pgno is not in the wal file, set *piRead +** to zero. +** +** Return SQLITE_OK if successful, or an error code if an error occurs. If an +** error does occur, the final value of *piRead is undefined. +*/ +SQLITE_PRIVATE int sqlite3WalFindFrame( + Wal *pWal, /* WAL handle */ + Pgno pgno, /* Database page number to read data for */ + u32 *piRead /* OUT: Frame number (or zero) */ +){ + u32 iRead = 0; /* If !=0, WAL frame to return data from */ + u32 iLast = pWal->hdr.mxFrame; /* Last page in WAL for this reader */ + int iHash; /* Used to loop through N hash tables */ + int iMinHash; + + /* This routine is only be called from within a read transaction. */ + assert( pWal->readLock>=0 || pWal->lockError ); + + /* If the "last page" field of the wal-index header snapshot is 0, then + ** no data will be read from the wal under any circumstances. Return early + ** in this case as an optimization. Likewise, if pWal->readLock==0, + ** then the WAL is ignored by the reader so return early, as if the + ** WAL were empty. + */ + if( iLast==0 || (pWal->readLock==0 && pWal->bShmUnreliable==0) ){ + *piRead = 0; + return SQLITE_OK; + } + + /* Search the hash table or tables for an entry matching page number + ** pgno. Each iteration of the following for() loop searches one + ** hash table (each hash table indexes up to HASHTABLE_NPAGE frames). + ** + ** This code might run concurrently to the code in walIndexAppend() + ** that adds entries to the wal-index (and possibly to this hash + ** table). This means the value just read from the hash + ** slot (aHash[iKey]) may have been added before or after the + ** current read transaction was opened. Values added after the + ** read transaction was opened may have been written incorrectly - + ** i.e. these slots may contain garbage data. However, we assume + ** that any slots written before the current read transaction was + ** opened remain unmodified. + ** + ** For the reasons above, the if(...) condition featured in the inner + ** loop of the following block is more stringent that would be required + ** if we had exclusive access to the hash-table: + ** + ** (aPgno[iFrame]==pgno): + ** This condition filters out normal hash-table collisions. + ** + ** (iFrame<=iLast): + ** This condition filters out entries that were added to the hash + ** table after the current read-transaction had started. + */ + iMinHash = walFramePage(pWal->minFrame); + for(iHash=walFramePage(iLast); iHash>=iMinHash; iHash--){ + WalHashLoc sLoc; /* Hash table location */ + int iKey; /* Hash slot index */ + int nCollide; /* Number of hash collisions remaining */ + int rc; /* Error code */ + + rc = walHashGet(pWal, iHash, &sLoc); + if( rc!=SQLITE_OK ){ + return rc; + } + nCollide = HASHTABLE_NSLOT; + for(iKey=walHash(pgno); sLoc.aHash[iKey]; iKey=walNextHash(iKey)){ + u32 iH = sLoc.aHash[iKey]; + u32 iFrame = iH + sLoc.iZero; + if( iFrame<=iLast && iFrame>=pWal->minFrame && sLoc.aPgno[iH]==pgno ){ + assert( iFrame>iRead || CORRUPT_DB ); + iRead = iFrame; + } + if( (nCollide--)==0 ){ + return SQLITE_CORRUPT_BKPT; + } + } + if( iRead ) break; + } + +#ifdef SQLITE_ENABLE_EXPENSIVE_ASSERT + /* If expensive assert() statements are available, do a linear search + ** of the wal-index file content. Make sure the results agree with the + ** result obtained using the hash indexes above. */ + { + u32 iRead2 = 0; + u32 iTest; + assert( pWal->bShmUnreliable || pWal->minFrame>0 ); + for(iTest=iLast; iTest>=pWal->minFrame && iTest>0; iTest--){ + if( walFramePgno(pWal, iTest)==pgno ){ + iRead2 = iTest; + break; + } + } + assert( iRead==iRead2 ); + } +#endif + + *piRead = iRead; + return SQLITE_OK; +} + +/* +** Read the contents of frame iRead from the wal file into buffer pOut +** (which is nOut bytes in size). Return SQLITE_OK if successful, or an +** error code otherwise. +*/ +SQLITE_PRIVATE int sqlite3WalReadFrame( + Wal *pWal, /* WAL handle */ + u32 iRead, /* Frame to read */ + int nOut, /* Size of buffer pOut in bytes */ + u8 *pOut /* Buffer to write page data to */ +){ + int sz; + i64 iOffset; + sz = pWal->hdr.szPage; + sz = (sz&0xfe00) + ((sz&0x0001)<<16); + testcase( sz<=32768 ); + testcase( sz>=65536 ); + iOffset = walFrameOffset(iRead, sz) + WAL_FRAME_HDRSIZE; + /* testcase( IS_BIG_INT(iOffset) ); // requires a 4GiB WAL */ + return sqlite3OsRead(pWal->pWalFd, pOut, (nOut>sz ? sz : nOut), iOffset); +} + +/* +** Return the size of the database in pages (or zero, if unknown). +*/ +SQLITE_PRIVATE Pgno sqlite3WalDbsize(Wal *pWal){ + if( pWal && ALWAYS(pWal->readLock>=0) ){ + return pWal->hdr.nPage; + } + return 0; +} + + +/* +** This function starts a write transaction on the WAL. +** +** A read transaction must have already been started by a prior call +** to sqlite3WalBeginReadTransaction(). +** +** If another thread or process has written into the database since +** the read transaction was started, then it is not possible for this +** thread to write as doing so would cause a fork. So this routine +** returns SQLITE_BUSY in that case and no write transaction is started. +** +** There can only be a single writer active at a time. +*/ +SQLITE_PRIVATE int sqlite3WalBeginWriteTransaction(Wal *pWal){ + int rc; + + /* Cannot start a write transaction without first holding a read + ** transaction. */ + assert( pWal->readLock>=0 ); + assert( pWal->writeLock==0 && pWal->iReCksum==0 ); + + if( pWal->readOnly ){ + return SQLITE_READONLY; + } + + /* Only one writer allowed at a time. Get the write lock. Return + ** SQLITE_BUSY if unable. + */ + rc = walLockExclusive(pWal, WAL_WRITE_LOCK, 1); + if( rc ){ + return rc; + } + pWal->writeLock = 1; + + /* If another connection has written to the database file since the + ** time the read transaction on this connection was started, then + ** the write is disallowed. + */ + if( memcmp(&pWal->hdr, (void *)walIndexHdr(pWal), sizeof(WalIndexHdr))!=0 ){ + walUnlockExclusive(pWal, WAL_WRITE_LOCK, 1); + pWal->writeLock = 0; + rc = SQLITE_BUSY_SNAPSHOT; + } + + return rc; +} + +/* +** End a write transaction. The commit has already been done. This +** routine merely releases the lock. +*/ +SQLITE_PRIVATE int sqlite3WalEndWriteTransaction(Wal *pWal){ + if( pWal->writeLock ){ + walUnlockExclusive(pWal, WAL_WRITE_LOCK, 1); + pWal->writeLock = 0; + pWal->iReCksum = 0; + pWal->truncateOnCommit = 0; + } + return SQLITE_OK; +} + +/* +** If any data has been written (but not committed) to the log file, this +** function moves the write-pointer back to the start of the transaction. +** +** Additionally, the callback function is invoked for each frame written +** to the WAL since the start of the transaction. If the callback returns +** other than SQLITE_OK, it is not invoked again and the error code is +** returned to the caller. +** +** Otherwise, if the callback function does not return an error, this +** function returns SQLITE_OK. +*/ +SQLITE_PRIVATE int sqlite3WalUndo(Wal *pWal, int (*xUndo)(void *, Pgno), void *pUndoCtx){ + int rc = SQLITE_OK; + if( ALWAYS(pWal->writeLock) ){ + Pgno iMax = pWal->hdr.mxFrame; + Pgno iFrame; + + /* Restore the clients cache of the wal-index header to the state it + ** was in before the client began writing to the database. + */ + memcpy(&pWal->hdr, (void *)walIndexHdr(pWal), sizeof(WalIndexHdr)); + + for(iFrame=pWal->hdr.mxFrame+1; + ALWAYS(rc==SQLITE_OK) && iFrame<=iMax; + iFrame++ + ){ + /* This call cannot fail. Unless the page for which the page number + ** is passed as the second argument is (a) in the cache and + ** (b) has an outstanding reference, then xUndo is either a no-op + ** (if (a) is false) or simply expels the page from the cache (if (b) + ** is false). + ** + ** If the upper layer is doing a rollback, it is guaranteed that there + ** are no outstanding references to any page other than page 1. And + ** page 1 is never written to the log until the transaction is + ** committed. As a result, the call to xUndo may not fail. + */ + assert( walFramePgno(pWal, iFrame)!=1 ); + rc = xUndo(pUndoCtx, walFramePgno(pWal, iFrame)); + } + if( iMax!=pWal->hdr.mxFrame ) walCleanupHash(pWal); + } + return rc; +} + +/* +** Argument aWalData must point to an array of WAL_SAVEPOINT_NDATA u32 +** values. This function populates the array with values required to +** "rollback" the write position of the WAL handle back to the current +** point in the event of a savepoint rollback (via WalSavepointUndo()). +*/ +SQLITE_PRIVATE void sqlite3WalSavepoint(Wal *pWal, u32 *aWalData){ + assert( pWal->writeLock ); + aWalData[0] = pWal->hdr.mxFrame; + aWalData[1] = pWal->hdr.aFrameCksum[0]; + aWalData[2] = pWal->hdr.aFrameCksum[1]; + aWalData[3] = pWal->nCkpt; +} + +/* +** Move the write position of the WAL back to the point identified by +** the values in the aWalData[] array. aWalData must point to an array +** of WAL_SAVEPOINT_NDATA u32 values that has been previously populated +** by a call to WalSavepoint(). +*/ +SQLITE_PRIVATE int sqlite3WalSavepointUndo(Wal *pWal, u32 *aWalData){ + int rc = SQLITE_OK; + + assert( pWal->writeLock ); + assert( aWalData[3]!=pWal->nCkpt || aWalData[0]<=pWal->hdr.mxFrame ); + + if( aWalData[3]!=pWal->nCkpt ){ + /* This savepoint was opened immediately after the write-transaction + ** was started. Right after that, the writer decided to wrap around + ** to the start of the log. Update the savepoint values to match. + */ + aWalData[0] = 0; + aWalData[3] = pWal->nCkpt; + } + + if( aWalData[0]hdr.mxFrame ){ + pWal->hdr.mxFrame = aWalData[0]; + pWal->hdr.aFrameCksum[0] = aWalData[1]; + pWal->hdr.aFrameCksum[1] = aWalData[2]; + walCleanupHash(pWal); + } + + return rc; +} + +/* +** This function is called just before writing a set of frames to the log +** file (see sqlite3WalFrames()). It checks to see if, instead of appending +** to the current log file, it is possible to overwrite the start of the +** existing log file with the new frames (i.e. "reset" the log). If so, +** it sets pWal->hdr.mxFrame to 0. Otherwise, pWal->hdr.mxFrame is left +** unchanged. +** +** SQLITE_OK is returned if no error is encountered (regardless of whether +** or not pWal->hdr.mxFrame is modified). An SQLite error code is returned +** if an error occurs. +*/ +static int walRestartLog(Wal *pWal){ + int rc = SQLITE_OK; + int cnt; + + if( pWal->readLock==0 ){ + volatile WalCkptInfo *pInfo = walCkptInfo(pWal); + assert( pInfo->nBackfill==pWal->hdr.mxFrame ); + if( pInfo->nBackfill>0 ){ + u32 salt1; + sqlite3_randomness(4, &salt1); + rc = walLockExclusive(pWal, WAL_READ_LOCK(1), WAL_NREADER-1); + if( rc==SQLITE_OK ){ + /* If all readers are using WAL_READ_LOCK(0) (in other words if no + ** readers are currently using the WAL), then the transactions + ** frames will overwrite the start of the existing log. Update the + ** wal-index header to reflect this. + ** + ** In theory it would be Ok to update the cache of the header only + ** at this point. But updating the actual wal-index header is also + ** safe and means there is no special case for sqlite3WalUndo() + ** to handle if this transaction is rolled back. */ + walRestartHdr(pWal, salt1); + walUnlockExclusive(pWal, WAL_READ_LOCK(1), WAL_NREADER-1); + }else if( rc!=SQLITE_BUSY ){ + return rc; + } + } + walUnlockShared(pWal, WAL_READ_LOCK(0)); + pWal->readLock = -1; + cnt = 0; + do{ + int notUsed; + rc = walTryBeginRead(pWal, ¬Used, 1, ++cnt); + }while( rc==WAL_RETRY ); + assert( (rc&0xff)!=SQLITE_BUSY ); /* BUSY not possible when useWal==1 */ + testcase( (rc&0xff)==SQLITE_IOERR ); + testcase( rc==SQLITE_PROTOCOL ); + testcase( rc==SQLITE_OK ); + } + return rc; +} + +/* +** Information about the current state of the WAL file and where +** the next fsync should occur - passed from sqlite3WalFrames() into +** walWriteToLog(). +*/ +typedef struct WalWriter { + Wal *pWal; /* The complete WAL information */ + sqlite3_file *pFd; /* The WAL file to which we write */ + sqlite3_int64 iSyncPoint; /* Fsync at this offset */ + int syncFlags; /* Flags for the fsync */ + int szPage; /* Size of one page */ +} WalWriter; + +/* +** Write iAmt bytes of content into the WAL file beginning at iOffset. +** Do a sync when crossing the p->iSyncPoint boundary. +** +** In other words, if iSyncPoint is in between iOffset and iOffset+iAmt, +** first write the part before iSyncPoint, then sync, then write the +** rest. +*/ +static int walWriteToLog( + WalWriter *p, /* WAL to write to */ + void *pContent, /* Content to be written */ + int iAmt, /* Number of bytes to write */ + sqlite3_int64 iOffset /* Start writing at this offset */ +){ + int rc; + if( iOffsetiSyncPoint && iOffset+iAmt>=p->iSyncPoint ){ + int iFirstAmt = (int)(p->iSyncPoint - iOffset); + rc = sqlite3OsWrite(p->pFd, pContent, iFirstAmt, iOffset); + if( rc ) return rc; + iOffset += iFirstAmt; + iAmt -= iFirstAmt; + pContent = (void*)(iFirstAmt + (char*)pContent); + assert( WAL_SYNC_FLAGS(p->syncFlags)!=0 ); + rc = sqlite3OsSync(p->pFd, WAL_SYNC_FLAGS(p->syncFlags)); + if( iAmt==0 || rc ) return rc; + } + rc = sqlite3OsWrite(p->pFd, pContent, iAmt, iOffset); + return rc; +} + +/* +** Write out a single frame of the WAL +*/ +static int walWriteOneFrame( + WalWriter *p, /* Where to write the frame */ + PgHdr *pPage, /* The page of the frame to be written */ + int nTruncate, /* The commit flag. Usually 0. >0 for commit */ + sqlite3_int64 iOffset /* Byte offset at which to write */ +){ + int rc; /* Result code from subfunctions */ + void *pData; /* Data actually written */ + u8 aFrame[WAL_FRAME_HDRSIZE]; /* Buffer to assemble frame-header in */ +#if defined(SQLITE_HAS_CODEC) + if( (pData = sqlite3PagerCodec(pPage))==0 ) return SQLITE_NOMEM_BKPT; +#else + pData = pPage->pData; +#endif + walEncodeFrame(p->pWal, pPage->pgno, nTruncate, pData, aFrame); + rc = walWriteToLog(p, aFrame, sizeof(aFrame), iOffset); + if( rc ) return rc; + /* Write the page data */ + rc = walWriteToLog(p, pData, p->szPage, iOffset+sizeof(aFrame)); + return rc; +} + +/* +** This function is called as part of committing a transaction within which +** one or more frames have been overwritten. It updates the checksums for +** all frames written to the wal file by the current transaction starting +** with the earliest to have been overwritten. +** +** SQLITE_OK is returned if successful, or an SQLite error code otherwise. +*/ +static int walRewriteChecksums(Wal *pWal, u32 iLast){ + const int szPage = pWal->szPage;/* Database page size */ + int rc = SQLITE_OK; /* Return code */ + u8 *aBuf; /* Buffer to load data from wal file into */ + u8 aFrame[WAL_FRAME_HDRSIZE]; /* Buffer to assemble frame-headers in */ + u32 iRead; /* Next frame to read from wal file */ + i64 iCksumOff; + + aBuf = sqlite3_malloc(szPage + WAL_FRAME_HDRSIZE); + if( aBuf==0 ) return SQLITE_NOMEM_BKPT; + + /* Find the checksum values to use as input for the recalculating the + ** first checksum. If the first frame is frame 1 (implying that the current + ** transaction restarted the wal file), these values must be read from the + ** wal-file header. Otherwise, read them from the frame header of the + ** previous frame. */ + assert( pWal->iReCksum>0 ); + if( pWal->iReCksum==1 ){ + iCksumOff = 24; + }else{ + iCksumOff = walFrameOffset(pWal->iReCksum-1, szPage) + 16; + } + rc = sqlite3OsRead(pWal->pWalFd, aBuf, sizeof(u32)*2, iCksumOff); + pWal->hdr.aFrameCksum[0] = sqlite3Get4byte(aBuf); + pWal->hdr.aFrameCksum[1] = sqlite3Get4byte(&aBuf[sizeof(u32)]); + + iRead = pWal->iReCksum; + pWal->iReCksum = 0; + for(; rc==SQLITE_OK && iRead<=iLast; iRead++){ + i64 iOff = walFrameOffset(iRead, szPage); + rc = sqlite3OsRead(pWal->pWalFd, aBuf, szPage+WAL_FRAME_HDRSIZE, iOff); + if( rc==SQLITE_OK ){ + u32 iPgno, nDbSize; + iPgno = sqlite3Get4byte(aBuf); + nDbSize = sqlite3Get4byte(&aBuf[4]); + + walEncodeFrame(pWal, iPgno, nDbSize, &aBuf[WAL_FRAME_HDRSIZE], aFrame); + rc = sqlite3OsWrite(pWal->pWalFd, aFrame, sizeof(aFrame), iOff); + } + } + + sqlite3_free(aBuf); + return rc; +} + +/* +** Write a set of frames to the log. The caller must hold the write-lock +** on the log file (obtained using sqlite3WalBeginWriteTransaction()). +*/ +SQLITE_PRIVATE int sqlite3WalFrames( + Wal *pWal, /* Wal handle to write to */ + int szPage, /* Database page-size in bytes */ + PgHdr *pList, /* List of dirty pages to write */ + Pgno nTruncate, /* Database size after this commit */ + int isCommit, /* True if this is a commit */ + int sync_flags /* Flags to pass to OsSync() (or 0) */ +){ + int rc; /* Used to catch return codes */ + u32 iFrame; /* Next frame address */ + PgHdr *p; /* Iterator to run through pList with. */ + PgHdr *pLast = 0; /* Last frame in list */ + int nExtra = 0; /* Number of extra copies of last page */ + int szFrame; /* The size of a single frame */ + i64 iOffset; /* Next byte to write in WAL file */ + WalWriter w; /* The writer */ + u32 iFirst = 0; /* First frame that may be overwritten */ + WalIndexHdr *pLive; /* Pointer to shared header */ + + assert( pList ); + assert( pWal->writeLock ); + + /* If this frame set completes a transaction, then nTruncate>0. If + ** nTruncate==0 then this frame set does not complete the transaction. */ + assert( (isCommit!=0)==(nTruncate!=0) ); + +#if defined(SQLITE_TEST) && defined(SQLITE_DEBUG) + { int cnt; for(cnt=0, p=pList; p; p=p->pDirty, cnt++){} + WALTRACE(("WAL%p: frame write begin. %d frames. mxFrame=%d. %s\n", + pWal, cnt, pWal->hdr.mxFrame, isCommit ? "Commit" : "Spill")); + } +#endif + + pLive = (WalIndexHdr*)walIndexHdr(pWal); + if( memcmp(&pWal->hdr, (void *)pLive, sizeof(WalIndexHdr))!=0 ){ + iFirst = pLive->mxFrame+1; + } + + /* See if it is possible to write these frames into the start of the + ** log file, instead of appending to it at pWal->hdr.mxFrame. + */ + if( SQLITE_OK!=(rc = walRestartLog(pWal)) ){ + return rc; + } + + /* If this is the first frame written into the log, write the WAL + ** header to the start of the WAL file. See comments at the top of + ** this source file for a description of the WAL header format. + */ + iFrame = pWal->hdr.mxFrame; + if( iFrame==0 ){ + u8 aWalHdr[WAL_HDRSIZE]; /* Buffer to assemble wal-header in */ + u32 aCksum[2]; /* Checksum for wal-header */ + + sqlite3Put4byte(&aWalHdr[0], (WAL_MAGIC | SQLITE_BIGENDIAN)); + sqlite3Put4byte(&aWalHdr[4], WAL_MAX_VERSION); + sqlite3Put4byte(&aWalHdr[8], szPage); + sqlite3Put4byte(&aWalHdr[12], pWal->nCkpt); + if( pWal->nCkpt==0 ) sqlite3_randomness(8, pWal->hdr.aSalt); + memcpy(&aWalHdr[16], pWal->hdr.aSalt, 8); + walChecksumBytes(1, aWalHdr, WAL_HDRSIZE-2*4, 0, aCksum); + sqlite3Put4byte(&aWalHdr[24], aCksum[0]); + sqlite3Put4byte(&aWalHdr[28], aCksum[1]); + + pWal->szPage = szPage; + pWal->hdr.bigEndCksum = SQLITE_BIGENDIAN; + pWal->hdr.aFrameCksum[0] = aCksum[0]; + pWal->hdr.aFrameCksum[1] = aCksum[1]; + pWal->truncateOnCommit = 1; + + rc = sqlite3OsWrite(pWal->pWalFd, aWalHdr, sizeof(aWalHdr), 0); + WALTRACE(("WAL%p: wal-header write %s\n", pWal, rc ? "failed" : "ok")); + if( rc!=SQLITE_OK ){ + return rc; + } + + /* Sync the header (unless SQLITE_IOCAP_SEQUENTIAL is true or unless + ** all syncing is turned off by PRAGMA synchronous=OFF). Otherwise + ** an out-of-order write following a WAL restart could result in + ** database corruption. See the ticket: + ** + ** https://sqlite.org/src/info/ff5be73dee + */ + if( pWal->syncHeader ){ + rc = sqlite3OsSync(pWal->pWalFd, CKPT_SYNC_FLAGS(sync_flags)); + if( rc ) return rc; + } + } + assert( (int)pWal->szPage==szPage ); + + /* Setup information needed to write frames into the WAL */ + w.pWal = pWal; + w.pFd = pWal->pWalFd; + w.iSyncPoint = 0; + w.syncFlags = sync_flags; + w.szPage = szPage; + iOffset = walFrameOffset(iFrame+1, szPage); + szFrame = szPage + WAL_FRAME_HDRSIZE; + + /* Write all frames into the log file exactly once */ + for(p=pList; p; p=p->pDirty){ + int nDbSize; /* 0 normally. Positive == commit flag */ + + /* Check if this page has already been written into the wal file by + ** the current transaction. If so, overwrite the existing frame and + ** set Wal.writeLock to WAL_WRITELOCK_RECKSUM - indicating that + ** checksums must be recomputed when the transaction is committed. */ + if( iFirst && (p->pDirty || isCommit==0) ){ + u32 iWrite = 0; + VVA_ONLY(rc =) sqlite3WalFindFrame(pWal, p->pgno, &iWrite); + assert( rc==SQLITE_OK || iWrite==0 ); + if( iWrite>=iFirst ){ + i64 iOff = walFrameOffset(iWrite, szPage) + WAL_FRAME_HDRSIZE; + void *pData; + if( pWal->iReCksum==0 || iWriteiReCksum ){ + pWal->iReCksum = iWrite; + } +#if defined(SQLITE_HAS_CODEC) + if( (pData = sqlite3PagerCodec(p))==0 ) return SQLITE_NOMEM; +#else + pData = p->pData; +#endif + rc = sqlite3OsWrite(pWal->pWalFd, pData, szPage, iOff); + if( rc ) return rc; + p->flags &= ~PGHDR_WAL_APPEND; + continue; + } + } + + iFrame++; + assert( iOffset==walFrameOffset(iFrame, szPage) ); + nDbSize = (isCommit && p->pDirty==0) ? nTruncate : 0; + rc = walWriteOneFrame(&w, p, nDbSize, iOffset); + if( rc ) return rc; + pLast = p; + iOffset += szFrame; + p->flags |= PGHDR_WAL_APPEND; + } + + /* Recalculate checksums within the wal file if required. */ + if( isCommit && pWal->iReCksum ){ + rc = walRewriteChecksums(pWal, iFrame); + if( rc ) return rc; + } + + /* If this is the end of a transaction, then we might need to pad + ** the transaction and/or sync the WAL file. + ** + ** Padding and syncing only occur if this set of frames complete a + ** transaction and if PRAGMA synchronous=FULL. If synchronous==NORMAL + ** or synchronous==OFF, then no padding or syncing are needed. + ** + ** If SQLITE_IOCAP_POWERSAFE_OVERWRITE is defined, then padding is not + ** needed and only the sync is done. If padding is needed, then the + ** final frame is repeated (with its commit mark) until the next sector + ** boundary is crossed. Only the part of the WAL prior to the last + ** sector boundary is synced; the part of the last frame that extends + ** past the sector boundary is written after the sync. + */ + if( isCommit && WAL_SYNC_FLAGS(sync_flags)!=0 ){ + int bSync = 1; + if( pWal->padToSectorBoundary ){ + int sectorSize = sqlite3SectorSize(pWal->pWalFd); + w.iSyncPoint = ((iOffset+sectorSize-1)/sectorSize)*sectorSize; + bSync = (w.iSyncPoint==iOffset); + testcase( bSync ); + while( iOffsettruncateOnCommit && pWal->mxWalSize>=0 ){ + i64 sz = pWal->mxWalSize; + if( walFrameOffset(iFrame+nExtra+1, szPage)>pWal->mxWalSize ){ + sz = walFrameOffset(iFrame+nExtra+1, szPage); + } + walLimitSize(pWal, sz); + pWal->truncateOnCommit = 0; + } + + /* Append data to the wal-index. It is not necessary to lock the + ** wal-index to do this as the SQLITE_SHM_WRITE lock held on the wal-index + ** guarantees that there are no other writers, and no data that may + ** be in use by existing readers is being overwritten. + */ + iFrame = pWal->hdr.mxFrame; + for(p=pList; p && rc==SQLITE_OK; p=p->pDirty){ + if( (p->flags & PGHDR_WAL_APPEND)==0 ) continue; + iFrame++; + rc = walIndexAppend(pWal, iFrame, p->pgno); + } + assert( pLast!=0 || nExtra==0 ); + while( rc==SQLITE_OK && nExtra>0 ){ + iFrame++; + nExtra--; + rc = walIndexAppend(pWal, iFrame, pLast->pgno); + } + + if( rc==SQLITE_OK ){ + /* Update the private copy of the header. */ + pWal->hdr.szPage = (u16)((szPage&0xff00) | (szPage>>16)); + testcase( szPage<=32768 ); + testcase( szPage>=65536 ); + pWal->hdr.mxFrame = iFrame; + if( isCommit ){ + pWal->hdr.iChange++; + pWal->hdr.nPage = nTruncate; + } + /* If this is a commit, update the wal-index header too. */ + if( isCommit ){ + walIndexWriteHdr(pWal); + pWal->iCallback = iFrame; + } + } + + WALTRACE(("WAL%p: frame write %s\n", pWal, rc ? "failed" : "ok")); + return rc; +} + +/* +** This routine is called to implement sqlite3_wal_checkpoint() and +** related interfaces. +** +** Obtain a CHECKPOINT lock and then backfill as much information as +** we can from WAL into the database. +** +** If parameter xBusy is not NULL, it is a pointer to a busy-handler +** callback. In this case this function runs a blocking checkpoint. +*/ +SQLITE_PRIVATE int sqlite3WalCheckpoint( + Wal *pWal, /* Wal connection */ + sqlite3 *db, /* Check this handle's interrupt flag */ + int eMode, /* PASSIVE, FULL, RESTART, or TRUNCATE */ + int (*xBusy)(void*), /* Function to call when busy */ + void *pBusyArg, /* Context argument for xBusyHandler */ + int sync_flags, /* Flags to sync db file with (or 0) */ + int nBuf, /* Size of temporary buffer */ + u8 *zBuf, /* Temporary buffer to use */ + int *pnLog, /* OUT: Number of frames in WAL */ + int *pnCkpt /* OUT: Number of backfilled frames in WAL */ +){ + int rc; /* Return code */ + int isChanged = 0; /* True if a new wal-index header is loaded */ + int eMode2 = eMode; /* Mode to pass to walCheckpoint() */ + int (*xBusy2)(void*) = xBusy; /* Busy handler for eMode2 */ + + assert( pWal->ckptLock==0 ); + assert( pWal->writeLock==0 ); + + /* EVIDENCE-OF: R-62920-47450 The busy-handler callback is never invoked + ** in the SQLITE_CHECKPOINT_PASSIVE mode. */ + assert( eMode!=SQLITE_CHECKPOINT_PASSIVE || xBusy==0 ); + + if( pWal->readOnly ) return SQLITE_READONLY; + WALTRACE(("WAL%p: checkpoint begins\n", pWal)); + + /* IMPLEMENTATION-OF: R-62028-47212 All calls obtain an exclusive + ** "checkpoint" lock on the database file. */ + rc = walLockExclusive(pWal, WAL_CKPT_LOCK, 1); + if( rc ){ + /* EVIDENCE-OF: R-10421-19736 If any other process is running a + ** checkpoint operation at the same time, the lock cannot be obtained and + ** SQLITE_BUSY is returned. + ** EVIDENCE-OF: R-53820-33897 Even if there is a busy-handler configured, + ** it will not be invoked in this case. + */ + testcase( rc==SQLITE_BUSY ); + testcase( xBusy!=0 ); + return rc; + } + pWal->ckptLock = 1; + + /* IMPLEMENTATION-OF: R-59782-36818 The SQLITE_CHECKPOINT_FULL, RESTART and + ** TRUNCATE modes also obtain the exclusive "writer" lock on the database + ** file. + ** + ** EVIDENCE-OF: R-60642-04082 If the writer lock cannot be obtained + ** immediately, and a busy-handler is configured, it is invoked and the + ** writer lock retried until either the busy-handler returns 0 or the + ** lock is successfully obtained. + */ + if( eMode!=SQLITE_CHECKPOINT_PASSIVE ){ + rc = walBusyLock(pWal, xBusy, pBusyArg, WAL_WRITE_LOCK, 1); + if( rc==SQLITE_OK ){ + pWal->writeLock = 1; + }else if( rc==SQLITE_BUSY ){ + eMode2 = SQLITE_CHECKPOINT_PASSIVE; + xBusy2 = 0; + rc = SQLITE_OK; + } + } + + /* Read the wal-index header. */ + if( rc==SQLITE_OK ){ + rc = walIndexReadHdr(pWal, &isChanged); + if( isChanged && pWal->pDbFd->pMethods->iVersion>=3 ){ + sqlite3OsUnfetch(pWal->pDbFd, 0, 0); + } + } + + /* Copy data from the log to the database file. */ + if( rc==SQLITE_OK ){ + + if( pWal->hdr.mxFrame && walPagesize(pWal)!=nBuf ){ + rc = SQLITE_CORRUPT_BKPT; + }else{ + rc = walCheckpoint(pWal, db, eMode2, xBusy2, pBusyArg, sync_flags, zBuf); + } + + /* If no error occurred, set the output variables. */ + if( rc==SQLITE_OK || rc==SQLITE_BUSY ){ + if( pnLog ) *pnLog = (int)pWal->hdr.mxFrame; + if( pnCkpt ) *pnCkpt = (int)(walCkptInfo(pWal)->nBackfill); + } + } + + if( isChanged ){ + /* If a new wal-index header was loaded before the checkpoint was + ** performed, then the pager-cache associated with pWal is now + ** out of date. So zero the cached wal-index header to ensure that + ** next time the pager opens a snapshot on this database it knows that + ** the cache needs to be reset. + */ + memset(&pWal->hdr, 0, sizeof(WalIndexHdr)); + } + + /* Release the locks. */ + sqlite3WalEndWriteTransaction(pWal); + walUnlockExclusive(pWal, WAL_CKPT_LOCK, 1); + pWal->ckptLock = 0; + WALTRACE(("WAL%p: checkpoint %s\n", pWal, rc ? "failed" : "ok")); + return (rc==SQLITE_OK && eMode!=eMode2 ? SQLITE_BUSY : rc); +} + +/* Return the value to pass to a sqlite3_wal_hook callback, the +** number of frames in the WAL at the point of the last commit since +** sqlite3WalCallback() was called. If no commits have occurred since +** the last call, then return 0. +*/ +SQLITE_PRIVATE int sqlite3WalCallback(Wal *pWal){ + u32 ret = 0; + if( pWal ){ + ret = pWal->iCallback; + pWal->iCallback = 0; + } + return (int)ret; +} + +/* +** This function is called to change the WAL subsystem into or out +** of locking_mode=EXCLUSIVE. +** +** If op is zero, then attempt to change from locking_mode=EXCLUSIVE +** into locking_mode=NORMAL. This means that we must acquire a lock +** on the pWal->readLock byte. If the WAL is already in locking_mode=NORMAL +** or if the acquisition of the lock fails, then return 0. If the +** transition out of exclusive-mode is successful, return 1. This +** operation must occur while the pager is still holding the exclusive +** lock on the main database file. +** +** If op is one, then change from locking_mode=NORMAL into +** locking_mode=EXCLUSIVE. This means that the pWal->readLock must +** be released. Return 1 if the transition is made and 0 if the +** WAL is already in exclusive-locking mode - meaning that this +** routine is a no-op. The pager must already hold the exclusive lock +** on the main database file before invoking this operation. +** +** If op is negative, then do a dry-run of the op==1 case but do +** not actually change anything. The pager uses this to see if it +** should acquire the database exclusive lock prior to invoking +** the op==1 case. +*/ +SQLITE_PRIVATE int sqlite3WalExclusiveMode(Wal *pWal, int op){ + int rc; + assert( pWal->writeLock==0 ); + assert( pWal->exclusiveMode!=WAL_HEAPMEMORY_MODE || op==-1 ); + + /* pWal->readLock is usually set, but might be -1 if there was a + ** prior error while attempting to acquire are read-lock. This cannot + ** happen if the connection is actually in exclusive mode (as no xShmLock + ** locks are taken in this case). Nor should the pager attempt to + ** upgrade to exclusive-mode following such an error. + */ + assert( pWal->readLock>=0 || pWal->lockError ); + assert( pWal->readLock>=0 || (op<=0 && pWal->exclusiveMode==0) ); + + if( op==0 ){ + if( pWal->exclusiveMode!=WAL_NORMAL_MODE ){ + pWal->exclusiveMode = WAL_NORMAL_MODE; + if( walLockShared(pWal, WAL_READ_LOCK(pWal->readLock))!=SQLITE_OK ){ + pWal->exclusiveMode = WAL_EXCLUSIVE_MODE; + } + rc = pWal->exclusiveMode==WAL_NORMAL_MODE; + }else{ + /* Already in locking_mode=NORMAL */ + rc = 0; + } + }else if( op>0 ){ + assert( pWal->exclusiveMode==WAL_NORMAL_MODE ); + assert( pWal->readLock>=0 ); + walUnlockShared(pWal, WAL_READ_LOCK(pWal->readLock)); + pWal->exclusiveMode = WAL_EXCLUSIVE_MODE; + rc = 1; + }else{ + rc = pWal->exclusiveMode==WAL_NORMAL_MODE; + } + return rc; +} + +/* +** Return true if the argument is non-NULL and the WAL module is using +** heap-memory for the wal-index. Otherwise, if the argument is NULL or the +** WAL module is using shared-memory, return false. +*/ +SQLITE_PRIVATE int sqlite3WalHeapMemory(Wal *pWal){ + return (pWal && pWal->exclusiveMode==WAL_HEAPMEMORY_MODE ); +} + +#ifdef SQLITE_ENABLE_SNAPSHOT +/* Create a snapshot object. The content of a snapshot is opaque to +** every other subsystem, so the WAL module can put whatever it needs +** in the object. +*/ +SQLITE_PRIVATE int sqlite3WalSnapshotGet(Wal *pWal, sqlite3_snapshot **ppSnapshot){ + int rc = SQLITE_OK; + WalIndexHdr *pRet; + static const u32 aZero[4] = { 0, 0, 0, 0 }; + + assert( pWal->readLock>=0 && pWal->writeLock==0 ); + + if( memcmp(&pWal->hdr.aFrameCksum[0],aZero,16)==0 ){ + *ppSnapshot = 0; + return SQLITE_ERROR; + } + pRet = (WalIndexHdr*)sqlite3_malloc(sizeof(WalIndexHdr)); + if( pRet==0 ){ + rc = SQLITE_NOMEM_BKPT; + }else{ + memcpy(pRet, &pWal->hdr, sizeof(WalIndexHdr)); + *ppSnapshot = (sqlite3_snapshot*)pRet; + } + + return rc; +} + +/* Try to open on pSnapshot when the next read-transaction starts +*/ +SQLITE_PRIVATE void sqlite3WalSnapshotOpen(Wal *pWal, sqlite3_snapshot *pSnapshot){ + pWal->pSnapshot = (WalIndexHdr*)pSnapshot; +} + +/* +** Return a +ve value if snapshot p1 is newer than p2. A -ve value if +** p1 is older than p2 and zero if p1 and p2 are the same snapshot. +*/ +SQLITE_API int sqlite3_snapshot_cmp(sqlite3_snapshot *p1, sqlite3_snapshot *p2){ + WalIndexHdr *pHdr1 = (WalIndexHdr*)p1; + WalIndexHdr *pHdr2 = (WalIndexHdr*)p2; + + /* aSalt[0] is a copy of the value stored in the wal file header. It + ** is incremented each time the wal file is restarted. */ + if( pHdr1->aSalt[0]aSalt[0] ) return -1; + if( pHdr1->aSalt[0]>pHdr2->aSalt[0] ) return +1; + if( pHdr1->mxFramemxFrame ) return -1; + if( pHdr1->mxFrame>pHdr2->mxFrame ) return +1; + return 0; +} + +/* +** The caller currently has a read transaction open on the database. +** This function takes a SHARED lock on the CHECKPOINTER slot and then +** checks if the snapshot passed as the second argument is still +** available. If so, SQLITE_OK is returned. +** +** If the snapshot is not available, SQLITE_ERROR is returned. Or, if +** the CHECKPOINTER lock cannot be obtained, SQLITE_BUSY. If any error +** occurs (any value other than SQLITE_OK is returned), the CHECKPOINTER +** lock is released before returning. +*/ +SQLITE_PRIVATE int sqlite3WalSnapshotCheck(Wal *pWal, sqlite3_snapshot *pSnapshot){ + int rc; + rc = walLockShared(pWal, WAL_CKPT_LOCK); + if( rc==SQLITE_OK ){ + WalIndexHdr *pNew = (WalIndexHdr*)pSnapshot; + if( memcmp(pNew->aSalt, pWal->hdr.aSalt, sizeof(pWal->hdr.aSalt)) + || pNew->mxFramenBackfillAttempted + ){ + rc = SQLITE_ERROR_SNAPSHOT; + walUnlockShared(pWal, WAL_CKPT_LOCK); + } + } + return rc; +} + +/* +** Release a lock obtained by an earlier successful call to +** sqlite3WalSnapshotCheck(). +*/ +SQLITE_PRIVATE void sqlite3WalSnapshotUnlock(Wal *pWal){ + assert( pWal ); + walUnlockShared(pWal, WAL_CKPT_LOCK); +} + + +#endif /* SQLITE_ENABLE_SNAPSHOT */ + +#ifdef SQLITE_ENABLE_ZIPVFS +/* +** If the argument is not NULL, it points to a Wal object that holds a +** read-lock. This function returns the database page-size if it is known, +** or zero if it is not (or if pWal is NULL). +*/ +SQLITE_PRIVATE int sqlite3WalFramesize(Wal *pWal){ + assert( pWal==0 || pWal->readLock>=0 ); + return (pWal ? pWal->szPage : 0); +} +#endif + +/* Return the sqlite3_file object for the WAL file +*/ +SQLITE_PRIVATE sqlite3_file *sqlite3WalFile(Wal *pWal){ + return pWal->pWalFd; +} + +#endif /* #ifndef SQLITE_OMIT_WAL */ + +/************** End of wal.c *************************************************/ +/************** Begin file btmutex.c *****************************************/ +/* +** 2007 August 27 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** +** This file contains code used to implement mutexes on Btree objects. +** This code really belongs in btree.c. But btree.c is getting too +** big and we want to break it down some. This packaged seemed like +** a good breakout. +*/ +/************** Include btreeInt.h in the middle of btmutex.c ****************/ +/************** Begin file btreeInt.h ****************************************/ +/* +** 2004 April 6 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** This file implements an external (disk-based) database using BTrees. +** For a detailed discussion of BTrees, refer to +** +** Donald E. Knuth, THE ART OF COMPUTER PROGRAMMING, Volume 3: +** "Sorting And Searching", pages 473-480. Addison-Wesley +** Publishing Company, Reading, Massachusetts. +** +** The basic idea is that each page of the file contains N database +** entries and N+1 pointers to subpages. +** +** ---------------------------------------------------------------- +** | Ptr(0) | Key(0) | Ptr(1) | Key(1) | ... | Key(N-1) | Ptr(N) | +** ---------------------------------------------------------------- +** +** All of the keys on the page that Ptr(0) points to have values less +** than Key(0). All of the keys on page Ptr(1) and its subpages have +** values greater than Key(0) and less than Key(1). All of the keys +** on Ptr(N) and its subpages have values greater than Key(N-1). And +** so forth. +** +** Finding a particular key requires reading O(log(M)) pages from the +** disk where M is the number of entries in the tree. +** +** In this implementation, a single file can hold one or more separate +** BTrees. Each BTree is identified by the index of its root page. The +** key and data for any entry are combined to form the "payload". A +** fixed amount of payload can be carried directly on the database +** page. If the payload is larger than the preset amount then surplus +** bytes are stored on overflow pages. The payload for an entry +** and the preceding pointer are combined to form a "Cell". Each +** page has a small header which contains the Ptr(N) pointer and other +** information such as the size of key and data. +** +** FORMAT DETAILS +** +** The file is divided into pages. The first page is called page 1, +** the second is page 2, and so forth. A page number of zero indicates +** "no such page". The page size can be any power of 2 between 512 and 65536. +** Each page can be either a btree page, a freelist page, an overflow +** page, or a pointer-map page. +** +** The first page is always a btree page. The first 100 bytes of the first +** page contain a special header (the "file header") that describes the file. +** The format of the file header is as follows: +** +** OFFSET SIZE DESCRIPTION +** 0 16 Header string: "SQLite format 3\000" +** 16 2 Page size in bytes. (1 means 65536) +** 18 1 File format write version +** 19 1 File format read version +** 20 1 Bytes of unused space at the end of each page +** 21 1 Max embedded payload fraction (must be 64) +** 22 1 Min embedded payload fraction (must be 32) +** 23 1 Min leaf payload fraction (must be 32) +** 24 4 File change counter +** 28 4 Reserved for future use +** 32 4 First freelist page +** 36 4 Number of freelist pages in the file +** 40 60 15 4-byte meta values passed to higher layers +** +** 40 4 Schema cookie +** 44 4 File format of schema layer +** 48 4 Size of page cache +** 52 4 Largest root-page (auto/incr_vacuum) +** 56 4 1=UTF-8 2=UTF16le 3=UTF16be +** 60 4 User version +** 64 4 Incremental vacuum mode +** 68 4 Application-ID +** 72 20 unused +** 92 4 The version-valid-for number +** 96 4 SQLITE_VERSION_NUMBER +** +** All of the integer values are big-endian (most significant byte first). +** +** The file change counter is incremented when the database is changed +** This counter allows other processes to know when the file has changed +** and thus when they need to flush their cache. +** +** The max embedded payload fraction is the amount of the total usable +** space in a page that can be consumed by a single cell for standard +** B-tree (non-LEAFDATA) tables. A value of 255 means 100%. The default +** is to limit the maximum cell size so that at least 4 cells will fit +** on one page. Thus the default max embedded payload fraction is 64. +** +** If the payload for a cell is larger than the max payload, then extra +** payload is spilled to overflow pages. Once an overflow page is allocated, +** as many bytes as possible are moved into the overflow pages without letting +** the cell size drop below the min embedded payload fraction. +** +** The min leaf payload fraction is like the min embedded payload fraction +** except that it applies to leaf nodes in a LEAFDATA tree. The maximum +** payload fraction for a LEAFDATA tree is always 100% (or 255) and it +** not specified in the header. +** +** Each btree pages is divided into three sections: The header, the +** cell pointer array, and the cell content area. Page 1 also has a 100-byte +** file header that occurs before the page header. +** +** |----------------| +** | file header | 100 bytes. Page 1 only. +** |----------------| +** | page header | 8 bytes for leaves. 12 bytes for interior nodes +** |----------------| +** | cell pointer | | 2 bytes per cell. Sorted order. +** | array | | Grows downward +** | | v +** |----------------| +** | unallocated | +** | space | +** |----------------| ^ Grows upwards +** | cell content | | Arbitrary order interspersed with freeblocks. +** | area | | and free space fragments. +** |----------------| +** +** The page headers looks like this: +** +** OFFSET SIZE DESCRIPTION +** 0 1 Flags. 1: intkey, 2: zerodata, 4: leafdata, 8: leaf +** 1 2 byte offset to the first freeblock +** 3 2 number of cells on this page +** 5 2 first byte of the cell content area +** 7 1 number of fragmented free bytes +** 8 4 Right child (the Ptr(N) value). Omitted on leaves. +** +** The flags define the format of this btree page. The leaf flag means that +** this page has no children. The zerodata flag means that this page carries +** only keys and no data. The intkey flag means that the key is an integer +** which is stored in the key size entry of the cell header rather than in +** the payload area. +** +** The cell pointer array begins on the first byte after the page header. +** The cell pointer array contains zero or more 2-byte numbers which are +** offsets from the beginning of the page to the cell content in the cell +** content area. The cell pointers occur in sorted order. The system strives +** to keep free space after the last cell pointer so that new cells can +** be easily added without having to defragment the page. +** +** Cell content is stored at the very end of the page and grows toward the +** beginning of the page. +** +** Unused space within the cell content area is collected into a linked list of +** freeblocks. Each freeblock is at least 4 bytes in size. The byte offset +** to the first freeblock is given in the header. Freeblocks occur in +** increasing order. Because a freeblock must be at least 4 bytes in size, +** any group of 3 or fewer unused bytes in the cell content area cannot +** exist on the freeblock chain. A group of 3 or fewer free bytes is called +** a fragment. The total number of bytes in all fragments is recorded. +** in the page header at offset 7. +** +** SIZE DESCRIPTION +** 2 Byte offset of the next freeblock +** 2 Bytes in this freeblock +** +** Cells are of variable length. Cells are stored in the cell content area at +** the end of the page. Pointers to the cells are in the cell pointer array +** that immediately follows the page header. Cells is not necessarily +** contiguous or in order, but cell pointers are contiguous and in order. +** +** Cell content makes use of variable length integers. A variable +** length integer is 1 to 9 bytes where the lower 7 bits of each +** byte are used. The integer consists of all bytes that have bit 8 set and +** the first byte with bit 8 clear. The most significant byte of the integer +** appears first. A variable-length integer may not be more than 9 bytes long. +** As a special case, all 8 bytes of the 9th byte are used as data. This +** allows a 64-bit integer to be encoded in 9 bytes. +** +** 0x00 becomes 0x00000000 +** 0x7f becomes 0x0000007f +** 0x81 0x00 becomes 0x00000080 +** 0x82 0x00 becomes 0x00000100 +** 0x80 0x7f becomes 0x0000007f +** 0x8a 0x91 0xd1 0xac 0x78 becomes 0x12345678 +** 0x81 0x81 0x81 0x81 0x01 becomes 0x10204081 +** +** Variable length integers are used for rowids and to hold the number of +** bytes of key and data in a btree cell. +** +** The content of a cell looks like this: +** +** SIZE DESCRIPTION +** 4 Page number of the left child. Omitted if leaf flag is set. +** var Number of bytes of data. Omitted if the zerodata flag is set. +** var Number of bytes of key. Or the key itself if intkey flag is set. +** * Payload +** 4 First page of the overflow chain. Omitted if no overflow +** +** Overflow pages form a linked list. Each page except the last is completely +** filled with data (pagesize - 4 bytes). The last page can have as little +** as 1 byte of data. +** +** SIZE DESCRIPTION +** 4 Page number of next overflow page +** * Data +** +** Freelist pages come in two subtypes: trunk pages and leaf pages. The +** file header points to the first in a linked list of trunk page. Each trunk +** page points to multiple leaf pages. The content of a leaf page is +** unspecified. A trunk page looks like this: +** +** SIZE DESCRIPTION +** 4 Page number of next trunk page +** 4 Number of leaf pointers on this page +** * zero or more pages numbers of leaves +*/ +/* #include "sqliteInt.h" */ + + +/* The following value is the maximum cell size assuming a maximum page +** size give above. +*/ +#define MX_CELL_SIZE(pBt) ((int)(pBt->pageSize-8)) + +/* The maximum number of cells on a single page of the database. This +** assumes a minimum cell size of 6 bytes (4 bytes for the cell itself +** plus 2 bytes for the index to the cell in the page header). Such +** small cells will be rare, but they are possible. +*/ +#define MX_CELL(pBt) ((pBt->pageSize-8)/6) + +/* Forward declarations */ +typedef struct MemPage MemPage; +typedef struct BtLock BtLock; +typedef struct CellInfo CellInfo; + +/* +** This is a magic string that appears at the beginning of every +** SQLite database in order to identify the file as a real database. +** +** You can change this value at compile-time by specifying a +** -DSQLITE_FILE_HEADER="..." on the compiler command-line. The +** header must be exactly 16 bytes including the zero-terminator so +** the string itself should be 15 characters long. If you change +** the header, then your custom library will not be able to read +** databases generated by the standard tools and the standard tools +** will not be able to read databases created by your custom library. +*/ +#ifndef SQLITE_FILE_HEADER /* 123456789 123456 */ +# define SQLITE_FILE_HEADER "SQLite format 3" +#endif + +/* +** Page type flags. An ORed combination of these flags appear as the +** first byte of on-disk image of every BTree page. +*/ +#define PTF_INTKEY 0x01 +#define PTF_ZERODATA 0x02 +#define PTF_LEAFDATA 0x04 +#define PTF_LEAF 0x08 + +/* +** An instance of this object stores information about each a single database +** page that has been loaded into memory. The information in this object +** is derived from the raw on-disk page content. +** +** As each database page is loaded into memory, the pager allocats an +** instance of this object and zeros the first 8 bytes. (This is the +** "extra" information associated with each page of the pager.) +** +** Access to all fields of this structure is controlled by the mutex +** stored in MemPage.pBt->mutex. +*/ +struct MemPage { + u8 isInit; /* True if previously initialized. MUST BE FIRST! */ + u8 bBusy; /* Prevent endless loops on corrupt database files */ + u8 intKey; /* True if table b-trees. False for index b-trees */ + u8 intKeyLeaf; /* True if the leaf of an intKey table */ + Pgno pgno; /* Page number for this page */ + /* Only the first 8 bytes (above) are zeroed by pager.c when a new page + ** is allocated. All fields that follow must be initialized before use */ + u8 leaf; /* True if a leaf page */ + u8 hdrOffset; /* 100 for page 1. 0 otherwise */ + u8 childPtrSize; /* 0 if leaf==1. 4 if leaf==0 */ + u8 max1bytePayload; /* min(maxLocal,127) */ + u8 nOverflow; /* Number of overflow cell bodies in aCell[] */ + u16 maxLocal; /* Copy of BtShared.maxLocal or BtShared.maxLeaf */ + u16 minLocal; /* Copy of BtShared.minLocal or BtShared.minLeaf */ + u16 cellOffset; /* Index in aData of first cell pointer */ + int nFree; /* Number of free bytes on the page. -1 for unknown */ + u16 nCell; /* Number of cells on this page, local and ovfl */ + u16 maskPage; /* Mask for page offset */ + u16 aiOvfl[4]; /* Insert the i-th overflow cell before the aiOvfl-th + ** non-overflow cell */ + u8 *apOvfl[4]; /* Pointers to the body of overflow cells */ + BtShared *pBt; /* Pointer to BtShared that this page is part of */ + u8 *aData; /* Pointer to disk image of the page data */ + u8 *aDataEnd; /* One byte past the end of usable data */ + u8 *aCellIdx; /* The cell index area */ + u8 *aDataOfst; /* Same as aData for leaves. aData+4 for interior */ + DbPage *pDbPage; /* Pager page handle */ + u16 (*xCellSize)(MemPage*,u8*); /* cellSizePtr method */ + void (*xParseCell)(MemPage*,u8*,CellInfo*); /* btreeParseCell method */ +}; + +/* +** A linked list of the following structures is stored at BtShared.pLock. +** Locks are added (or upgraded from READ_LOCK to WRITE_LOCK) when a cursor +** is opened on the table with root page BtShared.iTable. Locks are removed +** from this list when a transaction is committed or rolled back, or when +** a btree handle is closed. +*/ +struct BtLock { + Btree *pBtree; /* Btree handle holding this lock */ + Pgno iTable; /* Root page of table */ + u8 eLock; /* READ_LOCK or WRITE_LOCK */ + BtLock *pNext; /* Next in BtShared.pLock list */ +}; + +/* Candidate values for BtLock.eLock */ +#define READ_LOCK 1 +#define WRITE_LOCK 2 + +/* A Btree handle +** +** A database connection contains a pointer to an instance of +** this object for every database file that it has open. This structure +** is opaque to the database connection. The database connection cannot +** see the internals of this structure and only deals with pointers to +** this structure. +** +** For some database files, the same underlying database cache might be +** shared between multiple connections. In that case, each connection +** has it own instance of this object. But each instance of this object +** points to the same BtShared object. The database cache and the +** schema associated with the database file are all contained within +** the BtShared object. +** +** All fields in this structure are accessed under sqlite3.mutex. +** The pBt pointer itself may not be changed while there exists cursors +** in the referenced BtShared that point back to this Btree since those +** cursors have to go through this Btree to find their BtShared and +** they often do so without holding sqlite3.mutex. +*/ +struct Btree { + sqlite3 *db; /* The database connection holding this btree */ + BtShared *pBt; /* Sharable content of this btree */ + u8 inTrans; /* TRANS_NONE, TRANS_READ or TRANS_WRITE */ + u8 sharable; /* True if we can share pBt with another db */ + u8 locked; /* True if db currently has pBt locked */ + u8 hasIncrblobCur; /* True if there are one or more Incrblob cursors */ + int wantToLock; /* Number of nested calls to sqlite3BtreeEnter() */ + int nBackup; /* Number of backup operations reading this btree */ + u32 iDataVersion; /* Combines with pBt->pPager->iDataVersion */ + Btree *pNext; /* List of other sharable Btrees from the same db */ + Btree *pPrev; /* Back pointer of the same list */ +#ifndef SQLITE_OMIT_SHARED_CACHE + BtLock lock; /* Object used to lock page 1 */ +#endif +}; + +/* +** Btree.inTrans may take one of the following values. +** +** If the shared-data extension is enabled, there may be multiple users +** of the Btree structure. At most one of these may open a write transaction, +** but any number may have active read transactions. +*/ +#define TRANS_NONE 0 +#define TRANS_READ 1 +#define TRANS_WRITE 2 + +/* +** An instance of this object represents a single database file. +** +** A single database file can be in use at the same time by two +** or more database connections. When two or more connections are +** sharing the same database file, each connection has it own +** private Btree object for the file and each of those Btrees points +** to this one BtShared object. BtShared.nRef is the number of +** connections currently sharing this database file. +** +** Fields in this structure are accessed under the BtShared.mutex +** mutex, except for nRef and pNext which are accessed under the +** global SQLITE_MUTEX_STATIC_MASTER mutex. The pPager field +** may not be modified once it is initially set as long as nRef>0. +** The pSchema field may be set once under BtShared.mutex and +** thereafter is unchanged as long as nRef>0. +** +** isPending: +** +** If a BtShared client fails to obtain a write-lock on a database +** table (because there exists one or more read-locks on the table), +** the shared-cache enters 'pending-lock' state and isPending is +** set to true. +** +** The shared-cache leaves the 'pending lock' state when either of +** the following occur: +** +** 1) The current writer (BtShared.pWriter) concludes its transaction, OR +** 2) The number of locks held by other connections drops to zero. +** +** while in the 'pending-lock' state, no connection may start a new +** transaction. +** +** This feature is included to help prevent writer-starvation. +*/ +struct BtShared { + Pager *pPager; /* The page cache */ + sqlite3 *db; /* Database connection currently using this Btree */ + BtCursor *pCursor; /* A list of all open cursors */ + MemPage *pPage1; /* First page of the database */ + u8 openFlags; /* Flags to sqlite3BtreeOpen() */ +#ifndef SQLITE_OMIT_AUTOVACUUM + u8 autoVacuum; /* True if auto-vacuum is enabled */ + u8 incrVacuum; /* True if incr-vacuum is enabled */ + u8 bDoTruncate; /* True to truncate db on commit */ +#endif + u8 inTransaction; /* Transaction state */ + u8 max1bytePayload; /* Maximum first byte of cell for a 1-byte payload */ +#ifdef SQLITE_HAS_CODEC + u8 optimalReserve; /* Desired amount of reserved space per page */ +#endif + u16 btsFlags; /* Boolean parameters. See BTS_* macros below */ + u16 maxLocal; /* Maximum local payload in non-LEAFDATA tables */ + u16 minLocal; /* Minimum local payload in non-LEAFDATA tables */ + u16 maxLeaf; /* Maximum local payload in a LEAFDATA table */ + u16 minLeaf; /* Minimum local payload in a LEAFDATA table */ + u32 pageSize; /* Total number of bytes on a page */ + u32 usableSize; /* Number of usable bytes on each page */ + int nTransaction; /* Number of open transactions (read + write) */ + u32 nPage; /* Number of pages in the database */ + void *pSchema; /* Pointer to space allocated by sqlite3BtreeSchema() */ + void (*xFreeSchema)(void*); /* Destructor for BtShared.pSchema */ + sqlite3_mutex *mutex; /* Non-recursive mutex required to access this object */ + Bitvec *pHasContent; /* Set of pages moved to free-list this transaction */ +#ifndef SQLITE_OMIT_SHARED_CACHE + int nRef; /* Number of references to this structure */ + BtShared *pNext; /* Next on a list of sharable BtShared structs */ + BtLock *pLock; /* List of locks held on this shared-btree struct */ + Btree *pWriter; /* Btree with currently open write transaction */ +#endif + u8 *pTmpSpace; /* Temp space sufficient to hold a single cell */ +}; + +/* +** Allowed values for BtShared.btsFlags +*/ +#define BTS_READ_ONLY 0x0001 /* Underlying file is readonly */ +#define BTS_PAGESIZE_FIXED 0x0002 /* Page size can no longer be changed */ +#define BTS_SECURE_DELETE 0x0004 /* PRAGMA secure_delete is enabled */ +#define BTS_OVERWRITE 0x0008 /* Overwrite deleted content with zeros */ +#define BTS_FAST_SECURE 0x000c /* Combination of the previous two */ +#define BTS_INITIALLY_EMPTY 0x0010 /* Database was empty at trans start */ +#define BTS_NO_WAL 0x0020 /* Do not open write-ahead-log files */ +#define BTS_EXCLUSIVE 0x0040 /* pWriter has an exclusive lock */ +#define BTS_PENDING 0x0080 /* Waiting for read-locks to clear */ + +/* +** An instance of the following structure is used to hold information +** about a cell. The parseCellPtr() function fills in this structure +** based on information extract from the raw disk page. +*/ +struct CellInfo { + i64 nKey; /* The key for INTKEY tables, or nPayload otherwise */ + u8 *pPayload; /* Pointer to the start of payload */ + u32 nPayload; /* Bytes of payload */ + u16 nLocal; /* Amount of payload held locally, not on overflow */ + u16 nSize; /* Size of the cell content on the main b-tree page */ +}; + +/* +** Maximum depth of an SQLite B-Tree structure. Any B-Tree deeper than +** this will be declared corrupt. This value is calculated based on a +** maximum database size of 2^31 pages a minimum fanout of 2 for a +** root-node and 3 for all other internal nodes. +** +** If a tree that appears to be taller than this is encountered, it is +** assumed that the database is corrupt. +*/ +#define BTCURSOR_MAX_DEPTH 20 + +/* +** A cursor is a pointer to a particular entry within a particular +** b-tree within a database file. +** +** The entry is identified by its MemPage and the index in +** MemPage.aCell[] of the entry. +** +** A single database file can be shared by two more database connections, +** but cursors cannot be shared. Each cursor is associated with a +** particular database connection identified BtCursor.pBtree.db. +** +** Fields in this structure are accessed under the BtShared.mutex +** found at self->pBt->mutex. +** +** skipNext meaning: +** The meaning of skipNext depends on the value of eState: +** +** eState Meaning of skipNext +** VALID skipNext is meaningless and is ignored +** INVALID skipNext is meaningless and is ignored +** SKIPNEXT sqlite3BtreeNext() is a no-op if skipNext>0 and +** sqlite3BtreePrevious() is no-op if skipNext<0. +** REQUIRESEEK restoreCursorPosition() restores the cursor to +** eState=SKIPNEXT if skipNext!=0 +** FAULT skipNext holds the cursor fault error code. +*/ +struct BtCursor { + u8 eState; /* One of the CURSOR_XXX constants (see below) */ + u8 curFlags; /* zero or more BTCF_* flags defined below */ + u8 curPagerFlags; /* Flags to send to sqlite3PagerGet() */ + u8 hints; /* As configured by CursorSetHints() */ + int skipNext; /* Prev() is noop if negative. Next() is noop if positive. + ** Error code if eState==CURSOR_FAULT */ + Btree *pBtree; /* The Btree to which this cursor belongs */ + Pgno *aOverflow; /* Cache of overflow page locations */ + void *pKey; /* Saved key that was cursor last known position */ + /* All fields above are zeroed when the cursor is allocated. See + ** sqlite3BtreeCursorZero(). Fields that follow must be manually + ** initialized. */ +#define BTCURSOR_FIRST_UNINIT pBt /* Name of first uninitialized field */ + BtShared *pBt; /* The BtShared this cursor points to */ + BtCursor *pNext; /* Forms a linked list of all cursors */ + CellInfo info; /* A parse of the cell we are pointing at */ + i64 nKey; /* Size of pKey, or last integer key */ + Pgno pgnoRoot; /* The root page of this tree */ + i8 iPage; /* Index of current page in apPage */ + u8 curIntKey; /* Value of apPage[0]->intKey */ + u16 ix; /* Current index for apPage[iPage] */ + u16 aiIdx[BTCURSOR_MAX_DEPTH-1]; /* Current index in apPage[i] */ + struct KeyInfo *pKeyInfo; /* Arg passed to comparison function */ + MemPage *pPage; /* Current page */ + MemPage *apPage[BTCURSOR_MAX_DEPTH-1]; /* Stack of parents of current page */ +}; + +/* +** Legal values for BtCursor.curFlags +*/ +#define BTCF_WriteFlag 0x01 /* True if a write cursor */ +#define BTCF_ValidNKey 0x02 /* True if info.nKey is valid */ +#define BTCF_ValidOvfl 0x04 /* True if aOverflow is valid */ +#define BTCF_AtLast 0x08 /* Cursor is pointing ot the last entry */ +#define BTCF_Incrblob 0x10 /* True if an incremental I/O handle */ +#define BTCF_Multiple 0x20 /* Maybe another cursor on the same btree */ + +/* +** Potential values for BtCursor.eState. +** +** CURSOR_INVALID: +** Cursor does not point to a valid entry. This can happen (for example) +** because the table is empty or because BtreeCursorFirst() has not been +** called. +** +** CURSOR_VALID: +** Cursor points to a valid entry. getPayload() etc. may be called. +** +** CURSOR_SKIPNEXT: +** Cursor is valid except that the Cursor.skipNext field is non-zero +** indicating that the next sqlite3BtreeNext() or sqlite3BtreePrevious() +** operation should be a no-op. +** +** CURSOR_REQUIRESEEK: +** The table that this cursor was opened on still exists, but has been +** modified since the cursor was last used. The cursor position is saved +** in variables BtCursor.pKey and BtCursor.nKey. When a cursor is in +** this state, restoreCursorPosition() can be called to attempt to +** seek the cursor to the saved position. +** +** CURSOR_FAULT: +** An unrecoverable error (an I/O error or a malloc failure) has occurred +** on a different connection that shares the BtShared cache with this +** cursor. The error has left the cache in an inconsistent state. +** Do nothing else with this cursor. Any attempt to use the cursor +** should return the error code stored in BtCursor.skipNext +*/ +#define CURSOR_VALID 0 +#define CURSOR_INVALID 1 +#define CURSOR_SKIPNEXT 2 +#define CURSOR_REQUIRESEEK 3 +#define CURSOR_FAULT 4 + +/* +** The database page the PENDING_BYTE occupies. This page is never used. +*/ +# define PENDING_BYTE_PAGE(pBt) PAGER_MJ_PGNO(pBt) + +/* +** These macros define the location of the pointer-map entry for a +** database page. The first argument to each is the number of usable +** bytes on each page of the database (often 1024). The second is the +** page number to look up in the pointer map. +** +** PTRMAP_PAGENO returns the database page number of the pointer-map +** page that stores the required pointer. PTRMAP_PTROFFSET returns +** the offset of the requested map entry. +** +** If the pgno argument passed to PTRMAP_PAGENO is a pointer-map page, +** then pgno is returned. So (pgno==PTRMAP_PAGENO(pgsz, pgno)) can be +** used to test if pgno is a pointer-map page. PTRMAP_ISPAGE implements +** this test. +*/ +#define PTRMAP_PAGENO(pBt, pgno) ptrmapPageno(pBt, pgno) +#define PTRMAP_PTROFFSET(pgptrmap, pgno) (5*(pgno-pgptrmap-1)) +#define PTRMAP_ISPAGE(pBt, pgno) (PTRMAP_PAGENO((pBt),(pgno))==(pgno)) + +/* +** The pointer map is a lookup table that identifies the parent page for +** each child page in the database file. The parent page is the page that +** contains a pointer to the child. Every page in the database contains +** 0 or 1 parent pages. (In this context 'database page' refers +** to any page that is not part of the pointer map itself.) Each pointer map +** entry consists of a single byte 'type' and a 4 byte parent page number. +** The PTRMAP_XXX identifiers below are the valid types. +** +** The purpose of the pointer map is to facility moving pages from one +** position in the file to another as part of autovacuum. When a page +** is moved, the pointer in its parent must be updated to point to the +** new location. The pointer map is used to locate the parent page quickly. +** +** PTRMAP_ROOTPAGE: The database page is a root-page. The page-number is not +** used in this case. +** +** PTRMAP_FREEPAGE: The database page is an unused (free) page. The page-number +** is not used in this case. +** +** PTRMAP_OVERFLOW1: The database page is the first page in a list of +** overflow pages. The page number identifies the page that +** contains the cell with a pointer to this overflow page. +** +** PTRMAP_OVERFLOW2: The database page is the second or later page in a list of +** overflow pages. The page-number identifies the previous +** page in the overflow page list. +** +** PTRMAP_BTREE: The database page is a non-root btree page. The page number +** identifies the parent page in the btree. +*/ +#define PTRMAP_ROOTPAGE 1 +#define PTRMAP_FREEPAGE 2 +#define PTRMAP_OVERFLOW1 3 +#define PTRMAP_OVERFLOW2 4 +#define PTRMAP_BTREE 5 + +/* A bunch of assert() statements to check the transaction state variables +** of handle p (type Btree*) are internally consistent. +*/ +#define btreeIntegrity(p) \ + assert( p->pBt->inTransaction!=TRANS_NONE || p->pBt->nTransaction==0 ); \ + assert( p->pBt->inTransaction>=p->inTrans ); + + +/* +** The ISAUTOVACUUM macro is used within balance_nonroot() to determine +** if the database supports auto-vacuum or not. Because it is used +** within an expression that is an argument to another macro +** (sqliteMallocRaw), it is not possible to use conditional compilation. +** So, this macro is defined instead. +*/ +#ifndef SQLITE_OMIT_AUTOVACUUM +#define ISAUTOVACUUM (pBt->autoVacuum) +#else +#define ISAUTOVACUUM 0 +#endif + + +/* +** This structure is passed around through all the sanity checking routines +** in order to keep track of some global state information. +** +** The aRef[] array is allocated so that there is 1 bit for each page in +** the database. As the integrity-check proceeds, for each page used in +** the database the corresponding bit is set. This allows integrity-check to +** detect pages that are used twice and orphaned pages (both of which +** indicate corruption). +*/ +typedef struct IntegrityCk IntegrityCk; +struct IntegrityCk { + BtShared *pBt; /* The tree being checked out */ + Pager *pPager; /* The associated pager. Also accessible by pBt->pPager */ + u8 *aPgRef; /* 1 bit per page in the db (see above) */ + Pgno nPage; /* Number of pages in the database */ + int mxErr; /* Stop accumulating errors when this reaches zero */ + int nErr; /* Number of messages written to zErrMsg so far */ + int mallocFailed; /* A memory allocation error has occurred */ + const char *zPfx; /* Error message prefix */ + int v1, v2; /* Values for up to two %d fields in zPfx */ + StrAccum errMsg; /* Accumulate the error message text here */ + u32 *heap; /* Min-heap used for analyzing cell coverage */ +}; + +/* +** Routines to read or write a two- and four-byte big-endian integer values. +*/ +#define get2byte(x) ((x)[0]<<8 | (x)[1]) +#define put2byte(p,v) ((p)[0] = (u8)((v)>>8), (p)[1] = (u8)(v)) +#define get4byte sqlite3Get4byte +#define put4byte sqlite3Put4byte + +/* +** get2byteAligned(), unlike get2byte(), requires that its argument point to a +** two-byte aligned address. get2bytea() is only used for accessing the +** cell addresses in a btree header. +*/ +#if SQLITE_BYTEORDER==4321 +# define get2byteAligned(x) (*(u16*)(x)) +#elif SQLITE_BYTEORDER==1234 && GCC_VERSION>=4008000 +# define get2byteAligned(x) __builtin_bswap16(*(u16*)(x)) +#elif SQLITE_BYTEORDER==1234 && MSVC_VERSION>=1300 +# define get2byteAligned(x) _byteswap_ushort(*(u16*)(x)) +#else +# define get2byteAligned(x) ((x)[0]<<8 | (x)[1]) +#endif + +/************** End of btreeInt.h ********************************************/ +/************** Continuing where we left off in btmutex.c ********************/ +#ifndef SQLITE_OMIT_SHARED_CACHE +#if SQLITE_THREADSAFE + +/* +** Obtain the BtShared mutex associated with B-Tree handle p. Also, +** set BtShared.db to the database handle associated with p and the +** p->locked boolean to true. +*/ +static void lockBtreeMutex(Btree *p){ + assert( p->locked==0 ); + assert( sqlite3_mutex_notheld(p->pBt->mutex) ); + assert( sqlite3_mutex_held(p->db->mutex) ); + + sqlite3_mutex_enter(p->pBt->mutex); + p->pBt->db = p->db; + p->locked = 1; +} + +/* +** Release the BtShared mutex associated with B-Tree handle p and +** clear the p->locked boolean. +*/ +static void SQLITE_NOINLINE unlockBtreeMutex(Btree *p){ + BtShared *pBt = p->pBt; + assert( p->locked==1 ); + assert( sqlite3_mutex_held(pBt->mutex) ); + assert( sqlite3_mutex_held(p->db->mutex) ); + assert( p->db==pBt->db ); + + sqlite3_mutex_leave(pBt->mutex); + p->locked = 0; +} + +/* Forward reference */ +static void SQLITE_NOINLINE btreeLockCarefully(Btree *p); + +/* +** Enter a mutex on the given BTree object. +** +** If the object is not sharable, then no mutex is ever required +** and this routine is a no-op. The underlying mutex is non-recursive. +** But we keep a reference count in Btree.wantToLock so the behavior +** of this interface is recursive. +** +** To avoid deadlocks, multiple Btrees are locked in the same order +** by all database connections. The p->pNext is a list of other +** Btrees belonging to the same database connection as the p Btree +** which need to be locked after p. If we cannot get a lock on +** p, then first unlock all of the others on p->pNext, then wait +** for the lock to become available on p, then relock all of the +** subsequent Btrees that desire a lock. +*/ +SQLITE_PRIVATE void sqlite3BtreeEnter(Btree *p){ + /* Some basic sanity checking on the Btree. The list of Btrees + ** connected by pNext and pPrev should be in sorted order by + ** Btree.pBt value. All elements of the list should belong to + ** the same connection. Only shared Btrees are on the list. */ + assert( p->pNext==0 || p->pNext->pBt>p->pBt ); + assert( p->pPrev==0 || p->pPrev->pBtpBt ); + assert( p->pNext==0 || p->pNext->db==p->db ); + assert( p->pPrev==0 || p->pPrev->db==p->db ); + assert( p->sharable || (p->pNext==0 && p->pPrev==0) ); + + /* Check for locking consistency */ + assert( !p->locked || p->wantToLock>0 ); + assert( p->sharable || p->wantToLock==0 ); + + /* We should already hold a lock on the database connection */ + assert( sqlite3_mutex_held(p->db->mutex) ); + + /* Unless the database is sharable and unlocked, then BtShared.db + ** should already be set correctly. */ + assert( (p->locked==0 && p->sharable) || p->pBt->db==p->db ); + + if( !p->sharable ) return; + p->wantToLock++; + if( p->locked ) return; + btreeLockCarefully(p); +} + +/* This is a helper function for sqlite3BtreeLock(). By moving +** complex, but seldom used logic, out of sqlite3BtreeLock() and +** into this routine, we avoid unnecessary stack pointer changes +** and thus help the sqlite3BtreeLock() routine to run much faster +** in the common case. +*/ +static void SQLITE_NOINLINE btreeLockCarefully(Btree *p){ + Btree *pLater; + + /* In most cases, we should be able to acquire the lock we + ** want without having to go through the ascending lock + ** procedure that follows. Just be sure not to block. + */ + if( sqlite3_mutex_try(p->pBt->mutex)==SQLITE_OK ){ + p->pBt->db = p->db; + p->locked = 1; + return; + } + + /* To avoid deadlock, first release all locks with a larger + ** BtShared address. Then acquire our lock. Then reacquire + ** the other BtShared locks that we used to hold in ascending + ** order. + */ + for(pLater=p->pNext; pLater; pLater=pLater->pNext){ + assert( pLater->sharable ); + assert( pLater->pNext==0 || pLater->pNext->pBt>pLater->pBt ); + assert( !pLater->locked || pLater->wantToLock>0 ); + if( pLater->locked ){ + unlockBtreeMutex(pLater); + } + } + lockBtreeMutex(p); + for(pLater=p->pNext; pLater; pLater=pLater->pNext){ + if( pLater->wantToLock ){ + lockBtreeMutex(pLater); + } + } +} + + +/* +** Exit the recursive mutex on a Btree. +*/ +SQLITE_PRIVATE void sqlite3BtreeLeave(Btree *p){ + assert( sqlite3_mutex_held(p->db->mutex) ); + if( p->sharable ){ + assert( p->wantToLock>0 ); + p->wantToLock--; + if( p->wantToLock==0 ){ + unlockBtreeMutex(p); + } + } +} + +#ifndef NDEBUG +/* +** Return true if the BtShared mutex is held on the btree, or if the +** B-Tree is not marked as sharable. +** +** This routine is used only from within assert() statements. +*/ +SQLITE_PRIVATE int sqlite3BtreeHoldsMutex(Btree *p){ + assert( p->sharable==0 || p->locked==0 || p->wantToLock>0 ); + assert( p->sharable==0 || p->locked==0 || p->db==p->pBt->db ); + assert( p->sharable==0 || p->locked==0 || sqlite3_mutex_held(p->pBt->mutex) ); + assert( p->sharable==0 || p->locked==0 || sqlite3_mutex_held(p->db->mutex) ); + + return (p->sharable==0 || p->locked); +} +#endif + + +/* +** Enter the mutex on every Btree associated with a database +** connection. This is needed (for example) prior to parsing +** a statement since we will be comparing table and column names +** against all schemas and we do not want those schemas being +** reset out from under us. +** +** There is a corresponding leave-all procedures. +** +** Enter the mutexes in accending order by BtShared pointer address +** to avoid the possibility of deadlock when two threads with +** two or more btrees in common both try to lock all their btrees +** at the same instant. +*/ +static void SQLITE_NOINLINE btreeEnterAll(sqlite3 *db){ + int i; + int skipOk = 1; + Btree *p; + assert( sqlite3_mutex_held(db->mutex) ); + for(i=0; inDb; i++){ + p = db->aDb[i].pBt; + if( p && p->sharable ){ + sqlite3BtreeEnter(p); + skipOk = 0; + } + } + db->noSharedCache = skipOk; +} +SQLITE_PRIVATE void sqlite3BtreeEnterAll(sqlite3 *db){ + if( db->noSharedCache==0 ) btreeEnterAll(db); +} +static void SQLITE_NOINLINE btreeLeaveAll(sqlite3 *db){ + int i; + Btree *p; + assert( sqlite3_mutex_held(db->mutex) ); + for(i=0; inDb; i++){ + p = db->aDb[i].pBt; + if( p ) sqlite3BtreeLeave(p); + } +} +SQLITE_PRIVATE void sqlite3BtreeLeaveAll(sqlite3 *db){ + if( db->noSharedCache==0 ) btreeLeaveAll(db); +} + +#ifndef NDEBUG +/* +** Return true if the current thread holds the database connection +** mutex and all required BtShared mutexes. +** +** This routine is used inside assert() statements only. +*/ +SQLITE_PRIVATE int sqlite3BtreeHoldsAllMutexes(sqlite3 *db){ + int i; + if( !sqlite3_mutex_held(db->mutex) ){ + return 0; + } + for(i=0; inDb; i++){ + Btree *p; + p = db->aDb[i].pBt; + if( p && p->sharable && + (p->wantToLock==0 || !sqlite3_mutex_held(p->pBt->mutex)) ){ + return 0; + } + } + return 1; +} +#endif /* NDEBUG */ + +#ifndef NDEBUG +/* +** Return true if the correct mutexes are held for accessing the +** db->aDb[iDb].pSchema structure. The mutexes required for schema +** access are: +** +** (1) The mutex on db +** (2) if iDb!=1, then the mutex on db->aDb[iDb].pBt. +** +** If pSchema is not NULL, then iDb is computed from pSchema and +** db using sqlite3SchemaToIndex(). +*/ +SQLITE_PRIVATE int sqlite3SchemaMutexHeld(sqlite3 *db, int iDb, Schema *pSchema){ + Btree *p; + assert( db!=0 ); + if( pSchema ) iDb = sqlite3SchemaToIndex(db, pSchema); + assert( iDb>=0 && iDbnDb ); + if( !sqlite3_mutex_held(db->mutex) ) return 0; + if( iDb==1 ) return 1; + p = db->aDb[iDb].pBt; + assert( p!=0 ); + return p->sharable==0 || p->locked==1; +} +#endif /* NDEBUG */ + +#else /* SQLITE_THREADSAFE>0 above. SQLITE_THREADSAFE==0 below */ +/* +** The following are special cases for mutex enter routines for use +** in single threaded applications that use shared cache. Except for +** these two routines, all mutex operations are no-ops in that case and +** are null #defines in btree.h. +** +** If shared cache is disabled, then all btree mutex routines, including +** the ones below, are no-ops and are null #defines in btree.h. +*/ + +SQLITE_PRIVATE void sqlite3BtreeEnter(Btree *p){ + p->pBt->db = p->db; +} +SQLITE_PRIVATE void sqlite3BtreeEnterAll(sqlite3 *db){ + int i; + for(i=0; inDb; i++){ + Btree *p = db->aDb[i].pBt; + if( p ){ + p->pBt->db = p->db; + } + } +} +#endif /* if SQLITE_THREADSAFE */ + +#ifndef SQLITE_OMIT_INCRBLOB +/* +** Enter a mutex on a Btree given a cursor owned by that Btree. +** +** These entry points are used by incremental I/O only. Enter() is required +** any time OMIT_SHARED_CACHE is not defined, regardless of whether or not +** the build is threadsafe. Leave() is only required by threadsafe builds. +*/ +SQLITE_PRIVATE void sqlite3BtreeEnterCursor(BtCursor *pCur){ + sqlite3BtreeEnter(pCur->pBtree); +} +# if SQLITE_THREADSAFE +SQLITE_PRIVATE void sqlite3BtreeLeaveCursor(BtCursor *pCur){ + sqlite3BtreeLeave(pCur->pBtree); +} +# endif +#endif /* ifndef SQLITE_OMIT_INCRBLOB */ + +#endif /* ifndef SQLITE_OMIT_SHARED_CACHE */ + +/************** End of btmutex.c *********************************************/ +/************** Begin file btree.c *******************************************/ +/* +** 2004 April 6 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** This file implements an external (disk-based) database using BTrees. +** See the header comment on "btreeInt.h" for additional information. +** Including a description of file format and an overview of operation. +*/ +/* #include "btreeInt.h" */ + +/* +** The header string that appears at the beginning of every +** SQLite database. +*/ +static const char zMagicHeader[] = SQLITE_FILE_HEADER; + +/* +** Set this global variable to 1 to enable tracing using the TRACE +** macro. +*/ +#if 0 +int sqlite3BtreeTrace=1; /* True to enable tracing */ +# define TRACE(X) if(sqlite3BtreeTrace){printf X;fflush(stdout);} +#else +# define TRACE(X) +#endif + +/* +** Extract a 2-byte big-endian integer from an array of unsigned bytes. +** But if the value is zero, make it 65536. +** +** This routine is used to extract the "offset to cell content area" value +** from the header of a btree page. If the page size is 65536 and the page +** is empty, the offset should be 65536, but the 2-byte value stores zero. +** This routine makes the necessary adjustment to 65536. +*/ +#define get2byteNotZero(X) (((((int)get2byte(X))-1)&0xffff)+1) + +/* +** Values passed as the 5th argument to allocateBtreePage() +*/ +#define BTALLOC_ANY 0 /* Allocate any page */ +#define BTALLOC_EXACT 1 /* Allocate exact page if possible */ +#define BTALLOC_LE 2 /* Allocate any page <= the parameter */ + +/* +** Macro IfNotOmitAV(x) returns (x) if SQLITE_OMIT_AUTOVACUUM is not +** defined, or 0 if it is. For example: +** +** bIncrVacuum = IfNotOmitAV(pBtShared->incrVacuum); +*/ +#ifndef SQLITE_OMIT_AUTOVACUUM +#define IfNotOmitAV(expr) (expr) +#else +#define IfNotOmitAV(expr) 0 +#endif + +#ifndef SQLITE_OMIT_SHARED_CACHE +/* +** A list of BtShared objects that are eligible for participation +** in shared cache. This variable has file scope during normal builds, +** but the test harness needs to access it so we make it global for +** test builds. +** +** Access to this variable is protected by SQLITE_MUTEX_STATIC_MASTER. +*/ +#ifdef SQLITE_TEST +SQLITE_PRIVATE BtShared *SQLITE_WSD sqlite3SharedCacheList = 0; +#else +static BtShared *SQLITE_WSD sqlite3SharedCacheList = 0; +#endif +#endif /* SQLITE_OMIT_SHARED_CACHE */ + +#ifndef SQLITE_OMIT_SHARED_CACHE +/* +** Enable or disable the shared pager and schema features. +** +** This routine has no effect on existing database connections. +** The shared cache setting effects only future calls to +** sqlite3_open(), sqlite3_open16(), or sqlite3_open_v2(). +*/ +SQLITE_API int sqlite3_enable_shared_cache(int enable){ + sqlite3GlobalConfig.sharedCacheEnabled = enable; + return SQLITE_OK; +} +#endif + + + +#ifdef SQLITE_OMIT_SHARED_CACHE + /* + ** The functions querySharedCacheTableLock(), setSharedCacheTableLock(), + ** and clearAllSharedCacheTableLocks() + ** manipulate entries in the BtShared.pLock linked list used to store + ** shared-cache table level locks. If the library is compiled with the + ** shared-cache feature disabled, then there is only ever one user + ** of each BtShared structure and so this locking is not necessary. + ** So define the lock related functions as no-ops. + */ + #define querySharedCacheTableLock(a,b,c) SQLITE_OK + #define setSharedCacheTableLock(a,b,c) SQLITE_OK + #define clearAllSharedCacheTableLocks(a) + #define downgradeAllSharedCacheTableLocks(a) + #define hasSharedCacheTableLock(a,b,c,d) 1 + #define hasReadConflicts(a, b) 0 +#endif + +/* +** Implementation of the SQLITE_CORRUPT_PAGE() macro. Takes a single +** (MemPage*) as an argument. The (MemPage*) must not be NULL. +** +** If SQLITE_DEBUG is not defined, then this macro is equivalent to +** SQLITE_CORRUPT_BKPT. Or, if SQLITE_DEBUG is set, then the log message +** normally produced as a side-effect of SQLITE_CORRUPT_BKPT is augmented +** with the page number and filename associated with the (MemPage*). +*/ +#ifdef SQLITE_DEBUG +int corruptPageError(int lineno, MemPage *p){ + char *zMsg; + sqlite3BeginBenignMalloc(); + zMsg = sqlite3_mprintf("database corruption page %d of %s", + (int)p->pgno, sqlite3PagerFilename(p->pBt->pPager, 0) + ); + sqlite3EndBenignMalloc(); + if( zMsg ){ + sqlite3ReportError(SQLITE_CORRUPT, lineno, zMsg); + } + sqlite3_free(zMsg); + return SQLITE_CORRUPT_BKPT; +} +# define SQLITE_CORRUPT_PAGE(pMemPage) corruptPageError(__LINE__, pMemPage) +#else +# define SQLITE_CORRUPT_PAGE(pMemPage) SQLITE_CORRUPT_PGNO(pMemPage->pgno) +#endif + +#ifndef SQLITE_OMIT_SHARED_CACHE + +#ifdef SQLITE_DEBUG +/* +**** This function is only used as part of an assert() statement. *** +** +** Check to see if pBtree holds the required locks to read or write to the +** table with root page iRoot. Return 1 if it does and 0 if not. +** +** For example, when writing to a table with root-page iRoot via +** Btree connection pBtree: +** +** assert( hasSharedCacheTableLock(pBtree, iRoot, 0, WRITE_LOCK) ); +** +** When writing to an index that resides in a sharable database, the +** caller should have first obtained a lock specifying the root page of +** the corresponding table. This makes things a bit more complicated, +** as this module treats each table as a separate structure. To determine +** the table corresponding to the index being written, this +** function has to search through the database schema. +** +** Instead of a lock on the table/index rooted at page iRoot, the caller may +** hold a write-lock on the schema table (root page 1). This is also +** acceptable. +*/ +static int hasSharedCacheTableLock( + Btree *pBtree, /* Handle that must hold lock */ + Pgno iRoot, /* Root page of b-tree */ + int isIndex, /* True if iRoot is the root of an index b-tree */ + int eLockType /* Required lock type (READ_LOCK or WRITE_LOCK) */ +){ + Schema *pSchema = (Schema *)pBtree->pBt->pSchema; + Pgno iTab = 0; + BtLock *pLock; + + /* If this database is not shareable, or if the client is reading + ** and has the read-uncommitted flag set, then no lock is required. + ** Return true immediately. + */ + if( (pBtree->sharable==0) + || (eLockType==READ_LOCK && (pBtree->db->flags & SQLITE_ReadUncommit)) + ){ + return 1; + } + + /* If the client is reading or writing an index and the schema is + ** not loaded, then it is too difficult to actually check to see if + ** the correct locks are held. So do not bother - just return true. + ** This case does not come up very often anyhow. + */ + if( isIndex && (!pSchema || (pSchema->schemaFlags&DB_SchemaLoaded)==0) ){ + return 1; + } + + /* Figure out the root-page that the lock should be held on. For table + ** b-trees, this is just the root page of the b-tree being read or + ** written. For index b-trees, it is the root page of the associated + ** table. */ + if( isIndex ){ + HashElem *p; + for(p=sqliteHashFirst(&pSchema->idxHash); p; p=sqliteHashNext(p)){ + Index *pIdx = (Index *)sqliteHashData(p); + if( pIdx->tnum==(int)iRoot ){ + if( iTab ){ + /* Two or more indexes share the same root page. There must + ** be imposter tables. So just return true. The assert is not + ** useful in that case. */ + return 1; + } + iTab = pIdx->pTable->tnum; + } + } + }else{ + iTab = iRoot; + } + + /* Search for the required lock. Either a write-lock on root-page iTab, a + ** write-lock on the schema table, or (if the client is reading) a + ** read-lock on iTab will suffice. Return 1 if any of these are found. */ + for(pLock=pBtree->pBt->pLock; pLock; pLock=pLock->pNext){ + if( pLock->pBtree==pBtree + && (pLock->iTable==iTab || (pLock->eLock==WRITE_LOCK && pLock->iTable==1)) + && pLock->eLock>=eLockType + ){ + return 1; + } + } + + /* Failed to find the required lock. */ + return 0; +} +#endif /* SQLITE_DEBUG */ + +#ifdef SQLITE_DEBUG +/* +**** This function may be used as part of assert() statements only. **** +** +** Return true if it would be illegal for pBtree to write into the +** table or index rooted at iRoot because other shared connections are +** simultaneously reading that same table or index. +** +** It is illegal for pBtree to write if some other Btree object that +** shares the same BtShared object is currently reading or writing +** the iRoot table. Except, if the other Btree object has the +** read-uncommitted flag set, then it is OK for the other object to +** have a read cursor. +** +** For example, before writing to any part of the table or index +** rooted at page iRoot, one should call: +** +** assert( !hasReadConflicts(pBtree, iRoot) ); +*/ +static int hasReadConflicts(Btree *pBtree, Pgno iRoot){ + BtCursor *p; + for(p=pBtree->pBt->pCursor; p; p=p->pNext){ + if( p->pgnoRoot==iRoot + && p->pBtree!=pBtree + && 0==(p->pBtree->db->flags & SQLITE_ReadUncommit) + ){ + return 1; + } + } + return 0; +} +#endif /* #ifdef SQLITE_DEBUG */ + +/* +** Query to see if Btree handle p may obtain a lock of type eLock +** (READ_LOCK or WRITE_LOCK) on the table with root-page iTab. Return +** SQLITE_OK if the lock may be obtained (by calling +** setSharedCacheTableLock()), or SQLITE_LOCKED if not. +*/ +static int querySharedCacheTableLock(Btree *p, Pgno iTab, u8 eLock){ + BtShared *pBt = p->pBt; + BtLock *pIter; + + assert( sqlite3BtreeHoldsMutex(p) ); + assert( eLock==READ_LOCK || eLock==WRITE_LOCK ); + assert( p->db!=0 ); + assert( !(p->db->flags&SQLITE_ReadUncommit)||eLock==WRITE_LOCK||iTab==1 ); + + /* If requesting a write-lock, then the Btree must have an open write + ** transaction on this file. And, obviously, for this to be so there + ** must be an open write transaction on the file itself. + */ + assert( eLock==READ_LOCK || (p==pBt->pWriter && p->inTrans==TRANS_WRITE) ); + assert( eLock==READ_LOCK || pBt->inTransaction==TRANS_WRITE ); + + /* This routine is a no-op if the shared-cache is not enabled */ + if( !p->sharable ){ + return SQLITE_OK; + } + + /* If some other connection is holding an exclusive lock, the + ** requested lock may not be obtained. + */ + if( pBt->pWriter!=p && (pBt->btsFlags & BTS_EXCLUSIVE)!=0 ){ + sqlite3ConnectionBlocked(p->db, pBt->pWriter->db); + return SQLITE_LOCKED_SHAREDCACHE; + } + + for(pIter=pBt->pLock; pIter; pIter=pIter->pNext){ + /* The condition (pIter->eLock!=eLock) in the following if(...) + ** statement is a simplification of: + ** + ** (eLock==WRITE_LOCK || pIter->eLock==WRITE_LOCK) + ** + ** since we know that if eLock==WRITE_LOCK, then no other connection + ** may hold a WRITE_LOCK on any table in this file (since there can + ** only be a single writer). + */ + assert( pIter->eLock==READ_LOCK || pIter->eLock==WRITE_LOCK ); + assert( eLock==READ_LOCK || pIter->pBtree==p || pIter->eLock==READ_LOCK); + if( pIter->pBtree!=p && pIter->iTable==iTab && pIter->eLock!=eLock ){ + sqlite3ConnectionBlocked(p->db, pIter->pBtree->db); + if( eLock==WRITE_LOCK ){ + assert( p==pBt->pWriter ); + pBt->btsFlags |= BTS_PENDING; + } + return SQLITE_LOCKED_SHAREDCACHE; + } + } + return SQLITE_OK; +} +#endif /* !SQLITE_OMIT_SHARED_CACHE */ + +#ifndef SQLITE_OMIT_SHARED_CACHE +/* +** Add a lock on the table with root-page iTable to the shared-btree used +** by Btree handle p. Parameter eLock must be either READ_LOCK or +** WRITE_LOCK. +** +** This function assumes the following: +** +** (a) The specified Btree object p is connected to a sharable +** database (one with the BtShared.sharable flag set), and +** +** (b) No other Btree objects hold a lock that conflicts +** with the requested lock (i.e. querySharedCacheTableLock() has +** already been called and returned SQLITE_OK). +** +** SQLITE_OK is returned if the lock is added successfully. SQLITE_NOMEM +** is returned if a malloc attempt fails. +*/ +static int setSharedCacheTableLock(Btree *p, Pgno iTable, u8 eLock){ + BtShared *pBt = p->pBt; + BtLock *pLock = 0; + BtLock *pIter; + + assert( sqlite3BtreeHoldsMutex(p) ); + assert( eLock==READ_LOCK || eLock==WRITE_LOCK ); + assert( p->db!=0 ); + + /* A connection with the read-uncommitted flag set will never try to + ** obtain a read-lock using this function. The only read-lock obtained + ** by a connection in read-uncommitted mode is on the sqlite_master + ** table, and that lock is obtained in BtreeBeginTrans(). */ + assert( 0==(p->db->flags&SQLITE_ReadUncommit) || eLock==WRITE_LOCK ); + + /* This function should only be called on a sharable b-tree after it + ** has been determined that no other b-tree holds a conflicting lock. */ + assert( p->sharable ); + assert( SQLITE_OK==querySharedCacheTableLock(p, iTable, eLock) ); + + /* First search the list for an existing lock on this table. */ + for(pIter=pBt->pLock; pIter; pIter=pIter->pNext){ + if( pIter->iTable==iTable && pIter->pBtree==p ){ + pLock = pIter; + break; + } + } + + /* If the above search did not find a BtLock struct associating Btree p + ** with table iTable, allocate one and link it into the list. + */ + if( !pLock ){ + pLock = (BtLock *)sqlite3MallocZero(sizeof(BtLock)); + if( !pLock ){ + return SQLITE_NOMEM_BKPT; + } + pLock->iTable = iTable; + pLock->pBtree = p; + pLock->pNext = pBt->pLock; + pBt->pLock = pLock; + } + + /* Set the BtLock.eLock variable to the maximum of the current lock + ** and the requested lock. This means if a write-lock was already held + ** and a read-lock requested, we don't incorrectly downgrade the lock. + */ + assert( WRITE_LOCK>READ_LOCK ); + if( eLock>pLock->eLock ){ + pLock->eLock = eLock; + } + + return SQLITE_OK; +} +#endif /* !SQLITE_OMIT_SHARED_CACHE */ + +#ifndef SQLITE_OMIT_SHARED_CACHE +/* +** Release all the table locks (locks obtained via calls to +** the setSharedCacheTableLock() procedure) held by Btree object p. +** +** This function assumes that Btree p has an open read or write +** transaction. If it does not, then the BTS_PENDING flag +** may be incorrectly cleared. +*/ +static void clearAllSharedCacheTableLocks(Btree *p){ + BtShared *pBt = p->pBt; + BtLock **ppIter = &pBt->pLock; + + assert( sqlite3BtreeHoldsMutex(p) ); + assert( p->sharable || 0==*ppIter ); + assert( p->inTrans>0 ); + + while( *ppIter ){ + BtLock *pLock = *ppIter; + assert( (pBt->btsFlags & BTS_EXCLUSIVE)==0 || pBt->pWriter==pLock->pBtree ); + assert( pLock->pBtree->inTrans>=pLock->eLock ); + if( pLock->pBtree==p ){ + *ppIter = pLock->pNext; + assert( pLock->iTable!=1 || pLock==&p->lock ); + if( pLock->iTable!=1 ){ + sqlite3_free(pLock); + } + }else{ + ppIter = &pLock->pNext; + } + } + + assert( (pBt->btsFlags & BTS_PENDING)==0 || pBt->pWriter ); + if( pBt->pWriter==p ){ + pBt->pWriter = 0; + pBt->btsFlags &= ~(BTS_EXCLUSIVE|BTS_PENDING); + }else if( pBt->nTransaction==2 ){ + /* This function is called when Btree p is concluding its + ** transaction. If there currently exists a writer, and p is not + ** that writer, then the number of locks held by connections other + ** than the writer must be about to drop to zero. In this case + ** set the BTS_PENDING flag to 0. + ** + ** If there is not currently a writer, then BTS_PENDING must + ** be zero already. So this next line is harmless in that case. + */ + pBt->btsFlags &= ~BTS_PENDING; + } +} + +/* +** This function changes all write-locks held by Btree p into read-locks. +*/ +static void downgradeAllSharedCacheTableLocks(Btree *p){ + BtShared *pBt = p->pBt; + if( pBt->pWriter==p ){ + BtLock *pLock; + pBt->pWriter = 0; + pBt->btsFlags &= ~(BTS_EXCLUSIVE|BTS_PENDING); + for(pLock=pBt->pLock; pLock; pLock=pLock->pNext){ + assert( pLock->eLock==READ_LOCK || pLock->pBtree==p ); + pLock->eLock = READ_LOCK; + } + } +} + +#endif /* SQLITE_OMIT_SHARED_CACHE */ + +static void releasePage(MemPage *pPage); /* Forward reference */ +static void releasePageOne(MemPage *pPage); /* Forward reference */ +static void releasePageNotNull(MemPage *pPage); /* Forward reference */ + +/* +***** This routine is used inside of assert() only **** +** +** Verify that the cursor holds the mutex on its BtShared +*/ +#ifdef SQLITE_DEBUG +static int cursorHoldsMutex(BtCursor *p){ + return sqlite3_mutex_held(p->pBt->mutex); +} + +/* Verify that the cursor and the BtShared agree about what is the current +** database connetion. This is important in shared-cache mode. If the database +** connection pointers get out-of-sync, it is possible for routines like +** btreeInitPage() to reference an stale connection pointer that references a +** a connection that has already closed. This routine is used inside assert() +** statements only and for the purpose of double-checking that the btree code +** does keep the database connection pointers up-to-date. +*/ +static int cursorOwnsBtShared(BtCursor *p){ + assert( cursorHoldsMutex(p) ); + return (p->pBtree->db==p->pBt->db); +} +#endif + +/* +** Invalidate the overflow cache of the cursor passed as the first argument. +** on the shared btree structure pBt. +*/ +#define invalidateOverflowCache(pCur) (pCur->curFlags &= ~BTCF_ValidOvfl) + +/* +** Invalidate the overflow page-list cache for all cursors opened +** on the shared btree structure pBt. +*/ +static void invalidateAllOverflowCache(BtShared *pBt){ + BtCursor *p; + assert( sqlite3_mutex_held(pBt->mutex) ); + for(p=pBt->pCursor; p; p=p->pNext){ + invalidateOverflowCache(p); + } +} + +#ifndef SQLITE_OMIT_INCRBLOB +/* +** This function is called before modifying the contents of a table +** to invalidate any incrblob cursors that are open on the +** row or one of the rows being modified. +** +** If argument isClearTable is true, then the entire contents of the +** table is about to be deleted. In this case invalidate all incrblob +** cursors open on any row within the table with root-page pgnoRoot. +** +** Otherwise, if argument isClearTable is false, then the row with +** rowid iRow is being replaced or deleted. In this case invalidate +** only those incrblob cursors open on that specific row. +*/ +static void invalidateIncrblobCursors( + Btree *pBtree, /* The database file to check */ + Pgno pgnoRoot, /* The table that might be changing */ + i64 iRow, /* The rowid that might be changing */ + int isClearTable /* True if all rows are being deleted */ +){ + BtCursor *p; + if( pBtree->hasIncrblobCur==0 ) return; + assert( sqlite3BtreeHoldsMutex(pBtree) ); + pBtree->hasIncrblobCur = 0; + for(p=pBtree->pBt->pCursor; p; p=p->pNext){ + if( (p->curFlags & BTCF_Incrblob)!=0 ){ + pBtree->hasIncrblobCur = 1; + if( p->pgnoRoot==pgnoRoot && (isClearTable || p->info.nKey==iRow) ){ + p->eState = CURSOR_INVALID; + } + } + } +} + +#else + /* Stub function when INCRBLOB is omitted */ + #define invalidateIncrblobCursors(w,x,y,z) +#endif /* SQLITE_OMIT_INCRBLOB */ + +/* +** Set bit pgno of the BtShared.pHasContent bitvec. This is called +** when a page that previously contained data becomes a free-list leaf +** page. +** +** The BtShared.pHasContent bitvec exists to work around an obscure +** bug caused by the interaction of two useful IO optimizations surrounding +** free-list leaf pages: +** +** 1) When all data is deleted from a page and the page becomes +** a free-list leaf page, the page is not written to the database +** (as free-list leaf pages contain no meaningful data). Sometimes +** such a page is not even journalled (as it will not be modified, +** why bother journalling it?). +** +** 2) When a free-list leaf page is reused, its content is not read +** from the database or written to the journal file (why should it +** be, if it is not at all meaningful?). +** +** By themselves, these optimizations work fine and provide a handy +** performance boost to bulk delete or insert operations. However, if +** a page is moved to the free-list and then reused within the same +** transaction, a problem comes up. If the page is not journalled when +** it is moved to the free-list and it is also not journalled when it +** is extracted from the free-list and reused, then the original data +** may be lost. In the event of a rollback, it may not be possible +** to restore the database to its original configuration. +** +** The solution is the BtShared.pHasContent bitvec. Whenever a page is +** moved to become a free-list leaf page, the corresponding bit is +** set in the bitvec. Whenever a leaf page is extracted from the free-list, +** optimization 2 above is omitted if the corresponding bit is already +** set in BtShared.pHasContent. The contents of the bitvec are cleared +** at the end of every transaction. +*/ +static int btreeSetHasContent(BtShared *pBt, Pgno pgno){ + int rc = SQLITE_OK; + if( !pBt->pHasContent ){ + assert( pgno<=pBt->nPage ); + pBt->pHasContent = sqlite3BitvecCreate(pBt->nPage); + if( !pBt->pHasContent ){ + rc = SQLITE_NOMEM_BKPT; + } + } + if( rc==SQLITE_OK && pgno<=sqlite3BitvecSize(pBt->pHasContent) ){ + rc = sqlite3BitvecSet(pBt->pHasContent, pgno); + } + return rc; +} + +/* +** Query the BtShared.pHasContent vector. +** +** This function is called when a free-list leaf page is removed from the +** free-list for reuse. It returns false if it is safe to retrieve the +** page from the pager layer with the 'no-content' flag set. True otherwise. +*/ +static int btreeGetHasContent(BtShared *pBt, Pgno pgno){ + Bitvec *p = pBt->pHasContent; + return (p && (pgno>sqlite3BitvecSize(p) || sqlite3BitvecTest(p, pgno))); +} + +/* +** Clear (destroy) the BtShared.pHasContent bitvec. This should be +** invoked at the conclusion of each write-transaction. +*/ +static void btreeClearHasContent(BtShared *pBt){ + sqlite3BitvecDestroy(pBt->pHasContent); + pBt->pHasContent = 0; +} + +/* +** Release all of the apPage[] pages for a cursor. +*/ +static void btreeReleaseAllCursorPages(BtCursor *pCur){ + int i; + if( pCur->iPage>=0 ){ + for(i=0; iiPage; i++){ + releasePageNotNull(pCur->apPage[i]); + } + releasePageNotNull(pCur->pPage); + pCur->iPage = -1; + } +} + +/* +** The cursor passed as the only argument must point to a valid entry +** when this function is called (i.e. have eState==CURSOR_VALID). This +** function saves the current cursor key in variables pCur->nKey and +** pCur->pKey. SQLITE_OK is returned if successful or an SQLite error +** code otherwise. +** +** If the cursor is open on an intkey table, then the integer key +** (the rowid) is stored in pCur->nKey and pCur->pKey is left set to +** NULL. If the cursor is open on a non-intkey table, then pCur->pKey is +** set to point to a malloced buffer pCur->nKey bytes in size containing +** the key. +*/ +static int saveCursorKey(BtCursor *pCur){ + int rc = SQLITE_OK; + assert( CURSOR_VALID==pCur->eState ); + assert( 0==pCur->pKey ); + assert( cursorHoldsMutex(pCur) ); + + if( pCur->curIntKey ){ + /* Only the rowid is required for a table btree */ + pCur->nKey = sqlite3BtreeIntegerKey(pCur); + }else{ + /* For an index btree, save the complete key content. It is possible + ** that the current key is corrupt. In that case, it is possible that + ** the sqlite3VdbeRecordUnpack() function may overread the buffer by + ** up to the size of 1 varint plus 1 8-byte value when the cursor + ** position is restored. Hence the 17 bytes of padding allocated + ** below. */ + void *pKey; + pCur->nKey = sqlite3BtreePayloadSize(pCur); + pKey = sqlite3Malloc( pCur->nKey + 9 + 8 ); + if( pKey ){ + rc = sqlite3BtreePayload(pCur, 0, (int)pCur->nKey, pKey); + if( rc==SQLITE_OK ){ + memset(((u8*)pKey)+pCur->nKey, 0, 9+8); + pCur->pKey = pKey; + }else{ + sqlite3_free(pKey); + } + }else{ + rc = SQLITE_NOMEM_BKPT; + } + } + assert( !pCur->curIntKey || !pCur->pKey ); + return rc; +} + +/* +** Save the current cursor position in the variables BtCursor.nKey +** and BtCursor.pKey. The cursor's state is set to CURSOR_REQUIRESEEK. +** +** The caller must ensure that the cursor is valid (has eState==CURSOR_VALID) +** prior to calling this routine. +*/ +static int saveCursorPosition(BtCursor *pCur){ + int rc; + + assert( CURSOR_VALID==pCur->eState || CURSOR_SKIPNEXT==pCur->eState ); + assert( 0==pCur->pKey ); + assert( cursorHoldsMutex(pCur) ); + + if( pCur->eState==CURSOR_SKIPNEXT ){ + pCur->eState = CURSOR_VALID; + }else{ + pCur->skipNext = 0; + } + + rc = saveCursorKey(pCur); + if( rc==SQLITE_OK ){ + btreeReleaseAllCursorPages(pCur); + pCur->eState = CURSOR_REQUIRESEEK; + } + + pCur->curFlags &= ~(BTCF_ValidNKey|BTCF_ValidOvfl|BTCF_AtLast); + return rc; +} + +/* Forward reference */ +static int SQLITE_NOINLINE saveCursorsOnList(BtCursor*,Pgno,BtCursor*); + +/* +** Save the positions of all cursors (except pExcept) that are open on +** the table with root-page iRoot. "Saving the cursor position" means that +** the location in the btree is remembered in such a way that it can be +** moved back to the same spot after the btree has been modified. This +** routine is called just before cursor pExcept is used to modify the +** table, for example in BtreeDelete() or BtreeInsert(). +** +** If there are two or more cursors on the same btree, then all such +** cursors should have their BTCF_Multiple flag set. The btreeCursor() +** routine enforces that rule. This routine only needs to be called in +** the uncommon case when pExpect has the BTCF_Multiple flag set. +** +** If pExpect!=NULL and if no other cursors are found on the same root-page, +** then the BTCF_Multiple flag on pExpect is cleared, to avoid another +** pointless call to this routine. +** +** Implementation note: This routine merely checks to see if any cursors +** need to be saved. It calls out to saveCursorsOnList() in the (unusual) +** event that cursors are in need to being saved. +*/ +static int saveAllCursors(BtShared *pBt, Pgno iRoot, BtCursor *pExcept){ + BtCursor *p; + assert( sqlite3_mutex_held(pBt->mutex) ); + assert( pExcept==0 || pExcept->pBt==pBt ); + for(p=pBt->pCursor; p; p=p->pNext){ + if( p!=pExcept && (0==iRoot || p->pgnoRoot==iRoot) ) break; + } + if( p ) return saveCursorsOnList(p, iRoot, pExcept); + if( pExcept ) pExcept->curFlags &= ~BTCF_Multiple; + return SQLITE_OK; +} + +/* This helper routine to saveAllCursors does the actual work of saving +** the cursors if and when a cursor is found that actually requires saving. +** The common case is that no cursors need to be saved, so this routine is +** broken out from its caller to avoid unnecessary stack pointer movement. +*/ +static int SQLITE_NOINLINE saveCursorsOnList( + BtCursor *p, /* The first cursor that needs saving */ + Pgno iRoot, /* Only save cursor with this iRoot. Save all if zero */ + BtCursor *pExcept /* Do not save this cursor */ +){ + do{ + if( p!=pExcept && (0==iRoot || p->pgnoRoot==iRoot) ){ + if( p->eState==CURSOR_VALID || p->eState==CURSOR_SKIPNEXT ){ + int rc = saveCursorPosition(p); + if( SQLITE_OK!=rc ){ + return rc; + } + }else{ + testcase( p->iPage>=0 ); + btreeReleaseAllCursorPages(p); + } + } + p = p->pNext; + }while( p ); + return SQLITE_OK; +} + +/* +** Clear the current cursor position. +*/ +SQLITE_PRIVATE void sqlite3BtreeClearCursor(BtCursor *pCur){ + assert( cursorHoldsMutex(pCur) ); + sqlite3_free(pCur->pKey); + pCur->pKey = 0; + pCur->eState = CURSOR_INVALID; +} + +/* +** In this version of BtreeMoveto, pKey is a packed index record +** such as is generated by the OP_MakeRecord opcode. Unpack the +** record and then call BtreeMovetoUnpacked() to do the work. +*/ +static int btreeMoveto( + BtCursor *pCur, /* Cursor open on the btree to be searched */ + const void *pKey, /* Packed key if the btree is an index */ + i64 nKey, /* Integer key for tables. Size of pKey for indices */ + int bias, /* Bias search to the high end */ + int *pRes /* Write search results here */ +){ + int rc; /* Status code */ + UnpackedRecord *pIdxKey; /* Unpacked index key */ + + if( pKey ){ + KeyInfo *pKeyInfo = pCur->pKeyInfo; + assert( nKey==(i64)(int)nKey ); + pIdxKey = sqlite3VdbeAllocUnpackedRecord(pKeyInfo); + if( pIdxKey==0 ) return SQLITE_NOMEM_BKPT; + sqlite3VdbeRecordUnpack(pKeyInfo, (int)nKey, pKey, pIdxKey); + if( pIdxKey->nField==0 || pIdxKey->nField>pKeyInfo->nAllField ){ + rc = SQLITE_CORRUPT_BKPT; + goto moveto_done; + } + }else{ + pIdxKey = 0; + } + rc = sqlite3BtreeMovetoUnpacked(pCur, pIdxKey, nKey, bias, pRes); +moveto_done: + if( pIdxKey ){ + sqlite3DbFree(pCur->pKeyInfo->db, pIdxKey); + } + return rc; +} + +/* +** Restore the cursor to the position it was in (or as close to as possible) +** when saveCursorPosition() was called. Note that this call deletes the +** saved position info stored by saveCursorPosition(), so there can be +** at most one effective restoreCursorPosition() call after each +** saveCursorPosition(). +*/ +static int btreeRestoreCursorPosition(BtCursor *pCur){ + int rc; + int skipNext = 0; + assert( cursorOwnsBtShared(pCur) ); + assert( pCur->eState>=CURSOR_REQUIRESEEK ); + if( pCur->eState==CURSOR_FAULT ){ + return pCur->skipNext; + } + pCur->eState = CURSOR_INVALID; + if( sqlite3FaultSim(410) ){ + rc = SQLITE_IOERR; + }else{ + rc = btreeMoveto(pCur, pCur->pKey, pCur->nKey, 0, &skipNext); + } + if( rc==SQLITE_OK ){ + sqlite3_free(pCur->pKey); + pCur->pKey = 0; + assert( pCur->eState==CURSOR_VALID || pCur->eState==CURSOR_INVALID ); + if( skipNext ) pCur->skipNext = skipNext; + if( pCur->skipNext && pCur->eState==CURSOR_VALID ){ + pCur->eState = CURSOR_SKIPNEXT; + } + } + return rc; +} + +#define restoreCursorPosition(p) \ + (p->eState>=CURSOR_REQUIRESEEK ? \ + btreeRestoreCursorPosition(p) : \ + SQLITE_OK) + +/* +** Determine whether or not a cursor has moved from the position where +** it was last placed, or has been invalidated for any other reason. +** Cursors can move when the row they are pointing at is deleted out +** from under them, for example. Cursor might also move if a btree +** is rebalanced. +** +** Calling this routine with a NULL cursor pointer returns false. +** +** Use the separate sqlite3BtreeCursorRestore() routine to restore a cursor +** back to where it ought to be if this routine returns true. +*/ +SQLITE_PRIVATE int sqlite3BtreeCursorHasMoved(BtCursor *pCur){ + assert( EIGHT_BYTE_ALIGNMENT(pCur) + || pCur==sqlite3BtreeFakeValidCursor() ); + assert( offsetof(BtCursor, eState)==0 ); + assert( sizeof(pCur->eState)==1 ); + return CURSOR_VALID != *(u8*)pCur; +} + +/* +** Return a pointer to a fake BtCursor object that will always answer +** false to the sqlite3BtreeCursorHasMoved() routine above. The fake +** cursor returned must not be used with any other Btree interface. +*/ +SQLITE_PRIVATE BtCursor *sqlite3BtreeFakeValidCursor(void){ + static u8 fakeCursor = CURSOR_VALID; + assert( offsetof(BtCursor, eState)==0 ); + return (BtCursor*)&fakeCursor; +} + +/* +** This routine restores a cursor back to its original position after it +** has been moved by some outside activity (such as a btree rebalance or +** a row having been deleted out from under the cursor). +** +** On success, the *pDifferentRow parameter is false if the cursor is left +** pointing at exactly the same row. *pDifferntRow is the row the cursor +** was pointing to has been deleted, forcing the cursor to point to some +** nearby row. +** +** This routine should only be called for a cursor that just returned +** TRUE from sqlite3BtreeCursorHasMoved(). +*/ +SQLITE_PRIVATE int sqlite3BtreeCursorRestore(BtCursor *pCur, int *pDifferentRow){ + int rc; + + assert( pCur!=0 ); + assert( pCur->eState!=CURSOR_VALID ); + rc = restoreCursorPosition(pCur); + if( rc ){ + *pDifferentRow = 1; + return rc; + } + if( pCur->eState!=CURSOR_VALID ){ + *pDifferentRow = 1; + }else{ + *pDifferentRow = 0; + } + return SQLITE_OK; +} + +#ifdef SQLITE_ENABLE_CURSOR_HINTS +/* +** Provide hints to the cursor. The particular hint given (and the type +** and number of the varargs parameters) is determined by the eHintType +** parameter. See the definitions of the BTREE_HINT_* macros for details. +*/ +SQLITE_PRIVATE void sqlite3BtreeCursorHint(BtCursor *pCur, int eHintType, ...){ + /* Used only by system that substitute their own storage engine */ +} +#endif + +/* +** Provide flag hints to the cursor. +*/ +SQLITE_PRIVATE void sqlite3BtreeCursorHintFlags(BtCursor *pCur, unsigned x){ + assert( x==BTREE_SEEK_EQ || x==BTREE_BULKLOAD || x==0 ); + pCur->hints = x; +} + + +#ifndef SQLITE_OMIT_AUTOVACUUM +/* +** Given a page number of a regular database page, return the page +** number for the pointer-map page that contains the entry for the +** input page number. +** +** Return 0 (not a valid page) for pgno==1 since there is +** no pointer map associated with page 1. The integrity_check logic +** requires that ptrmapPageno(*,1)!=1. +*/ +static Pgno ptrmapPageno(BtShared *pBt, Pgno pgno){ + int nPagesPerMapPage; + Pgno iPtrMap, ret; + assert( sqlite3_mutex_held(pBt->mutex) ); + if( pgno<2 ) return 0; + nPagesPerMapPage = (pBt->usableSize/5)+1; + iPtrMap = (pgno-2)/nPagesPerMapPage; + ret = (iPtrMap*nPagesPerMapPage) + 2; + if( ret==PENDING_BYTE_PAGE(pBt) ){ + ret++; + } + return ret; +} + +/* +** Write an entry into the pointer map. +** +** This routine updates the pointer map entry for page number 'key' +** so that it maps to type 'eType' and parent page number 'pgno'. +** +** If *pRC is initially non-zero (non-SQLITE_OK) then this routine is +** a no-op. If an error occurs, the appropriate error code is written +** into *pRC. +*/ +static void ptrmapPut(BtShared *pBt, Pgno key, u8 eType, Pgno parent, int *pRC){ + DbPage *pDbPage; /* The pointer map page */ + u8 *pPtrmap; /* The pointer map data */ + Pgno iPtrmap; /* The pointer map page number */ + int offset; /* Offset in pointer map page */ + int rc; /* Return code from subfunctions */ + + if( *pRC ) return; + + assert( sqlite3_mutex_held(pBt->mutex) ); + /* The master-journal page number must never be used as a pointer map page */ + assert( 0==PTRMAP_ISPAGE(pBt, PENDING_BYTE_PAGE(pBt)) ); + + assert( pBt->autoVacuum ); + if( key==0 ){ + *pRC = SQLITE_CORRUPT_BKPT; + return; + } + iPtrmap = PTRMAP_PAGENO(pBt, key); + rc = sqlite3PagerGet(pBt->pPager, iPtrmap, &pDbPage, 0); + if( rc!=SQLITE_OK ){ + *pRC = rc; + return; + } + if( ((char*)sqlite3PagerGetExtra(pDbPage))[0]!=0 ){ + /* The first byte of the extra data is the MemPage.isInit byte. + ** If that byte is set, it means this page is also being used + ** as a btree page. */ + *pRC = SQLITE_CORRUPT_BKPT; + goto ptrmap_exit; + } + offset = PTRMAP_PTROFFSET(iPtrmap, key); + if( offset<0 ){ + *pRC = SQLITE_CORRUPT_BKPT; + goto ptrmap_exit; + } + assert( offset <= (int)pBt->usableSize-5 ); + pPtrmap = (u8 *)sqlite3PagerGetData(pDbPage); + + if( eType!=pPtrmap[offset] || get4byte(&pPtrmap[offset+1])!=parent ){ + TRACE(("PTRMAP_UPDATE: %d->(%d,%d)\n", key, eType, parent)); + *pRC= rc = sqlite3PagerWrite(pDbPage); + if( rc==SQLITE_OK ){ + pPtrmap[offset] = eType; + put4byte(&pPtrmap[offset+1], parent); + } + } + +ptrmap_exit: + sqlite3PagerUnref(pDbPage); +} + +/* +** Read an entry from the pointer map. +** +** This routine retrieves the pointer map entry for page 'key', writing +** the type and parent page number to *pEType and *pPgno respectively. +** An error code is returned if something goes wrong, otherwise SQLITE_OK. +*/ +static int ptrmapGet(BtShared *pBt, Pgno key, u8 *pEType, Pgno *pPgno){ + DbPage *pDbPage; /* The pointer map page */ + int iPtrmap; /* Pointer map page index */ + u8 *pPtrmap; /* Pointer map page data */ + int offset; /* Offset of entry in pointer map */ + int rc; + + assert( sqlite3_mutex_held(pBt->mutex) ); + + iPtrmap = PTRMAP_PAGENO(pBt, key); + rc = sqlite3PagerGet(pBt->pPager, iPtrmap, &pDbPage, 0); + if( rc!=0 ){ + return rc; + } + pPtrmap = (u8 *)sqlite3PagerGetData(pDbPage); + + offset = PTRMAP_PTROFFSET(iPtrmap, key); + if( offset<0 ){ + sqlite3PagerUnref(pDbPage); + return SQLITE_CORRUPT_BKPT; + } + assert( offset <= (int)pBt->usableSize-5 ); + assert( pEType!=0 ); + *pEType = pPtrmap[offset]; + if( pPgno ) *pPgno = get4byte(&pPtrmap[offset+1]); + + sqlite3PagerUnref(pDbPage); + if( *pEType<1 || *pEType>5 ) return SQLITE_CORRUPT_PGNO(iPtrmap); + return SQLITE_OK; +} + +#else /* if defined SQLITE_OMIT_AUTOVACUUM */ + #define ptrmapPut(w,x,y,z,rc) + #define ptrmapGet(w,x,y,z) SQLITE_OK + #define ptrmapPutOvflPtr(x, y, z, rc) +#endif + +/* +** Given a btree page and a cell index (0 means the first cell on +** the page, 1 means the second cell, and so forth) return a pointer +** to the cell content. +** +** findCellPastPtr() does the same except it skips past the initial +** 4-byte child pointer found on interior pages, if there is one. +** +** This routine works only for pages that do not contain overflow cells. +*/ +#define findCell(P,I) \ + ((P)->aData + ((P)->maskPage & get2byteAligned(&(P)->aCellIdx[2*(I)]))) +#define findCellPastPtr(P,I) \ + ((P)->aDataOfst + ((P)->maskPage & get2byteAligned(&(P)->aCellIdx[2*(I)]))) + + +/* +** This is common tail processing for btreeParseCellPtr() and +** btreeParseCellPtrIndex() for the case when the cell does not fit entirely +** on a single B-tree page. Make necessary adjustments to the CellInfo +** structure. +*/ +static SQLITE_NOINLINE void btreeParseCellAdjustSizeForOverflow( + MemPage *pPage, /* Page containing the cell */ + u8 *pCell, /* Pointer to the cell text. */ + CellInfo *pInfo /* Fill in this structure */ +){ + /* If the payload will not fit completely on the local page, we have + ** to decide how much to store locally and how much to spill onto + ** overflow pages. The strategy is to minimize the amount of unused + ** space on overflow pages while keeping the amount of local storage + ** in between minLocal and maxLocal. + ** + ** Warning: changing the way overflow payload is distributed in any + ** way will result in an incompatible file format. + */ + int minLocal; /* Minimum amount of payload held locally */ + int maxLocal; /* Maximum amount of payload held locally */ + int surplus; /* Overflow payload available for local storage */ + + minLocal = pPage->minLocal; + maxLocal = pPage->maxLocal; + surplus = minLocal + (pInfo->nPayload - minLocal)%(pPage->pBt->usableSize-4); + testcase( surplus==maxLocal ); + testcase( surplus==maxLocal+1 ); + if( surplus <= maxLocal ){ + pInfo->nLocal = (u16)surplus; + }else{ + pInfo->nLocal = (u16)minLocal; + } + pInfo->nSize = (u16)(&pInfo->pPayload[pInfo->nLocal] - pCell) + 4; +} + +/* +** The following routines are implementations of the MemPage.xParseCell() +** method. +** +** Parse a cell content block and fill in the CellInfo structure. +** +** btreeParseCellPtr() => table btree leaf nodes +** btreeParseCellNoPayload() => table btree internal nodes +** btreeParseCellPtrIndex() => index btree nodes +** +** There is also a wrapper function btreeParseCell() that works for +** all MemPage types and that references the cell by index rather than +** by pointer. +*/ +static void btreeParseCellPtrNoPayload( + MemPage *pPage, /* Page containing the cell */ + u8 *pCell, /* Pointer to the cell text. */ + CellInfo *pInfo /* Fill in this structure */ +){ + assert( sqlite3_mutex_held(pPage->pBt->mutex) ); + assert( pPage->leaf==0 ); + assert( pPage->childPtrSize==4 ); +#ifndef SQLITE_DEBUG + UNUSED_PARAMETER(pPage); +#endif + pInfo->nSize = 4 + getVarint(&pCell[4], (u64*)&pInfo->nKey); + pInfo->nPayload = 0; + pInfo->nLocal = 0; + pInfo->pPayload = 0; + return; +} +static void btreeParseCellPtr( + MemPage *pPage, /* Page containing the cell */ + u8 *pCell, /* Pointer to the cell text. */ + CellInfo *pInfo /* Fill in this structure */ +){ + u8 *pIter; /* For scanning through pCell */ + u32 nPayload; /* Number of bytes of cell payload */ + u64 iKey; /* Extracted Key value */ + + assert( sqlite3_mutex_held(pPage->pBt->mutex) ); + assert( pPage->leaf==0 || pPage->leaf==1 ); + assert( pPage->intKeyLeaf ); + assert( pPage->childPtrSize==0 ); + pIter = pCell; + + /* The next block of code is equivalent to: + ** + ** pIter += getVarint32(pIter, nPayload); + ** + ** The code is inlined to avoid a function call. + */ + nPayload = *pIter; + if( nPayload>=0x80 ){ + u8 *pEnd = &pIter[8]; + nPayload &= 0x7f; + do{ + nPayload = (nPayload<<7) | (*++pIter & 0x7f); + }while( (*pIter)>=0x80 && pIternKey); + ** + ** The code is inlined to avoid a function call. + */ + iKey = *pIter; + if( iKey>=0x80 ){ + u8 *pEnd = &pIter[7]; + iKey &= 0x7f; + while(1){ + iKey = (iKey<<7) | (*++pIter & 0x7f); + if( (*pIter)<0x80 ) break; + if( pIter>=pEnd ){ + iKey = (iKey<<8) | *++pIter; + break; + } + } + } + pIter++; + + pInfo->nKey = *(i64*)&iKey; + pInfo->nPayload = nPayload; + pInfo->pPayload = pIter; + testcase( nPayload==pPage->maxLocal ); + testcase( nPayload==pPage->maxLocal+1 ); + if( nPayload<=pPage->maxLocal ){ + /* This is the (easy) common case where the entire payload fits + ** on the local page. No overflow is required. + */ + pInfo->nSize = nPayload + (u16)(pIter - pCell); + if( pInfo->nSize<4 ) pInfo->nSize = 4; + pInfo->nLocal = (u16)nPayload; + }else{ + btreeParseCellAdjustSizeForOverflow(pPage, pCell, pInfo); + } +} +static void btreeParseCellPtrIndex( + MemPage *pPage, /* Page containing the cell */ + u8 *pCell, /* Pointer to the cell text. */ + CellInfo *pInfo /* Fill in this structure */ +){ + u8 *pIter; /* For scanning through pCell */ + u32 nPayload; /* Number of bytes of cell payload */ + + assert( sqlite3_mutex_held(pPage->pBt->mutex) ); + assert( pPage->leaf==0 || pPage->leaf==1 ); + assert( pPage->intKeyLeaf==0 ); + pIter = pCell + pPage->childPtrSize; + nPayload = *pIter; + if( nPayload>=0x80 ){ + u8 *pEnd = &pIter[8]; + nPayload &= 0x7f; + do{ + nPayload = (nPayload<<7) | (*++pIter & 0x7f); + }while( *(pIter)>=0x80 && pIternKey = nPayload; + pInfo->nPayload = nPayload; + pInfo->pPayload = pIter; + testcase( nPayload==pPage->maxLocal ); + testcase( nPayload==pPage->maxLocal+1 ); + if( nPayload<=pPage->maxLocal ){ + /* This is the (easy) common case where the entire payload fits + ** on the local page. No overflow is required. + */ + pInfo->nSize = nPayload + (u16)(pIter - pCell); + if( pInfo->nSize<4 ) pInfo->nSize = 4; + pInfo->nLocal = (u16)nPayload; + }else{ + btreeParseCellAdjustSizeForOverflow(pPage, pCell, pInfo); + } +} +static void btreeParseCell( + MemPage *pPage, /* Page containing the cell */ + int iCell, /* The cell index. First cell is 0 */ + CellInfo *pInfo /* Fill in this structure */ +){ + pPage->xParseCell(pPage, findCell(pPage, iCell), pInfo); +} + +/* +** The following routines are implementations of the MemPage.xCellSize +** method. +** +** Compute the total number of bytes that a Cell needs in the cell +** data area of the btree-page. The return number includes the cell +** data header and the local payload, but not any overflow page or +** the space used by the cell pointer. +** +** cellSizePtrNoPayload() => table internal nodes +** cellSizePtr() => all index nodes & table leaf nodes +*/ +static u16 cellSizePtr(MemPage *pPage, u8 *pCell){ + u8 *pIter = pCell + pPage->childPtrSize; /* For looping over bytes of pCell */ + u8 *pEnd; /* End mark for a varint */ + u32 nSize; /* Size value to return */ + +#ifdef SQLITE_DEBUG + /* The value returned by this function should always be the same as + ** the (CellInfo.nSize) value found by doing a full parse of the + ** cell. If SQLITE_DEBUG is defined, an assert() at the bottom of + ** this function verifies that this invariant is not violated. */ + CellInfo debuginfo; + pPage->xParseCell(pPage, pCell, &debuginfo); +#endif + + nSize = *pIter; + if( nSize>=0x80 ){ + pEnd = &pIter[8]; + nSize &= 0x7f; + do{ + nSize = (nSize<<7) | (*++pIter & 0x7f); + }while( *(pIter)>=0x80 && pIterintKey ){ + /* pIter now points at the 64-bit integer key value, a variable length + ** integer. The following block moves pIter to point at the first byte + ** past the end of the key value. */ + pEnd = &pIter[9]; + while( (*pIter++)&0x80 && pItermaxLocal ); + testcase( nSize==pPage->maxLocal+1 ); + if( nSize<=pPage->maxLocal ){ + nSize += (u32)(pIter - pCell); + if( nSize<4 ) nSize = 4; + }else{ + int minLocal = pPage->minLocal; + nSize = minLocal + (nSize - minLocal) % (pPage->pBt->usableSize - 4); + testcase( nSize==pPage->maxLocal ); + testcase( nSize==pPage->maxLocal+1 ); + if( nSize>pPage->maxLocal ){ + nSize = minLocal; + } + nSize += 4 + (u16)(pIter - pCell); + } + assert( nSize==debuginfo.nSize || CORRUPT_DB ); + return (u16)nSize; +} +static u16 cellSizePtrNoPayload(MemPage *pPage, u8 *pCell){ + u8 *pIter = pCell + 4; /* For looping over bytes of pCell */ + u8 *pEnd; /* End mark for a varint */ + +#ifdef SQLITE_DEBUG + /* The value returned by this function should always be the same as + ** the (CellInfo.nSize) value found by doing a full parse of the + ** cell. If SQLITE_DEBUG is defined, an assert() at the bottom of + ** this function verifies that this invariant is not violated. */ + CellInfo debuginfo; + pPage->xParseCell(pPage, pCell, &debuginfo); +#else + UNUSED_PARAMETER(pPage); +#endif + + assert( pPage->childPtrSize==4 ); + pEnd = pIter + 9; + while( (*pIter++)&0x80 && pIterxCellSize(pPage, findCell(pPage, iCell)); +} +#endif + +#ifndef SQLITE_OMIT_AUTOVACUUM +/* +** The cell pCell is currently part of page pSrc but will ultimately be part +** of pPage. (pSrc and pPager are often the same.) If pCell contains a +** pointer to an overflow page, insert an entry into the pointer-map for +** the overflow page that will be valid after pCell has been moved to pPage. +*/ +static void ptrmapPutOvflPtr(MemPage *pPage, MemPage *pSrc, u8 *pCell,int *pRC){ + CellInfo info; + if( *pRC ) return; + assert( pCell!=0 ); + pPage->xParseCell(pPage, pCell, &info); + if( info.nLocalaDataEnd, pCell, pCell+info.nLocal) ){ + testcase( pSrc!=pPage ); + *pRC = SQLITE_CORRUPT_BKPT; + return; + } + ovfl = get4byte(&pCell[info.nSize-4]); + ptrmapPut(pPage->pBt, ovfl, PTRMAP_OVERFLOW1, pPage->pgno, pRC); + } +} +#endif + + +/* +** Defragment the page given. This routine reorganizes cells within the +** page so that there are no free-blocks on the free-block list. +** +** Parameter nMaxFrag is the maximum amount of fragmented space that may be +** present in the page after this routine returns. +** +** EVIDENCE-OF: R-44582-60138 SQLite may from time to time reorganize a +** b-tree page so that there are no freeblocks or fragment bytes, all +** unused bytes are contained in the unallocated space region, and all +** cells are packed tightly at the end of the page. +*/ +static int defragmentPage(MemPage *pPage, int nMaxFrag){ + int i; /* Loop counter */ + int pc; /* Address of the i-th cell */ + int hdr; /* Offset to the page header */ + int size; /* Size of a cell */ + int usableSize; /* Number of usable bytes on a page */ + int cellOffset; /* Offset to the cell pointer array */ + int cbrk; /* Offset to the cell content area */ + int nCell; /* Number of cells on the page */ + unsigned char *data; /* The page data */ + unsigned char *temp; /* Temp area for cell content */ + unsigned char *src; /* Source of content */ + int iCellFirst; /* First allowable cell index */ + int iCellLast; /* Last possible cell index */ + + assert( sqlite3PagerIswriteable(pPage->pDbPage) ); + assert( pPage->pBt!=0 ); + assert( pPage->pBt->usableSize <= SQLITE_MAX_PAGE_SIZE ); + assert( pPage->nOverflow==0 ); + assert( sqlite3_mutex_held(pPage->pBt->mutex) ); + temp = 0; + src = data = pPage->aData; + hdr = pPage->hdrOffset; + cellOffset = pPage->cellOffset; + nCell = pPage->nCell; + assert( nCell==get2byte(&data[hdr+3]) || CORRUPT_DB ); + iCellFirst = cellOffset + 2*nCell; + usableSize = pPage->pBt->usableSize; + + /* This block handles pages with two or fewer free blocks and nMaxFrag + ** or fewer fragmented bytes. In this case it is faster to move the + ** two (or one) blocks of cells using memmove() and add the required + ** offsets to each pointer in the cell-pointer array than it is to + ** reconstruct the entire page. */ + if( (int)data[hdr+7]<=nMaxFrag ){ + int iFree = get2byte(&data[hdr+1]); + if( iFree>usableSize-4 ) return SQLITE_CORRUPT_PAGE(pPage); + if( iFree ){ + int iFree2 = get2byte(&data[iFree]); + if( iFree2>usableSize-4 ) return SQLITE_CORRUPT_PAGE(pPage); + if( 0==iFree2 || (data[iFree2]==0 && data[iFree2+1]==0) ){ + u8 *pEnd = &data[cellOffset + nCell*2]; + u8 *pAddr; + int sz2 = 0; + int sz = get2byte(&data[iFree+2]); + int top = get2byte(&data[hdr+5]); + if( top>=iFree ){ + return SQLITE_CORRUPT_PAGE(pPage); + } + if( iFree2 ){ + if( iFree+sz>iFree2 ) return SQLITE_CORRUPT_PAGE(pPage); + sz2 = get2byte(&data[iFree2+2]); + if( iFree2+sz2 > usableSize ) return SQLITE_CORRUPT_PAGE(pPage); + memmove(&data[iFree+sz+sz2], &data[iFree+sz], iFree2-(iFree+sz)); + sz += sz2; + }else if( iFree+sz>usableSize ){ + return SQLITE_CORRUPT_PAGE(pPage); + } + + cbrk = top+sz; + assert( cbrk+(iFree-top) <= usableSize ); + memmove(&data[cbrk], &data[top], iFree-top); + for(pAddr=&data[cellOffset]; pAddriCellLast ){ + return SQLITE_CORRUPT_PAGE(pPage); + } + assert( pc>=iCellFirst && pc<=iCellLast ); + size = pPage->xCellSize(pPage, &src[pc]); + cbrk -= size; + if( cbrkusableSize ){ + return SQLITE_CORRUPT_PAGE(pPage); + } + assert( cbrk+size<=usableSize && cbrk>=iCellFirst ); + testcase( cbrk+size==usableSize ); + testcase( pc+size==usableSize ); + put2byte(pAddr, cbrk); + if( temp==0 ){ + int x; + if( cbrk==pc ) continue; + temp = sqlite3PagerTempSpace(pPage->pBt->pPager); + x = get2byte(&data[hdr+5]); + memcpy(&temp[x], &data[x], (cbrk+size) - x); + src = temp; + } + memcpy(&data[cbrk], &src[pc], size); + } + data[hdr+7] = 0; + + defragment_out: + assert( pPage->nFree>=0 ); + if( data[hdr+7]+cbrk-iCellFirst!=pPage->nFree ){ + return SQLITE_CORRUPT_PAGE(pPage); + } + assert( cbrk>=iCellFirst ); + put2byte(&data[hdr+5], cbrk); + data[hdr+1] = 0; + data[hdr+2] = 0; + memset(&data[iCellFirst], 0, cbrk-iCellFirst); + assert( sqlite3PagerIswriteable(pPage->pDbPage) ); + return SQLITE_OK; +} + +/* +** Search the free-list on page pPg for space to store a cell nByte bytes in +** size. If one can be found, return a pointer to the space and remove it +** from the free-list. +** +** If no suitable space can be found on the free-list, return NULL. +** +** This function may detect corruption within pPg. If corruption is +** detected then *pRc is set to SQLITE_CORRUPT and NULL is returned. +** +** Slots on the free list that are between 1 and 3 bytes larger than nByte +** will be ignored if adding the extra space to the fragmentation count +** causes the fragmentation count to exceed 60. +*/ +static u8 *pageFindSlot(MemPage *pPg, int nByte, int *pRc){ + const int hdr = pPg->hdrOffset; /* Offset to page header */ + u8 * const aData = pPg->aData; /* Page data */ + int iAddr = hdr + 1; /* Address of ptr to pc */ + int pc = get2byte(&aData[iAddr]); /* Address of a free slot */ + int x; /* Excess size of the slot */ + int maxPC = pPg->pBt->usableSize - nByte; /* Max address for a usable slot */ + int size; /* Size of the free slot */ + + assert( pc>0 ); + while( pc<=maxPC ){ + /* EVIDENCE-OF: R-22710-53328 The third and fourth bytes of each + ** freeblock form a big-endian integer which is the size of the freeblock + ** in bytes, including the 4-byte header. */ + size = get2byte(&aData[pc+2]); + if( (x = size - nByte)>=0 ){ + testcase( x==4 ); + testcase( x==3 ); + if( x<4 ){ + /* EVIDENCE-OF: R-11498-58022 In a well-formed b-tree page, the total + ** number of bytes in fragments may not exceed 60. */ + if( aData[hdr+7]>57 ) return 0; + + /* Remove the slot from the free-list. Update the number of + ** fragmented bytes within the page. */ + memcpy(&aData[iAddr], &aData[pc], 2); + aData[hdr+7] += (u8)x; + }else if( x+pc > maxPC ){ + /* This slot extends off the end of the usable part of the page */ + *pRc = SQLITE_CORRUPT_PAGE(pPg); + return 0; + }else{ + /* The slot remains on the free-list. Reduce its size to account + ** for the portion used by the new allocation. */ + put2byte(&aData[pc+2], x); + } + return &aData[pc + x]; + } + iAddr = pc; + pc = get2byte(&aData[pc]); + if( pc<=iAddr+size ){ + if( pc ){ + /* The next slot in the chain is not past the end of the current slot */ + *pRc = SQLITE_CORRUPT_PAGE(pPg); + } + return 0; + } + } + if( pc>maxPC+nByte-4 ){ + /* The free slot chain extends off the end of the page */ + *pRc = SQLITE_CORRUPT_PAGE(pPg); + } + return 0; +} + +/* +** Allocate nByte bytes of space from within the B-Tree page passed +** as the first argument. Write into *pIdx the index into pPage->aData[] +** of the first byte of allocated space. Return either SQLITE_OK or +** an error code (usually SQLITE_CORRUPT). +** +** The caller guarantees that there is sufficient space to make the +** allocation. This routine might need to defragment in order to bring +** all the space together, however. This routine will avoid using +** the first two bytes past the cell pointer area since presumably this +** allocation is being made in order to insert a new cell, so we will +** also end up needing a new cell pointer. +*/ +static int allocateSpace(MemPage *pPage, int nByte, int *pIdx){ + const int hdr = pPage->hdrOffset; /* Local cache of pPage->hdrOffset */ + u8 * const data = pPage->aData; /* Local cache of pPage->aData */ + int top; /* First byte of cell content area */ + int rc = SQLITE_OK; /* Integer return code */ + int gap; /* First byte of gap between cell pointers and cell content */ + + assert( sqlite3PagerIswriteable(pPage->pDbPage) ); + assert( pPage->pBt ); + assert( sqlite3_mutex_held(pPage->pBt->mutex) ); + assert( nByte>=0 ); /* Minimum cell size is 4 */ + assert( pPage->nFree>=nByte ); + assert( pPage->nOverflow==0 ); + assert( nByte < (int)(pPage->pBt->usableSize-8) ); + + assert( pPage->cellOffset == hdr + 12 - 4*pPage->leaf ); + gap = pPage->cellOffset + 2*pPage->nCell; + assert( gap<=65536 ); + /* EVIDENCE-OF: R-29356-02391 If the database uses a 65536-byte page size + ** and the reserved space is zero (the usual value for reserved space) + ** then the cell content offset of an empty page wants to be 65536. + ** However, that integer is too large to be stored in a 2-byte unsigned + ** integer, so a value of 0 is used in its place. */ + top = get2byte(&data[hdr+5]); + assert( top<=(int)pPage->pBt->usableSize ); /* by btreeComputeFreeSpace() */ + if( gap>top ){ + if( top==0 && pPage->pBt->usableSize==65536 ){ + top = 65536; + }else{ + return SQLITE_CORRUPT_PAGE(pPage); + } + } + + /* If there is enough space between gap and top for one more cell pointer, + ** and if the freelist is not empty, then search the + ** freelist looking for a slot big enough to satisfy the request. + */ + testcase( gap+2==top ); + testcase( gap+1==top ); + testcase( gap==top ); + if( (data[hdr+2] || data[hdr+1]) && gap+2<=top ){ + u8 *pSpace = pageFindSlot(pPage, nByte, &rc); + if( pSpace ){ + assert( pSpace+nByte<=data+pPage->pBt->usableSize ); + if( (*pIdx = (int)(pSpace-data))<=gap ){ + return SQLITE_CORRUPT_PAGE(pPage); + }else{ + return SQLITE_OK; + } + }else if( rc ){ + return rc; + } + } + + /* The request could not be fulfilled using a freelist slot. Check + ** to see if defragmentation is necessary. + */ + testcase( gap+2+nByte==top ); + if( gap+2+nByte>top ){ + assert( pPage->nCell>0 || CORRUPT_DB ); + assert( pPage->nFree>=0 ); + rc = defragmentPage(pPage, MIN(4, pPage->nFree - (2+nByte))); + if( rc ) return rc; + top = get2byteNotZero(&data[hdr+5]); + assert( gap+2+nByte<=top ); + } + + + /* Allocate memory from the gap in between the cell pointer array + ** and the cell content area. The btreeComputeFreeSpace() call has already + ** validated the freelist. Given that the freelist is valid, there + ** is no way that the allocation can extend off the end of the page. + ** The assert() below verifies the previous sentence. + */ + top -= nByte; + put2byte(&data[hdr+5], top); + assert( top+nByte <= (int)pPage->pBt->usableSize ); + *pIdx = top; + return SQLITE_OK; +} + +/* +** Return a section of the pPage->aData to the freelist. +** The first byte of the new free block is pPage->aData[iStart] +** and the size of the block is iSize bytes. +** +** Adjacent freeblocks are coalesced. +** +** Even though the freeblock list was checked by btreeComputeFreeSpace(), +** that routine will not detect overlap between cells or freeblocks. Nor +** does it detect cells or freeblocks that encrouch into the reserved bytes +** at the end of the page. So do additional corruption checks inside this +** routine and return SQLITE_CORRUPT if any problems are found. +*/ +static int freeSpace(MemPage *pPage, u16 iStart, u16 iSize){ + u16 iPtr; /* Address of ptr to next freeblock */ + u16 iFreeBlk; /* Address of the next freeblock */ + u8 hdr; /* Page header size. 0 or 100 */ + u8 nFrag = 0; /* Reduction in fragmentation */ + u16 iOrigSize = iSize; /* Original value of iSize */ + u16 x; /* Offset to cell content area */ + u32 iEnd = iStart + iSize; /* First byte past the iStart buffer */ + unsigned char *data = pPage->aData; /* Page content */ + + assert( pPage->pBt!=0 ); + assert( sqlite3PagerIswriteable(pPage->pDbPage) ); + assert( CORRUPT_DB || iStart>=pPage->hdrOffset+6+pPage->childPtrSize ); + assert( CORRUPT_DB || iEnd <= pPage->pBt->usableSize ); + assert( sqlite3_mutex_held(pPage->pBt->mutex) ); + assert( iSize>=4 ); /* Minimum cell size is 4 */ + assert( iStart<=pPage->pBt->usableSize-4 ); + + /* The list of freeblocks must be in ascending order. Find the + ** spot on the list where iStart should be inserted. + */ + hdr = pPage->hdrOffset; + iPtr = hdr + 1; + if( data[iPtr+1]==0 && data[iPtr]==0 ){ + iFreeBlk = 0; /* Shortcut for the case when the freelist is empty */ + }else{ + while( (iFreeBlk = get2byte(&data[iPtr]))pPage->pBt->usableSize-4 ){ + return SQLITE_CORRUPT_PAGE(pPage); + } + assert( iFreeBlk>iPtr || iFreeBlk==0 ); + + /* At this point: + ** iFreeBlk: First freeblock after iStart, or zero if none + ** iPtr: The address of a pointer to iFreeBlk + ** + ** Check to see if iFreeBlk should be coalesced onto the end of iStart. + */ + if( iFreeBlk && iEnd+3>=iFreeBlk ){ + nFrag = iFreeBlk - iEnd; + if( iEnd>iFreeBlk ) return SQLITE_CORRUPT_PAGE(pPage); + iEnd = iFreeBlk + get2byte(&data[iFreeBlk+2]); + if( iEnd > pPage->pBt->usableSize ){ + return SQLITE_CORRUPT_PAGE(pPage); + } + iSize = iEnd - iStart; + iFreeBlk = get2byte(&data[iFreeBlk]); + } + + /* If iPtr is another freeblock (that is, if iPtr is not the freelist + ** pointer in the page header) then check to see if iStart should be + ** coalesced onto the end of iPtr. + */ + if( iPtr>hdr+1 ){ + int iPtrEnd = iPtr + get2byte(&data[iPtr+2]); + if( iPtrEnd+3>=iStart ){ + if( iPtrEnd>iStart ) return SQLITE_CORRUPT_PAGE(pPage); + nFrag += iStart - iPtrEnd; + iSize = iEnd - iPtr; + iStart = iPtr; + } + } + if( nFrag>data[hdr+7] ) return SQLITE_CORRUPT_PAGE(pPage); + data[hdr+7] -= nFrag; + } + x = get2byte(&data[hdr+5]); + if( iStart<=x ){ + /* The new freeblock is at the beginning of the cell content area, + ** so just extend the cell content area rather than create another + ** freelist entry */ + if( iStartpBt->btsFlags & BTS_FAST_SECURE ){ + /* Overwrite deleted information with zeros when the secure_delete + ** option is enabled */ + memset(&data[iStart], 0, iSize); + } + put2byte(&data[iStart], iFreeBlk); + put2byte(&data[iStart+2], iSize); + pPage->nFree += iOrigSize; + return SQLITE_OK; +} + +/* +** Decode the flags byte (the first byte of the header) for a page +** and initialize fields of the MemPage structure accordingly. +** +** Only the following combinations are supported. Anything different +** indicates a corrupt database files: +** +** PTF_ZERODATA +** PTF_ZERODATA | PTF_LEAF +** PTF_LEAFDATA | PTF_INTKEY +** PTF_LEAFDATA | PTF_INTKEY | PTF_LEAF +*/ +static int decodeFlags(MemPage *pPage, int flagByte){ + BtShared *pBt; /* A copy of pPage->pBt */ + + assert( pPage->hdrOffset==(pPage->pgno==1 ? 100 : 0) ); + assert( sqlite3_mutex_held(pPage->pBt->mutex) ); + pPage->leaf = (u8)(flagByte>>3); assert( PTF_LEAF == 1<<3 ); + flagByte &= ~PTF_LEAF; + pPage->childPtrSize = 4-4*pPage->leaf; + pPage->xCellSize = cellSizePtr; + pBt = pPage->pBt; + if( flagByte==(PTF_LEAFDATA | PTF_INTKEY) ){ + /* EVIDENCE-OF: R-07291-35328 A value of 5 (0x05) means the page is an + ** interior table b-tree page. */ + assert( (PTF_LEAFDATA|PTF_INTKEY)==5 ); + /* EVIDENCE-OF: R-26900-09176 A value of 13 (0x0d) means the page is a + ** leaf table b-tree page. */ + assert( (PTF_LEAFDATA|PTF_INTKEY|PTF_LEAF)==13 ); + pPage->intKey = 1; + if( pPage->leaf ){ + pPage->intKeyLeaf = 1; + pPage->xParseCell = btreeParseCellPtr; + }else{ + pPage->intKeyLeaf = 0; + pPage->xCellSize = cellSizePtrNoPayload; + pPage->xParseCell = btreeParseCellPtrNoPayload; + } + pPage->maxLocal = pBt->maxLeaf; + pPage->minLocal = pBt->minLeaf; + }else if( flagByte==PTF_ZERODATA ){ + /* EVIDENCE-OF: R-43316-37308 A value of 2 (0x02) means the page is an + ** interior index b-tree page. */ + assert( (PTF_ZERODATA)==2 ); + /* EVIDENCE-OF: R-59615-42828 A value of 10 (0x0a) means the page is a + ** leaf index b-tree page. */ + assert( (PTF_ZERODATA|PTF_LEAF)==10 ); + pPage->intKey = 0; + pPage->intKeyLeaf = 0; + pPage->xParseCell = btreeParseCellPtrIndex; + pPage->maxLocal = pBt->maxLocal; + pPage->minLocal = pBt->minLocal; + }else{ + /* EVIDENCE-OF: R-47608-56469 Any other value for the b-tree page type is + ** an error. */ + return SQLITE_CORRUPT_PAGE(pPage); + } + pPage->max1bytePayload = pBt->max1bytePayload; + return SQLITE_OK; +} + +/* +** Compute the amount of freespace on the page. In other words, fill +** in the pPage->nFree field. +*/ +static int btreeComputeFreeSpace(MemPage *pPage){ + int pc; /* Address of a freeblock within pPage->aData[] */ + u8 hdr; /* Offset to beginning of page header */ + u8 *data; /* Equal to pPage->aData */ + int usableSize; /* Amount of usable space on each page */ + int nFree; /* Number of unused bytes on the page */ + int top; /* First byte of the cell content area */ + int iCellFirst; /* First allowable cell or freeblock offset */ + int iCellLast; /* Last possible cell or freeblock offset */ + + assert( pPage->pBt!=0 ); + assert( pPage->pBt->db!=0 ); + assert( sqlite3_mutex_held(pPage->pBt->mutex) ); + assert( pPage->pgno==sqlite3PagerPagenumber(pPage->pDbPage) ); + assert( pPage == sqlite3PagerGetExtra(pPage->pDbPage) ); + assert( pPage->aData == sqlite3PagerGetData(pPage->pDbPage) ); + assert( pPage->isInit==1 ); + assert( pPage->nFree<0 ); + + usableSize = pPage->pBt->usableSize; + hdr = pPage->hdrOffset; + data = pPage->aData; + /* EVIDENCE-OF: R-58015-48175 The two-byte integer at offset 5 designates + ** the start of the cell content area. A zero value for this integer is + ** interpreted as 65536. */ + top = get2byteNotZero(&data[hdr+5]); + iCellFirst = hdr + 8 + pPage->childPtrSize + 2*pPage->nCell; + iCellLast = usableSize - 4; + + /* Compute the total free space on the page + ** EVIDENCE-OF: R-23588-34450 The two-byte integer at offset 1 gives the + ** start of the first freeblock on the page, or is zero if there are no + ** freeblocks. */ + pc = get2byte(&data[hdr+1]); + nFree = data[hdr+7] + top; /* Init nFree to non-freeblock free space */ + if( pc>0 ){ + u32 next, size; + if( pciCellLast ){ + /* Freeblock off the end of the page */ + return SQLITE_CORRUPT_PAGE(pPage); + } + next = get2byte(&data[pc]); + size = get2byte(&data[pc+2]); + nFree = nFree + size; + if( next<=pc+size+3 ) break; + pc = next; + } + if( next>0 ){ + /* Freeblock not in ascending order */ + return SQLITE_CORRUPT_PAGE(pPage); + } + if( pc+size>(unsigned int)usableSize ){ + /* Last freeblock extends past page end */ + return SQLITE_CORRUPT_PAGE(pPage); + } + } + + /* At this point, nFree contains the sum of the offset to the start + ** of the cell-content area plus the number of free bytes within + ** the cell-content area. If this is greater than the usable-size + ** of the page, then the page must be corrupted. This check also + ** serves to verify that the offset to the start of the cell-content + ** area, according to the page header, lies within the page. + */ + if( nFree>usableSize || nFreenFree = (u16)(nFree - iCellFirst); + return SQLITE_OK; +} + +/* +** Do additional sanity check after btreeInitPage() if +** PRAGMA cell_size_check=ON +*/ +static SQLITE_NOINLINE int btreeCellSizeCheck(MemPage *pPage){ + int iCellFirst; /* First allowable cell or freeblock offset */ + int iCellLast; /* Last possible cell or freeblock offset */ + int i; /* Index into the cell pointer array */ + int sz; /* Size of a cell */ + int pc; /* Address of a freeblock within pPage->aData[] */ + u8 *data; /* Equal to pPage->aData */ + int usableSize; /* Maximum usable space on the page */ + int cellOffset; /* Start of cell content area */ + + iCellFirst = pPage->cellOffset + 2*pPage->nCell; + usableSize = pPage->pBt->usableSize; + iCellLast = usableSize - 4; + data = pPage->aData; + cellOffset = pPage->cellOffset; + if( !pPage->leaf ) iCellLast--; + for(i=0; inCell; i++){ + pc = get2byteAligned(&data[cellOffset+i*2]); + testcase( pc==iCellFirst ); + testcase( pc==iCellLast ); + if( pciCellLast ){ + return SQLITE_CORRUPT_PAGE(pPage); + } + sz = pPage->xCellSize(pPage, &data[pc]); + testcase( pc+sz==usableSize ); + if( pc+sz>usableSize ){ + return SQLITE_CORRUPT_PAGE(pPage); + } + } + return SQLITE_OK; +} + +/* +** Initialize the auxiliary information for a disk block. +** +** Return SQLITE_OK on success. If we see that the page does +** not contain a well-formed database page, then return +** SQLITE_CORRUPT. Note that a return of SQLITE_OK does not +** guarantee that the page is well-formed. It only shows that +** we failed to detect any corruption. +*/ +static int btreeInitPage(MemPage *pPage){ + u8 *data; /* Equal to pPage->aData */ + BtShared *pBt; /* The main btree structure */ + + assert( pPage->pBt!=0 ); + assert( pPage->pBt->db!=0 ); + assert( sqlite3_mutex_held(pPage->pBt->mutex) ); + assert( pPage->pgno==sqlite3PagerPagenumber(pPage->pDbPage) ); + assert( pPage == sqlite3PagerGetExtra(pPage->pDbPage) ); + assert( pPage->aData == sqlite3PagerGetData(pPage->pDbPage) ); + assert( pPage->isInit==0 ); + + pBt = pPage->pBt; + data = pPage->aData + pPage->hdrOffset; + /* EVIDENCE-OF: R-28594-02890 The one-byte flag at offset 0 indicating + ** the b-tree page type. */ + if( decodeFlags(pPage, data[0]) ){ + return SQLITE_CORRUPT_PAGE(pPage); + } + assert( pBt->pageSize>=512 && pBt->pageSize<=65536 ); + pPage->maskPage = (u16)(pBt->pageSize - 1); + pPage->nOverflow = 0; + pPage->cellOffset = pPage->hdrOffset + 8 + pPage->childPtrSize; + pPage->aCellIdx = data + pPage->childPtrSize + 8; + pPage->aDataEnd = pPage->aData + pBt->usableSize; + pPage->aDataOfst = pPage->aData + pPage->childPtrSize; + /* EVIDENCE-OF: R-37002-32774 The two-byte integer at offset 3 gives the + ** number of cells on the page. */ + pPage->nCell = get2byte(&data[3]); + if( pPage->nCell>MX_CELL(pBt) ){ + /* To many cells for a single page. The page must be corrupt */ + return SQLITE_CORRUPT_PAGE(pPage); + } + testcase( pPage->nCell==MX_CELL(pBt) ); + /* EVIDENCE-OF: R-24089-57979 If a page contains no cells (which is only + ** possible for a root page of a table that contains no rows) then the + ** offset to the cell content area will equal the page size minus the + ** bytes of reserved space. */ + assert( pPage->nCell>0 + || get2byteNotZero(&data[5])==(int)pBt->usableSize + || CORRUPT_DB ); + pPage->nFree = -1; /* Indicate that this value is yet uncomputed */ + pPage->isInit = 1; + if( pBt->db->flags & SQLITE_CellSizeCk ){ + return btreeCellSizeCheck(pPage); + } + return SQLITE_OK; +} + +/* +** Set up a raw page so that it looks like a database page holding +** no entries. +*/ +static void zeroPage(MemPage *pPage, int flags){ + unsigned char *data = pPage->aData; + BtShared *pBt = pPage->pBt; + u8 hdr = pPage->hdrOffset; + u16 first; + + assert( sqlite3PagerPagenumber(pPage->pDbPage)==pPage->pgno ); + assert( sqlite3PagerGetExtra(pPage->pDbPage) == (void*)pPage ); + assert( sqlite3PagerGetData(pPage->pDbPage) == data ); + assert( sqlite3PagerIswriteable(pPage->pDbPage) ); + assert( sqlite3_mutex_held(pBt->mutex) ); + if( pBt->btsFlags & BTS_FAST_SECURE ){ + memset(&data[hdr], 0, pBt->usableSize - hdr); + } + data[hdr] = (char)flags; + first = hdr + ((flags&PTF_LEAF)==0 ? 12 : 8); + memset(&data[hdr+1], 0, 4); + data[hdr+7] = 0; + put2byte(&data[hdr+5], pBt->usableSize); + pPage->nFree = (u16)(pBt->usableSize - first); + decodeFlags(pPage, flags); + pPage->cellOffset = first; + pPage->aDataEnd = &data[pBt->usableSize]; + pPage->aCellIdx = &data[first]; + pPage->aDataOfst = &data[pPage->childPtrSize]; + pPage->nOverflow = 0; + assert( pBt->pageSize>=512 && pBt->pageSize<=65536 ); + pPage->maskPage = (u16)(pBt->pageSize - 1); + pPage->nCell = 0; + pPage->isInit = 1; +} + + +/* +** Convert a DbPage obtained from the pager into a MemPage used by +** the btree layer. +*/ +static MemPage *btreePageFromDbPage(DbPage *pDbPage, Pgno pgno, BtShared *pBt){ + MemPage *pPage = (MemPage*)sqlite3PagerGetExtra(pDbPage); + if( pgno!=pPage->pgno ){ + pPage->aData = sqlite3PagerGetData(pDbPage); + pPage->pDbPage = pDbPage; + pPage->pBt = pBt; + pPage->pgno = pgno; + pPage->hdrOffset = pgno==1 ? 100 : 0; + } + assert( pPage->aData==sqlite3PagerGetData(pDbPage) ); + return pPage; +} + +/* +** Get a page from the pager. Initialize the MemPage.pBt and +** MemPage.aData elements if needed. See also: btreeGetUnusedPage(). +** +** If the PAGER_GET_NOCONTENT flag is set, it means that we do not care +** about the content of the page at this time. So do not go to the disk +** to fetch the content. Just fill in the content with zeros for now. +** If in the future we call sqlite3PagerWrite() on this page, that +** means we have started to be concerned about content and the disk +** read should occur at that point. +*/ +static int btreeGetPage( + BtShared *pBt, /* The btree */ + Pgno pgno, /* Number of the page to fetch */ + MemPage **ppPage, /* Return the page in this parameter */ + int flags /* PAGER_GET_NOCONTENT or PAGER_GET_READONLY */ +){ + int rc; + DbPage *pDbPage; + + assert( flags==0 || flags==PAGER_GET_NOCONTENT || flags==PAGER_GET_READONLY ); + assert( sqlite3_mutex_held(pBt->mutex) ); + rc = sqlite3PagerGet(pBt->pPager, pgno, (DbPage**)&pDbPage, flags); + if( rc ) return rc; + *ppPage = btreePageFromDbPage(pDbPage, pgno, pBt); + return SQLITE_OK; +} + +/* +** Retrieve a page from the pager cache. If the requested page is not +** already in the pager cache return NULL. Initialize the MemPage.pBt and +** MemPage.aData elements if needed. +*/ +static MemPage *btreePageLookup(BtShared *pBt, Pgno pgno){ + DbPage *pDbPage; + assert( sqlite3_mutex_held(pBt->mutex) ); + pDbPage = sqlite3PagerLookup(pBt->pPager, pgno); + if( pDbPage ){ + return btreePageFromDbPage(pDbPage, pgno, pBt); + } + return 0; +} + +/* +** Return the size of the database file in pages. If there is any kind of +** error, return ((unsigned int)-1). +*/ +static Pgno btreePagecount(BtShared *pBt){ + return pBt->nPage; +} +SQLITE_PRIVATE u32 sqlite3BtreeLastPage(Btree *p){ + assert( sqlite3BtreeHoldsMutex(p) ); + assert( ((p->pBt->nPage)&0x80000000)==0 ); + return btreePagecount(p->pBt); +} + +/* +** Get a page from the pager and initialize it. +** +** If pCur!=0 then the page is being fetched as part of a moveToChild() +** call. Do additional sanity checking on the page in this case. +** And if the fetch fails, this routine must decrement pCur->iPage. +** +** The page is fetched as read-write unless pCur is not NULL and is +** a read-only cursor. +** +** If an error occurs, then *ppPage is undefined. It +** may remain unchanged, or it may be set to an invalid value. +*/ +static int getAndInitPage( + BtShared *pBt, /* The database file */ + Pgno pgno, /* Number of the page to get */ + MemPage **ppPage, /* Write the page pointer here */ + BtCursor *pCur, /* Cursor to receive the page, or NULL */ + int bReadOnly /* True for a read-only page */ +){ + int rc; + DbPage *pDbPage; + assert( sqlite3_mutex_held(pBt->mutex) ); + assert( pCur==0 || ppPage==&pCur->pPage ); + assert( pCur==0 || bReadOnly==pCur->curPagerFlags ); + assert( pCur==0 || pCur->iPage>0 ); + + if( pgno>btreePagecount(pBt) ){ + rc = SQLITE_CORRUPT_BKPT; + goto getAndInitPage_error1; + } + rc = sqlite3PagerGet(pBt->pPager, pgno, (DbPage**)&pDbPage, bReadOnly); + if( rc ){ + goto getAndInitPage_error1; + } + *ppPage = (MemPage*)sqlite3PagerGetExtra(pDbPage); + if( (*ppPage)->isInit==0 ){ + btreePageFromDbPage(pDbPage, pgno, pBt); + rc = btreeInitPage(*ppPage); + if( rc!=SQLITE_OK ){ + goto getAndInitPage_error2; + } + } + assert( (*ppPage)->pgno==pgno ); + assert( (*ppPage)->aData==sqlite3PagerGetData(pDbPage) ); + + /* If obtaining a child page for a cursor, we must verify that the page is + ** compatible with the root page. */ + if( pCur && ((*ppPage)->nCell<1 || (*ppPage)->intKey!=pCur->curIntKey) ){ + rc = SQLITE_CORRUPT_PGNO(pgno); + goto getAndInitPage_error2; + } + return SQLITE_OK; + +getAndInitPage_error2: + releasePage(*ppPage); +getAndInitPage_error1: + if( pCur ){ + pCur->iPage--; + pCur->pPage = pCur->apPage[pCur->iPage]; + } + testcase( pgno==0 ); + assert( pgno!=0 || rc==SQLITE_CORRUPT ); + return rc; +} + +/* +** Release a MemPage. This should be called once for each prior +** call to btreeGetPage. +** +** Page1 is a special case and must be released using releasePageOne(). +*/ +static void releasePageNotNull(MemPage *pPage){ + assert( pPage->aData ); + assert( pPage->pBt ); + assert( pPage->pDbPage!=0 ); + assert( sqlite3PagerGetExtra(pPage->pDbPage) == (void*)pPage ); + assert( sqlite3PagerGetData(pPage->pDbPage)==pPage->aData ); + assert( sqlite3_mutex_held(pPage->pBt->mutex) ); + sqlite3PagerUnrefNotNull(pPage->pDbPage); +} +static void releasePage(MemPage *pPage){ + if( pPage ) releasePageNotNull(pPage); +} +static void releasePageOne(MemPage *pPage){ + assert( pPage!=0 ); + assert( pPage->aData ); + assert( pPage->pBt ); + assert( pPage->pDbPage!=0 ); + assert( sqlite3PagerGetExtra(pPage->pDbPage) == (void*)pPage ); + assert( sqlite3PagerGetData(pPage->pDbPage)==pPage->aData ); + assert( sqlite3_mutex_held(pPage->pBt->mutex) ); + sqlite3PagerUnrefPageOne(pPage->pDbPage); +} + +/* +** Get an unused page. +** +** This works just like btreeGetPage() with the addition: +** +** * If the page is already in use for some other purpose, immediately +** release it and return an SQLITE_CURRUPT error. +** * Make sure the isInit flag is clear +*/ +static int btreeGetUnusedPage( + BtShared *pBt, /* The btree */ + Pgno pgno, /* Number of the page to fetch */ + MemPage **ppPage, /* Return the page in this parameter */ + int flags /* PAGER_GET_NOCONTENT or PAGER_GET_READONLY */ +){ + int rc = btreeGetPage(pBt, pgno, ppPage, flags); + if( rc==SQLITE_OK ){ + if( sqlite3PagerPageRefcount((*ppPage)->pDbPage)>1 ){ + releasePage(*ppPage); + *ppPage = 0; + return SQLITE_CORRUPT_BKPT; + } + (*ppPage)->isInit = 0; + }else{ + *ppPage = 0; + } + return rc; +} + + +/* +** During a rollback, when the pager reloads information into the cache +** so that the cache is restored to its original state at the start of +** the transaction, for each page restored this routine is called. +** +** This routine needs to reset the extra data section at the end of the +** page to agree with the restored data. +*/ +static void pageReinit(DbPage *pData){ + MemPage *pPage; + pPage = (MemPage *)sqlite3PagerGetExtra(pData); + assert( sqlite3PagerPageRefcount(pData)>0 ); + if( pPage->isInit ){ + assert( sqlite3_mutex_held(pPage->pBt->mutex) ); + pPage->isInit = 0; + if( sqlite3PagerPageRefcount(pData)>1 ){ + /* pPage might not be a btree page; it might be an overflow page + ** or ptrmap page or a free page. In those cases, the following + ** call to btreeInitPage() will likely return SQLITE_CORRUPT. + ** But no harm is done by this. And it is very important that + ** btreeInitPage() be called on every btree page so we make + ** the call for every page that comes in for re-initing. */ + btreeInitPage(pPage); + } + } +} + +/* +** Invoke the busy handler for a btree. +*/ +static int btreeInvokeBusyHandler(void *pArg){ + BtShared *pBt = (BtShared*)pArg; + assert( pBt->db ); + assert( sqlite3_mutex_held(pBt->db->mutex) ); + return sqlite3InvokeBusyHandler(&pBt->db->busyHandler, + sqlite3PagerFile(pBt->pPager)); +} + +/* +** Open a database file. +** +** zFilename is the name of the database file. If zFilename is NULL +** then an ephemeral database is created. The ephemeral database might +** be exclusively in memory, or it might use a disk-based memory cache. +** Either way, the ephemeral database will be automatically deleted +** when sqlite3BtreeClose() is called. +** +** If zFilename is ":memory:" then an in-memory database is created +** that is automatically destroyed when it is closed. +** +** The "flags" parameter is a bitmask that might contain bits like +** BTREE_OMIT_JOURNAL and/or BTREE_MEMORY. +** +** If the database is already opened in the same database connection +** and we are in shared cache mode, then the open will fail with an +** SQLITE_CONSTRAINT error. We cannot allow two or more BtShared +** objects in the same database connection since doing so will lead +** to problems with locking. +*/ +SQLITE_PRIVATE int sqlite3BtreeOpen( + sqlite3_vfs *pVfs, /* VFS to use for this b-tree */ + const char *zFilename, /* Name of the file containing the BTree database */ + sqlite3 *db, /* Associated database handle */ + Btree **ppBtree, /* Pointer to new Btree object written here */ + int flags, /* Options */ + int vfsFlags /* Flags passed through to sqlite3_vfs.xOpen() */ +){ + BtShared *pBt = 0; /* Shared part of btree structure */ + Btree *p; /* Handle to return */ + sqlite3_mutex *mutexOpen = 0; /* Prevents a race condition. Ticket #3537 */ + int rc = SQLITE_OK; /* Result code from this function */ + u8 nReserve; /* Byte of unused space on each page */ + unsigned char zDbHeader[100]; /* Database header content */ + + /* True if opening an ephemeral, temporary database */ + const int isTempDb = zFilename==0 || zFilename[0]==0; + + /* Set the variable isMemdb to true for an in-memory database, or + ** false for a file-based database. + */ +#ifdef SQLITE_OMIT_MEMORYDB + const int isMemdb = 0; +#else + const int isMemdb = (zFilename && strcmp(zFilename, ":memory:")==0) + || (isTempDb && sqlite3TempInMemory(db)) + || (vfsFlags & SQLITE_OPEN_MEMORY)!=0; +#endif + + assert( db!=0 ); + assert( pVfs!=0 ); + assert( sqlite3_mutex_held(db->mutex) ); + assert( (flags&0xff)==flags ); /* flags fit in 8 bits */ + + /* Only a BTREE_SINGLE database can be BTREE_UNORDERED */ + assert( (flags & BTREE_UNORDERED)==0 || (flags & BTREE_SINGLE)!=0 ); + + /* A BTREE_SINGLE database is always a temporary and/or ephemeral */ + assert( (flags & BTREE_SINGLE)==0 || isTempDb ); + + if( isMemdb ){ + flags |= BTREE_MEMORY; + } + if( (vfsFlags & SQLITE_OPEN_MAIN_DB)!=0 && (isMemdb || isTempDb) ){ + vfsFlags = (vfsFlags & ~SQLITE_OPEN_MAIN_DB) | SQLITE_OPEN_TEMP_DB; + } + p = sqlite3MallocZero(sizeof(Btree)); + if( !p ){ + return SQLITE_NOMEM_BKPT; + } + p->inTrans = TRANS_NONE; + p->db = db; +#ifndef SQLITE_OMIT_SHARED_CACHE + p->lock.pBtree = p; + p->lock.iTable = 1; +#endif + +#if !defined(SQLITE_OMIT_SHARED_CACHE) && !defined(SQLITE_OMIT_DISKIO) + /* + ** If this Btree is a candidate for shared cache, try to find an + ** existing BtShared object that we can share with + */ + if( isTempDb==0 && (isMemdb==0 || (vfsFlags&SQLITE_OPEN_URI)!=0) ){ + if( vfsFlags & SQLITE_OPEN_SHAREDCACHE ){ + int nFilename = sqlite3Strlen30(zFilename)+1; + int nFullPathname = pVfs->mxPathname+1; + char *zFullPathname = sqlite3Malloc(MAX(nFullPathname,nFilename)); + MUTEX_LOGIC( sqlite3_mutex *mutexShared; ) + + p->sharable = 1; + if( !zFullPathname ){ + sqlite3_free(p); + return SQLITE_NOMEM_BKPT; + } + if( isMemdb ){ + memcpy(zFullPathname, zFilename, nFilename); + }else{ + rc = sqlite3OsFullPathname(pVfs, zFilename, + nFullPathname, zFullPathname); + if( rc ){ + sqlite3_free(zFullPathname); + sqlite3_free(p); + return rc; + } + } +#if SQLITE_THREADSAFE + mutexOpen = sqlite3MutexAlloc(SQLITE_MUTEX_STATIC_OPEN); + sqlite3_mutex_enter(mutexOpen); + mutexShared = sqlite3MutexAlloc(SQLITE_MUTEX_STATIC_MASTER); + sqlite3_mutex_enter(mutexShared); +#endif + for(pBt=GLOBAL(BtShared*,sqlite3SharedCacheList); pBt; pBt=pBt->pNext){ + assert( pBt->nRef>0 ); + if( 0==strcmp(zFullPathname, sqlite3PagerFilename(pBt->pPager, 0)) + && sqlite3PagerVfs(pBt->pPager)==pVfs ){ + int iDb; + for(iDb=db->nDb-1; iDb>=0; iDb--){ + Btree *pExisting = db->aDb[iDb].pBt; + if( pExisting && pExisting->pBt==pBt ){ + sqlite3_mutex_leave(mutexShared); + sqlite3_mutex_leave(mutexOpen); + sqlite3_free(zFullPathname); + sqlite3_free(p); + return SQLITE_CONSTRAINT; + } + } + p->pBt = pBt; + pBt->nRef++; + break; + } + } + sqlite3_mutex_leave(mutexShared); + sqlite3_free(zFullPathname); + } +#ifdef SQLITE_DEBUG + else{ + /* In debug mode, we mark all persistent databases as sharable + ** even when they are not. This exercises the locking code and + ** gives more opportunity for asserts(sqlite3_mutex_held()) + ** statements to find locking problems. + */ + p->sharable = 1; + } +#endif + } +#endif + if( pBt==0 ){ + /* + ** The following asserts make sure that structures used by the btree are + ** the right size. This is to guard against size changes that result + ** when compiling on a different architecture. + */ + assert( sizeof(i64)==8 ); + assert( sizeof(u64)==8 ); + assert( sizeof(u32)==4 ); + assert( sizeof(u16)==2 ); + assert( sizeof(Pgno)==4 ); + + pBt = sqlite3MallocZero( sizeof(*pBt) ); + if( pBt==0 ){ + rc = SQLITE_NOMEM_BKPT; + goto btree_open_out; + } + rc = sqlite3PagerOpen(pVfs, &pBt->pPager, zFilename, + sizeof(MemPage), flags, vfsFlags, pageReinit); + if( rc==SQLITE_OK ){ + sqlite3PagerSetMmapLimit(pBt->pPager, db->szMmap); + rc = sqlite3PagerReadFileheader(pBt->pPager,sizeof(zDbHeader),zDbHeader); + } + if( rc!=SQLITE_OK ){ + goto btree_open_out; + } + pBt->openFlags = (u8)flags; + pBt->db = db; + sqlite3PagerSetBusyHandler(pBt->pPager, btreeInvokeBusyHandler, pBt); + p->pBt = pBt; + + pBt->pCursor = 0; + pBt->pPage1 = 0; + if( sqlite3PagerIsreadonly(pBt->pPager) ) pBt->btsFlags |= BTS_READ_ONLY; +#if defined(SQLITE_SECURE_DELETE) + pBt->btsFlags |= BTS_SECURE_DELETE; +#elif defined(SQLITE_FAST_SECURE_DELETE) + pBt->btsFlags |= BTS_OVERWRITE; +#endif + /* EVIDENCE-OF: R-51873-39618 The page size for a database file is + ** determined by the 2-byte integer located at an offset of 16 bytes from + ** the beginning of the database file. */ + pBt->pageSize = (zDbHeader[16]<<8) | (zDbHeader[17]<<16); + if( pBt->pageSize<512 || pBt->pageSize>SQLITE_MAX_PAGE_SIZE + || ((pBt->pageSize-1)&pBt->pageSize)!=0 ){ + pBt->pageSize = 0; +#ifndef SQLITE_OMIT_AUTOVACUUM + /* If the magic name ":memory:" will create an in-memory database, then + ** leave the autoVacuum mode at 0 (do not auto-vacuum), even if + ** SQLITE_DEFAULT_AUTOVACUUM is true. On the other hand, if + ** SQLITE_OMIT_MEMORYDB has been defined, then ":memory:" is just a + ** regular file-name. In this case the auto-vacuum applies as per normal. + */ + if( zFilename && !isMemdb ){ + pBt->autoVacuum = (SQLITE_DEFAULT_AUTOVACUUM ? 1 : 0); + pBt->incrVacuum = (SQLITE_DEFAULT_AUTOVACUUM==2 ? 1 : 0); + } +#endif + nReserve = 0; + }else{ + /* EVIDENCE-OF: R-37497-42412 The size of the reserved region is + ** determined by the one-byte unsigned integer found at an offset of 20 + ** into the database file header. */ + nReserve = zDbHeader[20]; + pBt->btsFlags |= BTS_PAGESIZE_FIXED; +#ifndef SQLITE_OMIT_AUTOVACUUM + pBt->autoVacuum = (get4byte(&zDbHeader[36 + 4*4])?1:0); + pBt->incrVacuum = (get4byte(&zDbHeader[36 + 7*4])?1:0); +#endif + } + rc = sqlite3PagerSetPagesize(pBt->pPager, &pBt->pageSize, nReserve); + if( rc ) goto btree_open_out; + pBt->usableSize = pBt->pageSize - nReserve; + assert( (pBt->pageSize & 7)==0 ); /* 8-byte alignment of pageSize */ + +#if !defined(SQLITE_OMIT_SHARED_CACHE) && !defined(SQLITE_OMIT_DISKIO) + /* Add the new BtShared object to the linked list sharable BtShareds. + */ + pBt->nRef = 1; + if( p->sharable ){ + MUTEX_LOGIC( sqlite3_mutex *mutexShared; ) + MUTEX_LOGIC( mutexShared = sqlite3MutexAlloc(SQLITE_MUTEX_STATIC_MASTER);) + if( SQLITE_THREADSAFE && sqlite3GlobalConfig.bCoreMutex ){ + pBt->mutex = sqlite3MutexAlloc(SQLITE_MUTEX_FAST); + if( pBt->mutex==0 ){ + rc = SQLITE_NOMEM_BKPT; + goto btree_open_out; + } + } + sqlite3_mutex_enter(mutexShared); + pBt->pNext = GLOBAL(BtShared*,sqlite3SharedCacheList); + GLOBAL(BtShared*,sqlite3SharedCacheList) = pBt; + sqlite3_mutex_leave(mutexShared); + } +#endif + } + +#if !defined(SQLITE_OMIT_SHARED_CACHE) && !defined(SQLITE_OMIT_DISKIO) + /* If the new Btree uses a sharable pBtShared, then link the new + ** Btree into the list of all sharable Btrees for the same connection. + ** The list is kept in ascending order by pBt address. + */ + if( p->sharable ){ + int i; + Btree *pSib; + for(i=0; inDb; i++){ + if( (pSib = db->aDb[i].pBt)!=0 && pSib->sharable ){ + while( pSib->pPrev ){ pSib = pSib->pPrev; } + if( (uptr)p->pBt<(uptr)pSib->pBt ){ + p->pNext = pSib; + p->pPrev = 0; + pSib->pPrev = p; + }else{ + while( pSib->pNext && (uptr)pSib->pNext->pBt<(uptr)p->pBt ){ + pSib = pSib->pNext; + } + p->pNext = pSib->pNext; + p->pPrev = pSib; + if( p->pNext ){ + p->pNext->pPrev = p; + } + pSib->pNext = p; + } + break; + } + } + } +#endif + *ppBtree = p; + +btree_open_out: + if( rc!=SQLITE_OK ){ + if( pBt && pBt->pPager ){ + sqlite3PagerClose(pBt->pPager, 0); + } + sqlite3_free(pBt); + sqlite3_free(p); + *ppBtree = 0; + }else{ + sqlite3_file *pFile; + + /* If the B-Tree was successfully opened, set the pager-cache size to the + ** default value. Except, when opening on an existing shared pager-cache, + ** do not change the pager-cache size. + */ + if( sqlite3BtreeSchema(p, 0, 0)==0 ){ + sqlite3PagerSetCachesize(p->pBt->pPager, SQLITE_DEFAULT_CACHE_SIZE); + } + + pFile = sqlite3PagerFile(pBt->pPager); + if( pFile->pMethods ){ + sqlite3OsFileControlHint(pFile, SQLITE_FCNTL_PDB, (void*)&pBt->db); + } + } + if( mutexOpen ){ + assert( sqlite3_mutex_held(mutexOpen) ); + sqlite3_mutex_leave(mutexOpen); + } + assert( rc!=SQLITE_OK || sqlite3BtreeConnectionCount(*ppBtree)>0 ); + return rc; +} + +/* +** Decrement the BtShared.nRef counter. When it reaches zero, +** remove the BtShared structure from the sharing list. Return +** true if the BtShared.nRef counter reaches zero and return +** false if it is still positive. +*/ +static int removeFromSharingList(BtShared *pBt){ +#ifndef SQLITE_OMIT_SHARED_CACHE + MUTEX_LOGIC( sqlite3_mutex *pMaster; ) + BtShared *pList; + int removed = 0; + + assert( sqlite3_mutex_notheld(pBt->mutex) ); + MUTEX_LOGIC( pMaster = sqlite3MutexAlloc(SQLITE_MUTEX_STATIC_MASTER); ) + sqlite3_mutex_enter(pMaster); + pBt->nRef--; + if( pBt->nRef<=0 ){ + if( GLOBAL(BtShared*,sqlite3SharedCacheList)==pBt ){ + GLOBAL(BtShared*,sqlite3SharedCacheList) = pBt->pNext; + }else{ + pList = GLOBAL(BtShared*,sqlite3SharedCacheList); + while( ALWAYS(pList) && pList->pNext!=pBt ){ + pList=pList->pNext; + } + if( ALWAYS(pList) ){ + pList->pNext = pBt->pNext; + } + } + if( SQLITE_THREADSAFE ){ + sqlite3_mutex_free(pBt->mutex); + } + removed = 1; + } + sqlite3_mutex_leave(pMaster); + return removed; +#else + return 1; +#endif +} + +/* +** Make sure pBt->pTmpSpace points to an allocation of +** MX_CELL_SIZE(pBt) bytes with a 4-byte prefix for a left-child +** pointer. +*/ +static void allocateTempSpace(BtShared *pBt){ + if( !pBt->pTmpSpace ){ + pBt->pTmpSpace = sqlite3PageMalloc( pBt->pageSize ); + + /* One of the uses of pBt->pTmpSpace is to format cells before + ** inserting them into a leaf page (function fillInCell()). If + ** a cell is less than 4 bytes in size, it is rounded up to 4 bytes + ** by the various routines that manipulate binary cells. Which + ** can mean that fillInCell() only initializes the first 2 or 3 + ** bytes of pTmpSpace, but that the first 4 bytes are copied from + ** it into a database page. This is not actually a problem, but it + ** does cause a valgrind error when the 1 or 2 bytes of unitialized + ** data is passed to system call write(). So to avoid this error, + ** zero the first 4 bytes of temp space here. + ** + ** Also: Provide four bytes of initialized space before the + ** beginning of pTmpSpace as an area available to prepend the + ** left-child pointer to the beginning of a cell. + */ + if( pBt->pTmpSpace ){ + memset(pBt->pTmpSpace, 0, 8); + pBt->pTmpSpace += 4; + } + } +} + +/* +** Free the pBt->pTmpSpace allocation +*/ +static void freeTempSpace(BtShared *pBt){ + if( pBt->pTmpSpace ){ + pBt->pTmpSpace -= 4; + sqlite3PageFree(pBt->pTmpSpace); + pBt->pTmpSpace = 0; + } +} + +/* +** Close an open database and invalidate all cursors. +*/ +SQLITE_PRIVATE int sqlite3BtreeClose(Btree *p){ + BtShared *pBt = p->pBt; + BtCursor *pCur; + + /* Close all cursors opened via this handle. */ + assert( sqlite3_mutex_held(p->db->mutex) ); + sqlite3BtreeEnter(p); + pCur = pBt->pCursor; + while( pCur ){ + BtCursor *pTmp = pCur; + pCur = pCur->pNext; + if( pTmp->pBtree==p ){ + sqlite3BtreeCloseCursor(pTmp); + } + } + + /* Rollback any active transaction and free the handle structure. + ** The call to sqlite3BtreeRollback() drops any table-locks held by + ** this handle. + */ + sqlite3BtreeRollback(p, SQLITE_OK, 0); + sqlite3BtreeLeave(p); + + /* If there are still other outstanding references to the shared-btree + ** structure, return now. The remainder of this procedure cleans + ** up the shared-btree. + */ + assert( p->wantToLock==0 && p->locked==0 ); + if( !p->sharable || removeFromSharingList(pBt) ){ + /* The pBt is no longer on the sharing list, so we can access + ** it without having to hold the mutex. + ** + ** Clean out and delete the BtShared object. + */ + assert( !pBt->pCursor ); + sqlite3PagerClose(pBt->pPager, p->db); + if( pBt->xFreeSchema && pBt->pSchema ){ + pBt->xFreeSchema(pBt->pSchema); + } + sqlite3DbFree(0, pBt->pSchema); + freeTempSpace(pBt); + sqlite3_free(pBt); + } + +#ifndef SQLITE_OMIT_SHARED_CACHE + assert( p->wantToLock==0 ); + assert( p->locked==0 ); + if( p->pPrev ) p->pPrev->pNext = p->pNext; + if( p->pNext ) p->pNext->pPrev = p->pPrev; +#endif + + sqlite3_free(p); + return SQLITE_OK; +} + +/* +** Change the "soft" limit on the number of pages in the cache. +** Unused and unmodified pages will be recycled when the number of +** pages in the cache exceeds this soft limit. But the size of the +** cache is allowed to grow larger than this limit if it contains +** dirty pages or pages still in active use. +*/ +SQLITE_PRIVATE int sqlite3BtreeSetCacheSize(Btree *p, int mxPage){ + BtShared *pBt = p->pBt; + assert( sqlite3_mutex_held(p->db->mutex) ); + sqlite3BtreeEnter(p); + sqlite3PagerSetCachesize(pBt->pPager, mxPage); + sqlite3BtreeLeave(p); + return SQLITE_OK; +} + +/* +** Change the "spill" limit on the number of pages in the cache. +** If the number of pages exceeds this limit during a write transaction, +** the pager might attempt to "spill" pages to the journal early in +** order to free up memory. +** +** The value returned is the current spill size. If zero is passed +** as an argument, no changes are made to the spill size setting, so +** using mxPage of 0 is a way to query the current spill size. +*/ +SQLITE_PRIVATE int sqlite3BtreeSetSpillSize(Btree *p, int mxPage){ + BtShared *pBt = p->pBt; + int res; + assert( sqlite3_mutex_held(p->db->mutex) ); + sqlite3BtreeEnter(p); + res = sqlite3PagerSetSpillsize(pBt->pPager, mxPage); + sqlite3BtreeLeave(p); + return res; +} + +#if SQLITE_MAX_MMAP_SIZE>0 +/* +** Change the limit on the amount of the database file that may be +** memory mapped. +*/ +SQLITE_PRIVATE int sqlite3BtreeSetMmapLimit(Btree *p, sqlite3_int64 szMmap){ + BtShared *pBt = p->pBt; + assert( sqlite3_mutex_held(p->db->mutex) ); + sqlite3BtreeEnter(p); + sqlite3PagerSetMmapLimit(pBt->pPager, szMmap); + sqlite3BtreeLeave(p); + return SQLITE_OK; +} +#endif /* SQLITE_MAX_MMAP_SIZE>0 */ + +/* +** Change the way data is synced to disk in order to increase or decrease +** how well the database resists damage due to OS crashes and power +** failures. Level 1 is the same as asynchronous (no syncs() occur and +** there is a high probability of damage) Level 2 is the default. There +** is a very low but non-zero probability of damage. Level 3 reduces the +** probability of damage to near zero but with a write performance reduction. +*/ +#ifndef SQLITE_OMIT_PAGER_PRAGMAS +SQLITE_PRIVATE int sqlite3BtreeSetPagerFlags( + Btree *p, /* The btree to set the safety level on */ + unsigned pgFlags /* Various PAGER_* flags */ +){ + BtShared *pBt = p->pBt; + assert( sqlite3_mutex_held(p->db->mutex) ); + sqlite3BtreeEnter(p); + sqlite3PagerSetFlags(pBt->pPager, pgFlags); + sqlite3BtreeLeave(p); + return SQLITE_OK; +} +#endif + +/* +** Change the default pages size and the number of reserved bytes per page. +** Or, if the page size has already been fixed, return SQLITE_READONLY +** without changing anything. +** +** The page size must be a power of 2 between 512 and 65536. If the page +** size supplied does not meet this constraint then the page size is not +** changed. +** +** Page sizes are constrained to be a power of two so that the region +** of the database file used for locking (beginning at PENDING_BYTE, +** the first byte past the 1GB boundary, 0x40000000) needs to occur +** at the beginning of a page. +** +** If parameter nReserve is less than zero, then the number of reserved +** bytes per page is left unchanged. +** +** If the iFix!=0 then the BTS_PAGESIZE_FIXED flag is set so that the page size +** and autovacuum mode can no longer be changed. +*/ +SQLITE_PRIVATE int sqlite3BtreeSetPageSize(Btree *p, int pageSize, int nReserve, int iFix){ + int rc = SQLITE_OK; + BtShared *pBt = p->pBt; + assert( nReserve>=-1 && nReserve<=255 ); + sqlite3BtreeEnter(p); +#if SQLITE_HAS_CODEC + if( nReserve>pBt->optimalReserve ) pBt->optimalReserve = (u8)nReserve; +#endif + if( pBt->btsFlags & BTS_PAGESIZE_FIXED ){ + sqlite3BtreeLeave(p); + return SQLITE_READONLY; + } + if( nReserve<0 ){ + nReserve = pBt->pageSize - pBt->usableSize; + } + assert( nReserve>=0 && nReserve<=255 ); + if( pageSize>=512 && pageSize<=SQLITE_MAX_PAGE_SIZE && + ((pageSize-1)&pageSize)==0 ){ + assert( (pageSize & 7)==0 ); + assert( !pBt->pCursor ); + pBt->pageSize = (u32)pageSize; + freeTempSpace(pBt); + } + rc = sqlite3PagerSetPagesize(pBt->pPager, &pBt->pageSize, nReserve); + pBt->usableSize = pBt->pageSize - (u16)nReserve; + if( iFix ) pBt->btsFlags |= BTS_PAGESIZE_FIXED; + sqlite3BtreeLeave(p); + return rc; +} + +/* +** Return the currently defined page size +*/ +SQLITE_PRIVATE int sqlite3BtreeGetPageSize(Btree *p){ + return p->pBt->pageSize; +} + +/* +** This function is similar to sqlite3BtreeGetReserve(), except that it +** may only be called if it is guaranteed that the b-tree mutex is already +** held. +** +** This is useful in one special case in the backup API code where it is +** known that the shared b-tree mutex is held, but the mutex on the +** database handle that owns *p is not. In this case if sqlite3BtreeEnter() +** were to be called, it might collide with some other operation on the +** database handle that owns *p, causing undefined behavior. +*/ +SQLITE_PRIVATE int sqlite3BtreeGetReserveNoMutex(Btree *p){ + int n; + assert( sqlite3_mutex_held(p->pBt->mutex) ); + n = p->pBt->pageSize - p->pBt->usableSize; + return n; +} + +/* +** Return the number of bytes of space at the end of every page that +** are intentually left unused. This is the "reserved" space that is +** sometimes used by extensions. +** +** If SQLITE_HAS_MUTEX is defined then the number returned is the +** greater of the current reserved space and the maximum requested +** reserve space. +*/ +SQLITE_PRIVATE int sqlite3BtreeGetOptimalReserve(Btree *p){ + int n; + sqlite3BtreeEnter(p); + n = sqlite3BtreeGetReserveNoMutex(p); +#ifdef SQLITE_HAS_CODEC + if( npBt->optimalReserve ) n = p->pBt->optimalReserve; +#endif + sqlite3BtreeLeave(p); + return n; +} + + +/* +** Set the maximum page count for a database if mxPage is positive. +** No changes are made if mxPage is 0 or negative. +** Regardless of the value of mxPage, return the maximum page count. +*/ +SQLITE_PRIVATE int sqlite3BtreeMaxPageCount(Btree *p, int mxPage){ + int n; + sqlite3BtreeEnter(p); + n = sqlite3PagerMaxPageCount(p->pBt->pPager, mxPage); + sqlite3BtreeLeave(p); + return n; +} + +/* +** Change the values for the BTS_SECURE_DELETE and BTS_OVERWRITE flags: +** +** newFlag==0 Both BTS_SECURE_DELETE and BTS_OVERWRITE are cleared +** newFlag==1 BTS_SECURE_DELETE set and BTS_OVERWRITE is cleared +** newFlag==2 BTS_SECURE_DELETE cleared and BTS_OVERWRITE is set +** newFlag==(-1) No changes +** +** This routine acts as a query if newFlag is less than zero +** +** With BTS_OVERWRITE set, deleted content is overwritten by zeros, but +** freelist leaf pages are not written back to the database. Thus in-page +** deleted content is cleared, but freelist deleted content is not. +** +** With BTS_SECURE_DELETE, operation is like BTS_OVERWRITE with the addition +** that freelist leaf pages are written back into the database, increasing +** the amount of disk I/O. +*/ +SQLITE_PRIVATE int sqlite3BtreeSecureDelete(Btree *p, int newFlag){ + int b; + if( p==0 ) return 0; + sqlite3BtreeEnter(p); + assert( BTS_OVERWRITE==BTS_SECURE_DELETE*2 ); + assert( BTS_FAST_SECURE==(BTS_OVERWRITE|BTS_SECURE_DELETE) ); + if( newFlag>=0 ){ + p->pBt->btsFlags &= ~BTS_FAST_SECURE; + p->pBt->btsFlags |= BTS_SECURE_DELETE*newFlag; + } + b = (p->pBt->btsFlags & BTS_FAST_SECURE)/BTS_SECURE_DELETE; + sqlite3BtreeLeave(p); + return b; +} + +/* +** Change the 'auto-vacuum' property of the database. If the 'autoVacuum' +** parameter is non-zero, then auto-vacuum mode is enabled. If zero, it +** is disabled. The default value for the auto-vacuum property is +** determined by the SQLITE_DEFAULT_AUTOVACUUM macro. +*/ +SQLITE_PRIVATE int sqlite3BtreeSetAutoVacuum(Btree *p, int autoVacuum){ +#ifdef SQLITE_OMIT_AUTOVACUUM + return SQLITE_READONLY; +#else + BtShared *pBt = p->pBt; + int rc = SQLITE_OK; + u8 av = (u8)autoVacuum; + + sqlite3BtreeEnter(p); + if( (pBt->btsFlags & BTS_PAGESIZE_FIXED)!=0 && (av ?1:0)!=pBt->autoVacuum ){ + rc = SQLITE_READONLY; + }else{ + pBt->autoVacuum = av ?1:0; + pBt->incrVacuum = av==2 ?1:0; + } + sqlite3BtreeLeave(p); + return rc; +#endif +} + +/* +** Return the value of the 'auto-vacuum' property. If auto-vacuum is +** enabled 1 is returned. Otherwise 0. +*/ +SQLITE_PRIVATE int sqlite3BtreeGetAutoVacuum(Btree *p){ +#ifdef SQLITE_OMIT_AUTOVACUUM + return BTREE_AUTOVACUUM_NONE; +#else + int rc; + sqlite3BtreeEnter(p); + rc = ( + (!p->pBt->autoVacuum)?BTREE_AUTOVACUUM_NONE: + (!p->pBt->incrVacuum)?BTREE_AUTOVACUUM_FULL: + BTREE_AUTOVACUUM_INCR + ); + sqlite3BtreeLeave(p); + return rc; +#endif +} + +/* +** If the user has not set the safety-level for this database connection +** using "PRAGMA synchronous", and if the safety-level is not already +** set to the value passed to this function as the second parameter, +** set it so. +*/ +#if SQLITE_DEFAULT_SYNCHRONOUS!=SQLITE_DEFAULT_WAL_SYNCHRONOUS \ + && !defined(SQLITE_OMIT_WAL) +static void setDefaultSyncFlag(BtShared *pBt, u8 safety_level){ + sqlite3 *db; + Db *pDb; + if( (db=pBt->db)!=0 && (pDb=db->aDb)!=0 ){ + while( pDb->pBt==0 || pDb->pBt->pBt!=pBt ){ pDb++; } + if( pDb->bSyncSet==0 + && pDb->safety_level!=safety_level + && pDb!=&db->aDb[1] + ){ + pDb->safety_level = safety_level; + sqlite3PagerSetFlags(pBt->pPager, + pDb->safety_level | (db->flags & PAGER_FLAGS_MASK)); + } + } +} +#else +# define setDefaultSyncFlag(pBt,safety_level) +#endif + +/* Forward declaration */ +static int newDatabase(BtShared*); + + +/* +** Get a reference to pPage1 of the database file. This will +** also acquire a readlock on that file. +** +** SQLITE_OK is returned on success. If the file is not a +** well-formed database file, then SQLITE_CORRUPT is returned. +** SQLITE_BUSY is returned if the database is locked. SQLITE_NOMEM +** is returned if we run out of memory. +*/ +static int lockBtree(BtShared *pBt){ + int rc; /* Result code from subfunctions */ + MemPage *pPage1; /* Page 1 of the database file */ + u32 nPage; /* Number of pages in the database */ + u32 nPageFile = 0; /* Number of pages in the database file */ + u32 nPageHeader; /* Number of pages in the database according to hdr */ + + assert( sqlite3_mutex_held(pBt->mutex) ); + assert( pBt->pPage1==0 ); + rc = sqlite3PagerSharedLock(pBt->pPager); + if( rc!=SQLITE_OK ) return rc; + rc = btreeGetPage(pBt, 1, &pPage1, 0); + if( rc!=SQLITE_OK ) return rc; + + /* Do some checking to help insure the file we opened really is + ** a valid database file. + */ + nPage = nPageHeader = get4byte(28+(u8*)pPage1->aData); + sqlite3PagerPagecount(pBt->pPager, (int*)&nPageFile); + if( nPage==0 || memcmp(24+(u8*)pPage1->aData, 92+(u8*)pPage1->aData,4)!=0 ){ + nPage = nPageFile; + } + if( (pBt->db->flags & SQLITE_ResetDatabase)!=0 ){ + nPage = 0; + } + if( nPage>0 ){ + u32 pageSize; + u32 usableSize; + u8 *page1 = pPage1->aData; + rc = SQLITE_NOTADB; + /* EVIDENCE-OF: R-43737-39999 Every valid SQLite database file begins + ** with the following 16 bytes (in hex): 53 51 4c 69 74 65 20 66 6f 72 6d + ** 61 74 20 33 00. */ + if( memcmp(page1, zMagicHeader, 16)!=0 ){ + goto page1_init_failed; + } + +#ifdef SQLITE_OMIT_WAL + if( page1[18]>1 ){ + pBt->btsFlags |= BTS_READ_ONLY; + } + if( page1[19]>1 ){ + goto page1_init_failed; + } +#else + if( page1[18]>2 ){ + pBt->btsFlags |= BTS_READ_ONLY; + } + if( page1[19]>2 ){ + goto page1_init_failed; + } + + /* If the write version is set to 2, this database should be accessed + ** in WAL mode. If the log is not already open, open it now. Then + ** return SQLITE_OK and return without populating BtShared.pPage1. + ** The caller detects this and calls this function again. This is + ** required as the version of page 1 currently in the page1 buffer + ** may not be the latest version - there may be a newer one in the log + ** file. + */ + if( page1[19]==2 && (pBt->btsFlags & BTS_NO_WAL)==0 ){ + int isOpen = 0; + rc = sqlite3PagerOpenWal(pBt->pPager, &isOpen); + if( rc!=SQLITE_OK ){ + goto page1_init_failed; + }else{ + setDefaultSyncFlag(pBt, SQLITE_DEFAULT_WAL_SYNCHRONOUS+1); + if( isOpen==0 ){ + releasePageOne(pPage1); + return SQLITE_OK; + } + } + rc = SQLITE_NOTADB; + }else{ + setDefaultSyncFlag(pBt, SQLITE_DEFAULT_SYNCHRONOUS+1); + } +#endif + + /* EVIDENCE-OF: R-15465-20813 The maximum and minimum embedded payload + ** fractions and the leaf payload fraction values must be 64, 32, and 32. + ** + ** The original design allowed these amounts to vary, but as of + ** version 3.6.0, we require them to be fixed. + */ + if( memcmp(&page1[21], "\100\040\040",3)!=0 ){ + goto page1_init_failed; + } + /* EVIDENCE-OF: R-51873-39618 The page size for a database file is + ** determined by the 2-byte integer located at an offset of 16 bytes from + ** the beginning of the database file. */ + pageSize = (page1[16]<<8) | (page1[17]<<16); + /* EVIDENCE-OF: R-25008-21688 The size of a page is a power of two + ** between 512 and 65536 inclusive. */ + if( ((pageSize-1)&pageSize)!=0 + || pageSize>SQLITE_MAX_PAGE_SIZE + || pageSize<=256 + ){ + goto page1_init_failed; + } + pBt->btsFlags |= BTS_PAGESIZE_FIXED; + assert( (pageSize & 7)==0 ); + /* EVIDENCE-OF: R-59310-51205 The "reserved space" size in the 1-byte + ** integer at offset 20 is the number of bytes of space at the end of + ** each page to reserve for extensions. + ** + ** EVIDENCE-OF: R-37497-42412 The size of the reserved region is + ** determined by the one-byte unsigned integer found at an offset of 20 + ** into the database file header. */ + usableSize = pageSize - page1[20]; + if( (u32)pageSize!=pBt->pageSize ){ + /* After reading the first page of the database assuming a page size + ** of BtShared.pageSize, we have discovered that the page-size is + ** actually pageSize. Unlock the database, leave pBt->pPage1 at + ** zero and return SQLITE_OK. The caller will call this function + ** again with the correct page-size. + */ + releasePageOne(pPage1); + pBt->usableSize = usableSize; + pBt->pageSize = pageSize; + freeTempSpace(pBt); + rc = sqlite3PagerSetPagesize(pBt->pPager, &pBt->pageSize, + pageSize-usableSize); + return rc; + } + if( sqlite3WritableSchema(pBt->db)==0 && nPage>nPageFile ){ + rc = SQLITE_CORRUPT_BKPT; + goto page1_init_failed; + } + /* EVIDENCE-OF: R-28312-64704 However, the usable size is not allowed to + ** be less than 480. In other words, if the page size is 512, then the + ** reserved space size cannot exceed 32. */ + if( usableSize<480 ){ + goto page1_init_failed; + } + pBt->pageSize = pageSize; + pBt->usableSize = usableSize; +#ifndef SQLITE_OMIT_AUTOVACUUM + pBt->autoVacuum = (get4byte(&page1[36 + 4*4])?1:0); + pBt->incrVacuum = (get4byte(&page1[36 + 7*4])?1:0); +#endif + } + + /* maxLocal is the maximum amount of payload to store locally for + ** a cell. Make sure it is small enough so that at least minFanout + ** cells can will fit on one page. We assume a 10-byte page header. + ** Besides the payload, the cell must store: + ** 2-byte pointer to the cell + ** 4-byte child pointer + ** 9-byte nKey value + ** 4-byte nData value + ** 4-byte overflow page pointer + ** So a cell consists of a 2-byte pointer, a header which is as much as + ** 17 bytes long, 0 to N bytes of payload, and an optional 4 byte overflow + ** page pointer. + */ + pBt->maxLocal = (u16)((pBt->usableSize-12)*64/255 - 23); + pBt->minLocal = (u16)((pBt->usableSize-12)*32/255 - 23); + pBt->maxLeaf = (u16)(pBt->usableSize - 35); + pBt->minLeaf = (u16)((pBt->usableSize-12)*32/255 - 23); + if( pBt->maxLocal>127 ){ + pBt->max1bytePayload = 127; + }else{ + pBt->max1bytePayload = (u8)pBt->maxLocal; + } + assert( pBt->maxLeaf + 23 <= MX_CELL_SIZE(pBt) ); + pBt->pPage1 = pPage1; + pBt->nPage = nPage; + return SQLITE_OK; + +page1_init_failed: + releasePageOne(pPage1); + pBt->pPage1 = 0; + return rc; +} + +#ifndef NDEBUG +/* +** Return the number of cursors open on pBt. This is for use +** in assert() expressions, so it is only compiled if NDEBUG is not +** defined. +** +** Only write cursors are counted if wrOnly is true. If wrOnly is +** false then all cursors are counted. +** +** For the purposes of this routine, a cursor is any cursor that +** is capable of reading or writing to the database. Cursors that +** have been tripped into the CURSOR_FAULT state are not counted. +*/ +static int countValidCursors(BtShared *pBt, int wrOnly){ + BtCursor *pCur; + int r = 0; + for(pCur=pBt->pCursor; pCur; pCur=pCur->pNext){ + if( (wrOnly==0 || (pCur->curFlags & BTCF_WriteFlag)!=0) + && pCur->eState!=CURSOR_FAULT ) r++; + } + return r; +} +#endif + +/* +** If there are no outstanding cursors and we are not in the middle +** of a transaction but there is a read lock on the database, then +** this routine unrefs the first page of the database file which +** has the effect of releasing the read lock. +** +** If there is a transaction in progress, this routine is a no-op. +*/ +static void unlockBtreeIfUnused(BtShared *pBt){ + assert( sqlite3_mutex_held(pBt->mutex) ); + assert( countValidCursors(pBt,0)==0 || pBt->inTransaction>TRANS_NONE ); + if( pBt->inTransaction==TRANS_NONE && pBt->pPage1!=0 ){ + MemPage *pPage1 = pBt->pPage1; + assert( pPage1->aData ); + assert( sqlite3PagerRefcount(pBt->pPager)==1 ); + pBt->pPage1 = 0; + releasePageOne(pPage1); + } +} + +/* +** If pBt points to an empty file then convert that empty file +** into a new empty database by initializing the first page of +** the database. +*/ +static int newDatabase(BtShared *pBt){ + MemPage *pP1; + unsigned char *data; + int rc; + + assert( sqlite3_mutex_held(pBt->mutex) ); + if( pBt->nPage>0 ){ + return SQLITE_OK; + } + pP1 = pBt->pPage1; + assert( pP1!=0 ); + data = pP1->aData; + rc = sqlite3PagerWrite(pP1->pDbPage); + if( rc ) return rc; + memcpy(data, zMagicHeader, sizeof(zMagicHeader)); + assert( sizeof(zMagicHeader)==16 ); + data[16] = (u8)((pBt->pageSize>>8)&0xff); + data[17] = (u8)((pBt->pageSize>>16)&0xff); + data[18] = 1; + data[19] = 1; + assert( pBt->usableSize<=pBt->pageSize && pBt->usableSize+255>=pBt->pageSize); + data[20] = (u8)(pBt->pageSize - pBt->usableSize); + data[21] = 64; + data[22] = 32; + data[23] = 32; + memset(&data[24], 0, 100-24); + zeroPage(pP1, PTF_INTKEY|PTF_LEAF|PTF_LEAFDATA ); + pBt->btsFlags |= BTS_PAGESIZE_FIXED; +#ifndef SQLITE_OMIT_AUTOVACUUM + assert( pBt->autoVacuum==1 || pBt->autoVacuum==0 ); + assert( pBt->incrVacuum==1 || pBt->incrVacuum==0 ); + put4byte(&data[36 + 4*4], pBt->autoVacuum); + put4byte(&data[36 + 7*4], pBt->incrVacuum); +#endif + pBt->nPage = 1; + data[31] = 1; + return SQLITE_OK; +} + +/* +** Initialize the first page of the database file (creating a database +** consisting of a single page and no schema objects). Return SQLITE_OK +** if successful, or an SQLite error code otherwise. +*/ +SQLITE_PRIVATE int sqlite3BtreeNewDb(Btree *p){ + int rc; + sqlite3BtreeEnter(p); + p->pBt->nPage = 0; + rc = newDatabase(p->pBt); + sqlite3BtreeLeave(p); + return rc; +} + +/* +** Attempt to start a new transaction. A write-transaction +** is started if the second argument is nonzero, otherwise a read- +** transaction. If the second argument is 2 or more and exclusive +** transaction is started, meaning that no other process is allowed +** to access the database. A preexisting transaction may not be +** upgraded to exclusive by calling this routine a second time - the +** exclusivity flag only works for a new transaction. +** +** A write-transaction must be started before attempting any +** changes to the database. None of the following routines +** will work unless a transaction is started first: +** +** sqlite3BtreeCreateTable() +** sqlite3BtreeCreateIndex() +** sqlite3BtreeClearTable() +** sqlite3BtreeDropTable() +** sqlite3BtreeInsert() +** sqlite3BtreeDelete() +** sqlite3BtreeUpdateMeta() +** +** If an initial attempt to acquire the lock fails because of lock contention +** and the database was previously unlocked, then invoke the busy handler +** if there is one. But if there was previously a read-lock, do not +** invoke the busy handler - just return SQLITE_BUSY. SQLITE_BUSY is +** returned when there is already a read-lock in order to avoid a deadlock. +** +** Suppose there are two processes A and B. A has a read lock and B has +** a reserved lock. B tries to promote to exclusive but is blocked because +** of A's read lock. A tries to promote to reserved but is blocked by B. +** One or the other of the two processes must give way or there can be +** no progress. By returning SQLITE_BUSY and not invoking the busy callback +** when A already has a read lock, we encourage A to give up and let B +** proceed. +*/ +SQLITE_PRIVATE int sqlite3BtreeBeginTrans(Btree *p, int wrflag, int *pSchemaVersion){ + BtShared *pBt = p->pBt; + int rc = SQLITE_OK; + + sqlite3BtreeEnter(p); + btreeIntegrity(p); + + /* If the btree is already in a write-transaction, or it + ** is already in a read-transaction and a read-transaction + ** is requested, this is a no-op. + */ + if( p->inTrans==TRANS_WRITE || (p->inTrans==TRANS_READ && !wrflag) ){ + goto trans_begun; + } + assert( pBt->inTransaction==TRANS_WRITE || IfNotOmitAV(pBt->bDoTruncate)==0 ); + + if( (p->db->flags & SQLITE_ResetDatabase) + && sqlite3PagerIsreadonly(pBt->pPager)==0 + ){ + pBt->btsFlags &= ~BTS_READ_ONLY; + } + + /* Write transactions are not possible on a read-only database */ + if( (pBt->btsFlags & BTS_READ_ONLY)!=0 && wrflag ){ + rc = SQLITE_READONLY; + goto trans_begun; + } + +#ifndef SQLITE_OMIT_SHARED_CACHE + { + sqlite3 *pBlock = 0; + /* If another database handle has already opened a write transaction + ** on this shared-btree structure and a second write transaction is + ** requested, return SQLITE_LOCKED. + */ + if( (wrflag && pBt->inTransaction==TRANS_WRITE) + || (pBt->btsFlags & BTS_PENDING)!=0 + ){ + pBlock = pBt->pWriter->db; + }else if( wrflag>1 ){ + BtLock *pIter; + for(pIter=pBt->pLock; pIter; pIter=pIter->pNext){ + if( pIter->pBtree!=p ){ + pBlock = pIter->pBtree->db; + break; + } + } + } + if( pBlock ){ + sqlite3ConnectionBlocked(p->db, pBlock); + rc = SQLITE_LOCKED_SHAREDCACHE; + goto trans_begun; + } + } +#endif + + /* Any read-only or read-write transaction implies a read-lock on + ** page 1. So if some other shared-cache client already has a write-lock + ** on page 1, the transaction cannot be opened. */ + rc = querySharedCacheTableLock(p, MASTER_ROOT, READ_LOCK); + if( SQLITE_OK!=rc ) goto trans_begun; + + pBt->btsFlags &= ~BTS_INITIALLY_EMPTY; + if( pBt->nPage==0 ) pBt->btsFlags |= BTS_INITIALLY_EMPTY; + do { + /* Call lockBtree() until either pBt->pPage1 is populated or + ** lockBtree() returns something other than SQLITE_OK. lockBtree() + ** may return SQLITE_OK but leave pBt->pPage1 set to 0 if after + ** reading page 1 it discovers that the page-size of the database + ** file is not pBt->pageSize. In this case lockBtree() will update + ** pBt->pageSize to the page-size of the file on disk. + */ + while( pBt->pPage1==0 && SQLITE_OK==(rc = lockBtree(pBt)) ); + + if( rc==SQLITE_OK && wrflag ){ + if( (pBt->btsFlags & BTS_READ_ONLY)!=0 ){ + rc = SQLITE_READONLY; + }else{ + rc = sqlite3PagerBegin(pBt->pPager,wrflag>1,sqlite3TempInMemory(p->db)); + if( rc==SQLITE_OK ){ + rc = newDatabase(pBt); + }else if( rc==SQLITE_BUSY_SNAPSHOT && pBt->inTransaction==TRANS_NONE ){ + /* if there was no transaction opened when this function was + ** called and SQLITE_BUSY_SNAPSHOT is returned, change the error + ** code to SQLITE_BUSY. */ + rc = SQLITE_BUSY; + } + } + } + + if( rc!=SQLITE_OK ){ + unlockBtreeIfUnused(pBt); + } + }while( (rc&0xFF)==SQLITE_BUSY && pBt->inTransaction==TRANS_NONE && + btreeInvokeBusyHandler(pBt) ); + sqlite3PagerResetLockTimeout(pBt->pPager); + + if( rc==SQLITE_OK ){ + if( p->inTrans==TRANS_NONE ){ + pBt->nTransaction++; +#ifndef SQLITE_OMIT_SHARED_CACHE + if( p->sharable ){ + assert( p->lock.pBtree==p && p->lock.iTable==1 ); + p->lock.eLock = READ_LOCK; + p->lock.pNext = pBt->pLock; + pBt->pLock = &p->lock; + } +#endif + } + p->inTrans = (wrflag?TRANS_WRITE:TRANS_READ); + if( p->inTrans>pBt->inTransaction ){ + pBt->inTransaction = p->inTrans; + } + if( wrflag ){ + MemPage *pPage1 = pBt->pPage1; +#ifndef SQLITE_OMIT_SHARED_CACHE + assert( !pBt->pWriter ); + pBt->pWriter = p; + pBt->btsFlags &= ~BTS_EXCLUSIVE; + if( wrflag>1 ) pBt->btsFlags |= BTS_EXCLUSIVE; +#endif + + /* If the db-size header field is incorrect (as it may be if an old + ** client has been writing the database file), update it now. Doing + ** this sooner rather than later means the database size can safely + ** re-read the database size from page 1 if a savepoint or transaction + ** rollback occurs within the transaction. + */ + if( pBt->nPage!=get4byte(&pPage1->aData[28]) ){ + rc = sqlite3PagerWrite(pPage1->pDbPage); + if( rc==SQLITE_OK ){ + put4byte(&pPage1->aData[28], pBt->nPage); + } + } + } + } + +trans_begun: + if( rc==SQLITE_OK ){ + if( pSchemaVersion ){ + *pSchemaVersion = get4byte(&pBt->pPage1->aData[40]); + } + if( wrflag ){ + /* This call makes sure that the pager has the correct number of + ** open savepoints. If the second parameter is greater than 0 and + ** the sub-journal is not already open, then it will be opened here. + */ + rc = sqlite3PagerOpenSavepoint(pBt->pPager, p->db->nSavepoint); + } + } + + btreeIntegrity(p); + sqlite3BtreeLeave(p); + return rc; +} + +#ifndef SQLITE_OMIT_AUTOVACUUM + +/* +** Set the pointer-map entries for all children of page pPage. Also, if +** pPage contains cells that point to overflow pages, set the pointer +** map entries for the overflow pages as well. +*/ +static int setChildPtrmaps(MemPage *pPage){ + int i; /* Counter variable */ + int nCell; /* Number of cells in page pPage */ + int rc; /* Return code */ + BtShared *pBt = pPage->pBt; + Pgno pgno = pPage->pgno; + + assert( sqlite3_mutex_held(pPage->pBt->mutex) ); + rc = pPage->isInit ? SQLITE_OK : btreeInitPage(pPage); + if( rc!=SQLITE_OK ) return rc; + nCell = pPage->nCell; + + for(i=0; ileaf ){ + Pgno childPgno = get4byte(pCell); + ptrmapPut(pBt, childPgno, PTRMAP_BTREE, pgno, &rc); + } + } + + if( !pPage->leaf ){ + Pgno childPgno = get4byte(&pPage->aData[pPage->hdrOffset+8]); + ptrmapPut(pBt, childPgno, PTRMAP_BTREE, pgno, &rc); + } + + return rc; +} + +/* +** Somewhere on pPage is a pointer to page iFrom. Modify this pointer so +** that it points to iTo. Parameter eType describes the type of pointer to +** be modified, as follows: +** +** PTRMAP_BTREE: pPage is a btree-page. The pointer points at a child +** page of pPage. +** +** PTRMAP_OVERFLOW1: pPage is a btree-page. The pointer points at an overflow +** page pointed to by one of the cells on pPage. +** +** PTRMAP_OVERFLOW2: pPage is an overflow-page. The pointer points at the next +** overflow page in the list. +*/ +static int modifyPagePointer(MemPage *pPage, Pgno iFrom, Pgno iTo, u8 eType){ + assert( sqlite3_mutex_held(pPage->pBt->mutex) ); + assert( sqlite3PagerIswriteable(pPage->pDbPage) ); + if( eType==PTRMAP_OVERFLOW2 ){ + /* The pointer is always the first 4 bytes of the page in this case. */ + if( get4byte(pPage->aData)!=iFrom ){ + return SQLITE_CORRUPT_PAGE(pPage); + } + put4byte(pPage->aData, iTo); + }else{ + int i; + int nCell; + int rc; + + rc = pPage->isInit ? SQLITE_OK : btreeInitPage(pPage); + if( rc ) return rc; + nCell = pPage->nCell; + + for(i=0; ixParseCell(pPage, pCell, &info); + if( info.nLocal pPage->aData+pPage->pBt->usableSize ){ + return SQLITE_CORRUPT_PAGE(pPage); + } + if( iFrom==get4byte(pCell+info.nSize-4) ){ + put4byte(pCell+info.nSize-4, iTo); + break; + } + } + }else{ + if( get4byte(pCell)==iFrom ){ + put4byte(pCell, iTo); + break; + } + } + } + + if( i==nCell ){ + if( eType!=PTRMAP_BTREE || + get4byte(&pPage->aData[pPage->hdrOffset+8])!=iFrom ){ + return SQLITE_CORRUPT_PAGE(pPage); + } + put4byte(&pPage->aData[pPage->hdrOffset+8], iTo); + } + } + return SQLITE_OK; +} + + +/* +** Move the open database page pDbPage to location iFreePage in the +** database. The pDbPage reference remains valid. +** +** The isCommit flag indicates that there is no need to remember that +** the journal needs to be sync()ed before database page pDbPage->pgno +** can be written to. The caller has already promised not to write to that +** page. +*/ +static int relocatePage( + BtShared *pBt, /* Btree */ + MemPage *pDbPage, /* Open page to move */ + u8 eType, /* Pointer map 'type' entry for pDbPage */ + Pgno iPtrPage, /* Pointer map 'page-no' entry for pDbPage */ + Pgno iFreePage, /* The location to move pDbPage to */ + int isCommit /* isCommit flag passed to sqlite3PagerMovepage */ +){ + MemPage *pPtrPage; /* The page that contains a pointer to pDbPage */ + Pgno iDbPage = pDbPage->pgno; + Pager *pPager = pBt->pPager; + int rc; + + assert( eType==PTRMAP_OVERFLOW2 || eType==PTRMAP_OVERFLOW1 || + eType==PTRMAP_BTREE || eType==PTRMAP_ROOTPAGE ); + assert( sqlite3_mutex_held(pBt->mutex) ); + assert( pDbPage->pBt==pBt ); + if( iDbPage<3 ) return SQLITE_CORRUPT_BKPT; + + /* Move page iDbPage from its current location to page number iFreePage */ + TRACE(("AUTOVACUUM: Moving %d to free page %d (ptr page %d type %d)\n", + iDbPage, iFreePage, iPtrPage, eType)); + rc = sqlite3PagerMovepage(pPager, pDbPage->pDbPage, iFreePage, isCommit); + if( rc!=SQLITE_OK ){ + return rc; + } + pDbPage->pgno = iFreePage; + + /* If pDbPage was a btree-page, then it may have child pages and/or cells + ** that point to overflow pages. The pointer map entries for all these + ** pages need to be changed. + ** + ** If pDbPage is an overflow page, then the first 4 bytes may store a + ** pointer to a subsequent overflow page. If this is the case, then + ** the pointer map needs to be updated for the subsequent overflow page. + */ + if( eType==PTRMAP_BTREE || eType==PTRMAP_ROOTPAGE ){ + rc = setChildPtrmaps(pDbPage); + if( rc!=SQLITE_OK ){ + return rc; + } + }else{ + Pgno nextOvfl = get4byte(pDbPage->aData); + if( nextOvfl!=0 ){ + ptrmapPut(pBt, nextOvfl, PTRMAP_OVERFLOW2, iFreePage, &rc); + if( rc!=SQLITE_OK ){ + return rc; + } + } + } + + /* Fix the database pointer on page iPtrPage that pointed at iDbPage so + ** that it points at iFreePage. Also fix the pointer map entry for + ** iPtrPage. + */ + if( eType!=PTRMAP_ROOTPAGE ){ + rc = btreeGetPage(pBt, iPtrPage, &pPtrPage, 0); + if( rc!=SQLITE_OK ){ + return rc; + } + rc = sqlite3PagerWrite(pPtrPage->pDbPage); + if( rc!=SQLITE_OK ){ + releasePage(pPtrPage); + return rc; + } + rc = modifyPagePointer(pPtrPage, iDbPage, iFreePage, eType); + releasePage(pPtrPage); + if( rc==SQLITE_OK ){ + ptrmapPut(pBt, iFreePage, eType, iPtrPage, &rc); + } + } + return rc; +} + +/* Forward declaration required by incrVacuumStep(). */ +static int allocateBtreePage(BtShared *, MemPage **, Pgno *, Pgno, u8); + +/* +** Perform a single step of an incremental-vacuum. If successful, return +** SQLITE_OK. If there is no work to do (and therefore no point in +** calling this function again), return SQLITE_DONE. Or, if an error +** occurs, return some other error code. +** +** More specifically, this function attempts to re-organize the database so +** that the last page of the file currently in use is no longer in use. +** +** Parameter nFin is the number of pages that this database would contain +** were this function called until it returns SQLITE_DONE. +** +** If the bCommit parameter is non-zero, this function assumes that the +** caller will keep calling incrVacuumStep() until it returns SQLITE_DONE +** or an error. bCommit is passed true for an auto-vacuum-on-commit +** operation, or false for an incremental vacuum. +*/ +static int incrVacuumStep(BtShared *pBt, Pgno nFin, Pgno iLastPg, int bCommit){ + Pgno nFreeList; /* Number of pages still on the free-list */ + int rc; + + assert( sqlite3_mutex_held(pBt->mutex) ); + assert( iLastPg>nFin ); + + if( !PTRMAP_ISPAGE(pBt, iLastPg) && iLastPg!=PENDING_BYTE_PAGE(pBt) ){ + u8 eType; + Pgno iPtrPage; + + nFreeList = get4byte(&pBt->pPage1->aData[36]); + if( nFreeList==0 ){ + return SQLITE_DONE; + } + + rc = ptrmapGet(pBt, iLastPg, &eType, &iPtrPage); + if( rc!=SQLITE_OK ){ + return rc; + } + if( eType==PTRMAP_ROOTPAGE ){ + return SQLITE_CORRUPT_BKPT; + } + + if( eType==PTRMAP_FREEPAGE ){ + if( bCommit==0 ){ + /* Remove the page from the files free-list. This is not required + ** if bCommit is non-zero. In that case, the free-list will be + ** truncated to zero after this function returns, so it doesn't + ** matter if it still contains some garbage entries. + */ + Pgno iFreePg; + MemPage *pFreePg; + rc = allocateBtreePage(pBt, &pFreePg, &iFreePg, iLastPg, BTALLOC_EXACT); + if( rc!=SQLITE_OK ){ + return rc; + } + assert( iFreePg==iLastPg ); + releasePage(pFreePg); + } + } else { + Pgno iFreePg; /* Index of free page to move pLastPg to */ + MemPage *pLastPg; + u8 eMode = BTALLOC_ANY; /* Mode parameter for allocateBtreePage() */ + Pgno iNear = 0; /* nearby parameter for allocateBtreePage() */ + + rc = btreeGetPage(pBt, iLastPg, &pLastPg, 0); + if( rc!=SQLITE_OK ){ + return rc; + } + + /* If bCommit is zero, this loop runs exactly once and page pLastPg + ** is swapped with the first free page pulled off the free list. + ** + ** On the other hand, if bCommit is greater than zero, then keep + ** looping until a free-page located within the first nFin pages + ** of the file is found. + */ + if( bCommit==0 ){ + eMode = BTALLOC_LE; + iNear = nFin; + } + do { + MemPage *pFreePg; + rc = allocateBtreePage(pBt, &pFreePg, &iFreePg, iNear, eMode); + if( rc!=SQLITE_OK ){ + releasePage(pLastPg); + return rc; + } + releasePage(pFreePg); + }while( bCommit && iFreePg>nFin ); + assert( iFreePgbDoTruncate = 1; + pBt->nPage = iLastPg; + } + return SQLITE_OK; +} + +/* +** The database opened by the first argument is an auto-vacuum database +** nOrig pages in size containing nFree free pages. Return the expected +** size of the database in pages following an auto-vacuum operation. +*/ +static Pgno finalDbSize(BtShared *pBt, Pgno nOrig, Pgno nFree){ + int nEntry; /* Number of entries on one ptrmap page */ + Pgno nPtrmap; /* Number of PtrMap pages to be freed */ + Pgno nFin; /* Return value */ + + nEntry = pBt->usableSize/5; + nPtrmap = (nFree-nOrig+PTRMAP_PAGENO(pBt, nOrig)+nEntry)/nEntry; + nFin = nOrig - nFree - nPtrmap; + if( nOrig>PENDING_BYTE_PAGE(pBt) && nFinpBt; + + sqlite3BtreeEnter(p); + assert( pBt->inTransaction==TRANS_WRITE && p->inTrans==TRANS_WRITE ); + if( !pBt->autoVacuum ){ + rc = SQLITE_DONE; + }else{ + Pgno nOrig = btreePagecount(pBt); + Pgno nFree = get4byte(&pBt->pPage1->aData[36]); + Pgno nFin = finalDbSize(pBt, nOrig, nFree); + + if( nOrig0 ){ + rc = saveAllCursors(pBt, 0, 0); + if( rc==SQLITE_OK ){ + invalidateAllOverflowCache(pBt); + rc = incrVacuumStep(pBt, nFin, nOrig, 0); + } + if( rc==SQLITE_OK ){ + rc = sqlite3PagerWrite(pBt->pPage1->pDbPage); + put4byte(&pBt->pPage1->aData[28], pBt->nPage); + } + }else{ + rc = SQLITE_DONE; + } + } + sqlite3BtreeLeave(p); + return rc; +} + +/* +** This routine is called prior to sqlite3PagerCommit when a transaction +** is committed for an auto-vacuum database. +** +** If SQLITE_OK is returned, then *pnTrunc is set to the number of pages +** the database file should be truncated to during the commit process. +** i.e. the database has been reorganized so that only the first *pnTrunc +** pages are in use. +*/ +static int autoVacuumCommit(BtShared *pBt){ + int rc = SQLITE_OK; + Pager *pPager = pBt->pPager; + VVA_ONLY( int nRef = sqlite3PagerRefcount(pPager); ) + + assert( sqlite3_mutex_held(pBt->mutex) ); + invalidateAllOverflowCache(pBt); + assert(pBt->autoVacuum); + if( !pBt->incrVacuum ){ + Pgno nFin; /* Number of pages in database after autovacuuming */ + Pgno nFree; /* Number of pages on the freelist initially */ + Pgno iFree; /* The next page to be freed */ + Pgno nOrig; /* Database size before freeing */ + + nOrig = btreePagecount(pBt); + if( PTRMAP_ISPAGE(pBt, nOrig) || nOrig==PENDING_BYTE_PAGE(pBt) ){ + /* It is not possible to create a database for which the final page + ** is either a pointer-map page or the pending-byte page. If one + ** is encountered, this indicates corruption. + */ + return SQLITE_CORRUPT_BKPT; + } + + nFree = get4byte(&pBt->pPage1->aData[36]); + nFin = finalDbSize(pBt, nOrig, nFree); + if( nFin>nOrig ) return SQLITE_CORRUPT_BKPT; + if( nFinnFin && rc==SQLITE_OK; iFree--){ + rc = incrVacuumStep(pBt, nFin, iFree, 1); + } + if( (rc==SQLITE_DONE || rc==SQLITE_OK) && nFree>0 ){ + rc = sqlite3PagerWrite(pBt->pPage1->pDbPage); + put4byte(&pBt->pPage1->aData[32], 0); + put4byte(&pBt->pPage1->aData[36], 0); + put4byte(&pBt->pPage1->aData[28], nFin); + pBt->bDoTruncate = 1; + pBt->nPage = nFin; + } + if( rc!=SQLITE_OK ){ + sqlite3PagerRollback(pPager); + } + } + + assert( nRef>=sqlite3PagerRefcount(pPager) ); + return rc; +} + +#else /* ifndef SQLITE_OMIT_AUTOVACUUM */ +# define setChildPtrmaps(x) SQLITE_OK +#endif + +/* +** This routine does the first phase of a two-phase commit. This routine +** causes a rollback journal to be created (if it does not already exist) +** and populated with enough information so that if a power loss occurs +** the database can be restored to its original state by playing back +** the journal. Then the contents of the journal are flushed out to +** the disk. After the journal is safely on oxide, the changes to the +** database are written into the database file and flushed to oxide. +** At the end of this call, the rollback journal still exists on the +** disk and we are still holding all locks, so the transaction has not +** committed. See sqlite3BtreeCommitPhaseTwo() for the second phase of the +** commit process. +** +** This call is a no-op if no write-transaction is currently active on pBt. +** +** Otherwise, sync the database file for the btree pBt. zMaster points to +** the name of a master journal file that should be written into the +** individual journal file, or is NULL, indicating no master journal file +** (single database transaction). +** +** When this is called, the master journal should already have been +** created, populated with this journal pointer and synced to disk. +** +** Once this is routine has returned, the only thing required to commit +** the write-transaction for this database file is to delete the journal. +*/ +SQLITE_PRIVATE int sqlite3BtreeCommitPhaseOne(Btree *p, const char *zMaster){ + int rc = SQLITE_OK; + if( p->inTrans==TRANS_WRITE ){ + BtShared *pBt = p->pBt; + sqlite3BtreeEnter(p); +#ifndef SQLITE_OMIT_AUTOVACUUM + if( pBt->autoVacuum ){ + rc = autoVacuumCommit(pBt); + if( rc!=SQLITE_OK ){ + sqlite3BtreeLeave(p); + return rc; + } + } + if( pBt->bDoTruncate ){ + sqlite3PagerTruncateImage(pBt->pPager, pBt->nPage); + } +#endif + rc = sqlite3PagerCommitPhaseOne(pBt->pPager, zMaster, 0); + sqlite3BtreeLeave(p); + } + return rc; +} + +/* +** This function is called from both BtreeCommitPhaseTwo() and BtreeRollback() +** at the conclusion of a transaction. +*/ +static void btreeEndTransaction(Btree *p){ + BtShared *pBt = p->pBt; + sqlite3 *db = p->db; + assert( sqlite3BtreeHoldsMutex(p) ); + +#ifndef SQLITE_OMIT_AUTOVACUUM + pBt->bDoTruncate = 0; +#endif + if( p->inTrans>TRANS_NONE && db->nVdbeRead>1 ){ + /* If there are other active statements that belong to this database + ** handle, downgrade to a read-only transaction. The other statements + ** may still be reading from the database. */ + downgradeAllSharedCacheTableLocks(p); + p->inTrans = TRANS_READ; + }else{ + /* If the handle had any kind of transaction open, decrement the + ** transaction count of the shared btree. If the transaction count + ** reaches 0, set the shared state to TRANS_NONE. The unlockBtreeIfUnused() + ** call below will unlock the pager. */ + if( p->inTrans!=TRANS_NONE ){ + clearAllSharedCacheTableLocks(p); + pBt->nTransaction--; + if( 0==pBt->nTransaction ){ + pBt->inTransaction = TRANS_NONE; + } + } + + /* Set the current transaction state to TRANS_NONE and unlock the + ** pager if this call closed the only read or write transaction. */ + p->inTrans = TRANS_NONE; + unlockBtreeIfUnused(pBt); + } + + btreeIntegrity(p); +} + +/* +** Commit the transaction currently in progress. +** +** This routine implements the second phase of a 2-phase commit. The +** sqlite3BtreeCommitPhaseOne() routine does the first phase and should +** be invoked prior to calling this routine. The sqlite3BtreeCommitPhaseOne() +** routine did all the work of writing information out to disk and flushing the +** contents so that they are written onto the disk platter. All this +** routine has to do is delete or truncate or zero the header in the +** the rollback journal (which causes the transaction to commit) and +** drop locks. +** +** Normally, if an error occurs while the pager layer is attempting to +** finalize the underlying journal file, this function returns an error and +** the upper layer will attempt a rollback. However, if the second argument +** is non-zero then this b-tree transaction is part of a multi-file +** transaction. In this case, the transaction has already been committed +** (by deleting a master journal file) and the caller will ignore this +** functions return code. So, even if an error occurs in the pager layer, +** reset the b-tree objects internal state to indicate that the write +** transaction has been closed. This is quite safe, as the pager will have +** transitioned to the error state. +** +** This will release the write lock on the database file. If there +** are no active cursors, it also releases the read lock. +*/ +SQLITE_PRIVATE int sqlite3BtreeCommitPhaseTwo(Btree *p, int bCleanup){ + + if( p->inTrans==TRANS_NONE ) return SQLITE_OK; + sqlite3BtreeEnter(p); + btreeIntegrity(p); + + /* If the handle has a write-transaction open, commit the shared-btrees + ** transaction and set the shared state to TRANS_READ. + */ + if( p->inTrans==TRANS_WRITE ){ + int rc; + BtShared *pBt = p->pBt; + assert( pBt->inTransaction==TRANS_WRITE ); + assert( pBt->nTransaction>0 ); + rc = sqlite3PagerCommitPhaseTwo(pBt->pPager); + if( rc!=SQLITE_OK && bCleanup==0 ){ + sqlite3BtreeLeave(p); + return rc; + } + p->iDataVersion--; /* Compensate for pPager->iDataVersion++; */ + pBt->inTransaction = TRANS_READ; + btreeClearHasContent(pBt); + } + + btreeEndTransaction(p); + sqlite3BtreeLeave(p); + return SQLITE_OK; +} + +/* +** Do both phases of a commit. +*/ +SQLITE_PRIVATE int sqlite3BtreeCommit(Btree *p){ + int rc; + sqlite3BtreeEnter(p); + rc = sqlite3BtreeCommitPhaseOne(p, 0); + if( rc==SQLITE_OK ){ + rc = sqlite3BtreeCommitPhaseTwo(p, 0); + } + sqlite3BtreeLeave(p); + return rc; +} + +/* +** This routine sets the state to CURSOR_FAULT and the error +** code to errCode for every cursor on any BtShared that pBtree +** references. Or if the writeOnly flag is set to 1, then only +** trip write cursors and leave read cursors unchanged. +** +** Every cursor is a candidate to be tripped, including cursors +** that belong to other database connections that happen to be +** sharing the cache with pBtree. +** +** This routine gets called when a rollback occurs. If the writeOnly +** flag is true, then only write-cursors need be tripped - read-only +** cursors save their current positions so that they may continue +** following the rollback. Or, if writeOnly is false, all cursors are +** tripped. In general, writeOnly is false if the transaction being +** rolled back modified the database schema. In this case b-tree root +** pages may be moved or deleted from the database altogether, making +** it unsafe for read cursors to continue. +** +** If the writeOnly flag is true and an error is encountered while +** saving the current position of a read-only cursor, all cursors, +** including all read-cursors are tripped. +** +** SQLITE_OK is returned if successful, or if an error occurs while +** saving a cursor position, an SQLite error code. +*/ +SQLITE_PRIVATE int sqlite3BtreeTripAllCursors(Btree *pBtree, int errCode, int writeOnly){ + BtCursor *p; + int rc = SQLITE_OK; + + assert( (writeOnly==0 || writeOnly==1) && BTCF_WriteFlag==1 ); + if( pBtree ){ + sqlite3BtreeEnter(pBtree); + for(p=pBtree->pBt->pCursor; p; p=p->pNext){ + if( writeOnly && (p->curFlags & BTCF_WriteFlag)==0 ){ + if( p->eState==CURSOR_VALID || p->eState==CURSOR_SKIPNEXT ){ + rc = saveCursorPosition(p); + if( rc!=SQLITE_OK ){ + (void)sqlite3BtreeTripAllCursors(pBtree, rc, 0); + break; + } + } + }else{ + sqlite3BtreeClearCursor(p); + p->eState = CURSOR_FAULT; + p->skipNext = errCode; + } + btreeReleaseAllCursorPages(p); + } + sqlite3BtreeLeave(pBtree); + } + return rc; +} + +/* +** Set the pBt->nPage field correctly, according to the current +** state of the database. Assume pBt->pPage1 is valid. +*/ +static void btreeSetNPage(BtShared *pBt, MemPage *pPage1){ + int nPage = get4byte(&pPage1->aData[28]); + testcase( nPage==0 ); + if( nPage==0 ) sqlite3PagerPagecount(pBt->pPager, &nPage); + testcase( pBt->nPage!=nPage ); + pBt->nPage = nPage; +} + +/* +** Rollback the transaction in progress. +** +** If tripCode is not SQLITE_OK then cursors will be invalidated (tripped). +** Only write cursors are tripped if writeOnly is true but all cursors are +** tripped if writeOnly is false. Any attempt to use +** a tripped cursor will result in an error. +** +** This will release the write lock on the database file. If there +** are no active cursors, it also releases the read lock. +*/ +SQLITE_PRIVATE int sqlite3BtreeRollback(Btree *p, int tripCode, int writeOnly){ + int rc; + BtShared *pBt = p->pBt; + MemPage *pPage1; + + assert( writeOnly==1 || writeOnly==0 ); + assert( tripCode==SQLITE_ABORT_ROLLBACK || tripCode==SQLITE_OK ); + sqlite3BtreeEnter(p); + if( tripCode==SQLITE_OK ){ + rc = tripCode = saveAllCursors(pBt, 0, 0); + if( rc ) writeOnly = 0; + }else{ + rc = SQLITE_OK; + } + if( tripCode ){ + int rc2 = sqlite3BtreeTripAllCursors(p, tripCode, writeOnly); + assert( rc==SQLITE_OK || (writeOnly==0 && rc2==SQLITE_OK) ); + if( rc2!=SQLITE_OK ) rc = rc2; + } + btreeIntegrity(p); + + if( p->inTrans==TRANS_WRITE ){ + int rc2; + + assert( TRANS_WRITE==pBt->inTransaction ); + rc2 = sqlite3PagerRollback(pBt->pPager); + if( rc2!=SQLITE_OK ){ + rc = rc2; + } + + /* The rollback may have destroyed the pPage1->aData value. So + ** call btreeGetPage() on page 1 again to make + ** sure pPage1->aData is set correctly. */ + if( btreeGetPage(pBt, 1, &pPage1, 0)==SQLITE_OK ){ + btreeSetNPage(pBt, pPage1); + releasePageOne(pPage1); + } + assert( countValidCursors(pBt, 1)==0 ); + pBt->inTransaction = TRANS_READ; + btreeClearHasContent(pBt); + } + + btreeEndTransaction(p); + sqlite3BtreeLeave(p); + return rc; +} + +/* +** Start a statement subtransaction. The subtransaction can be rolled +** back independently of the main transaction. You must start a transaction +** before starting a subtransaction. The subtransaction is ended automatically +** if the main transaction commits or rolls back. +** +** Statement subtransactions are used around individual SQL statements +** that are contained within a BEGIN...COMMIT block. If a constraint +** error occurs within the statement, the effect of that one statement +** can be rolled back without having to rollback the entire transaction. +** +** A statement sub-transaction is implemented as an anonymous savepoint. The +** value passed as the second parameter is the total number of savepoints, +** including the new anonymous savepoint, open on the B-Tree. i.e. if there +** are no active savepoints and no other statement-transactions open, +** iStatement is 1. This anonymous savepoint can be released or rolled back +** using the sqlite3BtreeSavepoint() function. +*/ +SQLITE_PRIVATE int sqlite3BtreeBeginStmt(Btree *p, int iStatement){ + int rc; + BtShared *pBt = p->pBt; + sqlite3BtreeEnter(p); + assert( p->inTrans==TRANS_WRITE ); + assert( (pBt->btsFlags & BTS_READ_ONLY)==0 ); + assert( iStatement>0 ); + assert( iStatement>p->db->nSavepoint ); + assert( pBt->inTransaction==TRANS_WRITE ); + /* At the pager level, a statement transaction is a savepoint with + ** an index greater than all savepoints created explicitly using + ** SQL statements. It is illegal to open, release or rollback any + ** such savepoints while the statement transaction savepoint is active. + */ + rc = sqlite3PagerOpenSavepoint(pBt->pPager, iStatement); + sqlite3BtreeLeave(p); + return rc; +} + +/* +** The second argument to this function, op, is always SAVEPOINT_ROLLBACK +** or SAVEPOINT_RELEASE. This function either releases or rolls back the +** savepoint identified by parameter iSavepoint, depending on the value +** of op. +** +** Normally, iSavepoint is greater than or equal to zero. However, if op is +** SAVEPOINT_ROLLBACK, then iSavepoint may also be -1. In this case the +** contents of the entire transaction are rolled back. This is different +** from a normal transaction rollback, as no locks are released and the +** transaction remains open. +*/ +SQLITE_PRIVATE int sqlite3BtreeSavepoint(Btree *p, int op, int iSavepoint){ + int rc = SQLITE_OK; + if( p && p->inTrans==TRANS_WRITE ){ + BtShared *pBt = p->pBt; + assert( op==SAVEPOINT_RELEASE || op==SAVEPOINT_ROLLBACK ); + assert( iSavepoint>=0 || (iSavepoint==-1 && op==SAVEPOINT_ROLLBACK) ); + sqlite3BtreeEnter(p); + if( op==SAVEPOINT_ROLLBACK ){ + rc = saveAllCursors(pBt, 0, 0); + } + if( rc==SQLITE_OK ){ + rc = sqlite3PagerSavepoint(pBt->pPager, op, iSavepoint); + } + if( rc==SQLITE_OK ){ + if( iSavepoint<0 && (pBt->btsFlags & BTS_INITIALLY_EMPTY)!=0 ){ + pBt->nPage = 0; + } + rc = newDatabase(pBt); + btreeSetNPage(pBt, pBt->pPage1); + + /* pBt->nPage might be zero if the database was corrupt when + ** the transaction was started. Otherwise, it must be at least 1. */ + assert( CORRUPT_DB || pBt->nPage>0 ); + } + sqlite3BtreeLeave(p); + } + return rc; +} + +/* +** Create a new cursor for the BTree whose root is on the page +** iTable. If a read-only cursor is requested, it is assumed that +** the caller already has at least a read-only transaction open +** on the database already. If a write-cursor is requested, then +** the caller is assumed to have an open write transaction. +** +** If the BTREE_WRCSR bit of wrFlag is clear, then the cursor can only +** be used for reading. If the BTREE_WRCSR bit is set, then the cursor +** can be used for reading or for writing if other conditions for writing +** are also met. These are the conditions that must be met in order +** for writing to be allowed: +** +** 1: The cursor must have been opened with wrFlag containing BTREE_WRCSR +** +** 2: Other database connections that share the same pager cache +** but which are not in the READ_UNCOMMITTED state may not have +** cursors open with wrFlag==0 on the same table. Otherwise +** the changes made by this write cursor would be visible to +** the read cursors in the other database connection. +** +** 3: The database must be writable (not on read-only media) +** +** 4: There must be an active transaction. +** +** The BTREE_FORDELETE bit of wrFlag may optionally be set if BTREE_WRCSR +** is set. If FORDELETE is set, that is a hint to the implementation that +** this cursor will only be used to seek to and delete entries of an index +** as part of a larger DELETE statement. The FORDELETE hint is not used by +** this implementation. But in a hypothetical alternative storage engine +** in which index entries are automatically deleted when corresponding table +** rows are deleted, the FORDELETE flag is a hint that all SEEK and DELETE +** operations on this cursor can be no-ops and all READ operations can +** return a null row (2-bytes: 0x01 0x00). +** +** No checking is done to make sure that page iTable really is the +** root page of a b-tree. If it is not, then the cursor acquired +** will not work correctly. +** +** It is assumed that the sqlite3BtreeCursorZero() has been called +** on pCur to initialize the memory space prior to invoking this routine. +*/ +static int btreeCursor( + Btree *p, /* The btree */ + int iTable, /* Root page of table to open */ + int wrFlag, /* 1 to write. 0 read-only */ + struct KeyInfo *pKeyInfo, /* First arg to comparison function */ + BtCursor *pCur /* Space for new cursor */ +){ + BtShared *pBt = p->pBt; /* Shared b-tree handle */ + BtCursor *pX; /* Looping over other all cursors */ + + assert( sqlite3BtreeHoldsMutex(p) ); + assert( wrFlag==0 + || wrFlag==BTREE_WRCSR + || wrFlag==(BTREE_WRCSR|BTREE_FORDELETE) + ); + + /* The following assert statements verify that if this is a sharable + ** b-tree database, the connection is holding the required table locks, + ** and that no other connection has any open cursor that conflicts with + ** this lock. */ + assert( hasSharedCacheTableLock(p, iTable, pKeyInfo!=0, (wrFlag?2:1)) ); + assert( wrFlag==0 || !hasReadConflicts(p, iTable) ); + + /* Assert that the caller has opened the required transaction. */ + assert( p->inTrans>TRANS_NONE ); + assert( wrFlag==0 || p->inTrans==TRANS_WRITE ); + assert( pBt->pPage1 && pBt->pPage1->aData ); + assert( wrFlag==0 || (pBt->btsFlags & BTS_READ_ONLY)==0 ); + + if( wrFlag ){ + allocateTempSpace(pBt); + if( pBt->pTmpSpace==0 ) return SQLITE_NOMEM_BKPT; + } + if( iTable==1 && btreePagecount(pBt)==0 ){ + assert( wrFlag==0 ); + iTable = 0; + } + + /* Now that no other errors can occur, finish filling in the BtCursor + ** variables and link the cursor into the BtShared list. */ + pCur->pgnoRoot = (Pgno)iTable; + pCur->iPage = -1; + pCur->pKeyInfo = pKeyInfo; + pCur->pBtree = p; + pCur->pBt = pBt; + pCur->curFlags = wrFlag ? BTCF_WriteFlag : 0; + pCur->curPagerFlags = wrFlag ? 0 : PAGER_GET_READONLY; + /* If there are two or more cursors on the same btree, then all such + ** cursors *must* have the BTCF_Multiple flag set. */ + for(pX=pBt->pCursor; pX; pX=pX->pNext){ + if( pX->pgnoRoot==(Pgno)iTable ){ + pX->curFlags |= BTCF_Multiple; + pCur->curFlags |= BTCF_Multiple; + } + } + pCur->pNext = pBt->pCursor; + pBt->pCursor = pCur; + pCur->eState = CURSOR_INVALID; + return SQLITE_OK; +} +SQLITE_PRIVATE int sqlite3BtreeCursor( + Btree *p, /* The btree */ + int iTable, /* Root page of table to open */ + int wrFlag, /* 1 to write. 0 read-only */ + struct KeyInfo *pKeyInfo, /* First arg to xCompare() */ + BtCursor *pCur /* Write new cursor here */ +){ + int rc; + if( iTable<1 ){ + rc = SQLITE_CORRUPT_BKPT; + }else{ + sqlite3BtreeEnter(p); + rc = btreeCursor(p, iTable, wrFlag, pKeyInfo, pCur); + sqlite3BtreeLeave(p); + } + return rc; +} + +/* +** Return the size of a BtCursor object in bytes. +** +** This interfaces is needed so that users of cursors can preallocate +** sufficient storage to hold a cursor. The BtCursor object is opaque +** to users so they cannot do the sizeof() themselves - they must call +** this routine. +*/ +SQLITE_PRIVATE int sqlite3BtreeCursorSize(void){ + return ROUND8(sizeof(BtCursor)); +} + +/* +** Initialize memory that will be converted into a BtCursor object. +** +** The simple approach here would be to memset() the entire object +** to zero. But it turns out that the apPage[] and aiIdx[] arrays +** do not need to be zeroed and they are large, so we can save a lot +** of run-time by skipping the initialization of those elements. +*/ +SQLITE_PRIVATE void sqlite3BtreeCursorZero(BtCursor *p){ + memset(p, 0, offsetof(BtCursor, BTCURSOR_FIRST_UNINIT)); +} + +/* +** Close a cursor. The read lock on the database file is released +** when the last cursor is closed. +*/ +SQLITE_PRIVATE int sqlite3BtreeCloseCursor(BtCursor *pCur){ + Btree *pBtree = pCur->pBtree; + if( pBtree ){ + BtShared *pBt = pCur->pBt; + sqlite3BtreeEnter(pBtree); + assert( pBt->pCursor!=0 ); + if( pBt->pCursor==pCur ){ + pBt->pCursor = pCur->pNext; + }else{ + BtCursor *pPrev = pBt->pCursor; + do{ + if( pPrev->pNext==pCur ){ + pPrev->pNext = pCur->pNext; + break; + } + pPrev = pPrev->pNext; + }while( ALWAYS(pPrev) ); + } + btreeReleaseAllCursorPages(pCur); + unlockBtreeIfUnused(pBt); + sqlite3_free(pCur->aOverflow); + sqlite3_free(pCur->pKey); + sqlite3BtreeLeave(pBtree); + pCur->pBtree = 0; + } + return SQLITE_OK; +} + +/* +** Make sure the BtCursor* given in the argument has a valid +** BtCursor.info structure. If it is not already valid, call +** btreeParseCell() to fill it in. +** +** BtCursor.info is a cache of the information in the current cell. +** Using this cache reduces the number of calls to btreeParseCell(). +*/ +#ifndef NDEBUG + static int cellInfoEqual(CellInfo *a, CellInfo *b){ + if( a->nKey!=b->nKey ) return 0; + if( a->pPayload!=b->pPayload ) return 0; + if( a->nPayload!=b->nPayload ) return 0; + if( a->nLocal!=b->nLocal ) return 0; + if( a->nSize!=b->nSize ) return 0; + return 1; + } + static void assertCellInfo(BtCursor *pCur){ + CellInfo info; + memset(&info, 0, sizeof(info)); + btreeParseCell(pCur->pPage, pCur->ix, &info); + assert( CORRUPT_DB || cellInfoEqual(&info, &pCur->info) ); + } +#else + #define assertCellInfo(x) +#endif +static SQLITE_NOINLINE void getCellInfo(BtCursor *pCur){ + if( pCur->info.nSize==0 ){ + pCur->curFlags |= BTCF_ValidNKey; + btreeParseCell(pCur->pPage,pCur->ix,&pCur->info); + }else{ + assertCellInfo(pCur); + } +} + +#ifndef NDEBUG /* The next routine used only within assert() statements */ +/* +** Return true if the given BtCursor is valid. A valid cursor is one +** that is currently pointing to a row in a (non-empty) table. +** This is a verification routine is used only within assert() statements. +*/ +SQLITE_PRIVATE int sqlite3BtreeCursorIsValid(BtCursor *pCur){ + return pCur && pCur->eState==CURSOR_VALID; +} +#endif /* NDEBUG */ +SQLITE_PRIVATE int sqlite3BtreeCursorIsValidNN(BtCursor *pCur){ + assert( pCur!=0 ); + return pCur->eState==CURSOR_VALID; +} + +/* +** Return the value of the integer key or "rowid" for a table btree. +** This routine is only valid for a cursor that is pointing into a +** ordinary table btree. If the cursor points to an index btree or +** is invalid, the result of this routine is undefined. +*/ +SQLITE_PRIVATE i64 sqlite3BtreeIntegerKey(BtCursor *pCur){ + assert( cursorHoldsMutex(pCur) ); + assert( pCur->eState==CURSOR_VALID ); + assert( pCur->curIntKey ); + getCellInfo(pCur); + return pCur->info.nKey; +} + +#ifdef SQLITE_ENABLE_OFFSET_SQL_FUNC +/* +** Return the offset into the database file for the start of the +** payload to which the cursor is pointing. +*/ +SQLITE_PRIVATE i64 sqlite3BtreeOffset(BtCursor *pCur){ + assert( cursorHoldsMutex(pCur) ); + assert( pCur->eState==CURSOR_VALID ); + getCellInfo(pCur); + return (i64)pCur->pBt->pageSize*((i64)pCur->pPage->pgno - 1) + + (i64)(pCur->info.pPayload - pCur->pPage->aData); +} +#endif /* SQLITE_ENABLE_OFFSET_SQL_FUNC */ + +/* +** Return the number of bytes of payload for the entry that pCur is +** currently pointing to. For table btrees, this will be the amount +** of data. For index btrees, this will be the size of the key. +** +** The caller must guarantee that the cursor is pointing to a non-NULL +** valid entry. In other words, the calling procedure must guarantee +** that the cursor has Cursor.eState==CURSOR_VALID. +*/ +SQLITE_PRIVATE u32 sqlite3BtreePayloadSize(BtCursor *pCur){ + assert( cursorHoldsMutex(pCur) ); + assert( pCur->eState==CURSOR_VALID ); + getCellInfo(pCur); + return pCur->info.nPayload; +} + +/* +** Return an upper bound on the size of any record for the table +** that the cursor is pointing into. +** +** This is an optimization. Everything will still work if this +** routine always returns 2147483647 (which is the largest record +** that SQLite can handle) or more. But returning a smaller value might +** prevent large memory allocations when trying to interpret a +** corrupt datrabase. +** +** The current implementation merely returns the size of the underlying +** database file. +*/ +SQLITE_PRIVATE sqlite3_int64 sqlite3BtreeMaxRecordSize(BtCursor *pCur){ + assert( cursorHoldsMutex(pCur) ); + assert( pCur->eState==CURSOR_VALID ); + return pCur->pBt->pageSize * (sqlite3_int64)pCur->pBt->nPage; +} + +/* +** Given the page number of an overflow page in the database (parameter +** ovfl), this function finds the page number of the next page in the +** linked list of overflow pages. If possible, it uses the auto-vacuum +** pointer-map data instead of reading the content of page ovfl to do so. +** +** If an error occurs an SQLite error code is returned. Otherwise: +** +** The page number of the next overflow page in the linked list is +** written to *pPgnoNext. If page ovfl is the last page in its linked +** list, *pPgnoNext is set to zero. +** +** If ppPage is not NULL, and a reference to the MemPage object corresponding +** to page number pOvfl was obtained, then *ppPage is set to point to that +** reference. It is the responsibility of the caller to call releasePage() +** on *ppPage to free the reference. In no reference was obtained (because +** the pointer-map was used to obtain the value for *pPgnoNext), then +** *ppPage is set to zero. +*/ +static int getOverflowPage( + BtShared *pBt, /* The database file */ + Pgno ovfl, /* Current overflow page number */ + MemPage **ppPage, /* OUT: MemPage handle (may be NULL) */ + Pgno *pPgnoNext /* OUT: Next overflow page number */ +){ + Pgno next = 0; + MemPage *pPage = 0; + int rc = SQLITE_OK; + + assert( sqlite3_mutex_held(pBt->mutex) ); + assert(pPgnoNext); + +#ifndef SQLITE_OMIT_AUTOVACUUM + /* Try to find the next page in the overflow list using the + ** autovacuum pointer-map pages. Guess that the next page in + ** the overflow list is page number (ovfl+1). If that guess turns + ** out to be wrong, fall back to loading the data of page + ** number ovfl to determine the next page number. + */ + if( pBt->autoVacuum ){ + Pgno pgno; + Pgno iGuess = ovfl+1; + u8 eType; + + while( PTRMAP_ISPAGE(pBt, iGuess) || iGuess==PENDING_BYTE_PAGE(pBt) ){ + iGuess++; + } + + if( iGuess<=btreePagecount(pBt) ){ + rc = ptrmapGet(pBt, iGuess, &eType, &pgno); + if( rc==SQLITE_OK && eType==PTRMAP_OVERFLOW2 && pgno==ovfl ){ + next = iGuess; + rc = SQLITE_DONE; + } + } + } +#endif + + assert( next==0 || rc==SQLITE_DONE ); + if( rc==SQLITE_OK ){ + rc = btreeGetPage(pBt, ovfl, &pPage, (ppPage==0) ? PAGER_GET_READONLY : 0); + assert( rc==SQLITE_OK || pPage==0 ); + if( rc==SQLITE_OK ){ + next = get4byte(pPage->aData); + } + } + + *pPgnoNext = next; + if( ppPage ){ + *ppPage = pPage; + }else{ + releasePage(pPage); + } + return (rc==SQLITE_DONE ? SQLITE_OK : rc); +} + +/* +** Copy data from a buffer to a page, or from a page to a buffer. +** +** pPayload is a pointer to data stored on database page pDbPage. +** If argument eOp is false, then nByte bytes of data are copied +** from pPayload to the buffer pointed at by pBuf. If eOp is true, +** then sqlite3PagerWrite() is called on pDbPage and nByte bytes +** of data are copied from the buffer pBuf to pPayload. +** +** SQLITE_OK is returned on success, otherwise an error code. +*/ +static int copyPayload( + void *pPayload, /* Pointer to page data */ + void *pBuf, /* Pointer to buffer */ + int nByte, /* Number of bytes to copy */ + int eOp, /* 0 -> copy from page, 1 -> copy to page */ + DbPage *pDbPage /* Page containing pPayload */ +){ + if( eOp ){ + /* Copy data from buffer to page (a write operation) */ + int rc = sqlite3PagerWrite(pDbPage); + if( rc!=SQLITE_OK ){ + return rc; + } + memcpy(pPayload, pBuf, nByte); + }else{ + /* Copy data from page to buffer (a read operation) */ + memcpy(pBuf, pPayload, nByte); + } + return SQLITE_OK; +} + +/* +** This function is used to read or overwrite payload information +** for the entry that the pCur cursor is pointing to. The eOp +** argument is interpreted as follows: +** +** 0: The operation is a read. Populate the overflow cache. +** 1: The operation is a write. Populate the overflow cache. +** +** A total of "amt" bytes are read or written beginning at "offset". +** Data is read to or from the buffer pBuf. +** +** The content being read or written might appear on the main page +** or be scattered out on multiple overflow pages. +** +** If the current cursor entry uses one or more overflow pages +** this function may allocate space for and lazily populate +** the overflow page-list cache array (BtCursor.aOverflow). +** Subsequent calls use this cache to make seeking to the supplied offset +** more efficient. +** +** Once an overflow page-list cache has been allocated, it must be +** invalidated if some other cursor writes to the same table, or if +** the cursor is moved to a different row. Additionally, in auto-vacuum +** mode, the following events may invalidate an overflow page-list cache. +** +** * An incremental vacuum, +** * A commit in auto_vacuum="full" mode, +** * Creating a table (may require moving an overflow page). +*/ +static int accessPayload( + BtCursor *pCur, /* Cursor pointing to entry to read from */ + u32 offset, /* Begin reading this far into payload */ + u32 amt, /* Read this many bytes */ + unsigned char *pBuf, /* Write the bytes into this buffer */ + int eOp /* zero to read. non-zero to write. */ +){ + unsigned char *aPayload; + int rc = SQLITE_OK; + int iIdx = 0; + MemPage *pPage = pCur->pPage; /* Btree page of current entry */ + BtShared *pBt = pCur->pBt; /* Btree this cursor belongs to */ +#ifdef SQLITE_DIRECT_OVERFLOW_READ + unsigned char * const pBufStart = pBuf; /* Start of original out buffer */ +#endif + + assert( pPage ); + assert( eOp==0 || eOp==1 ); + assert( pCur->eState==CURSOR_VALID ); + assert( pCur->ixnCell ); + assert( cursorHoldsMutex(pCur) ); + + getCellInfo(pCur); + aPayload = pCur->info.pPayload; + assert( offset+amt <= pCur->info.nPayload ); + + assert( aPayload > pPage->aData ); + if( (uptr)(aPayload - pPage->aData) > (pBt->usableSize - pCur->info.nLocal) ){ + /* Trying to read or write past the end of the data is an error. The + ** conditional above is really: + ** &aPayload[pCur->info.nLocal] > &pPage->aData[pBt->usableSize] + ** but is recast into its current form to avoid integer overflow problems + */ + return SQLITE_CORRUPT_PAGE(pPage); + } + + /* Check if data must be read/written to/from the btree page itself. */ + if( offsetinfo.nLocal ){ + int a = amt; + if( a+offset>pCur->info.nLocal ){ + a = pCur->info.nLocal - offset; + } + rc = copyPayload(&aPayload[offset], pBuf, a, eOp, pPage->pDbPage); + offset = 0; + pBuf += a; + amt -= a; + }else{ + offset -= pCur->info.nLocal; + } + + + if( rc==SQLITE_OK && amt>0 ){ + const u32 ovflSize = pBt->usableSize - 4; /* Bytes content per ovfl page */ + Pgno nextPage; + + nextPage = get4byte(&aPayload[pCur->info.nLocal]); + + /* If the BtCursor.aOverflow[] has not been allocated, allocate it now. + ** + ** The aOverflow[] array is sized at one entry for each overflow page + ** in the overflow chain. The page number of the first overflow page is + ** stored in aOverflow[0], etc. A value of 0 in the aOverflow[] array + ** means "not yet known" (the cache is lazily populated). + */ + if( (pCur->curFlags & BTCF_ValidOvfl)==0 ){ + int nOvfl = (pCur->info.nPayload-pCur->info.nLocal+ovflSize-1)/ovflSize; + if( pCur->aOverflow==0 + || nOvfl*(int)sizeof(Pgno) > sqlite3MallocSize(pCur->aOverflow) + ){ + Pgno *aNew = (Pgno*)sqlite3Realloc( + pCur->aOverflow, nOvfl*2*sizeof(Pgno) + ); + if( aNew==0 ){ + return SQLITE_NOMEM_BKPT; + }else{ + pCur->aOverflow = aNew; + } + } + memset(pCur->aOverflow, 0, nOvfl*sizeof(Pgno)); + pCur->curFlags |= BTCF_ValidOvfl; + }else{ + /* If the overflow page-list cache has been allocated and the + ** entry for the first required overflow page is valid, skip + ** directly to it. + */ + if( pCur->aOverflow[offset/ovflSize] ){ + iIdx = (offset/ovflSize); + nextPage = pCur->aOverflow[iIdx]; + offset = (offset%ovflSize); + } + } + + assert( rc==SQLITE_OK && amt>0 ); + while( nextPage ){ + /* If required, populate the overflow page-list cache. */ + assert( pCur->aOverflow[iIdx]==0 + || pCur->aOverflow[iIdx]==nextPage + || CORRUPT_DB ); + pCur->aOverflow[iIdx] = nextPage; + + if( offset>=ovflSize ){ + /* The only reason to read this page is to obtain the page + ** number for the next page in the overflow chain. The page + ** data is not required. So first try to lookup the overflow + ** page-list cache, if any, then fall back to the getOverflowPage() + ** function. + */ + assert( pCur->curFlags & BTCF_ValidOvfl ); + assert( pCur->pBtree->db==pBt->db ); + if( pCur->aOverflow[iIdx+1] ){ + nextPage = pCur->aOverflow[iIdx+1]; + }else{ + rc = getOverflowPage(pBt, nextPage, 0, &nextPage); + } + offset -= ovflSize; + }else{ + /* Need to read this page properly. It contains some of the + ** range of data that is being read (eOp==0) or written (eOp!=0). + */ + int a = amt; + if( a + offset > ovflSize ){ + a = ovflSize - offset; + } + +#ifdef SQLITE_DIRECT_OVERFLOW_READ + /* If all the following are true: + ** + ** 1) this is a read operation, and + ** 2) data is required from the start of this overflow page, and + ** 3) there are no dirty pages in the page-cache + ** 4) the database is file-backed, and + ** 5) the page is not in the WAL file + ** 6) at least 4 bytes have already been read into the output buffer + ** + ** then data can be read directly from the database file into the + ** output buffer, bypassing the page-cache altogether. This speeds + ** up loading large records that span many overflow pages. + */ + if( eOp==0 /* (1) */ + && offset==0 /* (2) */ + && sqlite3PagerDirectReadOk(pBt->pPager, nextPage) /* (3,4,5) */ + && &pBuf[-4]>=pBufStart /* (6) */ + ){ + sqlite3_file *fd = sqlite3PagerFile(pBt->pPager); + u8 aSave[4]; + u8 *aWrite = &pBuf[-4]; + assert( aWrite>=pBufStart ); /* due to (6) */ + memcpy(aSave, aWrite, 4); + rc = sqlite3OsRead(fd, aWrite, a+4, (i64)pBt->pageSize*(nextPage-1)); + if( rc && nextPage>pBt->nPage ) rc = SQLITE_CORRUPT_BKPT; + nextPage = get4byte(aWrite); + memcpy(aWrite, aSave, 4); + }else +#endif + + { + DbPage *pDbPage; + rc = sqlite3PagerGet(pBt->pPager, nextPage, &pDbPage, + (eOp==0 ? PAGER_GET_READONLY : 0) + ); + if( rc==SQLITE_OK ){ + aPayload = sqlite3PagerGetData(pDbPage); + nextPage = get4byte(aPayload); + rc = copyPayload(&aPayload[offset+4], pBuf, a, eOp, pDbPage); + sqlite3PagerUnref(pDbPage); + offset = 0; + } + } + amt -= a; + if( amt==0 ) return rc; + pBuf += a; + } + if( rc ) break; + iIdx++; + } + } + + if( rc==SQLITE_OK && amt>0 ){ + /* Overflow chain ends prematurely */ + return SQLITE_CORRUPT_PAGE(pPage); + } + return rc; +} + +/* +** Read part of the payload for the row at which that cursor pCur is currently +** pointing. "amt" bytes will be transferred into pBuf[]. The transfer +** begins at "offset". +** +** pCur can be pointing to either a table or an index b-tree. +** If pointing to a table btree, then the content section is read. If +** pCur is pointing to an index b-tree then the key section is read. +** +** For sqlite3BtreePayload(), the caller must ensure that pCur is pointing +** to a valid row in the table. For sqlite3BtreePayloadChecked(), the +** cursor might be invalid or might need to be restored before being read. +** +** Return SQLITE_OK on success or an error code if anything goes +** wrong. An error is returned if "offset+amt" is larger than +** the available payload. +*/ +SQLITE_PRIVATE int sqlite3BtreePayload(BtCursor *pCur, u32 offset, u32 amt, void *pBuf){ + assert( cursorHoldsMutex(pCur) ); + assert( pCur->eState==CURSOR_VALID ); + assert( pCur->iPage>=0 && pCur->pPage ); + assert( pCur->ixpPage->nCell ); + return accessPayload(pCur, offset, amt, (unsigned char*)pBuf, 0); +} + +/* +** This variant of sqlite3BtreePayload() works even if the cursor has not +** in the CURSOR_VALID state. It is only used by the sqlite3_blob_read() +** interface. +*/ +#ifndef SQLITE_OMIT_INCRBLOB +static SQLITE_NOINLINE int accessPayloadChecked( + BtCursor *pCur, + u32 offset, + u32 amt, + void *pBuf +){ + int rc; + if ( pCur->eState==CURSOR_INVALID ){ + return SQLITE_ABORT; + } + assert( cursorOwnsBtShared(pCur) ); + rc = btreeRestoreCursorPosition(pCur); + return rc ? rc : accessPayload(pCur, offset, amt, pBuf, 0); +} +SQLITE_PRIVATE int sqlite3BtreePayloadChecked(BtCursor *pCur, u32 offset, u32 amt, void *pBuf){ + if( pCur->eState==CURSOR_VALID ){ + assert( cursorOwnsBtShared(pCur) ); + return accessPayload(pCur, offset, amt, pBuf, 0); + }else{ + return accessPayloadChecked(pCur, offset, amt, pBuf); + } +} +#endif /* SQLITE_OMIT_INCRBLOB */ + +/* +** Return a pointer to payload information from the entry that the +** pCur cursor is pointing to. The pointer is to the beginning of +** the key if index btrees (pPage->intKey==0) and is the data for +** table btrees (pPage->intKey==1). The number of bytes of available +** key/data is written into *pAmt. If *pAmt==0, then the value +** returned will not be a valid pointer. +** +** This routine is an optimization. It is common for the entire key +** and data to fit on the local page and for there to be no overflow +** pages. When that is so, this routine can be used to access the +** key and data without making a copy. If the key and/or data spills +** onto overflow pages, then accessPayload() must be used to reassemble +** the key/data and copy it into a preallocated buffer. +** +** The pointer returned by this routine looks directly into the cached +** page of the database. The data might change or move the next time +** any btree routine is called. +*/ +static const void *fetchPayload( + BtCursor *pCur, /* Cursor pointing to entry to read from */ + u32 *pAmt /* Write the number of available bytes here */ +){ + int amt; + assert( pCur!=0 && pCur->iPage>=0 && pCur->pPage); + assert( pCur->eState==CURSOR_VALID ); + assert( sqlite3_mutex_held(pCur->pBtree->db->mutex) ); + assert( cursorOwnsBtShared(pCur) ); + assert( pCur->ixpPage->nCell ); + assert( pCur->info.nSize>0 ); + assert( pCur->info.pPayload>pCur->pPage->aData || CORRUPT_DB ); + assert( pCur->info.pPayloadpPage->aDataEnd ||CORRUPT_DB); + amt = pCur->info.nLocal; + if( amt>(int)(pCur->pPage->aDataEnd - pCur->info.pPayload) ){ + /* There is too little space on the page for the expected amount + ** of local content. Database must be corrupt. */ + assert( CORRUPT_DB ); + amt = MAX(0, (int)(pCur->pPage->aDataEnd - pCur->info.pPayload)); + } + *pAmt = (u32)amt; + return (void*)pCur->info.pPayload; +} + + +/* +** For the entry that cursor pCur is point to, return as +** many bytes of the key or data as are available on the local +** b-tree page. Write the number of available bytes into *pAmt. +** +** The pointer returned is ephemeral. The key/data may move +** or be destroyed on the next call to any Btree routine, +** including calls from other threads against the same cache. +** Hence, a mutex on the BtShared should be held prior to calling +** this routine. +** +** These routines is used to get quick access to key and data +** in the common case where no overflow pages are used. +*/ +SQLITE_PRIVATE const void *sqlite3BtreePayloadFetch(BtCursor *pCur, u32 *pAmt){ + return fetchPayload(pCur, pAmt); +} + + +/* +** Move the cursor down to a new child page. The newPgno argument is the +** page number of the child page to move to. +** +** This function returns SQLITE_CORRUPT if the page-header flags field of +** the new child page does not match the flags field of the parent (i.e. +** if an intkey page appears to be the parent of a non-intkey page, or +** vice-versa). +*/ +static int moveToChild(BtCursor *pCur, u32 newPgno){ + BtShared *pBt = pCur->pBt; + + assert( cursorOwnsBtShared(pCur) ); + assert( pCur->eState==CURSOR_VALID ); + assert( pCur->iPageiPage>=0 ); + if( pCur->iPage>=(BTCURSOR_MAX_DEPTH-1) ){ + return SQLITE_CORRUPT_BKPT; + } + pCur->info.nSize = 0; + pCur->curFlags &= ~(BTCF_ValidNKey|BTCF_ValidOvfl); + pCur->aiIdx[pCur->iPage] = pCur->ix; + pCur->apPage[pCur->iPage] = pCur->pPage; + pCur->ix = 0; + pCur->iPage++; + return getAndInitPage(pBt, newPgno, &pCur->pPage, pCur, pCur->curPagerFlags); +} + +#ifdef SQLITE_DEBUG +/* +** Page pParent is an internal (non-leaf) tree page. This function +** asserts that page number iChild is the left-child if the iIdx'th +** cell in page pParent. Or, if iIdx is equal to the total number of +** cells in pParent, that page number iChild is the right-child of +** the page. +*/ +static void assertParentIndex(MemPage *pParent, int iIdx, Pgno iChild){ + if( CORRUPT_DB ) return; /* The conditions tested below might not be true + ** in a corrupt database */ + assert( iIdx<=pParent->nCell ); + if( iIdx==pParent->nCell ){ + assert( get4byte(&pParent->aData[pParent->hdrOffset+8])==iChild ); + }else{ + assert( get4byte(findCell(pParent, iIdx))==iChild ); + } +} +#else +# define assertParentIndex(x,y,z) +#endif + +/* +** Move the cursor up to the parent page. +** +** pCur->idx is set to the cell index that contains the pointer +** to the page we are coming from. If we are coming from the +** right-most child page then pCur->idx is set to one more than +** the largest cell index. +*/ +static void moveToParent(BtCursor *pCur){ + MemPage *pLeaf; + assert( cursorOwnsBtShared(pCur) ); + assert( pCur->eState==CURSOR_VALID ); + assert( pCur->iPage>0 ); + assert( pCur->pPage ); + assertParentIndex( + pCur->apPage[pCur->iPage-1], + pCur->aiIdx[pCur->iPage-1], + pCur->pPage->pgno + ); + testcase( pCur->aiIdx[pCur->iPage-1] > pCur->apPage[pCur->iPage-1]->nCell ); + pCur->info.nSize = 0; + pCur->curFlags &= ~(BTCF_ValidNKey|BTCF_ValidOvfl); + pCur->ix = pCur->aiIdx[pCur->iPage-1]; + pLeaf = pCur->pPage; + pCur->pPage = pCur->apPage[--pCur->iPage]; + releasePageNotNull(pLeaf); +} + +/* +** Move the cursor to point to the root page of its b-tree structure. +** +** If the table has a virtual root page, then the cursor is moved to point +** to the virtual root page instead of the actual root page. A table has a +** virtual root page when the actual root page contains no cells and a +** single child page. This can only happen with the table rooted at page 1. +** +** If the b-tree structure is empty, the cursor state is set to +** CURSOR_INVALID and this routine returns SQLITE_EMPTY. Otherwise, +** the cursor is set to point to the first cell located on the root +** (or virtual root) page and the cursor state is set to CURSOR_VALID. +** +** If this function returns successfully, it may be assumed that the +** page-header flags indicate that the [virtual] root-page is the expected +** kind of b-tree page (i.e. if when opening the cursor the caller did not +** specify a KeyInfo structure the flags byte is set to 0x05 or 0x0D, +** indicating a table b-tree, or if the caller did specify a KeyInfo +** structure the flags byte is set to 0x02 or 0x0A, indicating an index +** b-tree). +*/ +static int moveToRoot(BtCursor *pCur){ + MemPage *pRoot; + int rc = SQLITE_OK; + + assert( cursorOwnsBtShared(pCur) ); + assert( CURSOR_INVALID < CURSOR_REQUIRESEEK ); + assert( CURSOR_VALID < CURSOR_REQUIRESEEK ); + assert( CURSOR_FAULT > CURSOR_REQUIRESEEK ); + assert( pCur->eState < CURSOR_REQUIRESEEK || pCur->iPage<0 ); + assert( pCur->pgnoRoot>0 || pCur->iPage<0 ); + + if( pCur->iPage>=0 ){ + if( pCur->iPage ){ + releasePageNotNull(pCur->pPage); + while( --pCur->iPage ){ + releasePageNotNull(pCur->apPage[pCur->iPage]); + } + pCur->pPage = pCur->apPage[0]; + goto skip_init; + } + }else if( pCur->pgnoRoot==0 ){ + pCur->eState = CURSOR_INVALID; + return SQLITE_EMPTY; + }else{ + assert( pCur->iPage==(-1) ); + if( pCur->eState>=CURSOR_REQUIRESEEK ){ + if( pCur->eState==CURSOR_FAULT ){ + assert( pCur->skipNext!=SQLITE_OK ); + return pCur->skipNext; + } + sqlite3BtreeClearCursor(pCur); + } + rc = getAndInitPage(pCur->pBtree->pBt, pCur->pgnoRoot, &pCur->pPage, + 0, pCur->curPagerFlags); + if( rc!=SQLITE_OK ){ + pCur->eState = CURSOR_INVALID; + return rc; + } + pCur->iPage = 0; + pCur->curIntKey = pCur->pPage->intKey; + } + pRoot = pCur->pPage; + assert( pRoot->pgno==pCur->pgnoRoot ); + + /* If pCur->pKeyInfo is not NULL, then the caller that opened this cursor + ** expected to open it on an index b-tree. Otherwise, if pKeyInfo is + ** NULL, the caller expects a table b-tree. If this is not the case, + ** return an SQLITE_CORRUPT error. + ** + ** Earlier versions of SQLite assumed that this test could not fail + ** if the root page was already loaded when this function was called (i.e. + ** if pCur->iPage>=0). But this is not so if the database is corrupted + ** in such a way that page pRoot is linked into a second b-tree table + ** (or the freelist). */ + assert( pRoot->intKey==1 || pRoot->intKey==0 ); + if( pRoot->isInit==0 || (pCur->pKeyInfo==0)!=pRoot->intKey ){ + return SQLITE_CORRUPT_PAGE(pCur->pPage); + } + +skip_init: + pCur->ix = 0; + pCur->info.nSize = 0; + pCur->curFlags &= ~(BTCF_AtLast|BTCF_ValidNKey|BTCF_ValidOvfl); + + pRoot = pCur->pPage; + if( pRoot->nCell>0 ){ + pCur->eState = CURSOR_VALID; + }else if( !pRoot->leaf ){ + Pgno subpage; + if( pRoot->pgno!=1 ) return SQLITE_CORRUPT_BKPT; + subpage = get4byte(&pRoot->aData[pRoot->hdrOffset+8]); + pCur->eState = CURSOR_VALID; + rc = moveToChild(pCur, subpage); + }else{ + pCur->eState = CURSOR_INVALID; + rc = SQLITE_EMPTY; + } + return rc; +} + +/* +** Move the cursor down to the left-most leaf entry beneath the +** entry to which it is currently pointing. +** +** The left-most leaf is the one with the smallest key - the first +** in ascending order. +*/ +static int moveToLeftmost(BtCursor *pCur){ + Pgno pgno; + int rc = SQLITE_OK; + MemPage *pPage; + + assert( cursorOwnsBtShared(pCur) ); + assert( pCur->eState==CURSOR_VALID ); + while( rc==SQLITE_OK && !(pPage = pCur->pPage)->leaf ){ + assert( pCur->ixnCell ); + pgno = get4byte(findCell(pPage, pCur->ix)); + rc = moveToChild(pCur, pgno); + } + return rc; +} + +/* +** Move the cursor down to the right-most leaf entry beneath the +** page to which it is currently pointing. Notice the difference +** between moveToLeftmost() and moveToRightmost(). moveToLeftmost() +** finds the left-most entry beneath the *entry* whereas moveToRightmost() +** finds the right-most entry beneath the *page*. +** +** The right-most entry is the one with the largest key - the last +** key in ascending order. +*/ +static int moveToRightmost(BtCursor *pCur){ + Pgno pgno; + int rc = SQLITE_OK; + MemPage *pPage = 0; + + assert( cursorOwnsBtShared(pCur) ); + assert( pCur->eState==CURSOR_VALID ); + while( !(pPage = pCur->pPage)->leaf ){ + pgno = get4byte(&pPage->aData[pPage->hdrOffset+8]); + pCur->ix = pPage->nCell; + rc = moveToChild(pCur, pgno); + if( rc ) return rc; + } + pCur->ix = pPage->nCell-1; + assert( pCur->info.nSize==0 ); + assert( (pCur->curFlags & BTCF_ValidNKey)==0 ); + return SQLITE_OK; +} + +/* Move the cursor to the first entry in the table. Return SQLITE_OK +** on success. Set *pRes to 0 if the cursor actually points to something +** or set *pRes to 1 if the table is empty. +*/ +SQLITE_PRIVATE int sqlite3BtreeFirst(BtCursor *pCur, int *pRes){ + int rc; + + assert( cursorOwnsBtShared(pCur) ); + assert( sqlite3_mutex_held(pCur->pBtree->db->mutex) ); + rc = moveToRoot(pCur); + if( rc==SQLITE_OK ){ + assert( pCur->pPage->nCell>0 ); + *pRes = 0; + rc = moveToLeftmost(pCur); + }else if( rc==SQLITE_EMPTY ){ + assert( pCur->pgnoRoot==0 || pCur->pPage->nCell==0 ); + *pRes = 1; + rc = SQLITE_OK; + } + return rc; +} + +/* Move the cursor to the last entry in the table. Return SQLITE_OK +** on success. Set *pRes to 0 if the cursor actually points to something +** or set *pRes to 1 if the table is empty. +*/ +SQLITE_PRIVATE int sqlite3BtreeLast(BtCursor *pCur, int *pRes){ + int rc; + + assert( cursorOwnsBtShared(pCur) ); + assert( sqlite3_mutex_held(pCur->pBtree->db->mutex) ); + + /* If the cursor already points to the last entry, this is a no-op. */ + if( CURSOR_VALID==pCur->eState && (pCur->curFlags & BTCF_AtLast)!=0 ){ +#ifdef SQLITE_DEBUG + /* This block serves to assert() that the cursor really does point + ** to the last entry in the b-tree. */ + int ii; + for(ii=0; iiiPage; ii++){ + assert( pCur->aiIdx[ii]==pCur->apPage[ii]->nCell ); + } + assert( pCur->ix==pCur->pPage->nCell-1 ); + assert( pCur->pPage->leaf ); +#endif + *pRes = 0; + return SQLITE_OK; + } + + rc = moveToRoot(pCur); + if( rc==SQLITE_OK ){ + assert( pCur->eState==CURSOR_VALID ); + *pRes = 0; + rc = moveToRightmost(pCur); + if( rc==SQLITE_OK ){ + pCur->curFlags |= BTCF_AtLast; + }else{ + pCur->curFlags &= ~BTCF_AtLast; + } + }else if( rc==SQLITE_EMPTY ){ + assert( pCur->pgnoRoot==0 || pCur->pPage->nCell==0 ); + *pRes = 1; + rc = SQLITE_OK; + } + return rc; +} + +/* Move the cursor so that it points to an entry near the key +** specified by pIdxKey or intKey. Return a success code. +** +** For INTKEY tables, the intKey parameter is used. pIdxKey +** must be NULL. For index tables, pIdxKey is used and intKey +** is ignored. +** +** If an exact match is not found, then the cursor is always +** left pointing at a leaf page which would hold the entry if it +** were present. The cursor might point to an entry that comes +** before or after the key. +** +** An integer is written into *pRes which is the result of +** comparing the key with the entry to which the cursor is +** pointing. The meaning of the integer written into +** *pRes is as follows: +** +** *pRes<0 The cursor is left pointing at an entry that +** is smaller than intKey/pIdxKey or if the table is empty +** and the cursor is therefore left point to nothing. +** +** *pRes==0 The cursor is left pointing at an entry that +** exactly matches intKey/pIdxKey. +** +** *pRes>0 The cursor is left pointing at an entry that +** is larger than intKey/pIdxKey. +** +** For index tables, the pIdxKey->eqSeen field is set to 1 if there +** exists an entry in the table that exactly matches pIdxKey. +*/ +SQLITE_PRIVATE int sqlite3BtreeMovetoUnpacked( + BtCursor *pCur, /* The cursor to be moved */ + UnpackedRecord *pIdxKey, /* Unpacked index key */ + i64 intKey, /* The table key */ + int biasRight, /* If true, bias the search to the high end */ + int *pRes /* Write search results here */ +){ + int rc; + RecordCompare xRecordCompare; + + assert( cursorOwnsBtShared(pCur) ); + assert( sqlite3_mutex_held(pCur->pBtree->db->mutex) ); + assert( pRes ); + assert( (pIdxKey==0)==(pCur->pKeyInfo==0) ); + assert( pCur->eState!=CURSOR_VALID || (pIdxKey==0)==(pCur->curIntKey!=0) ); + + /* If the cursor is already positioned at the point we are trying + ** to move to, then just return without doing any work */ + if( pIdxKey==0 + && pCur->eState==CURSOR_VALID && (pCur->curFlags & BTCF_ValidNKey)!=0 + ){ + if( pCur->info.nKey==intKey ){ + *pRes = 0; + return SQLITE_OK; + } + if( pCur->info.nKeycurFlags & BTCF_AtLast)!=0 ){ + *pRes = -1; + return SQLITE_OK; + } + /* If the requested key is one more than the previous key, then + ** try to get there using sqlite3BtreeNext() rather than a full + ** binary search. This is an optimization only. The correct answer + ** is still obtained without this case, only a little more slowely */ + if( pCur->info.nKey+1==intKey ){ + *pRes = 0; + rc = sqlite3BtreeNext(pCur, 0); + if( rc==SQLITE_OK ){ + getCellInfo(pCur); + if( pCur->info.nKey==intKey ){ + return SQLITE_OK; + } + }else if( rc==SQLITE_DONE ){ + rc = SQLITE_OK; + }else{ + return rc; + } + } + } + } + + if( pIdxKey ){ + xRecordCompare = sqlite3VdbeFindCompare(pIdxKey); + pIdxKey->errCode = 0; + assert( pIdxKey->default_rc==1 + || pIdxKey->default_rc==0 + || pIdxKey->default_rc==-1 + ); + }else{ + xRecordCompare = 0; /* All keys are integers */ + } + + rc = moveToRoot(pCur); + if( rc ){ + if( rc==SQLITE_EMPTY ){ + assert( pCur->pgnoRoot==0 || pCur->pPage->nCell==0 ); + *pRes = -1; + return SQLITE_OK; + } + return rc; + } + assert( pCur->pPage ); + assert( pCur->pPage->isInit ); + assert( pCur->eState==CURSOR_VALID ); + assert( pCur->pPage->nCell > 0 ); + assert( pCur->iPage==0 || pCur->apPage[0]->intKey==pCur->curIntKey ); + assert( pCur->curIntKey || pIdxKey ); + for(;;){ + int lwr, upr, idx, c; + Pgno chldPg; + MemPage *pPage = pCur->pPage; + u8 *pCell; /* Pointer to current cell in pPage */ + + /* pPage->nCell must be greater than zero. If this is the root-page + ** the cursor would have been INVALID above and this for(;;) loop + ** not run. If this is not the root-page, then the moveToChild() routine + ** would have already detected db corruption. Similarly, pPage must + ** be the right kind (index or table) of b-tree page. Otherwise + ** a moveToChild() or moveToRoot() call would have detected corruption. */ + assert( pPage->nCell>0 ); + assert( pPage->intKey==(pIdxKey==0) ); + lwr = 0; + upr = pPage->nCell-1; + assert( biasRight==0 || biasRight==1 ); + idx = upr>>(1-biasRight); /* idx = biasRight ? upr : (lwr+upr)/2; */ + pCur->ix = (u16)idx; + if( xRecordCompare==0 ){ + for(;;){ + i64 nCellKey; + pCell = findCellPastPtr(pPage, idx); + if( pPage->intKeyLeaf ){ + while( 0x80 <= *(pCell++) ){ + if( pCell>=pPage->aDataEnd ){ + return SQLITE_CORRUPT_PAGE(pPage); + } + } + } + getVarint(pCell, (u64*)&nCellKey); + if( nCellKeyupr ){ c = -1; break; } + }else if( nCellKey>intKey ){ + upr = idx-1; + if( lwr>upr ){ c = +1; break; } + }else{ + assert( nCellKey==intKey ); + pCur->ix = (u16)idx; + if( !pPage->leaf ){ + lwr = idx; + goto moveto_next_layer; + }else{ + pCur->curFlags |= BTCF_ValidNKey; + pCur->info.nKey = nCellKey; + pCur->info.nSize = 0; + *pRes = 0; + return SQLITE_OK; + } + } + assert( lwr+upr>=0 ); + idx = (lwr+upr)>>1; /* idx = (lwr+upr)/2; */ + } + }else{ + for(;;){ + int nCell; /* Size of the pCell cell in bytes */ + pCell = findCellPastPtr(pPage, idx); + + /* The maximum supported page-size is 65536 bytes. This means that + ** the maximum number of record bytes stored on an index B-Tree + ** page is less than 16384 bytes and may be stored as a 2-byte + ** varint. This information is used to attempt to avoid parsing + ** the entire cell by checking for the cases where the record is + ** stored entirely within the b-tree page by inspecting the first + ** 2 bytes of the cell. + */ + nCell = pCell[0]; + if( nCell<=pPage->max1bytePayload ){ + /* This branch runs if the record-size field of the cell is a + ** single byte varint and the record fits entirely on the main + ** b-tree page. */ + testcase( pCell+nCell+1==pPage->aDataEnd ); + c = xRecordCompare(nCell, (void*)&pCell[1], pIdxKey); + }else if( !(pCell[1] & 0x80) + && (nCell = ((nCell&0x7f)<<7) + pCell[1])<=pPage->maxLocal + ){ + /* The record-size field is a 2 byte varint and the record + ** fits entirely on the main b-tree page. */ + testcase( pCell+nCell+2==pPage->aDataEnd ); + c = xRecordCompare(nCell, (void*)&pCell[2], pIdxKey); + }else{ + /* The record flows over onto one or more overflow pages. In + ** this case the whole cell needs to be parsed, a buffer allocated + ** and accessPayload() used to retrieve the record into the + ** buffer before VdbeRecordCompare() can be called. + ** + ** If the record is corrupt, the xRecordCompare routine may read + ** up to two varints past the end of the buffer. An extra 18 + ** bytes of padding is allocated at the end of the buffer in + ** case this happens. */ + void *pCellKey; + u8 * const pCellBody = pCell - pPage->childPtrSize; + const int nOverrun = 18; /* Size of the overrun padding */ + pPage->xParseCell(pPage, pCellBody, &pCur->info); + nCell = (int)pCur->info.nKey; + testcase( nCell<0 ); /* True if key size is 2^32 or more */ + testcase( nCell==0 ); /* Invalid key size: 0x80 0x80 0x00 */ + testcase( nCell==1 ); /* Invalid key size: 0x80 0x80 0x01 */ + testcase( nCell==2 ); /* Minimum legal index key size */ + if( nCell<2 || nCell/pCur->pBt->usableSize>pCur->pBt->nPage ){ + rc = SQLITE_CORRUPT_PAGE(pPage); + goto moveto_finish; + } + pCellKey = sqlite3Malloc( nCell+nOverrun ); + if( pCellKey==0 ){ + rc = SQLITE_NOMEM_BKPT; + goto moveto_finish; + } + pCur->ix = (u16)idx; + rc = accessPayload(pCur, 0, nCell, (unsigned char*)pCellKey, 0); + memset(((u8*)pCellKey)+nCell,0,nOverrun); /* Fix uninit warnings */ + pCur->curFlags &= ~BTCF_ValidOvfl; + if( rc ){ + sqlite3_free(pCellKey); + goto moveto_finish; + } + c = sqlite3VdbeRecordCompare(nCell, pCellKey, pIdxKey); + sqlite3_free(pCellKey); + } + assert( + (pIdxKey->errCode!=SQLITE_CORRUPT || c==0) + && (pIdxKey->errCode!=SQLITE_NOMEM || pCur->pBtree->db->mallocFailed) + ); + if( c<0 ){ + lwr = idx+1; + }else if( c>0 ){ + upr = idx-1; + }else{ + assert( c==0 ); + *pRes = 0; + rc = SQLITE_OK; + pCur->ix = (u16)idx; + if( pIdxKey->errCode ) rc = SQLITE_CORRUPT_BKPT; + goto moveto_finish; + } + if( lwr>upr ) break; + assert( lwr+upr>=0 ); + idx = (lwr+upr)>>1; /* idx = (lwr+upr)/2 */ + } + } + assert( lwr==upr+1 || (pPage->intKey && !pPage->leaf) ); + assert( pPage->isInit ); + if( pPage->leaf ){ + assert( pCur->ixpPage->nCell ); + pCur->ix = (u16)idx; + *pRes = c; + rc = SQLITE_OK; + goto moveto_finish; + } +moveto_next_layer: + if( lwr>=pPage->nCell ){ + chldPg = get4byte(&pPage->aData[pPage->hdrOffset+8]); + }else{ + chldPg = get4byte(findCell(pPage, lwr)); + } + pCur->ix = (u16)lwr; + rc = moveToChild(pCur, chldPg); + if( rc ) break; + } +moveto_finish: + pCur->info.nSize = 0; + assert( (pCur->curFlags & BTCF_ValidOvfl)==0 ); + return rc; +} + + +/* +** Return TRUE if the cursor is not pointing at an entry of the table. +** +** TRUE will be returned after a call to sqlite3BtreeNext() moves +** past the last entry in the table or sqlite3BtreePrev() moves past +** the first entry. TRUE is also returned if the table is empty. +*/ +SQLITE_PRIVATE int sqlite3BtreeEof(BtCursor *pCur){ + /* TODO: What if the cursor is in CURSOR_REQUIRESEEK but all table entries + ** have been deleted? This API will need to change to return an error code + ** as well as the boolean result value. + */ + return (CURSOR_VALID!=pCur->eState); +} + +/* +** Return an estimate for the number of rows in the table that pCur is +** pointing to. Return a negative number if no estimate is currently +** available. +*/ +SQLITE_PRIVATE i64 sqlite3BtreeRowCountEst(BtCursor *pCur){ + i64 n; + u8 i; + + assert( cursorOwnsBtShared(pCur) ); + assert( sqlite3_mutex_held(pCur->pBtree->db->mutex) ); + + /* Currently this interface is only called by the OP_IfSmaller + ** opcode, and it that case the cursor will always be valid and + ** will always point to a leaf node. */ + if( NEVER(pCur->eState!=CURSOR_VALID) ) return -1; + if( NEVER(pCur->pPage->leaf==0) ) return -1; + + n = pCur->pPage->nCell; + for(i=0; iiPage; i++){ + n *= pCur->apPage[i]->nCell; + } + return n; +} + +/* +** Advance the cursor to the next entry in the database. +** Return value: +** +** SQLITE_OK success +** SQLITE_DONE cursor is already pointing at the last element +** otherwise some kind of error occurred +** +** The main entry point is sqlite3BtreeNext(). That routine is optimized +** for the common case of merely incrementing the cell counter BtCursor.aiIdx +** to the next cell on the current page. The (slower) btreeNext() helper +** routine is called when it is necessary to move to a different page or +** to restore the cursor. +** +** If bit 0x01 of the F argument in sqlite3BtreeNext(C,F) is 1, then the +** cursor corresponds to an SQL index and this routine could have been +** skipped if the SQL index had been a unique index. The F argument +** is a hint to the implement. SQLite btree implementation does not use +** this hint, but COMDB2 does. +*/ +static SQLITE_NOINLINE int btreeNext(BtCursor *pCur){ + int rc; + int idx; + MemPage *pPage; + + assert( cursorOwnsBtShared(pCur) ); + if( pCur->eState!=CURSOR_VALID ){ + assert( (pCur->curFlags & BTCF_ValidOvfl)==0 ); + rc = restoreCursorPosition(pCur); + if( rc!=SQLITE_OK ){ + return rc; + } + if( CURSOR_INVALID==pCur->eState ){ + return SQLITE_DONE; + } + if( pCur->eState==CURSOR_SKIPNEXT ){ + pCur->eState = CURSOR_VALID; + if( pCur->skipNext>0 ) return SQLITE_OK; + } + } + + pPage = pCur->pPage; + idx = ++pCur->ix; + if( !pPage->isInit ){ + /* The only known way for this to happen is for there to be a + ** recursive SQL function that does a DELETE operation as part of a + ** SELECT which deletes content out from under an active cursor + ** in a corrupt database file where the table being DELETE-ed from + ** has pages in common with the table being queried. See TH3 + ** module cov1/btree78.test testcase 220 (2018-06-08) for an + ** example. */ + return SQLITE_CORRUPT_BKPT; + } + + /* If the database file is corrupt, it is possible for the value of idx + ** to be invalid here. This can only occur if a second cursor modifies + ** the page while cursor pCur is holding a reference to it. Which can + ** only happen if the database is corrupt in such a way as to link the + ** page into more than one b-tree structure. */ + testcase( idx>pPage->nCell ); + + if( idx>=pPage->nCell ){ + if( !pPage->leaf ){ + rc = moveToChild(pCur, get4byte(&pPage->aData[pPage->hdrOffset+8])); + if( rc ) return rc; + return moveToLeftmost(pCur); + } + do{ + if( pCur->iPage==0 ){ + pCur->eState = CURSOR_INVALID; + return SQLITE_DONE; + } + moveToParent(pCur); + pPage = pCur->pPage; + }while( pCur->ix>=pPage->nCell ); + if( pPage->intKey ){ + return sqlite3BtreeNext(pCur, 0); + }else{ + return SQLITE_OK; + } + } + if( pPage->leaf ){ + return SQLITE_OK; + }else{ + return moveToLeftmost(pCur); + } +} +SQLITE_PRIVATE int sqlite3BtreeNext(BtCursor *pCur, int flags){ + MemPage *pPage; + UNUSED_PARAMETER( flags ); /* Used in COMDB2 but not native SQLite */ + assert( cursorOwnsBtShared(pCur) ); + assert( flags==0 || flags==1 ); + pCur->info.nSize = 0; + pCur->curFlags &= ~(BTCF_ValidNKey|BTCF_ValidOvfl); + if( pCur->eState!=CURSOR_VALID ) return btreeNext(pCur); + pPage = pCur->pPage; + if( (++pCur->ix)>=pPage->nCell ){ + pCur->ix--; + return btreeNext(pCur); + } + if( pPage->leaf ){ + return SQLITE_OK; + }else{ + return moveToLeftmost(pCur); + } +} + +/* +** Step the cursor to the back to the previous entry in the database. +** Return values: +** +** SQLITE_OK success +** SQLITE_DONE the cursor is already on the first element of the table +** otherwise some kind of error occurred +** +** The main entry point is sqlite3BtreePrevious(). That routine is optimized +** for the common case of merely decrementing the cell counter BtCursor.aiIdx +** to the previous cell on the current page. The (slower) btreePrevious() +** helper routine is called when it is necessary to move to a different page +** or to restore the cursor. +** +** If bit 0x01 of the F argument to sqlite3BtreePrevious(C,F) is 1, then +** the cursor corresponds to an SQL index and this routine could have been +** skipped if the SQL index had been a unique index. The F argument is a +** hint to the implement. The native SQLite btree implementation does not +** use this hint, but COMDB2 does. +*/ +static SQLITE_NOINLINE int btreePrevious(BtCursor *pCur){ + int rc; + MemPage *pPage; + + assert( cursorOwnsBtShared(pCur) ); + assert( (pCur->curFlags & (BTCF_AtLast|BTCF_ValidOvfl|BTCF_ValidNKey))==0 ); + assert( pCur->info.nSize==0 ); + if( pCur->eState!=CURSOR_VALID ){ + rc = restoreCursorPosition(pCur); + if( rc!=SQLITE_OK ){ + return rc; + } + if( CURSOR_INVALID==pCur->eState ){ + return SQLITE_DONE; + } + if( CURSOR_SKIPNEXT==pCur->eState ){ + pCur->eState = CURSOR_VALID; + if( pCur->skipNext<0 ) return SQLITE_OK; + } + } + + pPage = pCur->pPage; + assert( pPage->isInit ); + if( !pPage->leaf ){ + int idx = pCur->ix; + rc = moveToChild(pCur, get4byte(findCell(pPage, idx))); + if( rc ) return rc; + rc = moveToRightmost(pCur); + }else{ + while( pCur->ix==0 ){ + if( pCur->iPage==0 ){ + pCur->eState = CURSOR_INVALID; + return SQLITE_DONE; + } + moveToParent(pCur); + } + assert( pCur->info.nSize==0 ); + assert( (pCur->curFlags & (BTCF_ValidOvfl))==0 ); + + pCur->ix--; + pPage = pCur->pPage; + if( pPage->intKey && !pPage->leaf ){ + rc = sqlite3BtreePrevious(pCur, 0); + }else{ + rc = SQLITE_OK; + } + } + return rc; +} +SQLITE_PRIVATE int sqlite3BtreePrevious(BtCursor *pCur, int flags){ + assert( cursorOwnsBtShared(pCur) ); + assert( flags==0 || flags==1 ); + UNUSED_PARAMETER( flags ); /* Used in COMDB2 but not native SQLite */ + pCur->curFlags &= ~(BTCF_AtLast|BTCF_ValidOvfl|BTCF_ValidNKey); + pCur->info.nSize = 0; + if( pCur->eState!=CURSOR_VALID + || pCur->ix==0 + || pCur->pPage->leaf==0 + ){ + return btreePrevious(pCur); + } + pCur->ix--; + return SQLITE_OK; +} + +/* +** Allocate a new page from the database file. +** +** The new page is marked as dirty. (In other words, sqlite3PagerWrite() +** has already been called on the new page.) The new page has also +** been referenced and the calling routine is responsible for calling +** sqlite3PagerUnref() on the new page when it is done. +** +** SQLITE_OK is returned on success. Any other return value indicates +** an error. *ppPage is set to NULL in the event of an error. +** +** If the "nearby" parameter is not 0, then an effort is made to +** locate a page close to the page number "nearby". This can be used in an +** attempt to keep related pages close to each other in the database file, +** which in turn can make database access faster. +** +** If the eMode parameter is BTALLOC_EXACT and the nearby page exists +** anywhere on the free-list, then it is guaranteed to be returned. If +** eMode is BTALLOC_LT then the page returned will be less than or equal +** to nearby if any such page exists. If eMode is BTALLOC_ANY then there +** are no restrictions on which page is returned. +*/ +static int allocateBtreePage( + BtShared *pBt, /* The btree */ + MemPage **ppPage, /* Store pointer to the allocated page here */ + Pgno *pPgno, /* Store the page number here */ + Pgno nearby, /* Search for a page near this one */ + u8 eMode /* BTALLOC_EXACT, BTALLOC_LT, or BTALLOC_ANY */ +){ + MemPage *pPage1; + int rc; + u32 n; /* Number of pages on the freelist */ + u32 k; /* Number of leaves on the trunk of the freelist */ + MemPage *pTrunk = 0; + MemPage *pPrevTrunk = 0; + Pgno mxPage; /* Total size of the database file */ + + assert( sqlite3_mutex_held(pBt->mutex) ); + assert( eMode==BTALLOC_ANY || (nearby>0 && IfNotOmitAV(pBt->autoVacuum)) ); + pPage1 = pBt->pPage1; + mxPage = btreePagecount(pBt); + /* EVIDENCE-OF: R-05119-02637 The 4-byte big-endian integer at offset 36 + ** stores stores the total number of pages on the freelist. */ + n = get4byte(&pPage1->aData[36]); + testcase( n==mxPage-1 ); + if( n>=mxPage ){ + return SQLITE_CORRUPT_BKPT; + } + if( n>0 ){ + /* There are pages on the freelist. Reuse one of those pages. */ + Pgno iTrunk; + u8 searchList = 0; /* If the free-list must be searched for 'nearby' */ + u32 nSearch = 0; /* Count of the number of search attempts */ + + /* If eMode==BTALLOC_EXACT and a query of the pointer-map + ** shows that the page 'nearby' is somewhere on the free-list, then + ** the entire-list will be searched for that page. + */ +#ifndef SQLITE_OMIT_AUTOVACUUM + if( eMode==BTALLOC_EXACT ){ + if( nearby<=mxPage ){ + u8 eType; + assert( nearby>0 ); + assert( pBt->autoVacuum ); + rc = ptrmapGet(pBt, nearby, &eType, 0); + if( rc ) return rc; + if( eType==PTRMAP_FREEPAGE ){ + searchList = 1; + } + } + }else if( eMode==BTALLOC_LE ){ + searchList = 1; + } +#endif + + /* Decrement the free-list count by 1. Set iTrunk to the index of the + ** first free-list trunk page. iPrevTrunk is initially 1. + */ + rc = sqlite3PagerWrite(pPage1->pDbPage); + if( rc ) return rc; + put4byte(&pPage1->aData[36], n-1); + + /* The code within this loop is run only once if the 'searchList' variable + ** is not true. Otherwise, it runs once for each trunk-page on the + ** free-list until the page 'nearby' is located (eMode==BTALLOC_EXACT) + ** or until a page less than 'nearby' is located (eMode==BTALLOC_LT) + */ + do { + pPrevTrunk = pTrunk; + if( pPrevTrunk ){ + /* EVIDENCE-OF: R-01506-11053 The first integer on a freelist trunk page + ** is the page number of the next freelist trunk page in the list or + ** zero if this is the last freelist trunk page. */ + iTrunk = get4byte(&pPrevTrunk->aData[0]); + }else{ + /* EVIDENCE-OF: R-59841-13798 The 4-byte big-endian integer at offset 32 + ** stores the page number of the first page of the freelist, or zero if + ** the freelist is empty. */ + iTrunk = get4byte(&pPage1->aData[32]); + } + testcase( iTrunk==mxPage ); + if( iTrunk>mxPage || nSearch++ > n ){ + rc = SQLITE_CORRUPT_PGNO(pPrevTrunk ? pPrevTrunk->pgno : 1); + }else{ + rc = btreeGetUnusedPage(pBt, iTrunk, &pTrunk, 0); + } + if( rc ){ + pTrunk = 0; + goto end_allocate_page; + } + assert( pTrunk!=0 ); + assert( pTrunk->aData!=0 ); + /* EVIDENCE-OF: R-13523-04394 The second integer on a freelist trunk page + ** is the number of leaf page pointers to follow. */ + k = get4byte(&pTrunk->aData[4]); + if( k==0 && !searchList ){ + /* The trunk has no leaves and the list is not being searched. + ** So extract the trunk page itself and use it as the newly + ** allocated page */ + assert( pPrevTrunk==0 ); + rc = sqlite3PagerWrite(pTrunk->pDbPage); + if( rc ){ + goto end_allocate_page; + } + *pPgno = iTrunk; + memcpy(&pPage1->aData[32], &pTrunk->aData[0], 4); + *ppPage = pTrunk; + pTrunk = 0; + TRACE(("ALLOCATE: %d trunk - %d free pages left\n", *pPgno, n-1)); + }else if( k>(u32)(pBt->usableSize/4 - 2) ){ + /* Value of k is out of range. Database corruption */ + rc = SQLITE_CORRUPT_PGNO(iTrunk); + goto end_allocate_page; +#ifndef SQLITE_OMIT_AUTOVACUUM + }else if( searchList + && (nearby==iTrunk || (iTrunkpDbPage); + if( rc ){ + goto end_allocate_page; + } + if( k==0 ){ + if( !pPrevTrunk ){ + memcpy(&pPage1->aData[32], &pTrunk->aData[0], 4); + }else{ + rc = sqlite3PagerWrite(pPrevTrunk->pDbPage); + if( rc!=SQLITE_OK ){ + goto end_allocate_page; + } + memcpy(&pPrevTrunk->aData[0], &pTrunk->aData[0], 4); + } + }else{ + /* The trunk page is required by the caller but it contains + ** pointers to free-list leaves. The first leaf becomes a trunk + ** page in this case. + */ + MemPage *pNewTrunk; + Pgno iNewTrunk = get4byte(&pTrunk->aData[8]); + if( iNewTrunk>mxPage ){ + rc = SQLITE_CORRUPT_PGNO(iTrunk); + goto end_allocate_page; + } + testcase( iNewTrunk==mxPage ); + rc = btreeGetUnusedPage(pBt, iNewTrunk, &pNewTrunk, 0); + if( rc!=SQLITE_OK ){ + goto end_allocate_page; + } + rc = sqlite3PagerWrite(pNewTrunk->pDbPage); + if( rc!=SQLITE_OK ){ + releasePage(pNewTrunk); + goto end_allocate_page; + } + memcpy(&pNewTrunk->aData[0], &pTrunk->aData[0], 4); + put4byte(&pNewTrunk->aData[4], k-1); + memcpy(&pNewTrunk->aData[8], &pTrunk->aData[12], (k-1)*4); + releasePage(pNewTrunk); + if( !pPrevTrunk ){ + assert( sqlite3PagerIswriteable(pPage1->pDbPage) ); + put4byte(&pPage1->aData[32], iNewTrunk); + }else{ + rc = sqlite3PagerWrite(pPrevTrunk->pDbPage); + if( rc ){ + goto end_allocate_page; + } + put4byte(&pPrevTrunk->aData[0], iNewTrunk); + } + } + pTrunk = 0; + TRACE(("ALLOCATE: %d trunk - %d free pages left\n", *pPgno, n-1)); +#endif + }else if( k>0 ){ + /* Extract a leaf from the trunk */ + u32 closest; + Pgno iPage; + unsigned char *aData = pTrunk->aData; + if( nearby>0 ){ + u32 i; + closest = 0; + if( eMode==BTALLOC_LE ){ + for(i=0; imxPage ){ + rc = SQLITE_CORRUPT_PGNO(iTrunk); + goto end_allocate_page; + } + testcase( iPage==mxPage ); + if( !searchList + || (iPage==nearby || (iPagepgno, n-1)); + rc = sqlite3PagerWrite(pTrunk->pDbPage); + if( rc ) goto end_allocate_page; + if( closestpDbPage); + if( rc!=SQLITE_OK ){ + releasePage(*ppPage); + *ppPage = 0; + } + } + searchList = 0; + } + } + releasePage(pPrevTrunk); + pPrevTrunk = 0; + }while( searchList ); + }else{ + /* There are no pages on the freelist, so append a new page to the + ** database image. + ** + ** Normally, new pages allocated by this block can be requested from the + ** pager layer with the 'no-content' flag set. This prevents the pager + ** from trying to read the pages content from disk. However, if the + ** current transaction has already run one or more incremental-vacuum + ** steps, then the page we are about to allocate may contain content + ** that is required in the event of a rollback. In this case, do + ** not set the no-content flag. This causes the pager to load and journal + ** the current page content before overwriting it. + ** + ** Note that the pager will not actually attempt to load or journal + ** content for any page that really does lie past the end of the database + ** file on disk. So the effects of disabling the no-content optimization + ** here are confined to those pages that lie between the end of the + ** database image and the end of the database file. + */ + int bNoContent = (0==IfNotOmitAV(pBt->bDoTruncate))? PAGER_GET_NOCONTENT:0; + + rc = sqlite3PagerWrite(pBt->pPage1->pDbPage); + if( rc ) return rc; + pBt->nPage++; + if( pBt->nPage==PENDING_BYTE_PAGE(pBt) ) pBt->nPage++; + +#ifndef SQLITE_OMIT_AUTOVACUUM + if( pBt->autoVacuum && PTRMAP_ISPAGE(pBt, pBt->nPage) ){ + /* If *pPgno refers to a pointer-map page, allocate two new pages + ** at the end of the file instead of one. The first allocated page + ** becomes a new pointer-map page, the second is used by the caller. + */ + MemPage *pPg = 0; + TRACE(("ALLOCATE: %d from end of file (pointer-map page)\n", pBt->nPage)); + assert( pBt->nPage!=PENDING_BYTE_PAGE(pBt) ); + rc = btreeGetUnusedPage(pBt, pBt->nPage, &pPg, bNoContent); + if( rc==SQLITE_OK ){ + rc = sqlite3PagerWrite(pPg->pDbPage); + releasePage(pPg); + } + if( rc ) return rc; + pBt->nPage++; + if( pBt->nPage==PENDING_BYTE_PAGE(pBt) ){ pBt->nPage++; } + } +#endif + put4byte(28 + (u8*)pBt->pPage1->aData, pBt->nPage); + *pPgno = pBt->nPage; + + assert( *pPgno!=PENDING_BYTE_PAGE(pBt) ); + rc = btreeGetUnusedPage(pBt, *pPgno, ppPage, bNoContent); + if( rc ) return rc; + rc = sqlite3PagerWrite((*ppPage)->pDbPage); + if( rc!=SQLITE_OK ){ + releasePage(*ppPage); + *ppPage = 0; + } + TRACE(("ALLOCATE: %d from end of file\n", *pPgno)); + } + + assert( CORRUPT_DB || *pPgno!=PENDING_BYTE_PAGE(pBt) ); + +end_allocate_page: + releasePage(pTrunk); + releasePage(pPrevTrunk); + assert( rc!=SQLITE_OK || sqlite3PagerPageRefcount((*ppPage)->pDbPage)<=1 ); + assert( rc!=SQLITE_OK || (*ppPage)->isInit==0 ); + return rc; +} + +/* +** This function is used to add page iPage to the database file free-list. +** It is assumed that the page is not already a part of the free-list. +** +** The value passed as the second argument to this function is optional. +** If the caller happens to have a pointer to the MemPage object +** corresponding to page iPage handy, it may pass it as the second value. +** Otherwise, it may pass NULL. +** +** If a pointer to a MemPage object is passed as the second argument, +** its reference count is not altered by this function. +*/ +static int freePage2(BtShared *pBt, MemPage *pMemPage, Pgno iPage){ + MemPage *pTrunk = 0; /* Free-list trunk page */ + Pgno iTrunk = 0; /* Page number of free-list trunk page */ + MemPage *pPage1 = pBt->pPage1; /* Local reference to page 1 */ + MemPage *pPage; /* Page being freed. May be NULL. */ + int rc; /* Return Code */ + u32 nFree; /* Initial number of pages on free-list */ + + assert( sqlite3_mutex_held(pBt->mutex) ); + assert( CORRUPT_DB || iPage>1 ); + assert( !pMemPage || pMemPage->pgno==iPage ); + + if( iPage<2 || iPage>pBt->nPage ){ + return SQLITE_CORRUPT_BKPT; + } + if( pMemPage ){ + pPage = pMemPage; + sqlite3PagerRef(pPage->pDbPage); + }else{ + pPage = btreePageLookup(pBt, iPage); + } + + /* Increment the free page count on pPage1 */ + rc = sqlite3PagerWrite(pPage1->pDbPage); + if( rc ) goto freepage_out; + nFree = get4byte(&pPage1->aData[36]); + put4byte(&pPage1->aData[36], nFree+1); + + if( pBt->btsFlags & BTS_SECURE_DELETE ){ + /* If the secure_delete option is enabled, then + ** always fully overwrite deleted information with zeros. + */ + if( (!pPage && ((rc = btreeGetPage(pBt, iPage, &pPage, 0))!=0) ) + || ((rc = sqlite3PagerWrite(pPage->pDbPage))!=0) + ){ + goto freepage_out; + } + memset(pPage->aData, 0, pPage->pBt->pageSize); + } + + /* If the database supports auto-vacuum, write an entry in the pointer-map + ** to indicate that the page is free. + */ + if( ISAUTOVACUUM ){ + ptrmapPut(pBt, iPage, PTRMAP_FREEPAGE, 0, &rc); + if( rc ) goto freepage_out; + } + + /* Now manipulate the actual database free-list structure. There are two + ** possibilities. If the free-list is currently empty, or if the first + ** trunk page in the free-list is full, then this page will become a + ** new free-list trunk page. Otherwise, it will become a leaf of the + ** first trunk page in the current free-list. This block tests if it + ** is possible to add the page as a new free-list leaf. + */ + if( nFree!=0 ){ + u32 nLeaf; /* Initial number of leaf cells on trunk page */ + + iTrunk = get4byte(&pPage1->aData[32]); + rc = btreeGetPage(pBt, iTrunk, &pTrunk, 0); + if( rc!=SQLITE_OK ){ + goto freepage_out; + } + + nLeaf = get4byte(&pTrunk->aData[4]); + assert( pBt->usableSize>32 ); + if( nLeaf > (u32)pBt->usableSize/4 - 2 ){ + rc = SQLITE_CORRUPT_BKPT; + goto freepage_out; + } + if( nLeaf < (u32)pBt->usableSize/4 - 8 ){ + /* In this case there is room on the trunk page to insert the page + ** being freed as a new leaf. + ** + ** Note that the trunk page is not really full until it contains + ** usableSize/4 - 2 entries, not usableSize/4 - 8 entries as we have + ** coded. But due to a coding error in versions of SQLite prior to + ** 3.6.0, databases with freelist trunk pages holding more than + ** usableSize/4 - 8 entries will be reported as corrupt. In order + ** to maintain backwards compatibility with older versions of SQLite, + ** we will continue to restrict the number of entries to usableSize/4 - 8 + ** for now. At some point in the future (once everyone has upgraded + ** to 3.6.0 or later) we should consider fixing the conditional above + ** to read "usableSize/4-2" instead of "usableSize/4-8". + ** + ** EVIDENCE-OF: R-19920-11576 However, newer versions of SQLite still + ** avoid using the last six entries in the freelist trunk page array in + ** order that database files created by newer versions of SQLite can be + ** read by older versions of SQLite. + */ + rc = sqlite3PagerWrite(pTrunk->pDbPage); + if( rc==SQLITE_OK ){ + put4byte(&pTrunk->aData[4], nLeaf+1); + put4byte(&pTrunk->aData[8+nLeaf*4], iPage); + if( pPage && (pBt->btsFlags & BTS_SECURE_DELETE)==0 ){ + sqlite3PagerDontWrite(pPage->pDbPage); + } + rc = btreeSetHasContent(pBt, iPage); + } + TRACE(("FREE-PAGE: %d leaf on trunk page %d\n",pPage->pgno,pTrunk->pgno)); + goto freepage_out; + } + } + + /* If control flows to this point, then it was not possible to add the + ** the page being freed as a leaf page of the first trunk in the free-list. + ** Possibly because the free-list is empty, or possibly because the + ** first trunk in the free-list is full. Either way, the page being freed + ** will become the new first trunk page in the free-list. + */ + if( pPage==0 && SQLITE_OK!=(rc = btreeGetPage(pBt, iPage, &pPage, 0)) ){ + goto freepage_out; + } + rc = sqlite3PagerWrite(pPage->pDbPage); + if( rc!=SQLITE_OK ){ + goto freepage_out; + } + put4byte(pPage->aData, iTrunk); + put4byte(&pPage->aData[4], 0); + put4byte(&pPage1->aData[32], iPage); + TRACE(("FREE-PAGE: %d new trunk page replacing %d\n", pPage->pgno, iTrunk)); + +freepage_out: + if( pPage ){ + pPage->isInit = 0; + } + releasePage(pPage); + releasePage(pTrunk); + return rc; +} +static void freePage(MemPage *pPage, int *pRC){ + if( (*pRC)==SQLITE_OK ){ + *pRC = freePage2(pPage->pBt, pPage, pPage->pgno); + } +} + +/* +** Free any overflow pages associated with the given Cell. Store +** size information about the cell in pInfo. +*/ +static int clearCell( + MemPage *pPage, /* The page that contains the Cell */ + unsigned char *pCell, /* First byte of the Cell */ + CellInfo *pInfo /* Size information about the cell */ +){ + BtShared *pBt; + Pgno ovflPgno; + int rc; + int nOvfl; + u32 ovflPageSize; + + assert( sqlite3_mutex_held(pPage->pBt->mutex) ); + pPage->xParseCell(pPage, pCell, pInfo); + if( pInfo->nLocal==pInfo->nPayload ){ + return SQLITE_OK; /* No overflow pages. Return without doing anything */ + } + testcase( pCell + pInfo->nSize == pPage->aDataEnd ); + testcase( pCell + (pInfo->nSize-1) == pPage->aDataEnd ); + if( pCell + pInfo->nSize > pPage->aDataEnd ){ + /* Cell extends past end of page */ + return SQLITE_CORRUPT_PAGE(pPage); + } + ovflPgno = get4byte(pCell + pInfo->nSize - 4); + pBt = pPage->pBt; + assert( pBt->usableSize > 4 ); + ovflPageSize = pBt->usableSize - 4; + nOvfl = (pInfo->nPayload - pInfo->nLocal + ovflPageSize - 1)/ovflPageSize; + assert( nOvfl>0 || + (CORRUPT_DB && (pInfo->nPayload + ovflPageSize)btreePagecount(pBt) ){ + /* 0 is not a legal page number and page 1 cannot be an + ** overflow page. Therefore if ovflPgno<2 or past the end of the + ** file the database must be corrupt. */ + return SQLITE_CORRUPT_BKPT; + } + if( nOvfl ){ + rc = getOverflowPage(pBt, ovflPgno, &pOvfl, &iNext); + if( rc ) return rc; + } + + if( ( pOvfl || ((pOvfl = btreePageLookup(pBt, ovflPgno))!=0) ) + && sqlite3PagerPageRefcount(pOvfl->pDbPage)!=1 + ){ + /* There is no reason any cursor should have an outstanding reference + ** to an overflow page belonging to a cell that is being deleted/updated. + ** So if there exists more than one reference to this page, then it + ** must not really be an overflow page and the database must be corrupt. + ** It is helpful to detect this before calling freePage2(), as + ** freePage2() may zero the page contents if secure-delete mode is + ** enabled. If this 'overflow' page happens to be a page that the + ** caller is iterating through or using in some other way, this + ** can be problematic. + */ + rc = SQLITE_CORRUPT_BKPT; + }else{ + rc = freePage2(pBt, pOvfl, ovflPgno); + } + + if( pOvfl ){ + sqlite3PagerUnref(pOvfl->pDbPage); + } + if( rc ) return rc; + ovflPgno = iNext; + } + return SQLITE_OK; +} + +/* +** Create the byte sequence used to represent a cell on page pPage +** and write that byte sequence into pCell[]. Overflow pages are +** allocated and filled in as necessary. The calling procedure +** is responsible for making sure sufficient space has been allocated +** for pCell[]. +** +** Note that pCell does not necessary need to point to the pPage->aData +** area. pCell might point to some temporary storage. The cell will +** be constructed in this temporary area then copied into pPage->aData +** later. +*/ +static int fillInCell( + MemPage *pPage, /* The page that contains the cell */ + unsigned char *pCell, /* Complete text of the cell */ + const BtreePayload *pX, /* Payload with which to construct the cell */ + int *pnSize /* Write cell size here */ +){ + int nPayload; + const u8 *pSrc; + int nSrc, n, rc, mn; + int spaceLeft; + MemPage *pToRelease; + unsigned char *pPrior; + unsigned char *pPayload; + BtShared *pBt; + Pgno pgnoOvfl; + int nHeader; + + assert( sqlite3_mutex_held(pPage->pBt->mutex) ); + + /* pPage is not necessarily writeable since pCell might be auxiliary + ** buffer space that is separate from the pPage buffer area */ + assert( pCellaData || pCell>=&pPage->aData[pPage->pBt->pageSize] + || sqlite3PagerIswriteable(pPage->pDbPage) ); + + /* Fill in the header. */ + nHeader = pPage->childPtrSize; + if( pPage->intKey ){ + nPayload = pX->nData + pX->nZero; + pSrc = pX->pData; + nSrc = pX->nData; + assert( pPage->intKeyLeaf ); /* fillInCell() only called for leaves */ + nHeader += putVarint32(&pCell[nHeader], nPayload); + nHeader += putVarint(&pCell[nHeader], *(u64*)&pX->nKey); + }else{ + assert( pX->nKey<=0x7fffffff && pX->pKey!=0 ); + nSrc = nPayload = (int)pX->nKey; + pSrc = pX->pKey; + nHeader += putVarint32(&pCell[nHeader], nPayload); + } + + /* Fill in the payload */ + pPayload = &pCell[nHeader]; + if( nPayload<=pPage->maxLocal ){ + /* This is the common case where everything fits on the btree page + ** and no overflow pages are required. */ + n = nHeader + nPayload; + testcase( n==3 ); + testcase( n==4 ); + if( n<4 ) n = 4; + *pnSize = n; + assert( nSrc<=nPayload ); + testcase( nSrcminLocal; + n = mn + (nPayload - mn) % (pPage->pBt->usableSize - 4); + testcase( n==pPage->maxLocal ); + testcase( n==pPage->maxLocal+1 ); + if( n > pPage->maxLocal ) n = mn; + spaceLeft = n; + *pnSize = n + nHeader + 4; + pPrior = &pCell[nHeader+n]; + pToRelease = 0; + pgnoOvfl = 0; + pBt = pPage->pBt; + + /* At this point variables should be set as follows: + ** + ** nPayload Total payload size in bytes + ** pPayload Begin writing payload here + ** spaceLeft Space available at pPayload. If nPayload>spaceLeft, + ** that means content must spill into overflow pages. + ** *pnSize Size of the local cell (not counting overflow pages) + ** pPrior Where to write the pgno of the first overflow page + ** + ** Use a call to btreeParseCellPtr() to verify that the values above + ** were computed correctly. + */ +#ifdef SQLITE_DEBUG + { + CellInfo info; + pPage->xParseCell(pPage, pCell, &info); + assert( nHeader==(int)(info.pPayload - pCell) ); + assert( info.nKey==pX->nKey ); + assert( *pnSize == info.nSize ); + assert( spaceLeft == info.nLocal ); + } +#endif + + /* Write the payload into the local Cell and any extra into overflow pages */ + while( 1 ){ + n = nPayload; + if( n>spaceLeft ) n = spaceLeft; + + /* If pToRelease is not zero than pPayload points into the data area + ** of pToRelease. Make sure pToRelease is still writeable. */ + assert( pToRelease==0 || sqlite3PagerIswriteable(pToRelease->pDbPage) ); + + /* If pPayload is part of the data area of pPage, then make sure pPage + ** is still writeable */ + assert( pPayloadaData || pPayload>=&pPage->aData[pBt->pageSize] + || sqlite3PagerIswriteable(pPage->pDbPage) ); + + if( nSrc>=n ){ + memcpy(pPayload, pSrc, n); + }else if( nSrc>0 ){ + n = nSrc; + memcpy(pPayload, pSrc, n); + }else{ + memset(pPayload, 0, n); + } + nPayload -= n; + if( nPayload<=0 ) break; + pPayload += n; + pSrc += n; + nSrc -= n; + spaceLeft -= n; + if( spaceLeft==0 ){ + MemPage *pOvfl = 0; +#ifndef SQLITE_OMIT_AUTOVACUUM + Pgno pgnoPtrmap = pgnoOvfl; /* Overflow page pointer-map entry page */ + if( pBt->autoVacuum ){ + do{ + pgnoOvfl++; + } while( + PTRMAP_ISPAGE(pBt, pgnoOvfl) || pgnoOvfl==PENDING_BYTE_PAGE(pBt) + ); + } +#endif + rc = allocateBtreePage(pBt, &pOvfl, &pgnoOvfl, pgnoOvfl, 0); +#ifndef SQLITE_OMIT_AUTOVACUUM + /* If the database supports auto-vacuum, and the second or subsequent + ** overflow page is being allocated, add an entry to the pointer-map + ** for that page now. + ** + ** If this is the first overflow page, then write a partial entry + ** to the pointer-map. If we write nothing to this pointer-map slot, + ** then the optimistic overflow chain processing in clearCell() + ** may misinterpret the uninitialized values and delete the + ** wrong pages from the database. + */ + if( pBt->autoVacuum && rc==SQLITE_OK ){ + u8 eType = (pgnoPtrmap?PTRMAP_OVERFLOW2:PTRMAP_OVERFLOW1); + ptrmapPut(pBt, pgnoOvfl, eType, pgnoPtrmap, &rc); + if( rc ){ + releasePage(pOvfl); + } + } +#endif + if( rc ){ + releasePage(pToRelease); + return rc; + } + + /* If pToRelease is not zero than pPrior points into the data area + ** of pToRelease. Make sure pToRelease is still writeable. */ + assert( pToRelease==0 || sqlite3PagerIswriteable(pToRelease->pDbPage) ); + + /* If pPrior is part of the data area of pPage, then make sure pPage + ** is still writeable */ + assert( pPrioraData || pPrior>=&pPage->aData[pBt->pageSize] + || sqlite3PagerIswriteable(pPage->pDbPage) ); + + put4byte(pPrior, pgnoOvfl); + releasePage(pToRelease); + pToRelease = pOvfl; + pPrior = pOvfl->aData; + put4byte(pPrior, 0); + pPayload = &pOvfl->aData[4]; + spaceLeft = pBt->usableSize - 4; + } + } + releasePage(pToRelease); + return SQLITE_OK; +} + +/* +** Remove the i-th cell from pPage. This routine effects pPage only. +** The cell content is not freed or deallocated. It is assumed that +** the cell content has been copied someplace else. This routine just +** removes the reference to the cell from pPage. +** +** "sz" must be the number of bytes in the cell. +*/ +static void dropCell(MemPage *pPage, int idx, int sz, int *pRC){ + u32 pc; /* Offset to cell content of cell being deleted */ + u8 *data; /* pPage->aData */ + u8 *ptr; /* Used to move bytes around within data[] */ + int rc; /* The return code */ + int hdr; /* Beginning of the header. 0 most pages. 100 page 1 */ + + if( *pRC ) return; + assert( idx>=0 && idxnCell ); + assert( CORRUPT_DB || sz==cellSize(pPage, idx) ); + assert( sqlite3PagerIswriteable(pPage->pDbPage) ); + assert( sqlite3_mutex_held(pPage->pBt->mutex) ); + assert( pPage->nFree>=0 ); + data = pPage->aData; + ptr = &pPage->aCellIdx[2*idx]; + pc = get2byte(ptr); + hdr = pPage->hdrOffset; + testcase( pc==get2byte(&data[hdr+5]) ); + testcase( pc+sz==pPage->pBt->usableSize ); + if( pc+sz > pPage->pBt->usableSize ){ + *pRC = SQLITE_CORRUPT_BKPT; + return; + } + rc = freeSpace(pPage, pc, sz); + if( rc ){ + *pRC = rc; + return; + } + pPage->nCell--; + if( pPage->nCell==0 ){ + memset(&data[hdr+1], 0, 4); + data[hdr+7] = 0; + put2byte(&data[hdr+5], pPage->pBt->usableSize); + pPage->nFree = pPage->pBt->usableSize - pPage->hdrOffset + - pPage->childPtrSize - 8; + }else{ + memmove(ptr, ptr+2, 2*(pPage->nCell - idx)); + put2byte(&data[hdr+3], pPage->nCell); + pPage->nFree += 2; + } +} + +/* +** Insert a new cell on pPage at cell index "i". pCell points to the +** content of the cell. +** +** If the cell content will fit on the page, then put it there. If it +** will not fit, then make a copy of the cell content into pTemp if +** pTemp is not null. Regardless of pTemp, allocate a new entry +** in pPage->apOvfl[] and make it point to the cell content (either +** in pTemp or the original pCell) and also record its index. +** Allocating a new entry in pPage->aCell[] implies that +** pPage->nOverflow is incremented. +** +** *pRC must be SQLITE_OK when this routine is called. +*/ +static void insertCell( + MemPage *pPage, /* Page into which we are copying */ + int i, /* New cell becomes the i-th cell of the page */ + u8 *pCell, /* Content of the new cell */ + int sz, /* Bytes of content in pCell */ + u8 *pTemp, /* Temp storage space for pCell, if needed */ + Pgno iChild, /* If non-zero, replace first 4 bytes with this value */ + int *pRC /* Read and write return code from here */ +){ + int idx = 0; /* Where to write new cell content in data[] */ + int j; /* Loop counter */ + u8 *data; /* The content of the whole page */ + u8 *pIns; /* The point in pPage->aCellIdx[] where no cell inserted */ + + assert( *pRC==SQLITE_OK ); + assert( i>=0 && i<=pPage->nCell+pPage->nOverflow ); + assert( MX_CELL(pPage->pBt)<=10921 ); + assert( pPage->nCell<=MX_CELL(pPage->pBt) || CORRUPT_DB ); + assert( pPage->nOverflow<=ArraySize(pPage->apOvfl) ); + assert( ArraySize(pPage->apOvfl)==ArraySize(pPage->aiOvfl) ); + assert( sqlite3_mutex_held(pPage->pBt->mutex) ); + assert( sz==pPage->xCellSize(pPage, pCell) || CORRUPT_DB ); + assert( pPage->nFree>=0 ); + if( pPage->nOverflow || sz+2>pPage->nFree ){ + if( pTemp ){ + memcpy(pTemp, pCell, sz); + pCell = pTemp; + } + if( iChild ){ + put4byte(pCell, iChild); + } + j = pPage->nOverflow++; + /* Comparison against ArraySize-1 since we hold back one extra slot + ** as a contingency. In other words, never need more than 3 overflow + ** slots but 4 are allocated, just to be safe. */ + assert( j < ArraySize(pPage->apOvfl)-1 ); + pPage->apOvfl[j] = pCell; + pPage->aiOvfl[j] = (u16)i; + + /* When multiple overflows occur, they are always sequential and in + ** sorted order. This invariants arise because multiple overflows can + ** only occur when inserting divider cells into the parent page during + ** balancing, and the dividers are adjacent and sorted. + */ + assert( j==0 || pPage->aiOvfl[j-1]<(u16)i ); /* Overflows in sorted order */ + assert( j==0 || i==pPage->aiOvfl[j-1]+1 ); /* Overflows are sequential */ + }else{ + int rc = sqlite3PagerWrite(pPage->pDbPage); + if( rc!=SQLITE_OK ){ + *pRC = rc; + return; + } + assert( sqlite3PagerIswriteable(pPage->pDbPage) ); + data = pPage->aData; + assert( &data[pPage->cellOffset]==pPage->aCellIdx ); + rc = allocateSpace(pPage, sz, &idx); + if( rc ){ *pRC = rc; return; } + /* The allocateSpace() routine guarantees the following properties + ** if it returns successfully */ + assert( idx >= 0 ); + assert( idx >= pPage->cellOffset+2*pPage->nCell+2 || CORRUPT_DB ); + assert( idx+sz <= (int)pPage->pBt->usableSize ); + pPage->nFree -= (u16)(2 + sz); + if( iChild ){ + /* In a corrupt database where an entry in the cell index section of + ** a btree page has a value of 3 or less, the pCell value might point + ** as many as 4 bytes in front of the start of the aData buffer for + ** the source page. Make sure this does not cause problems by not + ** reading the first 4 bytes */ + memcpy(&data[idx+4], pCell+4, sz-4); + put4byte(&data[idx], iChild); + }else{ + memcpy(&data[idx], pCell, sz); + } + pIns = pPage->aCellIdx + i*2; + memmove(pIns+2, pIns, 2*(pPage->nCell - i)); + put2byte(pIns, idx); + pPage->nCell++; + /* increment the cell count */ + if( (++data[pPage->hdrOffset+4])==0 ) data[pPage->hdrOffset+3]++; + assert( get2byte(&data[pPage->hdrOffset+3])==pPage->nCell || CORRUPT_DB ); +#ifndef SQLITE_OMIT_AUTOVACUUM + if( pPage->pBt->autoVacuum ){ + /* The cell may contain a pointer to an overflow page. If so, write + ** the entry for the overflow page into the pointer map. + */ + ptrmapPutOvflPtr(pPage, pPage, pCell, pRC); + } +#endif + } +} + +/* +** The following parameters determine how many adjacent pages get involved +** in a balancing operation. NN is the number of neighbors on either side +** of the page that participate in the balancing operation. NB is the +** total number of pages that participate, including the target page and +** NN neighbors on either side. +** +** The minimum value of NN is 1 (of course). Increasing NN above 1 +** (to 2 or 3) gives a modest improvement in SELECT and DELETE performance +** in exchange for a larger degradation in INSERT and UPDATE performance. +** The value of NN appears to give the best results overall. +** +** (Later:) The description above makes it seem as if these values are +** tunable - as if you could change them and recompile and it would all work. +** But that is unlikely. NB has been 3 since the inception of SQLite and +** we have never tested any other value. +*/ +#define NN 1 /* Number of neighbors on either side of pPage */ +#define NB 3 /* (NN*2+1): Total pages involved in the balance */ + +/* +** A CellArray object contains a cache of pointers and sizes for a +** consecutive sequence of cells that might be held on multiple pages. +** +** The cells in this array are the divider cell or cells from the pParent +** page plus up to three child pages. There are a total of nCell cells. +** +** pRef is a pointer to one of the pages that contributes cells. This is +** used to access information such as MemPage.intKey and MemPage.pBt->pageSize +** which should be common to all pages that contribute cells to this array. +** +** apCell[] and szCell[] hold, respectively, pointers to the start of each +** cell and the size of each cell. Some of the apCell[] pointers might refer +** to overflow cells. In other words, some apCel[] pointers might not point +** to content area of the pages. +** +** A szCell[] of zero means the size of that cell has not yet been computed. +** +** The cells come from as many as four different pages: +** +** ----------- +** | Parent | +** ----------- +** / | \ +** / | \ +** --------- --------- --------- +** |Child-1| |Child-2| |Child-3| +** --------- --------- --------- +** +** The order of cells is in the array is for an index btree is: +** +** 1. All cells from Child-1 in order +** 2. The first divider cell from Parent +** 3. All cells from Child-2 in order +** 4. The second divider cell from Parent +** 5. All cells from Child-3 in order +** +** For a table-btree (with rowids) the items 2 and 4 are empty because +** content exists only in leaves and there are no divider cells. +** +** For an index btree, the apEnd[] array holds pointer to the end of page +** for Child-1, the Parent, Child-2, the Parent (again), and Child-3, +** respectively. The ixNx[] array holds the number of cells contained in +** each of these 5 stages, and all stages to the left. Hence: +** +** ixNx[0] = Number of cells in Child-1. +** ixNx[1] = Number of cells in Child-1 plus 1 for first divider. +** ixNx[2] = Number of cells in Child-1 and Child-2 + 1 for 1st divider. +** ixNx[3] = Number of cells in Child-1 and Child-2 + both divider cells +** ixNx[4] = Total number of cells. +** +** For a table-btree, the concept is similar, except only apEnd[0]..apEnd[2] +** are used and they point to the leaf pages only, and the ixNx value are: +** +** ixNx[0] = Number of cells in Child-1. +** ixNx[1] = Number of cells in Child-1 and Child-2. +** ixNx[2] = Total number of cells. +** +** Sometimes when deleting, a child page can have zero cells. In those +** cases, ixNx[] entries with higher indexes, and the corresponding apEnd[] +** entries, shift down. The end result is that each ixNx[] entry should +** be larger than the previous +*/ +typedef struct CellArray CellArray; +struct CellArray { + int nCell; /* Number of cells in apCell[] */ + MemPage *pRef; /* Reference page */ + u8 **apCell; /* All cells begin balanced */ + u16 *szCell; /* Local size of all cells in apCell[] */ + u8 *apEnd[NB*2]; /* MemPage.aDataEnd values */ + int ixNx[NB*2]; /* Index of at which we move to the next apEnd[] */ +}; + +/* +** Make sure the cell sizes at idx, idx+1, ..., idx+N-1 have been +** computed. +*/ +static void populateCellCache(CellArray *p, int idx, int N){ + assert( idx>=0 && idx+N<=p->nCell ); + while( N>0 ){ + assert( p->apCell[idx]!=0 ); + if( p->szCell[idx]==0 ){ + p->szCell[idx] = p->pRef->xCellSize(p->pRef, p->apCell[idx]); + }else{ + assert( CORRUPT_DB || + p->szCell[idx]==p->pRef->xCellSize(p->pRef, p->apCell[idx]) ); + } + idx++; + N--; + } +} + +/* +** Return the size of the Nth element of the cell array +*/ +static SQLITE_NOINLINE u16 computeCellSize(CellArray *p, int N){ + assert( N>=0 && NnCell ); + assert( p->szCell[N]==0 ); + p->szCell[N] = p->pRef->xCellSize(p->pRef, p->apCell[N]); + return p->szCell[N]; +} +static u16 cachedCellSize(CellArray *p, int N){ + assert( N>=0 && NnCell ); + if( p->szCell[N] ) return p->szCell[N]; + return computeCellSize(p, N); +} + +/* +** Array apCell[] contains pointers to nCell b-tree page cells. The +** szCell[] array contains the size in bytes of each cell. This function +** replaces the current contents of page pPg with the contents of the cell +** array. +** +** Some of the cells in apCell[] may currently be stored in pPg. This +** function works around problems caused by this by making a copy of any +** such cells before overwriting the page data. +** +** The MemPage.nFree field is invalidated by this function. It is the +** responsibility of the caller to set it correctly. +*/ +static int rebuildPage( + CellArray *pCArray, /* Content to be added to page pPg */ + int iFirst, /* First cell in pCArray to use */ + int nCell, /* Final number of cells on page */ + MemPage *pPg /* The page to be reconstructed */ +){ + const int hdr = pPg->hdrOffset; /* Offset of header on pPg */ + u8 * const aData = pPg->aData; /* Pointer to data for pPg */ + const int usableSize = pPg->pBt->usableSize; + u8 * const pEnd = &aData[usableSize]; + int i = iFirst; /* Which cell to copy from pCArray*/ + u32 j; /* Start of cell content area */ + int iEnd = i+nCell; /* Loop terminator */ + u8 *pCellptr = pPg->aCellIdx; + u8 *pTmp = sqlite3PagerTempSpace(pPg->pBt->pPager); + u8 *pData; + int k; /* Current slot in pCArray->apEnd[] */ + u8 *pSrcEnd; /* Current pCArray->apEnd[k] value */ + + assert( i(u32)usableSize ){ j = 0; } + memcpy(&pTmp[j], &aData[j], usableSize - j); + + for(k=0; pCArray->ixNx[k]<=i && ALWAYS(kapEnd[k]; + + pData = pEnd; + while( 1/*exit by break*/ ){ + u8 *pCell = pCArray->apCell[i]; + u16 sz = pCArray->szCell[i]; + assert( sz>0 ); + if( SQLITE_WITHIN(pCell,aData,pEnd) ){ + if( ((uptr)(pCell+sz))>(uptr)pEnd ) return SQLITE_CORRUPT_BKPT; + pCell = &pTmp[pCell - aData]; + }else if( (uptr)(pCell+sz)>(uptr)pSrcEnd + && (uptr)(pCell)<(uptr)pSrcEnd + ){ + return SQLITE_CORRUPT_BKPT; + } + + pData -= sz; + put2byte(pCellptr, (pData - aData)); + pCellptr += 2; + if( pData < pCellptr ) return SQLITE_CORRUPT_BKPT; + memcpy(pData, pCell, sz); + assert( sz==pPg->xCellSize(pPg, pCell) || CORRUPT_DB ); + testcase( sz!=pPg->xCellSize(pPg,pCell) ); + i++; + if( i>=iEnd ) break; + if( pCArray->ixNx[k]<=i ){ + k++; + pSrcEnd = pCArray->apEnd[k]; + } + } + + /* The pPg->nFree field is now set incorrectly. The caller will fix it. */ + pPg->nCell = nCell; + pPg->nOverflow = 0; + + put2byte(&aData[hdr+1], 0); + put2byte(&aData[hdr+3], pPg->nCell); + put2byte(&aData[hdr+5], pData - aData); + aData[hdr+7] = 0x00; + return SQLITE_OK; +} + +/* +** The pCArray objects contains pointers to b-tree cells and the cell sizes. +** This function attempts to add the cells stored in the array to page pPg. +** If it cannot (because the page needs to be defragmented before the cells +** will fit), non-zero is returned. Otherwise, if the cells are added +** successfully, zero is returned. +** +** Argument pCellptr points to the first entry in the cell-pointer array +** (part of page pPg) to populate. After cell apCell[0] is written to the +** page body, a 16-bit offset is written to pCellptr. And so on, for each +** cell in the array. It is the responsibility of the caller to ensure +** that it is safe to overwrite this part of the cell-pointer array. +** +** When this function is called, *ppData points to the start of the +** content area on page pPg. If the size of the content area is extended, +** *ppData is updated to point to the new start of the content area +** before returning. +** +** Finally, argument pBegin points to the byte immediately following the +** end of the space required by this page for the cell-pointer area (for +** all cells - not just those inserted by the current call). If the content +** area must be extended to before this point in order to accomodate all +** cells in apCell[], then the cells do not fit and non-zero is returned. +*/ +static int pageInsertArray( + MemPage *pPg, /* Page to add cells to */ + u8 *pBegin, /* End of cell-pointer array */ + u8 **ppData, /* IN/OUT: Page content-area pointer */ + u8 *pCellptr, /* Pointer to cell-pointer area */ + int iFirst, /* Index of first cell to add */ + int nCell, /* Number of cells to add to pPg */ + CellArray *pCArray /* Array of cells */ +){ + int i = iFirst; /* Loop counter - cell index to insert */ + u8 *aData = pPg->aData; /* Complete page */ + u8 *pData = *ppData; /* Content area. A subset of aData[] */ + int iEnd = iFirst + nCell; /* End of loop. One past last cell to ins */ + int k; /* Current slot in pCArray->apEnd[] */ + u8 *pEnd; /* Maximum extent of cell data */ + assert( CORRUPT_DB || pPg->hdrOffset==0 ); /* Never called on page 1 */ + if( iEnd<=iFirst ) return 0; + for(k=0; pCArray->ixNx[k]<=i && ALWAYS(kapEnd[k]; + while( 1 /*Exit by break*/ ){ + int sz, rc; + u8 *pSlot; + assert( pCArray->szCell[i]!=0 ); + sz = pCArray->szCell[i]; + if( (aData[1]==0 && aData[2]==0) || (pSlot = pageFindSlot(pPg,sz,&rc))==0 ){ + if( (pData - pBegin)apCell[i] will never overlap on a well-formed + ** database. But they might for a corrupt database. Hence use memmove() + ** since memcpy() sends SIGABORT with overlapping buffers on OpenBSD */ + assert( (pSlot+sz)<=pCArray->apCell[i] + || pSlot>=(pCArray->apCell[i]+sz) + || CORRUPT_DB ); + if( (uptr)(pCArray->apCell[i]+sz)>(uptr)pEnd + && (uptr)(pCArray->apCell[i])<(uptr)pEnd + ){ + assert( CORRUPT_DB ); + (void)SQLITE_CORRUPT_BKPT; + return 1; + } + memmove(pSlot, pCArray->apCell[i], sz); + put2byte(pCellptr, (pSlot - aData)); + pCellptr += 2; + i++; + if( i>=iEnd ) break; + if( pCArray->ixNx[k]<=i ){ + k++; + pEnd = pCArray->apEnd[k]; + } + } + *ppData = pData; + return 0; +} + +/* +** The pCArray object contains pointers to b-tree cells and their sizes. +** +** This function adds the space associated with each cell in the array +** that is currently stored within the body of pPg to the pPg free-list. +** The cell-pointers and other fields of the page are not updated. +** +** This function returns the total number of cells added to the free-list. +*/ +static int pageFreeArray( + MemPage *pPg, /* Page to edit */ + int iFirst, /* First cell to delete */ + int nCell, /* Cells to delete */ + CellArray *pCArray /* Array of cells */ +){ + u8 * const aData = pPg->aData; + u8 * const pEnd = &aData[pPg->pBt->usableSize]; + u8 * const pStart = &aData[pPg->hdrOffset + 8 + pPg->childPtrSize]; + int nRet = 0; + int i; + int iEnd = iFirst + nCell; + u8 *pFree = 0; + int szFree = 0; + + for(i=iFirst; iapCell[i]; + if( SQLITE_WITHIN(pCell, pStart, pEnd) ){ + int sz; + /* No need to use cachedCellSize() here. The sizes of all cells that + ** are to be freed have already been computing while deciding which + ** cells need freeing */ + sz = pCArray->szCell[i]; assert( sz>0 ); + if( pFree!=(pCell + sz) ){ + if( pFree ){ + assert( pFree>aData && (pFree - aData)<65536 ); + freeSpace(pPg, (u16)(pFree - aData), szFree); + } + pFree = pCell; + szFree = sz; + if( pFree+sz>pEnd ) return 0; + }else{ + pFree = pCell; + szFree += sz; + } + nRet++; + } + } + if( pFree ){ + assert( pFree>aData && (pFree - aData)<65536 ); + freeSpace(pPg, (u16)(pFree - aData), szFree); + } + return nRet; +} + +/* +** pCArray contains pointers to and sizes of all cells in the page being +** balanced. The current page, pPg, has pPg->nCell cells starting with +** pCArray->apCell[iOld]. After balancing, this page should hold nNew cells +** starting at apCell[iNew]. +** +** This routine makes the necessary adjustments to pPg so that it contains +** the correct cells after being balanced. +** +** The pPg->nFree field is invalid when this function returns. It is the +** responsibility of the caller to set it correctly. +*/ +static int editPage( + MemPage *pPg, /* Edit this page */ + int iOld, /* Index of first cell currently on page */ + int iNew, /* Index of new first cell on page */ + int nNew, /* Final number of cells on page */ + CellArray *pCArray /* Array of cells and sizes */ +){ + u8 * const aData = pPg->aData; + const int hdr = pPg->hdrOffset; + u8 *pBegin = &pPg->aCellIdx[nNew * 2]; + int nCell = pPg->nCell; /* Cells stored on pPg */ + u8 *pData; + u8 *pCellptr; + int i; + int iOldEnd = iOld + pPg->nCell + pPg->nOverflow; + int iNewEnd = iNew + nNew; + +#ifdef SQLITE_DEBUG + u8 *pTmp = sqlite3PagerTempSpace(pPg->pBt->pPager); + memcpy(pTmp, aData, pPg->pBt->usableSize); +#endif + + /* Remove cells from the start and end of the page */ + assert( nCell>=0 ); + if( iOldnCell ) return SQLITE_CORRUPT_BKPT; + memmove(pPg->aCellIdx, &pPg->aCellIdx[nShift*2], nCell*2); + nCell -= nShift; + } + if( iNewEnd < iOldEnd ){ + int nTail = pageFreeArray(pPg, iNewEnd, iOldEnd - iNewEnd, pCArray); + assert( nCell>=nTail ); + nCell -= nTail; + } + + pData = &aData[get2byteNotZero(&aData[hdr+5])]; + if( pData=0 ); + pCellptr = pPg->aCellIdx; + memmove(&pCellptr[nAdd*2], pCellptr, nCell*2); + if( pageInsertArray( + pPg, pBegin, &pData, pCellptr, + iNew, nAdd, pCArray + ) ) goto editpage_fail; + nCell += nAdd; + } + + /* Add any overflow cells */ + for(i=0; inOverflow; i++){ + int iCell = (iOld + pPg->aiOvfl[i]) - iNew; + if( iCell>=0 && iCellaCellIdx[iCell * 2]; + if( nCell>iCell ){ + memmove(&pCellptr[2], pCellptr, (nCell - iCell) * 2); + } + nCell++; + cachedCellSize(pCArray, iCell+iNew); + if( pageInsertArray( + pPg, pBegin, &pData, pCellptr, + iCell+iNew, 1, pCArray + ) ) goto editpage_fail; + } + } + + /* Append cells to the end of the page */ + assert( nCell>=0 ); + pCellptr = &pPg->aCellIdx[nCell*2]; + if( pageInsertArray( + pPg, pBegin, &pData, pCellptr, + iNew+nCell, nNew-nCell, pCArray + ) ) goto editpage_fail; + + pPg->nCell = nNew; + pPg->nOverflow = 0; + + put2byte(&aData[hdr+3], pPg->nCell); + put2byte(&aData[hdr+5], pData - aData); + +#ifdef SQLITE_DEBUG + for(i=0; iapCell[i+iNew]; + int iOff = get2byteAligned(&pPg->aCellIdx[i*2]); + if( SQLITE_WITHIN(pCell, aData, &aData[pPg->pBt->usableSize]) ){ + pCell = &pTmp[pCell - aData]; + } + assert( 0==memcmp(pCell, &aData[iOff], + pCArray->pRef->xCellSize(pCArray->pRef, pCArray->apCell[i+iNew])) ); + } +#endif + + return SQLITE_OK; + editpage_fail: + /* Unable to edit this page. Rebuild it from scratch instead. */ + populateCellCache(pCArray, iNew, nNew); + return rebuildPage(pCArray, iNew, nNew, pPg); +} + + +#ifndef SQLITE_OMIT_QUICKBALANCE +/* +** This version of balance() handles the common special case where +** a new entry is being inserted on the extreme right-end of the +** tree, in other words, when the new entry will become the largest +** entry in the tree. +** +** Instead of trying to balance the 3 right-most leaf pages, just add +** a new page to the right-hand side and put the one new entry in +** that page. This leaves the right side of the tree somewhat +** unbalanced. But odds are that we will be inserting new entries +** at the end soon afterwards so the nearly empty page will quickly +** fill up. On average. +** +** pPage is the leaf page which is the right-most page in the tree. +** pParent is its parent. pPage must have a single overflow entry +** which is also the right-most entry on the page. +** +** The pSpace buffer is used to store a temporary copy of the divider +** cell that will be inserted into pParent. Such a cell consists of a 4 +** byte page number followed by a variable length integer. In other +** words, at most 13 bytes. Hence the pSpace buffer must be at +** least 13 bytes in size. +*/ +static int balance_quick(MemPage *pParent, MemPage *pPage, u8 *pSpace){ + BtShared *const pBt = pPage->pBt; /* B-Tree Database */ + MemPage *pNew; /* Newly allocated page */ + int rc; /* Return Code */ + Pgno pgnoNew; /* Page number of pNew */ + + assert( sqlite3_mutex_held(pPage->pBt->mutex) ); + assert( sqlite3PagerIswriteable(pParent->pDbPage) ); + assert( pPage->nOverflow==1 ); + + if( pPage->nCell==0 ) return SQLITE_CORRUPT_BKPT; /* dbfuzz001.test */ + assert( pPage->nFree>=0 ); + assert( pParent->nFree>=0 ); + + /* Allocate a new page. This page will become the right-sibling of + ** pPage. Make the parent page writable, so that the new divider cell + ** may be inserted. If both these operations are successful, proceed. + */ + rc = allocateBtreePage(pBt, &pNew, &pgnoNew, 0, 0); + + if( rc==SQLITE_OK ){ + + u8 *pOut = &pSpace[4]; + u8 *pCell = pPage->apOvfl[0]; + u16 szCell = pPage->xCellSize(pPage, pCell); + u8 *pStop; + CellArray b; + + assert( sqlite3PagerIswriteable(pNew->pDbPage) ); + assert( CORRUPT_DB || pPage->aData[0]==(PTF_INTKEY|PTF_LEAFDATA|PTF_LEAF) ); + zeroPage(pNew, PTF_INTKEY|PTF_LEAFDATA|PTF_LEAF); + b.nCell = 1; + b.pRef = pPage; + b.apCell = &pCell; + b.szCell = &szCell; + b.apEnd[0] = pPage->aDataEnd; + b.ixNx[0] = 2; + rc = rebuildPage(&b, 0, 1, pNew); + if( NEVER(rc) ){ + releasePage(pNew); + return rc; + } + pNew->nFree = pBt->usableSize - pNew->cellOffset - 2 - szCell; + + /* If this is an auto-vacuum database, update the pointer map + ** with entries for the new page, and any pointer from the + ** cell on the page to an overflow page. If either of these + ** operations fails, the return code is set, but the contents + ** of the parent page are still manipulated by thh code below. + ** That is Ok, at this point the parent page is guaranteed to + ** be marked as dirty. Returning an error code will cause a + ** rollback, undoing any changes made to the parent page. + */ + if( ISAUTOVACUUM ){ + ptrmapPut(pBt, pgnoNew, PTRMAP_BTREE, pParent->pgno, &rc); + if( szCell>pNew->minLocal ){ + ptrmapPutOvflPtr(pNew, pNew, pCell, &rc); + } + } + + /* Create a divider cell to insert into pParent. The divider cell + ** consists of a 4-byte page number (the page number of pPage) and + ** a variable length key value (which must be the same value as the + ** largest key on pPage). + ** + ** To find the largest key value on pPage, first find the right-most + ** cell on pPage. The first two fields of this cell are the + ** record-length (a variable length integer at most 32-bits in size) + ** and the key value (a variable length integer, may have any value). + ** The first of the while(...) loops below skips over the record-length + ** field. The second while(...) loop copies the key value from the + ** cell on pPage into the pSpace buffer. + */ + pCell = findCell(pPage, pPage->nCell-1); + pStop = &pCell[9]; + while( (*(pCell++)&0x80) && pCellnCell, pSpace, (int)(pOut-pSpace), + 0, pPage->pgno, &rc); + } + + /* Set the right-child pointer of pParent to point to the new page. */ + put4byte(&pParent->aData[pParent->hdrOffset+8], pgnoNew); + + /* Release the reference to the new page. */ + releasePage(pNew); + } + + return rc; +} +#endif /* SQLITE_OMIT_QUICKBALANCE */ + +#if 0 +/* +** This function does not contribute anything to the operation of SQLite. +** it is sometimes activated temporarily while debugging code responsible +** for setting pointer-map entries. +*/ +static int ptrmapCheckPages(MemPage **apPage, int nPage){ + int i, j; + for(i=0; ipBt; + assert( pPage->isInit ); + + for(j=0; jnCell; j++){ + CellInfo info; + u8 *z; + + z = findCell(pPage, j); + pPage->xParseCell(pPage, z, &info); + if( info.nLocalpgno && e==PTRMAP_OVERFLOW1 ); + } + if( !pPage->leaf ){ + Pgno child = get4byte(z); + ptrmapGet(pBt, child, &e, &n); + assert( n==pPage->pgno && e==PTRMAP_BTREE ); + } + } + if( !pPage->leaf ){ + Pgno child = get4byte(&pPage->aData[pPage->hdrOffset+8]); + ptrmapGet(pBt, child, &e, &n); + assert( n==pPage->pgno && e==PTRMAP_BTREE ); + } + } + return 1; +} +#endif + +/* +** This function is used to copy the contents of the b-tree node stored +** on page pFrom to page pTo. If page pFrom was not a leaf page, then +** the pointer-map entries for each child page are updated so that the +** parent page stored in the pointer map is page pTo. If pFrom contained +** any cells with overflow page pointers, then the corresponding pointer +** map entries are also updated so that the parent page is page pTo. +** +** If pFrom is currently carrying any overflow cells (entries in the +** MemPage.apOvfl[] array), they are not copied to pTo. +** +** Before returning, page pTo is reinitialized using btreeInitPage(). +** +** The performance of this function is not critical. It is only used by +** the balance_shallower() and balance_deeper() procedures, neither of +** which are called often under normal circumstances. +*/ +static void copyNodeContent(MemPage *pFrom, MemPage *pTo, int *pRC){ + if( (*pRC)==SQLITE_OK ){ + BtShared * const pBt = pFrom->pBt; + u8 * const aFrom = pFrom->aData; + u8 * const aTo = pTo->aData; + int const iFromHdr = pFrom->hdrOffset; + int const iToHdr = ((pTo->pgno==1) ? 100 : 0); + int rc; + int iData; + + + assert( pFrom->isInit ); + assert( pFrom->nFree>=iToHdr ); + assert( get2byte(&aFrom[iFromHdr+5]) <= (int)pBt->usableSize ); + + /* Copy the b-tree node content from page pFrom to page pTo. */ + iData = get2byte(&aFrom[iFromHdr+5]); + memcpy(&aTo[iData], &aFrom[iData], pBt->usableSize-iData); + memcpy(&aTo[iToHdr], &aFrom[iFromHdr], pFrom->cellOffset + 2*pFrom->nCell); + + /* Reinitialize page pTo so that the contents of the MemPage structure + ** match the new data. The initialization of pTo can actually fail under + ** fairly obscure circumstances, even though it is a copy of initialized + ** page pFrom. + */ + pTo->isInit = 0; + rc = btreeInitPage(pTo); + if( rc==SQLITE_OK ) rc = btreeComputeFreeSpace(pTo); + if( rc!=SQLITE_OK ){ + *pRC = rc; + return; + } + + /* If this is an auto-vacuum database, update the pointer-map entries + ** for any b-tree or overflow pages that pTo now contains the pointers to. + */ + if( ISAUTOVACUUM ){ + *pRC = setChildPtrmaps(pTo); + } + } +} + +/* +** This routine redistributes cells on the iParentIdx'th child of pParent +** (hereafter "the page") and up to 2 siblings so that all pages have about the +** same amount of free space. Usually a single sibling on either side of the +** page are used in the balancing, though both siblings might come from one +** side if the page is the first or last child of its parent. If the page +** has fewer than 2 siblings (something which can only happen if the page +** is a root page or a child of a root page) then all available siblings +** participate in the balancing. +** +** The number of siblings of the page might be increased or decreased by +** one or two in an effort to keep pages nearly full but not over full. +** +** Note that when this routine is called, some of the cells on the page +** might not actually be stored in MemPage.aData[]. This can happen +** if the page is overfull. This routine ensures that all cells allocated +** to the page and its siblings fit into MemPage.aData[] before returning. +** +** In the course of balancing the page and its siblings, cells may be +** inserted into or removed from the parent page (pParent). Doing so +** may cause the parent page to become overfull or underfull. If this +** happens, it is the responsibility of the caller to invoke the correct +** balancing routine to fix this problem (see the balance() routine). +** +** If this routine fails for any reason, it might leave the database +** in a corrupted state. So if this routine fails, the database should +** be rolled back. +** +** The third argument to this function, aOvflSpace, is a pointer to a +** buffer big enough to hold one page. If while inserting cells into the parent +** page (pParent) the parent page becomes overfull, this buffer is +** used to store the parent's overflow cells. Because this function inserts +** a maximum of four divider cells into the parent page, and the maximum +** size of a cell stored within an internal node is always less than 1/4 +** of the page-size, the aOvflSpace[] buffer is guaranteed to be large +** enough for all overflow cells. +** +** If aOvflSpace is set to a null pointer, this function returns +** SQLITE_NOMEM. +*/ +static int balance_nonroot( + MemPage *pParent, /* Parent page of siblings being balanced */ + int iParentIdx, /* Index of "the page" in pParent */ + u8 *aOvflSpace, /* page-size bytes of space for parent ovfl */ + int isRoot, /* True if pParent is a root-page */ + int bBulk /* True if this call is part of a bulk load */ +){ + BtShared *pBt; /* The whole database */ + int nMaxCells = 0; /* Allocated size of apCell, szCell, aFrom. */ + int nNew = 0; /* Number of pages in apNew[] */ + int nOld; /* Number of pages in apOld[] */ + int i, j, k; /* Loop counters */ + int nxDiv; /* Next divider slot in pParent->aCell[] */ + int rc = SQLITE_OK; /* The return code */ + u16 leafCorrection; /* 4 if pPage is a leaf. 0 if not */ + int leafData; /* True if pPage is a leaf of a LEAFDATA tree */ + int usableSpace; /* Bytes in pPage beyond the header */ + int pageFlags; /* Value of pPage->aData[0] */ + int iSpace1 = 0; /* First unused byte of aSpace1[] */ + int iOvflSpace = 0; /* First unused byte of aOvflSpace[] */ + int szScratch; /* Size of scratch memory requested */ + MemPage *apOld[NB]; /* pPage and up to two siblings */ + MemPage *apNew[NB+2]; /* pPage and up to NB siblings after balancing */ + u8 *pRight; /* Location in parent of right-sibling pointer */ + u8 *apDiv[NB-1]; /* Divider cells in pParent */ + int cntNew[NB+2]; /* Index in b.paCell[] of cell after i-th page */ + int cntOld[NB+2]; /* Old index in b.apCell[] */ + int szNew[NB+2]; /* Combined size of cells placed on i-th page */ + u8 *aSpace1; /* Space for copies of dividers cells */ + Pgno pgno; /* Temp var to store a page number in */ + u8 abDone[NB+2]; /* True after i'th new page is populated */ + Pgno aPgno[NB+2]; /* Page numbers of new pages before shuffling */ + Pgno aPgOrder[NB+2]; /* Copy of aPgno[] used for sorting pages */ + u16 aPgFlags[NB+2]; /* flags field of new pages before shuffling */ + CellArray b; /* Parsed information on cells being balanced */ + + memset(abDone, 0, sizeof(abDone)); + b.nCell = 0; + b.apCell = 0; + pBt = pParent->pBt; + assert( sqlite3_mutex_held(pBt->mutex) ); + assert( sqlite3PagerIswriteable(pParent->pDbPage) ); + + /* At this point pParent may have at most one overflow cell. And if + ** this overflow cell is present, it must be the cell with + ** index iParentIdx. This scenario comes about when this function + ** is called (indirectly) from sqlite3BtreeDelete(). + */ + assert( pParent->nOverflow==0 || pParent->nOverflow==1 ); + assert( pParent->nOverflow==0 || pParent->aiOvfl[0]==iParentIdx ); + + if( !aOvflSpace ){ + return SQLITE_NOMEM_BKPT; + } + assert( pParent->nFree>=0 ); + + /* Find the sibling pages to balance. Also locate the cells in pParent + ** that divide the siblings. An attempt is made to find NN siblings on + ** either side of pPage. More siblings are taken from one side, however, + ** if there are fewer than NN siblings on the other side. If pParent + ** has NB or fewer children then all children of pParent are taken. + ** + ** This loop also drops the divider cells from the parent page. This + ** way, the remainder of the function does not have to deal with any + ** overflow cells in the parent page, since if any existed they will + ** have already been removed. + */ + i = pParent->nOverflow + pParent->nCell; + if( i<2 ){ + nxDiv = 0; + }else{ + assert( bBulk==0 || bBulk==1 ); + if( iParentIdx==0 ){ + nxDiv = 0; + }else if( iParentIdx==i ){ + nxDiv = i-2+bBulk; + }else{ + nxDiv = iParentIdx-1; + } + i = 2-bBulk; + } + nOld = i+1; + if( (i+nxDiv-pParent->nOverflow)==pParent->nCell ){ + pRight = &pParent->aData[pParent->hdrOffset+8]; + }else{ + pRight = findCell(pParent, i+nxDiv-pParent->nOverflow); + } + pgno = get4byte(pRight); + while( 1 ){ + rc = getAndInitPage(pBt, pgno, &apOld[i], 0, 0); + if( rc ){ + memset(apOld, 0, (i+1)*sizeof(MemPage*)); + goto balance_cleanup; + } + if( apOld[i]->nFree<0 ){ + rc = btreeComputeFreeSpace(apOld[i]); + if( rc ){ + memset(apOld, 0, (i)*sizeof(MemPage*)); + goto balance_cleanup; + } + } + if( (i--)==0 ) break; + + if( pParent->nOverflow && i+nxDiv==pParent->aiOvfl[0] ){ + apDiv[i] = pParent->apOvfl[0]; + pgno = get4byte(apDiv[i]); + szNew[i] = pParent->xCellSize(pParent, apDiv[i]); + pParent->nOverflow = 0; + }else{ + apDiv[i] = findCell(pParent, i+nxDiv-pParent->nOverflow); + pgno = get4byte(apDiv[i]); + szNew[i] = pParent->xCellSize(pParent, apDiv[i]); + + /* Drop the cell from the parent page. apDiv[i] still points to + ** the cell within the parent, even though it has been dropped. + ** This is safe because dropping a cell only overwrites the first + ** four bytes of it, and this function does not need the first + ** four bytes of the divider cell. So the pointer is safe to use + ** later on. + ** + ** But not if we are in secure-delete mode. In secure-delete mode, + ** the dropCell() routine will overwrite the entire cell with zeroes. + ** In this case, temporarily copy the cell into the aOvflSpace[] + ** buffer. It will be copied out again as soon as the aSpace[] buffer + ** is allocated. */ + if( pBt->btsFlags & BTS_FAST_SECURE ){ + int iOff; + + iOff = SQLITE_PTR_TO_INT(apDiv[i]) - SQLITE_PTR_TO_INT(pParent->aData); + if( (iOff+szNew[i])>(int)pBt->usableSize ){ + rc = SQLITE_CORRUPT_BKPT; + memset(apOld, 0, (i+1)*sizeof(MemPage*)); + goto balance_cleanup; + }else{ + memcpy(&aOvflSpace[iOff], apDiv[i], szNew[i]); + apDiv[i] = &aOvflSpace[apDiv[i]-pParent->aData]; + } + } + dropCell(pParent, i+nxDiv-pParent->nOverflow, szNew[i], &rc); + } + } + + /* Make nMaxCells a multiple of 4 in order to preserve 8-byte + ** alignment */ + nMaxCells = nOld*(MX_CELL(pBt) + ArraySize(pParent->apOvfl)); + nMaxCells = (nMaxCells + 3)&~3; + + /* + ** Allocate space for memory structures + */ + szScratch = + nMaxCells*sizeof(u8*) /* b.apCell */ + + nMaxCells*sizeof(u16) /* b.szCell */ + + pBt->pageSize; /* aSpace1 */ + + assert( szScratch<=7*(int)pBt->pageSize ); + b.apCell = sqlite3StackAllocRaw(0, szScratch ); + if( b.apCell==0 ){ + rc = SQLITE_NOMEM_BKPT; + goto balance_cleanup; + } + b.szCell = (u16*)&b.apCell[nMaxCells]; + aSpace1 = (u8*)&b.szCell[nMaxCells]; + assert( EIGHT_BYTE_ALIGNMENT(aSpace1) ); + + /* + ** Load pointers to all cells on sibling pages and the divider cells + ** into the local b.apCell[] array. Make copies of the divider cells + ** into space obtained from aSpace1[]. The divider cells have already + ** been removed from pParent. + ** + ** If the siblings are on leaf pages, then the child pointers of the + ** divider cells are stripped from the cells before they are copied + ** into aSpace1[]. In this way, all cells in b.apCell[] are without + ** child pointers. If siblings are not leaves, then all cell in + ** b.apCell[] include child pointers. Either way, all cells in b.apCell[] + ** are alike. + ** + ** leafCorrection: 4 if pPage is a leaf. 0 if pPage is not a leaf. + ** leafData: 1 if pPage holds key+data and pParent holds only keys. + */ + b.pRef = apOld[0]; + leafCorrection = b.pRef->leaf*4; + leafData = b.pRef->intKeyLeaf; + for(i=0; inCell; + u8 *aData = pOld->aData; + u16 maskPage = pOld->maskPage; + u8 *piCell = aData + pOld->cellOffset; + u8 *piEnd; + VVA_ONLY( int nCellAtStart = b.nCell; ) + + /* Verify that all sibling pages are of the same "type" (table-leaf, + ** table-interior, index-leaf, or index-interior). + */ + if( pOld->aData[0]!=apOld[0]->aData[0] ){ + rc = SQLITE_CORRUPT_BKPT; + goto balance_cleanup; + } + + /* Load b.apCell[] with pointers to all cells in pOld. If pOld + ** contains overflow cells, include them in the b.apCell[] array + ** in the correct spot. + ** + ** Note that when there are multiple overflow cells, it is always the + ** case that they are sequential and adjacent. This invariant arises + ** because multiple overflows can only occurs when inserting divider + ** cells into a parent on a prior balance, and divider cells are always + ** adjacent and are inserted in order. There is an assert() tagged + ** with "NOTE 1" in the overflow cell insertion loop to prove this + ** invariant. + ** + ** This must be done in advance. Once the balance starts, the cell + ** offset section of the btree page will be overwritten and we will no + ** long be able to find the cells if a pointer to each cell is not saved + ** first. + */ + memset(&b.szCell[b.nCell], 0, sizeof(b.szCell[0])*(limit+pOld->nOverflow)); + if( pOld->nOverflow>0 ){ + if( NEVER(limitaiOvfl[0]) ){ + rc = SQLITE_CORRUPT_BKPT; + goto balance_cleanup; + } + limit = pOld->aiOvfl[0]; + for(j=0; jnOverflow; k++){ + assert( k==0 || pOld->aiOvfl[k-1]+1==pOld->aiOvfl[k] );/* NOTE 1 */ + b.apCell[b.nCell] = pOld->apOvfl[k]; + b.nCell++; + } + } + piEnd = aData + pOld->cellOffset + 2*pOld->nCell; + while( piCellnCell+pOld->nOverflow) ); + + cntOld[i] = b.nCell; + if( imaxLocal+23 ); + assert( iSpace1 <= (int)pBt->pageSize ); + memcpy(pTemp, apDiv[i], sz); + b.apCell[b.nCell] = pTemp+leafCorrection; + assert( leafCorrection==0 || leafCorrection==4 ); + b.szCell[b.nCell] = b.szCell[b.nCell] - leafCorrection; + if( !pOld->leaf ){ + assert( leafCorrection==0 ); + assert( pOld->hdrOffset==0 ); + /* The right pointer of the child page pOld becomes the left + ** pointer of the divider cell */ + memcpy(b.apCell[b.nCell], &pOld->aData[8], 4); + }else{ + assert( leafCorrection==4 ); + while( b.szCell[b.nCell]<4 ){ + /* Do not allow any cells smaller than 4 bytes. If a smaller cell + ** does exist, pad it with 0x00 bytes. */ + assert( b.szCell[b.nCell]==3 || CORRUPT_DB ); + assert( b.apCell[b.nCell]==&aSpace1[iSpace1-3] || CORRUPT_DB ); + aSpace1[iSpace1++] = 0x00; + b.szCell[b.nCell]++; + } + } + b.nCell++; + } + } + + /* + ** Figure out the number of pages needed to hold all b.nCell cells. + ** Store this number in "k". Also compute szNew[] which is the total + ** size of all cells on the i-th page and cntNew[] which is the index + ** in b.apCell[] of the cell that divides page i from page i+1. + ** cntNew[k] should equal b.nCell. + ** + ** Values computed by this block: + ** + ** k: The total number of sibling pages + ** szNew[i]: Spaced used on the i-th sibling page. + ** cntNew[i]: Index in b.apCell[] and b.szCell[] for the first cell to + ** the right of the i-th sibling page. + ** usableSpace: Number of bytes of space available on each sibling. + ** + */ + usableSpace = pBt->usableSize - 12 + leafCorrection; + for(i=k=0; iaDataEnd; + b.ixNx[k] = cntOld[i]; + if( k && b.ixNx[k]==b.ixNx[k-1] ){ + k--; /* Omit b.ixNx[] entry for child pages with no cells */ + } + if( !leafData ){ + k++; + b.apEnd[k] = pParent->aDataEnd; + b.ixNx[k] = cntOld[i]+1; + } + assert( p->nFree>=0 ); + szNew[i] = usableSpace - p->nFree; + for(j=0; jnOverflow; j++){ + szNew[i] += 2 + p->xCellSize(p, p->apOvfl[j]); + } + cntNew[i] = cntOld[i]; + } + k = nOld; + for(i=0; iusableSpace ){ + if( i+1>=k ){ + k = i+2; + if( k>NB+2 ){ rc = SQLITE_CORRUPT_BKPT; goto balance_cleanup; } + szNew[k-1] = 0; + cntNew[k-1] = b.nCell; + } + sz = 2 + cachedCellSize(&b, cntNew[i]-1); + szNew[i] -= sz; + if( !leafData ){ + if( cntNew[i]usableSpace ) break; + szNew[i] += sz; + cntNew[i]++; + if( !leafData ){ + if( cntNew[i]=b.nCell ){ + k = i+1; + }else if( cntNew[i] <= (i>0 ? cntNew[i-1] : 0) ){ + rc = SQLITE_CORRUPT_BKPT; + goto balance_cleanup; + } + } + + /* + ** The packing computed by the previous block is biased toward the siblings + ** on the left side (siblings with smaller keys). The left siblings are + ** always nearly full, while the right-most sibling might be nearly empty. + ** The next block of code attempts to adjust the packing of siblings to + ** get a better balance. + ** + ** This adjustment is more than an optimization. The packing above might + ** be so out of balance as to be illegal. For example, the right-most + ** sibling might be completely empty. This adjustment is not optional. + */ + for(i=k-1; i>0; i--){ + int szRight = szNew[i]; /* Size of sibling on the right */ + int szLeft = szNew[i-1]; /* Size of sibling on the left */ + int r; /* Index of right-most cell in left sibling */ + int d; /* Index of first cell to the left of right sibling */ + + r = cntNew[i-1] - 1; + d = r + 1 - leafData; + (void)cachedCellSize(&b, d); + do{ + assert( d szLeft-(b.szCell[r]+(i==k-1?0:2)))){ + break; + } + szRight += b.szCell[d] + 2; + szLeft -= b.szCell[r] + 2; + cntNew[i-1] = r; + r--; + d--; + }while( r>=0 ); + szNew[i] = szRight; + szNew[i-1] = szLeft; + if( cntNew[i-1] <= (i>1 ? cntNew[i-2] : 0) ){ + rc = SQLITE_CORRUPT_BKPT; + goto balance_cleanup; + } + } + + /* Sanity check: For a non-corrupt database file one of the follwing + ** must be true: + ** (1) We found one or more cells (cntNew[0])>0), or + ** (2) pPage is a virtual root page. A virtual root page is when + ** the real root page is page 1 and we are the only child of + ** that page. + */ + assert( cntNew[0]>0 || (pParent->pgno==1 && pParent->nCell==0) || CORRUPT_DB); + TRACE(("BALANCE: old: %d(nc=%d) %d(nc=%d) %d(nc=%d)\n", + apOld[0]->pgno, apOld[0]->nCell, + nOld>=2 ? apOld[1]->pgno : 0, nOld>=2 ? apOld[1]->nCell : 0, + nOld>=3 ? apOld[2]->pgno : 0, nOld>=3 ? apOld[2]->nCell : 0 + )); + + /* + ** Allocate k new pages. Reuse old pages where possible. + */ + pageFlags = apOld[0]->aData[0]; + for(i=0; ipDbPage); + nNew++; + if( rc ) goto balance_cleanup; + }else{ + assert( i>0 ); + rc = allocateBtreePage(pBt, &pNew, &pgno, (bBulk ? 1 : pgno), 0); + if( rc ) goto balance_cleanup; + zeroPage(pNew, pageFlags); + apNew[i] = pNew; + nNew++; + cntOld[i] = b.nCell; + + /* Set the pointer-map entry for the new sibling page. */ + if( ISAUTOVACUUM ){ + ptrmapPut(pBt, pNew->pgno, PTRMAP_BTREE, pParent->pgno, &rc); + if( rc!=SQLITE_OK ){ + goto balance_cleanup; + } + } + } + } + + /* + ** Reassign page numbers so that the new pages are in ascending order. + ** This helps to keep entries in the disk file in order so that a scan + ** of the table is closer to a linear scan through the file. That in turn + ** helps the operating system to deliver pages from the disk more rapidly. + ** + ** An O(n^2) insertion sort algorithm is used, but since n is never more + ** than (NB+2) (a small constant), that should not be a problem. + ** + ** When NB==3, this one optimization makes the database about 25% faster + ** for large insertions and deletions. + */ + for(i=0; ipgno; + aPgFlags[i] = apNew[i]->pDbPage->flags; + for(j=0; ji ){ + sqlite3PagerRekey(apNew[iBest]->pDbPage, pBt->nPage+iBest+1, 0); + } + sqlite3PagerRekey(apNew[i]->pDbPage, pgno, aPgFlags[iBest]); + apNew[i]->pgno = pgno; + } + } + + TRACE(("BALANCE: new: %d(%d nc=%d) %d(%d nc=%d) %d(%d nc=%d) " + "%d(%d nc=%d) %d(%d nc=%d)\n", + apNew[0]->pgno, szNew[0], cntNew[0], + nNew>=2 ? apNew[1]->pgno : 0, nNew>=2 ? szNew[1] : 0, + nNew>=2 ? cntNew[1] - cntNew[0] - !leafData : 0, + nNew>=3 ? apNew[2]->pgno : 0, nNew>=3 ? szNew[2] : 0, + nNew>=3 ? cntNew[2] - cntNew[1] - !leafData : 0, + nNew>=4 ? apNew[3]->pgno : 0, nNew>=4 ? szNew[3] : 0, + nNew>=4 ? cntNew[3] - cntNew[2] - !leafData : 0, + nNew>=5 ? apNew[4]->pgno : 0, nNew>=5 ? szNew[4] : 0, + nNew>=5 ? cntNew[4] - cntNew[3] - !leafData : 0 + )); + + assert( sqlite3PagerIswriteable(pParent->pDbPage) ); + assert( nNew>=1 && nNew<=ArraySize(apNew) ); + assert( apNew[nNew-1]!=0 ); + put4byte(pRight, apNew[nNew-1]->pgno); + + /* If the sibling pages are not leaves, ensure that the right-child pointer + ** of the right-most new sibling page is set to the value that was + ** originally in the same field of the right-most old sibling page. */ + if( (pageFlags & PTF_LEAF)==0 && nOld!=nNew ){ + MemPage *pOld = (nNew>nOld ? apNew : apOld)[nOld-1]; + memcpy(&apNew[nNew-1]->aData[8], &pOld->aData[8], 4); + } + + /* Make any required updates to pointer map entries associated with + ** cells stored on sibling pages following the balance operation. Pointer + ** map entries associated with divider cells are set by the insertCell() + ** routine. The associated pointer map entries are: + ** + ** a) if the cell contains a reference to an overflow chain, the + ** entry associated with the first page in the overflow chain, and + ** + ** b) if the sibling pages are not leaves, the child page associated + ** with the cell. + ** + ** If the sibling pages are not leaves, then the pointer map entry + ** associated with the right-child of each sibling may also need to be + ** updated. This happens below, after the sibling pages have been + ** populated, not here. + */ + if( ISAUTOVACUUM ){ + MemPage *pOld; + MemPage *pNew = pOld = apNew[0]; + int cntOldNext = pNew->nCell + pNew->nOverflow; + int iNew = 0; + int iOld = 0; + + for(i=0; i=0 && iOldnCell + pOld->nOverflow + !leafData; + } + if( i==cntNew[iNew] ){ + pNew = apNew[++iNew]; + if( !leafData ) continue; + } + + /* Cell pCell is destined for new sibling page pNew. Originally, it + ** was either part of sibling page iOld (possibly an overflow cell), + ** or else the divider cell to the left of sibling page iOld. So, + ** if sibling page iOld had the same page number as pNew, and if + ** pCell really was a part of sibling page iOld (not a divider or + ** overflow cell), we can skip updating the pointer map entries. */ + if( iOld>=nNew + || pNew->pgno!=aPgno[iOld] + || !SQLITE_WITHIN(pCell,pOld->aData,pOld->aDataEnd) + ){ + if( !leafCorrection ){ + ptrmapPut(pBt, get4byte(pCell), PTRMAP_BTREE, pNew->pgno, &rc); + } + if( cachedCellSize(&b,i)>pNew->minLocal ){ + ptrmapPutOvflPtr(pNew, pOld, pCell, &rc); + } + if( rc ) goto balance_cleanup; + } + } + } + + /* Insert new divider cells into pParent. */ + for(i=0; ileaf ){ + memcpy(&pNew->aData[8], pCell, 4); + }else if( leafData ){ + /* If the tree is a leaf-data tree, and the siblings are leaves, + ** then there is no divider cell in b.apCell[]. Instead, the divider + ** cell consists of the integer key for the right-most cell of + ** the sibling-page assembled above only. + */ + CellInfo info; + j--; + pNew->xParseCell(pNew, b.apCell[j], &info); + pCell = pTemp; + sz = 4 + putVarint(&pCell[4], info.nKey); + pTemp = 0; + }else{ + pCell -= 4; + /* Obscure case for non-leaf-data trees: If the cell at pCell was + ** previously stored on a leaf node, and its reported size was 4 + ** bytes, then it may actually be smaller than this + ** (see btreeParseCellPtr(), 4 bytes is the minimum size of + ** any cell). But it is important to pass the correct size to + ** insertCell(), so reparse the cell now. + ** + ** This can only happen for b-trees used to evaluate "IN (SELECT ...)" + ** and WITHOUT ROWID tables with exactly one column which is the + ** primary key. + */ + if( b.szCell[j]==4 ){ + assert(leafCorrection==4); + sz = pParent->xCellSize(pParent, pCell); + } + } + iOvflSpace += sz; + assert( sz<=pBt->maxLocal+23 ); + assert( iOvflSpace <= (int)pBt->pageSize ); + insertCell(pParent, nxDiv+i, pCell, sz, pTemp, pNew->pgno, &rc); + if( rc!=SQLITE_OK ) goto balance_cleanup; + assert( sqlite3PagerIswriteable(pParent->pDbPage) ); + } + + /* Now update the actual sibling pages. The order in which they are updated + ** is important, as this code needs to avoid disrupting any page from which + ** cells may still to be read. In practice, this means: + ** + ** (1) If cells are moving left (from apNew[iPg] to apNew[iPg-1]) + ** then it is not safe to update page apNew[iPg] until after + ** the left-hand sibling apNew[iPg-1] has been updated. + ** + ** (2) If cells are moving right (from apNew[iPg] to apNew[iPg+1]) + ** then it is not safe to update page apNew[iPg] until after + ** the right-hand sibling apNew[iPg+1] has been updated. + ** + ** If neither of the above apply, the page is safe to update. + ** + ** The iPg value in the following loop starts at nNew-1 goes down + ** to 0, then back up to nNew-1 again, thus making two passes over + ** the pages. On the initial downward pass, only condition (1) above + ** needs to be tested because (2) will always be true from the previous + ** step. On the upward pass, both conditions are always true, so the + ** upwards pass simply processes pages that were missed on the downward + ** pass. + */ + for(i=1-nNew; i=0 && iPg=0 /* On the upwards pass, or... */ + || cntOld[iPg-1]>=cntNew[iPg-1] /* Condition (1) is true */ + ){ + int iNew; + int iOld; + int nNewCell; + + /* Verify condition (1): If cells are moving left, update iPg + ** only after iPg-1 has already been updated. */ + assert( iPg==0 || cntOld[iPg-1]>=cntNew[iPg-1] || abDone[iPg-1] ); + + /* Verify condition (2): If cells are moving right, update iPg + ** only after iPg+1 has already been updated. */ + assert( cntNew[iPg]>=cntOld[iPg] || abDone[iPg+1] ); + + if( iPg==0 ){ + iNew = iOld = 0; + nNewCell = cntNew[0]; + }else{ + iOld = iPgnFree = usableSpace-szNew[iPg]; + assert( apNew[iPg]->nOverflow==0 ); + assert( apNew[iPg]->nCell==nNewCell ); + } + } + + /* All pages have been processed exactly once */ + assert( memcmp(abDone, "\01\01\01\01\01", nNew)==0 ); + + assert( nOld>0 ); + assert( nNew>0 ); + + if( isRoot && pParent->nCell==0 && pParent->hdrOffset<=apNew[0]->nFree ){ + /* The root page of the b-tree now contains no cells. The only sibling + ** page is the right-child of the parent. Copy the contents of the + ** child page into the parent, decreasing the overall height of the + ** b-tree structure by one. This is described as the "balance-shallower" + ** sub-algorithm in some documentation. + ** + ** If this is an auto-vacuum database, the call to copyNodeContent() + ** sets all pointer-map entries corresponding to database image pages + ** for which the pointer is stored within the content being copied. + ** + ** It is critical that the child page be defragmented before being + ** copied into the parent, because if the parent is page 1 then it will + ** by smaller than the child due to the database header, and so all the + ** free space needs to be up front. + */ + assert( nNew==1 || CORRUPT_DB ); + rc = defragmentPage(apNew[0], -1); + testcase( rc!=SQLITE_OK ); + assert( apNew[0]->nFree == + (get2byteNotZero(&apNew[0]->aData[5]) - apNew[0]->cellOffset + - apNew[0]->nCell*2) + || rc!=SQLITE_OK + ); + copyNodeContent(apNew[0], pParent, &rc); + freePage(apNew[0], &rc); + }else if( ISAUTOVACUUM && !leafCorrection ){ + /* Fix the pointer map entries associated with the right-child of each + ** sibling page. All other pointer map entries have already been taken + ** care of. */ + for(i=0; iaData[8]); + ptrmapPut(pBt, key, PTRMAP_BTREE, apNew[i]->pgno, &rc); + } + } + + assert( pParent->isInit ); + TRACE(("BALANCE: finished: old=%d new=%d cells=%d\n", + nOld, nNew, b.nCell)); + + /* Free any old pages that were not reused as new pages. + */ + for(i=nNew; iisInit ){ + /* The ptrmapCheckPages() contains assert() statements that verify that + ** all pointer map pages are set correctly. This is helpful while + ** debugging. This is usually disabled because a corrupt database may + ** cause an assert() statement to fail. */ + ptrmapCheckPages(apNew, nNew); + ptrmapCheckPages(&pParent, 1); + } +#endif + + /* + ** Cleanup before returning. + */ +balance_cleanup: + sqlite3StackFree(0, b.apCell); + for(i=0; ipBt; /* The BTree */ + + assert( pRoot->nOverflow>0 ); + assert( sqlite3_mutex_held(pBt->mutex) ); + + /* Make pRoot, the root page of the b-tree, writable. Allocate a new + ** page that will become the new right-child of pPage. Copy the contents + ** of the node stored on pRoot into the new child page. + */ + rc = sqlite3PagerWrite(pRoot->pDbPage); + if( rc==SQLITE_OK ){ + rc = allocateBtreePage(pBt,&pChild,&pgnoChild,pRoot->pgno,0); + copyNodeContent(pRoot, pChild, &rc); + if( ISAUTOVACUUM ){ + ptrmapPut(pBt, pgnoChild, PTRMAP_BTREE, pRoot->pgno, &rc); + } + } + if( rc ){ + *ppChild = 0; + releasePage(pChild); + return rc; + } + assert( sqlite3PagerIswriteable(pChild->pDbPage) ); + assert( sqlite3PagerIswriteable(pRoot->pDbPage) ); + assert( pChild->nCell==pRoot->nCell || CORRUPT_DB ); + + TRACE(("BALANCE: copy root %d into %d\n", pRoot->pgno, pChild->pgno)); + + /* Copy the overflow cells from pRoot to pChild */ + memcpy(pChild->aiOvfl, pRoot->aiOvfl, + pRoot->nOverflow*sizeof(pRoot->aiOvfl[0])); + memcpy(pChild->apOvfl, pRoot->apOvfl, + pRoot->nOverflow*sizeof(pRoot->apOvfl[0])); + pChild->nOverflow = pRoot->nOverflow; + + /* Zero the contents of pRoot. Then install pChild as the right-child. */ + zeroPage(pRoot, pChild->aData[0] & ~PTF_LEAF); + put4byte(&pRoot->aData[pRoot->hdrOffset+8], pgnoChild); + + *ppChild = pChild; + return SQLITE_OK; +} + +/* +** The page that pCur currently points to has just been modified in +** some way. This function figures out if this modification means the +** tree needs to be balanced, and if so calls the appropriate balancing +** routine. Balancing routines are: +** +** balance_quick() +** balance_deeper() +** balance_nonroot() +*/ +static int balance(BtCursor *pCur){ + int rc = SQLITE_OK; + const int nMin = pCur->pBt->usableSize * 2 / 3; + u8 aBalanceQuickSpace[13]; + u8 *pFree = 0; + + VVA_ONLY( int balance_quick_called = 0 ); + VVA_ONLY( int balance_deeper_called = 0 ); + + do { + int iPage; + MemPage *pPage = pCur->pPage; + + if( NEVER(pPage->nFree<0) && btreeComputeFreeSpace(pPage) ) break; + if( pPage->nOverflow==0 && pPage->nFree<=nMin ){ + break; + }else if( (iPage = pCur->iPage)==0 ){ + if( pPage->nOverflow ){ + /* The root page of the b-tree is overfull. In this case call the + ** balance_deeper() function to create a new child for the root-page + ** and copy the current contents of the root-page to it. The + ** next iteration of the do-loop will balance the child page. + */ + assert( balance_deeper_called==0 ); + VVA_ONLY( balance_deeper_called++ ); + rc = balance_deeper(pPage, &pCur->apPage[1]); + if( rc==SQLITE_OK ){ + pCur->iPage = 1; + pCur->ix = 0; + pCur->aiIdx[0] = 0; + pCur->apPage[0] = pPage; + pCur->pPage = pCur->apPage[1]; + assert( pCur->pPage->nOverflow ); + } + }else{ + break; + } + }else{ + MemPage * const pParent = pCur->apPage[iPage-1]; + int const iIdx = pCur->aiIdx[iPage-1]; + + rc = sqlite3PagerWrite(pParent->pDbPage); + if( rc==SQLITE_OK && pParent->nFree<0 ){ + rc = btreeComputeFreeSpace(pParent); + } + if( rc==SQLITE_OK ){ +#ifndef SQLITE_OMIT_QUICKBALANCE + if( pPage->intKeyLeaf + && pPage->nOverflow==1 + && pPage->aiOvfl[0]==pPage->nCell + && pParent->pgno!=1 + && pParent->nCell==iIdx + ){ + /* Call balance_quick() to create a new sibling of pPage on which + ** to store the overflow cell. balance_quick() inserts a new cell + ** into pParent, which may cause pParent overflow. If this + ** happens, the next iteration of the do-loop will balance pParent + ** use either balance_nonroot() or balance_deeper(). Until this + ** happens, the overflow cell is stored in the aBalanceQuickSpace[] + ** buffer. + ** + ** The purpose of the following assert() is to check that only a + ** single call to balance_quick() is made for each call to this + ** function. If this were not verified, a subtle bug involving reuse + ** of the aBalanceQuickSpace[] might sneak in. + */ + assert( balance_quick_called==0 ); + VVA_ONLY( balance_quick_called++ ); + rc = balance_quick(pParent, pPage, aBalanceQuickSpace); + }else +#endif + { + /* In this case, call balance_nonroot() to redistribute cells + ** between pPage and up to 2 of its sibling pages. This involves + ** modifying the contents of pParent, which may cause pParent to + ** become overfull or underfull. The next iteration of the do-loop + ** will balance the parent page to correct this. + ** + ** If the parent page becomes overfull, the overflow cell or cells + ** are stored in the pSpace buffer allocated immediately below. + ** A subsequent iteration of the do-loop will deal with this by + ** calling balance_nonroot() (balance_deeper() may be called first, + ** but it doesn't deal with overflow cells - just moves them to a + ** different page). Once this subsequent call to balance_nonroot() + ** has completed, it is safe to release the pSpace buffer used by + ** the previous call, as the overflow cell data will have been + ** copied either into the body of a database page or into the new + ** pSpace buffer passed to the latter call to balance_nonroot(). + */ + u8 *pSpace = sqlite3PageMalloc(pCur->pBt->pageSize); + rc = balance_nonroot(pParent, iIdx, pSpace, iPage==1, + pCur->hints&BTREE_BULKLOAD); + if( pFree ){ + /* If pFree is not NULL, it points to the pSpace buffer used + ** by a previous call to balance_nonroot(). Its contents are + ** now stored either on real database pages or within the + ** new pSpace buffer, so it may be safely freed here. */ + sqlite3PageFree(pFree); + } + + /* The pSpace buffer will be freed after the next call to + ** balance_nonroot(), or just before this function returns, whichever + ** comes first. */ + pFree = pSpace; + } + } + + pPage->nOverflow = 0; + + /* The next iteration of the do-loop balances the parent page. */ + releasePage(pPage); + pCur->iPage--; + assert( pCur->iPage>=0 ); + pCur->pPage = pCur->apPage[pCur->iPage]; + } + }while( rc==SQLITE_OK ); + + if( pFree ){ + sqlite3PageFree(pFree); + } + return rc; +} + +/* Overwrite content from pX into pDest. Only do the write if the +** content is different from what is already there. +*/ +static int btreeOverwriteContent( + MemPage *pPage, /* MemPage on which writing will occur */ + u8 *pDest, /* Pointer to the place to start writing */ + const BtreePayload *pX, /* Source of data to write */ + int iOffset, /* Offset of first byte to write */ + int iAmt /* Number of bytes to be written */ +){ + int nData = pX->nData - iOffset; + if( nData<=0 ){ + /* Overwritting with zeros */ + int i; + for(i=0; ipDbPage); + if( rc ) return rc; + memset(pDest + i, 0, iAmt - i); + } + }else{ + if( nDatapData) + iOffset, iAmt)!=0 ){ + int rc = sqlite3PagerWrite(pPage->pDbPage); + if( rc ) return rc; + /* In a corrupt database, it is possible for the source and destination + ** buffers to overlap. This is harmless since the database is already + ** corrupt but it does cause valgrind and ASAN warnings. So use + ** memmove(). */ + memmove(pDest, ((u8*)pX->pData) + iOffset, iAmt); + } + } + return SQLITE_OK; +} + +/* +** Overwrite the cell that cursor pCur is pointing to with fresh content +** contained in pX. +*/ +static int btreeOverwriteCell(BtCursor *pCur, const BtreePayload *pX){ + int iOffset; /* Next byte of pX->pData to write */ + int nTotal = pX->nData + pX->nZero; /* Total bytes of to write */ + int rc; /* Return code */ + MemPage *pPage = pCur->pPage; /* Page being written */ + BtShared *pBt; /* Btree */ + Pgno ovflPgno; /* Next overflow page to write */ + u32 ovflPageSize; /* Size to write on overflow page */ + + if( pCur->info.pPayload + pCur->info.nLocal > pPage->aDataEnd + || pCur->info.pPayload < pPage->aData + pPage->cellOffset + ){ + return SQLITE_CORRUPT_BKPT; + } + /* Overwrite the local portion first */ + rc = btreeOverwriteContent(pPage, pCur->info.pPayload, pX, + 0, pCur->info.nLocal); + if( rc ) return rc; + if( pCur->info.nLocal==nTotal ) return SQLITE_OK; + + /* Now overwrite the overflow pages */ + iOffset = pCur->info.nLocal; + assert( nTotal>=0 ); + assert( iOffset>=0 ); + ovflPgno = get4byte(pCur->info.pPayload + iOffset); + pBt = pPage->pBt; + ovflPageSize = pBt->usableSize - 4; + do{ + rc = btreeGetPage(pBt, ovflPgno, &pPage, 0); + if( rc ) return rc; + if( sqlite3PagerPageRefcount(pPage->pDbPage)!=1 ){ + rc = SQLITE_CORRUPT_BKPT; + }else{ + if( iOffset+ovflPageSize<(u32)nTotal ){ + ovflPgno = get4byte(pPage->aData); + }else{ + ovflPageSize = nTotal - iOffset; + } + rc = btreeOverwriteContent(pPage, pPage->aData+4, pX, + iOffset, ovflPageSize); + } + sqlite3PagerUnref(pPage->pDbPage); + if( rc ) return rc; + iOffset += ovflPageSize; + }while( iOffset0 then pCur points to a cell +** that is larger than (pKey,nKey). +** +** If seekResult==0, that means pCur is pointing at some unknown location. +** In that case, this routine must seek the cursor to the correct insertion +** point for (pKey,nKey) before doing the insertion. For index btrees, +** if pX->nMem is non-zero, then pX->aMem contains pointers to the unpacked +** key values and pX->aMem can be used instead of pX->pKey to avoid having +** to decode the key. +*/ +SQLITE_PRIVATE int sqlite3BtreeInsert( + BtCursor *pCur, /* Insert data into the table of this cursor */ + const BtreePayload *pX, /* Content of the row to be inserted */ + int flags, /* True if this is likely an append */ + int seekResult /* Result of prior MovetoUnpacked() call */ +){ + int rc; + int loc = seekResult; /* -1: before desired location +1: after */ + int szNew = 0; + int idx; + MemPage *pPage; + Btree *p = pCur->pBtree; + BtShared *pBt = p->pBt; + unsigned char *oldCell; + unsigned char *newCell = 0; + + assert( (flags & (BTREE_SAVEPOSITION|BTREE_APPEND))==flags ); + + if( pCur->eState==CURSOR_FAULT ){ + assert( pCur->skipNext!=SQLITE_OK ); + return pCur->skipNext; + } + + assert( cursorOwnsBtShared(pCur) ); + assert( (pCur->curFlags & BTCF_WriteFlag)!=0 + && pBt->inTransaction==TRANS_WRITE + && (pBt->btsFlags & BTS_READ_ONLY)==0 ); + assert( hasSharedCacheTableLock(p, pCur->pgnoRoot, pCur->pKeyInfo!=0, 2) ); + + /* Assert that the caller has been consistent. If this cursor was opened + ** expecting an index b-tree, then the caller should be inserting blob + ** keys with no associated data. If the cursor was opened expecting an + ** intkey table, the caller should be inserting integer keys with a + ** blob of associated data. */ + assert( (pX->pKey==0)==(pCur->pKeyInfo==0) ); + + /* Save the positions of any other cursors open on this table. + ** + ** In some cases, the call to btreeMoveto() below is a no-op. For + ** example, when inserting data into a table with auto-generated integer + ** keys, the VDBE layer invokes sqlite3BtreeLast() to figure out the + ** integer key to use. It then calls this function to actually insert the + ** data into the intkey B-Tree. In this case btreeMoveto() recognizes + ** that the cursor is already where it needs to be and returns without + ** doing any work. To avoid thwarting these optimizations, it is important + ** not to clear the cursor here. + */ + if( pCur->curFlags & BTCF_Multiple ){ + rc = saveAllCursors(pBt, pCur->pgnoRoot, pCur); + if( rc ) return rc; + } + + if( pCur->pKeyInfo==0 ){ + assert( pX->pKey==0 ); + /* If this is an insert into a table b-tree, invalidate any incrblob + ** cursors open on the row being replaced */ + invalidateIncrblobCursors(p, pCur->pgnoRoot, pX->nKey, 0); + + /* If BTREE_SAVEPOSITION is set, the cursor must already be pointing + ** to a row with the same key as the new entry being inserted. + */ +#ifdef SQLITE_DEBUG + if( flags & BTREE_SAVEPOSITION ){ + assert( pCur->curFlags & BTCF_ValidNKey ); + assert( pX->nKey==pCur->info.nKey ); + assert( pCur->info.nSize!=0 ); + assert( loc==0 ); + } +#endif + + /* On the other hand, BTREE_SAVEPOSITION==0 does not imply + ** that the cursor is not pointing to a row to be overwritten. + ** So do a complete check. + */ + if( (pCur->curFlags&BTCF_ValidNKey)!=0 && pX->nKey==pCur->info.nKey ){ + /* The cursor is pointing to the entry that is to be + ** overwritten */ + assert( pX->nData>=0 && pX->nZero>=0 ); + if( pCur->info.nSize!=0 + && pCur->info.nPayload==(u32)pX->nData+pX->nZero + ){ + /* New entry is the same size as the old. Do an overwrite */ + return btreeOverwriteCell(pCur, pX); + } + assert( loc==0 ); + }else if( loc==0 ){ + /* The cursor is *not* pointing to the cell to be overwritten, nor + ** to an adjacent cell. Move the cursor so that it is pointing either + ** to the cell to be overwritten or an adjacent cell. + */ + rc = sqlite3BtreeMovetoUnpacked(pCur, 0, pX->nKey, flags!=0, &loc); + if( rc ) return rc; + } + }else{ + /* This is an index or a WITHOUT ROWID table */ + + /* If BTREE_SAVEPOSITION is set, the cursor must already be pointing + ** to a row with the same key as the new entry being inserted. + */ + assert( (flags & BTREE_SAVEPOSITION)==0 || loc==0 ); + + /* If the cursor is not already pointing either to the cell to be + ** overwritten, or if a new cell is being inserted, if the cursor is + ** not pointing to an immediately adjacent cell, then move the cursor + ** so that it does. + */ + if( loc==0 && (flags & BTREE_SAVEPOSITION)==0 ){ + if( pX->nMem ){ + UnpackedRecord r; + r.pKeyInfo = pCur->pKeyInfo; + r.aMem = pX->aMem; + r.nField = pX->nMem; + r.default_rc = 0; + r.errCode = 0; + r.r1 = 0; + r.r2 = 0; + r.eqSeen = 0; + rc = sqlite3BtreeMovetoUnpacked(pCur, &r, 0, flags!=0, &loc); + }else{ + rc = btreeMoveto(pCur, pX->pKey, pX->nKey, flags!=0, &loc); + } + if( rc ) return rc; + } + + /* If the cursor is currently pointing to an entry to be overwritten + ** and the new content is the same as as the old, then use the + ** overwrite optimization. + */ + if( loc==0 ){ + getCellInfo(pCur); + if( pCur->info.nKey==pX->nKey ){ + BtreePayload x2; + x2.pData = pX->pKey; + x2.nData = pX->nKey; + x2.nZero = 0; + return btreeOverwriteCell(pCur, &x2); + } + } + + } + assert( pCur->eState==CURSOR_VALID || (pCur->eState==CURSOR_INVALID && loc) ); + + pPage = pCur->pPage; + assert( pPage->intKey || pX->nKey>=0 ); + assert( pPage->leaf || !pPage->intKey ); + if( pPage->nFree<0 ){ + rc = btreeComputeFreeSpace(pPage); + if( rc ) return rc; + } + + TRACE(("INSERT: table=%d nkey=%lld ndata=%d page=%d %s\n", + pCur->pgnoRoot, pX->nKey, pX->nData, pPage->pgno, + loc==0 ? "overwrite" : "new entry")); + assert( pPage->isInit ); + newCell = pBt->pTmpSpace; + assert( newCell!=0 ); + rc = fillInCell(pPage, newCell, pX, &szNew); + if( rc ) goto end_insert; + assert( szNew==pPage->xCellSize(pPage, newCell) ); + assert( szNew <= MX_CELL_SIZE(pBt) ); + idx = pCur->ix; + if( loc==0 ){ + CellInfo info; + assert( idxnCell ); + rc = sqlite3PagerWrite(pPage->pDbPage); + if( rc ){ + goto end_insert; + } + oldCell = findCell(pPage, idx); + if( !pPage->leaf ){ + memcpy(newCell, oldCell, 4); + } + rc = clearCell(pPage, oldCell, &info); + testcase( pCur->curFlags & BTCF_ValidOvfl ); + invalidateOverflowCache(pCur); + if( info.nSize==szNew && info.nLocal==info.nPayload + && (!ISAUTOVACUUM || szNewminLocal) + ){ + /* Overwrite the old cell with the new if they are the same size. + ** We could also try to do this if the old cell is smaller, then add + ** the leftover space to the free list. But experiments show that + ** doing that is no faster then skipping this optimization and just + ** calling dropCell() and insertCell(). + ** + ** This optimization cannot be used on an autovacuum database if the + ** new entry uses overflow pages, as the insertCell() call below is + ** necessary to add the PTRMAP_OVERFLOW1 pointer-map entry. */ + assert( rc==SQLITE_OK ); /* clearCell never fails when nLocal==nPayload */ + if( oldCell < pPage->aData+pPage->hdrOffset+10 ){ + return SQLITE_CORRUPT_BKPT; + } + if( oldCell+szNew > pPage->aDataEnd ){ + return SQLITE_CORRUPT_BKPT; + } + memcpy(oldCell, newCell, szNew); + return SQLITE_OK; + } + dropCell(pPage, idx, info.nSize, &rc); + if( rc ) goto end_insert; + }else if( loc<0 && pPage->nCell>0 ){ + assert( pPage->leaf ); + idx = ++pCur->ix; + pCur->curFlags &= ~BTCF_ValidNKey; + }else{ + assert( pPage->leaf ); + } + insertCell(pPage, idx, newCell, szNew, 0, 0, &rc); + assert( pPage->nOverflow==0 || rc==SQLITE_OK ); + assert( rc!=SQLITE_OK || pPage->nCell>0 || pPage->nOverflow>0 ); + + /* If no error has occurred and pPage has an overflow cell, call balance() + ** to redistribute the cells within the tree. Since balance() may move + ** the cursor, zero the BtCursor.info.nSize and BTCF_ValidNKey + ** variables. + ** + ** Previous versions of SQLite called moveToRoot() to move the cursor + ** back to the root page as balance() used to invalidate the contents + ** of BtCursor.apPage[] and BtCursor.aiIdx[]. Instead of doing that, + ** set the cursor state to "invalid". This makes common insert operations + ** slightly faster. + ** + ** There is a subtle but important optimization here too. When inserting + ** multiple records into an intkey b-tree using a single cursor (as can + ** happen while processing an "INSERT INTO ... SELECT" statement), it + ** is advantageous to leave the cursor pointing to the last entry in + ** the b-tree if possible. If the cursor is left pointing to the last + ** entry in the table, and the next row inserted has an integer key + ** larger than the largest existing key, it is possible to insert the + ** row without seeking the cursor. This can be a big performance boost. + */ + pCur->info.nSize = 0; + if( pPage->nOverflow ){ + assert( rc==SQLITE_OK ); + pCur->curFlags &= ~(BTCF_ValidNKey); + rc = balance(pCur); + + /* Must make sure nOverflow is reset to zero even if the balance() + ** fails. Internal data structure corruption will result otherwise. + ** Also, set the cursor state to invalid. This stops saveCursorPosition() + ** from trying to save the current position of the cursor. */ + pCur->pPage->nOverflow = 0; + pCur->eState = CURSOR_INVALID; + if( (flags & BTREE_SAVEPOSITION) && rc==SQLITE_OK ){ + btreeReleaseAllCursorPages(pCur); + if( pCur->pKeyInfo ){ + assert( pCur->pKey==0 ); + pCur->pKey = sqlite3Malloc( pX->nKey ); + if( pCur->pKey==0 ){ + rc = SQLITE_NOMEM; + }else{ + memcpy(pCur->pKey, pX->pKey, pX->nKey); + } + } + pCur->eState = CURSOR_REQUIRESEEK; + pCur->nKey = pX->nKey; + } + } + assert( pCur->iPage<0 || pCur->pPage->nOverflow==0 ); + +end_insert: + return rc; +} + +/* +** Delete the entry that the cursor is pointing to. +** +** If the BTREE_SAVEPOSITION bit of the flags parameter is zero, then +** the cursor is left pointing at an arbitrary location after the delete. +** But if that bit is set, then the cursor is left in a state such that +** the next call to BtreeNext() or BtreePrev() moves it to the same row +** as it would have been on if the call to BtreeDelete() had been omitted. +** +** The BTREE_AUXDELETE bit of flags indicates that is one of several deletes +** associated with a single table entry and its indexes. Only one of those +** deletes is considered the "primary" delete. The primary delete occurs +** on a cursor that is not a BTREE_FORDELETE cursor. All but one delete +** operation on non-FORDELETE cursors is tagged with the AUXDELETE flag. +** The BTREE_AUXDELETE bit is a hint that is not used by this implementation, +** but which might be used by alternative storage engines. +*/ +SQLITE_PRIVATE int sqlite3BtreeDelete(BtCursor *pCur, u8 flags){ + Btree *p = pCur->pBtree; + BtShared *pBt = p->pBt; + int rc; /* Return code */ + MemPage *pPage; /* Page to delete cell from */ + unsigned char *pCell; /* Pointer to cell to delete */ + int iCellIdx; /* Index of cell to delete */ + int iCellDepth; /* Depth of node containing pCell */ + CellInfo info; /* Size of the cell being deleted */ + int bSkipnext = 0; /* Leaf cursor in SKIPNEXT state */ + u8 bPreserve = flags & BTREE_SAVEPOSITION; /* Keep cursor valid */ + + assert( cursorOwnsBtShared(pCur) ); + assert( pBt->inTransaction==TRANS_WRITE ); + assert( (pBt->btsFlags & BTS_READ_ONLY)==0 ); + assert( pCur->curFlags & BTCF_WriteFlag ); + assert( hasSharedCacheTableLock(p, pCur->pgnoRoot, pCur->pKeyInfo!=0, 2) ); + assert( !hasReadConflicts(p, pCur->pgnoRoot) ); + assert( (flags & ~(BTREE_SAVEPOSITION | BTREE_AUXDELETE))==0 ); + if( pCur->eState==CURSOR_REQUIRESEEK ){ + rc = btreeRestoreCursorPosition(pCur); + if( rc ) return rc; + } + assert( pCur->eState==CURSOR_VALID ); + + iCellDepth = pCur->iPage; + iCellIdx = pCur->ix; + pPage = pCur->pPage; + pCell = findCell(pPage, iCellIdx); + if( pPage->nFree<0 && btreeComputeFreeSpace(pPage) ) return SQLITE_CORRUPT; + + /* If the bPreserve flag is set to true, then the cursor position must + ** be preserved following this delete operation. If the current delete + ** will cause a b-tree rebalance, then this is done by saving the cursor + ** key and leaving the cursor in CURSOR_REQUIRESEEK state before + ** returning. + ** + ** Or, if the current delete will not cause a rebalance, then the cursor + ** will be left in CURSOR_SKIPNEXT state pointing to the entry immediately + ** before or after the deleted entry. In this case set bSkipnext to true. */ + if( bPreserve ){ + if( !pPage->leaf + || (pPage->nFree+cellSizePtr(pPage,pCell)+2)>(int)(pBt->usableSize*2/3) + || pPage->nCell==1 /* See dbfuzz001.test for a test case */ + ){ + /* A b-tree rebalance will be required after deleting this entry. + ** Save the cursor key. */ + rc = saveCursorKey(pCur); + if( rc ) return rc; + }else{ + bSkipnext = 1; + } + } + + /* If the page containing the entry to delete is not a leaf page, move + ** the cursor to the largest entry in the tree that is smaller than + ** the entry being deleted. This cell will replace the cell being deleted + ** from the internal node. The 'previous' entry is used for this instead + ** of the 'next' entry, as the previous entry is always a part of the + ** sub-tree headed by the child page of the cell being deleted. This makes + ** balancing the tree following the delete operation easier. */ + if( !pPage->leaf ){ + rc = sqlite3BtreePrevious(pCur, 0); + assert( rc!=SQLITE_DONE ); + if( rc ) return rc; + } + + /* Save the positions of any other cursors open on this table before + ** making any modifications. */ + if( pCur->curFlags & BTCF_Multiple ){ + rc = saveAllCursors(pBt, pCur->pgnoRoot, pCur); + if( rc ) return rc; + } + + /* If this is a delete operation to remove a row from a table b-tree, + ** invalidate any incrblob cursors open on the row being deleted. */ + if( pCur->pKeyInfo==0 ){ + invalidateIncrblobCursors(p, pCur->pgnoRoot, pCur->info.nKey, 0); + } + + /* Make the page containing the entry to be deleted writable. Then free any + ** overflow pages associated with the entry and finally remove the cell + ** itself from within the page. */ + rc = sqlite3PagerWrite(pPage->pDbPage); + if( rc ) return rc; + rc = clearCell(pPage, pCell, &info); + dropCell(pPage, iCellIdx, info.nSize, &rc); + if( rc ) return rc; + + /* If the cell deleted was not located on a leaf page, then the cursor + ** is currently pointing to the largest entry in the sub-tree headed + ** by the child-page of the cell that was just deleted from an internal + ** node. The cell from the leaf node needs to be moved to the internal + ** node to replace the deleted cell. */ + if( !pPage->leaf ){ + MemPage *pLeaf = pCur->pPage; + int nCell; + Pgno n; + unsigned char *pTmp; + + if( pLeaf->nFree<0 ){ + rc = btreeComputeFreeSpace(pLeaf); + if( rc ) return rc; + } + if( iCellDepthiPage-1 ){ + n = pCur->apPage[iCellDepth+1]->pgno; + }else{ + n = pCur->pPage->pgno; + } + pCell = findCell(pLeaf, pLeaf->nCell-1); + if( pCell<&pLeaf->aData[4] ) return SQLITE_CORRUPT_BKPT; + nCell = pLeaf->xCellSize(pLeaf, pCell); + assert( MX_CELL_SIZE(pBt) >= nCell ); + pTmp = pBt->pTmpSpace; + assert( pTmp!=0 ); + rc = sqlite3PagerWrite(pLeaf->pDbPage); + if( rc==SQLITE_OK ){ + insertCell(pPage, iCellIdx, pCell-4, nCell+4, pTmp, n, &rc); + } + dropCell(pLeaf, pLeaf->nCell-1, nCell, &rc); + if( rc ) return rc; + } + + /* Balance the tree. If the entry deleted was located on a leaf page, + ** then the cursor still points to that page. In this case the first + ** call to balance() repairs the tree, and the if(...) condition is + ** never true. + ** + ** Otherwise, if the entry deleted was on an internal node page, then + ** pCur is pointing to the leaf page from which a cell was removed to + ** replace the cell deleted from the internal node. This is slightly + ** tricky as the leaf node may be underfull, and the internal node may + ** be either under or overfull. In this case run the balancing algorithm + ** on the leaf node first. If the balance proceeds far enough up the + ** tree that we can be sure that any problem in the internal node has + ** been corrected, so be it. Otherwise, after balancing the leaf node, + ** walk the cursor up the tree to the internal node and balance it as + ** well. */ + rc = balance(pCur); + if( rc==SQLITE_OK && pCur->iPage>iCellDepth ){ + releasePageNotNull(pCur->pPage); + pCur->iPage--; + while( pCur->iPage>iCellDepth ){ + releasePage(pCur->apPage[pCur->iPage--]); + } + pCur->pPage = pCur->apPage[pCur->iPage]; + rc = balance(pCur); + } + + if( rc==SQLITE_OK ){ + if( bSkipnext ){ + assert( bPreserve && (pCur->iPage==iCellDepth || CORRUPT_DB) ); + assert( pPage==pCur->pPage || CORRUPT_DB ); + assert( (pPage->nCell>0 || CORRUPT_DB) && iCellIdx<=pPage->nCell ); + pCur->eState = CURSOR_SKIPNEXT; + if( iCellIdx>=pPage->nCell ){ + pCur->skipNext = -1; + pCur->ix = pPage->nCell-1; + }else{ + pCur->skipNext = 1; + } + }else{ + rc = moveToRoot(pCur); + if( bPreserve ){ + btreeReleaseAllCursorPages(pCur); + pCur->eState = CURSOR_REQUIRESEEK; + } + if( rc==SQLITE_EMPTY ) rc = SQLITE_OK; + } + } + return rc; +} + +/* +** Create a new BTree table. Write into *piTable the page +** number for the root page of the new table. +** +** The type of type is determined by the flags parameter. Only the +** following values of flags are currently in use. Other values for +** flags might not work: +** +** BTREE_INTKEY|BTREE_LEAFDATA Used for SQL tables with rowid keys +** BTREE_ZERODATA Used for SQL indices +*/ +static int btreeCreateTable(Btree *p, int *piTable, int createTabFlags){ + BtShared *pBt = p->pBt; + MemPage *pRoot; + Pgno pgnoRoot; + int rc; + int ptfFlags; /* Page-type flage for the root page of new table */ + + assert( sqlite3BtreeHoldsMutex(p) ); + assert( pBt->inTransaction==TRANS_WRITE ); + assert( (pBt->btsFlags & BTS_READ_ONLY)==0 ); + +#ifdef SQLITE_OMIT_AUTOVACUUM + rc = allocateBtreePage(pBt, &pRoot, &pgnoRoot, 1, 0); + if( rc ){ + return rc; + } +#else + if( pBt->autoVacuum ){ + Pgno pgnoMove; /* Move a page here to make room for the root-page */ + MemPage *pPageMove; /* The page to move to. */ + + /* Creating a new table may probably require moving an existing database + ** to make room for the new tables root page. In case this page turns + ** out to be an overflow page, delete all overflow page-map caches + ** held by open cursors. + */ + invalidateAllOverflowCache(pBt); + + /* Read the value of meta[3] from the database to determine where the + ** root page of the new table should go. meta[3] is the largest root-page + ** created so far, so the new root-page is (meta[3]+1). + */ + sqlite3BtreeGetMeta(p, BTREE_LARGEST_ROOT_PAGE, &pgnoRoot); + pgnoRoot++; + + /* The new root-page may not be allocated on a pointer-map page, or the + ** PENDING_BYTE page. + */ + while( pgnoRoot==PTRMAP_PAGENO(pBt, pgnoRoot) || + pgnoRoot==PENDING_BYTE_PAGE(pBt) ){ + pgnoRoot++; + } + assert( pgnoRoot>=3 || CORRUPT_DB ); + testcase( pgnoRoot<3 ); + + /* Allocate a page. The page that currently resides at pgnoRoot will + ** be moved to the allocated page (unless the allocated page happens + ** to reside at pgnoRoot). + */ + rc = allocateBtreePage(pBt, &pPageMove, &pgnoMove, pgnoRoot, BTALLOC_EXACT); + if( rc!=SQLITE_OK ){ + return rc; + } + + if( pgnoMove!=pgnoRoot ){ + /* pgnoRoot is the page that will be used for the root-page of + ** the new table (assuming an error did not occur). But we were + ** allocated pgnoMove. If required (i.e. if it was not allocated + ** by extending the file), the current page at position pgnoMove + ** is already journaled. + */ + u8 eType = 0; + Pgno iPtrPage = 0; + + /* Save the positions of any open cursors. This is required in + ** case they are holding a reference to an xFetch reference + ** corresponding to page pgnoRoot. */ + rc = saveAllCursors(pBt, 0, 0); + releasePage(pPageMove); + if( rc!=SQLITE_OK ){ + return rc; + } + + /* Move the page currently at pgnoRoot to pgnoMove. */ + rc = btreeGetPage(pBt, pgnoRoot, &pRoot, 0); + if( rc!=SQLITE_OK ){ + return rc; + } + rc = ptrmapGet(pBt, pgnoRoot, &eType, &iPtrPage); + if( eType==PTRMAP_ROOTPAGE || eType==PTRMAP_FREEPAGE ){ + rc = SQLITE_CORRUPT_BKPT; + } + if( rc!=SQLITE_OK ){ + releasePage(pRoot); + return rc; + } + assert( eType!=PTRMAP_ROOTPAGE ); + assert( eType!=PTRMAP_FREEPAGE ); + rc = relocatePage(pBt, pRoot, eType, iPtrPage, pgnoMove, 0); + releasePage(pRoot); + + /* Obtain the page at pgnoRoot */ + if( rc!=SQLITE_OK ){ + return rc; + } + rc = btreeGetPage(pBt, pgnoRoot, &pRoot, 0); + if( rc!=SQLITE_OK ){ + return rc; + } + rc = sqlite3PagerWrite(pRoot->pDbPage); + if( rc!=SQLITE_OK ){ + releasePage(pRoot); + return rc; + } + }else{ + pRoot = pPageMove; + } + + /* Update the pointer-map and meta-data with the new root-page number. */ + ptrmapPut(pBt, pgnoRoot, PTRMAP_ROOTPAGE, 0, &rc); + if( rc ){ + releasePage(pRoot); + return rc; + } + + /* When the new root page was allocated, page 1 was made writable in + ** order either to increase the database filesize, or to decrement the + ** freelist count. Hence, the sqlite3BtreeUpdateMeta() call cannot fail. + */ + assert( sqlite3PagerIswriteable(pBt->pPage1->pDbPage) ); + rc = sqlite3BtreeUpdateMeta(p, 4, pgnoRoot); + if( NEVER(rc) ){ + releasePage(pRoot); + return rc; + } + + }else{ + rc = allocateBtreePage(pBt, &pRoot, &pgnoRoot, 1, 0); + if( rc ) return rc; + } +#endif + assert( sqlite3PagerIswriteable(pRoot->pDbPage) ); + if( createTabFlags & BTREE_INTKEY ){ + ptfFlags = PTF_INTKEY | PTF_LEAFDATA | PTF_LEAF; + }else{ + ptfFlags = PTF_ZERODATA | PTF_LEAF; + } + zeroPage(pRoot, ptfFlags); + sqlite3PagerUnref(pRoot->pDbPage); + assert( (pBt->openFlags & BTREE_SINGLE)==0 || pgnoRoot==2 ); + *piTable = (int)pgnoRoot; + return SQLITE_OK; +} +SQLITE_PRIVATE int sqlite3BtreeCreateTable(Btree *p, int *piTable, int flags){ + int rc; + sqlite3BtreeEnter(p); + rc = btreeCreateTable(p, piTable, flags); + sqlite3BtreeLeave(p); + return rc; +} + +/* +** Erase the given database page and all its children. Return +** the page to the freelist. +*/ +static int clearDatabasePage( + BtShared *pBt, /* The BTree that contains the table */ + Pgno pgno, /* Page number to clear */ + int freePageFlag, /* Deallocate page if true */ + int *pnChange /* Add number of Cells freed to this counter */ +){ + MemPage *pPage; + int rc; + unsigned char *pCell; + int i; + int hdr; + CellInfo info; + + assert( sqlite3_mutex_held(pBt->mutex) ); + if( pgno>btreePagecount(pBt) ){ + return SQLITE_CORRUPT_BKPT; + } + rc = getAndInitPage(pBt, pgno, &pPage, 0, 0); + if( rc ) return rc; + if( pPage->bBusy ){ + rc = SQLITE_CORRUPT_BKPT; + goto cleardatabasepage_out; + } + pPage->bBusy = 1; + hdr = pPage->hdrOffset; + for(i=0; inCell; i++){ + pCell = findCell(pPage, i); + if( !pPage->leaf ){ + rc = clearDatabasePage(pBt, get4byte(pCell), 1, pnChange); + if( rc ) goto cleardatabasepage_out; + } + rc = clearCell(pPage, pCell, &info); + if( rc ) goto cleardatabasepage_out; + } + if( !pPage->leaf ){ + rc = clearDatabasePage(pBt, get4byte(&pPage->aData[hdr+8]), 1, pnChange); + if( rc ) goto cleardatabasepage_out; + }else if( pnChange ){ + assert( pPage->intKey || CORRUPT_DB ); + testcase( !pPage->intKey ); + *pnChange += pPage->nCell; + } + if( freePageFlag ){ + freePage(pPage, &rc); + }else if( (rc = sqlite3PagerWrite(pPage->pDbPage))==0 ){ + zeroPage(pPage, pPage->aData[hdr] | PTF_LEAF); + } + +cleardatabasepage_out: + pPage->bBusy = 0; + releasePage(pPage); + return rc; +} + +/* +** Delete all information from a single table in the database. iTable is +** the page number of the root of the table. After this routine returns, +** the root page is empty, but still exists. +** +** This routine will fail with SQLITE_LOCKED if there are any open +** read cursors on the table. Open write cursors are moved to the +** root of the table. +** +** If pnChange is not NULL, then table iTable must be an intkey table. The +** integer value pointed to by pnChange is incremented by the number of +** entries in the table. +*/ +SQLITE_PRIVATE int sqlite3BtreeClearTable(Btree *p, int iTable, int *pnChange){ + int rc; + BtShared *pBt = p->pBt; + sqlite3BtreeEnter(p); + assert( p->inTrans==TRANS_WRITE ); + + rc = saveAllCursors(pBt, (Pgno)iTable, 0); + + if( SQLITE_OK==rc ){ + /* Invalidate all incrblob cursors open on table iTable (assuming iTable + ** is the root of a table b-tree - if it is not, the following call is + ** a no-op). */ + invalidateIncrblobCursors(p, (Pgno)iTable, 0, 1); + rc = clearDatabasePage(pBt, (Pgno)iTable, 0, pnChange); + } + sqlite3BtreeLeave(p); + return rc; +} + +/* +** Delete all information from the single table that pCur is open on. +** +** This routine only work for pCur on an ephemeral table. +*/ +SQLITE_PRIVATE int sqlite3BtreeClearTableOfCursor(BtCursor *pCur){ + return sqlite3BtreeClearTable(pCur->pBtree, pCur->pgnoRoot, 0); +} + +/* +** Erase all information in a table and add the root of the table to +** the freelist. Except, the root of the principle table (the one on +** page 1) is never added to the freelist. +** +** This routine will fail with SQLITE_LOCKED if there are any open +** cursors on the table. +** +** If AUTOVACUUM is enabled and the page at iTable is not the last +** root page in the database file, then the last root page +** in the database file is moved into the slot formerly occupied by +** iTable and that last slot formerly occupied by the last root page +** is added to the freelist instead of iTable. In this say, all +** root pages are kept at the beginning of the database file, which +** is necessary for AUTOVACUUM to work right. *piMoved is set to the +** page number that used to be the last root page in the file before +** the move. If no page gets moved, *piMoved is set to 0. +** The last root page is recorded in meta[3] and the value of +** meta[3] is updated by this procedure. +*/ +static int btreeDropTable(Btree *p, Pgno iTable, int *piMoved){ + int rc; + MemPage *pPage = 0; + BtShared *pBt = p->pBt; + + assert( sqlite3BtreeHoldsMutex(p) ); + assert( p->inTrans==TRANS_WRITE ); + assert( iTable>=2 ); + if( iTable>btreePagecount(pBt) ){ + return SQLITE_CORRUPT_BKPT; + } + + rc = btreeGetPage(pBt, (Pgno)iTable, &pPage, 0); + if( rc ) return rc; + rc = sqlite3BtreeClearTable(p, iTable, 0); + if( rc ){ + releasePage(pPage); + return rc; + } + + *piMoved = 0; + +#ifdef SQLITE_OMIT_AUTOVACUUM + freePage(pPage, &rc); + releasePage(pPage); +#else + if( pBt->autoVacuum ){ + Pgno maxRootPgno; + sqlite3BtreeGetMeta(p, BTREE_LARGEST_ROOT_PAGE, &maxRootPgno); + + if( iTable==maxRootPgno ){ + /* If the table being dropped is the table with the largest root-page + ** number in the database, put the root page on the free list. + */ + freePage(pPage, &rc); + releasePage(pPage); + if( rc!=SQLITE_OK ){ + return rc; + } + }else{ + /* The table being dropped does not have the largest root-page + ** number in the database. So move the page that does into the + ** gap left by the deleted root-page. + */ + MemPage *pMove; + releasePage(pPage); + rc = btreeGetPage(pBt, maxRootPgno, &pMove, 0); + if( rc!=SQLITE_OK ){ + return rc; + } + rc = relocatePage(pBt, pMove, PTRMAP_ROOTPAGE, 0, iTable, 0); + releasePage(pMove); + if( rc!=SQLITE_OK ){ + return rc; + } + pMove = 0; + rc = btreeGetPage(pBt, maxRootPgno, &pMove, 0); + freePage(pMove, &rc); + releasePage(pMove); + if( rc!=SQLITE_OK ){ + return rc; + } + *piMoved = maxRootPgno; + } + + /* Set the new 'max-root-page' value in the database header. This + ** is the old value less one, less one more if that happens to + ** be a root-page number, less one again if that is the + ** PENDING_BYTE_PAGE. + */ + maxRootPgno--; + while( maxRootPgno==PENDING_BYTE_PAGE(pBt) + || PTRMAP_ISPAGE(pBt, maxRootPgno) ){ + maxRootPgno--; + } + assert( maxRootPgno!=PENDING_BYTE_PAGE(pBt) ); + + rc = sqlite3BtreeUpdateMeta(p, 4, maxRootPgno); + }else{ + freePage(pPage, &rc); + releasePage(pPage); + } +#endif + return rc; +} +SQLITE_PRIVATE int sqlite3BtreeDropTable(Btree *p, int iTable, int *piMoved){ + int rc; + sqlite3BtreeEnter(p); + rc = btreeDropTable(p, iTable, piMoved); + sqlite3BtreeLeave(p); + return rc; +} + + +/* +** This function may only be called if the b-tree connection already +** has a read or write transaction open on the database. +** +** Read the meta-information out of a database file. Meta[0] +** is the number of free pages currently in the database. Meta[1] +** through meta[15] are available for use by higher layers. Meta[0] +** is read-only, the others are read/write. +** +** The schema layer numbers meta values differently. At the schema +** layer (and the SetCookie and ReadCookie opcodes) the number of +** free pages is not visible. So Cookie[0] is the same as Meta[1]. +** +** This routine treats Meta[BTREE_DATA_VERSION] as a special case. Instead +** of reading the value out of the header, it instead loads the "DataVersion" +** from the pager. The BTREE_DATA_VERSION value is not actually stored in the +** database file. It is a number computed by the pager. But its access +** pattern is the same as header meta values, and so it is convenient to +** read it from this routine. +*/ +SQLITE_PRIVATE void sqlite3BtreeGetMeta(Btree *p, int idx, u32 *pMeta){ + BtShared *pBt = p->pBt; + + sqlite3BtreeEnter(p); + assert( p->inTrans>TRANS_NONE ); + assert( SQLITE_OK==querySharedCacheTableLock(p, MASTER_ROOT, READ_LOCK) ); + assert( pBt->pPage1 ); + assert( idx>=0 && idx<=15 ); + + if( idx==BTREE_DATA_VERSION ){ + *pMeta = sqlite3PagerDataVersion(pBt->pPager) + p->iDataVersion; + }else{ + *pMeta = get4byte(&pBt->pPage1->aData[36 + idx*4]); + } + + /* If auto-vacuum is disabled in this build and this is an auto-vacuum + ** database, mark the database as read-only. */ +#ifdef SQLITE_OMIT_AUTOVACUUM + if( idx==BTREE_LARGEST_ROOT_PAGE && *pMeta>0 ){ + pBt->btsFlags |= BTS_READ_ONLY; + } +#endif + + sqlite3BtreeLeave(p); +} + +/* +** Write meta-information back into the database. Meta[0] is +** read-only and may not be written. +*/ +SQLITE_PRIVATE int sqlite3BtreeUpdateMeta(Btree *p, int idx, u32 iMeta){ + BtShared *pBt = p->pBt; + unsigned char *pP1; + int rc; + assert( idx>=1 && idx<=15 ); + sqlite3BtreeEnter(p); + assert( p->inTrans==TRANS_WRITE ); + assert( pBt->pPage1!=0 ); + pP1 = pBt->pPage1->aData; + rc = sqlite3PagerWrite(pBt->pPage1->pDbPage); + if( rc==SQLITE_OK ){ + put4byte(&pP1[36 + idx*4], iMeta); +#ifndef SQLITE_OMIT_AUTOVACUUM + if( idx==BTREE_INCR_VACUUM ){ + assert( pBt->autoVacuum || iMeta==0 ); + assert( iMeta==0 || iMeta==1 ); + pBt->incrVacuum = (u8)iMeta; + } +#endif + } + sqlite3BtreeLeave(p); + return rc; +} + +#ifndef SQLITE_OMIT_BTREECOUNT +/* +** The first argument, pCur, is a cursor opened on some b-tree. Count the +** number of entries in the b-tree and write the result to *pnEntry. +** +** SQLITE_OK is returned if the operation is successfully executed. +** Otherwise, if an error is encountered (i.e. an IO error or database +** corruption) an SQLite error code is returned. +*/ +SQLITE_PRIVATE int sqlite3BtreeCount(BtCursor *pCur, i64 *pnEntry){ + i64 nEntry = 0; /* Value to return in *pnEntry */ + int rc; /* Return code */ + + rc = moveToRoot(pCur); + if( rc==SQLITE_EMPTY ){ + *pnEntry = 0; + return SQLITE_OK; + } + + /* Unless an error occurs, the following loop runs one iteration for each + ** page in the B-Tree structure (not including overflow pages). + */ + while( rc==SQLITE_OK ){ + int iIdx; /* Index of child node in parent */ + MemPage *pPage; /* Current page of the b-tree */ + + /* If this is a leaf page or the tree is not an int-key tree, then + ** this page contains countable entries. Increment the entry counter + ** accordingly. + */ + pPage = pCur->pPage; + if( pPage->leaf || !pPage->intKey ){ + nEntry += pPage->nCell; + } + + /* pPage is a leaf node. This loop navigates the cursor so that it + ** points to the first interior cell that it points to the parent of + ** the next page in the tree that has not yet been visited. The + ** pCur->aiIdx[pCur->iPage] value is set to the index of the parent cell + ** of the page, or to the number of cells in the page if the next page + ** to visit is the right-child of its parent. + ** + ** If all pages in the tree have been visited, return SQLITE_OK to the + ** caller. + */ + if( pPage->leaf ){ + do { + if( pCur->iPage==0 ){ + /* All pages of the b-tree have been visited. Return successfully. */ + *pnEntry = nEntry; + return moveToRoot(pCur); + } + moveToParent(pCur); + }while ( pCur->ix>=pCur->pPage->nCell ); + + pCur->ix++; + pPage = pCur->pPage; + } + + /* Descend to the child node of the cell that the cursor currently + ** points at. This is the right-child if (iIdx==pPage->nCell). + */ + iIdx = pCur->ix; + if( iIdx==pPage->nCell ){ + rc = moveToChild(pCur, get4byte(&pPage->aData[pPage->hdrOffset+8])); + }else{ + rc = moveToChild(pCur, get4byte(findCell(pPage, iIdx))); + } + } + + /* An error has occurred. Return an error code. */ + return rc; +} +#endif + +/* +** Return the pager associated with a BTree. This routine is used for +** testing and debugging only. +*/ +SQLITE_PRIVATE Pager *sqlite3BtreePager(Btree *p){ + return p->pBt->pPager; +} + +#ifndef SQLITE_OMIT_INTEGRITY_CHECK +/* +** Append a message to the error message string. +*/ +static void checkAppendMsg( + IntegrityCk *pCheck, + const char *zFormat, + ... +){ + va_list ap; + if( !pCheck->mxErr ) return; + pCheck->mxErr--; + pCheck->nErr++; + va_start(ap, zFormat); + if( pCheck->errMsg.nChar ){ + sqlite3_str_append(&pCheck->errMsg, "\n", 1); + } + if( pCheck->zPfx ){ + sqlite3_str_appendf(&pCheck->errMsg, pCheck->zPfx, pCheck->v1, pCheck->v2); + } + sqlite3_str_vappendf(&pCheck->errMsg, zFormat, ap); + va_end(ap); + if( pCheck->errMsg.accError==SQLITE_NOMEM ){ + pCheck->mallocFailed = 1; + } +} +#endif /* SQLITE_OMIT_INTEGRITY_CHECK */ + +#ifndef SQLITE_OMIT_INTEGRITY_CHECK + +/* +** Return non-zero if the bit in the IntegrityCk.aPgRef[] array that +** corresponds to page iPg is already set. +*/ +static int getPageReferenced(IntegrityCk *pCheck, Pgno iPg){ + assert( iPg<=pCheck->nPage && sizeof(pCheck->aPgRef[0])==1 ); + return (pCheck->aPgRef[iPg/8] & (1 << (iPg & 0x07))); +} + +/* +** Set the bit in the IntegrityCk.aPgRef[] array that corresponds to page iPg. +*/ +static void setPageReferenced(IntegrityCk *pCheck, Pgno iPg){ + assert( iPg<=pCheck->nPage && sizeof(pCheck->aPgRef[0])==1 ); + pCheck->aPgRef[iPg/8] |= (1 << (iPg & 0x07)); +} + + +/* +** Add 1 to the reference count for page iPage. If this is the second +** reference to the page, add an error message to pCheck->zErrMsg. +** Return 1 if there are 2 or more references to the page and 0 if +** if this is the first reference to the page. +** +** Also check that the page number is in bounds. +*/ +static int checkRef(IntegrityCk *pCheck, Pgno iPage){ + if( iPage>pCheck->nPage || iPage==0 ){ + checkAppendMsg(pCheck, "invalid page number %d", iPage); + return 1; + } + if( getPageReferenced(pCheck, iPage) ){ + checkAppendMsg(pCheck, "2nd reference to page %d", iPage); + return 1; + } + setPageReferenced(pCheck, iPage); + return 0; +} + +#ifndef SQLITE_OMIT_AUTOVACUUM +/* +** Check that the entry in the pointer-map for page iChild maps to +** page iParent, pointer type ptrType. If not, append an error message +** to pCheck. +*/ +static void checkPtrmap( + IntegrityCk *pCheck, /* Integrity check context */ + Pgno iChild, /* Child page number */ + u8 eType, /* Expected pointer map type */ + Pgno iParent /* Expected pointer map parent page number */ +){ + int rc; + u8 ePtrmapType; + Pgno iPtrmapParent; + + rc = ptrmapGet(pCheck->pBt, iChild, &ePtrmapType, &iPtrmapParent); + if( rc!=SQLITE_OK ){ + if( rc==SQLITE_NOMEM || rc==SQLITE_IOERR_NOMEM ) pCheck->mallocFailed = 1; + checkAppendMsg(pCheck, "Failed to read ptrmap key=%d", iChild); + return; + } + + if( ePtrmapType!=eType || iPtrmapParent!=iParent ){ + checkAppendMsg(pCheck, + "Bad ptr map entry key=%d expected=(%d,%d) got=(%d,%d)", + iChild, eType, iParent, ePtrmapType, iPtrmapParent); + } +} +#endif + +/* +** Check the integrity of the freelist or of an overflow page list. +** Verify that the number of pages on the list is N. +*/ +static void checkList( + IntegrityCk *pCheck, /* Integrity checking context */ + int isFreeList, /* True for a freelist. False for overflow page list */ + int iPage, /* Page number for first page in the list */ + u32 N /* Expected number of pages in the list */ +){ + int i; + u32 expected = N; + int nErrAtStart = pCheck->nErr; + while( iPage!=0 && pCheck->mxErr ){ + DbPage *pOvflPage; + unsigned char *pOvflData; + if( checkRef(pCheck, iPage) ) break; + N--; + if( sqlite3PagerGet(pCheck->pPager, (Pgno)iPage, &pOvflPage, 0) ){ + checkAppendMsg(pCheck, "failed to get page %d", iPage); + break; + } + pOvflData = (unsigned char *)sqlite3PagerGetData(pOvflPage); + if( isFreeList ){ + u32 n = (u32)get4byte(&pOvflData[4]); +#ifndef SQLITE_OMIT_AUTOVACUUM + if( pCheck->pBt->autoVacuum ){ + checkPtrmap(pCheck, iPage, PTRMAP_FREEPAGE, 0); + } +#endif + if( n>pCheck->pBt->usableSize/4-2 ){ + checkAppendMsg(pCheck, + "freelist leaf count too big on page %d", iPage); + N--; + }else{ + for(i=0; i<(int)n; i++){ + Pgno iFreePage = get4byte(&pOvflData[8+i*4]); +#ifndef SQLITE_OMIT_AUTOVACUUM + if( pCheck->pBt->autoVacuum ){ + checkPtrmap(pCheck, iFreePage, PTRMAP_FREEPAGE, 0); + } +#endif + checkRef(pCheck, iFreePage); + } + N -= n; + } + } +#ifndef SQLITE_OMIT_AUTOVACUUM + else{ + /* If this database supports auto-vacuum and iPage is not the last + ** page in this overflow list, check that the pointer-map entry for + ** the following page matches iPage. + */ + if( pCheck->pBt->autoVacuum && N>0 ){ + i = get4byte(pOvflData); + checkPtrmap(pCheck, i, PTRMAP_OVERFLOW2, iPage); + } + } +#endif + iPage = get4byte(pOvflData); + sqlite3PagerUnref(pOvflPage); + } + if( N && nErrAtStart==pCheck->nErr ){ + checkAppendMsg(pCheck, + "%s is %d but should be %d", + isFreeList ? "size" : "overflow list length", + expected-N, expected); + } +} +#endif /* SQLITE_OMIT_INTEGRITY_CHECK */ + +/* +** An implementation of a min-heap. +** +** aHeap[0] is the number of elements on the heap. aHeap[1] is the +** root element. The daughter nodes of aHeap[N] are aHeap[N*2] +** and aHeap[N*2+1]. +** +** The heap property is this: Every node is less than or equal to both +** of its daughter nodes. A consequence of the heap property is that the +** root node aHeap[1] is always the minimum value currently in the heap. +** +** The btreeHeapInsert() routine inserts an unsigned 32-bit number onto +** the heap, preserving the heap property. The btreeHeapPull() routine +** removes the root element from the heap (the minimum value in the heap) +** and then moves other nodes around as necessary to preserve the heap +** property. +** +** This heap is used for cell overlap and coverage testing. Each u32 +** entry represents the span of a cell or freeblock on a btree page. +** The upper 16 bits are the index of the first byte of a range and the +** lower 16 bits are the index of the last byte of that range. +*/ +static void btreeHeapInsert(u32 *aHeap, u32 x){ + u32 j, i = ++aHeap[0]; + aHeap[i] = x; + while( (j = i/2)>0 && aHeap[j]>aHeap[i] ){ + x = aHeap[j]; + aHeap[j] = aHeap[i]; + aHeap[i] = x; + i = j; + } +} +static int btreeHeapPull(u32 *aHeap, u32 *pOut){ + u32 j, i, x; + if( (x = aHeap[0])==0 ) return 0; + *pOut = aHeap[1]; + aHeap[1] = aHeap[x]; + aHeap[x] = 0xffffffff; + aHeap[0]--; + i = 1; + while( (j = i*2)<=aHeap[0] ){ + if( aHeap[j]>aHeap[j+1] ) j++; + if( aHeap[i]zPfx; + int saved_v1 = pCheck->v1; + int saved_v2 = pCheck->v2; + u8 savedIsInit = 0; + + /* Check that the page exists + */ + pBt = pCheck->pBt; + usableSize = pBt->usableSize; + if( iPage==0 ) return 0; + if( checkRef(pCheck, iPage) ) return 0; + pCheck->zPfx = "Page %d: "; + pCheck->v1 = iPage; + if( (rc = btreeGetPage(pBt, (Pgno)iPage, &pPage, 0))!=0 ){ + checkAppendMsg(pCheck, + "unable to get the page. error code=%d", rc); + goto end_of_check; + } + + /* Clear MemPage.isInit to make sure the corruption detection code in + ** btreeInitPage() is executed. */ + savedIsInit = pPage->isInit; + pPage->isInit = 0; + if( (rc = btreeInitPage(pPage))!=0 ){ + assert( rc==SQLITE_CORRUPT ); /* The only possible error from InitPage */ + checkAppendMsg(pCheck, + "btreeInitPage() returns error code %d", rc); + goto end_of_check; + } + if( (rc = btreeComputeFreeSpace(pPage))!=0 ){ + assert( rc==SQLITE_CORRUPT ); + checkAppendMsg(pCheck, "free space corruption", rc); + goto end_of_check; + } + data = pPage->aData; + hdr = pPage->hdrOffset; + + /* Set up for cell analysis */ + pCheck->zPfx = "On tree page %d cell %d: "; + contentOffset = get2byteNotZero(&data[hdr+5]); + assert( contentOffset<=usableSize ); /* Enforced by btreeInitPage() */ + + /* EVIDENCE-OF: R-37002-32774 The two-byte integer at offset 3 gives the + ** number of cells on the page. */ + nCell = get2byte(&data[hdr+3]); + assert( pPage->nCell==nCell ); + + /* EVIDENCE-OF: R-23882-45353 The cell pointer array of a b-tree page + ** immediately follows the b-tree page header. */ + cellStart = hdr + 12 - 4*pPage->leaf; + assert( pPage->aCellIdx==&data[cellStart] ); + pCellIdx = &data[cellStart + 2*(nCell-1)]; + + if( !pPage->leaf ){ + /* Analyze the right-child page of internal pages */ + pgno = get4byte(&data[hdr+8]); +#ifndef SQLITE_OMIT_AUTOVACUUM + if( pBt->autoVacuum ){ + pCheck->zPfx = "On page %d at right child: "; + checkPtrmap(pCheck, pgno, PTRMAP_BTREE, iPage); + } +#endif + depth = checkTreePage(pCheck, pgno, &maxKey, maxKey); + keyCanBeEqual = 0; + }else{ + /* For leaf pages, the coverage check will occur in the same loop + ** as the other cell checks, so initialize the heap. */ + heap = pCheck->heap; + heap[0] = 0; + } + + /* EVIDENCE-OF: R-02776-14802 The cell pointer array consists of K 2-byte + ** integer offsets to the cell contents. */ + for(i=nCell-1; i>=0 && pCheck->mxErr; i--){ + CellInfo info; + + /* Check cell size */ + pCheck->v2 = i; + assert( pCellIdx==&data[cellStart + i*2] ); + pc = get2byteAligned(pCellIdx); + pCellIdx -= 2; + if( pcusableSize-4 ){ + checkAppendMsg(pCheck, "Offset %d out of range %d..%d", + pc, contentOffset, usableSize-4); + doCoverageCheck = 0; + continue; + } + pCell = &data[pc]; + pPage->xParseCell(pPage, pCell, &info); + if( pc+info.nSize>usableSize ){ + checkAppendMsg(pCheck, "Extends off end of page"); + doCoverageCheck = 0; + continue; + } + + /* Check for integer primary key out of range */ + if( pPage->intKey ){ + if( keyCanBeEqual ? (info.nKey > maxKey) : (info.nKey >= maxKey) ){ + checkAppendMsg(pCheck, "Rowid %lld out of order", info.nKey); + } + maxKey = info.nKey; + keyCanBeEqual = 0; /* Only the first key on the page may ==maxKey */ + } + + /* Check the content overflow list */ + if( info.nPayload>info.nLocal ){ + u32 nPage; /* Number of pages on the overflow chain */ + Pgno pgnoOvfl; /* First page of the overflow chain */ + assert( pc + info.nSize - 4 <= usableSize ); + nPage = (info.nPayload - info.nLocal + usableSize - 5)/(usableSize - 4); + pgnoOvfl = get4byte(&pCell[info.nSize - 4]); +#ifndef SQLITE_OMIT_AUTOVACUUM + if( pBt->autoVacuum ){ + checkPtrmap(pCheck, pgnoOvfl, PTRMAP_OVERFLOW1, iPage); + } +#endif + checkList(pCheck, 0, pgnoOvfl, nPage); + } + + if( !pPage->leaf ){ + /* Check sanity of left child page for internal pages */ + pgno = get4byte(pCell); +#ifndef SQLITE_OMIT_AUTOVACUUM + if( pBt->autoVacuum ){ + checkPtrmap(pCheck, pgno, PTRMAP_BTREE, iPage); + } +#endif + d2 = checkTreePage(pCheck, pgno, &maxKey, maxKey); + keyCanBeEqual = 0; + if( d2!=depth ){ + checkAppendMsg(pCheck, "Child page depth differs"); + depth = d2; + } + }else{ + /* Populate the coverage-checking heap for leaf pages */ + btreeHeapInsert(heap, (pc<<16)|(pc+info.nSize-1)); + } + } + *piMinKey = maxKey; + + /* Check for complete coverage of the page + */ + pCheck->zPfx = 0; + if( doCoverageCheck && pCheck->mxErr>0 ){ + /* For leaf pages, the min-heap has already been initialized and the + ** cells have already been inserted. But for internal pages, that has + ** not yet been done, so do it now */ + if( !pPage->leaf ){ + heap = pCheck->heap; + heap[0] = 0; + for(i=nCell-1; i>=0; i--){ + u32 size; + pc = get2byteAligned(&data[cellStart+i*2]); + size = pPage->xCellSize(pPage, &data[pc]); + btreeHeapInsert(heap, (pc<<16)|(pc+size-1)); + } + } + /* Add the freeblocks to the min-heap + ** + ** EVIDENCE-OF: R-20690-50594 The second field of the b-tree page header + ** is the offset of the first freeblock, or zero if there are no + ** freeblocks on the page. + */ + i = get2byte(&data[hdr+1]); + while( i>0 ){ + int size, j; + assert( (u32)i<=usableSize-4 ); /* Enforced by btreeComputeFreeSpace() */ + size = get2byte(&data[i+2]); + assert( (u32)(i+size)<=usableSize ); /* due to btreeComputeFreeSpace() */ + btreeHeapInsert(heap, (((u32)i)<<16)|(i+size-1)); + /* EVIDENCE-OF: R-58208-19414 The first 2 bytes of a freeblock are a + ** big-endian integer which is the offset in the b-tree page of the next + ** freeblock in the chain, or zero if the freeblock is the last on the + ** chain. */ + j = get2byte(&data[i]); + /* EVIDENCE-OF: R-06866-39125 Freeblocks are always connected in order of + ** increasing offset. */ + assert( j==0 || j>i+size ); /* Enforced by btreeComputeFreeSpace() */ + assert( (u32)j<=usableSize-4 ); /* Enforced by btreeComputeFreeSpace() */ + i = j; + } + /* Analyze the min-heap looking for overlap between cells and/or + ** freeblocks, and counting the number of untracked bytes in nFrag. + ** + ** Each min-heap entry is of the form: (start_address<<16)|end_address. + ** There is an implied first entry the covers the page header, the cell + ** pointer index, and the gap between the cell pointer index and the start + ** of cell content. + ** + ** The loop below pulls entries from the min-heap in order and compares + ** the start_address against the previous end_address. If there is an + ** overlap, that means bytes are used multiple times. If there is a gap, + ** that gap is added to the fragmentation count. + */ + nFrag = 0; + prev = contentOffset - 1; /* Implied first min-heap entry */ + while( btreeHeapPull(heap,&x) ){ + if( (prev&0xffff)>=(x>>16) ){ + checkAppendMsg(pCheck, + "Multiple uses for byte %u of page %d", x>>16, iPage); + break; + }else{ + nFrag += (x>>16) - (prev&0xffff) - 1; + prev = x; + } + } + nFrag += usableSize - (prev&0xffff) - 1; + /* EVIDENCE-OF: R-43263-13491 The total number of bytes in all fragments + ** is stored in the fifth field of the b-tree page header. + ** EVIDENCE-OF: R-07161-27322 The one-byte integer at offset 7 gives the + ** number of fragmented free bytes within the cell content area. + */ + if( heap[0]==0 && nFrag!=data[hdr+7] ){ + checkAppendMsg(pCheck, + "Fragmentation of %d bytes reported as %d on page %d", + nFrag, data[hdr+7], iPage); + } + } + +end_of_check: + if( !doCoverageCheck ) pPage->isInit = savedIsInit; + releasePage(pPage); + pCheck->zPfx = saved_zPfx; + pCheck->v1 = saved_v1; + pCheck->v2 = saved_v2; + return depth+1; +} +#endif /* SQLITE_OMIT_INTEGRITY_CHECK */ + +#ifndef SQLITE_OMIT_INTEGRITY_CHECK +/* +** This routine does a complete check of the given BTree file. aRoot[] is +** an array of pages numbers were each page number is the root page of +** a table. nRoot is the number of entries in aRoot. +** +** A read-only or read-write transaction must be opened before calling +** this function. +** +** Write the number of error seen in *pnErr. Except for some memory +** allocation errors, an error message held in memory obtained from +** malloc is returned if *pnErr is non-zero. If *pnErr==0 then NULL is +** returned. If a memory allocation error occurs, NULL is returned. +*/ +SQLITE_PRIVATE char *sqlite3BtreeIntegrityCheck( + Btree *p, /* The btree to be checked */ + int *aRoot, /* An array of root pages numbers for individual trees */ + int nRoot, /* Number of entries in aRoot[] */ + int mxErr, /* Stop reporting errors after this many */ + int *pnErr /* Write number of errors seen to this variable */ +){ + Pgno i; + IntegrityCk sCheck; + BtShared *pBt = p->pBt; + u64 savedDbFlags = pBt->db->flags; + char zErr[100]; + VVA_ONLY( int nRef ); + + sqlite3BtreeEnter(p); + assert( p->inTrans>TRANS_NONE && pBt->inTransaction>TRANS_NONE ); + VVA_ONLY( nRef = sqlite3PagerRefcount(pBt->pPager) ); + assert( nRef>=0 ); + sCheck.pBt = pBt; + sCheck.pPager = pBt->pPager; + sCheck.nPage = btreePagecount(sCheck.pBt); + sCheck.mxErr = mxErr; + sCheck.nErr = 0; + sCheck.mallocFailed = 0; + sCheck.zPfx = 0; + sCheck.v1 = 0; + sCheck.v2 = 0; + sCheck.aPgRef = 0; + sCheck.heap = 0; + sqlite3StrAccumInit(&sCheck.errMsg, 0, zErr, sizeof(zErr), SQLITE_MAX_LENGTH); + sCheck.errMsg.printfFlags = SQLITE_PRINTF_INTERNAL; + if( sCheck.nPage==0 ){ + goto integrity_ck_cleanup; + } + + sCheck.aPgRef = sqlite3MallocZero((sCheck.nPage / 8)+ 1); + if( !sCheck.aPgRef ){ + sCheck.mallocFailed = 1; + goto integrity_ck_cleanup; + } + sCheck.heap = (u32*)sqlite3PageMalloc( pBt->pageSize ); + if( sCheck.heap==0 ){ + sCheck.mallocFailed = 1; + goto integrity_ck_cleanup; + } + + i = PENDING_BYTE_PAGE(pBt); + if( i<=sCheck.nPage ) setPageReferenced(&sCheck, i); + + /* Check the integrity of the freelist + */ + sCheck.zPfx = "Main freelist: "; + checkList(&sCheck, 1, get4byte(&pBt->pPage1->aData[32]), + get4byte(&pBt->pPage1->aData[36])); + sCheck.zPfx = 0; + + /* Check all the tables. + */ +#ifndef SQLITE_OMIT_AUTOVACUUM + if( pBt->autoVacuum ){ + int mx = 0; + int mxInHdr; + for(i=0; (int)ipPage1->aData[52]); + if( mx!=mxInHdr ){ + checkAppendMsg(&sCheck, + "max rootpage (%d) disagrees with header (%d)", + mx, mxInHdr + ); + } + }else if( get4byte(&pBt->pPage1->aData[64])!=0 ){ + checkAppendMsg(&sCheck, + "incremental_vacuum enabled with a max rootpage of zero" + ); + } +#endif + testcase( pBt->db->flags & SQLITE_CellSizeCk ); + pBt->db->flags &= ~(u64)SQLITE_CellSizeCk; + for(i=0; (int)iautoVacuum && aRoot[i]>1 ){ + checkPtrmap(&sCheck, aRoot[i], PTRMAP_ROOTPAGE, 0); + } +#endif + checkTreePage(&sCheck, aRoot[i], ¬Used, LARGEST_INT64); + } + pBt->db->flags = savedDbFlags; + + /* Make sure every page in the file is referenced + */ + for(i=1; i<=sCheck.nPage && sCheck.mxErr; i++){ +#ifdef SQLITE_OMIT_AUTOVACUUM + if( getPageReferenced(&sCheck, i)==0 ){ + checkAppendMsg(&sCheck, "Page %d is never used", i); + } +#else + /* If the database supports auto-vacuum, make sure no tables contain + ** references to pointer-map pages. + */ + if( getPageReferenced(&sCheck, i)==0 && + (PTRMAP_PAGENO(pBt, i)!=i || !pBt->autoVacuum) ){ + checkAppendMsg(&sCheck, "Page %d is never used", i); + } + if( getPageReferenced(&sCheck, i)!=0 && + (PTRMAP_PAGENO(pBt, i)==i && pBt->autoVacuum) ){ + checkAppendMsg(&sCheck, "Pointer map page %d is referenced", i); + } +#endif + } + + /* Clean up and report errors. + */ +integrity_ck_cleanup: + sqlite3PageFree(sCheck.heap); + sqlite3_free(sCheck.aPgRef); + if( sCheck.mallocFailed ){ + sqlite3_str_reset(&sCheck.errMsg); + sCheck.nErr++; + } + *pnErr = sCheck.nErr; + if( sCheck.nErr==0 ) sqlite3_str_reset(&sCheck.errMsg); + /* Make sure this analysis did not leave any unref() pages. */ + assert( nRef==sqlite3PagerRefcount(pBt->pPager) ); + sqlite3BtreeLeave(p); + return sqlite3StrAccumFinish(&sCheck.errMsg); +} +#endif /* SQLITE_OMIT_INTEGRITY_CHECK */ + +/* +** Return the full pathname of the underlying database file. Return +** an empty string if the database is in-memory or a TEMP database. +** +** The pager filename is invariant as long as the pager is +** open so it is safe to access without the BtShared mutex. +*/ +SQLITE_PRIVATE const char *sqlite3BtreeGetFilename(Btree *p){ + assert( p->pBt->pPager!=0 ); + return sqlite3PagerFilename(p->pBt->pPager, 1); +} + +/* +** Return the pathname of the journal file for this database. The return +** value of this routine is the same regardless of whether the journal file +** has been created or not. +** +** The pager journal filename is invariant as long as the pager is +** open so it is safe to access without the BtShared mutex. +*/ +SQLITE_PRIVATE const char *sqlite3BtreeGetJournalname(Btree *p){ + assert( p->pBt->pPager!=0 ); + return sqlite3PagerJournalname(p->pBt->pPager); +} + +/* +** Return non-zero if a transaction is active. +*/ +SQLITE_PRIVATE int sqlite3BtreeIsInTrans(Btree *p){ + assert( p==0 || sqlite3_mutex_held(p->db->mutex) ); + return (p && (p->inTrans==TRANS_WRITE)); +} + +#ifndef SQLITE_OMIT_WAL +/* +** Run a checkpoint on the Btree passed as the first argument. +** +** Return SQLITE_LOCKED if this or any other connection has an open +** transaction on the shared-cache the argument Btree is connected to. +** +** Parameter eMode is one of SQLITE_CHECKPOINT_PASSIVE, FULL or RESTART. +*/ +SQLITE_PRIVATE int sqlite3BtreeCheckpoint(Btree *p, int eMode, int *pnLog, int *pnCkpt){ + int rc = SQLITE_OK; + if( p ){ + BtShared *pBt = p->pBt; + sqlite3BtreeEnter(p); + if( pBt->inTransaction!=TRANS_NONE ){ + rc = SQLITE_LOCKED; + }else{ + rc = sqlite3PagerCheckpoint(pBt->pPager, p->db, eMode, pnLog, pnCkpt); + } + sqlite3BtreeLeave(p); + } + return rc; +} +#endif + +/* +** Return non-zero if a read (or write) transaction is active. +*/ +SQLITE_PRIVATE int sqlite3BtreeIsInReadTrans(Btree *p){ + assert( p ); + assert( sqlite3_mutex_held(p->db->mutex) ); + return p->inTrans!=TRANS_NONE; +} + +SQLITE_PRIVATE int sqlite3BtreeIsInBackup(Btree *p){ + assert( p ); + assert( sqlite3_mutex_held(p->db->mutex) ); + return p->nBackup!=0; +} + +/* +** This function returns a pointer to a blob of memory associated with +** a single shared-btree. The memory is used by client code for its own +** purposes (for example, to store a high-level schema associated with +** the shared-btree). The btree layer manages reference counting issues. +** +** The first time this is called on a shared-btree, nBytes bytes of memory +** are allocated, zeroed, and returned to the caller. For each subsequent +** call the nBytes parameter is ignored and a pointer to the same blob +** of memory returned. +** +** If the nBytes parameter is 0 and the blob of memory has not yet been +** allocated, a null pointer is returned. If the blob has already been +** allocated, it is returned as normal. +** +** Just before the shared-btree is closed, the function passed as the +** xFree argument when the memory allocation was made is invoked on the +** blob of allocated memory. The xFree function should not call sqlite3_free() +** on the memory, the btree layer does that. +*/ +SQLITE_PRIVATE void *sqlite3BtreeSchema(Btree *p, int nBytes, void(*xFree)(void *)){ + BtShared *pBt = p->pBt; + sqlite3BtreeEnter(p); + if( !pBt->pSchema && nBytes ){ + pBt->pSchema = sqlite3DbMallocZero(0, nBytes); + pBt->xFreeSchema = xFree; + } + sqlite3BtreeLeave(p); + return pBt->pSchema; +} + +/* +** Return SQLITE_LOCKED_SHAREDCACHE if another user of the same shared +** btree as the argument handle holds an exclusive lock on the +** sqlite_master table. Otherwise SQLITE_OK. +*/ +SQLITE_PRIVATE int sqlite3BtreeSchemaLocked(Btree *p){ + int rc; + assert( sqlite3_mutex_held(p->db->mutex) ); + sqlite3BtreeEnter(p); + rc = querySharedCacheTableLock(p, MASTER_ROOT, READ_LOCK); + assert( rc==SQLITE_OK || rc==SQLITE_LOCKED_SHAREDCACHE ); + sqlite3BtreeLeave(p); + return rc; +} + + +#ifndef SQLITE_OMIT_SHARED_CACHE +/* +** Obtain a lock on the table whose root page is iTab. The +** lock is a write lock if isWritelock is true or a read lock +** if it is false. +*/ +SQLITE_PRIVATE int sqlite3BtreeLockTable(Btree *p, int iTab, u8 isWriteLock){ + int rc = SQLITE_OK; + assert( p->inTrans!=TRANS_NONE ); + if( p->sharable ){ + u8 lockType = READ_LOCK + isWriteLock; + assert( READ_LOCK+1==WRITE_LOCK ); + assert( isWriteLock==0 || isWriteLock==1 ); + + sqlite3BtreeEnter(p); + rc = querySharedCacheTableLock(p, iTab, lockType); + if( rc==SQLITE_OK ){ + rc = setSharedCacheTableLock(p, iTab, lockType); + } + sqlite3BtreeLeave(p); + } + return rc; +} +#endif + +#ifndef SQLITE_OMIT_INCRBLOB +/* +** Argument pCsr must be a cursor opened for writing on an +** INTKEY table currently pointing at a valid table entry. +** This function modifies the data stored as part of that entry. +** +** Only the data content may only be modified, it is not possible to +** change the length of the data stored. If this function is called with +** parameters that attempt to write past the end of the existing data, +** no modifications are made and SQLITE_CORRUPT is returned. +*/ +SQLITE_PRIVATE int sqlite3BtreePutData(BtCursor *pCsr, u32 offset, u32 amt, void *z){ + int rc; + assert( cursorOwnsBtShared(pCsr) ); + assert( sqlite3_mutex_held(pCsr->pBtree->db->mutex) ); + assert( pCsr->curFlags & BTCF_Incrblob ); + + rc = restoreCursorPosition(pCsr); + if( rc!=SQLITE_OK ){ + return rc; + } + assert( pCsr->eState!=CURSOR_REQUIRESEEK ); + if( pCsr->eState!=CURSOR_VALID ){ + return SQLITE_ABORT; + } + + /* Save the positions of all other cursors open on this table. This is + ** required in case any of them are holding references to an xFetch + ** version of the b-tree page modified by the accessPayload call below. + ** + ** Note that pCsr must be open on a INTKEY table and saveCursorPosition() + ** and hence saveAllCursors() cannot fail on a BTREE_INTKEY table, hence + ** saveAllCursors can only return SQLITE_OK. + */ + VVA_ONLY(rc =) saveAllCursors(pCsr->pBt, pCsr->pgnoRoot, pCsr); + assert( rc==SQLITE_OK ); + + /* Check some assumptions: + ** (a) the cursor is open for writing, + ** (b) there is a read/write transaction open, + ** (c) the connection holds a write-lock on the table (if required), + ** (d) there are no conflicting read-locks, and + ** (e) the cursor points at a valid row of an intKey table. + */ + if( (pCsr->curFlags & BTCF_WriteFlag)==0 ){ + return SQLITE_READONLY; + } + assert( (pCsr->pBt->btsFlags & BTS_READ_ONLY)==0 + && pCsr->pBt->inTransaction==TRANS_WRITE ); + assert( hasSharedCacheTableLock(pCsr->pBtree, pCsr->pgnoRoot, 0, 2) ); + assert( !hasReadConflicts(pCsr->pBtree, pCsr->pgnoRoot) ); + assert( pCsr->pPage->intKey ); + + return accessPayload(pCsr, offset, amt, (unsigned char *)z, 1); +} + +/* +** Mark this cursor as an incremental blob cursor. +*/ +SQLITE_PRIVATE void sqlite3BtreeIncrblobCursor(BtCursor *pCur){ + pCur->curFlags |= BTCF_Incrblob; + pCur->pBtree->hasIncrblobCur = 1; +} +#endif + +/* +** Set both the "read version" (single byte at byte offset 18) and +** "write version" (single byte at byte offset 19) fields in the database +** header to iVersion. +*/ +SQLITE_PRIVATE int sqlite3BtreeSetVersion(Btree *pBtree, int iVersion){ + BtShared *pBt = pBtree->pBt; + int rc; /* Return code */ + + assert( iVersion==1 || iVersion==2 ); + + /* If setting the version fields to 1, do not automatically open the + ** WAL connection, even if the version fields are currently set to 2. + */ + pBt->btsFlags &= ~BTS_NO_WAL; + if( iVersion==1 ) pBt->btsFlags |= BTS_NO_WAL; + + rc = sqlite3BtreeBeginTrans(pBtree, 0, 0); + if( rc==SQLITE_OK ){ + u8 *aData = pBt->pPage1->aData; + if( aData[18]!=(u8)iVersion || aData[19]!=(u8)iVersion ){ + rc = sqlite3BtreeBeginTrans(pBtree, 2, 0); + if( rc==SQLITE_OK ){ + rc = sqlite3PagerWrite(pBt->pPage1->pDbPage); + if( rc==SQLITE_OK ){ + aData[18] = (u8)iVersion; + aData[19] = (u8)iVersion; + } + } + } + } + + pBt->btsFlags &= ~BTS_NO_WAL; + return rc; +} + +/* +** Return true if the cursor has a hint specified. This routine is +** only used from within assert() statements +*/ +SQLITE_PRIVATE int sqlite3BtreeCursorHasHint(BtCursor *pCsr, unsigned int mask){ + return (pCsr->hints & mask)!=0; +} + +/* +** Return true if the given Btree is read-only. +*/ +SQLITE_PRIVATE int sqlite3BtreeIsReadonly(Btree *p){ + return (p->pBt->btsFlags & BTS_READ_ONLY)!=0; +} + +/* +** Return the size of the header added to each page by this module. +*/ +SQLITE_PRIVATE int sqlite3HeaderSizeBtree(void){ return ROUND8(sizeof(MemPage)); } + +#if !defined(SQLITE_OMIT_SHARED_CACHE) +/* +** Return true if the Btree passed as the only argument is sharable. +*/ +SQLITE_PRIVATE int sqlite3BtreeSharable(Btree *p){ + return p->sharable; +} + +/* +** Return the number of connections to the BtShared object accessed by +** the Btree handle passed as the only argument. For private caches +** this is always 1. For shared caches it may be 1 or greater. +*/ +SQLITE_PRIVATE int sqlite3BtreeConnectionCount(Btree *p){ + testcase( p->sharable ); + return p->pBt->nRef; +} +#endif + +/************** End of btree.c ***********************************************/ +/************** Begin file backup.c ******************************************/ +/* +** 2009 January 28 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** This file contains the implementation of the sqlite3_backup_XXX() +** API functions and the related features. +*/ +/* #include "sqliteInt.h" */ +/* #include "btreeInt.h" */ + +/* +** Structure allocated for each backup operation. +*/ +struct sqlite3_backup { + sqlite3* pDestDb; /* Destination database handle */ + Btree *pDest; /* Destination b-tree file */ + u32 iDestSchema; /* Original schema cookie in destination */ + int bDestLocked; /* True once a write-transaction is open on pDest */ + + Pgno iNext; /* Page number of the next source page to copy */ + sqlite3* pSrcDb; /* Source database handle */ + Btree *pSrc; /* Source b-tree file */ + + int rc; /* Backup process error code */ + + /* These two variables are set by every call to backup_step(). They are + ** read by calls to backup_remaining() and backup_pagecount(). + */ + Pgno nRemaining; /* Number of pages left to copy */ + Pgno nPagecount; /* Total number of pages to copy */ + + int isAttached; /* True once backup has been registered with pager */ + sqlite3_backup *pNext; /* Next backup associated with source pager */ +}; + +/* +** THREAD SAFETY NOTES: +** +** Once it has been created using backup_init(), a single sqlite3_backup +** structure may be accessed via two groups of thread-safe entry points: +** +** * Via the sqlite3_backup_XXX() API function backup_step() and +** backup_finish(). Both these functions obtain the source database +** handle mutex and the mutex associated with the source BtShared +** structure, in that order. +** +** * Via the BackupUpdate() and BackupRestart() functions, which are +** invoked by the pager layer to report various state changes in +** the page cache associated with the source database. The mutex +** associated with the source database BtShared structure will always +** be held when either of these functions are invoked. +** +** The other sqlite3_backup_XXX() API functions, backup_remaining() and +** backup_pagecount() are not thread-safe functions. If they are called +** while some other thread is calling backup_step() or backup_finish(), +** the values returned may be invalid. There is no way for a call to +** BackupUpdate() or BackupRestart() to interfere with backup_remaining() +** or backup_pagecount(). +** +** Depending on the SQLite configuration, the database handles and/or +** the Btree objects may have their own mutexes that require locking. +** Non-sharable Btrees (in-memory databases for example), do not have +** associated mutexes. +*/ + +/* +** Return a pointer corresponding to database zDb (i.e. "main", "temp") +** in connection handle pDb. If such a database cannot be found, return +** a NULL pointer and write an error message to pErrorDb. +** +** If the "temp" database is requested, it may need to be opened by this +** function. If an error occurs while doing so, return 0 and write an +** error message to pErrorDb. +*/ +static Btree *findBtree(sqlite3 *pErrorDb, sqlite3 *pDb, const char *zDb){ + int i = sqlite3FindDbName(pDb, zDb); + + if( i==1 ){ + Parse sParse; + int rc = 0; + memset(&sParse, 0, sizeof(sParse)); + sParse.db = pDb; + if( sqlite3OpenTempDatabase(&sParse) ){ + sqlite3ErrorWithMsg(pErrorDb, sParse.rc, "%s", sParse.zErrMsg); + rc = SQLITE_ERROR; + } + sqlite3DbFree(pErrorDb, sParse.zErrMsg); + sqlite3ParserReset(&sParse); + if( rc ){ + return 0; + } + } + + if( i<0 ){ + sqlite3ErrorWithMsg(pErrorDb, SQLITE_ERROR, "unknown database %s", zDb); + return 0; + } + + return pDb->aDb[i].pBt; +} + +/* +** Attempt to set the page size of the destination to match the page size +** of the source. +*/ +static int setDestPgsz(sqlite3_backup *p){ + int rc; + rc = sqlite3BtreeSetPageSize(p->pDest,sqlite3BtreeGetPageSize(p->pSrc),-1,0); + return rc; +} + +/* +** Check that there is no open read-transaction on the b-tree passed as the +** second argument. If there is not, return SQLITE_OK. Otherwise, if there +** is an open read-transaction, return SQLITE_ERROR and leave an error +** message in database handle db. +*/ +static int checkReadTransaction(sqlite3 *db, Btree *p){ + if( sqlite3BtreeIsInReadTrans(p) ){ + sqlite3ErrorWithMsg(db, SQLITE_ERROR, "destination database is in use"); + return SQLITE_ERROR; + } + return SQLITE_OK; +} + +/* +** Create an sqlite3_backup process to copy the contents of zSrcDb from +** connection handle pSrcDb to zDestDb in pDestDb. If successful, return +** a pointer to the new sqlite3_backup object. +** +** If an error occurs, NULL is returned and an error code and error message +** stored in database handle pDestDb. +*/ +SQLITE_API sqlite3_backup *sqlite3_backup_init( + sqlite3* pDestDb, /* Database to write to */ + const char *zDestDb, /* Name of database within pDestDb */ + sqlite3* pSrcDb, /* Database connection to read from */ + const char *zSrcDb /* Name of database within pSrcDb */ +){ + sqlite3_backup *p; /* Value to return */ + +#ifdef SQLITE_ENABLE_API_ARMOR + if( !sqlite3SafetyCheckOk(pSrcDb)||!sqlite3SafetyCheckOk(pDestDb) ){ + (void)SQLITE_MISUSE_BKPT; + return 0; + } +#endif + + /* Lock the source database handle. The destination database + ** handle is not locked in this routine, but it is locked in + ** sqlite3_backup_step(). The user is required to ensure that no + ** other thread accesses the destination handle for the duration + ** of the backup operation. Any attempt to use the destination + ** database connection while a backup is in progress may cause + ** a malfunction or a deadlock. + */ + sqlite3_mutex_enter(pSrcDb->mutex); + sqlite3_mutex_enter(pDestDb->mutex); + + if( pSrcDb==pDestDb ){ + sqlite3ErrorWithMsg( + pDestDb, SQLITE_ERROR, "source and destination must be distinct" + ); + p = 0; + }else { + /* Allocate space for a new sqlite3_backup object... + ** EVIDENCE-OF: R-64852-21591 The sqlite3_backup object is created by a + ** call to sqlite3_backup_init() and is destroyed by a call to + ** sqlite3_backup_finish(). */ + p = (sqlite3_backup *)sqlite3MallocZero(sizeof(sqlite3_backup)); + if( !p ){ + sqlite3Error(pDestDb, SQLITE_NOMEM_BKPT); + } + } + + /* If the allocation succeeded, populate the new object. */ + if( p ){ + p->pSrc = findBtree(pDestDb, pSrcDb, zSrcDb); + p->pDest = findBtree(pDestDb, pDestDb, zDestDb); + p->pDestDb = pDestDb; + p->pSrcDb = pSrcDb; + p->iNext = 1; + p->isAttached = 0; + + if( 0==p->pSrc || 0==p->pDest + || checkReadTransaction(pDestDb, p->pDest)!=SQLITE_OK + ){ + /* One (or both) of the named databases did not exist or an OOM + ** error was hit. Or there is a transaction open on the destination + ** database. The error has already been written into the pDestDb + ** handle. All that is left to do here is free the sqlite3_backup + ** structure. */ + sqlite3_free(p); + p = 0; + } + } + if( p ){ + p->pSrc->nBackup++; + } + + sqlite3_mutex_leave(pDestDb->mutex); + sqlite3_mutex_leave(pSrcDb->mutex); + return p; +} + +/* +** Argument rc is an SQLite error code. Return true if this error is +** considered fatal if encountered during a backup operation. All errors +** are considered fatal except for SQLITE_BUSY and SQLITE_LOCKED. +*/ +static int isFatalError(int rc){ + return (rc!=SQLITE_OK && rc!=SQLITE_BUSY && ALWAYS(rc!=SQLITE_LOCKED)); +} + +/* +** Parameter zSrcData points to a buffer containing the data for +** page iSrcPg from the source database. Copy this data into the +** destination database. +*/ +static int backupOnePage( + sqlite3_backup *p, /* Backup handle */ + Pgno iSrcPg, /* Source database page to backup */ + const u8 *zSrcData, /* Source database page data */ + int bUpdate /* True for an update, false otherwise */ +){ + Pager * const pDestPager = sqlite3BtreePager(p->pDest); + const int nSrcPgsz = sqlite3BtreeGetPageSize(p->pSrc); + int nDestPgsz = sqlite3BtreeGetPageSize(p->pDest); + const int nCopy = MIN(nSrcPgsz, nDestPgsz); + const i64 iEnd = (i64)iSrcPg*(i64)nSrcPgsz; +#ifdef SQLITE_HAS_CODEC + /* Use BtreeGetReserveNoMutex() for the source b-tree, as although it is + ** guaranteed that the shared-mutex is held by this thread, handle + ** p->pSrc may not actually be the owner. */ + int nSrcReserve = sqlite3BtreeGetReserveNoMutex(p->pSrc); + int nDestReserve = sqlite3BtreeGetOptimalReserve(p->pDest); +#endif + int rc = SQLITE_OK; + i64 iOff; + + assert( sqlite3BtreeGetReserveNoMutex(p->pSrc)>=0 ); + assert( p->bDestLocked ); + assert( !isFatalError(p->rc) ); + assert( iSrcPg!=PENDING_BYTE_PAGE(p->pSrc->pBt) ); + assert( zSrcData ); + + /* Catch the case where the destination is an in-memory database and the + ** page sizes of the source and destination differ. + */ + if( nSrcPgsz!=nDestPgsz && sqlite3PagerIsMemdb(pDestPager) ){ + rc = SQLITE_READONLY; + } + +#ifdef SQLITE_HAS_CODEC + /* Backup is not possible if the page size of the destination is changing + ** and a codec is in use. + */ + if( nSrcPgsz!=nDestPgsz && sqlite3PagerGetCodec(pDestPager)!=0 ){ + rc = SQLITE_READONLY; + } + + /* Backup is not possible if the number of bytes of reserve space differ + ** between source and destination. If there is a difference, try to + ** fix the destination to agree with the source. If that is not possible, + ** then the backup cannot proceed. + */ + if( nSrcReserve!=nDestReserve ){ + u32 newPgsz = nSrcPgsz; + rc = sqlite3PagerSetPagesize(pDestPager, &newPgsz, nSrcReserve); + if( rc==SQLITE_OK && newPgsz!=(u32)nSrcPgsz ) rc = SQLITE_READONLY; + } +#endif + + /* This loop runs once for each destination page spanned by the source + ** page. For each iteration, variable iOff is set to the byte offset + ** of the destination page. + */ + for(iOff=iEnd-(i64)nSrcPgsz; rc==SQLITE_OK && iOffpDest->pBt) ) continue; + if( SQLITE_OK==(rc = sqlite3PagerGet(pDestPager, iDest, &pDestPg, 0)) + && SQLITE_OK==(rc = sqlite3PagerWrite(pDestPg)) + ){ + const u8 *zIn = &zSrcData[iOff%nSrcPgsz]; + u8 *zDestData = sqlite3PagerGetData(pDestPg); + u8 *zOut = &zDestData[iOff%nDestPgsz]; + + /* Copy the data from the source page into the destination page. + ** Then clear the Btree layer MemPage.isInit flag. Both this module + ** and the pager code use this trick (clearing the first byte + ** of the page 'extra' space to invalidate the Btree layers + ** cached parse of the page). MemPage.isInit is marked + ** "MUST BE FIRST" for this purpose. + */ + memcpy(zOut, zIn, nCopy); + ((u8 *)sqlite3PagerGetExtra(pDestPg))[0] = 0; + if( iOff==0 && bUpdate==0 ){ + sqlite3Put4byte(&zOut[28], sqlite3BtreeLastPage(p->pSrc)); + } + } + sqlite3PagerUnref(pDestPg); + } + + return rc; +} + +/* +** If pFile is currently larger than iSize bytes, then truncate it to +** exactly iSize bytes. If pFile is not larger than iSize bytes, then +** this function is a no-op. +** +** Return SQLITE_OK if everything is successful, or an SQLite error +** code if an error occurs. +*/ +static int backupTruncateFile(sqlite3_file *pFile, i64 iSize){ + i64 iCurrent; + int rc = sqlite3OsFileSize(pFile, &iCurrent); + if( rc==SQLITE_OK && iCurrent>iSize ){ + rc = sqlite3OsTruncate(pFile, iSize); + } + return rc; +} + +/* +** Register this backup object with the associated source pager for +** callbacks when pages are changed or the cache invalidated. +*/ +static void attachBackupObject(sqlite3_backup *p){ + sqlite3_backup **pp; + assert( sqlite3BtreeHoldsMutex(p->pSrc) ); + pp = sqlite3PagerBackupPtr(sqlite3BtreePager(p->pSrc)); + p->pNext = *pp; + *pp = p; + p->isAttached = 1; +} + +/* +** Copy nPage pages from the source b-tree to the destination. +*/ +SQLITE_API int sqlite3_backup_step(sqlite3_backup *p, int nPage){ + int rc; + int destMode; /* Destination journal mode */ + int pgszSrc = 0; /* Source page size */ + int pgszDest = 0; /* Destination page size */ + +#ifdef SQLITE_ENABLE_API_ARMOR + if( p==0 ) return SQLITE_MISUSE_BKPT; +#endif + sqlite3_mutex_enter(p->pSrcDb->mutex); + sqlite3BtreeEnter(p->pSrc); + if( p->pDestDb ){ + sqlite3_mutex_enter(p->pDestDb->mutex); + } + + rc = p->rc; + if( !isFatalError(rc) ){ + Pager * const pSrcPager = sqlite3BtreePager(p->pSrc); /* Source pager */ + Pager * const pDestPager = sqlite3BtreePager(p->pDest); /* Dest pager */ + int ii; /* Iterator variable */ + int nSrcPage = -1; /* Size of source db in pages */ + int bCloseTrans = 0; /* True if src db requires unlocking */ + + /* If the source pager is currently in a write-transaction, return + ** SQLITE_BUSY immediately. + */ + if( p->pDestDb && p->pSrc->pBt->inTransaction==TRANS_WRITE ){ + rc = SQLITE_BUSY; + }else{ + rc = SQLITE_OK; + } + + /* If there is no open read-transaction on the source database, open + ** one now. If a transaction is opened here, then it will be closed + ** before this function exits. + */ + if( rc==SQLITE_OK && 0==sqlite3BtreeIsInReadTrans(p->pSrc) ){ + rc = sqlite3BtreeBeginTrans(p->pSrc, 0, 0); + bCloseTrans = 1; + } + + /* If the destination database has not yet been locked (i.e. if this + ** is the first call to backup_step() for the current backup operation), + ** try to set its page size to the same as the source database. This + ** is especially important on ZipVFS systems, as in that case it is + ** not possible to create a database file that uses one page size by + ** writing to it with another. */ + if( p->bDestLocked==0 && rc==SQLITE_OK && setDestPgsz(p)==SQLITE_NOMEM ){ + rc = SQLITE_NOMEM; + } + + /* Lock the destination database, if it is not locked already. */ + if( SQLITE_OK==rc && p->bDestLocked==0 + && SQLITE_OK==(rc = sqlite3BtreeBeginTrans(p->pDest, 2, + (int*)&p->iDestSchema)) + ){ + p->bDestLocked = 1; + } + + /* Do not allow backup if the destination database is in WAL mode + ** and the page sizes are different between source and destination */ + pgszSrc = sqlite3BtreeGetPageSize(p->pSrc); + pgszDest = sqlite3BtreeGetPageSize(p->pDest); + destMode = sqlite3PagerGetJournalMode(sqlite3BtreePager(p->pDest)); + if( SQLITE_OK==rc && destMode==PAGER_JOURNALMODE_WAL && pgszSrc!=pgszDest ){ + rc = SQLITE_READONLY; + } + + /* Now that there is a read-lock on the source database, query the + ** source pager for the number of pages in the database. + */ + nSrcPage = (int)sqlite3BtreeLastPage(p->pSrc); + assert( nSrcPage>=0 ); + for(ii=0; (nPage<0 || iiiNext<=(Pgno)nSrcPage && !rc; ii++){ + const Pgno iSrcPg = p->iNext; /* Source page number */ + if( iSrcPg!=PENDING_BYTE_PAGE(p->pSrc->pBt) ){ + DbPage *pSrcPg; /* Source page object */ + rc = sqlite3PagerGet(pSrcPager, iSrcPg, &pSrcPg,PAGER_GET_READONLY); + if( rc==SQLITE_OK ){ + rc = backupOnePage(p, iSrcPg, sqlite3PagerGetData(pSrcPg), 0); + sqlite3PagerUnref(pSrcPg); + } + } + p->iNext++; + } + if( rc==SQLITE_OK ){ + p->nPagecount = nSrcPage; + p->nRemaining = nSrcPage+1-p->iNext; + if( p->iNext>(Pgno)nSrcPage ){ + rc = SQLITE_DONE; + }else if( !p->isAttached ){ + attachBackupObject(p); + } + } + + /* Update the schema version field in the destination database. This + ** is to make sure that the schema-version really does change in + ** the case where the source and destination databases have the + ** same schema version. + */ + if( rc==SQLITE_DONE ){ + if( nSrcPage==0 ){ + rc = sqlite3BtreeNewDb(p->pDest); + nSrcPage = 1; + } + if( rc==SQLITE_OK || rc==SQLITE_DONE ){ + rc = sqlite3BtreeUpdateMeta(p->pDest,1,p->iDestSchema+1); + } + if( rc==SQLITE_OK ){ + if( p->pDestDb ){ + sqlite3ResetAllSchemasOfConnection(p->pDestDb); + } + if( destMode==PAGER_JOURNALMODE_WAL ){ + rc = sqlite3BtreeSetVersion(p->pDest, 2); + } + } + if( rc==SQLITE_OK ){ + int nDestTruncate; + /* Set nDestTruncate to the final number of pages in the destination + ** database. The complication here is that the destination page + ** size may be different to the source page size. + ** + ** If the source page size is smaller than the destination page size, + ** round up. In this case the call to sqlite3OsTruncate() below will + ** fix the size of the file. However it is important to call + ** sqlite3PagerTruncateImage() here so that any pages in the + ** destination file that lie beyond the nDestTruncate page mark are + ** journalled by PagerCommitPhaseOne() before they are destroyed + ** by the file truncation. + */ + assert( pgszSrc==sqlite3BtreeGetPageSize(p->pSrc) ); + assert( pgszDest==sqlite3BtreeGetPageSize(p->pDest) ); + if( pgszSrcpDest->pBt) ){ + nDestTruncate--; + } + }else{ + nDestTruncate = nSrcPage * (pgszSrc/pgszDest); + } + assert( nDestTruncate>0 ); + + if( pgszSrc= iSize || ( + nDestTruncate==(int)(PENDING_BYTE_PAGE(p->pDest->pBt)-1) + && iSize>=PENDING_BYTE && iSize<=PENDING_BYTE+pgszDest + )); + + /* This block ensures that all data required to recreate the original + ** database has been stored in the journal for pDestPager and the + ** journal synced to disk. So at this point we may safely modify + ** the database file in any way, knowing that if a power failure + ** occurs, the original database will be reconstructed from the + ** journal file. */ + sqlite3PagerPagecount(pDestPager, &nDstPage); + for(iPg=nDestTruncate; rc==SQLITE_OK && iPg<=(Pgno)nDstPage; iPg++){ + if( iPg!=PENDING_BYTE_PAGE(p->pDest->pBt) ){ + DbPage *pPg; + rc = sqlite3PagerGet(pDestPager, iPg, &pPg, 0); + if( rc==SQLITE_OK ){ + rc = sqlite3PagerWrite(pPg); + sqlite3PagerUnref(pPg); + } + } + } + if( rc==SQLITE_OK ){ + rc = sqlite3PagerCommitPhaseOne(pDestPager, 0, 1); + } + + /* Write the extra pages and truncate the database file as required */ + iEnd = MIN(PENDING_BYTE + pgszDest, iSize); + for( + iOff=PENDING_BYTE+pgszSrc; + rc==SQLITE_OK && iOffpDest, 0)) + ){ + rc = SQLITE_DONE; + } + } + } + + /* If bCloseTrans is true, then this function opened a read transaction + ** on the source database. Close the read transaction here. There is + ** no need to check the return values of the btree methods here, as + ** "committing" a read-only transaction cannot fail. + */ + if( bCloseTrans ){ + TESTONLY( int rc2 ); + TESTONLY( rc2 = ) sqlite3BtreeCommitPhaseOne(p->pSrc, 0); + TESTONLY( rc2 |= ) sqlite3BtreeCommitPhaseTwo(p->pSrc, 0); + assert( rc2==SQLITE_OK ); + } + + if( rc==SQLITE_IOERR_NOMEM ){ + rc = SQLITE_NOMEM_BKPT; + } + p->rc = rc; + } + if( p->pDestDb ){ + sqlite3_mutex_leave(p->pDestDb->mutex); + } + sqlite3BtreeLeave(p->pSrc); + sqlite3_mutex_leave(p->pSrcDb->mutex); + return rc; +} + +/* +** Release all resources associated with an sqlite3_backup* handle. +*/ +SQLITE_API int sqlite3_backup_finish(sqlite3_backup *p){ + sqlite3_backup **pp; /* Ptr to head of pagers backup list */ + sqlite3 *pSrcDb; /* Source database connection */ + int rc; /* Value to return */ + + /* Enter the mutexes */ + if( p==0 ) return SQLITE_OK; + pSrcDb = p->pSrcDb; + sqlite3_mutex_enter(pSrcDb->mutex); + sqlite3BtreeEnter(p->pSrc); + if( p->pDestDb ){ + sqlite3_mutex_enter(p->pDestDb->mutex); + } + + /* Detach this backup from the source pager. */ + if( p->pDestDb ){ + p->pSrc->nBackup--; + } + if( p->isAttached ){ + pp = sqlite3PagerBackupPtr(sqlite3BtreePager(p->pSrc)); + assert( pp!=0 ); + while( *pp!=p ){ + pp = &(*pp)->pNext; + assert( pp!=0 ); + } + *pp = p->pNext; + } + + /* If a transaction is still open on the Btree, roll it back. */ + sqlite3BtreeRollback(p->pDest, SQLITE_OK, 0); + + /* Set the error code of the destination database handle. */ + rc = (p->rc==SQLITE_DONE) ? SQLITE_OK : p->rc; + if( p->pDestDb ){ + sqlite3Error(p->pDestDb, rc); + + /* Exit the mutexes and free the backup context structure. */ + sqlite3LeaveMutexAndCloseZombie(p->pDestDb); + } + sqlite3BtreeLeave(p->pSrc); + if( p->pDestDb ){ + /* EVIDENCE-OF: R-64852-21591 The sqlite3_backup object is created by a + ** call to sqlite3_backup_init() and is destroyed by a call to + ** sqlite3_backup_finish(). */ + sqlite3_free(p); + } + sqlite3LeaveMutexAndCloseZombie(pSrcDb); + return rc; +} + +/* +** Return the number of pages still to be backed up as of the most recent +** call to sqlite3_backup_step(). +*/ +SQLITE_API int sqlite3_backup_remaining(sqlite3_backup *p){ +#ifdef SQLITE_ENABLE_API_ARMOR + if( p==0 ){ + (void)SQLITE_MISUSE_BKPT; + return 0; + } +#endif + return p->nRemaining; +} + +/* +** Return the total number of pages in the source database as of the most +** recent call to sqlite3_backup_step(). +*/ +SQLITE_API int sqlite3_backup_pagecount(sqlite3_backup *p){ +#ifdef SQLITE_ENABLE_API_ARMOR + if( p==0 ){ + (void)SQLITE_MISUSE_BKPT; + return 0; + } +#endif + return p->nPagecount; +} + +/* +** This function is called after the contents of page iPage of the +** source database have been modified. If page iPage has already been +** copied into the destination database, then the data written to the +** destination is now invalidated. The destination copy of iPage needs +** to be updated with the new data before the backup operation is +** complete. +** +** It is assumed that the mutex associated with the BtShared object +** corresponding to the source database is held when this function is +** called. +*/ +static SQLITE_NOINLINE void backupUpdate( + sqlite3_backup *p, + Pgno iPage, + const u8 *aData +){ + assert( p!=0 ); + do{ + assert( sqlite3_mutex_held(p->pSrc->pBt->mutex) ); + if( !isFatalError(p->rc) && iPageiNext ){ + /* The backup process p has already copied page iPage. But now it + ** has been modified by a transaction on the source pager. Copy + ** the new data into the backup. + */ + int rc; + assert( p->pDestDb ); + sqlite3_mutex_enter(p->pDestDb->mutex); + rc = backupOnePage(p, iPage, aData, 1); + sqlite3_mutex_leave(p->pDestDb->mutex); + assert( rc!=SQLITE_BUSY && rc!=SQLITE_LOCKED ); + if( rc!=SQLITE_OK ){ + p->rc = rc; + } + } + }while( (p = p->pNext)!=0 ); +} +SQLITE_PRIVATE void sqlite3BackupUpdate(sqlite3_backup *pBackup, Pgno iPage, const u8 *aData){ + if( pBackup ) backupUpdate(pBackup, iPage, aData); +} + +/* +** Restart the backup process. This is called when the pager layer +** detects that the database has been modified by an external database +** connection. In this case there is no way of knowing which of the +** pages that have been copied into the destination database are still +** valid and which are not, so the entire process needs to be restarted. +** +** It is assumed that the mutex associated with the BtShared object +** corresponding to the source database is held when this function is +** called. +*/ +SQLITE_PRIVATE void sqlite3BackupRestart(sqlite3_backup *pBackup){ + sqlite3_backup *p; /* Iterator variable */ + for(p=pBackup; p; p=p->pNext){ + assert( sqlite3_mutex_held(p->pSrc->pBt->mutex) ); + p->iNext = 1; + } +} + +#ifndef SQLITE_OMIT_VACUUM +/* +** Copy the complete content of pBtFrom into pBtTo. A transaction +** must be active for both files. +** +** The size of file pTo may be reduced by this operation. If anything +** goes wrong, the transaction on pTo is rolled back. If successful, the +** transaction is committed before returning. +*/ +SQLITE_PRIVATE int sqlite3BtreeCopyFile(Btree *pTo, Btree *pFrom){ + int rc; + sqlite3_file *pFd; /* File descriptor for database pTo */ + sqlite3_backup b; + sqlite3BtreeEnter(pTo); + sqlite3BtreeEnter(pFrom); + + assert( sqlite3BtreeIsInTrans(pTo) ); + pFd = sqlite3PagerFile(sqlite3BtreePager(pTo)); + if( pFd->pMethods ){ + i64 nByte = sqlite3BtreeGetPageSize(pFrom)*(i64)sqlite3BtreeLastPage(pFrom); + rc = sqlite3OsFileControl(pFd, SQLITE_FCNTL_OVERWRITE, &nByte); + if( rc==SQLITE_NOTFOUND ) rc = SQLITE_OK; + if( rc ) goto copy_finished; + } + + /* Set up an sqlite3_backup object. sqlite3_backup.pDestDb must be set + ** to 0. This is used by the implementations of sqlite3_backup_step() + ** and sqlite3_backup_finish() to detect that they are being called + ** from this function, not directly by the user. + */ + memset(&b, 0, sizeof(b)); + b.pSrcDb = pFrom->db; + b.pSrc = pFrom; + b.pDest = pTo; + b.iNext = 1; + +#ifdef SQLITE_HAS_CODEC + sqlite3PagerAlignReserve(sqlite3BtreePager(pTo), sqlite3BtreePager(pFrom)); +#endif + + /* 0x7FFFFFFF is the hard limit for the number of pages in a database + ** file. By passing this as the number of pages to copy to + ** sqlite3_backup_step(), we can guarantee that the copy finishes + ** within a single call (unless an error occurs). The assert() statement + ** checks this assumption - (p->rc) should be set to either SQLITE_DONE + ** or an error code. */ + sqlite3_backup_step(&b, 0x7FFFFFFF); + assert( b.rc!=SQLITE_OK ); + + rc = sqlite3_backup_finish(&b); + if( rc==SQLITE_OK ){ + pTo->pBt->btsFlags &= ~BTS_PAGESIZE_FIXED; + }else{ + sqlite3PagerClearCache(sqlite3BtreePager(b.pDest)); + } + + assert( sqlite3BtreeIsInTrans(pTo)==0 ); +copy_finished: + sqlite3BtreeLeave(pFrom); + sqlite3BtreeLeave(pTo); + return rc; +} +#endif /* SQLITE_OMIT_VACUUM */ + +/************** End of backup.c **********************************************/ +/************** Begin file vdbemem.c *****************************************/ +/* +** 2004 May 26 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** +** This file contains code use to manipulate "Mem" structure. A "Mem" +** stores a single value in the VDBE. Mem is an opaque structure visible +** only within the VDBE. Interface routines refer to a Mem using the +** name sqlite_value +*/ +/* #include "sqliteInt.h" */ +/* #include "vdbeInt.h" */ + +/* True if X is a power of two. 0 is considered a power of two here. +** In other words, return true if X has at most one bit set. +*/ +#define ISPOWEROF2(X) (((X)&((X)-1))==0) + +#ifdef SQLITE_DEBUG +/* +** Check invariants on a Mem object. +** +** This routine is intended for use inside of assert() statements, like +** this: assert( sqlite3VdbeCheckMemInvariants(pMem) ); +*/ +SQLITE_PRIVATE int sqlite3VdbeCheckMemInvariants(Mem *p){ + /* If MEM_Dyn is set then Mem.xDel!=0. + ** Mem.xDel might not be initialized if MEM_Dyn is clear. + */ + assert( (p->flags & MEM_Dyn)==0 || p->xDel!=0 ); + + /* MEM_Dyn may only be set if Mem.szMalloc==0. In this way we + ** ensure that if Mem.szMalloc>0 then it is safe to do + ** Mem.z = Mem.zMalloc without having to check Mem.flags&MEM_Dyn. + ** That saves a few cycles in inner loops. */ + assert( (p->flags & MEM_Dyn)==0 || p->szMalloc==0 ); + + /* Cannot have more than one of MEM_Int, MEM_Real, or MEM_IntReal */ + assert( ISPOWEROF2(p->flags & (MEM_Int|MEM_Real|MEM_IntReal)) ); + + if( p->flags & MEM_Null ){ + /* Cannot be both MEM_Null and some other type */ + assert( (p->flags & (MEM_Int|MEM_Real|MEM_Str|MEM_Blob|MEM_Agg))==0 ); + + /* If MEM_Null is set, then either the value is a pure NULL (the usual + ** case) or it is a pointer set using sqlite3_bind_pointer() or + ** sqlite3_result_pointer(). If a pointer, then MEM_Term must also be + ** set. + */ + if( (p->flags & (MEM_Term|MEM_Subtype))==(MEM_Term|MEM_Subtype) ){ + /* This is a pointer type. There may be a flag to indicate what to + ** do with the pointer. */ + assert( ((p->flags&MEM_Dyn)!=0 ? 1 : 0) + + ((p->flags&MEM_Ephem)!=0 ? 1 : 0) + + ((p->flags&MEM_Static)!=0 ? 1 : 0) <= 1 ); + + /* No other bits set */ + assert( (p->flags & ~(MEM_Null|MEM_Term|MEM_Subtype|MEM_FromBind + |MEM_Dyn|MEM_Ephem|MEM_Static))==0 ); + }else{ + /* A pure NULL might have other flags, such as MEM_Static, MEM_Dyn, + ** MEM_Ephem, MEM_Cleared, or MEM_Subtype */ + } + }else{ + /* The MEM_Cleared bit is only allowed on NULLs */ + assert( (p->flags & MEM_Cleared)==0 ); + } + + /* The szMalloc field holds the correct memory allocation size */ + assert( p->szMalloc==0 + || p->szMalloc==sqlite3DbMallocSize(p->db,p->zMalloc) ); + + /* If p holds a string or blob, the Mem.z must point to exactly + ** one of the following: + ** + ** (1) Memory in Mem.zMalloc and managed by the Mem object + ** (2) Memory to be freed using Mem.xDel + ** (3) An ephemeral string or blob + ** (4) A static string or blob + */ + if( (p->flags & (MEM_Str|MEM_Blob)) && p->n>0 ){ + assert( + ((p->szMalloc>0 && p->z==p->zMalloc)? 1 : 0) + + ((p->flags&MEM_Dyn)!=0 ? 1 : 0) + + ((p->flags&MEM_Ephem)!=0 ? 1 : 0) + + ((p->flags&MEM_Static)!=0 ? 1 : 0) == 1 + ); + } + return 1; +} +#endif + +/* +** Render a Mem object which is one of MEM_Int, MEM_Real, or MEM_IntReal +** into a buffer. +*/ +static void vdbeMemRenderNum(int sz, char *zBuf, Mem *p){ + StrAccum acc; + assert( p->flags & (MEM_Int|MEM_Real|MEM_IntReal) ); + sqlite3StrAccumInit(&acc, 0, zBuf, sz, 0); + if( p->flags & MEM_Int ){ + sqlite3_str_appendf(&acc, "%lld", p->u.i); + }else if( p->flags & MEM_IntReal ){ + sqlite3_str_appendf(&acc, "%!.15g", (double)p->u.i); + }else{ + sqlite3_str_appendf(&acc, "%!.15g", p->u.r); + } + assert( acc.zText==zBuf && acc.mxAlloc<=0 ); + zBuf[acc.nChar] = 0; /* Fast version of sqlite3StrAccumFinish(&acc) */ +} + +#ifdef SQLITE_DEBUG +/* +** Validity checks on pMem. pMem holds a string. +** +** (1) Check that string value of pMem agrees with its integer or real value. +** (2) Check that the string is correctly zero terminated +** +** A single int or real value always converts to the same strings. But +** many different strings can be converted into the same int or real. +** If a table contains a numeric value and an index is based on the +** corresponding string value, then it is important that the string be +** derived from the numeric value, not the other way around, to ensure +** that the index and table are consistent. See ticket +** https://www.sqlite.org/src/info/343634942dd54ab (2018-01-31) for +** an example. +** +** This routine looks at pMem to verify that if it has both a numeric +** representation and a string representation then the string rep has +** been derived from the numeric and not the other way around. It returns +** true if everything is ok and false if there is a problem. +** +** This routine is for use inside of assert() statements only. +*/ +SQLITE_PRIVATE int sqlite3VdbeMemValidStrRep(Mem *p){ + char zBuf[100]; + char *z; + int i, j, incr; + if( (p->flags & MEM_Str)==0 ) return 1; + if( p->flags & MEM_Term ){ + /* Insure that the string is properly zero-terminated. Pay particular + ** attention to the case where p->n is odd */ + if( p->szMalloc>0 && p->z==p->zMalloc ){ + assert( p->enc==SQLITE_UTF8 || p->szMalloc >= ((p->n+1)&~1)+2 ); + assert( p->enc!=SQLITE_UTF8 || p->szMalloc >= p->n+1 ); + } + assert( p->z[p->n]==0 ); + assert( p->enc==SQLITE_UTF8 || p->z[(p->n+1)&~1]==0 ); + assert( p->enc==SQLITE_UTF8 || p->z[((p->n+1)&~1)+1]==0 ); + } + if( (p->flags & (MEM_Int|MEM_Real|MEM_IntReal))==0 ) return 1; + vdbeMemRenderNum(sizeof(zBuf), zBuf, p); + z = p->z; + i = j = 0; + incr = 1; + if( p->enc!=SQLITE_UTF8 ){ + incr = 2; + if( p->enc==SQLITE_UTF16BE ) z++; + } + while( zBuf[j] ){ + if( zBuf[j++]!=z[i] ) return 0; + i += incr; + } + return 1; +} +#endif /* SQLITE_DEBUG */ + +/* +** If pMem is an object with a valid string representation, this routine +** ensures the internal encoding for the string representation is +** 'desiredEnc', one of SQLITE_UTF8, SQLITE_UTF16LE or SQLITE_UTF16BE. +** +** If pMem is not a string object, or the encoding of the string +** representation is already stored using the requested encoding, then this +** routine is a no-op. +** +** SQLITE_OK is returned if the conversion is successful (or not required). +** SQLITE_NOMEM may be returned if a malloc() fails during conversion +** between formats. +*/ +SQLITE_PRIVATE int sqlite3VdbeChangeEncoding(Mem *pMem, int desiredEnc){ +#ifndef SQLITE_OMIT_UTF16 + int rc; +#endif + assert( !sqlite3VdbeMemIsRowSet(pMem) ); + assert( desiredEnc==SQLITE_UTF8 || desiredEnc==SQLITE_UTF16LE + || desiredEnc==SQLITE_UTF16BE ); + if( !(pMem->flags&MEM_Str) || pMem->enc==desiredEnc ){ + return SQLITE_OK; + } + assert( pMem->db==0 || sqlite3_mutex_held(pMem->db->mutex) ); +#ifdef SQLITE_OMIT_UTF16 + return SQLITE_ERROR; +#else + + /* MemTranslate() may return SQLITE_OK or SQLITE_NOMEM. If NOMEM is returned, + ** then the encoding of the value may not have changed. + */ + rc = sqlite3VdbeMemTranslate(pMem, (u8)desiredEnc); + assert(rc==SQLITE_OK || rc==SQLITE_NOMEM); + assert(rc==SQLITE_OK || pMem->enc!=desiredEnc); + assert(rc==SQLITE_NOMEM || pMem->enc==desiredEnc); + return rc; +#endif +} + +/* +** Make sure pMem->z points to a writable allocation of at least n bytes. +** +** If the bPreserve argument is true, then copy of the content of +** pMem->z into the new allocation. pMem must be either a string or +** blob if bPreserve is true. If bPreserve is false, any prior content +** in pMem->z is discarded. +*/ +SQLITE_PRIVATE SQLITE_NOINLINE int sqlite3VdbeMemGrow(Mem *pMem, int n, int bPreserve){ + assert( sqlite3VdbeCheckMemInvariants(pMem) ); + assert( !sqlite3VdbeMemIsRowSet(pMem) ); + testcase( pMem->db==0 ); + + /* If the bPreserve flag is set to true, then the memory cell must already + ** contain a valid string or blob value. */ + assert( bPreserve==0 || pMem->flags&(MEM_Blob|MEM_Str) ); + testcase( bPreserve && pMem->z==0 ); + + assert( pMem->szMalloc==0 + || pMem->szMalloc==sqlite3DbMallocSize(pMem->db, pMem->zMalloc) ); + if( pMem->szMalloc>0 && bPreserve && pMem->z==pMem->zMalloc ){ + if( pMem->db ){ + pMem->z = pMem->zMalloc = sqlite3DbReallocOrFree(pMem->db, pMem->z, n); + }else{ + pMem->zMalloc = sqlite3Realloc(pMem->z, n); + if( pMem->zMalloc==0 ) sqlite3_free(pMem->z); + pMem->z = pMem->zMalloc; + } + bPreserve = 0; + }else{ + if( pMem->szMalloc>0 ) sqlite3DbFreeNN(pMem->db, pMem->zMalloc); + pMem->zMalloc = sqlite3DbMallocRaw(pMem->db, n); + } + if( pMem->zMalloc==0 ){ + sqlite3VdbeMemSetNull(pMem); + pMem->z = 0; + pMem->szMalloc = 0; + return SQLITE_NOMEM_BKPT; + }else{ + pMem->szMalloc = sqlite3DbMallocSize(pMem->db, pMem->zMalloc); + } + + if( bPreserve && pMem->z ){ + assert( pMem->z!=pMem->zMalloc ); + memcpy(pMem->zMalloc, pMem->z, pMem->n); + } + if( (pMem->flags&MEM_Dyn)!=0 ){ + assert( pMem->xDel!=0 && pMem->xDel!=SQLITE_DYNAMIC ); + pMem->xDel((void *)(pMem->z)); + } + + pMem->z = pMem->zMalloc; + pMem->flags &= ~(MEM_Dyn|MEM_Ephem|MEM_Static); + return SQLITE_OK; +} + +/* +** Change the pMem->zMalloc allocation to be at least szNew bytes. +** If pMem->zMalloc already meets or exceeds the requested size, this +** routine is a no-op. +** +** Any prior string or blob content in the pMem object may be discarded. +** The pMem->xDel destructor is called, if it exists. Though MEM_Str +** and MEM_Blob values may be discarded, MEM_Int, MEM_Real, MEM_IntReal, +** and MEM_Null values are preserved. +** +** Return SQLITE_OK on success or an error code (probably SQLITE_NOMEM) +** if unable to complete the resizing. +*/ +SQLITE_PRIVATE int sqlite3VdbeMemClearAndResize(Mem *pMem, int szNew){ + assert( CORRUPT_DB || szNew>0 ); + assert( (pMem->flags & MEM_Dyn)==0 || pMem->szMalloc==0 ); + if( pMem->szMallocflags & MEM_Dyn)==0 ); + pMem->z = pMem->zMalloc; + pMem->flags &= (MEM_Null|MEM_Int|MEM_Real|MEM_IntReal); + return SQLITE_OK; +} + +/* +** It is already known that pMem contains an unterminated string. +** Add the zero terminator. +** +** Three bytes of zero are added. In this way, there is guaranteed +** to be a double-zero byte at an even byte boundary in order to +** terminate a UTF16 string, even if the initial size of the buffer +** is an odd number of bytes. +*/ +static SQLITE_NOINLINE int vdbeMemAddTerminator(Mem *pMem){ + if( sqlite3VdbeMemGrow(pMem, pMem->n+3, 1) ){ + return SQLITE_NOMEM_BKPT; + } + pMem->z[pMem->n] = 0; + pMem->z[pMem->n+1] = 0; + pMem->z[pMem->n+2] = 0; + pMem->flags |= MEM_Term; + return SQLITE_OK; +} + +/* +** Change pMem so that its MEM_Str or MEM_Blob value is stored in +** MEM.zMalloc, where it can be safely written. +** +** Return SQLITE_OK on success or SQLITE_NOMEM if malloc fails. +*/ +SQLITE_PRIVATE int sqlite3VdbeMemMakeWriteable(Mem *pMem){ + assert( pMem->db==0 || sqlite3_mutex_held(pMem->db->mutex) ); + assert( !sqlite3VdbeMemIsRowSet(pMem) ); + if( (pMem->flags & (MEM_Str|MEM_Blob))!=0 ){ + if( ExpandBlob(pMem) ) return SQLITE_NOMEM; + if( pMem->szMalloc==0 || pMem->z!=pMem->zMalloc ){ + int rc = vdbeMemAddTerminator(pMem); + if( rc ) return rc; + } + } + pMem->flags &= ~MEM_Ephem; +#ifdef SQLITE_DEBUG + pMem->pScopyFrom = 0; +#endif + + return SQLITE_OK; +} + +/* +** If the given Mem* has a zero-filled tail, turn it into an ordinary +** blob stored in dynamically allocated space. +*/ +#ifndef SQLITE_OMIT_INCRBLOB +SQLITE_PRIVATE int sqlite3VdbeMemExpandBlob(Mem *pMem){ + int nByte; + assert( pMem->flags & MEM_Zero ); + assert( (pMem->flags&MEM_Blob)!=0 || MemNullNochng(pMem) ); + testcase( sqlite3_value_nochange(pMem) ); + assert( !sqlite3VdbeMemIsRowSet(pMem) ); + assert( pMem->db==0 || sqlite3_mutex_held(pMem->db->mutex) ); + + /* Set nByte to the number of bytes required to store the expanded blob. */ + nByte = pMem->n + pMem->u.nZero; + if( nByte<=0 ){ + if( (pMem->flags & MEM_Blob)==0 ) return SQLITE_OK; + nByte = 1; + } + if( sqlite3VdbeMemGrow(pMem, nByte, 1) ){ + return SQLITE_NOMEM_BKPT; + } + + memset(&pMem->z[pMem->n], 0, pMem->u.nZero); + pMem->n += pMem->u.nZero; + pMem->flags &= ~(MEM_Zero|MEM_Term); + return SQLITE_OK; +} +#endif + +/* +** Make sure the given Mem is \u0000 terminated. +*/ +SQLITE_PRIVATE int sqlite3VdbeMemNulTerminate(Mem *pMem){ + assert( pMem->db==0 || sqlite3_mutex_held(pMem->db->mutex) ); + testcase( (pMem->flags & (MEM_Term|MEM_Str))==(MEM_Term|MEM_Str) ); + testcase( (pMem->flags & (MEM_Term|MEM_Str))==0 ); + if( (pMem->flags & (MEM_Term|MEM_Str))!=MEM_Str ){ + return SQLITE_OK; /* Nothing to do */ + }else{ + return vdbeMemAddTerminator(pMem); + } +} + +/* +** Add MEM_Str to the set of representations for the given Mem. This +** routine is only called if pMem is a number of some kind, not a NULL +** or a BLOB. +** +** Existing representations MEM_Int, MEM_Real, or MEM_IntReal are invalidated +** if bForce is true but are retained if bForce is false. +** +** A MEM_Null value will never be passed to this function. This function is +** used for converting values to text for returning to the user (i.e. via +** sqlite3_value_text()), or for ensuring that values to be used as btree +** keys are strings. In the former case a NULL pointer is returned the +** user and the latter is an internal programming error. +*/ +SQLITE_PRIVATE int sqlite3VdbeMemStringify(Mem *pMem, u8 enc, u8 bForce){ + const int nByte = 32; + + assert( pMem->db==0 || sqlite3_mutex_held(pMem->db->mutex) ); + assert( !(pMem->flags&MEM_Zero) ); + assert( !(pMem->flags&(MEM_Str|MEM_Blob)) ); + assert( pMem->flags&(MEM_Int|MEM_Real|MEM_IntReal) ); + assert( !sqlite3VdbeMemIsRowSet(pMem) ); + assert( EIGHT_BYTE_ALIGNMENT(pMem) ); + + + if( sqlite3VdbeMemClearAndResize(pMem, nByte) ){ + pMem->enc = 0; + return SQLITE_NOMEM_BKPT; + } + + vdbeMemRenderNum(nByte, pMem->z, pMem); + assert( pMem->z!=0 ); + pMem->n = sqlite3Strlen30NN(pMem->z); + pMem->enc = SQLITE_UTF8; + pMem->flags |= MEM_Str|MEM_Term; + if( bForce ) pMem->flags &= ~(MEM_Int|MEM_Real|MEM_IntReal); + sqlite3VdbeChangeEncoding(pMem, enc); + return SQLITE_OK; +} + +/* +** Memory cell pMem contains the context of an aggregate function. +** This routine calls the finalize method for that function. The +** result of the aggregate is stored back into pMem. +** +** Return SQLITE_ERROR if the finalizer reports an error. SQLITE_OK +** otherwise. +*/ +SQLITE_PRIVATE int sqlite3VdbeMemFinalize(Mem *pMem, FuncDef *pFunc){ + sqlite3_context ctx; + Mem t; + assert( pFunc!=0 ); + assert( pFunc->xFinalize!=0 ); + assert( (pMem->flags & MEM_Null)!=0 || pFunc==pMem->u.pDef ); + assert( pMem->db==0 || sqlite3_mutex_held(pMem->db->mutex) ); + memset(&ctx, 0, sizeof(ctx)); + memset(&t, 0, sizeof(t)); + t.flags = MEM_Null; + t.db = pMem->db; + ctx.pOut = &t; + ctx.pMem = pMem; + ctx.pFunc = pFunc; + pFunc->xFinalize(&ctx); /* IMP: R-24505-23230 */ + assert( (pMem->flags & MEM_Dyn)==0 ); + if( pMem->szMalloc>0 ) sqlite3DbFreeNN(pMem->db, pMem->zMalloc); + memcpy(pMem, &t, sizeof(t)); + return ctx.isError; +} + +/* +** Memory cell pAccum contains the context of an aggregate function. +** This routine calls the xValue method for that function and stores +** the results in memory cell pMem. +** +** SQLITE_ERROR is returned if xValue() reports an error. SQLITE_OK +** otherwise. +*/ +#ifndef SQLITE_OMIT_WINDOWFUNC +SQLITE_PRIVATE int sqlite3VdbeMemAggValue(Mem *pAccum, Mem *pOut, FuncDef *pFunc){ + sqlite3_context ctx; + Mem t; + assert( pFunc!=0 ); + assert( pFunc->xValue!=0 ); + assert( (pAccum->flags & MEM_Null)!=0 || pFunc==pAccum->u.pDef ); + assert( pAccum->db==0 || sqlite3_mutex_held(pAccum->db->mutex) ); + memset(&ctx, 0, sizeof(ctx)); + memset(&t, 0, sizeof(t)); + t.flags = MEM_Null; + t.db = pAccum->db; + sqlite3VdbeMemSetNull(pOut); + ctx.pOut = pOut; + ctx.pMem = pAccum; + ctx.pFunc = pFunc; + pFunc->xValue(&ctx); + return ctx.isError; +} +#endif /* SQLITE_OMIT_WINDOWFUNC */ + +/* +** If the memory cell contains a value that must be freed by +** invoking the external callback in Mem.xDel, then this routine +** will free that value. It also sets Mem.flags to MEM_Null. +** +** This is a helper routine for sqlite3VdbeMemSetNull() and +** for sqlite3VdbeMemRelease(). Use those other routines as the +** entry point for releasing Mem resources. +*/ +static SQLITE_NOINLINE void vdbeMemClearExternAndSetNull(Mem *p){ + assert( p->db==0 || sqlite3_mutex_held(p->db->mutex) ); + assert( VdbeMemDynamic(p) ); + if( p->flags&MEM_Agg ){ + sqlite3VdbeMemFinalize(p, p->u.pDef); + assert( (p->flags & MEM_Agg)==0 ); + testcase( p->flags & MEM_Dyn ); + } + if( p->flags&MEM_Dyn ){ + assert( p->xDel!=SQLITE_DYNAMIC && p->xDel!=0 ); + p->xDel((void *)p->z); + } + p->flags = MEM_Null; +} + +/* +** Release memory held by the Mem p, both external memory cleared +** by p->xDel and memory in p->zMalloc. +** +** This is a helper routine invoked by sqlite3VdbeMemRelease() in +** the unusual case where there really is memory in p that needs +** to be freed. +*/ +static SQLITE_NOINLINE void vdbeMemClear(Mem *p){ + if( VdbeMemDynamic(p) ){ + vdbeMemClearExternAndSetNull(p); + } + if( p->szMalloc ){ + sqlite3DbFreeNN(p->db, p->zMalloc); + p->szMalloc = 0; + } + p->z = 0; +} + +/* +** Release any memory resources held by the Mem. Both the memory that is +** free by Mem.xDel and the Mem.zMalloc allocation are freed. +** +** Use this routine prior to clean up prior to abandoning a Mem, or to +** reset a Mem back to its minimum memory utilization. +** +** Use sqlite3VdbeMemSetNull() to release just the Mem.xDel space +** prior to inserting new content into the Mem. +*/ +SQLITE_PRIVATE void sqlite3VdbeMemRelease(Mem *p){ + assert( sqlite3VdbeCheckMemInvariants(p) ); + if( VdbeMemDynamic(p) || p->szMalloc ){ + vdbeMemClear(p); + } +} + +/* +** Convert a 64-bit IEEE double into a 64-bit signed integer. +** If the double is out of range of a 64-bit signed integer then +** return the closest available 64-bit signed integer. +*/ +static SQLITE_NOINLINE i64 doubleToInt64(double r){ +#ifdef SQLITE_OMIT_FLOATING_POINT + /* When floating-point is omitted, double and int64 are the same thing */ + return r; +#else + /* + ** Many compilers we encounter do not define constants for the + ** minimum and maximum 64-bit integers, or they define them + ** inconsistently. And many do not understand the "LL" notation. + ** So we define our own static constants here using nothing + ** larger than a 32-bit integer constant. + */ + static const i64 maxInt = LARGEST_INT64; + static const i64 minInt = SMALLEST_INT64; + + if( r<=(double)minInt ){ + return minInt; + }else if( r>=(double)maxInt ){ + return maxInt; + }else{ + return (i64)r; + } +#endif +} + +/* +** Return some kind of integer value which is the best we can do +** at representing the value that *pMem describes as an integer. +** If pMem is an integer, then the value is exact. If pMem is +** a floating-point then the value returned is the integer part. +** If pMem is a string or blob, then we make an attempt to convert +** it into an integer and return that. If pMem represents an +** an SQL-NULL value, return 0. +** +** If pMem represents a string value, its encoding might be changed. +*/ +static SQLITE_NOINLINE i64 memIntValue(Mem *pMem){ + i64 value = 0; + sqlite3Atoi64(pMem->z, &value, pMem->n, pMem->enc); + return value; +} +SQLITE_PRIVATE i64 sqlite3VdbeIntValue(Mem *pMem){ + int flags; + assert( pMem->db==0 || sqlite3_mutex_held(pMem->db->mutex) ); + assert( EIGHT_BYTE_ALIGNMENT(pMem) ); + flags = pMem->flags; + if( flags & (MEM_Int|MEM_IntReal) ){ + testcase( flags & MEM_IntReal ); + return pMem->u.i; + }else if( flags & MEM_Real ){ + return doubleToInt64(pMem->u.r); + }else if( flags & (MEM_Str|MEM_Blob) ){ + assert( pMem->z || pMem->n==0 ); + return memIntValue(pMem); + }else{ + return 0; + } +} + +/* +** Return the best representation of pMem that we can get into a +** double. If pMem is already a double or an integer, return its +** value. If it is a string or blob, try to convert it to a double. +** If it is a NULL, return 0.0. +*/ +static SQLITE_NOINLINE double memRealValue(Mem *pMem){ + /* (double)0 In case of SQLITE_OMIT_FLOATING_POINT... */ + double val = (double)0; + sqlite3AtoF(pMem->z, &val, pMem->n, pMem->enc); + return val; +} +SQLITE_PRIVATE double sqlite3VdbeRealValue(Mem *pMem){ + assert( pMem->db==0 || sqlite3_mutex_held(pMem->db->mutex) ); + assert( EIGHT_BYTE_ALIGNMENT(pMem) ); + if( pMem->flags & MEM_Real ){ + return pMem->u.r; + }else if( pMem->flags & (MEM_Int|MEM_IntReal) ){ + testcase( pMem->flags & MEM_IntReal ); + return (double)pMem->u.i; + }else if( pMem->flags & (MEM_Str|MEM_Blob) ){ + return memRealValue(pMem); + }else{ + /* (double)0 In case of SQLITE_OMIT_FLOATING_POINT... */ + return (double)0; + } +} + +/* +** Return 1 if pMem represents true, and return 0 if pMem represents false. +** Return the value ifNull if pMem is NULL. +*/ +SQLITE_PRIVATE int sqlite3VdbeBooleanValue(Mem *pMem, int ifNull){ + testcase( pMem->flags & MEM_IntReal ); + if( pMem->flags & (MEM_Int|MEM_IntReal) ) return pMem->u.i!=0; + if( pMem->flags & MEM_Null ) return ifNull; + return sqlite3VdbeRealValue(pMem)!=0.0; +} + +/* +** The MEM structure is already a MEM_Real. Try to also make it a +** MEM_Int if we can. +*/ +SQLITE_PRIVATE void sqlite3VdbeIntegerAffinity(Mem *pMem){ + i64 ix; + assert( pMem->flags & MEM_Real ); + assert( !sqlite3VdbeMemIsRowSet(pMem) ); + assert( pMem->db==0 || sqlite3_mutex_held(pMem->db->mutex) ); + assert( EIGHT_BYTE_ALIGNMENT(pMem) ); + + ix = doubleToInt64(pMem->u.r); + + /* Only mark the value as an integer if + ** + ** (1) the round-trip conversion real->int->real is a no-op, and + ** (2) The integer is neither the largest nor the smallest + ** possible integer (ticket #3922) + ** + ** The second and third terms in the following conditional enforces + ** the second condition under the assumption that addition overflow causes + ** values to wrap around. + */ + if( pMem->u.r==ix && ix>SMALLEST_INT64 && ixu.i = ix; + MemSetTypeFlag(pMem, MEM_Int); + } +} + +/* +** Convert pMem to type integer. Invalidate any prior representations. +*/ +SQLITE_PRIVATE int sqlite3VdbeMemIntegerify(Mem *pMem){ + assert( pMem->db==0 || sqlite3_mutex_held(pMem->db->mutex) ); + assert( !sqlite3VdbeMemIsRowSet(pMem) ); + assert( EIGHT_BYTE_ALIGNMENT(pMem) ); + + pMem->u.i = sqlite3VdbeIntValue(pMem); + MemSetTypeFlag(pMem, MEM_Int); + return SQLITE_OK; +} + +/* +** Convert pMem so that it is of type MEM_Real. +** Invalidate any prior representations. +*/ +SQLITE_PRIVATE int sqlite3VdbeMemRealify(Mem *pMem){ + assert( pMem->db==0 || sqlite3_mutex_held(pMem->db->mutex) ); + assert( EIGHT_BYTE_ALIGNMENT(pMem) ); + + pMem->u.r = sqlite3VdbeRealValue(pMem); + MemSetTypeFlag(pMem, MEM_Real); + return SQLITE_OK; +} + +/* Compare a floating point value to an integer. Return true if the two +** values are the same within the precision of the floating point value. +** +** This function assumes that i was obtained by assignment from r1. +** +** For some versions of GCC on 32-bit machines, if you do the more obvious +** comparison of "r1==(double)i" you sometimes get an answer of false even +** though the r1 and (double)i values are bit-for-bit the same. +*/ +SQLITE_PRIVATE int sqlite3RealSameAsInt(double r1, sqlite3_int64 i){ + double r2 = (double)i; + return r1==0.0 + || (memcmp(&r1, &r2, sizeof(r1))==0 + && i >= -2251799813685248LL && i < 2251799813685248LL); +} + +/* +** Convert pMem so that it has type MEM_Real or MEM_Int. +** Invalidate any prior representations. +** +** Every effort is made to force the conversion, even if the input +** is a string that does not look completely like a number. Convert +** as much of the string as we can and ignore the rest. +*/ +SQLITE_PRIVATE int sqlite3VdbeMemNumerify(Mem *pMem){ + testcase( pMem->flags & MEM_Int ); + testcase( pMem->flags & MEM_Real ); + testcase( pMem->flags & MEM_IntReal ); + testcase( pMem->flags & MEM_Null ); + if( (pMem->flags & (MEM_Int|MEM_Real|MEM_IntReal|MEM_Null))==0 ){ + int rc; + sqlite3_int64 ix; + assert( (pMem->flags & (MEM_Blob|MEM_Str))!=0 ); + assert( pMem->db==0 || sqlite3_mutex_held(pMem->db->mutex) ); + rc = sqlite3AtoF(pMem->z, &pMem->u.r, pMem->n, pMem->enc); + if( ((rc==0 || rc==1) && sqlite3Atoi64(pMem->z, &ix, pMem->n, pMem->enc)<=1) + || sqlite3RealSameAsInt(pMem->u.r, (ix = (i64)pMem->u.r)) + ){ + pMem->u.i = ix; + MemSetTypeFlag(pMem, MEM_Int); + }else{ + MemSetTypeFlag(pMem, MEM_Real); + } + } + assert( (pMem->flags & (MEM_Int|MEM_Real|MEM_IntReal|MEM_Null))!=0 ); + pMem->flags &= ~(MEM_Str|MEM_Blob|MEM_Zero); + return SQLITE_OK; +} + +/* +** Cast the datatype of the value in pMem according to the affinity +** "aff". Casting is different from applying affinity in that a cast +** is forced. In other words, the value is converted into the desired +** affinity even if that results in loss of data. This routine is +** used (for example) to implement the SQL "cast()" operator. +*/ +SQLITE_PRIVATE void sqlite3VdbeMemCast(Mem *pMem, u8 aff, u8 encoding){ + if( pMem->flags & MEM_Null ) return; + switch( aff ){ + case SQLITE_AFF_BLOB: { /* Really a cast to BLOB */ + if( (pMem->flags & MEM_Blob)==0 ){ + sqlite3ValueApplyAffinity(pMem, SQLITE_AFF_TEXT, encoding); + assert( pMem->flags & MEM_Str || pMem->db->mallocFailed ); + if( pMem->flags & MEM_Str ) MemSetTypeFlag(pMem, MEM_Blob); + }else{ + pMem->flags &= ~(MEM_TypeMask&~MEM_Blob); + } + break; + } + case SQLITE_AFF_NUMERIC: { + sqlite3VdbeMemNumerify(pMem); + break; + } + case SQLITE_AFF_INTEGER: { + sqlite3VdbeMemIntegerify(pMem); + break; + } + case SQLITE_AFF_REAL: { + sqlite3VdbeMemRealify(pMem); + break; + } + default: { + assert( aff==SQLITE_AFF_TEXT ); + assert( MEM_Str==(MEM_Blob>>3) ); + pMem->flags |= (pMem->flags&MEM_Blob)>>3; + sqlite3ValueApplyAffinity(pMem, SQLITE_AFF_TEXT, encoding); + assert( pMem->flags & MEM_Str || pMem->db->mallocFailed ); + pMem->flags &= ~(MEM_Int|MEM_Real|MEM_IntReal|MEM_Blob|MEM_Zero); + break; + } + } +} + +/* +** Initialize bulk memory to be a consistent Mem object. +** +** The minimum amount of initialization feasible is performed. +*/ +SQLITE_PRIVATE void sqlite3VdbeMemInit(Mem *pMem, sqlite3 *db, u16 flags){ + assert( (flags & ~MEM_TypeMask)==0 ); + pMem->flags = flags; + pMem->db = db; + pMem->szMalloc = 0; +} + + +/* +** Delete any previous value and set the value stored in *pMem to NULL. +** +** This routine calls the Mem.xDel destructor to dispose of values that +** require the destructor. But it preserves the Mem.zMalloc memory allocation. +** To free all resources, use sqlite3VdbeMemRelease(), which both calls this +** routine to invoke the destructor and deallocates Mem.zMalloc. +** +** Use this routine to reset the Mem prior to insert a new value. +** +** Use sqlite3VdbeMemRelease() to complete erase the Mem prior to abandoning it. +*/ +SQLITE_PRIVATE void sqlite3VdbeMemSetNull(Mem *pMem){ + if( VdbeMemDynamic(pMem) ){ + vdbeMemClearExternAndSetNull(pMem); + }else{ + pMem->flags = MEM_Null; + } +} +SQLITE_PRIVATE void sqlite3ValueSetNull(sqlite3_value *p){ + sqlite3VdbeMemSetNull((Mem*)p); +} + +/* +** Delete any previous value and set the value to be a BLOB of length +** n containing all zeros. +*/ +SQLITE_PRIVATE void sqlite3VdbeMemSetZeroBlob(Mem *pMem, int n){ + sqlite3VdbeMemRelease(pMem); + pMem->flags = MEM_Blob|MEM_Zero; + pMem->n = 0; + if( n<0 ) n = 0; + pMem->u.nZero = n; + pMem->enc = SQLITE_UTF8; + pMem->z = 0; +} + +/* +** The pMem is known to contain content that needs to be destroyed prior +** to a value change. So invoke the destructor, then set the value to +** a 64-bit integer. +*/ +static SQLITE_NOINLINE void vdbeReleaseAndSetInt64(Mem *pMem, i64 val){ + sqlite3VdbeMemSetNull(pMem); + pMem->u.i = val; + pMem->flags = MEM_Int; +} + +/* +** Delete any previous value and set the value stored in *pMem to val, +** manifest type INTEGER. +*/ +SQLITE_PRIVATE void sqlite3VdbeMemSetInt64(Mem *pMem, i64 val){ + if( VdbeMemDynamic(pMem) ){ + vdbeReleaseAndSetInt64(pMem, val); + }else{ + pMem->u.i = val; + pMem->flags = MEM_Int; + } +} + +/* A no-op destructor */ +SQLITE_PRIVATE void sqlite3NoopDestructor(void *p){ UNUSED_PARAMETER(p); } + +/* +** Set the value stored in *pMem should already be a NULL. +** Also store a pointer to go with it. +*/ +SQLITE_PRIVATE void sqlite3VdbeMemSetPointer( + Mem *pMem, + void *pPtr, + const char *zPType, + void (*xDestructor)(void*) +){ + assert( pMem->flags==MEM_Null ); + pMem->u.zPType = zPType ? zPType : ""; + pMem->z = pPtr; + pMem->flags = MEM_Null|MEM_Dyn|MEM_Subtype|MEM_Term; + pMem->eSubtype = 'p'; + pMem->xDel = xDestructor ? xDestructor : sqlite3NoopDestructor; +} + +#ifndef SQLITE_OMIT_FLOATING_POINT +/* +** Delete any previous value and set the value stored in *pMem to val, +** manifest type REAL. +*/ +SQLITE_PRIVATE void sqlite3VdbeMemSetDouble(Mem *pMem, double val){ + sqlite3VdbeMemSetNull(pMem); + if( !sqlite3IsNaN(val) ){ + pMem->u.r = val; + pMem->flags = MEM_Real; + } +} +#endif + +#ifdef SQLITE_DEBUG +/* +** Return true if the Mem holds a RowSet object. This routine is intended +** for use inside of assert() statements. +*/ +SQLITE_PRIVATE int sqlite3VdbeMemIsRowSet(const Mem *pMem){ + return (pMem->flags&(MEM_Blob|MEM_Dyn))==(MEM_Blob|MEM_Dyn) + && pMem->xDel==sqlite3RowSetDelete; +} +#endif + +/* +** Delete any previous value and set the value of pMem to be an +** empty boolean index. +** +** Return SQLITE_OK on success and SQLITE_NOMEM if a memory allocation +** error occurs. +*/ +SQLITE_PRIVATE int sqlite3VdbeMemSetRowSet(Mem *pMem){ + sqlite3 *db = pMem->db; + RowSet *p; + assert( db!=0 ); + assert( !sqlite3VdbeMemIsRowSet(pMem) ); + sqlite3VdbeMemRelease(pMem); + p = sqlite3RowSetInit(db); + if( p==0 ) return SQLITE_NOMEM; + pMem->z = (char*)p; + pMem->flags = MEM_Blob|MEM_Dyn; + pMem->xDel = sqlite3RowSetDelete; + return SQLITE_OK; +} + +/* +** Return true if the Mem object contains a TEXT or BLOB that is +** too large - whose size exceeds SQLITE_MAX_LENGTH. +*/ +SQLITE_PRIVATE int sqlite3VdbeMemTooBig(Mem *p){ + assert( p->db!=0 ); + if( p->flags & (MEM_Str|MEM_Blob) ){ + int n = p->n; + if( p->flags & MEM_Zero ){ + n += p->u.nZero; + } + return n>p->db->aLimit[SQLITE_LIMIT_LENGTH]; + } + return 0; +} + +#ifdef SQLITE_DEBUG +/* +** This routine prepares a memory cell for modification by breaking +** its link to a shallow copy and by marking any current shallow +** copies of this cell as invalid. +** +** This is used for testing and debugging only - to make sure shallow +** copies are not misused. +*/ +SQLITE_PRIVATE void sqlite3VdbeMemAboutToChange(Vdbe *pVdbe, Mem *pMem){ + int i; + Mem *pX; + for(i=0, pX=pVdbe->aMem; inMem; i++, pX++){ + if( pX->pScopyFrom==pMem ){ + /* If pX is marked as a shallow copy of pMem, then verify that + ** no significant changes have been made to pX since the OP_SCopy. + ** A significant change would indicated a missed call to this + ** function for pX. Minor changes, such as adding or removing a + ** dual type, are allowed, as long as the underlying value is the + ** same. */ + u16 mFlags = pMem->flags & pX->flags & pX->mScopyFlags; + assert( (mFlags&(MEM_Int|MEM_IntReal))==0 || pMem->u.i==pX->u.i ); + assert( (mFlags&MEM_Real)==0 || pMem->u.r==pX->u.r ); + assert( (mFlags&MEM_Str)==0 || (pMem->n==pX->n && pMem->z==pX->z) ); + assert( (mFlags&MEM_Blob)==0 || sqlite3BlobCompare(pMem,pX)==0 ); + + /* pMem is the register that is changing. But also mark pX as + ** undefined so that we can quickly detect the shallow-copy error */ + pX->flags = MEM_Undefined; + pX->pScopyFrom = 0; + } + } + pMem->pScopyFrom = 0; +} +#endif /* SQLITE_DEBUG */ + + +/* +** Make an shallow copy of pFrom into pTo. Prior contents of +** pTo are freed. The pFrom->z field is not duplicated. If +** pFrom->z is used, then pTo->z points to the same thing as pFrom->z +** and flags gets srcType (either MEM_Ephem or MEM_Static). +*/ +static SQLITE_NOINLINE void vdbeClrCopy(Mem *pTo, const Mem *pFrom, int eType){ + vdbeMemClearExternAndSetNull(pTo); + assert( !VdbeMemDynamic(pTo) ); + sqlite3VdbeMemShallowCopy(pTo, pFrom, eType); +} +SQLITE_PRIVATE void sqlite3VdbeMemShallowCopy(Mem *pTo, const Mem *pFrom, int srcType){ + assert( !sqlite3VdbeMemIsRowSet(pFrom) ); + assert( pTo->db==pFrom->db ); + if( VdbeMemDynamic(pTo) ){ vdbeClrCopy(pTo,pFrom,srcType); return; } + memcpy(pTo, pFrom, MEMCELLSIZE); + if( (pFrom->flags&MEM_Static)==0 ){ + pTo->flags &= ~(MEM_Dyn|MEM_Static|MEM_Ephem); + assert( srcType==MEM_Ephem || srcType==MEM_Static ); + pTo->flags |= srcType; + } +} + +/* +** Make a full copy of pFrom into pTo. Prior contents of pTo are +** freed before the copy is made. +*/ +SQLITE_PRIVATE int sqlite3VdbeMemCopy(Mem *pTo, const Mem *pFrom){ + int rc = SQLITE_OK; + + assert( !sqlite3VdbeMemIsRowSet(pFrom) ); + if( VdbeMemDynamic(pTo) ) vdbeMemClearExternAndSetNull(pTo); + memcpy(pTo, pFrom, MEMCELLSIZE); + pTo->flags &= ~MEM_Dyn; + if( pTo->flags&(MEM_Str|MEM_Blob) ){ + if( 0==(pFrom->flags&MEM_Static) ){ + pTo->flags |= MEM_Ephem; + rc = sqlite3VdbeMemMakeWriteable(pTo); + } + } + + return rc; +} + +/* +** Transfer the contents of pFrom to pTo. Any existing value in pTo is +** freed. If pFrom contains ephemeral data, a copy is made. +** +** pFrom contains an SQL NULL when this routine returns. +*/ +SQLITE_PRIVATE void sqlite3VdbeMemMove(Mem *pTo, Mem *pFrom){ + assert( pFrom->db==0 || sqlite3_mutex_held(pFrom->db->mutex) ); + assert( pTo->db==0 || sqlite3_mutex_held(pTo->db->mutex) ); + assert( pFrom->db==0 || pTo->db==0 || pFrom->db==pTo->db ); + + sqlite3VdbeMemRelease(pTo); + memcpy(pTo, pFrom, sizeof(Mem)); + pFrom->flags = MEM_Null; + pFrom->szMalloc = 0; +} + +/* +** Change the value of a Mem to be a string or a BLOB. +** +** The memory management strategy depends on the value of the xDel +** parameter. If the value passed is SQLITE_TRANSIENT, then the +** string is copied into a (possibly existing) buffer managed by the +** Mem structure. Otherwise, any existing buffer is freed and the +** pointer copied. +** +** If the string is too large (if it exceeds the SQLITE_LIMIT_LENGTH +** size limit) then no memory allocation occurs. If the string can be +** stored without allocating memory, then it is. If a memory allocation +** is required to store the string, then value of pMem is unchanged. In +** either case, SQLITE_TOOBIG is returned. +*/ +SQLITE_PRIVATE int sqlite3VdbeMemSetStr( + Mem *pMem, /* Memory cell to set to string value */ + const char *z, /* String pointer */ + int n, /* Bytes in string, or negative */ + u8 enc, /* Encoding of z. 0 for BLOBs */ + void (*xDel)(void*) /* Destructor function */ +){ + int nByte = n; /* New value for pMem->n */ + int iLimit; /* Maximum allowed string or blob size */ + u16 flags = 0; /* New value for pMem->flags */ + + assert( pMem->db==0 || sqlite3_mutex_held(pMem->db->mutex) ); + assert( !sqlite3VdbeMemIsRowSet(pMem) ); + + /* If z is a NULL pointer, set pMem to contain an SQL NULL. */ + if( !z ){ + sqlite3VdbeMemSetNull(pMem); + return SQLITE_OK; + } + + if( pMem->db ){ + iLimit = pMem->db->aLimit[SQLITE_LIMIT_LENGTH]; + }else{ + iLimit = SQLITE_MAX_LENGTH; + } + flags = (enc==0?MEM_Blob:MEM_Str); + if( nByte<0 ){ + assert( enc!=0 ); + if( enc==SQLITE_UTF8 ){ + nByte = 0x7fffffff & (int)strlen(z); + }else{ + for(nByte=0; nByte<=iLimit && (z[nByte] | z[nByte+1]); nByte+=2){} + } + flags |= MEM_Term; + } + + /* The following block sets the new values of Mem.z and Mem.xDel. It + ** also sets a flag in local variable "flags" to indicate the memory + ** management (one of MEM_Dyn or MEM_Static). + */ + if( xDel==SQLITE_TRANSIENT ){ + u32 nAlloc = nByte; + if( flags&MEM_Term ){ + nAlloc += (enc==SQLITE_UTF8?1:2); + } + if( nByte>iLimit ){ + return sqlite3ErrorToParser(pMem->db, SQLITE_TOOBIG); + } + testcase( nAlloc==0 ); + testcase( nAlloc==31 ); + testcase( nAlloc==32 ); + if( sqlite3VdbeMemClearAndResize(pMem, (int)MAX(nAlloc,32)) ){ + return SQLITE_NOMEM_BKPT; + } + memcpy(pMem->z, z, nAlloc); + }else{ + sqlite3VdbeMemRelease(pMem); + pMem->z = (char *)z; + if( xDel==SQLITE_DYNAMIC ){ + pMem->zMalloc = pMem->z; + pMem->szMalloc = sqlite3DbMallocSize(pMem->db, pMem->zMalloc); + }else{ + pMem->xDel = xDel; + flags |= ((xDel==SQLITE_STATIC)?MEM_Static:MEM_Dyn); + } + } + + pMem->n = nByte; + pMem->flags = flags; + pMem->enc = (enc==0 ? SQLITE_UTF8 : enc); + +#ifndef SQLITE_OMIT_UTF16 + if( pMem->enc!=SQLITE_UTF8 && sqlite3VdbeMemHandleBom(pMem) ){ + return SQLITE_NOMEM_BKPT; + } +#endif + + if( nByte>iLimit ){ + return SQLITE_TOOBIG; + } + + return SQLITE_OK; +} + +/* +** Move data out of a btree key or data field and into a Mem structure. +** The data is payload from the entry that pCur is currently pointing +** to. offset and amt determine what portion of the data or key to retrieve. +** The result is written into the pMem element. +** +** The pMem object must have been initialized. This routine will use +** pMem->zMalloc to hold the content from the btree, if possible. New +** pMem->zMalloc space will be allocated if necessary. The calling routine +** is responsible for making sure that the pMem object is eventually +** destroyed. +** +** If this routine fails for any reason (malloc returns NULL or unable +** to read from the disk) then the pMem is left in an inconsistent state. +*/ +static SQLITE_NOINLINE int vdbeMemFromBtreeResize( + BtCursor *pCur, /* Cursor pointing at record to retrieve. */ + u32 offset, /* Offset from the start of data to return bytes from. */ + u32 amt, /* Number of bytes to return. */ + Mem *pMem /* OUT: Return data in this Mem structure. */ +){ + int rc; + pMem->flags = MEM_Null; + if( sqlite3BtreeMaxRecordSize(pCur)z); + if( rc==SQLITE_OK ){ + pMem->z[amt] = 0; /* Overrun area used when reading malformed records */ + pMem->flags = MEM_Blob; + pMem->n = (int)amt; + }else{ + sqlite3VdbeMemRelease(pMem); + } + } + return rc; +} +SQLITE_PRIVATE int sqlite3VdbeMemFromBtree( + BtCursor *pCur, /* Cursor pointing at record to retrieve. */ + u32 offset, /* Offset from the start of data to return bytes from. */ + u32 amt, /* Number of bytes to return. */ + Mem *pMem /* OUT: Return data in this Mem structure. */ +){ + char *zData; /* Data from the btree layer */ + u32 available = 0; /* Number of bytes available on the local btree page */ + int rc = SQLITE_OK; /* Return code */ + + assert( sqlite3BtreeCursorIsValid(pCur) ); + assert( !VdbeMemDynamic(pMem) ); + + /* Note: the calls to BtreeKeyFetch() and DataFetch() below assert() + ** that both the BtShared and database handle mutexes are held. */ + assert( !sqlite3VdbeMemIsRowSet(pMem) ); + zData = (char *)sqlite3BtreePayloadFetch(pCur, &available); + assert( zData!=0 ); + + if( offset+amt<=available ){ + pMem->z = &zData[offset]; + pMem->flags = MEM_Blob|MEM_Ephem; + pMem->n = (int)amt; + }else{ + rc = vdbeMemFromBtreeResize(pCur, offset, amt, pMem); + } + + return rc; +} + +/* +** The pVal argument is known to be a value other than NULL. +** Convert it into a string with encoding enc and return a pointer +** to a zero-terminated version of that string. +*/ +static SQLITE_NOINLINE const void *valueToText(sqlite3_value* pVal, u8 enc){ + assert( pVal!=0 ); + assert( pVal->db==0 || sqlite3_mutex_held(pVal->db->mutex) ); + assert( (enc&3)==(enc&~SQLITE_UTF16_ALIGNED) ); + assert( !sqlite3VdbeMemIsRowSet(pVal) ); + assert( (pVal->flags & (MEM_Null))==0 ); + if( pVal->flags & (MEM_Blob|MEM_Str) ){ + if( ExpandBlob(pVal) ) return 0; + pVal->flags |= MEM_Str; + if( pVal->enc != (enc & ~SQLITE_UTF16_ALIGNED) ){ + sqlite3VdbeChangeEncoding(pVal, enc & ~SQLITE_UTF16_ALIGNED); + } + if( (enc & SQLITE_UTF16_ALIGNED)!=0 && 1==(1&SQLITE_PTR_TO_INT(pVal->z)) ){ + assert( (pVal->flags & (MEM_Ephem|MEM_Static))!=0 ); + if( sqlite3VdbeMemMakeWriteable(pVal)!=SQLITE_OK ){ + return 0; + } + } + sqlite3VdbeMemNulTerminate(pVal); /* IMP: R-31275-44060 */ + }else{ + sqlite3VdbeMemStringify(pVal, enc, 0); + assert( 0==(1&SQLITE_PTR_TO_INT(pVal->z)) ); + } + assert(pVal->enc==(enc & ~SQLITE_UTF16_ALIGNED) || pVal->db==0 + || pVal->db->mallocFailed ); + if( pVal->enc==(enc & ~SQLITE_UTF16_ALIGNED) ){ + assert( sqlite3VdbeMemValidStrRep(pVal) ); + return pVal->z; + }else{ + return 0; + } +} + +/* This function is only available internally, it is not part of the +** external API. It works in a similar way to sqlite3_value_text(), +** except the data returned is in the encoding specified by the second +** parameter, which must be one of SQLITE_UTF16BE, SQLITE_UTF16LE or +** SQLITE_UTF8. +** +** (2006-02-16:) The enc value can be or-ed with SQLITE_UTF16_ALIGNED. +** If that is the case, then the result must be aligned on an even byte +** boundary. +*/ +SQLITE_PRIVATE const void *sqlite3ValueText(sqlite3_value* pVal, u8 enc){ + if( !pVal ) return 0; + assert( pVal->db==0 || sqlite3_mutex_held(pVal->db->mutex) ); + assert( (enc&3)==(enc&~SQLITE_UTF16_ALIGNED) ); + assert( !sqlite3VdbeMemIsRowSet(pVal) ); + if( (pVal->flags&(MEM_Str|MEM_Term))==(MEM_Str|MEM_Term) && pVal->enc==enc ){ + assert( sqlite3VdbeMemValidStrRep(pVal) ); + return pVal->z; + } + if( pVal->flags&MEM_Null ){ + return 0; + } + return valueToText(pVal, enc); +} + +/* +** Create a new sqlite3_value object. +*/ +SQLITE_PRIVATE sqlite3_value *sqlite3ValueNew(sqlite3 *db){ + Mem *p = sqlite3DbMallocZero(db, sizeof(*p)); + if( p ){ + p->flags = MEM_Null; + p->db = db; + } + return p; +} + +/* +** Context object passed by sqlite3Stat4ProbeSetValue() through to +** valueNew(). See comments above valueNew() for details. +*/ +struct ValueNewStat4Ctx { + Parse *pParse; + Index *pIdx; + UnpackedRecord **ppRec; + int iVal; +}; + +/* +** Allocate and return a pointer to a new sqlite3_value object. If +** the second argument to this function is NULL, the object is allocated +** by calling sqlite3ValueNew(). +** +** Otherwise, if the second argument is non-zero, then this function is +** being called indirectly by sqlite3Stat4ProbeSetValue(). If it has not +** already been allocated, allocate the UnpackedRecord structure that +** that function will return to its caller here. Then return a pointer to +** an sqlite3_value within the UnpackedRecord.a[] array. +*/ +static sqlite3_value *valueNew(sqlite3 *db, struct ValueNewStat4Ctx *p){ +#ifdef SQLITE_ENABLE_STAT4 + if( p ){ + UnpackedRecord *pRec = p->ppRec[0]; + + if( pRec==0 ){ + Index *pIdx = p->pIdx; /* Index being probed */ + int nByte; /* Bytes of space to allocate */ + int i; /* Counter variable */ + int nCol = pIdx->nColumn; /* Number of index columns including rowid */ + + nByte = sizeof(Mem) * nCol + ROUND8(sizeof(UnpackedRecord)); + pRec = (UnpackedRecord*)sqlite3DbMallocZero(db, nByte); + if( pRec ){ + pRec->pKeyInfo = sqlite3KeyInfoOfIndex(p->pParse, pIdx); + if( pRec->pKeyInfo ){ + assert( pRec->pKeyInfo->nAllField==nCol ); + assert( pRec->pKeyInfo->enc==ENC(db) ); + pRec->aMem = (Mem *)((u8*)pRec + ROUND8(sizeof(UnpackedRecord))); + for(i=0; iaMem[i].flags = MEM_Null; + pRec->aMem[i].db = db; + } + }else{ + sqlite3DbFreeNN(db, pRec); + pRec = 0; + } + } + if( pRec==0 ) return 0; + p->ppRec[0] = pRec; + } + + pRec->nField = p->iVal+1; + return &pRec->aMem[p->iVal]; + } +#else + UNUSED_PARAMETER(p); +#endif /* defined(SQLITE_ENABLE_STAT4) */ + return sqlite3ValueNew(db); +} + +/* +** The expression object indicated by the second argument is guaranteed +** to be a scalar SQL function. If +** +** * all function arguments are SQL literals, +** * one of the SQLITE_FUNC_CONSTANT or _SLOCHNG function flags is set, and +** * the SQLITE_FUNC_NEEDCOLL function flag is not set, +** +** then this routine attempts to invoke the SQL function. Assuming no +** error occurs, output parameter (*ppVal) is set to point to a value +** object containing the result before returning SQLITE_OK. +** +** Affinity aff is applied to the result of the function before returning. +** If the result is a text value, the sqlite3_value object uses encoding +** enc. +** +** If the conditions above are not met, this function returns SQLITE_OK +** and sets (*ppVal) to NULL. Or, if an error occurs, (*ppVal) is set to +** NULL and an SQLite error code returned. +*/ +#ifdef SQLITE_ENABLE_STAT4 +static int valueFromFunction( + sqlite3 *db, /* The database connection */ + Expr *p, /* The expression to evaluate */ + u8 enc, /* Encoding to use */ + u8 aff, /* Affinity to use */ + sqlite3_value **ppVal, /* Write the new value here */ + struct ValueNewStat4Ctx *pCtx /* Second argument for valueNew() */ +){ + sqlite3_context ctx; /* Context object for function invocation */ + sqlite3_value **apVal = 0; /* Function arguments */ + int nVal = 0; /* Size of apVal[] array */ + FuncDef *pFunc = 0; /* Function definition */ + sqlite3_value *pVal = 0; /* New value */ + int rc = SQLITE_OK; /* Return code */ + ExprList *pList = 0; /* Function arguments */ + int i; /* Iterator variable */ + + assert( pCtx!=0 ); + assert( (p->flags & EP_TokenOnly)==0 ); + pList = p->x.pList; + if( pList ) nVal = pList->nExpr; + pFunc = sqlite3FindFunction(db, p->u.zToken, nVal, enc, 0); + assert( pFunc ); + if( (pFunc->funcFlags & (SQLITE_FUNC_CONSTANT|SQLITE_FUNC_SLOCHNG))==0 + || (pFunc->funcFlags & SQLITE_FUNC_NEEDCOLL) + ){ + return SQLITE_OK; + } + + if( pList ){ + apVal = (sqlite3_value**)sqlite3DbMallocZero(db, sizeof(apVal[0]) * nVal); + if( apVal==0 ){ + rc = SQLITE_NOMEM_BKPT; + goto value_from_function_out; + } + for(i=0; ia[i].pExpr, enc, aff, &apVal[i]); + if( apVal[i]==0 || rc!=SQLITE_OK ) goto value_from_function_out; + } + } + + pVal = valueNew(db, pCtx); + if( pVal==0 ){ + rc = SQLITE_NOMEM_BKPT; + goto value_from_function_out; + } + + assert( pCtx->pParse->rc==SQLITE_OK ); + memset(&ctx, 0, sizeof(ctx)); + ctx.pOut = pVal; + ctx.pFunc = pFunc; + pFunc->xSFunc(&ctx, nVal, apVal); + if( ctx.isError ){ + rc = ctx.isError; + sqlite3ErrorMsg(pCtx->pParse, "%s", sqlite3_value_text(pVal)); + }else{ + sqlite3ValueApplyAffinity(pVal, aff, SQLITE_UTF8); + assert( rc==SQLITE_OK ); + rc = sqlite3VdbeChangeEncoding(pVal, enc); + if( rc==SQLITE_OK && sqlite3VdbeMemTooBig(pVal) ){ + rc = SQLITE_TOOBIG; + pCtx->pParse->nErr++; + } + } + pCtx->pParse->rc = rc; + + value_from_function_out: + if( rc!=SQLITE_OK ){ + pVal = 0; + } + if( apVal ){ + for(i=0; iop)==TK_UPLUS || op==TK_SPAN ) pExpr = pExpr->pLeft; +#if defined(SQLITE_ENABLE_STAT4) + if( op==TK_REGISTER ) op = pExpr->op2; +#else + if( NEVER(op==TK_REGISTER) ) op = pExpr->op2; +#endif + + /* Compressed expressions only appear when parsing the DEFAULT clause + ** on a table column definition, and hence only when pCtx==0. This + ** check ensures that an EP_TokenOnly expression is never passed down + ** into valueFromFunction(). */ + assert( (pExpr->flags & EP_TokenOnly)==0 || pCtx==0 ); + + if( op==TK_CAST ){ + u8 aff = sqlite3AffinityType(pExpr->u.zToken,0); + rc = valueFromExpr(db, pExpr->pLeft, enc, aff, ppVal, pCtx); + testcase( rc!=SQLITE_OK ); + if( *ppVal ){ + sqlite3VdbeMemCast(*ppVal, aff, SQLITE_UTF8); + sqlite3ValueApplyAffinity(*ppVal, affinity, SQLITE_UTF8); + } + return rc; + } + + /* Handle negative integers in a single step. This is needed in the + ** case when the value is -9223372036854775808. + */ + if( op==TK_UMINUS + && (pExpr->pLeft->op==TK_INTEGER || pExpr->pLeft->op==TK_FLOAT) ){ + pExpr = pExpr->pLeft; + op = pExpr->op; + negInt = -1; + zNeg = "-"; + } + + if( op==TK_STRING || op==TK_FLOAT || op==TK_INTEGER ){ + pVal = valueNew(db, pCtx); + if( pVal==0 ) goto no_mem; + if( ExprHasProperty(pExpr, EP_IntValue) ){ + sqlite3VdbeMemSetInt64(pVal, (i64)pExpr->u.iValue*negInt); + }else{ + zVal = sqlite3MPrintf(db, "%s%s", zNeg, pExpr->u.zToken); + if( zVal==0 ) goto no_mem; + sqlite3ValueSetStr(pVal, -1, zVal, SQLITE_UTF8, SQLITE_DYNAMIC); + } + if( (op==TK_INTEGER || op==TK_FLOAT ) && affinity==SQLITE_AFF_BLOB ){ + sqlite3ValueApplyAffinity(pVal, SQLITE_AFF_NUMERIC, SQLITE_UTF8); + }else{ + sqlite3ValueApplyAffinity(pVal, affinity, SQLITE_UTF8); + } + assert( (pVal->flags & MEM_IntReal)==0 ); + if( pVal->flags & (MEM_Int|MEM_IntReal|MEM_Real) ){ + testcase( pVal->flags & MEM_Int ); + testcase( pVal->flags & MEM_Real ); + pVal->flags &= ~MEM_Str; + } + if( enc!=SQLITE_UTF8 ){ + rc = sqlite3VdbeChangeEncoding(pVal, enc); + } + }else if( op==TK_UMINUS ) { + /* This branch happens for multiple negative signs. Ex: -(-5) */ + if( SQLITE_OK==valueFromExpr(db,pExpr->pLeft,enc,affinity,&pVal,pCtx) + && pVal!=0 + ){ + sqlite3VdbeMemNumerify(pVal); + if( pVal->flags & MEM_Real ){ + pVal->u.r = -pVal->u.r; + }else if( pVal->u.i==SMALLEST_INT64 ){ + pVal->u.r = -(double)SMALLEST_INT64; + MemSetTypeFlag(pVal, MEM_Real); + }else{ + pVal->u.i = -pVal->u.i; + } + sqlite3ValueApplyAffinity(pVal, affinity, enc); + } + }else if( op==TK_NULL ){ + pVal = valueNew(db, pCtx); + if( pVal==0 ) goto no_mem; + sqlite3VdbeMemSetNull(pVal); + } +#ifndef SQLITE_OMIT_BLOB_LITERAL + else if( op==TK_BLOB ){ + int nVal; + assert( pExpr->u.zToken[0]=='x' || pExpr->u.zToken[0]=='X' ); + assert( pExpr->u.zToken[1]=='\'' ); + pVal = valueNew(db, pCtx); + if( !pVal ) goto no_mem; + zVal = &pExpr->u.zToken[2]; + nVal = sqlite3Strlen30(zVal)-1; + assert( zVal[nVal]=='\'' ); + sqlite3VdbeMemSetStr(pVal, sqlite3HexToBlob(db, zVal, nVal), nVal/2, + 0, SQLITE_DYNAMIC); + } +#endif +#ifdef SQLITE_ENABLE_STAT4 + else if( op==TK_FUNCTION && pCtx!=0 ){ + rc = valueFromFunction(db, pExpr, enc, affinity, &pVal, pCtx); + } +#endif + else if( op==TK_TRUEFALSE ){ + pVal = valueNew(db, pCtx); + if( pVal ){ + pVal->flags = MEM_Int; + pVal->u.i = pExpr->u.zToken[4]==0; + } + } + + *ppVal = pVal; + return rc; + +no_mem: +#ifdef SQLITE_ENABLE_STAT4 + if( pCtx==0 || pCtx->pParse->nErr==0 ) +#endif + sqlite3OomFault(db); + sqlite3DbFree(db, zVal); + assert( *ppVal==0 ); +#ifdef SQLITE_ENABLE_STAT4 + if( pCtx==0 ) sqlite3ValueFree(pVal); +#else + assert( pCtx==0 ); sqlite3ValueFree(pVal); +#endif + return SQLITE_NOMEM_BKPT; +} + +/* +** Create a new sqlite3_value object, containing the value of pExpr. +** +** This only works for very simple expressions that consist of one constant +** token (i.e. "5", "5.1", "'a string'"). If the expression can +** be converted directly into a value, then the value is allocated and +** a pointer written to *ppVal. The caller is responsible for deallocating +** the value by passing it to sqlite3ValueFree() later on. If the expression +** cannot be converted to a value, then *ppVal is set to NULL. +*/ +SQLITE_PRIVATE int sqlite3ValueFromExpr( + sqlite3 *db, /* The database connection */ + Expr *pExpr, /* The expression to evaluate */ + u8 enc, /* Encoding to use */ + u8 affinity, /* Affinity to use */ + sqlite3_value **ppVal /* Write the new value here */ +){ + return pExpr ? valueFromExpr(db, pExpr, enc, affinity, ppVal, 0) : 0; +} + +#ifdef SQLITE_ENABLE_STAT4 +/* +** Attempt to extract a value from pExpr and use it to construct *ppVal. +** +** If pAlloc is not NULL, then an UnpackedRecord object is created for +** pAlloc if one does not exist and the new value is added to the +** UnpackedRecord object. +** +** A value is extracted in the following cases: +** +** * (pExpr==0). In this case the value is assumed to be an SQL NULL, +** +** * The expression is a bound variable, and this is a reprepare, or +** +** * The expression is a literal value. +** +** On success, *ppVal is made to point to the extracted value. The caller +** is responsible for ensuring that the value is eventually freed. +*/ +static int stat4ValueFromExpr( + Parse *pParse, /* Parse context */ + Expr *pExpr, /* The expression to extract a value from */ + u8 affinity, /* Affinity to use */ + struct ValueNewStat4Ctx *pAlloc,/* How to allocate space. Or NULL */ + sqlite3_value **ppVal /* OUT: New value object (or NULL) */ +){ + int rc = SQLITE_OK; + sqlite3_value *pVal = 0; + sqlite3 *db = pParse->db; + + /* Skip over any TK_COLLATE nodes */ + pExpr = sqlite3ExprSkipCollate(pExpr); + + assert( pExpr==0 || pExpr->op!=TK_REGISTER || pExpr->op2!=TK_VARIABLE ); + if( !pExpr ){ + pVal = valueNew(db, pAlloc); + if( pVal ){ + sqlite3VdbeMemSetNull((Mem*)pVal); + } + }else if( pExpr->op==TK_VARIABLE && (db->flags & SQLITE_EnableQPSG)==0 ){ + Vdbe *v; + int iBindVar = pExpr->iColumn; + sqlite3VdbeSetVarmask(pParse->pVdbe, iBindVar); + if( (v = pParse->pReprepare)!=0 ){ + pVal = valueNew(db, pAlloc); + if( pVal ){ + rc = sqlite3VdbeMemCopy((Mem*)pVal, &v->aVar[iBindVar-1]); + sqlite3ValueApplyAffinity(pVal, affinity, ENC(db)); + pVal->db = pParse->db; + } + } + }else{ + rc = valueFromExpr(db, pExpr, ENC(db), affinity, &pVal, pAlloc); + } + + assert( pVal==0 || pVal->db==db ); + *ppVal = pVal; + return rc; +} + +/* +** This function is used to allocate and populate UnpackedRecord +** structures intended to be compared against sample index keys stored +** in the sqlite_stat4 table. +** +** A single call to this function populates zero or more fields of the +** record starting with field iVal (fields are numbered from left to +** right starting with 0). A single field is populated if: +** +** * (pExpr==0). In this case the value is assumed to be an SQL NULL, +** +** * The expression is a bound variable, and this is a reprepare, or +** +** * The sqlite3ValueFromExpr() function is able to extract a value +** from the expression (i.e. the expression is a literal value). +** +** Or, if pExpr is a TK_VECTOR, one field is populated for each of the +** vector components that match either of the two latter criteria listed +** above. +** +** Before any value is appended to the record, the affinity of the +** corresponding column within index pIdx is applied to it. Before +** this function returns, output parameter *pnExtract is set to the +** number of values appended to the record. +** +** When this function is called, *ppRec must either point to an object +** allocated by an earlier call to this function, or must be NULL. If it +** is NULL and a value can be successfully extracted, a new UnpackedRecord +** is allocated (and *ppRec set to point to it) before returning. +** +** Unless an error is encountered, SQLITE_OK is returned. It is not an +** error if a value cannot be extracted from pExpr. If an error does +** occur, an SQLite error code is returned. +*/ +SQLITE_PRIVATE int sqlite3Stat4ProbeSetValue( + Parse *pParse, /* Parse context */ + Index *pIdx, /* Index being probed */ + UnpackedRecord **ppRec, /* IN/OUT: Probe record */ + Expr *pExpr, /* The expression to extract a value from */ + int nElem, /* Maximum number of values to append */ + int iVal, /* Array element to populate */ + int *pnExtract /* OUT: Values appended to the record */ +){ + int rc = SQLITE_OK; + int nExtract = 0; + + if( pExpr==0 || pExpr->op!=TK_SELECT ){ + int i; + struct ValueNewStat4Ctx alloc; + + alloc.pParse = pParse; + alloc.pIdx = pIdx; + alloc.ppRec = ppRec; + + for(i=0; idb, pIdx, iVal+i); + alloc.iVal = iVal+i; + rc = stat4ValueFromExpr(pParse, pElem, aff, &alloc, &pVal); + if( !pVal ) break; + nExtract++; + } + } + + *pnExtract = nExtract; + return rc; +} + +/* +** Attempt to extract a value from expression pExpr using the methods +** as described for sqlite3Stat4ProbeSetValue() above. +** +** If successful, set *ppVal to point to a new value object and return +** SQLITE_OK. If no value can be extracted, but no other error occurs +** (e.g. OOM), return SQLITE_OK and set *ppVal to NULL. Or, if an error +** does occur, return an SQLite error code. The final value of *ppVal +** is undefined in this case. +*/ +SQLITE_PRIVATE int sqlite3Stat4ValueFromExpr( + Parse *pParse, /* Parse context */ + Expr *pExpr, /* The expression to extract a value from */ + u8 affinity, /* Affinity to use */ + sqlite3_value **ppVal /* OUT: New value object (or NULL) */ +){ + return stat4ValueFromExpr(pParse, pExpr, affinity, 0, ppVal); +} + +/* +** Extract the iCol-th column from the nRec-byte record in pRec. Write +** the column value into *ppVal. If *ppVal is initially NULL then a new +** sqlite3_value object is allocated. +** +** If *ppVal is initially NULL then the caller is responsible for +** ensuring that the value written into *ppVal is eventually freed. +*/ +SQLITE_PRIVATE int sqlite3Stat4Column( + sqlite3 *db, /* Database handle */ + const void *pRec, /* Pointer to buffer containing record */ + int nRec, /* Size of buffer pRec in bytes */ + int iCol, /* Column to extract */ + sqlite3_value **ppVal /* OUT: Extracted value */ +){ + u32 t = 0; /* a column type code */ + int nHdr; /* Size of the header in the record */ + int iHdr; /* Next unread header byte */ + int iField; /* Next unread data byte */ + int szField = 0; /* Size of the current data field */ + int i; /* Column index */ + u8 *a = (u8*)pRec; /* Typecast byte array */ + Mem *pMem = *ppVal; /* Write result into this Mem object */ + + assert( iCol>0 ); + iHdr = getVarint32(a, nHdr); + if( nHdr>nRec || iHdr>=nHdr ) return SQLITE_CORRUPT_BKPT; + iField = nHdr; + for(i=0; i<=iCol; i++){ + iHdr += getVarint32(&a[iHdr], t); + testcase( iHdr==nHdr ); + testcase( iHdr==nHdr+1 ); + if( iHdr>nHdr ) return SQLITE_CORRUPT_BKPT; + szField = sqlite3VdbeSerialTypeLen(t); + iField += szField; + } + testcase( iField==nRec ); + testcase( iField==nRec+1 ); + if( iField>nRec ) return SQLITE_CORRUPT_BKPT; + if( pMem==0 ){ + pMem = *ppVal = sqlite3ValueNew(db); + if( pMem==0 ) return SQLITE_NOMEM_BKPT; + } + sqlite3VdbeSerialGet(&a[iField-szField], t, pMem); + pMem->enc = ENC(db); + return SQLITE_OK; +} + +/* +** Unless it is NULL, the argument must be an UnpackedRecord object returned +** by an earlier call to sqlite3Stat4ProbeSetValue(). This call deletes +** the object. +*/ +SQLITE_PRIVATE void sqlite3Stat4ProbeFree(UnpackedRecord *pRec){ + if( pRec ){ + int i; + int nCol = pRec->pKeyInfo->nAllField; + Mem *aMem = pRec->aMem; + sqlite3 *db = aMem[0].db; + for(i=0; ipKeyInfo); + sqlite3DbFreeNN(db, pRec); + } +} +#endif /* ifdef SQLITE_ENABLE_STAT4 */ + +/* +** Change the string value of an sqlite3_value object +*/ +SQLITE_PRIVATE void sqlite3ValueSetStr( + sqlite3_value *v, /* Value to be set */ + int n, /* Length of string z */ + const void *z, /* Text of the new string */ + u8 enc, /* Encoding to use */ + void (*xDel)(void*) /* Destructor for the string */ +){ + if( v ) sqlite3VdbeMemSetStr((Mem *)v, z, n, enc, xDel); +} + +/* +** Free an sqlite3_value object +*/ +SQLITE_PRIVATE void sqlite3ValueFree(sqlite3_value *v){ + if( !v ) return; + sqlite3VdbeMemRelease((Mem *)v); + sqlite3DbFreeNN(((Mem*)v)->db, v); +} + +/* +** The sqlite3ValueBytes() routine returns the number of bytes in the +** sqlite3_value object assuming that it uses the encoding "enc". +** The valueBytes() routine is a helper function. +*/ +static SQLITE_NOINLINE int valueBytes(sqlite3_value *pVal, u8 enc){ + return valueToText(pVal, enc)!=0 ? pVal->n : 0; +} +SQLITE_PRIVATE int sqlite3ValueBytes(sqlite3_value *pVal, u8 enc){ + Mem *p = (Mem*)pVal; + assert( (p->flags & MEM_Null)==0 || (p->flags & (MEM_Str|MEM_Blob))==0 ); + if( (p->flags & MEM_Str)!=0 && pVal->enc==enc ){ + return p->n; + } + if( (p->flags & MEM_Blob)!=0 ){ + if( p->flags & MEM_Zero ){ + return p->n + p->u.nZero; + }else{ + return p->n; + } + } + if( p->flags & MEM_Null ) return 0; + return valueBytes(pVal, enc); +} + +/************** End of vdbemem.c *********************************************/ +/************** Begin file vdbeaux.c *****************************************/ +/* +** 2003 September 6 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** This file contains code used for creating, destroying, and populating +** a VDBE (or an "sqlite3_stmt" as it is known to the outside world.) +*/ +/* #include "sqliteInt.h" */ +/* #include "vdbeInt.h" */ + +/* +** Create a new virtual database engine. +*/ +SQLITE_PRIVATE Vdbe *sqlite3VdbeCreate(Parse *pParse){ + sqlite3 *db = pParse->db; + Vdbe *p; + p = sqlite3DbMallocRawNN(db, sizeof(Vdbe) ); + if( p==0 ) return 0; + memset(&p->aOp, 0, sizeof(Vdbe)-offsetof(Vdbe,aOp)); + p->db = db; + if( db->pVdbe ){ + db->pVdbe->pPrev = p; + } + p->pNext = db->pVdbe; + p->pPrev = 0; + db->pVdbe = p; + p->magic = VDBE_MAGIC_INIT; + p->pParse = pParse; + pParse->pVdbe = p; + assert( pParse->aLabel==0 ); + assert( pParse->nLabel==0 ); + assert( p->nOpAlloc==0 ); + assert( pParse->szOpAlloc==0 ); + sqlite3VdbeAddOp2(p, OP_Init, 0, 1); + return p; +} + +/* +** Change the error string stored in Vdbe.zErrMsg +*/ +SQLITE_PRIVATE void sqlite3VdbeError(Vdbe *p, const char *zFormat, ...){ + va_list ap; + sqlite3DbFree(p->db, p->zErrMsg); + va_start(ap, zFormat); + p->zErrMsg = sqlite3VMPrintf(p->db, zFormat, ap); + va_end(ap); +} + +/* +** Remember the SQL string for a prepared statement. +*/ +SQLITE_PRIVATE void sqlite3VdbeSetSql(Vdbe *p, const char *z, int n, u8 prepFlags){ + if( p==0 ) return; + p->prepFlags = prepFlags; + if( (prepFlags & SQLITE_PREPARE_SAVESQL)==0 ){ + p->expmask = 0; + } + assert( p->zSql==0 ); + p->zSql = sqlite3DbStrNDup(p->db, z, n); +} + +#ifdef SQLITE_ENABLE_NORMALIZE +/* +** Add a new element to the Vdbe->pDblStr list. +*/ +SQLITE_PRIVATE void sqlite3VdbeAddDblquoteStr(sqlite3 *db, Vdbe *p, const char *z){ + if( p ){ + int n = sqlite3Strlen30(z); + DblquoteStr *pStr = sqlite3DbMallocRawNN(db, + sizeof(*pStr)+n+1-sizeof(pStr->z)); + if( pStr ){ + pStr->pNextStr = p->pDblStr; + p->pDblStr = pStr; + memcpy(pStr->z, z, n+1); + } + } +} +#endif + +#ifdef SQLITE_ENABLE_NORMALIZE +/* +** zId of length nId is a double-quoted identifier. Check to see if +** that identifier is really used as a string literal. +*/ +SQLITE_PRIVATE int sqlite3VdbeUsesDoubleQuotedString( + Vdbe *pVdbe, /* The prepared statement */ + const char *zId /* The double-quoted identifier, already dequoted */ +){ + DblquoteStr *pStr; + assert( zId!=0 ); + if( pVdbe->pDblStr==0 ) return 0; + for(pStr=pVdbe->pDblStr; pStr; pStr=pStr->pNextStr){ + if( strcmp(zId, pStr->z)==0 ) return 1; + } + return 0; +} +#endif + +/* +** Swap all content between two VDBE structures. +*/ +SQLITE_PRIVATE void sqlite3VdbeSwap(Vdbe *pA, Vdbe *pB){ + Vdbe tmp, *pTmp; + char *zTmp; + assert( pA->db==pB->db ); + tmp = *pA; + *pA = *pB; + *pB = tmp; + pTmp = pA->pNext; + pA->pNext = pB->pNext; + pB->pNext = pTmp; + pTmp = pA->pPrev; + pA->pPrev = pB->pPrev; + pB->pPrev = pTmp; + zTmp = pA->zSql; + pA->zSql = pB->zSql; + pB->zSql = zTmp; +#if 0 + zTmp = pA->zNormSql; + pA->zNormSql = pB->zNormSql; + pB->zNormSql = zTmp; +#endif + pB->expmask = pA->expmask; + pB->prepFlags = pA->prepFlags; + memcpy(pB->aCounter, pA->aCounter, sizeof(pB->aCounter)); + pB->aCounter[SQLITE_STMTSTATUS_REPREPARE]++; +} + +/* +** Resize the Vdbe.aOp array so that it is at least nOp elements larger +** than its current size. nOp is guaranteed to be less than or equal +** to 1024/sizeof(Op). +** +** If an out-of-memory error occurs while resizing the array, return +** SQLITE_NOMEM. In this case Vdbe.aOp and Vdbe.nOpAlloc remain +** unchanged (this is so that any opcodes already allocated can be +** correctly deallocated along with the rest of the Vdbe). +*/ +static int growOpArray(Vdbe *v, int nOp){ + VdbeOp *pNew; + Parse *p = v->pParse; + + /* The SQLITE_TEST_REALLOC_STRESS compile-time option is designed to force + ** more frequent reallocs and hence provide more opportunities for + ** simulated OOM faults. SQLITE_TEST_REALLOC_STRESS is generally used + ** during testing only. With SQLITE_TEST_REALLOC_STRESS grow the op array + ** by the minimum* amount required until the size reaches 512. Normal + ** operation (without SQLITE_TEST_REALLOC_STRESS) is to double the current + ** size of the op array or add 1KB of space, whichever is smaller. */ +#ifdef SQLITE_TEST_REALLOC_STRESS + sqlite3_int64 nNew = (v->nOpAlloc>=512 ? 2*(sqlite3_int64)v->nOpAlloc + : (sqlite3_int64)v->nOpAlloc+nOp); +#else + sqlite3_int64 nNew = (v->nOpAlloc ? 2*(sqlite3_int64)v->nOpAlloc + : (sqlite3_int64)(1024/sizeof(Op))); + UNUSED_PARAMETER(nOp); +#endif + + /* Ensure that the size of a VDBE does not grow too large */ + if( nNew > p->db->aLimit[SQLITE_LIMIT_VDBE_OP] ){ + sqlite3OomFault(p->db); + return SQLITE_NOMEM; + } + + assert( nOp<=(1024/sizeof(Op)) ); + assert( nNew>=(v->nOpAlloc+nOp) ); + pNew = sqlite3DbRealloc(p->db, v->aOp, nNew*sizeof(Op)); + if( pNew ){ + p->szOpAlloc = sqlite3DbMallocSize(p->db, pNew); + v->nOpAlloc = p->szOpAlloc/sizeof(Op); + v->aOp = pNew; + } + return (pNew ? SQLITE_OK : SQLITE_NOMEM_BKPT); +} + +#ifdef SQLITE_DEBUG +/* This routine is just a convenient place to set a breakpoint that will +** fire after each opcode is inserted and displayed using +** "PRAGMA vdbe_addoptrace=on". +*/ +static void test_addop_breakpoint(void){ + static int n = 0; + n++; +} +#endif + +/* +** Add a new instruction to the list of instructions current in the +** VDBE. Return the address of the new instruction. +** +** Parameters: +** +** p Pointer to the VDBE +** +** op The opcode for this instruction +** +** p1, p2, p3 Operands +** +** Use the sqlite3VdbeResolveLabel() function to fix an address and +** the sqlite3VdbeChangeP4() function to change the value of the P4 +** operand. +*/ +static SQLITE_NOINLINE int growOp3(Vdbe *p, int op, int p1, int p2, int p3){ + assert( p->nOpAlloc<=p->nOp ); + if( growOpArray(p, 1) ) return 1; + assert( p->nOpAlloc>p->nOp ); + return sqlite3VdbeAddOp3(p, op, p1, p2, p3); +} +SQLITE_PRIVATE int sqlite3VdbeAddOp3(Vdbe *p, int op, int p1, int p2, int p3){ + int i; + VdbeOp *pOp; + + i = p->nOp; + assert( p->magic==VDBE_MAGIC_INIT ); + assert( op>=0 && op<0xff ); + if( p->nOpAlloc<=i ){ + return growOp3(p, op, p1, p2, p3); + } + p->nOp++; + pOp = &p->aOp[i]; + pOp->opcode = (u8)op; + pOp->p5 = 0; + pOp->p1 = p1; + pOp->p2 = p2; + pOp->p3 = p3; + pOp->p4.p = 0; + pOp->p4type = P4_NOTUSED; +#ifdef SQLITE_ENABLE_EXPLAIN_COMMENTS + pOp->zComment = 0; +#endif +#ifdef SQLITE_DEBUG + if( p->db->flags & SQLITE_VdbeAddopTrace ){ + sqlite3VdbePrintOp(0, i, &p->aOp[i]); + test_addop_breakpoint(); + } +#endif +#ifdef VDBE_PROFILE + pOp->cycles = 0; + pOp->cnt = 0; +#endif +#ifdef SQLITE_VDBE_COVERAGE + pOp->iSrcLine = 0; +#endif + return i; +} +SQLITE_PRIVATE int sqlite3VdbeAddOp0(Vdbe *p, int op){ + return sqlite3VdbeAddOp3(p, op, 0, 0, 0); +} +SQLITE_PRIVATE int sqlite3VdbeAddOp1(Vdbe *p, int op, int p1){ + return sqlite3VdbeAddOp3(p, op, p1, 0, 0); +} +SQLITE_PRIVATE int sqlite3VdbeAddOp2(Vdbe *p, int op, int p1, int p2){ + return sqlite3VdbeAddOp3(p, op, p1, p2, 0); +} + +/* Generate code for an unconditional jump to instruction iDest +*/ +SQLITE_PRIVATE int sqlite3VdbeGoto(Vdbe *p, int iDest){ + return sqlite3VdbeAddOp3(p, OP_Goto, 0, iDest, 0); +} + +/* Generate code to cause the string zStr to be loaded into +** register iDest +*/ +SQLITE_PRIVATE int sqlite3VdbeLoadString(Vdbe *p, int iDest, const char *zStr){ + return sqlite3VdbeAddOp4(p, OP_String8, 0, iDest, 0, zStr, 0); +} + +/* +** Generate code that initializes multiple registers to string or integer +** constants. The registers begin with iDest and increase consecutively. +** One register is initialized for each characgter in zTypes[]. For each +** "s" character in zTypes[], the register is a string if the argument is +** not NULL, or OP_Null if the value is a null pointer. For each "i" character +** in zTypes[], the register is initialized to an integer. +** +** If the input string does not end with "X" then an OP_ResultRow instruction +** is generated for the values inserted. +*/ +SQLITE_PRIVATE void sqlite3VdbeMultiLoad(Vdbe *p, int iDest, const char *zTypes, ...){ + va_list ap; + int i; + char c; + va_start(ap, zTypes); + for(i=0; (c = zTypes[i])!=0; i++){ + if( c=='s' ){ + const char *z = va_arg(ap, const char*); + sqlite3VdbeAddOp4(p, z==0 ? OP_Null : OP_String8, 0, iDest+i, 0, z, 0); + }else if( c=='i' ){ + sqlite3VdbeAddOp2(p, OP_Integer, va_arg(ap, int), iDest+i); + }else{ + goto skip_op_resultrow; + } + } + sqlite3VdbeAddOp2(p, OP_ResultRow, iDest, i); +skip_op_resultrow: + va_end(ap); +} + +/* +** Add an opcode that includes the p4 value as a pointer. +*/ +SQLITE_PRIVATE int sqlite3VdbeAddOp4( + Vdbe *p, /* Add the opcode to this VM */ + int op, /* The new opcode */ + int p1, /* The P1 operand */ + int p2, /* The P2 operand */ + int p3, /* The P3 operand */ + const char *zP4, /* The P4 operand */ + int p4type /* P4 operand type */ +){ + int addr = sqlite3VdbeAddOp3(p, op, p1, p2, p3); + sqlite3VdbeChangeP4(p, addr, zP4, p4type); + return addr; +} + +/* +** Add an opcode that includes the p4 value with a P4_INT64 or +** P4_REAL type. +*/ +SQLITE_PRIVATE int sqlite3VdbeAddOp4Dup8( + Vdbe *p, /* Add the opcode to this VM */ + int op, /* The new opcode */ + int p1, /* The P1 operand */ + int p2, /* The P2 operand */ + int p3, /* The P3 operand */ + const u8 *zP4, /* The P4 operand */ + int p4type /* P4 operand type */ +){ + char *p4copy = sqlite3DbMallocRawNN(sqlite3VdbeDb(p), 8); + if( p4copy ) memcpy(p4copy, zP4, 8); + return sqlite3VdbeAddOp4(p, op, p1, p2, p3, p4copy, p4type); +} + +#ifndef SQLITE_OMIT_EXPLAIN +/* +** Return the address of the current EXPLAIN QUERY PLAN baseline. +** 0 means "none". +*/ +SQLITE_PRIVATE int sqlite3VdbeExplainParent(Parse *pParse){ + VdbeOp *pOp; + if( pParse->addrExplain==0 ) return 0; + pOp = sqlite3VdbeGetOp(pParse->pVdbe, pParse->addrExplain); + return pOp->p2; +} + +/* +** Set a debugger breakpoint on the following routine in order to +** monitor the EXPLAIN QUERY PLAN code generation. +*/ +#if defined(SQLITE_DEBUG) +SQLITE_PRIVATE void sqlite3ExplainBreakpoint(const char *z1, const char *z2){ + (void)z1; + (void)z2; +} +#endif + +/* +** Add a new OP_ opcode. +** +** If the bPush flag is true, then make this opcode the parent for +** subsequent Explains until sqlite3VdbeExplainPop() is called. +*/ +SQLITE_PRIVATE void sqlite3VdbeExplain(Parse *pParse, u8 bPush, const char *zFmt, ...){ +#ifndef SQLITE_DEBUG + /* Always include the OP_Explain opcodes if SQLITE_DEBUG is defined. + ** But omit them (for performance) during production builds */ + if( pParse->explain==2 ) +#endif + { + char *zMsg; + Vdbe *v; + va_list ap; + int iThis; + va_start(ap, zFmt); + zMsg = sqlite3VMPrintf(pParse->db, zFmt, ap); + va_end(ap); + v = pParse->pVdbe; + iThis = v->nOp; + sqlite3VdbeAddOp4(v, OP_Explain, iThis, pParse->addrExplain, 0, + zMsg, P4_DYNAMIC); + sqlite3ExplainBreakpoint(bPush?"PUSH":"", sqlite3VdbeGetOp(v,-1)->p4.z); + if( bPush){ + pParse->addrExplain = iThis; + } + } +} + +/* +** Pop the EXPLAIN QUERY PLAN stack one level. +*/ +SQLITE_PRIVATE void sqlite3VdbeExplainPop(Parse *pParse){ + sqlite3ExplainBreakpoint("POP", 0); + pParse->addrExplain = sqlite3VdbeExplainParent(pParse); +} +#endif /* SQLITE_OMIT_EXPLAIN */ + +/* +** Add an OP_ParseSchema opcode. This routine is broken out from +** sqlite3VdbeAddOp4() since it needs to also needs to mark all btrees +** as having been used. +** +** The zWhere string must have been obtained from sqlite3_malloc(). +** This routine will take ownership of the allocated memory. +*/ +SQLITE_PRIVATE void sqlite3VdbeAddParseSchemaOp(Vdbe *p, int iDb, char *zWhere){ + int j; + sqlite3VdbeAddOp4(p, OP_ParseSchema, iDb, 0, 0, zWhere, P4_DYNAMIC); + for(j=0; jdb->nDb; j++) sqlite3VdbeUsesBtree(p, j); +} + +/* +** Add an opcode that includes the p4 value as an integer. +*/ +SQLITE_PRIVATE int sqlite3VdbeAddOp4Int( + Vdbe *p, /* Add the opcode to this VM */ + int op, /* The new opcode */ + int p1, /* The P1 operand */ + int p2, /* The P2 operand */ + int p3, /* The P3 operand */ + int p4 /* The P4 operand as an integer */ +){ + int addr = sqlite3VdbeAddOp3(p, op, p1, p2, p3); + if( p->db->mallocFailed==0 ){ + VdbeOp *pOp = &p->aOp[addr]; + pOp->p4type = P4_INT32; + pOp->p4.i = p4; + } + return addr; +} + +/* Insert the end of a co-routine +*/ +SQLITE_PRIVATE void sqlite3VdbeEndCoroutine(Vdbe *v, int regYield){ + sqlite3VdbeAddOp1(v, OP_EndCoroutine, regYield); + + /* Clear the temporary register cache, thereby ensuring that each + ** co-routine has its own independent set of registers, because co-routines + ** might expect their registers to be preserved across an OP_Yield, and + ** that could cause problems if two or more co-routines are using the same + ** temporary register. + */ + v->pParse->nTempReg = 0; + v->pParse->nRangeReg = 0; +} + +/* +** Create a new symbolic label for an instruction that has yet to be +** coded. The symbolic label is really just a negative number. The +** label can be used as the P2 value of an operation. Later, when +** the label is resolved to a specific address, the VDBE will scan +** through its operation list and change all values of P2 which match +** the label into the resolved address. +** +** The VDBE knows that a P2 value is a label because labels are +** always negative and P2 values are suppose to be non-negative. +** Hence, a negative P2 value is a label that has yet to be resolved. +** (Later:) This is only true for opcodes that have the OPFLG_JUMP +** property. +** +** Variable usage notes: +** +** Parse.aLabel[x] Stores the address that the x-th label resolves +** into. For testing (SQLITE_DEBUG), unresolved +** labels stores -1, but that is not required. +** Parse.nLabelAlloc Number of slots allocated to Parse.aLabel[] +** Parse.nLabel The *negative* of the number of labels that have +** been issued. The negative is stored because +** that gives a performance improvement over storing +** the equivalent positive value. +*/ +SQLITE_PRIVATE int sqlite3VdbeMakeLabel(Parse *pParse){ + return --pParse->nLabel; +} + +/* +** Resolve label "x" to be the address of the next instruction to +** be inserted. The parameter "x" must have been obtained from +** a prior call to sqlite3VdbeMakeLabel(). +*/ +static SQLITE_NOINLINE void resizeResolveLabel(Parse *p, Vdbe *v, int j){ + int nNewSize = 10 - p->nLabel; + p->aLabel = sqlite3DbReallocOrFree(p->db, p->aLabel, + nNewSize*sizeof(p->aLabel[0])); + if( p->aLabel==0 ){ + p->nLabelAlloc = 0; + }else{ +#ifdef SQLITE_DEBUG + int i; + for(i=p->nLabelAlloc; iaLabel[i] = -1; +#endif + p->nLabelAlloc = nNewSize; + p->aLabel[j] = v->nOp; + } +} +SQLITE_PRIVATE void sqlite3VdbeResolveLabel(Vdbe *v, int x){ + Parse *p = v->pParse; + int j = ADDR(x); + assert( v->magic==VDBE_MAGIC_INIT ); + assert( j<-p->nLabel ); + assert( j>=0 ); +#ifdef SQLITE_DEBUG + if( p->db->flags & SQLITE_VdbeAddopTrace ){ + printf("RESOLVE LABEL %d to %d\n", x, v->nOp); + } +#endif + if( p->nLabelAlloc + p->nLabel < 0 ){ + resizeResolveLabel(p,v,j); + }else{ + assert( p->aLabel[j]==(-1) ); /* Labels may only be resolved once */ + p->aLabel[j] = v->nOp; + } +} + +/* +** Mark the VDBE as one that can only be run one time. +*/ +SQLITE_PRIVATE void sqlite3VdbeRunOnlyOnce(Vdbe *p){ + p->runOnlyOnce = 1; +} + +/* +** Mark the VDBE as one that can only be run multiple times. +*/ +SQLITE_PRIVATE void sqlite3VdbeReusable(Vdbe *p){ + p->runOnlyOnce = 0; +} + +#ifdef SQLITE_DEBUG /* sqlite3AssertMayAbort() logic */ + +/* +** The following type and function are used to iterate through all opcodes +** in a Vdbe main program and each of the sub-programs (triggers) it may +** invoke directly or indirectly. It should be used as follows: +** +** Op *pOp; +** VdbeOpIter sIter; +** +** memset(&sIter, 0, sizeof(sIter)); +** sIter.v = v; // v is of type Vdbe* +** while( (pOp = opIterNext(&sIter)) ){ +** // Do something with pOp +** } +** sqlite3DbFree(v->db, sIter.apSub); +** +*/ +typedef struct VdbeOpIter VdbeOpIter; +struct VdbeOpIter { + Vdbe *v; /* Vdbe to iterate through the opcodes of */ + SubProgram **apSub; /* Array of subprograms */ + int nSub; /* Number of entries in apSub */ + int iAddr; /* Address of next instruction to return */ + int iSub; /* 0 = main program, 1 = first sub-program etc. */ +}; +static Op *opIterNext(VdbeOpIter *p){ + Vdbe *v = p->v; + Op *pRet = 0; + Op *aOp; + int nOp; + + if( p->iSub<=p->nSub ){ + + if( p->iSub==0 ){ + aOp = v->aOp; + nOp = v->nOp; + }else{ + aOp = p->apSub[p->iSub-1]->aOp; + nOp = p->apSub[p->iSub-1]->nOp; + } + assert( p->iAddriAddr]; + p->iAddr++; + if( p->iAddr==nOp ){ + p->iSub++; + p->iAddr = 0; + } + + if( pRet->p4type==P4_SUBPROGRAM ){ + int nByte = (p->nSub+1)*sizeof(SubProgram*); + int j; + for(j=0; jnSub; j++){ + if( p->apSub[j]==pRet->p4.pProgram ) break; + } + if( j==p->nSub ){ + p->apSub = sqlite3DbReallocOrFree(v->db, p->apSub, nByte); + if( !p->apSub ){ + pRet = 0; + }else{ + p->apSub[p->nSub++] = pRet->p4.pProgram; + } + } + } + } + + return pRet; +} + +/* +** Check if the program stored in the VM associated with pParse may +** throw an ABORT exception (causing the statement, but not entire transaction +** to be rolled back). This condition is true if the main program or any +** sub-programs contains any of the following: +** +** * OP_Halt with P1=SQLITE_CONSTRAINT and P2=OE_Abort. +** * OP_HaltIfNull with P1=SQLITE_CONSTRAINT and P2=OE_Abort. +** * OP_Destroy +** * OP_VUpdate +** * OP_VRename +** * OP_FkCounter with P2==0 (immediate foreign key constraint) +** * OP_CreateBtree/BTREE_INTKEY and OP_InitCoroutine +** (for CREATE TABLE AS SELECT ...) +** +** Then check that the value of Parse.mayAbort is true if an +** ABORT may be thrown, or false otherwise. Return true if it does +** match, or false otherwise. This function is intended to be used as +** part of an assert statement in the compiler. Similar to: +** +** assert( sqlite3VdbeAssertMayAbort(pParse->pVdbe, pParse->mayAbort) ); +*/ +SQLITE_PRIVATE int sqlite3VdbeAssertMayAbort(Vdbe *v, int mayAbort){ + int hasAbort = 0; + int hasFkCounter = 0; + int hasCreateTable = 0; + int hasCreateIndex = 0; + int hasInitCoroutine = 0; + Op *pOp; + VdbeOpIter sIter; + memset(&sIter, 0, sizeof(sIter)); + sIter.v = v; + + while( (pOp = opIterNext(&sIter))!=0 ){ + int opcode = pOp->opcode; + if( opcode==OP_Destroy || opcode==OP_VUpdate || opcode==OP_VRename + || opcode==OP_VDestroy + || (opcode==OP_ParseSchema && pOp->p4.z==0) + || ((opcode==OP_Halt || opcode==OP_HaltIfNull) + && ((pOp->p1)!=SQLITE_OK && pOp->p2==OE_Abort)) + ){ + hasAbort = 1; + break; + } + if( opcode==OP_CreateBtree && pOp->p3==BTREE_INTKEY ) hasCreateTable = 1; + if( mayAbort ){ + /* hasCreateIndex may also be set for some DELETE statements that use + ** OP_Clear. So this routine may end up returning true in the case + ** where a "DELETE FROM tbl" has a statement-journal but does not + ** require one. This is not so bad - it is an inefficiency, not a bug. */ + if( opcode==OP_CreateBtree && pOp->p3==BTREE_BLOBKEY ) hasCreateIndex = 1; + if( opcode==OP_Clear ) hasCreateIndex = 1; + } + if( opcode==OP_InitCoroutine ) hasInitCoroutine = 1; +#ifndef SQLITE_OMIT_FOREIGN_KEY + if( opcode==OP_FkCounter && pOp->p1==0 && pOp->p2==1 ){ + hasFkCounter = 1; + } +#endif + } + sqlite3DbFree(v->db, sIter.apSub); + + /* Return true if hasAbort==mayAbort. Or if a malloc failure occurred. + ** If malloc failed, then the while() loop above may not have iterated + ** through all opcodes and hasAbort may be set incorrectly. Return + ** true for this case to prevent the assert() in the callers frame + ** from failing. */ + return ( v->db->mallocFailed || hasAbort==mayAbort || hasFkCounter + || (hasCreateTable && hasInitCoroutine) || hasCreateIndex + ); +} +#endif /* SQLITE_DEBUG - the sqlite3AssertMayAbort() function */ + +#ifdef SQLITE_DEBUG +/* +** Increment the nWrite counter in the VDBE if the cursor is not an +** ephemeral cursor, or if the cursor argument is NULL. +*/ +SQLITE_PRIVATE void sqlite3VdbeIncrWriteCounter(Vdbe *p, VdbeCursor *pC){ + if( pC==0 + || (pC->eCurType!=CURTYPE_SORTER + && pC->eCurType!=CURTYPE_PSEUDO + && !pC->isEphemeral) + ){ + p->nWrite++; + } +} +#endif + +#ifdef SQLITE_DEBUG +/* +** Assert if an Abort at this point in time might result in a corrupt +** database. +*/ +SQLITE_PRIVATE void sqlite3VdbeAssertAbortable(Vdbe *p){ + assert( p->nWrite==0 || p->usesStmtJournal ); +} +#endif + +/* +** This routine is called after all opcodes have been inserted. It loops +** through all the opcodes and fixes up some details. +** +** (1) For each jump instruction with a negative P2 value (a label) +** resolve the P2 value to an actual address. +** +** (2) Compute the maximum number of arguments used by any SQL function +** and store that value in *pMaxFuncArgs. +** +** (3) Update the Vdbe.readOnly and Vdbe.bIsReader flags to accurately +** indicate what the prepared statement actually does. +** +** (4) Initialize the p4.xAdvance pointer on opcodes that use it. +** +** (5) Reclaim the memory allocated for storing labels. +** +** This routine will only function correctly if the mkopcodeh.tcl generator +** script numbers the opcodes correctly. Changes to this routine must be +** coordinated with changes to mkopcodeh.tcl. +*/ +static void resolveP2Values(Vdbe *p, int *pMaxFuncArgs){ + int nMaxArgs = *pMaxFuncArgs; + Op *pOp; + Parse *pParse = p->pParse; + int *aLabel = pParse->aLabel; + p->readOnly = 1; + p->bIsReader = 0; + pOp = &p->aOp[p->nOp-1]; + while(1){ + + /* Only JUMP opcodes and the short list of special opcodes in the switch + ** below need to be considered. The mkopcodeh.tcl generator script groups + ** all these opcodes together near the front of the opcode list. Skip + ** any opcode that does not need processing by virtual of the fact that + ** it is larger than SQLITE_MX_JUMP_OPCODE, as a performance optimization. + */ + if( pOp->opcode<=SQLITE_MX_JUMP_OPCODE ){ + /* NOTE: Be sure to update mkopcodeh.tcl when adding or removing + ** cases from this switch! */ + switch( pOp->opcode ){ + case OP_Transaction: { + if( pOp->p2!=0 ) p->readOnly = 0; + /* fall thru */ + } + case OP_AutoCommit: + case OP_Savepoint: { + p->bIsReader = 1; + break; + } +#ifndef SQLITE_OMIT_WAL + case OP_Checkpoint: +#endif + case OP_Vacuum: + case OP_JournalMode: { + p->readOnly = 0; + p->bIsReader = 1; + break; + } + case OP_Next: + case OP_SorterNext: { + pOp->p4.xAdvance = sqlite3BtreeNext; + pOp->p4type = P4_ADVANCE; + /* The code generator never codes any of these opcodes as a jump + ** to a label. They are always coded as a jump backwards to a + ** known address */ + assert( pOp->p2>=0 ); + break; + } + case OP_Prev: { + pOp->p4.xAdvance = sqlite3BtreePrevious; + pOp->p4type = P4_ADVANCE; + /* The code generator never codes any of these opcodes as a jump + ** to a label. They are always coded as a jump backwards to a + ** known address */ + assert( pOp->p2>=0 ); + break; + } +#ifndef SQLITE_OMIT_VIRTUALTABLE + case OP_VUpdate: { + if( pOp->p2>nMaxArgs ) nMaxArgs = pOp->p2; + break; + } + case OP_VFilter: { + int n; + assert( (pOp - p->aOp) >= 3 ); + assert( pOp[-1].opcode==OP_Integer ); + n = pOp[-1].p1; + if( n>nMaxArgs ) nMaxArgs = n; + /* Fall through into the default case */ + } +#endif + default: { + if( pOp->p2<0 ){ + /* The mkopcodeh.tcl script has so arranged things that the only + ** non-jump opcodes less than SQLITE_MX_JUMP_CODE are guaranteed to + ** have non-negative values for P2. */ + assert( (sqlite3OpcodeProperty[pOp->opcode] & OPFLG_JUMP)!=0 ); + assert( ADDR(pOp->p2)<-pParse->nLabel ); + pOp->p2 = aLabel[ADDR(pOp->p2)]; + } + break; + } + } + /* The mkopcodeh.tcl script has so arranged things that the only + ** non-jump opcodes less than SQLITE_MX_JUMP_CODE are guaranteed to + ** have non-negative values for P2. */ + assert( (sqlite3OpcodeProperty[pOp->opcode]&OPFLG_JUMP)==0 || pOp->p2>=0); + } + if( pOp==p->aOp ) break; + pOp--; + } + sqlite3DbFree(p->db, pParse->aLabel); + pParse->aLabel = 0; + pParse->nLabel = 0; + *pMaxFuncArgs = nMaxArgs; + assert( p->bIsReader!=0 || DbMaskAllZero(p->btreeMask) ); +} + +/* +** Return the address of the next instruction to be inserted. +*/ +SQLITE_PRIVATE int sqlite3VdbeCurrentAddr(Vdbe *p){ + assert( p->magic==VDBE_MAGIC_INIT ); + return p->nOp; +} + +/* +** Verify that at least N opcode slots are available in p without +** having to malloc for more space (except when compiled using +** SQLITE_TEST_REALLOC_STRESS). This interface is used during testing +** to verify that certain calls to sqlite3VdbeAddOpList() can never +** fail due to a OOM fault and hence that the return value from +** sqlite3VdbeAddOpList() will always be non-NULL. +*/ +#if defined(SQLITE_DEBUG) && !defined(SQLITE_TEST_REALLOC_STRESS) +SQLITE_PRIVATE void sqlite3VdbeVerifyNoMallocRequired(Vdbe *p, int N){ + assert( p->nOp + N <= p->nOpAlloc ); +} +#endif + +/* +** Verify that the VM passed as the only argument does not contain +** an OP_ResultRow opcode. Fail an assert() if it does. This is used +** by code in pragma.c to ensure that the implementation of certain +** pragmas comports with the flags specified in the mkpragmatab.tcl +** script. +*/ +#if defined(SQLITE_DEBUG) && !defined(SQLITE_TEST_REALLOC_STRESS) +SQLITE_PRIVATE void sqlite3VdbeVerifyNoResultRow(Vdbe *p){ + int i; + for(i=0; inOp; i++){ + assert( p->aOp[i].opcode!=OP_ResultRow ); + } +} +#endif + +/* +** Generate code (a single OP_Abortable opcode) that will +** verify that the VDBE program can safely call Abort in the current +** context. +*/ +#if defined(SQLITE_DEBUG) +SQLITE_PRIVATE void sqlite3VdbeVerifyAbortable(Vdbe *p, int onError){ + if( onError==OE_Abort ) sqlite3VdbeAddOp0(p, OP_Abortable); +} +#endif + +/* +** This function returns a pointer to the array of opcodes associated with +** the Vdbe passed as the first argument. It is the callers responsibility +** to arrange for the returned array to be eventually freed using the +** vdbeFreeOpArray() function. +** +** Before returning, *pnOp is set to the number of entries in the returned +** array. Also, *pnMaxArg is set to the larger of its current value and +** the number of entries in the Vdbe.apArg[] array required to execute the +** returned program. +*/ +SQLITE_PRIVATE VdbeOp *sqlite3VdbeTakeOpArray(Vdbe *p, int *pnOp, int *pnMaxArg){ + VdbeOp *aOp = p->aOp; + assert( aOp && !p->db->mallocFailed ); + + /* Check that sqlite3VdbeUsesBtree() was not called on this VM */ + assert( DbMaskAllZero(p->btreeMask) ); + + resolveP2Values(p, pnMaxArg); + *pnOp = p->nOp; + p->aOp = 0; + return aOp; +} + +/* +** Add a whole list of operations to the operation stack. Return a +** pointer to the first operation inserted. +** +** Non-zero P2 arguments to jump instructions are automatically adjusted +** so that the jump target is relative to the first operation inserted. +*/ +SQLITE_PRIVATE VdbeOp *sqlite3VdbeAddOpList( + Vdbe *p, /* Add opcodes to the prepared statement */ + int nOp, /* Number of opcodes to add */ + VdbeOpList const *aOp, /* The opcodes to be added */ + int iLineno /* Source-file line number of first opcode */ +){ + int i; + VdbeOp *pOut, *pFirst; + assert( nOp>0 ); + assert( p->magic==VDBE_MAGIC_INIT ); + if( p->nOp + nOp > p->nOpAlloc && growOpArray(p, nOp) ){ + return 0; + } + pFirst = pOut = &p->aOp[p->nOp]; + for(i=0; iopcode = aOp->opcode; + pOut->p1 = aOp->p1; + pOut->p2 = aOp->p2; + assert( aOp->p2>=0 ); + if( (sqlite3OpcodeProperty[aOp->opcode] & OPFLG_JUMP)!=0 && aOp->p2>0 ){ + pOut->p2 += p->nOp; + } + pOut->p3 = aOp->p3; + pOut->p4type = P4_NOTUSED; + pOut->p4.p = 0; + pOut->p5 = 0; +#ifdef SQLITE_ENABLE_EXPLAIN_COMMENTS + pOut->zComment = 0; +#endif +#ifdef SQLITE_VDBE_COVERAGE + pOut->iSrcLine = iLineno+i; +#else + (void)iLineno; +#endif +#ifdef SQLITE_DEBUG + if( p->db->flags & SQLITE_VdbeAddopTrace ){ + sqlite3VdbePrintOp(0, i+p->nOp, &p->aOp[i+p->nOp]); + } +#endif + } + p->nOp += nOp; + return pFirst; +} + +#if defined(SQLITE_ENABLE_STMT_SCANSTATUS) +/* +** Add an entry to the array of counters managed by sqlite3_stmt_scanstatus(). +*/ +SQLITE_PRIVATE void sqlite3VdbeScanStatus( + Vdbe *p, /* VM to add scanstatus() to */ + int addrExplain, /* Address of OP_Explain (or 0) */ + int addrLoop, /* Address of loop counter */ + int addrVisit, /* Address of rows visited counter */ + LogEst nEst, /* Estimated number of output rows */ + const char *zName /* Name of table or index being scanned */ +){ + sqlite3_int64 nByte = (p->nScan+1) * sizeof(ScanStatus); + ScanStatus *aNew; + aNew = (ScanStatus*)sqlite3DbRealloc(p->db, p->aScan, nByte); + if( aNew ){ + ScanStatus *pNew = &aNew[p->nScan++]; + pNew->addrExplain = addrExplain; + pNew->addrLoop = addrLoop; + pNew->addrVisit = addrVisit; + pNew->nEst = nEst; + pNew->zName = sqlite3DbStrDup(p->db, zName); + p->aScan = aNew; + } +} +#endif + + +/* +** Change the value of the opcode, or P1, P2, P3, or P5 operands +** for a specific instruction. +*/ +SQLITE_PRIVATE void sqlite3VdbeChangeOpcode(Vdbe *p, int addr, u8 iNewOpcode){ + sqlite3VdbeGetOp(p,addr)->opcode = iNewOpcode; +} +SQLITE_PRIVATE void sqlite3VdbeChangeP1(Vdbe *p, int addr, int val){ + sqlite3VdbeGetOp(p,addr)->p1 = val; +} +SQLITE_PRIVATE void sqlite3VdbeChangeP2(Vdbe *p, int addr, int val){ + sqlite3VdbeGetOp(p,addr)->p2 = val; +} +SQLITE_PRIVATE void sqlite3VdbeChangeP3(Vdbe *p, int addr, int val){ + sqlite3VdbeGetOp(p,addr)->p3 = val; +} +SQLITE_PRIVATE void sqlite3VdbeChangeP5(Vdbe *p, u16 p5){ + assert( p->nOp>0 || p->db->mallocFailed ); + if( p->nOp>0 ) p->aOp[p->nOp-1].p5 = p5; +} + +/* +** Change the P2 operand of instruction addr so that it points to +** the address of the next instruction to be coded. +*/ +SQLITE_PRIVATE void sqlite3VdbeJumpHere(Vdbe *p, int addr){ + sqlite3VdbeChangeP2(p, addr, p->nOp); +} + + +/* +** If the input FuncDef structure is ephemeral, then free it. If +** the FuncDef is not ephermal, then do nothing. +*/ +static void freeEphemeralFunction(sqlite3 *db, FuncDef *pDef){ + if( (pDef->funcFlags & SQLITE_FUNC_EPHEM)!=0 ){ + sqlite3DbFreeNN(db, pDef); + } +} + +static void vdbeFreeOpArray(sqlite3 *, Op *, int); + +/* +** Delete a P4 value if necessary. +*/ +static SQLITE_NOINLINE void freeP4Mem(sqlite3 *db, Mem *p){ + if( p->szMalloc ) sqlite3DbFree(db, p->zMalloc); + sqlite3DbFreeNN(db, p); +} +static SQLITE_NOINLINE void freeP4FuncCtx(sqlite3 *db, sqlite3_context *p){ + freeEphemeralFunction(db, p->pFunc); + sqlite3DbFreeNN(db, p); +} +static void freeP4(sqlite3 *db, int p4type, void *p4){ + assert( db ); + switch( p4type ){ + case P4_FUNCCTX: { + freeP4FuncCtx(db, (sqlite3_context*)p4); + break; + } + case P4_REAL: + case P4_INT64: + case P4_DYNAMIC: + case P4_DYNBLOB: + case P4_INTARRAY: { + sqlite3DbFree(db, p4); + break; + } + case P4_KEYINFO: { + if( db->pnBytesFreed==0 ) sqlite3KeyInfoUnref((KeyInfo*)p4); + break; + } +#ifdef SQLITE_ENABLE_CURSOR_HINTS + case P4_EXPR: { + sqlite3ExprDelete(db, (Expr*)p4); + break; + } +#endif + case P4_FUNCDEF: { + freeEphemeralFunction(db, (FuncDef*)p4); + break; + } + case P4_MEM: { + if( db->pnBytesFreed==0 ){ + sqlite3ValueFree((sqlite3_value*)p4); + }else{ + freeP4Mem(db, (Mem*)p4); + } + break; + } + case P4_VTAB : { + if( db->pnBytesFreed==0 ) sqlite3VtabUnlock((VTable *)p4); + break; + } + } +} + +/* +** Free the space allocated for aOp and any p4 values allocated for the +** opcodes contained within. If aOp is not NULL it is assumed to contain +** nOp entries. +*/ +static void vdbeFreeOpArray(sqlite3 *db, Op *aOp, int nOp){ + if( aOp ){ + Op *pOp; + for(pOp=&aOp[nOp-1]; pOp>=aOp; pOp--){ + if( pOp->p4type <= P4_FREE_IF_LE ) freeP4(db, pOp->p4type, pOp->p4.p); +#ifdef SQLITE_ENABLE_EXPLAIN_COMMENTS + sqlite3DbFree(db, pOp->zComment); +#endif + } + sqlite3DbFreeNN(db, aOp); + } +} + +/* +** Link the SubProgram object passed as the second argument into the linked +** list at Vdbe.pSubProgram. This list is used to delete all sub-program +** objects when the VM is no longer required. +*/ +SQLITE_PRIVATE void sqlite3VdbeLinkSubProgram(Vdbe *pVdbe, SubProgram *p){ + p->pNext = pVdbe->pProgram; + pVdbe->pProgram = p; +} + +/* +** Change the opcode at addr into OP_Noop +*/ +SQLITE_PRIVATE int sqlite3VdbeChangeToNoop(Vdbe *p, int addr){ + VdbeOp *pOp; + if( p->db->mallocFailed ) return 0; + assert( addr>=0 && addrnOp ); + pOp = &p->aOp[addr]; + freeP4(p->db, pOp->p4type, pOp->p4.p); + pOp->p4type = P4_NOTUSED; + pOp->p4.z = 0; + pOp->opcode = OP_Noop; + return 1; +} + +/* +** If the last opcode is "op" and it is not a jump destination, +** then remove it. Return true if and only if an opcode was removed. +*/ +SQLITE_PRIVATE int sqlite3VdbeDeletePriorOpcode(Vdbe *p, u8 op){ + if( p->nOp>0 && p->aOp[p->nOp-1].opcode==op ){ + return sqlite3VdbeChangeToNoop(p, p->nOp-1); + }else{ + return 0; + } +} + +/* +** Change the value of the P4 operand for a specific instruction. +** This routine is useful when a large program is loaded from a +** static array using sqlite3VdbeAddOpList but we want to make a +** few minor changes to the program. +** +** If n>=0 then the P4 operand is dynamic, meaning that a copy of +** the string is made into memory obtained from sqlite3_malloc(). +** A value of n==0 means copy bytes of zP4 up to and including the +** first null byte. If n>0 then copy n+1 bytes of zP4. +** +** Other values of n (P4_STATIC, P4_COLLSEQ etc.) indicate that zP4 points +** to a string or structure that is guaranteed to exist for the lifetime of +** the Vdbe. In these cases we can just copy the pointer. +** +** If addr<0 then change P4 on the most recently inserted instruction. +*/ +static void SQLITE_NOINLINE vdbeChangeP4Full( + Vdbe *p, + Op *pOp, + const char *zP4, + int n +){ + if( pOp->p4type ){ + freeP4(p->db, pOp->p4type, pOp->p4.p); + pOp->p4type = 0; + pOp->p4.p = 0; + } + if( n<0 ){ + sqlite3VdbeChangeP4(p, (int)(pOp - p->aOp), zP4, n); + }else{ + if( n==0 ) n = sqlite3Strlen30(zP4); + pOp->p4.z = sqlite3DbStrNDup(p->db, zP4, n); + pOp->p4type = P4_DYNAMIC; + } +} +SQLITE_PRIVATE void sqlite3VdbeChangeP4(Vdbe *p, int addr, const char *zP4, int n){ + Op *pOp; + sqlite3 *db; + assert( p!=0 ); + db = p->db; + assert( p->magic==VDBE_MAGIC_INIT ); + assert( p->aOp!=0 || db->mallocFailed ); + if( db->mallocFailed ){ + if( n!=P4_VTAB ) freeP4(db, n, (void*)*(char**)&zP4); + return; + } + assert( p->nOp>0 ); + assert( addrnOp ); + if( addr<0 ){ + addr = p->nOp - 1; + } + pOp = &p->aOp[addr]; + if( n>=0 || pOp->p4type ){ + vdbeChangeP4Full(p, pOp, zP4, n); + return; + } + if( n==P4_INT32 ){ + /* Note: this cast is safe, because the origin data point was an int + ** that was cast to a (const char *). */ + pOp->p4.i = SQLITE_PTR_TO_INT(zP4); + pOp->p4type = P4_INT32; + }else if( zP4!=0 ){ + assert( n<0 ); + pOp->p4.p = (void*)zP4; + pOp->p4type = (signed char)n; + if( n==P4_VTAB ) sqlite3VtabLock((VTable*)zP4); + } +} + +/* +** Change the P4 operand of the most recently coded instruction +** to the value defined by the arguments. This is a high-speed +** version of sqlite3VdbeChangeP4(). +** +** The P4 operand must not have been previously defined. And the new +** P4 must not be P4_INT32. Use sqlite3VdbeChangeP4() in either of +** those cases. +*/ +SQLITE_PRIVATE void sqlite3VdbeAppendP4(Vdbe *p, void *pP4, int n){ + VdbeOp *pOp; + assert( n!=P4_INT32 && n!=P4_VTAB ); + assert( n<=0 ); + if( p->db->mallocFailed ){ + freeP4(p->db, n, pP4); + }else{ + assert( pP4!=0 ); + assert( p->nOp>0 ); + pOp = &p->aOp[p->nOp-1]; + assert( pOp->p4type==P4_NOTUSED ); + pOp->p4type = n; + pOp->p4.p = pP4; + } +} + +/* +** Set the P4 on the most recently added opcode to the KeyInfo for the +** index given. +*/ +SQLITE_PRIVATE void sqlite3VdbeSetP4KeyInfo(Parse *pParse, Index *pIdx){ + Vdbe *v = pParse->pVdbe; + KeyInfo *pKeyInfo; + assert( v!=0 ); + assert( pIdx!=0 ); + pKeyInfo = sqlite3KeyInfoOfIndex(pParse, pIdx); + if( pKeyInfo ) sqlite3VdbeAppendP4(v, pKeyInfo, P4_KEYINFO); +} + +#ifdef SQLITE_ENABLE_EXPLAIN_COMMENTS +/* +** Change the comment on the most recently coded instruction. Or +** insert a No-op and add the comment to that new instruction. This +** makes the code easier to read during debugging. None of this happens +** in a production build. +*/ +static void vdbeVComment(Vdbe *p, const char *zFormat, va_list ap){ + assert( p->nOp>0 || p->aOp==0 ); + assert( p->aOp==0 || p->aOp[p->nOp-1].zComment==0 || p->db->mallocFailed ); + if( p->nOp ){ + assert( p->aOp ); + sqlite3DbFree(p->db, p->aOp[p->nOp-1].zComment); + p->aOp[p->nOp-1].zComment = sqlite3VMPrintf(p->db, zFormat, ap); + } +} +SQLITE_PRIVATE void sqlite3VdbeComment(Vdbe *p, const char *zFormat, ...){ + va_list ap; + if( p ){ + va_start(ap, zFormat); + vdbeVComment(p, zFormat, ap); + va_end(ap); + } +} +SQLITE_PRIVATE void sqlite3VdbeNoopComment(Vdbe *p, const char *zFormat, ...){ + va_list ap; + if( p ){ + sqlite3VdbeAddOp0(p, OP_Noop); + va_start(ap, zFormat); + vdbeVComment(p, zFormat, ap); + va_end(ap); + } +} +#endif /* NDEBUG */ + +#ifdef SQLITE_VDBE_COVERAGE +/* +** Set the value if the iSrcLine field for the previously coded instruction. +*/ +SQLITE_PRIVATE void sqlite3VdbeSetLineNumber(Vdbe *v, int iLine){ + sqlite3VdbeGetOp(v,-1)->iSrcLine = iLine; +} +#endif /* SQLITE_VDBE_COVERAGE */ + +/* +** Return the opcode for a given address. If the address is -1, then +** return the most recently inserted opcode. +** +** If a memory allocation error has occurred prior to the calling of this +** routine, then a pointer to a dummy VdbeOp will be returned. That opcode +** is readable but not writable, though it is cast to a writable value. +** The return of a dummy opcode allows the call to continue functioning +** after an OOM fault without having to check to see if the return from +** this routine is a valid pointer. But because the dummy.opcode is 0, +** dummy will never be written to. This is verified by code inspection and +** by running with Valgrind. +*/ +SQLITE_PRIVATE VdbeOp *sqlite3VdbeGetOp(Vdbe *p, int addr){ + /* C89 specifies that the constant "dummy" will be initialized to all + ** zeros, which is correct. MSVC generates a warning, nevertheless. */ + static VdbeOp dummy; /* Ignore the MSVC warning about no initializer */ + assert( p->magic==VDBE_MAGIC_INIT ); + if( addr<0 ){ + addr = p->nOp - 1; + } + assert( (addr>=0 && addrnOp) || p->db->mallocFailed ); + if( p->db->mallocFailed ){ + return (VdbeOp*)&dummy; + }else{ + return &p->aOp[addr]; + } +} + +#if defined(SQLITE_ENABLE_EXPLAIN_COMMENTS) +/* +** Return an integer value for one of the parameters to the opcode pOp +** determined by character c. +*/ +static int translateP(char c, const Op *pOp){ + if( c=='1' ) return pOp->p1; + if( c=='2' ) return pOp->p2; + if( c=='3' ) return pOp->p3; + if( c=='4' ) return pOp->p4.i; + return pOp->p5; +} + +/* +** Compute a string for the "comment" field of a VDBE opcode listing. +** +** The Synopsis: field in comments in the vdbe.c source file gets converted +** to an extra string that is appended to the sqlite3OpcodeName(). In the +** absence of other comments, this synopsis becomes the comment on the opcode. +** Some translation occurs: +** +** "PX" -> "r[X]" +** "PX@PY" -> "r[X..X+Y-1]" or "r[x]" if y is 0 or 1 +** "PX@PY+1" -> "r[X..X+Y]" or "r[x]" if y is 0 +** "PY..PY" -> "r[X..Y]" or "r[x]" if y<=x +*/ +static int displayComment( + const Op *pOp, /* The opcode to be commented */ + const char *zP4, /* Previously obtained value for P4 */ + char *zTemp, /* Write result here */ + int nTemp /* Space available in zTemp[] */ +){ + const char *zOpName; + const char *zSynopsis; + int nOpName; + int ii, jj; + char zAlt[50]; + zOpName = sqlite3OpcodeName(pOp->opcode); + nOpName = sqlite3Strlen30(zOpName); + if( zOpName[nOpName+1] ){ + int seenCom = 0; + char c; + zSynopsis = zOpName += nOpName + 1; + if( strncmp(zSynopsis,"IF ",3)==0 ){ + if( pOp->p5 & SQLITE_STOREP2 ){ + sqlite3_snprintf(sizeof(zAlt), zAlt, "r[P2] = (%s)", zSynopsis+3); + }else{ + sqlite3_snprintf(sizeof(zAlt), zAlt, "if %s goto P2", zSynopsis+3); + } + zSynopsis = zAlt; + } + for(ii=jj=0; jjzComment); + seenCom = 1; + }else{ + int v1 = translateP(c, pOp); + int v2; + sqlite3_snprintf(nTemp-jj, zTemp+jj, "%d", v1); + if( strncmp(zSynopsis+ii+1, "@P", 2)==0 ){ + ii += 3; + jj += sqlite3Strlen30(zTemp+jj); + v2 = translateP(zSynopsis[ii], pOp); + if( strncmp(zSynopsis+ii+1,"+1",2)==0 ){ + ii += 2; + v2++; + } + if( v2>1 ){ + sqlite3_snprintf(nTemp-jj, zTemp+jj, "..%d", v1+v2-1); + } + }else if( strncmp(zSynopsis+ii+1, "..P3", 4)==0 && pOp->p3==0 ){ + ii += 4; + } + } + jj += sqlite3Strlen30(zTemp+jj); + }else{ + zTemp[jj++] = c; + } + } + if( !seenCom && jjzComment ){ + sqlite3_snprintf(nTemp-jj, zTemp+jj, "; %s", pOp->zComment); + jj += sqlite3Strlen30(zTemp+jj); + } + if( jjzComment ){ + sqlite3_snprintf(nTemp, zTemp, "%s", pOp->zComment); + jj = sqlite3Strlen30(zTemp); + }else{ + zTemp[0] = 0; + jj = 0; + } + return jj; +} +#endif /* SQLITE_DEBUG */ + +#if VDBE_DISPLAY_P4 && defined(SQLITE_ENABLE_CURSOR_HINTS) +/* +** Translate the P4.pExpr value for an OP_CursorHint opcode into text +** that can be displayed in the P4 column of EXPLAIN output. +*/ +static void displayP4Expr(StrAccum *p, Expr *pExpr){ + const char *zOp = 0; + switch( pExpr->op ){ + case TK_STRING: + sqlite3_str_appendf(p, "%Q", pExpr->u.zToken); + break; + case TK_INTEGER: + sqlite3_str_appendf(p, "%d", pExpr->u.iValue); + break; + case TK_NULL: + sqlite3_str_appendf(p, "NULL"); + break; + case TK_REGISTER: { + sqlite3_str_appendf(p, "r[%d]", pExpr->iTable); + break; + } + case TK_COLUMN: { + if( pExpr->iColumn<0 ){ + sqlite3_str_appendf(p, "rowid"); + }else{ + sqlite3_str_appendf(p, "c%d", (int)pExpr->iColumn); + } + break; + } + case TK_LT: zOp = "LT"; break; + case TK_LE: zOp = "LE"; break; + case TK_GT: zOp = "GT"; break; + case TK_GE: zOp = "GE"; break; + case TK_NE: zOp = "NE"; break; + case TK_EQ: zOp = "EQ"; break; + case TK_IS: zOp = "IS"; break; + case TK_ISNOT: zOp = "ISNOT"; break; + case TK_AND: zOp = "AND"; break; + case TK_OR: zOp = "OR"; break; + case TK_PLUS: zOp = "ADD"; break; + case TK_STAR: zOp = "MUL"; break; + case TK_MINUS: zOp = "SUB"; break; + case TK_REM: zOp = "REM"; break; + case TK_BITAND: zOp = "BITAND"; break; + case TK_BITOR: zOp = "BITOR"; break; + case TK_SLASH: zOp = "DIV"; break; + case TK_LSHIFT: zOp = "LSHIFT"; break; + case TK_RSHIFT: zOp = "RSHIFT"; break; + case TK_CONCAT: zOp = "CONCAT"; break; + case TK_UMINUS: zOp = "MINUS"; break; + case TK_UPLUS: zOp = "PLUS"; break; + case TK_BITNOT: zOp = "BITNOT"; break; + case TK_NOT: zOp = "NOT"; break; + case TK_ISNULL: zOp = "ISNULL"; break; + case TK_NOTNULL: zOp = "NOTNULL"; break; + + default: + sqlite3_str_appendf(p, "%s", "expr"); + break; + } + + if( zOp ){ + sqlite3_str_appendf(p, "%s(", zOp); + displayP4Expr(p, pExpr->pLeft); + if( pExpr->pRight ){ + sqlite3_str_append(p, ",", 1); + displayP4Expr(p, pExpr->pRight); + } + sqlite3_str_append(p, ")", 1); + } +} +#endif /* VDBE_DISPLAY_P4 && defined(SQLITE_ENABLE_CURSOR_HINTS) */ + + +#if VDBE_DISPLAY_P4 +/* +** Compute a string that describes the P4 parameter for an opcode. +** Use zTemp for any required temporary buffer space. +*/ +static char *displayP4(Op *pOp, char *zTemp, int nTemp){ + char *zP4 = zTemp; + StrAccum x; + assert( nTemp>=20 ); + sqlite3StrAccumInit(&x, 0, zTemp, nTemp, 0); + switch( pOp->p4type ){ + case P4_KEYINFO: { + int j; + KeyInfo *pKeyInfo = pOp->p4.pKeyInfo; + assert( pKeyInfo->aSortFlags!=0 ); + sqlite3_str_appendf(&x, "k(%d", pKeyInfo->nKeyField); + for(j=0; jnKeyField; j++){ + CollSeq *pColl = pKeyInfo->aColl[j]; + const char *zColl = pColl ? pColl->zName : ""; + if( strcmp(zColl, "BINARY")==0 ) zColl = "B"; + sqlite3_str_appendf(&x, ",%s%s%s", + (pKeyInfo->aSortFlags[j] & KEYINFO_ORDER_DESC) ? "-" : "", + (pKeyInfo->aSortFlags[j] & KEYINFO_ORDER_BIGNULL)? "N." : "", + zColl); + } + sqlite3_str_append(&x, ")", 1); + break; + } +#ifdef SQLITE_ENABLE_CURSOR_HINTS + case P4_EXPR: { + displayP4Expr(&x, pOp->p4.pExpr); + break; + } +#endif + case P4_COLLSEQ: { + CollSeq *pColl = pOp->p4.pColl; + sqlite3_str_appendf(&x, "(%.20s)", pColl->zName); + break; + } + case P4_FUNCDEF: { + FuncDef *pDef = pOp->p4.pFunc; + sqlite3_str_appendf(&x, "%s(%d)", pDef->zName, pDef->nArg); + break; + } +#if defined(SQLITE_DEBUG) || defined(VDBE_PROFILE) + case P4_FUNCCTX: { + FuncDef *pDef = pOp->p4.pCtx->pFunc; + sqlite3_str_appendf(&x, "%s(%d)", pDef->zName, pDef->nArg); + break; + } +#endif + case P4_INT64: { + sqlite3_str_appendf(&x, "%lld", *pOp->p4.pI64); + break; + } + case P4_INT32: { + sqlite3_str_appendf(&x, "%d", pOp->p4.i); + break; + } + case P4_REAL: { + sqlite3_str_appendf(&x, "%.16g", *pOp->p4.pReal); + break; + } + case P4_MEM: { + Mem *pMem = pOp->p4.pMem; + if( pMem->flags & MEM_Str ){ + zP4 = pMem->z; + }else if( pMem->flags & (MEM_Int|MEM_IntReal) ){ + sqlite3_str_appendf(&x, "%lld", pMem->u.i); + }else if( pMem->flags & MEM_Real ){ + sqlite3_str_appendf(&x, "%.16g", pMem->u.r); + }else if( pMem->flags & MEM_Null ){ + zP4 = "NULL"; + }else{ + assert( pMem->flags & MEM_Blob ); + zP4 = "(blob)"; + } + break; + } +#ifndef SQLITE_OMIT_VIRTUALTABLE + case P4_VTAB: { + sqlite3_vtab *pVtab = pOp->p4.pVtab->pVtab; + sqlite3_str_appendf(&x, "vtab:%p", pVtab); + break; + } +#endif + case P4_INTARRAY: { + int i; + int *ai = pOp->p4.ai; + int n = ai[0]; /* The first element of an INTARRAY is always the + ** count of the number of elements to follow */ + for(i=1; i<=n; i++){ + sqlite3_str_appendf(&x, ",%d", ai[i]); + } + zTemp[0] = '['; + sqlite3_str_append(&x, "]", 1); + break; + } + case P4_SUBPROGRAM: { + sqlite3_str_appendf(&x, "program"); + break; + } + case P4_DYNBLOB: + case P4_ADVANCE: { + zTemp[0] = 0; + break; + } + case P4_TABLE: { + sqlite3_str_appendf(&x, "%s", pOp->p4.pTab->zName); + break; + } + default: { + zP4 = pOp->p4.z; + if( zP4==0 ){ + zP4 = zTemp; + zTemp[0] = 0; + } + } + } + sqlite3StrAccumFinish(&x); + assert( zP4!=0 ); + return zP4; +} +#endif /* VDBE_DISPLAY_P4 */ + +/* +** Declare to the Vdbe that the BTree object at db->aDb[i] is used. +** +** The prepared statements need to know in advance the complete set of +** attached databases that will be use. A mask of these databases +** is maintained in p->btreeMask. The p->lockMask value is the subset of +** p->btreeMask of databases that will require a lock. +*/ +SQLITE_PRIVATE void sqlite3VdbeUsesBtree(Vdbe *p, int i){ + assert( i>=0 && idb->nDb && i<(int)sizeof(yDbMask)*8 ); + assert( i<(int)sizeof(p->btreeMask)*8 ); + DbMaskSet(p->btreeMask, i); + if( i!=1 && sqlite3BtreeSharable(p->db->aDb[i].pBt) ){ + DbMaskSet(p->lockMask, i); + } +} + +#if !defined(SQLITE_OMIT_SHARED_CACHE) +/* +** If SQLite is compiled to support shared-cache mode and to be threadsafe, +** this routine obtains the mutex associated with each BtShared structure +** that may be accessed by the VM passed as an argument. In doing so it also +** sets the BtShared.db member of each of the BtShared structures, ensuring +** that the correct busy-handler callback is invoked if required. +** +** If SQLite is not threadsafe but does support shared-cache mode, then +** sqlite3BtreeEnter() is invoked to set the BtShared.db variables +** of all of BtShared structures accessible via the database handle +** associated with the VM. +** +** If SQLite is not threadsafe and does not support shared-cache mode, this +** function is a no-op. +** +** The p->btreeMask field is a bitmask of all btrees that the prepared +** statement p will ever use. Let N be the number of bits in p->btreeMask +** corresponding to btrees that use shared cache. Then the runtime of +** this routine is N*N. But as N is rarely more than 1, this should not +** be a problem. +*/ +SQLITE_PRIVATE void sqlite3VdbeEnter(Vdbe *p){ + int i; + sqlite3 *db; + Db *aDb; + int nDb; + if( DbMaskAllZero(p->lockMask) ) return; /* The common case */ + db = p->db; + aDb = db->aDb; + nDb = db->nDb; + for(i=0; ilockMask,i) && ALWAYS(aDb[i].pBt!=0) ){ + sqlite3BtreeEnter(aDb[i].pBt); + } + } +} +#endif + +#if !defined(SQLITE_OMIT_SHARED_CACHE) && SQLITE_THREADSAFE>0 +/* +** Unlock all of the btrees previously locked by a call to sqlite3VdbeEnter(). +*/ +static SQLITE_NOINLINE void vdbeLeave(Vdbe *p){ + int i; + sqlite3 *db; + Db *aDb; + int nDb; + db = p->db; + aDb = db->aDb; + nDb = db->nDb; + for(i=0; ilockMask,i) && ALWAYS(aDb[i].pBt!=0) ){ + sqlite3BtreeLeave(aDb[i].pBt); + } + } +} +SQLITE_PRIVATE void sqlite3VdbeLeave(Vdbe *p){ + if( DbMaskAllZero(p->lockMask) ) return; /* The common case */ + vdbeLeave(p); +} +#endif + +#if defined(VDBE_PROFILE) || defined(SQLITE_DEBUG) +/* +** Print a single opcode. This routine is used for debugging only. +*/ +SQLITE_PRIVATE void sqlite3VdbePrintOp(FILE *pOut, int pc, VdbeOp *pOp){ + char *zP4; + char zPtr[50]; + char zCom[100]; + static const char *zFormat1 = "%4d %-13s %4d %4d %4d %-13s %.2X %s\n"; + if( pOut==0 ) pOut = stdout; + zP4 = displayP4(pOp, zPtr, sizeof(zPtr)); +#ifdef SQLITE_ENABLE_EXPLAIN_COMMENTS + displayComment(pOp, zP4, zCom, sizeof(zCom)); +#else + zCom[0] = 0; +#endif + /* NB: The sqlite3OpcodeName() function is implemented by code created + ** by the mkopcodeh.awk and mkopcodec.awk scripts which extract the + ** information from the vdbe.c source text */ + fprintf(pOut, zFormat1, pc, + sqlite3OpcodeName(pOp->opcode), pOp->p1, pOp->p2, pOp->p3, zP4, pOp->p5, + zCom + ); + fflush(pOut); +} +#endif + +/* +** Initialize an array of N Mem element. +*/ +static void initMemArray(Mem *p, int N, sqlite3 *db, u16 flags){ + while( (N--)>0 ){ + p->db = db; + p->flags = flags; + p->szMalloc = 0; +#ifdef SQLITE_DEBUG + p->pScopyFrom = 0; +#endif + p++; + } +} + +/* +** Release an array of N Mem elements +*/ +static void releaseMemArray(Mem *p, int N){ + if( p && N ){ + Mem *pEnd = &p[N]; + sqlite3 *db = p->db; + if( db->pnBytesFreed ){ + do{ + if( p->szMalloc ) sqlite3DbFree(db, p->zMalloc); + }while( (++p)flags & MEM_Agg ); + testcase( p->flags & MEM_Dyn ); + testcase( p->xDel==sqlite3VdbeFrameMemDel ); + if( p->flags&(MEM_Agg|MEM_Dyn) ){ + sqlite3VdbeMemRelease(p); + }else if( p->szMalloc ){ + sqlite3DbFreeNN(db, p->zMalloc); + p->szMalloc = 0; + } + + p->flags = MEM_Undefined; + }while( (++p)iFrameMagic!=SQLITE_FRAME_MAGIC ) return 0; + return 1; +} +#endif + + +/* +** This is a destructor on a Mem object (which is really an sqlite3_value) +** that deletes the Frame object that is attached to it as a blob. +** +** This routine does not delete the Frame right away. It merely adds the +** frame to a list of frames to be deleted when the Vdbe halts. +*/ +SQLITE_PRIVATE void sqlite3VdbeFrameMemDel(void *pArg){ + VdbeFrame *pFrame = (VdbeFrame*)pArg; + assert( sqlite3VdbeFrameIsValid(pFrame) ); + pFrame->pParent = pFrame->v->pDelFrame; + pFrame->v->pDelFrame = pFrame; +} + + +/* +** Delete a VdbeFrame object and its contents. VdbeFrame objects are +** allocated by the OP_Program opcode in sqlite3VdbeExec(). +*/ +SQLITE_PRIVATE void sqlite3VdbeFrameDelete(VdbeFrame *p){ + int i; + Mem *aMem = VdbeFrameMem(p); + VdbeCursor **apCsr = (VdbeCursor **)&aMem[p->nChildMem]; + assert( sqlite3VdbeFrameIsValid(p) ); + for(i=0; inChildCsr; i++){ + sqlite3VdbeFreeCursor(p->v, apCsr[i]); + } + releaseMemArray(aMem, p->nChildMem); + sqlite3VdbeDeleteAuxData(p->v->db, &p->pAuxData, -1, 0); + sqlite3DbFree(p->v->db, p); +} + +#ifndef SQLITE_OMIT_EXPLAIN +/* +** Give a listing of the program in the virtual machine. +** +** The interface is the same as sqlite3VdbeExec(). But instead of +** running the code, it invokes the callback once for each instruction. +** This feature is used to implement "EXPLAIN". +** +** When p->explain==1, each instruction is listed. When +** p->explain==2, only OP_Explain instructions are listed and these +** are shown in a different format. p->explain==2 is used to implement +** EXPLAIN QUERY PLAN. +** 2018-04-24: In p->explain==2 mode, the OP_Init opcodes of triggers +** are also shown, so that the boundaries between the main program and +** each trigger are clear. +** +** When p->explain==1, first the main program is listed, then each of +** the trigger subprograms are listed one by one. +*/ +SQLITE_PRIVATE int sqlite3VdbeList( + Vdbe *p /* The VDBE */ +){ + int nRow; /* Stop when row count reaches this */ + int nSub = 0; /* Number of sub-vdbes seen so far */ + SubProgram **apSub = 0; /* Array of sub-vdbes */ + Mem *pSub = 0; /* Memory cell hold array of subprogs */ + sqlite3 *db = p->db; /* The database connection */ + int i; /* Loop counter */ + int rc = SQLITE_OK; /* Return code */ + Mem *pMem = &p->aMem[1]; /* First Mem of result set */ + int bListSubprogs = (p->explain==1 || (db->flags & SQLITE_TriggerEQP)!=0); + Op *pOp = 0; + + assert( p->explain ); + assert( p->magic==VDBE_MAGIC_RUN ); + assert( p->rc==SQLITE_OK || p->rc==SQLITE_BUSY || p->rc==SQLITE_NOMEM ); + + /* Even though this opcode does not use dynamic strings for + ** the result, result columns may become dynamic if the user calls + ** sqlite3_column_text16(), causing a translation to UTF-16 encoding. + */ + releaseMemArray(pMem, 8); + p->pResultSet = 0; + + if( p->rc==SQLITE_NOMEM ){ + /* This happens if a malloc() inside a call to sqlite3_column_text() or + ** sqlite3_column_text16() failed. */ + sqlite3OomFault(db); + return SQLITE_ERROR; + } + + /* When the number of output rows reaches nRow, that means the + ** listing has finished and sqlite3_step() should return SQLITE_DONE. + ** nRow is the sum of the number of rows in the main program, plus + ** the sum of the number of rows in all trigger subprograms encountered + ** so far. The nRow value will increase as new trigger subprograms are + ** encountered, but p->pc will eventually catch up to nRow. + */ + nRow = p->nOp; + if( bListSubprogs ){ + /* The first 8 memory cells are used for the result set. So we will + ** commandeer the 9th cell to use as storage for an array of pointers + ** to trigger subprograms. The VDBE is guaranteed to have at least 9 + ** cells. */ + assert( p->nMem>9 ); + pSub = &p->aMem[9]; + if( pSub->flags&MEM_Blob ){ + /* On the first call to sqlite3_step(), pSub will hold a NULL. It is + ** initialized to a BLOB by the P4_SUBPROGRAM processing logic below */ + nSub = pSub->n/sizeof(Vdbe*); + apSub = (SubProgram **)pSub->z; + } + for(i=0; inOp; + } + } + + while(1){ /* Loop exits via break */ + i = p->pc++; + if( i>=nRow ){ + p->rc = SQLITE_OK; + rc = SQLITE_DONE; + break; + } + if( inOp ){ + /* The output line number is small enough that we are still in the + ** main program. */ + pOp = &p->aOp[i]; + }else{ + /* We are currently listing subprograms. Figure out which one and + ** pick up the appropriate opcode. */ + int j; + i -= p->nOp; + assert( apSub!=0 ); + assert( nSub>0 ); + for(j=0; i>=apSub[j]->nOp; j++){ + i -= apSub[j]->nOp; + assert( inOp || j+1aOp[i]; + } + + /* When an OP_Program opcode is encounter (the only opcode that has + ** a P4_SUBPROGRAM argument), expand the size of the array of subprograms + ** kept in p->aMem[9].z to hold the new program - assuming this subprogram + ** has not already been seen. + */ + if( bListSubprogs && pOp->p4type==P4_SUBPROGRAM ){ + int nByte = (nSub+1)*sizeof(SubProgram*); + int j; + for(j=0; jp4.pProgram ) break; + } + if( j==nSub ){ + p->rc = sqlite3VdbeMemGrow(pSub, nByte, nSub!=0); + if( p->rc!=SQLITE_OK ){ + rc = SQLITE_ERROR; + break; + } + apSub = (SubProgram **)pSub->z; + apSub[nSub++] = pOp->p4.pProgram; + pSub->flags |= MEM_Blob; + pSub->n = nSub*sizeof(SubProgram*); + nRow += pOp->p4.pProgram->nOp; + } + } + if( p->explain<2 ) break; + if( pOp->opcode==OP_Explain ) break; + if( pOp->opcode==OP_Init && p->pc>1 ) break; + } + + if( rc==SQLITE_OK ){ + if( db->u1.isInterrupted ){ + p->rc = SQLITE_INTERRUPT; + rc = SQLITE_ERROR; + sqlite3VdbeError(p, sqlite3ErrStr(p->rc)); + }else{ + char *zP4; + if( p->explain==1 ){ + pMem->flags = MEM_Int; + pMem->u.i = i; /* Program counter */ + pMem++; + + pMem->flags = MEM_Static|MEM_Str|MEM_Term; + pMem->z = (char*)sqlite3OpcodeName(pOp->opcode); /* Opcode */ + assert( pMem->z!=0 ); + pMem->n = sqlite3Strlen30(pMem->z); + pMem->enc = SQLITE_UTF8; + pMem++; + } + + pMem->flags = MEM_Int; + pMem->u.i = pOp->p1; /* P1 */ + pMem++; + + pMem->flags = MEM_Int; + pMem->u.i = pOp->p2; /* P2 */ + pMem++; + + pMem->flags = MEM_Int; + pMem->u.i = pOp->p3; /* P3 */ + pMem++; + + if( sqlite3VdbeMemClearAndResize(pMem, 100) ){ /* P4 */ + assert( p->db->mallocFailed ); + return SQLITE_ERROR; + } + pMem->flags = MEM_Str|MEM_Term; + zP4 = displayP4(pOp, pMem->z, pMem->szMalloc); + if( zP4!=pMem->z ){ + pMem->n = 0; + sqlite3VdbeMemSetStr(pMem, zP4, -1, SQLITE_UTF8, 0); + }else{ + assert( pMem->z!=0 ); + pMem->n = sqlite3Strlen30(pMem->z); + pMem->enc = SQLITE_UTF8; + } + pMem++; + + if( p->explain==1 ){ + if( sqlite3VdbeMemClearAndResize(pMem, 4) ){ + assert( p->db->mallocFailed ); + return SQLITE_ERROR; + } + pMem->flags = MEM_Str|MEM_Term; + pMem->n = 2; + sqlite3_snprintf(3, pMem->z, "%.2x", pOp->p5); /* P5 */ + pMem->enc = SQLITE_UTF8; + pMem++; + +#ifdef SQLITE_ENABLE_EXPLAIN_COMMENTS + if( sqlite3VdbeMemClearAndResize(pMem, 500) ){ + assert( p->db->mallocFailed ); + return SQLITE_ERROR; + } + pMem->flags = MEM_Str|MEM_Term; + pMem->n = displayComment(pOp, zP4, pMem->z, 500); + pMem->enc = SQLITE_UTF8; +#else + pMem->flags = MEM_Null; /* Comment */ +#endif + } + + p->nResColumn = 8 - 4*(p->explain-1); + p->pResultSet = &p->aMem[1]; + p->rc = SQLITE_OK; + rc = SQLITE_ROW; + } + } + return rc; +} +#endif /* SQLITE_OMIT_EXPLAIN */ + +#ifdef SQLITE_DEBUG +/* +** Print the SQL that was used to generate a VDBE program. +*/ +SQLITE_PRIVATE void sqlite3VdbePrintSql(Vdbe *p){ + const char *z = 0; + if( p->zSql ){ + z = p->zSql; + }else if( p->nOp>=1 ){ + const VdbeOp *pOp = &p->aOp[0]; + if( pOp->opcode==OP_Init && pOp->p4.z!=0 ){ + z = pOp->p4.z; + while( sqlite3Isspace(*z) ) z++; + } + } + if( z ) printf("SQL: [%s]\n", z); +} +#endif + +#if !defined(SQLITE_OMIT_TRACE) && defined(SQLITE_ENABLE_IOTRACE) +/* +** Print an IOTRACE message showing SQL content. +*/ +SQLITE_PRIVATE void sqlite3VdbeIOTraceSql(Vdbe *p){ + int nOp = p->nOp; + VdbeOp *pOp; + if( sqlite3IoTrace==0 ) return; + if( nOp<1 ) return; + pOp = &p->aOp[0]; + if( pOp->opcode==OP_Init && pOp->p4.z!=0 ){ + int i, j; + char z[1000]; + sqlite3_snprintf(sizeof(z), z, "%s", pOp->p4.z); + for(i=0; sqlite3Isspace(z[i]); i++){} + for(j=0; z[i]; i++){ + if( sqlite3Isspace(z[i]) ){ + if( z[i-1]!=' ' ){ + z[j++] = ' '; + } + }else{ + z[j++] = z[i]; + } + } + z[j] = 0; + sqlite3IoTrace("SQL %s\n", z); + } +} +#endif /* !SQLITE_OMIT_TRACE && SQLITE_ENABLE_IOTRACE */ + +/* An instance of this object describes bulk memory available for use +** by subcomponents of a prepared statement. Space is allocated out +** of a ReusableSpace object by the allocSpace() routine below. +*/ +struct ReusableSpace { + u8 *pSpace; /* Available memory */ + sqlite3_int64 nFree; /* Bytes of available memory */ + sqlite3_int64 nNeeded; /* Total bytes that could not be allocated */ +}; + +/* Try to allocate nByte bytes of 8-byte aligned bulk memory for pBuf +** from the ReusableSpace object. Return a pointer to the allocated +** memory on success. If insufficient memory is available in the +** ReusableSpace object, increase the ReusableSpace.nNeeded +** value by the amount needed and return NULL. +** +** If pBuf is not initially NULL, that means that the memory has already +** been allocated by a prior call to this routine, so just return a copy +** of pBuf and leave ReusableSpace unchanged. +** +** This allocator is employed to repurpose unused slots at the end of the +** opcode array of prepared state for other memory needs of the prepared +** statement. +*/ +static void *allocSpace( + struct ReusableSpace *p, /* Bulk memory available for allocation */ + void *pBuf, /* Pointer to a prior allocation */ + sqlite3_int64 nByte /* Bytes of memory needed */ +){ + assert( EIGHT_BYTE_ALIGNMENT(p->pSpace) ); + if( pBuf==0 ){ + nByte = ROUND8(nByte); + if( nByte <= p->nFree ){ + p->nFree -= nByte; + pBuf = &p->pSpace[p->nFree]; + }else{ + p->nNeeded += nByte; + } + } + assert( EIGHT_BYTE_ALIGNMENT(pBuf) ); + return pBuf; +} + +/* +** Rewind the VDBE back to the beginning in preparation for +** running it. +*/ +SQLITE_PRIVATE void sqlite3VdbeRewind(Vdbe *p){ +#if defined(SQLITE_DEBUG) || defined(VDBE_PROFILE) + int i; +#endif + assert( p!=0 ); + assert( p->magic==VDBE_MAGIC_INIT || p->magic==VDBE_MAGIC_RESET ); + + /* There should be at least one opcode. + */ + assert( p->nOp>0 ); + + /* Set the magic to VDBE_MAGIC_RUN sooner rather than later. */ + p->magic = VDBE_MAGIC_RUN; + +#ifdef SQLITE_DEBUG + for(i=0; inMem; i++){ + assert( p->aMem[i].db==p->db ); + } +#endif + p->pc = -1; + p->rc = SQLITE_OK; + p->errorAction = OE_Abort; + p->nChange = 0; + p->cacheCtr = 1; + p->minWriteFileFormat = 255; + p->iStatement = 0; + p->nFkConstraint = 0; +#ifdef VDBE_PROFILE + for(i=0; inOp; i++){ + p->aOp[i].cnt = 0; + p->aOp[i].cycles = 0; + } +#endif +} + +/* +** Prepare a virtual machine for execution for the first time after +** creating the virtual machine. This involves things such +** as allocating registers and initializing the program counter. +** After the VDBE has be prepped, it can be executed by one or more +** calls to sqlite3VdbeExec(). +** +** This function may be called exactly once on each virtual machine. +** After this routine is called the VM has been "packaged" and is ready +** to run. After this routine is called, further calls to +** sqlite3VdbeAddOp() functions are prohibited. This routine disconnects +** the Vdbe from the Parse object that helped generate it so that the +** the Vdbe becomes an independent entity and the Parse object can be +** destroyed. +** +** Use the sqlite3VdbeRewind() procedure to restore a virtual machine back +** to its initial state after it has been run. +*/ +SQLITE_PRIVATE void sqlite3VdbeMakeReady( + Vdbe *p, /* The VDBE */ + Parse *pParse /* Parsing context */ +){ + sqlite3 *db; /* The database connection */ + int nVar; /* Number of parameters */ + int nMem; /* Number of VM memory registers */ + int nCursor; /* Number of cursors required */ + int nArg; /* Number of arguments in subprograms */ + int n; /* Loop counter */ + struct ReusableSpace x; /* Reusable bulk memory */ + + assert( p!=0 ); + assert( p->nOp>0 ); + assert( pParse!=0 ); + assert( p->magic==VDBE_MAGIC_INIT ); + assert( pParse==p->pParse ); + db = p->db; + assert( db->mallocFailed==0 ); + nVar = pParse->nVar; + nMem = pParse->nMem; + nCursor = pParse->nTab; + nArg = pParse->nMaxArg; + + /* Each cursor uses a memory cell. The first cursor (cursor 0) can + ** use aMem[0] which is not otherwise used by the VDBE program. Allocate + ** space at the end of aMem[] for cursors 1 and greater. + ** See also: allocateCursor(). + */ + nMem += nCursor; + if( nCursor==0 && nMem>0 ) nMem++; /* Space for aMem[0] even if not used */ + + /* Figure out how much reusable memory is available at the end of the + ** opcode array. This extra memory will be reallocated for other elements + ** of the prepared statement. + */ + n = ROUND8(sizeof(Op)*p->nOp); /* Bytes of opcode memory used */ + x.pSpace = &((u8*)p->aOp)[n]; /* Unused opcode memory */ + assert( EIGHT_BYTE_ALIGNMENT(x.pSpace) ); + x.nFree = ROUNDDOWN8(pParse->szOpAlloc - n); /* Bytes of unused memory */ + assert( x.nFree>=0 ); + assert( EIGHT_BYTE_ALIGNMENT(&x.pSpace[x.nFree]) ); + + resolveP2Values(p, &nArg); + p->usesStmtJournal = (u8)(pParse->isMultiWrite && pParse->mayAbort); + if( pParse->explain && nMem<10 ){ + nMem = 10; + } + p->expired = 0; + + /* Memory for registers, parameters, cursor, etc, is allocated in one or two + ** passes. On the first pass, we try to reuse unused memory at the + ** end of the opcode array. If we are unable to satisfy all memory + ** requirements by reusing the opcode array tail, then the second + ** pass will fill in the remainder using a fresh memory allocation. + ** + ** This two-pass approach that reuses as much memory as possible from + ** the leftover memory at the end of the opcode array. This can significantly + ** reduce the amount of memory held by a prepared statement. + */ + x.nNeeded = 0; + p->aMem = allocSpace(&x, 0, nMem*sizeof(Mem)); + p->aVar = allocSpace(&x, 0, nVar*sizeof(Mem)); + p->apArg = allocSpace(&x, 0, nArg*sizeof(Mem*)); + p->apCsr = allocSpace(&x, 0, nCursor*sizeof(VdbeCursor*)); +#ifdef SQLITE_ENABLE_STMT_SCANSTATUS + p->anExec = allocSpace(&x, 0, p->nOp*sizeof(i64)); +#endif + if( x.nNeeded ){ + x.pSpace = p->pFree = sqlite3DbMallocRawNN(db, x.nNeeded); + x.nFree = x.nNeeded; + if( !db->mallocFailed ){ + p->aMem = allocSpace(&x, p->aMem, nMem*sizeof(Mem)); + p->aVar = allocSpace(&x, p->aVar, nVar*sizeof(Mem)); + p->apArg = allocSpace(&x, p->apArg, nArg*sizeof(Mem*)); + p->apCsr = allocSpace(&x, p->apCsr, nCursor*sizeof(VdbeCursor*)); +#ifdef SQLITE_ENABLE_STMT_SCANSTATUS + p->anExec = allocSpace(&x, p->anExec, p->nOp*sizeof(i64)); +#endif + } + } + + p->pVList = pParse->pVList; + pParse->pVList = 0; + p->explain = pParse->explain; + if( db->mallocFailed ){ + p->nVar = 0; + p->nCursor = 0; + p->nMem = 0; + }else{ + p->nCursor = nCursor; + p->nVar = (ynVar)nVar; + initMemArray(p->aVar, nVar, db, MEM_Null); + p->nMem = nMem; + initMemArray(p->aMem, nMem, db, MEM_Undefined); + memset(p->apCsr, 0, nCursor*sizeof(VdbeCursor*)); +#ifdef SQLITE_ENABLE_STMT_SCANSTATUS + memset(p->anExec, 0, p->nOp*sizeof(i64)); +#endif + } + sqlite3VdbeRewind(p); +} + +/* +** Close a VDBE cursor and release all the resources that cursor +** happens to hold. +*/ +SQLITE_PRIVATE void sqlite3VdbeFreeCursor(Vdbe *p, VdbeCursor *pCx){ + if( pCx==0 ){ + return; + } + assert( pCx->pBtx==0 || pCx->eCurType==CURTYPE_BTREE ); + switch( pCx->eCurType ){ + case CURTYPE_SORTER: { + sqlite3VdbeSorterClose(p->db, pCx); + break; + } + case CURTYPE_BTREE: { + if( pCx->isEphemeral ){ + if( pCx->pBtx ) sqlite3BtreeClose(pCx->pBtx); + /* The pCx->pCursor will be close automatically, if it exists, by + ** the call above. */ + }else{ + assert( pCx->uc.pCursor!=0 ); + sqlite3BtreeCloseCursor(pCx->uc.pCursor); + } + break; + } +#ifndef SQLITE_OMIT_VIRTUALTABLE + case CURTYPE_VTAB: { + sqlite3_vtab_cursor *pVCur = pCx->uc.pVCur; + const sqlite3_module *pModule = pVCur->pVtab->pModule; + assert( pVCur->pVtab->nRef>0 ); + pVCur->pVtab->nRef--; + pModule->xClose(pVCur); + break; + } +#endif + } +} + +/* +** Close all cursors in the current frame. +*/ +static void closeCursorsInFrame(Vdbe *p){ + if( p->apCsr ){ + int i; + for(i=0; inCursor; i++){ + VdbeCursor *pC = p->apCsr[i]; + if( pC ){ + sqlite3VdbeFreeCursor(p, pC); + p->apCsr[i] = 0; + } + } + } +} + +/* +** Copy the values stored in the VdbeFrame structure to its Vdbe. This +** is used, for example, when a trigger sub-program is halted to restore +** control to the main program. +*/ +SQLITE_PRIVATE int sqlite3VdbeFrameRestore(VdbeFrame *pFrame){ + Vdbe *v = pFrame->v; + closeCursorsInFrame(v); +#ifdef SQLITE_ENABLE_STMT_SCANSTATUS + v->anExec = pFrame->anExec; +#endif + v->aOp = pFrame->aOp; + v->nOp = pFrame->nOp; + v->aMem = pFrame->aMem; + v->nMem = pFrame->nMem; + v->apCsr = pFrame->apCsr; + v->nCursor = pFrame->nCursor; + v->db->lastRowid = pFrame->lastRowid; + v->nChange = pFrame->nChange; + v->db->nChange = pFrame->nDbChange; + sqlite3VdbeDeleteAuxData(v->db, &v->pAuxData, -1, 0); + v->pAuxData = pFrame->pAuxData; + pFrame->pAuxData = 0; + return pFrame->pc; +} + +/* +** Close all cursors. +** +** Also release any dynamic memory held by the VM in the Vdbe.aMem memory +** cell array. This is necessary as the memory cell array may contain +** pointers to VdbeFrame objects, which may in turn contain pointers to +** open cursors. +*/ +static void closeAllCursors(Vdbe *p){ + if( p->pFrame ){ + VdbeFrame *pFrame; + for(pFrame=p->pFrame; pFrame->pParent; pFrame=pFrame->pParent); + sqlite3VdbeFrameRestore(pFrame); + p->pFrame = 0; + p->nFrame = 0; + } + assert( p->nFrame==0 ); + closeCursorsInFrame(p); + if( p->aMem ){ + releaseMemArray(p->aMem, p->nMem); + } + while( p->pDelFrame ){ + VdbeFrame *pDel = p->pDelFrame; + p->pDelFrame = pDel->pParent; + sqlite3VdbeFrameDelete(pDel); + } + + /* Delete any auxdata allocations made by the VM */ + if( p->pAuxData ) sqlite3VdbeDeleteAuxData(p->db, &p->pAuxData, -1, 0); + assert( p->pAuxData==0 ); +} + +/* +** Set the number of result columns that will be returned by this SQL +** statement. This is now set at compile time, rather than during +** execution of the vdbe program so that sqlite3_column_count() can +** be called on an SQL statement before sqlite3_step(). +*/ +SQLITE_PRIVATE void sqlite3VdbeSetNumCols(Vdbe *p, int nResColumn){ + int n; + sqlite3 *db = p->db; + + if( p->nResColumn ){ + releaseMemArray(p->aColName, p->nResColumn*COLNAME_N); + sqlite3DbFree(db, p->aColName); + } + n = nResColumn*COLNAME_N; + p->nResColumn = (u16)nResColumn; + p->aColName = (Mem*)sqlite3DbMallocRawNN(db, sizeof(Mem)*n ); + if( p->aColName==0 ) return; + initMemArray(p->aColName, n, db, MEM_Null); +} + +/* +** Set the name of the idx'th column to be returned by the SQL statement. +** zName must be a pointer to a nul terminated string. +** +** This call must be made after a call to sqlite3VdbeSetNumCols(). +** +** The final parameter, xDel, must be one of SQLITE_DYNAMIC, SQLITE_STATIC +** or SQLITE_TRANSIENT. If it is SQLITE_DYNAMIC, then the buffer pointed +** to by zName will be freed by sqlite3DbFree() when the vdbe is destroyed. +*/ +SQLITE_PRIVATE int sqlite3VdbeSetColName( + Vdbe *p, /* Vdbe being configured */ + int idx, /* Index of column zName applies to */ + int var, /* One of the COLNAME_* constants */ + const char *zName, /* Pointer to buffer containing name */ + void (*xDel)(void*) /* Memory management strategy for zName */ +){ + int rc; + Mem *pColName; + assert( idxnResColumn ); + assert( vardb->mallocFailed ){ + assert( !zName || xDel!=SQLITE_DYNAMIC ); + return SQLITE_NOMEM_BKPT; + } + assert( p->aColName!=0 ); + pColName = &(p->aColName[idx+var*p->nResColumn]); + rc = sqlite3VdbeMemSetStr(pColName, zName, -1, SQLITE_UTF8, xDel); + assert( rc!=0 || !zName || (pColName->flags&MEM_Term)!=0 ); + return rc; +} + +/* +** A read or write transaction may or may not be active on database handle +** db. If a transaction is active, commit it. If there is a +** write-transaction spanning more than one database file, this routine +** takes care of the master journal trickery. +*/ +static int vdbeCommit(sqlite3 *db, Vdbe *p){ + int i; + int nTrans = 0; /* Number of databases with an active write-transaction + ** that are candidates for a two-phase commit using a + ** master-journal */ + int rc = SQLITE_OK; + int needXcommit = 0; + +#ifdef SQLITE_OMIT_VIRTUALTABLE + /* With this option, sqlite3VtabSync() is defined to be simply + ** SQLITE_OK so p is not used. + */ + UNUSED_PARAMETER(p); +#endif + + /* Before doing anything else, call the xSync() callback for any + ** virtual module tables written in this transaction. This has to + ** be done before determining whether a master journal file is + ** required, as an xSync() callback may add an attached database + ** to the transaction. + */ + rc = sqlite3VtabSync(db, p); + + /* This loop determines (a) if the commit hook should be invoked and + ** (b) how many database files have open write transactions, not + ** including the temp database. (b) is important because if more than + ** one database file has an open write transaction, a master journal + ** file is required for an atomic commit. + */ + for(i=0; rc==SQLITE_OK && inDb; i++){ + Btree *pBt = db->aDb[i].pBt; + if( sqlite3BtreeIsInTrans(pBt) ){ + /* Whether or not a database might need a master journal depends upon + ** its journal mode (among other things). This matrix determines which + ** journal modes use a master journal and which do not */ + static const u8 aMJNeeded[] = { + /* DELETE */ 1, + /* PERSIST */ 1, + /* OFF */ 0, + /* TRUNCATE */ 1, + /* MEMORY */ 0, + /* WAL */ 0 + }; + Pager *pPager; /* Pager associated with pBt */ + needXcommit = 1; + sqlite3BtreeEnter(pBt); + pPager = sqlite3BtreePager(pBt); + if( db->aDb[i].safety_level!=PAGER_SYNCHRONOUS_OFF + && aMJNeeded[sqlite3PagerGetJournalMode(pPager)] + && sqlite3PagerIsMemdb(pPager)==0 + ){ + assert( i!=1 ); + nTrans++; + } + rc = sqlite3PagerExclusiveLock(pPager); + sqlite3BtreeLeave(pBt); + } + } + if( rc!=SQLITE_OK ){ + return rc; + } + + /* If there are any write-transactions at all, invoke the commit hook */ + if( needXcommit && db->xCommitCallback ){ + rc = db->xCommitCallback(db->pCommitArg); + if( rc ){ + return SQLITE_CONSTRAINT_COMMITHOOK; + } + } + + /* The simple case - no more than one database file (not counting the + ** TEMP database) has a transaction active. There is no need for the + ** master-journal. + ** + ** If the return value of sqlite3BtreeGetFilename() is a zero length + ** string, it means the main database is :memory: or a temp file. In + ** that case we do not support atomic multi-file commits, so use the + ** simple case then too. + */ + if( 0==sqlite3Strlen30(sqlite3BtreeGetFilename(db->aDb[0].pBt)) + || nTrans<=1 + ){ + for(i=0; rc==SQLITE_OK && inDb; i++){ + Btree *pBt = db->aDb[i].pBt; + if( pBt ){ + rc = sqlite3BtreeCommitPhaseOne(pBt, 0); + } + } + + /* Do the commit only if all databases successfully complete phase 1. + ** If one of the BtreeCommitPhaseOne() calls fails, this indicates an + ** IO error while deleting or truncating a journal file. It is unlikely, + ** but could happen. In this case abandon processing and return the error. + */ + for(i=0; rc==SQLITE_OK && inDb; i++){ + Btree *pBt = db->aDb[i].pBt; + if( pBt ){ + rc = sqlite3BtreeCommitPhaseTwo(pBt, 0); + } + } + if( rc==SQLITE_OK ){ + sqlite3VtabCommit(db); + } + } + + /* The complex case - There is a multi-file write-transaction active. + ** This requires a master journal file to ensure the transaction is + ** committed atomically. + */ +#ifndef SQLITE_OMIT_DISKIO + else{ + sqlite3_vfs *pVfs = db->pVfs; + char *zMaster = 0; /* File-name for the master journal */ + char const *zMainFile = sqlite3BtreeGetFilename(db->aDb[0].pBt); + sqlite3_file *pMaster = 0; + i64 offset = 0; + int res; + int retryCount = 0; + int nMainFile; + + /* Select a master journal file name */ + nMainFile = sqlite3Strlen30(zMainFile); + zMaster = sqlite3MPrintf(db, "%s-mjXXXXXX9XXz", zMainFile); + if( zMaster==0 ) return SQLITE_NOMEM_BKPT; + do { + u32 iRandom; + if( retryCount ){ + if( retryCount>100 ){ + sqlite3_log(SQLITE_FULL, "MJ delete: %s", zMaster); + sqlite3OsDelete(pVfs, zMaster, 0); + break; + }else if( retryCount==1 ){ + sqlite3_log(SQLITE_FULL, "MJ collide: %s", zMaster); + } + } + retryCount++; + sqlite3_randomness(sizeof(iRandom), &iRandom); + sqlite3_snprintf(13, &zMaster[nMainFile], "-mj%06X9%02X", + (iRandom>>8)&0xffffff, iRandom&0xff); + /* The antipenultimate character of the master journal name must + ** be "9" to avoid name collisions when using 8+3 filenames. */ + assert( zMaster[sqlite3Strlen30(zMaster)-3]=='9' ); + sqlite3FileSuffix3(zMainFile, zMaster); + rc = sqlite3OsAccess(pVfs, zMaster, SQLITE_ACCESS_EXISTS, &res); + }while( rc==SQLITE_OK && res ); + if( rc==SQLITE_OK ){ + /* Open the master journal. */ + rc = sqlite3OsOpenMalloc(pVfs, zMaster, &pMaster, + SQLITE_OPEN_READWRITE|SQLITE_OPEN_CREATE| + SQLITE_OPEN_EXCLUSIVE|SQLITE_OPEN_MASTER_JOURNAL, 0 + ); + } + if( rc!=SQLITE_OK ){ + sqlite3DbFree(db, zMaster); + return rc; + } + + /* Write the name of each database file in the transaction into the new + ** master journal file. If an error occurs at this point close + ** and delete the master journal file. All the individual journal files + ** still have 'null' as the master journal pointer, so they will roll + ** back independently if a failure occurs. + */ + for(i=0; inDb; i++){ + Btree *pBt = db->aDb[i].pBt; + if( sqlite3BtreeIsInTrans(pBt) ){ + char const *zFile = sqlite3BtreeGetJournalname(pBt); + if( zFile==0 ){ + continue; /* Ignore TEMP and :memory: databases */ + } + assert( zFile[0]!=0 ); + rc = sqlite3OsWrite(pMaster, zFile, sqlite3Strlen30(zFile)+1, offset); + offset += sqlite3Strlen30(zFile)+1; + if( rc!=SQLITE_OK ){ + sqlite3OsCloseFree(pMaster); + sqlite3OsDelete(pVfs, zMaster, 0); + sqlite3DbFree(db, zMaster); + return rc; + } + } + } + + /* Sync the master journal file. If the IOCAP_SEQUENTIAL device + ** flag is set this is not required. + */ + if( 0==(sqlite3OsDeviceCharacteristics(pMaster)&SQLITE_IOCAP_SEQUENTIAL) + && SQLITE_OK!=(rc = sqlite3OsSync(pMaster, SQLITE_SYNC_NORMAL)) + ){ + sqlite3OsCloseFree(pMaster); + sqlite3OsDelete(pVfs, zMaster, 0); + sqlite3DbFree(db, zMaster); + return rc; + } + + /* Sync all the db files involved in the transaction. The same call + ** sets the master journal pointer in each individual journal. If + ** an error occurs here, do not delete the master journal file. + ** + ** If the error occurs during the first call to + ** sqlite3BtreeCommitPhaseOne(), then there is a chance that the + ** master journal file will be orphaned. But we cannot delete it, + ** in case the master journal file name was written into the journal + ** file before the failure occurred. + */ + for(i=0; rc==SQLITE_OK && inDb; i++){ + Btree *pBt = db->aDb[i].pBt; + if( pBt ){ + rc = sqlite3BtreeCommitPhaseOne(pBt, zMaster); + } + } + sqlite3OsCloseFree(pMaster); + assert( rc!=SQLITE_BUSY ); + if( rc!=SQLITE_OK ){ + sqlite3DbFree(db, zMaster); + return rc; + } + + /* Delete the master journal file. This commits the transaction. After + ** doing this the directory is synced again before any individual + ** transaction files are deleted. + */ + rc = sqlite3OsDelete(pVfs, zMaster, 1); + sqlite3DbFree(db, zMaster); + zMaster = 0; + if( rc ){ + return rc; + } + + /* All files and directories have already been synced, so the following + ** calls to sqlite3BtreeCommitPhaseTwo() are only closing files and + ** deleting or truncating journals. If something goes wrong while + ** this is happening we don't really care. The integrity of the + ** transaction is already guaranteed, but some stray 'cold' journals + ** may be lying around. Returning an error code won't help matters. + */ + disable_simulated_io_errors(); + sqlite3BeginBenignMalloc(); + for(i=0; inDb; i++){ + Btree *pBt = db->aDb[i].pBt; + if( pBt ){ + sqlite3BtreeCommitPhaseTwo(pBt, 1); + } + } + sqlite3EndBenignMalloc(); + enable_simulated_io_errors(); + + sqlite3VtabCommit(db); + } +#endif + + return rc; +} + +/* +** This routine checks that the sqlite3.nVdbeActive count variable +** matches the number of vdbe's in the list sqlite3.pVdbe that are +** currently active. An assertion fails if the two counts do not match. +** This is an internal self-check only - it is not an essential processing +** step. +** +** This is a no-op if NDEBUG is defined. +*/ +#ifndef NDEBUG +static void checkActiveVdbeCnt(sqlite3 *db){ + Vdbe *p; + int cnt = 0; + int nWrite = 0; + int nRead = 0; + p = db->pVdbe; + while( p ){ + if( sqlite3_stmt_busy((sqlite3_stmt*)p) ){ + cnt++; + if( p->readOnly==0 ) nWrite++; + if( p->bIsReader ) nRead++; + } + p = p->pNext; + } + assert( cnt==db->nVdbeActive ); + assert( nWrite==db->nVdbeWrite ); + assert( nRead==db->nVdbeRead ); +} +#else +#define checkActiveVdbeCnt(x) +#endif + +/* +** If the Vdbe passed as the first argument opened a statement-transaction, +** close it now. Argument eOp must be either SAVEPOINT_ROLLBACK or +** SAVEPOINT_RELEASE. If it is SAVEPOINT_ROLLBACK, then the statement +** transaction is rolled back. If eOp is SAVEPOINT_RELEASE, then the +** statement transaction is committed. +** +** If an IO error occurs, an SQLITE_IOERR_XXX error code is returned. +** Otherwise SQLITE_OK. +*/ +static SQLITE_NOINLINE int vdbeCloseStatement(Vdbe *p, int eOp){ + sqlite3 *const db = p->db; + int rc = SQLITE_OK; + int i; + const int iSavepoint = p->iStatement-1; + + assert( eOp==SAVEPOINT_ROLLBACK || eOp==SAVEPOINT_RELEASE); + assert( db->nStatement>0 ); + assert( p->iStatement==(db->nStatement+db->nSavepoint) ); + + for(i=0; inDb; i++){ + int rc2 = SQLITE_OK; + Btree *pBt = db->aDb[i].pBt; + if( pBt ){ + if( eOp==SAVEPOINT_ROLLBACK ){ + rc2 = sqlite3BtreeSavepoint(pBt, SAVEPOINT_ROLLBACK, iSavepoint); + } + if( rc2==SQLITE_OK ){ + rc2 = sqlite3BtreeSavepoint(pBt, SAVEPOINT_RELEASE, iSavepoint); + } + if( rc==SQLITE_OK ){ + rc = rc2; + } + } + } + db->nStatement--; + p->iStatement = 0; + + if( rc==SQLITE_OK ){ + if( eOp==SAVEPOINT_ROLLBACK ){ + rc = sqlite3VtabSavepoint(db, SAVEPOINT_ROLLBACK, iSavepoint); + } + if( rc==SQLITE_OK ){ + rc = sqlite3VtabSavepoint(db, SAVEPOINT_RELEASE, iSavepoint); + } + } + + /* If the statement transaction is being rolled back, also restore the + ** database handles deferred constraint counter to the value it had when + ** the statement transaction was opened. */ + if( eOp==SAVEPOINT_ROLLBACK ){ + db->nDeferredCons = p->nStmtDefCons; + db->nDeferredImmCons = p->nStmtDefImmCons; + } + return rc; +} +SQLITE_PRIVATE int sqlite3VdbeCloseStatement(Vdbe *p, int eOp){ + if( p->db->nStatement && p->iStatement ){ + return vdbeCloseStatement(p, eOp); + } + return SQLITE_OK; +} + + +/* +** This function is called when a transaction opened by the database +** handle associated with the VM passed as an argument is about to be +** committed. If there are outstanding deferred foreign key constraint +** violations, return SQLITE_ERROR. Otherwise, SQLITE_OK. +** +** If there are outstanding FK violations and this function returns +** SQLITE_ERROR, set the result of the VM to SQLITE_CONSTRAINT_FOREIGNKEY +** and write an error message to it. Then return SQLITE_ERROR. +*/ +#ifndef SQLITE_OMIT_FOREIGN_KEY +SQLITE_PRIVATE int sqlite3VdbeCheckFk(Vdbe *p, int deferred){ + sqlite3 *db = p->db; + if( (deferred && (db->nDeferredCons+db->nDeferredImmCons)>0) + || (!deferred && p->nFkConstraint>0) + ){ + p->rc = SQLITE_CONSTRAINT_FOREIGNKEY; + p->errorAction = OE_Abort; + sqlite3VdbeError(p, "FOREIGN KEY constraint failed"); + return SQLITE_ERROR; + } + return SQLITE_OK; +} +#endif + +/* +** This routine is called the when a VDBE tries to halt. If the VDBE +** has made changes and is in autocommit mode, then commit those +** changes. If a rollback is needed, then do the rollback. +** +** This routine is the only way to move the state of a VM from +** SQLITE_MAGIC_RUN to SQLITE_MAGIC_HALT. It is harmless to +** call this on a VM that is in the SQLITE_MAGIC_HALT state. +** +** Return an error code. If the commit could not complete because of +** lock contention, return SQLITE_BUSY. If SQLITE_BUSY is returned, it +** means the close did not happen and needs to be repeated. +*/ +SQLITE_PRIVATE int sqlite3VdbeHalt(Vdbe *p){ + int rc; /* Used to store transient return codes */ + sqlite3 *db = p->db; + + /* This function contains the logic that determines if a statement or + ** transaction will be committed or rolled back as a result of the + ** execution of this virtual machine. + ** + ** If any of the following errors occur: + ** + ** SQLITE_NOMEM + ** SQLITE_IOERR + ** SQLITE_FULL + ** SQLITE_INTERRUPT + ** + ** Then the internal cache might have been left in an inconsistent + ** state. We need to rollback the statement transaction, if there is + ** one, or the complete transaction if there is no statement transaction. + */ + + if( p->magic!=VDBE_MAGIC_RUN ){ + return SQLITE_OK; + } + if( db->mallocFailed ){ + p->rc = SQLITE_NOMEM_BKPT; + } + closeAllCursors(p); + checkActiveVdbeCnt(db); + + /* No commit or rollback needed if the program never started or if the + ** SQL statement does not read or write a database file. */ + if( p->pc>=0 && p->bIsReader ){ + int mrc; /* Primary error code from p->rc */ + int eStatementOp = 0; + int isSpecialError; /* Set to true if a 'special' error */ + + /* Lock all btrees used by the statement */ + sqlite3VdbeEnter(p); + + /* Check for one of the special errors */ + mrc = p->rc & 0xff; + isSpecialError = mrc==SQLITE_NOMEM || mrc==SQLITE_IOERR + || mrc==SQLITE_INTERRUPT || mrc==SQLITE_FULL; + if( isSpecialError ){ + /* If the query was read-only and the error code is SQLITE_INTERRUPT, + ** no rollback is necessary. Otherwise, at least a savepoint + ** transaction must be rolled back to restore the database to a + ** consistent state. + ** + ** Even if the statement is read-only, it is important to perform + ** a statement or transaction rollback operation. If the error + ** occurred while writing to the journal, sub-journal or database + ** file as part of an effort to free up cache space (see function + ** pagerStress() in pager.c), the rollback is required to restore + ** the pager to a consistent state. + */ + if( !p->readOnly || mrc!=SQLITE_INTERRUPT ){ + if( (mrc==SQLITE_NOMEM || mrc==SQLITE_FULL) && p->usesStmtJournal ){ + eStatementOp = SAVEPOINT_ROLLBACK; + }else{ + /* We are forced to roll back the active transaction. Before doing + ** so, abort any other statements this handle currently has active. + */ + sqlite3RollbackAll(db, SQLITE_ABORT_ROLLBACK); + sqlite3CloseSavepoints(db); + db->autoCommit = 1; + p->nChange = 0; + } + } + } + + /* Check for immediate foreign key violations. */ + if( p->rc==SQLITE_OK || (p->errorAction==OE_Fail && !isSpecialError) ){ + sqlite3VdbeCheckFk(p, 0); + } + + /* If the auto-commit flag is set and this is the only active writer + ** VM, then we do either a commit or rollback of the current transaction. + ** + ** Note: This block also runs if one of the special errors handled + ** above has occurred. + */ + if( !sqlite3VtabInSync(db) + && db->autoCommit + && db->nVdbeWrite==(p->readOnly==0) + ){ + if( p->rc==SQLITE_OK || (p->errorAction==OE_Fail && !isSpecialError) ){ + rc = sqlite3VdbeCheckFk(p, 1); + if( rc!=SQLITE_OK ){ + if( NEVER(p->readOnly) ){ + sqlite3VdbeLeave(p); + return SQLITE_ERROR; + } + rc = SQLITE_CONSTRAINT_FOREIGNKEY; + }else{ + /* The auto-commit flag is true, the vdbe program was successful + ** or hit an 'OR FAIL' constraint and there are no deferred foreign + ** key constraints to hold up the transaction. This means a commit + ** is required. */ + rc = vdbeCommit(db, p); + } + if( rc==SQLITE_BUSY && p->readOnly ){ + sqlite3VdbeLeave(p); + return SQLITE_BUSY; + }else if( rc!=SQLITE_OK ){ + p->rc = rc; + sqlite3RollbackAll(db, SQLITE_OK); + p->nChange = 0; + }else{ + db->nDeferredCons = 0; + db->nDeferredImmCons = 0; + db->flags &= ~(u64)SQLITE_DeferFKs; + sqlite3CommitInternalChanges(db); + } + }else{ + sqlite3RollbackAll(db, SQLITE_OK); + p->nChange = 0; + } + db->nStatement = 0; + }else if( eStatementOp==0 ){ + if( p->rc==SQLITE_OK || p->errorAction==OE_Fail ){ + eStatementOp = SAVEPOINT_RELEASE; + }else if( p->errorAction==OE_Abort ){ + eStatementOp = SAVEPOINT_ROLLBACK; + }else{ + sqlite3RollbackAll(db, SQLITE_ABORT_ROLLBACK); + sqlite3CloseSavepoints(db); + db->autoCommit = 1; + p->nChange = 0; + } + } + + /* If eStatementOp is non-zero, then a statement transaction needs to + ** be committed or rolled back. Call sqlite3VdbeCloseStatement() to + ** do so. If this operation returns an error, and the current statement + ** error code is SQLITE_OK or SQLITE_CONSTRAINT, then promote the + ** current statement error code. + */ + if( eStatementOp ){ + rc = sqlite3VdbeCloseStatement(p, eStatementOp); + if( rc ){ + if( p->rc==SQLITE_OK || (p->rc&0xff)==SQLITE_CONSTRAINT ){ + p->rc = rc; + sqlite3DbFree(db, p->zErrMsg); + p->zErrMsg = 0; + } + sqlite3RollbackAll(db, SQLITE_ABORT_ROLLBACK); + sqlite3CloseSavepoints(db); + db->autoCommit = 1; + p->nChange = 0; + } + } + + /* If this was an INSERT, UPDATE or DELETE and no statement transaction + ** has been rolled back, update the database connection change-counter. + */ + if( p->changeCntOn ){ + if( eStatementOp!=SAVEPOINT_ROLLBACK ){ + sqlite3VdbeSetChanges(db, p->nChange); + }else{ + sqlite3VdbeSetChanges(db, 0); + } + p->nChange = 0; + } + + /* Release the locks */ + sqlite3VdbeLeave(p); + } + + /* We have successfully halted and closed the VM. Record this fact. */ + if( p->pc>=0 ){ + db->nVdbeActive--; + if( !p->readOnly ) db->nVdbeWrite--; + if( p->bIsReader ) db->nVdbeRead--; + assert( db->nVdbeActive>=db->nVdbeRead ); + assert( db->nVdbeRead>=db->nVdbeWrite ); + assert( db->nVdbeWrite>=0 ); + } + p->magic = VDBE_MAGIC_HALT; + checkActiveVdbeCnt(db); + if( db->mallocFailed ){ + p->rc = SQLITE_NOMEM_BKPT; + } + + /* If the auto-commit flag is set to true, then any locks that were held + ** by connection db have now been released. Call sqlite3ConnectionUnlocked() + ** to invoke any required unlock-notify callbacks. + */ + if( db->autoCommit ){ + sqlite3ConnectionUnlocked(db); + } + + assert( db->nVdbeActive>0 || db->autoCommit==0 || db->nStatement==0 ); + return (p->rc==SQLITE_BUSY ? SQLITE_BUSY : SQLITE_OK); +} + + +/* +** Each VDBE holds the result of the most recent sqlite3_step() call +** in p->rc. This routine sets that result back to SQLITE_OK. +*/ +SQLITE_PRIVATE void sqlite3VdbeResetStepResult(Vdbe *p){ + p->rc = SQLITE_OK; +} + +/* +** Copy the error code and error message belonging to the VDBE passed +** as the first argument to its database handle (so that they will be +** returned by calls to sqlite3_errcode() and sqlite3_errmsg()). +** +** This function does not clear the VDBE error code or message, just +** copies them to the database handle. +*/ +SQLITE_PRIVATE int sqlite3VdbeTransferError(Vdbe *p){ + sqlite3 *db = p->db; + int rc = p->rc; + if( p->zErrMsg ){ + db->bBenignMalloc++; + sqlite3BeginBenignMalloc(); + if( db->pErr==0 ) db->pErr = sqlite3ValueNew(db); + sqlite3ValueSetStr(db->pErr, -1, p->zErrMsg, SQLITE_UTF8, SQLITE_TRANSIENT); + sqlite3EndBenignMalloc(); + db->bBenignMalloc--; + }else if( db->pErr ){ + sqlite3ValueSetNull(db->pErr); + } + db->errCode = rc; + return rc; +} + +#ifdef SQLITE_ENABLE_SQLLOG +/* +** If an SQLITE_CONFIG_SQLLOG hook is registered and the VM has been run, +** invoke it. +*/ +static void vdbeInvokeSqllog(Vdbe *v){ + if( sqlite3GlobalConfig.xSqllog && v->rc==SQLITE_OK && v->zSql && v->pc>=0 ){ + char *zExpanded = sqlite3VdbeExpandSql(v, v->zSql); + assert( v->db->init.busy==0 ); + if( zExpanded ){ + sqlite3GlobalConfig.xSqllog( + sqlite3GlobalConfig.pSqllogArg, v->db, zExpanded, 1 + ); + sqlite3DbFree(v->db, zExpanded); + } + } +} +#else +# define vdbeInvokeSqllog(x) +#endif + +/* +** Clean up a VDBE after execution but do not delete the VDBE just yet. +** Write any error messages into *pzErrMsg. Return the result code. +** +** After this routine is run, the VDBE should be ready to be executed +** again. +** +** To look at it another way, this routine resets the state of the +** virtual machine from VDBE_MAGIC_RUN or VDBE_MAGIC_HALT back to +** VDBE_MAGIC_INIT. +*/ +SQLITE_PRIVATE int sqlite3VdbeReset(Vdbe *p){ +#if defined(SQLITE_DEBUG) || defined(VDBE_PROFILE) + int i; +#endif + + sqlite3 *db; + db = p->db; + + /* If the VM did not run to completion or if it encountered an + ** error, then it might not have been halted properly. So halt + ** it now. + */ + sqlite3VdbeHalt(p); + + /* If the VDBE has been run even partially, then transfer the error code + ** and error message from the VDBE into the main database structure. But + ** if the VDBE has just been set to run but has not actually executed any + ** instructions yet, leave the main database error information unchanged. + */ + if( p->pc>=0 ){ + vdbeInvokeSqllog(p); + sqlite3VdbeTransferError(p); + if( p->runOnlyOnce ) p->expired = 1; + }else if( p->rc && p->expired ){ + /* The expired flag was set on the VDBE before the first call + ** to sqlite3_step(). For consistency (since sqlite3_step() was + ** called), set the database error in this case as well. + */ + sqlite3ErrorWithMsg(db, p->rc, p->zErrMsg ? "%s" : 0, p->zErrMsg); + } + + /* Reset register contents and reclaim error message memory. + */ +#ifdef SQLITE_DEBUG + /* Execute assert() statements to ensure that the Vdbe.apCsr[] and + ** Vdbe.aMem[] arrays have already been cleaned up. */ + if( p->apCsr ) for(i=0; inCursor; i++) assert( p->apCsr[i]==0 ); + if( p->aMem ){ + for(i=0; inMem; i++) assert( p->aMem[i].flags==MEM_Undefined ); + } +#endif + sqlite3DbFree(db, p->zErrMsg); + p->zErrMsg = 0; + p->pResultSet = 0; +#ifdef SQLITE_DEBUG + p->nWrite = 0; +#endif + + /* Save profiling information from this VDBE run. + */ +#ifdef VDBE_PROFILE + { + FILE *out = fopen("vdbe_profile.out", "a"); + if( out ){ + fprintf(out, "---- "); + for(i=0; inOp; i++){ + fprintf(out, "%02x", p->aOp[i].opcode); + } + fprintf(out, "\n"); + if( p->zSql ){ + char c, pc = 0; + fprintf(out, "-- "); + for(i=0; (c = p->zSql[i])!=0; i++){ + if( pc=='\n' ) fprintf(out, "-- "); + putc(c, out); + pc = c; + } + if( pc!='\n' ) fprintf(out, "\n"); + } + for(i=0; inOp; i++){ + char zHdr[100]; + sqlite3_snprintf(sizeof(zHdr), zHdr, "%6u %12llu %8llu ", + p->aOp[i].cnt, + p->aOp[i].cycles, + p->aOp[i].cnt>0 ? p->aOp[i].cycles/p->aOp[i].cnt : 0 + ); + fprintf(out, "%s", zHdr); + sqlite3VdbePrintOp(out, i, &p->aOp[i]); + } + fclose(out); + } + } +#endif + p->magic = VDBE_MAGIC_RESET; + return p->rc & db->errMask; +} + +/* +** Clean up and delete a VDBE after execution. Return an integer which is +** the result code. Write any error message text into *pzErrMsg. +*/ +SQLITE_PRIVATE int sqlite3VdbeFinalize(Vdbe *p){ + int rc = SQLITE_OK; + if( p->magic==VDBE_MAGIC_RUN || p->magic==VDBE_MAGIC_HALT ){ + rc = sqlite3VdbeReset(p); + assert( (rc & p->db->errMask)==rc ); + } + sqlite3VdbeDelete(p); + return rc; +} + +/* +** If parameter iOp is less than zero, then invoke the destructor for +** all auxiliary data pointers currently cached by the VM passed as +** the first argument. +** +** Or, if iOp is greater than or equal to zero, then the destructor is +** only invoked for those auxiliary data pointers created by the user +** function invoked by the OP_Function opcode at instruction iOp of +** VM pVdbe, and only then if: +** +** * the associated function parameter is the 32nd or later (counting +** from left to right), or +** +** * the corresponding bit in argument mask is clear (where the first +** function parameter corresponds to bit 0 etc.). +*/ +SQLITE_PRIVATE void sqlite3VdbeDeleteAuxData(sqlite3 *db, AuxData **pp, int iOp, int mask){ + while( *pp ){ + AuxData *pAux = *pp; + if( (iOp<0) + || (pAux->iAuxOp==iOp + && pAux->iAuxArg>=0 + && (pAux->iAuxArg>31 || !(mask & MASKBIT32(pAux->iAuxArg)))) + ){ + testcase( pAux->iAuxArg==31 ); + if( pAux->xDeleteAux ){ + pAux->xDeleteAux(pAux->pAux); + } + *pp = pAux->pNextAux; + sqlite3DbFree(db, pAux); + }else{ + pp= &pAux->pNextAux; + } + } +} + +/* +** Free all memory associated with the Vdbe passed as the second argument, +** except for object itself, which is preserved. +** +** The difference between this function and sqlite3VdbeDelete() is that +** VdbeDelete() also unlinks the Vdbe from the list of VMs associated with +** the database connection and frees the object itself. +*/ +SQLITE_PRIVATE void sqlite3VdbeClearObject(sqlite3 *db, Vdbe *p){ + SubProgram *pSub, *pNext; + assert( p->db==0 || p->db==db ); + releaseMemArray(p->aColName, p->nResColumn*COLNAME_N); + for(pSub=p->pProgram; pSub; pSub=pNext){ + pNext = pSub->pNext; + vdbeFreeOpArray(db, pSub->aOp, pSub->nOp); + sqlite3DbFree(db, pSub); + } + if( p->magic!=VDBE_MAGIC_INIT ){ + releaseMemArray(p->aVar, p->nVar); + sqlite3DbFree(db, p->pVList); + sqlite3DbFree(db, p->pFree); + } + vdbeFreeOpArray(db, p->aOp, p->nOp); + sqlite3DbFree(db, p->aColName); + sqlite3DbFree(db, p->zSql); +#ifdef SQLITE_ENABLE_NORMALIZE + sqlite3DbFree(db, p->zNormSql); + { + DblquoteStr *pThis, *pNext; + for(pThis=p->pDblStr; pThis; pThis=pNext){ + pNext = pThis->pNextStr; + sqlite3DbFree(db, pThis); + } + } +#endif +#ifdef SQLITE_ENABLE_STMT_SCANSTATUS + { + int i; + for(i=0; inScan; i++){ + sqlite3DbFree(db, p->aScan[i].zName); + } + sqlite3DbFree(db, p->aScan); + } +#endif +} + +/* +** Delete an entire VDBE. +*/ +SQLITE_PRIVATE void sqlite3VdbeDelete(Vdbe *p){ + sqlite3 *db; + + assert( p!=0 ); + db = p->db; + assert( sqlite3_mutex_held(db->mutex) ); + sqlite3VdbeClearObject(db, p); + if( p->pPrev ){ + p->pPrev->pNext = p->pNext; + }else{ + assert( db->pVdbe==p ); + db->pVdbe = p->pNext; + } + if( p->pNext ){ + p->pNext->pPrev = p->pPrev; + } + p->magic = VDBE_MAGIC_DEAD; + p->db = 0; + sqlite3DbFreeNN(db, p); +} + +/* +** The cursor "p" has a pending seek operation that has not yet been +** carried out. Seek the cursor now. If an error occurs, return +** the appropriate error code. +*/ +static int SQLITE_NOINLINE handleDeferredMoveto(VdbeCursor *p){ + int res, rc; +#ifdef SQLITE_TEST + extern int sqlite3_search_count; +#endif + assert( p->deferredMoveto ); + assert( p->isTable ); + assert( p->eCurType==CURTYPE_BTREE ); + rc = sqlite3BtreeMovetoUnpacked(p->uc.pCursor, 0, p->movetoTarget, 0, &res); + if( rc ) return rc; + if( res!=0 ) return SQLITE_CORRUPT_BKPT; +#ifdef SQLITE_TEST + sqlite3_search_count++; +#endif + p->deferredMoveto = 0; + p->cacheStatus = CACHE_STALE; + return SQLITE_OK; +} + +/* +** Something has moved cursor "p" out of place. Maybe the row it was +** pointed to was deleted out from under it. Or maybe the btree was +** rebalanced. Whatever the cause, try to restore "p" to the place it +** is supposed to be pointing. If the row was deleted out from under the +** cursor, set the cursor to point to a NULL row. +*/ +static int SQLITE_NOINLINE handleMovedCursor(VdbeCursor *p){ + int isDifferentRow, rc; + assert( p->eCurType==CURTYPE_BTREE ); + assert( p->uc.pCursor!=0 ); + assert( sqlite3BtreeCursorHasMoved(p->uc.pCursor) ); + rc = sqlite3BtreeCursorRestore(p->uc.pCursor, &isDifferentRow); + p->cacheStatus = CACHE_STALE; + if( isDifferentRow ) p->nullRow = 1; + return rc; +} + +/* +** Check to ensure that the cursor is valid. Restore the cursor +** if need be. Return any I/O error from the restore operation. +*/ +SQLITE_PRIVATE int sqlite3VdbeCursorRestore(VdbeCursor *p){ + assert( p->eCurType==CURTYPE_BTREE ); + if( sqlite3BtreeCursorHasMoved(p->uc.pCursor) ){ + return handleMovedCursor(p); + } + return SQLITE_OK; +} + +/* +** Make sure the cursor p is ready to read or write the row to which it +** was last positioned. Return an error code if an OOM fault or I/O error +** prevents us from positioning the cursor to its correct position. +** +** If a MoveTo operation is pending on the given cursor, then do that +** MoveTo now. If no move is pending, check to see if the row has been +** deleted out from under the cursor and if it has, mark the row as +** a NULL row. +** +** If the cursor is already pointing to the correct row and that row has +** not been deleted out from under the cursor, then this routine is a no-op. +*/ +SQLITE_PRIVATE int sqlite3VdbeCursorMoveto(VdbeCursor **pp, int *piCol){ + VdbeCursor *p = *pp; + assert( p->eCurType==CURTYPE_BTREE || p->eCurType==CURTYPE_PSEUDO ); + if( p->deferredMoveto ){ + int iMap; + if( p->aAltMap && (iMap = p->aAltMap[1+*piCol])>0 ){ + *pp = p->pAltCursor; + *piCol = iMap - 1; + return SQLITE_OK; + } + return handleDeferredMoveto(p); + } + if( sqlite3BtreeCursorHasMoved(p->uc.pCursor) ){ + return handleMovedCursor(p); + } + return SQLITE_OK; +} + +/* +** The following functions: +** +** sqlite3VdbeSerialType() +** sqlite3VdbeSerialTypeLen() +** sqlite3VdbeSerialLen() +** sqlite3VdbeSerialPut() +** sqlite3VdbeSerialGet() +** +** encapsulate the code that serializes values for storage in SQLite +** data and index records. Each serialized value consists of a +** 'serial-type' and a blob of data. The serial type is an 8-byte unsigned +** integer, stored as a varint. +** +** In an SQLite index record, the serial type is stored directly before +** the blob of data that it corresponds to. In a table record, all serial +** types are stored at the start of the record, and the blobs of data at +** the end. Hence these functions allow the caller to handle the +** serial-type and data blob separately. +** +** The following table describes the various storage classes for data: +** +** serial type bytes of data type +** -------------- --------------- --------------- +** 0 0 NULL +** 1 1 signed integer +** 2 2 signed integer +** 3 3 signed integer +** 4 4 signed integer +** 5 6 signed integer +** 6 8 signed integer +** 7 8 IEEE float +** 8 0 Integer constant 0 +** 9 0 Integer constant 1 +** 10,11 reserved for expansion +** N>=12 and even (N-12)/2 BLOB +** N>=13 and odd (N-13)/2 text +** +** The 8 and 9 types were added in 3.3.0, file format 4. Prior versions +** of SQLite will not understand those serial types. +*/ + +#if 0 /* Inlined into the OP_MakeRecord opcode */ +/* +** Return the serial-type for the value stored in pMem. +** +** This routine might convert a large MEM_IntReal value into MEM_Real. +** +** 2019-07-11: The primary user of this subroutine was the OP_MakeRecord +** opcode in the byte-code engine. But by moving this routine in-line, we +** can omit some redundant tests and make that opcode a lot faster. So +** this routine is now only used by the STAT3 logic and STAT3 support has +** ended. The code is kept here for historical reference only. +*/ +SQLITE_PRIVATE u32 sqlite3VdbeSerialType(Mem *pMem, int file_format, u32 *pLen){ + int flags = pMem->flags; + u32 n; + + assert( pLen!=0 ); + if( flags&MEM_Null ){ + *pLen = 0; + return 0; + } + if( flags&(MEM_Int|MEM_IntReal) ){ + /* Figure out whether to use 1, 2, 4, 6 or 8 bytes. */ +# define MAX_6BYTE ((((i64)0x00008000)<<32)-1) + i64 i = pMem->u.i; + u64 u; + testcase( flags & MEM_Int ); + testcase( flags & MEM_IntReal ); + if( i<0 ){ + u = ~i; + }else{ + u = i; + } + if( u<=127 ){ + if( (i&1)==i && file_format>=4 ){ + *pLen = 0; + return 8+(u32)u; + }else{ + *pLen = 1; + return 1; + } + } + if( u<=32767 ){ *pLen = 2; return 2; } + if( u<=8388607 ){ *pLen = 3; return 3; } + if( u<=2147483647 ){ *pLen = 4; return 4; } + if( u<=MAX_6BYTE ){ *pLen = 6; return 5; } + *pLen = 8; + if( flags&MEM_IntReal ){ + /* If the value is IntReal and is going to take up 8 bytes to store + ** as an integer, then we might as well make it an 8-byte floating + ** point value */ + pMem->u.r = (double)pMem->u.i; + pMem->flags &= ~MEM_IntReal; + pMem->flags |= MEM_Real; + return 7; + } + return 6; + } + if( flags&MEM_Real ){ + *pLen = 8; + return 7; + } + assert( pMem->db->mallocFailed || flags&(MEM_Str|MEM_Blob) ); + assert( pMem->n>=0 ); + n = (u32)pMem->n; + if( flags & MEM_Zero ){ + n += pMem->u.nZero; + } + *pLen = n; + return ((n*2) + 12 + ((flags&MEM_Str)!=0)); +} +#endif /* inlined into OP_MakeRecord */ + +/* +** The sizes for serial types less than 128 +*/ +static const u8 sqlite3SmallTypeSizes[] = { + /* 0 1 2 3 4 5 6 7 8 9 */ +/* 0 */ 0, 1, 2, 3, 4, 6, 8, 8, 0, 0, +/* 10 */ 0, 0, 0, 0, 1, 1, 2, 2, 3, 3, +/* 20 */ 4, 4, 5, 5, 6, 6, 7, 7, 8, 8, +/* 30 */ 9, 9, 10, 10, 11, 11, 12, 12, 13, 13, +/* 40 */ 14, 14, 15, 15, 16, 16, 17, 17, 18, 18, +/* 50 */ 19, 19, 20, 20, 21, 21, 22, 22, 23, 23, +/* 60 */ 24, 24, 25, 25, 26, 26, 27, 27, 28, 28, +/* 70 */ 29, 29, 30, 30, 31, 31, 32, 32, 33, 33, +/* 80 */ 34, 34, 35, 35, 36, 36, 37, 37, 38, 38, +/* 90 */ 39, 39, 40, 40, 41, 41, 42, 42, 43, 43, +/* 100 */ 44, 44, 45, 45, 46, 46, 47, 47, 48, 48, +/* 110 */ 49, 49, 50, 50, 51, 51, 52, 52, 53, 53, +/* 120 */ 54, 54, 55, 55, 56, 56, 57, 57 +}; + +/* +** Return the length of the data corresponding to the supplied serial-type. +*/ +SQLITE_PRIVATE u32 sqlite3VdbeSerialTypeLen(u32 serial_type){ + if( serial_type>=128 ){ + return (serial_type-12)/2; + }else{ + assert( serial_type<12 + || sqlite3SmallTypeSizes[serial_type]==(serial_type - 12)/2 ); + return sqlite3SmallTypeSizes[serial_type]; + } +} +SQLITE_PRIVATE u8 sqlite3VdbeOneByteSerialTypeLen(u8 serial_type){ + assert( serial_type<128 ); + return sqlite3SmallTypeSizes[serial_type]; +} + +/* +** If we are on an architecture with mixed-endian floating +** points (ex: ARM7) then swap the lower 4 bytes with the +** upper 4 bytes. Return the result. +** +** For most architectures, this is a no-op. +** +** (later): It is reported to me that the mixed-endian problem +** on ARM7 is an issue with GCC, not with the ARM7 chip. It seems +** that early versions of GCC stored the two words of a 64-bit +** float in the wrong order. And that error has been propagated +** ever since. The blame is not necessarily with GCC, though. +** GCC might have just copying the problem from a prior compiler. +** I am also told that newer versions of GCC that follow a different +** ABI get the byte order right. +** +** Developers using SQLite on an ARM7 should compile and run their +** application using -DSQLITE_DEBUG=1 at least once. With DEBUG +** enabled, some asserts below will ensure that the byte order of +** floating point values is correct. +** +** (2007-08-30) Frank van Vugt has studied this problem closely +** and has send his findings to the SQLite developers. Frank +** writes that some Linux kernels offer floating point hardware +** emulation that uses only 32-bit mantissas instead of a full +** 48-bits as required by the IEEE standard. (This is the +** CONFIG_FPE_FASTFPE option.) On such systems, floating point +** byte swapping becomes very complicated. To avoid problems, +** the necessary byte swapping is carried out using a 64-bit integer +** rather than a 64-bit float. Frank assures us that the code here +** works for him. We, the developers, have no way to independently +** verify this, but Frank seems to know what he is talking about +** so we trust him. +*/ +#ifdef SQLITE_MIXED_ENDIAN_64BIT_FLOAT +static u64 floatSwap(u64 in){ + union { + u64 r; + u32 i[2]; + } u; + u32 t; + + u.r = in; + t = u.i[0]; + u.i[0] = u.i[1]; + u.i[1] = t; + return u.r; +} +# define swapMixedEndianFloat(X) X = floatSwap(X) +#else +# define swapMixedEndianFloat(X) +#endif + +/* +** Write the serialized data blob for the value stored in pMem into +** buf. It is assumed that the caller has allocated sufficient space. +** Return the number of bytes written. +** +** nBuf is the amount of space left in buf[]. The caller is responsible +** for allocating enough space to buf[] to hold the entire field, exclusive +** of the pMem->u.nZero bytes for a MEM_Zero value. +** +** Return the number of bytes actually written into buf[]. The number +** of bytes in the zero-filled tail is included in the return value only +** if those bytes were zeroed in buf[]. +*/ +SQLITE_PRIVATE u32 sqlite3VdbeSerialPut(u8 *buf, Mem *pMem, u32 serial_type){ + u32 len; + + /* Integer and Real */ + if( serial_type<=7 && serial_type>0 ){ + u64 v; + u32 i; + if( serial_type==7 ){ + assert( sizeof(v)==sizeof(pMem->u.r) ); + memcpy(&v, &pMem->u.r, sizeof(v)); + swapMixedEndianFloat(v); + }else{ + v = pMem->u.i; + } + len = i = sqlite3SmallTypeSizes[serial_type]; + assert( i>0 ); + do{ + buf[--i] = (u8)(v&0xFF); + v >>= 8; + }while( i ); + return len; + } + + /* String or blob */ + if( serial_type>=12 ){ + assert( pMem->n + ((pMem->flags & MEM_Zero)?pMem->u.nZero:0) + == (int)sqlite3VdbeSerialTypeLen(serial_type) ); + len = pMem->n; + if( len>0 ) memcpy(buf, pMem->z, len); + return len; + } + + /* NULL or constants 0 or 1 */ + return 0; +} + +/* Input "x" is a sequence of unsigned characters that represent a +** big-endian integer. Return the equivalent native integer +*/ +#define ONE_BYTE_INT(x) ((i8)(x)[0]) +#define TWO_BYTE_INT(x) (256*(i8)((x)[0])|(x)[1]) +#define THREE_BYTE_INT(x) (65536*(i8)((x)[0])|((x)[1]<<8)|(x)[2]) +#define FOUR_BYTE_UINT(x) (((u32)(x)[0]<<24)|((x)[1]<<16)|((x)[2]<<8)|(x)[3]) +#define FOUR_BYTE_INT(x) (16777216*(i8)((x)[0])|((x)[1]<<16)|((x)[2]<<8)|(x)[3]) + +/* +** Deserialize the data blob pointed to by buf as serial type serial_type +** and store the result in pMem. Return the number of bytes read. +** +** This function is implemented as two separate routines for performance. +** The few cases that require local variables are broken out into a separate +** routine so that in most cases the overhead of moving the stack pointer +** is avoided. +*/ +static u32 serialGet( + const unsigned char *buf, /* Buffer to deserialize from */ + u32 serial_type, /* Serial type to deserialize */ + Mem *pMem /* Memory cell to write value into */ +){ + u64 x = FOUR_BYTE_UINT(buf); + u32 y = FOUR_BYTE_UINT(buf+4); + x = (x<<32) + y; + if( serial_type==6 ){ + /* EVIDENCE-OF: R-29851-52272 Value is a big-endian 64-bit + ** twos-complement integer. */ + pMem->u.i = *(i64*)&x; + pMem->flags = MEM_Int; + testcase( pMem->u.i<0 ); + }else{ + /* EVIDENCE-OF: R-57343-49114 Value is a big-endian IEEE 754-2008 64-bit + ** floating point number. */ +#if !defined(NDEBUG) && !defined(SQLITE_OMIT_FLOATING_POINT) + /* Verify that integers and floating point values use the same + ** byte order. Or, that if SQLITE_MIXED_ENDIAN_64BIT_FLOAT is + ** defined that 64-bit floating point values really are mixed + ** endian. + */ + static const u64 t1 = ((u64)0x3ff00000)<<32; + static const double r1 = 1.0; + u64 t2 = t1; + swapMixedEndianFloat(t2); + assert( sizeof(r1)==sizeof(t2) && memcmp(&r1, &t2, sizeof(r1))==0 ); +#endif + assert( sizeof(x)==8 && sizeof(pMem->u.r)==8 ); + swapMixedEndianFloat(x); + memcpy(&pMem->u.r, &x, sizeof(x)); + pMem->flags = IsNaN(x) ? MEM_Null : MEM_Real; + } + return 8; +} +SQLITE_PRIVATE u32 sqlite3VdbeSerialGet( + const unsigned char *buf, /* Buffer to deserialize from */ + u32 serial_type, /* Serial type to deserialize */ + Mem *pMem /* Memory cell to write value into */ +){ + switch( serial_type ){ + case 10: { /* Internal use only: NULL with virtual table + ** UPDATE no-change flag set */ + pMem->flags = MEM_Null|MEM_Zero; + pMem->n = 0; + pMem->u.nZero = 0; + break; + } + case 11: /* Reserved for future use */ + case 0: { /* Null */ + /* EVIDENCE-OF: R-24078-09375 Value is a NULL. */ + pMem->flags = MEM_Null; + break; + } + case 1: { + /* EVIDENCE-OF: R-44885-25196 Value is an 8-bit twos-complement + ** integer. */ + pMem->u.i = ONE_BYTE_INT(buf); + pMem->flags = MEM_Int; + testcase( pMem->u.i<0 ); + return 1; + } + case 2: { /* 2-byte signed integer */ + /* EVIDENCE-OF: R-49794-35026 Value is a big-endian 16-bit + ** twos-complement integer. */ + pMem->u.i = TWO_BYTE_INT(buf); + pMem->flags = MEM_Int; + testcase( pMem->u.i<0 ); + return 2; + } + case 3: { /* 3-byte signed integer */ + /* EVIDENCE-OF: R-37839-54301 Value is a big-endian 24-bit + ** twos-complement integer. */ + pMem->u.i = THREE_BYTE_INT(buf); + pMem->flags = MEM_Int; + testcase( pMem->u.i<0 ); + return 3; + } + case 4: { /* 4-byte signed integer */ + /* EVIDENCE-OF: R-01849-26079 Value is a big-endian 32-bit + ** twos-complement integer. */ + pMem->u.i = FOUR_BYTE_INT(buf); +#ifdef __HP_cc + /* Work around a sign-extension bug in the HP compiler for HP/UX */ + if( buf[0]&0x80 ) pMem->u.i |= 0xffffffff80000000LL; +#endif + pMem->flags = MEM_Int; + testcase( pMem->u.i<0 ); + return 4; + } + case 5: { /* 6-byte signed integer */ + /* EVIDENCE-OF: R-50385-09674 Value is a big-endian 48-bit + ** twos-complement integer. */ + pMem->u.i = FOUR_BYTE_UINT(buf+2) + (((i64)1)<<32)*TWO_BYTE_INT(buf); + pMem->flags = MEM_Int; + testcase( pMem->u.i<0 ); + return 6; + } + case 6: /* 8-byte signed integer */ + case 7: { /* IEEE floating point */ + /* These use local variables, so do them in a separate routine + ** to avoid having to move the frame pointer in the common case */ + return serialGet(buf,serial_type,pMem); + } + case 8: /* Integer 0 */ + case 9: { /* Integer 1 */ + /* EVIDENCE-OF: R-12976-22893 Value is the integer 0. */ + /* EVIDENCE-OF: R-18143-12121 Value is the integer 1. */ + pMem->u.i = serial_type-8; + pMem->flags = MEM_Int; + return 0; + } + default: { + /* EVIDENCE-OF: R-14606-31564 Value is a BLOB that is (N-12)/2 bytes in + ** length. + ** EVIDENCE-OF: R-28401-00140 Value is a string in the text encoding and + ** (N-13)/2 bytes in length. */ + static const u16 aFlag[] = { MEM_Blob|MEM_Ephem, MEM_Str|MEM_Ephem }; + pMem->z = (char *)buf; + pMem->n = (serial_type-12)/2; + pMem->flags = aFlag[serial_type&1]; + return pMem->n; + } + } + return 0; +} +/* +** This routine is used to allocate sufficient space for an UnpackedRecord +** structure large enough to be used with sqlite3VdbeRecordUnpack() if +** the first argument is a pointer to KeyInfo structure pKeyInfo. +** +** The space is either allocated using sqlite3DbMallocRaw() or from within +** the unaligned buffer passed via the second and third arguments (presumably +** stack space). If the former, then *ppFree is set to a pointer that should +** be eventually freed by the caller using sqlite3DbFree(). Or, if the +** allocation comes from the pSpace/szSpace buffer, *ppFree is set to NULL +** before returning. +** +** If an OOM error occurs, NULL is returned. +*/ +SQLITE_PRIVATE UnpackedRecord *sqlite3VdbeAllocUnpackedRecord( + KeyInfo *pKeyInfo /* Description of the record */ +){ + UnpackedRecord *p; /* Unpacked record to return */ + int nByte; /* Number of bytes required for *p */ + nByte = ROUND8(sizeof(UnpackedRecord)) + sizeof(Mem)*(pKeyInfo->nKeyField+1); + p = (UnpackedRecord *)sqlite3DbMallocRaw(pKeyInfo->db, nByte); + if( !p ) return 0; + p->aMem = (Mem*)&((char*)p)[ROUND8(sizeof(UnpackedRecord))]; + assert( pKeyInfo->aSortFlags!=0 ); + p->pKeyInfo = pKeyInfo; + p->nField = pKeyInfo->nKeyField + 1; + return p; +} + +/* +** Given the nKey-byte encoding of a record in pKey[], populate the +** UnpackedRecord structure indicated by the fourth argument with the +** contents of the decoded record. +*/ +SQLITE_PRIVATE void sqlite3VdbeRecordUnpack( + KeyInfo *pKeyInfo, /* Information about the record format */ + int nKey, /* Size of the binary record */ + const void *pKey, /* The binary record */ + UnpackedRecord *p /* Populate this structure before returning. */ +){ + const unsigned char *aKey = (const unsigned char *)pKey; + u32 d; + u32 idx; /* Offset in aKey[] to read from */ + u16 u; /* Unsigned loop counter */ + u32 szHdr; + Mem *pMem = p->aMem; + + p->default_rc = 0; + assert( EIGHT_BYTE_ALIGNMENT(pMem) ); + idx = getVarint32(aKey, szHdr); + d = szHdr; + u = 0; + while( idxenc = pKeyInfo->enc; + pMem->db = pKeyInfo->db; + /* pMem->flags = 0; // sqlite3VdbeSerialGet() will set this for us */ + pMem->szMalloc = 0; + pMem->z = 0; + d += sqlite3VdbeSerialGet(&aKey[d], serial_type, pMem); + pMem++; + if( (++u)>=p->nField ) break; + } + if( d>(u32)nKey && u ){ + assert( CORRUPT_DB ); + /* In a corrupt record entry, the last pMem might have been set up using + ** uninitialized memory. Overwrite its value with NULL, to prevent + ** warnings from MSAN. */ + sqlite3VdbeMemSetNull(pMem-1); + } + assert( u<=pKeyInfo->nKeyField + 1 ); + p->nField = u; +} + +#ifdef SQLITE_DEBUG +/* +** This function compares two index or table record keys in the same way +** as the sqlite3VdbeRecordCompare() routine. Unlike VdbeRecordCompare(), +** this function deserializes and compares values using the +** sqlite3VdbeSerialGet() and sqlite3MemCompare() functions. It is used +** in assert() statements to ensure that the optimized code in +** sqlite3VdbeRecordCompare() returns results with these two primitives. +** +** Return true if the result of comparison is equivalent to desiredResult. +** Return false if there is a disagreement. +*/ +static int vdbeRecordCompareDebug( + int nKey1, const void *pKey1, /* Left key */ + const UnpackedRecord *pPKey2, /* Right key */ + int desiredResult /* Correct answer */ +){ + u32 d1; /* Offset into aKey[] of next data element */ + u32 idx1; /* Offset into aKey[] of next header element */ + u32 szHdr1; /* Number of bytes in header */ + int i = 0; + int rc = 0; + const unsigned char *aKey1 = (const unsigned char *)pKey1; + KeyInfo *pKeyInfo; + Mem mem1; + + pKeyInfo = pPKey2->pKeyInfo; + if( pKeyInfo->db==0 ) return 1; + mem1.enc = pKeyInfo->enc; + mem1.db = pKeyInfo->db; + /* mem1.flags = 0; // Will be initialized by sqlite3VdbeSerialGet() */ + VVA_ONLY( mem1.szMalloc = 0; ) /* Only needed by assert() statements */ + + /* Compilers may complain that mem1.u.i is potentially uninitialized. + ** We could initialize it, as shown here, to silence those complaints. + ** But in fact, mem1.u.i will never actually be used uninitialized, and doing + ** the unnecessary initialization has a measurable negative performance + ** impact, since this routine is a very high runner. And so, we choose + ** to ignore the compiler warnings and leave this variable uninitialized. + */ + /* mem1.u.i = 0; // not needed, here to silence compiler warning */ + + idx1 = getVarint32(aKey1, szHdr1); + if( szHdr1>98307 ) return SQLITE_CORRUPT; + d1 = szHdr1; + assert( pKeyInfo->nAllField>=pPKey2->nField || CORRUPT_DB ); + assert( pKeyInfo->aSortFlags!=0 ); + assert( pKeyInfo->nKeyField>0 ); + assert( idx1<=szHdr1 || CORRUPT_DB ); + do{ + u32 serial_type1; + + /* Read the serial types for the next element in each key. */ + idx1 += getVarint32( aKey1+idx1, serial_type1 ); + + /* Verify that there is enough key space remaining to avoid + ** a buffer overread. The "d1+serial_type1+2" subexpression will + ** always be greater than or equal to the amount of required key space. + ** Use that approximation to avoid the more expensive call to + ** sqlite3VdbeSerialTypeLen() in the common case. + */ + if( d1+(u64)serial_type1+2>(u64)nKey1 + && d1+(u64)sqlite3VdbeSerialTypeLen(serial_type1)>(u64)nKey1 + ){ + break; + } + + /* Extract the values to be compared. + */ + d1 += sqlite3VdbeSerialGet(&aKey1[d1], serial_type1, &mem1); + + /* Do the comparison + */ + rc = sqlite3MemCompare(&mem1, &pPKey2->aMem[i], + pKeyInfo->nAllField>i ? pKeyInfo->aColl[i] : 0); + if( rc!=0 ){ + assert( mem1.szMalloc==0 ); /* See comment below */ + if( (pKeyInfo->aSortFlags[i] & KEYINFO_ORDER_BIGNULL) + && ((mem1.flags & MEM_Null) || (pPKey2->aMem[i].flags & MEM_Null)) + ){ + rc = -rc; + } + if( pKeyInfo->aSortFlags[i] & KEYINFO_ORDER_DESC ){ + rc = -rc; /* Invert the result for DESC sort order. */ + } + goto debugCompareEnd; + } + i++; + }while( idx1nField ); + + /* No memory allocation is ever used on mem1. Prove this using + ** the following assert(). If the assert() fails, it indicates a + ** memory leak and a need to call sqlite3VdbeMemRelease(&mem1). + */ + assert( mem1.szMalloc==0 ); + + /* rc==0 here means that one of the keys ran out of fields and + ** all the fields up to that point were equal. Return the default_rc + ** value. */ + rc = pPKey2->default_rc; + +debugCompareEnd: + if( desiredResult==0 && rc==0 ) return 1; + if( desiredResult<0 && rc<0 ) return 1; + if( desiredResult>0 && rc>0 ) return 1; + if( CORRUPT_DB ) return 1; + if( pKeyInfo->db->mallocFailed ) return 1; + return 0; +} +#endif + +#ifdef SQLITE_DEBUG +/* +** Count the number of fields (a.k.a. columns) in the record given by +** pKey,nKey. The verify that this count is less than or equal to the +** limit given by pKeyInfo->nAllField. +** +** If this constraint is not satisfied, it means that the high-speed +** vdbeRecordCompareInt() and vdbeRecordCompareString() routines will +** not work correctly. If this assert() ever fires, it probably means +** that the KeyInfo.nKeyField or KeyInfo.nAllField values were computed +** incorrectly. +*/ +static void vdbeAssertFieldCountWithinLimits( + int nKey, const void *pKey, /* The record to verify */ + const KeyInfo *pKeyInfo /* Compare size with this KeyInfo */ +){ + int nField = 0; + u32 szHdr; + u32 idx; + u32 notUsed; + const unsigned char *aKey = (const unsigned char*)pKey; + + if( CORRUPT_DB ) return; + idx = getVarint32(aKey, szHdr); + assert( nKey>=0 ); + assert( szHdr<=(u32)nKey ); + while( idxnAllField ); +} +#else +# define vdbeAssertFieldCountWithinLimits(A,B,C) +#endif + +/* +** Both *pMem1 and *pMem2 contain string values. Compare the two values +** using the collation sequence pColl. As usual, return a negative , zero +** or positive value if *pMem1 is less than, equal to or greater than +** *pMem2, respectively. Similar in spirit to "rc = (*pMem1) - (*pMem2);". +*/ +static int vdbeCompareMemString( + const Mem *pMem1, + const Mem *pMem2, + const CollSeq *pColl, + u8 *prcErr /* If an OOM occurs, set to SQLITE_NOMEM */ +){ + if( pMem1->enc==pColl->enc ){ + /* The strings are already in the correct encoding. Call the + ** comparison function directly */ + return pColl->xCmp(pColl->pUser,pMem1->n,pMem1->z,pMem2->n,pMem2->z); + }else{ + int rc; + const void *v1, *v2; + Mem c1; + Mem c2; + sqlite3VdbeMemInit(&c1, pMem1->db, MEM_Null); + sqlite3VdbeMemInit(&c2, pMem1->db, MEM_Null); + sqlite3VdbeMemShallowCopy(&c1, pMem1, MEM_Ephem); + sqlite3VdbeMemShallowCopy(&c2, pMem2, MEM_Ephem); + v1 = sqlite3ValueText((sqlite3_value*)&c1, pColl->enc); + v2 = sqlite3ValueText((sqlite3_value*)&c2, pColl->enc); + if( (v1==0 || v2==0) ){ + if( prcErr ) *prcErr = SQLITE_NOMEM_BKPT; + rc = 0; + }else{ + rc = pColl->xCmp(pColl->pUser, c1.n, v1, c2.n, v2); + } + sqlite3VdbeMemRelease(&c1); + sqlite3VdbeMemRelease(&c2); + return rc; + } +} + +/* +** The input pBlob is guaranteed to be a Blob that is not marked +** with MEM_Zero. Return true if it could be a zero-blob. +*/ +static int isAllZero(const char *z, int n){ + int i; + for(i=0; in; + int n2 = pB2->n; + + /* It is possible to have a Blob value that has some non-zero content + ** followed by zero content. But that only comes up for Blobs formed + ** by the OP_MakeRecord opcode, and such Blobs never get passed into + ** sqlite3MemCompare(). */ + assert( (pB1->flags & MEM_Zero)==0 || n1==0 ); + assert( (pB2->flags & MEM_Zero)==0 || n2==0 ); + + if( (pB1->flags|pB2->flags) & MEM_Zero ){ + if( pB1->flags & pB2->flags & MEM_Zero ){ + return pB1->u.nZero - pB2->u.nZero; + }else if( pB1->flags & MEM_Zero ){ + if( !isAllZero(pB2->z, pB2->n) ) return -1; + return pB1->u.nZero - n2; + }else{ + if( !isAllZero(pB1->z, pB1->n) ) return +1; + return n1 - pB2->u.nZero; + } + } + c = memcmp(pB1->z, pB2->z, n1>n2 ? n2 : n1); + if( c ) return c; + return n1 - n2; +} + +/* +** Do a comparison between a 64-bit signed integer and a 64-bit floating-point +** number. Return negative, zero, or positive if the first (i64) is less than, +** equal to, or greater than the second (double). +*/ +static int sqlite3IntFloatCompare(i64 i, double r){ + if( sizeof(LONGDOUBLE_TYPE)>8 ){ + LONGDOUBLE_TYPE x = (LONGDOUBLE_TYPE)i; + if( xr ) return +1; + return 0; + }else{ + i64 y; + double s; + if( r<-9223372036854775808.0 ) return +1; + if( r>=9223372036854775808.0 ) return -1; + y = (i64)r; + if( iy ) return +1; + s = (double)i; + if( sr ) return +1; + return 0; + } +} + +/* +** Compare the values contained by the two memory cells, returning +** negative, zero or positive if pMem1 is less than, equal to, or greater +** than pMem2. Sorting order is NULL's first, followed by numbers (integers +** and reals) sorted numerically, followed by text ordered by the collating +** sequence pColl and finally blob's ordered by memcmp(). +** +** Two NULL values are considered equal by this function. +*/ +SQLITE_PRIVATE int sqlite3MemCompare(const Mem *pMem1, const Mem *pMem2, const CollSeq *pColl){ + int f1, f2; + int combined_flags; + + f1 = pMem1->flags; + f2 = pMem2->flags; + combined_flags = f1|f2; + assert( !sqlite3VdbeMemIsRowSet(pMem1) && !sqlite3VdbeMemIsRowSet(pMem2) ); + + /* If one value is NULL, it is less than the other. If both values + ** are NULL, return 0. + */ + if( combined_flags&MEM_Null ){ + return (f2&MEM_Null) - (f1&MEM_Null); + } + + /* At least one of the two values is a number + */ + if( combined_flags&(MEM_Int|MEM_Real|MEM_IntReal) ){ + testcase( combined_flags & MEM_Int ); + testcase( combined_flags & MEM_Real ); + testcase( combined_flags & MEM_IntReal ); + if( (f1 & f2 & (MEM_Int|MEM_IntReal))!=0 ){ + testcase( f1 & f2 & MEM_Int ); + testcase( f1 & f2 & MEM_IntReal ); + if( pMem1->u.i < pMem2->u.i ) return -1; + if( pMem1->u.i > pMem2->u.i ) return +1; + return 0; + } + if( (f1 & f2 & MEM_Real)!=0 ){ + if( pMem1->u.r < pMem2->u.r ) return -1; + if( pMem1->u.r > pMem2->u.r ) return +1; + return 0; + } + if( (f1&(MEM_Int|MEM_IntReal))!=0 ){ + testcase( f1 & MEM_Int ); + testcase( f1 & MEM_IntReal ); + if( (f2&MEM_Real)!=0 ){ + return sqlite3IntFloatCompare(pMem1->u.i, pMem2->u.r); + }else if( (f2&(MEM_Int|MEM_IntReal))!=0 ){ + if( pMem1->u.i < pMem2->u.i ) return -1; + if( pMem1->u.i > pMem2->u.i ) return +1; + return 0; + }else{ + return -1; + } + } + if( (f1&MEM_Real)!=0 ){ + if( (f2&(MEM_Int|MEM_IntReal))!=0 ){ + testcase( f2 & MEM_Int ); + testcase( f2 & MEM_IntReal ); + return -sqlite3IntFloatCompare(pMem2->u.i, pMem1->u.r); + }else{ + return -1; + } + } + return +1; + } + + /* If one value is a string and the other is a blob, the string is less. + ** If both are strings, compare using the collating functions. + */ + if( combined_flags&MEM_Str ){ + if( (f1 & MEM_Str)==0 ){ + return 1; + } + if( (f2 & MEM_Str)==0 ){ + return -1; + } + + assert( pMem1->enc==pMem2->enc || pMem1->db->mallocFailed ); + assert( pMem1->enc==SQLITE_UTF8 || + pMem1->enc==SQLITE_UTF16LE || pMem1->enc==SQLITE_UTF16BE ); + + /* The collation sequence must be defined at this point, even if + ** the user deletes the collation sequence after the vdbe program is + ** compiled (this was not always the case). + */ + assert( !pColl || pColl->xCmp ); + + if( pColl ){ + return vdbeCompareMemString(pMem1, pMem2, pColl, 0); + } + /* If a NULL pointer was passed as the collate function, fall through + ** to the blob case and use memcmp(). */ + } + + /* Both values must be blobs. Compare using memcmp(). */ + return sqlite3BlobCompare(pMem1, pMem2); +} + + +/* +** The first argument passed to this function is a serial-type that +** corresponds to an integer - all values between 1 and 9 inclusive +** except 7. The second points to a buffer containing an integer value +** serialized according to serial_type. This function deserializes +** and returns the value. +*/ +static i64 vdbeRecordDecodeInt(u32 serial_type, const u8 *aKey){ + u32 y; + assert( CORRUPT_DB || (serial_type>=1 && serial_type<=9 && serial_type!=7) ); + switch( serial_type ){ + case 0: + case 1: + testcase( aKey[0]&0x80 ); + return ONE_BYTE_INT(aKey); + case 2: + testcase( aKey[0]&0x80 ); + return TWO_BYTE_INT(aKey); + case 3: + testcase( aKey[0]&0x80 ); + return THREE_BYTE_INT(aKey); + case 4: { + testcase( aKey[0]&0x80 ); + y = FOUR_BYTE_UINT(aKey); + return (i64)*(int*)&y; + } + case 5: { + testcase( aKey[0]&0x80 ); + return FOUR_BYTE_UINT(aKey+2) + (((i64)1)<<32)*TWO_BYTE_INT(aKey); + } + case 6: { + u64 x = FOUR_BYTE_UINT(aKey); + testcase( aKey[0]&0x80 ); + x = (x<<32) | FOUR_BYTE_UINT(aKey+4); + return (i64)*(i64*)&x; + } + } + + return (serial_type - 8); +} + +/* +** This function compares the two table rows or index records +** specified by {nKey1, pKey1} and pPKey2. It returns a negative, zero +** or positive integer if key1 is less than, equal to or +** greater than key2. The {nKey1, pKey1} key must be a blob +** created by the OP_MakeRecord opcode of the VDBE. The pPKey2 +** key must be a parsed key such as obtained from +** sqlite3VdbeParseRecord. +** +** If argument bSkip is non-zero, it is assumed that the caller has already +** determined that the first fields of the keys are equal. +** +** Key1 and Key2 do not have to contain the same number of fields. If all +** fields that appear in both keys are equal, then pPKey2->default_rc is +** returned. +** +** If database corruption is discovered, set pPKey2->errCode to +** SQLITE_CORRUPT and return 0. If an OOM error is encountered, +** pPKey2->errCode is set to SQLITE_NOMEM and, if it is not NULL, the +** malloc-failed flag set on database handle (pPKey2->pKeyInfo->db). +*/ +SQLITE_PRIVATE int sqlite3VdbeRecordCompareWithSkip( + int nKey1, const void *pKey1, /* Left key */ + UnpackedRecord *pPKey2, /* Right key */ + int bSkip /* If true, skip the first field */ +){ + u32 d1; /* Offset into aKey[] of next data element */ + int i; /* Index of next field to compare */ + u32 szHdr1; /* Size of record header in bytes */ + u32 idx1; /* Offset of first type in header */ + int rc = 0; /* Return value */ + Mem *pRhs = pPKey2->aMem; /* Next field of pPKey2 to compare */ + KeyInfo *pKeyInfo; + const unsigned char *aKey1 = (const unsigned char *)pKey1; + Mem mem1; + + /* If bSkip is true, then the caller has already determined that the first + ** two elements in the keys are equal. Fix the various stack variables so + ** that this routine begins comparing at the second field. */ + if( bSkip ){ + u32 s1; + idx1 = 1 + getVarint32(&aKey1[1], s1); + szHdr1 = aKey1[0]; + d1 = szHdr1 + sqlite3VdbeSerialTypeLen(s1); + i = 1; + pRhs++; + }else{ + idx1 = getVarint32(aKey1, szHdr1); + d1 = szHdr1; + i = 0; + } + if( d1>(unsigned)nKey1 ){ + pPKey2->errCode = (u8)SQLITE_CORRUPT_BKPT; + return 0; /* Corruption */ + } + + VVA_ONLY( mem1.szMalloc = 0; ) /* Only needed by assert() statements */ + assert( pPKey2->pKeyInfo->nAllField>=pPKey2->nField + || CORRUPT_DB ); + assert( pPKey2->pKeyInfo->aSortFlags!=0 ); + assert( pPKey2->pKeyInfo->nKeyField>0 ); + assert( idx1<=szHdr1 || CORRUPT_DB ); + do{ + u32 serial_type; + + /* RHS is an integer */ + if( pRhs->flags & (MEM_Int|MEM_IntReal) ){ + testcase( pRhs->flags & MEM_Int ); + testcase( pRhs->flags & MEM_IntReal ); + serial_type = aKey1[idx1]; + testcase( serial_type==12 ); + if( serial_type>=10 ){ + rc = +1; + }else if( serial_type==0 ){ + rc = -1; + }else if( serial_type==7 ){ + sqlite3VdbeSerialGet(&aKey1[d1], serial_type, &mem1); + rc = -sqlite3IntFloatCompare(pRhs->u.i, mem1.u.r); + }else{ + i64 lhs = vdbeRecordDecodeInt(serial_type, &aKey1[d1]); + i64 rhs = pRhs->u.i; + if( lhsrhs ){ + rc = +1; + } + } + } + + /* RHS is real */ + else if( pRhs->flags & MEM_Real ){ + serial_type = aKey1[idx1]; + if( serial_type>=10 ){ + /* Serial types 12 or greater are strings and blobs (greater than + ** numbers). Types 10 and 11 are currently "reserved for future + ** use", so it doesn't really matter what the results of comparing + ** them to numberic values are. */ + rc = +1; + }else if( serial_type==0 ){ + rc = -1; + }else{ + sqlite3VdbeSerialGet(&aKey1[d1], serial_type, &mem1); + if( serial_type==7 ){ + if( mem1.u.ru.r ){ + rc = -1; + }else if( mem1.u.r>pRhs->u.r ){ + rc = +1; + } + }else{ + rc = sqlite3IntFloatCompare(mem1.u.i, pRhs->u.r); + } + } + } + + /* RHS is a string */ + else if( pRhs->flags & MEM_Str ){ + getVarint32(&aKey1[idx1], serial_type); + testcase( serial_type==12 ); + if( serial_type<12 ){ + rc = -1; + }else if( !(serial_type & 0x01) ){ + rc = +1; + }else{ + mem1.n = (serial_type - 12) / 2; + testcase( (d1+mem1.n)==(unsigned)nKey1 ); + testcase( (d1+mem1.n+1)==(unsigned)nKey1 ); + if( (d1+mem1.n) > (unsigned)nKey1 + || (pKeyInfo = pPKey2->pKeyInfo)->nAllField<=i + ){ + pPKey2->errCode = (u8)SQLITE_CORRUPT_BKPT; + return 0; /* Corruption */ + }else if( pKeyInfo->aColl[i] ){ + mem1.enc = pKeyInfo->enc; + mem1.db = pKeyInfo->db; + mem1.flags = MEM_Str; + mem1.z = (char*)&aKey1[d1]; + rc = vdbeCompareMemString( + &mem1, pRhs, pKeyInfo->aColl[i], &pPKey2->errCode + ); + }else{ + int nCmp = MIN(mem1.n, pRhs->n); + rc = memcmp(&aKey1[d1], pRhs->z, nCmp); + if( rc==0 ) rc = mem1.n - pRhs->n; + } + } + } + + /* RHS is a blob */ + else if( pRhs->flags & MEM_Blob ){ + assert( (pRhs->flags & MEM_Zero)==0 || pRhs->n==0 ); + getVarint32(&aKey1[idx1], serial_type); + testcase( serial_type==12 ); + if( serial_type<12 || (serial_type & 0x01) ){ + rc = -1; + }else{ + int nStr = (serial_type - 12) / 2; + testcase( (d1+nStr)==(unsigned)nKey1 ); + testcase( (d1+nStr+1)==(unsigned)nKey1 ); + if( (d1+nStr) > (unsigned)nKey1 ){ + pPKey2->errCode = (u8)SQLITE_CORRUPT_BKPT; + return 0; /* Corruption */ + }else if( pRhs->flags & MEM_Zero ){ + if( !isAllZero((const char*)&aKey1[d1],nStr) ){ + rc = 1; + }else{ + rc = nStr - pRhs->u.nZero; + } + }else{ + int nCmp = MIN(nStr, pRhs->n); + rc = memcmp(&aKey1[d1], pRhs->z, nCmp); + if( rc==0 ) rc = nStr - pRhs->n; + } + } + } + + /* RHS is null */ + else{ + serial_type = aKey1[idx1]; + rc = (serial_type!=0); + } + + if( rc!=0 ){ + int sortFlags = pPKey2->pKeyInfo->aSortFlags[i]; + if( sortFlags ){ + if( (sortFlags & KEYINFO_ORDER_BIGNULL)==0 + || ((sortFlags & KEYINFO_ORDER_DESC) + !=(serial_type==0 || (pRhs->flags&MEM_Null))) + ){ + rc = -rc; + } + } + assert( vdbeRecordCompareDebug(nKey1, pKey1, pPKey2, rc) ); + assert( mem1.szMalloc==0 ); /* See comment below */ + return rc; + } + + i++; + if( i==pPKey2->nField ) break; + pRhs++; + d1 += sqlite3VdbeSerialTypeLen(serial_type); + idx1 += sqlite3VarintLen(serial_type); + }while( idx1<(unsigned)szHdr1 && d1<=(unsigned)nKey1 ); + + /* No memory allocation is ever used on mem1. Prove this using + ** the following assert(). If the assert() fails, it indicates a + ** memory leak and a need to call sqlite3VdbeMemRelease(&mem1). */ + assert( mem1.szMalloc==0 ); + + /* rc==0 here means that one or both of the keys ran out of fields and + ** all the fields up to that point were equal. Return the default_rc + ** value. */ + assert( CORRUPT_DB + || vdbeRecordCompareDebug(nKey1, pKey1, pPKey2, pPKey2->default_rc) + || pPKey2->pKeyInfo->db->mallocFailed + ); + pPKey2->eqSeen = 1; + return pPKey2->default_rc; +} +SQLITE_PRIVATE int sqlite3VdbeRecordCompare( + int nKey1, const void *pKey1, /* Left key */ + UnpackedRecord *pPKey2 /* Right key */ +){ + return sqlite3VdbeRecordCompareWithSkip(nKey1, pKey1, pPKey2, 0); +} + + +/* +** This function is an optimized version of sqlite3VdbeRecordCompare() +** that (a) the first field of pPKey2 is an integer, and (b) the +** size-of-header varint at the start of (pKey1/nKey1) fits in a single +** byte (i.e. is less than 128). +** +** To avoid concerns about buffer overreads, this routine is only used +** on schemas where the maximum valid header size is 63 bytes or less. +*/ +static int vdbeRecordCompareInt( + int nKey1, const void *pKey1, /* Left key */ + UnpackedRecord *pPKey2 /* Right key */ +){ + const u8 *aKey = &((const u8*)pKey1)[*(const u8*)pKey1 & 0x3F]; + int serial_type = ((const u8*)pKey1)[1]; + int res; + u32 y; + u64 x; + i64 v; + i64 lhs; + + vdbeAssertFieldCountWithinLimits(nKey1, pKey1, pPKey2->pKeyInfo); + assert( (*(u8*)pKey1)<=0x3F || CORRUPT_DB ); + switch( serial_type ){ + case 1: { /* 1-byte signed integer */ + lhs = ONE_BYTE_INT(aKey); + testcase( lhs<0 ); + break; + } + case 2: { /* 2-byte signed integer */ + lhs = TWO_BYTE_INT(aKey); + testcase( lhs<0 ); + break; + } + case 3: { /* 3-byte signed integer */ + lhs = THREE_BYTE_INT(aKey); + testcase( lhs<0 ); + break; + } + case 4: { /* 4-byte signed integer */ + y = FOUR_BYTE_UINT(aKey); + lhs = (i64)*(int*)&y; + testcase( lhs<0 ); + break; + } + case 5: { /* 6-byte signed integer */ + lhs = FOUR_BYTE_UINT(aKey+2) + (((i64)1)<<32)*TWO_BYTE_INT(aKey); + testcase( lhs<0 ); + break; + } + case 6: { /* 8-byte signed integer */ + x = FOUR_BYTE_UINT(aKey); + x = (x<<32) | FOUR_BYTE_UINT(aKey+4); + lhs = *(i64*)&x; + testcase( lhs<0 ); + break; + } + case 8: + lhs = 0; + break; + case 9: + lhs = 1; + break; + + /* This case could be removed without changing the results of running + ** this code. Including it causes gcc to generate a faster switch + ** statement (since the range of switch targets now starts at zero and + ** is contiguous) but does not cause any duplicate code to be generated + ** (as gcc is clever enough to combine the two like cases). Other + ** compilers might be similar. */ + case 0: case 7: + return sqlite3VdbeRecordCompare(nKey1, pKey1, pPKey2); + + default: + return sqlite3VdbeRecordCompare(nKey1, pKey1, pPKey2); + } + + v = pPKey2->aMem[0].u.i; + if( v>lhs ){ + res = pPKey2->r1; + }else if( vr2; + }else if( pPKey2->nField>1 ){ + /* The first fields of the two keys are equal. Compare the trailing + ** fields. */ + res = sqlite3VdbeRecordCompareWithSkip(nKey1, pKey1, pPKey2, 1); + }else{ + /* The first fields of the two keys are equal and there are no trailing + ** fields. Return pPKey2->default_rc in this case. */ + res = pPKey2->default_rc; + pPKey2->eqSeen = 1; + } + + assert( vdbeRecordCompareDebug(nKey1, pKey1, pPKey2, res) ); + return res; +} + +/* +** This function is an optimized version of sqlite3VdbeRecordCompare() +** that (a) the first field of pPKey2 is a string, that (b) the first field +** uses the collation sequence BINARY and (c) that the size-of-header varint +** at the start of (pKey1/nKey1) fits in a single byte. +*/ +static int vdbeRecordCompareString( + int nKey1, const void *pKey1, /* Left key */ + UnpackedRecord *pPKey2 /* Right key */ +){ + const u8 *aKey1 = (const u8*)pKey1; + int serial_type; + int res; + + assert( pPKey2->aMem[0].flags & MEM_Str ); + vdbeAssertFieldCountWithinLimits(nKey1, pKey1, pPKey2->pKeyInfo); + getVarint32(&aKey1[1], serial_type); + if( serial_type<12 ){ + res = pPKey2->r1; /* (pKey1/nKey1) is a number or a null */ + }else if( !(serial_type & 0x01) ){ + res = pPKey2->r2; /* (pKey1/nKey1) is a blob */ + }else{ + int nCmp; + int nStr; + int szHdr = aKey1[0]; + + nStr = (serial_type-12) / 2; + if( (szHdr + nStr) > nKey1 ){ + pPKey2->errCode = (u8)SQLITE_CORRUPT_BKPT; + return 0; /* Corruption */ + } + nCmp = MIN( pPKey2->aMem[0].n, nStr ); + res = memcmp(&aKey1[szHdr], pPKey2->aMem[0].z, nCmp); + + if( res>0 ){ + res = pPKey2->r2; + }else if( res<0 ){ + res = pPKey2->r1; + }else{ + res = nStr - pPKey2->aMem[0].n; + if( res==0 ){ + if( pPKey2->nField>1 ){ + res = sqlite3VdbeRecordCompareWithSkip(nKey1, pKey1, pPKey2, 1); + }else{ + res = pPKey2->default_rc; + pPKey2->eqSeen = 1; + } + }else if( res>0 ){ + res = pPKey2->r2; + }else{ + res = pPKey2->r1; + } + } + } + + assert( vdbeRecordCompareDebug(nKey1, pKey1, pPKey2, res) + || CORRUPT_DB + || pPKey2->pKeyInfo->db->mallocFailed + ); + return res; +} + +/* +** Return a pointer to an sqlite3VdbeRecordCompare() compatible function +** suitable for comparing serialized records to the unpacked record passed +** as the only argument. +*/ +SQLITE_PRIVATE RecordCompare sqlite3VdbeFindCompare(UnpackedRecord *p){ + /* varintRecordCompareInt() and varintRecordCompareString() both assume + ** that the size-of-header varint that occurs at the start of each record + ** fits in a single byte (i.e. is 127 or less). varintRecordCompareInt() + ** also assumes that it is safe to overread a buffer by at least the + ** maximum possible legal header size plus 8 bytes. Because there is + ** guaranteed to be at least 74 (but not 136) bytes of padding following each + ** buffer passed to varintRecordCompareInt() this makes it convenient to + ** limit the size of the header to 64 bytes in cases where the first field + ** is an integer. + ** + ** The easiest way to enforce this limit is to consider only records with + ** 13 fields or less. If the first field is an integer, the maximum legal + ** header size is (12*5 + 1 + 1) bytes. */ + if( p->pKeyInfo->nAllField<=13 ){ + int flags = p->aMem[0].flags; + if( p->pKeyInfo->aSortFlags[0] ){ + if( p->pKeyInfo->aSortFlags[0] & KEYINFO_ORDER_BIGNULL ){ + return sqlite3VdbeRecordCompare; + } + p->r1 = 1; + p->r2 = -1; + }else{ + p->r1 = -1; + p->r2 = 1; + } + if( (flags & MEM_Int) ){ + return vdbeRecordCompareInt; + } + testcase( flags & MEM_Real ); + testcase( flags & MEM_Null ); + testcase( flags & MEM_Blob ); + if( (flags & (MEM_Real|MEM_IntReal|MEM_Null|MEM_Blob))==0 + && p->pKeyInfo->aColl[0]==0 + ){ + assert( flags & MEM_Str ); + return vdbeRecordCompareString; + } + } + + return sqlite3VdbeRecordCompare; +} + +/* +** pCur points at an index entry created using the OP_MakeRecord opcode. +** Read the rowid (the last field in the record) and store it in *rowid. +** Return SQLITE_OK if everything works, or an error code otherwise. +** +** pCur might be pointing to text obtained from a corrupt database file. +** So the content cannot be trusted. Do appropriate checks on the content. +*/ +SQLITE_PRIVATE int sqlite3VdbeIdxRowid(sqlite3 *db, BtCursor *pCur, i64 *rowid){ + i64 nCellKey = 0; + int rc; + u32 szHdr; /* Size of the header */ + u32 typeRowid; /* Serial type of the rowid */ + u32 lenRowid; /* Size of the rowid */ + Mem m, v; + + /* Get the size of the index entry. Only indices entries of less + ** than 2GiB are support - anything large must be database corruption. + ** Any corruption is detected in sqlite3BtreeParseCellPtr(), though, so + ** this code can safely assume that nCellKey is 32-bits + */ + assert( sqlite3BtreeCursorIsValid(pCur) ); + nCellKey = sqlite3BtreePayloadSize(pCur); + assert( (nCellKey & SQLITE_MAX_U32)==(u64)nCellKey ); + + /* Read in the complete content of the index entry */ + sqlite3VdbeMemInit(&m, db, 0); + rc = sqlite3VdbeMemFromBtree(pCur, 0, (u32)nCellKey, &m); + if( rc ){ + return rc; + } + + /* The index entry must begin with a header size */ + (void)getVarint32((u8*)m.z, szHdr); + testcase( szHdr==3 ); + testcase( szHdr==m.n ); + testcase( szHdr>0x7fffffff ); + assert( m.n>=0 ); + if( unlikely(szHdr<3 || szHdr>(unsigned)m.n) ){ + goto idx_rowid_corruption; + } + + /* The last field of the index should be an integer - the ROWID. + ** Verify that the last entry really is an integer. */ + (void)getVarint32((u8*)&m.z[szHdr-1], typeRowid); + testcase( typeRowid==1 ); + testcase( typeRowid==2 ); + testcase( typeRowid==3 ); + testcase( typeRowid==4 ); + testcase( typeRowid==5 ); + testcase( typeRowid==6 ); + testcase( typeRowid==8 ); + testcase( typeRowid==9 ); + if( unlikely(typeRowid<1 || typeRowid>9 || typeRowid==7) ){ + goto idx_rowid_corruption; + } + lenRowid = sqlite3SmallTypeSizes[typeRowid]; + testcase( (u32)m.n==szHdr+lenRowid ); + if( unlikely((u32)m.neCurType==CURTYPE_BTREE ); + pCur = pC->uc.pCursor; + assert( sqlite3BtreeCursorIsValid(pCur) ); + nCellKey = sqlite3BtreePayloadSize(pCur); + /* nCellKey will always be between 0 and 0xffffffff because of the way + ** that btreeParseCellPtr() and sqlite3GetVarint32() are implemented */ + if( nCellKey<=0 || nCellKey>0x7fffffff ){ + *res = 0; + return SQLITE_CORRUPT_BKPT; + } + sqlite3VdbeMemInit(&m, db, 0); + rc = sqlite3VdbeMemFromBtree(pCur, 0, (u32)nCellKey, &m); + if( rc ){ + return rc; + } + *res = sqlite3VdbeRecordCompareWithSkip(m.n, m.z, pUnpacked, 0); + sqlite3VdbeMemRelease(&m); + return SQLITE_OK; +} + +/* +** This routine sets the value to be returned by subsequent calls to +** sqlite3_changes() on the database handle 'db'. +*/ +SQLITE_PRIVATE void sqlite3VdbeSetChanges(sqlite3 *db, int nChange){ + assert( sqlite3_mutex_held(db->mutex) ); + db->nChange = nChange; + db->nTotalChange += nChange; +} + +/* +** Set a flag in the vdbe to update the change counter when it is finalised +** or reset. +*/ +SQLITE_PRIVATE void sqlite3VdbeCountChanges(Vdbe *v){ + v->changeCntOn = 1; +} + +/* +** Mark every prepared statement associated with a database connection +** as expired. +** +** An expired statement means that recompilation of the statement is +** recommend. Statements expire when things happen that make their +** programs obsolete. Removing user-defined functions or collating +** sequences, or changing an authorization function are the types of +** things that make prepared statements obsolete. +** +** If iCode is 1, then expiration is advisory. The statement should +** be reprepared before being restarted, but if it is already running +** it is allowed to run to completion. +** +** Internally, this function just sets the Vdbe.expired flag on all +** prepared statements. The flag is set to 1 for an immediate expiration +** and set to 2 for an advisory expiration. +*/ +SQLITE_PRIVATE void sqlite3ExpirePreparedStatements(sqlite3 *db, int iCode){ + Vdbe *p; + for(p = db->pVdbe; p; p=p->pNext){ + p->expired = iCode+1; + } +} + +/* +** Return the database associated with the Vdbe. +*/ +SQLITE_PRIVATE sqlite3 *sqlite3VdbeDb(Vdbe *v){ + return v->db; +} + +/* +** Return the SQLITE_PREPARE flags for a Vdbe. +*/ +SQLITE_PRIVATE u8 sqlite3VdbePrepareFlags(Vdbe *v){ + return v->prepFlags; +} + +/* +** Return a pointer to an sqlite3_value structure containing the value bound +** parameter iVar of VM v. Except, if the value is an SQL NULL, return +** 0 instead. Unless it is NULL, apply affinity aff (one of the SQLITE_AFF_* +** constants) to the value before returning it. +** +** The returned value must be freed by the caller using sqlite3ValueFree(). +*/ +SQLITE_PRIVATE sqlite3_value *sqlite3VdbeGetBoundValue(Vdbe *v, int iVar, u8 aff){ + assert( iVar>0 ); + if( v ){ + Mem *pMem = &v->aVar[iVar-1]; + assert( (v->db->flags & SQLITE_EnableQPSG)==0 ); + if( 0==(pMem->flags & MEM_Null) ){ + sqlite3_value *pRet = sqlite3ValueNew(v->db); + if( pRet ){ + sqlite3VdbeMemCopy((Mem *)pRet, pMem); + sqlite3ValueApplyAffinity(pRet, aff, SQLITE_UTF8); + } + return pRet; + } + } + return 0; +} + +/* +** Configure SQL variable iVar so that binding a new value to it signals +** to sqlite3_reoptimize() that re-preparing the statement may result +** in a better query plan. +*/ +SQLITE_PRIVATE void sqlite3VdbeSetVarmask(Vdbe *v, int iVar){ + assert( iVar>0 ); + assert( (v->db->flags & SQLITE_EnableQPSG)==0 ); + if( iVar>=32 ){ + v->expmask |= 0x80000000; + }else{ + v->expmask |= ((u32)1 << (iVar-1)); + } +} + +/* +** Cause a function to throw an error if it was call from OP_PureFunc +** rather than OP_Function. +** +** OP_PureFunc means that the function must be deterministic, and should +** throw an error if it is given inputs that would make it non-deterministic. +** This routine is invoked by date/time functions that use non-deterministic +** features such as 'now'. +*/ +SQLITE_PRIVATE int sqlite3NotPureFunc(sqlite3_context *pCtx){ +#ifdef SQLITE_ENABLE_STAT4 + if( pCtx->pVdbe==0 ) return 1; +#endif + if( pCtx->pVdbe->aOp[pCtx->iOp].opcode==OP_PureFunc ){ + sqlite3_result_error(pCtx, + "non-deterministic function in index expression or CHECK constraint", + -1); + return 0; + } + return 1; +} + +#ifndef SQLITE_OMIT_VIRTUALTABLE +/* +** Transfer error message text from an sqlite3_vtab.zErrMsg (text stored +** in memory obtained from sqlite3_malloc) into a Vdbe.zErrMsg (text stored +** in memory obtained from sqlite3DbMalloc). +*/ +SQLITE_PRIVATE void sqlite3VtabImportErrmsg(Vdbe *p, sqlite3_vtab *pVtab){ + if( pVtab->zErrMsg ){ + sqlite3 *db = p->db; + sqlite3DbFree(db, p->zErrMsg); + p->zErrMsg = sqlite3DbStrDup(db, pVtab->zErrMsg); + sqlite3_free(pVtab->zErrMsg); + pVtab->zErrMsg = 0; + } +} +#endif /* SQLITE_OMIT_VIRTUALTABLE */ + +#ifdef SQLITE_ENABLE_PREUPDATE_HOOK + +/* +** If the second argument is not NULL, release any allocations associated +** with the memory cells in the p->aMem[] array. Also free the UnpackedRecord +** structure itself, using sqlite3DbFree(). +** +** This function is used to free UnpackedRecord structures allocated by +** the vdbeUnpackRecord() function found in vdbeapi.c. +*/ +static void vdbeFreeUnpacked(sqlite3 *db, int nField, UnpackedRecord *p){ + if( p ){ + int i; + for(i=0; iaMem[i]; + if( pMem->zMalloc ) sqlite3VdbeMemRelease(pMem); + } + sqlite3DbFreeNN(db, p); + } +} +#endif /* SQLITE_ENABLE_PREUPDATE_HOOK */ + +#ifdef SQLITE_ENABLE_PREUPDATE_HOOK +/* +** Invoke the pre-update hook. If this is an UPDATE or DELETE pre-update call, +** then cursor passed as the second argument should point to the row about +** to be update or deleted. If the application calls sqlite3_preupdate_old(), +** the required value will be read from the row the cursor points to. +*/ +SQLITE_PRIVATE void sqlite3VdbePreUpdateHook( + Vdbe *v, /* Vdbe pre-update hook is invoked by */ + VdbeCursor *pCsr, /* Cursor to grab old.* values from */ + int op, /* SQLITE_INSERT, UPDATE or DELETE */ + const char *zDb, /* Database name */ + Table *pTab, /* Modified table */ + i64 iKey1, /* Initial key value */ + int iReg /* Register for new.* record */ +){ + sqlite3 *db = v->db; + i64 iKey2; + PreUpdate preupdate; + const char *zTbl = pTab->zName; + static const u8 fakeSortOrder = 0; + + assert( db->pPreUpdate==0 ); + memset(&preupdate, 0, sizeof(PreUpdate)); + if( HasRowid(pTab)==0 ){ + iKey1 = iKey2 = 0; + preupdate.pPk = sqlite3PrimaryKeyIndex(pTab); + }else{ + if( op==SQLITE_UPDATE ){ + iKey2 = v->aMem[iReg].u.i; + }else{ + iKey2 = iKey1; + } + } + + assert( pCsr->nField==pTab->nCol + || (pCsr->nField==pTab->nCol+1 && op==SQLITE_DELETE && iReg==-1) + ); + + preupdate.v = v; + preupdate.pCsr = pCsr; + preupdate.op = op; + preupdate.iNewReg = iReg; + preupdate.keyinfo.db = db; + preupdate.keyinfo.enc = ENC(db); + preupdate.keyinfo.nKeyField = pTab->nCol; + preupdate.keyinfo.aSortFlags = (u8*)&fakeSortOrder; + preupdate.iKey1 = iKey1; + preupdate.iKey2 = iKey2; + preupdate.pTab = pTab; + + db->pPreUpdate = &preupdate; + db->xPreUpdateCallback(db->pPreUpdateArg, db, op, zDb, zTbl, iKey1, iKey2); + db->pPreUpdate = 0; + sqlite3DbFree(db, preupdate.aRecord); + vdbeFreeUnpacked(db, preupdate.keyinfo.nKeyField+1, preupdate.pUnpacked); + vdbeFreeUnpacked(db, preupdate.keyinfo.nKeyField+1, preupdate.pNewUnpacked); + if( preupdate.aNew ){ + int i; + for(i=0; inField; i++){ + sqlite3VdbeMemRelease(&preupdate.aNew[i]); + } + sqlite3DbFreeNN(db, preupdate.aNew); + } +} +#endif /* SQLITE_ENABLE_PREUPDATE_HOOK */ + +/************** End of vdbeaux.c *********************************************/ +/************** Begin file vdbeapi.c *****************************************/ +/* +** 2004 May 26 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** +** This file contains code use to implement APIs that are part of the +** VDBE. +*/ +/* #include "sqliteInt.h" */ +/* #include "vdbeInt.h" */ + +#ifndef SQLITE_OMIT_DEPRECATED +/* +** Return TRUE (non-zero) of the statement supplied as an argument needs +** to be recompiled. A statement needs to be recompiled whenever the +** execution environment changes in a way that would alter the program +** that sqlite3_prepare() generates. For example, if new functions or +** collating sequences are registered or if an authorizer function is +** added or changed. +*/ +SQLITE_API int sqlite3_expired(sqlite3_stmt *pStmt){ + Vdbe *p = (Vdbe*)pStmt; + return p==0 || p->expired; +} +#endif + +/* +** Check on a Vdbe to make sure it has not been finalized. Log +** an error and return true if it has been finalized (or is otherwise +** invalid). Return false if it is ok. +*/ +static int vdbeSafety(Vdbe *p){ + if( p->db==0 ){ + sqlite3_log(SQLITE_MISUSE, "API called with finalized prepared statement"); + return 1; + }else{ + return 0; + } +} +static int vdbeSafetyNotNull(Vdbe *p){ + if( p==0 ){ + sqlite3_log(SQLITE_MISUSE, "API called with NULL prepared statement"); + return 1; + }else{ + return vdbeSafety(p); + } +} + +#ifndef SQLITE_OMIT_TRACE +/* +** Invoke the profile callback. This routine is only called if we already +** know that the profile callback is defined and needs to be invoked. +*/ +static SQLITE_NOINLINE void invokeProfileCallback(sqlite3 *db, Vdbe *p){ + sqlite3_int64 iNow; + sqlite3_int64 iElapse; + assert( p->startTime>0 ); + assert( (db->mTrace & (SQLITE_TRACE_PROFILE|SQLITE_TRACE_XPROFILE))!=0 ); + assert( db->init.busy==0 ); + assert( p->zSql!=0 ); + sqlite3OsCurrentTimeInt64(db->pVfs, &iNow); + iElapse = (iNow - p->startTime)*1000000; +#ifndef SQLITE_OMIT_DEPRECATED + if( db->xProfile ){ + db->xProfile(db->pProfileArg, p->zSql, iElapse); + } +#endif + if( db->mTrace & SQLITE_TRACE_PROFILE ){ + db->xTrace(SQLITE_TRACE_PROFILE, db->pTraceArg, p, (void*)&iElapse); + } + p->startTime = 0; +} +/* +** The checkProfileCallback(DB,P) macro checks to see if a profile callback +** is needed, and it invokes the callback if it is needed. +*/ +# define checkProfileCallback(DB,P) \ + if( ((P)->startTime)>0 ){ invokeProfileCallback(DB,P); } +#else +# define checkProfileCallback(DB,P) /*no-op*/ +#endif + +/* +** The following routine destroys a virtual machine that is created by +** the sqlite3_compile() routine. The integer returned is an SQLITE_ +** success/failure code that describes the result of executing the virtual +** machine. +** +** This routine sets the error code and string returned by +** sqlite3_errcode(), sqlite3_errmsg() and sqlite3_errmsg16(). +*/ +SQLITE_API int sqlite3_finalize(sqlite3_stmt *pStmt){ + int rc; + if( pStmt==0 ){ + /* IMPLEMENTATION-OF: R-57228-12904 Invoking sqlite3_finalize() on a NULL + ** pointer is a harmless no-op. */ + rc = SQLITE_OK; + }else{ + Vdbe *v = (Vdbe*)pStmt; + sqlite3 *db = v->db; + if( vdbeSafety(v) ) return SQLITE_MISUSE_BKPT; + sqlite3_mutex_enter(db->mutex); + checkProfileCallback(db, v); + rc = sqlite3VdbeFinalize(v); + rc = sqlite3ApiExit(db, rc); + sqlite3LeaveMutexAndCloseZombie(db); + } + return rc; +} + +/* +** Terminate the current execution of an SQL statement and reset it +** back to its starting state so that it can be reused. A success code from +** the prior execution is returned. +** +** This routine sets the error code and string returned by +** sqlite3_errcode(), sqlite3_errmsg() and sqlite3_errmsg16(). +*/ +SQLITE_API int sqlite3_reset(sqlite3_stmt *pStmt){ + int rc; + if( pStmt==0 ){ + rc = SQLITE_OK; + }else{ + Vdbe *v = (Vdbe*)pStmt; + sqlite3 *db = v->db; + sqlite3_mutex_enter(db->mutex); + checkProfileCallback(db, v); + rc = sqlite3VdbeReset(v); + sqlite3VdbeRewind(v); + assert( (rc & (db->errMask))==rc ); + rc = sqlite3ApiExit(db, rc); + sqlite3_mutex_leave(db->mutex); + } + return rc; +} + +/* +** Set all the parameters in the compiled SQL statement to NULL. +*/ +SQLITE_API int sqlite3_clear_bindings(sqlite3_stmt *pStmt){ + int i; + int rc = SQLITE_OK; + Vdbe *p = (Vdbe*)pStmt; +#if SQLITE_THREADSAFE + sqlite3_mutex *mutex = ((Vdbe*)pStmt)->db->mutex; +#endif + sqlite3_mutex_enter(mutex); + for(i=0; inVar; i++){ + sqlite3VdbeMemRelease(&p->aVar[i]); + p->aVar[i].flags = MEM_Null; + } + assert( (p->prepFlags & SQLITE_PREPARE_SAVESQL)!=0 || p->expmask==0 ); + if( p->expmask ){ + p->expired = 1; + } + sqlite3_mutex_leave(mutex); + return rc; +} + + +/**************************** sqlite3_value_ ******************************* +** The following routines extract information from a Mem or sqlite3_value +** structure. +*/ +SQLITE_API const void *sqlite3_value_blob(sqlite3_value *pVal){ + Mem *p = (Mem*)pVal; + if( p->flags & (MEM_Blob|MEM_Str) ){ + if( ExpandBlob(p)!=SQLITE_OK ){ + assert( p->flags==MEM_Null && p->z==0 ); + return 0; + } + p->flags |= MEM_Blob; + return p->n ? p->z : 0; + }else{ + return sqlite3_value_text(pVal); + } +} +SQLITE_API int sqlite3_value_bytes(sqlite3_value *pVal){ + return sqlite3ValueBytes(pVal, SQLITE_UTF8); +} +SQLITE_API int sqlite3_value_bytes16(sqlite3_value *pVal){ + return sqlite3ValueBytes(pVal, SQLITE_UTF16NATIVE); +} +SQLITE_API double sqlite3_value_double(sqlite3_value *pVal){ + return sqlite3VdbeRealValue((Mem*)pVal); +} +SQLITE_API int sqlite3_value_int(sqlite3_value *pVal){ + return (int)sqlite3VdbeIntValue((Mem*)pVal); +} +SQLITE_API sqlite_int64 sqlite3_value_int64(sqlite3_value *pVal){ + return sqlite3VdbeIntValue((Mem*)pVal); +} +SQLITE_API unsigned int sqlite3_value_subtype(sqlite3_value *pVal){ + Mem *pMem = (Mem*)pVal; + return ((pMem->flags & MEM_Subtype) ? pMem->eSubtype : 0); +} +SQLITE_API void *sqlite3_value_pointer(sqlite3_value *pVal, const char *zPType){ + Mem *p = (Mem*)pVal; + if( (p->flags&(MEM_TypeMask|MEM_Term|MEM_Subtype)) == + (MEM_Null|MEM_Term|MEM_Subtype) + && zPType!=0 + && p->eSubtype=='p' + && strcmp(p->u.zPType, zPType)==0 + ){ + return (void*)p->z; + }else{ + return 0; + } +} +SQLITE_API const unsigned char *sqlite3_value_text(sqlite3_value *pVal){ + return (const unsigned char *)sqlite3ValueText(pVal, SQLITE_UTF8); +} +#ifndef SQLITE_OMIT_UTF16 +SQLITE_API const void *sqlite3_value_text16(sqlite3_value* pVal){ + return sqlite3ValueText(pVal, SQLITE_UTF16NATIVE); +} +SQLITE_API const void *sqlite3_value_text16be(sqlite3_value *pVal){ + return sqlite3ValueText(pVal, SQLITE_UTF16BE); +} +SQLITE_API const void *sqlite3_value_text16le(sqlite3_value *pVal){ + return sqlite3ValueText(pVal, SQLITE_UTF16LE); +} +#endif /* SQLITE_OMIT_UTF16 */ +/* EVIDENCE-OF: R-12793-43283 Every value in SQLite has one of five +** fundamental datatypes: 64-bit signed integer 64-bit IEEE floating +** point number string BLOB NULL +*/ +SQLITE_API int sqlite3_value_type(sqlite3_value* pVal){ + static const u8 aType[] = { + SQLITE_BLOB, /* 0x00 (not possible) */ + SQLITE_NULL, /* 0x01 NULL */ + SQLITE_TEXT, /* 0x02 TEXT */ + SQLITE_NULL, /* 0x03 (not possible) */ + SQLITE_INTEGER, /* 0x04 INTEGER */ + SQLITE_NULL, /* 0x05 (not possible) */ + SQLITE_INTEGER, /* 0x06 INTEGER + TEXT */ + SQLITE_NULL, /* 0x07 (not possible) */ + SQLITE_FLOAT, /* 0x08 FLOAT */ + SQLITE_NULL, /* 0x09 (not possible) */ + SQLITE_FLOAT, /* 0x0a FLOAT + TEXT */ + SQLITE_NULL, /* 0x0b (not possible) */ + SQLITE_INTEGER, /* 0x0c (not possible) */ + SQLITE_NULL, /* 0x0d (not possible) */ + SQLITE_INTEGER, /* 0x0e (not possible) */ + SQLITE_NULL, /* 0x0f (not possible) */ + SQLITE_BLOB, /* 0x10 BLOB */ + SQLITE_NULL, /* 0x11 (not possible) */ + SQLITE_TEXT, /* 0x12 (not possible) */ + SQLITE_NULL, /* 0x13 (not possible) */ + SQLITE_INTEGER, /* 0x14 INTEGER + BLOB */ + SQLITE_NULL, /* 0x15 (not possible) */ + SQLITE_INTEGER, /* 0x16 (not possible) */ + SQLITE_NULL, /* 0x17 (not possible) */ + SQLITE_FLOAT, /* 0x18 FLOAT + BLOB */ + SQLITE_NULL, /* 0x19 (not possible) */ + SQLITE_FLOAT, /* 0x1a (not possible) */ + SQLITE_NULL, /* 0x1b (not possible) */ + SQLITE_INTEGER, /* 0x1c (not possible) */ + SQLITE_NULL, /* 0x1d (not possible) */ + SQLITE_INTEGER, /* 0x1e (not possible) */ + SQLITE_NULL, /* 0x1f (not possible) */ + SQLITE_FLOAT, /* 0x20 INTREAL */ + SQLITE_NULL, /* 0x21 (not possible) */ + SQLITE_TEXT, /* 0x22 INTREAL + TEXT */ + SQLITE_NULL, /* 0x23 (not possible) */ + SQLITE_FLOAT, /* 0x24 (not possible) */ + SQLITE_NULL, /* 0x25 (not possible) */ + SQLITE_FLOAT, /* 0x26 (not possible) */ + SQLITE_NULL, /* 0x27 (not possible) */ + SQLITE_FLOAT, /* 0x28 (not possible) */ + SQLITE_NULL, /* 0x29 (not possible) */ + SQLITE_FLOAT, /* 0x2a (not possible) */ + SQLITE_NULL, /* 0x2b (not possible) */ + SQLITE_FLOAT, /* 0x2c (not possible) */ + SQLITE_NULL, /* 0x2d (not possible) */ + SQLITE_FLOAT, /* 0x2e (not possible) */ + SQLITE_NULL, /* 0x2f (not possible) */ + SQLITE_BLOB, /* 0x30 (not possible) */ + SQLITE_NULL, /* 0x31 (not possible) */ + SQLITE_TEXT, /* 0x32 (not possible) */ + SQLITE_NULL, /* 0x33 (not possible) */ + SQLITE_FLOAT, /* 0x34 (not possible) */ + SQLITE_NULL, /* 0x35 (not possible) */ + SQLITE_FLOAT, /* 0x36 (not possible) */ + SQLITE_NULL, /* 0x37 (not possible) */ + SQLITE_FLOAT, /* 0x38 (not possible) */ + SQLITE_NULL, /* 0x39 (not possible) */ + SQLITE_FLOAT, /* 0x3a (not possible) */ + SQLITE_NULL, /* 0x3b (not possible) */ + SQLITE_FLOAT, /* 0x3c (not possible) */ + SQLITE_NULL, /* 0x3d (not possible) */ + SQLITE_FLOAT, /* 0x3e (not possible) */ + SQLITE_NULL, /* 0x3f (not possible) */ + }; +#ifdef SQLITE_DEBUG + { + int eType = SQLITE_BLOB; + if( pVal->flags & MEM_Null ){ + eType = SQLITE_NULL; + }else if( pVal->flags & (MEM_Real|MEM_IntReal) ){ + eType = SQLITE_FLOAT; + }else if( pVal->flags & MEM_Int ){ + eType = SQLITE_INTEGER; + }else if( pVal->flags & MEM_Str ){ + eType = SQLITE_TEXT; + } + assert( eType == aType[pVal->flags&MEM_AffMask] ); + } +#endif + return aType[pVal->flags&MEM_AffMask]; +} + +/* Return true if a parameter to xUpdate represents an unchanged column */ +SQLITE_API int sqlite3_value_nochange(sqlite3_value *pVal){ + return (pVal->flags&(MEM_Null|MEM_Zero))==(MEM_Null|MEM_Zero); +} + +/* Return true if a parameter value originated from an sqlite3_bind() */ +SQLITE_API int sqlite3_value_frombind(sqlite3_value *pVal){ + return (pVal->flags&MEM_FromBind)!=0; +} + +/* Make a copy of an sqlite3_value object +*/ +SQLITE_API sqlite3_value *sqlite3_value_dup(const sqlite3_value *pOrig){ + sqlite3_value *pNew; + if( pOrig==0 ) return 0; + pNew = sqlite3_malloc( sizeof(*pNew) ); + if( pNew==0 ) return 0; + memset(pNew, 0, sizeof(*pNew)); + memcpy(pNew, pOrig, MEMCELLSIZE); + pNew->flags &= ~MEM_Dyn; + pNew->db = 0; + if( pNew->flags&(MEM_Str|MEM_Blob) ){ + pNew->flags &= ~(MEM_Static|MEM_Dyn); + pNew->flags |= MEM_Ephem; + if( sqlite3VdbeMemMakeWriteable(pNew)!=SQLITE_OK ){ + sqlite3ValueFree(pNew); + pNew = 0; + } + } + return pNew; +} + +/* Destroy an sqlite3_value object previously obtained from +** sqlite3_value_dup(). +*/ +SQLITE_API void sqlite3_value_free(sqlite3_value *pOld){ + sqlite3ValueFree(pOld); +} + + +/**************************** sqlite3_result_ ******************************* +** The following routines are used by user-defined functions to specify +** the function result. +** +** The setStrOrError() function calls sqlite3VdbeMemSetStr() to store the +** result as a string or blob but if the string or blob is too large, it +** then sets the error code to SQLITE_TOOBIG +** +** The invokeValueDestructor(P,X) routine invokes destructor function X() +** on value P is not going to be used and need to be destroyed. +*/ +static void setResultStrOrError( + sqlite3_context *pCtx, /* Function context */ + const char *z, /* String pointer */ + int n, /* Bytes in string, or negative */ + u8 enc, /* Encoding of z. 0 for BLOBs */ + void (*xDel)(void*) /* Destructor function */ +){ + if( sqlite3VdbeMemSetStr(pCtx->pOut, z, n, enc, xDel)==SQLITE_TOOBIG ){ + sqlite3_result_error_toobig(pCtx); + } +} +static int invokeValueDestructor( + const void *p, /* Value to destroy */ + void (*xDel)(void*), /* The destructor */ + sqlite3_context *pCtx /* Set a SQLITE_TOOBIG error if no NULL */ +){ + assert( xDel!=SQLITE_DYNAMIC ); + if( xDel==0 ){ + /* noop */ + }else if( xDel==SQLITE_TRANSIENT ){ + /* noop */ + }else{ + xDel((void*)p); + } + if( pCtx ) sqlite3_result_error_toobig(pCtx); + return SQLITE_TOOBIG; +} +SQLITE_API void sqlite3_result_blob( + sqlite3_context *pCtx, + const void *z, + int n, + void (*xDel)(void *) +){ + assert( n>=0 ); + assert( sqlite3_mutex_held(pCtx->pOut->db->mutex) ); + setResultStrOrError(pCtx, z, n, 0, xDel); +} +SQLITE_API void sqlite3_result_blob64( + sqlite3_context *pCtx, + const void *z, + sqlite3_uint64 n, + void (*xDel)(void *) +){ + assert( sqlite3_mutex_held(pCtx->pOut->db->mutex) ); + assert( xDel!=SQLITE_DYNAMIC ); + if( n>0x7fffffff ){ + (void)invokeValueDestructor(z, xDel, pCtx); + }else{ + setResultStrOrError(pCtx, z, (int)n, 0, xDel); + } +} +SQLITE_API void sqlite3_result_double(sqlite3_context *pCtx, double rVal){ + assert( sqlite3_mutex_held(pCtx->pOut->db->mutex) ); + sqlite3VdbeMemSetDouble(pCtx->pOut, rVal); +} +SQLITE_API void sqlite3_result_error(sqlite3_context *pCtx, const char *z, int n){ + assert( sqlite3_mutex_held(pCtx->pOut->db->mutex) ); + pCtx->isError = SQLITE_ERROR; + sqlite3VdbeMemSetStr(pCtx->pOut, z, n, SQLITE_UTF8, SQLITE_TRANSIENT); +} +#ifndef SQLITE_OMIT_UTF16 +SQLITE_API void sqlite3_result_error16(sqlite3_context *pCtx, const void *z, int n){ + assert( sqlite3_mutex_held(pCtx->pOut->db->mutex) ); + pCtx->isError = SQLITE_ERROR; + sqlite3VdbeMemSetStr(pCtx->pOut, z, n, SQLITE_UTF16NATIVE, SQLITE_TRANSIENT); +} +#endif +SQLITE_API void sqlite3_result_int(sqlite3_context *pCtx, int iVal){ + assert( sqlite3_mutex_held(pCtx->pOut->db->mutex) ); + sqlite3VdbeMemSetInt64(pCtx->pOut, (i64)iVal); +} +SQLITE_API void sqlite3_result_int64(sqlite3_context *pCtx, i64 iVal){ + assert( sqlite3_mutex_held(pCtx->pOut->db->mutex) ); + sqlite3VdbeMemSetInt64(pCtx->pOut, iVal); +} +SQLITE_API void sqlite3_result_null(sqlite3_context *pCtx){ + assert( sqlite3_mutex_held(pCtx->pOut->db->mutex) ); + sqlite3VdbeMemSetNull(pCtx->pOut); +} +SQLITE_API void sqlite3_result_pointer( + sqlite3_context *pCtx, + void *pPtr, + const char *zPType, + void (*xDestructor)(void*) +){ + Mem *pOut = pCtx->pOut; + assert( sqlite3_mutex_held(pOut->db->mutex) ); + sqlite3VdbeMemRelease(pOut); + pOut->flags = MEM_Null; + sqlite3VdbeMemSetPointer(pOut, pPtr, zPType, xDestructor); +} +SQLITE_API void sqlite3_result_subtype(sqlite3_context *pCtx, unsigned int eSubtype){ + Mem *pOut = pCtx->pOut; + assert( sqlite3_mutex_held(pOut->db->mutex) ); + pOut->eSubtype = eSubtype & 0xff; + pOut->flags |= MEM_Subtype; +} +SQLITE_API void sqlite3_result_text( + sqlite3_context *pCtx, + const char *z, + int n, + void (*xDel)(void *) +){ + assert( sqlite3_mutex_held(pCtx->pOut->db->mutex) ); + setResultStrOrError(pCtx, z, n, SQLITE_UTF8, xDel); +} +SQLITE_API void sqlite3_result_text64( + sqlite3_context *pCtx, + const char *z, + sqlite3_uint64 n, + void (*xDel)(void *), + unsigned char enc +){ + assert( sqlite3_mutex_held(pCtx->pOut->db->mutex) ); + assert( xDel!=SQLITE_DYNAMIC ); + if( enc==SQLITE_UTF16 ) enc = SQLITE_UTF16NATIVE; + if( n>0x7fffffff ){ + (void)invokeValueDestructor(z, xDel, pCtx); + }else{ + setResultStrOrError(pCtx, z, (int)n, enc, xDel); + } +} +#ifndef SQLITE_OMIT_UTF16 +SQLITE_API void sqlite3_result_text16( + sqlite3_context *pCtx, + const void *z, + int n, + void (*xDel)(void *) +){ + assert( sqlite3_mutex_held(pCtx->pOut->db->mutex) ); + setResultStrOrError(pCtx, z, n, SQLITE_UTF16NATIVE, xDel); +} +SQLITE_API void sqlite3_result_text16be( + sqlite3_context *pCtx, + const void *z, + int n, + void (*xDel)(void *) +){ + assert( sqlite3_mutex_held(pCtx->pOut->db->mutex) ); + setResultStrOrError(pCtx, z, n, SQLITE_UTF16BE, xDel); +} +SQLITE_API void sqlite3_result_text16le( + sqlite3_context *pCtx, + const void *z, + int n, + void (*xDel)(void *) +){ + assert( sqlite3_mutex_held(pCtx->pOut->db->mutex) ); + setResultStrOrError(pCtx, z, n, SQLITE_UTF16LE, xDel); +} +#endif /* SQLITE_OMIT_UTF16 */ +SQLITE_API void sqlite3_result_value(sqlite3_context *pCtx, sqlite3_value *pValue){ + assert( sqlite3_mutex_held(pCtx->pOut->db->mutex) ); + sqlite3VdbeMemCopy(pCtx->pOut, pValue); +} +SQLITE_API void sqlite3_result_zeroblob(sqlite3_context *pCtx, int n){ + assert( sqlite3_mutex_held(pCtx->pOut->db->mutex) ); + sqlite3VdbeMemSetZeroBlob(pCtx->pOut, n); +} +SQLITE_API int sqlite3_result_zeroblob64(sqlite3_context *pCtx, u64 n){ + Mem *pOut = pCtx->pOut; + assert( sqlite3_mutex_held(pOut->db->mutex) ); + if( n>(u64)pOut->db->aLimit[SQLITE_LIMIT_LENGTH] ){ + return SQLITE_TOOBIG; + } + sqlite3VdbeMemSetZeroBlob(pCtx->pOut, (int)n); + return SQLITE_OK; +} +SQLITE_API void sqlite3_result_error_code(sqlite3_context *pCtx, int errCode){ + pCtx->isError = errCode ? errCode : -1; +#ifdef SQLITE_DEBUG + if( pCtx->pVdbe ) pCtx->pVdbe->rcApp = errCode; +#endif + if( pCtx->pOut->flags & MEM_Null ){ + sqlite3VdbeMemSetStr(pCtx->pOut, sqlite3ErrStr(errCode), -1, + SQLITE_UTF8, SQLITE_STATIC); + } +} + +/* Force an SQLITE_TOOBIG error. */ +SQLITE_API void sqlite3_result_error_toobig(sqlite3_context *pCtx){ + assert( sqlite3_mutex_held(pCtx->pOut->db->mutex) ); + pCtx->isError = SQLITE_TOOBIG; + sqlite3VdbeMemSetStr(pCtx->pOut, "string or blob too big", -1, + SQLITE_UTF8, SQLITE_STATIC); +} + +/* An SQLITE_NOMEM error. */ +SQLITE_API void sqlite3_result_error_nomem(sqlite3_context *pCtx){ + assert( sqlite3_mutex_held(pCtx->pOut->db->mutex) ); + sqlite3VdbeMemSetNull(pCtx->pOut); + pCtx->isError = SQLITE_NOMEM_BKPT; + sqlite3OomFault(pCtx->pOut->db); +} + +#ifndef SQLITE_UNTESTABLE +/* Force the INT64 value currently stored as the result to be +** a MEM_IntReal value. See the SQLITE_TESTCTRL_RESULT_INTREAL +** test-control. +*/ +SQLITE_PRIVATE void sqlite3ResultIntReal(sqlite3_context *pCtx){ + assert( sqlite3_mutex_held(pCtx->pOut->db->mutex) ); + if( pCtx->pOut->flags & MEM_Int ){ + pCtx->pOut->flags &= ~MEM_Int; + pCtx->pOut->flags |= MEM_IntReal; + } +} +#endif + + +/* +** This function is called after a transaction has been committed. It +** invokes callbacks registered with sqlite3_wal_hook() as required. +*/ +static int doWalCallbacks(sqlite3 *db){ + int rc = SQLITE_OK; +#ifndef SQLITE_OMIT_WAL + int i; + for(i=0; inDb; i++){ + Btree *pBt = db->aDb[i].pBt; + if( pBt ){ + int nEntry; + sqlite3BtreeEnter(pBt); + nEntry = sqlite3PagerWalCallback(sqlite3BtreePager(pBt)); + sqlite3BtreeLeave(pBt); + if( nEntry>0 && db->xWalCallback && rc==SQLITE_OK ){ + rc = db->xWalCallback(db->pWalArg, db, db->aDb[i].zDbSName, nEntry); + } + } + } +#endif + return rc; +} + + +/* +** Execute the statement pStmt, either until a row of data is ready, the +** statement is completely executed or an error occurs. +** +** This routine implements the bulk of the logic behind the sqlite_step() +** API. The only thing omitted is the automatic recompile if a +** schema change has occurred. That detail is handled by the +** outer sqlite3_step() wrapper procedure. +*/ +static int sqlite3Step(Vdbe *p){ + sqlite3 *db; + int rc; + + assert(p); + if( p->magic!=VDBE_MAGIC_RUN ){ + /* We used to require that sqlite3_reset() be called before retrying + ** sqlite3_step() after any error or after SQLITE_DONE. But beginning + ** with version 3.7.0, we changed this so that sqlite3_reset() would + ** be called automatically instead of throwing the SQLITE_MISUSE error. + ** This "automatic-reset" change is not technically an incompatibility, + ** since any application that receives an SQLITE_MISUSE is broken by + ** definition. + ** + ** Nevertheless, some published applications that were originally written + ** for version 3.6.23 or earlier do in fact depend on SQLITE_MISUSE + ** returns, and those were broken by the automatic-reset change. As a + ** a work-around, the SQLITE_OMIT_AUTORESET compile-time restores the + ** legacy behavior of returning SQLITE_MISUSE for cases where the + ** previous sqlite3_step() returned something other than a SQLITE_LOCKED + ** or SQLITE_BUSY error. + */ +#ifdef SQLITE_OMIT_AUTORESET + if( (rc = p->rc&0xff)==SQLITE_BUSY || rc==SQLITE_LOCKED ){ + sqlite3_reset((sqlite3_stmt*)p); + }else{ + return SQLITE_MISUSE_BKPT; + } +#else + sqlite3_reset((sqlite3_stmt*)p); +#endif + } + + /* Check that malloc() has not failed. If it has, return early. */ + db = p->db; + if( db->mallocFailed ){ + p->rc = SQLITE_NOMEM; + return SQLITE_NOMEM_BKPT; + } + + if( p->pc<0 && p->expired ){ + p->rc = SQLITE_SCHEMA; + rc = SQLITE_ERROR; + goto end_of_step; + } + if( p->pc<0 ){ + /* If there are no other statements currently running, then + ** reset the interrupt flag. This prevents a call to sqlite3_interrupt + ** from interrupting a statement that has not yet started. + */ + if( db->nVdbeActive==0 ){ + db->u1.isInterrupted = 0; + } + + assert( db->nVdbeWrite>0 || db->autoCommit==0 + || (db->nDeferredCons==0 && db->nDeferredImmCons==0) + ); + +#ifndef SQLITE_OMIT_TRACE + if( (db->mTrace & (SQLITE_TRACE_PROFILE|SQLITE_TRACE_XPROFILE))!=0 + && !db->init.busy && p->zSql ){ + sqlite3OsCurrentTimeInt64(db->pVfs, &p->startTime); + }else{ + assert( p->startTime==0 ); + } +#endif + + db->nVdbeActive++; + if( p->readOnly==0 ) db->nVdbeWrite++; + if( p->bIsReader ) db->nVdbeRead++; + p->pc = 0; + } +#ifdef SQLITE_DEBUG + p->rcApp = SQLITE_OK; +#endif +#ifndef SQLITE_OMIT_EXPLAIN + if( p->explain ){ + rc = sqlite3VdbeList(p); + }else +#endif /* SQLITE_OMIT_EXPLAIN */ + { + db->nVdbeExec++; + rc = sqlite3VdbeExec(p); + db->nVdbeExec--; + } + + if( rc!=SQLITE_ROW ){ +#ifndef SQLITE_OMIT_TRACE + /* If the statement completed successfully, invoke the profile callback */ + checkProfileCallback(db, p); +#endif + + if( rc==SQLITE_DONE && db->autoCommit ){ + assert( p->rc==SQLITE_OK ); + p->rc = doWalCallbacks(db); + if( p->rc!=SQLITE_OK ){ + rc = SQLITE_ERROR; + } + } + } + + db->errCode = rc; + if( SQLITE_NOMEM==sqlite3ApiExit(p->db, p->rc) ){ + p->rc = SQLITE_NOMEM_BKPT; + } +end_of_step: + /* At this point local variable rc holds the value that should be + ** returned if this statement was compiled using the legacy + ** sqlite3_prepare() interface. According to the docs, this can only + ** be one of the values in the first assert() below. Variable p->rc + ** contains the value that would be returned if sqlite3_finalize() + ** were called on statement p. + */ + assert( rc==SQLITE_ROW || rc==SQLITE_DONE || rc==SQLITE_ERROR + || (rc&0xff)==SQLITE_BUSY || rc==SQLITE_MISUSE + ); + assert( (p->rc!=SQLITE_ROW && p->rc!=SQLITE_DONE) || p->rc==p->rcApp ); + if( rc!=SQLITE_ROW + && rc!=SQLITE_DONE + && (p->prepFlags & SQLITE_PREPARE_SAVESQL)!=0 + ){ + /* If this statement was prepared using saved SQL and an + ** error has occurred, then return the error code in p->rc to the + ** caller. Set the error code in the database handle to the same value. + */ + rc = sqlite3VdbeTransferError(p); + } + return (rc&db->errMask); +} + +/* +** This is the top-level implementation of sqlite3_step(). Call +** sqlite3Step() to do most of the work. If a schema error occurs, +** call sqlite3Reprepare() and try again. +*/ +SQLITE_API int sqlite3_step(sqlite3_stmt *pStmt){ + int rc = SQLITE_OK; /* Result from sqlite3Step() */ + Vdbe *v = (Vdbe*)pStmt; /* the prepared statement */ + int cnt = 0; /* Counter to prevent infinite loop of reprepares */ + sqlite3 *db; /* The database connection */ + + if( vdbeSafetyNotNull(v) ){ + return SQLITE_MISUSE_BKPT; + } + db = v->db; + sqlite3_mutex_enter(db->mutex); + v->doingRerun = 0; + while( (rc = sqlite3Step(v))==SQLITE_SCHEMA + && cnt++ < SQLITE_MAX_SCHEMA_RETRY ){ + int savedPc = v->pc; + rc = sqlite3Reprepare(v); + if( rc!=SQLITE_OK ){ + /* This case occurs after failing to recompile an sql statement. + ** The error message from the SQL compiler has already been loaded + ** into the database handle. This block copies the error message + ** from the database handle into the statement and sets the statement + ** program counter to 0 to ensure that when the statement is + ** finalized or reset the parser error message is available via + ** sqlite3_errmsg() and sqlite3_errcode(). + */ + const char *zErr = (const char *)sqlite3_value_text(db->pErr); + sqlite3DbFree(db, v->zErrMsg); + if( !db->mallocFailed ){ + v->zErrMsg = sqlite3DbStrDup(db, zErr); + v->rc = rc = sqlite3ApiExit(db, rc); + } else { + v->zErrMsg = 0; + v->rc = rc = SQLITE_NOMEM_BKPT; + } + break; + } + sqlite3_reset(pStmt); + if( savedPc>=0 ) v->doingRerun = 1; + assert( v->expired==0 ); + } + sqlite3_mutex_leave(db->mutex); + return rc; +} + + +/* +** Extract the user data from a sqlite3_context structure and return a +** pointer to it. +*/ +SQLITE_API void *sqlite3_user_data(sqlite3_context *p){ + assert( p && p->pFunc ); + return p->pFunc->pUserData; +} + +/* +** Extract the user data from a sqlite3_context structure and return a +** pointer to it. +** +** IMPLEMENTATION-OF: R-46798-50301 The sqlite3_context_db_handle() interface +** returns a copy of the pointer to the database connection (the 1st +** parameter) of the sqlite3_create_function() and +** sqlite3_create_function16() routines that originally registered the +** application defined function. +*/ +SQLITE_API sqlite3 *sqlite3_context_db_handle(sqlite3_context *p){ + assert( p && p->pOut ); + return p->pOut->db; +} + +/* +** If this routine is invoked from within an xColumn method of a virtual +** table, then it returns true if and only if the the call is during an +** UPDATE operation and the value of the column will not be modified +** by the UPDATE. +** +** If this routine is called from any context other than within the +** xColumn method of a virtual table, then the return value is meaningless +** and arbitrary. +** +** Virtual table implements might use this routine to optimize their +** performance by substituting a NULL result, or some other light-weight +** value, as a signal to the xUpdate routine that the column is unchanged. +*/ +SQLITE_API int sqlite3_vtab_nochange(sqlite3_context *p){ + assert( p ); + return sqlite3_value_nochange(p->pOut); +} + +/* +** Return the current time for a statement. If the current time +** is requested more than once within the same run of a single prepared +** statement, the exact same time is returned for each invocation regardless +** of the amount of time that elapses between invocations. In other words, +** the time returned is always the time of the first call. +*/ +SQLITE_PRIVATE sqlite3_int64 sqlite3StmtCurrentTime(sqlite3_context *p){ + int rc; +#ifndef SQLITE_ENABLE_STAT4 + sqlite3_int64 *piTime = &p->pVdbe->iCurrentTime; + assert( p->pVdbe!=0 ); +#else + sqlite3_int64 iTime = 0; + sqlite3_int64 *piTime = p->pVdbe!=0 ? &p->pVdbe->iCurrentTime : &iTime; +#endif + if( *piTime==0 ){ + rc = sqlite3OsCurrentTimeInt64(p->pOut->db->pVfs, piTime); + if( rc ) *piTime = 0; + } + return *piTime; +} + +/* +** Create a new aggregate context for p and return a pointer to +** its pMem->z element. +*/ +static SQLITE_NOINLINE void *createAggContext(sqlite3_context *p, int nByte){ + Mem *pMem = p->pMem; + assert( (pMem->flags & MEM_Agg)==0 ); + if( nByte<=0 ){ + sqlite3VdbeMemSetNull(pMem); + pMem->z = 0; + }else{ + sqlite3VdbeMemClearAndResize(pMem, nByte); + pMem->flags = MEM_Agg; + pMem->u.pDef = p->pFunc; + if( pMem->z ){ + memset(pMem->z, 0, nByte); + } + } + return (void*)pMem->z; +} + +/* +** Allocate or return the aggregate context for a user function. A new +** context is allocated on the first call. Subsequent calls return the +** same context that was returned on prior calls. +*/ +SQLITE_API void *sqlite3_aggregate_context(sqlite3_context *p, int nByte){ + assert( p && p->pFunc && p->pFunc->xFinalize ); + assert( sqlite3_mutex_held(p->pOut->db->mutex) ); + testcase( nByte<0 ); + if( (p->pMem->flags & MEM_Agg)==0 ){ + return createAggContext(p, nByte); + }else{ + return (void*)p->pMem->z; + } +} + +/* +** Return the auxiliary data pointer, if any, for the iArg'th argument to +** the user-function defined by pCtx. +** +** The left-most argument is 0. +** +** Undocumented behavior: If iArg is negative then access a cache of +** auxiliary data pointers that is available to all functions within a +** single prepared statement. The iArg values must match. +*/ +SQLITE_API void *sqlite3_get_auxdata(sqlite3_context *pCtx, int iArg){ + AuxData *pAuxData; + + assert( sqlite3_mutex_held(pCtx->pOut->db->mutex) ); +#if SQLITE_ENABLE_STAT4 + if( pCtx->pVdbe==0 ) return 0; +#else + assert( pCtx->pVdbe!=0 ); +#endif + for(pAuxData=pCtx->pVdbe->pAuxData; pAuxData; pAuxData=pAuxData->pNextAux){ + if( pAuxData->iAuxArg==iArg && (pAuxData->iAuxOp==pCtx->iOp || iArg<0) ){ + return pAuxData->pAux; + } + } + return 0; +} + +/* +** Set the auxiliary data pointer and delete function, for the iArg'th +** argument to the user-function defined by pCtx. Any previous value is +** deleted by calling the delete function specified when it was set. +** +** The left-most argument is 0. +** +** Undocumented behavior: If iArg is negative then make the data available +** to all functions within the current prepared statement using iArg as an +** access code. +*/ +SQLITE_API void sqlite3_set_auxdata( + sqlite3_context *pCtx, + int iArg, + void *pAux, + void (*xDelete)(void*) +){ + AuxData *pAuxData; + Vdbe *pVdbe = pCtx->pVdbe; + + assert( sqlite3_mutex_held(pCtx->pOut->db->mutex) ); +#ifdef SQLITE_ENABLE_STAT4 + if( pVdbe==0 ) goto failed; +#else + assert( pVdbe!=0 ); +#endif + + for(pAuxData=pVdbe->pAuxData; pAuxData; pAuxData=pAuxData->pNextAux){ + if( pAuxData->iAuxArg==iArg && (pAuxData->iAuxOp==pCtx->iOp || iArg<0) ){ + break; + } + } + if( pAuxData==0 ){ + pAuxData = sqlite3DbMallocZero(pVdbe->db, sizeof(AuxData)); + if( !pAuxData ) goto failed; + pAuxData->iAuxOp = pCtx->iOp; + pAuxData->iAuxArg = iArg; + pAuxData->pNextAux = pVdbe->pAuxData; + pVdbe->pAuxData = pAuxData; + if( pCtx->isError==0 ) pCtx->isError = -1; + }else if( pAuxData->xDeleteAux ){ + pAuxData->xDeleteAux(pAuxData->pAux); + } + + pAuxData->pAux = pAux; + pAuxData->xDeleteAux = xDelete; + return; + +failed: + if( xDelete ){ + xDelete(pAux); + } +} + +#ifndef SQLITE_OMIT_DEPRECATED +/* +** Return the number of times the Step function of an aggregate has been +** called. +** +** This function is deprecated. Do not use it for new code. It is +** provide only to avoid breaking legacy code. New aggregate function +** implementations should keep their own counts within their aggregate +** context. +*/ +SQLITE_API int sqlite3_aggregate_count(sqlite3_context *p){ + assert( p && p->pMem && p->pFunc && p->pFunc->xFinalize ); + return p->pMem->n; +} +#endif + +/* +** Return the number of columns in the result set for the statement pStmt. +*/ +SQLITE_API int sqlite3_column_count(sqlite3_stmt *pStmt){ + Vdbe *pVm = (Vdbe *)pStmt; + return pVm ? pVm->nResColumn : 0; +} + +/* +** Return the number of values available from the current row of the +** currently executing statement pStmt. +*/ +SQLITE_API int sqlite3_data_count(sqlite3_stmt *pStmt){ + Vdbe *pVm = (Vdbe *)pStmt; + if( pVm==0 || pVm->pResultSet==0 ) return 0; + return pVm->nResColumn; +} + +/* +** Return a pointer to static memory containing an SQL NULL value. +*/ +static const Mem *columnNullValue(void){ + /* Even though the Mem structure contains an element + ** of type i64, on certain architectures (x86) with certain compiler + ** switches (-Os), gcc may align this Mem object on a 4-byte boundary + ** instead of an 8-byte one. This all works fine, except that when + ** running with SQLITE_DEBUG defined the SQLite code sometimes assert()s + ** that a Mem structure is located on an 8-byte boundary. To prevent + ** these assert()s from failing, when building with SQLITE_DEBUG defined + ** using gcc, we force nullMem to be 8-byte aligned using the magical + ** __attribute__((aligned(8))) macro. */ + static const Mem nullMem +#if defined(SQLITE_DEBUG) && defined(__GNUC__) + __attribute__((aligned(8))) +#endif + = { + /* .u = */ {0}, + /* .flags = */ (u16)MEM_Null, + /* .enc = */ (u8)0, + /* .eSubtype = */ (u8)0, + /* .n = */ (int)0, + /* .z = */ (char*)0, + /* .zMalloc = */ (char*)0, + /* .szMalloc = */ (int)0, + /* .uTemp = */ (u32)0, + /* .db = */ (sqlite3*)0, + /* .xDel = */ (void(*)(void*))0, +#ifdef SQLITE_DEBUG + /* .pScopyFrom = */ (Mem*)0, + /* .mScopyFlags= */ 0, +#endif + }; + return &nullMem; +} + +/* +** Check to see if column iCol of the given statement is valid. If +** it is, return a pointer to the Mem for the value of that column. +** If iCol is not valid, return a pointer to a Mem which has a value +** of NULL. +*/ +static Mem *columnMem(sqlite3_stmt *pStmt, int i){ + Vdbe *pVm; + Mem *pOut; + + pVm = (Vdbe *)pStmt; + if( pVm==0 ) return (Mem*)columnNullValue(); + assert( pVm->db ); + sqlite3_mutex_enter(pVm->db->mutex); + if( pVm->pResultSet!=0 && inResColumn && i>=0 ){ + pOut = &pVm->pResultSet[i]; + }else{ + sqlite3Error(pVm->db, SQLITE_RANGE); + pOut = (Mem*)columnNullValue(); + } + return pOut; +} + +/* +** This function is called after invoking an sqlite3_value_XXX function on a +** column value (i.e. a value returned by evaluating an SQL expression in the +** select list of a SELECT statement) that may cause a malloc() failure. If +** malloc() has failed, the threads mallocFailed flag is cleared and the result +** code of statement pStmt set to SQLITE_NOMEM. +** +** Specifically, this is called from within: +** +** sqlite3_column_int() +** sqlite3_column_int64() +** sqlite3_column_text() +** sqlite3_column_text16() +** sqlite3_column_real() +** sqlite3_column_bytes() +** sqlite3_column_bytes16() +** sqiite3_column_blob() +*/ +static void columnMallocFailure(sqlite3_stmt *pStmt) +{ + /* If malloc() failed during an encoding conversion within an + ** sqlite3_column_XXX API, then set the return code of the statement to + ** SQLITE_NOMEM. The next call to _step() (if any) will return SQLITE_ERROR + ** and _finalize() will return NOMEM. + */ + Vdbe *p = (Vdbe *)pStmt; + if( p ){ + assert( p->db!=0 ); + assert( sqlite3_mutex_held(p->db->mutex) ); + p->rc = sqlite3ApiExit(p->db, p->rc); + sqlite3_mutex_leave(p->db->mutex); + } +} + +/**************************** sqlite3_column_ ******************************* +** The following routines are used to access elements of the current row +** in the result set. +*/ +SQLITE_API const void *sqlite3_column_blob(sqlite3_stmt *pStmt, int i){ + const void *val; + val = sqlite3_value_blob( columnMem(pStmt,i) ); + /* Even though there is no encoding conversion, value_blob() might + ** need to call malloc() to expand the result of a zeroblob() + ** expression. + */ + columnMallocFailure(pStmt); + return val; +} +SQLITE_API int sqlite3_column_bytes(sqlite3_stmt *pStmt, int i){ + int val = sqlite3_value_bytes( columnMem(pStmt,i) ); + columnMallocFailure(pStmt); + return val; +} +SQLITE_API int sqlite3_column_bytes16(sqlite3_stmt *pStmt, int i){ + int val = sqlite3_value_bytes16( columnMem(pStmt,i) ); + columnMallocFailure(pStmt); + return val; +} +SQLITE_API double sqlite3_column_double(sqlite3_stmt *pStmt, int i){ + double val = sqlite3_value_double( columnMem(pStmt,i) ); + columnMallocFailure(pStmt); + return val; +} +SQLITE_API int sqlite3_column_int(sqlite3_stmt *pStmt, int i){ + int val = sqlite3_value_int( columnMem(pStmt,i) ); + columnMallocFailure(pStmt); + return val; +} +SQLITE_API sqlite_int64 sqlite3_column_int64(sqlite3_stmt *pStmt, int i){ + sqlite_int64 val = sqlite3_value_int64( columnMem(pStmt,i) ); + columnMallocFailure(pStmt); + return val; +} +SQLITE_API const unsigned char *sqlite3_column_text(sqlite3_stmt *pStmt, int i){ + const unsigned char *val = sqlite3_value_text( columnMem(pStmt,i) ); + columnMallocFailure(pStmt); + return val; +} +SQLITE_API sqlite3_value *sqlite3_column_value(sqlite3_stmt *pStmt, int i){ + Mem *pOut = columnMem(pStmt, i); + if( pOut->flags&MEM_Static ){ + pOut->flags &= ~MEM_Static; + pOut->flags |= MEM_Ephem; + } + columnMallocFailure(pStmt); + return (sqlite3_value *)pOut; +} +#ifndef SQLITE_OMIT_UTF16 +SQLITE_API const void *sqlite3_column_text16(sqlite3_stmt *pStmt, int i){ + const void *val = sqlite3_value_text16( columnMem(pStmt,i) ); + columnMallocFailure(pStmt); + return val; +} +#endif /* SQLITE_OMIT_UTF16 */ +SQLITE_API int sqlite3_column_type(sqlite3_stmt *pStmt, int i){ + int iType = sqlite3_value_type( columnMem(pStmt,i) ); + columnMallocFailure(pStmt); + return iType; +} + +/* +** Convert the N-th element of pStmt->pColName[] into a string using +** xFunc() then return that string. If N is out of range, return 0. +** +** There are up to 5 names for each column. useType determines which +** name is returned. Here are the names: +** +** 0 The column name as it should be displayed for output +** 1 The datatype name for the column +** 2 The name of the database that the column derives from +** 3 The name of the table that the column derives from +** 4 The name of the table column that the result column derives from +** +** If the result is not a simple column reference (if it is an expression +** or a constant) then useTypes 2, 3, and 4 return NULL. +*/ +static const void *columnName( + sqlite3_stmt *pStmt, /* The statement */ + int N, /* Which column to get the name for */ + int useUtf16, /* True to return the name as UTF16 */ + int useType /* What type of name */ +){ + const void *ret; + Vdbe *p; + int n; + sqlite3 *db; +#ifdef SQLITE_ENABLE_API_ARMOR + if( pStmt==0 ){ + (void)SQLITE_MISUSE_BKPT; + return 0; + } +#endif + ret = 0; + p = (Vdbe *)pStmt; + db = p->db; + assert( db!=0 ); + n = sqlite3_column_count(pStmt); + if( N=0 ){ + N += useType*n; + sqlite3_mutex_enter(db->mutex); + assert( db->mallocFailed==0 ); +#ifndef SQLITE_OMIT_UTF16 + if( useUtf16 ){ + ret = sqlite3_value_text16((sqlite3_value*)&p->aColName[N]); + }else +#endif + { + ret = sqlite3_value_text((sqlite3_value*)&p->aColName[N]); + } + /* A malloc may have failed inside of the _text() call. If this + ** is the case, clear the mallocFailed flag and return NULL. + */ + if( db->mallocFailed ){ + sqlite3OomClear(db); + ret = 0; + } + sqlite3_mutex_leave(db->mutex); + } + return ret; +} + +/* +** Return the name of the Nth column of the result set returned by SQL +** statement pStmt. +*/ +SQLITE_API const char *sqlite3_column_name(sqlite3_stmt *pStmt, int N){ + return columnName(pStmt, N, 0, COLNAME_NAME); +} +#ifndef SQLITE_OMIT_UTF16 +SQLITE_API const void *sqlite3_column_name16(sqlite3_stmt *pStmt, int N){ + return columnName(pStmt, N, 1, COLNAME_NAME); +} +#endif + +/* +** Constraint: If you have ENABLE_COLUMN_METADATA then you must +** not define OMIT_DECLTYPE. +*/ +#if defined(SQLITE_OMIT_DECLTYPE) && defined(SQLITE_ENABLE_COLUMN_METADATA) +# error "Must not define both SQLITE_OMIT_DECLTYPE \ + and SQLITE_ENABLE_COLUMN_METADATA" +#endif + +#ifndef SQLITE_OMIT_DECLTYPE +/* +** Return the column declaration type (if applicable) of the 'i'th column +** of the result set of SQL statement pStmt. +*/ +SQLITE_API const char *sqlite3_column_decltype(sqlite3_stmt *pStmt, int N){ + return columnName(pStmt, N, 0, COLNAME_DECLTYPE); +} +#ifndef SQLITE_OMIT_UTF16 +SQLITE_API const void *sqlite3_column_decltype16(sqlite3_stmt *pStmt, int N){ + return columnName(pStmt, N, 1, COLNAME_DECLTYPE); +} +#endif /* SQLITE_OMIT_UTF16 */ +#endif /* SQLITE_OMIT_DECLTYPE */ + +#ifdef SQLITE_ENABLE_COLUMN_METADATA +/* +** Return the name of the database from which a result column derives. +** NULL is returned if the result column is an expression or constant or +** anything else which is not an unambiguous reference to a database column. +*/ +SQLITE_API const char *sqlite3_column_database_name(sqlite3_stmt *pStmt, int N){ + return columnName(pStmt, N, 0, COLNAME_DATABASE); +} +#ifndef SQLITE_OMIT_UTF16 +SQLITE_API const void *sqlite3_column_database_name16(sqlite3_stmt *pStmt, int N){ + return columnName(pStmt, N, 1, COLNAME_DATABASE); +} +#endif /* SQLITE_OMIT_UTF16 */ + +/* +** Return the name of the table from which a result column derives. +** NULL is returned if the result column is an expression or constant or +** anything else which is not an unambiguous reference to a database column. +*/ +SQLITE_API const char *sqlite3_column_table_name(sqlite3_stmt *pStmt, int N){ + return columnName(pStmt, N, 0, COLNAME_TABLE); +} +#ifndef SQLITE_OMIT_UTF16 +SQLITE_API const void *sqlite3_column_table_name16(sqlite3_stmt *pStmt, int N){ + return columnName(pStmt, N, 1, COLNAME_TABLE); +} +#endif /* SQLITE_OMIT_UTF16 */ + +/* +** Return the name of the table column from which a result column derives. +** NULL is returned if the result column is an expression or constant or +** anything else which is not an unambiguous reference to a database column. +*/ +SQLITE_API const char *sqlite3_column_origin_name(sqlite3_stmt *pStmt, int N){ + return columnName(pStmt, N, 0, COLNAME_COLUMN); +} +#ifndef SQLITE_OMIT_UTF16 +SQLITE_API const void *sqlite3_column_origin_name16(sqlite3_stmt *pStmt, int N){ + return columnName(pStmt, N, 1, COLNAME_COLUMN); +} +#endif /* SQLITE_OMIT_UTF16 */ +#endif /* SQLITE_ENABLE_COLUMN_METADATA */ + + +/******************************* sqlite3_bind_ *************************** +** +** Routines used to attach values to wildcards in a compiled SQL statement. +*/ +/* +** Unbind the value bound to variable i in virtual machine p. This is the +** the same as binding a NULL value to the column. If the "i" parameter is +** out of range, then SQLITE_RANGE is returned. Othewise SQLITE_OK. +** +** A successful evaluation of this routine acquires the mutex on p. +** the mutex is released if any kind of error occurs. +** +** The error code stored in database p->db is overwritten with the return +** value in any case. +*/ +static int vdbeUnbind(Vdbe *p, int i){ + Mem *pVar; + if( vdbeSafetyNotNull(p) ){ + return SQLITE_MISUSE_BKPT; + } + sqlite3_mutex_enter(p->db->mutex); + if( p->magic!=VDBE_MAGIC_RUN || p->pc>=0 ){ + sqlite3Error(p->db, SQLITE_MISUSE); + sqlite3_mutex_leave(p->db->mutex); + sqlite3_log(SQLITE_MISUSE, + "bind on a busy prepared statement: [%s]", p->zSql); + return SQLITE_MISUSE_BKPT; + } + if( i<1 || i>p->nVar ){ + sqlite3Error(p->db, SQLITE_RANGE); + sqlite3_mutex_leave(p->db->mutex); + return SQLITE_RANGE; + } + i--; + pVar = &p->aVar[i]; + sqlite3VdbeMemRelease(pVar); + pVar->flags = MEM_Null; + p->db->errCode = SQLITE_OK; + + /* If the bit corresponding to this variable in Vdbe.expmask is set, then + ** binding a new value to this variable invalidates the current query plan. + ** + ** IMPLEMENTATION-OF: R-48440-37595 If the specific value bound to host + ** parameter in the WHERE clause might influence the choice of query plan + ** for a statement, then the statement will be automatically recompiled, + ** as if there had been a schema change, on the first sqlite3_step() call + ** following any change to the bindings of that parameter. + */ + assert( (p->prepFlags & SQLITE_PREPARE_SAVESQL)!=0 || p->expmask==0 ); + if( p->expmask!=0 && (p->expmask & (i>=31 ? 0x80000000 : (u32)1<expired = 1; + } + return SQLITE_OK; +} + +/* +** Bind a text or BLOB value. +*/ +static int bindText( + sqlite3_stmt *pStmt, /* The statement to bind against */ + int i, /* Index of the parameter to bind */ + const void *zData, /* Pointer to the data to be bound */ + int nData, /* Number of bytes of data to be bound */ + void (*xDel)(void*), /* Destructor for the data */ + u8 encoding /* Encoding for the data */ +){ + Vdbe *p = (Vdbe *)pStmt; + Mem *pVar; + int rc; + + rc = vdbeUnbind(p, i); + if( rc==SQLITE_OK ){ + if( zData!=0 ){ + pVar = &p->aVar[i-1]; + rc = sqlite3VdbeMemSetStr(pVar, zData, nData, encoding, xDel); + if( rc==SQLITE_OK && encoding!=0 ){ + rc = sqlite3VdbeChangeEncoding(pVar, ENC(p->db)); + } + if( rc ){ + sqlite3Error(p->db, rc); + rc = sqlite3ApiExit(p->db, rc); + } + } + sqlite3_mutex_leave(p->db->mutex); + }else if( xDel!=SQLITE_STATIC && xDel!=SQLITE_TRANSIENT ){ + xDel((void*)zData); + } + return rc; +} + + +/* +** Bind a blob value to an SQL statement variable. +*/ +SQLITE_API int sqlite3_bind_blob( + sqlite3_stmt *pStmt, + int i, + const void *zData, + int nData, + void (*xDel)(void*) +){ +#ifdef SQLITE_ENABLE_API_ARMOR + if( nData<0 ) return SQLITE_MISUSE_BKPT; +#endif + return bindText(pStmt, i, zData, nData, xDel, 0); +} +SQLITE_API int sqlite3_bind_blob64( + sqlite3_stmt *pStmt, + int i, + const void *zData, + sqlite3_uint64 nData, + void (*xDel)(void*) +){ + assert( xDel!=SQLITE_DYNAMIC ); + if( nData>0x7fffffff ){ + return invokeValueDestructor(zData, xDel, 0); + }else{ + return bindText(pStmt, i, zData, (int)nData, xDel, 0); + } +} +SQLITE_API int sqlite3_bind_double(sqlite3_stmt *pStmt, int i, double rValue){ + int rc; + Vdbe *p = (Vdbe *)pStmt; + rc = vdbeUnbind(p, i); + if( rc==SQLITE_OK ){ + sqlite3VdbeMemSetDouble(&p->aVar[i-1], rValue); + sqlite3_mutex_leave(p->db->mutex); + } + return rc; +} +SQLITE_API int sqlite3_bind_int(sqlite3_stmt *p, int i, int iValue){ + return sqlite3_bind_int64(p, i, (i64)iValue); +} +SQLITE_API int sqlite3_bind_int64(sqlite3_stmt *pStmt, int i, sqlite_int64 iValue){ + int rc; + Vdbe *p = (Vdbe *)pStmt; + rc = vdbeUnbind(p, i); + if( rc==SQLITE_OK ){ + sqlite3VdbeMemSetInt64(&p->aVar[i-1], iValue); + sqlite3_mutex_leave(p->db->mutex); + } + return rc; +} +SQLITE_API int sqlite3_bind_null(sqlite3_stmt *pStmt, int i){ + int rc; + Vdbe *p = (Vdbe*)pStmt; + rc = vdbeUnbind(p, i); + if( rc==SQLITE_OK ){ + sqlite3_mutex_leave(p->db->mutex); + } + return rc; +} +SQLITE_API int sqlite3_bind_pointer( + sqlite3_stmt *pStmt, + int i, + void *pPtr, + const char *zPTtype, + void (*xDestructor)(void*) +){ + int rc; + Vdbe *p = (Vdbe*)pStmt; + rc = vdbeUnbind(p, i); + if( rc==SQLITE_OK ){ + sqlite3VdbeMemSetPointer(&p->aVar[i-1], pPtr, zPTtype, xDestructor); + sqlite3_mutex_leave(p->db->mutex); + }else if( xDestructor ){ + xDestructor(pPtr); + } + return rc; +} +SQLITE_API int sqlite3_bind_text( + sqlite3_stmt *pStmt, + int i, + const char *zData, + int nData, + void (*xDel)(void*) +){ + return bindText(pStmt, i, zData, nData, xDel, SQLITE_UTF8); +} +SQLITE_API int sqlite3_bind_text64( + sqlite3_stmt *pStmt, + int i, + const char *zData, + sqlite3_uint64 nData, + void (*xDel)(void*), + unsigned char enc +){ + assert( xDel!=SQLITE_DYNAMIC ); + if( nData>0x7fffffff ){ + return invokeValueDestructor(zData, xDel, 0); + }else{ + if( enc==SQLITE_UTF16 ) enc = SQLITE_UTF16NATIVE; + return bindText(pStmt, i, zData, (int)nData, xDel, enc); + } +} +#ifndef SQLITE_OMIT_UTF16 +SQLITE_API int sqlite3_bind_text16( + sqlite3_stmt *pStmt, + int i, + const void *zData, + int nData, + void (*xDel)(void*) +){ + return bindText(pStmt, i, zData, nData, xDel, SQLITE_UTF16NATIVE); +} +#endif /* SQLITE_OMIT_UTF16 */ +SQLITE_API int sqlite3_bind_value(sqlite3_stmt *pStmt, int i, const sqlite3_value *pValue){ + int rc; + switch( sqlite3_value_type((sqlite3_value*)pValue) ){ + case SQLITE_INTEGER: { + rc = sqlite3_bind_int64(pStmt, i, pValue->u.i); + break; + } + case SQLITE_FLOAT: { + rc = sqlite3_bind_double(pStmt, i, pValue->u.r); + break; + } + case SQLITE_BLOB: { + if( pValue->flags & MEM_Zero ){ + rc = sqlite3_bind_zeroblob(pStmt, i, pValue->u.nZero); + }else{ + rc = sqlite3_bind_blob(pStmt, i, pValue->z, pValue->n,SQLITE_TRANSIENT); + } + break; + } + case SQLITE_TEXT: { + rc = bindText(pStmt,i, pValue->z, pValue->n, SQLITE_TRANSIENT, + pValue->enc); + break; + } + default: { + rc = sqlite3_bind_null(pStmt, i); + break; + } + } + return rc; +} +SQLITE_API int sqlite3_bind_zeroblob(sqlite3_stmt *pStmt, int i, int n){ + int rc; + Vdbe *p = (Vdbe *)pStmt; + rc = vdbeUnbind(p, i); + if( rc==SQLITE_OK ){ + sqlite3VdbeMemSetZeroBlob(&p->aVar[i-1], n); + sqlite3_mutex_leave(p->db->mutex); + } + return rc; +} +SQLITE_API int sqlite3_bind_zeroblob64(sqlite3_stmt *pStmt, int i, sqlite3_uint64 n){ + int rc; + Vdbe *p = (Vdbe *)pStmt; + sqlite3_mutex_enter(p->db->mutex); + if( n>(u64)p->db->aLimit[SQLITE_LIMIT_LENGTH] ){ + rc = SQLITE_TOOBIG; + }else{ + assert( (n & 0x7FFFFFFF)==n ); + rc = sqlite3_bind_zeroblob(pStmt, i, n); + } + rc = sqlite3ApiExit(p->db, rc); + sqlite3_mutex_leave(p->db->mutex); + return rc; +} + +/* +** Return the number of wildcards that can be potentially bound to. +** This routine is added to support DBD::SQLite. +*/ +SQLITE_API int sqlite3_bind_parameter_count(sqlite3_stmt *pStmt){ + Vdbe *p = (Vdbe*)pStmt; + return p ? p->nVar : 0; +} + +/* +** Return the name of a wildcard parameter. Return NULL if the index +** is out of range or if the wildcard is unnamed. +** +** The result is always UTF-8. +*/ +SQLITE_API const char *sqlite3_bind_parameter_name(sqlite3_stmt *pStmt, int i){ + Vdbe *p = (Vdbe*)pStmt; + if( p==0 ) return 0; + return sqlite3VListNumToName(p->pVList, i); +} + +/* +** Given a wildcard parameter name, return the index of the variable +** with that name. If there is no variable with the given name, +** return 0. +*/ +SQLITE_PRIVATE int sqlite3VdbeParameterIndex(Vdbe *p, const char *zName, int nName){ + if( p==0 || zName==0 ) return 0; + return sqlite3VListNameToNum(p->pVList, zName, nName); +} +SQLITE_API int sqlite3_bind_parameter_index(sqlite3_stmt *pStmt, const char *zName){ + return sqlite3VdbeParameterIndex((Vdbe*)pStmt, zName, sqlite3Strlen30(zName)); +} + +/* +** Transfer all bindings from the first statement over to the second. +*/ +SQLITE_PRIVATE int sqlite3TransferBindings(sqlite3_stmt *pFromStmt, sqlite3_stmt *pToStmt){ + Vdbe *pFrom = (Vdbe*)pFromStmt; + Vdbe *pTo = (Vdbe*)pToStmt; + int i; + assert( pTo->db==pFrom->db ); + assert( pTo->nVar==pFrom->nVar ); + sqlite3_mutex_enter(pTo->db->mutex); + for(i=0; inVar; i++){ + sqlite3VdbeMemMove(&pTo->aVar[i], &pFrom->aVar[i]); + } + sqlite3_mutex_leave(pTo->db->mutex); + return SQLITE_OK; +} + +#ifndef SQLITE_OMIT_DEPRECATED +/* +** Deprecated external interface. Internal/core SQLite code +** should call sqlite3TransferBindings. +** +** It is misuse to call this routine with statements from different +** database connections. But as this is a deprecated interface, we +** will not bother to check for that condition. +** +** If the two statements contain a different number of bindings, then +** an SQLITE_ERROR is returned. Nothing else can go wrong, so otherwise +** SQLITE_OK is returned. +*/ +SQLITE_API int sqlite3_transfer_bindings(sqlite3_stmt *pFromStmt, sqlite3_stmt *pToStmt){ + Vdbe *pFrom = (Vdbe*)pFromStmt; + Vdbe *pTo = (Vdbe*)pToStmt; + if( pFrom->nVar!=pTo->nVar ){ + return SQLITE_ERROR; + } + assert( (pTo->prepFlags & SQLITE_PREPARE_SAVESQL)!=0 || pTo->expmask==0 ); + if( pTo->expmask ){ + pTo->expired = 1; + } + assert( (pFrom->prepFlags & SQLITE_PREPARE_SAVESQL)!=0 || pFrom->expmask==0 ); + if( pFrom->expmask ){ + pFrom->expired = 1; + } + return sqlite3TransferBindings(pFromStmt, pToStmt); +} +#endif + +/* +** Return the sqlite3* database handle to which the prepared statement given +** in the argument belongs. This is the same database handle that was +** the first argument to the sqlite3_prepare() that was used to create +** the statement in the first place. +*/ +SQLITE_API sqlite3 *sqlite3_db_handle(sqlite3_stmt *pStmt){ + return pStmt ? ((Vdbe*)pStmt)->db : 0; +} + +/* +** Return true if the prepared statement is guaranteed to not modify the +** database. +*/ +SQLITE_API int sqlite3_stmt_readonly(sqlite3_stmt *pStmt){ + return pStmt ? ((Vdbe*)pStmt)->readOnly : 1; +} + +/* +** Return 1 if the statement is an EXPLAIN and return 2 if the +** statement is an EXPLAIN QUERY PLAN +*/ +SQLITE_API int sqlite3_stmt_isexplain(sqlite3_stmt *pStmt){ + return pStmt ? ((Vdbe*)pStmt)->explain : 0; +} + +/* +** Return true if the prepared statement is in need of being reset. +*/ +SQLITE_API int sqlite3_stmt_busy(sqlite3_stmt *pStmt){ + Vdbe *v = (Vdbe*)pStmt; + return v!=0 && v->magic==VDBE_MAGIC_RUN && v->pc>=0; +} + +/* +** Return a pointer to the next prepared statement after pStmt associated +** with database connection pDb. If pStmt is NULL, return the first +** prepared statement for the database connection. Return NULL if there +** are no more. +*/ +SQLITE_API sqlite3_stmt *sqlite3_next_stmt(sqlite3 *pDb, sqlite3_stmt *pStmt){ + sqlite3_stmt *pNext; +#ifdef SQLITE_ENABLE_API_ARMOR + if( !sqlite3SafetyCheckOk(pDb) ){ + (void)SQLITE_MISUSE_BKPT; + return 0; + } +#endif + sqlite3_mutex_enter(pDb->mutex); + if( pStmt==0 ){ + pNext = (sqlite3_stmt*)pDb->pVdbe; + }else{ + pNext = (sqlite3_stmt*)((Vdbe*)pStmt)->pNext; + } + sqlite3_mutex_leave(pDb->mutex); + return pNext; +} + +/* +** Return the value of a status counter for a prepared statement +*/ +SQLITE_API int sqlite3_stmt_status(sqlite3_stmt *pStmt, int op, int resetFlag){ + Vdbe *pVdbe = (Vdbe*)pStmt; + u32 v; +#ifdef SQLITE_ENABLE_API_ARMOR + if( !pStmt + || (op!=SQLITE_STMTSTATUS_MEMUSED && (op<0||op>=ArraySize(pVdbe->aCounter))) + ){ + (void)SQLITE_MISUSE_BKPT; + return 0; + } +#endif + if( op==SQLITE_STMTSTATUS_MEMUSED ){ + sqlite3 *db = pVdbe->db; + sqlite3_mutex_enter(db->mutex); + v = 0; + db->pnBytesFreed = (int*)&v; + sqlite3VdbeClearObject(db, pVdbe); + sqlite3DbFree(db, pVdbe); + db->pnBytesFreed = 0; + sqlite3_mutex_leave(db->mutex); + }else{ + v = pVdbe->aCounter[op]; + if( resetFlag ) pVdbe->aCounter[op] = 0; + } + return (int)v; +} + +/* +** Return the SQL associated with a prepared statement +*/ +SQLITE_API const char *sqlite3_sql(sqlite3_stmt *pStmt){ + Vdbe *p = (Vdbe *)pStmt; + return p ? p->zSql : 0; +} + +/* +** Return the SQL associated with a prepared statement with +** bound parameters expanded. Space to hold the returned string is +** obtained from sqlite3_malloc(). The caller is responsible for +** freeing the returned string by passing it to sqlite3_free(). +** +** The SQLITE_TRACE_SIZE_LIMIT puts an upper bound on the size of +** expanded bound parameters. +*/ +SQLITE_API char *sqlite3_expanded_sql(sqlite3_stmt *pStmt){ +#ifdef SQLITE_OMIT_TRACE + return 0; +#else + char *z = 0; + const char *zSql = sqlite3_sql(pStmt); + if( zSql ){ + Vdbe *p = (Vdbe *)pStmt; + sqlite3_mutex_enter(p->db->mutex); + z = sqlite3VdbeExpandSql(p, zSql); + sqlite3_mutex_leave(p->db->mutex); + } + return z; +#endif +} + +#ifdef SQLITE_ENABLE_NORMALIZE +/* +** Return the normalized SQL associated with a prepared statement. +*/ +SQLITE_API const char *sqlite3_normalized_sql(sqlite3_stmt *pStmt){ + Vdbe *p = (Vdbe *)pStmt; + if( p==0 ) return 0; + if( p->zNormSql==0 && ALWAYS(p->zSql!=0) ){ + sqlite3_mutex_enter(p->db->mutex); + p->zNormSql = sqlite3Normalize(p, p->zSql); + sqlite3_mutex_leave(p->db->mutex); + } + return p->zNormSql; +} +#endif /* SQLITE_ENABLE_NORMALIZE */ + +#ifdef SQLITE_ENABLE_PREUPDATE_HOOK +/* +** Allocate and populate an UnpackedRecord structure based on the serialized +** record in nKey/pKey. Return a pointer to the new UnpackedRecord structure +** if successful, or a NULL pointer if an OOM error is encountered. +*/ +static UnpackedRecord *vdbeUnpackRecord( + KeyInfo *pKeyInfo, + int nKey, + const void *pKey +){ + UnpackedRecord *pRet; /* Return value */ + + pRet = sqlite3VdbeAllocUnpackedRecord(pKeyInfo); + if( pRet ){ + memset(pRet->aMem, 0, sizeof(Mem)*(pKeyInfo->nKeyField+1)); + sqlite3VdbeRecordUnpack(pKeyInfo, nKey, pKey, pRet); + } + return pRet; +} + +/* +** This function is called from within a pre-update callback to retrieve +** a field of the row currently being updated or deleted. +*/ +SQLITE_API int sqlite3_preupdate_old(sqlite3 *db, int iIdx, sqlite3_value **ppValue){ + PreUpdate *p = db->pPreUpdate; + Mem *pMem; + int rc = SQLITE_OK; + + /* Test that this call is being made from within an SQLITE_DELETE or + ** SQLITE_UPDATE pre-update callback, and that iIdx is within range. */ + if( !p || p->op==SQLITE_INSERT ){ + rc = SQLITE_MISUSE_BKPT; + goto preupdate_old_out; + } + if( p->pPk ){ + iIdx = sqlite3ColumnOfIndex(p->pPk, iIdx); + } + if( iIdx>=p->pCsr->nField || iIdx<0 ){ + rc = SQLITE_RANGE; + goto preupdate_old_out; + } + + /* If the old.* record has not yet been loaded into memory, do so now. */ + if( p->pUnpacked==0 ){ + u32 nRec; + u8 *aRec; + + nRec = sqlite3BtreePayloadSize(p->pCsr->uc.pCursor); + aRec = sqlite3DbMallocRaw(db, nRec); + if( !aRec ) goto preupdate_old_out; + rc = sqlite3BtreePayload(p->pCsr->uc.pCursor, 0, nRec, aRec); + if( rc==SQLITE_OK ){ + p->pUnpacked = vdbeUnpackRecord(&p->keyinfo, nRec, aRec); + if( !p->pUnpacked ) rc = SQLITE_NOMEM; + } + if( rc!=SQLITE_OK ){ + sqlite3DbFree(db, aRec); + goto preupdate_old_out; + } + p->aRecord = aRec; + } + + pMem = *ppValue = &p->pUnpacked->aMem[iIdx]; + if( iIdx==p->pTab->iPKey ){ + sqlite3VdbeMemSetInt64(pMem, p->iKey1); + }else if( iIdx>=p->pUnpacked->nField ){ + *ppValue = (sqlite3_value *)columnNullValue(); + }else if( p->pTab->aCol[iIdx].affinity==SQLITE_AFF_REAL ){ + if( pMem->flags & (MEM_Int|MEM_IntReal) ){ + testcase( pMem->flags & MEM_Int ); + testcase( pMem->flags & MEM_IntReal ); + sqlite3VdbeMemRealify(pMem); + } + } + + preupdate_old_out: + sqlite3Error(db, rc); + return sqlite3ApiExit(db, rc); +} +#endif /* SQLITE_ENABLE_PREUPDATE_HOOK */ + +#ifdef SQLITE_ENABLE_PREUPDATE_HOOK +/* +** This function is called from within a pre-update callback to retrieve +** the number of columns in the row being updated, deleted or inserted. +*/ +SQLITE_API int sqlite3_preupdate_count(sqlite3 *db){ + PreUpdate *p = db->pPreUpdate; + return (p ? p->keyinfo.nKeyField : 0); +} +#endif /* SQLITE_ENABLE_PREUPDATE_HOOK */ + +#ifdef SQLITE_ENABLE_PREUPDATE_HOOK +/* +** This function is designed to be called from within a pre-update callback +** only. It returns zero if the change that caused the callback was made +** immediately by a user SQL statement. Or, if the change was made by a +** trigger program, it returns the number of trigger programs currently +** on the stack (1 for a top-level trigger, 2 for a trigger fired by a +** top-level trigger etc.). +** +** For the purposes of the previous paragraph, a foreign key CASCADE, SET NULL +** or SET DEFAULT action is considered a trigger. +*/ +SQLITE_API int sqlite3_preupdate_depth(sqlite3 *db){ + PreUpdate *p = db->pPreUpdate; + return (p ? p->v->nFrame : 0); +} +#endif /* SQLITE_ENABLE_PREUPDATE_HOOK */ + +#ifdef SQLITE_ENABLE_PREUPDATE_HOOK +/* +** This function is called from within a pre-update callback to retrieve +** a field of the row currently being updated or inserted. +*/ +SQLITE_API int sqlite3_preupdate_new(sqlite3 *db, int iIdx, sqlite3_value **ppValue){ + PreUpdate *p = db->pPreUpdate; + int rc = SQLITE_OK; + Mem *pMem; + + if( !p || p->op==SQLITE_DELETE ){ + rc = SQLITE_MISUSE_BKPT; + goto preupdate_new_out; + } + if( p->pPk && p->op!=SQLITE_UPDATE ){ + iIdx = sqlite3ColumnOfIndex(p->pPk, iIdx); + } + if( iIdx>=p->pCsr->nField || iIdx<0 ){ + rc = SQLITE_RANGE; + goto preupdate_new_out; + } + + if( p->op==SQLITE_INSERT ){ + /* For an INSERT, memory cell p->iNewReg contains the serialized record + ** that is being inserted. Deserialize it. */ + UnpackedRecord *pUnpack = p->pNewUnpacked; + if( !pUnpack ){ + Mem *pData = &p->v->aMem[p->iNewReg]; + rc = ExpandBlob(pData); + if( rc!=SQLITE_OK ) goto preupdate_new_out; + pUnpack = vdbeUnpackRecord(&p->keyinfo, pData->n, pData->z); + if( !pUnpack ){ + rc = SQLITE_NOMEM; + goto preupdate_new_out; + } + p->pNewUnpacked = pUnpack; + } + pMem = &pUnpack->aMem[iIdx]; + if( iIdx==p->pTab->iPKey ){ + sqlite3VdbeMemSetInt64(pMem, p->iKey2); + }else if( iIdx>=pUnpack->nField ){ + pMem = (sqlite3_value *)columnNullValue(); + } + }else{ + /* For an UPDATE, memory cell (p->iNewReg+1+iIdx) contains the required + ** value. Make a copy of the cell contents and return a pointer to it. + ** It is not safe to return a pointer to the memory cell itself as the + ** caller may modify the value text encoding. + */ + assert( p->op==SQLITE_UPDATE ); + if( !p->aNew ){ + p->aNew = (Mem *)sqlite3DbMallocZero(db, sizeof(Mem) * p->pCsr->nField); + if( !p->aNew ){ + rc = SQLITE_NOMEM; + goto preupdate_new_out; + } + } + assert( iIdx>=0 && iIdxpCsr->nField ); + pMem = &p->aNew[iIdx]; + if( pMem->flags==0 ){ + if( iIdx==p->pTab->iPKey ){ + sqlite3VdbeMemSetInt64(pMem, p->iKey2); + }else{ + rc = sqlite3VdbeMemCopy(pMem, &p->v->aMem[p->iNewReg+1+iIdx]); + if( rc!=SQLITE_OK ) goto preupdate_new_out; + } + } + } + *ppValue = pMem; + + preupdate_new_out: + sqlite3Error(db, rc); + return sqlite3ApiExit(db, rc); +} +#endif /* SQLITE_ENABLE_PREUPDATE_HOOK */ + +#ifdef SQLITE_ENABLE_STMT_SCANSTATUS +/* +** Return status data for a single loop within query pStmt. +*/ +SQLITE_API int sqlite3_stmt_scanstatus( + sqlite3_stmt *pStmt, /* Prepared statement being queried */ + int idx, /* Index of loop to report on */ + int iScanStatusOp, /* Which metric to return */ + void *pOut /* OUT: Write the answer here */ +){ + Vdbe *p = (Vdbe*)pStmt; + ScanStatus *pScan; + if( idx<0 || idx>=p->nScan ) return 1; + pScan = &p->aScan[idx]; + switch( iScanStatusOp ){ + case SQLITE_SCANSTAT_NLOOP: { + *(sqlite3_int64*)pOut = p->anExec[pScan->addrLoop]; + break; + } + case SQLITE_SCANSTAT_NVISIT: { + *(sqlite3_int64*)pOut = p->anExec[pScan->addrVisit]; + break; + } + case SQLITE_SCANSTAT_EST: { + double r = 1.0; + LogEst x = pScan->nEst; + while( x<100 ){ + x += 10; + r *= 0.5; + } + *(double*)pOut = r*sqlite3LogEstToInt(x); + break; + } + case SQLITE_SCANSTAT_NAME: { + *(const char**)pOut = pScan->zName; + break; + } + case SQLITE_SCANSTAT_EXPLAIN: { + if( pScan->addrExplain ){ + *(const char**)pOut = p->aOp[ pScan->addrExplain ].p4.z; + }else{ + *(const char**)pOut = 0; + } + break; + } + case SQLITE_SCANSTAT_SELECTID: { + if( pScan->addrExplain ){ + *(int*)pOut = p->aOp[ pScan->addrExplain ].p1; + }else{ + *(int*)pOut = -1; + } + break; + } + default: { + return 1; + } + } + return 0; +} + +/* +** Zero all counters associated with the sqlite3_stmt_scanstatus() data. +*/ +SQLITE_API void sqlite3_stmt_scanstatus_reset(sqlite3_stmt *pStmt){ + Vdbe *p = (Vdbe*)pStmt; + memset(p->anExec, 0, p->nOp * sizeof(i64)); +} +#endif /* SQLITE_ENABLE_STMT_SCANSTATUS */ + +/************** End of vdbeapi.c *********************************************/ +/************** Begin file vdbetrace.c ***************************************/ +/* +** 2009 November 25 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** +** This file contains code used to insert the values of host parameters +** (aka "wildcards") into the SQL text output by sqlite3_trace(). +** +** The Vdbe parse-tree explainer is also found here. +*/ +/* #include "sqliteInt.h" */ +/* #include "vdbeInt.h" */ + +#ifndef SQLITE_OMIT_TRACE + +/* +** zSql is a zero-terminated string of UTF-8 SQL text. Return the number of +** bytes in this text up to but excluding the first character in +** a host parameter. If the text contains no host parameters, return +** the total number of bytes in the text. +*/ +static int findNextHostParameter(const char *zSql, int *pnToken){ + int tokenType; + int nTotal = 0; + int n; + + *pnToken = 0; + while( zSql[0] ){ + n = sqlite3GetToken((u8*)zSql, &tokenType); + assert( n>0 && tokenType!=TK_ILLEGAL ); + if( tokenType==TK_VARIABLE ){ + *pnToken = n; + break; + } + nTotal += n; + zSql += n; + } + return nTotal; +} + +/* +** This function returns a pointer to a nul-terminated string in memory +** obtained from sqlite3DbMalloc(). If sqlite3.nVdbeExec is 1, then the +** string contains a copy of zRawSql but with host parameters expanded to +** their current bindings. Or, if sqlite3.nVdbeExec is greater than 1, +** then the returned string holds a copy of zRawSql with "-- " prepended +** to each line of text. +** +** If the SQLITE_TRACE_SIZE_LIMIT macro is defined to an integer, then +** then long strings and blobs are truncated to that many bytes. This +** can be used to prevent unreasonably large trace strings when dealing +** with large (multi-megabyte) strings and blobs. +** +** The calling function is responsible for making sure the memory returned +** is eventually freed. +** +** ALGORITHM: Scan the input string looking for host parameters in any of +** these forms: ?, ?N, $A, @A, :A. Take care to avoid text within +** string literals, quoted identifier names, and comments. For text forms, +** the host parameter index is found by scanning the prepared +** statement for the corresponding OP_Variable opcode. Once the host +** parameter index is known, locate the value in p->aVar[]. Then render +** the value as a literal in place of the host parameter name. +*/ +SQLITE_PRIVATE char *sqlite3VdbeExpandSql( + Vdbe *p, /* The prepared statement being evaluated */ + const char *zRawSql /* Raw text of the SQL statement */ +){ + sqlite3 *db; /* The database connection */ + int idx = 0; /* Index of a host parameter */ + int nextIndex = 1; /* Index of next ? host parameter */ + int n; /* Length of a token prefix */ + int nToken; /* Length of the parameter token */ + int i; /* Loop counter */ + Mem *pVar; /* Value of a host parameter */ + StrAccum out; /* Accumulate the output here */ +#ifndef SQLITE_OMIT_UTF16 + Mem utf8; /* Used to convert UTF16 into UTF8 for display */ +#endif + char zBase[100]; /* Initial working space */ + + db = p->db; + sqlite3StrAccumInit(&out, 0, zBase, sizeof(zBase), + db->aLimit[SQLITE_LIMIT_LENGTH]); + if( db->nVdbeExec>1 ){ + while( *zRawSql ){ + const char *zStart = zRawSql; + while( *(zRawSql++)!='\n' && *zRawSql ); + sqlite3_str_append(&out, "-- ", 3); + assert( (zRawSql - zStart) > 0 ); + sqlite3_str_append(&out, zStart, (int)(zRawSql-zStart)); + } + }else if( p->nVar==0 ){ + sqlite3_str_append(&out, zRawSql, sqlite3Strlen30(zRawSql)); + }else{ + while( zRawSql[0] ){ + n = findNextHostParameter(zRawSql, &nToken); + assert( n>0 ); + sqlite3_str_append(&out, zRawSql, n); + zRawSql += n; + assert( zRawSql[0] || nToken==0 ); + if( nToken==0 ) break; + if( zRawSql[0]=='?' ){ + if( nToken>1 ){ + assert( sqlite3Isdigit(zRawSql[1]) ); + sqlite3GetInt32(&zRawSql[1], &idx); + }else{ + idx = nextIndex; + } + }else{ + assert( zRawSql[0]==':' || zRawSql[0]=='$' || + zRawSql[0]=='@' || zRawSql[0]=='#' ); + testcase( zRawSql[0]==':' ); + testcase( zRawSql[0]=='$' ); + testcase( zRawSql[0]=='@' ); + testcase( zRawSql[0]=='#' ); + idx = sqlite3VdbeParameterIndex(p, zRawSql, nToken); + assert( idx>0 ); + } + zRawSql += nToken; + nextIndex = idx + 1; + assert( idx>0 && idx<=p->nVar ); + pVar = &p->aVar[idx-1]; + if( pVar->flags & MEM_Null ){ + sqlite3_str_append(&out, "NULL", 4); + }else if( pVar->flags & (MEM_Int|MEM_IntReal) ){ + sqlite3_str_appendf(&out, "%lld", pVar->u.i); + }else if( pVar->flags & MEM_Real ){ + sqlite3_str_appendf(&out, "%!.15g", pVar->u.r); + }else if( pVar->flags & MEM_Str ){ + int nOut; /* Number of bytes of the string text to include in output */ +#ifndef SQLITE_OMIT_UTF16 + u8 enc = ENC(db); + if( enc!=SQLITE_UTF8 ){ + memset(&utf8, 0, sizeof(utf8)); + utf8.db = db; + sqlite3VdbeMemSetStr(&utf8, pVar->z, pVar->n, enc, SQLITE_STATIC); + if( SQLITE_NOMEM==sqlite3VdbeChangeEncoding(&utf8, SQLITE_UTF8) ){ + out.accError = SQLITE_NOMEM; + out.nAlloc = 0; + } + pVar = &utf8; + } +#endif + nOut = pVar->n; +#ifdef SQLITE_TRACE_SIZE_LIMIT + if( nOut>SQLITE_TRACE_SIZE_LIMIT ){ + nOut = SQLITE_TRACE_SIZE_LIMIT; + while( nOutn && (pVar->z[nOut]&0xc0)==0x80 ){ nOut++; } + } +#endif + sqlite3_str_appendf(&out, "'%.*q'", nOut, pVar->z); +#ifdef SQLITE_TRACE_SIZE_LIMIT + if( nOutn ){ + sqlite3_str_appendf(&out, "/*+%d bytes*/", pVar->n-nOut); + } +#endif +#ifndef SQLITE_OMIT_UTF16 + if( enc!=SQLITE_UTF8 ) sqlite3VdbeMemRelease(&utf8); +#endif + }else if( pVar->flags & MEM_Zero ){ + sqlite3_str_appendf(&out, "zeroblob(%d)", pVar->u.nZero); + }else{ + int nOut; /* Number of bytes of the blob to include in output */ + assert( pVar->flags & MEM_Blob ); + sqlite3_str_append(&out, "x'", 2); + nOut = pVar->n; +#ifdef SQLITE_TRACE_SIZE_LIMIT + if( nOut>SQLITE_TRACE_SIZE_LIMIT ) nOut = SQLITE_TRACE_SIZE_LIMIT; +#endif + for(i=0; iz[i]&0xff); + } + sqlite3_str_append(&out, "'", 1); +#ifdef SQLITE_TRACE_SIZE_LIMIT + if( nOutn ){ + sqlite3_str_appendf(&out, "/*+%d bytes*/", pVar->n-nOut); + } +#endif + } + } + } + if( out.accError ) sqlite3_str_reset(&out); + return sqlite3StrAccumFinish(&out); +} + +#endif /* #ifndef SQLITE_OMIT_TRACE */ + +/************** End of vdbetrace.c *******************************************/ +/************** Begin file vdbe.c ********************************************/ +/* +** 2001 September 15 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** The code in this file implements the function that runs the +** bytecode of a prepared statement. +** +** Various scripts scan this source file in order to generate HTML +** documentation, headers files, or other derived files. The formatting +** of the code in this file is, therefore, important. See other comments +** in this file for details. If in doubt, do not deviate from existing +** commenting and indentation practices when changing or adding code. +*/ +/* #include "sqliteInt.h" */ +/* #include "vdbeInt.h" */ + +/* +** Invoke this macro on memory cells just prior to changing the +** value of the cell. This macro verifies that shallow copies are +** not misused. A shallow copy of a string or blob just copies a +** pointer to the string or blob, not the content. If the original +** is changed while the copy is still in use, the string or blob might +** be changed out from under the copy. This macro verifies that nothing +** like that ever happens. +*/ +#ifdef SQLITE_DEBUG +# define memAboutToChange(P,M) sqlite3VdbeMemAboutToChange(P,M) +#else +# define memAboutToChange(P,M) +#endif + +/* +** The following global variable is incremented every time a cursor +** moves, either by the OP_SeekXX, OP_Next, or OP_Prev opcodes. The test +** procedures use this information to make sure that indices are +** working correctly. This variable has no function other than to +** help verify the correct operation of the library. +*/ +#ifdef SQLITE_TEST +SQLITE_API int sqlite3_search_count = 0; +#endif + +/* +** When this global variable is positive, it gets decremented once before +** each instruction in the VDBE. When it reaches zero, the u1.isInterrupted +** field of the sqlite3 structure is set in order to simulate an interrupt. +** +** This facility is used for testing purposes only. It does not function +** in an ordinary build. +*/ +#ifdef SQLITE_TEST +SQLITE_API int sqlite3_interrupt_count = 0; +#endif + +/* +** The next global variable is incremented each type the OP_Sort opcode +** is executed. The test procedures use this information to make sure that +** sorting is occurring or not occurring at appropriate times. This variable +** has no function other than to help verify the correct operation of the +** library. +*/ +#ifdef SQLITE_TEST +SQLITE_API int sqlite3_sort_count = 0; +#endif + +/* +** The next global variable records the size of the largest MEM_Blob +** or MEM_Str that has been used by a VDBE opcode. The test procedures +** use this information to make sure that the zero-blob functionality +** is working correctly. This variable has no function other than to +** help verify the correct operation of the library. +*/ +#ifdef SQLITE_TEST +SQLITE_API int sqlite3_max_blobsize = 0; +static void updateMaxBlobsize(Mem *p){ + if( (p->flags & (MEM_Str|MEM_Blob))!=0 && p->n>sqlite3_max_blobsize ){ + sqlite3_max_blobsize = p->n; + } +} +#endif + +/* +** This macro evaluates to true if either the update hook or the preupdate +** hook are enabled for database connect DB. +*/ +#ifdef SQLITE_ENABLE_PREUPDATE_HOOK +# define HAS_UPDATE_HOOK(DB) ((DB)->xPreUpdateCallback||(DB)->xUpdateCallback) +#else +# define HAS_UPDATE_HOOK(DB) ((DB)->xUpdateCallback) +#endif + +/* +** The next global variable is incremented each time the OP_Found opcode +** is executed. This is used to test whether or not the foreign key +** operation implemented using OP_FkIsZero is working. This variable +** has no function other than to help verify the correct operation of the +** library. +*/ +#ifdef SQLITE_TEST +SQLITE_API int sqlite3_found_count = 0; +#endif + +/* +** Test a register to see if it exceeds the current maximum blob size. +** If it does, record the new maximum blob size. +*/ +#if defined(SQLITE_TEST) && !defined(SQLITE_UNTESTABLE) +# define UPDATE_MAX_BLOBSIZE(P) updateMaxBlobsize(P) +#else +# define UPDATE_MAX_BLOBSIZE(P) +#endif + +/* +** Invoke the VDBE coverage callback, if that callback is defined. This +** feature is used for test suite validation only and does not appear an +** production builds. +** +** M is the type of branch. I is the direction taken for this instance of +** the branch. +** +** M: 2 - two-way branch (I=0: fall-thru 1: jump ) +** 3 - two-way + NULL (I=0: fall-thru 1: jump 2: NULL ) +** 4 - OP_Jump (I=0: jump p1 1: jump p2 2: jump p3) +** +** In other words, if M is 2, then I is either 0 (for fall-through) or +** 1 (for when the branch is taken). If M is 3, the I is 0 for an +** ordinary fall-through, I is 1 if the branch was taken, and I is 2 +** if the result of comparison is NULL. For M=3, I=2 the jump may or +** may not be taken, depending on the SQLITE_JUMPIFNULL flags in p5. +** When M is 4, that means that an OP_Jump is being run. I is 0, 1, or 2 +** depending on if the operands are less than, equal, or greater than. +** +** iSrcLine is the source code line (from the __LINE__ macro) that +** generated the VDBE instruction combined with flag bits. The source +** code line number is in the lower 24 bits of iSrcLine and the upper +** 8 bytes are flags. The lower three bits of the flags indicate +** values for I that should never occur. For example, if the branch is +** always taken, the flags should be 0x05 since the fall-through and +** alternate branch are never taken. If a branch is never taken then +** flags should be 0x06 since only the fall-through approach is allowed. +** +** Bit 0x08 of the flags indicates an OP_Jump opcode that is only +** interested in equal or not-equal. In other words, I==0 and I==2 +** should be treated as equivalent +** +** Since only a line number is retained, not the filename, this macro +** only works for amalgamation builds. But that is ok, since these macros +** should be no-ops except for special builds used to measure test coverage. +*/ +#if !defined(SQLITE_VDBE_COVERAGE) +# define VdbeBranchTaken(I,M) +#else +# define VdbeBranchTaken(I,M) vdbeTakeBranch(pOp->iSrcLine,I,M) + static void vdbeTakeBranch(u32 iSrcLine, u8 I, u8 M){ + u8 mNever; + assert( I<=2 ); /* 0: fall through, 1: taken, 2: alternate taken */ + assert( M<=4 ); /* 2: two-way branch, 3: three-way branch, 4: OP_Jump */ + assert( I> 24; + assert( (I & mNever)==0 ); + if( sqlite3GlobalConfig.xVdbeBranch==0 ) return; /*NO_TEST*/ + /* Invoke the branch coverage callback with three arguments: + ** iSrcLine - the line number of the VdbeCoverage() macro, with + ** flags removed. + ** I - Mask of bits 0x07 indicating which cases are are + ** fulfilled by this instance of the jump. 0x01 means + ** fall-thru, 0x02 means taken, 0x04 means NULL. Any + ** impossible cases (ex: if the comparison is never NULL) + ** are filled in automatically so that the coverage + ** measurement logic does not flag those impossible cases + ** as missed coverage. + ** M - Type of jump. Same as M argument above + */ + I |= mNever; + if( M==2 ) I |= 0x04; + if( M==4 ){ + I |= 0x08; + if( (mNever&0x08)!=0 && (I&0x05)!=0) I |= 0x05; /*NO_TEST*/ + } + sqlite3GlobalConfig.xVdbeBranch(sqlite3GlobalConfig.pVdbeBranchArg, + iSrcLine&0xffffff, I, M); + } +#endif + +/* +** An ephemeral string value (signified by the MEM_Ephem flag) contains +** a pointer to a dynamically allocated string where some other entity +** is responsible for deallocating that string. Because the register +** does not control the string, it might be deleted without the register +** knowing it. +** +** This routine converts an ephemeral string into a dynamically allocated +** string that the register itself controls. In other words, it +** converts an MEM_Ephem string into a string with P.z==P.zMalloc. +*/ +#define Deephemeralize(P) \ + if( ((P)->flags&MEM_Ephem)!=0 \ + && sqlite3VdbeMemMakeWriteable(P) ){ goto no_mem;} + +/* Return true if the cursor was opened using the OP_OpenSorter opcode. */ +#define isSorter(x) ((x)->eCurType==CURTYPE_SORTER) + +/* +** Allocate VdbeCursor number iCur. Return a pointer to it. Return NULL +** if we run out of memory. +*/ +static VdbeCursor *allocateCursor( + Vdbe *p, /* The virtual machine */ + int iCur, /* Index of the new VdbeCursor */ + int nField, /* Number of fields in the table or index */ + int iDb, /* Database the cursor belongs to, or -1 */ + u8 eCurType /* Type of the new cursor */ +){ + /* Find the memory cell that will be used to store the blob of memory + ** required for this VdbeCursor structure. It is convenient to use a + ** vdbe memory cell to manage the memory allocation required for a + ** VdbeCursor structure for the following reasons: + ** + ** * Sometimes cursor numbers are used for a couple of different + ** purposes in a vdbe program. The different uses might require + ** different sized allocations. Memory cells provide growable + ** allocations. + ** + ** * When using ENABLE_MEMORY_MANAGEMENT, memory cell buffers can + ** be freed lazily via the sqlite3_release_memory() API. This + ** minimizes the number of malloc calls made by the system. + ** + ** The memory cell for cursor 0 is aMem[0]. The rest are allocated from + ** the top of the register space. Cursor 1 is at Mem[p->nMem-1]. + ** Cursor 2 is at Mem[p->nMem-2]. And so forth. + */ + Mem *pMem = iCur>0 ? &p->aMem[p->nMem-iCur] : p->aMem; + + int nByte; + VdbeCursor *pCx = 0; + nByte = + ROUND8(sizeof(VdbeCursor)) + 2*sizeof(u32)*nField + + (eCurType==CURTYPE_BTREE?sqlite3BtreeCursorSize():0); + + assert( iCur>=0 && iCurnCursor ); + if( p->apCsr[iCur] ){ /*OPTIMIZATION-IF-FALSE*/ + /* Before calling sqlite3VdbeFreeCursor(), ensure the isEphemeral flag + ** is clear. Otherwise, if this is an ephemeral cursor created by + ** OP_OpenDup, the cursor will not be closed and will still be part + ** of a BtShared.pCursor list. */ + if( p->apCsr[iCur]->pBtx==0 ) p->apCsr[iCur]->isEphemeral = 0; + sqlite3VdbeFreeCursor(p, p->apCsr[iCur]); + p->apCsr[iCur] = 0; + } + if( SQLITE_OK==sqlite3VdbeMemClearAndResize(pMem, nByte) ){ + p->apCsr[iCur] = pCx = (VdbeCursor*)pMem->z; + memset(pCx, 0, offsetof(VdbeCursor,pAltCursor)); + pCx->eCurType = eCurType; + pCx->iDb = iDb; + pCx->nField = nField; + pCx->aOffset = &pCx->aType[nField]; + if( eCurType==CURTYPE_BTREE ){ + pCx->uc.pCursor = (BtCursor*) + &pMem->z[ROUND8(sizeof(VdbeCursor))+2*sizeof(u32)*nField]; + sqlite3BtreeCursorZero(pCx->uc.pCursor); + } + } + return pCx; +} + +/* +** The string in pRec is known to look like an integer and to have a +** floating point value of rValue. Return true and set *piValue to the +** integer value if the string is in range to be an integer. Otherwise, +** return false. +*/ +static int alsoAnInt(Mem *pRec, double rValue, i64 *piValue){ + i64 iValue = (double)rValue; + if( sqlite3RealSameAsInt(rValue,iValue) ){ + *piValue = iValue; + return 1; + } + return 0==sqlite3Atoi64(pRec->z, piValue, pRec->n, pRec->enc); +} + +/* +** Try to convert a value into a numeric representation if we can +** do so without loss of information. In other words, if the string +** looks like a number, convert it into a number. If it does not +** look like a number, leave it alone. +** +** If the bTryForInt flag is true, then extra effort is made to give +** an integer representation. Strings that look like floating point +** values but which have no fractional component (example: '48.00') +** will have a MEM_Int representation when bTryForInt is true. +** +** If bTryForInt is false, then if the input string contains a decimal +** point or exponential notation, the result is only MEM_Real, even +** if there is an exact integer representation of the quantity. +*/ +static void applyNumericAffinity(Mem *pRec, int bTryForInt){ + double rValue; + u8 enc = pRec->enc; + int rc; + assert( (pRec->flags & (MEM_Str|MEM_Int|MEM_Real|MEM_IntReal))==MEM_Str ); + rc = sqlite3AtoF(pRec->z, &rValue, pRec->n, enc); + if( rc<=0 ) return; + if( rc==1 && alsoAnInt(pRec, rValue, &pRec->u.i) ){ + pRec->flags |= MEM_Int; + }else{ + pRec->u.r = rValue; + pRec->flags |= MEM_Real; + if( bTryForInt ) sqlite3VdbeIntegerAffinity(pRec); + } + /* TEXT->NUMERIC is many->one. Hence, it is important to invalidate the + ** string representation after computing a numeric equivalent, because the + ** string representation might not be the canonical representation for the + ** numeric value. Ticket [343634942dd54ab57b7024] 2018-01-31. */ + pRec->flags &= ~MEM_Str; +} + +/* +** Processing is determine by the affinity parameter: +** +** SQLITE_AFF_INTEGER: +** SQLITE_AFF_REAL: +** SQLITE_AFF_NUMERIC: +** Try to convert pRec to an integer representation or a +** floating-point representation if an integer representation +** is not possible. Note that the integer representation is +** always preferred, even if the affinity is REAL, because +** an integer representation is more space efficient on disk. +** +** SQLITE_AFF_TEXT: +** Convert pRec to a text representation. +** +** SQLITE_AFF_BLOB: +** SQLITE_AFF_NONE: +** No-op. pRec is unchanged. +*/ +static void applyAffinity( + Mem *pRec, /* The value to apply affinity to */ + char affinity, /* The affinity to be applied */ + u8 enc /* Use this text encoding */ +){ + if( affinity>=SQLITE_AFF_NUMERIC ){ + assert( affinity==SQLITE_AFF_INTEGER || affinity==SQLITE_AFF_REAL + || affinity==SQLITE_AFF_NUMERIC ); + if( (pRec->flags & MEM_Int)==0 ){ /*OPTIMIZATION-IF-FALSE*/ + if( (pRec->flags & MEM_Real)==0 ){ + if( pRec->flags & MEM_Str ) applyNumericAffinity(pRec,1); + }else{ + sqlite3VdbeIntegerAffinity(pRec); + } + } + }else if( affinity==SQLITE_AFF_TEXT ){ + /* Only attempt the conversion to TEXT if there is an integer or real + ** representation (blob and NULL do not get converted) but no string + ** representation. It would be harmless to repeat the conversion if + ** there is already a string rep, but it is pointless to waste those + ** CPU cycles. */ + if( 0==(pRec->flags&MEM_Str) ){ /*OPTIMIZATION-IF-FALSE*/ + if( (pRec->flags&(MEM_Real|MEM_Int|MEM_IntReal)) ){ + testcase( pRec->flags & MEM_Int ); + testcase( pRec->flags & MEM_Real ); + testcase( pRec->flags & MEM_IntReal ); + sqlite3VdbeMemStringify(pRec, enc, 1); + } + } + pRec->flags &= ~(MEM_Real|MEM_Int|MEM_IntReal); + } +} + +/* +** Try to convert the type of a function argument or a result column +** into a numeric representation. Use either INTEGER or REAL whichever +** is appropriate. But only do the conversion if it is possible without +** loss of information and return the revised type of the argument. +*/ +SQLITE_API int sqlite3_value_numeric_type(sqlite3_value *pVal){ + int eType = sqlite3_value_type(pVal); + if( eType==SQLITE_TEXT ){ + Mem *pMem = (Mem*)pVal; + applyNumericAffinity(pMem, 0); + eType = sqlite3_value_type(pVal); + } + return eType; +} + +/* +** Exported version of applyAffinity(). This one works on sqlite3_value*, +** not the internal Mem* type. +*/ +SQLITE_PRIVATE void sqlite3ValueApplyAffinity( + sqlite3_value *pVal, + u8 affinity, + u8 enc +){ + applyAffinity((Mem *)pVal, affinity, enc); +} + +/* +** pMem currently only holds a string type (or maybe a BLOB that we can +** interpret as a string if we want to). Compute its corresponding +** numeric type, if has one. Set the pMem->u.r and pMem->u.i fields +** accordingly. +*/ +static u16 SQLITE_NOINLINE computeNumericType(Mem *pMem){ + int rc; + sqlite3_int64 ix; + assert( (pMem->flags & (MEM_Int|MEM_Real|MEM_IntReal))==0 ); + assert( (pMem->flags & (MEM_Str|MEM_Blob))!=0 ); + ExpandBlob(pMem); + rc = sqlite3AtoF(pMem->z, &pMem->u.r, pMem->n, pMem->enc); + if( rc<=0 ){ + if( rc==0 && sqlite3Atoi64(pMem->z, &ix, pMem->n, pMem->enc)<=1 ){ + pMem->u.i = ix; + return MEM_Int; + }else{ + return MEM_Real; + } + }else if( rc==1 && sqlite3Atoi64(pMem->z, &ix, pMem->n, pMem->enc)==0 ){ + pMem->u.i = ix; + return MEM_Int; + } + return MEM_Real; +} + +/* +** Return the numeric type for pMem, either MEM_Int or MEM_Real or both or +** none. +** +** Unlike applyNumericAffinity(), this routine does not modify pMem->flags. +** But it does set pMem->u.r and pMem->u.i appropriately. +*/ +static u16 numericType(Mem *pMem){ + if( pMem->flags & (MEM_Int|MEM_Real|MEM_IntReal) ){ + testcase( pMem->flags & MEM_Int ); + testcase( pMem->flags & MEM_Real ); + testcase( pMem->flags & MEM_IntReal ); + return pMem->flags & (MEM_Int|MEM_Real|MEM_IntReal); + } + if( pMem->flags & (MEM_Str|MEM_Blob) ){ + testcase( pMem->flags & MEM_Str ); + testcase( pMem->flags & MEM_Blob ); + return computeNumericType(pMem); + } + return 0; +} + +#ifdef SQLITE_DEBUG +/* +** Write a nice string representation of the contents of cell pMem +** into buffer zBuf, length nBuf. +*/ +SQLITE_PRIVATE void sqlite3VdbeMemPrettyPrint(Mem *pMem, char *zBuf){ + char *zCsr = zBuf; + int f = pMem->flags; + + static const char *const encnames[] = {"(X)", "(8)", "(16LE)", "(16BE)"}; + + if( f&MEM_Blob ){ + int i; + char c; + if( f & MEM_Dyn ){ + c = 'z'; + assert( (f & (MEM_Static|MEM_Ephem))==0 ); + }else if( f & MEM_Static ){ + c = 't'; + assert( (f & (MEM_Dyn|MEM_Ephem))==0 ); + }else if( f & MEM_Ephem ){ + c = 'e'; + assert( (f & (MEM_Static|MEM_Dyn))==0 ); + }else{ + c = 's'; + } + *(zCsr++) = c; + *(zCsr++) = 'x'; + sqlite3_snprintf(100, zCsr, "%d[", pMem->n); + zCsr += sqlite3Strlen30(zCsr); + for(i=0; i<25 && in; i++){ + sqlite3_snprintf(100, zCsr, "%02X", ((int)pMem->z[i] & 0xFF)); + zCsr += sqlite3Strlen30(zCsr); + } + *zCsr++ = '|'; + for(i=0; i<25 && in; i++){ + char z = pMem->z[i]; + if( z<32 || z>126 ) *zCsr++ = '.'; + else *zCsr++ = z; + } + *(zCsr++) = ']'; + if( f & MEM_Zero ){ + sqlite3_snprintf(100, zCsr,"+%dz",pMem->u.nZero); + zCsr += sqlite3Strlen30(zCsr); + } + *zCsr = '\0'; + }else if( f & MEM_Str ){ + int j, k; + zBuf[0] = ' '; + if( f & MEM_Dyn ){ + zBuf[1] = 'z'; + assert( (f & (MEM_Static|MEM_Ephem))==0 ); + }else if( f & MEM_Static ){ + zBuf[1] = 't'; + assert( (f & (MEM_Dyn|MEM_Ephem))==0 ); + }else if( f & MEM_Ephem ){ + zBuf[1] = 'e'; + assert( (f & (MEM_Static|MEM_Dyn))==0 ); + }else{ + zBuf[1] = 's'; + } + k = 2; + sqlite3_snprintf(100, &zBuf[k], "%d", pMem->n); + k += sqlite3Strlen30(&zBuf[k]); + zBuf[k++] = '['; + for(j=0; j<25 && jn; j++){ + u8 c = pMem->z[j]; + if( c>=0x20 && c<0x7f ){ + zBuf[k++] = c; + }else{ + zBuf[k++] = '.'; + } + } + zBuf[k++] = ']'; + sqlite3_snprintf(100,&zBuf[k], encnames[pMem->enc]); + k += sqlite3Strlen30(&zBuf[k]); + zBuf[k++] = 0; + } +} +#endif + +#ifdef SQLITE_DEBUG +/* +** Print the value of a register for tracing purposes: +*/ +static void memTracePrint(Mem *p){ + if( p->flags & MEM_Undefined ){ + printf(" undefined"); + }else if( p->flags & MEM_Null ){ + printf(p->flags & MEM_Zero ? " NULL-nochng" : " NULL"); + }else if( (p->flags & (MEM_Int|MEM_Str))==(MEM_Int|MEM_Str) ){ + printf(" si:%lld", p->u.i); + }else if( (p->flags & (MEM_IntReal))!=0 ){ + printf(" ir:%lld", p->u.i); + }else if( p->flags & MEM_Int ){ + printf(" i:%lld", p->u.i); +#ifndef SQLITE_OMIT_FLOATING_POINT + }else if( p->flags & MEM_Real ){ + printf(" r:%.17g", p->u.r); +#endif + }else if( sqlite3VdbeMemIsRowSet(p) ){ + printf(" (rowset)"); + }else{ + char zBuf[200]; + sqlite3VdbeMemPrettyPrint(p, zBuf); + printf(" %s", zBuf); + } + if( p->flags & MEM_Subtype ) printf(" subtype=0x%02x", p->eSubtype); +} +static void registerTrace(int iReg, Mem *p){ + printf("REG[%d] = ", iReg); + memTracePrint(p); + printf("\n"); + sqlite3VdbeCheckMemInvariants(p); +} +#endif + +#ifdef SQLITE_DEBUG +# define REGISTER_TRACE(R,M) if(db->flags&SQLITE_VdbeTrace)registerTrace(R,M) +#else +# define REGISTER_TRACE(R,M) +#endif + + +#ifdef VDBE_PROFILE + +/* +** hwtime.h contains inline assembler code for implementing +** high-performance timing routines. +*/ +/************** Include hwtime.h in the middle of vdbe.c *********************/ +/************** Begin file hwtime.h ******************************************/ +/* +** 2008 May 27 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +****************************************************************************** +** +** This file contains inline asm code for retrieving "high-performance" +** counters for x86 class CPUs. +*/ +#ifndef SQLITE_HWTIME_H +#define SQLITE_HWTIME_H + +/* +** The following routine only works on pentium-class (or newer) processors. +** It uses the RDTSC opcode to read the cycle count value out of the +** processor and returns that value. This can be used for high-res +** profiling. +*/ +#if (defined(__GNUC__) || defined(_MSC_VER)) && \ + (defined(i386) || defined(__i386__) || defined(_M_IX86)) + + #if defined(__GNUC__) + + __inline__ sqlite_uint64 sqlite3Hwtime(void){ + unsigned int lo, hi; + __asm__ __volatile__ ("rdtsc" : "=a" (lo), "=d" (hi)); + return (sqlite_uint64)hi << 32 | lo; + } + + #elif defined(_MSC_VER) + + __declspec(naked) __inline sqlite_uint64 __cdecl sqlite3Hwtime(void){ + __asm { + rdtsc + ret ; return value at EDX:EAX + } + } + + #endif + +#elif (defined(__GNUC__) && defined(__x86_64__)) + + __inline__ sqlite_uint64 sqlite3Hwtime(void){ + unsigned long val; + __asm__ __volatile__ ("rdtsc" : "=A" (val)); + return val; + } + +#elif (defined(__GNUC__) && defined(__ppc__)) + + __inline__ sqlite_uint64 sqlite3Hwtime(void){ + unsigned long long retval; + unsigned long junk; + __asm__ __volatile__ ("\n\ + 1: mftbu %1\n\ + mftb %L0\n\ + mftbu %0\n\ + cmpw %0,%1\n\ + bne 1b" + : "=r" (retval), "=r" (junk)); + return retval; + } + +#else + + #error Need implementation of sqlite3Hwtime() for your platform. + + /* + ** To compile without implementing sqlite3Hwtime() for your platform, + ** you can remove the above #error and use the following + ** stub function. You will lose timing support for many + ** of the debugging and testing utilities, but it should at + ** least compile and run. + */ +SQLITE_PRIVATE sqlite_uint64 sqlite3Hwtime(void){ return ((sqlite_uint64)0); } + +#endif + +#endif /* !defined(SQLITE_HWTIME_H) */ + +/************** End of hwtime.h **********************************************/ +/************** Continuing where we left off in vdbe.c ***********************/ + +#endif + +#ifndef NDEBUG +/* +** This function is only called from within an assert() expression. It +** checks that the sqlite3.nTransaction variable is correctly set to +** the number of non-transaction savepoints currently in the +** linked list starting at sqlite3.pSavepoint. +** +** Usage: +** +** assert( checkSavepointCount(db) ); +*/ +static int checkSavepointCount(sqlite3 *db){ + int n = 0; + Savepoint *p; + for(p=db->pSavepoint; p; p=p->pNext) n++; + assert( n==(db->nSavepoint + db->isTransactionSavepoint) ); + return 1; +} +#endif + +/* +** Return the register of pOp->p2 after first preparing it to be +** overwritten with an integer value. +*/ +static SQLITE_NOINLINE Mem *out2PrereleaseWithClear(Mem *pOut){ + sqlite3VdbeMemSetNull(pOut); + pOut->flags = MEM_Int; + return pOut; +} +static Mem *out2Prerelease(Vdbe *p, VdbeOp *pOp){ + Mem *pOut; + assert( pOp->p2>0 ); + assert( pOp->p2<=(p->nMem+1 - p->nCursor) ); + pOut = &p->aMem[pOp->p2]; + memAboutToChange(p, pOut); + if( VdbeMemDynamic(pOut) ){ /*OPTIMIZATION-IF-FALSE*/ + return out2PrereleaseWithClear(pOut); + }else{ + pOut->flags = MEM_Int; + return pOut; + } +} + + +/* +** Execute as much of a VDBE program as we can. +** This is the core of sqlite3_step(). +*/ +SQLITE_PRIVATE int sqlite3VdbeExec( + Vdbe *p /* The VDBE */ +){ + Op *aOp = p->aOp; /* Copy of p->aOp */ + Op *pOp = aOp; /* Current operation */ +#if defined(SQLITE_DEBUG) || defined(VDBE_PROFILE) + Op *pOrigOp; /* Value of pOp at the top of the loop */ +#endif +#ifdef SQLITE_DEBUG + int nExtraDelete = 0; /* Verifies FORDELETE and AUXDELETE flags */ +#endif + int rc = SQLITE_OK; /* Value to return */ + sqlite3 *db = p->db; /* The database */ + u8 resetSchemaOnFault = 0; /* Reset schema after an error if positive */ + u8 encoding = ENC(db); /* The database encoding */ + int iCompare = 0; /* Result of last comparison */ + unsigned nVmStep = 0; /* Number of virtual machine steps */ +#ifndef SQLITE_OMIT_PROGRESS_CALLBACK + unsigned nProgressLimit; /* Invoke xProgress() when nVmStep reaches this */ +#endif + Mem *aMem = p->aMem; /* Copy of p->aMem */ + Mem *pIn1 = 0; /* 1st input operand */ + Mem *pIn2 = 0; /* 2nd input operand */ + Mem *pIn3 = 0; /* 3rd input operand */ + Mem *pOut = 0; /* Output operand */ +#ifdef VDBE_PROFILE + u64 start; /* CPU clock count at start of opcode */ +#endif + /*** INSERT STACK UNION HERE ***/ + + assert( p->magic==VDBE_MAGIC_RUN ); /* sqlite3_step() verifies this */ + sqlite3VdbeEnter(p); +#ifndef SQLITE_OMIT_PROGRESS_CALLBACK + if( db->xProgress ){ + u32 iPrior = p->aCounter[SQLITE_STMTSTATUS_VM_STEP]; + assert( 0 < db->nProgressOps ); + nProgressLimit = db->nProgressOps - (iPrior % db->nProgressOps); + }else{ + nProgressLimit = 0xffffffff; + } +#endif + if( p->rc==SQLITE_NOMEM ){ + /* This happens if a malloc() inside a call to sqlite3_column_text() or + ** sqlite3_column_text16() failed. */ + goto no_mem; + } + assert( p->rc==SQLITE_OK || (p->rc&0xff)==SQLITE_BUSY ); + assert( p->bIsReader || p->readOnly!=0 ); + p->iCurrentTime = 0; + assert( p->explain==0 ); + p->pResultSet = 0; + db->busyHandler.nBusy = 0; + if( db->u1.isInterrupted ) goto abort_due_to_interrupt; + sqlite3VdbeIOTraceSql(p); +#ifdef SQLITE_DEBUG + sqlite3BeginBenignMalloc(); + if( p->pc==0 + && (p->db->flags & (SQLITE_VdbeListing|SQLITE_VdbeEQP|SQLITE_VdbeTrace))!=0 + ){ + int i; + int once = 1; + sqlite3VdbePrintSql(p); + if( p->db->flags & SQLITE_VdbeListing ){ + printf("VDBE Program Listing:\n"); + for(i=0; inOp; i++){ + sqlite3VdbePrintOp(stdout, i, &aOp[i]); + } + } + if( p->db->flags & SQLITE_VdbeEQP ){ + for(i=0; inOp; i++){ + if( aOp[i].opcode==OP_Explain ){ + if( once ) printf("VDBE Query Plan:\n"); + printf("%s\n", aOp[i].p4.z); + once = 0; + } + } + } + if( p->db->flags & SQLITE_VdbeTrace ) printf("VDBE Trace:\n"); + } + sqlite3EndBenignMalloc(); +#endif + for(pOp=&aOp[p->pc]; 1; pOp++){ + /* Errors are detected by individual opcodes, with an immediate + ** jumps to abort_due_to_error. */ + assert( rc==SQLITE_OK ); + + assert( pOp>=aOp && pOp<&aOp[p->nOp]); +#ifdef VDBE_PROFILE + start = sqlite3NProfileCnt ? sqlite3NProfileCnt : sqlite3Hwtime(); +#endif + nVmStep++; +#ifdef SQLITE_ENABLE_STMT_SCANSTATUS + if( p->anExec ) p->anExec[(int)(pOp-aOp)]++; +#endif + + /* Only allow tracing if SQLITE_DEBUG is defined. + */ +#ifdef SQLITE_DEBUG + if( db->flags & SQLITE_VdbeTrace ){ + sqlite3VdbePrintOp(stdout, (int)(pOp - aOp), pOp); + } +#endif + + + /* Check to see if we need to simulate an interrupt. This only happens + ** if we have a special test build. + */ +#ifdef SQLITE_TEST + if( sqlite3_interrupt_count>0 ){ + sqlite3_interrupt_count--; + if( sqlite3_interrupt_count==0 ){ + sqlite3_interrupt(db); + } + } +#endif + + /* Sanity checking on other operands */ +#ifdef SQLITE_DEBUG + { + u8 opProperty = sqlite3OpcodeProperty[pOp->opcode]; + if( (opProperty & OPFLG_IN1)!=0 ){ + assert( pOp->p1>0 ); + assert( pOp->p1<=(p->nMem+1 - p->nCursor) ); + assert( memIsValid(&aMem[pOp->p1]) ); + assert( sqlite3VdbeCheckMemInvariants(&aMem[pOp->p1]) ); + REGISTER_TRACE(pOp->p1, &aMem[pOp->p1]); + } + if( (opProperty & OPFLG_IN2)!=0 ){ + assert( pOp->p2>0 ); + assert( pOp->p2<=(p->nMem+1 - p->nCursor) ); + assert( memIsValid(&aMem[pOp->p2]) ); + assert( sqlite3VdbeCheckMemInvariants(&aMem[pOp->p2]) ); + REGISTER_TRACE(pOp->p2, &aMem[pOp->p2]); + } + if( (opProperty & OPFLG_IN3)!=0 ){ + assert( pOp->p3>0 ); + assert( pOp->p3<=(p->nMem+1 - p->nCursor) ); + assert( memIsValid(&aMem[pOp->p3]) ); + assert( sqlite3VdbeCheckMemInvariants(&aMem[pOp->p3]) ); + REGISTER_TRACE(pOp->p3, &aMem[pOp->p3]); + } + if( (opProperty & OPFLG_OUT2)!=0 ){ + assert( pOp->p2>0 ); + assert( pOp->p2<=(p->nMem+1 - p->nCursor) ); + memAboutToChange(p, &aMem[pOp->p2]); + } + if( (opProperty & OPFLG_OUT3)!=0 ){ + assert( pOp->p3>0 ); + assert( pOp->p3<=(p->nMem+1 - p->nCursor) ); + memAboutToChange(p, &aMem[pOp->p3]); + } + } +#endif +#if defined(SQLITE_DEBUG) || defined(VDBE_PROFILE) + pOrigOp = pOp; +#endif + + switch( pOp->opcode ){ + +/***************************************************************************** +** What follows is a massive switch statement where each case implements a +** separate instruction in the virtual machine. If we follow the usual +** indentation conventions, each case should be indented by 6 spaces. But +** that is a lot of wasted space on the left margin. So the code within +** the switch statement will break with convention and be flush-left. Another +** big comment (similar to this one) will mark the point in the code where +** we transition back to normal indentation. +** +** The formatting of each case is important. The makefile for SQLite +** generates two C files "opcodes.h" and "opcodes.c" by scanning this +** file looking for lines that begin with "case OP_". The opcodes.h files +** will be filled with #defines that give unique integer values to each +** opcode and the opcodes.c file is filled with an array of strings where +** each string is the symbolic name for the corresponding opcode. If the +** case statement is followed by a comment of the form "/# same as ... #/" +** that comment is used to determine the particular value of the opcode. +** +** Other keywords in the comment that follows each case are used to +** construct the OPFLG_INITIALIZER value that initializes opcodeProperty[]. +** Keywords include: in1, in2, in3, out2, out3. See +** the mkopcodeh.awk script for additional information. +** +** Documentation about VDBE opcodes is generated by scanning this file +** for lines of that contain "Opcode:". That line and all subsequent +** comment lines are used in the generation of the opcode.html documentation +** file. +** +** SUMMARY: +** +** Formatting is important to scripts that scan this file. +** Do not deviate from the formatting style currently in use. +** +*****************************************************************************/ + +/* Opcode: Goto * P2 * * * +** +** An unconditional jump to address P2. +** The next instruction executed will be +** the one at index P2 from the beginning of +** the program. +** +** The P1 parameter is not actually used by this opcode. However, it +** is sometimes set to 1 instead of 0 as a hint to the command-line shell +** that this Goto is the bottom of a loop and that the lines from P2 down +** to the current line should be indented for EXPLAIN output. +*/ +case OP_Goto: { /* jump */ +jump_to_p2_and_check_for_interrupt: + pOp = &aOp[pOp->p2 - 1]; + + /* Opcodes that are used as the bottom of a loop (OP_Next, OP_Prev, + ** OP_VNext, or OP_SorterNext) all jump here upon + ** completion. Check to see if sqlite3_interrupt() has been called + ** or if the progress callback needs to be invoked. + ** + ** This code uses unstructured "goto" statements and does not look clean. + ** But that is not due to sloppy coding habits. The code is written this + ** way for performance, to avoid having to run the interrupt and progress + ** checks on every opcode. This helps sqlite3_step() to run about 1.5% + ** faster according to "valgrind --tool=cachegrind" */ +check_for_interrupt: + if( db->u1.isInterrupted ) goto abort_due_to_interrupt; +#ifndef SQLITE_OMIT_PROGRESS_CALLBACK + /* Call the progress callback if it is configured and the required number + ** of VDBE ops have been executed (either since this invocation of + ** sqlite3VdbeExec() or since last time the progress callback was called). + ** If the progress callback returns non-zero, exit the virtual machine with + ** a return code SQLITE_ABORT. + */ + while( nVmStep>=nProgressLimit && db->xProgress!=0 ){ + assert( db->nProgressOps!=0 ); + nProgressLimit += db->nProgressOps; + if( db->xProgress(db->pProgressArg) ){ + nProgressLimit = 0xffffffff; + rc = SQLITE_INTERRUPT; + goto abort_due_to_error; + } + } +#endif + + break; +} + +/* Opcode: Gosub P1 P2 * * * +** +** Write the current address onto register P1 +** and then jump to address P2. +*/ +case OP_Gosub: { /* jump */ + assert( pOp->p1>0 && pOp->p1<=(p->nMem+1 - p->nCursor) ); + pIn1 = &aMem[pOp->p1]; + assert( VdbeMemDynamic(pIn1)==0 ); + memAboutToChange(p, pIn1); + pIn1->flags = MEM_Int; + pIn1->u.i = (int)(pOp-aOp); + REGISTER_TRACE(pOp->p1, pIn1); + + /* Most jump operations do a goto to this spot in order to update + ** the pOp pointer. */ +jump_to_p2: + pOp = &aOp[pOp->p2 - 1]; + break; +} + +/* Opcode: Return P1 * * * * +** +** Jump to the next instruction after the address in register P1. After +** the jump, register P1 becomes undefined. +*/ +case OP_Return: { /* in1 */ + pIn1 = &aMem[pOp->p1]; + assert( pIn1->flags==MEM_Int ); + pOp = &aOp[pIn1->u.i]; + pIn1->flags = MEM_Undefined; + break; +} + +/* Opcode: InitCoroutine P1 P2 P3 * * +** +** Set up register P1 so that it will Yield to the coroutine +** located at address P3. +** +** If P2!=0 then the coroutine implementation immediately follows +** this opcode. So jump over the coroutine implementation to +** address P2. +** +** See also: EndCoroutine +*/ +case OP_InitCoroutine: { /* jump */ + assert( pOp->p1>0 && pOp->p1<=(p->nMem+1 - p->nCursor) ); + assert( pOp->p2>=0 && pOp->p2nOp ); + assert( pOp->p3>=0 && pOp->p3nOp ); + pOut = &aMem[pOp->p1]; + assert( !VdbeMemDynamic(pOut) ); + pOut->u.i = pOp->p3 - 1; + pOut->flags = MEM_Int; + if( pOp->p2 ) goto jump_to_p2; + break; +} + +/* Opcode: EndCoroutine P1 * * * * +** +** The instruction at the address in register P1 is a Yield. +** Jump to the P2 parameter of that Yield. +** After the jump, register P1 becomes undefined. +** +** See also: InitCoroutine +*/ +case OP_EndCoroutine: { /* in1 */ + VdbeOp *pCaller; + pIn1 = &aMem[pOp->p1]; + assert( pIn1->flags==MEM_Int ); + assert( pIn1->u.i>=0 && pIn1->u.inOp ); + pCaller = &aOp[pIn1->u.i]; + assert( pCaller->opcode==OP_Yield ); + assert( pCaller->p2>=0 && pCaller->p2nOp ); + pOp = &aOp[pCaller->p2 - 1]; + pIn1->flags = MEM_Undefined; + break; +} + +/* Opcode: Yield P1 P2 * * * +** +** Swap the program counter with the value in register P1. This +** has the effect of yielding to a coroutine. +** +** If the coroutine that is launched by this instruction ends with +** Yield or Return then continue to the next instruction. But if +** the coroutine launched by this instruction ends with +** EndCoroutine, then jump to P2 rather than continuing with the +** next instruction. +** +** See also: InitCoroutine +*/ +case OP_Yield: { /* in1, jump */ + int pcDest; + pIn1 = &aMem[pOp->p1]; + assert( VdbeMemDynamic(pIn1)==0 ); + pIn1->flags = MEM_Int; + pcDest = (int)pIn1->u.i; + pIn1->u.i = (int)(pOp - aOp); + REGISTER_TRACE(pOp->p1, pIn1); + pOp = &aOp[pcDest]; + break; +} + +/* Opcode: HaltIfNull P1 P2 P3 P4 P5 +** Synopsis: if r[P3]=null halt +** +** Check the value in register P3. If it is NULL then Halt using +** parameter P1, P2, and P4 as if this were a Halt instruction. If the +** value in register P3 is not NULL, then this routine is a no-op. +** The P5 parameter should be 1. +*/ +case OP_HaltIfNull: { /* in3 */ + pIn3 = &aMem[pOp->p3]; +#ifdef SQLITE_DEBUG + if( pOp->p2==OE_Abort ){ sqlite3VdbeAssertAbortable(p); } +#endif + if( (pIn3->flags & MEM_Null)==0 ) break; + /* Fall through into OP_Halt */ +} + +/* Opcode: Halt P1 P2 * P4 P5 +** +** Exit immediately. All open cursors, etc are closed +** automatically. +** +** P1 is the result code returned by sqlite3_exec(), sqlite3_reset(), +** or sqlite3_finalize(). For a normal halt, this should be SQLITE_OK (0). +** For errors, it can be some other value. If P1!=0 then P2 will determine +** whether or not to rollback the current transaction. Do not rollback +** if P2==OE_Fail. Do the rollback if P2==OE_Rollback. If P2==OE_Abort, +** then back out all changes that have occurred during this execution of the +** VDBE, but do not rollback the transaction. +** +** If P4 is not null then it is an error message string. +** +** P5 is a value between 0 and 4, inclusive, that modifies the P4 string. +** +** 0: (no change) +** 1: NOT NULL contraint failed: P4 +** 2: UNIQUE constraint failed: P4 +** 3: CHECK constraint failed: P4 +** 4: FOREIGN KEY constraint failed: P4 +** +** If P5 is not zero and P4 is NULL, then everything after the ":" is +** omitted. +** +** There is an implied "Halt 0 0 0" instruction inserted at the very end of +** every program. So a jump past the last instruction of the program +** is the same as executing Halt. +*/ +case OP_Halt: { + VdbeFrame *pFrame; + int pcx; + + pcx = (int)(pOp - aOp); +#ifdef SQLITE_DEBUG + if( pOp->p2==OE_Abort ){ sqlite3VdbeAssertAbortable(p); } +#endif + if( pOp->p1==SQLITE_OK && p->pFrame ){ + /* Halt the sub-program. Return control to the parent frame. */ + pFrame = p->pFrame; + p->pFrame = pFrame->pParent; + p->nFrame--; + sqlite3VdbeSetChanges(db, p->nChange); + pcx = sqlite3VdbeFrameRestore(pFrame); + if( pOp->p2==OE_Ignore ){ + /* Instruction pcx is the OP_Program that invoked the sub-program + ** currently being halted. If the p2 instruction of this OP_Halt + ** instruction is set to OE_Ignore, then the sub-program is throwing + ** an IGNORE exception. In this case jump to the address specified + ** as the p2 of the calling OP_Program. */ + pcx = p->aOp[pcx].p2-1; + } + aOp = p->aOp; + aMem = p->aMem; + pOp = &aOp[pcx]; + break; + } + p->rc = pOp->p1; + p->errorAction = (u8)pOp->p2; + p->pc = pcx; + assert( pOp->p5<=4 ); + if( p->rc ){ + if( pOp->p5 ){ + static const char * const azType[] = { "NOT NULL", "UNIQUE", "CHECK", + "FOREIGN KEY" }; + testcase( pOp->p5==1 ); + testcase( pOp->p5==2 ); + testcase( pOp->p5==3 ); + testcase( pOp->p5==4 ); + sqlite3VdbeError(p, "%s constraint failed", azType[pOp->p5-1]); + if( pOp->p4.z ){ + p->zErrMsg = sqlite3MPrintf(db, "%z: %s", p->zErrMsg, pOp->p4.z); + } + }else{ + sqlite3VdbeError(p, "%s", pOp->p4.z); + } + sqlite3_log(pOp->p1, "abort at %d in [%s]: %s", pcx, p->zSql, p->zErrMsg); + } + rc = sqlite3VdbeHalt(p); + assert( rc==SQLITE_BUSY || rc==SQLITE_OK || rc==SQLITE_ERROR ); + if( rc==SQLITE_BUSY ){ + p->rc = SQLITE_BUSY; + }else{ + assert( rc==SQLITE_OK || (p->rc&0xff)==SQLITE_CONSTRAINT ); + assert( rc==SQLITE_OK || db->nDeferredCons>0 || db->nDeferredImmCons>0 ); + rc = p->rc ? SQLITE_ERROR : SQLITE_DONE; + } + goto vdbe_return; +} + +/* Opcode: Integer P1 P2 * * * +** Synopsis: r[P2]=P1 +** +** The 32-bit integer value P1 is written into register P2. +*/ +case OP_Integer: { /* out2 */ + pOut = out2Prerelease(p, pOp); + pOut->u.i = pOp->p1; + break; +} + +/* Opcode: Int64 * P2 * P4 * +** Synopsis: r[P2]=P4 +** +** P4 is a pointer to a 64-bit integer value. +** Write that value into register P2. +*/ +case OP_Int64: { /* out2 */ + pOut = out2Prerelease(p, pOp); + assert( pOp->p4.pI64!=0 ); + pOut->u.i = *pOp->p4.pI64; + break; +} + +#ifndef SQLITE_OMIT_FLOATING_POINT +/* Opcode: Real * P2 * P4 * +** Synopsis: r[P2]=P4 +** +** P4 is a pointer to a 64-bit floating point value. +** Write that value into register P2. +*/ +case OP_Real: { /* same as TK_FLOAT, out2 */ + pOut = out2Prerelease(p, pOp); + pOut->flags = MEM_Real; + assert( !sqlite3IsNaN(*pOp->p4.pReal) ); + pOut->u.r = *pOp->p4.pReal; + break; +} +#endif + +/* Opcode: String8 * P2 * P4 * +** Synopsis: r[P2]='P4' +** +** P4 points to a nul terminated UTF-8 string. This opcode is transformed +** into a String opcode before it is executed for the first time. During +** this transformation, the length of string P4 is computed and stored +** as the P1 parameter. +*/ +case OP_String8: { /* same as TK_STRING, out2 */ + assert( pOp->p4.z!=0 ); + pOut = out2Prerelease(p, pOp); + pOp->p1 = sqlite3Strlen30(pOp->p4.z); + +#ifndef SQLITE_OMIT_UTF16 + if( encoding!=SQLITE_UTF8 ){ + rc = sqlite3VdbeMemSetStr(pOut, pOp->p4.z, -1, SQLITE_UTF8, SQLITE_STATIC); + assert( rc==SQLITE_OK || rc==SQLITE_TOOBIG ); + if( rc ) goto too_big; + if( SQLITE_OK!=sqlite3VdbeChangeEncoding(pOut, encoding) ) goto no_mem; + assert( pOut->szMalloc>0 && pOut->zMalloc==pOut->z ); + assert( VdbeMemDynamic(pOut)==0 ); + pOut->szMalloc = 0; + pOut->flags |= MEM_Static; + if( pOp->p4type==P4_DYNAMIC ){ + sqlite3DbFree(db, pOp->p4.z); + } + pOp->p4type = P4_DYNAMIC; + pOp->p4.z = pOut->z; + pOp->p1 = pOut->n; + } +#endif + if( pOp->p1>db->aLimit[SQLITE_LIMIT_LENGTH] ){ + goto too_big; + } + pOp->opcode = OP_String; + assert( rc==SQLITE_OK ); + /* Fall through to the next case, OP_String */ +} + +/* Opcode: String P1 P2 P3 P4 P5 +** Synopsis: r[P2]='P4' (len=P1) +** +** The string value P4 of length P1 (bytes) is stored in register P2. +** +** If P3 is not zero and the content of register P3 is equal to P5, then +** the datatype of the register P2 is converted to BLOB. The content is +** the same sequence of bytes, it is merely interpreted as a BLOB instead +** of a string, as if it had been CAST. In other words: +** +** if( P3!=0 and reg[P3]==P5 ) reg[P2] := CAST(reg[P2] as BLOB) +*/ +case OP_String: { /* out2 */ + assert( pOp->p4.z!=0 ); + pOut = out2Prerelease(p, pOp); + pOut->flags = MEM_Str|MEM_Static|MEM_Term; + pOut->z = pOp->p4.z; + pOut->n = pOp->p1; + pOut->enc = encoding; + UPDATE_MAX_BLOBSIZE(pOut); +#ifndef SQLITE_LIKE_DOESNT_MATCH_BLOBS + if( pOp->p3>0 ){ + assert( pOp->p3<=(p->nMem+1 - p->nCursor) ); + pIn3 = &aMem[pOp->p3]; + assert( pIn3->flags & MEM_Int ); + if( pIn3->u.i==pOp->p5 ) pOut->flags = MEM_Blob|MEM_Static|MEM_Term; + } +#endif + break; +} + +/* Opcode: Null P1 P2 P3 * * +** Synopsis: r[P2..P3]=NULL +** +** Write a NULL into registers P2. If P3 greater than P2, then also write +** NULL into register P3 and every register in between P2 and P3. If P3 +** is less than P2 (typically P3 is zero) then only register P2 is +** set to NULL. +** +** If the P1 value is non-zero, then also set the MEM_Cleared flag so that +** NULL values will not compare equal even if SQLITE_NULLEQ is set on +** OP_Ne or OP_Eq. +*/ +case OP_Null: { /* out2 */ + int cnt; + u16 nullFlag; + pOut = out2Prerelease(p, pOp); + cnt = pOp->p3-pOp->p2; + assert( pOp->p3<=(p->nMem+1 - p->nCursor) ); + pOut->flags = nullFlag = pOp->p1 ? (MEM_Null|MEM_Cleared) : MEM_Null; + pOut->n = 0; +#ifdef SQLITE_DEBUG + pOut->uTemp = 0; +#endif + while( cnt>0 ){ + pOut++; + memAboutToChange(p, pOut); + sqlite3VdbeMemSetNull(pOut); + pOut->flags = nullFlag; + pOut->n = 0; + cnt--; + } + break; +} + +/* Opcode: SoftNull P1 * * * * +** Synopsis: r[P1]=NULL +** +** Set register P1 to have the value NULL as seen by the OP_MakeRecord +** instruction, but do not free any string or blob memory associated with +** the register, so that if the value was a string or blob that was +** previously copied using OP_SCopy, the copies will continue to be valid. +*/ +case OP_SoftNull: { + assert( pOp->p1>0 && pOp->p1<=(p->nMem+1 - p->nCursor) ); + pOut = &aMem[pOp->p1]; + pOut->flags = (pOut->flags&~(MEM_Undefined|MEM_AffMask))|MEM_Null; + break; +} + +/* Opcode: Blob P1 P2 * P4 * +** Synopsis: r[P2]=P4 (len=P1) +** +** P4 points to a blob of data P1 bytes long. Store this +** blob in register P2. +*/ +case OP_Blob: { /* out2 */ + assert( pOp->p1 <= SQLITE_MAX_LENGTH ); + pOut = out2Prerelease(p, pOp); + sqlite3VdbeMemSetStr(pOut, pOp->p4.z, pOp->p1, 0, 0); + pOut->enc = encoding; + UPDATE_MAX_BLOBSIZE(pOut); + break; +} + +/* Opcode: Variable P1 P2 * P4 * +** Synopsis: r[P2]=parameter(P1,P4) +** +** Transfer the values of bound parameter P1 into register P2 +** +** If the parameter is named, then its name appears in P4. +** The P4 value is used by sqlite3_bind_parameter_name(). +*/ +case OP_Variable: { /* out2 */ + Mem *pVar; /* Value being transferred */ + + assert( pOp->p1>0 && pOp->p1<=p->nVar ); + assert( pOp->p4.z==0 || pOp->p4.z==sqlite3VListNumToName(p->pVList,pOp->p1) ); + pVar = &p->aVar[pOp->p1 - 1]; + if( sqlite3VdbeMemTooBig(pVar) ){ + goto too_big; + } + pOut = &aMem[pOp->p2]; + if( VdbeMemDynamic(pOut) ) sqlite3VdbeMemSetNull(pOut); + memcpy(pOut, pVar, MEMCELLSIZE); + pOut->flags &= ~(MEM_Dyn|MEM_Ephem); + pOut->flags |= MEM_Static|MEM_FromBind; + UPDATE_MAX_BLOBSIZE(pOut); + break; +} + +/* Opcode: Move P1 P2 P3 * * +** Synopsis: r[P2@P3]=r[P1@P3] +** +** Move the P3 values in register P1..P1+P3-1 over into +** registers P2..P2+P3-1. Registers P1..P1+P3-1 are +** left holding a NULL. It is an error for register ranges +** P1..P1+P3-1 and P2..P2+P3-1 to overlap. It is an error +** for P3 to be less than 1. +*/ +case OP_Move: { + int n; /* Number of registers left to copy */ + int p1; /* Register to copy from */ + int p2; /* Register to copy to */ + + n = pOp->p3; + p1 = pOp->p1; + p2 = pOp->p2; + assert( n>0 && p1>0 && p2>0 ); + assert( p1+n<=p2 || p2+n<=p1 ); + + pIn1 = &aMem[p1]; + pOut = &aMem[p2]; + do{ + assert( pOut<=&aMem[(p->nMem+1 - p->nCursor)] ); + assert( pIn1<=&aMem[(p->nMem+1 - p->nCursor)] ); + assert( memIsValid(pIn1) ); + memAboutToChange(p, pOut); + sqlite3VdbeMemMove(pOut, pIn1); +#ifdef SQLITE_DEBUG + if( pOut->pScopyFrom>=&aMem[p1] && pOut->pScopyFrompScopyFrom += pOp->p2 - p1; + } +#endif + Deephemeralize(pOut); + REGISTER_TRACE(p2++, pOut); + pIn1++; + pOut++; + }while( --n ); + break; +} + +/* Opcode: Copy P1 P2 P3 * * +** Synopsis: r[P2@P3+1]=r[P1@P3+1] +** +** Make a copy of registers P1..P1+P3 into registers P2..P2+P3. +** +** This instruction makes a deep copy of the value. A duplicate +** is made of any string or blob constant. See also OP_SCopy. +*/ +case OP_Copy: { + int n; + + n = pOp->p3; + pIn1 = &aMem[pOp->p1]; + pOut = &aMem[pOp->p2]; + assert( pOut!=pIn1 ); + while( 1 ){ + memAboutToChange(p, pOut); + sqlite3VdbeMemShallowCopy(pOut, pIn1, MEM_Ephem); + Deephemeralize(pOut); +#ifdef SQLITE_DEBUG + pOut->pScopyFrom = 0; +#endif + REGISTER_TRACE(pOp->p2+pOp->p3-n, pOut); + if( (n--)==0 ) break; + pOut++; + pIn1++; + } + break; +} + +/* Opcode: SCopy P1 P2 * * * +** Synopsis: r[P2]=r[P1] +** +** Make a shallow copy of register P1 into register P2. +** +** This instruction makes a shallow copy of the value. If the value +** is a string or blob, then the copy is only a pointer to the +** original and hence if the original changes so will the copy. +** Worse, if the original is deallocated, the copy becomes invalid. +** Thus the program must guarantee that the original will not change +** during the lifetime of the copy. Use OP_Copy to make a complete +** copy. +*/ +case OP_SCopy: { /* out2 */ + pIn1 = &aMem[pOp->p1]; + pOut = &aMem[pOp->p2]; + assert( pOut!=pIn1 ); + sqlite3VdbeMemShallowCopy(pOut, pIn1, MEM_Ephem); +#ifdef SQLITE_DEBUG + pOut->pScopyFrom = pIn1; + pOut->mScopyFlags = pIn1->flags; +#endif + break; +} + +/* Opcode: IntCopy P1 P2 * * * +** Synopsis: r[P2]=r[P1] +** +** Transfer the integer value held in register P1 into register P2. +** +** This is an optimized version of SCopy that works only for integer +** values. +*/ +case OP_IntCopy: { /* out2 */ + pIn1 = &aMem[pOp->p1]; + assert( (pIn1->flags & MEM_Int)!=0 ); + pOut = &aMem[pOp->p2]; + sqlite3VdbeMemSetInt64(pOut, pIn1->u.i); + break; +} + +/* Opcode: ResultRow P1 P2 * * * +** Synopsis: output=r[P1@P2] +** +** The registers P1 through P1+P2-1 contain a single row of +** results. This opcode causes the sqlite3_step() call to terminate +** with an SQLITE_ROW return code and it sets up the sqlite3_stmt +** structure to provide access to the r(P1)..r(P1+P2-1) values as +** the result row. +*/ +case OP_ResultRow: { + Mem *pMem; + int i; + assert( p->nResColumn==pOp->p2 ); + assert( pOp->p1>0 ); + assert( pOp->p1+pOp->p2<=(p->nMem+1 - p->nCursor)+1 ); + + /* If this statement has violated immediate foreign key constraints, do + ** not return the number of rows modified. And do not RELEASE the statement + ** transaction. It needs to be rolled back. */ + if( SQLITE_OK!=(rc = sqlite3VdbeCheckFk(p, 0)) ){ + assert( db->flags&SQLITE_CountRows ); + assert( p->usesStmtJournal ); + goto abort_due_to_error; + } + + /* If the SQLITE_CountRows flag is set in sqlite3.flags mask, then + ** DML statements invoke this opcode to return the number of rows + ** modified to the user. This is the only way that a VM that + ** opens a statement transaction may invoke this opcode. + ** + ** In case this is such a statement, close any statement transaction + ** opened by this VM before returning control to the user. This is to + ** ensure that statement-transactions are always nested, not overlapping. + ** If the open statement-transaction is not closed here, then the user + ** may step another VM that opens its own statement transaction. This + ** may lead to overlapping statement transactions. + ** + ** The statement transaction is never a top-level transaction. Hence + ** the RELEASE call below can never fail. + */ + assert( p->iStatement==0 || db->flags&SQLITE_CountRows ); + rc = sqlite3VdbeCloseStatement(p, SAVEPOINT_RELEASE); + assert( rc==SQLITE_OK ); + + /* Invalidate all ephemeral cursor row caches */ + p->cacheCtr = (p->cacheCtr + 2)|1; + + /* Make sure the results of the current row are \000 terminated + ** and have an assigned type. The results are de-ephemeralized as + ** a side effect. + */ + pMem = p->pResultSet = &aMem[pOp->p1]; + for(i=0; ip2; i++){ + assert( memIsValid(&pMem[i]) ); + Deephemeralize(&pMem[i]); + assert( (pMem[i].flags & MEM_Ephem)==0 + || (pMem[i].flags & (MEM_Str|MEM_Blob))==0 ); + sqlite3VdbeMemNulTerminate(&pMem[i]); + REGISTER_TRACE(pOp->p1+i, &pMem[i]); + } + if( db->mallocFailed ) goto no_mem; + + if( db->mTrace & SQLITE_TRACE_ROW ){ + db->xTrace(SQLITE_TRACE_ROW, db->pTraceArg, p, 0); + } + + /* Return SQLITE_ROW + */ + p->pc = (int)(pOp - aOp) + 1; + rc = SQLITE_ROW; + goto vdbe_return; +} + +/* Opcode: Concat P1 P2 P3 * * +** Synopsis: r[P3]=r[P2]+r[P1] +** +** Add the text in register P1 onto the end of the text in +** register P2 and store the result in register P3. +** If either the P1 or P2 text are NULL then store NULL in P3. +** +** P3 = P2 || P1 +** +** It is illegal for P1 and P3 to be the same register. Sometimes, +** if P3 is the same register as P2, the implementation is able +** to avoid a memcpy(). +*/ +case OP_Concat: { /* same as TK_CONCAT, in1, in2, out3 */ + i64 nByte; /* Total size of the output string or blob */ + u16 flags1; /* Initial flags for P1 */ + u16 flags2; /* Initial flags for P2 */ + + pIn1 = &aMem[pOp->p1]; + pIn2 = &aMem[pOp->p2]; + pOut = &aMem[pOp->p3]; + testcase( pIn1==pIn2 ); + testcase( pOut==pIn2 ); + assert( pIn1!=pOut ); + flags1 = pIn1->flags; + testcase( flags1 & MEM_Null ); + testcase( pIn2->flags & MEM_Null ); + if( (flags1 | pIn2->flags) & MEM_Null ){ + sqlite3VdbeMemSetNull(pOut); + break; + } + if( (flags1 & (MEM_Str|MEM_Blob))==0 ){ + if( sqlite3VdbeMemStringify(pIn1,encoding,0) ) goto no_mem; + flags1 = pIn1->flags & ~MEM_Str; + }else if( (flags1 & MEM_Zero)!=0 ){ + if( sqlite3VdbeMemExpandBlob(pIn1) ) goto no_mem; + flags1 = pIn1->flags & ~MEM_Str; + } + flags2 = pIn2->flags; + if( (flags2 & (MEM_Str|MEM_Blob))==0 ){ + if( sqlite3VdbeMemStringify(pIn2,encoding,0) ) goto no_mem; + flags2 = pIn2->flags & ~MEM_Str; + }else if( (flags2 & MEM_Zero)!=0 ){ + if( sqlite3VdbeMemExpandBlob(pIn2) ) goto no_mem; + flags2 = pIn2->flags & ~MEM_Str; + } + nByte = pIn1->n + pIn2->n; + if( nByte>db->aLimit[SQLITE_LIMIT_LENGTH] ){ + goto too_big; + } + if( sqlite3VdbeMemGrow(pOut, (int)nByte+3, pOut==pIn2) ){ + goto no_mem; + } + MemSetTypeFlag(pOut, MEM_Str); + if( pOut!=pIn2 ){ + memcpy(pOut->z, pIn2->z, pIn2->n); + assert( (pIn2->flags & MEM_Dyn) == (flags2 & MEM_Dyn) ); + pIn2->flags = flags2; + } + memcpy(&pOut->z[pIn2->n], pIn1->z, pIn1->n); + assert( (pIn1->flags & MEM_Dyn) == (flags1 & MEM_Dyn) ); + pIn1->flags = flags1; + pOut->z[nByte]=0; + pOut->z[nByte+1] = 0; + pOut->z[nByte+2] = 0; + pOut->flags |= MEM_Term; + pOut->n = (int)nByte; + pOut->enc = encoding; + UPDATE_MAX_BLOBSIZE(pOut); + break; +} + +/* Opcode: Add P1 P2 P3 * * +** Synopsis: r[P3]=r[P1]+r[P2] +** +** Add the value in register P1 to the value in register P2 +** and store the result in register P3. +** If either input is NULL, the result is NULL. +*/ +/* Opcode: Multiply P1 P2 P3 * * +** Synopsis: r[P3]=r[P1]*r[P2] +** +** +** Multiply the value in register P1 by the value in register P2 +** and store the result in register P3. +** If either input is NULL, the result is NULL. +*/ +/* Opcode: Subtract P1 P2 P3 * * +** Synopsis: r[P3]=r[P2]-r[P1] +** +** Subtract the value in register P1 from the value in register P2 +** and store the result in register P3. +** If either input is NULL, the result is NULL. +*/ +/* Opcode: Divide P1 P2 P3 * * +** Synopsis: r[P3]=r[P2]/r[P1] +** +** Divide the value in register P1 by the value in register P2 +** and store the result in register P3 (P3=P2/P1). If the value in +** register P1 is zero, then the result is NULL. If either input is +** NULL, the result is NULL. +*/ +/* Opcode: Remainder P1 P2 P3 * * +** Synopsis: r[P3]=r[P2]%r[P1] +** +** Compute the remainder after integer register P2 is divided by +** register P1 and store the result in register P3. +** If the value in register P1 is zero the result is NULL. +** If either operand is NULL, the result is NULL. +*/ +case OP_Add: /* same as TK_PLUS, in1, in2, out3 */ +case OP_Subtract: /* same as TK_MINUS, in1, in2, out3 */ +case OP_Multiply: /* same as TK_STAR, in1, in2, out3 */ +case OP_Divide: /* same as TK_SLASH, in1, in2, out3 */ +case OP_Remainder: { /* same as TK_REM, in1, in2, out3 */ + u16 flags; /* Combined MEM_* flags from both inputs */ + u16 type1; /* Numeric type of left operand */ + u16 type2; /* Numeric type of right operand */ + i64 iA; /* Integer value of left operand */ + i64 iB; /* Integer value of right operand */ + double rA; /* Real value of left operand */ + double rB; /* Real value of right operand */ + + pIn1 = &aMem[pOp->p1]; + type1 = numericType(pIn1); + pIn2 = &aMem[pOp->p2]; + type2 = numericType(pIn2); + pOut = &aMem[pOp->p3]; + flags = pIn1->flags | pIn2->flags; + if( (type1 & type2 & MEM_Int)!=0 ){ + iA = pIn1->u.i; + iB = pIn2->u.i; + switch( pOp->opcode ){ + case OP_Add: if( sqlite3AddInt64(&iB,iA) ) goto fp_math; break; + case OP_Subtract: if( sqlite3SubInt64(&iB,iA) ) goto fp_math; break; + case OP_Multiply: if( sqlite3MulInt64(&iB,iA) ) goto fp_math; break; + case OP_Divide: { + if( iA==0 ) goto arithmetic_result_is_null; + if( iA==-1 && iB==SMALLEST_INT64 ) goto fp_math; + iB /= iA; + break; + } + default: { + if( iA==0 ) goto arithmetic_result_is_null; + if( iA==-1 ) iA = 1; + iB %= iA; + break; + } + } + pOut->u.i = iB; + MemSetTypeFlag(pOut, MEM_Int); + }else if( (flags & MEM_Null)!=0 ){ + goto arithmetic_result_is_null; + }else{ +fp_math: + rA = sqlite3VdbeRealValue(pIn1); + rB = sqlite3VdbeRealValue(pIn2); + switch( pOp->opcode ){ + case OP_Add: rB += rA; break; + case OP_Subtract: rB -= rA; break; + case OP_Multiply: rB *= rA; break; + case OP_Divide: { + /* (double)0 In case of SQLITE_OMIT_FLOATING_POINT... */ + if( rA==(double)0 ) goto arithmetic_result_is_null; + rB /= rA; + break; + } + default: { + iA = sqlite3VdbeIntValue(pIn1); + iB = sqlite3VdbeIntValue(pIn2); + if( iA==0 ) goto arithmetic_result_is_null; + if( iA==-1 ) iA = 1; + rB = (double)(iB % iA); + break; + } + } +#ifdef SQLITE_OMIT_FLOATING_POINT + pOut->u.i = rB; + MemSetTypeFlag(pOut, MEM_Int); +#else + if( sqlite3IsNaN(rB) ){ + goto arithmetic_result_is_null; + } + pOut->u.r = rB; + MemSetTypeFlag(pOut, MEM_Real); +#endif + } + break; + +arithmetic_result_is_null: + sqlite3VdbeMemSetNull(pOut); + break; +} + +/* Opcode: CollSeq P1 * * P4 +** +** P4 is a pointer to a CollSeq object. If the next call to a user function +** or aggregate calls sqlite3GetFuncCollSeq(), this collation sequence will +** be returned. This is used by the built-in min(), max() and nullif() +** functions. +** +** If P1 is not zero, then it is a register that a subsequent min() or +** max() aggregate will set to 1 if the current row is not the minimum or +** maximum. The P1 register is initialized to 0 by this instruction. +** +** The interface used by the implementation of the aforementioned functions +** to retrieve the collation sequence set by this opcode is not available +** publicly. Only built-in functions have access to this feature. +*/ +case OP_CollSeq: { + assert( pOp->p4type==P4_COLLSEQ ); + if( pOp->p1 ){ + sqlite3VdbeMemSetInt64(&aMem[pOp->p1], 0); + } + break; +} + +/* Opcode: BitAnd P1 P2 P3 * * +** Synopsis: r[P3]=r[P1]&r[P2] +** +** Take the bit-wise AND of the values in register P1 and P2 and +** store the result in register P3. +** If either input is NULL, the result is NULL. +*/ +/* Opcode: BitOr P1 P2 P3 * * +** Synopsis: r[P3]=r[P1]|r[P2] +** +** Take the bit-wise OR of the values in register P1 and P2 and +** store the result in register P3. +** If either input is NULL, the result is NULL. +*/ +/* Opcode: ShiftLeft P1 P2 P3 * * +** Synopsis: r[P3]=r[P2]<>r[P1] +** +** Shift the integer value in register P2 to the right by the +** number of bits specified by the integer in register P1. +** Store the result in register P3. +** If either input is NULL, the result is NULL. +*/ +case OP_BitAnd: /* same as TK_BITAND, in1, in2, out3 */ +case OP_BitOr: /* same as TK_BITOR, in1, in2, out3 */ +case OP_ShiftLeft: /* same as TK_LSHIFT, in1, in2, out3 */ +case OP_ShiftRight: { /* same as TK_RSHIFT, in1, in2, out3 */ + i64 iA; + u64 uA; + i64 iB; + u8 op; + + pIn1 = &aMem[pOp->p1]; + pIn2 = &aMem[pOp->p2]; + pOut = &aMem[pOp->p3]; + if( (pIn1->flags | pIn2->flags) & MEM_Null ){ + sqlite3VdbeMemSetNull(pOut); + break; + } + iA = sqlite3VdbeIntValue(pIn2); + iB = sqlite3VdbeIntValue(pIn1); + op = pOp->opcode; + if( op==OP_BitAnd ){ + iA &= iB; + }else if( op==OP_BitOr ){ + iA |= iB; + }else if( iB!=0 ){ + assert( op==OP_ShiftRight || op==OP_ShiftLeft ); + + /* If shifting by a negative amount, shift in the other direction */ + if( iB<0 ){ + assert( OP_ShiftRight==OP_ShiftLeft+1 ); + op = 2*OP_ShiftLeft + 1 - op; + iB = iB>(-64) ? -iB : 64; + } + + if( iB>=64 ){ + iA = (iA>=0 || op==OP_ShiftLeft) ? 0 : -1; + }else{ + memcpy(&uA, &iA, sizeof(uA)); + if( op==OP_ShiftLeft ){ + uA <<= iB; + }else{ + uA >>= iB; + /* Sign-extend on a right shift of a negative number */ + if( iA<0 ) uA |= ((((u64)0xffffffff)<<32)|0xffffffff) << (64-iB); + } + memcpy(&iA, &uA, sizeof(iA)); + } + } + pOut->u.i = iA; + MemSetTypeFlag(pOut, MEM_Int); + break; +} + +/* Opcode: AddImm P1 P2 * * * +** Synopsis: r[P1]=r[P1]+P2 +** +** Add the constant P2 to the value in register P1. +** The result is always an integer. +** +** To force any register to be an integer, just add 0. +*/ +case OP_AddImm: { /* in1 */ + pIn1 = &aMem[pOp->p1]; + memAboutToChange(p, pIn1); + sqlite3VdbeMemIntegerify(pIn1); + pIn1->u.i += pOp->p2; + break; +} + +/* Opcode: MustBeInt P1 P2 * * * +** +** Force the value in register P1 to be an integer. If the value +** in P1 is not an integer and cannot be converted into an integer +** without data loss, then jump immediately to P2, or if P2==0 +** raise an SQLITE_MISMATCH exception. +*/ +case OP_MustBeInt: { /* jump, in1 */ + pIn1 = &aMem[pOp->p1]; + if( (pIn1->flags & MEM_Int)==0 ){ + applyAffinity(pIn1, SQLITE_AFF_NUMERIC, encoding); + if( (pIn1->flags & MEM_Int)==0 ){ + VdbeBranchTaken(1, 2); + if( pOp->p2==0 ){ + rc = SQLITE_MISMATCH; + goto abort_due_to_error; + }else{ + goto jump_to_p2; + } + } + } + VdbeBranchTaken(0, 2); + MemSetTypeFlag(pIn1, MEM_Int); + break; +} + +#ifndef SQLITE_OMIT_FLOATING_POINT +/* Opcode: RealAffinity P1 * * * * +** +** If register P1 holds an integer convert it to a real value. +** +** This opcode is used when extracting information from a column that +** has REAL affinity. Such column values may still be stored as +** integers, for space efficiency, but after extraction we want them +** to have only a real value. +*/ +case OP_RealAffinity: { /* in1 */ + pIn1 = &aMem[pOp->p1]; + if( pIn1->flags & (MEM_Int|MEM_IntReal) ){ + testcase( pIn1->flags & MEM_Int ); + testcase( pIn1->flags & MEM_IntReal ); + sqlite3VdbeMemRealify(pIn1); + REGISTER_TRACE(pOp->p1, pIn1); + } + break; +} +#endif + +#ifndef SQLITE_OMIT_CAST +/* Opcode: Cast P1 P2 * * * +** Synopsis: affinity(r[P1]) +** +** Force the value in register P1 to be the type defined by P2. +** +**
      +**
    • P2=='A' → BLOB +**
    • P2=='B' → TEXT +**
    • P2=='C' → NUMERIC +**
    • P2=='D' → INTEGER +**
    • P2=='E' → REAL +**
    +** +** A NULL value is not changed by this routine. It remains NULL. +*/ +case OP_Cast: { /* in1 */ + assert( pOp->p2>=SQLITE_AFF_BLOB && pOp->p2<=SQLITE_AFF_REAL ); + testcase( pOp->p2==SQLITE_AFF_TEXT ); + testcase( pOp->p2==SQLITE_AFF_BLOB ); + testcase( pOp->p2==SQLITE_AFF_NUMERIC ); + testcase( pOp->p2==SQLITE_AFF_INTEGER ); + testcase( pOp->p2==SQLITE_AFF_REAL ); + pIn1 = &aMem[pOp->p1]; + memAboutToChange(p, pIn1); + rc = ExpandBlob(pIn1); + sqlite3VdbeMemCast(pIn1, pOp->p2, encoding); + UPDATE_MAX_BLOBSIZE(pIn1); + if( rc ) goto abort_due_to_error; + break; +} +#endif /* SQLITE_OMIT_CAST */ + +/* Opcode: Eq P1 P2 P3 P4 P5 +** Synopsis: IF r[P3]==r[P1] +** +** Compare the values in register P1 and P3. If reg(P3)==reg(P1) then +** jump to address P2. Or if the SQLITE_STOREP2 flag is set in P5, then +** store the result of comparison in register P2. +** +** The SQLITE_AFF_MASK portion of P5 must be an affinity character - +** SQLITE_AFF_TEXT, SQLITE_AFF_INTEGER, and so forth. An attempt is made +** to coerce both inputs according to this affinity before the +** comparison is made. If the SQLITE_AFF_MASK is 0x00, then numeric +** affinity is used. Note that the affinity conversions are stored +** back into the input registers P1 and P3. So this opcode can cause +** persistent changes to registers P1 and P3. +** +** Once any conversions have taken place, and neither value is NULL, +** the values are compared. If both values are blobs then memcmp() is +** used to determine the results of the comparison. If both values +** are text, then the appropriate collating function specified in +** P4 is used to do the comparison. If P4 is not specified then +** memcmp() is used to compare text string. If both values are +** numeric, then a numeric comparison is used. If the two values +** are of different types, then numbers are considered less than +** strings and strings are considered less than blobs. +** +** If SQLITE_NULLEQ is set in P5 then the result of comparison is always either +** true or false and is never NULL. If both operands are NULL then the result +** of comparison is true. If either operand is NULL then the result is false. +** If neither operand is NULL the result is the same as it would be if +** the SQLITE_NULLEQ flag were omitted from P5. +** +** If both SQLITE_STOREP2 and SQLITE_KEEPNULL flags are set then the +** content of r[P2] is only changed if the new value is NULL or 0 (false). +** In other words, a prior r[P2] value will not be overwritten by 1 (true). +*/ +/* Opcode: Ne P1 P2 P3 P4 P5 +** Synopsis: IF r[P3]!=r[P1] +** +** This works just like the Eq opcode except that the jump is taken if +** the operands in registers P1 and P3 are not equal. See the Eq opcode for +** additional information. +** +** If both SQLITE_STOREP2 and SQLITE_KEEPNULL flags are set then the +** content of r[P2] is only changed if the new value is NULL or 1 (true). +** In other words, a prior r[P2] value will not be overwritten by 0 (false). +*/ +/* Opcode: Lt P1 P2 P3 P4 P5 +** Synopsis: IF r[P3]r[P1] +** +** This works just like the Lt opcode except that the jump is taken if +** the content of register P3 is greater than the content of +** register P1. See the Lt opcode for additional information. +*/ +/* Opcode: Ge P1 P2 P3 P4 P5 +** Synopsis: IF r[P3]>=r[P1] +** +** This works just like the Lt opcode except that the jump is taken if +** the content of register P3 is greater than or equal to the content of +** register P1. See the Lt opcode for additional information. +*/ +case OP_Eq: /* same as TK_EQ, jump, in1, in3 */ +case OP_Ne: /* same as TK_NE, jump, in1, in3 */ +case OP_Lt: /* same as TK_LT, jump, in1, in3 */ +case OP_Le: /* same as TK_LE, jump, in1, in3 */ +case OP_Gt: /* same as TK_GT, jump, in1, in3 */ +case OP_Ge: { /* same as TK_GE, jump, in1, in3 */ + int res, res2; /* Result of the comparison of pIn1 against pIn3 */ + char affinity; /* Affinity to use for comparison */ + u16 flags1; /* Copy of initial value of pIn1->flags */ + u16 flags3; /* Copy of initial value of pIn3->flags */ + + pIn1 = &aMem[pOp->p1]; + pIn3 = &aMem[pOp->p3]; + flags1 = pIn1->flags; + flags3 = pIn3->flags; + if( (flags1 | flags3)&MEM_Null ){ + /* One or both operands are NULL */ + if( pOp->p5 & SQLITE_NULLEQ ){ + /* If SQLITE_NULLEQ is set (which will only happen if the operator is + ** OP_Eq or OP_Ne) then take the jump or not depending on whether + ** or not both operands are null. + */ + assert( (flags1 & MEM_Cleared)==0 ); + assert( (pOp->p5 & SQLITE_JUMPIFNULL)==0 || CORRUPT_DB ); + testcase( (pOp->p5 & SQLITE_JUMPIFNULL)!=0 ); + if( (flags1&flags3&MEM_Null)!=0 + && (flags3&MEM_Cleared)==0 + ){ + res = 0; /* Operands are equal */ + }else{ + res = ((flags3 & MEM_Null) ? -1 : +1); /* Operands are not equal */ + } + }else{ + /* SQLITE_NULLEQ is clear and at least one operand is NULL, + ** then the result is always NULL. + ** The jump is taken if the SQLITE_JUMPIFNULL bit is set. + */ + if( pOp->p5 & SQLITE_STOREP2 ){ + pOut = &aMem[pOp->p2]; + iCompare = 1; /* Operands are not equal */ + memAboutToChange(p, pOut); + MemSetTypeFlag(pOut, MEM_Null); + REGISTER_TRACE(pOp->p2, pOut); + }else{ + VdbeBranchTaken(2,3); + if( pOp->p5 & SQLITE_JUMPIFNULL ){ + goto jump_to_p2; + } + } + break; + } + }else{ + /* Neither operand is NULL. Do a comparison. */ + affinity = pOp->p5 & SQLITE_AFF_MASK; + if( affinity>=SQLITE_AFF_NUMERIC ){ + if( (flags1 | flags3)&MEM_Str ){ + if( (flags1 & (MEM_Int|MEM_IntReal|MEM_Real|MEM_Str))==MEM_Str ){ + applyNumericAffinity(pIn1,0); + assert( flags3==pIn3->flags ); + /* testcase( flags3!=pIn3->flags ); + ** this used to be possible with pIn1==pIn3, but not since + ** the column cache was removed. The following assignment + ** is essentially a no-op. But, it provides defense-in-depth + ** in case our analysis is incorrect, so it is left in. */ + flags3 = pIn3->flags; + } + if( (flags3 & (MEM_Int|MEM_IntReal|MEM_Real|MEM_Str))==MEM_Str ){ + applyNumericAffinity(pIn3,0); + } + } + /* Handle the common case of integer comparison here, as an + ** optimization, to avoid a call to sqlite3MemCompare() */ + if( (pIn1->flags & pIn3->flags & MEM_Int)!=0 ){ + if( pIn3->u.i > pIn1->u.i ){ res = +1; goto compare_op; } + if( pIn3->u.i < pIn1->u.i ){ res = -1; goto compare_op; } + res = 0; + goto compare_op; + } + }else if( affinity==SQLITE_AFF_TEXT ){ + if( (flags1 & MEM_Str)==0 && (flags1&(MEM_Int|MEM_Real|MEM_IntReal))!=0 ){ + testcase( pIn1->flags & MEM_Int ); + testcase( pIn1->flags & MEM_Real ); + testcase( pIn1->flags & MEM_IntReal ); + sqlite3VdbeMemStringify(pIn1, encoding, 1); + testcase( (flags1&MEM_Dyn) != (pIn1->flags&MEM_Dyn) ); + flags1 = (pIn1->flags & ~MEM_TypeMask) | (flags1 & MEM_TypeMask); + assert( pIn1!=pIn3 ); + } + if( (flags3 & MEM_Str)==0 && (flags3&(MEM_Int|MEM_Real|MEM_IntReal))!=0 ){ + testcase( pIn3->flags & MEM_Int ); + testcase( pIn3->flags & MEM_Real ); + testcase( pIn3->flags & MEM_IntReal ); + sqlite3VdbeMemStringify(pIn3, encoding, 1); + testcase( (flags3&MEM_Dyn) != (pIn3->flags&MEM_Dyn) ); + flags3 = (pIn3->flags & ~MEM_TypeMask) | (flags3 & MEM_TypeMask); + } + } + assert( pOp->p4type==P4_COLLSEQ || pOp->p4.pColl==0 ); + res = sqlite3MemCompare(pIn3, pIn1, pOp->p4.pColl); + } +compare_op: + /* At this point, res is negative, zero, or positive if reg[P1] is + ** less than, equal to, or greater than reg[P3], respectively. Compute + ** the answer to this operator in res2, depending on what the comparison + ** operator actually is. The next block of code depends on the fact + ** that the 6 comparison operators are consecutive integers in this + ** order: NE, EQ, GT, LE, LT, GE */ + assert( OP_Eq==OP_Ne+1 ); assert( OP_Gt==OP_Ne+2 ); assert( OP_Le==OP_Ne+3 ); + assert( OP_Lt==OP_Ne+4 ); assert( OP_Ge==OP_Ne+5 ); + if( res<0 ){ /* ne, eq, gt, le, lt, ge */ + static const unsigned char aLTb[] = { 1, 0, 0, 1, 1, 0 }; + res2 = aLTb[pOp->opcode - OP_Ne]; + }else if( res==0 ){ + static const unsigned char aEQb[] = { 0, 1, 0, 1, 0, 1 }; + res2 = aEQb[pOp->opcode - OP_Ne]; + }else{ + static const unsigned char aGTb[] = { 1, 0, 1, 0, 0, 1 }; + res2 = aGTb[pOp->opcode - OP_Ne]; + } + + /* Undo any changes made by applyAffinity() to the input registers. */ + assert( (pIn1->flags & MEM_Dyn) == (flags1 & MEM_Dyn) ); + pIn1->flags = flags1; + assert( (pIn3->flags & MEM_Dyn) == (flags3 & MEM_Dyn) ); + pIn3->flags = flags3; + + if( pOp->p5 & SQLITE_STOREP2 ){ + pOut = &aMem[pOp->p2]; + iCompare = res; + if( (pOp->p5 & SQLITE_KEEPNULL)!=0 ){ + /* The KEEPNULL flag prevents OP_Eq from overwriting a NULL with 1 + ** and prevents OP_Ne from overwriting NULL with 0. This flag + ** is only used in contexts where either: + ** (1) op==OP_Eq && (r[P2]==NULL || r[P2]==0) + ** (2) op==OP_Ne && (r[P2]==NULL || r[P2]==1) + ** Therefore it is not necessary to check the content of r[P2] for + ** NULL. */ + assert( pOp->opcode==OP_Ne || pOp->opcode==OP_Eq ); + assert( res2==0 || res2==1 ); + testcase( res2==0 && pOp->opcode==OP_Eq ); + testcase( res2==1 && pOp->opcode==OP_Eq ); + testcase( res2==0 && pOp->opcode==OP_Ne ); + testcase( res2==1 && pOp->opcode==OP_Ne ); + if( (pOp->opcode==OP_Eq)==res2 ) break; + } + memAboutToChange(p, pOut); + MemSetTypeFlag(pOut, MEM_Int); + pOut->u.i = res2; + REGISTER_TRACE(pOp->p2, pOut); + }else{ + VdbeBranchTaken(res2!=0, (pOp->p5 & SQLITE_NULLEQ)?2:3); + if( res2 ){ + goto jump_to_p2; + } + } + break; +} + +/* Opcode: ElseNotEq * P2 * * * +** +** This opcode must immediately follow an OP_Lt or OP_Gt comparison operator. +** If result of an OP_Eq comparison on the same two operands +** would have be NULL or false (0), then then jump to P2. +** If the result of an OP_Eq comparison on the two previous operands +** would have been true (1), then fall through. +*/ +case OP_ElseNotEq: { /* same as TK_ESCAPE, jump */ + assert( pOp>aOp ); + assert( pOp[-1].opcode==OP_Lt || pOp[-1].opcode==OP_Gt ); + assert( pOp[-1].p5 & SQLITE_STOREP2 ); + VdbeBranchTaken(iCompare!=0, 2); + if( iCompare!=0 ) goto jump_to_p2; + break; +} + + +/* Opcode: Permutation * * * P4 * +** +** Set the permutation used by the OP_Compare operator in the next +** instruction. The permutation is stored in the P4 operand. +** +** The permutation is only valid until the next OP_Compare that has +** the OPFLAG_PERMUTE bit set in P5. Typically the OP_Permutation should +** occur immediately prior to the OP_Compare. +** +** The first integer in the P4 integer array is the length of the array +** and does not become part of the permutation. +*/ +case OP_Permutation: { + assert( pOp->p4type==P4_INTARRAY ); + assert( pOp->p4.ai ); + assert( pOp[1].opcode==OP_Compare ); + assert( pOp[1].p5 & OPFLAG_PERMUTE ); + break; +} + +/* Opcode: Compare P1 P2 P3 P4 P5 +** Synopsis: r[P1@P3] <-> r[P2@P3] +** +** Compare two vectors of registers in reg(P1)..reg(P1+P3-1) (call this +** vector "A") and in reg(P2)..reg(P2+P3-1) ("B"). Save the result of +** the comparison for use by the next OP_Jump instruct. +** +** If P5 has the OPFLAG_PERMUTE bit set, then the order of comparison is +** determined by the most recent OP_Permutation operator. If the +** OPFLAG_PERMUTE bit is clear, then register are compared in sequential +** order. +** +** P4 is a KeyInfo structure that defines collating sequences and sort +** orders for the comparison. The permutation applies to registers +** only. The KeyInfo elements are used sequentially. +** +** The comparison is a sort comparison, so NULLs compare equal, +** NULLs are less than numbers, numbers are less than strings, +** and strings are less than blobs. +*/ +case OP_Compare: { + int n; + int i; + int p1; + int p2; + const KeyInfo *pKeyInfo; + int idx; + CollSeq *pColl; /* Collating sequence to use on this term */ + int bRev; /* True for DESCENDING sort order */ + int *aPermute; /* The permutation */ + + if( (pOp->p5 & OPFLAG_PERMUTE)==0 ){ + aPermute = 0; + }else{ + assert( pOp>aOp ); + assert( pOp[-1].opcode==OP_Permutation ); + assert( pOp[-1].p4type==P4_INTARRAY ); + aPermute = pOp[-1].p4.ai + 1; + assert( aPermute!=0 ); + } + n = pOp->p3; + pKeyInfo = pOp->p4.pKeyInfo; + assert( n>0 ); + assert( pKeyInfo!=0 ); + p1 = pOp->p1; + p2 = pOp->p2; +#ifdef SQLITE_DEBUG + if( aPermute ){ + int k, mx = 0; + for(k=0; kmx ) mx = aPermute[k]; + assert( p1>0 && p1+mx<=(p->nMem+1 - p->nCursor)+1 ); + assert( p2>0 && p2+mx<=(p->nMem+1 - p->nCursor)+1 ); + }else{ + assert( p1>0 && p1+n<=(p->nMem+1 - p->nCursor)+1 ); + assert( p2>0 && p2+n<=(p->nMem+1 - p->nCursor)+1 ); + } +#endif /* SQLITE_DEBUG */ + for(i=0; inKeyField ); + pColl = pKeyInfo->aColl[i]; + bRev = (pKeyInfo->aSortFlags[i] & KEYINFO_ORDER_DESC); + iCompare = sqlite3MemCompare(&aMem[p1+idx], &aMem[p2+idx], pColl); + if( iCompare ){ + if( (pKeyInfo->aSortFlags[i] & KEYINFO_ORDER_BIGNULL) + && ((aMem[p1+idx].flags & MEM_Null) || (aMem[p2+idx].flags & MEM_Null)) + ){ + iCompare = -iCompare; + } + if( bRev ) iCompare = -iCompare; + break; + } + } + break; +} + +/* Opcode: Jump P1 P2 P3 * * +** +** Jump to the instruction at address P1, P2, or P3 depending on whether +** in the most recent OP_Compare instruction the P1 vector was less than +** equal to, or greater than the P2 vector, respectively. +*/ +case OP_Jump: { /* jump */ + if( iCompare<0 ){ + VdbeBranchTaken(0,4); pOp = &aOp[pOp->p1 - 1]; + }else if( iCompare==0 ){ + VdbeBranchTaken(1,4); pOp = &aOp[pOp->p2 - 1]; + }else{ + VdbeBranchTaken(2,4); pOp = &aOp[pOp->p3 - 1]; + } + break; +} + +/* Opcode: And P1 P2 P3 * * +** Synopsis: r[P3]=(r[P1] && r[P2]) +** +** Take the logical AND of the values in registers P1 and P2 and +** write the result into register P3. +** +** If either P1 or P2 is 0 (false) then the result is 0 even if +** the other input is NULL. A NULL and true or two NULLs give +** a NULL output. +*/ +/* Opcode: Or P1 P2 P3 * * +** Synopsis: r[P3]=(r[P1] || r[P2]) +** +** Take the logical OR of the values in register P1 and P2 and +** store the answer in register P3. +** +** If either P1 or P2 is nonzero (true) then the result is 1 (true) +** even if the other input is NULL. A NULL and false or two NULLs +** give a NULL output. +*/ +case OP_And: /* same as TK_AND, in1, in2, out3 */ +case OP_Or: { /* same as TK_OR, in1, in2, out3 */ + int v1; /* Left operand: 0==FALSE, 1==TRUE, 2==UNKNOWN or NULL */ + int v2; /* Right operand: 0==FALSE, 1==TRUE, 2==UNKNOWN or NULL */ + + v1 = sqlite3VdbeBooleanValue(&aMem[pOp->p1], 2); + v2 = sqlite3VdbeBooleanValue(&aMem[pOp->p2], 2); + if( pOp->opcode==OP_And ){ + static const unsigned char and_logic[] = { 0, 0, 0, 0, 1, 2, 0, 2, 2 }; + v1 = and_logic[v1*3+v2]; + }else{ + static const unsigned char or_logic[] = { 0, 1, 2, 1, 1, 1, 2, 1, 2 }; + v1 = or_logic[v1*3+v2]; + } + pOut = &aMem[pOp->p3]; + if( v1==2 ){ + MemSetTypeFlag(pOut, MEM_Null); + }else{ + pOut->u.i = v1; + MemSetTypeFlag(pOut, MEM_Int); + } + break; +} + +/* Opcode: IsTrue P1 P2 P3 P4 * +** Synopsis: r[P2] = coalesce(r[P1]==TRUE,P3) ^ P4 +** +** This opcode implements the IS TRUE, IS FALSE, IS NOT TRUE, and +** IS NOT FALSE operators. +** +** Interpret the value in register P1 as a boolean value. Store that +** boolean (a 0 or 1) in register P2. Or if the value in register P1 is +** NULL, then the P3 is stored in register P2. Invert the answer if P4 +** is 1. +** +** The logic is summarized like this: +** +**
      +**
    • If P3==0 and P4==0 then r[P2] := r[P1] IS TRUE +**
    • If P3==1 and P4==1 then r[P2] := r[P1] IS FALSE +**
    • If P3==0 and P4==1 then r[P2] := r[P1] IS NOT TRUE +**
    • If P3==1 and P4==0 then r[P2] := r[P1] IS NOT FALSE +**
    +*/ +case OP_IsTrue: { /* in1, out2 */ + assert( pOp->p4type==P4_INT32 ); + assert( pOp->p4.i==0 || pOp->p4.i==1 ); + assert( pOp->p3==0 || pOp->p3==1 ); + sqlite3VdbeMemSetInt64(&aMem[pOp->p2], + sqlite3VdbeBooleanValue(&aMem[pOp->p1], pOp->p3) ^ pOp->p4.i); + break; +} + +/* Opcode: Not P1 P2 * * * +** Synopsis: r[P2]= !r[P1] +** +** Interpret the value in register P1 as a boolean value. Store the +** boolean complement in register P2. If the value in register P1 is +** NULL, then a NULL is stored in P2. +*/ +case OP_Not: { /* same as TK_NOT, in1, out2 */ + pIn1 = &aMem[pOp->p1]; + pOut = &aMem[pOp->p2]; + if( (pIn1->flags & MEM_Null)==0 ){ + sqlite3VdbeMemSetInt64(pOut, !sqlite3VdbeBooleanValue(pIn1,0)); + }else{ + sqlite3VdbeMemSetNull(pOut); + } + break; +} + +/* Opcode: BitNot P1 P2 * * * +** Synopsis: r[P2]= ~r[P1] +** +** Interpret the content of register P1 as an integer. Store the +** ones-complement of the P1 value into register P2. If P1 holds +** a NULL then store a NULL in P2. +*/ +case OP_BitNot: { /* same as TK_BITNOT, in1, out2 */ + pIn1 = &aMem[pOp->p1]; + pOut = &aMem[pOp->p2]; + sqlite3VdbeMemSetNull(pOut); + if( (pIn1->flags & MEM_Null)==0 ){ + pOut->flags = MEM_Int; + pOut->u.i = ~sqlite3VdbeIntValue(pIn1); + } + break; +} + +/* Opcode: Once P1 P2 * * * +** +** Fall through to the next instruction the first time this opcode is +** encountered on each invocation of the byte-code program. Jump to P2 +** on the second and all subsequent encounters during the same invocation. +** +** Top-level programs determine first invocation by comparing the P1 +** operand against the P1 operand on the OP_Init opcode at the beginning +** of the program. If the P1 values differ, then fall through and make +** the P1 of this opcode equal to the P1 of OP_Init. If P1 values are +** the same then take the jump. +** +** For subprograms, there is a bitmask in the VdbeFrame that determines +** whether or not the jump should be taken. The bitmask is necessary +** because the self-altering code trick does not work for recursive +** triggers. +*/ +case OP_Once: { /* jump */ + u32 iAddr; /* Address of this instruction */ + assert( p->aOp[0].opcode==OP_Init ); + if( p->pFrame ){ + iAddr = (int)(pOp - p->aOp); + if( (p->pFrame->aOnce[iAddr/8] & (1<<(iAddr & 7)))!=0 ){ + VdbeBranchTaken(1, 2); + goto jump_to_p2; + } + p->pFrame->aOnce[iAddr/8] |= 1<<(iAddr & 7); + }else{ + if( p->aOp[0].p1==pOp->p1 ){ + VdbeBranchTaken(1, 2); + goto jump_to_p2; + } + } + VdbeBranchTaken(0, 2); + pOp->p1 = p->aOp[0].p1; + break; +} + +/* Opcode: If P1 P2 P3 * * +** +** Jump to P2 if the value in register P1 is true. The value +** is considered true if it is numeric and non-zero. If the value +** in P1 is NULL then take the jump if and only if P3 is non-zero. +*/ +case OP_If: { /* jump, in1 */ + int c; + c = sqlite3VdbeBooleanValue(&aMem[pOp->p1], pOp->p3); + VdbeBranchTaken(c!=0, 2); + if( c ) goto jump_to_p2; + break; +} + +/* Opcode: IfNot P1 P2 P3 * * +** +** Jump to P2 if the value in register P1 is False. The value +** is considered false if it has a numeric value of zero. If the value +** in P1 is NULL then take the jump if and only if P3 is non-zero. +*/ +case OP_IfNot: { /* jump, in1 */ + int c; + c = !sqlite3VdbeBooleanValue(&aMem[pOp->p1], !pOp->p3); + VdbeBranchTaken(c!=0, 2); + if( c ) goto jump_to_p2; + break; +} + +/* Opcode: IsNull P1 P2 * * * +** Synopsis: if r[P1]==NULL goto P2 +** +** Jump to P2 if the value in register P1 is NULL. +*/ +case OP_IsNull: { /* same as TK_ISNULL, jump, in1 */ + pIn1 = &aMem[pOp->p1]; + VdbeBranchTaken( (pIn1->flags & MEM_Null)!=0, 2); + if( (pIn1->flags & MEM_Null)!=0 ){ + goto jump_to_p2; + } + break; +} + +/* Opcode: NotNull P1 P2 * * * +** Synopsis: if r[P1]!=NULL goto P2 +** +** Jump to P2 if the value in register P1 is not NULL. +*/ +case OP_NotNull: { /* same as TK_NOTNULL, jump, in1 */ + pIn1 = &aMem[pOp->p1]; + VdbeBranchTaken( (pIn1->flags & MEM_Null)==0, 2); + if( (pIn1->flags & MEM_Null)==0 ){ + goto jump_to_p2; + } + break; +} + +/* Opcode: IfNullRow P1 P2 P3 * * +** Synopsis: if P1.nullRow then r[P3]=NULL, goto P2 +** +** Check the cursor P1 to see if it is currently pointing at a NULL row. +** If it is, then set register P3 to NULL and jump immediately to P2. +** If P1 is not on a NULL row, then fall through without making any +** changes. +*/ +case OP_IfNullRow: { /* jump */ + assert( pOp->p1>=0 && pOp->p1nCursor ); + assert( p->apCsr[pOp->p1]!=0 ); + if( p->apCsr[pOp->p1]->nullRow ){ + sqlite3VdbeMemSetNull(aMem + pOp->p3); + goto jump_to_p2; + } + break; +} + +#ifdef SQLITE_ENABLE_OFFSET_SQL_FUNC +/* Opcode: Offset P1 P2 P3 * * +** Synopsis: r[P3] = sqlite_offset(P1) +** +** Store in register r[P3] the byte offset into the database file that is the +** start of the payload for the record at which that cursor P1 is currently +** pointing. +** +** P2 is the column number for the argument to the sqlite_offset() function. +** This opcode does not use P2 itself, but the P2 value is used by the +** code generator. The P1, P2, and P3 operands to this opcode are the +** same as for OP_Column. +** +** This opcode is only available if SQLite is compiled with the +** -DSQLITE_ENABLE_OFFSET_SQL_FUNC option. +*/ +case OP_Offset: { /* out3 */ + VdbeCursor *pC; /* The VDBE cursor */ + assert( pOp->p1>=0 && pOp->p1nCursor ); + pC = p->apCsr[pOp->p1]; + pOut = &p->aMem[pOp->p3]; + if( NEVER(pC==0) || pC->eCurType!=CURTYPE_BTREE ){ + sqlite3VdbeMemSetNull(pOut); + }else{ + sqlite3VdbeMemSetInt64(pOut, sqlite3BtreeOffset(pC->uc.pCursor)); + } + break; +} +#endif /* SQLITE_ENABLE_OFFSET_SQL_FUNC */ + +/* Opcode: Column P1 P2 P3 P4 P5 +** Synopsis: r[P3]=PX +** +** Interpret the data that cursor P1 points to as a structure built using +** the MakeRecord instruction. (See the MakeRecord opcode for additional +** information about the format of the data.) Extract the P2-th column +** from this record. If there are less that (P2+1) +** values in the record, extract a NULL. +** +** The value extracted is stored in register P3. +** +** If the record contains fewer than P2 fields, then extract a NULL. Or, +** if the P4 argument is a P4_MEM use the value of the P4 argument as +** the result. +** +** If the OPFLAG_LENGTHARG and OPFLAG_TYPEOFARG bits are set on P5 then +** the result is guaranteed to only be used as the argument of a length() +** or typeof() function, respectively. The loading of large blobs can be +** skipped for length() and all content loading can be skipped for typeof(). +*/ +case OP_Column: { + int p2; /* column number to retrieve */ + VdbeCursor *pC; /* The VDBE cursor */ + BtCursor *pCrsr; /* The BTree cursor */ + u32 *aOffset; /* aOffset[i] is offset to start of data for i-th column */ + int len; /* The length of the serialized data for the column */ + int i; /* Loop counter */ + Mem *pDest; /* Where to write the extracted value */ + Mem sMem; /* For storing the record being decoded */ + const u8 *zData; /* Part of the record being decoded */ + const u8 *zHdr; /* Next unparsed byte of the header */ + const u8 *zEndHdr; /* Pointer to first byte after the header */ + u64 offset64; /* 64-bit offset */ + u32 t; /* A type code from the record header */ + Mem *pReg; /* PseudoTable input register */ + + pC = p->apCsr[pOp->p1]; + p2 = pOp->p2; + + /* If the cursor cache is stale (meaning it is not currently point at + ** the correct row) then bring it up-to-date by doing the necessary + ** B-Tree seek. */ + rc = sqlite3VdbeCursorMoveto(&pC, &p2); + if( rc ) goto abort_due_to_error; + + assert( pOp->p3>0 && pOp->p3<=(p->nMem+1 - p->nCursor) ); + pDest = &aMem[pOp->p3]; + memAboutToChange(p, pDest); + assert( pOp->p1>=0 && pOp->p1nCursor ); + assert( pC!=0 ); + assert( p2nField ); + aOffset = pC->aOffset; + assert( pC->eCurType!=CURTYPE_VTAB ); + assert( pC->eCurType!=CURTYPE_PSEUDO || pC->nullRow ); + assert( pC->eCurType!=CURTYPE_SORTER ); + + if( pC->cacheStatus!=p->cacheCtr ){ /*OPTIMIZATION-IF-FALSE*/ + if( pC->nullRow ){ + if( pC->eCurType==CURTYPE_PSEUDO ){ + /* For the special case of as pseudo-cursor, the seekResult field + ** identifies the register that holds the record */ + assert( pC->seekResult>0 ); + pReg = &aMem[pC->seekResult]; + assert( pReg->flags & MEM_Blob ); + assert( memIsValid(pReg) ); + pC->payloadSize = pC->szRow = pReg->n; + pC->aRow = (u8*)pReg->z; + }else{ + sqlite3VdbeMemSetNull(pDest); + goto op_column_out; + } + }else{ + pCrsr = pC->uc.pCursor; + assert( pC->eCurType==CURTYPE_BTREE ); + assert( pCrsr ); + assert( sqlite3BtreeCursorIsValid(pCrsr) ); + pC->payloadSize = sqlite3BtreePayloadSize(pCrsr); + pC->aRow = sqlite3BtreePayloadFetch(pCrsr, &pC->szRow); + assert( pC->szRow<=pC->payloadSize ); + assert( pC->szRow<=65536 ); /* Maximum page size is 64KiB */ + if( pC->payloadSize > (u32)db->aLimit[SQLITE_LIMIT_LENGTH] ){ + goto too_big; + } + } + pC->cacheStatus = p->cacheCtr; + pC->iHdrOffset = getVarint32(pC->aRow, aOffset[0]); + pC->nHdrParsed = 0; + + + if( pC->szRowaRow does not have to hold the entire row, but it does at least + ** need to cover the header of the record. If pC->aRow does not contain + ** the complete header, then set it to zero, forcing the header to be + ** dynamically allocated. */ + pC->aRow = 0; + pC->szRow = 0; + + /* Make sure a corrupt database has not given us an oversize header. + ** Do this now to avoid an oversize memory allocation. + ** + ** Type entries can be between 1 and 5 bytes each. But 4 and 5 byte + ** types use so much data space that there can only be 4096 and 32 of + ** them, respectively. So the maximum header length results from a + ** 3-byte type for each of the maximum of 32768 columns plus three + ** extra bytes for the header length itself. 32768*3 + 3 = 98307. + */ + if( aOffset[0] > 98307 || aOffset[0] > pC->payloadSize ){ + goto op_column_corrupt; + } + }else{ + /* This is an optimization. By skipping over the first few tests + ** (ex: pC->nHdrParsed<=p2) in the next section, we achieve a + ** measurable performance gain. + ** + ** This branch is taken even if aOffset[0]==0. Such a record is never + ** generated by SQLite, and could be considered corruption, but we + ** accept it for historical reasons. When aOffset[0]==0, the code this + ** branch jumps to reads past the end of the record, but never more + ** than a few bytes. Even if the record occurs at the end of the page + ** content area, the "page header" comes after the page content and so + ** this overread is harmless. Similar overreads can occur for a corrupt + ** database file. + */ + zData = pC->aRow; + assert( pC->nHdrParsed<=p2 ); /* Conditional skipped */ + testcase( aOffset[0]==0 ); + goto op_column_read_header; + } + } + + /* Make sure at least the first p2+1 entries of the header have been + ** parsed and valid information is in aOffset[] and pC->aType[]. + */ + if( pC->nHdrParsed<=p2 ){ + /* If there is more header available for parsing in the record, try + ** to extract additional fields up through the p2+1-th field + */ + if( pC->iHdrOffsetaRow==0 ){ + memset(&sMem, 0, sizeof(sMem)); + rc = sqlite3VdbeMemFromBtree(pC->uc.pCursor, 0, aOffset[0], &sMem); + if( rc!=SQLITE_OK ) goto abort_due_to_error; + zData = (u8*)sMem.z; + }else{ + zData = pC->aRow; + } + + /* Fill in pC->aType[i] and aOffset[i] values through the p2-th field. */ + op_column_read_header: + i = pC->nHdrParsed; + offset64 = aOffset[i]; + zHdr = zData + pC->iHdrOffset; + zEndHdr = zData + aOffset[0]; + testcase( zHdr>=zEndHdr ); + do{ + if( (pC->aType[i] = t = zHdr[0])<0x80 ){ + zHdr++; + offset64 += sqlite3VdbeOneByteSerialTypeLen(t); + }else{ + zHdr += sqlite3GetVarint32(zHdr, &t); + pC->aType[i] = t; + offset64 += sqlite3VdbeSerialTypeLen(t); + } + aOffset[++i] = (u32)(offset64 & 0xffffffff); + }while( i<=p2 && zHdr=zEndHdr && (zHdr>zEndHdr || offset64!=pC->payloadSize)) + || (offset64 > pC->payloadSize) + ){ + if( aOffset[0]==0 ){ + i = 0; + zHdr = zEndHdr; + }else{ + if( pC->aRow==0 ) sqlite3VdbeMemRelease(&sMem); + goto op_column_corrupt; + } + } + + pC->nHdrParsed = i; + pC->iHdrOffset = (u32)(zHdr - zData); + if( pC->aRow==0 ) sqlite3VdbeMemRelease(&sMem); + }else{ + t = 0; + } + + /* If after trying to extract new entries from the header, nHdrParsed is + ** still not up to p2, that means that the record has fewer than p2 + ** columns. So the result will be either the default value or a NULL. + */ + if( pC->nHdrParsed<=p2 ){ + if( pOp->p4type==P4_MEM ){ + sqlite3VdbeMemShallowCopy(pDest, pOp->p4.pMem, MEM_Static); + }else{ + sqlite3VdbeMemSetNull(pDest); + } + goto op_column_out; + } + }else{ + t = pC->aType[p2]; + } + + /* Extract the content for the p2+1-th column. Control can only + ** reach this point if aOffset[p2], aOffset[p2+1], and pC->aType[p2] are + ** all valid. + */ + assert( p2nHdrParsed ); + assert( rc==SQLITE_OK ); + assert( sqlite3VdbeCheckMemInvariants(pDest) ); + if( VdbeMemDynamic(pDest) ){ + sqlite3VdbeMemSetNull(pDest); + } + assert( t==pC->aType[p2] ); + if( pC->szRow>=aOffset[p2+1] ){ + /* This is the common case where the desired content fits on the original + ** page - where the content is not on an overflow page */ + zData = pC->aRow + aOffset[p2]; + if( t<12 ){ + sqlite3VdbeSerialGet(zData, t, pDest); + }else{ + /* If the column value is a string, we need a persistent value, not + ** a MEM_Ephem value. This branch is a fast short-cut that is equivalent + ** to calling sqlite3VdbeSerialGet() and sqlite3VdbeDeephemeralize(). + */ + static const u16 aFlag[] = { MEM_Blob, MEM_Str|MEM_Term }; + pDest->n = len = (t-12)/2; + pDest->enc = encoding; + if( pDest->szMalloc < len+2 ){ + pDest->flags = MEM_Null; + if( sqlite3VdbeMemGrow(pDest, len+2, 0) ) goto no_mem; + }else{ + pDest->z = pDest->zMalloc; + } + memcpy(pDest->z, zData, len); + pDest->z[len] = 0; + pDest->z[len+1] = 0; + pDest->flags = aFlag[t&1]; + } + }else{ + pDest->enc = encoding; + /* This branch happens only when content is on overflow pages */ + if( ((pOp->p5 & (OPFLAG_LENGTHARG|OPFLAG_TYPEOFARG))!=0 + && ((t>=12 && (t&1)==0) || (pOp->p5 & OPFLAG_TYPEOFARG)!=0)) + || (len = sqlite3VdbeSerialTypeLen(t))==0 + ){ + /* Content is irrelevant for + ** 1. the typeof() function, + ** 2. the length(X) function if X is a blob, and + ** 3. if the content length is zero. + ** So we might as well use bogus content rather than reading + ** content from disk. + ** + ** Although sqlite3VdbeSerialGet() may read at most 8 bytes from the + ** buffer passed to it, debugging function VdbeMemPrettyPrint() may + ** read up to 16. So 16 bytes of bogus content is supplied. + */ + static u8 aZero[16]; /* This is the bogus content */ + sqlite3VdbeSerialGet(aZero, t, pDest); + }else{ + rc = sqlite3VdbeMemFromBtree(pC->uc.pCursor, aOffset[p2], len, pDest); + if( rc!=SQLITE_OK ) goto abort_due_to_error; + sqlite3VdbeSerialGet((const u8*)pDest->z, t, pDest); + pDest->flags &= ~MEM_Ephem; + } + } + +op_column_out: + UPDATE_MAX_BLOBSIZE(pDest); + REGISTER_TRACE(pOp->p3, pDest); + break; + +op_column_corrupt: + if( aOp[0].p3>0 ){ + pOp = &aOp[aOp[0].p3-1]; + break; + }else{ + rc = SQLITE_CORRUPT_BKPT; + goto abort_due_to_error; + } +} + +/* Opcode: Affinity P1 P2 * P4 * +** Synopsis: affinity(r[P1@P2]) +** +** Apply affinities to a range of P2 registers starting with P1. +** +** P4 is a string that is P2 characters long. The N-th character of the +** string indicates the column affinity that should be used for the N-th +** memory cell in the range. +*/ +case OP_Affinity: { + const char *zAffinity; /* The affinity to be applied */ + + zAffinity = pOp->p4.z; + assert( zAffinity!=0 ); + assert( pOp->p2>0 ); + assert( zAffinity[pOp->p2]==0 ); + pIn1 = &aMem[pOp->p1]; + while( 1 /*exit-by-break*/ ){ + assert( pIn1 <= &p->aMem[(p->nMem+1 - p->nCursor)] ); + assert( memIsValid(pIn1) ); + applyAffinity(pIn1, zAffinity[0], encoding); + if( zAffinity[0]==SQLITE_AFF_REAL && (pIn1->flags & MEM_Int)!=0 ){ + /* When applying REAL affinity, if the result is still an MEM_Int + ** that will fit in 6 bytes, then change the type to MEM_IntReal + ** so that we keep the high-resolution integer value but know that + ** the type really wants to be REAL. */ + testcase( pIn1->u.i==140737488355328LL ); + testcase( pIn1->u.i==140737488355327LL ); + testcase( pIn1->u.i==-140737488355328LL ); + testcase( pIn1->u.i==-140737488355329LL ); + if( pIn1->u.i<=140737488355327LL && pIn1->u.i>=-140737488355328LL ){ + pIn1->flags |= MEM_IntReal; + pIn1->flags &= ~MEM_Int; + }else{ + pIn1->u.r = (double)pIn1->u.i; + pIn1->flags |= MEM_Real; + pIn1->flags &= ~MEM_Int; + } + } + REGISTER_TRACE((int)(pIn1-aMem), pIn1); + zAffinity++; + if( zAffinity[0]==0 ) break; + pIn1++; + } + break; +} + +/* Opcode: MakeRecord P1 P2 P3 P4 * +** Synopsis: r[P3]=mkrec(r[P1@P2]) +** +** Convert P2 registers beginning with P1 into the [record format] +** use as a data record in a database table or as a key +** in an index. The OP_Column opcode can decode the record later. +** +** P4 may be a string that is P2 characters long. The N-th character of the +** string indicates the column affinity that should be used for the N-th +** field of the index key. +** +** The mapping from character to affinity is given by the SQLITE_AFF_ +** macros defined in sqliteInt.h. +** +** If P4 is NULL then all index fields have the affinity BLOB. +*/ +case OP_MakeRecord: { + Mem *pRec; /* The new record */ + u64 nData; /* Number of bytes of data space */ + int nHdr; /* Number of bytes of header space */ + i64 nByte; /* Data space required for this record */ + i64 nZero; /* Number of zero bytes at the end of the record */ + int nVarint; /* Number of bytes in a varint */ + u32 serial_type; /* Type field */ + Mem *pData0; /* First field to be combined into the record */ + Mem *pLast; /* Last field of the record */ + int nField; /* Number of fields in the record */ + char *zAffinity; /* The affinity string for the record */ + int file_format; /* File format to use for encoding */ + u32 len; /* Length of a field */ + u8 *zHdr; /* Where to write next byte of the header */ + u8 *zPayload; /* Where to write next byte of the payload */ + + /* Assuming the record contains N fields, the record format looks + ** like this: + ** + ** ------------------------------------------------------------------------ + ** | hdr-size | type 0 | type 1 | ... | type N-1 | data0 | ... | data N-1 | + ** ------------------------------------------------------------------------ + ** + ** Data(0) is taken from register P1. Data(1) comes from register P1+1 + ** and so forth. + ** + ** Each type field is a varint representing the serial type of the + ** corresponding data element (see sqlite3VdbeSerialType()). The + ** hdr-size field is also a varint which is the offset from the beginning + ** of the record to data0. + */ + nData = 0; /* Number of bytes of data space */ + nHdr = 0; /* Number of bytes of header space */ + nZero = 0; /* Number of zero bytes at the end of the record */ + nField = pOp->p1; + zAffinity = pOp->p4.z; + assert( nField>0 && pOp->p2>0 && pOp->p2+nField<=(p->nMem+1 - p->nCursor)+1 ); + pData0 = &aMem[nField]; + nField = pOp->p2; + pLast = &pData0[nField-1]; + file_format = p->minWriteFileFormat; + + /* Identify the output register */ + assert( pOp->p3p1 || pOp->p3>=pOp->p1+pOp->p2 ); + pOut = &aMem[pOp->p3]; + memAboutToChange(p, pOut); + + /* Apply the requested affinity to all inputs + */ + assert( pData0<=pLast ); + if( zAffinity ){ + pRec = pData0; + do{ + applyAffinity(pRec, zAffinity[0], encoding); + if( zAffinity[0]==SQLITE_AFF_REAL && (pRec->flags & MEM_Int) ){ + pRec->flags |= MEM_IntReal; + pRec->flags &= ~(MEM_Int); + } + REGISTER_TRACE((int)(pRec-aMem), pRec); + zAffinity++; + pRec++; + assert( zAffinity[0]==0 || pRec<=pLast ); + }while( zAffinity[0] ); + } + +#ifdef SQLITE_ENABLE_NULL_TRIM + /* NULLs can be safely trimmed from the end of the record, as long as + ** as the schema format is 2 or more and none of the omitted columns + ** have a non-NULL default value. Also, the record must be left with + ** at least one field. If P5>0 then it will be one more than the + ** index of the right-most column with a non-NULL default value */ + if( pOp->p5 ){ + while( (pLast->flags & MEM_Null)!=0 && nField>pOp->p5 ){ + pLast--; + nField--; + } + } +#endif + + /* Loop through the elements that will make up the record to figure + ** out how much space is required for the new record. After this loop, + ** the Mem.uTemp field of each term should hold the serial-type that will + ** be used for that term in the generated record: + ** + ** Mem.uTemp value type + ** --------------- --------------- + ** 0 NULL + ** 1 1-byte signed integer + ** 2 2-byte signed integer + ** 3 3-byte signed integer + ** 4 4-byte signed integer + ** 5 6-byte signed integer + ** 6 8-byte signed integer + ** 7 IEEE float + ** 8 Integer constant 0 + ** 9 Integer constant 1 + ** 10,11 reserved for expansion + ** N>=12 and even BLOB + ** N>=13 and odd text + ** + ** The following additional values are computed: + ** nHdr Number of bytes needed for the record header + ** nData Number of bytes of data space needed for the record + ** nZero Zero bytes at the end of the record + */ + pRec = pLast; + do{ + assert( memIsValid(pRec) ); + if( pRec->flags & MEM_Null ){ + if( pRec->flags & MEM_Zero ){ + /* Values with MEM_Null and MEM_Zero are created by xColumn virtual + ** table methods that never invoke sqlite3_result_xxxxx() while + ** computing an unchanging column value in an UPDATE statement. + ** Give such values a special internal-use-only serial-type of 10 + ** so that they can be passed through to xUpdate and have + ** a true sqlite3_value_nochange(). */ + assert( pOp->p5==OPFLAG_NOCHNG_MAGIC || CORRUPT_DB ); + pRec->uTemp = 10; + }else{ + pRec->uTemp = 0; + } + nHdr++; + }else if( pRec->flags & (MEM_Int|MEM_IntReal) ){ + /* Figure out whether to use 1, 2, 4, 6 or 8 bytes. */ + i64 i = pRec->u.i; + u64 uu; + testcase( pRec->flags & MEM_Int ); + testcase( pRec->flags & MEM_IntReal ); + if( i<0 ){ + uu = ~i; + }else{ + uu = i; + } + nHdr++; + testcase( uu==127 ); testcase( uu==128 ); + testcase( uu==32767 ); testcase( uu==32768 ); + testcase( uu==8388607 ); testcase( uu==8388608 ); + testcase( uu==2147483647 ); testcase( uu==2147483648 ); + testcase( uu==140737488355327LL ); testcase( uu==140737488355328LL ); + if( uu<=127 ){ + if( (i&1)==i && file_format>=4 ){ + pRec->uTemp = 8+(u32)uu; + }else{ + nData++; + pRec->uTemp = 1; + } + }else if( uu<=32767 ){ + nData += 2; + pRec->uTemp = 2; + }else if( uu<=8388607 ){ + nData += 3; + pRec->uTemp = 3; + }else if( uu<=2147483647 ){ + nData += 4; + pRec->uTemp = 4; + }else if( uu<=140737488355327LL ){ + nData += 6; + pRec->uTemp = 5; + }else{ + nData += 8; + if( pRec->flags & MEM_IntReal ){ + /* If the value is IntReal and is going to take up 8 bytes to store + ** as an integer, then we might as well make it an 8-byte floating + ** point value */ + pRec->u.r = (double)pRec->u.i; + pRec->flags &= ~MEM_IntReal; + pRec->flags |= MEM_Real; + pRec->uTemp = 7; + }else{ + pRec->uTemp = 6; + } + } + }else if( pRec->flags & MEM_Real ){ + nHdr++; + nData += 8; + pRec->uTemp = 7; + }else{ + assert( db->mallocFailed || pRec->flags&(MEM_Str|MEM_Blob) ); + assert( pRec->n>=0 ); + len = (u32)pRec->n; + serial_type = (len*2) + 12 + ((pRec->flags & MEM_Str)!=0); + if( pRec->flags & MEM_Zero ){ + serial_type += pRec->u.nZero*2; + if( nData ){ + if( sqlite3VdbeMemExpandBlob(pRec) ) goto no_mem; + len += pRec->u.nZero; + }else{ + nZero += pRec->u.nZero; + } + } + nData += len; + nHdr += sqlite3VarintLen(serial_type); + pRec->uTemp = serial_type; + } + if( pRec==pData0 ) break; + pRec--; + }while(1); + + /* EVIDENCE-OF: R-22564-11647 The header begins with a single varint + ** which determines the total number of bytes in the header. The varint + ** value is the size of the header in bytes including the size varint + ** itself. */ + testcase( nHdr==126 ); + testcase( nHdr==127 ); + if( nHdr<=126 ){ + /* The common case */ + nHdr += 1; + }else{ + /* Rare case of a really large header */ + nVarint = sqlite3VarintLen(nHdr); + nHdr += nVarint; + if( nVarintp3) is not allowed to + ** be one of the input registers (because the following call to + ** sqlite3VdbeMemClearAndResize() could clobber the value before it is used). + */ + if( nByte+nZero<=pOut->szMalloc ){ + /* The output register is already large enough to hold the record. + ** No error checks or buffer enlargement is required */ + pOut->z = pOut->zMalloc; + }else{ + /* Need to make sure that the output is not too big and then enlarge + ** the output register to hold the full result */ + if( nByte+nZero>db->aLimit[SQLITE_LIMIT_LENGTH] ){ + goto too_big; + } + if( sqlite3VdbeMemClearAndResize(pOut, (int)nByte) ){ + goto no_mem; + } + } + pOut->n = (int)nByte; + pOut->flags = MEM_Blob; + if( nZero ){ + pOut->u.nZero = nZero; + pOut->flags |= MEM_Zero; + } + UPDATE_MAX_BLOBSIZE(pOut); + zHdr = (u8 *)pOut->z; + zPayload = zHdr + nHdr; + + /* Write the record */ + zHdr += putVarint32(zHdr, nHdr); + assert( pData0<=pLast ); + pRec = pData0; + do{ + serial_type = pRec->uTemp; + /* EVIDENCE-OF: R-06529-47362 Following the size varint are one or more + ** additional varints, one per column. */ + zHdr += putVarint32(zHdr, serial_type); /* serial type */ + /* EVIDENCE-OF: R-64536-51728 The values for each column in the record + ** immediately follow the header. */ + zPayload += sqlite3VdbeSerialPut(zPayload, pRec, serial_type); /* content */ + }while( (++pRec)<=pLast ); + assert( nHdr==(int)(zHdr - (u8*)pOut->z) ); + assert( nByte==(int)(zPayload - (u8*)pOut->z) ); + + assert( pOp->p3>0 && pOp->p3<=(p->nMem+1 - p->nCursor) ); + REGISTER_TRACE(pOp->p3, pOut); + break; +} + +/* Opcode: Count P1 P2 * * * +** Synopsis: r[P2]=count() +** +** Store the number of entries (an integer value) in the table or index +** opened by cursor P1 in register P2 +*/ +#ifndef SQLITE_OMIT_BTREECOUNT +case OP_Count: { /* out2 */ + i64 nEntry; + BtCursor *pCrsr; + + assert( p->apCsr[pOp->p1]->eCurType==CURTYPE_BTREE ); + pCrsr = p->apCsr[pOp->p1]->uc.pCursor; + assert( pCrsr ); + nEntry = 0; /* Not needed. Only used to silence a warning. */ + rc = sqlite3BtreeCount(pCrsr, &nEntry); + if( rc ) goto abort_due_to_error; + pOut = out2Prerelease(p, pOp); + pOut->u.i = nEntry; + break; +} +#endif + +/* Opcode: Savepoint P1 * * P4 * +** +** Open, release or rollback the savepoint named by parameter P4, depending +** on the value of P1. To open a new savepoint set P1==0 (SAVEPOINT_BEGIN). +** To release (commit) an existing savepoint set P1==1 (SAVEPOINT_RELEASE). +** To rollback an existing savepoint set P1==2 (SAVEPOINT_ROLLBACK). +*/ +case OP_Savepoint: { + int p1; /* Value of P1 operand */ + char *zName; /* Name of savepoint */ + int nName; + Savepoint *pNew; + Savepoint *pSavepoint; + Savepoint *pTmp; + int iSavepoint; + int ii; + + p1 = pOp->p1; + zName = pOp->p4.z; + + /* Assert that the p1 parameter is valid. Also that if there is no open + ** transaction, then there cannot be any savepoints. + */ + assert( db->pSavepoint==0 || db->autoCommit==0 ); + assert( p1==SAVEPOINT_BEGIN||p1==SAVEPOINT_RELEASE||p1==SAVEPOINT_ROLLBACK ); + assert( db->pSavepoint || db->isTransactionSavepoint==0 ); + assert( checkSavepointCount(db) ); + assert( p->bIsReader ); + + if( p1==SAVEPOINT_BEGIN ){ + if( db->nVdbeWrite>0 ){ + /* A new savepoint cannot be created if there are active write + ** statements (i.e. open read/write incremental blob handles). + */ + sqlite3VdbeError(p, "cannot open savepoint - SQL statements in progress"); + rc = SQLITE_BUSY; + }else{ + nName = sqlite3Strlen30(zName); + +#ifndef SQLITE_OMIT_VIRTUALTABLE + /* This call is Ok even if this savepoint is actually a transaction + ** savepoint (and therefore should not prompt xSavepoint()) callbacks. + ** If this is a transaction savepoint being opened, it is guaranteed + ** that the db->aVTrans[] array is empty. */ + assert( db->autoCommit==0 || db->nVTrans==0 ); + rc = sqlite3VtabSavepoint(db, SAVEPOINT_BEGIN, + db->nStatement+db->nSavepoint); + if( rc!=SQLITE_OK ) goto abort_due_to_error; +#endif + + /* Create a new savepoint structure. */ + pNew = sqlite3DbMallocRawNN(db, sizeof(Savepoint)+nName+1); + if( pNew ){ + pNew->zName = (char *)&pNew[1]; + memcpy(pNew->zName, zName, nName+1); + + /* If there is no open transaction, then mark this as a special + ** "transaction savepoint". */ + if( db->autoCommit ){ + db->autoCommit = 0; + db->isTransactionSavepoint = 1; + }else{ + db->nSavepoint++; + } + + /* Link the new savepoint into the database handle's list. */ + pNew->pNext = db->pSavepoint; + db->pSavepoint = pNew; + pNew->nDeferredCons = db->nDeferredCons; + pNew->nDeferredImmCons = db->nDeferredImmCons; + } + } + }else{ + assert( p1==SAVEPOINT_RELEASE || p1==SAVEPOINT_ROLLBACK ); + iSavepoint = 0; + + /* Find the named savepoint. If there is no such savepoint, then an + ** an error is returned to the user. */ + for( + pSavepoint = db->pSavepoint; + pSavepoint && sqlite3StrICmp(pSavepoint->zName, zName); + pSavepoint = pSavepoint->pNext + ){ + iSavepoint++; + } + if( !pSavepoint ){ + sqlite3VdbeError(p, "no such savepoint: %s", zName); + rc = SQLITE_ERROR; + }else if( db->nVdbeWrite>0 && p1==SAVEPOINT_RELEASE ){ + /* It is not possible to release (commit) a savepoint if there are + ** active write statements. + */ + sqlite3VdbeError(p, "cannot release savepoint - " + "SQL statements in progress"); + rc = SQLITE_BUSY; + }else{ + + /* Determine whether or not this is a transaction savepoint. If so, + ** and this is a RELEASE command, then the current transaction + ** is committed. + */ + int isTransaction = pSavepoint->pNext==0 && db->isTransactionSavepoint; + if( isTransaction && p1==SAVEPOINT_RELEASE ){ + if( (rc = sqlite3VdbeCheckFk(p, 1))!=SQLITE_OK ){ + goto vdbe_return; + } + db->autoCommit = 1; + if( sqlite3VdbeHalt(p)==SQLITE_BUSY ){ + p->pc = (int)(pOp - aOp); + db->autoCommit = 0; + p->rc = rc = SQLITE_BUSY; + goto vdbe_return; + } + db->isTransactionSavepoint = 0; + rc = p->rc; + }else{ + int isSchemaChange; + iSavepoint = db->nSavepoint - iSavepoint - 1; + if( p1==SAVEPOINT_ROLLBACK ){ + isSchemaChange = (db->mDbFlags & DBFLAG_SchemaChange)!=0; + for(ii=0; iinDb; ii++){ + rc = sqlite3BtreeTripAllCursors(db->aDb[ii].pBt, + SQLITE_ABORT_ROLLBACK, + isSchemaChange==0); + if( rc!=SQLITE_OK ) goto abort_due_to_error; + } + }else{ + assert( p1==SAVEPOINT_RELEASE ); + isSchemaChange = 0; + } + for(ii=0; iinDb; ii++){ + rc = sqlite3BtreeSavepoint(db->aDb[ii].pBt, p1, iSavepoint); + if( rc!=SQLITE_OK ){ + goto abort_due_to_error; + } + } + if( isSchemaChange ){ + sqlite3ExpirePreparedStatements(db, 0); + sqlite3ResetAllSchemasOfConnection(db); + db->mDbFlags |= DBFLAG_SchemaChange; + } + } + + /* Regardless of whether this is a RELEASE or ROLLBACK, destroy all + ** savepoints nested inside of the savepoint being operated on. */ + while( db->pSavepoint!=pSavepoint ){ + pTmp = db->pSavepoint; + db->pSavepoint = pTmp->pNext; + sqlite3DbFree(db, pTmp); + db->nSavepoint--; + } + + /* If it is a RELEASE, then destroy the savepoint being operated on + ** too. If it is a ROLLBACK TO, then set the number of deferred + ** constraint violations present in the database to the value stored + ** when the savepoint was created. */ + if( p1==SAVEPOINT_RELEASE ){ + assert( pSavepoint==db->pSavepoint ); + db->pSavepoint = pSavepoint->pNext; + sqlite3DbFree(db, pSavepoint); + if( !isTransaction ){ + db->nSavepoint--; + } + }else{ + assert( p1==SAVEPOINT_ROLLBACK ); + db->nDeferredCons = pSavepoint->nDeferredCons; + db->nDeferredImmCons = pSavepoint->nDeferredImmCons; + } + + if( !isTransaction || p1==SAVEPOINT_ROLLBACK ){ + rc = sqlite3VtabSavepoint(db, p1, iSavepoint); + if( rc!=SQLITE_OK ) goto abort_due_to_error; + } + } + } + if( rc ) goto abort_due_to_error; + + break; +} + +/* Opcode: AutoCommit P1 P2 * * * +** +** Set the database auto-commit flag to P1 (1 or 0). If P2 is true, roll +** back any currently active btree transactions. If there are any active +** VMs (apart from this one), then a ROLLBACK fails. A COMMIT fails if +** there are active writing VMs or active VMs that use shared cache. +** +** This instruction causes the VM to halt. +*/ +case OP_AutoCommit: { + int desiredAutoCommit; + int iRollback; + + desiredAutoCommit = pOp->p1; + iRollback = pOp->p2; + assert( desiredAutoCommit==1 || desiredAutoCommit==0 ); + assert( desiredAutoCommit==1 || iRollback==0 ); + assert( db->nVdbeActive>0 ); /* At least this one VM is active */ + assert( p->bIsReader ); + + if( desiredAutoCommit!=db->autoCommit ){ + if( iRollback ){ + assert( desiredAutoCommit==1 ); + sqlite3RollbackAll(db, SQLITE_ABORT_ROLLBACK); + db->autoCommit = 1; + }else if( desiredAutoCommit && db->nVdbeWrite>0 ){ + /* If this instruction implements a COMMIT and other VMs are writing + ** return an error indicating that the other VMs must complete first. + */ + sqlite3VdbeError(p, "cannot commit transaction - " + "SQL statements in progress"); + rc = SQLITE_BUSY; + goto abort_due_to_error; + }else if( (rc = sqlite3VdbeCheckFk(p, 1))!=SQLITE_OK ){ + goto vdbe_return; + }else{ + db->autoCommit = (u8)desiredAutoCommit; + } + if( sqlite3VdbeHalt(p)==SQLITE_BUSY ){ + p->pc = (int)(pOp - aOp); + db->autoCommit = (u8)(1-desiredAutoCommit); + p->rc = rc = SQLITE_BUSY; + goto vdbe_return; + } + assert( db->nStatement==0 ); + sqlite3CloseSavepoints(db); + if( p->rc==SQLITE_OK ){ + rc = SQLITE_DONE; + }else{ + rc = SQLITE_ERROR; + } + goto vdbe_return; + }else{ + sqlite3VdbeError(p, + (!desiredAutoCommit)?"cannot start a transaction within a transaction":( + (iRollback)?"cannot rollback - no transaction is active": + "cannot commit - no transaction is active")); + + rc = SQLITE_ERROR; + goto abort_due_to_error; + } + /*NOTREACHED*/ assert(0); +} + +/* Opcode: Transaction P1 P2 P3 P4 P5 +** +** Begin a transaction on database P1 if a transaction is not already +** active. +** If P2 is non-zero, then a write-transaction is started, or if a +** read-transaction is already active, it is upgraded to a write-transaction. +** If P2 is zero, then a read-transaction is started. +** +** P1 is the index of the database file on which the transaction is +** started. Index 0 is the main database file and index 1 is the +** file used for temporary tables. Indices of 2 or more are used for +** attached databases. +** +** If a write-transaction is started and the Vdbe.usesStmtJournal flag is +** true (this flag is set if the Vdbe may modify more than one row and may +** throw an ABORT exception), a statement transaction may also be opened. +** More specifically, a statement transaction is opened iff the database +** connection is currently not in autocommit mode, or if there are other +** active statements. A statement transaction allows the changes made by this +** VDBE to be rolled back after an error without having to roll back the +** entire transaction. If no error is encountered, the statement transaction +** will automatically commit when the VDBE halts. +** +** If P5!=0 then this opcode also checks the schema cookie against P3 +** and the schema generation counter against P4. +** The cookie changes its value whenever the database schema changes. +** This operation is used to detect when that the cookie has changed +** and that the current process needs to reread the schema. If the schema +** cookie in P3 differs from the schema cookie in the database header or +** if the schema generation counter in P4 differs from the current +** generation counter, then an SQLITE_SCHEMA error is raised and execution +** halts. The sqlite3_step() wrapper function might then reprepare the +** statement and rerun it from the beginning. +*/ +case OP_Transaction: { + Btree *pBt; + int iMeta = 0; + + assert( p->bIsReader ); + assert( p->readOnly==0 || pOp->p2==0 ); + assert( pOp->p1>=0 && pOp->p1nDb ); + assert( DbMaskTest(p->btreeMask, pOp->p1) ); + if( pOp->p2 && (db->flags & SQLITE_QueryOnly)!=0 ){ + rc = SQLITE_READONLY; + goto abort_due_to_error; + } + pBt = db->aDb[pOp->p1].pBt; + + if( pBt ){ + rc = sqlite3BtreeBeginTrans(pBt, pOp->p2, &iMeta); + testcase( rc==SQLITE_BUSY_SNAPSHOT ); + testcase( rc==SQLITE_BUSY_RECOVERY ); + if( rc!=SQLITE_OK ){ + if( (rc&0xff)==SQLITE_BUSY ){ + p->pc = (int)(pOp - aOp); + p->rc = rc; + goto vdbe_return; + } + goto abort_due_to_error; + } + + if( pOp->p2 && p->usesStmtJournal + && (db->autoCommit==0 || db->nVdbeRead>1) + ){ + assert( sqlite3BtreeIsInTrans(pBt) ); + if( p->iStatement==0 ){ + assert( db->nStatement>=0 && db->nSavepoint>=0 ); + db->nStatement++; + p->iStatement = db->nSavepoint + db->nStatement; + } + + rc = sqlite3VtabSavepoint(db, SAVEPOINT_BEGIN, p->iStatement-1); + if( rc==SQLITE_OK ){ + rc = sqlite3BtreeBeginStmt(pBt, p->iStatement); + } + + /* Store the current value of the database handles deferred constraint + ** counter. If the statement transaction needs to be rolled back, + ** the value of this counter needs to be restored too. */ + p->nStmtDefCons = db->nDeferredCons; + p->nStmtDefImmCons = db->nDeferredImmCons; + } + } + assert( pOp->p5==0 || pOp->p4type==P4_INT32 ); + if( pOp->p5 + && (iMeta!=pOp->p3 + || db->aDb[pOp->p1].pSchema->iGeneration!=pOp->p4.i) + ){ + /* + ** IMPLEMENTATION-OF: R-03189-51135 As each SQL statement runs, the schema + ** version is checked to ensure that the schema has not changed since the + ** SQL statement was prepared. + */ + sqlite3DbFree(db, p->zErrMsg); + p->zErrMsg = sqlite3DbStrDup(db, "database schema has changed"); + /* If the schema-cookie from the database file matches the cookie + ** stored with the in-memory representation of the schema, do + ** not reload the schema from the database file. + ** + ** If virtual-tables are in use, this is not just an optimization. + ** Often, v-tables store their data in other SQLite tables, which + ** are queried from within xNext() and other v-table methods using + ** prepared queries. If such a query is out-of-date, we do not want to + ** discard the database schema, as the user code implementing the + ** v-table would have to be ready for the sqlite3_vtab structure itself + ** to be invalidated whenever sqlite3_step() is called from within + ** a v-table method. + */ + if( db->aDb[pOp->p1].pSchema->schema_cookie!=iMeta ){ + sqlite3ResetOneSchema(db, pOp->p1); + } + p->expired = 1; + rc = SQLITE_SCHEMA; + } + if( rc ) goto abort_due_to_error; + break; +} + +/* Opcode: ReadCookie P1 P2 P3 * * +** +** Read cookie number P3 from database P1 and write it into register P2. +** P3==1 is the schema version. P3==2 is the database format. +** P3==3 is the recommended pager cache size, and so forth. P1==0 is +** the main database file and P1==1 is the database file used to store +** temporary tables. +** +** There must be a read-lock on the database (either a transaction +** must be started or there must be an open cursor) before +** executing this instruction. +*/ +case OP_ReadCookie: { /* out2 */ + int iMeta; + int iDb; + int iCookie; + + assert( p->bIsReader ); + iDb = pOp->p1; + iCookie = pOp->p3; + assert( pOp->p3=0 && iDbnDb ); + assert( db->aDb[iDb].pBt!=0 ); + assert( DbMaskTest(p->btreeMask, iDb) ); + + sqlite3BtreeGetMeta(db->aDb[iDb].pBt, iCookie, (u32 *)&iMeta); + pOut = out2Prerelease(p, pOp); + pOut->u.i = iMeta; + break; +} + +/* Opcode: SetCookie P1 P2 P3 * * +** +** Write the integer value P3 into cookie number P2 of database P1. +** P2==1 is the schema version. P2==2 is the database format. +** P2==3 is the recommended pager cache +** size, and so forth. P1==0 is the main database file and P1==1 is the +** database file used to store temporary tables. +** +** A transaction must be started before executing this opcode. +*/ +case OP_SetCookie: { + Db *pDb; + + sqlite3VdbeIncrWriteCounter(p, 0); + assert( pOp->p2p1>=0 && pOp->p1nDb ); + assert( DbMaskTest(p->btreeMask, pOp->p1) ); + assert( p->readOnly==0 ); + pDb = &db->aDb[pOp->p1]; + assert( pDb->pBt!=0 ); + assert( sqlite3SchemaMutexHeld(db, pOp->p1, 0) ); + /* See note about index shifting on OP_ReadCookie */ + rc = sqlite3BtreeUpdateMeta(pDb->pBt, pOp->p2, pOp->p3); + if( pOp->p2==BTREE_SCHEMA_VERSION ){ + /* When the schema cookie changes, record the new cookie internally */ + pDb->pSchema->schema_cookie = pOp->p3; + db->mDbFlags |= DBFLAG_SchemaChange; + }else if( pOp->p2==BTREE_FILE_FORMAT ){ + /* Record changes in the file format */ + pDb->pSchema->file_format = pOp->p3; + } + if( pOp->p1==1 ){ + /* Invalidate all prepared statements whenever the TEMP database + ** schema is changed. Ticket #1644 */ + sqlite3ExpirePreparedStatements(db, 0); + p->expired = 0; + } + if( rc ) goto abort_due_to_error; + break; +} + +/* Opcode: OpenRead P1 P2 P3 P4 P5 +** Synopsis: root=P2 iDb=P3 +** +** Open a read-only cursor for the database table whose root page is +** P2 in a database file. The database file is determined by P3. +** P3==0 means the main database, P3==1 means the database used for +** temporary tables, and P3>1 means used the corresponding attached +** database. Give the new cursor an identifier of P1. The P1 +** values need not be contiguous but all P1 values should be small integers. +** It is an error for P1 to be negative. +** +** Allowed P5 bits: +**
      +**
    • 0x02 OPFLAG_SEEKEQ: This cursor will only be used for +** equality lookups (implemented as a pair of opcodes OP_SeekGE/OP_IdxGT +** of OP_SeekLE/OP_IdxGT) +**
    +** +** The P4 value may be either an integer (P4_INT32) or a pointer to +** a KeyInfo structure (P4_KEYINFO). If it is a pointer to a KeyInfo +** object, then table being opened must be an [index b-tree] where the +** KeyInfo object defines the content and collating +** sequence of that index b-tree. Otherwise, if P4 is an integer +** value, then the table being opened must be a [table b-tree] with a +** number of columns no less than the value of P4. +** +** See also: OpenWrite, ReopenIdx +*/ +/* Opcode: ReopenIdx P1 P2 P3 P4 P5 +** Synopsis: root=P2 iDb=P3 +** +** The ReopenIdx opcode works like OP_OpenRead except that it first +** checks to see if the cursor on P1 is already open on the same +** b-tree and if it is this opcode becomes a no-op. In other words, +** if the cursor is already open, do not reopen it. +** +** The ReopenIdx opcode may only be used with P5==0 or P5==OPFLAG_SEEKEQ +** and with P4 being a P4_KEYINFO object. Furthermore, the P3 value must +** be the same as every other ReopenIdx or OpenRead for the same cursor +** number. +** +** Allowed P5 bits: +**
      +**
    • 0x02 OPFLAG_SEEKEQ: This cursor will only be used for +** equality lookups (implemented as a pair of opcodes OP_SeekGE/OP_IdxGT +** of OP_SeekLE/OP_IdxGT) +**
    +** +** See also: OP_OpenRead, OP_OpenWrite +*/ +/* Opcode: OpenWrite P1 P2 P3 P4 P5 +** Synopsis: root=P2 iDb=P3 +** +** Open a read/write cursor named P1 on the table or index whose root +** page is P2 (or whose root page is held in register P2 if the +** OPFLAG_P2ISREG bit is set in P5 - see below). +** +** The P4 value may be either an integer (P4_INT32) or a pointer to +** a KeyInfo structure (P4_KEYINFO). If it is a pointer to a KeyInfo +** object, then table being opened must be an [index b-tree] where the +** KeyInfo object defines the content and collating +** sequence of that index b-tree. Otherwise, if P4 is an integer +** value, then the table being opened must be a [table b-tree] with a +** number of columns no less than the value of P4. +** +** Allowed P5 bits: +**
      +**
    • 0x02 OPFLAG_SEEKEQ: This cursor will only be used for +** equality lookups (implemented as a pair of opcodes OP_SeekGE/OP_IdxGT +** of OP_SeekLE/OP_IdxGT) +**
    • 0x08 OPFLAG_FORDELETE: This cursor is used only to seek +** and subsequently delete entries in an index btree. This is a +** hint to the storage engine that the storage engine is allowed to +** ignore. The hint is not used by the official SQLite b*tree storage +** engine, but is used by COMDB2. +**
    • 0x10 OPFLAG_P2ISREG: Use the content of register P2 +** as the root page, not the value of P2 itself. +**
    +** +** This instruction works like OpenRead except that it opens the cursor +** in read/write mode. +** +** See also: OP_OpenRead, OP_ReopenIdx +*/ +case OP_ReopenIdx: { + int nField; + KeyInfo *pKeyInfo; + int p2; + int iDb; + int wrFlag; + Btree *pX; + VdbeCursor *pCur; + Db *pDb; + + assert( pOp->p5==0 || pOp->p5==OPFLAG_SEEKEQ ); + assert( pOp->p4type==P4_KEYINFO ); + pCur = p->apCsr[pOp->p1]; + if( pCur && pCur->pgnoRoot==(u32)pOp->p2 ){ + assert( pCur->iDb==pOp->p3 ); /* Guaranteed by the code generator */ + goto open_cursor_set_hints; + } + /* If the cursor is not currently open or is open on a different + ** index, then fall through into OP_OpenRead to force a reopen */ +case OP_OpenRead: +case OP_OpenWrite: + + assert( pOp->opcode==OP_OpenWrite || pOp->p5==0 || pOp->p5==OPFLAG_SEEKEQ ); + assert( p->bIsReader ); + assert( pOp->opcode==OP_OpenRead || pOp->opcode==OP_ReopenIdx + || p->readOnly==0 ); + + if( p->expired==1 ){ + rc = SQLITE_ABORT_ROLLBACK; + goto abort_due_to_error; + } + + nField = 0; + pKeyInfo = 0; + p2 = pOp->p2; + iDb = pOp->p3; + assert( iDb>=0 && iDbnDb ); + assert( DbMaskTest(p->btreeMask, iDb) ); + pDb = &db->aDb[iDb]; + pX = pDb->pBt; + assert( pX!=0 ); + if( pOp->opcode==OP_OpenWrite ){ + assert( OPFLAG_FORDELETE==BTREE_FORDELETE ); + wrFlag = BTREE_WRCSR | (pOp->p5 & OPFLAG_FORDELETE); + assert( sqlite3SchemaMutexHeld(db, iDb, 0) ); + if( pDb->pSchema->file_format < p->minWriteFileFormat ){ + p->minWriteFileFormat = pDb->pSchema->file_format; + } + }else{ + wrFlag = 0; + } + if( pOp->p5 & OPFLAG_P2ISREG ){ + assert( p2>0 ); + assert( p2<=(p->nMem+1 - p->nCursor) ); + assert( pOp->opcode==OP_OpenWrite ); + pIn2 = &aMem[p2]; + assert( memIsValid(pIn2) ); + assert( (pIn2->flags & MEM_Int)!=0 ); + sqlite3VdbeMemIntegerify(pIn2); + p2 = (int)pIn2->u.i; + /* The p2 value always comes from a prior OP_CreateBtree opcode and + ** that opcode will always set the p2 value to 2 or more or else fail. + ** If there were a failure, the prepared statement would have halted + ** before reaching this instruction. */ + assert( p2>=2 ); + } + if( pOp->p4type==P4_KEYINFO ){ + pKeyInfo = pOp->p4.pKeyInfo; + assert( pKeyInfo->enc==ENC(db) ); + assert( pKeyInfo->db==db ); + nField = pKeyInfo->nAllField; + }else if( pOp->p4type==P4_INT32 ){ + nField = pOp->p4.i; + } + assert( pOp->p1>=0 ); + assert( nField>=0 ); + testcase( nField==0 ); /* Table with INTEGER PRIMARY KEY and nothing else */ + pCur = allocateCursor(p, pOp->p1, nField, iDb, CURTYPE_BTREE); + if( pCur==0 ) goto no_mem; + pCur->nullRow = 1; + pCur->isOrdered = 1; + pCur->pgnoRoot = p2; +#ifdef SQLITE_DEBUG + pCur->wrFlag = wrFlag; +#endif + rc = sqlite3BtreeCursor(pX, p2, wrFlag, pKeyInfo, pCur->uc.pCursor); + pCur->pKeyInfo = pKeyInfo; + /* Set the VdbeCursor.isTable variable. Previous versions of + ** SQLite used to check if the root-page flags were sane at this point + ** and report database corruption if they were not, but this check has + ** since moved into the btree layer. */ + pCur->isTable = pOp->p4type!=P4_KEYINFO; + +open_cursor_set_hints: + assert( OPFLAG_BULKCSR==BTREE_BULKLOAD ); + assert( OPFLAG_SEEKEQ==BTREE_SEEK_EQ ); + testcase( pOp->p5 & OPFLAG_BULKCSR ); +#ifdef SQLITE_ENABLE_CURSOR_HINTS + testcase( pOp->p2 & OPFLAG_SEEKEQ ); +#endif + sqlite3BtreeCursorHintFlags(pCur->uc.pCursor, + (pOp->p5 & (OPFLAG_BULKCSR|OPFLAG_SEEKEQ))); + if( rc ) goto abort_due_to_error; + break; +} + +/* Opcode: OpenDup P1 P2 * * * +** +** Open a new cursor P1 that points to the same ephemeral table as +** cursor P2. The P2 cursor must have been opened by a prior OP_OpenEphemeral +** opcode. Only ephemeral cursors may be duplicated. +** +** Duplicate ephemeral cursors are used for self-joins of materialized views. +*/ +case OP_OpenDup: { + VdbeCursor *pOrig; /* The original cursor to be duplicated */ + VdbeCursor *pCx; /* The new cursor */ + + pOrig = p->apCsr[pOp->p2]; + assert( pOrig->pBtx!=0 ); /* Only ephemeral cursors can be duplicated */ + + pCx = allocateCursor(p, pOp->p1, pOrig->nField, -1, CURTYPE_BTREE); + if( pCx==0 ) goto no_mem; + pCx->nullRow = 1; + pCx->isEphemeral = 1; + pCx->pKeyInfo = pOrig->pKeyInfo; + pCx->isTable = pOrig->isTable; + pCx->pgnoRoot = pOrig->pgnoRoot; + pCx->isOrdered = pOrig->isOrdered; + rc = sqlite3BtreeCursor(pOrig->pBtx, pCx->pgnoRoot, BTREE_WRCSR, + pCx->pKeyInfo, pCx->uc.pCursor); + /* The sqlite3BtreeCursor() routine can only fail for the first cursor + ** opened for a database. Since there is already an open cursor when this + ** opcode is run, the sqlite3BtreeCursor() cannot fail */ + assert( rc==SQLITE_OK ); + break; +} + + +/* Opcode: OpenEphemeral P1 P2 * P4 P5 +** Synopsis: nColumn=P2 +** +** Open a new cursor P1 to a transient table. +** The cursor is always opened read/write even if +** the main database is read-only. The ephemeral +** table is deleted automatically when the cursor is closed. +** +** If the cursor P1 is already opened on an ephemeral table, the table +** is cleared (all content is erased). +** +** P2 is the number of columns in the ephemeral table. +** The cursor points to a BTree table if P4==0 and to a BTree index +** if P4 is not 0. If P4 is not NULL, it points to a KeyInfo structure +** that defines the format of keys in the index. +** +** The P5 parameter can be a mask of the BTREE_* flags defined +** in btree.h. These flags control aspects of the operation of +** the btree. The BTREE_OMIT_JOURNAL and BTREE_SINGLE flags are +** added automatically. +*/ +/* Opcode: OpenAutoindex P1 P2 * P4 * +** Synopsis: nColumn=P2 +** +** This opcode works the same as OP_OpenEphemeral. It has a +** different name to distinguish its use. Tables created using +** by this opcode will be used for automatically created transient +** indices in joins. +*/ +case OP_OpenAutoindex: +case OP_OpenEphemeral: { + VdbeCursor *pCx; + KeyInfo *pKeyInfo; + + static const int vfsFlags = + SQLITE_OPEN_READWRITE | + SQLITE_OPEN_CREATE | + SQLITE_OPEN_EXCLUSIVE | + SQLITE_OPEN_DELETEONCLOSE | + SQLITE_OPEN_TRANSIENT_DB; + assert( pOp->p1>=0 ); + assert( pOp->p2>=0 ); + pCx = p->apCsr[pOp->p1]; + if( pCx ){ + /* If the ephermeral table is already open, erase all existing content + ** so that the table is empty again, rather than creating a new table. */ + assert( pCx->isEphemeral ); + pCx->seqCount = 0; + pCx->cacheStatus = CACHE_STALE; + if( pCx->pBtx ){ + rc = sqlite3BtreeClearTable(pCx->pBtx, pCx->pgnoRoot, 0); + } + }else{ + pCx = allocateCursor(p, pOp->p1, pOp->p2, -1, CURTYPE_BTREE); + if( pCx==0 ) goto no_mem; + pCx->isEphemeral = 1; + rc = sqlite3BtreeOpen(db->pVfs, 0, db, &pCx->pBtx, + BTREE_OMIT_JOURNAL | BTREE_SINGLE | pOp->p5, + vfsFlags); + if( rc==SQLITE_OK ){ + rc = sqlite3BtreeBeginTrans(pCx->pBtx, 1, 0); + } + if( rc==SQLITE_OK ){ + /* If a transient index is required, create it by calling + ** sqlite3BtreeCreateTable() with the BTREE_BLOBKEY flag before + ** opening it. If a transient table is required, just use the + ** automatically created table with root-page 1 (an BLOB_INTKEY table). + */ + if( (pCx->pKeyInfo = pKeyInfo = pOp->p4.pKeyInfo)!=0 ){ + assert( pOp->p4type==P4_KEYINFO ); + rc = sqlite3BtreeCreateTable(pCx->pBtx, (int*)&pCx->pgnoRoot, + BTREE_BLOBKEY | pOp->p5); + if( rc==SQLITE_OK ){ + assert( pCx->pgnoRoot==MASTER_ROOT+1 ); + assert( pKeyInfo->db==db ); + assert( pKeyInfo->enc==ENC(db) ); + rc = sqlite3BtreeCursor(pCx->pBtx, pCx->pgnoRoot, BTREE_WRCSR, + pKeyInfo, pCx->uc.pCursor); + } + pCx->isTable = 0; + }else{ + pCx->pgnoRoot = MASTER_ROOT; + rc = sqlite3BtreeCursor(pCx->pBtx, MASTER_ROOT, BTREE_WRCSR, + 0, pCx->uc.pCursor); + pCx->isTable = 1; + } + } + pCx->isOrdered = (pOp->p5!=BTREE_UNORDERED); + } + if( rc ) goto abort_due_to_error; + pCx->nullRow = 1; + break; +} + +/* Opcode: SorterOpen P1 P2 P3 P4 * +** +** This opcode works like OP_OpenEphemeral except that it opens +** a transient index that is specifically designed to sort large +** tables using an external merge-sort algorithm. +** +** If argument P3 is non-zero, then it indicates that the sorter may +** assume that a stable sort considering the first P3 fields of each +** key is sufficient to produce the required results. +*/ +case OP_SorterOpen: { + VdbeCursor *pCx; + + assert( pOp->p1>=0 ); + assert( pOp->p2>=0 ); + pCx = allocateCursor(p, pOp->p1, pOp->p2, -1, CURTYPE_SORTER); + if( pCx==0 ) goto no_mem; + pCx->pKeyInfo = pOp->p4.pKeyInfo; + assert( pCx->pKeyInfo->db==db ); + assert( pCx->pKeyInfo->enc==ENC(db) ); + rc = sqlite3VdbeSorterInit(db, pOp->p3, pCx); + if( rc ) goto abort_due_to_error; + break; +} + +/* Opcode: SequenceTest P1 P2 * * * +** Synopsis: if( cursor[P1].ctr++ ) pc = P2 +** +** P1 is a sorter cursor. If the sequence counter is currently zero, jump +** to P2. Regardless of whether or not the jump is taken, increment the +** the sequence value. +*/ +case OP_SequenceTest: { + VdbeCursor *pC; + assert( pOp->p1>=0 && pOp->p1nCursor ); + pC = p->apCsr[pOp->p1]; + assert( isSorter(pC) ); + if( (pC->seqCount++)==0 ){ + goto jump_to_p2; + } + break; +} + +/* Opcode: OpenPseudo P1 P2 P3 * * +** Synopsis: P3 columns in r[P2] +** +** Open a new cursor that points to a fake table that contains a single +** row of data. The content of that one row is the content of memory +** register P2. In other words, cursor P1 becomes an alias for the +** MEM_Blob content contained in register P2. +** +** A pseudo-table created by this opcode is used to hold a single +** row output from the sorter so that the row can be decomposed into +** individual columns using the OP_Column opcode. The OP_Column opcode +** is the only cursor opcode that works with a pseudo-table. +** +** P3 is the number of fields in the records that will be stored by +** the pseudo-table. +*/ +case OP_OpenPseudo: { + VdbeCursor *pCx; + + assert( pOp->p1>=0 ); + assert( pOp->p3>=0 ); + pCx = allocateCursor(p, pOp->p1, pOp->p3, -1, CURTYPE_PSEUDO); + if( pCx==0 ) goto no_mem; + pCx->nullRow = 1; + pCx->seekResult = pOp->p2; + pCx->isTable = 1; + /* Give this pseudo-cursor a fake BtCursor pointer so that pCx + ** can be safely passed to sqlite3VdbeCursorMoveto(). This avoids a test + ** for pCx->eCurType==CURTYPE_BTREE inside of sqlite3VdbeCursorMoveto() + ** which is a performance optimization */ + pCx->uc.pCursor = sqlite3BtreeFakeValidCursor(); + assert( pOp->p5==0 ); + break; +} + +/* Opcode: Close P1 * * * * +** +** Close a cursor previously opened as P1. If P1 is not +** currently open, this instruction is a no-op. +*/ +case OP_Close: { + assert( pOp->p1>=0 && pOp->p1nCursor ); + sqlite3VdbeFreeCursor(p, p->apCsr[pOp->p1]); + p->apCsr[pOp->p1] = 0; + break; +} + +#ifdef SQLITE_ENABLE_COLUMN_USED_MASK +/* Opcode: ColumnsUsed P1 * * P4 * +** +** This opcode (which only exists if SQLite was compiled with +** SQLITE_ENABLE_COLUMN_USED_MASK) identifies which columns of the +** table or index for cursor P1 are used. P4 is a 64-bit integer +** (P4_INT64) in which the first 63 bits are one for each of the +** first 63 columns of the table or index that are actually used +** by the cursor. The high-order bit is set if any column after +** the 64th is used. +*/ +case OP_ColumnsUsed: { + VdbeCursor *pC; + pC = p->apCsr[pOp->p1]; + assert( pC->eCurType==CURTYPE_BTREE ); + pC->maskUsed = *(u64*)pOp->p4.pI64; + break; +} +#endif + +/* Opcode: SeekGE P1 P2 P3 P4 * +** Synopsis: key=r[P3@P4] +** +** If cursor P1 refers to an SQL table (B-Tree that uses integer keys), +** use the value in register P3 as the key. If cursor P1 refers +** to an SQL index, then P3 is the first in an array of P4 registers +** that are used as an unpacked index key. +** +** Reposition cursor P1 so that it points to the smallest entry that +** is greater than or equal to the key value. If there are no records +** greater than or equal to the key and P2 is not zero, then jump to P2. +** +** If the cursor P1 was opened using the OPFLAG_SEEKEQ flag, then this +** opcode will always land on a record that equally equals the key, or +** else jump immediately to P2. When the cursor is OPFLAG_SEEKEQ, this +** opcode must be followed by an IdxLE opcode with the same arguments. +** The IdxLE opcode will be skipped if this opcode succeeds, but the +** IdxLE opcode will be used on subsequent loop iterations. +** +** This opcode leaves the cursor configured to move in forward order, +** from the beginning toward the end. In other words, the cursor is +** configured to use Next, not Prev. +** +** See also: Found, NotFound, SeekLt, SeekGt, SeekLe +*/ +/* Opcode: SeekGT P1 P2 P3 P4 * +** Synopsis: key=r[P3@P4] +** +** If cursor P1 refers to an SQL table (B-Tree that uses integer keys), +** use the value in register P3 as a key. If cursor P1 refers +** to an SQL index, then P3 is the first in an array of P4 registers +** that are used as an unpacked index key. +** +** Reposition cursor P1 so that it points to the smallest entry that +** is greater than the key value. If there are no records greater than +** the key and P2 is not zero, then jump to P2. +** +** This opcode leaves the cursor configured to move in forward order, +** from the beginning toward the end. In other words, the cursor is +** configured to use Next, not Prev. +** +** See also: Found, NotFound, SeekLt, SeekGe, SeekLe +*/ +/* Opcode: SeekLT P1 P2 P3 P4 * +** Synopsis: key=r[P3@P4] +** +** If cursor P1 refers to an SQL table (B-Tree that uses integer keys), +** use the value in register P3 as a key. If cursor P1 refers +** to an SQL index, then P3 is the first in an array of P4 registers +** that are used as an unpacked index key. +** +** Reposition cursor P1 so that it points to the largest entry that +** is less than the key value. If there are no records less than +** the key and P2 is not zero, then jump to P2. +** +** This opcode leaves the cursor configured to move in reverse order, +** from the end toward the beginning. In other words, the cursor is +** configured to use Prev, not Next. +** +** See also: Found, NotFound, SeekGt, SeekGe, SeekLe +*/ +/* Opcode: SeekLE P1 P2 P3 P4 * +** Synopsis: key=r[P3@P4] +** +** If cursor P1 refers to an SQL table (B-Tree that uses integer keys), +** use the value in register P3 as a key. If cursor P1 refers +** to an SQL index, then P3 is the first in an array of P4 registers +** that are used as an unpacked index key. +** +** Reposition cursor P1 so that it points to the largest entry that +** is less than or equal to the key value. If there are no records +** less than or equal to the key and P2 is not zero, then jump to P2. +** +** This opcode leaves the cursor configured to move in reverse order, +** from the end toward the beginning. In other words, the cursor is +** configured to use Prev, not Next. +** +** If the cursor P1 was opened using the OPFLAG_SEEKEQ flag, then this +** opcode will always land on a record that equally equals the key, or +** else jump immediately to P2. When the cursor is OPFLAG_SEEKEQ, this +** opcode must be followed by an IdxGE opcode with the same arguments. +** The IdxGE opcode will be skipped if this opcode succeeds, but the +** IdxGE opcode will be used on subsequent loop iterations. +** +** See also: Found, NotFound, SeekGt, SeekGe, SeekLt +*/ +case OP_SeekLT: /* jump, in3, group */ +case OP_SeekLE: /* jump, in3, group */ +case OP_SeekGE: /* jump, in3, group */ +case OP_SeekGT: { /* jump, in3, group */ + int res; /* Comparison result */ + int oc; /* Opcode */ + VdbeCursor *pC; /* The cursor to seek */ + UnpackedRecord r; /* The key to seek for */ + int nField; /* Number of columns or fields in the key */ + i64 iKey; /* The rowid we are to seek to */ + int eqOnly; /* Only interested in == results */ + + assert( pOp->p1>=0 && pOp->p1nCursor ); + assert( pOp->p2!=0 ); + pC = p->apCsr[pOp->p1]; + assert( pC!=0 ); + assert( pC->eCurType==CURTYPE_BTREE ); + assert( OP_SeekLE == OP_SeekLT+1 ); + assert( OP_SeekGE == OP_SeekLT+2 ); + assert( OP_SeekGT == OP_SeekLT+3 ); + assert( pC->isOrdered ); + assert( pC->uc.pCursor!=0 ); + oc = pOp->opcode; + eqOnly = 0; + pC->nullRow = 0; +#ifdef SQLITE_DEBUG + pC->seekOp = pOp->opcode; +#endif + + pC->deferredMoveto = 0; + pC->cacheStatus = CACHE_STALE; + if( pC->isTable ){ + u16 flags3, newType; + /* The BTREE_SEEK_EQ flag is only set on index cursors */ + assert( sqlite3BtreeCursorHasHint(pC->uc.pCursor, BTREE_SEEK_EQ)==0 + || CORRUPT_DB ); + + /* The input value in P3 might be of any type: integer, real, string, + ** blob, or NULL. But it needs to be an integer before we can do + ** the seek, so convert it. */ + pIn3 = &aMem[pOp->p3]; + flags3 = pIn3->flags; + if( (flags3 & (MEM_Int|MEM_Real|MEM_IntReal|MEM_Str))==MEM_Str ){ + applyNumericAffinity(pIn3, 0); + } + iKey = sqlite3VdbeIntValue(pIn3); /* Get the integer key value */ + newType = pIn3->flags; /* Record the type after applying numeric affinity */ + pIn3->flags = flags3; /* But convert the type back to its original */ + + /* If the P3 value could not be converted into an integer without + ** loss of information, then special processing is required... */ + if( (newType & (MEM_Int|MEM_IntReal))==0 ){ + if( (newType & MEM_Real)==0 ){ + if( (newType & MEM_Null) || oc>=OP_SeekGE ){ + VdbeBranchTaken(1,2); + goto jump_to_p2; + }else{ + rc = sqlite3BtreeLast(pC->uc.pCursor, &res); + if( rc!=SQLITE_OK ) goto abort_due_to_error; + goto seek_not_found; + } + }else + + /* If the approximation iKey is larger than the actual real search + ** term, substitute >= for > and < for <=. e.g. if the search term + ** is 4.9 and the integer approximation 5: + ** + ** (x > 4.9) -> (x >= 5) + ** (x <= 4.9) -> (x < 5) + */ + if( pIn3->u.r<(double)iKey ){ + assert( OP_SeekGE==(OP_SeekGT-1) ); + assert( OP_SeekLT==(OP_SeekLE-1) ); + assert( (OP_SeekLE & 0x0001)==(OP_SeekGT & 0x0001) ); + if( (oc & 0x0001)==(OP_SeekGT & 0x0001) ) oc--; + } + + /* If the approximation iKey is smaller than the actual real search + ** term, substitute <= for < and > for >=. */ + else if( pIn3->u.r>(double)iKey ){ + assert( OP_SeekLE==(OP_SeekLT+1) ); + assert( OP_SeekGT==(OP_SeekGE+1) ); + assert( (OP_SeekLT & 0x0001)==(OP_SeekGE & 0x0001) ); + if( (oc & 0x0001)==(OP_SeekLT & 0x0001) ) oc++; + } + } + rc = sqlite3BtreeMovetoUnpacked(pC->uc.pCursor, 0, (u64)iKey, 0, &res); + pC->movetoTarget = iKey; /* Used by OP_Delete */ + if( rc!=SQLITE_OK ){ + goto abort_due_to_error; + } + }else{ + /* For a cursor with the BTREE_SEEK_EQ hint, only the OP_SeekGE and + ** OP_SeekLE opcodes are allowed, and these must be immediately followed + ** by an OP_IdxGT or OP_IdxLT opcode, respectively, with the same key. + */ + if( sqlite3BtreeCursorHasHint(pC->uc.pCursor, BTREE_SEEK_EQ) ){ + eqOnly = 1; + assert( pOp->opcode==OP_SeekGE || pOp->opcode==OP_SeekLE ); + assert( pOp[1].opcode==OP_IdxLT || pOp[1].opcode==OP_IdxGT ); + assert( pOp[1].p1==pOp[0].p1 ); + assert( pOp[1].p2==pOp[0].p2 ); + assert( pOp[1].p3==pOp[0].p3 ); + assert( pOp[1].p4.i==pOp[0].p4.i ); + } + + nField = pOp->p4.i; + assert( pOp->p4type==P4_INT32 ); + assert( nField>0 ); + r.pKeyInfo = pC->pKeyInfo; + r.nField = (u16)nField; + + /* The next line of code computes as follows, only faster: + ** if( oc==OP_SeekGT || oc==OP_SeekLE ){ + ** r.default_rc = -1; + ** }else{ + ** r.default_rc = +1; + ** } + */ + r.default_rc = ((1 & (oc - OP_SeekLT)) ? -1 : +1); + assert( oc!=OP_SeekGT || r.default_rc==-1 ); + assert( oc!=OP_SeekLE || r.default_rc==-1 ); + assert( oc!=OP_SeekGE || r.default_rc==+1 ); + assert( oc!=OP_SeekLT || r.default_rc==+1 ); + + r.aMem = &aMem[pOp->p3]; +#ifdef SQLITE_DEBUG + { int i; for(i=0; iuc.pCursor, &r, 0, 0, &res); + if( rc!=SQLITE_OK ){ + goto abort_due_to_error; + } + if( eqOnly && r.eqSeen==0 ){ + assert( res!=0 ); + goto seek_not_found; + } + } +#ifdef SQLITE_TEST + sqlite3_search_count++; +#endif + if( oc>=OP_SeekGE ){ assert( oc==OP_SeekGE || oc==OP_SeekGT ); + if( res<0 || (res==0 && oc==OP_SeekGT) ){ + res = 0; + rc = sqlite3BtreeNext(pC->uc.pCursor, 0); + if( rc!=SQLITE_OK ){ + if( rc==SQLITE_DONE ){ + rc = SQLITE_OK; + res = 1; + }else{ + goto abort_due_to_error; + } + } + }else{ + res = 0; + } + }else{ + assert( oc==OP_SeekLT || oc==OP_SeekLE ); + if( res>0 || (res==0 && oc==OP_SeekLT) ){ + res = 0; + rc = sqlite3BtreePrevious(pC->uc.pCursor, 0); + if( rc!=SQLITE_OK ){ + if( rc==SQLITE_DONE ){ + rc = SQLITE_OK; + res = 1; + }else{ + goto abort_due_to_error; + } + } + }else{ + /* res might be negative because the table is empty. Check to + ** see if this is the case. + */ + res = sqlite3BtreeEof(pC->uc.pCursor); + } + } +seek_not_found: + assert( pOp->p2>0 ); + VdbeBranchTaken(res!=0,2); + if( res ){ + goto jump_to_p2; + }else if( eqOnly ){ + assert( pOp[1].opcode==OP_IdxLT || pOp[1].opcode==OP_IdxGT ); + pOp++; /* Skip the OP_IdxLt or OP_IdxGT that follows */ + } + break; +} + +/* Opcode: SeekHit P1 P2 * * * +** Synopsis: seekHit=P2 +** +** Set the seekHit flag on cursor P1 to the value in P2. +** The seekHit flag is used by the IfNoHope opcode. +** +** P1 must be a valid b-tree cursor. P2 must be a boolean value, +** either 0 or 1. +*/ +case OP_SeekHit: { + VdbeCursor *pC; + assert( pOp->p1>=0 && pOp->p1nCursor ); + pC = p->apCsr[pOp->p1]; + assert( pC!=0 ); + assert( pOp->p2==0 || pOp->p2==1 ); + pC->seekHit = pOp->p2 & 1; + break; +} + +/* Opcode: Found P1 P2 P3 P4 * +** Synopsis: key=r[P3@P4] +** +** If P4==0 then register P3 holds a blob constructed by MakeRecord. If +** P4>0 then register P3 is the first of P4 registers that form an unpacked +** record. +** +** Cursor P1 is on an index btree. If the record identified by P3 and P4 +** is a prefix of any entry in P1 then a jump is made to P2 and +** P1 is left pointing at the matching entry. +** +** This operation leaves the cursor in a state where it can be +** advanced in the forward direction. The Next instruction will work, +** but not the Prev instruction. +** +** See also: NotFound, NoConflict, NotExists. SeekGe +*/ +/* Opcode: NotFound P1 P2 P3 P4 * +** Synopsis: key=r[P3@P4] +** +** If P4==0 then register P3 holds a blob constructed by MakeRecord. If +** P4>0 then register P3 is the first of P4 registers that form an unpacked +** record. +** +** Cursor P1 is on an index btree. If the record identified by P3 and P4 +** is not the prefix of any entry in P1 then a jump is made to P2. If P1 +** does contain an entry whose prefix matches the P3/P4 record then control +** falls through to the next instruction and P1 is left pointing at the +** matching entry. +** +** This operation leaves the cursor in a state where it cannot be +** advanced in either direction. In other words, the Next and Prev +** opcodes do not work after this operation. +** +** See also: Found, NotExists, NoConflict, IfNoHope +*/ +/* Opcode: IfNoHope P1 P2 P3 P4 * +** Synopsis: key=r[P3@P4] +** +** Register P3 is the first of P4 registers that form an unpacked +** record. +** +** Cursor P1 is on an index btree. If the seekHit flag is set on P1, then +** this opcode is a no-op. But if the seekHit flag of P1 is clear, then +** check to see if there is any entry in P1 that matches the +** prefix identified by P3 and P4. If no entry matches the prefix, +** jump to P2. Otherwise fall through. +** +** This opcode behaves like OP_NotFound if the seekHit +** flag is clear and it behaves like OP_Noop if the seekHit flag is set. +** +** This opcode is used in IN clause processing for a multi-column key. +** If an IN clause is attached to an element of the key other than the +** left-most element, and if there are no matches on the most recent +** seek over the whole key, then it might be that one of the key element +** to the left is prohibiting a match, and hence there is "no hope" of +** any match regardless of how many IN clause elements are checked. +** In such a case, we abandon the IN clause search early, using this +** opcode. The opcode name comes from the fact that the +** jump is taken if there is "no hope" of achieving a match. +** +** See also: NotFound, SeekHit +*/ +/* Opcode: NoConflict P1 P2 P3 P4 * +** Synopsis: key=r[P3@P4] +** +** If P4==0 then register P3 holds a blob constructed by MakeRecord. If +** P4>0 then register P3 is the first of P4 registers that form an unpacked +** record. +** +** Cursor P1 is on an index btree. If the record identified by P3 and P4 +** contains any NULL value, jump immediately to P2. If all terms of the +** record are not-NULL then a check is done to determine if any row in the +** P1 index btree has a matching key prefix. If there are no matches, jump +** immediately to P2. If there is a match, fall through and leave the P1 +** cursor pointing to the matching row. +** +** This opcode is similar to OP_NotFound with the exceptions that the +** branch is always taken if any part of the search key input is NULL. +** +** This operation leaves the cursor in a state where it cannot be +** advanced in either direction. In other words, the Next and Prev +** opcodes do not work after this operation. +** +** See also: NotFound, Found, NotExists +*/ +case OP_IfNoHope: { /* jump, in3 */ + VdbeCursor *pC; + assert( pOp->p1>=0 && pOp->p1nCursor ); + pC = p->apCsr[pOp->p1]; + assert( pC!=0 ); + if( pC->seekHit ) break; + /* Fall through into OP_NotFound */ +} +case OP_NoConflict: /* jump, in3 */ +case OP_NotFound: /* jump, in3 */ +case OP_Found: { /* jump, in3 */ + int alreadyExists; + int takeJump; + int ii; + VdbeCursor *pC; + int res; + UnpackedRecord *pFree; + UnpackedRecord *pIdxKey; + UnpackedRecord r; + +#ifdef SQLITE_TEST + if( pOp->opcode!=OP_NoConflict ) sqlite3_found_count++; +#endif + + assert( pOp->p1>=0 && pOp->p1nCursor ); + assert( pOp->p4type==P4_INT32 ); + pC = p->apCsr[pOp->p1]; + assert( pC!=0 ); +#ifdef SQLITE_DEBUG + pC->seekOp = pOp->opcode; +#endif + pIn3 = &aMem[pOp->p3]; + assert( pC->eCurType==CURTYPE_BTREE ); + assert( pC->uc.pCursor!=0 ); + assert( pC->isTable==0 ); + if( pOp->p4.i>0 ){ + r.pKeyInfo = pC->pKeyInfo; + r.nField = (u16)pOp->p4.i; + r.aMem = pIn3; +#ifdef SQLITE_DEBUG + for(ii=0; iip3+ii, &r.aMem[ii]); + } +#endif + pIdxKey = &r; + pFree = 0; + }else{ + assert( pIn3->flags & MEM_Blob ); + rc = ExpandBlob(pIn3); + assert( rc==SQLITE_OK || rc==SQLITE_NOMEM ); + if( rc ) goto no_mem; + pFree = pIdxKey = sqlite3VdbeAllocUnpackedRecord(pC->pKeyInfo); + if( pIdxKey==0 ) goto no_mem; + sqlite3VdbeRecordUnpack(pC->pKeyInfo, pIn3->n, pIn3->z, pIdxKey); + } + pIdxKey->default_rc = 0; + takeJump = 0; + if( pOp->opcode==OP_NoConflict ){ + /* For the OP_NoConflict opcode, take the jump if any of the + ** input fields are NULL, since any key with a NULL will not + ** conflict */ + for(ii=0; iinField; ii++){ + if( pIdxKey->aMem[ii].flags & MEM_Null ){ + takeJump = 1; + break; + } + } + } + rc = sqlite3BtreeMovetoUnpacked(pC->uc.pCursor, pIdxKey, 0, 0, &res); + if( pFree ) sqlite3DbFreeNN(db, pFree); + if( rc!=SQLITE_OK ){ + goto abort_due_to_error; + } + pC->seekResult = res; + alreadyExists = (res==0); + pC->nullRow = 1-alreadyExists; + pC->deferredMoveto = 0; + pC->cacheStatus = CACHE_STALE; + if( pOp->opcode==OP_Found ){ + VdbeBranchTaken(alreadyExists!=0,2); + if( alreadyExists ) goto jump_to_p2; + }else{ + VdbeBranchTaken(takeJump||alreadyExists==0,2); + if( takeJump || !alreadyExists ) goto jump_to_p2; + } + break; +} + +/* Opcode: SeekRowid P1 P2 P3 * * +** Synopsis: intkey=r[P3] +** +** P1 is the index of a cursor open on an SQL table btree (with integer +** keys). If register P3 does not contain an integer or if P1 does not +** contain a record with rowid P3 then jump immediately to P2. +** Or, if P2 is 0, raise an SQLITE_CORRUPT error. If P1 does contain +** a record with rowid P3 then +** leave the cursor pointing at that record and fall through to the next +** instruction. +** +** The OP_NotExists opcode performs the same operation, but with OP_NotExists +** the P3 register must be guaranteed to contain an integer value. With this +** opcode, register P3 might not contain an integer. +** +** The OP_NotFound opcode performs the same operation on index btrees +** (with arbitrary multi-value keys). +** +** This opcode leaves the cursor in a state where it cannot be advanced +** in either direction. In other words, the Next and Prev opcodes will +** not work following this opcode. +** +** See also: Found, NotFound, NoConflict, SeekRowid +*/ +/* Opcode: NotExists P1 P2 P3 * * +** Synopsis: intkey=r[P3] +** +** P1 is the index of a cursor open on an SQL table btree (with integer +** keys). P3 is an integer rowid. If P1 does not contain a record with +** rowid P3 then jump immediately to P2. Or, if P2 is 0, raise an +** SQLITE_CORRUPT error. If P1 does contain a record with rowid P3 then +** leave the cursor pointing at that record and fall through to the next +** instruction. +** +** The OP_SeekRowid opcode performs the same operation but also allows the +** P3 register to contain a non-integer value, in which case the jump is +** always taken. This opcode requires that P3 always contain an integer. +** +** The OP_NotFound opcode performs the same operation on index btrees +** (with arbitrary multi-value keys). +** +** This opcode leaves the cursor in a state where it cannot be advanced +** in either direction. In other words, the Next and Prev opcodes will +** not work following this opcode. +** +** See also: Found, NotFound, NoConflict, SeekRowid +*/ +case OP_SeekRowid: { /* jump, in3 */ + VdbeCursor *pC; + BtCursor *pCrsr; + int res; + u64 iKey; + + pIn3 = &aMem[pOp->p3]; + testcase( pIn3->flags & MEM_Int ); + testcase( pIn3->flags & MEM_IntReal ); + testcase( pIn3->flags & MEM_Real ); + testcase( (pIn3->flags & (MEM_Str|MEM_Int))==MEM_Str ); + if( (pIn3->flags & (MEM_Int|MEM_IntReal))==0 ){ + /* If pIn3->u.i does not contain an integer, compute iKey as the + ** integer value of pIn3. Jump to P2 if pIn3 cannot be converted + ** into an integer without loss of information. Take care to avoid + ** changing the datatype of pIn3, however, as it is used by other + ** parts of the prepared statement. */ + Mem x = pIn3[0]; + applyAffinity(&x, SQLITE_AFF_NUMERIC, encoding); + if( (x.flags & MEM_Int)==0 ) goto jump_to_p2; + iKey = x.u.i; + goto notExistsWithKey; + } + /* Fall through into OP_NotExists */ +case OP_NotExists: /* jump, in3 */ + pIn3 = &aMem[pOp->p3]; + assert( (pIn3->flags & MEM_Int)!=0 || pOp->opcode==OP_SeekRowid ); + assert( pOp->p1>=0 && pOp->p1nCursor ); + iKey = pIn3->u.i; +notExistsWithKey: + pC = p->apCsr[pOp->p1]; + assert( pC!=0 ); +#ifdef SQLITE_DEBUG + if( pOp->opcode==OP_SeekRowid ) pC->seekOp = OP_SeekRowid; +#endif + assert( pC->isTable ); + assert( pC->eCurType==CURTYPE_BTREE ); + pCrsr = pC->uc.pCursor; + assert( pCrsr!=0 ); + res = 0; + rc = sqlite3BtreeMovetoUnpacked(pCrsr, 0, iKey, 0, &res); + assert( rc==SQLITE_OK || res==0 ); + pC->movetoTarget = iKey; /* Used by OP_Delete */ + pC->nullRow = 0; + pC->cacheStatus = CACHE_STALE; + pC->deferredMoveto = 0; + VdbeBranchTaken(res!=0,2); + pC->seekResult = res; + if( res!=0 ){ + assert( rc==SQLITE_OK ); + if( pOp->p2==0 ){ + rc = SQLITE_CORRUPT_BKPT; + }else{ + goto jump_to_p2; + } + } + if( rc ) goto abort_due_to_error; + break; +} + +/* Opcode: Sequence P1 P2 * * * +** Synopsis: r[P2]=cursor[P1].ctr++ +** +** Find the next available sequence number for cursor P1. +** Write the sequence number into register P2. +** The sequence number on the cursor is incremented after this +** instruction. +*/ +case OP_Sequence: { /* out2 */ + assert( pOp->p1>=0 && pOp->p1nCursor ); + assert( p->apCsr[pOp->p1]!=0 ); + assert( p->apCsr[pOp->p1]->eCurType!=CURTYPE_VTAB ); + pOut = out2Prerelease(p, pOp); + pOut->u.i = p->apCsr[pOp->p1]->seqCount++; + break; +} + + +/* Opcode: NewRowid P1 P2 P3 * * +** Synopsis: r[P2]=rowid +** +** Get a new integer record number (a.k.a "rowid") used as the key to a table. +** The record number is not previously used as a key in the database +** table that cursor P1 points to. The new record number is written +** written to register P2. +** +** If P3>0 then P3 is a register in the root frame of this VDBE that holds +** the largest previously generated record number. No new record numbers are +** allowed to be less than this value. When this value reaches its maximum, +** an SQLITE_FULL error is generated. The P3 register is updated with the ' +** generated record number. This P3 mechanism is used to help implement the +** AUTOINCREMENT feature. +*/ +case OP_NewRowid: { /* out2 */ + i64 v; /* The new rowid */ + VdbeCursor *pC; /* Cursor of table to get the new rowid */ + int res; /* Result of an sqlite3BtreeLast() */ + int cnt; /* Counter to limit the number of searches */ + Mem *pMem; /* Register holding largest rowid for AUTOINCREMENT */ + VdbeFrame *pFrame; /* Root frame of VDBE */ + + v = 0; + res = 0; + pOut = out2Prerelease(p, pOp); + assert( pOp->p1>=0 && pOp->p1nCursor ); + pC = p->apCsr[pOp->p1]; + assert( pC!=0 ); + assert( pC->isTable ); + assert( pC->eCurType==CURTYPE_BTREE ); + assert( pC->uc.pCursor!=0 ); + { + /* The next rowid or record number (different terms for the same + ** thing) is obtained in a two-step algorithm. + ** + ** First we attempt to find the largest existing rowid and add one + ** to that. But if the largest existing rowid is already the maximum + ** positive integer, we have to fall through to the second + ** probabilistic algorithm + ** + ** The second algorithm is to select a rowid at random and see if + ** it already exists in the table. If it does not exist, we have + ** succeeded. If the random rowid does exist, we select a new one + ** and try again, up to 100 times. + */ + assert( pC->isTable ); + +#ifdef SQLITE_32BIT_ROWID +# define MAX_ROWID 0x7fffffff +#else + /* Some compilers complain about constants of the form 0x7fffffffffffffff. + ** Others complain about 0x7ffffffffffffffffLL. The following macro seems + ** to provide the constant while making all compilers happy. + */ +# define MAX_ROWID (i64)( (((u64)0x7fffffff)<<32) | (u64)0xffffffff ) +#endif + + if( !pC->useRandomRowid ){ + rc = sqlite3BtreeLast(pC->uc.pCursor, &res); + if( rc!=SQLITE_OK ){ + goto abort_due_to_error; + } + if( res ){ + v = 1; /* IMP: R-61914-48074 */ + }else{ + assert( sqlite3BtreeCursorIsValid(pC->uc.pCursor) ); + v = sqlite3BtreeIntegerKey(pC->uc.pCursor); + if( v>=MAX_ROWID ){ + pC->useRandomRowid = 1; + }else{ + v++; /* IMP: R-29538-34987 */ + } + } + } + +#ifndef SQLITE_OMIT_AUTOINCREMENT + if( pOp->p3 ){ + /* Assert that P3 is a valid memory cell. */ + assert( pOp->p3>0 ); + if( p->pFrame ){ + for(pFrame=p->pFrame; pFrame->pParent; pFrame=pFrame->pParent); + /* Assert that P3 is a valid memory cell. */ + assert( pOp->p3<=pFrame->nMem ); + pMem = &pFrame->aMem[pOp->p3]; + }else{ + /* Assert that P3 is a valid memory cell. */ + assert( pOp->p3<=(p->nMem+1 - p->nCursor) ); + pMem = &aMem[pOp->p3]; + memAboutToChange(p, pMem); + } + assert( memIsValid(pMem) ); + + REGISTER_TRACE(pOp->p3, pMem); + sqlite3VdbeMemIntegerify(pMem); + assert( (pMem->flags & MEM_Int)!=0 ); /* mem(P3) holds an integer */ + if( pMem->u.i==MAX_ROWID || pC->useRandomRowid ){ + rc = SQLITE_FULL; /* IMP: R-17817-00630 */ + goto abort_due_to_error; + } + if( vu.i+1 ){ + v = pMem->u.i + 1; + } + pMem->u.i = v; + } +#endif + if( pC->useRandomRowid ){ + /* IMPLEMENTATION-OF: R-07677-41881 If the largest ROWID is equal to the + ** largest possible integer (9223372036854775807) then the database + ** engine starts picking positive candidate ROWIDs at random until + ** it finds one that is not previously used. */ + assert( pOp->p3==0 ); /* We cannot be in random rowid mode if this is + ** an AUTOINCREMENT table. */ + cnt = 0; + do{ + sqlite3_randomness(sizeof(v), &v); + v &= (MAX_ROWID>>1); v++; /* Ensure that v is greater than zero */ + }while( ((rc = sqlite3BtreeMovetoUnpacked(pC->uc.pCursor, 0, (u64)v, + 0, &res))==SQLITE_OK) + && (res==0) + && (++cnt<100)); + if( rc ) goto abort_due_to_error; + if( res==0 ){ + rc = SQLITE_FULL; /* IMP: R-38219-53002 */ + goto abort_due_to_error; + } + assert( v>0 ); /* EV: R-40812-03570 */ + } + pC->deferredMoveto = 0; + pC->cacheStatus = CACHE_STALE; + } + pOut->u.i = v; + break; +} + +/* Opcode: Insert P1 P2 P3 P4 P5 +** Synopsis: intkey=r[P3] data=r[P2] +** +** Write an entry into the table of cursor P1. A new entry is +** created if it doesn't already exist or the data for an existing +** entry is overwritten. The data is the value MEM_Blob stored in register +** number P2. The key is stored in register P3. The key must +** be a MEM_Int. +** +** If the OPFLAG_NCHANGE flag of P5 is set, then the row change count is +** incremented (otherwise not). If the OPFLAG_LASTROWID flag of P5 is set, +** then rowid is stored for subsequent return by the +** sqlite3_last_insert_rowid() function (otherwise it is unmodified). +** +** If the OPFLAG_USESEEKRESULT flag of P5 is set, the implementation might +** run faster by avoiding an unnecessary seek on cursor P1. However, +** the OPFLAG_USESEEKRESULT flag must only be set if there have been no prior +** seeks on the cursor or if the most recent seek used a key equal to P3. +** +** If the OPFLAG_ISUPDATE flag is set, then this opcode is part of an +** UPDATE operation. Otherwise (if the flag is clear) then this opcode +** is part of an INSERT operation. The difference is only important to +** the update hook. +** +** Parameter P4 may point to a Table structure, or may be NULL. If it is +** not NULL, then the update-hook (sqlite3.xUpdateCallback) is invoked +** following a successful insert. +** +** (WARNING/TODO: If P1 is a pseudo-cursor and P2 is dynamically +** allocated, then ownership of P2 is transferred to the pseudo-cursor +** and register P2 becomes ephemeral. If the cursor is changed, the +** value of register P2 will then change. Make sure this does not +** cause any problems.) +** +** This instruction only works on tables. The equivalent instruction +** for indices is OP_IdxInsert. +*/ +case OP_Insert: { + Mem *pData; /* MEM cell holding data for the record to be inserted */ + Mem *pKey; /* MEM cell holding key for the record */ + VdbeCursor *pC; /* Cursor to table into which insert is written */ + int seekResult; /* Result of prior seek or 0 if no USESEEKRESULT flag */ + const char *zDb; /* database name - used by the update hook */ + Table *pTab; /* Table structure - used by update and pre-update hooks */ + BtreePayload x; /* Payload to be inserted */ + + pData = &aMem[pOp->p2]; + assert( pOp->p1>=0 && pOp->p1nCursor ); + assert( memIsValid(pData) ); + pC = p->apCsr[pOp->p1]; + assert( pC!=0 ); + assert( pC->eCurType==CURTYPE_BTREE ); + assert( pC->uc.pCursor!=0 ); + assert( (pOp->p5 & OPFLAG_ISNOOP) || pC->isTable ); + assert( pOp->p4type==P4_TABLE || pOp->p4type>=P4_STATIC ); + REGISTER_TRACE(pOp->p2, pData); + sqlite3VdbeIncrWriteCounter(p, pC); + + pKey = &aMem[pOp->p3]; + assert( pKey->flags & MEM_Int ); + assert( memIsValid(pKey) ); + REGISTER_TRACE(pOp->p3, pKey); + x.nKey = pKey->u.i; + + if( pOp->p4type==P4_TABLE && HAS_UPDATE_HOOK(db) ){ + assert( pC->iDb>=0 ); + zDb = db->aDb[pC->iDb].zDbSName; + pTab = pOp->p4.pTab; + assert( (pOp->p5 & OPFLAG_ISNOOP) || HasRowid(pTab) ); + }else{ + pTab = 0; + zDb = 0; /* Not needed. Silence a compiler warning. */ + } + +#ifdef SQLITE_ENABLE_PREUPDATE_HOOK + /* Invoke the pre-update hook, if any */ + if( pTab ){ + if( db->xPreUpdateCallback && !(pOp->p5 & OPFLAG_ISUPDATE) ){ + sqlite3VdbePreUpdateHook(p, pC, SQLITE_INSERT, zDb, pTab, x.nKey,pOp->p2); + } + if( db->xUpdateCallback==0 || pTab->aCol==0 ){ + /* Prevent post-update hook from running in cases when it should not */ + pTab = 0; + } + } + if( pOp->p5 & OPFLAG_ISNOOP ) break; +#endif + + if( pOp->p5 & OPFLAG_NCHANGE ) p->nChange++; + if( pOp->p5 & OPFLAG_LASTROWID ) db->lastRowid = x.nKey; + assert( pData->flags & (MEM_Blob|MEM_Str) ); + x.pData = pData->z; + x.nData = pData->n; + seekResult = ((pOp->p5 & OPFLAG_USESEEKRESULT) ? pC->seekResult : 0); + if( pData->flags & MEM_Zero ){ + x.nZero = pData->u.nZero; + }else{ + x.nZero = 0; + } + x.pKey = 0; + rc = sqlite3BtreeInsert(pC->uc.pCursor, &x, + (pOp->p5 & (OPFLAG_APPEND|OPFLAG_SAVEPOSITION)), seekResult + ); + pC->deferredMoveto = 0; + pC->cacheStatus = CACHE_STALE; + + /* Invoke the update-hook if required. */ + if( rc ) goto abort_due_to_error; + if( pTab ){ + assert( db->xUpdateCallback!=0 ); + assert( pTab->aCol!=0 ); + db->xUpdateCallback(db->pUpdateArg, + (pOp->p5 & OPFLAG_ISUPDATE) ? SQLITE_UPDATE : SQLITE_INSERT, + zDb, pTab->zName, x.nKey); + } + break; +} + +/* Opcode: Delete P1 P2 P3 P4 P5 +** +** Delete the record at which the P1 cursor is currently pointing. +** +** If the OPFLAG_SAVEPOSITION bit of the P5 parameter is set, then +** the cursor will be left pointing at either the next or the previous +** record in the table. If it is left pointing at the next record, then +** the next Next instruction will be a no-op. As a result, in this case +** it is ok to delete a record from within a Next loop. If +** OPFLAG_SAVEPOSITION bit of P5 is clear, then the cursor will be +** left in an undefined state. +** +** If the OPFLAG_AUXDELETE bit is set on P5, that indicates that this +** delete one of several associated with deleting a table row and all its +** associated index entries. Exactly one of those deletes is the "primary" +** delete. The others are all on OPFLAG_FORDELETE cursors or else are +** marked with the AUXDELETE flag. +** +** If the OPFLAG_NCHANGE flag of P2 (NB: P2 not P5) is set, then the row +** change count is incremented (otherwise not). +** +** P1 must not be pseudo-table. It has to be a real table with +** multiple rows. +** +** If P4 is not NULL then it points to a Table object. In this case either +** the update or pre-update hook, or both, may be invoked. The P1 cursor must +** have been positioned using OP_NotFound prior to invoking this opcode in +** this case. Specifically, if one is configured, the pre-update hook is +** invoked if P4 is not NULL. The update-hook is invoked if one is configured, +** P4 is not NULL, and the OPFLAG_NCHANGE flag is set in P2. +** +** If the OPFLAG_ISUPDATE flag is set in P2, then P3 contains the address +** of the memory cell that contains the value that the rowid of the row will +** be set to by the update. +*/ +case OP_Delete: { + VdbeCursor *pC; + const char *zDb; + Table *pTab; + int opflags; + + opflags = pOp->p2; + assert( pOp->p1>=0 && pOp->p1nCursor ); + pC = p->apCsr[pOp->p1]; + assert( pC!=0 ); + assert( pC->eCurType==CURTYPE_BTREE ); + assert( pC->uc.pCursor!=0 ); + assert( pC->deferredMoveto==0 ); + sqlite3VdbeIncrWriteCounter(p, pC); + +#ifdef SQLITE_DEBUG + if( pOp->p4type==P4_TABLE && HasRowid(pOp->p4.pTab) && pOp->p5==0 ){ + /* If p5 is zero, the seek operation that positioned the cursor prior to + ** OP_Delete will have also set the pC->movetoTarget field to the rowid of + ** the row that is being deleted */ + i64 iKey = sqlite3BtreeIntegerKey(pC->uc.pCursor); + assert( CORRUPT_DB || pC->movetoTarget==iKey ); + } +#endif + + /* If the update-hook or pre-update-hook will be invoked, set zDb to + ** the name of the db to pass as to it. Also set local pTab to a copy + ** of p4.pTab. Finally, if p5 is true, indicating that this cursor was + ** last moved with OP_Next or OP_Prev, not Seek or NotFound, set + ** VdbeCursor.movetoTarget to the current rowid. */ + if( pOp->p4type==P4_TABLE && HAS_UPDATE_HOOK(db) ){ + assert( pC->iDb>=0 ); + assert( pOp->p4.pTab!=0 ); + zDb = db->aDb[pC->iDb].zDbSName; + pTab = pOp->p4.pTab; + if( (pOp->p5 & OPFLAG_SAVEPOSITION)!=0 && pC->isTable ){ + pC->movetoTarget = sqlite3BtreeIntegerKey(pC->uc.pCursor); + } + }else{ + zDb = 0; /* Not needed. Silence a compiler warning. */ + pTab = 0; /* Not needed. Silence a compiler warning. */ + } + +#ifdef SQLITE_ENABLE_PREUPDATE_HOOK + /* Invoke the pre-update-hook if required. */ + if( db->xPreUpdateCallback && pOp->p4.pTab ){ + assert( !(opflags & OPFLAG_ISUPDATE) + || HasRowid(pTab)==0 + || (aMem[pOp->p3].flags & MEM_Int) + ); + sqlite3VdbePreUpdateHook(p, pC, + (opflags & OPFLAG_ISUPDATE) ? SQLITE_UPDATE : SQLITE_DELETE, + zDb, pTab, pC->movetoTarget, + pOp->p3 + ); + } + if( opflags & OPFLAG_ISNOOP ) break; +#endif + + /* Only flags that can be set are SAVEPOISTION and AUXDELETE */ + assert( (pOp->p5 & ~(OPFLAG_SAVEPOSITION|OPFLAG_AUXDELETE))==0 ); + assert( OPFLAG_SAVEPOSITION==BTREE_SAVEPOSITION ); + assert( OPFLAG_AUXDELETE==BTREE_AUXDELETE ); + +#ifdef SQLITE_DEBUG + if( p->pFrame==0 ){ + if( pC->isEphemeral==0 + && (pOp->p5 & OPFLAG_AUXDELETE)==0 + && (pC->wrFlag & OPFLAG_FORDELETE)==0 + ){ + nExtraDelete++; + } + if( pOp->p2 & OPFLAG_NCHANGE ){ + nExtraDelete--; + } + } +#endif + + rc = sqlite3BtreeDelete(pC->uc.pCursor, pOp->p5); + pC->cacheStatus = CACHE_STALE; + pC->seekResult = 0; + if( rc ) goto abort_due_to_error; + + /* Invoke the update-hook if required. */ + if( opflags & OPFLAG_NCHANGE ){ + p->nChange++; + if( db->xUpdateCallback && HasRowid(pTab) ){ + db->xUpdateCallback(db->pUpdateArg, SQLITE_DELETE, zDb, pTab->zName, + pC->movetoTarget); + assert( pC->iDb>=0 ); + } + } + + break; +} +/* Opcode: ResetCount * * * * * +** +** The value of the change counter is copied to the database handle +** change counter (returned by subsequent calls to sqlite3_changes()). +** Then the VMs internal change counter resets to 0. +** This is used by trigger programs. +*/ +case OP_ResetCount: { + sqlite3VdbeSetChanges(db, p->nChange); + p->nChange = 0; + break; +} + +/* Opcode: SorterCompare P1 P2 P3 P4 +** Synopsis: if key(P1)!=trim(r[P3],P4) goto P2 +** +** P1 is a sorter cursor. This instruction compares a prefix of the +** record blob in register P3 against a prefix of the entry that +** the sorter cursor currently points to. Only the first P4 fields +** of r[P3] and the sorter record are compared. +** +** If either P3 or the sorter contains a NULL in one of their significant +** fields (not counting the P4 fields at the end which are ignored) then +** the comparison is assumed to be equal. +** +** Fall through to next instruction if the two records compare equal to +** each other. Jump to P2 if they are different. +*/ +case OP_SorterCompare: { + VdbeCursor *pC; + int res; + int nKeyCol; + + pC = p->apCsr[pOp->p1]; + assert( isSorter(pC) ); + assert( pOp->p4type==P4_INT32 ); + pIn3 = &aMem[pOp->p3]; + nKeyCol = pOp->p4.i; + res = 0; + rc = sqlite3VdbeSorterCompare(pC, pIn3, nKeyCol, &res); + VdbeBranchTaken(res!=0,2); + if( rc ) goto abort_due_to_error; + if( res ) goto jump_to_p2; + break; +}; + +/* Opcode: SorterData P1 P2 P3 * * +** Synopsis: r[P2]=data +** +** Write into register P2 the current sorter data for sorter cursor P1. +** Then clear the column header cache on cursor P3. +** +** This opcode is normally use to move a record out of the sorter and into +** a register that is the source for a pseudo-table cursor created using +** OpenPseudo. That pseudo-table cursor is the one that is identified by +** parameter P3. Clearing the P3 column cache as part of this opcode saves +** us from having to issue a separate NullRow instruction to clear that cache. +*/ +case OP_SorterData: { + VdbeCursor *pC; + + pOut = &aMem[pOp->p2]; + pC = p->apCsr[pOp->p1]; + assert( isSorter(pC) ); + rc = sqlite3VdbeSorterRowkey(pC, pOut); + assert( rc!=SQLITE_OK || (pOut->flags & MEM_Blob) ); + assert( pOp->p1>=0 && pOp->p1nCursor ); + if( rc ) goto abort_due_to_error; + p->apCsr[pOp->p3]->cacheStatus = CACHE_STALE; + break; +} + +/* Opcode: RowData P1 P2 P3 * * +** Synopsis: r[P2]=data +** +** Write into register P2 the complete row content for the row at +** which cursor P1 is currently pointing. +** There is no interpretation of the data. +** It is just copied onto the P2 register exactly as +** it is found in the database file. +** +** If cursor P1 is an index, then the content is the key of the row. +** If cursor P2 is a table, then the content extracted is the data. +** +** If the P1 cursor must be pointing to a valid row (not a NULL row) +** of a real table, not a pseudo-table. +** +** If P3!=0 then this opcode is allowed to make an ephemeral pointer +** into the database page. That means that the content of the output +** register will be invalidated as soon as the cursor moves - including +** moves caused by other cursors that "save" the current cursors +** position in order that they can write to the same table. If P3==0 +** then a copy of the data is made into memory. P3!=0 is faster, but +** P3==0 is safer. +** +** If P3!=0 then the content of the P2 register is unsuitable for use +** in OP_Result and any OP_Result will invalidate the P2 register content. +** The P2 register content is invalidated by opcodes like OP_Function or +** by any use of another cursor pointing to the same table. +*/ +case OP_RowData: { + VdbeCursor *pC; + BtCursor *pCrsr; + u32 n; + + pOut = out2Prerelease(p, pOp); + + assert( pOp->p1>=0 && pOp->p1nCursor ); + pC = p->apCsr[pOp->p1]; + assert( pC!=0 ); + assert( pC->eCurType==CURTYPE_BTREE ); + assert( isSorter(pC)==0 ); + assert( pC->nullRow==0 ); + assert( pC->uc.pCursor!=0 ); + pCrsr = pC->uc.pCursor; + + /* The OP_RowData opcodes always follow OP_NotExists or + ** OP_SeekRowid or OP_Rewind/Op_Next with no intervening instructions + ** that might invalidate the cursor. + ** If this where not the case, on of the following assert()s + ** would fail. Should this ever change (because of changes in the code + ** generator) then the fix would be to insert a call to + ** sqlite3VdbeCursorMoveto(). + */ + assert( pC->deferredMoveto==0 ); + assert( sqlite3BtreeCursorIsValid(pCrsr) ); +#if 0 /* Not required due to the previous to assert() statements */ + rc = sqlite3VdbeCursorMoveto(pC); + if( rc!=SQLITE_OK ) goto abort_due_to_error; +#endif + + n = sqlite3BtreePayloadSize(pCrsr); + if( n>(u32)db->aLimit[SQLITE_LIMIT_LENGTH] ){ + goto too_big; + } + testcase( n==0 ); + rc = sqlite3VdbeMemFromBtree(pCrsr, 0, n, pOut); + if( rc ) goto abort_due_to_error; + if( !pOp->p3 ) Deephemeralize(pOut); + UPDATE_MAX_BLOBSIZE(pOut); + REGISTER_TRACE(pOp->p2, pOut); + break; +} + +/* Opcode: Rowid P1 P2 * * * +** Synopsis: r[P2]=rowid +** +** Store in register P2 an integer which is the key of the table entry that +** P1 is currently point to. +** +** P1 can be either an ordinary table or a virtual table. There used to +** be a separate OP_VRowid opcode for use with virtual tables, but this +** one opcode now works for both table types. +*/ +case OP_Rowid: { /* out2 */ + VdbeCursor *pC; + i64 v; + sqlite3_vtab *pVtab; + const sqlite3_module *pModule; + + pOut = out2Prerelease(p, pOp); + assert( pOp->p1>=0 && pOp->p1nCursor ); + pC = p->apCsr[pOp->p1]; + assert( pC!=0 ); + assert( pC->eCurType!=CURTYPE_PSEUDO || pC->nullRow ); + if( pC->nullRow ){ + pOut->flags = MEM_Null; + break; + }else if( pC->deferredMoveto ){ + v = pC->movetoTarget; +#ifndef SQLITE_OMIT_VIRTUALTABLE + }else if( pC->eCurType==CURTYPE_VTAB ){ + assert( pC->uc.pVCur!=0 ); + pVtab = pC->uc.pVCur->pVtab; + pModule = pVtab->pModule; + assert( pModule->xRowid ); + rc = pModule->xRowid(pC->uc.pVCur, &v); + sqlite3VtabImportErrmsg(p, pVtab); + if( rc ) goto abort_due_to_error; +#endif /* SQLITE_OMIT_VIRTUALTABLE */ + }else{ + assert( pC->eCurType==CURTYPE_BTREE ); + assert( pC->uc.pCursor!=0 ); + rc = sqlite3VdbeCursorRestore(pC); + if( rc ) goto abort_due_to_error; + if( pC->nullRow ){ + pOut->flags = MEM_Null; + break; + } + v = sqlite3BtreeIntegerKey(pC->uc.pCursor); + } + pOut->u.i = v; + break; +} + +/* Opcode: NullRow P1 * * * * +** +** Move the cursor P1 to a null row. Any OP_Column operations +** that occur while the cursor is on the null row will always +** write a NULL. +*/ +case OP_NullRow: { + VdbeCursor *pC; + + assert( pOp->p1>=0 && pOp->p1nCursor ); + pC = p->apCsr[pOp->p1]; + assert( pC!=0 ); + pC->nullRow = 1; + pC->cacheStatus = CACHE_STALE; + if( pC->eCurType==CURTYPE_BTREE ){ + assert( pC->uc.pCursor!=0 ); + sqlite3BtreeClearCursor(pC->uc.pCursor); + } +#ifdef SQLITE_DEBUG + if( pC->seekOp==0 ) pC->seekOp = OP_NullRow; +#endif + break; +} + +/* Opcode: SeekEnd P1 * * * * +** +** Position cursor P1 at the end of the btree for the purpose of +** appending a new entry onto the btree. +** +** It is assumed that the cursor is used only for appending and so +** if the cursor is valid, then the cursor must already be pointing +** at the end of the btree and so no changes are made to +** the cursor. +*/ +/* Opcode: Last P1 P2 * * * +** +** The next use of the Rowid or Column or Prev instruction for P1 +** will refer to the last entry in the database table or index. +** If the table or index is empty and P2>0, then jump immediately to P2. +** If P2 is 0 or if the table or index is not empty, fall through +** to the following instruction. +** +** This opcode leaves the cursor configured to move in reverse order, +** from the end toward the beginning. In other words, the cursor is +** configured to use Prev, not Next. +*/ +case OP_SeekEnd: +case OP_Last: { /* jump */ + VdbeCursor *pC; + BtCursor *pCrsr; + int res; + + assert( pOp->p1>=0 && pOp->p1nCursor ); + pC = p->apCsr[pOp->p1]; + assert( pC!=0 ); + assert( pC->eCurType==CURTYPE_BTREE ); + pCrsr = pC->uc.pCursor; + res = 0; + assert( pCrsr!=0 ); +#ifdef SQLITE_DEBUG + pC->seekOp = pOp->opcode; +#endif + if( pOp->opcode==OP_SeekEnd ){ + assert( pOp->p2==0 ); + pC->seekResult = -1; + if( sqlite3BtreeCursorIsValidNN(pCrsr) ){ + break; + } + } + rc = sqlite3BtreeLast(pCrsr, &res); + pC->nullRow = (u8)res; + pC->deferredMoveto = 0; + pC->cacheStatus = CACHE_STALE; + if( rc ) goto abort_due_to_error; + if( pOp->p2>0 ){ + VdbeBranchTaken(res!=0,2); + if( res ) goto jump_to_p2; + } + break; +} + +/* Opcode: IfSmaller P1 P2 P3 * * +** +** Estimate the number of rows in the table P1. Jump to P2 if that +** estimate is less than approximately 2**(0.1*P3). +*/ +case OP_IfSmaller: { /* jump */ + VdbeCursor *pC; + BtCursor *pCrsr; + int res; + i64 sz; + + assert( pOp->p1>=0 && pOp->p1nCursor ); + pC = p->apCsr[pOp->p1]; + assert( pC!=0 ); + pCrsr = pC->uc.pCursor; + assert( pCrsr ); + rc = sqlite3BtreeFirst(pCrsr, &res); + if( rc ) goto abort_due_to_error; + if( res==0 ){ + sz = sqlite3BtreeRowCountEst(pCrsr); + if( ALWAYS(sz>=0) && sqlite3LogEst((u64)sz)p3 ) res = 1; + } + VdbeBranchTaken(res!=0,2); + if( res ) goto jump_to_p2; + break; +} + + +/* Opcode: SorterSort P1 P2 * * * +** +** After all records have been inserted into the Sorter object +** identified by P1, invoke this opcode to actually do the sorting. +** Jump to P2 if there are no records to be sorted. +** +** This opcode is an alias for OP_Sort and OP_Rewind that is used +** for Sorter objects. +*/ +/* Opcode: Sort P1 P2 * * * +** +** This opcode does exactly the same thing as OP_Rewind except that +** it increments an undocumented global variable used for testing. +** +** Sorting is accomplished by writing records into a sorting index, +** then rewinding that index and playing it back from beginning to +** end. We use the OP_Sort opcode instead of OP_Rewind to do the +** rewinding so that the global variable will be incremented and +** regression tests can determine whether or not the optimizer is +** correctly optimizing out sorts. +*/ +case OP_SorterSort: /* jump */ +case OP_Sort: { /* jump */ +#ifdef SQLITE_TEST + sqlite3_sort_count++; + sqlite3_search_count--; +#endif + p->aCounter[SQLITE_STMTSTATUS_SORT]++; + /* Fall through into OP_Rewind */ +} +/* Opcode: Rewind P1 P2 * * * +** +** The next use of the Rowid or Column or Next instruction for P1 +** will refer to the first entry in the database table or index. +** If the table or index is empty, jump immediately to P2. +** If the table or index is not empty, fall through to the following +** instruction. +** +** This opcode leaves the cursor configured to move in forward order, +** from the beginning toward the end. In other words, the cursor is +** configured to use Next, not Prev. +*/ +case OP_Rewind: { /* jump */ + VdbeCursor *pC; + BtCursor *pCrsr; + int res; + + assert( pOp->p1>=0 && pOp->p1nCursor ); + assert( pOp->p5==0 ); + pC = p->apCsr[pOp->p1]; + assert( pC!=0 ); + assert( isSorter(pC)==(pOp->opcode==OP_SorterSort) ); + res = 1; +#ifdef SQLITE_DEBUG + pC->seekOp = OP_Rewind; +#endif + if( isSorter(pC) ){ + rc = sqlite3VdbeSorterRewind(pC, &res); + }else{ + assert( pC->eCurType==CURTYPE_BTREE ); + pCrsr = pC->uc.pCursor; + assert( pCrsr ); + rc = sqlite3BtreeFirst(pCrsr, &res); + pC->deferredMoveto = 0; + pC->cacheStatus = CACHE_STALE; + } + if( rc ) goto abort_due_to_error; + pC->nullRow = (u8)res; + assert( pOp->p2>0 && pOp->p2nOp ); + VdbeBranchTaken(res!=0,2); + if( res ) goto jump_to_p2; + break; +} + +/* Opcode: Next P1 P2 P3 P4 P5 +** +** Advance cursor P1 so that it points to the next key/data pair in its +** table or index. If there are no more key/value pairs then fall through +** to the following instruction. But if the cursor advance was successful, +** jump immediately to P2. +** +** The Next opcode is only valid following an SeekGT, SeekGE, or +** OP_Rewind opcode used to position the cursor. Next is not allowed +** to follow SeekLT, SeekLE, or OP_Last. +** +** The P1 cursor must be for a real table, not a pseudo-table. P1 must have +** been opened prior to this opcode or the program will segfault. +** +** The P3 value is a hint to the btree implementation. If P3==1, that +** means P1 is an SQL index and that this instruction could have been +** omitted if that index had been unique. P3 is usually 0. P3 is +** always either 0 or 1. +** +** P4 is always of type P4_ADVANCE. The function pointer points to +** sqlite3BtreeNext(). +** +** If P5 is positive and the jump is taken, then event counter +** number P5-1 in the prepared statement is incremented. +** +** See also: Prev +*/ +/* Opcode: Prev P1 P2 P3 P4 P5 +** +** Back up cursor P1 so that it points to the previous key/data pair in its +** table or index. If there is no previous key/value pairs then fall through +** to the following instruction. But if the cursor backup was successful, +** jump immediately to P2. +** +** +** The Prev opcode is only valid following an SeekLT, SeekLE, or +** OP_Last opcode used to position the cursor. Prev is not allowed +** to follow SeekGT, SeekGE, or OP_Rewind. +** +** The P1 cursor must be for a real table, not a pseudo-table. If P1 is +** not open then the behavior is undefined. +** +** The P3 value is a hint to the btree implementation. If P3==1, that +** means P1 is an SQL index and that this instruction could have been +** omitted if that index had been unique. P3 is usually 0. P3 is +** always either 0 or 1. +** +** P4 is always of type P4_ADVANCE. The function pointer points to +** sqlite3BtreePrevious(). +** +** If P5 is positive and the jump is taken, then event counter +** number P5-1 in the prepared statement is incremented. +*/ +/* Opcode: SorterNext P1 P2 * * P5 +** +** This opcode works just like OP_Next except that P1 must be a +** sorter object for which the OP_SorterSort opcode has been +** invoked. This opcode advances the cursor to the next sorted +** record, or jumps to P2 if there are no more sorted records. +*/ +case OP_SorterNext: { /* jump */ + VdbeCursor *pC; + + pC = p->apCsr[pOp->p1]; + assert( isSorter(pC) ); + rc = sqlite3VdbeSorterNext(db, pC); + goto next_tail; +case OP_Prev: /* jump */ +case OP_Next: /* jump */ + assert( pOp->p1>=0 && pOp->p1nCursor ); + assert( pOp->p5aCounter) ); + pC = p->apCsr[pOp->p1]; + assert( pC!=0 ); + assert( pC->deferredMoveto==0 ); + assert( pC->eCurType==CURTYPE_BTREE ); + assert( pOp->opcode!=OP_Next || pOp->p4.xAdvance==sqlite3BtreeNext ); + assert( pOp->opcode!=OP_Prev || pOp->p4.xAdvance==sqlite3BtreePrevious ); + + /* The Next opcode is only used after SeekGT, SeekGE, Rewind, and Found. + ** The Prev opcode is only used after SeekLT, SeekLE, and Last. */ + assert( pOp->opcode!=OP_Next + || pC->seekOp==OP_SeekGT || pC->seekOp==OP_SeekGE + || pC->seekOp==OP_Rewind || pC->seekOp==OP_Found + || pC->seekOp==OP_NullRow|| pC->seekOp==OP_SeekRowid + || pC->seekOp==OP_IfNoHope); + assert( pOp->opcode!=OP_Prev + || pC->seekOp==OP_SeekLT || pC->seekOp==OP_SeekLE + || pC->seekOp==OP_Last || pC->seekOp==OP_IfNoHope + || pC->seekOp==OP_NullRow); + + rc = pOp->p4.xAdvance(pC->uc.pCursor, pOp->p3); +next_tail: + pC->cacheStatus = CACHE_STALE; + VdbeBranchTaken(rc==SQLITE_OK,2); + if( rc==SQLITE_OK ){ + pC->nullRow = 0; + p->aCounter[pOp->p5]++; +#ifdef SQLITE_TEST + sqlite3_search_count++; +#endif + goto jump_to_p2_and_check_for_interrupt; + } + if( rc!=SQLITE_DONE ) goto abort_due_to_error; + rc = SQLITE_OK; + pC->nullRow = 1; + goto check_for_interrupt; +} + +/* Opcode: IdxInsert P1 P2 P3 P4 P5 +** Synopsis: key=r[P2] +** +** Register P2 holds an SQL index key made using the +** MakeRecord instructions. This opcode writes that key +** into the index P1. Data for the entry is nil. +** +** If P4 is not zero, then it is the number of values in the unpacked +** key of reg(P2). In that case, P3 is the index of the first register +** for the unpacked key. The availability of the unpacked key can sometimes +** be an optimization. +** +** If P5 has the OPFLAG_APPEND bit set, that is a hint to the b-tree layer +** that this insert is likely to be an append. +** +** If P5 has the OPFLAG_NCHANGE bit set, then the change counter is +** incremented by this instruction. If the OPFLAG_NCHANGE bit is clear, +** then the change counter is unchanged. +** +** If the OPFLAG_USESEEKRESULT flag of P5 is set, the implementation might +** run faster by avoiding an unnecessary seek on cursor P1. However, +** the OPFLAG_USESEEKRESULT flag must only be set if there have been no prior +** seeks on the cursor or if the most recent seek used a key equivalent +** to P2. +** +** This instruction only works for indices. The equivalent instruction +** for tables is OP_Insert. +*/ +/* Opcode: SorterInsert P1 P2 * * * +** Synopsis: key=r[P2] +** +** Register P2 holds an SQL index key made using the +** MakeRecord instructions. This opcode writes that key +** into the sorter P1. Data for the entry is nil. +*/ +case OP_SorterInsert: /* in2 */ +case OP_IdxInsert: { /* in2 */ + VdbeCursor *pC; + BtreePayload x; + + assert( pOp->p1>=0 && pOp->p1nCursor ); + pC = p->apCsr[pOp->p1]; + sqlite3VdbeIncrWriteCounter(p, pC); + assert( pC!=0 ); + assert( isSorter(pC)==(pOp->opcode==OP_SorterInsert) ); + pIn2 = &aMem[pOp->p2]; + assert( pIn2->flags & MEM_Blob ); + if( pOp->p5 & OPFLAG_NCHANGE ) p->nChange++; + assert( pC->eCurType==CURTYPE_BTREE || pOp->opcode==OP_SorterInsert ); + assert( pC->isTable==0 ); + rc = ExpandBlob(pIn2); + if( rc ) goto abort_due_to_error; + if( pOp->opcode==OP_SorterInsert ){ + rc = sqlite3VdbeSorterWrite(pC, pIn2); + }else{ + x.nKey = pIn2->n; + x.pKey = pIn2->z; + x.aMem = aMem + pOp->p3; + x.nMem = (u16)pOp->p4.i; + rc = sqlite3BtreeInsert(pC->uc.pCursor, &x, + (pOp->p5 & (OPFLAG_APPEND|OPFLAG_SAVEPOSITION)), + ((pOp->p5 & OPFLAG_USESEEKRESULT) ? pC->seekResult : 0) + ); + assert( pC->deferredMoveto==0 ); + pC->cacheStatus = CACHE_STALE; + } + if( rc) goto abort_due_to_error; + break; +} + +/* Opcode: IdxDelete P1 P2 P3 * * +** Synopsis: key=r[P2@P3] +** +** The content of P3 registers starting at register P2 form +** an unpacked index key. This opcode removes that entry from the +** index opened by cursor P1. +*/ +case OP_IdxDelete: { + VdbeCursor *pC; + BtCursor *pCrsr; + int res; + UnpackedRecord r; + + assert( pOp->p3>0 ); + assert( pOp->p2>0 && pOp->p2+pOp->p3<=(p->nMem+1 - p->nCursor)+1 ); + assert( pOp->p1>=0 && pOp->p1nCursor ); + pC = p->apCsr[pOp->p1]; + assert( pC!=0 ); + assert( pC->eCurType==CURTYPE_BTREE ); + sqlite3VdbeIncrWriteCounter(p, pC); + pCrsr = pC->uc.pCursor; + assert( pCrsr!=0 ); + assert( pOp->p5==0 ); + r.pKeyInfo = pC->pKeyInfo; + r.nField = (u16)pOp->p3; + r.default_rc = 0; + r.aMem = &aMem[pOp->p2]; + rc = sqlite3BtreeMovetoUnpacked(pCrsr, &r, 0, 0, &res); + if( rc ) goto abort_due_to_error; + if( res==0 ){ + rc = sqlite3BtreeDelete(pCrsr, BTREE_AUXDELETE); + if( rc ) goto abort_due_to_error; + } + assert( pC->deferredMoveto==0 ); + pC->cacheStatus = CACHE_STALE; + pC->seekResult = 0; + break; +} + +/* Opcode: DeferredSeek P1 * P3 P4 * +** Synopsis: Move P3 to P1.rowid if needed +** +** P1 is an open index cursor and P3 is a cursor on the corresponding +** table. This opcode does a deferred seek of the P3 table cursor +** to the row that corresponds to the current row of P1. +** +** This is a deferred seek. Nothing actually happens until +** the cursor is used to read a record. That way, if no reads +** occur, no unnecessary I/O happens. +** +** P4 may be an array of integers (type P4_INTARRAY) containing +** one entry for each column in the P3 table. If array entry a(i) +** is non-zero, then reading column a(i)-1 from cursor P3 is +** equivalent to performing the deferred seek and then reading column i +** from P1. This information is stored in P3 and used to redirect +** reads against P3 over to P1, thus possibly avoiding the need to +** seek and read cursor P3. +*/ +/* Opcode: IdxRowid P1 P2 * * * +** Synopsis: r[P2]=rowid +** +** Write into register P2 an integer which is the last entry in the record at +** the end of the index key pointed to by cursor P1. This integer should be +** the rowid of the table entry to which this index entry points. +** +** See also: Rowid, MakeRecord. +*/ +case OP_DeferredSeek: +case OP_IdxRowid: { /* out2 */ + VdbeCursor *pC; /* The P1 index cursor */ + VdbeCursor *pTabCur; /* The P2 table cursor (OP_DeferredSeek only) */ + i64 rowid; /* Rowid that P1 current points to */ + + assert( pOp->p1>=0 && pOp->p1nCursor ); + pC = p->apCsr[pOp->p1]; + assert( pC!=0 ); + assert( pC->eCurType==CURTYPE_BTREE ); + assert( pC->uc.pCursor!=0 ); + assert( pC->isTable==0 ); + assert( pC->deferredMoveto==0 ); + assert( !pC->nullRow || pOp->opcode==OP_IdxRowid ); + + /* The IdxRowid and Seek opcodes are combined because of the commonality + ** of sqlite3VdbeCursorRestore() and sqlite3VdbeIdxRowid(). */ + rc = sqlite3VdbeCursorRestore(pC); + + /* sqlite3VbeCursorRestore() can only fail if the record has been deleted + ** out from under the cursor. That will never happens for an IdxRowid + ** or Seek opcode */ + if( NEVER(rc!=SQLITE_OK) ) goto abort_due_to_error; + + if( !pC->nullRow ){ + rowid = 0; /* Not needed. Only used to silence a warning. */ + rc = sqlite3VdbeIdxRowid(db, pC->uc.pCursor, &rowid); + if( rc!=SQLITE_OK ){ + goto abort_due_to_error; + } + if( pOp->opcode==OP_DeferredSeek ){ + assert( pOp->p3>=0 && pOp->p3nCursor ); + pTabCur = p->apCsr[pOp->p3]; + assert( pTabCur!=0 ); + assert( pTabCur->eCurType==CURTYPE_BTREE ); + assert( pTabCur->uc.pCursor!=0 ); + assert( pTabCur->isTable ); + pTabCur->nullRow = 0; + pTabCur->movetoTarget = rowid; + pTabCur->deferredMoveto = 1; + assert( pOp->p4type==P4_INTARRAY || pOp->p4.ai==0 ); + pTabCur->aAltMap = pOp->p4.ai; + pTabCur->pAltCursor = pC; + }else{ + pOut = out2Prerelease(p, pOp); + pOut->u.i = rowid; + } + }else{ + assert( pOp->opcode==OP_IdxRowid ); + sqlite3VdbeMemSetNull(&aMem[pOp->p2]); + } + break; +} + +/* Opcode: IdxGE P1 P2 P3 P4 P5 +** Synopsis: key=r[P3@P4] +** +** The P4 register values beginning with P3 form an unpacked index +** key that omits the PRIMARY KEY. Compare this key value against the index +** that P1 is currently pointing to, ignoring the PRIMARY KEY or ROWID +** fields at the end. +** +** If the P1 index entry is greater than or equal to the key value +** then jump to P2. Otherwise fall through to the next instruction. +*/ +/* Opcode: IdxGT P1 P2 P3 P4 P5 +** Synopsis: key=r[P3@P4] +** +** The P4 register values beginning with P3 form an unpacked index +** key that omits the PRIMARY KEY. Compare this key value against the index +** that P1 is currently pointing to, ignoring the PRIMARY KEY or ROWID +** fields at the end. +** +** If the P1 index entry is greater than the key value +** then jump to P2. Otherwise fall through to the next instruction. +*/ +/* Opcode: IdxLT P1 P2 P3 P4 P5 +** Synopsis: key=r[P3@P4] +** +** The P4 register values beginning with P3 form an unpacked index +** key that omits the PRIMARY KEY or ROWID. Compare this key value against +** the index that P1 is currently pointing to, ignoring the PRIMARY KEY or +** ROWID on the P1 index. +** +** If the P1 index entry is less than the key value then jump to P2. +** Otherwise fall through to the next instruction. +*/ +/* Opcode: IdxLE P1 P2 P3 P4 P5 +** Synopsis: key=r[P3@P4] +** +** The P4 register values beginning with P3 form an unpacked index +** key that omits the PRIMARY KEY or ROWID. Compare this key value against +** the index that P1 is currently pointing to, ignoring the PRIMARY KEY or +** ROWID on the P1 index. +** +** If the P1 index entry is less than or equal to the key value then jump +** to P2. Otherwise fall through to the next instruction. +*/ +case OP_IdxLE: /* jump */ +case OP_IdxGT: /* jump */ +case OP_IdxLT: /* jump */ +case OP_IdxGE: { /* jump */ + VdbeCursor *pC; + int res; + UnpackedRecord r; + + assert( pOp->p1>=0 && pOp->p1nCursor ); + pC = p->apCsr[pOp->p1]; + assert( pC!=0 ); + assert( pC->isOrdered ); + assert( pC->eCurType==CURTYPE_BTREE ); + assert( pC->uc.pCursor!=0); + assert( pC->deferredMoveto==0 ); + assert( pOp->p5==0 || pOp->p5==1 ); + assert( pOp->p4type==P4_INT32 ); + r.pKeyInfo = pC->pKeyInfo; + r.nField = (u16)pOp->p4.i; + if( pOp->opcodeopcode==OP_IdxLE || pOp->opcode==OP_IdxGT ); + r.default_rc = -1; + }else{ + assert( pOp->opcode==OP_IdxGE || pOp->opcode==OP_IdxLT ); + r.default_rc = 0; + } + r.aMem = &aMem[pOp->p3]; +#ifdef SQLITE_DEBUG + { + int i; + for(i=0; ip3+i, &aMem[pOp->p3+i]); + } + } +#endif + res = 0; /* Not needed. Only used to silence a warning. */ + rc = sqlite3VdbeIdxKeyCompare(db, pC, &r, &res); + assert( (OP_IdxLE&1)==(OP_IdxLT&1) && (OP_IdxGE&1)==(OP_IdxGT&1) ); + if( (pOp->opcode&1)==(OP_IdxLT&1) ){ + assert( pOp->opcode==OP_IdxLE || pOp->opcode==OP_IdxLT ); + res = -res; + }else{ + assert( pOp->opcode==OP_IdxGE || pOp->opcode==OP_IdxGT ); + res++; + } + VdbeBranchTaken(res>0,2); + if( rc ) goto abort_due_to_error; + if( res>0 ) goto jump_to_p2; + break; +} + +/* Opcode: Destroy P1 P2 P3 * * +** +** Delete an entire database table or index whose root page in the database +** file is given by P1. +** +** The table being destroyed is in the main database file if P3==0. If +** P3==1 then the table to be clear is in the auxiliary database file +** that is used to store tables create using CREATE TEMPORARY TABLE. +** +** If AUTOVACUUM is enabled then it is possible that another root page +** might be moved into the newly deleted root page in order to keep all +** root pages contiguous at the beginning of the database. The former +** value of the root page that moved - its value before the move occurred - +** is stored in register P2. If no page movement was required (because the +** table being dropped was already the last one in the database) then a +** zero is stored in register P2. If AUTOVACUUM is disabled then a zero +** is stored in register P2. +** +** This opcode throws an error if there are any active reader VMs when +** it is invoked. This is done to avoid the difficulty associated with +** updating existing cursors when a root page is moved in an AUTOVACUUM +** database. This error is thrown even if the database is not an AUTOVACUUM +** db in order to avoid introducing an incompatibility between autovacuum +** and non-autovacuum modes. +** +** See also: Clear +*/ +case OP_Destroy: { /* out2 */ + int iMoved; + int iDb; + + sqlite3VdbeIncrWriteCounter(p, 0); + assert( p->readOnly==0 ); + assert( pOp->p1>1 ); + pOut = out2Prerelease(p, pOp); + pOut->flags = MEM_Null; + if( db->nVdbeRead > db->nVDestroy+1 ){ + rc = SQLITE_LOCKED; + p->errorAction = OE_Abort; + goto abort_due_to_error; + }else{ + iDb = pOp->p3; + assert( DbMaskTest(p->btreeMask, iDb) ); + iMoved = 0; /* Not needed. Only to silence a warning. */ + rc = sqlite3BtreeDropTable(db->aDb[iDb].pBt, pOp->p1, &iMoved); + pOut->flags = MEM_Int; + pOut->u.i = iMoved; + if( rc ) goto abort_due_to_error; +#ifndef SQLITE_OMIT_AUTOVACUUM + if( iMoved!=0 ){ + sqlite3RootPageMoved(db, iDb, iMoved, pOp->p1); + /* All OP_Destroy operations occur on the same btree */ + assert( resetSchemaOnFault==0 || resetSchemaOnFault==iDb+1 ); + resetSchemaOnFault = iDb+1; + } +#endif + } + break; +} + +/* Opcode: Clear P1 P2 P3 +** +** Delete all contents of the database table or index whose root page +** in the database file is given by P1. But, unlike Destroy, do not +** remove the table or index from the database file. +** +** The table being clear is in the main database file if P2==0. If +** P2==1 then the table to be clear is in the auxiliary database file +** that is used to store tables create using CREATE TEMPORARY TABLE. +** +** If the P3 value is non-zero, then the table referred to must be an +** intkey table (an SQL table, not an index). In this case the row change +** count is incremented by the number of rows in the table being cleared. +** If P3 is greater than zero, then the value stored in register P3 is +** also incremented by the number of rows in the table being cleared. +** +** See also: Destroy +*/ +case OP_Clear: { + int nChange; + + sqlite3VdbeIncrWriteCounter(p, 0); + nChange = 0; + assert( p->readOnly==0 ); + assert( DbMaskTest(p->btreeMask, pOp->p2) ); + rc = sqlite3BtreeClearTable( + db->aDb[pOp->p2].pBt, pOp->p1, (pOp->p3 ? &nChange : 0) + ); + if( pOp->p3 ){ + p->nChange += nChange; + if( pOp->p3>0 ){ + assert( memIsValid(&aMem[pOp->p3]) ); + memAboutToChange(p, &aMem[pOp->p3]); + aMem[pOp->p3].u.i += nChange; + } + } + if( rc ) goto abort_due_to_error; + break; +} + +/* Opcode: ResetSorter P1 * * * * +** +** Delete all contents from the ephemeral table or sorter +** that is open on cursor P1. +** +** This opcode only works for cursors used for sorting and +** opened with OP_OpenEphemeral or OP_SorterOpen. +*/ +case OP_ResetSorter: { + VdbeCursor *pC; + + assert( pOp->p1>=0 && pOp->p1nCursor ); + pC = p->apCsr[pOp->p1]; + assert( pC!=0 ); + if( isSorter(pC) ){ + sqlite3VdbeSorterReset(db, pC->uc.pSorter); + }else{ + assert( pC->eCurType==CURTYPE_BTREE ); + assert( pC->isEphemeral ); + rc = sqlite3BtreeClearTableOfCursor(pC->uc.pCursor); + if( rc ) goto abort_due_to_error; + } + break; +} + +/* Opcode: CreateBtree P1 P2 P3 * * +** Synopsis: r[P2]=root iDb=P1 flags=P3 +** +** Allocate a new b-tree in the main database file if P1==0 or in the +** TEMP database file if P1==1 or in an attached database if +** P1>1. The P3 argument must be 1 (BTREE_INTKEY) for a rowid table +** it must be 2 (BTREE_BLOBKEY) for an index or WITHOUT ROWID table. +** The root page number of the new b-tree is stored in register P2. +*/ +case OP_CreateBtree: { /* out2 */ + int pgno; + Db *pDb; + + sqlite3VdbeIncrWriteCounter(p, 0); + pOut = out2Prerelease(p, pOp); + pgno = 0; + assert( pOp->p3==BTREE_INTKEY || pOp->p3==BTREE_BLOBKEY ); + assert( pOp->p1>=0 && pOp->p1nDb ); + assert( DbMaskTest(p->btreeMask, pOp->p1) ); + assert( p->readOnly==0 ); + pDb = &db->aDb[pOp->p1]; + assert( pDb->pBt!=0 ); + rc = sqlite3BtreeCreateTable(pDb->pBt, &pgno, pOp->p3); + if( rc ) goto abort_due_to_error; + pOut->u.i = pgno; + break; +} + +/* Opcode: SqlExec * * * P4 * +** +** Run the SQL statement or statements specified in the P4 string. +*/ +case OP_SqlExec: { + sqlite3VdbeIncrWriteCounter(p, 0); + db->nSqlExec++; + rc = sqlite3_exec(db, pOp->p4.z, 0, 0, 0); + db->nSqlExec--; + if( rc ) goto abort_due_to_error; + break; +} + +/* Opcode: ParseSchema P1 * * P4 * +** +** Read and parse all entries from the SQLITE_MASTER table of database P1 +** that match the WHERE clause P4. If P4 is a NULL pointer, then the +** entire schema for P1 is reparsed. +** +** This opcode invokes the parser to create a new virtual machine, +** then runs the new virtual machine. It is thus a re-entrant opcode. +*/ +case OP_ParseSchema: { + int iDb; + const char *zMaster; + char *zSql; + InitData initData; + + /* Any prepared statement that invokes this opcode will hold mutexes + ** on every btree. This is a prerequisite for invoking + ** sqlite3InitCallback(). + */ +#ifdef SQLITE_DEBUG + for(iDb=0; iDbnDb; iDb++){ + assert( iDb==1 || sqlite3BtreeHoldsMutex(db->aDb[iDb].pBt) ); + } +#endif + + iDb = pOp->p1; + assert( iDb>=0 && iDbnDb ); + assert( DbHasProperty(db, iDb, DB_SchemaLoaded) ); + +#ifndef SQLITE_OMIT_ALTERTABLE + if( pOp->p4.z==0 ){ + sqlite3SchemaClear(db->aDb[iDb].pSchema); + db->mDbFlags &= ~DBFLAG_SchemaKnownOk; + rc = sqlite3InitOne(db, iDb, &p->zErrMsg, INITFLAG_AlterTable); + db->mDbFlags |= DBFLAG_SchemaChange; + p->expired = 0; + }else +#endif + { + zMaster = MASTER_NAME; + initData.db = db; + initData.iDb = iDb; + initData.pzErrMsg = &p->zErrMsg; + initData.mInitFlags = 0; + zSql = sqlite3MPrintf(db, + "SELECT*FROM\"%w\".%s WHERE %s ORDER BY rowid", + db->aDb[iDb].zDbSName, zMaster, pOp->p4.z); + if( zSql==0 ){ + rc = SQLITE_NOMEM_BKPT; + }else{ + assert( db->init.busy==0 ); + db->init.busy = 1; + initData.rc = SQLITE_OK; + initData.nInitRow = 0; + assert( !db->mallocFailed ); + rc = sqlite3_exec(db, zSql, sqlite3InitCallback, &initData, 0); + if( rc==SQLITE_OK ) rc = initData.rc; + if( rc==SQLITE_OK && initData.nInitRow==0 ){ + /* The OP_ParseSchema opcode with a non-NULL P4 argument should parse + ** at least one SQL statement. Any less than that indicates that + ** the sqlite_master table is corrupt. */ + rc = SQLITE_CORRUPT_BKPT; + } + sqlite3DbFreeNN(db, zSql); + db->init.busy = 0; + } + } + if( rc ){ + sqlite3ResetAllSchemasOfConnection(db); + if( rc==SQLITE_NOMEM ){ + goto no_mem; + } + goto abort_due_to_error; + } + break; +} + +#if !defined(SQLITE_OMIT_ANALYZE) +/* Opcode: LoadAnalysis P1 * * * * +** +** Read the sqlite_stat1 table for database P1 and load the content +** of that table into the internal index hash table. This will cause +** the analysis to be used when preparing all subsequent queries. +*/ +case OP_LoadAnalysis: { + assert( pOp->p1>=0 && pOp->p1nDb ); + rc = sqlite3AnalysisLoad(db, pOp->p1); + if( rc ) goto abort_due_to_error; + break; +} +#endif /* !defined(SQLITE_OMIT_ANALYZE) */ + +/* Opcode: DropTable P1 * * P4 * +** +** Remove the internal (in-memory) data structures that describe +** the table named P4 in database P1. This is called after a table +** is dropped from disk (using the Destroy opcode) in order to keep +** the internal representation of the +** schema consistent with what is on disk. +*/ +case OP_DropTable: { + sqlite3VdbeIncrWriteCounter(p, 0); + sqlite3UnlinkAndDeleteTable(db, pOp->p1, pOp->p4.z); + break; +} + +/* Opcode: DropIndex P1 * * P4 * +** +** Remove the internal (in-memory) data structures that describe +** the index named P4 in database P1. This is called after an index +** is dropped from disk (using the Destroy opcode) +** in order to keep the internal representation of the +** schema consistent with what is on disk. +*/ +case OP_DropIndex: { + sqlite3VdbeIncrWriteCounter(p, 0); + sqlite3UnlinkAndDeleteIndex(db, pOp->p1, pOp->p4.z); + break; +} + +/* Opcode: DropTrigger P1 * * P4 * +** +** Remove the internal (in-memory) data structures that describe +** the trigger named P4 in database P1. This is called after a trigger +** is dropped from disk (using the Destroy opcode) in order to keep +** the internal representation of the +** schema consistent with what is on disk. +*/ +case OP_DropTrigger: { + sqlite3VdbeIncrWriteCounter(p, 0); + sqlite3UnlinkAndDeleteTrigger(db, pOp->p1, pOp->p4.z); + break; +} + + +#ifndef SQLITE_OMIT_INTEGRITY_CHECK +/* Opcode: IntegrityCk P1 P2 P3 P4 P5 +** +** Do an analysis of the currently open database. Store in +** register P1 the text of an error message describing any problems. +** If no problems are found, store a NULL in register P1. +** +** The register P3 contains one less than the maximum number of allowed errors. +** At most reg(P3) errors will be reported. +** In other words, the analysis stops as soon as reg(P1) errors are +** seen. Reg(P1) is updated with the number of errors remaining. +** +** The root page numbers of all tables in the database are integers +** stored in P4_INTARRAY argument. +** +** If P5 is not zero, the check is done on the auxiliary database +** file, not the main database file. +** +** This opcode is used to implement the integrity_check pragma. +*/ +case OP_IntegrityCk: { + int nRoot; /* Number of tables to check. (Number of root pages.) */ + int *aRoot; /* Array of rootpage numbers for tables to be checked */ + int nErr; /* Number of errors reported */ + char *z; /* Text of the error report */ + Mem *pnErr; /* Register keeping track of errors remaining */ + + assert( p->bIsReader ); + nRoot = pOp->p2; + aRoot = pOp->p4.ai; + assert( nRoot>0 ); + assert( aRoot[0]==nRoot ); + assert( pOp->p3>0 && pOp->p3<=(p->nMem+1 - p->nCursor) ); + pnErr = &aMem[pOp->p3]; + assert( (pnErr->flags & MEM_Int)!=0 ); + assert( (pnErr->flags & (MEM_Str|MEM_Blob))==0 ); + pIn1 = &aMem[pOp->p1]; + assert( pOp->p5nDb ); + assert( DbMaskTest(p->btreeMask, pOp->p5) ); + z = sqlite3BtreeIntegrityCheck(db->aDb[pOp->p5].pBt, &aRoot[1], nRoot, + (int)pnErr->u.i+1, &nErr); + sqlite3VdbeMemSetNull(pIn1); + if( nErr==0 ){ + assert( z==0 ); + }else if( z==0 ){ + goto no_mem; + }else{ + pnErr->u.i -= nErr-1; + sqlite3VdbeMemSetStr(pIn1, z, -1, SQLITE_UTF8, sqlite3_free); + } + UPDATE_MAX_BLOBSIZE(pIn1); + sqlite3VdbeChangeEncoding(pIn1, encoding); + break; +} +#endif /* SQLITE_OMIT_INTEGRITY_CHECK */ + +/* Opcode: RowSetAdd P1 P2 * * * +** Synopsis: rowset(P1)=r[P2] +** +** Insert the integer value held by register P2 into a RowSet object +** held in register P1. +** +** An assertion fails if P2 is not an integer. +*/ +case OP_RowSetAdd: { /* in1, in2 */ + pIn1 = &aMem[pOp->p1]; + pIn2 = &aMem[pOp->p2]; + assert( (pIn2->flags & MEM_Int)!=0 ); + if( (pIn1->flags & MEM_Blob)==0 ){ + if( sqlite3VdbeMemSetRowSet(pIn1) ) goto no_mem; + } + assert( sqlite3VdbeMemIsRowSet(pIn1) ); + sqlite3RowSetInsert((RowSet*)pIn1->z, pIn2->u.i); + break; +} + +/* Opcode: RowSetRead P1 P2 P3 * * +** Synopsis: r[P3]=rowset(P1) +** +** Extract the smallest value from the RowSet object in P1 +** and put that value into register P3. +** Or, if RowSet object P1 is initially empty, leave P3 +** unchanged and jump to instruction P2. +*/ +case OP_RowSetRead: { /* jump, in1, out3 */ + i64 val; + + pIn1 = &aMem[pOp->p1]; + assert( (pIn1->flags & MEM_Blob)==0 || sqlite3VdbeMemIsRowSet(pIn1) ); + if( (pIn1->flags & MEM_Blob)==0 + || sqlite3RowSetNext((RowSet*)pIn1->z, &val)==0 + ){ + /* The boolean index is empty */ + sqlite3VdbeMemSetNull(pIn1); + VdbeBranchTaken(1,2); + goto jump_to_p2_and_check_for_interrupt; + }else{ + /* A value was pulled from the index */ + VdbeBranchTaken(0,2); + sqlite3VdbeMemSetInt64(&aMem[pOp->p3], val); + } + goto check_for_interrupt; +} + +/* Opcode: RowSetTest P1 P2 P3 P4 +** Synopsis: if r[P3] in rowset(P1) goto P2 +** +** Register P3 is assumed to hold a 64-bit integer value. If register P1 +** contains a RowSet object and that RowSet object contains +** the value held in P3, jump to register P2. Otherwise, insert the +** integer in P3 into the RowSet and continue on to the +** next opcode. +** +** The RowSet object is optimized for the case where sets of integers +** are inserted in distinct phases, which each set contains no duplicates. +** Each set is identified by a unique P4 value. The first set +** must have P4==0, the final set must have P4==-1, and for all other sets +** must have P4>0. +** +** This allows optimizations: (a) when P4==0 there is no need to test +** the RowSet object for P3, as it is guaranteed not to contain it, +** (b) when P4==-1 there is no need to insert the value, as it will +** never be tested for, and (c) when a value that is part of set X is +** inserted, there is no need to search to see if the same value was +** previously inserted as part of set X (only if it was previously +** inserted as part of some other set). +*/ +case OP_RowSetTest: { /* jump, in1, in3 */ + int iSet; + int exists; + + pIn1 = &aMem[pOp->p1]; + pIn3 = &aMem[pOp->p3]; + iSet = pOp->p4.i; + assert( pIn3->flags&MEM_Int ); + + /* If there is anything other than a rowset object in memory cell P1, + ** delete it now and initialize P1 with an empty rowset + */ + if( (pIn1->flags & MEM_Blob)==0 ){ + if( sqlite3VdbeMemSetRowSet(pIn1) ) goto no_mem; + } + assert( sqlite3VdbeMemIsRowSet(pIn1) ); + assert( pOp->p4type==P4_INT32 ); + assert( iSet==-1 || iSet>=0 ); + if( iSet ){ + exists = sqlite3RowSetTest((RowSet*)pIn1->z, iSet, pIn3->u.i); + VdbeBranchTaken(exists!=0,2); + if( exists ) goto jump_to_p2; + } + if( iSet>=0 ){ + sqlite3RowSetInsert((RowSet*)pIn1->z, pIn3->u.i); + } + break; +} + + +#ifndef SQLITE_OMIT_TRIGGER + +/* Opcode: Program P1 P2 P3 P4 P5 +** +** Execute the trigger program passed as P4 (type P4_SUBPROGRAM). +** +** P1 contains the address of the memory cell that contains the first memory +** cell in an array of values used as arguments to the sub-program. P2 +** contains the address to jump to if the sub-program throws an IGNORE +** exception using the RAISE() function. Register P3 contains the address +** of a memory cell in this (the parent) VM that is used to allocate the +** memory required by the sub-vdbe at runtime. +** +** P4 is a pointer to the VM containing the trigger program. +** +** If P5 is non-zero, then recursive program invocation is enabled. +*/ +case OP_Program: { /* jump */ + int nMem; /* Number of memory registers for sub-program */ + int nByte; /* Bytes of runtime space required for sub-program */ + Mem *pRt; /* Register to allocate runtime space */ + Mem *pMem; /* Used to iterate through memory cells */ + Mem *pEnd; /* Last memory cell in new array */ + VdbeFrame *pFrame; /* New vdbe frame to execute in */ + SubProgram *pProgram; /* Sub-program to execute */ + void *t; /* Token identifying trigger */ + + pProgram = pOp->p4.pProgram; + pRt = &aMem[pOp->p3]; + assert( pProgram->nOp>0 ); + + /* If the p5 flag is clear, then recursive invocation of triggers is + ** disabled for backwards compatibility (p5 is set if this sub-program + ** is really a trigger, not a foreign key action, and the flag set + ** and cleared by the "PRAGMA recursive_triggers" command is clear). + ** + ** It is recursive invocation of triggers, at the SQL level, that is + ** disabled. In some cases a single trigger may generate more than one + ** SubProgram (if the trigger may be executed with more than one different + ** ON CONFLICT algorithm). SubProgram structures associated with a + ** single trigger all have the same value for the SubProgram.token + ** variable. */ + if( pOp->p5 ){ + t = pProgram->token; + for(pFrame=p->pFrame; pFrame && pFrame->token!=t; pFrame=pFrame->pParent); + if( pFrame ) break; + } + + if( p->nFrame>=db->aLimit[SQLITE_LIMIT_TRIGGER_DEPTH] ){ + rc = SQLITE_ERROR; + sqlite3VdbeError(p, "too many levels of trigger recursion"); + goto abort_due_to_error; + } + + /* Register pRt is used to store the memory required to save the state + ** of the current program, and the memory required at runtime to execute + ** the trigger program. If this trigger has been fired before, then pRt + ** is already allocated. Otherwise, it must be initialized. */ + if( (pRt->flags&MEM_Blob)==0 ){ + /* SubProgram.nMem is set to the number of memory cells used by the + ** program stored in SubProgram.aOp. As well as these, one memory + ** cell is required for each cursor used by the program. Set local + ** variable nMem (and later, VdbeFrame.nChildMem) to this value. + */ + nMem = pProgram->nMem + pProgram->nCsr; + assert( nMem>0 ); + if( pProgram->nCsr==0 ) nMem++; + nByte = ROUND8(sizeof(VdbeFrame)) + + nMem * sizeof(Mem) + + pProgram->nCsr * sizeof(VdbeCursor*) + + (pProgram->nOp + 7)/8; + pFrame = sqlite3DbMallocZero(db, nByte); + if( !pFrame ){ + goto no_mem; + } + sqlite3VdbeMemRelease(pRt); + pRt->flags = MEM_Blob|MEM_Dyn; + pRt->z = (char*)pFrame; + pRt->n = nByte; + pRt->xDel = sqlite3VdbeFrameMemDel; + + pFrame->v = p; + pFrame->nChildMem = nMem; + pFrame->nChildCsr = pProgram->nCsr; + pFrame->pc = (int)(pOp - aOp); + pFrame->aMem = p->aMem; + pFrame->nMem = p->nMem; + pFrame->apCsr = p->apCsr; + pFrame->nCursor = p->nCursor; + pFrame->aOp = p->aOp; + pFrame->nOp = p->nOp; + pFrame->token = pProgram->token; +#ifdef SQLITE_ENABLE_STMT_SCANSTATUS + pFrame->anExec = p->anExec; +#endif +#ifdef SQLITE_DEBUG + pFrame->iFrameMagic = SQLITE_FRAME_MAGIC; +#endif + + pEnd = &VdbeFrameMem(pFrame)[pFrame->nChildMem]; + for(pMem=VdbeFrameMem(pFrame); pMem!=pEnd; pMem++){ + pMem->flags = MEM_Undefined; + pMem->db = db; + } + }else{ + pFrame = (VdbeFrame*)pRt->z; + assert( pRt->xDel==sqlite3VdbeFrameMemDel ); + assert( pProgram->nMem+pProgram->nCsr==pFrame->nChildMem + || (pProgram->nCsr==0 && pProgram->nMem+1==pFrame->nChildMem) ); + assert( pProgram->nCsr==pFrame->nChildCsr ); + assert( (int)(pOp - aOp)==pFrame->pc ); + } + + p->nFrame++; + pFrame->pParent = p->pFrame; + pFrame->lastRowid = db->lastRowid; + pFrame->nChange = p->nChange; + pFrame->nDbChange = p->db->nChange; + assert( pFrame->pAuxData==0 ); + pFrame->pAuxData = p->pAuxData; + p->pAuxData = 0; + p->nChange = 0; + p->pFrame = pFrame; + p->aMem = aMem = VdbeFrameMem(pFrame); + p->nMem = pFrame->nChildMem; + p->nCursor = (u16)pFrame->nChildCsr; + p->apCsr = (VdbeCursor **)&aMem[p->nMem]; + pFrame->aOnce = (u8*)&p->apCsr[pProgram->nCsr]; + memset(pFrame->aOnce, 0, (pProgram->nOp + 7)/8); + p->aOp = aOp = pProgram->aOp; + p->nOp = pProgram->nOp; +#ifdef SQLITE_ENABLE_STMT_SCANSTATUS + p->anExec = 0; +#endif +#ifdef SQLITE_DEBUG + /* Verify that second and subsequent executions of the same trigger do not + ** try to reuse register values from the first use. */ + { + int i; + for(i=0; inMem; i++){ + aMem[i].pScopyFrom = 0; /* Prevent false-positive AboutToChange() errs */ + aMem[i].flags |= MEM_Undefined; /* Cause a fault if this reg is reused */ + } + } +#endif + pOp = &aOp[-1]; + goto check_for_interrupt; +} + +/* Opcode: Param P1 P2 * * * +** +** This opcode is only ever present in sub-programs called via the +** OP_Program instruction. Copy a value currently stored in a memory +** cell of the calling (parent) frame to cell P2 in the current frames +** address space. This is used by trigger programs to access the new.* +** and old.* values. +** +** The address of the cell in the parent frame is determined by adding +** the value of the P1 argument to the value of the P1 argument to the +** calling OP_Program instruction. +*/ +case OP_Param: { /* out2 */ + VdbeFrame *pFrame; + Mem *pIn; + pOut = out2Prerelease(p, pOp); + pFrame = p->pFrame; + pIn = &pFrame->aMem[pOp->p1 + pFrame->aOp[pFrame->pc].p1]; + sqlite3VdbeMemShallowCopy(pOut, pIn, MEM_Ephem); + break; +} + +#endif /* #ifndef SQLITE_OMIT_TRIGGER */ + +#ifndef SQLITE_OMIT_FOREIGN_KEY +/* Opcode: FkCounter P1 P2 * * * +** Synopsis: fkctr[P1]+=P2 +** +** Increment a "constraint counter" by P2 (P2 may be negative or positive). +** If P1 is non-zero, the database constraint counter is incremented +** (deferred foreign key constraints). Otherwise, if P1 is zero, the +** statement counter is incremented (immediate foreign key constraints). +*/ +case OP_FkCounter: { + if( db->flags & SQLITE_DeferFKs ){ + db->nDeferredImmCons += pOp->p2; + }else if( pOp->p1 ){ + db->nDeferredCons += pOp->p2; + }else{ + p->nFkConstraint += pOp->p2; + } + break; +} + +/* Opcode: FkIfZero P1 P2 * * * +** Synopsis: if fkctr[P1]==0 goto P2 +** +** This opcode tests if a foreign key constraint-counter is currently zero. +** If so, jump to instruction P2. Otherwise, fall through to the next +** instruction. +** +** If P1 is non-zero, then the jump is taken if the database constraint-counter +** is zero (the one that counts deferred constraint violations). If P1 is +** zero, the jump is taken if the statement constraint-counter is zero +** (immediate foreign key constraint violations). +*/ +case OP_FkIfZero: { /* jump */ + if( pOp->p1 ){ + VdbeBranchTaken(db->nDeferredCons==0 && db->nDeferredImmCons==0, 2); + if( db->nDeferredCons==0 && db->nDeferredImmCons==0 ) goto jump_to_p2; + }else{ + VdbeBranchTaken(p->nFkConstraint==0 && db->nDeferredImmCons==0, 2); + if( p->nFkConstraint==0 && db->nDeferredImmCons==0 ) goto jump_to_p2; + } + break; +} +#endif /* #ifndef SQLITE_OMIT_FOREIGN_KEY */ + +#ifndef SQLITE_OMIT_AUTOINCREMENT +/* Opcode: MemMax P1 P2 * * * +** Synopsis: r[P1]=max(r[P1],r[P2]) +** +** P1 is a register in the root frame of this VM (the root frame is +** different from the current frame if this instruction is being executed +** within a sub-program). Set the value of register P1 to the maximum of +** its current value and the value in register P2. +** +** This instruction throws an error if the memory cell is not initially +** an integer. +*/ +case OP_MemMax: { /* in2 */ + VdbeFrame *pFrame; + if( p->pFrame ){ + for(pFrame=p->pFrame; pFrame->pParent; pFrame=pFrame->pParent); + pIn1 = &pFrame->aMem[pOp->p1]; + }else{ + pIn1 = &aMem[pOp->p1]; + } + assert( memIsValid(pIn1) ); + sqlite3VdbeMemIntegerify(pIn1); + pIn2 = &aMem[pOp->p2]; + sqlite3VdbeMemIntegerify(pIn2); + if( pIn1->u.iu.i){ + pIn1->u.i = pIn2->u.i; + } + break; +} +#endif /* SQLITE_OMIT_AUTOINCREMENT */ + +/* Opcode: IfPos P1 P2 P3 * * +** Synopsis: if r[P1]>0 then r[P1]-=P3, goto P2 +** +** Register P1 must contain an integer. +** If the value of register P1 is 1 or greater, subtract P3 from the +** value in P1 and jump to P2. +** +** If the initial value of register P1 is less than 1, then the +** value is unchanged and control passes through to the next instruction. +*/ +case OP_IfPos: { /* jump, in1 */ + pIn1 = &aMem[pOp->p1]; + assert( pIn1->flags&MEM_Int ); + VdbeBranchTaken( pIn1->u.i>0, 2); + if( pIn1->u.i>0 ){ + pIn1->u.i -= pOp->p3; + goto jump_to_p2; + } + break; +} + +/* Opcode: OffsetLimit P1 P2 P3 * * +** Synopsis: if r[P1]>0 then r[P2]=r[P1]+max(0,r[P3]) else r[P2]=(-1) +** +** This opcode performs a commonly used computation associated with +** LIMIT and OFFSET process. r[P1] holds the limit counter. r[P3] +** holds the offset counter. The opcode computes the combined value +** of the LIMIT and OFFSET and stores that value in r[P2]. The r[P2] +** value computed is the total number of rows that will need to be +** visited in order to complete the query. +** +** If r[P3] is zero or negative, that means there is no OFFSET +** and r[P2] is set to be the value of the LIMIT, r[P1]. +** +** if r[P1] is zero or negative, that means there is no LIMIT +** and r[P2] is set to -1. +** +** Otherwise, r[P2] is set to the sum of r[P1] and r[P3]. +*/ +case OP_OffsetLimit: { /* in1, out2, in3 */ + i64 x; + pIn1 = &aMem[pOp->p1]; + pIn3 = &aMem[pOp->p3]; + pOut = out2Prerelease(p, pOp); + assert( pIn1->flags & MEM_Int ); + assert( pIn3->flags & MEM_Int ); + x = pIn1->u.i; + if( x<=0 || sqlite3AddInt64(&x, pIn3->u.i>0?pIn3->u.i:0) ){ + /* If the LIMIT is less than or equal to zero, loop forever. This + ** is documented. But also, if the LIMIT+OFFSET exceeds 2^63 then + ** also loop forever. This is undocumented. In fact, one could argue + ** that the loop should terminate. But assuming 1 billion iterations + ** per second (far exceeding the capabilities of any current hardware) + ** it would take nearly 300 years to actually reach the limit. So + ** looping forever is a reasonable approximation. */ + pOut->u.i = -1; + }else{ + pOut->u.i = x; + } + break; +} + +/* Opcode: IfNotZero P1 P2 * * * +** Synopsis: if r[P1]!=0 then r[P1]--, goto P2 +** +** Register P1 must contain an integer. If the content of register P1 is +** initially greater than zero, then decrement the value in register P1. +** If it is non-zero (negative or positive) and then also jump to P2. +** If register P1 is initially zero, leave it unchanged and fall through. +*/ +case OP_IfNotZero: { /* jump, in1 */ + pIn1 = &aMem[pOp->p1]; + assert( pIn1->flags&MEM_Int ); + VdbeBranchTaken(pIn1->u.i<0, 2); + if( pIn1->u.i ){ + if( pIn1->u.i>0 ) pIn1->u.i--; + goto jump_to_p2; + } + break; +} + +/* Opcode: DecrJumpZero P1 P2 * * * +** Synopsis: if (--r[P1])==0 goto P2 +** +** Register P1 must hold an integer. Decrement the value in P1 +** and jump to P2 if the new value is exactly zero. +*/ +case OP_DecrJumpZero: { /* jump, in1 */ + pIn1 = &aMem[pOp->p1]; + assert( pIn1->flags&MEM_Int ); + if( pIn1->u.i>SMALLEST_INT64 ) pIn1->u.i--; + VdbeBranchTaken(pIn1->u.i==0, 2); + if( pIn1->u.i==0 ) goto jump_to_p2; + break; +} + + +/* Opcode: AggStep * P2 P3 P4 P5 +** Synopsis: accum=r[P3] step(r[P2@P5]) +** +** Execute the xStep function for an aggregate. +** The function has P5 arguments. P4 is a pointer to the +** FuncDef structure that specifies the function. Register P3 is the +** accumulator. +** +** The P5 arguments are taken from register P2 and its +** successors. +*/ +/* Opcode: AggInverse * P2 P3 P4 P5 +** Synopsis: accum=r[P3] inverse(r[P2@P5]) +** +** Execute the xInverse function for an aggregate. +** The function has P5 arguments. P4 is a pointer to the +** FuncDef structure that specifies the function. Register P3 is the +** accumulator. +** +** The P5 arguments are taken from register P2 and its +** successors. +*/ +/* Opcode: AggStep1 P1 P2 P3 P4 P5 +** Synopsis: accum=r[P3] step(r[P2@P5]) +** +** Execute the xStep (if P1==0) or xInverse (if P1!=0) function for an +** aggregate. The function has P5 arguments. P4 is a pointer to the +** FuncDef structure that specifies the function. Register P3 is the +** accumulator. +** +** The P5 arguments are taken from register P2 and its +** successors. +** +** This opcode is initially coded as OP_AggStep0. On first evaluation, +** the FuncDef stored in P4 is converted into an sqlite3_context and +** the opcode is changed. In this way, the initialization of the +** sqlite3_context only happens once, instead of on each call to the +** step function. +*/ +case OP_AggInverse: +case OP_AggStep: { + int n; + sqlite3_context *pCtx; + + assert( pOp->p4type==P4_FUNCDEF ); + n = pOp->p5; + assert( pOp->p3>0 && pOp->p3<=(p->nMem+1 - p->nCursor) ); + assert( n==0 || (pOp->p2>0 && pOp->p2+n<=(p->nMem+1 - p->nCursor)+1) ); + assert( pOp->p3p2 || pOp->p3>=pOp->p2+n ); + pCtx = sqlite3DbMallocRawNN(db, n*sizeof(sqlite3_value*) + + (sizeof(pCtx[0]) + sizeof(Mem) - sizeof(sqlite3_value*))); + if( pCtx==0 ) goto no_mem; + pCtx->pMem = 0; + pCtx->pOut = (Mem*)&(pCtx->argv[n]); + sqlite3VdbeMemInit(pCtx->pOut, db, MEM_Null); + pCtx->pFunc = pOp->p4.pFunc; + pCtx->iOp = (int)(pOp - aOp); + pCtx->pVdbe = p; + pCtx->skipFlag = 0; + pCtx->isError = 0; + pCtx->argc = n; + pOp->p4type = P4_FUNCCTX; + pOp->p4.pCtx = pCtx; + + /* OP_AggInverse must have P1==1 and OP_AggStep must have P1==0 */ + assert( pOp->p1==(pOp->opcode==OP_AggInverse) ); + + pOp->opcode = OP_AggStep1; + /* Fall through into OP_AggStep */ +} +case OP_AggStep1: { + int i; + sqlite3_context *pCtx; + Mem *pMem; + + assert( pOp->p4type==P4_FUNCCTX ); + pCtx = pOp->p4.pCtx; + pMem = &aMem[pOp->p3]; + +#ifdef SQLITE_DEBUG + if( pOp->p1 ){ + /* This is an OP_AggInverse call. Verify that xStep has always + ** been called at least once prior to any xInverse call. */ + assert( pMem->uTemp==0x1122e0e3 ); + }else{ + /* This is an OP_AggStep call. Mark it as such. */ + pMem->uTemp = 0x1122e0e3; + } +#endif + + /* If this function is inside of a trigger, the register array in aMem[] + ** might change from one evaluation to the next. The next block of code + ** checks to see if the register array has changed, and if so it + ** reinitializes the relavant parts of the sqlite3_context object */ + if( pCtx->pMem != pMem ){ + pCtx->pMem = pMem; + for(i=pCtx->argc-1; i>=0; i--) pCtx->argv[i] = &aMem[pOp->p2+i]; + } + +#ifdef SQLITE_DEBUG + for(i=0; iargc; i++){ + assert( memIsValid(pCtx->argv[i]) ); + REGISTER_TRACE(pOp->p2+i, pCtx->argv[i]); + } +#endif + + pMem->n++; + assert( pCtx->pOut->flags==MEM_Null ); + assert( pCtx->isError==0 ); + assert( pCtx->skipFlag==0 ); +#ifndef SQLITE_OMIT_WINDOWFUNC + if( pOp->p1 ){ + (pCtx->pFunc->xInverse)(pCtx,pCtx->argc,pCtx->argv); + }else +#endif + (pCtx->pFunc->xSFunc)(pCtx,pCtx->argc,pCtx->argv); /* IMP: R-24505-23230 */ + + if( pCtx->isError ){ + if( pCtx->isError>0 ){ + sqlite3VdbeError(p, "%s", sqlite3_value_text(pCtx->pOut)); + rc = pCtx->isError; + } + if( pCtx->skipFlag ){ + assert( pOp[-1].opcode==OP_CollSeq ); + i = pOp[-1].p1; + if( i ) sqlite3VdbeMemSetInt64(&aMem[i], 1); + pCtx->skipFlag = 0; + } + sqlite3VdbeMemRelease(pCtx->pOut); + pCtx->pOut->flags = MEM_Null; + pCtx->isError = 0; + if( rc ) goto abort_due_to_error; + } + assert( pCtx->pOut->flags==MEM_Null ); + assert( pCtx->skipFlag==0 ); + break; +} + +/* Opcode: AggFinal P1 P2 * P4 * +** Synopsis: accum=r[P1] N=P2 +** +** P1 is the memory location that is the accumulator for an aggregate +** or window function. Execute the finalizer function +** for an aggregate and store the result in P1. +** +** P2 is the number of arguments that the step function takes and +** P4 is a pointer to the FuncDef for this function. The P2 +** argument is not used by this opcode. It is only there to disambiguate +** functions that can take varying numbers of arguments. The +** P4 argument is only needed for the case where +** the step function was not previously called. +*/ +/* Opcode: AggValue * P2 P3 P4 * +** Synopsis: r[P3]=value N=P2 +** +** Invoke the xValue() function and store the result in register P3. +** +** P2 is the number of arguments that the step function takes and +** P4 is a pointer to the FuncDef for this function. The P2 +** argument is not used by this opcode. It is only there to disambiguate +** functions that can take varying numbers of arguments. The +** P4 argument is only needed for the case where +** the step function was not previously called. +*/ +case OP_AggValue: +case OP_AggFinal: { + Mem *pMem; + assert( pOp->p1>0 && pOp->p1<=(p->nMem+1 - p->nCursor) ); + assert( pOp->p3==0 || pOp->opcode==OP_AggValue ); + pMem = &aMem[pOp->p1]; + assert( (pMem->flags & ~(MEM_Null|MEM_Agg))==0 ); +#ifndef SQLITE_OMIT_WINDOWFUNC + if( pOp->p3 ){ + memAboutToChange(p, &aMem[pOp->p3]); + rc = sqlite3VdbeMemAggValue(pMem, &aMem[pOp->p3], pOp->p4.pFunc); + pMem = &aMem[pOp->p3]; + }else +#endif + { + rc = sqlite3VdbeMemFinalize(pMem, pOp->p4.pFunc); + } + + if( rc ){ + sqlite3VdbeError(p, "%s", sqlite3_value_text(pMem)); + goto abort_due_to_error; + } + sqlite3VdbeChangeEncoding(pMem, encoding); + UPDATE_MAX_BLOBSIZE(pMem); + if( sqlite3VdbeMemTooBig(pMem) ){ + goto too_big; + } + break; +} + +#ifndef SQLITE_OMIT_WAL +/* Opcode: Checkpoint P1 P2 P3 * * +** +** Checkpoint database P1. This is a no-op if P1 is not currently in +** WAL mode. Parameter P2 is one of SQLITE_CHECKPOINT_PASSIVE, FULL, +** RESTART, or TRUNCATE. Write 1 or 0 into mem[P3] if the checkpoint returns +** SQLITE_BUSY or not, respectively. Write the number of pages in the +** WAL after the checkpoint into mem[P3+1] and the number of pages +** in the WAL that have been checkpointed after the checkpoint +** completes into mem[P3+2]. However on an error, mem[P3+1] and +** mem[P3+2] are initialized to -1. +*/ +case OP_Checkpoint: { + int i; /* Loop counter */ + int aRes[3]; /* Results */ + Mem *pMem; /* Write results here */ + + assert( p->readOnly==0 ); + aRes[0] = 0; + aRes[1] = aRes[2] = -1; + assert( pOp->p2==SQLITE_CHECKPOINT_PASSIVE + || pOp->p2==SQLITE_CHECKPOINT_FULL + || pOp->p2==SQLITE_CHECKPOINT_RESTART + || pOp->p2==SQLITE_CHECKPOINT_TRUNCATE + ); + rc = sqlite3Checkpoint(db, pOp->p1, pOp->p2, &aRes[1], &aRes[2]); + if( rc ){ + if( rc!=SQLITE_BUSY ) goto abort_due_to_error; + rc = SQLITE_OK; + aRes[0] = 1; + } + for(i=0, pMem = &aMem[pOp->p3]; i<3; i++, pMem++){ + sqlite3VdbeMemSetInt64(pMem, (i64)aRes[i]); + } + break; +}; +#endif + +#ifndef SQLITE_OMIT_PRAGMA +/* Opcode: JournalMode P1 P2 P3 * * +** +** Change the journal mode of database P1 to P3. P3 must be one of the +** PAGER_JOURNALMODE_XXX values. If changing between the various rollback +** modes (delete, truncate, persist, off and memory), this is a simple +** operation. No IO is required. +** +** If changing into or out of WAL mode the procedure is more complicated. +** +** Write a string containing the final journal-mode to register P2. +*/ +case OP_JournalMode: { /* out2 */ + Btree *pBt; /* Btree to change journal mode of */ + Pager *pPager; /* Pager associated with pBt */ + int eNew; /* New journal mode */ + int eOld; /* The old journal mode */ +#ifndef SQLITE_OMIT_WAL + const char *zFilename; /* Name of database file for pPager */ +#endif + + pOut = out2Prerelease(p, pOp); + eNew = pOp->p3; + assert( eNew==PAGER_JOURNALMODE_DELETE + || eNew==PAGER_JOURNALMODE_TRUNCATE + || eNew==PAGER_JOURNALMODE_PERSIST + || eNew==PAGER_JOURNALMODE_OFF + || eNew==PAGER_JOURNALMODE_MEMORY + || eNew==PAGER_JOURNALMODE_WAL + || eNew==PAGER_JOURNALMODE_QUERY + ); + assert( pOp->p1>=0 && pOp->p1nDb ); + assert( p->readOnly==0 ); + + pBt = db->aDb[pOp->p1].pBt; + pPager = sqlite3BtreePager(pBt); + eOld = sqlite3PagerGetJournalMode(pPager); + if( eNew==PAGER_JOURNALMODE_QUERY ) eNew = eOld; + if( !sqlite3PagerOkToChangeJournalMode(pPager) ) eNew = eOld; + +#ifndef SQLITE_OMIT_WAL + zFilename = sqlite3PagerFilename(pPager, 1); + + /* Do not allow a transition to journal_mode=WAL for a database + ** in temporary storage or if the VFS does not support shared memory + */ + if( eNew==PAGER_JOURNALMODE_WAL + && (sqlite3Strlen30(zFilename)==0 /* Temp file */ + || !sqlite3PagerWalSupported(pPager)) /* No shared-memory support */ + ){ + eNew = eOld; + } + + if( (eNew!=eOld) + && (eOld==PAGER_JOURNALMODE_WAL || eNew==PAGER_JOURNALMODE_WAL) + ){ + if( !db->autoCommit || db->nVdbeRead>1 ){ + rc = SQLITE_ERROR; + sqlite3VdbeError(p, + "cannot change %s wal mode from within a transaction", + (eNew==PAGER_JOURNALMODE_WAL ? "into" : "out of") + ); + goto abort_due_to_error; + }else{ + + if( eOld==PAGER_JOURNALMODE_WAL ){ + /* If leaving WAL mode, close the log file. If successful, the call + ** to PagerCloseWal() checkpoints and deletes the write-ahead-log + ** file. An EXCLUSIVE lock may still be held on the database file + ** after a successful return. + */ + rc = sqlite3PagerCloseWal(pPager, db); + if( rc==SQLITE_OK ){ + sqlite3PagerSetJournalMode(pPager, eNew); + } + }else if( eOld==PAGER_JOURNALMODE_MEMORY ){ + /* Cannot transition directly from MEMORY to WAL. Use mode OFF + ** as an intermediate */ + sqlite3PagerSetJournalMode(pPager, PAGER_JOURNALMODE_OFF); + } + + /* Open a transaction on the database file. Regardless of the journal + ** mode, this transaction always uses a rollback journal. + */ + assert( sqlite3BtreeIsInTrans(pBt)==0 ); + if( rc==SQLITE_OK ){ + rc = sqlite3BtreeSetVersion(pBt, (eNew==PAGER_JOURNALMODE_WAL ? 2 : 1)); + } + } + } +#endif /* ifndef SQLITE_OMIT_WAL */ + + if( rc ) eNew = eOld; + eNew = sqlite3PagerSetJournalMode(pPager, eNew); + + pOut->flags = MEM_Str|MEM_Static|MEM_Term; + pOut->z = (char *)sqlite3JournalModename(eNew); + pOut->n = sqlite3Strlen30(pOut->z); + pOut->enc = SQLITE_UTF8; + sqlite3VdbeChangeEncoding(pOut, encoding); + if( rc ) goto abort_due_to_error; + break; +}; +#endif /* SQLITE_OMIT_PRAGMA */ + +#if !defined(SQLITE_OMIT_VACUUM) && !defined(SQLITE_OMIT_ATTACH) +/* Opcode: Vacuum P1 P2 * * * +** +** Vacuum the entire database P1. P1 is 0 for "main", and 2 or more +** for an attached database. The "temp" database may not be vacuumed. +** +** If P2 is not zero, then it is a register holding a string which is +** the file into which the result of vacuum should be written. When +** P2 is zero, the vacuum overwrites the original database. +*/ +case OP_Vacuum: { + assert( p->readOnly==0 ); + rc = sqlite3RunVacuum(&p->zErrMsg, db, pOp->p1, + pOp->p2 ? &aMem[pOp->p2] : 0); + if( rc ) goto abort_due_to_error; + break; +} +#endif + +#if !defined(SQLITE_OMIT_AUTOVACUUM) +/* Opcode: IncrVacuum P1 P2 * * * +** +** Perform a single step of the incremental vacuum procedure on +** the P1 database. If the vacuum has finished, jump to instruction +** P2. Otherwise, fall through to the next instruction. +*/ +case OP_IncrVacuum: { /* jump */ + Btree *pBt; + + assert( pOp->p1>=0 && pOp->p1nDb ); + assert( DbMaskTest(p->btreeMask, pOp->p1) ); + assert( p->readOnly==0 ); + pBt = db->aDb[pOp->p1].pBt; + rc = sqlite3BtreeIncrVacuum(pBt); + VdbeBranchTaken(rc==SQLITE_DONE,2); + if( rc ){ + if( rc!=SQLITE_DONE ) goto abort_due_to_error; + rc = SQLITE_OK; + goto jump_to_p2; + } + break; +} +#endif + +/* Opcode: Expire P1 P2 * * * +** +** Cause precompiled statements to expire. When an expired statement +** is executed using sqlite3_step() it will either automatically +** reprepare itself (if it was originally created using sqlite3_prepare_v2()) +** or it will fail with SQLITE_SCHEMA. +** +** If P1 is 0, then all SQL statements become expired. If P1 is non-zero, +** then only the currently executing statement is expired. +** +** If P2 is 0, then SQL statements are expired immediately. If P2 is 1, +** then running SQL statements are allowed to continue to run to completion. +** The P2==1 case occurs when a CREATE INDEX or similar schema change happens +** that might help the statement run faster but which does not affect the +** correctness of operation. +*/ +case OP_Expire: { + assert( pOp->p2==0 || pOp->p2==1 ); + if( !pOp->p1 ){ + sqlite3ExpirePreparedStatements(db, pOp->p2); + }else{ + p->expired = pOp->p2+1; + } + break; +} + +#ifndef SQLITE_OMIT_SHARED_CACHE +/* Opcode: TableLock P1 P2 P3 P4 * +** Synopsis: iDb=P1 root=P2 write=P3 +** +** Obtain a lock on a particular table. This instruction is only used when +** the shared-cache feature is enabled. +** +** P1 is the index of the database in sqlite3.aDb[] of the database +** on which the lock is acquired. A readlock is obtained if P3==0 or +** a write lock if P3==1. +** +** P2 contains the root-page of the table to lock. +** +** P4 contains a pointer to the name of the table being locked. This is only +** used to generate an error message if the lock cannot be obtained. +*/ +case OP_TableLock: { + u8 isWriteLock = (u8)pOp->p3; + if( isWriteLock || 0==(db->flags&SQLITE_ReadUncommit) ){ + int p1 = pOp->p1; + assert( p1>=0 && p1nDb ); + assert( DbMaskTest(p->btreeMask, p1) ); + assert( isWriteLock==0 || isWriteLock==1 ); + rc = sqlite3BtreeLockTable(db->aDb[p1].pBt, pOp->p2, isWriteLock); + if( rc ){ + if( (rc&0xFF)==SQLITE_LOCKED ){ + const char *z = pOp->p4.z; + sqlite3VdbeError(p, "database table is locked: %s", z); + } + goto abort_due_to_error; + } + } + break; +} +#endif /* SQLITE_OMIT_SHARED_CACHE */ + +#ifndef SQLITE_OMIT_VIRTUALTABLE +/* Opcode: VBegin * * * P4 * +** +** P4 may be a pointer to an sqlite3_vtab structure. If so, call the +** xBegin method for that table. +** +** Also, whether or not P4 is set, check that this is not being called from +** within a callback to a virtual table xSync() method. If it is, the error +** code will be set to SQLITE_LOCKED. +*/ +case OP_VBegin: { + VTable *pVTab; + pVTab = pOp->p4.pVtab; + rc = sqlite3VtabBegin(db, pVTab); + if( pVTab ) sqlite3VtabImportErrmsg(p, pVTab->pVtab); + if( rc ) goto abort_due_to_error; + break; +} +#endif /* SQLITE_OMIT_VIRTUALTABLE */ + +#ifndef SQLITE_OMIT_VIRTUALTABLE +/* Opcode: VCreate P1 P2 * * * +** +** P2 is a register that holds the name of a virtual table in database +** P1. Call the xCreate method for that table. +*/ +case OP_VCreate: { + Mem sMem; /* For storing the record being decoded */ + const char *zTab; /* Name of the virtual table */ + + memset(&sMem, 0, sizeof(sMem)); + sMem.db = db; + /* Because P2 is always a static string, it is impossible for the + ** sqlite3VdbeMemCopy() to fail */ + assert( (aMem[pOp->p2].flags & MEM_Str)!=0 ); + assert( (aMem[pOp->p2].flags & MEM_Static)!=0 ); + rc = sqlite3VdbeMemCopy(&sMem, &aMem[pOp->p2]); + assert( rc==SQLITE_OK ); + zTab = (const char*)sqlite3_value_text(&sMem); + assert( zTab || db->mallocFailed ); + if( zTab ){ + rc = sqlite3VtabCallCreate(db, pOp->p1, zTab, &p->zErrMsg); + } + sqlite3VdbeMemRelease(&sMem); + if( rc ) goto abort_due_to_error; + break; +} +#endif /* SQLITE_OMIT_VIRTUALTABLE */ + +#ifndef SQLITE_OMIT_VIRTUALTABLE +/* Opcode: VDestroy P1 * * P4 * +** +** P4 is the name of a virtual table in database P1. Call the xDestroy method +** of that table. +*/ +case OP_VDestroy: { + db->nVDestroy++; + rc = sqlite3VtabCallDestroy(db, pOp->p1, pOp->p4.z); + db->nVDestroy--; + assert( p->errorAction==OE_Abort && p->usesStmtJournal ); + if( rc ) goto abort_due_to_error; + break; +} +#endif /* SQLITE_OMIT_VIRTUALTABLE */ + +#ifndef SQLITE_OMIT_VIRTUALTABLE +/* Opcode: VOpen P1 * * P4 * +** +** P4 is a pointer to a virtual table object, an sqlite3_vtab structure. +** P1 is a cursor number. This opcode opens a cursor to the virtual +** table and stores that cursor in P1. +*/ +case OP_VOpen: { + VdbeCursor *pCur; + sqlite3_vtab_cursor *pVCur; + sqlite3_vtab *pVtab; + const sqlite3_module *pModule; + + assert( p->bIsReader ); + pCur = 0; + pVCur = 0; + pVtab = pOp->p4.pVtab->pVtab; + if( pVtab==0 || NEVER(pVtab->pModule==0) ){ + rc = SQLITE_LOCKED; + goto abort_due_to_error; + } + pModule = pVtab->pModule; + rc = pModule->xOpen(pVtab, &pVCur); + sqlite3VtabImportErrmsg(p, pVtab); + if( rc ) goto abort_due_to_error; + + /* Initialize sqlite3_vtab_cursor base class */ + pVCur->pVtab = pVtab; + + /* Initialize vdbe cursor object */ + pCur = allocateCursor(p, pOp->p1, 0, -1, CURTYPE_VTAB); + if( pCur ){ + pCur->uc.pVCur = pVCur; + pVtab->nRef++; + }else{ + assert( db->mallocFailed ); + pModule->xClose(pVCur); + goto no_mem; + } + break; +} +#endif /* SQLITE_OMIT_VIRTUALTABLE */ + +#ifndef SQLITE_OMIT_VIRTUALTABLE +/* Opcode: VFilter P1 P2 P3 P4 * +** Synopsis: iplan=r[P3] zplan='P4' +** +** P1 is a cursor opened using VOpen. P2 is an address to jump to if +** the filtered result set is empty. +** +** P4 is either NULL or a string that was generated by the xBestIndex +** method of the module. The interpretation of the P4 string is left +** to the module implementation. +** +** This opcode invokes the xFilter method on the virtual table specified +** by P1. The integer query plan parameter to xFilter is stored in register +** P3. Register P3+1 stores the argc parameter to be passed to the +** xFilter method. Registers P3+2..P3+1+argc are the argc +** additional parameters which are passed to +** xFilter as argv. Register P3+2 becomes argv[0] when passed to xFilter. +** +** A jump is made to P2 if the result set after filtering would be empty. +*/ +case OP_VFilter: { /* jump */ + int nArg; + int iQuery; + const sqlite3_module *pModule; + Mem *pQuery; + Mem *pArgc; + sqlite3_vtab_cursor *pVCur; + sqlite3_vtab *pVtab; + VdbeCursor *pCur; + int res; + int i; + Mem **apArg; + + pQuery = &aMem[pOp->p3]; + pArgc = &pQuery[1]; + pCur = p->apCsr[pOp->p1]; + assert( memIsValid(pQuery) ); + REGISTER_TRACE(pOp->p3, pQuery); + assert( pCur->eCurType==CURTYPE_VTAB ); + pVCur = pCur->uc.pVCur; + pVtab = pVCur->pVtab; + pModule = pVtab->pModule; + + /* Grab the index number and argc parameters */ + assert( (pQuery->flags&MEM_Int)!=0 && pArgc->flags==MEM_Int ); + nArg = (int)pArgc->u.i; + iQuery = (int)pQuery->u.i; + + /* Invoke the xFilter method */ + res = 0; + apArg = p->apArg; + for(i = 0; ixFilter(pVCur, iQuery, pOp->p4.z, nArg, apArg); + sqlite3VtabImportErrmsg(p, pVtab); + if( rc ) goto abort_due_to_error; + res = pModule->xEof(pVCur); + pCur->nullRow = 0; + VdbeBranchTaken(res!=0,2); + if( res ) goto jump_to_p2; + break; +} +#endif /* SQLITE_OMIT_VIRTUALTABLE */ + +#ifndef SQLITE_OMIT_VIRTUALTABLE +/* Opcode: VColumn P1 P2 P3 * P5 +** Synopsis: r[P3]=vcolumn(P2) +** +** Store in register P3 the value of the P2-th column of +** the current row of the virtual-table of cursor P1. +** +** If the VColumn opcode is being used to fetch the value of +** an unchanging column during an UPDATE operation, then the P5 +** value is OPFLAG_NOCHNG. This will cause the sqlite3_vtab_nochange() +** function to return true inside the xColumn method of the virtual +** table implementation. The P5 column might also contain other +** bits (OPFLAG_LENGTHARG or OPFLAG_TYPEOFARG) but those bits are +** unused by OP_VColumn. +*/ +case OP_VColumn: { + sqlite3_vtab *pVtab; + const sqlite3_module *pModule; + Mem *pDest; + sqlite3_context sContext; + + VdbeCursor *pCur = p->apCsr[pOp->p1]; + assert( pCur->eCurType==CURTYPE_VTAB ); + assert( pOp->p3>0 && pOp->p3<=(p->nMem+1 - p->nCursor) ); + pDest = &aMem[pOp->p3]; + memAboutToChange(p, pDest); + if( pCur->nullRow ){ + sqlite3VdbeMemSetNull(pDest); + break; + } + pVtab = pCur->uc.pVCur->pVtab; + pModule = pVtab->pModule; + assert( pModule->xColumn ); + memset(&sContext, 0, sizeof(sContext)); + sContext.pOut = pDest; + testcase( (pOp->p5 & OPFLAG_NOCHNG)==0 && pOp->p5!=0 ); + if( pOp->p5 & OPFLAG_NOCHNG ){ + sqlite3VdbeMemSetNull(pDest); + pDest->flags = MEM_Null|MEM_Zero; + pDest->u.nZero = 0; + }else{ + MemSetTypeFlag(pDest, MEM_Null); + } + rc = pModule->xColumn(pCur->uc.pVCur, &sContext, pOp->p2); + sqlite3VtabImportErrmsg(p, pVtab); + if( sContext.isError>0 ){ + sqlite3VdbeError(p, "%s", sqlite3_value_text(pDest)); + rc = sContext.isError; + } + sqlite3VdbeChangeEncoding(pDest, encoding); + REGISTER_TRACE(pOp->p3, pDest); + UPDATE_MAX_BLOBSIZE(pDest); + + if( sqlite3VdbeMemTooBig(pDest) ){ + goto too_big; + } + if( rc ) goto abort_due_to_error; + break; +} +#endif /* SQLITE_OMIT_VIRTUALTABLE */ + +#ifndef SQLITE_OMIT_VIRTUALTABLE +/* Opcode: VNext P1 P2 * * * +** +** Advance virtual table P1 to the next row in its result set and +** jump to instruction P2. Or, if the virtual table has reached +** the end of its result set, then fall through to the next instruction. +*/ +case OP_VNext: { /* jump */ + sqlite3_vtab *pVtab; + const sqlite3_module *pModule; + int res; + VdbeCursor *pCur; + + res = 0; + pCur = p->apCsr[pOp->p1]; + assert( pCur->eCurType==CURTYPE_VTAB ); + if( pCur->nullRow ){ + break; + } + pVtab = pCur->uc.pVCur->pVtab; + pModule = pVtab->pModule; + assert( pModule->xNext ); + + /* Invoke the xNext() method of the module. There is no way for the + ** underlying implementation to return an error if one occurs during + ** xNext(). Instead, if an error occurs, true is returned (indicating that + ** data is available) and the error code returned when xColumn or + ** some other method is next invoked on the save virtual table cursor. + */ + rc = pModule->xNext(pCur->uc.pVCur); + sqlite3VtabImportErrmsg(p, pVtab); + if( rc ) goto abort_due_to_error; + res = pModule->xEof(pCur->uc.pVCur); + VdbeBranchTaken(!res,2); + if( !res ){ + /* If there is data, jump to P2 */ + goto jump_to_p2_and_check_for_interrupt; + } + goto check_for_interrupt; +} +#endif /* SQLITE_OMIT_VIRTUALTABLE */ + +#ifndef SQLITE_OMIT_VIRTUALTABLE +/* Opcode: VRename P1 * * P4 * +** +** P4 is a pointer to a virtual table object, an sqlite3_vtab structure. +** This opcode invokes the corresponding xRename method. The value +** in register P1 is passed as the zName argument to the xRename method. +*/ +case OP_VRename: { + sqlite3_vtab *pVtab; + Mem *pName; + int isLegacy; + + isLegacy = (db->flags & SQLITE_LegacyAlter); + db->flags |= SQLITE_LegacyAlter; + pVtab = pOp->p4.pVtab->pVtab; + pName = &aMem[pOp->p1]; + assert( pVtab->pModule->xRename ); + assert( memIsValid(pName) ); + assert( p->readOnly==0 ); + REGISTER_TRACE(pOp->p1, pName); + assert( pName->flags & MEM_Str ); + testcase( pName->enc==SQLITE_UTF8 ); + testcase( pName->enc==SQLITE_UTF16BE ); + testcase( pName->enc==SQLITE_UTF16LE ); + rc = sqlite3VdbeChangeEncoding(pName, SQLITE_UTF8); + if( rc ) goto abort_due_to_error; + rc = pVtab->pModule->xRename(pVtab, pName->z); + if( isLegacy==0 ) db->flags &= ~(u64)SQLITE_LegacyAlter; + sqlite3VtabImportErrmsg(p, pVtab); + p->expired = 0; + if( rc ) goto abort_due_to_error; + break; +} +#endif + +#ifndef SQLITE_OMIT_VIRTUALTABLE +/* Opcode: VUpdate P1 P2 P3 P4 P5 +** Synopsis: data=r[P3@P2] +** +** P4 is a pointer to a virtual table object, an sqlite3_vtab structure. +** This opcode invokes the corresponding xUpdate method. P2 values +** are contiguous memory cells starting at P3 to pass to the xUpdate +** invocation. The value in register (P3+P2-1) corresponds to the +** p2th element of the argv array passed to xUpdate. +** +** The xUpdate method will do a DELETE or an INSERT or both. +** The argv[0] element (which corresponds to memory cell P3) +** is the rowid of a row to delete. If argv[0] is NULL then no +** deletion occurs. The argv[1] element is the rowid of the new +** row. This can be NULL to have the virtual table select the new +** rowid for itself. The subsequent elements in the array are +** the values of columns in the new row. +** +** If P2==1 then no insert is performed. argv[0] is the rowid of +** a row to delete. +** +** P1 is a boolean flag. If it is set to true and the xUpdate call +** is successful, then the value returned by sqlite3_last_insert_rowid() +** is set to the value of the rowid for the row just inserted. +** +** P5 is the error actions (OE_Replace, OE_Fail, OE_Ignore, etc) to +** apply in the case of a constraint failure on an insert or update. +*/ +case OP_VUpdate: { + sqlite3_vtab *pVtab; + const sqlite3_module *pModule; + int nArg; + int i; + sqlite_int64 rowid; + Mem **apArg; + Mem *pX; + + assert( pOp->p2==1 || pOp->p5==OE_Fail || pOp->p5==OE_Rollback + || pOp->p5==OE_Abort || pOp->p5==OE_Ignore || pOp->p5==OE_Replace + ); + assert( p->readOnly==0 ); + if( db->mallocFailed ) goto no_mem; + sqlite3VdbeIncrWriteCounter(p, 0); + pVtab = pOp->p4.pVtab->pVtab; + if( pVtab==0 || NEVER(pVtab->pModule==0) ){ + rc = SQLITE_LOCKED; + goto abort_due_to_error; + } + pModule = pVtab->pModule; + nArg = pOp->p2; + assert( pOp->p4type==P4_VTAB ); + if( ALWAYS(pModule->xUpdate) ){ + u8 vtabOnConflict = db->vtabOnConflict; + apArg = p->apArg; + pX = &aMem[pOp->p3]; + for(i=0; ivtabOnConflict = pOp->p5; + rc = pModule->xUpdate(pVtab, nArg, apArg, &rowid); + db->vtabOnConflict = vtabOnConflict; + sqlite3VtabImportErrmsg(p, pVtab); + if( rc==SQLITE_OK && pOp->p1 ){ + assert( nArg>1 && apArg[0] && (apArg[0]->flags&MEM_Null) ); + db->lastRowid = rowid; + } + if( (rc&0xff)==SQLITE_CONSTRAINT && pOp->p4.pVtab->bConstraint ){ + if( pOp->p5==OE_Ignore ){ + rc = SQLITE_OK; + }else{ + p->errorAction = ((pOp->p5==OE_Replace) ? OE_Abort : pOp->p5); + } + }else{ + p->nChange++; + } + if( rc ) goto abort_due_to_error; + } + break; +} +#endif /* SQLITE_OMIT_VIRTUALTABLE */ + +#ifndef SQLITE_OMIT_PAGER_PRAGMAS +/* Opcode: Pagecount P1 P2 * * * +** +** Write the current number of pages in database P1 to memory cell P2. +*/ +case OP_Pagecount: { /* out2 */ + pOut = out2Prerelease(p, pOp); + pOut->u.i = sqlite3BtreeLastPage(db->aDb[pOp->p1].pBt); + break; +} +#endif + + +#ifndef SQLITE_OMIT_PAGER_PRAGMAS +/* Opcode: MaxPgcnt P1 P2 P3 * * +** +** Try to set the maximum page count for database P1 to the value in P3. +** Do not let the maximum page count fall below the current page count and +** do not change the maximum page count value if P3==0. +** +** Store the maximum page count after the change in register P2. +*/ +case OP_MaxPgcnt: { /* out2 */ + unsigned int newMax; + Btree *pBt; + + pOut = out2Prerelease(p, pOp); + pBt = db->aDb[pOp->p1].pBt; + newMax = 0; + if( pOp->p3 ){ + newMax = sqlite3BtreeLastPage(pBt); + if( newMax < (unsigned)pOp->p3 ) newMax = (unsigned)pOp->p3; + } + pOut->u.i = sqlite3BtreeMaxPageCount(pBt, newMax); + break; +} +#endif + +/* Opcode: Function0 P1 P2 P3 P4 P5 +** Synopsis: r[P3]=func(r[P2@P5]) +** +** Invoke a user function (P4 is a pointer to a FuncDef object that +** defines the function) with P5 arguments taken from register P2 and +** successors. The result of the function is stored in register P3. +** Register P3 must not be one of the function inputs. +** +** P1 is a 32-bit bitmask indicating whether or not each argument to the +** function was determined to be constant at compile time. If the first +** argument was constant then bit 0 of P1 is set. This is used to determine +** whether meta data associated with a user function argument using the +** sqlite3_set_auxdata() API may be safely retained until the next +** invocation of this opcode. +** +** See also: Function, AggStep, AggFinal +*/ +/* Opcode: Function P1 P2 P3 P4 P5 +** Synopsis: r[P3]=func(r[P2@P5]) +** +** Invoke a user function (P4 is a pointer to an sqlite3_context object that +** contains a pointer to the function to be run) with P5 arguments taken +** from register P2 and successors. The result of the function is stored +** in register P3. Register P3 must not be one of the function inputs. +** +** P1 is a 32-bit bitmask indicating whether or not each argument to the +** function was determined to be constant at compile time. If the first +** argument was constant then bit 0 of P1 is set. This is used to determine +** whether meta data associated with a user function argument using the +** sqlite3_set_auxdata() API may be safely retained until the next +** invocation of this opcode. +** +** SQL functions are initially coded as OP_Function0 with P4 pointing +** to a FuncDef object. But on first evaluation, the P4 operand is +** automatically converted into an sqlite3_context object and the operation +** changed to this OP_Function opcode. In this way, the initialization of +** the sqlite3_context object occurs only once, rather than once for each +** evaluation of the function. +** +** See also: Function0, AggStep, AggFinal +*/ +case OP_PureFunc0: /* group */ +case OP_Function0: { /* group */ + int n; + sqlite3_context *pCtx; + + assert( pOp->p4type==P4_FUNCDEF ); + n = pOp->p5; + assert( pOp->p3>0 && pOp->p3<=(p->nMem+1 - p->nCursor) ); + assert( n==0 || (pOp->p2>0 && pOp->p2+n<=(p->nMem+1 - p->nCursor)+1) ); + assert( pOp->p3p2 || pOp->p3>=pOp->p2+n ); + pCtx = sqlite3DbMallocRawNN(db, sizeof(*pCtx) + (n-1)*sizeof(sqlite3_value*)); + if( pCtx==0 ) goto no_mem; + pCtx->pOut = 0; + pCtx->pFunc = pOp->p4.pFunc; + pCtx->iOp = (int)(pOp - aOp); + pCtx->pVdbe = p; + pCtx->isError = 0; + pCtx->argc = n; + pOp->p4type = P4_FUNCCTX; + pOp->p4.pCtx = pCtx; + assert( OP_PureFunc == OP_PureFunc0+2 ); + assert( OP_Function == OP_Function0+2 ); + pOp->opcode += 2; + /* Fall through into OP_Function */ +} +case OP_PureFunc: /* group */ +case OP_Function: { /* group */ + int i; + sqlite3_context *pCtx; + + assert( pOp->p4type==P4_FUNCCTX ); + pCtx = pOp->p4.pCtx; + + /* If this function is inside of a trigger, the register array in aMem[] + ** might change from one evaluation to the next. The next block of code + ** checks to see if the register array has changed, and if so it + ** reinitializes the relavant parts of the sqlite3_context object */ + pOut = &aMem[pOp->p3]; + if( pCtx->pOut != pOut ){ + pCtx->pOut = pOut; + for(i=pCtx->argc-1; i>=0; i--) pCtx->argv[i] = &aMem[pOp->p2+i]; + } + + memAboutToChange(p, pOut); +#ifdef SQLITE_DEBUG + for(i=0; iargc; i++){ + assert( memIsValid(pCtx->argv[i]) ); + REGISTER_TRACE(pOp->p2+i, pCtx->argv[i]); + } +#endif + MemSetTypeFlag(pOut, MEM_Null); + assert( pCtx->isError==0 ); + (*pCtx->pFunc->xSFunc)(pCtx, pCtx->argc, pCtx->argv);/* IMP: R-24505-23230 */ + + /* If the function returned an error, throw an exception */ + if( pCtx->isError ){ + if( pCtx->isError>0 ){ + sqlite3VdbeError(p, "%s", sqlite3_value_text(pOut)); + rc = pCtx->isError; + } + sqlite3VdbeDeleteAuxData(db, &p->pAuxData, pCtx->iOp, pOp->p1); + pCtx->isError = 0; + if( rc ) goto abort_due_to_error; + } + + /* Copy the result of the function into register P3 */ + if( pOut->flags & (MEM_Str|MEM_Blob) ){ + sqlite3VdbeChangeEncoding(pOut, encoding); + if( sqlite3VdbeMemTooBig(pOut) ) goto too_big; + } + + REGISTER_TRACE(pOp->p3, pOut); + UPDATE_MAX_BLOBSIZE(pOut); + break; +} + +/* Opcode: Trace P1 P2 * P4 * +** +** Write P4 on the statement trace output if statement tracing is +** enabled. +** +** Operand P1 must be 0x7fffffff and P2 must positive. +*/ +/* Opcode: Init P1 P2 P3 P4 * +** Synopsis: Start at P2 +** +** Programs contain a single instance of this opcode as the very first +** opcode. +** +** If tracing is enabled (by the sqlite3_trace()) interface, then +** the UTF-8 string contained in P4 is emitted on the trace callback. +** Or if P4 is blank, use the string returned by sqlite3_sql(). +** +** If P2 is not zero, jump to instruction P2. +** +** Increment the value of P1 so that OP_Once opcodes will jump the +** first time they are evaluated for this run. +** +** If P3 is not zero, then it is an address to jump to if an SQLITE_CORRUPT +** error is encountered. +*/ +case OP_Trace: +case OP_Init: { /* jump */ + int i; +#ifndef SQLITE_OMIT_TRACE + char *zTrace; +#endif + + /* If the P4 argument is not NULL, then it must be an SQL comment string. + ** The "--" string is broken up to prevent false-positives with srcck1.c. + ** + ** This assert() provides evidence for: + ** EVIDENCE-OF: R-50676-09860 The callback can compute the same text that + ** would have been returned by the legacy sqlite3_trace() interface by + ** using the X argument when X begins with "--" and invoking + ** sqlite3_expanded_sql(P) otherwise. + */ + assert( pOp->p4.z==0 || strncmp(pOp->p4.z, "-" "- ", 3)==0 ); + + /* OP_Init is always instruction 0 */ + assert( pOp==p->aOp || pOp->opcode==OP_Trace ); + +#ifndef SQLITE_OMIT_TRACE + if( (db->mTrace & (SQLITE_TRACE_STMT|SQLITE_TRACE_LEGACY))!=0 + && !p->doingRerun + && (zTrace = (pOp->p4.z ? pOp->p4.z : p->zSql))!=0 + ){ +#ifndef SQLITE_OMIT_DEPRECATED + if( db->mTrace & SQLITE_TRACE_LEGACY ){ + void (*x)(void*,const char*) = (void(*)(void*,const char*))db->xTrace; + char *z = sqlite3VdbeExpandSql(p, zTrace); + x(db->pTraceArg, z); + sqlite3_free(z); + }else +#endif + if( db->nVdbeExec>1 ){ + char *z = sqlite3MPrintf(db, "-- %s", zTrace); + (void)db->xTrace(SQLITE_TRACE_STMT, db->pTraceArg, p, z); + sqlite3DbFree(db, z); + }else{ + (void)db->xTrace(SQLITE_TRACE_STMT, db->pTraceArg, p, zTrace); + } + } +#ifdef SQLITE_USE_FCNTL_TRACE + zTrace = (pOp->p4.z ? pOp->p4.z : p->zSql); + if( zTrace ){ + int j; + for(j=0; jnDb; j++){ + if( DbMaskTest(p->btreeMask, j)==0 ) continue; + sqlite3_file_control(db, db->aDb[j].zDbSName, SQLITE_FCNTL_TRACE, zTrace); + } + } +#endif /* SQLITE_USE_FCNTL_TRACE */ +#ifdef SQLITE_DEBUG + if( (db->flags & SQLITE_SqlTrace)!=0 + && (zTrace = (pOp->p4.z ? pOp->p4.z : p->zSql))!=0 + ){ + sqlite3DebugPrintf("SQL-trace: %s\n", zTrace); + } +#endif /* SQLITE_DEBUG */ +#endif /* SQLITE_OMIT_TRACE */ + assert( pOp->p2>0 ); + if( pOp->p1>=sqlite3GlobalConfig.iOnceResetThreshold ){ + if( pOp->opcode==OP_Trace ) break; + for(i=1; inOp; i++){ + if( p->aOp[i].opcode==OP_Once ) p->aOp[i].p1 = 0; + } + pOp->p1 = 0; + } + pOp->p1++; + p->aCounter[SQLITE_STMTSTATUS_RUN]++; + goto jump_to_p2; +} + +#ifdef SQLITE_ENABLE_CURSOR_HINTS +/* Opcode: CursorHint P1 * * P4 * +** +** Provide a hint to cursor P1 that it only needs to return rows that +** satisfy the Expr in P4. TK_REGISTER terms in the P4 expression refer +** to values currently held in registers. TK_COLUMN terms in the P4 +** expression refer to columns in the b-tree to which cursor P1 is pointing. +*/ +case OP_CursorHint: { + VdbeCursor *pC; + + assert( pOp->p1>=0 && pOp->p1nCursor ); + assert( pOp->p4type==P4_EXPR ); + pC = p->apCsr[pOp->p1]; + if( pC ){ + assert( pC->eCurType==CURTYPE_BTREE ); + sqlite3BtreeCursorHint(pC->uc.pCursor, BTREE_HINT_RANGE, + pOp->p4.pExpr, aMem); + } + break; +} +#endif /* SQLITE_ENABLE_CURSOR_HINTS */ + +#ifdef SQLITE_DEBUG +/* Opcode: Abortable * * * * * +** +** Verify that an Abort can happen. Assert if an Abort at this point +** might cause database corruption. This opcode only appears in debugging +** builds. +** +** An Abort is safe if either there have been no writes, or if there is +** an active statement journal. +*/ +case OP_Abortable: { + sqlite3VdbeAssertAbortable(p); + break; +} +#endif + +/* Opcode: Noop * * * * * +** +** Do nothing. This instruction is often useful as a jump +** destination. +*/ +/* +** The magic Explain opcode are only inserted when explain==2 (which +** is to say when the EXPLAIN QUERY PLAN syntax is used.) +** This opcode records information from the optimizer. It is the +** the same as a no-op. This opcodesnever appears in a real VM program. +*/ +default: { /* This is really OP_Noop, OP_Explain */ + assert( pOp->opcode==OP_Noop || pOp->opcode==OP_Explain ); + + break; +} + +/***************************************************************************** +** The cases of the switch statement above this line should all be indented +** by 6 spaces. But the left-most 6 spaces have been removed to improve the +** readability. From this point on down, the normal indentation rules are +** restored. +*****************************************************************************/ + } + +#ifdef VDBE_PROFILE + { + u64 endTime = sqlite3NProfileCnt ? sqlite3NProfileCnt : sqlite3Hwtime(); + if( endTime>start ) pOrigOp->cycles += endTime - start; + pOrigOp->cnt++; + } +#endif + + /* The following code adds nothing to the actual functionality + ** of the program. It is only here for testing and debugging. + ** On the other hand, it does burn CPU cycles every time through + ** the evaluator loop. So we can leave it out when NDEBUG is defined. + */ +#ifndef NDEBUG + assert( pOp>=&aOp[-1] && pOp<&aOp[p->nOp-1] ); + +#ifdef SQLITE_DEBUG + if( db->flags & SQLITE_VdbeTrace ){ + u8 opProperty = sqlite3OpcodeProperty[pOrigOp->opcode]; + if( rc!=0 ) printf("rc=%d\n",rc); + if( opProperty & (OPFLG_OUT2) ){ + registerTrace(pOrigOp->p2, &aMem[pOrigOp->p2]); + } + if( opProperty & OPFLG_OUT3 ){ + registerTrace(pOrigOp->p3, &aMem[pOrigOp->p3]); + } + } +#endif /* SQLITE_DEBUG */ +#endif /* NDEBUG */ + } /* The end of the for(;;) loop the loops through opcodes */ + + /* If we reach this point, it means that execution is finished with + ** an error of some kind. + */ +abort_due_to_error: + if( db->mallocFailed ) rc = SQLITE_NOMEM_BKPT; + assert( rc ); + if( p->zErrMsg==0 && rc!=SQLITE_IOERR_NOMEM ){ + sqlite3VdbeError(p, "%s", sqlite3ErrStr(rc)); + } + p->rc = rc; + sqlite3SystemError(db, rc); + testcase( sqlite3GlobalConfig.xLog!=0 ); + sqlite3_log(rc, "statement aborts at %d: [%s] %s", + (int)(pOp - aOp), p->zSql, p->zErrMsg); + sqlite3VdbeHalt(p); + if( rc==SQLITE_IOERR_NOMEM ) sqlite3OomFault(db); + rc = SQLITE_ERROR; + if( resetSchemaOnFault>0 ){ + sqlite3ResetOneSchema(db, resetSchemaOnFault-1); + } + + /* This is the only way out of this procedure. We have to + ** release the mutexes on btrees that were acquired at the + ** top. */ +vdbe_return: +#ifndef SQLITE_OMIT_PROGRESS_CALLBACK + while( nVmStep>=nProgressLimit && db->xProgress!=0 ){ + nProgressLimit += db->nProgressOps; + if( db->xProgress(db->pProgressArg) ){ + nProgressLimit = 0xffffffff; + rc = SQLITE_INTERRUPT; + goto abort_due_to_error; + } + } +#endif + p->aCounter[SQLITE_STMTSTATUS_VM_STEP] += (int)nVmStep; + sqlite3VdbeLeave(p); + assert( rc!=SQLITE_OK || nExtraDelete==0 + || sqlite3_strlike("DELETE%",p->zSql,0)!=0 + ); + return rc; + + /* Jump to here if a string or blob larger than SQLITE_MAX_LENGTH + ** is encountered. + */ +too_big: + sqlite3VdbeError(p, "string or blob too big"); + rc = SQLITE_TOOBIG; + goto abort_due_to_error; + + /* Jump to here if a malloc() fails. + */ +no_mem: + sqlite3OomFault(db); + sqlite3VdbeError(p, "out of memory"); + rc = SQLITE_NOMEM_BKPT; + goto abort_due_to_error; + + /* Jump to here if the sqlite3_interrupt() API sets the interrupt + ** flag. + */ +abort_due_to_interrupt: + assert( db->u1.isInterrupted ); + rc = db->mallocFailed ? SQLITE_NOMEM_BKPT : SQLITE_INTERRUPT; + p->rc = rc; + sqlite3VdbeError(p, "%s", sqlite3ErrStr(rc)); + goto abort_due_to_error; +} + + +/************** End of vdbe.c ************************************************/ +/************** Begin file vdbeblob.c ****************************************/ +/* +** 2007 May 1 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** +** This file contains code used to implement incremental BLOB I/O. +*/ + +/* #include "sqliteInt.h" */ +/* #include "vdbeInt.h" */ + +#ifndef SQLITE_OMIT_INCRBLOB + +/* +** Valid sqlite3_blob* handles point to Incrblob structures. +*/ +typedef struct Incrblob Incrblob; +struct Incrblob { + int nByte; /* Size of open blob, in bytes */ + int iOffset; /* Byte offset of blob in cursor data */ + u16 iCol; /* Table column this handle is open on */ + BtCursor *pCsr; /* Cursor pointing at blob row */ + sqlite3_stmt *pStmt; /* Statement holding cursor open */ + sqlite3 *db; /* The associated database */ + char *zDb; /* Database name */ + Table *pTab; /* Table object */ +}; + + +/* +** This function is used by both blob_open() and blob_reopen(). It seeks +** the b-tree cursor associated with blob handle p to point to row iRow. +** If successful, SQLITE_OK is returned and subsequent calls to +** sqlite3_blob_read() or sqlite3_blob_write() access the specified row. +** +** If an error occurs, or if the specified row does not exist or does not +** contain a value of type TEXT or BLOB in the column nominated when the +** blob handle was opened, then an error code is returned and *pzErr may +** be set to point to a buffer containing an error message. It is the +** responsibility of the caller to free the error message buffer using +** sqlite3DbFree(). +** +** If an error does occur, then the b-tree cursor is closed. All subsequent +** calls to sqlite3_blob_read(), blob_write() or blob_reopen() will +** immediately return SQLITE_ABORT. +*/ +static int blobSeekToRow(Incrblob *p, sqlite3_int64 iRow, char **pzErr){ + int rc; /* Error code */ + char *zErr = 0; /* Error message */ + Vdbe *v = (Vdbe *)p->pStmt; + + /* Set the value of register r[1] in the SQL statement to integer iRow. + ** This is done directly as a performance optimization + */ + v->aMem[1].flags = MEM_Int; + v->aMem[1].u.i = iRow; + + /* If the statement has been run before (and is paused at the OP_ResultRow) + ** then back it up to the point where it does the OP_NotExists. This could + ** have been down with an extra OP_Goto, but simply setting the program + ** counter is faster. */ + if( v->pc>4 ){ + v->pc = 4; + assert( v->aOp[v->pc].opcode==OP_NotExists ); + rc = sqlite3VdbeExec(v); + }else{ + rc = sqlite3_step(p->pStmt); + } + if( rc==SQLITE_ROW ){ + VdbeCursor *pC = v->apCsr[0]; + u32 type = pC->nHdrParsed>p->iCol ? pC->aType[p->iCol] : 0; + testcase( pC->nHdrParsed==p->iCol ); + testcase( pC->nHdrParsed==p->iCol+1 ); + if( type<12 ){ + zErr = sqlite3MPrintf(p->db, "cannot open value of type %s", + type==0?"null": type==7?"real": "integer" + ); + rc = SQLITE_ERROR; + sqlite3_finalize(p->pStmt); + p->pStmt = 0; + }else{ + p->iOffset = pC->aType[p->iCol + pC->nField]; + p->nByte = sqlite3VdbeSerialTypeLen(type); + p->pCsr = pC->uc.pCursor; + sqlite3BtreeIncrblobCursor(p->pCsr); + } + } + + if( rc==SQLITE_ROW ){ + rc = SQLITE_OK; + }else if( p->pStmt ){ + rc = sqlite3_finalize(p->pStmt); + p->pStmt = 0; + if( rc==SQLITE_OK ){ + zErr = sqlite3MPrintf(p->db, "no such rowid: %lld", iRow); + rc = SQLITE_ERROR; + }else{ + zErr = sqlite3MPrintf(p->db, "%s", sqlite3_errmsg(p->db)); + } + } + + assert( rc!=SQLITE_OK || zErr==0 ); + assert( rc!=SQLITE_ROW && rc!=SQLITE_DONE ); + + *pzErr = zErr; + return rc; +} + +/* +** Open a blob handle. +*/ +SQLITE_API int sqlite3_blob_open( + sqlite3* db, /* The database connection */ + const char *zDb, /* The attached database containing the blob */ + const char *zTable, /* The table containing the blob */ + const char *zColumn, /* The column containing the blob */ + sqlite_int64 iRow, /* The row containing the glob */ + int wrFlag, /* True -> read/write access, false -> read-only */ + sqlite3_blob **ppBlob /* Handle for accessing the blob returned here */ +){ + int nAttempt = 0; + int iCol; /* Index of zColumn in row-record */ + int rc = SQLITE_OK; + char *zErr = 0; + Table *pTab; + Incrblob *pBlob = 0; + Parse sParse; + +#ifdef SQLITE_ENABLE_API_ARMOR + if( ppBlob==0 ){ + return SQLITE_MISUSE_BKPT; + } +#endif + *ppBlob = 0; +#ifdef SQLITE_ENABLE_API_ARMOR + if( !sqlite3SafetyCheckOk(db) || zTable==0 ){ + return SQLITE_MISUSE_BKPT; + } +#endif + wrFlag = !!wrFlag; /* wrFlag = (wrFlag ? 1 : 0); */ + + sqlite3_mutex_enter(db->mutex); + + pBlob = (Incrblob *)sqlite3DbMallocZero(db, sizeof(Incrblob)); + do { + memset(&sParse, 0, sizeof(Parse)); + if( !pBlob ) goto blob_open_out; + sParse.db = db; + sqlite3DbFree(db, zErr); + zErr = 0; + + sqlite3BtreeEnterAll(db); + pTab = sqlite3LocateTable(&sParse, 0, zTable, zDb); + if( pTab && IsVirtual(pTab) ){ + pTab = 0; + sqlite3ErrorMsg(&sParse, "cannot open virtual table: %s", zTable); + } + if( pTab && !HasRowid(pTab) ){ + pTab = 0; + sqlite3ErrorMsg(&sParse, "cannot open table without rowid: %s", zTable); + } +#ifndef SQLITE_OMIT_VIEW + if( pTab && pTab->pSelect ){ + pTab = 0; + sqlite3ErrorMsg(&sParse, "cannot open view: %s", zTable); + } +#endif + if( !pTab ){ + if( sParse.zErrMsg ){ + sqlite3DbFree(db, zErr); + zErr = sParse.zErrMsg; + sParse.zErrMsg = 0; + } + rc = SQLITE_ERROR; + sqlite3BtreeLeaveAll(db); + goto blob_open_out; + } + pBlob->pTab = pTab; + pBlob->zDb = db->aDb[sqlite3SchemaToIndex(db, pTab->pSchema)].zDbSName; + + /* Now search pTab for the exact column. */ + for(iCol=0; iColnCol; iCol++) { + if( sqlite3StrICmp(pTab->aCol[iCol].zName, zColumn)==0 ){ + break; + } + } + if( iCol==pTab->nCol ){ + sqlite3DbFree(db, zErr); + zErr = sqlite3MPrintf(db, "no such column: \"%s\"", zColumn); + rc = SQLITE_ERROR; + sqlite3BtreeLeaveAll(db); + goto blob_open_out; + } + + /* If the value is being opened for writing, check that the + ** column is not indexed, and that it is not part of a foreign key. + */ + if( wrFlag ){ + const char *zFault = 0; + Index *pIdx; +#ifndef SQLITE_OMIT_FOREIGN_KEY + if( db->flags&SQLITE_ForeignKeys ){ + /* Check that the column is not part of an FK child key definition. It + ** is not necessary to check if it is part of a parent key, as parent + ** key columns must be indexed. The check below will pick up this + ** case. */ + FKey *pFKey; + for(pFKey=pTab->pFKey; pFKey; pFKey=pFKey->pNextFrom){ + int j; + for(j=0; jnCol; j++){ + if( pFKey->aCol[j].iFrom==iCol ){ + zFault = "foreign key"; + } + } + } + } +#endif + for(pIdx=pTab->pIndex; pIdx; pIdx=pIdx->pNext){ + int j; + for(j=0; jnKeyCol; j++){ + /* FIXME: Be smarter about indexes that use expressions */ + if( pIdx->aiColumn[j]==iCol || pIdx->aiColumn[j]==XN_EXPR ){ + zFault = "indexed"; + } + } + } + if( zFault ){ + sqlite3DbFree(db, zErr); + zErr = sqlite3MPrintf(db, "cannot open %s column for writing", zFault); + rc = SQLITE_ERROR; + sqlite3BtreeLeaveAll(db); + goto blob_open_out; + } + } + + pBlob->pStmt = (sqlite3_stmt *)sqlite3VdbeCreate(&sParse); + assert( pBlob->pStmt || db->mallocFailed ); + if( pBlob->pStmt ){ + + /* This VDBE program seeks a btree cursor to the identified + ** db/table/row entry. The reason for using a vdbe program instead + ** of writing code to use the b-tree layer directly is that the + ** vdbe program will take advantage of the various transaction, + ** locking and error handling infrastructure built into the vdbe. + ** + ** After seeking the cursor, the vdbe executes an OP_ResultRow. + ** Code external to the Vdbe then "borrows" the b-tree cursor and + ** uses it to implement the blob_read(), blob_write() and + ** blob_bytes() functions. + ** + ** The sqlite3_blob_close() function finalizes the vdbe program, + ** which closes the b-tree cursor and (possibly) commits the + ** transaction. + */ + static const int iLn = VDBE_OFFSET_LINENO(2); + static const VdbeOpList openBlob[] = { + {OP_TableLock, 0, 0, 0}, /* 0: Acquire a read or write lock */ + {OP_OpenRead, 0, 0, 0}, /* 1: Open a cursor */ + /* blobSeekToRow() will initialize r[1] to the desired rowid */ + {OP_NotExists, 0, 5, 1}, /* 2: Seek the cursor to rowid=r[1] */ + {OP_Column, 0, 0, 1}, /* 3 */ + {OP_ResultRow, 1, 0, 0}, /* 4 */ + {OP_Halt, 0, 0, 0}, /* 5 */ + }; + Vdbe *v = (Vdbe *)pBlob->pStmt; + int iDb = sqlite3SchemaToIndex(db, pTab->pSchema); + VdbeOp *aOp; + + sqlite3VdbeAddOp4Int(v, OP_Transaction, iDb, wrFlag, + pTab->pSchema->schema_cookie, + pTab->pSchema->iGeneration); + sqlite3VdbeChangeP5(v, 1); + assert( sqlite3VdbeCurrentAddr(v)==2 || db->mallocFailed ); + aOp = sqlite3VdbeAddOpList(v, ArraySize(openBlob), openBlob, iLn); + + /* Make sure a mutex is held on the table to be accessed */ + sqlite3VdbeUsesBtree(v, iDb); + + if( db->mallocFailed==0 ){ + assert( aOp!=0 ); + /* Configure the OP_TableLock instruction */ +#ifdef SQLITE_OMIT_SHARED_CACHE + aOp[0].opcode = OP_Noop; +#else + aOp[0].p1 = iDb; + aOp[0].p2 = pTab->tnum; + aOp[0].p3 = wrFlag; + sqlite3VdbeChangeP4(v, 2, pTab->zName, P4_TRANSIENT); + } + if( db->mallocFailed==0 ){ +#endif + + /* Remove either the OP_OpenWrite or OpenRead. Set the P2 + ** parameter of the other to pTab->tnum. */ + if( wrFlag ) aOp[1].opcode = OP_OpenWrite; + aOp[1].p2 = pTab->tnum; + aOp[1].p3 = iDb; + + /* Configure the number of columns. Configure the cursor to + ** think that the table has one more column than it really + ** does. An OP_Column to retrieve this imaginary column will + ** always return an SQL NULL. This is useful because it means + ** we can invoke OP_Column to fill in the vdbe cursors type + ** and offset cache without causing any IO. + */ + aOp[1].p4type = P4_INT32; + aOp[1].p4.i = pTab->nCol+1; + aOp[3].p2 = pTab->nCol; + + sParse.nVar = 0; + sParse.nMem = 1; + sParse.nTab = 1; + sqlite3VdbeMakeReady(v, &sParse); + } + } + + pBlob->iCol = iCol; + pBlob->db = db; + sqlite3BtreeLeaveAll(db); + if( db->mallocFailed ){ + goto blob_open_out; + } + rc = blobSeekToRow(pBlob, iRow, &zErr); + } while( (++nAttempt)mallocFailed==0 ){ + *ppBlob = (sqlite3_blob *)pBlob; + }else{ + if( pBlob && pBlob->pStmt ) sqlite3VdbeFinalize((Vdbe *)pBlob->pStmt); + sqlite3DbFree(db, pBlob); + } + sqlite3ErrorWithMsg(db, rc, (zErr ? "%s" : 0), zErr); + sqlite3DbFree(db, zErr); + sqlite3ParserReset(&sParse); + rc = sqlite3ApiExit(db, rc); + sqlite3_mutex_leave(db->mutex); + return rc; +} + +/* +** Close a blob handle that was previously created using +** sqlite3_blob_open(). +*/ +SQLITE_API int sqlite3_blob_close(sqlite3_blob *pBlob){ + Incrblob *p = (Incrblob *)pBlob; + int rc; + sqlite3 *db; + + if( p ){ + sqlite3_stmt *pStmt = p->pStmt; + db = p->db; + sqlite3_mutex_enter(db->mutex); + sqlite3DbFree(db, p); + sqlite3_mutex_leave(db->mutex); + rc = sqlite3_finalize(pStmt); + }else{ + rc = SQLITE_OK; + } + return rc; +} + +/* +** Perform a read or write operation on a blob +*/ +static int blobReadWrite( + sqlite3_blob *pBlob, + void *z, + int n, + int iOffset, + int (*xCall)(BtCursor*, u32, u32, void*) +){ + int rc; + Incrblob *p = (Incrblob *)pBlob; + Vdbe *v; + sqlite3 *db; + + if( p==0 ) return SQLITE_MISUSE_BKPT; + db = p->db; + sqlite3_mutex_enter(db->mutex); + v = (Vdbe*)p->pStmt; + + if( n<0 || iOffset<0 || ((sqlite3_int64)iOffset+n)>p->nByte ){ + /* Request is out of range. Return a transient error. */ + rc = SQLITE_ERROR; + }else if( v==0 ){ + /* If there is no statement handle, then the blob-handle has + ** already been invalidated. Return SQLITE_ABORT in this case. + */ + rc = SQLITE_ABORT; + }else{ + /* Call either BtreeData() or BtreePutData(). If SQLITE_ABORT is + ** returned, clean-up the statement handle. + */ + assert( db == v->db ); + sqlite3BtreeEnterCursor(p->pCsr); + +#ifdef SQLITE_ENABLE_PREUPDATE_HOOK + if( xCall==sqlite3BtreePutData && db->xPreUpdateCallback ){ + /* If a pre-update hook is registered and this is a write cursor, + ** invoke it here. + ** + ** TODO: The preupdate-hook is passed SQLITE_DELETE, even though this + ** operation should really be an SQLITE_UPDATE. This is probably + ** incorrect, but is convenient because at this point the new.* values + ** are not easily obtainable. And for the sessions module, an + ** SQLITE_UPDATE where the PK columns do not change is handled in the + ** same way as an SQLITE_DELETE (the SQLITE_DELETE code is actually + ** slightly more efficient). Since you cannot write to a PK column + ** using the incremental-blob API, this works. For the sessions module + ** anyhow. + */ + sqlite3_int64 iKey; + iKey = sqlite3BtreeIntegerKey(p->pCsr); + sqlite3VdbePreUpdateHook( + v, v->apCsr[0], SQLITE_DELETE, p->zDb, p->pTab, iKey, -1 + ); + } +#endif + + rc = xCall(p->pCsr, iOffset+p->iOffset, n, z); + sqlite3BtreeLeaveCursor(p->pCsr); + if( rc==SQLITE_ABORT ){ + sqlite3VdbeFinalize(v); + p->pStmt = 0; + }else{ + v->rc = rc; + } + } + sqlite3Error(db, rc); + rc = sqlite3ApiExit(db, rc); + sqlite3_mutex_leave(db->mutex); + return rc; +} + +/* +** Read data from a blob handle. +*/ +SQLITE_API int sqlite3_blob_read(sqlite3_blob *pBlob, void *z, int n, int iOffset){ + return blobReadWrite(pBlob, z, n, iOffset, sqlite3BtreePayloadChecked); +} + +/* +** Write data to a blob handle. +*/ +SQLITE_API int sqlite3_blob_write(sqlite3_blob *pBlob, const void *z, int n, int iOffset){ + return blobReadWrite(pBlob, (void *)z, n, iOffset, sqlite3BtreePutData); +} + +/* +** Query a blob handle for the size of the data. +** +** The Incrblob.nByte field is fixed for the lifetime of the Incrblob +** so no mutex is required for access. +*/ +SQLITE_API int sqlite3_blob_bytes(sqlite3_blob *pBlob){ + Incrblob *p = (Incrblob *)pBlob; + return (p && p->pStmt) ? p->nByte : 0; +} + +/* +** Move an existing blob handle to point to a different row of the same +** database table. +** +** If an error occurs, or if the specified row does not exist or does not +** contain a blob or text value, then an error code is returned and the +** database handle error code and message set. If this happens, then all +** subsequent calls to sqlite3_blob_xxx() functions (except blob_close()) +** immediately return SQLITE_ABORT. +*/ +SQLITE_API int sqlite3_blob_reopen(sqlite3_blob *pBlob, sqlite3_int64 iRow){ + int rc; + Incrblob *p = (Incrblob *)pBlob; + sqlite3 *db; + + if( p==0 ) return SQLITE_MISUSE_BKPT; + db = p->db; + sqlite3_mutex_enter(db->mutex); + + if( p->pStmt==0 ){ + /* If there is no statement handle, then the blob-handle has + ** already been invalidated. Return SQLITE_ABORT in this case. + */ + rc = SQLITE_ABORT; + }else{ + char *zErr; + rc = blobSeekToRow(p, iRow, &zErr); + if( rc!=SQLITE_OK ){ + sqlite3ErrorWithMsg(db, rc, (zErr ? "%s" : 0), zErr); + sqlite3DbFree(db, zErr); + } + assert( rc!=SQLITE_SCHEMA ); + } + + rc = sqlite3ApiExit(db, rc); + assert( rc==SQLITE_OK || p->pStmt==0 ); + sqlite3_mutex_leave(db->mutex); + return rc; +} + +#endif /* #ifndef SQLITE_OMIT_INCRBLOB */ + +/************** End of vdbeblob.c ********************************************/ +/************** Begin file vdbesort.c ****************************************/ +/* +** 2011-07-09 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** This file contains code for the VdbeSorter object, used in concert with +** a VdbeCursor to sort large numbers of keys for CREATE INDEX statements +** or by SELECT statements with ORDER BY clauses that cannot be satisfied +** using indexes and without LIMIT clauses. +** +** The VdbeSorter object implements a multi-threaded external merge sort +** algorithm that is efficient even if the number of elements being sorted +** exceeds the available memory. +** +** Here is the (internal, non-API) interface between this module and the +** rest of the SQLite system: +** +** sqlite3VdbeSorterInit() Create a new VdbeSorter object. +** +** sqlite3VdbeSorterWrite() Add a single new row to the VdbeSorter +** object. The row is a binary blob in the +** OP_MakeRecord format that contains both +** the ORDER BY key columns and result columns +** in the case of a SELECT w/ ORDER BY, or +** the complete record for an index entry +** in the case of a CREATE INDEX. +** +** sqlite3VdbeSorterRewind() Sort all content previously added. +** Position the read cursor on the +** first sorted element. +** +** sqlite3VdbeSorterNext() Advance the read cursor to the next sorted +** element. +** +** sqlite3VdbeSorterRowkey() Return the complete binary blob for the +** row currently under the read cursor. +** +** sqlite3VdbeSorterCompare() Compare the binary blob for the row +** currently under the read cursor against +** another binary blob X and report if +** X is strictly less than the read cursor. +** Used to enforce uniqueness in a +** CREATE UNIQUE INDEX statement. +** +** sqlite3VdbeSorterClose() Close the VdbeSorter object and reclaim +** all resources. +** +** sqlite3VdbeSorterReset() Refurbish the VdbeSorter for reuse. This +** is like Close() followed by Init() only +** much faster. +** +** The interfaces above must be called in a particular order. Write() can +** only occur in between Init()/Reset() and Rewind(). Next(), Rowkey(), and +** Compare() can only occur in between Rewind() and Close()/Reset(). i.e. +** +** Init() +** for each record: Write() +** Rewind() +** Rowkey()/Compare() +** Next() +** Close() +** +** Algorithm: +** +** Records passed to the sorter via calls to Write() are initially held +** unsorted in main memory. Assuming the amount of memory used never exceeds +** a threshold, when Rewind() is called the set of records is sorted using +** an in-memory merge sort. In this case, no temporary files are required +** and subsequent calls to Rowkey(), Next() and Compare() read records +** directly from main memory. +** +** If the amount of space used to store records in main memory exceeds the +** threshold, then the set of records currently in memory are sorted and +** written to a temporary file in "Packed Memory Array" (PMA) format. +** A PMA created at this point is known as a "level-0 PMA". Higher levels +** of PMAs may be created by merging existing PMAs together - for example +** merging two or more level-0 PMAs together creates a level-1 PMA. +** +** The threshold for the amount of main memory to use before flushing +** records to a PMA is roughly the same as the limit configured for the +** page-cache of the main database. Specifically, the threshold is set to +** the value returned by "PRAGMA main.page_size" multipled by +** that returned by "PRAGMA main.cache_size", in bytes. +** +** If the sorter is running in single-threaded mode, then all PMAs generated +** are appended to a single temporary file. Or, if the sorter is running in +** multi-threaded mode then up to (N+1) temporary files may be opened, where +** N is the configured number of worker threads. In this case, instead of +** sorting the records and writing the PMA to a temporary file itself, the +** calling thread usually launches a worker thread to do so. Except, if +** there are already N worker threads running, the main thread does the work +** itself. +** +** The sorter is running in multi-threaded mode if (a) the library was built +** with pre-processor symbol SQLITE_MAX_WORKER_THREADS set to a value greater +** than zero, and (b) worker threads have been enabled at runtime by calling +** "PRAGMA threads=N" with some value of N greater than 0. +** +** When Rewind() is called, any data remaining in memory is flushed to a +** final PMA. So at this point the data is stored in some number of sorted +** PMAs within temporary files on disk. +** +** If there are fewer than SORTER_MAX_MERGE_COUNT PMAs in total and the +** sorter is running in single-threaded mode, then these PMAs are merged +** incrementally as keys are retreived from the sorter by the VDBE. The +** MergeEngine object, described in further detail below, performs this +** merge. +** +** Or, if running in multi-threaded mode, then a background thread is +** launched to merge the existing PMAs. Once the background thread has +** merged T bytes of data into a single sorted PMA, the main thread +** begins reading keys from that PMA while the background thread proceeds +** with merging the next T bytes of data. And so on. +** +** Parameter T is set to half the value of the memory threshold used +** by Write() above to determine when to create a new PMA. +** +** If there are more than SORTER_MAX_MERGE_COUNT PMAs in total when +** Rewind() is called, then a hierarchy of incremental-merges is used. +** First, T bytes of data from the first SORTER_MAX_MERGE_COUNT PMAs on +** disk are merged together. Then T bytes of data from the second set, and +** so on, such that no operation ever merges more than SORTER_MAX_MERGE_COUNT +** PMAs at a time. This done is to improve locality. +** +** If running in multi-threaded mode and there are more than +** SORTER_MAX_MERGE_COUNT PMAs on disk when Rewind() is called, then more +** than one background thread may be created. Specifically, there may be +** one background thread for each temporary file on disk, and one background +** thread to merge the output of each of the others to a single PMA for +** the main thread to read from. +*/ +/* #include "sqliteInt.h" */ +/* #include "vdbeInt.h" */ + +/* +** If SQLITE_DEBUG_SORTER_THREADS is defined, this module outputs various +** messages to stderr that may be helpful in understanding the performance +** characteristics of the sorter in multi-threaded mode. +*/ +#if 0 +# define SQLITE_DEBUG_SORTER_THREADS 1 +#endif + +/* +** Hard-coded maximum amount of data to accumulate in memory before flushing +** to a level 0 PMA. The purpose of this limit is to prevent various integer +** overflows. 512MiB. +*/ +#define SQLITE_MAX_PMASZ (1<<29) + +/* +** Private objects used by the sorter +*/ +typedef struct MergeEngine MergeEngine; /* Merge PMAs together */ +typedef struct PmaReader PmaReader; /* Incrementally read one PMA */ +typedef struct PmaWriter PmaWriter; /* Incrementally write one PMA */ +typedef struct SorterRecord SorterRecord; /* A record being sorted */ +typedef struct SortSubtask SortSubtask; /* A sub-task in the sort process */ +typedef struct SorterFile SorterFile; /* Temporary file object wrapper */ +typedef struct SorterList SorterList; /* In-memory list of records */ +typedef struct IncrMerger IncrMerger; /* Read & merge multiple PMAs */ + +/* +** A container for a temp file handle and the current amount of data +** stored in the file. +*/ +struct SorterFile { + sqlite3_file *pFd; /* File handle */ + i64 iEof; /* Bytes of data stored in pFd */ +}; + +/* +** An in-memory list of objects to be sorted. +** +** If aMemory==0 then each object is allocated separately and the objects +** are connected using SorterRecord.u.pNext. If aMemory!=0 then all objects +** are stored in the aMemory[] bulk memory, one right after the other, and +** are connected using SorterRecord.u.iNext. +*/ +struct SorterList { + SorterRecord *pList; /* Linked list of records */ + u8 *aMemory; /* If non-NULL, bulk memory to hold pList */ + int szPMA; /* Size of pList as PMA in bytes */ +}; + +/* +** The MergeEngine object is used to combine two or more smaller PMAs into +** one big PMA using a merge operation. Separate PMAs all need to be +** combined into one big PMA in order to be able to step through the sorted +** records in order. +** +** The aReadr[] array contains a PmaReader object for each of the PMAs being +** merged. An aReadr[] object either points to a valid key or else is at EOF. +** ("EOF" means "End Of File". When aReadr[] is at EOF there is no more data.) +** For the purposes of the paragraphs below, we assume that the array is +** actually N elements in size, where N is the smallest power of 2 greater +** to or equal to the number of PMAs being merged. The extra aReadr[] elements +** are treated as if they are empty (always at EOF). +** +** The aTree[] array is also N elements in size. The value of N is stored in +** the MergeEngine.nTree variable. +** +** The final (N/2) elements of aTree[] contain the results of comparing +** pairs of PMA keys together. Element i contains the result of +** comparing aReadr[2*i-N] and aReadr[2*i-N+1]. Whichever key is smaller, the +** aTree element is set to the index of it. +** +** For the purposes of this comparison, EOF is considered greater than any +** other key value. If the keys are equal (only possible with two EOF +** values), it doesn't matter which index is stored. +** +** The (N/4) elements of aTree[] that precede the final (N/2) described +** above contains the index of the smallest of each block of 4 PmaReaders +** And so on. So that aTree[1] contains the index of the PmaReader that +** currently points to the smallest key value. aTree[0] is unused. +** +** Example: +** +** aReadr[0] -> Banana +** aReadr[1] -> Feijoa +** aReadr[2] -> Elderberry +** aReadr[3] -> Currant +** aReadr[4] -> Grapefruit +** aReadr[5] -> Apple +** aReadr[6] -> Durian +** aReadr[7] -> EOF +** +** aTree[] = { X, 5 0, 5 0, 3, 5, 6 } +** +** The current element is "Apple" (the value of the key indicated by +** PmaReader 5). When the Next() operation is invoked, PmaReader 5 will +** be advanced to the next key in its segment. Say the next key is +** "Eggplant": +** +** aReadr[5] -> Eggplant +** +** The contents of aTree[] are updated first by comparing the new PmaReader +** 5 key to the current key of PmaReader 4 (still "Grapefruit"). The PmaReader +** 5 value is still smaller, so aTree[6] is set to 5. And so on up the tree. +** The value of PmaReader 6 - "Durian" - is now smaller than that of PmaReader +** 5, so aTree[3] is set to 6. Key 0 is smaller than key 6 (Bananafile2. And instead of using a +** background thread to prepare data for the PmaReader, with a single +** threaded IncrMerger the allocate part of pTask->file2 is "refilled" with +** keys from pMerger by the calling thread whenever the PmaReader runs out +** of data. +*/ +struct IncrMerger { + SortSubtask *pTask; /* Task that owns this merger */ + MergeEngine *pMerger; /* Merge engine thread reads data from */ + i64 iStartOff; /* Offset to start writing file at */ + int mxSz; /* Maximum bytes of data to store */ + int bEof; /* Set to true when merge is finished */ + int bUseThread; /* True to use a bg thread for this object */ + SorterFile aFile[2]; /* aFile[0] for reading, [1] for writing */ +}; + +/* +** An instance of this object is used for writing a PMA. +** +** The PMA is written one record at a time. Each record is of an arbitrary +** size. But I/O is more efficient if it occurs in page-sized blocks where +** each block is aligned on a page boundary. This object caches writes to +** the PMA so that aligned, page-size blocks are written. +*/ +struct PmaWriter { + int eFWErr; /* Non-zero if in an error state */ + u8 *aBuffer; /* Pointer to write buffer */ + int nBuffer; /* Size of write buffer in bytes */ + int iBufStart; /* First byte of buffer to write */ + int iBufEnd; /* Last byte of buffer to write */ + i64 iWriteOff; /* Offset of start of buffer in file */ + sqlite3_file *pFd; /* File handle to write to */ +}; + +/* +** This object is the header on a single record while that record is being +** held in memory and prior to being written out as part of a PMA. +** +** How the linked list is connected depends on how memory is being managed +** by this module. If using a separate allocation for each in-memory record +** (VdbeSorter.list.aMemory==0), then the list is always connected using the +** SorterRecord.u.pNext pointers. +** +** Or, if using the single large allocation method (VdbeSorter.list.aMemory!=0), +** then while records are being accumulated the list is linked using the +** SorterRecord.u.iNext offset. This is because the aMemory[] array may +** be sqlite3Realloc()ed while records are being accumulated. Once the VM +** has finished passing records to the sorter, or when the in-memory buffer +** is full, the list is sorted. As part of the sorting process, it is +** converted to use the SorterRecord.u.pNext pointers. See function +** vdbeSorterSort() for details. +*/ +struct SorterRecord { + int nVal; /* Size of the record in bytes */ + union { + SorterRecord *pNext; /* Pointer to next record in list */ + int iNext; /* Offset within aMemory of next record */ + } u; + /* The data for the record immediately follows this header */ +}; + +/* Return a pointer to the buffer containing the record data for SorterRecord +** object p. Should be used as if: +** +** void *SRVAL(SorterRecord *p) { return (void*)&p[1]; } +*/ +#define SRVAL(p) ((void*)((SorterRecord*)(p) + 1)) + + +/* Maximum number of PMAs that a single MergeEngine can merge */ +#define SORTER_MAX_MERGE_COUNT 16 + +static int vdbeIncrSwap(IncrMerger*); +static void vdbeIncrFree(IncrMerger *); + +/* +** Free all memory belonging to the PmaReader object passed as the +** argument. All structure fields are set to zero before returning. +*/ +static void vdbePmaReaderClear(PmaReader *pReadr){ + sqlite3_free(pReadr->aAlloc); + sqlite3_free(pReadr->aBuffer); + if( pReadr->aMap ) sqlite3OsUnfetch(pReadr->pFd, 0, pReadr->aMap); + vdbeIncrFree(pReadr->pIncr); + memset(pReadr, 0, sizeof(PmaReader)); +} + +/* +** Read the next nByte bytes of data from the PMA p. +** If successful, set *ppOut to point to a buffer containing the data +** and return SQLITE_OK. Otherwise, if an error occurs, return an SQLite +** error code. +** +** The buffer returned in *ppOut is only valid until the +** next call to this function. +*/ +static int vdbePmaReadBlob( + PmaReader *p, /* PmaReader from which to take the blob */ + int nByte, /* Bytes of data to read */ + u8 **ppOut /* OUT: Pointer to buffer containing data */ +){ + int iBuf; /* Offset within buffer to read from */ + int nAvail; /* Bytes of data available in buffer */ + + if( p->aMap ){ + *ppOut = &p->aMap[p->iReadOff]; + p->iReadOff += nByte; + return SQLITE_OK; + } + + assert( p->aBuffer ); + + /* If there is no more data to be read from the buffer, read the next + ** p->nBuffer bytes of data from the file into it. Or, if there are less + ** than p->nBuffer bytes remaining in the PMA, read all remaining data. */ + iBuf = p->iReadOff % p->nBuffer; + if( iBuf==0 ){ + int nRead; /* Bytes to read from disk */ + int rc; /* sqlite3OsRead() return code */ + + /* Determine how many bytes of data to read. */ + if( (p->iEof - p->iReadOff) > (i64)p->nBuffer ){ + nRead = p->nBuffer; + }else{ + nRead = (int)(p->iEof - p->iReadOff); + } + assert( nRead>0 ); + + /* Readr data from the file. Return early if an error occurs. */ + rc = sqlite3OsRead(p->pFd, p->aBuffer, nRead, p->iReadOff); + assert( rc!=SQLITE_IOERR_SHORT_READ ); + if( rc!=SQLITE_OK ) return rc; + } + nAvail = p->nBuffer - iBuf; + + if( nByte<=nAvail ){ + /* The requested data is available in the in-memory buffer. In this + ** case there is no need to make a copy of the data, just return a + ** pointer into the buffer to the caller. */ + *ppOut = &p->aBuffer[iBuf]; + p->iReadOff += nByte; + }else{ + /* The requested data is not all available in the in-memory buffer. + ** In this case, allocate space at p->aAlloc[] to copy the requested + ** range into. Then return a copy of pointer p->aAlloc to the caller. */ + int nRem; /* Bytes remaining to copy */ + + /* Extend the p->aAlloc[] allocation if required. */ + if( p->nAllocnAlloc); + while( nByte>nNew ) nNew = nNew*2; + aNew = sqlite3Realloc(p->aAlloc, nNew); + if( !aNew ) return SQLITE_NOMEM_BKPT; + p->nAlloc = nNew; + p->aAlloc = aNew; + } + + /* Copy as much data as is available in the buffer into the start of + ** p->aAlloc[]. */ + memcpy(p->aAlloc, &p->aBuffer[iBuf], nAvail); + p->iReadOff += nAvail; + nRem = nByte - nAvail; + + /* The following loop copies up to p->nBuffer bytes per iteration into + ** the p->aAlloc[] buffer. */ + while( nRem>0 ){ + int rc; /* vdbePmaReadBlob() return code */ + int nCopy; /* Number of bytes to copy */ + u8 *aNext; /* Pointer to buffer to copy data from */ + + nCopy = nRem; + if( nRem>p->nBuffer ) nCopy = p->nBuffer; + rc = vdbePmaReadBlob(p, nCopy, &aNext); + if( rc!=SQLITE_OK ) return rc; + assert( aNext!=p->aAlloc ); + memcpy(&p->aAlloc[nByte - nRem], aNext, nCopy); + nRem -= nCopy; + } + + *ppOut = p->aAlloc; + } + + return SQLITE_OK; +} + +/* +** Read a varint from the stream of data accessed by p. Set *pnOut to +** the value read. +*/ +static int vdbePmaReadVarint(PmaReader *p, u64 *pnOut){ + int iBuf; + + if( p->aMap ){ + p->iReadOff += sqlite3GetVarint(&p->aMap[p->iReadOff], pnOut); + }else{ + iBuf = p->iReadOff % p->nBuffer; + if( iBuf && (p->nBuffer-iBuf)>=9 ){ + p->iReadOff += sqlite3GetVarint(&p->aBuffer[iBuf], pnOut); + }else{ + u8 aVarint[16], *a; + int i = 0, rc; + do{ + rc = vdbePmaReadBlob(p, 1, &a); + if( rc ) return rc; + aVarint[(i++)&0xf] = a[0]; + }while( (a[0]&0x80)!=0 ); + sqlite3GetVarint(aVarint, pnOut); + } + } + + return SQLITE_OK; +} + +/* +** Attempt to memory map file pFile. If successful, set *pp to point to the +** new mapping and return SQLITE_OK. If the mapping is not attempted +** (because the file is too large or the VFS layer is configured not to use +** mmap), return SQLITE_OK and set *pp to NULL. +** +** Or, if an error occurs, return an SQLite error code. The final value of +** *pp is undefined in this case. +*/ +static int vdbeSorterMapFile(SortSubtask *pTask, SorterFile *pFile, u8 **pp){ + int rc = SQLITE_OK; + if( pFile->iEof<=(i64)(pTask->pSorter->db->nMaxSorterMmap) ){ + sqlite3_file *pFd = pFile->pFd; + if( pFd->pMethods->iVersion>=3 ){ + rc = sqlite3OsFetch(pFd, 0, (int)pFile->iEof, (void**)pp); + testcase( rc!=SQLITE_OK ); + } + } + return rc; +} + +/* +** Attach PmaReader pReadr to file pFile (if it is not already attached to +** that file) and seek it to offset iOff within the file. Return SQLITE_OK +** if successful, or an SQLite error code if an error occurs. +*/ +static int vdbePmaReaderSeek( + SortSubtask *pTask, /* Task context */ + PmaReader *pReadr, /* Reader whose cursor is to be moved */ + SorterFile *pFile, /* Sorter file to read from */ + i64 iOff /* Offset in pFile */ +){ + int rc = SQLITE_OK; + + assert( pReadr->pIncr==0 || pReadr->pIncr->bEof==0 ); + + if( sqlite3FaultSim(201) ) return SQLITE_IOERR_READ; + if( pReadr->aMap ){ + sqlite3OsUnfetch(pReadr->pFd, 0, pReadr->aMap); + pReadr->aMap = 0; + } + pReadr->iReadOff = iOff; + pReadr->iEof = pFile->iEof; + pReadr->pFd = pFile->pFd; + + rc = vdbeSorterMapFile(pTask, pFile, &pReadr->aMap); + if( rc==SQLITE_OK && pReadr->aMap==0 ){ + int pgsz = pTask->pSorter->pgsz; + int iBuf = pReadr->iReadOff % pgsz; + if( pReadr->aBuffer==0 ){ + pReadr->aBuffer = (u8*)sqlite3Malloc(pgsz); + if( pReadr->aBuffer==0 ) rc = SQLITE_NOMEM_BKPT; + pReadr->nBuffer = pgsz; + } + if( rc==SQLITE_OK && iBuf ){ + int nRead = pgsz - iBuf; + if( (pReadr->iReadOff + nRead) > pReadr->iEof ){ + nRead = (int)(pReadr->iEof - pReadr->iReadOff); + } + rc = sqlite3OsRead( + pReadr->pFd, &pReadr->aBuffer[iBuf], nRead, pReadr->iReadOff + ); + testcase( rc!=SQLITE_OK ); + } + } + + return rc; +} + +/* +** Advance PmaReader pReadr to the next key in its PMA. Return SQLITE_OK if +** no error occurs, or an SQLite error code if one does. +*/ +static int vdbePmaReaderNext(PmaReader *pReadr){ + int rc = SQLITE_OK; /* Return Code */ + u64 nRec = 0; /* Size of record in bytes */ + + + if( pReadr->iReadOff>=pReadr->iEof ){ + IncrMerger *pIncr = pReadr->pIncr; + int bEof = 1; + if( pIncr ){ + rc = vdbeIncrSwap(pIncr); + if( rc==SQLITE_OK && pIncr->bEof==0 ){ + rc = vdbePmaReaderSeek( + pIncr->pTask, pReadr, &pIncr->aFile[0], pIncr->iStartOff + ); + bEof = 0; + } + } + + if( bEof ){ + /* This is an EOF condition */ + vdbePmaReaderClear(pReadr); + testcase( rc!=SQLITE_OK ); + return rc; + } + } + + if( rc==SQLITE_OK ){ + rc = vdbePmaReadVarint(pReadr, &nRec); + } + if( rc==SQLITE_OK ){ + pReadr->nKey = (int)nRec; + rc = vdbePmaReadBlob(pReadr, (int)nRec, &pReadr->aKey); + testcase( rc!=SQLITE_OK ); + } + + return rc; +} + +/* +** Initialize PmaReader pReadr to scan through the PMA stored in file pFile +** starting at offset iStart and ending at offset iEof-1. This function +** leaves the PmaReader pointing to the first key in the PMA (or EOF if the +** PMA is empty). +** +** If the pnByte parameter is NULL, then it is assumed that the file +** contains a single PMA, and that that PMA omits the initial length varint. +*/ +static int vdbePmaReaderInit( + SortSubtask *pTask, /* Task context */ + SorterFile *pFile, /* Sorter file to read from */ + i64 iStart, /* Start offset in pFile */ + PmaReader *pReadr, /* PmaReader to populate */ + i64 *pnByte /* IN/OUT: Increment this value by PMA size */ +){ + int rc; + + assert( pFile->iEof>iStart ); + assert( pReadr->aAlloc==0 && pReadr->nAlloc==0 ); + assert( pReadr->aBuffer==0 ); + assert( pReadr->aMap==0 ); + + rc = vdbePmaReaderSeek(pTask, pReadr, pFile, iStart); + if( rc==SQLITE_OK ){ + u64 nByte = 0; /* Size of PMA in bytes */ + rc = vdbePmaReadVarint(pReadr, &nByte); + pReadr->iEof = pReadr->iReadOff + nByte; + *pnByte += nByte; + } + + if( rc==SQLITE_OK ){ + rc = vdbePmaReaderNext(pReadr); + } + return rc; +} + +/* +** A version of vdbeSorterCompare() that assumes that it has already been +** determined that the first field of key1 is equal to the first field of +** key2. +*/ +static int vdbeSorterCompareTail( + SortSubtask *pTask, /* Subtask context (for pKeyInfo) */ + int *pbKey2Cached, /* True if pTask->pUnpacked is pKey2 */ + const void *pKey1, int nKey1, /* Left side of comparison */ + const void *pKey2, int nKey2 /* Right side of comparison */ +){ + UnpackedRecord *r2 = pTask->pUnpacked; + if( *pbKey2Cached==0 ){ + sqlite3VdbeRecordUnpack(pTask->pSorter->pKeyInfo, nKey2, pKey2, r2); + *pbKey2Cached = 1; + } + return sqlite3VdbeRecordCompareWithSkip(nKey1, pKey1, r2, 1); +} + +/* +** Compare key1 (buffer pKey1, size nKey1 bytes) with key2 (buffer pKey2, +** size nKey2 bytes). Use (pTask->pKeyInfo) for the collation sequences +** used by the comparison. Return the result of the comparison. +** +** If IN/OUT parameter *pbKey2Cached is true when this function is called, +** it is assumed that (pTask->pUnpacked) contains the unpacked version +** of key2. If it is false, (pTask->pUnpacked) is populated with the unpacked +** version of key2 and *pbKey2Cached set to true before returning. +** +** If an OOM error is encountered, (pTask->pUnpacked->error_rc) is set +** to SQLITE_NOMEM. +*/ +static int vdbeSorterCompare( + SortSubtask *pTask, /* Subtask context (for pKeyInfo) */ + int *pbKey2Cached, /* True if pTask->pUnpacked is pKey2 */ + const void *pKey1, int nKey1, /* Left side of comparison */ + const void *pKey2, int nKey2 /* Right side of comparison */ +){ + UnpackedRecord *r2 = pTask->pUnpacked; + if( !*pbKey2Cached ){ + sqlite3VdbeRecordUnpack(pTask->pSorter->pKeyInfo, nKey2, pKey2, r2); + *pbKey2Cached = 1; + } + return sqlite3VdbeRecordCompare(nKey1, pKey1, r2); +} + +/* +** A specially optimized version of vdbeSorterCompare() that assumes that +** the first field of each key is a TEXT value and that the collation +** sequence to compare them with is BINARY. +*/ +static int vdbeSorterCompareText( + SortSubtask *pTask, /* Subtask context (for pKeyInfo) */ + int *pbKey2Cached, /* True if pTask->pUnpacked is pKey2 */ + const void *pKey1, int nKey1, /* Left side of comparison */ + const void *pKey2, int nKey2 /* Right side of comparison */ +){ + const u8 * const p1 = (const u8 * const)pKey1; + const u8 * const p2 = (const u8 * const)pKey2; + const u8 * const v1 = &p1[ p1[0] ]; /* Pointer to value 1 */ + const u8 * const v2 = &p2[ p2[0] ]; /* Pointer to value 2 */ + + int n1; + int n2; + int res; + + getVarint32(&p1[1], n1); + getVarint32(&p2[1], n2); + res = memcmp(v1, v2, (MIN(n1, n2) - 13)/2); + if( res==0 ){ + res = n1 - n2; + } + + if( res==0 ){ + if( pTask->pSorter->pKeyInfo->nKeyField>1 ){ + res = vdbeSorterCompareTail( + pTask, pbKey2Cached, pKey1, nKey1, pKey2, nKey2 + ); + } + }else{ + assert( !(pTask->pSorter->pKeyInfo->aSortFlags[0]&KEYINFO_ORDER_BIGNULL) ); + if( pTask->pSorter->pKeyInfo->aSortFlags[0] ){ + res = res * -1; + } + } + + return res; +} + +/* +** A specially optimized version of vdbeSorterCompare() that assumes that +** the first field of each key is an INTEGER value. +*/ +static int vdbeSorterCompareInt( + SortSubtask *pTask, /* Subtask context (for pKeyInfo) */ + int *pbKey2Cached, /* True if pTask->pUnpacked is pKey2 */ + const void *pKey1, int nKey1, /* Left side of comparison */ + const void *pKey2, int nKey2 /* Right side of comparison */ +){ + const u8 * const p1 = (const u8 * const)pKey1; + const u8 * const p2 = (const u8 * const)pKey2; + const int s1 = p1[1]; /* Left hand serial type */ + const int s2 = p2[1]; /* Right hand serial type */ + const u8 * const v1 = &p1[ p1[0] ]; /* Pointer to value 1 */ + const u8 * const v2 = &p2[ p2[0] ]; /* Pointer to value 2 */ + int res; /* Return value */ + + assert( (s1>0 && s1<7) || s1==8 || s1==9 ); + assert( (s2>0 && s2<7) || s2==8 || s2==9 ); + + if( s1==s2 ){ + /* The two values have the same sign. Compare using memcmp(). */ + static const u8 aLen[] = {0, 1, 2, 3, 4, 6, 8, 0, 0, 0 }; + const u8 n = aLen[s1]; + int i; + res = 0; + for(i=0; i7 && s2>7 ){ + res = s1 - s2; + }else{ + if( s2>7 ){ + res = +1; + }else if( s1>7 ){ + res = -1; + }else{ + res = s1 - s2; + } + assert( res!=0 ); + + if( res>0 ){ + if( *v1 & 0x80 ) res = -1; + }else{ + if( *v2 & 0x80 ) res = +1; + } + } + + if( res==0 ){ + if( pTask->pSorter->pKeyInfo->nKeyField>1 ){ + res = vdbeSorterCompareTail( + pTask, pbKey2Cached, pKey1, nKey1, pKey2, nKey2 + ); + } + }else if( pTask->pSorter->pKeyInfo->aSortFlags[0] ){ + assert( !(pTask->pSorter->pKeyInfo->aSortFlags[0]&KEYINFO_ORDER_BIGNULL) ); + res = res * -1; + } + + return res; +} + +/* +** Initialize the temporary index cursor just opened as a sorter cursor. +** +** Usually, the sorter module uses the value of (pCsr->pKeyInfo->nKeyField) +** to determine the number of fields that should be compared from the +** records being sorted. However, if the value passed as argument nField +** is non-zero and the sorter is able to guarantee a stable sort, nField +** is used instead. This is used when sorting records for a CREATE INDEX +** statement. In this case, keys are always delivered to the sorter in +** order of the primary key, which happens to be make up the final part +** of the records being sorted. So if the sort is stable, there is never +** any reason to compare PK fields and they can be ignored for a small +** performance boost. +** +** The sorter can guarantee a stable sort when running in single-threaded +** mode, but not in multi-threaded mode. +** +** SQLITE_OK is returned if successful, or an SQLite error code otherwise. +*/ +SQLITE_PRIVATE int sqlite3VdbeSorterInit( + sqlite3 *db, /* Database connection (for malloc()) */ + int nField, /* Number of key fields in each record */ + VdbeCursor *pCsr /* Cursor that holds the new sorter */ +){ + int pgsz; /* Page size of main database */ + int i; /* Used to iterate through aTask[] */ + VdbeSorter *pSorter; /* The new sorter */ + KeyInfo *pKeyInfo; /* Copy of pCsr->pKeyInfo with db==0 */ + int szKeyInfo; /* Size of pCsr->pKeyInfo in bytes */ + int sz; /* Size of pSorter in bytes */ + int rc = SQLITE_OK; +#if SQLITE_MAX_WORKER_THREADS==0 +# define nWorker 0 +#else + int nWorker; +#endif + + /* Initialize the upper limit on the number of worker threads */ +#if SQLITE_MAX_WORKER_THREADS>0 + if( sqlite3TempInMemory(db) || sqlite3GlobalConfig.bCoreMutex==0 ){ + nWorker = 0; + }else{ + nWorker = db->aLimit[SQLITE_LIMIT_WORKER_THREADS]; + } +#endif + + /* Do not allow the total number of threads (main thread + all workers) + ** to exceed the maximum merge count */ +#if SQLITE_MAX_WORKER_THREADS>=SORTER_MAX_MERGE_COUNT + if( nWorker>=SORTER_MAX_MERGE_COUNT ){ + nWorker = SORTER_MAX_MERGE_COUNT-1; + } +#endif + + assert( pCsr->pKeyInfo && pCsr->pBtx==0 ); + assert( pCsr->eCurType==CURTYPE_SORTER ); + szKeyInfo = sizeof(KeyInfo) + (pCsr->pKeyInfo->nKeyField-1)*sizeof(CollSeq*); + sz = sizeof(VdbeSorter) + nWorker * sizeof(SortSubtask); + + pSorter = (VdbeSorter*)sqlite3DbMallocZero(db, sz + szKeyInfo); + pCsr->uc.pSorter = pSorter; + if( pSorter==0 ){ + rc = SQLITE_NOMEM_BKPT; + }else{ + pSorter->pKeyInfo = pKeyInfo = (KeyInfo*)((u8*)pSorter + sz); + memcpy(pKeyInfo, pCsr->pKeyInfo, szKeyInfo); + pKeyInfo->db = 0; + if( nField && nWorker==0 ){ + pKeyInfo->nKeyField = nField; + } + pSorter->pgsz = pgsz = sqlite3BtreeGetPageSize(db->aDb[0].pBt); + pSorter->nTask = nWorker + 1; + pSorter->iPrev = (u8)(nWorker - 1); + pSorter->bUseThreads = (pSorter->nTask>1); + pSorter->db = db; + for(i=0; inTask; i++){ + SortSubtask *pTask = &pSorter->aTask[i]; + pTask->pSorter = pSorter; + } + + if( !sqlite3TempInMemory(db) ){ + i64 mxCache; /* Cache size in bytes*/ + u32 szPma = sqlite3GlobalConfig.szPma; + pSorter->mnPmaSize = szPma * pgsz; + + mxCache = db->aDb[0].pSchema->cache_size; + if( mxCache<0 ){ + /* A negative cache-size value C indicates that the cache is abs(C) + ** KiB in size. */ + mxCache = mxCache * -1024; + }else{ + mxCache = mxCache * pgsz; + } + mxCache = MIN(mxCache, SQLITE_MAX_PMASZ); + pSorter->mxPmaSize = MAX(pSorter->mnPmaSize, (int)mxCache); + + /* Avoid large memory allocations if the application has requested + ** SQLITE_CONFIG_SMALL_MALLOC. */ + if( sqlite3GlobalConfig.bSmallMalloc==0 ){ + assert( pSorter->iMemory==0 ); + pSorter->nMemory = pgsz; + pSorter->list.aMemory = (u8*)sqlite3Malloc(pgsz); + if( !pSorter->list.aMemory ) rc = SQLITE_NOMEM_BKPT; + } + } + + if( pKeyInfo->nAllField<13 + && (pKeyInfo->aColl[0]==0 || pKeyInfo->aColl[0]==db->pDfltColl) + && (pKeyInfo->aSortFlags[0] & KEYINFO_ORDER_BIGNULL)==0 + ){ + pSorter->typeMask = SORTER_TYPE_INTEGER | SORTER_TYPE_TEXT; + } + } + + return rc; +} +#undef nWorker /* Defined at the top of this function */ + +/* +** Free the list of sorted records starting at pRecord. +*/ +static void vdbeSorterRecordFree(sqlite3 *db, SorterRecord *pRecord){ + SorterRecord *p; + SorterRecord *pNext; + for(p=pRecord; p; p=pNext){ + pNext = p->u.pNext; + sqlite3DbFree(db, p); + } +} + +/* +** Free all resources owned by the object indicated by argument pTask. All +** fields of *pTask are zeroed before returning. +*/ +static void vdbeSortSubtaskCleanup(sqlite3 *db, SortSubtask *pTask){ + sqlite3DbFree(db, pTask->pUnpacked); +#if SQLITE_MAX_WORKER_THREADS>0 + /* pTask->list.aMemory can only be non-zero if it was handed memory + ** from the main thread. That only occurs SQLITE_MAX_WORKER_THREADS>0 */ + if( pTask->list.aMemory ){ + sqlite3_free(pTask->list.aMemory); + }else +#endif + { + assert( pTask->list.aMemory==0 ); + vdbeSorterRecordFree(0, pTask->list.pList); + } + if( pTask->file.pFd ){ + sqlite3OsCloseFree(pTask->file.pFd); + } + if( pTask->file2.pFd ){ + sqlite3OsCloseFree(pTask->file2.pFd); + } + memset(pTask, 0, sizeof(SortSubtask)); +} + +#ifdef SQLITE_DEBUG_SORTER_THREADS +static void vdbeSorterWorkDebug(SortSubtask *pTask, const char *zEvent){ + i64 t; + int iTask = (pTask - pTask->pSorter->aTask); + sqlite3OsCurrentTimeInt64(pTask->pSorter->db->pVfs, &t); + fprintf(stderr, "%lld:%d %s\n", t, iTask, zEvent); +} +static void vdbeSorterRewindDebug(const char *zEvent){ + i64 t; + sqlite3OsCurrentTimeInt64(sqlite3_vfs_find(0), &t); + fprintf(stderr, "%lld:X %s\n", t, zEvent); +} +static void vdbeSorterPopulateDebug( + SortSubtask *pTask, + const char *zEvent +){ + i64 t; + int iTask = (pTask - pTask->pSorter->aTask); + sqlite3OsCurrentTimeInt64(pTask->pSorter->db->pVfs, &t); + fprintf(stderr, "%lld:bg%d %s\n", t, iTask, zEvent); +} +static void vdbeSorterBlockDebug( + SortSubtask *pTask, + int bBlocked, + const char *zEvent +){ + if( bBlocked ){ + i64 t; + sqlite3OsCurrentTimeInt64(pTask->pSorter->db->pVfs, &t); + fprintf(stderr, "%lld:main %s\n", t, zEvent); + } +} +#else +# define vdbeSorterWorkDebug(x,y) +# define vdbeSorterRewindDebug(y) +# define vdbeSorterPopulateDebug(x,y) +# define vdbeSorterBlockDebug(x,y,z) +#endif + +#if SQLITE_MAX_WORKER_THREADS>0 +/* +** Join thread pTask->thread. +*/ +static int vdbeSorterJoinThread(SortSubtask *pTask){ + int rc = SQLITE_OK; + if( pTask->pThread ){ +#ifdef SQLITE_DEBUG_SORTER_THREADS + int bDone = pTask->bDone; +#endif + void *pRet = SQLITE_INT_TO_PTR(SQLITE_ERROR); + vdbeSorterBlockDebug(pTask, !bDone, "enter"); + (void)sqlite3ThreadJoin(pTask->pThread, &pRet); + vdbeSorterBlockDebug(pTask, !bDone, "exit"); + rc = SQLITE_PTR_TO_INT(pRet); + assert( pTask->bDone==1 ); + pTask->bDone = 0; + pTask->pThread = 0; + } + return rc; +} + +/* +** Launch a background thread to run xTask(pIn). +*/ +static int vdbeSorterCreateThread( + SortSubtask *pTask, /* Thread will use this task object */ + void *(*xTask)(void*), /* Routine to run in a separate thread */ + void *pIn /* Argument passed into xTask() */ +){ + assert( pTask->pThread==0 && pTask->bDone==0 ); + return sqlite3ThreadCreate(&pTask->pThread, xTask, pIn); +} + +/* +** Join all outstanding threads launched by SorterWrite() to create +** level-0 PMAs. +*/ +static int vdbeSorterJoinAll(VdbeSorter *pSorter, int rcin){ + int rc = rcin; + int i; + + /* This function is always called by the main user thread. + ** + ** If this function is being called after SorterRewind() has been called, + ** it is possible that thread pSorter->aTask[pSorter->nTask-1].pThread + ** is currently attempt to join one of the other threads. To avoid a race + ** condition where this thread also attempts to join the same object, join + ** thread pSorter->aTask[pSorter->nTask-1].pThread first. */ + for(i=pSorter->nTask-1; i>=0; i--){ + SortSubtask *pTask = &pSorter->aTask[i]; + int rc2 = vdbeSorterJoinThread(pTask); + if( rc==SQLITE_OK ) rc = rc2; + } + return rc; +} +#else +# define vdbeSorterJoinAll(x,rcin) (rcin) +# define vdbeSorterJoinThread(pTask) SQLITE_OK +#endif + +/* +** Allocate a new MergeEngine object capable of handling up to +** nReader PmaReader inputs. +** +** nReader is automatically rounded up to the next power of two. +** nReader may not exceed SORTER_MAX_MERGE_COUNT even after rounding up. +*/ +static MergeEngine *vdbeMergeEngineNew(int nReader){ + int N = 2; /* Smallest power of two >= nReader */ + int nByte; /* Total bytes of space to allocate */ + MergeEngine *pNew; /* Pointer to allocated object to return */ + + assert( nReader<=SORTER_MAX_MERGE_COUNT ); + + while( NnTree = N; + pNew->pTask = 0; + pNew->aReadr = (PmaReader*)&pNew[1]; + pNew->aTree = (int*)&pNew->aReadr[N]; + } + return pNew; +} + +/* +** Free the MergeEngine object passed as the only argument. +*/ +static void vdbeMergeEngineFree(MergeEngine *pMerger){ + int i; + if( pMerger ){ + for(i=0; inTree; i++){ + vdbePmaReaderClear(&pMerger->aReadr[i]); + } + } + sqlite3_free(pMerger); +} + +/* +** Free all resources associated with the IncrMerger object indicated by +** the first argument. +*/ +static void vdbeIncrFree(IncrMerger *pIncr){ + if( pIncr ){ +#if SQLITE_MAX_WORKER_THREADS>0 + if( pIncr->bUseThread ){ + vdbeSorterJoinThread(pIncr->pTask); + if( pIncr->aFile[0].pFd ) sqlite3OsCloseFree(pIncr->aFile[0].pFd); + if( pIncr->aFile[1].pFd ) sqlite3OsCloseFree(pIncr->aFile[1].pFd); + } +#endif + vdbeMergeEngineFree(pIncr->pMerger); + sqlite3_free(pIncr); + } +} + +/* +** Reset a sorting cursor back to its original empty state. +*/ +SQLITE_PRIVATE void sqlite3VdbeSorterReset(sqlite3 *db, VdbeSorter *pSorter){ + int i; + (void)vdbeSorterJoinAll(pSorter, SQLITE_OK); + assert( pSorter->bUseThreads || pSorter->pReader==0 ); +#if SQLITE_MAX_WORKER_THREADS>0 + if( pSorter->pReader ){ + vdbePmaReaderClear(pSorter->pReader); + sqlite3DbFree(db, pSorter->pReader); + pSorter->pReader = 0; + } +#endif + vdbeMergeEngineFree(pSorter->pMerger); + pSorter->pMerger = 0; + for(i=0; inTask; i++){ + SortSubtask *pTask = &pSorter->aTask[i]; + vdbeSortSubtaskCleanup(db, pTask); + pTask->pSorter = pSorter; + } + if( pSorter->list.aMemory==0 ){ + vdbeSorterRecordFree(0, pSorter->list.pList); + } + pSorter->list.pList = 0; + pSorter->list.szPMA = 0; + pSorter->bUsePMA = 0; + pSorter->iMemory = 0; + pSorter->mxKeysize = 0; + sqlite3DbFree(db, pSorter->pUnpacked); + pSorter->pUnpacked = 0; +} + +/* +** Free any cursor components allocated by sqlite3VdbeSorterXXX routines. +*/ +SQLITE_PRIVATE void sqlite3VdbeSorterClose(sqlite3 *db, VdbeCursor *pCsr){ + VdbeSorter *pSorter; + assert( pCsr->eCurType==CURTYPE_SORTER ); + pSorter = pCsr->uc.pSorter; + if( pSorter ){ + sqlite3VdbeSorterReset(db, pSorter); + sqlite3_free(pSorter->list.aMemory); + sqlite3DbFree(db, pSorter); + pCsr->uc.pSorter = 0; + } +} + +#if SQLITE_MAX_MMAP_SIZE>0 +/* +** The first argument is a file-handle open on a temporary file. The file +** is guaranteed to be nByte bytes or smaller in size. This function +** attempts to extend the file to nByte bytes in size and to ensure that +** the VFS has memory mapped it. +** +** Whether or not the file does end up memory mapped of course depends on +** the specific VFS implementation. +*/ +static void vdbeSorterExtendFile(sqlite3 *db, sqlite3_file *pFd, i64 nByte){ + if( nByte<=(i64)(db->nMaxSorterMmap) && pFd->pMethods->iVersion>=3 ){ + void *p = 0; + int chunksize = 4*1024; + sqlite3OsFileControlHint(pFd, SQLITE_FCNTL_CHUNK_SIZE, &chunksize); + sqlite3OsFileControlHint(pFd, SQLITE_FCNTL_SIZE_HINT, &nByte); + sqlite3OsFetch(pFd, 0, (int)nByte, &p); + sqlite3OsUnfetch(pFd, 0, p); + } +} +#else +# define vdbeSorterExtendFile(x,y,z) +#endif + +/* +** Allocate space for a file-handle and open a temporary file. If successful, +** set *ppFd to point to the malloc'd file-handle and return SQLITE_OK. +** Otherwise, set *ppFd to 0 and return an SQLite error code. +*/ +static int vdbeSorterOpenTempFile( + sqlite3 *db, /* Database handle doing sort */ + i64 nExtend, /* Attempt to extend file to this size */ + sqlite3_file **ppFd +){ + int rc; + if( sqlite3FaultSim(202) ) return SQLITE_IOERR_ACCESS; + rc = sqlite3OsOpenMalloc(db->pVfs, 0, ppFd, + SQLITE_OPEN_TEMP_JOURNAL | + SQLITE_OPEN_READWRITE | SQLITE_OPEN_CREATE | + SQLITE_OPEN_EXCLUSIVE | SQLITE_OPEN_DELETEONCLOSE, &rc + ); + if( rc==SQLITE_OK ){ + i64 max = SQLITE_MAX_MMAP_SIZE; + sqlite3OsFileControlHint(*ppFd, SQLITE_FCNTL_MMAP_SIZE, (void*)&max); + if( nExtend>0 ){ + vdbeSorterExtendFile(db, *ppFd, nExtend); + } + } + return rc; +} + +/* +** If it has not already been allocated, allocate the UnpackedRecord +** structure at pTask->pUnpacked. Return SQLITE_OK if successful (or +** if no allocation was required), or SQLITE_NOMEM otherwise. +*/ +static int vdbeSortAllocUnpacked(SortSubtask *pTask){ + if( pTask->pUnpacked==0 ){ + pTask->pUnpacked = sqlite3VdbeAllocUnpackedRecord(pTask->pSorter->pKeyInfo); + if( pTask->pUnpacked==0 ) return SQLITE_NOMEM_BKPT; + pTask->pUnpacked->nField = pTask->pSorter->pKeyInfo->nKeyField; + pTask->pUnpacked->errCode = 0; + } + return SQLITE_OK; +} + + +/* +** Merge the two sorted lists p1 and p2 into a single list. +*/ +static SorterRecord *vdbeSorterMerge( + SortSubtask *pTask, /* Calling thread context */ + SorterRecord *p1, /* First list to merge */ + SorterRecord *p2 /* Second list to merge */ +){ + SorterRecord *pFinal = 0; + SorterRecord **pp = &pFinal; + int bCached = 0; + + assert( p1!=0 && p2!=0 ); + for(;;){ + int res; + res = pTask->xCompare( + pTask, &bCached, SRVAL(p1), p1->nVal, SRVAL(p2), p2->nVal + ); + + if( res<=0 ){ + *pp = p1; + pp = &p1->u.pNext; + p1 = p1->u.pNext; + if( p1==0 ){ + *pp = p2; + break; + } + }else{ + *pp = p2; + pp = &p2->u.pNext; + p2 = p2->u.pNext; + bCached = 0; + if( p2==0 ){ + *pp = p1; + break; + } + } + } + return pFinal; +} + +/* +** Return the SorterCompare function to compare values collected by the +** sorter object passed as the only argument. +*/ +static SorterCompare vdbeSorterGetCompare(VdbeSorter *p){ + if( p->typeMask==SORTER_TYPE_INTEGER ){ + return vdbeSorterCompareInt; + }else if( p->typeMask==SORTER_TYPE_TEXT ){ + return vdbeSorterCompareText; + } + return vdbeSorterCompare; +} + +/* +** Sort the linked list of records headed at pTask->pList. Return +** SQLITE_OK if successful, or an SQLite error code (i.e. SQLITE_NOMEM) if +** an error occurs. +*/ +static int vdbeSorterSort(SortSubtask *pTask, SorterList *pList){ + int i; + SorterRecord **aSlot; + SorterRecord *p; + int rc; + + rc = vdbeSortAllocUnpacked(pTask); + if( rc!=SQLITE_OK ) return rc; + + p = pList->pList; + pTask->xCompare = vdbeSorterGetCompare(pTask->pSorter); + + aSlot = (SorterRecord **)sqlite3MallocZero(64 * sizeof(SorterRecord *)); + if( !aSlot ){ + return SQLITE_NOMEM_BKPT; + } + + while( p ){ + SorterRecord *pNext; + if( pList->aMemory ){ + if( (u8*)p==pList->aMemory ){ + pNext = 0; + }else{ + assert( p->u.iNextaMemory) ); + pNext = (SorterRecord*)&pList->aMemory[p->u.iNext]; + } + }else{ + pNext = p->u.pNext; + } + + p->u.pNext = 0; + for(i=0; aSlot[i]; i++){ + p = vdbeSorterMerge(pTask, p, aSlot[i]); + aSlot[i] = 0; + } + aSlot[i] = p; + p = pNext; + } + + p = 0; + for(i=0; i<64; i++){ + if( aSlot[i]==0 ) continue; + p = p ? vdbeSorterMerge(pTask, p, aSlot[i]) : aSlot[i]; + } + pList->pList = p; + + sqlite3_free(aSlot); + assert( pTask->pUnpacked->errCode==SQLITE_OK + || pTask->pUnpacked->errCode==SQLITE_NOMEM + ); + return pTask->pUnpacked->errCode; +} + +/* +** Initialize a PMA-writer object. +*/ +static void vdbePmaWriterInit( + sqlite3_file *pFd, /* File handle to write to */ + PmaWriter *p, /* Object to populate */ + int nBuf, /* Buffer size */ + i64 iStart /* Offset of pFd to begin writing at */ +){ + memset(p, 0, sizeof(PmaWriter)); + p->aBuffer = (u8*)sqlite3Malloc(nBuf); + if( !p->aBuffer ){ + p->eFWErr = SQLITE_NOMEM_BKPT; + }else{ + p->iBufEnd = p->iBufStart = (iStart % nBuf); + p->iWriteOff = iStart - p->iBufStart; + p->nBuffer = nBuf; + p->pFd = pFd; + } +} + +/* +** Write nData bytes of data to the PMA. Return SQLITE_OK +** if successful, or an SQLite error code if an error occurs. +*/ +static void vdbePmaWriteBlob(PmaWriter *p, u8 *pData, int nData){ + int nRem = nData; + while( nRem>0 && p->eFWErr==0 ){ + int nCopy = nRem; + if( nCopy>(p->nBuffer - p->iBufEnd) ){ + nCopy = p->nBuffer - p->iBufEnd; + } + + memcpy(&p->aBuffer[p->iBufEnd], &pData[nData-nRem], nCopy); + p->iBufEnd += nCopy; + if( p->iBufEnd==p->nBuffer ){ + p->eFWErr = sqlite3OsWrite(p->pFd, + &p->aBuffer[p->iBufStart], p->iBufEnd - p->iBufStart, + p->iWriteOff + p->iBufStart + ); + p->iBufStart = p->iBufEnd = 0; + p->iWriteOff += p->nBuffer; + } + assert( p->iBufEndnBuffer ); + + nRem -= nCopy; + } +} + +/* +** Flush any buffered data to disk and clean up the PMA-writer object. +** The results of using the PMA-writer after this call are undefined. +** Return SQLITE_OK if flushing the buffered data succeeds or is not +** required. Otherwise, return an SQLite error code. +** +** Before returning, set *piEof to the offset immediately following the +** last byte written to the file. +*/ +static int vdbePmaWriterFinish(PmaWriter *p, i64 *piEof){ + int rc; + if( p->eFWErr==0 && ALWAYS(p->aBuffer) && p->iBufEnd>p->iBufStart ){ + p->eFWErr = sqlite3OsWrite(p->pFd, + &p->aBuffer[p->iBufStart], p->iBufEnd - p->iBufStart, + p->iWriteOff + p->iBufStart + ); + } + *piEof = (p->iWriteOff + p->iBufEnd); + sqlite3_free(p->aBuffer); + rc = p->eFWErr; + memset(p, 0, sizeof(PmaWriter)); + return rc; +} + +/* +** Write value iVal encoded as a varint to the PMA. Return +** SQLITE_OK if successful, or an SQLite error code if an error occurs. +*/ +static void vdbePmaWriteVarint(PmaWriter *p, u64 iVal){ + int nByte; + u8 aByte[10]; + nByte = sqlite3PutVarint(aByte, iVal); + vdbePmaWriteBlob(p, aByte, nByte); +} + +/* +** Write the current contents of in-memory linked-list pList to a level-0 +** PMA in the temp file belonging to sub-task pTask. Return SQLITE_OK if +** successful, or an SQLite error code otherwise. +** +** The format of a PMA is: +** +** * A varint. This varint contains the total number of bytes of content +** in the PMA (not including the varint itself). +** +** * One or more records packed end-to-end in order of ascending keys. +** Each record consists of a varint followed by a blob of data (the +** key). The varint is the number of bytes in the blob of data. +*/ +static int vdbeSorterListToPMA(SortSubtask *pTask, SorterList *pList){ + sqlite3 *db = pTask->pSorter->db; + int rc = SQLITE_OK; /* Return code */ + PmaWriter writer; /* Object used to write to the file */ + +#ifdef SQLITE_DEBUG + /* Set iSz to the expected size of file pTask->file after writing the PMA. + ** This is used by an assert() statement at the end of this function. */ + i64 iSz = pList->szPMA + sqlite3VarintLen(pList->szPMA) + pTask->file.iEof; +#endif + + vdbeSorterWorkDebug(pTask, "enter"); + memset(&writer, 0, sizeof(PmaWriter)); + assert( pList->szPMA>0 ); + + /* If the first temporary PMA file has not been opened, open it now. */ + if( pTask->file.pFd==0 ){ + rc = vdbeSorterOpenTempFile(db, 0, &pTask->file.pFd); + assert( rc!=SQLITE_OK || pTask->file.pFd ); + assert( pTask->file.iEof==0 ); + assert( pTask->nPMA==0 ); + } + + /* Try to get the file to memory map */ + if( rc==SQLITE_OK ){ + vdbeSorterExtendFile(db, pTask->file.pFd, pTask->file.iEof+pList->szPMA+9); + } + + /* Sort the list */ + if( rc==SQLITE_OK ){ + rc = vdbeSorterSort(pTask, pList); + } + + if( rc==SQLITE_OK ){ + SorterRecord *p; + SorterRecord *pNext = 0; + + vdbePmaWriterInit(pTask->file.pFd, &writer, pTask->pSorter->pgsz, + pTask->file.iEof); + pTask->nPMA++; + vdbePmaWriteVarint(&writer, pList->szPMA); + for(p=pList->pList; p; p=pNext){ + pNext = p->u.pNext; + vdbePmaWriteVarint(&writer, p->nVal); + vdbePmaWriteBlob(&writer, SRVAL(p), p->nVal); + if( pList->aMemory==0 ) sqlite3_free(p); + } + pList->pList = p; + rc = vdbePmaWriterFinish(&writer, &pTask->file.iEof); + } + + vdbeSorterWorkDebug(pTask, "exit"); + assert( rc!=SQLITE_OK || pList->pList==0 ); + assert( rc!=SQLITE_OK || pTask->file.iEof==iSz ); + return rc; +} + +/* +** Advance the MergeEngine to its next entry. +** Set *pbEof to true there is no next entry because +** the MergeEngine has reached the end of all its inputs. +** +** Return SQLITE_OK if successful or an error code if an error occurs. +*/ +static int vdbeMergeEngineStep( + MergeEngine *pMerger, /* The merge engine to advance to the next row */ + int *pbEof /* Set TRUE at EOF. Set false for more content */ +){ + int rc; + int iPrev = pMerger->aTree[1];/* Index of PmaReader to advance */ + SortSubtask *pTask = pMerger->pTask; + + /* Advance the current PmaReader */ + rc = vdbePmaReaderNext(&pMerger->aReadr[iPrev]); + + /* Update contents of aTree[] */ + if( rc==SQLITE_OK ){ + int i; /* Index of aTree[] to recalculate */ + PmaReader *pReadr1; /* First PmaReader to compare */ + PmaReader *pReadr2; /* Second PmaReader to compare */ + int bCached = 0; + + /* Find the first two PmaReaders to compare. The one that was just + ** advanced (iPrev) and the one next to it in the array. */ + pReadr1 = &pMerger->aReadr[(iPrev & 0xFFFE)]; + pReadr2 = &pMerger->aReadr[(iPrev | 0x0001)]; + + for(i=(pMerger->nTree+iPrev)/2; i>0; i=i/2){ + /* Compare pReadr1 and pReadr2. Store the result in variable iRes. */ + int iRes; + if( pReadr1->pFd==0 ){ + iRes = +1; + }else if( pReadr2->pFd==0 ){ + iRes = -1; + }else{ + iRes = pTask->xCompare(pTask, &bCached, + pReadr1->aKey, pReadr1->nKey, pReadr2->aKey, pReadr2->nKey + ); + } + + /* If pReadr1 contained the smaller value, set aTree[i] to its index. + ** Then set pReadr2 to the next PmaReader to compare to pReadr1. In this + ** case there is no cache of pReadr2 in pTask->pUnpacked, so set + ** pKey2 to point to the record belonging to pReadr2. + ** + ** Alternatively, if pReadr2 contains the smaller of the two values, + ** set aTree[i] to its index and update pReadr1. If vdbeSorterCompare() + ** was actually called above, then pTask->pUnpacked now contains + ** a value equivalent to pReadr2. So set pKey2 to NULL to prevent + ** vdbeSorterCompare() from decoding pReadr2 again. + ** + ** If the two values were equal, then the value from the oldest + ** PMA should be considered smaller. The VdbeSorter.aReadr[] array + ** is sorted from oldest to newest, so pReadr1 contains older values + ** than pReadr2 iff (pReadr1aTree[i] = (int)(pReadr1 - pMerger->aReadr); + pReadr2 = &pMerger->aReadr[ pMerger->aTree[i ^ 0x0001] ]; + bCached = 0; + }else{ + if( pReadr1->pFd ) bCached = 0; + pMerger->aTree[i] = (int)(pReadr2 - pMerger->aReadr); + pReadr1 = &pMerger->aReadr[ pMerger->aTree[i ^ 0x0001] ]; + } + } + *pbEof = (pMerger->aReadr[pMerger->aTree[1]].pFd==0); + } + + return (rc==SQLITE_OK ? pTask->pUnpacked->errCode : rc); +} + +#if SQLITE_MAX_WORKER_THREADS>0 +/* +** The main routine for background threads that write level-0 PMAs. +*/ +static void *vdbeSorterFlushThread(void *pCtx){ + SortSubtask *pTask = (SortSubtask*)pCtx; + int rc; /* Return code */ + assert( pTask->bDone==0 ); + rc = vdbeSorterListToPMA(pTask, &pTask->list); + pTask->bDone = 1; + return SQLITE_INT_TO_PTR(rc); +} +#endif /* SQLITE_MAX_WORKER_THREADS>0 */ + +/* +** Flush the current contents of VdbeSorter.list to a new PMA, possibly +** using a background thread. +*/ +static int vdbeSorterFlushPMA(VdbeSorter *pSorter){ +#if SQLITE_MAX_WORKER_THREADS==0 + pSorter->bUsePMA = 1; + return vdbeSorterListToPMA(&pSorter->aTask[0], &pSorter->list); +#else + int rc = SQLITE_OK; + int i; + SortSubtask *pTask = 0; /* Thread context used to create new PMA */ + int nWorker = (pSorter->nTask-1); + + /* Set the flag to indicate that at least one PMA has been written. + ** Or will be, anyhow. */ + pSorter->bUsePMA = 1; + + /* Select a sub-task to sort and flush the current list of in-memory + ** records to disk. If the sorter is running in multi-threaded mode, + ** round-robin between the first (pSorter->nTask-1) tasks. Except, if + ** the background thread from a sub-tasks previous turn is still running, + ** skip it. If the first (pSorter->nTask-1) sub-tasks are all still busy, + ** fall back to using the final sub-task. The first (pSorter->nTask-1) + ** sub-tasks are prefered as they use background threads - the final + ** sub-task uses the main thread. */ + for(i=0; iiPrev + i + 1) % nWorker; + pTask = &pSorter->aTask[iTest]; + if( pTask->bDone ){ + rc = vdbeSorterJoinThread(pTask); + } + if( rc!=SQLITE_OK || pTask->pThread==0 ) break; + } + + if( rc==SQLITE_OK ){ + if( i==nWorker ){ + /* Use the foreground thread for this operation */ + rc = vdbeSorterListToPMA(&pSorter->aTask[nWorker], &pSorter->list); + }else{ + /* Launch a background thread for this operation */ + u8 *aMem; + void *pCtx; + + assert( pTask!=0 ); + assert( pTask->pThread==0 && pTask->bDone==0 ); + assert( pTask->list.pList==0 ); + assert( pTask->list.aMemory==0 || pSorter->list.aMemory!=0 ); + + aMem = pTask->list.aMemory; + pCtx = (void*)pTask; + pSorter->iPrev = (u8)(pTask - pSorter->aTask); + pTask->list = pSorter->list; + pSorter->list.pList = 0; + pSorter->list.szPMA = 0; + if( aMem ){ + pSorter->list.aMemory = aMem; + pSorter->nMemory = sqlite3MallocSize(aMem); + }else if( pSorter->list.aMemory ){ + pSorter->list.aMemory = sqlite3Malloc(pSorter->nMemory); + if( !pSorter->list.aMemory ) return SQLITE_NOMEM_BKPT; + } + + rc = vdbeSorterCreateThread(pTask, vdbeSorterFlushThread, pCtx); + } + } + + return rc; +#endif /* SQLITE_MAX_WORKER_THREADS!=0 */ +} + +/* +** Add a record to the sorter. +*/ +SQLITE_PRIVATE int sqlite3VdbeSorterWrite( + const VdbeCursor *pCsr, /* Sorter cursor */ + Mem *pVal /* Memory cell containing record */ +){ + VdbeSorter *pSorter; + int rc = SQLITE_OK; /* Return Code */ + SorterRecord *pNew; /* New list element */ + int bFlush; /* True to flush contents of memory to PMA */ + int nReq; /* Bytes of memory required */ + int nPMA; /* Bytes of PMA space required */ + int t; /* serial type of first record field */ + + assert( pCsr->eCurType==CURTYPE_SORTER ); + pSorter = pCsr->uc.pSorter; + getVarint32((const u8*)&pVal->z[1], t); + if( t>0 && t<10 && t!=7 ){ + pSorter->typeMask &= SORTER_TYPE_INTEGER; + }else if( t>10 && (t & 0x01) ){ + pSorter->typeMask &= SORTER_TYPE_TEXT; + }else{ + pSorter->typeMask = 0; + } + + assert( pSorter ); + + /* Figure out whether or not the current contents of memory should be + ** flushed to a PMA before continuing. If so, do so. + ** + ** If using the single large allocation mode (pSorter->aMemory!=0), then + ** flush the contents of memory to a new PMA if (a) at least one value is + ** already in memory and (b) the new value will not fit in memory. + ** + ** Or, if using separate allocations for each record, flush the contents + ** of memory to a PMA if either of the following are true: + ** + ** * The total memory allocated for the in-memory list is greater + ** than (page-size * cache-size), or + ** + ** * The total memory allocated for the in-memory list is greater + ** than (page-size * 10) and sqlite3HeapNearlyFull() returns true. + */ + nReq = pVal->n + sizeof(SorterRecord); + nPMA = pVal->n + sqlite3VarintLen(pVal->n); + if( pSorter->mxPmaSize ){ + if( pSorter->list.aMemory ){ + bFlush = pSorter->iMemory && (pSorter->iMemory+nReq) > pSorter->mxPmaSize; + }else{ + bFlush = ( + (pSorter->list.szPMA > pSorter->mxPmaSize) + || (pSorter->list.szPMA > pSorter->mnPmaSize && sqlite3HeapNearlyFull()) + ); + } + if( bFlush ){ + rc = vdbeSorterFlushPMA(pSorter); + pSorter->list.szPMA = 0; + pSorter->iMemory = 0; + assert( rc!=SQLITE_OK || pSorter->list.pList==0 ); + } + } + + pSorter->list.szPMA += nPMA; + if( nPMA>pSorter->mxKeysize ){ + pSorter->mxKeysize = nPMA; + } + + if( pSorter->list.aMemory ){ + int nMin = pSorter->iMemory + nReq; + + if( nMin>pSorter->nMemory ){ + u8 *aNew; + sqlite3_int64 nNew = 2 * (sqlite3_int64)pSorter->nMemory; + int iListOff = -1; + if( pSorter->list.pList ){ + iListOff = (u8*)pSorter->list.pList - pSorter->list.aMemory; + } + while( nNew < nMin ) nNew = nNew*2; + if( nNew > pSorter->mxPmaSize ) nNew = pSorter->mxPmaSize; + if( nNew < nMin ) nNew = nMin; + aNew = sqlite3Realloc(pSorter->list.aMemory, nNew); + if( !aNew ) return SQLITE_NOMEM_BKPT; + if( iListOff>=0 ){ + pSorter->list.pList = (SorterRecord*)&aNew[iListOff]; + } + pSorter->list.aMemory = aNew; + pSorter->nMemory = nNew; + } + + pNew = (SorterRecord*)&pSorter->list.aMemory[pSorter->iMemory]; + pSorter->iMemory += ROUND8(nReq); + if( pSorter->list.pList ){ + pNew->u.iNext = (int)((u8*)(pSorter->list.pList) - pSorter->list.aMemory); + } + }else{ + pNew = (SorterRecord *)sqlite3Malloc(nReq); + if( pNew==0 ){ + return SQLITE_NOMEM_BKPT; + } + pNew->u.pNext = pSorter->list.pList; + } + + memcpy(SRVAL(pNew), pVal->z, pVal->n); + pNew->nVal = pVal->n; + pSorter->list.pList = pNew; + + return rc; +} + +/* +** Read keys from pIncr->pMerger and populate pIncr->aFile[1]. The format +** of the data stored in aFile[1] is the same as that used by regular PMAs, +** except that the number-of-bytes varint is omitted from the start. +*/ +static int vdbeIncrPopulate(IncrMerger *pIncr){ + int rc = SQLITE_OK; + int rc2; + i64 iStart = pIncr->iStartOff; + SorterFile *pOut = &pIncr->aFile[1]; + SortSubtask *pTask = pIncr->pTask; + MergeEngine *pMerger = pIncr->pMerger; + PmaWriter writer; + assert( pIncr->bEof==0 ); + + vdbeSorterPopulateDebug(pTask, "enter"); + + vdbePmaWriterInit(pOut->pFd, &writer, pTask->pSorter->pgsz, iStart); + while( rc==SQLITE_OK ){ + int dummy; + PmaReader *pReader = &pMerger->aReadr[ pMerger->aTree[1] ]; + int nKey = pReader->nKey; + i64 iEof = writer.iWriteOff + writer.iBufEnd; + + /* Check if the output file is full or if the input has been exhausted. + ** In either case exit the loop. */ + if( pReader->pFd==0 ) break; + if( (iEof + nKey + sqlite3VarintLen(nKey))>(iStart + pIncr->mxSz) ) break; + + /* Write the next key to the output. */ + vdbePmaWriteVarint(&writer, nKey); + vdbePmaWriteBlob(&writer, pReader->aKey, nKey); + assert( pIncr->pMerger->pTask==pTask ); + rc = vdbeMergeEngineStep(pIncr->pMerger, &dummy); + } + + rc2 = vdbePmaWriterFinish(&writer, &pOut->iEof); + if( rc==SQLITE_OK ) rc = rc2; + vdbeSorterPopulateDebug(pTask, "exit"); + return rc; +} + +#if SQLITE_MAX_WORKER_THREADS>0 +/* +** The main routine for background threads that populate aFile[1] of +** multi-threaded IncrMerger objects. +*/ +static void *vdbeIncrPopulateThread(void *pCtx){ + IncrMerger *pIncr = (IncrMerger*)pCtx; + void *pRet = SQLITE_INT_TO_PTR( vdbeIncrPopulate(pIncr) ); + pIncr->pTask->bDone = 1; + return pRet; +} + +/* +** Launch a background thread to populate aFile[1] of pIncr. +*/ +static int vdbeIncrBgPopulate(IncrMerger *pIncr){ + void *p = (void*)pIncr; + assert( pIncr->bUseThread ); + return vdbeSorterCreateThread(pIncr->pTask, vdbeIncrPopulateThread, p); +} +#endif + +/* +** This function is called when the PmaReader corresponding to pIncr has +** finished reading the contents of aFile[0]. Its purpose is to "refill" +** aFile[0] such that the PmaReader should start rereading it from the +** beginning. +** +** For single-threaded objects, this is accomplished by literally reading +** keys from pIncr->pMerger and repopulating aFile[0]. +** +** For multi-threaded objects, all that is required is to wait until the +** background thread is finished (if it is not already) and then swap +** aFile[0] and aFile[1] in place. If the contents of pMerger have not +** been exhausted, this function also launches a new background thread +** to populate the new aFile[1]. +** +** SQLITE_OK is returned on success, or an SQLite error code otherwise. +*/ +static int vdbeIncrSwap(IncrMerger *pIncr){ + int rc = SQLITE_OK; + +#if SQLITE_MAX_WORKER_THREADS>0 + if( pIncr->bUseThread ){ + rc = vdbeSorterJoinThread(pIncr->pTask); + + if( rc==SQLITE_OK ){ + SorterFile f0 = pIncr->aFile[0]; + pIncr->aFile[0] = pIncr->aFile[1]; + pIncr->aFile[1] = f0; + } + + if( rc==SQLITE_OK ){ + if( pIncr->aFile[0].iEof==pIncr->iStartOff ){ + pIncr->bEof = 1; + }else{ + rc = vdbeIncrBgPopulate(pIncr); + } + } + }else +#endif + { + rc = vdbeIncrPopulate(pIncr); + pIncr->aFile[0] = pIncr->aFile[1]; + if( pIncr->aFile[0].iEof==pIncr->iStartOff ){ + pIncr->bEof = 1; + } + } + + return rc; +} + +/* +** Allocate and return a new IncrMerger object to read data from pMerger. +** +** If an OOM condition is encountered, return NULL. In this case free the +** pMerger argument before returning. +*/ +static int vdbeIncrMergerNew( + SortSubtask *pTask, /* The thread that will be using the new IncrMerger */ + MergeEngine *pMerger, /* The MergeEngine that the IncrMerger will control */ + IncrMerger **ppOut /* Write the new IncrMerger here */ +){ + int rc = SQLITE_OK; + IncrMerger *pIncr = *ppOut = (IncrMerger*) + (sqlite3FaultSim(100) ? 0 : sqlite3MallocZero(sizeof(*pIncr))); + if( pIncr ){ + pIncr->pMerger = pMerger; + pIncr->pTask = pTask; + pIncr->mxSz = MAX(pTask->pSorter->mxKeysize+9,pTask->pSorter->mxPmaSize/2); + pTask->file2.iEof += pIncr->mxSz; + }else{ + vdbeMergeEngineFree(pMerger); + rc = SQLITE_NOMEM_BKPT; + } + return rc; +} + +#if SQLITE_MAX_WORKER_THREADS>0 +/* +** Set the "use-threads" flag on object pIncr. +*/ +static void vdbeIncrMergerSetThreads(IncrMerger *pIncr){ + pIncr->bUseThread = 1; + pIncr->pTask->file2.iEof -= pIncr->mxSz; +} +#endif /* SQLITE_MAX_WORKER_THREADS>0 */ + + + +/* +** Recompute pMerger->aTree[iOut] by comparing the next keys on the +** two PmaReaders that feed that entry. Neither of the PmaReaders +** are advanced. This routine merely does the comparison. +*/ +static void vdbeMergeEngineCompare( + MergeEngine *pMerger, /* Merge engine containing PmaReaders to compare */ + int iOut /* Store the result in pMerger->aTree[iOut] */ +){ + int i1; + int i2; + int iRes; + PmaReader *p1; + PmaReader *p2; + + assert( iOutnTree && iOut>0 ); + + if( iOut>=(pMerger->nTree/2) ){ + i1 = (iOut - pMerger->nTree/2) * 2; + i2 = i1 + 1; + }else{ + i1 = pMerger->aTree[iOut*2]; + i2 = pMerger->aTree[iOut*2+1]; + } + + p1 = &pMerger->aReadr[i1]; + p2 = &pMerger->aReadr[i2]; + + if( p1->pFd==0 ){ + iRes = i2; + }else if( p2->pFd==0 ){ + iRes = i1; + }else{ + SortSubtask *pTask = pMerger->pTask; + int bCached = 0; + int res; + assert( pTask->pUnpacked!=0 ); /* from vdbeSortSubtaskMain() */ + res = pTask->xCompare( + pTask, &bCached, p1->aKey, p1->nKey, p2->aKey, p2->nKey + ); + if( res<=0 ){ + iRes = i1; + }else{ + iRes = i2; + } + } + + pMerger->aTree[iOut] = iRes; +} + +/* +** Allowed values for the eMode parameter to vdbeMergeEngineInit() +** and vdbePmaReaderIncrMergeInit(). +** +** Only INCRINIT_NORMAL is valid in single-threaded builds (when +** SQLITE_MAX_WORKER_THREADS==0). The other values are only used +** when there exists one or more separate worker threads. +*/ +#define INCRINIT_NORMAL 0 +#define INCRINIT_TASK 1 +#define INCRINIT_ROOT 2 + +/* +** Forward reference required as the vdbeIncrMergeInit() and +** vdbePmaReaderIncrInit() routines are called mutually recursively when +** building a merge tree. +*/ +static int vdbePmaReaderIncrInit(PmaReader *pReadr, int eMode); + +/* +** Initialize the MergeEngine object passed as the second argument. Once this +** function returns, the first key of merged data may be read from the +** MergeEngine object in the usual fashion. +** +** If argument eMode is INCRINIT_ROOT, then it is assumed that any IncrMerge +** objects attached to the PmaReader objects that the merger reads from have +** already been populated, but that they have not yet populated aFile[0] and +** set the PmaReader objects up to read from it. In this case all that is +** required is to call vdbePmaReaderNext() on each PmaReader to point it at +** its first key. +** +** Otherwise, if eMode is any value other than INCRINIT_ROOT, then use +** vdbePmaReaderIncrMergeInit() to initialize each PmaReader that feeds data +** to pMerger. +** +** SQLITE_OK is returned if successful, or an SQLite error code otherwise. +*/ +static int vdbeMergeEngineInit( + SortSubtask *pTask, /* Thread that will run pMerger */ + MergeEngine *pMerger, /* MergeEngine to initialize */ + int eMode /* One of the INCRINIT_XXX constants */ +){ + int rc = SQLITE_OK; /* Return code */ + int i; /* For looping over PmaReader objects */ + int nTree; /* Number of subtrees to merge */ + + /* Failure to allocate the merge would have been detected prior to + ** invoking this routine */ + assert( pMerger!=0 ); + + /* eMode is always INCRINIT_NORMAL in single-threaded mode */ + assert( SQLITE_MAX_WORKER_THREADS>0 || eMode==INCRINIT_NORMAL ); + + /* Verify that the MergeEngine is assigned to a single thread */ + assert( pMerger->pTask==0 ); + pMerger->pTask = pTask; + + nTree = pMerger->nTree; + for(i=0; i0 && eMode==INCRINIT_ROOT ){ + /* PmaReaders should be normally initialized in order, as if they are + ** reading from the same temp file this makes for more linear file IO. + ** However, in the INCRINIT_ROOT case, if PmaReader aReadr[nTask-1] is + ** in use it will block the vdbePmaReaderNext() call while it uses + ** the main thread to fill its buffer. So calling PmaReaderNext() + ** on this PmaReader before any of the multi-threaded PmaReaders takes + ** better advantage of multi-processor hardware. */ + rc = vdbePmaReaderNext(&pMerger->aReadr[nTree-i-1]); + }else{ + rc = vdbePmaReaderIncrInit(&pMerger->aReadr[i], INCRINIT_NORMAL); + } + if( rc!=SQLITE_OK ) return rc; + } + + for(i=pMerger->nTree-1; i>0; i--){ + vdbeMergeEngineCompare(pMerger, i); + } + return pTask->pUnpacked->errCode; +} + +/* +** The PmaReader passed as the first argument is guaranteed to be an +** incremental-reader (pReadr->pIncr!=0). This function serves to open +** and/or initialize the temp file related fields of the IncrMerge +** object at (pReadr->pIncr). +** +** If argument eMode is set to INCRINIT_NORMAL, then all PmaReaders +** in the sub-tree headed by pReadr are also initialized. Data is then +** loaded into the buffers belonging to pReadr and it is set to point to +** the first key in its range. +** +** If argument eMode is set to INCRINIT_TASK, then pReadr is guaranteed +** to be a multi-threaded PmaReader and this function is being called in a +** background thread. In this case all PmaReaders in the sub-tree are +** initialized as for INCRINIT_NORMAL and the aFile[1] buffer belonging to +** pReadr is populated. However, pReadr itself is not set up to point +** to its first key. A call to vdbePmaReaderNext() is still required to do +** that. +** +** The reason this function does not call vdbePmaReaderNext() immediately +** in the INCRINIT_TASK case is that vdbePmaReaderNext() assumes that it has +** to block on thread (pTask->thread) before accessing aFile[1]. But, since +** this entire function is being run by thread (pTask->thread), that will +** lead to the current background thread attempting to join itself. +** +** Finally, if argument eMode is set to INCRINIT_ROOT, it may be assumed +** that pReadr->pIncr is a multi-threaded IncrMerge objects, and that all +** child-trees have already been initialized using IncrInit(INCRINIT_TASK). +** In this case vdbePmaReaderNext() is called on all child PmaReaders and +** the current PmaReader set to point to the first key in its range. +** +** SQLITE_OK is returned if successful, or an SQLite error code otherwise. +*/ +static int vdbePmaReaderIncrMergeInit(PmaReader *pReadr, int eMode){ + int rc = SQLITE_OK; + IncrMerger *pIncr = pReadr->pIncr; + SortSubtask *pTask = pIncr->pTask; + sqlite3 *db = pTask->pSorter->db; + + /* eMode is always INCRINIT_NORMAL in single-threaded mode */ + assert( SQLITE_MAX_WORKER_THREADS>0 || eMode==INCRINIT_NORMAL ); + + rc = vdbeMergeEngineInit(pTask, pIncr->pMerger, eMode); + + /* Set up the required files for pIncr. A multi-theaded IncrMerge object + ** requires two temp files to itself, whereas a single-threaded object + ** only requires a region of pTask->file2. */ + if( rc==SQLITE_OK ){ + int mxSz = pIncr->mxSz; +#if SQLITE_MAX_WORKER_THREADS>0 + if( pIncr->bUseThread ){ + rc = vdbeSorterOpenTempFile(db, mxSz, &pIncr->aFile[0].pFd); + if( rc==SQLITE_OK ){ + rc = vdbeSorterOpenTempFile(db, mxSz, &pIncr->aFile[1].pFd); + } + }else +#endif + /*if( !pIncr->bUseThread )*/{ + if( pTask->file2.pFd==0 ){ + assert( pTask->file2.iEof>0 ); + rc = vdbeSorterOpenTempFile(db, pTask->file2.iEof, &pTask->file2.pFd); + pTask->file2.iEof = 0; + } + if( rc==SQLITE_OK ){ + pIncr->aFile[1].pFd = pTask->file2.pFd; + pIncr->iStartOff = pTask->file2.iEof; + pTask->file2.iEof += mxSz; + } + } + } + +#if SQLITE_MAX_WORKER_THREADS>0 + if( rc==SQLITE_OK && pIncr->bUseThread ){ + /* Use the current thread to populate aFile[1], even though this + ** PmaReader is multi-threaded. If this is an INCRINIT_TASK object, + ** then this function is already running in background thread + ** pIncr->pTask->thread. + ** + ** If this is the INCRINIT_ROOT object, then it is running in the + ** main VDBE thread. But that is Ok, as that thread cannot return + ** control to the VDBE or proceed with anything useful until the + ** first results are ready from this merger object anyway. + */ + assert( eMode==INCRINIT_ROOT || eMode==INCRINIT_TASK ); + rc = vdbeIncrPopulate(pIncr); + } +#endif + + if( rc==SQLITE_OK && (SQLITE_MAX_WORKER_THREADS==0 || eMode!=INCRINIT_TASK) ){ + rc = vdbePmaReaderNext(pReadr); + } + + return rc; +} + +#if SQLITE_MAX_WORKER_THREADS>0 +/* +** The main routine for vdbePmaReaderIncrMergeInit() operations run in +** background threads. +*/ +static void *vdbePmaReaderBgIncrInit(void *pCtx){ + PmaReader *pReader = (PmaReader*)pCtx; + void *pRet = SQLITE_INT_TO_PTR( + vdbePmaReaderIncrMergeInit(pReader,INCRINIT_TASK) + ); + pReader->pIncr->pTask->bDone = 1; + return pRet; +} +#endif + +/* +** If the PmaReader passed as the first argument is not an incremental-reader +** (if pReadr->pIncr==0), then this function is a no-op. Otherwise, it invokes +** the vdbePmaReaderIncrMergeInit() function with the parameters passed to +** this routine to initialize the incremental merge. +** +** If the IncrMerger object is multi-threaded (IncrMerger.bUseThread==1), +** then a background thread is launched to call vdbePmaReaderIncrMergeInit(). +** Or, if the IncrMerger is single threaded, the same function is called +** using the current thread. +*/ +static int vdbePmaReaderIncrInit(PmaReader *pReadr, int eMode){ + IncrMerger *pIncr = pReadr->pIncr; /* Incremental merger */ + int rc = SQLITE_OK; /* Return code */ + if( pIncr ){ +#if SQLITE_MAX_WORKER_THREADS>0 + assert( pIncr->bUseThread==0 || eMode==INCRINIT_TASK ); + if( pIncr->bUseThread ){ + void *pCtx = (void*)pReadr; + rc = vdbeSorterCreateThread(pIncr->pTask, vdbePmaReaderBgIncrInit, pCtx); + }else +#endif + { + rc = vdbePmaReaderIncrMergeInit(pReadr, eMode); + } + } + return rc; +} + +/* +** Allocate a new MergeEngine object to merge the contents of nPMA level-0 +** PMAs from pTask->file. If no error occurs, set *ppOut to point to +** the new object and return SQLITE_OK. Or, if an error does occur, set *ppOut +** to NULL and return an SQLite error code. +** +** When this function is called, *piOffset is set to the offset of the +** first PMA to read from pTask->file. Assuming no error occurs, it is +** set to the offset immediately following the last byte of the last +** PMA before returning. If an error does occur, then the final value of +** *piOffset is undefined. +*/ +static int vdbeMergeEngineLevel0( + SortSubtask *pTask, /* Sorter task to read from */ + int nPMA, /* Number of PMAs to read */ + i64 *piOffset, /* IN/OUT: Readr offset in pTask->file */ + MergeEngine **ppOut /* OUT: New merge-engine */ +){ + MergeEngine *pNew; /* Merge engine to return */ + i64 iOff = *piOffset; + int i; + int rc = SQLITE_OK; + + *ppOut = pNew = vdbeMergeEngineNew(nPMA); + if( pNew==0 ) rc = SQLITE_NOMEM_BKPT; + + for(i=0; iaReadr[i]; + rc = vdbePmaReaderInit(pTask, &pTask->file, iOff, pReadr, &nDummy); + iOff = pReadr->iEof; + } + + if( rc!=SQLITE_OK ){ + vdbeMergeEngineFree(pNew); + *ppOut = 0; + } + *piOffset = iOff; + return rc; +} + +/* +** Return the depth of a tree comprising nPMA PMAs, assuming a fanout of +** SORTER_MAX_MERGE_COUNT. The returned value does not include leaf nodes. +** +** i.e. +** +** nPMA<=16 -> TreeDepth() == 0 +** nPMA<=256 -> TreeDepth() == 1 +** nPMA<=65536 -> TreeDepth() == 2 +*/ +static int vdbeSorterTreeDepth(int nPMA){ + int nDepth = 0; + i64 nDiv = SORTER_MAX_MERGE_COUNT; + while( nDiv < (i64)nPMA ){ + nDiv = nDiv * SORTER_MAX_MERGE_COUNT; + nDepth++; + } + return nDepth; +} + +/* +** pRoot is the root of an incremental merge-tree with depth nDepth (according +** to vdbeSorterTreeDepth()). pLeaf is the iSeq'th leaf to be added to the +** tree, counting from zero. This function adds pLeaf to the tree. +** +** If successful, SQLITE_OK is returned. If an error occurs, an SQLite error +** code is returned and pLeaf is freed. +*/ +static int vdbeSorterAddToTree( + SortSubtask *pTask, /* Task context */ + int nDepth, /* Depth of tree according to TreeDepth() */ + int iSeq, /* Sequence number of leaf within tree */ + MergeEngine *pRoot, /* Root of tree */ + MergeEngine *pLeaf /* Leaf to add to tree */ +){ + int rc = SQLITE_OK; + int nDiv = 1; + int i; + MergeEngine *p = pRoot; + IncrMerger *pIncr; + + rc = vdbeIncrMergerNew(pTask, pLeaf, &pIncr); + + for(i=1; iaReadr[iIter]; + + if( pReadr->pIncr==0 ){ + MergeEngine *pNew = vdbeMergeEngineNew(SORTER_MAX_MERGE_COUNT); + if( pNew==0 ){ + rc = SQLITE_NOMEM_BKPT; + }else{ + rc = vdbeIncrMergerNew(pTask, pNew, &pReadr->pIncr); + } + } + if( rc==SQLITE_OK ){ + p = pReadr->pIncr->pMerger; + nDiv = nDiv / SORTER_MAX_MERGE_COUNT; + } + } + + if( rc==SQLITE_OK ){ + p->aReadr[iSeq % SORTER_MAX_MERGE_COUNT].pIncr = pIncr; + }else{ + vdbeIncrFree(pIncr); + } + return rc; +} + +/* +** This function is called as part of a SorterRewind() operation on a sorter +** that has already written two or more level-0 PMAs to one or more temp +** files. It builds a tree of MergeEngine/IncrMerger/PmaReader objects that +** can be used to incrementally merge all PMAs on disk. +** +** If successful, SQLITE_OK is returned and *ppOut set to point to the +** MergeEngine object at the root of the tree before returning. Or, if an +** error occurs, an SQLite error code is returned and the final value +** of *ppOut is undefined. +*/ +static int vdbeSorterMergeTreeBuild( + VdbeSorter *pSorter, /* The VDBE cursor that implements the sort */ + MergeEngine **ppOut /* Write the MergeEngine here */ +){ + MergeEngine *pMain = 0; + int rc = SQLITE_OK; + int iTask; + +#if SQLITE_MAX_WORKER_THREADS>0 + /* If the sorter uses more than one task, then create the top-level + ** MergeEngine here. This MergeEngine will read data from exactly + ** one PmaReader per sub-task. */ + assert( pSorter->bUseThreads || pSorter->nTask==1 ); + if( pSorter->nTask>1 ){ + pMain = vdbeMergeEngineNew(pSorter->nTask); + if( pMain==0 ) rc = SQLITE_NOMEM_BKPT; + } +#endif + + for(iTask=0; rc==SQLITE_OK && iTasknTask; iTask++){ + SortSubtask *pTask = &pSorter->aTask[iTask]; + assert( pTask->nPMA>0 || SQLITE_MAX_WORKER_THREADS>0 ); + if( SQLITE_MAX_WORKER_THREADS==0 || pTask->nPMA ){ + MergeEngine *pRoot = 0; /* Root node of tree for this task */ + int nDepth = vdbeSorterTreeDepth(pTask->nPMA); + i64 iReadOff = 0; + + if( pTask->nPMA<=SORTER_MAX_MERGE_COUNT ){ + rc = vdbeMergeEngineLevel0(pTask, pTask->nPMA, &iReadOff, &pRoot); + }else{ + int i; + int iSeq = 0; + pRoot = vdbeMergeEngineNew(SORTER_MAX_MERGE_COUNT); + if( pRoot==0 ) rc = SQLITE_NOMEM_BKPT; + for(i=0; inPMA && rc==SQLITE_OK; i += SORTER_MAX_MERGE_COUNT){ + MergeEngine *pMerger = 0; /* New level-0 PMA merger */ + int nReader; /* Number of level-0 PMAs to merge */ + + nReader = MIN(pTask->nPMA - i, SORTER_MAX_MERGE_COUNT); + rc = vdbeMergeEngineLevel0(pTask, nReader, &iReadOff, &pMerger); + if( rc==SQLITE_OK ){ + rc = vdbeSorterAddToTree(pTask, nDepth, iSeq++, pRoot, pMerger); + } + } + } + + if( rc==SQLITE_OK ){ +#if SQLITE_MAX_WORKER_THREADS>0 + if( pMain!=0 ){ + rc = vdbeIncrMergerNew(pTask, pRoot, &pMain->aReadr[iTask].pIncr); + }else +#endif + { + assert( pMain==0 ); + pMain = pRoot; + } + }else{ + vdbeMergeEngineFree(pRoot); + } + } + } + + if( rc!=SQLITE_OK ){ + vdbeMergeEngineFree(pMain); + pMain = 0; + } + *ppOut = pMain; + return rc; +} + +/* +** This function is called as part of an sqlite3VdbeSorterRewind() operation +** on a sorter that has written two or more PMAs to temporary files. It sets +** up either VdbeSorter.pMerger (for single threaded sorters) or pReader +** (for multi-threaded sorters) so that it can be used to iterate through +** all records stored in the sorter. +** +** SQLITE_OK is returned if successful, or an SQLite error code otherwise. +*/ +static int vdbeSorterSetupMerge(VdbeSorter *pSorter){ + int rc; /* Return code */ + SortSubtask *pTask0 = &pSorter->aTask[0]; + MergeEngine *pMain = 0; +#if SQLITE_MAX_WORKER_THREADS + sqlite3 *db = pTask0->pSorter->db; + int i; + SorterCompare xCompare = vdbeSorterGetCompare(pSorter); + for(i=0; inTask; i++){ + pSorter->aTask[i].xCompare = xCompare; + } +#endif + + rc = vdbeSorterMergeTreeBuild(pSorter, &pMain); + if( rc==SQLITE_OK ){ +#if SQLITE_MAX_WORKER_THREADS + assert( pSorter->bUseThreads==0 || pSorter->nTask>1 ); + if( pSorter->bUseThreads ){ + int iTask; + PmaReader *pReadr = 0; + SortSubtask *pLast = &pSorter->aTask[pSorter->nTask-1]; + rc = vdbeSortAllocUnpacked(pLast); + if( rc==SQLITE_OK ){ + pReadr = (PmaReader*)sqlite3DbMallocZero(db, sizeof(PmaReader)); + pSorter->pReader = pReadr; + if( pReadr==0 ) rc = SQLITE_NOMEM_BKPT; + } + if( rc==SQLITE_OK ){ + rc = vdbeIncrMergerNew(pLast, pMain, &pReadr->pIncr); + if( rc==SQLITE_OK ){ + vdbeIncrMergerSetThreads(pReadr->pIncr); + for(iTask=0; iTask<(pSorter->nTask-1); iTask++){ + IncrMerger *pIncr; + if( (pIncr = pMain->aReadr[iTask].pIncr) ){ + vdbeIncrMergerSetThreads(pIncr); + assert( pIncr->pTask!=pLast ); + } + } + for(iTask=0; rc==SQLITE_OK && iTasknTask; iTask++){ + /* Check that: + ** + ** a) The incremental merge object is configured to use the + ** right task, and + ** b) If it is using task (nTask-1), it is configured to run + ** in single-threaded mode. This is important, as the + ** root merge (INCRINIT_ROOT) will be using the same task + ** object. + */ + PmaReader *p = &pMain->aReadr[iTask]; + assert( p->pIncr==0 || ( + (p->pIncr->pTask==&pSorter->aTask[iTask]) /* a */ + && (iTask!=pSorter->nTask-1 || p->pIncr->bUseThread==0) /* b */ + )); + rc = vdbePmaReaderIncrInit(p, INCRINIT_TASK); + } + } + pMain = 0; + } + if( rc==SQLITE_OK ){ + rc = vdbePmaReaderIncrMergeInit(pReadr, INCRINIT_ROOT); + } + }else +#endif + { + rc = vdbeMergeEngineInit(pTask0, pMain, INCRINIT_NORMAL); + pSorter->pMerger = pMain; + pMain = 0; + } + } + + if( rc!=SQLITE_OK ){ + vdbeMergeEngineFree(pMain); + } + return rc; +} + + +/* +** Once the sorter has been populated by calls to sqlite3VdbeSorterWrite, +** this function is called to prepare for iterating through the records +** in sorted order. +*/ +SQLITE_PRIVATE int sqlite3VdbeSorterRewind(const VdbeCursor *pCsr, int *pbEof){ + VdbeSorter *pSorter; + int rc = SQLITE_OK; /* Return code */ + + assert( pCsr->eCurType==CURTYPE_SORTER ); + pSorter = pCsr->uc.pSorter; + assert( pSorter ); + + /* If no data has been written to disk, then do not do so now. Instead, + ** sort the VdbeSorter.pRecord list. The vdbe layer will read data directly + ** from the in-memory list. */ + if( pSorter->bUsePMA==0 ){ + if( pSorter->list.pList ){ + *pbEof = 0; + rc = vdbeSorterSort(&pSorter->aTask[0], &pSorter->list); + }else{ + *pbEof = 1; + } + return rc; + } + + /* Write the current in-memory list to a PMA. When the VdbeSorterWrite() + ** function flushes the contents of memory to disk, it immediately always + ** creates a new list consisting of a single key immediately afterwards. + ** So the list is never empty at this point. */ + assert( pSorter->list.pList ); + rc = vdbeSorterFlushPMA(pSorter); + + /* Join all threads */ + rc = vdbeSorterJoinAll(pSorter, rc); + + vdbeSorterRewindDebug("rewind"); + + /* Assuming no errors have occurred, set up a merger structure to + ** incrementally read and merge all remaining PMAs. */ + assert( pSorter->pReader==0 ); + if( rc==SQLITE_OK ){ + rc = vdbeSorterSetupMerge(pSorter); + *pbEof = 0; + } + + vdbeSorterRewindDebug("rewinddone"); + return rc; +} + +/* +** Advance to the next element in the sorter. Return value: +** +** SQLITE_OK success +** SQLITE_DONE end of data +** otherwise some kind of error. +*/ +SQLITE_PRIVATE int sqlite3VdbeSorterNext(sqlite3 *db, const VdbeCursor *pCsr){ + VdbeSorter *pSorter; + int rc; /* Return code */ + + assert( pCsr->eCurType==CURTYPE_SORTER ); + pSorter = pCsr->uc.pSorter; + assert( pSorter->bUsePMA || (pSorter->pReader==0 && pSorter->pMerger==0) ); + if( pSorter->bUsePMA ){ + assert( pSorter->pReader==0 || pSorter->pMerger==0 ); + assert( pSorter->bUseThreads==0 || pSorter->pReader ); + assert( pSorter->bUseThreads==1 || pSorter->pMerger ); +#if SQLITE_MAX_WORKER_THREADS>0 + if( pSorter->bUseThreads ){ + rc = vdbePmaReaderNext(pSorter->pReader); + if( rc==SQLITE_OK && pSorter->pReader->pFd==0 ) rc = SQLITE_DONE; + }else +#endif + /*if( !pSorter->bUseThreads )*/ { + int res = 0; + assert( pSorter->pMerger!=0 ); + assert( pSorter->pMerger->pTask==(&pSorter->aTask[0]) ); + rc = vdbeMergeEngineStep(pSorter->pMerger, &res); + if( rc==SQLITE_OK && res ) rc = SQLITE_DONE; + } + }else{ + SorterRecord *pFree = pSorter->list.pList; + pSorter->list.pList = pFree->u.pNext; + pFree->u.pNext = 0; + if( pSorter->list.aMemory==0 ) vdbeSorterRecordFree(db, pFree); + rc = pSorter->list.pList ? SQLITE_OK : SQLITE_DONE; + } + return rc; +} + +/* +** Return a pointer to a buffer owned by the sorter that contains the +** current key. +*/ +static void *vdbeSorterRowkey( + const VdbeSorter *pSorter, /* Sorter object */ + int *pnKey /* OUT: Size of current key in bytes */ +){ + void *pKey; + if( pSorter->bUsePMA ){ + PmaReader *pReader; +#if SQLITE_MAX_WORKER_THREADS>0 + if( pSorter->bUseThreads ){ + pReader = pSorter->pReader; + }else +#endif + /*if( !pSorter->bUseThreads )*/{ + pReader = &pSorter->pMerger->aReadr[pSorter->pMerger->aTree[1]]; + } + *pnKey = pReader->nKey; + pKey = pReader->aKey; + }else{ + *pnKey = pSorter->list.pList->nVal; + pKey = SRVAL(pSorter->list.pList); + } + return pKey; +} + +/* +** Copy the current sorter key into the memory cell pOut. +*/ +SQLITE_PRIVATE int sqlite3VdbeSorterRowkey(const VdbeCursor *pCsr, Mem *pOut){ + VdbeSorter *pSorter; + void *pKey; int nKey; /* Sorter key to copy into pOut */ + + assert( pCsr->eCurType==CURTYPE_SORTER ); + pSorter = pCsr->uc.pSorter; + pKey = vdbeSorterRowkey(pSorter, &nKey); + if( sqlite3VdbeMemClearAndResize(pOut, nKey) ){ + return SQLITE_NOMEM_BKPT; + } + pOut->n = nKey; + MemSetTypeFlag(pOut, MEM_Blob); + memcpy(pOut->z, pKey, nKey); + + return SQLITE_OK; +} + +/* +** Compare the key in memory cell pVal with the key that the sorter cursor +** passed as the first argument currently points to. For the purposes of +** the comparison, ignore the rowid field at the end of each record. +** +** If the sorter cursor key contains any NULL values, consider it to be +** less than pVal. Even if pVal also contains NULL values. +** +** If an error occurs, return an SQLite error code (i.e. SQLITE_NOMEM). +** Otherwise, set *pRes to a negative, zero or positive value if the +** key in pVal is smaller than, equal to or larger than the current sorter +** key. +** +** This routine forms the core of the OP_SorterCompare opcode, which in +** turn is used to verify uniqueness when constructing a UNIQUE INDEX. +*/ +SQLITE_PRIVATE int sqlite3VdbeSorterCompare( + const VdbeCursor *pCsr, /* Sorter cursor */ + Mem *pVal, /* Value to compare to current sorter key */ + int nKeyCol, /* Compare this many columns */ + int *pRes /* OUT: Result of comparison */ +){ + VdbeSorter *pSorter; + UnpackedRecord *r2; + KeyInfo *pKeyInfo; + int i; + void *pKey; int nKey; /* Sorter key to compare pVal with */ + + assert( pCsr->eCurType==CURTYPE_SORTER ); + pSorter = pCsr->uc.pSorter; + r2 = pSorter->pUnpacked; + pKeyInfo = pCsr->pKeyInfo; + if( r2==0 ){ + r2 = pSorter->pUnpacked = sqlite3VdbeAllocUnpackedRecord(pKeyInfo); + if( r2==0 ) return SQLITE_NOMEM_BKPT; + r2->nField = nKeyCol; + } + assert( r2->nField==nKeyCol ); + + pKey = vdbeSorterRowkey(pSorter, &nKey); + sqlite3VdbeRecordUnpack(pKeyInfo, nKey, pKey, r2); + for(i=0; iaMem[i].flags & MEM_Null ){ + *pRes = -1; + return SQLITE_OK; + } + } + + *pRes = sqlite3VdbeRecordCompare(pVal->n, pVal->z, r2); + return SQLITE_OK; +} + +/************** End of vdbesort.c ********************************************/ +/************** Begin file memjournal.c **************************************/ +/* +** 2008 October 7 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** +** This file contains code use to implement an in-memory rollback journal. +** The in-memory rollback journal is used to journal transactions for +** ":memory:" databases and when the journal_mode=MEMORY pragma is used. +** +** Update: The in-memory journal is also used to temporarily cache +** smaller journals that are not critical for power-loss recovery. +** For example, statement journals that are not too big will be held +** entirely in memory, thus reducing the number of file I/O calls, and +** more importantly, reducing temporary file creation events. If these +** journals become too large for memory, they are spilled to disk. But +** in the common case, they are usually small and no file I/O needs to +** occur. +*/ +/* #include "sqliteInt.h" */ + +/* Forward references to internal structures */ +typedef struct MemJournal MemJournal; +typedef struct FilePoint FilePoint; +typedef struct FileChunk FileChunk; + +/* +** The rollback journal is composed of a linked list of these structures. +** +** The zChunk array is always at least 8 bytes in size - usually much more. +** Its actual size is stored in the MemJournal.nChunkSize variable. +*/ +struct FileChunk { + FileChunk *pNext; /* Next chunk in the journal */ + u8 zChunk[8]; /* Content of this chunk */ +}; + +/* +** By default, allocate this many bytes of memory for each FileChunk object. +*/ +#define MEMJOURNAL_DFLT_FILECHUNKSIZE 1024 + +/* +** For chunk size nChunkSize, return the number of bytes that should +** be allocated for each FileChunk structure. +*/ +#define fileChunkSize(nChunkSize) (sizeof(FileChunk) + ((nChunkSize)-8)) + +/* +** An instance of this object serves as a cursor into the rollback journal. +** The cursor can be either for reading or writing. +*/ +struct FilePoint { + sqlite3_int64 iOffset; /* Offset from the beginning of the file */ + FileChunk *pChunk; /* Specific chunk into which cursor points */ +}; + +/* +** This structure is a subclass of sqlite3_file. Each open memory-journal +** is an instance of this class. +*/ +struct MemJournal { + const sqlite3_io_methods *pMethod; /* Parent class. MUST BE FIRST */ + int nChunkSize; /* In-memory chunk-size */ + + int nSpill; /* Bytes of data before flushing */ + int nSize; /* Bytes of data currently in memory */ + FileChunk *pFirst; /* Head of in-memory chunk-list */ + FilePoint endpoint; /* Pointer to the end of the file */ + FilePoint readpoint; /* Pointer to the end of the last xRead() */ + + int flags; /* xOpen flags */ + sqlite3_vfs *pVfs; /* The "real" underlying VFS */ + const char *zJournal; /* Name of the journal file */ +}; + +/* +** Read data from the in-memory journal file. This is the implementation +** of the sqlite3_vfs.xRead method. +*/ +static int memjrnlRead( + sqlite3_file *pJfd, /* The journal file from which to read */ + void *zBuf, /* Put the results here */ + int iAmt, /* Number of bytes to read */ + sqlite_int64 iOfst /* Begin reading at this offset */ +){ + MemJournal *p = (MemJournal *)pJfd; + u8 *zOut = zBuf; + int nRead = iAmt; + int iChunkOffset; + FileChunk *pChunk; + + if( (iAmt+iOfst)>p->endpoint.iOffset ){ + return SQLITE_IOERR_SHORT_READ; + } + assert( p->readpoint.iOffset==0 || p->readpoint.pChunk!=0 ); + if( p->readpoint.iOffset!=iOfst || iOfst==0 ){ + sqlite3_int64 iOff = 0; + for(pChunk=p->pFirst; + ALWAYS(pChunk) && (iOff+p->nChunkSize)<=iOfst; + pChunk=pChunk->pNext + ){ + iOff += p->nChunkSize; + } + }else{ + pChunk = p->readpoint.pChunk; + assert( pChunk!=0 ); + } + + iChunkOffset = (int)(iOfst%p->nChunkSize); + do { + int iSpace = p->nChunkSize - iChunkOffset; + int nCopy = MIN(nRead, (p->nChunkSize - iChunkOffset)); + memcpy(zOut, (u8*)pChunk->zChunk + iChunkOffset, nCopy); + zOut += nCopy; + nRead -= iSpace; + iChunkOffset = 0; + } while( nRead>=0 && (pChunk=pChunk->pNext)!=0 && nRead>0 ); + p->readpoint.iOffset = pChunk ? iOfst+iAmt : 0; + p->readpoint.pChunk = pChunk; + + return SQLITE_OK; +} + +/* +** Free the list of FileChunk structures headed at MemJournal.pFirst. +*/ +static void memjrnlFreeChunks(MemJournal *p){ + FileChunk *pIter; + FileChunk *pNext; + for(pIter=p->pFirst; pIter; pIter=pNext){ + pNext = pIter->pNext; + sqlite3_free(pIter); + } + p->pFirst = 0; +} + +/* +** Flush the contents of memory to a real file on disk. +*/ +static int memjrnlCreateFile(MemJournal *p){ + int rc; + sqlite3_file *pReal = (sqlite3_file*)p; + MemJournal copy = *p; + + memset(p, 0, sizeof(MemJournal)); + rc = sqlite3OsOpen(copy.pVfs, copy.zJournal, pReal, copy.flags, 0); + if( rc==SQLITE_OK ){ + int nChunk = copy.nChunkSize; + i64 iOff = 0; + FileChunk *pIter; + for(pIter=copy.pFirst; pIter; pIter=pIter->pNext){ + if( iOff + nChunk > copy.endpoint.iOffset ){ + nChunk = copy.endpoint.iOffset - iOff; + } + rc = sqlite3OsWrite(pReal, (u8*)pIter->zChunk, nChunk, iOff); + if( rc ) break; + iOff += nChunk; + } + if( rc==SQLITE_OK ){ + /* No error has occurred. Free the in-memory buffers. */ + memjrnlFreeChunks(©); + } + } + if( rc!=SQLITE_OK ){ + /* If an error occurred while creating or writing to the file, restore + ** the original before returning. This way, SQLite uses the in-memory + ** journal data to roll back changes made to the internal page-cache + ** before this function was called. */ + sqlite3OsClose(pReal); + *p = copy; + } + return rc; +} + + +/* +** Write data to the file. +*/ +static int memjrnlWrite( + sqlite3_file *pJfd, /* The journal file into which to write */ + const void *zBuf, /* Take data to be written from here */ + int iAmt, /* Number of bytes to write */ + sqlite_int64 iOfst /* Begin writing at this offset into the file */ +){ + MemJournal *p = (MemJournal *)pJfd; + int nWrite = iAmt; + u8 *zWrite = (u8 *)zBuf; + + /* If the file should be created now, create it and write the new data + ** into the file on disk. */ + if( p->nSpill>0 && (iAmt+iOfst)>p->nSpill ){ + int rc = memjrnlCreateFile(p); + if( rc==SQLITE_OK ){ + rc = sqlite3OsWrite(pJfd, zBuf, iAmt, iOfst); + } + return rc; + } + + /* If the contents of this write should be stored in memory */ + else{ + /* An in-memory journal file should only ever be appended to. Random + ** access writes are not required. The only exception to this is when + ** the in-memory journal is being used by a connection using the + ** atomic-write optimization. In this case the first 28 bytes of the + ** journal file may be written as part of committing the transaction. */ + assert( iOfst==p->endpoint.iOffset || iOfst==0 ); +#if defined(SQLITE_ENABLE_ATOMIC_WRITE) \ + || defined(SQLITE_ENABLE_BATCH_ATOMIC_WRITE) + if( iOfst==0 && p->pFirst ){ + assert( p->nChunkSize>iAmt ); + memcpy((u8*)p->pFirst->zChunk, zBuf, iAmt); + }else +#else + assert( iOfst>0 || p->pFirst==0 ); +#endif + { + while( nWrite>0 ){ + FileChunk *pChunk = p->endpoint.pChunk; + int iChunkOffset = (int)(p->endpoint.iOffset%p->nChunkSize); + int iSpace = MIN(nWrite, p->nChunkSize - iChunkOffset); + + if( iChunkOffset==0 ){ + /* New chunk is required to extend the file. */ + FileChunk *pNew = sqlite3_malloc(fileChunkSize(p->nChunkSize)); + if( !pNew ){ + return SQLITE_IOERR_NOMEM_BKPT; + } + pNew->pNext = 0; + if( pChunk ){ + assert( p->pFirst ); + pChunk->pNext = pNew; + }else{ + assert( !p->pFirst ); + p->pFirst = pNew; + } + p->endpoint.pChunk = pNew; + } + + memcpy((u8*)p->endpoint.pChunk->zChunk + iChunkOffset, zWrite, iSpace); + zWrite += iSpace; + nWrite -= iSpace; + p->endpoint.iOffset += iSpace; + } + p->nSize = iAmt + iOfst; + } + } + + return SQLITE_OK; +} + +/* +** Truncate the file. +** +** If the journal file is already on disk, truncate it there. Or, if it +** is still in main memory but is being truncated to zero bytes in size, +** ignore +*/ +static int memjrnlTruncate(sqlite3_file *pJfd, sqlite_int64 size){ + MemJournal *p = (MemJournal *)pJfd; + if( ALWAYS(size==0) ){ + memjrnlFreeChunks(p); + p->nSize = 0; + p->endpoint.pChunk = 0; + p->endpoint.iOffset = 0; + p->readpoint.pChunk = 0; + p->readpoint.iOffset = 0; + } + return SQLITE_OK; +} + +/* +** Close the file. +*/ +static int memjrnlClose(sqlite3_file *pJfd){ + MemJournal *p = (MemJournal *)pJfd; + memjrnlFreeChunks(p); + return SQLITE_OK; +} + +/* +** Sync the file. +** +** If the real file has been created, call its xSync method. Otherwise, +** syncing an in-memory journal is a no-op. +*/ +static int memjrnlSync(sqlite3_file *pJfd, int flags){ + UNUSED_PARAMETER2(pJfd, flags); + return SQLITE_OK; +} + +/* +** Query the size of the file in bytes. +*/ +static int memjrnlFileSize(sqlite3_file *pJfd, sqlite_int64 *pSize){ + MemJournal *p = (MemJournal *)pJfd; + *pSize = (sqlite_int64) p->endpoint.iOffset; + return SQLITE_OK; +} + +/* +** Table of methods for MemJournal sqlite3_file object. +*/ +static const struct sqlite3_io_methods MemJournalMethods = { + 1, /* iVersion */ + memjrnlClose, /* xClose */ + memjrnlRead, /* xRead */ + memjrnlWrite, /* xWrite */ + memjrnlTruncate, /* xTruncate */ + memjrnlSync, /* xSync */ + memjrnlFileSize, /* xFileSize */ + 0, /* xLock */ + 0, /* xUnlock */ + 0, /* xCheckReservedLock */ + 0, /* xFileControl */ + 0, /* xSectorSize */ + 0, /* xDeviceCharacteristics */ + 0, /* xShmMap */ + 0, /* xShmLock */ + 0, /* xShmBarrier */ + 0, /* xShmUnmap */ + 0, /* xFetch */ + 0 /* xUnfetch */ +}; + +/* +** Open a journal file. +** +** The behaviour of the journal file depends on the value of parameter +** nSpill. If nSpill is 0, then the journal file is always create and +** accessed using the underlying VFS. If nSpill is less than zero, then +** all content is always stored in main-memory. Finally, if nSpill is a +** positive value, then the journal file is initially created in-memory +** but may be flushed to disk later on. In this case the journal file is +** flushed to disk either when it grows larger than nSpill bytes in size, +** or when sqlite3JournalCreate() is called. +*/ +SQLITE_PRIVATE int sqlite3JournalOpen( + sqlite3_vfs *pVfs, /* The VFS to use for actual file I/O */ + const char *zName, /* Name of the journal file */ + sqlite3_file *pJfd, /* Preallocated, blank file handle */ + int flags, /* Opening flags */ + int nSpill /* Bytes buffered before opening the file */ +){ + MemJournal *p = (MemJournal*)pJfd; + + /* Zero the file-handle object. If nSpill was passed zero, initialize + ** it using the sqlite3OsOpen() function of the underlying VFS. In this + ** case none of the code in this module is executed as a result of calls + ** made on the journal file-handle. */ + memset(p, 0, sizeof(MemJournal)); + if( nSpill==0 ){ + return sqlite3OsOpen(pVfs, zName, pJfd, flags, 0); + } + + if( nSpill>0 ){ + p->nChunkSize = nSpill; + }else{ + p->nChunkSize = 8 + MEMJOURNAL_DFLT_FILECHUNKSIZE - sizeof(FileChunk); + assert( MEMJOURNAL_DFLT_FILECHUNKSIZE==fileChunkSize(p->nChunkSize) ); + } + + p->pMethod = (const sqlite3_io_methods*)&MemJournalMethods; + p->nSpill = nSpill; + p->flags = flags; + p->zJournal = zName; + p->pVfs = pVfs; + return SQLITE_OK; +} + +/* +** Open an in-memory journal file. +*/ +SQLITE_PRIVATE void sqlite3MemJournalOpen(sqlite3_file *pJfd){ + sqlite3JournalOpen(0, 0, pJfd, 0, -1); +} + +#if defined(SQLITE_ENABLE_ATOMIC_WRITE) \ + || defined(SQLITE_ENABLE_BATCH_ATOMIC_WRITE) +/* +** If the argument p points to a MemJournal structure that is not an +** in-memory-only journal file (i.e. is one that was opened with a +ve +** nSpill parameter or as SQLITE_OPEN_MAIN_JOURNAL), and the underlying +** file has not yet been created, create it now. +*/ +SQLITE_PRIVATE int sqlite3JournalCreate(sqlite3_file *pJfd){ + int rc = SQLITE_OK; + MemJournal *p = (MemJournal*)pJfd; + if( p->pMethod==&MemJournalMethods && ( +#ifdef SQLITE_ENABLE_ATOMIC_WRITE + p->nSpill>0 +#else + /* While this appears to not be possible without ATOMIC_WRITE, the + ** paths are complex, so it seems prudent to leave the test in as + ** a NEVER(), in case our analysis is subtly flawed. */ + NEVER(p->nSpill>0) +#endif +#ifdef SQLITE_ENABLE_BATCH_ATOMIC_WRITE + || (p->flags & SQLITE_OPEN_MAIN_JOURNAL) +#endif + )){ + rc = memjrnlCreateFile(p); + } + return rc; +} +#endif + +/* +** The file-handle passed as the only argument is open on a journal file. +** Return true if this "journal file" is currently stored in heap memory, +** or false otherwise. +*/ +SQLITE_PRIVATE int sqlite3JournalIsInMemory(sqlite3_file *p){ + return p->pMethods==&MemJournalMethods; +} + +/* +** Return the number of bytes required to store a JournalFile that uses vfs +** pVfs to create the underlying on-disk files. +*/ +SQLITE_PRIVATE int sqlite3JournalSize(sqlite3_vfs *pVfs){ + return MAX(pVfs->szOsFile, (int)sizeof(MemJournal)); +} + +/************** End of memjournal.c ******************************************/ +/************** Begin file walker.c ******************************************/ +/* +** 2008 August 16 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** This file contains routines used for walking the parser tree for +** an SQL statement. +*/ +/* #include "sqliteInt.h" */ +/* #include */ +/* #include */ + + +#if !defined(SQLITE_OMIT_WINDOWFUNC) +/* +** Walk all expressions linked into the list of Window objects passed +** as the second argument. +*/ +static int walkWindowList(Walker *pWalker, Window *pList){ + Window *pWin; + for(pWin=pList; pWin; pWin=pWin->pNextWin){ + int rc; + rc = sqlite3WalkExprList(pWalker, pWin->pOrderBy); + if( rc ) return WRC_Abort; + rc = sqlite3WalkExprList(pWalker, pWin->pPartition); + if( rc ) return WRC_Abort; + rc = sqlite3WalkExpr(pWalker, pWin->pFilter); + if( rc ) return WRC_Abort; + + /* The next two are purely for calls to sqlite3RenameExprUnmap() + ** within sqlite3WindowOffsetExpr(). Because of constraints imposed + ** by sqlite3WindowOffsetExpr(), they can never fail. The results do + ** not matter anyhow. */ + rc = sqlite3WalkExpr(pWalker, pWin->pStart); + if( NEVER(rc) ) return WRC_Abort; + rc = sqlite3WalkExpr(pWalker, pWin->pEnd); + if( NEVER(rc) ) return WRC_Abort; + } + return WRC_Continue; +} +#endif + +/* +** Walk an expression tree. Invoke the callback once for each node +** of the expression, while descending. (In other words, the callback +** is invoked before visiting children.) +** +** The return value from the callback should be one of the WRC_* +** constants to specify how to proceed with the walk. +** +** WRC_Continue Continue descending down the tree. +** +** WRC_Prune Do not descend into child nodes, but allow +** the walk to continue with sibling nodes. +** +** WRC_Abort Do no more callbacks. Unwind the stack and +** return from the top-level walk call. +** +** The return value from this routine is WRC_Abort to abandon the tree walk +** and WRC_Continue to continue. +*/ +static SQLITE_NOINLINE int walkExpr(Walker *pWalker, Expr *pExpr){ + int rc; + testcase( ExprHasProperty(pExpr, EP_TokenOnly) ); + testcase( ExprHasProperty(pExpr, EP_Reduced) ); + while(1){ + rc = pWalker->xExprCallback(pWalker, pExpr); + if( rc ) return rc & WRC_Abort; + if( !ExprHasProperty(pExpr,(EP_TokenOnly|EP_Leaf)) ){ + if( pExpr->pLeft && walkExpr(pWalker, pExpr->pLeft) ) return WRC_Abort; + assert( pExpr->x.pList==0 || pExpr->pRight==0 ); + if( pExpr->pRight ){ + assert( !ExprHasProperty(pExpr, EP_WinFunc) ); + pExpr = pExpr->pRight; + continue; + }else if( ExprHasProperty(pExpr, EP_xIsSelect) ){ + assert( !ExprHasProperty(pExpr, EP_WinFunc) ); + if( sqlite3WalkSelect(pWalker, pExpr->x.pSelect) ) return WRC_Abort; + }else{ + if( pExpr->x.pList ){ + if( sqlite3WalkExprList(pWalker, pExpr->x.pList) ) return WRC_Abort; + } +#ifndef SQLITE_OMIT_WINDOWFUNC + if( ExprHasProperty(pExpr, EP_WinFunc) ){ + if( walkWindowList(pWalker, pExpr->y.pWin) ) return WRC_Abort; + } +#endif + } + } + break; + } + return WRC_Continue; +} +SQLITE_PRIVATE int sqlite3WalkExpr(Walker *pWalker, Expr *pExpr){ + return pExpr ? walkExpr(pWalker,pExpr) : WRC_Continue; +} + +/* +** Call sqlite3WalkExpr() for every expression in list p or until +** an abort request is seen. +*/ +SQLITE_PRIVATE int sqlite3WalkExprList(Walker *pWalker, ExprList *p){ + int i; + struct ExprList_item *pItem; + if( p ){ + for(i=p->nExpr, pItem=p->a; i>0; i--, pItem++){ + if( sqlite3WalkExpr(pWalker, pItem->pExpr) ) return WRC_Abort; + } + } + return WRC_Continue; +} + +/* +** Walk all expressions associated with SELECT statement p. Do +** not invoke the SELECT callback on p, but do (of course) invoke +** any expr callbacks and SELECT callbacks that come from subqueries. +** Return WRC_Abort or WRC_Continue. +*/ +SQLITE_PRIVATE int sqlite3WalkSelectExpr(Walker *pWalker, Select *p){ + if( sqlite3WalkExprList(pWalker, p->pEList) ) return WRC_Abort; + if( sqlite3WalkExpr(pWalker, p->pWhere) ) return WRC_Abort; + if( sqlite3WalkExprList(pWalker, p->pGroupBy) ) return WRC_Abort; + if( sqlite3WalkExpr(pWalker, p->pHaving) ) return WRC_Abort; + if( sqlite3WalkExprList(pWalker, p->pOrderBy) ) return WRC_Abort; + if( sqlite3WalkExpr(pWalker, p->pLimit) ) return WRC_Abort; +#if !defined(SQLITE_OMIT_WINDOWFUNC) && !defined(SQLITE_OMIT_ALTERTABLE) + { + Parse *pParse = pWalker->pParse; + if( pParse && IN_RENAME_OBJECT ){ + /* The following may return WRC_Abort if there are unresolvable + ** symbols (e.g. a table that does not exist) in a window definition. */ + int rc = walkWindowList(pWalker, p->pWinDefn); + return rc; + } + } +#endif + return WRC_Continue; +} + +/* +** Walk the parse trees associated with all subqueries in the +** FROM clause of SELECT statement p. Do not invoke the select +** callback on p, but do invoke it on each FROM clause subquery +** and on any subqueries further down in the tree. Return +** WRC_Abort or WRC_Continue; +*/ +SQLITE_PRIVATE int sqlite3WalkSelectFrom(Walker *pWalker, Select *p){ + SrcList *pSrc; + int i; + struct SrcList_item *pItem; + + pSrc = p->pSrc; + assert( pSrc!=0 ); + for(i=pSrc->nSrc, pItem=pSrc->a; i>0; i--, pItem++){ + if( pItem->pSelect && sqlite3WalkSelect(pWalker, pItem->pSelect) ){ + return WRC_Abort; + } + if( pItem->fg.isTabFunc + && sqlite3WalkExprList(pWalker, pItem->u1.pFuncArg) + ){ + return WRC_Abort; + } + } + return WRC_Continue; +} + +/* +** Call sqlite3WalkExpr() for every expression in Select statement p. +** Invoke sqlite3WalkSelect() for subqueries in the FROM clause and +** on the compound select chain, p->pPrior. +** +** If it is not NULL, the xSelectCallback() callback is invoked before +** the walk of the expressions and FROM clause. The xSelectCallback2() +** method is invoked following the walk of the expressions and FROM clause, +** but only if both xSelectCallback and xSelectCallback2 are both non-NULL +** and if the expressions and FROM clause both return WRC_Continue; +** +** Return WRC_Continue under normal conditions. Return WRC_Abort if +** there is an abort request. +** +** If the Walker does not have an xSelectCallback() then this routine +** is a no-op returning WRC_Continue. +*/ +SQLITE_PRIVATE int sqlite3WalkSelect(Walker *pWalker, Select *p){ + int rc; + if( p==0 ) return WRC_Continue; + if( pWalker->xSelectCallback==0 ) return WRC_Continue; + do{ + rc = pWalker->xSelectCallback(pWalker, p); + if( rc ) return rc & WRC_Abort; + if( sqlite3WalkSelectExpr(pWalker, p) + || sqlite3WalkSelectFrom(pWalker, p) + ){ + return WRC_Abort; + } + if( pWalker->xSelectCallback2 ){ + pWalker->xSelectCallback2(pWalker, p); + } + p = p->pPrior; + }while( p!=0 ); + return WRC_Continue; +} + +/************** End of walker.c **********************************************/ +/************** Begin file resolve.c *****************************************/ +/* +** 2008 August 18 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** +** This file contains routines used for walking the parser tree and +** resolve all identifiers by associating them with a particular +** table and column. +*/ +/* #include "sqliteInt.h" */ + +/* +** Walk the expression tree pExpr and increase the aggregate function +** depth (the Expr.op2 field) by N on every TK_AGG_FUNCTION node. +** This needs to occur when copying a TK_AGG_FUNCTION node from an +** outer query into an inner subquery. +** +** incrAggFunctionDepth(pExpr,n) is the main routine. incrAggDepth(..) +** is a helper function - a callback for the tree walker. +*/ +static int incrAggDepth(Walker *pWalker, Expr *pExpr){ + if( pExpr->op==TK_AGG_FUNCTION ) pExpr->op2 += pWalker->u.n; + return WRC_Continue; +} +static void incrAggFunctionDepth(Expr *pExpr, int N){ + if( N>0 ){ + Walker w; + memset(&w, 0, sizeof(w)); + w.xExprCallback = incrAggDepth; + w.u.n = N; + sqlite3WalkExpr(&w, pExpr); + } +} + +/* +** Turn the pExpr expression into an alias for the iCol-th column of the +** result set in pEList. +** +** If the reference is followed by a COLLATE operator, then make sure +** the COLLATE operator is preserved. For example: +** +** SELECT a+b, c+d FROM t1 ORDER BY 1 COLLATE nocase; +** +** Should be transformed into: +** +** SELECT a+b, c+d FROM t1 ORDER BY (a+b) COLLATE nocase; +** +** The nSubquery parameter specifies how many levels of subquery the +** alias is removed from the original expression. The usual value is +** zero but it might be more if the alias is contained within a subquery +** of the original expression. The Expr.op2 field of TK_AGG_FUNCTION +** structures must be increased by the nSubquery amount. +*/ +static void resolveAlias( + Parse *pParse, /* Parsing context */ + ExprList *pEList, /* A result set */ + int iCol, /* A column in the result set. 0..pEList->nExpr-1 */ + Expr *pExpr, /* Transform this into an alias to the result set */ + const char *zType, /* "GROUP" or "ORDER" or "" */ + int nSubquery /* Number of subqueries that the label is moving */ +){ + Expr *pOrig; /* The iCol-th column of the result set */ + Expr *pDup; /* Copy of pOrig */ + sqlite3 *db; /* The database connection */ + + assert( iCol>=0 && iColnExpr ); + pOrig = pEList->a[iCol].pExpr; + assert( pOrig!=0 ); + db = pParse->db; + pDup = sqlite3ExprDup(db, pOrig, 0); + if( pDup!=0 ){ + if( zType[0]!='G' ) incrAggFunctionDepth(pDup, nSubquery); + if( pExpr->op==TK_COLLATE ){ + pDup = sqlite3ExprAddCollateString(pParse, pDup, pExpr->u.zToken); + } + + /* Before calling sqlite3ExprDelete(), set the EP_Static flag. This + ** prevents ExprDelete() from deleting the Expr structure itself, + ** allowing it to be repopulated by the memcpy() on the following line. + ** The pExpr->u.zToken might point into memory that will be freed by the + ** sqlite3DbFree(db, pDup) on the last line of this block, so be sure to + ** make a copy of the token before doing the sqlite3DbFree(). + */ + ExprSetProperty(pExpr, EP_Static); + sqlite3ExprDelete(db, pExpr); + memcpy(pExpr, pDup, sizeof(*pExpr)); + if( !ExprHasProperty(pExpr, EP_IntValue) && pExpr->u.zToken!=0 ){ + assert( (pExpr->flags & (EP_Reduced|EP_TokenOnly))==0 ); + pExpr->u.zToken = sqlite3DbStrDup(db, pExpr->u.zToken); + pExpr->flags |= EP_MemToken; + } + if( ExprHasProperty(pExpr, EP_WinFunc) ){ + if( pExpr->y.pWin!=0 ){ + pExpr->y.pWin->pOwner = pExpr; + }else{ + assert( db->mallocFailed ); + } + } + sqlite3DbFree(db, pDup); + } + ExprSetProperty(pExpr, EP_Alias); +} + + +/* +** Return TRUE if the name zCol occurs anywhere in the USING clause. +** +** Return FALSE if the USING clause is NULL or if it does not contain +** zCol. +*/ +static int nameInUsingClause(IdList *pUsing, const char *zCol){ + if( pUsing ){ + int k; + for(k=0; knId; k++){ + if( sqlite3StrICmp(pUsing->a[k].zName, zCol)==0 ) return 1; + } + } + return 0; +} + +/* +** Subqueries stores the original database, table and column names for their +** result sets in ExprList.a[].zSpan, in the form "DATABASE.TABLE.COLUMN". +** Check to see if the zSpan given to this routine matches the zDb, zTab, +** and zCol. If any of zDb, zTab, and zCol are NULL then those fields will +** match anything. +*/ +SQLITE_PRIVATE int sqlite3MatchSpanName( + const char *zSpan, + const char *zCol, + const char *zTab, + const char *zDb +){ + int n; + for(n=0; ALWAYS(zSpan[n]) && zSpan[n]!='.'; n++){} + if( zDb && (sqlite3StrNICmp(zSpan, zDb, n)!=0 || zDb[n]!=0) ){ + return 0; + } + zSpan += n+1; + for(n=0; ALWAYS(zSpan[n]) && zSpan[n]!='.'; n++){} + if( zTab && (sqlite3StrNICmp(zSpan, zTab, n)!=0 || zTab[n]!=0) ){ + return 0; + } + zSpan += n+1; + if( zCol && sqlite3StrICmp(zSpan, zCol)!=0 ){ + return 0; + } + return 1; +} + +/* +** Return TRUE if the double-quoted string mis-feature should be supported. +*/ +static int areDoubleQuotedStringsEnabled(sqlite3 *db, NameContext *pTopNC){ + if( db->init.busy ) return 1; /* Always support for legacy schemas */ + if( pTopNC->ncFlags & NC_IsDDL ){ + /* Currently parsing a DDL statement */ + if( sqlite3WritableSchema(db) && (db->flags & SQLITE_DqsDML)!=0 ){ + return 1; + } + return (db->flags & SQLITE_DqsDDL)!=0; + }else{ + /* Currently parsing a DML statement */ + return (db->flags & SQLITE_DqsDML)!=0; + } +} + +/* +** Given the name of a column of the form X.Y.Z or Y.Z or just Z, look up +** that name in the set of source tables in pSrcList and make the pExpr +** expression node refer back to that source column. The following changes +** are made to pExpr: +** +** pExpr->iDb Set the index in db->aDb[] of the database X +** (even if X is implied). +** pExpr->iTable Set to the cursor number for the table obtained +** from pSrcList. +** pExpr->y.pTab Points to the Table structure of X.Y (even if +** X and/or Y are implied.) +** pExpr->iColumn Set to the column number within the table. +** pExpr->op Set to TK_COLUMN. +** pExpr->pLeft Any expression this points to is deleted +** pExpr->pRight Any expression this points to is deleted. +** +** The zDb variable is the name of the database (the "X"). This value may be +** NULL meaning that name is of the form Y.Z or Z. Any available database +** can be used. The zTable variable is the name of the table (the "Y"). This +** value can be NULL if zDb is also NULL. If zTable is NULL it +** means that the form of the name is Z and that columns from any table +** can be used. +** +** If the name cannot be resolved unambiguously, leave an error message +** in pParse and return WRC_Abort. Return WRC_Prune on success. +*/ +static int lookupName( + Parse *pParse, /* The parsing context */ + const char *zDb, /* Name of the database containing table, or NULL */ + const char *zTab, /* Name of table containing column, or NULL */ + const char *zCol, /* Name of the column. */ + NameContext *pNC, /* The name context used to resolve the name */ + Expr *pExpr /* Make this EXPR node point to the selected column */ +){ + int i, j; /* Loop counters */ + int cnt = 0; /* Number of matching column names */ + int cntTab = 0; /* Number of matching table names */ + int nSubquery = 0; /* How many levels of subquery */ + sqlite3 *db = pParse->db; /* The database connection */ + struct SrcList_item *pItem; /* Use for looping over pSrcList items */ + struct SrcList_item *pMatch = 0; /* The matching pSrcList item */ + NameContext *pTopNC = pNC; /* First namecontext in the list */ + Schema *pSchema = 0; /* Schema of the expression */ + int eNewExprOp = TK_COLUMN; /* New value for pExpr->op on success */ + Table *pTab = 0; /* Table hold the row */ + Column *pCol; /* A column of pTab */ + + assert( pNC ); /* the name context cannot be NULL. */ + assert( zCol ); /* The Z in X.Y.Z cannot be NULL */ + assert( !ExprHasProperty(pExpr, EP_TokenOnly|EP_Reduced) ); + + /* Initialize the node to no-match */ + pExpr->iTable = -1; + ExprSetVVAProperty(pExpr, EP_NoReduce); + + /* Translate the schema name in zDb into a pointer to the corresponding + ** schema. If not found, pSchema will remain NULL and nothing will match + ** resulting in an appropriate error message toward the end of this routine + */ + if( zDb ){ + testcase( pNC->ncFlags & NC_PartIdx ); + testcase( pNC->ncFlags & NC_IsCheck ); + if( (pNC->ncFlags & (NC_PartIdx|NC_IsCheck))!=0 ){ + /* Silently ignore database qualifiers inside CHECK constraints and + ** partial indices. Do not raise errors because that might break + ** legacy and because it does not hurt anything to just ignore the + ** database name. */ + zDb = 0; + }else{ + for(i=0; inDb; i++){ + assert( db->aDb[i].zDbSName ); + if( sqlite3StrICmp(db->aDb[i].zDbSName,zDb)==0 ){ + pSchema = db->aDb[i].pSchema; + break; + } + } + } + } + + /* Start at the inner-most context and move outward until a match is found */ + assert( pNC && cnt==0 ); + do{ + ExprList *pEList; + SrcList *pSrcList = pNC->pSrcList; + + if( pSrcList ){ + for(i=0, pItem=pSrcList->a; inSrc; i++, pItem++){ + pTab = pItem->pTab; + assert( pTab!=0 && pTab->zName!=0 ); + assert( pTab->nCol>0 ); + if( pItem->pSelect && (pItem->pSelect->selFlags & SF_NestedFrom)!=0 ){ + int hit = 0; + pEList = pItem->pSelect->pEList; + for(j=0; jnExpr; j++){ + if( sqlite3MatchSpanName(pEList->a[j].zSpan, zCol, zTab, zDb) ){ + cnt++; + cntTab = 2; + pMatch = pItem; + pExpr->iColumn = j; + hit = 1; + } + } + if( hit || zTab==0 ) continue; + } + if( zDb && pTab->pSchema!=pSchema ){ + continue; + } + if( zTab ){ + const char *zTabName = pItem->zAlias ? pItem->zAlias : pTab->zName; + assert( zTabName!=0 ); + if( sqlite3StrICmp(zTabName, zTab)!=0 ){ + continue; + } + if( IN_RENAME_OBJECT && pItem->zAlias ){ + sqlite3RenameTokenRemap(pParse, 0, (void*)&pExpr->y.pTab); + } + } + if( 0==(cntTab++) ){ + pMatch = pItem; + } + for(j=0, pCol=pTab->aCol; jnCol; j++, pCol++){ + if( sqlite3StrICmp(pCol->zName, zCol)==0 ){ + /* If there has been exactly one prior match and this match + ** is for the right-hand table of a NATURAL JOIN or is in a + ** USING clause, then skip this match. + */ + if( cnt==1 ){ + if( pItem->fg.jointype & JT_NATURAL ) continue; + if( nameInUsingClause(pItem->pUsing, zCol) ) continue; + } + cnt++; + pMatch = pItem; + /* Substitute the rowid (column -1) for the INTEGER PRIMARY KEY */ + pExpr->iColumn = j==pTab->iPKey ? -1 : (i16)j; + break; + } + } + } + if( pMatch ){ + pExpr->iTable = pMatch->iCursor; + pExpr->y.pTab = pMatch->pTab; + /* RIGHT JOIN not (yet) supported */ + assert( (pMatch->fg.jointype & JT_RIGHT)==0 ); + if( (pMatch->fg.jointype & JT_LEFT)!=0 ){ + ExprSetProperty(pExpr, EP_CanBeNull); + } + pSchema = pExpr->y.pTab->pSchema; + } + } /* if( pSrcList ) */ + +#if !defined(SQLITE_OMIT_TRIGGER) || !defined(SQLITE_OMIT_UPSERT) + /* If we have not already resolved the name, then maybe + ** it is a new.* or old.* trigger argument reference. Or + ** maybe it is an excluded.* from an upsert. + */ + if( zDb==0 && zTab!=0 && cntTab==0 ){ + pTab = 0; +#ifndef SQLITE_OMIT_TRIGGER + if( pParse->pTriggerTab!=0 ){ + int op = pParse->eTriggerOp; + assert( op==TK_DELETE || op==TK_UPDATE || op==TK_INSERT ); + if( op!=TK_DELETE && sqlite3StrICmp("new",zTab) == 0 ){ + pExpr->iTable = 1; + pTab = pParse->pTriggerTab; + }else if( op!=TK_INSERT && sqlite3StrICmp("old",zTab)==0 ){ + pExpr->iTable = 0; + pTab = pParse->pTriggerTab; + } + } +#endif /* SQLITE_OMIT_TRIGGER */ +#ifndef SQLITE_OMIT_UPSERT + if( (pNC->ncFlags & NC_UUpsert)!=0 ){ + Upsert *pUpsert = pNC->uNC.pUpsert; + if( pUpsert && sqlite3StrICmp("excluded",zTab)==0 ){ + pTab = pUpsert->pUpsertSrc->a[0].pTab; + pExpr->iTable = 2; + } + } +#endif /* SQLITE_OMIT_UPSERT */ + + if( pTab ){ + int iCol; + pSchema = pTab->pSchema; + cntTab++; + for(iCol=0, pCol=pTab->aCol; iColnCol; iCol++, pCol++){ + if( sqlite3StrICmp(pCol->zName, zCol)==0 ){ + if( iCol==pTab->iPKey ){ + iCol = -1; + } + break; + } + } + if( iCol>=pTab->nCol && sqlite3IsRowid(zCol) && VisibleRowid(pTab) ){ + /* IMP: R-51414-32910 */ + iCol = -1; + } + if( iColnCol ){ + cnt++; +#ifndef SQLITE_OMIT_UPSERT + if( pExpr->iTable==2 ){ + testcase( iCol==(-1) ); + if( IN_RENAME_OBJECT ){ + pExpr->iColumn = iCol; + pExpr->y.pTab = pTab; + eNewExprOp = TK_COLUMN; + }else{ + pExpr->iTable = pNC->uNC.pUpsert->regData + iCol; + eNewExprOp = TK_REGISTER; + ExprSetProperty(pExpr, EP_Alias); + } + }else +#endif /* SQLITE_OMIT_UPSERT */ + { +#ifndef SQLITE_OMIT_TRIGGER + if( iCol<0 ){ + pExpr->affExpr = SQLITE_AFF_INTEGER; + }else if( pExpr->iTable==0 ){ + testcase( iCol==31 ); + testcase( iCol==32 ); + pParse->oldmask |= (iCol>=32 ? 0xffffffff : (((u32)1)<newmask |= (iCol>=32 ? 0xffffffff : (((u32)1)<y.pTab = pTab; + pExpr->iColumn = (i16)iCol; + eNewExprOp = TK_TRIGGER; +#endif /* SQLITE_OMIT_TRIGGER */ + } + } + } + } +#endif /* !defined(SQLITE_OMIT_TRIGGER) || !defined(SQLITE_OMIT_UPSERT) */ + + /* + ** Perhaps the name is a reference to the ROWID + */ + if( cnt==0 + && cntTab==1 + && pMatch + && (pNC->ncFlags & NC_IdxExpr)==0 + && sqlite3IsRowid(zCol) + && VisibleRowid(pMatch->pTab) + ){ + cnt = 1; + pExpr->iColumn = -1; + pExpr->affExpr = SQLITE_AFF_INTEGER; + } + + /* + ** If the input is of the form Z (not Y.Z or X.Y.Z) then the name Z + ** might refer to an result-set alias. This happens, for example, when + ** we are resolving names in the WHERE clause of the following command: + ** + ** SELECT a+b AS x FROM table WHERE x<10; + ** + ** In cases like this, replace pExpr with a copy of the expression that + ** forms the result set entry ("a+b" in the example) and return immediately. + ** Note that the expression in the result set should have already been + ** resolved by the time the WHERE clause is resolved. + ** + ** The ability to use an output result-set column in the WHERE, GROUP BY, + ** or HAVING clauses, or as part of a larger expression in the ORDER BY + ** clause is not standard SQL. This is a (goofy) SQLite extension, that + ** is supported for backwards compatibility only. Hence, we issue a warning + ** on sqlite3_log() whenever the capability is used. + */ + if( (pNC->ncFlags & NC_UEList)!=0 + && cnt==0 + && zTab==0 + ){ + pEList = pNC->uNC.pEList; + assert( pEList!=0 ); + for(j=0; jnExpr; j++){ + char *zAs = pEList->a[j].zName; + if( zAs!=0 && sqlite3StrICmp(zAs, zCol)==0 ){ + Expr *pOrig; + assert( pExpr->pLeft==0 && pExpr->pRight==0 ); + assert( pExpr->x.pList==0 ); + assert( pExpr->x.pSelect==0 ); + pOrig = pEList->a[j].pExpr; + if( (pNC->ncFlags&NC_AllowAgg)==0 && ExprHasProperty(pOrig, EP_Agg) ){ + sqlite3ErrorMsg(pParse, "misuse of aliased aggregate %s", zAs); + return WRC_Abort; + } + if( (pNC->ncFlags&NC_AllowWin)==0 && ExprHasProperty(pOrig, EP_Win) ){ + sqlite3ErrorMsg(pParse, "misuse of aliased window function %s",zAs); + return WRC_Abort; + } + if( sqlite3ExprVectorSize(pOrig)!=1 ){ + sqlite3ErrorMsg(pParse, "row value misused"); + return WRC_Abort; + } + resolveAlias(pParse, pEList, j, pExpr, "", nSubquery); + cnt = 1; + pMatch = 0; + assert( zTab==0 && zDb==0 ); + if( IN_RENAME_OBJECT ){ + sqlite3RenameTokenRemap(pParse, 0, (void*)pExpr); + } + goto lookupname_end; + } + } + } + + /* Advance to the next name context. The loop will exit when either + ** we have a match (cnt>0) or when we run out of name contexts. + */ + if( cnt ) break; + pNC = pNC->pNext; + nSubquery++; + }while( pNC ); + + + /* + ** If X and Y are NULL (in other words if only the column name Z is + ** supplied) and the value of Z is enclosed in double-quotes, then + ** Z is a string literal if it doesn't match any column names. In that + ** case, we need to return right away and not make any changes to + ** pExpr. + ** + ** Because no reference was made to outer contexts, the pNC->nRef + ** fields are not changed in any context. + */ + if( cnt==0 && zTab==0 ){ + assert( pExpr->op==TK_ID ); + if( ExprHasProperty(pExpr,EP_DblQuoted) + && areDoubleQuotedStringsEnabled(db, pTopNC) + ){ + /* If a double-quoted identifier does not match any known column name, + ** then treat it as a string. + ** + ** This hack was added in the early days of SQLite in a misguided attempt + ** to be compatible with MySQL 3.x, which used double-quotes for strings. + ** I now sorely regret putting in this hack. The effect of this hack is + ** that misspelled identifier names are silently converted into strings + ** rather than causing an error, to the frustration of countless + ** programmers. To all those frustrated programmers, my apologies. + ** + ** Someday, I hope to get rid of this hack. Unfortunately there is + ** a huge amount of legacy SQL that uses it. So for now, we just + ** issue a warning. + */ + sqlite3_log(SQLITE_WARNING, + "double-quoted string literal: \"%w\"", zCol); +#ifdef SQLITE_ENABLE_NORMALIZE + sqlite3VdbeAddDblquoteStr(db, pParse->pVdbe, zCol); +#endif + pExpr->op = TK_STRING; + pExpr->y.pTab = 0; + return WRC_Prune; + } + if( sqlite3ExprIdToTrueFalse(pExpr) ){ + return WRC_Prune; + } + } + + /* + ** cnt==0 means there was not match. cnt>1 means there were two or + ** more matches. Either way, we have an error. + */ + if( cnt!=1 ){ + const char *zErr; + zErr = cnt==0 ? "no such column" : "ambiguous column name"; + if( zDb ){ + sqlite3ErrorMsg(pParse, "%s: %s.%s.%s", zErr, zDb, zTab, zCol); + }else if( zTab ){ + sqlite3ErrorMsg(pParse, "%s: %s.%s", zErr, zTab, zCol); + }else{ + sqlite3ErrorMsg(pParse, "%s: %s", zErr, zCol); + } + pParse->checkSchema = 1; + pTopNC->nErr++; + } + + /* If a column from a table in pSrcList is referenced, then record + ** this fact in the pSrcList.a[].colUsed bitmask. Column 0 causes + ** bit 0 to be set. Column 1 sets bit 1. And so forth. If the + ** column number is greater than the number of bits in the bitmask + ** then set the high-order bit of the bitmask. + */ + if( pExpr->iColumn>=0 && pMatch!=0 ){ + int n = pExpr->iColumn; + testcase( n==BMS-1 ); + if( n>=BMS ){ + n = BMS-1; + } + assert( pMatch->iCursor==pExpr->iTable ); + pMatch->colUsed |= ((Bitmask)1)<pLeft); + pExpr->pLeft = 0; + sqlite3ExprDelete(db, pExpr->pRight); + pExpr->pRight = 0; + pExpr->op = eNewExprOp; + ExprSetProperty(pExpr, EP_Leaf); +lookupname_end: + if( cnt==1 ){ + assert( pNC!=0 ); + if( !ExprHasProperty(pExpr, EP_Alias) ){ + sqlite3AuthRead(pParse, pExpr, pSchema, pNC->pSrcList); + } + /* Increment the nRef value on all name contexts from TopNC up to + ** the point where the name matched. */ + for(;;){ + assert( pTopNC!=0 ); + pTopNC->nRef++; + if( pTopNC==pNC ) break; + pTopNC = pTopNC->pNext; + } + return WRC_Prune; + } else { + return WRC_Abort; + } +} + +/* +** Allocate and return a pointer to an expression to load the column iCol +** from datasource iSrc in SrcList pSrc. +*/ +SQLITE_PRIVATE Expr *sqlite3CreateColumnExpr(sqlite3 *db, SrcList *pSrc, int iSrc, int iCol){ + Expr *p = sqlite3ExprAlloc(db, TK_COLUMN, 0, 0); + if( p ){ + struct SrcList_item *pItem = &pSrc->a[iSrc]; + p->y.pTab = pItem->pTab; + p->iTable = pItem->iCursor; + if( p->y.pTab->iPKey==iCol ){ + p->iColumn = -1; + }else{ + p->iColumn = (ynVar)iCol; + testcase( iCol==BMS ); + testcase( iCol==BMS-1 ); + pItem->colUsed |= ((Bitmask)1)<<(iCol>=BMS ? BMS-1 : iCol); + } + } + return p; +} + +/* +** Report an error that an expression is not valid for some set of +** pNC->ncFlags values determined by validMask. +*/ +static void notValid( + Parse *pParse, /* Leave error message here */ + NameContext *pNC, /* The name context */ + const char *zMsg, /* Type of error */ + int validMask /* Set of contexts for which prohibited */ +){ + assert( (validMask&~(NC_IsCheck|NC_PartIdx|NC_IdxExpr))==0 ); + if( (pNC->ncFlags & validMask)!=0 ){ + const char *zIn = "partial index WHERE clauses"; + if( pNC->ncFlags & NC_IdxExpr ) zIn = "index expressions"; +#ifndef SQLITE_OMIT_CHECK + else if( pNC->ncFlags & NC_IsCheck ) zIn = "CHECK constraints"; +#endif + sqlite3ErrorMsg(pParse, "%s prohibited in %s", zMsg, zIn); + } +} + +/* +** Expression p should encode a floating point value between 1.0 and 0.0. +** Return 1024 times this value. Or return -1 if p is not a floating point +** value between 1.0 and 0.0. +*/ +static int exprProbability(Expr *p){ + double r = -1.0; + if( p->op!=TK_FLOAT ) return -1; + sqlite3AtoF(p->u.zToken, &r, sqlite3Strlen30(p->u.zToken), SQLITE_UTF8); + assert( r>=0.0 ); + if( r>1.0 ) return -1; + return (int)(r*134217728.0); +} + +/* +** This routine is callback for sqlite3WalkExpr(). +** +** Resolve symbolic names into TK_COLUMN operators for the current +** node in the expression tree. Return 0 to continue the search down +** the tree or 2 to abort the tree walk. +** +** This routine also does error checking and name resolution for +** function names. The operator for aggregate functions is changed +** to TK_AGG_FUNCTION. +*/ +static int resolveExprStep(Walker *pWalker, Expr *pExpr){ + NameContext *pNC; + Parse *pParse; + + pNC = pWalker->u.pNC; + assert( pNC!=0 ); + pParse = pNC->pParse; + assert( pParse==pWalker->pParse ); + +#ifndef NDEBUG + if( pNC->pSrcList && pNC->pSrcList->nAlloc>0 ){ + SrcList *pSrcList = pNC->pSrcList; + int i; + for(i=0; ipSrcList->nSrc; i++){ + assert( pSrcList->a[i].iCursor>=0 && pSrcList->a[i].iCursornTab); + } + } +#endif + switch( pExpr->op ){ + +#if defined(SQLITE_ENABLE_UPDATE_DELETE_LIMIT) && !defined(SQLITE_OMIT_SUBQUERY) + /* The special operator TK_ROW means use the rowid for the first + ** column in the FROM clause. This is used by the LIMIT and ORDER BY + ** clause processing on UPDATE and DELETE statements. + */ + case TK_ROW: { + SrcList *pSrcList = pNC->pSrcList; + struct SrcList_item *pItem; + assert( pSrcList && pSrcList->nSrc==1 ); + pItem = pSrcList->a; + assert( HasRowid(pItem->pTab) && pItem->pTab->pSelect==0 ); + pExpr->op = TK_COLUMN; + pExpr->y.pTab = pItem->pTab; + pExpr->iTable = pItem->iCursor; + pExpr->iColumn = -1; + pExpr->affExpr = SQLITE_AFF_INTEGER; + break; + } +#endif /* defined(SQLITE_ENABLE_UPDATE_DELETE_LIMIT) + && !defined(SQLITE_OMIT_SUBQUERY) */ + + /* A column name: ID + ** Or table name and column name: ID.ID + ** Or a database, table and column: ID.ID.ID + ** + ** The TK_ID and TK_OUT cases are combined so that there will only + ** be one call to lookupName(). Then the compiler will in-line + ** lookupName() for a size reduction and performance increase. + */ + case TK_ID: + case TK_DOT: { + const char *zColumn; + const char *zTable; + const char *zDb; + Expr *pRight; + + if( pExpr->op==TK_ID ){ + zDb = 0; + zTable = 0; + zColumn = pExpr->u.zToken; + }else{ + Expr *pLeft = pExpr->pLeft; + notValid(pParse, pNC, "the \".\" operator", NC_IdxExpr); + pRight = pExpr->pRight; + if( pRight->op==TK_ID ){ + zDb = 0; + }else{ + assert( pRight->op==TK_DOT ); + zDb = pLeft->u.zToken; + pLeft = pRight->pLeft; + pRight = pRight->pRight; + } + zTable = pLeft->u.zToken; + zColumn = pRight->u.zToken; + if( IN_RENAME_OBJECT ){ + sqlite3RenameTokenRemap(pParse, (void*)pExpr, (void*)pRight); + sqlite3RenameTokenRemap(pParse, (void*)&pExpr->y.pTab, (void*)pLeft); + } + } + return lookupName(pParse, zDb, zTable, zColumn, pNC, pExpr); + } + + /* Resolve function names + */ + case TK_FUNCTION: { + ExprList *pList = pExpr->x.pList; /* The argument list */ + int n = pList ? pList->nExpr : 0; /* Number of arguments */ + int no_such_func = 0; /* True if no such function exists */ + int wrong_num_args = 0; /* True if wrong number of arguments */ + int is_agg = 0; /* True if is an aggregate function */ + int nId; /* Number of characters in function name */ + const char *zId; /* The function name. */ + FuncDef *pDef; /* Information about the function */ + u8 enc = ENC(pParse->db); /* The database encoding */ + int savedAllowFlags = (pNC->ncFlags & (NC_AllowAgg | NC_AllowWin)); +#ifndef SQLITE_OMIT_WINDOWFUNC + Window *pWin = (IsWindowFunc(pExpr) ? pExpr->y.pWin : 0); +#endif + assert( !ExprHasProperty(pExpr, EP_xIsSelect) ); + zId = pExpr->u.zToken; + nId = sqlite3Strlen30(zId); + pDef = sqlite3FindFunction(pParse->db, zId, n, enc, 0); + if( pDef==0 ){ + pDef = sqlite3FindFunction(pParse->db, zId, -2, enc, 0); + if( pDef==0 ){ + no_such_func = 1; + }else{ + wrong_num_args = 1; + } + }else{ + is_agg = pDef->xFinalize!=0; + if( pDef->funcFlags & SQLITE_FUNC_UNLIKELY ){ + ExprSetProperty(pExpr, EP_Unlikely); + if( n==2 ){ + pExpr->iTable = exprProbability(pList->a[1].pExpr); + if( pExpr->iTable<0 ){ + sqlite3ErrorMsg(pParse, + "second argument to likelihood() must be a " + "constant between 0.0 and 1.0"); + pNC->nErr++; + } + }else{ + /* EVIDENCE-OF: R-61304-29449 The unlikely(X) function is + ** equivalent to likelihood(X, 0.0625). + ** EVIDENCE-OF: R-01283-11636 The unlikely(X) function is + ** short-hand for likelihood(X,0.0625). + ** EVIDENCE-OF: R-36850-34127 The likely(X) function is short-hand + ** for likelihood(X,0.9375). + ** EVIDENCE-OF: R-53436-40973 The likely(X) function is equivalent + ** to likelihood(X,0.9375). */ + /* TUNING: unlikely() probability is 0.0625. likely() is 0.9375 */ + pExpr->iTable = pDef->zName[0]=='u' ? 8388608 : 125829120; + } + } +#ifndef SQLITE_OMIT_AUTHORIZATION + { + int auth = sqlite3AuthCheck(pParse, SQLITE_FUNCTION, 0,pDef->zName,0); + if( auth!=SQLITE_OK ){ + if( auth==SQLITE_DENY ){ + sqlite3ErrorMsg(pParse, "not authorized to use function: %s", + pDef->zName); + pNC->nErr++; + } + pExpr->op = TK_NULL; + return WRC_Prune; + } + } +#endif + if( pDef->funcFlags & (SQLITE_FUNC_CONSTANT|SQLITE_FUNC_SLOCHNG) ){ + /* For the purposes of the EP_ConstFunc flag, date and time + ** functions and other functions that change slowly are considered + ** constant because they are constant for the duration of one query */ + ExprSetProperty(pExpr,EP_ConstFunc); + } + if( (pDef->funcFlags & SQLITE_FUNC_CONSTANT)==0 ){ + /* Date/time functions that use 'now', and other functions like + ** sqlite_version() that might change over time cannot be used + ** in an index. */ + notValid(pParse, pNC, "non-deterministic functions", + NC_IdxExpr|NC_PartIdx); + } + if( (pDef->funcFlags & SQLITE_FUNC_INTERNAL)!=0 + && pParse->nested==0 + && sqlite3Config.bInternalFunctions==0 + ){ + /* Internal-use-only functions are disallowed unless the + ** SQL is being compiled using sqlite3NestedParse() */ + no_such_func = 1; + pDef = 0; + }else + if( (pDef->funcFlags & SQLITE_FUNC_DIRECT)!=0 + && ExprHasProperty(pExpr, EP_Indirect) + && !IN_RENAME_OBJECT + ){ + /* Functions tagged with SQLITE_DIRECTONLY may not be used + ** inside of triggers and views */ + sqlite3ErrorMsg(pParse, "%s() prohibited in triggers and views", + pDef->zName); + } + } + + if( 0==IN_RENAME_OBJECT ){ +#ifndef SQLITE_OMIT_WINDOWFUNC + assert( is_agg==0 || (pDef->funcFlags & SQLITE_FUNC_MINMAX) + || (pDef->xValue==0 && pDef->xInverse==0) + || (pDef->xValue && pDef->xInverse && pDef->xSFunc && pDef->xFinalize) + ); + if( pDef && pDef->xValue==0 && pWin ){ + sqlite3ErrorMsg(pParse, + "%.*s() may not be used as a window function", nId, zId + ); + pNC->nErr++; + }else if( + (is_agg && (pNC->ncFlags & NC_AllowAgg)==0) + || (is_agg && (pDef->funcFlags&SQLITE_FUNC_WINDOW) && !pWin) + || (is_agg && pWin && (pNC->ncFlags & NC_AllowWin)==0) + ){ + const char *zType; + if( (pDef->funcFlags & SQLITE_FUNC_WINDOW) || pWin ){ + zType = "window"; + }else{ + zType = "aggregate"; + } + sqlite3ErrorMsg(pParse, "misuse of %s function %.*s()",zType,nId,zId); + pNC->nErr++; + is_agg = 0; + } +#else + if( (is_agg && (pNC->ncFlags & NC_AllowAgg)==0) ){ + sqlite3ErrorMsg(pParse,"misuse of aggregate function %.*s()",nId,zId); + pNC->nErr++; + is_agg = 0; + } +#endif + else if( no_such_func && pParse->db->init.busy==0 +#ifdef SQLITE_ENABLE_UNKNOWN_SQL_FUNCTION + && pParse->explain==0 +#endif + ){ + sqlite3ErrorMsg(pParse, "no such function: %.*s", nId, zId); + pNC->nErr++; + }else if( wrong_num_args ){ + sqlite3ErrorMsg(pParse,"wrong number of arguments to function %.*s()", + nId, zId); + pNC->nErr++; + } +#ifndef SQLITE_OMIT_WINDOWFUNC + else if( is_agg==0 && ExprHasProperty(pExpr, EP_WinFunc) ){ + sqlite3ErrorMsg(pParse, + "FILTER may not be used with non-aggregate %.*s()", + nId, zId + ); + pNC->nErr++; + } +#endif + if( is_agg ){ + /* Window functions may not be arguments of aggregate functions. + ** Or arguments of other window functions. But aggregate functions + ** may be arguments for window functions. */ +#ifndef SQLITE_OMIT_WINDOWFUNC + pNC->ncFlags &= ~(NC_AllowWin | (!pWin ? NC_AllowAgg : 0)); +#else + pNC->ncFlags &= ~NC_AllowAgg; +#endif + } + } +#ifndef SQLITE_OMIT_WINDOWFUNC + else if( ExprHasProperty(pExpr, EP_WinFunc) ){ + is_agg = 1; + } +#endif + sqlite3WalkExprList(pWalker, pList); + if( is_agg ){ +#ifndef SQLITE_OMIT_WINDOWFUNC + if( pWin ){ + Select *pSel = pNC->pWinSelect; + assert( pWin==pExpr->y.pWin ); + if( IN_RENAME_OBJECT==0 ){ + sqlite3WindowUpdate(pParse, pSel->pWinDefn, pWin, pDef); + } + sqlite3WalkExprList(pWalker, pWin->pPartition); + sqlite3WalkExprList(pWalker, pWin->pOrderBy); + sqlite3WalkExpr(pWalker, pWin->pFilter); + sqlite3WindowLink(pSel, pWin); + pNC->ncFlags |= NC_HasWin; + }else +#endif /* SQLITE_OMIT_WINDOWFUNC */ + { + NameContext *pNC2 = pNC; + pExpr->op = TK_AGG_FUNCTION; + pExpr->op2 = 0; +#ifndef SQLITE_OMIT_WINDOWFUNC + if( ExprHasProperty(pExpr, EP_WinFunc) ){ + sqlite3WalkExpr(pWalker, pExpr->y.pWin->pFilter); + } +#endif + while( pNC2 && !sqlite3FunctionUsesThisSrc(pExpr, pNC2->pSrcList) ){ + pExpr->op2++; + pNC2 = pNC2->pNext; + } + assert( pDef!=0 || IN_RENAME_OBJECT ); + if( pNC2 && pDef ){ + assert( SQLITE_FUNC_MINMAX==NC_MinMaxAgg ); + testcase( (pDef->funcFlags & SQLITE_FUNC_MINMAX)!=0 ); + pNC2->ncFlags |= NC_HasAgg | (pDef->funcFlags & SQLITE_FUNC_MINMAX); + + } + } + pNC->ncFlags |= savedAllowFlags; + } + /* FIX ME: Compute pExpr->affinity based on the expected return + ** type of the function + */ + return WRC_Prune; + } +#ifndef SQLITE_OMIT_SUBQUERY + case TK_SELECT: + case TK_EXISTS: testcase( pExpr->op==TK_EXISTS ); +#endif + case TK_IN: { + testcase( pExpr->op==TK_IN ); + if( ExprHasProperty(pExpr, EP_xIsSelect) ){ + int nRef = pNC->nRef; + notValid(pParse, pNC, "subqueries", NC_IsCheck|NC_PartIdx|NC_IdxExpr); + sqlite3WalkSelect(pWalker, pExpr->x.pSelect); + assert( pNC->nRef>=nRef ); + if( nRef!=pNC->nRef ){ + ExprSetProperty(pExpr, EP_VarSelect); + pNC->ncFlags |= NC_VarSelect; + } + } + break; + } + case TK_VARIABLE: { + notValid(pParse, pNC, "parameters", NC_IsCheck|NC_PartIdx|NC_IdxExpr); + break; + } + case TK_IS: + case TK_ISNOT: { + Expr *pRight = sqlite3ExprSkipCollateAndLikely(pExpr->pRight); + assert( !ExprHasProperty(pExpr, EP_Reduced) ); + /* Handle special cases of "x IS TRUE", "x IS FALSE", "x IS NOT TRUE", + ** and "x IS NOT FALSE". */ + if( pRight->op==TK_ID ){ + int rc = resolveExprStep(pWalker, pRight); + if( rc==WRC_Abort ) return WRC_Abort; + if( pRight->op==TK_TRUEFALSE ){ + pExpr->op2 = pExpr->op; + pExpr->op = TK_TRUTH; + return WRC_Continue; + } + } + /* Fall thru */ + } + case TK_BETWEEN: + case TK_EQ: + case TK_NE: + case TK_LT: + case TK_LE: + case TK_GT: + case TK_GE: { + int nLeft, nRight; + if( pParse->db->mallocFailed ) break; + assert( pExpr->pLeft!=0 ); + nLeft = sqlite3ExprVectorSize(pExpr->pLeft); + if( pExpr->op==TK_BETWEEN ){ + nRight = sqlite3ExprVectorSize(pExpr->x.pList->a[0].pExpr); + if( nRight==nLeft ){ + nRight = sqlite3ExprVectorSize(pExpr->x.pList->a[1].pExpr); + } + }else{ + assert( pExpr->pRight!=0 ); + nRight = sqlite3ExprVectorSize(pExpr->pRight); + } + if( nLeft!=nRight ){ + testcase( pExpr->op==TK_EQ ); + testcase( pExpr->op==TK_NE ); + testcase( pExpr->op==TK_LT ); + testcase( pExpr->op==TK_LE ); + testcase( pExpr->op==TK_GT ); + testcase( pExpr->op==TK_GE ); + testcase( pExpr->op==TK_IS ); + testcase( pExpr->op==TK_ISNOT ); + testcase( pExpr->op==TK_BETWEEN ); + sqlite3ErrorMsg(pParse, "row value misused"); + } + break; + } + } + return (pParse->nErr || pParse->db->mallocFailed) ? WRC_Abort : WRC_Continue; +} + +/* +** pEList is a list of expressions which are really the result set of the +** a SELECT statement. pE is a term in an ORDER BY or GROUP BY clause. +** This routine checks to see if pE is a simple identifier which corresponds +** to the AS-name of one of the terms of the expression list. If it is, +** this routine return an integer between 1 and N where N is the number of +** elements in pEList, corresponding to the matching entry. If there is +** no match, or if pE is not a simple identifier, then this routine +** return 0. +** +** pEList has been resolved. pE has not. +*/ +static int resolveAsName( + Parse *pParse, /* Parsing context for error messages */ + ExprList *pEList, /* List of expressions to scan */ + Expr *pE /* Expression we are trying to match */ +){ + int i; /* Loop counter */ + + UNUSED_PARAMETER(pParse); + + if( pE->op==TK_ID ){ + char *zCol = pE->u.zToken; + for(i=0; inExpr; i++){ + char *zAs = pEList->a[i].zName; + if( zAs!=0 && sqlite3StrICmp(zAs, zCol)==0 ){ + return i+1; + } + } + } + return 0; +} + +/* +** pE is a pointer to an expression which is a single term in the +** ORDER BY of a compound SELECT. The expression has not been +** name resolved. +** +** At the point this routine is called, we already know that the +** ORDER BY term is not an integer index into the result set. That +** case is handled by the calling routine. +** +** Attempt to match pE against result set columns in the left-most +** SELECT statement. Return the index i of the matching column, +** as an indication to the caller that it should sort by the i-th column. +** The left-most column is 1. In other words, the value returned is the +** same integer value that would be used in the SQL statement to indicate +** the column. +** +** If there is no match, return 0. Return -1 if an error occurs. +*/ +static int resolveOrderByTermToExprList( + Parse *pParse, /* Parsing context for error messages */ + Select *pSelect, /* The SELECT statement with the ORDER BY clause */ + Expr *pE /* The specific ORDER BY term */ +){ + int i; /* Loop counter */ + ExprList *pEList; /* The columns of the result set */ + NameContext nc; /* Name context for resolving pE */ + sqlite3 *db; /* Database connection */ + int rc; /* Return code from subprocedures */ + u8 savedSuppErr; /* Saved value of db->suppressErr */ + + assert( sqlite3ExprIsInteger(pE, &i)==0 ); + pEList = pSelect->pEList; + + /* Resolve all names in the ORDER BY term expression + */ + memset(&nc, 0, sizeof(nc)); + nc.pParse = pParse; + nc.pSrcList = pSelect->pSrc; + nc.uNC.pEList = pEList; + nc.ncFlags = NC_AllowAgg|NC_UEList; + nc.nErr = 0; + db = pParse->db; + savedSuppErr = db->suppressErr; + db->suppressErr = 1; + rc = sqlite3ResolveExprNames(&nc, pE); + db->suppressErr = savedSuppErr; + if( rc ) return 0; + + /* Try to match the ORDER BY expression against an expression + ** in the result set. Return an 1-based index of the matching + ** result-set entry. + */ + for(i=0; inExpr; i++){ + if( sqlite3ExprCompare(0, pEList->a[i].pExpr, pE, -1)<2 ){ + return i+1; + } + } + + /* If no match, return 0. */ + return 0; +} + +/* +** Generate an ORDER BY or GROUP BY term out-of-range error. +*/ +static void resolveOutOfRangeError( + Parse *pParse, /* The error context into which to write the error */ + const char *zType, /* "ORDER" or "GROUP" */ + int i, /* The index (1-based) of the term out of range */ + int mx /* Largest permissible value of i */ +){ + sqlite3ErrorMsg(pParse, + "%r %s BY term out of range - should be " + "between 1 and %d", i, zType, mx); +} + +/* +** Analyze the ORDER BY clause in a compound SELECT statement. Modify +** each term of the ORDER BY clause is a constant integer between 1 +** and N where N is the number of columns in the compound SELECT. +** +** ORDER BY terms that are already an integer between 1 and N are +** unmodified. ORDER BY terms that are integers outside the range of +** 1 through N generate an error. ORDER BY terms that are expressions +** are matched against result set expressions of compound SELECT +** beginning with the left-most SELECT and working toward the right. +** At the first match, the ORDER BY expression is transformed into +** the integer column number. +** +** Return the number of errors seen. +*/ +static int resolveCompoundOrderBy( + Parse *pParse, /* Parsing context. Leave error messages here */ + Select *pSelect /* The SELECT statement containing the ORDER BY */ +){ + int i; + ExprList *pOrderBy; + ExprList *pEList; + sqlite3 *db; + int moreToDo = 1; + + pOrderBy = pSelect->pOrderBy; + if( pOrderBy==0 ) return 0; + db = pParse->db; + if( pOrderBy->nExpr>db->aLimit[SQLITE_LIMIT_COLUMN] ){ + sqlite3ErrorMsg(pParse, "too many terms in ORDER BY clause"); + return 1; + } + for(i=0; inExpr; i++){ + pOrderBy->a[i].done = 0; + } + pSelect->pNext = 0; + while( pSelect->pPrior ){ + pSelect->pPrior->pNext = pSelect; + pSelect = pSelect->pPrior; + } + while( pSelect && moreToDo ){ + struct ExprList_item *pItem; + moreToDo = 0; + pEList = pSelect->pEList; + assert( pEList!=0 ); + for(i=0, pItem=pOrderBy->a; inExpr; i++, pItem++){ + int iCol = -1; + Expr *pE, *pDup; + if( pItem->done ) continue; + pE = sqlite3ExprSkipCollateAndLikely(pItem->pExpr); + if( sqlite3ExprIsInteger(pE, &iCol) ){ + if( iCol<=0 || iCol>pEList->nExpr ){ + resolveOutOfRangeError(pParse, "ORDER", i+1, pEList->nExpr); + return 1; + } + }else{ + iCol = resolveAsName(pParse, pEList, pE); + if( iCol==0 ){ + /* Now test if expression pE matches one of the values returned + ** by pSelect. In the usual case this is done by duplicating the + ** expression, resolving any symbols in it, and then comparing + ** it against each expression returned by the SELECT statement. + ** Once the comparisons are finished, the duplicate expression + ** is deleted. + ** + ** Or, if this is running as part of an ALTER TABLE operation, + ** resolve the symbols in the actual expression, not a duplicate. + ** And, if one of the comparisons is successful, leave the expression + ** as is instead of transforming it to an integer as in the usual + ** case. This allows the code in alter.c to modify column + ** refererences within the ORDER BY expression as required. */ + if( IN_RENAME_OBJECT ){ + pDup = pE; + }else{ + pDup = sqlite3ExprDup(db, pE, 0); + } + if( !db->mallocFailed ){ + assert(pDup); + iCol = resolveOrderByTermToExprList(pParse, pSelect, pDup); + } + if( !IN_RENAME_OBJECT ){ + sqlite3ExprDelete(db, pDup); + } + } + } + if( iCol>0 ){ + /* Convert the ORDER BY term into an integer column number iCol, + ** taking care to preserve the COLLATE clause if it exists */ + if( !IN_RENAME_OBJECT ){ + Expr *pNew = sqlite3Expr(db, TK_INTEGER, 0); + if( pNew==0 ) return 1; + pNew->flags |= EP_IntValue; + pNew->u.iValue = iCol; + if( pItem->pExpr==pE ){ + pItem->pExpr = pNew; + }else{ + Expr *pParent = pItem->pExpr; + assert( pParent->op==TK_COLLATE ); + while( pParent->pLeft->op==TK_COLLATE ) pParent = pParent->pLeft; + assert( pParent->pLeft==pE ); + pParent->pLeft = pNew; + } + sqlite3ExprDelete(db, pE); + pItem->u.x.iOrderByCol = (u16)iCol; + } + pItem->done = 1; + }else{ + moreToDo = 1; + } + } + pSelect = pSelect->pNext; + } + for(i=0; inExpr; i++){ + if( pOrderBy->a[i].done==0 ){ + sqlite3ErrorMsg(pParse, "%r ORDER BY term does not match any " + "column in the result set", i+1); + return 1; + } + } + return 0; +} + +/* +** Check every term in the ORDER BY or GROUP BY clause pOrderBy of +** the SELECT statement pSelect. If any term is reference to a +** result set expression (as determined by the ExprList.a.u.x.iOrderByCol +** field) then convert that term into a copy of the corresponding result set +** column. +** +** If any errors are detected, add an error message to pParse and +** return non-zero. Return zero if no errors are seen. +*/ +SQLITE_PRIVATE int sqlite3ResolveOrderGroupBy( + Parse *pParse, /* Parsing context. Leave error messages here */ + Select *pSelect, /* The SELECT statement containing the clause */ + ExprList *pOrderBy, /* The ORDER BY or GROUP BY clause to be processed */ + const char *zType /* "ORDER" or "GROUP" */ +){ + int i; + sqlite3 *db = pParse->db; + ExprList *pEList; + struct ExprList_item *pItem; + + if( pOrderBy==0 || pParse->db->mallocFailed || IN_RENAME_OBJECT ) return 0; + if( pOrderBy->nExpr>db->aLimit[SQLITE_LIMIT_COLUMN] ){ + sqlite3ErrorMsg(pParse, "too many terms in %s BY clause", zType); + return 1; + } + pEList = pSelect->pEList; + assert( pEList!=0 ); /* sqlite3SelectNew() guarantees this */ + for(i=0, pItem=pOrderBy->a; inExpr; i++, pItem++){ + if( pItem->u.x.iOrderByCol ){ + if( pItem->u.x.iOrderByCol>pEList->nExpr ){ + resolveOutOfRangeError(pParse, zType, i+1, pEList->nExpr); + return 1; + } + resolveAlias(pParse, pEList, pItem->u.x.iOrderByCol-1, pItem->pExpr, + zType,0); + } + } + return 0; +} + +#ifndef SQLITE_OMIT_WINDOWFUNC +/* +** Walker callback for windowRemoveExprFromSelect(). +*/ +static int resolveRemoveWindowsCb(Walker *pWalker, Expr *pExpr){ + UNUSED_PARAMETER(pWalker); + if( ExprHasProperty(pExpr, EP_WinFunc) ){ + Window *pWin = pExpr->y.pWin; + sqlite3WindowUnlinkFromSelect(pWin); + } + return WRC_Continue; +} + +/* +** Remove any Window objects owned by the expression pExpr from the +** Select.pWin list of Select object pSelect. +*/ +static void windowRemoveExprFromSelect(Select *pSelect, Expr *pExpr){ + if( pSelect->pWin ){ + Walker sWalker; + memset(&sWalker, 0, sizeof(Walker)); + sWalker.xExprCallback = resolveRemoveWindowsCb; + sWalker.u.pSelect = pSelect; + sqlite3WalkExpr(&sWalker, pExpr); + } +} +#else +# define windowRemoveExprFromSelect(a, b) +#endif /* SQLITE_OMIT_WINDOWFUNC */ + +/* +** pOrderBy is an ORDER BY or GROUP BY clause in SELECT statement pSelect. +** The Name context of the SELECT statement is pNC. zType is either +** "ORDER" or "GROUP" depending on which type of clause pOrderBy is. +** +** This routine resolves each term of the clause into an expression. +** If the order-by term is an integer I between 1 and N (where N is the +** number of columns in the result set of the SELECT) then the expression +** in the resolution is a copy of the I-th result-set expression. If +** the order-by term is an identifier that corresponds to the AS-name of +** a result-set expression, then the term resolves to a copy of the +** result-set expression. Otherwise, the expression is resolved in +** the usual way - using sqlite3ResolveExprNames(). +** +** This routine returns the number of errors. If errors occur, then +** an appropriate error message might be left in pParse. (OOM errors +** excepted.) +*/ +static int resolveOrderGroupBy( + NameContext *pNC, /* The name context of the SELECT statement */ + Select *pSelect, /* The SELECT statement holding pOrderBy */ + ExprList *pOrderBy, /* An ORDER BY or GROUP BY clause to resolve */ + const char *zType /* Either "ORDER" or "GROUP", as appropriate */ +){ + int i, j; /* Loop counters */ + int iCol; /* Column number */ + struct ExprList_item *pItem; /* A term of the ORDER BY clause */ + Parse *pParse; /* Parsing context */ + int nResult; /* Number of terms in the result set */ + + if( pOrderBy==0 ) return 0; + nResult = pSelect->pEList->nExpr; + pParse = pNC->pParse; + for(i=0, pItem=pOrderBy->a; inExpr; i++, pItem++){ + Expr *pE = pItem->pExpr; + Expr *pE2 = sqlite3ExprSkipCollateAndLikely(pE); + if( zType[0]!='G' ){ + iCol = resolveAsName(pParse, pSelect->pEList, pE2); + if( iCol>0 ){ + /* If an AS-name match is found, mark this ORDER BY column as being + ** a copy of the iCol-th result-set column. The subsequent call to + ** sqlite3ResolveOrderGroupBy() will convert the expression to a + ** copy of the iCol-th result-set expression. */ + pItem->u.x.iOrderByCol = (u16)iCol; + continue; + } + } + if( sqlite3ExprIsInteger(pE2, &iCol) ){ + /* The ORDER BY term is an integer constant. Again, set the column + ** number so that sqlite3ResolveOrderGroupBy() will convert the + ** order-by term to a copy of the result-set expression */ + if( iCol<1 || iCol>0xffff ){ + resolveOutOfRangeError(pParse, zType, i+1, nResult); + return 1; + } + pItem->u.x.iOrderByCol = (u16)iCol; + continue; + } + + /* Otherwise, treat the ORDER BY term as an ordinary expression */ + pItem->u.x.iOrderByCol = 0; + if( sqlite3ResolveExprNames(pNC, pE) ){ + return 1; + } + for(j=0; jpEList->nExpr; j++){ + if( sqlite3ExprCompare(0, pE, pSelect->pEList->a[j].pExpr, -1)==0 ){ + /* Since this expresion is being changed into a reference + ** to an identical expression in the result set, remove all Window + ** objects belonging to the expression from the Select.pWin list. */ + windowRemoveExprFromSelect(pSelect, pE); + pItem->u.x.iOrderByCol = j+1; + } + } + } + return sqlite3ResolveOrderGroupBy(pParse, pSelect, pOrderBy, zType); +} + +/* +** Resolve names in the SELECT statement p and all of its descendants. +*/ +static int resolveSelectStep(Walker *pWalker, Select *p){ + NameContext *pOuterNC; /* Context that contains this SELECT */ + NameContext sNC; /* Name context of this SELECT */ + int isCompound; /* True if p is a compound select */ + int nCompound; /* Number of compound terms processed so far */ + Parse *pParse; /* Parsing context */ + int i; /* Loop counter */ + ExprList *pGroupBy; /* The GROUP BY clause */ + Select *pLeftmost; /* Left-most of SELECT of a compound */ + sqlite3 *db; /* Database connection */ + + + assert( p!=0 ); + if( p->selFlags & SF_Resolved ){ + return WRC_Prune; + } + pOuterNC = pWalker->u.pNC; + pParse = pWalker->pParse; + db = pParse->db; + + /* Normally sqlite3SelectExpand() will be called first and will have + ** already expanded this SELECT. However, if this is a subquery within + ** an expression, sqlite3ResolveExprNames() will be called without a + ** prior call to sqlite3SelectExpand(). When that happens, let + ** sqlite3SelectPrep() do all of the processing for this SELECT. + ** sqlite3SelectPrep() will invoke both sqlite3SelectExpand() and + ** this routine in the correct order. + */ + if( (p->selFlags & SF_Expanded)==0 ){ + sqlite3SelectPrep(pParse, p, pOuterNC); + return (pParse->nErr || db->mallocFailed) ? WRC_Abort : WRC_Prune; + } + + isCompound = p->pPrior!=0; + nCompound = 0; + pLeftmost = p; + while( p ){ + assert( (p->selFlags & SF_Expanded)!=0 ); + assert( (p->selFlags & SF_Resolved)==0 ); + p->selFlags |= SF_Resolved; + + /* Resolve the expressions in the LIMIT and OFFSET clauses. These + ** are not allowed to refer to any names, so pass an empty NameContext. + */ + memset(&sNC, 0, sizeof(sNC)); + sNC.pParse = pParse; + sNC.pWinSelect = p; + if( sqlite3ResolveExprNames(&sNC, p->pLimit) ){ + return WRC_Abort; + } + + /* If the SF_Converted flags is set, then this Select object was + ** was created by the convertCompoundSelectToSubquery() function. + ** In this case the ORDER BY clause (p->pOrderBy) should be resolved + ** as if it were part of the sub-query, not the parent. This block + ** moves the pOrderBy down to the sub-query. It will be moved back + ** after the names have been resolved. */ + if( p->selFlags & SF_Converted ){ + Select *pSub = p->pSrc->a[0].pSelect; + assert( p->pSrc->nSrc==1 && p->pOrderBy ); + assert( pSub->pPrior && pSub->pOrderBy==0 ); + pSub->pOrderBy = p->pOrderBy; + p->pOrderBy = 0; + } + + /* Recursively resolve names in all subqueries + */ + for(i=0; ipSrc->nSrc; i++){ + struct SrcList_item *pItem = &p->pSrc->a[i]; + if( pItem->pSelect && (pItem->pSelect->selFlags & SF_Resolved)==0 ){ + NameContext *pNC; /* Used to iterate name contexts */ + int nRef = 0; /* Refcount for pOuterNC and outer contexts */ + const char *zSavedContext = pParse->zAuthContext; + + /* Count the total number of references to pOuterNC and all of its + ** parent contexts. After resolving references to expressions in + ** pItem->pSelect, check if this value has changed. If so, then + ** SELECT statement pItem->pSelect must be correlated. Set the + ** pItem->fg.isCorrelated flag if this is the case. */ + for(pNC=pOuterNC; pNC; pNC=pNC->pNext) nRef += pNC->nRef; + + if( pItem->zName ) pParse->zAuthContext = pItem->zName; + sqlite3ResolveSelectNames(pParse, pItem->pSelect, pOuterNC); + pParse->zAuthContext = zSavedContext; + if( pParse->nErr || db->mallocFailed ) return WRC_Abort; + + for(pNC=pOuterNC; pNC; pNC=pNC->pNext) nRef -= pNC->nRef; + assert( pItem->fg.isCorrelated==0 && nRef<=0 ); + pItem->fg.isCorrelated = (nRef!=0); + } + } + + /* Set up the local name-context to pass to sqlite3ResolveExprNames() to + ** resolve the result-set expression list. + */ + sNC.ncFlags = NC_AllowAgg|NC_AllowWin; + sNC.pSrcList = p->pSrc; + sNC.pNext = pOuterNC; + + /* Resolve names in the result set. */ + if( sqlite3ResolveExprListNames(&sNC, p->pEList) ) return WRC_Abort; + sNC.ncFlags &= ~NC_AllowWin; + + /* If there are no aggregate functions in the result-set, and no GROUP BY + ** expression, do not allow aggregates in any of the other expressions. + */ + assert( (p->selFlags & SF_Aggregate)==0 ); + pGroupBy = p->pGroupBy; + if( pGroupBy || (sNC.ncFlags & NC_HasAgg)!=0 ){ + assert( NC_MinMaxAgg==SF_MinMaxAgg ); + p->selFlags |= SF_Aggregate | (sNC.ncFlags&NC_MinMaxAgg); + }else{ + sNC.ncFlags &= ~NC_AllowAgg; + } + + /* If a HAVING clause is present, then there must be a GROUP BY clause. + */ + if( p->pHaving && !pGroupBy ){ + sqlite3ErrorMsg(pParse, "a GROUP BY clause is required before HAVING"); + return WRC_Abort; + } + + /* Add the output column list to the name-context before parsing the + ** other expressions in the SELECT statement. This is so that + ** expressions in the WHERE clause (etc.) can refer to expressions by + ** aliases in the result set. + ** + ** Minor point: If this is the case, then the expression will be + ** re-evaluated for each reference to it. + */ + assert( (sNC.ncFlags & (NC_UAggInfo|NC_UUpsert))==0 ); + sNC.uNC.pEList = p->pEList; + sNC.ncFlags |= NC_UEList; + if( sqlite3ResolveExprNames(&sNC, p->pHaving) ) return WRC_Abort; + if( sqlite3ResolveExprNames(&sNC, p->pWhere) ) return WRC_Abort; + + /* Resolve names in table-valued-function arguments */ + for(i=0; ipSrc->nSrc; i++){ + struct SrcList_item *pItem = &p->pSrc->a[i]; + if( pItem->fg.isTabFunc + && sqlite3ResolveExprListNames(&sNC, pItem->u1.pFuncArg) + ){ + return WRC_Abort; + } + } + + /* The ORDER BY and GROUP BY clauses may not refer to terms in + ** outer queries + */ + sNC.pNext = 0; + sNC.ncFlags |= NC_AllowAgg|NC_AllowWin; + + /* If this is a converted compound query, move the ORDER BY clause from + ** the sub-query back to the parent query. At this point each term + ** within the ORDER BY clause has been transformed to an integer value. + ** These integers will be replaced by copies of the corresponding result + ** set expressions by the call to resolveOrderGroupBy() below. */ + if( p->selFlags & SF_Converted ){ + Select *pSub = p->pSrc->a[0].pSelect; + p->pOrderBy = pSub->pOrderBy; + pSub->pOrderBy = 0; + } + + /* Process the ORDER BY clause for singleton SELECT statements. + ** The ORDER BY clause for compounds SELECT statements is handled + ** below, after all of the result-sets for all of the elements of + ** the compound have been resolved. + ** + ** If there is an ORDER BY clause on a term of a compound-select other + ** than the right-most term, then that is a syntax error. But the error + ** is not detected until much later, and so we need to go ahead and + ** resolve those symbols on the incorrect ORDER BY for consistency. + */ + if( isCompound<=nCompound /* Defer right-most ORDER BY of a compound */ + && resolveOrderGroupBy(&sNC, p, p->pOrderBy, "ORDER") + ){ + return WRC_Abort; + } + if( db->mallocFailed ){ + return WRC_Abort; + } + sNC.ncFlags &= ~NC_AllowWin; + + /* Resolve the GROUP BY clause. At the same time, make sure + ** the GROUP BY clause does not contain aggregate functions. + */ + if( pGroupBy ){ + struct ExprList_item *pItem; + + if( resolveOrderGroupBy(&sNC, p, pGroupBy, "GROUP") || db->mallocFailed ){ + return WRC_Abort; + } + for(i=0, pItem=pGroupBy->a; inExpr; i++, pItem++){ + if( ExprHasProperty(pItem->pExpr, EP_Agg) ){ + sqlite3ErrorMsg(pParse, "aggregate functions are not allowed in " + "the GROUP BY clause"); + return WRC_Abort; + } + } + } + +#ifndef SQLITE_OMIT_WINDOWFUNC + if( IN_RENAME_OBJECT ){ + Window *pWin; + for(pWin=p->pWinDefn; pWin; pWin=pWin->pNextWin){ + if( sqlite3ResolveExprListNames(&sNC, pWin->pOrderBy) + || sqlite3ResolveExprListNames(&sNC, pWin->pPartition) + ){ + return WRC_Abort; + } + } + } +#endif + + /* If this is part of a compound SELECT, check that it has the right + ** number of expressions in the select list. */ + if( p->pNext && p->pEList->nExpr!=p->pNext->pEList->nExpr ){ + sqlite3SelectWrongNumTermsError(pParse, p->pNext); + return WRC_Abort; + } + + /* Advance to the next term of the compound + */ + p = p->pPrior; + nCompound++; + } + + /* Resolve the ORDER BY on a compound SELECT after all terms of + ** the compound have been resolved. + */ + if( isCompound && resolveCompoundOrderBy(pParse, pLeftmost) ){ + return WRC_Abort; + } + + return WRC_Prune; +} + +/* +** This routine walks an expression tree and resolves references to +** table columns and result-set columns. At the same time, do error +** checking on function usage and set a flag if any aggregate functions +** are seen. +** +** To resolve table columns references we look for nodes (or subtrees) of the +** form X.Y.Z or Y.Z or just Z where +** +** X: The name of a database. Ex: "main" or "temp" or +** the symbolic name assigned to an ATTACH-ed database. +** +** Y: The name of a table in a FROM clause. Or in a trigger +** one of the special names "old" or "new". +** +** Z: The name of a column in table Y. +** +** The node at the root of the subtree is modified as follows: +** +** Expr.op Changed to TK_COLUMN +** Expr.pTab Points to the Table object for X.Y +** Expr.iColumn The column index in X.Y. -1 for the rowid. +** Expr.iTable The VDBE cursor number for X.Y +** +** +** To resolve result-set references, look for expression nodes of the +** form Z (with no X and Y prefix) where the Z matches the right-hand +** size of an AS clause in the result-set of a SELECT. The Z expression +** is replaced by a copy of the left-hand side of the result-set expression. +** Table-name and function resolution occurs on the substituted expression +** tree. For example, in: +** +** SELECT a+b AS x, c+d AS y FROM t1 ORDER BY x; +** +** The "x" term of the order by is replaced by "a+b" to render: +** +** SELECT a+b AS x, c+d AS y FROM t1 ORDER BY a+b; +** +** Function calls are checked to make sure that the function is +** defined and that the correct number of arguments are specified. +** If the function is an aggregate function, then the NC_HasAgg flag is +** set and the opcode is changed from TK_FUNCTION to TK_AGG_FUNCTION. +** If an expression contains aggregate functions then the EP_Agg +** property on the expression is set. +** +** An error message is left in pParse if anything is amiss. The number +** if errors is returned. +*/ +SQLITE_PRIVATE int sqlite3ResolveExprNames( + NameContext *pNC, /* Namespace to resolve expressions in. */ + Expr *pExpr /* The expression to be analyzed. */ +){ + int savedHasAgg; + Walker w; + + if( pExpr==0 ) return SQLITE_OK; + savedHasAgg = pNC->ncFlags & (NC_HasAgg|NC_MinMaxAgg|NC_HasWin); + pNC->ncFlags &= ~(NC_HasAgg|NC_MinMaxAgg|NC_HasWin); + w.pParse = pNC->pParse; + w.xExprCallback = resolveExprStep; + w.xSelectCallback = resolveSelectStep; + w.xSelectCallback2 = 0; + w.u.pNC = pNC; +#if SQLITE_MAX_EXPR_DEPTH>0 + w.pParse->nHeight += pExpr->nHeight; + if( sqlite3ExprCheckHeight(w.pParse, w.pParse->nHeight) ){ + return SQLITE_ERROR; + } +#endif + sqlite3WalkExpr(&w, pExpr); +#if SQLITE_MAX_EXPR_DEPTH>0 + w.pParse->nHeight -= pExpr->nHeight; +#endif + assert( EP_Agg==NC_HasAgg ); + assert( EP_Win==NC_HasWin ); + testcase( pNC->ncFlags & NC_HasAgg ); + testcase( pNC->ncFlags & NC_HasWin ); + ExprSetProperty(pExpr, pNC->ncFlags & (NC_HasAgg|NC_HasWin) ); + pNC->ncFlags |= savedHasAgg; + return pNC->nErr>0 || w.pParse->nErr>0; +} + +/* +** Resolve all names for all expression in an expression list. This is +** just like sqlite3ResolveExprNames() except that it works for an expression +** list rather than a single expression. +*/ +SQLITE_PRIVATE int sqlite3ResolveExprListNames( + NameContext *pNC, /* Namespace to resolve expressions in. */ + ExprList *pList /* The expression list to be analyzed. */ +){ + int i; + if( pList ){ + for(i=0; inExpr; i++){ + if( sqlite3ResolveExprNames(pNC, pList->a[i].pExpr) ) return WRC_Abort; + } + } + return WRC_Continue; +} + +/* +** Resolve all names in all expressions of a SELECT and in all +** decendents of the SELECT, including compounds off of p->pPrior, +** subqueries in expressions, and subqueries used as FROM clause +** terms. +** +** See sqlite3ResolveExprNames() for a description of the kinds of +** transformations that occur. +** +** All SELECT statements should have been expanded using +** sqlite3SelectExpand() prior to invoking this routine. +*/ +SQLITE_PRIVATE void sqlite3ResolveSelectNames( + Parse *pParse, /* The parser context */ + Select *p, /* The SELECT statement being coded. */ + NameContext *pOuterNC /* Name context for parent SELECT statement */ +){ + Walker w; + + assert( p!=0 ); + w.xExprCallback = resolveExprStep; + w.xSelectCallback = resolveSelectStep; + w.xSelectCallback2 = 0; + w.pParse = pParse; + w.u.pNC = pOuterNC; + sqlite3WalkSelect(&w, p); +} + +/* +** Resolve names in expressions that can only reference a single table +** or which cannot reference any tables at all. Examples: +** +** (1) CHECK constraints +** (2) WHERE clauses on partial indices +** (3) Expressions in indexes on expressions +** (4) Expression arguments to VACUUM INTO. +** +** In all cases except (4), the Expr.iTable value for Expr.op==TK_COLUMN +** nodes of the expression is set to -1 and the Expr.iColumn value is +** set to the column number. In case (4), TK_COLUMN nodes cause an error. +** +** Any errors cause an error message to be set in pParse. +*/ +SQLITE_PRIVATE int sqlite3ResolveSelfReference( + Parse *pParse, /* Parsing context */ + Table *pTab, /* The table being referenced, or NULL */ + int type, /* NC_IsCheck or NC_PartIdx or NC_IdxExpr, or 0 */ + Expr *pExpr, /* Expression to resolve. May be NULL. */ + ExprList *pList /* Expression list to resolve. May be NULL. */ +){ + SrcList sSrc; /* Fake SrcList for pParse->pNewTable */ + NameContext sNC; /* Name context for pParse->pNewTable */ + int rc; + + assert( type==0 || pTab!=0 ); + assert( type==NC_IsCheck || type==NC_PartIdx || type==NC_IdxExpr || pTab==0 ); + memset(&sNC, 0, sizeof(sNC)); + memset(&sSrc, 0, sizeof(sSrc)); + if( pTab ){ + sSrc.nSrc = 1; + sSrc.a[0].zName = pTab->zName; + sSrc.a[0].pTab = pTab; + sSrc.a[0].iCursor = -1; + } + sNC.pParse = pParse; + sNC.pSrcList = &sSrc; + sNC.ncFlags = type | NC_IsDDL; + if( (rc = sqlite3ResolveExprNames(&sNC, pExpr))!=SQLITE_OK ) return rc; + if( pList ) rc = sqlite3ResolveExprListNames(&sNC, pList); + return rc; +} + +/************** End of resolve.c *********************************************/ +/************** Begin file expr.c ********************************************/ +/* +** 2001 September 15 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** This file contains routines used for analyzing expressions and +** for generating VDBE code that evaluates expressions in SQLite. +*/ +/* #include "sqliteInt.h" */ + +/* Forward declarations */ +static void exprCodeBetween(Parse*,Expr*,int,void(*)(Parse*,Expr*,int,int),int); +static int exprCodeVector(Parse *pParse, Expr *p, int *piToFree); + +/* +** Return the affinity character for a single column of a table. +*/ +SQLITE_PRIVATE char sqlite3TableColumnAffinity(Table *pTab, int iCol){ + assert( iColnCol ); + return iCol>=0 ? pTab->aCol[iCol].affinity : SQLITE_AFF_INTEGER; +} + +/* +** Return the 'affinity' of the expression pExpr if any. +** +** If pExpr is a column, a reference to a column via an 'AS' alias, +** or a sub-select with a column as the return value, then the +** affinity of that column is returned. Otherwise, 0x00 is returned, +** indicating no affinity for the expression. +** +** i.e. the WHERE clause expressions in the following statements all +** have an affinity: +** +** CREATE TABLE t1(a); +** SELECT * FROM t1 WHERE a; +** SELECT a AS b FROM t1 WHERE b; +** SELECT * FROM t1 WHERE (select a from t1); +*/ +SQLITE_PRIVATE char sqlite3ExprAffinity(Expr *pExpr){ + int op; + while( ExprHasProperty(pExpr, EP_Skip) ){ + assert( pExpr->op==TK_COLLATE ); + pExpr = pExpr->pLeft; + assert( pExpr!=0 ); + } + op = pExpr->op; + if( op==TK_SELECT ){ + assert( pExpr->flags&EP_xIsSelect ); + return sqlite3ExprAffinity(pExpr->x.pSelect->pEList->a[0].pExpr); + } + if( op==TK_REGISTER ) op = pExpr->op2; +#ifndef SQLITE_OMIT_CAST + if( op==TK_CAST ){ + assert( !ExprHasProperty(pExpr, EP_IntValue) ); + return sqlite3AffinityType(pExpr->u.zToken, 0); + } +#endif + if( (op==TK_AGG_COLUMN || op==TK_COLUMN) && pExpr->y.pTab ){ + return sqlite3TableColumnAffinity(pExpr->y.pTab, pExpr->iColumn); + } + if( op==TK_SELECT_COLUMN ){ + assert( pExpr->pLeft->flags&EP_xIsSelect ); + return sqlite3ExprAffinity( + pExpr->pLeft->x.pSelect->pEList->a[pExpr->iColumn].pExpr + ); + } + return pExpr->affExpr; +} + +/* +** Set the collating sequence for expression pExpr to be the collating +** sequence named by pToken. Return a pointer to a new Expr node that +** implements the COLLATE operator. +** +** If a memory allocation error occurs, that fact is recorded in pParse->db +** and the pExpr parameter is returned unchanged. +*/ +SQLITE_PRIVATE Expr *sqlite3ExprAddCollateToken( + Parse *pParse, /* Parsing context */ + Expr *pExpr, /* Add the "COLLATE" clause to this expression */ + const Token *pCollName, /* Name of collating sequence */ + int dequote /* True to dequote pCollName */ +){ + if( pCollName->n>0 ){ + Expr *pNew = sqlite3ExprAlloc(pParse->db, TK_COLLATE, pCollName, dequote); + if( pNew ){ + pNew->pLeft = pExpr; + pNew->flags |= EP_Collate|EP_Skip; + pExpr = pNew; + } + } + return pExpr; +} +SQLITE_PRIVATE Expr *sqlite3ExprAddCollateString(Parse *pParse, Expr *pExpr, const char *zC){ + Token s; + assert( zC!=0 ); + sqlite3TokenInit(&s, (char*)zC); + return sqlite3ExprAddCollateToken(pParse, pExpr, &s, 0); +} + +/* +** Skip over any TK_COLLATE operators. +*/ +SQLITE_PRIVATE Expr *sqlite3ExprSkipCollate(Expr *pExpr){ + while( pExpr && ExprHasProperty(pExpr, EP_Skip) ){ + assert( pExpr->op==TK_COLLATE ); + pExpr = pExpr->pLeft; + } + return pExpr; +} + +/* +** Skip over any TK_COLLATE operators and/or any unlikely() +** or likelihood() or likely() functions at the root of an +** expression. +*/ +SQLITE_PRIVATE Expr *sqlite3ExprSkipCollateAndLikely(Expr *pExpr){ + while( pExpr && ExprHasProperty(pExpr, EP_Skip|EP_Unlikely) ){ + if( ExprHasProperty(pExpr, EP_Unlikely) ){ + assert( !ExprHasProperty(pExpr, EP_xIsSelect) ); + assert( pExpr->x.pList->nExpr>0 ); + assert( pExpr->op==TK_FUNCTION ); + pExpr = pExpr->x.pList->a[0].pExpr; + }else{ + assert( pExpr->op==TK_COLLATE ); + pExpr = pExpr->pLeft; + } + } + return pExpr; +} + +/* +** Return the collation sequence for the expression pExpr. If +** there is no defined collating sequence, return NULL. +** +** See also: sqlite3ExprNNCollSeq() +** +** The sqlite3ExprNNCollSeq() works the same exact that it returns the +** default collation if pExpr has no defined collation. +** +** The collating sequence might be determined by a COLLATE operator +** or by the presence of a column with a defined collating sequence. +** COLLATE operators take first precedence. Left operands take +** precedence over right operands. +*/ +SQLITE_PRIVATE CollSeq *sqlite3ExprCollSeq(Parse *pParse, Expr *pExpr){ + sqlite3 *db = pParse->db; + CollSeq *pColl = 0; + Expr *p = pExpr; + while( p ){ + int op = p->op; + if( op==TK_REGISTER ) op = p->op2; + if( (op==TK_AGG_COLUMN || op==TK_COLUMN || op==TK_TRIGGER) + && p->y.pTab!=0 + ){ + /* op==TK_REGISTER && p->y.pTab!=0 happens when pExpr was originally + ** a TK_COLUMN but was previously evaluated and cached in a register */ + int j = p->iColumn; + if( j>=0 ){ + const char *zColl = p->y.pTab->aCol[j].zColl; + pColl = sqlite3FindCollSeq(db, ENC(db), zColl, 0); + } + break; + } + if( op==TK_CAST || op==TK_UPLUS ){ + p = p->pLeft; + continue; + } + if( op==TK_COLLATE ){ + pColl = sqlite3GetCollSeq(pParse, ENC(db), 0, p->u.zToken); + break; + } + if( p->flags & EP_Collate ){ + if( p->pLeft && (p->pLeft->flags & EP_Collate)!=0 ){ + p = p->pLeft; + }else{ + Expr *pNext = p->pRight; + /* The Expr.x union is never used at the same time as Expr.pRight */ + assert( p->x.pList==0 || p->pRight==0 ); + /* p->flags holds EP_Collate and p->pLeft->flags does not. And + ** p->x.pSelect cannot. So if p->x.pLeft exists, it must hold at + ** least one EP_Collate. Thus the following two ALWAYS. */ + if( p->x.pList!=0 && ALWAYS(!ExprHasProperty(p, EP_xIsSelect)) ){ + int i; + for(i=0; ALWAYS(ix.pList->nExpr); i++){ + if( ExprHasProperty(p->x.pList->a[i].pExpr, EP_Collate) ){ + pNext = p->x.pList->a[i].pExpr; + break; + } + } + } + p = pNext; + } + }else{ + break; + } + } + if( sqlite3CheckCollSeq(pParse, pColl) ){ + pColl = 0; + } + return pColl; +} + +/* +** Return the collation sequence for the expression pExpr. If +** there is no defined collating sequence, return a pointer to the +** defautl collation sequence. +** +** See also: sqlite3ExprCollSeq() +** +** The sqlite3ExprCollSeq() routine works the same except that it +** returns NULL if there is no defined collation. +*/ +SQLITE_PRIVATE CollSeq *sqlite3ExprNNCollSeq(Parse *pParse, Expr *pExpr){ + CollSeq *p = sqlite3ExprCollSeq(pParse, pExpr); + if( p==0 ) p = pParse->db->pDfltColl; + assert( p!=0 ); + return p; +} + +/* +** Return TRUE if the two expressions have equivalent collating sequences. +*/ +SQLITE_PRIVATE int sqlite3ExprCollSeqMatch(Parse *pParse, Expr *pE1, Expr *pE2){ + CollSeq *pColl1 = sqlite3ExprNNCollSeq(pParse, pE1); + CollSeq *pColl2 = sqlite3ExprNNCollSeq(pParse, pE2); + return sqlite3StrICmp(pColl1->zName, pColl2->zName)==0; +} + +/* +** pExpr is an operand of a comparison operator. aff2 is the +** type affinity of the other operand. This routine returns the +** type affinity that should be used for the comparison operator. +*/ +SQLITE_PRIVATE char sqlite3CompareAffinity(Expr *pExpr, char aff2){ + char aff1 = sqlite3ExprAffinity(pExpr); + if( aff1>SQLITE_AFF_NONE && aff2>SQLITE_AFF_NONE ){ + /* Both sides of the comparison are columns. If one has numeric + ** affinity, use that. Otherwise use no affinity. + */ + if( sqlite3IsNumericAffinity(aff1) || sqlite3IsNumericAffinity(aff2) ){ + return SQLITE_AFF_NUMERIC; + }else{ + return SQLITE_AFF_BLOB; + } + }else{ + /* One side is a column, the other is not. Use the columns affinity. */ + assert( aff1<=SQLITE_AFF_NONE || aff2<=SQLITE_AFF_NONE ); + return (aff1<=SQLITE_AFF_NONE ? aff2 : aff1) | SQLITE_AFF_NONE; + } +} + +/* +** pExpr is a comparison operator. Return the type affinity that should +** be applied to both operands prior to doing the comparison. +*/ +static char comparisonAffinity(Expr *pExpr){ + char aff; + assert( pExpr->op==TK_EQ || pExpr->op==TK_IN || pExpr->op==TK_LT || + pExpr->op==TK_GT || pExpr->op==TK_GE || pExpr->op==TK_LE || + pExpr->op==TK_NE || pExpr->op==TK_IS || pExpr->op==TK_ISNOT ); + assert( pExpr->pLeft ); + aff = sqlite3ExprAffinity(pExpr->pLeft); + if( pExpr->pRight ){ + aff = sqlite3CompareAffinity(pExpr->pRight, aff); + }else if( ExprHasProperty(pExpr, EP_xIsSelect) ){ + aff = sqlite3CompareAffinity(pExpr->x.pSelect->pEList->a[0].pExpr, aff); + }else if( aff==0 ){ + aff = SQLITE_AFF_BLOB; + } + return aff; +} + +/* +** pExpr is a comparison expression, eg. '=', '<', IN(...) etc. +** idx_affinity is the affinity of an indexed column. Return true +** if the index with affinity idx_affinity may be used to implement +** the comparison in pExpr. +*/ +SQLITE_PRIVATE int sqlite3IndexAffinityOk(Expr *pExpr, char idx_affinity){ + char aff = comparisonAffinity(pExpr); + if( affflags & EP_Collate ){ + pColl = sqlite3ExprCollSeq(pParse, pLeft); + }else if( pRight && (pRight->flags & EP_Collate)!=0 ){ + pColl = sqlite3ExprCollSeq(pParse, pRight); + }else{ + pColl = sqlite3ExprCollSeq(pParse, pLeft); + if( !pColl ){ + pColl = sqlite3ExprCollSeq(pParse, pRight); + } + } + return pColl; +} + +/* +** Generate code for a comparison operator. +*/ +static int codeCompare( + Parse *pParse, /* The parsing (and code generating) context */ + Expr *pLeft, /* The left operand */ + Expr *pRight, /* The right operand */ + int opcode, /* The comparison opcode */ + int in1, int in2, /* Register holding operands */ + int dest, /* Jump here if true. */ + int jumpIfNull /* If true, jump if either operand is NULL */ +){ + int p5; + int addr; + CollSeq *p4; + + p4 = sqlite3BinaryCompareCollSeq(pParse, pLeft, pRight); + p5 = binaryCompareP5(pLeft, pRight, jumpIfNull); + addr = sqlite3VdbeAddOp4(pParse->pVdbe, opcode, in2, dest, in1, + (void*)p4, P4_COLLSEQ); + sqlite3VdbeChangeP5(pParse->pVdbe, (u8)p5); + return addr; +} + +/* +** Return true if expression pExpr is a vector, or false otherwise. +** +** A vector is defined as any expression that results in two or more +** columns of result. Every TK_VECTOR node is an vector because the +** parser will not generate a TK_VECTOR with fewer than two entries. +** But a TK_SELECT might be either a vector or a scalar. It is only +** considered a vector if it has two or more result columns. +*/ +SQLITE_PRIVATE int sqlite3ExprIsVector(Expr *pExpr){ + return sqlite3ExprVectorSize(pExpr)>1; +} + +/* +** If the expression passed as the only argument is of type TK_VECTOR +** return the number of expressions in the vector. Or, if the expression +** is a sub-select, return the number of columns in the sub-select. For +** any other type of expression, return 1. +*/ +SQLITE_PRIVATE int sqlite3ExprVectorSize(Expr *pExpr){ + u8 op = pExpr->op; + if( op==TK_REGISTER ) op = pExpr->op2; + if( op==TK_VECTOR ){ + return pExpr->x.pList->nExpr; + }else if( op==TK_SELECT ){ + return pExpr->x.pSelect->pEList->nExpr; + }else{ + return 1; + } +} + +/* +** Return a pointer to a subexpression of pVector that is the i-th +** column of the vector (numbered starting with 0). The caller must +** ensure that i is within range. +** +** If pVector is really a scalar (and "scalar" here includes subqueries +** that return a single column!) then return pVector unmodified. +** +** pVector retains ownership of the returned subexpression. +** +** If the vector is a (SELECT ...) then the expression returned is +** just the expression for the i-th term of the result set, and may +** not be ready for evaluation because the table cursor has not yet +** been positioned. +*/ +SQLITE_PRIVATE Expr *sqlite3VectorFieldSubexpr(Expr *pVector, int i){ + assert( iop2==0 || pVector->op==TK_REGISTER ); + if( pVector->op==TK_SELECT || pVector->op2==TK_SELECT ){ + return pVector->x.pSelect->pEList->a[i].pExpr; + }else{ + return pVector->x.pList->a[i].pExpr; + } + } + return pVector; +} + +/* +** Compute and return a new Expr object which when passed to +** sqlite3ExprCode() will generate all necessary code to compute +** the iField-th column of the vector expression pVector. +** +** It is ok for pVector to be a scalar (as long as iField==0). +** In that case, this routine works like sqlite3ExprDup(). +** +** The caller owns the returned Expr object and is responsible for +** ensuring that the returned value eventually gets freed. +** +** The caller retains ownership of pVector. If pVector is a TK_SELECT, +** then the returned object will reference pVector and so pVector must remain +** valid for the life of the returned object. If pVector is a TK_VECTOR +** or a scalar expression, then it can be deleted as soon as this routine +** returns. +** +** A trick to cause a TK_SELECT pVector to be deleted together with +** the returned Expr object is to attach the pVector to the pRight field +** of the returned TK_SELECT_COLUMN Expr object. +*/ +SQLITE_PRIVATE Expr *sqlite3ExprForVectorField( + Parse *pParse, /* Parsing context */ + Expr *pVector, /* The vector. List of expressions or a sub-SELECT */ + int iField /* Which column of the vector to return */ +){ + Expr *pRet; + if( pVector->op==TK_SELECT ){ + assert( pVector->flags & EP_xIsSelect ); + /* The TK_SELECT_COLUMN Expr node: + ** + ** pLeft: pVector containing TK_SELECT. Not deleted. + ** pRight: not used. But recursively deleted. + ** iColumn: Index of a column in pVector + ** iTable: 0 or the number of columns on the LHS of an assignment + ** pLeft->iTable: First in an array of register holding result, or 0 + ** if the result is not yet computed. + ** + ** sqlite3ExprDelete() specifically skips the recursive delete of + ** pLeft on TK_SELECT_COLUMN nodes. But pRight is followed, so pVector + ** can be attached to pRight to cause this node to take ownership of + ** pVector. Typically there will be multiple TK_SELECT_COLUMN nodes + ** with the same pLeft pointer to the pVector, but only one of them + ** will own the pVector. + */ + pRet = sqlite3PExpr(pParse, TK_SELECT_COLUMN, 0, 0); + if( pRet ){ + pRet->iColumn = iField; + pRet->pLeft = pVector; + } + assert( pRet==0 || pRet->iTable==0 ); + }else{ + if( pVector->op==TK_VECTOR ) pVector = pVector->x.pList->a[iField].pExpr; + pRet = sqlite3ExprDup(pParse->db, pVector, 0); + sqlite3RenameTokenRemap(pParse, pRet, pVector); + } + return pRet; +} + +/* +** If expression pExpr is of type TK_SELECT, generate code to evaluate +** it. Return the register in which the result is stored (or, if the +** sub-select returns more than one column, the first in an array +** of registers in which the result is stored). +** +** If pExpr is not a TK_SELECT expression, return 0. +*/ +static int exprCodeSubselect(Parse *pParse, Expr *pExpr){ + int reg = 0; +#ifndef SQLITE_OMIT_SUBQUERY + if( pExpr->op==TK_SELECT ){ + reg = sqlite3CodeSubselect(pParse, pExpr); + } +#endif + return reg; +} + +/* +** Argument pVector points to a vector expression - either a TK_VECTOR +** or TK_SELECT that returns more than one column. This function returns +** the register number of a register that contains the value of +** element iField of the vector. +** +** If pVector is a TK_SELECT expression, then code for it must have +** already been generated using the exprCodeSubselect() routine. In this +** case parameter regSelect should be the first in an array of registers +** containing the results of the sub-select. +** +** If pVector is of type TK_VECTOR, then code for the requested field +** is generated. In this case (*pRegFree) may be set to the number of +** a temporary register to be freed by the caller before returning. +** +** Before returning, output parameter (*ppExpr) is set to point to the +** Expr object corresponding to element iElem of the vector. +*/ +static int exprVectorRegister( + Parse *pParse, /* Parse context */ + Expr *pVector, /* Vector to extract element from */ + int iField, /* Field to extract from pVector */ + int regSelect, /* First in array of registers */ + Expr **ppExpr, /* OUT: Expression element */ + int *pRegFree /* OUT: Temp register to free */ +){ + u8 op = pVector->op; + assert( op==TK_VECTOR || op==TK_REGISTER || op==TK_SELECT ); + if( op==TK_REGISTER ){ + *ppExpr = sqlite3VectorFieldSubexpr(pVector, iField); + return pVector->iTable+iField; + } + if( op==TK_SELECT ){ + *ppExpr = pVector->x.pSelect->pEList->a[iField].pExpr; + return regSelect+iField; + } + *ppExpr = pVector->x.pList->a[iField].pExpr; + return sqlite3ExprCodeTemp(pParse, *ppExpr, pRegFree); +} + +/* +** Expression pExpr is a comparison between two vector values. Compute +** the result of the comparison (1, 0, or NULL) and write that +** result into register dest. +** +** The caller must satisfy the following preconditions: +** +** if pExpr->op==TK_IS: op==TK_EQ and p5==SQLITE_NULLEQ +** if pExpr->op==TK_ISNOT: op==TK_NE and p5==SQLITE_NULLEQ +** otherwise: op==pExpr->op and p5==0 +*/ +static void codeVectorCompare( + Parse *pParse, /* Code generator context */ + Expr *pExpr, /* The comparison operation */ + int dest, /* Write results into this register */ + u8 op, /* Comparison operator */ + u8 p5 /* SQLITE_NULLEQ or zero */ +){ + Vdbe *v = pParse->pVdbe; + Expr *pLeft = pExpr->pLeft; + Expr *pRight = pExpr->pRight; + int nLeft = sqlite3ExprVectorSize(pLeft); + int i; + int regLeft = 0; + int regRight = 0; + u8 opx = op; + int addrDone = sqlite3VdbeMakeLabel(pParse); + + if( nLeft!=sqlite3ExprVectorSize(pRight) ){ + sqlite3ErrorMsg(pParse, "row value misused"); + return; + } + assert( pExpr->op==TK_EQ || pExpr->op==TK_NE + || pExpr->op==TK_IS || pExpr->op==TK_ISNOT + || pExpr->op==TK_LT || pExpr->op==TK_GT + || pExpr->op==TK_LE || pExpr->op==TK_GE + ); + assert( pExpr->op==op || (pExpr->op==TK_IS && op==TK_EQ) + || (pExpr->op==TK_ISNOT && op==TK_NE) ); + assert( p5==0 || pExpr->op!=op ); + assert( p5==SQLITE_NULLEQ || pExpr->op==op ); + + p5 |= SQLITE_STOREP2; + if( opx==TK_LE ) opx = TK_LT; + if( opx==TK_GE ) opx = TK_GT; + + regLeft = exprCodeSubselect(pParse, pLeft); + regRight = exprCodeSubselect(pParse, pRight); + + for(i=0; 1 /*Loop exits by "break"*/; i++){ + int regFree1 = 0, regFree2 = 0; + Expr *pL, *pR; + int r1, r2; + assert( i>=0 && i0 +/* +** Check that argument nHeight is less than or equal to the maximum +** expression depth allowed. If it is not, leave an error message in +** pParse. +*/ +SQLITE_PRIVATE int sqlite3ExprCheckHeight(Parse *pParse, int nHeight){ + int rc = SQLITE_OK; + int mxHeight = pParse->db->aLimit[SQLITE_LIMIT_EXPR_DEPTH]; + if( nHeight>mxHeight ){ + sqlite3ErrorMsg(pParse, + "Expression tree is too large (maximum depth %d)", mxHeight + ); + rc = SQLITE_ERROR; + } + return rc; +} + +/* The following three functions, heightOfExpr(), heightOfExprList() +** and heightOfSelect(), are used to determine the maximum height +** of any expression tree referenced by the structure passed as the +** first argument. +** +** If this maximum height is greater than the current value pointed +** to by pnHeight, the second parameter, then set *pnHeight to that +** value. +*/ +static void heightOfExpr(Expr *p, int *pnHeight){ + if( p ){ + if( p->nHeight>*pnHeight ){ + *pnHeight = p->nHeight; + } + } +} +static void heightOfExprList(ExprList *p, int *pnHeight){ + if( p ){ + int i; + for(i=0; inExpr; i++){ + heightOfExpr(p->a[i].pExpr, pnHeight); + } + } +} +static void heightOfSelect(Select *pSelect, int *pnHeight){ + Select *p; + for(p=pSelect; p; p=p->pPrior){ + heightOfExpr(p->pWhere, pnHeight); + heightOfExpr(p->pHaving, pnHeight); + heightOfExpr(p->pLimit, pnHeight); + heightOfExprList(p->pEList, pnHeight); + heightOfExprList(p->pGroupBy, pnHeight); + heightOfExprList(p->pOrderBy, pnHeight); + } +} + +/* +** Set the Expr.nHeight variable in the structure passed as an +** argument. An expression with no children, Expr.pList or +** Expr.pSelect member has a height of 1. Any other expression +** has a height equal to the maximum height of any other +** referenced Expr plus one. +** +** Also propagate EP_Propagate flags up from Expr.x.pList to Expr.flags, +** if appropriate. +*/ +static void exprSetHeight(Expr *p){ + int nHeight = 0; + heightOfExpr(p->pLeft, &nHeight); + heightOfExpr(p->pRight, &nHeight); + if( ExprHasProperty(p, EP_xIsSelect) ){ + heightOfSelect(p->x.pSelect, &nHeight); + }else if( p->x.pList ){ + heightOfExprList(p->x.pList, &nHeight); + p->flags |= EP_Propagate & sqlite3ExprListFlags(p->x.pList); + } + p->nHeight = nHeight + 1; +} + +/* +** Set the Expr.nHeight variable using the exprSetHeight() function. If +** the height is greater than the maximum allowed expression depth, +** leave an error in pParse. +** +** Also propagate all EP_Propagate flags from the Expr.x.pList into +** Expr.flags. +*/ +SQLITE_PRIVATE void sqlite3ExprSetHeightAndFlags(Parse *pParse, Expr *p){ + if( pParse->nErr ) return; + exprSetHeight(p); + sqlite3ExprCheckHeight(pParse, p->nHeight); +} + +/* +** Return the maximum height of any expression tree referenced +** by the select statement passed as an argument. +*/ +SQLITE_PRIVATE int sqlite3SelectExprHeight(Select *p){ + int nHeight = 0; + heightOfSelect(p, &nHeight); + return nHeight; +} +#else /* ABOVE: Height enforcement enabled. BELOW: Height enforcement off */ +/* +** Propagate all EP_Propagate flags from the Expr.x.pList into +** Expr.flags. +*/ +SQLITE_PRIVATE void sqlite3ExprSetHeightAndFlags(Parse *pParse, Expr *p){ + if( p && p->x.pList && !ExprHasProperty(p, EP_xIsSelect) ){ + p->flags |= EP_Propagate & sqlite3ExprListFlags(p->x.pList); + } +} +#define exprSetHeight(y) +#endif /* SQLITE_MAX_EXPR_DEPTH>0 */ + +/* +** This routine is the core allocator for Expr nodes. +** +** Construct a new expression node and return a pointer to it. Memory +** for this node and for the pToken argument is a single allocation +** obtained from sqlite3DbMalloc(). The calling function +** is responsible for making sure the node eventually gets freed. +** +** If dequote is true, then the token (if it exists) is dequoted. +** If dequote is false, no dequoting is performed. The deQuote +** parameter is ignored if pToken is NULL or if the token does not +** appear to be quoted. If the quotes were of the form "..." (double-quotes) +** then the EP_DblQuoted flag is set on the expression node. +** +** Special case: If op==TK_INTEGER and pToken points to a string that +** can be translated into a 32-bit integer, then the token is not +** stored in u.zToken. Instead, the integer values is written +** into u.iValue and the EP_IntValue flag is set. No extra storage +** is allocated to hold the integer text and the dequote flag is ignored. +*/ +SQLITE_PRIVATE Expr *sqlite3ExprAlloc( + sqlite3 *db, /* Handle for sqlite3DbMallocRawNN() */ + int op, /* Expression opcode */ + const Token *pToken, /* Token argument. Might be NULL */ + int dequote /* True to dequote */ +){ + Expr *pNew; + int nExtra = 0; + int iValue = 0; + + assert( db!=0 ); + if( pToken ){ + if( op!=TK_INTEGER || pToken->z==0 + || sqlite3GetInt32(pToken->z, &iValue)==0 ){ + nExtra = pToken->n+1; + assert( iValue>=0 ); + } + } + pNew = sqlite3DbMallocRawNN(db, sizeof(Expr)+nExtra); + if( pNew ){ + memset(pNew, 0, sizeof(Expr)); + pNew->op = (u8)op; + pNew->iAgg = -1; + if( pToken ){ + if( nExtra==0 ){ + pNew->flags |= EP_IntValue|EP_Leaf|(iValue?EP_IsTrue:EP_IsFalse); + pNew->u.iValue = iValue; + }else{ + pNew->u.zToken = (char*)&pNew[1]; + assert( pToken->z!=0 || pToken->n==0 ); + if( pToken->n ) memcpy(pNew->u.zToken, pToken->z, pToken->n); + pNew->u.zToken[pToken->n] = 0; + if( dequote && sqlite3Isquote(pNew->u.zToken[0]) ){ + sqlite3DequoteExpr(pNew); + } + } + } +#if SQLITE_MAX_EXPR_DEPTH>0 + pNew->nHeight = 1; +#endif + } + return pNew; +} + +/* +** Allocate a new expression node from a zero-terminated token that has +** already been dequoted. +*/ +SQLITE_PRIVATE Expr *sqlite3Expr( + sqlite3 *db, /* Handle for sqlite3DbMallocZero() (may be null) */ + int op, /* Expression opcode */ + const char *zToken /* Token argument. Might be NULL */ +){ + Token x; + x.z = zToken; + x.n = sqlite3Strlen30(zToken); + return sqlite3ExprAlloc(db, op, &x, 0); +} + +/* +** Attach subtrees pLeft and pRight to the Expr node pRoot. +** +** If pRoot==NULL that means that a memory allocation error has occurred. +** In that case, delete the subtrees pLeft and pRight. +*/ +SQLITE_PRIVATE void sqlite3ExprAttachSubtrees( + sqlite3 *db, + Expr *pRoot, + Expr *pLeft, + Expr *pRight +){ + if( pRoot==0 ){ + assert( db->mallocFailed ); + sqlite3ExprDelete(db, pLeft); + sqlite3ExprDelete(db, pRight); + }else{ + if( pRight ){ + pRoot->pRight = pRight; + pRoot->flags |= EP_Propagate & pRight->flags; + } + if( pLeft ){ + pRoot->pLeft = pLeft; + pRoot->flags |= EP_Propagate & pLeft->flags; + } + exprSetHeight(pRoot); + } +} + +/* +** Allocate an Expr node which joins as many as two subtrees. +** +** One or both of the subtrees can be NULL. Return a pointer to the new +** Expr node. Or, if an OOM error occurs, set pParse->db->mallocFailed, +** free the subtrees and return NULL. +*/ +SQLITE_PRIVATE Expr *sqlite3PExpr( + Parse *pParse, /* Parsing context */ + int op, /* Expression opcode */ + Expr *pLeft, /* Left operand */ + Expr *pRight /* Right operand */ +){ + Expr *p; + p = sqlite3DbMallocRawNN(pParse->db, sizeof(Expr)); + if( p ){ + memset(p, 0, sizeof(Expr)); + p->op = op & 0xff; + p->iAgg = -1; + sqlite3ExprAttachSubtrees(pParse->db, p, pLeft, pRight); + sqlite3ExprCheckHeight(pParse, p->nHeight); + }else{ + sqlite3ExprDelete(pParse->db, pLeft); + sqlite3ExprDelete(pParse->db, pRight); + } + return p; +} + +/* +** Add pSelect to the Expr.x.pSelect field. Or, if pExpr is NULL (due +** do a memory allocation failure) then delete the pSelect object. +*/ +SQLITE_PRIVATE void sqlite3PExprAddSelect(Parse *pParse, Expr *pExpr, Select *pSelect){ + if( pExpr ){ + pExpr->x.pSelect = pSelect; + ExprSetProperty(pExpr, EP_xIsSelect|EP_Subquery); + sqlite3ExprSetHeightAndFlags(pParse, pExpr); + }else{ + assert( pParse->db->mallocFailed ); + sqlite3SelectDelete(pParse->db, pSelect); + } +} + + +/* +** Join two expressions using an AND operator. If either expression is +** NULL, then just return the other expression. +** +** If one side or the other of the AND is known to be false, then instead +** of returning an AND expression, just return a constant expression with +** a value of false. +*/ +SQLITE_PRIVATE Expr *sqlite3ExprAnd(Parse *pParse, Expr *pLeft, Expr *pRight){ + sqlite3 *db = pParse->db; + if( pLeft==0 ){ + return pRight; + }else if( pRight==0 ){ + return pLeft; + }else if( ExprAlwaysFalse(pLeft) || ExprAlwaysFalse(pRight) ){ + sqlite3ExprUnmapAndDelete(pParse, pLeft); + sqlite3ExprUnmapAndDelete(pParse, pRight); + return sqlite3Expr(db, TK_INTEGER, "0"); + }else{ + return sqlite3PExpr(pParse, TK_AND, pLeft, pRight); + } +} + +/* +** Construct a new expression node for a function with multiple +** arguments. +*/ +SQLITE_PRIVATE Expr *sqlite3ExprFunction( + Parse *pParse, /* Parsing context */ + ExprList *pList, /* Argument list */ + Token *pToken, /* Name of the function */ + int eDistinct /* SF_Distinct or SF_ALL or 0 */ +){ + Expr *pNew; + sqlite3 *db = pParse->db; + assert( pToken ); + pNew = sqlite3ExprAlloc(db, TK_FUNCTION, pToken, 1); + if( pNew==0 ){ + sqlite3ExprListDelete(db, pList); /* Avoid memory leak when malloc fails */ + return 0; + } + if( pList && pList->nExpr > pParse->db->aLimit[SQLITE_LIMIT_FUNCTION_ARG] ){ + sqlite3ErrorMsg(pParse, "too many arguments on function %T", pToken); + } + pNew->x.pList = pList; + ExprSetProperty(pNew, EP_HasFunc); + assert( !ExprHasProperty(pNew, EP_xIsSelect) ); + sqlite3ExprSetHeightAndFlags(pParse, pNew); + if( eDistinct==SF_Distinct ) ExprSetProperty(pNew, EP_Distinct); + return pNew; +} + +/* +** Assign a variable number to an expression that encodes a wildcard +** in the original SQL statement. +** +** Wildcards consisting of a single "?" are assigned the next sequential +** variable number. +** +** Wildcards of the form "?nnn" are assigned the number "nnn". We make +** sure "nnn" is not too big to avoid a denial of service attack when +** the SQL statement comes from an external source. +** +** Wildcards of the form ":aaa", "@aaa", or "$aaa" are assigned the same number +** as the previous instance of the same wildcard. Or if this is the first +** instance of the wildcard, the next sequential variable number is +** assigned. +*/ +SQLITE_PRIVATE void sqlite3ExprAssignVarNumber(Parse *pParse, Expr *pExpr, u32 n){ + sqlite3 *db = pParse->db; + const char *z; + ynVar x; + + if( pExpr==0 ) return; + assert( !ExprHasProperty(pExpr, EP_IntValue|EP_Reduced|EP_TokenOnly) ); + z = pExpr->u.zToken; + assert( z!=0 ); + assert( z[0]!=0 ); + assert( n==(u32)sqlite3Strlen30(z) ); + if( z[1]==0 ){ + /* Wildcard of the form "?". Assign the next variable number */ + assert( z[0]=='?' ); + x = (ynVar)(++pParse->nVar); + }else{ + int doAdd = 0; + if( z[0]=='?' ){ + /* Wildcard of the form "?nnn". Convert "nnn" to an integer and + ** use it as the variable number */ + i64 i; + int bOk; + if( n==2 ){ /*OPTIMIZATION-IF-TRUE*/ + i = z[1]-'0'; /* The common case of ?N for a single digit N */ + bOk = 1; + }else{ + bOk = 0==sqlite3Atoi64(&z[1], &i, n-1, SQLITE_UTF8); + } + testcase( i==0 ); + testcase( i==1 ); + testcase( i==db->aLimit[SQLITE_LIMIT_VARIABLE_NUMBER]-1 ); + testcase( i==db->aLimit[SQLITE_LIMIT_VARIABLE_NUMBER] ); + if( bOk==0 || i<1 || i>db->aLimit[SQLITE_LIMIT_VARIABLE_NUMBER] ){ + sqlite3ErrorMsg(pParse, "variable number must be between ?1 and ?%d", + db->aLimit[SQLITE_LIMIT_VARIABLE_NUMBER]); + return; + } + x = (ynVar)i; + if( x>pParse->nVar ){ + pParse->nVar = (int)x; + doAdd = 1; + }else if( sqlite3VListNumToName(pParse->pVList, x)==0 ){ + doAdd = 1; + } + }else{ + /* Wildcards like ":aaa", "$aaa" or "@aaa". Reuse the same variable + ** number as the prior appearance of the same name, or if the name + ** has never appeared before, reuse the same variable number + */ + x = (ynVar)sqlite3VListNameToNum(pParse->pVList, z, n); + if( x==0 ){ + x = (ynVar)(++pParse->nVar); + doAdd = 1; + } + } + if( doAdd ){ + pParse->pVList = sqlite3VListAdd(db, pParse->pVList, z, n, x); + } + } + pExpr->iColumn = x; + if( x>db->aLimit[SQLITE_LIMIT_VARIABLE_NUMBER] ){ + sqlite3ErrorMsg(pParse, "too many SQL variables"); + } +} + +/* +** Recursively delete an expression tree. +*/ +static SQLITE_NOINLINE void sqlite3ExprDeleteNN(sqlite3 *db, Expr *p){ + assert( p!=0 ); + /* Sanity check: Assert that the IntValue is non-negative if it exists */ + assert( !ExprHasProperty(p, EP_IntValue) || p->u.iValue>=0 ); + + assert( !ExprHasProperty(p, EP_WinFunc) || p->y.pWin!=0 || db->mallocFailed ); + assert( p->op!=TK_FUNCTION || ExprHasProperty(p, EP_TokenOnly|EP_Reduced) + || p->y.pWin==0 || ExprHasProperty(p, EP_WinFunc) ); +#ifdef SQLITE_DEBUG + if( ExprHasProperty(p, EP_Leaf) && !ExprHasProperty(p, EP_TokenOnly) ){ + assert( p->pLeft==0 ); + assert( p->pRight==0 ); + assert( p->x.pSelect==0 ); + } +#endif + if( !ExprHasProperty(p, (EP_TokenOnly|EP_Leaf)) ){ + /* The Expr.x union is never used at the same time as Expr.pRight */ + assert( p->x.pList==0 || p->pRight==0 ); + if( p->pLeft && p->op!=TK_SELECT_COLUMN ) sqlite3ExprDeleteNN(db, p->pLeft); + if( p->pRight ){ + assert( !ExprHasProperty(p, EP_WinFunc) ); + sqlite3ExprDeleteNN(db, p->pRight); + }else if( ExprHasProperty(p, EP_xIsSelect) ){ + assert( !ExprHasProperty(p, EP_WinFunc) ); + sqlite3SelectDelete(db, p->x.pSelect); + }else{ + sqlite3ExprListDelete(db, p->x.pList); +#ifndef SQLITE_OMIT_WINDOWFUNC + if( ExprHasProperty(p, EP_WinFunc) ){ + sqlite3WindowDelete(db, p->y.pWin); + } +#endif + } + } + if( ExprHasProperty(p, EP_MemToken) ) sqlite3DbFree(db, p->u.zToken); + if( !ExprHasProperty(p, EP_Static) ){ + sqlite3DbFreeNN(db, p); + } +} +SQLITE_PRIVATE void sqlite3ExprDelete(sqlite3 *db, Expr *p){ + if( p ) sqlite3ExprDeleteNN(db, p); +} + +/* Invoke sqlite3RenameExprUnmap() and sqlite3ExprDelete() on the +** expression. +*/ +SQLITE_PRIVATE void sqlite3ExprUnmapAndDelete(Parse *pParse, Expr *p){ + if( p ){ + if( IN_RENAME_OBJECT ){ + sqlite3RenameExprUnmap(pParse, p); + } + sqlite3ExprDeleteNN(pParse->db, p); + } +} + +/* +** Return the number of bytes allocated for the expression structure +** passed as the first argument. This is always one of EXPR_FULLSIZE, +** EXPR_REDUCEDSIZE or EXPR_TOKENONLYSIZE. +*/ +static int exprStructSize(Expr *p){ + if( ExprHasProperty(p, EP_TokenOnly) ) return EXPR_TOKENONLYSIZE; + if( ExprHasProperty(p, EP_Reduced) ) return EXPR_REDUCEDSIZE; + return EXPR_FULLSIZE; +} + +/* +** The dupedExpr*Size() routines each return the number of bytes required +** to store a copy of an expression or expression tree. They differ in +** how much of the tree is measured. +** +** dupedExprStructSize() Size of only the Expr structure +** dupedExprNodeSize() Size of Expr + space for token +** dupedExprSize() Expr + token + subtree components +** +*************************************************************************** +** +** The dupedExprStructSize() function returns two values OR-ed together: +** (1) the space required for a copy of the Expr structure only and +** (2) the EP_xxx flags that indicate what the structure size should be. +** The return values is always one of: +** +** EXPR_FULLSIZE +** EXPR_REDUCEDSIZE | EP_Reduced +** EXPR_TOKENONLYSIZE | EP_TokenOnly +** +** The size of the structure can be found by masking the return value +** of this routine with 0xfff. The flags can be found by masking the +** return value with EP_Reduced|EP_TokenOnly. +** +** Note that with flags==EXPRDUP_REDUCE, this routines works on full-size +** (unreduced) Expr objects as they or originally constructed by the parser. +** During expression analysis, extra information is computed and moved into +** later parts of the Expr object and that extra information might get chopped +** off if the expression is reduced. Note also that it does not work to +** make an EXPRDUP_REDUCE copy of a reduced expression. It is only legal +** to reduce a pristine expression tree from the parser. The implementation +** of dupedExprStructSize() contain multiple assert() statements that attempt +** to enforce this constraint. +*/ +static int dupedExprStructSize(Expr *p, int flags){ + int nSize; + assert( flags==EXPRDUP_REDUCE || flags==0 ); /* Only one flag value allowed */ + assert( EXPR_FULLSIZE<=0xfff ); + assert( (0xfff & (EP_Reduced|EP_TokenOnly))==0 ); + if( 0==flags || p->op==TK_SELECT_COLUMN +#ifndef SQLITE_OMIT_WINDOWFUNC + || ExprHasProperty(p, EP_WinFunc) +#endif + ){ + nSize = EXPR_FULLSIZE; + }else{ + assert( !ExprHasProperty(p, EP_TokenOnly|EP_Reduced) ); + assert( !ExprHasProperty(p, EP_FromJoin) ); + assert( !ExprHasProperty(p, EP_MemToken) ); + assert( !ExprHasProperty(p, EP_NoReduce) ); + if( p->pLeft || p->x.pList ){ + nSize = EXPR_REDUCEDSIZE | EP_Reduced; + }else{ + assert( p->pRight==0 ); + nSize = EXPR_TOKENONLYSIZE | EP_TokenOnly; + } + } + return nSize; +} + +/* +** This function returns the space in bytes required to store the copy +** of the Expr structure and a copy of the Expr.u.zToken string (if that +** string is defined.) +*/ +static int dupedExprNodeSize(Expr *p, int flags){ + int nByte = dupedExprStructSize(p, flags) & 0xfff; + if( !ExprHasProperty(p, EP_IntValue) && p->u.zToken ){ + nByte += sqlite3Strlen30NN(p->u.zToken)+1; + } + return ROUND8(nByte); +} + +/* +** Return the number of bytes required to create a duplicate of the +** expression passed as the first argument. The second argument is a +** mask containing EXPRDUP_XXX flags. +** +** The value returned includes space to create a copy of the Expr struct +** itself and the buffer referred to by Expr.u.zToken, if any. +** +** If the EXPRDUP_REDUCE flag is set, then the return value includes +** space to duplicate all Expr nodes in the tree formed by Expr.pLeft +** and Expr.pRight variables (but not for any structures pointed to or +** descended from the Expr.x.pList or Expr.x.pSelect variables). +*/ +static int dupedExprSize(Expr *p, int flags){ + int nByte = 0; + if( p ){ + nByte = dupedExprNodeSize(p, flags); + if( flags&EXPRDUP_REDUCE ){ + nByte += dupedExprSize(p->pLeft, flags) + dupedExprSize(p->pRight, flags); + } + } + return nByte; +} + +/* +** This function is similar to sqlite3ExprDup(), except that if pzBuffer +** is not NULL then *pzBuffer is assumed to point to a buffer large enough +** to store the copy of expression p, the copies of p->u.zToken +** (if applicable), and the copies of the p->pLeft and p->pRight expressions, +** if any. Before returning, *pzBuffer is set to the first byte past the +** portion of the buffer copied into by this function. +*/ +static Expr *exprDup(sqlite3 *db, Expr *p, int dupFlags, u8 **pzBuffer){ + Expr *pNew; /* Value to return */ + u8 *zAlloc; /* Memory space from which to build Expr object */ + u32 staticFlag; /* EP_Static if space not obtained from malloc */ + + assert( db!=0 ); + assert( p ); + assert( dupFlags==0 || dupFlags==EXPRDUP_REDUCE ); + assert( pzBuffer==0 || dupFlags==EXPRDUP_REDUCE ); + + /* Figure out where to write the new Expr structure. */ + if( pzBuffer ){ + zAlloc = *pzBuffer; + staticFlag = EP_Static; + }else{ + zAlloc = sqlite3DbMallocRawNN(db, dupedExprSize(p, dupFlags)); + staticFlag = 0; + } + pNew = (Expr *)zAlloc; + + if( pNew ){ + /* Set nNewSize to the size allocated for the structure pointed to + ** by pNew. This is either EXPR_FULLSIZE, EXPR_REDUCEDSIZE or + ** EXPR_TOKENONLYSIZE. nToken is set to the number of bytes consumed + ** by the copy of the p->u.zToken string (if any). + */ + const unsigned nStructSize = dupedExprStructSize(p, dupFlags); + const int nNewSize = nStructSize & 0xfff; + int nToken; + if( !ExprHasProperty(p, EP_IntValue) && p->u.zToken ){ + nToken = sqlite3Strlen30(p->u.zToken) + 1; + }else{ + nToken = 0; + } + if( dupFlags ){ + assert( ExprHasProperty(p, EP_Reduced)==0 ); + memcpy(zAlloc, p, nNewSize); + }else{ + u32 nSize = (u32)exprStructSize(p); + memcpy(zAlloc, p, nSize); + if( nSizeflags &= ~(EP_Reduced|EP_TokenOnly|EP_Static|EP_MemToken); + pNew->flags |= nStructSize & (EP_Reduced|EP_TokenOnly); + pNew->flags |= staticFlag; + + /* Copy the p->u.zToken string, if any. */ + if( nToken ){ + char *zToken = pNew->u.zToken = (char*)&zAlloc[nNewSize]; + memcpy(zToken, p->u.zToken, nToken); + } + + if( 0==((p->flags|pNew->flags) & (EP_TokenOnly|EP_Leaf)) ){ + /* Fill in the pNew->x.pSelect or pNew->x.pList member. */ + if( ExprHasProperty(p, EP_xIsSelect) ){ + pNew->x.pSelect = sqlite3SelectDup(db, p->x.pSelect, dupFlags); + }else{ + pNew->x.pList = sqlite3ExprListDup(db, p->x.pList, dupFlags); + } + } + + /* Fill in pNew->pLeft and pNew->pRight. */ + if( ExprHasProperty(pNew, EP_Reduced|EP_TokenOnly|EP_WinFunc) ){ + zAlloc += dupedExprNodeSize(p, dupFlags); + if( !ExprHasProperty(pNew, EP_TokenOnly|EP_Leaf) ){ + pNew->pLeft = p->pLeft ? + exprDup(db, p->pLeft, EXPRDUP_REDUCE, &zAlloc) : 0; + pNew->pRight = p->pRight ? + exprDup(db, p->pRight, EXPRDUP_REDUCE, &zAlloc) : 0; + } +#ifndef SQLITE_OMIT_WINDOWFUNC + if( ExprHasProperty(p, EP_WinFunc) ){ + pNew->y.pWin = sqlite3WindowDup(db, pNew, p->y.pWin); + assert( ExprHasProperty(pNew, EP_WinFunc) ); + } +#endif /* SQLITE_OMIT_WINDOWFUNC */ + if( pzBuffer ){ + *pzBuffer = zAlloc; + } + }else{ + if( !ExprHasProperty(p, EP_TokenOnly|EP_Leaf) ){ + if( pNew->op==TK_SELECT_COLUMN ){ + pNew->pLeft = p->pLeft; + assert( p->iColumn==0 || p->pRight==0 ); + assert( p->pRight==0 || p->pRight==p->pLeft ); + }else{ + pNew->pLeft = sqlite3ExprDup(db, p->pLeft, 0); + } + pNew->pRight = sqlite3ExprDup(db, p->pRight, 0); + } + } + } + return pNew; +} + +/* +** Create and return a deep copy of the object passed as the second +** argument. If an OOM condition is encountered, NULL is returned +** and the db->mallocFailed flag set. +*/ +#ifndef SQLITE_OMIT_CTE +static With *withDup(sqlite3 *db, With *p){ + With *pRet = 0; + if( p ){ + sqlite3_int64 nByte = sizeof(*p) + sizeof(p->a[0]) * (p->nCte-1); + pRet = sqlite3DbMallocZero(db, nByte); + if( pRet ){ + int i; + pRet->nCte = p->nCte; + for(i=0; inCte; i++){ + pRet->a[i].pSelect = sqlite3SelectDup(db, p->a[i].pSelect, 0); + pRet->a[i].pCols = sqlite3ExprListDup(db, p->a[i].pCols, 0); + pRet->a[i].zName = sqlite3DbStrDup(db, p->a[i].zName); + } + } + } + return pRet; +} +#else +# define withDup(x,y) 0 +#endif + +#ifndef SQLITE_OMIT_WINDOWFUNC +/* +** The gatherSelectWindows() procedure and its helper routine +** gatherSelectWindowsCallback() are used to scan all the expressions +** an a newly duplicated SELECT statement and gather all of the Window +** objects found there, assembling them onto the linked list at Select->pWin. +*/ +static int gatherSelectWindowsCallback(Walker *pWalker, Expr *pExpr){ + if( pExpr->op==TK_FUNCTION && ExprHasProperty(pExpr, EP_WinFunc) ){ + Select *pSelect = pWalker->u.pSelect; + Window *pWin = pExpr->y.pWin; + assert( pWin ); + assert( IsWindowFunc(pExpr) ); + assert( pWin->ppThis==0 ); + sqlite3WindowLink(pSelect, pWin); + } + return WRC_Continue; +} +static int gatherSelectWindowsSelectCallback(Walker *pWalker, Select *p){ + return p==pWalker->u.pSelect ? WRC_Continue : WRC_Prune; +} +static void gatherSelectWindows(Select *p){ + Walker w; + w.xExprCallback = gatherSelectWindowsCallback; + w.xSelectCallback = gatherSelectWindowsSelectCallback; + w.xSelectCallback2 = 0; + w.pParse = 0; + w.u.pSelect = p; + sqlite3WalkSelect(&w, p); +} +#endif + + +/* +** The following group of routines make deep copies of expressions, +** expression lists, ID lists, and select statements. The copies can +** be deleted (by being passed to their respective ...Delete() routines) +** without effecting the originals. +** +** The expression list, ID, and source lists return by sqlite3ExprListDup(), +** sqlite3IdListDup(), and sqlite3SrcListDup() can not be further expanded +** by subsequent calls to sqlite*ListAppend() routines. +** +** Any tables that the SrcList might point to are not duplicated. +** +** The flags parameter contains a combination of the EXPRDUP_XXX flags. +** If the EXPRDUP_REDUCE flag is set, then the structure returned is a +** truncated version of the usual Expr structure that will be stored as +** part of the in-memory representation of the database schema. +*/ +SQLITE_PRIVATE Expr *sqlite3ExprDup(sqlite3 *db, Expr *p, int flags){ + assert( flags==0 || flags==EXPRDUP_REDUCE ); + return p ? exprDup(db, p, flags, 0) : 0; +} +SQLITE_PRIVATE ExprList *sqlite3ExprListDup(sqlite3 *db, ExprList *p, int flags){ + ExprList *pNew; + struct ExprList_item *pItem, *pOldItem; + int i; + Expr *pPriorSelectCol = 0; + assert( db!=0 ); + if( p==0 ) return 0; + pNew = sqlite3DbMallocRawNN(db, sqlite3DbMallocSize(db, p)); + if( pNew==0 ) return 0; + pNew->nExpr = p->nExpr; + pItem = pNew->a; + pOldItem = p->a; + for(i=0; inExpr; i++, pItem++, pOldItem++){ + Expr *pOldExpr = pOldItem->pExpr; + Expr *pNewExpr; + pItem->pExpr = sqlite3ExprDup(db, pOldExpr, flags); + if( pOldExpr + && pOldExpr->op==TK_SELECT_COLUMN + && (pNewExpr = pItem->pExpr)!=0 + ){ + assert( pNewExpr->iColumn==0 || i>0 ); + if( pNewExpr->iColumn==0 ){ + assert( pOldExpr->pLeft==pOldExpr->pRight ); + pPriorSelectCol = pNewExpr->pLeft = pNewExpr->pRight; + }else{ + assert( i>0 ); + assert( pItem[-1].pExpr!=0 ); + assert( pNewExpr->iColumn==pItem[-1].pExpr->iColumn+1 ); + assert( pPriorSelectCol==pItem[-1].pExpr->pLeft ); + pNewExpr->pLeft = pPriorSelectCol; + } + } + pItem->zName = sqlite3DbStrDup(db, pOldItem->zName); + pItem->zSpan = sqlite3DbStrDup(db, pOldItem->zSpan); + pItem->sortFlags = pOldItem->sortFlags; + pItem->done = 0; + pItem->bNulls = pOldItem->bNulls; + pItem->bSpanIsTab = pOldItem->bSpanIsTab; + pItem->bSorterRef = pOldItem->bSorterRef; + pItem->u = pOldItem->u; + } + return pNew; +} + +/* +** If cursors, triggers, views and subqueries are all omitted from +** the build, then none of the following routines, except for +** sqlite3SelectDup(), can be called. sqlite3SelectDup() is sometimes +** called with a NULL argument. +*/ +#if !defined(SQLITE_OMIT_VIEW) || !defined(SQLITE_OMIT_TRIGGER) \ + || !defined(SQLITE_OMIT_SUBQUERY) +SQLITE_PRIVATE SrcList *sqlite3SrcListDup(sqlite3 *db, SrcList *p, int flags){ + SrcList *pNew; + int i; + int nByte; + assert( db!=0 ); + if( p==0 ) return 0; + nByte = sizeof(*p) + (p->nSrc>0 ? sizeof(p->a[0]) * (p->nSrc-1) : 0); + pNew = sqlite3DbMallocRawNN(db, nByte ); + if( pNew==0 ) return 0; + pNew->nSrc = pNew->nAlloc = p->nSrc; + for(i=0; inSrc; i++){ + struct SrcList_item *pNewItem = &pNew->a[i]; + struct SrcList_item *pOldItem = &p->a[i]; + Table *pTab; + pNewItem->pSchema = pOldItem->pSchema; + pNewItem->zDatabase = sqlite3DbStrDup(db, pOldItem->zDatabase); + pNewItem->zName = sqlite3DbStrDup(db, pOldItem->zName); + pNewItem->zAlias = sqlite3DbStrDup(db, pOldItem->zAlias); + pNewItem->fg = pOldItem->fg; + pNewItem->iCursor = pOldItem->iCursor; + pNewItem->addrFillSub = pOldItem->addrFillSub; + pNewItem->regReturn = pOldItem->regReturn; + if( pNewItem->fg.isIndexedBy ){ + pNewItem->u1.zIndexedBy = sqlite3DbStrDup(db, pOldItem->u1.zIndexedBy); + } + pNewItem->pIBIndex = pOldItem->pIBIndex; + if( pNewItem->fg.isTabFunc ){ + pNewItem->u1.pFuncArg = + sqlite3ExprListDup(db, pOldItem->u1.pFuncArg, flags); + } + pTab = pNewItem->pTab = pOldItem->pTab; + if( pTab ){ + pTab->nTabRef++; + } + pNewItem->pSelect = sqlite3SelectDup(db, pOldItem->pSelect, flags); + pNewItem->pOn = sqlite3ExprDup(db, pOldItem->pOn, flags); + pNewItem->pUsing = sqlite3IdListDup(db, pOldItem->pUsing); + pNewItem->colUsed = pOldItem->colUsed; + } + return pNew; +} +SQLITE_PRIVATE IdList *sqlite3IdListDup(sqlite3 *db, IdList *p){ + IdList *pNew; + int i; + assert( db!=0 ); + if( p==0 ) return 0; + pNew = sqlite3DbMallocRawNN(db, sizeof(*pNew) ); + if( pNew==0 ) return 0; + pNew->nId = p->nId; + pNew->a = sqlite3DbMallocRawNN(db, p->nId*sizeof(p->a[0]) ); + if( pNew->a==0 ){ + sqlite3DbFreeNN(db, pNew); + return 0; + } + /* Note that because the size of the allocation for p->a[] is not + ** necessarily a power of two, sqlite3IdListAppend() may not be called + ** on the duplicate created by this function. */ + for(i=0; inId; i++){ + struct IdList_item *pNewItem = &pNew->a[i]; + struct IdList_item *pOldItem = &p->a[i]; + pNewItem->zName = sqlite3DbStrDup(db, pOldItem->zName); + pNewItem->idx = pOldItem->idx; + } + return pNew; +} +SQLITE_PRIVATE Select *sqlite3SelectDup(sqlite3 *db, Select *pDup, int flags){ + Select *pRet = 0; + Select *pNext = 0; + Select **pp = &pRet; + Select *p; + + assert( db!=0 ); + for(p=pDup; p; p=p->pPrior){ + Select *pNew = sqlite3DbMallocRawNN(db, sizeof(*p) ); + if( pNew==0 ) break; + pNew->pEList = sqlite3ExprListDup(db, p->pEList, flags); + pNew->pSrc = sqlite3SrcListDup(db, p->pSrc, flags); + pNew->pWhere = sqlite3ExprDup(db, p->pWhere, flags); + pNew->pGroupBy = sqlite3ExprListDup(db, p->pGroupBy, flags); + pNew->pHaving = sqlite3ExprDup(db, p->pHaving, flags); + pNew->pOrderBy = sqlite3ExprListDup(db, p->pOrderBy, flags); + pNew->op = p->op; + pNew->pNext = pNext; + pNew->pPrior = 0; + pNew->pLimit = sqlite3ExprDup(db, p->pLimit, flags); + pNew->iLimit = 0; + pNew->iOffset = 0; + pNew->selFlags = p->selFlags & ~SF_UsesEphemeral; + pNew->addrOpenEphm[0] = -1; + pNew->addrOpenEphm[1] = -1; + pNew->nSelectRow = p->nSelectRow; + pNew->pWith = withDup(db, p->pWith); +#ifndef SQLITE_OMIT_WINDOWFUNC + pNew->pWin = 0; + pNew->pWinDefn = sqlite3WindowListDup(db, p->pWinDefn); + if( p->pWin && db->mallocFailed==0 ) gatherSelectWindows(pNew); +#endif + pNew->selId = p->selId; + *pp = pNew; + pp = &pNew->pPrior; + pNext = pNew; + } + + return pRet; +} +#else +SQLITE_PRIVATE Select *sqlite3SelectDup(sqlite3 *db, Select *p, int flags){ + assert( p==0 ); + return 0; +} +#endif + + +/* +** Add a new element to the end of an expression list. If pList is +** initially NULL, then create a new expression list. +** +** The pList argument must be either NULL or a pointer to an ExprList +** obtained from a prior call to sqlite3ExprListAppend(). This routine +** may not be used with an ExprList obtained from sqlite3ExprListDup(). +** Reason: This routine assumes that the number of slots in pList->a[] +** is a power of two. That is true for sqlite3ExprListAppend() returns +** but is not necessarily true from the return value of sqlite3ExprListDup(). +** +** If a memory allocation error occurs, the entire list is freed and +** NULL is returned. If non-NULL is returned, then it is guaranteed +** that the new entry was successfully appended. +*/ +SQLITE_PRIVATE ExprList *sqlite3ExprListAppend( + Parse *pParse, /* Parsing context */ + ExprList *pList, /* List to which to append. Might be NULL */ + Expr *pExpr /* Expression to be appended. Might be NULL */ +){ + struct ExprList_item *pItem; + sqlite3 *db = pParse->db; + assert( db!=0 ); + if( pList==0 ){ + pList = sqlite3DbMallocRawNN(db, sizeof(ExprList) ); + if( pList==0 ){ + goto no_mem; + } + pList->nExpr = 0; + }else if( (pList->nExpr & (pList->nExpr-1))==0 ){ + ExprList *pNew; + pNew = sqlite3DbRealloc(db, pList, + sizeof(*pList)+(2*(sqlite3_int64)pList->nExpr-1)*sizeof(pList->a[0])); + if( pNew==0 ){ + goto no_mem; + } + pList = pNew; + } + pItem = &pList->a[pList->nExpr++]; + assert( offsetof(struct ExprList_item,zName)==sizeof(pItem->pExpr) ); + assert( offsetof(struct ExprList_item,pExpr)==0 ); + memset(&pItem->zName,0,sizeof(*pItem)-offsetof(struct ExprList_item,zName)); + pItem->pExpr = pExpr; + return pList; + +no_mem: + /* Avoid leaking memory if malloc has failed. */ + sqlite3ExprDelete(db, pExpr); + sqlite3ExprListDelete(db, pList); + return 0; +} + +/* +** pColumns and pExpr form a vector assignment which is part of the SET +** clause of an UPDATE statement. Like this: +** +** (a,b,c) = (expr1,expr2,expr3) +** Or: (a,b,c) = (SELECT x,y,z FROM ....) +** +** For each term of the vector assignment, append new entries to the +** expression list pList. In the case of a subquery on the RHS, append +** TK_SELECT_COLUMN expressions. +*/ +SQLITE_PRIVATE ExprList *sqlite3ExprListAppendVector( + Parse *pParse, /* Parsing context */ + ExprList *pList, /* List to which to append. Might be NULL */ + IdList *pColumns, /* List of names of LHS of the assignment */ + Expr *pExpr /* Vector expression to be appended. Might be NULL */ +){ + sqlite3 *db = pParse->db; + int n; + int i; + int iFirst = pList ? pList->nExpr : 0; + /* pColumns can only be NULL due to an OOM but an OOM will cause an + ** exit prior to this routine being invoked */ + if( NEVER(pColumns==0) ) goto vector_append_error; + if( pExpr==0 ) goto vector_append_error; + + /* If the RHS is a vector, then we can immediately check to see that + ** the size of the RHS and LHS match. But if the RHS is a SELECT, + ** wildcards ("*") in the result set of the SELECT must be expanded before + ** we can do the size check, so defer the size check until code generation. + */ + if( pExpr->op!=TK_SELECT && pColumns->nId!=(n=sqlite3ExprVectorSize(pExpr)) ){ + sqlite3ErrorMsg(pParse, "%d columns assigned %d values", + pColumns->nId, n); + goto vector_append_error; + } + + for(i=0; inId; i++){ + Expr *pSubExpr = sqlite3ExprForVectorField(pParse, pExpr, i); + assert( pSubExpr!=0 || db->mallocFailed ); + assert( pSubExpr==0 || pSubExpr->iTable==0 ); + if( pSubExpr==0 ) continue; + pSubExpr->iTable = pColumns->nId; + pList = sqlite3ExprListAppend(pParse, pList, pSubExpr); + if( pList ){ + assert( pList->nExpr==iFirst+i+1 ); + pList->a[pList->nExpr-1].zName = pColumns->a[i].zName; + pColumns->a[i].zName = 0; + } + } + + if( !db->mallocFailed && pExpr->op==TK_SELECT && ALWAYS(pList!=0) ){ + Expr *pFirst = pList->a[iFirst].pExpr; + assert( pFirst!=0 ); + assert( pFirst->op==TK_SELECT_COLUMN ); + + /* Store the SELECT statement in pRight so it will be deleted when + ** sqlite3ExprListDelete() is called */ + pFirst->pRight = pExpr; + pExpr = 0; + + /* Remember the size of the LHS in iTable so that we can check that + ** the RHS and LHS sizes match during code generation. */ + pFirst->iTable = pColumns->nId; + } + +vector_append_error: + sqlite3ExprUnmapAndDelete(pParse, pExpr); + sqlite3IdListDelete(db, pColumns); + return pList; +} + +/* +** Set the sort order for the last element on the given ExprList. +*/ +SQLITE_PRIVATE void sqlite3ExprListSetSortOrder(ExprList *p, int iSortOrder, int eNulls){ + struct ExprList_item *pItem; + if( p==0 ) return; + assert( p->nExpr>0 ); + + assert( SQLITE_SO_UNDEFINED<0 && SQLITE_SO_ASC==0 && SQLITE_SO_DESC>0 ); + assert( iSortOrder==SQLITE_SO_UNDEFINED + || iSortOrder==SQLITE_SO_ASC + || iSortOrder==SQLITE_SO_DESC + ); + assert( eNulls==SQLITE_SO_UNDEFINED + || eNulls==SQLITE_SO_ASC + || eNulls==SQLITE_SO_DESC + ); + + pItem = &p->a[p->nExpr-1]; + assert( pItem->bNulls==0 ); + if( iSortOrder==SQLITE_SO_UNDEFINED ){ + iSortOrder = SQLITE_SO_ASC; + } + pItem->sortFlags = (u8)iSortOrder; + + if( eNulls!=SQLITE_SO_UNDEFINED ){ + pItem->bNulls = 1; + if( iSortOrder!=eNulls ){ + pItem->sortFlags |= KEYINFO_ORDER_BIGNULL; + } + } +} + +/* +** Set the ExprList.a[].zName element of the most recently added item +** on the expression list. +** +** pList might be NULL following an OOM error. But pName should never be +** NULL. If a memory allocation fails, the pParse->db->mallocFailed flag +** is set. +*/ +SQLITE_PRIVATE void sqlite3ExprListSetName( + Parse *pParse, /* Parsing context */ + ExprList *pList, /* List to which to add the span. */ + Token *pName, /* Name to be added */ + int dequote /* True to cause the name to be dequoted */ +){ + assert( pList!=0 || pParse->db->mallocFailed!=0 ); + if( pList ){ + struct ExprList_item *pItem; + assert( pList->nExpr>0 ); + pItem = &pList->a[pList->nExpr-1]; + assert( pItem->zName==0 ); + pItem->zName = sqlite3DbStrNDup(pParse->db, pName->z, pName->n); + if( dequote ) sqlite3Dequote(pItem->zName); + if( IN_RENAME_OBJECT ){ + sqlite3RenameTokenMap(pParse, (void*)pItem->zName, pName); + } + } +} + +/* +** Set the ExprList.a[].zSpan element of the most recently added item +** on the expression list. +** +** pList might be NULL following an OOM error. But pSpan should never be +** NULL. If a memory allocation fails, the pParse->db->mallocFailed flag +** is set. +*/ +SQLITE_PRIVATE void sqlite3ExprListSetSpan( + Parse *pParse, /* Parsing context */ + ExprList *pList, /* List to which to add the span. */ + const char *zStart, /* Start of the span */ + const char *zEnd /* End of the span */ +){ + sqlite3 *db = pParse->db; + assert( pList!=0 || db->mallocFailed!=0 ); + if( pList ){ + struct ExprList_item *pItem = &pList->a[pList->nExpr-1]; + assert( pList->nExpr>0 ); + sqlite3DbFree(db, pItem->zSpan); + pItem->zSpan = sqlite3DbSpanDup(db, zStart, zEnd); + } +} + +/* +** If the expression list pEList contains more than iLimit elements, +** leave an error message in pParse. +*/ +SQLITE_PRIVATE void sqlite3ExprListCheckLength( + Parse *pParse, + ExprList *pEList, + const char *zObject +){ + int mx = pParse->db->aLimit[SQLITE_LIMIT_COLUMN]; + testcase( pEList && pEList->nExpr==mx ); + testcase( pEList && pEList->nExpr==mx+1 ); + if( pEList && pEList->nExpr>mx ){ + sqlite3ErrorMsg(pParse, "too many columns in %s", zObject); + } +} + +/* +** Delete an entire expression list. +*/ +static SQLITE_NOINLINE void exprListDeleteNN(sqlite3 *db, ExprList *pList){ + int i = pList->nExpr; + struct ExprList_item *pItem = pList->a; + assert( pList->nExpr>0 ); + do{ + sqlite3ExprDelete(db, pItem->pExpr); + sqlite3DbFree(db, pItem->zName); + sqlite3DbFree(db, pItem->zSpan); + pItem++; + }while( --i>0 ); + sqlite3DbFreeNN(db, pList); +} +SQLITE_PRIVATE void sqlite3ExprListDelete(sqlite3 *db, ExprList *pList){ + if( pList ) exprListDeleteNN(db, pList); +} + +/* +** Return the bitwise-OR of all Expr.flags fields in the given +** ExprList. +*/ +SQLITE_PRIVATE u32 sqlite3ExprListFlags(const ExprList *pList){ + int i; + u32 m = 0; + assert( pList!=0 ); + for(i=0; inExpr; i++){ + Expr *pExpr = pList->a[i].pExpr; + assert( pExpr!=0 ); + m |= pExpr->flags; + } + return m; +} + +/* +** This is a SELECT-node callback for the expression walker that +** always "fails". By "fail" in this case, we mean set +** pWalker->eCode to zero and abort. +** +** This callback is used by multiple expression walkers. +*/ +SQLITE_PRIVATE int sqlite3SelectWalkFail(Walker *pWalker, Select *NotUsed){ + UNUSED_PARAMETER(NotUsed); + pWalker->eCode = 0; + return WRC_Abort; +} + +/* +** If the input expression is an ID with the name "true" or "false" +** then convert it into an TK_TRUEFALSE term. Return non-zero if +** the conversion happened, and zero if the expression is unaltered. +*/ +SQLITE_PRIVATE int sqlite3ExprIdToTrueFalse(Expr *pExpr){ + assert( pExpr->op==TK_ID || pExpr->op==TK_STRING ); + if( !ExprHasProperty(pExpr, EP_Quoted) + && (sqlite3StrICmp(pExpr->u.zToken, "true")==0 + || sqlite3StrICmp(pExpr->u.zToken, "false")==0) + ){ + pExpr->op = TK_TRUEFALSE; + ExprSetProperty(pExpr, pExpr->u.zToken[4]==0 ? EP_IsTrue : EP_IsFalse); + return 1; + } + return 0; +} + +/* +** The argument must be a TK_TRUEFALSE Expr node. Return 1 if it is TRUE +** and 0 if it is FALSE. +*/ +SQLITE_PRIVATE int sqlite3ExprTruthValue(const Expr *pExpr){ + pExpr = sqlite3ExprSkipCollate((Expr*)pExpr); + assert( pExpr->op==TK_TRUEFALSE ); + assert( sqlite3StrICmp(pExpr->u.zToken,"true")==0 + || sqlite3StrICmp(pExpr->u.zToken,"false")==0 ); + return pExpr->u.zToken[4]==0; +} + +/* +** If pExpr is an AND or OR expression, try to simplify it by eliminating +** terms that are always true or false. Return the simplified expression. +** Or return the original expression if no simplification is possible. +** +** Examples: +** +** (x<10) AND true => (x<10) +** (x<10) AND false => false +** (x<10) AND (y=22 OR false) => (x<10) AND (y=22) +** (x<10) AND (y=22 OR true) => (x<10) +** (y=22) OR true => true +*/ +SQLITE_PRIVATE Expr *sqlite3ExprSimplifiedAndOr(Expr *pExpr){ + assert( pExpr!=0 ); + if( pExpr->op==TK_AND || pExpr->op==TK_OR ){ + Expr *pRight = sqlite3ExprSimplifiedAndOr(pExpr->pRight); + Expr *pLeft = sqlite3ExprSimplifiedAndOr(pExpr->pLeft); + if( ExprAlwaysTrue(pLeft) || ExprAlwaysFalse(pRight) ){ + pExpr = pExpr->op==TK_AND ? pRight : pLeft; + }else if( ExprAlwaysTrue(pRight) || ExprAlwaysFalse(pLeft) ){ + pExpr = pExpr->op==TK_AND ? pLeft : pRight; + } + } + return pExpr; +} + + +/* +** These routines are Walker callbacks used to check expressions to +** see if they are "constant" for some definition of constant. The +** Walker.eCode value determines the type of "constant" we are looking +** for. +** +** These callback routines are used to implement the following: +** +** sqlite3ExprIsConstant() pWalker->eCode==1 +** sqlite3ExprIsConstantNotJoin() pWalker->eCode==2 +** sqlite3ExprIsTableConstant() pWalker->eCode==3 +** sqlite3ExprIsConstantOrFunction() pWalker->eCode==4 or 5 +** +** In all cases, the callbacks set Walker.eCode=0 and abort if the expression +** is found to not be a constant. +** +** The sqlite3ExprIsConstantOrFunction() is used for evaluating expressions +** in a CREATE TABLE statement. The Walker.eCode value is 5 when parsing +** an existing schema and 4 when processing a new statement. A bound +** parameter raises an error for new statements, but is silently converted +** to NULL for existing schemas. This allows sqlite_master tables that +** contain a bound parameter because they were generated by older versions +** of SQLite to be parsed by newer versions of SQLite without raising a +** malformed schema error. +*/ +static int exprNodeIsConstant(Walker *pWalker, Expr *pExpr){ + + /* If pWalker->eCode is 2 then any term of the expression that comes from + ** the ON or USING clauses of a left join disqualifies the expression + ** from being considered constant. */ + if( pWalker->eCode==2 && ExprHasProperty(pExpr, EP_FromJoin) ){ + pWalker->eCode = 0; + return WRC_Abort; + } + + switch( pExpr->op ){ + /* Consider functions to be constant if all their arguments are constant + ** and either pWalker->eCode==4 or 5 or the function has the + ** SQLITE_FUNC_CONST flag. */ + case TK_FUNCTION: + if( pWalker->eCode>=4 || ExprHasProperty(pExpr,EP_ConstFunc) ){ + return WRC_Continue; + }else{ + pWalker->eCode = 0; + return WRC_Abort; + } + case TK_ID: + /* Convert "true" or "false" in a DEFAULT clause into the + ** appropriate TK_TRUEFALSE operator */ + if( sqlite3ExprIdToTrueFalse(pExpr) ){ + return WRC_Prune; + } + /* Fall thru */ + case TK_COLUMN: + case TK_AGG_FUNCTION: + case TK_AGG_COLUMN: + testcase( pExpr->op==TK_ID ); + testcase( pExpr->op==TK_COLUMN ); + testcase( pExpr->op==TK_AGG_FUNCTION ); + testcase( pExpr->op==TK_AGG_COLUMN ); + if( ExprHasProperty(pExpr, EP_FixedCol) && pWalker->eCode!=2 ){ + return WRC_Continue; + } + if( pWalker->eCode==3 && pExpr->iTable==pWalker->u.iCur ){ + return WRC_Continue; + } + /* Fall through */ + case TK_IF_NULL_ROW: + case TK_REGISTER: + testcase( pExpr->op==TK_REGISTER ); + testcase( pExpr->op==TK_IF_NULL_ROW ); + pWalker->eCode = 0; + return WRC_Abort; + case TK_VARIABLE: + if( pWalker->eCode==5 ){ + /* Silently convert bound parameters that appear inside of CREATE + ** statements into a NULL when parsing the CREATE statement text out + ** of the sqlite_master table */ + pExpr->op = TK_NULL; + }else if( pWalker->eCode==4 ){ + /* A bound parameter in a CREATE statement that originates from + ** sqlite3_prepare() causes an error */ + pWalker->eCode = 0; + return WRC_Abort; + } + /* Fall through */ + default: + testcase( pExpr->op==TK_SELECT ); /* sqlite3SelectWalkFail() disallows */ + testcase( pExpr->op==TK_EXISTS ); /* sqlite3SelectWalkFail() disallows */ + return WRC_Continue; + } +} +static int exprIsConst(Expr *p, int initFlag, int iCur){ + Walker w; + w.eCode = initFlag; + w.xExprCallback = exprNodeIsConstant; + w.xSelectCallback = sqlite3SelectWalkFail; +#ifdef SQLITE_DEBUG + w.xSelectCallback2 = sqlite3SelectWalkAssert2; +#endif + w.u.iCur = iCur; + sqlite3WalkExpr(&w, p); + return w.eCode; +} + +/* +** Walk an expression tree. Return non-zero if the expression is constant +** and 0 if it involves variables or function calls. +** +** For the purposes of this function, a double-quoted string (ex: "abc") +** is considered a variable but a single-quoted string (ex: 'abc') is +** a constant. +*/ +SQLITE_PRIVATE int sqlite3ExprIsConstant(Expr *p){ + return exprIsConst(p, 1, 0); +} + +/* +** Walk an expression tree. Return non-zero if +** +** (1) the expression is constant, and +** (2) the expression does originate in the ON or USING clause +** of a LEFT JOIN, and +** (3) the expression does not contain any EP_FixedCol TK_COLUMN +** operands created by the constant propagation optimization. +** +** When this routine returns true, it indicates that the expression +** can be added to the pParse->pConstExpr list and evaluated once when +** the prepared statement starts up. See sqlite3ExprCodeAtInit(). +*/ +SQLITE_PRIVATE int sqlite3ExprIsConstantNotJoin(Expr *p){ + return exprIsConst(p, 2, 0); +} + +/* +** Walk an expression tree. Return non-zero if the expression is constant +** for any single row of the table with cursor iCur. In other words, the +** expression must not refer to any non-deterministic function nor any +** table other than iCur. +*/ +SQLITE_PRIVATE int sqlite3ExprIsTableConstant(Expr *p, int iCur){ + return exprIsConst(p, 3, iCur); +} + + +/* +** sqlite3WalkExpr() callback used by sqlite3ExprIsConstantOrGroupBy(). +*/ +static int exprNodeIsConstantOrGroupBy(Walker *pWalker, Expr *pExpr){ + ExprList *pGroupBy = pWalker->u.pGroupBy; + int i; + + /* Check if pExpr is identical to any GROUP BY term. If so, consider + ** it constant. */ + for(i=0; inExpr; i++){ + Expr *p = pGroupBy->a[i].pExpr; + if( sqlite3ExprCompare(0, pExpr, p, -1)<2 ){ + CollSeq *pColl = sqlite3ExprNNCollSeq(pWalker->pParse, p); + if( sqlite3IsBinary(pColl) ){ + return WRC_Prune; + } + } + } + + /* Check if pExpr is a sub-select. If so, consider it variable. */ + if( ExprHasProperty(pExpr, EP_xIsSelect) ){ + pWalker->eCode = 0; + return WRC_Abort; + } + + return exprNodeIsConstant(pWalker, pExpr); +} + +/* +** Walk the expression tree passed as the first argument. Return non-zero +** if the expression consists entirely of constants or copies of terms +** in pGroupBy that sort with the BINARY collation sequence. +** +** This routine is used to determine if a term of the HAVING clause can +** be promoted into the WHERE clause. In order for such a promotion to work, +** the value of the HAVING clause term must be the same for all members of +** a "group". The requirement that the GROUP BY term must be BINARY +** assumes that no other collating sequence will have a finer-grained +** grouping than binary. In other words (A=B COLLATE binary) implies +** A=B in every other collating sequence. The requirement that the +** GROUP BY be BINARY is stricter than necessary. It would also work +** to promote HAVING clauses that use the same alternative collating +** sequence as the GROUP BY term, but that is much harder to check, +** alternative collating sequences are uncommon, and this is only an +** optimization, so we take the easy way out and simply require the +** GROUP BY to use the BINARY collating sequence. +*/ +SQLITE_PRIVATE int sqlite3ExprIsConstantOrGroupBy(Parse *pParse, Expr *p, ExprList *pGroupBy){ + Walker w; + w.eCode = 1; + w.xExprCallback = exprNodeIsConstantOrGroupBy; + w.xSelectCallback = 0; + w.u.pGroupBy = pGroupBy; + w.pParse = pParse; + sqlite3WalkExpr(&w, p); + return w.eCode; +} + +/* +** Walk an expression tree. Return non-zero if the expression is constant +** or a function call with constant arguments. Return and 0 if there +** are any variables. +** +** For the purposes of this function, a double-quoted string (ex: "abc") +** is considered a variable but a single-quoted string (ex: 'abc') is +** a constant. +*/ +SQLITE_PRIVATE int sqlite3ExprIsConstantOrFunction(Expr *p, u8 isInit){ + assert( isInit==0 || isInit==1 ); + return exprIsConst(p, 4+isInit, 0); +} + +#ifdef SQLITE_ENABLE_CURSOR_HINTS +/* +** Walk an expression tree. Return 1 if the expression contains a +** subquery of some kind. Return 0 if there are no subqueries. +*/ +SQLITE_PRIVATE int sqlite3ExprContainsSubquery(Expr *p){ + Walker w; + w.eCode = 1; + w.xExprCallback = sqlite3ExprWalkNoop; + w.xSelectCallback = sqlite3SelectWalkFail; +#ifdef SQLITE_DEBUG + w.xSelectCallback2 = sqlite3SelectWalkAssert2; +#endif + sqlite3WalkExpr(&w, p); + return w.eCode==0; +} +#endif + +/* +** If the expression p codes a constant integer that is small enough +** to fit in a 32-bit integer, return 1 and put the value of the integer +** in *pValue. If the expression is not an integer or if it is too big +** to fit in a signed 32-bit integer, return 0 and leave *pValue unchanged. +*/ +SQLITE_PRIVATE int sqlite3ExprIsInteger(Expr *p, int *pValue){ + int rc = 0; + if( NEVER(p==0) ) return 0; /* Used to only happen following on OOM */ + + /* If an expression is an integer literal that fits in a signed 32-bit + ** integer, then the EP_IntValue flag will have already been set */ + assert( p->op!=TK_INTEGER || (p->flags & EP_IntValue)!=0 + || sqlite3GetInt32(p->u.zToken, &rc)==0 ); + + if( p->flags & EP_IntValue ){ + *pValue = p->u.iValue; + return 1; + } + switch( p->op ){ + case TK_UPLUS: { + rc = sqlite3ExprIsInteger(p->pLeft, pValue); + break; + } + case TK_UMINUS: { + int v; + if( sqlite3ExprIsInteger(p->pLeft, &v) ){ + assert( v!=(-2147483647-1) ); + *pValue = -v; + rc = 1; + } + break; + } + default: break; + } + return rc; +} + +/* +** Return FALSE if there is no chance that the expression can be NULL. +** +** If the expression might be NULL or if the expression is too complex +** to tell return TRUE. +** +** This routine is used as an optimization, to skip OP_IsNull opcodes +** when we know that a value cannot be NULL. Hence, a false positive +** (returning TRUE when in fact the expression can never be NULL) might +** be a small performance hit but is otherwise harmless. On the other +** hand, a false negative (returning FALSE when the result could be NULL) +** will likely result in an incorrect answer. So when in doubt, return +** TRUE. +*/ +SQLITE_PRIVATE int sqlite3ExprCanBeNull(const Expr *p){ + u8 op; + while( p->op==TK_UPLUS || p->op==TK_UMINUS ){ + p = p->pLeft; + } + op = p->op; + if( op==TK_REGISTER ) op = p->op2; + switch( op ){ + case TK_INTEGER: + case TK_STRING: + case TK_FLOAT: + case TK_BLOB: + return 0; + case TK_COLUMN: + return ExprHasProperty(p, EP_CanBeNull) || + p->y.pTab==0 || /* Reference to column of index on expression */ + (p->iColumn>=0 && p->y.pTab->aCol[p->iColumn].notNull==0); + default: + return 1; + } +} + +/* +** Return TRUE if the given expression is a constant which would be +** unchanged by OP_Affinity with the affinity given in the second +** argument. +** +** This routine is used to determine if the OP_Affinity operation +** can be omitted. When in doubt return FALSE. A false negative +** is harmless. A false positive, however, can result in the wrong +** answer. +*/ +SQLITE_PRIVATE int sqlite3ExprNeedsNoAffinityChange(const Expr *p, char aff){ + u8 op; + int unaryMinus = 0; + if( aff==SQLITE_AFF_BLOB ) return 1; + while( p->op==TK_UPLUS || p->op==TK_UMINUS ){ + if( p->op==TK_UMINUS ) unaryMinus = 1; + p = p->pLeft; + } + op = p->op; + if( op==TK_REGISTER ) op = p->op2; + switch( op ){ + case TK_INTEGER: { + return aff>=SQLITE_AFF_NUMERIC; + } + case TK_FLOAT: { + return aff>=SQLITE_AFF_NUMERIC; + } + case TK_STRING: { + return !unaryMinus && aff==SQLITE_AFF_TEXT; + } + case TK_BLOB: { + return !unaryMinus; + } + case TK_COLUMN: { + assert( p->iTable>=0 ); /* p cannot be part of a CHECK constraint */ + return aff>=SQLITE_AFF_NUMERIC && p->iColumn<0; + } + default: { + return 0; + } + } +} + +/* +** Return TRUE if the given string is a row-id column name. +*/ +SQLITE_PRIVATE int sqlite3IsRowid(const char *z){ + if( sqlite3StrICmp(z, "_ROWID_")==0 ) return 1; + if( sqlite3StrICmp(z, "ROWID")==0 ) return 1; + if( sqlite3StrICmp(z, "OID")==0 ) return 1; + return 0; +} + +/* +** pX is the RHS of an IN operator. If pX is a SELECT statement +** that can be simplified to a direct table access, then return +** a pointer to the SELECT statement. If pX is not a SELECT statement, +** or if the SELECT statement needs to be manifested into a transient +** table, then return NULL. +*/ +#ifndef SQLITE_OMIT_SUBQUERY +static Select *isCandidateForInOpt(Expr *pX){ + Select *p; + SrcList *pSrc; + ExprList *pEList; + Table *pTab; + int i; + if( !ExprHasProperty(pX, EP_xIsSelect) ) return 0; /* Not a subquery */ + if( ExprHasProperty(pX, EP_VarSelect) ) return 0; /* Correlated subq */ + p = pX->x.pSelect; + if( p->pPrior ) return 0; /* Not a compound SELECT */ + if( p->selFlags & (SF_Distinct|SF_Aggregate) ){ + testcase( (p->selFlags & (SF_Distinct|SF_Aggregate))==SF_Distinct ); + testcase( (p->selFlags & (SF_Distinct|SF_Aggregate))==SF_Aggregate ); + return 0; /* No DISTINCT keyword and no aggregate functions */ + } + assert( p->pGroupBy==0 ); /* Has no GROUP BY clause */ + if( p->pLimit ) return 0; /* Has no LIMIT clause */ + if( p->pWhere ) return 0; /* Has no WHERE clause */ + pSrc = p->pSrc; + assert( pSrc!=0 ); + if( pSrc->nSrc!=1 ) return 0; /* Single term in FROM clause */ + if( pSrc->a[0].pSelect ) return 0; /* FROM is not a subquery or view */ + pTab = pSrc->a[0].pTab; + assert( pTab!=0 ); + assert( pTab->pSelect==0 ); /* FROM clause is not a view */ + if( IsVirtual(pTab) ) return 0; /* FROM clause not a virtual table */ + pEList = p->pEList; + assert( pEList!=0 ); + /* All SELECT results must be columns. */ + for(i=0; inExpr; i++){ + Expr *pRes = pEList->a[i].pExpr; + if( pRes->op!=TK_COLUMN ) return 0; + assert( pRes->iTable==pSrc->a[0].iCursor ); /* Not a correlated subquery */ + } + return p; +} +#endif /* SQLITE_OMIT_SUBQUERY */ + +#ifndef SQLITE_OMIT_SUBQUERY +/* +** Generate code that checks the left-most column of index table iCur to see if +** it contains any NULL entries. Cause the register at regHasNull to be set +** to a non-NULL value if iCur contains no NULLs. Cause register regHasNull +** to be set to NULL if iCur contains one or more NULL values. +*/ +static void sqlite3SetHasNullFlag(Vdbe *v, int iCur, int regHasNull){ + int addr1; + sqlite3VdbeAddOp2(v, OP_Integer, 0, regHasNull); + addr1 = sqlite3VdbeAddOp1(v, OP_Rewind, iCur); VdbeCoverage(v); + sqlite3VdbeAddOp3(v, OP_Column, iCur, 0, regHasNull); + sqlite3VdbeChangeP5(v, OPFLAG_TYPEOFARG); + VdbeComment((v, "first_entry_in(%d)", iCur)); + sqlite3VdbeJumpHere(v, addr1); +} +#endif + + +#ifndef SQLITE_OMIT_SUBQUERY +/* +** The argument is an IN operator with a list (not a subquery) on the +** right-hand side. Return TRUE if that list is constant. +*/ +static int sqlite3InRhsIsConstant(Expr *pIn){ + Expr *pLHS; + int res; + assert( !ExprHasProperty(pIn, EP_xIsSelect) ); + pLHS = pIn->pLeft; + pIn->pLeft = 0; + res = sqlite3ExprIsConstant(pIn); + pIn->pLeft = pLHS; + return res; +} +#endif + +/* +** This function is used by the implementation of the IN (...) operator. +** The pX parameter is the expression on the RHS of the IN operator, which +** might be either a list of expressions or a subquery. +** +** The job of this routine is to find or create a b-tree object that can +** be used either to test for membership in the RHS set or to iterate through +** all members of the RHS set, skipping duplicates. +** +** A cursor is opened on the b-tree object that is the RHS of the IN operator +** and pX->iTable is set to the index of that cursor. +** +** The returned value of this function indicates the b-tree type, as follows: +** +** IN_INDEX_ROWID - The cursor was opened on a database table. +** IN_INDEX_INDEX_ASC - The cursor was opened on an ascending index. +** IN_INDEX_INDEX_DESC - The cursor was opened on a descending index. +** IN_INDEX_EPH - The cursor was opened on a specially created and +** populated epheremal table. +** IN_INDEX_NOOP - No cursor was allocated. The IN operator must be +** implemented as a sequence of comparisons. +** +** An existing b-tree might be used if the RHS expression pX is a simple +** subquery such as: +** +** SELECT , ... FROM +** +** If the RHS of the IN operator is a list or a more complex subquery, then +** an ephemeral table might need to be generated from the RHS and then +** pX->iTable made to point to the ephemeral table instead of an +** existing table. +** +** The inFlags parameter must contain, at a minimum, one of the bits +** IN_INDEX_MEMBERSHIP or IN_INDEX_LOOP but not both. If inFlags contains +** IN_INDEX_MEMBERSHIP, then the generated table will be used for a fast +** membership test. When the IN_INDEX_LOOP bit is set, the IN index will +** be used to loop over all values of the RHS of the IN operator. +** +** When IN_INDEX_LOOP is used (and the b-tree will be used to iterate +** through the set members) then the b-tree must not contain duplicates. +** An epheremal table will be created unless the selected columns are guaranteed +** to be unique - either because it is an INTEGER PRIMARY KEY or due to +** a UNIQUE constraint or index. +** +** When IN_INDEX_MEMBERSHIP is used (and the b-tree will be used +** for fast set membership tests) then an epheremal table must +** be used unless is a single INTEGER PRIMARY KEY column or an +** index can be found with the specified as its left-most. +** +** If the IN_INDEX_NOOP_OK and IN_INDEX_MEMBERSHIP are both set and +** if the RHS of the IN operator is a list (not a subquery) then this +** routine might decide that creating an ephemeral b-tree for membership +** testing is too expensive and return IN_INDEX_NOOP. In that case, the +** calling routine should implement the IN operator using a sequence +** of Eq or Ne comparison operations. +** +** When the b-tree is being used for membership tests, the calling function +** might need to know whether or not the RHS side of the IN operator +** contains a NULL. If prRhsHasNull is not a NULL pointer and +** if there is any chance that the (...) might contain a NULL value at +** runtime, then a register is allocated and the register number written +** to *prRhsHasNull. If there is no chance that the (...) contains a +** NULL value, then *prRhsHasNull is left unchanged. +** +** If a register is allocated and its location stored in *prRhsHasNull, then +** the value in that register will be NULL if the b-tree contains one or more +** NULL values, and it will be some non-NULL value if the b-tree contains no +** NULL values. +** +** If the aiMap parameter is not NULL, it must point to an array containing +** one element for each column returned by the SELECT statement on the RHS +** of the IN(...) operator. The i'th entry of the array is populated with the +** offset of the index column that matches the i'th column returned by the +** SELECT. For example, if the expression and selected index are: +** +** (?,?,?) IN (SELECT a, b, c FROM t1) +** CREATE INDEX i1 ON t1(b, c, a); +** +** then aiMap[] is populated with {2, 0, 1}. +*/ +#ifndef SQLITE_OMIT_SUBQUERY +SQLITE_PRIVATE int sqlite3FindInIndex( + Parse *pParse, /* Parsing context */ + Expr *pX, /* The IN expression */ + u32 inFlags, /* IN_INDEX_LOOP, _MEMBERSHIP, and/or _NOOP_OK */ + int *prRhsHasNull, /* Register holding NULL status. See notes */ + int *aiMap, /* Mapping from Index fields to RHS fields */ + int *piTab /* OUT: index to use */ +){ + Select *p; /* SELECT to the right of IN operator */ + int eType = 0; /* Type of RHS table. IN_INDEX_* */ + int iTab = pParse->nTab++; /* Cursor of the RHS table */ + int mustBeUnique; /* True if RHS must be unique */ + Vdbe *v = sqlite3GetVdbe(pParse); /* Virtual machine being coded */ + + assert( pX->op==TK_IN ); + mustBeUnique = (inFlags & IN_INDEX_LOOP)!=0; + + /* If the RHS of this IN(...) operator is a SELECT, and if it matters + ** whether or not the SELECT result contains NULL values, check whether + ** or not NULL is actually possible (it may not be, for example, due + ** to NOT NULL constraints in the schema). If no NULL values are possible, + ** set prRhsHasNull to 0 before continuing. */ + if( prRhsHasNull && (pX->flags & EP_xIsSelect) ){ + int i; + ExprList *pEList = pX->x.pSelect->pEList; + for(i=0; inExpr; i++){ + if( sqlite3ExprCanBeNull(pEList->a[i].pExpr) ) break; + } + if( i==pEList->nExpr ){ + prRhsHasNull = 0; + } + } + + /* Check to see if an existing table or index can be used to + ** satisfy the query. This is preferable to generating a new + ** ephemeral table. */ + if( pParse->nErr==0 && (p = isCandidateForInOpt(pX))!=0 ){ + sqlite3 *db = pParse->db; /* Database connection */ + Table *pTab; /* Table
    . */ + i16 iDb; /* Database idx for pTab */ + ExprList *pEList = p->pEList; + int nExpr = pEList->nExpr; + + assert( p->pEList!=0 ); /* Because of isCandidateForInOpt(p) */ + assert( p->pEList->a[0].pExpr!=0 ); /* Because of isCandidateForInOpt(p) */ + assert( p->pSrc!=0 ); /* Because of isCandidateForInOpt(p) */ + pTab = p->pSrc->a[0].pTab; + + /* Code an OP_Transaction and OP_TableLock for
    . */ + iDb = sqlite3SchemaToIndex(db, pTab->pSchema); + sqlite3CodeVerifySchema(pParse, iDb); + sqlite3TableLock(pParse, iDb, pTab->tnum, 0, pTab->zName); + + assert(v); /* sqlite3GetVdbe() has always been previously called */ + if( nExpr==1 && pEList->a[0].pExpr->iColumn<0 ){ + /* The "x IN (SELECT rowid FROM table)" case */ + int iAddr = sqlite3VdbeAddOp0(v, OP_Once); + VdbeCoverage(v); + + sqlite3OpenTable(pParse, iTab, iDb, pTab, OP_OpenRead); + eType = IN_INDEX_ROWID; + ExplainQueryPlan((pParse, 0, + "USING ROWID SEARCH ON TABLE %s FOR IN-OPERATOR",pTab->zName)); + sqlite3VdbeJumpHere(v, iAddr); + }else{ + Index *pIdx; /* Iterator variable */ + int affinity_ok = 1; + int i; + + /* Check that the affinity that will be used to perform each + ** comparison is the same as the affinity of each column in table + ** on the RHS of the IN operator. If it not, it is not possible to + ** use any index of the RHS table. */ + for(i=0; ipLeft, i); + int iCol = pEList->a[i].pExpr->iColumn; + char idxaff = sqlite3TableColumnAffinity(pTab,iCol); /* RHS table */ + char cmpaff = sqlite3CompareAffinity(pLhs, idxaff); + testcase( cmpaff==SQLITE_AFF_BLOB ); + testcase( cmpaff==SQLITE_AFF_TEXT ); + switch( cmpaff ){ + case SQLITE_AFF_BLOB: + break; + case SQLITE_AFF_TEXT: + /* sqlite3CompareAffinity() only returns TEXT if one side or the + ** other has no affinity and the other side is TEXT. Hence, + ** the only way for cmpaff to be TEXT is for idxaff to be TEXT + ** and for the term on the LHS of the IN to have no affinity. */ + assert( idxaff==SQLITE_AFF_TEXT ); + break; + default: + affinity_ok = sqlite3IsNumericAffinity(idxaff); + } + } + + if( affinity_ok ){ + /* Search for an existing index that will work for this IN operator */ + for(pIdx=pTab->pIndex; pIdx && eType==0; pIdx=pIdx->pNext){ + Bitmask colUsed; /* Columns of the index used */ + Bitmask mCol; /* Mask for the current column */ + if( pIdx->nColumnpPartIdxWhere!=0 ) continue; + /* Maximum nColumn is BMS-2, not BMS-1, so that we can compute + ** BITMASK(nExpr) without overflowing */ + testcase( pIdx->nColumn==BMS-2 ); + testcase( pIdx->nColumn==BMS-1 ); + if( pIdx->nColumn>=BMS-1 ) continue; + if( mustBeUnique ){ + if( pIdx->nKeyCol>nExpr + ||(pIdx->nColumn>nExpr && !IsUniqueIndex(pIdx)) + ){ + continue; /* This index is not unique over the IN RHS columns */ + } + } + + colUsed = 0; /* Columns of index used so far */ + for(i=0; ipLeft, i); + Expr *pRhs = pEList->a[i].pExpr; + CollSeq *pReq = sqlite3BinaryCompareCollSeq(pParse, pLhs, pRhs); + int j; + + assert( pReq!=0 || pRhs->iColumn==XN_ROWID || pParse->nErr ); + for(j=0; jaiColumn[j]!=pRhs->iColumn ) continue; + assert( pIdx->azColl[j] ); + if( pReq!=0 && sqlite3StrICmp(pReq->zName, pIdx->azColl[j])!=0 ){ + continue; + } + break; + } + if( j==nExpr ) break; + mCol = MASKBIT(j); + if( mCol & colUsed ) break; /* Each column used only once */ + colUsed |= mCol; + if( aiMap ) aiMap[i] = j; + } + + assert( i==nExpr || colUsed!=(MASKBIT(nExpr)-1) ); + if( colUsed==(MASKBIT(nExpr)-1) ){ + /* If we reach this point, that means the index pIdx is usable */ + int iAddr = sqlite3VdbeAddOp0(v, OP_Once); VdbeCoverage(v); + ExplainQueryPlan((pParse, 0, + "USING INDEX %s FOR IN-OPERATOR",pIdx->zName)); + sqlite3VdbeAddOp3(v, OP_OpenRead, iTab, pIdx->tnum, iDb); + sqlite3VdbeSetP4KeyInfo(pParse, pIdx); + VdbeComment((v, "%s", pIdx->zName)); + assert( IN_INDEX_INDEX_DESC == IN_INDEX_INDEX_ASC+1 ); + eType = IN_INDEX_INDEX_ASC + pIdx->aSortOrder[0]; + + if( prRhsHasNull ){ +#ifdef SQLITE_ENABLE_COLUMN_USED_MASK + i64 mask = (1<nMem; + if( nExpr==1 ){ + sqlite3SetHasNullFlag(v, iTab, *prRhsHasNull); + } + } + sqlite3VdbeJumpHere(v, iAddr); + } + } /* End loop over indexes */ + } /* End if( affinity_ok ) */ + } /* End if not an rowid index */ + } /* End attempt to optimize using an index */ + + /* If no preexisting index is available for the IN clause + ** and IN_INDEX_NOOP is an allowed reply + ** and the RHS of the IN operator is a list, not a subquery + ** and the RHS is not constant or has two or fewer terms, + ** then it is not worth creating an ephemeral table to evaluate + ** the IN operator so return IN_INDEX_NOOP. + */ + if( eType==0 + && (inFlags & IN_INDEX_NOOP_OK) + && !ExprHasProperty(pX, EP_xIsSelect) + && (!sqlite3InRhsIsConstant(pX) || pX->x.pList->nExpr<=2) + ){ + eType = IN_INDEX_NOOP; + } + + if( eType==0 ){ + /* Could not find an existing table or index to use as the RHS b-tree. + ** We will have to generate an ephemeral table to do the job. + */ + u32 savedNQueryLoop = pParse->nQueryLoop; + int rMayHaveNull = 0; + eType = IN_INDEX_EPH; + if( inFlags & IN_INDEX_LOOP ){ + pParse->nQueryLoop = 0; + }else if( prRhsHasNull ){ + *prRhsHasNull = rMayHaveNull = ++pParse->nMem; + } + assert( pX->op==TK_IN ); + sqlite3CodeRhsOfIN(pParse, pX, iTab); + if( rMayHaveNull ){ + sqlite3SetHasNullFlag(v, iTab, rMayHaveNull); + } + pParse->nQueryLoop = savedNQueryLoop; + } + + if( aiMap && eType!=IN_INDEX_INDEX_ASC && eType!=IN_INDEX_INDEX_DESC ){ + int i, n; + n = sqlite3ExprVectorSize(pX->pLeft); + for(i=0; ipLeft; + int nVal = sqlite3ExprVectorSize(pLeft); + Select *pSelect = (pExpr->flags & EP_xIsSelect) ? pExpr->x.pSelect : 0; + char *zRet; + + assert( pExpr->op==TK_IN ); + zRet = sqlite3DbMallocRaw(pParse->db, nVal+1); + if( zRet ){ + int i; + for(i=0; ipEList->a[i].pExpr, a); + }else{ + zRet[i] = a; + } + } + zRet[nVal] = '\0'; + } + return zRet; +} +#endif + +#ifndef SQLITE_OMIT_SUBQUERY +/* +** Load the Parse object passed as the first argument with an error +** message of the form: +** +** "sub-select returns N columns - expected M" +*/ +SQLITE_PRIVATE void sqlite3SubselectError(Parse *pParse, int nActual, int nExpect){ + const char *zFmt = "sub-select returns %d columns - expected %d"; + sqlite3ErrorMsg(pParse, zFmt, nActual, nExpect); +} +#endif + +/* +** Expression pExpr is a vector that has been used in a context where +** it is not permitted. If pExpr is a sub-select vector, this routine +** loads the Parse object with a message of the form: +** +** "sub-select returns N columns - expected 1" +** +** Or, if it is a regular scalar vector: +** +** "row value misused" +*/ +SQLITE_PRIVATE void sqlite3VectorErrorMsg(Parse *pParse, Expr *pExpr){ +#ifndef SQLITE_OMIT_SUBQUERY + if( pExpr->flags & EP_xIsSelect ){ + sqlite3SubselectError(pParse, pExpr->x.pSelect->pEList->nExpr, 1); + }else +#endif + { + sqlite3ErrorMsg(pParse, "row value misused"); + } +} + +#ifndef SQLITE_OMIT_SUBQUERY +/* +** Generate code that will construct an ephemeral table containing all terms +** in the RHS of an IN operator. The IN operator can be in either of two +** forms: +** +** x IN (4,5,11) -- IN operator with list on right-hand side +** x IN (SELECT a FROM b) -- IN operator with subquery on the right +** +** The pExpr parameter is the IN operator. The cursor number for the +** constructed ephermeral table is returned. The first time the ephemeral +** table is computed, the cursor number is also stored in pExpr->iTable, +** however the cursor number returned might not be the same, as it might +** have been duplicated using OP_OpenDup. +** +** If the LHS expression ("x" in the examples) is a column value, or +** the SELECT statement returns a column value, then the affinity of that +** column is used to build the index keys. If both 'x' and the +** SELECT... statement are columns, then numeric affinity is used +** if either column has NUMERIC or INTEGER affinity. If neither +** 'x' nor the SELECT... statement are columns, then numeric affinity +** is used. +*/ +SQLITE_PRIVATE void sqlite3CodeRhsOfIN( + Parse *pParse, /* Parsing context */ + Expr *pExpr, /* The IN operator */ + int iTab /* Use this cursor number */ +){ + int addrOnce = 0; /* Address of the OP_Once instruction at top */ + int addr; /* Address of OP_OpenEphemeral instruction */ + Expr *pLeft; /* the LHS of the IN operator */ + KeyInfo *pKeyInfo = 0; /* Key information */ + int nVal; /* Size of vector pLeft */ + Vdbe *v; /* The prepared statement under construction */ + + v = pParse->pVdbe; + assert( v!=0 ); + + /* The evaluation of the IN must be repeated every time it + ** is encountered if any of the following is true: + ** + ** * The right-hand side is a correlated subquery + ** * The right-hand side is an expression list containing variables + ** * We are inside a trigger + ** + ** If all of the above are false, then we can compute the RHS just once + ** and reuse it many names. + */ + if( !ExprHasProperty(pExpr, EP_VarSelect) && pParse->iSelfTab==0 ){ + /* Reuse of the RHS is allowed */ + /* If this routine has already been coded, but the previous code + ** might not have been invoked yet, so invoke it now as a subroutine. + */ + if( ExprHasProperty(pExpr, EP_Subrtn) ){ + addrOnce = sqlite3VdbeAddOp0(v, OP_Once); VdbeCoverage(v); + if( ExprHasProperty(pExpr, EP_xIsSelect) ){ + ExplainQueryPlan((pParse, 0, "REUSE LIST SUBQUERY %d", + pExpr->x.pSelect->selId)); + } + sqlite3VdbeAddOp2(v, OP_Gosub, pExpr->y.sub.regReturn, + pExpr->y.sub.iAddr); + sqlite3VdbeAddOp2(v, OP_OpenDup, iTab, pExpr->iTable); + sqlite3VdbeJumpHere(v, addrOnce); + return; + } + + /* Begin coding the subroutine */ + ExprSetProperty(pExpr, EP_Subrtn); + pExpr->y.sub.regReturn = ++pParse->nMem; + pExpr->y.sub.iAddr = + sqlite3VdbeAddOp2(v, OP_Integer, 0, pExpr->y.sub.regReturn) + 1; + VdbeComment((v, "return address")); + + addrOnce = sqlite3VdbeAddOp0(v, OP_Once); VdbeCoverage(v); + } + + /* Check to see if this is a vector IN operator */ + pLeft = pExpr->pLeft; + nVal = sqlite3ExprVectorSize(pLeft); + + /* Construct the ephemeral table that will contain the content of + ** RHS of the IN operator. + */ + pExpr->iTable = iTab; + addr = sqlite3VdbeAddOp2(v, OP_OpenEphemeral, pExpr->iTable, nVal); +#ifdef SQLITE_ENABLE_EXPLAIN_COMMENTS + if( ExprHasProperty(pExpr, EP_xIsSelect) ){ + VdbeComment((v, "Result of SELECT %u", pExpr->x.pSelect->selId)); + }else{ + VdbeComment((v, "RHS of IN operator")); + } +#endif + pKeyInfo = sqlite3KeyInfoAlloc(pParse->db, nVal, 1); + + if( ExprHasProperty(pExpr, EP_xIsSelect) ){ + /* Case 1: expr IN (SELECT ...) + ** + ** Generate code to write the results of the select into the temporary + ** table allocated and opened above. + */ + Select *pSelect = pExpr->x.pSelect; + ExprList *pEList = pSelect->pEList; + + ExplainQueryPlan((pParse, 1, "%sLIST SUBQUERY %d", + addrOnce?"":"CORRELATED ", pSelect->selId + )); + /* If the LHS and RHS of the IN operator do not match, that + ** error will have been caught long before we reach this point. */ + if( ALWAYS(pEList->nExpr==nVal) ){ + SelectDest dest; + int i; + sqlite3SelectDestInit(&dest, SRT_Set, iTab); + dest.zAffSdst = exprINAffinity(pParse, pExpr); + pSelect->iLimit = 0; + testcase( pSelect->selFlags & SF_Distinct ); + testcase( pKeyInfo==0 ); /* Caused by OOM in sqlite3KeyInfoAlloc() */ + if( sqlite3Select(pParse, pSelect, &dest) ){ + sqlite3DbFree(pParse->db, dest.zAffSdst); + sqlite3KeyInfoUnref(pKeyInfo); + return; + } + sqlite3DbFree(pParse->db, dest.zAffSdst); + assert( pKeyInfo!=0 ); /* OOM will cause exit after sqlite3Select() */ + assert( pEList!=0 ); + assert( pEList->nExpr>0 ); + assert( sqlite3KeyInfoIsWriteable(pKeyInfo) ); + for(i=0; iaColl[i] = sqlite3BinaryCompareCollSeq( + pParse, p, pEList->a[i].pExpr + ); + } + } + }else if( ALWAYS(pExpr->x.pList!=0) ){ + /* Case 2: expr IN (exprlist) + ** + ** For each expression, build an index key from the evaluation and + ** store it in the temporary table. If is a column, then use + ** that columns affinity when building index keys. If is not + ** a column, use numeric affinity. + */ + char affinity; /* Affinity of the LHS of the IN */ + int i; + ExprList *pList = pExpr->x.pList; + struct ExprList_item *pItem; + int r1, r2; + affinity = sqlite3ExprAffinity(pLeft); + if( affinity<=SQLITE_AFF_NONE ){ + affinity = SQLITE_AFF_BLOB; + } + if( pKeyInfo ){ + assert( sqlite3KeyInfoIsWriteable(pKeyInfo) ); + pKeyInfo->aColl[0] = sqlite3ExprCollSeq(pParse, pExpr->pLeft); + } + + /* Loop through each expression in . */ + r1 = sqlite3GetTempReg(pParse); + r2 = sqlite3GetTempReg(pParse); + for(i=pList->nExpr, pItem=pList->a; i>0; i--, pItem++){ + Expr *pE2 = pItem->pExpr; + + /* If the expression is not constant then we will need to + ** disable the test that was generated above that makes sure + ** this code only executes once. Because for a non-constant + ** expression we need to rerun this code each time. + */ + if( addrOnce && !sqlite3ExprIsConstant(pE2) ){ + sqlite3VdbeChangeToNoop(v, addrOnce); + ExprClearProperty(pExpr, EP_Subrtn); + addrOnce = 0; + } + + /* Evaluate the expression and insert it into the temp table */ + sqlite3ExprCode(pParse, pE2, r1); + sqlite3VdbeAddOp4(v, OP_MakeRecord, r1, 1, r2, &affinity, 1); + sqlite3VdbeAddOp4Int(v, OP_IdxInsert, iTab, r2, r1, 1); + } + sqlite3ReleaseTempReg(pParse, r1); + sqlite3ReleaseTempReg(pParse, r2); + } + if( pKeyInfo ){ + sqlite3VdbeChangeP4(v, addr, (void *)pKeyInfo, P4_KEYINFO); + } + if( addrOnce ){ + sqlite3VdbeJumpHere(v, addrOnce); + /* Subroutine return */ + sqlite3VdbeAddOp1(v, OP_Return, pExpr->y.sub.regReturn); + sqlite3VdbeChangeP1(v, pExpr->y.sub.iAddr-1, sqlite3VdbeCurrentAddr(v)-1); + sqlite3ClearTempRegCache(pParse); + } +} +#endif /* SQLITE_OMIT_SUBQUERY */ + +/* +** Generate code for scalar subqueries used as a subquery expression +** or EXISTS operator: +** +** (SELECT a FROM b) -- subquery +** EXISTS (SELECT a FROM b) -- EXISTS subquery +** +** The pExpr parameter is the SELECT or EXISTS operator to be coded. +** +** Return the register that holds the result. For a multi-column SELECT, +** the result is stored in a contiguous array of registers and the +** return value is the register of the left-most result column. +** Return 0 if an error occurs. +*/ +#ifndef SQLITE_OMIT_SUBQUERY +SQLITE_PRIVATE int sqlite3CodeSubselect(Parse *pParse, Expr *pExpr){ + int addrOnce = 0; /* Address of OP_Once at top of subroutine */ + int rReg = 0; /* Register storing resulting */ + Select *pSel; /* SELECT statement to encode */ + SelectDest dest; /* How to deal with SELECT result */ + int nReg; /* Registers to allocate */ + Expr *pLimit; /* New limit expression */ + + Vdbe *v = pParse->pVdbe; + assert( v!=0 ); + testcase( pExpr->op==TK_EXISTS ); + testcase( pExpr->op==TK_SELECT ); + assert( pExpr->op==TK_EXISTS || pExpr->op==TK_SELECT ); + assert( ExprHasProperty(pExpr, EP_xIsSelect) ); + pSel = pExpr->x.pSelect; + + /* The evaluation of the EXISTS/SELECT must be repeated every time it + ** is encountered if any of the following is true: + ** + ** * The right-hand side is a correlated subquery + ** * The right-hand side is an expression list containing variables + ** * We are inside a trigger + ** + ** If all of the above are false, then we can run this code just once + ** save the results, and reuse the same result on subsequent invocations. + */ + if( !ExprHasProperty(pExpr, EP_VarSelect) ){ + /* If this routine has already been coded, then invoke it as a + ** subroutine. */ + if( ExprHasProperty(pExpr, EP_Subrtn) ){ + ExplainQueryPlan((pParse, 0, "REUSE SUBQUERY %d", pSel->selId)); + sqlite3VdbeAddOp2(v, OP_Gosub, pExpr->y.sub.regReturn, + pExpr->y.sub.iAddr); + return pExpr->iTable; + } + + /* Begin coding the subroutine */ + ExprSetProperty(pExpr, EP_Subrtn); + pExpr->y.sub.regReturn = ++pParse->nMem; + pExpr->y.sub.iAddr = + sqlite3VdbeAddOp2(v, OP_Integer, 0, pExpr->y.sub.regReturn) + 1; + VdbeComment((v, "return address")); + + addrOnce = sqlite3VdbeAddOp0(v, OP_Once); VdbeCoverage(v); + } + + /* For a SELECT, generate code to put the values for all columns of + ** the first row into an array of registers and return the index of + ** the first register. + ** + ** If this is an EXISTS, write an integer 0 (not exists) or 1 (exists) + ** into a register and return that register number. + ** + ** In both cases, the query is augmented with "LIMIT 1". Any + ** preexisting limit is discarded in place of the new LIMIT 1. + */ + ExplainQueryPlan((pParse, 1, "%sSCALAR SUBQUERY %d", + addrOnce?"":"CORRELATED ", pSel->selId)); + nReg = pExpr->op==TK_SELECT ? pSel->pEList->nExpr : 1; + sqlite3SelectDestInit(&dest, 0, pParse->nMem+1); + pParse->nMem += nReg; + if( pExpr->op==TK_SELECT ){ + dest.eDest = SRT_Mem; + dest.iSdst = dest.iSDParm; + dest.nSdst = nReg; + sqlite3VdbeAddOp3(v, OP_Null, 0, dest.iSDParm, dest.iSDParm+nReg-1); + VdbeComment((v, "Init subquery result")); + }else{ + dest.eDest = SRT_Exists; + sqlite3VdbeAddOp2(v, OP_Integer, 0, dest.iSDParm); + VdbeComment((v, "Init EXISTS result")); + } + if( pSel->pLimit ){ + /* The subquery already has a limit. If the pre-existing limit is X + ** then make the new limit X<>0 so that the new limit is either 1 or 0 */ + sqlite3 *db = pParse->db; + pLimit = sqlite3Expr(db, TK_INTEGER, "0"); + if( pLimit ){ + pLimit->affExpr = SQLITE_AFF_NUMERIC; + pLimit = sqlite3PExpr(pParse, TK_NE, + sqlite3ExprDup(db, pSel->pLimit->pLeft, 0), pLimit); + } + sqlite3ExprDelete(db, pSel->pLimit->pLeft); + pSel->pLimit->pLeft = pLimit; + }else{ + /* If there is no pre-existing limit add a limit of 1 */ + pLimit = sqlite3Expr(pParse->db, TK_INTEGER, "1"); + pSel->pLimit = sqlite3PExpr(pParse, TK_LIMIT, pLimit, 0); + } + pSel->iLimit = 0; + if( sqlite3Select(pParse, pSel, &dest) ){ + return 0; + } + pExpr->iTable = rReg = dest.iSDParm; + ExprSetVVAProperty(pExpr, EP_NoReduce); + if( addrOnce ){ + sqlite3VdbeJumpHere(v, addrOnce); + + /* Subroutine return */ + sqlite3VdbeAddOp1(v, OP_Return, pExpr->y.sub.regReturn); + sqlite3VdbeChangeP1(v, pExpr->y.sub.iAddr-1, sqlite3VdbeCurrentAddr(v)-1); + sqlite3ClearTempRegCache(pParse); + } + + return rReg; +} +#endif /* SQLITE_OMIT_SUBQUERY */ + +#ifndef SQLITE_OMIT_SUBQUERY +/* +** Expr pIn is an IN(...) expression. This function checks that the +** sub-select on the RHS of the IN() operator has the same number of +** columns as the vector on the LHS. Or, if the RHS of the IN() is not +** a sub-query, that the LHS is a vector of size 1. +*/ +SQLITE_PRIVATE int sqlite3ExprCheckIN(Parse *pParse, Expr *pIn){ + int nVector = sqlite3ExprVectorSize(pIn->pLeft); + if( (pIn->flags & EP_xIsSelect) ){ + if( nVector!=pIn->x.pSelect->pEList->nExpr ){ + sqlite3SubselectError(pParse, pIn->x.pSelect->pEList->nExpr, nVector); + return 1; + } + }else if( nVector!=1 ){ + sqlite3VectorErrorMsg(pParse, pIn->pLeft); + return 1; + } + return 0; +} +#endif + +#ifndef SQLITE_OMIT_SUBQUERY +/* +** Generate code for an IN expression. +** +** x IN (SELECT ...) +** x IN (value, value, ...) +** +** The left-hand side (LHS) is a scalar or vector expression. The +** right-hand side (RHS) is an array of zero or more scalar values, or a +** subquery. If the RHS is a subquery, the number of result columns must +** match the number of columns in the vector on the LHS. If the RHS is +** a list of values, the LHS must be a scalar. +** +** The IN operator is true if the LHS value is contained within the RHS. +** The result is false if the LHS is definitely not in the RHS. The +** result is NULL if the presence of the LHS in the RHS cannot be +** determined due to NULLs. +** +** This routine generates code that jumps to destIfFalse if the LHS is not +** contained within the RHS. If due to NULLs we cannot determine if the LHS +** is contained in the RHS then jump to destIfNull. If the LHS is contained +** within the RHS then fall through. +** +** See the separate in-operator.md documentation file in the canonical +** SQLite source tree for additional information. +*/ +static void sqlite3ExprCodeIN( + Parse *pParse, /* Parsing and code generating context */ + Expr *pExpr, /* The IN expression */ + int destIfFalse, /* Jump here if LHS is not contained in the RHS */ + int destIfNull /* Jump here if the results are unknown due to NULLs */ +){ + int rRhsHasNull = 0; /* Register that is true if RHS contains NULL values */ + int eType; /* Type of the RHS */ + int rLhs; /* Register(s) holding the LHS values */ + int rLhsOrig; /* LHS values prior to reordering by aiMap[] */ + Vdbe *v; /* Statement under construction */ + int *aiMap = 0; /* Map from vector field to index column */ + char *zAff = 0; /* Affinity string for comparisons */ + int nVector; /* Size of vectors for this IN operator */ + int iDummy; /* Dummy parameter to exprCodeVector() */ + Expr *pLeft; /* The LHS of the IN operator */ + int i; /* loop counter */ + int destStep2; /* Where to jump when NULLs seen in step 2 */ + int destStep6 = 0; /* Start of code for Step 6 */ + int addrTruthOp; /* Address of opcode that determines the IN is true */ + int destNotNull; /* Jump here if a comparison is not true in step 6 */ + int addrTop; /* Top of the step-6 loop */ + int iTab = 0; /* Index to use */ + + pLeft = pExpr->pLeft; + if( sqlite3ExprCheckIN(pParse, pExpr) ) return; + zAff = exprINAffinity(pParse, pExpr); + nVector = sqlite3ExprVectorSize(pExpr->pLeft); + aiMap = (int*)sqlite3DbMallocZero( + pParse->db, nVector*(sizeof(int) + sizeof(char)) + 1 + ); + if( pParse->db->mallocFailed ) goto sqlite3ExprCodeIN_oom_error; + + /* Attempt to compute the RHS. After this step, if anything other than + ** IN_INDEX_NOOP is returned, the table opened with cursor iTab + ** contains the values that make up the RHS. If IN_INDEX_NOOP is returned, + ** the RHS has not yet been coded. */ + v = pParse->pVdbe; + assert( v!=0 ); /* OOM detected prior to this routine */ + VdbeNoopComment((v, "begin IN expr")); + eType = sqlite3FindInIndex(pParse, pExpr, + IN_INDEX_MEMBERSHIP | IN_INDEX_NOOP_OK, + destIfFalse==destIfNull ? 0 : &rRhsHasNull, + aiMap, &iTab); + + assert( pParse->nErr || nVector==1 || eType==IN_INDEX_EPH + || eType==IN_INDEX_INDEX_ASC || eType==IN_INDEX_INDEX_DESC + ); +#ifdef SQLITE_DEBUG + /* Confirm that aiMap[] contains nVector integer values between 0 and + ** nVector-1. */ + for(i=0; i from " IN (...)". If the LHS is a + ** vector, then it is stored in an array of nVector registers starting + ** at r1. + ** + ** sqlite3FindInIndex() might have reordered the fields of the LHS vector + ** so that the fields are in the same order as an existing index. The + ** aiMap[] array contains a mapping from the original LHS field order to + ** the field order that matches the RHS index. + */ + rLhsOrig = exprCodeVector(pParse, pLeft, &iDummy); + for(i=0; ix.pList; + CollSeq *pColl = sqlite3ExprCollSeq(pParse, pExpr->pLeft); + int labelOk = sqlite3VdbeMakeLabel(pParse); + int r2, regToFree; + int regCkNull = 0; + int ii; + int bLhsReal; /* True if the LHS of the IN has REAL affinity */ + assert( !ExprHasProperty(pExpr, EP_xIsSelect) ); + if( destIfNull!=destIfFalse ){ + regCkNull = sqlite3GetTempReg(pParse); + sqlite3VdbeAddOp3(v, OP_BitAnd, rLhs, rLhs, regCkNull); + } + bLhsReal = sqlite3ExprAffinity(pExpr->pLeft)==SQLITE_AFF_REAL; + for(ii=0; iinExpr; ii++){ + if( bLhsReal ){ + r2 = regToFree = sqlite3GetTempReg(pParse); + sqlite3ExprCode(pParse, pList->a[ii].pExpr, r2); + sqlite3VdbeAddOp4(v, OP_Affinity, r2, 1, 0, "E", P4_STATIC); + }else{ + r2 = sqlite3ExprCodeTemp(pParse, pList->a[ii].pExpr, ®ToFree); + } + if( regCkNull && sqlite3ExprCanBeNull(pList->a[ii].pExpr) ){ + sqlite3VdbeAddOp3(v, OP_BitAnd, regCkNull, r2, regCkNull); + } + if( iinExpr-1 || destIfNull!=destIfFalse ){ + sqlite3VdbeAddOp4(v, OP_Eq, rLhs, labelOk, r2, + (void*)pColl, P4_COLLSEQ); + VdbeCoverageIf(v, iinExpr-1); + VdbeCoverageIf(v, ii==pList->nExpr-1); + sqlite3VdbeChangeP5(v, zAff[0]); + }else{ + assert( destIfNull==destIfFalse ); + sqlite3VdbeAddOp4(v, OP_Ne, rLhs, destIfFalse, r2, + (void*)pColl, P4_COLLSEQ); VdbeCoverage(v); + sqlite3VdbeChangeP5(v, zAff[0] | SQLITE_JUMPIFNULL); + } + sqlite3ReleaseTempReg(pParse, regToFree); + } + if( regCkNull ){ + sqlite3VdbeAddOp2(v, OP_IsNull, regCkNull, destIfNull); VdbeCoverage(v); + sqlite3VdbeGoto(v, destIfFalse); + } + sqlite3VdbeResolveLabel(v, labelOk); + sqlite3ReleaseTempReg(pParse, regCkNull); + goto sqlite3ExprCodeIN_finished; + } + + /* Step 2: Check to see if the LHS contains any NULL columns. If the + ** LHS does contain NULLs then the result must be either FALSE or NULL. + ** We will then skip the binary search of the RHS. + */ + if( destIfNull==destIfFalse ){ + destStep2 = destIfFalse; + }else{ + destStep2 = destStep6 = sqlite3VdbeMakeLabel(pParse); + } + for(i=0; ipLeft, i); + if( sqlite3ExprCanBeNull(p) ){ + sqlite3VdbeAddOp2(v, OP_IsNull, rLhs+i, destStep2); + VdbeCoverage(v); + } + } + + /* Step 3. The LHS is now known to be non-NULL. Do the binary search + ** of the RHS using the LHS as a probe. If found, the result is + ** true. + */ + if( eType==IN_INDEX_ROWID ){ + /* In this case, the RHS is the ROWID of table b-tree and so we also + ** know that the RHS is non-NULL. Hence, we combine steps 3 and 4 + ** into a single opcode. */ + sqlite3VdbeAddOp3(v, OP_SeekRowid, iTab, destIfFalse, rLhs); + VdbeCoverage(v); + addrTruthOp = sqlite3VdbeAddOp0(v, OP_Goto); /* Return True */ + }else{ + sqlite3VdbeAddOp4(v, OP_Affinity, rLhs, nVector, 0, zAff, nVector); + if( destIfFalse==destIfNull ){ + /* Combine Step 3 and Step 5 into a single opcode */ + sqlite3VdbeAddOp4Int(v, OP_NotFound, iTab, destIfFalse, + rLhs, nVector); VdbeCoverage(v); + goto sqlite3ExprCodeIN_finished; + } + /* Ordinary Step 3, for the case where FALSE and NULL are distinct */ + addrTruthOp = sqlite3VdbeAddOp4Int(v, OP_Found, iTab, 0, + rLhs, nVector); VdbeCoverage(v); + } + + /* Step 4. If the RHS is known to be non-NULL and we did not find + ** an match on the search above, then the result must be FALSE. + */ + if( rRhsHasNull && nVector==1 ){ + sqlite3VdbeAddOp2(v, OP_NotNull, rRhsHasNull, destIfFalse); + VdbeCoverage(v); + } + + /* Step 5. If we do not care about the difference between NULL and + ** FALSE, then just return false. + */ + if( destIfFalse==destIfNull ) sqlite3VdbeGoto(v, destIfFalse); + + /* Step 6: Loop through rows of the RHS. Compare each row to the LHS. + ** If any comparison is NULL, then the result is NULL. If all + ** comparisons are FALSE then the final result is FALSE. + ** + ** For a scalar LHS, it is sufficient to check just the first row + ** of the RHS. + */ + if( destStep6 ) sqlite3VdbeResolveLabel(v, destStep6); + addrTop = sqlite3VdbeAddOp2(v, OP_Rewind, iTab, destIfFalse); + VdbeCoverage(v); + if( nVector>1 ){ + destNotNull = sqlite3VdbeMakeLabel(pParse); + }else{ + /* For nVector==1, combine steps 6 and 7 by immediately returning + ** FALSE if the first comparison is not NULL */ + destNotNull = destIfFalse; + } + for(i=0; i1 ){ + sqlite3VdbeResolveLabel(v, destNotNull); + sqlite3VdbeAddOp2(v, OP_Next, iTab, addrTop+1); + VdbeCoverage(v); + + /* Step 7: If we reach this point, we know that the result must + ** be false. */ + sqlite3VdbeAddOp2(v, OP_Goto, 0, destIfFalse); + } + + /* Jumps here in order to return true. */ + sqlite3VdbeJumpHere(v, addrTruthOp); + +sqlite3ExprCodeIN_finished: + if( rLhs!=rLhsOrig ) sqlite3ReleaseTempReg(pParse, rLhs); + VdbeComment((v, "end IN expr")); +sqlite3ExprCodeIN_oom_error: + sqlite3DbFree(pParse->db, aiMap); + sqlite3DbFree(pParse->db, zAff); +} +#endif /* SQLITE_OMIT_SUBQUERY */ + +#ifndef SQLITE_OMIT_FLOATING_POINT +/* +** Generate an instruction that will put the floating point +** value described by z[0..n-1] into register iMem. +** +** The z[] string will probably not be zero-terminated. But the +** z[n] character is guaranteed to be something that does not look +** like the continuation of the number. +*/ +static void codeReal(Vdbe *v, const char *z, int negateFlag, int iMem){ + if( ALWAYS(z!=0) ){ + double value; + sqlite3AtoF(z, &value, sqlite3Strlen30(z), SQLITE_UTF8); + assert( !sqlite3IsNaN(value) ); /* The new AtoF never returns NaN */ + if( negateFlag ) value = -value; + sqlite3VdbeAddOp4Dup8(v, OP_Real, 0, iMem, 0, (u8*)&value, P4_REAL); + } +} +#endif + + +/* +** Generate an instruction that will put the integer describe by +** text z[0..n-1] into register iMem. +** +** Expr.u.zToken is always UTF8 and zero-terminated. +*/ +static void codeInteger(Parse *pParse, Expr *pExpr, int negFlag, int iMem){ + Vdbe *v = pParse->pVdbe; + if( pExpr->flags & EP_IntValue ){ + int i = pExpr->u.iValue; + assert( i>=0 ); + if( negFlag ) i = -i; + sqlite3VdbeAddOp2(v, OP_Integer, i, iMem); + }else{ + int c; + i64 value; + const char *z = pExpr->u.zToken; + assert( z!=0 ); + c = sqlite3DecOrHexToI64(z, &value); + if( (c==3 && !negFlag) || (c==2) || (negFlag && value==SMALLEST_INT64)){ +#ifdef SQLITE_OMIT_FLOATING_POINT + sqlite3ErrorMsg(pParse, "oversized integer: %s%s", negFlag ? "-" : "", z); +#else +#ifndef SQLITE_OMIT_HEX_INTEGER + if( sqlite3_strnicmp(z,"0x",2)==0 ){ + sqlite3ErrorMsg(pParse, "hex literal too big: %s%s", negFlag?"-":"",z); + }else +#endif + { + codeReal(v, z, negFlag, iMem); + } +#endif + }else{ + if( negFlag ){ value = c==3 ? SMALLEST_INT64 : -value; } + sqlite3VdbeAddOp4Dup8(v, OP_Int64, 0, iMem, 0, (u8*)&value, P4_INT64); + } + } +} + + +/* Generate code that will load into register regOut a value that is +** appropriate for the iIdxCol-th column of index pIdx. +*/ +SQLITE_PRIVATE void sqlite3ExprCodeLoadIndexColumn( + Parse *pParse, /* The parsing context */ + Index *pIdx, /* The index whose column is to be loaded */ + int iTabCur, /* Cursor pointing to a table row */ + int iIdxCol, /* The column of the index to be loaded */ + int regOut /* Store the index column value in this register */ +){ + i16 iTabCol = pIdx->aiColumn[iIdxCol]; + if( iTabCol==XN_EXPR ){ + assert( pIdx->aColExpr ); + assert( pIdx->aColExpr->nExpr>iIdxCol ); + pParse->iSelfTab = iTabCur + 1; + sqlite3ExprCodeCopy(pParse, pIdx->aColExpr->a[iIdxCol].pExpr, regOut); + pParse->iSelfTab = 0; + }else{ + sqlite3ExprCodeGetColumnOfTable(pParse->pVdbe, pIdx->pTable, iTabCur, + iTabCol, regOut); + } +} + +/* +** Generate code to extract the value of the iCol-th column of a table. +*/ +SQLITE_PRIVATE void sqlite3ExprCodeGetColumnOfTable( + Vdbe *v, /* The VDBE under construction */ + Table *pTab, /* The table containing the value */ + int iTabCur, /* The table cursor. Or the PK cursor for WITHOUT ROWID */ + int iCol, /* Index of the column to extract */ + int regOut /* Extract the value into this register */ +){ + if( pTab==0 ){ + sqlite3VdbeAddOp3(v, OP_Column, iTabCur, iCol, regOut); + return; + } + if( iCol<0 || iCol==pTab->iPKey ){ + sqlite3VdbeAddOp2(v, OP_Rowid, iTabCur, regOut); + }else{ + int op = IsVirtual(pTab) ? OP_VColumn : OP_Column; + int x = iCol; + if( !HasRowid(pTab) && !IsVirtual(pTab) ){ + x = sqlite3ColumnOfIndex(sqlite3PrimaryKeyIndex(pTab), iCol); + } + sqlite3VdbeAddOp3(v, op, iTabCur, x, regOut); + } + if( iCol>=0 ){ + sqlite3ColumnDefault(v, pTab, iCol, regOut); + } +} + +/* +** Generate code that will extract the iColumn-th column from +** table pTab and store the column value in register iReg. +** +** There must be an open cursor to pTab in iTable when this routine +** is called. If iColumn<0 then code is generated that extracts the rowid. +*/ +SQLITE_PRIVATE int sqlite3ExprCodeGetColumn( + Parse *pParse, /* Parsing and code generating context */ + Table *pTab, /* Description of the table we are reading from */ + int iColumn, /* Index of the table column */ + int iTable, /* The cursor pointing to the table */ + int iReg, /* Store results here */ + u8 p5 /* P5 value for OP_Column + FLAGS */ +){ + Vdbe *v = pParse->pVdbe; + assert( v!=0 ); + sqlite3ExprCodeGetColumnOfTable(v, pTab, iTable, iColumn, iReg); + if( p5 ){ + sqlite3VdbeChangeP5(v, p5); + } + return iReg; +} + +/* +** Generate code to move content from registers iFrom...iFrom+nReg-1 +** over to iTo..iTo+nReg-1. +*/ +SQLITE_PRIVATE void sqlite3ExprCodeMove(Parse *pParse, int iFrom, int iTo, int nReg){ + assert( iFrom>=iTo+nReg || iFrom+nReg<=iTo ); + sqlite3VdbeAddOp3(pParse->pVdbe, OP_Move, iFrom, iTo, nReg); +} + +/* +** Convert a scalar expression node to a TK_REGISTER referencing +** register iReg. The caller must ensure that iReg already contains +** the correct value for the expression. +*/ +static void exprToRegister(Expr *pExpr, int iReg){ + Expr *p = sqlite3ExprSkipCollateAndLikely(pExpr); + p->op2 = p->op; + p->op = TK_REGISTER; + p->iTable = iReg; + ExprClearProperty(p, EP_Skip); +} + +/* +** Evaluate an expression (either a vector or a scalar expression) and store +** the result in continguous temporary registers. Return the index of +** the first register used to store the result. +** +** If the returned result register is a temporary scalar, then also write +** that register number into *piFreeable. If the returned result register +** is not a temporary or if the expression is a vector set *piFreeable +** to 0. +*/ +static int exprCodeVector(Parse *pParse, Expr *p, int *piFreeable){ + int iResult; + int nResult = sqlite3ExprVectorSize(p); + if( nResult==1 ){ + iResult = sqlite3ExprCodeTemp(pParse, p, piFreeable); + }else{ + *piFreeable = 0; + if( p->op==TK_SELECT ){ +#if SQLITE_OMIT_SUBQUERY + iResult = 0; +#else + iResult = sqlite3CodeSubselect(pParse, p); +#endif + }else{ + int i; + iResult = pParse->nMem+1; + pParse->nMem += nResult; + for(i=0; ix.pList->a[i].pExpr, i+iResult); + } + } + } + return iResult; +} + + +/* +** Generate code into the current Vdbe to evaluate the given +** expression. Attempt to store the results in register "target". +** Return the register where results are stored. +** +** With this routine, there is no guarantee that results will +** be stored in target. The result might be stored in some other +** register if it is convenient to do so. The calling function +** must check the return code and move the results to the desired +** register. +*/ +SQLITE_PRIVATE int sqlite3ExprCodeTarget(Parse *pParse, Expr *pExpr, int target){ + Vdbe *v = pParse->pVdbe; /* The VM under construction */ + int op; /* The opcode being coded */ + int inReg = target; /* Results stored in register inReg */ + int regFree1 = 0; /* If non-zero free this temporary register */ + int regFree2 = 0; /* If non-zero free this temporary register */ + int r1, r2; /* Various register numbers */ + Expr tempX; /* Temporary expression node */ + int p5 = 0; + + assert( target>0 && target<=pParse->nMem ); + if( v==0 ){ + assert( pParse->db->mallocFailed ); + return 0; + } + +expr_code_doover: + if( pExpr==0 ){ + op = TK_NULL; + }else{ + op = pExpr->op; + } + switch( op ){ + case TK_AGG_COLUMN: { + AggInfo *pAggInfo = pExpr->pAggInfo; + struct AggInfo_col *pCol = &pAggInfo->aCol[pExpr->iAgg]; + if( !pAggInfo->directMode ){ + assert( pCol->iMem>0 ); + return pCol->iMem; + }else if( pAggInfo->useSortingIdx ){ + sqlite3VdbeAddOp3(v, OP_Column, pAggInfo->sortingIdxPTab, + pCol->iSorterColumn, target); + return target; + } + /* Otherwise, fall thru into the TK_COLUMN case */ + } + case TK_COLUMN: { + int iTab = pExpr->iTable; + if( ExprHasProperty(pExpr, EP_FixedCol) ){ + /* This COLUMN expression is really a constant due to WHERE clause + ** constraints, and that constant is coded by the pExpr->pLeft + ** expresssion. However, make sure the constant has the correct + ** datatype by applying the Affinity of the table column to the + ** constant. + */ + int iReg = sqlite3ExprCodeTarget(pParse, pExpr->pLeft,target); + int aff = sqlite3TableColumnAffinity(pExpr->y.pTab, pExpr->iColumn); + if( aff>SQLITE_AFF_BLOB ){ + static const char zAff[] = "B\000C\000D\000E"; + assert( SQLITE_AFF_BLOB=='A' ); + assert( SQLITE_AFF_TEXT=='B' ); + if( iReg!=target ){ + sqlite3VdbeAddOp2(v, OP_SCopy, iReg, target); + iReg = target; + } + sqlite3VdbeAddOp4(v, OP_Affinity, iReg, 1, 0, + &zAff[(aff-'B')*2], P4_STATIC); + } + return iReg; + } + if( iTab<0 ){ + if( pParse->iSelfTab<0 ){ + /* Generating CHECK constraints or inserting into partial index */ + assert( pExpr->y.pTab!=0 ); + assert( pExpr->iColumn>=XN_ROWID ); + assert( pExpr->iColumny.pTab->nCol ); + if( pExpr->iColumn>=0 + && pExpr->y.pTab->aCol[pExpr->iColumn].affinity==SQLITE_AFF_REAL + ){ + sqlite3VdbeAddOp2(v, OP_SCopy, pExpr->iColumn - pParse->iSelfTab, + target); + sqlite3VdbeAddOp1(v, OP_RealAffinity, target); + return target; + }else{ + return pExpr->iColumn - pParse->iSelfTab; + } + }else{ + /* Coding an expression that is part of an index where column names + ** in the index refer to the table to which the index belongs */ + iTab = pParse->iSelfTab - 1; + } + } + return sqlite3ExprCodeGetColumn(pParse, pExpr->y.pTab, + pExpr->iColumn, iTab, target, + pExpr->op2); + } + case TK_INTEGER: { + codeInteger(pParse, pExpr, 0, target); + return target; + } + case TK_TRUEFALSE: { + sqlite3VdbeAddOp2(v, OP_Integer, sqlite3ExprTruthValue(pExpr), target); + return target; + } +#ifndef SQLITE_OMIT_FLOATING_POINT + case TK_FLOAT: { + assert( !ExprHasProperty(pExpr, EP_IntValue) ); + codeReal(v, pExpr->u.zToken, 0, target); + return target; + } +#endif + case TK_STRING: { + assert( !ExprHasProperty(pExpr, EP_IntValue) ); + sqlite3VdbeLoadString(v, target, pExpr->u.zToken); + return target; + } + case TK_NULL: { + sqlite3VdbeAddOp2(v, OP_Null, 0, target); + return target; + } +#ifndef SQLITE_OMIT_BLOB_LITERAL + case TK_BLOB: { + int n; + const char *z; + char *zBlob; + assert( !ExprHasProperty(pExpr, EP_IntValue) ); + assert( pExpr->u.zToken[0]=='x' || pExpr->u.zToken[0]=='X' ); + assert( pExpr->u.zToken[1]=='\'' ); + z = &pExpr->u.zToken[2]; + n = sqlite3Strlen30(z) - 1; + assert( z[n]=='\'' ); + zBlob = sqlite3HexToBlob(sqlite3VdbeDb(v), z, n); + sqlite3VdbeAddOp4(v, OP_Blob, n/2, target, 0, zBlob, P4_DYNAMIC); + return target; + } +#endif + case TK_VARIABLE: { + assert( !ExprHasProperty(pExpr, EP_IntValue) ); + assert( pExpr->u.zToken!=0 ); + assert( pExpr->u.zToken[0]!=0 ); + sqlite3VdbeAddOp2(v, OP_Variable, pExpr->iColumn, target); + if( pExpr->u.zToken[1]!=0 ){ + const char *z = sqlite3VListNumToName(pParse->pVList, pExpr->iColumn); + assert( pExpr->u.zToken[0]=='?' || strcmp(pExpr->u.zToken, z)==0 ); + pParse->pVList[0] = 0; /* Indicate VList may no longer be enlarged */ + sqlite3VdbeAppendP4(v, (char*)z, P4_STATIC); + } + return target; + } + case TK_REGISTER: { + return pExpr->iTable; + } +#ifndef SQLITE_OMIT_CAST + case TK_CAST: { + /* Expressions of the form: CAST(pLeft AS token) */ + inReg = sqlite3ExprCodeTarget(pParse, pExpr->pLeft, target); + if( inReg!=target ){ + sqlite3VdbeAddOp2(v, OP_SCopy, inReg, target); + inReg = target; + } + sqlite3VdbeAddOp2(v, OP_Cast, target, + sqlite3AffinityType(pExpr->u.zToken, 0)); + return inReg; + } +#endif /* SQLITE_OMIT_CAST */ + case TK_IS: + case TK_ISNOT: + op = (op==TK_IS) ? TK_EQ : TK_NE; + p5 = SQLITE_NULLEQ; + /* fall-through */ + case TK_LT: + case TK_LE: + case TK_GT: + case TK_GE: + case TK_NE: + case TK_EQ: { + Expr *pLeft = pExpr->pLeft; + if( sqlite3ExprIsVector(pLeft) ){ + codeVectorCompare(pParse, pExpr, target, op, p5); + }else{ + r1 = sqlite3ExprCodeTemp(pParse, pLeft, ®Free1); + r2 = sqlite3ExprCodeTemp(pParse, pExpr->pRight, ®Free2); + codeCompare(pParse, pLeft, pExpr->pRight, op, + r1, r2, inReg, SQLITE_STOREP2 | p5); + assert(TK_LT==OP_Lt); testcase(op==OP_Lt); VdbeCoverageIf(v,op==OP_Lt); + assert(TK_LE==OP_Le); testcase(op==OP_Le); VdbeCoverageIf(v,op==OP_Le); + assert(TK_GT==OP_Gt); testcase(op==OP_Gt); VdbeCoverageIf(v,op==OP_Gt); + assert(TK_GE==OP_Ge); testcase(op==OP_Ge); VdbeCoverageIf(v,op==OP_Ge); + assert(TK_EQ==OP_Eq); testcase(op==OP_Eq); VdbeCoverageIf(v,op==OP_Eq); + assert(TK_NE==OP_Ne); testcase(op==OP_Ne); VdbeCoverageIf(v,op==OP_Ne); + testcase( regFree1==0 ); + testcase( regFree2==0 ); + } + break; + } + case TK_AND: + case TK_OR: + case TK_PLUS: + case TK_STAR: + case TK_MINUS: + case TK_REM: + case TK_BITAND: + case TK_BITOR: + case TK_SLASH: + case TK_LSHIFT: + case TK_RSHIFT: + case TK_CONCAT: { + assert( TK_AND==OP_And ); testcase( op==TK_AND ); + assert( TK_OR==OP_Or ); testcase( op==TK_OR ); + assert( TK_PLUS==OP_Add ); testcase( op==TK_PLUS ); + assert( TK_MINUS==OP_Subtract ); testcase( op==TK_MINUS ); + assert( TK_REM==OP_Remainder ); testcase( op==TK_REM ); + assert( TK_BITAND==OP_BitAnd ); testcase( op==TK_BITAND ); + assert( TK_BITOR==OP_BitOr ); testcase( op==TK_BITOR ); + assert( TK_SLASH==OP_Divide ); testcase( op==TK_SLASH ); + assert( TK_LSHIFT==OP_ShiftLeft ); testcase( op==TK_LSHIFT ); + assert( TK_RSHIFT==OP_ShiftRight ); testcase( op==TK_RSHIFT ); + assert( TK_CONCAT==OP_Concat ); testcase( op==TK_CONCAT ); + r1 = sqlite3ExprCodeTemp(pParse, pExpr->pLeft, ®Free1); + r2 = sqlite3ExprCodeTemp(pParse, pExpr->pRight, ®Free2); + sqlite3VdbeAddOp3(v, op, r2, r1, target); + testcase( regFree1==0 ); + testcase( regFree2==0 ); + break; + } + case TK_UMINUS: { + Expr *pLeft = pExpr->pLeft; + assert( pLeft ); + if( pLeft->op==TK_INTEGER ){ + codeInteger(pParse, pLeft, 1, target); + return target; +#ifndef SQLITE_OMIT_FLOATING_POINT + }else if( pLeft->op==TK_FLOAT ){ + assert( !ExprHasProperty(pExpr, EP_IntValue) ); + codeReal(v, pLeft->u.zToken, 1, target); + return target; +#endif + }else{ + tempX.op = TK_INTEGER; + tempX.flags = EP_IntValue|EP_TokenOnly; + tempX.u.iValue = 0; + r1 = sqlite3ExprCodeTemp(pParse, &tempX, ®Free1); + r2 = sqlite3ExprCodeTemp(pParse, pExpr->pLeft, ®Free2); + sqlite3VdbeAddOp3(v, OP_Subtract, r2, r1, target); + testcase( regFree2==0 ); + } + break; + } + case TK_BITNOT: + case TK_NOT: { + assert( TK_BITNOT==OP_BitNot ); testcase( op==TK_BITNOT ); + assert( TK_NOT==OP_Not ); testcase( op==TK_NOT ); + r1 = sqlite3ExprCodeTemp(pParse, pExpr->pLeft, ®Free1); + testcase( regFree1==0 ); + sqlite3VdbeAddOp2(v, op, r1, inReg); + break; + } + case TK_TRUTH: { + int isTrue; /* IS TRUE or IS NOT TRUE */ + int bNormal; /* IS TRUE or IS FALSE */ + r1 = sqlite3ExprCodeTemp(pParse, pExpr->pLeft, ®Free1); + testcase( regFree1==0 ); + isTrue = sqlite3ExprTruthValue(pExpr->pRight); + bNormal = pExpr->op2==TK_IS; + testcase( isTrue && bNormal); + testcase( !isTrue && bNormal); + sqlite3VdbeAddOp4Int(v, OP_IsTrue, r1, inReg, !isTrue, isTrue ^ bNormal); + break; + } + case TK_ISNULL: + case TK_NOTNULL: { + int addr; + assert( TK_ISNULL==OP_IsNull ); testcase( op==TK_ISNULL ); + assert( TK_NOTNULL==OP_NotNull ); testcase( op==TK_NOTNULL ); + sqlite3VdbeAddOp2(v, OP_Integer, 1, target); + r1 = sqlite3ExprCodeTemp(pParse, pExpr->pLeft, ®Free1); + testcase( regFree1==0 ); + addr = sqlite3VdbeAddOp1(v, op, r1); + VdbeCoverageIf(v, op==TK_ISNULL); + VdbeCoverageIf(v, op==TK_NOTNULL); + sqlite3VdbeAddOp2(v, OP_Integer, 0, target); + sqlite3VdbeJumpHere(v, addr); + break; + } + case TK_AGG_FUNCTION: { + AggInfo *pInfo = pExpr->pAggInfo; + if( pInfo==0 ){ + assert( !ExprHasProperty(pExpr, EP_IntValue) ); + sqlite3ErrorMsg(pParse, "misuse of aggregate: %s()", pExpr->u.zToken); + }else{ + return pInfo->aFunc[pExpr->iAgg].iMem; + } + break; + } + case TK_FUNCTION: { + ExprList *pFarg; /* List of function arguments */ + int nFarg; /* Number of function arguments */ + FuncDef *pDef; /* The function definition object */ + const char *zId; /* The function name */ + u32 constMask = 0; /* Mask of function arguments that are constant */ + int i; /* Loop counter */ + sqlite3 *db = pParse->db; /* The database connection */ + u8 enc = ENC(db); /* The text encoding used by this database */ + CollSeq *pColl = 0; /* A collating sequence */ + +#ifndef SQLITE_OMIT_WINDOWFUNC + if( ExprHasProperty(pExpr, EP_WinFunc) ){ + return pExpr->y.pWin->regResult; + } +#endif + + if( ConstFactorOk(pParse) && sqlite3ExprIsConstantNotJoin(pExpr) ){ + /* SQL functions can be expensive. So try to move constant functions + ** out of the inner loop, even if that means an extra OP_Copy. */ + return sqlite3ExprCodeAtInit(pParse, pExpr, -1); + } + assert( !ExprHasProperty(pExpr, EP_xIsSelect) ); + if( ExprHasProperty(pExpr, EP_TokenOnly) ){ + pFarg = 0; + }else{ + pFarg = pExpr->x.pList; + } + nFarg = pFarg ? pFarg->nExpr : 0; + assert( !ExprHasProperty(pExpr, EP_IntValue) ); + zId = pExpr->u.zToken; + pDef = sqlite3FindFunction(db, zId, nFarg, enc, 0); +#ifdef SQLITE_ENABLE_UNKNOWN_SQL_FUNCTION + if( pDef==0 && pParse->explain ){ + pDef = sqlite3FindFunction(db, "unknown", nFarg, enc, 0); + } +#endif + if( pDef==0 || pDef->xFinalize!=0 ){ + sqlite3ErrorMsg(pParse, "unknown function: %s()", zId); + break; + } + + /* Attempt a direct implementation of the built-in COALESCE() and + ** IFNULL() functions. This avoids unnecessary evaluation of + ** arguments past the first non-NULL argument. + */ + if( pDef->funcFlags & SQLITE_FUNC_COALESCE ){ + int endCoalesce = sqlite3VdbeMakeLabel(pParse); + assert( nFarg>=2 ); + sqlite3ExprCode(pParse, pFarg->a[0].pExpr, target); + for(i=1; ia[i].pExpr, target); + } + sqlite3VdbeResolveLabel(v, endCoalesce); + break; + } + + /* The UNLIKELY() function is a no-op. The result is the value + ** of the first argument. + */ + if( pDef->funcFlags & SQLITE_FUNC_UNLIKELY ){ + assert( nFarg>=1 ); + return sqlite3ExprCodeTarget(pParse, pFarg->a[0].pExpr, target); + } + +#ifdef SQLITE_DEBUG + /* The AFFINITY() function evaluates to a string that describes + ** the type affinity of the argument. This is used for testing of + ** the SQLite type logic. + */ + if( pDef->funcFlags & SQLITE_FUNC_AFFINITY ){ + const char *azAff[] = { "blob", "text", "numeric", "integer", "real" }; + char aff; + assert( nFarg==1 ); + aff = sqlite3ExprAffinity(pFarg->a[0].pExpr); + sqlite3VdbeLoadString(v, target, + (aff<=SQLITE_AFF_NONE) ? "none" : azAff[aff-SQLITE_AFF_BLOB]); + return target; + } +#endif + + for(i=0; ia[i].pExpr) ){ + testcase( i==31 ); + constMask |= MASKBIT32(i); + } + if( (pDef->funcFlags & SQLITE_FUNC_NEEDCOLL)!=0 && !pColl ){ + pColl = sqlite3ExprCollSeq(pParse, pFarg->a[i].pExpr); + } + } + if( pFarg ){ + if( constMask ){ + r1 = pParse->nMem+1; + pParse->nMem += nFarg; + }else{ + r1 = sqlite3GetTempRange(pParse, nFarg); + } + + /* For length() and typeof() functions with a column argument, + ** set the P5 parameter to the OP_Column opcode to OPFLAG_LENGTHARG + ** or OPFLAG_TYPEOFARG respectively, to avoid unnecessary data + ** loading. + */ + if( (pDef->funcFlags & (SQLITE_FUNC_LENGTH|SQLITE_FUNC_TYPEOF))!=0 ){ + u8 exprOp; + assert( nFarg==1 ); + assert( pFarg->a[0].pExpr!=0 ); + exprOp = pFarg->a[0].pExpr->op; + if( exprOp==TK_COLUMN || exprOp==TK_AGG_COLUMN ){ + assert( SQLITE_FUNC_LENGTH==OPFLAG_LENGTHARG ); + assert( SQLITE_FUNC_TYPEOF==OPFLAG_TYPEOFARG ); + testcase( pDef->funcFlags & OPFLAG_LENGTHARG ); + pFarg->a[0].pExpr->op2 = + pDef->funcFlags & (OPFLAG_LENGTHARG|OPFLAG_TYPEOFARG); + } + } + + sqlite3ExprCodeExprList(pParse, pFarg, r1, 0, + SQLITE_ECEL_DUP|SQLITE_ECEL_FACTOR); + }else{ + r1 = 0; + } +#ifndef SQLITE_OMIT_VIRTUALTABLE + /* Possibly overload the function if the first argument is + ** a virtual table column. + ** + ** For infix functions (LIKE, GLOB, REGEXP, and MATCH) use the + ** second argument, not the first, as the argument to test to + ** see if it is a column in a virtual table. This is done because + ** the left operand of infix functions (the operand we want to + ** control overloading) ends up as the second argument to the + ** function. The expression "A glob B" is equivalent to + ** "glob(B,A). We want to use the A in "A glob B" to test + ** for function overloading. But we use the B term in "glob(B,A)". + */ + if( nFarg>=2 && ExprHasProperty(pExpr, EP_InfixFunc) ){ + pDef = sqlite3VtabOverloadFunction(db, pDef, nFarg, pFarg->a[1].pExpr); + }else if( nFarg>0 ){ + pDef = sqlite3VtabOverloadFunction(db, pDef, nFarg, pFarg->a[0].pExpr); + } +#endif + if( pDef->funcFlags & SQLITE_FUNC_NEEDCOLL ){ + if( !pColl ) pColl = db->pDfltColl; + sqlite3VdbeAddOp4(v, OP_CollSeq, 0, 0, 0, (char *)pColl, P4_COLLSEQ); + } +#ifdef SQLITE_ENABLE_OFFSET_SQL_FUNC + if( pDef->funcFlags & SQLITE_FUNC_OFFSET ){ + Expr *pArg = pFarg->a[0].pExpr; + if( pArg->op==TK_COLUMN ){ + sqlite3VdbeAddOp3(v, OP_Offset, pArg->iTable, pArg->iColumn, target); + }else{ + sqlite3VdbeAddOp2(v, OP_Null, 0, target); + } + }else +#endif + { + sqlite3VdbeAddOp4(v, pParse->iSelfTab ? OP_PureFunc0 : OP_Function0, + constMask, r1, target, (char*)pDef, P4_FUNCDEF); + sqlite3VdbeChangeP5(v, (u8)nFarg); + } + if( nFarg && constMask==0 ){ + sqlite3ReleaseTempRange(pParse, r1, nFarg); + } + return target; + } +#ifndef SQLITE_OMIT_SUBQUERY + case TK_EXISTS: + case TK_SELECT: { + int nCol; + testcase( op==TK_EXISTS ); + testcase( op==TK_SELECT ); + if( op==TK_SELECT && (nCol = pExpr->x.pSelect->pEList->nExpr)!=1 ){ + sqlite3SubselectError(pParse, nCol, 1); + }else{ + return sqlite3CodeSubselect(pParse, pExpr); + } + break; + } + case TK_SELECT_COLUMN: { + int n; + if( pExpr->pLeft->iTable==0 ){ + pExpr->pLeft->iTable = sqlite3CodeSubselect(pParse, pExpr->pLeft); + } + assert( pExpr->iTable==0 || pExpr->pLeft->op==TK_SELECT ); + if( pExpr->iTable!=0 + && pExpr->iTable!=(n = sqlite3ExprVectorSize(pExpr->pLeft)) + ){ + sqlite3ErrorMsg(pParse, "%d columns assigned %d values", + pExpr->iTable, n); + } + return pExpr->pLeft->iTable + pExpr->iColumn; + } + case TK_IN: { + int destIfFalse = sqlite3VdbeMakeLabel(pParse); + int destIfNull = sqlite3VdbeMakeLabel(pParse); + sqlite3VdbeAddOp2(v, OP_Null, 0, target); + sqlite3ExprCodeIN(pParse, pExpr, destIfFalse, destIfNull); + sqlite3VdbeAddOp2(v, OP_Integer, 1, target); + sqlite3VdbeResolveLabel(v, destIfFalse); + sqlite3VdbeAddOp2(v, OP_AddImm, target, 0); + sqlite3VdbeResolveLabel(v, destIfNull); + return target; + } +#endif /* SQLITE_OMIT_SUBQUERY */ + + + /* + ** x BETWEEN y AND z + ** + ** This is equivalent to + ** + ** x>=y AND x<=z + ** + ** X is stored in pExpr->pLeft. + ** Y is stored in pExpr->pList->a[0].pExpr. + ** Z is stored in pExpr->pList->a[1].pExpr. + */ + case TK_BETWEEN: { + exprCodeBetween(pParse, pExpr, target, 0, 0); + return target; + } + case TK_SPAN: + case TK_COLLATE: + case TK_UPLUS: { + pExpr = pExpr->pLeft; + goto expr_code_doover; /* 2018-04-28: Prevent deep recursion. OSSFuzz. */ + } + + case TK_TRIGGER: { + /* If the opcode is TK_TRIGGER, then the expression is a reference + ** to a column in the new.* or old.* pseudo-tables available to + ** trigger programs. In this case Expr.iTable is set to 1 for the + ** new.* pseudo-table, or 0 for the old.* pseudo-table. Expr.iColumn + ** is set to the column of the pseudo-table to read, or to -1 to + ** read the rowid field. + ** + ** The expression is implemented using an OP_Param opcode. The p1 + ** parameter is set to 0 for an old.rowid reference, or to (i+1) + ** to reference another column of the old.* pseudo-table, where + ** i is the index of the column. For a new.rowid reference, p1 is + ** set to (n+1), where n is the number of columns in each pseudo-table. + ** For a reference to any other column in the new.* pseudo-table, p1 + ** is set to (n+2+i), where n and i are as defined previously. For + ** example, if the table on which triggers are being fired is + ** declared as: + ** + ** CREATE TABLE t1(a, b); + ** + ** Then p1 is interpreted as follows: + ** + ** p1==0 -> old.rowid p1==3 -> new.rowid + ** p1==1 -> old.a p1==4 -> new.a + ** p1==2 -> old.b p1==5 -> new.b + */ + Table *pTab = pExpr->y.pTab; + int p1 = pExpr->iTable * (pTab->nCol+1) + 1 + pExpr->iColumn; + + assert( pExpr->iTable==0 || pExpr->iTable==1 ); + assert( pExpr->iColumn>=-1 && pExpr->iColumnnCol ); + assert( pTab->iPKey<0 || pExpr->iColumn!=pTab->iPKey ); + assert( p1>=0 && p1<(pTab->nCol*2+2) ); + + sqlite3VdbeAddOp2(v, OP_Param, p1, target); + VdbeComment((v, "r[%d]=%s.%s", target, + (pExpr->iTable ? "new" : "old"), + (pExpr->iColumn<0 ? "rowid" : pExpr->y.pTab->aCol[pExpr->iColumn].zName) + )); + +#ifndef SQLITE_OMIT_FLOATING_POINT + /* If the column has REAL affinity, it may currently be stored as an + ** integer. Use OP_RealAffinity to make sure it is really real. + ** + ** EVIDENCE-OF: R-60985-57662 SQLite will convert the value back to + ** floating point when extracting it from the record. */ + if( pExpr->iColumn>=0 + && pTab->aCol[pExpr->iColumn].affinity==SQLITE_AFF_REAL + ){ + sqlite3VdbeAddOp1(v, OP_RealAffinity, target); + } +#endif + break; + } + + case TK_VECTOR: { + sqlite3ErrorMsg(pParse, "row value misused"); + break; + } + + /* TK_IF_NULL_ROW Expr nodes are inserted ahead of expressions + ** that derive from the right-hand table of a LEFT JOIN. The + ** Expr.iTable value is the table number for the right-hand table. + ** The expression is only evaluated if that table is not currently + ** on a LEFT JOIN NULL row. + */ + case TK_IF_NULL_ROW: { + int addrINR; + u8 okConstFactor = pParse->okConstFactor; + addrINR = sqlite3VdbeAddOp1(v, OP_IfNullRow, pExpr->iTable); + /* Temporarily disable factoring of constant expressions, since + ** even though expressions may appear to be constant, they are not + ** really constant because they originate from the right-hand side + ** of a LEFT JOIN. */ + pParse->okConstFactor = 0; + inReg = sqlite3ExprCodeTarget(pParse, pExpr->pLeft, target); + pParse->okConstFactor = okConstFactor; + sqlite3VdbeJumpHere(v, addrINR); + sqlite3VdbeChangeP3(v, addrINR, inReg); + break; + } + + /* + ** Form A: + ** CASE x WHEN e1 THEN r1 WHEN e2 THEN r2 ... WHEN eN THEN rN ELSE y END + ** + ** Form B: + ** CASE WHEN e1 THEN r1 WHEN e2 THEN r2 ... WHEN eN THEN rN ELSE y END + ** + ** Form A is can be transformed into the equivalent form B as follows: + ** CASE WHEN x=e1 THEN r1 WHEN x=e2 THEN r2 ... + ** WHEN x=eN THEN rN ELSE y END + ** + ** X (if it exists) is in pExpr->pLeft. + ** Y is in the last element of pExpr->x.pList if pExpr->x.pList->nExpr is + ** odd. The Y is also optional. If the number of elements in x.pList + ** is even, then Y is omitted and the "otherwise" result is NULL. + ** Ei is in pExpr->pList->a[i*2] and Ri is pExpr->pList->a[i*2+1]. + ** + ** The result of the expression is the Ri for the first matching Ei, + ** or if there is no matching Ei, the ELSE term Y, or if there is + ** no ELSE term, NULL. + */ + default: assert( op==TK_CASE ); { + int endLabel; /* GOTO label for end of CASE stmt */ + int nextCase; /* GOTO label for next WHEN clause */ + int nExpr; /* 2x number of WHEN terms */ + int i; /* Loop counter */ + ExprList *pEList; /* List of WHEN terms */ + struct ExprList_item *aListelem; /* Array of WHEN terms */ + Expr opCompare; /* The X==Ei expression */ + Expr *pX; /* The X expression */ + Expr *pTest = 0; /* X==Ei (form A) or just Ei (form B) */ + Expr *pDel = 0; + sqlite3 *db = pParse->db; + + assert( !ExprHasProperty(pExpr, EP_xIsSelect) && pExpr->x.pList ); + assert(pExpr->x.pList->nExpr > 0); + pEList = pExpr->x.pList; + aListelem = pEList->a; + nExpr = pEList->nExpr; + endLabel = sqlite3VdbeMakeLabel(pParse); + if( (pX = pExpr->pLeft)!=0 ){ + pDel = sqlite3ExprDup(db, pX, 0); + if( db->mallocFailed ){ + sqlite3ExprDelete(db, pDel); + break; + } + testcase( pX->op==TK_COLUMN ); + exprToRegister(pDel, exprCodeVector(pParse, pDel, ®Free1)); + testcase( regFree1==0 ); + memset(&opCompare, 0, sizeof(opCompare)); + opCompare.op = TK_EQ; + opCompare.pLeft = pDel; + pTest = &opCompare; + /* Ticket b351d95f9cd5ef17e9d9dbae18f5ca8611190001: + ** The value in regFree1 might get SCopy-ed into the file result. + ** So make sure that the regFree1 register is not reused for other + ** purposes and possibly overwritten. */ + regFree1 = 0; + } + for(i=0; iop==TK_COLUMN ); + sqlite3ExprIfFalse(pParse, pTest, nextCase, SQLITE_JUMPIFNULL); + testcase( aListelem[i+1].pExpr->op==TK_COLUMN ); + sqlite3ExprCode(pParse, aListelem[i+1].pExpr, target); + sqlite3VdbeGoto(v, endLabel); + sqlite3VdbeResolveLabel(v, nextCase); + } + if( (nExpr&1)!=0 ){ + sqlite3ExprCode(pParse, pEList->a[nExpr-1].pExpr, target); + }else{ + sqlite3VdbeAddOp2(v, OP_Null, 0, target); + } + sqlite3ExprDelete(db, pDel); + sqlite3VdbeResolveLabel(v, endLabel); + break; + } +#ifndef SQLITE_OMIT_TRIGGER + case TK_RAISE: { + assert( pExpr->affExpr==OE_Rollback + || pExpr->affExpr==OE_Abort + || pExpr->affExpr==OE_Fail + || pExpr->affExpr==OE_Ignore + ); + if( !pParse->pTriggerTab ){ + sqlite3ErrorMsg(pParse, + "RAISE() may only be used within a trigger-program"); + return 0; + } + if( pExpr->affExpr==OE_Abort ){ + sqlite3MayAbort(pParse); + } + assert( !ExprHasProperty(pExpr, EP_IntValue) ); + if( pExpr->affExpr==OE_Ignore ){ + sqlite3VdbeAddOp4( + v, OP_Halt, SQLITE_OK, OE_Ignore, 0, pExpr->u.zToken,0); + VdbeCoverage(v); + }else{ + sqlite3HaltConstraint(pParse, SQLITE_CONSTRAINT_TRIGGER, + pExpr->affExpr, pExpr->u.zToken, 0, 0); + } + + break; + } +#endif + } + sqlite3ReleaseTempReg(pParse, regFree1); + sqlite3ReleaseTempReg(pParse, regFree2); + return inReg; +} + +/* +** Factor out the code of the given expression to initialization time. +** +** If regDest>=0 then the result is always stored in that register and the +** result is not reusable. If regDest<0 then this routine is free to +** store the value whereever it wants. The register where the expression +** is stored is returned. When regDest<0, two identical expressions will +** code to the same register. +*/ +SQLITE_PRIVATE int sqlite3ExprCodeAtInit( + Parse *pParse, /* Parsing context */ + Expr *pExpr, /* The expression to code when the VDBE initializes */ + int regDest /* Store the value in this register */ +){ + ExprList *p; + assert( ConstFactorOk(pParse) ); + p = pParse->pConstExpr; + if( regDest<0 && p ){ + struct ExprList_item *pItem; + int i; + for(pItem=p->a, i=p->nExpr; i>0; pItem++, i--){ + if( pItem->reusable && sqlite3ExprCompare(0,pItem->pExpr,pExpr,-1)==0 ){ + return pItem->u.iConstExprReg; + } + } + } + pExpr = sqlite3ExprDup(pParse->db, pExpr, 0); + p = sqlite3ExprListAppend(pParse, p, pExpr); + if( p ){ + struct ExprList_item *pItem = &p->a[p->nExpr-1]; + pItem->reusable = regDest<0; + if( regDest<0 ) regDest = ++pParse->nMem; + pItem->u.iConstExprReg = regDest; + } + pParse->pConstExpr = p; + return regDest; +} + +/* +** Generate code to evaluate an expression and store the results +** into a register. Return the register number where the results +** are stored. +** +** If the register is a temporary register that can be deallocated, +** then write its number into *pReg. If the result register is not +** a temporary, then set *pReg to zero. +** +** If pExpr is a constant, then this routine might generate this +** code to fill the register in the initialization section of the +** VDBE program, in order to factor it out of the evaluation loop. +*/ +SQLITE_PRIVATE int sqlite3ExprCodeTemp(Parse *pParse, Expr *pExpr, int *pReg){ + int r2; + pExpr = sqlite3ExprSkipCollateAndLikely(pExpr); + if( ConstFactorOk(pParse) + && pExpr->op!=TK_REGISTER + && sqlite3ExprIsConstantNotJoin(pExpr) + ){ + *pReg = 0; + r2 = sqlite3ExprCodeAtInit(pParse, pExpr, -1); + }else{ + int r1 = sqlite3GetTempReg(pParse); + r2 = sqlite3ExprCodeTarget(pParse, pExpr, r1); + if( r2==r1 ){ + *pReg = r1; + }else{ + sqlite3ReleaseTempReg(pParse, r1); + *pReg = 0; + } + } + return r2; +} + +/* +** Generate code that will evaluate expression pExpr and store the +** results in register target. The results are guaranteed to appear +** in register target. +*/ +SQLITE_PRIVATE void sqlite3ExprCode(Parse *pParse, Expr *pExpr, int target){ + int inReg; + + assert( target>0 && target<=pParse->nMem ); + if( pExpr && pExpr->op==TK_REGISTER ){ + sqlite3VdbeAddOp2(pParse->pVdbe, OP_Copy, pExpr->iTable, target); + }else{ + inReg = sqlite3ExprCodeTarget(pParse, pExpr, target); + assert( pParse->pVdbe!=0 || pParse->db->mallocFailed ); + if( inReg!=target && pParse->pVdbe ){ + sqlite3VdbeAddOp2(pParse->pVdbe, OP_SCopy, inReg, target); + } + } +} + +/* +** Make a transient copy of expression pExpr and then code it using +** sqlite3ExprCode(). This routine works just like sqlite3ExprCode() +** except that the input expression is guaranteed to be unchanged. +*/ +SQLITE_PRIVATE void sqlite3ExprCodeCopy(Parse *pParse, Expr *pExpr, int target){ + sqlite3 *db = pParse->db; + pExpr = sqlite3ExprDup(db, pExpr, 0); + if( !db->mallocFailed ) sqlite3ExprCode(pParse, pExpr, target); + sqlite3ExprDelete(db, pExpr); +} + +/* +** Generate code that will evaluate expression pExpr and store the +** results in register target. The results are guaranteed to appear +** in register target. If the expression is constant, then this routine +** might choose to code the expression at initialization time. +*/ +SQLITE_PRIVATE void sqlite3ExprCodeFactorable(Parse *pParse, Expr *pExpr, int target){ + if( pParse->okConstFactor && sqlite3ExprIsConstantNotJoin(pExpr) ){ + sqlite3ExprCodeAtInit(pParse, pExpr, target); + }else{ + sqlite3ExprCode(pParse, pExpr, target); + } +} + +/* +** Generate code that evaluates the given expression and puts the result +** in register target. +** +** Also make a copy of the expression results into another "cache" register +** and modify the expression so that the next time it is evaluated, +** the result is a copy of the cache register. +** +** This routine is used for expressions that are used multiple +** times. They are evaluated once and the results of the expression +** are reused. +*/ +SQLITE_PRIVATE void sqlite3ExprCodeAndCache(Parse *pParse, Expr *pExpr, int target){ + Vdbe *v = pParse->pVdbe; + int iMem; + + assert( target>0 ); + assert( pExpr->op!=TK_REGISTER ); + sqlite3ExprCode(pParse, pExpr, target); + iMem = ++pParse->nMem; + sqlite3VdbeAddOp2(v, OP_Copy, target, iMem); + exprToRegister(pExpr, iMem); +} + +/* +** Generate code that pushes the value of every element of the given +** expression list into a sequence of registers beginning at target. +** +** Return the number of elements evaluated. The number returned will +** usually be pList->nExpr but might be reduced if SQLITE_ECEL_OMITREF +** is defined. +** +** The SQLITE_ECEL_DUP flag prevents the arguments from being +** filled using OP_SCopy. OP_Copy must be used instead. +** +** The SQLITE_ECEL_FACTOR argument allows constant arguments to be +** factored out into initialization code. +** +** The SQLITE_ECEL_REF flag means that expressions in the list with +** ExprList.a[].u.x.iOrderByCol>0 have already been evaluated and stored +** in registers at srcReg, and so the value can be copied from there. +** If SQLITE_ECEL_OMITREF is also set, then the values with u.x.iOrderByCol>0 +** are simply omitted rather than being copied from srcReg. +*/ +SQLITE_PRIVATE int sqlite3ExprCodeExprList( + Parse *pParse, /* Parsing context */ + ExprList *pList, /* The expression list to be coded */ + int target, /* Where to write results */ + int srcReg, /* Source registers if SQLITE_ECEL_REF */ + u8 flags /* SQLITE_ECEL_* flags */ +){ + struct ExprList_item *pItem; + int i, j, n; + u8 copyOp = (flags & SQLITE_ECEL_DUP) ? OP_Copy : OP_SCopy; + Vdbe *v = pParse->pVdbe; + assert( pList!=0 ); + assert( target>0 ); + assert( pParse->pVdbe!=0 ); /* Never gets this far otherwise */ + n = pList->nExpr; + if( !ConstFactorOk(pParse) ) flags &= ~SQLITE_ECEL_FACTOR; + for(pItem=pList->a, i=0; ipExpr; +#ifdef SQLITE_ENABLE_SORTER_REFERENCES + if( pItem->bSorterRef ){ + i--; + n--; + }else +#endif + if( (flags & SQLITE_ECEL_REF)!=0 && (j = pItem->u.x.iOrderByCol)>0 ){ + if( flags & SQLITE_ECEL_OMITREF ){ + i--; + n--; + }else{ + sqlite3VdbeAddOp2(v, copyOp, j+srcReg-1, target+i); + } + }else if( (flags & SQLITE_ECEL_FACTOR)!=0 + && sqlite3ExprIsConstantNotJoin(pExpr) + ){ + sqlite3ExprCodeAtInit(pParse, pExpr, target+i); + }else{ + int inReg = sqlite3ExprCodeTarget(pParse, pExpr, target+i); + if( inReg!=target+i ){ + VdbeOp *pOp; + if( copyOp==OP_Copy + && (pOp=sqlite3VdbeGetOp(v, -1))->opcode==OP_Copy + && pOp->p1+pOp->p3+1==inReg + && pOp->p2+pOp->p3+1==target+i + ){ + pOp->p3++; + }else{ + sqlite3VdbeAddOp2(v, copyOp, inReg, target+i); + } + } + } + } + return n; +} + +/* +** Generate code for a BETWEEN operator. +** +** x BETWEEN y AND z +** +** The above is equivalent to +** +** x>=y AND x<=z +** +** Code it as such, taking care to do the common subexpression +** elimination of x. +** +** The xJumpIf parameter determines details: +** +** NULL: Store the boolean result in reg[dest] +** sqlite3ExprIfTrue: Jump to dest if true +** sqlite3ExprIfFalse: Jump to dest if false +** +** The jumpIfNull parameter is ignored if xJumpIf is NULL. +*/ +static void exprCodeBetween( + Parse *pParse, /* Parsing and code generating context */ + Expr *pExpr, /* The BETWEEN expression */ + int dest, /* Jump destination or storage location */ + void (*xJump)(Parse*,Expr*,int,int), /* Action to take */ + int jumpIfNull /* Take the jump if the BETWEEN is NULL */ +){ + Expr exprAnd; /* The AND operator in x>=y AND x<=z */ + Expr compLeft; /* The x>=y term */ + Expr compRight; /* The x<=z term */ + int regFree1 = 0; /* Temporary use register */ + Expr *pDel = 0; + sqlite3 *db = pParse->db; + + memset(&compLeft, 0, sizeof(Expr)); + memset(&compRight, 0, sizeof(Expr)); + memset(&exprAnd, 0, sizeof(Expr)); + + assert( !ExprHasProperty(pExpr, EP_xIsSelect) ); + pDel = sqlite3ExprDup(db, pExpr->pLeft, 0); + if( db->mallocFailed==0 ){ + exprAnd.op = TK_AND; + exprAnd.pLeft = &compLeft; + exprAnd.pRight = &compRight; + compLeft.op = TK_GE; + compLeft.pLeft = pDel; + compLeft.pRight = pExpr->x.pList->a[0].pExpr; + compRight.op = TK_LE; + compRight.pLeft = pDel; + compRight.pRight = pExpr->x.pList->a[1].pExpr; + exprToRegister(pDel, exprCodeVector(pParse, pDel, ®Free1)); + if( xJump ){ + xJump(pParse, &exprAnd, dest, jumpIfNull); + }else{ + /* Mark the expression is being from the ON or USING clause of a join + ** so that the sqlite3ExprCodeTarget() routine will not attempt to move + ** it into the Parse.pConstExpr list. We should use a new bit for this, + ** for clarity, but we are out of bits in the Expr.flags field so we + ** have to reuse the EP_FromJoin bit. Bummer. */ + pDel->flags |= EP_FromJoin; + sqlite3ExprCodeTarget(pParse, &exprAnd, dest); + } + sqlite3ReleaseTempReg(pParse, regFree1); + } + sqlite3ExprDelete(db, pDel); + + /* Ensure adequate test coverage */ + testcase( xJump==sqlite3ExprIfTrue && jumpIfNull==0 && regFree1==0 ); + testcase( xJump==sqlite3ExprIfTrue && jumpIfNull==0 && regFree1!=0 ); + testcase( xJump==sqlite3ExprIfTrue && jumpIfNull!=0 && regFree1==0 ); + testcase( xJump==sqlite3ExprIfTrue && jumpIfNull!=0 && regFree1!=0 ); + testcase( xJump==sqlite3ExprIfFalse && jumpIfNull==0 && regFree1==0 ); + testcase( xJump==sqlite3ExprIfFalse && jumpIfNull==0 && regFree1!=0 ); + testcase( xJump==sqlite3ExprIfFalse && jumpIfNull!=0 && regFree1==0 ); + testcase( xJump==sqlite3ExprIfFalse && jumpIfNull!=0 && regFree1!=0 ); + testcase( xJump==0 ); +} + +/* +** Generate code for a boolean expression such that a jump is made +** to the label "dest" if the expression is true but execution +** continues straight thru if the expression is false. +** +** If the expression evaluates to NULL (neither true nor false), then +** take the jump if the jumpIfNull flag is SQLITE_JUMPIFNULL. +** +** This code depends on the fact that certain token values (ex: TK_EQ) +** are the same as opcode values (ex: OP_Eq) that implement the corresponding +** operation. Special comments in vdbe.c and the mkopcodeh.awk script in +** the make process cause these values to align. Assert()s in the code +** below verify that the numbers are aligned correctly. +*/ +SQLITE_PRIVATE void sqlite3ExprIfTrue(Parse *pParse, Expr *pExpr, int dest, int jumpIfNull){ + Vdbe *v = pParse->pVdbe; + int op = 0; + int regFree1 = 0; + int regFree2 = 0; + int r1, r2; + + assert( jumpIfNull==SQLITE_JUMPIFNULL || jumpIfNull==0 ); + if( NEVER(v==0) ) return; /* Existence of VDBE checked by caller */ + if( NEVER(pExpr==0) ) return; /* No way this can happen */ + op = pExpr->op; + switch( op ){ + case TK_AND: + case TK_OR: { + Expr *pAlt = sqlite3ExprSimplifiedAndOr(pExpr); + if( pAlt!=pExpr ){ + sqlite3ExprIfTrue(pParse, pAlt, dest, jumpIfNull); + }else if( op==TK_AND ){ + int d2 = sqlite3VdbeMakeLabel(pParse); + testcase( jumpIfNull==0 ); + sqlite3ExprIfFalse(pParse, pExpr->pLeft, d2, + jumpIfNull^SQLITE_JUMPIFNULL); + sqlite3ExprIfTrue(pParse, pExpr->pRight, dest, jumpIfNull); + sqlite3VdbeResolveLabel(v, d2); + }else{ + testcase( jumpIfNull==0 ); + sqlite3ExprIfTrue(pParse, pExpr->pLeft, dest, jumpIfNull); + sqlite3ExprIfTrue(pParse, pExpr->pRight, dest, jumpIfNull); + } + break; + } + case TK_NOT: { + testcase( jumpIfNull==0 ); + sqlite3ExprIfFalse(pParse, pExpr->pLeft, dest, jumpIfNull); + break; + } + case TK_TRUTH: { + int isNot; /* IS NOT TRUE or IS NOT FALSE */ + int isTrue; /* IS TRUE or IS NOT TRUE */ + testcase( jumpIfNull==0 ); + isNot = pExpr->op2==TK_ISNOT; + isTrue = sqlite3ExprTruthValue(pExpr->pRight); + testcase( isTrue && isNot ); + testcase( !isTrue && isNot ); + if( isTrue ^ isNot ){ + sqlite3ExprIfTrue(pParse, pExpr->pLeft, dest, + isNot ? SQLITE_JUMPIFNULL : 0); + }else{ + sqlite3ExprIfFalse(pParse, pExpr->pLeft, dest, + isNot ? SQLITE_JUMPIFNULL : 0); + } + break; + } + case TK_IS: + case TK_ISNOT: + testcase( op==TK_IS ); + testcase( op==TK_ISNOT ); + op = (op==TK_IS) ? TK_EQ : TK_NE; + jumpIfNull = SQLITE_NULLEQ; + /* Fall thru */ + case TK_LT: + case TK_LE: + case TK_GT: + case TK_GE: + case TK_NE: + case TK_EQ: { + if( sqlite3ExprIsVector(pExpr->pLeft) ) goto default_expr; + testcase( jumpIfNull==0 ); + r1 = sqlite3ExprCodeTemp(pParse, pExpr->pLeft, ®Free1); + r2 = sqlite3ExprCodeTemp(pParse, pExpr->pRight, ®Free2); + codeCompare(pParse, pExpr->pLeft, pExpr->pRight, op, + r1, r2, dest, jumpIfNull); + assert(TK_LT==OP_Lt); testcase(op==OP_Lt); VdbeCoverageIf(v,op==OP_Lt); + assert(TK_LE==OP_Le); testcase(op==OP_Le); VdbeCoverageIf(v,op==OP_Le); + assert(TK_GT==OP_Gt); testcase(op==OP_Gt); VdbeCoverageIf(v,op==OP_Gt); + assert(TK_GE==OP_Ge); testcase(op==OP_Ge); VdbeCoverageIf(v,op==OP_Ge); + assert(TK_EQ==OP_Eq); testcase(op==OP_Eq); + VdbeCoverageIf(v, op==OP_Eq && jumpIfNull==SQLITE_NULLEQ); + VdbeCoverageIf(v, op==OP_Eq && jumpIfNull!=SQLITE_NULLEQ); + assert(TK_NE==OP_Ne); testcase(op==OP_Ne); + VdbeCoverageIf(v, op==OP_Ne && jumpIfNull==SQLITE_NULLEQ); + VdbeCoverageIf(v, op==OP_Ne && jumpIfNull!=SQLITE_NULLEQ); + testcase( regFree1==0 ); + testcase( regFree2==0 ); + break; + } + case TK_ISNULL: + case TK_NOTNULL: { + assert( TK_ISNULL==OP_IsNull ); testcase( op==TK_ISNULL ); + assert( TK_NOTNULL==OP_NotNull ); testcase( op==TK_NOTNULL ); + r1 = sqlite3ExprCodeTemp(pParse, pExpr->pLeft, ®Free1); + sqlite3VdbeAddOp2(v, op, r1, dest); + VdbeCoverageIf(v, op==TK_ISNULL); + VdbeCoverageIf(v, op==TK_NOTNULL); + testcase( regFree1==0 ); + break; + } + case TK_BETWEEN: { + testcase( jumpIfNull==0 ); + exprCodeBetween(pParse, pExpr, dest, sqlite3ExprIfTrue, jumpIfNull); + break; + } +#ifndef SQLITE_OMIT_SUBQUERY + case TK_IN: { + int destIfFalse = sqlite3VdbeMakeLabel(pParse); + int destIfNull = jumpIfNull ? dest : destIfFalse; + sqlite3ExprCodeIN(pParse, pExpr, destIfFalse, destIfNull); + sqlite3VdbeGoto(v, dest); + sqlite3VdbeResolveLabel(v, destIfFalse); + break; + } +#endif + default: { + default_expr: + if( ExprAlwaysTrue(pExpr) ){ + sqlite3VdbeGoto(v, dest); + }else if( ExprAlwaysFalse(pExpr) ){ + /* No-op */ + }else{ + r1 = sqlite3ExprCodeTemp(pParse, pExpr, ®Free1); + sqlite3VdbeAddOp3(v, OP_If, r1, dest, jumpIfNull!=0); + VdbeCoverage(v); + testcase( regFree1==0 ); + testcase( jumpIfNull==0 ); + } + break; + } + } + sqlite3ReleaseTempReg(pParse, regFree1); + sqlite3ReleaseTempReg(pParse, regFree2); +} + +/* +** Generate code for a boolean expression such that a jump is made +** to the label "dest" if the expression is false but execution +** continues straight thru if the expression is true. +** +** If the expression evaluates to NULL (neither true nor false) then +** jump if jumpIfNull is SQLITE_JUMPIFNULL or fall through if jumpIfNull +** is 0. +*/ +SQLITE_PRIVATE void sqlite3ExprIfFalse(Parse *pParse, Expr *pExpr, int dest, int jumpIfNull){ + Vdbe *v = pParse->pVdbe; + int op = 0; + int regFree1 = 0; + int regFree2 = 0; + int r1, r2; + + assert( jumpIfNull==SQLITE_JUMPIFNULL || jumpIfNull==0 ); + if( NEVER(v==0) ) return; /* Existence of VDBE checked by caller */ + if( pExpr==0 ) return; + + /* The value of pExpr->op and op are related as follows: + ** + ** pExpr->op op + ** --------- ---------- + ** TK_ISNULL OP_NotNull + ** TK_NOTNULL OP_IsNull + ** TK_NE OP_Eq + ** TK_EQ OP_Ne + ** TK_GT OP_Le + ** TK_LE OP_Gt + ** TK_GE OP_Lt + ** TK_LT OP_Ge + ** + ** For other values of pExpr->op, op is undefined and unused. + ** The value of TK_ and OP_ constants are arranged such that we + ** can compute the mapping above using the following expression. + ** Assert()s verify that the computation is correct. + */ + op = ((pExpr->op+(TK_ISNULL&1))^1)-(TK_ISNULL&1); + + /* Verify correct alignment of TK_ and OP_ constants + */ + assert( pExpr->op!=TK_ISNULL || op==OP_NotNull ); + assert( pExpr->op!=TK_NOTNULL || op==OP_IsNull ); + assert( pExpr->op!=TK_NE || op==OP_Eq ); + assert( pExpr->op!=TK_EQ || op==OP_Ne ); + assert( pExpr->op!=TK_LT || op==OP_Ge ); + assert( pExpr->op!=TK_LE || op==OP_Gt ); + assert( pExpr->op!=TK_GT || op==OP_Le ); + assert( pExpr->op!=TK_GE || op==OP_Lt ); + + switch( pExpr->op ){ + case TK_AND: + case TK_OR: { + Expr *pAlt = sqlite3ExprSimplifiedAndOr(pExpr); + if( pAlt!=pExpr ){ + sqlite3ExprIfFalse(pParse, pAlt, dest, jumpIfNull); + }else if( pExpr->op==TK_AND ){ + testcase( jumpIfNull==0 ); + sqlite3ExprIfFalse(pParse, pExpr->pLeft, dest, jumpIfNull); + sqlite3ExprIfFalse(pParse, pExpr->pRight, dest, jumpIfNull); + }else{ + int d2 = sqlite3VdbeMakeLabel(pParse); + testcase( jumpIfNull==0 ); + sqlite3ExprIfTrue(pParse, pExpr->pLeft, d2, + jumpIfNull^SQLITE_JUMPIFNULL); + sqlite3ExprIfFalse(pParse, pExpr->pRight, dest, jumpIfNull); + sqlite3VdbeResolveLabel(v, d2); + } + break; + } + case TK_NOT: { + testcase( jumpIfNull==0 ); + sqlite3ExprIfTrue(pParse, pExpr->pLeft, dest, jumpIfNull); + break; + } + case TK_TRUTH: { + int isNot; /* IS NOT TRUE or IS NOT FALSE */ + int isTrue; /* IS TRUE or IS NOT TRUE */ + testcase( jumpIfNull==0 ); + isNot = pExpr->op2==TK_ISNOT; + isTrue = sqlite3ExprTruthValue(pExpr->pRight); + testcase( isTrue && isNot ); + testcase( !isTrue && isNot ); + if( isTrue ^ isNot ){ + /* IS TRUE and IS NOT FALSE */ + sqlite3ExprIfFalse(pParse, pExpr->pLeft, dest, + isNot ? 0 : SQLITE_JUMPIFNULL); + + }else{ + /* IS FALSE and IS NOT TRUE */ + sqlite3ExprIfTrue(pParse, pExpr->pLeft, dest, + isNot ? 0 : SQLITE_JUMPIFNULL); + } + break; + } + case TK_IS: + case TK_ISNOT: + testcase( pExpr->op==TK_IS ); + testcase( pExpr->op==TK_ISNOT ); + op = (pExpr->op==TK_IS) ? TK_NE : TK_EQ; + jumpIfNull = SQLITE_NULLEQ; + /* Fall thru */ + case TK_LT: + case TK_LE: + case TK_GT: + case TK_GE: + case TK_NE: + case TK_EQ: { + if( sqlite3ExprIsVector(pExpr->pLeft) ) goto default_expr; + testcase( jumpIfNull==0 ); + r1 = sqlite3ExprCodeTemp(pParse, pExpr->pLeft, ®Free1); + r2 = sqlite3ExprCodeTemp(pParse, pExpr->pRight, ®Free2); + codeCompare(pParse, pExpr->pLeft, pExpr->pRight, op, + r1, r2, dest, jumpIfNull); + assert(TK_LT==OP_Lt); testcase(op==OP_Lt); VdbeCoverageIf(v,op==OP_Lt); + assert(TK_LE==OP_Le); testcase(op==OP_Le); VdbeCoverageIf(v,op==OP_Le); + assert(TK_GT==OP_Gt); testcase(op==OP_Gt); VdbeCoverageIf(v,op==OP_Gt); + assert(TK_GE==OP_Ge); testcase(op==OP_Ge); VdbeCoverageIf(v,op==OP_Ge); + assert(TK_EQ==OP_Eq); testcase(op==OP_Eq); + VdbeCoverageIf(v, op==OP_Eq && jumpIfNull!=SQLITE_NULLEQ); + VdbeCoverageIf(v, op==OP_Eq && jumpIfNull==SQLITE_NULLEQ); + assert(TK_NE==OP_Ne); testcase(op==OP_Ne); + VdbeCoverageIf(v, op==OP_Ne && jumpIfNull!=SQLITE_NULLEQ); + VdbeCoverageIf(v, op==OP_Ne && jumpIfNull==SQLITE_NULLEQ); + testcase( regFree1==0 ); + testcase( regFree2==0 ); + break; + } + case TK_ISNULL: + case TK_NOTNULL: { + r1 = sqlite3ExprCodeTemp(pParse, pExpr->pLeft, ®Free1); + sqlite3VdbeAddOp2(v, op, r1, dest); + testcase( op==TK_ISNULL ); VdbeCoverageIf(v, op==TK_ISNULL); + testcase( op==TK_NOTNULL ); VdbeCoverageIf(v, op==TK_NOTNULL); + testcase( regFree1==0 ); + break; + } + case TK_BETWEEN: { + testcase( jumpIfNull==0 ); + exprCodeBetween(pParse, pExpr, dest, sqlite3ExprIfFalse, jumpIfNull); + break; + } +#ifndef SQLITE_OMIT_SUBQUERY + case TK_IN: { + if( jumpIfNull ){ + sqlite3ExprCodeIN(pParse, pExpr, dest, dest); + }else{ + int destIfNull = sqlite3VdbeMakeLabel(pParse); + sqlite3ExprCodeIN(pParse, pExpr, dest, destIfNull); + sqlite3VdbeResolveLabel(v, destIfNull); + } + break; + } +#endif + default: { + default_expr: + if( ExprAlwaysFalse(pExpr) ){ + sqlite3VdbeGoto(v, dest); + }else if( ExprAlwaysTrue(pExpr) ){ + /* no-op */ + }else{ + r1 = sqlite3ExprCodeTemp(pParse, pExpr, ®Free1); + sqlite3VdbeAddOp3(v, OP_IfNot, r1, dest, jumpIfNull!=0); + VdbeCoverage(v); + testcase( regFree1==0 ); + testcase( jumpIfNull==0 ); + } + break; + } + } + sqlite3ReleaseTempReg(pParse, regFree1); + sqlite3ReleaseTempReg(pParse, regFree2); +} + +/* +** Like sqlite3ExprIfFalse() except that a copy is made of pExpr before +** code generation, and that copy is deleted after code generation. This +** ensures that the original pExpr is unchanged. +*/ +SQLITE_PRIVATE void sqlite3ExprIfFalseDup(Parse *pParse, Expr *pExpr, int dest,int jumpIfNull){ + sqlite3 *db = pParse->db; + Expr *pCopy = sqlite3ExprDup(db, pExpr, 0); + if( db->mallocFailed==0 ){ + sqlite3ExprIfFalse(pParse, pCopy, dest, jumpIfNull); + } + sqlite3ExprDelete(db, pCopy); +} + +/* +** Expression pVar is guaranteed to be an SQL variable. pExpr may be any +** type of expression. +** +** If pExpr is a simple SQL value - an integer, real, string, blob +** or NULL value - then the VDBE currently being prepared is configured +** to re-prepare each time a new value is bound to variable pVar. +** +** Additionally, if pExpr is a simple SQL value and the value is the +** same as that currently bound to variable pVar, non-zero is returned. +** Otherwise, if the values are not the same or if pExpr is not a simple +** SQL value, zero is returned. +*/ +static int exprCompareVariable(Parse *pParse, Expr *pVar, Expr *pExpr){ + int res = 0; + int iVar; + sqlite3_value *pL, *pR = 0; + + sqlite3ValueFromExpr(pParse->db, pExpr, SQLITE_UTF8, SQLITE_AFF_BLOB, &pR); + if( pR ){ + iVar = pVar->iColumn; + sqlite3VdbeSetVarmask(pParse->pVdbe, iVar); + pL = sqlite3VdbeGetBoundValue(pParse->pReprepare, iVar, SQLITE_AFF_BLOB); + if( pL ){ + if( sqlite3_value_type(pL)==SQLITE_TEXT ){ + sqlite3_value_text(pL); /* Make sure the encoding is UTF-8 */ + } + res = 0==sqlite3MemCompare(pL, pR, 0); + } + sqlite3ValueFree(pR); + sqlite3ValueFree(pL); + } + + return res; +} + +/* +** Do a deep comparison of two expression trees. Return 0 if the two +** expressions are completely identical. Return 1 if they differ only +** by a COLLATE operator at the top level. Return 2 if there are differences +** other than the top-level COLLATE operator. +** +** If any subelement of pB has Expr.iTable==(-1) then it is allowed +** to compare equal to an equivalent element in pA with Expr.iTable==iTab. +** +** The pA side might be using TK_REGISTER. If that is the case and pB is +** not using TK_REGISTER but is otherwise equivalent, then still return 0. +** +** Sometimes this routine will return 2 even if the two expressions +** really are equivalent. If we cannot prove that the expressions are +** identical, we return 2 just to be safe. So if this routine +** returns 2, then you do not really know for certain if the two +** expressions are the same. But if you get a 0 or 1 return, then you +** can be sure the expressions are the same. In the places where +** this routine is used, it does not hurt to get an extra 2 - that +** just might result in some slightly slower code. But returning +** an incorrect 0 or 1 could lead to a malfunction. +** +** If pParse is not NULL then TK_VARIABLE terms in pA with bindings in +** pParse->pReprepare can be matched against literals in pB. The +** pParse->pVdbe->expmask bitmask is updated for each variable referenced. +** If pParse is NULL (the normal case) then any TK_VARIABLE term in +** Argument pParse should normally be NULL. If it is not NULL and pA or +** pB causes a return value of 2. +*/ +SQLITE_PRIVATE int sqlite3ExprCompare(Parse *pParse, Expr *pA, Expr *pB, int iTab){ + u32 combinedFlags; + if( pA==0 || pB==0 ){ + return pB==pA ? 0 : 2; + } + if( pParse && pA->op==TK_VARIABLE && exprCompareVariable(pParse, pA, pB) ){ + return 0; + } + combinedFlags = pA->flags | pB->flags; + if( combinedFlags & EP_IntValue ){ + if( (pA->flags&pB->flags&EP_IntValue)!=0 && pA->u.iValue==pB->u.iValue ){ + return 0; + } + return 2; + } + if( pA->op!=pB->op || pA->op==TK_RAISE ){ + if( pA->op==TK_COLLATE && sqlite3ExprCompare(pParse, pA->pLeft,pB,iTab)<2 ){ + return 1; + } + if( pB->op==TK_COLLATE && sqlite3ExprCompare(pParse, pA,pB->pLeft,iTab)<2 ){ + return 1; + } + return 2; + } + if( pA->op!=TK_COLUMN && pA->op!=TK_AGG_COLUMN && pA->u.zToken ){ + if( pA->op==TK_FUNCTION || pA->op==TK_AGG_FUNCTION ){ + if( sqlite3StrICmp(pA->u.zToken,pB->u.zToken)!=0 ) return 2; +#ifndef SQLITE_OMIT_WINDOWFUNC + assert( pA->op==pB->op ); + if( ExprHasProperty(pA,EP_WinFunc)!=ExprHasProperty(pB,EP_WinFunc) ){ + return 2; + } + if( ExprHasProperty(pA,EP_WinFunc) ){ + if( sqlite3WindowCompare(pParse, pA->y.pWin, pB->y.pWin, 1)!=0 ){ + return 2; + } + } +#endif + }else if( pA->op==TK_NULL ){ + return 0; + }else if( pA->op==TK_COLLATE ){ + if( sqlite3_stricmp(pA->u.zToken,pB->u.zToken)!=0 ) return 2; + }else if( ALWAYS(pB->u.zToken!=0) && strcmp(pA->u.zToken,pB->u.zToken)!=0 ){ + return 2; + } + } + if( (pA->flags & EP_Distinct)!=(pB->flags & EP_Distinct) ) return 2; + if( (combinedFlags & EP_TokenOnly)==0 ){ + if( combinedFlags & EP_xIsSelect ) return 2; + if( (combinedFlags & EP_FixedCol)==0 + && sqlite3ExprCompare(pParse, pA->pLeft, pB->pLeft, iTab) ) return 2; + if( sqlite3ExprCompare(pParse, pA->pRight, pB->pRight, iTab) ) return 2; + if( sqlite3ExprListCompare(pA->x.pList, pB->x.pList, iTab) ) return 2; + if( pA->op!=TK_STRING + && pA->op!=TK_TRUEFALSE + && (combinedFlags & EP_Reduced)==0 + ){ + if( pA->iColumn!=pB->iColumn ) return 2; + if( pA->op2!=pB->op2 ) return 2; + if( pA->op!=TK_IN + && pA->iTable!=pB->iTable + && (pA->iTable!=iTab || NEVER(pB->iTable>=0)) ) return 2; + } + } + return 0; +} + +/* +** Compare two ExprList objects. Return 0 if they are identical and +** non-zero if they differ in any way. +** +** If any subelement of pB has Expr.iTable==(-1) then it is allowed +** to compare equal to an equivalent element in pA with Expr.iTable==iTab. +** +** This routine might return non-zero for equivalent ExprLists. The +** only consequence will be disabled optimizations. But this routine +** must never return 0 if the two ExprList objects are different, or +** a malfunction will result. +** +** Two NULL pointers are considered to be the same. But a NULL pointer +** always differs from a non-NULL pointer. +*/ +SQLITE_PRIVATE int sqlite3ExprListCompare(ExprList *pA, ExprList *pB, int iTab){ + int i; + if( pA==0 && pB==0 ) return 0; + if( pA==0 || pB==0 ) return 1; + if( pA->nExpr!=pB->nExpr ) return 1; + for(i=0; inExpr; i++){ + Expr *pExprA = pA->a[i].pExpr; + Expr *pExprB = pB->a[i].pExpr; + if( pA->a[i].sortFlags!=pB->a[i].sortFlags ) return 1; + if( sqlite3ExprCompare(0, pExprA, pExprB, iTab) ) return 1; + } + return 0; +} + +/* +** Like sqlite3ExprCompare() except COLLATE operators at the top-level +** are ignored. +*/ +SQLITE_PRIVATE int sqlite3ExprCompareSkip(Expr *pA, Expr *pB, int iTab){ + return sqlite3ExprCompare(0, + sqlite3ExprSkipCollateAndLikely(pA), + sqlite3ExprSkipCollateAndLikely(pB), + iTab); +} + +/* +** Return non-zero if Expr p can only be true if pNN is not NULL. +** +** Or if seenNot is true, return non-zero if Expr p can only be +** non-NULL if pNN is not NULL +*/ +static int exprImpliesNotNull( + Parse *pParse, /* Parsing context */ + Expr *p, /* The expression to be checked */ + Expr *pNN, /* The expression that is NOT NULL */ + int iTab, /* Table being evaluated */ + int seenNot /* Return true only if p can be any non-NULL value */ +){ + assert( p ); + assert( pNN ); + if( sqlite3ExprCompare(pParse, p, pNN, iTab)==0 ){ + return pNN->op!=TK_NULL; + } + switch( p->op ){ + case TK_IN: { + if( seenNot && ExprHasProperty(p, EP_xIsSelect) ) return 0; + assert( ExprHasProperty(p,EP_xIsSelect) + || (p->x.pList!=0 && p->x.pList->nExpr>0) ); + return exprImpliesNotNull(pParse, p->pLeft, pNN, iTab, 1); + } + case TK_BETWEEN: { + ExprList *pList = p->x.pList; + assert( pList!=0 ); + assert( pList->nExpr==2 ); + if( seenNot ) return 0; + if( exprImpliesNotNull(pParse, pList->a[0].pExpr, pNN, iTab, 1) + || exprImpliesNotNull(pParse, pList->a[1].pExpr, pNN, iTab, 1) + ){ + return 1; + } + return exprImpliesNotNull(pParse, p->pLeft, pNN, iTab, 1); + } + case TK_EQ: + case TK_NE: + case TK_LT: + case TK_LE: + case TK_GT: + case TK_GE: + case TK_PLUS: + case TK_MINUS: + case TK_BITOR: + case TK_LSHIFT: + case TK_RSHIFT: + case TK_CONCAT: + seenNot = 1; + /* Fall thru */ + case TK_STAR: + case TK_REM: + case TK_BITAND: + case TK_SLASH: { + if( exprImpliesNotNull(pParse, p->pRight, pNN, iTab, seenNot) ) return 1; + /* Fall thru into the next case */ + } + case TK_SPAN: + case TK_COLLATE: + case TK_UPLUS: + case TK_UMINUS: { + return exprImpliesNotNull(pParse, p->pLeft, pNN, iTab, seenNot); + } + case TK_TRUTH: { + if( seenNot ) return 0; + if( p->op2!=TK_IS ) return 0; + return exprImpliesNotNull(pParse, p->pLeft, pNN, iTab, 1); + } + case TK_BITNOT: + case TK_NOT: { + return exprImpliesNotNull(pParse, p->pLeft, pNN, iTab, 1); + } + } + return 0; +} + +/* +** Return true if we can prove the pE2 will always be true if pE1 is +** true. Return false if we cannot complete the proof or if pE2 might +** be false. Examples: +** +** pE1: x==5 pE2: x==5 Result: true +** pE1: x>0 pE2: x==5 Result: false +** pE1: x=21 pE2: x=21 OR y=43 Result: true +** pE1: x!=123 pE2: x IS NOT NULL Result: true +** pE1: x!=?1 pE2: x IS NOT NULL Result: true +** pE1: x IS NULL pE2: x IS NOT NULL Result: false +** pE1: x IS ?2 pE2: x IS NOT NULL Reuslt: false +** +** When comparing TK_COLUMN nodes between pE1 and pE2, if pE2 has +** Expr.iTable<0 then assume a table number given by iTab. +** +** If pParse is not NULL, then the values of bound variables in pE1 are +** compared against literal values in pE2 and pParse->pVdbe->expmask is +** modified to record which bound variables are referenced. If pParse +** is NULL, then false will be returned if pE1 contains any bound variables. +** +** When in doubt, return false. Returning true might give a performance +** improvement. Returning false might cause a performance reduction, but +** it will always give the correct answer and is hence always safe. +*/ +SQLITE_PRIVATE int sqlite3ExprImpliesExpr(Parse *pParse, Expr *pE1, Expr *pE2, int iTab){ + if( sqlite3ExprCompare(pParse, pE1, pE2, iTab)==0 ){ + return 1; + } + if( pE2->op==TK_OR + && (sqlite3ExprImpliesExpr(pParse, pE1, pE2->pLeft, iTab) + || sqlite3ExprImpliesExpr(pParse, pE1, pE2->pRight, iTab) ) + ){ + return 1; + } + if( pE2->op==TK_NOTNULL + && exprImpliesNotNull(pParse, pE1, pE2->pLeft, iTab, 0) + ){ + return 1; + } + return 0; +} + +/* +** This is the Expr node callback for sqlite3ExprImpliesNotNullRow(). +** If the expression node requires that the table at pWalker->iCur +** have one or more non-NULL column, then set pWalker->eCode to 1 and abort. +** +** This routine controls an optimization. False positives (setting +** pWalker->eCode to 1 when it should not be) are deadly, but false-negatives +** (never setting pWalker->eCode) is a harmless missed optimization. +*/ +static int impliesNotNullRow(Walker *pWalker, Expr *pExpr){ + testcase( pExpr->op==TK_AGG_COLUMN ); + testcase( pExpr->op==TK_AGG_FUNCTION ); + if( ExprHasProperty(pExpr, EP_FromJoin) ) return WRC_Prune; + switch( pExpr->op ){ + case TK_ISNOT: + case TK_ISNULL: + case TK_NOTNULL: + case TK_IS: + case TK_OR: + case TK_CASE: + case TK_IN: + case TK_FUNCTION: + case TK_TRUTH: + testcase( pExpr->op==TK_ISNOT ); + testcase( pExpr->op==TK_ISNULL ); + testcase( pExpr->op==TK_NOTNULL ); + testcase( pExpr->op==TK_IS ); + testcase( pExpr->op==TK_OR ); + testcase( pExpr->op==TK_CASE ); + testcase( pExpr->op==TK_IN ); + testcase( pExpr->op==TK_FUNCTION ); + testcase( pExpr->op==TK_TRUTH ); + return WRC_Prune; + case TK_COLUMN: + if( pWalker->u.iCur==pExpr->iTable ){ + pWalker->eCode = 1; + return WRC_Abort; + } + return WRC_Prune; + + case TK_AND: + if( sqlite3ExprImpliesNonNullRow(pExpr->pLeft, pWalker->u.iCur) + && sqlite3ExprImpliesNonNullRow(pExpr->pRight, pWalker->u.iCur) + ){ + pWalker->eCode = 1; + } + return WRC_Prune; + + case TK_BETWEEN: + sqlite3WalkExpr(pWalker, pExpr->pLeft); + return WRC_Prune; + + /* Virtual tables are allowed to use constraints like x=NULL. So + ** a term of the form x=y does not prove that y is not null if x + ** is the column of a virtual table */ + case TK_EQ: + case TK_NE: + case TK_LT: + case TK_LE: + case TK_GT: + case TK_GE: + testcase( pExpr->op==TK_EQ ); + testcase( pExpr->op==TK_NE ); + testcase( pExpr->op==TK_LT ); + testcase( pExpr->op==TK_LE ); + testcase( pExpr->op==TK_GT ); + testcase( pExpr->op==TK_GE ); + if( (pExpr->pLeft->op==TK_COLUMN && IsVirtual(pExpr->pLeft->y.pTab)) + || (pExpr->pRight->op==TK_COLUMN && IsVirtual(pExpr->pRight->y.pTab)) + ){ + return WRC_Prune; + } + + default: + return WRC_Continue; + } +} + +/* +** Return true (non-zero) if expression p can only be true if at least +** one column of table iTab is non-null. In other words, return true +** if expression p will always be NULL or false if every column of iTab +** is NULL. +** +** False negatives are acceptable. In other words, it is ok to return +** zero even if expression p will never be true of every column of iTab +** is NULL. A false negative is merely a missed optimization opportunity. +** +** False positives are not allowed, however. A false positive may result +** in an incorrect answer. +** +** Terms of p that are marked with EP_FromJoin (and hence that come from +** the ON or USING clauses of LEFT JOINS) are excluded from the analysis. +** +** This routine is used to check if a LEFT JOIN can be converted into +** an ordinary JOIN. The p argument is the WHERE clause. If the WHERE +** clause requires that some column of the right table of the LEFT JOIN +** be non-NULL, then the LEFT JOIN can be safely converted into an +** ordinary join. +*/ +SQLITE_PRIVATE int sqlite3ExprImpliesNonNullRow(Expr *p, int iTab){ + Walker w; + p = sqlite3ExprSkipCollateAndLikely(p); + while( p ){ + if( p->op==TK_NOTNULL ){ + p = p->pLeft; + }else if( p->op==TK_AND ){ + if( sqlite3ExprImpliesNonNullRow(p->pLeft, iTab) ) return 1; + p = p->pRight; + }else{ + break; + } + } + w.xExprCallback = impliesNotNullRow; + w.xSelectCallback = 0; + w.xSelectCallback2 = 0; + w.eCode = 0; + w.u.iCur = iTab; + sqlite3WalkExpr(&w, p); + return w.eCode; +} + +/* +** An instance of the following structure is used by the tree walker +** to determine if an expression can be evaluated by reference to the +** index only, without having to do a search for the corresponding +** table entry. The IdxCover.pIdx field is the index. IdxCover.iCur +** is the cursor for the table. +*/ +struct IdxCover { + Index *pIdx; /* The index to be tested for coverage */ + int iCur; /* Cursor number for the table corresponding to the index */ +}; + +/* +** Check to see if there are references to columns in table +** pWalker->u.pIdxCover->iCur can be satisfied using the index +** pWalker->u.pIdxCover->pIdx. +*/ +static int exprIdxCover(Walker *pWalker, Expr *pExpr){ + if( pExpr->op==TK_COLUMN + && pExpr->iTable==pWalker->u.pIdxCover->iCur + && sqlite3ColumnOfIndex(pWalker->u.pIdxCover->pIdx, pExpr->iColumn)<0 + ){ + pWalker->eCode = 1; + return WRC_Abort; + } + return WRC_Continue; +} + +/* +** Determine if an index pIdx on table with cursor iCur contains will +** the expression pExpr. Return true if the index does cover the +** expression and false if the pExpr expression references table columns +** that are not found in the index pIdx. +** +** An index covering an expression means that the expression can be +** evaluated using only the index and without having to lookup the +** corresponding table entry. +*/ +SQLITE_PRIVATE int sqlite3ExprCoveredByIndex( + Expr *pExpr, /* The index to be tested */ + int iCur, /* The cursor number for the corresponding table */ + Index *pIdx /* The index that might be used for coverage */ +){ + Walker w; + struct IdxCover xcov; + memset(&w, 0, sizeof(w)); + xcov.iCur = iCur; + xcov.pIdx = pIdx; + w.xExprCallback = exprIdxCover; + w.u.pIdxCover = &xcov; + sqlite3WalkExpr(&w, pExpr); + return !w.eCode; +} + + +/* +** An instance of the following structure is used by the tree walker +** to count references to table columns in the arguments of an +** aggregate function, in order to implement the +** sqlite3FunctionThisSrc() routine. +*/ +struct SrcCount { + SrcList *pSrc; /* One particular FROM clause in a nested query */ + int nThis; /* Number of references to columns in pSrcList */ + int nOther; /* Number of references to columns in other FROM clauses */ +}; + +/* +** Count the number of references to columns. +*/ +static int exprSrcCount(Walker *pWalker, Expr *pExpr){ + /* The NEVER() on the second term is because sqlite3FunctionUsesThisSrc() + ** is always called before sqlite3ExprAnalyzeAggregates() and so the + ** TK_COLUMNs have not yet been converted into TK_AGG_COLUMN. If + ** sqlite3FunctionUsesThisSrc() is used differently in the future, the + ** NEVER() will need to be removed. */ + if( pExpr->op==TK_COLUMN || NEVER(pExpr->op==TK_AGG_COLUMN) ){ + int i; + struct SrcCount *p = pWalker->u.pSrcCount; + SrcList *pSrc = p->pSrc; + int nSrc = pSrc ? pSrc->nSrc : 0; + for(i=0; iiTable==pSrc->a[i].iCursor ) break; + } + if( inThis++; + }else if( nSrc==0 || pExpr->iTablea[0].iCursor ){ + /* In a well-formed parse tree (no name resolution errors), + ** TK_COLUMN nodes with smaller Expr.iTable values are in an + ** outer context. Those are the only ones to count as "other" */ + p->nOther++; + } + } + return WRC_Continue; +} + +/* +** Determine if any of the arguments to the pExpr Function reference +** pSrcList. Return true if they do. Also return true if the function +** has no arguments or has only constant arguments. Return false if pExpr +** references columns but not columns of tables found in pSrcList. +*/ +SQLITE_PRIVATE int sqlite3FunctionUsesThisSrc(Expr *pExpr, SrcList *pSrcList){ + Walker w; + struct SrcCount cnt; + assert( pExpr->op==TK_AGG_FUNCTION ); + memset(&w, 0, sizeof(w)); + w.xExprCallback = exprSrcCount; + w.xSelectCallback = sqlite3SelectWalkNoop; + w.u.pSrcCount = &cnt; + cnt.pSrc = pSrcList; + cnt.nThis = 0; + cnt.nOther = 0; + sqlite3WalkExprList(&w, pExpr->x.pList); + return cnt.nThis>0 || cnt.nOther==0; +} + +/* +** Add a new element to the pAggInfo->aCol[] array. Return the index of +** the new element. Return a negative number if malloc fails. +*/ +static int addAggInfoColumn(sqlite3 *db, AggInfo *pInfo){ + int i; + pInfo->aCol = sqlite3ArrayAllocate( + db, + pInfo->aCol, + sizeof(pInfo->aCol[0]), + &pInfo->nColumn, + &i + ); + return i; +} + +/* +** Add a new element to the pAggInfo->aFunc[] array. Return the index of +** the new element. Return a negative number if malloc fails. +*/ +static int addAggInfoFunc(sqlite3 *db, AggInfo *pInfo){ + int i; + pInfo->aFunc = sqlite3ArrayAllocate( + db, + pInfo->aFunc, + sizeof(pInfo->aFunc[0]), + &pInfo->nFunc, + &i + ); + return i; +} + +/* +** This is the xExprCallback for a tree walker. It is used to +** implement sqlite3ExprAnalyzeAggregates(). See sqlite3ExprAnalyzeAggregates +** for additional information. +*/ +static int analyzeAggregate(Walker *pWalker, Expr *pExpr){ + int i; + NameContext *pNC = pWalker->u.pNC; + Parse *pParse = pNC->pParse; + SrcList *pSrcList = pNC->pSrcList; + AggInfo *pAggInfo = pNC->uNC.pAggInfo; + + assert( pNC->ncFlags & NC_UAggInfo ); + switch( pExpr->op ){ + case TK_AGG_COLUMN: + case TK_COLUMN: { + testcase( pExpr->op==TK_AGG_COLUMN ); + testcase( pExpr->op==TK_COLUMN ); + /* Check to see if the column is in one of the tables in the FROM + ** clause of the aggregate query */ + if( ALWAYS(pSrcList!=0) ){ + struct SrcList_item *pItem = pSrcList->a; + for(i=0; inSrc; i++, pItem++){ + struct AggInfo_col *pCol; + assert( !ExprHasProperty(pExpr, EP_TokenOnly|EP_Reduced) ); + if( pExpr->iTable==pItem->iCursor ){ + /* If we reach this point, it means that pExpr refers to a table + ** that is in the FROM clause of the aggregate query. + ** + ** Make an entry for the column in pAggInfo->aCol[] if there + ** is not an entry there already. + */ + int k; + pCol = pAggInfo->aCol; + for(k=0; knColumn; k++, pCol++){ + if( pCol->iTable==pExpr->iTable && + pCol->iColumn==pExpr->iColumn ){ + break; + } + } + if( (k>=pAggInfo->nColumn) + && (k = addAggInfoColumn(pParse->db, pAggInfo))>=0 + ){ + pCol = &pAggInfo->aCol[k]; + pCol->pTab = pExpr->y.pTab; + pCol->iTable = pExpr->iTable; + pCol->iColumn = pExpr->iColumn; + pCol->iMem = ++pParse->nMem; + pCol->iSorterColumn = -1; + pCol->pExpr = pExpr; + if( pAggInfo->pGroupBy ){ + int j, n; + ExprList *pGB = pAggInfo->pGroupBy; + struct ExprList_item *pTerm = pGB->a; + n = pGB->nExpr; + for(j=0; jpExpr; + if( pE->op==TK_COLUMN && pE->iTable==pExpr->iTable && + pE->iColumn==pExpr->iColumn ){ + pCol->iSorterColumn = j; + break; + } + } + } + if( pCol->iSorterColumn<0 ){ + pCol->iSorterColumn = pAggInfo->nSortingColumn++; + } + } + /* There is now an entry for pExpr in pAggInfo->aCol[] (either + ** because it was there before or because we just created it). + ** Convert the pExpr to be a TK_AGG_COLUMN referring to that + ** pAggInfo->aCol[] entry. + */ + ExprSetVVAProperty(pExpr, EP_NoReduce); + pExpr->pAggInfo = pAggInfo; + pExpr->op = TK_AGG_COLUMN; + pExpr->iAgg = (i16)k; + break; + } /* endif pExpr->iTable==pItem->iCursor */ + } /* end loop over pSrcList */ + } + return WRC_Prune; + } + case TK_AGG_FUNCTION: { + if( (pNC->ncFlags & NC_InAggFunc)==0 + && pWalker->walkerDepth==pExpr->op2 + ){ + /* Check to see if pExpr is a duplicate of another aggregate + ** function that is already in the pAggInfo structure + */ + struct AggInfo_func *pItem = pAggInfo->aFunc; + for(i=0; inFunc; i++, pItem++){ + if( sqlite3ExprCompare(0, pItem->pExpr, pExpr, -1)==0 ){ + break; + } + } + if( i>=pAggInfo->nFunc ){ + /* pExpr is original. Make a new entry in pAggInfo->aFunc[] + */ + u8 enc = ENC(pParse->db); + i = addAggInfoFunc(pParse->db, pAggInfo); + if( i>=0 ){ + assert( !ExprHasProperty(pExpr, EP_xIsSelect) ); + pItem = &pAggInfo->aFunc[i]; + pItem->pExpr = pExpr; + pItem->iMem = ++pParse->nMem; + assert( !ExprHasProperty(pExpr, EP_IntValue) ); + pItem->pFunc = sqlite3FindFunction(pParse->db, + pExpr->u.zToken, + pExpr->x.pList ? pExpr->x.pList->nExpr : 0, enc, 0); + if( pExpr->flags & EP_Distinct ){ + pItem->iDistinct = pParse->nTab++; + }else{ + pItem->iDistinct = -1; + } + } + } + /* Make pExpr point to the appropriate pAggInfo->aFunc[] entry + */ + assert( !ExprHasProperty(pExpr, EP_TokenOnly|EP_Reduced) ); + ExprSetVVAProperty(pExpr, EP_NoReduce); + pExpr->iAgg = (i16)i; + pExpr->pAggInfo = pAggInfo; + return WRC_Prune; + }else{ + return WRC_Continue; + } + } + } + return WRC_Continue; +} +static int analyzeAggregatesInSelect(Walker *pWalker, Select *pSelect){ + UNUSED_PARAMETER(pSelect); + pWalker->walkerDepth++; + return WRC_Continue; +} +static void analyzeAggregatesInSelectEnd(Walker *pWalker, Select *pSelect){ + UNUSED_PARAMETER(pSelect); + pWalker->walkerDepth--; +} + +/* +** Analyze the pExpr expression looking for aggregate functions and +** for variables that need to be added to AggInfo object that pNC->pAggInfo +** points to. Additional entries are made on the AggInfo object as +** necessary. +** +** This routine should only be called after the expression has been +** analyzed by sqlite3ResolveExprNames(). +*/ +SQLITE_PRIVATE void sqlite3ExprAnalyzeAggregates(NameContext *pNC, Expr *pExpr){ + Walker w; + w.xExprCallback = analyzeAggregate; + w.xSelectCallback = analyzeAggregatesInSelect; + w.xSelectCallback2 = analyzeAggregatesInSelectEnd; + w.walkerDepth = 0; + w.u.pNC = pNC; + w.pParse = 0; + assert( pNC->pSrcList!=0 ); + sqlite3WalkExpr(&w, pExpr); +} + +/* +** Call sqlite3ExprAnalyzeAggregates() for every expression in an +** expression list. Return the number of errors. +** +** If an error is found, the analysis is cut short. +*/ +SQLITE_PRIVATE void sqlite3ExprAnalyzeAggList(NameContext *pNC, ExprList *pList){ + struct ExprList_item *pItem; + int i; + if( pList ){ + for(pItem=pList->a, i=0; inExpr; i++, pItem++){ + sqlite3ExprAnalyzeAggregates(pNC, pItem->pExpr); + } + } +} + +/* +** Allocate a single new register for use to hold some intermediate result. +*/ +SQLITE_PRIVATE int sqlite3GetTempReg(Parse *pParse){ + if( pParse->nTempReg==0 ){ + return ++pParse->nMem; + } + return pParse->aTempReg[--pParse->nTempReg]; +} + +/* +** Deallocate a register, making available for reuse for some other +** purpose. +*/ +SQLITE_PRIVATE void sqlite3ReleaseTempReg(Parse *pParse, int iReg){ + if( iReg && pParse->nTempRegaTempReg) ){ + pParse->aTempReg[pParse->nTempReg++] = iReg; + } +} + +/* +** Allocate or deallocate a block of nReg consecutive registers. +*/ +SQLITE_PRIVATE int sqlite3GetTempRange(Parse *pParse, int nReg){ + int i, n; + if( nReg==1 ) return sqlite3GetTempReg(pParse); + i = pParse->iRangeReg; + n = pParse->nRangeReg; + if( nReg<=n ){ + pParse->iRangeReg += nReg; + pParse->nRangeReg -= nReg; + }else{ + i = pParse->nMem+1; + pParse->nMem += nReg; + } + return i; +} +SQLITE_PRIVATE void sqlite3ReleaseTempRange(Parse *pParse, int iReg, int nReg){ + if( nReg==1 ){ + sqlite3ReleaseTempReg(pParse, iReg); + return; + } + if( nReg>pParse->nRangeReg ){ + pParse->nRangeReg = nReg; + pParse->iRangeReg = iReg; + } +} + +/* +** Mark all temporary registers as being unavailable for reuse. +** +** Always invoke this procedure after coding a subroutine or co-routine +** that might be invoked from other parts of the code, to ensure that +** the sub/co-routine does not use registers in common with the code that +** invokes the sub/co-routine. +*/ +SQLITE_PRIVATE void sqlite3ClearTempRegCache(Parse *pParse){ + pParse->nTempReg = 0; + pParse->nRangeReg = 0; +} + +/* +** Validate that no temporary register falls within the range of +** iFirst..iLast, inclusive. This routine is only call from within assert() +** statements. +*/ +#ifdef SQLITE_DEBUG +SQLITE_PRIVATE int sqlite3NoTempsInRange(Parse *pParse, int iFirst, int iLast){ + int i; + if( pParse->nRangeReg>0 + && pParse->iRangeReg+pParse->nRangeReg > iFirst + && pParse->iRangeReg <= iLast + ){ + return 0; + } + for(i=0; inTempReg; i++){ + if( pParse->aTempReg[i]>=iFirst && pParse->aTempReg[i]<=iLast ){ + return 0; + } + } + return 1; +} +#endif /* SQLITE_DEBUG */ + +/************** End of expr.c ************************************************/ +/************** Begin file alter.c *******************************************/ +/* +** 2005 February 15 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** This file contains C code routines that used to generate VDBE code +** that implements the ALTER TABLE command. +*/ +/* #include "sqliteInt.h" */ + +/* +** The code in this file only exists if we are not omitting the +** ALTER TABLE logic from the build. +*/ +#ifndef SQLITE_OMIT_ALTERTABLE + +/* +** Parameter zName is the name of a table that is about to be altered +** (either with ALTER TABLE ... RENAME TO or ALTER TABLE ... ADD COLUMN). +** If the table is a system table, this function leaves an error message +** in pParse->zErr (system tables may not be altered) and returns non-zero. +** +** Or, if zName is not a system table, zero is returned. +*/ +static int isAlterableTable(Parse *pParse, Table *pTab){ + if( 0==sqlite3StrNICmp(pTab->zName, "sqlite_", 7) +#ifndef SQLITE_OMIT_VIRTUALTABLE + || ( (pTab->tabFlags & TF_Shadow) + && (pParse->db->flags & SQLITE_Defensive) + && pParse->db->nVdbeExec==0 + ) +#endif + ){ + sqlite3ErrorMsg(pParse, "table %s may not be altered", pTab->zName); + return 1; + } + return 0; +} + +/* +** Generate code to verify that the schemas of database zDb and, if +** bTemp is not true, database "temp", can still be parsed. This is +** called at the end of the generation of an ALTER TABLE ... RENAME ... +** statement to ensure that the operation has not rendered any schema +** objects unusable. +*/ +static void renameTestSchema(Parse *pParse, const char *zDb, int bTemp){ + sqlite3NestedParse(pParse, + "SELECT 1 " + "FROM \"%w\".%s " + "WHERE name NOT LIKE 'sqliteX_%%' ESCAPE 'X'" + " AND sql NOT LIKE 'create virtual%%'" + " AND sqlite_rename_test(%Q, sql, type, name, %d)=NULL ", + zDb, MASTER_NAME, + zDb, bTemp + ); + + if( bTemp==0 ){ + sqlite3NestedParse(pParse, + "SELECT 1 " + "FROM temp.%s " + "WHERE name NOT LIKE 'sqliteX_%%' ESCAPE 'X'" + " AND sql NOT LIKE 'create virtual%%'" + " AND sqlite_rename_test(%Q, sql, type, name, 1)=NULL ", + MASTER_NAME, zDb + ); + } +} + +/* +** Generate code to reload the schema for database iDb. And, if iDb!=1, for +** the temp database as well. +*/ +static void renameReloadSchema(Parse *pParse, int iDb){ + Vdbe *v = pParse->pVdbe; + if( v ){ + sqlite3ChangeCookie(pParse, iDb); + sqlite3VdbeAddParseSchemaOp(pParse->pVdbe, iDb, 0); + if( iDb!=1 ) sqlite3VdbeAddParseSchemaOp(pParse->pVdbe, 1, 0); + } +} + +/* +** Generate code to implement the "ALTER TABLE xxx RENAME TO yyy" +** command. +*/ +SQLITE_PRIVATE void sqlite3AlterRenameTable( + Parse *pParse, /* Parser context. */ + SrcList *pSrc, /* The table to rename. */ + Token *pName /* The new table name. */ +){ + int iDb; /* Database that contains the table */ + char *zDb; /* Name of database iDb */ + Table *pTab; /* Table being renamed */ + char *zName = 0; /* NULL-terminated version of pName */ + sqlite3 *db = pParse->db; /* Database connection */ + int nTabName; /* Number of UTF-8 characters in zTabName */ + const char *zTabName; /* Original name of the table */ + Vdbe *v; + VTable *pVTab = 0; /* Non-zero if this is a v-tab with an xRename() */ + u32 savedDbFlags; /* Saved value of db->mDbFlags */ + + savedDbFlags = db->mDbFlags; + if( NEVER(db->mallocFailed) ) goto exit_rename_table; + assert( pSrc->nSrc==1 ); + assert( sqlite3BtreeHoldsAllMutexes(pParse->db) ); + + pTab = sqlite3LocateTableItem(pParse, 0, &pSrc->a[0]); + if( !pTab ) goto exit_rename_table; + iDb = sqlite3SchemaToIndex(pParse->db, pTab->pSchema); + zDb = db->aDb[iDb].zDbSName; + db->mDbFlags |= DBFLAG_PreferBuiltin; + + /* Get a NULL terminated version of the new table name. */ + zName = sqlite3NameFromToken(db, pName); + if( !zName ) goto exit_rename_table; + + /* Check that a table or index named 'zName' does not already exist + ** in database iDb. If so, this is an error. + */ + if( sqlite3FindTable(db, zName, zDb) || sqlite3FindIndex(db, zName, zDb) ){ + sqlite3ErrorMsg(pParse, + "there is already another table or index with this name: %s", zName); + goto exit_rename_table; + } + + /* Make sure it is not a system table being altered, or a reserved name + ** that the table is being renamed to. + */ + if( SQLITE_OK!=isAlterableTable(pParse, pTab) ){ + goto exit_rename_table; + } + if( SQLITE_OK!=sqlite3CheckObjectName(pParse,zName,"table",zName) ){ + goto exit_rename_table; + } + +#ifndef SQLITE_OMIT_VIEW + if( pTab->pSelect ){ + sqlite3ErrorMsg(pParse, "view %s may not be altered", pTab->zName); + goto exit_rename_table; + } +#endif + +#ifndef SQLITE_OMIT_AUTHORIZATION + /* Invoke the authorization callback. */ + if( sqlite3AuthCheck(pParse, SQLITE_ALTER_TABLE, zDb, pTab->zName, 0) ){ + goto exit_rename_table; + } +#endif + +#ifndef SQLITE_OMIT_VIRTUALTABLE + if( sqlite3ViewGetColumnNames(pParse, pTab) ){ + goto exit_rename_table; + } + if( IsVirtual(pTab) ){ + pVTab = sqlite3GetVTable(db, pTab); + if( pVTab->pVtab->pModule->xRename==0 ){ + pVTab = 0; + } + } +#endif + + /* Begin a transaction for database iDb. Then modify the schema cookie + ** (since the ALTER TABLE modifies the schema). Call sqlite3MayAbort(), + ** as the scalar functions (e.g. sqlite_rename_table()) invoked by the + ** nested SQL may raise an exception. */ + v = sqlite3GetVdbe(pParse); + if( v==0 ){ + goto exit_rename_table; + } + sqlite3MayAbort(pParse); + + /* figure out how many UTF-8 characters are in zName */ + zTabName = pTab->zName; + nTabName = sqlite3Utf8CharLen(zTabName, -1); + + /* Rewrite all CREATE TABLE, INDEX, TRIGGER or VIEW statements in + ** the schema to use the new table name. */ + sqlite3NestedParse(pParse, + "UPDATE \"%w\".%s SET " + "sql = sqlite_rename_table(%Q, type, name, sql, %Q, %Q, %d) " + "WHERE (type!='index' OR tbl_name=%Q COLLATE nocase)" + "AND name NOT LIKE 'sqliteX_%%' ESCAPE 'X'" + , zDb, MASTER_NAME, zDb, zTabName, zName, (iDb==1), zTabName + ); + + /* Update the tbl_name and name columns of the sqlite_master table + ** as required. */ + sqlite3NestedParse(pParse, + "UPDATE %Q.%s SET " + "tbl_name = %Q, " + "name = CASE " + "WHEN type='table' THEN %Q " + "WHEN name LIKE 'sqliteX_autoindex%%' ESCAPE 'X' " + " AND type='index' THEN " + "'sqlite_autoindex_' || %Q || substr(name,%d+18) " + "ELSE name END " + "WHERE tbl_name=%Q COLLATE nocase AND " + "(type='table' OR type='index' OR type='trigger');", + zDb, MASTER_NAME, + zName, zName, zName, + nTabName, zTabName + ); + +#ifndef SQLITE_OMIT_AUTOINCREMENT + /* If the sqlite_sequence table exists in this database, then update + ** it with the new table name. + */ + if( sqlite3FindTable(db, "sqlite_sequence", zDb) ){ + sqlite3NestedParse(pParse, + "UPDATE \"%w\".sqlite_sequence set name = %Q WHERE name = %Q", + zDb, zName, pTab->zName); + } +#endif + + /* If the table being renamed is not itself part of the temp database, + ** edit view and trigger definitions within the temp database + ** as required. */ + if( iDb!=1 ){ + sqlite3NestedParse(pParse, + "UPDATE sqlite_temp_master SET " + "sql = sqlite_rename_table(%Q, type, name, sql, %Q, %Q, 1), " + "tbl_name = " + "CASE WHEN tbl_name=%Q COLLATE nocase AND " + " sqlite_rename_test(%Q, sql, type, name, 1) " + "THEN %Q ELSE tbl_name END " + "WHERE type IN ('view', 'trigger')" + , zDb, zTabName, zName, zTabName, zDb, zName); + } + + /* If this is a virtual table, invoke the xRename() function if + ** one is defined. The xRename() callback will modify the names + ** of any resources used by the v-table implementation (including other + ** SQLite tables) that are identified by the name of the virtual table. + */ +#ifndef SQLITE_OMIT_VIRTUALTABLE + if( pVTab ){ + int i = ++pParse->nMem; + sqlite3VdbeLoadString(v, i, zName); + sqlite3VdbeAddOp4(v, OP_VRename, i, 0, 0,(const char*)pVTab, P4_VTAB); + } +#endif + + renameReloadSchema(pParse, iDb); + renameTestSchema(pParse, zDb, iDb==1); + +exit_rename_table: + sqlite3SrcListDelete(db, pSrc); + sqlite3DbFree(db, zName); + db->mDbFlags = savedDbFlags; +} + +/* +** This function is called after an "ALTER TABLE ... ADD" statement +** has been parsed. Argument pColDef contains the text of the new +** column definition. +** +** The Table structure pParse->pNewTable was extended to include +** the new column during parsing. +*/ +SQLITE_PRIVATE void sqlite3AlterFinishAddColumn(Parse *pParse, Token *pColDef){ + Table *pNew; /* Copy of pParse->pNewTable */ + Table *pTab; /* Table being altered */ + int iDb; /* Database number */ + const char *zDb; /* Database name */ + const char *zTab; /* Table name */ + char *zCol; /* Null-terminated column definition */ + Column *pCol; /* The new column */ + Expr *pDflt; /* Default value for the new column */ + sqlite3 *db; /* The database connection; */ + Vdbe *v; /* The prepared statement under construction */ + int r1; /* Temporary registers */ + + db = pParse->db; + if( pParse->nErr || db->mallocFailed ) return; + pNew = pParse->pNewTable; + assert( pNew ); + + assert( sqlite3BtreeHoldsAllMutexes(db) ); + iDb = sqlite3SchemaToIndex(db, pNew->pSchema); + zDb = db->aDb[iDb].zDbSName; + zTab = &pNew->zName[16]; /* Skip the "sqlite_altertab_" prefix on the name */ + pCol = &pNew->aCol[pNew->nCol-1]; + pDflt = pCol->pDflt; + pTab = sqlite3FindTable(db, zTab, zDb); + assert( pTab ); + +#ifndef SQLITE_OMIT_AUTHORIZATION + /* Invoke the authorization callback. */ + if( sqlite3AuthCheck(pParse, SQLITE_ALTER_TABLE, zDb, pTab->zName, 0) ){ + return; + } +#endif + + /* If the default value for the new column was specified with a + ** literal NULL, then set pDflt to 0. This simplifies checking + ** for an SQL NULL default below. + */ + assert( pDflt==0 || pDflt->op==TK_SPAN ); + if( pDflt && pDflt->pLeft->op==TK_NULL ){ + pDflt = 0; + } + + /* Check that the new column is not specified as PRIMARY KEY or UNIQUE. + ** If there is a NOT NULL constraint, then the default value for the + ** column must not be NULL. + */ + if( pCol->colFlags & COLFLAG_PRIMKEY ){ + sqlite3ErrorMsg(pParse, "Cannot add a PRIMARY KEY column"); + return; + } + if( pNew->pIndex ){ + sqlite3ErrorMsg(pParse, "Cannot add a UNIQUE column"); + return; + } + if( (db->flags&SQLITE_ForeignKeys) && pNew->pFKey && pDflt ){ + sqlite3ErrorMsg(pParse, + "Cannot add a REFERENCES column with non-NULL default value"); + return; + } + if( pCol->notNull && !pDflt ){ + sqlite3ErrorMsg(pParse, + "Cannot add a NOT NULL column with default value NULL"); + return; + } + + /* Ensure the default expression is something that sqlite3ValueFromExpr() + ** can handle (i.e. not CURRENT_TIME etc.) + */ + if( pDflt ){ + sqlite3_value *pVal = 0; + int rc; + rc = sqlite3ValueFromExpr(db, pDflt, SQLITE_UTF8, SQLITE_AFF_BLOB, &pVal); + assert( rc==SQLITE_OK || rc==SQLITE_NOMEM ); + if( rc!=SQLITE_OK ){ + assert( db->mallocFailed == 1 ); + return; + } + if( !pVal ){ + sqlite3ErrorMsg(pParse, "Cannot add a column with non-constant default"); + return; + } + sqlite3ValueFree(pVal); + } + + /* Modify the CREATE TABLE statement. */ + zCol = sqlite3DbStrNDup(db, (char*)pColDef->z, pColDef->n); + if( zCol ){ + char *zEnd = &zCol[pColDef->n-1]; + u32 savedDbFlags = db->mDbFlags; + while( zEnd>zCol && (*zEnd==';' || sqlite3Isspace(*zEnd)) ){ + *zEnd-- = '\0'; + } + db->mDbFlags |= DBFLAG_PreferBuiltin; + sqlite3NestedParse(pParse, + "UPDATE \"%w\".%s SET " + "sql = substr(sql,1,%d) || ', ' || %Q || substr(sql,%d) " + "WHERE type = 'table' AND name = %Q", + zDb, MASTER_NAME, pNew->addColOffset, zCol, pNew->addColOffset+1, + zTab + ); + sqlite3DbFree(db, zCol); + db->mDbFlags = savedDbFlags; + } + + /* Make sure the schema version is at least 3. But do not upgrade + ** from less than 3 to 4, as that will corrupt any preexisting DESC + ** index. + */ + v = sqlite3GetVdbe(pParse); + if( v ){ + r1 = sqlite3GetTempReg(pParse); + sqlite3VdbeAddOp3(v, OP_ReadCookie, iDb, r1, BTREE_FILE_FORMAT); + sqlite3VdbeUsesBtree(v, iDb); + sqlite3VdbeAddOp2(v, OP_AddImm, r1, -2); + sqlite3VdbeAddOp2(v, OP_IfPos, r1, sqlite3VdbeCurrentAddr(v)+2); + VdbeCoverage(v); + sqlite3VdbeAddOp3(v, OP_SetCookie, iDb, BTREE_FILE_FORMAT, 3); + sqlite3ReleaseTempReg(pParse, r1); + } + + /* Reload the table definition */ + renameReloadSchema(pParse, iDb); +} + +/* +** This function is called by the parser after the table-name in +** an "ALTER TABLE ADD" statement is parsed. Argument +** pSrc is the full-name of the table being altered. +** +** This routine makes a (partial) copy of the Table structure +** for the table being altered and sets Parse.pNewTable to point +** to it. Routines called by the parser as the column definition +** is parsed (i.e. sqlite3AddColumn()) add the new Column data to +** the copy. The copy of the Table structure is deleted by tokenize.c +** after parsing is finished. +** +** Routine sqlite3AlterFinishAddColumn() will be called to complete +** coding the "ALTER TABLE ... ADD" statement. +*/ +SQLITE_PRIVATE void sqlite3AlterBeginAddColumn(Parse *pParse, SrcList *pSrc){ + Table *pNew; + Table *pTab; + int iDb; + int i; + int nAlloc; + sqlite3 *db = pParse->db; + + /* Look up the table being altered. */ + assert( pParse->pNewTable==0 ); + assert( sqlite3BtreeHoldsAllMutexes(db) ); + if( db->mallocFailed ) goto exit_begin_add_column; + pTab = sqlite3LocateTableItem(pParse, 0, &pSrc->a[0]); + if( !pTab ) goto exit_begin_add_column; + +#ifndef SQLITE_OMIT_VIRTUALTABLE + if( IsVirtual(pTab) ){ + sqlite3ErrorMsg(pParse, "virtual tables may not be altered"); + goto exit_begin_add_column; + } +#endif + + /* Make sure this is not an attempt to ALTER a view. */ + if( pTab->pSelect ){ + sqlite3ErrorMsg(pParse, "Cannot add a column to a view"); + goto exit_begin_add_column; + } + if( SQLITE_OK!=isAlterableTable(pParse, pTab) ){ + goto exit_begin_add_column; + } + + sqlite3MayAbort(pParse); + assert( pTab->addColOffset>0 ); + iDb = sqlite3SchemaToIndex(db, pTab->pSchema); + + /* Put a copy of the Table struct in Parse.pNewTable for the + ** sqlite3AddColumn() function and friends to modify. But modify + ** the name by adding an "sqlite_altertab_" prefix. By adding this + ** prefix, we insure that the name will not collide with an existing + ** table because user table are not allowed to have the "sqlite_" + ** prefix on their name. + */ + pNew = (Table*)sqlite3DbMallocZero(db, sizeof(Table)); + if( !pNew ) goto exit_begin_add_column; + pParse->pNewTable = pNew; + pNew->nTabRef = 1; + pNew->nCol = pTab->nCol; + assert( pNew->nCol>0 ); + nAlloc = (((pNew->nCol-1)/8)*8)+8; + assert( nAlloc>=pNew->nCol && nAlloc%8==0 && nAlloc-pNew->nCol<8 ); + pNew->aCol = (Column*)sqlite3DbMallocZero(db, sizeof(Column)*nAlloc); + pNew->zName = sqlite3MPrintf(db, "sqlite_altertab_%s", pTab->zName); + if( !pNew->aCol || !pNew->zName ){ + assert( db->mallocFailed ); + goto exit_begin_add_column; + } + memcpy(pNew->aCol, pTab->aCol, sizeof(Column)*pNew->nCol); + for(i=0; inCol; i++){ + Column *pCol = &pNew->aCol[i]; + pCol->zName = sqlite3DbStrDup(db, pCol->zName); + pCol->zColl = 0; + pCol->pDflt = 0; + } + pNew->pSchema = db->aDb[iDb].pSchema; + pNew->addColOffset = pTab->addColOffset; + pNew->nTabRef = 1; + +exit_begin_add_column: + sqlite3SrcListDelete(db, pSrc); + return; +} + +/* +** Parameter pTab is the subject of an ALTER TABLE ... RENAME COLUMN +** command. This function checks if the table is a view or virtual +** table (columns of views or virtual tables may not be renamed). If so, +** it loads an error message into pParse and returns non-zero. +** +** Or, if pTab is not a view or virtual table, zero is returned. +*/ +#if !defined(SQLITE_OMIT_VIEW) || !defined(SQLITE_OMIT_VIRTUALTABLE) +static int isRealTable(Parse *pParse, Table *pTab){ + const char *zType = 0; +#ifndef SQLITE_OMIT_VIEW + if( pTab->pSelect ){ + zType = "view"; + } +#endif +#ifndef SQLITE_OMIT_VIRTUALTABLE + if( IsVirtual(pTab) ){ + zType = "virtual table"; + } +#endif + if( zType ){ + sqlite3ErrorMsg( + pParse, "cannot rename columns of %s \"%s\"", zType, pTab->zName + ); + return 1; + } + return 0; +} +#else /* !defined(SQLITE_OMIT_VIEW) || !defined(SQLITE_OMIT_VIRTUALTABLE) */ +# define isRealTable(x,y) (0) +#endif + +/* +** Handles the following parser reduction: +** +** cmd ::= ALTER TABLE pSrc RENAME COLUMN pOld TO pNew +*/ +SQLITE_PRIVATE void sqlite3AlterRenameColumn( + Parse *pParse, /* Parsing context */ + SrcList *pSrc, /* Table being altered. pSrc->nSrc==1 */ + Token *pOld, /* Name of column being changed */ + Token *pNew /* New column name */ +){ + sqlite3 *db = pParse->db; /* Database connection */ + Table *pTab; /* Table being updated */ + int iCol; /* Index of column being renamed */ + char *zOld = 0; /* Old column name */ + char *zNew = 0; /* New column name */ + const char *zDb; /* Name of schema containing the table */ + int iSchema; /* Index of the schema */ + int bQuote; /* True to quote the new name */ + + /* Locate the table to be altered */ + pTab = sqlite3LocateTableItem(pParse, 0, &pSrc->a[0]); + if( !pTab ) goto exit_rename_column; + + /* Cannot alter a system table */ + if( SQLITE_OK!=isAlterableTable(pParse, pTab) ) goto exit_rename_column; + if( SQLITE_OK!=isRealTable(pParse, pTab) ) goto exit_rename_column; + + /* Which schema holds the table to be altered */ + iSchema = sqlite3SchemaToIndex(db, pTab->pSchema); + assert( iSchema>=0 ); + zDb = db->aDb[iSchema].zDbSName; + +#ifndef SQLITE_OMIT_AUTHORIZATION + /* Invoke the authorization callback. */ + if( sqlite3AuthCheck(pParse, SQLITE_ALTER_TABLE, zDb, pTab->zName, 0) ){ + goto exit_rename_column; + } +#endif + + /* Make sure the old name really is a column name in the table to be + ** altered. Set iCol to be the index of the column being renamed */ + zOld = sqlite3NameFromToken(db, pOld); + if( !zOld ) goto exit_rename_column; + for(iCol=0; iColnCol; iCol++){ + if( 0==sqlite3StrICmp(pTab->aCol[iCol].zName, zOld) ) break; + } + if( iCol==pTab->nCol ){ + sqlite3ErrorMsg(pParse, "no such column: \"%s\"", zOld); + goto exit_rename_column; + } + + /* Do the rename operation using a recursive UPDATE statement that + ** uses the sqlite_rename_column() SQL function to compute the new + ** CREATE statement text for the sqlite_master table. + */ + sqlite3MayAbort(pParse); + zNew = sqlite3NameFromToken(db, pNew); + if( !zNew ) goto exit_rename_column; + assert( pNew->n>0 ); + bQuote = sqlite3Isquote(pNew->z[0]); + sqlite3NestedParse(pParse, + "UPDATE \"%w\".%s SET " + "sql = sqlite_rename_column(sql, type, name, %Q, %Q, %d, %Q, %d, %d) " + "WHERE name NOT LIKE 'sqliteX_%%' ESCAPE 'X' " + " AND (type != 'index' OR tbl_name = %Q)" + " AND sql NOT LIKE 'create virtual%%'", + zDb, MASTER_NAME, + zDb, pTab->zName, iCol, zNew, bQuote, iSchema==1, + pTab->zName + ); + + sqlite3NestedParse(pParse, + "UPDATE temp.%s SET " + "sql = sqlite_rename_column(sql, type, name, %Q, %Q, %d, %Q, %d, 1) " + "WHERE type IN ('trigger', 'view')", + MASTER_NAME, + zDb, pTab->zName, iCol, zNew, bQuote + ); + + /* Drop and reload the database schema. */ + renameReloadSchema(pParse, iSchema); + renameTestSchema(pParse, zDb, iSchema==1); + + exit_rename_column: + sqlite3SrcListDelete(db, pSrc); + sqlite3DbFree(db, zOld); + sqlite3DbFree(db, zNew); + return; +} + +/* +** Each RenameToken object maps an element of the parse tree into +** the token that generated that element. The parse tree element +** might be one of: +** +** * A pointer to an Expr that represents an ID +** * The name of a table column in Column.zName +** +** A list of RenameToken objects can be constructed during parsing. +** Each new object is created by sqlite3RenameTokenMap(). +** As the parse tree is transformed, the sqlite3RenameTokenRemap() +** routine is used to keep the mapping current. +** +** After the parse finishes, renameTokenFind() routine can be used +** to look up the actual token value that created some element in +** the parse tree. +*/ +struct RenameToken { + void *p; /* Parse tree element created by token t */ + Token t; /* The token that created parse tree element p */ + RenameToken *pNext; /* Next is a list of all RenameToken objects */ +}; + +/* +** The context of an ALTER TABLE RENAME COLUMN operation that gets passed +** down into the Walker. +*/ +typedef struct RenameCtx RenameCtx; +struct RenameCtx { + RenameToken *pList; /* List of tokens to overwrite */ + int nList; /* Number of tokens in pList */ + int iCol; /* Index of column being renamed */ + Table *pTab; /* Table being ALTERed */ + const char *zOld; /* Old column name */ +}; + +#ifdef SQLITE_DEBUG +/* +** This function is only for debugging. It performs two tasks: +** +** 1. Checks that pointer pPtr does not already appear in the +** rename-token list. +** +** 2. Dereferences each pointer in the rename-token list. +** +** The second is most effective when debugging under valgrind or +** address-sanitizer or similar. If any of these pointers no longer +** point to valid objects, an exception is raised by the memory-checking +** tool. +** +** The point of this is to prevent comparisons of invalid pointer values. +** Even though this always seems to work, it is undefined according to the +** C standard. Example of undefined comparison: +** +** sqlite3_free(x); +** if( x==y ) ... +** +** Technically, as x no longer points into a valid object or to the byte +** following a valid object, it may not be used in comparison operations. +*/ +static void renameTokenCheckAll(Parse *pParse, void *pPtr){ + if( pParse->nErr==0 && pParse->db->mallocFailed==0 ){ + RenameToken *p; + u8 i = 0; + for(p=pParse->pRename; p; p=p->pNext){ + if( p->p ){ + assert( p->p!=pPtr ); + i += *(u8*)(p->p); + } + } + } +} +#else +# define renameTokenCheckAll(x,y) +#endif + +/* +** Remember that the parser tree element pPtr was created using +** the token pToken. +** +** In other words, construct a new RenameToken object and add it +** to the list of RenameToken objects currently being built up +** in pParse->pRename. +** +** The pPtr argument is returned so that this routine can be used +** with tail recursion in tokenExpr() routine, for a small performance +** improvement. +*/ +SQLITE_PRIVATE void *sqlite3RenameTokenMap(Parse *pParse, void *pPtr, Token *pToken){ + RenameToken *pNew; + assert( pPtr || pParse->db->mallocFailed ); + renameTokenCheckAll(pParse, pPtr); + pNew = sqlite3DbMallocZero(pParse->db, sizeof(RenameToken)); + if( pNew ){ + pNew->p = pPtr; + pNew->t = *pToken; + pNew->pNext = pParse->pRename; + pParse->pRename = pNew; + } + + return pPtr; +} + +/* +** It is assumed that there is already a RenameToken object associated +** with parse tree element pFrom. This function remaps the associated token +** to parse tree element pTo. +*/ +SQLITE_PRIVATE void sqlite3RenameTokenRemap(Parse *pParse, void *pTo, void *pFrom){ + RenameToken *p; + renameTokenCheckAll(pParse, pTo); + for(p=pParse->pRename; p; p=p->pNext){ + if( p->p==pFrom ){ + p->p = pTo; + break; + } + } +} + +/* +** Walker callback used by sqlite3RenameExprUnmap(). +*/ +static int renameUnmapExprCb(Walker *pWalker, Expr *pExpr){ + Parse *pParse = pWalker->pParse; + sqlite3RenameTokenRemap(pParse, 0, (void*)pExpr); + return WRC_Continue; +} + +/* +** Walker callback used by sqlite3RenameExprUnmap(). +*/ +static int renameUnmapSelectCb(Walker *pWalker, Select *p){ + Parse *pParse = pWalker->pParse; + int i; + if( ALWAYS(p->pEList) ){ + ExprList *pList = p->pEList; + for(i=0; inExpr; i++){ + if( pList->a[i].zName ){ + sqlite3RenameTokenRemap(pParse, 0, (void*)pList->a[i].zName); + } + } + } + if( ALWAYS(p->pSrc) ){ /* Every Select as a SrcList, even if it is empty */ + SrcList *pSrc = p->pSrc; + for(i=0; inSrc; i++){ + sqlite3RenameTokenRemap(pParse, 0, (void*)pSrc->a[i].zName); + } + } + return WRC_Continue; +} + +/* +** Remove all nodes that are part of expression pExpr from the rename list. +*/ +SQLITE_PRIVATE void sqlite3RenameExprUnmap(Parse *pParse, Expr *pExpr){ + Walker sWalker; + memset(&sWalker, 0, sizeof(Walker)); + sWalker.pParse = pParse; + sWalker.xExprCallback = renameUnmapExprCb; + sWalker.xSelectCallback = renameUnmapSelectCb; + sqlite3WalkExpr(&sWalker, pExpr); +} + +/* +** Remove all nodes that are part of expression-list pEList from the +** rename list. +*/ +SQLITE_PRIVATE void sqlite3RenameExprlistUnmap(Parse *pParse, ExprList *pEList){ + if( pEList ){ + int i; + Walker sWalker; + memset(&sWalker, 0, sizeof(Walker)); + sWalker.pParse = pParse; + sWalker.xExprCallback = renameUnmapExprCb; + sqlite3WalkExprList(&sWalker, pEList); + for(i=0; inExpr; i++){ + sqlite3RenameTokenRemap(pParse, 0, (void*)pEList->a[i].zName); + } + } +} + +/* +** Free the list of RenameToken objects given in the second argument +*/ +static void renameTokenFree(sqlite3 *db, RenameToken *pToken){ + RenameToken *pNext; + RenameToken *p; + for(p=pToken; p; p=pNext){ + pNext = p->pNext; + sqlite3DbFree(db, p); + } +} + +/* +** Search the Parse object passed as the first argument for a RenameToken +** object associated with parse tree element pPtr. If found, remove it +** from the Parse object and add it to the list maintained by the +** RenameCtx object passed as the second argument. +*/ +static void renameTokenFind(Parse *pParse, struct RenameCtx *pCtx, void *pPtr){ + RenameToken **pp; + assert( pPtr!=0 ); + for(pp=&pParse->pRename; (*pp); pp=&(*pp)->pNext){ + if( (*pp)->p==pPtr ){ + RenameToken *pToken = *pp; + *pp = pToken->pNext; + pToken->pNext = pCtx->pList; + pCtx->pList = pToken; + pCtx->nList++; + break; + } + } +} + +/* +** Iterate through the Select objects that are part of WITH clauses attached +** to select statement pSelect. +*/ +static void renameWalkWith(Walker *pWalker, Select *pSelect){ + if( pSelect->pWith ){ + int i; + for(i=0; ipWith->nCte; i++){ + Select *p = pSelect->pWith->a[i].pSelect; + NameContext sNC; + memset(&sNC, 0, sizeof(sNC)); + sNC.pParse = pWalker->pParse; + sqlite3SelectPrep(sNC.pParse, p, &sNC); + sqlite3WalkSelect(pWalker, p); + } + } +} + +/* +** This is a Walker select callback. It does nothing. It is only required +** because without a dummy callback, sqlite3WalkExpr() and similar do not +** descend into sub-select statements. +*/ +static int renameColumnSelectCb(Walker *pWalker, Select *p){ + renameWalkWith(pWalker, p); + return WRC_Continue; +} + +/* +** This is a Walker expression callback. +** +** For every TK_COLUMN node in the expression tree, search to see +** if the column being references is the column being renamed by an +** ALTER TABLE statement. If it is, then attach its associated +** RenameToken object to the list of RenameToken objects being +** constructed in RenameCtx object at pWalker->u.pRename. +*/ +static int renameColumnExprCb(Walker *pWalker, Expr *pExpr){ + RenameCtx *p = pWalker->u.pRename; + if( pExpr->op==TK_TRIGGER + && pExpr->iColumn==p->iCol + && pWalker->pParse->pTriggerTab==p->pTab + ){ + renameTokenFind(pWalker->pParse, p, (void*)pExpr); + }else if( pExpr->op==TK_COLUMN + && pExpr->iColumn==p->iCol + && p->pTab==pExpr->y.pTab + ){ + renameTokenFind(pWalker->pParse, p, (void*)pExpr); + } + return WRC_Continue; +} + +/* +** The RenameCtx contains a list of tokens that reference a column that +** is being renamed by an ALTER TABLE statement. Return the "last" +** RenameToken in the RenameCtx and remove that RenameToken from the +** RenameContext. "Last" means the last RenameToken encountered when +** the input SQL is parsed from left to right. Repeated calls to this routine +** return all column name tokens in the order that they are encountered +** in the SQL statement. +*/ +static RenameToken *renameColumnTokenNext(RenameCtx *pCtx){ + RenameToken *pBest = pCtx->pList; + RenameToken *pToken; + RenameToken **pp; + + for(pToken=pBest->pNext; pToken; pToken=pToken->pNext){ + if( pToken->t.z>pBest->t.z ) pBest = pToken; + } + for(pp=&pCtx->pList; *pp!=pBest; pp=&(*pp)->pNext); + *pp = pBest->pNext; + + return pBest; +} + +/* +** An error occured while parsing or otherwise processing a database +** object (either pParse->pNewTable, pNewIndex or pNewTrigger) as part of an +** ALTER TABLE RENAME COLUMN program. The error message emitted by the +** sub-routine is currently stored in pParse->zErrMsg. This function +** adds context to the error message and then stores it in pCtx. +*/ +static void renameColumnParseError( + sqlite3_context *pCtx, + int bPost, + sqlite3_value *pType, + sqlite3_value *pObject, + Parse *pParse +){ + const char *zT = (const char*)sqlite3_value_text(pType); + const char *zN = (const char*)sqlite3_value_text(pObject); + char *zErr; + + zErr = sqlite3_mprintf("error in %s %s%s: %s", + zT, zN, (bPost ? " after rename" : ""), + pParse->zErrMsg + ); + sqlite3_result_error(pCtx, zErr, -1); + sqlite3_free(zErr); +} + +/* +** For each name in the the expression-list pEList (i.e. each +** pEList->a[i].zName) that matches the string in zOld, extract the +** corresponding rename-token from Parse object pParse and add it +** to the RenameCtx pCtx. +*/ +static void renameColumnElistNames( + Parse *pParse, + RenameCtx *pCtx, + ExprList *pEList, + const char *zOld +){ + if( pEList ){ + int i; + for(i=0; inExpr; i++){ + char *zName = pEList->a[i].zName; + if( 0==sqlite3_stricmp(zName, zOld) ){ + renameTokenFind(pParse, pCtx, (void*)zName); + } + } + } +} + +/* +** For each name in the the id-list pIdList (i.e. each pIdList->a[i].zName) +** that matches the string in zOld, extract the corresponding rename-token +** from Parse object pParse and add it to the RenameCtx pCtx. +*/ +static void renameColumnIdlistNames( + Parse *pParse, + RenameCtx *pCtx, + IdList *pIdList, + const char *zOld +){ + if( pIdList ){ + int i; + for(i=0; inId; i++){ + char *zName = pIdList->a[i].zName; + if( 0==sqlite3_stricmp(zName, zOld) ){ + renameTokenFind(pParse, pCtx, (void*)zName); + } + } + } +} + +/* +** Parse the SQL statement zSql using Parse object (*p). The Parse object +** is initialized by this function before it is used. +*/ +static int renameParseSql( + Parse *p, /* Memory to use for Parse object */ + const char *zDb, /* Name of schema SQL belongs to */ + int bTable, /* 1 -> RENAME TABLE, 0 -> RENAME COLUMN */ + sqlite3 *db, /* Database handle */ + const char *zSql, /* SQL to parse */ + int bTemp /* True if SQL is from temp schema */ +){ + int rc; + char *zErr = 0; + + db->init.iDb = bTemp ? 1 : sqlite3FindDbName(db, zDb); + + /* Parse the SQL statement passed as the first argument. If no error + ** occurs and the parse does not result in a new table, index or + ** trigger object, the database must be corrupt. */ + memset(p, 0, sizeof(Parse)); + p->eParseMode = (bTable ? PARSE_MODE_RENAME_TABLE : PARSE_MODE_RENAME_COLUMN); + p->db = db; + p->nQueryLoop = 1; + rc = sqlite3RunParser(p, zSql, &zErr); + assert( p->zErrMsg==0 ); + assert( rc!=SQLITE_OK || zErr==0 ); + p->zErrMsg = zErr; + if( db->mallocFailed ) rc = SQLITE_NOMEM; + if( rc==SQLITE_OK + && p->pNewTable==0 && p->pNewIndex==0 && p->pNewTrigger==0 + ){ + rc = SQLITE_CORRUPT_BKPT; + } + +#ifdef SQLITE_DEBUG + /* Ensure that all mappings in the Parse.pRename list really do map to + ** a part of the input string. */ + if( rc==SQLITE_OK ){ + int nSql = sqlite3Strlen30(zSql); + RenameToken *pToken; + for(pToken=p->pRename; pToken; pToken=pToken->pNext){ + assert( pToken->t.z>=zSql && &pToken->t.z[pToken->t.n]<=&zSql[nSql] ); + } + } +#endif + + db->init.iDb = 0; + return rc; +} + +/* +** This function edits SQL statement zSql, replacing each token identified +** by the linked list pRename with the text of zNew. If argument bQuote is +** true, then zNew is always quoted first. If no error occurs, the result +** is loaded into context object pCtx as the result. +** +** Or, if an error occurs (i.e. an OOM condition), an error is left in +** pCtx and an SQLite error code returned. +*/ +static int renameEditSql( + sqlite3_context *pCtx, /* Return result here */ + RenameCtx *pRename, /* Rename context */ + const char *zSql, /* SQL statement to edit */ + const char *zNew, /* New token text */ + int bQuote /* True to always quote token */ +){ + int nNew = sqlite3Strlen30(zNew); + int nSql = sqlite3Strlen30(zSql); + sqlite3 *db = sqlite3_context_db_handle(pCtx); + int rc = SQLITE_OK; + char *zQuot; + char *zOut; + int nQuot; + + /* Set zQuot to point to a buffer containing a quoted copy of the + ** identifier zNew. If the corresponding identifier in the original + ** ALTER TABLE statement was quoted (bQuote==1), then set zNew to + ** point to zQuot so that all substitutions are made using the + ** quoted version of the new column name. */ + zQuot = sqlite3MPrintf(db, "\"%w\"", zNew); + if( zQuot==0 ){ + return SQLITE_NOMEM; + }else{ + nQuot = sqlite3Strlen30(zQuot); + } + if( bQuote ){ + zNew = zQuot; + nNew = nQuot; + } + + /* At this point pRename->pList contains a list of RenameToken objects + ** corresponding to all tokens in the input SQL that must be replaced + ** with the new column name. All that remains is to construct and + ** return the edited SQL string. */ + assert( nQuot>=nNew ); + zOut = sqlite3DbMallocZero(db, nSql + pRename->nList*nQuot + 1); + if( zOut ){ + int nOut = nSql; + memcpy(zOut, zSql, nSql); + while( pRename->pList ){ + int iOff; /* Offset of token to replace in zOut */ + RenameToken *pBest = renameColumnTokenNext(pRename); + + u32 nReplace; + const char *zReplace; + if( sqlite3IsIdChar(*pBest->t.z) ){ + nReplace = nNew; + zReplace = zNew; + }else{ + nReplace = nQuot; + zReplace = zQuot; + } + + iOff = pBest->t.z - zSql; + if( pBest->t.n!=nReplace ){ + memmove(&zOut[iOff + nReplace], &zOut[iOff + pBest->t.n], + nOut - (iOff + pBest->t.n) + ); + nOut += nReplace - pBest->t.n; + zOut[nOut] = '\0'; + } + memcpy(&zOut[iOff], zReplace, nReplace); + sqlite3DbFree(db, pBest); + } + + sqlite3_result_text(pCtx, zOut, -1, SQLITE_TRANSIENT); + sqlite3DbFree(db, zOut); + }else{ + rc = SQLITE_NOMEM; + } + + sqlite3_free(zQuot); + return rc; +} + +/* +** Resolve all symbols in the trigger at pParse->pNewTrigger, assuming +** it was read from the schema of database zDb. Return SQLITE_OK if +** successful. Otherwise, return an SQLite error code and leave an error +** message in the Parse object. +*/ +static int renameResolveTrigger(Parse *pParse, const char *zDb){ + sqlite3 *db = pParse->db; + Trigger *pNew = pParse->pNewTrigger; + TriggerStep *pStep; + NameContext sNC; + int rc = SQLITE_OK; + + memset(&sNC, 0, sizeof(sNC)); + sNC.pParse = pParse; + assert( pNew->pTabSchema ); + pParse->pTriggerTab = sqlite3FindTable(db, pNew->table, + db->aDb[sqlite3SchemaToIndex(db, pNew->pTabSchema)].zDbSName + ); + pParse->eTriggerOp = pNew->op; + /* ALWAYS() because if the table of the trigger does not exist, the + ** error would have been hit before this point */ + if( ALWAYS(pParse->pTriggerTab) ){ + rc = sqlite3ViewGetColumnNames(pParse, pParse->pTriggerTab); + } + + /* Resolve symbols in WHEN clause */ + if( rc==SQLITE_OK && pNew->pWhen ){ + rc = sqlite3ResolveExprNames(&sNC, pNew->pWhen); + } + + for(pStep=pNew->step_list; rc==SQLITE_OK && pStep; pStep=pStep->pNext){ + if( pStep->pSelect ){ + sqlite3SelectPrep(pParse, pStep->pSelect, &sNC); + if( pParse->nErr ) rc = pParse->rc; + } + if( rc==SQLITE_OK && pStep->zTarget ){ + Table *pTarget = sqlite3LocateTable(pParse, 0, pStep->zTarget, zDb); + if( pTarget==0 ){ + rc = SQLITE_ERROR; + }else if( SQLITE_OK==(rc = sqlite3ViewGetColumnNames(pParse, pTarget)) ){ + SrcList sSrc; + memset(&sSrc, 0, sizeof(sSrc)); + sSrc.nSrc = 1; + sSrc.a[0].zName = pStep->zTarget; + sSrc.a[0].pTab = pTarget; + sNC.pSrcList = &sSrc; + if( pStep->pWhere ){ + rc = sqlite3ResolveExprNames(&sNC, pStep->pWhere); + } + if( rc==SQLITE_OK ){ + rc = sqlite3ResolveExprListNames(&sNC, pStep->pExprList); + } + assert( !pStep->pUpsert || (!pStep->pWhere && !pStep->pExprList) ); + if( pStep->pUpsert ){ + Upsert *pUpsert = pStep->pUpsert; + assert( rc==SQLITE_OK ); + pUpsert->pUpsertSrc = &sSrc; + sNC.uNC.pUpsert = pUpsert; + sNC.ncFlags = NC_UUpsert; + rc = sqlite3ResolveExprListNames(&sNC, pUpsert->pUpsertTarget); + if( rc==SQLITE_OK ){ + ExprList *pUpsertSet = pUpsert->pUpsertSet; + rc = sqlite3ResolveExprListNames(&sNC, pUpsertSet); + } + if( rc==SQLITE_OK ){ + rc = sqlite3ResolveExprNames(&sNC, pUpsert->pUpsertWhere); + } + if( rc==SQLITE_OK ){ + rc = sqlite3ResolveExprNames(&sNC, pUpsert->pUpsertTargetWhere); + } + sNC.ncFlags = 0; + } + sNC.pSrcList = 0; + } + } + } + return rc; +} + +/* +** Invoke sqlite3WalkExpr() or sqlite3WalkSelect() on all Select or Expr +** objects that are part of the trigger passed as the second argument. +*/ +static void renameWalkTrigger(Walker *pWalker, Trigger *pTrigger){ + TriggerStep *pStep; + + /* Find tokens to edit in WHEN clause */ + sqlite3WalkExpr(pWalker, pTrigger->pWhen); + + /* Find tokens to edit in trigger steps */ + for(pStep=pTrigger->step_list; pStep; pStep=pStep->pNext){ + sqlite3WalkSelect(pWalker, pStep->pSelect); + sqlite3WalkExpr(pWalker, pStep->pWhere); + sqlite3WalkExprList(pWalker, pStep->pExprList); + if( pStep->pUpsert ){ + Upsert *pUpsert = pStep->pUpsert; + sqlite3WalkExprList(pWalker, pUpsert->pUpsertTarget); + sqlite3WalkExprList(pWalker, pUpsert->pUpsertSet); + sqlite3WalkExpr(pWalker, pUpsert->pUpsertWhere); + sqlite3WalkExpr(pWalker, pUpsert->pUpsertTargetWhere); + } + } +} + +/* +** Free the contents of Parse object (*pParse). Do not free the memory +** occupied by the Parse object itself. +*/ +static void renameParseCleanup(Parse *pParse){ + sqlite3 *db = pParse->db; + Index *pIdx; + if( pParse->pVdbe ){ + sqlite3VdbeFinalize(pParse->pVdbe); + } + sqlite3DeleteTable(db, pParse->pNewTable); + while( (pIdx = pParse->pNewIndex)!=0 ){ + pParse->pNewIndex = pIdx->pNext; + sqlite3FreeIndex(db, pIdx); + } + sqlite3DeleteTrigger(db, pParse->pNewTrigger); + sqlite3DbFree(db, pParse->zErrMsg); + renameTokenFree(db, pParse->pRename); + sqlite3ParserReset(pParse); +} + +/* +** SQL function: +** +** sqlite_rename_column(zSql, iCol, bQuote, zNew, zTable, zOld) +** +** 0. zSql: SQL statement to rewrite +** 1. type: Type of object ("table", "view" etc.) +** 2. object: Name of object +** 3. Database: Database name (e.g. "main") +** 4. Table: Table name +** 5. iCol: Index of column to rename +** 6. zNew: New column name +** 7. bQuote: Non-zero if the new column name should be quoted. +** 8. bTemp: True if zSql comes from temp schema +** +** Do a column rename operation on the CREATE statement given in zSql. +** The iCol-th column (left-most is 0) of table zTable is renamed from zCol +** into zNew. The name should be quoted if bQuote is true. +** +** This function is used internally by the ALTER TABLE RENAME COLUMN command. +** It is only accessible to SQL created using sqlite3NestedParse(). It is +** not reachable from ordinary SQL passed into sqlite3_prepare(). +*/ +static void renameColumnFunc( + sqlite3_context *context, + int NotUsed, + sqlite3_value **argv +){ + sqlite3 *db = sqlite3_context_db_handle(context); + RenameCtx sCtx; + const char *zSql = (const char*)sqlite3_value_text(argv[0]); + const char *zDb = (const char*)sqlite3_value_text(argv[3]); + const char *zTable = (const char*)sqlite3_value_text(argv[4]); + int iCol = sqlite3_value_int(argv[5]); + const char *zNew = (const char*)sqlite3_value_text(argv[6]); + int bQuote = sqlite3_value_int(argv[7]); + int bTemp = sqlite3_value_int(argv[8]); + const char *zOld; + int rc; + Parse sParse; + Walker sWalker; + Index *pIdx; + int i; + Table *pTab; +#ifndef SQLITE_OMIT_AUTHORIZATION + sqlite3_xauth xAuth = db->xAuth; +#endif + + UNUSED_PARAMETER(NotUsed); + if( zSql==0 ) return; + if( zTable==0 ) return; + if( zNew==0 ) return; + if( iCol<0 ) return; + sqlite3BtreeEnterAll(db); + pTab = sqlite3FindTable(db, zTable, zDb); + if( pTab==0 || iCol>=pTab->nCol ){ + sqlite3BtreeLeaveAll(db); + return; + } + zOld = pTab->aCol[iCol].zName; + memset(&sCtx, 0, sizeof(sCtx)); + sCtx.iCol = ((iCol==pTab->iPKey) ? -1 : iCol); + +#ifndef SQLITE_OMIT_AUTHORIZATION + db->xAuth = 0; +#endif + rc = renameParseSql(&sParse, zDb, 0, db, zSql, bTemp); + + /* Find tokens that need to be replaced. */ + memset(&sWalker, 0, sizeof(Walker)); + sWalker.pParse = &sParse; + sWalker.xExprCallback = renameColumnExprCb; + sWalker.xSelectCallback = renameColumnSelectCb; + sWalker.u.pRename = &sCtx; + + sCtx.pTab = pTab; + if( rc!=SQLITE_OK ) goto renameColumnFunc_done; + if( sParse.pNewTable ){ + Select *pSelect = sParse.pNewTable->pSelect; + if( pSelect ){ + sParse.rc = SQLITE_OK; + sqlite3SelectPrep(&sParse, sParse.pNewTable->pSelect, 0); + rc = (db->mallocFailed ? SQLITE_NOMEM : sParse.rc); + if( rc==SQLITE_OK ){ + sqlite3WalkSelect(&sWalker, pSelect); + } + if( rc!=SQLITE_OK ) goto renameColumnFunc_done; + }else{ + /* A regular table */ + int bFKOnly = sqlite3_stricmp(zTable, sParse.pNewTable->zName); + FKey *pFKey; + assert( sParse.pNewTable->pSelect==0 ); + sCtx.pTab = sParse.pNewTable; + if( bFKOnly==0 ){ + renameTokenFind( + &sParse, &sCtx, (void*)sParse.pNewTable->aCol[iCol].zName + ); + if( sCtx.iCol<0 ){ + renameTokenFind(&sParse, &sCtx, (void*)&sParse.pNewTable->iPKey); + } + sqlite3WalkExprList(&sWalker, sParse.pNewTable->pCheck); + for(pIdx=sParse.pNewTable->pIndex; pIdx; pIdx=pIdx->pNext){ + sqlite3WalkExprList(&sWalker, pIdx->aColExpr); + } + for(pIdx=sParse.pNewIndex; pIdx; pIdx=pIdx->pNext){ + sqlite3WalkExprList(&sWalker, pIdx->aColExpr); + } + } + + for(pFKey=sParse.pNewTable->pFKey; pFKey; pFKey=pFKey->pNextFrom){ + for(i=0; inCol; i++){ + if( bFKOnly==0 && pFKey->aCol[i].iFrom==iCol ){ + renameTokenFind(&sParse, &sCtx, (void*)&pFKey->aCol[i]); + } + if( 0==sqlite3_stricmp(pFKey->zTo, zTable) + && 0==sqlite3_stricmp(pFKey->aCol[i].zCol, zOld) + ){ + renameTokenFind(&sParse, &sCtx, (void*)pFKey->aCol[i].zCol); + } + } + } + } + }else if( sParse.pNewIndex ){ + sqlite3WalkExprList(&sWalker, sParse.pNewIndex->aColExpr); + sqlite3WalkExpr(&sWalker, sParse.pNewIndex->pPartIdxWhere); + }else{ + /* A trigger */ + TriggerStep *pStep; + rc = renameResolveTrigger(&sParse, (bTemp ? 0 : zDb)); + if( rc!=SQLITE_OK ) goto renameColumnFunc_done; + + for(pStep=sParse.pNewTrigger->step_list; pStep; pStep=pStep->pNext){ + if( pStep->zTarget ){ + Table *pTarget = sqlite3LocateTable(&sParse, 0, pStep->zTarget, zDb); + if( pTarget==pTab ){ + if( pStep->pUpsert ){ + ExprList *pUpsertSet = pStep->pUpsert->pUpsertSet; + renameColumnElistNames(&sParse, &sCtx, pUpsertSet, zOld); + } + renameColumnIdlistNames(&sParse, &sCtx, pStep->pIdList, zOld); + renameColumnElistNames(&sParse, &sCtx, pStep->pExprList, zOld); + } + } + } + + + /* Find tokens to edit in UPDATE OF clause */ + if( sParse.pTriggerTab==pTab ){ + renameColumnIdlistNames(&sParse, &sCtx,sParse.pNewTrigger->pColumns,zOld); + } + + /* Find tokens to edit in various expressions and selects */ + renameWalkTrigger(&sWalker, sParse.pNewTrigger); + } + + assert( rc==SQLITE_OK ); + rc = renameEditSql(context, &sCtx, zSql, zNew, bQuote); + +renameColumnFunc_done: + if( rc!=SQLITE_OK ){ + if( sParse.zErrMsg ){ + renameColumnParseError(context, 0, argv[1], argv[2], &sParse); + }else{ + sqlite3_result_error_code(context, rc); + } + } + + renameParseCleanup(&sParse); + renameTokenFree(db, sCtx.pList); +#ifndef SQLITE_OMIT_AUTHORIZATION + db->xAuth = xAuth; +#endif + sqlite3BtreeLeaveAll(db); +} + +/* +** Walker expression callback used by "RENAME TABLE". +*/ +static int renameTableExprCb(Walker *pWalker, Expr *pExpr){ + RenameCtx *p = pWalker->u.pRename; + if( pExpr->op==TK_COLUMN && p->pTab==pExpr->y.pTab ){ + renameTokenFind(pWalker->pParse, p, (void*)&pExpr->y.pTab); + } + return WRC_Continue; +} + +/* +** Walker select callback used by "RENAME TABLE". +*/ +static int renameTableSelectCb(Walker *pWalker, Select *pSelect){ + int i; + RenameCtx *p = pWalker->u.pRename; + SrcList *pSrc = pSelect->pSrc; + if( pSrc==0 ){ + assert( pWalker->pParse->db->mallocFailed ); + return WRC_Abort; + } + for(i=0; inSrc; i++){ + struct SrcList_item *pItem = &pSrc->a[i]; + if( pItem->pTab==p->pTab ){ + renameTokenFind(pWalker->pParse, p, pItem->zName); + } + } + renameWalkWith(pWalker, pSelect); + + return WRC_Continue; +} + + +/* +** This C function implements an SQL user function that is used by SQL code +** generated by the ALTER TABLE ... RENAME command to modify the definition +** of any foreign key constraints that use the table being renamed as the +** parent table. It is passed three arguments: +** +** 0: The database containing the table being renamed. +** 1. type: Type of object ("table", "view" etc.) +** 2. object: Name of object +** 3: The complete text of the schema statement being modified, +** 4: The old name of the table being renamed, and +** 5: The new name of the table being renamed. +** 6: True if the schema statement comes from the temp db. +** +** It returns the new schema statement. For example: +** +** sqlite_rename_table('main', 'CREATE TABLE t1(a REFERENCES t2)','t2','t3',0) +** -> 'CREATE TABLE t1(a REFERENCES t3)' +*/ +static void renameTableFunc( + sqlite3_context *context, + int NotUsed, + sqlite3_value **argv +){ + sqlite3 *db = sqlite3_context_db_handle(context); + const char *zDb = (const char*)sqlite3_value_text(argv[0]); + const char *zInput = (const char*)sqlite3_value_text(argv[3]); + const char *zOld = (const char*)sqlite3_value_text(argv[4]); + const char *zNew = (const char*)sqlite3_value_text(argv[5]); + int bTemp = sqlite3_value_int(argv[6]); + UNUSED_PARAMETER(NotUsed); + + if( zInput && zOld && zNew ){ + Parse sParse; + int rc; + int bQuote = 1; + RenameCtx sCtx; + Walker sWalker; + +#ifndef SQLITE_OMIT_AUTHORIZATION + sqlite3_xauth xAuth = db->xAuth; + db->xAuth = 0; +#endif + + sqlite3BtreeEnterAll(db); + + memset(&sCtx, 0, sizeof(RenameCtx)); + sCtx.pTab = sqlite3FindTable(db, zOld, zDb); + memset(&sWalker, 0, sizeof(Walker)); + sWalker.pParse = &sParse; + sWalker.xExprCallback = renameTableExprCb; + sWalker.xSelectCallback = renameTableSelectCb; + sWalker.u.pRename = &sCtx; + + rc = renameParseSql(&sParse, zDb, 1, db, zInput, bTemp); + + if( rc==SQLITE_OK ){ + int isLegacy = (db->flags & SQLITE_LegacyAlter); + if( sParse.pNewTable ){ + Table *pTab = sParse.pNewTable; + + if( pTab->pSelect ){ + if( isLegacy==0 ){ + NameContext sNC; + memset(&sNC, 0, sizeof(sNC)); + sNC.pParse = &sParse; + + sqlite3SelectPrep(&sParse, pTab->pSelect, &sNC); + if( sParse.nErr ) rc = sParse.rc; + sqlite3WalkSelect(&sWalker, pTab->pSelect); + } + }else{ + /* Modify any FK definitions to point to the new table. */ +#ifndef SQLITE_OMIT_FOREIGN_KEY + if( isLegacy==0 || (db->flags & SQLITE_ForeignKeys) ){ + FKey *pFKey; + for(pFKey=pTab->pFKey; pFKey; pFKey=pFKey->pNextFrom){ + if( sqlite3_stricmp(pFKey->zTo, zOld)==0 ){ + renameTokenFind(&sParse, &sCtx, (void*)pFKey->zTo); + } + } + } +#endif + + /* If this is the table being altered, fix any table refs in CHECK + ** expressions. Also update the name that appears right after the + ** "CREATE [VIRTUAL] TABLE" bit. */ + if( sqlite3_stricmp(zOld, pTab->zName)==0 ){ + sCtx.pTab = pTab; + if( isLegacy==0 ){ + sqlite3WalkExprList(&sWalker, pTab->pCheck); + } + renameTokenFind(&sParse, &sCtx, pTab->zName); + } + } + } + + else if( sParse.pNewIndex ){ + renameTokenFind(&sParse, &sCtx, sParse.pNewIndex->zName); + if( isLegacy==0 ){ + sqlite3WalkExpr(&sWalker, sParse.pNewIndex->pPartIdxWhere); + } + } + +#ifndef SQLITE_OMIT_TRIGGER + else{ + Trigger *pTrigger = sParse.pNewTrigger; + TriggerStep *pStep; + if( 0==sqlite3_stricmp(sParse.pNewTrigger->table, zOld) + && sCtx.pTab->pSchema==pTrigger->pTabSchema + ){ + renameTokenFind(&sParse, &sCtx, sParse.pNewTrigger->table); + } + + if( isLegacy==0 ){ + rc = renameResolveTrigger(&sParse, bTemp ? 0 : zDb); + if( rc==SQLITE_OK ){ + renameWalkTrigger(&sWalker, pTrigger); + for(pStep=pTrigger->step_list; pStep; pStep=pStep->pNext){ + if( pStep->zTarget && 0==sqlite3_stricmp(pStep->zTarget, zOld) ){ + renameTokenFind(&sParse, &sCtx, pStep->zTarget); + } + } + } + } + } +#endif + } + + if( rc==SQLITE_OK ){ + rc = renameEditSql(context, &sCtx, zInput, zNew, bQuote); + } + if( rc!=SQLITE_OK ){ + if( sParse.zErrMsg ){ + renameColumnParseError(context, 0, argv[1], argv[2], &sParse); + }else{ + sqlite3_result_error_code(context, rc); + } + } + + renameParseCleanup(&sParse); + renameTokenFree(db, sCtx.pList); + sqlite3BtreeLeaveAll(db); +#ifndef SQLITE_OMIT_AUTHORIZATION + db->xAuth = xAuth; +#endif + } + + return; +} + +/* +** An SQL user function that checks that there are no parse or symbol +** resolution problems in a CREATE TRIGGER|TABLE|VIEW|INDEX statement. +** After an ALTER TABLE .. RENAME operation is performed and the schema +** reloaded, this function is called on each SQL statement in the schema +** to ensure that it is still usable. +** +** 0: Database name ("main", "temp" etc.). +** 1: SQL statement. +** 2: Object type ("view", "table", "trigger" or "index"). +** 3: Object name. +** 4: True if object is from temp schema. +** +** Unless it finds an error, this function normally returns NULL. However, it +** returns integer value 1 if: +** +** * the SQL argument creates a trigger, and +** * the table that the trigger is attached to is in database zDb. +*/ +static void renameTableTest( + sqlite3_context *context, + int NotUsed, + sqlite3_value **argv +){ + sqlite3 *db = sqlite3_context_db_handle(context); + char const *zDb = (const char*)sqlite3_value_text(argv[0]); + char const *zInput = (const char*)sqlite3_value_text(argv[1]); + int bTemp = sqlite3_value_int(argv[4]); + int isLegacy = (db->flags & SQLITE_LegacyAlter); + +#ifndef SQLITE_OMIT_AUTHORIZATION + sqlite3_xauth xAuth = db->xAuth; + db->xAuth = 0; +#endif + + UNUSED_PARAMETER(NotUsed); + if( zDb && zInput ){ + int rc; + Parse sParse; + rc = renameParseSql(&sParse, zDb, 1, db, zInput, bTemp); + if( rc==SQLITE_OK ){ + if( isLegacy==0 && sParse.pNewTable && sParse.pNewTable->pSelect ){ + NameContext sNC; + memset(&sNC, 0, sizeof(sNC)); + sNC.pParse = &sParse; + sqlite3SelectPrep(&sParse, sParse.pNewTable->pSelect, &sNC); + if( sParse.nErr ) rc = sParse.rc; + } + + else if( sParse.pNewTrigger ){ + if( isLegacy==0 ){ + rc = renameResolveTrigger(&sParse, bTemp ? 0 : zDb); + } + if( rc==SQLITE_OK ){ + int i1 = sqlite3SchemaToIndex(db, sParse.pNewTrigger->pTabSchema); + int i2 = sqlite3FindDbName(db, zDb); + if( i1==i2 ) sqlite3_result_int(context, 1); + } + } + } + + if( rc!=SQLITE_OK ){ + renameColumnParseError(context, 1, argv[2], argv[3], &sParse); + } + renameParseCleanup(&sParse); + } + +#ifndef SQLITE_OMIT_AUTHORIZATION + db->xAuth = xAuth; +#endif +} + +/* +** Register built-in functions used to help implement ALTER TABLE +*/ +SQLITE_PRIVATE void sqlite3AlterFunctions(void){ + static FuncDef aAlterTableFuncs[] = { + INTERNAL_FUNCTION(sqlite_rename_column, 9, renameColumnFunc), + INTERNAL_FUNCTION(sqlite_rename_table, 7, renameTableFunc), + INTERNAL_FUNCTION(sqlite_rename_test, 5, renameTableTest), + }; + sqlite3InsertBuiltinFuncs(aAlterTableFuncs, ArraySize(aAlterTableFuncs)); +} +#endif /* SQLITE_ALTER_TABLE */ + +/************** End of alter.c ***********************************************/ +/************** Begin file analyze.c *****************************************/ +/* +** 2005-07-08 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** This file contains code associated with the ANALYZE command. +** +** The ANALYZE command gather statistics about the content of tables +** and indices. These statistics are made available to the query planner +** to help it make better decisions about how to perform queries. +** +** The following system tables are or have been supported: +** +** CREATE TABLE sqlite_stat1(tbl, idx, stat); +** CREATE TABLE sqlite_stat2(tbl, idx, sampleno, sample); +** CREATE TABLE sqlite_stat3(tbl, idx, nEq, nLt, nDLt, sample); +** CREATE TABLE sqlite_stat4(tbl, idx, nEq, nLt, nDLt, sample); +** +** Additional tables might be added in future releases of SQLite. +** The sqlite_stat2 table is not created or used unless the SQLite version +** is between 3.6.18 and 3.7.8, inclusive, and unless SQLite is compiled +** with SQLITE_ENABLE_STAT2. The sqlite_stat2 table is deprecated. +** The sqlite_stat2 table is superseded by sqlite_stat3, which is only +** created and used by SQLite versions 3.7.9 through 3.29.0 when +** SQLITE_ENABLE_STAT3 defined. The functionality of sqlite_stat3 +** is a superset of sqlite_stat2 and is also now deprecated. The +** sqlite_stat4 is an enhanced version of sqlite_stat3 and is only +** available when compiled with SQLITE_ENABLE_STAT4 and in SQLite +** versions 3.8.1 and later. STAT4 is the only variant that is still +** supported. +** +** For most applications, sqlite_stat1 provides all the statistics required +** for the query planner to make good choices. +** +** Format of sqlite_stat1: +** +** There is normally one row per index, with the index identified by the +** name in the idx column. The tbl column is the name of the table to +** which the index belongs. In each such row, the stat column will be +** a string consisting of a list of integers. The first integer in this +** list is the number of rows in the index. (This is the same as the +** number of rows in the table, except for partial indices.) The second +** integer is the average number of rows in the index that have the same +** value in the first column of the index. The third integer is the average +** number of rows in the index that have the same value for the first two +** columns. The N-th integer (for N>1) is the average number of rows in +** the index which have the same value for the first N-1 columns. For +** a K-column index, there will be K+1 integers in the stat column. If +** the index is unique, then the last integer will be 1. +** +** The list of integers in the stat column can optionally be followed +** by the keyword "unordered". The "unordered" keyword, if it is present, +** must be separated from the last integer by a single space. If the +** "unordered" keyword is present, then the query planner assumes that +** the index is unordered and will not use the index for a range query. +** +** If the sqlite_stat1.idx column is NULL, then the sqlite_stat1.stat +** column contains a single integer which is the (estimated) number of +** rows in the table identified by sqlite_stat1.tbl. +** +** Format of sqlite_stat2: +** +** The sqlite_stat2 is only created and is only used if SQLite is compiled +** with SQLITE_ENABLE_STAT2 and if the SQLite version number is between +** 3.6.18 and 3.7.8. The "stat2" table contains additional information +** about the distribution of keys within an index. The index is identified by +** the "idx" column and the "tbl" column is the name of the table to which +** the index belongs. There are usually 10 rows in the sqlite_stat2 +** table for each index. +** +** The sqlite_stat2 entries for an index that have sampleno between 0 and 9 +** inclusive are samples of the left-most key value in the index taken at +** evenly spaced points along the index. Let the number of samples be S +** (10 in the standard build) and let C be the number of rows in the index. +** Then the sampled rows are given by: +** +** rownumber = (i*C*2 + C)/(S*2) +** +** For i between 0 and S-1. Conceptually, the index space is divided into +** S uniform buckets and the samples are the middle row from each bucket. +** +** The format for sqlite_stat2 is recorded here for legacy reference. This +** version of SQLite does not support sqlite_stat2. It neither reads nor +** writes the sqlite_stat2 table. This version of SQLite only supports +** sqlite_stat3. +** +** Format for sqlite_stat3: +** +** The sqlite_stat3 format is a subset of sqlite_stat4. Hence, the +** sqlite_stat4 format will be described first. Further information +** about sqlite_stat3 follows the sqlite_stat4 description. +** +** Format for sqlite_stat4: +** +** As with sqlite_stat2, the sqlite_stat4 table contains histogram data +** to aid the query planner in choosing good indices based on the values +** that indexed columns are compared against in the WHERE clauses of +** queries. +** +** The sqlite_stat4 table contains multiple entries for each index. +** The idx column names the index and the tbl column is the table of the +** index. If the idx and tbl columns are the same, then the sample is +** of the INTEGER PRIMARY KEY. The sample column is a blob which is the +** binary encoding of a key from the index. The nEq column is a +** list of integers. The first integer is the approximate number +** of entries in the index whose left-most column exactly matches +** the left-most column of the sample. The second integer in nEq +** is the approximate number of entries in the index where the +** first two columns match the first two columns of the sample. +** And so forth. nLt is another list of integers that show the approximate +** number of entries that are strictly less than the sample. The first +** integer in nLt contains the number of entries in the index where the +** left-most column is less than the left-most column of the sample. +** The K-th integer in the nLt entry is the number of index entries +** where the first K columns are less than the first K columns of the +** sample. The nDLt column is like nLt except that it contains the +** number of distinct entries in the index that are less than the +** sample. +** +** There can be an arbitrary number of sqlite_stat4 entries per index. +** The ANALYZE command will typically generate sqlite_stat4 tables +** that contain between 10 and 40 samples which are distributed across +** the key space, though not uniformly, and which include samples with +** large nEq values. +** +** Format for sqlite_stat3 redux: +** +** The sqlite_stat3 table is like sqlite_stat4 except that it only +** looks at the left-most column of the index. The sqlite_stat3.sample +** column contains the actual value of the left-most column instead +** of a blob encoding of the complete index key as is found in +** sqlite_stat4.sample. The nEq, nLt, and nDLt entries of sqlite_stat3 +** all contain just a single integer which is the same as the first +** integer in the equivalent columns in sqlite_stat4. +*/ +#ifndef SQLITE_OMIT_ANALYZE +/* #include "sqliteInt.h" */ + +#if defined(SQLITE_ENABLE_STAT4) +# define IsStat4 1 +#else +# define IsStat4 0 +# undef SQLITE_STAT4_SAMPLES +# define SQLITE_STAT4_SAMPLES 1 +#endif + +/* +** This routine generates code that opens the sqlite_statN tables. +** The sqlite_stat1 table is always relevant. sqlite_stat2 is now +** obsolete. sqlite_stat3 and sqlite_stat4 are only opened when +** appropriate compile-time options are provided. +** +** If the sqlite_statN tables do not previously exist, it is created. +** +** Argument zWhere may be a pointer to a buffer containing a table name, +** or it may be a NULL pointer. If it is not NULL, then all entries in +** the sqlite_statN tables associated with the named table are deleted. +** If zWhere==0, then code is generated to delete all stat table entries. +*/ +static void openStatTable( + Parse *pParse, /* Parsing context */ + int iDb, /* The database we are looking in */ + int iStatCur, /* Open the sqlite_stat1 table on this cursor */ + const char *zWhere, /* Delete entries for this table or index */ + const char *zWhereType /* Either "tbl" or "idx" */ +){ + static const struct { + const char *zName; + const char *zCols; + } aTable[] = { + { "sqlite_stat1", "tbl,idx,stat" }, +#if defined(SQLITE_ENABLE_STAT4) + { "sqlite_stat4", "tbl,idx,neq,nlt,ndlt,sample" }, +#else + { "sqlite_stat4", 0 }, +#endif + { "sqlite_stat3", 0 }, + }; + int i; + sqlite3 *db = pParse->db; + Db *pDb; + Vdbe *v = sqlite3GetVdbe(pParse); + int aRoot[ArraySize(aTable)]; + u8 aCreateTbl[ArraySize(aTable)]; + + if( v==0 ) return; + assert( sqlite3BtreeHoldsAllMutexes(db) ); + assert( sqlite3VdbeDb(v)==db ); + pDb = &db->aDb[iDb]; + + /* Create new statistic tables if they do not exist, or clear them + ** if they do already exist. + */ + for(i=0; izDbSName))==0 ){ + if( aTable[i].zCols ){ + /* The sqlite_statN table does not exist. Create it. Note that a + ** side-effect of the CREATE TABLE statement is to leave the rootpage + ** of the new table in register pParse->regRoot. This is important + ** because the OpenWrite opcode below will be needing it. */ + sqlite3NestedParse(pParse, + "CREATE TABLE %Q.%s(%s)", pDb->zDbSName, zTab, aTable[i].zCols + ); + aRoot[i] = pParse->regRoot; + aCreateTbl[i] = OPFLAG_P2ISREG; + } + }else{ + /* The table already exists. If zWhere is not NULL, delete all entries + ** associated with the table zWhere. If zWhere is NULL, delete the + ** entire contents of the table. */ + aRoot[i] = pStat->tnum; + aCreateTbl[i] = 0; + sqlite3TableLock(pParse, iDb, aRoot[i], 1, zTab); + if( zWhere ){ + sqlite3NestedParse(pParse, + "DELETE FROM %Q.%s WHERE %s=%Q", + pDb->zDbSName, zTab, zWhereType, zWhere + ); +#ifdef SQLITE_ENABLE_PREUPDATE_HOOK + }else if( db->xPreUpdateCallback ){ + sqlite3NestedParse(pParse, "DELETE FROM %Q.%s", pDb->zDbSName, zTab); +#endif + }else{ + /* The sqlite_stat[134] table already exists. Delete all rows. */ + sqlite3VdbeAddOp2(v, OP_Clear, aRoot[i], iDb); + } + } + } + + /* Open the sqlite_stat[134] tables for writing. */ + for(i=0; aTable[i].zCols; i++){ + assert( inRowid ){ + sqlite3DbFree(db, p->u.aRowid); + p->nRowid = 0; + } +} +#endif + +/* Initialize the BLOB value of a ROWID +*/ +#ifdef SQLITE_ENABLE_STAT4 +static void sampleSetRowid(sqlite3 *db, Stat4Sample *p, int n, const u8 *pData){ + assert( db!=0 ); + if( p->nRowid ) sqlite3DbFree(db, p->u.aRowid); + p->u.aRowid = sqlite3DbMallocRawNN(db, n); + if( p->u.aRowid ){ + p->nRowid = n; + memcpy(p->u.aRowid, pData, n); + }else{ + p->nRowid = 0; + } +} +#endif + +/* Initialize the INTEGER value of a ROWID. +*/ +#ifdef SQLITE_ENABLE_STAT4 +static void sampleSetRowidInt64(sqlite3 *db, Stat4Sample *p, i64 iRowid){ + assert( db!=0 ); + if( p->nRowid ) sqlite3DbFree(db, p->u.aRowid); + p->nRowid = 0; + p->u.iRowid = iRowid; +} +#endif + + +/* +** Copy the contents of object (*pFrom) into (*pTo). +*/ +#ifdef SQLITE_ENABLE_STAT4 +static void sampleCopy(Stat4Accum *p, Stat4Sample *pTo, Stat4Sample *pFrom){ + pTo->isPSample = pFrom->isPSample; + pTo->iCol = pFrom->iCol; + pTo->iHash = pFrom->iHash; + memcpy(pTo->anEq, pFrom->anEq, sizeof(tRowcnt)*p->nCol); + memcpy(pTo->anLt, pFrom->anLt, sizeof(tRowcnt)*p->nCol); + memcpy(pTo->anDLt, pFrom->anDLt, sizeof(tRowcnt)*p->nCol); + if( pFrom->nRowid ){ + sampleSetRowid(p->db, pTo, pFrom->nRowid, pFrom->u.aRowid); + }else{ + sampleSetRowidInt64(p->db, pTo, pFrom->u.iRowid); + } +} +#endif + +/* +** Reclaim all memory of a Stat4Accum structure. +*/ +static void stat4Destructor(void *pOld){ + Stat4Accum *p = (Stat4Accum*)pOld; +#ifdef SQLITE_ENABLE_STAT4 + int i; + for(i=0; inCol; i++) sampleClear(p->db, p->aBest+i); + for(i=0; imxSample; i++) sampleClear(p->db, p->a+i); + sampleClear(p->db, &p->current); +#endif + sqlite3DbFree(p->db, p); +} + +/* +** Implementation of the stat_init(N,K,C) SQL function. The three parameters +** are: +** N: The number of columns in the index including the rowid/pk (note 1) +** K: The number of columns in the index excluding the rowid/pk. +** C: The number of rows in the index (note 2) +** +** Note 1: In the special case of the covering index that implements a +** WITHOUT ROWID table, N is the number of PRIMARY KEY columns, not the +** total number of columns in the table. +** +** Note 2: C is only used for STAT4. +** +** For indexes on ordinary rowid tables, N==K+1. But for indexes on +** WITHOUT ROWID tables, N=K+P where P is the number of columns in the +** PRIMARY KEY of the table. The covering index that implements the +** original WITHOUT ROWID table as N==K as a special case. +** +** This routine allocates the Stat4Accum object in heap memory. The return +** value is a pointer to the Stat4Accum object. The datatype of the +** return value is BLOB, but it is really just a pointer to the Stat4Accum +** object. +*/ +static void statInit( + sqlite3_context *context, + int argc, + sqlite3_value **argv +){ + Stat4Accum *p; + int nCol; /* Number of columns in index being sampled */ + int nKeyCol; /* Number of key columns */ + int nColUp; /* nCol rounded up for alignment */ + int n; /* Bytes of space to allocate */ + sqlite3 *db; /* Database connection */ +#ifdef SQLITE_ENABLE_STAT4 + int mxSample = SQLITE_STAT4_SAMPLES; +#endif + + /* Decode the three function arguments */ + UNUSED_PARAMETER(argc); + nCol = sqlite3_value_int(argv[0]); + assert( nCol>0 ); + nColUp = sizeof(tRowcnt)<8 ? (nCol+1)&~1 : nCol; + nKeyCol = sqlite3_value_int(argv[1]); + assert( nKeyCol<=nCol ); + assert( nKeyCol>0 ); + + /* Allocate the space required for the Stat4Accum object */ + n = sizeof(*p) + + sizeof(tRowcnt)*nColUp /* Stat4Accum.anEq */ + + sizeof(tRowcnt)*nColUp /* Stat4Accum.anDLt */ +#ifdef SQLITE_ENABLE_STAT4 + + sizeof(tRowcnt)*nColUp /* Stat4Accum.anLt */ + + sizeof(Stat4Sample)*(nCol+mxSample) /* Stat4Accum.aBest[], a[] */ + + sizeof(tRowcnt)*3*nColUp*(nCol+mxSample) +#endif + ; + db = sqlite3_context_db_handle(context); + p = sqlite3DbMallocZero(db, n); + if( p==0 ){ + sqlite3_result_error_nomem(context); + return; + } + + p->db = db; + p->nRow = 0; + p->nCol = nCol; + p->nKeyCol = nKeyCol; + p->current.anDLt = (tRowcnt*)&p[1]; + p->current.anEq = &p->current.anDLt[nColUp]; + +#ifdef SQLITE_ENABLE_STAT4 + { + u8 *pSpace; /* Allocated space not yet assigned */ + int i; /* Used to iterate through p->aSample[] */ + + p->iGet = -1; + p->mxSample = mxSample; + p->nPSample = (tRowcnt)(sqlite3_value_int64(argv[2])/(mxSample/3+1) + 1); + p->current.anLt = &p->current.anEq[nColUp]; + p->iPrn = 0x689e962d*(u32)nCol ^ 0xd0944565*(u32)sqlite3_value_int(argv[2]); + + /* Set up the Stat4Accum.a[] and aBest[] arrays */ + p->a = (struct Stat4Sample*)&p->current.anLt[nColUp]; + p->aBest = &p->a[mxSample]; + pSpace = (u8*)(&p->a[mxSample+nCol]); + for(i=0; i<(mxSample+nCol); i++){ + p->a[i].anEq = (tRowcnt *)pSpace; pSpace += (sizeof(tRowcnt) * nColUp); + p->a[i].anLt = (tRowcnt *)pSpace; pSpace += (sizeof(tRowcnt) * nColUp); + p->a[i].anDLt = (tRowcnt *)pSpace; pSpace += (sizeof(tRowcnt) * nColUp); + } + assert( (pSpace - (u8*)p)==n ); + + for(i=0; iaBest[i].iCol = i; + } + } +#endif + + /* Return a pointer to the allocated object to the caller. Note that + ** only the pointer (the 2nd parameter) matters. The size of the object + ** (given by the 3rd parameter) is never used and can be any positive + ** value. */ + sqlite3_result_blob(context, p, sizeof(*p), stat4Destructor); +} +static const FuncDef statInitFuncdef = { + 2+IsStat4, /* nArg */ + SQLITE_UTF8, /* funcFlags */ + 0, /* pUserData */ + 0, /* pNext */ + statInit, /* xSFunc */ + 0, /* xFinalize */ + 0, 0, /* xValue, xInverse */ + "stat_init", /* zName */ + {0} +}; + +#ifdef SQLITE_ENABLE_STAT4 +/* +** pNew and pOld are both candidate non-periodic samples selected for +** the same column (pNew->iCol==pOld->iCol). Ignoring this column and +** considering only any trailing columns and the sample hash value, this +** function returns true if sample pNew is to be preferred over pOld. +** In other words, if we assume that the cardinalities of the selected +** column for pNew and pOld are equal, is pNew to be preferred over pOld. +** +** This function assumes that for each argument sample, the contents of +** the anEq[] array from pSample->anEq[pSample->iCol+1] onwards are valid. +*/ +static int sampleIsBetterPost( + Stat4Accum *pAccum, + Stat4Sample *pNew, + Stat4Sample *pOld +){ + int nCol = pAccum->nCol; + int i; + assert( pNew->iCol==pOld->iCol ); + for(i=pNew->iCol+1; ianEq[i]>pOld->anEq[i] ) return 1; + if( pNew->anEq[i]anEq[i] ) return 0; + } + if( pNew->iHash>pOld->iHash ) return 1; + return 0; +} +#endif + +#ifdef SQLITE_ENABLE_STAT4 +/* +** Return true if pNew is to be preferred over pOld. +** +** This function assumes that for each argument sample, the contents of +** the anEq[] array from pSample->anEq[pSample->iCol] onwards are valid. +*/ +static int sampleIsBetter( + Stat4Accum *pAccum, + Stat4Sample *pNew, + Stat4Sample *pOld +){ + tRowcnt nEqNew = pNew->anEq[pNew->iCol]; + tRowcnt nEqOld = pOld->anEq[pOld->iCol]; + + assert( pOld->isPSample==0 && pNew->isPSample==0 ); + assert( IsStat4 || (pNew->iCol==0 && pOld->iCol==0) ); + + if( (nEqNew>nEqOld) ) return 1; + if( nEqNew==nEqOld ){ + if( pNew->iColiCol ) return 1; + return (pNew->iCol==pOld->iCol && sampleIsBetterPost(pAccum, pNew, pOld)); + } + return 0; +} + +/* +** Copy the contents of sample *pNew into the p->a[] array. If necessary, +** remove the least desirable sample from p->a[] to make room. +*/ +static void sampleInsert(Stat4Accum *p, Stat4Sample *pNew, int nEqZero){ + Stat4Sample *pSample = 0; + int i; + + assert( IsStat4 || nEqZero==0 ); + + /* Stat4Accum.nMaxEqZero is set to the maximum number of leading 0 + ** values in the anEq[] array of any sample in Stat4Accum.a[]. In + ** other words, if nMaxEqZero is n, then it is guaranteed that there + ** are no samples with Stat4Sample.anEq[m]==0 for (m>=n). */ + if( nEqZero>p->nMaxEqZero ){ + p->nMaxEqZero = nEqZero; + } + if( pNew->isPSample==0 ){ + Stat4Sample *pUpgrade = 0; + assert( pNew->anEq[pNew->iCol]>0 ); + + /* This sample is being added because the prefix that ends in column + ** iCol occurs many times in the table. However, if we have already + ** added a sample that shares this prefix, there is no need to add + ** this one. Instead, upgrade the priority of the highest priority + ** existing sample that shares this prefix. */ + for(i=p->nSample-1; i>=0; i--){ + Stat4Sample *pOld = &p->a[i]; + if( pOld->anEq[pNew->iCol]==0 ){ + if( pOld->isPSample ) return; + assert( pOld->iCol>pNew->iCol ); + assert( sampleIsBetter(p, pNew, pOld) ); + if( pUpgrade==0 || sampleIsBetter(p, pOld, pUpgrade) ){ + pUpgrade = pOld; + } + } + } + if( pUpgrade ){ + pUpgrade->iCol = pNew->iCol; + pUpgrade->anEq[pUpgrade->iCol] = pNew->anEq[pUpgrade->iCol]; + goto find_new_min; + } + } + + /* If necessary, remove sample iMin to make room for the new sample. */ + if( p->nSample>=p->mxSample ){ + Stat4Sample *pMin = &p->a[p->iMin]; + tRowcnt *anEq = pMin->anEq; + tRowcnt *anLt = pMin->anLt; + tRowcnt *anDLt = pMin->anDLt; + sampleClear(p->db, pMin); + memmove(pMin, &pMin[1], sizeof(p->a[0])*(p->nSample-p->iMin-1)); + pSample = &p->a[p->nSample-1]; + pSample->nRowid = 0; + pSample->anEq = anEq; + pSample->anDLt = anDLt; + pSample->anLt = anLt; + p->nSample = p->mxSample-1; + } + + /* The "rows less-than" for the rowid column must be greater than that + ** for the last sample in the p->a[] array. Otherwise, the samples would + ** be out of order. */ + assert( p->nSample==0 + || pNew->anLt[p->nCol-1] > p->a[p->nSample-1].anLt[p->nCol-1] ); + + /* Insert the new sample */ + pSample = &p->a[p->nSample]; + sampleCopy(p, pSample, pNew); + p->nSample++; + + /* Zero the first nEqZero entries in the anEq[] array. */ + memset(pSample->anEq, 0, sizeof(tRowcnt)*nEqZero); + +find_new_min: + if( p->nSample>=p->mxSample ){ + int iMin = -1; + for(i=0; imxSample; i++){ + if( p->a[i].isPSample ) continue; + if( iMin<0 || sampleIsBetter(p, &p->a[iMin], &p->a[i]) ){ + iMin = i; + } + } + assert( iMin>=0 ); + p->iMin = iMin; + } +} +#endif /* SQLITE_ENABLE_STAT4 */ + +/* +** Field iChng of the index being scanned has changed. So at this point +** p->current contains a sample that reflects the previous row of the +** index. The value of anEq[iChng] and subsequent anEq[] elements are +** correct at this point. +*/ +static void samplePushPrevious(Stat4Accum *p, int iChng){ +#ifdef SQLITE_ENABLE_STAT4 + int i; + + /* Check if any samples from the aBest[] array should be pushed + ** into IndexSample.a[] at this point. */ + for(i=(p->nCol-2); i>=iChng; i--){ + Stat4Sample *pBest = &p->aBest[i]; + pBest->anEq[i] = p->current.anEq[i]; + if( p->nSamplemxSample || sampleIsBetter(p, pBest, &p->a[p->iMin]) ){ + sampleInsert(p, pBest, i); + } + } + + /* Check that no sample contains an anEq[] entry with an index of + ** p->nMaxEqZero or greater set to zero. */ + for(i=p->nSample-1; i>=0; i--){ + int j; + for(j=p->nMaxEqZero; jnCol; j++) assert( p->a[i].anEq[j]>0 ); + } + + /* Update the anEq[] fields of any samples already collected. */ + if( iChngnMaxEqZero ){ + for(i=p->nSample-1; i>=0; i--){ + int j; + for(j=iChng; jnCol; j++){ + if( p->a[i].anEq[j]==0 ) p->a[i].anEq[j] = p->current.anEq[j]; + } + } + p->nMaxEqZero = iChng; + } +#endif + +#ifndef SQLITE_ENABLE_STAT4 + UNUSED_PARAMETER( p ); + UNUSED_PARAMETER( iChng ); +#endif +} + +/* +** Implementation of the stat_push SQL function: stat_push(P,C,R) +** Arguments: +** +** P Pointer to the Stat4Accum object created by stat_init() +** C Index of left-most column to differ from previous row +** R Rowid for the current row. Might be a key record for +** WITHOUT ROWID tables. +** +** This SQL function always returns NULL. It's purpose it to accumulate +** statistical data and/or samples in the Stat4Accum object about the +** index being analyzed. The stat_get() SQL function will later be used to +** extract relevant information for constructing the sqlite_statN tables. +** +** The R parameter is only used for STAT4 +*/ +static void statPush( + sqlite3_context *context, + int argc, + sqlite3_value **argv +){ + int i; + + /* The three function arguments */ + Stat4Accum *p = (Stat4Accum*)sqlite3_value_blob(argv[0]); + int iChng = sqlite3_value_int(argv[1]); + + UNUSED_PARAMETER( argc ); + UNUSED_PARAMETER( context ); + assert( p->nCol>0 ); + assert( iChngnCol ); + + if( p->nRow==0 ){ + /* This is the first call to this function. Do initialization. */ + for(i=0; inCol; i++) p->current.anEq[i] = 1; + }else{ + /* Second and subsequent calls get processed here */ + samplePushPrevious(p, iChng); + + /* Update anDLt[], anLt[] and anEq[] to reflect the values that apply + ** to the current row of the index. */ + for(i=0; icurrent.anEq[i]++; + } + for(i=iChng; inCol; i++){ + p->current.anDLt[i]++; +#ifdef SQLITE_ENABLE_STAT4 + p->current.anLt[i] += p->current.anEq[i]; +#endif + p->current.anEq[i] = 1; + } + } + p->nRow++; +#ifdef SQLITE_ENABLE_STAT4 + if( sqlite3_value_type(argv[2])==SQLITE_INTEGER ){ + sampleSetRowidInt64(p->db, &p->current, sqlite3_value_int64(argv[2])); + }else{ + sampleSetRowid(p->db, &p->current, sqlite3_value_bytes(argv[2]), + sqlite3_value_blob(argv[2])); + } + p->current.iHash = p->iPrn = p->iPrn*1103515245 + 12345; +#endif + +#ifdef SQLITE_ENABLE_STAT4 + { + tRowcnt nLt = p->current.anLt[p->nCol-1]; + + /* Check if this is to be a periodic sample. If so, add it. */ + if( (nLt/p->nPSample)!=(nLt+1)/p->nPSample ){ + p->current.isPSample = 1; + p->current.iCol = 0; + sampleInsert(p, &p->current, p->nCol-1); + p->current.isPSample = 0; + } + + /* Update the aBest[] array. */ + for(i=0; i<(p->nCol-1); i++){ + p->current.iCol = i; + if( i>=iChng || sampleIsBetterPost(p, &p->current, &p->aBest[i]) ){ + sampleCopy(p, &p->aBest[i], &p->current); + } + } + } +#endif +} +static const FuncDef statPushFuncdef = { + 2+IsStat4, /* nArg */ + SQLITE_UTF8, /* funcFlags */ + 0, /* pUserData */ + 0, /* pNext */ + statPush, /* xSFunc */ + 0, /* xFinalize */ + 0, 0, /* xValue, xInverse */ + "stat_push", /* zName */ + {0} +}; + +#define STAT_GET_STAT1 0 /* "stat" column of stat1 table */ +#define STAT_GET_ROWID 1 /* "rowid" column of stat[34] entry */ +#define STAT_GET_NEQ 2 /* "neq" column of stat[34] entry */ +#define STAT_GET_NLT 3 /* "nlt" column of stat[34] entry */ +#define STAT_GET_NDLT 4 /* "ndlt" column of stat[34] entry */ + +/* +** Implementation of the stat_get(P,J) SQL function. This routine is +** used to query statistical information that has been gathered into +** the Stat4Accum object by prior calls to stat_push(). The P parameter +** has type BLOB but it is really just a pointer to the Stat4Accum object. +** The content to returned is determined by the parameter J +** which is one of the STAT_GET_xxxx values defined above. +** +** The stat_get(P,J) function is not available to generic SQL. It is +** inserted as part of a manually constructed bytecode program. (See +** the callStatGet() routine below.) It is guaranteed that the P +** parameter will always be a poiner to a Stat4Accum object, never a +** NULL. +** +** If STAT4 is not enabled, then J is always +** STAT_GET_STAT1 and is hence omitted and this routine becomes +** a one-parameter function, stat_get(P), that always returns the +** stat1 table entry information. +*/ +static void statGet( + sqlite3_context *context, + int argc, + sqlite3_value **argv +){ + Stat4Accum *p = (Stat4Accum*)sqlite3_value_blob(argv[0]); +#ifdef SQLITE_ENABLE_STAT4 + /* STAT4 has a parameter on this routine. */ + int eCall = sqlite3_value_int(argv[1]); + assert( argc==2 ); + assert( eCall==STAT_GET_STAT1 || eCall==STAT_GET_NEQ + || eCall==STAT_GET_ROWID || eCall==STAT_GET_NLT + || eCall==STAT_GET_NDLT + ); + if( eCall==STAT_GET_STAT1 ) +#else + assert( argc==1 ); +#endif + { + /* Return the value to store in the "stat" column of the sqlite_stat1 + ** table for this index. + ** + ** The value is a string composed of a list of integers describing + ** the index. The first integer in the list is the total number of + ** entries in the index. There is one additional integer in the list + ** for each indexed column. This additional integer is an estimate of + ** the number of rows matched by a stabbing query on the index using + ** a key with the corresponding number of fields. In other words, + ** if the index is on columns (a,b) and the sqlite_stat1 value is + ** "100 10 2", then SQLite estimates that: + ** + ** * the index contains 100 rows, + ** * "WHERE a=?" matches 10 rows, and + ** * "WHERE a=? AND b=?" matches 2 rows. + ** + ** If D is the count of distinct values and K is the total number of + ** rows, then each estimate is computed as: + ** + ** I = (K+D-1)/D + */ + char *z; + int i; + + char *zRet = sqlite3MallocZero( (p->nKeyCol+1)*25 ); + if( zRet==0 ){ + sqlite3_result_error_nomem(context); + return; + } + + sqlite3_snprintf(24, zRet, "%llu", (u64)p->nRow); + z = zRet + sqlite3Strlen30(zRet); + for(i=0; inKeyCol; i++){ + u64 nDistinct = p->current.anDLt[i] + 1; + u64 iVal = (p->nRow + nDistinct - 1) / nDistinct; + sqlite3_snprintf(24, z, " %llu", iVal); + z += sqlite3Strlen30(z); + assert( p->current.anEq[i] ); + } + assert( z[0]=='\0' && z>zRet ); + + sqlite3_result_text(context, zRet, -1, sqlite3_free); + } +#ifdef SQLITE_ENABLE_STAT4 + else if( eCall==STAT_GET_ROWID ){ + if( p->iGet<0 ){ + samplePushPrevious(p, 0); + p->iGet = 0; + } + if( p->iGetnSample ){ + Stat4Sample *pS = p->a + p->iGet; + if( pS->nRowid==0 ){ + sqlite3_result_int64(context, pS->u.iRowid); + }else{ + sqlite3_result_blob(context, pS->u.aRowid, pS->nRowid, + SQLITE_TRANSIENT); + } + } + }else{ + tRowcnt *aCnt = 0; + + assert( p->iGetnSample ); + switch( eCall ){ + case STAT_GET_NEQ: aCnt = p->a[p->iGet].anEq; break; + case STAT_GET_NLT: aCnt = p->a[p->iGet].anLt; break; + default: { + aCnt = p->a[p->iGet].anDLt; + p->iGet++; + break; + } + } + + { + char *zRet = sqlite3MallocZero(p->nCol * 25); + if( zRet==0 ){ + sqlite3_result_error_nomem(context); + }else{ + int i; + char *z = zRet; + for(i=0; inCol; i++){ + sqlite3_snprintf(24, z, "%llu ", (u64)aCnt[i]); + z += sqlite3Strlen30(z); + } + assert( z[0]=='\0' && z>zRet ); + z[-1] = '\0'; + sqlite3_result_text(context, zRet, -1, sqlite3_free); + } + } + } +#endif /* SQLITE_ENABLE_STAT4 */ +#ifndef SQLITE_DEBUG + UNUSED_PARAMETER( argc ); +#endif +} +static const FuncDef statGetFuncdef = { + 1+IsStat4, /* nArg */ + SQLITE_UTF8, /* funcFlags */ + 0, /* pUserData */ + 0, /* pNext */ + statGet, /* xSFunc */ + 0, /* xFinalize */ + 0, 0, /* xValue, xInverse */ + "stat_get", /* zName */ + {0} +}; + +static void callStatGet(Vdbe *v, int regStat4, int iParam, int regOut){ + assert( regOut!=regStat4 && regOut!=regStat4+1 ); +#ifdef SQLITE_ENABLE_STAT4 + sqlite3VdbeAddOp2(v, OP_Integer, iParam, regStat4+1); +#elif SQLITE_DEBUG + assert( iParam==STAT_GET_STAT1 ); +#else + UNUSED_PARAMETER( iParam ); +#endif + sqlite3VdbeAddOp4(v, OP_Function0, 0, regStat4, regOut, + (char*)&statGetFuncdef, P4_FUNCDEF); + sqlite3VdbeChangeP5(v, 1 + IsStat4); +} + +/* +** Generate code to do an analysis of all indices associated with +** a single table. +*/ +static void analyzeOneTable( + Parse *pParse, /* Parser context */ + Table *pTab, /* Table whose indices are to be analyzed */ + Index *pOnlyIdx, /* If not NULL, only analyze this one index */ + int iStatCur, /* Index of VdbeCursor that writes the sqlite_stat1 table */ + int iMem, /* Available memory locations begin here */ + int iTab /* Next available cursor */ +){ + sqlite3 *db = pParse->db; /* Database handle */ + Index *pIdx; /* An index to being analyzed */ + int iIdxCur; /* Cursor open on index being analyzed */ + int iTabCur; /* Table cursor */ + Vdbe *v; /* The virtual machine being built up */ + int i; /* Loop counter */ + int jZeroRows = -1; /* Jump from here if number of rows is zero */ + int iDb; /* Index of database containing pTab */ + u8 needTableCnt = 1; /* True to count the table */ + int regNewRowid = iMem++; /* Rowid for the inserted record */ + int regStat4 = iMem++; /* Register to hold Stat4Accum object */ + int regChng = iMem++; /* Index of changed index field */ +#ifdef SQLITE_ENABLE_STAT4 + int regRowid = iMem++; /* Rowid argument passed to stat_push() */ +#endif + int regTemp = iMem++; /* Temporary use register */ + int regTabname = iMem++; /* Register containing table name */ + int regIdxname = iMem++; /* Register containing index name */ + int regStat1 = iMem++; /* Value for the stat column of sqlite_stat1 */ + int regPrev = iMem; /* MUST BE LAST (see below) */ +#ifdef SQLITE_ENABLE_PREUPDATE_HOOK + Table *pStat1 = 0; +#endif + + pParse->nMem = MAX(pParse->nMem, iMem); + v = sqlite3GetVdbe(pParse); + if( v==0 || NEVER(pTab==0) ){ + return; + } + if( pTab->tnum==0 ){ + /* Do not gather statistics on views or virtual tables */ + return; + } + if( sqlite3_strlike("sqlite\\_%", pTab->zName, '\\')==0 ){ + /* Do not gather statistics on system tables */ + return; + } + assert( sqlite3BtreeHoldsAllMutexes(db) ); + iDb = sqlite3SchemaToIndex(db, pTab->pSchema); + assert( iDb>=0 ); + assert( sqlite3SchemaMutexHeld(db, iDb, 0) ); +#ifndef SQLITE_OMIT_AUTHORIZATION + if( sqlite3AuthCheck(pParse, SQLITE_ANALYZE, pTab->zName, 0, + db->aDb[iDb].zDbSName ) ){ + return; + } +#endif + +#ifdef SQLITE_ENABLE_PREUPDATE_HOOK + if( db->xPreUpdateCallback ){ + pStat1 = (Table*)sqlite3DbMallocZero(db, sizeof(Table) + 13); + if( pStat1==0 ) return; + pStat1->zName = (char*)&pStat1[1]; + memcpy(pStat1->zName, "sqlite_stat1", 13); + pStat1->nCol = 3; + pStat1->iPKey = -1; + sqlite3VdbeAddOp4(pParse->pVdbe, OP_Noop, 0, 0, 0,(char*)pStat1,P4_DYNBLOB); + } +#endif + + /* Establish a read-lock on the table at the shared-cache level. + ** Open a read-only cursor on the table. Also allocate a cursor number + ** to use for scanning indexes (iIdxCur). No index cursor is opened at + ** this time though. */ + sqlite3TableLock(pParse, iDb, pTab->tnum, 0, pTab->zName); + iTabCur = iTab++; + iIdxCur = iTab++; + pParse->nTab = MAX(pParse->nTab, iTab); + sqlite3OpenTable(pParse, iTabCur, iDb, pTab, OP_OpenRead); + sqlite3VdbeLoadString(v, regTabname, pTab->zName); + + for(pIdx=pTab->pIndex; pIdx; pIdx=pIdx->pNext){ + int nCol; /* Number of columns in pIdx. "N" */ + int addrRewind; /* Address of "OP_Rewind iIdxCur" */ + int addrNextRow; /* Address of "next_row:" */ + const char *zIdxName; /* Name of the index */ + int nColTest; /* Number of columns to test for changes */ + + if( pOnlyIdx && pOnlyIdx!=pIdx ) continue; + if( pIdx->pPartIdxWhere==0 ) needTableCnt = 0; + if( !HasRowid(pTab) && IsPrimaryKeyIndex(pIdx) ){ + nCol = pIdx->nKeyCol; + zIdxName = pTab->zName; + nColTest = nCol - 1; + }else{ + nCol = pIdx->nColumn; + zIdxName = pIdx->zName; + nColTest = pIdx->uniqNotNull ? pIdx->nKeyCol-1 : nCol-1; + } + + /* Populate the register containing the index name. */ + sqlite3VdbeLoadString(v, regIdxname, zIdxName); + VdbeComment((v, "Analysis for %s.%s", pTab->zName, zIdxName)); + + /* + ** Pseudo-code for loop that calls stat_push(): + ** + ** Rewind csr + ** if eof(csr) goto end_of_scan; + ** regChng = 0 + ** goto chng_addr_0; + ** + ** next_row: + ** regChng = 0 + ** if( idx(0) != regPrev(0) ) goto chng_addr_0 + ** regChng = 1 + ** if( idx(1) != regPrev(1) ) goto chng_addr_1 + ** ... + ** regChng = N + ** goto chng_addr_N + ** + ** chng_addr_0: + ** regPrev(0) = idx(0) + ** chng_addr_1: + ** regPrev(1) = idx(1) + ** ... + ** + ** endDistinctTest: + ** regRowid = idx(rowid) + ** stat_push(P, regChng, regRowid) + ** Next csr + ** if !eof(csr) goto next_row; + ** + ** end_of_scan: + */ + + /* Make sure there are enough memory cells allocated to accommodate + ** the regPrev array and a trailing rowid (the rowid slot is required + ** when building a record to insert into the sample column of + ** the sqlite_stat4 table. */ + pParse->nMem = MAX(pParse->nMem, regPrev+nColTest); + + /* Open a read-only cursor on the index being analyzed. */ + assert( iDb==sqlite3SchemaToIndex(db, pIdx->pSchema) ); + sqlite3VdbeAddOp3(v, OP_OpenRead, iIdxCur, pIdx->tnum, iDb); + sqlite3VdbeSetP4KeyInfo(pParse, pIdx); + VdbeComment((v, "%s", pIdx->zName)); + + /* Invoke the stat_init() function. The arguments are: + ** + ** (1) the number of columns in the index including the rowid + ** (or for a WITHOUT ROWID table, the number of PK columns), + ** (2) the number of columns in the key without the rowid/pk + ** (3) the number of rows in the index, + ** + ** + ** The third argument is only used for STAT4 + */ +#ifdef SQLITE_ENABLE_STAT4 + sqlite3VdbeAddOp2(v, OP_Count, iIdxCur, regStat4+3); +#endif + sqlite3VdbeAddOp2(v, OP_Integer, nCol, regStat4+1); + sqlite3VdbeAddOp2(v, OP_Integer, pIdx->nKeyCol, regStat4+2); + sqlite3VdbeAddOp4(v, OP_Function0, 0, regStat4+1, regStat4, + (char*)&statInitFuncdef, P4_FUNCDEF); + sqlite3VdbeChangeP5(v, 2+IsStat4); + + /* Implementation of the following: + ** + ** Rewind csr + ** if eof(csr) goto end_of_scan; + ** regChng = 0 + ** goto next_push_0; + ** + */ + addrRewind = sqlite3VdbeAddOp1(v, OP_Rewind, iIdxCur); + VdbeCoverage(v); + sqlite3VdbeAddOp2(v, OP_Integer, 0, regChng); + addrNextRow = sqlite3VdbeCurrentAddr(v); + + if( nColTest>0 ){ + int endDistinctTest = sqlite3VdbeMakeLabel(pParse); + int *aGotoChng; /* Array of jump instruction addresses */ + aGotoChng = sqlite3DbMallocRawNN(db, sizeof(int)*nColTest); + if( aGotoChng==0 ) continue; + + /* + ** next_row: + ** regChng = 0 + ** if( idx(0) != regPrev(0) ) goto chng_addr_0 + ** regChng = 1 + ** if( idx(1) != regPrev(1) ) goto chng_addr_1 + ** ... + ** regChng = N + ** goto endDistinctTest + */ + sqlite3VdbeAddOp0(v, OP_Goto); + addrNextRow = sqlite3VdbeCurrentAddr(v); + if( nColTest==1 && pIdx->nKeyCol==1 && IsUniqueIndex(pIdx) ){ + /* For a single-column UNIQUE index, once we have found a non-NULL + ** row, we know that all the rest will be distinct, so skip + ** subsequent distinctness tests. */ + sqlite3VdbeAddOp2(v, OP_NotNull, regPrev, endDistinctTest); + VdbeCoverage(v); + } + for(i=0; iazColl[i]); + sqlite3VdbeAddOp2(v, OP_Integer, i, regChng); + sqlite3VdbeAddOp3(v, OP_Column, iIdxCur, i, regTemp); + aGotoChng[i] = + sqlite3VdbeAddOp4(v, OP_Ne, regTemp, 0, regPrev+i, pColl, P4_COLLSEQ); + sqlite3VdbeChangeP5(v, SQLITE_NULLEQ); + VdbeCoverage(v); + } + sqlite3VdbeAddOp2(v, OP_Integer, nColTest, regChng); + sqlite3VdbeGoto(v, endDistinctTest); + + + /* + ** chng_addr_0: + ** regPrev(0) = idx(0) + ** chng_addr_1: + ** regPrev(1) = idx(1) + ** ... + */ + sqlite3VdbeJumpHere(v, addrNextRow-1); + for(i=0; ipTable); + int j, k, regKey; + regKey = sqlite3GetTempRange(pParse, pPk->nKeyCol); + for(j=0; jnKeyCol; j++){ + k = sqlite3ColumnOfIndex(pIdx, pPk->aiColumn[j]); + assert( k>=0 && knColumn ); + sqlite3VdbeAddOp3(v, OP_Column, iIdxCur, k, regKey+j); + VdbeComment((v, "%s", pTab->aCol[pPk->aiColumn[j]].zName)); + } + sqlite3VdbeAddOp3(v, OP_MakeRecord, regKey, pPk->nKeyCol, regRowid); + sqlite3ReleaseTempRange(pParse, regKey, pPk->nKeyCol); + } +#endif + assert( regChng==(regStat4+1) ); + sqlite3VdbeAddOp4(v, OP_Function0, 1, regStat4, regTemp, + (char*)&statPushFuncdef, P4_FUNCDEF); + sqlite3VdbeChangeP5(v, 2+IsStat4); + sqlite3VdbeAddOp2(v, OP_Next, iIdxCur, addrNextRow); VdbeCoverage(v); + + /* Add the entry to the stat1 table. */ + callStatGet(v, regStat4, STAT_GET_STAT1, regStat1); + assert( "BBB"[0]==SQLITE_AFF_TEXT ); + sqlite3VdbeAddOp4(v, OP_MakeRecord, regTabname, 3, regTemp, "BBB", 0); + sqlite3VdbeAddOp2(v, OP_NewRowid, iStatCur, regNewRowid); + sqlite3VdbeAddOp3(v, OP_Insert, iStatCur, regTemp, regNewRowid); +#ifdef SQLITE_ENABLE_PREUPDATE_HOOK + sqlite3VdbeChangeP4(v, -1, (char*)pStat1, P4_TABLE); +#endif + sqlite3VdbeChangeP5(v, OPFLAG_APPEND); + + /* Add the entries to the stat4 table. */ +#ifdef SQLITE_ENABLE_STAT4 + { + int regEq = regStat1; + int regLt = regStat1+1; + int regDLt = regStat1+2; + int regSample = regStat1+3; + int regCol = regStat1+4; + int regSampleRowid = regCol + nCol; + int addrNext; + int addrIsNull; + u8 seekOp = HasRowid(pTab) ? OP_NotExists : OP_NotFound; + + pParse->nMem = MAX(pParse->nMem, regCol+nCol); + + addrNext = sqlite3VdbeCurrentAddr(v); + callStatGet(v, regStat4, STAT_GET_ROWID, regSampleRowid); + addrIsNull = sqlite3VdbeAddOp1(v, OP_IsNull, regSampleRowid); + VdbeCoverage(v); + callStatGet(v, regStat4, STAT_GET_NEQ, regEq); + callStatGet(v, regStat4, STAT_GET_NLT, regLt); + callStatGet(v, regStat4, STAT_GET_NDLT, regDLt); + sqlite3VdbeAddOp4Int(v, seekOp, iTabCur, addrNext, regSampleRowid, 0); + VdbeCoverage(v); + for(i=0; izName)); + sqlite3VdbeAddOp2(v, OP_Count, iTabCur, regStat1); + jZeroRows = sqlite3VdbeAddOp1(v, OP_IfNot, regStat1); VdbeCoverage(v); + sqlite3VdbeAddOp2(v, OP_Null, 0, regIdxname); + assert( "BBB"[0]==SQLITE_AFF_TEXT ); + sqlite3VdbeAddOp4(v, OP_MakeRecord, regTabname, 3, regTemp, "BBB", 0); + sqlite3VdbeAddOp2(v, OP_NewRowid, iStatCur, regNewRowid); + sqlite3VdbeAddOp3(v, OP_Insert, iStatCur, regTemp, regNewRowid); + sqlite3VdbeChangeP5(v, OPFLAG_APPEND); +#ifdef SQLITE_ENABLE_PREUPDATE_HOOK + sqlite3VdbeChangeP4(v, -1, (char*)pStat1, P4_TABLE); +#endif + sqlite3VdbeJumpHere(v, jZeroRows); + } +} + + +/* +** Generate code that will cause the most recent index analysis to +** be loaded into internal hash tables where is can be used. +*/ +static void loadAnalysis(Parse *pParse, int iDb){ + Vdbe *v = sqlite3GetVdbe(pParse); + if( v ){ + sqlite3VdbeAddOp1(v, OP_LoadAnalysis, iDb); + } +} + +/* +** Generate code that will do an analysis of an entire database +*/ +static void analyzeDatabase(Parse *pParse, int iDb){ + sqlite3 *db = pParse->db; + Schema *pSchema = db->aDb[iDb].pSchema; /* Schema of database iDb */ + HashElem *k; + int iStatCur; + int iMem; + int iTab; + + sqlite3BeginWriteOperation(pParse, 0, iDb); + iStatCur = pParse->nTab; + pParse->nTab += 3; + openStatTable(pParse, iDb, iStatCur, 0, 0); + iMem = pParse->nMem+1; + iTab = pParse->nTab; + assert( sqlite3SchemaMutexHeld(db, iDb, 0) ); + for(k=sqliteHashFirst(&pSchema->tblHash); k; k=sqliteHashNext(k)){ + Table *pTab = (Table*)sqliteHashData(k); + analyzeOneTable(pParse, pTab, 0, iStatCur, iMem, iTab); + } + loadAnalysis(pParse, iDb); +} + +/* +** Generate code that will do an analysis of a single table in +** a database. If pOnlyIdx is not NULL then it is a single index +** in pTab that should be analyzed. +*/ +static void analyzeTable(Parse *pParse, Table *pTab, Index *pOnlyIdx){ + int iDb; + int iStatCur; + + assert( pTab!=0 ); + assert( sqlite3BtreeHoldsAllMutexes(pParse->db) ); + iDb = sqlite3SchemaToIndex(pParse->db, pTab->pSchema); + sqlite3BeginWriteOperation(pParse, 0, iDb); + iStatCur = pParse->nTab; + pParse->nTab += 3; + if( pOnlyIdx ){ + openStatTable(pParse, iDb, iStatCur, pOnlyIdx->zName, "idx"); + }else{ + openStatTable(pParse, iDb, iStatCur, pTab->zName, "tbl"); + } + analyzeOneTable(pParse, pTab, pOnlyIdx, iStatCur,pParse->nMem+1,pParse->nTab); + loadAnalysis(pParse, iDb); +} + +/* +** Generate code for the ANALYZE command. The parser calls this routine +** when it recognizes an ANALYZE command. +** +** ANALYZE -- 1 +** ANALYZE -- 2 +** ANALYZE ?.? -- 3 +** +** Form 1 causes all indices in all attached databases to be analyzed. +** Form 2 analyzes all indices the single database named. +** Form 3 analyzes all indices associated with the named table. +*/ +SQLITE_PRIVATE void sqlite3Analyze(Parse *pParse, Token *pName1, Token *pName2){ + sqlite3 *db = pParse->db; + int iDb; + int i; + char *z, *zDb; + Table *pTab; + Index *pIdx; + Token *pTableName; + Vdbe *v; + + /* Read the database schema. If an error occurs, leave an error message + ** and code in pParse and return NULL. */ + assert( sqlite3BtreeHoldsAllMutexes(pParse->db) ); + if( SQLITE_OK!=sqlite3ReadSchema(pParse) ){ + return; + } + + assert( pName2!=0 || pName1==0 ); + if( pName1==0 ){ + /* Form 1: Analyze everything */ + for(i=0; inDb; i++){ + if( i==1 ) continue; /* Do not analyze the TEMP database */ + analyzeDatabase(pParse, i); + } + }else if( pName2->n==0 && (iDb = sqlite3FindDb(db, pName1))>=0 ){ + /* Analyze the schema named as the argument */ + analyzeDatabase(pParse, iDb); + }else{ + /* Form 3: Analyze the table or index named as an argument */ + iDb = sqlite3TwoPartName(pParse, pName1, pName2, &pTableName); + if( iDb>=0 ){ + zDb = pName2->n ? db->aDb[iDb].zDbSName : 0; + z = sqlite3NameFromToken(db, pTableName); + if( z ){ + if( (pIdx = sqlite3FindIndex(db, z, zDb))!=0 ){ + analyzeTable(pParse, pIdx->pTable, pIdx); + }else if( (pTab = sqlite3LocateTable(pParse, 0, z, zDb))!=0 ){ + analyzeTable(pParse, pTab, 0); + } + sqlite3DbFree(db, z); + } + } + } + if( db->nSqlExec==0 && (v = sqlite3GetVdbe(pParse))!=0 ){ + sqlite3VdbeAddOp0(v, OP_Expire); + } +} + +/* +** Used to pass information from the analyzer reader through to the +** callback routine. +*/ +typedef struct analysisInfo analysisInfo; +struct analysisInfo { + sqlite3 *db; + const char *zDatabase; +}; + +/* +** The first argument points to a nul-terminated string containing a +** list of space separated integers. Read the first nOut of these into +** the array aOut[]. +*/ +static void decodeIntArray( + char *zIntArray, /* String containing int array to decode */ + int nOut, /* Number of slots in aOut[] */ + tRowcnt *aOut, /* Store integers here */ + LogEst *aLog, /* Or, if aOut==0, here */ + Index *pIndex /* Handle extra flags for this index, if not NULL */ +){ + char *z = zIntArray; + int c; + int i; + tRowcnt v; + +#ifdef SQLITE_ENABLE_STAT4 + if( z==0 ) z = ""; +#else + assert( z!=0 ); +#endif + for(i=0; *z && i='0' && c<='9' ){ + v = v*10 + c - '0'; + z++; + } +#ifdef SQLITE_ENABLE_STAT4 + if( aOut ) aOut[i] = v; + if( aLog ) aLog[i] = sqlite3LogEst(v); +#else + assert( aOut==0 ); + UNUSED_PARAMETER(aOut); + assert( aLog!=0 ); + aLog[i] = sqlite3LogEst(v); +#endif + if( *z==' ' ) z++; + } +#ifndef SQLITE_ENABLE_STAT4 + assert( pIndex!=0 ); { +#else + if( pIndex ){ +#endif + pIndex->bUnordered = 0; + pIndex->noSkipScan = 0; + while( z[0] ){ + if( sqlite3_strglob("unordered*", z)==0 ){ + pIndex->bUnordered = 1; + }else if( sqlite3_strglob("sz=[0-9]*", z)==0 ){ + int sz = sqlite3Atoi(z+3); + if( sz<2 ) sz = 2; + pIndex->szIdxRow = sqlite3LogEst(sz); + }else if( sqlite3_strglob("noskipscan*", z)==0 ){ + pIndex->noSkipScan = 1; + } +#ifdef SQLITE_ENABLE_COSTMULT + else if( sqlite3_strglob("costmult=[0-9]*",z)==0 ){ + pIndex->pTable->costMult = sqlite3LogEst(sqlite3Atoi(z+9)); + } +#endif + while( z[0]!=0 && z[0]!=' ' ) z++; + while( z[0]==' ' ) z++; + } + } +} + +/* +** This callback is invoked once for each index when reading the +** sqlite_stat1 table. +** +** argv[0] = name of the table +** argv[1] = name of the index (might be NULL) +** argv[2] = results of analysis - on integer for each column +** +** Entries for which argv[1]==NULL simply record the number of rows in +** the table. +*/ +static int analysisLoader(void *pData, int argc, char **argv, char **NotUsed){ + analysisInfo *pInfo = (analysisInfo*)pData; + Index *pIndex; + Table *pTable; + const char *z; + + assert( argc==3 ); + UNUSED_PARAMETER2(NotUsed, argc); + + if( argv==0 || argv[0]==0 || argv[2]==0 ){ + return 0; + } + pTable = sqlite3FindTable(pInfo->db, argv[0], pInfo->zDatabase); + if( pTable==0 ){ + return 0; + } + if( argv[1]==0 ){ + pIndex = 0; + }else if( sqlite3_stricmp(argv[0],argv[1])==0 ){ + pIndex = sqlite3PrimaryKeyIndex(pTable); + }else{ + pIndex = sqlite3FindIndex(pInfo->db, argv[1], pInfo->zDatabase); + } + z = argv[2]; + + if( pIndex ){ + tRowcnt *aiRowEst = 0; + int nCol = pIndex->nKeyCol+1; +#ifdef SQLITE_ENABLE_STAT4 + /* Index.aiRowEst may already be set here if there are duplicate + ** sqlite_stat1 entries for this index. In that case just clobber + ** the old data with the new instead of allocating a new array. */ + if( pIndex->aiRowEst==0 ){ + pIndex->aiRowEst = (tRowcnt*)sqlite3MallocZero(sizeof(tRowcnt) * nCol); + if( pIndex->aiRowEst==0 ) sqlite3OomFault(pInfo->db); + } + aiRowEst = pIndex->aiRowEst; +#endif + pIndex->bUnordered = 0; + decodeIntArray((char*)z, nCol, aiRowEst, pIndex->aiRowLogEst, pIndex); + pIndex->hasStat1 = 1; + if( pIndex->pPartIdxWhere==0 ){ + pTable->nRowLogEst = pIndex->aiRowLogEst[0]; + pTable->tabFlags |= TF_HasStat1; + } + }else{ + Index fakeIdx; + fakeIdx.szIdxRow = pTable->szTabRow; +#ifdef SQLITE_ENABLE_COSTMULT + fakeIdx.pTable = pTable; +#endif + decodeIntArray((char*)z, 1, 0, &pTable->nRowLogEst, &fakeIdx); + pTable->szTabRow = fakeIdx.szIdxRow; + pTable->tabFlags |= TF_HasStat1; + } + + return 0; +} + +/* +** If the Index.aSample variable is not NULL, delete the aSample[] array +** and its contents. +*/ +SQLITE_PRIVATE void sqlite3DeleteIndexSamples(sqlite3 *db, Index *pIdx){ +#ifdef SQLITE_ENABLE_STAT4 + if( pIdx->aSample ){ + int j; + for(j=0; jnSample; j++){ + IndexSample *p = &pIdx->aSample[j]; + sqlite3DbFree(db, p->p); + } + sqlite3DbFree(db, pIdx->aSample); + } + if( db && db->pnBytesFreed==0 ){ + pIdx->nSample = 0; + pIdx->aSample = 0; + } +#else + UNUSED_PARAMETER(db); + UNUSED_PARAMETER(pIdx); +#endif /* SQLITE_ENABLE_STAT4 */ +} + +#ifdef SQLITE_ENABLE_STAT4 +/* +** Populate the pIdx->aAvgEq[] array based on the samples currently +** stored in pIdx->aSample[]. +*/ +static void initAvgEq(Index *pIdx){ + if( pIdx ){ + IndexSample *aSample = pIdx->aSample; + IndexSample *pFinal = &aSample[pIdx->nSample-1]; + int iCol; + int nCol = 1; + if( pIdx->nSampleCol>1 ){ + /* If this is stat4 data, then calculate aAvgEq[] values for all + ** sample columns except the last. The last is always set to 1, as + ** once the trailing PK fields are considered all index keys are + ** unique. */ + nCol = pIdx->nSampleCol-1; + pIdx->aAvgEq[nCol] = 1; + } + for(iCol=0; iColnSample; + int i; /* Used to iterate through samples */ + tRowcnt sumEq = 0; /* Sum of the nEq values */ + tRowcnt avgEq = 0; + tRowcnt nRow; /* Number of rows in index */ + i64 nSum100 = 0; /* Number of terms contributing to sumEq */ + i64 nDist100; /* Number of distinct values in index */ + + if( !pIdx->aiRowEst || iCol>=pIdx->nKeyCol || pIdx->aiRowEst[iCol+1]==0 ){ + nRow = pFinal->anLt[iCol]; + nDist100 = (i64)100 * pFinal->anDLt[iCol]; + nSample--; + }else{ + nRow = pIdx->aiRowEst[0]; + nDist100 = ((i64)100 * pIdx->aiRowEst[0]) / pIdx->aiRowEst[iCol+1]; + } + pIdx->nRowEst0 = nRow; + + /* Set nSum to the number of distinct (iCol+1) field prefixes that + ** occur in the stat4 table for this index. Set sumEq to the sum of + ** the nEq values for column iCol for the same set (adding the value + ** only once where there exist duplicate prefixes). */ + for(i=0; inSample-1) + || aSample[i].anDLt[iCol]!=aSample[i+1].anDLt[iCol] + ){ + sumEq += aSample[i].anEq[iCol]; + nSum100 += 100; + } + } + + if( nDist100>nSum100 && sumEqaAvgEq[iCol] = avgEq; + } + } +} + +/* +** Look up an index by name. Or, if the name of a WITHOUT ROWID table +** is supplied instead, find the PRIMARY KEY index for that table. +*/ +static Index *findIndexOrPrimaryKey( + sqlite3 *db, + const char *zName, + const char *zDb +){ + Index *pIdx = sqlite3FindIndex(db, zName, zDb); + if( pIdx==0 ){ + Table *pTab = sqlite3FindTable(db, zName, zDb); + if( pTab && !HasRowid(pTab) ) pIdx = sqlite3PrimaryKeyIndex(pTab); + } + return pIdx; +} + +/* +** Load the content from either the sqlite_stat4 +** into the relevant Index.aSample[] arrays. +** +** Arguments zSql1 and zSql2 must point to SQL statements that return +** data equivalent to the following: +** +** zSql1: SELECT idx,count(*) FROM %Q.sqlite_stat4 GROUP BY idx +** zSql2: SELECT idx,neq,nlt,ndlt,sample FROM %Q.sqlite_stat4 +** +** where %Q is replaced with the database name before the SQL is executed. +*/ +static int loadStatTbl( + sqlite3 *db, /* Database handle */ + const char *zSql1, /* SQL statement 1 (see above) */ + const char *zSql2, /* SQL statement 2 (see above) */ + const char *zDb /* Database name (e.g. "main") */ +){ + int rc; /* Result codes from subroutines */ + sqlite3_stmt *pStmt = 0; /* An SQL statement being run */ + char *zSql; /* Text of the SQL statement */ + Index *pPrevIdx = 0; /* Previous index in the loop */ + IndexSample *pSample; /* A slot in pIdx->aSample[] */ + + assert( db->lookaside.bDisable ); + zSql = sqlite3MPrintf(db, zSql1, zDb); + if( !zSql ){ + return SQLITE_NOMEM_BKPT; + } + rc = sqlite3_prepare(db, zSql, -1, &pStmt, 0); + sqlite3DbFree(db, zSql); + if( rc ) return rc; + + while( sqlite3_step(pStmt)==SQLITE_ROW ){ + int nIdxCol = 1; /* Number of columns in stat4 records */ + + char *zIndex; /* Index name */ + Index *pIdx; /* Pointer to the index object */ + int nSample; /* Number of samples */ + int nByte; /* Bytes of space required */ + int i; /* Bytes of space required */ + tRowcnt *pSpace; + + zIndex = (char *)sqlite3_column_text(pStmt, 0); + if( zIndex==0 ) continue; + nSample = sqlite3_column_int(pStmt, 1); + pIdx = findIndexOrPrimaryKey(db, zIndex, zDb); + assert( pIdx==0 || pIdx->nSample==0 ); + if( pIdx==0 ) continue; + assert( !HasRowid(pIdx->pTable) || pIdx->nColumn==pIdx->nKeyCol+1 ); + if( !HasRowid(pIdx->pTable) && IsPrimaryKeyIndex(pIdx) ){ + nIdxCol = pIdx->nKeyCol; + }else{ + nIdxCol = pIdx->nColumn; + } + pIdx->nSampleCol = nIdxCol; + nByte = sizeof(IndexSample) * nSample; + nByte += sizeof(tRowcnt) * nIdxCol * 3 * nSample; + nByte += nIdxCol * sizeof(tRowcnt); /* Space for Index.aAvgEq[] */ + + pIdx->aSample = sqlite3DbMallocZero(db, nByte); + if( pIdx->aSample==0 ){ + sqlite3_finalize(pStmt); + return SQLITE_NOMEM_BKPT; + } + pSpace = (tRowcnt*)&pIdx->aSample[nSample]; + pIdx->aAvgEq = pSpace; pSpace += nIdxCol; + for(i=0; iaSample[i].anEq = pSpace; pSpace += nIdxCol; + pIdx->aSample[i].anLt = pSpace; pSpace += nIdxCol; + pIdx->aSample[i].anDLt = pSpace; pSpace += nIdxCol; + } + assert( ((u8*)pSpace)-nByte==(u8*)(pIdx->aSample) ); + } + rc = sqlite3_finalize(pStmt); + if( rc ) return rc; + + zSql = sqlite3MPrintf(db, zSql2, zDb); + if( !zSql ){ + return SQLITE_NOMEM_BKPT; + } + rc = sqlite3_prepare(db, zSql, -1, &pStmt, 0); + sqlite3DbFree(db, zSql); + if( rc ) return rc; + + while( sqlite3_step(pStmt)==SQLITE_ROW ){ + char *zIndex; /* Index name */ + Index *pIdx; /* Pointer to the index object */ + int nCol = 1; /* Number of columns in index */ + + zIndex = (char *)sqlite3_column_text(pStmt, 0); + if( zIndex==0 ) continue; + pIdx = findIndexOrPrimaryKey(db, zIndex, zDb); + if( pIdx==0 ) continue; + /* This next condition is true if data has already been loaded from + ** the sqlite_stat4 table. */ + nCol = pIdx->nSampleCol; + if( pIdx!=pPrevIdx ){ + initAvgEq(pPrevIdx); + pPrevIdx = pIdx; + } + pSample = &pIdx->aSample[pIdx->nSample]; + decodeIntArray((char*)sqlite3_column_text(pStmt,1),nCol,pSample->anEq,0,0); + decodeIntArray((char*)sqlite3_column_text(pStmt,2),nCol,pSample->anLt,0,0); + decodeIntArray((char*)sqlite3_column_text(pStmt,3),nCol,pSample->anDLt,0,0); + + /* Take a copy of the sample. Add two 0x00 bytes the end of the buffer. + ** This is in case the sample record is corrupted. In that case, the + ** sqlite3VdbeRecordCompare() may read up to two varints past the + ** end of the allocated buffer before it realizes it is dealing with + ** a corrupt record. Adding the two 0x00 bytes prevents this from causing + ** a buffer overread. */ + pSample->n = sqlite3_column_bytes(pStmt, 4); + pSample->p = sqlite3DbMallocZero(db, pSample->n + 2); + if( pSample->p==0 ){ + sqlite3_finalize(pStmt); + return SQLITE_NOMEM_BKPT; + } + if( pSample->n ){ + memcpy(pSample->p, sqlite3_column_blob(pStmt, 4), pSample->n); + } + pIdx->nSample++; + } + rc = sqlite3_finalize(pStmt); + if( rc==SQLITE_OK ) initAvgEq(pPrevIdx); + return rc; +} + +/* +** Load content from the sqlite_stat4 table into +** the Index.aSample[] arrays of all indices. +*/ +static int loadStat4(sqlite3 *db, const char *zDb){ + int rc = SQLITE_OK; /* Result codes from subroutines */ + + assert( db->lookaside.bDisable ); + if( sqlite3FindTable(db, "sqlite_stat4", zDb) ){ + rc = loadStatTbl(db, + "SELECT idx,count(*) FROM %Q.sqlite_stat4 GROUP BY idx", + "SELECT idx,neq,nlt,ndlt,sample FROM %Q.sqlite_stat4", + zDb + ); + } + return rc; +} +#endif /* SQLITE_ENABLE_STAT4 */ + +/* +** Load the content of the sqlite_stat1 and sqlite_stat4 tables. The +** contents of sqlite_stat1 are used to populate the Index.aiRowEst[] +** arrays. The contents of sqlite_stat4 are used to populate the +** Index.aSample[] arrays. +** +** If the sqlite_stat1 table is not present in the database, SQLITE_ERROR +** is returned. In this case, even if SQLITE_ENABLE_STAT4 was defined +** during compilation and the sqlite_stat4 table is present, no data is +** read from it. +** +** If SQLITE_ENABLE_STAT4 was defined during compilation and the +** sqlite_stat4 table is not present in the database, SQLITE_ERROR is +** returned. However, in this case, data is read from the sqlite_stat1 +** table (if it is present) before returning. +** +** If an OOM error occurs, this function always sets db->mallocFailed. +** This means if the caller does not care about other errors, the return +** code may be ignored. +*/ +SQLITE_PRIVATE int sqlite3AnalysisLoad(sqlite3 *db, int iDb){ + analysisInfo sInfo; + HashElem *i; + char *zSql; + int rc = SQLITE_OK; + Schema *pSchema = db->aDb[iDb].pSchema; + + assert( iDb>=0 && iDbnDb ); + assert( db->aDb[iDb].pBt!=0 ); + + /* Clear any prior statistics */ + assert( sqlite3SchemaMutexHeld(db, iDb, 0) ); + for(i=sqliteHashFirst(&pSchema->tblHash); i; i=sqliteHashNext(i)){ + Table *pTab = sqliteHashData(i); + pTab->tabFlags &= ~TF_HasStat1; + } + for(i=sqliteHashFirst(&pSchema->idxHash); i; i=sqliteHashNext(i)){ + Index *pIdx = sqliteHashData(i); + pIdx->hasStat1 = 0; +#ifdef SQLITE_ENABLE_STAT4 + sqlite3DeleteIndexSamples(db, pIdx); + pIdx->aSample = 0; +#endif + } + + /* Load new statistics out of the sqlite_stat1 table */ + sInfo.db = db; + sInfo.zDatabase = db->aDb[iDb].zDbSName; + if( sqlite3FindTable(db, "sqlite_stat1", sInfo.zDatabase)!=0 ){ + zSql = sqlite3MPrintf(db, + "SELECT tbl,idx,stat FROM %Q.sqlite_stat1", sInfo.zDatabase); + if( zSql==0 ){ + rc = SQLITE_NOMEM_BKPT; + }else{ + rc = sqlite3_exec(db, zSql, analysisLoader, &sInfo, 0); + sqlite3DbFree(db, zSql); + } + } + + /* Set appropriate defaults on all indexes not in the sqlite_stat1 table */ + assert( sqlite3SchemaMutexHeld(db, iDb, 0) ); + for(i=sqliteHashFirst(&pSchema->idxHash); i; i=sqliteHashNext(i)){ + Index *pIdx = sqliteHashData(i); + if( !pIdx->hasStat1 ) sqlite3DefaultRowEst(pIdx); + } + + /* Load the statistics from the sqlite_stat4 table. */ +#ifdef SQLITE_ENABLE_STAT4 + if( rc==SQLITE_OK ){ + db->lookaside.bDisable++; + rc = loadStat4(db, sInfo.zDatabase); + db->lookaside.bDisable--; + } + for(i=sqliteHashFirst(&pSchema->idxHash); i; i=sqliteHashNext(i)){ + Index *pIdx = sqliteHashData(i); + sqlite3_free(pIdx->aiRowEst); + pIdx->aiRowEst = 0; + } +#endif + + if( rc==SQLITE_NOMEM ){ + sqlite3OomFault(db); + } + return rc; +} + + +#endif /* SQLITE_OMIT_ANALYZE */ + +/************** End of analyze.c *********************************************/ +/************** Begin file attach.c ******************************************/ +/* +** 2003 April 6 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** This file contains code used to implement the ATTACH and DETACH commands. +*/ +/* #include "sqliteInt.h" */ + +#ifndef SQLITE_OMIT_ATTACH +/* +** Resolve an expression that was part of an ATTACH or DETACH statement. This +** is slightly different from resolving a normal SQL expression, because simple +** identifiers are treated as strings, not possible column names or aliases. +** +** i.e. if the parser sees: +** +** ATTACH DATABASE abc AS def +** +** it treats the two expressions as literal strings 'abc' and 'def' instead of +** looking for columns of the same name. +** +** This only applies to the root node of pExpr, so the statement: +** +** ATTACH DATABASE abc||def AS 'db2' +** +** will fail because neither abc or def can be resolved. +*/ +static int resolveAttachExpr(NameContext *pName, Expr *pExpr) +{ + int rc = SQLITE_OK; + if( pExpr ){ + if( pExpr->op!=TK_ID ){ + rc = sqlite3ResolveExprNames(pName, pExpr); + }else{ + pExpr->op = TK_STRING; + } + } + return rc; +} + +/* +** An SQL user-function registered to do the work of an ATTACH statement. The +** three arguments to the function come directly from an attach statement: +** +** ATTACH DATABASE x AS y KEY z +** +** SELECT sqlite_attach(x, y, z) +** +** If the optional "KEY z" syntax is omitted, an SQL NULL is passed as the +** third argument. +** +** If the db->init.reopenMemdb flags is set, then instead of attaching a +** new database, close the database on db->init.iDb and reopen it as an +** empty MemDB. +*/ +static void attachFunc( + sqlite3_context *context, + int NotUsed, + sqlite3_value **argv +){ + int i; + int rc = 0; + sqlite3 *db = sqlite3_context_db_handle(context); + const char *zName; + const char *zFile; + char *zPath = 0; + char *zErr = 0; + unsigned int flags; + Db *aNew; /* New array of Db pointers */ + Db *pNew; /* Db object for the newly attached database */ + char *zErrDyn = 0; + sqlite3_vfs *pVfs; + + UNUSED_PARAMETER(NotUsed); + zFile = (const char *)sqlite3_value_text(argv[0]); + zName = (const char *)sqlite3_value_text(argv[1]); + if( zFile==0 ) zFile = ""; + if( zName==0 ) zName = ""; + +#ifdef SQLITE_ENABLE_DESERIALIZE +# define REOPEN_AS_MEMDB(db) (db->init.reopenMemdb) +#else +# define REOPEN_AS_MEMDB(db) (0) +#endif + + if( REOPEN_AS_MEMDB(db) ){ + /* This is not a real ATTACH. Instead, this routine is being called + ** from sqlite3_deserialize() to close database db->init.iDb and + ** reopen it as a MemDB */ + pVfs = sqlite3_vfs_find("memdb"); + if( pVfs==0 ) return; + pNew = &db->aDb[db->init.iDb]; + if( pNew->pBt ) sqlite3BtreeClose(pNew->pBt); + pNew->pBt = 0; + pNew->pSchema = 0; + rc = sqlite3BtreeOpen(pVfs, "x\0", db, &pNew->pBt, 0, SQLITE_OPEN_MAIN_DB); + }else{ + /* This is a real ATTACH + ** + ** Check for the following errors: + ** + ** * Too many attached databases, + ** * Transaction currently open + ** * Specified database name already being used. + */ + if( db->nDb>=db->aLimit[SQLITE_LIMIT_ATTACHED]+2 ){ + zErrDyn = sqlite3MPrintf(db, "too many attached databases - max %d", + db->aLimit[SQLITE_LIMIT_ATTACHED] + ); + goto attach_error; + } + for(i=0; inDb; i++){ + char *z = db->aDb[i].zDbSName; + assert( z && zName ); + if( sqlite3StrICmp(z, zName)==0 ){ + zErrDyn = sqlite3MPrintf(db, "database %s is already in use", zName); + goto attach_error; + } + } + + /* Allocate the new entry in the db->aDb[] array and initialize the schema + ** hash tables. + */ + if( db->aDb==db->aDbStatic ){ + aNew = sqlite3DbMallocRawNN(db, sizeof(db->aDb[0])*3 ); + if( aNew==0 ) return; + memcpy(aNew, db->aDb, sizeof(db->aDb[0])*2); + }else{ + aNew = sqlite3DbRealloc(db, db->aDb, sizeof(db->aDb[0])*(db->nDb+1) ); + if( aNew==0 ) return; + } + db->aDb = aNew; + pNew = &db->aDb[db->nDb]; + memset(pNew, 0, sizeof(*pNew)); + + /* Open the database file. If the btree is successfully opened, use + ** it to obtain the database schema. At this point the schema may + ** or may not be initialized. + */ + flags = db->openFlags; + rc = sqlite3ParseUri(db->pVfs->zName, zFile, &flags, &pVfs, &zPath, &zErr); + if( rc!=SQLITE_OK ){ + if( rc==SQLITE_NOMEM ) sqlite3OomFault(db); + sqlite3_result_error(context, zErr, -1); + sqlite3_free(zErr); + return; + } + assert( pVfs ); + flags |= SQLITE_OPEN_MAIN_DB; + rc = sqlite3BtreeOpen(pVfs, zPath, db, &pNew->pBt, 0, flags); + db->nDb++; + pNew->zDbSName = sqlite3DbStrDup(db, zName); + } + db->noSharedCache = 0; + if( rc==SQLITE_CONSTRAINT ){ + rc = SQLITE_ERROR; + zErrDyn = sqlite3MPrintf(db, "database is already attached"); + }else if( rc==SQLITE_OK ){ + Pager *pPager; + pNew->pSchema = sqlite3SchemaGet(db, pNew->pBt); + if( !pNew->pSchema ){ + rc = SQLITE_NOMEM_BKPT; + }else if( pNew->pSchema->file_format && pNew->pSchema->enc!=ENC(db) ){ + zErrDyn = sqlite3MPrintf(db, + "attached databases must use the same text encoding as main database"); + rc = SQLITE_ERROR; + } + sqlite3BtreeEnter(pNew->pBt); + pPager = sqlite3BtreePager(pNew->pBt); + sqlite3PagerLockingMode(pPager, db->dfltLockMode); + sqlite3BtreeSecureDelete(pNew->pBt, + sqlite3BtreeSecureDelete(db->aDb[0].pBt,-1) ); +#ifndef SQLITE_OMIT_PAGER_PRAGMAS + sqlite3BtreeSetPagerFlags(pNew->pBt, + PAGER_SYNCHRONOUS_FULL | (db->flags & PAGER_FLAGS_MASK)); +#endif + sqlite3BtreeLeave(pNew->pBt); + } + pNew->safety_level = SQLITE_DEFAULT_SYNCHRONOUS+1; + if( rc==SQLITE_OK && pNew->zDbSName==0 ){ + rc = SQLITE_NOMEM_BKPT; + } + + +#ifdef SQLITE_HAS_CODEC + if( rc==SQLITE_OK ){ + extern int sqlite3CodecAttach(sqlite3*, int, const void*, int); + extern void sqlite3CodecGetKey(sqlite3*, int, void**, int*); + int nKey; + char *zKey; + int t = sqlite3_value_type(argv[2]); + switch( t ){ + case SQLITE_INTEGER: + case SQLITE_FLOAT: + zErrDyn = sqlite3DbStrDup(db, "Invalid key value"); + rc = SQLITE_ERROR; + break; + + case SQLITE_TEXT: + case SQLITE_BLOB: + nKey = sqlite3_value_bytes(argv[2]); + zKey = (char *)sqlite3_value_blob(argv[2]); + rc = sqlite3CodecAttach(db, db->nDb-1, zKey, nKey); + break; + + case SQLITE_NULL: + /* No key specified. Use the key from URI filename, or if none, + ** use the key from the main database. */ + if( sqlite3CodecQueryParameters(db, zName, zPath)==0 ){ + sqlite3CodecGetKey(db, 0, (void**)&zKey, &nKey); + if( nKey || sqlite3BtreeGetOptimalReserve(db->aDb[0].pBt)>0 ){ + rc = sqlite3CodecAttach(db, db->nDb-1, zKey, nKey); + } + } + break; + } + } +#endif + sqlite3_free( zPath ); + + /* If the file was opened successfully, read the schema for the new database. + ** If this fails, or if opening the file failed, then close the file and + ** remove the entry from the db->aDb[] array. i.e. put everything back the + ** way we found it. + */ + if( rc==SQLITE_OK ){ + sqlite3BtreeEnterAll(db); + db->init.iDb = 0; + db->mDbFlags &= ~(DBFLAG_SchemaKnownOk); + if( !REOPEN_AS_MEMDB(db) ){ + rc = sqlite3Init(db, &zErrDyn); + } + sqlite3BtreeLeaveAll(db); + assert( zErrDyn==0 || rc!=SQLITE_OK ); + } +#ifdef SQLITE_USER_AUTHENTICATION + if( rc==SQLITE_OK && !REOPEN_AS_MEMDB(db) ){ + u8 newAuth = 0; + rc = sqlite3UserAuthCheckLogin(db, zName, &newAuth); + if( newAuthauth.authLevel ){ + rc = SQLITE_AUTH_USER; + } + } +#endif + if( rc ){ + if( !REOPEN_AS_MEMDB(db) ){ + int iDb = db->nDb - 1; + assert( iDb>=2 ); + if( db->aDb[iDb].pBt ){ + sqlite3BtreeClose(db->aDb[iDb].pBt); + db->aDb[iDb].pBt = 0; + db->aDb[iDb].pSchema = 0; + } + sqlite3ResetAllSchemasOfConnection(db); + db->nDb = iDb; + if( rc==SQLITE_NOMEM || rc==SQLITE_IOERR_NOMEM ){ + sqlite3OomFault(db); + sqlite3DbFree(db, zErrDyn); + zErrDyn = sqlite3MPrintf(db, "out of memory"); + }else if( zErrDyn==0 ){ + zErrDyn = sqlite3MPrintf(db, "unable to open database: %s", zFile); + } + } + goto attach_error; + } + + return; + +attach_error: + /* Return an error if we get here */ + if( zErrDyn ){ + sqlite3_result_error(context, zErrDyn, -1); + sqlite3DbFree(db, zErrDyn); + } + if( rc ) sqlite3_result_error_code(context, rc); +} + +/* +** An SQL user-function registered to do the work of an DETACH statement. The +** three arguments to the function come directly from a detach statement: +** +** DETACH DATABASE x +** +** SELECT sqlite_detach(x) +*/ +static void detachFunc( + sqlite3_context *context, + int NotUsed, + sqlite3_value **argv +){ + const char *zName = (const char *)sqlite3_value_text(argv[0]); + sqlite3 *db = sqlite3_context_db_handle(context); + int i; + Db *pDb = 0; + HashElem *pEntry; + char zErr[128]; + + UNUSED_PARAMETER(NotUsed); + + if( zName==0 ) zName = ""; + for(i=0; inDb; i++){ + pDb = &db->aDb[i]; + if( pDb->pBt==0 ) continue; + if( sqlite3StrICmp(pDb->zDbSName, zName)==0 ) break; + } + + if( i>=db->nDb ){ + sqlite3_snprintf(sizeof(zErr),zErr, "no such database: %s", zName); + goto detach_error; + } + if( i<2 ){ + sqlite3_snprintf(sizeof(zErr),zErr, "cannot detach database %s", zName); + goto detach_error; + } + if( sqlite3BtreeIsInReadTrans(pDb->pBt) || sqlite3BtreeIsInBackup(pDb->pBt) ){ + sqlite3_snprintf(sizeof(zErr),zErr, "database %s is locked", zName); + goto detach_error; + } + + /* If any TEMP triggers reference the schema being detached, move those + ** triggers to reference the TEMP schema itself. */ + assert( db->aDb[1].pSchema ); + pEntry = sqliteHashFirst(&db->aDb[1].pSchema->trigHash); + while( pEntry ){ + Trigger *pTrig = (Trigger*)sqliteHashData(pEntry); + if( pTrig->pTabSchema==pDb->pSchema ){ + pTrig->pTabSchema = pTrig->pSchema; + } + pEntry = sqliteHashNext(pEntry); + } + + sqlite3BtreeClose(pDb->pBt); + pDb->pBt = 0; + pDb->pSchema = 0; + sqlite3CollapseDatabaseArray(db); + return; + +detach_error: + sqlite3_result_error(context, zErr, -1); +} + +/* +** This procedure generates VDBE code for a single invocation of either the +** sqlite_detach() or sqlite_attach() SQL user functions. +*/ +static void codeAttach( + Parse *pParse, /* The parser context */ + int type, /* Either SQLITE_ATTACH or SQLITE_DETACH */ + FuncDef const *pFunc,/* FuncDef wrapper for detachFunc() or attachFunc() */ + Expr *pAuthArg, /* Expression to pass to authorization callback */ + Expr *pFilename, /* Name of database file */ + Expr *pDbname, /* Name of the database to use internally */ + Expr *pKey /* Database key for encryption extension */ +){ + int rc; + NameContext sName; + Vdbe *v; + sqlite3* db = pParse->db; + int regArgs; + + if( pParse->nErr ) goto attach_end; + memset(&sName, 0, sizeof(NameContext)); + sName.pParse = pParse; + + if( + SQLITE_OK!=(rc = resolveAttachExpr(&sName, pFilename)) || + SQLITE_OK!=(rc = resolveAttachExpr(&sName, pDbname)) || + SQLITE_OK!=(rc = resolveAttachExpr(&sName, pKey)) + ){ + goto attach_end; + } + +#ifndef SQLITE_OMIT_AUTHORIZATION + if( pAuthArg ){ + char *zAuthArg; + if( pAuthArg->op==TK_STRING ){ + zAuthArg = pAuthArg->u.zToken; + }else{ + zAuthArg = 0; + } + rc = sqlite3AuthCheck(pParse, type, zAuthArg, 0, 0); + if(rc!=SQLITE_OK ){ + goto attach_end; + } + } +#endif /* SQLITE_OMIT_AUTHORIZATION */ + + + v = sqlite3GetVdbe(pParse); + regArgs = sqlite3GetTempRange(pParse, 4); + sqlite3ExprCode(pParse, pFilename, regArgs); + sqlite3ExprCode(pParse, pDbname, regArgs+1); + sqlite3ExprCode(pParse, pKey, regArgs+2); + + assert( v || db->mallocFailed ); + if( v ){ + sqlite3VdbeAddOp4(v, OP_Function0, 0, regArgs+3-pFunc->nArg, regArgs+3, + (char *)pFunc, P4_FUNCDEF); + assert( pFunc->nArg==-1 || (pFunc->nArg&0xff)==pFunc->nArg ); + sqlite3VdbeChangeP5(v, (u8)(pFunc->nArg)); + + /* Code an OP_Expire. For an ATTACH statement, set P1 to true (expire this + ** statement only). For DETACH, set it to false (expire all existing + ** statements). + */ + sqlite3VdbeAddOp1(v, OP_Expire, (type==SQLITE_ATTACH)); + } + +attach_end: + sqlite3ExprDelete(db, pFilename); + sqlite3ExprDelete(db, pDbname); + sqlite3ExprDelete(db, pKey); +} + +/* +** Called by the parser to compile a DETACH statement. +** +** DETACH pDbname +*/ +SQLITE_PRIVATE void sqlite3Detach(Parse *pParse, Expr *pDbname){ + static const FuncDef detach_func = { + 1, /* nArg */ + SQLITE_UTF8, /* funcFlags */ + 0, /* pUserData */ + 0, /* pNext */ + detachFunc, /* xSFunc */ + 0, /* xFinalize */ + 0, 0, /* xValue, xInverse */ + "sqlite_detach", /* zName */ + {0} + }; + codeAttach(pParse, SQLITE_DETACH, &detach_func, pDbname, 0, 0, pDbname); +} + +/* +** Called by the parser to compile an ATTACH statement. +** +** ATTACH p AS pDbname KEY pKey +*/ +SQLITE_PRIVATE void sqlite3Attach(Parse *pParse, Expr *p, Expr *pDbname, Expr *pKey){ + static const FuncDef attach_func = { + 3, /* nArg */ + SQLITE_UTF8, /* funcFlags */ + 0, /* pUserData */ + 0, /* pNext */ + attachFunc, /* xSFunc */ + 0, /* xFinalize */ + 0, 0, /* xValue, xInverse */ + "sqlite_attach", /* zName */ + {0} + }; + codeAttach(pParse, SQLITE_ATTACH, &attach_func, p, p, pDbname, pKey); +} +#endif /* SQLITE_OMIT_ATTACH */ + +/* +** Initialize a DbFixer structure. This routine must be called prior +** to passing the structure to one of the sqliteFixAAAA() routines below. +*/ +SQLITE_PRIVATE void sqlite3FixInit( + DbFixer *pFix, /* The fixer to be initialized */ + Parse *pParse, /* Error messages will be written here */ + int iDb, /* This is the database that must be used */ + const char *zType, /* "view", "trigger", or "index" */ + const Token *pName /* Name of the view, trigger, or index */ +){ + sqlite3 *db; + + db = pParse->db; + assert( db->nDb>iDb ); + pFix->pParse = pParse; + pFix->zDb = db->aDb[iDb].zDbSName; + pFix->pSchema = db->aDb[iDb].pSchema; + pFix->zType = zType; + pFix->pName = pName; + pFix->bVarOnly = (iDb==1); +} + +/* +** The following set of routines walk through the parse tree and assign +** a specific database to all table references where the database name +** was left unspecified in the original SQL statement. The pFix structure +** must have been initialized by a prior call to sqlite3FixInit(). +** +** These routines are used to make sure that an index, trigger, or +** view in one database does not refer to objects in a different database. +** (Exception: indices, triggers, and views in the TEMP database are +** allowed to refer to anything.) If a reference is explicitly made +** to an object in a different database, an error message is added to +** pParse->zErrMsg and these routines return non-zero. If everything +** checks out, these routines return 0. +*/ +SQLITE_PRIVATE int sqlite3FixSrcList( + DbFixer *pFix, /* Context of the fixation */ + SrcList *pList /* The Source list to check and modify */ +){ + int i; + const char *zDb; + struct SrcList_item *pItem; + + if( NEVER(pList==0) ) return 0; + zDb = pFix->zDb; + for(i=0, pItem=pList->a; inSrc; i++, pItem++){ + if( pFix->bVarOnly==0 ){ + if( pItem->zDatabase && sqlite3StrICmp(pItem->zDatabase, zDb) ){ + sqlite3ErrorMsg(pFix->pParse, + "%s %T cannot reference objects in database %s", + pFix->zType, pFix->pName, pItem->zDatabase); + return 1; + } + sqlite3DbFree(pFix->pParse->db, pItem->zDatabase); + pItem->zDatabase = 0; + pItem->pSchema = pFix->pSchema; + } +#if !defined(SQLITE_OMIT_VIEW) || !defined(SQLITE_OMIT_TRIGGER) + if( sqlite3FixSelect(pFix, pItem->pSelect) ) return 1; + if( sqlite3FixExpr(pFix, pItem->pOn) ) return 1; +#endif + if( pItem->fg.isTabFunc && sqlite3FixExprList(pFix, pItem->u1.pFuncArg) ){ + return 1; + } + } + return 0; +} +#if !defined(SQLITE_OMIT_VIEW) || !defined(SQLITE_OMIT_TRIGGER) +SQLITE_PRIVATE int sqlite3FixSelect( + DbFixer *pFix, /* Context of the fixation */ + Select *pSelect /* The SELECT statement to be fixed to one database */ +){ + while( pSelect ){ + if( sqlite3FixExprList(pFix, pSelect->pEList) ){ + return 1; + } + if( sqlite3FixSrcList(pFix, pSelect->pSrc) ){ + return 1; + } + if( sqlite3FixExpr(pFix, pSelect->pWhere) ){ + return 1; + } + if( sqlite3FixExprList(pFix, pSelect->pGroupBy) ){ + return 1; + } + if( sqlite3FixExpr(pFix, pSelect->pHaving) ){ + return 1; + } + if( sqlite3FixExprList(pFix, pSelect->pOrderBy) ){ + return 1; + } + if( sqlite3FixExpr(pFix, pSelect->pLimit) ){ + return 1; + } + if( pSelect->pWith ){ + int i; + for(i=0; ipWith->nCte; i++){ + if( sqlite3FixSelect(pFix, pSelect->pWith->a[i].pSelect) ){ + return 1; + } + } + } + pSelect = pSelect->pPrior; + } + return 0; +} +SQLITE_PRIVATE int sqlite3FixExpr( + DbFixer *pFix, /* Context of the fixation */ + Expr *pExpr /* The expression to be fixed to one database */ +){ + while( pExpr ){ + ExprSetProperty(pExpr, EP_Indirect); + if( pExpr->op==TK_VARIABLE ){ + if( pFix->pParse->db->init.busy ){ + pExpr->op = TK_NULL; + }else{ + sqlite3ErrorMsg(pFix->pParse, "%s cannot use variables", pFix->zType); + return 1; + } + } + if( ExprHasProperty(pExpr, EP_TokenOnly|EP_Leaf) ) break; + if( ExprHasProperty(pExpr, EP_xIsSelect) ){ + if( sqlite3FixSelect(pFix, pExpr->x.pSelect) ) return 1; + }else{ + if( sqlite3FixExprList(pFix, pExpr->x.pList) ) return 1; + } + if( sqlite3FixExpr(pFix, pExpr->pRight) ){ + return 1; + } + pExpr = pExpr->pLeft; + } + return 0; +} +SQLITE_PRIVATE int sqlite3FixExprList( + DbFixer *pFix, /* Context of the fixation */ + ExprList *pList /* The expression to be fixed to one database */ +){ + int i; + struct ExprList_item *pItem; + if( pList==0 ) return 0; + for(i=0, pItem=pList->a; inExpr; i++, pItem++){ + if( sqlite3FixExpr(pFix, pItem->pExpr) ){ + return 1; + } + } + return 0; +} +#endif + +#ifndef SQLITE_OMIT_TRIGGER +SQLITE_PRIVATE int sqlite3FixTriggerStep( + DbFixer *pFix, /* Context of the fixation */ + TriggerStep *pStep /* The trigger step be fixed to one database */ +){ + while( pStep ){ + if( sqlite3FixSelect(pFix, pStep->pSelect) ){ + return 1; + } + if( sqlite3FixExpr(pFix, pStep->pWhere) ){ + return 1; + } + if( sqlite3FixExprList(pFix, pStep->pExprList) ){ + return 1; + } +#ifndef SQLITE_OMIT_UPSERT + if( pStep->pUpsert ){ + Upsert *pUp = pStep->pUpsert; + if( sqlite3FixExprList(pFix, pUp->pUpsertTarget) + || sqlite3FixExpr(pFix, pUp->pUpsertTargetWhere) + || sqlite3FixExprList(pFix, pUp->pUpsertSet) + || sqlite3FixExpr(pFix, pUp->pUpsertWhere) + ){ + return 1; + } + } +#endif + pStep = pStep->pNext; + } + return 0; +} +#endif + +/************** End of attach.c **********************************************/ +/************** Begin file auth.c ********************************************/ +/* +** 2003 January 11 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** This file contains code used to implement the sqlite3_set_authorizer() +** API. This facility is an optional feature of the library. Embedded +** systems that do not need this facility may omit it by recompiling +** the library with -DSQLITE_OMIT_AUTHORIZATION=1 +*/ +/* #include "sqliteInt.h" */ + +/* +** All of the code in this file may be omitted by defining a single +** macro. +*/ +#ifndef SQLITE_OMIT_AUTHORIZATION + +/* +** Set or clear the access authorization function. +** +** The access authorization function is be called during the compilation +** phase to verify that the user has read and/or write access permission on +** various fields of the database. The first argument to the auth function +** is a copy of the 3rd argument to this routine. The second argument +** to the auth function is one of these constants: +** +** SQLITE_CREATE_INDEX +** SQLITE_CREATE_TABLE +** SQLITE_CREATE_TEMP_INDEX +** SQLITE_CREATE_TEMP_TABLE +** SQLITE_CREATE_TEMP_TRIGGER +** SQLITE_CREATE_TEMP_VIEW +** SQLITE_CREATE_TRIGGER +** SQLITE_CREATE_VIEW +** SQLITE_DELETE +** SQLITE_DROP_INDEX +** SQLITE_DROP_TABLE +** SQLITE_DROP_TEMP_INDEX +** SQLITE_DROP_TEMP_TABLE +** SQLITE_DROP_TEMP_TRIGGER +** SQLITE_DROP_TEMP_VIEW +** SQLITE_DROP_TRIGGER +** SQLITE_DROP_VIEW +** SQLITE_INSERT +** SQLITE_PRAGMA +** SQLITE_READ +** SQLITE_SELECT +** SQLITE_TRANSACTION +** SQLITE_UPDATE +** +** The third and fourth arguments to the auth function are the name of +** the table and the column that are being accessed. The auth function +** should return either SQLITE_OK, SQLITE_DENY, or SQLITE_IGNORE. If +** SQLITE_OK is returned, it means that access is allowed. SQLITE_DENY +** means that the SQL statement will never-run - the sqlite3_exec() call +** will return with an error. SQLITE_IGNORE means that the SQL statement +** should run but attempts to read the specified column will return NULL +** and attempts to write the column will be ignored. +** +** Setting the auth function to NULL disables this hook. The default +** setting of the auth function is NULL. +*/ +SQLITE_API int sqlite3_set_authorizer( + sqlite3 *db, + int (*xAuth)(void*,int,const char*,const char*,const char*,const char*), + void *pArg +){ +#ifdef SQLITE_ENABLE_API_ARMOR + if( !sqlite3SafetyCheckOk(db) ) return SQLITE_MISUSE_BKPT; +#endif + sqlite3_mutex_enter(db->mutex); + db->xAuth = (sqlite3_xauth)xAuth; + db->pAuthArg = pArg; + if( db->xAuth ) sqlite3ExpirePreparedStatements(db, 1); + sqlite3_mutex_leave(db->mutex); + return SQLITE_OK; +} + +/* +** Write an error message into pParse->zErrMsg that explains that the +** user-supplied authorization function returned an illegal value. +*/ +static void sqliteAuthBadReturnCode(Parse *pParse){ + sqlite3ErrorMsg(pParse, "authorizer malfunction"); + pParse->rc = SQLITE_ERROR; +} + +/* +** Invoke the authorization callback for permission to read column zCol from +** table zTab in database zDb. This function assumes that an authorization +** callback has been registered (i.e. that sqlite3.xAuth is not NULL). +** +** If SQLITE_IGNORE is returned and pExpr is not NULL, then pExpr is changed +** to an SQL NULL expression. Otherwise, if pExpr is NULL, then SQLITE_IGNORE +** is treated as SQLITE_DENY. In this case an error is left in pParse. +*/ +SQLITE_PRIVATE int sqlite3AuthReadCol( + Parse *pParse, /* The parser context */ + const char *zTab, /* Table name */ + const char *zCol, /* Column name */ + int iDb /* Index of containing database. */ +){ + sqlite3 *db = pParse->db; /* Database handle */ + char *zDb = db->aDb[iDb].zDbSName; /* Schema name of attached database */ + int rc; /* Auth callback return code */ + + if( db->init.busy ) return SQLITE_OK; + rc = db->xAuth(db->pAuthArg, SQLITE_READ, zTab,zCol,zDb,pParse->zAuthContext +#ifdef SQLITE_USER_AUTHENTICATION + ,db->auth.zAuthUser +#endif + ); + if( rc==SQLITE_DENY ){ + char *z = sqlite3_mprintf("%s.%s", zTab, zCol); + if( db->nDb>2 || iDb!=0 ) z = sqlite3_mprintf("%s.%z", zDb, z); + sqlite3ErrorMsg(pParse, "access to %z is prohibited", z); + pParse->rc = SQLITE_AUTH; + }else if( rc!=SQLITE_IGNORE && rc!=SQLITE_OK ){ + sqliteAuthBadReturnCode(pParse); + } + return rc; +} + +/* +** The pExpr should be a TK_COLUMN expression. The table referred to +** is in pTabList or else it is the NEW or OLD table of a trigger. +** Check to see if it is OK to read this particular column. +** +** If the auth function returns SQLITE_IGNORE, change the TK_COLUMN +** instruction into a TK_NULL. If the auth function returns SQLITE_DENY, +** then generate an error. +*/ +SQLITE_PRIVATE void sqlite3AuthRead( + Parse *pParse, /* The parser context */ + Expr *pExpr, /* The expression to check authorization on */ + Schema *pSchema, /* The schema of the expression */ + SrcList *pTabList /* All table that pExpr might refer to */ +){ + sqlite3 *db = pParse->db; + Table *pTab = 0; /* The table being read */ + const char *zCol; /* Name of the column of the table */ + int iSrc; /* Index in pTabList->a[] of table being read */ + int iDb; /* The index of the database the expression refers to */ + int iCol; /* Index of column in table */ + + assert( pExpr->op==TK_COLUMN || pExpr->op==TK_TRIGGER ); + assert( !IN_RENAME_OBJECT || db->xAuth==0 ); + if( db->xAuth==0 ) return; + iDb = sqlite3SchemaToIndex(pParse->db, pSchema); + if( iDb<0 ){ + /* An attempt to read a column out of a subquery or other + ** temporary table. */ + return; + } + + if( pExpr->op==TK_TRIGGER ){ + pTab = pParse->pTriggerTab; + }else{ + assert( pTabList ); + for(iSrc=0; ALWAYS(iSrcnSrc); iSrc++){ + if( pExpr->iTable==pTabList->a[iSrc].iCursor ){ + pTab = pTabList->a[iSrc].pTab; + break; + } + } + } + iCol = pExpr->iColumn; + if( NEVER(pTab==0) ) return; + + if( iCol>=0 ){ + assert( iColnCol ); + zCol = pTab->aCol[iCol].zName; + }else if( pTab->iPKey>=0 ){ + assert( pTab->iPKeynCol ); + zCol = pTab->aCol[pTab->iPKey].zName; + }else{ + zCol = "ROWID"; + } + assert( iDb>=0 && iDbnDb ); + if( SQLITE_IGNORE==sqlite3AuthReadCol(pParse, pTab->zName, zCol, iDb) ){ + pExpr->op = TK_NULL; + } +} + +/* +** Do an authorization check using the code and arguments given. Return +** either SQLITE_OK (zero) or SQLITE_IGNORE or SQLITE_DENY. If SQLITE_DENY +** is returned, then the error count and error message in pParse are +** modified appropriately. +*/ +SQLITE_PRIVATE int sqlite3AuthCheck( + Parse *pParse, + int code, + const char *zArg1, + const char *zArg2, + const char *zArg3 +){ + sqlite3 *db = pParse->db; + int rc; + + /* Don't do any authorization checks if the database is initialising + ** or if the parser is being invoked from within sqlite3_declare_vtab. + */ + assert( !IN_RENAME_OBJECT || db->xAuth==0 ); + if( db->init.busy || IN_SPECIAL_PARSE ){ + return SQLITE_OK; + } + + if( db->xAuth==0 ){ + return SQLITE_OK; + } + + /* EVIDENCE-OF: R-43249-19882 The third through sixth parameters to the + ** callback are either NULL pointers or zero-terminated strings that + ** contain additional details about the action to be authorized. + ** + ** The following testcase() macros show that any of the 3rd through 6th + ** parameters can be either NULL or a string. */ + testcase( zArg1==0 ); + testcase( zArg2==0 ); + testcase( zArg3==0 ); + testcase( pParse->zAuthContext==0 ); + + rc = db->xAuth(db->pAuthArg, code, zArg1, zArg2, zArg3, pParse->zAuthContext +#ifdef SQLITE_USER_AUTHENTICATION + ,db->auth.zAuthUser +#endif + ); + if( rc==SQLITE_DENY ){ + sqlite3ErrorMsg(pParse, "not authorized"); + pParse->rc = SQLITE_AUTH; + }else if( rc!=SQLITE_OK && rc!=SQLITE_IGNORE ){ + rc = SQLITE_DENY; + sqliteAuthBadReturnCode(pParse); + } + return rc; +} + +/* +** Push an authorization context. After this routine is called, the +** zArg3 argument to authorization callbacks will be zContext until +** popped. Or if pParse==0, this routine is a no-op. +*/ +SQLITE_PRIVATE void sqlite3AuthContextPush( + Parse *pParse, + AuthContext *pContext, + const char *zContext +){ + assert( pParse ); + pContext->pParse = pParse; + pContext->zAuthContext = pParse->zAuthContext; + pParse->zAuthContext = zContext; +} + +/* +** Pop an authorization context that was previously pushed +** by sqlite3AuthContextPush +*/ +SQLITE_PRIVATE void sqlite3AuthContextPop(AuthContext *pContext){ + if( pContext->pParse ){ + pContext->pParse->zAuthContext = pContext->zAuthContext; + pContext->pParse = 0; + } +} + +#endif /* SQLITE_OMIT_AUTHORIZATION */ + +/************** End of auth.c ************************************************/ +/************** Begin file build.c *******************************************/ +/* +** 2001 September 15 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** This file contains C code routines that are called by the SQLite parser +** when syntax rules are reduced. The routines in this file handle the +** following kinds of SQL syntax: +** +** CREATE TABLE +** DROP TABLE +** CREATE INDEX +** DROP INDEX +** creating ID lists +** BEGIN TRANSACTION +** COMMIT +** ROLLBACK +*/ +/* #include "sqliteInt.h" */ + +#ifndef SQLITE_OMIT_SHARED_CACHE +/* +** The TableLock structure is only used by the sqlite3TableLock() and +** codeTableLocks() functions. +*/ +struct TableLock { + int iDb; /* The database containing the table to be locked */ + int iTab; /* The root page of the table to be locked */ + u8 isWriteLock; /* True for write lock. False for a read lock */ + const char *zLockName; /* Name of the table */ +}; + +/* +** Record the fact that we want to lock a table at run-time. +** +** The table to be locked has root page iTab and is found in database iDb. +** A read or a write lock can be taken depending on isWritelock. +** +** This routine just records the fact that the lock is desired. The +** code to make the lock occur is generated by a later call to +** codeTableLocks() which occurs during sqlite3FinishCoding(). +*/ +SQLITE_PRIVATE void sqlite3TableLock( + Parse *pParse, /* Parsing context */ + int iDb, /* Index of the database containing the table to lock */ + int iTab, /* Root page number of the table to be locked */ + u8 isWriteLock, /* True for a write lock */ + const char *zName /* Name of the table to be locked */ +){ + Parse *pToplevel = sqlite3ParseToplevel(pParse); + int i; + int nBytes; + TableLock *p; + assert( iDb>=0 ); + + if( iDb==1 ) return; + if( !sqlite3BtreeSharable(pParse->db->aDb[iDb].pBt) ) return; + for(i=0; inTableLock; i++){ + p = &pToplevel->aTableLock[i]; + if( p->iDb==iDb && p->iTab==iTab ){ + p->isWriteLock = (p->isWriteLock || isWriteLock); + return; + } + } + + nBytes = sizeof(TableLock) * (pToplevel->nTableLock+1); + pToplevel->aTableLock = + sqlite3DbReallocOrFree(pToplevel->db, pToplevel->aTableLock, nBytes); + if( pToplevel->aTableLock ){ + p = &pToplevel->aTableLock[pToplevel->nTableLock++]; + p->iDb = iDb; + p->iTab = iTab; + p->isWriteLock = isWriteLock; + p->zLockName = zName; + }else{ + pToplevel->nTableLock = 0; + sqlite3OomFault(pToplevel->db); + } +} + +/* +** Code an OP_TableLock instruction for each table locked by the +** statement (configured by calls to sqlite3TableLock()). +*/ +static void codeTableLocks(Parse *pParse){ + int i; + Vdbe *pVdbe; + + pVdbe = sqlite3GetVdbe(pParse); + assert( pVdbe!=0 ); /* sqlite3GetVdbe cannot fail: VDBE already allocated */ + + for(i=0; inTableLock; i++){ + TableLock *p = &pParse->aTableLock[i]; + int p1 = p->iDb; + sqlite3VdbeAddOp4(pVdbe, OP_TableLock, p1, p->iTab, p->isWriteLock, + p->zLockName, P4_STATIC); + } +} +#else + #define codeTableLocks(x) +#endif + +/* +** Return TRUE if the given yDbMask object is empty - if it contains no +** 1 bits. This routine is used by the DbMaskAllZero() and DbMaskNotZero() +** macros when SQLITE_MAX_ATTACHED is greater than 30. +*/ +#if SQLITE_MAX_ATTACHED>30 +SQLITE_PRIVATE int sqlite3DbMaskAllZero(yDbMask m){ + int i; + for(i=0; ipToplevel==0 ); + db = pParse->db; + if( pParse->nested ) return; + if( db->mallocFailed || pParse->nErr ){ + if( pParse->rc==SQLITE_OK ) pParse->rc = SQLITE_ERROR; + return; + } + + /* Begin by generating some termination code at the end of the + ** vdbe program + */ + v = sqlite3GetVdbe(pParse); + assert( !pParse->isMultiWrite + || sqlite3VdbeAssertMayAbort(v, pParse->mayAbort)); + if( v ){ + sqlite3VdbeAddOp0(v, OP_Halt); + +#if SQLITE_USER_AUTHENTICATION + if( pParse->nTableLock>0 && db->init.busy==0 ){ + sqlite3UserAuthInit(db); + if( db->auth.authLevelrc = SQLITE_AUTH_USER; + return; + } + } +#endif + + /* The cookie mask contains one bit for each database file open. + ** (Bit 0 is for main, bit 1 is for temp, and so forth.) Bits are + ** set for each database that is used. Generate code to start a + ** transaction on each used database and to verify the schema cookie + ** on each used database. + */ + if( db->mallocFailed==0 + && (DbMaskNonZero(pParse->cookieMask) || pParse->pConstExpr) + ){ + int iDb, i; + assert( sqlite3VdbeGetOp(v, 0)->opcode==OP_Init ); + sqlite3VdbeJumpHere(v, 0); + for(iDb=0; iDbnDb; iDb++){ + Schema *pSchema; + if( DbMaskTest(pParse->cookieMask, iDb)==0 ) continue; + sqlite3VdbeUsesBtree(v, iDb); + pSchema = db->aDb[iDb].pSchema; + sqlite3VdbeAddOp4Int(v, + OP_Transaction, /* Opcode */ + iDb, /* P1 */ + DbMaskTest(pParse->writeMask,iDb), /* P2 */ + pSchema->schema_cookie, /* P3 */ + pSchema->iGeneration /* P4 */ + ); + if( db->init.busy==0 ) sqlite3VdbeChangeP5(v, 1); + VdbeComment((v, + "usesStmtJournal=%d", pParse->mayAbort && pParse->isMultiWrite)); + } +#ifndef SQLITE_OMIT_VIRTUALTABLE + for(i=0; inVtabLock; i++){ + char *vtab = (char *)sqlite3GetVTable(db, pParse->apVtabLock[i]); + sqlite3VdbeAddOp4(v, OP_VBegin, 0, 0, 0, vtab, P4_VTAB); + } + pParse->nVtabLock = 0; +#endif + + /* Once all the cookies have been verified and transactions opened, + ** obtain the required table-locks. This is a no-op unless the + ** shared-cache feature is enabled. + */ + codeTableLocks(pParse); + + /* Initialize any AUTOINCREMENT data structures required. + */ + sqlite3AutoincrementBegin(pParse); + + /* Code constant expressions that where factored out of inner loops */ + if( pParse->pConstExpr ){ + ExprList *pEL = pParse->pConstExpr; + pParse->okConstFactor = 0; + for(i=0; inExpr; i++){ + sqlite3ExprCode(pParse, pEL->a[i].pExpr, pEL->a[i].u.iConstExprReg); + } + } + + /* Finally, jump back to the beginning of the executable code. */ + sqlite3VdbeGoto(v, 1); + } + } + + + /* Get the VDBE program ready for execution + */ + if( v && pParse->nErr==0 && !db->mallocFailed ){ + /* A minimum of one cursor is required if autoincrement is used + * See ticket [a696379c1f08866] */ + assert( pParse->pAinc==0 || pParse->nTab>0 ); + sqlite3VdbeMakeReady(v, pParse); + pParse->rc = SQLITE_DONE; + }else{ + pParse->rc = SQLITE_ERROR; + } +} + +/* +** Run the parser and code generator recursively in order to generate +** code for the SQL statement given onto the end of the pParse context +** currently under construction. When the parser is run recursively +** this way, the final OP_Halt is not appended and other initialization +** and finalization steps are omitted because those are handling by the +** outermost parser. +** +** Not everything is nestable. This facility is designed to permit +** INSERT, UPDATE, and DELETE operations against SQLITE_MASTER. Use +** care if you decide to try to use this routine for some other purposes. +*/ +SQLITE_PRIVATE void sqlite3NestedParse(Parse *pParse, const char *zFormat, ...){ + va_list ap; + char *zSql; + char *zErrMsg = 0; + sqlite3 *db = pParse->db; + char saveBuf[PARSE_TAIL_SZ]; + + if( pParse->nErr ) return; + assert( pParse->nested<10 ); /* Nesting should only be of limited depth */ + va_start(ap, zFormat); + zSql = sqlite3VMPrintf(db, zFormat, ap); + va_end(ap); + if( zSql==0 ){ + /* This can result either from an OOM or because the formatted string + ** exceeds SQLITE_LIMIT_LENGTH. In the latter case, we need to set + ** an error */ + if( !db->mallocFailed ) pParse->rc = SQLITE_TOOBIG; + pParse->nErr++; + return; + } + pParse->nested++; + memcpy(saveBuf, PARSE_TAIL(pParse), PARSE_TAIL_SZ); + memset(PARSE_TAIL(pParse), 0, PARSE_TAIL_SZ); + sqlite3RunParser(pParse, zSql, &zErrMsg); + sqlite3DbFree(db, zErrMsg); + sqlite3DbFree(db, zSql); + memcpy(PARSE_TAIL(pParse), saveBuf, PARSE_TAIL_SZ); + pParse->nested--; +} + +#if SQLITE_USER_AUTHENTICATION +/* +** Return TRUE if zTable is the name of the system table that stores the +** list of users and their access credentials. +*/ +SQLITE_PRIVATE int sqlite3UserAuthTable(const char *zTable){ + return sqlite3_stricmp(zTable, "sqlite_user")==0; +} +#endif + +/* +** Locate the in-memory structure that describes a particular database +** table given the name of that table and (optionally) the name of the +** database containing the table. Return NULL if not found. +** +** If zDatabase is 0, all databases are searched for the table and the +** first matching table is returned. (No checking for duplicate table +** names is done.) The search order is TEMP first, then MAIN, then any +** auxiliary databases added using the ATTACH command. +** +** See also sqlite3LocateTable(). +*/ +SQLITE_PRIVATE Table *sqlite3FindTable(sqlite3 *db, const char *zName, const char *zDatabase){ + Table *p = 0; + int i; + + /* All mutexes are required for schema access. Make sure we hold them. */ + assert( zDatabase!=0 || sqlite3BtreeHoldsAllMutexes(db) ); +#if SQLITE_USER_AUTHENTICATION + /* Only the admin user is allowed to know that the sqlite_user table + ** exists */ + if( db->auth.authLevelnDb; i++){ + int j = (i<2) ? i^1 : i; /* Search TEMP before MAIN */ + if( zDatabase==0 || sqlite3StrICmp(zDatabase, db->aDb[j].zDbSName)==0 ){ + assert( sqlite3SchemaMutexHeld(db, j, 0) ); + p = sqlite3HashFind(&db->aDb[j].pSchema->tblHash, zName); + if( p ) return p; + } + } + /* Not found. If the name we were looking for was temp.sqlite_master + ** then change the name to sqlite_temp_master and try again. */ + if( sqlite3StrICmp(zName, MASTER_NAME)!=0 ) break; + if( sqlite3_stricmp(zDatabase, db->aDb[1].zDbSName)!=0 ) break; + zName = TEMP_MASTER_NAME; + } + return 0; +} + +/* +** Locate the in-memory structure that describes a particular database +** table given the name of that table and (optionally) the name of the +** database containing the table. Return NULL if not found. Also leave an +** error message in pParse->zErrMsg. +** +** The difference between this routine and sqlite3FindTable() is that this +** routine leaves an error message in pParse->zErrMsg where +** sqlite3FindTable() does not. +*/ +SQLITE_PRIVATE Table *sqlite3LocateTable( + Parse *pParse, /* context in which to report errors */ + u32 flags, /* LOCATE_VIEW or LOCATE_NOERR */ + const char *zName, /* Name of the table we are looking for */ + const char *zDbase /* Name of the database. Might be NULL */ +){ + Table *p; + sqlite3 *db = pParse->db; + + /* Read the database schema. If an error occurs, leave an error message + ** and code in pParse and return NULL. */ + if( (db->mDbFlags & DBFLAG_SchemaKnownOk)==0 + && SQLITE_OK!=sqlite3ReadSchema(pParse) + ){ + return 0; + } + + p = sqlite3FindTable(db, zName, zDbase); + if( p==0 ){ +#ifndef SQLITE_OMIT_VIRTUALTABLE + /* If zName is the not the name of a table in the schema created using + ** CREATE, then check to see if it is the name of an virtual table that + ** can be an eponymous virtual table. */ + if( pParse->disableVtab==0 ){ + Module *pMod = (Module*)sqlite3HashFind(&db->aModule, zName); + if( pMod==0 && sqlite3_strnicmp(zName, "pragma_", 7)==0 ){ + pMod = sqlite3PragmaVtabRegister(db, zName); + } + if( pMod && sqlite3VtabEponymousTableInit(pParse, pMod) ){ + return pMod->pEpoTab; + } + } +#endif + if( flags & LOCATE_NOERR ) return 0; + pParse->checkSchema = 1; + }else if( IsVirtual(p) && pParse->disableVtab ){ + p = 0; + } + + if( p==0 ){ + const char *zMsg = flags & LOCATE_VIEW ? "no such view" : "no such table"; + if( zDbase ){ + sqlite3ErrorMsg(pParse, "%s: %s.%s", zMsg, zDbase, zName); + }else{ + sqlite3ErrorMsg(pParse, "%s: %s", zMsg, zName); + } + } + + return p; +} + +/* +** Locate the table identified by *p. +** +** This is a wrapper around sqlite3LocateTable(). The difference between +** sqlite3LocateTable() and this function is that this function restricts +** the search to schema (p->pSchema) if it is not NULL. p->pSchema may be +** non-NULL if it is part of a view or trigger program definition. See +** sqlite3FixSrcList() for details. +*/ +SQLITE_PRIVATE Table *sqlite3LocateTableItem( + Parse *pParse, + u32 flags, + struct SrcList_item *p +){ + const char *zDb; + assert( p->pSchema==0 || p->zDatabase==0 ); + if( p->pSchema ){ + int iDb = sqlite3SchemaToIndex(pParse->db, p->pSchema); + zDb = pParse->db->aDb[iDb].zDbSName; + }else{ + zDb = p->zDatabase; + } + return sqlite3LocateTable(pParse, flags, p->zName, zDb); +} + +/* +** Locate the in-memory structure that describes +** a particular index given the name of that index +** and the name of the database that contains the index. +** Return NULL if not found. +** +** If zDatabase is 0, all databases are searched for the +** table and the first matching index is returned. (No checking +** for duplicate index names is done.) The search order is +** TEMP first, then MAIN, then any auxiliary databases added +** using the ATTACH command. +*/ +SQLITE_PRIVATE Index *sqlite3FindIndex(sqlite3 *db, const char *zName, const char *zDb){ + Index *p = 0; + int i; + /* All mutexes are required for schema access. Make sure we hold them. */ + assert( zDb!=0 || sqlite3BtreeHoldsAllMutexes(db) ); + for(i=OMIT_TEMPDB; inDb; i++){ + int j = (i<2) ? i^1 : i; /* Search TEMP before MAIN */ + Schema *pSchema = db->aDb[j].pSchema; + assert( pSchema ); + if( zDb && sqlite3StrICmp(zDb, db->aDb[j].zDbSName) ) continue; + assert( sqlite3SchemaMutexHeld(db, j, 0) ); + p = sqlite3HashFind(&pSchema->idxHash, zName); + if( p ) break; + } + return p; +} + +/* +** Reclaim the memory used by an index +*/ +SQLITE_PRIVATE void sqlite3FreeIndex(sqlite3 *db, Index *p){ +#ifndef SQLITE_OMIT_ANALYZE + sqlite3DeleteIndexSamples(db, p); +#endif + sqlite3ExprDelete(db, p->pPartIdxWhere); + sqlite3ExprListDelete(db, p->aColExpr); + sqlite3DbFree(db, p->zColAff); + if( p->isResized ) sqlite3DbFree(db, (void *)p->azColl); +#ifdef SQLITE_ENABLE_STAT4 + sqlite3_free(p->aiRowEst); +#endif + sqlite3DbFree(db, p); +} + +/* +** For the index called zIdxName which is found in the database iDb, +** unlike that index from its Table then remove the index from +** the index hash table and free all memory structures associated +** with the index. +*/ +SQLITE_PRIVATE void sqlite3UnlinkAndDeleteIndex(sqlite3 *db, int iDb, const char *zIdxName){ + Index *pIndex; + Hash *pHash; + + assert( sqlite3SchemaMutexHeld(db, iDb, 0) ); + pHash = &db->aDb[iDb].pSchema->idxHash; + pIndex = sqlite3HashInsert(pHash, zIdxName, 0); + if( ALWAYS(pIndex) ){ + if( pIndex->pTable->pIndex==pIndex ){ + pIndex->pTable->pIndex = pIndex->pNext; + }else{ + Index *p; + /* Justification of ALWAYS(); The index must be on the list of + ** indices. */ + p = pIndex->pTable->pIndex; + while( ALWAYS(p) && p->pNext!=pIndex ){ p = p->pNext; } + if( ALWAYS(p && p->pNext==pIndex) ){ + p->pNext = pIndex->pNext; + } + } + sqlite3FreeIndex(db, pIndex); + } + db->mDbFlags |= DBFLAG_SchemaChange; +} + +/* +** Look through the list of open database files in db->aDb[] and if +** any have been closed, remove them from the list. Reallocate the +** db->aDb[] structure to a smaller size, if possible. +** +** Entry 0 (the "main" database) and entry 1 (the "temp" database) +** are never candidates for being collapsed. +*/ +SQLITE_PRIVATE void sqlite3CollapseDatabaseArray(sqlite3 *db){ + int i, j; + for(i=j=2; inDb; i++){ + struct Db *pDb = &db->aDb[i]; + if( pDb->pBt==0 ){ + sqlite3DbFree(db, pDb->zDbSName); + pDb->zDbSName = 0; + continue; + } + if( jaDb[j] = db->aDb[i]; + } + j++; + } + db->nDb = j; + if( db->nDb<=2 && db->aDb!=db->aDbStatic ){ + memcpy(db->aDbStatic, db->aDb, 2*sizeof(db->aDb[0])); + sqlite3DbFree(db, db->aDb); + db->aDb = db->aDbStatic; + } +} + +/* +** Reset the schema for the database at index iDb. Also reset the +** TEMP schema. The reset is deferred if db->nSchemaLock is not zero. +** Deferred resets may be run by calling with iDb<0. +*/ +SQLITE_PRIVATE void sqlite3ResetOneSchema(sqlite3 *db, int iDb){ + int i; + assert( iDbnDb ); + + if( iDb>=0 ){ + assert( sqlite3SchemaMutexHeld(db, iDb, 0) ); + DbSetProperty(db, iDb, DB_ResetWanted); + DbSetProperty(db, 1, DB_ResetWanted); + db->mDbFlags &= ~DBFLAG_SchemaKnownOk; + } + + if( db->nSchemaLock==0 ){ + for(i=0; inDb; i++){ + if( DbHasProperty(db, i, DB_ResetWanted) ){ + sqlite3SchemaClear(db->aDb[i].pSchema); + } + } + } +} + +/* +** Erase all schema information from all attached databases (including +** "main" and "temp") for a single database connection. +*/ +SQLITE_PRIVATE void sqlite3ResetAllSchemasOfConnection(sqlite3 *db){ + int i; + sqlite3BtreeEnterAll(db); + for(i=0; inDb; i++){ + Db *pDb = &db->aDb[i]; + if( pDb->pSchema ){ + if( db->nSchemaLock==0 ){ + sqlite3SchemaClear(pDb->pSchema); + }else{ + DbSetProperty(db, i, DB_ResetWanted); + } + } + } + db->mDbFlags &= ~(DBFLAG_SchemaChange|DBFLAG_SchemaKnownOk); + sqlite3VtabUnlockList(db); + sqlite3BtreeLeaveAll(db); + if( db->nSchemaLock==0 ){ + sqlite3CollapseDatabaseArray(db); + } +} + +/* +** This routine is called when a commit occurs. +*/ +SQLITE_PRIVATE void sqlite3CommitInternalChanges(sqlite3 *db){ + db->mDbFlags &= ~DBFLAG_SchemaChange; +} + +/* +** Delete memory allocated for the column names of a table or view (the +** Table.aCol[] array). +*/ +SQLITE_PRIVATE void sqlite3DeleteColumnNames(sqlite3 *db, Table *pTable){ + int i; + Column *pCol; + assert( pTable!=0 ); + if( (pCol = pTable->aCol)!=0 ){ + for(i=0; inCol; i++, pCol++){ + sqlite3DbFree(db, pCol->zName); + sqlite3ExprDelete(db, pCol->pDflt); + sqlite3DbFree(db, pCol->zColl); + } + sqlite3DbFree(db, pTable->aCol); + } +} + +/* +** Remove the memory data structures associated with the given +** Table. No changes are made to disk by this routine. +** +** This routine just deletes the data structure. It does not unlink +** the table data structure from the hash table. But it does destroy +** memory structures of the indices and foreign keys associated with +** the table. +** +** The db parameter is optional. It is needed if the Table object +** contains lookaside memory. (Table objects in the schema do not use +** lookaside memory, but some ephemeral Table objects do.) Or the +** db parameter can be used with db->pnBytesFreed to measure the memory +** used by the Table object. +*/ +static void SQLITE_NOINLINE deleteTable(sqlite3 *db, Table *pTable){ + Index *pIndex, *pNext; + +#ifdef SQLITE_DEBUG + /* Record the number of outstanding lookaside allocations in schema Tables + ** prior to doing any free() operations. Since schema Tables do not use + ** lookaside, this number should not change. + ** + ** If malloc has already failed, it may be that it failed while allocating + ** a Table object that was going to be marked ephemeral. So do not check + ** that no lookaside memory is used in this case either. */ + int nLookaside = 0; + if( db && !db->mallocFailed && (pTable->tabFlags & TF_Ephemeral)==0 ){ + nLookaside = sqlite3LookasideUsed(db, 0); + } +#endif + + /* Delete all indices associated with this table. */ + for(pIndex = pTable->pIndex; pIndex; pIndex=pNext){ + pNext = pIndex->pNext; + assert( pIndex->pSchema==pTable->pSchema + || (IsVirtual(pTable) && pIndex->idxType!=SQLITE_IDXTYPE_APPDEF) ); + if( (db==0 || db->pnBytesFreed==0) && !IsVirtual(pTable) ){ + char *zName = pIndex->zName; + TESTONLY ( Index *pOld = ) sqlite3HashInsert( + &pIndex->pSchema->idxHash, zName, 0 + ); + assert( db==0 || sqlite3SchemaMutexHeld(db, 0, pIndex->pSchema) ); + assert( pOld==pIndex || pOld==0 ); + } + sqlite3FreeIndex(db, pIndex); + } + + /* Delete any foreign keys attached to this table. */ + sqlite3FkDelete(db, pTable); + + /* Delete the Table structure itself. + */ + sqlite3DeleteColumnNames(db, pTable); + sqlite3DbFree(db, pTable->zName); + sqlite3DbFree(db, pTable->zColAff); + sqlite3SelectDelete(db, pTable->pSelect); + sqlite3ExprListDelete(db, pTable->pCheck); +#ifndef SQLITE_OMIT_VIRTUALTABLE + sqlite3VtabClear(db, pTable); +#endif + sqlite3DbFree(db, pTable); + + /* Verify that no lookaside memory was used by schema tables */ + assert( nLookaside==0 || nLookaside==sqlite3LookasideUsed(db,0) ); +} +SQLITE_PRIVATE void sqlite3DeleteTable(sqlite3 *db, Table *pTable){ + /* Do not delete the table until the reference count reaches zero. */ + if( !pTable ) return; + if( ((!db || db->pnBytesFreed==0) && (--pTable->nTabRef)>0) ) return; + deleteTable(db, pTable); +} + + +/* +** Unlink the given table from the hash tables and the delete the +** table structure with all its indices and foreign keys. +*/ +SQLITE_PRIVATE void sqlite3UnlinkAndDeleteTable(sqlite3 *db, int iDb, const char *zTabName){ + Table *p; + Db *pDb; + + assert( db!=0 ); + assert( iDb>=0 && iDbnDb ); + assert( zTabName ); + assert( sqlite3SchemaMutexHeld(db, iDb, 0) ); + testcase( zTabName[0]==0 ); /* Zero-length table names are allowed */ + pDb = &db->aDb[iDb]; + p = sqlite3HashInsert(&pDb->pSchema->tblHash, zTabName, 0); + sqlite3DeleteTable(db, p); + db->mDbFlags |= DBFLAG_SchemaChange; +} + +/* +** Given a token, return a string that consists of the text of that +** token. Space to hold the returned string +** is obtained from sqliteMalloc() and must be freed by the calling +** function. +** +** Any quotation marks (ex: "name", 'name', [name], or `name`) that +** surround the body of the token are removed. +** +** Tokens are often just pointers into the original SQL text and so +** are not \000 terminated and are not persistent. The returned string +** is \000 terminated and is persistent. +*/ +SQLITE_PRIVATE char *sqlite3NameFromToken(sqlite3 *db, Token *pName){ + char *zName; + if( pName ){ + zName = sqlite3DbStrNDup(db, (char*)pName->z, pName->n); + sqlite3Dequote(zName); + }else{ + zName = 0; + } + return zName; +} + +/* +** Open the sqlite_master table stored in database number iDb for +** writing. The table is opened using cursor 0. +*/ +SQLITE_PRIVATE void sqlite3OpenMasterTable(Parse *p, int iDb){ + Vdbe *v = sqlite3GetVdbe(p); + sqlite3TableLock(p, iDb, MASTER_ROOT, 1, MASTER_NAME); + sqlite3VdbeAddOp4Int(v, OP_OpenWrite, 0, MASTER_ROOT, iDb, 5); + if( p->nTab==0 ){ + p->nTab = 1; + } +} + +/* +** Parameter zName points to a nul-terminated buffer containing the name +** of a database ("main", "temp" or the name of an attached db). This +** function returns the index of the named database in db->aDb[], or +** -1 if the named db cannot be found. +*/ +SQLITE_PRIVATE int sqlite3FindDbName(sqlite3 *db, const char *zName){ + int i = -1; /* Database number */ + if( zName ){ + Db *pDb; + for(i=(db->nDb-1), pDb=&db->aDb[i]; i>=0; i--, pDb--){ + if( 0==sqlite3_stricmp(pDb->zDbSName, zName) ) break; + /* "main" is always an acceptable alias for the primary database + ** even if it has been renamed using SQLITE_DBCONFIG_MAINDBNAME. */ + if( i==0 && 0==sqlite3_stricmp("main", zName) ) break; + } + } + return i; +} + +/* +** The token *pName contains the name of a database (either "main" or +** "temp" or the name of an attached db). This routine returns the +** index of the named database in db->aDb[], or -1 if the named db +** does not exist. +*/ +SQLITE_PRIVATE int sqlite3FindDb(sqlite3 *db, Token *pName){ + int i; /* Database number */ + char *zName; /* Name we are searching for */ + zName = sqlite3NameFromToken(db, pName); + i = sqlite3FindDbName(db, zName); + sqlite3DbFree(db, zName); + return i; +} + +/* The table or view or trigger name is passed to this routine via tokens +** pName1 and pName2. If the table name was fully qualified, for example: +** +** CREATE TABLE xxx.yyy (...); +** +** Then pName1 is set to "xxx" and pName2 "yyy". On the other hand if +** the table name is not fully qualified, i.e.: +** +** CREATE TABLE yyy(...); +** +** Then pName1 is set to "yyy" and pName2 is "". +** +** This routine sets the *ppUnqual pointer to point at the token (pName1 or +** pName2) that stores the unqualified table name. The index of the +** database "xxx" is returned. +*/ +SQLITE_PRIVATE int sqlite3TwoPartName( + Parse *pParse, /* Parsing and code generating context */ + Token *pName1, /* The "xxx" in the name "xxx.yyy" or "xxx" */ + Token *pName2, /* The "yyy" in the name "xxx.yyy" */ + Token **pUnqual /* Write the unqualified object name here */ +){ + int iDb; /* Database holding the object */ + sqlite3 *db = pParse->db; + + assert( pName2!=0 ); + if( pName2->n>0 ){ + if( db->init.busy ) { + sqlite3ErrorMsg(pParse, "corrupt database"); + return -1; + } + *pUnqual = pName2; + iDb = sqlite3FindDb(db, pName1); + if( iDb<0 ){ + sqlite3ErrorMsg(pParse, "unknown database %T", pName1); + return -1; + } + }else{ + assert( db->init.iDb==0 || db->init.busy || IN_RENAME_OBJECT + || (db->mDbFlags & DBFLAG_Vacuum)!=0); + iDb = db->init.iDb; + *pUnqual = pName1; + } + return iDb; +} + +/* +** True if PRAGMA writable_schema is ON +*/ +SQLITE_PRIVATE int sqlite3WritableSchema(sqlite3 *db){ + testcase( (db->flags&(SQLITE_WriteSchema|SQLITE_Defensive))==0 ); + testcase( (db->flags&(SQLITE_WriteSchema|SQLITE_Defensive))== + SQLITE_WriteSchema ); + testcase( (db->flags&(SQLITE_WriteSchema|SQLITE_Defensive))== + SQLITE_Defensive ); + testcase( (db->flags&(SQLITE_WriteSchema|SQLITE_Defensive))== + (SQLITE_WriteSchema|SQLITE_Defensive) ); + return (db->flags&(SQLITE_WriteSchema|SQLITE_Defensive))==SQLITE_WriteSchema; +} + +/* +** This routine is used to check if the UTF-8 string zName is a legal +** unqualified name for a new schema object (table, index, view or +** trigger). All names are legal except those that begin with the string +** "sqlite_" (in upper, lower or mixed case). This portion of the namespace +** is reserved for internal use. +** +** When parsing the sqlite_master table, this routine also checks to +** make sure the "type", "name", and "tbl_name" columns are consistent +** with the SQL. +*/ +SQLITE_PRIVATE int sqlite3CheckObjectName( + Parse *pParse, /* Parsing context */ + const char *zName, /* Name of the object to check */ + const char *zType, /* Type of this object */ + const char *zTblName /* Parent table name for triggers and indexes */ +){ + sqlite3 *db = pParse->db; + if( sqlite3WritableSchema(db) || db->init.imposterTable ){ + /* Skip these error checks for writable_schema=ON */ + return SQLITE_OK; + } + if( db->init.busy ){ + if( sqlite3_stricmp(zType, db->init.azInit[0]) + || sqlite3_stricmp(zName, db->init.azInit[1]) + || sqlite3_stricmp(zTblName, db->init.azInit[2]) + ){ + if( sqlite3Config.bExtraSchemaChecks ){ + sqlite3ErrorMsg(pParse, ""); /* corruptSchema() will supply the error */ + return SQLITE_ERROR; + } + } + }else{ + if( pParse->nested==0 + && 0==sqlite3StrNICmp(zName, "sqlite_", 7) + ){ + sqlite3ErrorMsg(pParse, "object name reserved for internal use: %s", + zName); + return SQLITE_ERROR; + } + } + return SQLITE_OK; +} + +/* +** Return the PRIMARY KEY index of a table +*/ +SQLITE_PRIVATE Index *sqlite3PrimaryKeyIndex(Table *pTab){ + Index *p; + for(p=pTab->pIndex; p && !IsPrimaryKeyIndex(p); p=p->pNext){} + return p; +} + +/* +** Return the column of index pIdx that corresponds to table +** column iCol. Return -1 if not found. +*/ +SQLITE_PRIVATE i16 sqlite3ColumnOfIndex(Index *pIdx, i16 iCol){ + int i; + for(i=0; inColumn; i++){ + if( iCol==pIdx->aiColumn[i] ) return i; + } + return -1; +} + +/* +** Begin constructing a new table representation in memory. This is +** the first of several action routines that get called in response +** to a CREATE TABLE statement. In particular, this routine is called +** after seeing tokens "CREATE" and "TABLE" and the table name. The isTemp +** flag is true if the table should be stored in the auxiliary database +** file instead of in the main database file. This is normally the case +** when the "TEMP" or "TEMPORARY" keyword occurs in between +** CREATE and TABLE. +** +** The new table record is initialized and put in pParse->pNewTable. +** As more of the CREATE TABLE statement is parsed, additional action +** routines will be called to add more information to this record. +** At the end of the CREATE TABLE statement, the sqlite3EndTable() routine +** is called to complete the construction of the new table record. +*/ +SQLITE_PRIVATE void sqlite3StartTable( + Parse *pParse, /* Parser context */ + Token *pName1, /* First part of the name of the table or view */ + Token *pName2, /* Second part of the name of the table or view */ + int isTemp, /* True if this is a TEMP table */ + int isView, /* True if this is a VIEW */ + int isVirtual, /* True if this is a VIRTUAL table */ + int noErr /* Do nothing if table already exists */ +){ + Table *pTable; + char *zName = 0; /* The name of the new table */ + sqlite3 *db = pParse->db; + Vdbe *v; + int iDb; /* Database number to create the table in */ + Token *pName; /* Unqualified name of the table to create */ + + if( db->init.busy && db->init.newTnum==1 ){ + /* Special case: Parsing the sqlite_master or sqlite_temp_master schema */ + iDb = db->init.iDb; + zName = sqlite3DbStrDup(db, SCHEMA_TABLE(iDb)); + pName = pName1; + }else{ + /* The common case */ + iDb = sqlite3TwoPartName(pParse, pName1, pName2, &pName); + if( iDb<0 ) return; + if( !OMIT_TEMPDB && isTemp && pName2->n>0 && iDb!=1 ){ + /* If creating a temp table, the name may not be qualified. Unless + ** the database name is "temp" anyway. */ + sqlite3ErrorMsg(pParse, "temporary table name must be unqualified"); + return; + } + if( !OMIT_TEMPDB && isTemp ) iDb = 1; + zName = sqlite3NameFromToken(db, pName); + if( IN_RENAME_OBJECT ){ + sqlite3RenameTokenMap(pParse, (void*)zName, pName); + } + } + pParse->sNameToken = *pName; + if( zName==0 ) return; + if( sqlite3CheckObjectName(pParse, zName, isView?"view":"table", zName) ){ + goto begin_table_error; + } + if( db->init.iDb==1 ) isTemp = 1; +#ifndef SQLITE_OMIT_AUTHORIZATION + assert( isTemp==0 || isTemp==1 ); + assert( isView==0 || isView==1 ); + { + static const u8 aCode[] = { + SQLITE_CREATE_TABLE, + SQLITE_CREATE_TEMP_TABLE, + SQLITE_CREATE_VIEW, + SQLITE_CREATE_TEMP_VIEW + }; + char *zDb = db->aDb[iDb].zDbSName; + if( sqlite3AuthCheck(pParse, SQLITE_INSERT, SCHEMA_TABLE(isTemp), 0, zDb) ){ + goto begin_table_error; + } + if( !isVirtual && sqlite3AuthCheck(pParse, (int)aCode[isTemp+2*isView], + zName, 0, zDb) ){ + goto begin_table_error; + } + } +#endif + + /* Make sure the new table name does not collide with an existing + ** index or table name in the same database. Issue an error message if + ** it does. The exception is if the statement being parsed was passed + ** to an sqlite3_declare_vtab() call. In that case only the column names + ** and types will be used, so there is no need to test for namespace + ** collisions. + */ + if( !IN_SPECIAL_PARSE ){ + char *zDb = db->aDb[iDb].zDbSName; + if( SQLITE_OK!=sqlite3ReadSchema(pParse) ){ + goto begin_table_error; + } + pTable = sqlite3FindTable(db, zName, zDb); + if( pTable ){ + if( !noErr ){ + sqlite3ErrorMsg(pParse, "table %T already exists", pName); + }else{ + assert( !db->init.busy || CORRUPT_DB ); + sqlite3CodeVerifySchema(pParse, iDb); + } + goto begin_table_error; + } + if( sqlite3FindIndex(db, zName, zDb)!=0 ){ + sqlite3ErrorMsg(pParse, "there is already an index named %s", zName); + goto begin_table_error; + } + } + + pTable = sqlite3DbMallocZero(db, sizeof(Table)); + if( pTable==0 ){ + assert( db->mallocFailed ); + pParse->rc = SQLITE_NOMEM_BKPT; + pParse->nErr++; + goto begin_table_error; + } + pTable->zName = zName; + pTable->iPKey = -1; + pTable->pSchema = db->aDb[iDb].pSchema; + pTable->nTabRef = 1; +#ifdef SQLITE_DEFAULT_ROWEST + pTable->nRowLogEst = sqlite3LogEst(SQLITE_DEFAULT_ROWEST); +#else + pTable->nRowLogEst = 200; assert( 200==sqlite3LogEst(1048576) ); +#endif + assert( pParse->pNewTable==0 ); + pParse->pNewTable = pTable; + + /* If this is the magic sqlite_sequence table used by autoincrement, + ** then record a pointer to this table in the main database structure + ** so that INSERT can find the table easily. + */ +#ifndef SQLITE_OMIT_AUTOINCREMENT + if( !pParse->nested && strcmp(zName, "sqlite_sequence")==0 ){ + assert( sqlite3SchemaMutexHeld(db, iDb, 0) ); + pTable->pSchema->pSeqTab = pTable; + } +#endif + + /* Begin generating the code that will insert the table record into + ** the SQLITE_MASTER table. Note in particular that we must go ahead + ** and allocate the record number for the table entry now. Before any + ** PRIMARY KEY or UNIQUE keywords are parsed. Those keywords will cause + ** indices to be created and the table record must come before the + ** indices. Hence, the record number for the table must be allocated + ** now. + */ + if( !db->init.busy && (v = sqlite3GetVdbe(pParse))!=0 ){ + int addr1; + int fileFormat; + int reg1, reg2, reg3; + /* nullRow[] is an OP_Record encoding of a row containing 5 NULLs */ + static const char nullRow[] = { 6, 0, 0, 0, 0, 0 }; + sqlite3BeginWriteOperation(pParse, 1, iDb); + +#ifndef SQLITE_OMIT_VIRTUALTABLE + if( isVirtual ){ + sqlite3VdbeAddOp0(v, OP_VBegin); + } +#endif + + /* If the file format and encoding in the database have not been set, + ** set them now. + */ + reg1 = pParse->regRowid = ++pParse->nMem; + reg2 = pParse->regRoot = ++pParse->nMem; + reg3 = ++pParse->nMem; + sqlite3VdbeAddOp3(v, OP_ReadCookie, iDb, reg3, BTREE_FILE_FORMAT); + sqlite3VdbeUsesBtree(v, iDb); + addr1 = sqlite3VdbeAddOp1(v, OP_If, reg3); VdbeCoverage(v); + fileFormat = (db->flags & SQLITE_LegacyFileFmt)!=0 ? + 1 : SQLITE_MAX_FILE_FORMAT; + sqlite3VdbeAddOp3(v, OP_SetCookie, iDb, BTREE_FILE_FORMAT, fileFormat); + sqlite3VdbeAddOp3(v, OP_SetCookie, iDb, BTREE_TEXT_ENCODING, ENC(db)); + sqlite3VdbeJumpHere(v, addr1); + + /* This just creates a place-holder record in the sqlite_master table. + ** The record created does not contain anything yet. It will be replaced + ** by the real entry in code generated at sqlite3EndTable(). + ** + ** The rowid for the new entry is left in register pParse->regRowid. + ** The root page number of the new table is left in reg pParse->regRoot. + ** The rowid and root page number values are needed by the code that + ** sqlite3EndTable will generate. + */ +#if !defined(SQLITE_OMIT_VIEW) || !defined(SQLITE_OMIT_VIRTUALTABLE) + if( isView || isVirtual ){ + sqlite3VdbeAddOp2(v, OP_Integer, 0, reg2); + }else +#endif + { + pParse->addrCrTab = + sqlite3VdbeAddOp3(v, OP_CreateBtree, iDb, reg2, BTREE_INTKEY); + } + sqlite3OpenMasterTable(pParse, iDb); + sqlite3VdbeAddOp2(v, OP_NewRowid, 0, reg1); + sqlite3VdbeAddOp4(v, OP_Blob, 6, reg3, 0, nullRow, P4_STATIC); + sqlite3VdbeAddOp3(v, OP_Insert, 0, reg3, reg1); + sqlite3VdbeChangeP5(v, OPFLAG_APPEND); + sqlite3VdbeAddOp0(v, OP_Close); + } + + /* Normal (non-error) return. */ + return; + + /* If an error occurs, we jump here */ +begin_table_error: + sqlite3DbFree(db, zName); + return; +} + +/* Set properties of a table column based on the (magical) +** name of the column. +*/ +#if SQLITE_ENABLE_HIDDEN_COLUMNS +SQLITE_PRIVATE void sqlite3ColumnPropertiesFromName(Table *pTab, Column *pCol){ + if( sqlite3_strnicmp(pCol->zName, "__hidden__", 10)==0 ){ + pCol->colFlags |= COLFLAG_HIDDEN; + }else if( pTab && pCol!=pTab->aCol && (pCol[-1].colFlags & COLFLAG_HIDDEN) ){ + pTab->tabFlags |= TF_OOOHidden; + } +} +#endif + + +/* +** Add a new column to the table currently being constructed. +** +** The parser calls this routine once for each column declaration +** in a CREATE TABLE statement. sqlite3StartTable() gets called +** first to get things going. Then this routine is called for each +** column. +*/ +SQLITE_PRIVATE void sqlite3AddColumn(Parse *pParse, Token *pName, Token *pType){ + Table *p; + int i; + char *z; + char *zType; + Column *pCol; + sqlite3 *db = pParse->db; + if( (p = pParse->pNewTable)==0 ) return; + if( p->nCol+1>db->aLimit[SQLITE_LIMIT_COLUMN] ){ + sqlite3ErrorMsg(pParse, "too many columns on %s", p->zName); + return; + } + z = sqlite3DbMallocRaw(db, pName->n + pType->n + 2); + if( z==0 ) return; + if( IN_RENAME_OBJECT ) sqlite3RenameTokenMap(pParse, (void*)z, pName); + memcpy(z, pName->z, pName->n); + z[pName->n] = 0; + sqlite3Dequote(z); + for(i=0; inCol; i++){ + if( sqlite3_stricmp(z, p->aCol[i].zName)==0 ){ + sqlite3ErrorMsg(pParse, "duplicate column name: %s", z); + sqlite3DbFree(db, z); + return; + } + } + if( (p->nCol & 0x7)==0 ){ + Column *aNew; + aNew = sqlite3DbRealloc(db,p->aCol,(p->nCol+8)*sizeof(p->aCol[0])); + if( aNew==0 ){ + sqlite3DbFree(db, z); + return; + } + p->aCol = aNew; + } + pCol = &p->aCol[p->nCol]; + memset(pCol, 0, sizeof(p->aCol[0])); + pCol->zName = z; + sqlite3ColumnPropertiesFromName(p, pCol); + + if( pType->n==0 ){ + /* If there is no type specified, columns have the default affinity + ** 'BLOB' with a default size of 4 bytes. */ + pCol->affinity = SQLITE_AFF_BLOB; + pCol->szEst = 1; +#ifdef SQLITE_ENABLE_SORTER_REFERENCES + if( 4>=sqlite3GlobalConfig.szSorterRef ){ + pCol->colFlags |= COLFLAG_SORTERREF; + } +#endif + }else{ + zType = z + sqlite3Strlen30(z) + 1; + memcpy(zType, pType->z, pType->n); + zType[pType->n] = 0; + sqlite3Dequote(zType); + pCol->affinity = sqlite3AffinityType(zType, pCol); + pCol->colFlags |= COLFLAG_HASTYPE; + } + p->nCol++; + pParse->constraintName.n = 0; +} + +/* +** This routine is called by the parser while in the middle of +** parsing a CREATE TABLE statement. A "NOT NULL" constraint has +** been seen on a column. This routine sets the notNull flag on +** the column currently under construction. +*/ +SQLITE_PRIVATE void sqlite3AddNotNull(Parse *pParse, int onError){ + Table *p; + Column *pCol; + p = pParse->pNewTable; + if( p==0 || NEVER(p->nCol<1) ) return; + pCol = &p->aCol[p->nCol-1]; + pCol->notNull = (u8)onError; + p->tabFlags |= TF_HasNotNull; + + /* Set the uniqNotNull flag on any UNIQUE or PK indexes already created + ** on this column. */ + if( pCol->colFlags & COLFLAG_UNIQUE ){ + Index *pIdx; + for(pIdx=p->pIndex; pIdx; pIdx=pIdx->pNext){ + assert( pIdx->nKeyCol==1 && pIdx->onError!=OE_None ); + if( pIdx->aiColumn[0]==p->nCol-1 ){ + pIdx->uniqNotNull = 1; + } + } + } +} + +/* +** Scan the column type name zType (length nType) and return the +** associated affinity type. +** +** This routine does a case-independent search of zType for the +** substrings in the following table. If one of the substrings is +** found, the corresponding affinity is returned. If zType contains +** more than one of the substrings, entries toward the top of +** the table take priority. For example, if zType is 'BLOBINT', +** SQLITE_AFF_INTEGER is returned. +** +** Substring | Affinity +** -------------------------------- +** 'INT' | SQLITE_AFF_INTEGER +** 'CHAR' | SQLITE_AFF_TEXT +** 'CLOB' | SQLITE_AFF_TEXT +** 'TEXT' | SQLITE_AFF_TEXT +** 'BLOB' | SQLITE_AFF_BLOB +** 'REAL' | SQLITE_AFF_REAL +** 'FLOA' | SQLITE_AFF_REAL +** 'DOUB' | SQLITE_AFF_REAL +** +** If none of the substrings in the above table are found, +** SQLITE_AFF_NUMERIC is returned. +*/ +SQLITE_PRIVATE char sqlite3AffinityType(const char *zIn, Column *pCol){ + u32 h = 0; + char aff = SQLITE_AFF_NUMERIC; + const char *zChar = 0; + + assert( zIn!=0 ); + while( zIn[0] ){ + h = (h<<8) + sqlite3UpperToLower[(*zIn)&0xff]; + zIn++; + if( h==(('c'<<24)+('h'<<16)+('a'<<8)+'r') ){ /* CHAR */ + aff = SQLITE_AFF_TEXT; + zChar = zIn; + }else if( h==(('c'<<24)+('l'<<16)+('o'<<8)+'b') ){ /* CLOB */ + aff = SQLITE_AFF_TEXT; + }else if( h==(('t'<<24)+('e'<<16)+('x'<<8)+'t') ){ /* TEXT */ + aff = SQLITE_AFF_TEXT; + }else if( h==(('b'<<24)+('l'<<16)+('o'<<8)+'b') /* BLOB */ + && (aff==SQLITE_AFF_NUMERIC || aff==SQLITE_AFF_REAL) ){ + aff = SQLITE_AFF_BLOB; + if( zIn[0]=='(' ) zChar = zIn; +#ifndef SQLITE_OMIT_FLOATING_POINT + }else if( h==(('r'<<24)+('e'<<16)+('a'<<8)+'l') /* REAL */ + && aff==SQLITE_AFF_NUMERIC ){ + aff = SQLITE_AFF_REAL; + }else if( h==(('f'<<24)+('l'<<16)+('o'<<8)+'a') /* FLOA */ + && aff==SQLITE_AFF_NUMERIC ){ + aff = SQLITE_AFF_REAL; + }else if( h==(('d'<<24)+('o'<<16)+('u'<<8)+'b') /* DOUB */ + && aff==SQLITE_AFF_NUMERIC ){ + aff = SQLITE_AFF_REAL; +#endif + }else if( (h&0x00FFFFFF)==(('i'<<16)+('n'<<8)+'t') ){ /* INT */ + aff = SQLITE_AFF_INTEGER; + break; + } + } + + /* If pCol is not NULL, store an estimate of the field size. The + ** estimate is scaled so that the size of an integer is 1. */ + if( pCol ){ + int v = 0; /* default size is approx 4 bytes */ + if( aff r=(k/4+1) */ + sqlite3GetInt32(zChar, &v); + break; + } + zChar++; + } + }else{ + v = 16; /* BLOB, TEXT, CLOB -> r=5 (approx 20 bytes)*/ + } + } +#ifdef SQLITE_ENABLE_SORTER_REFERENCES + if( v>=sqlite3GlobalConfig.szSorterRef ){ + pCol->colFlags |= COLFLAG_SORTERREF; + } +#endif + v = v/4 + 1; + if( v>255 ) v = 255; + pCol->szEst = v; + } + return aff; +} + +/* +** The expression is the default value for the most recently added column +** of the table currently under construction. +** +** Default value expressions must be constant. Raise an exception if this +** is not the case. +** +** This routine is called by the parser while in the middle of +** parsing a CREATE TABLE statement. +*/ +SQLITE_PRIVATE void sqlite3AddDefaultValue( + Parse *pParse, /* Parsing context */ + Expr *pExpr, /* The parsed expression of the default value */ + const char *zStart, /* Start of the default value text */ + const char *zEnd /* First character past end of defaut value text */ +){ + Table *p; + Column *pCol; + sqlite3 *db = pParse->db; + p = pParse->pNewTable; + if( p!=0 ){ + pCol = &(p->aCol[p->nCol-1]); + if( !sqlite3ExprIsConstantOrFunction(pExpr, db->init.busy) ){ + sqlite3ErrorMsg(pParse, "default value of column [%s] is not constant", + pCol->zName); + }else{ + /* A copy of pExpr is used instead of the original, as pExpr contains + ** tokens that point to volatile memory. + */ + Expr x; + sqlite3ExprDelete(db, pCol->pDflt); + memset(&x, 0, sizeof(x)); + x.op = TK_SPAN; + x.u.zToken = sqlite3DbSpanDup(db, zStart, zEnd); + x.pLeft = pExpr; + x.flags = EP_Skip; + pCol->pDflt = sqlite3ExprDup(db, &x, EXPRDUP_REDUCE); + sqlite3DbFree(db, x.u.zToken); + } + } + if( IN_RENAME_OBJECT ){ + sqlite3RenameExprUnmap(pParse, pExpr); + } + sqlite3ExprDelete(db, pExpr); +} + +/* +** Backwards Compatibility Hack: +** +** Historical versions of SQLite accepted strings as column names in +** indexes and PRIMARY KEY constraints and in UNIQUE constraints. Example: +** +** CREATE TABLE xyz(a,b,c,d,e,PRIMARY KEY('a'),UNIQUE('b','c' COLLATE trim) +** CREATE INDEX abc ON xyz('c','d' DESC,'e' COLLATE nocase DESC); +** +** This is goofy. But to preserve backwards compatibility we continue to +** accept it. This routine does the necessary conversion. It converts +** the expression given in its argument from a TK_STRING into a TK_ID +** if the expression is just a TK_STRING with an optional COLLATE clause. +** If the expression is anything other than TK_STRING, the expression is +** unchanged. +*/ +static void sqlite3StringToId(Expr *p){ + if( p->op==TK_STRING ){ + p->op = TK_ID; + }else if( p->op==TK_COLLATE && p->pLeft->op==TK_STRING ){ + p->pLeft->op = TK_ID; + } +} + +/* +** Designate the PRIMARY KEY for the table. pList is a list of names +** of columns that form the primary key. If pList is NULL, then the +** most recently added column of the table is the primary key. +** +** A table can have at most one primary key. If the table already has +** a primary key (and this is the second primary key) then create an +** error. +** +** If the PRIMARY KEY is on a single column whose datatype is INTEGER, +** then we will try to use that column as the rowid. Set the Table.iPKey +** field of the table under construction to be the index of the +** INTEGER PRIMARY KEY column. Table.iPKey is set to -1 if there is +** no INTEGER PRIMARY KEY. +** +** If the key is not an INTEGER PRIMARY KEY, then create a unique +** index for the key. No index is created for INTEGER PRIMARY KEYs. +*/ +SQLITE_PRIVATE void sqlite3AddPrimaryKey( + Parse *pParse, /* Parsing context */ + ExprList *pList, /* List of field names to be indexed */ + int onError, /* What to do with a uniqueness conflict */ + int autoInc, /* True if the AUTOINCREMENT keyword is present */ + int sortOrder /* SQLITE_SO_ASC or SQLITE_SO_DESC */ +){ + Table *pTab = pParse->pNewTable; + Column *pCol = 0; + int iCol = -1, i; + int nTerm; + if( pTab==0 ) goto primary_key_exit; + if( pTab->tabFlags & TF_HasPrimaryKey ){ + sqlite3ErrorMsg(pParse, + "table \"%s\" has more than one primary key", pTab->zName); + goto primary_key_exit; + } + pTab->tabFlags |= TF_HasPrimaryKey; + if( pList==0 ){ + iCol = pTab->nCol - 1; + pCol = &pTab->aCol[iCol]; + pCol->colFlags |= COLFLAG_PRIMKEY; + nTerm = 1; + }else{ + nTerm = pList->nExpr; + for(i=0; ia[i].pExpr); + assert( pCExpr!=0 ); + sqlite3StringToId(pCExpr); + if( pCExpr->op==TK_ID ){ + const char *zCName = pCExpr->u.zToken; + for(iCol=0; iColnCol; iCol++){ + if( sqlite3StrICmp(zCName, pTab->aCol[iCol].zName)==0 ){ + pCol = &pTab->aCol[iCol]; + pCol->colFlags |= COLFLAG_PRIMKEY; + break; + } + } + } + } + } + if( nTerm==1 + && pCol + && sqlite3StrICmp(sqlite3ColumnType(pCol,""), "INTEGER")==0 + && sortOrder!=SQLITE_SO_DESC + ){ + if( IN_RENAME_OBJECT && pList ){ + Expr *pCExpr = sqlite3ExprSkipCollate(pList->a[0].pExpr); + sqlite3RenameTokenRemap(pParse, &pTab->iPKey, pCExpr); + } + pTab->iPKey = iCol; + pTab->keyConf = (u8)onError; + assert( autoInc==0 || autoInc==1 ); + pTab->tabFlags |= autoInc*TF_Autoincrement; + if( pList ) pParse->iPkSortOrder = pList->a[0].sortFlags; + }else if( autoInc ){ +#ifndef SQLITE_OMIT_AUTOINCREMENT + sqlite3ErrorMsg(pParse, "AUTOINCREMENT is only allowed on an " + "INTEGER PRIMARY KEY"); +#endif + }else{ + sqlite3CreateIndex(pParse, 0, 0, 0, pList, onError, 0, + 0, sortOrder, 0, SQLITE_IDXTYPE_PRIMARYKEY); + pList = 0; + } + +primary_key_exit: + sqlite3ExprListDelete(pParse->db, pList); + return; +} + +/* +** Add a new CHECK constraint to the table currently under construction. +*/ +SQLITE_PRIVATE void sqlite3AddCheckConstraint( + Parse *pParse, /* Parsing context */ + Expr *pCheckExpr /* The check expression */ +){ +#ifndef SQLITE_OMIT_CHECK + Table *pTab = pParse->pNewTable; + sqlite3 *db = pParse->db; + if( pTab && !IN_DECLARE_VTAB + && !sqlite3BtreeIsReadonly(db->aDb[db->init.iDb].pBt) + ){ + pTab->pCheck = sqlite3ExprListAppend(pParse, pTab->pCheck, pCheckExpr); + if( pParse->constraintName.n ){ + sqlite3ExprListSetName(pParse, pTab->pCheck, &pParse->constraintName, 1); + } + }else +#endif + { + sqlite3ExprDelete(pParse->db, pCheckExpr); + } +} + +/* +** Set the collation function of the most recently parsed table column +** to the CollSeq given. +*/ +SQLITE_PRIVATE void sqlite3AddCollateType(Parse *pParse, Token *pToken){ + Table *p; + int i; + char *zColl; /* Dequoted name of collation sequence */ + sqlite3 *db; + + if( (p = pParse->pNewTable)==0 ) return; + i = p->nCol-1; + db = pParse->db; + zColl = sqlite3NameFromToken(db, pToken); + if( !zColl ) return; + + if( sqlite3LocateCollSeq(pParse, zColl) ){ + Index *pIdx; + sqlite3DbFree(db, p->aCol[i].zColl); + p->aCol[i].zColl = zColl; + + /* If the column is declared as " PRIMARY KEY COLLATE ", + ** then an index may have been created on this column before the + ** collation type was added. Correct this if it is the case. + */ + for(pIdx=p->pIndex; pIdx; pIdx=pIdx->pNext){ + assert( pIdx->nKeyCol==1 ); + if( pIdx->aiColumn[0]==i ){ + pIdx->azColl[0] = p->aCol[i].zColl; + } + } + }else{ + sqlite3DbFree(db, zColl); + } +} + +/* +** This function returns the collation sequence for database native text +** encoding identified by the string zName, length nName. +** +** If the requested collation sequence is not available, or not available +** in the database native encoding, the collation factory is invoked to +** request it. If the collation factory does not supply such a sequence, +** and the sequence is available in another text encoding, then that is +** returned instead. +** +** If no versions of the requested collations sequence are available, or +** another error occurs, NULL is returned and an error message written into +** pParse. +** +** This routine is a wrapper around sqlite3FindCollSeq(). This routine +** invokes the collation factory if the named collation cannot be found +** and generates an error message. +** +** See also: sqlite3FindCollSeq(), sqlite3GetCollSeq() +*/ +SQLITE_PRIVATE CollSeq *sqlite3LocateCollSeq(Parse *pParse, const char *zName){ + sqlite3 *db = pParse->db; + u8 enc = ENC(db); + u8 initbusy = db->init.busy; + CollSeq *pColl; + + pColl = sqlite3FindCollSeq(db, enc, zName, initbusy); + if( !initbusy && (!pColl || !pColl->xCmp) ){ + pColl = sqlite3GetCollSeq(pParse, enc, pColl, zName); + } + + return pColl; +} + + +/* +** Generate code that will increment the schema cookie. +** +** The schema cookie is used to determine when the schema for the +** database changes. After each schema change, the cookie value +** changes. When a process first reads the schema it records the +** cookie. Thereafter, whenever it goes to access the database, +** it checks the cookie to make sure the schema has not changed +** since it was last read. +** +** This plan is not completely bullet-proof. It is possible for +** the schema to change multiple times and for the cookie to be +** set back to prior value. But schema changes are infrequent +** and the probability of hitting the same cookie value is only +** 1 chance in 2^32. So we're safe enough. +** +** IMPLEMENTATION-OF: R-34230-56049 SQLite automatically increments +** the schema-version whenever the schema changes. +*/ +SQLITE_PRIVATE void sqlite3ChangeCookie(Parse *pParse, int iDb){ + sqlite3 *db = pParse->db; + Vdbe *v = pParse->pVdbe; + assert( sqlite3SchemaMutexHeld(db, iDb, 0) ); + sqlite3VdbeAddOp3(v, OP_SetCookie, iDb, BTREE_SCHEMA_VERSION, + (int)(1+(unsigned)db->aDb[iDb].pSchema->schema_cookie)); +} + +/* +** Measure the number of characters needed to output the given +** identifier. The number returned includes any quotes used +** but does not include the null terminator. +** +** The estimate is conservative. It might be larger that what is +** really needed. +*/ +static int identLength(const char *z){ + int n; + for(n=0; *z; n++, z++){ + if( *z=='"' ){ n++; } + } + return n + 2; +} + +/* +** The first parameter is a pointer to an output buffer. The second +** parameter is a pointer to an integer that contains the offset at +** which to write into the output buffer. This function copies the +** nul-terminated string pointed to by the third parameter, zSignedIdent, +** to the specified offset in the buffer and updates *pIdx to refer +** to the first byte after the last byte written before returning. +** +** If the string zSignedIdent consists entirely of alpha-numeric +** characters, does not begin with a digit and is not an SQL keyword, +** then it is copied to the output buffer exactly as it is. Otherwise, +** it is quoted using double-quotes. +*/ +static void identPut(char *z, int *pIdx, char *zSignedIdent){ + unsigned char *zIdent = (unsigned char*)zSignedIdent; + int i, j, needQuote; + i = *pIdx; + + for(j=0; zIdent[j]; j++){ + if( !sqlite3Isalnum(zIdent[j]) && zIdent[j]!='_' ) break; + } + needQuote = sqlite3Isdigit(zIdent[0]) + || sqlite3KeywordCode(zIdent, j)!=TK_ID + || zIdent[j]!=0 + || j==0; + + if( needQuote ) z[i++] = '"'; + for(j=0; zIdent[j]; j++){ + z[i++] = zIdent[j]; + if( zIdent[j]=='"' ) z[i++] = '"'; + } + if( needQuote ) z[i++] = '"'; + z[i] = 0; + *pIdx = i; +} + +/* +** Generate a CREATE TABLE statement appropriate for the given +** table. Memory to hold the text of the statement is obtained +** from sqliteMalloc() and must be freed by the calling function. +*/ +static char *createTableStmt(sqlite3 *db, Table *p){ + int i, k, n; + char *zStmt; + char *zSep, *zSep2, *zEnd; + Column *pCol; + n = 0; + for(pCol = p->aCol, i=0; inCol; i++, pCol++){ + n += identLength(pCol->zName) + 5; + } + n += identLength(p->zName); + if( n<50 ){ + zSep = ""; + zSep2 = ","; + zEnd = ")"; + }else{ + zSep = "\n "; + zSep2 = ",\n "; + zEnd = "\n)"; + } + n += 35 + 6*p->nCol; + zStmt = sqlite3DbMallocRaw(0, n); + if( zStmt==0 ){ + sqlite3OomFault(db); + return 0; + } + sqlite3_snprintf(n, zStmt, "CREATE TABLE "); + k = sqlite3Strlen30(zStmt); + identPut(zStmt, &k, p->zName); + zStmt[k++] = '('; + for(pCol=p->aCol, i=0; inCol; i++, pCol++){ + static const char * const azType[] = { + /* SQLITE_AFF_BLOB */ "", + /* SQLITE_AFF_TEXT */ " TEXT", + /* SQLITE_AFF_NUMERIC */ " NUM", + /* SQLITE_AFF_INTEGER */ " INT", + /* SQLITE_AFF_REAL */ " REAL" + }; + int len; + const char *zType; + + sqlite3_snprintf(n-k, &zStmt[k], zSep); + k += sqlite3Strlen30(&zStmt[k]); + zSep = zSep2; + identPut(zStmt, &k, pCol->zName); + assert( pCol->affinity-SQLITE_AFF_BLOB >= 0 ); + assert( pCol->affinity-SQLITE_AFF_BLOB < ArraySize(azType) ); + testcase( pCol->affinity==SQLITE_AFF_BLOB ); + testcase( pCol->affinity==SQLITE_AFF_TEXT ); + testcase( pCol->affinity==SQLITE_AFF_NUMERIC ); + testcase( pCol->affinity==SQLITE_AFF_INTEGER ); + testcase( pCol->affinity==SQLITE_AFF_REAL ); + + zType = azType[pCol->affinity - SQLITE_AFF_BLOB]; + len = sqlite3Strlen30(zType); + assert( pCol->affinity==SQLITE_AFF_BLOB + || pCol->affinity==sqlite3AffinityType(zType, 0) ); + memcpy(&zStmt[k], zType, len); + k += len; + assert( k<=n ); + } + sqlite3_snprintf(n-k, &zStmt[k], "%s", zEnd); + return zStmt; +} + +/* +** Resize an Index object to hold N columns total. Return SQLITE_OK +** on success and SQLITE_NOMEM on an OOM error. +*/ +static int resizeIndexObject(sqlite3 *db, Index *pIdx, int N){ + char *zExtra; + int nByte; + if( pIdx->nColumn>=N ) return SQLITE_OK; + assert( pIdx->isResized==0 ); + nByte = (sizeof(char*) + sizeof(i16) + 1)*N; + zExtra = sqlite3DbMallocZero(db, nByte); + if( zExtra==0 ) return SQLITE_NOMEM_BKPT; + memcpy(zExtra, pIdx->azColl, sizeof(char*)*pIdx->nColumn); + pIdx->azColl = (const char**)zExtra; + zExtra += sizeof(char*)*N; + memcpy(zExtra, pIdx->aiColumn, sizeof(i16)*pIdx->nColumn); + pIdx->aiColumn = (i16*)zExtra; + zExtra += sizeof(i16)*N; + memcpy(zExtra, pIdx->aSortOrder, pIdx->nColumn); + pIdx->aSortOrder = (u8*)zExtra; + pIdx->nColumn = N; + pIdx->isResized = 1; + return SQLITE_OK; +} + +/* +** Estimate the total row width for a table. +*/ +static void estimateTableWidth(Table *pTab){ + unsigned wTable = 0; + const Column *pTabCol; + int i; + for(i=pTab->nCol, pTabCol=pTab->aCol; i>0; i--, pTabCol++){ + wTable += pTabCol->szEst; + } + if( pTab->iPKey<0 ) wTable++; + pTab->szTabRow = sqlite3LogEst(wTable*4); +} + +/* +** Estimate the average size of a row for an index. +*/ +static void estimateIndexWidth(Index *pIdx){ + unsigned wIndex = 0; + int i; + const Column *aCol = pIdx->pTable->aCol; + for(i=0; inColumn; i++){ + i16 x = pIdx->aiColumn[i]; + assert( xpTable->nCol ); + wIndex += x<0 ? 1 : aCol[pIdx->aiColumn[i]].szEst; + } + pIdx->szIdxRow = sqlite3LogEst(wIndex*4); +} + +/* Return true if column number x is any of the first nCol entries of aiCol[]. +** This is used to determine if the column number x appears in any of the +** first nCol entries of an index. +*/ +static int hasColumn(const i16 *aiCol, int nCol, int x){ + while( nCol-- > 0 ){ + assert( aiCol[0]>=0 ); + if( x==*(aiCol++) ){ + return 1; + } + } + return 0; +} + +/* +** Return true if any of the first nKey entries of index pIdx exactly +** match the iCol-th entry of pPk. pPk is always a WITHOUT ROWID +** PRIMARY KEY index. pIdx is an index on the same table. pIdx may +** or may not be the same index as pPk. +** +** The first nKey entries of pIdx are guaranteed to be ordinary columns, +** not a rowid or expression. +** +** This routine differs from hasColumn() in that both the column and the +** collating sequence must match for this routine, but for hasColumn() only +** the column name must match. +*/ +static int isDupColumn(Index *pIdx, int nKey, Index *pPk, int iCol){ + int i, j; + assert( nKey<=pIdx->nColumn ); + assert( iColnColumn,pPk->nKeyCol) ); + assert( pPk->idxType==SQLITE_IDXTYPE_PRIMARYKEY ); + assert( pPk->pTable->tabFlags & TF_WithoutRowid ); + assert( pPk->pTable==pIdx->pTable ); + testcase( pPk==pIdx ); + j = pPk->aiColumn[iCol]; + assert( j!=XN_ROWID && j!=XN_EXPR ); + for(i=0; iaiColumn[i]>=0 || j>=0 ); + if( pIdx->aiColumn[i]==j + && sqlite3StrICmp(pIdx->azColl[i], pPk->azColl[iCol])==0 + ){ + return 1; + } + } + return 0; +} + +/* Recompute the colNotIdxed field of the Index. +** +** colNotIdxed is a bitmask that has a 0 bit representing each indexed +** columns that are within the first 63 columns of the table. The +** high-order bit of colNotIdxed is always 1. All unindexed columns +** of the table have a 1. +** +** The colNotIdxed mask is AND-ed with the SrcList.a[].colUsed mask +** to determine if the index is covering index. +*/ +static void recomputeColumnsNotIndexed(Index *pIdx){ + Bitmask m = 0; + int j; + for(j=pIdx->nColumn-1; j>=0; j--){ + int x = pIdx->aiColumn[j]; + if( x>=0 ){ + testcase( x==BMS-1 ); + testcase( x==BMS-2 ); + if( xcolNotIdxed = ~m; + assert( (pIdx->colNotIdxed>>63)==1 ); +} + +/* +** This routine runs at the end of parsing a CREATE TABLE statement that +** has a WITHOUT ROWID clause. The job of this routine is to convert both +** internal schema data structures and the generated VDBE code so that they +** are appropriate for a WITHOUT ROWID table instead of a rowid table. +** Changes include: +** +** (1) Set all columns of the PRIMARY KEY schema object to be NOT NULL. +** (2) Convert P3 parameter of the OP_CreateBtree from BTREE_INTKEY +** into BTREE_BLOBKEY. +** (3) Bypass the creation of the sqlite_master table entry +** for the PRIMARY KEY as the primary key index is now +** identified by the sqlite_master table entry of the table itself. +** (4) Set the Index.tnum of the PRIMARY KEY Index object in the +** schema to the rootpage from the main table. +** (5) Add all table columns to the PRIMARY KEY Index object +** so that the PRIMARY KEY is a covering index. The surplus +** columns are part of KeyInfo.nAllField and are not used for +** sorting or lookup or uniqueness checks. +** (6) Replace the rowid tail on all automatically generated UNIQUE +** indices with the PRIMARY KEY columns. +** +** For virtual tables, only (1) is performed. +*/ +static void convertToWithoutRowidTable(Parse *pParse, Table *pTab){ + Index *pIdx; + Index *pPk; + int nPk; + int nExtra; + int i, j; + sqlite3 *db = pParse->db; + Vdbe *v = pParse->pVdbe; + + /* Mark every PRIMARY KEY column as NOT NULL (except for imposter tables) + */ + if( !db->init.imposterTable ){ + for(i=0; inCol; i++){ + if( (pTab->aCol[i].colFlags & COLFLAG_PRIMKEY)!=0 ){ + pTab->aCol[i].notNull = OE_Abort; + } + } + } + + /* Convert the P3 operand of the OP_CreateBtree opcode from BTREE_INTKEY + ** into BTREE_BLOBKEY. + */ + if( pParse->addrCrTab ){ + assert( v ); + sqlite3VdbeChangeP3(v, pParse->addrCrTab, BTREE_BLOBKEY); + } + + /* Locate the PRIMARY KEY index. Or, if this table was originally + ** an INTEGER PRIMARY KEY table, create a new PRIMARY KEY index. + */ + if( pTab->iPKey>=0 ){ + ExprList *pList; + Token ipkToken; + sqlite3TokenInit(&ipkToken, pTab->aCol[pTab->iPKey].zName); + pList = sqlite3ExprListAppend(pParse, 0, + sqlite3ExprAlloc(db, TK_ID, &ipkToken, 0)); + if( pList==0 ) return; + if( IN_RENAME_OBJECT ){ + sqlite3RenameTokenRemap(pParse, pList->a[0].pExpr, &pTab->iPKey); + } + pList->a[0].sortFlags = pParse->iPkSortOrder; + assert( pParse->pNewTable==pTab ); + pTab->iPKey = -1; + sqlite3CreateIndex(pParse, 0, 0, 0, pList, pTab->keyConf, 0, 0, 0, 0, + SQLITE_IDXTYPE_PRIMARYKEY); + if( db->mallocFailed || pParse->nErr ) return; + pPk = sqlite3PrimaryKeyIndex(pTab); + assert( pPk->nKeyCol==1 ); + }else{ + pPk = sqlite3PrimaryKeyIndex(pTab); + assert( pPk!=0 ); + + /* + ** Remove all redundant columns from the PRIMARY KEY. For example, change + ** "PRIMARY KEY(a,b,a,b,c,b,c,d)" into just "PRIMARY KEY(a,b,c,d)". Later + ** code assumes the PRIMARY KEY contains no repeated columns. + */ + for(i=j=1; inKeyCol; i++){ + if( isDupColumn(pPk, j, pPk, i) ){ + pPk->nColumn--; + }else{ + testcase( hasColumn(pPk->aiColumn, j, pPk->aiColumn[i]) ); + pPk->azColl[j] = pPk->azColl[i]; + pPk->aSortOrder[j] = pPk->aSortOrder[i]; + pPk->aiColumn[j++] = pPk->aiColumn[i]; + } + } + pPk->nKeyCol = j; + } + assert( pPk!=0 ); + pPk->isCovering = 1; + if( !db->init.imposterTable ) pPk->uniqNotNull = 1; + nPk = pPk->nColumn = pPk->nKeyCol; + + /* Bypass the creation of the PRIMARY KEY btree and the sqlite_master + ** table entry. This is only required if currently generating VDBE + ** code for a CREATE TABLE (not when parsing one as part of reading + ** a database schema). */ + if( v && pPk->tnum>0 ){ + assert( db->init.busy==0 ); + sqlite3VdbeChangeOpcode(v, pPk->tnum, OP_Goto); + } + + /* The root page of the PRIMARY KEY is the table root page */ + pPk->tnum = pTab->tnum; + + /* Update the in-memory representation of all UNIQUE indices by converting + ** the final rowid column into one or more columns of the PRIMARY KEY. + */ + for(pIdx=pTab->pIndex; pIdx; pIdx=pIdx->pNext){ + int n; + if( IsPrimaryKeyIndex(pIdx) ) continue; + for(i=n=0; inKeyCol, pPk, i) ){ + testcase( hasColumn(pIdx->aiColumn, pIdx->nKeyCol, pPk->aiColumn[i]) ); + n++; + } + } + if( n==0 ){ + /* This index is a superset of the primary key */ + pIdx->nColumn = pIdx->nKeyCol; + continue; + } + if( resizeIndexObject(db, pIdx, pIdx->nKeyCol+n) ) return; + for(i=0, j=pIdx->nKeyCol; inKeyCol, pPk, i) ){ + testcase( hasColumn(pIdx->aiColumn, pIdx->nKeyCol, pPk->aiColumn[i]) ); + pIdx->aiColumn[j] = pPk->aiColumn[i]; + pIdx->azColl[j] = pPk->azColl[i]; + if( pPk->aSortOrder[i] ){ + /* See ticket https://www.sqlite.org/src/info/bba7b69f9849b5bf */ + pIdx->bAscKeyBug = 1; + } + j++; + } + } + assert( pIdx->nColumn>=pIdx->nKeyCol+n ); + assert( pIdx->nColumn>=j ); + } + + /* Add all table columns to the PRIMARY KEY index + */ + nExtra = 0; + for(i=0; inCol; i++){ + if( !hasColumn(pPk->aiColumn, nPk, i) ) nExtra++; + } + if( resizeIndexObject(db, pPk, nPk+nExtra) ) return; + for(i=0, j=nPk; inCol; i++){ + if( !hasColumn(pPk->aiColumn, j, i) ){ + assert( jnColumn ); + pPk->aiColumn[j] = i; + pPk->azColl[j] = sqlite3StrBINARY; + j++; + } + } + assert( pPk->nColumn==j ); + assert( pTab->nCol<=j ); + recomputeColumnsNotIndexed(pPk); +} + +#ifndef SQLITE_OMIT_VIRTUALTABLE +/* +** Return true if zName is a shadow table name in the current database +** connection. +** +** zName is temporarily modified while this routine is running, but is +** restored to its original value prior to this routine returning. +*/ +static int isShadowTableName(sqlite3 *db, char *zName){ + char *zTail; /* Pointer to the last "_" in zName */ + Table *pTab; /* Table that zName is a shadow of */ + Module *pMod; /* Module for the virtual table */ + + zTail = strrchr(zName, '_'); + if( zTail==0 ) return 0; + *zTail = 0; + pTab = sqlite3FindTable(db, zName, 0); + *zTail = '_'; + if( pTab==0 ) return 0; + if( !IsVirtual(pTab) ) return 0; + pMod = (Module*)sqlite3HashFind(&db->aModule, pTab->azModuleArg[0]); + if( pMod==0 ) return 0; + if( pMod->pModule->iVersion<3 ) return 0; + if( pMod->pModule->xShadowName==0 ) return 0; + return pMod->pModule->xShadowName(zTail+1); +} +#else +# define isShadowTableName(x,y) 0 +#endif /* ifndef SQLITE_OMIT_VIRTUALTABLE */ + +/* +** This routine is called to report the final ")" that terminates +** a CREATE TABLE statement. +** +** The table structure that other action routines have been building +** is added to the internal hash tables, assuming no errors have +** occurred. +** +** An entry for the table is made in the master table on disk, unless +** this is a temporary table or db->init.busy==1. When db->init.busy==1 +** it means we are reading the sqlite_master table because we just +** connected to the database or because the sqlite_master table has +** recently changed, so the entry for this table already exists in +** the sqlite_master table. We do not want to create it again. +** +** If the pSelect argument is not NULL, it means that this routine +** was called to create a table generated from a +** "CREATE TABLE ... AS SELECT ..." statement. The column names of +** the new table will match the result set of the SELECT. +*/ +SQLITE_PRIVATE void sqlite3EndTable( + Parse *pParse, /* Parse context */ + Token *pCons, /* The ',' token after the last column defn. */ + Token *pEnd, /* The ')' before options in the CREATE TABLE */ + u8 tabOpts, /* Extra table options. Usually 0. */ + Select *pSelect /* Select from a "CREATE ... AS SELECT" */ +){ + Table *p; /* The new table */ + sqlite3 *db = pParse->db; /* The database connection */ + int iDb; /* Database in which the table lives */ + Index *pIdx; /* An implied index of the table */ + + if( pEnd==0 && pSelect==0 ){ + return; + } + assert( !db->mallocFailed ); + p = pParse->pNewTable; + if( p==0 ) return; + + if( pSelect==0 && isShadowTableName(db, p->zName) ){ + p->tabFlags |= TF_Shadow; + } + + /* If the db->init.busy is 1 it means we are reading the SQL off the + ** "sqlite_master" or "sqlite_temp_master" table on the disk. + ** So do not write to the disk again. Extract the root page number + ** for the table from the db->init.newTnum field. (The page number + ** should have been put there by the sqliteOpenCb routine.) + ** + ** If the root page number is 1, that means this is the sqlite_master + ** table itself. So mark it read-only. + */ + if( db->init.busy ){ + if( pSelect ){ + sqlite3ErrorMsg(pParse, ""); + return; + } + p->tnum = db->init.newTnum; + if( p->tnum==1 ) p->tabFlags |= TF_Readonly; + } + + assert( (p->tabFlags & TF_HasPrimaryKey)==0 + || p->iPKey>=0 || sqlite3PrimaryKeyIndex(p)!=0 ); + assert( (p->tabFlags & TF_HasPrimaryKey)!=0 + || (p->iPKey<0 && sqlite3PrimaryKeyIndex(p)==0) ); + + /* Special processing for WITHOUT ROWID Tables */ + if( tabOpts & TF_WithoutRowid ){ + if( (p->tabFlags & TF_Autoincrement) ){ + sqlite3ErrorMsg(pParse, + "AUTOINCREMENT not allowed on WITHOUT ROWID tables"); + return; + } + if( (p->tabFlags & TF_HasPrimaryKey)==0 ){ + sqlite3ErrorMsg(pParse, "PRIMARY KEY missing on table %s", p->zName); + }else{ + p->tabFlags |= TF_WithoutRowid | TF_NoVisibleRowid; + convertToWithoutRowidTable(pParse, p); + } + } + + iDb = sqlite3SchemaToIndex(db, p->pSchema); + +#ifndef SQLITE_OMIT_CHECK + /* Resolve names in all CHECK constraint expressions. + */ + if( p->pCheck ){ + sqlite3ResolveSelfReference(pParse, p, NC_IsCheck, 0, p->pCheck); + } +#endif /* !defined(SQLITE_OMIT_CHECK) */ + + /* Estimate the average row size for the table and for all implied indices */ + estimateTableWidth(p); + for(pIdx=p->pIndex; pIdx; pIdx=pIdx->pNext){ + estimateIndexWidth(pIdx); + } + + /* If not initializing, then create a record for the new table + ** in the SQLITE_MASTER table of the database. + ** + ** If this is a TEMPORARY table, write the entry into the auxiliary + ** file instead of into the main database file. + */ + if( !db->init.busy ){ + int n; + Vdbe *v; + char *zType; /* "view" or "table" */ + char *zType2; /* "VIEW" or "TABLE" */ + char *zStmt; /* Text of the CREATE TABLE or CREATE VIEW statement */ + + v = sqlite3GetVdbe(pParse); + if( NEVER(v==0) ) return; + + sqlite3VdbeAddOp1(v, OP_Close, 0); + + /* + ** Initialize zType for the new view or table. + */ + if( p->pSelect==0 ){ + /* A regular table */ + zType = "table"; + zType2 = "TABLE"; +#ifndef SQLITE_OMIT_VIEW + }else{ + /* A view */ + zType = "view"; + zType2 = "VIEW"; +#endif + } + + /* If this is a CREATE TABLE xx AS SELECT ..., execute the SELECT + ** statement to populate the new table. The root-page number for the + ** new table is in register pParse->regRoot. + ** + ** Once the SELECT has been coded by sqlite3Select(), it is in a + ** suitable state to query for the column names and types to be used + ** by the new table. + ** + ** A shared-cache write-lock is not required to write to the new table, + ** as a schema-lock must have already been obtained to create it. Since + ** a schema-lock excludes all other database users, the write-lock would + ** be redundant. + */ + if( pSelect ){ + SelectDest dest; /* Where the SELECT should store results */ + int regYield; /* Register holding co-routine entry-point */ + int addrTop; /* Top of the co-routine */ + int regRec; /* A record to be insert into the new table */ + int regRowid; /* Rowid of the next row to insert */ + int addrInsLoop; /* Top of the loop for inserting rows */ + Table *pSelTab; /* A table that describes the SELECT results */ + + regYield = ++pParse->nMem; + regRec = ++pParse->nMem; + regRowid = ++pParse->nMem; + assert(pParse->nTab==1); + sqlite3MayAbort(pParse); + sqlite3VdbeAddOp3(v, OP_OpenWrite, 1, pParse->regRoot, iDb); + sqlite3VdbeChangeP5(v, OPFLAG_P2ISREG); + pParse->nTab = 2; + addrTop = sqlite3VdbeCurrentAddr(v) + 1; + sqlite3VdbeAddOp3(v, OP_InitCoroutine, regYield, 0, addrTop); + if( pParse->nErr ) return; + pSelTab = sqlite3ResultSetOfSelect(pParse, pSelect, SQLITE_AFF_BLOB); + if( pSelTab==0 ) return; + assert( p->aCol==0 ); + p->nCol = pSelTab->nCol; + p->aCol = pSelTab->aCol; + pSelTab->nCol = 0; + pSelTab->aCol = 0; + sqlite3DeleteTable(db, pSelTab); + sqlite3SelectDestInit(&dest, SRT_Coroutine, regYield); + sqlite3Select(pParse, pSelect, &dest); + if( pParse->nErr ) return; + sqlite3VdbeEndCoroutine(v, regYield); + sqlite3VdbeJumpHere(v, addrTop - 1); + addrInsLoop = sqlite3VdbeAddOp1(v, OP_Yield, dest.iSDParm); + VdbeCoverage(v); + sqlite3VdbeAddOp3(v, OP_MakeRecord, dest.iSdst, dest.nSdst, regRec); + sqlite3TableAffinity(v, p, 0); + sqlite3VdbeAddOp2(v, OP_NewRowid, 1, regRowid); + sqlite3VdbeAddOp3(v, OP_Insert, 1, regRec, regRowid); + sqlite3VdbeGoto(v, addrInsLoop); + sqlite3VdbeJumpHere(v, addrInsLoop); + sqlite3VdbeAddOp1(v, OP_Close, 1); + } + + /* Compute the complete text of the CREATE statement */ + if( pSelect ){ + zStmt = createTableStmt(db, p); + }else{ + Token *pEnd2 = tabOpts ? &pParse->sLastToken : pEnd; + n = (int)(pEnd2->z - pParse->sNameToken.z); + if( pEnd2->z[0]!=';' ) n += pEnd2->n; + zStmt = sqlite3MPrintf(db, + "CREATE %s %.*s", zType2, n, pParse->sNameToken.z + ); + } + + /* A slot for the record has already been allocated in the + ** SQLITE_MASTER table. We just need to update that slot with all + ** the information we've collected. + */ + sqlite3NestedParse(pParse, + "UPDATE %Q.%s " + "SET type='%s', name=%Q, tbl_name=%Q, rootpage=#%d, sql=%Q " + "WHERE rowid=#%d", + db->aDb[iDb].zDbSName, MASTER_NAME, + zType, + p->zName, + p->zName, + pParse->regRoot, + zStmt, + pParse->regRowid + ); + sqlite3DbFree(db, zStmt); + sqlite3ChangeCookie(pParse, iDb); + +#ifndef SQLITE_OMIT_AUTOINCREMENT + /* Check to see if we need to create an sqlite_sequence table for + ** keeping track of autoincrement keys. + */ + if( (p->tabFlags & TF_Autoincrement)!=0 ){ + Db *pDb = &db->aDb[iDb]; + assert( sqlite3SchemaMutexHeld(db, iDb, 0) ); + if( pDb->pSchema->pSeqTab==0 ){ + sqlite3NestedParse(pParse, + "CREATE TABLE %Q.sqlite_sequence(name,seq)", + pDb->zDbSName + ); + } + } +#endif + + /* Reparse everything to update our internal data structures */ + sqlite3VdbeAddParseSchemaOp(v, iDb, + sqlite3MPrintf(db, "tbl_name='%q' AND type!='trigger'", p->zName)); + } + + + /* Add the table to the in-memory representation of the database. + */ + if( db->init.busy ){ + Table *pOld; + Schema *pSchema = p->pSchema; + assert( sqlite3SchemaMutexHeld(db, iDb, 0) ); + pOld = sqlite3HashInsert(&pSchema->tblHash, p->zName, p); + if( pOld ){ + assert( p==pOld ); /* Malloc must have failed inside HashInsert() */ + sqlite3OomFault(db); + return; + } + pParse->pNewTable = 0; + db->mDbFlags |= DBFLAG_SchemaChange; + +#ifndef SQLITE_OMIT_ALTERTABLE + if( !p->pSelect ){ + const char *zName = (const char *)pParse->sNameToken.z; + int nName; + assert( !pSelect && pCons && pEnd ); + if( pCons->z==0 ){ + pCons = pEnd; + } + nName = (int)((const char *)pCons->z - zName); + p->addColOffset = 13 + sqlite3Utf8CharLen(zName, nName); + } +#endif + } +} + +#ifndef SQLITE_OMIT_VIEW +/* +** The parser calls this routine in order to create a new VIEW +*/ +SQLITE_PRIVATE void sqlite3CreateView( + Parse *pParse, /* The parsing context */ + Token *pBegin, /* The CREATE token that begins the statement */ + Token *pName1, /* The token that holds the name of the view */ + Token *pName2, /* The token that holds the name of the view */ + ExprList *pCNames, /* Optional list of view column names */ + Select *pSelect, /* A SELECT statement that will become the new view */ + int isTemp, /* TRUE for a TEMPORARY view */ + int noErr /* Suppress error messages if VIEW already exists */ +){ + Table *p; + int n; + const char *z; + Token sEnd; + DbFixer sFix; + Token *pName = 0; + int iDb; + sqlite3 *db = pParse->db; + + if( pParse->nVar>0 ){ + sqlite3ErrorMsg(pParse, "parameters are not allowed in views"); + goto create_view_fail; + } + sqlite3StartTable(pParse, pName1, pName2, isTemp, 1, 0, noErr); + p = pParse->pNewTable; + if( p==0 || pParse->nErr ) goto create_view_fail; + sqlite3TwoPartName(pParse, pName1, pName2, &pName); + iDb = sqlite3SchemaToIndex(db, p->pSchema); + sqlite3FixInit(&sFix, pParse, iDb, "view", pName); + if( sqlite3FixSelect(&sFix, pSelect) ) goto create_view_fail; + + /* Make a copy of the entire SELECT statement that defines the view. + ** This will force all the Expr.token.z values to be dynamically + ** allocated rather than point to the input string - which means that + ** they will persist after the current sqlite3_exec() call returns. + */ + if( IN_RENAME_OBJECT ){ + p->pSelect = pSelect; + pSelect = 0; + }else{ + p->pSelect = sqlite3SelectDup(db, pSelect, EXPRDUP_REDUCE); + } + p->pCheck = sqlite3ExprListDup(db, pCNames, EXPRDUP_REDUCE); + if( db->mallocFailed ) goto create_view_fail; + + /* Locate the end of the CREATE VIEW statement. Make sEnd point to + ** the end. + */ + sEnd = pParse->sLastToken; + assert( sEnd.z[0]!=0 || sEnd.n==0 ); + if( sEnd.z[0]!=';' ){ + sEnd.z += sEnd.n; + } + sEnd.n = 0; + n = (int)(sEnd.z - pBegin->z); + assert( n>0 ); + z = pBegin->z; + while( sqlite3Isspace(z[n-1]) ){ n--; } + sEnd.z = &z[n-1]; + sEnd.n = 1; + + /* Use sqlite3EndTable() to add the view to the SQLITE_MASTER table */ + sqlite3EndTable(pParse, 0, &sEnd, 0, 0); + +create_view_fail: + sqlite3SelectDelete(db, pSelect); + if( IN_RENAME_OBJECT ){ + sqlite3RenameExprlistUnmap(pParse, pCNames); + } + sqlite3ExprListDelete(db, pCNames); + return; +} +#endif /* SQLITE_OMIT_VIEW */ + +#if !defined(SQLITE_OMIT_VIEW) || !defined(SQLITE_OMIT_VIRTUALTABLE) +/* +** The Table structure pTable is really a VIEW. Fill in the names of +** the columns of the view in the pTable structure. Return the number +** of errors. If an error is seen leave an error message in pParse->zErrMsg. +*/ +SQLITE_PRIVATE int sqlite3ViewGetColumnNames(Parse *pParse, Table *pTable){ + Table *pSelTab; /* A fake table from which we get the result set */ + Select *pSel; /* Copy of the SELECT that implements the view */ + int nErr = 0; /* Number of errors encountered */ + int n; /* Temporarily holds the number of cursors assigned */ + sqlite3 *db = pParse->db; /* Database connection for malloc errors */ +#ifndef SQLITE_OMIT_VIRTUALTABLE + int rc; +#endif +#ifndef SQLITE_OMIT_AUTHORIZATION + sqlite3_xauth xAuth; /* Saved xAuth pointer */ +#endif + + assert( pTable ); + +#ifndef SQLITE_OMIT_VIRTUALTABLE + db->nSchemaLock++; + rc = sqlite3VtabCallConnect(pParse, pTable); + db->nSchemaLock--; + if( rc ){ + return 1; + } + if( IsVirtual(pTable) ) return 0; +#endif + +#ifndef SQLITE_OMIT_VIEW + /* A positive nCol means the columns names for this view are + ** already known. + */ + if( pTable->nCol>0 ) return 0; + + /* A negative nCol is a special marker meaning that we are currently + ** trying to compute the column names. If we enter this routine with + ** a negative nCol, it means two or more views form a loop, like this: + ** + ** CREATE VIEW one AS SELECT * FROM two; + ** CREATE VIEW two AS SELECT * FROM one; + ** + ** Actually, the error above is now caught prior to reaching this point. + ** But the following test is still important as it does come up + ** in the following: + ** + ** CREATE TABLE main.ex1(a); + ** CREATE TEMP VIEW ex1 AS SELECT a FROM ex1; + ** SELECT * FROM temp.ex1; + */ + if( pTable->nCol<0 ){ + sqlite3ErrorMsg(pParse, "view %s is circularly defined", pTable->zName); + return 1; + } + assert( pTable->nCol>=0 ); + + /* If we get this far, it means we need to compute the table names. + ** Note that the call to sqlite3ResultSetOfSelect() will expand any + ** "*" elements in the results set of the view and will assign cursors + ** to the elements of the FROM clause. But we do not want these changes + ** to be permanent. So the computation is done on a copy of the SELECT + ** statement that defines the view. + */ + assert( pTable->pSelect ); + pSel = sqlite3SelectDup(db, pTable->pSelect, 0); + if( pSel ){ +#ifndef SQLITE_OMIT_ALTERTABLE + u8 eParseMode = pParse->eParseMode; + pParse->eParseMode = PARSE_MODE_NORMAL; +#endif + n = pParse->nTab; + sqlite3SrcListAssignCursors(pParse, pSel->pSrc); + pTable->nCol = -1; + db->lookaside.bDisable++; +#ifndef SQLITE_OMIT_AUTHORIZATION + xAuth = db->xAuth; + db->xAuth = 0; + pSelTab = sqlite3ResultSetOfSelect(pParse, pSel, SQLITE_AFF_NONE); + db->xAuth = xAuth; +#else + pSelTab = sqlite3ResultSetOfSelect(pParse, pSel, SQLITE_AFF_NONE); +#endif + pParse->nTab = n; + if( pTable->pCheck ){ + /* CREATE VIEW name(arglist) AS ... + ** The names of the columns in the table are taken from + ** arglist which is stored in pTable->pCheck. The pCheck field + ** normally holds CHECK constraints on an ordinary table, but for + ** a VIEW it holds the list of column names. + */ + sqlite3ColumnsFromExprList(pParse, pTable->pCheck, + &pTable->nCol, &pTable->aCol); + if( db->mallocFailed==0 + && pParse->nErr==0 + && pTable->nCol==pSel->pEList->nExpr + ){ + sqlite3SelectAddColumnTypeAndCollation(pParse, pTable, pSel, + SQLITE_AFF_NONE); + } + }else if( pSelTab ){ + /* CREATE VIEW name AS... without an argument list. Construct + ** the column names from the SELECT statement that defines the view. + */ + assert( pTable->aCol==0 ); + pTable->nCol = pSelTab->nCol; + pTable->aCol = pSelTab->aCol; + pSelTab->nCol = 0; + pSelTab->aCol = 0; + assert( sqlite3SchemaMutexHeld(db, 0, pTable->pSchema) ); + }else{ + pTable->nCol = 0; + nErr++; + } + sqlite3DeleteTable(db, pSelTab); + sqlite3SelectDelete(db, pSel); + db->lookaside.bDisable--; +#ifndef SQLITE_OMIT_ALTERTABLE + pParse->eParseMode = eParseMode; +#endif + } else { + nErr++; + } + pTable->pSchema->schemaFlags |= DB_UnresetViews; + if( db->mallocFailed ){ + sqlite3DeleteColumnNames(db, pTable); + pTable->aCol = 0; + pTable->nCol = 0; + } +#endif /* SQLITE_OMIT_VIEW */ + return nErr; +} +#endif /* !defined(SQLITE_OMIT_VIEW) || !defined(SQLITE_OMIT_VIRTUALTABLE) */ + +#ifndef SQLITE_OMIT_VIEW +/* +** Clear the column names from every VIEW in database idx. +*/ +static void sqliteViewResetAll(sqlite3 *db, int idx){ + HashElem *i; + assert( sqlite3SchemaMutexHeld(db, idx, 0) ); + if( !DbHasProperty(db, idx, DB_UnresetViews) ) return; + for(i=sqliteHashFirst(&db->aDb[idx].pSchema->tblHash); i;i=sqliteHashNext(i)){ + Table *pTab = sqliteHashData(i); + if( pTab->pSelect ){ + sqlite3DeleteColumnNames(db, pTab); + pTab->aCol = 0; + pTab->nCol = 0; + } + } + DbClearProperty(db, idx, DB_UnresetViews); +} +#else +# define sqliteViewResetAll(A,B) +#endif /* SQLITE_OMIT_VIEW */ + +/* +** This function is called by the VDBE to adjust the internal schema +** used by SQLite when the btree layer moves a table root page. The +** root-page of a table or index in database iDb has changed from iFrom +** to iTo. +** +** Ticket #1728: The symbol table might still contain information +** on tables and/or indices that are the process of being deleted. +** If you are unlucky, one of those deleted indices or tables might +** have the same rootpage number as the real table or index that is +** being moved. So we cannot stop searching after the first match +** because the first match might be for one of the deleted indices +** or tables and not the table/index that is actually being moved. +** We must continue looping until all tables and indices with +** rootpage==iFrom have been converted to have a rootpage of iTo +** in order to be certain that we got the right one. +*/ +#ifndef SQLITE_OMIT_AUTOVACUUM +SQLITE_PRIVATE void sqlite3RootPageMoved(sqlite3 *db, int iDb, int iFrom, int iTo){ + HashElem *pElem; + Hash *pHash; + Db *pDb; + + assert( sqlite3SchemaMutexHeld(db, iDb, 0) ); + pDb = &db->aDb[iDb]; + pHash = &pDb->pSchema->tblHash; + for(pElem=sqliteHashFirst(pHash); pElem; pElem=sqliteHashNext(pElem)){ + Table *pTab = sqliteHashData(pElem); + if( pTab->tnum==iFrom ){ + pTab->tnum = iTo; + } + } + pHash = &pDb->pSchema->idxHash; + for(pElem=sqliteHashFirst(pHash); pElem; pElem=sqliteHashNext(pElem)){ + Index *pIdx = sqliteHashData(pElem); + if( pIdx->tnum==iFrom ){ + pIdx->tnum = iTo; + } + } +} +#endif + +/* +** Write code to erase the table with root-page iTable from database iDb. +** Also write code to modify the sqlite_master table and internal schema +** if a root-page of another table is moved by the btree-layer whilst +** erasing iTable (this can happen with an auto-vacuum database). +*/ +static void destroyRootPage(Parse *pParse, int iTable, int iDb){ + Vdbe *v = sqlite3GetVdbe(pParse); + int r1 = sqlite3GetTempReg(pParse); + if( iTable<2 ) sqlite3ErrorMsg(pParse, "corrupt schema"); + sqlite3VdbeAddOp3(v, OP_Destroy, iTable, r1, iDb); + sqlite3MayAbort(pParse); +#ifndef SQLITE_OMIT_AUTOVACUUM + /* OP_Destroy stores an in integer r1. If this integer + ** is non-zero, then it is the root page number of a table moved to + ** location iTable. The following code modifies the sqlite_master table to + ** reflect this. + ** + ** The "#NNN" in the SQL is a special constant that means whatever value + ** is in register NNN. See grammar rules associated with the TK_REGISTER + ** token for additional information. + */ + sqlite3NestedParse(pParse, + "UPDATE %Q.%s SET rootpage=%d WHERE #%d AND rootpage=#%d", + pParse->db->aDb[iDb].zDbSName, MASTER_NAME, iTable, r1, r1); +#endif + sqlite3ReleaseTempReg(pParse, r1); +} + +/* +** Write VDBE code to erase table pTab and all associated indices on disk. +** Code to update the sqlite_master tables and internal schema definitions +** in case a root-page belonging to another table is moved by the btree layer +** is also added (this can happen with an auto-vacuum database). +*/ +static void destroyTable(Parse *pParse, Table *pTab){ + /* If the database may be auto-vacuum capable (if SQLITE_OMIT_AUTOVACUUM + ** is not defined), then it is important to call OP_Destroy on the + ** table and index root-pages in order, starting with the numerically + ** largest root-page number. This guarantees that none of the root-pages + ** to be destroyed is relocated by an earlier OP_Destroy. i.e. if the + ** following were coded: + ** + ** OP_Destroy 4 0 + ** ... + ** OP_Destroy 5 0 + ** + ** and root page 5 happened to be the largest root-page number in the + ** database, then root page 5 would be moved to page 4 by the + ** "OP_Destroy 4 0" opcode. The subsequent "OP_Destroy 5 0" would hit + ** a free-list page. + */ + int iTab = pTab->tnum; + int iDestroyed = 0; + + while( 1 ){ + Index *pIdx; + int iLargest = 0; + + if( iDestroyed==0 || iTabpIndex; pIdx; pIdx=pIdx->pNext){ + int iIdx = pIdx->tnum; + assert( pIdx->pSchema==pTab->pSchema ); + if( (iDestroyed==0 || (iIdxiLargest ){ + iLargest = iIdx; + } + } + if( iLargest==0 ){ + return; + }else{ + int iDb = sqlite3SchemaToIndex(pParse->db, pTab->pSchema); + assert( iDb>=0 && iDbdb->nDb ); + destroyRootPage(pParse, iLargest, iDb); + iDestroyed = iLargest; + } + } +} + +/* +** Remove entries from the sqlite_statN tables (for N in (1,2,3)) +** after a DROP INDEX or DROP TABLE command. +*/ +static void sqlite3ClearStatTables( + Parse *pParse, /* The parsing context */ + int iDb, /* The database number */ + const char *zType, /* "idx" or "tbl" */ + const char *zName /* Name of index or table */ +){ + int i; + const char *zDbName = pParse->db->aDb[iDb].zDbSName; + for(i=1; i<=4; i++){ + char zTab[24]; + sqlite3_snprintf(sizeof(zTab),zTab,"sqlite_stat%d",i); + if( sqlite3FindTable(pParse->db, zTab, zDbName) ){ + sqlite3NestedParse(pParse, + "DELETE FROM %Q.%s WHERE %s=%Q", + zDbName, zTab, zType, zName + ); + } + } +} + +/* +** Generate code to drop a table. +*/ +SQLITE_PRIVATE void sqlite3CodeDropTable(Parse *pParse, Table *pTab, int iDb, int isView){ + Vdbe *v; + sqlite3 *db = pParse->db; + Trigger *pTrigger; + Db *pDb = &db->aDb[iDb]; + + v = sqlite3GetVdbe(pParse); + assert( v!=0 ); + sqlite3BeginWriteOperation(pParse, 1, iDb); + +#ifndef SQLITE_OMIT_VIRTUALTABLE + if( IsVirtual(pTab) ){ + sqlite3VdbeAddOp0(v, OP_VBegin); + } +#endif + + /* Drop all triggers associated with the table being dropped. Code + ** is generated to remove entries from sqlite_master and/or + ** sqlite_temp_master if required. + */ + pTrigger = sqlite3TriggerList(pParse, pTab); + while( pTrigger ){ + assert( pTrigger->pSchema==pTab->pSchema || + pTrigger->pSchema==db->aDb[1].pSchema ); + sqlite3DropTriggerPtr(pParse, pTrigger); + pTrigger = pTrigger->pNext; + } + +#ifndef SQLITE_OMIT_AUTOINCREMENT + /* Remove any entries of the sqlite_sequence table associated with + ** the table being dropped. This is done before the table is dropped + ** at the btree level, in case the sqlite_sequence table needs to + ** move as a result of the drop (can happen in auto-vacuum mode). + */ + if( pTab->tabFlags & TF_Autoincrement ){ + sqlite3NestedParse(pParse, + "DELETE FROM %Q.sqlite_sequence WHERE name=%Q", + pDb->zDbSName, pTab->zName + ); + } +#endif + + /* Drop all SQLITE_MASTER table and index entries that refer to the + ** table. The program name loops through the master table and deletes + ** every row that refers to a table of the same name as the one being + ** dropped. Triggers are handled separately because a trigger can be + ** created in the temp database that refers to a table in another + ** database. + */ + sqlite3NestedParse(pParse, + "DELETE FROM %Q.%s WHERE tbl_name=%Q and type!='trigger'", + pDb->zDbSName, MASTER_NAME, pTab->zName); + if( !isView && !IsVirtual(pTab) ){ + destroyTable(pParse, pTab); + } + + /* Remove the table entry from SQLite's internal schema and modify + ** the schema cookie. + */ + if( IsVirtual(pTab) ){ + sqlite3VdbeAddOp4(v, OP_VDestroy, iDb, 0, 0, pTab->zName, 0); + sqlite3MayAbort(pParse); + } + sqlite3VdbeAddOp4(v, OP_DropTable, iDb, 0, 0, pTab->zName, 0); + sqlite3ChangeCookie(pParse, iDb); + sqliteViewResetAll(db, iDb); +} + +/* +** This routine is called to do the work of a DROP TABLE statement. +** pName is the name of the table to be dropped. +*/ +SQLITE_PRIVATE void sqlite3DropTable(Parse *pParse, SrcList *pName, int isView, int noErr){ + Table *pTab; + Vdbe *v; + sqlite3 *db = pParse->db; + int iDb; + + if( db->mallocFailed ){ + goto exit_drop_table; + } + assert( pParse->nErr==0 ); + assert( pName->nSrc==1 ); + if( sqlite3ReadSchema(pParse) ) goto exit_drop_table; + if( noErr ) db->suppressErr++; + assert( isView==0 || isView==LOCATE_VIEW ); + pTab = sqlite3LocateTableItem(pParse, isView, &pName->a[0]); + if( noErr ) db->suppressErr--; + + if( pTab==0 ){ + if( noErr ) sqlite3CodeVerifyNamedSchema(pParse, pName->a[0].zDatabase); + goto exit_drop_table; + } + iDb = sqlite3SchemaToIndex(db, pTab->pSchema); + assert( iDb>=0 && iDbnDb ); + + /* If pTab is a virtual table, call ViewGetColumnNames() to ensure + ** it is initialized. + */ + if( IsVirtual(pTab) && sqlite3ViewGetColumnNames(pParse, pTab) ){ + goto exit_drop_table; + } +#ifndef SQLITE_OMIT_AUTHORIZATION + { + int code; + const char *zTab = SCHEMA_TABLE(iDb); + const char *zDb = db->aDb[iDb].zDbSName; + const char *zArg2 = 0; + if( sqlite3AuthCheck(pParse, SQLITE_DELETE, zTab, 0, zDb)){ + goto exit_drop_table; + } + if( isView ){ + if( !OMIT_TEMPDB && iDb==1 ){ + code = SQLITE_DROP_TEMP_VIEW; + }else{ + code = SQLITE_DROP_VIEW; + } +#ifndef SQLITE_OMIT_VIRTUALTABLE + }else if( IsVirtual(pTab) ){ + code = SQLITE_DROP_VTABLE; + zArg2 = sqlite3GetVTable(db, pTab)->pMod->zName; +#endif + }else{ + if( !OMIT_TEMPDB && iDb==1 ){ + code = SQLITE_DROP_TEMP_TABLE; + }else{ + code = SQLITE_DROP_TABLE; + } + } + if( sqlite3AuthCheck(pParse, code, pTab->zName, zArg2, zDb) ){ + goto exit_drop_table; + } + if( sqlite3AuthCheck(pParse, SQLITE_DELETE, pTab->zName, 0, zDb) ){ + goto exit_drop_table; + } + } +#endif + if( sqlite3StrNICmp(pTab->zName, "sqlite_", 7)==0 + && sqlite3StrNICmp(pTab->zName+7, "stat", 4)!=0 + && sqlite3StrNICmp(pTab->zName+7, "parameters", 10)!=0 ){ + sqlite3ErrorMsg(pParse, "table %s may not be dropped", pTab->zName); + goto exit_drop_table; + } + +#ifndef SQLITE_OMIT_VIEW + /* Ensure DROP TABLE is not used on a view, and DROP VIEW is not used + ** on a table. + */ + if( isView && pTab->pSelect==0 ){ + sqlite3ErrorMsg(pParse, "use DROP TABLE to delete table %s", pTab->zName); + goto exit_drop_table; + } + if( !isView && pTab->pSelect ){ + sqlite3ErrorMsg(pParse, "use DROP VIEW to delete view %s", pTab->zName); + goto exit_drop_table; + } +#endif + + /* Generate code to remove the table from the master table + ** on disk. + */ + v = sqlite3GetVdbe(pParse); + if( v ){ + sqlite3BeginWriteOperation(pParse, 1, iDb); + if( !isView ){ + sqlite3ClearStatTables(pParse, iDb, "tbl", pTab->zName); + sqlite3FkDropTable(pParse, pName, pTab); + } + sqlite3CodeDropTable(pParse, pTab, iDb, isView); + } + +exit_drop_table: + sqlite3SrcListDelete(db, pName); +} + +/* +** This routine is called to create a new foreign key on the table +** currently under construction. pFromCol determines which columns +** in the current table point to the foreign key. If pFromCol==0 then +** connect the key to the last column inserted. pTo is the name of +** the table referred to (a.k.a the "parent" table). pToCol is a list +** of tables in the parent pTo table. flags contains all +** information about the conflict resolution algorithms specified +** in the ON DELETE, ON UPDATE and ON INSERT clauses. +** +** An FKey structure is created and added to the table currently +** under construction in the pParse->pNewTable field. +** +** The foreign key is set for IMMEDIATE processing. A subsequent call +** to sqlite3DeferForeignKey() might change this to DEFERRED. +*/ +SQLITE_PRIVATE void sqlite3CreateForeignKey( + Parse *pParse, /* Parsing context */ + ExprList *pFromCol, /* Columns in this table that point to other table */ + Token *pTo, /* Name of the other table */ + ExprList *pToCol, /* Columns in the other table */ + int flags /* Conflict resolution algorithms. */ +){ + sqlite3 *db = pParse->db; +#ifndef SQLITE_OMIT_FOREIGN_KEY + FKey *pFKey = 0; + FKey *pNextTo; + Table *p = pParse->pNewTable; + int nByte; + int i; + int nCol; + char *z; + + assert( pTo!=0 ); + if( p==0 || IN_DECLARE_VTAB ) goto fk_end; + if( pFromCol==0 ){ + int iCol = p->nCol-1; + if( NEVER(iCol<0) ) goto fk_end; + if( pToCol && pToCol->nExpr!=1 ){ + sqlite3ErrorMsg(pParse, "foreign key on %s" + " should reference only one column of table %T", + p->aCol[iCol].zName, pTo); + goto fk_end; + } + nCol = 1; + }else if( pToCol && pToCol->nExpr!=pFromCol->nExpr ){ + sqlite3ErrorMsg(pParse, + "number of columns in foreign key does not match the number of " + "columns in the referenced table"); + goto fk_end; + }else{ + nCol = pFromCol->nExpr; + } + nByte = sizeof(*pFKey) + (nCol-1)*sizeof(pFKey->aCol[0]) + pTo->n + 1; + if( pToCol ){ + for(i=0; inExpr; i++){ + nByte += sqlite3Strlen30(pToCol->a[i].zName) + 1; + } + } + pFKey = sqlite3DbMallocZero(db, nByte ); + if( pFKey==0 ){ + goto fk_end; + } + pFKey->pFrom = p; + pFKey->pNextFrom = p->pFKey; + z = (char*)&pFKey->aCol[nCol]; + pFKey->zTo = z; + if( IN_RENAME_OBJECT ){ + sqlite3RenameTokenMap(pParse, (void*)z, pTo); + } + memcpy(z, pTo->z, pTo->n); + z[pTo->n] = 0; + sqlite3Dequote(z); + z += pTo->n+1; + pFKey->nCol = nCol; + if( pFromCol==0 ){ + pFKey->aCol[0].iFrom = p->nCol-1; + }else{ + for(i=0; inCol; j++){ + if( sqlite3StrICmp(p->aCol[j].zName, pFromCol->a[i].zName)==0 ){ + pFKey->aCol[i].iFrom = j; + break; + } + } + if( j>=p->nCol ){ + sqlite3ErrorMsg(pParse, + "unknown column \"%s\" in foreign key definition", + pFromCol->a[i].zName); + goto fk_end; + } + if( IN_RENAME_OBJECT ){ + sqlite3RenameTokenRemap(pParse, &pFKey->aCol[i], pFromCol->a[i].zName); + } + } + } + if( pToCol ){ + for(i=0; ia[i].zName); + pFKey->aCol[i].zCol = z; + if( IN_RENAME_OBJECT ){ + sqlite3RenameTokenRemap(pParse, z, pToCol->a[i].zName); + } + memcpy(z, pToCol->a[i].zName, n); + z[n] = 0; + z += n+1; + } + } + pFKey->isDeferred = 0; + pFKey->aAction[0] = (u8)(flags & 0xff); /* ON DELETE action */ + pFKey->aAction[1] = (u8)((flags >> 8 ) & 0xff); /* ON UPDATE action */ + + assert( sqlite3SchemaMutexHeld(db, 0, p->pSchema) ); + pNextTo = (FKey *)sqlite3HashInsert(&p->pSchema->fkeyHash, + pFKey->zTo, (void *)pFKey + ); + if( pNextTo==pFKey ){ + sqlite3OomFault(db); + goto fk_end; + } + if( pNextTo ){ + assert( pNextTo->pPrevTo==0 ); + pFKey->pNextTo = pNextTo; + pNextTo->pPrevTo = pFKey; + } + + /* Link the foreign key to the table as the last step. + */ + p->pFKey = pFKey; + pFKey = 0; + +fk_end: + sqlite3DbFree(db, pFKey); +#endif /* !defined(SQLITE_OMIT_FOREIGN_KEY) */ + sqlite3ExprListDelete(db, pFromCol); + sqlite3ExprListDelete(db, pToCol); +} + +/* +** This routine is called when an INITIALLY IMMEDIATE or INITIALLY DEFERRED +** clause is seen as part of a foreign key definition. The isDeferred +** parameter is 1 for INITIALLY DEFERRED and 0 for INITIALLY IMMEDIATE. +** The behavior of the most recently created foreign key is adjusted +** accordingly. +*/ +SQLITE_PRIVATE void sqlite3DeferForeignKey(Parse *pParse, int isDeferred){ +#ifndef SQLITE_OMIT_FOREIGN_KEY + Table *pTab; + FKey *pFKey; + if( (pTab = pParse->pNewTable)==0 || (pFKey = pTab->pFKey)==0 ) return; + assert( isDeferred==0 || isDeferred==1 ); /* EV: R-30323-21917 */ + pFKey->isDeferred = (u8)isDeferred; +#endif +} + +/* +** Generate code that will erase and refill index *pIdx. This is +** used to initialize a newly created index or to recompute the +** content of an index in response to a REINDEX command. +** +** if memRootPage is not negative, it means that the index is newly +** created. The register specified by memRootPage contains the +** root page number of the index. If memRootPage is negative, then +** the index already exists and must be cleared before being refilled and +** the root page number of the index is taken from pIndex->tnum. +*/ +static void sqlite3RefillIndex(Parse *pParse, Index *pIndex, int memRootPage){ + Table *pTab = pIndex->pTable; /* The table that is indexed */ + int iTab = pParse->nTab++; /* Btree cursor used for pTab */ + int iIdx = pParse->nTab++; /* Btree cursor used for pIndex */ + int iSorter; /* Cursor opened by OpenSorter (if in use) */ + int addr1; /* Address of top of loop */ + int addr2; /* Address to jump to for next iteration */ + int tnum; /* Root page of index */ + int iPartIdxLabel; /* Jump to this label to skip a row */ + Vdbe *v; /* Generate code into this virtual machine */ + KeyInfo *pKey; /* KeyInfo for index */ + int regRecord; /* Register holding assembled index record */ + sqlite3 *db = pParse->db; /* The database connection */ + int iDb = sqlite3SchemaToIndex(db, pIndex->pSchema); + +#ifndef SQLITE_OMIT_AUTHORIZATION + if( sqlite3AuthCheck(pParse, SQLITE_REINDEX, pIndex->zName, 0, + db->aDb[iDb].zDbSName ) ){ + return; + } +#endif + + /* Require a write-lock on the table to perform this operation */ + sqlite3TableLock(pParse, iDb, pTab->tnum, 1, pTab->zName); + + v = sqlite3GetVdbe(pParse); + if( v==0 ) return; + if( memRootPage>=0 ){ + tnum = memRootPage; + }else{ + tnum = pIndex->tnum; + } + pKey = sqlite3KeyInfoOfIndex(pParse, pIndex); + assert( pKey!=0 || db->mallocFailed || pParse->nErr ); + + /* Open the sorter cursor if we are to use one. */ + iSorter = pParse->nTab++; + sqlite3VdbeAddOp4(v, OP_SorterOpen, iSorter, 0, pIndex->nKeyCol, (char*) + sqlite3KeyInfoRef(pKey), P4_KEYINFO); + + /* Open the table. Loop through all rows of the table, inserting index + ** records into the sorter. */ + sqlite3OpenTable(pParse, iTab, iDb, pTab, OP_OpenRead); + addr1 = sqlite3VdbeAddOp2(v, OP_Rewind, iTab, 0); VdbeCoverage(v); + regRecord = sqlite3GetTempReg(pParse); + sqlite3MultiWrite(pParse); + + sqlite3GenerateIndexKey(pParse,pIndex,iTab,regRecord,0,&iPartIdxLabel,0,0); + sqlite3VdbeAddOp2(v, OP_SorterInsert, iSorter, regRecord); + sqlite3ResolvePartIdxLabel(pParse, iPartIdxLabel); + sqlite3VdbeAddOp2(v, OP_Next, iTab, addr1+1); VdbeCoverage(v); + sqlite3VdbeJumpHere(v, addr1); + if( memRootPage<0 ) sqlite3VdbeAddOp2(v, OP_Clear, tnum, iDb); + sqlite3VdbeAddOp4(v, OP_OpenWrite, iIdx, tnum, iDb, + (char *)pKey, P4_KEYINFO); + sqlite3VdbeChangeP5(v, OPFLAG_BULKCSR|((memRootPage>=0)?OPFLAG_P2ISREG:0)); + + addr1 = sqlite3VdbeAddOp2(v, OP_SorterSort, iSorter, 0); VdbeCoverage(v); + if( IsUniqueIndex(pIndex) ){ + int j2 = sqlite3VdbeGoto(v, 1); + addr2 = sqlite3VdbeCurrentAddr(v); + sqlite3VdbeVerifyAbortable(v, OE_Abort); + sqlite3VdbeAddOp4Int(v, OP_SorterCompare, iSorter, j2, regRecord, + pIndex->nKeyCol); VdbeCoverage(v); + sqlite3UniqueConstraint(pParse, OE_Abort, pIndex); + sqlite3VdbeJumpHere(v, j2); + }else{ + /* Most CREATE INDEX and REINDEX statements that are not UNIQUE can not + ** abort. The exception is if one of the indexed expressions contains a + ** user function that throws an exception when it is evaluated. But the + ** overhead of adding a statement journal to a CREATE INDEX statement is + ** very small (since most of the pages written do not contain content that + ** needs to be restored if the statement aborts), so we call + ** sqlite3MayAbort() for all CREATE INDEX statements. */ + sqlite3MayAbort(pParse); + addr2 = sqlite3VdbeCurrentAddr(v); + } + sqlite3VdbeAddOp3(v, OP_SorterData, iSorter, regRecord, iIdx); + if( !pIndex->bAscKeyBug ){ + /* This OP_SeekEnd opcode makes index insert for a REINDEX go much + ** faster by avoiding unnecessary seeks. But the optimization does + ** not work for UNIQUE constraint indexes on WITHOUT ROWID tables + ** with DESC primary keys, since those indexes have there keys in + ** a different order from the main table. + ** See ticket: https://www.sqlite.org/src/info/bba7b69f9849b5bf + */ + sqlite3VdbeAddOp1(v, OP_SeekEnd, iIdx); + } + sqlite3VdbeAddOp2(v, OP_IdxInsert, iIdx, regRecord); + sqlite3VdbeChangeP5(v, OPFLAG_USESEEKRESULT); + sqlite3ReleaseTempReg(pParse, regRecord); + sqlite3VdbeAddOp2(v, OP_SorterNext, iSorter, addr2); VdbeCoverage(v); + sqlite3VdbeJumpHere(v, addr1); + + sqlite3VdbeAddOp1(v, OP_Close, iTab); + sqlite3VdbeAddOp1(v, OP_Close, iIdx); + sqlite3VdbeAddOp1(v, OP_Close, iSorter); +} + +/* +** Allocate heap space to hold an Index object with nCol columns. +** +** Increase the allocation size to provide an extra nExtra bytes +** of 8-byte aligned space after the Index object and return a +** pointer to this extra space in *ppExtra. +*/ +SQLITE_PRIVATE Index *sqlite3AllocateIndexObject( + sqlite3 *db, /* Database connection */ + i16 nCol, /* Total number of columns in the index */ + int nExtra, /* Number of bytes of extra space to alloc */ + char **ppExtra /* Pointer to the "extra" space */ +){ + Index *p; /* Allocated index object */ + int nByte; /* Bytes of space for Index object + arrays */ + + nByte = ROUND8(sizeof(Index)) + /* Index structure */ + ROUND8(sizeof(char*)*nCol) + /* Index.azColl */ + ROUND8(sizeof(LogEst)*(nCol+1) + /* Index.aiRowLogEst */ + sizeof(i16)*nCol + /* Index.aiColumn */ + sizeof(u8)*nCol); /* Index.aSortOrder */ + p = sqlite3DbMallocZero(db, nByte + nExtra); + if( p ){ + char *pExtra = ((char*)p)+ROUND8(sizeof(Index)); + p->azColl = (const char**)pExtra; pExtra += ROUND8(sizeof(char*)*nCol); + p->aiRowLogEst = (LogEst*)pExtra; pExtra += sizeof(LogEst)*(nCol+1); + p->aiColumn = (i16*)pExtra; pExtra += sizeof(i16)*nCol; + p->aSortOrder = (u8*)pExtra; + p->nColumn = nCol; + p->nKeyCol = nCol - 1; + *ppExtra = ((char*)p) + nByte; + } + return p; +} + +/* +** If expression list pList contains an expression that was parsed with +** an explicit "NULLS FIRST" or "NULLS LAST" clause, leave an error in +** pParse and return non-zero. Otherwise, return zero. +*/ +SQLITE_PRIVATE int sqlite3HasExplicitNulls(Parse *pParse, ExprList *pList){ + if( pList ){ + int i; + for(i=0; inExpr; i++){ + if( pList->a[i].bNulls ){ + u8 sf = pList->a[i].sortFlags; + sqlite3ErrorMsg(pParse, "unsupported use of NULLS %s", + (sf==0 || sf==3) ? "FIRST" : "LAST" + ); + return 1; + } + } + } + return 0; +} + +/* +** Create a new index for an SQL table. pName1.pName2 is the name of the index +** and pTblList is the name of the table that is to be indexed. Both will +** be NULL for a primary key or an index that is created to satisfy a +** UNIQUE constraint. If pTable and pIndex are NULL, use pParse->pNewTable +** as the table to be indexed. pParse->pNewTable is a table that is +** currently being constructed by a CREATE TABLE statement. +** +** pList is a list of columns to be indexed. pList will be NULL if this +** is a primary key or unique-constraint on the most recent column added +** to the table currently under construction. +*/ +SQLITE_PRIVATE void sqlite3CreateIndex( + Parse *pParse, /* All information about this parse */ + Token *pName1, /* First part of index name. May be NULL */ + Token *pName2, /* Second part of index name. May be NULL */ + SrcList *pTblName, /* Table to index. Use pParse->pNewTable if 0 */ + ExprList *pList, /* A list of columns to be indexed */ + int onError, /* OE_Abort, OE_Ignore, OE_Replace, or OE_None */ + Token *pStart, /* The CREATE token that begins this statement */ + Expr *pPIWhere, /* WHERE clause for partial indices */ + int sortOrder, /* Sort order of primary key when pList==NULL */ + int ifNotExist, /* Omit error if index already exists */ + u8 idxType /* The index type */ +){ + Table *pTab = 0; /* Table to be indexed */ + Index *pIndex = 0; /* The index to be created */ + char *zName = 0; /* Name of the index */ + int nName; /* Number of characters in zName */ + int i, j; + DbFixer sFix; /* For assigning database names to pTable */ + int sortOrderMask; /* 1 to honor DESC in index. 0 to ignore. */ + sqlite3 *db = pParse->db; + Db *pDb; /* The specific table containing the indexed database */ + int iDb; /* Index of the database that is being written */ + Token *pName = 0; /* Unqualified name of the index to create */ + struct ExprList_item *pListItem; /* For looping over pList */ + int nExtra = 0; /* Space allocated for zExtra[] */ + int nExtraCol; /* Number of extra columns needed */ + char *zExtra = 0; /* Extra space after the Index object */ + Index *pPk = 0; /* PRIMARY KEY index for WITHOUT ROWID tables */ + + if( db->mallocFailed || pParse->nErr>0 ){ + goto exit_create_index; + } + if( IN_DECLARE_VTAB && idxType!=SQLITE_IDXTYPE_PRIMARYKEY ){ + goto exit_create_index; + } + if( SQLITE_OK!=sqlite3ReadSchema(pParse) ){ + goto exit_create_index; + } + if( sqlite3HasExplicitNulls(pParse, pList) ){ + goto exit_create_index; + } + + /* + ** Find the table that is to be indexed. Return early if not found. + */ + if( pTblName!=0 ){ + + /* Use the two-part index name to determine the database + ** to search for the table. 'Fix' the table name to this db + ** before looking up the table. + */ + assert( pName1 && pName2 ); + iDb = sqlite3TwoPartName(pParse, pName1, pName2, &pName); + if( iDb<0 ) goto exit_create_index; + assert( pName && pName->z ); + +#ifndef SQLITE_OMIT_TEMPDB + /* If the index name was unqualified, check if the table + ** is a temp table. If so, set the database to 1. Do not do this + ** if initialising a database schema. + */ + if( !db->init.busy ){ + pTab = sqlite3SrcListLookup(pParse, pTblName); + if( pName2->n==0 && pTab && pTab->pSchema==db->aDb[1].pSchema ){ + iDb = 1; + } + } +#endif + + sqlite3FixInit(&sFix, pParse, iDb, "index", pName); + if( sqlite3FixSrcList(&sFix, pTblName) ){ + /* Because the parser constructs pTblName from a single identifier, + ** sqlite3FixSrcList can never fail. */ + assert(0); + } + pTab = sqlite3LocateTableItem(pParse, 0, &pTblName->a[0]); + assert( db->mallocFailed==0 || pTab==0 ); + if( pTab==0 ) goto exit_create_index; + if( iDb==1 && db->aDb[iDb].pSchema!=pTab->pSchema ){ + sqlite3ErrorMsg(pParse, + "cannot create a TEMP index on non-TEMP table \"%s\"", + pTab->zName); + goto exit_create_index; + } + if( !HasRowid(pTab) ) pPk = sqlite3PrimaryKeyIndex(pTab); + }else{ + assert( pName==0 ); + assert( pStart==0 ); + pTab = pParse->pNewTable; + if( !pTab ) goto exit_create_index; + iDb = sqlite3SchemaToIndex(db, pTab->pSchema); + } + pDb = &db->aDb[iDb]; + + assert( pTab!=0 ); + assert( pParse->nErr==0 ); + if( sqlite3StrNICmp(pTab->zName, "sqlite_", 7)==0 + && db->init.busy==0 + && pTblName!=0 +#if SQLITE_USER_AUTHENTICATION + && sqlite3UserAuthTable(pTab->zName)==0 +#endif +#ifdef SQLITE_ALLOW_SQLITE_MASTER_INDEX + && sqlite3StrICmp(&pTab->zName[7],"master")!=0 +#endif + ){ + sqlite3ErrorMsg(pParse, "table %s may not be indexed", pTab->zName); + goto exit_create_index; + } +#ifndef SQLITE_OMIT_VIEW + if( pTab->pSelect ){ + sqlite3ErrorMsg(pParse, "views may not be indexed"); + goto exit_create_index; + } +#endif +#ifndef SQLITE_OMIT_VIRTUALTABLE + if( IsVirtual(pTab) ){ + sqlite3ErrorMsg(pParse, "virtual tables may not be indexed"); + goto exit_create_index; + } +#endif + + /* + ** Find the name of the index. Make sure there is not already another + ** index or table with the same name. + ** + ** Exception: If we are reading the names of permanent indices from the + ** sqlite_master table (because some other process changed the schema) and + ** one of the index names collides with the name of a temporary table or + ** index, then we will continue to process this index. + ** + ** If pName==0 it means that we are + ** dealing with a primary key or UNIQUE constraint. We have to invent our + ** own name. + */ + if( pName ){ + zName = sqlite3NameFromToken(db, pName); + if( zName==0 ) goto exit_create_index; + assert( pName->z!=0 ); + if( SQLITE_OK!=sqlite3CheckObjectName(pParse, zName,"index",pTab->zName) ){ + goto exit_create_index; + } + if( !IN_RENAME_OBJECT ){ + if( !db->init.busy ){ + if( sqlite3FindTable(db, zName, 0)!=0 ){ + sqlite3ErrorMsg(pParse, "there is already a table named %s", zName); + goto exit_create_index; + } + } + if( sqlite3FindIndex(db, zName, pDb->zDbSName)!=0 ){ + if( !ifNotExist ){ + sqlite3ErrorMsg(pParse, "index %s already exists", zName); + }else{ + assert( !db->init.busy ); + sqlite3CodeVerifySchema(pParse, iDb); + } + goto exit_create_index; + } + } + }else{ + int n; + Index *pLoop; + for(pLoop=pTab->pIndex, n=1; pLoop; pLoop=pLoop->pNext, n++){} + zName = sqlite3MPrintf(db, "sqlite_autoindex_%s_%d", pTab->zName, n); + if( zName==0 ){ + goto exit_create_index; + } + + /* Automatic index names generated from within sqlite3_declare_vtab() + ** must have names that are distinct from normal automatic index names. + ** The following statement converts "sqlite3_autoindex..." into + ** "sqlite3_butoindex..." in order to make the names distinct. + ** The "vtab_err.test" test demonstrates the need of this statement. */ + if( IN_SPECIAL_PARSE ) zName[7]++; + } + + /* Check for authorization to create an index. + */ +#ifndef SQLITE_OMIT_AUTHORIZATION + if( !IN_RENAME_OBJECT ){ + const char *zDb = pDb->zDbSName; + if( sqlite3AuthCheck(pParse, SQLITE_INSERT, SCHEMA_TABLE(iDb), 0, zDb) ){ + goto exit_create_index; + } + i = SQLITE_CREATE_INDEX; + if( !OMIT_TEMPDB && iDb==1 ) i = SQLITE_CREATE_TEMP_INDEX; + if( sqlite3AuthCheck(pParse, i, zName, pTab->zName, zDb) ){ + goto exit_create_index; + } + } +#endif + + /* If pList==0, it means this routine was called to make a primary + ** key out of the last column added to the table under construction. + ** So create a fake list to simulate this. + */ + if( pList==0 ){ + Token prevCol; + Column *pCol = &pTab->aCol[pTab->nCol-1]; + pCol->colFlags |= COLFLAG_UNIQUE; + sqlite3TokenInit(&prevCol, pCol->zName); + pList = sqlite3ExprListAppend(pParse, 0, + sqlite3ExprAlloc(db, TK_ID, &prevCol, 0)); + if( pList==0 ) goto exit_create_index; + assert( pList->nExpr==1 ); + sqlite3ExprListSetSortOrder(pList, sortOrder, SQLITE_SO_UNDEFINED); + }else{ + sqlite3ExprListCheckLength(pParse, pList, "index"); + if( pParse->nErr ) goto exit_create_index; + } + + /* Figure out how many bytes of space are required to store explicitly + ** specified collation sequence names. + */ + for(i=0; inExpr; i++){ + Expr *pExpr = pList->a[i].pExpr; + assert( pExpr!=0 ); + if( pExpr->op==TK_COLLATE ){ + nExtra += (1 + sqlite3Strlen30(pExpr->u.zToken)); + } + } + + /* + ** Allocate the index structure. + */ + nName = sqlite3Strlen30(zName); + nExtraCol = pPk ? pPk->nKeyCol : 1; + assert( pList->nExpr + nExtraCol <= 32767 /* Fits in i16 */ ); + pIndex = sqlite3AllocateIndexObject(db, pList->nExpr + nExtraCol, + nName + nExtra + 1, &zExtra); + if( db->mallocFailed ){ + goto exit_create_index; + } + assert( EIGHT_BYTE_ALIGNMENT(pIndex->aiRowLogEst) ); + assert( EIGHT_BYTE_ALIGNMENT(pIndex->azColl) ); + pIndex->zName = zExtra; + zExtra += nName + 1; + memcpy(pIndex->zName, zName, nName+1); + pIndex->pTable = pTab; + pIndex->onError = (u8)onError; + pIndex->uniqNotNull = onError!=OE_None; + pIndex->idxType = idxType; + pIndex->pSchema = db->aDb[iDb].pSchema; + pIndex->nKeyCol = pList->nExpr; + if( pPIWhere ){ + sqlite3ResolveSelfReference(pParse, pTab, NC_PartIdx, pPIWhere, 0); + pIndex->pPartIdxWhere = pPIWhere; + pPIWhere = 0; + } + assert( sqlite3SchemaMutexHeld(db, iDb, 0) ); + + /* Check to see if we should honor DESC requests on index columns + */ + if( pDb->pSchema->file_format>=4 ){ + sortOrderMask = -1; /* Honor DESC */ + }else{ + sortOrderMask = 0; /* Ignore DESC */ + } + + /* Analyze the list of expressions that form the terms of the index and + ** report any errors. In the common case where the expression is exactly + ** a table column, store that column in aiColumn[]. For general expressions, + ** populate pIndex->aColExpr and store XN_EXPR (-2) in aiColumn[]. + ** + ** TODO: Issue a warning if two or more columns of the index are identical. + ** TODO: Issue a warning if the table primary key is used as part of the + ** index key. + */ + pListItem = pList->a; + if( IN_RENAME_OBJECT ){ + pIndex->aColExpr = pList; + pList = 0; + } + for(i=0; inKeyCol; i++, pListItem++){ + Expr *pCExpr; /* The i-th index expression */ + int requestedSortOrder; /* ASC or DESC on the i-th expression */ + const char *zColl; /* Collation sequence name */ + + sqlite3StringToId(pListItem->pExpr); + sqlite3ResolveSelfReference(pParse, pTab, NC_IdxExpr, pListItem->pExpr, 0); + if( pParse->nErr ) goto exit_create_index; + pCExpr = sqlite3ExprSkipCollate(pListItem->pExpr); + if( pCExpr->op!=TK_COLUMN ){ + if( pTab==pParse->pNewTable ){ + sqlite3ErrorMsg(pParse, "expressions prohibited in PRIMARY KEY and " + "UNIQUE constraints"); + goto exit_create_index; + } + if( pIndex->aColExpr==0 ){ + pIndex->aColExpr = pList; + pList = 0; + } + j = XN_EXPR; + pIndex->aiColumn[i] = XN_EXPR; + pIndex->uniqNotNull = 0; + }else{ + j = pCExpr->iColumn; + assert( j<=0x7fff ); + if( j<0 ){ + j = pTab->iPKey; + }else if( pTab->aCol[j].notNull==0 ){ + pIndex->uniqNotNull = 0; + } + pIndex->aiColumn[i] = (i16)j; + } + zColl = 0; + if( pListItem->pExpr->op==TK_COLLATE ){ + int nColl; + zColl = pListItem->pExpr->u.zToken; + nColl = sqlite3Strlen30(zColl) + 1; + assert( nExtra>=nColl ); + memcpy(zExtra, zColl, nColl); + zColl = zExtra; + zExtra += nColl; + nExtra -= nColl; + }else if( j>=0 ){ + zColl = pTab->aCol[j].zColl; + } + if( !zColl ) zColl = sqlite3StrBINARY; + if( !db->init.busy && !sqlite3LocateCollSeq(pParse, zColl) ){ + goto exit_create_index; + } + pIndex->azColl[i] = zColl; + requestedSortOrder = pListItem->sortFlags & sortOrderMask; + pIndex->aSortOrder[i] = (u8)requestedSortOrder; + } + + /* Append the table key to the end of the index. For WITHOUT ROWID + ** tables (when pPk!=0) this will be the declared PRIMARY KEY. For + ** normal tables (when pPk==0) this will be the rowid. + */ + if( pPk ){ + for(j=0; jnKeyCol; j++){ + int x = pPk->aiColumn[j]; + assert( x>=0 ); + if( isDupColumn(pIndex, pIndex->nKeyCol, pPk, j) ){ + pIndex->nColumn--; + }else{ + testcase( hasColumn(pIndex->aiColumn,pIndex->nKeyCol,x) ); + pIndex->aiColumn[i] = x; + pIndex->azColl[i] = pPk->azColl[j]; + pIndex->aSortOrder[i] = pPk->aSortOrder[j]; + i++; + } + } + assert( i==pIndex->nColumn ); + }else{ + pIndex->aiColumn[i] = XN_ROWID; + pIndex->azColl[i] = sqlite3StrBINARY; + } + sqlite3DefaultRowEst(pIndex); + if( pParse->pNewTable==0 ) estimateIndexWidth(pIndex); + + /* If this index contains every column of its table, then mark + ** it as a covering index */ + assert( HasRowid(pTab) + || pTab->iPKey<0 || sqlite3ColumnOfIndex(pIndex, pTab->iPKey)>=0 ); + recomputeColumnsNotIndexed(pIndex); + if( pTblName!=0 && pIndex->nColumn>=pTab->nCol ){ + pIndex->isCovering = 1; + for(j=0; jnCol; j++){ + if( j==pTab->iPKey ) continue; + if( sqlite3ColumnOfIndex(pIndex,j)>=0 ) continue; + pIndex->isCovering = 0; + break; + } + } + + if( pTab==pParse->pNewTable ){ + /* This routine has been called to create an automatic index as a + ** result of a PRIMARY KEY or UNIQUE clause on a column definition, or + ** a PRIMARY KEY or UNIQUE clause following the column definitions. + ** i.e. one of: + ** + ** CREATE TABLE t(x PRIMARY KEY, y); + ** CREATE TABLE t(x, y, UNIQUE(x, y)); + ** + ** Either way, check to see if the table already has such an index. If + ** so, don't bother creating this one. This only applies to + ** automatically created indices. Users can do as they wish with + ** explicit indices. + ** + ** Two UNIQUE or PRIMARY KEY constraints are considered equivalent + ** (and thus suppressing the second one) even if they have different + ** sort orders. + ** + ** If there are different collating sequences or if the columns of + ** the constraint occur in different orders, then the constraints are + ** considered distinct and both result in separate indices. + */ + Index *pIdx; + for(pIdx=pTab->pIndex; pIdx; pIdx=pIdx->pNext){ + int k; + assert( IsUniqueIndex(pIdx) ); + assert( pIdx->idxType!=SQLITE_IDXTYPE_APPDEF ); + assert( IsUniqueIndex(pIndex) ); + + if( pIdx->nKeyCol!=pIndex->nKeyCol ) continue; + for(k=0; knKeyCol; k++){ + const char *z1; + const char *z2; + assert( pIdx->aiColumn[k]>=0 ); + if( pIdx->aiColumn[k]!=pIndex->aiColumn[k] ) break; + z1 = pIdx->azColl[k]; + z2 = pIndex->azColl[k]; + if( sqlite3StrICmp(z1, z2) ) break; + } + if( k==pIdx->nKeyCol ){ + if( pIdx->onError!=pIndex->onError ){ + /* This constraint creates the same index as a previous + ** constraint specified somewhere in the CREATE TABLE statement. + ** However the ON CONFLICT clauses are different. If both this + ** constraint and the previous equivalent constraint have explicit + ** ON CONFLICT clauses this is an error. Otherwise, use the + ** explicitly specified behavior for the index. + */ + if( !(pIdx->onError==OE_Default || pIndex->onError==OE_Default) ){ + sqlite3ErrorMsg(pParse, + "conflicting ON CONFLICT clauses specified", 0); + } + if( pIdx->onError==OE_Default ){ + pIdx->onError = pIndex->onError; + } + } + if( idxType==SQLITE_IDXTYPE_PRIMARYKEY ) pIdx->idxType = idxType; + if( IN_RENAME_OBJECT ){ + pIndex->pNext = pParse->pNewIndex; + pParse->pNewIndex = pIndex; + pIndex = 0; + } + goto exit_create_index; + } + } + } + + if( !IN_RENAME_OBJECT ){ + + /* Link the new Index structure to its table and to the other + ** in-memory database structures. + */ + assert( pParse->nErr==0 ); + if( db->init.busy ){ + Index *p; + assert( !IN_SPECIAL_PARSE ); + assert( sqlite3SchemaMutexHeld(db, 0, pIndex->pSchema) ); + if( pTblName!=0 ){ + pIndex->tnum = db->init.newTnum; + if( sqlite3IndexHasDuplicateRootPage(pIndex) ){ + sqlite3ErrorMsg(pParse, "invalid rootpage"); + pParse->rc = SQLITE_CORRUPT_BKPT; + goto exit_create_index; + } + } + p = sqlite3HashInsert(&pIndex->pSchema->idxHash, + pIndex->zName, pIndex); + if( p ){ + assert( p==pIndex ); /* Malloc must have failed */ + sqlite3OomFault(db); + goto exit_create_index; + } + db->mDbFlags |= DBFLAG_SchemaChange; + } + + /* If this is the initial CREATE INDEX statement (or CREATE TABLE if the + ** index is an implied index for a UNIQUE or PRIMARY KEY constraint) then + ** emit code to allocate the index rootpage on disk and make an entry for + ** the index in the sqlite_master table and populate the index with + ** content. But, do not do this if we are simply reading the sqlite_master + ** table to parse the schema, or if this index is the PRIMARY KEY index + ** of a WITHOUT ROWID table. + ** + ** If pTblName==0 it means this index is generated as an implied PRIMARY KEY + ** or UNIQUE index in a CREATE TABLE statement. Since the table + ** has just been created, it contains no data and the index initialization + ** step can be skipped. + */ + else if( HasRowid(pTab) || pTblName!=0 ){ + Vdbe *v; + char *zStmt; + int iMem = ++pParse->nMem; + + v = sqlite3GetVdbe(pParse); + if( v==0 ) goto exit_create_index; + + sqlite3BeginWriteOperation(pParse, 1, iDb); + + /* Create the rootpage for the index using CreateIndex. But before + ** doing so, code a Noop instruction and store its address in + ** Index.tnum. This is required in case this index is actually a + ** PRIMARY KEY and the table is actually a WITHOUT ROWID table. In + ** that case the convertToWithoutRowidTable() routine will replace + ** the Noop with a Goto to jump over the VDBE code generated below. */ + pIndex->tnum = sqlite3VdbeAddOp0(v, OP_Noop); + sqlite3VdbeAddOp3(v, OP_CreateBtree, iDb, iMem, BTREE_BLOBKEY); + + /* Gather the complete text of the CREATE INDEX statement into + ** the zStmt variable + */ + assert( pName!=0 || pStart==0 ); + if( pStart ){ + int n = (int)(pParse->sLastToken.z - pName->z) + pParse->sLastToken.n; + if( pName->z[n-1]==';' ) n--; + /* A named index with an explicit CREATE INDEX statement */ + zStmt = sqlite3MPrintf(db, "CREATE%s INDEX %.*s", + onError==OE_None ? "" : " UNIQUE", n, pName->z); + }else{ + /* An automatic index created by a PRIMARY KEY or UNIQUE constraint */ + /* zStmt = sqlite3MPrintf(""); */ + zStmt = 0; + } + + /* Add an entry in sqlite_master for this index + */ + sqlite3NestedParse(pParse, + "INSERT INTO %Q.%s VALUES('index',%Q,%Q,#%d,%Q);", + db->aDb[iDb].zDbSName, MASTER_NAME, + pIndex->zName, + pTab->zName, + iMem, + zStmt + ); + sqlite3DbFree(db, zStmt); + + /* Fill the index with data and reparse the schema. Code an OP_Expire + ** to invalidate all pre-compiled statements. + */ + if( pTblName ){ + sqlite3RefillIndex(pParse, pIndex, iMem); + sqlite3ChangeCookie(pParse, iDb); + sqlite3VdbeAddParseSchemaOp(v, iDb, + sqlite3MPrintf(db, "name='%q' AND type='index'", pIndex->zName)); + sqlite3VdbeAddOp2(v, OP_Expire, 0, 1); + } + + sqlite3VdbeJumpHere(v, pIndex->tnum); + } + } + + /* When adding an index to the list of indices for a table, make + ** sure all indices labeled OE_Replace come after all those labeled + ** OE_Ignore. This is necessary for the correct constraint check + ** processing (in sqlite3GenerateConstraintChecks()) as part of + ** UPDATE and INSERT statements. + */ + if( db->init.busy || pTblName==0 ){ + if( onError!=OE_Replace || pTab->pIndex==0 + || pTab->pIndex->onError==OE_Replace){ + pIndex->pNext = pTab->pIndex; + pTab->pIndex = pIndex; + }else{ + Index *pOther = pTab->pIndex; + while( pOther->pNext && pOther->pNext->onError!=OE_Replace ){ + pOther = pOther->pNext; + } + pIndex->pNext = pOther->pNext; + pOther->pNext = pIndex; + } + pIndex = 0; + } + else if( IN_RENAME_OBJECT ){ + assert( pParse->pNewIndex==0 ); + pParse->pNewIndex = pIndex; + pIndex = 0; + } + + /* Clean up before exiting */ +exit_create_index: + if( pIndex ) sqlite3FreeIndex(db, pIndex); + sqlite3ExprDelete(db, pPIWhere); + sqlite3ExprListDelete(db, pList); + sqlite3SrcListDelete(db, pTblName); + sqlite3DbFree(db, zName); +} + +/* +** Fill the Index.aiRowEst[] array with default information - information +** to be used when we have not run the ANALYZE command. +** +** aiRowEst[0] is supposed to contain the number of elements in the index. +** Since we do not know, guess 1 million. aiRowEst[1] is an estimate of the +** number of rows in the table that match any particular value of the +** first column of the index. aiRowEst[2] is an estimate of the number +** of rows that match any particular combination of the first 2 columns +** of the index. And so forth. It must always be the case that +* +** aiRowEst[N]<=aiRowEst[N-1] +** aiRowEst[N]>=1 +** +** Apart from that, we have little to go on besides intuition as to +** how aiRowEst[] should be initialized. The numbers generated here +** are based on typical values found in actual indices. +*/ +SQLITE_PRIVATE void sqlite3DefaultRowEst(Index *pIdx){ + /* 10, 9, 8, 7, 6 */ + LogEst aVal[] = { 33, 32, 30, 28, 26 }; + LogEst *a = pIdx->aiRowLogEst; + int nCopy = MIN(ArraySize(aVal), pIdx->nKeyCol); + int i; + + /* Indexes with default row estimates should not have stat1 data */ + assert( !pIdx->hasStat1 ); + + /* Set the first entry (number of rows in the index) to the estimated + ** number of rows in the table, or half the number of rows in the table + ** for a partial index. But do not let the estimate drop below 10. */ + a[0] = pIdx->pTable->nRowLogEst; + if( pIdx->pPartIdxWhere!=0 ) a[0] -= 10; assert( 10==sqlite3LogEst(2) ); + if( a[0]<33 ) a[0] = 33; assert( 33==sqlite3LogEst(10) ); + + /* Estimate that a[1] is 10, a[2] is 9, a[3] is 8, a[4] is 7, a[5] is + ** 6 and each subsequent value (if any) is 5. */ + memcpy(&a[1], aVal, nCopy*sizeof(LogEst)); + for(i=nCopy+1; i<=pIdx->nKeyCol; i++){ + a[i] = 23; assert( 23==sqlite3LogEst(5) ); + } + + assert( 0==sqlite3LogEst(1) ); + if( IsUniqueIndex(pIdx) ) a[pIdx->nKeyCol] = 0; +} + +/* +** This routine will drop an existing named index. This routine +** implements the DROP INDEX statement. +*/ +SQLITE_PRIVATE void sqlite3DropIndex(Parse *pParse, SrcList *pName, int ifExists){ + Index *pIndex; + Vdbe *v; + sqlite3 *db = pParse->db; + int iDb; + + assert( pParse->nErr==0 ); /* Never called with prior errors */ + if( db->mallocFailed ){ + goto exit_drop_index; + } + assert( pName->nSrc==1 ); + if( SQLITE_OK!=sqlite3ReadSchema(pParse) ){ + goto exit_drop_index; + } + pIndex = sqlite3FindIndex(db, pName->a[0].zName, pName->a[0].zDatabase); + if( pIndex==0 ){ + if( !ifExists ){ + sqlite3ErrorMsg(pParse, "no such index: %S", pName, 0); + }else{ + sqlite3CodeVerifyNamedSchema(pParse, pName->a[0].zDatabase); + } + pParse->checkSchema = 1; + goto exit_drop_index; + } + if( pIndex->idxType!=SQLITE_IDXTYPE_APPDEF ){ + sqlite3ErrorMsg(pParse, "index associated with UNIQUE " + "or PRIMARY KEY constraint cannot be dropped", 0); + goto exit_drop_index; + } + iDb = sqlite3SchemaToIndex(db, pIndex->pSchema); +#ifndef SQLITE_OMIT_AUTHORIZATION + { + int code = SQLITE_DROP_INDEX; + Table *pTab = pIndex->pTable; + const char *zDb = db->aDb[iDb].zDbSName; + const char *zTab = SCHEMA_TABLE(iDb); + if( sqlite3AuthCheck(pParse, SQLITE_DELETE, zTab, 0, zDb) ){ + goto exit_drop_index; + } + if( !OMIT_TEMPDB && iDb ) code = SQLITE_DROP_TEMP_INDEX; + if( sqlite3AuthCheck(pParse, code, pIndex->zName, pTab->zName, zDb) ){ + goto exit_drop_index; + } + } +#endif + + /* Generate code to remove the index and from the master table */ + v = sqlite3GetVdbe(pParse); + if( v ){ + sqlite3BeginWriteOperation(pParse, 1, iDb); + sqlite3NestedParse(pParse, + "DELETE FROM %Q.%s WHERE name=%Q AND type='index'", + db->aDb[iDb].zDbSName, MASTER_NAME, pIndex->zName + ); + sqlite3ClearStatTables(pParse, iDb, "idx", pIndex->zName); + sqlite3ChangeCookie(pParse, iDb); + destroyRootPage(pParse, pIndex->tnum, iDb); + sqlite3VdbeAddOp4(v, OP_DropIndex, iDb, 0, 0, pIndex->zName, 0); + } + +exit_drop_index: + sqlite3SrcListDelete(db, pName); +} + +/* +** pArray is a pointer to an array of objects. Each object in the +** array is szEntry bytes in size. This routine uses sqlite3DbRealloc() +** to extend the array so that there is space for a new object at the end. +** +** When this function is called, *pnEntry contains the current size of +** the array (in entries - so the allocation is ((*pnEntry) * szEntry) bytes +** in total). +** +** If the realloc() is successful (i.e. if no OOM condition occurs), the +** space allocated for the new object is zeroed, *pnEntry updated to +** reflect the new size of the array and a pointer to the new allocation +** returned. *pIdx is set to the index of the new array entry in this case. +** +** Otherwise, if the realloc() fails, *pIdx is set to -1, *pnEntry remains +** unchanged and a copy of pArray returned. +*/ +SQLITE_PRIVATE void *sqlite3ArrayAllocate( + sqlite3 *db, /* Connection to notify of malloc failures */ + void *pArray, /* Array of objects. Might be reallocated */ + int szEntry, /* Size of each object in the array */ + int *pnEntry, /* Number of objects currently in use */ + int *pIdx /* Write the index of a new slot here */ +){ + char *z; + sqlite3_int64 n = *pIdx = *pnEntry; + if( (n & (n-1))==0 ){ + sqlite3_int64 sz = (n==0) ? 1 : 2*n; + void *pNew = sqlite3DbRealloc(db, pArray, sz*szEntry); + if( pNew==0 ){ + *pIdx = -1; + return pArray; + } + pArray = pNew; + } + z = (char*)pArray; + memset(&z[n * szEntry], 0, szEntry); + ++*pnEntry; + return pArray; +} + +/* +** Append a new element to the given IdList. Create a new IdList if +** need be. +** +** A new IdList is returned, or NULL if malloc() fails. +*/ +SQLITE_PRIVATE IdList *sqlite3IdListAppend(Parse *pParse, IdList *pList, Token *pToken){ + sqlite3 *db = pParse->db; + int i; + if( pList==0 ){ + pList = sqlite3DbMallocZero(db, sizeof(IdList) ); + if( pList==0 ) return 0; + } + pList->a = sqlite3ArrayAllocate( + db, + pList->a, + sizeof(pList->a[0]), + &pList->nId, + &i + ); + if( i<0 ){ + sqlite3IdListDelete(db, pList); + return 0; + } + pList->a[i].zName = sqlite3NameFromToken(db, pToken); + if( IN_RENAME_OBJECT && pList->a[i].zName ){ + sqlite3RenameTokenMap(pParse, (void*)pList->a[i].zName, pToken); + } + return pList; +} + +/* +** Delete an IdList. +*/ +SQLITE_PRIVATE void sqlite3IdListDelete(sqlite3 *db, IdList *pList){ + int i; + if( pList==0 ) return; + for(i=0; inId; i++){ + sqlite3DbFree(db, pList->a[i].zName); + } + sqlite3DbFree(db, pList->a); + sqlite3DbFreeNN(db, pList); +} + +/* +** Return the index in pList of the identifier named zId. Return -1 +** if not found. +*/ +SQLITE_PRIVATE int sqlite3IdListIndex(IdList *pList, const char *zName){ + int i; + if( pList==0 ) return -1; + for(i=0; inId; i++){ + if( sqlite3StrICmp(pList->a[i].zName, zName)==0 ) return i; + } + return -1; +} + +/* +** Maximum size of a SrcList object. +** The SrcList object is used to represent the FROM clause of a +** SELECT statement, and the query planner cannot deal with more +** than 64 tables in a join. So any value larger than 64 here +** is sufficient for most uses. Smaller values, like say 10, are +** appropriate for small and memory-limited applications. +*/ +#ifndef SQLITE_MAX_SRCLIST +# define SQLITE_MAX_SRCLIST 200 +#endif + +/* +** Expand the space allocated for the given SrcList object by +** creating nExtra new slots beginning at iStart. iStart is zero based. +** New slots are zeroed. +** +** For example, suppose a SrcList initially contains two entries: A,B. +** To append 3 new entries onto the end, do this: +** +** sqlite3SrcListEnlarge(db, pSrclist, 3, 2); +** +** After the call above it would contain: A, B, nil, nil, nil. +** If the iStart argument had been 1 instead of 2, then the result +** would have been: A, nil, nil, nil, B. To prepend the new slots, +** the iStart value would be 0. The result then would +** be: nil, nil, nil, A, B. +** +** If a memory allocation fails or the SrcList becomes too large, leave +** the original SrcList unchanged, return NULL, and leave an error message +** in pParse. +*/ +SQLITE_PRIVATE SrcList *sqlite3SrcListEnlarge( + Parse *pParse, /* Parsing context into which errors are reported */ + SrcList *pSrc, /* The SrcList to be enlarged */ + int nExtra, /* Number of new slots to add to pSrc->a[] */ + int iStart /* Index in pSrc->a[] of first new slot */ +){ + int i; + + /* Sanity checking on calling parameters */ + assert( iStart>=0 ); + assert( nExtra>=1 ); + assert( pSrc!=0 ); + assert( iStart<=pSrc->nSrc ); + + /* Allocate additional space if needed */ + if( (u32)pSrc->nSrc+nExtra>pSrc->nAlloc ){ + SrcList *pNew; + sqlite3_int64 nAlloc = 2*(sqlite3_int64)pSrc->nSrc+nExtra; + sqlite3 *db = pParse->db; + + if( pSrc->nSrc+nExtra>=SQLITE_MAX_SRCLIST ){ + sqlite3ErrorMsg(pParse, "too many FROM clause terms, max: %d", + SQLITE_MAX_SRCLIST); + return 0; + } + if( nAlloc>SQLITE_MAX_SRCLIST ) nAlloc = SQLITE_MAX_SRCLIST; + pNew = sqlite3DbRealloc(db, pSrc, + sizeof(*pSrc) + (nAlloc-1)*sizeof(pSrc->a[0]) ); + if( pNew==0 ){ + assert( db->mallocFailed ); + return 0; + } + pSrc = pNew; + pSrc->nAlloc = nAlloc; + } + + /* Move existing slots that come after the newly inserted slots + ** out of the way */ + for(i=pSrc->nSrc-1; i>=iStart; i--){ + pSrc->a[i+nExtra] = pSrc->a[i]; + } + pSrc->nSrc += nExtra; + + /* Zero the newly allocated slots */ + memset(&pSrc->a[iStart], 0, sizeof(pSrc->a[0])*nExtra); + for(i=iStart; ia[i].iCursor = -1; + } + + /* Return a pointer to the enlarged SrcList */ + return pSrc; +} + + +/* +** Append a new table name to the given SrcList. Create a new SrcList if +** need be. A new entry is created in the SrcList even if pTable is NULL. +** +** A SrcList is returned, or NULL if there is an OOM error or if the +** SrcList grows to large. The returned +** SrcList might be the same as the SrcList that was input or it might be +** a new one. If an OOM error does occurs, then the prior value of pList +** that is input to this routine is automatically freed. +** +** If pDatabase is not null, it means that the table has an optional +** database name prefix. Like this: "database.table". The pDatabase +** points to the table name and the pTable points to the database name. +** The SrcList.a[].zName field is filled with the table name which might +** come from pTable (if pDatabase is NULL) or from pDatabase. +** SrcList.a[].zDatabase is filled with the database name from pTable, +** or with NULL if no database is specified. +** +** In other words, if call like this: +** +** sqlite3SrcListAppend(D,A,B,0); +** +** Then B is a table name and the database name is unspecified. If called +** like this: +** +** sqlite3SrcListAppend(D,A,B,C); +** +** Then C is the table name and B is the database name. If C is defined +** then so is B. In other words, we never have a case where: +** +** sqlite3SrcListAppend(D,A,0,C); +** +** Both pTable and pDatabase are assumed to be quoted. They are dequoted +** before being added to the SrcList. +*/ +SQLITE_PRIVATE SrcList *sqlite3SrcListAppend( + Parse *pParse, /* Parsing context, in which errors are reported */ + SrcList *pList, /* Append to this SrcList. NULL creates a new SrcList */ + Token *pTable, /* Table to append */ + Token *pDatabase /* Database of the table */ +){ + struct SrcList_item *pItem; + sqlite3 *db; + assert( pDatabase==0 || pTable!=0 ); /* Cannot have C without B */ + assert( pParse!=0 ); + assert( pParse->db!=0 ); + db = pParse->db; + if( pList==0 ){ + pList = sqlite3DbMallocRawNN(pParse->db, sizeof(SrcList) ); + if( pList==0 ) return 0; + pList->nAlloc = 1; + pList->nSrc = 1; + memset(&pList->a[0], 0, sizeof(pList->a[0])); + pList->a[0].iCursor = -1; + }else{ + SrcList *pNew = sqlite3SrcListEnlarge(pParse, pList, 1, pList->nSrc); + if( pNew==0 ){ + sqlite3SrcListDelete(db, pList); + return 0; + }else{ + pList = pNew; + } + } + pItem = &pList->a[pList->nSrc-1]; + if( pDatabase && pDatabase->z==0 ){ + pDatabase = 0; + } + if( pDatabase ){ + pItem->zName = sqlite3NameFromToken(db, pDatabase); + pItem->zDatabase = sqlite3NameFromToken(db, pTable); + }else{ + pItem->zName = sqlite3NameFromToken(db, pTable); + pItem->zDatabase = 0; + } + return pList; +} + +/* +** Assign VdbeCursor index numbers to all tables in a SrcList +*/ +SQLITE_PRIVATE void sqlite3SrcListAssignCursors(Parse *pParse, SrcList *pList){ + int i; + struct SrcList_item *pItem; + assert(pList || pParse->db->mallocFailed ); + if( pList ){ + for(i=0, pItem=pList->a; inSrc; i++, pItem++){ + if( pItem->iCursor>=0 ) break; + pItem->iCursor = pParse->nTab++; + if( pItem->pSelect ){ + sqlite3SrcListAssignCursors(pParse, pItem->pSelect->pSrc); + } + } + } +} + +/* +** Delete an entire SrcList including all its substructure. +*/ +SQLITE_PRIVATE void sqlite3SrcListDelete(sqlite3 *db, SrcList *pList){ + int i; + struct SrcList_item *pItem; + if( pList==0 ) return; + for(pItem=pList->a, i=0; inSrc; i++, pItem++){ + sqlite3DbFree(db, pItem->zDatabase); + sqlite3DbFree(db, pItem->zName); + sqlite3DbFree(db, pItem->zAlias); + if( pItem->fg.isIndexedBy ) sqlite3DbFree(db, pItem->u1.zIndexedBy); + if( pItem->fg.isTabFunc ) sqlite3ExprListDelete(db, pItem->u1.pFuncArg); + sqlite3DeleteTable(db, pItem->pTab); + sqlite3SelectDelete(db, pItem->pSelect); + sqlite3ExprDelete(db, pItem->pOn); + sqlite3IdListDelete(db, pItem->pUsing); + } + sqlite3DbFreeNN(db, pList); +} + +/* +** This routine is called by the parser to add a new term to the +** end of a growing FROM clause. The "p" parameter is the part of +** the FROM clause that has already been constructed. "p" is NULL +** if this is the first term of the FROM clause. pTable and pDatabase +** are the name of the table and database named in the FROM clause term. +** pDatabase is NULL if the database name qualifier is missing - the +** usual case. If the term has an alias, then pAlias points to the +** alias token. If the term is a subquery, then pSubquery is the +** SELECT statement that the subquery encodes. The pTable and +** pDatabase parameters are NULL for subqueries. The pOn and pUsing +** parameters are the content of the ON and USING clauses. +** +** Return a new SrcList which encodes is the FROM with the new +** term added. +*/ +SQLITE_PRIVATE SrcList *sqlite3SrcListAppendFromTerm( + Parse *pParse, /* Parsing context */ + SrcList *p, /* The left part of the FROM clause already seen */ + Token *pTable, /* Name of the table to add to the FROM clause */ + Token *pDatabase, /* Name of the database containing pTable */ + Token *pAlias, /* The right-hand side of the AS subexpression */ + Select *pSubquery, /* A subquery used in place of a table name */ + Expr *pOn, /* The ON clause of a join */ + IdList *pUsing /* The USING clause of a join */ +){ + struct SrcList_item *pItem; + sqlite3 *db = pParse->db; + if( !p && (pOn || pUsing) ){ + sqlite3ErrorMsg(pParse, "a JOIN clause is required before %s", + (pOn ? "ON" : "USING") + ); + goto append_from_error; + } + p = sqlite3SrcListAppend(pParse, p, pTable, pDatabase); + if( p==0 ){ + goto append_from_error; + } + assert( p->nSrc>0 ); + pItem = &p->a[p->nSrc-1]; + assert( (pTable==0)==(pDatabase==0) ); + assert( pItem->zName==0 || pDatabase!=0 ); + if( IN_RENAME_OBJECT && pItem->zName ){ + Token *pToken = (ALWAYS(pDatabase) && pDatabase->z) ? pDatabase : pTable; + sqlite3RenameTokenMap(pParse, pItem->zName, pToken); + } + assert( pAlias!=0 ); + if( pAlias->n ){ + pItem->zAlias = sqlite3NameFromToken(db, pAlias); + } + pItem->pSelect = pSubquery; + pItem->pOn = pOn; + pItem->pUsing = pUsing; + return p; + + append_from_error: + assert( p==0 ); + sqlite3ExprDelete(db, pOn); + sqlite3IdListDelete(db, pUsing); + sqlite3SelectDelete(db, pSubquery); + return 0; +} + +/* +** Add an INDEXED BY or NOT INDEXED clause to the most recently added +** element of the source-list passed as the second argument. +*/ +SQLITE_PRIVATE void sqlite3SrcListIndexedBy(Parse *pParse, SrcList *p, Token *pIndexedBy){ + assert( pIndexedBy!=0 ); + if( p && pIndexedBy->n>0 ){ + struct SrcList_item *pItem; + assert( p->nSrc>0 ); + pItem = &p->a[p->nSrc-1]; + assert( pItem->fg.notIndexed==0 ); + assert( pItem->fg.isIndexedBy==0 ); + assert( pItem->fg.isTabFunc==0 ); + if( pIndexedBy->n==1 && !pIndexedBy->z ){ + /* A "NOT INDEXED" clause was supplied. See parse.y + ** construct "indexed_opt" for details. */ + pItem->fg.notIndexed = 1; + }else{ + pItem->u1.zIndexedBy = sqlite3NameFromToken(pParse->db, pIndexedBy); + pItem->fg.isIndexedBy = 1; + } + } +} + +/* +** Add the list of function arguments to the SrcList entry for a +** table-valued-function. +*/ +SQLITE_PRIVATE void sqlite3SrcListFuncArgs(Parse *pParse, SrcList *p, ExprList *pList){ + if( p ){ + struct SrcList_item *pItem = &p->a[p->nSrc-1]; + assert( pItem->fg.notIndexed==0 ); + assert( pItem->fg.isIndexedBy==0 ); + assert( pItem->fg.isTabFunc==0 ); + pItem->u1.pFuncArg = pList; + pItem->fg.isTabFunc = 1; + }else{ + sqlite3ExprListDelete(pParse->db, pList); + } +} + +/* +** When building up a FROM clause in the parser, the join operator +** is initially attached to the left operand. But the code generator +** expects the join operator to be on the right operand. This routine +** Shifts all join operators from left to right for an entire FROM +** clause. +** +** Example: Suppose the join is like this: +** +** A natural cross join B +** +** The operator is "natural cross join". The A and B operands are stored +** in p->a[0] and p->a[1], respectively. The parser initially stores the +** operator with A. This routine shifts that operator over to B. +*/ +SQLITE_PRIVATE void sqlite3SrcListShiftJoinType(SrcList *p){ + if( p ){ + int i; + for(i=p->nSrc-1; i>0; i--){ + p->a[i].fg.jointype = p->a[i-1].fg.jointype; + } + p->a[0].fg.jointype = 0; + } +} + +/* +** Generate VDBE code for a BEGIN statement. +*/ +SQLITE_PRIVATE void sqlite3BeginTransaction(Parse *pParse, int type){ + sqlite3 *db; + Vdbe *v; + int i; + + assert( pParse!=0 ); + db = pParse->db; + assert( db!=0 ); + if( sqlite3AuthCheck(pParse, SQLITE_TRANSACTION, "BEGIN", 0, 0) ){ + return; + } + v = sqlite3GetVdbe(pParse); + if( !v ) return; + if( type!=TK_DEFERRED ){ + for(i=0; inDb; i++){ + sqlite3VdbeAddOp2(v, OP_Transaction, i, (type==TK_EXCLUSIVE)+1); + sqlite3VdbeUsesBtree(v, i); + } + } + sqlite3VdbeAddOp0(v, OP_AutoCommit); +} + +/* +** Generate VDBE code for a COMMIT or ROLLBACK statement. +** Code for ROLLBACK is generated if eType==TK_ROLLBACK. Otherwise +** code is generated for a COMMIT. +*/ +SQLITE_PRIVATE void sqlite3EndTransaction(Parse *pParse, int eType){ + Vdbe *v; + int isRollback; + + assert( pParse!=0 ); + assert( pParse->db!=0 ); + assert( eType==TK_COMMIT || eType==TK_END || eType==TK_ROLLBACK ); + isRollback = eType==TK_ROLLBACK; + if( sqlite3AuthCheck(pParse, SQLITE_TRANSACTION, + isRollback ? "ROLLBACK" : "COMMIT", 0, 0) ){ + return; + } + v = sqlite3GetVdbe(pParse); + if( v ){ + sqlite3VdbeAddOp2(v, OP_AutoCommit, 1, isRollback); + } +} + +/* +** This function is called by the parser when it parses a command to create, +** release or rollback an SQL savepoint. +*/ +SQLITE_PRIVATE void sqlite3Savepoint(Parse *pParse, int op, Token *pName){ + char *zName = sqlite3NameFromToken(pParse->db, pName); + if( zName ){ + Vdbe *v = sqlite3GetVdbe(pParse); +#ifndef SQLITE_OMIT_AUTHORIZATION + static const char * const az[] = { "BEGIN", "RELEASE", "ROLLBACK" }; + assert( !SAVEPOINT_BEGIN && SAVEPOINT_RELEASE==1 && SAVEPOINT_ROLLBACK==2 ); +#endif + if( !v || sqlite3AuthCheck(pParse, SQLITE_SAVEPOINT, az[op], zName, 0) ){ + sqlite3DbFree(pParse->db, zName); + return; + } + sqlite3VdbeAddOp4(v, OP_Savepoint, op, 0, 0, zName, P4_DYNAMIC); + } +} + +/* +** Make sure the TEMP database is open and available for use. Return +** the number of errors. Leave any error messages in the pParse structure. +*/ +SQLITE_PRIVATE int sqlite3OpenTempDatabase(Parse *pParse){ + sqlite3 *db = pParse->db; + if( db->aDb[1].pBt==0 && !pParse->explain ){ + int rc; + Btree *pBt; + static const int flags = + SQLITE_OPEN_READWRITE | + SQLITE_OPEN_CREATE | + SQLITE_OPEN_EXCLUSIVE | + SQLITE_OPEN_DELETEONCLOSE | + SQLITE_OPEN_TEMP_DB; + + rc = sqlite3BtreeOpen(db->pVfs, 0, db, &pBt, 0, flags); + if( rc!=SQLITE_OK ){ + sqlite3ErrorMsg(pParse, "unable to open a temporary database " + "file for storing temporary tables"); + pParse->rc = rc; + return 1; + } + db->aDb[1].pBt = pBt; + assert( db->aDb[1].pSchema ); + if( SQLITE_NOMEM==sqlite3BtreeSetPageSize(pBt, db->nextPagesize, -1, 0) ){ + sqlite3OomFault(db); + return 1; + } + } + return 0; +} + +/* +** Record the fact that the schema cookie will need to be verified +** for database iDb. The code to actually verify the schema cookie +** will occur at the end of the top-level VDBE and will be generated +** later, by sqlite3FinishCoding(). +*/ +SQLITE_PRIVATE void sqlite3CodeVerifySchema(Parse *pParse, int iDb){ + Parse *pToplevel = sqlite3ParseToplevel(pParse); + + assert( iDb>=0 && iDbdb->nDb ); + assert( pParse->db->aDb[iDb].pBt!=0 || iDb==1 ); + assert( iDbdb, iDb, 0) ); + if( DbMaskTest(pToplevel->cookieMask, iDb)==0 ){ + DbMaskSet(pToplevel->cookieMask, iDb); + if( !OMIT_TEMPDB && iDb==1 ){ + sqlite3OpenTempDatabase(pToplevel); + } + } +} + +/* +** If argument zDb is NULL, then call sqlite3CodeVerifySchema() for each +** attached database. Otherwise, invoke it for the database named zDb only. +*/ +SQLITE_PRIVATE void sqlite3CodeVerifyNamedSchema(Parse *pParse, const char *zDb){ + sqlite3 *db = pParse->db; + int i; + for(i=0; inDb; i++){ + Db *pDb = &db->aDb[i]; + if( pDb->pBt && (!zDb || 0==sqlite3StrICmp(zDb, pDb->zDbSName)) ){ + sqlite3CodeVerifySchema(pParse, i); + } + } +} + +/* +** Generate VDBE code that prepares for doing an operation that +** might change the database. +** +** This routine starts a new transaction if we are not already within +** a transaction. If we are already within a transaction, then a checkpoint +** is set if the setStatement parameter is true. A checkpoint should +** be set for operations that might fail (due to a constraint) part of +** the way through and which will need to undo some writes without having to +** rollback the whole transaction. For operations where all constraints +** can be checked before any changes are made to the database, it is never +** necessary to undo a write and the checkpoint should not be set. +*/ +SQLITE_PRIVATE void sqlite3BeginWriteOperation(Parse *pParse, int setStatement, int iDb){ + Parse *pToplevel = sqlite3ParseToplevel(pParse); + sqlite3CodeVerifySchema(pParse, iDb); + DbMaskSet(pToplevel->writeMask, iDb); + pToplevel->isMultiWrite |= setStatement; +} + +/* +** Indicate that the statement currently under construction might write +** more than one entry (example: deleting one row then inserting another, +** inserting multiple rows in a table, or inserting a row and index entries.) +** If an abort occurs after some of these writes have completed, then it will +** be necessary to undo the completed writes. +*/ +SQLITE_PRIVATE void sqlite3MultiWrite(Parse *pParse){ + Parse *pToplevel = sqlite3ParseToplevel(pParse); + pToplevel->isMultiWrite = 1; +} + +/* +** The code generator calls this routine if is discovers that it is +** possible to abort a statement prior to completion. In order to +** perform this abort without corrupting the database, we need to make +** sure that the statement is protected by a statement transaction. +** +** Technically, we only need to set the mayAbort flag if the +** isMultiWrite flag was previously set. There is a time dependency +** such that the abort must occur after the multiwrite. This makes +** some statements involving the REPLACE conflict resolution algorithm +** go a little faster. But taking advantage of this time dependency +** makes it more difficult to prove that the code is correct (in +** particular, it prevents us from writing an effective +** implementation of sqlite3AssertMayAbort()) and so we have chosen +** to take the safe route and skip the optimization. +*/ +SQLITE_PRIVATE void sqlite3MayAbort(Parse *pParse){ + Parse *pToplevel = sqlite3ParseToplevel(pParse); + pToplevel->mayAbort = 1; +} + +/* +** Code an OP_Halt that causes the vdbe to return an SQLITE_CONSTRAINT +** error. The onError parameter determines which (if any) of the statement +** and/or current transaction is rolled back. +*/ +SQLITE_PRIVATE void sqlite3HaltConstraint( + Parse *pParse, /* Parsing context */ + int errCode, /* extended error code */ + int onError, /* Constraint type */ + char *p4, /* Error message */ + i8 p4type, /* P4_STATIC or P4_TRANSIENT */ + u8 p5Errmsg /* P5_ErrMsg type */ +){ + Vdbe *v = sqlite3GetVdbe(pParse); + assert( (errCode&0xff)==SQLITE_CONSTRAINT ); + if( onError==OE_Abort ){ + sqlite3MayAbort(pParse); + } + sqlite3VdbeAddOp4(v, OP_Halt, errCode, onError, 0, p4, p4type); + sqlite3VdbeChangeP5(v, p5Errmsg); +} + +/* +** Code an OP_Halt due to UNIQUE or PRIMARY KEY constraint violation. +*/ +SQLITE_PRIVATE void sqlite3UniqueConstraint( + Parse *pParse, /* Parsing context */ + int onError, /* Constraint type */ + Index *pIdx /* The index that triggers the constraint */ +){ + char *zErr; + int j; + StrAccum errMsg; + Table *pTab = pIdx->pTable; + + sqlite3StrAccumInit(&errMsg, pParse->db, 0, 0, + pParse->db->aLimit[SQLITE_LIMIT_LENGTH]); + if( pIdx->aColExpr ){ + sqlite3_str_appendf(&errMsg, "index '%q'", pIdx->zName); + }else{ + for(j=0; jnKeyCol; j++){ + char *zCol; + assert( pIdx->aiColumn[j]>=0 ); + zCol = pTab->aCol[pIdx->aiColumn[j]].zName; + if( j ) sqlite3_str_append(&errMsg, ", ", 2); + sqlite3_str_appendall(&errMsg, pTab->zName); + sqlite3_str_append(&errMsg, ".", 1); + sqlite3_str_appendall(&errMsg, zCol); + } + } + zErr = sqlite3StrAccumFinish(&errMsg); + sqlite3HaltConstraint(pParse, + IsPrimaryKeyIndex(pIdx) ? SQLITE_CONSTRAINT_PRIMARYKEY + : SQLITE_CONSTRAINT_UNIQUE, + onError, zErr, P4_DYNAMIC, P5_ConstraintUnique); +} + + +/* +** Code an OP_Halt due to non-unique rowid. +*/ +SQLITE_PRIVATE void sqlite3RowidConstraint( + Parse *pParse, /* Parsing context */ + int onError, /* Conflict resolution algorithm */ + Table *pTab /* The table with the non-unique rowid */ +){ + char *zMsg; + int rc; + if( pTab->iPKey>=0 ){ + zMsg = sqlite3MPrintf(pParse->db, "%s.%s", pTab->zName, + pTab->aCol[pTab->iPKey].zName); + rc = SQLITE_CONSTRAINT_PRIMARYKEY; + }else{ + zMsg = sqlite3MPrintf(pParse->db, "%s.rowid", pTab->zName); + rc = SQLITE_CONSTRAINT_ROWID; + } + sqlite3HaltConstraint(pParse, rc, onError, zMsg, P4_DYNAMIC, + P5_ConstraintUnique); +} + +/* +** Check to see if pIndex uses the collating sequence pColl. Return +** true if it does and false if it does not. +*/ +#ifndef SQLITE_OMIT_REINDEX +static int collationMatch(const char *zColl, Index *pIndex){ + int i; + assert( zColl!=0 ); + for(i=0; inColumn; i++){ + const char *z = pIndex->azColl[i]; + assert( z!=0 || pIndex->aiColumn[i]<0 ); + if( pIndex->aiColumn[i]>=0 && 0==sqlite3StrICmp(z, zColl) ){ + return 1; + } + } + return 0; +} +#endif + +/* +** Recompute all indices of pTab that use the collating sequence pColl. +** If pColl==0 then recompute all indices of pTab. +*/ +#ifndef SQLITE_OMIT_REINDEX +static void reindexTable(Parse *pParse, Table *pTab, char const *zColl){ + if( !IsVirtual(pTab) ){ + Index *pIndex; /* An index associated with pTab */ + + for(pIndex=pTab->pIndex; pIndex; pIndex=pIndex->pNext){ + if( zColl==0 || collationMatch(zColl, pIndex) ){ + int iDb = sqlite3SchemaToIndex(pParse->db, pTab->pSchema); + sqlite3BeginWriteOperation(pParse, 0, iDb); + sqlite3RefillIndex(pParse, pIndex, -1); + } + } + } +} +#endif + +/* +** Recompute all indices of all tables in all databases where the +** indices use the collating sequence pColl. If pColl==0 then recompute +** all indices everywhere. +*/ +#ifndef SQLITE_OMIT_REINDEX +static void reindexDatabases(Parse *pParse, char const *zColl){ + Db *pDb; /* A single database */ + int iDb; /* The database index number */ + sqlite3 *db = pParse->db; /* The database connection */ + HashElem *k; /* For looping over tables in pDb */ + Table *pTab; /* A table in the database */ + + assert( sqlite3BtreeHoldsAllMutexes(db) ); /* Needed for schema access */ + for(iDb=0, pDb=db->aDb; iDbnDb; iDb++, pDb++){ + assert( pDb!=0 ); + for(k=sqliteHashFirst(&pDb->pSchema->tblHash); k; k=sqliteHashNext(k)){ + pTab = (Table*)sqliteHashData(k); + reindexTable(pParse, pTab, zColl); + } + } +} +#endif + +/* +** Generate code for the REINDEX command. +** +** REINDEX -- 1 +** REINDEX -- 2 +** REINDEX ?.? -- 3 +** REINDEX ?.? -- 4 +** +** Form 1 causes all indices in all attached databases to be rebuilt. +** Form 2 rebuilds all indices in all databases that use the named +** collating function. Forms 3 and 4 rebuild the named index or all +** indices associated with the named table. +*/ +#ifndef SQLITE_OMIT_REINDEX +SQLITE_PRIVATE void sqlite3Reindex(Parse *pParse, Token *pName1, Token *pName2){ + CollSeq *pColl; /* Collating sequence to be reindexed, or NULL */ + char *z; /* Name of a table or index */ + const char *zDb; /* Name of the database */ + Table *pTab; /* A table in the database */ + Index *pIndex; /* An index associated with pTab */ + int iDb; /* The database index number */ + sqlite3 *db = pParse->db; /* The database connection */ + Token *pObjName; /* Name of the table or index to be reindexed */ + + /* Read the database schema. If an error occurs, leave an error message + ** and code in pParse and return NULL. */ + if( SQLITE_OK!=sqlite3ReadSchema(pParse) ){ + return; + } + + if( pName1==0 ){ + reindexDatabases(pParse, 0); + return; + }else if( NEVER(pName2==0) || pName2->z==0 ){ + char *zColl; + assert( pName1->z ); + zColl = sqlite3NameFromToken(pParse->db, pName1); + if( !zColl ) return; + pColl = sqlite3FindCollSeq(db, ENC(db), zColl, 0); + if( pColl ){ + reindexDatabases(pParse, zColl); + sqlite3DbFree(db, zColl); + return; + } + sqlite3DbFree(db, zColl); + } + iDb = sqlite3TwoPartName(pParse, pName1, pName2, &pObjName); + if( iDb<0 ) return; + z = sqlite3NameFromToken(db, pObjName); + if( z==0 ) return; + zDb = db->aDb[iDb].zDbSName; + pTab = sqlite3FindTable(db, z, zDb); + if( pTab ){ + reindexTable(pParse, pTab, 0); + sqlite3DbFree(db, z); + return; + } + pIndex = sqlite3FindIndex(db, z, zDb); + sqlite3DbFree(db, z); + if( pIndex ){ + sqlite3BeginWriteOperation(pParse, 0, iDb); + sqlite3RefillIndex(pParse, pIndex, -1); + return; + } + sqlite3ErrorMsg(pParse, "unable to identify the object to be reindexed"); +} +#endif + +/* +** Return a KeyInfo structure that is appropriate for the given Index. +** +** The caller should invoke sqlite3KeyInfoUnref() on the returned object +** when it has finished using it. +*/ +SQLITE_PRIVATE KeyInfo *sqlite3KeyInfoOfIndex(Parse *pParse, Index *pIdx){ + int i; + int nCol = pIdx->nColumn; + int nKey = pIdx->nKeyCol; + KeyInfo *pKey; + if( pParse->nErr ) return 0; + if( pIdx->uniqNotNull ){ + pKey = sqlite3KeyInfoAlloc(pParse->db, nKey, nCol-nKey); + }else{ + pKey = sqlite3KeyInfoAlloc(pParse->db, nCol, 0); + } + if( pKey ){ + assert( sqlite3KeyInfoIsWriteable(pKey) ); + for(i=0; iazColl[i]; + pKey->aColl[i] = zColl==sqlite3StrBINARY ? 0 : + sqlite3LocateCollSeq(pParse, zColl); + pKey->aSortFlags[i] = pIdx->aSortOrder[i]; + assert( 0==(pKey->aSortFlags[i] & KEYINFO_ORDER_BIGNULL) ); + } + if( pParse->nErr ){ + assert( pParse->rc==SQLITE_ERROR_MISSING_COLLSEQ ); + if( pIdx->bNoQuery==0 ){ + /* Deactivate the index because it contains an unknown collating + ** sequence. The only way to reactive the index is to reload the + ** schema. Adding the missing collating sequence later does not + ** reactive the index. The application had the chance to register + ** the missing index using the collation-needed callback. For + ** simplicity, SQLite will not give the application a second chance. + */ + pIdx->bNoQuery = 1; + pParse->rc = SQLITE_ERROR_RETRY; + } + sqlite3KeyInfoUnref(pKey); + pKey = 0; + } + } + return pKey; +} + +#ifndef SQLITE_OMIT_CTE +/* +** This routine is invoked once per CTE by the parser while parsing a +** WITH clause. +*/ +SQLITE_PRIVATE With *sqlite3WithAdd( + Parse *pParse, /* Parsing context */ + With *pWith, /* Existing WITH clause, or NULL */ + Token *pName, /* Name of the common-table */ + ExprList *pArglist, /* Optional column name list for the table */ + Select *pQuery /* Query used to initialize the table */ +){ + sqlite3 *db = pParse->db; + With *pNew; + char *zName; + + /* Check that the CTE name is unique within this WITH clause. If + ** not, store an error in the Parse structure. */ + zName = sqlite3NameFromToken(pParse->db, pName); + if( zName && pWith ){ + int i; + for(i=0; inCte; i++){ + if( sqlite3StrICmp(zName, pWith->a[i].zName)==0 ){ + sqlite3ErrorMsg(pParse, "duplicate WITH table name: %s", zName); + } + } + } + + if( pWith ){ + sqlite3_int64 nByte = sizeof(*pWith) + (sizeof(pWith->a[1]) * pWith->nCte); + pNew = sqlite3DbRealloc(db, pWith, nByte); + }else{ + pNew = sqlite3DbMallocZero(db, sizeof(*pWith)); + } + assert( (pNew!=0 && zName!=0) || db->mallocFailed ); + + if( db->mallocFailed ){ + sqlite3ExprListDelete(db, pArglist); + sqlite3SelectDelete(db, pQuery); + sqlite3DbFree(db, zName); + pNew = pWith; + }else{ + pNew->a[pNew->nCte].pSelect = pQuery; + pNew->a[pNew->nCte].pCols = pArglist; + pNew->a[pNew->nCte].zName = zName; + pNew->a[pNew->nCte].zCteErr = 0; + pNew->nCte++; + } + + return pNew; +} + +/* +** Free the contents of the With object passed as the second argument. +*/ +SQLITE_PRIVATE void sqlite3WithDelete(sqlite3 *db, With *pWith){ + if( pWith ){ + int i; + for(i=0; inCte; i++){ + struct Cte *pCte = &pWith->a[i]; + sqlite3ExprListDelete(db, pCte->pCols); + sqlite3SelectDelete(db, pCte->pSelect); + sqlite3DbFree(db, pCte->zName); + } + sqlite3DbFree(db, pWith); + } +} +#endif /* !defined(SQLITE_OMIT_CTE) */ + +/************** End of build.c ***********************************************/ +/************** Begin file callback.c ****************************************/ +/* +** 2005 May 23 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** +** This file contains functions used to access the internal hash tables +** of user defined functions and collation sequences. +*/ + +/* #include "sqliteInt.h" */ + +/* +** Invoke the 'collation needed' callback to request a collation sequence +** in the encoding enc of name zName, length nName. +*/ +static void callCollNeeded(sqlite3 *db, int enc, const char *zName){ + assert( !db->xCollNeeded || !db->xCollNeeded16 ); + if( db->xCollNeeded ){ + char *zExternal = sqlite3DbStrDup(db, zName); + if( !zExternal ) return; + db->xCollNeeded(db->pCollNeededArg, db, enc, zExternal); + sqlite3DbFree(db, zExternal); + } +#ifndef SQLITE_OMIT_UTF16 + if( db->xCollNeeded16 ){ + char const *zExternal; + sqlite3_value *pTmp = sqlite3ValueNew(db); + sqlite3ValueSetStr(pTmp, -1, zName, SQLITE_UTF8, SQLITE_STATIC); + zExternal = sqlite3ValueText(pTmp, SQLITE_UTF16NATIVE); + if( zExternal ){ + db->xCollNeeded16(db->pCollNeededArg, db, (int)ENC(db), zExternal); + } + sqlite3ValueFree(pTmp); + } +#endif +} + +/* +** This routine is called if the collation factory fails to deliver a +** collation function in the best encoding but there may be other versions +** of this collation function (for other text encodings) available. Use one +** of these instead if they exist. Avoid a UTF-8 <-> UTF-16 conversion if +** possible. +*/ +static int synthCollSeq(sqlite3 *db, CollSeq *pColl){ + CollSeq *pColl2; + char *z = pColl->zName; + int i; + static const u8 aEnc[] = { SQLITE_UTF16BE, SQLITE_UTF16LE, SQLITE_UTF8 }; + for(i=0; i<3; i++){ + pColl2 = sqlite3FindCollSeq(db, aEnc[i], z, 0); + if( pColl2->xCmp!=0 ){ + memcpy(pColl, pColl2, sizeof(CollSeq)); + pColl->xDel = 0; /* Do not copy the destructor */ + return SQLITE_OK; + } + } + return SQLITE_ERROR; +} + +/* +** This function is responsible for invoking the collation factory callback +** or substituting a collation sequence of a different encoding when the +** requested collation sequence is not available in the desired encoding. +** +** If it is not NULL, then pColl must point to the database native encoding +** collation sequence with name zName, length nName. +** +** The return value is either the collation sequence to be used in database +** db for collation type name zName, length nName, or NULL, if no collation +** sequence can be found. If no collation is found, leave an error message. +** +** See also: sqlite3LocateCollSeq(), sqlite3FindCollSeq() +*/ +SQLITE_PRIVATE CollSeq *sqlite3GetCollSeq( + Parse *pParse, /* Parsing context */ + u8 enc, /* The desired encoding for the collating sequence */ + CollSeq *pColl, /* Collating sequence with native encoding, or NULL */ + const char *zName /* Collating sequence name */ +){ + CollSeq *p; + sqlite3 *db = pParse->db; + + p = pColl; + if( !p ){ + p = sqlite3FindCollSeq(db, enc, zName, 0); + } + if( !p || !p->xCmp ){ + /* No collation sequence of this type for this encoding is registered. + ** Call the collation factory to see if it can supply us with one. + */ + callCollNeeded(db, enc, zName); + p = sqlite3FindCollSeq(db, enc, zName, 0); + } + if( p && !p->xCmp && synthCollSeq(db, p) ){ + p = 0; + } + assert( !p || p->xCmp ); + if( p==0 ){ + sqlite3ErrorMsg(pParse, "no such collation sequence: %s", zName); + pParse->rc = SQLITE_ERROR_MISSING_COLLSEQ; + } + return p; +} + +/* +** This routine is called on a collation sequence before it is used to +** check that it is defined. An undefined collation sequence exists when +** a database is loaded that contains references to collation sequences +** that have not been defined by sqlite3_create_collation() etc. +** +** If required, this routine calls the 'collation needed' callback to +** request a definition of the collating sequence. If this doesn't work, +** an equivalent collating sequence that uses a text encoding different +** from the main database is substituted, if one is available. +*/ +SQLITE_PRIVATE int sqlite3CheckCollSeq(Parse *pParse, CollSeq *pColl){ + if( pColl && pColl->xCmp==0 ){ + const char *zName = pColl->zName; + sqlite3 *db = pParse->db; + CollSeq *p = sqlite3GetCollSeq(pParse, ENC(db), pColl, zName); + if( !p ){ + return SQLITE_ERROR; + } + assert( p==pColl ); + } + return SQLITE_OK; +} + + + +/* +** Locate and return an entry from the db.aCollSeq hash table. If the entry +** specified by zName and nName is not found and parameter 'create' is +** true, then create a new entry. Otherwise return NULL. +** +** Each pointer stored in the sqlite3.aCollSeq hash table contains an +** array of three CollSeq structures. The first is the collation sequence +** preferred for UTF-8, the second UTF-16le, and the third UTF-16be. +** +** Stored immediately after the three collation sequences is a copy of +** the collation sequence name. A pointer to this string is stored in +** each collation sequence structure. +*/ +static CollSeq *findCollSeqEntry( + sqlite3 *db, /* Database connection */ + const char *zName, /* Name of the collating sequence */ + int create /* Create a new entry if true */ +){ + CollSeq *pColl; + pColl = sqlite3HashFind(&db->aCollSeq, zName); + + if( 0==pColl && create ){ + int nName = sqlite3Strlen30(zName) + 1; + pColl = sqlite3DbMallocZero(db, 3*sizeof(*pColl) + nName); + if( pColl ){ + CollSeq *pDel = 0; + pColl[0].zName = (char*)&pColl[3]; + pColl[0].enc = SQLITE_UTF8; + pColl[1].zName = (char*)&pColl[3]; + pColl[1].enc = SQLITE_UTF16LE; + pColl[2].zName = (char*)&pColl[3]; + pColl[2].enc = SQLITE_UTF16BE; + memcpy(pColl[0].zName, zName, nName); + pDel = sqlite3HashInsert(&db->aCollSeq, pColl[0].zName, pColl); + + /* If a malloc() failure occurred in sqlite3HashInsert(), it will + ** return the pColl pointer to be deleted (because it wasn't added + ** to the hash table). + */ + assert( pDel==0 || pDel==pColl ); + if( pDel!=0 ){ + sqlite3OomFault(db); + sqlite3DbFree(db, pDel); + pColl = 0; + } + } + } + return pColl; +} + +/* +** Parameter zName points to a UTF-8 encoded string nName bytes long. +** Return the CollSeq* pointer for the collation sequence named zName +** for the encoding 'enc' from the database 'db'. +** +** If the entry specified is not found and 'create' is true, then create a +** new entry. Otherwise return NULL. +** +** A separate function sqlite3LocateCollSeq() is a wrapper around +** this routine. sqlite3LocateCollSeq() invokes the collation factory +** if necessary and generates an error message if the collating sequence +** cannot be found. +** +** See also: sqlite3LocateCollSeq(), sqlite3GetCollSeq() +*/ +SQLITE_PRIVATE CollSeq *sqlite3FindCollSeq( + sqlite3 *db, + u8 enc, + const char *zName, + int create +){ + CollSeq *pColl; + if( zName ){ + pColl = findCollSeqEntry(db, zName, create); + }else{ + pColl = db->pDfltColl; + } + assert( SQLITE_UTF8==1 && SQLITE_UTF16LE==2 && SQLITE_UTF16BE==3 ); + assert( enc>=SQLITE_UTF8 && enc<=SQLITE_UTF16BE ); + if( pColl ) pColl += enc-1; + return pColl; +} + +/* During the search for the best function definition, this procedure +** is called to test how well the function passed as the first argument +** matches the request for a function with nArg arguments in a system +** that uses encoding enc. The value returned indicates how well the +** request is matched. A higher value indicates a better match. +** +** If nArg is -1 that means to only return a match (non-zero) if p->nArg +** is also -1. In other words, we are searching for a function that +** takes a variable number of arguments. +** +** If nArg is -2 that means that we are searching for any function +** regardless of the number of arguments it uses, so return a positive +** match score for any +** +** The returned value is always between 0 and 6, as follows: +** +** 0: Not a match. +** 1: UTF8/16 conversion required and function takes any number of arguments. +** 2: UTF16 byte order change required and function takes any number of args. +** 3: encoding matches and function takes any number of arguments +** 4: UTF8/16 conversion required - argument count matches exactly +** 5: UTF16 byte order conversion required - argument count matches exactly +** 6: Perfect match: encoding and argument count match exactly. +** +** If nArg==(-2) then any function with a non-null xSFunc is +** a perfect match and any function with xSFunc NULL is +** a non-match. +*/ +#define FUNC_PERFECT_MATCH 6 /* The score for a perfect match */ +static int matchQuality( + FuncDef *p, /* The function we are evaluating for match quality */ + int nArg, /* Desired number of arguments. (-1)==any */ + u8 enc /* Desired text encoding */ +){ + int match; + + /* nArg of -2 is a special case */ + if( nArg==(-2) ) return (p->xSFunc==0) ? 0 : FUNC_PERFECT_MATCH; + + /* Wrong number of arguments means "no match" */ + if( p->nArg!=nArg && p->nArg>=0 ) return 0; + + /* Give a better score to a function with a specific number of arguments + ** than to function that accepts any number of arguments. */ + if( p->nArg==nArg ){ + match = 4; + }else{ + match = 1; + } + + /* Bonus points if the text encoding matches */ + if( enc==(p->funcFlags & SQLITE_FUNC_ENCMASK) ){ + match += 2; /* Exact encoding match */ + }else if( (enc & p->funcFlags & 2)!=0 ){ + match += 1; /* Both are UTF16, but with different byte orders */ + } + + return match; +} + +/* +** Search a FuncDefHash for a function with the given name. Return +** a pointer to the matching FuncDef if found, or 0 if there is no match. +*/ +SQLITE_PRIVATE FuncDef *sqlite3FunctionSearch( + int h, /* Hash of the name */ + const char *zFunc /* Name of function */ +){ + FuncDef *p; + for(p=sqlite3BuiltinFunctions.a[h]; p; p=p->u.pHash){ + if( sqlite3StrICmp(p->zName, zFunc)==0 ){ + return p; + } + } + return 0; +} + +/* +** Insert a new FuncDef into a FuncDefHash hash table. +*/ +SQLITE_PRIVATE void sqlite3InsertBuiltinFuncs( + FuncDef *aDef, /* List of global functions to be inserted */ + int nDef /* Length of the apDef[] list */ +){ + int i; + for(i=0; i='a' && zName[0]<='z' ); + pOther = sqlite3FunctionSearch(h, zName); + if( pOther ){ + assert( pOther!=&aDef[i] && pOther->pNext!=&aDef[i] ); + aDef[i].pNext = pOther->pNext; + pOther->pNext = &aDef[i]; + }else{ + aDef[i].pNext = 0; + aDef[i].u.pHash = sqlite3BuiltinFunctions.a[h]; + sqlite3BuiltinFunctions.a[h] = &aDef[i]; + } + } +} + + + +/* +** Locate a user function given a name, a number of arguments and a flag +** indicating whether the function prefers UTF-16 over UTF-8. Return a +** pointer to the FuncDef structure that defines that function, or return +** NULL if the function does not exist. +** +** If the createFlag argument is true, then a new (blank) FuncDef +** structure is created and liked into the "db" structure if a +** no matching function previously existed. +** +** If nArg is -2, then the first valid function found is returned. A +** function is valid if xSFunc is non-zero. The nArg==(-2) +** case is used to see if zName is a valid function name for some number +** of arguments. If nArg is -2, then createFlag must be 0. +** +** If createFlag is false, then a function with the required name and +** number of arguments may be returned even if the eTextRep flag does not +** match that requested. +*/ +SQLITE_PRIVATE FuncDef *sqlite3FindFunction( + sqlite3 *db, /* An open database */ + const char *zName, /* Name of the function. zero-terminated */ + int nArg, /* Number of arguments. -1 means any number */ + u8 enc, /* Preferred text encoding */ + u8 createFlag /* Create new entry if true and does not otherwise exist */ +){ + FuncDef *p; /* Iterator variable */ + FuncDef *pBest = 0; /* Best match found so far */ + int bestScore = 0; /* Score of best match */ + int h; /* Hash value */ + int nName; /* Length of the name */ + + assert( nArg>=(-2) ); + assert( nArg>=(-1) || createFlag==0 ); + nName = sqlite3Strlen30(zName); + + /* First search for a match amongst the application-defined functions. + */ + p = (FuncDef*)sqlite3HashFind(&db->aFunc, zName); + while( p ){ + int score = matchQuality(p, nArg, enc); + if( score>bestScore ){ + pBest = p; + bestScore = score; + } + p = p->pNext; + } + + /* If no match is found, search the built-in functions. + ** + ** If the DBFLAG_PreferBuiltin flag is set, then search the built-in + ** functions even if a prior app-defined function was found. And give + ** priority to built-in functions. + ** + ** Except, if createFlag is true, that means that we are trying to + ** install a new function. Whatever FuncDef structure is returned it will + ** have fields overwritten with new information appropriate for the + ** new function. But the FuncDefs for built-in functions are read-only. + ** So we must not search for built-ins when creating a new function. + */ + if( !createFlag && (pBest==0 || (db->mDbFlags & DBFLAG_PreferBuiltin)!=0) ){ + bestScore = 0; + h = SQLITE_FUNC_HASH(sqlite3UpperToLower[(u8)zName[0]], nName); + p = sqlite3FunctionSearch(h, zName); + while( p ){ + int score = matchQuality(p, nArg, enc); + if( score>bestScore ){ + pBest = p; + bestScore = score; + } + p = p->pNext; + } + } + + /* If the createFlag parameter is true and the search did not reveal an + ** exact match for the name, number of arguments and encoding, then add a + ** new entry to the hash table and return it. + */ + if( createFlag && bestScorezName = (const char*)&pBest[1]; + pBest->nArg = (u16)nArg; + pBest->funcFlags = enc; + memcpy((char*)&pBest[1], zName, nName+1); + for(z=(u8*)pBest->zName; *z; z++) *z = sqlite3UpperToLower[*z]; + pOther = (FuncDef*)sqlite3HashInsert(&db->aFunc, pBest->zName, pBest); + if( pOther==pBest ){ + sqlite3DbFree(db, pBest); + sqlite3OomFault(db); + return 0; + }else{ + pBest->pNext = pOther; + } + } + + if( pBest && (pBest->xSFunc || createFlag) ){ + return pBest; + } + return 0; +} + +/* +** Free all resources held by the schema structure. The void* argument points +** at a Schema struct. This function does not call sqlite3DbFree(db, ) on the +** pointer itself, it just cleans up subsidiary resources (i.e. the contents +** of the schema hash tables). +** +** The Schema.cache_size variable is not cleared. +*/ +SQLITE_PRIVATE void sqlite3SchemaClear(void *p){ + Hash temp1; + Hash temp2; + HashElem *pElem; + Schema *pSchema = (Schema *)p; + + temp1 = pSchema->tblHash; + temp2 = pSchema->trigHash; + sqlite3HashInit(&pSchema->trigHash); + sqlite3HashClear(&pSchema->idxHash); + for(pElem=sqliteHashFirst(&temp2); pElem; pElem=sqliteHashNext(pElem)){ + sqlite3DeleteTrigger(0, (Trigger*)sqliteHashData(pElem)); + } + sqlite3HashClear(&temp2); + sqlite3HashInit(&pSchema->tblHash); + for(pElem=sqliteHashFirst(&temp1); pElem; pElem=sqliteHashNext(pElem)){ + Table *pTab = sqliteHashData(pElem); + sqlite3DeleteTable(0, pTab); + } + sqlite3HashClear(&temp1); + sqlite3HashClear(&pSchema->fkeyHash); + pSchema->pSeqTab = 0; + if( pSchema->schemaFlags & DB_SchemaLoaded ){ + pSchema->iGeneration++; + } + pSchema->schemaFlags &= ~(DB_SchemaLoaded|DB_ResetWanted); +} + +/* +** Find and return the schema associated with a BTree. Create +** a new one if necessary. +*/ +SQLITE_PRIVATE Schema *sqlite3SchemaGet(sqlite3 *db, Btree *pBt){ + Schema * p; + if( pBt ){ + p = (Schema *)sqlite3BtreeSchema(pBt, sizeof(Schema), sqlite3SchemaClear); + }else{ + p = (Schema *)sqlite3DbMallocZero(0, sizeof(Schema)); + } + if( !p ){ + sqlite3OomFault(db); + }else if ( 0==p->file_format ){ + sqlite3HashInit(&p->tblHash); + sqlite3HashInit(&p->idxHash); + sqlite3HashInit(&p->trigHash); + sqlite3HashInit(&p->fkeyHash); + p->enc = SQLITE_UTF8; + } + return p; +} + +/************** End of callback.c ********************************************/ +/************** Begin file delete.c ******************************************/ +/* +** 2001 September 15 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** This file contains C code routines that are called by the parser +** in order to generate code for DELETE FROM statements. +*/ +/* #include "sqliteInt.h" */ + +/* +** While a SrcList can in general represent multiple tables and subqueries +** (as in the FROM clause of a SELECT statement) in this case it contains +** the name of a single table, as one might find in an INSERT, DELETE, +** or UPDATE statement. Look up that table in the symbol table and +** return a pointer. Set an error message and return NULL if the table +** name is not found or if any other error occurs. +** +** The following fields are initialized appropriate in pSrc: +** +** pSrc->a[0].pTab Pointer to the Table object +** pSrc->a[0].pIndex Pointer to the INDEXED BY index, if there is one +** +*/ +SQLITE_PRIVATE Table *sqlite3SrcListLookup(Parse *pParse, SrcList *pSrc){ + struct SrcList_item *pItem = pSrc->a; + Table *pTab; + assert( pItem && pSrc->nSrc==1 ); + pTab = sqlite3LocateTableItem(pParse, 0, pItem); + sqlite3DeleteTable(pParse->db, pItem->pTab); + pItem->pTab = pTab; + if( pTab ){ + pTab->nTabRef++; + } + if( sqlite3IndexedByLookup(pParse, pItem) ){ + pTab = 0; + } + return pTab; +} + +/* Return true if table pTab is read-only. +** +** A table is read-only if any of the following are true: +** +** 1) It is a virtual table and no implementation of the xUpdate method +** has been provided +** +** 2) It is a system table (i.e. sqlite_master), this call is not +** part of a nested parse and writable_schema pragma has not +** been specified +** +** 3) The table is a shadow table, the database connection is in +** defensive mode, and the current sqlite3_prepare() +** is for a top-level SQL statement. +*/ +static int tabIsReadOnly(Parse *pParse, Table *pTab){ + sqlite3 *db; + if( IsVirtual(pTab) ){ + return sqlite3GetVTable(pParse->db, pTab)->pMod->pModule->xUpdate==0; + } + if( (pTab->tabFlags & (TF_Readonly|TF_Shadow))==0 ) return 0; + db = pParse->db; + if( (pTab->tabFlags & TF_Readonly)!=0 ){ + return sqlite3WritableSchema(db)==0 && pParse->nested==0; + } + assert( pTab->tabFlags & TF_Shadow ); + return (db->flags & SQLITE_Defensive)!=0 +#ifndef SQLITE_OMIT_VIRTUALTABLE + && db->pVtabCtx==0 +#endif + && db->nVdbeExec==0; +} + +/* +** Check to make sure the given table is writable. If it is not +** writable, generate an error message and return 1. If it is +** writable return 0; +*/ +SQLITE_PRIVATE int sqlite3IsReadOnly(Parse *pParse, Table *pTab, int viewOk){ + if( tabIsReadOnly(pParse, pTab) ){ + sqlite3ErrorMsg(pParse, "table %s may not be modified", pTab->zName); + return 1; + } +#ifndef SQLITE_OMIT_VIEW + if( !viewOk && pTab->pSelect ){ + sqlite3ErrorMsg(pParse,"cannot modify %s because it is a view",pTab->zName); + return 1; + } +#endif + return 0; +} + + +#if !defined(SQLITE_OMIT_VIEW) && !defined(SQLITE_OMIT_TRIGGER) +/* +** Evaluate a view and store its result in an ephemeral table. The +** pWhere argument is an optional WHERE clause that restricts the +** set of rows in the view that are to be added to the ephemeral table. +*/ +SQLITE_PRIVATE void sqlite3MaterializeView( + Parse *pParse, /* Parsing context */ + Table *pView, /* View definition */ + Expr *pWhere, /* Optional WHERE clause to be added */ + ExprList *pOrderBy, /* Optional ORDER BY clause */ + Expr *pLimit, /* Optional LIMIT clause */ + int iCur /* Cursor number for ephemeral table */ +){ + SelectDest dest; + Select *pSel; + SrcList *pFrom; + sqlite3 *db = pParse->db; + int iDb = sqlite3SchemaToIndex(db, pView->pSchema); + pWhere = sqlite3ExprDup(db, pWhere, 0); + pFrom = sqlite3SrcListAppend(pParse, 0, 0, 0); + if( pFrom ){ + assert( pFrom->nSrc==1 ); + pFrom->a[0].zName = sqlite3DbStrDup(db, pView->zName); + pFrom->a[0].zDatabase = sqlite3DbStrDup(db, db->aDb[iDb].zDbSName); + assert( pFrom->a[0].pOn==0 ); + assert( pFrom->a[0].pUsing==0 ); + } + pSel = sqlite3SelectNew(pParse, 0, pFrom, pWhere, 0, 0, pOrderBy, + SF_IncludeHidden, pLimit); + sqlite3SelectDestInit(&dest, SRT_EphemTab, iCur); + sqlite3Select(pParse, pSel, &dest); + sqlite3SelectDelete(db, pSel); +} +#endif /* !defined(SQLITE_OMIT_VIEW) && !defined(SQLITE_OMIT_TRIGGER) */ + +#if defined(SQLITE_ENABLE_UPDATE_DELETE_LIMIT) && !defined(SQLITE_OMIT_SUBQUERY) +/* +** Generate an expression tree to implement the WHERE, ORDER BY, +** and LIMIT/OFFSET portion of DELETE and UPDATE statements. +** +** DELETE FROM table_wxyz WHERE a<5 ORDER BY a LIMIT 1; +** \__________________________/ +** pLimitWhere (pInClause) +*/ +SQLITE_PRIVATE Expr *sqlite3LimitWhere( + Parse *pParse, /* The parser context */ + SrcList *pSrc, /* the FROM clause -- which tables to scan */ + Expr *pWhere, /* The WHERE clause. May be null */ + ExprList *pOrderBy, /* The ORDER BY clause. May be null */ + Expr *pLimit, /* The LIMIT clause. May be null */ + char *zStmtType /* Either DELETE or UPDATE. For err msgs. */ +){ + sqlite3 *db = pParse->db; + Expr *pLhs = NULL; /* LHS of IN(SELECT...) operator */ + Expr *pInClause = NULL; /* WHERE rowid IN ( select ) */ + ExprList *pEList = NULL; /* Expression list contaning only pSelectRowid */ + SrcList *pSelectSrc = NULL; /* SELECT rowid FROM x ... (dup of pSrc) */ + Select *pSelect = NULL; /* Complete SELECT tree */ + Table *pTab; + + /* Check that there isn't an ORDER BY without a LIMIT clause. + */ + if( pOrderBy && pLimit==0 ) { + sqlite3ErrorMsg(pParse, "ORDER BY without LIMIT on %s", zStmtType); + sqlite3ExprDelete(pParse->db, pWhere); + sqlite3ExprListDelete(pParse->db, pOrderBy); + return 0; + } + + /* We only need to generate a select expression if there + ** is a limit/offset term to enforce. + */ + if( pLimit == 0 ) { + return pWhere; + } + + /* Generate a select expression tree to enforce the limit/offset + ** term for the DELETE or UPDATE statement. For example: + ** DELETE FROM table_a WHERE col1=1 ORDER BY col2 LIMIT 1 OFFSET 1 + ** becomes: + ** DELETE FROM table_a WHERE rowid IN ( + ** SELECT rowid FROM table_a WHERE col1=1 ORDER BY col2 LIMIT 1 OFFSET 1 + ** ); + */ + + pTab = pSrc->a[0].pTab; + if( HasRowid(pTab) ){ + pLhs = sqlite3PExpr(pParse, TK_ROW, 0, 0); + pEList = sqlite3ExprListAppend( + pParse, 0, sqlite3PExpr(pParse, TK_ROW, 0, 0) + ); + }else{ + Index *pPk = sqlite3PrimaryKeyIndex(pTab); + if( pPk->nKeyCol==1 ){ + const char *zName = pTab->aCol[pPk->aiColumn[0]].zName; + pLhs = sqlite3Expr(db, TK_ID, zName); + pEList = sqlite3ExprListAppend(pParse, 0, sqlite3Expr(db, TK_ID, zName)); + }else{ + int i; + for(i=0; inKeyCol; i++){ + Expr *p = sqlite3Expr(db, TK_ID, pTab->aCol[pPk->aiColumn[i]].zName); + pEList = sqlite3ExprListAppend(pParse, pEList, p); + } + pLhs = sqlite3PExpr(pParse, TK_VECTOR, 0, 0); + if( pLhs ){ + pLhs->x.pList = sqlite3ExprListDup(db, pEList, 0); + } + } + } + + /* duplicate the FROM clause as it is needed by both the DELETE/UPDATE tree + ** and the SELECT subtree. */ + pSrc->a[0].pTab = 0; + pSelectSrc = sqlite3SrcListDup(pParse->db, pSrc, 0); + pSrc->a[0].pTab = pTab; + pSrc->a[0].pIBIndex = 0; + + /* generate the SELECT expression tree. */ + pSelect = sqlite3SelectNew(pParse, pEList, pSelectSrc, pWhere, 0 ,0, + pOrderBy,0,pLimit + ); + + /* now generate the new WHERE rowid IN clause for the DELETE/UDPATE */ + pInClause = sqlite3PExpr(pParse, TK_IN, pLhs, 0); + sqlite3PExprAddSelect(pParse, pInClause, pSelect); + return pInClause; +} +#endif /* defined(SQLITE_ENABLE_UPDATE_DELETE_LIMIT) */ + /* && !defined(SQLITE_OMIT_SUBQUERY) */ + +/* +** Generate code for a DELETE FROM statement. +** +** DELETE FROM table_wxyz WHERE a<5 AND b NOT NULL; +** \________/ \________________/ +** pTabList pWhere +*/ +SQLITE_PRIVATE void sqlite3DeleteFrom( + Parse *pParse, /* The parser context */ + SrcList *pTabList, /* The table from which we should delete things */ + Expr *pWhere, /* The WHERE clause. May be null */ + ExprList *pOrderBy, /* ORDER BY clause. May be null */ + Expr *pLimit /* LIMIT clause. May be null */ +){ + Vdbe *v; /* The virtual database engine */ + Table *pTab; /* The table from which records will be deleted */ + int i; /* Loop counter */ + WhereInfo *pWInfo; /* Information about the WHERE clause */ + Index *pIdx; /* For looping over indices of the table */ + int iTabCur; /* Cursor number for the table */ + int iDataCur = 0; /* VDBE cursor for the canonical data source */ + int iIdxCur = 0; /* Cursor number of the first index */ + int nIdx; /* Number of indices */ + sqlite3 *db; /* Main database structure */ + AuthContext sContext; /* Authorization context */ + NameContext sNC; /* Name context to resolve expressions in */ + int iDb; /* Database number */ + int memCnt = 0; /* Memory cell used for change counting */ + int rcauth; /* Value returned by authorization callback */ + int eOnePass; /* ONEPASS_OFF or _SINGLE or _MULTI */ + int aiCurOnePass[2]; /* The write cursors opened by WHERE_ONEPASS */ + u8 *aToOpen = 0; /* Open cursor iTabCur+j if aToOpen[j] is true */ + Index *pPk; /* The PRIMARY KEY index on the table */ + int iPk = 0; /* First of nPk registers holding PRIMARY KEY value */ + i16 nPk = 1; /* Number of columns in the PRIMARY KEY */ + int iKey; /* Memory cell holding key of row to be deleted */ + i16 nKey; /* Number of memory cells in the row key */ + int iEphCur = 0; /* Ephemeral table holding all primary key values */ + int iRowSet = 0; /* Register for rowset of rows to delete */ + int addrBypass = 0; /* Address of jump over the delete logic */ + int addrLoop = 0; /* Top of the delete loop */ + int addrEphOpen = 0; /* Instruction to open the Ephemeral table */ + int bComplex; /* True if there are triggers or FKs or + ** subqueries in the WHERE clause */ + +#ifndef SQLITE_OMIT_TRIGGER + int isView; /* True if attempting to delete from a view */ + Trigger *pTrigger; /* List of table triggers, if required */ +#endif + + memset(&sContext, 0, sizeof(sContext)); + db = pParse->db; + if( pParse->nErr || db->mallocFailed ){ + goto delete_from_cleanup; + } + assert( pTabList->nSrc==1 ); + + + /* Locate the table which we want to delete. This table has to be + ** put in an SrcList structure because some of the subroutines we + ** will be calling are designed to work with multiple tables and expect + ** an SrcList* parameter instead of just a Table* parameter. + */ + pTab = sqlite3SrcListLookup(pParse, pTabList); + if( pTab==0 ) goto delete_from_cleanup; + + /* Figure out if we have any triggers and if the table being + ** deleted from is a view + */ +#ifndef SQLITE_OMIT_TRIGGER + pTrigger = sqlite3TriggersExist(pParse, pTab, TK_DELETE, 0, 0); + isView = pTab->pSelect!=0; +#else +# define pTrigger 0 +# define isView 0 +#endif + bComplex = pTrigger || sqlite3FkRequired(pParse, pTab, 0, 0); +#ifdef SQLITE_OMIT_VIEW +# undef isView +# define isView 0 +#endif + +#ifdef SQLITE_ENABLE_UPDATE_DELETE_LIMIT + if( !isView ){ + pWhere = sqlite3LimitWhere( + pParse, pTabList, pWhere, pOrderBy, pLimit, "DELETE" + ); + pOrderBy = 0; + pLimit = 0; + } +#endif + + /* If pTab is really a view, make sure it has been initialized. + */ + if( sqlite3ViewGetColumnNames(pParse, pTab) ){ + goto delete_from_cleanup; + } + + if( sqlite3IsReadOnly(pParse, pTab, (pTrigger?1:0)) ){ + goto delete_from_cleanup; + } + iDb = sqlite3SchemaToIndex(db, pTab->pSchema); + assert( iDbnDb ); + rcauth = sqlite3AuthCheck(pParse, SQLITE_DELETE, pTab->zName, 0, + db->aDb[iDb].zDbSName); + assert( rcauth==SQLITE_OK || rcauth==SQLITE_DENY || rcauth==SQLITE_IGNORE ); + if( rcauth==SQLITE_DENY ){ + goto delete_from_cleanup; + } + assert(!isView || pTrigger); + + /* Assign cursor numbers to the table and all its indices. + */ + assert( pTabList->nSrc==1 ); + iTabCur = pTabList->a[0].iCursor = pParse->nTab++; + for(nIdx=0, pIdx=pTab->pIndex; pIdx; pIdx=pIdx->pNext, nIdx++){ + pParse->nTab++; + } + + /* Start the view context + */ + if( isView ){ + sqlite3AuthContextPush(pParse, &sContext, pTab->zName); + } + + /* Begin generating code. + */ + v = sqlite3GetVdbe(pParse); + if( v==0 ){ + goto delete_from_cleanup; + } + if( pParse->nested==0 ) sqlite3VdbeCountChanges(v); + sqlite3BeginWriteOperation(pParse, bComplex, iDb); + + /* If we are trying to delete from a view, realize that view into + ** an ephemeral table. + */ +#if !defined(SQLITE_OMIT_VIEW) && !defined(SQLITE_OMIT_TRIGGER) + if( isView ){ + sqlite3MaterializeView(pParse, pTab, + pWhere, pOrderBy, pLimit, iTabCur + ); + iDataCur = iIdxCur = iTabCur; + pOrderBy = 0; + pLimit = 0; + } +#endif + + /* Resolve the column names in the WHERE clause. + */ + memset(&sNC, 0, sizeof(sNC)); + sNC.pParse = pParse; + sNC.pSrcList = pTabList; + if( sqlite3ResolveExprNames(&sNC, pWhere) ){ + goto delete_from_cleanup; + } + + /* Initialize the counter of the number of rows deleted, if + ** we are counting rows. + */ + if( (db->flags & SQLITE_CountRows)!=0 + && !pParse->nested + && !pParse->pTriggerTab + ){ + memCnt = ++pParse->nMem; + sqlite3VdbeAddOp2(v, OP_Integer, 0, memCnt); + } + +#ifndef SQLITE_OMIT_TRUNCATE_OPTIMIZATION + /* Special case: A DELETE without a WHERE clause deletes everything. + ** It is easier just to erase the whole table. Prior to version 3.6.5, + ** this optimization caused the row change count (the value returned by + ** API function sqlite3_count_changes) to be set incorrectly. + ** + ** The "rcauth==SQLITE_OK" terms is the + ** IMPLEMENTATION-OF: R-17228-37124 If the action code is SQLITE_DELETE and + ** the callback returns SQLITE_IGNORE then the DELETE operation proceeds but + ** the truncate optimization is disabled and all rows are deleted + ** individually. + */ + if( rcauth==SQLITE_OK + && pWhere==0 + && !bComplex + && !IsVirtual(pTab) +#ifdef SQLITE_ENABLE_PREUPDATE_HOOK + && db->xPreUpdateCallback==0 +#endif + ){ + assert( !isView ); + sqlite3TableLock(pParse, iDb, pTab->tnum, 1, pTab->zName); + if( HasRowid(pTab) ){ + sqlite3VdbeAddOp4(v, OP_Clear, pTab->tnum, iDb, memCnt ? memCnt : -1, + pTab->zName, P4_STATIC); + } + for(pIdx=pTab->pIndex; pIdx; pIdx=pIdx->pNext){ + assert( pIdx->pSchema==pTab->pSchema ); + sqlite3VdbeAddOp2(v, OP_Clear, pIdx->tnum, iDb); + } + }else +#endif /* SQLITE_OMIT_TRUNCATE_OPTIMIZATION */ + { + u16 wcf = WHERE_ONEPASS_DESIRED|WHERE_DUPLICATES_OK|WHERE_SEEK_TABLE; + if( sNC.ncFlags & NC_VarSelect ) bComplex = 1; + wcf |= (bComplex ? 0 : WHERE_ONEPASS_MULTIROW); + if( HasRowid(pTab) ){ + /* For a rowid table, initialize the RowSet to an empty set */ + pPk = 0; + nPk = 1; + iRowSet = ++pParse->nMem; + sqlite3VdbeAddOp2(v, OP_Null, 0, iRowSet); + }else{ + /* For a WITHOUT ROWID table, create an ephemeral table used to + ** hold all primary keys for rows to be deleted. */ + pPk = sqlite3PrimaryKeyIndex(pTab); + assert( pPk!=0 ); + nPk = pPk->nKeyCol; + iPk = pParse->nMem+1; + pParse->nMem += nPk; + iEphCur = pParse->nTab++; + addrEphOpen = sqlite3VdbeAddOp2(v, OP_OpenEphemeral, iEphCur, nPk); + sqlite3VdbeSetP4KeyInfo(pParse, pPk); + } + + /* Construct a query to find the rowid or primary key for every row + ** to be deleted, based on the WHERE clause. Set variable eOnePass + ** to indicate the strategy used to implement this delete: + ** + ** ONEPASS_OFF: Two-pass approach - use a FIFO for rowids/PK values. + ** ONEPASS_SINGLE: One-pass approach - at most one row deleted. + ** ONEPASS_MULTI: One-pass approach - any number of rows may be deleted. + */ + pWInfo = sqlite3WhereBegin(pParse, pTabList, pWhere, 0, 0, wcf, iTabCur+1); + if( pWInfo==0 ) goto delete_from_cleanup; + eOnePass = sqlite3WhereOkOnePass(pWInfo, aiCurOnePass); + assert( IsVirtual(pTab)==0 || eOnePass!=ONEPASS_MULTI ); + assert( IsVirtual(pTab) || bComplex || eOnePass!=ONEPASS_OFF ); + if( eOnePass!=ONEPASS_SINGLE ) sqlite3MultiWrite(pParse); + + /* Keep track of the number of rows to be deleted */ + if( memCnt ){ + sqlite3VdbeAddOp2(v, OP_AddImm, memCnt, 1); + } + + /* Extract the rowid or primary key for the current row */ + if( pPk ){ + for(i=0; iaiColumn[i]>=0 ); + sqlite3ExprCodeGetColumnOfTable(v, pTab, iTabCur, + pPk->aiColumn[i], iPk+i); + } + iKey = iPk; + }else{ + iKey = ++pParse->nMem; + sqlite3ExprCodeGetColumnOfTable(v, pTab, iTabCur, -1, iKey); + } + + if( eOnePass!=ONEPASS_OFF ){ + /* For ONEPASS, no need to store the rowid/primary-key. There is only + ** one, so just keep it in its register(s) and fall through to the + ** delete code. */ + nKey = nPk; /* OP_Found will use an unpacked key */ + aToOpen = sqlite3DbMallocRawNN(db, nIdx+2); + if( aToOpen==0 ){ + sqlite3WhereEnd(pWInfo); + goto delete_from_cleanup; + } + memset(aToOpen, 1, nIdx+1); + aToOpen[nIdx+1] = 0; + if( aiCurOnePass[0]>=0 ) aToOpen[aiCurOnePass[0]-iTabCur] = 0; + if( aiCurOnePass[1]>=0 ) aToOpen[aiCurOnePass[1]-iTabCur] = 0; + if( addrEphOpen ) sqlite3VdbeChangeToNoop(v, addrEphOpen); + }else{ + if( pPk ){ + /* Add the PK key for this row to the temporary table */ + iKey = ++pParse->nMem; + nKey = 0; /* Zero tells OP_Found to use a composite key */ + sqlite3VdbeAddOp4(v, OP_MakeRecord, iPk, nPk, iKey, + sqlite3IndexAffinityStr(pParse->db, pPk), nPk); + sqlite3VdbeAddOp4Int(v, OP_IdxInsert, iEphCur, iKey, iPk, nPk); + }else{ + /* Add the rowid of the row to be deleted to the RowSet */ + nKey = 1; /* OP_DeferredSeek always uses a single rowid */ + sqlite3VdbeAddOp2(v, OP_RowSetAdd, iRowSet, iKey); + } + } + + /* If this DELETE cannot use the ONEPASS strategy, this is the + ** end of the WHERE loop */ + if( eOnePass!=ONEPASS_OFF ){ + addrBypass = sqlite3VdbeMakeLabel(pParse); + }else{ + sqlite3WhereEnd(pWInfo); + } + + /* Unless this is a view, open cursors for the table we are + ** deleting from and all its indices. If this is a view, then the + ** only effect this statement has is to fire the INSTEAD OF + ** triggers. + */ + if( !isView ){ + int iAddrOnce = 0; + if( eOnePass==ONEPASS_MULTI ){ + iAddrOnce = sqlite3VdbeAddOp0(v, OP_Once); VdbeCoverage(v); + } + testcase( IsVirtual(pTab) ); + sqlite3OpenTableAndIndices(pParse, pTab, OP_OpenWrite, OPFLAG_FORDELETE, + iTabCur, aToOpen, &iDataCur, &iIdxCur); + assert( pPk || IsVirtual(pTab) || iDataCur==iTabCur ); + assert( pPk || IsVirtual(pTab) || iIdxCur==iDataCur+1 ); + if( eOnePass==ONEPASS_MULTI ) sqlite3VdbeJumpHere(v, iAddrOnce); + } + + /* Set up a loop over the rowids/primary-keys that were found in the + ** where-clause loop above. + */ + if( eOnePass!=ONEPASS_OFF ){ + assert( nKey==nPk ); /* OP_Found will use an unpacked key */ + if( !IsVirtual(pTab) && aToOpen[iDataCur-iTabCur] ){ + assert( pPk!=0 || pTab->pSelect!=0 ); + sqlite3VdbeAddOp4Int(v, OP_NotFound, iDataCur, addrBypass, iKey, nKey); + VdbeCoverage(v); + } + }else if( pPk ){ + addrLoop = sqlite3VdbeAddOp1(v, OP_Rewind, iEphCur); VdbeCoverage(v); + if( IsVirtual(pTab) ){ + sqlite3VdbeAddOp3(v, OP_Column, iEphCur, 0, iKey); + }else{ + sqlite3VdbeAddOp2(v, OP_RowData, iEphCur, iKey); + } + assert( nKey==0 ); /* OP_Found will use a composite key */ + }else{ + addrLoop = sqlite3VdbeAddOp3(v, OP_RowSetRead, iRowSet, 0, iKey); + VdbeCoverage(v); + assert( nKey==1 ); + } + + /* Delete the row */ +#ifndef SQLITE_OMIT_VIRTUALTABLE + if( IsVirtual(pTab) ){ + const char *pVTab = (const char *)sqlite3GetVTable(db, pTab); + sqlite3VtabMakeWritable(pParse, pTab); + assert( eOnePass==ONEPASS_OFF || eOnePass==ONEPASS_SINGLE ); + sqlite3MayAbort(pParse); + if( eOnePass==ONEPASS_SINGLE ){ + sqlite3VdbeAddOp1(v, OP_Close, iTabCur); + if( sqlite3IsToplevel(pParse) ){ + pParse->isMultiWrite = 0; + } + } + sqlite3VdbeAddOp4(v, OP_VUpdate, 0, 1, iKey, pVTab, P4_VTAB); + sqlite3VdbeChangeP5(v, OE_Abort); + }else +#endif + { + int count = (pParse->nested==0); /* True to count changes */ + sqlite3GenerateRowDelete(pParse, pTab, pTrigger, iDataCur, iIdxCur, + iKey, nKey, count, OE_Default, eOnePass, aiCurOnePass[1]); + } + + /* End of the loop over all rowids/primary-keys. */ + if( eOnePass!=ONEPASS_OFF ){ + sqlite3VdbeResolveLabel(v, addrBypass); + sqlite3WhereEnd(pWInfo); + }else if( pPk ){ + sqlite3VdbeAddOp2(v, OP_Next, iEphCur, addrLoop+1); VdbeCoverage(v); + sqlite3VdbeJumpHere(v, addrLoop); + }else{ + sqlite3VdbeGoto(v, addrLoop); + sqlite3VdbeJumpHere(v, addrLoop); + } + } /* End non-truncate path */ + + /* Update the sqlite_sequence table by storing the content of the + ** maximum rowid counter values recorded while inserting into + ** autoincrement tables. + */ + if( pParse->nested==0 && pParse->pTriggerTab==0 ){ + sqlite3AutoincrementEnd(pParse); + } + + /* Return the number of rows that were deleted. If this routine is + ** generating code because of a call to sqlite3NestedParse(), do not + ** invoke the callback function. + */ + if( memCnt ){ + sqlite3VdbeAddOp2(v, OP_ResultRow, memCnt, 1); + sqlite3VdbeSetNumCols(v, 1); + sqlite3VdbeSetColName(v, 0, COLNAME_NAME, "rows deleted", SQLITE_STATIC); + } + +delete_from_cleanup: + sqlite3AuthContextPop(&sContext); + sqlite3SrcListDelete(db, pTabList); + sqlite3ExprDelete(db, pWhere); +#if defined(SQLITE_ENABLE_UPDATE_DELETE_LIMIT) + sqlite3ExprListDelete(db, pOrderBy); + sqlite3ExprDelete(db, pLimit); +#endif + sqlite3DbFree(db, aToOpen); + return; +} +/* Make sure "isView" and other macros defined above are undefined. Otherwise +** they may interfere with compilation of other functions in this file +** (or in another file, if this file becomes part of the amalgamation). */ +#ifdef isView + #undef isView +#endif +#ifdef pTrigger + #undef pTrigger +#endif + +/* +** This routine generates VDBE code that causes a single row of a +** single table to be deleted. Both the original table entry and +** all indices are removed. +** +** Preconditions: +** +** 1. iDataCur is an open cursor on the btree that is the canonical data +** store for the table. (This will be either the table itself, +** in the case of a rowid table, or the PRIMARY KEY index in the case +** of a WITHOUT ROWID table.) +** +** 2. Read/write cursors for all indices of pTab must be open as +** cursor number iIdxCur+i for the i-th index. +** +** 3. The primary key for the row to be deleted must be stored in a +** sequence of nPk memory cells starting at iPk. If nPk==0 that means +** that a search record formed from OP_MakeRecord is contained in the +** single memory location iPk. +** +** eMode: +** Parameter eMode may be passed either ONEPASS_OFF (0), ONEPASS_SINGLE, or +** ONEPASS_MULTI. If eMode is not ONEPASS_OFF, then the cursor +** iDataCur already points to the row to delete. If eMode is ONEPASS_OFF +** then this function must seek iDataCur to the entry identified by iPk +** and nPk before reading from it. +** +** If eMode is ONEPASS_MULTI, then this call is being made as part +** of a ONEPASS delete that affects multiple rows. In this case, if +** iIdxNoSeek is a valid cursor number (>=0) and is not the same as +** iDataCur, then its position should be preserved following the delete +** operation. Or, if iIdxNoSeek is not a valid cursor number, the +** position of iDataCur should be preserved instead. +** +** iIdxNoSeek: +** If iIdxNoSeek is a valid cursor number (>=0) not equal to iDataCur, +** then it identifies an index cursor (from within array of cursors +** starting at iIdxCur) that already points to the index entry to be deleted. +** Except, this optimization is disabled if there are BEFORE triggers since +** the trigger body might have moved the cursor. +*/ +SQLITE_PRIVATE void sqlite3GenerateRowDelete( + Parse *pParse, /* Parsing context */ + Table *pTab, /* Table containing the row to be deleted */ + Trigger *pTrigger, /* List of triggers to (potentially) fire */ + int iDataCur, /* Cursor from which column data is extracted */ + int iIdxCur, /* First index cursor */ + int iPk, /* First memory cell containing the PRIMARY KEY */ + i16 nPk, /* Number of PRIMARY KEY memory cells */ + u8 count, /* If non-zero, increment the row change counter */ + u8 onconf, /* Default ON CONFLICT policy for triggers */ + u8 eMode, /* ONEPASS_OFF, _SINGLE, or _MULTI. See above */ + int iIdxNoSeek /* Cursor number of cursor that does not need seeking */ +){ + Vdbe *v = pParse->pVdbe; /* Vdbe */ + int iOld = 0; /* First register in OLD.* array */ + int iLabel; /* Label resolved to end of generated code */ + u8 opSeek; /* Seek opcode */ + + /* Vdbe is guaranteed to have been allocated by this stage. */ + assert( v ); + VdbeModuleComment((v, "BEGIN: GenRowDel(%d,%d,%d,%d)", + iDataCur, iIdxCur, iPk, (int)nPk)); + + /* Seek cursor iCur to the row to delete. If this row no longer exists + ** (this can happen if a trigger program has already deleted it), do + ** not attempt to delete it or fire any DELETE triggers. */ + iLabel = sqlite3VdbeMakeLabel(pParse); + opSeek = HasRowid(pTab) ? OP_NotExists : OP_NotFound; + if( eMode==ONEPASS_OFF ){ + sqlite3VdbeAddOp4Int(v, opSeek, iDataCur, iLabel, iPk, nPk); + VdbeCoverageIf(v, opSeek==OP_NotExists); + VdbeCoverageIf(v, opSeek==OP_NotFound); + } + + /* If there are any triggers to fire, allocate a range of registers to + ** use for the old.* references in the triggers. */ + if( sqlite3FkRequired(pParse, pTab, 0, 0) || pTrigger ){ + u32 mask; /* Mask of OLD.* columns in use */ + int iCol; /* Iterator used while populating OLD.* */ + int addrStart; /* Start of BEFORE trigger programs */ + + /* TODO: Could use temporary registers here. Also could attempt to + ** avoid copying the contents of the rowid register. */ + mask = sqlite3TriggerColmask( + pParse, pTrigger, 0, 0, TRIGGER_BEFORE|TRIGGER_AFTER, pTab, onconf + ); + mask |= sqlite3FkOldmask(pParse, pTab); + iOld = pParse->nMem+1; + pParse->nMem += (1 + pTab->nCol); + + /* Populate the OLD.* pseudo-table register array. These values will be + ** used by any BEFORE and AFTER triggers that exist. */ + sqlite3VdbeAddOp2(v, OP_Copy, iPk, iOld); + for(iCol=0; iColnCol; iCol++){ + testcase( mask!=0xffffffff && iCol==31 ); + testcase( mask!=0xffffffff && iCol==32 ); + if( mask==0xffffffff || (iCol<=31 && (mask & MASKBIT32(iCol))!=0) ){ + sqlite3ExprCodeGetColumnOfTable(v, pTab, iDataCur, iCol, iOld+iCol+1); + } + } + + /* Invoke BEFORE DELETE trigger programs. */ + addrStart = sqlite3VdbeCurrentAddr(v); + sqlite3CodeRowTrigger(pParse, pTrigger, + TK_DELETE, 0, TRIGGER_BEFORE, pTab, iOld, onconf, iLabel + ); + + /* If any BEFORE triggers were coded, then seek the cursor to the + ** row to be deleted again. It may be that the BEFORE triggers moved + ** the cursor or already deleted the row that the cursor was + ** pointing to. + ** + ** Also disable the iIdxNoSeek optimization since the BEFORE trigger + ** may have moved that cursor. + */ + if( addrStart=0 ); + iIdxNoSeek = -1; + } + + /* Do FK processing. This call checks that any FK constraints that + ** refer to this table (i.e. constraints attached to other tables) + ** are not violated by deleting this row. */ + sqlite3FkCheck(pParse, pTab, iOld, 0, 0, 0); + } + + /* Delete the index and table entries. Skip this step if pTab is really + ** a view (in which case the only effect of the DELETE statement is to + ** fire the INSTEAD OF triggers). + ** + ** If variable 'count' is non-zero, then this OP_Delete instruction should + ** invoke the update-hook. The pre-update-hook, on the other hand should + ** be invoked unless table pTab is a system table. The difference is that + ** the update-hook is not invoked for rows removed by REPLACE, but the + ** pre-update-hook is. + */ + if( pTab->pSelect==0 ){ + u8 p5 = 0; + sqlite3GenerateRowIndexDelete(pParse, pTab, iDataCur, iIdxCur,0,iIdxNoSeek); + sqlite3VdbeAddOp2(v, OP_Delete, iDataCur, (count?OPFLAG_NCHANGE:0)); + if( pParse->nested==0 || 0==sqlite3_stricmp(pTab->zName, "sqlite_stat1") ){ + sqlite3VdbeAppendP4(v, (char*)pTab, P4_TABLE); + } + if( eMode!=ONEPASS_OFF ){ + sqlite3VdbeChangeP5(v, OPFLAG_AUXDELETE); + } + if( iIdxNoSeek>=0 && iIdxNoSeek!=iDataCur ){ + sqlite3VdbeAddOp1(v, OP_Delete, iIdxNoSeek); + } + if( eMode==ONEPASS_MULTI ) p5 |= OPFLAG_SAVEPOSITION; + sqlite3VdbeChangeP5(v, p5); + } + + /* Do any ON CASCADE, SET NULL or SET DEFAULT operations required to + ** handle rows (possibly in other tables) that refer via a foreign key + ** to the row just deleted. */ + sqlite3FkActions(pParse, pTab, 0, iOld, 0, 0); + + /* Invoke AFTER DELETE trigger programs. */ + sqlite3CodeRowTrigger(pParse, pTrigger, + TK_DELETE, 0, TRIGGER_AFTER, pTab, iOld, onconf, iLabel + ); + + /* Jump here if the row had already been deleted before any BEFORE + ** trigger programs were invoked. Or if a trigger program throws a + ** RAISE(IGNORE) exception. */ + sqlite3VdbeResolveLabel(v, iLabel); + VdbeModuleComment((v, "END: GenRowDel()")); +} + +/* +** This routine generates VDBE code that causes the deletion of all +** index entries associated with a single row of a single table, pTab +** +** Preconditions: +** +** 1. A read/write cursor "iDataCur" must be open on the canonical storage +** btree for the table pTab. (This will be either the table itself +** for rowid tables or to the primary key index for WITHOUT ROWID +** tables.) +** +** 2. Read/write cursors for all indices of pTab must be open as +** cursor number iIdxCur+i for the i-th index. (The pTab->pIndex +** index is the 0-th index.) +** +** 3. The "iDataCur" cursor must be already be positioned on the row +** that is to be deleted. +*/ +SQLITE_PRIVATE void sqlite3GenerateRowIndexDelete( + Parse *pParse, /* Parsing and code generating context */ + Table *pTab, /* Table containing the row to be deleted */ + int iDataCur, /* Cursor of table holding data. */ + int iIdxCur, /* First index cursor */ + int *aRegIdx, /* Only delete if aRegIdx!=0 && aRegIdx[i]>0 */ + int iIdxNoSeek /* Do not delete from this cursor */ +){ + int i; /* Index loop counter */ + int r1 = -1; /* Register holding an index key */ + int iPartIdxLabel; /* Jump destination for skipping partial index entries */ + Index *pIdx; /* Current index */ + Index *pPrior = 0; /* Prior index */ + Vdbe *v; /* The prepared statement under construction */ + Index *pPk; /* PRIMARY KEY index, or NULL for rowid tables */ + + v = pParse->pVdbe; + pPk = HasRowid(pTab) ? 0 : sqlite3PrimaryKeyIndex(pTab); + for(i=0, pIdx=pTab->pIndex; pIdx; i++, pIdx=pIdx->pNext){ + assert( iIdxCur+i!=iDataCur || pPk==pIdx ); + if( aRegIdx!=0 && aRegIdx[i]==0 ) continue; + if( pIdx==pPk ) continue; + if( iIdxCur+i==iIdxNoSeek ) continue; + VdbeModuleComment((v, "GenRowIdxDel for %s", pIdx->zName)); + r1 = sqlite3GenerateIndexKey(pParse, pIdx, iDataCur, 0, 1, + &iPartIdxLabel, pPrior, r1); + sqlite3VdbeAddOp3(v, OP_IdxDelete, iIdxCur+i, r1, + pIdx->uniqNotNull ? pIdx->nKeyCol : pIdx->nColumn); + sqlite3ResolvePartIdxLabel(pParse, iPartIdxLabel); + pPrior = pIdx; + } +} + +/* +** Generate code that will assemble an index key and stores it in register +** regOut. The key with be for index pIdx which is an index on pTab. +** iCur is the index of a cursor open on the pTab table and pointing to +** the entry that needs indexing. If pTab is a WITHOUT ROWID table, then +** iCur must be the cursor of the PRIMARY KEY index. +** +** Return a register number which is the first in a block of +** registers that holds the elements of the index key. The +** block of registers has already been deallocated by the time +** this routine returns. +** +** If *piPartIdxLabel is not NULL, fill it in with a label and jump +** to that label if pIdx is a partial index that should be skipped. +** The label should be resolved using sqlite3ResolvePartIdxLabel(). +** A partial index should be skipped if its WHERE clause evaluates +** to false or null. If pIdx is not a partial index, *piPartIdxLabel +** will be set to zero which is an empty label that is ignored by +** sqlite3ResolvePartIdxLabel(). +** +** The pPrior and regPrior parameters are used to implement a cache to +** avoid unnecessary register loads. If pPrior is not NULL, then it is +** a pointer to a different index for which an index key has just been +** computed into register regPrior. If the current pIdx index is generating +** its key into the same sequence of registers and if pPrior and pIdx share +** a column in common, then the register corresponding to that column already +** holds the correct value and the loading of that register is skipped. +** This optimization is helpful when doing a DELETE or an INTEGRITY_CHECK +** on a table with multiple indices, and especially with the ROWID or +** PRIMARY KEY columns of the index. +*/ +SQLITE_PRIVATE int sqlite3GenerateIndexKey( + Parse *pParse, /* Parsing context */ + Index *pIdx, /* The index for which to generate a key */ + int iDataCur, /* Cursor number from which to take column data */ + int regOut, /* Put the new key into this register if not 0 */ + int prefixOnly, /* Compute only a unique prefix of the key */ + int *piPartIdxLabel, /* OUT: Jump to this label to skip partial index */ + Index *pPrior, /* Previously generated index key */ + int regPrior /* Register holding previous generated key */ +){ + Vdbe *v = pParse->pVdbe; + int j; + int regBase; + int nCol; + + if( piPartIdxLabel ){ + if( pIdx->pPartIdxWhere ){ + *piPartIdxLabel = sqlite3VdbeMakeLabel(pParse); + pParse->iSelfTab = iDataCur + 1; + sqlite3ExprIfFalseDup(pParse, pIdx->pPartIdxWhere, *piPartIdxLabel, + SQLITE_JUMPIFNULL); + pParse->iSelfTab = 0; + }else{ + *piPartIdxLabel = 0; + } + } + nCol = (prefixOnly && pIdx->uniqNotNull) ? pIdx->nKeyCol : pIdx->nColumn; + regBase = sqlite3GetTempRange(pParse, nCol); + if( pPrior && (regBase!=regPrior || pPrior->pPartIdxWhere) ) pPrior = 0; + for(j=0; jaiColumn[j]==pIdx->aiColumn[j] + && pPrior->aiColumn[j]!=XN_EXPR + ){ + /* This column was already computed by the previous index */ + continue; + } + sqlite3ExprCodeLoadIndexColumn(pParse, pIdx, iDataCur, j, regBase+j); + /* If the column affinity is REAL but the number is an integer, then it + ** might be stored in the table as an integer (using a compact + ** representation) then converted to REAL by an OP_RealAffinity opcode. + ** But we are getting ready to store this value back into an index, where + ** it should be converted by to INTEGER again. So omit the OP_RealAffinity + ** opcode if it is present */ + sqlite3VdbeDeletePriorOpcode(v, OP_RealAffinity); + } + if( regOut ){ + sqlite3VdbeAddOp3(v, OP_MakeRecord, regBase, nCol, regOut); + if( pIdx->pTable->pSelect ){ + const char *zAff = sqlite3IndexAffinityStr(pParse->db, pIdx); + sqlite3VdbeChangeP4(v, -1, zAff, P4_TRANSIENT); + } + } + sqlite3ReleaseTempRange(pParse, regBase, nCol); + return regBase; +} + +/* +** If a prior call to sqlite3GenerateIndexKey() generated a jump-over label +** because it was a partial index, then this routine should be called to +** resolve that label. +*/ +SQLITE_PRIVATE void sqlite3ResolvePartIdxLabel(Parse *pParse, int iLabel){ + if( iLabel ){ + sqlite3VdbeResolveLabel(pParse->pVdbe, iLabel); + } +} + +/************** End of delete.c **********************************************/ +/************** Begin file func.c ********************************************/ +/* +** 2002 February 23 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** This file contains the C-language implementations for many of the SQL +** functions of SQLite. (Some function, and in particular the date and +** time functions, are implemented separately.) +*/ +/* #include "sqliteInt.h" */ +/* #include */ +/* #include */ +/* #include */ +/* #include "vdbeInt.h" */ + +/* +** Return the collating function associated with a function. +*/ +static CollSeq *sqlite3GetFuncCollSeq(sqlite3_context *context){ + VdbeOp *pOp; + assert( context->pVdbe!=0 ); + pOp = &context->pVdbe->aOp[context->iOp-1]; + assert( pOp->opcode==OP_CollSeq ); + assert( pOp->p4type==P4_COLLSEQ ); + return pOp->p4.pColl; +} + +/* +** Indicate that the accumulator load should be skipped on this +** iteration of the aggregate loop. +*/ +static void sqlite3SkipAccumulatorLoad(sqlite3_context *context){ + assert( context->isError<=0 ); + context->isError = -1; + context->skipFlag = 1; +} + +/* +** Implementation of the non-aggregate min() and max() functions +*/ +static void minmaxFunc( + sqlite3_context *context, + int argc, + sqlite3_value **argv +){ + int i; + int mask; /* 0 for min() or 0xffffffff for max() */ + int iBest; + CollSeq *pColl; + + assert( argc>1 ); + mask = sqlite3_user_data(context)==0 ? 0 : -1; + pColl = sqlite3GetFuncCollSeq(context); + assert( pColl ); + assert( mask==-1 || mask==0 ); + iBest = 0; + if( sqlite3_value_type(argv[0])==SQLITE_NULL ) return; + for(i=1; i=0 ){ + testcase( mask==0 ); + iBest = i; + } + } + sqlite3_result_value(context, argv[iBest]); +} + +/* +** Return the type of the argument. +*/ +static void typeofFunc( + sqlite3_context *context, + int NotUsed, + sqlite3_value **argv +){ + static const char *azType[] = { "integer", "real", "text", "blob", "null" }; + int i = sqlite3_value_type(argv[0]) - 1; + UNUSED_PARAMETER(NotUsed); + assert( i>=0 && i=0xc0 ){ + while( (*z & 0xc0)==0x80 ){ z++; z0++; } + } + } + sqlite3_result_int(context, (int)(z-z0)); + break; + } + default: { + sqlite3_result_null(context); + break; + } + } +} + +/* +** Implementation of the abs() function. +** +** IMP: R-23979-26855 The abs(X) function returns the absolute value of +** the numeric argument X. +*/ +static void absFunc(sqlite3_context *context, int argc, sqlite3_value **argv){ + assert( argc==1 ); + UNUSED_PARAMETER(argc); + switch( sqlite3_value_type(argv[0]) ){ + case SQLITE_INTEGER: { + i64 iVal = sqlite3_value_int64(argv[0]); + if( iVal<0 ){ + if( iVal==SMALLEST_INT64 ){ + /* IMP: R-31676-45509 If X is the integer -9223372036854775808 + ** then abs(X) throws an integer overflow error since there is no + ** equivalent positive 64-bit two complement value. */ + sqlite3_result_error(context, "integer overflow", -1); + return; + } + iVal = -iVal; + } + sqlite3_result_int64(context, iVal); + break; + } + case SQLITE_NULL: { + /* IMP: R-37434-19929 Abs(X) returns NULL if X is NULL. */ + sqlite3_result_null(context); + break; + } + default: { + /* Because sqlite3_value_double() returns 0.0 if the argument is not + ** something that can be converted into a number, we have: + ** IMP: R-01992-00519 Abs(X) returns 0.0 if X is a string or blob + ** that cannot be converted to a numeric value. + */ + double rVal = sqlite3_value_double(argv[0]); + if( rVal<0 ) rVal = -rVal; + sqlite3_result_double(context, rVal); + break; + } + } +} + +/* +** Implementation of the instr() function. +** +** instr(haystack,needle) finds the first occurrence of needle +** in haystack and returns the number of previous characters plus 1, +** or 0 if needle does not occur within haystack. +** +** If both haystack and needle are BLOBs, then the result is one more than +** the number of bytes in haystack prior to the first occurrence of needle, +** or 0 if needle never occurs in haystack. +*/ +static void instrFunc( + sqlite3_context *context, + int argc, + sqlite3_value **argv +){ + const unsigned char *zHaystack; + const unsigned char *zNeedle; + int nHaystack; + int nNeedle; + int typeHaystack, typeNeedle; + int N = 1; + int isText; + unsigned char firstChar; + sqlite3_value *pC1 = 0; + sqlite3_value *pC2 = 0; + + UNUSED_PARAMETER(argc); + typeHaystack = sqlite3_value_type(argv[0]); + typeNeedle = sqlite3_value_type(argv[1]); + if( typeHaystack==SQLITE_NULL || typeNeedle==SQLITE_NULL ) return; + nHaystack = sqlite3_value_bytes(argv[0]); + nNeedle = sqlite3_value_bytes(argv[1]); + if( nNeedle>0 ){ + if( typeHaystack==SQLITE_BLOB && typeNeedle==SQLITE_BLOB ){ + zHaystack = sqlite3_value_blob(argv[0]); + zNeedle = sqlite3_value_blob(argv[1]); + isText = 0; + }else if( typeHaystack!=SQLITE_BLOB && typeNeedle!=SQLITE_BLOB ){ + zHaystack = sqlite3_value_text(argv[0]); + zNeedle = sqlite3_value_text(argv[1]); + isText = 1; + }else{ + pC1 = sqlite3_value_dup(argv[0]); + zHaystack = sqlite3_value_text(pC1); + if( zHaystack==0 ) goto endInstrOOM; + nHaystack = sqlite3_value_bytes(pC1); + pC2 = sqlite3_value_dup(argv[1]); + zNeedle = sqlite3_value_text(pC2); + if( zNeedle==0 ) goto endInstrOOM; + nNeedle = sqlite3_value_bytes(pC2); + isText = 1; + } + if( zNeedle==0 || (nHaystack && zHaystack==0) ) goto endInstrOOM; + firstChar = zNeedle[0]; + while( nNeedle<=nHaystack + && (zHaystack[0]!=firstChar || memcmp(zHaystack, zNeedle, nNeedle)!=0) + ){ + N++; + do{ + nHaystack--; + zHaystack++; + }while( isText && (zHaystack[0]&0xc0)==0x80 ); + } + if( nNeedle>nHaystack ) N = 0; + } + sqlite3_result_int(context, N); +endInstr: + sqlite3_value_free(pC1); + sqlite3_value_free(pC2); + return; +endInstrOOM: + sqlite3_result_error_nomem(context); + goto endInstr; +} + +/* +** Implementation of the printf() function. +*/ +static void printfFunc( + sqlite3_context *context, + int argc, + sqlite3_value **argv +){ + PrintfArguments x; + StrAccum str; + const char *zFormat; + int n; + sqlite3 *db = sqlite3_context_db_handle(context); + + if( argc>=1 && (zFormat = (const char*)sqlite3_value_text(argv[0]))!=0 ){ + x.nArg = argc-1; + x.nUsed = 0; + x.apArg = argv+1; + sqlite3StrAccumInit(&str, db, 0, 0, db->aLimit[SQLITE_LIMIT_LENGTH]); + str.printfFlags = SQLITE_PRINTF_SQLFUNC; + sqlite3_str_appendf(&str, zFormat, &x); + n = str.nChar; + sqlite3_result_text(context, sqlite3StrAccumFinish(&str), n, + SQLITE_DYNAMIC); + } +} + +/* +** Implementation of the substr() function. +** +** substr(x,p1,p2) returns p2 characters of x[] beginning with p1. +** p1 is 1-indexed. So substr(x,1,1) returns the first character +** of x. If x is text, then we actually count UTF-8 characters. +** If x is a blob, then we count bytes. +** +** If p1 is negative, then we begin abs(p1) from the end of x[]. +** +** If p2 is negative, return the p2 characters preceding p1. +*/ +static void substrFunc( + sqlite3_context *context, + int argc, + sqlite3_value **argv +){ + const unsigned char *z; + const unsigned char *z2; + int len; + int p0type; + i64 p1, p2; + int negP2 = 0; + + assert( argc==3 || argc==2 ); + if( sqlite3_value_type(argv[1])==SQLITE_NULL + || (argc==3 && sqlite3_value_type(argv[2])==SQLITE_NULL) + ){ + return; + } + p0type = sqlite3_value_type(argv[0]); + p1 = sqlite3_value_int(argv[1]); + if( p0type==SQLITE_BLOB ){ + len = sqlite3_value_bytes(argv[0]); + z = sqlite3_value_blob(argv[0]); + if( z==0 ) return; + assert( len==sqlite3_value_bytes(argv[0]) ); + }else{ + z = sqlite3_value_text(argv[0]); + if( z==0 ) return; + len = 0; + if( p1<0 ){ + for(z2=z; *z2; len++){ + SQLITE_SKIP_UTF8(z2); + } + } + } +#ifdef SQLITE_SUBSTR_COMPATIBILITY + /* If SUBSTR_COMPATIBILITY is defined then substr(X,0,N) work the same as + ** as substr(X,1,N) - it returns the first N characters of X. This + ** is essentially a back-out of the bug-fix in check-in [5fc125d362df4b8] + ** from 2009-02-02 for compatibility of applications that exploited the + ** old buggy behavior. */ + if( p1==0 ) p1 = 1; /* */ +#endif + if( argc==3 ){ + p2 = sqlite3_value_int(argv[2]); + if( p2<0 ){ + p2 = -p2; + negP2 = 1; + } + }else{ + p2 = sqlite3_context_db_handle(context)->aLimit[SQLITE_LIMIT_LENGTH]; + } + if( p1<0 ){ + p1 += len; + if( p1<0 ){ + p2 += p1; + if( p2<0 ) p2 = 0; + p1 = 0; + } + }else if( p1>0 ){ + p1--; + }else if( p2>0 ){ + p2--; + } + if( negP2 ){ + p1 -= p2; + if( p1<0 ){ + p2 += p1; + p1 = 0; + } + } + assert( p1>=0 && p2>=0 ); + if( p0type!=SQLITE_BLOB ){ + while( *z && p1 ){ + SQLITE_SKIP_UTF8(z); + p1--; + } + for(z2=z; *z2 && p2; p2--){ + SQLITE_SKIP_UTF8(z2); + } + sqlite3_result_text64(context, (char*)z, z2-z, SQLITE_TRANSIENT, + SQLITE_UTF8); + }else{ + if( p1+p2>len ){ + p2 = len-p1; + if( p2<0 ) p2 = 0; + } + sqlite3_result_blob64(context, (char*)&z[p1], (u64)p2, SQLITE_TRANSIENT); + } +} + +/* +** Implementation of the round() function +*/ +#ifndef SQLITE_OMIT_FLOATING_POINT +static void roundFunc(sqlite3_context *context, int argc, sqlite3_value **argv){ + int n = 0; + double r; + char *zBuf; + assert( argc==1 || argc==2 ); + if( argc==2 ){ + if( SQLITE_NULL==sqlite3_value_type(argv[1]) ) return; + n = sqlite3_value_int(argv[1]); + if( n>30 ) n = 30; + if( n<0 ) n = 0; + } + if( sqlite3_value_type(argv[0])==SQLITE_NULL ) return; + r = sqlite3_value_double(argv[0]); + /* If Y==0 and X will fit in a 64-bit int, + ** handle the rounding directly, + ** otherwise use printf. + */ + if( r<-4503599627370496.0 || r>+4503599627370496.0 ){ + /* The value has no fractional part so there is nothing to round */ + }else if( n==0 ){ + r = (double)((sqlite_int64)(r+(r<0?-0.5:+0.5))); + }else{ + zBuf = sqlite3_mprintf("%.*f",n,r); + if( zBuf==0 ){ + sqlite3_result_error_nomem(context); + return; + } + sqlite3AtoF(zBuf, &r, sqlite3Strlen30(zBuf), SQLITE_UTF8); + sqlite3_free(zBuf); + } + sqlite3_result_double(context, r); +} +#endif + +/* +** Allocate nByte bytes of space using sqlite3Malloc(). If the +** allocation fails, call sqlite3_result_error_nomem() to notify +** the database handle that malloc() has failed and return NULL. +** If nByte is larger than the maximum string or blob length, then +** raise an SQLITE_TOOBIG exception and return NULL. +*/ +static void *contextMalloc(sqlite3_context *context, i64 nByte){ + char *z; + sqlite3 *db = sqlite3_context_db_handle(context); + assert( nByte>0 ); + testcase( nByte==db->aLimit[SQLITE_LIMIT_LENGTH] ); + testcase( nByte==db->aLimit[SQLITE_LIMIT_LENGTH]+1 ); + if( nByte>db->aLimit[SQLITE_LIMIT_LENGTH] ){ + sqlite3_result_error_toobig(context); + z = 0; + }else{ + z = sqlite3Malloc(nByte); + if( !z ){ + sqlite3_result_error_nomem(context); + } + } + return z; +} + +/* +** Implementation of the upper() and lower() SQL functions. +*/ +static void upperFunc(sqlite3_context *context, int argc, sqlite3_value **argv){ + char *z1; + const char *z2; + int i, n; + UNUSED_PARAMETER(argc); + z2 = (char*)sqlite3_value_text(argv[0]); + n = sqlite3_value_bytes(argv[0]); + /* Verify that the call to _bytes() does not invalidate the _text() pointer */ + assert( z2==(char*)sqlite3_value_text(argv[0]) ); + if( z2 ){ + z1 = contextMalloc(context, ((i64)n)+1); + if( z1 ){ + for(i=0; imatchOne; /* "?" or "_" */ + u32 matchAll = pInfo->matchAll; /* "*" or "%" */ + u8 noCase = pInfo->noCase; /* True if uppercase==lowercase */ + const u8 *zEscaped = 0; /* One past the last escaped input char */ + + while( (c = Utf8Read(zPattern))!=0 ){ + if( c==matchAll ){ /* Match "*" */ + /* Skip over multiple "*" characters in the pattern. If there + ** are also "?" characters, skip those as well, but consume a + ** single character of the input string for each "?" skipped */ + while( (c=Utf8Read(zPattern)) == matchAll || c == matchOne ){ + if( c==matchOne && sqlite3Utf8Read(&zString)==0 ){ + return SQLITE_NOWILDCARDMATCH; + } + } + if( c==0 ){ + return SQLITE_MATCH; /* "*" at the end of the pattern matches */ + }else if( c==matchOther ){ + if( pInfo->matchSet==0 ){ + c = sqlite3Utf8Read(&zPattern); + if( c==0 ) return SQLITE_NOWILDCARDMATCH; + }else{ + /* "[...]" immediately follows the "*". We have to do a slow + ** recursive search in this case, but it is an unusual case. */ + assert( matchOther<0x80 ); /* '[' is a single-byte character */ + while( *zString ){ + int bMatch = patternCompare(&zPattern[-1],zString,pInfo,matchOther); + if( bMatch!=SQLITE_NOMATCH ) return bMatch; + SQLITE_SKIP_UTF8(zString); + } + return SQLITE_NOWILDCARDMATCH; + } + } + + /* At this point variable c contains the first character of the + ** pattern string past the "*". Search in the input string for the + ** first matching character and recursively continue the match from + ** that point. + ** + ** For a case-insensitive search, set variable cx to be the same as + ** c but in the other case and search the input string for either + ** c or cx. + */ + if( c<=0x80 ){ + char zStop[3]; + int bMatch; + if( noCase ){ + zStop[0] = sqlite3Toupper(c); + zStop[1] = sqlite3Tolower(c); + zStop[2] = 0; + }else{ + zStop[0] = c; + zStop[1] = 0; + } + while(1){ + zString += strcspn((const char*)zString, zStop); + if( zString[0]==0 ) break; + zString++; + bMatch = patternCompare(zPattern,zString,pInfo,matchOther); + if( bMatch!=SQLITE_NOMATCH ) return bMatch; + } + }else{ + int bMatch; + while( (c2 = Utf8Read(zString))!=0 ){ + if( c2!=c ) continue; + bMatch = patternCompare(zPattern,zString,pInfo,matchOther); + if( bMatch!=SQLITE_NOMATCH ) return bMatch; + } + } + return SQLITE_NOWILDCARDMATCH; + } + if( c==matchOther ){ + if( pInfo->matchSet==0 ){ + c = sqlite3Utf8Read(&zPattern); + if( c==0 ) return SQLITE_NOMATCH; + zEscaped = zPattern; + }else{ + u32 prior_c = 0; + int seen = 0; + int invert = 0; + c = sqlite3Utf8Read(&zString); + if( c==0 ) return SQLITE_NOMATCH; + c2 = sqlite3Utf8Read(&zPattern); + if( c2=='^' ){ + invert = 1; + c2 = sqlite3Utf8Read(&zPattern); + } + if( c2==']' ){ + if( c==']' ) seen = 1; + c2 = sqlite3Utf8Read(&zPattern); + } + while( c2 && c2!=']' ){ + if( c2=='-' && zPattern[0]!=']' && zPattern[0]!=0 && prior_c>0 ){ + c2 = sqlite3Utf8Read(&zPattern); + if( c>=prior_c && c<=c2 ) seen = 1; + prior_c = 0; + }else{ + if( c==c2 ){ + seen = 1; + } + prior_c = c2; + } + c2 = sqlite3Utf8Read(&zPattern); + } + if( c2==0 || (seen ^ invert)==0 ){ + return SQLITE_NOMATCH; + } + continue; + } + } + c2 = Utf8Read(zString); + if( c==c2 ) continue; + if( noCase && sqlite3Tolower(c)==sqlite3Tolower(c2) && c<0x80 && c2<0x80 ){ + continue; + } + if( c==matchOne && zPattern!=zEscaped && c2!=0 ) continue; + return SQLITE_NOMATCH; + } + return *zString==0 ? SQLITE_MATCH : SQLITE_NOMATCH; +} + +/* +** The sqlite3_strglob() interface. Return 0 on a match (like strcmp()) and +** non-zero if there is no match. +*/ +SQLITE_API int sqlite3_strglob(const char *zGlobPattern, const char *zString){ + return patternCompare((u8*)zGlobPattern, (u8*)zString, &globInfo, '['); +} + +/* +** The sqlite3_strlike() interface. Return 0 on a match and non-zero for +** a miss - like strcmp(). +*/ +SQLITE_API int sqlite3_strlike(const char *zPattern, const char *zStr, unsigned int esc){ + return patternCompare((u8*)zPattern, (u8*)zStr, &likeInfoNorm, esc); +} + +/* +** Count the number of times that the LIKE operator (or GLOB which is +** just a variation of LIKE) gets called. This is used for testing +** only. +*/ +#ifdef SQLITE_TEST +SQLITE_API int sqlite3_like_count = 0; +#endif + + +/* +** Implementation of the like() SQL function. This function implements +** the build-in LIKE operator. The first argument to the function is the +** pattern and the second argument is the string. So, the SQL statements: +** +** A LIKE B +** +** is implemented as like(B,A). +** +** This same function (with a different compareInfo structure) computes +** the GLOB operator. +*/ +static void likeFunc( + sqlite3_context *context, + int argc, + sqlite3_value **argv +){ + const unsigned char *zA, *zB; + u32 escape; + int nPat; + sqlite3 *db = sqlite3_context_db_handle(context); + struct compareInfo *pInfo = sqlite3_user_data(context); + +#ifdef SQLITE_LIKE_DOESNT_MATCH_BLOBS + if( sqlite3_value_type(argv[0])==SQLITE_BLOB + || sqlite3_value_type(argv[1])==SQLITE_BLOB + ){ +#ifdef SQLITE_TEST + sqlite3_like_count++; +#endif + sqlite3_result_int(context, 0); + return; + } +#endif + + /* Limit the length of the LIKE or GLOB pattern to avoid problems + ** of deep recursion and N*N behavior in patternCompare(). + */ + nPat = sqlite3_value_bytes(argv[0]); + testcase( nPat==db->aLimit[SQLITE_LIMIT_LIKE_PATTERN_LENGTH] ); + testcase( nPat==db->aLimit[SQLITE_LIMIT_LIKE_PATTERN_LENGTH]+1 ); + if( nPat > db->aLimit[SQLITE_LIMIT_LIKE_PATTERN_LENGTH] ){ + sqlite3_result_error(context, "LIKE or GLOB pattern too complex", -1); + return; + } + if( argc==3 ){ + /* The escape character string must consist of a single UTF-8 character. + ** Otherwise, return an error. + */ + const unsigned char *zEsc = sqlite3_value_text(argv[2]); + if( zEsc==0 ) return; + if( sqlite3Utf8CharLen((char*)zEsc, -1)!=1 ){ + sqlite3_result_error(context, + "ESCAPE expression must be a single character", -1); + return; + } + escape = sqlite3Utf8Read(&zEsc); + }else{ + escape = pInfo->matchSet; + } + zB = sqlite3_value_text(argv[0]); + zA = sqlite3_value_text(argv[1]); + if( zA && zB ){ +#ifdef SQLITE_TEST + sqlite3_like_count++; +#endif + sqlite3_result_int(context, + patternCompare(zB, zA, pInfo, escape)==SQLITE_MATCH); + } +} + +/* +** Implementation of the NULLIF(x,y) function. The result is the first +** argument if the arguments are different. The result is NULL if the +** arguments are equal to each other. +*/ +static void nullifFunc( + sqlite3_context *context, + int NotUsed, + sqlite3_value **argv +){ + CollSeq *pColl = sqlite3GetFuncCollSeq(context); + UNUSED_PARAMETER(NotUsed); + if( sqlite3MemCompare(argv[0], argv[1], pColl)!=0 ){ + sqlite3_result_value(context, argv[0]); + } +} + +/* +** Implementation of the sqlite_version() function. The result is the version +** of the SQLite library that is running. +*/ +static void versionFunc( + sqlite3_context *context, + int NotUsed, + sqlite3_value **NotUsed2 +){ + UNUSED_PARAMETER2(NotUsed, NotUsed2); + /* IMP: R-48699-48617 This function is an SQL wrapper around the + ** sqlite3_libversion() C-interface. */ + sqlite3_result_text(context, sqlite3_libversion(), -1, SQLITE_STATIC); +} + +/* +** Implementation of the sqlite_source_id() function. The result is a string +** that identifies the particular version of the source code used to build +** SQLite. +*/ +static void sourceidFunc( + sqlite3_context *context, + int NotUsed, + sqlite3_value **NotUsed2 +){ + UNUSED_PARAMETER2(NotUsed, NotUsed2); + /* IMP: R-24470-31136 This function is an SQL wrapper around the + ** sqlite3_sourceid() C interface. */ + sqlite3_result_text(context, sqlite3_sourceid(), -1, SQLITE_STATIC); +} + +/* +** Implementation of the sqlite_log() function. This is a wrapper around +** sqlite3_log(). The return value is NULL. The function exists purely for +** its side-effects. +*/ +static void errlogFunc( + sqlite3_context *context, + int argc, + sqlite3_value **argv +){ + UNUSED_PARAMETER(argc); + UNUSED_PARAMETER(context); + sqlite3_log(sqlite3_value_int(argv[0]), "%s", sqlite3_value_text(argv[1])); +} + +/* +** Implementation of the sqlite_compileoption_used() function. +** The result is an integer that identifies if the compiler option +** was used to build SQLite. +*/ +#ifndef SQLITE_OMIT_COMPILEOPTION_DIAGS +static void compileoptionusedFunc( + sqlite3_context *context, + int argc, + sqlite3_value **argv +){ + const char *zOptName; + assert( argc==1 ); + UNUSED_PARAMETER(argc); + /* IMP: R-39564-36305 The sqlite_compileoption_used() SQL + ** function is a wrapper around the sqlite3_compileoption_used() C/C++ + ** function. + */ + if( (zOptName = (const char*)sqlite3_value_text(argv[0]))!=0 ){ + sqlite3_result_int(context, sqlite3_compileoption_used(zOptName)); + } +} +#endif /* SQLITE_OMIT_COMPILEOPTION_DIAGS */ + +/* +** Implementation of the sqlite_compileoption_get() function. +** The result is a string that identifies the compiler options +** used to build SQLite. +*/ +#ifndef SQLITE_OMIT_COMPILEOPTION_DIAGS +static void compileoptiongetFunc( + sqlite3_context *context, + int argc, + sqlite3_value **argv +){ + int n; + assert( argc==1 ); + UNUSED_PARAMETER(argc); + /* IMP: R-04922-24076 The sqlite_compileoption_get() SQL function + ** is a wrapper around the sqlite3_compileoption_get() C/C++ function. + */ + n = sqlite3_value_int(argv[0]); + sqlite3_result_text(context, sqlite3_compileoption_get(n), -1, SQLITE_STATIC); +} +#endif /* SQLITE_OMIT_COMPILEOPTION_DIAGS */ + +/* Array for converting from half-bytes (nybbles) into ASCII hex +** digits. */ +static const char hexdigits[] = { + '0', '1', '2', '3', '4', '5', '6', '7', + '8', '9', 'A', 'B', 'C', 'D', 'E', 'F' +}; + +/* +** Implementation of the QUOTE() function. This function takes a single +** argument. If the argument is numeric, the return value is the same as +** the argument. If the argument is NULL, the return value is the string +** "NULL". Otherwise, the argument is enclosed in single quotes with +** single-quote escapes. +*/ +static void quoteFunc(sqlite3_context *context, int argc, sqlite3_value **argv){ + assert( argc==1 ); + UNUSED_PARAMETER(argc); + switch( sqlite3_value_type(argv[0]) ){ + case SQLITE_FLOAT: { + double r1, r2; + char zBuf[50]; + r1 = sqlite3_value_double(argv[0]); + sqlite3_snprintf(sizeof(zBuf), zBuf, "%!.15g", r1); + sqlite3AtoF(zBuf, &r2, 20, SQLITE_UTF8); + if( r1!=r2 ){ + sqlite3_snprintf(sizeof(zBuf), zBuf, "%!.20e", r1); + } + sqlite3_result_text(context, zBuf, -1, SQLITE_TRANSIENT); + break; + } + case SQLITE_INTEGER: { + sqlite3_result_value(context, argv[0]); + break; + } + case SQLITE_BLOB: { + char *zText = 0; + char const *zBlob = sqlite3_value_blob(argv[0]); + int nBlob = sqlite3_value_bytes(argv[0]); + assert( zBlob==sqlite3_value_blob(argv[0]) ); /* No encoding change */ + zText = (char *)contextMalloc(context, (2*(i64)nBlob)+4); + if( zText ){ + int i; + for(i=0; i>4)&0x0F]; + zText[(i*2)+3] = hexdigits[(zBlob[i])&0x0F]; + } + zText[(nBlob*2)+2] = '\''; + zText[(nBlob*2)+3] = '\0'; + zText[0] = 'X'; + zText[1] = '\''; + sqlite3_result_text(context, zText, -1, SQLITE_TRANSIENT); + sqlite3_free(zText); + } + break; + } + case SQLITE_TEXT: { + int i,j; + u64 n; + const unsigned char *zArg = sqlite3_value_text(argv[0]); + char *z; + + if( zArg==0 ) return; + for(i=0, n=0; zArg[i]; i++){ if( zArg[i]=='\'' ) n++; } + z = contextMalloc(context, ((i64)i)+((i64)n)+3); + if( z ){ + z[0] = '\''; + for(i=0, j=1; zArg[i]; i++){ + z[j++] = zArg[i]; + if( zArg[i]=='\'' ){ + z[j++] = '\''; + } + } + z[j++] = '\''; + z[j] = 0; + sqlite3_result_text(context, z, j, sqlite3_free); + } + break; + } + default: { + assert( sqlite3_value_type(argv[0])==SQLITE_NULL ); + sqlite3_result_text(context, "NULL", 4, SQLITE_STATIC); + break; + } + } +} + +/* +** The unicode() function. Return the integer unicode code-point value +** for the first character of the input string. +*/ +static void unicodeFunc( + sqlite3_context *context, + int argc, + sqlite3_value **argv +){ + const unsigned char *z = sqlite3_value_text(argv[0]); + (void)argc; + if( z && z[0] ) sqlite3_result_int(context, sqlite3Utf8Read(&z)); +} + +/* +** The char() function takes zero or more arguments, each of which is +** an integer. It constructs a string where each character of the string +** is the unicode character for the corresponding integer argument. +*/ +static void charFunc( + sqlite3_context *context, + int argc, + sqlite3_value **argv +){ + unsigned char *z, *zOut; + int i; + zOut = z = sqlite3_malloc64( argc*4+1 ); + if( z==0 ){ + sqlite3_result_error_nomem(context); + return; + } + for(i=0; i0x10ffff ) x = 0xfffd; + c = (unsigned)(x & 0x1fffff); + if( c<0x00080 ){ + *zOut++ = (u8)(c&0xFF); + }else if( c<0x00800 ){ + *zOut++ = 0xC0 + (u8)((c>>6)&0x1F); + *zOut++ = 0x80 + (u8)(c & 0x3F); + }else if( c<0x10000 ){ + *zOut++ = 0xE0 + (u8)((c>>12)&0x0F); + *zOut++ = 0x80 + (u8)((c>>6) & 0x3F); + *zOut++ = 0x80 + (u8)(c & 0x3F); + }else{ + *zOut++ = 0xF0 + (u8)((c>>18) & 0x07); + *zOut++ = 0x80 + (u8)((c>>12) & 0x3F); + *zOut++ = 0x80 + (u8)((c>>6) & 0x3F); + *zOut++ = 0x80 + (u8)(c & 0x3F); + } \ + } + sqlite3_result_text64(context, (char*)z, zOut-z, sqlite3_free, SQLITE_UTF8); +} + +/* +** The hex() function. Interpret the argument as a blob. Return +** a hexadecimal rendering as text. +*/ +static void hexFunc( + sqlite3_context *context, + int argc, + sqlite3_value **argv +){ + int i, n; + const unsigned char *pBlob; + char *zHex, *z; + assert( argc==1 ); + UNUSED_PARAMETER(argc); + pBlob = sqlite3_value_blob(argv[0]); + n = sqlite3_value_bytes(argv[0]); + assert( pBlob==sqlite3_value_blob(argv[0]) ); /* No encoding change */ + z = zHex = contextMalloc(context, ((i64)n)*2 + 1); + if( zHex ){ + for(i=0; i>4)&0xf]; + *(z++) = hexdigits[c&0xf]; + } + *z = 0; + sqlite3_result_text(context, zHex, n*2, sqlite3_free); + } +} + +/* +** The zeroblob(N) function returns a zero-filled blob of size N bytes. +*/ +static void zeroblobFunc( + sqlite3_context *context, + int argc, + sqlite3_value **argv +){ + i64 n; + int rc; + assert( argc==1 ); + UNUSED_PARAMETER(argc); + n = sqlite3_value_int64(argv[0]); + if( n<0 ) n = 0; + rc = sqlite3_result_zeroblob64(context, n); /* IMP: R-00293-64994 */ + if( rc ){ + sqlite3_result_error_code(context, rc); + } +} + +/* +** The replace() function. Three arguments are all strings: call +** them A, B, and C. The result is also a string which is derived +** from A by replacing every occurrence of B with C. The match +** must be exact. Collating sequences are not used. +*/ +static void replaceFunc( + sqlite3_context *context, + int argc, + sqlite3_value **argv +){ + const unsigned char *zStr; /* The input string A */ + const unsigned char *zPattern; /* The pattern string B */ + const unsigned char *zRep; /* The replacement string C */ + unsigned char *zOut; /* The output */ + int nStr; /* Size of zStr */ + int nPattern; /* Size of zPattern */ + int nRep; /* Size of zRep */ + i64 nOut; /* Maximum size of zOut */ + int loopLimit; /* Last zStr[] that might match zPattern[] */ + int i, j; /* Loop counters */ + unsigned cntExpand; /* Number zOut expansions */ + sqlite3 *db = sqlite3_context_db_handle(context); + + assert( argc==3 ); + UNUSED_PARAMETER(argc); + zStr = sqlite3_value_text(argv[0]); + if( zStr==0 ) return; + nStr = sqlite3_value_bytes(argv[0]); + assert( zStr==sqlite3_value_text(argv[0]) ); /* No encoding change */ + zPattern = sqlite3_value_text(argv[1]); + if( zPattern==0 ){ + assert( sqlite3_value_type(argv[1])==SQLITE_NULL + || sqlite3_context_db_handle(context)->mallocFailed ); + return; + } + if( zPattern[0]==0 ){ + assert( sqlite3_value_type(argv[1])!=SQLITE_NULL ); + sqlite3_result_value(context, argv[0]); + return; + } + nPattern = sqlite3_value_bytes(argv[1]); + assert( zPattern==sqlite3_value_text(argv[1]) ); /* No encoding change */ + zRep = sqlite3_value_text(argv[2]); + if( zRep==0 ) return; + nRep = sqlite3_value_bytes(argv[2]); + assert( zRep==sqlite3_value_text(argv[2]) ); + nOut = nStr + 1; + assert( nOutnPattern ){ + nOut += nRep - nPattern; + testcase( nOut-1==db->aLimit[SQLITE_LIMIT_LENGTH] ); + testcase( nOut-2==db->aLimit[SQLITE_LIMIT_LENGTH] ); + if( nOut-1>db->aLimit[SQLITE_LIMIT_LENGTH] ){ + sqlite3_result_error_toobig(context); + sqlite3_free(zOut); + return; + } + cntExpand++; + if( (cntExpand&(cntExpand-1))==0 ){ + /* Grow the size of the output buffer only on substitutions + ** whose index is a power of two: 1, 2, 4, 8, 16, 32, ... */ + u8 *zOld; + zOld = zOut; + zOut = sqlite3_realloc64(zOut, (int)nOut + (nOut - nStr - 1)); + if( zOut==0 ){ + sqlite3_result_error_nomem(context); + sqlite3_free(zOld); + return; + } + } + } + memcpy(&zOut[j], zRep, nRep); + j += nRep; + i += nPattern-1; + } + } + assert( j+nStr-i+1<=nOut ); + memcpy(&zOut[j], &zStr[i], nStr-i); + j += nStr - i; + assert( j<=nOut ); + zOut[j] = 0; + sqlite3_result_text(context, (char*)zOut, j, sqlite3_free); +} + +/* +** Implementation of the TRIM(), LTRIM(), and RTRIM() functions. +** The userdata is 0x1 for left trim, 0x2 for right trim, 0x3 for both. +*/ +static void trimFunc( + sqlite3_context *context, + int argc, + sqlite3_value **argv +){ + const unsigned char *zIn; /* Input string */ + const unsigned char *zCharSet; /* Set of characters to trim */ + int nIn; /* Number of bytes in input */ + int flags; /* 1: trimleft 2: trimright 3: trim */ + int i; /* Loop counter */ + unsigned char *aLen = 0; /* Length of each character in zCharSet */ + unsigned char **azChar = 0; /* Individual characters in zCharSet */ + int nChar; /* Number of characters in zCharSet */ + + if( sqlite3_value_type(argv[0])==SQLITE_NULL ){ + return; + } + zIn = sqlite3_value_text(argv[0]); + if( zIn==0 ) return; + nIn = sqlite3_value_bytes(argv[0]); + assert( zIn==sqlite3_value_text(argv[0]) ); + if( argc==1 ){ + static const unsigned char lenOne[] = { 1 }; + static unsigned char * const azOne[] = { (u8*)" " }; + nChar = 1; + aLen = (u8*)lenOne; + azChar = (unsigned char **)azOne; + zCharSet = 0; + }else if( (zCharSet = sqlite3_value_text(argv[1]))==0 ){ + return; + }else{ + const unsigned char *z; + for(z=zCharSet, nChar=0; *z; nChar++){ + SQLITE_SKIP_UTF8(z); + } + if( nChar>0 ){ + azChar = contextMalloc(context, ((i64)nChar)*(sizeof(char*)+1)); + if( azChar==0 ){ + return; + } + aLen = (unsigned char*)&azChar[nChar]; + for(z=zCharSet, nChar=0; *z; nChar++){ + azChar[nChar] = (unsigned char *)z; + SQLITE_SKIP_UTF8(z); + aLen[nChar] = (u8)(z - azChar[nChar]); + } + } + } + if( nChar>0 ){ + flags = SQLITE_PTR_TO_INT(sqlite3_user_data(context)); + if( flags & 1 ){ + while( nIn>0 ){ + int len = 0; + for(i=0; i=nChar ) break; + zIn += len; + nIn -= len; + } + } + if( flags & 2 ){ + while( nIn>0 ){ + int len = 0; + for(i=0; i=nChar ) break; + nIn -= len; + } + } + if( zCharSet ){ + sqlite3_free(azChar); + } + } + sqlite3_result_text(context, (char*)zIn, nIn, SQLITE_TRANSIENT); +} + + +#ifdef SQLITE_ENABLE_UNKNOWN_SQL_FUNCTION +/* +** The "unknown" function is automatically substituted in place of +** any unrecognized function name when doing an EXPLAIN or EXPLAIN QUERY PLAN +** when the SQLITE_ENABLE_UNKNOWN_FUNCTION compile-time option is used. +** When the "sqlite3" command-line shell is built using this functionality, +** that allows an EXPLAIN or EXPLAIN QUERY PLAN for complex queries +** involving application-defined functions to be examined in a generic +** sqlite3 shell. +*/ +static void unknownFunc( + sqlite3_context *context, + int argc, + sqlite3_value **argv +){ + /* no-op */ +} +#endif /*SQLITE_ENABLE_UNKNOWN_SQL_FUNCTION*/ + + +/* IMP: R-25361-16150 This function is omitted from SQLite by default. It +** is only available if the SQLITE_SOUNDEX compile-time option is used +** when SQLite is built. +*/ +#ifdef SQLITE_SOUNDEX +/* +** Compute the soundex encoding of a word. +** +** IMP: R-59782-00072 The soundex(X) function returns a string that is the +** soundex encoding of the string X. +*/ +static void soundexFunc( + sqlite3_context *context, + int argc, + sqlite3_value **argv +){ + char zResult[8]; + const u8 *zIn; + int i, j; + static const unsigned char iCode[] = { + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 1, 2, 3, 0, 1, 2, 0, 0, 2, 2, 4, 5, 5, 0, + 1, 2, 6, 2, 3, 0, 1, 0, 2, 0, 2, 0, 0, 0, 0, 0, + 0, 0, 1, 2, 3, 0, 1, 2, 0, 0, 2, 2, 4, 5, 5, 0, + 1, 2, 6, 2, 3, 0, 1, 0, 2, 0, 2, 0, 0, 0, 0, 0, + }; + assert( argc==1 ); + zIn = (u8*)sqlite3_value_text(argv[0]); + if( zIn==0 ) zIn = (u8*)""; + for(i=0; zIn[i] && !sqlite3Isalpha(zIn[i]); i++){} + if( zIn[i] ){ + u8 prevcode = iCode[zIn[i]&0x7f]; + zResult[0] = sqlite3Toupper(zIn[i]); + for(j=1; j<4 && zIn[i]; i++){ + int code = iCode[zIn[i]&0x7f]; + if( code>0 ){ + if( code!=prevcode ){ + prevcode = code; + zResult[j++] = code + '0'; + } + }else{ + prevcode = 0; + } + } + while( j<4 ){ + zResult[j++] = '0'; + } + zResult[j] = 0; + sqlite3_result_text(context, zResult, 4, SQLITE_TRANSIENT); + }else{ + /* IMP: R-64894-50321 The string "?000" is returned if the argument + ** is NULL or contains no ASCII alphabetic characters. */ + sqlite3_result_text(context, "?000", 4, SQLITE_STATIC); + } +} +#endif /* SQLITE_SOUNDEX */ + +#ifndef SQLITE_OMIT_LOAD_EXTENSION +/* +** A function that loads a shared-library extension then returns NULL. +*/ +static void loadExt(sqlite3_context *context, int argc, sqlite3_value **argv){ + const char *zFile = (const char *)sqlite3_value_text(argv[0]); + const char *zProc; + sqlite3 *db = sqlite3_context_db_handle(context); + char *zErrMsg = 0; + + /* Disallow the load_extension() SQL function unless the SQLITE_LoadExtFunc + ** flag is set. See the sqlite3_enable_load_extension() API. + */ + if( (db->flags & SQLITE_LoadExtFunc)==0 ){ + sqlite3_result_error(context, "not authorized", -1); + return; + } + + if( argc==2 ){ + zProc = (const char *)sqlite3_value_text(argv[1]); + }else{ + zProc = 0; + } + if( zFile && sqlite3_load_extension(db, zFile, zProc, &zErrMsg) ){ + sqlite3_result_error(context, zErrMsg, -1); + sqlite3_free(zErrMsg); + } +} +#endif + + +/* +** An instance of the following structure holds the context of a +** sum() or avg() aggregate computation. +*/ +typedef struct SumCtx SumCtx; +struct SumCtx { + double rSum; /* Floating point sum */ + i64 iSum; /* Integer sum */ + i64 cnt; /* Number of elements summed */ + u8 overflow; /* True if integer overflow seen */ + u8 approx; /* True if non-integer value was input to the sum */ +}; + +/* +** Routines used to compute the sum, average, and total. +** +** The SUM() function follows the (broken) SQL standard which means +** that it returns NULL if it sums over no inputs. TOTAL returns +** 0.0 in that case. In addition, TOTAL always returns a float where +** SUM might return an integer if it never encounters a floating point +** value. TOTAL never fails, but SUM might through an exception if +** it overflows an integer. +*/ +static void sumStep(sqlite3_context *context, int argc, sqlite3_value **argv){ + SumCtx *p; + int type; + assert( argc==1 ); + UNUSED_PARAMETER(argc); + p = sqlite3_aggregate_context(context, sizeof(*p)); + type = sqlite3_value_numeric_type(argv[0]); + if( p && type!=SQLITE_NULL ){ + p->cnt++; + if( type==SQLITE_INTEGER ){ + i64 v = sqlite3_value_int64(argv[0]); + p->rSum += v; + if( (p->approx|p->overflow)==0 && sqlite3AddInt64(&p->iSum, v) ){ + p->approx = p->overflow = 1; + } + }else{ + p->rSum += sqlite3_value_double(argv[0]); + p->approx = 1; + } + } +} +#ifndef SQLITE_OMIT_WINDOWFUNC +static void sumInverse(sqlite3_context *context, int argc, sqlite3_value**argv){ + SumCtx *p; + int type; + assert( argc==1 ); + UNUSED_PARAMETER(argc); + p = sqlite3_aggregate_context(context, sizeof(*p)); + type = sqlite3_value_numeric_type(argv[0]); + /* p is always non-NULL because sumStep() will have been called first + ** to initialize it */ + if( ALWAYS(p) && type!=SQLITE_NULL ){ + assert( p->cnt>0 ); + p->cnt--; + assert( type==SQLITE_INTEGER || p->approx ); + if( type==SQLITE_INTEGER && p->approx==0 ){ + i64 v = sqlite3_value_int64(argv[0]); + p->rSum -= v; + p->iSum -= v; + }else{ + p->rSum -= sqlite3_value_double(argv[0]); + } + } +} +#else +# define sumInverse 0 +#endif /* SQLITE_OMIT_WINDOWFUNC */ +static void sumFinalize(sqlite3_context *context){ + SumCtx *p; + p = sqlite3_aggregate_context(context, 0); + if( p && p->cnt>0 ){ + if( p->overflow ){ + sqlite3_result_error(context,"integer overflow",-1); + }else if( p->approx ){ + sqlite3_result_double(context, p->rSum); + }else{ + sqlite3_result_int64(context, p->iSum); + } + } +} +static void avgFinalize(sqlite3_context *context){ + SumCtx *p; + p = sqlite3_aggregate_context(context, 0); + if( p && p->cnt>0 ){ + sqlite3_result_double(context, p->rSum/(double)p->cnt); + } +} +static void totalFinalize(sqlite3_context *context){ + SumCtx *p; + p = sqlite3_aggregate_context(context, 0); + /* (double)0 In case of SQLITE_OMIT_FLOATING_POINT... */ + sqlite3_result_double(context, p ? p->rSum : (double)0); +} + +/* +** The following structure keeps track of state information for the +** count() aggregate function. +*/ +typedef struct CountCtx CountCtx; +struct CountCtx { + i64 n; +#ifdef SQLITE_DEBUG + int bInverse; /* True if xInverse() ever called */ +#endif +}; + +/* +** Routines to implement the count() aggregate function. +*/ +static void countStep(sqlite3_context *context, int argc, sqlite3_value **argv){ + CountCtx *p; + p = sqlite3_aggregate_context(context, sizeof(*p)); + if( (argc==0 || SQLITE_NULL!=sqlite3_value_type(argv[0])) && p ){ + p->n++; + } + +#ifndef SQLITE_OMIT_DEPRECATED + /* The sqlite3_aggregate_count() function is deprecated. But just to make + ** sure it still operates correctly, verify that its count agrees with our + ** internal count when using count(*) and when the total count can be + ** expressed as a 32-bit integer. */ + assert( argc==1 || p==0 || p->n>0x7fffffff || p->bInverse + || p->n==sqlite3_aggregate_count(context) ); +#endif +} +static void countFinalize(sqlite3_context *context){ + CountCtx *p; + p = sqlite3_aggregate_context(context, 0); + sqlite3_result_int64(context, p ? p->n : 0); +} +#ifndef SQLITE_OMIT_WINDOWFUNC +static void countInverse(sqlite3_context *ctx, int argc, sqlite3_value **argv){ + CountCtx *p; + p = sqlite3_aggregate_context(ctx, sizeof(*p)); + /* p is always non-NULL since countStep() will have been called first */ + if( (argc==0 || SQLITE_NULL!=sqlite3_value_type(argv[0])) && ALWAYS(p) ){ + p->n--; +#ifdef SQLITE_DEBUG + p->bInverse = 1; +#endif + } +} +#else +# define countInverse 0 +#endif /* SQLITE_OMIT_WINDOWFUNC */ + +/* +** Routines to implement min() and max() aggregate functions. +*/ +static void minmaxStep( + sqlite3_context *context, + int NotUsed, + sqlite3_value **argv +){ + Mem *pArg = (Mem *)argv[0]; + Mem *pBest; + UNUSED_PARAMETER(NotUsed); + + pBest = (Mem *)sqlite3_aggregate_context(context, sizeof(*pBest)); + if( !pBest ) return; + + if( sqlite3_value_type(pArg)==SQLITE_NULL ){ + if( pBest->flags ) sqlite3SkipAccumulatorLoad(context); + }else if( pBest->flags ){ + int max; + int cmp; + CollSeq *pColl = sqlite3GetFuncCollSeq(context); + /* This step function is used for both the min() and max() aggregates, + ** the only difference between the two being that the sense of the + ** comparison is inverted. For the max() aggregate, the + ** sqlite3_user_data() function returns (void *)-1. For min() it + ** returns (void *)db, where db is the sqlite3* database pointer. + ** Therefore the next statement sets variable 'max' to 1 for the max() + ** aggregate, or 0 for min(). + */ + max = sqlite3_user_data(context)!=0; + cmp = sqlite3MemCompare(pBest, pArg, pColl); + if( (max && cmp<0) || (!max && cmp>0) ){ + sqlite3VdbeMemCopy(pBest, pArg); + }else{ + sqlite3SkipAccumulatorLoad(context); + } + }else{ + pBest->db = sqlite3_context_db_handle(context); + sqlite3VdbeMemCopy(pBest, pArg); + } +} +static void minMaxValueFinalize(sqlite3_context *context, int bValue){ + sqlite3_value *pRes; + pRes = (sqlite3_value *)sqlite3_aggregate_context(context, 0); + if( pRes ){ + if( pRes->flags ){ + sqlite3_result_value(context, pRes); + } + if( bValue==0 ) sqlite3VdbeMemRelease(pRes); + } +} +#ifndef SQLITE_OMIT_WINDOWFUNC +static void minMaxValue(sqlite3_context *context){ + minMaxValueFinalize(context, 1); +} +#else +# define minMaxValue 0 +#endif /* SQLITE_OMIT_WINDOWFUNC */ +static void minMaxFinalize(sqlite3_context *context){ + minMaxValueFinalize(context, 0); +} + +/* +** group_concat(EXPR, ?SEPARATOR?) +*/ +static void groupConcatStep( + sqlite3_context *context, + int argc, + sqlite3_value **argv +){ + const char *zVal; + StrAccum *pAccum; + const char *zSep; + int nVal, nSep; + assert( argc==1 || argc==2 ); + if( sqlite3_value_type(argv[0])==SQLITE_NULL ) return; + pAccum = (StrAccum*)sqlite3_aggregate_context(context, sizeof(*pAccum)); + + if( pAccum ){ + sqlite3 *db = sqlite3_context_db_handle(context); + int firstTerm = pAccum->mxAlloc==0; + pAccum->mxAlloc = db->aLimit[SQLITE_LIMIT_LENGTH]; + if( !firstTerm ){ + if( argc==2 ){ + zSep = (char*)sqlite3_value_text(argv[1]); + nSep = sqlite3_value_bytes(argv[1]); + }else{ + zSep = ","; + nSep = 1; + } + if( zSep ) sqlite3_str_append(pAccum, zSep, nSep); + } + zVal = (char*)sqlite3_value_text(argv[0]); + nVal = sqlite3_value_bytes(argv[0]); + if( zVal ) sqlite3_str_append(pAccum, zVal, nVal); + } +} +#ifndef SQLITE_OMIT_WINDOWFUNC +static void groupConcatInverse( + sqlite3_context *context, + int argc, + sqlite3_value **argv +){ + int n; + StrAccum *pAccum; + assert( argc==1 || argc==2 ); + if( sqlite3_value_type(argv[0])==SQLITE_NULL ) return; + pAccum = (StrAccum*)sqlite3_aggregate_context(context, sizeof(*pAccum)); + /* pAccum is always non-NULL since groupConcatStep() will have always + ** run frist to initialize it */ + if( ALWAYS(pAccum) ){ + n = sqlite3_value_bytes(argv[0]); + if( argc==2 ){ + n += sqlite3_value_bytes(argv[1]); + }else{ + n++; + } + if( n>=(int)pAccum->nChar ){ + pAccum->nChar = 0; + }else{ + pAccum->nChar -= n; + memmove(pAccum->zText, &pAccum->zText[n], pAccum->nChar); + } + if( pAccum->nChar==0 ) pAccum->mxAlloc = 0; + } +} +#else +# define groupConcatInverse 0 +#endif /* SQLITE_OMIT_WINDOWFUNC */ +static void groupConcatFinalize(sqlite3_context *context){ + StrAccum *pAccum; + pAccum = sqlite3_aggregate_context(context, 0); + if( pAccum ){ + if( pAccum->accError==SQLITE_TOOBIG ){ + sqlite3_result_error_toobig(context); + }else if( pAccum->accError==SQLITE_NOMEM ){ + sqlite3_result_error_nomem(context); + }else{ + sqlite3_result_text(context, sqlite3StrAccumFinish(pAccum), -1, + sqlite3_free); + } + } +} +#ifndef SQLITE_OMIT_WINDOWFUNC +static void groupConcatValue(sqlite3_context *context){ + sqlite3_str *pAccum; + pAccum = (sqlite3_str*)sqlite3_aggregate_context(context, 0); + if( pAccum ){ + if( pAccum->accError==SQLITE_TOOBIG ){ + sqlite3_result_error_toobig(context); + }else if( pAccum->accError==SQLITE_NOMEM ){ + sqlite3_result_error_nomem(context); + }else{ + const char *zText = sqlite3_str_value(pAccum); + sqlite3_result_text(context, zText, -1, SQLITE_TRANSIENT); + } + } +} +#else +# define groupConcatValue 0 +#endif /* SQLITE_OMIT_WINDOWFUNC */ + +/* +** This routine does per-connection function registration. Most +** of the built-in functions above are part of the global function set. +** This routine only deals with those that are not global. +*/ +SQLITE_PRIVATE void sqlite3RegisterPerConnectionBuiltinFunctions(sqlite3 *db){ + int rc = sqlite3_overload_function(db, "MATCH", 2); + assert( rc==SQLITE_NOMEM || rc==SQLITE_OK ); + if( rc==SQLITE_NOMEM ){ + sqlite3OomFault(db); + } +} + +/* +** Re-register the built-in LIKE functions. The caseSensitive +** parameter determines whether or not the LIKE operator is case +** sensitive. +*/ +SQLITE_PRIVATE void sqlite3RegisterLikeFunctions(sqlite3 *db, int caseSensitive){ + struct compareInfo *pInfo; + int flags; + if( caseSensitive ){ + pInfo = (struct compareInfo*)&likeInfoAlt; + flags = SQLITE_FUNC_LIKE | SQLITE_FUNC_CASE; + }else{ + pInfo = (struct compareInfo*)&likeInfoNorm; + flags = SQLITE_FUNC_LIKE; + } + sqlite3CreateFunc(db, "like", 2, SQLITE_UTF8, pInfo, likeFunc, 0, 0, 0, 0, 0); + sqlite3CreateFunc(db, "like", 3, SQLITE_UTF8, pInfo, likeFunc, 0, 0, 0, 0, 0); + sqlite3FindFunction(db, "like", 2, SQLITE_UTF8, 0)->funcFlags |= flags; + sqlite3FindFunction(db, "like", 3, SQLITE_UTF8, 0)->funcFlags |= flags; +} + +/* +** pExpr points to an expression which implements a function. If +** it is appropriate to apply the LIKE optimization to that function +** then set aWc[0] through aWc[2] to the wildcard characters and the +** escape character and then return TRUE. If the function is not a +** LIKE-style function then return FALSE. +** +** The expression "a LIKE b ESCAPE c" is only considered a valid LIKE +** operator if c is a string literal that is exactly one byte in length. +** That one byte is stored in aWc[3]. aWc[3] is set to zero if there is +** no ESCAPE clause. +** +** *pIsNocase is set to true if uppercase and lowercase are equivalent for +** the function (default for LIKE). If the function makes the distinction +** between uppercase and lowercase (as does GLOB) then *pIsNocase is set to +** false. +*/ +SQLITE_PRIVATE int sqlite3IsLikeFunction(sqlite3 *db, Expr *pExpr, int *pIsNocase, char *aWc){ + FuncDef *pDef; + int nExpr; + if( pExpr->op!=TK_FUNCTION || !pExpr->x.pList ){ + return 0; + } + assert( !ExprHasProperty(pExpr, EP_xIsSelect) ); + nExpr = pExpr->x.pList->nExpr; + pDef = sqlite3FindFunction(db, pExpr->u.zToken, nExpr, SQLITE_UTF8, 0); + if( NEVER(pDef==0) || (pDef->funcFlags & SQLITE_FUNC_LIKE)==0 ){ + return 0; + } + if( nExpr<3 ){ + aWc[3] = 0; + }else{ + Expr *pEscape = pExpr->x.pList->a[2].pExpr; + char *zEscape; + if( pEscape->op!=TK_STRING ) return 0; + zEscape = pEscape->u.zToken; + if( zEscape[0]==0 || zEscape[1]!=0 ) return 0; + aWc[3] = zEscape[0]; + } + + /* The memcpy() statement assumes that the wildcard characters are + ** the first three statements in the compareInfo structure. The + ** asserts() that follow verify that assumption + */ + memcpy(aWc, pDef->pUserData, 3); + assert( (char*)&likeInfoAlt == (char*)&likeInfoAlt.matchAll ); + assert( &((char*)&likeInfoAlt)[1] == (char*)&likeInfoAlt.matchOne ); + assert( &((char*)&likeInfoAlt)[2] == (char*)&likeInfoAlt.matchSet ); + *pIsNocase = (pDef->funcFlags & SQLITE_FUNC_CASE)==0; + return 1; +} + +/* +** All of the FuncDef structures in the aBuiltinFunc[] array above +** to the global function hash table. This occurs at start-time (as +** a consequence of calling sqlite3_initialize()). +** +** After this routine runs +*/ +SQLITE_PRIVATE void sqlite3RegisterBuiltinFunctions(void){ + /* + ** The following array holds FuncDef structures for all of the functions + ** defined in this file. + ** + ** The array cannot be constant since changes are made to the + ** FuncDef.pHash elements at start-time. The elements of this array + ** are read-only after initialization is complete. + ** + ** For peak efficiency, put the most frequently used function last. + */ + static FuncDef aBuiltinFunc[] = { +#ifdef SQLITE_SOUNDEX + FUNCTION(soundex, 1, 0, 0, soundexFunc ), +#endif +#ifndef SQLITE_OMIT_LOAD_EXTENSION + VFUNCTION(load_extension, 1, 0, 0, loadExt ), + VFUNCTION(load_extension, 2, 0, 0, loadExt ), +#endif +#if SQLITE_USER_AUTHENTICATION + FUNCTION(sqlite_crypt, 2, 0, 0, sqlite3CryptFunc ), +#endif +#ifndef SQLITE_OMIT_COMPILEOPTION_DIAGS + DFUNCTION(sqlite_compileoption_used,1, 0, 0, compileoptionusedFunc ), + DFUNCTION(sqlite_compileoption_get, 1, 0, 0, compileoptiongetFunc ), +#endif /* SQLITE_OMIT_COMPILEOPTION_DIAGS */ + FUNCTION2(unlikely, 1, 0, 0, noopFunc, SQLITE_FUNC_UNLIKELY), + FUNCTION2(likelihood, 2, 0, 0, noopFunc, SQLITE_FUNC_UNLIKELY), + FUNCTION2(likely, 1, 0, 0, noopFunc, SQLITE_FUNC_UNLIKELY), +#ifdef SQLITE_DEBUG + FUNCTION2(affinity, 1, 0, 0, noopFunc, SQLITE_FUNC_AFFINITY), +#endif +#ifdef SQLITE_ENABLE_OFFSET_SQL_FUNC + FUNCTION2(sqlite_offset, 1, 0, 0, noopFunc, SQLITE_FUNC_OFFSET| + SQLITE_FUNC_TYPEOF), +#endif + FUNCTION(ltrim, 1, 1, 0, trimFunc ), + FUNCTION(ltrim, 2, 1, 0, trimFunc ), + FUNCTION(rtrim, 1, 2, 0, trimFunc ), + FUNCTION(rtrim, 2, 2, 0, trimFunc ), + FUNCTION(trim, 1, 3, 0, trimFunc ), + FUNCTION(trim, 2, 3, 0, trimFunc ), + FUNCTION(min, -1, 0, 1, minmaxFunc ), + FUNCTION(min, 0, 0, 1, 0 ), + WAGGREGATE(min, 1, 0, 1, minmaxStep, minMaxFinalize, minMaxValue, 0, + SQLITE_FUNC_MINMAX ), + FUNCTION(max, -1, 1, 1, minmaxFunc ), + FUNCTION(max, 0, 1, 1, 0 ), + WAGGREGATE(max, 1, 1, 1, minmaxStep, minMaxFinalize, minMaxValue, 0, + SQLITE_FUNC_MINMAX ), + FUNCTION2(typeof, 1, 0, 0, typeofFunc, SQLITE_FUNC_TYPEOF), + FUNCTION2(length, 1, 0, 0, lengthFunc, SQLITE_FUNC_LENGTH), + FUNCTION(instr, 2, 0, 0, instrFunc ), + FUNCTION(printf, -1, 0, 0, printfFunc ), + FUNCTION(unicode, 1, 0, 0, unicodeFunc ), + FUNCTION(char, -1, 0, 0, charFunc ), + FUNCTION(abs, 1, 0, 0, absFunc ), +#ifndef SQLITE_OMIT_FLOATING_POINT + FUNCTION(round, 1, 0, 0, roundFunc ), + FUNCTION(round, 2, 0, 0, roundFunc ), +#endif + FUNCTION(upper, 1, 0, 0, upperFunc ), + FUNCTION(lower, 1, 0, 0, lowerFunc ), + FUNCTION(hex, 1, 0, 0, hexFunc ), + FUNCTION2(ifnull, 2, 0, 0, noopFunc, SQLITE_FUNC_COALESCE), + VFUNCTION(random, 0, 0, 0, randomFunc ), + VFUNCTION(randomblob, 1, 0, 0, randomBlob ), + FUNCTION(nullif, 2, 0, 1, nullifFunc ), + DFUNCTION(sqlite_version, 0, 0, 0, versionFunc ), + DFUNCTION(sqlite_source_id, 0, 0, 0, sourceidFunc ), + FUNCTION(sqlite_log, 2, 0, 0, errlogFunc ), + FUNCTION(quote, 1, 0, 0, quoteFunc ), + VFUNCTION(last_insert_rowid, 0, 0, 0, last_insert_rowid), + VFUNCTION(changes, 0, 0, 0, changes ), + VFUNCTION(total_changes, 0, 0, 0, total_changes ), + FUNCTION(replace, 3, 0, 0, replaceFunc ), + FUNCTION(zeroblob, 1, 0, 0, zeroblobFunc ), + FUNCTION(substr, 2, 0, 0, substrFunc ), + FUNCTION(substr, 3, 0, 0, substrFunc ), + WAGGREGATE(sum, 1,0,0, sumStep, sumFinalize, sumFinalize, sumInverse, 0), + WAGGREGATE(total, 1,0,0, sumStep,totalFinalize,totalFinalize,sumInverse, 0), + WAGGREGATE(avg, 1,0,0, sumStep, avgFinalize, avgFinalize, sumInverse, 0), + WAGGREGATE(count, 0,0,0, countStep, + countFinalize, countFinalize, countInverse, SQLITE_FUNC_COUNT ), + WAGGREGATE(count, 1,0,0, countStep, + countFinalize, countFinalize, countInverse, 0 ), + WAGGREGATE(group_concat, 1, 0, 0, groupConcatStep, + groupConcatFinalize, groupConcatValue, groupConcatInverse, 0), + WAGGREGATE(group_concat, 2, 0, 0, groupConcatStep, + groupConcatFinalize, groupConcatValue, groupConcatInverse, 0), + + LIKEFUNC(glob, 2, &globInfo, SQLITE_FUNC_LIKE|SQLITE_FUNC_CASE), +#ifdef SQLITE_CASE_SENSITIVE_LIKE + LIKEFUNC(like, 2, &likeInfoAlt, SQLITE_FUNC_LIKE|SQLITE_FUNC_CASE), + LIKEFUNC(like, 3, &likeInfoAlt, SQLITE_FUNC_LIKE|SQLITE_FUNC_CASE), +#else + LIKEFUNC(like, 2, &likeInfoNorm, SQLITE_FUNC_LIKE), + LIKEFUNC(like, 3, &likeInfoNorm, SQLITE_FUNC_LIKE), +#endif +#ifdef SQLITE_ENABLE_UNKNOWN_SQL_FUNCTION + FUNCTION(unknown, -1, 0, 0, unknownFunc ), +#endif + FUNCTION(coalesce, 1, 0, 0, 0 ), + FUNCTION(coalesce, 0, 0, 0, 0 ), + FUNCTION2(coalesce, -1, 0, 0, noopFunc, SQLITE_FUNC_COALESCE), + }; +#ifndef SQLITE_OMIT_ALTERTABLE + sqlite3AlterFunctions(); +#endif + sqlite3WindowFunctions(); + sqlite3RegisterDateTimeFunctions(); + sqlite3InsertBuiltinFuncs(aBuiltinFunc, ArraySize(aBuiltinFunc)); + +#if 0 /* Enable to print out how the built-in functions are hashed */ + { + int i; + FuncDef *p; + for(i=0; iu.pHash){ + int n = sqlite3Strlen30(p->zName); + int h = p->zName[0] + n; + printf(" %s(%d)", p->zName, h); + } + printf("\n"); + } + } +#endif +} + +/************** End of func.c ************************************************/ +/************** Begin file fkey.c ********************************************/ +/* +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** This file contains code used by the compiler to add foreign key +** support to compiled SQL statements. +*/ +/* #include "sqliteInt.h" */ + +#ifndef SQLITE_OMIT_FOREIGN_KEY +#ifndef SQLITE_OMIT_TRIGGER + +/* +** Deferred and Immediate FKs +** -------------------------- +** +** Foreign keys in SQLite come in two flavours: deferred and immediate. +** If an immediate foreign key constraint is violated, +** SQLITE_CONSTRAINT_FOREIGNKEY is returned and the current +** statement transaction rolled back. If a +** deferred foreign key constraint is violated, no action is taken +** immediately. However if the application attempts to commit the +** transaction before fixing the constraint violation, the attempt fails. +** +** Deferred constraints are implemented using a simple counter associated +** with the database handle. The counter is set to zero each time a +** database transaction is opened. Each time a statement is executed +** that causes a foreign key violation, the counter is incremented. Each +** time a statement is executed that removes an existing violation from +** the database, the counter is decremented. When the transaction is +** committed, the commit fails if the current value of the counter is +** greater than zero. This scheme has two big drawbacks: +** +** * When a commit fails due to a deferred foreign key constraint, +** there is no way to tell which foreign constraint is not satisfied, +** or which row it is not satisfied for. +** +** * If the database contains foreign key violations when the +** transaction is opened, this may cause the mechanism to malfunction. +** +** Despite these problems, this approach is adopted as it seems simpler +** than the alternatives. +** +** INSERT operations: +** +** I.1) For each FK for which the table is the child table, search +** the parent table for a match. If none is found increment the +** constraint counter. +** +** I.2) For each FK for which the table is the parent table, +** search the child table for rows that correspond to the new +** row in the parent table. Decrement the counter for each row +** found (as the constraint is now satisfied). +** +** DELETE operations: +** +** D.1) For each FK for which the table is the child table, +** search the parent table for a row that corresponds to the +** deleted row in the child table. If such a row is not found, +** decrement the counter. +** +** D.2) For each FK for which the table is the parent table, search +** the child table for rows that correspond to the deleted row +** in the parent table. For each found increment the counter. +** +** UPDATE operations: +** +** An UPDATE command requires that all 4 steps above are taken, but only +** for FK constraints for which the affected columns are actually +** modified (values must be compared at runtime). +** +** Note that I.1 and D.1 are very similar operations, as are I.2 and D.2. +** This simplifies the implementation a bit. +** +** For the purposes of immediate FK constraints, the OR REPLACE conflict +** resolution is considered to delete rows before the new row is inserted. +** If a delete caused by OR REPLACE violates an FK constraint, an exception +** is thrown, even if the FK constraint would be satisfied after the new +** row is inserted. +** +** Immediate constraints are usually handled similarly. The only difference +** is that the counter used is stored as part of each individual statement +** object (struct Vdbe). If, after the statement has run, its immediate +** constraint counter is greater than zero, +** it returns SQLITE_CONSTRAINT_FOREIGNKEY +** and the statement transaction is rolled back. An exception is an INSERT +** statement that inserts a single row only (no triggers). In this case, +** instead of using a counter, an exception is thrown immediately if the +** INSERT violates a foreign key constraint. This is necessary as such +** an INSERT does not open a statement transaction. +** +** TODO: How should dropping a table be handled? How should renaming a +** table be handled? +** +** +** Query API Notes +** --------------- +** +** Before coding an UPDATE or DELETE row operation, the code-generator +** for those two operations needs to know whether or not the operation +** requires any FK processing and, if so, which columns of the original +** row are required by the FK processing VDBE code (i.e. if FKs were +** implemented using triggers, which of the old.* columns would be +** accessed). No information is required by the code-generator before +** coding an INSERT operation. The functions used by the UPDATE/DELETE +** generation code to query for this information are: +** +** sqlite3FkRequired() - Test to see if FK processing is required. +** sqlite3FkOldmask() - Query for the set of required old.* columns. +** +** +** Externally accessible module functions +** -------------------------------------- +** +** sqlite3FkCheck() - Check for foreign key violations. +** sqlite3FkActions() - Code triggers for ON UPDATE/ON DELETE actions. +** sqlite3FkDelete() - Delete an FKey structure. +*/ + +/* +** VDBE Calling Convention +** ----------------------- +** +** Example: +** +** For the following INSERT statement: +** +** CREATE TABLE t1(a, b INTEGER PRIMARY KEY, c); +** INSERT INTO t1 VALUES(1, 2, 3.1); +** +** Register (x): 2 (type integer) +** Register (x+1): 1 (type integer) +** Register (x+2): NULL (type NULL) +** Register (x+3): 3.1 (type real) +*/ + +/* +** A foreign key constraint requires that the key columns in the parent +** table are collectively subject to a UNIQUE or PRIMARY KEY constraint. +** Given that pParent is the parent table for foreign key constraint pFKey, +** search the schema for a unique index on the parent key columns. +** +** If successful, zero is returned. If the parent key is an INTEGER PRIMARY +** KEY column, then output variable *ppIdx is set to NULL. Otherwise, *ppIdx +** is set to point to the unique index. +** +** If the parent key consists of a single column (the foreign key constraint +** is not a composite foreign key), output variable *paiCol is set to NULL. +** Otherwise, it is set to point to an allocated array of size N, where +** N is the number of columns in the parent key. The first element of the +** array is the index of the child table column that is mapped by the FK +** constraint to the parent table column stored in the left-most column +** of index *ppIdx. The second element of the array is the index of the +** child table column that corresponds to the second left-most column of +** *ppIdx, and so on. +** +** If the required index cannot be found, either because: +** +** 1) The named parent key columns do not exist, or +** +** 2) The named parent key columns do exist, but are not subject to a +** UNIQUE or PRIMARY KEY constraint, or +** +** 3) No parent key columns were provided explicitly as part of the +** foreign key definition, and the parent table does not have a +** PRIMARY KEY, or +** +** 4) No parent key columns were provided explicitly as part of the +** foreign key definition, and the PRIMARY KEY of the parent table +** consists of a different number of columns to the child key in +** the child table. +** +** then non-zero is returned, and a "foreign key mismatch" error loaded +** into pParse. If an OOM error occurs, non-zero is returned and the +** pParse->db->mallocFailed flag is set. +*/ +SQLITE_PRIVATE int sqlite3FkLocateIndex( + Parse *pParse, /* Parse context to store any error in */ + Table *pParent, /* Parent table of FK constraint pFKey */ + FKey *pFKey, /* Foreign key to find index for */ + Index **ppIdx, /* OUT: Unique index on parent table */ + int **paiCol /* OUT: Map of index columns in pFKey */ +){ + Index *pIdx = 0; /* Value to return via *ppIdx */ + int *aiCol = 0; /* Value to return via *paiCol */ + int nCol = pFKey->nCol; /* Number of columns in parent key */ + char *zKey = pFKey->aCol[0].zCol; /* Name of left-most parent key column */ + + /* The caller is responsible for zeroing output parameters. */ + assert( ppIdx && *ppIdx==0 ); + assert( !paiCol || *paiCol==0 ); + assert( pParse ); + + /* If this is a non-composite (single column) foreign key, check if it + ** maps to the INTEGER PRIMARY KEY of table pParent. If so, leave *ppIdx + ** and *paiCol set to zero and return early. + ** + ** Otherwise, for a composite foreign key (more than one column), allocate + ** space for the aiCol array (returned via output parameter *paiCol). + ** Non-composite foreign keys do not require the aiCol array. + */ + if( nCol==1 ){ + /* The FK maps to the IPK if any of the following are true: + ** + ** 1) There is an INTEGER PRIMARY KEY column and the FK is implicitly + ** mapped to the primary key of table pParent, or + ** 2) The FK is explicitly mapped to a column declared as INTEGER + ** PRIMARY KEY. + */ + if( pParent->iPKey>=0 ){ + if( !zKey ) return 0; + if( !sqlite3StrICmp(pParent->aCol[pParent->iPKey].zName, zKey) ) return 0; + } + }else if( paiCol ){ + assert( nCol>1 ); + aiCol = (int *)sqlite3DbMallocRawNN(pParse->db, nCol*sizeof(int)); + if( !aiCol ) return 1; + *paiCol = aiCol; + } + + for(pIdx=pParent->pIndex; pIdx; pIdx=pIdx->pNext){ + if( pIdx->nKeyCol==nCol && IsUniqueIndex(pIdx) && pIdx->pPartIdxWhere==0 ){ + /* pIdx is a UNIQUE index (or a PRIMARY KEY) and has the right number + ** of columns. If each indexed column corresponds to a foreign key + ** column of pFKey, then this index is a winner. */ + + if( zKey==0 ){ + /* If zKey is NULL, then this foreign key is implicitly mapped to + ** the PRIMARY KEY of table pParent. The PRIMARY KEY index may be + ** identified by the test. */ + if( IsPrimaryKeyIndex(pIdx) ){ + if( aiCol ){ + int i; + for(i=0; iaCol[i].iFrom; + } + break; + } + }else{ + /* If zKey is non-NULL, then this foreign key was declared to + ** map to an explicit list of columns in table pParent. Check if this + ** index matches those columns. Also, check that the index uses + ** the default collation sequences for each column. */ + int i, j; + for(i=0; iaiColumn[i]; /* Index of column in parent tbl */ + const char *zDfltColl; /* Def. collation for column */ + char *zIdxCol; /* Name of indexed column */ + + if( iCol<0 ) break; /* No foreign keys against expression indexes */ + + /* If the index uses a collation sequence that is different from + ** the default collation sequence for the column, this index is + ** unusable. Bail out early in this case. */ + zDfltColl = pParent->aCol[iCol].zColl; + if( !zDfltColl ) zDfltColl = sqlite3StrBINARY; + if( sqlite3StrICmp(pIdx->azColl[i], zDfltColl) ) break; + + zIdxCol = pParent->aCol[iCol].zName; + for(j=0; jaCol[j].zCol, zIdxCol)==0 ){ + if( aiCol ) aiCol[i] = pFKey->aCol[j].iFrom; + break; + } + } + if( j==nCol ) break; + } + if( i==nCol ) break; /* pIdx is usable */ + } + } + } + + if( !pIdx ){ + if( !pParse->disableTriggers ){ + sqlite3ErrorMsg(pParse, + "foreign key mismatch - \"%w\" referencing \"%w\"", + pFKey->pFrom->zName, pFKey->zTo); + } + sqlite3DbFree(pParse->db, aiCol); + return 1; + } + + *ppIdx = pIdx; + return 0; +} + +/* +** This function is called when a row is inserted into or deleted from the +** child table of foreign key constraint pFKey. If an SQL UPDATE is executed +** on the child table of pFKey, this function is invoked twice for each row +** affected - once to "delete" the old row, and then again to "insert" the +** new row. +** +** Each time it is called, this function generates VDBE code to locate the +** row in the parent table that corresponds to the row being inserted into +** or deleted from the child table. If the parent row can be found, no +** special action is taken. Otherwise, if the parent row can *not* be +** found in the parent table: +** +** Operation | FK type | Action taken +** -------------------------------------------------------------------------- +** INSERT immediate Increment the "immediate constraint counter". +** +** DELETE immediate Decrement the "immediate constraint counter". +** +** INSERT deferred Increment the "deferred constraint counter". +** +** DELETE deferred Decrement the "deferred constraint counter". +** +** These operations are identified in the comment at the top of this file +** (fkey.c) as "I.1" and "D.1". +*/ +static void fkLookupParent( + Parse *pParse, /* Parse context */ + int iDb, /* Index of database housing pTab */ + Table *pTab, /* Parent table of FK pFKey */ + Index *pIdx, /* Unique index on parent key columns in pTab */ + FKey *pFKey, /* Foreign key constraint */ + int *aiCol, /* Map from parent key columns to child table columns */ + int regData, /* Address of array containing child table row */ + int nIncr, /* Increment constraint counter by this */ + int isIgnore /* If true, pretend pTab contains all NULL values */ +){ + int i; /* Iterator variable */ + Vdbe *v = sqlite3GetVdbe(pParse); /* Vdbe to add code to */ + int iCur = pParse->nTab - 1; /* Cursor number to use */ + int iOk = sqlite3VdbeMakeLabel(pParse); /* jump here if parent key found */ + + sqlite3VdbeVerifyAbortable(v, + (!pFKey->isDeferred + && !(pParse->db->flags & SQLITE_DeferFKs) + && !pParse->pToplevel + && !pParse->isMultiWrite) ? OE_Abort : OE_Ignore); + + /* If nIncr is less than zero, then check at runtime if there are any + ** outstanding constraints to resolve. If there are not, there is no need + ** to check if deleting this row resolves any outstanding violations. + ** + ** Check if any of the key columns in the child table row are NULL. If + ** any are, then the constraint is considered satisfied. No need to + ** search for a matching row in the parent table. */ + if( nIncr<0 ){ + sqlite3VdbeAddOp2(v, OP_FkIfZero, pFKey->isDeferred, iOk); + VdbeCoverage(v); + } + for(i=0; inCol; i++){ + int iReg = aiCol[i] + regData + 1; + sqlite3VdbeAddOp2(v, OP_IsNull, iReg, iOk); VdbeCoverage(v); + } + + if( isIgnore==0 ){ + if( pIdx==0 ){ + /* If pIdx is NULL, then the parent key is the INTEGER PRIMARY KEY + ** column of the parent table (table pTab). */ + int iMustBeInt; /* Address of MustBeInt instruction */ + int regTemp = sqlite3GetTempReg(pParse); + + /* Invoke MustBeInt to coerce the child key value to an integer (i.e. + ** apply the affinity of the parent key). If this fails, then there + ** is no matching parent key. Before using MustBeInt, make a copy of + ** the value. Otherwise, the value inserted into the child key column + ** will have INTEGER affinity applied to it, which may not be correct. */ + sqlite3VdbeAddOp2(v, OP_SCopy, aiCol[0]+1+regData, regTemp); + iMustBeInt = sqlite3VdbeAddOp2(v, OP_MustBeInt, regTemp, 0); + VdbeCoverage(v); + + /* If the parent table is the same as the child table, and we are about + ** to increment the constraint-counter (i.e. this is an INSERT operation), + ** then check if the row being inserted matches itself. If so, do not + ** increment the constraint-counter. */ + if( pTab==pFKey->pFrom && nIncr==1 ){ + sqlite3VdbeAddOp3(v, OP_Eq, regData, iOk, regTemp); VdbeCoverage(v); + sqlite3VdbeChangeP5(v, SQLITE_NOTNULL); + } + + sqlite3OpenTable(pParse, iCur, iDb, pTab, OP_OpenRead); + sqlite3VdbeAddOp3(v, OP_NotExists, iCur, 0, regTemp); VdbeCoverage(v); + sqlite3VdbeGoto(v, iOk); + sqlite3VdbeJumpHere(v, sqlite3VdbeCurrentAddr(v)-2); + sqlite3VdbeJumpHere(v, iMustBeInt); + sqlite3ReleaseTempReg(pParse, regTemp); + }else{ + int nCol = pFKey->nCol; + int regTemp = sqlite3GetTempRange(pParse, nCol); + int regRec = sqlite3GetTempReg(pParse); + + sqlite3VdbeAddOp3(v, OP_OpenRead, iCur, pIdx->tnum, iDb); + sqlite3VdbeSetP4KeyInfo(pParse, pIdx); + for(i=0; ipFrom && nIncr==1 ){ + int iJump = sqlite3VdbeCurrentAddr(v) + nCol + 1; + for(i=0; iaiColumn[i]+1+regData; + assert( pIdx->aiColumn[i]>=0 ); + assert( aiCol[i]!=pTab->iPKey ); + if( pIdx->aiColumn[i]==pTab->iPKey ){ + /* The parent key is a composite key that includes the IPK column */ + iParent = regData; + } + sqlite3VdbeAddOp3(v, OP_Ne, iChild, iJump, iParent); VdbeCoverage(v); + sqlite3VdbeChangeP5(v, SQLITE_JUMPIFNULL); + } + sqlite3VdbeGoto(v, iOk); + } + + sqlite3VdbeAddOp4(v, OP_MakeRecord, regTemp, nCol, regRec, + sqlite3IndexAffinityStr(pParse->db,pIdx), nCol); + sqlite3VdbeAddOp4Int(v, OP_Found, iCur, iOk, regRec, 0); VdbeCoverage(v); + + sqlite3ReleaseTempReg(pParse, regRec); + sqlite3ReleaseTempRange(pParse, regTemp, nCol); + } + } + + if( !pFKey->isDeferred && !(pParse->db->flags & SQLITE_DeferFKs) + && !pParse->pToplevel + && !pParse->isMultiWrite + ){ + /* Special case: If this is an INSERT statement that will insert exactly + ** one row into the table, raise a constraint immediately instead of + ** incrementing a counter. This is necessary as the VM code is being + ** generated for will not open a statement transaction. */ + assert( nIncr==1 ); + sqlite3HaltConstraint(pParse, SQLITE_CONSTRAINT_FOREIGNKEY, + OE_Abort, 0, P4_STATIC, P5_ConstraintFK); + }else{ + if( nIncr>0 && pFKey->isDeferred==0 ){ + sqlite3MayAbort(pParse); + } + sqlite3VdbeAddOp2(v, OP_FkCounter, pFKey->isDeferred, nIncr); + } + + sqlite3VdbeResolveLabel(v, iOk); + sqlite3VdbeAddOp1(v, OP_Close, iCur); +} + + +/* +** Return an Expr object that refers to a memory register corresponding +** to column iCol of table pTab. +** +** regBase is the first of an array of register that contains the data +** for pTab. regBase itself holds the rowid. regBase+1 holds the first +** column. regBase+2 holds the second column, and so forth. +*/ +static Expr *exprTableRegister( + Parse *pParse, /* Parsing and code generating context */ + Table *pTab, /* The table whose content is at r[regBase]... */ + int regBase, /* Contents of table pTab */ + i16 iCol /* Which column of pTab is desired */ +){ + Expr *pExpr; + Column *pCol; + const char *zColl; + sqlite3 *db = pParse->db; + + pExpr = sqlite3Expr(db, TK_REGISTER, 0); + if( pExpr ){ + if( iCol>=0 && iCol!=pTab->iPKey ){ + pCol = &pTab->aCol[iCol]; + pExpr->iTable = regBase + iCol + 1; + pExpr->affExpr = pCol->affinity; + zColl = pCol->zColl; + if( zColl==0 ) zColl = db->pDfltColl->zName; + pExpr = sqlite3ExprAddCollateString(pParse, pExpr, zColl); + }else{ + pExpr->iTable = regBase; + pExpr->affExpr = SQLITE_AFF_INTEGER; + } + } + return pExpr; +} + +/* +** Return an Expr object that refers to column iCol of table pTab which +** has cursor iCur. +*/ +static Expr *exprTableColumn( + sqlite3 *db, /* The database connection */ + Table *pTab, /* The table whose column is desired */ + int iCursor, /* The open cursor on the table */ + i16 iCol /* The column that is wanted */ +){ + Expr *pExpr = sqlite3Expr(db, TK_COLUMN, 0); + if( pExpr ){ + pExpr->y.pTab = pTab; + pExpr->iTable = iCursor; + pExpr->iColumn = iCol; + } + return pExpr; +} + +/* +** This function is called to generate code executed when a row is deleted +** from the parent table of foreign key constraint pFKey and, if pFKey is +** deferred, when a row is inserted into the same table. When generating +** code for an SQL UPDATE operation, this function may be called twice - +** once to "delete" the old row and once to "insert" the new row. +** +** Parameter nIncr is passed -1 when inserting a row (as this may decrease +** the number of FK violations in the db) or +1 when deleting one (as this +** may increase the number of FK constraint problems). +** +** The code generated by this function scans through the rows in the child +** table that correspond to the parent table row being deleted or inserted. +** For each child row found, one of the following actions is taken: +** +** Operation | FK type | Action taken +** -------------------------------------------------------------------------- +** DELETE immediate Increment the "immediate constraint counter". +** Or, if the ON (UPDATE|DELETE) action is RESTRICT, +** throw a "FOREIGN KEY constraint failed" exception. +** +** INSERT immediate Decrement the "immediate constraint counter". +** +** DELETE deferred Increment the "deferred constraint counter". +** Or, if the ON (UPDATE|DELETE) action is RESTRICT, +** throw a "FOREIGN KEY constraint failed" exception. +** +** INSERT deferred Decrement the "deferred constraint counter". +** +** These operations are identified in the comment at the top of this file +** (fkey.c) as "I.2" and "D.2". +*/ +static void fkScanChildren( + Parse *pParse, /* Parse context */ + SrcList *pSrc, /* The child table to be scanned */ + Table *pTab, /* The parent table */ + Index *pIdx, /* Index on parent covering the foreign key */ + FKey *pFKey, /* The foreign key linking pSrc to pTab */ + int *aiCol, /* Map from pIdx cols to child table cols */ + int regData, /* Parent row data starts here */ + int nIncr /* Amount to increment deferred counter by */ +){ + sqlite3 *db = pParse->db; /* Database handle */ + int i; /* Iterator variable */ + Expr *pWhere = 0; /* WHERE clause to scan with */ + NameContext sNameContext; /* Context used to resolve WHERE clause */ + WhereInfo *pWInfo; /* Context used by sqlite3WhereXXX() */ + int iFkIfZero = 0; /* Address of OP_FkIfZero */ + Vdbe *v = sqlite3GetVdbe(pParse); + + assert( pIdx==0 || pIdx->pTable==pTab ); + assert( pIdx==0 || pIdx->nKeyCol==pFKey->nCol ); + assert( pIdx!=0 || pFKey->nCol==1 ); + assert( pIdx!=0 || HasRowid(pTab) ); + + if( nIncr<0 ){ + iFkIfZero = sqlite3VdbeAddOp2(v, OP_FkIfZero, pFKey->isDeferred, 0); + VdbeCoverage(v); + } + + /* Create an Expr object representing an SQL expression like: + ** + ** = AND = ... + ** + ** The collation sequence used for the comparison should be that of + ** the parent key columns. The affinity of the parent key column should + ** be applied to each child key value before the comparison takes place. + */ + for(i=0; inCol; i++){ + Expr *pLeft; /* Value from parent table row */ + Expr *pRight; /* Column ref to child table */ + Expr *pEq; /* Expression (pLeft = pRight) */ + i16 iCol; /* Index of column in child table */ + const char *zCol; /* Name of column in child table */ + + iCol = pIdx ? pIdx->aiColumn[i] : -1; + pLeft = exprTableRegister(pParse, pTab, regData, iCol); + iCol = aiCol ? aiCol[i] : pFKey->aCol[0].iFrom; + assert( iCol>=0 ); + zCol = pFKey->pFrom->aCol[iCol].zName; + pRight = sqlite3Expr(db, TK_ID, zCol); + pEq = sqlite3PExpr(pParse, TK_EQ, pLeft, pRight); + pWhere = sqlite3ExprAnd(pParse, pWhere, pEq); + } + + /* If the child table is the same as the parent table, then add terms + ** to the WHERE clause that prevent this entry from being scanned. + ** The added WHERE clause terms are like this: + ** + ** $current_rowid!=rowid + ** NOT( $current_a==a AND $current_b==b AND ... ) + ** + ** The first form is used for rowid tables. The second form is used + ** for WITHOUT ROWID tables. In the second form, the *parent* key is + ** (a,b,...). Either the parent or primary key could be used to + ** uniquely identify the current row, but the parent key is more convenient + ** as the required values have already been loaded into registers + ** by the caller. + */ + if( pTab==pFKey->pFrom && nIncr>0 ){ + Expr *pNe; /* Expression (pLeft != pRight) */ + Expr *pLeft; /* Value from parent table row */ + Expr *pRight; /* Column ref to child table */ + if( HasRowid(pTab) ){ + pLeft = exprTableRegister(pParse, pTab, regData, -1); + pRight = exprTableColumn(db, pTab, pSrc->a[0].iCursor, -1); + pNe = sqlite3PExpr(pParse, TK_NE, pLeft, pRight); + }else{ + Expr *pEq, *pAll = 0; + assert( pIdx!=0 ); + for(i=0; inKeyCol; i++){ + i16 iCol = pIdx->aiColumn[i]; + assert( iCol>=0 ); + pLeft = exprTableRegister(pParse, pTab, regData, iCol); + pRight = sqlite3Expr(db, TK_ID, pTab->aCol[iCol].zName); + pEq = sqlite3PExpr(pParse, TK_IS, pLeft, pRight); + pAll = sqlite3ExprAnd(pParse, pAll, pEq); + } + pNe = sqlite3PExpr(pParse, TK_NOT, pAll, 0); + } + pWhere = sqlite3ExprAnd(pParse, pWhere, pNe); + } + + /* Resolve the references in the WHERE clause. */ + memset(&sNameContext, 0, sizeof(NameContext)); + sNameContext.pSrcList = pSrc; + sNameContext.pParse = pParse; + sqlite3ResolveExprNames(&sNameContext, pWhere); + + /* Create VDBE to loop through the entries in pSrc that match the WHERE + ** clause. For each row found, increment either the deferred or immediate + ** foreign key constraint counter. */ + if( pParse->nErr==0 ){ + pWInfo = sqlite3WhereBegin(pParse, pSrc, pWhere, 0, 0, 0, 0); + sqlite3VdbeAddOp2(v, OP_FkCounter, pFKey->isDeferred, nIncr); + if( pWInfo ){ + sqlite3WhereEnd(pWInfo); + } + } + + /* Clean up the WHERE clause constructed above. */ + sqlite3ExprDelete(db, pWhere); + if( iFkIfZero ){ + sqlite3VdbeJumpHere(v, iFkIfZero); + } +} + +/* +** This function returns a linked list of FKey objects (connected by +** FKey.pNextTo) holding all children of table pTab. For example, +** given the following schema: +** +** CREATE TABLE t1(a PRIMARY KEY); +** CREATE TABLE t2(b REFERENCES t1(a); +** +** Calling this function with table "t1" as an argument returns a pointer +** to the FKey structure representing the foreign key constraint on table +** "t2". Calling this function with "t2" as the argument would return a +** NULL pointer (as there are no FK constraints for which t2 is the parent +** table). +*/ +SQLITE_PRIVATE FKey *sqlite3FkReferences(Table *pTab){ + return (FKey *)sqlite3HashFind(&pTab->pSchema->fkeyHash, pTab->zName); +} + +/* +** The second argument is a Trigger structure allocated by the +** fkActionTrigger() routine. This function deletes the Trigger structure +** and all of its sub-components. +** +** The Trigger structure or any of its sub-components may be allocated from +** the lookaside buffer belonging to database handle dbMem. +*/ +static void fkTriggerDelete(sqlite3 *dbMem, Trigger *p){ + if( p ){ + TriggerStep *pStep = p->step_list; + sqlite3ExprDelete(dbMem, pStep->pWhere); + sqlite3ExprListDelete(dbMem, pStep->pExprList); + sqlite3SelectDelete(dbMem, pStep->pSelect); + sqlite3ExprDelete(dbMem, p->pWhen); + sqlite3DbFree(dbMem, p); + } +} + +/* +** This function is called to generate code that runs when table pTab is +** being dropped from the database. The SrcList passed as the second argument +** to this function contains a single entry guaranteed to resolve to +** table pTab. +** +** Normally, no code is required. However, if either +** +** (a) The table is the parent table of a FK constraint, or +** (b) The table is the child table of a deferred FK constraint and it is +** determined at runtime that there are outstanding deferred FK +** constraint violations in the database, +** +** then the equivalent of "DELETE FROM " is executed before dropping +** the table from the database. Triggers are disabled while running this +** DELETE, but foreign key actions are not. +*/ +SQLITE_PRIVATE void sqlite3FkDropTable(Parse *pParse, SrcList *pName, Table *pTab){ + sqlite3 *db = pParse->db; + if( (db->flags&SQLITE_ForeignKeys) && !IsVirtual(pTab) ){ + int iSkip = 0; + Vdbe *v = sqlite3GetVdbe(pParse); + + assert( v ); /* VDBE has already been allocated */ + assert( pTab->pSelect==0 ); /* Not a view */ + if( sqlite3FkReferences(pTab)==0 ){ + /* Search for a deferred foreign key constraint for which this table + ** is the child table. If one cannot be found, return without + ** generating any VDBE code. If one can be found, then jump over + ** the entire DELETE if there are no outstanding deferred constraints + ** when this statement is run. */ + FKey *p; + for(p=pTab->pFKey; p; p=p->pNextFrom){ + if( p->isDeferred || (db->flags & SQLITE_DeferFKs) ) break; + } + if( !p ) return; + iSkip = sqlite3VdbeMakeLabel(pParse); + sqlite3VdbeAddOp2(v, OP_FkIfZero, 1, iSkip); VdbeCoverage(v); + } + + pParse->disableTriggers = 1; + sqlite3DeleteFrom(pParse, sqlite3SrcListDup(db, pName, 0), 0, 0, 0); + pParse->disableTriggers = 0; + + /* If the DELETE has generated immediate foreign key constraint + ** violations, halt the VDBE and return an error at this point, before + ** any modifications to the schema are made. This is because statement + ** transactions are not able to rollback schema changes. + ** + ** If the SQLITE_DeferFKs flag is set, then this is not required, as + ** the statement transaction will not be rolled back even if FK + ** constraints are violated. + */ + if( (db->flags & SQLITE_DeferFKs)==0 ){ + sqlite3VdbeVerifyAbortable(v, OE_Abort); + sqlite3VdbeAddOp2(v, OP_FkIfZero, 0, sqlite3VdbeCurrentAddr(v)+2); + VdbeCoverage(v); + sqlite3HaltConstraint(pParse, SQLITE_CONSTRAINT_FOREIGNKEY, + OE_Abort, 0, P4_STATIC, P5_ConstraintFK); + } + + if( iSkip ){ + sqlite3VdbeResolveLabel(v, iSkip); + } + } +} + + +/* +** The second argument points to an FKey object representing a foreign key +** for which pTab is the child table. An UPDATE statement against pTab +** is currently being processed. For each column of the table that is +** actually updated, the corresponding element in the aChange[] array +** is zero or greater (if a column is unmodified the corresponding element +** is set to -1). If the rowid column is modified by the UPDATE statement +** the bChngRowid argument is non-zero. +** +** This function returns true if any of the columns that are part of the +** child key for FK constraint *p are modified. +*/ +static int fkChildIsModified( + Table *pTab, /* Table being updated */ + FKey *p, /* Foreign key for which pTab is the child */ + int *aChange, /* Array indicating modified columns */ + int bChngRowid /* True if rowid is modified by this update */ +){ + int i; + for(i=0; inCol; i++){ + int iChildKey = p->aCol[i].iFrom; + if( aChange[iChildKey]>=0 ) return 1; + if( iChildKey==pTab->iPKey && bChngRowid ) return 1; + } + return 0; +} + +/* +** The second argument points to an FKey object representing a foreign key +** for which pTab is the parent table. An UPDATE statement against pTab +** is currently being processed. For each column of the table that is +** actually updated, the corresponding element in the aChange[] array +** is zero or greater (if a column is unmodified the corresponding element +** is set to -1). If the rowid column is modified by the UPDATE statement +** the bChngRowid argument is non-zero. +** +** This function returns true if any of the columns that are part of the +** parent key for FK constraint *p are modified. +*/ +static int fkParentIsModified( + Table *pTab, + FKey *p, + int *aChange, + int bChngRowid +){ + int i; + for(i=0; inCol; i++){ + char *zKey = p->aCol[i].zCol; + int iKey; + for(iKey=0; iKeynCol; iKey++){ + if( aChange[iKey]>=0 || (iKey==pTab->iPKey && bChngRowid) ){ + Column *pCol = &pTab->aCol[iKey]; + if( zKey ){ + if( 0==sqlite3StrICmp(pCol->zName, zKey) ) return 1; + }else if( pCol->colFlags & COLFLAG_PRIMKEY ){ + return 1; + } + } + } + } + return 0; +} + +/* +** Return true if the parser passed as the first argument is being +** used to code a trigger that is really a "SET NULL" action belonging +** to trigger pFKey. +*/ +static int isSetNullAction(Parse *pParse, FKey *pFKey){ + Parse *pTop = sqlite3ParseToplevel(pParse); + if( pTop->pTriggerPrg ){ + Trigger *p = pTop->pTriggerPrg->pTrigger; + if( (p==pFKey->apTrigger[0] && pFKey->aAction[0]==OE_SetNull) + || (p==pFKey->apTrigger[1] && pFKey->aAction[1]==OE_SetNull) + ){ + return 1; + } + } + return 0; +} + +/* +** This function is called when inserting, deleting or updating a row of +** table pTab to generate VDBE code to perform foreign key constraint +** processing for the operation. +** +** For a DELETE operation, parameter regOld is passed the index of the +** first register in an array of (pTab->nCol+1) registers containing the +** rowid of the row being deleted, followed by each of the column values +** of the row being deleted, from left to right. Parameter regNew is passed +** zero in this case. +** +** For an INSERT operation, regOld is passed zero and regNew is passed the +** first register of an array of (pTab->nCol+1) registers containing the new +** row data. +** +** For an UPDATE operation, this function is called twice. Once before +** the original record is deleted from the table using the calling convention +** described for DELETE. Then again after the original record is deleted +** but before the new record is inserted using the INSERT convention. +*/ +SQLITE_PRIVATE void sqlite3FkCheck( + Parse *pParse, /* Parse context */ + Table *pTab, /* Row is being deleted from this table */ + int regOld, /* Previous row data is stored here */ + int regNew, /* New row data is stored here */ + int *aChange, /* Array indicating UPDATEd columns (or 0) */ + int bChngRowid /* True if rowid is UPDATEd */ +){ + sqlite3 *db = pParse->db; /* Database handle */ + FKey *pFKey; /* Used to iterate through FKs */ + int iDb; /* Index of database containing pTab */ + const char *zDb; /* Name of database containing pTab */ + int isIgnoreErrors = pParse->disableTriggers; + + /* Exactly one of regOld and regNew should be non-zero. */ + assert( (regOld==0)!=(regNew==0) ); + + /* If foreign-keys are disabled, this function is a no-op. */ + if( (db->flags&SQLITE_ForeignKeys)==0 ) return; + + iDb = sqlite3SchemaToIndex(db, pTab->pSchema); + zDb = db->aDb[iDb].zDbSName; + + /* Loop through all the foreign key constraints for which pTab is the + ** child table (the table that the foreign key definition is part of). */ + for(pFKey=pTab->pFKey; pFKey; pFKey=pFKey->pNextFrom){ + Table *pTo; /* Parent table of foreign key pFKey */ + Index *pIdx = 0; /* Index on key columns in pTo */ + int *aiFree = 0; + int *aiCol; + int iCol; + int i; + int bIgnore = 0; + + if( aChange + && sqlite3_stricmp(pTab->zName, pFKey->zTo)!=0 + && fkChildIsModified(pTab, pFKey, aChange, bChngRowid)==0 + ){ + continue; + } + + /* Find the parent table of this foreign key. Also find a unique index + ** on the parent key columns in the parent table. If either of these + ** schema items cannot be located, set an error in pParse and return + ** early. */ + if( pParse->disableTriggers ){ + pTo = sqlite3FindTable(db, pFKey->zTo, zDb); + }else{ + pTo = sqlite3LocateTable(pParse, 0, pFKey->zTo, zDb); + } + if( !pTo || sqlite3FkLocateIndex(pParse, pTo, pFKey, &pIdx, &aiFree) ){ + assert( isIgnoreErrors==0 || (regOld!=0 && regNew==0) ); + if( !isIgnoreErrors || db->mallocFailed ) return; + if( pTo==0 ){ + /* If isIgnoreErrors is true, then a table is being dropped. In this + ** case SQLite runs a "DELETE FROM xxx" on the table being dropped + ** before actually dropping it in order to check FK constraints. + ** If the parent table of an FK constraint on the current table is + ** missing, behave as if it is empty. i.e. decrement the relevant + ** FK counter for each row of the current table with non-NULL keys. + */ + Vdbe *v = sqlite3GetVdbe(pParse); + int iJump = sqlite3VdbeCurrentAddr(v) + pFKey->nCol + 1; + for(i=0; inCol; i++){ + int iReg = pFKey->aCol[i].iFrom + regOld + 1; + sqlite3VdbeAddOp2(v, OP_IsNull, iReg, iJump); VdbeCoverage(v); + } + sqlite3VdbeAddOp2(v, OP_FkCounter, pFKey->isDeferred, -1); + } + continue; + } + assert( pFKey->nCol==1 || (aiFree && pIdx) ); + + if( aiFree ){ + aiCol = aiFree; + }else{ + iCol = pFKey->aCol[0].iFrom; + aiCol = &iCol; + } + for(i=0; inCol; i++){ + if( aiCol[i]==pTab->iPKey ){ + aiCol[i] = -1; + } + assert( pIdx==0 || pIdx->aiColumn[i]>=0 ); +#ifndef SQLITE_OMIT_AUTHORIZATION + /* Request permission to read the parent key columns. If the + ** authorization callback returns SQLITE_IGNORE, behave as if any + ** values read from the parent table are NULL. */ + if( db->xAuth ){ + int rcauth; + char *zCol = pTo->aCol[pIdx ? pIdx->aiColumn[i] : pTo->iPKey].zName; + rcauth = sqlite3AuthReadCol(pParse, pTo->zName, zCol, iDb); + bIgnore = (rcauth==SQLITE_IGNORE); + } +#endif + } + + /* Take a shared-cache advisory read-lock on the parent table. Allocate + ** a cursor to use to search the unique index on the parent key columns + ** in the parent table. */ + sqlite3TableLock(pParse, iDb, pTo->tnum, 0, pTo->zName); + pParse->nTab++; + + if( regOld!=0 ){ + /* A row is being removed from the child table. Search for the parent. + ** If the parent does not exist, removing the child row resolves an + ** outstanding foreign key constraint violation. */ + fkLookupParent(pParse, iDb, pTo, pIdx, pFKey, aiCol, regOld, -1, bIgnore); + } + if( regNew!=0 && !isSetNullAction(pParse, pFKey) ){ + /* A row is being added to the child table. If a parent row cannot + ** be found, adding the child row has violated the FK constraint. + ** + ** If this operation is being performed as part of a trigger program + ** that is actually a "SET NULL" action belonging to this very + ** foreign key, then omit this scan altogether. As all child key + ** values are guaranteed to be NULL, it is not possible for adding + ** this row to cause an FK violation. */ + fkLookupParent(pParse, iDb, pTo, pIdx, pFKey, aiCol, regNew, +1, bIgnore); + } + + sqlite3DbFree(db, aiFree); + } + + /* Loop through all the foreign key constraints that refer to this table. + ** (the "child" constraints) */ + for(pFKey = sqlite3FkReferences(pTab); pFKey; pFKey=pFKey->pNextTo){ + Index *pIdx = 0; /* Foreign key index for pFKey */ + SrcList *pSrc; + int *aiCol = 0; + + if( aChange && fkParentIsModified(pTab, pFKey, aChange, bChngRowid)==0 ){ + continue; + } + + if( !pFKey->isDeferred && !(db->flags & SQLITE_DeferFKs) + && !pParse->pToplevel && !pParse->isMultiWrite + ){ + assert( regOld==0 && regNew!=0 ); + /* Inserting a single row into a parent table cannot cause (or fix) + ** an immediate foreign key violation. So do nothing in this case. */ + continue; + } + + if( sqlite3FkLocateIndex(pParse, pTab, pFKey, &pIdx, &aiCol) ){ + if( !isIgnoreErrors || db->mallocFailed ) return; + continue; + } + assert( aiCol || pFKey->nCol==1 ); + + /* Create a SrcList structure containing the child table. We need the + ** child table as a SrcList for sqlite3WhereBegin() */ + pSrc = sqlite3SrcListAppend(pParse, 0, 0, 0); + if( pSrc ){ + struct SrcList_item *pItem = pSrc->a; + pItem->pTab = pFKey->pFrom; + pItem->zName = pFKey->pFrom->zName; + pItem->pTab->nTabRef++; + pItem->iCursor = pParse->nTab++; + + if( regNew!=0 ){ + fkScanChildren(pParse, pSrc, pTab, pIdx, pFKey, aiCol, regNew, -1); + } + if( regOld!=0 ){ + int eAction = pFKey->aAction[aChange!=0]; + fkScanChildren(pParse, pSrc, pTab, pIdx, pFKey, aiCol, regOld, 1); + /* If this is a deferred FK constraint, or a CASCADE or SET NULL + ** action applies, then any foreign key violations caused by + ** removing the parent key will be rectified by the action trigger. + ** So do not set the "may-abort" flag in this case. + ** + ** Note 1: If the FK is declared "ON UPDATE CASCADE", then the + ** may-abort flag will eventually be set on this statement anyway + ** (when this function is called as part of processing the UPDATE + ** within the action trigger). + ** + ** Note 2: At first glance it may seem like SQLite could simply omit + ** all OP_FkCounter related scans when either CASCADE or SET NULL + ** applies. The trouble starts if the CASCADE or SET NULL action + ** trigger causes other triggers or action rules attached to the + ** child table to fire. In these cases the fk constraint counters + ** might be set incorrectly if any OP_FkCounter related scans are + ** omitted. */ + if( !pFKey->isDeferred && eAction!=OE_Cascade && eAction!=OE_SetNull ){ + sqlite3MayAbort(pParse); + } + } + pItem->zName = 0; + sqlite3SrcListDelete(db, pSrc); + } + sqlite3DbFree(db, aiCol); + } +} + +#define COLUMN_MASK(x) (((x)>31) ? 0xffffffff : ((u32)1<<(x))) + +/* +** This function is called before generating code to update or delete a +** row contained in table pTab. +*/ +SQLITE_PRIVATE u32 sqlite3FkOldmask( + Parse *pParse, /* Parse context */ + Table *pTab /* Table being modified */ +){ + u32 mask = 0; + if( pParse->db->flags&SQLITE_ForeignKeys ){ + FKey *p; + int i; + for(p=pTab->pFKey; p; p=p->pNextFrom){ + for(i=0; inCol; i++) mask |= COLUMN_MASK(p->aCol[i].iFrom); + } + for(p=sqlite3FkReferences(pTab); p; p=p->pNextTo){ + Index *pIdx = 0; + sqlite3FkLocateIndex(pParse, pTab, p, &pIdx, 0); + if( pIdx ){ + for(i=0; inKeyCol; i++){ + assert( pIdx->aiColumn[i]>=0 ); + mask |= COLUMN_MASK(pIdx->aiColumn[i]); + } + } + } + } + return mask; +} + + +/* +** This function is called before generating code to update or delete a +** row contained in table pTab. If the operation is a DELETE, then +** parameter aChange is passed a NULL value. For an UPDATE, aChange points +** to an array of size N, where N is the number of columns in table pTab. +** If the i'th column is not modified by the UPDATE, then the corresponding +** entry in the aChange[] array is set to -1. If the column is modified, +** the value is 0 or greater. Parameter chngRowid is set to true if the +** UPDATE statement modifies the rowid fields of the table. +** +** If any foreign key processing will be required, this function returns +** non-zero. If there is no foreign key related processing, this function +** returns zero. +** +** For an UPDATE, this function returns 2 if: +** +** * There are any FKs for which pTab is the child and the parent table, or +** * the UPDATE modifies one or more parent keys for which the action is +** not "NO ACTION" (i.e. is CASCADE, SET DEFAULT or SET NULL). +** +** Or, assuming some other foreign key processing is required, 1. +*/ +SQLITE_PRIVATE int sqlite3FkRequired( + Parse *pParse, /* Parse context */ + Table *pTab, /* Table being modified */ + int *aChange, /* Non-NULL for UPDATE operations */ + int chngRowid /* True for UPDATE that affects rowid */ +){ + int eRet = 0; + if( pParse->db->flags&SQLITE_ForeignKeys ){ + if( !aChange ){ + /* A DELETE operation. Foreign key processing is required if the + ** table in question is either the child or parent table for any + ** foreign key constraint. */ + eRet = (sqlite3FkReferences(pTab) || pTab->pFKey); + }else{ + /* This is an UPDATE. Foreign key processing is only required if the + ** operation modifies one or more child or parent key columns. */ + FKey *p; + + /* Check if any child key columns are being modified. */ + for(p=pTab->pFKey; p; p=p->pNextFrom){ + if( 0==sqlite3_stricmp(pTab->zName, p->zTo) ) return 2; + if( fkChildIsModified(pTab, p, aChange, chngRowid) ){ + eRet = 1; + } + } + + /* Check if any parent key columns are being modified. */ + for(p=sqlite3FkReferences(pTab); p; p=p->pNextTo){ + if( fkParentIsModified(pTab, p, aChange, chngRowid) ){ + if( p->aAction[1]!=OE_None ) return 2; + eRet = 1; + } + } + } + } + return eRet; +} + +/* +** This function is called when an UPDATE or DELETE operation is being +** compiled on table pTab, which is the parent table of foreign-key pFKey. +** If the current operation is an UPDATE, then the pChanges parameter is +** passed a pointer to the list of columns being modified. If it is a +** DELETE, pChanges is passed a NULL pointer. +** +** It returns a pointer to a Trigger structure containing a trigger +** equivalent to the ON UPDATE or ON DELETE action specified by pFKey. +** If the action is "NO ACTION" or "RESTRICT", then a NULL pointer is +** returned (these actions require no special handling by the triggers +** sub-system, code for them is created by fkScanChildren()). +** +** For example, if pFKey is the foreign key and pTab is table "p" in +** the following schema: +** +** CREATE TABLE p(pk PRIMARY KEY); +** CREATE TABLE c(ck REFERENCES p ON DELETE CASCADE); +** +** then the returned trigger structure is equivalent to: +** +** CREATE TRIGGER ... DELETE ON p BEGIN +** DELETE FROM c WHERE ck = old.pk; +** END; +** +** The returned pointer is cached as part of the foreign key object. It +** is eventually freed along with the rest of the foreign key object by +** sqlite3FkDelete(). +*/ +static Trigger *fkActionTrigger( + Parse *pParse, /* Parse context */ + Table *pTab, /* Table being updated or deleted from */ + FKey *pFKey, /* Foreign key to get action for */ + ExprList *pChanges /* Change-list for UPDATE, NULL for DELETE */ +){ + sqlite3 *db = pParse->db; /* Database handle */ + int action; /* One of OE_None, OE_Cascade etc. */ + Trigger *pTrigger; /* Trigger definition to return */ + int iAction = (pChanges!=0); /* 1 for UPDATE, 0 for DELETE */ + + action = pFKey->aAction[iAction]; + if( action==OE_Restrict && (db->flags & SQLITE_DeferFKs) ){ + return 0; + } + pTrigger = pFKey->apTrigger[iAction]; + + if( action!=OE_None && !pTrigger ){ + char const *zFrom; /* Name of child table */ + int nFrom; /* Length in bytes of zFrom */ + Index *pIdx = 0; /* Parent key index for this FK */ + int *aiCol = 0; /* child table cols -> parent key cols */ + TriggerStep *pStep = 0; /* First (only) step of trigger program */ + Expr *pWhere = 0; /* WHERE clause of trigger step */ + ExprList *pList = 0; /* Changes list if ON UPDATE CASCADE */ + Select *pSelect = 0; /* If RESTRICT, "SELECT RAISE(...)" */ + int i; /* Iterator variable */ + Expr *pWhen = 0; /* WHEN clause for the trigger */ + + if( sqlite3FkLocateIndex(pParse, pTab, pFKey, &pIdx, &aiCol) ) return 0; + assert( aiCol || pFKey->nCol==1 ); + + for(i=0; inCol; i++){ + Token tOld = { "old", 3 }; /* Literal "old" token */ + Token tNew = { "new", 3 }; /* Literal "new" token */ + Token tFromCol; /* Name of column in child table */ + Token tToCol; /* Name of column in parent table */ + int iFromCol; /* Idx of column in child table */ + Expr *pEq; /* tFromCol = OLD.tToCol */ + + iFromCol = aiCol ? aiCol[i] : pFKey->aCol[0].iFrom; + assert( iFromCol>=0 ); + assert( pIdx!=0 || (pTab->iPKey>=0 && pTab->iPKeynCol) ); + assert( pIdx==0 || pIdx->aiColumn[i]>=0 ); + sqlite3TokenInit(&tToCol, + pTab->aCol[pIdx ? pIdx->aiColumn[i] : pTab->iPKey].zName); + sqlite3TokenInit(&tFromCol, pFKey->pFrom->aCol[iFromCol].zName); + + /* Create the expression "OLD.zToCol = zFromCol". It is important + ** that the "OLD.zToCol" term is on the LHS of the = operator, so + ** that the affinity and collation sequence associated with the + ** parent table are used for the comparison. */ + pEq = sqlite3PExpr(pParse, TK_EQ, + sqlite3PExpr(pParse, TK_DOT, + sqlite3ExprAlloc(db, TK_ID, &tOld, 0), + sqlite3ExprAlloc(db, TK_ID, &tToCol, 0)), + sqlite3ExprAlloc(db, TK_ID, &tFromCol, 0) + ); + pWhere = sqlite3ExprAnd(pParse, pWhere, pEq); + + /* For ON UPDATE, construct the next term of the WHEN clause. + ** The final WHEN clause will be like this: + ** + ** WHEN NOT(old.col1 IS new.col1 AND ... AND old.colN IS new.colN) + */ + if( pChanges ){ + pEq = sqlite3PExpr(pParse, TK_IS, + sqlite3PExpr(pParse, TK_DOT, + sqlite3ExprAlloc(db, TK_ID, &tOld, 0), + sqlite3ExprAlloc(db, TK_ID, &tToCol, 0)), + sqlite3PExpr(pParse, TK_DOT, + sqlite3ExprAlloc(db, TK_ID, &tNew, 0), + sqlite3ExprAlloc(db, TK_ID, &tToCol, 0)) + ); + pWhen = sqlite3ExprAnd(pParse, pWhen, pEq); + } + + if( action!=OE_Restrict && (action!=OE_Cascade || pChanges) ){ + Expr *pNew; + if( action==OE_Cascade ){ + pNew = sqlite3PExpr(pParse, TK_DOT, + sqlite3ExprAlloc(db, TK_ID, &tNew, 0), + sqlite3ExprAlloc(db, TK_ID, &tToCol, 0)); + }else if( action==OE_SetDflt ){ + Expr *pDflt = pFKey->pFrom->aCol[iFromCol].pDflt; + if( pDflt ){ + pNew = sqlite3ExprDup(db, pDflt, 0); + }else{ + pNew = sqlite3ExprAlloc(db, TK_NULL, 0, 0); + } + }else{ + pNew = sqlite3ExprAlloc(db, TK_NULL, 0, 0); + } + pList = sqlite3ExprListAppend(pParse, pList, pNew); + sqlite3ExprListSetName(pParse, pList, &tFromCol, 0); + } + } + sqlite3DbFree(db, aiCol); + + zFrom = pFKey->pFrom->zName; + nFrom = sqlite3Strlen30(zFrom); + + if( action==OE_Restrict ){ + Token tFrom; + Expr *pRaise; + + tFrom.z = zFrom; + tFrom.n = nFrom; + pRaise = sqlite3Expr(db, TK_RAISE, "FOREIGN KEY constraint failed"); + if( pRaise ){ + pRaise->affExpr = OE_Abort; + } + pSelect = sqlite3SelectNew(pParse, + sqlite3ExprListAppend(pParse, 0, pRaise), + sqlite3SrcListAppend(pParse, 0, &tFrom, 0), + pWhere, + 0, 0, 0, 0, 0 + ); + pWhere = 0; + } + + /* Disable lookaside memory allocation */ + db->lookaside.bDisable++; + + pTrigger = (Trigger *)sqlite3DbMallocZero(db, + sizeof(Trigger) + /* struct Trigger */ + sizeof(TriggerStep) + /* Single step in trigger program */ + nFrom + 1 /* Space for pStep->zTarget */ + ); + if( pTrigger ){ + pStep = pTrigger->step_list = (TriggerStep *)&pTrigger[1]; + pStep->zTarget = (char *)&pStep[1]; + memcpy((char *)pStep->zTarget, zFrom, nFrom); + + pStep->pWhere = sqlite3ExprDup(db, pWhere, EXPRDUP_REDUCE); + pStep->pExprList = sqlite3ExprListDup(db, pList, EXPRDUP_REDUCE); + pStep->pSelect = sqlite3SelectDup(db, pSelect, EXPRDUP_REDUCE); + if( pWhen ){ + pWhen = sqlite3PExpr(pParse, TK_NOT, pWhen, 0); + pTrigger->pWhen = sqlite3ExprDup(db, pWhen, EXPRDUP_REDUCE); + } + } + + /* Re-enable the lookaside buffer, if it was disabled earlier. */ + db->lookaside.bDisable--; + + sqlite3ExprDelete(db, pWhere); + sqlite3ExprDelete(db, pWhen); + sqlite3ExprListDelete(db, pList); + sqlite3SelectDelete(db, pSelect); + if( db->mallocFailed==1 ){ + fkTriggerDelete(db, pTrigger); + return 0; + } + assert( pStep!=0 ); + assert( pTrigger!=0 ); + + switch( action ){ + case OE_Restrict: + pStep->op = TK_SELECT; + break; + case OE_Cascade: + if( !pChanges ){ + pStep->op = TK_DELETE; + break; + } + default: + pStep->op = TK_UPDATE; + } + pStep->pTrig = pTrigger; + pTrigger->pSchema = pTab->pSchema; + pTrigger->pTabSchema = pTab->pSchema; + pFKey->apTrigger[iAction] = pTrigger; + pTrigger->op = (pChanges ? TK_UPDATE : TK_DELETE); + } + + return pTrigger; +} + +/* +** This function is called when deleting or updating a row to implement +** any required CASCADE, SET NULL or SET DEFAULT actions. +*/ +SQLITE_PRIVATE void sqlite3FkActions( + Parse *pParse, /* Parse context */ + Table *pTab, /* Table being updated or deleted from */ + ExprList *pChanges, /* Change-list for UPDATE, NULL for DELETE */ + int regOld, /* Address of array containing old row */ + int *aChange, /* Array indicating UPDATEd columns (or 0) */ + int bChngRowid /* True if rowid is UPDATEd */ +){ + /* If foreign-key support is enabled, iterate through all FKs that + ** refer to table pTab. If there is an action associated with the FK + ** for this operation (either update or delete), invoke the associated + ** trigger sub-program. */ + if( pParse->db->flags&SQLITE_ForeignKeys ){ + FKey *pFKey; /* Iterator variable */ + for(pFKey = sqlite3FkReferences(pTab); pFKey; pFKey=pFKey->pNextTo){ + if( aChange==0 || fkParentIsModified(pTab, pFKey, aChange, bChngRowid) ){ + Trigger *pAct = fkActionTrigger(pParse, pTab, pFKey, pChanges); + if( pAct ){ + sqlite3CodeRowTriggerDirect(pParse, pAct, pTab, regOld, OE_Abort, 0); + } + } + } + } +} + +#endif /* ifndef SQLITE_OMIT_TRIGGER */ + +/* +** Free all memory associated with foreign key definitions attached to +** table pTab. Remove the deleted foreign keys from the Schema.fkeyHash +** hash table. +*/ +SQLITE_PRIVATE void sqlite3FkDelete(sqlite3 *db, Table *pTab){ + FKey *pFKey; /* Iterator variable */ + FKey *pNext; /* Copy of pFKey->pNextFrom */ + + assert( db==0 || IsVirtual(pTab) + || sqlite3SchemaMutexHeld(db, 0, pTab->pSchema) ); + for(pFKey=pTab->pFKey; pFKey; pFKey=pNext){ + + /* Remove the FK from the fkeyHash hash table. */ + if( !db || db->pnBytesFreed==0 ){ + if( pFKey->pPrevTo ){ + pFKey->pPrevTo->pNextTo = pFKey->pNextTo; + }else{ + void *p = (void *)pFKey->pNextTo; + const char *z = (p ? pFKey->pNextTo->zTo : pFKey->zTo); + sqlite3HashInsert(&pTab->pSchema->fkeyHash, z, p); + } + if( pFKey->pNextTo ){ + pFKey->pNextTo->pPrevTo = pFKey->pPrevTo; + } + } + + /* EV: R-30323-21917 Each foreign key constraint in SQLite is + ** classified as either immediate or deferred. + */ + assert( pFKey->isDeferred==0 || pFKey->isDeferred==1 ); + + /* Delete any triggers created to implement actions for this FK. */ +#ifndef SQLITE_OMIT_TRIGGER + fkTriggerDelete(db, pFKey->apTrigger[0]); + fkTriggerDelete(db, pFKey->apTrigger[1]); +#endif + + pNext = pFKey->pNextFrom; + sqlite3DbFree(db, pFKey); + } +} +#endif /* ifndef SQLITE_OMIT_FOREIGN_KEY */ + +/************** End of fkey.c ************************************************/ +/************** Begin file insert.c ******************************************/ +/* +** 2001 September 15 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** This file contains C code routines that are called by the parser +** to handle INSERT statements in SQLite. +*/ +/* #include "sqliteInt.h" */ + +/* +** Generate code that will +** +** (1) acquire a lock for table pTab then +** (2) open pTab as cursor iCur. +** +** If pTab is a WITHOUT ROWID table, then it is the PRIMARY KEY index +** for that table that is actually opened. +*/ +SQLITE_PRIVATE void sqlite3OpenTable( + Parse *pParse, /* Generate code into this VDBE */ + int iCur, /* The cursor number of the table */ + int iDb, /* The database index in sqlite3.aDb[] */ + Table *pTab, /* The table to be opened */ + int opcode /* OP_OpenRead or OP_OpenWrite */ +){ + Vdbe *v; + assert( !IsVirtual(pTab) ); + v = sqlite3GetVdbe(pParse); + assert( opcode==OP_OpenWrite || opcode==OP_OpenRead ); + sqlite3TableLock(pParse, iDb, pTab->tnum, + (opcode==OP_OpenWrite)?1:0, pTab->zName); + if( HasRowid(pTab) ){ + sqlite3VdbeAddOp4Int(v, opcode, iCur, pTab->tnum, iDb, pTab->nCol); + VdbeComment((v, "%s", pTab->zName)); + }else{ + Index *pPk = sqlite3PrimaryKeyIndex(pTab); + assert( pPk!=0 ); + assert( pPk->tnum==pTab->tnum ); + sqlite3VdbeAddOp3(v, opcode, iCur, pPk->tnum, iDb); + sqlite3VdbeSetP4KeyInfo(pParse, pPk); + VdbeComment((v, "%s", pTab->zName)); + } +} + +/* +** Return a pointer to the column affinity string associated with index +** pIdx. A column affinity string has one character for each column in +** the table, according to the affinity of the column: +** +** Character Column affinity +** ------------------------------ +** 'A' BLOB +** 'B' TEXT +** 'C' NUMERIC +** 'D' INTEGER +** 'F' REAL +** +** An extra 'D' is appended to the end of the string to cover the +** rowid that appears as the last column in every index. +** +** Memory for the buffer containing the column index affinity string +** is managed along with the rest of the Index structure. It will be +** released when sqlite3DeleteIndex() is called. +*/ +SQLITE_PRIVATE const char *sqlite3IndexAffinityStr(sqlite3 *db, Index *pIdx){ + if( !pIdx->zColAff ){ + /* The first time a column affinity string for a particular index is + ** required, it is allocated and populated here. It is then stored as + ** a member of the Index structure for subsequent use. + ** + ** The column affinity string will eventually be deleted by + ** sqliteDeleteIndex() when the Index structure itself is cleaned + ** up. + */ + int n; + Table *pTab = pIdx->pTable; + pIdx->zColAff = (char *)sqlite3DbMallocRaw(0, pIdx->nColumn+1); + if( !pIdx->zColAff ){ + sqlite3OomFault(db); + return 0; + } + for(n=0; nnColumn; n++){ + i16 x = pIdx->aiColumn[n]; + char aff; + if( x>=0 ){ + aff = pTab->aCol[x].affinity; + }else if( x==XN_ROWID ){ + aff = SQLITE_AFF_INTEGER; + }else{ + assert( x==XN_EXPR ); + assert( pIdx->aColExpr!=0 ); + aff = sqlite3ExprAffinity(pIdx->aColExpr->a[n].pExpr); + } + if( affSQLITE_AFF_NUMERIC) aff = SQLITE_AFF_NUMERIC; + pIdx->zColAff[n] = aff; + } + pIdx->zColAff[n] = 0; + } + + return pIdx->zColAff; +} + +/* +** Compute the affinity string for table pTab, if it has not already been +** computed. As an optimization, omit trailing SQLITE_AFF_BLOB affinities. +** +** If the affinity exists (if it is no entirely SQLITE_AFF_BLOB values) and +** if iReg>0 then code an OP_Affinity opcode that will set the affinities +** for register iReg and following. Or if affinities exists and iReg==0, +** then just set the P4 operand of the previous opcode (which should be +** an OP_MakeRecord) to the affinity string. +** +** A column affinity string has one character per column: +** +** Character Column affinity +** ------------------------------ +** 'A' BLOB +** 'B' TEXT +** 'C' NUMERIC +** 'D' INTEGER +** 'E' REAL +*/ +SQLITE_PRIVATE void sqlite3TableAffinity(Vdbe *v, Table *pTab, int iReg){ + int i; + char *zColAff = pTab->zColAff; + if( zColAff==0 ){ + sqlite3 *db = sqlite3VdbeDb(v); + zColAff = (char *)sqlite3DbMallocRaw(0, pTab->nCol+1); + if( !zColAff ){ + sqlite3OomFault(db); + return; + } + + for(i=0; inCol; i++){ + assert( pTab->aCol[i].affinity!=0 ); + zColAff[i] = pTab->aCol[i].affinity; + } + do{ + zColAff[i--] = 0; + }while( i>=0 && zColAff[i]<=SQLITE_AFF_BLOB ); + pTab->zColAff = zColAff; + } + assert( zColAff!=0 ); + i = sqlite3Strlen30NN(zColAff); + if( i ){ + if( iReg ){ + sqlite3VdbeAddOp4(v, OP_Affinity, iReg, i, 0, zColAff, i); + }else{ + sqlite3VdbeChangeP4(v, -1, zColAff, i); + } + } +} + +/* +** Return non-zero if the table pTab in database iDb or any of its indices +** have been opened at any point in the VDBE program. This is used to see if +** a statement of the form "INSERT INTO SELECT ..." can +** run without using a temporary table for the results of the SELECT. +*/ +static int readsTable(Parse *p, int iDb, Table *pTab){ + Vdbe *v = sqlite3GetVdbe(p); + int i; + int iEnd = sqlite3VdbeCurrentAddr(v); +#ifndef SQLITE_OMIT_VIRTUALTABLE + VTable *pVTab = IsVirtual(pTab) ? sqlite3GetVTable(p->db, pTab) : 0; +#endif + + for(i=1; iopcode==OP_OpenRead && pOp->p3==iDb ){ + Index *pIndex; + int tnum = pOp->p2; + if( tnum==pTab->tnum ){ + return 1; + } + for(pIndex=pTab->pIndex; pIndex; pIndex=pIndex->pNext){ + if( tnum==pIndex->tnum ){ + return 1; + } + } + } +#ifndef SQLITE_OMIT_VIRTUALTABLE + if( pOp->opcode==OP_VOpen && pOp->p4.pVtab==pVTab ){ + assert( pOp->p4.pVtab!=0 ); + assert( pOp->p4type==P4_VTAB ); + return 1; + } +#endif + } + return 0; +} + +#ifndef SQLITE_OMIT_AUTOINCREMENT +/* +** Locate or create an AutoincInfo structure associated with table pTab +** which is in database iDb. Return the register number for the register +** that holds the maximum rowid. Return zero if pTab is not an AUTOINCREMENT +** table. (Also return zero when doing a VACUUM since we do not want to +** update the AUTOINCREMENT counters during a VACUUM.) +** +** There is at most one AutoincInfo structure per table even if the +** same table is autoincremented multiple times due to inserts within +** triggers. A new AutoincInfo structure is created if this is the +** first use of table pTab. On 2nd and subsequent uses, the original +** AutoincInfo structure is used. +** +** Four consecutive registers are allocated: +** +** (1) The name of the pTab table. +** (2) The maximum ROWID of pTab. +** (3) The rowid in sqlite_sequence of pTab +** (4) The original value of the max ROWID in pTab, or NULL if none +** +** The 2nd register is the one that is returned. That is all the +** insert routine needs to know about. +*/ +static int autoIncBegin( + Parse *pParse, /* Parsing context */ + int iDb, /* Index of the database holding pTab */ + Table *pTab /* The table we are writing to */ +){ + int memId = 0; /* Register holding maximum rowid */ + assert( pParse->db->aDb[iDb].pSchema!=0 ); + if( (pTab->tabFlags & TF_Autoincrement)!=0 + && (pParse->db->mDbFlags & DBFLAG_Vacuum)==0 + ){ + Parse *pToplevel = sqlite3ParseToplevel(pParse); + AutoincInfo *pInfo; + Table *pSeqTab = pParse->db->aDb[iDb].pSchema->pSeqTab; + + /* Verify that the sqlite_sequence table exists and is an ordinary + ** rowid table with exactly two columns. + ** Ticket d8dc2b3a58cd5dc2918a1d4acb 2018-05-23 */ + if( pSeqTab==0 + || !HasRowid(pSeqTab) + || IsVirtual(pSeqTab) + || pSeqTab->nCol!=2 + ){ + pParse->nErr++; + pParse->rc = SQLITE_CORRUPT_SEQUENCE; + return 0; + } + + pInfo = pToplevel->pAinc; + while( pInfo && pInfo->pTab!=pTab ){ pInfo = pInfo->pNext; } + if( pInfo==0 ){ + pInfo = sqlite3DbMallocRawNN(pParse->db, sizeof(*pInfo)); + if( pInfo==0 ) return 0; + pInfo->pNext = pToplevel->pAinc; + pToplevel->pAinc = pInfo; + pInfo->pTab = pTab; + pInfo->iDb = iDb; + pToplevel->nMem++; /* Register to hold name of table */ + pInfo->regCtr = ++pToplevel->nMem; /* Max rowid register */ + pToplevel->nMem +=2; /* Rowid in sqlite_sequence + orig max val */ + } + memId = pInfo->regCtr; + } + return memId; +} + +/* +** This routine generates code that will initialize all of the +** register used by the autoincrement tracker. +*/ +SQLITE_PRIVATE void sqlite3AutoincrementBegin(Parse *pParse){ + AutoincInfo *p; /* Information about an AUTOINCREMENT */ + sqlite3 *db = pParse->db; /* The database connection */ + Db *pDb; /* Database only autoinc table */ + int memId; /* Register holding max rowid */ + Vdbe *v = pParse->pVdbe; /* VDBE under construction */ + + /* This routine is never called during trigger-generation. It is + ** only called from the top-level */ + assert( pParse->pTriggerTab==0 ); + assert( sqlite3IsToplevel(pParse) ); + + assert( v ); /* We failed long ago if this is not so */ + for(p = pParse->pAinc; p; p = p->pNext){ + static const int iLn = VDBE_OFFSET_LINENO(2); + static const VdbeOpList autoInc[] = { + /* 0 */ {OP_Null, 0, 0, 0}, + /* 1 */ {OP_Rewind, 0, 10, 0}, + /* 2 */ {OP_Column, 0, 0, 0}, + /* 3 */ {OP_Ne, 0, 9, 0}, + /* 4 */ {OP_Rowid, 0, 0, 0}, + /* 5 */ {OP_Column, 0, 1, 0}, + /* 6 */ {OP_AddImm, 0, 0, 0}, + /* 7 */ {OP_Copy, 0, 0, 0}, + /* 8 */ {OP_Goto, 0, 11, 0}, + /* 9 */ {OP_Next, 0, 2, 0}, + /* 10 */ {OP_Integer, 0, 0, 0}, + /* 11 */ {OP_Close, 0, 0, 0} + }; + VdbeOp *aOp; + pDb = &db->aDb[p->iDb]; + memId = p->regCtr; + assert( sqlite3SchemaMutexHeld(db, 0, pDb->pSchema) ); + sqlite3OpenTable(pParse, 0, p->iDb, pDb->pSchema->pSeqTab, OP_OpenRead); + sqlite3VdbeLoadString(v, memId-1, p->pTab->zName); + aOp = sqlite3VdbeAddOpList(v, ArraySize(autoInc), autoInc, iLn); + if( aOp==0 ) break; + aOp[0].p2 = memId; + aOp[0].p3 = memId+2; + aOp[2].p3 = memId; + aOp[3].p1 = memId-1; + aOp[3].p3 = memId; + aOp[3].p5 = SQLITE_JUMPIFNULL; + aOp[4].p2 = memId+1; + aOp[5].p3 = memId; + aOp[6].p1 = memId; + aOp[7].p2 = memId+2; + aOp[7].p1 = memId; + aOp[10].p2 = memId; + if( pParse->nTab==0 ) pParse->nTab = 1; + } +} + +/* +** Update the maximum rowid for an autoincrement calculation. +** +** This routine should be called when the regRowid register holds a +** new rowid that is about to be inserted. If that new rowid is +** larger than the maximum rowid in the memId memory cell, then the +** memory cell is updated. +*/ +static void autoIncStep(Parse *pParse, int memId, int regRowid){ + if( memId>0 ){ + sqlite3VdbeAddOp2(pParse->pVdbe, OP_MemMax, memId, regRowid); + } +} + +/* +** This routine generates the code needed to write autoincrement +** maximum rowid values back into the sqlite_sequence register. +** Every statement that might do an INSERT into an autoincrement +** table (either directly or through triggers) needs to call this +** routine just before the "exit" code. +*/ +static SQLITE_NOINLINE void autoIncrementEnd(Parse *pParse){ + AutoincInfo *p; + Vdbe *v = pParse->pVdbe; + sqlite3 *db = pParse->db; + + assert( v ); + for(p = pParse->pAinc; p; p = p->pNext){ + static const int iLn = VDBE_OFFSET_LINENO(2); + static const VdbeOpList autoIncEnd[] = { + /* 0 */ {OP_NotNull, 0, 2, 0}, + /* 1 */ {OP_NewRowid, 0, 0, 0}, + /* 2 */ {OP_MakeRecord, 0, 2, 0}, + /* 3 */ {OP_Insert, 0, 0, 0}, + /* 4 */ {OP_Close, 0, 0, 0} + }; + VdbeOp *aOp; + Db *pDb = &db->aDb[p->iDb]; + int iRec; + int memId = p->regCtr; + + iRec = sqlite3GetTempReg(pParse); + assert( sqlite3SchemaMutexHeld(db, 0, pDb->pSchema) ); + sqlite3VdbeAddOp3(v, OP_Le, memId+2, sqlite3VdbeCurrentAddr(v)+7, memId); + VdbeCoverage(v); + sqlite3OpenTable(pParse, 0, p->iDb, pDb->pSchema->pSeqTab, OP_OpenWrite); + aOp = sqlite3VdbeAddOpList(v, ArraySize(autoIncEnd), autoIncEnd, iLn); + if( aOp==0 ) break; + aOp[0].p1 = memId+1; + aOp[1].p2 = memId+1; + aOp[2].p1 = memId-1; + aOp[2].p3 = iRec; + aOp[3].p2 = iRec; + aOp[3].p3 = memId+1; + aOp[3].p5 = OPFLAG_APPEND; + sqlite3ReleaseTempReg(pParse, iRec); + } +} +SQLITE_PRIVATE void sqlite3AutoincrementEnd(Parse *pParse){ + if( pParse->pAinc ) autoIncrementEnd(pParse); +} +#else +/* +** If SQLITE_OMIT_AUTOINCREMENT is defined, then the three routines +** above are all no-ops +*/ +# define autoIncBegin(A,B,C) (0) +# define autoIncStep(A,B,C) +#endif /* SQLITE_OMIT_AUTOINCREMENT */ + + +/* Forward declaration */ +static int xferOptimization( + Parse *pParse, /* Parser context */ + Table *pDest, /* The table we are inserting into */ + Select *pSelect, /* A SELECT statement to use as the data source */ + int onError, /* How to handle constraint errors */ + int iDbDest /* The database of pDest */ +); + +/* +** This routine is called to handle SQL of the following forms: +** +** insert into TABLE (IDLIST) values(EXPRLIST),(EXPRLIST),... +** insert into TABLE (IDLIST) select +** insert into TABLE (IDLIST) default values +** +** The IDLIST following the table name is always optional. If omitted, +** then a list of all (non-hidden) columns for the table is substituted. +** The IDLIST appears in the pColumn parameter. pColumn is NULL if IDLIST +** is omitted. +** +** For the pSelect parameter holds the values to be inserted for the +** first two forms shown above. A VALUES clause is really just short-hand +** for a SELECT statement that omits the FROM clause and everything else +** that follows. If the pSelect parameter is NULL, that means that the +** DEFAULT VALUES form of the INSERT statement is intended. +** +** The code generated follows one of four templates. For a simple +** insert with data coming from a single-row VALUES clause, the code executes +** once straight down through. Pseudo-code follows (we call this +** the "1st template"): +** +** open write cursor to
    and its indices +** put VALUES clause expressions into registers +** write the resulting record into
    +** cleanup +** +** The three remaining templates assume the statement is of the form +** +** INSERT INTO
    SELECT ... +** +** If the SELECT clause is of the restricted form "SELECT * FROM " - +** in other words if the SELECT pulls all columns from a single table +** and there is no WHERE or LIMIT or GROUP BY or ORDER BY clauses, and +** if and are distinct tables but have identical +** schemas, including all the same indices, then a special optimization +** is invoked that copies raw records from over to . +** See the xferOptimization() function for the implementation of this +** template. This is the 2nd template. +** +** open a write cursor to
    +** open read cursor on +** transfer all records in over to
    +** close cursors +** foreach index on
    +** open a write cursor on the
    index +** open a read cursor on the corresponding index +** transfer all records from the read to the write cursors +** close cursors +** end foreach +** +** The 3rd template is for when the second template does not apply +** and the SELECT clause does not read from
    at any time. +** The generated code follows this template: +** +** X <- A +** goto B +** A: setup for the SELECT +** loop over the rows in the SELECT +** load values into registers R..R+n +** yield X +** end loop +** cleanup after the SELECT +** end-coroutine X +** B: open write cursor to
    and its indices +** C: yield X, at EOF goto D +** insert the select result into
    from R..R+n +** goto C +** D: cleanup +** +** The 4th template is used if the insert statement takes its +** values from a SELECT but the data is being inserted into a table +** that is also read as part of the SELECT. In the third form, +** we have to use an intermediate table to store the results of +** the select. The template is like this: +** +** X <- A +** goto B +** A: setup for the SELECT +** loop over the tables in the SELECT +** load value into register R..R+n +** yield X +** end loop +** cleanup after the SELECT +** end co-routine R +** B: open temp table +** L: yield X, at EOF goto M +** insert row from R..R+n into temp table +** goto L +** M: open write cursor to
    and its indices +** rewind temp table +** C: loop over rows of intermediate table +** transfer values form intermediate table into
    +** end loop +** D: cleanup +*/ +SQLITE_PRIVATE void sqlite3Insert( + Parse *pParse, /* Parser context */ + SrcList *pTabList, /* Name of table into which we are inserting */ + Select *pSelect, /* A SELECT statement to use as the data source */ + IdList *pColumn, /* Column names corresponding to IDLIST. */ + int onError, /* How to handle constraint errors */ + Upsert *pUpsert /* ON CONFLICT clauses for upsert, or NULL */ +){ + sqlite3 *db; /* The main database structure */ + Table *pTab; /* The table to insert into. aka TABLE */ + int i, j; /* Loop counters */ + Vdbe *v; /* Generate code into this virtual machine */ + Index *pIdx; /* For looping over indices of the table */ + int nColumn; /* Number of columns in the data */ + int nHidden = 0; /* Number of hidden columns if TABLE is virtual */ + int iDataCur = 0; /* VDBE cursor that is the main data repository */ + int iIdxCur = 0; /* First index cursor */ + int ipkColumn = -1; /* Column that is the INTEGER PRIMARY KEY */ + int endOfLoop; /* Label for the end of the insertion loop */ + int srcTab = 0; /* Data comes from this temporary cursor if >=0 */ + int addrInsTop = 0; /* Jump to label "D" */ + int addrCont = 0; /* Top of insert loop. Label "C" in templates 3 and 4 */ + SelectDest dest; /* Destination for SELECT on rhs of INSERT */ + int iDb; /* Index of database holding TABLE */ + u8 useTempTable = 0; /* Store SELECT results in intermediate table */ + u8 appendFlag = 0; /* True if the insert is likely to be an append */ + u8 withoutRowid; /* 0 for normal table. 1 for WITHOUT ROWID table */ + u8 bIdListInOrder; /* True if IDLIST is in table order */ + ExprList *pList = 0; /* List of VALUES() to be inserted */ + + /* Register allocations */ + int regFromSelect = 0;/* Base register for data coming from SELECT */ + int regAutoinc = 0; /* Register holding the AUTOINCREMENT counter */ + int regRowCount = 0; /* Memory cell used for the row counter */ + int regIns; /* Block of regs holding rowid+data being inserted */ + int regRowid; /* registers holding insert rowid */ + int regData; /* register holding first column to insert */ + int *aRegIdx = 0; /* One register allocated to each index */ + +#ifndef SQLITE_OMIT_TRIGGER + int isView; /* True if attempting to insert into a view */ + Trigger *pTrigger; /* List of triggers on pTab, if required */ + int tmask; /* Mask of trigger times */ +#endif + + db = pParse->db; + if( pParse->nErr || db->mallocFailed ){ + goto insert_cleanup; + } + dest.iSDParm = 0; /* Suppress a harmless compiler warning */ + + /* If the Select object is really just a simple VALUES() list with a + ** single row (the common case) then keep that one row of values + ** and discard the other (unused) parts of the pSelect object + */ + if( pSelect && (pSelect->selFlags & SF_Values)!=0 && pSelect->pPrior==0 ){ + pList = pSelect->pEList; + pSelect->pEList = 0; + sqlite3SelectDelete(db, pSelect); + pSelect = 0; + } + + /* Locate the table into which we will be inserting new information. + */ + assert( pTabList->nSrc==1 ); + pTab = sqlite3SrcListLookup(pParse, pTabList); + if( pTab==0 ){ + goto insert_cleanup; + } + iDb = sqlite3SchemaToIndex(db, pTab->pSchema); + assert( iDbnDb ); + if( sqlite3AuthCheck(pParse, SQLITE_INSERT, pTab->zName, 0, + db->aDb[iDb].zDbSName) ){ + goto insert_cleanup; + } + withoutRowid = !HasRowid(pTab); + + /* Figure out if we have any triggers and if the table being + ** inserted into is a view + */ +#ifndef SQLITE_OMIT_TRIGGER + pTrigger = sqlite3TriggersExist(pParse, pTab, TK_INSERT, 0, &tmask); + isView = pTab->pSelect!=0; +#else +# define pTrigger 0 +# define tmask 0 +# define isView 0 +#endif +#ifdef SQLITE_OMIT_VIEW +# undef isView +# define isView 0 +#endif + assert( (pTrigger && tmask) || (pTrigger==0 && tmask==0) ); + + /* If pTab is really a view, make sure it has been initialized. + ** ViewGetColumnNames() is a no-op if pTab is not a view. + */ + if( sqlite3ViewGetColumnNames(pParse, pTab) ){ + goto insert_cleanup; + } + + /* Cannot insert into a read-only table. + */ + if( sqlite3IsReadOnly(pParse, pTab, tmask) ){ + goto insert_cleanup; + } + + /* Allocate a VDBE + */ + v = sqlite3GetVdbe(pParse); + if( v==0 ) goto insert_cleanup; + if( pParse->nested==0 ) sqlite3VdbeCountChanges(v); + sqlite3BeginWriteOperation(pParse, pSelect || pTrigger, iDb); + +#ifndef SQLITE_OMIT_XFER_OPT + /* If the statement is of the form + ** + ** INSERT INTO SELECT * FROM ; + ** + ** Then special optimizations can be applied that make the transfer + ** very fast and which reduce fragmentation of indices. + ** + ** This is the 2nd template. + */ + if( pColumn==0 && xferOptimization(pParse, pTab, pSelect, onError, iDb) ){ + assert( !pTrigger ); + assert( pList==0 ); + goto insert_end; + } +#endif /* SQLITE_OMIT_XFER_OPT */ + + /* If this is an AUTOINCREMENT table, look up the sequence number in the + ** sqlite_sequence table and store it in memory cell regAutoinc. + */ + regAutoinc = autoIncBegin(pParse, iDb, pTab); + + /* Allocate registers for holding the rowid of the new row, + ** the content of the new row, and the assembled row record. + */ + regRowid = regIns = pParse->nMem+1; + pParse->nMem += pTab->nCol + 1; + if( IsVirtual(pTab) ){ + regRowid++; + pParse->nMem++; + } + regData = regRowid+1; + + /* If the INSERT statement included an IDLIST term, then make sure + ** all elements of the IDLIST really are columns of the table and + ** remember the column indices. + ** + ** If the table has an INTEGER PRIMARY KEY column and that column + ** is named in the IDLIST, then record in the ipkColumn variable + ** the index into IDLIST of the primary key column. ipkColumn is + ** the index of the primary key as it appears in IDLIST, not as + ** is appears in the original table. (The index of the INTEGER + ** PRIMARY KEY in the original table is pTab->iPKey.) + */ + bIdListInOrder = (pTab->tabFlags & TF_OOOHidden)==0; + if( pColumn ){ + for(i=0; inId; i++){ + pColumn->a[i].idx = -1; + } + for(i=0; inId; i++){ + for(j=0; jnCol; j++){ + if( sqlite3StrICmp(pColumn->a[i].zName, pTab->aCol[j].zName)==0 ){ + pColumn->a[i].idx = j; + if( i!=j ) bIdListInOrder = 0; + if( j==pTab->iPKey ){ + ipkColumn = i; assert( !withoutRowid ); + } + break; + } + } + if( j>=pTab->nCol ){ + if( sqlite3IsRowid(pColumn->a[i].zName) && !withoutRowid ){ + ipkColumn = i; + bIdListInOrder = 0; + }else{ + sqlite3ErrorMsg(pParse, "table %S has no column named %s", + pTabList, 0, pColumn->a[i].zName); + pParse->checkSchema = 1; + goto insert_cleanup; + } + } + } + } + + /* Figure out how many columns of data are supplied. If the data + ** is coming from a SELECT statement, then generate a co-routine that + ** produces a single row of the SELECT on each invocation. The + ** co-routine is the common header to the 3rd and 4th templates. + */ + if( pSelect ){ + /* Data is coming from a SELECT or from a multi-row VALUES clause. + ** Generate a co-routine to run the SELECT. */ + int regYield; /* Register holding co-routine entry-point */ + int addrTop; /* Top of the co-routine */ + int rc; /* Result code */ + + regYield = ++pParse->nMem; + addrTop = sqlite3VdbeCurrentAddr(v) + 1; + sqlite3VdbeAddOp3(v, OP_InitCoroutine, regYield, 0, addrTop); + sqlite3SelectDestInit(&dest, SRT_Coroutine, regYield); + dest.iSdst = bIdListInOrder ? regData : 0; + dest.nSdst = pTab->nCol; + rc = sqlite3Select(pParse, pSelect, &dest); + regFromSelect = dest.iSdst; + if( rc || db->mallocFailed || pParse->nErr ) goto insert_cleanup; + sqlite3VdbeEndCoroutine(v, regYield); + sqlite3VdbeJumpHere(v, addrTop - 1); /* label B: */ + assert( pSelect->pEList ); + nColumn = pSelect->pEList->nExpr; + + /* Set useTempTable to TRUE if the result of the SELECT statement + ** should be written into a temporary table (template 4). Set to + ** FALSE if each output row of the SELECT can be written directly into + ** the destination table (template 3). + ** + ** A temp table must be used if the table being updated is also one + ** of the tables being read by the SELECT statement. Also use a + ** temp table in the case of row triggers. + */ + if( pTrigger || readsTable(pParse, iDb, pTab) ){ + useTempTable = 1; + } + + if( useTempTable ){ + /* Invoke the coroutine to extract information from the SELECT + ** and add it to a transient table srcTab. The code generated + ** here is from the 4th template: + ** + ** B: open temp table + ** L: yield X, goto M at EOF + ** insert row from R..R+n into temp table + ** goto L + ** M: ... + */ + int regRec; /* Register to hold packed record */ + int regTempRowid; /* Register to hold temp table ROWID */ + int addrL; /* Label "L" */ + + srcTab = pParse->nTab++; + regRec = sqlite3GetTempReg(pParse); + regTempRowid = sqlite3GetTempReg(pParse); + sqlite3VdbeAddOp2(v, OP_OpenEphemeral, srcTab, nColumn); + addrL = sqlite3VdbeAddOp1(v, OP_Yield, dest.iSDParm); VdbeCoverage(v); + sqlite3VdbeAddOp3(v, OP_MakeRecord, regFromSelect, nColumn, regRec); + sqlite3VdbeAddOp2(v, OP_NewRowid, srcTab, regTempRowid); + sqlite3VdbeAddOp3(v, OP_Insert, srcTab, regRec, regTempRowid); + sqlite3VdbeGoto(v, addrL); + sqlite3VdbeJumpHere(v, addrL); + sqlite3ReleaseTempReg(pParse, regRec); + sqlite3ReleaseTempReg(pParse, regTempRowid); + } + }else{ + /* This is the case if the data for the INSERT is coming from a + ** single-row VALUES clause + */ + NameContext sNC; + memset(&sNC, 0, sizeof(sNC)); + sNC.pParse = pParse; + srcTab = -1; + assert( useTempTable==0 ); + if( pList ){ + nColumn = pList->nExpr; + if( sqlite3ResolveExprListNames(&sNC, pList) ){ + goto insert_cleanup; + } + }else{ + nColumn = 0; + } + } + + /* If there is no IDLIST term but the table has an integer primary + ** key, the set the ipkColumn variable to the integer primary key + ** column index in the original table definition. + */ + if( pColumn==0 && nColumn>0 ){ + ipkColumn = pTab->iPKey; + } + + /* Make sure the number of columns in the source data matches the number + ** of columns to be inserted into the table. + */ + for(i=0; inCol; i++){ + nHidden += (IsHiddenColumn(&pTab->aCol[i]) ? 1 : 0); + } + if( pColumn==0 && nColumn && nColumn!=(pTab->nCol-nHidden) ){ + sqlite3ErrorMsg(pParse, + "table %S has %d columns but %d values were supplied", + pTabList, 0, pTab->nCol-nHidden, nColumn); + goto insert_cleanup; + } + if( pColumn!=0 && nColumn!=pColumn->nId ){ + sqlite3ErrorMsg(pParse, "%d values for %d columns", nColumn, pColumn->nId); + goto insert_cleanup; + } + + /* Initialize the count of rows to be inserted + */ + if( (db->flags & SQLITE_CountRows)!=0 + && !pParse->nested + && !pParse->pTriggerTab + ){ + regRowCount = ++pParse->nMem; + sqlite3VdbeAddOp2(v, OP_Integer, 0, regRowCount); + } + + /* If this is not a view, open the table and and all indices */ + if( !isView ){ + int nIdx; + nIdx = sqlite3OpenTableAndIndices(pParse, pTab, OP_OpenWrite, 0, -1, 0, + &iDataCur, &iIdxCur); + aRegIdx = sqlite3DbMallocRawNN(db, sizeof(int)*(nIdx+2)); + if( aRegIdx==0 ){ + goto insert_cleanup; + } + for(i=0, pIdx=pTab->pIndex; ipNext, i++){ + assert( pIdx ); + aRegIdx[i] = ++pParse->nMem; + pParse->nMem += pIdx->nColumn; + } + aRegIdx[i] = ++pParse->nMem; /* Register to store the table record */ + } +#ifndef SQLITE_OMIT_UPSERT + if( pUpsert ){ + if( IsVirtual(pTab) ){ + sqlite3ErrorMsg(pParse, "UPSERT not implemented for virtual table \"%s\"", + pTab->zName); + goto insert_cleanup; + } + if( sqlite3HasExplicitNulls(pParse, pUpsert->pUpsertTarget) ){ + goto insert_cleanup; + } + pTabList->a[0].iCursor = iDataCur; + pUpsert->pUpsertSrc = pTabList; + pUpsert->regData = regData; + pUpsert->iDataCur = iDataCur; + pUpsert->iIdxCur = iIdxCur; + if( pUpsert->pUpsertTarget ){ + sqlite3UpsertAnalyzeTarget(pParse, pTabList, pUpsert); + } + } +#endif + + + /* This is the top of the main insertion loop */ + if( useTempTable ){ + /* This block codes the top of loop only. The complete loop is the + ** following pseudocode (template 4): + ** + ** rewind temp table, if empty goto D + ** C: loop over rows of intermediate table + ** transfer values form intermediate table into
    + ** end loop + ** D: ... + */ + addrInsTop = sqlite3VdbeAddOp1(v, OP_Rewind, srcTab); VdbeCoverage(v); + addrCont = sqlite3VdbeCurrentAddr(v); + }else if( pSelect ){ + /* This block codes the top of loop only. The complete loop is the + ** following pseudocode (template 3): + ** + ** C: yield X, at EOF goto D + ** insert the select result into
    from R..R+n + ** goto C + ** D: ... + */ + addrInsTop = addrCont = sqlite3VdbeAddOp1(v, OP_Yield, dest.iSDParm); + VdbeCoverage(v); + } + + /* Run the BEFORE and INSTEAD OF triggers, if there are any + */ + endOfLoop = sqlite3VdbeMakeLabel(pParse); + if( tmask & TRIGGER_BEFORE ){ + int regCols = sqlite3GetTempRange(pParse, pTab->nCol+1); + + /* build the NEW.* reference row. Note that if there is an INTEGER + ** PRIMARY KEY into which a NULL is being inserted, that NULL will be + ** translated into a unique ID for the row. But on a BEFORE trigger, + ** we do not know what the unique ID will be (because the insert has + ** not happened yet) so we substitute a rowid of -1 + */ + if( ipkColumn<0 ){ + sqlite3VdbeAddOp2(v, OP_Integer, -1, regCols); + }else{ + int addr1; + assert( !withoutRowid ); + if( useTempTable ){ + sqlite3VdbeAddOp3(v, OP_Column, srcTab, ipkColumn, regCols); + }else{ + assert( pSelect==0 ); /* Otherwise useTempTable is true */ + sqlite3ExprCode(pParse, pList->a[ipkColumn].pExpr, regCols); + } + addr1 = sqlite3VdbeAddOp1(v, OP_NotNull, regCols); VdbeCoverage(v); + sqlite3VdbeAddOp2(v, OP_Integer, -1, regCols); + sqlite3VdbeJumpHere(v, addr1); + sqlite3VdbeAddOp1(v, OP_MustBeInt, regCols); VdbeCoverage(v); + } + + /* Cannot have triggers on a virtual table. If it were possible, + ** this block would have to account for hidden column. + */ + assert( !IsVirtual(pTab) ); + + /* Create the new column data + */ + for(i=j=0; inCol; i++){ + if( pColumn ){ + for(j=0; jnId; j++){ + if( pColumn->a[j].idx==i ) break; + } + } + if( (!useTempTable && !pList) || (pColumn && j>=pColumn->nId) + || (pColumn==0 && IsOrdinaryHiddenColumn(&pTab->aCol[i])) ){ + sqlite3ExprCode(pParse, pTab->aCol[i].pDflt, regCols+i+1); + }else if( useTempTable ){ + sqlite3VdbeAddOp3(v, OP_Column, srcTab, j, regCols+i+1); + }else{ + assert( pSelect==0 ); /* Otherwise useTempTable is true */ + sqlite3ExprCodeAndCache(pParse, pList->a[j].pExpr, regCols+i+1); + } + if( pColumn==0 && !IsOrdinaryHiddenColumn(&pTab->aCol[i]) ) j++; + } + + /* If this is an INSERT on a view with an INSTEAD OF INSERT trigger, + ** do not attempt any conversions before assembling the record. + ** If this is a real table, attempt conversions as required by the + ** table column affinities. + */ + if( !isView ){ + sqlite3TableAffinity(v, pTab, regCols+1); + } + + /* Fire BEFORE or INSTEAD OF triggers */ + sqlite3CodeRowTrigger(pParse, pTrigger, TK_INSERT, 0, TRIGGER_BEFORE, + pTab, regCols-pTab->nCol-1, onError, endOfLoop); + + sqlite3ReleaseTempRange(pParse, regCols, pTab->nCol+1); + } + + /* Compute the content of the next row to insert into a range of + ** registers beginning at regIns. + */ + if( !isView ){ + if( IsVirtual(pTab) ){ + /* The row that the VUpdate opcode will delete: none */ + sqlite3VdbeAddOp2(v, OP_Null, 0, regIns); + } + if( ipkColumn>=0 ){ + if( useTempTable ){ + sqlite3VdbeAddOp3(v, OP_Column, srcTab, ipkColumn, regRowid); + }else if( pSelect ){ + sqlite3VdbeAddOp2(v, OP_Copy, regFromSelect+ipkColumn, regRowid); + }else{ + Expr *pIpk = pList->a[ipkColumn].pExpr; + if( pIpk->op==TK_NULL && !IsVirtual(pTab) ){ + sqlite3VdbeAddOp3(v, OP_NewRowid, iDataCur, regRowid, regAutoinc); + appendFlag = 1; + }else{ + sqlite3ExprCode(pParse, pList->a[ipkColumn].pExpr, regRowid); + } + } + /* If the PRIMARY KEY expression is NULL, then use OP_NewRowid + ** to generate a unique primary key value. + */ + if( !appendFlag ){ + int addr1; + if( !IsVirtual(pTab) ){ + addr1 = sqlite3VdbeAddOp1(v, OP_NotNull, regRowid); VdbeCoverage(v); + sqlite3VdbeAddOp3(v, OP_NewRowid, iDataCur, regRowid, regAutoinc); + sqlite3VdbeJumpHere(v, addr1); + }else{ + addr1 = sqlite3VdbeCurrentAddr(v); + sqlite3VdbeAddOp2(v, OP_IsNull, regRowid, addr1+2); VdbeCoverage(v); + } + sqlite3VdbeAddOp1(v, OP_MustBeInt, regRowid); VdbeCoverage(v); + } + }else if( IsVirtual(pTab) || withoutRowid ){ + sqlite3VdbeAddOp2(v, OP_Null, 0, regRowid); + }else{ + sqlite3VdbeAddOp3(v, OP_NewRowid, iDataCur, regRowid, regAutoinc); + appendFlag = 1; + } + autoIncStep(pParse, regAutoinc, regRowid); + + /* Compute data for all columns of the new entry, beginning + ** with the first column. + */ + nHidden = 0; + for(i=0; inCol; i++){ + int iRegStore = regRowid+1+i; + if( i==pTab->iPKey ){ + /* The value of the INTEGER PRIMARY KEY column is always a NULL. + ** Whenever this column is read, the rowid will be substituted + ** in its place. Hence, fill this column with a NULL to avoid + ** taking up data space with information that will never be used. + ** As there may be shallow copies of this value, make it a soft-NULL */ + sqlite3VdbeAddOp1(v, OP_SoftNull, iRegStore); + continue; + } + if( pColumn==0 ){ + if( IsHiddenColumn(&pTab->aCol[i]) ){ + j = -1; + nHidden++; + }else{ + j = i - nHidden; + } + }else{ + for(j=0; jnId; j++){ + if( pColumn->a[j].idx==i ) break; + } + } + if( j<0 || nColumn==0 || (pColumn && j>=pColumn->nId) ){ + sqlite3ExprCodeFactorable(pParse, pTab->aCol[i].pDflt, iRegStore); + }else if( useTempTable ){ + sqlite3VdbeAddOp3(v, OP_Column, srcTab, j, iRegStore); + }else if( pSelect ){ + if( regFromSelect!=regData ){ + sqlite3VdbeAddOp2(v, OP_SCopy, regFromSelect+j, iRegStore); + } + }else{ + sqlite3ExprCode(pParse, pList->a[j].pExpr, iRegStore); + } + } + + /* Generate code to check constraints and generate index keys and + ** do the insertion. + */ +#ifndef SQLITE_OMIT_VIRTUALTABLE + if( IsVirtual(pTab) ){ + const char *pVTab = (const char *)sqlite3GetVTable(db, pTab); + sqlite3VtabMakeWritable(pParse, pTab); + sqlite3VdbeAddOp4(v, OP_VUpdate, 1, pTab->nCol+2, regIns, pVTab, P4_VTAB); + sqlite3VdbeChangeP5(v, onError==OE_Default ? OE_Abort : onError); + sqlite3MayAbort(pParse); + }else +#endif + { + int isReplace; /* Set to true if constraints may cause a replace */ + int bUseSeek; /* True to use OPFLAG_SEEKRESULT */ + sqlite3GenerateConstraintChecks(pParse, pTab, aRegIdx, iDataCur, iIdxCur, + regIns, 0, ipkColumn>=0, onError, endOfLoop, &isReplace, 0, pUpsert + ); + sqlite3FkCheck(pParse, pTab, 0, regIns, 0, 0); + + /* Set the OPFLAG_USESEEKRESULT flag if either (a) there are no REPLACE + ** constraints or (b) there are no triggers and this table is not a + ** parent table in a foreign key constraint. It is safe to set the + ** flag in the second case as if any REPLACE constraint is hit, an + ** OP_Delete or OP_IdxDelete instruction will be executed on each + ** cursor that is disturbed. And these instructions both clear the + ** VdbeCursor.seekResult variable, disabling the OPFLAG_USESEEKRESULT + ** functionality. */ + bUseSeek = (isReplace==0 || (pTrigger==0 && + ((db->flags & SQLITE_ForeignKeys)==0 || sqlite3FkReferences(pTab)==0) + )); + sqlite3CompleteInsertion(pParse, pTab, iDataCur, iIdxCur, + regIns, aRegIdx, 0, appendFlag, bUseSeek + ); + } + } + + /* Update the count of rows that are inserted + */ + if( regRowCount ){ + sqlite3VdbeAddOp2(v, OP_AddImm, regRowCount, 1); + } + + if( pTrigger ){ + /* Code AFTER triggers */ + sqlite3CodeRowTrigger(pParse, pTrigger, TK_INSERT, 0, TRIGGER_AFTER, + pTab, regData-2-pTab->nCol, onError, endOfLoop); + } + + /* The bottom of the main insertion loop, if the data source + ** is a SELECT statement. + */ + sqlite3VdbeResolveLabel(v, endOfLoop); + if( useTempTable ){ + sqlite3VdbeAddOp2(v, OP_Next, srcTab, addrCont); VdbeCoverage(v); + sqlite3VdbeJumpHere(v, addrInsTop); + sqlite3VdbeAddOp1(v, OP_Close, srcTab); + }else if( pSelect ){ + sqlite3VdbeGoto(v, addrCont); + sqlite3VdbeJumpHere(v, addrInsTop); + } + +insert_end: + /* Update the sqlite_sequence table by storing the content of the + ** maximum rowid counter values recorded while inserting into + ** autoincrement tables. + */ + if( pParse->nested==0 && pParse->pTriggerTab==0 ){ + sqlite3AutoincrementEnd(pParse); + } + + /* + ** Return the number of rows inserted. If this routine is + ** generating code because of a call to sqlite3NestedParse(), do not + ** invoke the callback function. + */ + if( regRowCount ){ + sqlite3VdbeAddOp2(v, OP_ResultRow, regRowCount, 1); + sqlite3VdbeSetNumCols(v, 1); + sqlite3VdbeSetColName(v, 0, COLNAME_NAME, "rows inserted", SQLITE_STATIC); + } + +insert_cleanup: + sqlite3SrcListDelete(db, pTabList); + sqlite3ExprListDelete(db, pList); + sqlite3UpsertDelete(db, pUpsert); + sqlite3SelectDelete(db, pSelect); + sqlite3IdListDelete(db, pColumn); + sqlite3DbFree(db, aRegIdx); +} + +/* Make sure "isView" and other macros defined above are undefined. Otherwise +** they may interfere with compilation of other functions in this file +** (or in another file, if this file becomes part of the amalgamation). */ +#ifdef isView + #undef isView +#endif +#ifdef pTrigger + #undef pTrigger +#endif +#ifdef tmask + #undef tmask +#endif + +/* +** Meanings of bits in of pWalker->eCode for +** sqlite3ExprReferencesUpdatedColumn() +*/ +#define CKCNSTRNT_COLUMN 0x01 /* CHECK constraint uses a changing column */ +#define CKCNSTRNT_ROWID 0x02 /* CHECK constraint references the ROWID */ + +/* This is the Walker callback from sqlite3ExprReferencesUpdatedColumn(). +* Set bit 0x01 of pWalker->eCode if pWalker->eCode to 0 and if this +** expression node references any of the +** columns that are being modifed by an UPDATE statement. +*/ +static int checkConstraintExprNode(Walker *pWalker, Expr *pExpr){ + if( pExpr->op==TK_COLUMN ){ + assert( pExpr->iColumn>=0 || pExpr->iColumn==-1 ); + if( pExpr->iColumn>=0 ){ + if( pWalker->u.aiCol[pExpr->iColumn]>=0 ){ + pWalker->eCode |= CKCNSTRNT_COLUMN; + } + }else{ + pWalker->eCode |= CKCNSTRNT_ROWID; + } + } + return WRC_Continue; +} + +/* +** pExpr is a CHECK constraint on a row that is being UPDATE-ed. The +** only columns that are modified by the UPDATE are those for which +** aiChng[i]>=0, and also the ROWID is modified if chngRowid is true. +** +** Return true if CHECK constraint pExpr uses any of the +** changing columns (or the rowid if it is changing). In other words, +** return true if this CHECK constraint must be validated for +** the new row in the UPDATE statement. +** +** 2018-09-15: pExpr might also be an expression for an index-on-expressions. +** The operation of this routine is the same - return true if an only if +** the expression uses one or more of columns identified by the second and +** third arguments. +*/ +SQLITE_PRIVATE int sqlite3ExprReferencesUpdatedColumn( + Expr *pExpr, /* The expression to be checked */ + int *aiChng, /* aiChng[x]>=0 if column x changed by the UPDATE */ + int chngRowid /* True if UPDATE changes the rowid */ +){ + Walker w; + memset(&w, 0, sizeof(w)); + w.eCode = 0; + w.xExprCallback = checkConstraintExprNode; + w.u.aiCol = aiChng; + sqlite3WalkExpr(&w, pExpr); + if( !chngRowid ){ + testcase( (w.eCode & CKCNSTRNT_ROWID)!=0 ); + w.eCode &= ~CKCNSTRNT_ROWID; + } + testcase( w.eCode==0 ); + testcase( w.eCode==CKCNSTRNT_COLUMN ); + testcase( w.eCode==CKCNSTRNT_ROWID ); + testcase( w.eCode==(CKCNSTRNT_ROWID|CKCNSTRNT_COLUMN) ); + return w.eCode!=0; +} + +/* +** Generate code to do constraint checks prior to an INSERT or an UPDATE +** on table pTab. +** +** The regNewData parameter is the first register in a range that contains +** the data to be inserted or the data after the update. There will be +** pTab->nCol+1 registers in this range. The first register (the one +** that regNewData points to) will contain the new rowid, or NULL in the +** case of a WITHOUT ROWID table. The second register in the range will +** contain the content of the first table column. The third register will +** contain the content of the second table column. And so forth. +** +** The regOldData parameter is similar to regNewData except that it contains +** the data prior to an UPDATE rather than afterwards. regOldData is zero +** for an INSERT. This routine can distinguish between UPDATE and INSERT by +** checking regOldData for zero. +** +** For an UPDATE, the pkChng boolean is true if the true primary key (the +** rowid for a normal table or the PRIMARY KEY for a WITHOUT ROWID table) +** might be modified by the UPDATE. If pkChng is false, then the key of +** the iDataCur content table is guaranteed to be unchanged by the UPDATE. +** +** For an INSERT, the pkChng boolean indicates whether or not the rowid +** was explicitly specified as part of the INSERT statement. If pkChng +** is zero, it means that the either rowid is computed automatically or +** that the table is a WITHOUT ROWID table and has no rowid. On an INSERT, +** pkChng will only be true if the INSERT statement provides an integer +** value for either the rowid column or its INTEGER PRIMARY KEY alias. +** +** The code generated by this routine will store new index entries into +** registers identified by aRegIdx[]. No index entry is created for +** indices where aRegIdx[i]==0. The order of indices in aRegIdx[] is +** the same as the order of indices on the linked list of indices +** at pTab->pIndex. +** +** (2019-05-07) The generated code also creates a new record for the +** main table, if pTab is a rowid table, and stores that record in the +** register identified by aRegIdx[nIdx] - in other words in the first +** entry of aRegIdx[] past the last index. It is important that the +** record be generated during constraint checks to avoid affinity changes +** to the register content that occur after constraint checks but before +** the new record is inserted. +** +** The caller must have already opened writeable cursors on the main +** table and all applicable indices (that is to say, all indices for which +** aRegIdx[] is not zero). iDataCur is the cursor for the main table when +** inserting or updating a rowid table, or the cursor for the PRIMARY KEY +** index when operating on a WITHOUT ROWID table. iIdxCur is the cursor +** for the first index in the pTab->pIndex list. Cursors for other indices +** are at iIdxCur+N for the N-th element of the pTab->pIndex list. +** +** This routine also generates code to check constraints. NOT NULL, +** CHECK, and UNIQUE constraints are all checked. If a constraint fails, +** then the appropriate action is performed. There are five possible +** actions: ROLLBACK, ABORT, FAIL, REPLACE, and IGNORE. +** +** Constraint type Action What Happens +** --------------- ---------- ---------------------------------------- +** any ROLLBACK The current transaction is rolled back and +** sqlite3_step() returns immediately with a +** return code of SQLITE_CONSTRAINT. +** +** any ABORT Back out changes from the current command +** only (do not do a complete rollback) then +** cause sqlite3_step() to return immediately +** with SQLITE_CONSTRAINT. +** +** any FAIL Sqlite3_step() returns immediately with a +** return code of SQLITE_CONSTRAINT. The +** transaction is not rolled back and any +** changes to prior rows are retained. +** +** any IGNORE The attempt in insert or update the current +** row is skipped, without throwing an error. +** Processing continues with the next row. +** (There is an immediate jump to ignoreDest.) +** +** NOT NULL REPLACE The NULL value is replace by the default +** value for that column. If the default value +** is NULL, the action is the same as ABORT. +** +** UNIQUE REPLACE The other row that conflicts with the row +** being inserted is removed. +** +** CHECK REPLACE Illegal. The results in an exception. +** +** Which action to take is determined by the overrideError parameter. +** Or if overrideError==OE_Default, then the pParse->onError parameter +** is used. Or if pParse->onError==OE_Default then the onError value +** for the constraint is used. +*/ +SQLITE_PRIVATE void sqlite3GenerateConstraintChecks( + Parse *pParse, /* The parser context */ + Table *pTab, /* The table being inserted or updated */ + int *aRegIdx, /* Use register aRegIdx[i] for index i. 0 for unused */ + int iDataCur, /* Canonical data cursor (main table or PK index) */ + int iIdxCur, /* First index cursor */ + int regNewData, /* First register in a range holding values to insert */ + int regOldData, /* Previous content. 0 for INSERTs */ + u8 pkChng, /* Non-zero if the rowid or PRIMARY KEY changed */ + u8 overrideError, /* Override onError to this if not OE_Default */ + int ignoreDest, /* Jump to this label on an OE_Ignore resolution */ + int *pbMayReplace, /* OUT: Set to true if constraint may cause a replace */ + int *aiChng, /* column i is unchanged if aiChng[i]<0 */ + Upsert *pUpsert /* ON CONFLICT clauses, if any. NULL otherwise */ +){ + Vdbe *v; /* VDBE under constrution */ + Index *pIdx; /* Pointer to one of the indices */ + Index *pPk = 0; /* The PRIMARY KEY index */ + sqlite3 *db; /* Database connection */ + int i; /* loop counter */ + int ix; /* Index loop counter */ + int nCol; /* Number of columns */ + int onError; /* Conflict resolution strategy */ + int addr1; /* Address of jump instruction */ + int seenReplace = 0; /* True if REPLACE is used to resolve INT PK conflict */ + int nPkField; /* Number of fields in PRIMARY KEY. 1 for ROWID tables */ + Index *pUpIdx = 0; /* Index to which to apply the upsert */ + u8 isUpdate; /* True if this is an UPDATE operation */ + u8 bAffinityDone = 0; /* True if the OP_Affinity operation has been run */ + int upsertBypass = 0; /* Address of Goto to bypass upsert subroutine */ + int upsertJump = 0; /* Address of Goto that jumps into upsert subroutine */ + int ipkTop = 0; /* Top of the IPK uniqueness check */ + int ipkBottom = 0; /* OP_Goto at the end of the IPK uniqueness check */ + + isUpdate = regOldData!=0; + db = pParse->db; + v = sqlite3GetVdbe(pParse); + assert( v!=0 ); + assert( pTab->pSelect==0 ); /* This table is not a VIEW */ + nCol = pTab->nCol; + + /* pPk is the PRIMARY KEY index for WITHOUT ROWID tables and NULL for + ** normal rowid tables. nPkField is the number of key fields in the + ** pPk index or 1 for a rowid table. In other words, nPkField is the + ** number of fields in the true primary key of the table. */ + if( HasRowid(pTab) ){ + pPk = 0; + nPkField = 1; + }else{ + pPk = sqlite3PrimaryKeyIndex(pTab); + nPkField = pPk->nKeyCol; + } + + /* Record that this module has started */ + VdbeModuleComment((v, "BEGIN: GenCnstCks(%d,%d,%d,%d,%d)", + iDataCur, iIdxCur, regNewData, regOldData, pkChng)); + + /* Test all NOT NULL constraints. + */ + for(i=0; iiPKey ){ + continue; /* ROWID is never NULL */ + } + if( aiChng && aiChng[i]<0 ){ + /* Don't bother checking for NOT NULL on columns that do not change */ + continue; + } + onError = pTab->aCol[i].notNull; + if( onError==OE_None ) continue; /* This column is allowed to be NULL */ + if( overrideError!=OE_Default ){ + onError = overrideError; + }else if( onError==OE_Default ){ + onError = OE_Abort; + } + if( onError==OE_Replace && pTab->aCol[i].pDflt==0 ){ + onError = OE_Abort; + } + assert( onError==OE_Rollback || onError==OE_Abort || onError==OE_Fail + || onError==OE_Ignore || onError==OE_Replace ); + addr1 = 0; + switch( onError ){ + case OE_Replace: { + assert( onError==OE_Replace ); + addr1 = sqlite3VdbeMakeLabel(pParse); + sqlite3VdbeAddOp2(v, OP_NotNull, regNewData+1+i, addr1); + VdbeCoverage(v); + sqlite3ExprCode(pParse, pTab->aCol[i].pDflt, regNewData+1+i); + sqlite3VdbeAddOp2(v, OP_NotNull, regNewData+1+i, addr1); + VdbeCoverage(v); + onError = OE_Abort; + /* Fall through into the OE_Abort case to generate code that runs + ** if both the input and the default value are NULL */ + } + case OE_Abort: + sqlite3MayAbort(pParse); + /* Fall through */ + case OE_Rollback: + case OE_Fail: { + char *zMsg = sqlite3MPrintf(db, "%s.%s", pTab->zName, + pTab->aCol[i].zName); + sqlite3VdbeAddOp3(v, OP_HaltIfNull, SQLITE_CONSTRAINT_NOTNULL, onError, + regNewData+1+i); + sqlite3VdbeAppendP4(v, zMsg, P4_DYNAMIC); + sqlite3VdbeChangeP5(v, P5_ConstraintNotNull); + VdbeCoverage(v); + if( addr1 ) sqlite3VdbeResolveLabel(v, addr1); + break; + } + default: { + assert( onError==OE_Ignore ); + sqlite3VdbeAddOp2(v, OP_IsNull, regNewData+1+i, ignoreDest); + VdbeCoverage(v); + break; + } + } + } + + /* Test all CHECK constraints + */ +#ifndef SQLITE_OMIT_CHECK + if( pTab->pCheck && (db->flags & SQLITE_IgnoreChecks)==0 ){ + ExprList *pCheck = pTab->pCheck; + pParse->iSelfTab = -(regNewData+1); + onError = overrideError!=OE_Default ? overrideError : OE_Abort; + for(i=0; inExpr; i++){ + int allOk; + Expr *pExpr = pCheck->a[i].pExpr; + if( aiChng + && !sqlite3ExprReferencesUpdatedColumn(pExpr, aiChng, pkChng) + ){ + /* The check constraints do not reference any of the columns being + ** updated so there is no point it verifying the check constraint */ + continue; + } + allOk = sqlite3VdbeMakeLabel(pParse); + sqlite3VdbeVerifyAbortable(v, onError); + sqlite3ExprIfTrue(pParse, pExpr, allOk, SQLITE_JUMPIFNULL); + if( onError==OE_Ignore ){ + sqlite3VdbeGoto(v, ignoreDest); + }else{ + char *zName = pCheck->a[i].zName; + if( zName==0 ) zName = pTab->zName; + if( onError==OE_Replace ) onError = OE_Abort; /* IMP: R-26383-51744 */ + sqlite3HaltConstraint(pParse, SQLITE_CONSTRAINT_CHECK, + onError, zName, P4_TRANSIENT, + P5_ConstraintCheck); + } + sqlite3VdbeResolveLabel(v, allOk); + } + pParse->iSelfTab = 0; + } +#endif /* !defined(SQLITE_OMIT_CHECK) */ + + /* UNIQUE and PRIMARY KEY constraints should be handled in the following + ** order: + ** + ** (1) OE_Update + ** (2) OE_Abort, OE_Fail, OE_Rollback, OE_Ignore + ** (3) OE_Replace + ** + ** OE_Fail and OE_Ignore must happen before any changes are made. + ** OE_Update guarantees that only a single row will change, so it + ** must happen before OE_Replace. Technically, OE_Abort and OE_Rollback + ** could happen in any order, but they are grouped up front for + ** convenience. + ** + ** 2018-08-14: Ticket https://www.sqlite.org/src/info/908f001483982c43 + ** The order of constraints used to have OE_Update as (2) and OE_Abort + ** and so forth as (1). But apparently PostgreSQL checks the OE_Update + ** constraint before any others, so it had to be moved. + ** + ** Constraint checking code is generated in this order: + ** (A) The rowid constraint + ** (B) Unique index constraints that do not have OE_Replace as their + ** default conflict resolution strategy + ** (C) Unique index that do use OE_Replace by default. + ** + ** The ordering of (2) and (3) is accomplished by making sure the linked + ** list of indexes attached to a table puts all OE_Replace indexes last + ** in the list. See sqlite3CreateIndex() for where that happens. + */ + + if( pUpsert ){ + if( pUpsert->pUpsertTarget==0 ){ + /* An ON CONFLICT DO NOTHING clause, without a constraint-target. + ** Make all unique constraint resolution be OE_Ignore */ + assert( pUpsert->pUpsertSet==0 ); + overrideError = OE_Ignore; + pUpsert = 0; + }else if( (pUpIdx = pUpsert->pUpsertIdx)!=0 ){ + /* If the constraint-target uniqueness check must be run first. + ** Jump to that uniqueness check now */ + upsertJump = sqlite3VdbeAddOp0(v, OP_Goto); + VdbeComment((v, "UPSERT constraint goes first")); + } + } + + /* If rowid is changing, make sure the new rowid does not previously + ** exist in the table. + */ + if( pkChng && pPk==0 ){ + int addrRowidOk = sqlite3VdbeMakeLabel(pParse); + + /* Figure out what action to take in case of a rowid collision */ + onError = pTab->keyConf; + if( overrideError!=OE_Default ){ + onError = overrideError; + }else if( onError==OE_Default ){ + onError = OE_Abort; + } + + /* figure out whether or not upsert applies in this case */ + if( pUpsert && pUpsert->pUpsertIdx==0 ){ + if( pUpsert->pUpsertSet==0 ){ + onError = OE_Ignore; /* DO NOTHING is the same as INSERT OR IGNORE */ + }else{ + onError = OE_Update; /* DO UPDATE */ + } + } + + /* If the response to a rowid conflict is REPLACE but the response + ** to some other UNIQUE constraint is FAIL or IGNORE, then we need + ** to defer the running of the rowid conflict checking until after + ** the UNIQUE constraints have run. + */ + if( onError==OE_Replace /* IPK rule is REPLACE */ + && onError!=overrideError /* Rules for other contraints are different */ + && pTab->pIndex /* There exist other constraints */ + ){ + ipkTop = sqlite3VdbeAddOp0(v, OP_Goto)+1; + VdbeComment((v, "defer IPK REPLACE until last")); + } + + if( isUpdate ){ + /* pkChng!=0 does not mean that the rowid has changed, only that + ** it might have changed. Skip the conflict logic below if the rowid + ** is unchanged. */ + sqlite3VdbeAddOp3(v, OP_Eq, regNewData, addrRowidOk, regOldData); + sqlite3VdbeChangeP5(v, SQLITE_NOTNULL); + VdbeCoverage(v); + } + + /* Check to see if the new rowid already exists in the table. Skip + ** the following conflict logic if it does not. */ + VdbeNoopComment((v, "uniqueness check for ROWID")); + sqlite3VdbeVerifyAbortable(v, onError); + sqlite3VdbeAddOp3(v, OP_NotExists, iDataCur, addrRowidOk, regNewData); + VdbeCoverage(v); + + switch( onError ){ + default: { + onError = OE_Abort; + /* Fall thru into the next case */ + } + case OE_Rollback: + case OE_Abort: + case OE_Fail: { + testcase( onError==OE_Rollback ); + testcase( onError==OE_Abort ); + testcase( onError==OE_Fail ); + sqlite3RowidConstraint(pParse, onError, pTab); + break; + } + case OE_Replace: { + /* If there are DELETE triggers on this table and the + ** recursive-triggers flag is set, call GenerateRowDelete() to + ** remove the conflicting row from the table. This will fire + ** the triggers and remove both the table and index b-tree entries. + ** + ** Otherwise, if there are no triggers or the recursive-triggers + ** flag is not set, but the table has one or more indexes, call + ** GenerateRowIndexDelete(). This removes the index b-tree entries + ** only. The table b-tree entry will be replaced by the new entry + ** when it is inserted. + ** + ** If either GenerateRowDelete() or GenerateRowIndexDelete() is called, + ** also invoke MultiWrite() to indicate that this VDBE may require + ** statement rollback (if the statement is aborted after the delete + ** takes place). Earlier versions called sqlite3MultiWrite() regardless, + ** but being more selective here allows statements like: + ** + ** REPLACE INTO t(rowid) VALUES($newrowid) + ** + ** to run without a statement journal if there are no indexes on the + ** table. + */ + Trigger *pTrigger = 0; + if( db->flags&SQLITE_RecTriggers ){ + pTrigger = sqlite3TriggersExist(pParse, pTab, TK_DELETE, 0, 0); + } + if( pTrigger || sqlite3FkRequired(pParse, pTab, 0, 0) ){ + sqlite3MultiWrite(pParse); + sqlite3GenerateRowDelete(pParse, pTab, pTrigger, iDataCur, iIdxCur, + regNewData, 1, 0, OE_Replace, 1, -1); + }else{ +#ifdef SQLITE_ENABLE_PREUPDATE_HOOK + assert( HasRowid(pTab) ); + /* This OP_Delete opcode fires the pre-update-hook only. It does + ** not modify the b-tree. It is more efficient to let the coming + ** OP_Insert replace the existing entry than it is to delete the + ** existing entry and then insert a new one. */ + sqlite3VdbeAddOp2(v, OP_Delete, iDataCur, OPFLAG_ISNOOP); + sqlite3VdbeAppendP4(v, pTab, P4_TABLE); +#endif /* SQLITE_ENABLE_PREUPDATE_HOOK */ + if( pTab->pIndex ){ + sqlite3MultiWrite(pParse); + sqlite3GenerateRowIndexDelete(pParse, pTab, iDataCur, iIdxCur,0,-1); + } + } + seenReplace = 1; + break; + } +#ifndef SQLITE_OMIT_UPSERT + case OE_Update: { + sqlite3UpsertDoUpdate(pParse, pUpsert, pTab, 0, iDataCur); + /* Fall through */ + } +#endif + case OE_Ignore: { + testcase( onError==OE_Ignore ); + sqlite3VdbeGoto(v, ignoreDest); + break; + } + } + sqlite3VdbeResolveLabel(v, addrRowidOk); + if( ipkTop ){ + ipkBottom = sqlite3VdbeAddOp0(v, OP_Goto); + sqlite3VdbeJumpHere(v, ipkTop-1); + } + } + + /* Test all UNIQUE constraints by creating entries for each UNIQUE + ** index and making sure that duplicate entries do not already exist. + ** Compute the revised record entries for indices as we go. + ** + ** This loop also handles the case of the PRIMARY KEY index for a + ** WITHOUT ROWID table. + */ + for(ix=0, pIdx=pTab->pIndex; pIdx; pIdx=pIdx->pNext, ix++){ + int regIdx; /* Range of registers hold conent for pIdx */ + int regR; /* Range of registers holding conflicting PK */ + int iThisCur; /* Cursor for this UNIQUE index */ + int addrUniqueOk; /* Jump here if the UNIQUE constraint is satisfied */ + + if( aRegIdx[ix]==0 ) continue; /* Skip indices that do not change */ + if( pUpIdx==pIdx ){ + addrUniqueOk = upsertJump+1; + upsertBypass = sqlite3VdbeGoto(v, 0); + VdbeComment((v, "Skip upsert subroutine")); + sqlite3VdbeJumpHere(v, upsertJump); + }else{ + addrUniqueOk = sqlite3VdbeMakeLabel(pParse); + } + if( bAffinityDone==0 && (pUpIdx==0 || pUpIdx==pIdx) ){ + sqlite3TableAffinity(v, pTab, regNewData+1); + bAffinityDone = 1; + } + VdbeNoopComment((v, "uniqueness check for %s", pIdx->zName)); + iThisCur = iIdxCur+ix; + + + /* Skip partial indices for which the WHERE clause is not true */ + if( pIdx->pPartIdxWhere ){ + sqlite3VdbeAddOp2(v, OP_Null, 0, aRegIdx[ix]); + pParse->iSelfTab = -(regNewData+1); + sqlite3ExprIfFalseDup(pParse, pIdx->pPartIdxWhere, addrUniqueOk, + SQLITE_JUMPIFNULL); + pParse->iSelfTab = 0; + } + + /* Create a record for this index entry as it should appear after + ** the insert or update. Store that record in the aRegIdx[ix] register + */ + regIdx = aRegIdx[ix]+1; + for(i=0; inColumn; i++){ + int iField = pIdx->aiColumn[i]; + int x; + if( iField==XN_EXPR ){ + pParse->iSelfTab = -(regNewData+1); + sqlite3ExprCodeCopy(pParse, pIdx->aColExpr->a[i].pExpr, regIdx+i); + pParse->iSelfTab = 0; + VdbeComment((v, "%s column %d", pIdx->zName, i)); + }else{ + if( iField==XN_ROWID || iField==pTab->iPKey ){ + x = regNewData; + }else{ + x = iField + regNewData + 1; + } + sqlite3VdbeAddOp2(v, iField<0 ? OP_IntCopy : OP_SCopy, x, regIdx+i); + VdbeComment((v, "%s", iField<0 ? "rowid" : pTab->aCol[iField].zName)); + } + } + sqlite3VdbeAddOp3(v, OP_MakeRecord, regIdx, pIdx->nColumn, aRegIdx[ix]); + VdbeComment((v, "for %s", pIdx->zName)); +#ifdef SQLITE_ENABLE_NULL_TRIM + if( pIdx->idxType==SQLITE_IDXTYPE_PRIMARYKEY ){ + sqlite3SetMakeRecordP5(v, pIdx->pTable); + } +#endif + + /* In an UPDATE operation, if this index is the PRIMARY KEY index + ** of a WITHOUT ROWID table and there has been no change the + ** primary key, then no collision is possible. The collision detection + ** logic below can all be skipped. */ + if( isUpdate && pPk==pIdx && pkChng==0 ){ + sqlite3VdbeResolveLabel(v, addrUniqueOk); + continue; + } + + /* Find out what action to take in case there is a uniqueness conflict */ + onError = pIdx->onError; + if( onError==OE_None ){ + sqlite3VdbeResolveLabel(v, addrUniqueOk); + continue; /* pIdx is not a UNIQUE index */ + } + if( overrideError!=OE_Default ){ + onError = overrideError; + }else if( onError==OE_Default ){ + onError = OE_Abort; + } + + /* Figure out if the upsert clause applies to this index */ + if( pUpIdx==pIdx ){ + if( pUpsert->pUpsertSet==0 ){ + onError = OE_Ignore; /* DO NOTHING is the same as INSERT OR IGNORE */ + }else{ + onError = OE_Update; /* DO UPDATE */ + } + } + + /* Collision detection may be omitted if all of the following are true: + ** (1) The conflict resolution algorithm is REPLACE + ** (2) The table is a WITHOUT ROWID table + ** (3) There are no secondary indexes on the table + ** (4) No delete triggers need to be fired if there is a conflict + ** (5) No FK constraint counters need to be updated if a conflict occurs. + ** + ** This is not possible for ENABLE_PREUPDATE_HOOK builds, as the row + ** must be explicitly deleted in order to ensure any pre-update hook + ** is invoked. */ +#ifndef SQLITE_ENABLE_PREUPDATE_HOOK + if( (ix==0 && pIdx->pNext==0) /* Condition 3 */ + && pPk==pIdx /* Condition 2 */ + && onError==OE_Replace /* Condition 1 */ + && ( 0==(db->flags&SQLITE_RecTriggers) || /* Condition 4 */ + 0==sqlite3TriggersExist(pParse, pTab, TK_DELETE, 0, 0)) + && ( 0==(db->flags&SQLITE_ForeignKeys) || /* Condition 5 */ + (0==pTab->pFKey && 0==sqlite3FkReferences(pTab))) + ){ + sqlite3VdbeResolveLabel(v, addrUniqueOk); + continue; + } +#endif /* ifndef SQLITE_ENABLE_PREUPDATE_HOOK */ + + /* Check to see if the new index entry will be unique */ + sqlite3VdbeVerifyAbortable(v, onError); + sqlite3VdbeAddOp4Int(v, OP_NoConflict, iThisCur, addrUniqueOk, + regIdx, pIdx->nKeyCol); VdbeCoverage(v); + + /* Generate code to handle collisions */ + regR = (pIdx==pPk) ? regIdx : sqlite3GetTempRange(pParse, nPkField); + if( isUpdate || onError==OE_Replace ){ + if( HasRowid(pTab) ){ + sqlite3VdbeAddOp2(v, OP_IdxRowid, iThisCur, regR); + /* Conflict only if the rowid of the existing index entry + ** is different from old-rowid */ + if( isUpdate ){ + sqlite3VdbeAddOp3(v, OP_Eq, regR, addrUniqueOk, regOldData); + sqlite3VdbeChangeP5(v, SQLITE_NOTNULL); + VdbeCoverage(v); + } + }else{ + int x; + /* Extract the PRIMARY KEY from the end of the index entry and + ** store it in registers regR..regR+nPk-1 */ + if( pIdx!=pPk ){ + for(i=0; inKeyCol; i++){ + assert( pPk->aiColumn[i]>=0 ); + x = sqlite3ColumnOfIndex(pIdx, pPk->aiColumn[i]); + sqlite3VdbeAddOp3(v, OP_Column, iThisCur, x, regR+i); + VdbeComment((v, "%s.%s", pTab->zName, + pTab->aCol[pPk->aiColumn[i]].zName)); + } + } + if( isUpdate ){ + /* If currently processing the PRIMARY KEY of a WITHOUT ROWID + ** table, only conflict if the new PRIMARY KEY values are actually + ** different from the old. + ** + ** For a UNIQUE index, only conflict if the PRIMARY KEY values + ** of the matched index row are different from the original PRIMARY + ** KEY values of this row before the update. */ + int addrJump = sqlite3VdbeCurrentAddr(v)+pPk->nKeyCol; + int op = OP_Ne; + int regCmp = (IsPrimaryKeyIndex(pIdx) ? regIdx : regR); + + for(i=0; inKeyCol; i++){ + char *p4 = (char*)sqlite3LocateCollSeq(pParse, pPk->azColl[i]); + x = pPk->aiColumn[i]; + assert( x>=0 ); + if( i==(pPk->nKeyCol-1) ){ + addrJump = addrUniqueOk; + op = OP_Eq; + } + sqlite3VdbeAddOp4(v, op, + regOldData+1+x, addrJump, regCmp+i, p4, P4_COLLSEQ + ); + sqlite3VdbeChangeP5(v, SQLITE_NOTNULL); + VdbeCoverageIf(v, op==OP_Eq); + VdbeCoverageIf(v, op==OP_Ne); + } + } + } + } + + /* Generate code that executes if the new index entry is not unique */ + assert( onError==OE_Rollback || onError==OE_Abort || onError==OE_Fail + || onError==OE_Ignore || onError==OE_Replace || onError==OE_Update ); + switch( onError ){ + case OE_Rollback: + case OE_Abort: + case OE_Fail: { + testcase( onError==OE_Rollback ); + testcase( onError==OE_Abort ); + testcase( onError==OE_Fail ); + sqlite3UniqueConstraint(pParse, onError, pIdx); + break; + } +#ifndef SQLITE_OMIT_UPSERT + case OE_Update: { + sqlite3UpsertDoUpdate(pParse, pUpsert, pTab, pIdx, iIdxCur+ix); + /* Fall through */ + } +#endif + case OE_Ignore: { + testcase( onError==OE_Ignore ); + sqlite3VdbeGoto(v, ignoreDest); + break; + } + default: { + Trigger *pTrigger = 0; + assert( onError==OE_Replace ); + if( db->flags&SQLITE_RecTriggers ){ + pTrigger = sqlite3TriggersExist(pParse, pTab, TK_DELETE, 0, 0); + } + if( pTrigger || sqlite3FkRequired(pParse, pTab, 0, 0) ){ + sqlite3MultiWrite(pParse); + } + sqlite3GenerateRowDelete(pParse, pTab, pTrigger, iDataCur, iIdxCur, + regR, nPkField, 0, OE_Replace, + (pIdx==pPk ? ONEPASS_SINGLE : ONEPASS_OFF), iThisCur); + seenReplace = 1; + break; + } + } + if( pUpIdx==pIdx ){ + sqlite3VdbeGoto(v, upsertJump+1); + sqlite3VdbeJumpHere(v, upsertBypass); + }else{ + sqlite3VdbeResolveLabel(v, addrUniqueOk); + } + if( regR!=regIdx ) sqlite3ReleaseTempRange(pParse, regR, nPkField); + } + + /* If the IPK constraint is a REPLACE, run it last */ + if( ipkTop ){ + sqlite3VdbeGoto(v, ipkTop); + VdbeComment((v, "Do IPK REPLACE")); + sqlite3VdbeJumpHere(v, ipkBottom); + } + + /* Generate the table record */ + if( HasRowid(pTab) ){ + int regRec = aRegIdx[ix]; + sqlite3VdbeAddOp3(v, OP_MakeRecord, regNewData+1, pTab->nCol, regRec); + sqlite3SetMakeRecordP5(v, pTab); + if( !bAffinityDone ){ + sqlite3TableAffinity(v, pTab, 0); + } + } + + *pbMayReplace = seenReplace; + VdbeModuleComment((v, "END: GenCnstCks(%d)", seenReplace)); +} + +#ifdef SQLITE_ENABLE_NULL_TRIM +/* +** Change the P5 operand on the last opcode (which should be an OP_MakeRecord) +** to be the number of columns in table pTab that must not be NULL-trimmed. +** +** Or if no columns of pTab may be NULL-trimmed, leave P5 at zero. +*/ +SQLITE_PRIVATE void sqlite3SetMakeRecordP5(Vdbe *v, Table *pTab){ + u16 i; + + /* Records with omitted columns are only allowed for schema format + ** version 2 and later (SQLite version 3.1.4, 2005-02-20). */ + if( pTab->pSchema->file_format<2 ) return; + + for(i=pTab->nCol-1; i>0; i--){ + if( pTab->aCol[i].pDflt!=0 ) break; + if( pTab->aCol[i].colFlags & COLFLAG_PRIMKEY ) break; + } + sqlite3VdbeChangeP5(v, i+1); +} +#endif + +/* +** This routine generates code to finish the INSERT or UPDATE operation +** that was started by a prior call to sqlite3GenerateConstraintChecks. +** A consecutive range of registers starting at regNewData contains the +** rowid and the content to be inserted. +** +** The arguments to this routine should be the same as the first six +** arguments to sqlite3GenerateConstraintChecks. +*/ +SQLITE_PRIVATE void sqlite3CompleteInsertion( + Parse *pParse, /* The parser context */ + Table *pTab, /* the table into which we are inserting */ + int iDataCur, /* Cursor of the canonical data source */ + int iIdxCur, /* First index cursor */ + int regNewData, /* Range of content */ + int *aRegIdx, /* Register used by each index. 0 for unused indices */ + int update_flags, /* True for UPDATE, False for INSERT */ + int appendBias, /* True if this is likely to be an append */ + int useSeekResult /* True to set the USESEEKRESULT flag on OP_[Idx]Insert */ +){ + Vdbe *v; /* Prepared statements under construction */ + Index *pIdx; /* An index being inserted or updated */ + u8 pik_flags; /* flag values passed to the btree insert */ + int i; /* Loop counter */ + + assert( update_flags==0 + || update_flags==OPFLAG_ISUPDATE + || update_flags==(OPFLAG_ISUPDATE|OPFLAG_SAVEPOSITION) + ); + + v = sqlite3GetVdbe(pParse); + assert( v!=0 ); + assert( pTab->pSelect==0 ); /* This table is not a VIEW */ + for(i=0, pIdx=pTab->pIndex; pIdx; pIdx=pIdx->pNext, i++){ + if( aRegIdx[i]==0 ) continue; + if( pIdx->pPartIdxWhere ){ + sqlite3VdbeAddOp2(v, OP_IsNull, aRegIdx[i], sqlite3VdbeCurrentAddr(v)+2); + VdbeCoverage(v); + } + pik_flags = (useSeekResult ? OPFLAG_USESEEKRESULT : 0); + if( IsPrimaryKeyIndex(pIdx) && !HasRowid(pTab) ){ + assert( pParse->nested==0 ); + pik_flags |= OPFLAG_NCHANGE; + pik_flags |= (update_flags & OPFLAG_SAVEPOSITION); +#ifdef SQLITE_ENABLE_PREUPDATE_HOOK + if( update_flags==0 ){ + int r = sqlite3GetTempReg(pParse); + sqlite3VdbeAddOp2(v, OP_Integer, 0, r); + sqlite3VdbeAddOp4(v, OP_Insert, + iIdxCur+i, aRegIdx[i], r, (char*)pTab, P4_TABLE + ); + sqlite3VdbeChangeP5(v, OPFLAG_ISNOOP); + sqlite3ReleaseTempReg(pParse, r); + } +#endif + } + sqlite3VdbeAddOp4Int(v, OP_IdxInsert, iIdxCur+i, aRegIdx[i], + aRegIdx[i]+1, + pIdx->uniqNotNull ? pIdx->nKeyCol: pIdx->nColumn); + sqlite3VdbeChangeP5(v, pik_flags); + } + if( !HasRowid(pTab) ) return; + if( pParse->nested ){ + pik_flags = 0; + }else{ + pik_flags = OPFLAG_NCHANGE; + pik_flags |= (update_flags?update_flags:OPFLAG_LASTROWID); + } + if( appendBias ){ + pik_flags |= OPFLAG_APPEND; + } + if( useSeekResult ){ + pik_flags |= OPFLAG_USESEEKRESULT; + } + sqlite3VdbeAddOp3(v, OP_Insert, iDataCur, aRegIdx[i], regNewData); + if( !pParse->nested ){ + sqlite3VdbeAppendP4(v, pTab, P4_TABLE); + } + sqlite3VdbeChangeP5(v, pik_flags); +} + +/* +** Allocate cursors for the pTab table and all its indices and generate +** code to open and initialized those cursors. +** +** The cursor for the object that contains the complete data (normally +** the table itself, but the PRIMARY KEY index in the case of a WITHOUT +** ROWID table) is returned in *piDataCur. The first index cursor is +** returned in *piIdxCur. The number of indices is returned. +** +** Use iBase as the first cursor (either the *piDataCur for rowid tables +** or the first index for WITHOUT ROWID tables) if it is non-negative. +** If iBase is negative, then allocate the next available cursor. +** +** For a rowid table, *piDataCur will be exactly one less than *piIdxCur. +** For a WITHOUT ROWID table, *piDataCur will be somewhere in the range +** of *piIdxCurs, depending on where the PRIMARY KEY index appears on the +** pTab->pIndex list. +** +** If pTab is a virtual table, then this routine is a no-op and the +** *piDataCur and *piIdxCur values are left uninitialized. +*/ +SQLITE_PRIVATE int sqlite3OpenTableAndIndices( + Parse *pParse, /* Parsing context */ + Table *pTab, /* Table to be opened */ + int op, /* OP_OpenRead or OP_OpenWrite */ + u8 p5, /* P5 value for OP_Open* opcodes (except on WITHOUT ROWID) */ + int iBase, /* Use this for the table cursor, if there is one */ + u8 *aToOpen, /* If not NULL: boolean for each table and index */ + int *piDataCur, /* Write the database source cursor number here */ + int *piIdxCur /* Write the first index cursor number here */ +){ + int i; + int iDb; + int iDataCur; + Index *pIdx; + Vdbe *v; + + assert( op==OP_OpenRead || op==OP_OpenWrite ); + assert( op==OP_OpenWrite || p5==0 ); + if( IsVirtual(pTab) ){ + /* This routine is a no-op for virtual tables. Leave the output + ** variables *piDataCur and *piIdxCur uninitialized so that valgrind + ** can detect if they are used by mistake in the caller. */ + return 0; + } + iDb = sqlite3SchemaToIndex(pParse->db, pTab->pSchema); + v = sqlite3GetVdbe(pParse); + assert( v!=0 ); + if( iBase<0 ) iBase = pParse->nTab; + iDataCur = iBase++; + if( piDataCur ) *piDataCur = iDataCur; + if( HasRowid(pTab) && (aToOpen==0 || aToOpen[0]) ){ + sqlite3OpenTable(pParse, iDataCur, iDb, pTab, op); + }else{ + sqlite3TableLock(pParse, iDb, pTab->tnum, op==OP_OpenWrite, pTab->zName); + } + if( piIdxCur ) *piIdxCur = iBase; + for(i=0, pIdx=pTab->pIndex; pIdx; pIdx=pIdx->pNext, i++){ + int iIdxCur = iBase++; + assert( pIdx->pSchema==pTab->pSchema ); + if( IsPrimaryKeyIndex(pIdx) && !HasRowid(pTab) ){ + if( piDataCur ) *piDataCur = iIdxCur; + p5 = 0; + } + if( aToOpen==0 || aToOpen[i+1] ){ + sqlite3VdbeAddOp3(v, op, iIdxCur, pIdx->tnum, iDb); + sqlite3VdbeSetP4KeyInfo(pParse, pIdx); + sqlite3VdbeChangeP5(v, p5); + VdbeComment((v, "%s", pIdx->zName)); + } + } + if( iBase>pParse->nTab ) pParse->nTab = iBase; + return i; +} + + +#ifdef SQLITE_TEST +/* +** The following global variable is incremented whenever the +** transfer optimization is used. This is used for testing +** purposes only - to make sure the transfer optimization really +** is happening when it is supposed to. +*/ +SQLITE_API int sqlite3_xferopt_count; +#endif /* SQLITE_TEST */ + + +#ifndef SQLITE_OMIT_XFER_OPT +/* +** Check to see if index pSrc is compatible as a source of data +** for index pDest in an insert transfer optimization. The rules +** for a compatible index: +** +** * The index is over the same set of columns +** * The same DESC and ASC markings occurs on all columns +** * The same onError processing (OE_Abort, OE_Ignore, etc) +** * The same collating sequence on each column +** * The index has the exact same WHERE clause +*/ +static int xferCompatibleIndex(Index *pDest, Index *pSrc){ + int i; + assert( pDest && pSrc ); + assert( pDest->pTable!=pSrc->pTable ); + if( pDest->nKeyCol!=pSrc->nKeyCol ){ + return 0; /* Different number of columns */ + } + if( pDest->onError!=pSrc->onError ){ + return 0; /* Different conflict resolution strategies */ + } + for(i=0; inKeyCol; i++){ + if( pSrc->aiColumn[i]!=pDest->aiColumn[i] ){ + return 0; /* Different columns indexed */ + } + if( pSrc->aiColumn[i]==XN_EXPR ){ + assert( pSrc->aColExpr!=0 && pDest->aColExpr!=0 ); + if( sqlite3ExprCompare(0, pSrc->aColExpr->a[i].pExpr, + pDest->aColExpr->a[i].pExpr, -1)!=0 ){ + return 0; /* Different expressions in the index */ + } + } + if( pSrc->aSortOrder[i]!=pDest->aSortOrder[i] ){ + return 0; /* Different sort orders */ + } + if( sqlite3_stricmp(pSrc->azColl[i],pDest->azColl[i])!=0 ){ + return 0; /* Different collating sequences */ + } + } + if( sqlite3ExprCompare(0, pSrc->pPartIdxWhere, pDest->pPartIdxWhere, -1) ){ + return 0; /* Different WHERE clauses */ + } + + /* If no test above fails then the indices must be compatible */ + return 1; +} + +/* +** Attempt the transfer optimization on INSERTs of the form +** +** INSERT INTO tab1 SELECT * FROM tab2; +** +** The xfer optimization transfers raw records from tab2 over to tab1. +** Columns are not decoded and reassembled, which greatly improves +** performance. Raw index records are transferred in the same way. +** +** The xfer optimization is only attempted if tab1 and tab2 are compatible. +** There are lots of rules for determining compatibility - see comments +** embedded in the code for details. +** +** This routine returns TRUE if the optimization is guaranteed to be used. +** Sometimes the xfer optimization will only work if the destination table +** is empty - a factor that can only be determined at run-time. In that +** case, this routine generates code for the xfer optimization but also +** does a test to see if the destination table is empty and jumps over the +** xfer optimization code if the test fails. In that case, this routine +** returns FALSE so that the caller will know to go ahead and generate +** an unoptimized transfer. This routine also returns FALSE if there +** is no chance that the xfer optimization can be applied. +** +** This optimization is particularly useful at making VACUUM run faster. +*/ +static int xferOptimization( + Parse *pParse, /* Parser context */ + Table *pDest, /* The table we are inserting into */ + Select *pSelect, /* A SELECT statement to use as the data source */ + int onError, /* How to handle constraint errors */ + int iDbDest /* The database of pDest */ +){ + sqlite3 *db = pParse->db; + ExprList *pEList; /* The result set of the SELECT */ + Table *pSrc; /* The table in the FROM clause of SELECT */ + Index *pSrcIdx, *pDestIdx; /* Source and destination indices */ + struct SrcList_item *pItem; /* An element of pSelect->pSrc */ + int i; /* Loop counter */ + int iDbSrc; /* The database of pSrc */ + int iSrc, iDest; /* Cursors from source and destination */ + int addr1, addr2; /* Loop addresses */ + int emptyDestTest = 0; /* Address of test for empty pDest */ + int emptySrcTest = 0; /* Address of test for empty pSrc */ + Vdbe *v; /* The VDBE we are building */ + int regAutoinc; /* Memory register used by AUTOINC */ + int destHasUniqueIdx = 0; /* True if pDest has a UNIQUE index */ + int regData, regRowid; /* Registers holding data and rowid */ + + if( pSelect==0 ){ + return 0; /* Must be of the form INSERT INTO ... SELECT ... */ + } + if( pParse->pWith || pSelect->pWith ){ + /* Do not attempt to process this query if there are an WITH clauses + ** attached to it. Proceeding may generate a false "no such table: xxx" + ** error if pSelect reads from a CTE named "xxx". */ + return 0; + } + if( sqlite3TriggerList(pParse, pDest) ){ + return 0; /* tab1 must not have triggers */ + } +#ifndef SQLITE_OMIT_VIRTUALTABLE + if( IsVirtual(pDest) ){ + return 0; /* tab1 must not be a virtual table */ + } +#endif + if( onError==OE_Default ){ + if( pDest->iPKey>=0 ) onError = pDest->keyConf; + if( onError==OE_Default ) onError = OE_Abort; + } + assert(pSelect->pSrc); /* allocated even if there is no FROM clause */ + if( pSelect->pSrc->nSrc!=1 ){ + return 0; /* FROM clause must have exactly one term */ + } + if( pSelect->pSrc->a[0].pSelect ){ + return 0; /* FROM clause cannot contain a subquery */ + } + if( pSelect->pWhere ){ + return 0; /* SELECT may not have a WHERE clause */ + } + if( pSelect->pOrderBy ){ + return 0; /* SELECT may not have an ORDER BY clause */ + } + /* Do not need to test for a HAVING clause. If HAVING is present but + ** there is no ORDER BY, we will get an error. */ + if( pSelect->pGroupBy ){ + return 0; /* SELECT may not have a GROUP BY clause */ + } + if( pSelect->pLimit ){ + return 0; /* SELECT may not have a LIMIT clause */ + } + if( pSelect->pPrior ){ + return 0; /* SELECT may not be a compound query */ + } + if( pSelect->selFlags & SF_Distinct ){ + return 0; /* SELECT may not be DISTINCT */ + } + pEList = pSelect->pEList; + assert( pEList!=0 ); + if( pEList->nExpr!=1 ){ + return 0; /* The result set must have exactly one column */ + } + assert( pEList->a[0].pExpr ); + if( pEList->a[0].pExpr->op!=TK_ASTERISK ){ + return 0; /* The result set must be the special operator "*" */ + } + + /* At this point we have established that the statement is of the + ** correct syntactic form to participate in this optimization. Now + ** we have to check the semantics. + */ + pItem = pSelect->pSrc->a; + pSrc = sqlite3LocateTableItem(pParse, 0, pItem); + if( pSrc==0 ){ + return 0; /* FROM clause does not contain a real table */ + } + if( pSrc->tnum==pDest->tnum && pSrc->pSchema==pDest->pSchema ){ + testcase( pSrc!=pDest ); /* Possible due to bad sqlite_master.rootpage */ + return 0; /* tab1 and tab2 may not be the same table */ + } + if( HasRowid(pDest)!=HasRowid(pSrc) ){ + return 0; /* source and destination must both be WITHOUT ROWID or not */ + } +#ifndef SQLITE_OMIT_VIRTUALTABLE + if( IsVirtual(pSrc) ){ + return 0; /* tab2 must not be a virtual table */ + } +#endif + if( pSrc->pSelect ){ + return 0; /* tab2 may not be a view */ + } + if( pDest->nCol!=pSrc->nCol ){ + return 0; /* Number of columns must be the same in tab1 and tab2 */ + } + if( pDest->iPKey!=pSrc->iPKey ){ + return 0; /* Both tables must have the same INTEGER PRIMARY KEY */ + } + for(i=0; inCol; i++){ + Column *pDestCol = &pDest->aCol[i]; + Column *pSrcCol = &pSrc->aCol[i]; +#ifdef SQLITE_ENABLE_HIDDEN_COLUMNS + if( (db->mDbFlags & DBFLAG_Vacuum)==0 + && (pDestCol->colFlags | pSrcCol->colFlags) & COLFLAG_HIDDEN + ){ + return 0; /* Neither table may have __hidden__ columns */ + } +#endif + if( pDestCol->affinity!=pSrcCol->affinity ){ + return 0; /* Affinity must be the same on all columns */ + } + if( sqlite3_stricmp(pDestCol->zColl, pSrcCol->zColl)!=0 ){ + return 0; /* Collating sequence must be the same on all columns */ + } + if( pDestCol->notNull && !pSrcCol->notNull ){ + return 0; /* tab2 must be NOT NULL if tab1 is */ + } + /* Default values for second and subsequent columns need to match. */ + if( i>0 ){ + assert( pDestCol->pDflt==0 || pDestCol->pDflt->op==TK_SPAN ); + assert( pSrcCol->pDflt==0 || pSrcCol->pDflt->op==TK_SPAN ); + if( (pDestCol->pDflt==0)!=(pSrcCol->pDflt==0) + || (pDestCol->pDflt && strcmp(pDestCol->pDflt->u.zToken, + pSrcCol->pDflt->u.zToken)!=0) + ){ + return 0; /* Default values must be the same for all columns */ + } + } + } + for(pDestIdx=pDest->pIndex; pDestIdx; pDestIdx=pDestIdx->pNext){ + if( IsUniqueIndex(pDestIdx) ){ + destHasUniqueIdx = 1; + } + for(pSrcIdx=pSrc->pIndex; pSrcIdx; pSrcIdx=pSrcIdx->pNext){ + if( xferCompatibleIndex(pDestIdx, pSrcIdx) ) break; + } + if( pSrcIdx==0 ){ + return 0; /* pDestIdx has no corresponding index in pSrc */ + } + if( pSrcIdx->tnum==pDestIdx->tnum && pSrc->pSchema==pDest->pSchema + && sqlite3FaultSim(411)==SQLITE_OK ){ + /* The sqlite3FaultSim() call allows this corruption test to be + ** bypassed during testing, in order to exercise other corruption tests + ** further downstream. */ + return 0; /* Corrupt schema - two indexes on the same btree */ + } + } +#ifndef SQLITE_OMIT_CHECK + if( pDest->pCheck && sqlite3ExprListCompare(pSrc->pCheck,pDest->pCheck,-1) ){ + return 0; /* Tables have different CHECK constraints. Ticket #2252 */ + } +#endif +#ifndef SQLITE_OMIT_FOREIGN_KEY + /* Disallow the transfer optimization if the destination table constains + ** any foreign key constraints. This is more restrictive than necessary. + ** But the main beneficiary of the transfer optimization is the VACUUM + ** command, and the VACUUM command disables foreign key constraints. So + ** the extra complication to make this rule less restrictive is probably + ** not worth the effort. Ticket [6284df89debdfa61db8073e062908af0c9b6118e] + */ + if( (db->flags & SQLITE_ForeignKeys)!=0 && pDest->pFKey!=0 ){ + return 0; + } +#endif + if( (db->flags & SQLITE_CountRows)!=0 ){ + return 0; /* xfer opt does not play well with PRAGMA count_changes */ + } + + /* If we get this far, it means that the xfer optimization is at + ** least a possibility, though it might only work if the destination + ** table (tab1) is initially empty. + */ +#ifdef SQLITE_TEST + sqlite3_xferopt_count++; +#endif + iDbSrc = sqlite3SchemaToIndex(db, pSrc->pSchema); + v = sqlite3GetVdbe(pParse); + sqlite3CodeVerifySchema(pParse, iDbSrc); + iSrc = pParse->nTab++; + iDest = pParse->nTab++; + regAutoinc = autoIncBegin(pParse, iDbDest, pDest); + regData = sqlite3GetTempReg(pParse); + regRowid = sqlite3GetTempReg(pParse); + sqlite3OpenTable(pParse, iDest, iDbDest, pDest, OP_OpenWrite); + assert( HasRowid(pDest) || destHasUniqueIdx ); + if( (db->mDbFlags & DBFLAG_Vacuum)==0 && ( + (pDest->iPKey<0 && pDest->pIndex!=0) /* (1) */ + || destHasUniqueIdx /* (2) */ + || (onError!=OE_Abort && onError!=OE_Rollback) /* (3) */ + )){ + /* In some circumstances, we are able to run the xfer optimization + ** only if the destination table is initially empty. Unless the + ** DBFLAG_Vacuum flag is set, this block generates code to make + ** that determination. If DBFLAG_Vacuum is set, then the destination + ** table is always empty. + ** + ** Conditions under which the destination must be empty: + ** + ** (1) There is no INTEGER PRIMARY KEY but there are indices. + ** (If the destination is not initially empty, the rowid fields + ** of index entries might need to change.) + ** + ** (2) The destination has a unique index. (The xfer optimization + ** is unable to test uniqueness.) + ** + ** (3) onError is something other than OE_Abort and OE_Rollback. + */ + addr1 = sqlite3VdbeAddOp2(v, OP_Rewind, iDest, 0); VdbeCoverage(v); + emptyDestTest = sqlite3VdbeAddOp0(v, OP_Goto); + sqlite3VdbeJumpHere(v, addr1); + } + if( HasRowid(pSrc) ){ + u8 insFlags; + sqlite3OpenTable(pParse, iSrc, iDbSrc, pSrc, OP_OpenRead); + emptySrcTest = sqlite3VdbeAddOp2(v, OP_Rewind, iSrc, 0); VdbeCoverage(v); + if( pDest->iPKey>=0 ){ + addr1 = sqlite3VdbeAddOp2(v, OP_Rowid, iSrc, regRowid); + sqlite3VdbeVerifyAbortable(v, onError); + addr2 = sqlite3VdbeAddOp3(v, OP_NotExists, iDest, 0, regRowid); + VdbeCoverage(v); + sqlite3RowidConstraint(pParse, onError, pDest); + sqlite3VdbeJumpHere(v, addr2); + autoIncStep(pParse, regAutoinc, regRowid); + }else if( pDest->pIndex==0 && !(db->mDbFlags & DBFLAG_VacuumInto) ){ + addr1 = sqlite3VdbeAddOp2(v, OP_NewRowid, iDest, regRowid); + }else{ + addr1 = sqlite3VdbeAddOp2(v, OP_Rowid, iSrc, regRowid); + assert( (pDest->tabFlags & TF_Autoincrement)==0 ); + } + sqlite3VdbeAddOp3(v, OP_RowData, iSrc, regData, 1); + if( db->mDbFlags & DBFLAG_Vacuum ){ + sqlite3VdbeAddOp1(v, OP_SeekEnd, iDest); + insFlags = OPFLAG_NCHANGE|OPFLAG_LASTROWID| + OPFLAG_APPEND|OPFLAG_USESEEKRESULT; + }else{ + insFlags = OPFLAG_NCHANGE|OPFLAG_LASTROWID|OPFLAG_APPEND; + } + sqlite3VdbeAddOp4(v, OP_Insert, iDest, regData, regRowid, + (char*)pDest, P4_TABLE); + sqlite3VdbeChangeP5(v, insFlags); + sqlite3VdbeAddOp2(v, OP_Next, iSrc, addr1); VdbeCoverage(v); + sqlite3VdbeAddOp2(v, OP_Close, iSrc, 0); + sqlite3VdbeAddOp2(v, OP_Close, iDest, 0); + }else{ + sqlite3TableLock(pParse, iDbDest, pDest->tnum, 1, pDest->zName); + sqlite3TableLock(pParse, iDbSrc, pSrc->tnum, 0, pSrc->zName); + } + for(pDestIdx=pDest->pIndex; pDestIdx; pDestIdx=pDestIdx->pNext){ + u8 idxInsFlags = 0; + for(pSrcIdx=pSrc->pIndex; ALWAYS(pSrcIdx); pSrcIdx=pSrcIdx->pNext){ + if( xferCompatibleIndex(pDestIdx, pSrcIdx) ) break; + } + assert( pSrcIdx ); + sqlite3VdbeAddOp3(v, OP_OpenRead, iSrc, pSrcIdx->tnum, iDbSrc); + sqlite3VdbeSetP4KeyInfo(pParse, pSrcIdx); + VdbeComment((v, "%s", pSrcIdx->zName)); + sqlite3VdbeAddOp3(v, OP_OpenWrite, iDest, pDestIdx->tnum, iDbDest); + sqlite3VdbeSetP4KeyInfo(pParse, pDestIdx); + sqlite3VdbeChangeP5(v, OPFLAG_BULKCSR); + VdbeComment((v, "%s", pDestIdx->zName)); + addr1 = sqlite3VdbeAddOp2(v, OP_Rewind, iSrc, 0); VdbeCoverage(v); + sqlite3VdbeAddOp3(v, OP_RowData, iSrc, regData, 1); + if( db->mDbFlags & DBFLAG_Vacuum ){ + /* This INSERT command is part of a VACUUM operation, which guarantees + ** that the destination table is empty. If all indexed columns use + ** collation sequence BINARY, then it can also be assumed that the + ** index will be populated by inserting keys in strictly sorted + ** order. In this case, instead of seeking within the b-tree as part + ** of every OP_IdxInsert opcode, an OP_SeekEnd is added before the + ** OP_IdxInsert to seek to the point within the b-tree where each key + ** should be inserted. This is faster. + ** + ** If any of the indexed columns use a collation sequence other than + ** BINARY, this optimization is disabled. This is because the user + ** might change the definition of a collation sequence and then run + ** a VACUUM command. In that case keys may not be written in strictly + ** sorted order. */ + for(i=0; inColumn; i++){ + const char *zColl = pSrcIdx->azColl[i]; + if( sqlite3_stricmp(sqlite3StrBINARY, zColl) ) break; + } + if( i==pSrcIdx->nColumn ){ + idxInsFlags = OPFLAG_USESEEKRESULT; + sqlite3VdbeAddOp1(v, OP_SeekEnd, iDest); + } + } + if( !HasRowid(pSrc) && pDestIdx->idxType==SQLITE_IDXTYPE_PRIMARYKEY ){ + idxInsFlags |= OPFLAG_NCHANGE; + } + sqlite3VdbeAddOp2(v, OP_IdxInsert, iDest, regData); + sqlite3VdbeChangeP5(v, idxInsFlags|OPFLAG_APPEND); + sqlite3VdbeAddOp2(v, OP_Next, iSrc, addr1+1); VdbeCoverage(v); + sqlite3VdbeJumpHere(v, addr1); + sqlite3VdbeAddOp2(v, OP_Close, iSrc, 0); + sqlite3VdbeAddOp2(v, OP_Close, iDest, 0); + } + if( emptySrcTest ) sqlite3VdbeJumpHere(v, emptySrcTest); + sqlite3ReleaseTempReg(pParse, regRowid); + sqlite3ReleaseTempReg(pParse, regData); + if( emptyDestTest ){ + sqlite3AutoincrementEnd(pParse); + sqlite3VdbeAddOp2(v, OP_Halt, SQLITE_OK, 0); + sqlite3VdbeJumpHere(v, emptyDestTest); + sqlite3VdbeAddOp2(v, OP_Close, iDest, 0); + return 0; + }else{ + return 1; + } +} +#endif /* SQLITE_OMIT_XFER_OPT */ + +/************** End of insert.c **********************************************/ +/************** Begin file legacy.c ******************************************/ +/* +** 2001 September 15 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** Main file for the SQLite library. The routines in this file +** implement the programmer interface to the library. Routines in +** other files are for internal use by SQLite and should not be +** accessed by users of the library. +*/ + +/* #include "sqliteInt.h" */ + +/* +** Execute SQL code. Return one of the SQLITE_ success/failure +** codes. Also write an error message into memory obtained from +** malloc() and make *pzErrMsg point to that message. +** +** If the SQL is a query, then for each row in the query result +** the xCallback() function is called. pArg becomes the first +** argument to xCallback(). If xCallback=NULL then no callback +** is invoked, even for queries. +*/ +SQLITE_API int sqlite3_exec( + sqlite3 *db, /* The database on which the SQL executes */ + const char *zSql, /* The SQL to be executed */ + sqlite3_callback xCallback, /* Invoke this callback routine */ + void *pArg, /* First argument to xCallback() */ + char **pzErrMsg /* Write error messages here */ +){ + int rc = SQLITE_OK; /* Return code */ + const char *zLeftover; /* Tail of unprocessed SQL */ + sqlite3_stmt *pStmt = 0; /* The current SQL statement */ + char **azCols = 0; /* Names of result columns */ + int callbackIsInit; /* True if callback data is initialized */ + + if( !sqlite3SafetyCheckOk(db) ) return SQLITE_MISUSE_BKPT; + if( zSql==0 ) zSql = ""; + + sqlite3_mutex_enter(db->mutex); + sqlite3Error(db, SQLITE_OK); + while( rc==SQLITE_OK && zSql[0] ){ + int nCol = 0; + char **azVals = 0; + + pStmt = 0; + rc = sqlite3_prepare_v2(db, zSql, -1, &pStmt, &zLeftover); + assert( rc==SQLITE_OK || pStmt==0 ); + if( rc!=SQLITE_OK ){ + continue; + } + if( !pStmt ){ + /* this happens for a comment or white-space */ + zSql = zLeftover; + continue; + } + callbackIsInit = 0; + + while( 1 ){ + int i; + rc = sqlite3_step(pStmt); + + /* Invoke the callback function if required */ + if( xCallback && (SQLITE_ROW==rc || + (SQLITE_DONE==rc && !callbackIsInit + && db->flags&SQLITE_NullCallback)) ){ + if( !callbackIsInit ){ + nCol = sqlite3_column_count(pStmt); + azCols = sqlite3DbMallocRaw(db, (2*nCol+1)*sizeof(const char*)); + if( azCols==0 ){ + goto exec_out; + } + for(i=0; ierrMask)==rc ); + sqlite3_mutex_leave(db->mutex); + return rc; +} + +/************** End of legacy.c **********************************************/ +/************** Begin file loadext.c *****************************************/ +/* +** 2006 June 7 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** This file contains code used to dynamically load extensions into +** the SQLite library. +*/ + +#ifndef SQLITE_CORE + #define SQLITE_CORE 1 /* Disable the API redefinition in sqlite3ext.h */ +#endif +/************** Include sqlite3ext.h in the middle of loadext.c **************/ +/************** Begin file sqlite3ext.h **************************************/ +/* +** 2006 June 7 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** This header file defines the SQLite interface for use by +** shared libraries that want to be imported as extensions into +** an SQLite instance. Shared libraries that intend to be loaded +** as extensions by SQLite should #include this file instead of +** sqlite3.h. +*/ +#ifndef SQLITE3EXT_H +#define SQLITE3EXT_H +/* #include "sqlite3.h" */ + +/* +** The following structure holds pointers to all of the SQLite API +** routines. +** +** WARNING: In order to maintain backwards compatibility, add new +** interfaces to the end of this structure only. If you insert new +** interfaces in the middle of this structure, then older different +** versions of SQLite will not be able to load each other's shared +** libraries! +*/ +struct sqlite3_api_routines { + void * (*aggregate_context)(sqlite3_context*,int nBytes); + int (*aggregate_count)(sqlite3_context*); + int (*bind_blob)(sqlite3_stmt*,int,const void*,int n,void(*)(void*)); + int (*bind_double)(sqlite3_stmt*,int,double); + int (*bind_int)(sqlite3_stmt*,int,int); + int (*bind_int64)(sqlite3_stmt*,int,sqlite_int64); + int (*bind_null)(sqlite3_stmt*,int); + int (*bind_parameter_count)(sqlite3_stmt*); + int (*bind_parameter_index)(sqlite3_stmt*,const char*zName); + const char * (*bind_parameter_name)(sqlite3_stmt*,int); + int (*bind_text)(sqlite3_stmt*,int,const char*,int n,void(*)(void*)); + int (*bind_text16)(sqlite3_stmt*,int,const void*,int,void(*)(void*)); + int (*bind_value)(sqlite3_stmt*,int,const sqlite3_value*); + int (*busy_handler)(sqlite3*,int(*)(void*,int),void*); + int (*busy_timeout)(sqlite3*,int ms); + int (*changes)(sqlite3*); + int (*close)(sqlite3*); + int (*collation_needed)(sqlite3*,void*,void(*)(void*,sqlite3*, + int eTextRep,const char*)); + int (*collation_needed16)(sqlite3*,void*,void(*)(void*,sqlite3*, + int eTextRep,const void*)); + const void * (*column_blob)(sqlite3_stmt*,int iCol); + int (*column_bytes)(sqlite3_stmt*,int iCol); + int (*column_bytes16)(sqlite3_stmt*,int iCol); + int (*column_count)(sqlite3_stmt*pStmt); + const char * (*column_database_name)(sqlite3_stmt*,int); + const void * (*column_database_name16)(sqlite3_stmt*,int); + const char * (*column_decltype)(sqlite3_stmt*,int i); + const void * (*column_decltype16)(sqlite3_stmt*,int); + double (*column_double)(sqlite3_stmt*,int iCol); + int (*column_int)(sqlite3_stmt*,int iCol); + sqlite_int64 (*column_int64)(sqlite3_stmt*,int iCol); + const char * (*column_name)(sqlite3_stmt*,int); + const void * (*column_name16)(sqlite3_stmt*,int); + const char * (*column_origin_name)(sqlite3_stmt*,int); + const void * (*column_origin_name16)(sqlite3_stmt*,int); + const char * (*column_table_name)(sqlite3_stmt*,int); + const void * (*column_table_name16)(sqlite3_stmt*,int); + const unsigned char * (*column_text)(sqlite3_stmt*,int iCol); + const void * (*column_text16)(sqlite3_stmt*,int iCol); + int (*column_type)(sqlite3_stmt*,int iCol); + sqlite3_value* (*column_value)(sqlite3_stmt*,int iCol); + void * (*commit_hook)(sqlite3*,int(*)(void*),void*); + int (*complete)(const char*sql); + int (*complete16)(const void*sql); + int (*create_collation)(sqlite3*,const char*,int,void*, + int(*)(void*,int,const void*,int,const void*)); + int (*create_collation16)(sqlite3*,const void*,int,void*, + int(*)(void*,int,const void*,int,const void*)); + int (*create_function)(sqlite3*,const char*,int,int,void*, + void (*xFunc)(sqlite3_context*,int,sqlite3_value**), + void (*xStep)(sqlite3_context*,int,sqlite3_value**), + void (*xFinal)(sqlite3_context*)); + int (*create_function16)(sqlite3*,const void*,int,int,void*, + void (*xFunc)(sqlite3_context*,int,sqlite3_value**), + void (*xStep)(sqlite3_context*,int,sqlite3_value**), + void (*xFinal)(sqlite3_context*)); + int (*create_module)(sqlite3*,const char*,const sqlite3_module*,void*); + int (*data_count)(sqlite3_stmt*pStmt); + sqlite3 * (*db_handle)(sqlite3_stmt*); + int (*declare_vtab)(sqlite3*,const char*); + int (*enable_shared_cache)(int); + int (*errcode)(sqlite3*db); + const char * (*errmsg)(sqlite3*); + const void * (*errmsg16)(sqlite3*); + int (*exec)(sqlite3*,const char*,sqlite3_callback,void*,char**); + int (*expired)(sqlite3_stmt*); + int (*finalize)(sqlite3_stmt*pStmt); + void (*free)(void*); + void (*free_table)(char**result); + int (*get_autocommit)(sqlite3*); + void * (*get_auxdata)(sqlite3_context*,int); + int (*get_table)(sqlite3*,const char*,char***,int*,int*,char**); + int (*global_recover)(void); + void (*interruptx)(sqlite3*); + sqlite_int64 (*last_insert_rowid)(sqlite3*); + const char * (*libversion)(void); + int (*libversion_number)(void); + void *(*malloc)(int); + char * (*mprintf)(const char*,...); + int (*open)(const char*,sqlite3**); + int (*open16)(const void*,sqlite3**); + int (*prepare)(sqlite3*,const char*,int,sqlite3_stmt**,const char**); + int (*prepare16)(sqlite3*,const void*,int,sqlite3_stmt**,const void**); + void * (*profile)(sqlite3*,void(*)(void*,const char*,sqlite_uint64),void*); + void (*progress_handler)(sqlite3*,int,int(*)(void*),void*); + void *(*realloc)(void*,int); + int (*reset)(sqlite3_stmt*pStmt); + void (*result_blob)(sqlite3_context*,const void*,int,void(*)(void*)); + void (*result_double)(sqlite3_context*,double); + void (*result_error)(sqlite3_context*,const char*,int); + void (*result_error16)(sqlite3_context*,const void*,int); + void (*result_int)(sqlite3_context*,int); + void (*result_int64)(sqlite3_context*,sqlite_int64); + void (*result_null)(sqlite3_context*); + void (*result_text)(sqlite3_context*,const char*,int,void(*)(void*)); + void (*result_text16)(sqlite3_context*,const void*,int,void(*)(void*)); + void (*result_text16be)(sqlite3_context*,const void*,int,void(*)(void*)); + void (*result_text16le)(sqlite3_context*,const void*,int,void(*)(void*)); + void (*result_value)(sqlite3_context*,sqlite3_value*); + void * (*rollback_hook)(sqlite3*,void(*)(void*),void*); + int (*set_authorizer)(sqlite3*,int(*)(void*,int,const char*,const char*, + const char*,const char*),void*); + void (*set_auxdata)(sqlite3_context*,int,void*,void (*)(void*)); + char * (*xsnprintf)(int,char*,const char*,...); + int (*step)(sqlite3_stmt*); + int (*table_column_metadata)(sqlite3*,const char*,const char*,const char*, + char const**,char const**,int*,int*,int*); + void (*thread_cleanup)(void); + int (*total_changes)(sqlite3*); + void * (*trace)(sqlite3*,void(*xTrace)(void*,const char*),void*); + int (*transfer_bindings)(sqlite3_stmt*,sqlite3_stmt*); + void * (*update_hook)(sqlite3*,void(*)(void*,int ,char const*,char const*, + sqlite_int64),void*); + void * (*user_data)(sqlite3_context*); + const void * (*value_blob)(sqlite3_value*); + int (*value_bytes)(sqlite3_value*); + int (*value_bytes16)(sqlite3_value*); + double (*value_double)(sqlite3_value*); + int (*value_int)(sqlite3_value*); + sqlite_int64 (*value_int64)(sqlite3_value*); + int (*value_numeric_type)(sqlite3_value*); + const unsigned char * (*value_text)(sqlite3_value*); + const void * (*value_text16)(sqlite3_value*); + const void * (*value_text16be)(sqlite3_value*); + const void * (*value_text16le)(sqlite3_value*); + int (*value_type)(sqlite3_value*); + char *(*vmprintf)(const char*,va_list); + /* Added ??? */ + int (*overload_function)(sqlite3*, const char *zFuncName, int nArg); + /* Added by 3.3.13 */ + int (*prepare_v2)(sqlite3*,const char*,int,sqlite3_stmt**,const char**); + int (*prepare16_v2)(sqlite3*,const void*,int,sqlite3_stmt**,const void**); + int (*clear_bindings)(sqlite3_stmt*); + /* Added by 3.4.1 */ + int (*create_module_v2)(sqlite3*,const char*,const sqlite3_module*,void*, + void (*xDestroy)(void *)); + /* Added by 3.5.0 */ + int (*bind_zeroblob)(sqlite3_stmt*,int,int); + int (*blob_bytes)(sqlite3_blob*); + int (*blob_close)(sqlite3_blob*); + int (*blob_open)(sqlite3*,const char*,const char*,const char*,sqlite3_int64, + int,sqlite3_blob**); + int (*blob_read)(sqlite3_blob*,void*,int,int); + int (*blob_write)(sqlite3_blob*,const void*,int,int); + int (*create_collation_v2)(sqlite3*,const char*,int,void*, + int(*)(void*,int,const void*,int,const void*), + void(*)(void*)); + int (*file_control)(sqlite3*,const char*,int,void*); + sqlite3_int64 (*memory_highwater)(int); + sqlite3_int64 (*memory_used)(void); + sqlite3_mutex *(*mutex_alloc)(int); + void (*mutex_enter)(sqlite3_mutex*); + void (*mutex_free)(sqlite3_mutex*); + void (*mutex_leave)(sqlite3_mutex*); + int (*mutex_try)(sqlite3_mutex*); + int (*open_v2)(const char*,sqlite3**,int,const char*); + int (*release_memory)(int); + void (*result_error_nomem)(sqlite3_context*); + void (*result_error_toobig)(sqlite3_context*); + int (*sleep)(int); + void (*soft_heap_limit)(int); + sqlite3_vfs *(*vfs_find)(const char*); + int (*vfs_register)(sqlite3_vfs*,int); + int (*vfs_unregister)(sqlite3_vfs*); + int (*xthreadsafe)(void); + void (*result_zeroblob)(sqlite3_context*,int); + void (*result_error_code)(sqlite3_context*,int); + int (*test_control)(int, ...); + void (*randomness)(int,void*); + sqlite3 *(*context_db_handle)(sqlite3_context*); + int (*extended_result_codes)(sqlite3*,int); + int (*limit)(sqlite3*,int,int); + sqlite3_stmt *(*next_stmt)(sqlite3*,sqlite3_stmt*); + const char *(*sql)(sqlite3_stmt*); + int (*status)(int,int*,int*,int); + int (*backup_finish)(sqlite3_backup*); + sqlite3_backup *(*backup_init)(sqlite3*,const char*,sqlite3*,const char*); + int (*backup_pagecount)(sqlite3_backup*); + int (*backup_remaining)(sqlite3_backup*); + int (*backup_step)(sqlite3_backup*,int); + const char *(*compileoption_get)(int); + int (*compileoption_used)(const char*); + int (*create_function_v2)(sqlite3*,const char*,int,int,void*, + void (*xFunc)(sqlite3_context*,int,sqlite3_value**), + void (*xStep)(sqlite3_context*,int,sqlite3_value**), + void (*xFinal)(sqlite3_context*), + void(*xDestroy)(void*)); + int (*db_config)(sqlite3*,int,...); + sqlite3_mutex *(*db_mutex)(sqlite3*); + int (*db_status)(sqlite3*,int,int*,int*,int); + int (*extended_errcode)(sqlite3*); + void (*log)(int,const char*,...); + sqlite3_int64 (*soft_heap_limit64)(sqlite3_int64); + const char *(*sourceid)(void); + int (*stmt_status)(sqlite3_stmt*,int,int); + int (*strnicmp)(const char*,const char*,int); + int (*unlock_notify)(sqlite3*,void(*)(void**,int),void*); + int (*wal_autocheckpoint)(sqlite3*,int); + int (*wal_checkpoint)(sqlite3*,const char*); + void *(*wal_hook)(sqlite3*,int(*)(void*,sqlite3*,const char*,int),void*); + int (*blob_reopen)(sqlite3_blob*,sqlite3_int64); + int (*vtab_config)(sqlite3*,int op,...); + int (*vtab_on_conflict)(sqlite3*); + /* Version 3.7.16 and later */ + int (*close_v2)(sqlite3*); + const char *(*db_filename)(sqlite3*,const char*); + int (*db_readonly)(sqlite3*,const char*); + int (*db_release_memory)(sqlite3*); + const char *(*errstr)(int); + int (*stmt_busy)(sqlite3_stmt*); + int (*stmt_readonly)(sqlite3_stmt*); + int (*stricmp)(const char*,const char*); + int (*uri_boolean)(const char*,const char*,int); + sqlite3_int64 (*uri_int64)(const char*,const char*,sqlite3_int64); + const char *(*uri_parameter)(const char*,const char*); + char *(*xvsnprintf)(int,char*,const char*,va_list); + int (*wal_checkpoint_v2)(sqlite3*,const char*,int,int*,int*); + /* Version 3.8.7 and later */ + int (*auto_extension)(void(*)(void)); + int (*bind_blob64)(sqlite3_stmt*,int,const void*,sqlite3_uint64, + void(*)(void*)); + int (*bind_text64)(sqlite3_stmt*,int,const char*,sqlite3_uint64, + void(*)(void*),unsigned char); + int (*cancel_auto_extension)(void(*)(void)); + int (*load_extension)(sqlite3*,const char*,const char*,char**); + void *(*malloc64)(sqlite3_uint64); + sqlite3_uint64 (*msize)(void*); + void *(*realloc64)(void*,sqlite3_uint64); + void (*reset_auto_extension)(void); + void (*result_blob64)(sqlite3_context*,const void*,sqlite3_uint64, + void(*)(void*)); + void (*result_text64)(sqlite3_context*,const char*,sqlite3_uint64, + void(*)(void*), unsigned char); + int (*strglob)(const char*,const char*); + /* Version 3.8.11 and later */ + sqlite3_value *(*value_dup)(const sqlite3_value*); + void (*value_free)(sqlite3_value*); + int (*result_zeroblob64)(sqlite3_context*,sqlite3_uint64); + int (*bind_zeroblob64)(sqlite3_stmt*, int, sqlite3_uint64); + /* Version 3.9.0 and later */ + unsigned int (*value_subtype)(sqlite3_value*); + void (*result_subtype)(sqlite3_context*,unsigned int); + /* Version 3.10.0 and later */ + int (*status64)(int,sqlite3_int64*,sqlite3_int64*,int); + int (*strlike)(const char*,const char*,unsigned int); + int (*db_cacheflush)(sqlite3*); + /* Version 3.12.0 and later */ + int (*system_errno)(sqlite3*); + /* Version 3.14.0 and later */ + int (*trace_v2)(sqlite3*,unsigned,int(*)(unsigned,void*,void*,void*),void*); + char *(*expanded_sql)(sqlite3_stmt*); + /* Version 3.18.0 and later */ + void (*set_last_insert_rowid)(sqlite3*,sqlite3_int64); + /* Version 3.20.0 and later */ + int (*prepare_v3)(sqlite3*,const char*,int,unsigned int, + sqlite3_stmt**,const char**); + int (*prepare16_v3)(sqlite3*,const void*,int,unsigned int, + sqlite3_stmt**,const void**); + int (*bind_pointer)(sqlite3_stmt*,int,void*,const char*,void(*)(void*)); + void (*result_pointer)(sqlite3_context*,void*,const char*,void(*)(void*)); + void *(*value_pointer)(sqlite3_value*,const char*); + int (*vtab_nochange)(sqlite3_context*); + int (*value_nochange)(sqlite3_value*); + const char *(*vtab_collation)(sqlite3_index_info*,int); + /* Version 3.24.0 and later */ + int (*keyword_count)(void); + int (*keyword_name)(int,const char**,int*); + int (*keyword_check)(const char*,int); + sqlite3_str *(*str_new)(sqlite3*); + char *(*str_finish)(sqlite3_str*); + void (*str_appendf)(sqlite3_str*, const char *zFormat, ...); + void (*str_vappendf)(sqlite3_str*, const char *zFormat, va_list); + void (*str_append)(sqlite3_str*, const char *zIn, int N); + void (*str_appendall)(sqlite3_str*, const char *zIn); + void (*str_appendchar)(sqlite3_str*, int N, char C); + void (*str_reset)(sqlite3_str*); + int (*str_errcode)(sqlite3_str*); + int (*str_length)(sqlite3_str*); + char *(*str_value)(sqlite3_str*); + /* Version 3.25.0 and later */ + int (*create_window_function)(sqlite3*,const char*,int,int,void*, + void (*xStep)(sqlite3_context*,int,sqlite3_value**), + void (*xFinal)(sqlite3_context*), + void (*xValue)(sqlite3_context*), + void (*xInv)(sqlite3_context*,int,sqlite3_value**), + void(*xDestroy)(void*)); + /* Version 3.26.0 and later */ + const char *(*normalized_sql)(sqlite3_stmt*); + /* Version 3.28.0 and later */ + int (*stmt_isexplain)(sqlite3_stmt*); + int (*value_frombind)(sqlite3_value*); + /* Version 3.30.0 and later */ + int (*drop_modules)(sqlite3*,const char**); +}; + +/* +** This is the function signature used for all extension entry points. It +** is also defined in the file "loadext.c". +*/ +typedef int (*sqlite3_loadext_entry)( + sqlite3 *db, /* Handle to the database. */ + char **pzErrMsg, /* Used to set error string on failure. */ + const sqlite3_api_routines *pThunk /* Extension API function pointers. */ +); + +/* +** The following macros redefine the API routines so that they are +** redirected through the global sqlite3_api structure. +** +** This header file is also used by the loadext.c source file +** (part of the main SQLite library - not an extension) so that +** it can get access to the sqlite3_api_routines structure +** definition. But the main library does not want to redefine +** the API. So the redefinition macros are only valid if the +** SQLITE_CORE macros is undefined. +*/ +#if !defined(SQLITE_CORE) && !defined(SQLITE_OMIT_LOAD_EXTENSION) +#define sqlite3_aggregate_context sqlite3_api->aggregate_context +#ifndef SQLITE_OMIT_DEPRECATED +#define sqlite3_aggregate_count sqlite3_api->aggregate_count +#endif +#define sqlite3_bind_blob sqlite3_api->bind_blob +#define sqlite3_bind_double sqlite3_api->bind_double +#define sqlite3_bind_int sqlite3_api->bind_int +#define sqlite3_bind_int64 sqlite3_api->bind_int64 +#define sqlite3_bind_null sqlite3_api->bind_null +#define sqlite3_bind_parameter_count sqlite3_api->bind_parameter_count +#define sqlite3_bind_parameter_index sqlite3_api->bind_parameter_index +#define sqlite3_bind_parameter_name sqlite3_api->bind_parameter_name +#define sqlite3_bind_text sqlite3_api->bind_text +#define sqlite3_bind_text16 sqlite3_api->bind_text16 +#define sqlite3_bind_value sqlite3_api->bind_value +#define sqlite3_busy_handler sqlite3_api->busy_handler +#define sqlite3_busy_timeout sqlite3_api->busy_timeout +#define sqlite3_changes sqlite3_api->changes +#define sqlite3_close sqlite3_api->close +#define sqlite3_collation_needed sqlite3_api->collation_needed +#define sqlite3_collation_needed16 sqlite3_api->collation_needed16 +#define sqlite3_column_blob sqlite3_api->column_blob +#define sqlite3_column_bytes sqlite3_api->column_bytes +#define sqlite3_column_bytes16 sqlite3_api->column_bytes16 +#define sqlite3_column_count sqlite3_api->column_count +#define sqlite3_column_database_name sqlite3_api->column_database_name +#define sqlite3_column_database_name16 sqlite3_api->column_database_name16 +#define sqlite3_column_decltype sqlite3_api->column_decltype +#define sqlite3_column_decltype16 sqlite3_api->column_decltype16 +#define sqlite3_column_double sqlite3_api->column_double +#define sqlite3_column_int sqlite3_api->column_int +#define sqlite3_column_int64 sqlite3_api->column_int64 +#define sqlite3_column_name sqlite3_api->column_name +#define sqlite3_column_name16 sqlite3_api->column_name16 +#define sqlite3_column_origin_name sqlite3_api->column_origin_name +#define sqlite3_column_origin_name16 sqlite3_api->column_origin_name16 +#define sqlite3_column_table_name sqlite3_api->column_table_name +#define sqlite3_column_table_name16 sqlite3_api->column_table_name16 +#define sqlite3_column_text sqlite3_api->column_text +#define sqlite3_column_text16 sqlite3_api->column_text16 +#define sqlite3_column_type sqlite3_api->column_type +#define sqlite3_column_value sqlite3_api->column_value +#define sqlite3_commit_hook sqlite3_api->commit_hook +#define sqlite3_complete sqlite3_api->complete +#define sqlite3_complete16 sqlite3_api->complete16 +#define sqlite3_create_collation sqlite3_api->create_collation +#define sqlite3_create_collation16 sqlite3_api->create_collation16 +#define sqlite3_create_function sqlite3_api->create_function +#define sqlite3_create_function16 sqlite3_api->create_function16 +#define sqlite3_create_module sqlite3_api->create_module +#define sqlite3_create_module_v2 sqlite3_api->create_module_v2 +#define sqlite3_data_count sqlite3_api->data_count +#define sqlite3_db_handle sqlite3_api->db_handle +#define sqlite3_declare_vtab sqlite3_api->declare_vtab +#define sqlite3_enable_shared_cache sqlite3_api->enable_shared_cache +#define sqlite3_errcode sqlite3_api->errcode +#define sqlite3_errmsg sqlite3_api->errmsg +#define sqlite3_errmsg16 sqlite3_api->errmsg16 +#define sqlite3_exec sqlite3_api->exec +#ifndef SQLITE_OMIT_DEPRECATED +#define sqlite3_expired sqlite3_api->expired +#endif +#define sqlite3_finalize sqlite3_api->finalize +#define sqlite3_free sqlite3_api->free +#define sqlite3_free_table sqlite3_api->free_table +#define sqlite3_get_autocommit sqlite3_api->get_autocommit +#define sqlite3_get_auxdata sqlite3_api->get_auxdata +#define sqlite3_get_table sqlite3_api->get_table +#ifndef SQLITE_OMIT_DEPRECATED +#define sqlite3_global_recover sqlite3_api->global_recover +#endif +#define sqlite3_interrupt sqlite3_api->interruptx +#define sqlite3_last_insert_rowid sqlite3_api->last_insert_rowid +#define sqlite3_libversion sqlite3_api->libversion +#define sqlite3_libversion_number sqlite3_api->libversion_number +#define sqlite3_malloc sqlite3_api->malloc +#define sqlite3_mprintf sqlite3_api->mprintf +#define sqlite3_open sqlite3_api->open +#define sqlite3_open16 sqlite3_api->open16 +#define sqlite3_prepare sqlite3_api->prepare +#define sqlite3_prepare16 sqlite3_api->prepare16 +#define sqlite3_prepare_v2 sqlite3_api->prepare_v2 +#define sqlite3_prepare16_v2 sqlite3_api->prepare16_v2 +#define sqlite3_profile sqlite3_api->profile +#define sqlite3_progress_handler sqlite3_api->progress_handler +#define sqlite3_realloc sqlite3_api->realloc +#define sqlite3_reset sqlite3_api->reset +#define sqlite3_result_blob sqlite3_api->result_blob +#define sqlite3_result_double sqlite3_api->result_double +#define sqlite3_result_error sqlite3_api->result_error +#define sqlite3_result_error16 sqlite3_api->result_error16 +#define sqlite3_result_int sqlite3_api->result_int +#define sqlite3_result_int64 sqlite3_api->result_int64 +#define sqlite3_result_null sqlite3_api->result_null +#define sqlite3_result_text sqlite3_api->result_text +#define sqlite3_result_text16 sqlite3_api->result_text16 +#define sqlite3_result_text16be sqlite3_api->result_text16be +#define sqlite3_result_text16le sqlite3_api->result_text16le +#define sqlite3_result_value sqlite3_api->result_value +#define sqlite3_rollback_hook sqlite3_api->rollback_hook +#define sqlite3_set_authorizer sqlite3_api->set_authorizer +#define sqlite3_set_auxdata sqlite3_api->set_auxdata +#define sqlite3_snprintf sqlite3_api->xsnprintf +#define sqlite3_step sqlite3_api->step +#define sqlite3_table_column_metadata sqlite3_api->table_column_metadata +#define sqlite3_thread_cleanup sqlite3_api->thread_cleanup +#define sqlite3_total_changes sqlite3_api->total_changes +#define sqlite3_trace sqlite3_api->trace +#ifndef SQLITE_OMIT_DEPRECATED +#define sqlite3_transfer_bindings sqlite3_api->transfer_bindings +#endif +#define sqlite3_update_hook sqlite3_api->update_hook +#define sqlite3_user_data sqlite3_api->user_data +#define sqlite3_value_blob sqlite3_api->value_blob +#define sqlite3_value_bytes sqlite3_api->value_bytes +#define sqlite3_value_bytes16 sqlite3_api->value_bytes16 +#define sqlite3_value_double sqlite3_api->value_double +#define sqlite3_value_int sqlite3_api->value_int +#define sqlite3_value_int64 sqlite3_api->value_int64 +#define sqlite3_value_numeric_type sqlite3_api->value_numeric_type +#define sqlite3_value_text sqlite3_api->value_text +#define sqlite3_value_text16 sqlite3_api->value_text16 +#define sqlite3_value_text16be sqlite3_api->value_text16be +#define sqlite3_value_text16le sqlite3_api->value_text16le +#define sqlite3_value_type sqlite3_api->value_type +#define sqlite3_vmprintf sqlite3_api->vmprintf +#define sqlite3_vsnprintf sqlite3_api->xvsnprintf +#define sqlite3_overload_function sqlite3_api->overload_function +#define sqlite3_prepare_v2 sqlite3_api->prepare_v2 +#define sqlite3_prepare16_v2 sqlite3_api->prepare16_v2 +#define sqlite3_clear_bindings sqlite3_api->clear_bindings +#define sqlite3_bind_zeroblob sqlite3_api->bind_zeroblob +#define sqlite3_blob_bytes sqlite3_api->blob_bytes +#define sqlite3_blob_close sqlite3_api->blob_close +#define sqlite3_blob_open sqlite3_api->blob_open +#define sqlite3_blob_read sqlite3_api->blob_read +#define sqlite3_blob_write sqlite3_api->blob_write +#define sqlite3_create_collation_v2 sqlite3_api->create_collation_v2 +#define sqlite3_file_control sqlite3_api->file_control +#define sqlite3_memory_highwater sqlite3_api->memory_highwater +#define sqlite3_memory_used sqlite3_api->memory_used +#define sqlite3_mutex_alloc sqlite3_api->mutex_alloc +#define sqlite3_mutex_enter sqlite3_api->mutex_enter +#define sqlite3_mutex_free sqlite3_api->mutex_free +#define sqlite3_mutex_leave sqlite3_api->mutex_leave +#define sqlite3_mutex_try sqlite3_api->mutex_try +#define sqlite3_open_v2 sqlite3_api->open_v2 +#define sqlite3_release_memory sqlite3_api->release_memory +#define sqlite3_result_error_nomem sqlite3_api->result_error_nomem +#define sqlite3_result_error_toobig sqlite3_api->result_error_toobig +#define sqlite3_sleep sqlite3_api->sleep +#define sqlite3_soft_heap_limit sqlite3_api->soft_heap_limit +#define sqlite3_vfs_find sqlite3_api->vfs_find +#define sqlite3_vfs_register sqlite3_api->vfs_register +#define sqlite3_vfs_unregister sqlite3_api->vfs_unregister +#define sqlite3_threadsafe sqlite3_api->xthreadsafe +#define sqlite3_result_zeroblob sqlite3_api->result_zeroblob +#define sqlite3_result_error_code sqlite3_api->result_error_code +#define sqlite3_test_control sqlite3_api->test_control +#define sqlite3_randomness sqlite3_api->randomness +#define sqlite3_context_db_handle sqlite3_api->context_db_handle +#define sqlite3_extended_result_codes sqlite3_api->extended_result_codes +#define sqlite3_limit sqlite3_api->limit +#define sqlite3_next_stmt sqlite3_api->next_stmt +#define sqlite3_sql sqlite3_api->sql +#define sqlite3_status sqlite3_api->status +#define sqlite3_backup_finish sqlite3_api->backup_finish +#define sqlite3_backup_init sqlite3_api->backup_init +#define sqlite3_backup_pagecount sqlite3_api->backup_pagecount +#define sqlite3_backup_remaining sqlite3_api->backup_remaining +#define sqlite3_backup_step sqlite3_api->backup_step +#define sqlite3_compileoption_get sqlite3_api->compileoption_get +#define sqlite3_compileoption_used sqlite3_api->compileoption_used +#define sqlite3_create_function_v2 sqlite3_api->create_function_v2 +#define sqlite3_db_config sqlite3_api->db_config +#define sqlite3_db_mutex sqlite3_api->db_mutex +#define sqlite3_db_status sqlite3_api->db_status +#define sqlite3_extended_errcode sqlite3_api->extended_errcode +#define sqlite3_log sqlite3_api->log +#define sqlite3_soft_heap_limit64 sqlite3_api->soft_heap_limit64 +#define sqlite3_sourceid sqlite3_api->sourceid +#define sqlite3_stmt_status sqlite3_api->stmt_status +#define sqlite3_strnicmp sqlite3_api->strnicmp +#define sqlite3_unlock_notify sqlite3_api->unlock_notify +#define sqlite3_wal_autocheckpoint sqlite3_api->wal_autocheckpoint +#define sqlite3_wal_checkpoint sqlite3_api->wal_checkpoint +#define sqlite3_wal_hook sqlite3_api->wal_hook +#define sqlite3_blob_reopen sqlite3_api->blob_reopen +#define sqlite3_vtab_config sqlite3_api->vtab_config +#define sqlite3_vtab_on_conflict sqlite3_api->vtab_on_conflict +/* Version 3.7.16 and later */ +#define sqlite3_close_v2 sqlite3_api->close_v2 +#define sqlite3_db_filename sqlite3_api->db_filename +#define sqlite3_db_readonly sqlite3_api->db_readonly +#define sqlite3_db_release_memory sqlite3_api->db_release_memory +#define sqlite3_errstr sqlite3_api->errstr +#define sqlite3_stmt_busy sqlite3_api->stmt_busy +#define sqlite3_stmt_readonly sqlite3_api->stmt_readonly +#define sqlite3_stricmp sqlite3_api->stricmp +#define sqlite3_uri_boolean sqlite3_api->uri_boolean +#define sqlite3_uri_int64 sqlite3_api->uri_int64 +#define sqlite3_uri_parameter sqlite3_api->uri_parameter +#define sqlite3_uri_vsnprintf sqlite3_api->xvsnprintf +#define sqlite3_wal_checkpoint_v2 sqlite3_api->wal_checkpoint_v2 +/* Version 3.8.7 and later */ +#define sqlite3_auto_extension sqlite3_api->auto_extension +#define sqlite3_bind_blob64 sqlite3_api->bind_blob64 +#define sqlite3_bind_text64 sqlite3_api->bind_text64 +#define sqlite3_cancel_auto_extension sqlite3_api->cancel_auto_extension +#define sqlite3_load_extension sqlite3_api->load_extension +#define sqlite3_malloc64 sqlite3_api->malloc64 +#define sqlite3_msize sqlite3_api->msize +#define sqlite3_realloc64 sqlite3_api->realloc64 +#define sqlite3_reset_auto_extension sqlite3_api->reset_auto_extension +#define sqlite3_result_blob64 sqlite3_api->result_blob64 +#define sqlite3_result_text64 sqlite3_api->result_text64 +#define sqlite3_strglob sqlite3_api->strglob +/* Version 3.8.11 and later */ +#define sqlite3_value_dup sqlite3_api->value_dup +#define sqlite3_value_free sqlite3_api->value_free +#define sqlite3_result_zeroblob64 sqlite3_api->result_zeroblob64 +#define sqlite3_bind_zeroblob64 sqlite3_api->bind_zeroblob64 +/* Version 3.9.0 and later */ +#define sqlite3_value_subtype sqlite3_api->value_subtype +#define sqlite3_result_subtype sqlite3_api->result_subtype +/* Version 3.10.0 and later */ +#define sqlite3_status64 sqlite3_api->status64 +#define sqlite3_strlike sqlite3_api->strlike +#define sqlite3_db_cacheflush sqlite3_api->db_cacheflush +/* Version 3.12.0 and later */ +#define sqlite3_system_errno sqlite3_api->system_errno +/* Version 3.14.0 and later */ +#define sqlite3_trace_v2 sqlite3_api->trace_v2 +#define sqlite3_expanded_sql sqlite3_api->expanded_sql +/* Version 3.18.0 and later */ +#define sqlite3_set_last_insert_rowid sqlite3_api->set_last_insert_rowid +/* Version 3.20.0 and later */ +#define sqlite3_prepare_v3 sqlite3_api->prepare_v3 +#define sqlite3_prepare16_v3 sqlite3_api->prepare16_v3 +#define sqlite3_bind_pointer sqlite3_api->bind_pointer +#define sqlite3_result_pointer sqlite3_api->result_pointer +#define sqlite3_value_pointer sqlite3_api->value_pointer +/* Version 3.22.0 and later */ +#define sqlite3_vtab_nochange sqlite3_api->vtab_nochange +#define sqlite3_value_nochange sqlite3_api->value_nochange +#define sqlite3_vtab_collation sqlite3_api->vtab_collation +/* Version 3.24.0 and later */ +#define sqlite3_keyword_count sqlite3_api->keyword_count +#define sqlite3_keyword_name sqlite3_api->keyword_name +#define sqlite3_keyword_check sqlite3_api->keyword_check +#define sqlite3_str_new sqlite3_api->str_new +#define sqlite3_str_finish sqlite3_api->str_finish +#define sqlite3_str_appendf sqlite3_api->str_appendf +#define sqlite3_str_vappendf sqlite3_api->str_vappendf +#define sqlite3_str_append sqlite3_api->str_append +#define sqlite3_str_appendall sqlite3_api->str_appendall +#define sqlite3_str_appendchar sqlite3_api->str_appendchar +#define sqlite3_str_reset sqlite3_api->str_reset +#define sqlite3_str_errcode sqlite3_api->str_errcode +#define sqlite3_str_length sqlite3_api->str_length +#define sqlite3_str_value sqlite3_api->str_value +/* Version 3.25.0 and later */ +#define sqlite3_create_window_function sqlite3_api->create_window_function +/* Version 3.26.0 and later */ +#define sqlite3_normalized_sql sqlite3_api->normalized_sql +/* Version 3.28.0 and later */ +#define sqlite3_stmt_isexplain sqlite3_api->isexplain +#define sqlite3_value_frombind sqlite3_api->frombind +/* Version 3.30.0 and later */ +#define sqlite3_drop_modules sqlite3_api->drop_modules +#endif /* !defined(SQLITE_CORE) && !defined(SQLITE_OMIT_LOAD_EXTENSION) */ + +#if !defined(SQLITE_CORE) && !defined(SQLITE_OMIT_LOAD_EXTENSION) + /* This case when the file really is being compiled as a loadable + ** extension */ +# define SQLITE_EXTENSION_INIT1 const sqlite3_api_routines *sqlite3_api=0; +# define SQLITE_EXTENSION_INIT2(v) sqlite3_api=v; +# define SQLITE_EXTENSION_INIT3 \ + extern const sqlite3_api_routines *sqlite3_api; +#else + /* This case when the file is being statically linked into the + ** application */ +# define SQLITE_EXTENSION_INIT1 /*no-op*/ +# define SQLITE_EXTENSION_INIT2(v) (void)v; /* unused parameter */ +# define SQLITE_EXTENSION_INIT3 /*no-op*/ +#endif + +#endif /* SQLITE3EXT_H */ + +/************** End of sqlite3ext.h ******************************************/ +/************** Continuing where we left off in loadext.c ********************/ +/* #include "sqliteInt.h" */ + +#ifndef SQLITE_OMIT_LOAD_EXTENSION +/* +** Some API routines are omitted when various features are +** excluded from a build of SQLite. Substitute a NULL pointer +** for any missing APIs. +*/ +#ifndef SQLITE_ENABLE_COLUMN_METADATA +# define sqlite3_column_database_name 0 +# define sqlite3_column_database_name16 0 +# define sqlite3_column_table_name 0 +# define sqlite3_column_table_name16 0 +# define sqlite3_column_origin_name 0 +# define sqlite3_column_origin_name16 0 +#endif + +#ifdef SQLITE_OMIT_AUTHORIZATION +# define sqlite3_set_authorizer 0 +#endif + +#ifdef SQLITE_OMIT_UTF16 +# define sqlite3_bind_text16 0 +# define sqlite3_collation_needed16 0 +# define sqlite3_column_decltype16 0 +# define sqlite3_column_name16 0 +# define sqlite3_column_text16 0 +# define sqlite3_complete16 0 +# define sqlite3_create_collation16 0 +# define sqlite3_create_function16 0 +# define sqlite3_errmsg16 0 +# define sqlite3_open16 0 +# define sqlite3_prepare16 0 +# define sqlite3_prepare16_v2 0 +# define sqlite3_prepare16_v3 0 +# define sqlite3_result_error16 0 +# define sqlite3_result_text16 0 +# define sqlite3_result_text16be 0 +# define sqlite3_result_text16le 0 +# define sqlite3_value_text16 0 +# define sqlite3_value_text16be 0 +# define sqlite3_value_text16le 0 +# define sqlite3_column_database_name16 0 +# define sqlite3_column_table_name16 0 +# define sqlite3_column_origin_name16 0 +#endif + +#ifdef SQLITE_OMIT_COMPLETE +# define sqlite3_complete 0 +# define sqlite3_complete16 0 +#endif + +#ifdef SQLITE_OMIT_DECLTYPE +# define sqlite3_column_decltype16 0 +# define sqlite3_column_decltype 0 +#endif + +#ifdef SQLITE_OMIT_PROGRESS_CALLBACK +# define sqlite3_progress_handler 0 +#endif + +#ifdef SQLITE_OMIT_VIRTUALTABLE +# define sqlite3_create_module 0 +# define sqlite3_create_module_v2 0 +# define sqlite3_declare_vtab 0 +# define sqlite3_vtab_config 0 +# define sqlite3_vtab_on_conflict 0 +# define sqlite3_vtab_collation 0 +#endif + +#ifdef SQLITE_OMIT_SHARED_CACHE +# define sqlite3_enable_shared_cache 0 +#endif + +#if defined(SQLITE_OMIT_TRACE) || defined(SQLITE_OMIT_DEPRECATED) +# define sqlite3_profile 0 +# define sqlite3_trace 0 +#endif + +#ifdef SQLITE_OMIT_GET_TABLE +# define sqlite3_free_table 0 +# define sqlite3_get_table 0 +#endif + +#ifdef SQLITE_OMIT_INCRBLOB +#define sqlite3_bind_zeroblob 0 +#define sqlite3_blob_bytes 0 +#define sqlite3_blob_close 0 +#define sqlite3_blob_open 0 +#define sqlite3_blob_read 0 +#define sqlite3_blob_write 0 +#define sqlite3_blob_reopen 0 +#endif + +#if defined(SQLITE_OMIT_TRACE) +# define sqlite3_trace_v2 0 +#endif + +/* +** The following structure contains pointers to all SQLite API routines. +** A pointer to this structure is passed into extensions when they are +** loaded so that the extension can make calls back into the SQLite +** library. +** +** When adding new APIs, add them to the bottom of this structure +** in order to preserve backwards compatibility. +** +** Extensions that use newer APIs should first call the +** sqlite3_libversion_number() to make sure that the API they +** intend to use is supported by the library. Extensions should +** also check to make sure that the pointer to the function is +** not NULL before calling it. +*/ +static const sqlite3_api_routines sqlite3Apis = { + sqlite3_aggregate_context, +#ifndef SQLITE_OMIT_DEPRECATED + sqlite3_aggregate_count, +#else + 0, +#endif + sqlite3_bind_blob, + sqlite3_bind_double, + sqlite3_bind_int, + sqlite3_bind_int64, + sqlite3_bind_null, + sqlite3_bind_parameter_count, + sqlite3_bind_parameter_index, + sqlite3_bind_parameter_name, + sqlite3_bind_text, + sqlite3_bind_text16, + sqlite3_bind_value, + sqlite3_busy_handler, + sqlite3_busy_timeout, + sqlite3_changes, + sqlite3_close, + sqlite3_collation_needed, + sqlite3_collation_needed16, + sqlite3_column_blob, + sqlite3_column_bytes, + sqlite3_column_bytes16, + sqlite3_column_count, + sqlite3_column_database_name, + sqlite3_column_database_name16, + sqlite3_column_decltype, + sqlite3_column_decltype16, + sqlite3_column_double, + sqlite3_column_int, + sqlite3_column_int64, + sqlite3_column_name, + sqlite3_column_name16, + sqlite3_column_origin_name, + sqlite3_column_origin_name16, + sqlite3_column_table_name, + sqlite3_column_table_name16, + sqlite3_column_text, + sqlite3_column_text16, + sqlite3_column_type, + sqlite3_column_value, + sqlite3_commit_hook, + sqlite3_complete, + sqlite3_complete16, + sqlite3_create_collation, + sqlite3_create_collation16, + sqlite3_create_function, + sqlite3_create_function16, + sqlite3_create_module, + sqlite3_data_count, + sqlite3_db_handle, + sqlite3_declare_vtab, + sqlite3_enable_shared_cache, + sqlite3_errcode, + sqlite3_errmsg, + sqlite3_errmsg16, + sqlite3_exec, +#ifndef SQLITE_OMIT_DEPRECATED + sqlite3_expired, +#else + 0, +#endif + sqlite3_finalize, + sqlite3_free, + sqlite3_free_table, + sqlite3_get_autocommit, + sqlite3_get_auxdata, + sqlite3_get_table, + 0, /* Was sqlite3_global_recover(), but that function is deprecated */ + sqlite3_interrupt, + sqlite3_last_insert_rowid, + sqlite3_libversion, + sqlite3_libversion_number, + sqlite3_malloc, + sqlite3_mprintf, + sqlite3_open, + sqlite3_open16, + sqlite3_prepare, + sqlite3_prepare16, + sqlite3_profile, + sqlite3_progress_handler, + sqlite3_realloc, + sqlite3_reset, + sqlite3_result_blob, + sqlite3_result_double, + sqlite3_result_error, + sqlite3_result_error16, + sqlite3_result_int, + sqlite3_result_int64, + sqlite3_result_null, + sqlite3_result_text, + sqlite3_result_text16, + sqlite3_result_text16be, + sqlite3_result_text16le, + sqlite3_result_value, + sqlite3_rollback_hook, + sqlite3_set_authorizer, + sqlite3_set_auxdata, + sqlite3_snprintf, + sqlite3_step, + sqlite3_table_column_metadata, +#ifndef SQLITE_OMIT_DEPRECATED + sqlite3_thread_cleanup, +#else + 0, +#endif + sqlite3_total_changes, + sqlite3_trace, +#ifndef SQLITE_OMIT_DEPRECATED + sqlite3_transfer_bindings, +#else + 0, +#endif + sqlite3_update_hook, + sqlite3_user_data, + sqlite3_value_blob, + sqlite3_value_bytes, + sqlite3_value_bytes16, + sqlite3_value_double, + sqlite3_value_int, + sqlite3_value_int64, + sqlite3_value_numeric_type, + sqlite3_value_text, + sqlite3_value_text16, + sqlite3_value_text16be, + sqlite3_value_text16le, + sqlite3_value_type, + sqlite3_vmprintf, + /* + ** The original API set ends here. All extensions can call any + ** of the APIs above provided that the pointer is not NULL. But + ** before calling APIs that follow, extension should check the + ** sqlite3_libversion_number() to make sure they are dealing with + ** a library that is new enough to support that API. + ************************************************************************* + */ + sqlite3_overload_function, + + /* + ** Added after 3.3.13 + */ + sqlite3_prepare_v2, + sqlite3_prepare16_v2, + sqlite3_clear_bindings, + + /* + ** Added for 3.4.1 + */ + sqlite3_create_module_v2, + + /* + ** Added for 3.5.0 + */ + sqlite3_bind_zeroblob, + sqlite3_blob_bytes, + sqlite3_blob_close, + sqlite3_blob_open, + sqlite3_blob_read, + sqlite3_blob_write, + sqlite3_create_collation_v2, + sqlite3_file_control, + sqlite3_memory_highwater, + sqlite3_memory_used, +#ifdef SQLITE_MUTEX_OMIT + 0, + 0, + 0, + 0, + 0, +#else + sqlite3_mutex_alloc, + sqlite3_mutex_enter, + sqlite3_mutex_free, + sqlite3_mutex_leave, + sqlite3_mutex_try, +#endif + sqlite3_open_v2, + sqlite3_release_memory, + sqlite3_result_error_nomem, + sqlite3_result_error_toobig, + sqlite3_sleep, + sqlite3_soft_heap_limit, + sqlite3_vfs_find, + sqlite3_vfs_register, + sqlite3_vfs_unregister, + + /* + ** Added for 3.5.8 + */ + sqlite3_threadsafe, + sqlite3_result_zeroblob, + sqlite3_result_error_code, + sqlite3_test_control, + sqlite3_randomness, + sqlite3_context_db_handle, + + /* + ** Added for 3.6.0 + */ + sqlite3_extended_result_codes, + sqlite3_limit, + sqlite3_next_stmt, + sqlite3_sql, + sqlite3_status, + + /* + ** Added for 3.7.4 + */ + sqlite3_backup_finish, + sqlite3_backup_init, + sqlite3_backup_pagecount, + sqlite3_backup_remaining, + sqlite3_backup_step, +#ifndef SQLITE_OMIT_COMPILEOPTION_DIAGS + sqlite3_compileoption_get, + sqlite3_compileoption_used, +#else + 0, + 0, +#endif + sqlite3_create_function_v2, + sqlite3_db_config, + sqlite3_db_mutex, + sqlite3_db_status, + sqlite3_extended_errcode, + sqlite3_log, + sqlite3_soft_heap_limit64, + sqlite3_sourceid, + sqlite3_stmt_status, + sqlite3_strnicmp, +#ifdef SQLITE_ENABLE_UNLOCK_NOTIFY + sqlite3_unlock_notify, +#else + 0, +#endif +#ifndef SQLITE_OMIT_WAL + sqlite3_wal_autocheckpoint, + sqlite3_wal_checkpoint, + sqlite3_wal_hook, +#else + 0, + 0, + 0, +#endif + sqlite3_blob_reopen, + sqlite3_vtab_config, + sqlite3_vtab_on_conflict, + sqlite3_close_v2, + sqlite3_db_filename, + sqlite3_db_readonly, + sqlite3_db_release_memory, + sqlite3_errstr, + sqlite3_stmt_busy, + sqlite3_stmt_readonly, + sqlite3_stricmp, + sqlite3_uri_boolean, + sqlite3_uri_int64, + sqlite3_uri_parameter, + sqlite3_vsnprintf, + sqlite3_wal_checkpoint_v2, + /* Version 3.8.7 and later */ + sqlite3_auto_extension, + sqlite3_bind_blob64, + sqlite3_bind_text64, + sqlite3_cancel_auto_extension, + sqlite3_load_extension, + sqlite3_malloc64, + sqlite3_msize, + sqlite3_realloc64, + sqlite3_reset_auto_extension, + sqlite3_result_blob64, + sqlite3_result_text64, + sqlite3_strglob, + /* Version 3.8.11 and later */ + (sqlite3_value*(*)(const sqlite3_value*))sqlite3_value_dup, + sqlite3_value_free, + sqlite3_result_zeroblob64, + sqlite3_bind_zeroblob64, + /* Version 3.9.0 and later */ + sqlite3_value_subtype, + sqlite3_result_subtype, + /* Version 3.10.0 and later */ + sqlite3_status64, + sqlite3_strlike, + sqlite3_db_cacheflush, + /* Version 3.12.0 and later */ + sqlite3_system_errno, + /* Version 3.14.0 and later */ + sqlite3_trace_v2, + sqlite3_expanded_sql, + /* Version 3.18.0 and later */ + sqlite3_set_last_insert_rowid, + /* Version 3.20.0 and later */ + sqlite3_prepare_v3, + sqlite3_prepare16_v3, + sqlite3_bind_pointer, + sqlite3_result_pointer, + sqlite3_value_pointer, + /* Version 3.22.0 and later */ + sqlite3_vtab_nochange, + sqlite3_value_nochange, + sqlite3_vtab_collation, + /* Version 3.24.0 and later */ + sqlite3_keyword_count, + sqlite3_keyword_name, + sqlite3_keyword_check, + sqlite3_str_new, + sqlite3_str_finish, + sqlite3_str_appendf, + sqlite3_str_vappendf, + sqlite3_str_append, + sqlite3_str_appendall, + sqlite3_str_appendchar, + sqlite3_str_reset, + sqlite3_str_errcode, + sqlite3_str_length, + sqlite3_str_value, + /* Version 3.25.0 and later */ + sqlite3_create_window_function, + /* Version 3.26.0 and later */ +#ifdef SQLITE_ENABLE_NORMALIZE + sqlite3_normalized_sql, +#else + 0, +#endif + /* Version 3.28.0 and later */ + sqlite3_stmt_isexplain, + sqlite3_value_frombind, + /* Version 3.30.0 and later */ +#ifndef SQLITE_OMIT_VIRTUALTABLE + sqlite3_drop_modules, +#else + 0, +#endif +}; + +/* +** Attempt to load an SQLite extension library contained in the file +** zFile. The entry point is zProc. zProc may be 0 in which case a +** default entry point name (sqlite3_extension_init) is used. Use +** of the default name is recommended. +** +** Return SQLITE_OK on success and SQLITE_ERROR if something goes wrong. +** +** If an error occurs and pzErrMsg is not 0, then fill *pzErrMsg with +** error message text. The calling function should free this memory +** by calling sqlite3DbFree(db, ). +*/ +static int sqlite3LoadExtension( + sqlite3 *db, /* Load the extension into this database connection */ + const char *zFile, /* Name of the shared library containing extension */ + const char *zProc, /* Entry point. Use "sqlite3_extension_init" if 0 */ + char **pzErrMsg /* Put error message here if not 0 */ +){ + sqlite3_vfs *pVfs = db->pVfs; + void *handle; + sqlite3_loadext_entry xInit; + char *zErrmsg = 0; + const char *zEntry; + char *zAltEntry = 0; + void **aHandle; + u64 nMsg = 300 + sqlite3Strlen30(zFile); + int ii; + int rc; + + /* Shared library endings to try if zFile cannot be loaded as written */ + static const char *azEndings[] = { +#if SQLITE_OS_WIN + "dll" +#elif defined(__APPLE__) + "dylib" +#else + "so" +#endif + }; + + + if( pzErrMsg ) *pzErrMsg = 0; + + /* Ticket #1863. To avoid a creating security problems for older + ** applications that relink against newer versions of SQLite, the + ** ability to run load_extension is turned off by default. One + ** must call either sqlite3_enable_load_extension(db) or + ** sqlite3_db_config(db, SQLITE_DBCONFIG_ENABLE_LOAD_EXTENSION, 1, 0) + ** to turn on extension loading. + */ + if( (db->flags & SQLITE_LoadExtension)==0 ){ + if( pzErrMsg ){ + *pzErrMsg = sqlite3_mprintf("not authorized"); + } + return SQLITE_ERROR; + } + + zEntry = zProc ? zProc : "sqlite3_extension_init"; + + handle = sqlite3OsDlOpen(pVfs, zFile); +#if SQLITE_OS_UNIX || SQLITE_OS_WIN + for(ii=0; ii sqlite3_example_init + ** C:/lib/mathfuncs.dll ==> sqlite3_mathfuncs_init + */ + if( xInit==0 && zProc==0 ){ + int iFile, iEntry, c; + int ncFile = sqlite3Strlen30(zFile); + zAltEntry = sqlite3_malloc64(ncFile+30); + if( zAltEntry==0 ){ + sqlite3OsDlClose(pVfs, handle); + return SQLITE_NOMEM_BKPT; + } + memcpy(zAltEntry, "sqlite3_", 8); + for(iFile=ncFile-1; iFile>=0 && zFile[iFile]!='/'; iFile--){} + iFile++; + if( sqlite3_strnicmp(zFile+iFile, "lib", 3)==0 ) iFile += 3; + for(iEntry=8; (c = zFile[iFile])!=0 && c!='.'; iFile++){ + if( sqlite3Isalpha(c) ){ + zAltEntry[iEntry++] = (char)sqlite3UpperToLower[(unsigned)c]; + } + } + memcpy(zAltEntry+iEntry, "_init", 6); + zEntry = zAltEntry; + xInit = (sqlite3_loadext_entry)sqlite3OsDlSym(pVfs, handle, zEntry); + } + if( xInit==0 ){ + if( pzErrMsg ){ + nMsg += sqlite3Strlen30(zEntry); + *pzErrMsg = zErrmsg = sqlite3_malloc64(nMsg); + if( zErrmsg ){ + sqlite3_snprintf(nMsg, zErrmsg, + "no entry point [%s] in shared library [%s]", zEntry, zFile); + sqlite3OsDlError(pVfs, nMsg-1, zErrmsg); + } + } + sqlite3OsDlClose(pVfs, handle); + sqlite3_free(zAltEntry); + return SQLITE_ERROR; + } + sqlite3_free(zAltEntry); + rc = xInit(db, &zErrmsg, &sqlite3Apis); + if( rc ){ + if( rc==SQLITE_OK_LOAD_PERMANENTLY ) return SQLITE_OK; + if( pzErrMsg ){ + *pzErrMsg = sqlite3_mprintf("error during initialization: %s", zErrmsg); + } + sqlite3_free(zErrmsg); + sqlite3OsDlClose(pVfs, handle); + return SQLITE_ERROR; + } + + /* Append the new shared library handle to the db->aExtension array. */ + aHandle = sqlite3DbMallocZero(db, sizeof(handle)*(db->nExtension+1)); + if( aHandle==0 ){ + return SQLITE_NOMEM_BKPT; + } + if( db->nExtension>0 ){ + memcpy(aHandle, db->aExtension, sizeof(handle)*db->nExtension); + } + sqlite3DbFree(db, db->aExtension); + db->aExtension = aHandle; + + db->aExtension[db->nExtension++] = handle; + return SQLITE_OK; +} +SQLITE_API int sqlite3_load_extension( + sqlite3 *db, /* Load the extension into this database connection */ + const char *zFile, /* Name of the shared library containing extension */ + const char *zProc, /* Entry point. Use "sqlite3_extension_init" if 0 */ + char **pzErrMsg /* Put error message here if not 0 */ +){ + int rc; + sqlite3_mutex_enter(db->mutex); + rc = sqlite3LoadExtension(db, zFile, zProc, pzErrMsg); + rc = sqlite3ApiExit(db, rc); + sqlite3_mutex_leave(db->mutex); + return rc; +} + +/* +** Call this routine when the database connection is closing in order +** to clean up loaded extensions +*/ +SQLITE_PRIVATE void sqlite3CloseExtensions(sqlite3 *db){ + int i; + assert( sqlite3_mutex_held(db->mutex) ); + for(i=0; inExtension; i++){ + sqlite3OsDlClose(db->pVfs, db->aExtension[i]); + } + sqlite3DbFree(db, db->aExtension); +} + +/* +** Enable or disable extension loading. Extension loading is disabled by +** default so as not to open security holes in older applications. +*/ +SQLITE_API int sqlite3_enable_load_extension(sqlite3 *db, int onoff){ + sqlite3_mutex_enter(db->mutex); + if( onoff ){ + db->flags |= SQLITE_LoadExtension|SQLITE_LoadExtFunc; + }else{ + db->flags &= ~(u64)(SQLITE_LoadExtension|SQLITE_LoadExtFunc); + } + sqlite3_mutex_leave(db->mutex); + return SQLITE_OK; +} + +#endif /* !defined(SQLITE_OMIT_LOAD_EXTENSION) */ + +/* +** The following object holds the list of automatically loaded +** extensions. +** +** This list is shared across threads. The SQLITE_MUTEX_STATIC_MASTER +** mutex must be held while accessing this list. +*/ +typedef struct sqlite3AutoExtList sqlite3AutoExtList; +static SQLITE_WSD struct sqlite3AutoExtList { + u32 nExt; /* Number of entries in aExt[] */ + void (**aExt)(void); /* Pointers to the extension init functions */ +} sqlite3Autoext = { 0, 0 }; + +/* The "wsdAutoext" macro will resolve to the autoextension +** state vector. If writable static data is unsupported on the target, +** we have to locate the state vector at run-time. In the more common +** case where writable static data is supported, wsdStat can refer directly +** to the "sqlite3Autoext" state vector declared above. +*/ +#ifdef SQLITE_OMIT_WSD +# define wsdAutoextInit \ + sqlite3AutoExtList *x = &GLOBAL(sqlite3AutoExtList,sqlite3Autoext) +# define wsdAutoext x[0] +#else +# define wsdAutoextInit +# define wsdAutoext sqlite3Autoext +#endif + + +/* +** Register a statically linked extension that is automatically +** loaded by every new database connection. +*/ +SQLITE_API int sqlite3_auto_extension( + void (*xInit)(void) +){ + int rc = SQLITE_OK; +#ifndef SQLITE_OMIT_AUTOINIT + rc = sqlite3_initialize(); + if( rc ){ + return rc; + }else +#endif + { + u32 i; +#if SQLITE_THREADSAFE + sqlite3_mutex *mutex = sqlite3MutexAlloc(SQLITE_MUTEX_STATIC_MASTER); +#endif + wsdAutoextInit; + sqlite3_mutex_enter(mutex); + for(i=0; i=0; i--){ + if( wsdAutoext.aExt[i]==xInit ){ + wsdAutoext.nExt--; + wsdAutoext.aExt[i] = wsdAutoext.aExt[wsdAutoext.nExt]; + n++; + break; + } + } + sqlite3_mutex_leave(mutex); + return n; +} + +/* +** Reset the automatic extension loading mechanism. +*/ +SQLITE_API void sqlite3_reset_auto_extension(void){ +#ifndef SQLITE_OMIT_AUTOINIT + if( sqlite3_initialize()==SQLITE_OK ) +#endif + { +#if SQLITE_THREADSAFE + sqlite3_mutex *mutex = sqlite3MutexAlloc(SQLITE_MUTEX_STATIC_MASTER); +#endif + wsdAutoextInit; + sqlite3_mutex_enter(mutex); + sqlite3_free(wsdAutoext.aExt); + wsdAutoext.aExt = 0; + wsdAutoext.nExt = 0; + sqlite3_mutex_leave(mutex); + } +} + +/* +** Load all automatic extensions. +** +** If anything goes wrong, set an error in the database connection. +*/ +SQLITE_PRIVATE void sqlite3AutoLoadExtensions(sqlite3 *db){ + u32 i; + int go = 1; + int rc; + sqlite3_loadext_entry xInit; + + wsdAutoextInit; + if( wsdAutoext.nExt==0 ){ + /* Common case: early out without every having to acquire a mutex */ + return; + } + for(i=0; go; i++){ + char *zErrmsg; +#if SQLITE_THREADSAFE + sqlite3_mutex *mutex = sqlite3MutexAlloc(SQLITE_MUTEX_STATIC_MASTER); +#endif +#ifdef SQLITE_OMIT_LOAD_EXTENSION + const sqlite3_api_routines *pThunk = 0; +#else + const sqlite3_api_routines *pThunk = &sqlite3Apis; +#endif + sqlite3_mutex_enter(mutex); + if( i>=wsdAutoext.nExt ){ + xInit = 0; + go = 0; + }else{ + xInit = (sqlite3_loadext_entry)wsdAutoext.aExt[i]; + } + sqlite3_mutex_leave(mutex); + zErrmsg = 0; + if( xInit && (rc = xInit(db, &zErrmsg, pThunk))!=0 ){ + sqlite3ErrorWithMsg(db, rc, + "automatic extension loading failed: %s", zErrmsg); + go = 0; + } + sqlite3_free(zErrmsg); + } +} + +/************** End of loadext.c *********************************************/ +/************** Begin file pragma.c ******************************************/ +/* +** 2003 April 6 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** This file contains code used to implement the PRAGMA command. +*/ +/* #include "sqliteInt.h" */ + +#if !defined(SQLITE_ENABLE_LOCKING_STYLE) +# if defined(__APPLE__) +# define SQLITE_ENABLE_LOCKING_STYLE 1 +# else +# define SQLITE_ENABLE_LOCKING_STYLE 0 +# endif +#endif + +/*************************************************************************** +** The "pragma.h" include file is an automatically generated file that +** that includes the PragType_XXXX macro definitions and the aPragmaName[] +** object. This ensures that the aPragmaName[] table is arranged in +** lexicographical order to facility a binary search of the pragma name. +** Do not edit pragma.h directly. Edit and rerun the script in at +** ../tool/mkpragmatab.tcl. */ +/************** Include pragma.h in the middle of pragma.c *******************/ +/************** Begin file pragma.h ******************************************/ +/* DO NOT EDIT! +** This file is automatically generated by the script at +** ../tool/mkpragmatab.tcl. To update the set of pragmas, edit +** that script and rerun it. +*/ + +/* The various pragma types */ +#define PragTyp_HEADER_VALUE 0 +#define PragTyp_AUTO_VACUUM 1 +#define PragTyp_FLAG 2 +#define PragTyp_BUSY_TIMEOUT 3 +#define PragTyp_CACHE_SIZE 4 +#define PragTyp_CACHE_SPILL 5 +#define PragTyp_CASE_SENSITIVE_LIKE 6 +#define PragTyp_COLLATION_LIST 7 +#define PragTyp_COMPILE_OPTIONS 8 +#define PragTyp_DATA_STORE_DIRECTORY 9 +#define PragTyp_DATABASE_LIST 10 +#define PragTyp_DEFAULT_CACHE_SIZE 11 +#define PragTyp_ENCODING 12 +#define PragTyp_FOREIGN_KEY_CHECK 13 +#define PragTyp_FOREIGN_KEY_LIST 14 +#define PragTyp_FUNCTION_LIST 15 +#define PragTyp_INCREMENTAL_VACUUM 16 +#define PragTyp_INDEX_INFO 17 +#define PragTyp_INDEX_LIST 18 +#define PragTyp_INTEGRITY_CHECK 19 +#define PragTyp_JOURNAL_MODE 20 +#define PragTyp_JOURNAL_SIZE_LIMIT 21 +#define PragTyp_LOCK_PROXY_FILE 22 +#define PragTyp_LOCKING_MODE 23 +#define PragTyp_PAGE_COUNT 24 +#define PragTyp_MMAP_SIZE 25 +#define PragTyp_MODULE_LIST 26 +#define PragTyp_OPTIMIZE 27 +#define PragTyp_PAGE_SIZE 28 +#define PragTyp_PRAGMA_LIST 29 +#define PragTyp_SECURE_DELETE 30 +#define PragTyp_SHRINK_MEMORY 31 +#define PragTyp_SOFT_HEAP_LIMIT 32 +#define PragTyp_SYNCHRONOUS 33 +#define PragTyp_TABLE_INFO 34 +#define PragTyp_TEMP_STORE 35 +#define PragTyp_TEMP_STORE_DIRECTORY 36 +#define PragTyp_THREADS 37 +#define PragTyp_WAL_AUTOCHECKPOINT 38 +#define PragTyp_WAL_CHECKPOINT 39 +#define PragTyp_ACTIVATE_EXTENSIONS 40 +#define PragTyp_KEY 41 +#define PragTyp_LOCK_STATUS 42 +#define PragTyp_STATS 43 + +/* Property flags associated with various pragma. */ +#define PragFlg_NeedSchema 0x01 /* Force schema load before running */ +#define PragFlg_NoColumns 0x02 /* OP_ResultRow called with zero columns */ +#define PragFlg_NoColumns1 0x04 /* zero columns if RHS argument is present */ +#define PragFlg_ReadOnly 0x08 /* Read-only HEADER_VALUE */ +#define PragFlg_Result0 0x10 /* Acts as query when no argument */ +#define PragFlg_Result1 0x20 /* Acts as query when has one argument */ +#define PragFlg_SchemaOpt 0x40 /* Schema restricts name search if present */ +#define PragFlg_SchemaReq 0x80 /* Schema required - "main" is default */ + +/* Names of columns for pragmas that return multi-column result +** or that return single-column results where the name of the +** result column is different from the name of the pragma +*/ +static const char *const pragCName[] = { + /* 0 */ "id", /* Used by: foreign_key_list */ + /* 1 */ "seq", + /* 2 */ "table", + /* 3 */ "from", + /* 4 */ "to", + /* 5 */ "on_update", + /* 6 */ "on_delete", + /* 7 */ "match", + /* 8 */ "cid", /* Used by: table_xinfo */ + /* 9 */ "name", + /* 10 */ "type", + /* 11 */ "notnull", + /* 12 */ "dflt_value", + /* 13 */ "pk", + /* 14 */ "hidden", + /* table_info reuses 8 */ + /* 15 */ "seqno", /* Used by: index_xinfo */ + /* 16 */ "cid", + /* 17 */ "name", + /* 18 */ "desc", + /* 19 */ "coll", + /* 20 */ "key", + /* 21 */ "tbl", /* Used by: stats */ + /* 22 */ "idx", + /* 23 */ "wdth", + /* 24 */ "hght", + /* 25 */ "flgs", + /* 26 */ "seq", /* Used by: index_list */ + /* 27 */ "name", + /* 28 */ "unique", + /* 29 */ "origin", + /* 30 */ "partial", + /* 31 */ "table", /* Used by: foreign_key_check */ + /* 32 */ "rowid", + /* 33 */ "parent", + /* 34 */ "fkid", + /* index_info reuses 15 */ + /* 35 */ "seq", /* Used by: database_list */ + /* 36 */ "name", + /* 37 */ "file", + /* 38 */ "busy", /* Used by: wal_checkpoint */ + /* 39 */ "log", + /* 40 */ "checkpointed", + /* 41 */ "name", /* Used by: function_list */ + /* 42 */ "builtin", + /* collation_list reuses 26 */ + /* 43 */ "database", /* Used by: lock_status */ + /* 44 */ "status", + /* 45 */ "cache_size", /* Used by: default_cache_size */ + /* module_list pragma_list reuses 9 */ + /* 46 */ "timeout", /* Used by: busy_timeout */ +}; + +/* Definitions of all built-in pragmas */ +typedef struct PragmaName { + const char *const zName; /* Name of pragma */ + u8 ePragTyp; /* PragTyp_XXX value */ + u8 mPragFlg; /* Zero or more PragFlg_XXX values */ + u8 iPragCName; /* Start of column names in pragCName[] */ + u8 nPragCName; /* Num of col names. 0 means use pragma name */ + u64 iArg; /* Extra argument */ +} PragmaName; +static const PragmaName aPragmaName[] = { +#if defined(SQLITE_HAS_CODEC) || defined(SQLITE_ENABLE_CEROD) + {/* zName: */ "activate_extensions", + /* ePragTyp: */ PragTyp_ACTIVATE_EXTENSIONS, + /* ePragFlg: */ 0, + /* ColNames: */ 0, 0, + /* iArg: */ 0 }, +#endif +#if !defined(SQLITE_OMIT_SCHEMA_VERSION_PRAGMAS) + {/* zName: */ "application_id", + /* ePragTyp: */ PragTyp_HEADER_VALUE, + /* ePragFlg: */ PragFlg_NoColumns1|PragFlg_Result0, + /* ColNames: */ 0, 0, + /* iArg: */ BTREE_APPLICATION_ID }, +#endif +#if !defined(SQLITE_OMIT_AUTOVACUUM) + {/* zName: */ "auto_vacuum", + /* ePragTyp: */ PragTyp_AUTO_VACUUM, + /* ePragFlg: */ PragFlg_NeedSchema|PragFlg_Result0|PragFlg_SchemaReq|PragFlg_NoColumns1, + /* ColNames: */ 0, 0, + /* iArg: */ 0 }, +#endif +#if !defined(SQLITE_OMIT_FLAG_PRAGMAS) +#if !defined(SQLITE_OMIT_AUTOMATIC_INDEX) + {/* zName: */ "automatic_index", + /* ePragTyp: */ PragTyp_FLAG, + /* ePragFlg: */ PragFlg_Result0|PragFlg_NoColumns1, + /* ColNames: */ 0, 0, + /* iArg: */ SQLITE_AutoIndex }, +#endif +#endif + {/* zName: */ "busy_timeout", + /* ePragTyp: */ PragTyp_BUSY_TIMEOUT, + /* ePragFlg: */ PragFlg_Result0, + /* ColNames: */ 46, 1, + /* iArg: */ 0 }, +#if !defined(SQLITE_OMIT_PAGER_PRAGMAS) + {/* zName: */ "cache_size", + /* ePragTyp: */ PragTyp_CACHE_SIZE, + /* ePragFlg: */ PragFlg_NeedSchema|PragFlg_Result0|PragFlg_SchemaReq|PragFlg_NoColumns1, + /* ColNames: */ 0, 0, + /* iArg: */ 0 }, +#endif +#if !defined(SQLITE_OMIT_FLAG_PRAGMAS) + {/* zName: */ "cache_spill", + /* ePragTyp: */ PragTyp_CACHE_SPILL, + /* ePragFlg: */ PragFlg_Result0|PragFlg_SchemaReq|PragFlg_NoColumns1, + /* ColNames: */ 0, 0, + /* iArg: */ 0 }, +#endif +#if !defined(SQLITE_OMIT_CASE_SENSITIVE_LIKE_PRAGMA) + {/* zName: */ "case_sensitive_like", + /* ePragTyp: */ PragTyp_CASE_SENSITIVE_LIKE, + /* ePragFlg: */ PragFlg_NoColumns, + /* ColNames: */ 0, 0, + /* iArg: */ 0 }, +#endif + {/* zName: */ "cell_size_check", + /* ePragTyp: */ PragTyp_FLAG, + /* ePragFlg: */ PragFlg_Result0|PragFlg_NoColumns1, + /* ColNames: */ 0, 0, + /* iArg: */ SQLITE_CellSizeCk }, +#if !defined(SQLITE_OMIT_FLAG_PRAGMAS) + {/* zName: */ "checkpoint_fullfsync", + /* ePragTyp: */ PragTyp_FLAG, + /* ePragFlg: */ PragFlg_Result0|PragFlg_NoColumns1, + /* ColNames: */ 0, 0, + /* iArg: */ SQLITE_CkptFullFSync }, +#endif +#if !defined(SQLITE_OMIT_SCHEMA_PRAGMAS) + {/* zName: */ "collation_list", + /* ePragTyp: */ PragTyp_COLLATION_LIST, + /* ePragFlg: */ PragFlg_Result0, + /* ColNames: */ 26, 2, + /* iArg: */ 0 }, +#endif +#if !defined(SQLITE_OMIT_COMPILEOPTION_DIAGS) + {/* zName: */ "compile_options", + /* ePragTyp: */ PragTyp_COMPILE_OPTIONS, + /* ePragFlg: */ PragFlg_Result0, + /* ColNames: */ 0, 0, + /* iArg: */ 0 }, +#endif +#if !defined(SQLITE_OMIT_FLAG_PRAGMAS) + {/* zName: */ "count_changes", + /* ePragTyp: */ PragTyp_FLAG, + /* ePragFlg: */ PragFlg_Result0|PragFlg_NoColumns1, + /* ColNames: */ 0, 0, + /* iArg: */ SQLITE_CountRows }, +#endif +#if !defined(SQLITE_OMIT_PAGER_PRAGMAS) && SQLITE_OS_WIN + {/* zName: */ "data_store_directory", + /* ePragTyp: */ PragTyp_DATA_STORE_DIRECTORY, + /* ePragFlg: */ PragFlg_NoColumns1, + /* ColNames: */ 0, 0, + /* iArg: */ 0 }, +#endif +#if !defined(SQLITE_OMIT_SCHEMA_VERSION_PRAGMAS) + {/* zName: */ "data_version", + /* ePragTyp: */ PragTyp_HEADER_VALUE, + /* ePragFlg: */ PragFlg_ReadOnly|PragFlg_Result0, + /* ColNames: */ 0, 0, + /* iArg: */ BTREE_DATA_VERSION }, +#endif +#if !defined(SQLITE_OMIT_SCHEMA_PRAGMAS) + {/* zName: */ "database_list", + /* ePragTyp: */ PragTyp_DATABASE_LIST, + /* ePragFlg: */ PragFlg_NeedSchema|PragFlg_Result0, + /* ColNames: */ 35, 3, + /* iArg: */ 0 }, +#endif +#if !defined(SQLITE_OMIT_PAGER_PRAGMAS) && !defined(SQLITE_OMIT_DEPRECATED) + {/* zName: */ "default_cache_size", + /* ePragTyp: */ PragTyp_DEFAULT_CACHE_SIZE, + /* ePragFlg: */ PragFlg_NeedSchema|PragFlg_Result0|PragFlg_SchemaReq|PragFlg_NoColumns1, + /* ColNames: */ 45, 1, + /* iArg: */ 0 }, +#endif +#if !defined(SQLITE_OMIT_FLAG_PRAGMAS) +#if !defined(SQLITE_OMIT_FOREIGN_KEY) && !defined(SQLITE_OMIT_TRIGGER) + {/* zName: */ "defer_foreign_keys", + /* ePragTyp: */ PragTyp_FLAG, + /* ePragFlg: */ PragFlg_Result0|PragFlg_NoColumns1, + /* ColNames: */ 0, 0, + /* iArg: */ SQLITE_DeferFKs }, +#endif +#endif +#if !defined(SQLITE_OMIT_FLAG_PRAGMAS) + {/* zName: */ "empty_result_callbacks", + /* ePragTyp: */ PragTyp_FLAG, + /* ePragFlg: */ PragFlg_Result0|PragFlg_NoColumns1, + /* ColNames: */ 0, 0, + /* iArg: */ SQLITE_NullCallback }, +#endif +#if !defined(SQLITE_OMIT_UTF16) + {/* zName: */ "encoding", + /* ePragTyp: */ PragTyp_ENCODING, + /* ePragFlg: */ PragFlg_Result0|PragFlg_NoColumns1, + /* ColNames: */ 0, 0, + /* iArg: */ 0 }, +#endif +#if !defined(SQLITE_OMIT_FOREIGN_KEY) && !defined(SQLITE_OMIT_TRIGGER) + {/* zName: */ "foreign_key_check", + /* ePragTyp: */ PragTyp_FOREIGN_KEY_CHECK, + /* ePragFlg: */ PragFlg_NeedSchema|PragFlg_Result0, + /* ColNames: */ 31, 4, + /* iArg: */ 0 }, +#endif +#if !defined(SQLITE_OMIT_FOREIGN_KEY) + {/* zName: */ "foreign_key_list", + /* ePragTyp: */ PragTyp_FOREIGN_KEY_LIST, + /* ePragFlg: */ PragFlg_NeedSchema|PragFlg_Result1|PragFlg_SchemaOpt, + /* ColNames: */ 0, 8, + /* iArg: */ 0 }, +#endif +#if !defined(SQLITE_OMIT_FLAG_PRAGMAS) +#if !defined(SQLITE_OMIT_FOREIGN_KEY) && !defined(SQLITE_OMIT_TRIGGER) + {/* zName: */ "foreign_keys", + /* ePragTyp: */ PragTyp_FLAG, + /* ePragFlg: */ PragFlg_Result0|PragFlg_NoColumns1, + /* ColNames: */ 0, 0, + /* iArg: */ SQLITE_ForeignKeys }, +#endif +#endif +#if !defined(SQLITE_OMIT_SCHEMA_VERSION_PRAGMAS) + {/* zName: */ "freelist_count", + /* ePragTyp: */ PragTyp_HEADER_VALUE, + /* ePragFlg: */ PragFlg_ReadOnly|PragFlg_Result0, + /* ColNames: */ 0, 0, + /* iArg: */ BTREE_FREE_PAGE_COUNT }, +#endif +#if !defined(SQLITE_OMIT_FLAG_PRAGMAS) + {/* zName: */ "full_column_names", + /* ePragTyp: */ PragTyp_FLAG, + /* ePragFlg: */ PragFlg_Result0|PragFlg_NoColumns1, + /* ColNames: */ 0, 0, + /* iArg: */ SQLITE_FullColNames }, + {/* zName: */ "fullfsync", + /* ePragTyp: */ PragTyp_FLAG, + /* ePragFlg: */ PragFlg_Result0|PragFlg_NoColumns1, + /* ColNames: */ 0, 0, + /* iArg: */ SQLITE_FullFSync }, +#endif +#if !defined(SQLITE_OMIT_SCHEMA_PRAGMAS) +#if !defined(SQLITE_OMIT_INTROSPECTION_PRAGMAS) + {/* zName: */ "function_list", + /* ePragTyp: */ PragTyp_FUNCTION_LIST, + /* ePragFlg: */ PragFlg_Result0, + /* ColNames: */ 41, 2, + /* iArg: */ 0 }, +#endif +#endif +#if defined(SQLITE_HAS_CODEC) + {/* zName: */ "hexkey", + /* ePragTyp: */ PragTyp_KEY, + /* ePragFlg: */ 0, + /* ColNames: */ 0, 0, + /* iArg: */ 2 }, + {/* zName: */ "hexrekey", + /* ePragTyp: */ PragTyp_KEY, + /* ePragFlg: */ 0, + /* ColNames: */ 0, 0, + /* iArg: */ 3 }, +#endif +#if !defined(SQLITE_OMIT_FLAG_PRAGMAS) +#if !defined(SQLITE_OMIT_CHECK) + {/* zName: */ "ignore_check_constraints", + /* ePragTyp: */ PragTyp_FLAG, + /* ePragFlg: */ PragFlg_Result0|PragFlg_NoColumns1, + /* ColNames: */ 0, 0, + /* iArg: */ SQLITE_IgnoreChecks }, +#endif +#endif +#if !defined(SQLITE_OMIT_AUTOVACUUM) + {/* zName: */ "incremental_vacuum", + /* ePragTyp: */ PragTyp_INCREMENTAL_VACUUM, + /* ePragFlg: */ PragFlg_NeedSchema|PragFlg_NoColumns, + /* ColNames: */ 0, 0, + /* iArg: */ 0 }, +#endif +#if !defined(SQLITE_OMIT_SCHEMA_PRAGMAS) + {/* zName: */ "index_info", + /* ePragTyp: */ PragTyp_INDEX_INFO, + /* ePragFlg: */ PragFlg_NeedSchema|PragFlg_Result1|PragFlg_SchemaOpt, + /* ColNames: */ 15, 3, + /* iArg: */ 0 }, + {/* zName: */ "index_list", + /* ePragTyp: */ PragTyp_INDEX_LIST, + /* ePragFlg: */ PragFlg_NeedSchema|PragFlg_Result1|PragFlg_SchemaOpt, + /* ColNames: */ 26, 5, + /* iArg: */ 0 }, + {/* zName: */ "index_xinfo", + /* ePragTyp: */ PragTyp_INDEX_INFO, + /* ePragFlg: */ PragFlg_NeedSchema|PragFlg_Result1|PragFlg_SchemaOpt, + /* ColNames: */ 15, 6, + /* iArg: */ 1 }, +#endif +#if !defined(SQLITE_OMIT_INTEGRITY_CHECK) + {/* zName: */ "integrity_check", + /* ePragTyp: */ PragTyp_INTEGRITY_CHECK, + /* ePragFlg: */ PragFlg_NeedSchema|PragFlg_Result0|PragFlg_Result1, + /* ColNames: */ 0, 0, + /* iArg: */ 0 }, +#endif +#if !defined(SQLITE_OMIT_PAGER_PRAGMAS) + {/* zName: */ "journal_mode", + /* ePragTyp: */ PragTyp_JOURNAL_MODE, + /* ePragFlg: */ PragFlg_NeedSchema|PragFlg_Result0|PragFlg_SchemaReq, + /* ColNames: */ 0, 0, + /* iArg: */ 0 }, + {/* zName: */ "journal_size_limit", + /* ePragTyp: */ PragTyp_JOURNAL_SIZE_LIMIT, + /* ePragFlg: */ PragFlg_Result0|PragFlg_SchemaReq, + /* ColNames: */ 0, 0, + /* iArg: */ 0 }, +#endif +#if defined(SQLITE_HAS_CODEC) + {/* zName: */ "key", + /* ePragTyp: */ PragTyp_KEY, + /* ePragFlg: */ 0, + /* ColNames: */ 0, 0, + /* iArg: */ 0 }, +#endif +#if !defined(SQLITE_OMIT_FLAG_PRAGMAS) + {/* zName: */ "legacy_alter_table", + /* ePragTyp: */ PragTyp_FLAG, + /* ePragFlg: */ PragFlg_Result0|PragFlg_NoColumns1, + /* ColNames: */ 0, 0, + /* iArg: */ SQLITE_LegacyAlter }, + {/* zName: */ "legacy_file_format", + /* ePragTyp: */ PragTyp_FLAG, + /* ePragFlg: */ PragFlg_Result0|PragFlg_NoColumns1, + /* ColNames: */ 0, 0, + /* iArg: */ SQLITE_LegacyFileFmt }, +#endif +#if !defined(SQLITE_OMIT_PAGER_PRAGMAS) && SQLITE_ENABLE_LOCKING_STYLE + {/* zName: */ "lock_proxy_file", + /* ePragTyp: */ PragTyp_LOCK_PROXY_FILE, + /* ePragFlg: */ PragFlg_NoColumns1, + /* ColNames: */ 0, 0, + /* iArg: */ 0 }, +#endif +#if defined(SQLITE_DEBUG) || defined(SQLITE_TEST) + {/* zName: */ "lock_status", + /* ePragTyp: */ PragTyp_LOCK_STATUS, + /* ePragFlg: */ PragFlg_Result0, + /* ColNames: */ 43, 2, + /* iArg: */ 0 }, +#endif +#if !defined(SQLITE_OMIT_PAGER_PRAGMAS) + {/* zName: */ "locking_mode", + /* ePragTyp: */ PragTyp_LOCKING_MODE, + /* ePragFlg: */ PragFlg_Result0|PragFlg_SchemaReq, + /* ColNames: */ 0, 0, + /* iArg: */ 0 }, + {/* zName: */ "max_page_count", + /* ePragTyp: */ PragTyp_PAGE_COUNT, + /* ePragFlg: */ PragFlg_NeedSchema|PragFlg_Result0|PragFlg_SchemaReq, + /* ColNames: */ 0, 0, + /* iArg: */ 0 }, + {/* zName: */ "mmap_size", + /* ePragTyp: */ PragTyp_MMAP_SIZE, + /* ePragFlg: */ 0, + /* ColNames: */ 0, 0, + /* iArg: */ 0 }, +#endif +#if !defined(SQLITE_OMIT_SCHEMA_PRAGMAS) +#if !defined(SQLITE_OMIT_VIRTUALTABLE) +#if !defined(SQLITE_OMIT_INTROSPECTION_PRAGMAS) + {/* zName: */ "module_list", + /* ePragTyp: */ PragTyp_MODULE_LIST, + /* ePragFlg: */ PragFlg_Result0, + /* ColNames: */ 9, 1, + /* iArg: */ 0 }, +#endif +#endif +#endif + {/* zName: */ "optimize", + /* ePragTyp: */ PragTyp_OPTIMIZE, + /* ePragFlg: */ PragFlg_Result1|PragFlg_NeedSchema, + /* ColNames: */ 0, 0, + /* iArg: */ 0 }, +#if !defined(SQLITE_OMIT_PAGER_PRAGMAS) + {/* zName: */ "page_count", + /* ePragTyp: */ PragTyp_PAGE_COUNT, + /* ePragFlg: */ PragFlg_NeedSchema|PragFlg_Result0|PragFlg_SchemaReq, + /* ColNames: */ 0, 0, + /* iArg: */ 0 }, + {/* zName: */ "page_size", + /* ePragTyp: */ PragTyp_PAGE_SIZE, + /* ePragFlg: */ PragFlg_Result0|PragFlg_SchemaReq|PragFlg_NoColumns1, + /* ColNames: */ 0, 0, + /* iArg: */ 0 }, +#endif +#if !defined(SQLITE_OMIT_FLAG_PRAGMAS) +#if defined(SQLITE_DEBUG) + {/* zName: */ "parser_trace", + /* ePragTyp: */ PragTyp_FLAG, + /* ePragFlg: */ PragFlg_Result0|PragFlg_NoColumns1, + /* ColNames: */ 0, 0, + /* iArg: */ SQLITE_ParserTrace }, +#endif +#endif +#if !defined(SQLITE_OMIT_INTROSPECTION_PRAGMAS) + {/* zName: */ "pragma_list", + /* ePragTyp: */ PragTyp_PRAGMA_LIST, + /* ePragFlg: */ PragFlg_Result0, + /* ColNames: */ 9, 1, + /* iArg: */ 0 }, +#endif +#if !defined(SQLITE_OMIT_FLAG_PRAGMAS) + {/* zName: */ "query_only", + /* ePragTyp: */ PragTyp_FLAG, + /* ePragFlg: */ PragFlg_Result0|PragFlg_NoColumns1, + /* ColNames: */ 0, 0, + /* iArg: */ SQLITE_QueryOnly }, +#endif +#if !defined(SQLITE_OMIT_INTEGRITY_CHECK) + {/* zName: */ "quick_check", + /* ePragTyp: */ PragTyp_INTEGRITY_CHECK, + /* ePragFlg: */ PragFlg_NeedSchema|PragFlg_Result0|PragFlg_Result1, + /* ColNames: */ 0, 0, + /* iArg: */ 0 }, +#endif +#if !defined(SQLITE_OMIT_FLAG_PRAGMAS) + {/* zName: */ "read_uncommitted", + /* ePragTyp: */ PragTyp_FLAG, + /* ePragFlg: */ PragFlg_Result0|PragFlg_NoColumns1, + /* ColNames: */ 0, 0, + /* iArg: */ SQLITE_ReadUncommit }, + {/* zName: */ "recursive_triggers", + /* ePragTyp: */ PragTyp_FLAG, + /* ePragFlg: */ PragFlg_Result0|PragFlg_NoColumns1, + /* ColNames: */ 0, 0, + /* iArg: */ SQLITE_RecTriggers }, +#endif +#if defined(SQLITE_HAS_CODEC) + {/* zName: */ "rekey", + /* ePragTyp: */ PragTyp_KEY, + /* ePragFlg: */ 0, + /* ColNames: */ 0, 0, + /* iArg: */ 1 }, +#endif +#if !defined(SQLITE_OMIT_FLAG_PRAGMAS) + {/* zName: */ "reverse_unordered_selects", + /* ePragTyp: */ PragTyp_FLAG, + /* ePragFlg: */ PragFlg_Result0|PragFlg_NoColumns1, + /* ColNames: */ 0, 0, + /* iArg: */ SQLITE_ReverseOrder }, +#endif +#if !defined(SQLITE_OMIT_SCHEMA_VERSION_PRAGMAS) + {/* zName: */ "schema_version", + /* ePragTyp: */ PragTyp_HEADER_VALUE, + /* ePragFlg: */ PragFlg_NoColumns1|PragFlg_Result0, + /* ColNames: */ 0, 0, + /* iArg: */ BTREE_SCHEMA_VERSION }, +#endif +#if !defined(SQLITE_OMIT_PAGER_PRAGMAS) + {/* zName: */ "secure_delete", + /* ePragTyp: */ PragTyp_SECURE_DELETE, + /* ePragFlg: */ PragFlg_Result0, + /* ColNames: */ 0, 0, + /* iArg: */ 0 }, +#endif +#if !defined(SQLITE_OMIT_FLAG_PRAGMAS) + {/* zName: */ "short_column_names", + /* ePragTyp: */ PragTyp_FLAG, + /* ePragFlg: */ PragFlg_Result0|PragFlg_NoColumns1, + /* ColNames: */ 0, 0, + /* iArg: */ SQLITE_ShortColNames }, +#endif + {/* zName: */ "shrink_memory", + /* ePragTyp: */ PragTyp_SHRINK_MEMORY, + /* ePragFlg: */ PragFlg_NoColumns, + /* ColNames: */ 0, 0, + /* iArg: */ 0 }, + {/* zName: */ "soft_heap_limit", + /* ePragTyp: */ PragTyp_SOFT_HEAP_LIMIT, + /* ePragFlg: */ PragFlg_Result0, + /* ColNames: */ 0, 0, + /* iArg: */ 0 }, +#if !defined(SQLITE_OMIT_FLAG_PRAGMAS) +#if defined(SQLITE_DEBUG) + {/* zName: */ "sql_trace", + /* ePragTyp: */ PragTyp_FLAG, + /* ePragFlg: */ PragFlg_Result0|PragFlg_NoColumns1, + /* ColNames: */ 0, 0, + /* iArg: */ SQLITE_SqlTrace }, +#endif +#endif +#if !defined(SQLITE_OMIT_SCHEMA_PRAGMAS) && defined(SQLITE_DEBUG) + {/* zName: */ "stats", + /* ePragTyp: */ PragTyp_STATS, + /* ePragFlg: */ PragFlg_NeedSchema|PragFlg_Result0|PragFlg_SchemaReq, + /* ColNames: */ 21, 5, + /* iArg: */ 0 }, +#endif +#if !defined(SQLITE_OMIT_PAGER_PRAGMAS) + {/* zName: */ "synchronous", + /* ePragTyp: */ PragTyp_SYNCHRONOUS, + /* ePragFlg: */ PragFlg_NeedSchema|PragFlg_Result0|PragFlg_SchemaReq|PragFlg_NoColumns1, + /* ColNames: */ 0, 0, + /* iArg: */ 0 }, +#endif +#if !defined(SQLITE_OMIT_SCHEMA_PRAGMAS) + {/* zName: */ "table_info", + /* ePragTyp: */ PragTyp_TABLE_INFO, + /* ePragFlg: */ PragFlg_NeedSchema|PragFlg_Result1|PragFlg_SchemaOpt, + /* ColNames: */ 8, 6, + /* iArg: */ 0 }, + {/* zName: */ "table_xinfo", + /* ePragTyp: */ PragTyp_TABLE_INFO, + /* ePragFlg: */ PragFlg_NeedSchema|PragFlg_Result1|PragFlg_SchemaOpt, + /* ColNames: */ 8, 7, + /* iArg: */ 1 }, +#endif +#if !defined(SQLITE_OMIT_PAGER_PRAGMAS) + {/* zName: */ "temp_store", + /* ePragTyp: */ PragTyp_TEMP_STORE, + /* ePragFlg: */ PragFlg_Result0|PragFlg_NoColumns1, + /* ColNames: */ 0, 0, + /* iArg: */ 0 }, + {/* zName: */ "temp_store_directory", + /* ePragTyp: */ PragTyp_TEMP_STORE_DIRECTORY, + /* ePragFlg: */ PragFlg_NoColumns1, + /* ColNames: */ 0, 0, + /* iArg: */ 0 }, +#endif +#if defined(SQLITE_HAS_CODEC) + {/* zName: */ "textkey", + /* ePragTyp: */ PragTyp_KEY, + /* ePragFlg: */ 0, + /* ColNames: */ 0, 0, + /* iArg: */ 4 }, + {/* zName: */ "textrekey", + /* ePragTyp: */ PragTyp_KEY, + /* ePragFlg: */ 0, + /* ColNames: */ 0, 0, + /* iArg: */ 5 }, +#endif + {/* zName: */ "threads", + /* ePragTyp: */ PragTyp_THREADS, + /* ePragFlg: */ PragFlg_Result0, + /* ColNames: */ 0, 0, + /* iArg: */ 0 }, +#if !defined(SQLITE_OMIT_SCHEMA_VERSION_PRAGMAS) + {/* zName: */ "user_version", + /* ePragTyp: */ PragTyp_HEADER_VALUE, + /* ePragFlg: */ PragFlg_NoColumns1|PragFlg_Result0, + /* ColNames: */ 0, 0, + /* iArg: */ BTREE_USER_VERSION }, +#endif +#if !defined(SQLITE_OMIT_FLAG_PRAGMAS) +#if defined(SQLITE_DEBUG) + {/* zName: */ "vdbe_addoptrace", + /* ePragTyp: */ PragTyp_FLAG, + /* ePragFlg: */ PragFlg_Result0|PragFlg_NoColumns1, + /* ColNames: */ 0, 0, + /* iArg: */ SQLITE_VdbeAddopTrace }, + {/* zName: */ "vdbe_debug", + /* ePragTyp: */ PragTyp_FLAG, + /* ePragFlg: */ PragFlg_Result0|PragFlg_NoColumns1, + /* ColNames: */ 0, 0, + /* iArg: */ SQLITE_SqlTrace|SQLITE_VdbeListing|SQLITE_VdbeTrace }, + {/* zName: */ "vdbe_eqp", + /* ePragTyp: */ PragTyp_FLAG, + /* ePragFlg: */ PragFlg_Result0|PragFlg_NoColumns1, + /* ColNames: */ 0, 0, + /* iArg: */ SQLITE_VdbeEQP }, + {/* zName: */ "vdbe_listing", + /* ePragTyp: */ PragTyp_FLAG, + /* ePragFlg: */ PragFlg_Result0|PragFlg_NoColumns1, + /* ColNames: */ 0, 0, + /* iArg: */ SQLITE_VdbeListing }, + {/* zName: */ "vdbe_trace", + /* ePragTyp: */ PragTyp_FLAG, + /* ePragFlg: */ PragFlg_Result0|PragFlg_NoColumns1, + /* ColNames: */ 0, 0, + /* iArg: */ SQLITE_VdbeTrace }, +#endif +#endif +#if !defined(SQLITE_OMIT_WAL) + {/* zName: */ "wal_autocheckpoint", + /* ePragTyp: */ PragTyp_WAL_AUTOCHECKPOINT, + /* ePragFlg: */ 0, + /* ColNames: */ 0, 0, + /* iArg: */ 0 }, + {/* zName: */ "wal_checkpoint", + /* ePragTyp: */ PragTyp_WAL_CHECKPOINT, + /* ePragFlg: */ PragFlg_NeedSchema, + /* ColNames: */ 38, 3, + /* iArg: */ 0 }, +#endif +#if !defined(SQLITE_OMIT_FLAG_PRAGMAS) + {/* zName: */ "writable_schema", + /* ePragTyp: */ PragTyp_FLAG, + /* ePragFlg: */ PragFlg_Result0|PragFlg_NoColumns1, + /* ColNames: */ 0, 0, + /* iArg: */ SQLITE_WriteSchema|SQLITE_NoSchemaError }, +#endif +}; +/* Number of pragmas: 65 on by default, 81 total. */ + +/************** End of pragma.h **********************************************/ +/************** Continuing where we left off in pragma.c *********************/ + +/* +** Interpret the given string as a safety level. Return 0 for OFF, +** 1 for ON or NORMAL, 2 for FULL, and 3 for EXTRA. Return 1 for an empty or +** unrecognized string argument. The FULL and EXTRA option is disallowed +** if the omitFull parameter it 1. +** +** Note that the values returned are one less that the values that +** should be passed into sqlite3BtreeSetSafetyLevel(). The is done +** to support legacy SQL code. The safety level used to be boolean +** and older scripts may have used numbers 0 for OFF and 1 for ON. +*/ +static u8 getSafetyLevel(const char *z, int omitFull, u8 dflt){ + /* 123456789 123456789 123 */ + static const char zText[] = "onoffalseyestruextrafull"; + static const u8 iOffset[] = {0, 1, 2, 4, 9, 12, 15, 20}; + static const u8 iLength[] = {2, 2, 3, 5, 3, 4, 5, 4}; + static const u8 iValue[] = {1, 0, 0, 0, 1, 1, 3, 2}; + /* on no off false yes true extra full */ + int i, n; + if( sqlite3Isdigit(*z) ){ + return (u8)sqlite3Atoi(z); + } + n = sqlite3Strlen30(z); + for(i=0; i=0&&i<=2)?i:0); +} +#endif /* ifndef SQLITE_OMIT_AUTOVACUUM */ + +#ifndef SQLITE_OMIT_PAGER_PRAGMAS +/* +** Interpret the given string as a temp db location. Return 1 for file +** backed temporary databases, 2 for the Red-Black tree in memory database +** and 0 to use the compile-time default. +*/ +static int getTempStore(const char *z){ + if( z[0]>='0' && z[0]<='2' ){ + return z[0] - '0'; + }else if( sqlite3StrICmp(z, "file")==0 ){ + return 1; + }else if( sqlite3StrICmp(z, "memory")==0 ){ + return 2; + }else{ + return 0; + } +} +#endif /* SQLITE_PAGER_PRAGMAS */ + +#ifndef SQLITE_OMIT_PAGER_PRAGMAS +/* +** Invalidate temp storage, either when the temp storage is changed +** from default, or when 'file' and the temp_store_directory has changed +*/ +static int invalidateTempStorage(Parse *pParse){ + sqlite3 *db = pParse->db; + if( db->aDb[1].pBt!=0 ){ + if( !db->autoCommit || sqlite3BtreeIsInReadTrans(db->aDb[1].pBt) ){ + sqlite3ErrorMsg(pParse, "temporary storage cannot be changed " + "from within a transaction"); + return SQLITE_ERROR; + } + sqlite3BtreeClose(db->aDb[1].pBt); + db->aDb[1].pBt = 0; + sqlite3ResetAllSchemasOfConnection(db); + } + return SQLITE_OK; +} +#endif /* SQLITE_PAGER_PRAGMAS */ + +#ifndef SQLITE_OMIT_PAGER_PRAGMAS +/* +** If the TEMP database is open, close it and mark the database schema +** as needing reloading. This must be done when using the SQLITE_TEMP_STORE +** or DEFAULT_TEMP_STORE pragmas. +*/ +static int changeTempStorage(Parse *pParse, const char *zStorageType){ + int ts = getTempStore(zStorageType); + sqlite3 *db = pParse->db; + if( db->temp_store==ts ) return SQLITE_OK; + if( invalidateTempStorage( pParse ) != SQLITE_OK ){ + return SQLITE_ERROR; + } + db->temp_store = (u8)ts; + return SQLITE_OK; +} +#endif /* SQLITE_PAGER_PRAGMAS */ + +/* +** Set result column names for a pragma. +*/ +static void setPragmaResultColumnNames( + Vdbe *v, /* The query under construction */ + const PragmaName *pPragma /* The pragma */ +){ + u8 n = pPragma->nPragCName; + sqlite3VdbeSetNumCols(v, n==0 ? 1 : n); + if( n==0 ){ + sqlite3VdbeSetColName(v, 0, COLNAME_NAME, pPragma->zName, SQLITE_STATIC); + }else{ + int i, j; + for(i=0, j=pPragma->iPragCName; iautoCommit ){ + Db *pDb = db->aDb; + int n = db->nDb; + assert( SQLITE_FullFSync==PAGER_FULLFSYNC ); + assert( SQLITE_CkptFullFSync==PAGER_CKPT_FULLFSYNC ); + assert( SQLITE_CacheSpill==PAGER_CACHESPILL ); + assert( (PAGER_FULLFSYNC | PAGER_CKPT_FULLFSYNC | PAGER_CACHESPILL) + == PAGER_FLAGS_MASK ); + assert( (pDb->safety_level & PAGER_SYNCHRONOUS_MASK)==pDb->safety_level ); + while( (n--) > 0 ){ + if( pDb->pBt ){ + sqlite3BtreeSetPagerFlags(pDb->pBt, + pDb->safety_level | (db->flags & PAGER_FLAGS_MASK) ); + } + pDb++; + } + } +} +#else +# define setAllPagerFlags(X) /* no-op */ +#endif + + +/* +** Return a human-readable name for a constraint resolution action. +*/ +#ifndef SQLITE_OMIT_FOREIGN_KEY +static const char *actionName(u8 action){ + const char *zName; + switch( action ){ + case OE_SetNull: zName = "SET NULL"; break; + case OE_SetDflt: zName = "SET DEFAULT"; break; + case OE_Cascade: zName = "CASCADE"; break; + case OE_Restrict: zName = "RESTRICT"; break; + default: zName = "NO ACTION"; + assert( action==OE_None ); break; + } + return zName; +} +#endif + + +/* +** Parameter eMode must be one of the PAGER_JOURNALMODE_XXX constants +** defined in pager.h. This function returns the associated lowercase +** journal-mode name. +*/ +SQLITE_PRIVATE const char *sqlite3JournalModename(int eMode){ + static char * const azModeName[] = { + "delete", "persist", "off", "truncate", "memory" +#ifndef SQLITE_OMIT_WAL + , "wal" +#endif + }; + assert( PAGER_JOURNALMODE_DELETE==0 ); + assert( PAGER_JOURNALMODE_PERSIST==1 ); + assert( PAGER_JOURNALMODE_OFF==2 ); + assert( PAGER_JOURNALMODE_TRUNCATE==3 ); + assert( PAGER_JOURNALMODE_MEMORY==4 ); + assert( PAGER_JOURNALMODE_WAL==5 ); + assert( eMode>=0 && eMode<=ArraySize(azModeName) ); + + if( eMode==ArraySize(azModeName) ) return 0; + return azModeName[eMode]; +} + +/* +** Locate a pragma in the aPragmaName[] array. +*/ +static const PragmaName *pragmaLocate(const char *zName){ + int upr, lwr, mid = 0, rc; + lwr = 0; + upr = ArraySize(aPragmaName)-1; + while( lwr<=upr ){ + mid = (lwr+upr)/2; + rc = sqlite3_stricmp(zName, aPragmaName[mid].zName); + if( rc==0 ) break; + if( rc<0 ){ + upr = mid - 1; + }else{ + lwr = mid + 1; + } + } + return lwr>upr ? 0 : &aPragmaName[mid]; +} + +/* +** Helper subroutine for PRAGMA integrity_check: +** +** Generate code to output a single-column result row with a value of the +** string held in register 3. Decrement the result count in register 1 +** and halt if the maximum number of result rows have been issued. +*/ +static int integrityCheckResultRow(Vdbe *v){ + int addr; + sqlite3VdbeAddOp2(v, OP_ResultRow, 3, 1); + addr = sqlite3VdbeAddOp3(v, OP_IfPos, 1, sqlite3VdbeCurrentAddr(v)+2, 1); + VdbeCoverage(v); + sqlite3VdbeAddOp0(v, OP_Halt); + return addr; +} + +/* +** Process a pragma statement. +** +** Pragmas are of this form: +** +** PRAGMA [schema.]id [= value] +** +** The identifier might also be a string. The value is a string, and +** identifier, or a number. If minusFlag is true, then the value is +** a number that was preceded by a minus sign. +** +** If the left side is "database.id" then pId1 is the database name +** and pId2 is the id. If the left side is just "id" then pId1 is the +** id and pId2 is any empty string. +*/ +SQLITE_PRIVATE void sqlite3Pragma( + Parse *pParse, + Token *pId1, /* First part of [schema.]id field */ + Token *pId2, /* Second part of [schema.]id field, or NULL */ + Token *pValue, /* Token for , or NULL */ + int minusFlag /* True if a '-' sign preceded */ +){ + char *zLeft = 0; /* Nul-terminated UTF-8 string */ + char *zRight = 0; /* Nul-terminated UTF-8 string , or NULL */ + const char *zDb = 0; /* The database name */ + Token *pId; /* Pointer to token */ + char *aFcntl[4]; /* Argument to SQLITE_FCNTL_PRAGMA */ + int iDb; /* Database index for */ + int rc; /* return value form SQLITE_FCNTL_PRAGMA */ + sqlite3 *db = pParse->db; /* The database connection */ + Db *pDb; /* The specific database being pragmaed */ + Vdbe *v = sqlite3GetVdbe(pParse); /* Prepared statement */ + const PragmaName *pPragma; /* The pragma */ + + if( v==0 ) return; + sqlite3VdbeRunOnlyOnce(v); + pParse->nMem = 2; + + /* Interpret the [schema.] part of the pragma statement. iDb is the + ** index of the database this pragma is being applied to in db.aDb[]. */ + iDb = sqlite3TwoPartName(pParse, pId1, pId2, &pId); + if( iDb<0 ) return; + pDb = &db->aDb[iDb]; + + /* If the temp database has been explicitly named as part of the + ** pragma, make sure it is open. + */ + if( iDb==1 && sqlite3OpenTempDatabase(pParse) ){ + return; + } + + zLeft = sqlite3NameFromToken(db, pId); + if( !zLeft ) return; + if( minusFlag ){ + zRight = sqlite3MPrintf(db, "-%T", pValue); + }else{ + zRight = sqlite3NameFromToken(db, pValue); + } + + assert( pId2 ); + zDb = pId2->n>0 ? pDb->zDbSName : 0; + if( sqlite3AuthCheck(pParse, SQLITE_PRAGMA, zLeft, zRight, zDb) ){ + goto pragma_out; + } + + /* Send an SQLITE_FCNTL_PRAGMA file-control to the underlying VFS + ** connection. If it returns SQLITE_OK, then assume that the VFS + ** handled the pragma and generate a no-op prepared statement. + ** + ** IMPLEMENTATION-OF: R-12238-55120 Whenever a PRAGMA statement is parsed, + ** an SQLITE_FCNTL_PRAGMA file control is sent to the open sqlite3_file + ** object corresponding to the database file to which the pragma + ** statement refers. + ** + ** IMPLEMENTATION-OF: R-29875-31678 The argument to the SQLITE_FCNTL_PRAGMA + ** file control is an array of pointers to strings (char**) in which the + ** second element of the array is the name of the pragma and the third + ** element is the argument to the pragma or NULL if the pragma has no + ** argument. + */ + aFcntl[0] = 0; + aFcntl[1] = zLeft; + aFcntl[2] = zRight; + aFcntl[3] = 0; + db->busyHandler.nBusy = 0; + rc = sqlite3_file_control(db, zDb, SQLITE_FCNTL_PRAGMA, (void*)aFcntl); + if( rc==SQLITE_OK ){ + sqlite3VdbeSetNumCols(v, 1); + sqlite3VdbeSetColName(v, 0, COLNAME_NAME, aFcntl[0], SQLITE_TRANSIENT); + returnSingleText(v, aFcntl[0]); + sqlite3_free(aFcntl[0]); + goto pragma_out; + } + if( rc!=SQLITE_NOTFOUND ){ + if( aFcntl[0] ){ + sqlite3ErrorMsg(pParse, "%s", aFcntl[0]); + sqlite3_free(aFcntl[0]); + } + pParse->nErr++; + pParse->rc = rc; + goto pragma_out; + } + + /* Locate the pragma in the lookup table */ + pPragma = pragmaLocate(zLeft); + if( pPragma==0 ) goto pragma_out; + + /* Make sure the database schema is loaded if the pragma requires that */ + if( (pPragma->mPragFlg & PragFlg_NeedSchema)!=0 ){ + if( sqlite3ReadSchema(pParse) ) goto pragma_out; + } + + /* Register the result column names for pragmas that return results */ + if( (pPragma->mPragFlg & PragFlg_NoColumns)==0 + && ((pPragma->mPragFlg & PragFlg_NoColumns1)==0 || zRight==0) + ){ + setPragmaResultColumnNames(v, pPragma); + } + + /* Jump to the appropriate pragma handler */ + switch( pPragma->ePragTyp ){ + +#if !defined(SQLITE_OMIT_PAGER_PRAGMAS) && !defined(SQLITE_OMIT_DEPRECATED) + /* + ** PRAGMA [schema.]default_cache_size + ** PRAGMA [schema.]default_cache_size=N + ** + ** The first form reports the current persistent setting for the + ** page cache size. The value returned is the maximum number of + ** pages in the page cache. The second form sets both the current + ** page cache size value and the persistent page cache size value + ** stored in the database file. + ** + ** Older versions of SQLite would set the default cache size to a + ** negative number to indicate synchronous=OFF. These days, synchronous + ** is always on by default regardless of the sign of the default cache + ** size. But continue to take the absolute value of the default cache + ** size of historical compatibility. + */ + case PragTyp_DEFAULT_CACHE_SIZE: { + static const int iLn = VDBE_OFFSET_LINENO(2); + static const VdbeOpList getCacheSize[] = { + { OP_Transaction, 0, 0, 0}, /* 0 */ + { OP_ReadCookie, 0, 1, BTREE_DEFAULT_CACHE_SIZE}, /* 1 */ + { OP_IfPos, 1, 8, 0}, + { OP_Integer, 0, 2, 0}, + { OP_Subtract, 1, 2, 1}, + { OP_IfPos, 1, 8, 0}, + { OP_Integer, 0, 1, 0}, /* 6 */ + { OP_Noop, 0, 0, 0}, + { OP_ResultRow, 1, 1, 0}, + }; + VdbeOp *aOp; + sqlite3VdbeUsesBtree(v, iDb); + if( !zRight ){ + pParse->nMem += 2; + sqlite3VdbeVerifyNoMallocRequired(v, ArraySize(getCacheSize)); + aOp = sqlite3VdbeAddOpList(v, ArraySize(getCacheSize), getCacheSize, iLn); + if( ONLY_IF_REALLOC_STRESS(aOp==0) ) break; + aOp[0].p1 = iDb; + aOp[1].p1 = iDb; + aOp[6].p1 = SQLITE_DEFAULT_CACHE_SIZE; + }else{ + int size = sqlite3AbsInt32(sqlite3Atoi(zRight)); + sqlite3BeginWriteOperation(pParse, 0, iDb); + sqlite3VdbeAddOp3(v, OP_SetCookie, iDb, BTREE_DEFAULT_CACHE_SIZE, size); + assert( sqlite3SchemaMutexHeld(db, iDb, 0) ); + pDb->pSchema->cache_size = size; + sqlite3BtreeSetCacheSize(pDb->pBt, pDb->pSchema->cache_size); + } + break; + } +#endif /* !SQLITE_OMIT_PAGER_PRAGMAS && !SQLITE_OMIT_DEPRECATED */ + +#if !defined(SQLITE_OMIT_PAGER_PRAGMAS) + /* + ** PRAGMA [schema.]page_size + ** PRAGMA [schema.]page_size=N + ** + ** The first form reports the current setting for the + ** database page size in bytes. The second form sets the + ** database page size value. The value can only be set if + ** the database has not yet been created. + */ + case PragTyp_PAGE_SIZE: { + Btree *pBt = pDb->pBt; + assert( pBt!=0 ); + if( !zRight ){ + int size = ALWAYS(pBt) ? sqlite3BtreeGetPageSize(pBt) : 0; + returnSingleInt(v, size); + }else{ + /* Malloc may fail when setting the page-size, as there is an internal + ** buffer that the pager module resizes using sqlite3_realloc(). + */ + db->nextPagesize = sqlite3Atoi(zRight); + if( SQLITE_NOMEM==sqlite3BtreeSetPageSize(pBt, db->nextPagesize,-1,0) ){ + sqlite3OomFault(db); + } + } + break; + } + + /* + ** PRAGMA [schema.]secure_delete + ** PRAGMA [schema.]secure_delete=ON/OFF/FAST + ** + ** The first form reports the current setting for the + ** secure_delete flag. The second form changes the secure_delete + ** flag setting and reports the new value. + */ + case PragTyp_SECURE_DELETE: { + Btree *pBt = pDb->pBt; + int b = -1; + assert( pBt!=0 ); + if( zRight ){ + if( sqlite3_stricmp(zRight, "fast")==0 ){ + b = 2; + }else{ + b = sqlite3GetBoolean(zRight, 0); + } + } + if( pId2->n==0 && b>=0 ){ + int ii; + for(ii=0; iinDb; ii++){ + sqlite3BtreeSecureDelete(db->aDb[ii].pBt, b); + } + } + b = sqlite3BtreeSecureDelete(pBt, b); + returnSingleInt(v, b); + break; + } + + /* + ** PRAGMA [schema.]max_page_count + ** PRAGMA [schema.]max_page_count=N + ** + ** The first form reports the current setting for the + ** maximum number of pages in the database file. The + ** second form attempts to change this setting. Both + ** forms return the current setting. + ** + ** The absolute value of N is used. This is undocumented and might + ** change. The only purpose is to provide an easy way to test + ** the sqlite3AbsInt32() function. + ** + ** PRAGMA [schema.]page_count + ** + ** Return the number of pages in the specified database. + */ + case PragTyp_PAGE_COUNT: { + int iReg; + sqlite3CodeVerifySchema(pParse, iDb); + iReg = ++pParse->nMem; + if( sqlite3Tolower(zLeft[0])=='p' ){ + sqlite3VdbeAddOp2(v, OP_Pagecount, iDb, iReg); + }else{ + sqlite3VdbeAddOp3(v, OP_MaxPgcnt, iDb, iReg, + sqlite3AbsInt32(sqlite3Atoi(zRight))); + } + sqlite3VdbeAddOp2(v, OP_ResultRow, iReg, 1); + break; + } + + /* + ** PRAGMA [schema.]locking_mode + ** PRAGMA [schema.]locking_mode = (normal|exclusive) + */ + case PragTyp_LOCKING_MODE: { + const char *zRet = "normal"; + int eMode = getLockingMode(zRight); + + if( pId2->n==0 && eMode==PAGER_LOCKINGMODE_QUERY ){ + /* Simple "PRAGMA locking_mode;" statement. This is a query for + ** the current default locking mode (which may be different to + ** the locking-mode of the main database). + */ + eMode = db->dfltLockMode; + }else{ + Pager *pPager; + if( pId2->n==0 ){ + /* This indicates that no database name was specified as part + ** of the PRAGMA command. In this case the locking-mode must be + ** set on all attached databases, as well as the main db file. + ** + ** Also, the sqlite3.dfltLockMode variable is set so that + ** any subsequently attached databases also use the specified + ** locking mode. + */ + int ii; + assert(pDb==&db->aDb[0]); + for(ii=2; iinDb; ii++){ + pPager = sqlite3BtreePager(db->aDb[ii].pBt); + sqlite3PagerLockingMode(pPager, eMode); + } + db->dfltLockMode = (u8)eMode; + } + pPager = sqlite3BtreePager(pDb->pBt); + eMode = sqlite3PagerLockingMode(pPager, eMode); + } + + assert( eMode==PAGER_LOCKINGMODE_NORMAL + || eMode==PAGER_LOCKINGMODE_EXCLUSIVE ); + if( eMode==PAGER_LOCKINGMODE_EXCLUSIVE ){ + zRet = "exclusive"; + } + returnSingleText(v, zRet); + break; + } + + /* + ** PRAGMA [schema.]journal_mode + ** PRAGMA [schema.]journal_mode = + ** (delete|persist|off|truncate|memory|wal|off) + */ + case PragTyp_JOURNAL_MODE: { + int eMode; /* One of the PAGER_JOURNALMODE_XXX symbols */ + int ii; /* Loop counter */ + + if( zRight==0 ){ + /* If there is no "=MODE" part of the pragma, do a query for the + ** current mode */ + eMode = PAGER_JOURNALMODE_QUERY; + }else{ + const char *zMode; + int n = sqlite3Strlen30(zRight); + for(eMode=0; (zMode = sqlite3JournalModename(eMode))!=0; eMode++){ + if( sqlite3StrNICmp(zRight, zMode, n)==0 ) break; + } + if( !zMode ){ + /* If the "=MODE" part does not match any known journal mode, + ** then do a query */ + eMode = PAGER_JOURNALMODE_QUERY; + } + if( eMode==PAGER_JOURNALMODE_OFF && (db->flags & SQLITE_Defensive)!=0 ){ + /* Do not allow journal-mode "OFF" in defensive since the database + ** can become corrupted using ordinary SQL when the journal is off */ + eMode = PAGER_JOURNALMODE_QUERY; + } + } + if( eMode==PAGER_JOURNALMODE_QUERY && pId2->n==0 ){ + /* Convert "PRAGMA journal_mode" into "PRAGMA main.journal_mode" */ + iDb = 0; + pId2->n = 1; + } + for(ii=db->nDb-1; ii>=0; ii--){ + if( db->aDb[ii].pBt && (ii==iDb || pId2->n==0) ){ + sqlite3VdbeUsesBtree(v, ii); + sqlite3VdbeAddOp3(v, OP_JournalMode, ii, 1, eMode); + } + } + sqlite3VdbeAddOp2(v, OP_ResultRow, 1, 1); + break; + } + + /* + ** PRAGMA [schema.]journal_size_limit + ** PRAGMA [schema.]journal_size_limit=N + ** + ** Get or set the size limit on rollback journal files. + */ + case PragTyp_JOURNAL_SIZE_LIMIT: { + Pager *pPager = sqlite3BtreePager(pDb->pBt); + i64 iLimit = -2; + if( zRight ){ + sqlite3DecOrHexToI64(zRight, &iLimit); + if( iLimit<-1 ) iLimit = -1; + } + iLimit = sqlite3PagerJournalSizeLimit(pPager, iLimit); + returnSingleInt(v, iLimit); + break; + } + +#endif /* SQLITE_OMIT_PAGER_PRAGMAS */ + + /* + ** PRAGMA [schema.]auto_vacuum + ** PRAGMA [schema.]auto_vacuum=N + ** + ** Get or set the value of the database 'auto-vacuum' parameter. + ** The value is one of: 0 NONE 1 FULL 2 INCREMENTAL + */ +#ifndef SQLITE_OMIT_AUTOVACUUM + case PragTyp_AUTO_VACUUM: { + Btree *pBt = pDb->pBt; + assert( pBt!=0 ); + if( !zRight ){ + returnSingleInt(v, sqlite3BtreeGetAutoVacuum(pBt)); + }else{ + int eAuto = getAutoVacuum(zRight); + assert( eAuto>=0 && eAuto<=2 ); + db->nextAutovac = (u8)eAuto; + /* Call SetAutoVacuum() to set initialize the internal auto and + ** incr-vacuum flags. This is required in case this connection + ** creates the database file. It is important that it is created + ** as an auto-vacuum capable db. + */ + rc = sqlite3BtreeSetAutoVacuum(pBt, eAuto); + if( rc==SQLITE_OK && (eAuto==1 || eAuto==2) ){ + /* When setting the auto_vacuum mode to either "full" or + ** "incremental", write the value of meta[6] in the database + ** file. Before writing to meta[6], check that meta[3] indicates + ** that this really is an auto-vacuum capable database. + */ + static const int iLn = VDBE_OFFSET_LINENO(2); + static const VdbeOpList setMeta6[] = { + { OP_Transaction, 0, 1, 0}, /* 0 */ + { OP_ReadCookie, 0, 1, BTREE_LARGEST_ROOT_PAGE}, + { OP_If, 1, 0, 0}, /* 2 */ + { OP_Halt, SQLITE_OK, OE_Abort, 0}, /* 3 */ + { OP_SetCookie, 0, BTREE_INCR_VACUUM, 0}, /* 4 */ + }; + VdbeOp *aOp; + int iAddr = sqlite3VdbeCurrentAddr(v); + sqlite3VdbeVerifyNoMallocRequired(v, ArraySize(setMeta6)); + aOp = sqlite3VdbeAddOpList(v, ArraySize(setMeta6), setMeta6, iLn); + if( ONLY_IF_REALLOC_STRESS(aOp==0) ) break; + aOp[0].p1 = iDb; + aOp[1].p1 = iDb; + aOp[2].p2 = iAddr+4; + aOp[4].p1 = iDb; + aOp[4].p3 = eAuto - 1; + sqlite3VdbeUsesBtree(v, iDb); + } + } + break; + } +#endif + + /* + ** PRAGMA [schema.]incremental_vacuum(N) + ** + ** Do N steps of incremental vacuuming on a database. + */ +#ifndef SQLITE_OMIT_AUTOVACUUM + case PragTyp_INCREMENTAL_VACUUM: { + int iLimit, addr; + if( zRight==0 || !sqlite3GetInt32(zRight, &iLimit) || iLimit<=0 ){ + iLimit = 0x7fffffff; + } + sqlite3BeginWriteOperation(pParse, 0, iDb); + sqlite3VdbeAddOp2(v, OP_Integer, iLimit, 1); + addr = sqlite3VdbeAddOp1(v, OP_IncrVacuum, iDb); VdbeCoverage(v); + sqlite3VdbeAddOp1(v, OP_ResultRow, 1); + sqlite3VdbeAddOp2(v, OP_AddImm, 1, -1); + sqlite3VdbeAddOp2(v, OP_IfPos, 1, addr); VdbeCoverage(v); + sqlite3VdbeJumpHere(v, addr); + break; + } +#endif + +#ifndef SQLITE_OMIT_PAGER_PRAGMAS + /* + ** PRAGMA [schema.]cache_size + ** PRAGMA [schema.]cache_size=N + ** + ** The first form reports the current local setting for the + ** page cache size. The second form sets the local + ** page cache size value. If N is positive then that is the + ** number of pages in the cache. If N is negative, then the + ** number of pages is adjusted so that the cache uses -N kibibytes + ** of memory. + */ + case PragTyp_CACHE_SIZE: { + assert( sqlite3SchemaMutexHeld(db, iDb, 0) ); + if( !zRight ){ + returnSingleInt(v, pDb->pSchema->cache_size); + }else{ + int size = sqlite3Atoi(zRight); + pDb->pSchema->cache_size = size; + sqlite3BtreeSetCacheSize(pDb->pBt, pDb->pSchema->cache_size); + } + break; + } + + /* + ** PRAGMA [schema.]cache_spill + ** PRAGMA cache_spill=BOOLEAN + ** PRAGMA [schema.]cache_spill=N + ** + ** The first form reports the current local setting for the + ** page cache spill size. The second form turns cache spill on + ** or off. When turnning cache spill on, the size is set to the + ** current cache_size. The third form sets a spill size that + ** may be different form the cache size. + ** If N is positive then that is the + ** number of pages in the cache. If N is negative, then the + ** number of pages is adjusted so that the cache uses -N kibibytes + ** of memory. + ** + ** If the number of cache_spill pages is less then the number of + ** cache_size pages, no spilling occurs until the page count exceeds + ** the number of cache_size pages. + ** + ** The cache_spill=BOOLEAN setting applies to all attached schemas, + ** not just the schema specified. + */ + case PragTyp_CACHE_SPILL: { + assert( sqlite3SchemaMutexHeld(db, iDb, 0) ); + if( !zRight ){ + returnSingleInt(v, + (db->flags & SQLITE_CacheSpill)==0 ? 0 : + sqlite3BtreeSetSpillSize(pDb->pBt,0)); + }else{ + int size = 1; + if( sqlite3GetInt32(zRight, &size) ){ + sqlite3BtreeSetSpillSize(pDb->pBt, size); + } + if( sqlite3GetBoolean(zRight, size!=0) ){ + db->flags |= SQLITE_CacheSpill; + }else{ + db->flags &= ~(u64)SQLITE_CacheSpill; + } + setAllPagerFlags(db); + } + break; + } + + /* + ** PRAGMA [schema.]mmap_size(N) + ** + ** Used to set mapping size limit. The mapping size limit is + ** used to limit the aggregate size of all memory mapped regions of the + ** database file. If this parameter is set to zero, then memory mapping + ** is not used at all. If N is negative, then the default memory map + ** limit determined by sqlite3_config(SQLITE_CONFIG_MMAP_SIZE) is set. + ** The parameter N is measured in bytes. + ** + ** This value is advisory. The underlying VFS is free to memory map + ** as little or as much as it wants. Except, if N is set to 0 then the + ** upper layers will never invoke the xFetch interfaces to the VFS. + */ + case PragTyp_MMAP_SIZE: { + sqlite3_int64 sz; +#if SQLITE_MAX_MMAP_SIZE>0 + assert( sqlite3SchemaMutexHeld(db, iDb, 0) ); + if( zRight ){ + int ii; + sqlite3DecOrHexToI64(zRight, &sz); + if( sz<0 ) sz = sqlite3GlobalConfig.szMmap; + if( pId2->n==0 ) db->szMmap = sz; + for(ii=db->nDb-1; ii>=0; ii--){ + if( db->aDb[ii].pBt && (ii==iDb || pId2->n==0) ){ + sqlite3BtreeSetMmapLimit(db->aDb[ii].pBt, sz); + } + } + } + sz = -1; + rc = sqlite3_file_control(db, zDb, SQLITE_FCNTL_MMAP_SIZE, &sz); +#else + sz = 0; + rc = SQLITE_OK; +#endif + if( rc==SQLITE_OK ){ + returnSingleInt(v, sz); + }else if( rc!=SQLITE_NOTFOUND ){ + pParse->nErr++; + pParse->rc = rc; + } + break; + } + + /* + ** PRAGMA temp_store + ** PRAGMA temp_store = "default"|"memory"|"file" + ** + ** Return or set the local value of the temp_store flag. Changing + ** the local value does not make changes to the disk file and the default + ** value will be restored the next time the database is opened. + ** + ** Note that it is possible for the library compile-time options to + ** override this setting + */ + case PragTyp_TEMP_STORE: { + if( !zRight ){ + returnSingleInt(v, db->temp_store); + }else{ + changeTempStorage(pParse, zRight); + } + break; + } + + /* + ** PRAGMA temp_store_directory + ** PRAGMA temp_store_directory = ""|"directory_name" + ** + ** Return or set the local value of the temp_store_directory flag. Changing + ** the value sets a specific directory to be used for temporary files. + ** Setting to a null string reverts to the default temporary directory search. + ** If temporary directory is changed, then invalidateTempStorage. + ** + */ + case PragTyp_TEMP_STORE_DIRECTORY: { + if( !zRight ){ + returnSingleText(v, sqlite3_temp_directory); + }else{ +#ifndef SQLITE_OMIT_WSD + if( zRight[0] ){ + int res; + rc = sqlite3OsAccess(db->pVfs, zRight, SQLITE_ACCESS_READWRITE, &res); + if( rc!=SQLITE_OK || res==0 ){ + sqlite3ErrorMsg(pParse, "not a writable directory"); + goto pragma_out; + } + } + if( SQLITE_TEMP_STORE==0 + || (SQLITE_TEMP_STORE==1 && db->temp_store<=1) + || (SQLITE_TEMP_STORE==2 && db->temp_store==1) + ){ + invalidateTempStorage(pParse); + } + sqlite3_free(sqlite3_temp_directory); + if( zRight[0] ){ + sqlite3_temp_directory = sqlite3_mprintf("%s", zRight); + }else{ + sqlite3_temp_directory = 0; + } +#endif /* SQLITE_OMIT_WSD */ + } + break; + } + +#if SQLITE_OS_WIN + /* + ** PRAGMA data_store_directory + ** PRAGMA data_store_directory = ""|"directory_name" + ** + ** Return or set the local value of the data_store_directory flag. Changing + ** the value sets a specific directory to be used for database files that + ** were specified with a relative pathname. Setting to a null string reverts + ** to the default database directory, which for database files specified with + ** a relative path will probably be based on the current directory for the + ** process. Database file specified with an absolute path are not impacted + ** by this setting, regardless of its value. + ** + */ + case PragTyp_DATA_STORE_DIRECTORY: { + if( !zRight ){ + returnSingleText(v, sqlite3_data_directory); + }else{ +#ifndef SQLITE_OMIT_WSD + if( zRight[0] ){ + int res; + rc = sqlite3OsAccess(db->pVfs, zRight, SQLITE_ACCESS_READWRITE, &res); + if( rc!=SQLITE_OK || res==0 ){ + sqlite3ErrorMsg(pParse, "not a writable directory"); + goto pragma_out; + } + } + sqlite3_free(sqlite3_data_directory); + if( zRight[0] ){ + sqlite3_data_directory = sqlite3_mprintf("%s", zRight); + }else{ + sqlite3_data_directory = 0; + } +#endif /* SQLITE_OMIT_WSD */ + } + break; + } +#endif + +#if SQLITE_ENABLE_LOCKING_STYLE + /* + ** PRAGMA [schema.]lock_proxy_file + ** PRAGMA [schema.]lock_proxy_file = ":auto:"|"lock_file_path" + ** + ** Return or set the value of the lock_proxy_file flag. Changing + ** the value sets a specific file to be used for database access locks. + ** + */ + case PragTyp_LOCK_PROXY_FILE: { + if( !zRight ){ + Pager *pPager = sqlite3BtreePager(pDb->pBt); + char *proxy_file_path = NULL; + sqlite3_file *pFile = sqlite3PagerFile(pPager); + sqlite3OsFileControlHint(pFile, SQLITE_GET_LOCKPROXYFILE, + &proxy_file_path); + returnSingleText(v, proxy_file_path); + }else{ + Pager *pPager = sqlite3BtreePager(pDb->pBt); + sqlite3_file *pFile = sqlite3PagerFile(pPager); + int res; + if( zRight[0] ){ + res=sqlite3OsFileControl(pFile, SQLITE_SET_LOCKPROXYFILE, + zRight); + } else { + res=sqlite3OsFileControl(pFile, SQLITE_SET_LOCKPROXYFILE, + NULL); + } + if( res!=SQLITE_OK ){ + sqlite3ErrorMsg(pParse, "failed to set lock proxy file"); + goto pragma_out; + } + } + break; + } +#endif /* SQLITE_ENABLE_LOCKING_STYLE */ + + /* + ** PRAGMA [schema.]synchronous + ** PRAGMA [schema.]synchronous=OFF|ON|NORMAL|FULL|EXTRA + ** + ** Return or set the local value of the synchronous flag. Changing + ** the local value does not make changes to the disk file and the + ** default value will be restored the next time the database is + ** opened. + */ + case PragTyp_SYNCHRONOUS: { + if( !zRight ){ + returnSingleInt(v, pDb->safety_level-1); + }else{ + if( !db->autoCommit ){ + sqlite3ErrorMsg(pParse, + "Safety level may not be changed inside a transaction"); + }else if( iDb!=1 ){ + int iLevel = (getSafetyLevel(zRight,0,1)+1) & PAGER_SYNCHRONOUS_MASK; + if( iLevel==0 ) iLevel = 1; + pDb->safety_level = iLevel; + pDb->bSyncSet = 1; + setAllPagerFlags(db); + } + } + break; + } +#endif /* SQLITE_OMIT_PAGER_PRAGMAS */ + +#ifndef SQLITE_OMIT_FLAG_PRAGMAS + case PragTyp_FLAG: { + if( zRight==0 ){ + setPragmaResultColumnNames(v, pPragma); + returnSingleInt(v, (db->flags & pPragma->iArg)!=0 ); + }else{ + u64 mask = pPragma->iArg; /* Mask of bits to set or clear. */ + if( db->autoCommit==0 ){ + /* Foreign key support may not be enabled or disabled while not + ** in auto-commit mode. */ + mask &= ~(SQLITE_ForeignKeys); + } +#if SQLITE_USER_AUTHENTICATION + if( db->auth.authLevel==UAUTH_User ){ + /* Do not allow non-admin users to modify the schema arbitrarily */ + mask &= ~(SQLITE_WriteSchema); + } +#endif + + if( sqlite3GetBoolean(zRight, 0) ){ + db->flags |= mask; + }else{ + db->flags &= ~mask; + if( mask==SQLITE_DeferFKs ) db->nDeferredImmCons = 0; + } + + /* Many of the flag-pragmas modify the code generated by the SQL + ** compiler (eg. count_changes). So add an opcode to expire all + ** compiled SQL statements after modifying a pragma value. + */ + sqlite3VdbeAddOp0(v, OP_Expire); + setAllPagerFlags(db); + } + break; + } +#endif /* SQLITE_OMIT_FLAG_PRAGMAS */ + +#ifndef SQLITE_OMIT_SCHEMA_PRAGMAS + /* + ** PRAGMA table_info(
    ) + ** + ** Return a single row for each column of the named table. The columns of + ** the returned data set are: + ** + ** cid: Column id (numbered from left to right, starting at 0) + ** name: Column name + ** type: Column declaration type. + ** notnull: True if 'NOT NULL' is part of column declaration + ** dflt_value: The default value for the column, if any. + ** pk: Non-zero for PK fields. + */ + case PragTyp_TABLE_INFO: if( zRight ){ + Table *pTab; + pTab = sqlite3LocateTable(pParse, LOCATE_NOERR, zRight, zDb); + if( pTab ){ + int iTabDb = sqlite3SchemaToIndex(db, pTab->pSchema); + int i, k; + int nHidden = 0; + Column *pCol; + Index *pPk = sqlite3PrimaryKeyIndex(pTab); + pParse->nMem = 7; + sqlite3CodeVerifySchema(pParse, iTabDb); + sqlite3ViewGetColumnNames(pParse, pTab); + for(i=0, pCol=pTab->aCol; inCol; i++, pCol++){ + int isHidden = IsHiddenColumn(pCol); + if( isHidden && pPragma->iArg==0 ){ + nHidden++; + continue; + } + if( (pCol->colFlags & COLFLAG_PRIMKEY)==0 ){ + k = 0; + }else if( pPk==0 ){ + k = 1; + }else{ + for(k=1; k<=pTab->nCol && pPk->aiColumn[k-1]!=i; k++){} + } + assert( pCol->pDflt==0 || pCol->pDflt->op==TK_SPAN ); + sqlite3VdbeMultiLoad(v, 1, pPragma->iArg ? "issisii" : "issisi", + i-nHidden, + pCol->zName, + sqlite3ColumnType(pCol,""), + pCol->notNull ? 1 : 0, + pCol->pDflt ? pCol->pDflt->u.zToken : 0, + k, + isHidden); + } + } + } + break; + +#ifdef SQLITE_DEBUG + case PragTyp_STATS: { + Index *pIdx; + HashElem *i; + pParse->nMem = 5; + sqlite3CodeVerifySchema(pParse, iDb); + for(i=sqliteHashFirst(&pDb->pSchema->tblHash); i; i=sqliteHashNext(i)){ + Table *pTab = sqliteHashData(i); + sqlite3VdbeMultiLoad(v, 1, "ssiii", + pTab->zName, + 0, + pTab->szTabRow, + pTab->nRowLogEst, + pTab->tabFlags); + for(pIdx=pTab->pIndex; pIdx; pIdx=pIdx->pNext){ + sqlite3VdbeMultiLoad(v, 2, "siiiX", + pIdx->zName, + pIdx->szIdxRow, + pIdx->aiRowLogEst[0], + pIdx->hasStat1); + sqlite3VdbeAddOp2(v, OP_ResultRow, 1, 5); + } + } + } + break; +#endif + + case PragTyp_INDEX_INFO: if( zRight ){ + Index *pIdx; + Table *pTab; + pIdx = sqlite3FindIndex(db, zRight, zDb); + if( pIdx==0 ){ + /* If there is no index named zRight, check to see if there is a + ** WITHOUT ROWID table named zRight, and if there is, show the + ** structure of the PRIMARY KEY index for that table. */ + pTab = sqlite3LocateTable(pParse, LOCATE_NOERR, zRight, zDb); + if( pTab && !HasRowid(pTab) ){ + pIdx = sqlite3PrimaryKeyIndex(pTab); + } + } + if( pIdx ){ + int iIdxDb = sqlite3SchemaToIndex(db, pIdx->pSchema); + int i; + int mx; + if( pPragma->iArg ){ + /* PRAGMA index_xinfo (newer version with more rows and columns) */ + mx = pIdx->nColumn; + pParse->nMem = 6; + }else{ + /* PRAGMA index_info (legacy version) */ + mx = pIdx->nKeyCol; + pParse->nMem = 3; + } + pTab = pIdx->pTable; + sqlite3CodeVerifySchema(pParse, iIdxDb); + assert( pParse->nMem<=pPragma->nPragCName ); + for(i=0; iaiColumn[i]; + sqlite3VdbeMultiLoad(v, 1, "iisX", i, cnum, + cnum<0 ? 0 : pTab->aCol[cnum].zName); + if( pPragma->iArg ){ + sqlite3VdbeMultiLoad(v, 4, "isiX", + pIdx->aSortOrder[i], + pIdx->azColl[i], + inKeyCol); + } + sqlite3VdbeAddOp2(v, OP_ResultRow, 1, pParse->nMem); + } + } + } + break; + + case PragTyp_INDEX_LIST: if( zRight ){ + Index *pIdx; + Table *pTab; + int i; + pTab = sqlite3FindTable(db, zRight, zDb); + if( pTab ){ + int iTabDb = sqlite3SchemaToIndex(db, pTab->pSchema); + pParse->nMem = 5; + sqlite3CodeVerifySchema(pParse, iTabDb); + for(pIdx=pTab->pIndex, i=0; pIdx; pIdx=pIdx->pNext, i++){ + const char *azOrigin[] = { "c", "u", "pk" }; + sqlite3VdbeMultiLoad(v, 1, "isisi", + i, + pIdx->zName, + IsUniqueIndex(pIdx), + azOrigin[pIdx->idxType], + pIdx->pPartIdxWhere!=0); + } + } + } + break; + + case PragTyp_DATABASE_LIST: { + int i; + pParse->nMem = 3; + for(i=0; inDb; i++){ + if( db->aDb[i].pBt==0 ) continue; + assert( db->aDb[i].zDbSName!=0 ); + sqlite3VdbeMultiLoad(v, 1, "iss", + i, + db->aDb[i].zDbSName, + sqlite3BtreeGetFilename(db->aDb[i].pBt)); + } + } + break; + + case PragTyp_COLLATION_LIST: { + int i = 0; + HashElem *p; + pParse->nMem = 2; + for(p=sqliteHashFirst(&db->aCollSeq); p; p=sqliteHashNext(p)){ + CollSeq *pColl = (CollSeq *)sqliteHashData(p); + sqlite3VdbeMultiLoad(v, 1, "is", i++, pColl->zName); + } + } + break; + +#ifndef SQLITE_OMIT_INTROSPECTION_PRAGMAS + case PragTyp_FUNCTION_LIST: { + int i; + HashElem *j; + FuncDef *p; + pParse->nMem = 2; + for(i=0; iu.pHash ){ + if( p->funcFlags & SQLITE_FUNC_INTERNAL ) continue; + sqlite3VdbeMultiLoad(v, 1, "si", p->zName, 1); + } + } + for(j=sqliteHashFirst(&db->aFunc); j; j=sqliteHashNext(j)){ + p = (FuncDef*)sqliteHashData(j); + sqlite3VdbeMultiLoad(v, 1, "si", p->zName, 0); + } + } + break; + +#ifndef SQLITE_OMIT_VIRTUALTABLE + case PragTyp_MODULE_LIST: { + HashElem *j; + pParse->nMem = 1; + for(j=sqliteHashFirst(&db->aModule); j; j=sqliteHashNext(j)){ + Module *pMod = (Module*)sqliteHashData(j); + sqlite3VdbeMultiLoad(v, 1, "s", pMod->zName); + } + } + break; +#endif /* SQLITE_OMIT_VIRTUALTABLE */ + + case PragTyp_PRAGMA_LIST: { + int i; + for(i=0; ipFKey; + if( pFK ){ + int iTabDb = sqlite3SchemaToIndex(db, pTab->pSchema); + int i = 0; + pParse->nMem = 8; + sqlite3CodeVerifySchema(pParse, iTabDb); + while(pFK){ + int j; + for(j=0; jnCol; j++){ + sqlite3VdbeMultiLoad(v, 1, "iissssss", + i, + j, + pFK->zTo, + pTab->aCol[pFK->aCol[j].iFrom].zName, + pFK->aCol[j].zCol, + actionName(pFK->aAction[1]), /* ON UPDATE */ + actionName(pFK->aAction[0]), /* ON DELETE */ + "NONE"); + } + ++i; + pFK = pFK->pNextFrom; + } + } + } + } + break; +#endif /* !defined(SQLITE_OMIT_FOREIGN_KEY) */ + +#ifndef SQLITE_OMIT_FOREIGN_KEY +#ifndef SQLITE_OMIT_TRIGGER + case PragTyp_FOREIGN_KEY_CHECK: { + FKey *pFK; /* A foreign key constraint */ + Table *pTab; /* Child table contain "REFERENCES" keyword */ + Table *pParent; /* Parent table that child points to */ + Index *pIdx; /* Index in the parent table */ + int i; /* Loop counter: Foreign key number for pTab */ + int j; /* Loop counter: Field of the foreign key */ + HashElem *k; /* Loop counter: Next table in schema */ + int x; /* result variable */ + int regResult; /* 3 registers to hold a result row */ + int regKey; /* Register to hold key for checking the FK */ + int regRow; /* Registers to hold a row from pTab */ + int addrTop; /* Top of a loop checking foreign keys */ + int addrOk; /* Jump here if the key is OK */ + int *aiCols; /* child to parent column mapping */ + + regResult = pParse->nMem+1; + pParse->nMem += 4; + regKey = ++pParse->nMem; + regRow = ++pParse->nMem; + k = sqliteHashFirst(&db->aDb[iDb].pSchema->tblHash); + while( k ){ + int iTabDb; + if( zRight ){ + pTab = sqlite3LocateTable(pParse, 0, zRight, zDb); + k = 0; + }else{ + pTab = (Table*)sqliteHashData(k); + k = sqliteHashNext(k); + } + if( pTab==0 || pTab->pFKey==0 ) continue; + iTabDb = sqlite3SchemaToIndex(db, pTab->pSchema); + sqlite3CodeVerifySchema(pParse, iTabDb); + sqlite3TableLock(pParse, iTabDb, pTab->tnum, 0, pTab->zName); + if( pTab->nCol+regRow>pParse->nMem ) pParse->nMem = pTab->nCol + regRow; + sqlite3OpenTable(pParse, 0, iTabDb, pTab, OP_OpenRead); + sqlite3VdbeLoadString(v, regResult, pTab->zName); + for(i=1, pFK=pTab->pFKey; pFK; i++, pFK=pFK->pNextFrom){ + pParent = sqlite3FindTable(db, pFK->zTo, zDb); + if( pParent==0 ) continue; + pIdx = 0; + sqlite3TableLock(pParse, iTabDb, pParent->tnum, 0, pParent->zName); + x = sqlite3FkLocateIndex(pParse, pParent, pFK, &pIdx, 0); + if( x==0 ){ + if( pIdx==0 ){ + sqlite3OpenTable(pParse, i, iTabDb, pParent, OP_OpenRead); + }else{ + sqlite3VdbeAddOp3(v, OP_OpenRead, i, pIdx->tnum, iTabDb); + sqlite3VdbeSetP4KeyInfo(pParse, pIdx); + } + }else{ + k = 0; + break; + } + } + assert( pParse->nErr>0 || pFK==0 ); + if( pFK ) break; + if( pParse->nTabnTab = i; + addrTop = sqlite3VdbeAddOp1(v, OP_Rewind, 0); VdbeCoverage(v); + for(i=1, pFK=pTab->pFKey; pFK; i++, pFK=pFK->pNextFrom){ + pParent = sqlite3FindTable(db, pFK->zTo, zDb); + pIdx = 0; + aiCols = 0; + if( pParent ){ + x = sqlite3FkLocateIndex(pParse, pParent, pFK, &pIdx, &aiCols); + assert( x==0 ); + } + addrOk = sqlite3VdbeMakeLabel(pParse); + + /* Generate code to read the child key values into registers + ** regRow..regRow+n. If any of the child key values are NULL, this + ** row cannot cause an FK violation. Jump directly to addrOk in + ** this case. */ + for(j=0; jnCol; j++){ + int iCol = aiCols ? aiCols[j] : pFK->aCol[j].iFrom; + sqlite3ExprCodeGetColumnOfTable(v, pTab, 0, iCol, regRow+j); + sqlite3VdbeAddOp2(v, OP_IsNull, regRow+j, addrOk); VdbeCoverage(v); + } + + /* Generate code to query the parent index for a matching parent + ** key. If a match is found, jump to addrOk. */ + if( pIdx ){ + sqlite3VdbeAddOp4(v, OP_MakeRecord, regRow, pFK->nCol, regKey, + sqlite3IndexAffinityStr(db,pIdx), pFK->nCol); + sqlite3VdbeAddOp4Int(v, OP_Found, i, addrOk, regKey, 0); + VdbeCoverage(v); + }else if( pParent ){ + int jmp = sqlite3VdbeCurrentAddr(v)+2; + sqlite3VdbeAddOp3(v, OP_SeekRowid, i, jmp, regRow); VdbeCoverage(v); + sqlite3VdbeGoto(v, addrOk); + assert( pFK->nCol==1 ); + } + + /* Generate code to report an FK violation to the caller. */ + if( HasRowid(pTab) ){ + sqlite3VdbeAddOp2(v, OP_Rowid, 0, regResult+1); + }else{ + sqlite3VdbeAddOp2(v, OP_Null, 0, regResult+1); + } + sqlite3VdbeMultiLoad(v, regResult+2, "siX", pFK->zTo, i-1); + sqlite3VdbeAddOp2(v, OP_ResultRow, regResult, 4); + sqlite3VdbeResolveLabel(v, addrOk); + sqlite3DbFree(db, aiCols); + } + sqlite3VdbeAddOp2(v, OP_Next, 0, addrTop+1); VdbeCoverage(v); + sqlite3VdbeJumpHere(v, addrTop); + } + } + break; +#endif /* !defined(SQLITE_OMIT_TRIGGER) */ +#endif /* !defined(SQLITE_OMIT_FOREIGN_KEY) */ + +#ifndef SQLITE_OMIT_CASE_SENSITIVE_LIKE_PRAGMA + /* Reinstall the LIKE and GLOB functions. The variant of LIKE + ** used will be case sensitive or not depending on the RHS. + */ + case PragTyp_CASE_SENSITIVE_LIKE: { + if( zRight ){ + sqlite3RegisterLikeFunctions(db, sqlite3GetBoolean(zRight, 0)); + } + } + break; +#endif /* SQLITE_OMIT_CASE_SENSITIVE_LIKE_PRAGMA */ + +#ifndef SQLITE_INTEGRITY_CHECK_ERROR_MAX +# define SQLITE_INTEGRITY_CHECK_ERROR_MAX 100 +#endif + +#ifndef SQLITE_OMIT_INTEGRITY_CHECK + /* PRAGMA integrity_check + ** PRAGMA integrity_check(N) + ** PRAGMA quick_check + ** PRAGMA quick_check(N) + ** + ** Verify the integrity of the database. + ** + ** The "quick_check" is reduced version of + ** integrity_check designed to detect most database corruption + ** without the overhead of cross-checking indexes. Quick_check + ** is linear time wherease integrity_check is O(NlogN). + */ + case PragTyp_INTEGRITY_CHECK: { + int i, j, addr, mxErr; + + int isQuick = (sqlite3Tolower(zLeft[0])=='q'); + + /* If the PRAGMA command was of the form "PRAGMA .integrity_check", + ** then iDb is set to the index of the database identified by . + ** In this case, the integrity of database iDb only is verified by + ** the VDBE created below. + ** + ** Otherwise, if the command was simply "PRAGMA integrity_check" (or + ** "PRAGMA quick_check"), then iDb is set to 0. In this case, set iDb + ** to -1 here, to indicate that the VDBE should verify the integrity + ** of all attached databases. */ + assert( iDb>=0 ); + assert( iDb==0 || pId2->z ); + if( pId2->z==0 ) iDb = -1; + + /* Initialize the VDBE program */ + pParse->nMem = 6; + + /* Set the maximum error count */ + mxErr = SQLITE_INTEGRITY_CHECK_ERROR_MAX; + if( zRight ){ + sqlite3GetInt32(zRight, &mxErr); + if( mxErr<=0 ){ + mxErr = SQLITE_INTEGRITY_CHECK_ERROR_MAX; + } + } + sqlite3VdbeAddOp2(v, OP_Integer, mxErr-1, 1); /* reg[1] holds errors left */ + + /* Do an integrity check on each database file */ + for(i=0; inDb; i++){ + HashElem *x; /* For looping over tables in the schema */ + Hash *pTbls; /* Set of all tables in the schema */ + int *aRoot; /* Array of root page numbers of all btrees */ + int cnt = 0; /* Number of entries in aRoot[] */ + int mxIdx = 0; /* Maximum number of indexes for any table */ + + if( OMIT_TEMPDB && i==1 ) continue; + if( iDb>=0 && i!=iDb ) continue; + + sqlite3CodeVerifySchema(pParse, i); + + /* Do an integrity check of the B-Tree + ** + ** Begin by finding the root pages numbers + ** for all tables and indices in the database. + */ + assert( sqlite3SchemaMutexHeld(db, i, 0) ); + pTbls = &db->aDb[i].pSchema->tblHash; + for(cnt=0, x=sqliteHashFirst(pTbls); x; x=sqliteHashNext(x)){ + Table *pTab = sqliteHashData(x); /* Current table */ + Index *pIdx; /* An index on pTab */ + int nIdx; /* Number of indexes on pTab */ + if( HasRowid(pTab) ) cnt++; + for(nIdx=0, pIdx=pTab->pIndex; pIdx; pIdx=pIdx->pNext, nIdx++){ cnt++; } + if( nIdx>mxIdx ) mxIdx = nIdx; + } + aRoot = sqlite3DbMallocRawNN(db, sizeof(int)*(cnt+1)); + if( aRoot==0 ) break; + for(cnt=0, x=sqliteHashFirst(pTbls); x; x=sqliteHashNext(x)){ + Table *pTab = sqliteHashData(x); + Index *pIdx; + if( HasRowid(pTab) ) aRoot[++cnt] = pTab->tnum; + for(pIdx=pTab->pIndex; pIdx; pIdx=pIdx->pNext){ + aRoot[++cnt] = pIdx->tnum; + } + } + aRoot[0] = cnt; + + /* Make sure sufficient number of registers have been allocated */ + pParse->nMem = MAX( pParse->nMem, 8+mxIdx ); + sqlite3ClearTempRegCache(pParse); + + /* Do the b-tree integrity checks */ + sqlite3VdbeAddOp4(v, OP_IntegrityCk, 2, cnt, 1, (char*)aRoot,P4_INTARRAY); + sqlite3VdbeChangeP5(v, (u8)i); + addr = sqlite3VdbeAddOp1(v, OP_IsNull, 2); VdbeCoverage(v); + sqlite3VdbeAddOp4(v, OP_String8, 0, 3, 0, + sqlite3MPrintf(db, "*** in database %s ***\n", db->aDb[i].zDbSName), + P4_DYNAMIC); + sqlite3VdbeAddOp3(v, OP_Concat, 2, 3, 3); + integrityCheckResultRow(v); + sqlite3VdbeJumpHere(v, addr); + + /* Make sure all the indices are constructed correctly. + */ + for(x=sqliteHashFirst(pTbls); x; x=sqliteHashNext(x)){ + Table *pTab = sqliteHashData(x); + Index *pIdx, *pPk; + Index *pPrior = 0; + int loopTop; + int iDataCur, iIdxCur; + int r1 = -1; + + if( pTab->tnum<1 ) continue; /* Skip VIEWs or VIRTUAL TABLEs */ + pPk = HasRowid(pTab) ? 0 : sqlite3PrimaryKeyIndex(pTab); + sqlite3OpenTableAndIndices(pParse, pTab, OP_OpenRead, 0, + 1, 0, &iDataCur, &iIdxCur); + /* reg[7] counts the number of entries in the table. + ** reg[8+i] counts the number of entries in the i-th index + */ + sqlite3VdbeAddOp2(v, OP_Integer, 0, 7); + for(j=0, pIdx=pTab->pIndex; pIdx; pIdx=pIdx->pNext, j++){ + sqlite3VdbeAddOp2(v, OP_Integer, 0, 8+j); /* index entries counter */ + } + assert( pParse->nMem>=8+j ); + assert( sqlite3NoTempsInRange(pParse,1,7+j) ); + sqlite3VdbeAddOp2(v, OP_Rewind, iDataCur, 0); VdbeCoverage(v); + loopTop = sqlite3VdbeAddOp2(v, OP_AddImm, 7, 1); + if( !isQuick ){ + /* Sanity check on record header decoding */ + sqlite3VdbeAddOp3(v, OP_Column, iDataCur, pTab->nCol-1, 3); + sqlite3VdbeChangeP5(v, OPFLAG_TYPEOFARG); + } + /* Verify that all NOT NULL columns really are NOT NULL */ + for(j=0; jnCol; j++){ + char *zErr; + int jmp2; + if( j==pTab->iPKey ) continue; + if( pTab->aCol[j].notNull==0 ) continue; + sqlite3ExprCodeGetColumnOfTable(v, pTab, iDataCur, j, 3); + sqlite3VdbeChangeP5(v, OPFLAG_TYPEOFARG); + jmp2 = sqlite3VdbeAddOp1(v, OP_NotNull, 3); VdbeCoverage(v); + zErr = sqlite3MPrintf(db, "NULL value in %s.%s", pTab->zName, + pTab->aCol[j].zName); + sqlite3VdbeAddOp4(v, OP_String8, 0, 3, 0, zErr, P4_DYNAMIC); + integrityCheckResultRow(v); + sqlite3VdbeJumpHere(v, jmp2); + } + /* Verify CHECK constraints */ + if( pTab->pCheck && (db->flags & SQLITE_IgnoreChecks)==0 ){ + ExprList *pCheck = sqlite3ExprListDup(db, pTab->pCheck, 0); + if( db->mallocFailed==0 ){ + int addrCkFault = sqlite3VdbeMakeLabel(pParse); + int addrCkOk = sqlite3VdbeMakeLabel(pParse); + char *zErr; + int k; + pParse->iSelfTab = iDataCur + 1; + for(k=pCheck->nExpr-1; k>0; k--){ + sqlite3ExprIfFalse(pParse, pCheck->a[k].pExpr, addrCkFault, 0); + } + sqlite3ExprIfTrue(pParse, pCheck->a[0].pExpr, addrCkOk, + SQLITE_JUMPIFNULL); + sqlite3VdbeResolveLabel(v, addrCkFault); + pParse->iSelfTab = 0; + zErr = sqlite3MPrintf(db, "CHECK constraint failed in %s", + pTab->zName); + sqlite3VdbeAddOp4(v, OP_String8, 0, 3, 0, zErr, P4_DYNAMIC); + integrityCheckResultRow(v); + sqlite3VdbeResolveLabel(v, addrCkOk); + } + sqlite3ExprListDelete(db, pCheck); + } + if( !isQuick ){ /* Omit the remaining tests for quick_check */ + /* Validate index entries for the current row */ + for(j=0, pIdx=pTab->pIndex; pIdx; pIdx=pIdx->pNext, j++){ + int jmp2, jmp3, jmp4, jmp5; + int ckUniq = sqlite3VdbeMakeLabel(pParse); + if( pPk==pIdx ) continue; + r1 = sqlite3GenerateIndexKey(pParse, pIdx, iDataCur, 0, 0, &jmp3, + pPrior, r1); + pPrior = pIdx; + sqlite3VdbeAddOp2(v, OP_AddImm, 8+j, 1);/* increment entry count */ + /* Verify that an index entry exists for the current table row */ + jmp2 = sqlite3VdbeAddOp4Int(v, OP_Found, iIdxCur+j, ckUniq, r1, + pIdx->nColumn); VdbeCoverage(v); + sqlite3VdbeLoadString(v, 3, "row "); + sqlite3VdbeAddOp3(v, OP_Concat, 7, 3, 3); + sqlite3VdbeLoadString(v, 4, " missing from index "); + sqlite3VdbeAddOp3(v, OP_Concat, 4, 3, 3); + jmp5 = sqlite3VdbeLoadString(v, 4, pIdx->zName); + sqlite3VdbeAddOp3(v, OP_Concat, 4, 3, 3); + jmp4 = integrityCheckResultRow(v); + sqlite3VdbeJumpHere(v, jmp2); + /* For UNIQUE indexes, verify that only one entry exists with the + ** current key. The entry is unique if (1) any column is NULL + ** or (2) the next entry has a different key */ + if( IsUniqueIndex(pIdx) ){ + int uniqOk = sqlite3VdbeMakeLabel(pParse); + int jmp6; + int kk; + for(kk=0; kknKeyCol; kk++){ + int iCol = pIdx->aiColumn[kk]; + assert( iCol!=XN_ROWID && iColnCol ); + if( iCol>=0 && pTab->aCol[iCol].notNull ) continue; + sqlite3VdbeAddOp2(v, OP_IsNull, r1+kk, uniqOk); + VdbeCoverage(v); + } + jmp6 = sqlite3VdbeAddOp1(v, OP_Next, iIdxCur+j); VdbeCoverage(v); + sqlite3VdbeGoto(v, uniqOk); + sqlite3VdbeJumpHere(v, jmp6); + sqlite3VdbeAddOp4Int(v, OP_IdxGT, iIdxCur+j, uniqOk, r1, + pIdx->nKeyCol); VdbeCoverage(v); + sqlite3VdbeLoadString(v, 3, "non-unique entry in index "); + sqlite3VdbeGoto(v, jmp5); + sqlite3VdbeResolveLabel(v, uniqOk); + } + sqlite3VdbeJumpHere(v, jmp4); + sqlite3ResolvePartIdxLabel(pParse, jmp3); + } + } + sqlite3VdbeAddOp2(v, OP_Next, iDataCur, loopTop); VdbeCoverage(v); + sqlite3VdbeJumpHere(v, loopTop-1); +#ifndef SQLITE_OMIT_BTREECOUNT + if( !isQuick ){ + sqlite3VdbeLoadString(v, 2, "wrong # of entries in index "); + for(j=0, pIdx=pTab->pIndex; pIdx; pIdx=pIdx->pNext, j++){ + if( pPk==pIdx ) continue; + sqlite3VdbeAddOp2(v, OP_Count, iIdxCur+j, 3); + addr = sqlite3VdbeAddOp3(v, OP_Eq, 8+j, 0, 3); VdbeCoverage(v); + sqlite3VdbeChangeP5(v, SQLITE_NOTNULL); + sqlite3VdbeLoadString(v, 4, pIdx->zName); + sqlite3VdbeAddOp3(v, OP_Concat, 4, 2, 3); + integrityCheckResultRow(v); + sqlite3VdbeJumpHere(v, addr); + } + } +#endif /* SQLITE_OMIT_BTREECOUNT */ + } + } + { + static const int iLn = VDBE_OFFSET_LINENO(2); + static const VdbeOpList endCode[] = { + { OP_AddImm, 1, 0, 0}, /* 0 */ + { OP_IfNotZero, 1, 4, 0}, /* 1 */ + { OP_String8, 0, 3, 0}, /* 2 */ + { OP_ResultRow, 3, 1, 0}, /* 3 */ + { OP_Halt, 0, 0, 0}, /* 4 */ + { OP_String8, 0, 3, 0}, /* 5 */ + { OP_Goto, 0, 3, 0}, /* 6 */ + }; + VdbeOp *aOp; + + aOp = sqlite3VdbeAddOpList(v, ArraySize(endCode), endCode, iLn); + if( aOp ){ + aOp[0].p2 = 1-mxErr; + aOp[2].p4type = P4_STATIC; + aOp[2].p4.z = "ok"; + aOp[5].p4type = P4_STATIC; + aOp[5].p4.z = (char*)sqlite3ErrStr(SQLITE_CORRUPT); + } + sqlite3VdbeChangeP3(v, 0, sqlite3VdbeCurrentAddr(v)-2); + } + } + break; +#endif /* SQLITE_OMIT_INTEGRITY_CHECK */ + +#ifndef SQLITE_OMIT_UTF16 + /* + ** PRAGMA encoding + ** PRAGMA encoding = "utf-8"|"utf-16"|"utf-16le"|"utf-16be" + ** + ** In its first form, this pragma returns the encoding of the main + ** database. If the database is not initialized, it is initialized now. + ** + ** The second form of this pragma is a no-op if the main database file + ** has not already been initialized. In this case it sets the default + ** encoding that will be used for the main database file if a new file + ** is created. If an existing main database file is opened, then the + ** default text encoding for the existing database is used. + ** + ** In all cases new databases created using the ATTACH command are + ** created to use the same default text encoding as the main database. If + ** the main database has not been initialized and/or created when ATTACH + ** is executed, this is done before the ATTACH operation. + ** + ** In the second form this pragma sets the text encoding to be used in + ** new database files created using this database handle. It is only + ** useful if invoked immediately after the main database i + */ + case PragTyp_ENCODING: { + static const struct EncName { + char *zName; + u8 enc; + } encnames[] = { + { "UTF8", SQLITE_UTF8 }, + { "UTF-8", SQLITE_UTF8 }, /* Must be element [1] */ + { "UTF-16le", SQLITE_UTF16LE }, /* Must be element [2] */ + { "UTF-16be", SQLITE_UTF16BE }, /* Must be element [3] */ + { "UTF16le", SQLITE_UTF16LE }, + { "UTF16be", SQLITE_UTF16BE }, + { "UTF-16", 0 }, /* SQLITE_UTF16NATIVE */ + { "UTF16", 0 }, /* SQLITE_UTF16NATIVE */ + { 0, 0 } + }; + const struct EncName *pEnc; + if( !zRight ){ /* "PRAGMA encoding" */ + if( sqlite3ReadSchema(pParse) ) goto pragma_out; + assert( encnames[SQLITE_UTF8].enc==SQLITE_UTF8 ); + assert( encnames[SQLITE_UTF16LE].enc==SQLITE_UTF16LE ); + assert( encnames[SQLITE_UTF16BE].enc==SQLITE_UTF16BE ); + returnSingleText(v, encnames[ENC(pParse->db)].zName); + }else{ /* "PRAGMA encoding = XXX" */ + /* Only change the value of sqlite.enc if the database handle is not + ** initialized. If the main database exists, the new sqlite.enc value + ** will be overwritten when the schema is next loaded. If it does not + ** already exists, it will be created to use the new encoding value. + */ + if( + !(DbHasProperty(db, 0, DB_SchemaLoaded)) || + DbHasProperty(db, 0, DB_Empty) + ){ + for(pEnc=&encnames[0]; pEnc->zName; pEnc++){ + if( 0==sqlite3StrICmp(zRight, pEnc->zName) ){ + SCHEMA_ENC(db) = ENC(db) = + pEnc->enc ? pEnc->enc : SQLITE_UTF16NATIVE; + break; + } + } + if( !pEnc->zName ){ + sqlite3ErrorMsg(pParse, "unsupported encoding: %s", zRight); + } + } + } + } + break; +#endif /* SQLITE_OMIT_UTF16 */ + +#ifndef SQLITE_OMIT_SCHEMA_VERSION_PRAGMAS + /* + ** PRAGMA [schema.]schema_version + ** PRAGMA [schema.]schema_version = + ** + ** PRAGMA [schema.]user_version + ** PRAGMA [schema.]user_version = + ** + ** PRAGMA [schema.]freelist_count + ** + ** PRAGMA [schema.]data_version + ** + ** PRAGMA [schema.]application_id + ** PRAGMA [schema.]application_id = + ** + ** The pragma's schema_version and user_version are used to set or get + ** the value of the schema-version and user-version, respectively. Both + ** the schema-version and the user-version are 32-bit signed integers + ** stored in the database header. + ** + ** The schema-cookie is usually only manipulated internally by SQLite. It + ** is incremented by SQLite whenever the database schema is modified (by + ** creating or dropping a table or index). The schema version is used by + ** SQLite each time a query is executed to ensure that the internal cache + ** of the schema used when compiling the SQL query matches the schema of + ** the database against which the compiled query is actually executed. + ** Subverting this mechanism by using "PRAGMA schema_version" to modify + ** the schema-version is potentially dangerous and may lead to program + ** crashes or database corruption. Use with caution! + ** + ** The user-version is not used internally by SQLite. It may be used by + ** applications for any purpose. + */ + case PragTyp_HEADER_VALUE: { + int iCookie = pPragma->iArg; /* Which cookie to read or write */ + sqlite3VdbeUsesBtree(v, iDb); + if( zRight && (pPragma->mPragFlg & PragFlg_ReadOnly)==0 ){ + /* Write the specified cookie value */ + static const VdbeOpList setCookie[] = { + { OP_Transaction, 0, 1, 0}, /* 0 */ + { OP_SetCookie, 0, 0, 0}, /* 1 */ + }; + VdbeOp *aOp; + sqlite3VdbeVerifyNoMallocRequired(v, ArraySize(setCookie)); + aOp = sqlite3VdbeAddOpList(v, ArraySize(setCookie), setCookie, 0); + if( ONLY_IF_REALLOC_STRESS(aOp==0) ) break; + aOp[0].p1 = iDb; + aOp[1].p1 = iDb; + aOp[1].p2 = iCookie; + aOp[1].p3 = sqlite3Atoi(zRight); + }else{ + /* Read the specified cookie value */ + static const VdbeOpList readCookie[] = { + { OP_Transaction, 0, 0, 0}, /* 0 */ + { OP_ReadCookie, 0, 1, 0}, /* 1 */ + { OP_ResultRow, 1, 1, 0} + }; + VdbeOp *aOp; + sqlite3VdbeVerifyNoMallocRequired(v, ArraySize(readCookie)); + aOp = sqlite3VdbeAddOpList(v, ArraySize(readCookie),readCookie,0); + if( ONLY_IF_REALLOC_STRESS(aOp==0) ) break; + aOp[0].p1 = iDb; + aOp[1].p1 = iDb; + aOp[1].p3 = iCookie; + sqlite3VdbeReusable(v); + } + } + break; +#endif /* SQLITE_OMIT_SCHEMA_VERSION_PRAGMAS */ + +#ifndef SQLITE_OMIT_COMPILEOPTION_DIAGS + /* + ** PRAGMA compile_options + ** + ** Return the names of all compile-time options used in this build, + ** one option per row. + */ + case PragTyp_COMPILE_OPTIONS: { + int i = 0; + const char *zOpt; + pParse->nMem = 1; + while( (zOpt = sqlite3_compileoption_get(i++))!=0 ){ + sqlite3VdbeLoadString(v, 1, zOpt); + sqlite3VdbeAddOp2(v, OP_ResultRow, 1, 1); + } + sqlite3VdbeReusable(v); + } + break; +#endif /* SQLITE_OMIT_COMPILEOPTION_DIAGS */ + +#ifndef SQLITE_OMIT_WAL + /* + ** PRAGMA [schema.]wal_checkpoint = passive|full|restart|truncate + ** + ** Checkpoint the database. + */ + case PragTyp_WAL_CHECKPOINT: { + int iBt = (pId2->z?iDb:SQLITE_MAX_ATTACHED); + int eMode = SQLITE_CHECKPOINT_PASSIVE; + if( zRight ){ + if( sqlite3StrICmp(zRight, "full")==0 ){ + eMode = SQLITE_CHECKPOINT_FULL; + }else if( sqlite3StrICmp(zRight, "restart")==0 ){ + eMode = SQLITE_CHECKPOINT_RESTART; + }else if( sqlite3StrICmp(zRight, "truncate")==0 ){ + eMode = SQLITE_CHECKPOINT_TRUNCATE; + } + } + pParse->nMem = 3; + sqlite3VdbeAddOp3(v, OP_Checkpoint, iBt, eMode, 1); + sqlite3VdbeAddOp2(v, OP_ResultRow, 1, 3); + } + break; + + /* + ** PRAGMA wal_autocheckpoint + ** PRAGMA wal_autocheckpoint = N + ** + ** Configure a database connection to automatically checkpoint a database + ** after accumulating N frames in the log. Or query for the current value + ** of N. + */ + case PragTyp_WAL_AUTOCHECKPOINT: { + if( zRight ){ + sqlite3_wal_autocheckpoint(db, sqlite3Atoi(zRight)); + } + returnSingleInt(v, + db->xWalCallback==sqlite3WalDefaultHook ? + SQLITE_PTR_TO_INT(db->pWalArg) : 0); + } + break; +#endif + + /* + ** PRAGMA shrink_memory + ** + ** IMPLEMENTATION-OF: R-23445-46109 This pragma causes the database + ** connection on which it is invoked to free up as much memory as it + ** can, by calling sqlite3_db_release_memory(). + */ + case PragTyp_SHRINK_MEMORY: { + sqlite3_db_release_memory(db); + break; + } + + /* + ** PRAGMA optimize + ** PRAGMA optimize(MASK) + ** PRAGMA schema.optimize + ** PRAGMA schema.optimize(MASK) + ** + ** Attempt to optimize the database. All schemas are optimized in the first + ** two forms, and only the specified schema is optimized in the latter two. + ** + ** The details of optimizations performed by this pragma are expected + ** to change and improve over time. Applications should anticipate that + ** this pragma will perform new optimizations in future releases. + ** + ** The optional argument is a bitmask of optimizations to perform: + ** + ** 0x0001 Debugging mode. Do not actually perform any optimizations + ** but instead return one line of text for each optimization + ** that would have been done. Off by default. + ** + ** 0x0002 Run ANALYZE on tables that might benefit. On by default. + ** See below for additional information. + ** + ** 0x0004 (Not yet implemented) Record usage and performance + ** information from the current session in the + ** database file so that it will be available to "optimize" + ** pragmas run by future database connections. + ** + ** 0x0008 (Not yet implemented) Create indexes that might have + ** been helpful to recent queries + ** + ** The default MASK is and always shall be 0xfffe. 0xfffe means perform all + ** of the optimizations listed above except Debug Mode, including new + ** optimizations that have not yet been invented. If new optimizations are + ** ever added that should be off by default, those off-by-default + ** optimizations will have bitmasks of 0x10000 or larger. + ** + ** DETERMINATION OF WHEN TO RUN ANALYZE + ** + ** In the current implementation, a table is analyzed if only if all of + ** the following are true: + ** + ** (1) MASK bit 0x02 is set. + ** + ** (2) The query planner used sqlite_stat1-style statistics for one or + ** more indexes of the table at some point during the lifetime of + ** the current connection. + ** + ** (3) One or more indexes of the table are currently unanalyzed OR + ** the number of rows in the table has increased by 25 times or more + ** since the last time ANALYZE was run. + ** + ** The rules for when tables are analyzed are likely to change in + ** future releases. + */ + case PragTyp_OPTIMIZE: { + int iDbLast; /* Loop termination point for the schema loop */ + int iTabCur; /* Cursor for a table whose size needs checking */ + HashElem *k; /* Loop over tables of a schema */ + Schema *pSchema; /* The current schema */ + Table *pTab; /* A table in the schema */ + Index *pIdx; /* An index of the table */ + LogEst szThreshold; /* Size threshold above which reanalysis is needd */ + char *zSubSql; /* SQL statement for the OP_SqlExec opcode */ + u32 opMask; /* Mask of operations to perform */ + + if( zRight ){ + opMask = (u32)sqlite3Atoi(zRight); + if( (opMask & 0x02)==0 ) break; + }else{ + opMask = 0xfffe; + } + iTabCur = pParse->nTab++; + for(iDbLast = zDb?iDb:db->nDb-1; iDb<=iDbLast; iDb++){ + if( iDb==1 ) continue; + sqlite3CodeVerifySchema(pParse, iDb); + pSchema = db->aDb[iDb].pSchema; + for(k=sqliteHashFirst(&pSchema->tblHash); k; k=sqliteHashNext(k)){ + pTab = (Table*)sqliteHashData(k); + + /* If table pTab has not been used in a way that would benefit from + ** having analysis statistics during the current session, then skip it. + ** This also has the effect of skipping virtual tables and views */ + if( (pTab->tabFlags & TF_StatsUsed)==0 ) continue; + + /* Reanalyze if the table is 25 times larger than the last analysis */ + szThreshold = pTab->nRowLogEst + 46; assert( sqlite3LogEst(25)==46 ); + for(pIdx=pTab->pIndex; pIdx; pIdx=pIdx->pNext){ + if( !pIdx->hasStat1 ){ + szThreshold = 0; /* Always analyze if any index lacks statistics */ + break; + } + } + if( szThreshold ){ + sqlite3OpenTable(pParse, iTabCur, iDb, pTab, OP_OpenRead); + sqlite3VdbeAddOp3(v, OP_IfSmaller, iTabCur, + sqlite3VdbeCurrentAddr(v)+2+(opMask&1), szThreshold); + VdbeCoverage(v); + } + zSubSql = sqlite3MPrintf(db, "ANALYZE \"%w\".\"%w\"", + db->aDb[iDb].zDbSName, pTab->zName); + if( opMask & 0x01 ){ + int r1 = sqlite3GetTempReg(pParse); + sqlite3VdbeAddOp4(v, OP_String8, 0, r1, 0, zSubSql, P4_DYNAMIC); + sqlite3VdbeAddOp2(v, OP_ResultRow, r1, 1); + }else{ + sqlite3VdbeAddOp4(v, OP_SqlExec, 0, 0, 0, zSubSql, P4_DYNAMIC); + } + } + } + sqlite3VdbeAddOp0(v, OP_Expire); + break; + } + + /* + ** PRAGMA busy_timeout + ** PRAGMA busy_timeout = N + ** + ** Call sqlite3_busy_timeout(db, N). Return the current timeout value + ** if one is set. If no busy handler or a different busy handler is set + ** then 0 is returned. Setting the busy_timeout to 0 or negative + ** disables the timeout. + */ + /*case PragTyp_BUSY_TIMEOUT*/ default: { + assert( pPragma->ePragTyp==PragTyp_BUSY_TIMEOUT ); + if( zRight ){ + sqlite3_busy_timeout(db, sqlite3Atoi(zRight)); + } + returnSingleInt(v, db->busyTimeout); + break; + } + + /* + ** PRAGMA soft_heap_limit + ** PRAGMA soft_heap_limit = N + ** + ** IMPLEMENTATION-OF: R-26343-45930 This pragma invokes the + ** sqlite3_soft_heap_limit64() interface with the argument N, if N is + ** specified and is a non-negative integer. + ** IMPLEMENTATION-OF: R-64451-07163 The soft_heap_limit pragma always + ** returns the same integer that would be returned by the + ** sqlite3_soft_heap_limit64(-1) C-language function. + */ + case PragTyp_SOFT_HEAP_LIMIT: { + sqlite3_int64 N; + if( zRight && sqlite3DecOrHexToI64(zRight, &N)==SQLITE_OK ){ + sqlite3_soft_heap_limit64(N); + } + returnSingleInt(v, sqlite3_soft_heap_limit64(-1)); + break; + } + + /* + ** PRAGMA threads + ** PRAGMA threads = N + ** + ** Configure the maximum number of worker threads. Return the new + ** maximum, which might be less than requested. + */ + case PragTyp_THREADS: { + sqlite3_int64 N; + if( zRight + && sqlite3DecOrHexToI64(zRight, &N)==SQLITE_OK + && N>=0 + ){ + sqlite3_limit(db, SQLITE_LIMIT_WORKER_THREADS, (int)(N&0x7fffffff)); + } + returnSingleInt(v, sqlite3_limit(db, SQLITE_LIMIT_WORKER_THREADS, -1)); + break; + } + +#if defined(SQLITE_DEBUG) || defined(SQLITE_TEST) + /* + ** Report the current state of file logs for all databases + */ + case PragTyp_LOCK_STATUS: { + static const char *const azLockName[] = { + "unlocked", "shared", "reserved", "pending", "exclusive" + }; + int i; + pParse->nMem = 2; + for(i=0; inDb; i++){ + Btree *pBt; + const char *zState = "unknown"; + int j; + if( db->aDb[i].zDbSName==0 ) continue; + pBt = db->aDb[i].pBt; + if( pBt==0 || sqlite3BtreePager(pBt)==0 ){ + zState = "closed"; + }else if( sqlite3_file_control(db, i ? db->aDb[i].zDbSName : 0, + SQLITE_FCNTL_LOCKSTATE, &j)==SQLITE_OK ){ + zState = azLockName[j]; + } + sqlite3VdbeMultiLoad(v, 1, "ss", db->aDb[i].zDbSName, zState); + } + break; + } +#endif + +#ifdef SQLITE_HAS_CODEC + /* Pragma iArg + ** ---------- ------ + ** key 0 + ** rekey 1 + ** hexkey 2 + ** hexrekey 3 + ** textkey 4 + ** textrekey 5 + */ + case PragTyp_KEY: { + if( zRight ){ + char zBuf[40]; + const char *zKey = zRight; + int n; + if( pPragma->iArg==2 || pPragma->iArg==3 ){ + u8 iByte; + int i; + for(i=0, iByte=0; iiArg<4 ? sqlite3Strlen30(zRight) : -1; + } + if( (pPragma->iArg & 1)==0 ){ + rc = sqlite3_key_v2(db, zDb, zKey, n); + }else{ + rc = sqlite3_rekey_v2(db, zDb, zKey, n); + } + if( rc==SQLITE_OK && n!=0 ){ + sqlite3VdbeSetNumCols(v, 1); + sqlite3VdbeSetColName(v, 0, COLNAME_NAME, "ok", SQLITE_STATIC); + returnSingleText(v, "ok"); + } + } + break; + } +#endif +#if defined(SQLITE_HAS_CODEC) || defined(SQLITE_ENABLE_CEROD) + case PragTyp_ACTIVATE_EXTENSIONS: if( zRight ){ +#ifdef SQLITE_HAS_CODEC + if( sqlite3StrNICmp(zRight, "see-", 4)==0 ){ + sqlite3_activate_see(&zRight[4]); + } +#endif +#ifdef SQLITE_ENABLE_CEROD + if( sqlite3StrNICmp(zRight, "cerod-", 6)==0 ){ + sqlite3_activate_cerod(&zRight[6]); + } +#endif + } + break; +#endif + + } /* End of the PRAGMA switch */ + + /* The following block is a no-op unless SQLITE_DEBUG is defined. Its only + ** purpose is to execute assert() statements to verify that if the + ** PragFlg_NoColumns1 flag is set and the caller specified an argument + ** to the PRAGMA, the implementation has not added any OP_ResultRow + ** instructions to the VM. */ + if( (pPragma->mPragFlg & PragFlg_NoColumns1) && zRight ){ + sqlite3VdbeVerifyNoResultRow(v); + } + +pragma_out: + sqlite3DbFree(db, zLeft); + sqlite3DbFree(db, zRight); +} +#ifndef SQLITE_OMIT_VIRTUALTABLE +/***************************************************************************** +** Implementation of an eponymous virtual table that runs a pragma. +** +*/ +typedef struct PragmaVtab PragmaVtab; +typedef struct PragmaVtabCursor PragmaVtabCursor; +struct PragmaVtab { + sqlite3_vtab base; /* Base class. Must be first */ + sqlite3 *db; /* The database connection to which it belongs */ + const PragmaName *pName; /* Name of the pragma */ + u8 nHidden; /* Number of hidden columns */ + u8 iHidden; /* Index of the first hidden column */ +}; +struct PragmaVtabCursor { + sqlite3_vtab_cursor base; /* Base class. Must be first */ + sqlite3_stmt *pPragma; /* The pragma statement to run */ + sqlite_int64 iRowid; /* Current rowid */ + char *azArg[2]; /* Value of the argument and schema */ +}; + +/* +** Pragma virtual table module xConnect method. +*/ +static int pragmaVtabConnect( + sqlite3 *db, + void *pAux, + int argc, const char *const*argv, + sqlite3_vtab **ppVtab, + char **pzErr +){ + const PragmaName *pPragma = (const PragmaName*)pAux; + PragmaVtab *pTab = 0; + int rc; + int i, j; + char cSep = '('; + StrAccum acc; + char zBuf[200]; + + UNUSED_PARAMETER(argc); + UNUSED_PARAMETER(argv); + sqlite3StrAccumInit(&acc, 0, zBuf, sizeof(zBuf), 0); + sqlite3_str_appendall(&acc, "CREATE TABLE x"); + for(i=0, j=pPragma->iPragCName; inPragCName; i++, j++){ + sqlite3_str_appendf(&acc, "%c\"%s\"", cSep, pragCName[j]); + cSep = ','; + } + if( i==0 ){ + sqlite3_str_appendf(&acc, "(\"%s\"", pPragma->zName); + i++; + } + j = 0; + if( pPragma->mPragFlg & PragFlg_Result1 ){ + sqlite3_str_appendall(&acc, ",arg HIDDEN"); + j++; + } + if( pPragma->mPragFlg & (PragFlg_SchemaOpt|PragFlg_SchemaReq) ){ + sqlite3_str_appendall(&acc, ",schema HIDDEN"); + j++; + } + sqlite3_str_append(&acc, ")", 1); + sqlite3StrAccumFinish(&acc); + assert( strlen(zBuf) < sizeof(zBuf)-1 ); + rc = sqlite3_declare_vtab(db, zBuf); + if( rc==SQLITE_OK ){ + pTab = (PragmaVtab*)sqlite3_malloc(sizeof(PragmaVtab)); + if( pTab==0 ){ + rc = SQLITE_NOMEM; + }else{ + memset(pTab, 0, sizeof(PragmaVtab)); + pTab->pName = pPragma; + pTab->db = db; + pTab->iHidden = i; + pTab->nHidden = j; + } + }else{ + *pzErr = sqlite3_mprintf("%s", sqlite3_errmsg(db)); + } + + *ppVtab = (sqlite3_vtab*)pTab; + return rc; +} + +/* +** Pragma virtual table module xDisconnect method. +*/ +static int pragmaVtabDisconnect(sqlite3_vtab *pVtab){ + PragmaVtab *pTab = (PragmaVtab*)pVtab; + sqlite3_free(pTab); + return SQLITE_OK; +} + +/* Figure out the best index to use to search a pragma virtual table. +** +** There are not really any index choices. But we want to encourage the +** query planner to give == constraints on as many hidden parameters as +** possible, and especially on the first hidden parameter. So return a +** high cost if hidden parameters are unconstrained. +*/ +static int pragmaVtabBestIndex(sqlite3_vtab *tab, sqlite3_index_info *pIdxInfo){ + PragmaVtab *pTab = (PragmaVtab*)tab; + const struct sqlite3_index_constraint *pConstraint; + int i, j; + int seen[2]; + + pIdxInfo->estimatedCost = (double)1; + if( pTab->nHidden==0 ){ return SQLITE_OK; } + pConstraint = pIdxInfo->aConstraint; + seen[0] = 0; + seen[1] = 0; + for(i=0; inConstraint; i++, pConstraint++){ + if( pConstraint->usable==0 ) continue; + if( pConstraint->op!=SQLITE_INDEX_CONSTRAINT_EQ ) continue; + if( pConstraint->iColumn < pTab->iHidden ) continue; + j = pConstraint->iColumn - pTab->iHidden; + assert( j < 2 ); + seen[j] = i+1; + } + if( seen[0]==0 ){ + pIdxInfo->estimatedCost = (double)2147483647; + pIdxInfo->estimatedRows = 2147483647; + return SQLITE_OK; + } + j = seen[0]-1; + pIdxInfo->aConstraintUsage[j].argvIndex = 1; + pIdxInfo->aConstraintUsage[j].omit = 1; + if( seen[1]==0 ) return SQLITE_OK; + pIdxInfo->estimatedCost = (double)20; + pIdxInfo->estimatedRows = 20; + j = seen[1]-1; + pIdxInfo->aConstraintUsage[j].argvIndex = 2; + pIdxInfo->aConstraintUsage[j].omit = 1; + return SQLITE_OK; +} + +/* Create a new cursor for the pragma virtual table */ +static int pragmaVtabOpen(sqlite3_vtab *pVtab, sqlite3_vtab_cursor **ppCursor){ + PragmaVtabCursor *pCsr; + pCsr = (PragmaVtabCursor*)sqlite3_malloc(sizeof(*pCsr)); + if( pCsr==0 ) return SQLITE_NOMEM; + memset(pCsr, 0, sizeof(PragmaVtabCursor)); + pCsr->base.pVtab = pVtab; + *ppCursor = &pCsr->base; + return SQLITE_OK; +} + +/* Clear all content from pragma virtual table cursor. */ +static void pragmaVtabCursorClear(PragmaVtabCursor *pCsr){ + int i; + sqlite3_finalize(pCsr->pPragma); + pCsr->pPragma = 0; + for(i=0; iazArg); i++){ + sqlite3_free(pCsr->azArg[i]); + pCsr->azArg[i] = 0; + } +} + +/* Close a pragma virtual table cursor */ +static int pragmaVtabClose(sqlite3_vtab_cursor *cur){ + PragmaVtabCursor *pCsr = (PragmaVtabCursor*)cur; + pragmaVtabCursorClear(pCsr); + sqlite3_free(pCsr); + return SQLITE_OK; +} + +/* Advance the pragma virtual table cursor to the next row */ +static int pragmaVtabNext(sqlite3_vtab_cursor *pVtabCursor){ + PragmaVtabCursor *pCsr = (PragmaVtabCursor*)pVtabCursor; + int rc = SQLITE_OK; + + /* Increment the xRowid value */ + pCsr->iRowid++; + assert( pCsr->pPragma ); + if( SQLITE_ROW!=sqlite3_step(pCsr->pPragma) ){ + rc = sqlite3_finalize(pCsr->pPragma); + pCsr->pPragma = 0; + pragmaVtabCursorClear(pCsr); + } + return rc; +} + +/* +** Pragma virtual table module xFilter method. +*/ +static int pragmaVtabFilter( + sqlite3_vtab_cursor *pVtabCursor, + int idxNum, const char *idxStr, + int argc, sqlite3_value **argv +){ + PragmaVtabCursor *pCsr = (PragmaVtabCursor*)pVtabCursor; + PragmaVtab *pTab = (PragmaVtab*)(pVtabCursor->pVtab); + int rc; + int i, j; + StrAccum acc; + char *zSql; + + UNUSED_PARAMETER(idxNum); + UNUSED_PARAMETER(idxStr); + pragmaVtabCursorClear(pCsr); + j = (pTab->pName->mPragFlg & PragFlg_Result1)!=0 ? 0 : 1; + for(i=0; iazArg) ); + assert( pCsr->azArg[j]==0 ); + if( zText ){ + pCsr->azArg[j] = sqlite3_mprintf("%s", zText); + if( pCsr->azArg[j]==0 ){ + return SQLITE_NOMEM; + } + } + } + sqlite3StrAccumInit(&acc, 0, 0, 0, pTab->db->aLimit[SQLITE_LIMIT_SQL_LENGTH]); + sqlite3_str_appendall(&acc, "PRAGMA "); + if( pCsr->azArg[1] ){ + sqlite3_str_appendf(&acc, "%Q.", pCsr->azArg[1]); + } + sqlite3_str_appendall(&acc, pTab->pName->zName); + if( pCsr->azArg[0] ){ + sqlite3_str_appendf(&acc, "=%Q", pCsr->azArg[0]); + } + zSql = sqlite3StrAccumFinish(&acc); + if( zSql==0 ) return SQLITE_NOMEM; + rc = sqlite3_prepare_v2(pTab->db, zSql, -1, &pCsr->pPragma, 0); + sqlite3_free(zSql); + if( rc!=SQLITE_OK ){ + pTab->base.zErrMsg = sqlite3_mprintf("%s", sqlite3_errmsg(pTab->db)); + return rc; + } + return pragmaVtabNext(pVtabCursor); +} + +/* +** Pragma virtual table module xEof method. +*/ +static int pragmaVtabEof(sqlite3_vtab_cursor *pVtabCursor){ + PragmaVtabCursor *pCsr = (PragmaVtabCursor*)pVtabCursor; + return (pCsr->pPragma==0); +} + +/* The xColumn method simply returns the corresponding column from +** the PRAGMA. +*/ +static int pragmaVtabColumn( + sqlite3_vtab_cursor *pVtabCursor, + sqlite3_context *ctx, + int i +){ + PragmaVtabCursor *pCsr = (PragmaVtabCursor*)pVtabCursor; + PragmaVtab *pTab = (PragmaVtab*)(pVtabCursor->pVtab); + if( iiHidden ){ + sqlite3_result_value(ctx, sqlite3_column_value(pCsr->pPragma, i)); + }else{ + sqlite3_result_text(ctx, pCsr->azArg[i-pTab->iHidden],-1,SQLITE_TRANSIENT); + } + return SQLITE_OK; +} + +/* +** Pragma virtual table module xRowid method. +*/ +static int pragmaVtabRowid(sqlite3_vtab_cursor *pVtabCursor, sqlite_int64 *p){ + PragmaVtabCursor *pCsr = (PragmaVtabCursor*)pVtabCursor; + *p = pCsr->iRowid; + return SQLITE_OK; +} + +/* The pragma virtual table object */ +static const sqlite3_module pragmaVtabModule = { + 0, /* iVersion */ + 0, /* xCreate - create a table */ + pragmaVtabConnect, /* xConnect - connect to an existing table */ + pragmaVtabBestIndex, /* xBestIndex - Determine search strategy */ + pragmaVtabDisconnect, /* xDisconnect - Disconnect from a table */ + 0, /* xDestroy - Drop a table */ + pragmaVtabOpen, /* xOpen - open a cursor */ + pragmaVtabClose, /* xClose - close a cursor */ + pragmaVtabFilter, /* xFilter - configure scan constraints */ + pragmaVtabNext, /* xNext - advance a cursor */ + pragmaVtabEof, /* xEof */ + pragmaVtabColumn, /* xColumn - read data */ + pragmaVtabRowid, /* xRowid - read data */ + 0, /* xUpdate - write data */ + 0, /* xBegin - begin transaction */ + 0, /* xSync - sync transaction */ + 0, /* xCommit - commit transaction */ + 0, /* xRollback - rollback transaction */ + 0, /* xFindFunction - function overloading */ + 0, /* xRename - rename the table */ + 0, /* xSavepoint */ + 0, /* xRelease */ + 0, /* xRollbackTo */ + 0 /* xShadowName */ +}; + +/* +** Check to see if zTabName is really the name of a pragma. If it is, +** then register an eponymous virtual table for that pragma and return +** a pointer to the Module object for the new virtual table. +*/ +SQLITE_PRIVATE Module *sqlite3PragmaVtabRegister(sqlite3 *db, const char *zName){ + const PragmaName *pName; + assert( sqlite3_strnicmp(zName, "pragma_", 7)==0 ); + pName = pragmaLocate(zName+7); + if( pName==0 ) return 0; + if( (pName->mPragFlg & (PragFlg_Result0|PragFlg_Result1))==0 ) return 0; + assert( sqlite3HashFind(&db->aModule, zName)==0 ); + return sqlite3VtabCreateModule(db, zName, &pragmaVtabModule, (void*)pName, 0); +} + +#endif /* SQLITE_OMIT_VIRTUALTABLE */ + +#endif /* SQLITE_OMIT_PRAGMA */ + +/************** End of pragma.c **********************************************/ +/************** Begin file prepare.c *****************************************/ +/* +** 2005 May 25 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** This file contains the implementation of the sqlite3_prepare() +** interface, and routines that contribute to loading the database schema +** from disk. +*/ +/* #include "sqliteInt.h" */ + +/* +** Fill the InitData structure with an error message that indicates +** that the database is corrupt. +*/ +static void corruptSchema( + InitData *pData, /* Initialization context */ + const char *zObj, /* Object being parsed at the point of error */ + const char *zExtra /* Error information */ +){ + sqlite3 *db = pData->db; + if( db->mallocFailed ){ + pData->rc = SQLITE_NOMEM_BKPT; + }else if( pData->pzErrMsg[0]!=0 ){ + /* A error message has already been generated. Do not overwrite it */ + }else if( pData->mInitFlags & INITFLAG_AlterTable ){ + *pData->pzErrMsg = sqlite3DbStrDup(db, zExtra); + pData->rc = SQLITE_ERROR; + }else if( db->flags & SQLITE_WriteSchema ){ + pData->rc = SQLITE_CORRUPT_BKPT; + }else{ + char *z; + if( zObj==0 ) zObj = "?"; + z = sqlite3MPrintf(db, "malformed database schema (%s)", zObj); + if( zExtra && zExtra[0] ) z = sqlite3MPrintf(db, "%z - %s", z, zExtra); + *pData->pzErrMsg = z; + pData->rc = SQLITE_CORRUPT_BKPT; + } +} + +/* +** Check to see if any sibling index (another index on the same table) +** of pIndex has the same root page number, and if it does, return true. +** This would indicate a corrupt schema. +*/ +SQLITE_PRIVATE int sqlite3IndexHasDuplicateRootPage(Index *pIndex){ + Index *p; + for(p=pIndex->pTable->pIndex; p; p=p->pNext){ + if( p->tnum==pIndex->tnum && p!=pIndex ) return 1; + } + return 0; +} + +/* +** This is the callback routine for the code that initializes the +** database. See sqlite3Init() below for additional information. +** This routine is also called from the OP_ParseSchema opcode of the VDBE. +** +** Each callback contains the following information: +** +** argv[0] = type of object: "table", "index", "trigger", or "view". +** argv[1] = name of thing being created +** argv[2] = associated table if an index or trigger +** argv[3] = root page number for table or index. 0 for trigger or view. +** argv[4] = SQL text for the CREATE statement. +** +*/ +SQLITE_PRIVATE int sqlite3InitCallback(void *pInit, int argc, char **argv, char **NotUsed){ + InitData *pData = (InitData*)pInit; + sqlite3 *db = pData->db; + int iDb = pData->iDb; + + assert( argc==5 ); + UNUSED_PARAMETER2(NotUsed, argc); + assert( sqlite3_mutex_held(db->mutex) ); + DbClearProperty(db, iDb, DB_Empty); + pData->nInitRow++; + if( db->mallocFailed ){ + corruptSchema(pData, argv[1], 0); + return 1; + } + + assert( iDb>=0 && iDbnDb ); + if( argv==0 ) return 0; /* Might happen if EMPTY_RESULT_CALLBACKS are on */ + if( argv[3]==0 ){ + corruptSchema(pData, argv[1], 0); + }else if( sqlite3_strnicmp(argv[4],"create ",7)==0 ){ + /* Call the parser to process a CREATE TABLE, INDEX or VIEW. + ** But because db->init.busy is set to 1, no VDBE code is generated + ** or executed. All the parser does is build the internal data + ** structures that describe the table, index, or view. + */ + int rc; + u8 saved_iDb = db->init.iDb; + sqlite3_stmt *pStmt; + TESTONLY(int rcp); /* Return code from sqlite3_prepare() */ + + assert( db->init.busy ); + db->init.iDb = iDb; + db->init.newTnum = sqlite3Atoi(argv[3]); + db->init.orphanTrigger = 0; + db->init.azInit = argv; + TESTONLY(rcp = ) sqlite3_prepare(db, argv[4], -1, &pStmt, 0); + rc = db->errCode; + assert( (rc&0xFF)==(rcp&0xFF) ); + db->init.iDb = saved_iDb; + /* assert( saved_iDb==0 || (db->mDbFlags & DBFLAG_Vacuum)!=0 ); */ + if( SQLITE_OK!=rc ){ + if( db->init.orphanTrigger ){ + assert( iDb==1 ); + }else{ + if( rc > pData->rc ) pData->rc = rc; + if( rc==SQLITE_NOMEM ){ + sqlite3OomFault(db); + }else if( rc!=SQLITE_INTERRUPT && (rc&0xFF)!=SQLITE_LOCKED ){ + corruptSchema(pData, argv[1], sqlite3_errmsg(db)); + } + } + } + sqlite3_finalize(pStmt); + }else if( argv[1]==0 || (argv[4]!=0 && argv[4][0]!=0) ){ + corruptSchema(pData, argv[1], 0); + }else{ + /* If the SQL column is blank it means this is an index that + ** was created to be the PRIMARY KEY or to fulfill a UNIQUE + ** constraint for a CREATE TABLE. The index should have already + ** been created when we processed the CREATE TABLE. All we have + ** to do here is record the root page number for that index. + */ + Index *pIndex; + pIndex = sqlite3FindIndex(db, argv[1], db->aDb[iDb].zDbSName); + if( pIndex==0 + || sqlite3GetInt32(argv[3],&pIndex->tnum)==0 + || pIndex->tnum<2 + || sqlite3IndexHasDuplicateRootPage(pIndex) + ){ + corruptSchema(pData, argv[1], pIndex?"invalid rootpage":"orphan index"); + } + } + return 0; +} + +/* +** Attempt to read the database schema and initialize internal +** data structures for a single database file. The index of the +** database file is given by iDb. iDb==0 is used for the main +** database. iDb==1 should never be used. iDb>=2 is used for +** auxiliary databases. Return one of the SQLITE_ error codes to +** indicate success or failure. +*/ +SQLITE_PRIVATE int sqlite3InitOne(sqlite3 *db, int iDb, char **pzErrMsg, u32 mFlags){ + int rc; + int i; +#ifndef SQLITE_OMIT_DEPRECATED + int size; +#endif + Db *pDb; + char const *azArg[6]; + int meta[5]; + InitData initData; + const char *zMasterName; + int openedTransaction = 0; + + assert( (db->mDbFlags & DBFLAG_SchemaKnownOk)==0 ); + assert( iDb>=0 && iDbnDb ); + assert( db->aDb[iDb].pSchema ); + assert( sqlite3_mutex_held(db->mutex) ); + assert( iDb==1 || sqlite3BtreeHoldsMutex(db->aDb[iDb].pBt) ); + + db->init.busy = 1; + + /* Construct the in-memory representation schema tables (sqlite_master or + ** sqlite_temp_master) by invoking the parser directly. The appropriate + ** table name will be inserted automatically by the parser so we can just + ** use the abbreviation "x" here. The parser will also automatically tag + ** the schema table as read-only. */ + azArg[0] = "table"; + azArg[1] = zMasterName = SCHEMA_TABLE(iDb); + azArg[2] = azArg[1]; + azArg[3] = "1"; + azArg[4] = "CREATE TABLE x(type text,name text,tbl_name text," + "rootpage int,sql text)"; + azArg[5] = 0; + initData.db = db; + initData.iDb = iDb; + initData.rc = SQLITE_OK; + initData.pzErrMsg = pzErrMsg; + initData.mInitFlags = mFlags; + initData.nInitRow = 0; + sqlite3InitCallback(&initData, 5, (char **)azArg, 0); + if( initData.rc ){ + rc = initData.rc; + goto error_out; + } + + /* Create a cursor to hold the database open + */ + pDb = &db->aDb[iDb]; + if( pDb->pBt==0 ){ + assert( iDb==1 ); + DbSetProperty(db, 1, DB_SchemaLoaded); + rc = SQLITE_OK; + goto error_out; + } + + /* If there is not already a read-only (or read-write) transaction opened + ** on the b-tree database, open one now. If a transaction is opened, it + ** will be closed before this function returns. */ + sqlite3BtreeEnter(pDb->pBt); + if( !sqlite3BtreeIsInReadTrans(pDb->pBt) ){ + rc = sqlite3BtreeBeginTrans(pDb->pBt, 0, 0); + if( rc!=SQLITE_OK ){ + sqlite3SetString(pzErrMsg, db, sqlite3ErrStr(rc)); + goto initone_error_out; + } + openedTransaction = 1; + } + + /* Get the database meta information. + ** + ** Meta values are as follows: + ** meta[0] Schema cookie. Changes with each schema change. + ** meta[1] File format of schema layer. + ** meta[2] Size of the page cache. + ** meta[3] Largest rootpage (auto/incr_vacuum mode) + ** meta[4] Db text encoding. 1:UTF-8 2:UTF-16LE 3:UTF-16BE + ** meta[5] User version + ** meta[6] Incremental vacuum mode + ** meta[7] unused + ** meta[8] unused + ** meta[9] unused + ** + ** Note: The #defined SQLITE_UTF* symbols in sqliteInt.h correspond to + ** the possible values of meta[4]. + */ + for(i=0; ipBt, i+1, (u32 *)&meta[i]); + } + if( (db->flags & SQLITE_ResetDatabase)!=0 ){ + memset(meta, 0, sizeof(meta)); + } + pDb->pSchema->schema_cookie = meta[BTREE_SCHEMA_VERSION-1]; + + /* If opening a non-empty database, check the text encoding. For the + ** main database, set sqlite3.enc to the encoding of the main database. + ** For an attached db, it is an error if the encoding is not the same + ** as sqlite3.enc. + */ + if( meta[BTREE_TEXT_ENCODING-1] ){ /* text encoding */ + if( iDb==0 ){ +#ifndef SQLITE_OMIT_UTF16 + u8 encoding; + /* If opening the main database, set ENC(db). */ + encoding = (u8)meta[BTREE_TEXT_ENCODING-1] & 3; + if( encoding==0 ) encoding = SQLITE_UTF8; + ENC(db) = encoding; +#else + ENC(db) = SQLITE_UTF8; +#endif + }else{ + /* If opening an attached database, the encoding much match ENC(db) */ + if( meta[BTREE_TEXT_ENCODING-1]!=ENC(db) ){ + sqlite3SetString(pzErrMsg, db, "attached databases must use the same" + " text encoding as main database"); + rc = SQLITE_ERROR; + goto initone_error_out; + } + } + }else{ + DbSetProperty(db, iDb, DB_Empty); + } + pDb->pSchema->enc = ENC(db); + + if( pDb->pSchema->cache_size==0 ){ +#ifndef SQLITE_OMIT_DEPRECATED + size = sqlite3AbsInt32(meta[BTREE_DEFAULT_CACHE_SIZE-1]); + if( size==0 ){ size = SQLITE_DEFAULT_CACHE_SIZE; } + pDb->pSchema->cache_size = size; +#else + pDb->pSchema->cache_size = SQLITE_DEFAULT_CACHE_SIZE; +#endif + sqlite3BtreeSetCacheSize(pDb->pBt, pDb->pSchema->cache_size); + } + + /* + ** file_format==1 Version 3.0.0. + ** file_format==2 Version 3.1.3. // ALTER TABLE ADD COLUMN + ** file_format==3 Version 3.1.4. // ditto but with non-NULL defaults + ** file_format==4 Version 3.3.0. // DESC indices. Boolean constants + */ + pDb->pSchema->file_format = (u8)meta[BTREE_FILE_FORMAT-1]; + if( pDb->pSchema->file_format==0 ){ + pDb->pSchema->file_format = 1; + } + if( pDb->pSchema->file_format>SQLITE_MAX_FILE_FORMAT ){ + sqlite3SetString(pzErrMsg, db, "unsupported file format"); + rc = SQLITE_ERROR; + goto initone_error_out; + } + + /* Ticket #2804: When we open a database in the newer file format, + ** clear the legacy_file_format pragma flag so that a VACUUM will + ** not downgrade the database and thus invalidate any descending + ** indices that the user might have created. + */ + if( iDb==0 && meta[BTREE_FILE_FORMAT-1]>=4 ){ + db->flags &= ~(u64)SQLITE_LegacyFileFmt; + } + + /* Read the schema information out of the schema tables + */ + assert( db->init.busy ); + { + char *zSql; + zSql = sqlite3MPrintf(db, + "SELECT*FROM\"%w\".%s ORDER BY rowid", + db->aDb[iDb].zDbSName, zMasterName); +#ifndef SQLITE_OMIT_AUTHORIZATION + { + sqlite3_xauth xAuth; + xAuth = db->xAuth; + db->xAuth = 0; +#endif + rc = sqlite3_exec(db, zSql, sqlite3InitCallback, &initData, 0); +#ifndef SQLITE_OMIT_AUTHORIZATION + db->xAuth = xAuth; + } +#endif + if( rc==SQLITE_OK ) rc = initData.rc; + sqlite3DbFree(db, zSql); +#ifndef SQLITE_OMIT_ANALYZE + if( rc==SQLITE_OK ){ + sqlite3AnalysisLoad(db, iDb); + } +#endif + } + if( db->mallocFailed ){ + rc = SQLITE_NOMEM_BKPT; + sqlite3ResetAllSchemasOfConnection(db); + } + if( rc==SQLITE_OK || (db->flags&SQLITE_NoSchemaError)){ + /* Black magic: If the SQLITE_NoSchemaError flag is set, then consider + ** the schema loaded, even if errors occurred. In this situation the + ** current sqlite3_prepare() operation will fail, but the following one + ** will attempt to compile the supplied statement against whatever subset + ** of the schema was loaded before the error occurred. The primary + ** purpose of this is to allow access to the sqlite_master table + ** even when its contents have been corrupted. + */ + DbSetProperty(db, iDb, DB_SchemaLoaded); + rc = SQLITE_OK; + } + + /* Jump here for an error that occurs after successfully allocating + ** curMain and calling sqlite3BtreeEnter(). For an error that occurs + ** before that point, jump to error_out. + */ +initone_error_out: + if( openedTransaction ){ + sqlite3BtreeCommit(pDb->pBt); + } + sqlite3BtreeLeave(pDb->pBt); + +error_out: + if( rc ){ + if( rc==SQLITE_NOMEM || rc==SQLITE_IOERR_NOMEM ){ + sqlite3OomFault(db); + } + sqlite3ResetOneSchema(db, iDb); + } + db->init.busy = 0; + return rc; +} + +/* +** Initialize all database files - the main database file, the file +** used to store temporary tables, and any additional database files +** created using ATTACH statements. Return a success code. If an +** error occurs, write an error message into *pzErrMsg. +** +** After a database is initialized, the DB_SchemaLoaded bit is set +** bit is set in the flags field of the Db structure. If the database +** file was of zero-length, then the DB_Empty flag is also set. +*/ +SQLITE_PRIVATE int sqlite3Init(sqlite3 *db, char **pzErrMsg){ + int i, rc; + int commit_internal = !(db->mDbFlags&DBFLAG_SchemaChange); + + assert( sqlite3_mutex_held(db->mutex) ); + assert( sqlite3BtreeHoldsMutex(db->aDb[0].pBt) ); + assert( db->init.busy==0 ); + ENC(db) = SCHEMA_ENC(db); + assert( db->nDb>0 ); + /* Do the main schema first */ + if( !DbHasProperty(db, 0, DB_SchemaLoaded) ){ + rc = sqlite3InitOne(db, 0, pzErrMsg, 0); + if( rc ) return rc; + } + /* All other schemas after the main schema. The "temp" schema must be last */ + for(i=db->nDb-1; i>0; i--){ + assert( i==1 || sqlite3BtreeHoldsMutex(db->aDb[i].pBt) ); + if( !DbHasProperty(db, i, DB_SchemaLoaded) ){ + rc = sqlite3InitOne(db, i, pzErrMsg, 0); + if( rc ) return rc; + } + } + if( commit_internal ){ + sqlite3CommitInternalChanges(db); + } + return SQLITE_OK; +} + +/* +** This routine is a no-op if the database schema is already initialized. +** Otherwise, the schema is loaded. An error code is returned. +*/ +SQLITE_PRIVATE int sqlite3ReadSchema(Parse *pParse){ + int rc = SQLITE_OK; + sqlite3 *db = pParse->db; + assert( sqlite3_mutex_held(db->mutex) ); + if( !db->init.busy ){ + rc = sqlite3Init(db, &pParse->zErrMsg); + if( rc!=SQLITE_OK ){ + pParse->rc = rc; + pParse->nErr++; + }else if( db->noSharedCache ){ + db->mDbFlags |= DBFLAG_SchemaKnownOk; + } + } + return rc; +} + + +/* +** Check schema cookies in all databases. If any cookie is out +** of date set pParse->rc to SQLITE_SCHEMA. If all schema cookies +** make no changes to pParse->rc. +*/ +static void schemaIsValid(Parse *pParse){ + sqlite3 *db = pParse->db; + int iDb; + int rc; + int cookie; + + assert( pParse->checkSchema ); + assert( sqlite3_mutex_held(db->mutex) ); + for(iDb=0; iDbnDb; iDb++){ + int openedTransaction = 0; /* True if a transaction is opened */ + Btree *pBt = db->aDb[iDb].pBt; /* Btree database to read cookie from */ + if( pBt==0 ) continue; + + /* If there is not already a read-only (or read-write) transaction opened + ** on the b-tree database, open one now. If a transaction is opened, it + ** will be closed immediately after reading the meta-value. */ + if( !sqlite3BtreeIsInReadTrans(pBt) ){ + rc = sqlite3BtreeBeginTrans(pBt, 0, 0); + if( rc==SQLITE_NOMEM || rc==SQLITE_IOERR_NOMEM ){ + sqlite3OomFault(db); + } + if( rc!=SQLITE_OK ) return; + openedTransaction = 1; + } + + /* Read the schema cookie from the database. If it does not match the + ** value stored as part of the in-memory schema representation, + ** set Parse.rc to SQLITE_SCHEMA. */ + sqlite3BtreeGetMeta(pBt, BTREE_SCHEMA_VERSION, (u32 *)&cookie); + assert( sqlite3SchemaMutexHeld(db, iDb, 0) ); + if( cookie!=db->aDb[iDb].pSchema->schema_cookie ){ + sqlite3ResetOneSchema(db, iDb); + pParse->rc = SQLITE_SCHEMA; + } + + /* Close the transaction, if one was opened. */ + if( openedTransaction ){ + sqlite3BtreeCommit(pBt); + } + } +} + +/* +** Convert a schema pointer into the iDb index that indicates +** which database file in db->aDb[] the schema refers to. +** +** If the same database is attached more than once, the first +** attached database is returned. +*/ +SQLITE_PRIVATE int sqlite3SchemaToIndex(sqlite3 *db, Schema *pSchema){ + int i = -1000000; + + /* If pSchema is NULL, then return -1000000. This happens when code in + ** expr.c is trying to resolve a reference to a transient table (i.e. one + ** created by a sub-select). In this case the return value of this + ** function should never be used. + ** + ** We return -1000000 instead of the more usual -1 simply because using + ** -1000000 as the incorrect index into db->aDb[] is much + ** more likely to cause a segfault than -1 (of course there are assert() + ** statements too, but it never hurts to play the odds). + */ + assert( sqlite3_mutex_held(db->mutex) ); + if( pSchema ){ + for(i=0; 1; i++){ + assert( inDb ); + if( db->aDb[i].pSchema==pSchema ){ + break; + } + } + assert( i>=0 && inDb ); + } + return i; +} + +/* +** Free all memory allocations in the pParse object +*/ +SQLITE_PRIVATE void sqlite3ParserReset(Parse *pParse){ + sqlite3 *db = pParse->db; + sqlite3DbFree(db, pParse->aLabel); + sqlite3ExprListDelete(db, pParse->pConstExpr); + if( db ){ + assert( db->lookaside.bDisable >= pParse->disableLookaside ); + db->lookaside.bDisable -= pParse->disableLookaside; + } + pParse->disableLookaside = 0; +} + +/* +** Compile the UTF-8 encoded SQL statement zSql into a statement handle. +*/ +static int sqlite3Prepare( + sqlite3 *db, /* Database handle. */ + const char *zSql, /* UTF-8 encoded SQL statement. */ + int nBytes, /* Length of zSql in bytes. */ + u32 prepFlags, /* Zero or more SQLITE_PREPARE_* flags */ + Vdbe *pReprepare, /* VM being reprepared */ + sqlite3_stmt **ppStmt, /* OUT: A pointer to the prepared statement */ + const char **pzTail /* OUT: End of parsed string */ +){ + char *zErrMsg = 0; /* Error message */ + int rc = SQLITE_OK; /* Result code */ + int i; /* Loop counter */ + Parse sParse; /* Parsing context */ + + memset(&sParse, 0, PARSE_HDR_SZ); + memset(PARSE_TAIL(&sParse), 0, PARSE_TAIL_SZ); + sParse.pReprepare = pReprepare; + assert( ppStmt && *ppStmt==0 ); + /* assert( !db->mallocFailed ); // not true with SQLITE_USE_ALLOCA */ + assert( sqlite3_mutex_held(db->mutex) ); + + /* For a long-term use prepared statement avoid the use of + ** lookaside memory. + */ + if( prepFlags & SQLITE_PREPARE_PERSISTENT ){ + sParse.disableLookaside++; + db->lookaside.bDisable++; + } + sParse.disableVtab = (prepFlags & SQLITE_PREPARE_NO_VTAB)!=0; + + /* Check to verify that it is possible to get a read lock on all + ** database schemas. The inability to get a read lock indicates that + ** some other database connection is holding a write-lock, which in + ** turn means that the other connection has made uncommitted changes + ** to the schema. + ** + ** Were we to proceed and prepare the statement against the uncommitted + ** schema changes and if those schema changes are subsequently rolled + ** back and different changes are made in their place, then when this + ** prepared statement goes to run the schema cookie would fail to detect + ** the schema change. Disaster would follow. + ** + ** This thread is currently holding mutexes on all Btrees (because + ** of the sqlite3BtreeEnterAll() in sqlite3LockAndPrepare()) so it + ** is not possible for another thread to start a new schema change + ** while this routine is running. Hence, we do not need to hold + ** locks on the schema, we just need to make sure nobody else is + ** holding them. + ** + ** Note that setting READ_UNCOMMITTED overrides most lock detection, + ** but it does *not* override schema lock detection, so this all still + ** works even if READ_UNCOMMITTED is set. + */ + for(i=0; inDb; i++) { + Btree *pBt = db->aDb[i].pBt; + if( pBt ){ + assert( sqlite3BtreeHoldsMutex(pBt) ); + rc = sqlite3BtreeSchemaLocked(pBt); + if( rc ){ + const char *zDb = db->aDb[i].zDbSName; + sqlite3ErrorWithMsg(db, rc, "database schema is locked: %s", zDb); + testcase( db->flags & SQLITE_ReadUncommit ); + goto end_prepare; + } + } + } + + sqlite3VtabUnlockList(db); + + sParse.db = db; + if( nBytes>=0 && (nBytes==0 || zSql[nBytes-1]!=0) ){ + char *zSqlCopy; + int mxLen = db->aLimit[SQLITE_LIMIT_SQL_LENGTH]; + testcase( nBytes==mxLen ); + testcase( nBytes==mxLen+1 ); + if( nBytes>mxLen ){ + sqlite3ErrorWithMsg(db, SQLITE_TOOBIG, "statement too long"); + rc = sqlite3ApiExit(db, SQLITE_TOOBIG); + goto end_prepare; + } + zSqlCopy = sqlite3DbStrNDup(db, zSql, nBytes); + if( zSqlCopy ){ + sqlite3RunParser(&sParse, zSqlCopy, &zErrMsg); + sParse.zTail = &zSql[sParse.zTail-zSqlCopy]; + sqlite3DbFree(db, zSqlCopy); + }else{ + sParse.zTail = &zSql[nBytes]; + } + }else{ + sqlite3RunParser(&sParse, zSql, &zErrMsg); + } + assert( 0==sParse.nQueryLoop ); + + if( sParse.rc==SQLITE_DONE ) sParse.rc = SQLITE_OK; + if( sParse.checkSchema ){ + schemaIsValid(&sParse); + } + if( db->mallocFailed ){ + sParse.rc = SQLITE_NOMEM_BKPT; + } + if( pzTail ){ + *pzTail = sParse.zTail; + } + rc = sParse.rc; + +#ifndef SQLITE_OMIT_EXPLAIN + /* Justification for the ALWAYS(): The only way for rc to be SQLITE_OK and + ** sParse.pVdbe to be NULL is if the input SQL is an empty string, but in + ** that case, sParse.explain will be false. */ + if( sParse.explain && rc==SQLITE_OK && ALWAYS(sParse.pVdbe) ){ + static const char * const azColName[] = { + "addr", "opcode", "p1", "p2", "p3", "p4", "p5", "comment", + "id", "parent", "notused", "detail" + }; + int iFirst, mx; + if( sParse.explain==2 ){ + sqlite3VdbeSetNumCols(sParse.pVdbe, 4); + iFirst = 8; + mx = 12; + }else{ + sqlite3VdbeSetNumCols(sParse.pVdbe, 8); + iFirst = 0; + mx = 8; + } + for(i=iFirst; iinit.busy==0 ){ + sqlite3VdbeSetSql(sParse.pVdbe, zSql, (int)(sParse.zTail-zSql), prepFlags); + } + if( rc!=SQLITE_OK || db->mallocFailed ){ + if( sParse.pVdbe ) sqlite3VdbeFinalize(sParse.pVdbe); + assert(!(*ppStmt)); + }else{ + *ppStmt = (sqlite3_stmt*)sParse.pVdbe; + } + + if( zErrMsg ){ + sqlite3ErrorWithMsg(db, rc, "%s", zErrMsg); + sqlite3DbFree(db, zErrMsg); + }else{ + sqlite3Error(db, rc); + } + + /* Delete any TriggerPrg structures allocated while parsing this statement. */ + while( sParse.pTriggerPrg ){ + TriggerPrg *pT = sParse.pTriggerPrg; + sParse.pTriggerPrg = pT->pNext; + sqlite3DbFree(db, pT); + } + +end_prepare: + + sqlite3ParserReset(&sParse); + return rc; +} +static int sqlite3LockAndPrepare( + sqlite3 *db, /* Database handle. */ + const char *zSql, /* UTF-8 encoded SQL statement. */ + int nBytes, /* Length of zSql in bytes. */ + u32 prepFlags, /* Zero or more SQLITE_PREPARE_* flags */ + Vdbe *pOld, /* VM being reprepared */ + sqlite3_stmt **ppStmt, /* OUT: A pointer to the prepared statement */ + const char **pzTail /* OUT: End of parsed string */ +){ + int rc; + int cnt = 0; + +#ifdef SQLITE_ENABLE_API_ARMOR + if( ppStmt==0 ) return SQLITE_MISUSE_BKPT; +#endif + *ppStmt = 0; + if( !sqlite3SafetyCheckOk(db)||zSql==0 ){ + return SQLITE_MISUSE_BKPT; + } + sqlite3_mutex_enter(db->mutex); + sqlite3BtreeEnterAll(db); + do{ + /* Make multiple attempts to compile the SQL, until it either succeeds + ** or encounters a permanent error. A schema problem after one schema + ** reset is considered a permanent error. */ + rc = sqlite3Prepare(db, zSql, nBytes, prepFlags, pOld, ppStmt, pzTail); + assert( rc==SQLITE_OK || *ppStmt==0 ); + }while( rc==SQLITE_ERROR_RETRY + || (rc==SQLITE_SCHEMA && (sqlite3ResetOneSchema(db,-1), cnt++)==0) ); + sqlite3BtreeLeaveAll(db); + rc = sqlite3ApiExit(db, rc); + assert( (rc&db->errMask)==rc ); + sqlite3_mutex_leave(db->mutex); + return rc; +} + + +/* +** Rerun the compilation of a statement after a schema change. +** +** If the statement is successfully recompiled, return SQLITE_OK. Otherwise, +** if the statement cannot be recompiled because another connection has +** locked the sqlite3_master table, return SQLITE_LOCKED. If any other error +** occurs, return SQLITE_SCHEMA. +*/ +SQLITE_PRIVATE int sqlite3Reprepare(Vdbe *p){ + int rc; + sqlite3_stmt *pNew; + const char *zSql; + sqlite3 *db; + u8 prepFlags; + + assert( sqlite3_mutex_held(sqlite3VdbeDb(p)->mutex) ); + zSql = sqlite3_sql((sqlite3_stmt *)p); + assert( zSql!=0 ); /* Reprepare only called for prepare_v2() statements */ + db = sqlite3VdbeDb(p); + assert( sqlite3_mutex_held(db->mutex) ); + prepFlags = sqlite3VdbePrepareFlags(p); + rc = sqlite3LockAndPrepare(db, zSql, -1, prepFlags, p, &pNew, 0); + if( rc ){ + if( rc==SQLITE_NOMEM ){ + sqlite3OomFault(db); + } + assert( pNew==0 ); + return rc; + }else{ + assert( pNew!=0 ); + } + sqlite3VdbeSwap((Vdbe*)pNew, p); + sqlite3TransferBindings(pNew, (sqlite3_stmt*)p); + sqlite3VdbeResetStepResult((Vdbe*)pNew); + sqlite3VdbeFinalize((Vdbe*)pNew); + return SQLITE_OK; +} + + +/* +** Two versions of the official API. Legacy and new use. In the legacy +** version, the original SQL text is not saved in the prepared statement +** and so if a schema change occurs, SQLITE_SCHEMA is returned by +** sqlite3_step(). In the new version, the original SQL text is retained +** and the statement is automatically recompiled if an schema change +** occurs. +*/ +SQLITE_API int sqlite3_prepare( + sqlite3 *db, /* Database handle. */ + const char *zSql, /* UTF-8 encoded SQL statement. */ + int nBytes, /* Length of zSql in bytes. */ + sqlite3_stmt **ppStmt, /* OUT: A pointer to the prepared statement */ + const char **pzTail /* OUT: End of parsed string */ +){ + int rc; + rc = sqlite3LockAndPrepare(db,zSql,nBytes,0,0,ppStmt,pzTail); + assert( rc==SQLITE_OK || ppStmt==0 || *ppStmt==0 ); /* VERIFY: F13021 */ + return rc; +} +SQLITE_API int sqlite3_prepare_v2( + sqlite3 *db, /* Database handle. */ + const char *zSql, /* UTF-8 encoded SQL statement. */ + int nBytes, /* Length of zSql in bytes. */ + sqlite3_stmt **ppStmt, /* OUT: A pointer to the prepared statement */ + const char **pzTail /* OUT: End of parsed string */ +){ + int rc; + /* EVIDENCE-OF: R-37923-12173 The sqlite3_prepare_v2() interface works + ** exactly the same as sqlite3_prepare_v3() with a zero prepFlags + ** parameter. + ** + ** Proof in that the 5th parameter to sqlite3LockAndPrepare is 0 */ + rc = sqlite3LockAndPrepare(db,zSql,nBytes,SQLITE_PREPARE_SAVESQL,0, + ppStmt,pzTail); + assert( rc==SQLITE_OK || ppStmt==0 || *ppStmt==0 ); + return rc; +} +SQLITE_API int sqlite3_prepare_v3( + sqlite3 *db, /* Database handle. */ + const char *zSql, /* UTF-8 encoded SQL statement. */ + int nBytes, /* Length of zSql in bytes. */ + unsigned int prepFlags, /* Zero or more SQLITE_PREPARE_* flags */ + sqlite3_stmt **ppStmt, /* OUT: A pointer to the prepared statement */ + const char **pzTail /* OUT: End of parsed string */ +){ + int rc; + /* EVIDENCE-OF: R-56861-42673 sqlite3_prepare_v3() differs from + ** sqlite3_prepare_v2() only in having the extra prepFlags parameter, + ** which is a bit array consisting of zero or more of the + ** SQLITE_PREPARE_* flags. + ** + ** Proof by comparison to the implementation of sqlite3_prepare_v2() + ** directly above. */ + rc = sqlite3LockAndPrepare(db,zSql,nBytes, + SQLITE_PREPARE_SAVESQL|(prepFlags&SQLITE_PREPARE_MASK), + 0,ppStmt,pzTail); + assert( rc==SQLITE_OK || ppStmt==0 || *ppStmt==0 ); + return rc; +} + + +#ifndef SQLITE_OMIT_UTF16 +/* +** Compile the UTF-16 encoded SQL statement zSql into a statement handle. +*/ +static int sqlite3Prepare16( + sqlite3 *db, /* Database handle. */ + const void *zSql, /* UTF-16 encoded SQL statement. */ + int nBytes, /* Length of zSql in bytes. */ + u32 prepFlags, /* Zero or more SQLITE_PREPARE_* flags */ + sqlite3_stmt **ppStmt, /* OUT: A pointer to the prepared statement */ + const void **pzTail /* OUT: End of parsed string */ +){ + /* This function currently works by first transforming the UTF-16 + ** encoded string to UTF-8, then invoking sqlite3_prepare(). The + ** tricky bit is figuring out the pointer to return in *pzTail. + */ + char *zSql8; + const char *zTail8 = 0; + int rc = SQLITE_OK; + +#ifdef SQLITE_ENABLE_API_ARMOR + if( ppStmt==0 ) return SQLITE_MISUSE_BKPT; +#endif + *ppStmt = 0; + if( !sqlite3SafetyCheckOk(db)||zSql==0 ){ + return SQLITE_MISUSE_BKPT; + } + if( nBytes>=0 ){ + int sz; + const char *z = (const char*)zSql; + for(sz=0; szmutex); + zSql8 = sqlite3Utf16to8(db, zSql, nBytes, SQLITE_UTF16NATIVE); + if( zSql8 ){ + rc = sqlite3LockAndPrepare(db, zSql8, -1, prepFlags, 0, ppStmt, &zTail8); + } + + if( zTail8 && pzTail ){ + /* If sqlite3_prepare returns a tail pointer, we calculate the + ** equivalent pointer into the UTF-16 string by counting the unicode + ** characters between zSql8 and zTail8, and then returning a pointer + ** the same number of characters into the UTF-16 string. + */ + int chars_parsed = sqlite3Utf8CharLen(zSql8, (int)(zTail8-zSql8)); + *pzTail = (u8 *)zSql + sqlite3Utf16ByteLen(zSql, chars_parsed); + } + sqlite3DbFree(db, zSql8); + rc = sqlite3ApiExit(db, rc); + sqlite3_mutex_leave(db->mutex); + return rc; +} + +/* +** Two versions of the official API. Legacy and new use. In the legacy +** version, the original SQL text is not saved in the prepared statement +** and so if a schema change occurs, SQLITE_SCHEMA is returned by +** sqlite3_step(). In the new version, the original SQL text is retained +** and the statement is automatically recompiled if an schema change +** occurs. +*/ +SQLITE_API int sqlite3_prepare16( + sqlite3 *db, /* Database handle. */ + const void *zSql, /* UTF-16 encoded SQL statement. */ + int nBytes, /* Length of zSql in bytes. */ + sqlite3_stmt **ppStmt, /* OUT: A pointer to the prepared statement */ + const void **pzTail /* OUT: End of parsed string */ +){ + int rc; + rc = sqlite3Prepare16(db,zSql,nBytes,0,ppStmt,pzTail); + assert( rc==SQLITE_OK || ppStmt==0 || *ppStmt==0 ); /* VERIFY: F13021 */ + return rc; +} +SQLITE_API int sqlite3_prepare16_v2( + sqlite3 *db, /* Database handle. */ + const void *zSql, /* UTF-16 encoded SQL statement. */ + int nBytes, /* Length of zSql in bytes. */ + sqlite3_stmt **ppStmt, /* OUT: A pointer to the prepared statement */ + const void **pzTail /* OUT: End of parsed string */ +){ + int rc; + rc = sqlite3Prepare16(db,zSql,nBytes,SQLITE_PREPARE_SAVESQL,ppStmt,pzTail); + assert( rc==SQLITE_OK || ppStmt==0 || *ppStmt==0 ); /* VERIFY: F13021 */ + return rc; +} +SQLITE_API int sqlite3_prepare16_v3( + sqlite3 *db, /* Database handle. */ + const void *zSql, /* UTF-16 encoded SQL statement. */ + int nBytes, /* Length of zSql in bytes. */ + unsigned int prepFlags, /* Zero or more SQLITE_PREPARE_* flags */ + sqlite3_stmt **ppStmt, /* OUT: A pointer to the prepared statement */ + const void **pzTail /* OUT: End of parsed string */ +){ + int rc; + rc = sqlite3Prepare16(db,zSql,nBytes, + SQLITE_PREPARE_SAVESQL|(prepFlags&SQLITE_PREPARE_MASK), + ppStmt,pzTail); + assert( rc==SQLITE_OK || ppStmt==0 || *ppStmt==0 ); /* VERIFY: F13021 */ + return rc; +} + +#endif /* SQLITE_OMIT_UTF16 */ + +/************** End of prepare.c *********************************************/ +/************** Begin file select.c ******************************************/ +/* +** 2001 September 15 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** This file contains C code routines that are called by the parser +** to handle SELECT statements in SQLite. +*/ +/* #include "sqliteInt.h" */ + +/* +** Trace output macros +*/ +#if SELECTTRACE_ENABLED +/***/ int sqlite3SelectTrace = 0; +# define SELECTTRACE(K,P,S,X) \ + if(sqlite3SelectTrace&(K)) \ + sqlite3DebugPrintf("%u/%d/%p: ",(S)->selId,(P)->addrExplain,(S)),\ + sqlite3DebugPrintf X +#else +# define SELECTTRACE(K,P,S,X) +#endif + + +/* +** An instance of the following object is used to record information about +** how to process the DISTINCT keyword, to simplify passing that information +** into the selectInnerLoop() routine. +*/ +typedef struct DistinctCtx DistinctCtx; +struct DistinctCtx { + u8 isTnct; /* True if the DISTINCT keyword is present */ + u8 eTnctType; /* One of the WHERE_DISTINCT_* operators */ + int tabTnct; /* Ephemeral table used for DISTINCT processing */ + int addrTnct; /* Address of OP_OpenEphemeral opcode for tabTnct */ +}; + +/* +** An instance of the following object is used to record information about +** the ORDER BY (or GROUP BY) clause of query is being coded. +** +** The aDefer[] array is used by the sorter-references optimization. For +** example, assuming there is no index that can be used for the ORDER BY, +** for the query: +** +** SELECT a, bigblob FROM t1 ORDER BY a LIMIT 10; +** +** it may be more efficient to add just the "a" values to the sorter, and +** retrieve the associated "bigblob" values directly from table t1 as the +** 10 smallest "a" values are extracted from the sorter. +** +** When the sorter-reference optimization is used, there is one entry in the +** aDefer[] array for each database table that may be read as values are +** extracted from the sorter. +*/ +typedef struct SortCtx SortCtx; +struct SortCtx { + ExprList *pOrderBy; /* The ORDER BY (or GROUP BY clause) */ + int nOBSat; /* Number of ORDER BY terms satisfied by indices */ + int iECursor; /* Cursor number for the sorter */ + int regReturn; /* Register holding block-output return address */ + int labelBkOut; /* Start label for the block-output subroutine */ + int addrSortIndex; /* Address of the OP_SorterOpen or OP_OpenEphemeral */ + int labelDone; /* Jump here when done, ex: LIMIT reached */ + int labelOBLopt; /* Jump here when sorter is full */ + u8 sortFlags; /* Zero or more SORTFLAG_* bits */ +#ifdef SQLITE_ENABLE_SORTER_REFERENCES + u8 nDefer; /* Number of valid entries in aDefer[] */ + struct DeferredCsr { + Table *pTab; /* Table definition */ + int iCsr; /* Cursor number for table */ + int nKey; /* Number of PK columns for table pTab (>=1) */ + } aDefer[4]; +#endif + struct RowLoadInfo *pDeferredRowLoad; /* Deferred row loading info or NULL */ +}; +#define SORTFLAG_UseSorter 0x01 /* Use SorterOpen instead of OpenEphemeral */ + +/* +** Delete all the content of a Select structure. Deallocate the structure +** itself only if bFree is true. +*/ +static void clearSelect(sqlite3 *db, Select *p, int bFree){ + while( p ){ + Select *pPrior = p->pPrior; + sqlite3ExprListDelete(db, p->pEList); + sqlite3SrcListDelete(db, p->pSrc); + sqlite3ExprDelete(db, p->pWhere); + sqlite3ExprListDelete(db, p->pGroupBy); + sqlite3ExprDelete(db, p->pHaving); + sqlite3ExprListDelete(db, p->pOrderBy); + sqlite3ExprDelete(db, p->pLimit); +#ifndef SQLITE_OMIT_WINDOWFUNC + if( OK_IF_ALWAYS_TRUE(p->pWinDefn) ){ + sqlite3WindowListDelete(db, p->pWinDefn); + } + assert( p->pWin==0 ); +#endif + if( OK_IF_ALWAYS_TRUE(p->pWith) ) sqlite3WithDelete(db, p->pWith); + if( bFree ) sqlite3DbFreeNN(db, p); + p = pPrior; + bFree = 1; + } +} + +/* +** Initialize a SelectDest structure. +*/ +SQLITE_PRIVATE void sqlite3SelectDestInit(SelectDest *pDest, int eDest, int iParm){ + pDest->eDest = (u8)eDest; + pDest->iSDParm = iParm; + pDest->zAffSdst = 0; + pDest->iSdst = 0; + pDest->nSdst = 0; +} + + +/* +** Allocate a new Select structure and return a pointer to that +** structure. +*/ +SQLITE_PRIVATE Select *sqlite3SelectNew( + Parse *pParse, /* Parsing context */ + ExprList *pEList, /* which columns to include in the result */ + SrcList *pSrc, /* the FROM clause -- which tables to scan */ + Expr *pWhere, /* the WHERE clause */ + ExprList *pGroupBy, /* the GROUP BY clause */ + Expr *pHaving, /* the HAVING clause */ + ExprList *pOrderBy, /* the ORDER BY clause */ + u32 selFlags, /* Flag parameters, such as SF_Distinct */ + Expr *pLimit /* LIMIT value. NULL means not used */ +){ + Select *pNew; + Select standin; + pNew = sqlite3DbMallocRawNN(pParse->db, sizeof(*pNew) ); + if( pNew==0 ){ + assert( pParse->db->mallocFailed ); + pNew = &standin; + } + if( pEList==0 ){ + pEList = sqlite3ExprListAppend(pParse, 0, + sqlite3Expr(pParse->db,TK_ASTERISK,0)); + } + pNew->pEList = pEList; + pNew->op = TK_SELECT; + pNew->selFlags = selFlags; + pNew->iLimit = 0; + pNew->iOffset = 0; + pNew->selId = ++pParse->nSelect; + pNew->addrOpenEphm[0] = -1; + pNew->addrOpenEphm[1] = -1; + pNew->nSelectRow = 0; + if( pSrc==0 ) pSrc = sqlite3DbMallocZero(pParse->db, sizeof(*pSrc)); + pNew->pSrc = pSrc; + pNew->pWhere = pWhere; + pNew->pGroupBy = pGroupBy; + pNew->pHaving = pHaving; + pNew->pOrderBy = pOrderBy; + pNew->pPrior = 0; + pNew->pNext = 0; + pNew->pLimit = pLimit; + pNew->pWith = 0; +#ifndef SQLITE_OMIT_WINDOWFUNC + pNew->pWin = 0; + pNew->pWinDefn = 0; +#endif + if( pParse->db->mallocFailed ) { + clearSelect(pParse->db, pNew, pNew!=&standin); + pNew = 0; + }else{ + assert( pNew->pSrc!=0 || pParse->nErr>0 ); + } + assert( pNew!=&standin ); + return pNew; +} + + +/* +** Delete the given Select structure and all of its substructures. +*/ +SQLITE_PRIVATE void sqlite3SelectDelete(sqlite3 *db, Select *p){ + if( OK_IF_ALWAYS_TRUE(p) ) clearSelect(db, p, 1); +} + +/* +** Return a pointer to the right-most SELECT statement in a compound. +*/ +static Select *findRightmost(Select *p){ + while( p->pNext ) p = p->pNext; + return p; +} + +/* +** Given 1 to 3 identifiers preceding the JOIN keyword, determine the +** type of join. Return an integer constant that expresses that type +** in terms of the following bit values: +** +** JT_INNER +** JT_CROSS +** JT_OUTER +** JT_NATURAL +** JT_LEFT +** JT_RIGHT +** +** A full outer join is the combination of JT_LEFT and JT_RIGHT. +** +** If an illegal or unsupported join type is seen, then still return +** a join type, but put an error in the pParse structure. +*/ +SQLITE_PRIVATE int sqlite3JoinType(Parse *pParse, Token *pA, Token *pB, Token *pC){ + int jointype = 0; + Token *apAll[3]; + Token *p; + /* 0123456789 123456789 123456789 123 */ + static const char zKeyText[] = "naturaleftouterightfullinnercross"; + static const struct { + u8 i; /* Beginning of keyword text in zKeyText[] */ + u8 nChar; /* Length of the keyword in characters */ + u8 code; /* Join type mask */ + } aKeyword[] = { + /* natural */ { 0, 7, JT_NATURAL }, + /* left */ { 6, 4, JT_LEFT|JT_OUTER }, + /* outer */ { 10, 5, JT_OUTER }, + /* right */ { 14, 5, JT_RIGHT|JT_OUTER }, + /* full */ { 19, 4, JT_LEFT|JT_RIGHT|JT_OUTER }, + /* inner */ { 23, 5, JT_INNER }, + /* cross */ { 28, 5, JT_INNER|JT_CROSS }, + }; + int i, j; + apAll[0] = pA; + apAll[1] = pB; + apAll[2] = pC; + for(i=0; i<3 && apAll[i]; i++){ + p = apAll[i]; + for(j=0; jn==aKeyword[j].nChar + && sqlite3StrNICmp((char*)p->z, &zKeyText[aKeyword[j].i], p->n)==0 ){ + jointype |= aKeyword[j].code; + break; + } + } + testcase( j==0 || j==1 || j==2 || j==3 || j==4 || j==5 || j==6 ); + if( j>=ArraySize(aKeyword) ){ + jointype |= JT_ERROR; + break; + } + } + if( + (jointype & (JT_INNER|JT_OUTER))==(JT_INNER|JT_OUTER) || + (jointype & JT_ERROR)!=0 + ){ + const char *zSp = " "; + assert( pB!=0 ); + if( pC==0 ){ zSp++; } + sqlite3ErrorMsg(pParse, "unknown or unsupported join type: " + "%T %T%s%T", pA, pB, zSp, pC); + jointype = JT_INNER; + }else if( (jointype & JT_OUTER)!=0 + && (jointype & (JT_LEFT|JT_RIGHT))!=JT_LEFT ){ + sqlite3ErrorMsg(pParse, + "RIGHT and FULL OUTER JOINs are not currently supported"); + jointype = JT_INNER; + } + return jointype; +} + +/* +** Return the index of a column in a table. Return -1 if the column +** is not contained in the table. +*/ +static int columnIndex(Table *pTab, const char *zCol){ + int i; + for(i=0; inCol; i++){ + if( sqlite3StrICmp(pTab->aCol[i].zName, zCol)==0 ) return i; + } + return -1; +} + +/* +** Search the first N tables in pSrc, from left to right, looking for a +** table that has a column named zCol. +** +** When found, set *piTab and *piCol to the table index and column index +** of the matching column and return TRUE. +** +** If not found, return FALSE. +*/ +static int tableAndColumnIndex( + SrcList *pSrc, /* Array of tables to search */ + int N, /* Number of tables in pSrc->a[] to search */ + const char *zCol, /* Name of the column we are looking for */ + int *piTab, /* Write index of pSrc->a[] here */ + int *piCol /* Write index of pSrc->a[*piTab].pTab->aCol[] here */ +){ + int i; /* For looping over tables in pSrc */ + int iCol; /* Index of column matching zCol */ + + assert( (piTab==0)==(piCol==0) ); /* Both or neither are NULL */ + for(i=0; ia[i].pTab, zCol); + if( iCol>=0 ){ + if( piTab ){ + *piTab = i; + *piCol = iCol; + } + return 1; + } + } + return 0; +} + +/* +** This function is used to add terms implied by JOIN syntax to the +** WHERE clause expression of a SELECT statement. The new term, which +** is ANDed with the existing WHERE clause, is of the form: +** +** (tab1.col1 = tab2.col2) +** +** where tab1 is the iSrc'th table in SrcList pSrc and tab2 is the +** (iSrc+1)'th. Column col1 is column iColLeft of tab1, and col2 is +** column iColRight of tab2. +*/ +static void addWhereTerm( + Parse *pParse, /* Parsing context */ + SrcList *pSrc, /* List of tables in FROM clause */ + int iLeft, /* Index of first table to join in pSrc */ + int iColLeft, /* Index of column in first table */ + int iRight, /* Index of second table in pSrc */ + int iColRight, /* Index of column in second table */ + int isOuterJoin, /* True if this is an OUTER join */ + Expr **ppWhere /* IN/OUT: The WHERE clause to add to */ +){ + sqlite3 *db = pParse->db; + Expr *pE1; + Expr *pE2; + Expr *pEq; + + assert( iLeftnSrc>iRight ); + assert( pSrc->a[iLeft].pTab ); + assert( pSrc->a[iRight].pTab ); + + pE1 = sqlite3CreateColumnExpr(db, pSrc, iLeft, iColLeft); + pE2 = sqlite3CreateColumnExpr(db, pSrc, iRight, iColRight); + + pEq = sqlite3PExpr(pParse, TK_EQ, pE1, pE2); + if( pEq && isOuterJoin ){ + ExprSetProperty(pEq, EP_FromJoin); + assert( !ExprHasProperty(pEq, EP_TokenOnly|EP_Reduced) ); + ExprSetVVAProperty(pEq, EP_NoReduce); + pEq->iRightJoinTable = (i16)pE2->iTable; + } + *ppWhere = sqlite3ExprAnd(pParse, *ppWhere, pEq); +} + +/* +** Set the EP_FromJoin property on all terms of the given expression. +** And set the Expr.iRightJoinTable to iTable for every term in the +** expression. +** +** The EP_FromJoin property is used on terms of an expression to tell +** the LEFT OUTER JOIN processing logic that this term is part of the +** join restriction specified in the ON or USING clause and not a part +** of the more general WHERE clause. These terms are moved over to the +** WHERE clause during join processing but we need to remember that they +** originated in the ON or USING clause. +** +** The Expr.iRightJoinTable tells the WHERE clause processing that the +** expression depends on table iRightJoinTable even if that table is not +** explicitly mentioned in the expression. That information is needed +** for cases like this: +** +** SELECT * FROM t1 LEFT JOIN t2 ON t1.a=t2.b AND t1.x=5 +** +** The where clause needs to defer the handling of the t1.x=5 +** term until after the t2 loop of the join. In that way, a +** NULL t2 row will be inserted whenever t1.x!=5. If we do not +** defer the handling of t1.x=5, it will be processed immediately +** after the t1 loop and rows with t1.x!=5 will never appear in +** the output, which is incorrect. +*/ +static void setJoinExpr(Expr *p, int iTable){ + while( p ){ + ExprSetProperty(p, EP_FromJoin); + assert( !ExprHasProperty(p, EP_TokenOnly|EP_Reduced) ); + ExprSetVVAProperty(p, EP_NoReduce); + p->iRightJoinTable = (i16)iTable; + if( p->op==TK_FUNCTION && p->x.pList ){ + int i; + for(i=0; ix.pList->nExpr; i++){ + setJoinExpr(p->x.pList->a[i].pExpr, iTable); + } + } + setJoinExpr(p->pLeft, iTable); + p = p->pRight; + } +} + +/* Undo the work of setJoinExpr(). In the expression tree p, convert every +** term that is marked with EP_FromJoin and iRightJoinTable==iTable into +** an ordinary term that omits the EP_FromJoin mark. +** +** This happens when a LEFT JOIN is simplified into an ordinary JOIN. +*/ +static void unsetJoinExpr(Expr *p, int iTable){ + while( p ){ + if( ExprHasProperty(p, EP_FromJoin) + && (iTable<0 || p->iRightJoinTable==iTable) ){ + ExprClearProperty(p, EP_FromJoin); + } + if( p->op==TK_FUNCTION && p->x.pList ){ + int i; + for(i=0; ix.pList->nExpr; i++){ + unsetJoinExpr(p->x.pList->a[i].pExpr, iTable); + } + } + unsetJoinExpr(p->pLeft, iTable); + p = p->pRight; + } +} + +/* +** This routine processes the join information for a SELECT statement. +** ON and USING clauses are converted into extra terms of the WHERE clause. +** NATURAL joins also create extra WHERE clause terms. +** +** The terms of a FROM clause are contained in the Select.pSrc structure. +** The left most table is the first entry in Select.pSrc. The right-most +** table is the last entry. The join operator is held in the entry to +** the left. Thus entry 0 contains the join operator for the join between +** entries 0 and 1. Any ON or USING clauses associated with the join are +** also attached to the left entry. +** +** This routine returns the number of errors encountered. +*/ +static int sqliteProcessJoin(Parse *pParse, Select *p){ + SrcList *pSrc; /* All tables in the FROM clause */ + int i, j; /* Loop counters */ + struct SrcList_item *pLeft; /* Left table being joined */ + struct SrcList_item *pRight; /* Right table being joined */ + + pSrc = p->pSrc; + pLeft = &pSrc->a[0]; + pRight = &pLeft[1]; + for(i=0; inSrc-1; i++, pRight++, pLeft++){ + Table *pRightTab = pRight->pTab; + int isOuter; + + if( NEVER(pLeft->pTab==0 || pRightTab==0) ) continue; + isOuter = (pRight->fg.jointype & JT_OUTER)!=0; + + /* When the NATURAL keyword is present, add WHERE clause terms for + ** every column that the two tables have in common. + */ + if( pRight->fg.jointype & JT_NATURAL ){ + if( pRight->pOn || pRight->pUsing ){ + sqlite3ErrorMsg(pParse, "a NATURAL join may not have " + "an ON or USING clause", 0); + return 1; + } + for(j=0; jnCol; j++){ + char *zName; /* Name of column in the right table */ + int iLeft; /* Matching left table */ + int iLeftCol; /* Matching column in the left table */ + + zName = pRightTab->aCol[j].zName; + if( tableAndColumnIndex(pSrc, i+1, zName, &iLeft, &iLeftCol) ){ + addWhereTerm(pParse, pSrc, iLeft, iLeftCol, i+1, j, + isOuter, &p->pWhere); + } + } + } + + /* Disallow both ON and USING clauses in the same join + */ + if( pRight->pOn && pRight->pUsing ){ + sqlite3ErrorMsg(pParse, "cannot have both ON and USING " + "clauses in the same join"); + return 1; + } + + /* Add the ON clause to the end of the WHERE clause, connected by + ** an AND operator. + */ + if( pRight->pOn ){ + if( isOuter ) setJoinExpr(pRight->pOn, pRight->iCursor); + p->pWhere = sqlite3ExprAnd(pParse, p->pWhere, pRight->pOn); + pRight->pOn = 0; + } + + /* Create extra terms on the WHERE clause for each column named + ** in the USING clause. Example: If the two tables to be joined are + ** A and B and the USING clause names X, Y, and Z, then add this + ** to the WHERE clause: A.X=B.X AND A.Y=B.Y AND A.Z=B.Z + ** Report an error if any column mentioned in the USING clause is + ** not contained in both tables to be joined. + */ + if( pRight->pUsing ){ + IdList *pList = pRight->pUsing; + for(j=0; jnId; j++){ + char *zName; /* Name of the term in the USING clause */ + int iLeft; /* Table on the left with matching column name */ + int iLeftCol; /* Column number of matching column on the left */ + int iRightCol; /* Column number of matching column on the right */ + + zName = pList->a[j].zName; + iRightCol = columnIndex(pRightTab, zName); + if( iRightCol<0 + || !tableAndColumnIndex(pSrc, i+1, zName, &iLeft, &iLeftCol) + ){ + sqlite3ErrorMsg(pParse, "cannot join using column %s - column " + "not present in both tables", zName); + return 1; + } + addWhereTerm(pParse, pSrc, iLeft, iLeftCol, i+1, iRightCol, + isOuter, &p->pWhere); + } + } + } + return 0; +} + +/* +** An instance of this object holds information (beyond pParse and pSelect) +** needed to load the next result row that is to be added to the sorter. +*/ +typedef struct RowLoadInfo RowLoadInfo; +struct RowLoadInfo { + int regResult; /* Store results in array of registers here */ + u8 ecelFlags; /* Flag argument to ExprCodeExprList() */ +#ifdef SQLITE_ENABLE_SORTER_REFERENCES + ExprList *pExtra; /* Extra columns needed by sorter refs */ + int regExtraResult; /* Where to load the extra columns */ +#endif +}; + +/* +** This routine does the work of loading query data into an array of +** registers so that it can be added to the sorter. +*/ +static void innerLoopLoadRow( + Parse *pParse, /* Statement under construction */ + Select *pSelect, /* The query being coded */ + RowLoadInfo *pInfo /* Info needed to complete the row load */ +){ + sqlite3ExprCodeExprList(pParse, pSelect->pEList, pInfo->regResult, + 0, pInfo->ecelFlags); +#ifdef SQLITE_ENABLE_SORTER_REFERENCES + if( pInfo->pExtra ){ + sqlite3ExprCodeExprList(pParse, pInfo->pExtra, pInfo->regExtraResult, 0, 0); + sqlite3ExprListDelete(pParse->db, pInfo->pExtra); + } +#endif +} + +/* +** Code the OP_MakeRecord instruction that generates the entry to be +** added into the sorter. +** +** Return the register in which the result is stored. +*/ +static int makeSorterRecord( + Parse *pParse, + SortCtx *pSort, + Select *pSelect, + int regBase, + int nBase +){ + int nOBSat = pSort->nOBSat; + Vdbe *v = pParse->pVdbe; + int regOut = ++pParse->nMem; + if( pSort->pDeferredRowLoad ){ + innerLoopLoadRow(pParse, pSelect, pSort->pDeferredRowLoad); + } + sqlite3VdbeAddOp3(v, OP_MakeRecord, regBase+nOBSat, nBase-nOBSat, regOut); + return regOut; +} + +/* +** Generate code that will push the record in registers regData +** through regData+nData-1 onto the sorter. +*/ +static void pushOntoSorter( + Parse *pParse, /* Parser context */ + SortCtx *pSort, /* Information about the ORDER BY clause */ + Select *pSelect, /* The whole SELECT statement */ + int regData, /* First register holding data to be sorted */ + int regOrigData, /* First register holding data before packing */ + int nData, /* Number of elements in the regData data array */ + int nPrefixReg /* No. of reg prior to regData available for use */ +){ + Vdbe *v = pParse->pVdbe; /* Stmt under construction */ + int bSeq = ((pSort->sortFlags & SORTFLAG_UseSorter)==0); + int nExpr = pSort->pOrderBy->nExpr; /* No. of ORDER BY terms */ + int nBase = nExpr + bSeq + nData; /* Fields in sorter record */ + int regBase; /* Regs for sorter record */ + int regRecord = 0; /* Assembled sorter record */ + int nOBSat = pSort->nOBSat; /* ORDER BY terms to skip */ + int op; /* Opcode to add sorter record to sorter */ + int iLimit; /* LIMIT counter */ + int iSkip = 0; /* End of the sorter insert loop */ + + assert( bSeq==0 || bSeq==1 ); + + /* Three cases: + ** (1) The data to be sorted has already been packed into a Record + ** by a prior OP_MakeRecord. In this case nData==1 and regData + ** will be completely unrelated to regOrigData. + ** (2) All output columns are included in the sort record. In that + ** case regData==regOrigData. + ** (3) Some output columns are omitted from the sort record due to + ** the SQLITE_ENABLE_SORTER_REFERENCE optimization, or due to the + ** SQLITE_ECEL_OMITREF optimization, or due to the + ** SortCtx.pDeferredRowLoad optimiation. In any of these cases + ** regOrigData is 0 to prevent this routine from trying to copy + ** values that might not yet exist. + */ + assert( nData==1 || regData==regOrigData || regOrigData==0 ); + + if( nPrefixReg ){ + assert( nPrefixReg==nExpr+bSeq ); + regBase = regData - nPrefixReg; + }else{ + regBase = pParse->nMem + 1; + pParse->nMem += nBase; + } + assert( pSelect->iOffset==0 || pSelect->iLimit!=0 ); + iLimit = pSelect->iOffset ? pSelect->iOffset+1 : pSelect->iLimit; + pSort->labelDone = sqlite3VdbeMakeLabel(pParse); + sqlite3ExprCodeExprList(pParse, pSort->pOrderBy, regBase, regOrigData, + SQLITE_ECEL_DUP | (regOrigData? SQLITE_ECEL_REF : 0)); + if( bSeq ){ + sqlite3VdbeAddOp2(v, OP_Sequence, pSort->iECursor, regBase+nExpr); + } + if( nPrefixReg==0 && nData>0 ){ + sqlite3ExprCodeMove(pParse, regData, regBase+nExpr+bSeq, nData); + } + if( nOBSat>0 ){ + int regPrevKey; /* The first nOBSat columns of the previous row */ + int addrFirst; /* Address of the OP_IfNot opcode */ + int addrJmp; /* Address of the OP_Jump opcode */ + VdbeOp *pOp; /* Opcode that opens the sorter */ + int nKey; /* Number of sorting key columns, including OP_Sequence */ + KeyInfo *pKI; /* Original KeyInfo on the sorter table */ + + regRecord = makeSorterRecord(pParse, pSort, pSelect, regBase, nBase); + regPrevKey = pParse->nMem+1; + pParse->nMem += pSort->nOBSat; + nKey = nExpr - pSort->nOBSat + bSeq; + if( bSeq ){ + addrFirst = sqlite3VdbeAddOp1(v, OP_IfNot, regBase+nExpr); + }else{ + addrFirst = sqlite3VdbeAddOp1(v, OP_SequenceTest, pSort->iECursor); + } + VdbeCoverage(v); + sqlite3VdbeAddOp3(v, OP_Compare, regPrevKey, regBase, pSort->nOBSat); + pOp = sqlite3VdbeGetOp(v, pSort->addrSortIndex); + if( pParse->db->mallocFailed ) return; + pOp->p2 = nKey + nData; + pKI = pOp->p4.pKeyInfo; + memset(pKI->aSortFlags, 0, pKI->nKeyField); /* Makes OP_Jump testable */ + sqlite3VdbeChangeP4(v, -1, (char*)pKI, P4_KEYINFO); + testcase( pKI->nAllField > pKI->nKeyField+2 ); + pOp->p4.pKeyInfo = sqlite3KeyInfoFromExprList(pParse,pSort->pOrderBy,nOBSat, + pKI->nAllField-pKI->nKeyField-1); + addrJmp = sqlite3VdbeCurrentAddr(v); + sqlite3VdbeAddOp3(v, OP_Jump, addrJmp+1, 0, addrJmp+1); VdbeCoverage(v); + pSort->labelBkOut = sqlite3VdbeMakeLabel(pParse); + pSort->regReturn = ++pParse->nMem; + sqlite3VdbeAddOp2(v, OP_Gosub, pSort->regReturn, pSort->labelBkOut); + sqlite3VdbeAddOp1(v, OP_ResetSorter, pSort->iECursor); + if( iLimit ){ + sqlite3VdbeAddOp2(v, OP_IfNot, iLimit, pSort->labelDone); + VdbeCoverage(v); + } + sqlite3VdbeJumpHere(v, addrFirst); + sqlite3ExprCodeMove(pParse, regBase, regPrevKey, pSort->nOBSat); + sqlite3VdbeJumpHere(v, addrJmp); + } + if( iLimit ){ + /* At this point the values for the new sorter entry are stored + ** in an array of registers. They need to be composed into a record + ** and inserted into the sorter if either (a) there are currently + ** less than LIMIT+OFFSET items or (b) the new record is smaller than + ** the largest record currently in the sorter. If (b) is true and there + ** are already LIMIT+OFFSET items in the sorter, delete the largest + ** entry before inserting the new one. This way there are never more + ** than LIMIT+OFFSET items in the sorter. + ** + ** If the new record does not need to be inserted into the sorter, + ** jump to the next iteration of the loop. If the pSort->labelOBLopt + ** value is not zero, then it is a label of where to jump. Otherwise, + ** just bypass the row insert logic. See the header comment on the + ** sqlite3WhereOrderByLimitOptLabel() function for additional info. + */ + int iCsr = pSort->iECursor; + sqlite3VdbeAddOp2(v, OP_IfNotZero, iLimit, sqlite3VdbeCurrentAddr(v)+4); + VdbeCoverage(v); + sqlite3VdbeAddOp2(v, OP_Last, iCsr, 0); + iSkip = sqlite3VdbeAddOp4Int(v, OP_IdxLE, + iCsr, 0, regBase+nOBSat, nExpr-nOBSat); + VdbeCoverage(v); + sqlite3VdbeAddOp1(v, OP_Delete, iCsr); + } + if( regRecord==0 ){ + regRecord = makeSorterRecord(pParse, pSort, pSelect, regBase, nBase); + } + if( pSort->sortFlags & SORTFLAG_UseSorter ){ + op = OP_SorterInsert; + }else{ + op = OP_IdxInsert; + } + sqlite3VdbeAddOp4Int(v, op, pSort->iECursor, regRecord, + regBase+nOBSat, nBase-nOBSat); + if( iSkip ){ + sqlite3VdbeChangeP2(v, iSkip, + pSort->labelOBLopt ? pSort->labelOBLopt : sqlite3VdbeCurrentAddr(v)); + } +} + +/* +** Add code to implement the OFFSET +*/ +static void codeOffset( + Vdbe *v, /* Generate code into this VM */ + int iOffset, /* Register holding the offset counter */ + int iContinue /* Jump here to skip the current record */ +){ + if( iOffset>0 ){ + sqlite3VdbeAddOp3(v, OP_IfPos, iOffset, iContinue, 1); VdbeCoverage(v); + VdbeComment((v, "OFFSET")); + } +} + +/* +** Add code that will check to make sure the N registers starting at iMem +** form a distinct entry. iTab is a sorting index that holds previously +** seen combinations of the N values. A new entry is made in iTab +** if the current N values are new. +** +** A jump to addrRepeat is made and the N+1 values are popped from the +** stack if the top N elements are not distinct. +*/ +static void codeDistinct( + Parse *pParse, /* Parsing and code generating context */ + int iTab, /* A sorting index used to test for distinctness */ + int addrRepeat, /* Jump to here if not distinct */ + int N, /* Number of elements */ + int iMem /* First element */ +){ + Vdbe *v; + int r1; + + v = pParse->pVdbe; + r1 = sqlite3GetTempReg(pParse); + sqlite3VdbeAddOp4Int(v, OP_Found, iTab, addrRepeat, iMem, N); VdbeCoverage(v); + sqlite3VdbeAddOp3(v, OP_MakeRecord, iMem, N, r1); + sqlite3VdbeAddOp4Int(v, OP_IdxInsert, iTab, r1, iMem, N); + sqlite3VdbeChangeP5(v, OPFLAG_USESEEKRESULT); + sqlite3ReleaseTempReg(pParse, r1); +} + +#ifdef SQLITE_ENABLE_SORTER_REFERENCES +/* +** This function is called as part of inner-loop generation for a SELECT +** statement with an ORDER BY that is not optimized by an index. It +** determines the expressions, if any, that the sorter-reference +** optimization should be used for. The sorter-reference optimization +** is used for SELECT queries like: +** +** SELECT a, bigblob FROM t1 ORDER BY a LIMIT 10 +** +** If the optimization is used for expression "bigblob", then instead of +** storing values read from that column in the sorter records, the PK of +** the row from table t1 is stored instead. Then, as records are extracted from +** the sorter to return to the user, the required value of bigblob is +** retrieved directly from table t1. If the values are very large, this +** can be more efficient than storing them directly in the sorter records. +** +** The ExprList_item.bSorterRef flag is set for each expression in pEList +** for which the sorter-reference optimization should be enabled. +** Additionally, the pSort->aDefer[] array is populated with entries +** for all cursors required to evaluate all selected expressions. Finally. +** output variable (*ppExtra) is set to an expression list containing +** expressions for all extra PK values that should be stored in the +** sorter records. +*/ +static void selectExprDefer( + Parse *pParse, /* Leave any error here */ + SortCtx *pSort, /* Sorter context */ + ExprList *pEList, /* Expressions destined for sorter */ + ExprList **ppExtra /* Expressions to append to sorter record */ +){ + int i; + int nDefer = 0; + ExprList *pExtra = 0; + for(i=0; inExpr; i++){ + struct ExprList_item *pItem = &pEList->a[i]; + if( pItem->u.x.iOrderByCol==0 ){ + Expr *pExpr = pItem->pExpr; + Table *pTab = pExpr->y.pTab; + if( pExpr->op==TK_COLUMN && pExpr->iColumn>=0 && pTab && !IsVirtual(pTab) + && (pTab->aCol[pExpr->iColumn].colFlags & COLFLAG_SORTERREF) + ){ + int j; + for(j=0; jaDefer[j].iCsr==pExpr->iTable ) break; + } + if( j==nDefer ){ + if( nDefer==ArraySize(pSort->aDefer) ){ + continue; + }else{ + int nKey = 1; + int k; + Index *pPk = 0; + if( !HasRowid(pTab) ){ + pPk = sqlite3PrimaryKeyIndex(pTab); + nKey = pPk->nKeyCol; + } + for(k=0; kiTable = pExpr->iTable; + pNew->y.pTab = pExpr->y.pTab; + pNew->iColumn = pPk ? pPk->aiColumn[k] : -1; + pExtra = sqlite3ExprListAppend(pParse, pExtra, pNew); + } + } + pSort->aDefer[nDefer].pTab = pExpr->y.pTab; + pSort->aDefer[nDefer].iCsr = pExpr->iTable; + pSort->aDefer[nDefer].nKey = nKey; + nDefer++; + } + } + pItem->bSorterRef = 1; + } + } + } + pSort->nDefer = (u8)nDefer; + *ppExtra = pExtra; +} +#endif + +/* +** This routine generates the code for the inside of the inner loop +** of a SELECT. +** +** If srcTab is negative, then the p->pEList expressions +** are evaluated in order to get the data for this row. If srcTab is +** zero or more, then data is pulled from srcTab and p->pEList is used only +** to get the number of columns and the collation sequence for each column. +*/ +static void selectInnerLoop( + Parse *pParse, /* The parser context */ + Select *p, /* The complete select statement being coded */ + int srcTab, /* Pull data from this table if non-negative */ + SortCtx *pSort, /* If not NULL, info on how to process ORDER BY */ + DistinctCtx *pDistinct, /* If not NULL, info on how to process DISTINCT */ + SelectDest *pDest, /* How to dispose of the results */ + int iContinue, /* Jump here to continue with next row */ + int iBreak /* Jump here to break out of the inner loop */ +){ + Vdbe *v = pParse->pVdbe; + int i; + int hasDistinct; /* True if the DISTINCT keyword is present */ + int eDest = pDest->eDest; /* How to dispose of results */ + int iParm = pDest->iSDParm; /* First argument to disposal method */ + int nResultCol; /* Number of result columns */ + int nPrefixReg = 0; /* Number of extra registers before regResult */ + RowLoadInfo sRowLoadInfo; /* Info for deferred row loading */ + + /* Usually, regResult is the first cell in an array of memory cells + ** containing the current result row. In this case regOrig is set to the + ** same value. However, if the results are being sent to the sorter, the + ** values for any expressions that are also part of the sort-key are omitted + ** from this array. In this case regOrig is set to zero. */ + int regResult; /* Start of memory holding current results */ + int regOrig; /* Start of memory holding full result (or 0) */ + + assert( v ); + assert( p->pEList!=0 ); + hasDistinct = pDistinct ? pDistinct->eTnctType : WHERE_DISTINCT_NOOP; + if( pSort && pSort->pOrderBy==0 ) pSort = 0; + if( pSort==0 && !hasDistinct ){ + assert( iContinue!=0 ); + codeOffset(v, p->iOffset, iContinue); + } + + /* Pull the requested columns. + */ + nResultCol = p->pEList->nExpr; + + if( pDest->iSdst==0 ){ + if( pSort ){ + nPrefixReg = pSort->pOrderBy->nExpr; + if( !(pSort->sortFlags & SORTFLAG_UseSorter) ) nPrefixReg++; + pParse->nMem += nPrefixReg; + } + pDest->iSdst = pParse->nMem+1; + pParse->nMem += nResultCol; + }else if( pDest->iSdst+nResultCol > pParse->nMem ){ + /* This is an error condition that can result, for example, when a SELECT + ** on the right-hand side of an INSERT contains more result columns than + ** there are columns in the table on the left. The error will be caught + ** and reported later. But we need to make sure enough memory is allocated + ** to avoid other spurious errors in the meantime. */ + pParse->nMem += nResultCol; + } + pDest->nSdst = nResultCol; + regOrig = regResult = pDest->iSdst; + if( srcTab>=0 ){ + for(i=0; ipEList->a[i].zName)); + } + }else if( eDest!=SRT_Exists ){ +#ifdef SQLITE_ENABLE_SORTER_REFERENCES + ExprList *pExtra = 0; +#endif + /* If the destination is an EXISTS(...) expression, the actual + ** values returned by the SELECT are not required. + */ + u8 ecelFlags; /* "ecel" is an abbreviation of "ExprCodeExprList" */ + ExprList *pEList; + if( eDest==SRT_Mem || eDest==SRT_Output || eDest==SRT_Coroutine ){ + ecelFlags = SQLITE_ECEL_DUP; + }else{ + ecelFlags = 0; + } + if( pSort && hasDistinct==0 && eDest!=SRT_EphemTab && eDest!=SRT_Table ){ + /* For each expression in p->pEList that is a copy of an expression in + ** the ORDER BY clause (pSort->pOrderBy), set the associated + ** iOrderByCol value to one more than the index of the ORDER BY + ** expression within the sort-key that pushOntoSorter() will generate. + ** This allows the p->pEList field to be omitted from the sorted record, + ** saving space and CPU cycles. */ + ecelFlags |= (SQLITE_ECEL_OMITREF|SQLITE_ECEL_REF); + + for(i=pSort->nOBSat; ipOrderBy->nExpr; i++){ + int j; + if( (j = pSort->pOrderBy->a[i].u.x.iOrderByCol)>0 ){ + p->pEList->a[j-1].u.x.iOrderByCol = i+1-pSort->nOBSat; + } + } +#ifdef SQLITE_ENABLE_SORTER_REFERENCES + selectExprDefer(pParse, pSort, p->pEList, &pExtra); + if( pExtra && pParse->db->mallocFailed==0 ){ + /* If there are any extra PK columns to add to the sorter records, + ** allocate extra memory cells and adjust the OpenEphemeral + ** instruction to account for the larger records. This is only + ** required if there are one or more WITHOUT ROWID tables with + ** composite primary keys in the SortCtx.aDefer[] array. */ + VdbeOp *pOp = sqlite3VdbeGetOp(v, pSort->addrSortIndex); + pOp->p2 += (pExtra->nExpr - pSort->nDefer); + pOp->p4.pKeyInfo->nAllField += (pExtra->nExpr - pSort->nDefer); + pParse->nMem += pExtra->nExpr; + } +#endif + + /* Adjust nResultCol to account for columns that are omitted + ** from the sorter by the optimizations in this branch */ + pEList = p->pEList; + for(i=0; inExpr; i++){ + if( pEList->a[i].u.x.iOrderByCol>0 +#ifdef SQLITE_ENABLE_SORTER_REFERENCES + || pEList->a[i].bSorterRef +#endif + ){ + nResultCol--; + regOrig = 0; + } + } + + testcase( regOrig ); + testcase( eDest==SRT_Set ); + testcase( eDest==SRT_Mem ); + testcase( eDest==SRT_Coroutine ); + testcase( eDest==SRT_Output ); + assert( eDest==SRT_Set || eDest==SRT_Mem + || eDest==SRT_Coroutine || eDest==SRT_Output ); + } + sRowLoadInfo.regResult = regResult; + sRowLoadInfo.ecelFlags = ecelFlags; +#ifdef SQLITE_ENABLE_SORTER_REFERENCES + sRowLoadInfo.pExtra = pExtra; + sRowLoadInfo.regExtraResult = regResult + nResultCol; + if( pExtra ) nResultCol += pExtra->nExpr; +#endif + if( p->iLimit + && (ecelFlags & SQLITE_ECEL_OMITREF)!=0 + && nPrefixReg>0 + ){ + assert( pSort!=0 ); + assert( hasDistinct==0 ); + pSort->pDeferredRowLoad = &sRowLoadInfo; + regOrig = 0; + }else{ + innerLoopLoadRow(pParse, p, &sRowLoadInfo); + } + } + + /* If the DISTINCT keyword was present on the SELECT statement + ** and this row has been seen before, then do not make this row + ** part of the result. + */ + if( hasDistinct ){ + switch( pDistinct->eTnctType ){ + case WHERE_DISTINCT_ORDERED: { + VdbeOp *pOp; /* No longer required OpenEphemeral instr. */ + int iJump; /* Jump destination */ + int regPrev; /* Previous row content */ + + /* Allocate space for the previous row */ + regPrev = pParse->nMem+1; + pParse->nMem += nResultCol; + + /* Change the OP_OpenEphemeral coded earlier to an OP_Null + ** sets the MEM_Cleared bit on the first register of the + ** previous value. This will cause the OP_Ne below to always + ** fail on the first iteration of the loop even if the first + ** row is all NULLs. + */ + sqlite3VdbeChangeToNoop(v, pDistinct->addrTnct); + pOp = sqlite3VdbeGetOp(v, pDistinct->addrTnct); + pOp->opcode = OP_Null; + pOp->p1 = 1; + pOp->p2 = regPrev; + + iJump = sqlite3VdbeCurrentAddr(v) + nResultCol; + for(i=0; ipEList->a[i].pExpr); + if( idb->mallocFailed ); + sqlite3VdbeAddOp3(v, OP_Copy, regResult, regPrev, nResultCol-1); + break; + } + + case WHERE_DISTINCT_UNIQUE: { + sqlite3VdbeChangeToNoop(v, pDistinct->addrTnct); + break; + } + + default: { + assert( pDistinct->eTnctType==WHERE_DISTINCT_UNORDERED ); + codeDistinct(pParse, pDistinct->tabTnct, iContinue, nResultCol, + regResult); + break; + } + } + if( pSort==0 ){ + codeOffset(v, p->iOffset, iContinue); + } + } + + switch( eDest ){ + /* In this mode, write each query result to the key of the temporary + ** table iParm. + */ +#ifndef SQLITE_OMIT_COMPOUND_SELECT + case SRT_Union: { + int r1; + r1 = sqlite3GetTempReg(pParse); + sqlite3VdbeAddOp3(v, OP_MakeRecord, regResult, nResultCol, r1); + sqlite3VdbeAddOp4Int(v, OP_IdxInsert, iParm, r1, regResult, nResultCol); + sqlite3ReleaseTempReg(pParse, r1); + break; + } + + /* Construct a record from the query result, but instead of + ** saving that record, use it as a key to delete elements from + ** the temporary table iParm. + */ + case SRT_Except: { + sqlite3VdbeAddOp3(v, OP_IdxDelete, iParm, regResult, nResultCol); + break; + } +#endif /* SQLITE_OMIT_COMPOUND_SELECT */ + + /* Store the result as data using a unique key. + */ + case SRT_Fifo: + case SRT_DistFifo: + case SRT_Table: + case SRT_EphemTab: { + int r1 = sqlite3GetTempRange(pParse, nPrefixReg+1); + testcase( eDest==SRT_Table ); + testcase( eDest==SRT_EphemTab ); + testcase( eDest==SRT_Fifo ); + testcase( eDest==SRT_DistFifo ); + sqlite3VdbeAddOp3(v, OP_MakeRecord, regResult, nResultCol, r1+nPrefixReg); +#ifndef SQLITE_OMIT_CTE + if( eDest==SRT_DistFifo ){ + /* If the destination is DistFifo, then cursor (iParm+1) is open + ** on an ephemeral index. If the current row is already present + ** in the index, do not write it to the output. If not, add the + ** current row to the index and proceed with writing it to the + ** output table as well. */ + int addr = sqlite3VdbeCurrentAddr(v) + 4; + sqlite3VdbeAddOp4Int(v, OP_Found, iParm+1, addr, r1, 0); + VdbeCoverage(v); + sqlite3VdbeAddOp4Int(v, OP_IdxInsert, iParm+1, r1,regResult,nResultCol); + assert( pSort==0 ); + } +#endif + if( pSort ){ + assert( regResult==regOrig ); + pushOntoSorter(pParse, pSort, p, r1+nPrefixReg, regOrig, 1, nPrefixReg); + }else{ + int r2 = sqlite3GetTempReg(pParse); + sqlite3VdbeAddOp2(v, OP_NewRowid, iParm, r2); + sqlite3VdbeAddOp3(v, OP_Insert, iParm, r1, r2); + sqlite3VdbeChangeP5(v, OPFLAG_APPEND); + sqlite3ReleaseTempReg(pParse, r2); + } + sqlite3ReleaseTempRange(pParse, r1, nPrefixReg+1); + break; + } + +#ifndef SQLITE_OMIT_SUBQUERY + /* If we are creating a set for an "expr IN (SELECT ...)" construct, + ** then there should be a single item on the stack. Write this + ** item into the set table with bogus data. + */ + case SRT_Set: { + if( pSort ){ + /* At first glance you would think we could optimize out the + ** ORDER BY in this case since the order of entries in the set + ** does not matter. But there might be a LIMIT clause, in which + ** case the order does matter */ + pushOntoSorter( + pParse, pSort, p, regResult, regOrig, nResultCol, nPrefixReg); + }else{ + int r1 = sqlite3GetTempReg(pParse); + assert( sqlite3Strlen30(pDest->zAffSdst)==nResultCol ); + sqlite3VdbeAddOp4(v, OP_MakeRecord, regResult, nResultCol, + r1, pDest->zAffSdst, nResultCol); + sqlite3VdbeAddOp4Int(v, OP_IdxInsert, iParm, r1, regResult, nResultCol); + sqlite3ReleaseTempReg(pParse, r1); + } + break; + } + + /* If any row exist in the result set, record that fact and abort. + */ + case SRT_Exists: { + sqlite3VdbeAddOp2(v, OP_Integer, 1, iParm); + /* The LIMIT clause will terminate the loop for us */ + break; + } + + /* If this is a scalar select that is part of an expression, then + ** store the results in the appropriate memory cell or array of + ** memory cells and break out of the scan loop. + */ + case SRT_Mem: { + if( pSort ){ + assert( nResultCol<=pDest->nSdst ); + pushOntoSorter( + pParse, pSort, p, regResult, regOrig, nResultCol, nPrefixReg); + }else{ + assert( nResultCol==pDest->nSdst ); + assert( regResult==iParm ); + /* The LIMIT clause will jump out of the loop for us */ + } + break; + } +#endif /* #ifndef SQLITE_OMIT_SUBQUERY */ + + case SRT_Coroutine: /* Send data to a co-routine */ + case SRT_Output: { /* Return the results */ + testcase( eDest==SRT_Coroutine ); + testcase( eDest==SRT_Output ); + if( pSort ){ + pushOntoSorter(pParse, pSort, p, regResult, regOrig, nResultCol, + nPrefixReg); + }else if( eDest==SRT_Coroutine ){ + sqlite3VdbeAddOp1(v, OP_Yield, pDest->iSDParm); + }else{ + sqlite3VdbeAddOp2(v, OP_ResultRow, regResult, nResultCol); + } + break; + } + +#ifndef SQLITE_OMIT_CTE + /* Write the results into a priority queue that is order according to + ** pDest->pOrderBy (in pSO). pDest->iSDParm (in iParm) is the cursor for an + ** index with pSO->nExpr+2 columns. Build a key using pSO for the first + ** pSO->nExpr columns, then make sure all keys are unique by adding a + ** final OP_Sequence column. The last column is the record as a blob. + */ + case SRT_DistQueue: + case SRT_Queue: { + int nKey; + int r1, r2, r3; + int addrTest = 0; + ExprList *pSO; + pSO = pDest->pOrderBy; + assert( pSO ); + nKey = pSO->nExpr; + r1 = sqlite3GetTempReg(pParse); + r2 = sqlite3GetTempRange(pParse, nKey+2); + r3 = r2+nKey+1; + if( eDest==SRT_DistQueue ){ + /* If the destination is DistQueue, then cursor (iParm+1) is open + ** on a second ephemeral index that holds all values every previously + ** added to the queue. */ + addrTest = sqlite3VdbeAddOp4Int(v, OP_Found, iParm+1, 0, + regResult, nResultCol); + VdbeCoverage(v); + } + sqlite3VdbeAddOp3(v, OP_MakeRecord, regResult, nResultCol, r3); + if( eDest==SRT_DistQueue ){ + sqlite3VdbeAddOp2(v, OP_IdxInsert, iParm+1, r3); + sqlite3VdbeChangeP5(v, OPFLAG_USESEEKRESULT); + } + for(i=0; ia[i].u.x.iOrderByCol - 1, + r2+i); + } + sqlite3VdbeAddOp2(v, OP_Sequence, iParm, r2+nKey); + sqlite3VdbeAddOp3(v, OP_MakeRecord, r2, nKey+2, r1); + sqlite3VdbeAddOp4Int(v, OP_IdxInsert, iParm, r1, r2, nKey+2); + if( addrTest ) sqlite3VdbeJumpHere(v, addrTest); + sqlite3ReleaseTempReg(pParse, r1); + sqlite3ReleaseTempRange(pParse, r2, nKey+2); + break; + } +#endif /* SQLITE_OMIT_CTE */ + + + +#if !defined(SQLITE_OMIT_TRIGGER) + /* Discard the results. This is used for SELECT statements inside + ** the body of a TRIGGER. The purpose of such selects is to call + ** user-defined functions that have side effects. We do not care + ** about the actual results of the select. + */ + default: { + assert( eDest==SRT_Discard ); + break; + } +#endif + } + + /* Jump to the end of the loop if the LIMIT is reached. Except, if + ** there is a sorter, in which case the sorter has already limited + ** the output for us. + */ + if( pSort==0 && p->iLimit ){ + sqlite3VdbeAddOp2(v, OP_DecrJumpZero, p->iLimit, iBreak); VdbeCoverage(v); + } +} + +/* +** Allocate a KeyInfo object sufficient for an index of N key columns and +** X extra columns. +*/ +SQLITE_PRIVATE KeyInfo *sqlite3KeyInfoAlloc(sqlite3 *db, int N, int X){ + int nExtra = (N+X)*(sizeof(CollSeq*)+1) - sizeof(CollSeq*); + KeyInfo *p = sqlite3DbMallocRawNN(db, sizeof(KeyInfo) + nExtra); + if( p ){ + p->aSortFlags = (u8*)&p->aColl[N+X]; + p->nKeyField = (u16)N; + p->nAllField = (u16)(N+X); + p->enc = ENC(db); + p->db = db; + p->nRef = 1; + memset(&p[1], 0, nExtra); + }else{ + sqlite3OomFault(db); + } + return p; +} + +/* +** Deallocate a KeyInfo object +*/ +SQLITE_PRIVATE void sqlite3KeyInfoUnref(KeyInfo *p){ + if( p ){ + assert( p->nRef>0 ); + p->nRef--; + if( p->nRef==0 ) sqlite3DbFreeNN(p->db, p); + } +} + +/* +** Make a new pointer to a KeyInfo object +*/ +SQLITE_PRIVATE KeyInfo *sqlite3KeyInfoRef(KeyInfo *p){ + if( p ){ + assert( p->nRef>0 ); + p->nRef++; + } + return p; +} + +#ifdef SQLITE_DEBUG +/* +** Return TRUE if a KeyInfo object can be change. The KeyInfo object +** can only be changed if this is just a single reference to the object. +** +** This routine is used only inside of assert() statements. +*/ +SQLITE_PRIVATE int sqlite3KeyInfoIsWriteable(KeyInfo *p){ return p->nRef==1; } +#endif /* SQLITE_DEBUG */ + +/* +** Given an expression list, generate a KeyInfo structure that records +** the collating sequence for each expression in that expression list. +** +** If the ExprList is an ORDER BY or GROUP BY clause then the resulting +** KeyInfo structure is appropriate for initializing a virtual index to +** implement that clause. If the ExprList is the result set of a SELECT +** then the KeyInfo structure is appropriate for initializing a virtual +** index to implement a DISTINCT test. +** +** Space to hold the KeyInfo structure is obtained from malloc. The calling +** function is responsible for seeing that this structure is eventually +** freed. +*/ +SQLITE_PRIVATE KeyInfo *sqlite3KeyInfoFromExprList( + Parse *pParse, /* Parsing context */ + ExprList *pList, /* Form the KeyInfo object from this ExprList */ + int iStart, /* Begin with this column of pList */ + int nExtra /* Add this many extra columns to the end */ +){ + int nExpr; + KeyInfo *pInfo; + struct ExprList_item *pItem; + sqlite3 *db = pParse->db; + int i; + + nExpr = pList->nExpr; + pInfo = sqlite3KeyInfoAlloc(db, nExpr-iStart, nExtra+1); + if( pInfo ){ + assert( sqlite3KeyInfoIsWriteable(pInfo) ); + for(i=iStart, pItem=pList->a+iStart; iaColl[i-iStart] = sqlite3ExprNNCollSeq(pParse, pItem->pExpr); + pInfo->aSortFlags[i-iStart] = pItem->sortFlags; + } + } + return pInfo; +} + +/* +** Name of the connection operator, used for error messages. +*/ +static const char *selectOpName(int id){ + char *z; + switch( id ){ + case TK_ALL: z = "UNION ALL"; break; + case TK_INTERSECT: z = "INTERSECT"; break; + case TK_EXCEPT: z = "EXCEPT"; break; + default: z = "UNION"; break; + } + return z; +} + +#ifndef SQLITE_OMIT_EXPLAIN +/* +** Unless an "EXPLAIN QUERY PLAN" command is being processed, this function +** is a no-op. Otherwise, it adds a single row of output to the EQP result, +** where the caption is of the form: +** +** "USE TEMP B-TREE FOR xxx" +** +** where xxx is one of "DISTINCT", "ORDER BY" or "GROUP BY". Exactly which +** is determined by the zUsage argument. +*/ +static void explainTempTable(Parse *pParse, const char *zUsage){ + ExplainQueryPlan((pParse, 0, "USE TEMP B-TREE FOR %s", zUsage)); +} + +/* +** Assign expression b to lvalue a. A second, no-op, version of this macro +** is provided when SQLITE_OMIT_EXPLAIN is defined. This allows the code +** in sqlite3Select() to assign values to structure member variables that +** only exist if SQLITE_OMIT_EXPLAIN is not defined without polluting the +** code with #ifndef directives. +*/ +# define explainSetInteger(a, b) a = b + +#else +/* No-op versions of the explainXXX() functions and macros. */ +# define explainTempTable(y,z) +# define explainSetInteger(y,z) +#endif + + +/* +** If the inner loop was generated using a non-null pOrderBy argument, +** then the results were placed in a sorter. After the loop is terminated +** we need to run the sorter and output the results. The following +** routine generates the code needed to do that. +*/ +static void generateSortTail( + Parse *pParse, /* Parsing context */ + Select *p, /* The SELECT statement */ + SortCtx *pSort, /* Information on the ORDER BY clause */ + int nColumn, /* Number of columns of data */ + SelectDest *pDest /* Write the sorted results here */ +){ + Vdbe *v = pParse->pVdbe; /* The prepared statement */ + int addrBreak = pSort->labelDone; /* Jump here to exit loop */ + int addrContinue = sqlite3VdbeMakeLabel(pParse);/* Jump here for next cycle */ + int addr; /* Top of output loop. Jump for Next. */ + int addrOnce = 0; + int iTab; + ExprList *pOrderBy = pSort->pOrderBy; + int eDest = pDest->eDest; + int iParm = pDest->iSDParm; + int regRow; + int regRowid; + int iCol; + int nKey; /* Number of key columns in sorter record */ + int iSortTab; /* Sorter cursor to read from */ + int i; + int bSeq; /* True if sorter record includes seq. no. */ + int nRefKey = 0; + struct ExprList_item *aOutEx = p->pEList->a; + + assert( addrBreak<0 ); + if( pSort->labelBkOut ){ + sqlite3VdbeAddOp2(v, OP_Gosub, pSort->regReturn, pSort->labelBkOut); + sqlite3VdbeGoto(v, addrBreak); + sqlite3VdbeResolveLabel(v, pSort->labelBkOut); + } + +#ifdef SQLITE_ENABLE_SORTER_REFERENCES + /* Open any cursors needed for sorter-reference expressions */ + for(i=0; inDefer; i++){ + Table *pTab = pSort->aDefer[i].pTab; + int iDb = sqlite3SchemaToIndex(pParse->db, pTab->pSchema); + sqlite3OpenTable(pParse, pSort->aDefer[i].iCsr, iDb, pTab, OP_OpenRead); + nRefKey = MAX(nRefKey, pSort->aDefer[i].nKey); + } +#endif + + iTab = pSort->iECursor; + if( eDest==SRT_Output || eDest==SRT_Coroutine || eDest==SRT_Mem ){ + regRowid = 0; + regRow = pDest->iSdst; + }else{ + regRowid = sqlite3GetTempReg(pParse); + if( eDest==SRT_EphemTab || eDest==SRT_Table ){ + regRow = sqlite3GetTempReg(pParse); + nColumn = 0; + }else{ + regRow = sqlite3GetTempRange(pParse, nColumn); + } + } + nKey = pOrderBy->nExpr - pSort->nOBSat; + if( pSort->sortFlags & SORTFLAG_UseSorter ){ + int regSortOut = ++pParse->nMem; + iSortTab = pParse->nTab++; + if( pSort->labelBkOut ){ + addrOnce = sqlite3VdbeAddOp0(v, OP_Once); VdbeCoverage(v); + } + sqlite3VdbeAddOp3(v, OP_OpenPseudo, iSortTab, regSortOut, + nKey+1+nColumn+nRefKey); + if( addrOnce ) sqlite3VdbeJumpHere(v, addrOnce); + addr = 1 + sqlite3VdbeAddOp2(v, OP_SorterSort, iTab, addrBreak); + VdbeCoverage(v); + codeOffset(v, p->iOffset, addrContinue); + sqlite3VdbeAddOp3(v, OP_SorterData, iTab, regSortOut, iSortTab); + bSeq = 0; + }else{ + addr = 1 + sqlite3VdbeAddOp2(v, OP_Sort, iTab, addrBreak); VdbeCoverage(v); + codeOffset(v, p->iOffset, addrContinue); + iSortTab = iTab; + bSeq = 1; + } + for(i=0, iCol=nKey+bSeq-1; inDefer ){ + int iKey = iCol+1; + int regKey = sqlite3GetTempRange(pParse, nRefKey); + + for(i=0; inDefer; i++){ + int iCsr = pSort->aDefer[i].iCsr; + Table *pTab = pSort->aDefer[i].pTab; + int nKey = pSort->aDefer[i].nKey; + + sqlite3VdbeAddOp1(v, OP_NullRow, iCsr); + if( HasRowid(pTab) ){ + sqlite3VdbeAddOp3(v, OP_Column, iSortTab, iKey++, regKey); + sqlite3VdbeAddOp3(v, OP_SeekRowid, iCsr, + sqlite3VdbeCurrentAddr(v)+1, regKey); + }else{ + int k; + int iJmp; + assert( sqlite3PrimaryKeyIndex(pTab)->nKeyCol==nKey ); + for(k=0; k=0; i--){ +#ifdef SQLITE_ENABLE_SORTER_REFERENCES + if( aOutEx[i].bSorterRef ){ + sqlite3ExprCode(pParse, aOutEx[i].pExpr, regRow+i); + }else +#endif + { + int iRead; + if( aOutEx[i].u.x.iOrderByCol ){ + iRead = aOutEx[i].u.x.iOrderByCol-1; + }else{ + iRead = iCol--; + } + sqlite3VdbeAddOp3(v, OP_Column, iSortTab, iRead, regRow+i); + VdbeComment((v, "%s", aOutEx[i].zName?aOutEx[i].zName : aOutEx[i].zSpan)); + } + } + switch( eDest ){ + case SRT_Table: + case SRT_EphemTab: { + sqlite3VdbeAddOp3(v, OP_Column, iSortTab, nKey+bSeq, regRow); + sqlite3VdbeAddOp2(v, OP_NewRowid, iParm, regRowid); + sqlite3VdbeAddOp3(v, OP_Insert, iParm, regRow, regRowid); + sqlite3VdbeChangeP5(v, OPFLAG_APPEND); + break; + } +#ifndef SQLITE_OMIT_SUBQUERY + case SRT_Set: { + assert( nColumn==sqlite3Strlen30(pDest->zAffSdst) ); + sqlite3VdbeAddOp4(v, OP_MakeRecord, regRow, nColumn, regRowid, + pDest->zAffSdst, nColumn); + sqlite3VdbeAddOp4Int(v, OP_IdxInsert, iParm, regRowid, regRow, nColumn); + break; + } + case SRT_Mem: { + /* The LIMIT clause will terminate the loop for us */ + break; + } +#endif + default: { + assert( eDest==SRT_Output || eDest==SRT_Coroutine ); + testcase( eDest==SRT_Output ); + testcase( eDest==SRT_Coroutine ); + if( eDest==SRT_Output ){ + sqlite3VdbeAddOp2(v, OP_ResultRow, pDest->iSdst, nColumn); + }else{ + sqlite3VdbeAddOp1(v, OP_Yield, pDest->iSDParm); + } + break; + } + } + if( regRowid ){ + if( eDest==SRT_Set ){ + sqlite3ReleaseTempRange(pParse, regRow, nColumn); + }else{ + sqlite3ReleaseTempReg(pParse, regRow); + } + sqlite3ReleaseTempReg(pParse, regRowid); + } + /* The bottom of the loop + */ + sqlite3VdbeResolveLabel(v, addrContinue); + if( pSort->sortFlags & SORTFLAG_UseSorter ){ + sqlite3VdbeAddOp2(v, OP_SorterNext, iTab, addr); VdbeCoverage(v); + }else{ + sqlite3VdbeAddOp2(v, OP_Next, iTab, addr); VdbeCoverage(v); + } + if( pSort->regReturn ) sqlite3VdbeAddOp1(v, OP_Return, pSort->regReturn); + sqlite3VdbeResolveLabel(v, addrBreak); +} + +/* +** Return a pointer to a string containing the 'declaration type' of the +** expression pExpr. The string may be treated as static by the caller. +** +** Also try to estimate the size of the returned value and return that +** result in *pEstWidth. +** +** The declaration type is the exact datatype definition extracted from the +** original CREATE TABLE statement if the expression is a column. The +** declaration type for a ROWID field is INTEGER. Exactly when an expression +** is considered a column can be complex in the presence of subqueries. The +** result-set expression in all of the following SELECT statements is +** considered a column by this function. +** +** SELECT col FROM tbl; +** SELECT (SELECT col FROM tbl; +** SELECT (SELECT col FROM tbl); +** SELECT abc FROM (SELECT col AS abc FROM tbl); +** +** The declaration type for any expression other than a column is NULL. +** +** This routine has either 3 or 6 parameters depending on whether or not +** the SQLITE_ENABLE_COLUMN_METADATA compile-time option is used. +*/ +#ifdef SQLITE_ENABLE_COLUMN_METADATA +# define columnType(A,B,C,D,E) columnTypeImpl(A,B,C,D,E) +#else /* if !defined(SQLITE_ENABLE_COLUMN_METADATA) */ +# define columnType(A,B,C,D,E) columnTypeImpl(A,B) +#endif +static const char *columnTypeImpl( + NameContext *pNC, +#ifndef SQLITE_ENABLE_COLUMN_METADATA + Expr *pExpr +#else + Expr *pExpr, + const char **pzOrigDb, + const char **pzOrigTab, + const char **pzOrigCol +#endif +){ + char const *zType = 0; + int j; +#ifdef SQLITE_ENABLE_COLUMN_METADATA + char const *zOrigDb = 0; + char const *zOrigTab = 0; + char const *zOrigCol = 0; +#endif + + assert( pExpr!=0 ); + assert( pNC->pSrcList!=0 ); + switch( pExpr->op ){ + case TK_COLUMN: { + /* The expression is a column. Locate the table the column is being + ** extracted from in NameContext.pSrcList. This table may be real + ** database table or a subquery. + */ + Table *pTab = 0; /* Table structure column is extracted from */ + Select *pS = 0; /* Select the column is extracted from */ + int iCol = pExpr->iColumn; /* Index of column in pTab */ + while( pNC && !pTab ){ + SrcList *pTabList = pNC->pSrcList; + for(j=0;jnSrc && pTabList->a[j].iCursor!=pExpr->iTable;j++); + if( jnSrc ){ + pTab = pTabList->a[j].pTab; + pS = pTabList->a[j].pSelect; + }else{ + pNC = pNC->pNext; + } + } + + if( pTab==0 ){ + /* At one time, code such as "SELECT new.x" within a trigger would + ** cause this condition to run. Since then, we have restructured how + ** trigger code is generated and so this condition is no longer + ** possible. However, it can still be true for statements like + ** the following: + ** + ** CREATE TABLE t1(col INTEGER); + ** SELECT (SELECT t1.col) FROM FROM t1; + ** + ** when columnType() is called on the expression "t1.col" in the + ** sub-select. In this case, set the column type to NULL, even + ** though it should really be "INTEGER". + ** + ** This is not a problem, as the column type of "t1.col" is never + ** used. When columnType() is called on the expression + ** "(SELECT t1.col)", the correct type is returned (see the TK_SELECT + ** branch below. */ + break; + } + + assert( pTab && pExpr->y.pTab==pTab ); + if( pS ){ + /* The "table" is actually a sub-select or a view in the FROM clause + ** of the SELECT statement. Return the declaration type and origin + ** data for the result-set column of the sub-select. + */ + if( iCol>=0 && iColpEList->nExpr ){ + /* If iCol is less than zero, then the expression requests the + ** rowid of the sub-select or view. This expression is legal (see + ** test case misc2.2.2) - it always evaluates to NULL. + */ + NameContext sNC; + Expr *p = pS->pEList->a[iCol].pExpr; + sNC.pSrcList = pS->pSrc; + sNC.pNext = pNC; + sNC.pParse = pNC->pParse; + zType = columnType(&sNC, p,&zOrigDb,&zOrigTab,&zOrigCol); + } + }else{ + /* A real table or a CTE table */ + assert( !pS ); +#ifdef SQLITE_ENABLE_COLUMN_METADATA + if( iCol<0 ) iCol = pTab->iPKey; + assert( iCol==XN_ROWID || (iCol>=0 && iColnCol) ); + if( iCol<0 ){ + zType = "INTEGER"; + zOrigCol = "rowid"; + }else{ + zOrigCol = pTab->aCol[iCol].zName; + zType = sqlite3ColumnType(&pTab->aCol[iCol],0); + } + zOrigTab = pTab->zName; + if( pNC->pParse && pTab->pSchema ){ + int iDb = sqlite3SchemaToIndex(pNC->pParse->db, pTab->pSchema); + zOrigDb = pNC->pParse->db->aDb[iDb].zDbSName; + } +#else + assert( iCol==XN_ROWID || (iCol>=0 && iColnCol) ); + if( iCol<0 ){ + zType = "INTEGER"; + }else{ + zType = sqlite3ColumnType(&pTab->aCol[iCol],0); + } +#endif + } + break; + } +#ifndef SQLITE_OMIT_SUBQUERY + case TK_SELECT: { + /* The expression is a sub-select. Return the declaration type and + ** origin info for the single column in the result set of the SELECT + ** statement. + */ + NameContext sNC; + Select *pS = pExpr->x.pSelect; + Expr *p = pS->pEList->a[0].pExpr; + assert( ExprHasProperty(pExpr, EP_xIsSelect) ); + sNC.pSrcList = pS->pSrc; + sNC.pNext = pNC; + sNC.pParse = pNC->pParse; + zType = columnType(&sNC, p, &zOrigDb, &zOrigTab, &zOrigCol); + break; + } +#endif + } + +#ifdef SQLITE_ENABLE_COLUMN_METADATA + if( pzOrigDb ){ + assert( pzOrigTab && pzOrigCol ); + *pzOrigDb = zOrigDb; + *pzOrigTab = zOrigTab; + *pzOrigCol = zOrigCol; + } +#endif + return zType; +} + +/* +** Generate code that will tell the VDBE the declaration types of columns +** in the result set. +*/ +static void generateColumnTypes( + Parse *pParse, /* Parser context */ + SrcList *pTabList, /* List of tables */ + ExprList *pEList /* Expressions defining the result set */ +){ +#ifndef SQLITE_OMIT_DECLTYPE + Vdbe *v = pParse->pVdbe; + int i; + NameContext sNC; + sNC.pSrcList = pTabList; + sNC.pParse = pParse; + sNC.pNext = 0; + for(i=0; inExpr; i++){ + Expr *p = pEList->a[i].pExpr; + const char *zType; +#ifdef SQLITE_ENABLE_COLUMN_METADATA + const char *zOrigDb = 0; + const char *zOrigTab = 0; + const char *zOrigCol = 0; + zType = columnType(&sNC, p, &zOrigDb, &zOrigTab, &zOrigCol); + + /* The vdbe must make its own copy of the column-type and other + ** column specific strings, in case the schema is reset before this + ** virtual machine is deleted. + */ + sqlite3VdbeSetColName(v, i, COLNAME_DATABASE, zOrigDb, SQLITE_TRANSIENT); + sqlite3VdbeSetColName(v, i, COLNAME_TABLE, zOrigTab, SQLITE_TRANSIENT); + sqlite3VdbeSetColName(v, i, COLNAME_COLUMN, zOrigCol, SQLITE_TRANSIENT); +#else + zType = columnType(&sNC, p, 0, 0, 0); +#endif + sqlite3VdbeSetColName(v, i, COLNAME_DECLTYPE, zType, SQLITE_TRANSIENT); + } +#endif /* !defined(SQLITE_OMIT_DECLTYPE) */ +} + + +/* +** Compute the column names for a SELECT statement. +** +** The only guarantee that SQLite makes about column names is that if the +** column has an AS clause assigning it a name, that will be the name used. +** That is the only documented guarantee. However, countless applications +** developed over the years have made baseless assumptions about column names +** and will break if those assumptions changes. Hence, use extreme caution +** when modifying this routine to avoid breaking legacy. +** +** See Also: sqlite3ColumnsFromExprList() +** +** The PRAGMA short_column_names and PRAGMA full_column_names settings are +** deprecated. The default setting is short=ON, full=OFF. 99.9% of all +** applications should operate this way. Nevertheless, we need to support the +** other modes for legacy: +** +** short=OFF, full=OFF: Column name is the text of the expression has it +** originally appears in the SELECT statement. In +** other words, the zSpan of the result expression. +** +** short=ON, full=OFF: (This is the default setting). If the result +** refers directly to a table column, then the +** result column name is just the table column +** name: COLUMN. Otherwise use zSpan. +** +** full=ON, short=ANY: If the result refers directly to a table column, +** then the result column name with the table name +** prefix, ex: TABLE.COLUMN. Otherwise use zSpan. +*/ +static void generateColumnNames( + Parse *pParse, /* Parser context */ + Select *pSelect /* Generate column names for this SELECT statement */ +){ + Vdbe *v = pParse->pVdbe; + int i; + Table *pTab; + SrcList *pTabList; + ExprList *pEList; + sqlite3 *db = pParse->db; + int fullName; /* TABLE.COLUMN if no AS clause and is a direct table ref */ + int srcName; /* COLUMN or TABLE.COLUMN if no AS clause and is direct */ + +#ifndef SQLITE_OMIT_EXPLAIN + /* If this is an EXPLAIN, skip this step */ + if( pParse->explain ){ + return; + } +#endif + + if( pParse->colNamesSet ) return; + /* Column names are determined by the left-most term of a compound select */ + while( pSelect->pPrior ) pSelect = pSelect->pPrior; + SELECTTRACE(1,pParse,pSelect,("generating column names\n")); + pTabList = pSelect->pSrc; + pEList = pSelect->pEList; + assert( v!=0 ); + assert( pTabList!=0 ); + pParse->colNamesSet = 1; + fullName = (db->flags & SQLITE_FullColNames)!=0; + srcName = (db->flags & SQLITE_ShortColNames)!=0 || fullName; + sqlite3VdbeSetNumCols(v, pEList->nExpr); + for(i=0; inExpr; i++){ + Expr *p = pEList->a[i].pExpr; + + assert( p!=0 ); + assert( p->op!=TK_AGG_COLUMN ); /* Agg processing has not run yet */ + assert( p->op!=TK_COLUMN || p->y.pTab!=0 ); /* Covering idx not yet coded */ + if( pEList->a[i].zName ){ + /* An AS clause always takes first priority */ + char *zName = pEList->a[i].zName; + sqlite3VdbeSetColName(v, i, COLNAME_NAME, zName, SQLITE_TRANSIENT); + }else if( srcName && p->op==TK_COLUMN ){ + char *zCol; + int iCol = p->iColumn; + pTab = p->y.pTab; + assert( pTab!=0 ); + if( iCol<0 ) iCol = pTab->iPKey; + assert( iCol==-1 || (iCol>=0 && iColnCol) ); + if( iCol<0 ){ + zCol = "rowid"; + }else{ + zCol = pTab->aCol[iCol].zName; + } + if( fullName ){ + char *zName = 0; + zName = sqlite3MPrintf(db, "%s.%s", pTab->zName, zCol); + sqlite3VdbeSetColName(v, i, COLNAME_NAME, zName, SQLITE_DYNAMIC); + }else{ + sqlite3VdbeSetColName(v, i, COLNAME_NAME, zCol, SQLITE_TRANSIENT); + } + }else{ + const char *z = pEList->a[i].zSpan; + z = z==0 ? sqlite3MPrintf(db, "column%d", i+1) : sqlite3DbStrDup(db, z); + sqlite3VdbeSetColName(v, i, COLNAME_NAME, z, SQLITE_DYNAMIC); + } + } + generateColumnTypes(pParse, pTabList, pEList); +} + +/* +** Given an expression list (which is really the list of expressions +** that form the result set of a SELECT statement) compute appropriate +** column names for a table that would hold the expression list. +** +** All column names will be unique. +** +** Only the column names are computed. Column.zType, Column.zColl, +** and other fields of Column are zeroed. +** +** Return SQLITE_OK on success. If a memory allocation error occurs, +** store NULL in *paCol and 0 in *pnCol and return SQLITE_NOMEM. +** +** The only guarantee that SQLite makes about column names is that if the +** column has an AS clause assigning it a name, that will be the name used. +** That is the only documented guarantee. However, countless applications +** developed over the years have made baseless assumptions about column names +** and will break if those assumptions changes. Hence, use extreme caution +** when modifying this routine to avoid breaking legacy. +** +** See Also: generateColumnNames() +*/ +SQLITE_PRIVATE int sqlite3ColumnsFromExprList( + Parse *pParse, /* Parsing context */ + ExprList *pEList, /* Expr list from which to derive column names */ + i16 *pnCol, /* Write the number of columns here */ + Column **paCol /* Write the new column list here */ +){ + sqlite3 *db = pParse->db; /* Database connection */ + int i, j; /* Loop counters */ + u32 cnt; /* Index added to make the name unique */ + Column *aCol, *pCol; /* For looping over result columns */ + int nCol; /* Number of columns in the result set */ + char *zName; /* Column name */ + int nName; /* Size of name in zName[] */ + Hash ht; /* Hash table of column names */ + + sqlite3HashInit(&ht); + if( pEList ){ + nCol = pEList->nExpr; + aCol = sqlite3DbMallocZero(db, sizeof(aCol[0])*nCol); + testcase( aCol==0 ); + if( nCol>32767 ) nCol = 32767; + }else{ + nCol = 0; + aCol = 0; + } + assert( nCol==(i16)nCol ); + *pnCol = nCol; + *paCol = aCol; + + for(i=0, pCol=aCol; imallocFailed; i++, pCol++){ + /* Get an appropriate name for the column + */ + if( (zName = pEList->a[i].zName)!=0 ){ + /* If the column contains an "AS " phrase, use as the name */ + }else{ + Expr *pColExpr = sqlite3ExprSkipCollateAndLikely(pEList->a[i].pExpr); + while( pColExpr->op==TK_DOT ){ + pColExpr = pColExpr->pRight; + assert( pColExpr!=0 ); + } + if( pColExpr->op==TK_COLUMN ){ + /* For columns use the column name name */ + int iCol = pColExpr->iColumn; + Table *pTab = pColExpr->y.pTab; + assert( pTab!=0 ); + if( iCol<0 ) iCol = pTab->iPKey; + zName = iCol>=0 ? pTab->aCol[iCol].zName : "rowid"; + }else if( pColExpr->op==TK_ID ){ + assert( !ExprHasProperty(pColExpr, EP_IntValue) ); + zName = pColExpr->u.zToken; + }else{ + /* Use the original text of the column expression as its name */ + zName = pEList->a[i].zSpan; + } + } + if( zName ){ + zName = sqlite3DbStrDup(db, zName); + }else{ + zName = sqlite3MPrintf(db,"column%d",i+1); + } + + /* Make sure the column name is unique. If the name is not unique, + ** append an integer to the name so that it becomes unique. + */ + cnt = 0; + while( zName && sqlite3HashFind(&ht, zName)!=0 ){ + nName = sqlite3Strlen30(zName); + if( nName>0 ){ + for(j=nName-1; j>0 && sqlite3Isdigit(zName[j]); j--){} + if( zName[j]==':' ) nName = j; + } + zName = sqlite3MPrintf(db, "%.*z:%u", nName, zName, ++cnt); + if( cnt>3 ) sqlite3_randomness(sizeof(cnt), &cnt); + } + pCol->zName = zName; + sqlite3ColumnPropertiesFromName(0, pCol); + if( zName && sqlite3HashInsert(&ht, zName, pCol)==pCol ){ + sqlite3OomFault(db); + } + } + sqlite3HashClear(&ht); + if( db->mallocFailed ){ + for(j=0; jdb; + NameContext sNC; + Column *pCol; + CollSeq *pColl; + int i; + Expr *p; + struct ExprList_item *a; + + assert( pSelect!=0 ); + assert( (pSelect->selFlags & SF_Resolved)!=0 ); + assert( pTab->nCol==pSelect->pEList->nExpr || db->mallocFailed ); + if( db->mallocFailed ) return; + memset(&sNC, 0, sizeof(sNC)); + sNC.pSrcList = pSelect->pSrc; + a = pSelect->pEList->a; + for(i=0, pCol=pTab->aCol; inCol; i++, pCol++){ + const char *zType; + int n, m; + p = a[i].pExpr; + zType = columnType(&sNC, p, 0, 0, 0); + /* pCol->szEst = ... // Column size est for SELECT tables never used */ + pCol->affinity = sqlite3ExprAffinity(p); + if( zType ){ + m = sqlite3Strlen30(zType); + n = sqlite3Strlen30(pCol->zName); + pCol->zName = sqlite3DbReallocOrFree(db, pCol->zName, n+m+2); + if( pCol->zName ){ + memcpy(&pCol->zName[n+1], zType, m+1); + pCol->colFlags |= COLFLAG_HASTYPE; + } + } + if( pCol->affinity<=SQLITE_AFF_NONE ) pCol->affinity = aff; + pColl = sqlite3ExprCollSeq(pParse, p); + if( pColl && pCol->zColl==0 ){ + pCol->zColl = sqlite3DbStrDup(db, pColl->zName); + } + } + pTab->szTabRow = 1; /* Any non-zero value works */ +} + +/* +** Given a SELECT statement, generate a Table structure that describes +** the result set of that SELECT. +*/ +SQLITE_PRIVATE Table *sqlite3ResultSetOfSelect(Parse *pParse, Select *pSelect, char aff){ + Table *pTab; + sqlite3 *db = pParse->db; + u64 savedFlags; + + savedFlags = db->flags; + db->flags &= ~(u64)SQLITE_FullColNames; + db->flags |= SQLITE_ShortColNames; + sqlite3SelectPrep(pParse, pSelect, 0); + db->flags = savedFlags; + if( pParse->nErr ) return 0; + while( pSelect->pPrior ) pSelect = pSelect->pPrior; + pTab = sqlite3DbMallocZero(db, sizeof(Table) ); + if( pTab==0 ){ + return 0; + } + pTab->nTabRef = 1; + pTab->zName = 0; + pTab->nRowLogEst = 200; assert( 200==sqlite3LogEst(1048576) ); + sqlite3ColumnsFromExprList(pParse, pSelect->pEList, &pTab->nCol, &pTab->aCol); + sqlite3SelectAddColumnTypeAndCollation(pParse, pTab, pSelect, aff); + pTab->iPKey = -1; + if( db->mallocFailed ){ + sqlite3DeleteTable(db, pTab); + return 0; + } + return pTab; +} + +/* +** Get a VDBE for the given parser context. Create a new one if necessary. +** If an error occurs, return NULL and leave a message in pParse. +*/ +SQLITE_PRIVATE Vdbe *sqlite3GetVdbe(Parse *pParse){ + if( pParse->pVdbe ){ + return pParse->pVdbe; + } + if( pParse->pToplevel==0 + && OptimizationEnabled(pParse->db,SQLITE_FactorOutConst) + ){ + pParse->okConstFactor = 1; + } + return sqlite3VdbeCreate(pParse); +} + + +/* +** Compute the iLimit and iOffset fields of the SELECT based on the +** pLimit expressions. pLimit->pLeft and pLimit->pRight hold the expressions +** that appear in the original SQL statement after the LIMIT and OFFSET +** keywords. Or NULL if those keywords are omitted. iLimit and iOffset +** are the integer memory register numbers for counters used to compute +** the limit and offset. If there is no limit and/or offset, then +** iLimit and iOffset are negative. +** +** This routine changes the values of iLimit and iOffset only if +** a limit or offset is defined by pLimit->pLeft and pLimit->pRight. iLimit +** and iOffset should have been preset to appropriate default values (zero) +** prior to calling this routine. +** +** The iOffset register (if it exists) is initialized to the value +** of the OFFSET. The iLimit register is initialized to LIMIT. Register +** iOffset+1 is initialized to LIMIT+OFFSET. +** +** Only if pLimit->pLeft!=0 do the limit registers get +** redefined. The UNION ALL operator uses this property to force +** the reuse of the same limit and offset registers across multiple +** SELECT statements. +*/ +static void computeLimitRegisters(Parse *pParse, Select *p, int iBreak){ + Vdbe *v = 0; + int iLimit = 0; + int iOffset; + int n; + Expr *pLimit = p->pLimit; + + if( p->iLimit ) return; + + /* + ** "LIMIT -1" always shows all rows. There is some + ** controversy about what the correct behavior should be. + ** The current implementation interprets "LIMIT 0" to mean + ** no rows. + */ + if( pLimit ){ + assert( pLimit->op==TK_LIMIT ); + assert( pLimit->pLeft!=0 ); + p->iLimit = iLimit = ++pParse->nMem; + v = sqlite3GetVdbe(pParse); + assert( v!=0 ); + if( sqlite3ExprIsInteger(pLimit->pLeft, &n) ){ + sqlite3VdbeAddOp2(v, OP_Integer, n, iLimit); + VdbeComment((v, "LIMIT counter")); + if( n==0 ){ + sqlite3VdbeGoto(v, iBreak); + }else if( n>=0 && p->nSelectRow>sqlite3LogEst((u64)n) ){ + p->nSelectRow = sqlite3LogEst((u64)n); + p->selFlags |= SF_FixedLimit; + } + }else{ + sqlite3ExprCode(pParse, pLimit->pLeft, iLimit); + sqlite3VdbeAddOp1(v, OP_MustBeInt, iLimit); VdbeCoverage(v); + VdbeComment((v, "LIMIT counter")); + sqlite3VdbeAddOp2(v, OP_IfNot, iLimit, iBreak); VdbeCoverage(v); + } + if( pLimit->pRight ){ + p->iOffset = iOffset = ++pParse->nMem; + pParse->nMem++; /* Allocate an extra register for limit+offset */ + sqlite3ExprCode(pParse, pLimit->pRight, iOffset); + sqlite3VdbeAddOp1(v, OP_MustBeInt, iOffset); VdbeCoverage(v); + VdbeComment((v, "OFFSET counter")); + sqlite3VdbeAddOp3(v, OP_OffsetLimit, iLimit, iOffset+1, iOffset); + VdbeComment((v, "LIMIT+OFFSET")); + } + } +} + +#ifndef SQLITE_OMIT_COMPOUND_SELECT +/* +** Return the appropriate collating sequence for the iCol-th column of +** the result set for the compound-select statement "p". Return NULL if +** the column has no default collating sequence. +** +** The collating sequence for the compound select is taken from the +** left-most term of the select that has a collating sequence. +*/ +static CollSeq *multiSelectCollSeq(Parse *pParse, Select *p, int iCol){ + CollSeq *pRet; + if( p->pPrior ){ + pRet = multiSelectCollSeq(pParse, p->pPrior, iCol); + }else{ + pRet = 0; + } + assert( iCol>=0 ); + /* iCol must be less than p->pEList->nExpr. Otherwise an error would + ** have been thrown during name resolution and we would not have gotten + ** this far */ + if( pRet==0 && ALWAYS(iColpEList->nExpr) ){ + pRet = sqlite3ExprCollSeq(pParse, p->pEList->a[iCol].pExpr); + } + return pRet; +} + +/* +** The select statement passed as the second parameter is a compound SELECT +** with an ORDER BY clause. This function allocates and returns a KeyInfo +** structure suitable for implementing the ORDER BY. +** +** Space to hold the KeyInfo structure is obtained from malloc. The calling +** function is responsible for ensuring that this structure is eventually +** freed. +*/ +static KeyInfo *multiSelectOrderByKeyInfo(Parse *pParse, Select *p, int nExtra){ + ExprList *pOrderBy = p->pOrderBy; + int nOrderBy = p->pOrderBy->nExpr; + sqlite3 *db = pParse->db; + KeyInfo *pRet = sqlite3KeyInfoAlloc(db, nOrderBy+nExtra, 1); + if( pRet ){ + int i; + for(i=0; ia[i]; + Expr *pTerm = pItem->pExpr; + CollSeq *pColl; + + if( pTerm->flags & EP_Collate ){ + pColl = sqlite3ExprCollSeq(pParse, pTerm); + }else{ + pColl = multiSelectCollSeq(pParse, p, pItem->u.x.iOrderByCol-1); + if( pColl==0 ) pColl = db->pDfltColl; + pOrderBy->a[i].pExpr = + sqlite3ExprAddCollateString(pParse, pTerm, pColl->zName); + } + assert( sqlite3KeyInfoIsWriteable(pRet) ); + pRet->aColl[i] = pColl; + pRet->aSortFlags[i] = pOrderBy->a[i].sortFlags; + } + } + + return pRet; +} + +#ifndef SQLITE_OMIT_CTE +/* +** This routine generates VDBE code to compute the content of a WITH RECURSIVE +** query of the form: +** +** AS ( UNION [ALL] ) +** \___________/ \_______________/ +** p->pPrior p +** +** +** There is exactly one reference to the recursive-table in the FROM clause +** of recursive-query, marked with the SrcList->a[].fg.isRecursive flag. +** +** The setup-query runs once to generate an initial set of rows that go +** into a Queue table. Rows are extracted from the Queue table one by +** one. Each row extracted from Queue is output to pDest. Then the single +** extracted row (now in the iCurrent table) becomes the content of the +** recursive-table for a recursive-query run. The output of the recursive-query +** is added back into the Queue table. Then another row is extracted from Queue +** and the iteration continues until the Queue table is empty. +** +** If the compound query operator is UNION then no duplicate rows are ever +** inserted into the Queue table. The iDistinct table keeps a copy of all rows +** that have ever been inserted into Queue and causes duplicates to be +** discarded. If the operator is UNION ALL, then duplicates are allowed. +** +** If the query has an ORDER BY, then entries in the Queue table are kept in +** ORDER BY order and the first entry is extracted for each cycle. Without +** an ORDER BY, the Queue table is just a FIFO. +** +** If a LIMIT clause is provided, then the iteration stops after LIMIT rows +** have been output to pDest. A LIMIT of zero means to output no rows and a +** negative LIMIT means to output all rows. If there is also an OFFSET clause +** with a positive value, then the first OFFSET outputs are discarded rather +** than being sent to pDest. The LIMIT count does not begin until after OFFSET +** rows have been skipped. +*/ +static void generateWithRecursiveQuery( + Parse *pParse, /* Parsing context */ + Select *p, /* The recursive SELECT to be coded */ + SelectDest *pDest /* What to do with query results */ +){ + SrcList *pSrc = p->pSrc; /* The FROM clause of the recursive query */ + int nCol = p->pEList->nExpr; /* Number of columns in the recursive table */ + Vdbe *v = pParse->pVdbe; /* The prepared statement under construction */ + Select *pSetup = p->pPrior; /* The setup query */ + int addrTop; /* Top of the loop */ + int addrCont, addrBreak; /* CONTINUE and BREAK addresses */ + int iCurrent = 0; /* The Current table */ + int regCurrent; /* Register holding Current table */ + int iQueue; /* The Queue table */ + int iDistinct = 0; /* To ensure unique results if UNION */ + int eDest = SRT_Fifo; /* How to write to Queue */ + SelectDest destQueue; /* SelectDest targetting the Queue table */ + int i; /* Loop counter */ + int rc; /* Result code */ + ExprList *pOrderBy; /* The ORDER BY clause */ + Expr *pLimit; /* Saved LIMIT and OFFSET */ + int regLimit, regOffset; /* Registers used by LIMIT and OFFSET */ + +#ifndef SQLITE_OMIT_WINDOWFUNC + if( p->pWin ){ + sqlite3ErrorMsg(pParse, "cannot use window functions in recursive queries"); + return; + } +#endif + + /* Obtain authorization to do a recursive query */ + if( sqlite3AuthCheck(pParse, SQLITE_RECURSIVE, 0, 0, 0) ) return; + + /* Process the LIMIT and OFFSET clauses, if they exist */ + addrBreak = sqlite3VdbeMakeLabel(pParse); + p->nSelectRow = 320; /* 4 billion rows */ + computeLimitRegisters(pParse, p, addrBreak); + pLimit = p->pLimit; + regLimit = p->iLimit; + regOffset = p->iOffset; + p->pLimit = 0; + p->iLimit = p->iOffset = 0; + pOrderBy = p->pOrderBy; + + /* Locate the cursor number of the Current table */ + for(i=0; ALWAYS(inSrc); i++){ + if( pSrc->a[i].fg.isRecursive ){ + iCurrent = pSrc->a[i].iCursor; + break; + } + } + + /* Allocate cursors numbers for Queue and Distinct. The cursor number for + ** the Distinct table must be exactly one greater than Queue in order + ** for the SRT_DistFifo and SRT_DistQueue destinations to work. */ + iQueue = pParse->nTab++; + if( p->op==TK_UNION ){ + eDest = pOrderBy ? SRT_DistQueue : SRT_DistFifo; + iDistinct = pParse->nTab++; + }else{ + eDest = pOrderBy ? SRT_Queue : SRT_Fifo; + } + sqlite3SelectDestInit(&destQueue, eDest, iQueue); + + /* Allocate cursors for Current, Queue, and Distinct. */ + regCurrent = ++pParse->nMem; + sqlite3VdbeAddOp3(v, OP_OpenPseudo, iCurrent, regCurrent, nCol); + if( pOrderBy ){ + KeyInfo *pKeyInfo = multiSelectOrderByKeyInfo(pParse, p, 1); + sqlite3VdbeAddOp4(v, OP_OpenEphemeral, iQueue, pOrderBy->nExpr+2, 0, + (char*)pKeyInfo, P4_KEYINFO); + destQueue.pOrderBy = pOrderBy; + }else{ + sqlite3VdbeAddOp2(v, OP_OpenEphemeral, iQueue, nCol); + } + VdbeComment((v, "Queue table")); + if( iDistinct ){ + p->addrOpenEphm[0] = sqlite3VdbeAddOp2(v, OP_OpenEphemeral, iDistinct, 0); + p->selFlags |= SF_UsesEphemeral; + } + + /* Detach the ORDER BY clause from the compound SELECT */ + p->pOrderBy = 0; + + /* Store the results of the setup-query in Queue. */ + pSetup->pNext = 0; + ExplainQueryPlan((pParse, 1, "SETUP")); + rc = sqlite3Select(pParse, pSetup, &destQueue); + pSetup->pNext = p; + if( rc ) goto end_of_recursive_query; + + /* Find the next row in the Queue and output that row */ + addrTop = sqlite3VdbeAddOp2(v, OP_Rewind, iQueue, addrBreak); VdbeCoverage(v); + + /* Transfer the next row in Queue over to Current */ + sqlite3VdbeAddOp1(v, OP_NullRow, iCurrent); /* To reset column cache */ + if( pOrderBy ){ + sqlite3VdbeAddOp3(v, OP_Column, iQueue, pOrderBy->nExpr+1, regCurrent); + }else{ + sqlite3VdbeAddOp2(v, OP_RowData, iQueue, regCurrent); + } + sqlite3VdbeAddOp1(v, OP_Delete, iQueue); + + /* Output the single row in Current */ + addrCont = sqlite3VdbeMakeLabel(pParse); + codeOffset(v, regOffset, addrCont); + selectInnerLoop(pParse, p, iCurrent, + 0, 0, pDest, addrCont, addrBreak); + if( regLimit ){ + sqlite3VdbeAddOp2(v, OP_DecrJumpZero, regLimit, addrBreak); + VdbeCoverage(v); + } + sqlite3VdbeResolveLabel(v, addrCont); + + /* Execute the recursive SELECT taking the single row in Current as + ** the value for the recursive-table. Store the results in the Queue. + */ + if( p->selFlags & SF_Aggregate ){ + sqlite3ErrorMsg(pParse, "recursive aggregate queries not supported"); + }else{ + p->pPrior = 0; + ExplainQueryPlan((pParse, 1, "RECURSIVE STEP")); + sqlite3Select(pParse, p, &destQueue); + assert( p->pPrior==0 ); + p->pPrior = pSetup; + } + + /* Keep running the loop until the Queue is empty */ + sqlite3VdbeGoto(v, addrTop); + sqlite3VdbeResolveLabel(v, addrBreak); + +end_of_recursive_query: + sqlite3ExprListDelete(pParse->db, p->pOrderBy); + p->pOrderBy = pOrderBy; + p->pLimit = pLimit; + return; +} +#endif /* SQLITE_OMIT_CTE */ + +/* Forward references */ +static int multiSelectOrderBy( + Parse *pParse, /* Parsing context */ + Select *p, /* The right-most of SELECTs to be coded */ + SelectDest *pDest /* What to do with query results */ +); + +/* +** Handle the special case of a compound-select that originates from a +** VALUES clause. By handling this as a special case, we avoid deep +** recursion, and thus do not need to enforce the SQLITE_LIMIT_COMPOUND_SELECT +** on a VALUES clause. +** +** Because the Select object originates from a VALUES clause: +** (1) There is no LIMIT or OFFSET or else there is a LIMIT of exactly 1 +** (2) All terms are UNION ALL +** (3) There is no ORDER BY clause +** +** The "LIMIT of exactly 1" case of condition (1) comes about when a VALUES +** clause occurs within scalar expression (ex: "SELECT (VALUES(1),(2),(3))"). +** The sqlite3CodeSubselect will have added the LIMIT 1 clause in tht case. +** Since the limit is exactly 1, we only need to evalutes the left-most VALUES. +*/ +static int multiSelectValues( + Parse *pParse, /* Parsing context */ + Select *p, /* The right-most of SELECTs to be coded */ + SelectDest *pDest /* What to do with query results */ +){ + int nRow = 1; + int rc = 0; + int bShowAll = p->pLimit==0; + assert( p->selFlags & SF_MultiValue ); + do{ + assert( p->selFlags & SF_Values ); + assert( p->op==TK_ALL || (p->op==TK_SELECT && p->pPrior==0) ); + assert( p->pNext==0 || p->pEList->nExpr==p->pNext->pEList->nExpr ); + if( p->pPrior==0 ) break; + assert( p->pPrior->pNext==p ); + p = p->pPrior; + nRow += bShowAll; + }while(1); + ExplainQueryPlan((pParse, 0, "SCAN %d CONSTANT ROW%s", nRow, + nRow==1 ? "" : "S")); + while( p ){ + selectInnerLoop(pParse, p, -1, 0, 0, pDest, 1, 1); + if( !bShowAll ) break; + p->nSelectRow = nRow; + p = p->pNext; + } + return rc; +} + +/* +** This routine is called to process a compound query form from +** two or more separate queries using UNION, UNION ALL, EXCEPT, or +** INTERSECT +** +** "p" points to the right-most of the two queries. the query on the +** left is p->pPrior. The left query could also be a compound query +** in which case this routine will be called recursively. +** +** The results of the total query are to be written into a destination +** of type eDest with parameter iParm. +** +** Example 1: Consider a three-way compound SQL statement. +** +** SELECT a FROM t1 UNION SELECT b FROM t2 UNION SELECT c FROM t3 +** +** This statement is parsed up as follows: +** +** SELECT c FROM t3 +** | +** `-----> SELECT b FROM t2 +** | +** `------> SELECT a FROM t1 +** +** The arrows in the diagram above represent the Select.pPrior pointer. +** So if this routine is called with p equal to the t3 query, then +** pPrior will be the t2 query. p->op will be TK_UNION in this case. +** +** Notice that because of the way SQLite parses compound SELECTs, the +** individual selects always group from left to right. +*/ +static int multiSelect( + Parse *pParse, /* Parsing context */ + Select *p, /* The right-most of SELECTs to be coded */ + SelectDest *pDest /* What to do with query results */ +){ + int rc = SQLITE_OK; /* Success code from a subroutine */ + Select *pPrior; /* Another SELECT immediately to our left */ + Vdbe *v; /* Generate code to this VDBE */ + SelectDest dest; /* Alternative data destination */ + Select *pDelete = 0; /* Chain of simple selects to delete */ + sqlite3 *db; /* Database connection */ + + /* Make sure there is no ORDER BY or LIMIT clause on prior SELECTs. Only + ** the last (right-most) SELECT in the series may have an ORDER BY or LIMIT. + */ + assert( p && p->pPrior ); /* Calling function guarantees this much */ + assert( (p->selFlags & SF_Recursive)==0 || p->op==TK_ALL || p->op==TK_UNION ); + assert( p->selFlags & SF_Compound ); + db = pParse->db; + pPrior = p->pPrior; + dest = *pDest; + if( pPrior->pOrderBy || pPrior->pLimit ){ + sqlite3ErrorMsg(pParse,"%s clause should come after %s not before", + pPrior->pOrderBy!=0 ? "ORDER BY" : "LIMIT", selectOpName(p->op)); + rc = 1; + goto multi_select_end; + } + + v = sqlite3GetVdbe(pParse); + assert( v!=0 ); /* The VDBE already created by calling function */ + + /* Create the destination temporary table if necessary + */ + if( dest.eDest==SRT_EphemTab ){ + assert( p->pEList ); + sqlite3VdbeAddOp2(v, OP_OpenEphemeral, dest.iSDParm, p->pEList->nExpr); + dest.eDest = SRT_Table; + } + + /* Special handling for a compound-select that originates as a VALUES clause. + */ + if( p->selFlags & SF_MultiValue ){ + rc = multiSelectValues(pParse, p, &dest); + goto multi_select_end; + } + + /* Make sure all SELECTs in the statement have the same number of elements + ** in their result sets. + */ + assert( p->pEList && pPrior->pEList ); + assert( p->pEList->nExpr==pPrior->pEList->nExpr ); + +#ifndef SQLITE_OMIT_CTE + if( p->selFlags & SF_Recursive ){ + generateWithRecursiveQuery(pParse, p, &dest); + }else +#endif + + /* Compound SELECTs that have an ORDER BY clause are handled separately. + */ + if( p->pOrderBy ){ + return multiSelectOrderBy(pParse, p, pDest); + }else{ + +#ifndef SQLITE_OMIT_EXPLAIN + if( pPrior->pPrior==0 ){ + ExplainQueryPlan((pParse, 1, "COMPOUND QUERY")); + ExplainQueryPlan((pParse, 1, "LEFT-MOST SUBQUERY")); + } +#endif + + /* Generate code for the left and right SELECT statements. + */ + switch( p->op ){ + case TK_ALL: { + int addr = 0; + int nLimit; + assert( !pPrior->pLimit ); + pPrior->iLimit = p->iLimit; + pPrior->iOffset = p->iOffset; + pPrior->pLimit = p->pLimit; + rc = sqlite3Select(pParse, pPrior, &dest); + p->pLimit = 0; + if( rc ){ + goto multi_select_end; + } + p->pPrior = 0; + p->iLimit = pPrior->iLimit; + p->iOffset = pPrior->iOffset; + if( p->iLimit ){ + addr = sqlite3VdbeAddOp1(v, OP_IfNot, p->iLimit); VdbeCoverage(v); + VdbeComment((v, "Jump ahead if LIMIT reached")); + if( p->iOffset ){ + sqlite3VdbeAddOp3(v, OP_OffsetLimit, + p->iLimit, p->iOffset+1, p->iOffset); + } + } + ExplainQueryPlan((pParse, 1, "UNION ALL")); + rc = sqlite3Select(pParse, p, &dest); + testcase( rc!=SQLITE_OK ); + pDelete = p->pPrior; + p->pPrior = pPrior; + p->nSelectRow = sqlite3LogEstAdd(p->nSelectRow, pPrior->nSelectRow); + if( pPrior->pLimit + && sqlite3ExprIsInteger(pPrior->pLimit->pLeft, &nLimit) + && nLimit>0 && p->nSelectRow > sqlite3LogEst((u64)nLimit) + ){ + p->nSelectRow = sqlite3LogEst((u64)nLimit); + } + if( addr ){ + sqlite3VdbeJumpHere(v, addr); + } + break; + } + case TK_EXCEPT: + case TK_UNION: { + int unionTab; /* Cursor number of the temp table holding result */ + u8 op = 0; /* One of the SRT_ operations to apply to self */ + int priorOp; /* The SRT_ operation to apply to prior selects */ + Expr *pLimit; /* Saved values of p->nLimit */ + int addr; + SelectDest uniondest; + + testcase( p->op==TK_EXCEPT ); + testcase( p->op==TK_UNION ); + priorOp = SRT_Union; + if( dest.eDest==priorOp ){ + /* We can reuse a temporary table generated by a SELECT to our + ** right. + */ + assert( p->pLimit==0 ); /* Not allowed on leftward elements */ + unionTab = dest.iSDParm; + }else{ + /* We will need to create our own temporary table to hold the + ** intermediate results. + */ + unionTab = pParse->nTab++; + assert( p->pOrderBy==0 ); + addr = sqlite3VdbeAddOp2(v, OP_OpenEphemeral, unionTab, 0); + assert( p->addrOpenEphm[0] == -1 ); + p->addrOpenEphm[0] = addr; + findRightmost(p)->selFlags |= SF_UsesEphemeral; + assert( p->pEList ); + } + + /* Code the SELECT statements to our left + */ + assert( !pPrior->pOrderBy ); + sqlite3SelectDestInit(&uniondest, priorOp, unionTab); + rc = sqlite3Select(pParse, pPrior, &uniondest); + if( rc ){ + goto multi_select_end; + } + + /* Code the current SELECT statement + */ + if( p->op==TK_EXCEPT ){ + op = SRT_Except; + }else{ + assert( p->op==TK_UNION ); + op = SRT_Union; + } + p->pPrior = 0; + pLimit = p->pLimit; + p->pLimit = 0; + uniondest.eDest = op; + ExplainQueryPlan((pParse, 1, "%s USING TEMP B-TREE", + selectOpName(p->op))); + rc = sqlite3Select(pParse, p, &uniondest); + testcase( rc!=SQLITE_OK ); + /* Query flattening in sqlite3Select() might refill p->pOrderBy. + ** Be sure to delete p->pOrderBy, therefore, to avoid a memory leak. */ + sqlite3ExprListDelete(db, p->pOrderBy); + pDelete = p->pPrior; + p->pPrior = pPrior; + p->pOrderBy = 0; + if( p->op==TK_UNION ){ + p->nSelectRow = sqlite3LogEstAdd(p->nSelectRow, pPrior->nSelectRow); + } + sqlite3ExprDelete(db, p->pLimit); + p->pLimit = pLimit; + p->iLimit = 0; + p->iOffset = 0; + + /* Convert the data in the temporary table into whatever form + ** it is that we currently need. + */ + assert( unionTab==dest.iSDParm || dest.eDest!=priorOp ); + if( dest.eDest!=priorOp ){ + int iCont, iBreak, iStart; + assert( p->pEList ); + iBreak = sqlite3VdbeMakeLabel(pParse); + iCont = sqlite3VdbeMakeLabel(pParse); + computeLimitRegisters(pParse, p, iBreak); + sqlite3VdbeAddOp2(v, OP_Rewind, unionTab, iBreak); VdbeCoverage(v); + iStart = sqlite3VdbeCurrentAddr(v); + selectInnerLoop(pParse, p, unionTab, + 0, 0, &dest, iCont, iBreak); + sqlite3VdbeResolveLabel(v, iCont); + sqlite3VdbeAddOp2(v, OP_Next, unionTab, iStart); VdbeCoverage(v); + sqlite3VdbeResolveLabel(v, iBreak); + sqlite3VdbeAddOp2(v, OP_Close, unionTab, 0); + } + break; + } + default: assert( p->op==TK_INTERSECT ); { + int tab1, tab2; + int iCont, iBreak, iStart; + Expr *pLimit; + int addr; + SelectDest intersectdest; + int r1; + + /* INTERSECT is different from the others since it requires + ** two temporary tables. Hence it has its own case. Begin + ** by allocating the tables we will need. + */ + tab1 = pParse->nTab++; + tab2 = pParse->nTab++; + assert( p->pOrderBy==0 ); + + addr = sqlite3VdbeAddOp2(v, OP_OpenEphemeral, tab1, 0); + assert( p->addrOpenEphm[0] == -1 ); + p->addrOpenEphm[0] = addr; + findRightmost(p)->selFlags |= SF_UsesEphemeral; + assert( p->pEList ); + + /* Code the SELECTs to our left into temporary table "tab1". + */ + sqlite3SelectDestInit(&intersectdest, SRT_Union, tab1); + rc = sqlite3Select(pParse, pPrior, &intersectdest); + if( rc ){ + goto multi_select_end; + } + + /* Code the current SELECT into temporary table "tab2" + */ + addr = sqlite3VdbeAddOp2(v, OP_OpenEphemeral, tab2, 0); + assert( p->addrOpenEphm[1] == -1 ); + p->addrOpenEphm[1] = addr; + p->pPrior = 0; + pLimit = p->pLimit; + p->pLimit = 0; + intersectdest.iSDParm = tab2; + ExplainQueryPlan((pParse, 1, "%s USING TEMP B-TREE", + selectOpName(p->op))); + rc = sqlite3Select(pParse, p, &intersectdest); + testcase( rc!=SQLITE_OK ); + pDelete = p->pPrior; + p->pPrior = pPrior; + if( p->nSelectRow>pPrior->nSelectRow ){ + p->nSelectRow = pPrior->nSelectRow; + } + sqlite3ExprDelete(db, p->pLimit); + p->pLimit = pLimit; + + /* Generate code to take the intersection of the two temporary + ** tables. + */ + assert( p->pEList ); + iBreak = sqlite3VdbeMakeLabel(pParse); + iCont = sqlite3VdbeMakeLabel(pParse); + computeLimitRegisters(pParse, p, iBreak); + sqlite3VdbeAddOp2(v, OP_Rewind, tab1, iBreak); VdbeCoverage(v); + r1 = sqlite3GetTempReg(pParse); + iStart = sqlite3VdbeAddOp2(v, OP_RowData, tab1, r1); + sqlite3VdbeAddOp4Int(v, OP_NotFound, tab2, iCont, r1, 0); + VdbeCoverage(v); + sqlite3ReleaseTempReg(pParse, r1); + selectInnerLoop(pParse, p, tab1, + 0, 0, &dest, iCont, iBreak); + sqlite3VdbeResolveLabel(v, iCont); + sqlite3VdbeAddOp2(v, OP_Next, tab1, iStart); VdbeCoverage(v); + sqlite3VdbeResolveLabel(v, iBreak); + sqlite3VdbeAddOp2(v, OP_Close, tab2, 0); + sqlite3VdbeAddOp2(v, OP_Close, tab1, 0); + break; + } + } + + #ifndef SQLITE_OMIT_EXPLAIN + if( p->pNext==0 ){ + ExplainQueryPlanPop(pParse); + } + #endif + } + + /* Compute collating sequences used by + ** temporary tables needed to implement the compound select. + ** Attach the KeyInfo structure to all temporary tables. + ** + ** This section is run by the right-most SELECT statement only. + ** SELECT statements to the left always skip this part. The right-most + ** SELECT might also skip this part if it has no ORDER BY clause and + ** no temp tables are required. + */ + if( p->selFlags & SF_UsesEphemeral ){ + int i; /* Loop counter */ + KeyInfo *pKeyInfo; /* Collating sequence for the result set */ + Select *pLoop; /* For looping through SELECT statements */ + CollSeq **apColl; /* For looping through pKeyInfo->aColl[] */ + int nCol; /* Number of columns in result set */ + + assert( p->pNext==0 ); + nCol = p->pEList->nExpr; + pKeyInfo = sqlite3KeyInfoAlloc(db, nCol, 1); + if( !pKeyInfo ){ + rc = SQLITE_NOMEM_BKPT; + goto multi_select_end; + } + for(i=0, apColl=pKeyInfo->aColl; ipDfltColl; + } + } + + for(pLoop=p; pLoop; pLoop=pLoop->pPrior){ + for(i=0; i<2; i++){ + int addr = pLoop->addrOpenEphm[i]; + if( addr<0 ){ + /* If [0] is unused then [1] is also unused. So we can + ** always safely abort as soon as the first unused slot is found */ + assert( pLoop->addrOpenEphm[1]<0 ); + break; + } + sqlite3VdbeChangeP2(v, addr, nCol); + sqlite3VdbeChangeP4(v, addr, (char*)sqlite3KeyInfoRef(pKeyInfo), + P4_KEYINFO); + pLoop->addrOpenEphm[i] = -1; + } + } + sqlite3KeyInfoUnref(pKeyInfo); + } + +multi_select_end: + pDest->iSdst = dest.iSdst; + pDest->nSdst = dest.nSdst; + sqlite3SelectDelete(db, pDelete); + return rc; +} +#endif /* SQLITE_OMIT_COMPOUND_SELECT */ + +/* +** Error message for when two or more terms of a compound select have different +** size result sets. +*/ +SQLITE_PRIVATE void sqlite3SelectWrongNumTermsError(Parse *pParse, Select *p){ + if( p->selFlags & SF_Values ){ + sqlite3ErrorMsg(pParse, "all VALUES must have the same number of terms"); + }else{ + sqlite3ErrorMsg(pParse, "SELECTs to the left and right of %s" + " do not have the same number of result columns", selectOpName(p->op)); + } +} + +/* +** Code an output subroutine for a coroutine implementation of a +** SELECT statment. +** +** The data to be output is contained in pIn->iSdst. There are +** pIn->nSdst columns to be output. pDest is where the output should +** be sent. +** +** regReturn is the number of the register holding the subroutine +** return address. +** +** If regPrev>0 then it is the first register in a vector that +** records the previous output. mem[regPrev] is a flag that is false +** if there has been no previous output. If regPrev>0 then code is +** generated to suppress duplicates. pKeyInfo is used for comparing +** keys. +** +** If the LIMIT found in p->iLimit is reached, jump immediately to +** iBreak. +*/ +static int generateOutputSubroutine( + Parse *pParse, /* Parsing context */ + Select *p, /* The SELECT statement */ + SelectDest *pIn, /* Coroutine supplying data */ + SelectDest *pDest, /* Where to send the data */ + int regReturn, /* The return address register */ + int regPrev, /* Previous result register. No uniqueness if 0 */ + KeyInfo *pKeyInfo, /* For comparing with previous entry */ + int iBreak /* Jump here if we hit the LIMIT */ +){ + Vdbe *v = pParse->pVdbe; + int iContinue; + int addr; + + addr = sqlite3VdbeCurrentAddr(v); + iContinue = sqlite3VdbeMakeLabel(pParse); + + /* Suppress duplicates for UNION, EXCEPT, and INTERSECT + */ + if( regPrev ){ + int addr1, addr2; + addr1 = sqlite3VdbeAddOp1(v, OP_IfNot, regPrev); VdbeCoverage(v); + addr2 = sqlite3VdbeAddOp4(v, OP_Compare, pIn->iSdst, regPrev+1, pIn->nSdst, + (char*)sqlite3KeyInfoRef(pKeyInfo), P4_KEYINFO); + sqlite3VdbeAddOp3(v, OP_Jump, addr2+2, iContinue, addr2+2); VdbeCoverage(v); + sqlite3VdbeJumpHere(v, addr1); + sqlite3VdbeAddOp3(v, OP_Copy, pIn->iSdst, regPrev+1, pIn->nSdst-1); + sqlite3VdbeAddOp2(v, OP_Integer, 1, regPrev); + } + if( pParse->db->mallocFailed ) return 0; + + /* Suppress the first OFFSET entries if there is an OFFSET clause + */ + codeOffset(v, p->iOffset, iContinue); + + assert( pDest->eDest!=SRT_Exists ); + assert( pDest->eDest!=SRT_Table ); + switch( pDest->eDest ){ + /* Store the result as data using a unique key. + */ + case SRT_EphemTab: { + int r1 = sqlite3GetTempReg(pParse); + int r2 = sqlite3GetTempReg(pParse); + sqlite3VdbeAddOp3(v, OP_MakeRecord, pIn->iSdst, pIn->nSdst, r1); + sqlite3VdbeAddOp2(v, OP_NewRowid, pDest->iSDParm, r2); + sqlite3VdbeAddOp3(v, OP_Insert, pDest->iSDParm, r1, r2); + sqlite3VdbeChangeP5(v, OPFLAG_APPEND); + sqlite3ReleaseTempReg(pParse, r2); + sqlite3ReleaseTempReg(pParse, r1); + break; + } + +#ifndef SQLITE_OMIT_SUBQUERY + /* If we are creating a set for an "expr IN (SELECT ...)". + */ + case SRT_Set: { + int r1; + testcase( pIn->nSdst>1 ); + r1 = sqlite3GetTempReg(pParse); + sqlite3VdbeAddOp4(v, OP_MakeRecord, pIn->iSdst, pIn->nSdst, + r1, pDest->zAffSdst, pIn->nSdst); + sqlite3VdbeAddOp4Int(v, OP_IdxInsert, pDest->iSDParm, r1, + pIn->iSdst, pIn->nSdst); + sqlite3ReleaseTempReg(pParse, r1); + break; + } + + /* If this is a scalar select that is part of an expression, then + ** store the results in the appropriate memory cell and break out + ** of the scan loop. Note that the select might return multiple columns + ** if it is the RHS of a row-value IN operator. + */ + case SRT_Mem: { + if( pParse->nErr==0 ){ + testcase( pIn->nSdst>1 ); + sqlite3ExprCodeMove(pParse, pIn->iSdst, pDest->iSDParm, pIn->nSdst); + } + /* The LIMIT clause will jump out of the loop for us */ + break; + } +#endif /* #ifndef SQLITE_OMIT_SUBQUERY */ + + /* The results are stored in a sequence of registers + ** starting at pDest->iSdst. Then the co-routine yields. + */ + case SRT_Coroutine: { + if( pDest->iSdst==0 ){ + pDest->iSdst = sqlite3GetTempRange(pParse, pIn->nSdst); + pDest->nSdst = pIn->nSdst; + } + sqlite3ExprCodeMove(pParse, pIn->iSdst, pDest->iSdst, pIn->nSdst); + sqlite3VdbeAddOp1(v, OP_Yield, pDest->iSDParm); + break; + } + + /* If none of the above, then the result destination must be + ** SRT_Output. This routine is never called with any other + ** destination other than the ones handled above or SRT_Output. + ** + ** For SRT_Output, results are stored in a sequence of registers. + ** Then the OP_ResultRow opcode is used to cause sqlite3_step() to + ** return the next row of result. + */ + default: { + assert( pDest->eDest==SRT_Output ); + sqlite3VdbeAddOp2(v, OP_ResultRow, pIn->iSdst, pIn->nSdst); + break; + } + } + + /* Jump to the end of the loop if the LIMIT is reached. + */ + if( p->iLimit ){ + sqlite3VdbeAddOp2(v, OP_DecrJumpZero, p->iLimit, iBreak); VdbeCoverage(v); + } + + /* Generate the subroutine return + */ + sqlite3VdbeResolveLabel(v, iContinue); + sqlite3VdbeAddOp1(v, OP_Return, regReturn); + + return addr; +} + +/* +** Alternative compound select code generator for cases when there +** is an ORDER BY clause. +** +** We assume a query of the following form: +** +** ORDER BY +** +** is one of UNION ALL, UNION, EXCEPT, or INTERSECT. The idea +** is to code both and with the ORDER BY clause as +** co-routines. Then run the co-routines in parallel and merge the results +** into the output. In addition to the two coroutines (called selectA and +** selectB) there are 7 subroutines: +** +** outA: Move the output of the selectA coroutine into the output +** of the compound query. +** +** outB: Move the output of the selectB coroutine into the output +** of the compound query. (Only generated for UNION and +** UNION ALL. EXCEPT and INSERTSECT never output a row that +** appears only in B.) +** +** AltB: Called when there is data from both coroutines and AB. +** +** EofA: Called when data is exhausted from selectA. +** +** EofB: Called when data is exhausted from selectB. +** +** The implementation of the latter five subroutines depend on which +** is used: +** +** +** UNION ALL UNION EXCEPT INTERSECT +** ------------- ----------------- -------------- ----------------- +** AltB: outA, nextA outA, nextA outA, nextA nextA +** +** AeqB: outA, nextA nextA nextA outA, nextA +** +** AgtB: outB, nextB outB, nextB nextB nextB +** +** EofA: outB, nextB outB, nextB halt halt +** +** EofB: outA, nextA outA, nextA outA, nextA halt +** +** In the AltB, AeqB, and AgtB subroutines, an EOF on A following nextA +** causes an immediate jump to EofA and an EOF on B following nextB causes +** an immediate jump to EofB. Within EofA and EofB, and EOF on entry or +** following nextX causes a jump to the end of the select processing. +** +** Duplicate removal in the UNION, EXCEPT, and INTERSECT cases is handled +** within the output subroutine. The regPrev register set holds the previously +** output value. A comparison is made against this value and the output +** is skipped if the next results would be the same as the previous. +** +** The implementation plan is to implement the two coroutines and seven +** subroutines first, then put the control logic at the bottom. Like this: +** +** goto Init +** coA: coroutine for left query (A) +** coB: coroutine for right query (B) +** outA: output one row of A +** outB: output one row of B (UNION and UNION ALL only) +** EofA: ... +** EofB: ... +** AltB: ... +** AeqB: ... +** AgtB: ... +** Init: initialize coroutine registers +** yield coA +** if eof(A) goto EofA +** yield coB +** if eof(B) goto EofB +** Cmpr: Compare A, B +** Jump AltB, AeqB, AgtB +** End: ... +** +** We call AltB, AeqB, AgtB, EofA, and EofB "subroutines" but they are not +** actually called using Gosub and they do not Return. EofA and EofB loop +** until all data is exhausted then jump to the "end" labe. AltB, AeqB, +** and AgtB jump to either L2 or to one of EofA or EofB. +*/ +#ifndef SQLITE_OMIT_COMPOUND_SELECT +static int multiSelectOrderBy( + Parse *pParse, /* Parsing context */ + Select *p, /* The right-most of SELECTs to be coded */ + SelectDest *pDest /* What to do with query results */ +){ + int i, j; /* Loop counters */ + Select *pPrior; /* Another SELECT immediately to our left */ + Vdbe *v; /* Generate code to this VDBE */ + SelectDest destA; /* Destination for coroutine A */ + SelectDest destB; /* Destination for coroutine B */ + int regAddrA; /* Address register for select-A coroutine */ + int regAddrB; /* Address register for select-B coroutine */ + int addrSelectA; /* Address of the select-A coroutine */ + int addrSelectB; /* Address of the select-B coroutine */ + int regOutA; /* Address register for the output-A subroutine */ + int regOutB; /* Address register for the output-B subroutine */ + int addrOutA; /* Address of the output-A subroutine */ + int addrOutB = 0; /* Address of the output-B subroutine */ + int addrEofA; /* Address of the select-A-exhausted subroutine */ + int addrEofA_noB; /* Alternate addrEofA if B is uninitialized */ + int addrEofB; /* Address of the select-B-exhausted subroutine */ + int addrAltB; /* Address of the AB subroutine */ + int regLimitA; /* Limit register for select-A */ + int regLimitB; /* Limit register for select-A */ + int regPrev; /* A range of registers to hold previous output */ + int savedLimit; /* Saved value of p->iLimit */ + int savedOffset; /* Saved value of p->iOffset */ + int labelCmpr; /* Label for the start of the merge algorithm */ + int labelEnd; /* Label for the end of the overall SELECT stmt */ + int addr1; /* Jump instructions that get retargetted */ + int op; /* One of TK_ALL, TK_UNION, TK_EXCEPT, TK_INTERSECT */ + KeyInfo *pKeyDup = 0; /* Comparison information for duplicate removal */ + KeyInfo *pKeyMerge; /* Comparison information for merging rows */ + sqlite3 *db; /* Database connection */ + ExprList *pOrderBy; /* The ORDER BY clause */ + int nOrderBy; /* Number of terms in the ORDER BY clause */ + int *aPermute; /* Mapping from ORDER BY terms to result set columns */ + + assert( p->pOrderBy!=0 ); + assert( pKeyDup==0 ); /* "Managed" code needs this. Ticket #3382. */ + db = pParse->db; + v = pParse->pVdbe; + assert( v!=0 ); /* Already thrown the error if VDBE alloc failed */ + labelEnd = sqlite3VdbeMakeLabel(pParse); + labelCmpr = sqlite3VdbeMakeLabel(pParse); + + + /* Patch up the ORDER BY clause + */ + op = p->op; + pPrior = p->pPrior; + assert( pPrior->pOrderBy==0 ); + pOrderBy = p->pOrderBy; + assert( pOrderBy ); + nOrderBy = pOrderBy->nExpr; + + /* For operators other than UNION ALL we have to make sure that + ** the ORDER BY clause covers every term of the result set. Add + ** terms to the ORDER BY clause as necessary. + */ + if( op!=TK_ALL ){ + for(i=1; db->mallocFailed==0 && i<=p->pEList->nExpr; i++){ + struct ExprList_item *pItem; + for(j=0, pItem=pOrderBy->a; ju.x.iOrderByCol>0 ); + if( pItem->u.x.iOrderByCol==i ) break; + } + if( j==nOrderBy ){ + Expr *pNew = sqlite3Expr(db, TK_INTEGER, 0); + if( pNew==0 ) return SQLITE_NOMEM_BKPT; + pNew->flags |= EP_IntValue; + pNew->u.iValue = i; + p->pOrderBy = pOrderBy = sqlite3ExprListAppend(pParse, pOrderBy, pNew); + if( pOrderBy ) pOrderBy->a[nOrderBy++].u.x.iOrderByCol = (u16)i; + } + } + } + + /* Compute the comparison permutation and keyinfo that is used with + ** the permutation used to determine if the next + ** row of results comes from selectA or selectB. Also add explicit + ** collations to the ORDER BY clause terms so that when the subqueries + ** to the right and the left are evaluated, they use the correct + ** collation. + */ + aPermute = sqlite3DbMallocRawNN(db, sizeof(int)*(nOrderBy + 1)); + if( aPermute ){ + struct ExprList_item *pItem; + aPermute[0] = nOrderBy; + for(i=1, pItem=pOrderBy->a; i<=nOrderBy; i++, pItem++){ + assert( pItem->u.x.iOrderByCol>0 ); + assert( pItem->u.x.iOrderByCol<=p->pEList->nExpr ); + aPermute[i] = pItem->u.x.iOrderByCol - 1; + } + pKeyMerge = multiSelectOrderByKeyInfo(pParse, p, 1); + }else{ + pKeyMerge = 0; + } + + /* Reattach the ORDER BY clause to the query. + */ + p->pOrderBy = pOrderBy; + pPrior->pOrderBy = sqlite3ExprListDup(pParse->db, pOrderBy, 0); + + /* Allocate a range of temporary registers and the KeyInfo needed + ** for the logic that removes duplicate result rows when the + ** operator is UNION, EXCEPT, or INTERSECT (but not UNION ALL). + */ + if( op==TK_ALL ){ + regPrev = 0; + }else{ + int nExpr = p->pEList->nExpr; + assert( nOrderBy>=nExpr || db->mallocFailed ); + regPrev = pParse->nMem+1; + pParse->nMem += nExpr+1; + sqlite3VdbeAddOp2(v, OP_Integer, 0, regPrev); + pKeyDup = sqlite3KeyInfoAlloc(db, nExpr, 1); + if( pKeyDup ){ + assert( sqlite3KeyInfoIsWriteable(pKeyDup) ); + for(i=0; iaColl[i] = multiSelectCollSeq(pParse, p, i); + pKeyDup->aSortFlags[i] = 0; + } + } + } + + /* Separate the left and the right query from one another + */ + p->pPrior = 0; + pPrior->pNext = 0; + sqlite3ResolveOrderGroupBy(pParse, p, p->pOrderBy, "ORDER"); + if( pPrior->pPrior==0 ){ + sqlite3ResolveOrderGroupBy(pParse, pPrior, pPrior->pOrderBy, "ORDER"); + } + + /* Compute the limit registers */ + computeLimitRegisters(pParse, p, labelEnd); + if( p->iLimit && op==TK_ALL ){ + regLimitA = ++pParse->nMem; + regLimitB = ++pParse->nMem; + sqlite3VdbeAddOp2(v, OP_Copy, p->iOffset ? p->iOffset+1 : p->iLimit, + regLimitA); + sqlite3VdbeAddOp2(v, OP_Copy, regLimitA, regLimitB); + }else{ + regLimitA = regLimitB = 0; + } + sqlite3ExprDelete(db, p->pLimit); + p->pLimit = 0; + + regAddrA = ++pParse->nMem; + regAddrB = ++pParse->nMem; + regOutA = ++pParse->nMem; + regOutB = ++pParse->nMem; + sqlite3SelectDestInit(&destA, SRT_Coroutine, regAddrA); + sqlite3SelectDestInit(&destB, SRT_Coroutine, regAddrB); + + ExplainQueryPlan((pParse, 1, "MERGE (%s)", selectOpName(p->op))); + + /* Generate a coroutine to evaluate the SELECT statement to the + ** left of the compound operator - the "A" select. + */ + addrSelectA = sqlite3VdbeCurrentAddr(v) + 1; + addr1 = sqlite3VdbeAddOp3(v, OP_InitCoroutine, regAddrA, 0, addrSelectA); + VdbeComment((v, "left SELECT")); + pPrior->iLimit = regLimitA; + ExplainQueryPlan((pParse, 1, "LEFT")); + sqlite3Select(pParse, pPrior, &destA); + sqlite3VdbeEndCoroutine(v, regAddrA); + sqlite3VdbeJumpHere(v, addr1); + + /* Generate a coroutine to evaluate the SELECT statement on + ** the right - the "B" select + */ + addrSelectB = sqlite3VdbeCurrentAddr(v) + 1; + addr1 = sqlite3VdbeAddOp3(v, OP_InitCoroutine, regAddrB, 0, addrSelectB); + VdbeComment((v, "right SELECT")); + savedLimit = p->iLimit; + savedOffset = p->iOffset; + p->iLimit = regLimitB; + p->iOffset = 0; + ExplainQueryPlan((pParse, 1, "RIGHT")); + sqlite3Select(pParse, p, &destB); + p->iLimit = savedLimit; + p->iOffset = savedOffset; + sqlite3VdbeEndCoroutine(v, regAddrB); + + /* Generate a subroutine that outputs the current row of the A + ** select as the next output row of the compound select. + */ + VdbeNoopComment((v, "Output routine for A")); + addrOutA = generateOutputSubroutine(pParse, + p, &destA, pDest, regOutA, + regPrev, pKeyDup, labelEnd); + + /* Generate a subroutine that outputs the current row of the B + ** select as the next output row of the compound select. + */ + if( op==TK_ALL || op==TK_UNION ){ + VdbeNoopComment((v, "Output routine for B")); + addrOutB = generateOutputSubroutine(pParse, + p, &destB, pDest, regOutB, + regPrev, pKeyDup, labelEnd); + } + sqlite3KeyInfoUnref(pKeyDup); + + /* Generate a subroutine to run when the results from select A + ** are exhausted and only data in select B remains. + */ + if( op==TK_EXCEPT || op==TK_INTERSECT ){ + addrEofA_noB = addrEofA = labelEnd; + }else{ + VdbeNoopComment((v, "eof-A subroutine")); + addrEofA = sqlite3VdbeAddOp2(v, OP_Gosub, regOutB, addrOutB); + addrEofA_noB = sqlite3VdbeAddOp2(v, OP_Yield, regAddrB, labelEnd); + VdbeCoverage(v); + sqlite3VdbeGoto(v, addrEofA); + p->nSelectRow = sqlite3LogEstAdd(p->nSelectRow, pPrior->nSelectRow); + } + + /* Generate a subroutine to run when the results from select B + ** are exhausted and only data in select A remains. + */ + if( op==TK_INTERSECT ){ + addrEofB = addrEofA; + if( p->nSelectRow > pPrior->nSelectRow ) p->nSelectRow = pPrior->nSelectRow; + }else{ + VdbeNoopComment((v, "eof-B subroutine")); + addrEofB = sqlite3VdbeAddOp2(v, OP_Gosub, regOutA, addrOutA); + sqlite3VdbeAddOp2(v, OP_Yield, regAddrA, labelEnd); VdbeCoverage(v); + sqlite3VdbeGoto(v, addrEofB); + } + + /* Generate code to handle the case of AB + */ + VdbeNoopComment((v, "A-gt-B subroutine")); + addrAgtB = sqlite3VdbeCurrentAddr(v); + if( op==TK_ALL || op==TK_UNION ){ + sqlite3VdbeAddOp2(v, OP_Gosub, regOutB, addrOutB); + } + sqlite3VdbeAddOp2(v, OP_Yield, regAddrB, addrEofB); VdbeCoverage(v); + sqlite3VdbeGoto(v, labelCmpr); + + /* This code runs once to initialize everything. + */ + sqlite3VdbeJumpHere(v, addr1); + sqlite3VdbeAddOp2(v, OP_Yield, regAddrA, addrEofA_noB); VdbeCoverage(v); + sqlite3VdbeAddOp2(v, OP_Yield, regAddrB, addrEofB); VdbeCoverage(v); + + /* Implement the main merge loop + */ + sqlite3VdbeResolveLabel(v, labelCmpr); + sqlite3VdbeAddOp4(v, OP_Permutation, 0, 0, 0, (char*)aPermute, P4_INTARRAY); + sqlite3VdbeAddOp4(v, OP_Compare, destA.iSdst, destB.iSdst, nOrderBy, + (char*)pKeyMerge, P4_KEYINFO); + sqlite3VdbeChangeP5(v, OPFLAG_PERMUTE); + sqlite3VdbeAddOp3(v, OP_Jump, addrAltB, addrAeqB, addrAgtB); VdbeCoverage(v); + + /* Jump to the this point in order to terminate the query. + */ + sqlite3VdbeResolveLabel(v, labelEnd); + + /* Reassembly the compound query so that it will be freed correctly + ** by the calling function */ + if( p->pPrior ){ + sqlite3SelectDelete(db, p->pPrior); + } + p->pPrior = pPrior; + pPrior->pNext = p; + + /*** TBD: Insert subroutine calls to close cursors on incomplete + **** subqueries ****/ + ExplainQueryPlanPop(pParse); + return pParse->nErr!=0; +} +#endif + +#if !defined(SQLITE_OMIT_SUBQUERY) || !defined(SQLITE_OMIT_VIEW) + +/* An instance of the SubstContext object describes an substitution edit +** to be performed on a parse tree. +** +** All references to columns in table iTable are to be replaced by corresponding +** expressions in pEList. +*/ +typedef struct SubstContext { + Parse *pParse; /* The parsing context */ + int iTable; /* Replace references to this table */ + int iNewTable; /* New table number */ + int isLeftJoin; /* Add TK_IF_NULL_ROW opcodes on each replacement */ + ExprList *pEList; /* Replacement expressions */ +} SubstContext; + +/* Forward Declarations */ +static void substExprList(SubstContext*, ExprList*); +static void substSelect(SubstContext*, Select*, int); + +/* +** Scan through the expression pExpr. Replace every reference to +** a column in table number iTable with a copy of the iColumn-th +** entry in pEList. (But leave references to the ROWID column +** unchanged.) +** +** This routine is part of the flattening procedure. A subquery +** whose result set is defined by pEList appears as entry in the +** FROM clause of a SELECT such that the VDBE cursor assigned to that +** FORM clause entry is iTable. This routine makes the necessary +** changes to pExpr so that it refers directly to the source table +** of the subquery rather the result set of the subquery. +*/ +static Expr *substExpr( + SubstContext *pSubst, /* Description of the substitution */ + Expr *pExpr /* Expr in which substitution occurs */ +){ + if( pExpr==0 ) return 0; + if( ExprHasProperty(pExpr, EP_FromJoin) + && pExpr->iRightJoinTable==pSubst->iTable + ){ + pExpr->iRightJoinTable = pSubst->iNewTable; + } + if( pExpr->op==TK_COLUMN && pExpr->iTable==pSubst->iTable ){ + if( pExpr->iColumn<0 ){ + pExpr->op = TK_NULL; + }else{ + Expr *pNew; + Expr *pCopy = pSubst->pEList->a[pExpr->iColumn].pExpr; + Expr ifNullRow; + assert( pSubst->pEList!=0 && pExpr->iColumnpEList->nExpr ); + assert( pExpr->pRight==0 ); + if( sqlite3ExprIsVector(pCopy) ){ + sqlite3VectorErrorMsg(pSubst->pParse, pCopy); + }else{ + sqlite3 *db = pSubst->pParse->db; + if( pSubst->isLeftJoin && pCopy->op!=TK_COLUMN ){ + memset(&ifNullRow, 0, sizeof(ifNullRow)); + ifNullRow.op = TK_IF_NULL_ROW; + ifNullRow.pLeft = pCopy; + ifNullRow.iTable = pSubst->iNewTable; + pCopy = &ifNullRow; + } + testcase( ExprHasProperty(pCopy, EP_Subquery) ); + pNew = sqlite3ExprDup(db, pCopy, 0); + if( pNew && pSubst->isLeftJoin ){ + ExprSetProperty(pNew, EP_CanBeNull); + } + if( pNew && ExprHasProperty(pExpr,EP_FromJoin) ){ + pNew->iRightJoinTable = pExpr->iRightJoinTable; + ExprSetProperty(pNew, EP_FromJoin); + } + sqlite3ExprDelete(db, pExpr); + pExpr = pNew; + + /* Ensure that the expression now has an implicit collation sequence, + ** just as it did when it was a column of a view or sub-query. */ + if( pExpr ){ + if( pExpr->op!=TK_COLUMN && pExpr->op!=TK_COLLATE ){ + CollSeq *pColl = sqlite3ExprCollSeq(pSubst->pParse, pExpr); + pExpr = sqlite3ExprAddCollateString(pSubst->pParse, pExpr, + (pColl ? pColl->zName : "BINARY") + ); + } + ExprClearProperty(pExpr, EP_Collate); + } + } + } + }else{ + if( pExpr->op==TK_IF_NULL_ROW && pExpr->iTable==pSubst->iTable ){ + pExpr->iTable = pSubst->iNewTable; + } + pExpr->pLeft = substExpr(pSubst, pExpr->pLeft); + pExpr->pRight = substExpr(pSubst, pExpr->pRight); + if( ExprHasProperty(pExpr, EP_xIsSelect) ){ + substSelect(pSubst, pExpr->x.pSelect, 1); + }else{ + substExprList(pSubst, pExpr->x.pList); + } +#ifndef SQLITE_OMIT_WINDOWFUNC + if( ExprHasProperty(pExpr, EP_WinFunc) ){ + Window *pWin = pExpr->y.pWin; + pWin->pFilter = substExpr(pSubst, pWin->pFilter); + substExprList(pSubst, pWin->pPartition); + substExprList(pSubst, pWin->pOrderBy); + } +#endif + } + return pExpr; +} +static void substExprList( + SubstContext *pSubst, /* Description of the substitution */ + ExprList *pList /* List to scan and in which to make substitutes */ +){ + int i; + if( pList==0 ) return; + for(i=0; inExpr; i++){ + pList->a[i].pExpr = substExpr(pSubst, pList->a[i].pExpr); + } +} +static void substSelect( + SubstContext *pSubst, /* Description of the substitution */ + Select *p, /* SELECT statement in which to make substitutions */ + int doPrior /* Do substitutes on p->pPrior too */ +){ + SrcList *pSrc; + struct SrcList_item *pItem; + int i; + if( !p ) return; + do{ + substExprList(pSubst, p->pEList); + substExprList(pSubst, p->pGroupBy); + substExprList(pSubst, p->pOrderBy); + p->pHaving = substExpr(pSubst, p->pHaving); + p->pWhere = substExpr(pSubst, p->pWhere); + pSrc = p->pSrc; + assert( pSrc!=0 ); + for(i=pSrc->nSrc, pItem=pSrc->a; i>0; i--, pItem++){ + substSelect(pSubst, pItem->pSelect, 1); + if( pItem->fg.isTabFunc ){ + substExprList(pSubst, pItem->u1.pFuncArg); + } + } + }while( doPrior && (p = p->pPrior)!=0 ); +} +#endif /* !defined(SQLITE_OMIT_SUBQUERY) || !defined(SQLITE_OMIT_VIEW) */ + +#if !defined(SQLITE_OMIT_SUBQUERY) || !defined(SQLITE_OMIT_VIEW) +/* +** This routine attempts to flatten subqueries as a performance optimization. +** This routine returns 1 if it makes changes and 0 if no flattening occurs. +** +** To understand the concept of flattening, consider the following +** query: +** +** SELECT a FROM (SELECT x+y AS a FROM t1 WHERE z<100) WHERE a>5 +** +** The default way of implementing this query is to execute the +** subquery first and store the results in a temporary table, then +** run the outer query on that temporary table. This requires two +** passes over the data. Furthermore, because the temporary table +** has no indices, the WHERE clause on the outer query cannot be +** optimized. +** +** This routine attempts to rewrite queries such as the above into +** a single flat select, like this: +** +** SELECT x+y AS a FROM t1 WHERE z<100 AND a>5 +** +** The code generated for this simplification gives the same result +** but only has to scan the data once. And because indices might +** exist on the table t1, a complete scan of the data might be +** avoided. +** +** Flattening is subject to the following constraints: +** +** (**) We no longer attempt to flatten aggregate subqueries. Was: +** The subquery and the outer query cannot both be aggregates. +** +** (**) We no longer attempt to flatten aggregate subqueries. Was: +** (2) If the subquery is an aggregate then +** (2a) the outer query must not be a join and +** (2b) the outer query must not use subqueries +** other than the one FROM-clause subquery that is a candidate +** for flattening. (This is due to ticket [2f7170d73bf9abf80] +** from 2015-02-09.) +** +** (3) If the subquery is the right operand of a LEFT JOIN then +** (3a) the subquery may not be a join and +** (3b) the FROM clause of the subquery may not contain a virtual +** table and +** (3c) the outer query may not be an aggregate. +** +** (4) The subquery can not be DISTINCT. +** +** (**) At one point restrictions (4) and (5) defined a subset of DISTINCT +** sub-queries that were excluded from this optimization. Restriction +** (4) has since been expanded to exclude all DISTINCT subqueries. +** +** (**) We no longer attempt to flatten aggregate subqueries. Was: +** If the subquery is aggregate, the outer query may not be DISTINCT. +** +** (7) The subquery must have a FROM clause. TODO: For subqueries without +** A FROM clause, consider adding a FROM clause with the special +** table sqlite_once that consists of a single row containing a +** single NULL. +** +** (8) If the subquery uses LIMIT then the outer query may not be a join. +** +** (9) If the subquery uses LIMIT then the outer query may not be aggregate. +** +** (**) Restriction (10) was removed from the code on 2005-02-05 but we +** accidently carried the comment forward until 2014-09-15. Original +** constraint: "If the subquery is aggregate then the outer query +** may not use LIMIT." +** +** (11) The subquery and the outer query may not both have ORDER BY clauses. +** +** (**) Not implemented. Subsumed into restriction (3). Was previously +** a separate restriction deriving from ticket #350. +** +** (13) The subquery and outer query may not both use LIMIT. +** +** (14) The subquery may not use OFFSET. +** +** (15) If the outer query is part of a compound select, then the +** subquery may not use LIMIT. +** (See ticket #2339 and ticket [02a8e81d44]). +** +** (16) If the outer query is aggregate, then the subquery may not +** use ORDER BY. (Ticket #2942) This used to not matter +** until we introduced the group_concat() function. +** +** (17) If the subquery is a compound select, then +** (17a) all compound operators must be a UNION ALL, and +** (17b) no terms within the subquery compound may be aggregate +** or DISTINCT, and +** (17c) every term within the subquery compound must have a FROM clause +** (17d) the outer query may not be +** (17d1) aggregate, or +** (17d2) DISTINCT, or +** (17d3) a join. +** +** The parent and sub-query may contain WHERE clauses. Subject to +** rules (11), (13) and (14), they may also contain ORDER BY, +** LIMIT and OFFSET clauses. The subquery cannot use any compound +** operator other than UNION ALL because all the other compound +** operators have an implied DISTINCT which is disallowed by +** restriction (4). +** +** Also, each component of the sub-query must return the same number +** of result columns. This is actually a requirement for any compound +** SELECT statement, but all the code here does is make sure that no +** such (illegal) sub-query is flattened. The caller will detect the +** syntax error and return a detailed message. +** +** (18) If the sub-query is a compound select, then all terms of the +** ORDER BY clause of the parent must be simple references to +** columns of the sub-query. +** +** (19) If the subquery uses LIMIT then the outer query may not +** have a WHERE clause. +** +** (20) If the sub-query is a compound select, then it must not use +** an ORDER BY clause. Ticket #3773. We could relax this constraint +** somewhat by saying that the terms of the ORDER BY clause must +** appear as unmodified result columns in the outer query. But we +** have other optimizations in mind to deal with that case. +** +** (21) If the subquery uses LIMIT then the outer query may not be +** DISTINCT. (See ticket [752e1646fc]). +** +** (22) The subquery may not be a recursive CTE. +** +** (**) Subsumed into restriction (17d3). Was: If the outer query is +** a recursive CTE, then the sub-query may not be a compound query. +** This restriction is because transforming the +** parent to a compound query confuses the code that handles +** recursive queries in multiSelect(). +** +** (**) We no longer attempt to flatten aggregate subqueries. Was: +** The subquery may not be an aggregate that uses the built-in min() or +** or max() functions. (Without this restriction, a query like: +** "SELECT x FROM (SELECT max(y), x FROM t1)" would not necessarily +** return the value X for which Y was maximal.) +** +** (25) If either the subquery or the parent query contains a window +** function in the select list or ORDER BY clause, flattening +** is not attempted. +** +** +** In this routine, the "p" parameter is a pointer to the outer query. +** The subquery is p->pSrc->a[iFrom]. isAgg is true if the outer query +** uses aggregates. +** +** If flattening is not attempted, this routine is a no-op and returns 0. +** If flattening is attempted this routine returns 1. +** +** All of the expression analysis must occur on both the outer query and +** the subquery before this routine runs. +*/ +static int flattenSubquery( + Parse *pParse, /* Parsing context */ + Select *p, /* The parent or outer SELECT statement */ + int iFrom, /* Index in p->pSrc->a[] of the inner subquery */ + int isAgg /* True if outer SELECT uses aggregate functions */ +){ + const char *zSavedAuthContext = pParse->zAuthContext; + Select *pParent; /* Current UNION ALL term of the other query */ + Select *pSub; /* The inner query or "subquery" */ + Select *pSub1; /* Pointer to the rightmost select in sub-query */ + SrcList *pSrc; /* The FROM clause of the outer query */ + SrcList *pSubSrc; /* The FROM clause of the subquery */ + int iParent; /* VDBE cursor number of the pSub result set temp table */ + int iNewParent = -1;/* Replacement table for iParent */ + int isLeftJoin = 0; /* True if pSub is the right side of a LEFT JOIN */ + int i; /* Loop counter */ + Expr *pWhere; /* The WHERE clause */ + struct SrcList_item *pSubitem; /* The subquery */ + sqlite3 *db = pParse->db; + + /* Check to see if flattening is permitted. Return 0 if not. + */ + assert( p!=0 ); + assert( p->pPrior==0 ); + if( OptimizationDisabled(db, SQLITE_QueryFlattener) ) return 0; + pSrc = p->pSrc; + assert( pSrc && iFrom>=0 && iFromnSrc ); + pSubitem = &pSrc->a[iFrom]; + iParent = pSubitem->iCursor; + pSub = pSubitem->pSelect; + assert( pSub!=0 ); + +#ifndef SQLITE_OMIT_WINDOWFUNC + if( p->pWin || pSub->pWin ) return 0; /* Restriction (25) */ +#endif + + pSubSrc = pSub->pSrc; + assert( pSubSrc ); + /* Prior to version 3.1.2, when LIMIT and OFFSET had to be simple constants, + ** not arbitrary expressions, we allowed some combining of LIMIT and OFFSET + ** because they could be computed at compile-time. But when LIMIT and OFFSET + ** became arbitrary expressions, we were forced to add restrictions (13) + ** and (14). */ + if( pSub->pLimit && p->pLimit ) return 0; /* Restriction (13) */ + if( pSub->pLimit && pSub->pLimit->pRight ) return 0; /* Restriction (14) */ + if( (p->selFlags & SF_Compound)!=0 && pSub->pLimit ){ + return 0; /* Restriction (15) */ + } + if( pSubSrc->nSrc==0 ) return 0; /* Restriction (7) */ + if( pSub->selFlags & SF_Distinct ) return 0; /* Restriction (4) */ + if( pSub->pLimit && (pSrc->nSrc>1 || isAgg) ){ + return 0; /* Restrictions (8)(9) */ + } + if( p->pOrderBy && pSub->pOrderBy ){ + return 0; /* Restriction (11) */ + } + if( isAgg && pSub->pOrderBy ) return 0; /* Restriction (16) */ + if( pSub->pLimit && p->pWhere ) return 0; /* Restriction (19) */ + if( pSub->pLimit && (p->selFlags & SF_Distinct)!=0 ){ + return 0; /* Restriction (21) */ + } + if( pSub->selFlags & (SF_Recursive) ){ + return 0; /* Restrictions (22) */ + } + + /* + ** If the subquery is the right operand of a LEFT JOIN, then the + ** subquery may not be a join itself (3a). Example of why this is not + ** allowed: + ** + ** t1 LEFT OUTER JOIN (t2 JOIN t3) + ** + ** If we flatten the above, we would get + ** + ** (t1 LEFT OUTER JOIN t2) JOIN t3 + ** + ** which is not at all the same thing. + ** + ** If the subquery is the right operand of a LEFT JOIN, then the outer + ** query cannot be an aggregate. (3c) This is an artifact of the way + ** aggregates are processed - there is no mechanism to determine if + ** the LEFT JOIN table should be all-NULL. + ** + ** See also tickets #306, #350, and #3300. + */ + if( (pSubitem->fg.jointype & JT_OUTER)!=0 ){ + isLeftJoin = 1; + if( pSubSrc->nSrc>1 || isAgg || IsVirtual(pSubSrc->a[0].pTab) ){ + /* (3a) (3c) (3b) */ + return 0; + } + } +#ifdef SQLITE_EXTRA_IFNULLROW + else if( iFrom>0 && !isAgg ){ + /* Setting isLeftJoin to -1 causes OP_IfNullRow opcodes to be generated for + ** every reference to any result column from subquery in a join, even + ** though they are not necessary. This will stress-test the OP_IfNullRow + ** opcode. */ + isLeftJoin = -1; + } +#endif + + /* Restriction (17): If the sub-query is a compound SELECT, then it must + ** use only the UNION ALL operator. And none of the simple select queries + ** that make up the compound SELECT are allowed to be aggregate or distinct + ** queries. + */ + if( pSub->pPrior ){ + if( pSub->pOrderBy ){ + return 0; /* Restriction (20) */ + } + if( isAgg || (p->selFlags & SF_Distinct)!=0 || pSrc->nSrc!=1 ){ + return 0; /* (17d1), (17d2), or (17d3) */ + } + for(pSub1=pSub; pSub1; pSub1=pSub1->pPrior){ + testcase( (pSub1->selFlags & (SF_Distinct|SF_Aggregate))==SF_Distinct ); + testcase( (pSub1->selFlags & (SF_Distinct|SF_Aggregate))==SF_Aggregate ); + assert( pSub->pSrc!=0 ); + assert( pSub->pEList->nExpr==pSub1->pEList->nExpr ); + if( (pSub1->selFlags & (SF_Distinct|SF_Aggregate))!=0 /* (17b) */ + || (pSub1->pPrior && pSub1->op!=TK_ALL) /* (17a) */ + || pSub1->pSrc->nSrc<1 /* (17c) */ + ){ + return 0; + } + testcase( pSub1->pSrc->nSrc>1 ); + } + + /* Restriction (18). */ + if( p->pOrderBy ){ + int ii; + for(ii=0; iipOrderBy->nExpr; ii++){ + if( p->pOrderBy->a[ii].u.x.iOrderByCol==0 ) return 0; + } + } + } + + /* Ex-restriction (23): + ** The only way that the recursive part of a CTE can contain a compound + ** subquery is for the subquery to be one term of a join. But if the + ** subquery is a join, then the flattening has already been stopped by + ** restriction (17d3) + */ + assert( (p->selFlags & SF_Recursive)==0 || pSub->pPrior==0 ); + + /***** If we reach this point, flattening is permitted. *****/ + SELECTTRACE(1,pParse,p,("flatten %u.%p from term %d\n", + pSub->selId, pSub, iFrom)); + + /* Authorize the subquery */ + pParse->zAuthContext = pSubitem->zName; + TESTONLY(i =) sqlite3AuthCheck(pParse, SQLITE_SELECT, 0, 0, 0); + testcase( i==SQLITE_DENY ); + pParse->zAuthContext = zSavedAuthContext; + + /* If the sub-query is a compound SELECT statement, then (by restrictions + ** 17 and 18 above) it must be a UNION ALL and the parent query must + ** be of the form: + ** + ** SELECT FROM () + ** + ** followed by any ORDER BY, LIMIT and/or OFFSET clauses. This block + ** creates N-1 copies of the parent query without any ORDER BY, LIMIT or + ** OFFSET clauses and joins them to the left-hand-side of the original + ** using UNION ALL operators. In this case N is the number of simple + ** select statements in the compound sub-query. + ** + ** Example: + ** + ** SELECT a+1 FROM ( + ** SELECT x FROM tab + ** UNION ALL + ** SELECT y FROM tab + ** UNION ALL + ** SELECT abs(z*2) FROM tab2 + ** ) WHERE a!=5 ORDER BY 1 + ** + ** Transformed into: + ** + ** SELECT x+1 FROM tab WHERE x+1!=5 + ** UNION ALL + ** SELECT y+1 FROM tab WHERE y+1!=5 + ** UNION ALL + ** SELECT abs(z*2)+1 FROM tab2 WHERE abs(z*2)+1!=5 + ** ORDER BY 1 + ** + ** We call this the "compound-subquery flattening". + */ + for(pSub=pSub->pPrior; pSub; pSub=pSub->pPrior){ + Select *pNew; + ExprList *pOrderBy = p->pOrderBy; + Expr *pLimit = p->pLimit; + Select *pPrior = p->pPrior; + p->pOrderBy = 0; + p->pSrc = 0; + p->pPrior = 0; + p->pLimit = 0; + pNew = sqlite3SelectDup(db, p, 0); + p->pLimit = pLimit; + p->pOrderBy = pOrderBy; + p->pSrc = pSrc; + p->op = TK_ALL; + if( pNew==0 ){ + p->pPrior = pPrior; + }else{ + pNew->pPrior = pPrior; + if( pPrior ) pPrior->pNext = pNew; + pNew->pNext = p; + p->pPrior = pNew; + SELECTTRACE(2,pParse,p,("compound-subquery flattener" + " creates %u as peer\n",pNew->selId)); + } + if( db->mallocFailed ) return 1; + } + + /* Begin flattening the iFrom-th entry of the FROM clause + ** in the outer query. + */ + pSub = pSub1 = pSubitem->pSelect; + + /* Delete the transient table structure associated with the + ** subquery + */ + sqlite3DbFree(db, pSubitem->zDatabase); + sqlite3DbFree(db, pSubitem->zName); + sqlite3DbFree(db, pSubitem->zAlias); + pSubitem->zDatabase = 0; + pSubitem->zName = 0; + pSubitem->zAlias = 0; + pSubitem->pSelect = 0; + + /* Defer deleting the Table object associated with the + ** subquery until code generation is + ** complete, since there may still exist Expr.pTab entries that + ** refer to the subquery even after flattening. Ticket #3346. + ** + ** pSubitem->pTab is always non-NULL by test restrictions and tests above. + */ + if( ALWAYS(pSubitem->pTab!=0) ){ + Table *pTabToDel = pSubitem->pTab; + if( pTabToDel->nTabRef==1 ){ + Parse *pToplevel = sqlite3ParseToplevel(pParse); + pTabToDel->pNextZombie = pToplevel->pZombieTab; + pToplevel->pZombieTab = pTabToDel; + }else{ + pTabToDel->nTabRef--; + } + pSubitem->pTab = 0; + } + + /* The following loop runs once for each term in a compound-subquery + ** flattening (as described above). If we are doing a different kind + ** of flattening - a flattening other than a compound-subquery flattening - + ** then this loop only runs once. + ** + ** This loop moves all of the FROM elements of the subquery into the + ** the FROM clause of the outer query. Before doing this, remember + ** the cursor number for the original outer query FROM element in + ** iParent. The iParent cursor will never be used. Subsequent code + ** will scan expressions looking for iParent references and replace + ** those references with expressions that resolve to the subquery FROM + ** elements we are now copying in. + */ + for(pParent=p; pParent; pParent=pParent->pPrior, pSub=pSub->pPrior){ + int nSubSrc; + u8 jointype = 0; + assert( pSub!=0 ); + pSubSrc = pSub->pSrc; /* FROM clause of subquery */ + nSubSrc = pSubSrc->nSrc; /* Number of terms in subquery FROM clause */ + pSrc = pParent->pSrc; /* FROM clause of the outer query */ + + if( pSrc ){ + assert( pParent==p ); /* First time through the loop */ + jointype = pSubitem->fg.jointype; + }else{ + assert( pParent!=p ); /* 2nd and subsequent times through the loop */ + pSrc = sqlite3SrcListAppend(pParse, 0, 0, 0); + if( pSrc==0 ) break; + pParent->pSrc = pSrc; + } + + /* The subquery uses a single slot of the FROM clause of the outer + ** query. If the subquery has more than one element in its FROM clause, + ** then expand the outer query to make space for it to hold all elements + ** of the subquery. + ** + ** Example: + ** + ** SELECT * FROM tabA, (SELECT * FROM sub1, sub2), tabB; + ** + ** The outer query has 3 slots in its FROM clause. One slot of the + ** outer query (the middle slot) is used by the subquery. The next + ** block of code will expand the outer query FROM clause to 4 slots. + ** The middle slot is expanded to two slots in order to make space + ** for the two elements in the FROM clause of the subquery. + */ + if( nSubSrc>1 ){ + pSrc = sqlite3SrcListEnlarge(pParse, pSrc, nSubSrc-1,iFrom+1); + if( pSrc==0 ) break; + pParent->pSrc = pSrc; + } + + /* Transfer the FROM clause terms from the subquery into the + ** outer query. + */ + for(i=0; ia[i+iFrom].pUsing); + assert( pSrc->a[i+iFrom].fg.isTabFunc==0 ); + pSrc->a[i+iFrom] = pSubSrc->a[i]; + iNewParent = pSubSrc->a[i].iCursor; + memset(&pSubSrc->a[i], 0, sizeof(pSubSrc->a[i])); + } + pSrc->a[iFrom].fg.jointype = jointype; + + /* Now begin substituting subquery result set expressions for + ** references to the iParent in the outer query. + ** + ** Example: + ** + ** SELECT a+5, b*10 FROM (SELECT x*3 AS a, y+10 AS b FROM t1) WHERE a>b; + ** \ \_____________ subquery __________/ / + ** \_____________________ outer query ______________________________/ + ** + ** We look at every expression in the outer query and every place we see + ** "a" we substitute "x*3" and every place we see "b" we substitute "y+10". + */ + if( pSub->pOrderBy ){ + /* At this point, any non-zero iOrderByCol values indicate that the + ** ORDER BY column expression is identical to the iOrderByCol'th + ** expression returned by SELECT statement pSub. Since these values + ** do not necessarily correspond to columns in SELECT statement pParent, + ** zero them before transfering the ORDER BY clause. + ** + ** Not doing this may cause an error if a subsequent call to this + ** function attempts to flatten a compound sub-query into pParent + ** (the only way this can happen is if the compound sub-query is + ** currently part of pSub->pSrc). See ticket [d11a6e908f]. */ + ExprList *pOrderBy = pSub->pOrderBy; + for(i=0; inExpr; i++){ + pOrderBy->a[i].u.x.iOrderByCol = 0; + } + assert( pParent->pOrderBy==0 ); + pParent->pOrderBy = pOrderBy; + pSub->pOrderBy = 0; + } + pWhere = pSub->pWhere; + pSub->pWhere = 0; + if( isLeftJoin>0 ){ + setJoinExpr(pWhere, iNewParent); + } + pParent->pWhere = sqlite3ExprAnd(pParse, pWhere, pParent->pWhere); + if( db->mallocFailed==0 ){ + SubstContext x; + x.pParse = pParse; + x.iTable = iParent; + x.iNewTable = iNewParent; + x.isLeftJoin = isLeftJoin; + x.pEList = pSub->pEList; + substSelect(&x, pParent, 0); + } + + /* The flattened query is a compound if either the inner or the + ** outer query is a compound. */ + pParent->selFlags |= pSub->selFlags & SF_Compound; + assert( (pSub->selFlags & SF_Distinct)==0 ); /* restriction (17b) */ + + /* + ** SELECT ... FROM (SELECT ... LIMIT a OFFSET b) LIMIT x OFFSET y; + ** + ** One is tempted to try to add a and b to combine the limits. But this + ** does not work if either limit is negative. + */ + if( pSub->pLimit ){ + pParent->pLimit = pSub->pLimit; + pSub->pLimit = 0; + } + } + + /* Finially, delete what is left of the subquery and return + ** success. + */ + sqlite3SelectDelete(db, pSub1); + +#if SELECTTRACE_ENABLED + if( sqlite3SelectTrace & 0x100 ){ + SELECTTRACE(0x100,pParse,p,("After flattening:\n")); + sqlite3TreeViewSelect(0, p, 0); + } +#endif + + return 1; +} +#endif /* !defined(SQLITE_OMIT_SUBQUERY) || !defined(SQLITE_OMIT_VIEW) */ + +/* +** A structure to keep track of all of the column values that are fixed to +** a known value due to WHERE clause constraints of the form COLUMN=VALUE. +*/ +typedef struct WhereConst WhereConst; +struct WhereConst { + Parse *pParse; /* Parsing context */ + int nConst; /* Number for COLUMN=CONSTANT terms */ + int nChng; /* Number of times a constant is propagated */ + Expr **apExpr; /* [i*2] is COLUMN and [i*2+1] is VALUE */ +}; + +/* +** Add a new entry to the pConst object. Except, do not add duplicate +** pColumn entires. +*/ +static void constInsert( + WhereConst *pConst, /* The WhereConst into which we are inserting */ + Expr *pColumn, /* The COLUMN part of the constraint */ + Expr *pValue /* The VALUE part of the constraint */ +){ + int i; + assert( pColumn->op==TK_COLUMN ); + + /* 2018-10-25 ticket [cf5ed20f] + ** Make sure the same pColumn is not inserted more than once */ + for(i=0; inConst; i++){ + const Expr *pExpr = pConst->apExpr[i*2]; + assert( pExpr->op==TK_COLUMN ); + if( pExpr->iTable==pColumn->iTable + && pExpr->iColumn==pColumn->iColumn + ){ + return; /* Already present. Return without doing anything. */ + } + } + + pConst->nConst++; + pConst->apExpr = sqlite3DbReallocOrFree(pConst->pParse->db, pConst->apExpr, + pConst->nConst*2*sizeof(Expr*)); + if( pConst->apExpr==0 ){ + pConst->nConst = 0; + }else{ + if( ExprHasProperty(pValue, EP_FixedCol) ) pValue = pValue->pLeft; + pConst->apExpr[pConst->nConst*2-2] = pColumn; + pConst->apExpr[pConst->nConst*2-1] = pValue; + } +} + +/* +** Find all terms of COLUMN=VALUE or VALUE=COLUMN in pExpr where VALUE +** is a constant expression and where the term must be true because it +** is part of the AND-connected terms of the expression. For each term +** found, add it to the pConst structure. +*/ +static void findConstInWhere(WhereConst *pConst, Expr *pExpr){ + Expr *pRight, *pLeft; + if( pExpr==0 ) return; + if( ExprHasProperty(pExpr, EP_FromJoin) ) return; + if( pExpr->op==TK_AND ){ + findConstInWhere(pConst, pExpr->pRight); + findConstInWhere(pConst, pExpr->pLeft); + return; + } + if( pExpr->op!=TK_EQ ) return; + pRight = pExpr->pRight; + pLeft = pExpr->pLeft; + assert( pRight!=0 ); + assert( pLeft!=0 ); + if( pRight->op==TK_COLUMN + && !ExprHasProperty(pRight, EP_FixedCol) + && sqlite3ExprIsConstant(pLeft) + && sqlite3IsBinary(sqlite3BinaryCompareCollSeq(pConst->pParse,pLeft,pRight)) + ){ + constInsert(pConst, pRight, pLeft); + }else + if( pLeft->op==TK_COLUMN + && !ExprHasProperty(pLeft, EP_FixedCol) + && sqlite3ExprIsConstant(pRight) + && sqlite3IsBinary(sqlite3BinaryCompareCollSeq(pConst->pParse,pLeft,pRight)) + ){ + constInsert(pConst, pLeft, pRight); + } +} + +/* +** This is a Walker expression callback. pExpr is a candidate expression +** to be replaced by a value. If pExpr is equivalent to one of the +** columns named in pWalker->u.pConst, then overwrite it with its +** corresponding value. +*/ +static int propagateConstantExprRewrite(Walker *pWalker, Expr *pExpr){ + int i; + WhereConst *pConst; + if( pExpr->op!=TK_COLUMN ) return WRC_Continue; + if( ExprHasProperty(pExpr, EP_FixedCol) ) return WRC_Continue; + pConst = pWalker->u.pConst; + for(i=0; inConst; i++){ + Expr *pColumn = pConst->apExpr[i*2]; + if( pColumn==pExpr ) continue; + if( pColumn->iTable!=pExpr->iTable ) continue; + if( pColumn->iColumn!=pExpr->iColumn ) continue; + /* A match is found. Add the EP_FixedCol property */ + pConst->nChng++; + ExprClearProperty(pExpr, EP_Leaf); + ExprSetProperty(pExpr, EP_FixedCol); + assert( pExpr->pLeft==0 ); + pExpr->pLeft = sqlite3ExprDup(pConst->pParse->db, pConst->apExpr[i*2+1], 0); + break; + } + return WRC_Prune; +} + +/* +** The WHERE-clause constant propagation optimization. +** +** If the WHERE clause contains terms of the form COLUMN=CONSTANT or +** CONSTANT=COLUMN that must be tree (in other words, if the terms top-level +** AND-connected terms that are not part of a ON clause from a LEFT JOIN) +** then throughout the query replace all other occurrences of COLUMN +** with CONSTANT within the WHERE clause. +** +** For example, the query: +** +** SELECT * FROM t1, t2, t3 WHERE t1.a=39 AND t2.b=t1.a AND t3.c=t2.b +** +** Is transformed into +** +** SELECT * FROM t1, t2, t3 WHERE t1.a=39 AND t2.b=39 AND t3.c=39 +** +** Return true if any transformations where made and false if not. +** +** Implementation note: Constant propagation is tricky due to affinity +** and collating sequence interactions. Consider this example: +** +** CREATE TABLE t1(a INT,b TEXT); +** INSERT INTO t1 VALUES(123,'0123'); +** SELECT * FROM t1 WHERE a=123 AND b=a; +** SELECT * FROM t1 WHERE a=123 AND b=123; +** +** The two SELECT statements above should return different answers. b=a +** is alway true because the comparison uses numeric affinity, but b=123 +** is false because it uses text affinity and '0123' is not the same as '123'. +** To work around this, the expression tree is not actually changed from +** "b=a" to "b=123" but rather the "a" in "b=a" is tagged with EP_FixedCol +** and the "123" value is hung off of the pLeft pointer. Code generator +** routines know to generate the constant "123" instead of looking up the +** column value. Also, to avoid collation problems, this optimization is +** only attempted if the "a=123" term uses the default BINARY collation. +*/ +static int propagateConstants( + Parse *pParse, /* The parsing context */ + Select *p /* The query in which to propagate constants */ +){ + WhereConst x; + Walker w; + int nChng = 0; + x.pParse = pParse; + do{ + x.nConst = 0; + x.nChng = 0; + x.apExpr = 0; + findConstInWhere(&x, p->pWhere); + if( x.nConst ){ + memset(&w, 0, sizeof(w)); + w.pParse = pParse; + w.xExprCallback = propagateConstantExprRewrite; + w.xSelectCallback = sqlite3SelectWalkNoop; + w.xSelectCallback2 = 0; + w.walkerDepth = 0; + w.u.pConst = &x; + sqlite3WalkExpr(&w, p->pWhere); + sqlite3DbFree(x.pParse->db, x.apExpr); + nChng += x.nChng; + } + }while( x.nChng ); + return nChng; +} + +#if !defined(SQLITE_OMIT_SUBQUERY) || !defined(SQLITE_OMIT_VIEW) +/* +** Make copies of relevant WHERE clause terms of the outer query into +** the WHERE clause of subquery. Example: +** +** SELECT * FROM (SELECT a AS x, c-d AS y FROM t1) WHERE x=5 AND y=10; +** +** Transformed into: +** +** SELECT * FROM (SELECT a AS x, c-d AS y FROM t1 WHERE a=5 AND c-d=10) +** WHERE x=5 AND y=10; +** +** The hope is that the terms added to the inner query will make it more +** efficient. +** +** Do not attempt this optimization if: +** +** (1) (** This restriction was removed on 2017-09-29. We used to +** disallow this optimization for aggregate subqueries, but now +** it is allowed by putting the extra terms on the HAVING clause. +** The added HAVING clause is pointless if the subquery lacks +** a GROUP BY clause. But such a HAVING clause is also harmless +** so there does not appear to be any reason to add extra logic +** to suppress it. **) +** +** (2) The inner query is the recursive part of a common table expression. +** +** (3) The inner query has a LIMIT clause (since the changes to the WHERE +** clause would change the meaning of the LIMIT). +** +** (4) The inner query is the right operand of a LEFT JOIN and the +** expression to be pushed down does not come from the ON clause +** on that LEFT JOIN. +** +** (5) The WHERE clause expression originates in the ON or USING clause +** of a LEFT JOIN where iCursor is not the right-hand table of that +** left join. An example: +** +** SELECT * +** FROM (SELECT 1 AS a1 UNION ALL SELECT 2) AS aa +** JOIN (SELECT 1 AS b2 UNION ALL SELECT 2) AS bb ON (a1=b2) +** LEFT JOIN (SELECT 8 AS c3 UNION ALL SELECT 9) AS cc ON (b2=2); +** +** The correct answer is three rows: (1,1,NULL),(2,2,8),(2,2,9). +** But if the (b2=2) term were to be pushed down into the bb subquery, +** then the (1,1,NULL) row would be suppressed. +** +** (6) The inner query features one or more window-functions (since +** changes to the WHERE clause of the inner query could change the +** window over which window functions are calculated). +** +** Return 0 if no changes are made and non-zero if one or more WHERE clause +** terms are duplicated into the subquery. +*/ +static int pushDownWhereTerms( + Parse *pParse, /* Parse context (for malloc() and error reporting) */ + Select *pSubq, /* The subquery whose WHERE clause is to be augmented */ + Expr *pWhere, /* The WHERE clause of the outer query */ + int iCursor, /* Cursor number of the subquery */ + int isLeftJoin /* True if pSubq is the right term of a LEFT JOIN */ +){ + Expr *pNew; + int nChng = 0; + if( pWhere==0 ) return 0; + if( pSubq->selFlags & SF_Recursive ) return 0; /* restriction (2) */ + +#ifndef SQLITE_OMIT_WINDOWFUNC + if( pSubq->pWin ) return 0; /* restriction (6) */ +#endif + +#ifdef SQLITE_DEBUG + /* Only the first term of a compound can have a WITH clause. But make + ** sure no other terms are marked SF_Recursive in case something changes + ** in the future. + */ + { + Select *pX; + for(pX=pSubq; pX; pX=pX->pPrior){ + assert( (pX->selFlags & (SF_Recursive))==0 ); + } + } +#endif + + if( pSubq->pLimit!=0 ){ + return 0; /* restriction (3) */ + } + while( pWhere->op==TK_AND ){ + nChng += pushDownWhereTerms(pParse, pSubq, pWhere->pRight, + iCursor, isLeftJoin); + pWhere = pWhere->pLeft; + } + if( isLeftJoin + && (ExprHasProperty(pWhere,EP_FromJoin)==0 + || pWhere->iRightJoinTable!=iCursor) + ){ + return 0; /* restriction (4) */ + } + if( ExprHasProperty(pWhere,EP_FromJoin) && pWhere->iRightJoinTable!=iCursor ){ + return 0; /* restriction (5) */ + } + if( sqlite3ExprIsTableConstant(pWhere, iCursor) ){ + nChng++; + while( pSubq ){ + SubstContext x; + pNew = sqlite3ExprDup(pParse->db, pWhere, 0); + unsetJoinExpr(pNew, -1); + x.pParse = pParse; + x.iTable = iCursor; + x.iNewTable = iCursor; + x.isLeftJoin = 0; + x.pEList = pSubq->pEList; + pNew = substExpr(&x, pNew); + if( pSubq->selFlags & SF_Aggregate ){ + pSubq->pHaving = sqlite3ExprAnd(pParse, pSubq->pHaving, pNew); + }else{ + pSubq->pWhere = sqlite3ExprAnd(pParse, pSubq->pWhere, pNew); + } + pSubq = pSubq->pPrior; + } + } + return nChng; +} +#endif /* !defined(SQLITE_OMIT_SUBQUERY) || !defined(SQLITE_OMIT_VIEW) */ + +/* +** The pFunc is the only aggregate function in the query. Check to see +** if the query is a candidate for the min/max optimization. +** +** If the query is a candidate for the min/max optimization, then set +** *ppMinMax to be an ORDER BY clause to be used for the optimization +** and return either WHERE_ORDERBY_MIN or WHERE_ORDERBY_MAX depending on +** whether pFunc is a min() or max() function. +** +** If the query is not a candidate for the min/max optimization, return +** WHERE_ORDERBY_NORMAL (which must be zero). +** +** This routine must be called after aggregate functions have been +** located but before their arguments have been subjected to aggregate +** analysis. +*/ +static u8 minMaxQuery(sqlite3 *db, Expr *pFunc, ExprList **ppMinMax){ + int eRet = WHERE_ORDERBY_NORMAL; /* Return value */ + ExprList *pEList = pFunc->x.pList; /* Arguments to agg function */ + const char *zFunc; /* Name of aggregate function pFunc */ + ExprList *pOrderBy; + u8 sortFlags; + + assert( *ppMinMax==0 ); + assert( pFunc->op==TK_AGG_FUNCTION ); + assert( !IsWindowFunc(pFunc) ); + if( pEList==0 || pEList->nExpr!=1 || ExprHasProperty(pFunc, EP_WinFunc) ){ + return eRet; + } + zFunc = pFunc->u.zToken; + if( sqlite3StrICmp(zFunc, "min")==0 ){ + eRet = WHERE_ORDERBY_MIN; + sortFlags = KEYINFO_ORDER_BIGNULL; + }else if( sqlite3StrICmp(zFunc, "max")==0 ){ + eRet = WHERE_ORDERBY_MAX; + sortFlags = KEYINFO_ORDER_DESC; + }else{ + return eRet; + } + *ppMinMax = pOrderBy = sqlite3ExprListDup(db, pEList, 0); + assert( pOrderBy!=0 || db->mallocFailed ); + if( pOrderBy ) pOrderBy->a[0].sortFlags = sortFlags; + return eRet; +} + +/* +** The select statement passed as the first argument is an aggregate query. +** The second argument is the associated aggregate-info object. This +** function tests if the SELECT is of the form: +** +** SELECT count(*) FROM +** +** where table is a database table, not a sub-select or view. If the query +** does match this pattern, then a pointer to the Table object representing +** is returned. Otherwise, 0 is returned. +*/ +static Table *isSimpleCount(Select *p, AggInfo *pAggInfo){ + Table *pTab; + Expr *pExpr; + + assert( !p->pGroupBy ); + + if( p->pWhere || p->pEList->nExpr!=1 + || p->pSrc->nSrc!=1 || p->pSrc->a[0].pSelect + ){ + return 0; + } + pTab = p->pSrc->a[0].pTab; + pExpr = p->pEList->a[0].pExpr; + assert( pTab && !pTab->pSelect && pExpr ); + + if( IsVirtual(pTab) ) return 0; + if( pExpr->op!=TK_AGG_FUNCTION ) return 0; + if( NEVER(pAggInfo->nFunc==0) ) return 0; + if( (pAggInfo->aFunc[0].pFunc->funcFlags&SQLITE_FUNC_COUNT)==0 ) return 0; + if( ExprHasProperty(pExpr, EP_Distinct|EP_WinFunc) ) return 0; + + return pTab; +} + +/* +** If the source-list item passed as an argument was augmented with an +** INDEXED BY clause, then try to locate the specified index. If there +** was such a clause and the named index cannot be found, return +** SQLITE_ERROR and leave an error in pParse. Otherwise, populate +** pFrom->pIndex and return SQLITE_OK. +*/ +SQLITE_PRIVATE int sqlite3IndexedByLookup(Parse *pParse, struct SrcList_item *pFrom){ + if( pFrom->pTab && pFrom->fg.isIndexedBy ){ + Table *pTab = pFrom->pTab; + char *zIndexedBy = pFrom->u1.zIndexedBy; + Index *pIdx; + for(pIdx=pTab->pIndex; + pIdx && sqlite3StrICmp(pIdx->zName, zIndexedBy); + pIdx=pIdx->pNext + ); + if( !pIdx ){ + sqlite3ErrorMsg(pParse, "no such index: %s", zIndexedBy, 0); + pParse->checkSchema = 1; + return SQLITE_ERROR; + } + pFrom->pIBIndex = pIdx; + } + return SQLITE_OK; +} +/* +** Detect compound SELECT statements that use an ORDER BY clause with +** an alternative collating sequence. +** +** SELECT ... FROM t1 EXCEPT SELECT ... FROM t2 ORDER BY .. COLLATE ... +** +** These are rewritten as a subquery: +** +** SELECT * FROM (SELECT ... FROM t1 EXCEPT SELECT ... FROM t2) +** ORDER BY ... COLLATE ... +** +** This transformation is necessary because the multiSelectOrderBy() routine +** above that generates the code for a compound SELECT with an ORDER BY clause +** uses a merge algorithm that requires the same collating sequence on the +** result columns as on the ORDER BY clause. See ticket +** http://www.sqlite.org/src/info/6709574d2a +** +** This transformation is only needed for EXCEPT, INTERSECT, and UNION. +** The UNION ALL operator works fine with multiSelectOrderBy() even when +** there are COLLATE terms in the ORDER BY. +*/ +static int convertCompoundSelectToSubquery(Walker *pWalker, Select *p){ + int i; + Select *pNew; + Select *pX; + sqlite3 *db; + struct ExprList_item *a; + SrcList *pNewSrc; + Parse *pParse; + Token dummy; + + if( p->pPrior==0 ) return WRC_Continue; + if( p->pOrderBy==0 ) return WRC_Continue; + for(pX=p; pX && (pX->op==TK_ALL || pX->op==TK_SELECT); pX=pX->pPrior){} + if( pX==0 ) return WRC_Continue; + a = p->pOrderBy->a; + for(i=p->pOrderBy->nExpr-1; i>=0; i--){ + if( a[i].pExpr->flags & EP_Collate ) break; + } + if( i<0 ) return WRC_Continue; + + /* If we reach this point, that means the transformation is required. */ + + pParse = pWalker->pParse; + db = pParse->db; + pNew = sqlite3DbMallocZero(db, sizeof(*pNew) ); + if( pNew==0 ) return WRC_Abort; + memset(&dummy, 0, sizeof(dummy)); + pNewSrc = sqlite3SrcListAppendFromTerm(pParse,0,0,0,&dummy,pNew,0,0); + if( pNewSrc==0 ) return WRC_Abort; + *pNew = *p; + p->pSrc = pNewSrc; + p->pEList = sqlite3ExprListAppend(pParse, 0, sqlite3Expr(db, TK_ASTERISK, 0)); + p->op = TK_SELECT; + p->pWhere = 0; + pNew->pGroupBy = 0; + pNew->pHaving = 0; + pNew->pOrderBy = 0; + p->pPrior = 0; + p->pNext = 0; + p->pWith = 0; + p->selFlags &= ~SF_Compound; + assert( (p->selFlags & SF_Converted)==0 ); + p->selFlags |= SF_Converted; + assert( pNew->pPrior!=0 ); + pNew->pPrior->pNext = pNew; + pNew->pLimit = 0; + return WRC_Continue; +} + +/* +** Check to see if the FROM clause term pFrom has table-valued function +** arguments. If it does, leave an error message in pParse and return +** non-zero, since pFrom is not allowed to be a table-valued function. +*/ +static int cannotBeFunction(Parse *pParse, struct SrcList_item *pFrom){ + if( pFrom->fg.isTabFunc ){ + sqlite3ErrorMsg(pParse, "'%s' is not a function", pFrom->zName); + return 1; + } + return 0; +} + +#ifndef SQLITE_OMIT_CTE +/* +** Argument pWith (which may be NULL) points to a linked list of nested +** WITH contexts, from inner to outermost. If the table identified by +** FROM clause element pItem is really a common-table-expression (CTE) +** then return a pointer to the CTE definition for that table. Otherwise +** return NULL. +** +** If a non-NULL value is returned, set *ppContext to point to the With +** object that the returned CTE belongs to. +*/ +static struct Cte *searchWith( + With *pWith, /* Current innermost WITH clause */ + struct SrcList_item *pItem, /* FROM clause element to resolve */ + With **ppContext /* OUT: WITH clause return value belongs to */ +){ + const char *zName; + if( pItem->zDatabase==0 && (zName = pItem->zName)!=0 ){ + With *p; + for(p=pWith; p; p=p->pOuter){ + int i; + for(i=0; inCte; i++){ + if( sqlite3StrICmp(zName, p->a[i].zName)==0 ){ + *ppContext = p; + return &p->a[i]; + } + } + } + } + return 0; +} + +/* The code generator maintains a stack of active WITH clauses +** with the inner-most WITH clause being at the top of the stack. +** +** This routine pushes the WITH clause passed as the second argument +** onto the top of the stack. If argument bFree is true, then this +** WITH clause will never be popped from the stack. In this case it +** should be freed along with the Parse object. In other cases, when +** bFree==0, the With object will be freed along with the SELECT +** statement with which it is associated. +*/ +SQLITE_PRIVATE void sqlite3WithPush(Parse *pParse, With *pWith, u8 bFree){ + assert( bFree==0 || (pParse->pWith==0 && pParse->pWithToFree==0) ); + if( pWith ){ + assert( pParse->pWith!=pWith ); + pWith->pOuter = pParse->pWith; + pParse->pWith = pWith; + if( bFree ) pParse->pWithToFree = pWith; + } +} + +/* +** This function checks if argument pFrom refers to a CTE declared by +** a WITH clause on the stack currently maintained by the parser. And, +** if currently processing a CTE expression, if it is a recursive +** reference to the current CTE. +** +** If pFrom falls into either of the two categories above, pFrom->pTab +** and other fields are populated accordingly. The caller should check +** (pFrom->pTab!=0) to determine whether or not a successful match +** was found. +** +** Whether or not a match is found, SQLITE_OK is returned if no error +** occurs. If an error does occur, an error message is stored in the +** parser and some error code other than SQLITE_OK returned. +*/ +static int withExpand( + Walker *pWalker, + struct SrcList_item *pFrom +){ + Parse *pParse = pWalker->pParse; + sqlite3 *db = pParse->db; + struct Cte *pCte; /* Matched CTE (or NULL if no match) */ + With *pWith; /* WITH clause that pCte belongs to */ + + assert( pFrom->pTab==0 ); + + pCte = searchWith(pParse->pWith, pFrom, &pWith); + if( pCte ){ + Table *pTab; + ExprList *pEList; + Select *pSel; + Select *pLeft; /* Left-most SELECT statement */ + int bMayRecursive; /* True if compound joined by UNION [ALL] */ + With *pSavedWith; /* Initial value of pParse->pWith */ + + /* If pCte->zCteErr is non-NULL at this point, then this is an illegal + ** recursive reference to CTE pCte. Leave an error in pParse and return + ** early. If pCte->zCteErr is NULL, then this is not a recursive reference. + ** In this case, proceed. */ + if( pCte->zCteErr ){ + sqlite3ErrorMsg(pParse, pCte->zCteErr, pCte->zName); + return SQLITE_ERROR; + } + if( cannotBeFunction(pParse, pFrom) ) return SQLITE_ERROR; + + assert( pFrom->pTab==0 ); + pFrom->pTab = pTab = sqlite3DbMallocZero(db, sizeof(Table)); + if( pTab==0 ) return WRC_Abort; + pTab->nTabRef = 1; + pTab->zName = sqlite3DbStrDup(db, pCte->zName); + pTab->iPKey = -1; + pTab->nRowLogEst = 200; assert( 200==sqlite3LogEst(1048576) ); + pTab->tabFlags |= TF_Ephemeral | TF_NoVisibleRowid; + pFrom->pSelect = sqlite3SelectDup(db, pCte->pSelect, 0); + if( db->mallocFailed ) return SQLITE_NOMEM_BKPT; + assert( pFrom->pSelect ); + + /* Check if this is a recursive CTE. */ + pSel = pFrom->pSelect; + bMayRecursive = ( pSel->op==TK_ALL || pSel->op==TK_UNION ); + if( bMayRecursive ){ + int i; + SrcList *pSrc = pFrom->pSelect->pSrc; + for(i=0; inSrc; i++){ + struct SrcList_item *pItem = &pSrc->a[i]; + if( pItem->zDatabase==0 + && pItem->zName!=0 + && 0==sqlite3StrICmp(pItem->zName, pCte->zName) + ){ + pItem->pTab = pTab; + pItem->fg.isRecursive = 1; + pTab->nTabRef++; + pSel->selFlags |= SF_Recursive; + } + } + } + + /* Only one recursive reference is permitted. */ + if( pTab->nTabRef>2 ){ + sqlite3ErrorMsg( + pParse, "multiple references to recursive table: %s", pCte->zName + ); + return SQLITE_ERROR; + } + assert( pTab->nTabRef==1 || + ((pSel->selFlags&SF_Recursive) && pTab->nTabRef==2 )); + + pCte->zCteErr = "circular reference: %s"; + pSavedWith = pParse->pWith; + pParse->pWith = pWith; + if( bMayRecursive ){ + Select *pPrior = pSel->pPrior; + assert( pPrior->pWith==0 ); + pPrior->pWith = pSel->pWith; + sqlite3WalkSelect(pWalker, pPrior); + pPrior->pWith = 0; + }else{ + sqlite3WalkSelect(pWalker, pSel); + } + pParse->pWith = pWith; + + for(pLeft=pSel; pLeft->pPrior; pLeft=pLeft->pPrior); + pEList = pLeft->pEList; + if( pCte->pCols ){ + if( pEList && pEList->nExpr!=pCte->pCols->nExpr ){ + sqlite3ErrorMsg(pParse, "table %s has %d values for %d columns", + pCte->zName, pEList->nExpr, pCte->pCols->nExpr + ); + pParse->pWith = pSavedWith; + return SQLITE_ERROR; + } + pEList = pCte->pCols; + } + + sqlite3ColumnsFromExprList(pParse, pEList, &pTab->nCol, &pTab->aCol); + if( bMayRecursive ){ + if( pSel->selFlags & SF_Recursive ){ + pCte->zCteErr = "multiple recursive references: %s"; + }else{ + pCte->zCteErr = "recursive reference in a subquery: %s"; + } + sqlite3WalkSelect(pWalker, pSel); + } + pCte->zCteErr = 0; + pParse->pWith = pSavedWith; + } + + return SQLITE_OK; +} +#endif + +#ifndef SQLITE_OMIT_CTE +/* +** If the SELECT passed as the second argument has an associated WITH +** clause, pop it from the stack stored as part of the Parse object. +** +** This function is used as the xSelectCallback2() callback by +** sqlite3SelectExpand() when walking a SELECT tree to resolve table +** names and other FROM clause elements. +*/ +static void selectPopWith(Walker *pWalker, Select *p){ + Parse *pParse = pWalker->pParse; + if( OK_IF_ALWAYS_TRUE(pParse->pWith) && p->pPrior==0 ){ + With *pWith = findRightmost(p)->pWith; + if( pWith!=0 ){ + assert( pParse->pWith==pWith ); + pParse->pWith = pWith->pOuter; + } + } +} +#else +#define selectPopWith 0 +#endif + +/* +** The SrcList_item structure passed as the second argument represents a +** sub-query in the FROM clause of a SELECT statement. This function +** allocates and populates the SrcList_item.pTab object. If successful, +** SQLITE_OK is returned. Otherwise, if an OOM error is encountered, +** SQLITE_NOMEM. +*/ +SQLITE_PRIVATE int sqlite3ExpandSubquery(Parse *pParse, struct SrcList_item *pFrom){ + Select *pSel = pFrom->pSelect; + Table *pTab; + + assert( pSel ); + pFrom->pTab = pTab = sqlite3DbMallocZero(pParse->db, sizeof(Table)); + if( pTab==0 ) return SQLITE_NOMEM; + pTab->nTabRef = 1; + if( pFrom->zAlias ){ + pTab->zName = sqlite3DbStrDup(pParse->db, pFrom->zAlias); + }else{ + pTab->zName = sqlite3MPrintf(pParse->db, "subquery_%u", pSel->selId); + } + while( pSel->pPrior ){ pSel = pSel->pPrior; } + sqlite3ColumnsFromExprList(pParse, pSel->pEList,&pTab->nCol,&pTab->aCol); + pTab->iPKey = -1; + pTab->nRowLogEst = 200; assert( 200==sqlite3LogEst(1048576) ); + pTab->tabFlags |= TF_Ephemeral; + + return pParse->nErr ? SQLITE_ERROR : SQLITE_OK; +} + +/* +** This routine is a Walker callback for "expanding" a SELECT statement. +** "Expanding" means to do the following: +** +** (1) Make sure VDBE cursor numbers have been assigned to every +** element of the FROM clause. +** +** (2) Fill in the pTabList->a[].pTab fields in the SrcList that +** defines FROM clause. When views appear in the FROM clause, +** fill pTabList->a[].pSelect with a copy of the SELECT statement +** that implements the view. A copy is made of the view's SELECT +** statement so that we can freely modify or delete that statement +** without worrying about messing up the persistent representation +** of the view. +** +** (3) Add terms to the WHERE clause to accommodate the NATURAL keyword +** on joins and the ON and USING clause of joins. +** +** (4) Scan the list of columns in the result set (pEList) looking +** for instances of the "*" operator or the TABLE.* operator. +** If found, expand each "*" to be every column in every table +** and TABLE.* to be every column in TABLE. +** +*/ +static int selectExpander(Walker *pWalker, Select *p){ + Parse *pParse = pWalker->pParse; + int i, j, k; + SrcList *pTabList; + ExprList *pEList; + struct SrcList_item *pFrom; + sqlite3 *db = pParse->db; + Expr *pE, *pRight, *pExpr; + u16 selFlags = p->selFlags; + u32 elistFlags = 0; + + p->selFlags |= SF_Expanded; + if( db->mallocFailed ){ + return WRC_Abort; + } + assert( p->pSrc!=0 ); + if( (selFlags & SF_Expanded)!=0 ){ + return WRC_Prune; + } + if( pWalker->eCode ){ + /* Renumber selId because it has been copied from a view */ + p->selId = ++pParse->nSelect; + } + pTabList = p->pSrc; + pEList = p->pEList; + sqlite3WithPush(pParse, p->pWith, 0); + + /* Make sure cursor numbers have been assigned to all entries in + ** the FROM clause of the SELECT statement. + */ + sqlite3SrcListAssignCursors(pParse, pTabList); + + /* Look up every table named in the FROM clause of the select. If + ** an entry of the FROM clause is a subquery instead of a table or view, + ** then create a transient table structure to describe the subquery. + */ + for(i=0, pFrom=pTabList->a; inSrc; i++, pFrom++){ + Table *pTab; + assert( pFrom->fg.isRecursive==0 || pFrom->pTab!=0 ); + if( pFrom->fg.isRecursive ) continue; + assert( pFrom->pTab==0 ); +#ifndef SQLITE_OMIT_CTE + if( withExpand(pWalker, pFrom) ) return WRC_Abort; + if( pFrom->pTab ) {} else +#endif + if( pFrom->zName==0 ){ +#ifndef SQLITE_OMIT_SUBQUERY + Select *pSel = pFrom->pSelect; + /* A sub-query in the FROM clause of a SELECT */ + assert( pSel!=0 ); + assert( pFrom->pTab==0 ); + if( sqlite3WalkSelect(pWalker, pSel) ) return WRC_Abort; + if( sqlite3ExpandSubquery(pParse, pFrom) ) return WRC_Abort; +#endif + }else{ + /* An ordinary table or view name in the FROM clause */ + assert( pFrom->pTab==0 ); + pFrom->pTab = pTab = sqlite3LocateTableItem(pParse, 0, pFrom); + if( pTab==0 ) return WRC_Abort; + if( pTab->nTabRef>=0xffff ){ + sqlite3ErrorMsg(pParse, "too many references to \"%s\": max 65535", + pTab->zName); + pFrom->pTab = 0; + return WRC_Abort; + } + pTab->nTabRef++; + if( !IsVirtual(pTab) && cannotBeFunction(pParse, pFrom) ){ + return WRC_Abort; + } +#if !defined(SQLITE_OMIT_VIEW) || !defined (SQLITE_OMIT_VIRTUALTABLE) + if( IsVirtual(pTab) || pTab->pSelect ){ + i16 nCol; + u8 eCodeOrig = pWalker->eCode; + if( sqlite3ViewGetColumnNames(pParse, pTab) ) return WRC_Abort; + assert( pFrom->pSelect==0 ); + if( pTab->pSelect && (db->flags & SQLITE_EnableView)==0 ){ + sqlite3ErrorMsg(pParse, "access to view \"%s\" prohibited", + pTab->zName); + } + pFrom->pSelect = sqlite3SelectDup(db, pTab->pSelect, 0); + nCol = pTab->nCol; + pTab->nCol = -1; + pWalker->eCode = 1; /* Turn on Select.selId renumbering */ + sqlite3WalkSelect(pWalker, pFrom->pSelect); + pWalker->eCode = eCodeOrig; + pTab->nCol = nCol; + } +#endif + } + + /* Locate the index named by the INDEXED BY clause, if any. */ + if( sqlite3IndexedByLookup(pParse, pFrom) ){ + return WRC_Abort; + } + } + + /* Process NATURAL keywords, and ON and USING clauses of joins. + */ + if( db->mallocFailed || sqliteProcessJoin(pParse, p) ){ + return WRC_Abort; + } + + /* For every "*" that occurs in the column list, insert the names of + ** all columns in all tables. And for every TABLE.* insert the names + ** of all columns in TABLE. The parser inserted a special expression + ** with the TK_ASTERISK operator for each "*" that it found in the column + ** list. The following code just has to locate the TK_ASTERISK + ** expressions and expand each one to the list of all columns in + ** all tables. + ** + ** The first loop just checks to see if there are any "*" operators + ** that need expanding. + */ + for(k=0; knExpr; k++){ + pE = pEList->a[k].pExpr; + if( pE->op==TK_ASTERISK ) break; + assert( pE->op!=TK_DOT || pE->pRight!=0 ); + assert( pE->op!=TK_DOT || (pE->pLeft!=0 && pE->pLeft->op==TK_ID) ); + if( pE->op==TK_DOT && pE->pRight->op==TK_ASTERISK ) break; + elistFlags |= pE->flags; + } + if( knExpr ){ + /* + ** If we get here it means the result set contains one or more "*" + ** operators that need to be expanded. Loop through each expression + ** in the result set and expand them one by one. + */ + struct ExprList_item *a = pEList->a; + ExprList *pNew = 0; + int flags = pParse->db->flags; + int longNames = (flags & SQLITE_FullColNames)!=0 + && (flags & SQLITE_ShortColNames)==0; + + for(k=0; knExpr; k++){ + pE = a[k].pExpr; + elistFlags |= pE->flags; + pRight = pE->pRight; + assert( pE->op!=TK_DOT || pRight!=0 ); + if( pE->op!=TK_ASTERISK + && (pE->op!=TK_DOT || pRight->op!=TK_ASTERISK) + ){ + /* This particular expression does not need to be expanded. + */ + pNew = sqlite3ExprListAppend(pParse, pNew, a[k].pExpr); + if( pNew ){ + pNew->a[pNew->nExpr-1].zName = a[k].zName; + pNew->a[pNew->nExpr-1].zSpan = a[k].zSpan; + a[k].zName = 0; + a[k].zSpan = 0; + } + a[k].pExpr = 0; + }else{ + /* This expression is a "*" or a "TABLE.*" and needs to be + ** expanded. */ + int tableSeen = 0; /* Set to 1 when TABLE matches */ + char *zTName = 0; /* text of name of TABLE */ + if( pE->op==TK_DOT ){ + assert( pE->pLeft!=0 ); + assert( !ExprHasProperty(pE->pLeft, EP_IntValue) ); + zTName = pE->pLeft->u.zToken; + } + for(i=0, pFrom=pTabList->a; inSrc; i++, pFrom++){ + Table *pTab = pFrom->pTab; + Select *pSub = pFrom->pSelect; + char *zTabName = pFrom->zAlias; + const char *zSchemaName = 0; + int iDb; + if( zTabName==0 ){ + zTabName = pTab->zName; + } + if( db->mallocFailed ) break; + if( pSub==0 || (pSub->selFlags & SF_NestedFrom)==0 ){ + pSub = 0; + if( zTName && sqlite3StrICmp(zTName, zTabName)!=0 ){ + continue; + } + iDb = sqlite3SchemaToIndex(db, pTab->pSchema); + zSchemaName = iDb>=0 ? db->aDb[iDb].zDbSName : "*"; + } + for(j=0; jnCol; j++){ + char *zName = pTab->aCol[j].zName; + char *zColname; /* The computed column name */ + char *zToFree; /* Malloced string that needs to be freed */ + Token sColname; /* Computed column name as a token */ + + assert( zName ); + if( zTName && pSub + && sqlite3MatchSpanName(pSub->pEList->a[j].zSpan, 0, zTName, 0)==0 + ){ + continue; + } + + /* If a column is marked as 'hidden', omit it from the expanded + ** result-set list unless the SELECT has the SF_IncludeHidden + ** bit set. + */ + if( (p->selFlags & SF_IncludeHidden)==0 + && IsHiddenColumn(&pTab->aCol[j]) + ){ + continue; + } + tableSeen = 1; + + if( i>0 && zTName==0 ){ + if( (pFrom->fg.jointype & JT_NATURAL)!=0 + && tableAndColumnIndex(pTabList, i, zName, 0, 0) + ){ + /* In a NATURAL join, omit the join columns from the + ** table to the right of the join */ + continue; + } + if( sqlite3IdListIndex(pFrom->pUsing, zName)>=0 ){ + /* In a join with a USING clause, omit columns in the + ** using clause from the table on the right. */ + continue; + } + } + pRight = sqlite3Expr(db, TK_ID, zName); + zColname = zName; + zToFree = 0; + if( longNames || pTabList->nSrc>1 ){ + Expr *pLeft; + pLeft = sqlite3Expr(db, TK_ID, zTabName); + pExpr = sqlite3PExpr(pParse, TK_DOT, pLeft, pRight); + if( zSchemaName ){ + pLeft = sqlite3Expr(db, TK_ID, zSchemaName); + pExpr = sqlite3PExpr(pParse, TK_DOT, pLeft, pExpr); + } + if( longNames ){ + zColname = sqlite3MPrintf(db, "%s.%s", zTabName, zName); + zToFree = zColname; + } + }else{ + pExpr = pRight; + } + pNew = sqlite3ExprListAppend(pParse, pNew, pExpr); + sqlite3TokenInit(&sColname, zColname); + sqlite3ExprListSetName(pParse, pNew, &sColname, 0); + if( pNew && (p->selFlags & SF_NestedFrom)!=0 ){ + struct ExprList_item *pX = &pNew->a[pNew->nExpr-1]; + if( pSub ){ + pX->zSpan = sqlite3DbStrDup(db, pSub->pEList->a[j].zSpan); + testcase( pX->zSpan==0 ); + }else{ + pX->zSpan = sqlite3MPrintf(db, "%s.%s.%s", + zSchemaName, zTabName, zColname); + testcase( pX->zSpan==0 ); + } + pX->bSpanIsTab = 1; + } + sqlite3DbFree(db, zToFree); + } + } + if( !tableSeen ){ + if( zTName ){ + sqlite3ErrorMsg(pParse, "no such table: %s", zTName); + }else{ + sqlite3ErrorMsg(pParse, "no tables specified"); + } + } + } + } + sqlite3ExprListDelete(db, pEList); + p->pEList = pNew; + } + if( p->pEList ){ + if( p->pEList->nExpr>db->aLimit[SQLITE_LIMIT_COLUMN] ){ + sqlite3ErrorMsg(pParse, "too many columns in result set"); + return WRC_Abort; + } + if( (elistFlags & (EP_HasFunc|EP_Subquery))!=0 ){ + p->selFlags |= SF_ComplexResult; + } + } + return WRC_Continue; +} + +/* +** No-op routine for the parse-tree walker. +** +** When this routine is the Walker.xExprCallback then expression trees +** are walked without any actions being taken at each node. Presumably, +** when this routine is used for Walker.xExprCallback then +** Walker.xSelectCallback is set to do something useful for every +** subquery in the parser tree. +*/ +SQLITE_PRIVATE int sqlite3ExprWalkNoop(Walker *NotUsed, Expr *NotUsed2){ + UNUSED_PARAMETER2(NotUsed, NotUsed2); + return WRC_Continue; +} + +/* +** No-op routine for the parse-tree walker for SELECT statements. +** subquery in the parser tree. +*/ +SQLITE_PRIVATE int sqlite3SelectWalkNoop(Walker *NotUsed, Select *NotUsed2){ + UNUSED_PARAMETER2(NotUsed, NotUsed2); + return WRC_Continue; +} + +#if SQLITE_DEBUG +/* +** Always assert. This xSelectCallback2 implementation proves that the +** xSelectCallback2 is never invoked. +*/ +SQLITE_PRIVATE void sqlite3SelectWalkAssert2(Walker *NotUsed, Select *NotUsed2){ + UNUSED_PARAMETER2(NotUsed, NotUsed2); + assert( 0 ); +} +#endif +/* +** This routine "expands" a SELECT statement and all of its subqueries. +** For additional information on what it means to "expand" a SELECT +** statement, see the comment on the selectExpand worker callback above. +** +** Expanding a SELECT statement is the first step in processing a +** SELECT statement. The SELECT statement must be expanded before +** name resolution is performed. +** +** If anything goes wrong, an error message is written into pParse. +** The calling function can detect the problem by looking at pParse->nErr +** and/or pParse->db->mallocFailed. +*/ +static void sqlite3SelectExpand(Parse *pParse, Select *pSelect){ + Walker w; + w.xExprCallback = sqlite3ExprWalkNoop; + w.pParse = pParse; + if( OK_IF_ALWAYS_TRUE(pParse->hasCompound) ){ + w.xSelectCallback = convertCompoundSelectToSubquery; + w.xSelectCallback2 = 0; + sqlite3WalkSelect(&w, pSelect); + } + w.xSelectCallback = selectExpander; + w.xSelectCallback2 = selectPopWith; + w.eCode = 0; + sqlite3WalkSelect(&w, pSelect); +} + + +#ifndef SQLITE_OMIT_SUBQUERY +/* +** This is a Walker.xSelectCallback callback for the sqlite3SelectTypeInfo() +** interface. +** +** For each FROM-clause subquery, add Column.zType and Column.zColl +** information to the Table structure that represents the result set +** of that subquery. +** +** The Table structure that represents the result set was constructed +** by selectExpander() but the type and collation information was omitted +** at that point because identifiers had not yet been resolved. This +** routine is called after identifier resolution. +*/ +static void selectAddSubqueryTypeInfo(Walker *pWalker, Select *p){ + Parse *pParse; + int i; + SrcList *pTabList; + struct SrcList_item *pFrom; + + assert( p->selFlags & SF_Resolved ); + if( p->selFlags & SF_HasTypeInfo ) return; + p->selFlags |= SF_HasTypeInfo; + pParse = pWalker->pParse; + pTabList = p->pSrc; + for(i=0, pFrom=pTabList->a; inSrc; i++, pFrom++){ + Table *pTab = pFrom->pTab; + assert( pTab!=0 ); + if( (pTab->tabFlags & TF_Ephemeral)!=0 ){ + /* A sub-query in the FROM clause of a SELECT */ + Select *pSel = pFrom->pSelect; + if( pSel ){ + while( pSel->pPrior ) pSel = pSel->pPrior; + sqlite3SelectAddColumnTypeAndCollation(pParse, pTab, pSel, + SQLITE_AFF_NONE); + } + } + } +} +#endif + + +/* +** This routine adds datatype and collating sequence information to +** the Table structures of all FROM-clause subqueries in a +** SELECT statement. +** +** Use this routine after name resolution. +*/ +static void sqlite3SelectAddTypeInfo(Parse *pParse, Select *pSelect){ +#ifndef SQLITE_OMIT_SUBQUERY + Walker w; + w.xSelectCallback = sqlite3SelectWalkNoop; + w.xSelectCallback2 = selectAddSubqueryTypeInfo; + w.xExprCallback = sqlite3ExprWalkNoop; + w.pParse = pParse; + sqlite3WalkSelect(&w, pSelect); +#endif +} + + +/* +** This routine sets up a SELECT statement for processing. The +** following is accomplished: +** +** * VDBE Cursor numbers are assigned to all FROM-clause terms. +** * Ephemeral Table objects are created for all FROM-clause subqueries. +** * ON and USING clauses are shifted into WHERE statements +** * Wildcards "*" and "TABLE.*" in result sets are expanded. +** * Identifiers in expression are matched to tables. +** +** This routine acts recursively on all subqueries within the SELECT. +*/ +SQLITE_PRIVATE void sqlite3SelectPrep( + Parse *pParse, /* The parser context */ + Select *p, /* The SELECT statement being coded. */ + NameContext *pOuterNC /* Name context for container */ +){ + assert( p!=0 || pParse->db->mallocFailed ); + if( pParse->db->mallocFailed ) return; + if( p->selFlags & SF_HasTypeInfo ) return; + sqlite3SelectExpand(pParse, p); + if( pParse->nErr || pParse->db->mallocFailed ) return; + sqlite3ResolveSelectNames(pParse, p, pOuterNC); + if( pParse->nErr || pParse->db->mallocFailed ) return; + sqlite3SelectAddTypeInfo(pParse, p); +} + +/* +** Reset the aggregate accumulator. +** +** The aggregate accumulator is a set of memory cells that hold +** intermediate results while calculating an aggregate. This +** routine generates code that stores NULLs in all of those memory +** cells. +*/ +static void resetAccumulator(Parse *pParse, AggInfo *pAggInfo){ + Vdbe *v = pParse->pVdbe; + int i; + struct AggInfo_func *pFunc; + int nReg = pAggInfo->nFunc + pAggInfo->nColumn; + if( nReg==0 ) return; +#ifdef SQLITE_DEBUG + /* Verify that all AggInfo registers are within the range specified by + ** AggInfo.mnReg..AggInfo.mxReg */ + assert( nReg==pAggInfo->mxReg-pAggInfo->mnReg+1 ); + for(i=0; inColumn; i++){ + assert( pAggInfo->aCol[i].iMem>=pAggInfo->mnReg + && pAggInfo->aCol[i].iMem<=pAggInfo->mxReg ); + } + for(i=0; inFunc; i++){ + assert( pAggInfo->aFunc[i].iMem>=pAggInfo->mnReg + && pAggInfo->aFunc[i].iMem<=pAggInfo->mxReg ); + } +#endif + sqlite3VdbeAddOp3(v, OP_Null, 0, pAggInfo->mnReg, pAggInfo->mxReg); + for(pFunc=pAggInfo->aFunc, i=0; inFunc; i++, pFunc++){ + if( pFunc->iDistinct>=0 ){ + Expr *pE = pFunc->pExpr; + assert( !ExprHasProperty(pE, EP_xIsSelect) ); + if( pE->x.pList==0 || pE->x.pList->nExpr!=1 ){ + sqlite3ErrorMsg(pParse, "DISTINCT aggregates must have exactly one " + "argument"); + pFunc->iDistinct = -1; + }else{ + KeyInfo *pKeyInfo = sqlite3KeyInfoFromExprList(pParse, pE->x.pList,0,0); + sqlite3VdbeAddOp4(v, OP_OpenEphemeral, pFunc->iDistinct, 0, 0, + (char*)pKeyInfo, P4_KEYINFO); + } + } + } +} + +/* +** Invoke the OP_AggFinalize opcode for every aggregate function +** in the AggInfo structure. +*/ +static void finalizeAggFunctions(Parse *pParse, AggInfo *pAggInfo){ + Vdbe *v = pParse->pVdbe; + int i; + struct AggInfo_func *pF; + for(i=0, pF=pAggInfo->aFunc; inFunc; i++, pF++){ + ExprList *pList = pF->pExpr->x.pList; + assert( !ExprHasProperty(pF->pExpr, EP_xIsSelect) ); + sqlite3VdbeAddOp2(v, OP_AggFinal, pF->iMem, pList ? pList->nExpr : 0); + sqlite3VdbeAppendP4(v, pF->pFunc, P4_FUNCDEF); + } +} + + +/* +** Update the accumulator memory cells for an aggregate based on +** the current cursor position. +** +** If regAcc is non-zero and there are no min() or max() aggregates +** in pAggInfo, then only populate the pAggInfo->nAccumulator accumulator +** registers if register regAcc contains 0. The caller will take care +** of setting and clearing regAcc. +*/ +static void updateAccumulator(Parse *pParse, int regAcc, AggInfo *pAggInfo){ + Vdbe *v = pParse->pVdbe; + int i; + int regHit = 0; + int addrHitTest = 0; + struct AggInfo_func *pF; + struct AggInfo_col *pC; + + pAggInfo->directMode = 1; + for(i=0, pF=pAggInfo->aFunc; inFunc; i++, pF++){ + int nArg; + int addrNext = 0; + int regAgg; + ExprList *pList = pF->pExpr->x.pList; + assert( !ExprHasProperty(pF->pExpr, EP_xIsSelect) ); + assert( !IsWindowFunc(pF->pExpr) ); + if( ExprHasProperty(pF->pExpr, EP_WinFunc) ){ + Expr *pFilter = pF->pExpr->y.pWin->pFilter; + if( pAggInfo->nAccumulator + && (pF->pFunc->funcFlags & SQLITE_FUNC_NEEDCOLL) + ){ + if( regHit==0 ) regHit = ++pParse->nMem; + /* If this is the first row of the group (regAcc==0), clear the + ** "magnet" register regHit so that the accumulator registers + ** are populated if the FILTER clause jumps over the the + ** invocation of min() or max() altogether. Or, if this is not + ** the first row (regAcc==1), set the magnet register so that the + ** accumulators are not populated unless the min()/max() is invoked and + ** indicates that they should be. */ + sqlite3VdbeAddOp2(v, OP_Copy, regAcc, regHit); + } + addrNext = sqlite3VdbeMakeLabel(pParse); + sqlite3ExprIfFalse(pParse, pFilter, addrNext, SQLITE_JUMPIFNULL); + } + if( pList ){ + nArg = pList->nExpr; + regAgg = sqlite3GetTempRange(pParse, nArg); + sqlite3ExprCodeExprList(pParse, pList, regAgg, 0, SQLITE_ECEL_DUP); + }else{ + nArg = 0; + regAgg = 0; + } + if( pF->iDistinct>=0 ){ + if( addrNext==0 ){ + addrNext = sqlite3VdbeMakeLabel(pParse); + } + testcase( nArg==0 ); /* Error condition */ + testcase( nArg>1 ); /* Also an error */ + codeDistinct(pParse, pF->iDistinct, addrNext, 1, regAgg); + } + if( pF->pFunc->funcFlags & SQLITE_FUNC_NEEDCOLL ){ + CollSeq *pColl = 0; + struct ExprList_item *pItem; + int j; + assert( pList!=0 ); /* pList!=0 if pF->pFunc has NEEDCOLL */ + for(j=0, pItem=pList->a; !pColl && jpExpr); + } + if( !pColl ){ + pColl = pParse->db->pDfltColl; + } + if( regHit==0 && pAggInfo->nAccumulator ) regHit = ++pParse->nMem; + sqlite3VdbeAddOp4(v, OP_CollSeq, regHit, 0, 0, (char *)pColl, P4_COLLSEQ); + } + sqlite3VdbeAddOp3(v, OP_AggStep, 0, regAgg, pF->iMem); + sqlite3VdbeAppendP4(v, pF->pFunc, P4_FUNCDEF); + sqlite3VdbeChangeP5(v, (u8)nArg); + sqlite3ReleaseTempRange(pParse, regAgg, nArg); + if( addrNext ){ + sqlite3VdbeResolveLabel(v, addrNext); + } + } + if( regHit==0 && pAggInfo->nAccumulator ){ + regHit = regAcc; + } + if( regHit ){ + addrHitTest = sqlite3VdbeAddOp1(v, OP_If, regHit); VdbeCoverage(v); + } + for(i=0, pC=pAggInfo->aCol; inAccumulator; i++, pC++){ + sqlite3ExprCode(pParse, pC->pExpr, pC->iMem); + } + + pAggInfo->directMode = 0; + if( addrHitTest ){ + sqlite3VdbeJumpHere(v, addrHitTest); + } +} + +/* +** Add a single OP_Explain instruction to the VDBE to explain a simple +** count(*) query ("SELECT count(*) FROM pTab"). +*/ +#ifndef SQLITE_OMIT_EXPLAIN +static void explainSimpleCount( + Parse *pParse, /* Parse context */ + Table *pTab, /* Table being queried */ + Index *pIdx /* Index used to optimize scan, or NULL */ +){ + if( pParse->explain==2 ){ + int bCover = (pIdx!=0 && (HasRowid(pTab) || !IsPrimaryKeyIndex(pIdx))); + sqlite3VdbeExplain(pParse, 0, "SCAN TABLE %s%s%s", + pTab->zName, + bCover ? " USING COVERING INDEX " : "", + bCover ? pIdx->zName : "" + ); + } +} +#else +# define explainSimpleCount(a,b,c) +#endif + +/* +** sqlite3WalkExpr() callback used by havingToWhere(). +** +** If the node passed to the callback is a TK_AND node, return +** WRC_Continue to tell sqlite3WalkExpr() to iterate through child nodes. +** +** Otherwise, return WRC_Prune. In this case, also check if the +** sub-expression matches the criteria for being moved to the WHERE +** clause. If so, add it to the WHERE clause and replace the sub-expression +** within the HAVING expression with a constant "1". +*/ +static int havingToWhereExprCb(Walker *pWalker, Expr *pExpr){ + if( pExpr->op!=TK_AND ){ + Select *pS = pWalker->u.pSelect; + if( sqlite3ExprIsConstantOrGroupBy(pWalker->pParse, pExpr, pS->pGroupBy) ){ + sqlite3 *db = pWalker->pParse->db; + Expr *pNew = sqlite3Expr(db, TK_INTEGER, "1"); + if( pNew ){ + Expr *pWhere = pS->pWhere; + SWAP(Expr, *pNew, *pExpr); + pNew = sqlite3ExprAnd(pWalker->pParse, pWhere, pNew); + pS->pWhere = pNew; + pWalker->eCode = 1; + } + } + return WRC_Prune; + } + return WRC_Continue; +} + +/* +** Transfer eligible terms from the HAVING clause of a query, which is +** processed after grouping, to the WHERE clause, which is processed before +** grouping. For example, the query: +** +** SELECT * FROM WHERE a=? GROUP BY b HAVING b=? AND c=? +** +** can be rewritten as: +** +** SELECT * FROM WHERE a=? AND b=? GROUP BY b HAVING c=? +** +** A term of the HAVING expression is eligible for transfer if it consists +** entirely of constants and expressions that are also GROUP BY terms that +** use the "BINARY" collation sequence. +*/ +static void havingToWhere(Parse *pParse, Select *p){ + Walker sWalker; + memset(&sWalker, 0, sizeof(sWalker)); + sWalker.pParse = pParse; + sWalker.xExprCallback = havingToWhereExprCb; + sWalker.u.pSelect = p; + sqlite3WalkExpr(&sWalker, p->pHaving); +#if SELECTTRACE_ENABLED + if( sWalker.eCode && (sqlite3SelectTrace & 0x100)!=0 ){ + SELECTTRACE(0x100,pParse,p,("Move HAVING terms into WHERE:\n")); + sqlite3TreeViewSelect(0, p, 0); + } +#endif +} + +/* +** Check to see if the pThis entry of pTabList is a self-join of a prior view. +** If it is, then return the SrcList_item for the prior view. If it is not, +** then return 0. +*/ +static struct SrcList_item *isSelfJoinView( + SrcList *pTabList, /* Search for self-joins in this FROM clause */ + struct SrcList_item *pThis /* Search for prior reference to this subquery */ +){ + struct SrcList_item *pItem; + for(pItem = pTabList->a; pItempSelect==0 ) continue; + if( pItem->fg.viaCoroutine ) continue; + if( pItem->zName==0 ) continue; + assert( pItem->pTab!=0 ); + assert( pThis->pTab!=0 ); + if( pItem->pTab->pSchema!=pThis->pTab->pSchema ) continue; + if( sqlite3_stricmp(pItem->zName, pThis->zName)!=0 ) continue; + pS1 = pItem->pSelect; + if( pItem->pTab->pSchema==0 && pThis->pSelect->selId!=pS1->selId ){ + /* The query flattener left two different CTE tables with identical + ** names in the same FROM clause. */ + continue; + } + if( sqlite3ExprCompare(0, pThis->pSelect->pWhere, pS1->pWhere, -1) + || sqlite3ExprCompare(0, pThis->pSelect->pHaving, pS1->pHaving, -1) + ){ + /* The view was modified by some other optimization such as + ** pushDownWhereTerms() */ + continue; + } + return pItem; + } + return 0; +} + +#ifdef SQLITE_COUNTOFVIEW_OPTIMIZATION +/* +** Attempt to transform a query of the form +** +** SELECT count(*) FROM (SELECT x FROM t1 UNION ALL SELECT y FROM t2) +** +** Into this: +** +** SELECT (SELECT count(*) FROM t1)+(SELECT count(*) FROM t2) +** +** The transformation only works if all of the following are true: +** +** * The subquery is a UNION ALL of two or more terms +** * The subquery does not have a LIMIT clause +** * There is no WHERE or GROUP BY or HAVING clauses on the subqueries +** * The outer query is a simple count(*) with no WHERE clause or other +** extraneous syntax. +** +** Return TRUE if the optimization is undertaken. +*/ +static int countOfViewOptimization(Parse *pParse, Select *p){ + Select *pSub, *pPrior; + Expr *pExpr; + Expr *pCount; + sqlite3 *db; + if( (p->selFlags & SF_Aggregate)==0 ) return 0; /* This is an aggregate */ + if( p->pEList->nExpr!=1 ) return 0; /* Single result column */ + if( p->pWhere ) return 0; + if( p->pGroupBy ) return 0; + pExpr = p->pEList->a[0].pExpr; + if( pExpr->op!=TK_AGG_FUNCTION ) return 0; /* Result is an aggregate */ + if( sqlite3_stricmp(pExpr->u.zToken,"count") ) return 0; /* Is count() */ + if( pExpr->x.pList!=0 ) return 0; /* Must be count(*) */ + if( p->pSrc->nSrc!=1 ) return 0; /* One table in FROM */ + pSub = p->pSrc->a[0].pSelect; + if( pSub==0 ) return 0; /* The FROM is a subquery */ + if( pSub->pPrior==0 ) return 0; /* Must be a compound ry */ + do{ + if( pSub->op!=TK_ALL && pSub->pPrior ) return 0; /* Must be UNION ALL */ + if( pSub->pWhere ) return 0; /* No WHERE clause */ + if( pSub->pLimit ) return 0; /* No LIMIT clause */ + if( pSub->selFlags & SF_Aggregate ) return 0; /* Not an aggregate */ + pSub = pSub->pPrior; /* Repeat over compound */ + }while( pSub ); + + /* If we reach this point then it is OK to perform the transformation */ + + db = pParse->db; + pCount = pExpr; + pExpr = 0; + pSub = p->pSrc->a[0].pSelect; + p->pSrc->a[0].pSelect = 0; + sqlite3SrcListDelete(db, p->pSrc); + p->pSrc = sqlite3DbMallocZero(pParse->db, sizeof(*p->pSrc)); + while( pSub ){ + Expr *pTerm; + pPrior = pSub->pPrior; + pSub->pPrior = 0; + pSub->pNext = 0; + pSub->selFlags |= SF_Aggregate; + pSub->selFlags &= ~SF_Compound; + pSub->nSelectRow = 0; + sqlite3ExprListDelete(db, pSub->pEList); + pTerm = pPrior ? sqlite3ExprDup(db, pCount, 0) : pCount; + pSub->pEList = sqlite3ExprListAppend(pParse, 0, pTerm); + pTerm = sqlite3PExpr(pParse, TK_SELECT, 0, 0); + sqlite3PExprAddSelect(pParse, pTerm, pSub); + if( pExpr==0 ){ + pExpr = pTerm; + }else{ + pExpr = sqlite3PExpr(pParse, TK_PLUS, pTerm, pExpr); + } + pSub = pPrior; + } + p->pEList->a[0].pExpr = pExpr; + p->selFlags &= ~SF_Aggregate; + +#if SELECTTRACE_ENABLED + if( sqlite3SelectTrace & 0x400 ){ + SELECTTRACE(0x400,pParse,p,("After count-of-view optimization:\n")); + sqlite3TreeViewSelect(0, p, 0); + } +#endif + return 1; +} +#endif /* SQLITE_COUNTOFVIEW_OPTIMIZATION */ + +/* +** Generate code for the SELECT statement given in the p argument. +** +** The results are returned according to the SelectDest structure. +** See comments in sqliteInt.h for further information. +** +** This routine returns the number of errors. If any errors are +** encountered, then an appropriate error message is left in +** pParse->zErrMsg. +** +** This routine does NOT free the Select structure passed in. The +** calling function needs to do that. +*/ +SQLITE_PRIVATE int sqlite3Select( + Parse *pParse, /* The parser context */ + Select *p, /* The SELECT statement being coded. */ + SelectDest *pDest /* What to do with the query results */ +){ + int i, j; /* Loop counters */ + WhereInfo *pWInfo; /* Return from sqlite3WhereBegin() */ + Vdbe *v; /* The virtual machine under construction */ + int isAgg; /* True for select lists like "count(*)" */ + ExprList *pEList = 0; /* List of columns to extract. */ + SrcList *pTabList; /* List of tables to select from */ + Expr *pWhere; /* The WHERE clause. May be NULL */ + ExprList *pGroupBy; /* The GROUP BY clause. May be NULL */ + Expr *pHaving; /* The HAVING clause. May be NULL */ + int rc = 1; /* Value to return from this function */ + DistinctCtx sDistinct; /* Info on how to code the DISTINCT keyword */ + SortCtx sSort; /* Info on how to code the ORDER BY clause */ + AggInfo sAggInfo; /* Information used by aggregate queries */ + int iEnd; /* Address of the end of the query */ + sqlite3 *db; /* The database connection */ + ExprList *pMinMaxOrderBy = 0; /* Added ORDER BY for min/max queries */ + u8 minMaxFlag; /* Flag for min/max queries */ + + db = pParse->db; + v = sqlite3GetVdbe(pParse); + if( p==0 || db->mallocFailed || pParse->nErr ){ + return 1; + } + if( sqlite3AuthCheck(pParse, SQLITE_SELECT, 0, 0, 0) ) return 1; + memset(&sAggInfo, 0, sizeof(sAggInfo)); +#if SELECTTRACE_ENABLED + SELECTTRACE(1,pParse,p, ("begin processing:\n", pParse->addrExplain)); + if( sqlite3SelectTrace & 0x100 ){ + sqlite3TreeViewSelect(0, p, 0); + } +#endif + + assert( p->pOrderBy==0 || pDest->eDest!=SRT_DistFifo ); + assert( p->pOrderBy==0 || pDest->eDest!=SRT_Fifo ); + assert( p->pOrderBy==0 || pDest->eDest!=SRT_DistQueue ); + assert( p->pOrderBy==0 || pDest->eDest!=SRT_Queue ); + if( IgnorableOrderby(pDest) ){ + assert(pDest->eDest==SRT_Exists || pDest->eDest==SRT_Union || + pDest->eDest==SRT_Except || pDest->eDest==SRT_Discard || + pDest->eDest==SRT_Queue || pDest->eDest==SRT_DistFifo || + pDest->eDest==SRT_DistQueue || pDest->eDest==SRT_Fifo); + /* If ORDER BY makes no difference in the output then neither does + ** DISTINCT so it can be removed too. */ + sqlite3ExprListDelete(db, p->pOrderBy); + p->pOrderBy = 0; + p->selFlags &= ~SF_Distinct; + } + sqlite3SelectPrep(pParse, p, 0); + if( pParse->nErr || db->mallocFailed ){ + goto select_end; + } + assert( p->pEList!=0 ); +#if SELECTTRACE_ENABLED + if( sqlite3SelectTrace & 0x104 ){ + SELECTTRACE(0x104,pParse,p, ("after name resolution:\n")); + sqlite3TreeViewSelect(0, p, 0); + } +#endif + + if( pDest->eDest==SRT_Output ){ + generateColumnNames(pParse, p); + } + +#ifndef SQLITE_OMIT_WINDOWFUNC + if( sqlite3WindowRewrite(pParse, p) ){ + goto select_end; + } +#if SELECTTRACE_ENABLED + if( sqlite3SelectTrace & 0x108 ){ + SELECTTRACE(0x104,pParse,p, ("after window rewrite:\n")); + sqlite3TreeViewSelect(0, p, 0); + } +#endif +#endif /* SQLITE_OMIT_WINDOWFUNC */ + pTabList = p->pSrc; + isAgg = (p->selFlags & SF_Aggregate)!=0; + memset(&sSort, 0, sizeof(sSort)); + sSort.pOrderBy = p->pOrderBy; + + /* Try to various optimizations (flattening subqueries, and strength + ** reduction of join operators) in the FROM clause up into the main query + */ +#if !defined(SQLITE_OMIT_SUBQUERY) || !defined(SQLITE_OMIT_VIEW) + for(i=0; !p->pPrior && inSrc; i++){ + struct SrcList_item *pItem = &pTabList->a[i]; + Select *pSub = pItem->pSelect; + Table *pTab = pItem->pTab; + + /* Convert LEFT JOIN into JOIN if there are terms of the right table + ** of the LEFT JOIN used in the WHERE clause. + */ + if( (pItem->fg.jointype & JT_LEFT)!=0 + && sqlite3ExprImpliesNonNullRow(p->pWhere, pItem->iCursor) + && OptimizationEnabled(db, SQLITE_SimplifyJoin) + ){ + SELECTTRACE(0x100,pParse,p, + ("LEFT-JOIN simplifies to JOIN on term %d\n",i)); + pItem->fg.jointype &= ~(JT_LEFT|JT_OUTER); + unsetJoinExpr(p->pWhere, pItem->iCursor); + } + + /* No futher action if this term of the FROM clause is no a subquery */ + if( pSub==0 ) continue; + + /* Catch mismatch in the declared columns of a view and the number of + ** columns in the SELECT on the RHS */ + if( pTab->nCol!=pSub->pEList->nExpr ){ + sqlite3ErrorMsg(pParse, "expected %d columns for '%s' but got %d", + pTab->nCol, pTab->zName, pSub->pEList->nExpr); + goto select_end; + } + + /* Do not try to flatten an aggregate subquery. + ** + ** Flattening an aggregate subquery is only possible if the outer query + ** is not a join. But if the outer query is not a join, then the subquery + ** will be implemented as a co-routine and there is no advantage to + ** flattening in that case. + */ + if( (pSub->selFlags & SF_Aggregate)!=0 ) continue; + assert( pSub->pGroupBy==0 ); + + /* If the outer query contains a "complex" result set (that is, + ** if the result set of the outer query uses functions or subqueries) + ** and if the subquery contains an ORDER BY clause and if + ** it will be implemented as a co-routine, then do not flatten. This + ** restriction allows SQL constructs like this: + ** + ** SELECT expensive_function(x) + ** FROM (SELECT x FROM tab ORDER BY y LIMIT 10); + ** + ** The expensive_function() is only computed on the 10 rows that + ** are output, rather than every row of the table. + ** + ** The requirement that the outer query have a complex result set + ** means that flattening does occur on simpler SQL constraints without + ** the expensive_function() like: + ** + ** SELECT x FROM (SELECT x FROM tab ORDER BY y LIMIT 10); + */ + if( pSub->pOrderBy!=0 + && i==0 + && (p->selFlags & SF_ComplexResult)!=0 + && (pTabList->nSrc==1 + || (pTabList->a[1].fg.jointype&(JT_LEFT|JT_CROSS))!=0) + ){ + continue; + } + + if( flattenSubquery(pParse, p, i, isAgg) ){ + if( pParse->nErr ) goto select_end; + /* This subquery can be absorbed into its parent. */ + i = -1; + } + pTabList = p->pSrc; + if( db->mallocFailed ) goto select_end; + if( !IgnorableOrderby(pDest) ){ + sSort.pOrderBy = p->pOrderBy; + } + } +#endif + +#ifndef SQLITE_OMIT_COMPOUND_SELECT + /* Handle compound SELECT statements using the separate multiSelect() + ** procedure. + */ + if( p->pPrior ){ + rc = multiSelect(pParse, p, pDest); +#if SELECTTRACE_ENABLED + SELECTTRACE(0x1,pParse,p,("end compound-select processing\n")); + if( (sqlite3SelectTrace & 0x2000)!=0 && ExplainQueryPlanParent(pParse)==0 ){ + sqlite3TreeViewSelect(0, p, 0); + } +#endif + if( p->pNext==0 ) ExplainQueryPlanPop(pParse); + return rc; + } +#endif + + /* Do the WHERE-clause constant propagation optimization if this is + ** a join. No need to speed time on this operation for non-join queries + ** as the equivalent optimization will be handled by query planner in + ** sqlite3WhereBegin(). + */ + if( pTabList->nSrc>1 + && OptimizationEnabled(db, SQLITE_PropagateConst) + && propagateConstants(pParse, p) + ){ +#if SELECTTRACE_ENABLED + if( sqlite3SelectTrace & 0x100 ){ + SELECTTRACE(0x100,pParse,p,("After constant propagation:\n")); + sqlite3TreeViewSelect(0, p, 0); + } +#endif + }else{ + SELECTTRACE(0x100,pParse,p,("Constant propagation not helpful\n")); + } + +#ifdef SQLITE_COUNTOFVIEW_OPTIMIZATION + if( OptimizationEnabled(db, SQLITE_QueryFlattener|SQLITE_CountOfView) + && countOfViewOptimization(pParse, p) + ){ + if( db->mallocFailed ) goto select_end; + pEList = p->pEList; + pTabList = p->pSrc; + } +#endif + + /* For each term in the FROM clause, do two things: + ** (1) Authorized unreferenced tables + ** (2) Generate code for all sub-queries + */ + for(i=0; inSrc; i++){ + struct SrcList_item *pItem = &pTabList->a[i]; + SelectDest dest; + Select *pSub; +#if !defined(SQLITE_OMIT_SUBQUERY) || !defined(SQLITE_OMIT_VIEW) + const char *zSavedAuthContext; +#endif + + /* Issue SQLITE_READ authorizations with a fake column name for any + ** tables that are referenced but from which no values are extracted. + ** Examples of where these kinds of null SQLITE_READ authorizations + ** would occur: + ** + ** SELECT count(*) FROM t1; -- SQLITE_READ t1."" + ** SELECT t1.* FROM t1, t2; -- SQLITE_READ t2."" + ** + ** The fake column name is an empty string. It is possible for a table to + ** have a column named by the empty string, in which case there is no way to + ** distinguish between an unreferenced table and an actual reference to the + ** "" column. The original design was for the fake column name to be a NULL, + ** which would be unambiguous. But legacy authorization callbacks might + ** assume the column name is non-NULL and segfault. The use of an empty + ** string for the fake column name seems safer. + */ + if( pItem->colUsed==0 && pItem->zName!=0 ){ + sqlite3AuthCheck(pParse, SQLITE_READ, pItem->zName, "", pItem->zDatabase); + } + +#if !defined(SQLITE_OMIT_SUBQUERY) || !defined(SQLITE_OMIT_VIEW) + /* Generate code for all sub-queries in the FROM clause + */ + pSub = pItem->pSelect; + if( pSub==0 ) continue; + + /* The code for a subquery should only be generated once, though it is + ** technically harmless for it to be generated multiple times. The + ** following assert() will detect if something changes to cause + ** the same subquery to be coded multiple times, as a signal to the + ** developers to try to optimize the situation. + ** + ** Update 2019-07-24: + ** See ticket https://sqlite.org/src/tktview/c52b09c7f38903b1311cec40. + ** The dbsqlfuzz fuzzer found a case where the same subquery gets + ** coded twice. So this assert() now becomes a testcase(). It should + ** be very rare, though. + */ + testcase( pItem->addrFillSub!=0 ); + + /* Increment Parse.nHeight by the height of the largest expression + ** tree referred to by this, the parent select. The child select + ** may contain expression trees of at most + ** (SQLITE_MAX_EXPR_DEPTH-Parse.nHeight) height. This is a bit + ** more conservative than necessary, but much easier than enforcing + ** an exact limit. + */ + pParse->nHeight += sqlite3SelectExprHeight(p); + + /* Make copies of constant WHERE-clause terms in the outer query down + ** inside the subquery. This can help the subquery to run more efficiently. + */ + if( OptimizationEnabled(db, SQLITE_PushDown) + && pushDownWhereTerms(pParse, pSub, p->pWhere, pItem->iCursor, + (pItem->fg.jointype & JT_OUTER)!=0) + ){ +#if SELECTTRACE_ENABLED + if( sqlite3SelectTrace & 0x100 ){ + SELECTTRACE(0x100,pParse,p, + ("After WHERE-clause push-down into subquery %d:\n", pSub->selId)); + sqlite3TreeViewSelect(0, p, 0); + } +#endif + }else{ + SELECTTRACE(0x100,pParse,p,("Push-down not possible\n")); + } + + zSavedAuthContext = pParse->zAuthContext; + pParse->zAuthContext = pItem->zName; + + /* Generate code to implement the subquery + ** + ** The subquery is implemented as a co-routine if the subquery is + ** guaranteed to be the outer loop (so that it does not need to be + ** computed more than once) + ** + ** TODO: Are there other reasons beside (1) to use a co-routine + ** implementation? + */ + if( i==0 + && (pTabList->nSrc==1 + || (pTabList->a[1].fg.jointype&(JT_LEFT|JT_CROSS))!=0) /* (1) */ + ){ + /* Implement a co-routine that will return a single row of the result + ** set on each invocation. + */ + int addrTop = sqlite3VdbeCurrentAddr(v)+1; + + pItem->regReturn = ++pParse->nMem; + sqlite3VdbeAddOp3(v, OP_InitCoroutine, pItem->regReturn, 0, addrTop); + VdbeComment((v, "%s", pItem->pTab->zName)); + pItem->addrFillSub = addrTop; + sqlite3SelectDestInit(&dest, SRT_Coroutine, pItem->regReturn); + ExplainQueryPlan((pParse, 1, "CO-ROUTINE %u", pSub->selId)); + sqlite3Select(pParse, pSub, &dest); + pItem->pTab->nRowLogEst = pSub->nSelectRow; + pItem->fg.viaCoroutine = 1; + pItem->regResult = dest.iSdst; + sqlite3VdbeEndCoroutine(v, pItem->regReturn); + sqlite3VdbeJumpHere(v, addrTop-1); + sqlite3ClearTempRegCache(pParse); + }else{ + /* Generate a subroutine that will fill an ephemeral table with + ** the content of this subquery. pItem->addrFillSub will point + ** to the address of the generated subroutine. pItem->regReturn + ** is a register allocated to hold the subroutine return address + */ + int topAddr; + int onceAddr = 0; + int retAddr; + struct SrcList_item *pPrior; + + testcase( pItem->addrFillSub==0 ); /* Ticket c52b09c7f38903b1311 */ + pItem->regReturn = ++pParse->nMem; + topAddr = sqlite3VdbeAddOp2(v, OP_Integer, 0, pItem->regReturn); + pItem->addrFillSub = topAddr+1; + if( pItem->fg.isCorrelated==0 ){ + /* If the subquery is not correlated and if we are not inside of + ** a trigger, then we only need to compute the value of the subquery + ** once. */ + onceAddr = sqlite3VdbeAddOp0(v, OP_Once); VdbeCoverage(v); + VdbeComment((v, "materialize \"%s\"", pItem->pTab->zName)); + }else{ + VdbeNoopComment((v, "materialize \"%s\"", pItem->pTab->zName)); + } + pPrior = isSelfJoinView(pTabList, pItem); + if( pPrior ){ + sqlite3VdbeAddOp2(v, OP_OpenDup, pItem->iCursor, pPrior->iCursor); + assert( pPrior->pSelect!=0 ); + pSub->nSelectRow = pPrior->pSelect->nSelectRow; + }else{ + sqlite3SelectDestInit(&dest, SRT_EphemTab, pItem->iCursor); + ExplainQueryPlan((pParse, 1, "MATERIALIZE %u", pSub->selId)); + sqlite3Select(pParse, pSub, &dest); + } + pItem->pTab->nRowLogEst = pSub->nSelectRow; + if( onceAddr ) sqlite3VdbeJumpHere(v, onceAddr); + retAddr = sqlite3VdbeAddOp1(v, OP_Return, pItem->regReturn); + VdbeComment((v, "end %s", pItem->pTab->zName)); + sqlite3VdbeChangeP1(v, topAddr, retAddr); + sqlite3ClearTempRegCache(pParse); + } + if( db->mallocFailed ) goto select_end; + pParse->nHeight -= sqlite3SelectExprHeight(p); + pParse->zAuthContext = zSavedAuthContext; +#endif + } + + /* Various elements of the SELECT copied into local variables for + ** convenience */ + pEList = p->pEList; + pWhere = p->pWhere; + pGroupBy = p->pGroupBy; + pHaving = p->pHaving; + sDistinct.isTnct = (p->selFlags & SF_Distinct)!=0; + +#if SELECTTRACE_ENABLED + if( sqlite3SelectTrace & 0x400 ){ + SELECTTRACE(0x400,pParse,p,("After all FROM-clause analysis:\n")); + sqlite3TreeViewSelect(0, p, 0); + } +#endif + + /* If the query is DISTINCT with an ORDER BY but is not an aggregate, and + ** if the select-list is the same as the ORDER BY list, then this query + ** can be rewritten as a GROUP BY. In other words, this: + ** + ** SELECT DISTINCT xyz FROM ... ORDER BY xyz + ** + ** is transformed to: + ** + ** SELECT xyz FROM ... GROUP BY xyz ORDER BY xyz + ** + ** The second form is preferred as a single index (or temp-table) may be + ** used for both the ORDER BY and DISTINCT processing. As originally + ** written the query must use a temp-table for at least one of the ORDER + ** BY and DISTINCT, and an index or separate temp-table for the other. + */ + if( (p->selFlags & (SF_Distinct|SF_Aggregate))==SF_Distinct + && sqlite3ExprListCompare(sSort.pOrderBy, pEList, -1)==0 + ){ + p->selFlags &= ~SF_Distinct; + pGroupBy = p->pGroupBy = sqlite3ExprListDup(db, pEList, 0); + /* Notice that even thought SF_Distinct has been cleared from p->selFlags, + ** the sDistinct.isTnct is still set. Hence, isTnct represents the + ** original setting of the SF_Distinct flag, not the current setting */ + assert( sDistinct.isTnct ); + +#if SELECTTRACE_ENABLED + if( sqlite3SelectTrace & 0x400 ){ + SELECTTRACE(0x400,pParse,p,("Transform DISTINCT into GROUP BY:\n")); + sqlite3TreeViewSelect(0, p, 0); + } +#endif + } + + /* If there is an ORDER BY clause, then create an ephemeral index to + ** do the sorting. But this sorting ephemeral index might end up + ** being unused if the data can be extracted in pre-sorted order. + ** If that is the case, then the OP_OpenEphemeral instruction will be + ** changed to an OP_Noop once we figure out that the sorting index is + ** not needed. The sSort.addrSortIndex variable is used to facilitate + ** that change. + */ + if( sSort.pOrderBy ){ + KeyInfo *pKeyInfo; + pKeyInfo = sqlite3KeyInfoFromExprList( + pParse, sSort.pOrderBy, 0, pEList->nExpr); + sSort.iECursor = pParse->nTab++; + sSort.addrSortIndex = + sqlite3VdbeAddOp4(v, OP_OpenEphemeral, + sSort.iECursor, sSort.pOrderBy->nExpr+1+pEList->nExpr, 0, + (char*)pKeyInfo, P4_KEYINFO + ); + }else{ + sSort.addrSortIndex = -1; + } + + /* If the output is destined for a temporary table, open that table. + */ + if( pDest->eDest==SRT_EphemTab ){ + sqlite3VdbeAddOp2(v, OP_OpenEphemeral, pDest->iSDParm, pEList->nExpr); + } + + /* Set the limiter. + */ + iEnd = sqlite3VdbeMakeLabel(pParse); + if( (p->selFlags & SF_FixedLimit)==0 ){ + p->nSelectRow = 320; /* 4 billion rows */ + } + computeLimitRegisters(pParse, p, iEnd); + if( p->iLimit==0 && sSort.addrSortIndex>=0 ){ + sqlite3VdbeChangeOpcode(v, sSort.addrSortIndex, OP_SorterOpen); + sSort.sortFlags |= SORTFLAG_UseSorter; + } + + /* Open an ephemeral index to use for the distinct set. + */ + if( p->selFlags & SF_Distinct ){ + sDistinct.tabTnct = pParse->nTab++; + sDistinct.addrTnct = sqlite3VdbeAddOp4(v, OP_OpenEphemeral, + sDistinct.tabTnct, 0, 0, + (char*)sqlite3KeyInfoFromExprList(pParse, p->pEList,0,0), + P4_KEYINFO); + sqlite3VdbeChangeP5(v, BTREE_UNORDERED); + sDistinct.eTnctType = WHERE_DISTINCT_UNORDERED; + }else{ + sDistinct.eTnctType = WHERE_DISTINCT_NOOP; + } + + if( !isAgg && pGroupBy==0 ){ + /* No aggregate functions and no GROUP BY clause */ + u16 wctrlFlags = (sDistinct.isTnct ? WHERE_WANT_DISTINCT : 0) + | (p->selFlags & SF_FixedLimit); +#ifndef SQLITE_OMIT_WINDOWFUNC + Window *pWin = p->pWin; /* Master window object (or NULL) */ + if( pWin ){ + sqlite3WindowCodeInit(pParse, pWin); + } +#endif + assert( WHERE_USE_LIMIT==SF_FixedLimit ); + + + /* Begin the database scan. */ + SELECTTRACE(1,pParse,p,("WhereBegin\n")); + pWInfo = sqlite3WhereBegin(pParse, pTabList, pWhere, sSort.pOrderBy, + p->pEList, wctrlFlags, p->nSelectRow); + if( pWInfo==0 ) goto select_end; + if( sqlite3WhereOutputRowCount(pWInfo) < p->nSelectRow ){ + p->nSelectRow = sqlite3WhereOutputRowCount(pWInfo); + } + if( sDistinct.isTnct && sqlite3WhereIsDistinct(pWInfo) ){ + sDistinct.eTnctType = sqlite3WhereIsDistinct(pWInfo); + } + if( sSort.pOrderBy ){ + sSort.nOBSat = sqlite3WhereIsOrdered(pWInfo); + sSort.labelOBLopt = sqlite3WhereOrderByLimitOptLabel(pWInfo); + if( sSort.nOBSat==sSort.pOrderBy->nExpr ){ + sSort.pOrderBy = 0; + } + } + + /* If sorting index that was created by a prior OP_OpenEphemeral + ** instruction ended up not being needed, then change the OP_OpenEphemeral + ** into an OP_Noop. + */ + if( sSort.addrSortIndex>=0 && sSort.pOrderBy==0 ){ + sqlite3VdbeChangeToNoop(v, sSort.addrSortIndex); + } + + assert( p->pEList==pEList ); +#ifndef SQLITE_OMIT_WINDOWFUNC + if( pWin ){ + int addrGosub = sqlite3VdbeMakeLabel(pParse); + int iCont = sqlite3VdbeMakeLabel(pParse); + int iBreak = sqlite3VdbeMakeLabel(pParse); + int regGosub = ++pParse->nMem; + + sqlite3WindowCodeStep(pParse, p, pWInfo, regGosub, addrGosub); + + sqlite3VdbeAddOp2(v, OP_Goto, 0, iBreak); + sqlite3VdbeResolveLabel(v, addrGosub); + VdbeNoopComment((v, "inner-loop subroutine")); + sSort.labelOBLopt = 0; + selectInnerLoop(pParse, p, -1, &sSort, &sDistinct, pDest, iCont, iBreak); + sqlite3VdbeResolveLabel(v, iCont); + sqlite3VdbeAddOp1(v, OP_Return, regGosub); + VdbeComment((v, "end inner-loop subroutine")); + sqlite3VdbeResolveLabel(v, iBreak); + }else +#endif /* SQLITE_OMIT_WINDOWFUNC */ + { + /* Use the standard inner loop. */ + selectInnerLoop(pParse, p, -1, &sSort, &sDistinct, pDest, + sqlite3WhereContinueLabel(pWInfo), + sqlite3WhereBreakLabel(pWInfo)); + + /* End the database scan loop. + */ + sqlite3WhereEnd(pWInfo); + } + }else{ + /* This case when there exist aggregate functions or a GROUP BY clause + ** or both */ + NameContext sNC; /* Name context for processing aggregate information */ + int iAMem; /* First Mem address for storing current GROUP BY */ + int iBMem; /* First Mem address for previous GROUP BY */ + int iUseFlag; /* Mem address holding flag indicating that at least + ** one row of the input to the aggregator has been + ** processed */ + int iAbortFlag; /* Mem address which causes query abort if positive */ + int groupBySort; /* Rows come from source in GROUP BY order */ + int addrEnd; /* End of processing for this SELECT */ + int sortPTab = 0; /* Pseudotable used to decode sorting results */ + int sortOut = 0; /* Output register from the sorter */ + int orderByGrp = 0; /* True if the GROUP BY and ORDER BY are the same */ + + /* Remove any and all aliases between the result set and the + ** GROUP BY clause. + */ + if( pGroupBy ){ + int k; /* Loop counter */ + struct ExprList_item *pItem; /* For looping over expression in a list */ + + for(k=p->pEList->nExpr, pItem=p->pEList->a; k>0; k--, pItem++){ + pItem->u.x.iAlias = 0; + } + for(k=pGroupBy->nExpr, pItem=pGroupBy->a; k>0; k--, pItem++){ + pItem->u.x.iAlias = 0; + } + assert( 66==sqlite3LogEst(100) ); + if( p->nSelectRow>66 ) p->nSelectRow = 66; + + /* If there is both a GROUP BY and an ORDER BY clause and they are + ** identical, then it may be possible to disable the ORDER BY clause + ** on the grounds that the GROUP BY will cause elements to come out + ** in the correct order. It also may not - the GROUP BY might use a + ** database index that causes rows to be grouped together as required + ** but not actually sorted. Either way, record the fact that the + ** ORDER BY and GROUP BY clauses are the same by setting the orderByGrp + ** variable. */ + if( sSort.pOrderBy && pGroupBy->nExpr==sSort.pOrderBy->nExpr ){ + int ii; + /* The GROUP BY processing doesn't care whether rows are delivered in + ** ASC or DESC order - only that each group is returned contiguously. + ** So set the ASC/DESC flags in the GROUP BY to match those in the + ** ORDER BY to maximize the chances of rows being delivered in an + ** order that makes the ORDER BY redundant. */ + for(ii=0; iinExpr; ii++){ + u8 sortFlags = sSort.pOrderBy->a[ii].sortFlags & KEYINFO_ORDER_DESC; + pGroupBy->a[ii].sortFlags = sortFlags; + } + if( sqlite3ExprListCompare(pGroupBy, sSort.pOrderBy, -1)==0 ){ + orderByGrp = 1; + } + } + }else{ + assert( 0==sqlite3LogEst(1) ); + p->nSelectRow = 0; + } + + /* Create a label to jump to when we want to abort the query */ + addrEnd = sqlite3VdbeMakeLabel(pParse); + + /* Convert TK_COLUMN nodes into TK_AGG_COLUMN and make entries in + ** sAggInfo for all TK_AGG_FUNCTION nodes in expressions of the + ** SELECT statement. + */ + memset(&sNC, 0, sizeof(sNC)); + sNC.pParse = pParse; + sNC.pSrcList = pTabList; + sNC.uNC.pAggInfo = &sAggInfo; + VVA_ONLY( sNC.ncFlags = NC_UAggInfo; ) + sAggInfo.mnReg = pParse->nMem+1; + sAggInfo.nSortingColumn = pGroupBy ? pGroupBy->nExpr : 0; + sAggInfo.pGroupBy = pGroupBy; + sqlite3ExprAnalyzeAggList(&sNC, pEList); + sqlite3ExprAnalyzeAggList(&sNC, sSort.pOrderBy); + if( pHaving ){ + if( pGroupBy ){ + assert( pWhere==p->pWhere ); + assert( pHaving==p->pHaving ); + assert( pGroupBy==p->pGroupBy ); + havingToWhere(pParse, p); + pWhere = p->pWhere; + } + sqlite3ExprAnalyzeAggregates(&sNC, pHaving); + } + sAggInfo.nAccumulator = sAggInfo.nColumn; + if( p->pGroupBy==0 && p->pHaving==0 && sAggInfo.nFunc==1 ){ + minMaxFlag = minMaxQuery(db, sAggInfo.aFunc[0].pExpr, &pMinMaxOrderBy); + }else{ + minMaxFlag = WHERE_ORDERBY_NORMAL; + } + for(i=0; ix.pList); +#ifndef SQLITE_OMIT_WINDOWFUNC + assert( !IsWindowFunc(pExpr) ); + if( ExprHasProperty(pExpr, EP_WinFunc) ){ + sqlite3ExprAnalyzeAggregates(&sNC, pExpr->y.pWin->pFilter); + } +#endif + sNC.ncFlags &= ~NC_InAggFunc; + } + sAggInfo.mxReg = pParse->nMem; + if( db->mallocFailed ) goto select_end; +#if SELECTTRACE_ENABLED + if( sqlite3SelectTrace & 0x400 ){ + int ii; + SELECTTRACE(0x400,pParse,p,("After aggregate analysis:\n")); + sqlite3TreeViewSelect(0, p, 0); + for(ii=0; iinTab++; + pKeyInfo = sqlite3KeyInfoFromExprList(pParse,pGroupBy,0,sAggInfo.nColumn); + addrSortingIdx = sqlite3VdbeAddOp4(v, OP_SorterOpen, + sAggInfo.sortingIdx, sAggInfo.nSortingColumn, + 0, (char*)pKeyInfo, P4_KEYINFO); + + /* Initialize memory locations used by GROUP BY aggregate processing + */ + iUseFlag = ++pParse->nMem; + iAbortFlag = ++pParse->nMem; + regOutputRow = ++pParse->nMem; + addrOutputRow = sqlite3VdbeMakeLabel(pParse); + regReset = ++pParse->nMem; + addrReset = sqlite3VdbeMakeLabel(pParse); + iAMem = pParse->nMem + 1; + pParse->nMem += pGroupBy->nExpr; + iBMem = pParse->nMem + 1; + pParse->nMem += pGroupBy->nExpr; + sqlite3VdbeAddOp2(v, OP_Integer, 0, iAbortFlag); + VdbeComment((v, "clear abort flag")); + sqlite3VdbeAddOp3(v, OP_Null, 0, iAMem, iAMem+pGroupBy->nExpr-1); + + /* Begin a loop that will extract all source rows in GROUP BY order. + ** This might involve two separate loops with an OP_Sort in between, or + ** it might be a single loop that uses an index to extract information + ** in the right order to begin with. + */ + sqlite3VdbeAddOp2(v, OP_Gosub, regReset, addrReset); + SELECTTRACE(1,pParse,p,("WhereBegin\n")); + pWInfo = sqlite3WhereBegin(pParse, pTabList, pWhere, pGroupBy, 0, + WHERE_GROUPBY | (orderByGrp ? WHERE_SORTBYGROUP : 0), 0 + ); + if( pWInfo==0 ) goto select_end; + if( sqlite3WhereIsOrdered(pWInfo)==pGroupBy->nExpr ){ + /* The optimizer is able to deliver rows in group by order so + ** we do not have to sort. The OP_OpenEphemeral table will be + ** cancelled later because we still need to use the pKeyInfo + */ + groupBySort = 0; + }else{ + /* Rows are coming out in undetermined order. We have to push + ** each row into a sorting index, terminate the first loop, + ** then loop over the sorting index in order to get the output + ** in sorted order + */ + int regBase; + int regRecord; + int nCol; + int nGroupBy; + + explainTempTable(pParse, + (sDistinct.isTnct && (p->selFlags&SF_Distinct)==0) ? + "DISTINCT" : "GROUP BY"); + + groupBySort = 1; + nGroupBy = pGroupBy->nExpr; + nCol = nGroupBy; + j = nGroupBy; + for(i=0; i=j ){ + nCol++; + j++; + } + } + regBase = sqlite3GetTempRange(pParse, nCol); + sqlite3ExprCodeExprList(pParse, pGroupBy, regBase, 0, 0); + j = nGroupBy; + for(i=0; iiSorterColumn>=j ){ + int r1 = j + regBase; + sqlite3ExprCodeGetColumnOfTable(v, + pCol->pTab, pCol->iTable, pCol->iColumn, r1); + j++; + } + } + regRecord = sqlite3GetTempReg(pParse); + sqlite3VdbeAddOp3(v, OP_MakeRecord, regBase, nCol, regRecord); + sqlite3VdbeAddOp2(v, OP_SorterInsert, sAggInfo.sortingIdx, regRecord); + sqlite3ReleaseTempReg(pParse, regRecord); + sqlite3ReleaseTempRange(pParse, regBase, nCol); + sqlite3WhereEnd(pWInfo); + sAggInfo.sortingIdxPTab = sortPTab = pParse->nTab++; + sortOut = sqlite3GetTempReg(pParse); + sqlite3VdbeAddOp3(v, OP_OpenPseudo, sortPTab, sortOut, nCol); + sqlite3VdbeAddOp2(v, OP_SorterSort, sAggInfo.sortingIdx, addrEnd); + VdbeComment((v, "GROUP BY sort")); VdbeCoverage(v); + sAggInfo.useSortingIdx = 1; + } + + /* If the index or temporary table used by the GROUP BY sort + ** will naturally deliver rows in the order required by the ORDER BY + ** clause, cancel the ephemeral table open coded earlier. + ** + ** This is an optimization - the correct answer should result regardless. + ** Use the SQLITE_GroupByOrder flag with SQLITE_TESTCTRL_OPTIMIZER to + ** disable this optimization for testing purposes. */ + if( orderByGrp && OptimizationEnabled(db, SQLITE_GroupByOrder) + && (groupBySort || sqlite3WhereIsSorted(pWInfo)) + ){ + sSort.pOrderBy = 0; + sqlite3VdbeChangeToNoop(v, sSort.addrSortIndex); + } + + /* Evaluate the current GROUP BY terms and store in b0, b1, b2... + ** (b0 is memory location iBMem+0, b1 is iBMem+1, and so forth) + ** Then compare the current GROUP BY terms against the GROUP BY terms + ** from the previous row currently stored in a0, a1, a2... + */ + addrTopOfLoop = sqlite3VdbeCurrentAddr(v); + if( groupBySort ){ + sqlite3VdbeAddOp3(v, OP_SorterData, sAggInfo.sortingIdx, + sortOut, sortPTab); + } + for(j=0; jnExpr; j++){ + if( groupBySort ){ + sqlite3VdbeAddOp3(v, OP_Column, sortPTab, j, iBMem+j); + }else{ + sAggInfo.directMode = 1; + sqlite3ExprCode(pParse, pGroupBy->a[j].pExpr, iBMem+j); + } + } + sqlite3VdbeAddOp4(v, OP_Compare, iAMem, iBMem, pGroupBy->nExpr, + (char*)sqlite3KeyInfoRef(pKeyInfo), P4_KEYINFO); + addr1 = sqlite3VdbeCurrentAddr(v); + sqlite3VdbeAddOp3(v, OP_Jump, addr1+1, 0, addr1+1); VdbeCoverage(v); + + /* Generate code that runs whenever the GROUP BY changes. + ** Changes in the GROUP BY are detected by the previous code + ** block. If there were no changes, this block is skipped. + ** + ** This code copies current group by terms in b0,b1,b2,... + ** over to a0,a1,a2. It then calls the output subroutine + ** and resets the aggregate accumulator registers in preparation + ** for the next GROUP BY batch. + */ + sqlite3ExprCodeMove(pParse, iBMem, iAMem, pGroupBy->nExpr); + sqlite3VdbeAddOp2(v, OP_Gosub, regOutputRow, addrOutputRow); + VdbeComment((v, "output one row")); + sqlite3VdbeAddOp2(v, OP_IfPos, iAbortFlag, addrEnd); VdbeCoverage(v); + VdbeComment((v, "check abort flag")); + sqlite3VdbeAddOp2(v, OP_Gosub, regReset, addrReset); + VdbeComment((v, "reset accumulator")); + + /* Update the aggregate accumulators based on the content of + ** the current row + */ + sqlite3VdbeJumpHere(v, addr1); + updateAccumulator(pParse, iUseFlag, &sAggInfo); + sqlite3VdbeAddOp2(v, OP_Integer, 1, iUseFlag); + VdbeComment((v, "indicate data in accumulator")); + + /* End of the loop + */ + if( groupBySort ){ + sqlite3VdbeAddOp2(v, OP_SorterNext, sAggInfo.sortingIdx, addrTopOfLoop); + VdbeCoverage(v); + }else{ + sqlite3WhereEnd(pWInfo); + sqlite3VdbeChangeToNoop(v, addrSortingIdx); + } + + /* Output the final row of result + */ + sqlite3VdbeAddOp2(v, OP_Gosub, regOutputRow, addrOutputRow); + VdbeComment((v, "output final row")); + + /* Jump over the subroutines + */ + sqlite3VdbeGoto(v, addrEnd); + + /* Generate a subroutine that outputs a single row of the result + ** set. This subroutine first looks at the iUseFlag. If iUseFlag + ** is less than or equal to zero, the subroutine is a no-op. If + ** the processing calls for the query to abort, this subroutine + ** increments the iAbortFlag memory location before returning in + ** order to signal the caller to abort. + */ + addrSetAbort = sqlite3VdbeCurrentAddr(v); + sqlite3VdbeAddOp2(v, OP_Integer, 1, iAbortFlag); + VdbeComment((v, "set abort flag")); + sqlite3VdbeAddOp1(v, OP_Return, regOutputRow); + sqlite3VdbeResolveLabel(v, addrOutputRow); + addrOutputRow = sqlite3VdbeCurrentAddr(v); + sqlite3VdbeAddOp2(v, OP_IfPos, iUseFlag, addrOutputRow+2); + VdbeCoverage(v); + VdbeComment((v, "Groupby result generator entry point")); + sqlite3VdbeAddOp1(v, OP_Return, regOutputRow); + finalizeAggFunctions(pParse, &sAggInfo); + sqlite3ExprIfFalse(pParse, pHaving, addrOutputRow+1, SQLITE_JUMPIFNULL); + selectInnerLoop(pParse, p, -1, &sSort, + &sDistinct, pDest, + addrOutputRow+1, addrSetAbort); + sqlite3VdbeAddOp1(v, OP_Return, regOutputRow); + VdbeComment((v, "end groupby result generator")); + + /* Generate a subroutine that will reset the group-by accumulator + */ + sqlite3VdbeResolveLabel(v, addrReset); + resetAccumulator(pParse, &sAggInfo); + sqlite3VdbeAddOp2(v, OP_Integer, 0, iUseFlag); + VdbeComment((v, "indicate accumulator empty")); + sqlite3VdbeAddOp1(v, OP_Return, regReset); + + } /* endif pGroupBy. Begin aggregate queries without GROUP BY: */ + else { +#ifndef SQLITE_OMIT_BTREECOUNT + Table *pTab; + if( (pTab = isSimpleCount(p, &sAggInfo))!=0 ){ + /* If isSimpleCount() returns a pointer to a Table structure, then + ** the SQL statement is of the form: + ** + ** SELECT count(*) FROM + ** + ** where the Table structure returned represents table . + ** + ** This statement is so common that it is optimized specially. The + ** OP_Count instruction is executed either on the intkey table that + ** contains the data for table or on one of its indexes. It + ** is better to execute the op on an index, as indexes are almost + ** always spread across less pages than their corresponding tables. + */ + const int iDb = sqlite3SchemaToIndex(pParse->db, pTab->pSchema); + const int iCsr = pParse->nTab++; /* Cursor to scan b-tree */ + Index *pIdx; /* Iterator variable */ + KeyInfo *pKeyInfo = 0; /* Keyinfo for scanned index */ + Index *pBest = 0; /* Best index found so far */ + int iRoot = pTab->tnum; /* Root page of scanned b-tree */ + + sqlite3CodeVerifySchema(pParse, iDb); + sqlite3TableLock(pParse, iDb, pTab->tnum, 0, pTab->zName); + + /* Search for the index that has the lowest scan cost. + ** + ** (2011-04-15) Do not do a full scan of an unordered index. + ** + ** (2013-10-03) Do not count the entries in a partial index. + ** + ** In practice the KeyInfo structure will not be used. It is only + ** passed to keep OP_OpenRead happy. + */ + if( !HasRowid(pTab) ) pBest = sqlite3PrimaryKeyIndex(pTab); + for(pIdx=pTab->pIndex; pIdx; pIdx=pIdx->pNext){ + if( pIdx->bUnordered==0 + && pIdx->szIdxRowszTabRow + && pIdx->pPartIdxWhere==0 + && (!pBest || pIdx->szIdxRowszIdxRow) + ){ + pBest = pIdx; + } + } + if( pBest ){ + iRoot = pBest->tnum; + pKeyInfo = sqlite3KeyInfoOfIndex(pParse, pBest); + } + + /* Open a read-only cursor, execute the OP_Count, close the cursor. */ + sqlite3VdbeAddOp4Int(v, OP_OpenRead, iCsr, iRoot, iDb, 1); + if( pKeyInfo ){ + sqlite3VdbeChangeP4(v, -1, (char *)pKeyInfo, P4_KEYINFO); + } + sqlite3VdbeAddOp2(v, OP_Count, iCsr, sAggInfo.aFunc[0].iMem); + sqlite3VdbeAddOp1(v, OP_Close, iCsr); + explainSimpleCount(pParse, pTab, pBest); + }else +#endif /* SQLITE_OMIT_BTREECOUNT */ + { + int regAcc = 0; /* "populate accumulators" flag */ + + /* If there are accumulator registers but no min() or max() functions + ** without FILTER clauses, allocate register regAcc. Register regAcc + ** will contain 0 the first time the inner loop runs, and 1 thereafter. + ** The code generated by updateAccumulator() uses this to ensure + ** that the accumulator registers are (a) updated only once if + ** there are no min() or max functions or (b) always updated for the + ** first row visited by the aggregate, so that they are updated at + ** least once even if the FILTER clause means the min() or max() + ** function visits zero rows. */ + if( sAggInfo.nAccumulator ){ + for(i=0; ifuncFlags&SQLITE_FUNC_NEEDCOLL ) break; + } + if( i==sAggInfo.nFunc ){ + regAcc = ++pParse->nMem; + sqlite3VdbeAddOp2(v, OP_Integer, 0, regAcc); + } + } + + /* This case runs if the aggregate has no GROUP BY clause. The + ** processing is much simpler since there is only a single row + ** of output. + */ + assert( p->pGroupBy==0 ); + resetAccumulator(pParse, &sAggInfo); + + /* If this query is a candidate for the min/max optimization, then + ** minMaxFlag will have been previously set to either + ** WHERE_ORDERBY_MIN or WHERE_ORDERBY_MAX and pMinMaxOrderBy will + ** be an appropriate ORDER BY expression for the optimization. + */ + assert( minMaxFlag==WHERE_ORDERBY_NORMAL || pMinMaxOrderBy!=0 ); + assert( pMinMaxOrderBy==0 || pMinMaxOrderBy->nExpr==1 ); + + SELECTTRACE(1,pParse,p,("WhereBegin\n")); + pWInfo = sqlite3WhereBegin(pParse, pTabList, pWhere, pMinMaxOrderBy, + 0, minMaxFlag, 0); + if( pWInfo==0 ){ + goto select_end; + } + updateAccumulator(pParse, regAcc, &sAggInfo); + if( regAcc ) sqlite3VdbeAddOp2(v, OP_Integer, 1, regAcc); + if( sqlite3WhereIsOrdered(pWInfo)>0 ){ + sqlite3VdbeGoto(v, sqlite3WhereBreakLabel(pWInfo)); + VdbeComment((v, "%s() by index", + (minMaxFlag==WHERE_ORDERBY_MIN?"min":"max"))); + } + sqlite3WhereEnd(pWInfo); + finalizeAggFunctions(pParse, &sAggInfo); + } + + sSort.pOrderBy = 0; + sqlite3ExprIfFalse(pParse, pHaving, addrEnd, SQLITE_JUMPIFNULL); + selectInnerLoop(pParse, p, -1, 0, 0, + pDest, addrEnd, addrEnd); + } + sqlite3VdbeResolveLabel(v, addrEnd); + + } /* endif aggregate query */ + + if( sDistinct.eTnctType==WHERE_DISTINCT_UNORDERED ){ + explainTempTable(pParse, "DISTINCT"); + } + + /* If there is an ORDER BY clause, then we need to sort the results + ** and send them to the callback one by one. + */ + if( sSort.pOrderBy ){ + explainTempTable(pParse, + sSort.nOBSat>0 ? "RIGHT PART OF ORDER BY":"ORDER BY"); + assert( p->pEList==pEList ); + generateSortTail(pParse, p, &sSort, pEList->nExpr, pDest); + } + + /* Jump here to skip this query + */ + sqlite3VdbeResolveLabel(v, iEnd); + + /* The SELECT has been coded. If there is an error in the Parse structure, + ** set the return code to 1. Otherwise 0. */ + rc = (pParse->nErr>0); + + /* Control jumps to here if an error is encountered above, or upon + ** successful coding of the SELECT. + */ +select_end: + sqlite3ExprListDelete(db, pMinMaxOrderBy); + sqlite3DbFree(db, sAggInfo.aCol); + sqlite3DbFree(db, sAggInfo.aFunc); +#if SELECTTRACE_ENABLED + SELECTTRACE(0x1,pParse,p,("end processing\n")); + if( (sqlite3SelectTrace & 0x2000)!=0 && ExplainQueryPlanParent(pParse)==0 ){ + sqlite3TreeViewSelect(0, p, 0); + } +#endif + ExplainQueryPlanPop(pParse); + return rc; +} + +/************** End of select.c **********************************************/ +/************** Begin file table.c *******************************************/ +/* +** 2001 September 15 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** This file contains the sqlite3_get_table() and sqlite3_free_table() +** interface routines. These are just wrappers around the main +** interface routine of sqlite3_exec(). +** +** These routines are in a separate files so that they will not be linked +** if they are not used. +*/ +/* #include "sqliteInt.h" */ + +#ifndef SQLITE_OMIT_GET_TABLE + +/* +** This structure is used to pass data from sqlite3_get_table() through +** to the callback function is uses to build the result. +*/ +typedef struct TabResult { + char **azResult; /* Accumulated output */ + char *zErrMsg; /* Error message text, if an error occurs */ + u32 nAlloc; /* Slots allocated for azResult[] */ + u32 nRow; /* Number of rows in the result */ + u32 nColumn; /* Number of columns in the result */ + u32 nData; /* Slots used in azResult[]. (nRow+1)*nColumn */ + int rc; /* Return code from sqlite3_exec() */ +} TabResult; + +/* +** This routine is called once for each row in the result table. Its job +** is to fill in the TabResult structure appropriately, allocating new +** memory as necessary. +*/ +static int sqlite3_get_table_cb(void *pArg, int nCol, char **argv, char **colv){ + TabResult *p = (TabResult*)pArg; /* Result accumulator */ + int need; /* Slots needed in p->azResult[] */ + int i; /* Loop counter */ + char *z; /* A single column of result */ + + /* Make sure there is enough space in p->azResult to hold everything + ** we need to remember from this invocation of the callback. + */ + if( p->nRow==0 && argv!=0 ){ + need = nCol*2; + }else{ + need = nCol; + } + if( p->nData + need > p->nAlloc ){ + char **azNew; + p->nAlloc = p->nAlloc*2 + need; + azNew = sqlite3_realloc64( p->azResult, sizeof(char*)*p->nAlloc ); + if( azNew==0 ) goto malloc_failed; + p->azResult = azNew; + } + + /* If this is the first row, then generate an extra row containing + ** the names of all columns. + */ + if( p->nRow==0 ){ + p->nColumn = nCol; + for(i=0; iazResult[p->nData++] = z; + } + }else if( (int)p->nColumn!=nCol ){ + sqlite3_free(p->zErrMsg); + p->zErrMsg = sqlite3_mprintf( + "sqlite3_get_table() called with two or more incompatible queries" + ); + p->rc = SQLITE_ERROR; + return 1; + } + + /* Copy over the row data + */ + if( argv!=0 ){ + for(i=0; iazResult[p->nData++] = z; + } + p->nRow++; + } + return 0; + +malloc_failed: + p->rc = SQLITE_NOMEM_BKPT; + return 1; +} + +/* +** Query the database. But instead of invoking a callback for each row, +** malloc() for space to hold the result and return the entire results +** at the conclusion of the call. +** +** The result that is written to ***pazResult is held in memory obtained +** from malloc(). But the caller cannot free this memory directly. +** Instead, the entire table should be passed to sqlite3_free_table() when +** the calling procedure is finished using it. +*/ +SQLITE_API int sqlite3_get_table( + sqlite3 *db, /* The database on which the SQL executes */ + const char *zSql, /* The SQL to be executed */ + char ***pazResult, /* Write the result table here */ + int *pnRow, /* Write the number of rows in the result here */ + int *pnColumn, /* Write the number of columns of result here */ + char **pzErrMsg /* Write error messages here */ +){ + int rc; + TabResult res; + +#ifdef SQLITE_ENABLE_API_ARMOR + if( !sqlite3SafetyCheckOk(db) || pazResult==0 ) return SQLITE_MISUSE_BKPT; +#endif + *pazResult = 0; + if( pnColumn ) *pnColumn = 0; + if( pnRow ) *pnRow = 0; + if( pzErrMsg ) *pzErrMsg = 0; + res.zErrMsg = 0; + res.nRow = 0; + res.nColumn = 0; + res.nData = 1; + res.nAlloc = 20; + res.rc = SQLITE_OK; + res.azResult = sqlite3_malloc64(sizeof(char*)*res.nAlloc ); + if( res.azResult==0 ){ + db->errCode = SQLITE_NOMEM; + return SQLITE_NOMEM_BKPT; + } + res.azResult[0] = 0; + rc = sqlite3_exec(db, zSql, sqlite3_get_table_cb, &res, pzErrMsg); + assert( sizeof(res.azResult[0])>= sizeof(res.nData) ); + res.azResult[0] = SQLITE_INT_TO_PTR(res.nData); + if( (rc&0xff)==SQLITE_ABORT ){ + sqlite3_free_table(&res.azResult[1]); + if( res.zErrMsg ){ + if( pzErrMsg ){ + sqlite3_free(*pzErrMsg); + *pzErrMsg = sqlite3_mprintf("%s",res.zErrMsg); + } + sqlite3_free(res.zErrMsg); + } + db->errCode = res.rc; /* Assume 32-bit assignment is atomic */ + return res.rc; + } + sqlite3_free(res.zErrMsg); + if( rc!=SQLITE_OK ){ + sqlite3_free_table(&res.azResult[1]); + return rc; + } + if( res.nAlloc>res.nData ){ + char **azNew; + azNew = sqlite3_realloc64( res.azResult, sizeof(char*)*res.nData ); + if( azNew==0 ){ + sqlite3_free_table(&res.azResult[1]); + db->errCode = SQLITE_NOMEM; + return SQLITE_NOMEM_BKPT; + } + res.azResult = azNew; + } + *pazResult = &res.azResult[1]; + if( pnColumn ) *pnColumn = res.nColumn; + if( pnRow ) *pnRow = res.nRow; + return rc; +} + +/* +** This routine frees the space the sqlite3_get_table() malloced. +*/ +SQLITE_API void sqlite3_free_table( + char **azResult /* Result returned from sqlite3_get_table() */ +){ + if( azResult ){ + int i, n; + azResult--; + assert( azResult!=0 ); + n = SQLITE_PTR_TO_INT(azResult[0]); + for(i=1; ipNext; + + sqlite3ExprDelete(db, pTmp->pWhere); + sqlite3ExprListDelete(db, pTmp->pExprList); + sqlite3SelectDelete(db, pTmp->pSelect); + sqlite3IdListDelete(db, pTmp->pIdList); + sqlite3UpsertDelete(db, pTmp->pUpsert); + sqlite3DbFree(db, pTmp->zSpan); + + sqlite3DbFree(db, pTmp); + } +} + +/* +** Given table pTab, return a list of all the triggers attached to +** the table. The list is connected by Trigger.pNext pointers. +** +** All of the triggers on pTab that are in the same database as pTab +** are already attached to pTab->pTrigger. But there might be additional +** triggers on pTab in the TEMP schema. This routine prepends all +** TEMP triggers on pTab to the beginning of the pTab->pTrigger list +** and returns the combined list. +** +** To state it another way: This routine returns a list of all triggers +** that fire off of pTab. The list will include any TEMP triggers on +** pTab as well as the triggers lised in pTab->pTrigger. +*/ +SQLITE_PRIVATE Trigger *sqlite3TriggerList(Parse *pParse, Table *pTab){ + Schema * const pTmpSchema = pParse->db->aDb[1].pSchema; + Trigger *pList = 0; /* List of triggers to return */ + + if( pParse->disableTriggers ){ + return 0; + } + + if( pTmpSchema!=pTab->pSchema ){ + HashElem *p; + assert( sqlite3SchemaMutexHeld(pParse->db, 0, pTmpSchema) ); + for(p=sqliteHashFirst(&pTmpSchema->trigHash); p; p=sqliteHashNext(p)){ + Trigger *pTrig = (Trigger *)sqliteHashData(p); + if( pTrig->pTabSchema==pTab->pSchema + && 0==sqlite3StrICmp(pTrig->table, pTab->zName) + ){ + pTrig->pNext = (pList ? pList : pTab->pTrigger); + pList = pTrig; + } + } + } + + return (pList ? pList : pTab->pTrigger); +} + +/* +** This is called by the parser when it sees a CREATE TRIGGER statement +** up to the point of the BEGIN before the trigger actions. A Trigger +** structure is generated based on the information available and stored +** in pParse->pNewTrigger. After the trigger actions have been parsed, the +** sqlite3FinishTrigger() function is called to complete the trigger +** construction process. +*/ +SQLITE_PRIVATE void sqlite3BeginTrigger( + Parse *pParse, /* The parse context of the CREATE TRIGGER statement */ + Token *pName1, /* The name of the trigger */ + Token *pName2, /* The name of the trigger */ + int tr_tm, /* One of TK_BEFORE, TK_AFTER, TK_INSTEAD */ + int op, /* One of TK_INSERT, TK_UPDATE, TK_DELETE */ + IdList *pColumns, /* column list if this is an UPDATE OF trigger */ + SrcList *pTableName,/* The name of the table/view the trigger applies to */ + Expr *pWhen, /* WHEN clause */ + int isTemp, /* True if the TEMPORARY keyword is present */ + int noErr /* Suppress errors if the trigger already exists */ +){ + Trigger *pTrigger = 0; /* The new trigger */ + Table *pTab; /* Table that the trigger fires off of */ + char *zName = 0; /* Name of the trigger */ + sqlite3 *db = pParse->db; /* The database connection */ + int iDb; /* The database to store the trigger in */ + Token *pName; /* The unqualified db name */ + DbFixer sFix; /* State vector for the DB fixer */ + + assert( pName1!=0 ); /* pName1->z might be NULL, but not pName1 itself */ + assert( pName2!=0 ); + assert( op==TK_INSERT || op==TK_UPDATE || op==TK_DELETE ); + assert( op>0 && op<0xff ); + if( isTemp ){ + /* If TEMP was specified, then the trigger name may not be qualified. */ + if( pName2->n>0 ){ + sqlite3ErrorMsg(pParse, "temporary trigger may not have qualified name"); + goto trigger_cleanup; + } + iDb = 1; + pName = pName1; + }else{ + /* Figure out the db that the trigger will be created in */ + iDb = sqlite3TwoPartName(pParse, pName1, pName2, &pName); + if( iDb<0 ){ + goto trigger_cleanup; + } + } + if( !pTableName || db->mallocFailed ){ + goto trigger_cleanup; + } + + /* A long-standing parser bug is that this syntax was allowed: + ** + ** CREATE TRIGGER attached.demo AFTER INSERT ON attached.tab .... + ** ^^^^^^^^ + ** + ** To maintain backwards compatibility, ignore the database + ** name on pTableName if we are reparsing out of SQLITE_MASTER. + */ + if( db->init.busy && iDb!=1 ){ + sqlite3DbFree(db, pTableName->a[0].zDatabase); + pTableName->a[0].zDatabase = 0; + } + + /* If the trigger name was unqualified, and the table is a temp table, + ** then set iDb to 1 to create the trigger in the temporary database. + ** If sqlite3SrcListLookup() returns 0, indicating the table does not + ** exist, the error is caught by the block below. + */ + pTab = sqlite3SrcListLookup(pParse, pTableName); + if( db->init.busy==0 && pName2->n==0 && pTab + && pTab->pSchema==db->aDb[1].pSchema ){ + iDb = 1; + } + + /* Ensure the table name matches database name and that the table exists */ + if( db->mallocFailed ) goto trigger_cleanup; + assert( pTableName->nSrc==1 ); + sqlite3FixInit(&sFix, pParse, iDb, "trigger", pName); + if( sqlite3FixSrcList(&sFix, pTableName) ){ + goto trigger_cleanup; + } + pTab = sqlite3SrcListLookup(pParse, pTableName); + if( !pTab ){ + /* The table does not exist. */ + if( db->init.iDb==1 ){ + /* Ticket #3810. + ** Normally, whenever a table is dropped, all associated triggers are + ** dropped too. But if a TEMP trigger is created on a non-TEMP table + ** and the table is dropped by a different database connection, the + ** trigger is not visible to the database connection that does the + ** drop so the trigger cannot be dropped. This results in an + ** "orphaned trigger" - a trigger whose associated table is missing. + */ + db->init.orphanTrigger = 1; + } + goto trigger_cleanup; + } + if( IsVirtual(pTab) ){ + sqlite3ErrorMsg(pParse, "cannot create triggers on virtual tables"); + goto trigger_cleanup; + } + + /* Check that the trigger name is not reserved and that no trigger of the + ** specified name exists */ + zName = sqlite3NameFromToken(db, pName); + if( zName==0 ){ + assert( db->mallocFailed ); + goto trigger_cleanup; + } + if( sqlite3CheckObjectName(pParse, zName, "trigger", pTab->zName) ){ + goto trigger_cleanup; + } + assert( sqlite3SchemaMutexHeld(db, iDb, 0) ); + if( !IN_RENAME_OBJECT ){ + if( sqlite3HashFind(&(db->aDb[iDb].pSchema->trigHash),zName) ){ + if( !noErr ){ + sqlite3ErrorMsg(pParse, "trigger %T already exists", pName); + }else{ + assert( !db->init.busy ); + sqlite3CodeVerifySchema(pParse, iDb); + } + goto trigger_cleanup; + } + } + + /* Do not create a trigger on a system table */ + if( sqlite3StrNICmp(pTab->zName, "sqlite_", 7)==0 ){ + sqlite3ErrorMsg(pParse, "cannot create trigger on system table"); + goto trigger_cleanup; + } + + /* INSTEAD of triggers are only for views and views only support INSTEAD + ** of triggers. + */ + if( pTab->pSelect && tr_tm!=TK_INSTEAD ){ + sqlite3ErrorMsg(pParse, "cannot create %s trigger on view: %S", + (tr_tm == TK_BEFORE)?"BEFORE":"AFTER", pTableName, 0); + goto trigger_cleanup; + } + if( !pTab->pSelect && tr_tm==TK_INSTEAD ){ + sqlite3ErrorMsg(pParse, "cannot create INSTEAD OF" + " trigger on table: %S", pTableName, 0); + goto trigger_cleanup; + } + +#ifndef SQLITE_OMIT_AUTHORIZATION + if( !IN_RENAME_OBJECT ){ + int iTabDb = sqlite3SchemaToIndex(db, pTab->pSchema); + int code = SQLITE_CREATE_TRIGGER; + const char *zDb = db->aDb[iTabDb].zDbSName; + const char *zDbTrig = isTemp ? db->aDb[1].zDbSName : zDb; + if( iTabDb==1 || isTemp ) code = SQLITE_CREATE_TEMP_TRIGGER; + if( sqlite3AuthCheck(pParse, code, zName, pTab->zName, zDbTrig) ){ + goto trigger_cleanup; + } + if( sqlite3AuthCheck(pParse, SQLITE_INSERT, SCHEMA_TABLE(iTabDb),0,zDb)){ + goto trigger_cleanup; + } + } +#endif + + /* INSTEAD OF triggers can only appear on views and BEFORE triggers + ** cannot appear on views. So we might as well translate every + ** INSTEAD OF trigger into a BEFORE trigger. It simplifies code + ** elsewhere. + */ + if (tr_tm == TK_INSTEAD){ + tr_tm = TK_BEFORE; + } + + /* Build the Trigger object */ + pTrigger = (Trigger*)sqlite3DbMallocZero(db, sizeof(Trigger)); + if( pTrigger==0 ) goto trigger_cleanup; + pTrigger->zName = zName; + zName = 0; + pTrigger->table = sqlite3DbStrDup(db, pTableName->a[0].zName); + pTrigger->pSchema = db->aDb[iDb].pSchema; + pTrigger->pTabSchema = pTab->pSchema; + pTrigger->op = (u8)op; + pTrigger->tr_tm = tr_tm==TK_BEFORE ? TRIGGER_BEFORE : TRIGGER_AFTER; + if( IN_RENAME_OBJECT ){ + sqlite3RenameTokenRemap(pParse, pTrigger->table, pTableName->a[0].zName); + pTrigger->pWhen = pWhen; + pWhen = 0; + }else{ + pTrigger->pWhen = sqlite3ExprDup(db, pWhen, EXPRDUP_REDUCE); + } + pTrigger->pColumns = pColumns; + pColumns = 0; + assert( pParse->pNewTrigger==0 ); + pParse->pNewTrigger = pTrigger; + +trigger_cleanup: + sqlite3DbFree(db, zName); + sqlite3SrcListDelete(db, pTableName); + sqlite3IdListDelete(db, pColumns); + sqlite3ExprDelete(db, pWhen); + if( !pParse->pNewTrigger ){ + sqlite3DeleteTrigger(db, pTrigger); + }else{ + assert( pParse->pNewTrigger==pTrigger ); + } +} + +/* +** This routine is called after all of the trigger actions have been parsed +** in order to complete the process of building the trigger. +*/ +SQLITE_PRIVATE void sqlite3FinishTrigger( + Parse *pParse, /* Parser context */ + TriggerStep *pStepList, /* The triggered program */ + Token *pAll /* Token that describes the complete CREATE TRIGGER */ +){ + Trigger *pTrig = pParse->pNewTrigger; /* Trigger being finished */ + char *zName; /* Name of trigger */ + sqlite3 *db = pParse->db; /* The database */ + DbFixer sFix; /* Fixer object */ + int iDb; /* Database containing the trigger */ + Token nameToken; /* Trigger name for error reporting */ + + pParse->pNewTrigger = 0; + if( NEVER(pParse->nErr) || !pTrig ) goto triggerfinish_cleanup; + zName = pTrig->zName; + iDb = sqlite3SchemaToIndex(pParse->db, pTrig->pSchema); + pTrig->step_list = pStepList; + while( pStepList ){ + pStepList->pTrig = pTrig; + pStepList = pStepList->pNext; + } + sqlite3TokenInit(&nameToken, pTrig->zName); + sqlite3FixInit(&sFix, pParse, iDb, "trigger", &nameToken); + if( sqlite3FixTriggerStep(&sFix, pTrig->step_list) + || sqlite3FixExpr(&sFix, pTrig->pWhen) + ){ + goto triggerfinish_cleanup; + } + +#ifndef SQLITE_OMIT_ALTERTABLE + if( IN_RENAME_OBJECT ){ + assert( !db->init.busy ); + pParse->pNewTrigger = pTrig; + pTrig = 0; + }else +#endif + + /* if we are not initializing, + ** build the sqlite_master entry + */ + if( !db->init.busy ){ + Vdbe *v; + char *z; + + /* Make an entry in the sqlite_master table */ + v = sqlite3GetVdbe(pParse); + if( v==0 ) goto triggerfinish_cleanup; + sqlite3BeginWriteOperation(pParse, 0, iDb); + z = sqlite3DbStrNDup(db, (char*)pAll->z, pAll->n); + testcase( z==0 ); + sqlite3NestedParse(pParse, + "INSERT INTO %Q.%s VALUES('trigger',%Q,%Q,0,'CREATE TRIGGER %q')", + db->aDb[iDb].zDbSName, MASTER_NAME, zName, + pTrig->table, z); + sqlite3DbFree(db, z); + sqlite3ChangeCookie(pParse, iDb); + sqlite3VdbeAddParseSchemaOp(v, iDb, + sqlite3MPrintf(db, "type='trigger' AND name='%q'", zName)); + } + + if( db->init.busy ){ + Trigger *pLink = pTrig; + Hash *pHash = &db->aDb[iDb].pSchema->trigHash; + assert( sqlite3SchemaMutexHeld(db, iDb, 0) ); + assert( pLink!=0 ); + pTrig = sqlite3HashInsert(pHash, zName, pTrig); + if( pTrig ){ + sqlite3OomFault(db); + }else if( pLink->pSchema==pLink->pTabSchema ){ + Table *pTab; + pTab = sqlite3HashFind(&pLink->pTabSchema->tblHash, pLink->table); + assert( pTab!=0 ); + pLink->pNext = pTab->pTrigger; + pTab->pTrigger = pLink; + } + } + +triggerfinish_cleanup: + sqlite3DeleteTrigger(db, pTrig); + assert( IN_RENAME_OBJECT || !pParse->pNewTrigger ); + sqlite3DeleteTriggerStep(db, pStepList); +} + +/* +** Duplicate a range of text from an SQL statement, then convert all +** whitespace characters into ordinary space characters. +*/ +static char *triggerSpanDup(sqlite3 *db, const char *zStart, const char *zEnd){ + char *z = sqlite3DbSpanDup(db, zStart, zEnd); + int i; + if( z ) for(i=0; z[i]; i++) if( sqlite3Isspace(z[i]) ) z[i] = ' '; + return z; +} + +/* +** Turn a SELECT statement (that the pSelect parameter points to) into +** a trigger step. Return a pointer to a TriggerStep structure. +** +** The parser calls this routine when it finds a SELECT statement in +** body of a TRIGGER. +*/ +SQLITE_PRIVATE TriggerStep *sqlite3TriggerSelectStep( + sqlite3 *db, /* Database connection */ + Select *pSelect, /* The SELECT statement */ + const char *zStart, /* Start of SQL text */ + const char *zEnd /* End of SQL text */ +){ + TriggerStep *pTriggerStep = sqlite3DbMallocZero(db, sizeof(TriggerStep)); + if( pTriggerStep==0 ) { + sqlite3SelectDelete(db, pSelect); + return 0; + } + pTriggerStep->op = TK_SELECT; + pTriggerStep->pSelect = pSelect; + pTriggerStep->orconf = OE_Default; + pTriggerStep->zSpan = triggerSpanDup(db, zStart, zEnd); + return pTriggerStep; +} + +/* +** Allocate space to hold a new trigger step. The allocated space +** holds both the TriggerStep object and the TriggerStep.target.z string. +** +** If an OOM error occurs, NULL is returned and db->mallocFailed is set. +*/ +static TriggerStep *triggerStepAllocate( + Parse *pParse, /* Parser context */ + u8 op, /* Trigger opcode */ + Token *pName, /* The target name */ + const char *zStart, /* Start of SQL text */ + const char *zEnd /* End of SQL text */ +){ + sqlite3 *db = pParse->db; + TriggerStep *pTriggerStep; + + pTriggerStep = sqlite3DbMallocZero(db, sizeof(TriggerStep) + pName->n + 1); + if( pTriggerStep ){ + char *z = (char*)&pTriggerStep[1]; + memcpy(z, pName->z, pName->n); + sqlite3Dequote(z); + pTriggerStep->zTarget = z; + pTriggerStep->op = op; + pTriggerStep->zSpan = triggerSpanDup(db, zStart, zEnd); + if( IN_RENAME_OBJECT ){ + sqlite3RenameTokenMap(pParse, pTriggerStep->zTarget, pName); + } + } + return pTriggerStep; +} + +/* +** Build a trigger step out of an INSERT statement. Return a pointer +** to the new trigger step. +** +** The parser calls this routine when it sees an INSERT inside the +** body of a trigger. +*/ +SQLITE_PRIVATE TriggerStep *sqlite3TriggerInsertStep( + Parse *pParse, /* Parser */ + Token *pTableName, /* Name of the table into which we insert */ + IdList *pColumn, /* List of columns in pTableName to insert into */ + Select *pSelect, /* A SELECT statement that supplies values */ + u8 orconf, /* The conflict algorithm (OE_Abort, OE_Replace, etc.) */ + Upsert *pUpsert, /* ON CONFLICT clauses for upsert */ + const char *zStart, /* Start of SQL text */ + const char *zEnd /* End of SQL text */ +){ + sqlite3 *db = pParse->db; + TriggerStep *pTriggerStep; + + assert(pSelect != 0 || db->mallocFailed); + + pTriggerStep = triggerStepAllocate(pParse, TK_INSERT, pTableName,zStart,zEnd); + if( pTriggerStep ){ + if( IN_RENAME_OBJECT ){ + pTriggerStep->pSelect = pSelect; + pSelect = 0; + }else{ + pTriggerStep->pSelect = sqlite3SelectDup(db, pSelect, EXPRDUP_REDUCE); + } + pTriggerStep->pIdList = pColumn; + pTriggerStep->pUpsert = pUpsert; + pTriggerStep->orconf = orconf; + if( pUpsert ){ + sqlite3HasExplicitNulls(pParse, pUpsert->pUpsertTarget); + } + }else{ + testcase( pColumn ); + sqlite3IdListDelete(db, pColumn); + testcase( pUpsert ); + sqlite3UpsertDelete(db, pUpsert); + } + sqlite3SelectDelete(db, pSelect); + + return pTriggerStep; +} + +/* +** Construct a trigger step that implements an UPDATE statement and return +** a pointer to that trigger step. The parser calls this routine when it +** sees an UPDATE statement inside the body of a CREATE TRIGGER. +*/ +SQLITE_PRIVATE TriggerStep *sqlite3TriggerUpdateStep( + Parse *pParse, /* Parser */ + Token *pTableName, /* Name of the table to be updated */ + ExprList *pEList, /* The SET clause: list of column and new values */ + Expr *pWhere, /* The WHERE clause */ + u8 orconf, /* The conflict algorithm. (OE_Abort, OE_Ignore, etc) */ + const char *zStart, /* Start of SQL text */ + const char *zEnd /* End of SQL text */ +){ + sqlite3 *db = pParse->db; + TriggerStep *pTriggerStep; + + pTriggerStep = triggerStepAllocate(pParse, TK_UPDATE, pTableName,zStart,zEnd); + if( pTriggerStep ){ + if( IN_RENAME_OBJECT ){ + pTriggerStep->pExprList = pEList; + pTriggerStep->pWhere = pWhere; + pEList = 0; + pWhere = 0; + }else{ + pTriggerStep->pExprList = sqlite3ExprListDup(db, pEList, EXPRDUP_REDUCE); + pTriggerStep->pWhere = sqlite3ExprDup(db, pWhere, EXPRDUP_REDUCE); + } + pTriggerStep->orconf = orconf; + } + sqlite3ExprListDelete(db, pEList); + sqlite3ExprDelete(db, pWhere); + return pTriggerStep; +} + +/* +** Construct a trigger step that implements a DELETE statement and return +** a pointer to that trigger step. The parser calls this routine when it +** sees a DELETE statement inside the body of a CREATE TRIGGER. +*/ +SQLITE_PRIVATE TriggerStep *sqlite3TriggerDeleteStep( + Parse *pParse, /* Parser */ + Token *pTableName, /* The table from which rows are deleted */ + Expr *pWhere, /* The WHERE clause */ + const char *zStart, /* Start of SQL text */ + const char *zEnd /* End of SQL text */ +){ + sqlite3 *db = pParse->db; + TriggerStep *pTriggerStep; + + pTriggerStep = triggerStepAllocate(pParse, TK_DELETE, pTableName,zStart,zEnd); + if( pTriggerStep ){ + if( IN_RENAME_OBJECT ){ + pTriggerStep->pWhere = pWhere; + pWhere = 0; + }else{ + pTriggerStep->pWhere = sqlite3ExprDup(db, pWhere, EXPRDUP_REDUCE); + } + pTriggerStep->orconf = OE_Default; + } + sqlite3ExprDelete(db, pWhere); + return pTriggerStep; +} + +/* +** Recursively delete a Trigger structure +*/ +SQLITE_PRIVATE void sqlite3DeleteTrigger(sqlite3 *db, Trigger *pTrigger){ + if( pTrigger==0 ) return; + sqlite3DeleteTriggerStep(db, pTrigger->step_list); + sqlite3DbFree(db, pTrigger->zName); + sqlite3DbFree(db, pTrigger->table); + sqlite3ExprDelete(db, pTrigger->pWhen); + sqlite3IdListDelete(db, pTrigger->pColumns); + sqlite3DbFree(db, pTrigger); +} + +/* +** This function is called to drop a trigger from the database schema. +** +** This may be called directly from the parser and therefore identifies +** the trigger by name. The sqlite3DropTriggerPtr() routine does the +** same job as this routine except it takes a pointer to the trigger +** instead of the trigger name. +**/ +SQLITE_PRIVATE void sqlite3DropTrigger(Parse *pParse, SrcList *pName, int noErr){ + Trigger *pTrigger = 0; + int i; + const char *zDb; + const char *zName; + sqlite3 *db = pParse->db; + + if( db->mallocFailed ) goto drop_trigger_cleanup; + if( SQLITE_OK!=sqlite3ReadSchema(pParse) ){ + goto drop_trigger_cleanup; + } + + assert( pName->nSrc==1 ); + zDb = pName->a[0].zDatabase; + zName = pName->a[0].zName; + assert( zDb!=0 || sqlite3BtreeHoldsAllMutexes(db) ); + for(i=OMIT_TEMPDB; inDb; i++){ + int j = (i<2) ? i^1 : i; /* Search TEMP before MAIN */ + if( zDb && sqlite3StrICmp(db->aDb[j].zDbSName, zDb) ) continue; + assert( sqlite3SchemaMutexHeld(db, j, 0) ); + pTrigger = sqlite3HashFind(&(db->aDb[j].pSchema->trigHash), zName); + if( pTrigger ) break; + } + if( !pTrigger ){ + if( !noErr ){ + sqlite3ErrorMsg(pParse, "no such trigger: %S", pName, 0); + }else{ + sqlite3CodeVerifyNamedSchema(pParse, zDb); + } + pParse->checkSchema = 1; + goto drop_trigger_cleanup; + } + sqlite3DropTriggerPtr(pParse, pTrigger); + +drop_trigger_cleanup: + sqlite3SrcListDelete(db, pName); +} + +/* +** Return a pointer to the Table structure for the table that a trigger +** is set on. +*/ +static Table *tableOfTrigger(Trigger *pTrigger){ + return sqlite3HashFind(&pTrigger->pTabSchema->tblHash, pTrigger->table); +} + + +/* +** Drop a trigger given a pointer to that trigger. +*/ +SQLITE_PRIVATE void sqlite3DropTriggerPtr(Parse *pParse, Trigger *pTrigger){ + Table *pTable; + Vdbe *v; + sqlite3 *db = pParse->db; + int iDb; + + iDb = sqlite3SchemaToIndex(pParse->db, pTrigger->pSchema); + assert( iDb>=0 && iDbnDb ); + pTable = tableOfTrigger(pTrigger); + assert( (pTable && pTable->pSchema==pTrigger->pSchema) || iDb==1 ); +#ifndef SQLITE_OMIT_AUTHORIZATION + if( pTable ){ + int code = SQLITE_DROP_TRIGGER; + const char *zDb = db->aDb[iDb].zDbSName; + const char *zTab = SCHEMA_TABLE(iDb); + if( iDb==1 ) code = SQLITE_DROP_TEMP_TRIGGER; + if( sqlite3AuthCheck(pParse, code, pTrigger->zName, pTable->zName, zDb) || + sqlite3AuthCheck(pParse, SQLITE_DELETE, zTab, 0, zDb) ){ + return; + } + } +#endif + + /* Generate code to destroy the database record of the trigger. + */ + if( (v = sqlite3GetVdbe(pParse))!=0 ){ + sqlite3NestedParse(pParse, + "DELETE FROM %Q.%s WHERE name=%Q AND type='trigger'", + db->aDb[iDb].zDbSName, MASTER_NAME, pTrigger->zName + ); + sqlite3ChangeCookie(pParse, iDb); + sqlite3VdbeAddOp4(v, OP_DropTrigger, iDb, 0, 0, pTrigger->zName, 0); + } +} + +/* +** Remove a trigger from the hash tables of the sqlite* pointer. +*/ +SQLITE_PRIVATE void sqlite3UnlinkAndDeleteTrigger(sqlite3 *db, int iDb, const char *zName){ + Trigger *pTrigger; + Hash *pHash; + + assert( sqlite3SchemaMutexHeld(db, iDb, 0) ); + pHash = &(db->aDb[iDb].pSchema->trigHash); + pTrigger = sqlite3HashInsert(pHash, zName, 0); + if( ALWAYS(pTrigger) ){ + if( pTrigger->pSchema==pTrigger->pTabSchema ){ + Table *pTab = tableOfTrigger(pTrigger); + if( pTab ){ + Trigger **pp; + for(pp=&pTab->pTrigger; *pp!=pTrigger; pp=&((*pp)->pNext)); + *pp = (*pp)->pNext; + } + } + sqlite3DeleteTrigger(db, pTrigger); + db->mDbFlags |= DBFLAG_SchemaChange; + } +} + +/* +** pEList is the SET clause of an UPDATE statement. Each entry +** in pEList is of the format =. If any of the entries +** in pEList have an which matches an identifier in pIdList, +** then return TRUE. If pIdList==NULL, then it is considered a +** wildcard that matches anything. Likewise if pEList==NULL then +** it matches anything so always return true. Return false only +** if there is no match. +*/ +static int checkColumnOverlap(IdList *pIdList, ExprList *pEList){ + int e; + if( pIdList==0 || NEVER(pEList==0) ) return 1; + for(e=0; enExpr; e++){ + if( sqlite3IdListIndex(pIdList, pEList->a[e].zName)>=0 ) return 1; + } + return 0; +} + +/* +** Return a list of all triggers on table pTab if there exists at least +** one trigger that must be fired when an operation of type 'op' is +** performed on the table, and, if that operation is an UPDATE, if at +** least one of the columns in pChanges is being modified. +*/ +SQLITE_PRIVATE Trigger *sqlite3TriggersExist( + Parse *pParse, /* Parse context */ + Table *pTab, /* The table the contains the triggers */ + int op, /* one of TK_DELETE, TK_INSERT, TK_UPDATE */ + ExprList *pChanges, /* Columns that change in an UPDATE statement */ + int *pMask /* OUT: Mask of TRIGGER_BEFORE|TRIGGER_AFTER */ +){ + int mask = 0; + Trigger *pList = 0; + Trigger *p; + + if( (pParse->db->flags & SQLITE_EnableTrigger)!=0 ){ + pList = sqlite3TriggerList(pParse, pTab); + } + assert( pList==0 || IsVirtual(pTab)==0 ); + for(p=pList; p; p=p->pNext){ + if( p->op==op && checkColumnOverlap(p->pColumns, pChanges) ){ + mask |= p->tr_tm; + } + } + if( pMask ){ + *pMask = mask; + } + return (mask ? pList : 0); +} + +/* +** Convert the pStep->zTarget string into a SrcList and return a pointer +** to that SrcList. +** +** This routine adds a specific database name, if needed, to the target when +** forming the SrcList. This prevents a trigger in one database from +** referring to a target in another database. An exception is when the +** trigger is in TEMP in which case it can refer to any other database it +** wants. +*/ +static SrcList *targetSrcList( + Parse *pParse, /* The parsing context */ + TriggerStep *pStep /* The trigger containing the target token */ +){ + sqlite3 *db = pParse->db; + int iDb; /* Index of the database to use */ + SrcList *pSrc; /* SrcList to be returned */ + + pSrc = sqlite3SrcListAppend(pParse, 0, 0, 0); + if( pSrc ){ + assert( pSrc->nSrc>0 ); + pSrc->a[pSrc->nSrc-1].zName = sqlite3DbStrDup(db, pStep->zTarget); + iDb = sqlite3SchemaToIndex(db, pStep->pTrig->pSchema); + if( iDb==0 || iDb>=2 ){ + const char *zDb; + assert( iDbnDb ); + zDb = db->aDb[iDb].zDbSName; + pSrc->a[pSrc->nSrc-1].zDatabase = sqlite3DbStrDup(db, zDb); + } + } + return pSrc; +} + +/* +** Generate VDBE code for the statements inside the body of a single +** trigger. +*/ +static int codeTriggerProgram( + Parse *pParse, /* The parser context */ + TriggerStep *pStepList, /* List of statements inside the trigger body */ + int orconf /* Conflict algorithm. (OE_Abort, etc) */ +){ + TriggerStep *pStep; + Vdbe *v = pParse->pVdbe; + sqlite3 *db = pParse->db; + + assert( pParse->pTriggerTab && pParse->pToplevel ); + assert( pStepList ); + assert( v!=0 ); + for(pStep=pStepList; pStep; pStep=pStep->pNext){ + /* Figure out the ON CONFLICT policy that will be used for this step + ** of the trigger program. If the statement that caused this trigger + ** to fire had an explicit ON CONFLICT, then use it. Otherwise, use + ** the ON CONFLICT policy that was specified as part of the trigger + ** step statement. Example: + ** + ** CREATE TRIGGER AFTER INSERT ON t1 BEGIN; + ** INSERT OR REPLACE INTO t2 VALUES(new.a, new.b); + ** END; + ** + ** INSERT INTO t1 ... ; -- insert into t2 uses REPLACE policy + ** INSERT OR IGNORE INTO t1 ... ; -- insert into t2 uses IGNORE policy + */ + pParse->eOrconf = (orconf==OE_Default)?pStep->orconf:(u8)orconf; + assert( pParse->okConstFactor==0 ); + +#ifndef SQLITE_OMIT_TRACE + if( pStep->zSpan ){ + sqlite3VdbeAddOp4(v, OP_Trace, 0x7fffffff, 1, 0, + sqlite3MPrintf(db, "-- %s", pStep->zSpan), + P4_DYNAMIC); + } +#endif + + switch( pStep->op ){ + case TK_UPDATE: { + sqlite3Update(pParse, + targetSrcList(pParse, pStep), + sqlite3ExprListDup(db, pStep->pExprList, 0), + sqlite3ExprDup(db, pStep->pWhere, 0), + pParse->eOrconf, 0, 0, 0 + ); + break; + } + case TK_INSERT: { + sqlite3Insert(pParse, + targetSrcList(pParse, pStep), + sqlite3SelectDup(db, pStep->pSelect, 0), + sqlite3IdListDup(db, pStep->pIdList), + pParse->eOrconf, + sqlite3UpsertDup(db, pStep->pUpsert) + ); + break; + } + case TK_DELETE: { + sqlite3DeleteFrom(pParse, + targetSrcList(pParse, pStep), + sqlite3ExprDup(db, pStep->pWhere, 0), 0, 0 + ); + break; + } + default: assert( pStep->op==TK_SELECT ); { + SelectDest sDest; + Select *pSelect = sqlite3SelectDup(db, pStep->pSelect, 0); + sqlite3SelectDestInit(&sDest, SRT_Discard, 0); + sqlite3Select(pParse, pSelect, &sDest); + sqlite3SelectDelete(db, pSelect); + break; + } + } + if( pStep->op!=TK_SELECT ){ + sqlite3VdbeAddOp0(v, OP_ResetCount); + } + } + + return 0; +} + +#ifdef SQLITE_ENABLE_EXPLAIN_COMMENTS +/* +** This function is used to add VdbeComment() annotations to a VDBE +** program. It is not used in production code, only for debugging. +*/ +static const char *onErrorText(int onError){ + switch( onError ){ + case OE_Abort: return "abort"; + case OE_Rollback: return "rollback"; + case OE_Fail: return "fail"; + case OE_Replace: return "replace"; + case OE_Ignore: return "ignore"; + case OE_Default: return "default"; + } + return "n/a"; +} +#endif + +/* +** Parse context structure pFrom has just been used to create a sub-vdbe +** (trigger program). If an error has occurred, transfer error information +** from pFrom to pTo. +*/ +static void transferParseError(Parse *pTo, Parse *pFrom){ + assert( pFrom->zErrMsg==0 || pFrom->nErr ); + assert( pTo->zErrMsg==0 || pTo->nErr ); + if( pTo->nErr==0 ){ + pTo->zErrMsg = pFrom->zErrMsg; + pTo->nErr = pFrom->nErr; + pTo->rc = pFrom->rc; + }else{ + sqlite3DbFree(pFrom->db, pFrom->zErrMsg); + } +} + +/* +** Create and populate a new TriggerPrg object with a sub-program +** implementing trigger pTrigger with ON CONFLICT policy orconf. +*/ +static TriggerPrg *codeRowTrigger( + Parse *pParse, /* Current parse context */ + Trigger *pTrigger, /* Trigger to code */ + Table *pTab, /* The table pTrigger is attached to */ + int orconf /* ON CONFLICT policy to code trigger program with */ +){ + Parse *pTop = sqlite3ParseToplevel(pParse); + sqlite3 *db = pParse->db; /* Database handle */ + TriggerPrg *pPrg; /* Value to return */ + Expr *pWhen = 0; /* Duplicate of trigger WHEN expression */ + Vdbe *v; /* Temporary VM */ + NameContext sNC; /* Name context for sub-vdbe */ + SubProgram *pProgram = 0; /* Sub-vdbe for trigger program */ + Parse *pSubParse; /* Parse context for sub-vdbe */ + int iEndTrigger = 0; /* Label to jump to if WHEN is false */ + + assert( pTrigger->zName==0 || pTab==tableOfTrigger(pTrigger) ); + assert( pTop->pVdbe ); + + /* Allocate the TriggerPrg and SubProgram objects. To ensure that they + ** are freed if an error occurs, link them into the Parse.pTriggerPrg + ** list of the top-level Parse object sooner rather than later. */ + pPrg = sqlite3DbMallocZero(db, sizeof(TriggerPrg)); + if( !pPrg ) return 0; + pPrg->pNext = pTop->pTriggerPrg; + pTop->pTriggerPrg = pPrg; + pPrg->pProgram = pProgram = sqlite3DbMallocZero(db, sizeof(SubProgram)); + if( !pProgram ) return 0; + sqlite3VdbeLinkSubProgram(pTop->pVdbe, pProgram); + pPrg->pTrigger = pTrigger; + pPrg->orconf = orconf; + pPrg->aColmask[0] = 0xffffffff; + pPrg->aColmask[1] = 0xffffffff; + + /* Allocate and populate a new Parse context to use for coding the + ** trigger sub-program. */ + pSubParse = sqlite3StackAllocZero(db, sizeof(Parse)); + if( !pSubParse ) return 0; + memset(&sNC, 0, sizeof(sNC)); + sNC.pParse = pSubParse; + pSubParse->db = db; + pSubParse->pTriggerTab = pTab; + pSubParse->pToplevel = pTop; + pSubParse->zAuthContext = pTrigger->zName; + pSubParse->eTriggerOp = pTrigger->op; + pSubParse->nQueryLoop = pParse->nQueryLoop; + pSubParse->disableVtab = pParse->disableVtab; + + v = sqlite3GetVdbe(pSubParse); + if( v ){ + VdbeComment((v, "Start: %s.%s (%s %s%s%s ON %s)", + pTrigger->zName, onErrorText(orconf), + (pTrigger->tr_tm==TRIGGER_BEFORE ? "BEFORE" : "AFTER"), + (pTrigger->op==TK_UPDATE ? "UPDATE" : ""), + (pTrigger->op==TK_INSERT ? "INSERT" : ""), + (pTrigger->op==TK_DELETE ? "DELETE" : ""), + pTab->zName + )); +#ifndef SQLITE_OMIT_TRACE + if( pTrigger->zName ){ + sqlite3VdbeChangeP4(v, -1, + sqlite3MPrintf(db, "-- TRIGGER %s", pTrigger->zName), P4_DYNAMIC + ); + } +#endif + + /* If one was specified, code the WHEN clause. If it evaluates to false + ** (or NULL) the sub-vdbe is immediately halted by jumping to the + ** OP_Halt inserted at the end of the program. */ + if( pTrigger->pWhen ){ + pWhen = sqlite3ExprDup(db, pTrigger->pWhen, 0); + if( SQLITE_OK==sqlite3ResolveExprNames(&sNC, pWhen) + && db->mallocFailed==0 + ){ + iEndTrigger = sqlite3VdbeMakeLabel(pSubParse); + sqlite3ExprIfFalse(pSubParse, pWhen, iEndTrigger, SQLITE_JUMPIFNULL); + } + sqlite3ExprDelete(db, pWhen); + } + + /* Code the trigger program into the sub-vdbe. */ + codeTriggerProgram(pSubParse, pTrigger->step_list, orconf); + + /* Insert an OP_Halt at the end of the sub-program. */ + if( iEndTrigger ){ + sqlite3VdbeResolveLabel(v, iEndTrigger); + } + sqlite3VdbeAddOp0(v, OP_Halt); + VdbeComment((v, "End: %s.%s", pTrigger->zName, onErrorText(orconf))); + + transferParseError(pParse, pSubParse); + if( db->mallocFailed==0 && pParse->nErr==0 ){ + pProgram->aOp = sqlite3VdbeTakeOpArray(v, &pProgram->nOp, &pTop->nMaxArg); + } + pProgram->nMem = pSubParse->nMem; + pProgram->nCsr = pSubParse->nTab; + pProgram->token = (void *)pTrigger; + pPrg->aColmask[0] = pSubParse->oldmask; + pPrg->aColmask[1] = pSubParse->newmask; + sqlite3VdbeDelete(v); + } + + assert( !pSubParse->pAinc && !pSubParse->pZombieTab ); + assert( !pSubParse->pTriggerPrg && !pSubParse->nMaxArg ); + sqlite3ParserReset(pSubParse); + sqlite3StackFree(db, pSubParse); + + return pPrg; +} + +/* +** Return a pointer to a TriggerPrg object containing the sub-program for +** trigger pTrigger with default ON CONFLICT algorithm orconf. If no such +** TriggerPrg object exists, a new object is allocated and populated before +** being returned. +*/ +static TriggerPrg *getRowTrigger( + Parse *pParse, /* Current parse context */ + Trigger *pTrigger, /* Trigger to code */ + Table *pTab, /* The table trigger pTrigger is attached to */ + int orconf /* ON CONFLICT algorithm. */ +){ + Parse *pRoot = sqlite3ParseToplevel(pParse); + TriggerPrg *pPrg; + + assert( pTrigger->zName==0 || pTab==tableOfTrigger(pTrigger) ); + + /* It may be that this trigger has already been coded (or is in the + ** process of being coded). If this is the case, then an entry with + ** a matching TriggerPrg.pTrigger field will be present somewhere + ** in the Parse.pTriggerPrg list. Search for such an entry. */ + for(pPrg=pRoot->pTriggerPrg; + pPrg && (pPrg->pTrigger!=pTrigger || pPrg->orconf!=orconf); + pPrg=pPrg->pNext + ); + + /* If an existing TriggerPrg could not be located, create a new one. */ + if( !pPrg ){ + pPrg = codeRowTrigger(pParse, pTrigger, pTab, orconf); + } + + return pPrg; +} + +/* +** Generate code for the trigger program associated with trigger p on +** table pTab. The reg, orconf and ignoreJump parameters passed to this +** function are the same as those described in the header function for +** sqlite3CodeRowTrigger() +*/ +SQLITE_PRIVATE void sqlite3CodeRowTriggerDirect( + Parse *pParse, /* Parse context */ + Trigger *p, /* Trigger to code */ + Table *pTab, /* The table to code triggers from */ + int reg, /* Reg array containing OLD.* and NEW.* values */ + int orconf, /* ON CONFLICT policy */ + int ignoreJump /* Instruction to jump to for RAISE(IGNORE) */ +){ + Vdbe *v = sqlite3GetVdbe(pParse); /* Main VM */ + TriggerPrg *pPrg; + pPrg = getRowTrigger(pParse, p, pTab, orconf); + assert( pPrg || pParse->nErr || pParse->db->mallocFailed ); + + /* Code the OP_Program opcode in the parent VDBE. P4 of the OP_Program + ** is a pointer to the sub-vdbe containing the trigger program. */ + if( pPrg ){ + int bRecursive = (p->zName && 0==(pParse->db->flags&SQLITE_RecTriggers)); + + sqlite3VdbeAddOp4(v, OP_Program, reg, ignoreJump, ++pParse->nMem, + (const char *)pPrg->pProgram, P4_SUBPROGRAM); + VdbeComment( + (v, "Call: %s.%s", (p->zName?p->zName:"fkey"), onErrorText(orconf))); + + /* Set the P5 operand of the OP_Program instruction to non-zero if + ** recursive invocation of this trigger program is disallowed. Recursive + ** invocation is disallowed if (a) the sub-program is really a trigger, + ** not a foreign key action, and (b) the flag to enable recursive triggers + ** is clear. */ + sqlite3VdbeChangeP5(v, (u8)bRecursive); + } +} + +/* +** This is called to code the required FOR EACH ROW triggers for an operation +** on table pTab. The operation to code triggers for (INSERT, UPDATE or DELETE) +** is given by the op parameter. The tr_tm parameter determines whether the +** BEFORE or AFTER triggers are coded. If the operation is an UPDATE, then +** parameter pChanges is passed the list of columns being modified. +** +** If there are no triggers that fire at the specified time for the specified +** operation on pTab, this function is a no-op. +** +** The reg argument is the address of the first in an array of registers +** that contain the values substituted for the new.* and old.* references +** in the trigger program. If N is the number of columns in table pTab +** (a copy of pTab->nCol), then registers are populated as follows: +** +** Register Contains +** ------------------------------------------------------ +** reg+0 OLD.rowid +** reg+1 OLD.* value of left-most column of pTab +** ... ... +** reg+N OLD.* value of right-most column of pTab +** reg+N+1 NEW.rowid +** reg+N+2 OLD.* value of left-most column of pTab +** ... ... +** reg+N+N+1 NEW.* value of right-most column of pTab +** +** For ON DELETE triggers, the registers containing the NEW.* values will +** never be accessed by the trigger program, so they are not allocated or +** populated by the caller (there is no data to populate them with anyway). +** Similarly, for ON INSERT triggers the values stored in the OLD.* registers +** are never accessed, and so are not allocated by the caller. So, for an +** ON INSERT trigger, the value passed to this function as parameter reg +** is not a readable register, although registers (reg+N) through +** (reg+N+N+1) are. +** +** Parameter orconf is the default conflict resolution algorithm for the +** trigger program to use (REPLACE, IGNORE etc.). Parameter ignoreJump +** is the instruction that control should jump to if a trigger program +** raises an IGNORE exception. +*/ +SQLITE_PRIVATE void sqlite3CodeRowTrigger( + Parse *pParse, /* Parse context */ + Trigger *pTrigger, /* List of triggers on table pTab */ + int op, /* One of TK_UPDATE, TK_INSERT, TK_DELETE */ + ExprList *pChanges, /* Changes list for any UPDATE OF triggers */ + int tr_tm, /* One of TRIGGER_BEFORE, TRIGGER_AFTER */ + Table *pTab, /* The table to code triggers from */ + int reg, /* The first in an array of registers (see above) */ + int orconf, /* ON CONFLICT policy */ + int ignoreJump /* Instruction to jump to for RAISE(IGNORE) */ +){ + Trigger *p; /* Used to iterate through pTrigger list */ + + assert( op==TK_UPDATE || op==TK_INSERT || op==TK_DELETE ); + assert( tr_tm==TRIGGER_BEFORE || tr_tm==TRIGGER_AFTER ); + assert( (op==TK_UPDATE)==(pChanges!=0) ); + + for(p=pTrigger; p; p=p->pNext){ + + /* Sanity checking: The schema for the trigger and for the table are + ** always defined. The trigger must be in the same schema as the table + ** or else it must be a TEMP trigger. */ + assert( p->pSchema!=0 ); + assert( p->pTabSchema!=0 ); + assert( p->pSchema==p->pTabSchema + || p->pSchema==pParse->db->aDb[1].pSchema ); + + /* Determine whether we should code this trigger */ + if( p->op==op + && p->tr_tm==tr_tm + && checkColumnOverlap(p->pColumns, pChanges) + ){ + sqlite3CodeRowTriggerDirect(pParse, p, pTab, reg, orconf, ignoreJump); + } + } +} + +/* +** Triggers may access values stored in the old.* or new.* pseudo-table. +** This function returns a 32-bit bitmask indicating which columns of the +** old.* or new.* tables actually are used by triggers. This information +** may be used by the caller, for example, to avoid having to load the entire +** old.* record into memory when executing an UPDATE or DELETE command. +** +** Bit 0 of the returned mask is set if the left-most column of the +** table may be accessed using an [old|new].reference. Bit 1 is set if +** the second leftmost column value is required, and so on. If there +** are more than 32 columns in the table, and at least one of the columns +** with an index greater than 32 may be accessed, 0xffffffff is returned. +** +** It is not possible to determine if the old.rowid or new.rowid column is +** accessed by triggers. The caller must always assume that it is. +** +** Parameter isNew must be either 1 or 0. If it is 0, then the mask returned +** applies to the old.* table. If 1, the new.* table. +** +** Parameter tr_tm must be a mask with one or both of the TRIGGER_BEFORE +** and TRIGGER_AFTER bits set. Values accessed by BEFORE triggers are only +** included in the returned mask if the TRIGGER_BEFORE bit is set in the +** tr_tm parameter. Similarly, values accessed by AFTER triggers are only +** included in the returned mask if the TRIGGER_AFTER bit is set in tr_tm. +*/ +SQLITE_PRIVATE u32 sqlite3TriggerColmask( + Parse *pParse, /* Parse context */ + Trigger *pTrigger, /* List of triggers on table pTab */ + ExprList *pChanges, /* Changes list for any UPDATE OF triggers */ + int isNew, /* 1 for new.* ref mask, 0 for old.* ref mask */ + int tr_tm, /* Mask of TRIGGER_BEFORE|TRIGGER_AFTER */ + Table *pTab, /* The table to code triggers from */ + int orconf /* Default ON CONFLICT policy for trigger steps */ +){ + const int op = pChanges ? TK_UPDATE : TK_DELETE; + u32 mask = 0; + Trigger *p; + + assert( isNew==1 || isNew==0 ); + for(p=pTrigger; p; p=p->pNext){ + if( p->op==op && (tr_tm&p->tr_tm) + && checkColumnOverlap(p->pColumns,pChanges) + ){ + TriggerPrg *pPrg; + pPrg = getRowTrigger(pParse, p, pTab, orconf); + if( pPrg ){ + mask |= pPrg->aColmask[isNew]; + } + } + } + + return mask; +} + +#endif /* !defined(SQLITE_OMIT_TRIGGER) */ + +/************** End of trigger.c *********************************************/ +/************** Begin file update.c ******************************************/ +/* +** 2001 September 15 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** This file contains C code routines that are called by the parser +** to handle UPDATE statements. +*/ +/* #include "sqliteInt.h" */ + +#ifndef SQLITE_OMIT_VIRTUALTABLE +/* Forward declaration */ +static void updateVirtualTable( + Parse *pParse, /* The parsing context */ + SrcList *pSrc, /* The virtual table to be modified */ + Table *pTab, /* The virtual table */ + ExprList *pChanges, /* The columns to change in the UPDATE statement */ + Expr *pRowidExpr, /* Expression used to recompute the rowid */ + int *aXRef, /* Mapping from columns of pTab to entries in pChanges */ + Expr *pWhere, /* WHERE clause of the UPDATE statement */ + int onError /* ON CONFLICT strategy */ +); +#endif /* SQLITE_OMIT_VIRTUALTABLE */ + +/* +** The most recently coded instruction was an OP_Column to retrieve the +** i-th column of table pTab. This routine sets the P4 parameter of the +** OP_Column to the default value, if any. +** +** The default value of a column is specified by a DEFAULT clause in the +** column definition. This was either supplied by the user when the table +** was created, or added later to the table definition by an ALTER TABLE +** command. If the latter, then the row-records in the table btree on disk +** may not contain a value for the column and the default value, taken +** from the P4 parameter of the OP_Column instruction, is returned instead. +** If the former, then all row-records are guaranteed to include a value +** for the column and the P4 value is not required. +** +** Column definitions created by an ALTER TABLE command may only have +** literal default values specified: a number, null or a string. (If a more +** complicated default expression value was provided, it is evaluated +** when the ALTER TABLE is executed and one of the literal values written +** into the sqlite_master table.) +** +** Therefore, the P4 parameter is only required if the default value for +** the column is a literal number, string or null. The sqlite3ValueFromExpr() +** function is capable of transforming these types of expressions into +** sqlite3_value objects. +** +** If parameter iReg is not negative, code an OP_RealAffinity instruction +** on register iReg. This is used when an equivalent integer value is +** stored in place of an 8-byte floating point value in order to save +** space. +*/ +SQLITE_PRIVATE void sqlite3ColumnDefault(Vdbe *v, Table *pTab, int i, int iReg){ + assert( pTab!=0 ); + if( !pTab->pSelect ){ + sqlite3_value *pValue = 0; + u8 enc = ENC(sqlite3VdbeDb(v)); + Column *pCol = &pTab->aCol[i]; + VdbeComment((v, "%s.%s", pTab->zName, pCol->zName)); + assert( inCol ); + sqlite3ValueFromExpr(sqlite3VdbeDb(v), pCol->pDflt, enc, + pCol->affinity, &pValue); + if( pValue ){ + sqlite3VdbeAppendP4(v, pValue, P4_MEM); + } + } +#ifndef SQLITE_OMIT_FLOATING_POINT + if( pTab->aCol[i].affinity==SQLITE_AFF_REAL ){ + sqlite3VdbeAddOp1(v, OP_RealAffinity, iReg); + } +#endif +} + +/* +** Check to see if column iCol of index pIdx references any of the +** columns defined by aXRef and chngRowid. Return true if it does +** and false if not. This is an optimization. False-positives are a +** performance degradation, but false-negatives can result in a corrupt +** index and incorrect answers. +** +** aXRef[j] will be non-negative if column j of the original table is +** being updated. chngRowid will be true if the rowid of the table is +** being updated. +*/ +static int indexColumnIsBeingUpdated( + Index *pIdx, /* The index to check */ + int iCol, /* Which column of the index to check */ + int *aXRef, /* aXRef[j]>=0 if column j is being updated */ + int chngRowid /* true if the rowid is being updated */ +){ + i16 iIdxCol = pIdx->aiColumn[iCol]; + assert( iIdxCol!=XN_ROWID ); /* Cannot index rowid */ + if( iIdxCol>=0 ){ + return aXRef[iIdxCol]>=0; + } + assert( iIdxCol==XN_EXPR ); + assert( pIdx->aColExpr!=0 ); + assert( pIdx->aColExpr->a[iCol].pExpr!=0 ); + return sqlite3ExprReferencesUpdatedColumn(pIdx->aColExpr->a[iCol].pExpr, + aXRef,chngRowid); +} + +/* +** Check to see if index pIdx is a partial index whose conditional +** expression might change values due to an UPDATE. Return true if +** the index is subject to change and false if the index is guaranteed +** to be unchanged. This is an optimization. False-positives are a +** performance degradation, but false-negatives can result in a corrupt +** index and incorrect answers. +** +** aXRef[j] will be non-negative if column j of the original table is +** being updated. chngRowid will be true if the rowid of the table is +** being updated. +*/ +static int indexWhereClauseMightChange( + Index *pIdx, /* The index to check */ + int *aXRef, /* aXRef[j]>=0 if column j is being updated */ + int chngRowid /* true if the rowid is being updated */ +){ + if( pIdx->pPartIdxWhere==0 ) return 0; + return sqlite3ExprReferencesUpdatedColumn(pIdx->pPartIdxWhere, + aXRef, chngRowid); +} + +/* +** Process an UPDATE statement. +** +** UPDATE OR IGNORE table_wxyz SET a=b, c=d WHERE e<5 AND f NOT NULL; +** \_______/ \________/ \______/ \________________/ +* onError pTabList pChanges pWhere +*/ +SQLITE_PRIVATE void sqlite3Update( + Parse *pParse, /* The parser context */ + SrcList *pTabList, /* The table in which we should change things */ + ExprList *pChanges, /* Things to be changed */ + Expr *pWhere, /* The WHERE clause. May be null */ + int onError, /* How to handle constraint errors */ + ExprList *pOrderBy, /* ORDER BY clause. May be null */ + Expr *pLimit, /* LIMIT clause. May be null */ + Upsert *pUpsert /* ON CONFLICT clause, or null */ +){ + int i, j; /* Loop counters */ + Table *pTab; /* The table to be updated */ + int addrTop = 0; /* VDBE instruction address of the start of the loop */ + WhereInfo *pWInfo; /* Information about the WHERE clause */ + Vdbe *v; /* The virtual database engine */ + Index *pIdx; /* For looping over indices */ + Index *pPk; /* The PRIMARY KEY index for WITHOUT ROWID tables */ + int nIdx; /* Number of indices that need updating */ + int nAllIdx; /* Total number of indexes */ + int iBaseCur; /* Base cursor number */ + int iDataCur; /* Cursor for the canonical data btree */ + int iIdxCur; /* Cursor for the first index */ + sqlite3 *db; /* The database structure */ + int *aRegIdx = 0; /* Registers for to each index and the main table */ + int *aXRef = 0; /* aXRef[i] is the index in pChanges->a[] of the + ** an expression for the i-th column of the table. + ** aXRef[i]==-1 if the i-th column is not changed. */ + u8 *aToOpen; /* 1 for tables and indices to be opened */ + u8 chngPk; /* PRIMARY KEY changed in a WITHOUT ROWID table */ + u8 chngRowid; /* Rowid changed in a normal table */ + u8 chngKey; /* Either chngPk or chngRowid */ + Expr *pRowidExpr = 0; /* Expression defining the new record number */ + AuthContext sContext; /* The authorization context */ + NameContext sNC; /* The name-context to resolve expressions in */ + int iDb; /* Database containing the table being updated */ + int eOnePass; /* ONEPASS_XXX value from where.c */ + int hasFK; /* True if foreign key processing is required */ + int labelBreak; /* Jump here to break out of UPDATE loop */ + int labelContinue; /* Jump here to continue next step of UPDATE loop */ + int flags; /* Flags for sqlite3WhereBegin() */ + +#ifndef SQLITE_OMIT_TRIGGER + int isView; /* True when updating a view (INSTEAD OF trigger) */ + Trigger *pTrigger; /* List of triggers on pTab, if required */ + int tmask; /* Mask of TRIGGER_BEFORE|TRIGGER_AFTER */ +#endif + int newmask; /* Mask of NEW.* columns accessed by BEFORE triggers */ + int iEph = 0; /* Ephemeral table holding all primary key values */ + int nKey = 0; /* Number of elements in regKey for WITHOUT ROWID */ + int aiCurOnePass[2]; /* The write cursors opened by WHERE_ONEPASS */ + int addrOpen = 0; /* Address of OP_OpenEphemeral */ + int iPk = 0; /* First of nPk cells holding PRIMARY KEY value */ + i16 nPk = 0; /* Number of components of the PRIMARY KEY */ + int bReplace = 0; /* True if REPLACE conflict resolution might happen */ + + /* Register Allocations */ + int regRowCount = 0; /* A count of rows changed */ + int regOldRowid = 0; /* The old rowid */ + int regNewRowid = 0; /* The new rowid */ + int regNew = 0; /* Content of the NEW.* table in triggers */ + int regOld = 0; /* Content of OLD.* table in triggers */ + int regRowSet = 0; /* Rowset of rows to be updated */ + int regKey = 0; /* composite PRIMARY KEY value */ + + memset(&sContext, 0, sizeof(sContext)); + db = pParse->db; + if( pParse->nErr || db->mallocFailed ){ + goto update_cleanup; + } + assert( pTabList->nSrc==1 ); + + /* Locate the table which we want to update. + */ + pTab = sqlite3SrcListLookup(pParse, pTabList); + if( pTab==0 ) goto update_cleanup; + iDb = sqlite3SchemaToIndex(pParse->db, pTab->pSchema); + + /* Figure out if we have any triggers and if the table being + ** updated is a view. + */ +#ifndef SQLITE_OMIT_TRIGGER + pTrigger = sqlite3TriggersExist(pParse, pTab, TK_UPDATE, pChanges, &tmask); + isView = pTab->pSelect!=0; + assert( pTrigger || tmask==0 ); +#else +# define pTrigger 0 +# define isView 0 +# define tmask 0 +#endif +#ifdef SQLITE_OMIT_VIEW +# undef isView +# define isView 0 +#endif + +#ifdef SQLITE_ENABLE_UPDATE_DELETE_LIMIT + if( !isView ){ + pWhere = sqlite3LimitWhere( + pParse, pTabList, pWhere, pOrderBy, pLimit, "UPDATE" + ); + pOrderBy = 0; + pLimit = 0; + } +#endif + + if( sqlite3ViewGetColumnNames(pParse, pTab) ){ + goto update_cleanup; + } + if( sqlite3IsReadOnly(pParse, pTab, tmask) ){ + goto update_cleanup; + } + + /* Allocate a cursors for the main database table and for all indices. + ** The index cursors might not be used, but if they are used they + ** need to occur right after the database cursor. So go ahead and + ** allocate enough space, just in case. + */ + iBaseCur = iDataCur = pParse->nTab++; + iIdxCur = iDataCur+1; + pPk = HasRowid(pTab) ? 0 : sqlite3PrimaryKeyIndex(pTab); + testcase( pPk!=0 && pPk!=pTab->pIndex ); + for(nIdx=0, pIdx=pTab->pIndex; pIdx; pIdx=pIdx->pNext, nIdx++){ + if( pPk==pIdx ){ + iDataCur = pParse->nTab; + } + pParse->nTab++; + } + if( pUpsert ){ + /* On an UPSERT, reuse the same cursors already opened by INSERT */ + iDataCur = pUpsert->iDataCur; + iIdxCur = pUpsert->iIdxCur; + pParse->nTab = iBaseCur; + } + pTabList->a[0].iCursor = iDataCur; + + /* Allocate space for aXRef[], aRegIdx[], and aToOpen[]. + ** Initialize aXRef[] and aToOpen[] to their default values. + */ + aXRef = sqlite3DbMallocRawNN(db, sizeof(int) * (pTab->nCol+nIdx+1) + nIdx+2 ); + if( aXRef==0 ) goto update_cleanup; + aRegIdx = aXRef+pTab->nCol; + aToOpen = (u8*)(aRegIdx+nIdx+1); + memset(aToOpen, 1, nIdx+1); + aToOpen[nIdx+1] = 0; + for(i=0; inCol; i++) aXRef[i] = -1; + + /* Initialize the name-context */ + memset(&sNC, 0, sizeof(sNC)); + sNC.pParse = pParse; + sNC.pSrcList = pTabList; + sNC.uNC.pUpsert = pUpsert; + sNC.ncFlags = NC_UUpsert; + + /* Resolve the column names in all the expressions of the + ** of the UPDATE statement. Also find the column index + ** for each column to be updated in the pChanges array. For each + ** column to be updated, make sure we have authorization to change + ** that column. + */ + chngRowid = chngPk = 0; + for(i=0; inExpr; i++){ + if( sqlite3ResolveExprNames(&sNC, pChanges->a[i].pExpr) ){ + goto update_cleanup; + } + for(j=0; jnCol; j++){ + if( sqlite3StrICmp(pTab->aCol[j].zName, pChanges->a[i].zName)==0 ){ + if( j==pTab->iPKey ){ + chngRowid = 1; + pRowidExpr = pChanges->a[i].pExpr; + }else if( pPk && (pTab->aCol[j].colFlags & COLFLAG_PRIMKEY)!=0 ){ + chngPk = 1; + } + aXRef[j] = i; + break; + } + } + if( j>=pTab->nCol ){ + if( pPk==0 && sqlite3IsRowid(pChanges->a[i].zName) ){ + j = -1; + chngRowid = 1; + pRowidExpr = pChanges->a[i].pExpr; + }else{ + sqlite3ErrorMsg(pParse, "no such column: %s", pChanges->a[i].zName); + pParse->checkSchema = 1; + goto update_cleanup; + } + } +#ifndef SQLITE_OMIT_AUTHORIZATION + { + int rc; + rc = sqlite3AuthCheck(pParse, SQLITE_UPDATE, pTab->zName, + j<0 ? "ROWID" : pTab->aCol[j].zName, + db->aDb[iDb].zDbSName); + if( rc==SQLITE_DENY ){ + goto update_cleanup; + }else if( rc==SQLITE_IGNORE ){ + aXRef[j] = -1; + } + } +#endif + } + assert( (chngRowid & chngPk)==0 ); + assert( chngRowid==0 || chngRowid==1 ); + assert( chngPk==0 || chngPk==1 ); + chngKey = chngRowid + chngPk; + + /* The SET expressions are not actually used inside the WHERE loop. + ** So reset the colUsed mask. Unless this is a virtual table. In that + ** case, set all bits of the colUsed mask (to ensure that the virtual + ** table implementation makes all columns available). + */ + pTabList->a[0].colUsed = IsVirtual(pTab) ? ALLBITS : 0; + + hasFK = sqlite3FkRequired(pParse, pTab, aXRef, chngKey); + + /* There is one entry in the aRegIdx[] array for each index on the table + ** being updated. Fill in aRegIdx[] with a register number that will hold + ** the key for accessing each index. + */ + if( onError==OE_Replace ) bReplace = 1; + for(nAllIdx=0, pIdx=pTab->pIndex; pIdx; pIdx=pIdx->pNext, nAllIdx++){ + int reg; + if( chngKey || hasFK>1 || pIdx==pPk + || indexWhereClauseMightChange(pIdx,aXRef,chngRowid) + ){ + reg = ++pParse->nMem; + pParse->nMem += pIdx->nColumn; + }else{ + reg = 0; + for(i=0; inKeyCol; i++){ + if( indexColumnIsBeingUpdated(pIdx, i, aXRef, chngRowid) ){ + reg = ++pParse->nMem; + pParse->nMem += pIdx->nColumn; + if( onError==OE_Default && pIdx->onError==OE_Replace ){ + bReplace = 1; + } + break; + } + } + } + if( reg==0 ) aToOpen[nAllIdx+1] = 0; + aRegIdx[nAllIdx] = reg; + } + aRegIdx[nAllIdx] = ++pParse->nMem; /* Register storing the table record */ + if( bReplace ){ + /* If REPLACE conflict resolution might be invoked, open cursors on all + ** indexes in case they are needed to delete records. */ + memset(aToOpen, 1, nIdx+1); + } + + /* Begin generating code. */ + v = sqlite3GetVdbe(pParse); + if( v==0 ) goto update_cleanup; + if( pParse->nested==0 ) sqlite3VdbeCountChanges(v); + sqlite3BeginWriteOperation(pParse, pTrigger || hasFK, iDb); + + /* Allocate required registers. */ + if( !IsVirtual(pTab) ){ + /* For now, regRowSet and aRegIdx[nAllIdx] share the same register. + ** If regRowSet turns out to be needed, then aRegIdx[nAllIdx] will be + ** reallocated. aRegIdx[nAllIdx] is the register in which the main + ** table record is written. regRowSet holds the RowSet for the + ** two-pass update algorithm. */ + assert( aRegIdx[nAllIdx]==pParse->nMem ); + regRowSet = aRegIdx[nAllIdx]; + regOldRowid = regNewRowid = ++pParse->nMem; + if( chngPk || pTrigger || hasFK ){ + regOld = pParse->nMem + 1; + pParse->nMem += pTab->nCol; + } + if( chngKey || pTrigger || hasFK ){ + regNewRowid = ++pParse->nMem; + } + regNew = pParse->nMem + 1; + pParse->nMem += pTab->nCol; + } + + /* Start the view context. */ + if( isView ){ + sqlite3AuthContextPush(pParse, &sContext, pTab->zName); + } + + /* If we are trying to update a view, realize that view into + ** an ephemeral table. + */ +#if !defined(SQLITE_OMIT_VIEW) && !defined(SQLITE_OMIT_TRIGGER) + if( isView ){ + sqlite3MaterializeView(pParse, pTab, + pWhere, pOrderBy, pLimit, iDataCur + ); + pOrderBy = 0; + pLimit = 0; + } +#endif + + /* Resolve the column names in all the expressions in the + ** WHERE clause. + */ + if( sqlite3ResolveExprNames(&sNC, pWhere) ){ + goto update_cleanup; + } + +#ifndef SQLITE_OMIT_VIRTUALTABLE + /* Virtual tables must be handled separately */ + if( IsVirtual(pTab) ){ + updateVirtualTable(pParse, pTabList, pTab, pChanges, pRowidExpr, aXRef, + pWhere, onError); + goto update_cleanup; + } +#endif + + /* Jump to labelBreak to abandon further processing of this UPDATE */ + labelContinue = labelBreak = sqlite3VdbeMakeLabel(pParse); + + /* Not an UPSERT. Normal processing. Begin by + ** initialize the count of updated rows */ + if( (db->flags&SQLITE_CountRows)!=0 + && !pParse->pTriggerTab + && !pParse->nested + && pUpsert==0 + ){ + regRowCount = ++pParse->nMem; + sqlite3VdbeAddOp2(v, OP_Integer, 0, regRowCount); + } + + if( HasRowid(pTab) ){ + sqlite3VdbeAddOp3(v, OP_Null, 0, regRowSet, regOldRowid); + }else{ + assert( pPk!=0 ); + nPk = pPk->nKeyCol; + iPk = pParse->nMem+1; + pParse->nMem += nPk; + regKey = ++pParse->nMem; + if( pUpsert==0 ){ + iEph = pParse->nTab++; + sqlite3VdbeAddOp3(v, OP_Null, 0, iPk, iPk+nPk-1); + addrOpen = sqlite3VdbeAddOp2(v, OP_OpenEphemeral, iEph, nPk); + sqlite3VdbeSetP4KeyInfo(pParse, pPk); + } + } + + if( pUpsert ){ + /* If this is an UPSERT, then all cursors have already been opened by + ** the outer INSERT and the data cursor should be pointing at the row + ** that is to be updated. So bypass the code that searches for the + ** row(s) to be updated. + */ + pWInfo = 0; + eOnePass = ONEPASS_SINGLE; + sqlite3ExprIfFalse(pParse, pWhere, labelBreak, SQLITE_JUMPIFNULL); + }else{ + /* Begin the database scan. + ** + ** Do not consider a single-pass strategy for a multi-row update if + ** there are any triggers or foreign keys to process, or rows may + ** be deleted as a result of REPLACE conflict handling. Any of these + ** things might disturb a cursor being used to scan through the table + ** or index, causing a single-pass approach to malfunction. */ + flags = WHERE_ONEPASS_DESIRED|WHERE_SEEK_UNIQ_TABLE; + if( !pParse->nested && !pTrigger && !hasFK && !chngKey && !bReplace ){ + flags |= WHERE_ONEPASS_MULTIROW; + } + pWInfo = sqlite3WhereBegin(pParse, pTabList, pWhere, 0, 0, flags, iIdxCur); + if( pWInfo==0 ) goto update_cleanup; + + /* A one-pass strategy that might update more than one row may not + ** be used if any column of the index used for the scan is being + ** updated. Otherwise, if there is an index on "b", statements like + ** the following could create an infinite loop: + ** + ** UPDATE t1 SET b=b+1 WHERE b>? + ** + ** Fall back to ONEPASS_OFF if where.c has selected a ONEPASS_MULTI + ** strategy that uses an index for which one or more columns are being + ** updated. */ + eOnePass = sqlite3WhereOkOnePass(pWInfo, aiCurOnePass); + if( eOnePass!=ONEPASS_SINGLE ){ + sqlite3MultiWrite(pParse); + if( eOnePass==ONEPASS_MULTI ){ + int iCur = aiCurOnePass[1]; + if( iCur>=0 && iCur!=iDataCur && aToOpen[iCur-iBaseCur] ){ + eOnePass = ONEPASS_OFF; + } + assert( iCur!=iDataCur || !HasRowid(pTab) ); + } + } + } + + if( HasRowid(pTab) ){ + /* Read the rowid of the current row of the WHERE scan. In ONEPASS_OFF + ** mode, write the rowid into the FIFO. In either of the one-pass modes, + ** leave it in register regOldRowid. */ + sqlite3VdbeAddOp2(v, OP_Rowid, iDataCur, regOldRowid); + if( eOnePass==ONEPASS_OFF ){ + /* We need to use regRowSet, so reallocate aRegIdx[nAllIdx] */ + aRegIdx[nAllIdx] = ++pParse->nMem; + sqlite3VdbeAddOp2(v, OP_RowSetAdd, regRowSet, regOldRowid); + } + }else{ + /* Read the PK of the current row into an array of registers. In + ** ONEPASS_OFF mode, serialize the array into a record and store it in + ** the ephemeral table. Or, in ONEPASS_SINGLE or MULTI mode, change + ** the OP_OpenEphemeral instruction to a Noop (the ephemeral table + ** is not required) and leave the PK fields in the array of registers. */ + for(i=0; iaiColumn[i]>=0 ); + sqlite3ExprCodeGetColumnOfTable(v, pTab, iDataCur,pPk->aiColumn[i],iPk+i); + } + if( eOnePass ){ + if( addrOpen ) sqlite3VdbeChangeToNoop(v, addrOpen); + nKey = nPk; + regKey = iPk; + }else{ + sqlite3VdbeAddOp4(v, OP_MakeRecord, iPk, nPk, regKey, + sqlite3IndexAffinityStr(db, pPk), nPk); + sqlite3VdbeAddOp4Int(v, OP_IdxInsert, iEph, regKey, iPk, nPk); + } + } + + if( pUpsert==0 ){ + if( eOnePass!=ONEPASS_MULTI ){ + sqlite3WhereEnd(pWInfo); + } + + if( !isView ){ + int addrOnce = 0; + + /* Open every index that needs updating. */ + if( eOnePass!=ONEPASS_OFF ){ + if( aiCurOnePass[0]>=0 ) aToOpen[aiCurOnePass[0]-iBaseCur] = 0; + if( aiCurOnePass[1]>=0 ) aToOpen[aiCurOnePass[1]-iBaseCur] = 0; + } + + if( eOnePass==ONEPASS_MULTI && (nIdx-(aiCurOnePass[1]>=0))>0 ){ + addrOnce = sqlite3VdbeAddOp0(v, OP_Once); VdbeCoverage(v); + } + sqlite3OpenTableAndIndices(pParse, pTab, OP_OpenWrite, 0, iBaseCur, + aToOpen, 0, 0); + if( addrOnce ) sqlite3VdbeJumpHere(v, addrOnce); + } + + /* Top of the update loop */ + if( eOnePass!=ONEPASS_OFF ){ + if( !isView && aiCurOnePass[0]!=iDataCur && aiCurOnePass[1]!=iDataCur ){ + assert( pPk ); + sqlite3VdbeAddOp4Int(v, OP_NotFound, iDataCur, labelBreak, regKey,nKey); + VdbeCoverage(v); + } + if( eOnePass!=ONEPASS_SINGLE ){ + labelContinue = sqlite3VdbeMakeLabel(pParse); + } + sqlite3VdbeAddOp2(v, OP_IsNull, pPk ? regKey : regOldRowid, labelBreak); + VdbeCoverageIf(v, pPk==0); + VdbeCoverageIf(v, pPk!=0); + }else if( pPk ){ + labelContinue = sqlite3VdbeMakeLabel(pParse); + sqlite3VdbeAddOp2(v, OP_Rewind, iEph, labelBreak); VdbeCoverage(v); + addrTop = sqlite3VdbeAddOp2(v, OP_RowData, iEph, regKey); + sqlite3VdbeAddOp4Int(v, OP_NotFound, iDataCur, labelContinue, regKey, 0); + VdbeCoverage(v); + }else{ + labelContinue = sqlite3VdbeAddOp3(v, OP_RowSetRead, regRowSet,labelBreak, + regOldRowid); + VdbeCoverage(v); + sqlite3VdbeAddOp3(v, OP_NotExists, iDataCur, labelContinue, regOldRowid); + VdbeCoverage(v); + } + } + + /* If the rowid value will change, set register regNewRowid to + ** contain the new value. If the rowid is not being modified, + ** then regNewRowid is the same register as regOldRowid, which is + ** already populated. */ + assert( chngKey || pTrigger || hasFK || regOldRowid==regNewRowid ); + if( chngRowid ){ + sqlite3ExprCode(pParse, pRowidExpr, regNewRowid); + sqlite3VdbeAddOp1(v, OP_MustBeInt, regNewRowid); VdbeCoverage(v); + } + + /* Compute the old pre-UPDATE content of the row being changed, if that + ** information is needed */ + if( chngPk || hasFK || pTrigger ){ + u32 oldmask = (hasFK ? sqlite3FkOldmask(pParse, pTab) : 0); + oldmask |= sqlite3TriggerColmask(pParse, + pTrigger, pChanges, 0, TRIGGER_BEFORE|TRIGGER_AFTER, pTab, onError + ); + for(i=0; inCol; i++){ + if( oldmask==0xffffffff + || (i<32 && (oldmask & MASKBIT32(i))!=0) + || (pTab->aCol[i].colFlags & COLFLAG_PRIMKEY)!=0 + ){ + testcase( oldmask!=0xffffffff && i==31 ); + sqlite3ExprCodeGetColumnOfTable(v, pTab, iDataCur, i, regOld+i); + }else{ + sqlite3VdbeAddOp2(v, OP_Null, 0, regOld+i); + } + } + if( chngRowid==0 && pPk==0 ){ + sqlite3VdbeAddOp2(v, OP_Copy, regOldRowid, regNewRowid); + } + } + + /* Populate the array of registers beginning at regNew with the new + ** row data. This array is used to check constants, create the new + ** table and index records, and as the values for any new.* references + ** made by triggers. + ** + ** If there are one or more BEFORE triggers, then do not populate the + ** registers associated with columns that are (a) not modified by + ** this UPDATE statement and (b) not accessed by new.* references. The + ** values for registers not modified by the UPDATE must be reloaded from + ** the database after the BEFORE triggers are fired anyway (as the trigger + ** may have modified them). So not loading those that are not going to + ** be used eliminates some redundant opcodes. + */ + newmask = sqlite3TriggerColmask( + pParse, pTrigger, pChanges, 1, TRIGGER_BEFORE, pTab, onError + ); + for(i=0; inCol; i++){ + if( i==pTab->iPKey ){ + sqlite3VdbeAddOp2(v, OP_Null, 0, regNew+i); + }else{ + j = aXRef[i]; + if( j>=0 ){ + sqlite3ExprCode(pParse, pChanges->a[j].pExpr, regNew+i); + }else if( 0==(tmask&TRIGGER_BEFORE) || i>31 || (newmask & MASKBIT32(i)) ){ + /* This branch loads the value of a column that will not be changed + ** into a register. This is done if there are no BEFORE triggers, or + ** if there are one or more BEFORE triggers that use this value via + ** a new.* reference in a trigger program. + */ + testcase( i==31 ); + testcase( i==32 ); + sqlite3ExprCodeGetColumnOfTable(v, pTab, iDataCur, i, regNew+i); + }else{ + sqlite3VdbeAddOp2(v, OP_Null, 0, regNew+i); + } + } + } + + /* Fire any BEFORE UPDATE triggers. This happens before constraints are + ** verified. One could argue that this is wrong. + */ + if( tmask&TRIGGER_BEFORE ){ + sqlite3TableAffinity(v, pTab, regNew); + sqlite3CodeRowTrigger(pParse, pTrigger, TK_UPDATE, pChanges, + TRIGGER_BEFORE, pTab, regOldRowid, onError, labelContinue); + + /* The row-trigger may have deleted the row being updated. In this + ** case, jump to the next row. No updates or AFTER triggers are + ** required. This behavior - what happens when the row being updated + ** is deleted or renamed by a BEFORE trigger - is left undefined in the + ** documentation. + */ + if( pPk ){ + sqlite3VdbeAddOp4Int(v, OP_NotFound, iDataCur, labelContinue,regKey,nKey); + VdbeCoverage(v); + }else{ + sqlite3VdbeAddOp3(v, OP_NotExists, iDataCur, labelContinue, regOldRowid); + VdbeCoverage(v); + } + + /* After-BEFORE-trigger-reload-loop: + ** If it did not delete it, the BEFORE trigger may still have modified + ** some of the columns of the row being updated. Load the values for + ** all columns not modified by the update statement into their registers + ** in case this has happened. Only unmodified columns are reloaded. + ** The values computed for modified columns use the values before the + ** BEFORE trigger runs. See test case trigger1-18.0 (added 2018-04-26) + ** for an example. + */ + for(i=0; inCol; i++){ + if( aXRef[i]<0 && i!=pTab->iPKey ){ + sqlite3ExprCodeGetColumnOfTable(v, pTab, iDataCur, i, regNew+i); + } + } + } + + if( !isView ){ + /* Do constraint checks. */ + assert( regOldRowid>0 ); + sqlite3GenerateConstraintChecks(pParse, pTab, aRegIdx, iDataCur, iIdxCur, + regNewRowid, regOldRowid, chngKey, onError, labelContinue, &bReplace, + aXRef, 0); + + /* If REPLACE conflict handling may have been used, or if the PK of the + ** row is changing, then the GenerateConstraintChecks() above may have + ** moved cursor iDataCur. Reseek it. */ + if( bReplace || chngKey ){ + if( pPk ){ + sqlite3VdbeAddOp4Int(v, OP_NotFound,iDataCur,labelContinue,regKey,nKey); + }else{ + sqlite3VdbeAddOp3(v, OP_NotExists, iDataCur, labelContinue,regOldRowid); + } + VdbeCoverageNeverTaken(v); + } + + /* Do FK constraint checks. */ + if( hasFK ){ + sqlite3FkCheck(pParse, pTab, regOldRowid, 0, aXRef, chngKey); + } + + /* Delete the index entries associated with the current record. */ + sqlite3GenerateRowIndexDelete(pParse, pTab, iDataCur, iIdxCur, aRegIdx, -1); + + /* If changing the rowid value, or if there are foreign key constraints + ** to process, delete the old record. Otherwise, add a noop OP_Delete + ** to invoke the pre-update hook. + ** + ** That (regNew==regnewRowid+1) is true is also important for the + ** pre-update hook. If the caller invokes preupdate_new(), the returned + ** value is copied from memory cell (regNewRowid+1+iCol), where iCol + ** is the column index supplied by the user. + */ + assert( regNew==regNewRowid+1 ); +#ifdef SQLITE_ENABLE_PREUPDATE_HOOK + sqlite3VdbeAddOp3(v, OP_Delete, iDataCur, + OPFLAG_ISUPDATE | ((hasFK>1 || chngKey) ? 0 : OPFLAG_ISNOOP), + regNewRowid + ); + if( eOnePass==ONEPASS_MULTI ){ + assert( hasFK==0 && chngKey==0 ); + sqlite3VdbeChangeP5(v, OPFLAG_SAVEPOSITION); + } + if( !pParse->nested ){ + sqlite3VdbeAppendP4(v, pTab, P4_TABLE); + } +#else + if( hasFK>1 || chngKey ){ + sqlite3VdbeAddOp2(v, OP_Delete, iDataCur, 0); + } +#endif + + if( hasFK ){ + sqlite3FkCheck(pParse, pTab, 0, regNewRowid, aXRef, chngKey); + } + + /* Insert the new index entries and the new record. */ + sqlite3CompleteInsertion( + pParse, pTab, iDataCur, iIdxCur, regNewRowid, aRegIdx, + OPFLAG_ISUPDATE | (eOnePass==ONEPASS_MULTI ? OPFLAG_SAVEPOSITION : 0), + 0, 0 + ); + + /* Do any ON CASCADE, SET NULL or SET DEFAULT operations required to + ** handle rows (possibly in other tables) that refer via a foreign key + ** to the row just updated. */ + if( hasFK ){ + sqlite3FkActions(pParse, pTab, pChanges, regOldRowid, aXRef, chngKey); + } + } + + /* Increment the row counter + */ + if( regRowCount ){ + sqlite3VdbeAddOp2(v, OP_AddImm, regRowCount, 1); + } + + sqlite3CodeRowTrigger(pParse, pTrigger, TK_UPDATE, pChanges, + TRIGGER_AFTER, pTab, regOldRowid, onError, labelContinue); + + /* Repeat the above with the next record to be updated, until + ** all record selected by the WHERE clause have been updated. + */ + if( eOnePass==ONEPASS_SINGLE ){ + /* Nothing to do at end-of-loop for a single-pass */ + }else if( eOnePass==ONEPASS_MULTI ){ + sqlite3VdbeResolveLabel(v, labelContinue); + sqlite3WhereEnd(pWInfo); + }else if( pPk ){ + sqlite3VdbeResolveLabel(v, labelContinue); + sqlite3VdbeAddOp2(v, OP_Next, iEph, addrTop); VdbeCoverage(v); + }else{ + sqlite3VdbeGoto(v, labelContinue); + } + sqlite3VdbeResolveLabel(v, labelBreak); + + /* Update the sqlite_sequence table by storing the content of the + ** maximum rowid counter values recorded while inserting into + ** autoincrement tables. + */ + if( pParse->nested==0 && pParse->pTriggerTab==0 && pUpsert==0 ){ + sqlite3AutoincrementEnd(pParse); + } + + /* + ** Return the number of rows that were changed, if we are tracking + ** that information. + */ + if( regRowCount ){ + sqlite3VdbeAddOp2(v, OP_ResultRow, regRowCount, 1); + sqlite3VdbeSetNumCols(v, 1); + sqlite3VdbeSetColName(v, 0, COLNAME_NAME, "rows updated", SQLITE_STATIC); + } + +update_cleanup: + sqlite3AuthContextPop(&sContext); + sqlite3DbFree(db, aXRef); /* Also frees aRegIdx[] and aToOpen[] */ + sqlite3SrcListDelete(db, pTabList); + sqlite3ExprListDelete(db, pChanges); + sqlite3ExprDelete(db, pWhere); +#if defined(SQLITE_ENABLE_UPDATE_DELETE_LIMIT) + sqlite3ExprListDelete(db, pOrderBy); + sqlite3ExprDelete(db, pLimit); +#endif + return; +} +/* Make sure "isView" and other macros defined above are undefined. Otherwise +** they may interfere with compilation of other functions in this file +** (or in another file, if this file becomes part of the amalgamation). */ +#ifdef isView + #undef isView +#endif +#ifdef pTrigger + #undef pTrigger +#endif + +#ifndef SQLITE_OMIT_VIRTUALTABLE +/* +** Generate code for an UPDATE of a virtual table. +** +** There are two possible strategies - the default and the special +** "onepass" strategy. Onepass is only used if the virtual table +** implementation indicates that pWhere may match at most one row. +** +** The default strategy is to create an ephemeral table that contains +** for each row to be changed: +** +** (A) The original rowid of that row. +** (B) The revised rowid for the row. +** (C) The content of every column in the row. +** +** Then loop through the contents of this ephemeral table executing a +** VUpdate for each row. When finished, drop the ephemeral table. +** +** The "onepass" strategy does not use an ephemeral table. Instead, it +** stores the same values (A, B and C above) in a register array and +** makes a single invocation of VUpdate. +*/ +static void updateVirtualTable( + Parse *pParse, /* The parsing context */ + SrcList *pSrc, /* The virtual table to be modified */ + Table *pTab, /* The virtual table */ + ExprList *pChanges, /* The columns to change in the UPDATE statement */ + Expr *pRowid, /* Expression used to recompute the rowid */ + int *aXRef, /* Mapping from columns of pTab to entries in pChanges */ + Expr *pWhere, /* WHERE clause of the UPDATE statement */ + int onError /* ON CONFLICT strategy */ +){ + Vdbe *v = pParse->pVdbe; /* Virtual machine under construction */ + int ephemTab; /* Table holding the result of the SELECT */ + int i; /* Loop counter */ + sqlite3 *db = pParse->db; /* Database connection */ + const char *pVTab = (const char*)sqlite3GetVTable(db, pTab); + WhereInfo *pWInfo; + int nArg = 2 + pTab->nCol; /* Number of arguments to VUpdate */ + int regArg; /* First register in VUpdate arg array */ + int regRec; /* Register in which to assemble record */ + int regRowid; /* Register for ephem table rowid */ + int iCsr = pSrc->a[0].iCursor; /* Cursor used for virtual table scan */ + int aDummy[2]; /* Unused arg for sqlite3WhereOkOnePass() */ + int eOnePass; /* True to use onepass strategy */ + int addr; /* Address of OP_OpenEphemeral */ + + /* Allocate nArg registers in which to gather the arguments for VUpdate. Then + ** create and open the ephemeral table in which the records created from + ** these arguments will be temporarily stored. */ + assert( v ); + ephemTab = pParse->nTab++; + addr= sqlite3VdbeAddOp2(v, OP_OpenEphemeral, ephemTab, nArg); + regArg = pParse->nMem + 1; + pParse->nMem += nArg; + regRec = ++pParse->nMem; + regRowid = ++pParse->nMem; + + /* Start scanning the virtual table */ + pWInfo = sqlite3WhereBegin(pParse, pSrc, pWhere, 0,0,WHERE_ONEPASS_DESIRED,0); + if( pWInfo==0 ) return; + + /* Populate the argument registers. */ + for(i=0; inCol; i++){ + if( aXRef[i]>=0 ){ + sqlite3ExprCode(pParse, pChanges->a[aXRef[i]].pExpr, regArg+2+i); + }else{ + sqlite3VdbeAddOp3(v, OP_VColumn, iCsr, i, regArg+2+i); + sqlite3VdbeChangeP5(v, OPFLAG_NOCHNG);/* Enable sqlite3_vtab_nochange() */ + } + } + if( HasRowid(pTab) ){ + sqlite3VdbeAddOp2(v, OP_Rowid, iCsr, regArg); + if( pRowid ){ + sqlite3ExprCode(pParse, pRowid, regArg+1); + }else{ + sqlite3VdbeAddOp2(v, OP_Rowid, iCsr, regArg+1); + } + }else{ + Index *pPk; /* PRIMARY KEY index */ + i16 iPk; /* PRIMARY KEY column */ + pPk = sqlite3PrimaryKeyIndex(pTab); + assert( pPk!=0 ); + assert( pPk->nKeyCol==1 ); + iPk = pPk->aiColumn[0]; + sqlite3VdbeAddOp3(v, OP_VColumn, iCsr, iPk, regArg); + sqlite3VdbeAddOp2(v, OP_SCopy, regArg+2+iPk, regArg+1); + } + + eOnePass = sqlite3WhereOkOnePass(pWInfo, aDummy); + + /* There is no ONEPASS_MULTI on virtual tables */ + assert( eOnePass==ONEPASS_OFF || eOnePass==ONEPASS_SINGLE ); + + if( eOnePass ){ + /* If using the onepass strategy, no-op out the OP_OpenEphemeral coded + ** above. */ + sqlite3VdbeChangeToNoop(v, addr); + sqlite3VdbeAddOp1(v, OP_Close, iCsr); + }else{ + /* Create a record from the argument register contents and insert it into + ** the ephemeral table. */ + sqlite3MultiWrite(pParse); + sqlite3VdbeAddOp3(v, OP_MakeRecord, regArg, nArg, regRec); +#ifdef SQLITE_DEBUG + /* Signal an assert() within OP_MakeRecord that it is allowed to + ** accept no-change records with serial_type 10 */ + sqlite3VdbeChangeP5(v, OPFLAG_NOCHNG_MAGIC); +#endif + sqlite3VdbeAddOp2(v, OP_NewRowid, ephemTab, regRowid); + sqlite3VdbeAddOp3(v, OP_Insert, ephemTab, regRec, regRowid); + } + + + if( eOnePass==ONEPASS_OFF ){ + /* End the virtual table scan */ + sqlite3WhereEnd(pWInfo); + + /* Begin scannning through the ephemeral table. */ + addr = sqlite3VdbeAddOp1(v, OP_Rewind, ephemTab); VdbeCoverage(v); + + /* Extract arguments from the current row of the ephemeral table and + ** invoke the VUpdate method. */ + for(i=0; ipUpsertTarget); + sqlite3ExprDelete(db, p->pUpsertTargetWhere); + sqlite3ExprListDelete(db, p->pUpsertSet); + sqlite3ExprDelete(db, p->pUpsertWhere); + sqlite3DbFree(db, p); + } +} + +/* +** Duplicate an Upsert object. +*/ +SQLITE_PRIVATE Upsert *sqlite3UpsertDup(sqlite3 *db, Upsert *p){ + if( p==0 ) return 0; + return sqlite3UpsertNew(db, + sqlite3ExprListDup(db, p->pUpsertTarget, 0), + sqlite3ExprDup(db, p->pUpsertTargetWhere, 0), + sqlite3ExprListDup(db, p->pUpsertSet, 0), + sqlite3ExprDup(db, p->pUpsertWhere, 0) + ); +} + +/* +** Create a new Upsert object. +*/ +SQLITE_PRIVATE Upsert *sqlite3UpsertNew( + sqlite3 *db, /* Determines which memory allocator to use */ + ExprList *pTarget, /* Target argument to ON CONFLICT, or NULL */ + Expr *pTargetWhere, /* Optional WHERE clause on the target */ + ExprList *pSet, /* UPDATE columns, or NULL for a DO NOTHING */ + Expr *pWhere /* WHERE clause for the ON CONFLICT UPDATE */ +){ + Upsert *pNew; + pNew = sqlite3DbMallocRaw(db, sizeof(Upsert)); + if( pNew==0 ){ + sqlite3ExprListDelete(db, pTarget); + sqlite3ExprDelete(db, pTargetWhere); + sqlite3ExprListDelete(db, pSet); + sqlite3ExprDelete(db, pWhere); + return 0; + }else{ + pNew->pUpsertTarget = pTarget; + pNew->pUpsertTargetWhere = pTargetWhere; + pNew->pUpsertSet = pSet; + pNew->pUpsertWhere = pWhere; + pNew->pUpsertIdx = 0; + } + return pNew; +} + +/* +** Analyze the ON CONFLICT clause described by pUpsert. Resolve all +** symbols in the conflict-target. +** +** Return SQLITE_OK if everything works, or an error code is something +** is wrong. +*/ +SQLITE_PRIVATE int sqlite3UpsertAnalyzeTarget( + Parse *pParse, /* The parsing context */ + SrcList *pTabList, /* Table into which we are inserting */ + Upsert *pUpsert /* The ON CONFLICT clauses */ +){ + Table *pTab; /* That table into which we are inserting */ + int rc; /* Result code */ + int iCursor; /* Cursor used by pTab */ + Index *pIdx; /* One of the indexes of pTab */ + ExprList *pTarget; /* The conflict-target clause */ + Expr *pTerm; /* One term of the conflict-target clause */ + NameContext sNC; /* Context for resolving symbolic names */ + Expr sCol[2]; /* Index column converted into an Expr */ + + assert( pTabList->nSrc==1 ); + assert( pTabList->a[0].pTab!=0 ); + assert( pUpsert!=0 ); + assert( pUpsert->pUpsertTarget!=0 ); + + /* Resolve all symbolic names in the conflict-target clause, which + ** includes both the list of columns and the optional partial-index + ** WHERE clause. + */ + memset(&sNC, 0, sizeof(sNC)); + sNC.pParse = pParse; + sNC.pSrcList = pTabList; + rc = sqlite3ResolveExprListNames(&sNC, pUpsert->pUpsertTarget); + if( rc ) return rc; + rc = sqlite3ResolveExprNames(&sNC, pUpsert->pUpsertTargetWhere); + if( rc ) return rc; + + /* Check to see if the conflict target matches the rowid. */ + pTab = pTabList->a[0].pTab; + pTarget = pUpsert->pUpsertTarget; + iCursor = pTabList->a[0].iCursor; + if( HasRowid(pTab) + && pTarget->nExpr==1 + && (pTerm = pTarget->a[0].pExpr)->op==TK_COLUMN + && pTerm->iColumn==XN_ROWID + ){ + /* The conflict-target is the rowid of the primary table */ + assert( pUpsert->pUpsertIdx==0 ); + return SQLITE_OK; + } + + /* Initialize sCol[0..1] to be an expression parse tree for a + ** single column of an index. The sCol[0] node will be the TK_COLLATE + ** operator and sCol[1] will be the TK_COLUMN operator. Code below + ** will populate the specific collation and column number values + ** prior to comparing against the conflict-target expression. + */ + memset(sCol, 0, sizeof(sCol)); + sCol[0].op = TK_COLLATE; + sCol[0].pLeft = &sCol[1]; + sCol[1].op = TK_COLUMN; + sCol[1].iTable = pTabList->a[0].iCursor; + + /* Check for matches against other indexes */ + for(pIdx=pTab->pIndex; pIdx; pIdx=pIdx->pNext){ + int ii, jj, nn; + if( !IsUniqueIndex(pIdx) ) continue; + if( pTarget->nExpr!=pIdx->nKeyCol ) continue; + if( pIdx->pPartIdxWhere ){ + if( pUpsert->pUpsertTargetWhere==0 ) continue; + if( sqlite3ExprCompare(pParse, pUpsert->pUpsertTargetWhere, + pIdx->pPartIdxWhere, iCursor)!=0 ){ + continue; + } + } + nn = pIdx->nKeyCol; + for(ii=0; iiazColl[ii]; + if( pIdx->aiColumn[ii]==XN_EXPR ){ + assert( pIdx->aColExpr!=0 ); + assert( pIdx->aColExpr->nExpr>ii ); + pExpr = pIdx->aColExpr->a[ii].pExpr; + if( pExpr->op!=TK_COLLATE ){ + sCol[0].pLeft = pExpr; + pExpr = &sCol[0]; + } + }else{ + sCol[0].pLeft = &sCol[1]; + sCol[1].iColumn = pIdx->aiColumn[ii]; + pExpr = &sCol[0]; + } + for(jj=0; jja[jj].pExpr, pExpr,iCursor)<2 ){ + break; /* Column ii of the index matches column jj of target */ + } + } + if( jj>=nn ){ + /* The target contains no match for column jj of the index */ + break; + } + } + if( iipUpsertIdx = pIdx; + return SQLITE_OK; + } + sqlite3ErrorMsg(pParse, "ON CONFLICT clause does not match any " + "PRIMARY KEY or UNIQUE constraint"); + return SQLITE_ERROR; +} + +/* +** Generate bytecode that does an UPDATE as part of an upsert. +** +** If pIdx is NULL, then the UNIQUE constraint that failed was the IPK. +** In this case parameter iCur is a cursor open on the table b-tree that +** currently points to the conflicting table row. Otherwise, if pIdx +** is not NULL, then pIdx is the constraint that failed and iCur is a +** cursor points to the conflicting row. +*/ +SQLITE_PRIVATE void sqlite3UpsertDoUpdate( + Parse *pParse, /* The parsing and code-generating context */ + Upsert *pUpsert, /* The ON CONFLICT clause for the upsert */ + Table *pTab, /* The table being updated */ + Index *pIdx, /* The UNIQUE constraint that failed */ + int iCur /* Cursor for pIdx (or pTab if pIdx==NULL) */ +){ + Vdbe *v = pParse->pVdbe; + sqlite3 *db = pParse->db; + SrcList *pSrc; /* FROM clause for the UPDATE */ + int iDataCur; + int i; + + assert( v!=0 ); + assert( pUpsert!=0 ); + VdbeNoopComment((v, "Begin DO UPDATE of UPSERT")); + iDataCur = pUpsert->iDataCur; + if( pIdx && iCur!=iDataCur ){ + if( HasRowid(pTab) ){ + int regRowid = sqlite3GetTempReg(pParse); + sqlite3VdbeAddOp2(v, OP_IdxRowid, iCur, regRowid); + sqlite3VdbeAddOp3(v, OP_SeekRowid, iDataCur, 0, regRowid); + VdbeCoverage(v); + sqlite3ReleaseTempReg(pParse, regRowid); + }else{ + Index *pPk = sqlite3PrimaryKeyIndex(pTab); + int nPk = pPk->nKeyCol; + int iPk = pParse->nMem+1; + pParse->nMem += nPk; + for(i=0; iaiColumn[i]>=0 ); + k = sqlite3ColumnOfIndex(pIdx, pPk->aiColumn[i]); + sqlite3VdbeAddOp3(v, OP_Column, iCur, k, iPk+i); + VdbeComment((v, "%s.%s", pIdx->zName, + pTab->aCol[pPk->aiColumn[i]].zName)); + } + sqlite3VdbeVerifyAbortable(v, OE_Abort); + i = sqlite3VdbeAddOp4Int(v, OP_Found, iDataCur, 0, iPk, nPk); + VdbeCoverage(v); + sqlite3VdbeAddOp4(v, OP_Halt, SQLITE_CORRUPT, OE_Abort, 0, + "corrupt database", P4_STATIC); + sqlite3VdbeJumpHere(v, i); + } + } + /* pUpsert does not own pUpsertSrc - the outer INSERT statement does. So + ** we have to make a copy before passing it down into sqlite3Update() */ + pSrc = sqlite3SrcListDup(db, pUpsert->pUpsertSrc, 0); + /* excluded.* columns of type REAL need to be converted to a hard real */ + for(i=0; inCol; i++){ + if( pTab->aCol[i].affinity==SQLITE_AFF_REAL ){ + sqlite3VdbeAddOp1(v, OP_RealAffinity, pUpsert->regData+i); + } + } + sqlite3Update(pParse, pSrc, pUpsert->pUpsertSet, + pUpsert->pUpsertWhere, OE_Abort, 0, 0, pUpsert); + pUpsert->pUpsertSet = 0; /* Will have been deleted by sqlite3Update() */ + pUpsert->pUpsertWhere = 0; /* Will have been deleted by sqlite3Update() */ + VdbeNoopComment((v, "End DO UPDATE of UPSERT")); +} + +#endif /* SQLITE_OMIT_UPSERT */ + +/************** End of upsert.c **********************************************/ +/************** Begin file vacuum.c ******************************************/ +/* +** 2003 April 6 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** This file contains code used to implement the VACUUM command. +** +** Most of the code in this file may be omitted by defining the +** SQLITE_OMIT_VACUUM macro. +*/ +/* #include "sqliteInt.h" */ +/* #include "vdbeInt.h" */ + +#if !defined(SQLITE_OMIT_VACUUM) && !defined(SQLITE_OMIT_ATTACH) + +/* +** Execute zSql on database db. +** +** If zSql returns rows, then each row will have exactly one +** column. (This will only happen if zSql begins with "SELECT".) +** Take each row of result and call execSql() again recursively. +** +** The execSqlF() routine does the same thing, except it accepts +** a format string as its third argument +*/ +static int execSql(sqlite3 *db, char **pzErrMsg, const char *zSql){ + sqlite3_stmt *pStmt; + int rc; + + /* printf("SQL: [%s]\n", zSql); fflush(stdout); */ + rc = sqlite3_prepare_v2(db, zSql, -1, &pStmt, 0); + if( rc!=SQLITE_OK ) return rc; + while( SQLITE_ROW==(rc = sqlite3_step(pStmt)) ){ + const char *zSubSql = (const char*)sqlite3_column_text(pStmt,0); + assert( sqlite3_strnicmp(zSql,"SELECT",6)==0 ); + /* The secondary SQL must be one of CREATE TABLE, CREATE INDEX, + ** or INSERT. Historically there have been attacks that first + ** corrupt the sqlite_master.sql field with other kinds of statements + ** then run VACUUM to get those statements to execute at inappropriate + ** times. */ + if( zSubSql + && (strncmp(zSubSql,"CRE",3)==0 || strncmp(zSubSql,"INS",3)==0) + ){ + rc = execSql(db, pzErrMsg, zSubSql); + if( rc!=SQLITE_OK ) break; + } + } + assert( rc!=SQLITE_ROW ); + if( rc==SQLITE_DONE ) rc = SQLITE_OK; + if( rc ){ + sqlite3SetString(pzErrMsg, db, sqlite3_errmsg(db)); + } + (void)sqlite3_finalize(pStmt); + return rc; +} +static int execSqlF(sqlite3 *db, char **pzErrMsg, const char *zSql, ...){ + char *z; + va_list ap; + int rc; + va_start(ap, zSql); + z = sqlite3VMPrintf(db, zSql, ap); + va_end(ap); + if( z==0 ) return SQLITE_NOMEM; + rc = execSql(db, pzErrMsg, z); + sqlite3DbFree(db, z); + return rc; +} + +/* +** The VACUUM command is used to clean up the database, +** collapse free space, etc. It is modelled after the VACUUM command +** in PostgreSQL. The VACUUM command works as follows: +** +** (1) Create a new transient database file +** (2) Copy all content from the database being vacuumed into +** the new transient database file +** (3) Copy content from the transient database back into the +** original database. +** +** The transient database requires temporary disk space approximately +** equal to the size of the original database. The copy operation of +** step (3) requires additional temporary disk space approximately equal +** to the size of the original database for the rollback journal. +** Hence, temporary disk space that is approximately 2x the size of the +** original database is required. Every page of the database is written +** approximately 3 times: Once for step (2) and twice for step (3). +** Two writes per page are required in step (3) because the original +** database content must be written into the rollback journal prior to +** overwriting the database with the vacuumed content. +** +** Only 1x temporary space and only 1x writes would be required if +** the copy of step (3) were replaced by deleting the original database +** and renaming the transient database as the original. But that will +** not work if other processes are attached to the original database. +** And a power loss in between deleting the original and renaming the +** transient would cause the database file to appear to be deleted +** following reboot. +*/ +SQLITE_PRIVATE void sqlite3Vacuum(Parse *pParse, Token *pNm, Expr *pInto){ + Vdbe *v = sqlite3GetVdbe(pParse); + int iDb = 0; + if( v==0 ) goto build_vacuum_end; + if( pParse->nErr ) goto build_vacuum_end; + if( pNm ){ +#ifndef SQLITE_BUG_COMPATIBLE_20160819 + /* Default behavior: Report an error if the argument to VACUUM is + ** not recognized */ + iDb = sqlite3TwoPartName(pParse, pNm, pNm, &pNm); + if( iDb<0 ) goto build_vacuum_end; +#else + /* When SQLITE_BUG_COMPATIBLE_20160819 is defined, unrecognized arguments + ** to VACUUM are silently ignored. This is a back-out of a bug fix that + ** occurred on 2016-08-19 (https://www.sqlite.org/src/info/083f9e6270). + ** The buggy behavior is required for binary compatibility with some + ** legacy applications. */ + iDb = sqlite3FindDb(pParse->db, pNm); + if( iDb<0 ) iDb = 0; +#endif + } + if( iDb!=1 ){ + int iIntoReg = 0; + if( pInto && sqlite3ResolveSelfReference(pParse,0,0,pInto,0)==0 ){ + iIntoReg = ++pParse->nMem; + sqlite3ExprCode(pParse, pInto, iIntoReg); + } + sqlite3VdbeAddOp2(v, OP_Vacuum, iDb, iIntoReg); + sqlite3VdbeUsesBtree(v, iDb); + } +build_vacuum_end: + sqlite3ExprDelete(pParse->db, pInto); + return; +} + +/* +** This routine implements the OP_Vacuum opcode of the VDBE. +*/ +SQLITE_PRIVATE SQLITE_NOINLINE int sqlite3RunVacuum( + char **pzErrMsg, /* Write error message here */ + sqlite3 *db, /* Database connection */ + int iDb, /* Which attached DB to vacuum */ + sqlite3_value *pOut /* Write results here, if not NULL. VACUUM INTO */ +){ + int rc = SQLITE_OK; /* Return code from service routines */ + Btree *pMain; /* The database being vacuumed */ + Btree *pTemp; /* The temporary database we vacuum into */ + u32 saved_mDbFlags; /* Saved value of db->mDbFlags */ + u64 saved_flags; /* Saved value of db->flags */ + int saved_nChange; /* Saved value of db->nChange */ + int saved_nTotalChange; /* Saved value of db->nTotalChange */ + u32 saved_openFlags; /* Saved value of db->openFlags */ + u8 saved_mTrace; /* Saved trace settings */ + Db *pDb = 0; /* Database to detach at end of vacuum */ + int isMemDb; /* True if vacuuming a :memory: database */ + int nRes; /* Bytes of reserved space at the end of each page */ + int nDb; /* Number of attached databases */ + const char *zDbMain; /* Schema name of database to vacuum */ + const char *zOut; /* Name of output file */ + + if( !db->autoCommit ){ + sqlite3SetString(pzErrMsg, db, "cannot VACUUM from within a transaction"); + return SQLITE_ERROR; /* IMP: R-12218-18073 */ + } + if( db->nVdbeActive>1 ){ + sqlite3SetString(pzErrMsg, db,"cannot VACUUM - SQL statements in progress"); + return SQLITE_ERROR; /* IMP: R-15610-35227 */ + } + saved_openFlags = db->openFlags; + if( pOut ){ + if( sqlite3_value_type(pOut)!=SQLITE_TEXT ){ + sqlite3SetString(pzErrMsg, db, "non-text filename"); + return SQLITE_ERROR; + } + zOut = (const char*)sqlite3_value_text(pOut); + db->openFlags &= ~SQLITE_OPEN_READONLY; + db->openFlags |= SQLITE_OPEN_CREATE|SQLITE_OPEN_READWRITE; + }else{ + zOut = ""; + } + + /* Save the current value of the database flags so that it can be + ** restored before returning. Then set the writable-schema flag, and + ** disable CHECK and foreign key constraints. */ + saved_flags = db->flags; + saved_mDbFlags = db->mDbFlags; + saved_nChange = db->nChange; + saved_nTotalChange = db->nTotalChange; + saved_mTrace = db->mTrace; + db->flags |= SQLITE_WriteSchema | SQLITE_IgnoreChecks; + db->mDbFlags |= DBFLAG_PreferBuiltin | DBFLAG_Vacuum; + db->flags &= ~(u64)(SQLITE_ForeignKeys | SQLITE_ReverseOrder + | SQLITE_Defensive | SQLITE_CountRows); + db->mTrace = 0; + + zDbMain = db->aDb[iDb].zDbSName; + pMain = db->aDb[iDb].pBt; + isMemDb = sqlite3PagerIsMemdb(sqlite3BtreePager(pMain)); + + /* Attach the temporary database as 'vacuum_db'. The synchronous pragma + ** can be set to 'off' for this file, as it is not recovered if a crash + ** occurs anyway. The integrity of the database is maintained by a + ** (possibly synchronous) transaction opened on the main database before + ** sqlite3BtreeCopyFile() is called. + ** + ** An optimisation would be to use a non-journaled pager. + ** (Later:) I tried setting "PRAGMA vacuum_db.journal_mode=OFF" but + ** that actually made the VACUUM run slower. Very little journalling + ** actually occurs when doing a vacuum since the vacuum_db is initially + ** empty. Only the journal header is written. Apparently it takes more + ** time to parse and run the PRAGMA to turn journalling off than it does + ** to write the journal header file. + */ + nDb = db->nDb; + rc = execSqlF(db, pzErrMsg, "ATTACH %Q AS vacuum_db", zOut); + db->openFlags = saved_openFlags; + if( rc!=SQLITE_OK ) goto end_of_vacuum; + assert( (db->nDb-1)==nDb ); + pDb = &db->aDb[nDb]; + assert( strcmp(pDb->zDbSName,"vacuum_db")==0 ); + pTemp = pDb->pBt; + if( pOut ){ + sqlite3_file *id = sqlite3PagerFile(sqlite3BtreePager(pTemp)); + i64 sz = 0; + if( id->pMethods!=0 && (sqlite3OsFileSize(id, &sz)!=SQLITE_OK || sz>0) ){ + rc = SQLITE_ERROR; + sqlite3SetString(pzErrMsg, db, "output file already exists"); + goto end_of_vacuum; + } + db->mDbFlags |= DBFLAG_VacuumInto; + } + nRes = sqlite3BtreeGetOptimalReserve(pMain); + + /* A VACUUM cannot change the pagesize of an encrypted database. */ +#ifdef SQLITE_HAS_CODEC + if( db->nextPagesize ){ + extern void sqlite3CodecGetKey(sqlite3*, int, void**, int*); + int nKey; + char *zKey; + sqlite3CodecGetKey(db, iDb, (void**)&zKey, &nKey); + if( nKey ) db->nextPagesize = 0; + } +#endif + + sqlite3BtreeSetCacheSize(pTemp, db->aDb[iDb].pSchema->cache_size); + sqlite3BtreeSetSpillSize(pTemp, sqlite3BtreeSetSpillSize(pMain,0)); + sqlite3BtreeSetPagerFlags(pTemp, PAGER_SYNCHRONOUS_OFF|PAGER_CACHESPILL); + + /* Begin a transaction and take an exclusive lock on the main database + ** file. This is done before the sqlite3BtreeGetPageSize(pMain) call below, + ** to ensure that we do not try to change the page-size on a WAL database. + */ + rc = execSql(db, pzErrMsg, "BEGIN"); + if( rc!=SQLITE_OK ) goto end_of_vacuum; + rc = sqlite3BtreeBeginTrans(pMain, pOut==0 ? 2 : 0, 0); + if( rc!=SQLITE_OK ) goto end_of_vacuum; + + /* Do not attempt to change the page size for a WAL database */ + if( sqlite3PagerGetJournalMode(sqlite3BtreePager(pMain)) + ==PAGER_JOURNALMODE_WAL ){ + db->nextPagesize = 0; + } + + if( sqlite3BtreeSetPageSize(pTemp, sqlite3BtreeGetPageSize(pMain), nRes, 0) + || (!isMemDb && sqlite3BtreeSetPageSize(pTemp, db->nextPagesize, nRes, 0)) + || NEVER(db->mallocFailed) + ){ + rc = SQLITE_NOMEM_BKPT; + goto end_of_vacuum; + } + +#ifndef SQLITE_OMIT_AUTOVACUUM + sqlite3BtreeSetAutoVacuum(pTemp, db->nextAutovac>=0 ? db->nextAutovac : + sqlite3BtreeGetAutoVacuum(pMain)); +#endif + + /* Query the schema of the main database. Create a mirror schema + ** in the temporary database. + */ + db->init.iDb = nDb; /* force new CREATE statements into vacuum_db */ + rc = execSqlF(db, pzErrMsg, + "SELECT sql FROM \"%w\".sqlite_master" + " WHERE type='table'AND name<>'sqlite_sequence'" + " AND coalesce(rootpage,1)>0", + zDbMain + ); + if( rc!=SQLITE_OK ) goto end_of_vacuum; + rc = execSqlF(db, pzErrMsg, + "SELECT sql FROM \"%w\".sqlite_master" + " WHERE type='index'", + zDbMain + ); + if( rc!=SQLITE_OK ) goto end_of_vacuum; + db->init.iDb = 0; + + /* Loop through the tables in the main database. For each, do + ** an "INSERT INTO vacuum_db.xxx SELECT * FROM main.xxx;" to copy + ** the contents to the temporary database. + */ + rc = execSqlF(db, pzErrMsg, + "SELECT'INSERT INTO vacuum_db.'||quote(name)" + "||' SELECT*FROM\"%w\".'||quote(name)" + "FROM vacuum_db.sqlite_master " + "WHERE type='table'AND coalesce(rootpage,1)>0", + zDbMain + ); + assert( (db->mDbFlags & DBFLAG_Vacuum)!=0 ); + db->mDbFlags &= ~DBFLAG_Vacuum; + if( rc!=SQLITE_OK ) goto end_of_vacuum; + + /* Copy the triggers, views, and virtual tables from the main database + ** over to the temporary database. None of these objects has any + ** associated storage, so all we have to do is copy their entries + ** from the SQLITE_MASTER table. + */ + rc = execSqlF(db, pzErrMsg, + "INSERT INTO vacuum_db.sqlite_master" + " SELECT*FROM \"%w\".sqlite_master" + " WHERE type IN('view','trigger')" + " OR(type='table'AND rootpage=0)", + zDbMain + ); + if( rc ) goto end_of_vacuum; + + /* At this point, there is a write transaction open on both the + ** vacuum database and the main database. Assuming no error occurs, + ** both transactions are closed by this block - the main database + ** transaction by sqlite3BtreeCopyFile() and the other by an explicit + ** call to sqlite3BtreeCommit(). + */ + { + u32 meta; + int i; + + /* This array determines which meta meta values are preserved in the + ** vacuum. Even entries are the meta value number and odd entries + ** are an increment to apply to the meta value after the vacuum. + ** The increment is used to increase the schema cookie so that other + ** connections to the same database will know to reread the schema. + */ + static const unsigned char aCopy[] = { + BTREE_SCHEMA_VERSION, 1, /* Add one to the old schema cookie */ + BTREE_DEFAULT_CACHE_SIZE, 0, /* Preserve the default page cache size */ + BTREE_TEXT_ENCODING, 0, /* Preserve the text encoding */ + BTREE_USER_VERSION, 0, /* Preserve the user version */ + BTREE_APPLICATION_ID, 0, /* Preserve the application id */ + }; + + assert( 1==sqlite3BtreeIsInTrans(pTemp) ); + assert( pOut!=0 || 1==sqlite3BtreeIsInTrans(pMain) ); + + /* Copy Btree meta values */ + for(i=0; iflags */ + db->init.iDb = 0; + db->mDbFlags = saved_mDbFlags; + db->flags = saved_flags; + db->nChange = saved_nChange; + db->nTotalChange = saved_nTotalChange; + db->mTrace = saved_mTrace; + sqlite3BtreeSetPageSize(pMain, -1, -1, 1); + + /* Currently there is an SQL level transaction open on the vacuum + ** database. No locks are held on any other files (since the main file + ** was committed at the btree level). So it safe to end the transaction + ** by manually setting the autoCommit flag to true and detaching the + ** vacuum database. The vacuum_db journal file is deleted when the pager + ** is closed by the DETACH. + */ + db->autoCommit = 1; + + if( pDb ){ + sqlite3BtreeClose(pDb->pBt); + pDb->pBt = 0; + pDb->pSchema = 0; + } + + /* This both clears the schemas and reduces the size of the db->aDb[] + ** array. */ + sqlite3ResetAllSchemasOfConnection(db); + + return rc; +} + +#endif /* SQLITE_OMIT_VACUUM && SQLITE_OMIT_ATTACH */ + +/************** End of vacuum.c **********************************************/ +/************** Begin file vtab.c ********************************************/ +/* +** 2006 June 10 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** This file contains code used to help implement virtual tables. +*/ +#ifndef SQLITE_OMIT_VIRTUALTABLE +/* #include "sqliteInt.h" */ + +/* +** Before a virtual table xCreate() or xConnect() method is invoked, the +** sqlite3.pVtabCtx member variable is set to point to an instance of +** this struct allocated on the stack. It is used by the implementation of +** the sqlite3_declare_vtab() and sqlite3_vtab_config() APIs, both of which +** are invoked only from within xCreate and xConnect methods. +*/ +struct VtabCtx { + VTable *pVTable; /* The virtual table being constructed */ + Table *pTab; /* The Table object to which the virtual table belongs */ + VtabCtx *pPrior; /* Parent context (if any) */ + int bDeclared; /* True after sqlite3_declare_vtab() is called */ +}; + +/* +** Construct and install a Module object for a virtual table. When this +** routine is called, it is guaranteed that all appropriate locks are held +** and the module is not already part of the connection. +** +** If there already exists a module with zName, replace it with the new one. +** If pModule==0, then delete the module zName if it exists. +*/ +SQLITE_PRIVATE Module *sqlite3VtabCreateModule( + sqlite3 *db, /* Database in which module is registered */ + const char *zName, /* Name assigned to this module */ + const sqlite3_module *pModule, /* The definition of the module */ + void *pAux, /* Context pointer for xCreate/xConnect */ + void (*xDestroy)(void *) /* Module destructor function */ +){ + Module *pMod; + Module *pDel; + char *zCopy; + if( pModule==0 ){ + zCopy = (char*)zName; + pMod = 0; + }else{ + int nName = sqlite3Strlen30(zName); + pMod = (Module *)sqlite3Malloc(sizeof(Module) + nName + 1); + if( pMod==0 ){ + sqlite3OomFault(db); + return 0; + } + zCopy = (char *)(&pMod[1]); + memcpy(zCopy, zName, nName+1); + pMod->zName = zCopy; + pMod->pModule = pModule; + pMod->pAux = pAux; + pMod->xDestroy = xDestroy; + pMod->pEpoTab = 0; + pMod->nRefModule = 1; + } + pDel = (Module *)sqlite3HashInsert(&db->aModule,zCopy,(void*)pMod); + if( pDel ){ + if( pDel==pMod ){ + sqlite3OomFault(db); + sqlite3DbFree(db, pDel); + pMod = 0; + }else{ + sqlite3VtabEponymousTableClear(db, pDel); + sqlite3VtabModuleUnref(db, pDel); + } + } + return pMod; +} + +/* +** The actual function that does the work of creating a new module. +** This function implements the sqlite3_create_module() and +** sqlite3_create_module_v2() interfaces. +*/ +static int createModule( + sqlite3 *db, /* Database in which module is registered */ + const char *zName, /* Name assigned to this module */ + const sqlite3_module *pModule, /* The definition of the module */ + void *pAux, /* Context pointer for xCreate/xConnect */ + void (*xDestroy)(void *) /* Module destructor function */ +){ + int rc = SQLITE_OK; + + sqlite3_mutex_enter(db->mutex); + (void)sqlite3VtabCreateModule(db, zName, pModule, pAux, xDestroy); + rc = sqlite3ApiExit(db, rc); + if( rc!=SQLITE_OK && xDestroy ) xDestroy(pAux); + sqlite3_mutex_leave(db->mutex); + return rc; +} + + +/* +** External API function used to create a new virtual-table module. +*/ +SQLITE_API int sqlite3_create_module( + sqlite3 *db, /* Database in which module is registered */ + const char *zName, /* Name assigned to this module */ + const sqlite3_module *pModule, /* The definition of the module */ + void *pAux /* Context pointer for xCreate/xConnect */ +){ +#ifdef SQLITE_ENABLE_API_ARMOR + if( !sqlite3SafetyCheckOk(db) || zName==0 ) return SQLITE_MISUSE_BKPT; +#endif + return createModule(db, zName, pModule, pAux, 0); +} + +/* +** External API function used to create a new virtual-table module. +*/ +SQLITE_API int sqlite3_create_module_v2( + sqlite3 *db, /* Database in which module is registered */ + const char *zName, /* Name assigned to this module */ + const sqlite3_module *pModule, /* The definition of the module */ + void *pAux, /* Context pointer for xCreate/xConnect */ + void (*xDestroy)(void *) /* Module destructor function */ +){ +#ifdef SQLITE_ENABLE_API_ARMOR + if( !sqlite3SafetyCheckOk(db) || zName==0 ) return SQLITE_MISUSE_BKPT; +#endif + return createModule(db, zName, pModule, pAux, xDestroy); +} + +/* +** External API to drop all virtual-table modules, except those named +** on the azNames list. +*/ +SQLITE_API int sqlite3_drop_modules(sqlite3 *db, const char** azNames){ + HashElem *pThis, *pNext; +#ifdef SQLITE_ENABLE_API_ARMOR + if( !sqlite3SafetyCheckOk(db) ) return SQLITE_MISUSE_BKPT; +#endif + for(pThis=sqliteHashFirst(&db->aModule); pThis; pThis=pNext){ + Module *pMod = (Module*)sqliteHashData(pThis); + pNext = sqliteHashNext(pThis); + if( azNames ){ + int ii; + for(ii=0; azNames[ii]!=0 && strcmp(azNames[ii],pMod->zName)!=0; ii++){} + if( azNames[ii]!=0 ) continue; + } + createModule(db, pMod->zName, 0, 0, 0); + } + return SQLITE_OK; +} + +/* +** Decrement the reference count on a Module object. Destroy the +** module when the reference count reaches zero. +*/ +SQLITE_PRIVATE void sqlite3VtabModuleUnref(sqlite3 *db, Module *pMod){ + assert( pMod->nRefModule>0 ); + pMod->nRefModule--; + if( pMod->nRefModule==0 ){ + if( pMod->xDestroy ){ + pMod->xDestroy(pMod->pAux); + } + assert( pMod->pEpoTab==0 ); + sqlite3DbFree(db, pMod); + } +} + +/* +** Lock the virtual table so that it cannot be disconnected. +** Locks nest. Every lock should have a corresponding unlock. +** If an unlock is omitted, resources leaks will occur. +** +** If a disconnect is attempted while a virtual table is locked, +** the disconnect is deferred until all locks have been removed. +*/ +SQLITE_PRIVATE void sqlite3VtabLock(VTable *pVTab){ + pVTab->nRef++; +} + + +/* +** pTab is a pointer to a Table structure representing a virtual-table. +** Return a pointer to the VTable object used by connection db to access +** this virtual-table, if one has been created, or NULL otherwise. +*/ +SQLITE_PRIVATE VTable *sqlite3GetVTable(sqlite3 *db, Table *pTab){ + VTable *pVtab; + assert( IsVirtual(pTab) ); + for(pVtab=pTab->pVTable; pVtab && pVtab->db!=db; pVtab=pVtab->pNext); + return pVtab; +} + +/* +** Decrement the ref-count on a virtual table object. When the ref-count +** reaches zero, call the xDisconnect() method to delete the object. +*/ +SQLITE_PRIVATE void sqlite3VtabUnlock(VTable *pVTab){ + sqlite3 *db = pVTab->db; + + assert( db ); + assert( pVTab->nRef>0 ); + assert( db->magic==SQLITE_MAGIC_OPEN || db->magic==SQLITE_MAGIC_ZOMBIE ); + + pVTab->nRef--; + if( pVTab->nRef==0 ){ + sqlite3_vtab *p = pVTab->pVtab; + sqlite3VtabModuleUnref(pVTab->db, pVTab->pMod); + if( p ){ + p->pModule->xDisconnect(p); + } + sqlite3DbFree(db, pVTab); + } +} + +/* +** Table p is a virtual table. This function moves all elements in the +** p->pVTable list to the sqlite3.pDisconnect lists of their associated +** database connections to be disconnected at the next opportunity. +** Except, if argument db is not NULL, then the entry associated with +** connection db is left in the p->pVTable list. +*/ +static VTable *vtabDisconnectAll(sqlite3 *db, Table *p){ + VTable *pRet = 0; + VTable *pVTable = p->pVTable; + p->pVTable = 0; + + /* Assert that the mutex (if any) associated with the BtShared database + ** that contains table p is held by the caller. See header comments + ** above function sqlite3VtabUnlockList() for an explanation of why + ** this makes it safe to access the sqlite3.pDisconnect list of any + ** database connection that may have an entry in the p->pVTable list. + */ + assert( db==0 || sqlite3SchemaMutexHeld(db, 0, p->pSchema) ); + + while( pVTable ){ + sqlite3 *db2 = pVTable->db; + VTable *pNext = pVTable->pNext; + assert( db2 ); + if( db2==db ){ + pRet = pVTable; + p->pVTable = pRet; + pRet->pNext = 0; + }else{ + pVTable->pNext = db2->pDisconnect; + db2->pDisconnect = pVTable; + } + pVTable = pNext; + } + + assert( !db || pRet ); + return pRet; +} + +/* +** Table *p is a virtual table. This function removes the VTable object +** for table *p associated with database connection db from the linked +** list in p->pVTab. It also decrements the VTable ref count. This is +** used when closing database connection db to free all of its VTable +** objects without disturbing the rest of the Schema object (which may +** be being used by other shared-cache connections). +*/ +SQLITE_PRIVATE void sqlite3VtabDisconnect(sqlite3 *db, Table *p){ + VTable **ppVTab; + + assert( IsVirtual(p) ); + assert( sqlite3BtreeHoldsAllMutexes(db) ); + assert( sqlite3_mutex_held(db->mutex) ); + + for(ppVTab=&p->pVTable; *ppVTab; ppVTab=&(*ppVTab)->pNext){ + if( (*ppVTab)->db==db ){ + VTable *pVTab = *ppVTab; + *ppVTab = pVTab->pNext; + sqlite3VtabUnlock(pVTab); + break; + } + } +} + + +/* +** Disconnect all the virtual table objects in the sqlite3.pDisconnect list. +** +** This function may only be called when the mutexes associated with all +** shared b-tree databases opened using connection db are held by the +** caller. This is done to protect the sqlite3.pDisconnect list. The +** sqlite3.pDisconnect list is accessed only as follows: +** +** 1) By this function. In this case, all BtShared mutexes and the mutex +** associated with the database handle itself must be held. +** +** 2) By function vtabDisconnectAll(), when it adds a VTable entry to +** the sqlite3.pDisconnect list. In this case either the BtShared mutex +** associated with the database the virtual table is stored in is held +** or, if the virtual table is stored in a non-sharable database, then +** the database handle mutex is held. +** +** As a result, a sqlite3.pDisconnect cannot be accessed simultaneously +** by multiple threads. It is thread-safe. +*/ +SQLITE_PRIVATE void sqlite3VtabUnlockList(sqlite3 *db){ + VTable *p = db->pDisconnect; + db->pDisconnect = 0; + + assert( sqlite3BtreeHoldsAllMutexes(db) ); + assert( sqlite3_mutex_held(db->mutex) ); + + if( p ){ + sqlite3ExpirePreparedStatements(db, 0); + do { + VTable *pNext = p->pNext; + sqlite3VtabUnlock(p); + p = pNext; + }while( p ); + } +} + +/* +** Clear any and all virtual-table information from the Table record. +** This routine is called, for example, just before deleting the Table +** record. +** +** Since it is a virtual-table, the Table structure contains a pointer +** to the head of a linked list of VTable structures. Each VTable +** structure is associated with a single sqlite3* user of the schema. +** The reference count of the VTable structure associated with database +** connection db is decremented immediately (which may lead to the +** structure being xDisconnected and free). Any other VTable structures +** in the list are moved to the sqlite3.pDisconnect list of the associated +** database connection. +*/ +SQLITE_PRIVATE void sqlite3VtabClear(sqlite3 *db, Table *p){ + if( !db || db->pnBytesFreed==0 ) vtabDisconnectAll(0, p); + if( p->azModuleArg ){ + int i; + for(i=0; inModuleArg; i++){ + if( i!=1 ) sqlite3DbFree(db, p->azModuleArg[i]); + } + sqlite3DbFree(db, p->azModuleArg); + } +} + +/* +** Add a new module argument to pTable->azModuleArg[]. +** The string is not copied - the pointer is stored. The +** string will be freed automatically when the table is +** deleted. +*/ +static void addModuleArgument(Parse *pParse, Table *pTable, char *zArg){ + sqlite3_int64 nBytes = sizeof(char *)*(2+pTable->nModuleArg); + char **azModuleArg; + sqlite3 *db = pParse->db; + if( pTable->nModuleArg+3>=db->aLimit[SQLITE_LIMIT_COLUMN] ){ + sqlite3ErrorMsg(pParse, "too many columns on %s", pTable->zName); + } + azModuleArg = sqlite3DbRealloc(db, pTable->azModuleArg, nBytes); + if( azModuleArg==0 ){ + sqlite3DbFree(db, zArg); + }else{ + int i = pTable->nModuleArg++; + azModuleArg[i] = zArg; + azModuleArg[i+1] = 0; + pTable->azModuleArg = azModuleArg; + } +} + +/* +** The parser calls this routine when it first sees a CREATE VIRTUAL TABLE +** statement. The module name has been parsed, but the optional list +** of parameters that follow the module name are still pending. +*/ +SQLITE_PRIVATE void sqlite3VtabBeginParse( + Parse *pParse, /* Parsing context */ + Token *pName1, /* Name of new table, or database name */ + Token *pName2, /* Name of new table or NULL */ + Token *pModuleName, /* Name of the module for the virtual table */ + int ifNotExists /* No error if the table already exists */ +){ + Table *pTable; /* The new virtual table */ + sqlite3 *db; /* Database connection */ + + sqlite3StartTable(pParse, pName1, pName2, 0, 0, 1, ifNotExists); + pTable = pParse->pNewTable; + if( pTable==0 ) return; + assert( 0==pTable->pIndex ); + + db = pParse->db; + + assert( pTable->nModuleArg==0 ); + addModuleArgument(pParse, pTable, sqlite3NameFromToken(db, pModuleName)); + addModuleArgument(pParse, pTable, 0); + addModuleArgument(pParse, pTable, sqlite3DbStrDup(db, pTable->zName)); + assert( (pParse->sNameToken.z==pName2->z && pName2->z!=0) + || (pParse->sNameToken.z==pName1->z && pName2->z==0) + ); + pParse->sNameToken.n = (int)( + &pModuleName->z[pModuleName->n] - pParse->sNameToken.z + ); + +#ifndef SQLITE_OMIT_AUTHORIZATION + /* Creating a virtual table invokes the authorization callback twice. + ** The first invocation, to obtain permission to INSERT a row into the + ** sqlite_master table, has already been made by sqlite3StartTable(). + ** The second call, to obtain permission to create the table, is made now. + */ + if( pTable->azModuleArg ){ + int iDb = sqlite3SchemaToIndex(db, pTable->pSchema); + assert( iDb>=0 ); /* The database the table is being created in */ + sqlite3AuthCheck(pParse, SQLITE_CREATE_VTABLE, pTable->zName, + pTable->azModuleArg[0], pParse->db->aDb[iDb].zDbSName); + } +#endif +} + +/* +** This routine takes the module argument that has been accumulating +** in pParse->zArg[] and appends it to the list of arguments on the +** virtual table currently under construction in pParse->pTable. +*/ +static void addArgumentToVtab(Parse *pParse){ + if( pParse->sArg.z && pParse->pNewTable ){ + const char *z = (const char*)pParse->sArg.z; + int n = pParse->sArg.n; + sqlite3 *db = pParse->db; + addModuleArgument(pParse, pParse->pNewTable, sqlite3DbStrNDup(db, z, n)); + } +} + +/* +** The parser calls this routine after the CREATE VIRTUAL TABLE statement +** has been completely parsed. +*/ +SQLITE_PRIVATE void sqlite3VtabFinishParse(Parse *pParse, Token *pEnd){ + Table *pTab = pParse->pNewTable; /* The table being constructed */ + sqlite3 *db = pParse->db; /* The database connection */ + + if( pTab==0 ) return; + addArgumentToVtab(pParse); + pParse->sArg.z = 0; + if( pTab->nModuleArg<1 ) return; + + /* If the CREATE VIRTUAL TABLE statement is being entered for the + ** first time (in other words if the virtual table is actually being + ** created now instead of just being read out of sqlite_master) then + ** do additional initialization work and store the statement text + ** in the sqlite_master table. + */ + if( !db->init.busy ){ + char *zStmt; + char *zWhere; + int iDb; + int iReg; + Vdbe *v; + + /* Compute the complete text of the CREATE VIRTUAL TABLE statement */ + if( pEnd ){ + pParse->sNameToken.n = (int)(pEnd->z - pParse->sNameToken.z) + pEnd->n; + } + zStmt = sqlite3MPrintf(db, "CREATE VIRTUAL TABLE %T", &pParse->sNameToken); + + /* A slot for the record has already been allocated in the + ** SQLITE_MASTER table. We just need to update that slot with all + ** the information we've collected. + ** + ** The VM register number pParse->regRowid holds the rowid of an + ** entry in the sqlite_master table tht was created for this vtab + ** by sqlite3StartTable(). + */ + iDb = sqlite3SchemaToIndex(db, pTab->pSchema); + sqlite3NestedParse(pParse, + "UPDATE %Q.%s " + "SET type='table', name=%Q, tbl_name=%Q, rootpage=0, sql=%Q " + "WHERE rowid=#%d", + db->aDb[iDb].zDbSName, MASTER_NAME, + pTab->zName, + pTab->zName, + zStmt, + pParse->regRowid + ); + sqlite3DbFree(db, zStmt); + v = sqlite3GetVdbe(pParse); + sqlite3ChangeCookie(pParse, iDb); + + sqlite3VdbeAddOp0(v, OP_Expire); + zWhere = sqlite3MPrintf(db, "name='%q' AND type='table'", pTab->zName); + sqlite3VdbeAddParseSchemaOp(v, iDb, zWhere); + + iReg = ++pParse->nMem; + sqlite3VdbeLoadString(v, iReg, pTab->zName); + sqlite3VdbeAddOp2(v, OP_VCreate, iDb, iReg); + } + + /* If we are rereading the sqlite_master table create the in-memory + ** record of the table. The xConnect() method is not called until + ** the first time the virtual table is used in an SQL statement. This + ** allows a schema that contains virtual tables to be loaded before + ** the required virtual table implementations are registered. */ + else { + Table *pOld; + Schema *pSchema = pTab->pSchema; + const char *zName = pTab->zName; + assert( sqlite3SchemaMutexHeld(db, 0, pSchema) ); + pOld = sqlite3HashInsert(&pSchema->tblHash, zName, pTab); + if( pOld ){ + sqlite3OomFault(db); + assert( pTab==pOld ); /* Malloc must have failed inside HashInsert() */ + return; + } + pParse->pNewTable = 0; + } +} + +/* +** The parser calls this routine when it sees the first token +** of an argument to the module name in a CREATE VIRTUAL TABLE statement. +*/ +SQLITE_PRIVATE void sqlite3VtabArgInit(Parse *pParse){ + addArgumentToVtab(pParse); + pParse->sArg.z = 0; + pParse->sArg.n = 0; +} + +/* +** The parser calls this routine for each token after the first token +** in an argument to the module name in a CREATE VIRTUAL TABLE statement. +*/ +SQLITE_PRIVATE void sqlite3VtabArgExtend(Parse *pParse, Token *p){ + Token *pArg = &pParse->sArg; + if( pArg->z==0 ){ + pArg->z = p->z; + pArg->n = p->n; + }else{ + assert(pArg->z <= p->z); + pArg->n = (int)(&p->z[p->n] - pArg->z); + } +} + +/* +** Invoke a virtual table constructor (either xCreate or xConnect). The +** pointer to the function to invoke is passed as the fourth parameter +** to this procedure. +*/ +static int vtabCallConstructor( + sqlite3 *db, + Table *pTab, + Module *pMod, + int (*xConstruct)(sqlite3*,void*,int,const char*const*,sqlite3_vtab**,char**), + char **pzErr +){ + VtabCtx sCtx; + VTable *pVTable; + int rc; + const char *const*azArg = (const char *const*)pTab->azModuleArg; + int nArg = pTab->nModuleArg; + char *zErr = 0; + char *zModuleName; + int iDb; + VtabCtx *pCtx; + + /* Check that the virtual-table is not already being initialized */ + for(pCtx=db->pVtabCtx; pCtx; pCtx=pCtx->pPrior){ + if( pCtx->pTab==pTab ){ + *pzErr = sqlite3MPrintf(db, + "vtable constructor called recursively: %s", pTab->zName + ); + return SQLITE_LOCKED; + } + } + + zModuleName = sqlite3DbStrDup(db, pTab->zName); + if( !zModuleName ){ + return SQLITE_NOMEM_BKPT; + } + + pVTable = sqlite3MallocZero(sizeof(VTable)); + if( !pVTable ){ + sqlite3OomFault(db); + sqlite3DbFree(db, zModuleName); + return SQLITE_NOMEM_BKPT; + } + pVTable->db = db; + pVTable->pMod = pMod; + + iDb = sqlite3SchemaToIndex(db, pTab->pSchema); + pTab->azModuleArg[1] = db->aDb[iDb].zDbSName; + + /* Invoke the virtual table constructor */ + assert( &db->pVtabCtx ); + assert( xConstruct ); + sCtx.pTab = pTab; + sCtx.pVTable = pVTable; + sCtx.pPrior = db->pVtabCtx; + sCtx.bDeclared = 0; + db->pVtabCtx = &sCtx; + rc = xConstruct(db, pMod->pAux, nArg, azArg, &pVTable->pVtab, &zErr); + db->pVtabCtx = sCtx.pPrior; + if( rc==SQLITE_NOMEM ) sqlite3OomFault(db); + assert( sCtx.pTab==pTab ); + + if( SQLITE_OK!=rc ){ + if( zErr==0 ){ + *pzErr = sqlite3MPrintf(db, "vtable constructor failed: %s", zModuleName); + }else { + *pzErr = sqlite3MPrintf(db, "%s", zErr); + sqlite3_free(zErr); + } + sqlite3DbFree(db, pVTable); + }else if( ALWAYS(pVTable->pVtab) ){ + /* Justification of ALWAYS(): A correct vtab constructor must allocate + ** the sqlite3_vtab object if successful. */ + memset(pVTable->pVtab, 0, sizeof(pVTable->pVtab[0])); + pVTable->pVtab->pModule = pMod->pModule; + pMod->nRefModule++; + pVTable->nRef = 1; + if( sCtx.bDeclared==0 ){ + const char *zFormat = "vtable constructor did not declare schema: %s"; + *pzErr = sqlite3MPrintf(db, zFormat, pTab->zName); + sqlite3VtabUnlock(pVTable); + rc = SQLITE_ERROR; + }else{ + int iCol; + u8 oooHidden = 0; + /* If everything went according to plan, link the new VTable structure + ** into the linked list headed by pTab->pVTable. Then loop through the + ** columns of the table to see if any of them contain the token "hidden". + ** If so, set the Column COLFLAG_HIDDEN flag and remove the token from + ** the type string. */ + pVTable->pNext = pTab->pVTable; + pTab->pVTable = pVTable; + + for(iCol=0; iColnCol; iCol++){ + char *zType = sqlite3ColumnType(&pTab->aCol[iCol], ""); + int nType; + int i = 0; + nType = sqlite3Strlen30(zType); + for(i=0; i0 ){ + assert(zType[i-1]==' '); + zType[i-1] = '\0'; + } + pTab->aCol[iCol].colFlags |= COLFLAG_HIDDEN; + oooHidden = TF_OOOHidden; + }else{ + pTab->tabFlags |= oooHidden; + } + } + } + } + + sqlite3DbFree(db, zModuleName); + return rc; +} + +/* +** This function is invoked by the parser to call the xConnect() method +** of the virtual table pTab. If an error occurs, an error code is returned +** and an error left in pParse. +** +** This call is a no-op if table pTab is not a virtual table. +*/ +SQLITE_PRIVATE int sqlite3VtabCallConnect(Parse *pParse, Table *pTab){ + sqlite3 *db = pParse->db; + const char *zMod; + Module *pMod; + int rc; + + assert( pTab ); + if( !IsVirtual(pTab) || sqlite3GetVTable(db, pTab) ){ + return SQLITE_OK; + } + + /* Locate the required virtual table module */ + zMod = pTab->azModuleArg[0]; + pMod = (Module*)sqlite3HashFind(&db->aModule, zMod); + + if( !pMod ){ + const char *zModule = pTab->azModuleArg[0]; + sqlite3ErrorMsg(pParse, "no such module: %s", zModule); + rc = SQLITE_ERROR; + }else{ + char *zErr = 0; + rc = vtabCallConstructor(db, pTab, pMod, pMod->pModule->xConnect, &zErr); + if( rc!=SQLITE_OK ){ + sqlite3ErrorMsg(pParse, "%s", zErr); + pParse->rc = rc; + } + sqlite3DbFree(db, zErr); + } + + return rc; +} +/* +** Grow the db->aVTrans[] array so that there is room for at least one +** more v-table. Return SQLITE_NOMEM if a malloc fails, or SQLITE_OK otherwise. +*/ +static int growVTrans(sqlite3 *db){ + const int ARRAY_INCR = 5; + + /* Grow the sqlite3.aVTrans array if required */ + if( (db->nVTrans%ARRAY_INCR)==0 ){ + VTable **aVTrans; + sqlite3_int64 nBytes = sizeof(sqlite3_vtab*)* + ((sqlite3_int64)db->nVTrans + ARRAY_INCR); + aVTrans = sqlite3DbRealloc(db, (void *)db->aVTrans, nBytes); + if( !aVTrans ){ + return SQLITE_NOMEM_BKPT; + } + memset(&aVTrans[db->nVTrans], 0, sizeof(sqlite3_vtab *)*ARRAY_INCR); + db->aVTrans = aVTrans; + } + + return SQLITE_OK; +} + +/* +** Add the virtual table pVTab to the array sqlite3.aVTrans[]. Space should +** have already been reserved using growVTrans(). +*/ +static void addToVTrans(sqlite3 *db, VTable *pVTab){ + /* Add pVtab to the end of sqlite3.aVTrans */ + db->aVTrans[db->nVTrans++] = pVTab; + sqlite3VtabLock(pVTab); +} + +/* +** This function is invoked by the vdbe to call the xCreate method +** of the virtual table named zTab in database iDb. +** +** If an error occurs, *pzErr is set to point to an English language +** description of the error and an SQLITE_XXX error code is returned. +** In this case the caller must call sqlite3DbFree(db, ) on *pzErr. +*/ +SQLITE_PRIVATE int sqlite3VtabCallCreate(sqlite3 *db, int iDb, const char *zTab, char **pzErr){ + int rc = SQLITE_OK; + Table *pTab; + Module *pMod; + const char *zMod; + + pTab = sqlite3FindTable(db, zTab, db->aDb[iDb].zDbSName); + assert( pTab && IsVirtual(pTab) && !pTab->pVTable ); + + /* Locate the required virtual table module */ + zMod = pTab->azModuleArg[0]; + pMod = (Module*)sqlite3HashFind(&db->aModule, zMod); + + /* If the module has been registered and includes a Create method, + ** invoke it now. If the module has not been registered, return an + ** error. Otherwise, do nothing. + */ + if( pMod==0 || pMod->pModule->xCreate==0 || pMod->pModule->xDestroy==0 ){ + *pzErr = sqlite3MPrintf(db, "no such module: %s", zMod); + rc = SQLITE_ERROR; + }else{ + rc = vtabCallConstructor(db, pTab, pMod, pMod->pModule->xCreate, pzErr); + } + + /* Justification of ALWAYS(): The xConstructor method is required to + ** create a valid sqlite3_vtab if it returns SQLITE_OK. */ + if( rc==SQLITE_OK && ALWAYS(sqlite3GetVTable(db, pTab)) ){ + rc = growVTrans(db); + if( rc==SQLITE_OK ){ + addToVTrans(db, sqlite3GetVTable(db, pTab)); + } + } + + return rc; +} + +/* +** This function is used to set the schema of a virtual table. It is only +** valid to call this function from within the xCreate() or xConnect() of a +** virtual table module. +*/ +SQLITE_API int sqlite3_declare_vtab(sqlite3 *db, const char *zCreateTable){ + VtabCtx *pCtx; + int rc = SQLITE_OK; + Table *pTab; + char *zErr = 0; + Parse sParse; + +#ifdef SQLITE_ENABLE_API_ARMOR + if( !sqlite3SafetyCheckOk(db) || zCreateTable==0 ){ + return SQLITE_MISUSE_BKPT; + } +#endif + sqlite3_mutex_enter(db->mutex); + pCtx = db->pVtabCtx; + if( !pCtx || pCtx->bDeclared ){ + sqlite3Error(db, SQLITE_MISUSE); + sqlite3_mutex_leave(db->mutex); + return SQLITE_MISUSE_BKPT; + } + pTab = pCtx->pTab; + assert( IsVirtual(pTab) ); + + memset(&sParse, 0, sizeof(sParse)); + sParse.eParseMode = PARSE_MODE_DECLARE_VTAB; + sParse.db = db; + sParse.nQueryLoop = 1; + if( SQLITE_OK==sqlite3RunParser(&sParse, zCreateTable, &zErr) + && sParse.pNewTable + && !db->mallocFailed + && !sParse.pNewTable->pSelect + && !IsVirtual(sParse.pNewTable) + ){ + if( !pTab->aCol ){ + Table *pNew = sParse.pNewTable; + Index *pIdx; + pTab->aCol = pNew->aCol; + pTab->nCol = pNew->nCol; + pTab->tabFlags |= pNew->tabFlags & (TF_WithoutRowid|TF_NoVisibleRowid); + pNew->nCol = 0; + pNew->aCol = 0; + assert( pTab->pIndex==0 ); + assert( HasRowid(pNew) || sqlite3PrimaryKeyIndex(pNew)!=0 ); + if( !HasRowid(pNew) + && pCtx->pVTable->pMod->pModule->xUpdate!=0 + && sqlite3PrimaryKeyIndex(pNew)->nKeyCol!=1 + ){ + /* WITHOUT ROWID virtual tables must either be read-only (xUpdate==0) + ** or else must have a single-column PRIMARY KEY */ + rc = SQLITE_ERROR; + } + pIdx = pNew->pIndex; + if( pIdx ){ + assert( pIdx->pNext==0 ); + pTab->pIndex = pIdx; + pNew->pIndex = 0; + pIdx->pTable = pTab; + } + } + pCtx->bDeclared = 1; + }else{ + sqlite3ErrorWithMsg(db, SQLITE_ERROR, (zErr ? "%s" : 0), zErr); + sqlite3DbFree(db, zErr); + rc = SQLITE_ERROR; + } + sParse.eParseMode = PARSE_MODE_NORMAL; + + if( sParse.pVdbe ){ + sqlite3VdbeFinalize(sParse.pVdbe); + } + sqlite3DeleteTable(db, sParse.pNewTable); + sqlite3ParserReset(&sParse); + + assert( (rc&0xff)==rc ); + rc = sqlite3ApiExit(db, rc); + sqlite3_mutex_leave(db->mutex); + return rc; +} + +/* +** This function is invoked by the vdbe to call the xDestroy method +** of the virtual table named zTab in database iDb. This occurs +** when a DROP TABLE is mentioned. +** +** This call is a no-op if zTab is not a virtual table. +*/ +SQLITE_PRIVATE int sqlite3VtabCallDestroy(sqlite3 *db, int iDb, const char *zTab){ + int rc = SQLITE_OK; + Table *pTab; + + pTab = sqlite3FindTable(db, zTab, db->aDb[iDb].zDbSName); + if( pTab!=0 && ALWAYS(pTab->pVTable!=0) ){ + VTable *p; + int (*xDestroy)(sqlite3_vtab *); + for(p=pTab->pVTable; p; p=p->pNext){ + assert( p->pVtab ); + if( p->pVtab->nRef>0 ){ + return SQLITE_LOCKED; + } + } + p = vtabDisconnectAll(db, pTab); + xDestroy = p->pMod->pModule->xDestroy; + assert( xDestroy!=0 ); /* Checked before the virtual table is created */ + pTab->nTabRef++; + rc = xDestroy(p->pVtab); + /* Remove the sqlite3_vtab* from the aVTrans[] array, if applicable */ + if( rc==SQLITE_OK ){ + assert( pTab->pVTable==p && p->pNext==0 ); + p->pVtab = 0; + pTab->pVTable = 0; + sqlite3VtabUnlock(p); + } + sqlite3DeleteTable(db, pTab); + } + + return rc; +} + +/* +** This function invokes either the xRollback or xCommit method +** of each of the virtual tables in the sqlite3.aVTrans array. The method +** called is identified by the second argument, "offset", which is +** the offset of the method to call in the sqlite3_module structure. +** +** The array is cleared after invoking the callbacks. +*/ +static void callFinaliser(sqlite3 *db, int offset){ + int i; + if( db->aVTrans ){ + VTable **aVTrans = db->aVTrans; + db->aVTrans = 0; + for(i=0; inVTrans; i++){ + VTable *pVTab = aVTrans[i]; + sqlite3_vtab *p = pVTab->pVtab; + if( p ){ + int (*x)(sqlite3_vtab *); + x = *(int (**)(sqlite3_vtab *))((char *)p->pModule + offset); + if( x ) x(p); + } + pVTab->iSavepoint = 0; + sqlite3VtabUnlock(pVTab); + } + sqlite3DbFree(db, aVTrans); + db->nVTrans = 0; + } +} + +/* +** Invoke the xSync method of all virtual tables in the sqlite3.aVTrans +** array. Return the error code for the first error that occurs, or +** SQLITE_OK if all xSync operations are successful. +** +** If an error message is available, leave it in p->zErrMsg. +*/ +SQLITE_PRIVATE int sqlite3VtabSync(sqlite3 *db, Vdbe *p){ + int i; + int rc = SQLITE_OK; + VTable **aVTrans = db->aVTrans; + + db->aVTrans = 0; + for(i=0; rc==SQLITE_OK && inVTrans; i++){ + int (*x)(sqlite3_vtab *); + sqlite3_vtab *pVtab = aVTrans[i]->pVtab; + if( pVtab && (x = pVtab->pModule->xSync)!=0 ){ + rc = x(pVtab); + sqlite3VtabImportErrmsg(p, pVtab); + } + } + db->aVTrans = aVTrans; + return rc; +} + +/* +** Invoke the xRollback method of all virtual tables in the +** sqlite3.aVTrans array. Then clear the array itself. +*/ +SQLITE_PRIVATE int sqlite3VtabRollback(sqlite3 *db){ + callFinaliser(db, offsetof(sqlite3_module,xRollback)); + return SQLITE_OK; +} + +/* +** Invoke the xCommit method of all virtual tables in the +** sqlite3.aVTrans array. Then clear the array itself. +*/ +SQLITE_PRIVATE int sqlite3VtabCommit(sqlite3 *db){ + callFinaliser(db, offsetof(sqlite3_module,xCommit)); + return SQLITE_OK; +} + +/* +** If the virtual table pVtab supports the transaction interface +** (xBegin/xRollback/xCommit and optionally xSync) and a transaction is +** not currently open, invoke the xBegin method now. +** +** If the xBegin call is successful, place the sqlite3_vtab pointer +** in the sqlite3.aVTrans array. +*/ +SQLITE_PRIVATE int sqlite3VtabBegin(sqlite3 *db, VTable *pVTab){ + int rc = SQLITE_OK; + const sqlite3_module *pModule; + + /* Special case: If db->aVTrans is NULL and db->nVTrans is greater + ** than zero, then this function is being called from within a + ** virtual module xSync() callback. It is illegal to write to + ** virtual module tables in this case, so return SQLITE_LOCKED. + */ + if( sqlite3VtabInSync(db) ){ + return SQLITE_LOCKED; + } + if( !pVTab ){ + return SQLITE_OK; + } + pModule = pVTab->pVtab->pModule; + + if( pModule->xBegin ){ + int i; + + /* If pVtab is already in the aVTrans array, return early */ + for(i=0; inVTrans; i++){ + if( db->aVTrans[i]==pVTab ){ + return SQLITE_OK; + } + } + + /* Invoke the xBegin method. If successful, add the vtab to the + ** sqlite3.aVTrans[] array. */ + rc = growVTrans(db); + if( rc==SQLITE_OK ){ + rc = pModule->xBegin(pVTab->pVtab); + if( rc==SQLITE_OK ){ + int iSvpt = db->nStatement + db->nSavepoint; + addToVTrans(db, pVTab); + if( iSvpt && pModule->xSavepoint ){ + pVTab->iSavepoint = iSvpt; + rc = pModule->xSavepoint(pVTab->pVtab, iSvpt-1); + } + } + } + } + return rc; +} + +/* +** Invoke either the xSavepoint, xRollbackTo or xRelease method of all +** virtual tables that currently have an open transaction. Pass iSavepoint +** as the second argument to the virtual table method invoked. +** +** If op is SAVEPOINT_BEGIN, the xSavepoint method is invoked. If it is +** SAVEPOINT_ROLLBACK, the xRollbackTo method. Otherwise, if op is +** SAVEPOINT_RELEASE, then the xRelease method of each virtual table with +** an open transaction is invoked. +** +** If any virtual table method returns an error code other than SQLITE_OK, +** processing is abandoned and the error returned to the caller of this +** function immediately. If all calls to virtual table methods are successful, +** SQLITE_OK is returned. +*/ +SQLITE_PRIVATE int sqlite3VtabSavepoint(sqlite3 *db, int op, int iSavepoint){ + int rc = SQLITE_OK; + + assert( op==SAVEPOINT_RELEASE||op==SAVEPOINT_ROLLBACK||op==SAVEPOINT_BEGIN ); + assert( iSavepoint>=-1 ); + if( db->aVTrans ){ + int i; + for(i=0; rc==SQLITE_OK && inVTrans; i++){ + VTable *pVTab = db->aVTrans[i]; + const sqlite3_module *pMod = pVTab->pMod->pModule; + if( pVTab->pVtab && pMod->iVersion>=2 ){ + int (*xMethod)(sqlite3_vtab *, int); + sqlite3VtabLock(pVTab); + switch( op ){ + case SAVEPOINT_BEGIN: + xMethod = pMod->xSavepoint; + pVTab->iSavepoint = iSavepoint+1; + break; + case SAVEPOINT_ROLLBACK: + xMethod = pMod->xRollbackTo; + break; + default: + xMethod = pMod->xRelease; + break; + } + if( xMethod && pVTab->iSavepoint>iSavepoint ){ + rc = xMethod(pVTab->pVtab, iSavepoint); + } + sqlite3VtabUnlock(pVTab); + } + } + } + return rc; +} + +/* +** The first parameter (pDef) is a function implementation. The +** second parameter (pExpr) is the first argument to this function. +** If pExpr is a column in a virtual table, then let the virtual +** table implementation have an opportunity to overload the function. +** +** This routine is used to allow virtual table implementations to +** overload MATCH, LIKE, GLOB, and REGEXP operators. +** +** Return either the pDef argument (indicating no change) or a +** new FuncDef structure that is marked as ephemeral using the +** SQLITE_FUNC_EPHEM flag. +*/ +SQLITE_PRIVATE FuncDef *sqlite3VtabOverloadFunction( + sqlite3 *db, /* Database connection for reporting malloc problems */ + FuncDef *pDef, /* Function to possibly overload */ + int nArg, /* Number of arguments to the function */ + Expr *pExpr /* First argument to the function */ +){ + Table *pTab; + sqlite3_vtab *pVtab; + sqlite3_module *pMod; + void (*xSFunc)(sqlite3_context*,int,sqlite3_value**) = 0; + void *pArg = 0; + FuncDef *pNew; + int rc = 0; + + /* Check to see the left operand is a column in a virtual table */ + if( NEVER(pExpr==0) ) return pDef; + if( pExpr->op!=TK_COLUMN ) return pDef; + pTab = pExpr->y.pTab; + if( pTab==0 ) return pDef; + if( !IsVirtual(pTab) ) return pDef; + pVtab = sqlite3GetVTable(db, pTab)->pVtab; + assert( pVtab!=0 ); + assert( pVtab->pModule!=0 ); + pMod = (sqlite3_module *)pVtab->pModule; + if( pMod->xFindFunction==0 ) return pDef; + + /* Call the xFindFunction method on the virtual table implementation + ** to see if the implementation wants to overload this function. + ** + ** Though undocumented, we have historically always invoked xFindFunction + ** with an all lower-case function name. Continue in this tradition to + ** avoid any chance of an incompatibility. + */ +#ifdef SQLITE_DEBUG + { + int i; + for(i=0; pDef->zName[i]; i++){ + unsigned char x = (unsigned char)pDef->zName[i]; + assert( x==sqlite3UpperToLower[x] ); + } + } +#endif + rc = pMod->xFindFunction(pVtab, nArg, pDef->zName, &xSFunc, &pArg); + if( rc==0 ){ + return pDef; + } + + /* Create a new ephemeral function definition for the overloaded + ** function */ + pNew = sqlite3DbMallocZero(db, sizeof(*pNew) + + sqlite3Strlen30(pDef->zName) + 1); + if( pNew==0 ){ + return pDef; + } + *pNew = *pDef; + pNew->zName = (const char*)&pNew[1]; + memcpy((char*)&pNew[1], pDef->zName, sqlite3Strlen30(pDef->zName)+1); + pNew->xSFunc = xSFunc; + pNew->pUserData = pArg; + pNew->funcFlags |= SQLITE_FUNC_EPHEM; + return pNew; +} + +/* +** Make sure virtual table pTab is contained in the pParse->apVirtualLock[] +** array so that an OP_VBegin will get generated for it. Add pTab to the +** array if it is missing. If pTab is already in the array, this routine +** is a no-op. +*/ +SQLITE_PRIVATE void sqlite3VtabMakeWritable(Parse *pParse, Table *pTab){ + Parse *pToplevel = sqlite3ParseToplevel(pParse); + int i, n; + Table **apVtabLock; + + assert( IsVirtual(pTab) ); + for(i=0; inVtabLock; i++){ + if( pTab==pToplevel->apVtabLock[i] ) return; + } + n = (pToplevel->nVtabLock+1)*sizeof(pToplevel->apVtabLock[0]); + apVtabLock = sqlite3_realloc64(pToplevel->apVtabLock, n); + if( apVtabLock ){ + pToplevel->apVtabLock = apVtabLock; + pToplevel->apVtabLock[pToplevel->nVtabLock++] = pTab; + }else{ + sqlite3OomFault(pToplevel->db); + } +} + +/* +** Check to see if virtual table module pMod can be have an eponymous +** virtual table instance. If it can, create one if one does not already +** exist. Return non-zero if the eponymous virtual table instance exists +** when this routine returns, and return zero if it does not exist. +** +** An eponymous virtual table instance is one that is named after its +** module, and more importantly, does not require a CREATE VIRTUAL TABLE +** statement in order to come into existance. Eponymous virtual table +** instances always exist. They cannot be DROP-ed. +** +** Any virtual table module for which xConnect and xCreate are the same +** method can have an eponymous virtual table instance. +*/ +SQLITE_PRIVATE int sqlite3VtabEponymousTableInit(Parse *pParse, Module *pMod){ + const sqlite3_module *pModule = pMod->pModule; + Table *pTab; + char *zErr = 0; + int rc; + sqlite3 *db = pParse->db; + if( pMod->pEpoTab ) return 1; + if( pModule->xCreate!=0 && pModule->xCreate!=pModule->xConnect ) return 0; + pTab = sqlite3DbMallocZero(db, sizeof(Table)); + if( pTab==0 ) return 0; + pTab->zName = sqlite3DbStrDup(db, pMod->zName); + if( pTab->zName==0 ){ + sqlite3DbFree(db, pTab); + return 0; + } + pMod->pEpoTab = pTab; + pTab->nTabRef = 1; + pTab->pSchema = db->aDb[0].pSchema; + assert( pTab->nModuleArg==0 ); + pTab->iPKey = -1; + addModuleArgument(pParse, pTab, sqlite3DbStrDup(db, pTab->zName)); + addModuleArgument(pParse, pTab, 0); + addModuleArgument(pParse, pTab, sqlite3DbStrDup(db, pTab->zName)); + rc = vtabCallConstructor(db, pTab, pMod, pModule->xConnect, &zErr); + if( rc ){ + sqlite3ErrorMsg(pParse, "%s", zErr); + sqlite3DbFree(db, zErr); + sqlite3VtabEponymousTableClear(db, pMod); + return 0; + } + return 1; +} + +/* +** Erase the eponymous virtual table instance associated with +** virtual table module pMod, if it exists. +*/ +SQLITE_PRIVATE void sqlite3VtabEponymousTableClear(sqlite3 *db, Module *pMod){ + Table *pTab = pMod->pEpoTab; + if( pTab!=0 ){ + /* Mark the table as Ephemeral prior to deleting it, so that the + ** sqlite3DeleteTable() routine will know that it is not stored in + ** the schema. */ + pTab->tabFlags |= TF_Ephemeral; + sqlite3DeleteTable(db, pTab); + pMod->pEpoTab = 0; + } +} + +/* +** Return the ON CONFLICT resolution mode in effect for the virtual +** table update operation currently in progress. +** +** The results of this routine are undefined unless it is called from +** within an xUpdate method. +*/ +SQLITE_API int sqlite3_vtab_on_conflict(sqlite3 *db){ + static const unsigned char aMap[] = { + SQLITE_ROLLBACK, SQLITE_ABORT, SQLITE_FAIL, SQLITE_IGNORE, SQLITE_REPLACE + }; +#ifdef SQLITE_ENABLE_API_ARMOR + if( !sqlite3SafetyCheckOk(db) ) return SQLITE_MISUSE_BKPT; +#endif + assert( OE_Rollback==1 && OE_Abort==2 && OE_Fail==3 ); + assert( OE_Ignore==4 && OE_Replace==5 ); + assert( db->vtabOnConflict>=1 && db->vtabOnConflict<=5 ); + return (int)aMap[db->vtabOnConflict-1]; +} + +/* +** Call from within the xCreate() or xConnect() methods to provide +** the SQLite core with additional information about the behavior +** of the virtual table being implemented. +*/ +SQLITE_API int sqlite3_vtab_config(sqlite3 *db, int op, ...){ + va_list ap; + int rc = SQLITE_OK; + +#ifdef SQLITE_ENABLE_API_ARMOR + if( !sqlite3SafetyCheckOk(db) ) return SQLITE_MISUSE_BKPT; +#endif + sqlite3_mutex_enter(db->mutex); + va_start(ap, op); + switch( op ){ + case SQLITE_VTAB_CONSTRAINT_SUPPORT: { + VtabCtx *p = db->pVtabCtx; + if( !p ){ + rc = SQLITE_MISUSE_BKPT; + }else{ + assert( p->pTab==0 || IsVirtual(p->pTab) ); + p->pVTable->bConstraint = (u8)va_arg(ap, int); + } + break; + } + default: + rc = SQLITE_MISUSE_BKPT; + break; + } + va_end(ap); + + if( rc!=SQLITE_OK ) sqlite3Error(db, rc); + sqlite3_mutex_leave(db->mutex); + return rc; +} + +#endif /* SQLITE_OMIT_VIRTUALTABLE */ + +/************** End of vtab.c ************************************************/ +/************** Begin file wherecode.c ***************************************/ +/* +** 2015-06-06 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** This module contains C code that generates VDBE code used to process +** the WHERE clause of SQL statements. +** +** This file was split off from where.c on 2015-06-06 in order to reduce the +** size of where.c and make it easier to edit. This file contains the routines +** that actually generate the bulk of the WHERE loop code. The original where.c +** file retains the code that does query planning and analysis. +*/ +/* #include "sqliteInt.h" */ +/************** Include whereInt.h in the middle of wherecode.c **************/ +/************** Begin file whereInt.h ****************************************/ +/* +** 2013-11-12 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** +** This file contains structure and macro definitions for the query +** planner logic in "where.c". These definitions are broken out into +** a separate source file for easier editing. +*/ +#ifndef SQLITE_WHEREINT_H +#define SQLITE_WHEREINT_H + +/* +** Trace output macros +*/ +#if defined(SQLITE_TEST) || defined(SQLITE_DEBUG) +/***/ extern int sqlite3WhereTrace; +#endif +#if defined(SQLITE_DEBUG) \ + && (defined(SQLITE_TEST) || defined(SQLITE_ENABLE_WHERETRACE)) +# define WHERETRACE(K,X) if(sqlite3WhereTrace&(K)) sqlite3DebugPrintf X +# define WHERETRACE_ENABLED 1 +#else +# define WHERETRACE(K,X) +#endif + +/* Forward references +*/ +typedef struct WhereClause WhereClause; +typedef struct WhereMaskSet WhereMaskSet; +typedef struct WhereOrInfo WhereOrInfo; +typedef struct WhereAndInfo WhereAndInfo; +typedef struct WhereLevel WhereLevel; +typedef struct WhereLoop WhereLoop; +typedef struct WherePath WherePath; +typedef struct WhereTerm WhereTerm; +typedef struct WhereLoopBuilder WhereLoopBuilder; +typedef struct WhereScan WhereScan; +typedef struct WhereOrCost WhereOrCost; +typedef struct WhereOrSet WhereOrSet; + +/* +** This object contains information needed to implement a single nested +** loop in WHERE clause. +** +** Contrast this object with WhereLoop. This object describes the +** implementation of the loop. WhereLoop describes the algorithm. +** This object contains a pointer to the WhereLoop algorithm as one of +** its elements. +** +** The WhereInfo object contains a single instance of this object for +** each term in the FROM clause (which is to say, for each of the +** nested loops as implemented). The order of WhereLevel objects determines +** the loop nested order, with WhereInfo.a[0] being the outer loop and +** WhereInfo.a[WhereInfo.nLevel-1] being the inner loop. +*/ +struct WhereLevel { + int iLeftJoin; /* Memory cell used to implement LEFT OUTER JOIN */ + int iTabCur; /* The VDBE cursor used to access the table */ + int iIdxCur; /* The VDBE cursor used to access pIdx */ + int addrBrk; /* Jump here to break out of the loop */ + int addrNxt; /* Jump here to start the next IN combination */ + int addrSkip; /* Jump here for next iteration of skip-scan */ + int addrCont; /* Jump here to continue with the next loop cycle */ + int addrFirst; /* First instruction of interior of the loop */ + int addrBody; /* Beginning of the body of this loop */ + int regBignull; /* big-null flag reg. True if a NULL-scan is needed */ + int addrBignull; /* Jump here for next part of big-null scan */ +#ifndef SQLITE_LIKE_DOESNT_MATCH_BLOBS + u32 iLikeRepCntr; /* LIKE range processing counter register (times 2) */ + int addrLikeRep; /* LIKE range processing address */ +#endif + u8 iFrom; /* Which entry in the FROM clause */ + u8 op, p3, p5; /* Opcode, P3 & P5 of the opcode that ends the loop */ + int p1, p2; /* Operands of the opcode used to end the loop */ + union { /* Information that depends on pWLoop->wsFlags */ + struct { + int nIn; /* Number of entries in aInLoop[] */ + struct InLoop { + int iCur; /* The VDBE cursor used by this IN operator */ + int addrInTop; /* Top of the IN loop */ + int iBase; /* Base register of multi-key index record */ + int nPrefix; /* Number of prior entires in the key */ + u8 eEndLoopOp; /* IN Loop terminator. OP_Next or OP_Prev */ + } *aInLoop; /* Information about each nested IN operator */ + } in; /* Used when pWLoop->wsFlags&WHERE_IN_ABLE */ + Index *pCovidx; /* Possible covering index for WHERE_MULTI_OR */ + } u; + struct WhereLoop *pWLoop; /* The selected WhereLoop object */ + Bitmask notReady; /* FROM entries not usable at this level */ +#ifdef SQLITE_ENABLE_STMT_SCANSTATUS + int addrVisit; /* Address at which row is visited */ +#endif +}; + +/* +** Each instance of this object represents an algorithm for evaluating one +** term of a join. Every term of the FROM clause will have at least +** one corresponding WhereLoop object (unless INDEXED BY constraints +** prevent a query solution - which is an error) and many terms of the +** FROM clause will have multiple WhereLoop objects, each describing a +** potential way of implementing that FROM-clause term, together with +** dependencies and cost estimates for using the chosen algorithm. +** +** Query planning consists of building up a collection of these WhereLoop +** objects, then computing a particular sequence of WhereLoop objects, with +** one WhereLoop object per FROM clause term, that satisfy all dependencies +** and that minimize the overall cost. +*/ +struct WhereLoop { + Bitmask prereq; /* Bitmask of other loops that must run first */ + Bitmask maskSelf; /* Bitmask identifying table iTab */ +#ifdef SQLITE_DEBUG + char cId; /* Symbolic ID of this loop for debugging use */ +#endif + u8 iTab; /* Position in FROM clause of table for this loop */ + u8 iSortIdx; /* Sorting index number. 0==None */ + LogEst rSetup; /* One-time setup cost (ex: create transient index) */ + LogEst rRun; /* Cost of running each loop */ + LogEst nOut; /* Estimated number of output rows */ + union { + struct { /* Information for internal btree tables */ + u16 nEq; /* Number of equality constraints */ + u16 nBtm; /* Size of BTM vector */ + u16 nTop; /* Size of TOP vector */ + u16 nDistinctCol; /* Index columns used to sort for DISTINCT */ + Index *pIndex; /* Index used, or NULL */ + } btree; + struct { /* Information for virtual tables */ + int idxNum; /* Index number */ + u8 needFree; /* True if sqlite3_free(idxStr) is needed */ + i8 isOrdered; /* True if satisfies ORDER BY */ + u16 omitMask; /* Terms that may be omitted */ + char *idxStr; /* Index identifier string */ + } vtab; + } u; + u32 wsFlags; /* WHERE_* flags describing the plan */ + u16 nLTerm; /* Number of entries in aLTerm[] */ + u16 nSkip; /* Number of NULL aLTerm[] entries */ + /**** whereLoopXfer() copies fields above ***********************/ +# define WHERE_LOOP_XFER_SZ offsetof(WhereLoop,nLSlot) + u16 nLSlot; /* Number of slots allocated for aLTerm[] */ + WhereTerm **aLTerm; /* WhereTerms used */ + WhereLoop *pNextLoop; /* Next WhereLoop object in the WhereClause */ + WhereTerm *aLTermSpace[3]; /* Initial aLTerm[] space */ +}; + +/* This object holds the prerequisites and the cost of running a +** subquery on one operand of an OR operator in the WHERE clause. +** See WhereOrSet for additional information +*/ +struct WhereOrCost { + Bitmask prereq; /* Prerequisites */ + LogEst rRun; /* Cost of running this subquery */ + LogEst nOut; /* Number of outputs for this subquery */ +}; + +/* The WhereOrSet object holds a set of possible WhereOrCosts that +** correspond to the subquery(s) of OR-clause processing. Only the +** best N_OR_COST elements are retained. +*/ +#define N_OR_COST 3 +struct WhereOrSet { + u16 n; /* Number of valid a[] entries */ + WhereOrCost a[N_OR_COST]; /* Set of best costs */ +}; + +/* +** Each instance of this object holds a sequence of WhereLoop objects +** that implement some or all of a query plan. +** +** Think of each WhereLoop object as a node in a graph with arcs +** showing dependencies and costs for travelling between nodes. (That is +** not a completely accurate description because WhereLoop costs are a +** vector, not a scalar, and because dependencies are many-to-one, not +** one-to-one as are graph nodes. But it is a useful visualization aid.) +** Then a WherePath object is a path through the graph that visits some +** or all of the WhereLoop objects once. +** +** The "solver" works by creating the N best WherePath objects of length +** 1. Then using those as a basis to compute the N best WherePath objects +** of length 2. And so forth until the length of WherePaths equals the +** number of nodes in the FROM clause. The best (lowest cost) WherePath +** at the end is the chosen query plan. +*/ +struct WherePath { + Bitmask maskLoop; /* Bitmask of all WhereLoop objects in this path */ + Bitmask revLoop; /* aLoop[]s that should be reversed for ORDER BY */ + LogEst nRow; /* Estimated number of rows generated by this path */ + LogEst rCost; /* Total cost of this path */ + LogEst rUnsorted; /* Total cost of this path ignoring sorting costs */ + i8 isOrdered; /* No. of ORDER BY terms satisfied. -1 for unknown */ + WhereLoop **aLoop; /* Array of WhereLoop objects implementing this path */ +}; + +/* +** The query generator uses an array of instances of this structure to +** help it analyze the subexpressions of the WHERE clause. Each WHERE +** clause subexpression is separated from the others by AND operators, +** usually, or sometimes subexpressions separated by OR. +** +** All WhereTerms are collected into a single WhereClause structure. +** The following identity holds: +** +** WhereTerm.pWC->a[WhereTerm.idx] == WhereTerm +** +** When a term is of the form: +** +** X +** +** where X is a column name and is one of certain operators, +** then WhereTerm.leftCursor and WhereTerm.u.leftColumn record the +** cursor number and column number for X. WhereTerm.eOperator records +** the using a bitmask encoding defined by WO_xxx below. The +** use of a bitmask encoding for the operator allows us to search +** quickly for terms that match any of several different operators. +** +** A WhereTerm might also be two or more subterms connected by OR: +** +** (t1.X ) OR (t1.Y ) OR .... +** +** In this second case, wtFlag has the TERM_ORINFO bit set and eOperator==WO_OR +** and the WhereTerm.u.pOrInfo field points to auxiliary information that +** is collected about the OR clause. +** +** If a term in the WHERE clause does not match either of the two previous +** categories, then eOperator==0. The WhereTerm.pExpr field is still set +** to the original subexpression content and wtFlags is set up appropriately +** but no other fields in the WhereTerm object are meaningful. +** +** When eOperator!=0, prereqRight and prereqAll record sets of cursor numbers, +** but they do so indirectly. A single WhereMaskSet structure translates +** cursor number into bits and the translated bit is stored in the prereq +** fields. The translation is used in order to maximize the number of +** bits that will fit in a Bitmask. The VDBE cursor numbers might be +** spread out over the non-negative integers. For example, the cursor +** numbers might be 3, 8, 9, 10, 20, 23, 41, and 45. The WhereMaskSet +** translates these sparse cursor numbers into consecutive integers +** beginning with 0 in order to make the best possible use of the available +** bits in the Bitmask. So, in the example above, the cursor numbers +** would be mapped into integers 0 through 7. +** +** The number of terms in a join is limited by the number of bits +** in prereqRight and prereqAll. The default is 64 bits, hence SQLite +** is only able to process joins with 64 or fewer tables. +*/ +struct WhereTerm { + Expr *pExpr; /* Pointer to the subexpression that is this term */ + WhereClause *pWC; /* The clause this term is part of */ + LogEst truthProb; /* Probability of truth for this expression */ + u16 wtFlags; /* TERM_xxx bit flags. See below */ + u16 eOperator; /* A WO_xx value describing */ + u8 nChild; /* Number of children that must disable us */ + u8 eMatchOp; /* Op for vtab MATCH/LIKE/GLOB/REGEXP terms */ + int iParent; /* Disable pWC->a[iParent] when this term disabled */ + int leftCursor; /* Cursor number of X in "X " */ + int iField; /* Field in (?,?,?) IN (SELECT...) vector */ + union { + int leftColumn; /* Column number of X in "X " */ + WhereOrInfo *pOrInfo; /* Extra information if (eOperator & WO_OR)!=0 */ + WhereAndInfo *pAndInfo; /* Extra information if (eOperator& WO_AND)!=0 */ + } u; + Bitmask prereqRight; /* Bitmask of tables used by pExpr->pRight */ + Bitmask prereqAll; /* Bitmask of tables referenced by pExpr */ +}; + +/* +** Allowed values of WhereTerm.wtFlags +*/ +#define TERM_DYNAMIC 0x01 /* Need to call sqlite3ExprDelete(db, pExpr) */ +#define TERM_VIRTUAL 0x02 /* Added by the optimizer. Do not code */ +#define TERM_CODED 0x04 /* This term is already coded */ +#define TERM_COPIED 0x08 /* Has a child */ +#define TERM_ORINFO 0x10 /* Need to free the WhereTerm.u.pOrInfo object */ +#define TERM_ANDINFO 0x20 /* Need to free the WhereTerm.u.pAndInfo obj */ +#define TERM_OR_OK 0x40 /* Used during OR-clause processing */ +#ifdef SQLITE_ENABLE_STAT4 +# define TERM_VNULL 0x80 /* Manufactured x>NULL or x<=NULL term */ +#else +# define TERM_VNULL 0x00 /* Disabled if not using stat4 */ +#endif +#define TERM_LIKEOPT 0x100 /* Virtual terms from the LIKE optimization */ +#define TERM_LIKECOND 0x200 /* Conditionally this LIKE operator term */ +#define TERM_LIKE 0x400 /* The original LIKE operator */ +#define TERM_IS 0x800 /* Term.pExpr is an IS operator */ +#define TERM_VARSELECT 0x1000 /* Term.pExpr contains a correlated sub-query */ +#define TERM_NOPARTIDX 0x2000 /* Not for use to enable a partial index */ + +/* +** An instance of the WhereScan object is used as an iterator for locating +** terms in the WHERE clause that are useful to the query planner. +*/ +struct WhereScan { + WhereClause *pOrigWC; /* Original, innermost WhereClause */ + WhereClause *pWC; /* WhereClause currently being scanned */ + const char *zCollName; /* Required collating sequence, if not NULL */ + Expr *pIdxExpr; /* Search for this index expression */ + char idxaff; /* Must match this affinity, if zCollName!=NULL */ + unsigned char nEquiv; /* Number of entries in aEquiv[] */ + unsigned char iEquiv; /* Next unused slot in aEquiv[] */ + u32 opMask; /* Acceptable operators */ + int k; /* Resume scanning at this->pWC->a[this->k] */ + int aiCur[11]; /* Cursors in the equivalence class */ + i16 aiColumn[11]; /* Corresponding column number in the eq-class */ +}; + +/* +** An instance of the following structure holds all information about a +** WHERE clause. Mostly this is a container for one or more WhereTerms. +** +** Explanation of pOuter: For a WHERE clause of the form +** +** a AND ((b AND c) OR (d AND e)) AND f +** +** There are separate WhereClause objects for the whole clause and for +** the subclauses "(b AND c)" and "(d AND e)". The pOuter field of the +** subclauses points to the WhereClause object for the whole clause. +*/ +struct WhereClause { + WhereInfo *pWInfo; /* WHERE clause processing context */ + WhereClause *pOuter; /* Outer conjunction */ + u8 op; /* Split operator. TK_AND or TK_OR */ + u8 hasOr; /* True if any a[].eOperator is WO_OR */ + int nTerm; /* Number of terms */ + int nSlot; /* Number of entries in a[] */ + WhereTerm *a; /* Each a[] describes a term of the WHERE cluase */ +#if defined(SQLITE_SMALL_STACK) + WhereTerm aStatic[1]; /* Initial static space for a[] */ +#else + WhereTerm aStatic[8]; /* Initial static space for a[] */ +#endif +}; + +/* +** A WhereTerm with eOperator==WO_OR has its u.pOrInfo pointer set to +** a dynamically allocated instance of the following structure. +*/ +struct WhereOrInfo { + WhereClause wc; /* Decomposition into subterms */ + Bitmask indexable; /* Bitmask of all indexable tables in the clause */ +}; + +/* +** A WhereTerm with eOperator==WO_AND has its u.pAndInfo pointer set to +** a dynamically allocated instance of the following structure. +*/ +struct WhereAndInfo { + WhereClause wc; /* The subexpression broken out */ +}; + +/* +** An instance of the following structure keeps track of a mapping +** between VDBE cursor numbers and bits of the bitmasks in WhereTerm. +** +** The VDBE cursor numbers are small integers contained in +** SrcList_item.iCursor and Expr.iTable fields. For any given WHERE +** clause, the cursor numbers might not begin with 0 and they might +** contain gaps in the numbering sequence. But we want to make maximum +** use of the bits in our bitmasks. This structure provides a mapping +** from the sparse cursor numbers into consecutive integers beginning +** with 0. +** +** If WhereMaskSet.ix[A]==B it means that The A-th bit of a Bitmask +** corresponds VDBE cursor number B. The A-th bit of a bitmask is 1<3, 5->1, 8->2, 29->0, +** 57->5, 73->4. Or one of 719 other combinations might be used. It +** does not really matter. What is important is that sparse cursor +** numbers all get mapped into bit numbers that begin with 0 and contain +** no gaps. +*/ +struct WhereMaskSet { + int bVarSelect; /* Used by sqlite3WhereExprUsage() */ + int n; /* Number of assigned cursor values */ + int ix[BMS]; /* Cursor assigned to each bit */ +}; + +/* +** Initialize a WhereMaskSet object +*/ +#define initMaskSet(P) (P)->n=0 + +/* +** This object is a convenience wrapper holding all information needed +** to construct WhereLoop objects for a particular query. +*/ +struct WhereLoopBuilder { + WhereInfo *pWInfo; /* Information about this WHERE */ + WhereClause *pWC; /* WHERE clause terms */ + ExprList *pOrderBy; /* ORDER BY clause */ + WhereLoop *pNew; /* Template WhereLoop */ + WhereOrSet *pOrSet; /* Record best loops here, if not NULL */ +#ifdef SQLITE_ENABLE_STAT4 + UnpackedRecord *pRec; /* Probe for stat4 (if required) */ + int nRecValid; /* Number of valid fields currently in pRec */ +#endif + unsigned int bldFlags; /* SQLITE_BLDF_* flags */ + unsigned int iPlanLimit; /* Search limiter */ +}; + +/* Allowed values for WhereLoopBuider.bldFlags */ +#define SQLITE_BLDF_INDEXED 0x0001 /* An index is used */ +#define SQLITE_BLDF_UNIQUE 0x0002 /* All keys of a UNIQUE index used */ + +/* The WhereLoopBuilder.iPlanLimit is used to limit the number of +** index+constraint combinations the query planner will consider for a +** particular query. If this parameter is unlimited, then certain +** pathological queries can spend excess time in the sqlite3WhereBegin() +** routine. The limit is high enough that is should not impact real-world +** queries. +** +** SQLITE_QUERY_PLANNER_LIMIT is the baseline limit. The limit is +** increased by SQLITE_QUERY_PLANNER_LIMIT_INCR before each term of the FROM +** clause is processed, so that every table in a join is guaranteed to be +** able to propose a some index+constraint combinations even if the initial +** baseline limit was exhausted by prior tables of the join. +*/ +#ifndef SQLITE_QUERY_PLANNER_LIMIT +# define SQLITE_QUERY_PLANNER_LIMIT 20000 +#endif +#ifndef SQLITE_QUERY_PLANNER_LIMIT_INCR +# define SQLITE_QUERY_PLANNER_LIMIT_INCR 1000 +#endif + +/* +** The WHERE clause processing routine has two halves. The +** first part does the start of the WHERE loop and the second +** half does the tail of the WHERE loop. An instance of +** this structure is returned by the first half and passed +** into the second half to give some continuity. +** +** An instance of this object holds the complete state of the query +** planner. +*/ +struct WhereInfo { + Parse *pParse; /* Parsing and code generating context */ + SrcList *pTabList; /* List of tables in the join */ + ExprList *pOrderBy; /* The ORDER BY clause or NULL */ + ExprList *pResultSet; /* Result set of the query */ + Expr *pWhere; /* The complete WHERE clause */ + LogEst iLimit; /* LIMIT if wctrlFlags has WHERE_USE_LIMIT */ + int aiCurOnePass[2]; /* OP_OpenWrite cursors for the ONEPASS opt */ + int iContinue; /* Jump here to continue with next record */ + int iBreak; /* Jump here to break out of the loop */ + int savedNQueryLoop; /* pParse->nQueryLoop outside the WHERE loop */ + u16 wctrlFlags; /* Flags originally passed to sqlite3WhereBegin() */ + u8 nLevel; /* Number of nested loop */ + i8 nOBSat; /* Number of ORDER BY terms satisfied by indices */ + u8 sorted; /* True if really sorted (not just grouped) */ + u8 eOnePass; /* ONEPASS_OFF, or _SINGLE, or _MULTI */ + u8 untestedTerms; /* Not all WHERE terms resolved by outer loop */ + u8 eDistinct; /* One of the WHERE_DISTINCT_* values */ + u8 bOrderedInnerLoop; /* True if only the inner-most loop is ordered */ + int iTop; /* The very beginning of the WHERE loop */ + WhereLoop *pLoops; /* List of all WhereLoop objects */ + Bitmask revMask; /* Mask of ORDER BY terms that need reversing */ + LogEst nRowOut; /* Estimated number of output rows */ + WhereClause sWC; /* Decomposition of the WHERE clause */ + WhereMaskSet sMaskSet; /* Map cursor numbers to bitmasks */ + WhereLevel a[1]; /* Information about each nest loop in WHERE */ +}; + +/* +** Private interfaces - callable only by other where.c routines. +** +** where.c: +*/ +SQLITE_PRIVATE Bitmask sqlite3WhereGetMask(WhereMaskSet*,int); +#ifdef WHERETRACE_ENABLED +SQLITE_PRIVATE void sqlite3WhereClausePrint(WhereClause *pWC); +#endif +SQLITE_PRIVATE WhereTerm *sqlite3WhereFindTerm( + WhereClause *pWC, /* The WHERE clause to be searched */ + int iCur, /* Cursor number of LHS */ + int iColumn, /* Column number of LHS */ + Bitmask notReady, /* RHS must not overlap with this mask */ + u32 op, /* Mask of WO_xx values describing operator */ + Index *pIdx /* Must be compatible with this index, if not NULL */ +); + +/* wherecode.c: */ +#ifndef SQLITE_OMIT_EXPLAIN +SQLITE_PRIVATE int sqlite3WhereExplainOneScan( + Parse *pParse, /* Parse context */ + SrcList *pTabList, /* Table list this loop refers to */ + WhereLevel *pLevel, /* Scan to write OP_Explain opcode for */ + u16 wctrlFlags /* Flags passed to sqlite3WhereBegin() */ +); +#else +# define sqlite3WhereExplainOneScan(u,v,w,x) 0 +#endif /* SQLITE_OMIT_EXPLAIN */ +#ifdef SQLITE_ENABLE_STMT_SCANSTATUS +SQLITE_PRIVATE void sqlite3WhereAddScanStatus( + Vdbe *v, /* Vdbe to add scanstatus entry to */ + SrcList *pSrclist, /* FROM clause pLvl reads data from */ + WhereLevel *pLvl, /* Level to add scanstatus() entry for */ + int addrExplain /* Address of OP_Explain (or 0) */ +); +#else +# define sqlite3WhereAddScanStatus(a, b, c, d) ((void)d) +#endif +SQLITE_PRIVATE Bitmask sqlite3WhereCodeOneLoopStart( + Parse *pParse, /* Parsing context */ + Vdbe *v, /* Prepared statement under construction */ + WhereInfo *pWInfo, /* Complete information about the WHERE clause */ + int iLevel, /* Which level of pWInfo->a[] should be coded */ + WhereLevel *pLevel, /* The current level pointer */ + Bitmask notReady /* Which tables are currently available */ +); + +/* whereexpr.c: */ +SQLITE_PRIVATE void sqlite3WhereClauseInit(WhereClause*,WhereInfo*); +SQLITE_PRIVATE void sqlite3WhereClauseClear(WhereClause*); +SQLITE_PRIVATE void sqlite3WhereSplit(WhereClause*,Expr*,u8); +SQLITE_PRIVATE Bitmask sqlite3WhereExprUsage(WhereMaskSet*, Expr*); +SQLITE_PRIVATE Bitmask sqlite3WhereExprUsageNN(WhereMaskSet*, Expr*); +SQLITE_PRIVATE Bitmask sqlite3WhereExprListUsage(WhereMaskSet*, ExprList*); +SQLITE_PRIVATE void sqlite3WhereExprAnalyze(SrcList*, WhereClause*); +SQLITE_PRIVATE void sqlite3WhereTabFuncArgs(Parse*, struct SrcList_item*, WhereClause*); + + + + + +/* +** Bitmasks for the operators on WhereTerm objects. These are all +** operators that are of interest to the query planner. An +** OR-ed combination of these values can be used when searching for +** particular WhereTerms within a WhereClause. +** +** Value constraints: +** WO_EQ == SQLITE_INDEX_CONSTRAINT_EQ +** WO_LT == SQLITE_INDEX_CONSTRAINT_LT +** WO_LE == SQLITE_INDEX_CONSTRAINT_LE +** WO_GT == SQLITE_INDEX_CONSTRAINT_GT +** WO_GE == SQLITE_INDEX_CONSTRAINT_GE +*/ +#define WO_IN 0x0001 +#define WO_EQ 0x0002 +#define WO_LT (WO_EQ<<(TK_LT-TK_EQ)) +#define WO_LE (WO_EQ<<(TK_LE-TK_EQ)) +#define WO_GT (WO_EQ<<(TK_GT-TK_EQ)) +#define WO_GE (WO_EQ<<(TK_GE-TK_EQ)) +#define WO_AUX 0x0040 /* Op useful to virtual tables only */ +#define WO_IS 0x0080 +#define WO_ISNULL 0x0100 +#define WO_OR 0x0200 /* Two or more OR-connected terms */ +#define WO_AND 0x0400 /* Two or more AND-connected terms */ +#define WO_EQUIV 0x0800 /* Of the form A==B, both columns */ +#define WO_NOOP 0x1000 /* This term does not restrict search space */ + +#define WO_ALL 0x1fff /* Mask of all possible WO_* values */ +#define WO_SINGLE 0x01ff /* Mask of all non-compound WO_* values */ + +/* +** These are definitions of bits in the WhereLoop.wsFlags field. +** The particular combination of bits in each WhereLoop help to +** determine the algorithm that WhereLoop represents. +*/ +#define WHERE_COLUMN_EQ 0x00000001 /* x=EXPR */ +#define WHERE_COLUMN_RANGE 0x00000002 /* xEXPR */ +#define WHERE_COLUMN_IN 0x00000004 /* x IN (...) */ +#define WHERE_COLUMN_NULL 0x00000008 /* x IS NULL */ +#define WHERE_CONSTRAINT 0x0000000f /* Any of the WHERE_COLUMN_xxx values */ +#define WHERE_TOP_LIMIT 0x00000010 /* xEXPR or x>=EXPR constraint */ +#define WHERE_BOTH_LIMIT 0x00000030 /* Both x>EXPR and xaiColumn[i]; + if( i==XN_EXPR ) return ""; + if( i==XN_ROWID ) return "rowid"; + return pIdx->pTable->aCol[i].zName; +} + +/* +** This routine is a helper for explainIndexRange() below +** +** pStr holds the text of an expression that we are building up one term +** at a time. This routine adds a new term to the end of the expression. +** Terms are separated by AND so add the "AND" text for second and subsequent +** terms only. +*/ +static void explainAppendTerm( + StrAccum *pStr, /* The text expression being built */ + Index *pIdx, /* Index to read column names from */ + int nTerm, /* Number of terms */ + int iTerm, /* Zero-based index of first term. */ + int bAnd, /* Non-zero to append " AND " */ + const char *zOp /* Name of the operator */ +){ + int i; + + assert( nTerm>=1 ); + if( bAnd ) sqlite3_str_append(pStr, " AND ", 5); + + if( nTerm>1 ) sqlite3_str_append(pStr, "(", 1); + for(i=0; i1 ) sqlite3_str_append(pStr, ")", 1); + + sqlite3_str_append(pStr, zOp, 1); + + if( nTerm>1 ) sqlite3_str_append(pStr, "(", 1); + for(i=0; i1 ) sqlite3_str_append(pStr, ")", 1); +} + +/* +** Argument pLevel describes a strategy for scanning table pTab. This +** function appends text to pStr that describes the subset of table +** rows scanned by the strategy in the form of an SQL expression. +** +** For example, if the query: +** +** SELECT * FROM t1 WHERE a=1 AND b>2; +** +** is run and there is an index on (a, b), then this function returns a +** string similar to: +** +** "a=? AND b>?" +*/ +static void explainIndexRange(StrAccum *pStr, WhereLoop *pLoop){ + Index *pIndex = pLoop->u.btree.pIndex; + u16 nEq = pLoop->u.btree.nEq; + u16 nSkip = pLoop->nSkip; + int i, j; + + if( nEq==0 && (pLoop->wsFlags&(WHERE_BTM_LIMIT|WHERE_TOP_LIMIT))==0 ) return; + sqlite3_str_append(pStr, " (", 2); + for(i=0; i=nSkip ? "%s=?" : "ANY(%s)", z); + } + + j = i; + if( pLoop->wsFlags&WHERE_BTM_LIMIT ){ + explainAppendTerm(pStr, pIndex, pLoop->u.btree.nBtm, j, i, ">"); + i = 1; + } + if( pLoop->wsFlags&WHERE_TOP_LIMIT ){ + explainAppendTerm(pStr, pIndex, pLoop->u.btree.nTop, j, i, "<"); + } + sqlite3_str_append(pStr, ")", 1); +} + +/* +** This function is a no-op unless currently processing an EXPLAIN QUERY PLAN +** command, or if either SQLITE_DEBUG or SQLITE_ENABLE_STMT_SCANSTATUS was +** defined at compile-time. If it is not a no-op, a single OP_Explain opcode +** is added to the output to describe the table scan strategy in pLevel. +** +** If an OP_Explain opcode is added to the VM, its address is returned. +** Otherwise, if no OP_Explain is coded, zero is returned. +*/ +SQLITE_PRIVATE int sqlite3WhereExplainOneScan( + Parse *pParse, /* Parse context */ + SrcList *pTabList, /* Table list this loop refers to */ + WhereLevel *pLevel, /* Scan to write OP_Explain opcode for */ + u16 wctrlFlags /* Flags passed to sqlite3WhereBegin() */ +){ + int ret = 0; +#if !defined(SQLITE_DEBUG) && !defined(SQLITE_ENABLE_STMT_SCANSTATUS) + if( sqlite3ParseToplevel(pParse)->explain==2 ) +#endif + { + struct SrcList_item *pItem = &pTabList->a[pLevel->iFrom]; + Vdbe *v = pParse->pVdbe; /* VM being constructed */ + sqlite3 *db = pParse->db; /* Database handle */ + int isSearch; /* True for a SEARCH. False for SCAN. */ + WhereLoop *pLoop; /* The controlling WhereLoop object */ + u32 flags; /* Flags that describe this loop */ + char *zMsg; /* Text to add to EQP output */ + StrAccum str; /* EQP output string */ + char zBuf[100]; /* Initial space for EQP output string */ + + pLoop = pLevel->pWLoop; + flags = pLoop->wsFlags; + if( (flags&WHERE_MULTI_OR) || (wctrlFlags&WHERE_OR_SUBCLAUSE) ) return 0; + + isSearch = (flags&(WHERE_BTM_LIMIT|WHERE_TOP_LIMIT))!=0 + || ((flags&WHERE_VIRTUALTABLE)==0 && (pLoop->u.btree.nEq>0)) + || (wctrlFlags&(WHERE_ORDERBY_MIN|WHERE_ORDERBY_MAX)); + + sqlite3StrAccumInit(&str, db, zBuf, sizeof(zBuf), SQLITE_MAX_LENGTH); + sqlite3_str_appendall(&str, isSearch ? "SEARCH" : "SCAN"); + if( pItem->pSelect ){ + sqlite3_str_appendf(&str, " SUBQUERY %u", pItem->pSelect->selId); + }else{ + sqlite3_str_appendf(&str, " TABLE %s", pItem->zName); + } + + if( pItem->zAlias ){ + sqlite3_str_appendf(&str, " AS %s", pItem->zAlias); + } + if( (flags & (WHERE_IPK|WHERE_VIRTUALTABLE))==0 ){ + const char *zFmt = 0; + Index *pIdx; + + assert( pLoop->u.btree.pIndex!=0 ); + pIdx = pLoop->u.btree.pIndex; + assert( !(flags&WHERE_AUTO_INDEX) || (flags&WHERE_IDX_ONLY) ); + if( !HasRowid(pItem->pTab) && IsPrimaryKeyIndex(pIdx) ){ + if( isSearch ){ + zFmt = "PRIMARY KEY"; + } + }else if( flags & WHERE_PARTIALIDX ){ + zFmt = "AUTOMATIC PARTIAL COVERING INDEX"; + }else if( flags & WHERE_AUTO_INDEX ){ + zFmt = "AUTOMATIC COVERING INDEX"; + }else if( flags & WHERE_IDX_ONLY ){ + zFmt = "COVERING INDEX %s"; + }else{ + zFmt = "INDEX %s"; + } + if( zFmt ){ + sqlite3_str_append(&str, " USING ", 7); + sqlite3_str_appendf(&str, zFmt, pIdx->zName); + explainIndexRange(&str, pLoop); + } + }else if( (flags & WHERE_IPK)!=0 && (flags & WHERE_CONSTRAINT)!=0 ){ + const char *zRangeOp; + if( flags&(WHERE_COLUMN_EQ|WHERE_COLUMN_IN) ){ + zRangeOp = "="; + }else if( (flags&WHERE_BOTH_LIMIT)==WHERE_BOTH_LIMIT ){ + zRangeOp = ">? AND rowid<"; + }else if( flags&WHERE_BTM_LIMIT ){ + zRangeOp = ">"; + }else{ + assert( flags&WHERE_TOP_LIMIT); + zRangeOp = "<"; + } + sqlite3_str_appendf(&str, + " USING INTEGER PRIMARY KEY (rowid%s?)",zRangeOp); + } +#ifndef SQLITE_OMIT_VIRTUALTABLE + else if( (flags & WHERE_VIRTUALTABLE)!=0 ){ + sqlite3_str_appendf(&str, " VIRTUAL TABLE INDEX %d:%s", + pLoop->u.vtab.idxNum, pLoop->u.vtab.idxStr); + } +#endif +#ifdef SQLITE_EXPLAIN_ESTIMATED_ROWS + if( pLoop->nOut>=10 ){ + sqlite3_str_appendf(&str, " (~%llu rows)", + sqlite3LogEstToInt(pLoop->nOut)); + }else{ + sqlite3_str_append(&str, " (~1 row)", 9); + } +#endif + zMsg = sqlite3StrAccumFinish(&str); + sqlite3ExplainBreakpoint("",zMsg); + ret = sqlite3VdbeAddOp4(v, OP_Explain, sqlite3VdbeCurrentAddr(v), + pParse->addrExplain, 0, zMsg,P4_DYNAMIC); + } + return ret; +} +#endif /* SQLITE_OMIT_EXPLAIN */ + +#ifdef SQLITE_ENABLE_STMT_SCANSTATUS +/* +** Configure the VM passed as the first argument with an +** sqlite3_stmt_scanstatus() entry corresponding to the scan used to +** implement level pLvl. Argument pSrclist is a pointer to the FROM +** clause that the scan reads data from. +** +** If argument addrExplain is not 0, it must be the address of an +** OP_Explain instruction that describes the same loop. +*/ +SQLITE_PRIVATE void sqlite3WhereAddScanStatus( + Vdbe *v, /* Vdbe to add scanstatus entry to */ + SrcList *pSrclist, /* FROM clause pLvl reads data from */ + WhereLevel *pLvl, /* Level to add scanstatus() entry for */ + int addrExplain /* Address of OP_Explain (or 0) */ +){ + const char *zObj = 0; + WhereLoop *pLoop = pLvl->pWLoop; + if( (pLoop->wsFlags & WHERE_VIRTUALTABLE)==0 && pLoop->u.btree.pIndex!=0 ){ + zObj = pLoop->u.btree.pIndex->zName; + }else{ + zObj = pSrclist->a[pLvl->iFrom].zName; + } + sqlite3VdbeScanStatus( + v, addrExplain, pLvl->addrBody, pLvl->addrVisit, pLoop->nOut, zObj + ); +} +#endif + + +/* +** Disable a term in the WHERE clause. Except, do not disable the term +** if it controls a LEFT OUTER JOIN and it did not originate in the ON +** or USING clause of that join. +** +** Consider the term t2.z='ok' in the following queries: +** +** (1) SELECT * FROM t1 LEFT JOIN t2 ON t1.a=t2.x WHERE t2.z='ok' +** (2) SELECT * FROM t1 LEFT JOIN t2 ON t1.a=t2.x AND t2.z='ok' +** (3) SELECT * FROM t1, t2 WHERE t1.a=t2.x AND t2.z='ok' +** +** The t2.z='ok' is disabled in the in (2) because it originates +** in the ON clause. The term is disabled in (3) because it is not part +** of a LEFT OUTER JOIN. In (1), the term is not disabled. +** +** Disabling a term causes that term to not be tested in the inner loop +** of the join. Disabling is an optimization. When terms are satisfied +** by indices, we disable them to prevent redundant tests in the inner +** loop. We would get the correct results if nothing were ever disabled, +** but joins might run a little slower. The trick is to disable as much +** as we can without disabling too much. If we disabled in (1), we'd get +** the wrong answer. See ticket #813. +** +** If all the children of a term are disabled, then that term is also +** automatically disabled. In this way, terms get disabled if derived +** virtual terms are tested first. For example: +** +** x GLOB 'abc*' AND x>='abc' AND x<'acd' +** \___________/ \______/ \_____/ +** parent child1 child2 +** +** Only the parent term was in the original WHERE clause. The child1 +** and child2 terms were added by the LIKE optimization. If both of +** the virtual child terms are valid, then testing of the parent can be +** skipped. +** +** Usually the parent term is marked as TERM_CODED. But if the parent +** term was originally TERM_LIKE, then the parent gets TERM_LIKECOND instead. +** The TERM_LIKECOND marking indicates that the term should be coded inside +** a conditional such that is only evaluated on the second pass of a +** LIKE-optimization loop, when scanning BLOBs instead of strings. +*/ +static void disableTerm(WhereLevel *pLevel, WhereTerm *pTerm){ + int nLoop = 0; + assert( pTerm!=0 ); + while( (pTerm->wtFlags & TERM_CODED)==0 + && (pLevel->iLeftJoin==0 || ExprHasProperty(pTerm->pExpr, EP_FromJoin)) + && (pLevel->notReady & pTerm->prereqAll)==0 + ){ + if( nLoop && (pTerm->wtFlags & TERM_LIKE)!=0 ){ + pTerm->wtFlags |= TERM_LIKECOND; + }else{ + pTerm->wtFlags |= TERM_CODED; + } + if( pTerm->iParent<0 ) break; + pTerm = &pTerm->pWC->a[pTerm->iParent]; + assert( pTerm!=0 ); + pTerm->nChild--; + if( pTerm->nChild!=0 ) break; + nLoop++; + } +} + +/* +** Code an OP_Affinity opcode to apply the column affinity string zAff +** to the n registers starting at base. +** +** As an optimization, SQLITE_AFF_BLOB and SQLITE_AFF_NONE entries (which +** are no-ops) at the beginning and end of zAff are ignored. If all entries +** in zAff are SQLITE_AFF_BLOB or SQLITE_AFF_NONE, then no code gets generated. +** +** This routine makes its own copy of zAff so that the caller is free +** to modify zAff after this routine returns. +*/ +static void codeApplyAffinity(Parse *pParse, int base, int n, char *zAff){ + Vdbe *v = pParse->pVdbe; + if( zAff==0 ){ + assert( pParse->db->mallocFailed ); + return; + } + assert( v!=0 ); + + /* Adjust base and n to skip over SQLITE_AFF_BLOB and SQLITE_AFF_NONE + ** entries at the beginning and end of the affinity string. + */ + assert( SQLITE_AFF_NONE0 && zAff[0]<=SQLITE_AFF_BLOB ){ + n--; + base++; + zAff++; + } + while( n>1 && zAff[n-1]<=SQLITE_AFF_BLOB ){ + n--; + } + + /* Code the OP_Affinity opcode if there is anything left to do. */ + if( n>0 ){ + sqlite3VdbeAddOp4(v, OP_Affinity, base, n, 0, zAff, n); + } +} + +/* +** Expression pRight, which is the RHS of a comparison operation, is +** either a vector of n elements or, if n==1, a scalar expression. +** Before the comparison operation, affinity zAff is to be applied +** to the pRight values. This function modifies characters within the +** affinity string to SQLITE_AFF_BLOB if either: +** +** * the comparison will be performed with no affinity, or +** * the affinity change in zAff is guaranteed not to change the value. +*/ +static void updateRangeAffinityStr( + Expr *pRight, /* RHS of comparison */ + int n, /* Number of vector elements in comparison */ + char *zAff /* Affinity string to modify */ +){ + int i; + for(i=0; idb; + Expr *pNew = sqlite3ExprDup(db, pX, 0); + if( db->mallocFailed==0 ){ + ExprList *pOrigRhs = pNew->x.pSelect->pEList; /* Original unmodified RHS */ + ExprList *pOrigLhs = pNew->pLeft->x.pList; /* Original unmodified LHS */ + ExprList *pRhs = 0; /* New RHS after modifications */ + ExprList *pLhs = 0; /* New LHS after mods */ + int i; /* Loop counter */ + Select *pSelect; /* Pointer to the SELECT on the RHS */ + + for(i=iEq; inLTerm; i++){ + if( pLoop->aLTerm[i]->pExpr==pX ){ + int iField = pLoop->aLTerm[i]->iField - 1; + if( pOrigRhs->a[iField].pExpr==0 ) continue; /* Duplicate PK column */ + pRhs = sqlite3ExprListAppend(pParse, pRhs, pOrigRhs->a[iField].pExpr); + pOrigRhs->a[iField].pExpr = 0; + assert( pOrigLhs->a[iField].pExpr!=0 ); + pLhs = sqlite3ExprListAppend(pParse, pLhs, pOrigLhs->a[iField].pExpr); + pOrigLhs->a[iField].pExpr = 0; + } + } + sqlite3ExprListDelete(db, pOrigRhs); + sqlite3ExprListDelete(db, pOrigLhs); + pNew->pLeft->x.pList = pLhs; + pNew->x.pSelect->pEList = pRhs; + if( pLhs && pLhs->nExpr==1 ){ + /* Take care here not to generate a TK_VECTOR containing only a + ** single value. Since the parser never creates such a vector, some + ** of the subroutines do not handle this case. */ + Expr *p = pLhs->a[0].pExpr; + pLhs->a[0].pExpr = 0; + sqlite3ExprDelete(db, pNew->pLeft); + pNew->pLeft = p; + } + pSelect = pNew->x.pSelect; + if( pSelect->pOrderBy ){ + /* If the SELECT statement has an ORDER BY clause, zero the + ** iOrderByCol variables. These are set to non-zero when an + ** ORDER BY term exactly matches one of the terms of the + ** result-set. Since the result-set of the SELECT statement may + ** have been modified or reordered, these variables are no longer + ** set correctly. Since setting them is just an optimization, + ** it's easiest just to zero them here. */ + ExprList *pOrderBy = pSelect->pOrderBy; + for(i=0; inExpr; i++){ + pOrderBy->a[i].u.x.iOrderByCol = 0; + } + } + +#if 0 + printf("For indexing, change the IN expr:\n"); + sqlite3TreeViewExpr(0, pX, 0); + printf("Into:\n"); + sqlite3TreeViewExpr(0, pNew, 0); +#endif + } + return pNew; +} + + +/* +** Generate code for a single equality term of the WHERE clause. An equality +** term can be either X=expr or X IN (...). pTerm is the term to be +** coded. +** +** The current value for the constraint is left in a register, the index +** of which is returned. An attempt is made store the result in iTarget but +** this is only guaranteed for TK_ISNULL and TK_IN constraints. If the +** constraint is a TK_EQ or TK_IS, then the current value might be left in +** some other register and it is the caller's responsibility to compensate. +** +** For a constraint of the form X=expr, the expression is evaluated in +** straight-line code. For constraints of the form X IN (...) +** this routine sets up a loop that will iterate over all values of X. +*/ +static int codeEqualityTerm( + Parse *pParse, /* The parsing context */ + WhereTerm *pTerm, /* The term of the WHERE clause to be coded */ + WhereLevel *pLevel, /* The level of the FROM clause we are working on */ + int iEq, /* Index of the equality term within this level */ + int bRev, /* True for reverse-order IN operations */ + int iTarget /* Attempt to leave results in this register */ +){ + Expr *pX = pTerm->pExpr; + Vdbe *v = pParse->pVdbe; + int iReg; /* Register holding results */ + + assert( pLevel->pWLoop->aLTerm[iEq]==pTerm ); + assert( iTarget>0 ); + if( pX->op==TK_EQ || pX->op==TK_IS ){ + iReg = sqlite3ExprCodeTarget(pParse, pX->pRight, iTarget); + }else if( pX->op==TK_ISNULL ){ + iReg = iTarget; + sqlite3VdbeAddOp2(v, OP_Null, 0, iReg); +#ifndef SQLITE_OMIT_SUBQUERY + }else{ + int eType = IN_INDEX_NOOP; + int iTab; + struct InLoop *pIn; + WhereLoop *pLoop = pLevel->pWLoop; + int i; + int nEq = 0; + int *aiMap = 0; + + if( (pLoop->wsFlags & WHERE_VIRTUALTABLE)==0 + && pLoop->u.btree.pIndex!=0 + && pLoop->u.btree.pIndex->aSortOrder[iEq] + ){ + testcase( iEq==0 ); + testcase( bRev ); + bRev = !bRev; + } + assert( pX->op==TK_IN ); + iReg = iTarget; + + for(i=0; iaLTerm[i] && pLoop->aLTerm[i]->pExpr==pX ){ + disableTerm(pLevel, pTerm); + return iTarget; + } + } + for(i=iEq;inLTerm; i++){ + assert( pLoop->aLTerm[i]!=0 ); + if( pLoop->aLTerm[i]->pExpr==pX ) nEq++; + } + + iTab = 0; + if( (pX->flags & EP_xIsSelect)==0 || pX->x.pSelect->pEList->nExpr==1 ){ + eType = sqlite3FindInIndex(pParse, pX, IN_INDEX_LOOP, 0, 0, &iTab); + }else{ + sqlite3 *db = pParse->db; + pX = removeUnindexableInClauseTerms(pParse, iEq, pLoop, pX); + + if( !db->mallocFailed ){ + aiMap = (int*)sqlite3DbMallocZero(pParse->db, sizeof(int)*nEq); + eType = sqlite3FindInIndex(pParse, pX, IN_INDEX_LOOP, 0, aiMap, &iTab); + pTerm->pExpr->iTable = iTab; + } + sqlite3ExprDelete(db, pX); + pX = pTerm->pExpr; + } + + if( eType==IN_INDEX_INDEX_DESC ){ + testcase( bRev ); + bRev = !bRev; + } + sqlite3VdbeAddOp2(v, bRev ? OP_Last : OP_Rewind, iTab, 0); + VdbeCoverageIf(v, bRev); + VdbeCoverageIf(v, !bRev); + assert( (pLoop->wsFlags & WHERE_MULTI_OR)==0 ); + + pLoop->wsFlags |= WHERE_IN_ABLE; + if( pLevel->u.in.nIn==0 ){ + pLevel->addrNxt = sqlite3VdbeMakeLabel(pParse); + } + + i = pLevel->u.in.nIn; + pLevel->u.in.nIn += nEq; + pLevel->u.in.aInLoop = + sqlite3DbReallocOrFree(pParse->db, pLevel->u.in.aInLoop, + sizeof(pLevel->u.in.aInLoop[0])*pLevel->u.in.nIn); + pIn = pLevel->u.in.aInLoop; + if( pIn ){ + int iMap = 0; /* Index in aiMap[] */ + pIn += i; + for(i=iEq;inLTerm; i++){ + if( pLoop->aLTerm[i]->pExpr==pX ){ + int iOut = iReg + i - iEq; + if( eType==IN_INDEX_ROWID ){ + pIn->addrInTop = sqlite3VdbeAddOp2(v, OP_Rowid, iTab, iOut); + }else{ + int iCol = aiMap ? aiMap[iMap++] : 0; + pIn->addrInTop = sqlite3VdbeAddOp3(v,OP_Column,iTab, iCol, iOut); + } + sqlite3VdbeAddOp1(v, OP_IsNull, iOut); VdbeCoverage(v); + if( i==iEq ){ + pIn->iCur = iTab; + pIn->eEndLoopOp = bRev ? OP_Prev : OP_Next; + if( iEq>0 && (pLoop->wsFlags & WHERE_VIRTUALTABLE)==0 ){ + pIn->iBase = iReg - i; + pIn->nPrefix = i; + pLoop->wsFlags |= WHERE_IN_EARLYOUT; + }else{ + pIn->nPrefix = 0; + } + }else{ + pIn->eEndLoopOp = OP_Noop; + } + pIn++; + } + } + }else{ + pLevel->u.in.nIn = 0; + } + sqlite3DbFree(pParse->db, aiMap); +#endif + } + disableTerm(pLevel, pTerm); + return iReg; +} + +/* +** Generate code that will evaluate all == and IN constraints for an +** index scan. +** +** For example, consider table t1(a,b,c,d,e,f) with index i1(a,b,c). +** Suppose the WHERE clause is this: a==5 AND b IN (1,2,3) AND c>5 AND c<10 +** The index has as many as three equality constraints, but in this +** example, the third "c" value is an inequality. So only two +** constraints are coded. This routine will generate code to evaluate +** a==5 and b IN (1,2,3). The current values for a and b will be stored +** in consecutive registers and the index of the first register is returned. +** +** In the example above nEq==2. But this subroutine works for any value +** of nEq including 0. If nEq==0, this routine is nearly a no-op. +** The only thing it does is allocate the pLevel->iMem memory cell and +** compute the affinity string. +** +** The nExtraReg parameter is 0 or 1. It is 0 if all WHERE clause constraints +** are == or IN and are covered by the nEq. nExtraReg is 1 if there is +** an inequality constraint (such as the "c>=5 AND c<10" in the example) that +** occurs after the nEq quality constraints. +** +** This routine allocates a range of nEq+nExtraReg memory cells and returns +** the index of the first memory cell in that range. The code that +** calls this routine will use that memory range to store keys for +** start and termination conditions of the loop. +** key value of the loop. If one or more IN operators appear, then +** this routine allocates an additional nEq memory cells for internal +** use. +** +** Before returning, *pzAff is set to point to a buffer containing a +** copy of the column affinity string of the index allocated using +** sqlite3DbMalloc(). Except, entries in the copy of the string associated +** with equality constraints that use BLOB or NONE affinity are set to +** SQLITE_AFF_BLOB. This is to deal with SQL such as the following: +** +** CREATE TABLE t1(a TEXT PRIMARY KEY, b); +** SELECT ... FROM t1 AS t2, t1 WHERE t1.a = t2.b; +** +** In the example above, the index on t1(a) has TEXT affinity. But since +** the right hand side of the equality constraint (t2.b) has BLOB/NONE affinity, +** no conversion should be attempted before using a t2.b value as part of +** a key to search the index. Hence the first byte in the returned affinity +** string in this example would be set to SQLITE_AFF_BLOB. +*/ +static int codeAllEqualityTerms( + Parse *pParse, /* Parsing context */ + WhereLevel *pLevel, /* Which nested loop of the FROM we are coding */ + int bRev, /* Reverse the order of IN operators */ + int nExtraReg, /* Number of extra registers to allocate */ + char **pzAff /* OUT: Set to point to affinity string */ +){ + u16 nEq; /* The number of == or IN constraints to code */ + u16 nSkip; /* Number of left-most columns to skip */ + Vdbe *v = pParse->pVdbe; /* The vm under construction */ + Index *pIdx; /* The index being used for this loop */ + WhereTerm *pTerm; /* A single constraint term */ + WhereLoop *pLoop; /* The WhereLoop object */ + int j; /* Loop counter */ + int regBase; /* Base register */ + int nReg; /* Number of registers to allocate */ + char *zAff; /* Affinity string to return */ + + /* This module is only called on query plans that use an index. */ + pLoop = pLevel->pWLoop; + assert( (pLoop->wsFlags & WHERE_VIRTUALTABLE)==0 ); + nEq = pLoop->u.btree.nEq; + nSkip = pLoop->nSkip; + pIdx = pLoop->u.btree.pIndex; + assert( pIdx!=0 ); + + /* Figure out how many memory cells we will need then allocate them. + */ + regBase = pParse->nMem + 1; + nReg = pLoop->u.btree.nEq + nExtraReg; + pParse->nMem += nReg; + + zAff = sqlite3DbStrDup(pParse->db,sqlite3IndexAffinityStr(pParse->db,pIdx)); + assert( zAff!=0 || pParse->db->mallocFailed ); + + if( nSkip ){ + int iIdxCur = pLevel->iIdxCur; + sqlite3VdbeAddOp1(v, (bRev?OP_Last:OP_Rewind), iIdxCur); + VdbeCoverageIf(v, bRev==0); + VdbeCoverageIf(v, bRev!=0); + VdbeComment((v, "begin skip-scan on %s", pIdx->zName)); + j = sqlite3VdbeAddOp0(v, OP_Goto); + pLevel->addrSkip = sqlite3VdbeAddOp4Int(v, (bRev?OP_SeekLT:OP_SeekGT), + iIdxCur, 0, regBase, nSkip); + VdbeCoverageIf(v, bRev==0); + VdbeCoverageIf(v, bRev!=0); + sqlite3VdbeJumpHere(v, j); + for(j=0; jaiColumn[j]==XN_EXPR ); + VdbeComment((v, "%s", explainIndexColumnName(pIdx, j))); + } + } + + /* Evaluate the equality constraints + */ + assert( zAff==0 || (int)strlen(zAff)>=nEq ); + for(j=nSkip; jaLTerm[j]; + assert( pTerm!=0 ); + /* The following testcase is true for indices with redundant columns. + ** Ex: CREATE INDEX i1 ON t1(a,b,a); SELECT * FROM t1 WHERE a=0 AND b=0; */ + testcase( (pTerm->wtFlags & TERM_CODED)!=0 ); + testcase( pTerm->wtFlags & TERM_VIRTUAL ); + r1 = codeEqualityTerm(pParse, pTerm, pLevel, j, bRev, regBase+j); + if( r1!=regBase+j ){ + if( nReg==1 ){ + sqlite3ReleaseTempReg(pParse, regBase); + regBase = r1; + }else{ + sqlite3VdbeAddOp2(v, OP_SCopy, r1, regBase+j); + } + } + if( pTerm->eOperator & WO_IN ){ + if( pTerm->pExpr->flags & EP_xIsSelect ){ + /* No affinity ever needs to be (or should be) applied to a value + ** from the RHS of an "? IN (SELECT ...)" expression. The + ** sqlite3FindInIndex() routine has already ensured that the + ** affinity of the comparison has been applied to the value. */ + if( zAff ) zAff[j] = SQLITE_AFF_BLOB; + } + }else if( (pTerm->eOperator & WO_ISNULL)==0 ){ + Expr *pRight = pTerm->pExpr->pRight; + if( (pTerm->wtFlags & TERM_IS)==0 && sqlite3ExprCanBeNull(pRight) ){ + sqlite3VdbeAddOp2(v, OP_IsNull, regBase+j, pLevel->addrBrk); + VdbeCoverage(v); + } + if( zAff ){ + if( sqlite3CompareAffinity(pRight, zAff[j])==SQLITE_AFF_BLOB ){ + zAff[j] = SQLITE_AFF_BLOB; + } + if( sqlite3ExprNeedsNoAffinityChange(pRight, zAff[j]) ){ + zAff[j] = SQLITE_AFF_BLOB; + } + } + } + } + *pzAff = zAff; + return regBase; +} + +#ifndef SQLITE_LIKE_DOESNT_MATCH_BLOBS +/* +** If the most recently coded instruction is a constant range constraint +** (a string literal) that originated from the LIKE optimization, then +** set P3 and P5 on the OP_String opcode so that the string will be cast +** to a BLOB at appropriate times. +** +** The LIKE optimization trys to evaluate "x LIKE 'abc%'" as a range +** expression: "x>='ABC' AND x<'abd'". But this requires that the range +** scan loop run twice, once for strings and a second time for BLOBs. +** The OP_String opcodes on the second pass convert the upper and lower +** bound string constants to blobs. This routine makes the necessary changes +** to the OP_String opcodes for that to happen. +** +** Except, of course, if SQLITE_LIKE_DOESNT_MATCH_BLOBS is defined, then +** only the one pass through the string space is required, so this routine +** becomes a no-op. +*/ +static void whereLikeOptimizationStringFixup( + Vdbe *v, /* prepared statement under construction */ + WhereLevel *pLevel, /* The loop that contains the LIKE operator */ + WhereTerm *pTerm /* The upper or lower bound just coded */ +){ + if( pTerm->wtFlags & TERM_LIKEOPT ){ + VdbeOp *pOp; + assert( pLevel->iLikeRepCntr>0 ); + pOp = sqlite3VdbeGetOp(v, -1); + assert( pOp!=0 ); + assert( pOp->opcode==OP_String8 + || pTerm->pWC->pWInfo->pParse->db->mallocFailed ); + pOp->p3 = (int)(pLevel->iLikeRepCntr>>1); /* Register holding counter */ + pOp->p5 = (u8)(pLevel->iLikeRepCntr&1); /* ASC or DESC */ + } +} +#else +# define whereLikeOptimizationStringFixup(A,B,C) +#endif + +#ifdef SQLITE_ENABLE_CURSOR_HINTS +/* +** Information is passed from codeCursorHint() down to individual nodes of +** the expression tree (by sqlite3WalkExpr()) using an instance of this +** structure. +*/ +struct CCurHint { + int iTabCur; /* Cursor for the main table */ + int iIdxCur; /* Cursor for the index, if pIdx!=0. Unused otherwise */ + Index *pIdx; /* The index used to access the table */ +}; + +/* +** This function is called for every node of an expression that is a candidate +** for a cursor hint on an index cursor. For TK_COLUMN nodes that reference +** the table CCurHint.iTabCur, verify that the same column can be +** accessed through the index. If it cannot, then set pWalker->eCode to 1. +*/ +static int codeCursorHintCheckExpr(Walker *pWalker, Expr *pExpr){ + struct CCurHint *pHint = pWalker->u.pCCurHint; + assert( pHint->pIdx!=0 ); + if( pExpr->op==TK_COLUMN + && pExpr->iTable==pHint->iTabCur + && sqlite3ColumnOfIndex(pHint->pIdx, pExpr->iColumn)<0 + ){ + pWalker->eCode = 1; + } + return WRC_Continue; +} + +/* +** Test whether or not expression pExpr, which was part of a WHERE clause, +** should be included in the cursor-hint for a table that is on the rhs +** of a LEFT JOIN. Set Walker.eCode to non-zero before returning if the +** expression is not suitable. +** +** An expression is unsuitable if it might evaluate to non NULL even if +** a TK_COLUMN node that does affect the value of the expression is set +** to NULL. For example: +** +** col IS NULL +** col IS NOT NULL +** coalesce(col, 1) +** CASE WHEN col THEN 0 ELSE 1 END +*/ +static int codeCursorHintIsOrFunction(Walker *pWalker, Expr *pExpr){ + if( pExpr->op==TK_IS + || pExpr->op==TK_ISNULL || pExpr->op==TK_ISNOT + || pExpr->op==TK_NOTNULL || pExpr->op==TK_CASE + ){ + pWalker->eCode = 1; + }else if( pExpr->op==TK_FUNCTION ){ + int d1; + char d2[4]; + if( 0==sqlite3IsLikeFunction(pWalker->pParse->db, pExpr, &d1, d2) ){ + pWalker->eCode = 1; + } + } + + return WRC_Continue; +} + + +/* +** This function is called on every node of an expression tree used as an +** argument to the OP_CursorHint instruction. If the node is a TK_COLUMN +** that accesses any table other than the one identified by +** CCurHint.iTabCur, then do the following: +** +** 1) allocate a register and code an OP_Column instruction to read +** the specified column into the new register, and +** +** 2) transform the expression node to a TK_REGISTER node that reads +** from the newly populated register. +** +** Also, if the node is a TK_COLUMN that does access the table idenified +** by pCCurHint.iTabCur, and an index is being used (which we will +** know because CCurHint.pIdx!=0) then transform the TK_COLUMN into +** an access of the index rather than the original table. +*/ +static int codeCursorHintFixExpr(Walker *pWalker, Expr *pExpr){ + int rc = WRC_Continue; + struct CCurHint *pHint = pWalker->u.pCCurHint; + if( pExpr->op==TK_COLUMN ){ + if( pExpr->iTable!=pHint->iTabCur ){ + int reg = ++pWalker->pParse->nMem; /* Register for column value */ + sqlite3ExprCode(pWalker->pParse, pExpr, reg); + pExpr->op = TK_REGISTER; + pExpr->iTable = reg; + }else if( pHint->pIdx!=0 ){ + pExpr->iTable = pHint->iIdxCur; + pExpr->iColumn = sqlite3ColumnOfIndex(pHint->pIdx, pExpr->iColumn); + assert( pExpr->iColumn>=0 ); + } + }else if( pExpr->op==TK_AGG_FUNCTION ){ + /* An aggregate function in the WHERE clause of a query means this must + ** be a correlated sub-query, and expression pExpr is an aggregate from + ** the parent context. Do not walk the function arguments in this case. + ** + ** todo: It should be possible to replace this node with a TK_REGISTER + ** expression, as the result of the expression must be stored in a + ** register at this point. The same holds for TK_AGG_COLUMN nodes. */ + rc = WRC_Prune; + } + return rc; +} + +/* +** Insert an OP_CursorHint instruction if it is appropriate to do so. +*/ +static void codeCursorHint( + struct SrcList_item *pTabItem, /* FROM clause item */ + WhereInfo *pWInfo, /* The where clause */ + WhereLevel *pLevel, /* Which loop to provide hints for */ + WhereTerm *pEndRange /* Hint this end-of-scan boundary term if not NULL */ +){ + Parse *pParse = pWInfo->pParse; + sqlite3 *db = pParse->db; + Vdbe *v = pParse->pVdbe; + Expr *pExpr = 0; + WhereLoop *pLoop = pLevel->pWLoop; + int iCur; + WhereClause *pWC; + WhereTerm *pTerm; + int i, j; + struct CCurHint sHint; + Walker sWalker; + + if( OptimizationDisabled(db, SQLITE_CursorHints) ) return; + iCur = pLevel->iTabCur; + assert( iCur==pWInfo->pTabList->a[pLevel->iFrom].iCursor ); + sHint.iTabCur = iCur; + sHint.iIdxCur = pLevel->iIdxCur; + sHint.pIdx = pLoop->u.btree.pIndex; + memset(&sWalker, 0, sizeof(sWalker)); + sWalker.pParse = pParse; + sWalker.u.pCCurHint = &sHint; + pWC = &pWInfo->sWC; + for(i=0; inTerm; i++){ + pTerm = &pWC->a[i]; + if( pTerm->wtFlags & (TERM_VIRTUAL|TERM_CODED) ) continue; + if( pTerm->prereqAll & pLevel->notReady ) continue; + + /* Any terms specified as part of the ON(...) clause for any LEFT + ** JOIN for which the current table is not the rhs are omitted + ** from the cursor-hint. + ** + ** If this table is the rhs of a LEFT JOIN, "IS" or "IS NULL" terms + ** that were specified as part of the WHERE clause must be excluded. + ** This is to address the following: + ** + ** SELECT ... t1 LEFT JOIN t2 ON (t1.a=t2.b) WHERE t2.c IS NULL; + ** + ** Say there is a single row in t2 that matches (t1.a=t2.b), but its + ** t2.c values is not NULL. If the (t2.c IS NULL) constraint is + ** pushed down to the cursor, this row is filtered out, causing + ** SQLite to synthesize a row of NULL values. Which does match the + ** WHERE clause, and so the query returns a row. Which is incorrect. + ** + ** For the same reason, WHERE terms such as: + ** + ** WHERE 1 = (t2.c IS NULL) + ** + ** are also excluded. See codeCursorHintIsOrFunction() for details. + */ + if( pTabItem->fg.jointype & JT_LEFT ){ + Expr *pExpr = pTerm->pExpr; + if( !ExprHasProperty(pExpr, EP_FromJoin) + || pExpr->iRightJoinTable!=pTabItem->iCursor + ){ + sWalker.eCode = 0; + sWalker.xExprCallback = codeCursorHintIsOrFunction; + sqlite3WalkExpr(&sWalker, pTerm->pExpr); + if( sWalker.eCode ) continue; + } + }else{ + if( ExprHasProperty(pTerm->pExpr, EP_FromJoin) ) continue; + } + + /* All terms in pWLoop->aLTerm[] except pEndRange are used to initialize + ** the cursor. These terms are not needed as hints for a pure range + ** scan (that has no == terms) so omit them. */ + if( pLoop->u.btree.nEq==0 && pTerm!=pEndRange ){ + for(j=0; jnLTerm && pLoop->aLTerm[j]!=pTerm; j++){} + if( jnLTerm ) continue; + } + + /* No subqueries or non-deterministic functions allowed */ + if( sqlite3ExprContainsSubquery(pTerm->pExpr) ) continue; + + /* For an index scan, make sure referenced columns are actually in + ** the index. */ + if( sHint.pIdx!=0 ){ + sWalker.eCode = 0; + sWalker.xExprCallback = codeCursorHintCheckExpr; + sqlite3WalkExpr(&sWalker, pTerm->pExpr); + if( sWalker.eCode ) continue; + } + + /* If we survive all prior tests, that means this term is worth hinting */ + pExpr = sqlite3ExprAnd(pParse, pExpr, sqlite3ExprDup(db, pTerm->pExpr, 0)); + } + if( pExpr!=0 ){ + sWalker.xExprCallback = codeCursorHintFixExpr; + sqlite3WalkExpr(&sWalker, pExpr); + sqlite3VdbeAddOp4(v, OP_CursorHint, + (sHint.pIdx ? sHint.iIdxCur : sHint.iTabCur), 0, 0, + (const char*)pExpr, P4_EXPR); + } +} +#else +# define codeCursorHint(A,B,C,D) /* No-op */ +#endif /* SQLITE_ENABLE_CURSOR_HINTS */ + +/* +** Cursor iCur is open on an intkey b-tree (a table). Register iRowid contains +** a rowid value just read from cursor iIdxCur, open on index pIdx. This +** function generates code to do a deferred seek of cursor iCur to the +** rowid stored in register iRowid. +** +** Normally, this is just: +** +** OP_DeferredSeek $iCur $iRowid +** +** However, if the scan currently being coded is a branch of an OR-loop and +** the statement currently being coded is a SELECT, then P3 of OP_DeferredSeek +** is set to iIdxCur and P4 is set to point to an array of integers +** containing one entry for each column of the table cursor iCur is open +** on. For each table column, if the column is the i'th column of the +** index, then the corresponding array entry is set to (i+1). If the column +** does not appear in the index at all, the array entry is set to 0. +*/ +static void codeDeferredSeek( + WhereInfo *pWInfo, /* Where clause context */ + Index *pIdx, /* Index scan is using */ + int iCur, /* Cursor for IPK b-tree */ + int iIdxCur /* Index cursor */ +){ + Parse *pParse = pWInfo->pParse; /* Parse context */ + Vdbe *v = pParse->pVdbe; /* Vdbe to generate code within */ + + assert( iIdxCur>0 ); + assert( pIdx->aiColumn[pIdx->nColumn-1]==-1 ); + + sqlite3VdbeAddOp3(v, OP_DeferredSeek, iIdxCur, 0, iCur); + if( (pWInfo->wctrlFlags & WHERE_OR_SUBCLAUSE) + && DbMaskAllZero(sqlite3ParseToplevel(pParse)->writeMask) + ){ + int i; + Table *pTab = pIdx->pTable; + int *ai = (int*)sqlite3DbMallocZero(pParse->db, sizeof(int)*(pTab->nCol+1)); + if( ai ){ + ai[0] = pTab->nCol; + for(i=0; inColumn-1; i++){ + assert( pIdx->aiColumn[i]nCol ); + if( pIdx->aiColumn[i]>=0 ) ai[pIdx->aiColumn[i]+1] = i+1; + } + sqlite3VdbeChangeP4(v, -1, (char*)ai, P4_INTARRAY); + } + } +} + +/* +** If the expression passed as the second argument is a vector, generate +** code to write the first nReg elements of the vector into an array +** of registers starting with iReg. +** +** If the expression is not a vector, then nReg must be passed 1. In +** this case, generate code to evaluate the expression and leave the +** result in register iReg. +*/ +static void codeExprOrVector(Parse *pParse, Expr *p, int iReg, int nReg){ + assert( nReg>0 ); + if( p && sqlite3ExprIsVector(p) ){ +#ifndef SQLITE_OMIT_SUBQUERY + if( (p->flags & EP_xIsSelect) ){ + Vdbe *v = pParse->pVdbe; + int iSelect; + assert( p->op==TK_SELECT ); + iSelect = sqlite3CodeSubselect(pParse, p); + sqlite3VdbeAddOp3(v, OP_Copy, iSelect, iReg, nReg-1); + }else +#endif + { + int i; + ExprList *pList = p->x.pList; + assert( nReg<=pList->nExpr ); + for(i=0; ia[i].pExpr, iReg+i); + } + } + }else{ + assert( nReg==1 ); + sqlite3ExprCode(pParse, p, iReg); + } +} + +/* An instance of the IdxExprTrans object carries information about a +** mapping from an expression on table columns into a column in an index +** down through the Walker. +*/ +typedef struct IdxExprTrans { + Expr *pIdxExpr; /* The index expression */ + int iTabCur; /* The cursor of the corresponding table */ + int iIdxCur; /* The cursor for the index */ + int iIdxCol; /* The column for the index */ +} IdxExprTrans; + +/* The walker node callback used to transform matching expressions into +** a reference to an index column for an index on an expression. +** +** If pExpr matches, then transform it into a reference to the index column +** that contains the value of pExpr. +*/ +static int whereIndexExprTransNode(Walker *p, Expr *pExpr){ + IdxExprTrans *pX = p->u.pIdxTrans; + if( sqlite3ExprCompare(0, pExpr, pX->pIdxExpr, pX->iTabCur)==0 ){ + pExpr->affExpr = sqlite3ExprAffinity(pExpr); + pExpr->op = TK_COLUMN; + pExpr->iTable = pX->iIdxCur; + pExpr->iColumn = pX->iIdxCol; + pExpr->y.pTab = 0; + return WRC_Prune; + }else{ + return WRC_Continue; + } +} + +/* +** For an indexes on expression X, locate every instance of expression X +** in pExpr and change that subexpression into a reference to the appropriate +** column of the index. +*/ +static void whereIndexExprTrans( + Index *pIdx, /* The Index */ + int iTabCur, /* Cursor of the table that is being indexed */ + int iIdxCur, /* Cursor of the index itself */ + WhereInfo *pWInfo /* Transform expressions in this WHERE clause */ +){ + int iIdxCol; /* Column number of the index */ + ExprList *aColExpr; /* Expressions that are indexed */ + Walker w; + IdxExprTrans x; + aColExpr = pIdx->aColExpr; + if( aColExpr==0 ) return; /* Not an index on expressions */ + memset(&w, 0, sizeof(w)); + w.xExprCallback = whereIndexExprTransNode; + w.u.pIdxTrans = &x; + x.iTabCur = iTabCur; + x.iIdxCur = iIdxCur; + for(iIdxCol=0; iIdxColnExpr; iIdxCol++){ + if( pIdx->aiColumn[iIdxCol]!=XN_EXPR ) continue; + assert( aColExpr->a[iIdxCol].pExpr!=0 ); + x.iIdxCol = iIdxCol; + x.pIdxExpr = aColExpr->a[iIdxCol].pExpr; + sqlite3WalkExpr(&w, pWInfo->pWhere); + sqlite3WalkExprList(&w, pWInfo->pOrderBy); + sqlite3WalkExprList(&w, pWInfo->pResultSet); + } +} + +/* +** The pTruth expression is always true because it is the WHERE clause +** a partial index that is driving a query loop. Look through all of the +** WHERE clause terms on the query, and if any of those terms must be +** true because pTruth is true, then mark those WHERE clause terms as +** coded. +*/ +static void whereApplyPartialIndexConstraints( + Expr *pTruth, + int iTabCur, + WhereClause *pWC +){ + int i; + WhereTerm *pTerm; + while( pTruth->op==TK_AND ){ + whereApplyPartialIndexConstraints(pTruth->pLeft, iTabCur, pWC); + pTruth = pTruth->pRight; + } + for(i=0, pTerm=pWC->a; inTerm; i++, pTerm++){ + Expr *pExpr; + if( pTerm->wtFlags & TERM_CODED ) continue; + pExpr = pTerm->pExpr; + if( sqlite3ExprCompare(0, pExpr, pTruth, iTabCur)==0 ){ + pTerm->wtFlags |= TERM_CODED; + } + } +} + +/* +** Generate code for the start of the iLevel-th loop in the WHERE clause +** implementation described by pWInfo. +*/ +SQLITE_PRIVATE Bitmask sqlite3WhereCodeOneLoopStart( + Parse *pParse, /* Parsing context */ + Vdbe *v, /* Prepared statement under construction */ + WhereInfo *pWInfo, /* Complete information about the WHERE clause */ + int iLevel, /* Which level of pWInfo->a[] should be coded */ + WhereLevel *pLevel, /* The current level pointer */ + Bitmask notReady /* Which tables are currently available */ +){ + int j, k; /* Loop counters */ + int iCur; /* The VDBE cursor for the table */ + int addrNxt; /* Where to jump to continue with the next IN case */ + int bRev; /* True if we need to scan in reverse order */ + WhereLoop *pLoop; /* The WhereLoop object being coded */ + WhereClause *pWC; /* Decomposition of the entire WHERE clause */ + WhereTerm *pTerm; /* A WHERE clause term */ + sqlite3 *db; /* Database connection */ + struct SrcList_item *pTabItem; /* FROM clause term being coded */ + int addrBrk; /* Jump here to break out of the loop */ + int addrHalt; /* addrBrk for the outermost loop */ + int addrCont; /* Jump here to continue with next cycle */ + int iRowidReg = 0; /* Rowid is stored in this register, if not zero */ + int iReleaseReg = 0; /* Temp register to free before returning */ + Index *pIdx = 0; /* Index used by loop (if any) */ + int iLoop; /* Iteration of constraint generator loop */ + + pWC = &pWInfo->sWC; + db = pParse->db; + pLoop = pLevel->pWLoop; + pTabItem = &pWInfo->pTabList->a[pLevel->iFrom]; + iCur = pTabItem->iCursor; + pLevel->notReady = notReady & ~sqlite3WhereGetMask(&pWInfo->sMaskSet, iCur); + bRev = (pWInfo->revMask>>iLevel)&1; + VdbeModuleComment((v, "Begin WHERE-loop%d: %s",iLevel,pTabItem->pTab->zName)); + + /* Create labels for the "break" and "continue" instructions + ** for the current loop. Jump to addrBrk to break out of a loop. + ** Jump to cont to go immediately to the next iteration of the + ** loop. + ** + ** When there is an IN operator, we also have a "addrNxt" label that + ** means to continue with the next IN value combination. When + ** there are no IN operators in the constraints, the "addrNxt" label + ** is the same as "addrBrk". + */ + addrBrk = pLevel->addrBrk = pLevel->addrNxt = sqlite3VdbeMakeLabel(pParse); + addrCont = pLevel->addrCont = sqlite3VdbeMakeLabel(pParse); + + /* If this is the right table of a LEFT OUTER JOIN, allocate and + ** initialize a memory cell that records if this table matches any + ** row of the left table of the join. + */ + assert( (pWInfo->wctrlFlags & WHERE_OR_SUBCLAUSE) + || pLevel->iFrom>0 || (pTabItem[0].fg.jointype & JT_LEFT)==0 + ); + if( pLevel->iFrom>0 && (pTabItem[0].fg.jointype & JT_LEFT)!=0 ){ + pLevel->iLeftJoin = ++pParse->nMem; + sqlite3VdbeAddOp2(v, OP_Integer, 0, pLevel->iLeftJoin); + VdbeComment((v, "init LEFT JOIN no-match flag")); + } + + /* Compute a safe address to jump to if we discover that the table for + ** this loop is empty and can never contribute content. */ + for(j=iLevel; j>0 && pWInfo->a[j].iLeftJoin==0; j--){} + addrHalt = pWInfo->a[j].addrBrk; + + /* Special case of a FROM clause subquery implemented as a co-routine */ + if( pTabItem->fg.viaCoroutine ){ + int regYield = pTabItem->regReturn; + sqlite3VdbeAddOp3(v, OP_InitCoroutine, regYield, 0, pTabItem->addrFillSub); + pLevel->p2 = sqlite3VdbeAddOp2(v, OP_Yield, regYield, addrBrk); + VdbeCoverage(v); + VdbeComment((v, "next row of %s", pTabItem->pTab->zName)); + pLevel->op = OP_Goto; + }else + +#ifndef SQLITE_OMIT_VIRTUALTABLE + if( (pLoop->wsFlags & WHERE_VIRTUALTABLE)!=0 ){ + /* Case 1: The table is a virtual-table. Use the VFilter and VNext + ** to access the data. + */ + int iReg; /* P3 Value for OP_VFilter */ + int addrNotFound; + int nConstraint = pLoop->nLTerm; + int iIn; /* Counter for IN constraints */ + + iReg = sqlite3GetTempRange(pParse, nConstraint+2); + addrNotFound = pLevel->addrBrk; + for(j=0; jaLTerm[j]; + if( NEVER(pTerm==0) ) continue; + if( pTerm->eOperator & WO_IN ){ + codeEqualityTerm(pParse, pTerm, pLevel, j, bRev, iTarget); + addrNotFound = pLevel->addrNxt; + }else{ + Expr *pRight = pTerm->pExpr->pRight; + codeExprOrVector(pParse, pRight, iTarget, 1); + } + } + sqlite3VdbeAddOp2(v, OP_Integer, pLoop->u.vtab.idxNum, iReg); + sqlite3VdbeAddOp2(v, OP_Integer, nConstraint, iReg+1); + sqlite3VdbeAddOp4(v, OP_VFilter, iCur, addrNotFound, iReg, + pLoop->u.vtab.idxStr, + pLoop->u.vtab.needFree ? P4_DYNAMIC : P4_STATIC); + VdbeCoverage(v); + pLoop->u.vtab.needFree = 0; + pLevel->p1 = iCur; + pLevel->op = pWInfo->eOnePass ? OP_Noop : OP_VNext; + pLevel->p2 = sqlite3VdbeCurrentAddr(v); + iIn = pLevel->u.in.nIn; + for(j=nConstraint-1; j>=0; j--){ + pTerm = pLoop->aLTerm[j]; + if( j<16 && (pLoop->u.vtab.omitMask>>j)&1 ){ + disableTerm(pLevel, pTerm); + }else if( (pTerm->eOperator & WO_IN)!=0 ){ + Expr *pCompare; /* The comparison operator */ + Expr *pRight; /* RHS of the comparison */ + VdbeOp *pOp; /* Opcode to access the value of the IN constraint */ + + /* Reload the constraint value into reg[iReg+j+2]. The same value + ** was loaded into the same register prior to the OP_VFilter, but + ** the xFilter implementation might have changed the datatype or + ** encoding of the value in the register, so it *must* be reloaded. */ + assert( pLevel->u.in.aInLoop!=0 || db->mallocFailed ); + if( !db->mallocFailed ){ + assert( iIn>0 ); + pOp = sqlite3VdbeGetOp(v, pLevel->u.in.aInLoop[--iIn].addrInTop); + assert( pOp->opcode==OP_Column || pOp->opcode==OP_Rowid ); + assert( pOp->opcode!=OP_Column || pOp->p3==iReg+j+2 ); + assert( pOp->opcode!=OP_Rowid || pOp->p2==iReg+j+2 ); + testcase( pOp->opcode==OP_Rowid ); + sqlite3VdbeAddOp3(v, pOp->opcode, pOp->p1, pOp->p2, pOp->p3); + } + + /* Generate code that will continue to the next row if + ** the IN constraint is not satisfied */ + pCompare = sqlite3PExpr(pParse, TK_EQ, 0, 0); + assert( pCompare!=0 || db->mallocFailed ); + if( pCompare ){ + pCompare->pLeft = pTerm->pExpr->pLeft; + pCompare->pRight = pRight = sqlite3Expr(db, TK_REGISTER, 0); + if( pRight ){ + pRight->iTable = iReg+j+2; + sqlite3ExprIfFalse(pParse, pCompare, pLevel->addrCont, 0); + } + pCompare->pLeft = 0; + sqlite3ExprDelete(db, pCompare); + } + } + } + /* These registers need to be preserved in case there is an IN operator + ** loop. So we could deallocate the registers here (and potentially + ** reuse them later) if (pLoop->wsFlags & WHERE_IN_ABLE)==0. But it seems + ** simpler and safer to simply not reuse the registers. + ** + ** sqlite3ReleaseTempRange(pParse, iReg, nConstraint+2); + */ + }else +#endif /* SQLITE_OMIT_VIRTUALTABLE */ + + if( (pLoop->wsFlags & WHERE_IPK)!=0 + && (pLoop->wsFlags & (WHERE_COLUMN_IN|WHERE_COLUMN_EQ))!=0 + ){ + /* Case 2: We can directly reference a single row using an + ** equality comparison against the ROWID field. Or + ** we reference multiple rows using a "rowid IN (...)" + ** construct. + */ + assert( pLoop->u.btree.nEq==1 ); + pTerm = pLoop->aLTerm[0]; + assert( pTerm!=0 ); + assert( pTerm->pExpr!=0 ); + testcase( pTerm->wtFlags & TERM_VIRTUAL ); + iReleaseReg = ++pParse->nMem; + iRowidReg = codeEqualityTerm(pParse, pTerm, pLevel, 0, bRev, iReleaseReg); + if( iRowidReg!=iReleaseReg ) sqlite3ReleaseTempReg(pParse, iReleaseReg); + addrNxt = pLevel->addrNxt; + sqlite3VdbeAddOp3(v, OP_SeekRowid, iCur, addrNxt, iRowidReg); + VdbeCoverage(v); + pLevel->op = OP_Noop; + if( (pTerm->prereqAll & pLevel->notReady)==0 ){ + pTerm->wtFlags |= TERM_CODED; + } + }else if( (pLoop->wsFlags & WHERE_IPK)!=0 + && (pLoop->wsFlags & WHERE_COLUMN_RANGE)!=0 + ){ + /* Case 3: We have an inequality comparison against the ROWID field. + */ + int testOp = OP_Noop; + int start; + int memEndValue = 0; + WhereTerm *pStart, *pEnd; + + j = 0; + pStart = pEnd = 0; + if( pLoop->wsFlags & WHERE_BTM_LIMIT ) pStart = pLoop->aLTerm[j++]; + if( pLoop->wsFlags & WHERE_TOP_LIMIT ) pEnd = pLoop->aLTerm[j++]; + assert( pStart!=0 || pEnd!=0 ); + if( bRev ){ + pTerm = pStart; + pStart = pEnd; + pEnd = pTerm; + } + codeCursorHint(pTabItem, pWInfo, pLevel, pEnd); + if( pStart ){ + Expr *pX; /* The expression that defines the start bound */ + int r1, rTemp; /* Registers for holding the start boundary */ + int op; /* Cursor seek operation */ + + /* The following constant maps TK_xx codes into corresponding + ** seek opcodes. It depends on a particular ordering of TK_xx + */ + const u8 aMoveOp[] = { + /* TK_GT */ OP_SeekGT, + /* TK_LE */ OP_SeekLE, + /* TK_LT */ OP_SeekLT, + /* TK_GE */ OP_SeekGE + }; + assert( TK_LE==TK_GT+1 ); /* Make sure the ordering.. */ + assert( TK_LT==TK_GT+2 ); /* ... of the TK_xx values... */ + assert( TK_GE==TK_GT+3 ); /* ... is correcct. */ + + assert( (pStart->wtFlags & TERM_VNULL)==0 ); + testcase( pStart->wtFlags & TERM_VIRTUAL ); + pX = pStart->pExpr; + assert( pX!=0 ); + testcase( pStart->leftCursor!=iCur ); /* transitive constraints */ + if( sqlite3ExprIsVector(pX->pRight) ){ + r1 = rTemp = sqlite3GetTempReg(pParse); + codeExprOrVector(pParse, pX->pRight, r1, 1); + testcase( pX->op==TK_GT ); + testcase( pX->op==TK_GE ); + testcase( pX->op==TK_LT ); + testcase( pX->op==TK_LE ); + op = aMoveOp[((pX->op - TK_GT - 1) & 0x3) | 0x1]; + assert( pX->op!=TK_GT || op==OP_SeekGE ); + assert( pX->op!=TK_GE || op==OP_SeekGE ); + assert( pX->op!=TK_LT || op==OP_SeekLE ); + assert( pX->op!=TK_LE || op==OP_SeekLE ); + }else{ + r1 = sqlite3ExprCodeTemp(pParse, pX->pRight, &rTemp); + disableTerm(pLevel, pStart); + op = aMoveOp[(pX->op - TK_GT)]; + } + sqlite3VdbeAddOp3(v, op, iCur, addrBrk, r1); + VdbeComment((v, "pk")); + VdbeCoverageIf(v, pX->op==TK_GT); + VdbeCoverageIf(v, pX->op==TK_LE); + VdbeCoverageIf(v, pX->op==TK_LT); + VdbeCoverageIf(v, pX->op==TK_GE); + sqlite3ReleaseTempReg(pParse, rTemp); + }else{ + sqlite3VdbeAddOp2(v, bRev ? OP_Last : OP_Rewind, iCur, addrHalt); + VdbeCoverageIf(v, bRev==0); + VdbeCoverageIf(v, bRev!=0); + } + if( pEnd ){ + Expr *pX; + pX = pEnd->pExpr; + assert( pX!=0 ); + assert( (pEnd->wtFlags & TERM_VNULL)==0 ); + testcase( pEnd->leftCursor!=iCur ); /* Transitive constraints */ + testcase( pEnd->wtFlags & TERM_VIRTUAL ); + memEndValue = ++pParse->nMem; + codeExprOrVector(pParse, pX->pRight, memEndValue, 1); + if( 0==sqlite3ExprIsVector(pX->pRight) + && (pX->op==TK_LT || pX->op==TK_GT) + ){ + testOp = bRev ? OP_Le : OP_Ge; + }else{ + testOp = bRev ? OP_Lt : OP_Gt; + } + if( 0==sqlite3ExprIsVector(pX->pRight) ){ + disableTerm(pLevel, pEnd); + } + } + start = sqlite3VdbeCurrentAddr(v); + pLevel->op = bRev ? OP_Prev : OP_Next; + pLevel->p1 = iCur; + pLevel->p2 = start; + assert( pLevel->p5==0 ); + if( testOp!=OP_Noop ){ + iRowidReg = ++pParse->nMem; + sqlite3VdbeAddOp2(v, OP_Rowid, iCur, iRowidReg); + sqlite3VdbeAddOp3(v, testOp, memEndValue, addrBrk, iRowidReg); + VdbeCoverageIf(v, testOp==OP_Le); + VdbeCoverageIf(v, testOp==OP_Lt); + VdbeCoverageIf(v, testOp==OP_Ge); + VdbeCoverageIf(v, testOp==OP_Gt); + sqlite3VdbeChangeP5(v, SQLITE_AFF_NUMERIC | SQLITE_JUMPIFNULL); + } + }else if( pLoop->wsFlags & WHERE_INDEXED ){ + /* Case 4: A scan using an index. + ** + ** The WHERE clause may contain zero or more equality + ** terms ("==" or "IN" operators) that refer to the N + ** left-most columns of the index. It may also contain + ** inequality constraints (>, <, >= or <=) on the indexed + ** column that immediately follows the N equalities. Only + ** the right-most column can be an inequality - the rest must + ** use the "==" and "IN" operators. For example, if the + ** index is on (x,y,z), then the following clauses are all + ** optimized: + ** + ** x=5 + ** x=5 AND y=10 + ** x=5 AND y<10 + ** x=5 AND y>5 AND y<10 + ** x=5 AND y=5 AND z<=10 + ** + ** The z<10 term of the following cannot be used, only + ** the x=5 term: + ** + ** x=5 AND z<10 + ** + ** N may be zero if there are inequality constraints. + ** If there are no inequality constraints, then N is at + ** least one. + ** + ** This case is also used when there are no WHERE clause + ** constraints but an index is selected anyway, in order + ** to force the output order to conform to an ORDER BY. + */ + static const u8 aStartOp[] = { + 0, + 0, + OP_Rewind, /* 2: (!start_constraints && startEq && !bRev) */ + OP_Last, /* 3: (!start_constraints && startEq && bRev) */ + OP_SeekGT, /* 4: (start_constraints && !startEq && !bRev) */ + OP_SeekLT, /* 5: (start_constraints && !startEq && bRev) */ + OP_SeekGE, /* 6: (start_constraints && startEq && !bRev) */ + OP_SeekLE /* 7: (start_constraints && startEq && bRev) */ + }; + static const u8 aEndOp[] = { + OP_IdxGE, /* 0: (end_constraints && !bRev && !endEq) */ + OP_IdxGT, /* 1: (end_constraints && !bRev && endEq) */ + OP_IdxLE, /* 2: (end_constraints && bRev && !endEq) */ + OP_IdxLT, /* 3: (end_constraints && bRev && endEq) */ + }; + u16 nEq = pLoop->u.btree.nEq; /* Number of == or IN terms */ + u16 nBtm = pLoop->u.btree.nBtm; /* Length of BTM vector */ + u16 nTop = pLoop->u.btree.nTop; /* Length of TOP vector */ + int regBase; /* Base register holding constraint values */ + WhereTerm *pRangeStart = 0; /* Inequality constraint at range start */ + WhereTerm *pRangeEnd = 0; /* Inequality constraint at range end */ + int startEq; /* True if range start uses ==, >= or <= */ + int endEq; /* True if range end uses ==, >= or <= */ + int start_constraints; /* Start of range is constrained */ + int nConstraint; /* Number of constraint terms */ + int iIdxCur; /* The VDBE cursor for the index */ + int nExtraReg = 0; /* Number of extra registers needed */ + int op; /* Instruction opcode */ + char *zStartAff; /* Affinity for start of range constraint */ + char *zEndAff = 0; /* Affinity for end of range constraint */ + u8 bSeekPastNull = 0; /* True to seek past initial nulls */ + u8 bStopAtNull = 0; /* Add condition to terminate at NULLs */ + int omitTable; /* True if we use the index only */ + int regBignull = 0; /* big-null flag register */ + + pIdx = pLoop->u.btree.pIndex; + iIdxCur = pLevel->iIdxCur; + assert( nEq>=pLoop->nSkip ); + + /* Find any inequality constraint terms for the start and end + ** of the range. + */ + j = nEq; + if( pLoop->wsFlags & WHERE_BTM_LIMIT ){ + pRangeStart = pLoop->aLTerm[j++]; + nExtraReg = MAX(nExtraReg, pLoop->u.btree.nBtm); + /* Like optimization range constraints always occur in pairs */ + assert( (pRangeStart->wtFlags & TERM_LIKEOPT)==0 || + (pLoop->wsFlags & WHERE_TOP_LIMIT)!=0 ); + } + if( pLoop->wsFlags & WHERE_TOP_LIMIT ){ + pRangeEnd = pLoop->aLTerm[j++]; + nExtraReg = MAX(nExtraReg, pLoop->u.btree.nTop); +#ifndef SQLITE_LIKE_DOESNT_MATCH_BLOBS + if( (pRangeEnd->wtFlags & TERM_LIKEOPT)!=0 ){ + assert( pRangeStart!=0 ); /* LIKE opt constraints */ + assert( pRangeStart->wtFlags & TERM_LIKEOPT ); /* occur in pairs */ + pLevel->iLikeRepCntr = (u32)++pParse->nMem; + sqlite3VdbeAddOp2(v, OP_Integer, 1, (int)pLevel->iLikeRepCntr); + VdbeComment((v, "LIKE loop counter")); + pLevel->addrLikeRep = sqlite3VdbeCurrentAddr(v); + /* iLikeRepCntr actually stores 2x the counter register number. The + ** bottom bit indicates whether the search order is ASC or DESC. */ + testcase( bRev ); + testcase( pIdx->aSortOrder[nEq]==SQLITE_SO_DESC ); + assert( (bRev & ~1)==0 ); + pLevel->iLikeRepCntr <<=1; + pLevel->iLikeRepCntr |= bRev ^ (pIdx->aSortOrder[nEq]==SQLITE_SO_DESC); + } +#endif + if( pRangeStart==0 ){ + j = pIdx->aiColumn[nEq]; + if( (j>=0 && pIdx->pTable->aCol[j].notNull==0) || j==XN_EXPR ){ + bSeekPastNull = 1; + } + } + } + assert( pRangeEnd==0 || (pRangeEnd->wtFlags & TERM_VNULL)==0 ); + + /* If the WHERE_BIGNULL_SORT flag is set, then index column nEq uses + ** a non-default "big-null" sort (either ASC NULLS LAST or DESC NULLS + ** FIRST). In both cases separate ordered scans are made of those + ** index entries for which the column is null and for those for which + ** it is not. For an ASC sort, the non-NULL entries are scanned first. + ** For DESC, NULL entries are scanned first. + */ + if( (pLoop->wsFlags & (WHERE_TOP_LIMIT|WHERE_BTM_LIMIT))==0 + && (pLoop->wsFlags & WHERE_BIGNULL_SORT)!=0 + ){ + assert( bSeekPastNull==0 && nExtraReg==0 && nBtm==0 && nTop==0 ); + assert( pRangeEnd==0 && pRangeStart==0 ); + assert( pLoop->nSkip==0 ); + nExtraReg = 1; + bSeekPastNull = 1; + pLevel->regBignull = regBignull = ++pParse->nMem; + pLevel->addrBignull = sqlite3VdbeMakeLabel(pParse); + } + + /* If we are doing a reverse order scan on an ascending index, or + ** a forward order scan on a descending index, interchange the + ** start and end terms (pRangeStart and pRangeEnd). + */ + if( (nEqnKeyCol && bRev==(pIdx->aSortOrder[nEq]==SQLITE_SO_ASC)) + || (bRev && pIdx->nKeyCol==nEq) + ){ + SWAP(WhereTerm *, pRangeEnd, pRangeStart); + SWAP(u8, bSeekPastNull, bStopAtNull); + SWAP(u8, nBtm, nTop); + } + + /* Generate code to evaluate all constraint terms using == or IN + ** and store the values of those terms in an array of registers + ** starting at regBase. + */ + codeCursorHint(pTabItem, pWInfo, pLevel, pRangeEnd); + regBase = codeAllEqualityTerms(pParse,pLevel,bRev,nExtraReg,&zStartAff); + assert( zStartAff==0 || sqlite3Strlen30(zStartAff)>=nEq ); + if( zStartAff && nTop ){ + zEndAff = sqlite3DbStrDup(db, &zStartAff[nEq]); + } + addrNxt = (regBignull ? pLevel->addrBignull : pLevel->addrNxt); + + testcase( pRangeStart && (pRangeStart->eOperator & WO_LE)!=0 ); + testcase( pRangeStart && (pRangeStart->eOperator & WO_GE)!=0 ); + testcase( pRangeEnd && (pRangeEnd->eOperator & WO_LE)!=0 ); + testcase( pRangeEnd && (pRangeEnd->eOperator & WO_GE)!=0 ); + startEq = !pRangeStart || pRangeStart->eOperator & (WO_LE|WO_GE); + endEq = !pRangeEnd || pRangeEnd->eOperator & (WO_LE|WO_GE); + start_constraints = pRangeStart || nEq>0; + + /* Seek the index cursor to the start of the range. */ + nConstraint = nEq; + if( pRangeStart ){ + Expr *pRight = pRangeStart->pExpr->pRight; + codeExprOrVector(pParse, pRight, regBase+nEq, nBtm); + whereLikeOptimizationStringFixup(v, pLevel, pRangeStart); + if( (pRangeStart->wtFlags & TERM_VNULL)==0 + && sqlite3ExprCanBeNull(pRight) + ){ + sqlite3VdbeAddOp2(v, OP_IsNull, regBase+nEq, addrNxt); + VdbeCoverage(v); + } + if( zStartAff ){ + updateRangeAffinityStr(pRight, nBtm, &zStartAff[nEq]); + } + nConstraint += nBtm; + testcase( pRangeStart->wtFlags & TERM_VIRTUAL ); + if( sqlite3ExprIsVector(pRight)==0 ){ + disableTerm(pLevel, pRangeStart); + }else{ + startEq = 1; + } + bSeekPastNull = 0; + }else if( bSeekPastNull ){ + startEq = 0; + sqlite3VdbeAddOp2(v, OP_Null, 0, regBase+nEq); + start_constraints = 1; + nConstraint++; + }else if( regBignull ){ + sqlite3VdbeAddOp2(v, OP_Null, 0, regBase+nEq); + start_constraints = 1; + nConstraint++; + } + codeApplyAffinity(pParse, regBase, nConstraint - bSeekPastNull, zStartAff); + if( pLoop->nSkip>0 && nConstraint==pLoop->nSkip ){ + /* The skip-scan logic inside the call to codeAllEqualityConstraints() + ** above has already left the cursor sitting on the correct row, + ** so no further seeking is needed */ + }else{ + if( pLoop->wsFlags & WHERE_IN_EARLYOUT ){ + sqlite3VdbeAddOp1(v, OP_SeekHit, iIdxCur); + } + if( regBignull ){ + sqlite3VdbeAddOp2(v, OP_Integer, 1, regBignull); + VdbeComment((v, "NULL-scan pass ctr")); + } + + op = aStartOp[(start_constraints<<2) + (startEq<<1) + bRev]; + assert( op!=0 ); + sqlite3VdbeAddOp4Int(v, op, iIdxCur, addrNxt, regBase, nConstraint); + VdbeCoverage(v); + VdbeCoverageIf(v, op==OP_Rewind); testcase( op==OP_Rewind ); + VdbeCoverageIf(v, op==OP_Last); testcase( op==OP_Last ); + VdbeCoverageIf(v, op==OP_SeekGT); testcase( op==OP_SeekGT ); + VdbeCoverageIf(v, op==OP_SeekGE); testcase( op==OP_SeekGE ); + VdbeCoverageIf(v, op==OP_SeekLE); testcase( op==OP_SeekLE ); + VdbeCoverageIf(v, op==OP_SeekLT); testcase( op==OP_SeekLT ); + + assert( bSeekPastNull==0 || bStopAtNull==0 ); + if( regBignull ){ + assert( bSeekPastNull==1 || bStopAtNull==1 ); + assert( bSeekPastNull==!bStopAtNull ); + assert( bStopAtNull==startEq ); + sqlite3VdbeAddOp2(v, OP_Goto, 0, sqlite3VdbeCurrentAddr(v)+2); + op = aStartOp[(nConstraint>1)*4 + 2 + bRev]; + sqlite3VdbeAddOp4Int(v, op, iIdxCur, addrNxt, regBase, + nConstraint-startEq); + VdbeCoverage(v); + VdbeCoverageIf(v, op==OP_Rewind); testcase( op==OP_Rewind ); + VdbeCoverageIf(v, op==OP_Last); testcase( op==OP_Last ); + VdbeCoverageIf(v, op==OP_SeekGE); testcase( op==OP_SeekGE ); + VdbeCoverageIf(v, op==OP_SeekLE); testcase( op==OP_SeekLE ); + assert( op==OP_Rewind || op==OP_Last || op==OP_SeekGE || op==OP_SeekLE); + } + } + + /* Load the value for the inequality constraint at the end of the + ** range (if any). + */ + nConstraint = nEq; + if( pRangeEnd ){ + Expr *pRight = pRangeEnd->pExpr->pRight; + codeExprOrVector(pParse, pRight, regBase+nEq, nTop); + whereLikeOptimizationStringFixup(v, pLevel, pRangeEnd); + if( (pRangeEnd->wtFlags & TERM_VNULL)==0 + && sqlite3ExprCanBeNull(pRight) + ){ + sqlite3VdbeAddOp2(v, OP_IsNull, regBase+nEq, addrNxt); + VdbeCoverage(v); + } + if( zEndAff ){ + updateRangeAffinityStr(pRight, nTop, zEndAff); + codeApplyAffinity(pParse, regBase+nEq, nTop, zEndAff); + }else{ + assert( pParse->db->mallocFailed ); + } + nConstraint += nTop; + testcase( pRangeEnd->wtFlags & TERM_VIRTUAL ); + + if( sqlite3ExprIsVector(pRight)==0 ){ + disableTerm(pLevel, pRangeEnd); + }else{ + endEq = 1; + } + }else if( bStopAtNull ){ + if( regBignull==0 ){ + sqlite3VdbeAddOp2(v, OP_Null, 0, regBase+nEq); + endEq = 0; + } + nConstraint++; + } + sqlite3DbFree(db, zStartAff); + sqlite3DbFree(db, zEndAff); + + /* Top of the loop body */ + pLevel->p2 = sqlite3VdbeCurrentAddr(v); + + /* Check if the index cursor is past the end of the range. */ + if( nConstraint ){ + if( regBignull ){ + /* Except, skip the end-of-range check while doing the NULL-scan */ + sqlite3VdbeAddOp2(v, OP_IfNot, regBignull, sqlite3VdbeCurrentAddr(v)+3); + VdbeComment((v, "If NULL-scan 2nd pass")); + VdbeCoverage(v); + } + op = aEndOp[bRev*2 + endEq]; + sqlite3VdbeAddOp4Int(v, op, iIdxCur, addrNxt, regBase, nConstraint); + testcase( op==OP_IdxGT ); VdbeCoverageIf(v, op==OP_IdxGT ); + testcase( op==OP_IdxGE ); VdbeCoverageIf(v, op==OP_IdxGE ); + testcase( op==OP_IdxLT ); VdbeCoverageIf(v, op==OP_IdxLT ); + testcase( op==OP_IdxLE ); VdbeCoverageIf(v, op==OP_IdxLE ); + } + if( regBignull ){ + /* During a NULL-scan, check to see if we have reached the end of + ** the NULLs */ + assert( bSeekPastNull==!bStopAtNull ); + assert( bSeekPastNull+bStopAtNull==1 ); + assert( nConstraint+bSeekPastNull>0 ); + sqlite3VdbeAddOp2(v, OP_If, regBignull, sqlite3VdbeCurrentAddr(v)+2); + VdbeComment((v, "If NULL-scan 1st pass")); + VdbeCoverage(v); + op = aEndOp[bRev*2 + bSeekPastNull]; + sqlite3VdbeAddOp4Int(v, op, iIdxCur, addrNxt, regBase, + nConstraint+bSeekPastNull); + testcase( op==OP_IdxGT ); VdbeCoverageIf(v, op==OP_IdxGT ); + testcase( op==OP_IdxGE ); VdbeCoverageIf(v, op==OP_IdxGE ); + testcase( op==OP_IdxLT ); VdbeCoverageIf(v, op==OP_IdxLT ); + testcase( op==OP_IdxLE ); VdbeCoverageIf(v, op==OP_IdxLE ); + } + + if( pLoop->wsFlags & WHERE_IN_EARLYOUT ){ + sqlite3VdbeAddOp2(v, OP_SeekHit, iIdxCur, 1); + } + + /* Seek the table cursor, if required */ + omitTable = (pLoop->wsFlags & WHERE_IDX_ONLY)!=0 + && (pWInfo->wctrlFlags & WHERE_OR_SUBCLAUSE)==0; + if( omitTable ){ + /* pIdx is a covering index. No need to access the main table. */ + }else if( HasRowid(pIdx->pTable) ){ + if( (pWInfo->wctrlFlags & WHERE_SEEK_TABLE) || ( + (pWInfo->wctrlFlags & WHERE_SEEK_UNIQ_TABLE) + && (pWInfo->eOnePass==ONEPASS_SINGLE) + )){ + iRowidReg = ++pParse->nMem; + sqlite3VdbeAddOp2(v, OP_IdxRowid, iIdxCur, iRowidReg); + sqlite3VdbeAddOp3(v, OP_NotExists, iCur, 0, iRowidReg); + VdbeCoverage(v); + }else{ + codeDeferredSeek(pWInfo, pIdx, iCur, iIdxCur); + } + }else if( iCur!=iIdxCur ){ + Index *pPk = sqlite3PrimaryKeyIndex(pIdx->pTable); + iRowidReg = sqlite3GetTempRange(pParse, pPk->nKeyCol); + for(j=0; jnKeyCol; j++){ + k = sqlite3ColumnOfIndex(pIdx, pPk->aiColumn[j]); + sqlite3VdbeAddOp3(v, OP_Column, iIdxCur, k, iRowidReg+j); + } + sqlite3VdbeAddOp4Int(v, OP_NotFound, iCur, addrCont, + iRowidReg, pPk->nKeyCol); VdbeCoverage(v); + } + + /* If pIdx is an index on one or more expressions, then look through + ** all the expressions in pWInfo and try to transform matching expressions + ** into reference to index columns. + ** + ** Do not do this for the RHS of a LEFT JOIN. This is because the + ** expression may be evaluated after OP_NullRow has been executed on + ** the cursor. In this case it is important to do the full evaluation, + ** as the result of the expression may not be NULL, even if all table + ** column values are. https://www.sqlite.org/src/info/7fa8049685b50b5a + ** + ** Also, do not do this when processing one index an a multi-index + ** OR clause, since the transformation will become invalid once we + ** move forward to the next index. + ** https://sqlite.org/src/info/4e8e4857d32d401f + */ + if( pLevel->iLeftJoin==0 && (pWInfo->wctrlFlags & WHERE_OR_SUBCLAUSE)==0 ){ + whereIndexExprTrans(pIdx, iCur, iIdxCur, pWInfo); + } + + /* If a partial index is driving the loop, try to eliminate WHERE clause + ** terms from the query that must be true due to the WHERE clause of + ** the partial index + */ + if( pIdx->pPartIdxWhere ){ + whereApplyPartialIndexConstraints(pIdx->pPartIdxWhere, iCur, pWC); + } + + /* Record the instruction used to terminate the loop. */ + if( pLoop->wsFlags & WHERE_ONEROW ){ + pLevel->op = OP_Noop; + }else if( bRev ){ + pLevel->op = OP_Prev; + }else{ + pLevel->op = OP_Next; + } + pLevel->p1 = iIdxCur; + pLevel->p3 = (pLoop->wsFlags&WHERE_UNQ_WANTED)!=0 ? 1:0; + if( (pLoop->wsFlags & WHERE_CONSTRAINT)==0 ){ + pLevel->p5 = SQLITE_STMTSTATUS_FULLSCAN_STEP; + }else{ + assert( pLevel->p5==0 ); + } + if( omitTable ) pIdx = 0; + }else + +#ifndef SQLITE_OMIT_OR_OPTIMIZATION + if( pLoop->wsFlags & WHERE_MULTI_OR ){ + /* Case 5: Two or more separately indexed terms connected by OR + ** + ** Example: + ** + ** CREATE TABLE t1(a,b,c,d); + ** CREATE INDEX i1 ON t1(a); + ** CREATE INDEX i2 ON t1(b); + ** CREATE INDEX i3 ON t1(c); + ** + ** SELECT * FROM t1 WHERE a=5 OR b=7 OR (c=11 AND d=13) + ** + ** In the example, there are three indexed terms connected by OR. + ** The top of the loop looks like this: + ** + ** Null 1 # Zero the rowset in reg 1 + ** + ** Then, for each indexed term, the following. The arguments to + ** RowSetTest are such that the rowid of the current row is inserted + ** into the RowSet. If it is already present, control skips the + ** Gosub opcode and jumps straight to the code generated by WhereEnd(). + ** + ** sqlite3WhereBegin() + ** RowSetTest # Insert rowid into rowset + ** Gosub 2 A + ** sqlite3WhereEnd() + ** + ** Following the above, code to terminate the loop. Label A, the target + ** of the Gosub above, jumps to the instruction right after the Goto. + ** + ** Null 1 # Zero the rowset in reg 1 + ** Goto B # The loop is finished. + ** + ** A: # Return data, whatever. + ** + ** Return 2 # Jump back to the Gosub + ** + ** B: + ** + ** Added 2014-05-26: If the table is a WITHOUT ROWID table, then + ** use an ephemeral index instead of a RowSet to record the primary + ** keys of the rows we have already seen. + ** + */ + WhereClause *pOrWc; /* The OR-clause broken out into subterms */ + SrcList *pOrTab; /* Shortened table list or OR-clause generation */ + Index *pCov = 0; /* Potential covering index (or NULL) */ + int iCovCur = pParse->nTab++; /* Cursor used for index scans (if any) */ + + int regReturn = ++pParse->nMem; /* Register used with OP_Gosub */ + int regRowset = 0; /* Register for RowSet object */ + int regRowid = 0; /* Register holding rowid */ + int iLoopBody = sqlite3VdbeMakeLabel(pParse);/* Start of loop body */ + int iRetInit; /* Address of regReturn init */ + int untestedTerms = 0; /* Some terms not completely tested */ + int ii; /* Loop counter */ + u16 wctrlFlags; /* Flags for sub-WHERE clause */ + Expr *pAndExpr = 0; /* An ".. AND (...)" expression */ + Table *pTab = pTabItem->pTab; + + pTerm = pLoop->aLTerm[0]; + assert( pTerm!=0 ); + assert( pTerm->eOperator & WO_OR ); + assert( (pTerm->wtFlags & TERM_ORINFO)!=0 ); + pOrWc = &pTerm->u.pOrInfo->wc; + pLevel->op = OP_Return; + pLevel->p1 = regReturn; + + /* Set up a new SrcList in pOrTab containing the table being scanned + ** by this loop in the a[0] slot and all notReady tables in a[1..] slots. + ** This becomes the SrcList in the recursive call to sqlite3WhereBegin(). + */ + if( pWInfo->nLevel>1 ){ + int nNotReady; /* The number of notReady tables */ + struct SrcList_item *origSrc; /* Original list of tables */ + nNotReady = pWInfo->nLevel - iLevel - 1; + pOrTab = sqlite3StackAllocRaw(db, + sizeof(*pOrTab)+ nNotReady*sizeof(pOrTab->a[0])); + if( pOrTab==0 ) return notReady; + pOrTab->nAlloc = (u8)(nNotReady + 1); + pOrTab->nSrc = pOrTab->nAlloc; + memcpy(pOrTab->a, pTabItem, sizeof(*pTabItem)); + origSrc = pWInfo->pTabList->a; + for(k=1; k<=nNotReady; k++){ + memcpy(&pOrTab->a[k], &origSrc[pLevel[k].iFrom], sizeof(pOrTab->a[k])); + } + }else{ + pOrTab = pWInfo->pTabList; + } + + /* Initialize the rowset register to contain NULL. An SQL NULL is + ** equivalent to an empty rowset. Or, create an ephemeral index + ** capable of holding primary keys in the case of a WITHOUT ROWID. + ** + ** Also initialize regReturn to contain the address of the instruction + ** immediately following the OP_Return at the bottom of the loop. This + ** is required in a few obscure LEFT JOIN cases where control jumps + ** over the top of the loop into the body of it. In this case the + ** correct response for the end-of-loop code (the OP_Return) is to + ** fall through to the next instruction, just as an OP_Next does if + ** called on an uninitialized cursor. + */ + if( (pWInfo->wctrlFlags & WHERE_DUPLICATES_OK)==0 ){ + if( HasRowid(pTab) ){ + regRowset = ++pParse->nMem; + sqlite3VdbeAddOp2(v, OP_Null, 0, regRowset); + }else{ + Index *pPk = sqlite3PrimaryKeyIndex(pTab); + regRowset = pParse->nTab++; + sqlite3VdbeAddOp2(v, OP_OpenEphemeral, regRowset, pPk->nKeyCol); + sqlite3VdbeSetP4KeyInfo(pParse, pPk); + } + regRowid = ++pParse->nMem; + } + iRetInit = sqlite3VdbeAddOp2(v, OP_Integer, 0, regReturn); + + /* If the original WHERE clause is z of the form: (x1 OR x2 OR ...) AND y + ** Then for every term xN, evaluate as the subexpression: xN AND z + ** That way, terms in y that are factored into the disjunction will + ** be picked up by the recursive calls to sqlite3WhereBegin() below. + ** + ** Actually, each subexpression is converted to "xN AND w" where w is + ** the "interesting" terms of z - terms that did not originate in the + ** ON or USING clause of a LEFT JOIN, and terms that are usable as + ** indices. + ** + ** This optimization also only applies if the (x1 OR x2 OR ...) term + ** is not contained in the ON clause of a LEFT JOIN. + ** See ticket http://www.sqlite.org/src/info/f2369304e4 + */ + if( pWC->nTerm>1 ){ + int iTerm; + for(iTerm=0; iTermnTerm; iTerm++){ + Expr *pExpr = pWC->a[iTerm].pExpr; + if( &pWC->a[iTerm] == pTerm ) continue; + testcase( pWC->a[iTerm].wtFlags & TERM_VIRTUAL ); + testcase( pWC->a[iTerm].wtFlags & TERM_CODED ); + if( (pWC->a[iTerm].wtFlags & (TERM_VIRTUAL|TERM_CODED))!=0 ) continue; + if( (pWC->a[iTerm].eOperator & WO_ALL)==0 ) continue; + testcase( pWC->a[iTerm].wtFlags & TERM_ORINFO ); + pExpr = sqlite3ExprDup(db, pExpr, 0); + pAndExpr = sqlite3ExprAnd(pParse, pAndExpr, pExpr); + } + if( pAndExpr ){ + /* The extra 0x10000 bit on the opcode is masked off and does not + ** become part of the new Expr.op. However, it does make the + ** op==TK_AND comparison inside of sqlite3PExpr() false, and this + ** prevents sqlite3PExpr() from implementing AND short-circuit + ** optimization, which we do not want here. */ + pAndExpr = sqlite3PExpr(pParse, TK_AND|0x10000, 0, pAndExpr); + } + } + + /* Run a separate WHERE clause for each term of the OR clause. After + ** eliminating duplicates from other WHERE clauses, the action for each + ** sub-WHERE clause is to to invoke the main loop body as a subroutine. + */ + wctrlFlags = WHERE_OR_SUBCLAUSE | (pWInfo->wctrlFlags & WHERE_SEEK_TABLE); + ExplainQueryPlan((pParse, 1, "MULTI-INDEX OR")); + for(ii=0; iinTerm; ii++){ + WhereTerm *pOrTerm = &pOrWc->a[ii]; + if( pOrTerm->leftCursor==iCur || (pOrTerm->eOperator & WO_AND)!=0 ){ + WhereInfo *pSubWInfo; /* Info for single OR-term scan */ + Expr *pOrExpr = pOrTerm->pExpr; /* Current OR clause term */ + int jmp1 = 0; /* Address of jump operation */ + assert( (pTabItem[0].fg.jointype & JT_LEFT)==0 + || ExprHasProperty(pOrExpr, EP_FromJoin) + ); + if( pAndExpr ){ + pAndExpr->pLeft = pOrExpr; + pOrExpr = pAndExpr; + } + /* Loop through table entries that match term pOrTerm. */ + ExplainQueryPlan((pParse, 1, "INDEX %d", ii+1)); + WHERETRACE(0xffff, ("Subplan for OR-clause:\n")); + pSubWInfo = sqlite3WhereBegin(pParse, pOrTab, pOrExpr, 0, 0, + wctrlFlags, iCovCur); + assert( pSubWInfo || pParse->nErr || db->mallocFailed ); + if( pSubWInfo ){ + WhereLoop *pSubLoop; + int addrExplain = sqlite3WhereExplainOneScan( + pParse, pOrTab, &pSubWInfo->a[0], 0 + ); + sqlite3WhereAddScanStatus(v, pOrTab, &pSubWInfo->a[0], addrExplain); + + /* This is the sub-WHERE clause body. First skip over + ** duplicate rows from prior sub-WHERE clauses, and record the + ** rowid (or PRIMARY KEY) for the current row so that the same + ** row will be skipped in subsequent sub-WHERE clauses. + */ + if( (pWInfo->wctrlFlags & WHERE_DUPLICATES_OK)==0 ){ + int iSet = ((ii==pOrWc->nTerm-1)?-1:ii); + if( HasRowid(pTab) ){ + sqlite3ExprCodeGetColumnOfTable(v, pTab, iCur, -1, regRowid); + jmp1 = sqlite3VdbeAddOp4Int(v, OP_RowSetTest, regRowset, 0, + regRowid, iSet); + VdbeCoverage(v); + }else{ + Index *pPk = sqlite3PrimaryKeyIndex(pTab); + int nPk = pPk->nKeyCol; + int iPk; + int r; + + /* Read the PK into an array of temp registers. */ + r = sqlite3GetTempRange(pParse, nPk); + for(iPk=0; iPkaiColumn[iPk]; + sqlite3ExprCodeGetColumnOfTable(v, pTab, iCur, iCol, r+iPk); + } + + /* Check if the temp table already contains this key. If so, + ** the row has already been included in the result set and + ** can be ignored (by jumping past the Gosub below). Otherwise, + ** insert the key into the temp table and proceed with processing + ** the row. + ** + ** Use some of the same optimizations as OP_RowSetTest: If iSet + ** is zero, assume that the key cannot already be present in + ** the temp table. And if iSet is -1, assume that there is no + ** need to insert the key into the temp table, as it will never + ** be tested for. */ + if( iSet ){ + jmp1 = sqlite3VdbeAddOp4Int(v, OP_Found, regRowset, 0, r, nPk); + VdbeCoverage(v); + } + if( iSet>=0 ){ + sqlite3VdbeAddOp3(v, OP_MakeRecord, r, nPk, regRowid); + sqlite3VdbeAddOp4Int(v, OP_IdxInsert, regRowset, regRowid, + r, nPk); + if( iSet ) sqlite3VdbeChangeP5(v, OPFLAG_USESEEKRESULT); + } + + /* Release the array of temp registers */ + sqlite3ReleaseTempRange(pParse, r, nPk); + } + } + + /* Invoke the main loop body as a subroutine */ + sqlite3VdbeAddOp2(v, OP_Gosub, regReturn, iLoopBody); + + /* Jump here (skipping the main loop body subroutine) if the + ** current sub-WHERE row is a duplicate from prior sub-WHEREs. */ + if( jmp1 ) sqlite3VdbeJumpHere(v, jmp1); + + /* The pSubWInfo->untestedTerms flag means that this OR term + ** contained one or more AND term from a notReady table. The + ** terms from the notReady table could not be tested and will + ** need to be tested later. + */ + if( pSubWInfo->untestedTerms ) untestedTerms = 1; + + /* If all of the OR-connected terms are optimized using the same + ** index, and the index is opened using the same cursor number + ** by each call to sqlite3WhereBegin() made by this loop, it may + ** be possible to use that index as a covering index. + ** + ** If the call to sqlite3WhereBegin() above resulted in a scan that + ** uses an index, and this is either the first OR-connected term + ** processed or the index is the same as that used by all previous + ** terms, set pCov to the candidate covering index. Otherwise, set + ** pCov to NULL to indicate that no candidate covering index will + ** be available. + */ + pSubLoop = pSubWInfo->a[0].pWLoop; + assert( (pSubLoop->wsFlags & WHERE_AUTO_INDEX)==0 ); + if( (pSubLoop->wsFlags & WHERE_INDEXED)!=0 + && (ii==0 || pSubLoop->u.btree.pIndex==pCov) + && (HasRowid(pTab) || !IsPrimaryKeyIndex(pSubLoop->u.btree.pIndex)) + ){ + assert( pSubWInfo->a[0].iIdxCur==iCovCur ); + pCov = pSubLoop->u.btree.pIndex; + }else{ + pCov = 0; + } + + /* Finish the loop through table entries that match term pOrTerm. */ + sqlite3WhereEnd(pSubWInfo); + ExplainQueryPlanPop(pParse); + } + } + } + ExplainQueryPlanPop(pParse); + pLevel->u.pCovidx = pCov; + if( pCov ) pLevel->iIdxCur = iCovCur; + if( pAndExpr ){ + pAndExpr->pLeft = 0; + sqlite3ExprDelete(db, pAndExpr); + } + sqlite3VdbeChangeP1(v, iRetInit, sqlite3VdbeCurrentAddr(v)); + sqlite3VdbeGoto(v, pLevel->addrBrk); + sqlite3VdbeResolveLabel(v, iLoopBody); + + if( pWInfo->nLevel>1 ){ sqlite3StackFree(db, pOrTab); } + if( !untestedTerms ) disableTerm(pLevel, pTerm); + }else +#endif /* SQLITE_OMIT_OR_OPTIMIZATION */ + + { + /* Case 6: There is no usable index. We must do a complete + ** scan of the entire table. + */ + static const u8 aStep[] = { OP_Next, OP_Prev }; + static const u8 aStart[] = { OP_Rewind, OP_Last }; + assert( bRev==0 || bRev==1 ); + if( pTabItem->fg.isRecursive ){ + /* Tables marked isRecursive have only a single row that is stored in + ** a pseudo-cursor. No need to Rewind or Next such cursors. */ + pLevel->op = OP_Noop; + }else{ + codeCursorHint(pTabItem, pWInfo, pLevel, 0); + pLevel->op = aStep[bRev]; + pLevel->p1 = iCur; + pLevel->p2 = 1 + sqlite3VdbeAddOp2(v, aStart[bRev], iCur, addrHalt); + VdbeCoverageIf(v, bRev==0); + VdbeCoverageIf(v, bRev!=0); + pLevel->p5 = SQLITE_STMTSTATUS_FULLSCAN_STEP; + } + } + +#ifdef SQLITE_ENABLE_STMT_SCANSTATUS + pLevel->addrVisit = sqlite3VdbeCurrentAddr(v); +#endif + + /* Insert code to test every subexpression that can be completely + ** computed using the current set of tables. + ** + ** This loop may run between one and three times, depending on the + ** constraints to be generated. The value of stack variable iLoop + ** determines the constraints coded by each iteration, as follows: + ** + ** iLoop==1: Code only expressions that are entirely covered by pIdx. + ** iLoop==2: Code remaining expressions that do not contain correlated + ** sub-queries. + ** iLoop==3: Code all remaining expressions. + ** + ** An effort is made to skip unnecessary iterations of the loop. + */ + iLoop = (pIdx ? 1 : 2); + do{ + int iNext = 0; /* Next value for iLoop */ + for(pTerm=pWC->a, j=pWC->nTerm; j>0; j--, pTerm++){ + Expr *pE; + int skipLikeAddr = 0; + testcase( pTerm->wtFlags & TERM_VIRTUAL ); + testcase( pTerm->wtFlags & TERM_CODED ); + if( pTerm->wtFlags & (TERM_VIRTUAL|TERM_CODED) ) continue; + if( (pTerm->prereqAll & pLevel->notReady)!=0 ){ + testcase( pWInfo->untestedTerms==0 + && (pWInfo->wctrlFlags & WHERE_OR_SUBCLAUSE)!=0 ); + pWInfo->untestedTerms = 1; + continue; + } + pE = pTerm->pExpr; + assert( pE!=0 ); + if( (pTabItem->fg.jointype&JT_LEFT) && !ExprHasProperty(pE,EP_FromJoin) ){ + continue; + } + + if( iLoop==1 && !sqlite3ExprCoveredByIndex(pE, pLevel->iTabCur, pIdx) ){ + iNext = 2; + continue; + } + if( iLoop<3 && (pTerm->wtFlags & TERM_VARSELECT) ){ + if( iNext==0 ) iNext = 3; + continue; + } + + if( (pTerm->wtFlags & TERM_LIKECOND)!=0 ){ + /* If the TERM_LIKECOND flag is set, that means that the range search + ** is sufficient to guarantee that the LIKE operator is true, so we + ** can skip the call to the like(A,B) function. But this only works + ** for strings. So do not skip the call to the function on the pass + ** that compares BLOBs. */ +#ifdef SQLITE_LIKE_DOESNT_MATCH_BLOBS + continue; +#else + u32 x = pLevel->iLikeRepCntr; + if( x>0 ){ + skipLikeAddr = sqlite3VdbeAddOp1(v, (x&1)?OP_IfNot:OP_If,(int)(x>>1)); + VdbeCoverageIf(v, (x&1)==1); + VdbeCoverageIf(v, (x&1)==0); + } +#endif + } +#ifdef WHERETRACE_ENABLED /* 0xffff */ + if( sqlite3WhereTrace ){ + VdbeNoopComment((v, "WhereTerm[%d] (%p) priority=%d", + pWC->nTerm-j, pTerm, iLoop)); + } +#endif + sqlite3ExprIfFalse(pParse, pE, addrCont, SQLITE_JUMPIFNULL); + if( skipLikeAddr ) sqlite3VdbeJumpHere(v, skipLikeAddr); + pTerm->wtFlags |= TERM_CODED; + } + iLoop = iNext; + }while( iLoop>0 ); + + /* Insert code to test for implied constraints based on transitivity + ** of the "==" operator. + ** + ** Example: If the WHERE clause contains "t1.a=t2.b" and "t2.b=123" + ** and we are coding the t1 loop and the t2 loop has not yet coded, + ** then we cannot use the "t1.a=t2.b" constraint, but we can code + ** the implied "t1.a=123" constraint. + */ + for(pTerm=pWC->a, j=pWC->nTerm; j>0; j--, pTerm++){ + Expr *pE, sEAlt; + WhereTerm *pAlt; + if( pTerm->wtFlags & (TERM_VIRTUAL|TERM_CODED) ) continue; + if( (pTerm->eOperator & (WO_EQ|WO_IS))==0 ) continue; + if( (pTerm->eOperator & WO_EQUIV)==0 ) continue; + if( pTerm->leftCursor!=iCur ) continue; + if( pLevel->iLeftJoin ) continue; + pE = pTerm->pExpr; + assert( !ExprHasProperty(pE, EP_FromJoin) ); + assert( (pTerm->prereqRight & pLevel->notReady)!=0 ); + pAlt = sqlite3WhereFindTerm(pWC, iCur, pTerm->u.leftColumn, notReady, + WO_EQ|WO_IN|WO_IS, 0); + if( pAlt==0 ) continue; + if( pAlt->wtFlags & (TERM_CODED) ) continue; + if( (pAlt->eOperator & WO_IN) + && (pAlt->pExpr->flags & EP_xIsSelect) + && (pAlt->pExpr->x.pSelect->pEList->nExpr>1) + ){ + continue; + } + testcase( pAlt->eOperator & WO_EQ ); + testcase( pAlt->eOperator & WO_IS ); + testcase( pAlt->eOperator & WO_IN ); + VdbeModuleComment((v, "begin transitive constraint")); + sEAlt = *pAlt->pExpr; + sEAlt.pLeft = pE->pLeft; + sqlite3ExprIfFalse(pParse, &sEAlt, addrCont, SQLITE_JUMPIFNULL); + } + + /* For a LEFT OUTER JOIN, generate code that will record the fact that + ** at least one row of the right table has matched the left table. + */ + if( pLevel->iLeftJoin ){ + pLevel->addrFirst = sqlite3VdbeCurrentAddr(v); + sqlite3VdbeAddOp2(v, OP_Integer, 1, pLevel->iLeftJoin); + VdbeComment((v, "record LEFT JOIN hit")); + for(pTerm=pWC->a, j=0; jnTerm; j++, pTerm++){ + testcase( pTerm->wtFlags & TERM_VIRTUAL ); + testcase( pTerm->wtFlags & TERM_CODED ); + if( pTerm->wtFlags & (TERM_VIRTUAL|TERM_CODED) ) continue; + if( (pTerm->prereqAll & pLevel->notReady)!=0 ){ + assert( pWInfo->untestedTerms ); + continue; + } + assert( pTerm->pExpr ); + sqlite3ExprIfFalse(pParse, pTerm->pExpr, addrCont, SQLITE_JUMPIFNULL); + pTerm->wtFlags |= TERM_CODED; + } + } + + return pLevel->notReady; +} + +/************** End of wherecode.c *******************************************/ +/************** Begin file whereexpr.c ***************************************/ +/* +** 2015-06-08 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** This module contains C code that generates VDBE code used to process +** the WHERE clause of SQL statements. +** +** This file was originally part of where.c but was split out to improve +** readability and editabiliity. This file contains utility routines for +** analyzing Expr objects in the WHERE clause. +*/ +/* #include "sqliteInt.h" */ +/* #include "whereInt.h" */ + +/* Forward declarations */ +static void exprAnalyze(SrcList*, WhereClause*, int); + +/* +** Deallocate all memory associated with a WhereOrInfo object. +*/ +static void whereOrInfoDelete(sqlite3 *db, WhereOrInfo *p){ + sqlite3WhereClauseClear(&p->wc); + sqlite3DbFree(db, p); +} + +/* +** Deallocate all memory associated with a WhereAndInfo object. +*/ +static void whereAndInfoDelete(sqlite3 *db, WhereAndInfo *p){ + sqlite3WhereClauseClear(&p->wc); + sqlite3DbFree(db, p); +} + +/* +** Add a single new WhereTerm entry to the WhereClause object pWC. +** The new WhereTerm object is constructed from Expr p and with wtFlags. +** The index in pWC->a[] of the new WhereTerm is returned on success. +** 0 is returned if the new WhereTerm could not be added due to a memory +** allocation error. The memory allocation failure will be recorded in +** the db->mallocFailed flag so that higher-level functions can detect it. +** +** This routine will increase the size of the pWC->a[] array as necessary. +** +** If the wtFlags argument includes TERM_DYNAMIC, then responsibility +** for freeing the expression p is assumed by the WhereClause object pWC. +** This is true even if this routine fails to allocate a new WhereTerm. +** +** WARNING: This routine might reallocate the space used to store +** WhereTerms. All pointers to WhereTerms should be invalidated after +** calling this routine. Such pointers may be reinitialized by referencing +** the pWC->a[] array. +*/ +static int whereClauseInsert(WhereClause *pWC, Expr *p, u16 wtFlags){ + WhereTerm *pTerm; + int idx; + testcase( wtFlags & TERM_VIRTUAL ); + if( pWC->nTerm>=pWC->nSlot ){ + WhereTerm *pOld = pWC->a; + sqlite3 *db = pWC->pWInfo->pParse->db; + pWC->a = sqlite3DbMallocRawNN(db, sizeof(pWC->a[0])*pWC->nSlot*2 ); + if( pWC->a==0 ){ + if( wtFlags & TERM_DYNAMIC ){ + sqlite3ExprDelete(db, p); + } + pWC->a = pOld; + return 0; + } + memcpy(pWC->a, pOld, sizeof(pWC->a[0])*pWC->nTerm); + if( pOld!=pWC->aStatic ){ + sqlite3DbFree(db, pOld); + } + pWC->nSlot = sqlite3DbMallocSize(db, pWC->a)/sizeof(pWC->a[0]); + } + pTerm = &pWC->a[idx = pWC->nTerm++]; + if( p && ExprHasProperty(p, EP_Unlikely) ){ + pTerm->truthProb = sqlite3LogEst(p->iTable) - 270; + }else{ + pTerm->truthProb = 1; + } + pTerm->pExpr = sqlite3ExprSkipCollateAndLikely(p); + pTerm->wtFlags = wtFlags; + pTerm->pWC = pWC; + pTerm->iParent = -1; + memset(&pTerm->eOperator, 0, + sizeof(WhereTerm) - offsetof(WhereTerm,eOperator)); + return idx; +} + +/* +** Return TRUE if the given operator is one of the operators that is +** allowed for an indexable WHERE clause term. The allowed operators are +** "=", "<", ">", "<=", ">=", "IN", "IS", and "IS NULL" +*/ +static int allowedOp(int op){ + assert( TK_GT>TK_EQ && TK_GTTK_EQ && TK_LTTK_EQ && TK_LE=TK_EQ && op<=TK_GE) || op==TK_ISNULL || op==TK_IS; +} + +/* +** Commute a comparison operator. Expressions of the form "X op Y" +** are converted into "Y op X". +** +** If left/right precedence rules come into play when determining the +** collating sequence, then COLLATE operators are adjusted to ensure +** that the collating sequence does not change. For example: +** "Y collate NOCASE op X" becomes "X op Y" because any collation sequence on +** the left hand side of a comparison overrides any collation sequence +** attached to the right. For the same reason the EP_Collate flag +** is not commuted. +** +** The return value is extra flags that are added to the WhereTerm object +** after it is commuted. The only extra flag ever added is TERM_NOPARTIDX +** which prevents the term from being used to enable a partial index if +** COLLATE changes have been made. +*/ +static u16 exprCommute(Parse *pParse, Expr *pExpr){ + u16 expRight = (pExpr->pRight->flags & EP_Collate); + u16 expLeft = (pExpr->pLeft->flags & EP_Collate); + u16 wtFlags = 0; + assert( allowedOp(pExpr->op) && pExpr->op!=TK_IN ); + if( expRight==expLeft ){ + /* Either X and Y both have COLLATE operator or neither do */ + if( expRight ){ + /* Both X and Y have COLLATE operators. Make sure X is always + ** used by clearing the EP_Collate flag from Y. */ + pExpr->pRight->flags &= ~EP_Collate; + wtFlags |= TERM_NOPARTIDX; + }else if( sqlite3ExprCollSeq(pParse, pExpr->pLeft)!=0 ){ + /* Neither X nor Y have COLLATE operators, but X has a non-default + ** collating sequence. So add the EP_Collate marker on X to cause + ** it to be searched first. */ + pExpr->pLeft->flags |= EP_Collate; + wtFlags |= TERM_NOPARTIDX; + } + } + SWAP(Expr*,pExpr->pRight,pExpr->pLeft); + if( pExpr->op>=TK_GT ){ + assert( TK_LT==TK_GT+2 ); + assert( TK_GE==TK_LE+2 ); + assert( TK_GT>TK_EQ ); + assert( TK_GTop>=TK_GT && pExpr->op<=TK_GE ); + pExpr->op = ((pExpr->op-TK_GT)^2)+TK_GT; + } + return wtFlags; +} + +/* +** Translate from TK_xx operator to WO_xx bitmask. +*/ +static u16 operatorMask(int op){ + u16 c; + assert( allowedOp(op) ); + if( op==TK_IN ){ + c = WO_IN; + }else if( op==TK_ISNULL ){ + c = WO_ISNULL; + }else if( op==TK_IS ){ + c = WO_IS; + }else{ + assert( (WO_EQ<<(op-TK_EQ)) < 0x7fff ); + c = (u16)(WO_EQ<<(op-TK_EQ)); + } + assert( op!=TK_ISNULL || c==WO_ISNULL ); + assert( op!=TK_IN || c==WO_IN ); + assert( op!=TK_EQ || c==WO_EQ ); + assert( op!=TK_LT || c==WO_LT ); + assert( op!=TK_LE || c==WO_LE ); + assert( op!=TK_GT || c==WO_GT ); + assert( op!=TK_GE || c==WO_GE ); + assert( op!=TK_IS || c==WO_IS ); + return c; +} + + +#ifndef SQLITE_OMIT_LIKE_OPTIMIZATION +/* +** Check to see if the given expression is a LIKE or GLOB operator that +** can be optimized using inequality constraints. Return TRUE if it is +** so and false if not. +** +** In order for the operator to be optimizible, the RHS must be a string +** literal that does not begin with a wildcard. The LHS must be a column +** that may only be NULL, a string, or a BLOB, never a number. (This means +** that virtual tables cannot participate in the LIKE optimization.) The +** collating sequence for the column on the LHS must be appropriate for +** the operator. +*/ +static int isLikeOrGlob( + Parse *pParse, /* Parsing and code generating context */ + Expr *pExpr, /* Test this expression */ + Expr **ppPrefix, /* Pointer to TK_STRING expression with pattern prefix */ + int *pisComplete, /* True if the only wildcard is % in the last character */ + int *pnoCase /* True if uppercase is equivalent to lowercase */ +){ + const u8 *z = 0; /* String on RHS of LIKE operator */ + Expr *pRight, *pLeft; /* Right and left size of LIKE operator */ + ExprList *pList; /* List of operands to the LIKE operator */ + u8 c; /* One character in z[] */ + int cnt; /* Number of non-wildcard prefix characters */ + u8 wc[4]; /* Wildcard characters */ + sqlite3 *db = pParse->db; /* Database connection */ + sqlite3_value *pVal = 0; + int op; /* Opcode of pRight */ + int rc; /* Result code to return */ + + if( !sqlite3IsLikeFunction(db, pExpr, pnoCase, (char*)wc) ){ + return 0; + } +#ifdef SQLITE_EBCDIC + if( *pnoCase ) return 0; +#endif + pList = pExpr->x.pList; + pLeft = pList->a[1].pExpr; + + pRight = sqlite3ExprSkipCollate(pList->a[0].pExpr); + op = pRight->op; + if( op==TK_VARIABLE && (db->flags & SQLITE_EnableQPSG)==0 ){ + Vdbe *pReprepare = pParse->pReprepare; + int iCol = pRight->iColumn; + pVal = sqlite3VdbeGetBoundValue(pReprepare, iCol, SQLITE_AFF_BLOB); + if( pVal && sqlite3_value_type(pVal)==SQLITE_TEXT ){ + z = sqlite3_value_text(pVal); + } + sqlite3VdbeSetVarmask(pParse->pVdbe, iCol); + assert( pRight->op==TK_VARIABLE || pRight->op==TK_REGISTER ); + }else if( op==TK_STRING ){ + z = (u8*)pRight->u.zToken; + } + if( z ){ + + /* Count the number of prefix characters prior to the first wildcard */ + cnt = 0; + while( (c=z[cnt])!=0 && c!=wc[0] && c!=wc[1] && c!=wc[2] ){ + cnt++; + if( c==wc[3] && z[cnt]!=0 ) cnt++; + } + + /* The optimization is possible only if (1) the pattern does not begin + ** with a wildcard and if (2) the non-wildcard prefix does not end with + ** an (illegal 0xff) character, or (3) the pattern does not consist of + ** a single escape character. The second condition is necessary so + ** that we can increment the prefix key to find an upper bound for the + ** range search. The third is because the caller assumes that the pattern + ** consists of at least one character after all escapes have been + ** removed. */ + if( cnt!=0 && 255!=(u8)z[cnt-1] && (cnt>1 || z[0]!=wc[3]) ){ + Expr *pPrefix; + + /* A "complete" match if the pattern ends with "*" or "%" */ + *pisComplete = c==wc[0] && z[cnt+1]==0; + + /* Get the pattern prefix. Remove all escapes from the prefix. */ + pPrefix = sqlite3Expr(db, TK_STRING, (char*)z); + if( pPrefix ){ + int iFrom, iTo; + char *zNew = pPrefix->u.zToken; + zNew[cnt] = 0; + for(iFrom=iTo=0; iFrom0 ); + + /* If the LHS is not an ordinary column with TEXT affinity, then the + ** pattern prefix boundaries (both the start and end boundaries) must + ** not look like a number. Otherwise the pattern might be treated as + ** a number, which will invalidate the LIKE optimization. + ** + ** Getting this right has been a persistent source of bugs in the + ** LIKE optimization. See, for example: + ** 2018-09-10 https://sqlite.org/src/info/c94369cae9b561b1 + ** 2019-05-02 https://sqlite.org/src/info/b043a54c3de54b28 + ** 2019-06-10 https://sqlite.org/src/info/fd76310a5e843e07 + ** 2019-06-14 https://sqlite.org/src/info/ce8717f0885af975 + ** 2019-09-03 https://sqlite.org/src/info/0f0428096f17252a + */ + if( pLeft->op!=TK_COLUMN + || sqlite3ExprAffinity(pLeft)!=SQLITE_AFF_TEXT + || IsVirtual(pLeft->y.pTab) /* Value might be numeric */ + ){ + int isNum; + double rDummy; + isNum = sqlite3AtoF(zNew, &rDummy, iTo, SQLITE_UTF8); + if( isNum<=0 ){ + if( iTo==1 && zNew[0]=='-' ){ + isNum = +1; + }else{ + zNew[iTo-1]++; + isNum = sqlite3AtoF(zNew, &rDummy, iTo, SQLITE_UTF8); + zNew[iTo-1]--; + } + } + if( isNum>0 ){ + sqlite3ExprDelete(db, pPrefix); + sqlite3ValueFree(pVal); + return 0; + } + } + } + *ppPrefix = pPrefix; + + /* If the RHS pattern is a bound parameter, make arrangements to + ** reprepare the statement when that parameter is rebound */ + if( op==TK_VARIABLE ){ + Vdbe *v = pParse->pVdbe; + sqlite3VdbeSetVarmask(v, pRight->iColumn); + if( *pisComplete && pRight->u.zToken[1] ){ + /* If the rhs of the LIKE expression is a variable, and the current + ** value of the variable means there is no need to invoke the LIKE + ** function, then no OP_Variable will be added to the program. + ** This causes problems for the sqlite3_bind_parameter_name() + ** API. To work around them, add a dummy OP_Variable here. + */ + int r1 = sqlite3GetTempReg(pParse); + sqlite3ExprCodeTarget(pParse, pRight, r1); + sqlite3VdbeChangeP3(v, sqlite3VdbeCurrentAddr(v)-1, 0); + sqlite3ReleaseTempReg(pParse, r1); + } + } + }else{ + z = 0; + } + } + + rc = (z!=0); + sqlite3ValueFree(pVal); + return rc; +} +#endif /* SQLITE_OMIT_LIKE_OPTIMIZATION */ + + +#ifndef SQLITE_OMIT_VIRTUALTABLE +/* +** Check to see if the pExpr expression is a form that needs to be passed +** to the xBestIndex method of virtual tables. Forms of interest include: +** +** Expression Virtual Table Operator +** ----------------------- --------------------------------- +** 1. column MATCH expr SQLITE_INDEX_CONSTRAINT_MATCH +** 2. column GLOB expr SQLITE_INDEX_CONSTRAINT_GLOB +** 3. column LIKE expr SQLITE_INDEX_CONSTRAINT_LIKE +** 4. column REGEXP expr SQLITE_INDEX_CONSTRAINT_REGEXP +** 5. column != expr SQLITE_INDEX_CONSTRAINT_NE +** 6. expr != column SQLITE_INDEX_CONSTRAINT_NE +** 7. column IS NOT expr SQLITE_INDEX_CONSTRAINT_ISNOT +** 8. expr IS NOT column SQLITE_INDEX_CONSTRAINT_ISNOT +** 9. column IS NOT NULL SQLITE_INDEX_CONSTRAINT_ISNOTNULL +** +** In every case, "column" must be a column of a virtual table. If there +** is a match, set *ppLeft to the "column" expression, set *ppRight to the +** "expr" expression (even though in forms (6) and (8) the column is on the +** right and the expression is on the left). Also set *peOp2 to the +** appropriate virtual table operator. The return value is 1 or 2 if there +** is a match. The usual return is 1, but if the RHS is also a column +** of virtual table in forms (5) or (7) then return 2. +** +** If the expression matches none of the patterns above, return 0. +*/ +static int isAuxiliaryVtabOperator( + sqlite3 *db, /* Parsing context */ + Expr *pExpr, /* Test this expression */ + unsigned char *peOp2, /* OUT: 0 for MATCH, or else an op2 value */ + Expr **ppLeft, /* Column expression to left of MATCH/op2 */ + Expr **ppRight /* Expression to left of MATCH/op2 */ +){ + if( pExpr->op==TK_FUNCTION ){ + static const struct Op2 { + const char *zOp; + unsigned char eOp2; + } aOp[] = { + { "match", SQLITE_INDEX_CONSTRAINT_MATCH }, + { "glob", SQLITE_INDEX_CONSTRAINT_GLOB }, + { "like", SQLITE_INDEX_CONSTRAINT_LIKE }, + { "regexp", SQLITE_INDEX_CONSTRAINT_REGEXP } + }; + ExprList *pList; + Expr *pCol; /* Column reference */ + int i; + + pList = pExpr->x.pList; + if( pList==0 || pList->nExpr!=2 ){ + return 0; + } + + /* Built-in operators MATCH, GLOB, LIKE, and REGEXP attach to a + ** virtual table on their second argument, which is the same as + ** the left-hand side operand in their in-fix form. + ** + ** vtab_column MATCH expression + ** MATCH(expression,vtab_column) + */ + pCol = pList->a[1].pExpr; + if( pCol->op==TK_COLUMN && IsVirtual(pCol->y.pTab) ){ + for(i=0; iu.zToken, aOp[i].zOp)==0 ){ + *peOp2 = aOp[i].eOp2; + *ppRight = pList->a[0].pExpr; + *ppLeft = pCol; + return 1; + } + } + } + + /* We can also match against the first column of overloaded + ** functions where xFindFunction returns a value of at least + ** SQLITE_INDEX_CONSTRAINT_FUNCTION. + ** + ** OVERLOADED(vtab_column,expression) + ** + ** Historically, xFindFunction expected to see lower-case function + ** names. But for this use case, xFindFunction is expected to deal + ** with function names in an arbitrary case. + */ + pCol = pList->a[0].pExpr; + if( pCol->op==TK_COLUMN && IsVirtual(pCol->y.pTab) ){ + sqlite3_vtab *pVtab; + sqlite3_module *pMod; + void (*xNotUsed)(sqlite3_context*,int,sqlite3_value**); + void *pNotUsed; + pVtab = sqlite3GetVTable(db, pCol->y.pTab)->pVtab; + assert( pVtab!=0 ); + assert( pVtab->pModule!=0 ); + pMod = (sqlite3_module *)pVtab->pModule; + if( pMod->xFindFunction!=0 ){ + i = pMod->xFindFunction(pVtab,2, pExpr->u.zToken, &xNotUsed, &pNotUsed); + if( i>=SQLITE_INDEX_CONSTRAINT_FUNCTION ){ + *peOp2 = i; + *ppRight = pList->a[1].pExpr; + *ppLeft = pCol; + return 1; + } + } + } + }else if( pExpr->op==TK_NE || pExpr->op==TK_ISNOT || pExpr->op==TK_NOTNULL ){ + int res = 0; + Expr *pLeft = pExpr->pLeft; + Expr *pRight = pExpr->pRight; + if( pLeft->op==TK_COLUMN && IsVirtual(pLeft->y.pTab) ){ + res++; + } + if( pRight && pRight->op==TK_COLUMN && IsVirtual(pRight->y.pTab) ){ + res++; + SWAP(Expr*, pLeft, pRight); + } + *ppLeft = pLeft; + *ppRight = pRight; + if( pExpr->op==TK_NE ) *peOp2 = SQLITE_INDEX_CONSTRAINT_NE; + if( pExpr->op==TK_ISNOT ) *peOp2 = SQLITE_INDEX_CONSTRAINT_ISNOT; + if( pExpr->op==TK_NOTNULL ) *peOp2 = SQLITE_INDEX_CONSTRAINT_ISNOTNULL; + return res; + } + return 0; +} +#endif /* SQLITE_OMIT_VIRTUALTABLE */ + +/* +** If the pBase expression originated in the ON or USING clause of +** a join, then transfer the appropriate markings over to derived. +*/ +static void transferJoinMarkings(Expr *pDerived, Expr *pBase){ + if( pDerived ){ + pDerived->flags |= pBase->flags & EP_FromJoin; + pDerived->iRightJoinTable = pBase->iRightJoinTable; + } +} + +/* +** Mark term iChild as being a child of term iParent +*/ +static void markTermAsChild(WhereClause *pWC, int iChild, int iParent){ + pWC->a[iChild].iParent = iParent; + pWC->a[iChild].truthProb = pWC->a[iParent].truthProb; + pWC->a[iParent].nChild++; +} + +/* +** Return the N-th AND-connected subterm of pTerm. Or if pTerm is not +** a conjunction, then return just pTerm when N==0. If N is exceeds +** the number of available subterms, return NULL. +*/ +static WhereTerm *whereNthSubterm(WhereTerm *pTerm, int N){ + if( pTerm->eOperator!=WO_AND ){ + return N==0 ? pTerm : 0; + } + if( Nu.pAndInfo->wc.nTerm ){ + return &pTerm->u.pAndInfo->wc.a[N]; + } + return 0; +} + +/* +** Subterms pOne and pTwo are contained within WHERE clause pWC. The +** two subterms are in disjunction - they are OR-ed together. +** +** If these two terms are both of the form: "A op B" with the same +** A and B values but different operators and if the operators are +** compatible (if one is = and the other is <, for example) then +** add a new virtual AND term to pWC that is the combination of the +** two. +** +** Some examples: +** +** x x<=y +** x=y OR x=y --> x=y +** x<=y OR x x<=y +** +** The following is NOT generated: +** +** xy --> x!=y +*/ +static void whereCombineDisjuncts( + SrcList *pSrc, /* the FROM clause */ + WhereClause *pWC, /* The complete WHERE clause */ + WhereTerm *pOne, /* First disjunct */ + WhereTerm *pTwo /* Second disjunct */ +){ + u16 eOp = pOne->eOperator | pTwo->eOperator; + sqlite3 *db; /* Database connection (for malloc) */ + Expr *pNew; /* New virtual expression */ + int op; /* Operator for the combined expression */ + int idxNew; /* Index in pWC of the next virtual term */ + + if( (pOne->eOperator & (WO_EQ|WO_LT|WO_LE|WO_GT|WO_GE))==0 ) return; + if( (pTwo->eOperator & (WO_EQ|WO_LT|WO_LE|WO_GT|WO_GE))==0 ) return; + if( (eOp & (WO_EQ|WO_LT|WO_LE))!=eOp + && (eOp & (WO_EQ|WO_GT|WO_GE))!=eOp ) return; + assert( pOne->pExpr->pLeft!=0 && pOne->pExpr->pRight!=0 ); + assert( pTwo->pExpr->pLeft!=0 && pTwo->pExpr->pRight!=0 ); + if( sqlite3ExprCompare(0,pOne->pExpr->pLeft, pTwo->pExpr->pLeft, -1) ) return; + if( sqlite3ExprCompare(0,pOne->pExpr->pRight, pTwo->pExpr->pRight,-1) )return; + /* If we reach this point, it means the two subterms can be combined */ + if( (eOp & (eOp-1))!=0 ){ + if( eOp & (WO_LT|WO_LE) ){ + eOp = WO_LE; + }else{ + assert( eOp & (WO_GT|WO_GE) ); + eOp = WO_GE; + } + } + db = pWC->pWInfo->pParse->db; + pNew = sqlite3ExprDup(db, pOne->pExpr, 0); + if( pNew==0 ) return; + for(op=TK_EQ; eOp!=(WO_EQ<<(op-TK_EQ)); op++){ assert( opop = op; + idxNew = whereClauseInsert(pWC, pNew, TERM_VIRTUAL|TERM_DYNAMIC); + exprAnalyze(pSrc, pWC, idxNew); +} + +#if !defined(SQLITE_OMIT_OR_OPTIMIZATION) && !defined(SQLITE_OMIT_SUBQUERY) +/* +** Analyze a term that consists of two or more OR-connected +** subterms. So in: +** +** ... WHERE (a=5) AND (b=7 OR c=9 OR d=13) AND (d=13) +** ^^^^^^^^^^^^^^^^^^^^ +** +** This routine analyzes terms such as the middle term in the above example. +** A WhereOrTerm object is computed and attached to the term under +** analysis, regardless of the outcome of the analysis. Hence: +** +** WhereTerm.wtFlags |= TERM_ORINFO +** WhereTerm.u.pOrInfo = a dynamically allocated WhereOrTerm object +** +** The term being analyzed must have two or more of OR-connected subterms. +** A single subterm might be a set of AND-connected sub-subterms. +** Examples of terms under analysis: +** +** (A) t1.x=t2.y OR t1.x=t2.z OR t1.y=15 OR t1.z=t3.a+5 +** (B) x=expr1 OR expr2=x OR x=expr3 +** (C) t1.x=t2.y OR (t1.x=t2.z AND t1.y=15) +** (D) x=expr1 OR (y>11 AND y<22 AND z LIKE '*hello*') +** (E) (p.a=1 AND q.b=2 AND r.c=3) OR (p.x=4 AND q.y=5 AND r.z=6) +** (F) x>A OR (x=A AND y>=B) +** +** CASE 1: +** +** If all subterms are of the form T.C=expr for some single column of C and +** a single table T (as shown in example B above) then create a new virtual +** term that is an equivalent IN expression. In other words, if the term +** being analyzed is: +** +** x = expr1 OR expr2 = x OR x = expr3 +** +** then create a new virtual term like this: +** +** x IN (expr1,expr2,expr3) +** +** CASE 2: +** +** If there are exactly two disjuncts and one side has x>A and the other side +** has x=A (for the same x and A) then add a new virtual conjunct term to the +** WHERE clause of the form "x>=A". Example: +** +** x>A OR (x=A AND y>B) adds: x>=A +** +** The added conjunct can sometimes be helpful in query planning. +** +** CASE 3: +** +** If all subterms are indexable by a single table T, then set +** +** WhereTerm.eOperator = WO_OR +** WhereTerm.u.pOrInfo->indexable |= the cursor number for table T +** +** A subterm is "indexable" if it is of the form +** "T.C " where C is any column of table T and +** is one of "=", "<", "<=", ">", ">=", "IS NULL", or "IN". +** A subterm is also indexable if it is an AND of two or more +** subsubterms at least one of which is indexable. Indexable AND +** subterms have their eOperator set to WO_AND and they have +** u.pAndInfo set to a dynamically allocated WhereAndTerm object. +** +** From another point of view, "indexable" means that the subterm could +** potentially be used with an index if an appropriate index exists. +** This analysis does not consider whether or not the index exists; that +** is decided elsewhere. This analysis only looks at whether subterms +** appropriate for indexing exist. +** +** All examples A through E above satisfy case 3. But if a term +** also satisfies case 1 (such as B) we know that the optimizer will +** always prefer case 1, so in that case we pretend that case 3 is not +** satisfied. +** +** It might be the case that multiple tables are indexable. For example, +** (E) above is indexable on tables P, Q, and R. +** +** Terms that satisfy case 3 are candidates for lookup by using +** separate indices to find rowids for each subterm and composing +** the union of all rowids using a RowSet object. This is similar +** to "bitmap indices" in other database engines. +** +** OTHERWISE: +** +** If none of cases 1, 2, or 3 apply, then leave the eOperator set to +** zero. This term is not useful for search. +*/ +static void exprAnalyzeOrTerm( + SrcList *pSrc, /* the FROM clause */ + WhereClause *pWC, /* the complete WHERE clause */ + int idxTerm /* Index of the OR-term to be analyzed */ +){ + WhereInfo *pWInfo = pWC->pWInfo; /* WHERE clause processing context */ + Parse *pParse = pWInfo->pParse; /* Parser context */ + sqlite3 *db = pParse->db; /* Database connection */ + WhereTerm *pTerm = &pWC->a[idxTerm]; /* The term to be analyzed */ + Expr *pExpr = pTerm->pExpr; /* The expression of the term */ + int i; /* Loop counters */ + WhereClause *pOrWc; /* Breakup of pTerm into subterms */ + WhereTerm *pOrTerm; /* A Sub-term within the pOrWc */ + WhereOrInfo *pOrInfo; /* Additional information associated with pTerm */ + Bitmask chngToIN; /* Tables that might satisfy case 1 */ + Bitmask indexable; /* Tables that are indexable, satisfying case 2 */ + + /* + ** Break the OR clause into its separate subterms. The subterms are + ** stored in a WhereClause structure containing within the WhereOrInfo + ** object that is attached to the original OR clause term. + */ + assert( (pTerm->wtFlags & (TERM_DYNAMIC|TERM_ORINFO|TERM_ANDINFO))==0 ); + assert( pExpr->op==TK_OR ); + pTerm->u.pOrInfo = pOrInfo = sqlite3DbMallocZero(db, sizeof(*pOrInfo)); + if( pOrInfo==0 ) return; + pTerm->wtFlags |= TERM_ORINFO; + pOrWc = &pOrInfo->wc; + memset(pOrWc->aStatic, 0, sizeof(pOrWc->aStatic)); + sqlite3WhereClauseInit(pOrWc, pWInfo); + sqlite3WhereSplit(pOrWc, pExpr, TK_OR); + sqlite3WhereExprAnalyze(pSrc, pOrWc); + if( db->mallocFailed ) return; + assert( pOrWc->nTerm>=2 ); + + /* + ** Compute the set of tables that might satisfy cases 1 or 3. + */ + indexable = ~(Bitmask)0; + chngToIN = ~(Bitmask)0; + for(i=pOrWc->nTerm-1, pOrTerm=pOrWc->a; i>=0 && indexable; i--, pOrTerm++){ + if( (pOrTerm->eOperator & WO_SINGLE)==0 ){ + WhereAndInfo *pAndInfo; + assert( (pOrTerm->wtFlags & (TERM_ANDINFO|TERM_ORINFO))==0 ); + chngToIN = 0; + pAndInfo = sqlite3DbMallocRawNN(db, sizeof(*pAndInfo)); + if( pAndInfo ){ + WhereClause *pAndWC; + WhereTerm *pAndTerm; + int j; + Bitmask b = 0; + pOrTerm->u.pAndInfo = pAndInfo; + pOrTerm->wtFlags |= TERM_ANDINFO; + pOrTerm->eOperator = WO_AND; + pAndWC = &pAndInfo->wc; + memset(pAndWC->aStatic, 0, sizeof(pAndWC->aStatic)); + sqlite3WhereClauseInit(pAndWC, pWC->pWInfo); + sqlite3WhereSplit(pAndWC, pOrTerm->pExpr, TK_AND); + sqlite3WhereExprAnalyze(pSrc, pAndWC); + pAndWC->pOuter = pWC; + if( !db->mallocFailed ){ + for(j=0, pAndTerm=pAndWC->a; jnTerm; j++, pAndTerm++){ + assert( pAndTerm->pExpr ); + if( allowedOp(pAndTerm->pExpr->op) + || pAndTerm->eOperator==WO_AUX + ){ + b |= sqlite3WhereGetMask(&pWInfo->sMaskSet, pAndTerm->leftCursor); + } + } + } + indexable &= b; + } + }else if( pOrTerm->wtFlags & TERM_COPIED ){ + /* Skip this term for now. We revisit it when we process the + ** corresponding TERM_VIRTUAL term */ + }else{ + Bitmask b; + b = sqlite3WhereGetMask(&pWInfo->sMaskSet, pOrTerm->leftCursor); + if( pOrTerm->wtFlags & TERM_VIRTUAL ){ + WhereTerm *pOther = &pOrWc->a[pOrTerm->iParent]; + b |= sqlite3WhereGetMask(&pWInfo->sMaskSet, pOther->leftCursor); + } + indexable &= b; + if( (pOrTerm->eOperator & WO_EQ)==0 ){ + chngToIN = 0; + }else{ + chngToIN &= b; + } + } + } + + /* + ** Record the set of tables that satisfy case 3. The set might be + ** empty. + */ + pOrInfo->indexable = indexable; + if( indexable ){ + pTerm->eOperator = WO_OR; + pWC->hasOr = 1; + }else{ + pTerm->eOperator = WO_OR; + } + + /* For a two-way OR, attempt to implementation case 2. + */ + if( indexable && pOrWc->nTerm==2 ){ + int iOne = 0; + WhereTerm *pOne; + while( (pOne = whereNthSubterm(&pOrWc->a[0],iOne++))!=0 ){ + int iTwo = 0; + WhereTerm *pTwo; + while( (pTwo = whereNthSubterm(&pOrWc->a[1],iTwo++))!=0 ){ + whereCombineDisjuncts(pSrc, pWC, pOne, pTwo); + } + } + } + + /* + ** chngToIN holds a set of tables that *might* satisfy case 1. But + ** we have to do some additional checking to see if case 1 really + ** is satisfied. + ** + ** chngToIN will hold either 0, 1, or 2 bits. The 0-bit case means + ** that there is no possibility of transforming the OR clause into an + ** IN operator because one or more terms in the OR clause contain + ** something other than == on a column in the single table. The 1-bit + ** case means that every term of the OR clause is of the form + ** "table.column=expr" for some single table. The one bit that is set + ** will correspond to the common table. We still need to check to make + ** sure the same column is used on all terms. The 2-bit case is when + ** the all terms are of the form "table1.column=table2.column". It + ** might be possible to form an IN operator with either table1.column + ** or table2.column as the LHS if either is common to every term of + ** the OR clause. + ** + ** Note that terms of the form "table.column1=table.column2" (the + ** same table on both sizes of the ==) cannot be optimized. + */ + if( chngToIN ){ + int okToChngToIN = 0; /* True if the conversion to IN is valid */ + int iColumn = -1; /* Column index on lhs of IN operator */ + int iCursor = -1; /* Table cursor common to all terms */ + int j = 0; /* Loop counter */ + + /* Search for a table and column that appears on one side or the + ** other of the == operator in every subterm. That table and column + ** will be recorded in iCursor and iColumn. There might not be any + ** such table and column. Set okToChngToIN if an appropriate table + ** and column is found but leave okToChngToIN false if not found. + */ + for(j=0; j<2 && !okToChngToIN; j++){ + Expr *pLeft = 0; + pOrTerm = pOrWc->a; + for(i=pOrWc->nTerm-1; i>=0; i--, pOrTerm++){ + assert( pOrTerm->eOperator & WO_EQ ); + pOrTerm->wtFlags &= ~TERM_OR_OK; + if( pOrTerm->leftCursor==iCursor ){ + /* This is the 2-bit case and we are on the second iteration and + ** current term is from the first iteration. So skip this term. */ + assert( j==1 ); + continue; + } + if( (chngToIN & sqlite3WhereGetMask(&pWInfo->sMaskSet, + pOrTerm->leftCursor))==0 ){ + /* This term must be of the form t1.a==t2.b where t2 is in the + ** chngToIN set but t1 is not. This term will be either preceded + ** or follwed by an inverted copy (t2.b==t1.a). Skip this term + ** and use its inversion. */ + testcase( pOrTerm->wtFlags & TERM_COPIED ); + testcase( pOrTerm->wtFlags & TERM_VIRTUAL ); + assert( pOrTerm->wtFlags & (TERM_COPIED|TERM_VIRTUAL) ); + continue; + } + iColumn = pOrTerm->u.leftColumn; + iCursor = pOrTerm->leftCursor; + pLeft = pOrTerm->pExpr->pLeft; + break; + } + if( i<0 ){ + /* No candidate table+column was found. This can only occur + ** on the second iteration */ + assert( j==1 ); + assert( IsPowerOfTwo(chngToIN) ); + assert( chngToIN==sqlite3WhereGetMask(&pWInfo->sMaskSet, iCursor) ); + break; + } + testcase( j==1 ); + + /* We have found a candidate table and column. Check to see if that + ** table and column is common to every term in the OR clause */ + okToChngToIN = 1; + for(; i>=0 && okToChngToIN; i--, pOrTerm++){ + assert( pOrTerm->eOperator & WO_EQ ); + if( pOrTerm->leftCursor!=iCursor ){ + pOrTerm->wtFlags &= ~TERM_OR_OK; + }else if( pOrTerm->u.leftColumn!=iColumn || (iColumn==XN_EXPR + && sqlite3ExprCompare(pParse, pOrTerm->pExpr->pLeft, pLeft, -1) + )){ + okToChngToIN = 0; + }else{ + int affLeft, affRight; + /* If the right-hand side is also a column, then the affinities + ** of both right and left sides must be such that no type + ** conversions are required on the right. (Ticket #2249) + */ + affRight = sqlite3ExprAffinity(pOrTerm->pExpr->pRight); + affLeft = sqlite3ExprAffinity(pOrTerm->pExpr->pLeft); + if( affRight!=0 && affRight!=affLeft ){ + okToChngToIN = 0; + }else{ + pOrTerm->wtFlags |= TERM_OR_OK; + } + } + } + } + + /* At this point, okToChngToIN is true if original pTerm satisfies + ** case 1. In that case, construct a new virtual term that is + ** pTerm converted into an IN operator. + */ + if( okToChngToIN ){ + Expr *pDup; /* A transient duplicate expression */ + ExprList *pList = 0; /* The RHS of the IN operator */ + Expr *pLeft = 0; /* The LHS of the IN operator */ + Expr *pNew; /* The complete IN operator */ + + for(i=pOrWc->nTerm-1, pOrTerm=pOrWc->a; i>=0; i--, pOrTerm++){ + if( (pOrTerm->wtFlags & TERM_OR_OK)==0 ) continue; + assert( pOrTerm->eOperator & WO_EQ ); + assert( pOrTerm->leftCursor==iCursor ); + assert( pOrTerm->u.leftColumn==iColumn ); + pDup = sqlite3ExprDup(db, pOrTerm->pExpr->pRight, 0); + pList = sqlite3ExprListAppend(pWInfo->pParse, pList, pDup); + pLeft = pOrTerm->pExpr->pLeft; + } + assert( pLeft!=0 ); + pDup = sqlite3ExprDup(db, pLeft, 0); + pNew = sqlite3PExpr(pParse, TK_IN, pDup, 0); + if( pNew ){ + int idxNew; + transferJoinMarkings(pNew, pExpr); + assert( !ExprHasProperty(pNew, EP_xIsSelect) ); + pNew->x.pList = pList; + idxNew = whereClauseInsert(pWC, pNew, TERM_VIRTUAL|TERM_DYNAMIC); + testcase( idxNew==0 ); + exprAnalyze(pSrc, pWC, idxNew); + /* pTerm = &pWC->a[idxTerm]; // would be needed if pTerm where used again */ + markTermAsChild(pWC, idxNew, idxTerm); + }else{ + sqlite3ExprListDelete(db, pList); + } + } + } +} +#endif /* !SQLITE_OMIT_OR_OPTIMIZATION && !SQLITE_OMIT_SUBQUERY */ + +/* +** We already know that pExpr is a binary operator where both operands are +** column references. This routine checks to see if pExpr is an equivalence +** relation: +** 1. The SQLITE_Transitive optimization must be enabled +** 2. Must be either an == or an IS operator +** 3. Not originating in the ON clause of an OUTER JOIN +** 4. The affinities of A and B must be compatible +** 5a. Both operands use the same collating sequence OR +** 5b. The overall collating sequence is BINARY +** If this routine returns TRUE, that means that the RHS can be substituted +** for the LHS anyplace else in the WHERE clause where the LHS column occurs. +** This is an optimization. No harm comes from returning 0. But if 1 is +** returned when it should not be, then incorrect answers might result. +*/ +static int termIsEquivalence(Parse *pParse, Expr *pExpr){ + char aff1, aff2; + CollSeq *pColl; + if( !OptimizationEnabled(pParse->db, SQLITE_Transitive) ) return 0; + if( pExpr->op!=TK_EQ && pExpr->op!=TK_IS ) return 0; + if( ExprHasProperty(pExpr, EP_FromJoin) ) return 0; + aff1 = sqlite3ExprAffinity(pExpr->pLeft); + aff2 = sqlite3ExprAffinity(pExpr->pRight); + if( aff1!=aff2 + && (!sqlite3IsNumericAffinity(aff1) || !sqlite3IsNumericAffinity(aff2)) + ){ + return 0; + } + pColl = sqlite3BinaryCompareCollSeq(pParse, pExpr->pLeft, pExpr->pRight); + if( sqlite3IsBinary(pColl) ) return 1; + return sqlite3ExprCollSeqMatch(pParse, pExpr->pLeft, pExpr->pRight); +} + +/* +** Recursively walk the expressions of a SELECT statement and generate +** a bitmask indicating which tables are used in that expression +** tree. +*/ +static Bitmask exprSelectUsage(WhereMaskSet *pMaskSet, Select *pS){ + Bitmask mask = 0; + while( pS ){ + SrcList *pSrc = pS->pSrc; + mask |= sqlite3WhereExprListUsage(pMaskSet, pS->pEList); + mask |= sqlite3WhereExprListUsage(pMaskSet, pS->pGroupBy); + mask |= sqlite3WhereExprListUsage(pMaskSet, pS->pOrderBy); + mask |= sqlite3WhereExprUsage(pMaskSet, pS->pWhere); + mask |= sqlite3WhereExprUsage(pMaskSet, pS->pHaving); + if( ALWAYS(pSrc!=0) ){ + int i; + for(i=0; inSrc; i++){ + mask |= exprSelectUsage(pMaskSet, pSrc->a[i].pSelect); + mask |= sqlite3WhereExprUsage(pMaskSet, pSrc->a[i].pOn); + if( pSrc->a[i].fg.isTabFunc ){ + mask |= sqlite3WhereExprListUsage(pMaskSet, pSrc->a[i].u1.pFuncArg); + } + } + } + pS = pS->pPrior; + } + return mask; +} + +/* +** Expression pExpr is one operand of a comparison operator that might +** be useful for indexing. This routine checks to see if pExpr appears +** in any index. Return TRUE (1) if pExpr is an indexed term and return +** FALSE (0) if not. If TRUE is returned, also set aiCurCol[0] to the cursor +** number of the table that is indexed and aiCurCol[1] to the column number +** of the column that is indexed, or XN_EXPR (-2) if an expression is being +** indexed. +** +** If pExpr is a TK_COLUMN column reference, then this routine always returns +** true even if that particular column is not indexed, because the column +** might be added to an automatic index later. +*/ +static SQLITE_NOINLINE int exprMightBeIndexed2( + SrcList *pFrom, /* The FROM clause */ + Bitmask mPrereq, /* Bitmask of FROM clause terms referenced by pExpr */ + int *aiCurCol, /* Write the referenced table cursor and column here */ + Expr *pExpr /* An operand of a comparison operator */ +){ + Index *pIdx; + int i; + int iCur; + for(i=0; mPrereq>1; i++, mPrereq>>=1){} + iCur = pFrom->a[i].iCursor; + for(pIdx=pFrom->a[i].pTab->pIndex; pIdx; pIdx=pIdx->pNext){ + if( pIdx->aColExpr==0 ) continue; + for(i=0; inKeyCol; i++){ + if( pIdx->aiColumn[i]!=XN_EXPR ) continue; + if( sqlite3ExprCompareSkip(pExpr, pIdx->aColExpr->a[i].pExpr, iCur)==0 ){ + aiCurCol[0] = iCur; + aiCurCol[1] = XN_EXPR; + return 1; + } + } + } + return 0; +} +static int exprMightBeIndexed( + SrcList *pFrom, /* The FROM clause */ + Bitmask mPrereq, /* Bitmask of FROM clause terms referenced by pExpr */ + int *aiCurCol, /* Write the referenced table cursor & column here */ + Expr *pExpr, /* An operand of a comparison operator */ + int op /* The specific comparison operator */ +){ + /* If this expression is a vector to the left or right of a + ** inequality constraint (>, <, >= or <=), perform the processing + ** on the first element of the vector. */ + assert( TK_GT+1==TK_LE && TK_GT+2==TK_LT && TK_GT+3==TK_GE ); + assert( TK_ISop==TK_VECTOR && (op>=TK_GT && ALWAYS(op<=TK_GE)) ){ + pExpr = pExpr->x.pList->a[0].pExpr; + } + + if( pExpr->op==TK_COLUMN ){ + aiCurCol[0] = pExpr->iTable; + aiCurCol[1] = pExpr->iColumn; + return 1; + } + if( mPrereq==0 ) return 0; /* No table references */ + if( (mPrereq&(mPrereq-1))!=0 ) return 0; /* Refs more than one table */ + return exprMightBeIndexed2(pFrom,mPrereq,aiCurCol,pExpr); +} + +/* +** The input to this routine is an WhereTerm structure with only the +** "pExpr" field filled in. The job of this routine is to analyze the +** subexpression and populate all the other fields of the WhereTerm +** structure. +** +** If the expression is of the form " X" it gets commuted +** to the standard form of "X ". +** +** If the expression is of the form "X Y" where both X and Y are +** columns, then the original expression is unchanged and a new virtual +** term of the form "Y X" is added to the WHERE clause and +** analyzed separately. The original term is marked with TERM_COPIED +** and the new term is marked with TERM_DYNAMIC (because it's pExpr +** needs to be freed with the WhereClause) and TERM_VIRTUAL (because it +** is a commuted copy of a prior term.) The original term has nChild=1 +** and the copy has idxParent set to the index of the original term. +*/ +static void exprAnalyze( + SrcList *pSrc, /* the FROM clause */ + WhereClause *pWC, /* the WHERE clause */ + int idxTerm /* Index of the term to be analyzed */ +){ + WhereInfo *pWInfo = pWC->pWInfo; /* WHERE clause processing context */ + WhereTerm *pTerm; /* The term to be analyzed */ + WhereMaskSet *pMaskSet; /* Set of table index masks */ + Expr *pExpr; /* The expression to be analyzed */ + Bitmask prereqLeft; /* Prerequesites of the pExpr->pLeft */ + Bitmask prereqAll; /* Prerequesites of pExpr */ + Bitmask extraRight = 0; /* Extra dependencies on LEFT JOIN */ + Expr *pStr1 = 0; /* RHS of LIKE/GLOB operator */ + int isComplete = 0; /* RHS of LIKE/GLOB ends with wildcard */ + int noCase = 0; /* uppercase equivalent to lowercase */ + int op; /* Top-level operator. pExpr->op */ + Parse *pParse = pWInfo->pParse; /* Parsing context */ + sqlite3 *db = pParse->db; /* Database connection */ + unsigned char eOp2 = 0; /* op2 value for LIKE/REGEXP/GLOB */ + int nLeft; /* Number of elements on left side vector */ + + if( db->mallocFailed ){ + return; + } + pTerm = &pWC->a[idxTerm]; + pMaskSet = &pWInfo->sMaskSet; + pExpr = pTerm->pExpr; + assert( pExpr->op!=TK_AS && pExpr->op!=TK_COLLATE ); + prereqLeft = sqlite3WhereExprUsage(pMaskSet, pExpr->pLeft); + op = pExpr->op; + if( op==TK_IN ){ + assert( pExpr->pRight==0 ); + if( sqlite3ExprCheckIN(pParse, pExpr) ) return; + if( ExprHasProperty(pExpr, EP_xIsSelect) ){ + pTerm->prereqRight = exprSelectUsage(pMaskSet, pExpr->x.pSelect); + }else{ + pTerm->prereqRight = sqlite3WhereExprListUsage(pMaskSet, pExpr->x.pList); + } + }else if( op==TK_ISNULL ){ + pTerm->prereqRight = 0; + }else{ + pTerm->prereqRight = sqlite3WhereExprUsage(pMaskSet, pExpr->pRight); + } + pMaskSet->bVarSelect = 0; + prereqAll = sqlite3WhereExprUsageNN(pMaskSet, pExpr); + if( pMaskSet->bVarSelect ) pTerm->wtFlags |= TERM_VARSELECT; + if( ExprHasProperty(pExpr, EP_FromJoin) ){ + Bitmask x = sqlite3WhereGetMask(pMaskSet, pExpr->iRightJoinTable); + prereqAll |= x; + extraRight = x-1; /* ON clause terms may not be used with an index + ** on left table of a LEFT JOIN. Ticket #3015 */ + if( (prereqAll>>1)>=x ){ + sqlite3ErrorMsg(pParse, "ON clause references tables to its right"); + return; + } + } + pTerm->prereqAll = prereqAll; + pTerm->leftCursor = -1; + pTerm->iParent = -1; + pTerm->eOperator = 0; + if( allowedOp(op) ){ + int aiCurCol[2]; + Expr *pLeft = sqlite3ExprSkipCollate(pExpr->pLeft); + Expr *pRight = sqlite3ExprSkipCollate(pExpr->pRight); + u16 opMask = (pTerm->prereqRight & prereqLeft)==0 ? WO_ALL : WO_EQUIV; + + if( pTerm->iField>0 ){ + assert( op==TK_IN ); + assert( pLeft->op==TK_VECTOR ); + pLeft = pLeft->x.pList->a[pTerm->iField-1].pExpr; + } + + if( exprMightBeIndexed(pSrc, prereqLeft, aiCurCol, pLeft, op) ){ + pTerm->leftCursor = aiCurCol[0]; + pTerm->u.leftColumn = aiCurCol[1]; + pTerm->eOperator = operatorMask(op) & opMask; + } + if( op==TK_IS ) pTerm->wtFlags |= TERM_IS; + if( pRight + && exprMightBeIndexed(pSrc, pTerm->prereqRight, aiCurCol, pRight, op) + ){ + WhereTerm *pNew; + Expr *pDup; + u16 eExtraOp = 0; /* Extra bits for pNew->eOperator */ + assert( pTerm->iField==0 ); + if( pTerm->leftCursor>=0 ){ + int idxNew; + pDup = sqlite3ExprDup(db, pExpr, 0); + if( db->mallocFailed ){ + sqlite3ExprDelete(db, pDup); + return; + } + idxNew = whereClauseInsert(pWC, pDup, TERM_VIRTUAL|TERM_DYNAMIC); + if( idxNew==0 ) return; + pNew = &pWC->a[idxNew]; + markTermAsChild(pWC, idxNew, idxTerm); + if( op==TK_IS ) pNew->wtFlags |= TERM_IS; + pTerm = &pWC->a[idxTerm]; + pTerm->wtFlags |= TERM_COPIED; + + if( termIsEquivalence(pParse, pDup) ){ + pTerm->eOperator |= WO_EQUIV; + eExtraOp = WO_EQUIV; + } + }else{ + pDup = pExpr; + pNew = pTerm; + } + pNew->wtFlags |= exprCommute(pParse, pDup); + pNew->leftCursor = aiCurCol[0]; + pNew->u.leftColumn = aiCurCol[1]; + testcase( (prereqLeft | extraRight) != prereqLeft ); + pNew->prereqRight = prereqLeft | extraRight; + pNew->prereqAll = prereqAll; + pNew->eOperator = (operatorMask(pDup->op) + eExtraOp) & opMask; + } + } + +#ifndef SQLITE_OMIT_BETWEEN_OPTIMIZATION + /* If a term is the BETWEEN operator, create two new virtual terms + ** that define the range that the BETWEEN implements. For example: + ** + ** a BETWEEN b AND c + ** + ** is converted into: + ** + ** (a BETWEEN b AND c) AND (a>=b) AND (a<=c) + ** + ** The two new terms are added onto the end of the WhereClause object. + ** The new terms are "dynamic" and are children of the original BETWEEN + ** term. That means that if the BETWEEN term is coded, the children are + ** skipped. Or, if the children are satisfied by an index, the original + ** BETWEEN term is skipped. + */ + else if( pExpr->op==TK_BETWEEN && pWC->op==TK_AND ){ + ExprList *pList = pExpr->x.pList; + int i; + static const u8 ops[] = {TK_GE, TK_LE}; + assert( pList!=0 ); + assert( pList->nExpr==2 ); + for(i=0; i<2; i++){ + Expr *pNewExpr; + int idxNew; + pNewExpr = sqlite3PExpr(pParse, ops[i], + sqlite3ExprDup(db, pExpr->pLeft, 0), + sqlite3ExprDup(db, pList->a[i].pExpr, 0)); + transferJoinMarkings(pNewExpr, pExpr); + idxNew = whereClauseInsert(pWC, pNewExpr, TERM_VIRTUAL|TERM_DYNAMIC); + testcase( idxNew==0 ); + exprAnalyze(pSrc, pWC, idxNew); + pTerm = &pWC->a[idxTerm]; + markTermAsChild(pWC, idxNew, idxTerm); + } + } +#endif /* SQLITE_OMIT_BETWEEN_OPTIMIZATION */ + +#if !defined(SQLITE_OMIT_OR_OPTIMIZATION) && !defined(SQLITE_OMIT_SUBQUERY) + /* Analyze a term that is composed of two or more subterms connected by + ** an OR operator. + */ + else if( pExpr->op==TK_OR ){ + assert( pWC->op==TK_AND ); + exprAnalyzeOrTerm(pSrc, pWC, idxTerm); + pTerm = &pWC->a[idxTerm]; + } +#endif /* SQLITE_OMIT_OR_OPTIMIZATION */ + +#ifndef SQLITE_OMIT_LIKE_OPTIMIZATION + /* Add constraints to reduce the search space on a LIKE or GLOB + ** operator. + ** + ** A like pattern of the form "x LIKE 'aBc%'" is changed into constraints + ** + ** x>='ABC' AND x<'abd' AND x LIKE 'aBc%' + ** + ** The last character of the prefix "abc" is incremented to form the + ** termination condition "abd". If case is not significant (the default + ** for LIKE) then the lower-bound is made all uppercase and the upper- + ** bound is made all lowercase so that the bounds also work when comparing + ** BLOBs. + */ + if( pWC->op==TK_AND + && isLikeOrGlob(pParse, pExpr, &pStr1, &isComplete, &noCase) + ){ + Expr *pLeft; /* LHS of LIKE/GLOB operator */ + Expr *pStr2; /* Copy of pStr1 - RHS of LIKE/GLOB operator */ + Expr *pNewExpr1; + Expr *pNewExpr2; + int idxNew1; + int idxNew2; + const char *zCollSeqName; /* Name of collating sequence */ + const u16 wtFlags = TERM_LIKEOPT | TERM_VIRTUAL | TERM_DYNAMIC; + + pLeft = pExpr->x.pList->a[1].pExpr; + pStr2 = sqlite3ExprDup(db, pStr1, 0); + + /* Convert the lower bound to upper-case and the upper bound to + ** lower-case (upper-case is less than lower-case in ASCII) so that + ** the range constraints also work for BLOBs + */ + if( noCase && !pParse->db->mallocFailed ){ + int i; + char c; + pTerm->wtFlags |= TERM_LIKE; + for(i=0; (c = pStr1->u.zToken[i])!=0; i++){ + pStr1->u.zToken[i] = sqlite3Toupper(c); + pStr2->u.zToken[i] = sqlite3Tolower(c); + } + } + + if( !db->mallocFailed ){ + u8 c, *pC; /* Last character before the first wildcard */ + pC = (u8*)&pStr2->u.zToken[sqlite3Strlen30(pStr2->u.zToken)-1]; + c = *pC; + if( noCase ){ + /* The point is to increment the last character before the first + ** wildcard. But if we increment '@', that will push it into the + ** alphabetic range where case conversions will mess up the + ** inequality. To avoid this, make sure to also run the full + ** LIKE on all candidate expressions by clearing the isComplete flag + */ + if( c=='A'-1 ) isComplete = 0; + c = sqlite3UpperToLower[c]; + } + *pC = c + 1; + } + zCollSeqName = noCase ? "NOCASE" : sqlite3StrBINARY; + pNewExpr1 = sqlite3ExprDup(db, pLeft, 0); + pNewExpr1 = sqlite3PExpr(pParse, TK_GE, + sqlite3ExprAddCollateString(pParse,pNewExpr1,zCollSeqName), + pStr1); + transferJoinMarkings(pNewExpr1, pExpr); + idxNew1 = whereClauseInsert(pWC, pNewExpr1, wtFlags); + testcase( idxNew1==0 ); + exprAnalyze(pSrc, pWC, idxNew1); + pNewExpr2 = sqlite3ExprDup(db, pLeft, 0); + pNewExpr2 = sqlite3PExpr(pParse, TK_LT, + sqlite3ExprAddCollateString(pParse,pNewExpr2,zCollSeqName), + pStr2); + transferJoinMarkings(pNewExpr2, pExpr); + idxNew2 = whereClauseInsert(pWC, pNewExpr2, wtFlags); + testcase( idxNew2==0 ); + exprAnalyze(pSrc, pWC, idxNew2); + pTerm = &pWC->a[idxTerm]; + if( isComplete ){ + markTermAsChild(pWC, idxNew1, idxTerm); + markTermAsChild(pWC, idxNew2, idxTerm); + } + } +#endif /* SQLITE_OMIT_LIKE_OPTIMIZATION */ + +#ifndef SQLITE_OMIT_VIRTUALTABLE + /* Add a WO_AUX auxiliary term to the constraint set if the + ** current expression is of the form "column OP expr" where OP + ** is an operator that gets passed into virtual tables but which is + ** not normally optimized for ordinary tables. In other words, OP + ** is one of MATCH, LIKE, GLOB, REGEXP, !=, IS, IS NOT, or NOT NULL. + ** This information is used by the xBestIndex methods of + ** virtual tables. The native query optimizer does not attempt + ** to do anything with MATCH functions. + */ + if( pWC->op==TK_AND ){ + Expr *pRight = 0, *pLeft = 0; + int res = isAuxiliaryVtabOperator(db, pExpr, &eOp2, &pLeft, &pRight); + while( res-- > 0 ){ + int idxNew; + WhereTerm *pNewTerm; + Bitmask prereqColumn, prereqExpr; + + prereqExpr = sqlite3WhereExprUsage(pMaskSet, pRight); + prereqColumn = sqlite3WhereExprUsage(pMaskSet, pLeft); + if( (prereqExpr & prereqColumn)==0 ){ + Expr *pNewExpr; + pNewExpr = sqlite3PExpr(pParse, TK_MATCH, + 0, sqlite3ExprDup(db, pRight, 0)); + if( ExprHasProperty(pExpr, EP_FromJoin) && pNewExpr ){ + ExprSetProperty(pNewExpr, EP_FromJoin); + } + idxNew = whereClauseInsert(pWC, pNewExpr, TERM_VIRTUAL|TERM_DYNAMIC); + testcase( idxNew==0 ); + pNewTerm = &pWC->a[idxNew]; + pNewTerm->prereqRight = prereqExpr; + pNewTerm->leftCursor = pLeft->iTable; + pNewTerm->u.leftColumn = pLeft->iColumn; + pNewTerm->eOperator = WO_AUX; + pNewTerm->eMatchOp = eOp2; + markTermAsChild(pWC, idxNew, idxTerm); + pTerm = &pWC->a[idxTerm]; + pTerm->wtFlags |= TERM_COPIED; + pNewTerm->prereqAll = pTerm->prereqAll; + } + SWAP(Expr*, pLeft, pRight); + } + } +#endif /* SQLITE_OMIT_VIRTUALTABLE */ + + /* If there is a vector == or IS term - e.g. "(a, b) == (?, ?)" - create + ** new terms for each component comparison - "a = ?" and "b = ?". The + ** new terms completely replace the original vector comparison, which is + ** no longer used. + ** + ** This is only required if at least one side of the comparison operation + ** is not a sub-select. */ + if( pWC->op==TK_AND + && (pExpr->op==TK_EQ || pExpr->op==TK_IS) + && (nLeft = sqlite3ExprVectorSize(pExpr->pLeft))>1 + && sqlite3ExprVectorSize(pExpr->pRight)==nLeft + && ( (pExpr->pLeft->flags & EP_xIsSelect)==0 + || (pExpr->pRight->flags & EP_xIsSelect)==0) + ){ + int i; + for(i=0; ipLeft, i); + Expr *pRight = sqlite3ExprForVectorField(pParse, pExpr->pRight, i); + + pNew = sqlite3PExpr(pParse, pExpr->op, pLeft, pRight); + transferJoinMarkings(pNew, pExpr); + idxNew = whereClauseInsert(pWC, pNew, TERM_DYNAMIC); + exprAnalyze(pSrc, pWC, idxNew); + } + pTerm = &pWC->a[idxTerm]; + pTerm->wtFlags |= TERM_CODED|TERM_VIRTUAL; /* Disable the original */ + pTerm->eOperator = 0; + } + + /* If there is a vector IN term - e.g. "(a, b) IN (SELECT ...)" - create + ** a virtual term for each vector component. The expression object + ** used by each such virtual term is pExpr (the full vector IN(...) + ** expression). The WhereTerm.iField variable identifies the index within + ** the vector on the LHS that the virtual term represents. + ** + ** This only works if the RHS is a simple SELECT, not a compound + */ + if( pWC->op==TK_AND && pExpr->op==TK_IN && pTerm->iField==0 + && pExpr->pLeft->op==TK_VECTOR + && pExpr->x.pSelect->pPrior==0 + ){ + int i; + for(i=0; ipLeft); i++){ + int idxNew; + idxNew = whereClauseInsert(pWC, pExpr, TERM_VIRTUAL); + pWC->a[idxNew].iField = i+1; + exprAnalyze(pSrc, pWC, idxNew); + markTermAsChild(pWC, idxNew, idxTerm); + } + } + +#ifdef SQLITE_ENABLE_STAT4 + /* When sqlite_stat4 histogram data is available an operator of the + ** form "x IS NOT NULL" can sometimes be evaluated more efficiently + ** as "x>NULL" if x is not an INTEGER PRIMARY KEY. So construct a + ** virtual term of that form. + ** + ** Note that the virtual term must be tagged with TERM_VNULL. + */ + if( pExpr->op==TK_NOTNULL + && pExpr->pLeft->op==TK_COLUMN + && pExpr->pLeft->iColumn>=0 + && !ExprHasProperty(pExpr, EP_FromJoin) + && OptimizationEnabled(db, SQLITE_Stat4) + ){ + Expr *pNewExpr; + Expr *pLeft = pExpr->pLeft; + int idxNew; + WhereTerm *pNewTerm; + + pNewExpr = sqlite3PExpr(pParse, TK_GT, + sqlite3ExprDup(db, pLeft, 0), + sqlite3ExprAlloc(db, TK_NULL, 0, 0)); + + idxNew = whereClauseInsert(pWC, pNewExpr, + TERM_VIRTUAL|TERM_DYNAMIC|TERM_VNULL); + if( idxNew ){ + pNewTerm = &pWC->a[idxNew]; + pNewTerm->prereqRight = 0; + pNewTerm->leftCursor = pLeft->iTable; + pNewTerm->u.leftColumn = pLeft->iColumn; + pNewTerm->eOperator = WO_GT; + markTermAsChild(pWC, idxNew, idxTerm); + pTerm = &pWC->a[idxTerm]; + pTerm->wtFlags |= TERM_COPIED; + pNewTerm->prereqAll = pTerm->prereqAll; + } + } +#endif /* SQLITE_ENABLE_STAT4 */ + + /* Prevent ON clause terms of a LEFT JOIN from being used to drive + ** an index for tables to the left of the join. + */ + testcase( pTerm!=&pWC->a[idxTerm] ); + pTerm = &pWC->a[idxTerm]; + pTerm->prereqRight |= extraRight; +} + +/*************************************************************************** +** Routines with file scope above. Interface to the rest of the where.c +** subsystem follows. +***************************************************************************/ + +/* +** This routine identifies subexpressions in the WHERE clause where +** each subexpression is separated by the AND operator or some other +** operator specified in the op parameter. The WhereClause structure +** is filled with pointers to subexpressions. For example: +** +** WHERE a=='hello' AND coalesce(b,11)<10 AND (c+12!=d OR c==22) +** \________/ \_______________/ \________________/ +** slot[0] slot[1] slot[2] +** +** The original WHERE clause in pExpr is unaltered. All this routine +** does is make slot[] entries point to substructure within pExpr. +** +** In the previous sentence and in the diagram, "slot[]" refers to +** the WhereClause.a[] array. The slot[] array grows as needed to contain +** all terms of the WHERE clause. +*/ +SQLITE_PRIVATE void sqlite3WhereSplit(WhereClause *pWC, Expr *pExpr, u8 op){ + Expr *pE2 = sqlite3ExprSkipCollateAndLikely(pExpr); + pWC->op = op; + if( pE2==0 ) return; + if( pE2->op!=op ){ + whereClauseInsert(pWC, pExpr, 0); + }else{ + sqlite3WhereSplit(pWC, pE2->pLeft, op); + sqlite3WhereSplit(pWC, pE2->pRight, op); + } +} + +/* +** Initialize a preallocated WhereClause structure. +*/ +SQLITE_PRIVATE void sqlite3WhereClauseInit( + WhereClause *pWC, /* The WhereClause to be initialized */ + WhereInfo *pWInfo /* The WHERE processing context */ +){ + pWC->pWInfo = pWInfo; + pWC->hasOr = 0; + pWC->pOuter = 0; + pWC->nTerm = 0; + pWC->nSlot = ArraySize(pWC->aStatic); + pWC->a = pWC->aStatic; +} + +/* +** Deallocate a WhereClause structure. The WhereClause structure +** itself is not freed. This routine is the inverse of +** sqlite3WhereClauseInit(). +*/ +SQLITE_PRIVATE void sqlite3WhereClauseClear(WhereClause *pWC){ + int i; + WhereTerm *a; + sqlite3 *db = pWC->pWInfo->pParse->db; + for(i=pWC->nTerm-1, a=pWC->a; i>=0; i--, a++){ + if( a->wtFlags & TERM_DYNAMIC ){ + sqlite3ExprDelete(db, a->pExpr); + } + if( a->wtFlags & TERM_ORINFO ){ + whereOrInfoDelete(db, a->u.pOrInfo); + }else if( a->wtFlags & TERM_ANDINFO ){ + whereAndInfoDelete(db, a->u.pAndInfo); + } + } + if( pWC->a!=pWC->aStatic ){ + sqlite3DbFree(db, pWC->a); + } +} + + +/* +** These routines walk (recursively) an expression tree and generate +** a bitmask indicating which tables are used in that expression +** tree. +*/ +SQLITE_PRIVATE Bitmask sqlite3WhereExprUsageNN(WhereMaskSet *pMaskSet, Expr *p){ + Bitmask mask; + if( p->op==TK_COLUMN && !ExprHasProperty(p, EP_FixedCol) ){ + return sqlite3WhereGetMask(pMaskSet, p->iTable); + }else if( ExprHasProperty(p, EP_TokenOnly|EP_Leaf) ){ + assert( p->op!=TK_IF_NULL_ROW ); + return 0; + } + mask = (p->op==TK_IF_NULL_ROW) ? sqlite3WhereGetMask(pMaskSet, p->iTable) : 0; + if( p->pLeft ) mask |= sqlite3WhereExprUsageNN(pMaskSet, p->pLeft); + if( p->pRight ){ + mask |= sqlite3WhereExprUsageNN(pMaskSet, p->pRight); + assert( p->x.pList==0 ); + }else if( ExprHasProperty(p, EP_xIsSelect) ){ + if( ExprHasProperty(p, EP_VarSelect) ) pMaskSet->bVarSelect = 1; + mask |= exprSelectUsage(pMaskSet, p->x.pSelect); + }else if( p->x.pList ){ + mask |= sqlite3WhereExprListUsage(pMaskSet, p->x.pList); + } +#ifndef SQLITE_OMIT_WINDOWFUNC + if( p->op==TK_FUNCTION && p->y.pWin ){ + mask |= sqlite3WhereExprListUsage(pMaskSet, p->y.pWin->pPartition); + mask |= sqlite3WhereExprListUsage(pMaskSet, p->y.pWin->pOrderBy); + } +#endif + return mask; +} +SQLITE_PRIVATE Bitmask sqlite3WhereExprUsage(WhereMaskSet *pMaskSet, Expr *p){ + return p ? sqlite3WhereExprUsageNN(pMaskSet,p) : 0; +} +SQLITE_PRIVATE Bitmask sqlite3WhereExprListUsage(WhereMaskSet *pMaskSet, ExprList *pList){ + int i; + Bitmask mask = 0; + if( pList ){ + for(i=0; inExpr; i++){ + mask |= sqlite3WhereExprUsage(pMaskSet, pList->a[i].pExpr); + } + } + return mask; +} + + +/* +** Call exprAnalyze on all terms in a WHERE clause. +** +** Note that exprAnalyze() might add new virtual terms onto the +** end of the WHERE clause. We do not want to analyze these new +** virtual terms, so start analyzing at the end and work forward +** so that the added virtual terms are never processed. +*/ +SQLITE_PRIVATE void sqlite3WhereExprAnalyze( + SrcList *pTabList, /* the FROM clause */ + WhereClause *pWC /* the WHERE clause to be analyzed */ +){ + int i; + for(i=pWC->nTerm-1; i>=0; i--){ + exprAnalyze(pTabList, pWC, i); + } +} + +/* +** For table-valued-functions, transform the function arguments into +** new WHERE clause terms. +** +** Each function argument translates into an equality constraint against +** a HIDDEN column in the table. +*/ +SQLITE_PRIVATE void sqlite3WhereTabFuncArgs( + Parse *pParse, /* Parsing context */ + struct SrcList_item *pItem, /* The FROM clause term to process */ + WhereClause *pWC /* Xfer function arguments to here */ +){ + Table *pTab; + int j, k; + ExprList *pArgs; + Expr *pColRef; + Expr *pTerm; + if( pItem->fg.isTabFunc==0 ) return; + pTab = pItem->pTab; + assert( pTab!=0 ); + pArgs = pItem->u1.pFuncArg; + if( pArgs==0 ) return; + for(j=k=0; jnExpr; j++){ + Expr *pRhs; + while( knCol && (pTab->aCol[k].colFlags & COLFLAG_HIDDEN)==0 ){k++;} + if( k>=pTab->nCol ){ + sqlite3ErrorMsg(pParse, "too many arguments on %s() - max %d", + pTab->zName, j); + return; + } + pColRef = sqlite3ExprAlloc(pParse->db, TK_COLUMN, 0, 0); + if( pColRef==0 ) return; + pColRef->iTable = pItem->iCursor; + pColRef->iColumn = k++; + pColRef->y.pTab = pTab; + pRhs = sqlite3PExpr(pParse, TK_UPLUS, + sqlite3ExprDup(pParse->db, pArgs->a[j].pExpr, 0), 0); + pTerm = sqlite3PExpr(pParse, TK_EQ, pColRef, pRhs); + whereClauseInsert(pWC, pTerm, TERM_DYNAMIC); + } +} + +/************** End of whereexpr.c *******************************************/ +/************** Begin file where.c *******************************************/ +/* +** 2001 September 15 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** This module contains C code that generates VDBE code used to process +** the WHERE clause of SQL statements. This module is responsible for +** generating the code that loops through a table looking for applicable +** rows. Indices are selected and used to speed the search when doing +** so is applicable. Because this module is responsible for selecting +** indices, you might also think of this module as the "query optimizer". +*/ +/* #include "sqliteInt.h" */ +/* #include "whereInt.h" */ + +/* +** Extra information appended to the end of sqlite3_index_info but not +** visible to the xBestIndex function, at least not directly. The +** sqlite3_vtab_collation() interface knows how to reach it, however. +** +** This object is not an API and can be changed from one release to the +** next. As long as allocateIndexInfo() and sqlite3_vtab_collation() +** agree on the structure, all will be well. +*/ +typedef struct HiddenIndexInfo HiddenIndexInfo; +struct HiddenIndexInfo { + WhereClause *pWC; /* The Where clause being analyzed */ + Parse *pParse; /* The parsing context */ +}; + +/* Forward declaration of methods */ +static int whereLoopResize(sqlite3*, WhereLoop*, int); + +/* Test variable that can be set to enable WHERE tracing */ +#if defined(SQLITE_TEST) || defined(SQLITE_DEBUG) +/***/ int sqlite3WhereTrace = 0; +#endif + + +/* +** Return the estimated number of output rows from a WHERE clause +*/ +SQLITE_PRIVATE LogEst sqlite3WhereOutputRowCount(WhereInfo *pWInfo){ + return pWInfo->nRowOut; +} + +/* +** Return one of the WHERE_DISTINCT_xxxxx values to indicate how this +** WHERE clause returns outputs for DISTINCT processing. +*/ +SQLITE_PRIVATE int sqlite3WhereIsDistinct(WhereInfo *pWInfo){ + return pWInfo->eDistinct; +} + +/* +** Return TRUE if the WHERE clause returns rows in ORDER BY order. +** Return FALSE if the output needs to be sorted. +*/ +SQLITE_PRIVATE int sqlite3WhereIsOrdered(WhereInfo *pWInfo){ + return pWInfo->nOBSat; +} + +/* +** In the ORDER BY LIMIT optimization, if the inner-most loop is known +** to emit rows in increasing order, and if the last row emitted by the +** inner-most loop did not fit within the sorter, then we can skip all +** subsequent rows for the current iteration of the inner loop (because they +** will not fit in the sorter either) and continue with the second inner +** loop - the loop immediately outside the inner-most. +** +** When a row does not fit in the sorter (because the sorter already +** holds LIMIT+OFFSET rows that are smaller), then a jump is made to the +** label returned by this function. +** +** If the ORDER BY LIMIT optimization applies, the jump destination should +** be the continuation for the second-inner-most loop. If the ORDER BY +** LIMIT optimization does not apply, then the jump destination should +** be the continuation for the inner-most loop. +** +** It is always safe for this routine to return the continuation of the +** inner-most loop, in the sense that a correct answer will result. +** Returning the continuation the second inner loop is an optimization +** that might make the code run a little faster, but should not change +** the final answer. +*/ +SQLITE_PRIVATE int sqlite3WhereOrderByLimitOptLabel(WhereInfo *pWInfo){ + WhereLevel *pInner; + if( !pWInfo->bOrderedInnerLoop ){ + /* The ORDER BY LIMIT optimization does not apply. Jump to the + ** continuation of the inner-most loop. */ + return pWInfo->iContinue; + } + pInner = &pWInfo->a[pWInfo->nLevel-1]; + assert( pInner->addrNxt!=0 ); + return pInner->addrNxt; +} + +/* +** Return the VDBE address or label to jump to in order to continue +** immediately with the next row of a WHERE clause. +*/ +SQLITE_PRIVATE int sqlite3WhereContinueLabel(WhereInfo *pWInfo){ + assert( pWInfo->iContinue!=0 ); + return pWInfo->iContinue; +} + +/* +** Return the VDBE address or label to jump to in order to break +** out of a WHERE loop. +*/ +SQLITE_PRIVATE int sqlite3WhereBreakLabel(WhereInfo *pWInfo){ + return pWInfo->iBreak; +} + +/* +** Return ONEPASS_OFF (0) if an UPDATE or DELETE statement is unable to +** operate directly on the rowis returned by a WHERE clause. Return +** ONEPASS_SINGLE (1) if the statement can operation directly because only +** a single row is to be changed. Return ONEPASS_MULTI (2) if the one-pass +** optimization can be used on multiple +** +** If the ONEPASS optimization is used (if this routine returns true) +** then also write the indices of open cursors used by ONEPASS +** into aiCur[0] and aiCur[1]. iaCur[0] gets the cursor of the data +** table and iaCur[1] gets the cursor used by an auxiliary index. +** Either value may be -1, indicating that cursor is not used. +** Any cursors returned will have been opened for writing. +** +** aiCur[0] and aiCur[1] both get -1 if the where-clause logic is +** unable to use the ONEPASS optimization. +*/ +SQLITE_PRIVATE int sqlite3WhereOkOnePass(WhereInfo *pWInfo, int *aiCur){ + memcpy(aiCur, pWInfo->aiCurOnePass, sizeof(int)*2); +#ifdef WHERETRACE_ENABLED + if( sqlite3WhereTrace && pWInfo->eOnePass!=ONEPASS_OFF ){ + sqlite3DebugPrintf("%s cursors: %d %d\n", + pWInfo->eOnePass==ONEPASS_SINGLE ? "ONEPASS_SINGLE" : "ONEPASS_MULTI", + aiCur[0], aiCur[1]); + } +#endif + return pWInfo->eOnePass; +} + +/* +** Move the content of pSrc into pDest +*/ +static void whereOrMove(WhereOrSet *pDest, WhereOrSet *pSrc){ + pDest->n = pSrc->n; + memcpy(pDest->a, pSrc->a, pDest->n*sizeof(pDest->a[0])); +} + +/* +** Try to insert a new prerequisite/cost entry into the WhereOrSet pSet. +** +** The new entry might overwrite an existing entry, or it might be +** appended, or it might be discarded. Do whatever is the right thing +** so that pSet keeps the N_OR_COST best entries seen so far. +*/ +static int whereOrInsert( + WhereOrSet *pSet, /* The WhereOrSet to be updated */ + Bitmask prereq, /* Prerequisites of the new entry */ + LogEst rRun, /* Run-cost of the new entry */ + LogEst nOut /* Number of outputs for the new entry */ +){ + u16 i; + WhereOrCost *p; + for(i=pSet->n, p=pSet->a; i>0; i--, p++){ + if( rRun<=p->rRun && (prereq & p->prereq)==prereq ){ + goto whereOrInsert_done; + } + if( p->rRun<=rRun && (p->prereq & prereq)==p->prereq ){ + return 0; + } + } + if( pSet->na[pSet->n++]; + p->nOut = nOut; + }else{ + p = pSet->a; + for(i=1; in; i++){ + if( p->rRun>pSet->a[i].rRun ) p = pSet->a + i; + } + if( p->rRun<=rRun ) return 0; + } +whereOrInsert_done: + p->prereq = prereq; + p->rRun = rRun; + if( p->nOut>nOut ) p->nOut = nOut; + return 1; +} + +/* +** Return the bitmask for the given cursor number. Return 0 if +** iCursor is not in the set. +*/ +SQLITE_PRIVATE Bitmask sqlite3WhereGetMask(WhereMaskSet *pMaskSet, int iCursor){ + int i; + assert( pMaskSet->n<=(int)sizeof(Bitmask)*8 ); + for(i=0; in; i++){ + if( pMaskSet->ix[i]==iCursor ){ + return MASKBIT(i); + } + } + return 0; +} + +/* +** Create a new mask for cursor iCursor. +** +** There is one cursor per table in the FROM clause. The number of +** tables in the FROM clause is limited by a test early in the +** sqlite3WhereBegin() routine. So we know that the pMaskSet->ix[] +** array will never overflow. +*/ +static void createMask(WhereMaskSet *pMaskSet, int iCursor){ + assert( pMaskSet->n < ArraySize(pMaskSet->ix) ); + pMaskSet->ix[pMaskSet->n++] = iCursor; +} + +/* +** Advance to the next WhereTerm that matches according to the criteria +** established when the pScan object was initialized by whereScanInit(). +** Return NULL if there are no more matching WhereTerms. +*/ +static WhereTerm *whereScanNext(WhereScan *pScan){ + int iCur; /* The cursor on the LHS of the term */ + i16 iColumn; /* The column on the LHS of the term. -1 for IPK */ + Expr *pX; /* An expression being tested */ + WhereClause *pWC; /* Shorthand for pScan->pWC */ + WhereTerm *pTerm; /* The term being tested */ + int k = pScan->k; /* Where to start scanning */ + + assert( pScan->iEquiv<=pScan->nEquiv ); + pWC = pScan->pWC; + while(1){ + iColumn = pScan->aiColumn[pScan->iEquiv-1]; + iCur = pScan->aiCur[pScan->iEquiv-1]; + assert( pWC!=0 ); + do{ + for(pTerm=pWC->a+k; knTerm; k++, pTerm++){ + if( pTerm->leftCursor==iCur + && pTerm->u.leftColumn==iColumn + && (iColumn!=XN_EXPR + || sqlite3ExprCompareSkip(pTerm->pExpr->pLeft, + pScan->pIdxExpr,iCur)==0) + && (pScan->iEquiv<=1 || !ExprHasProperty(pTerm->pExpr, EP_FromJoin)) + ){ + if( (pTerm->eOperator & WO_EQUIV)!=0 + && pScan->nEquivaiCur) + && (pX = sqlite3ExprSkipCollateAndLikely(pTerm->pExpr->pRight))->op + ==TK_COLUMN + ){ + int j; + for(j=0; jnEquiv; j++){ + if( pScan->aiCur[j]==pX->iTable + && pScan->aiColumn[j]==pX->iColumn ){ + break; + } + } + if( j==pScan->nEquiv ){ + pScan->aiCur[j] = pX->iTable; + pScan->aiColumn[j] = pX->iColumn; + pScan->nEquiv++; + } + } + if( (pTerm->eOperator & pScan->opMask)!=0 ){ + /* Verify the affinity and collating sequence match */ + if( pScan->zCollName && (pTerm->eOperator & WO_ISNULL)==0 ){ + CollSeq *pColl; + Parse *pParse = pWC->pWInfo->pParse; + pX = pTerm->pExpr; + if( !sqlite3IndexAffinityOk(pX, pScan->idxaff) ){ + continue; + } + assert(pX->pLeft); + pColl = sqlite3BinaryCompareCollSeq(pParse, + pX->pLeft, pX->pRight); + if( pColl==0 ) pColl = pParse->db->pDfltColl; + if( sqlite3StrICmp(pColl->zName, pScan->zCollName) ){ + continue; + } + } + if( (pTerm->eOperator & (WO_EQ|WO_IS))!=0 + && (pX = pTerm->pExpr->pRight)->op==TK_COLUMN + && pX->iTable==pScan->aiCur[0] + && pX->iColumn==pScan->aiColumn[0] + ){ + testcase( pTerm->eOperator & WO_IS ); + continue; + } + pScan->pWC = pWC; + pScan->k = k+1; + return pTerm; + } + } + } + pWC = pWC->pOuter; + k = 0; + }while( pWC!=0 ); + if( pScan->iEquiv>=pScan->nEquiv ) break; + pWC = pScan->pOrigWC; + k = 0; + pScan->iEquiv++; + } + return 0; +} + +/* +** This is whereScanInit() for the case of an index on an expression. +** It is factored out into a separate tail-recursion subroutine so that +** the normal whereScanInit() routine, which is a high-runner, does not +** need to push registers onto the stack as part of its prologue. +*/ +static SQLITE_NOINLINE WhereTerm *whereScanInitIndexExpr(WhereScan *pScan){ + pScan->idxaff = sqlite3ExprAffinity(pScan->pIdxExpr); + return whereScanNext(pScan); +} + +/* +** Initialize a WHERE clause scanner object. Return a pointer to the +** first match. Return NULL if there are no matches. +** +** The scanner will be searching the WHERE clause pWC. It will look +** for terms of the form "X " where X is column iColumn of table +** iCur. Or if pIdx!=0 then X is column iColumn of index pIdx. pIdx +** must be one of the indexes of table iCur. +** +** The must be one of the operators described by opMask. +** +** If the search is for X and the WHERE clause contains terms of the +** form X=Y then this routine might also return terms of the form +** "Y ". The number of levels of transitivity is limited, +** but is enough to handle most commonly occurring SQL statements. +** +** If X is not the INTEGER PRIMARY KEY then X must be compatible with +** index pIdx. +*/ +static WhereTerm *whereScanInit( + WhereScan *pScan, /* The WhereScan object being initialized */ + WhereClause *pWC, /* The WHERE clause to be scanned */ + int iCur, /* Cursor to scan for */ + int iColumn, /* Column to scan for */ + u32 opMask, /* Operator(s) to scan for */ + Index *pIdx /* Must be compatible with this index */ +){ + pScan->pOrigWC = pWC; + pScan->pWC = pWC; + pScan->pIdxExpr = 0; + pScan->idxaff = 0; + pScan->zCollName = 0; + pScan->opMask = opMask; + pScan->k = 0; + pScan->aiCur[0] = iCur; + pScan->nEquiv = 1; + pScan->iEquiv = 1; + if( pIdx ){ + int j = iColumn; + iColumn = pIdx->aiColumn[j]; + if( iColumn==XN_EXPR ){ + pScan->pIdxExpr = pIdx->aColExpr->a[j].pExpr; + pScan->zCollName = pIdx->azColl[j]; + pScan->aiColumn[0] = XN_EXPR; + return whereScanInitIndexExpr(pScan); + }else if( iColumn==pIdx->pTable->iPKey ){ + iColumn = XN_ROWID; + }else if( iColumn>=0 ){ + pScan->idxaff = pIdx->pTable->aCol[iColumn].affinity; + pScan->zCollName = pIdx->azColl[j]; + } + }else if( iColumn==XN_EXPR ){ + return 0; + } + pScan->aiColumn[0] = iColumn; + return whereScanNext(pScan); +} + +/* +** Search for a term in the WHERE clause that is of the form "X " +** where X is a reference to the iColumn of table iCur or of index pIdx +** if pIdx!=0 and is one of the WO_xx operator codes specified by +** the op parameter. Return a pointer to the term. Return 0 if not found. +** +** If pIdx!=0 then it must be one of the indexes of table iCur. +** Search for terms matching the iColumn-th column of pIdx +** rather than the iColumn-th column of table iCur. +** +** The term returned might by Y= if there is another constraint in +** the WHERE clause that specifies that X=Y. Any such constraints will be +** identified by the WO_EQUIV bit in the pTerm->eOperator field. The +** aiCur[]/iaColumn[] arrays hold X and all its equivalents. There are 11 +** slots in aiCur[]/aiColumn[] so that means we can look for X plus up to 10 +** other equivalent values. Hence a search for X will return if X=A1 +** and A1=A2 and A2=A3 and ... and A9=A10 and A10=. +** +** If there are multiple terms in the WHERE clause of the form "X " +** then try for the one with no dependencies on - in other words where +** is a constant expression of some kind. Only return entries of +** the form "X Y" where Y is a column in another table if no terms of +** the form "X " exist. If no terms with a constant RHS +** exist, try to return a term that does not use WO_EQUIV. +*/ +SQLITE_PRIVATE WhereTerm *sqlite3WhereFindTerm( + WhereClause *pWC, /* The WHERE clause to be searched */ + int iCur, /* Cursor number of LHS */ + int iColumn, /* Column number of LHS */ + Bitmask notReady, /* RHS must not overlap with this mask */ + u32 op, /* Mask of WO_xx values describing operator */ + Index *pIdx /* Must be compatible with this index, if not NULL */ +){ + WhereTerm *pResult = 0; + WhereTerm *p; + WhereScan scan; + + p = whereScanInit(&scan, pWC, iCur, iColumn, op, pIdx); + op &= WO_EQ|WO_IS; + while( p ){ + if( (p->prereqRight & notReady)==0 ){ + if( p->prereqRight==0 && (p->eOperator&op)!=0 ){ + testcase( p->eOperator & WO_IS ); + return p; + } + if( pResult==0 ) pResult = p; + } + p = whereScanNext(&scan); + } + return pResult; +} + +/* +** This function searches pList for an entry that matches the iCol-th column +** of index pIdx. +** +** If such an expression is found, its index in pList->a[] is returned. If +** no expression is found, -1 is returned. +*/ +static int findIndexCol( + Parse *pParse, /* Parse context */ + ExprList *pList, /* Expression list to search */ + int iBase, /* Cursor for table associated with pIdx */ + Index *pIdx, /* Index to match column of */ + int iCol /* Column of index to match */ +){ + int i; + const char *zColl = pIdx->azColl[iCol]; + + for(i=0; inExpr; i++){ + Expr *p = sqlite3ExprSkipCollateAndLikely(pList->a[i].pExpr); + if( p->op==TK_COLUMN + && p->iColumn==pIdx->aiColumn[iCol] + && p->iTable==iBase + ){ + CollSeq *pColl = sqlite3ExprNNCollSeq(pParse, pList->a[i].pExpr); + if( 0==sqlite3StrICmp(pColl->zName, zColl) ){ + return i; + } + } + } + + return -1; +} + +/* +** Return TRUE if the iCol-th column of index pIdx is NOT NULL +*/ +static int indexColumnNotNull(Index *pIdx, int iCol){ + int j; + assert( pIdx!=0 ); + assert( iCol>=0 && iColnColumn ); + j = pIdx->aiColumn[iCol]; + if( j>=0 ){ + return pIdx->pTable->aCol[j].notNull; + }else if( j==(-1) ){ + return 1; + }else{ + assert( j==(-2) ); + return 0; /* Assume an indexed expression can always yield a NULL */ + + } +} + +/* +** Return true if the DISTINCT expression-list passed as the third argument +** is redundant. +** +** A DISTINCT list is redundant if any subset of the columns in the +** DISTINCT list are collectively unique and individually non-null. +*/ +static int isDistinctRedundant( + Parse *pParse, /* Parsing context */ + SrcList *pTabList, /* The FROM clause */ + WhereClause *pWC, /* The WHERE clause */ + ExprList *pDistinct /* The result set that needs to be DISTINCT */ +){ + Table *pTab; + Index *pIdx; + int i; + int iBase; + + /* If there is more than one table or sub-select in the FROM clause of + ** this query, then it will not be possible to show that the DISTINCT + ** clause is redundant. */ + if( pTabList->nSrc!=1 ) return 0; + iBase = pTabList->a[0].iCursor; + pTab = pTabList->a[0].pTab; + + /* If any of the expressions is an IPK column on table iBase, then return + ** true. Note: The (p->iTable==iBase) part of this test may be false if the + ** current SELECT is a correlated sub-query. + */ + for(i=0; inExpr; i++){ + Expr *p = sqlite3ExprSkipCollateAndLikely(pDistinct->a[i].pExpr); + if( p->op==TK_COLUMN && p->iTable==iBase && p->iColumn<0 ) return 1; + } + + /* Loop through all indices on the table, checking each to see if it makes + ** the DISTINCT qualifier redundant. It does so if: + ** + ** 1. The index is itself UNIQUE, and + ** + ** 2. All of the columns in the index are either part of the pDistinct + ** list, or else the WHERE clause contains a term of the form "col=X", + ** where X is a constant value. The collation sequences of the + ** comparison and select-list expressions must match those of the index. + ** + ** 3. All of those index columns for which the WHERE clause does not + ** contain a "col=X" term are subject to a NOT NULL constraint. + */ + for(pIdx=pTab->pIndex; pIdx; pIdx=pIdx->pNext){ + if( !IsUniqueIndex(pIdx) ) continue; + for(i=0; inKeyCol; i++){ + if( 0==sqlite3WhereFindTerm(pWC, iBase, i, ~(Bitmask)0, WO_EQ, pIdx) ){ + if( findIndexCol(pParse, pDistinct, iBase, pIdx, i)<0 ) break; + if( indexColumnNotNull(pIdx, i)==0 ) break; + } + } + if( i==pIdx->nKeyCol ){ + /* This index implies that the DISTINCT qualifier is redundant. */ + return 1; + } + } + + return 0; +} + + +/* +** Estimate the logarithm of the input value to base 2. +*/ +static LogEst estLog(LogEst N){ + return N<=10 ? 0 : sqlite3LogEst(N) - 33; +} + +/* +** Convert OP_Column opcodes to OP_Copy in previously generated code. +** +** This routine runs over generated VDBE code and translates OP_Column +** opcodes into OP_Copy when the table is being accessed via co-routine +** instead of via table lookup. +** +** If the iAutoidxCur is not zero, then any OP_Rowid instructions on +** cursor iTabCur are transformed into OP_Sequence opcode for the +** iAutoidxCur cursor, in order to generate unique rowids for the +** automatic index being generated. +*/ +static void translateColumnToCopy( + Parse *pParse, /* Parsing context */ + int iStart, /* Translate from this opcode to the end */ + int iTabCur, /* OP_Column/OP_Rowid references to this table */ + int iRegister, /* The first column is in this register */ + int iAutoidxCur /* If non-zero, cursor of autoindex being generated */ +){ + Vdbe *v = pParse->pVdbe; + VdbeOp *pOp = sqlite3VdbeGetOp(v, iStart); + int iEnd = sqlite3VdbeCurrentAddr(v); + if( pParse->db->mallocFailed ) return; + for(; iStartp1!=iTabCur ) continue; + if( pOp->opcode==OP_Column ){ + pOp->opcode = OP_Copy; + pOp->p1 = pOp->p2 + iRegister; + pOp->p2 = pOp->p3; + pOp->p3 = 0; + }else if( pOp->opcode==OP_Rowid ){ + if( iAutoidxCur ){ + pOp->opcode = OP_Sequence; + pOp->p1 = iAutoidxCur; + }else{ + pOp->opcode = OP_Null; + pOp->p1 = 0; + pOp->p3 = 0; + } + } + } +} + +/* +** Two routines for printing the content of an sqlite3_index_info +** structure. Used for testing and debugging only. If neither +** SQLITE_TEST or SQLITE_DEBUG are defined, then these routines +** are no-ops. +*/ +#if !defined(SQLITE_OMIT_VIRTUALTABLE) && defined(WHERETRACE_ENABLED) +static void TRACE_IDX_INPUTS(sqlite3_index_info *p){ + int i; + if( !sqlite3WhereTrace ) return; + for(i=0; inConstraint; i++){ + sqlite3DebugPrintf(" constraint[%d]: col=%d termid=%d op=%d usabled=%d\n", + i, + p->aConstraint[i].iColumn, + p->aConstraint[i].iTermOffset, + p->aConstraint[i].op, + p->aConstraint[i].usable); + } + for(i=0; inOrderBy; i++){ + sqlite3DebugPrintf(" orderby[%d]: col=%d desc=%d\n", + i, + p->aOrderBy[i].iColumn, + p->aOrderBy[i].desc); + } +} +static void TRACE_IDX_OUTPUTS(sqlite3_index_info *p){ + int i; + if( !sqlite3WhereTrace ) return; + for(i=0; inConstraint; i++){ + sqlite3DebugPrintf(" usage[%d]: argvIdx=%d omit=%d\n", + i, + p->aConstraintUsage[i].argvIndex, + p->aConstraintUsage[i].omit); + } + sqlite3DebugPrintf(" idxNum=%d\n", p->idxNum); + sqlite3DebugPrintf(" idxStr=%s\n", p->idxStr); + sqlite3DebugPrintf(" orderByConsumed=%d\n", p->orderByConsumed); + sqlite3DebugPrintf(" estimatedCost=%g\n", p->estimatedCost); + sqlite3DebugPrintf(" estimatedRows=%lld\n", p->estimatedRows); +} +#else +#define TRACE_IDX_INPUTS(A) +#define TRACE_IDX_OUTPUTS(A) +#endif + +#ifndef SQLITE_OMIT_AUTOMATIC_INDEX +/* +** Return TRUE if the WHERE clause term pTerm is of a form where it +** could be used with an index to access pSrc, assuming an appropriate +** index existed. +*/ +static int termCanDriveIndex( + WhereTerm *pTerm, /* WHERE clause term to check */ + struct SrcList_item *pSrc, /* Table we are trying to access */ + Bitmask notReady /* Tables in outer loops of the join */ +){ + char aff; + if( pTerm->leftCursor!=pSrc->iCursor ) return 0; + if( (pTerm->eOperator & (WO_EQ|WO_IS))==0 ) return 0; + if( (pSrc->fg.jointype & JT_LEFT) + && !ExprHasProperty(pTerm->pExpr, EP_FromJoin) + && (pTerm->eOperator & WO_IS) + ){ + /* Cannot use an IS term from the WHERE clause as an index driver for + ** the RHS of a LEFT JOIN. Such a term can only be used if it is from + ** the ON clause. */ + return 0; + } + if( (pTerm->prereqRight & notReady)!=0 ) return 0; + if( pTerm->u.leftColumn<0 ) return 0; + aff = pSrc->pTab->aCol[pTerm->u.leftColumn].affinity; + if( !sqlite3IndexAffinityOk(pTerm->pExpr, aff) ) return 0; + testcase( pTerm->pExpr->op==TK_IS ); + return 1; +} +#endif + + +#ifndef SQLITE_OMIT_AUTOMATIC_INDEX +/* +** Generate code to construct the Index object for an automatic index +** and to set up the WhereLevel object pLevel so that the code generator +** makes use of the automatic index. +*/ +static void constructAutomaticIndex( + Parse *pParse, /* The parsing context */ + WhereClause *pWC, /* The WHERE clause */ + struct SrcList_item *pSrc, /* The FROM clause term to get the next index */ + Bitmask notReady, /* Mask of cursors that are not available */ + WhereLevel *pLevel /* Write new index here */ +){ + int nKeyCol; /* Number of columns in the constructed index */ + WhereTerm *pTerm; /* A single term of the WHERE clause */ + WhereTerm *pWCEnd; /* End of pWC->a[] */ + Index *pIdx; /* Object describing the transient index */ + Vdbe *v; /* Prepared statement under construction */ + int addrInit; /* Address of the initialization bypass jump */ + Table *pTable; /* The table being indexed */ + int addrTop; /* Top of the index fill loop */ + int regRecord; /* Register holding an index record */ + int n; /* Column counter */ + int i; /* Loop counter */ + int mxBitCol; /* Maximum column in pSrc->colUsed */ + CollSeq *pColl; /* Collating sequence to on a column */ + WhereLoop *pLoop; /* The Loop object */ + char *zNotUsed; /* Extra space on the end of pIdx */ + Bitmask idxCols; /* Bitmap of columns used for indexing */ + Bitmask extraCols; /* Bitmap of additional columns */ + u8 sentWarning = 0; /* True if a warnning has been issued */ + Expr *pPartial = 0; /* Partial Index Expression */ + int iContinue = 0; /* Jump here to skip excluded rows */ + struct SrcList_item *pTabItem; /* FROM clause term being indexed */ + int addrCounter = 0; /* Address where integer counter is initialized */ + int regBase; /* Array of registers where record is assembled */ + + /* Generate code to skip over the creation and initialization of the + ** transient index on 2nd and subsequent iterations of the loop. */ + v = pParse->pVdbe; + assert( v!=0 ); + addrInit = sqlite3VdbeAddOp0(v, OP_Once); VdbeCoverage(v); + + /* Count the number of columns that will be added to the index + ** and used to match WHERE clause constraints */ + nKeyCol = 0; + pTable = pSrc->pTab; + pWCEnd = &pWC->a[pWC->nTerm]; + pLoop = pLevel->pWLoop; + idxCols = 0; + for(pTerm=pWC->a; pTermpExpr; + assert( !ExprHasProperty(pExpr, EP_FromJoin) /* prereq always non-zero */ + || pExpr->iRightJoinTable!=pSrc->iCursor /* for the right-hand */ + || pLoop->prereq!=0 ); /* table of a LEFT JOIN */ + if( pLoop->prereq==0 + && (pTerm->wtFlags & TERM_VIRTUAL)==0 + && !ExprHasProperty(pExpr, EP_FromJoin) + && sqlite3ExprIsTableConstant(pExpr, pSrc->iCursor) ){ + pPartial = sqlite3ExprAnd(pParse, pPartial, + sqlite3ExprDup(pParse->db, pExpr, 0)); + } + if( termCanDriveIndex(pTerm, pSrc, notReady) ){ + int iCol = pTerm->u.leftColumn; + Bitmask cMask = iCol>=BMS ? MASKBIT(BMS-1) : MASKBIT(iCol); + testcase( iCol==BMS ); + testcase( iCol==BMS-1 ); + if( !sentWarning ){ + sqlite3_log(SQLITE_WARNING_AUTOINDEX, + "automatic index on %s(%s)", pTable->zName, + pTable->aCol[iCol].zName); + sentWarning = 1; + } + if( (idxCols & cMask)==0 ){ + if( whereLoopResize(pParse->db, pLoop, nKeyCol+1) ){ + goto end_auto_index_create; + } + pLoop->aLTerm[nKeyCol++] = pTerm; + idxCols |= cMask; + } + } + } + assert( nKeyCol>0 ); + pLoop->u.btree.nEq = pLoop->nLTerm = nKeyCol; + pLoop->wsFlags = WHERE_COLUMN_EQ | WHERE_IDX_ONLY | WHERE_INDEXED + | WHERE_AUTO_INDEX; + + /* Count the number of additional columns needed to create a + ** covering index. A "covering index" is an index that contains all + ** columns that are needed by the query. With a covering index, the + ** original table never needs to be accessed. Automatic indices must + ** be a covering index because the index will not be updated if the + ** original table changes and the index and table cannot both be used + ** if they go out of sync. + */ + extraCols = pSrc->colUsed & (~idxCols | MASKBIT(BMS-1)); + mxBitCol = MIN(BMS-1,pTable->nCol); + testcase( pTable->nCol==BMS-1 ); + testcase( pTable->nCol==BMS-2 ); + for(i=0; icolUsed & MASKBIT(BMS-1) ){ + nKeyCol += pTable->nCol - BMS + 1; + } + + /* Construct the Index object to describe this index */ + pIdx = sqlite3AllocateIndexObject(pParse->db, nKeyCol+1, 0, &zNotUsed); + if( pIdx==0 ) goto end_auto_index_create; + pLoop->u.btree.pIndex = pIdx; + pIdx->zName = "auto-index"; + pIdx->pTable = pTable; + n = 0; + idxCols = 0; + for(pTerm=pWC->a; pTermu.leftColumn; + Bitmask cMask = iCol>=BMS ? MASKBIT(BMS-1) : MASKBIT(iCol); + testcase( iCol==BMS-1 ); + testcase( iCol==BMS ); + if( (idxCols & cMask)==0 ){ + Expr *pX = pTerm->pExpr; + idxCols |= cMask; + pIdx->aiColumn[n] = pTerm->u.leftColumn; + pColl = sqlite3BinaryCompareCollSeq(pParse, pX->pLeft, pX->pRight); + pIdx->azColl[n] = pColl ? pColl->zName : sqlite3StrBINARY; + n++; + } + } + } + assert( (u32)n==pLoop->u.btree.nEq ); + + /* Add additional columns needed to make the automatic index into + ** a covering index */ + for(i=0; iaiColumn[n] = i; + pIdx->azColl[n] = sqlite3StrBINARY; + n++; + } + } + if( pSrc->colUsed & MASKBIT(BMS-1) ){ + for(i=BMS-1; inCol; i++){ + pIdx->aiColumn[n] = i; + pIdx->azColl[n] = sqlite3StrBINARY; + n++; + } + } + assert( n==nKeyCol ); + pIdx->aiColumn[n] = XN_ROWID; + pIdx->azColl[n] = sqlite3StrBINARY; + + /* Create the automatic index */ + assert( pLevel->iIdxCur>=0 ); + pLevel->iIdxCur = pParse->nTab++; + sqlite3VdbeAddOp2(v, OP_OpenAutoindex, pLevel->iIdxCur, nKeyCol+1); + sqlite3VdbeSetP4KeyInfo(pParse, pIdx); + VdbeComment((v, "for %s", pTable->zName)); + + /* Fill the automatic index with content */ + pTabItem = &pWC->pWInfo->pTabList->a[pLevel->iFrom]; + if( pTabItem->fg.viaCoroutine ){ + int regYield = pTabItem->regReturn; + addrCounter = sqlite3VdbeAddOp2(v, OP_Integer, 0, 0); + sqlite3VdbeAddOp3(v, OP_InitCoroutine, regYield, 0, pTabItem->addrFillSub); + addrTop = sqlite3VdbeAddOp1(v, OP_Yield, regYield); + VdbeCoverage(v); + VdbeComment((v, "next row of %s", pTabItem->pTab->zName)); + }else{ + addrTop = sqlite3VdbeAddOp1(v, OP_Rewind, pLevel->iTabCur); VdbeCoverage(v); + } + if( pPartial ){ + iContinue = sqlite3VdbeMakeLabel(pParse); + sqlite3ExprIfFalse(pParse, pPartial, iContinue, SQLITE_JUMPIFNULL); + pLoop->wsFlags |= WHERE_PARTIALIDX; + } + regRecord = sqlite3GetTempReg(pParse); + regBase = sqlite3GenerateIndexKey( + pParse, pIdx, pLevel->iTabCur, regRecord, 0, 0, 0, 0 + ); + sqlite3VdbeAddOp2(v, OP_IdxInsert, pLevel->iIdxCur, regRecord); + sqlite3VdbeChangeP5(v, OPFLAG_USESEEKRESULT); + if( pPartial ) sqlite3VdbeResolveLabel(v, iContinue); + if( pTabItem->fg.viaCoroutine ){ + sqlite3VdbeChangeP2(v, addrCounter, regBase+n); + testcase( pParse->db->mallocFailed ); + assert( pLevel->iIdxCur>0 ); + translateColumnToCopy(pParse, addrTop, pLevel->iTabCur, + pTabItem->regResult, pLevel->iIdxCur); + sqlite3VdbeGoto(v, addrTop); + pTabItem->fg.viaCoroutine = 0; + }else{ + sqlite3VdbeAddOp2(v, OP_Next, pLevel->iTabCur, addrTop+1); VdbeCoverage(v); + } + sqlite3VdbeChangeP5(v, SQLITE_STMTSTATUS_AUTOINDEX); + sqlite3VdbeJumpHere(v, addrTop); + sqlite3ReleaseTempReg(pParse, regRecord); + + /* Jump here when skipping the initialization */ + sqlite3VdbeJumpHere(v, addrInit); + +end_auto_index_create: + sqlite3ExprDelete(pParse->db, pPartial); +} +#endif /* SQLITE_OMIT_AUTOMATIC_INDEX */ + +#ifndef SQLITE_OMIT_VIRTUALTABLE +/* +** Allocate and populate an sqlite3_index_info structure. It is the +** responsibility of the caller to eventually release the structure +** by passing the pointer returned by this function to sqlite3_free(). +*/ +static sqlite3_index_info *allocateIndexInfo( + Parse *pParse, /* The parsing context */ + WhereClause *pWC, /* The WHERE clause being analyzed */ + Bitmask mUnusable, /* Ignore terms with these prereqs */ + struct SrcList_item *pSrc, /* The FROM clause term that is the vtab */ + ExprList *pOrderBy, /* The ORDER BY clause */ + u16 *pmNoOmit /* Mask of terms not to omit */ +){ + int i, j; + int nTerm; + struct sqlite3_index_constraint *pIdxCons; + struct sqlite3_index_orderby *pIdxOrderBy; + struct sqlite3_index_constraint_usage *pUsage; + struct HiddenIndexInfo *pHidden; + WhereTerm *pTerm; + int nOrderBy; + sqlite3_index_info *pIdxInfo; + u16 mNoOmit = 0; + + /* Count the number of possible WHERE clause constraints referring + ** to this virtual table */ + for(i=nTerm=0, pTerm=pWC->a; inTerm; i++, pTerm++){ + if( pTerm->leftCursor != pSrc->iCursor ) continue; + if( pTerm->prereqRight & mUnusable ) continue; + assert( IsPowerOfTwo(pTerm->eOperator & ~WO_EQUIV) ); + testcase( pTerm->eOperator & WO_IN ); + testcase( pTerm->eOperator & WO_ISNULL ); + testcase( pTerm->eOperator & WO_IS ); + testcase( pTerm->eOperator & WO_ALL ); + if( (pTerm->eOperator & ~(WO_EQUIV))==0 ) continue; + if( pTerm->wtFlags & TERM_VNULL ) continue; + assert( pTerm->u.leftColumn>=(-1) ); + nTerm++; + } + + /* If the ORDER BY clause contains only columns in the current + ** virtual table then allocate space for the aOrderBy part of + ** the sqlite3_index_info structure. + */ + nOrderBy = 0; + if( pOrderBy ){ + int n = pOrderBy->nExpr; + for(i=0; ia[i].pExpr; + if( pExpr->op!=TK_COLUMN || pExpr->iTable!=pSrc->iCursor ) break; + if( pOrderBy->a[i].sortFlags & KEYINFO_ORDER_BIGNULL ) break; + } + if( i==n){ + nOrderBy = n; + } + } + + /* Allocate the sqlite3_index_info structure + */ + pIdxInfo = sqlite3DbMallocZero(pParse->db, sizeof(*pIdxInfo) + + (sizeof(*pIdxCons) + sizeof(*pUsage))*nTerm + + sizeof(*pIdxOrderBy)*nOrderBy + sizeof(*pHidden) ); + if( pIdxInfo==0 ){ + sqlite3ErrorMsg(pParse, "out of memory"); + return 0; + } + + /* Initialize the structure. The sqlite3_index_info structure contains + ** many fields that are declared "const" to prevent xBestIndex from + ** changing them. We have to do some funky casting in order to + ** initialize those fields. + */ + pHidden = (struct HiddenIndexInfo*)&pIdxInfo[1]; + pIdxCons = (struct sqlite3_index_constraint*)&pHidden[1]; + pIdxOrderBy = (struct sqlite3_index_orderby*)&pIdxCons[nTerm]; + pUsage = (struct sqlite3_index_constraint_usage*)&pIdxOrderBy[nOrderBy]; + *(int*)&pIdxInfo->nConstraint = nTerm; + *(int*)&pIdxInfo->nOrderBy = nOrderBy; + *(struct sqlite3_index_constraint**)&pIdxInfo->aConstraint = pIdxCons; + *(struct sqlite3_index_orderby**)&pIdxInfo->aOrderBy = pIdxOrderBy; + *(struct sqlite3_index_constraint_usage**)&pIdxInfo->aConstraintUsage = + pUsage; + + pHidden->pWC = pWC; + pHidden->pParse = pParse; + for(i=j=0, pTerm=pWC->a; inTerm; i++, pTerm++){ + u16 op; + if( pTerm->leftCursor != pSrc->iCursor ) continue; + if( pTerm->prereqRight & mUnusable ) continue; + assert( IsPowerOfTwo(pTerm->eOperator & ~WO_EQUIV) ); + testcase( pTerm->eOperator & WO_IN ); + testcase( pTerm->eOperator & WO_IS ); + testcase( pTerm->eOperator & WO_ISNULL ); + testcase( pTerm->eOperator & WO_ALL ); + if( (pTerm->eOperator & ~(WO_EQUIV))==0 ) continue; + if( pTerm->wtFlags & TERM_VNULL ) continue; + if( (pSrc->fg.jointype & JT_LEFT)!=0 + && !ExprHasProperty(pTerm->pExpr, EP_FromJoin) + && (pTerm->eOperator & (WO_IS|WO_ISNULL)) + ){ + /* An "IS" term in the WHERE clause where the virtual table is the rhs + ** of a LEFT JOIN. Do not pass this term to the virtual table + ** implementation, as this can lead to incorrect results from SQL such + ** as: + ** + ** "LEFT JOIN vtab WHERE vtab.col IS NULL" */ + testcase( pTerm->eOperator & WO_ISNULL ); + testcase( pTerm->eOperator & WO_IS ); + continue; + } + assert( pTerm->u.leftColumn>=(-1) ); + pIdxCons[j].iColumn = pTerm->u.leftColumn; + pIdxCons[j].iTermOffset = i; + op = pTerm->eOperator & WO_ALL; + if( op==WO_IN ) op = WO_EQ; + if( op==WO_AUX ){ + pIdxCons[j].op = pTerm->eMatchOp; + }else if( op & (WO_ISNULL|WO_IS) ){ + if( op==WO_ISNULL ){ + pIdxCons[j].op = SQLITE_INDEX_CONSTRAINT_ISNULL; + }else{ + pIdxCons[j].op = SQLITE_INDEX_CONSTRAINT_IS; + } + }else{ + pIdxCons[j].op = (u8)op; + /* The direct assignment in the previous line is possible only because + ** the WO_ and SQLITE_INDEX_CONSTRAINT_ codes are identical. The + ** following asserts verify this fact. */ + assert( WO_EQ==SQLITE_INDEX_CONSTRAINT_EQ ); + assert( WO_LT==SQLITE_INDEX_CONSTRAINT_LT ); + assert( WO_LE==SQLITE_INDEX_CONSTRAINT_LE ); + assert( WO_GT==SQLITE_INDEX_CONSTRAINT_GT ); + assert( WO_GE==SQLITE_INDEX_CONSTRAINT_GE ); + assert( pTerm->eOperator&(WO_IN|WO_EQ|WO_LT|WO_LE|WO_GT|WO_GE|WO_AUX) ); + + if( op & (WO_LT|WO_LE|WO_GT|WO_GE) + && sqlite3ExprIsVector(pTerm->pExpr->pRight) + ){ + if( i<16 ) mNoOmit |= (1 << i); + if( op==WO_LT ) pIdxCons[j].op = WO_LE; + if( op==WO_GT ) pIdxCons[j].op = WO_GE; + } + } + + j++; + } + for(i=0; ia[i].pExpr; + pIdxOrderBy[i].iColumn = pExpr->iColumn; + pIdxOrderBy[i].desc = pOrderBy->a[i].sortFlags & KEYINFO_ORDER_DESC; + } + + *pmNoOmit = mNoOmit; + return pIdxInfo; +} + +/* +** The table object reference passed as the second argument to this function +** must represent a virtual table. This function invokes the xBestIndex() +** method of the virtual table with the sqlite3_index_info object that +** comes in as the 3rd argument to this function. +** +** If an error occurs, pParse is populated with an error message and an +** appropriate error code is returned. A return of SQLITE_CONSTRAINT from +** xBestIndex is not considered an error. SQLITE_CONSTRAINT indicates that +** the current configuration of "unusable" flags in sqlite3_index_info can +** not result in a valid plan. +** +** Whether or not an error is returned, it is the responsibility of the +** caller to eventually free p->idxStr if p->needToFreeIdxStr indicates +** that this is required. +*/ +static int vtabBestIndex(Parse *pParse, Table *pTab, sqlite3_index_info *p){ + sqlite3_vtab *pVtab = sqlite3GetVTable(pParse->db, pTab)->pVtab; + int rc; + + TRACE_IDX_INPUTS(p); + rc = pVtab->pModule->xBestIndex(pVtab, p); + TRACE_IDX_OUTPUTS(p); + + if( rc!=SQLITE_OK && rc!=SQLITE_CONSTRAINT ){ + if( rc==SQLITE_NOMEM ){ + sqlite3OomFault(pParse->db); + }else if( !pVtab->zErrMsg ){ + sqlite3ErrorMsg(pParse, "%s", sqlite3ErrStr(rc)); + }else{ + sqlite3ErrorMsg(pParse, "%s", pVtab->zErrMsg); + } + } + sqlite3_free(pVtab->zErrMsg); + pVtab->zErrMsg = 0; + return rc; +} +#endif /* !defined(SQLITE_OMIT_VIRTUALTABLE) */ + +#ifdef SQLITE_ENABLE_STAT4 +/* +** Estimate the location of a particular key among all keys in an +** index. Store the results in aStat as follows: +** +** aStat[0] Est. number of rows less than pRec +** aStat[1] Est. number of rows equal to pRec +** +** Return the index of the sample that is the smallest sample that +** is greater than or equal to pRec. Note that this index is not an index +** into the aSample[] array - it is an index into a virtual set of samples +** based on the contents of aSample[] and the number of fields in record +** pRec. +*/ +static int whereKeyStats( + Parse *pParse, /* Database connection */ + Index *pIdx, /* Index to consider domain of */ + UnpackedRecord *pRec, /* Vector of values to consider */ + int roundUp, /* Round up if true. Round down if false */ + tRowcnt *aStat /* OUT: stats written here */ +){ + IndexSample *aSample = pIdx->aSample; + int iCol; /* Index of required stats in anEq[] etc. */ + int i; /* Index of first sample >= pRec */ + int iSample; /* Smallest sample larger than or equal to pRec */ + int iMin = 0; /* Smallest sample not yet tested */ + int iTest; /* Next sample to test */ + int res; /* Result of comparison operation */ + int nField; /* Number of fields in pRec */ + tRowcnt iLower = 0; /* anLt[] + anEq[] of largest sample pRec is > */ + +#ifndef SQLITE_DEBUG + UNUSED_PARAMETER( pParse ); +#endif + assert( pRec!=0 ); + assert( pIdx->nSample>0 ); + assert( pRec->nField>0 && pRec->nField<=pIdx->nSampleCol ); + + /* Do a binary search to find the first sample greater than or equal + ** to pRec. If pRec contains a single field, the set of samples to search + ** is simply the aSample[] array. If the samples in aSample[] contain more + ** than one fields, all fields following the first are ignored. + ** + ** If pRec contains N fields, where N is more than one, then as well as the + ** samples in aSample[] (truncated to N fields), the search also has to + ** consider prefixes of those samples. For example, if the set of samples + ** in aSample is: + ** + ** aSample[0] = (a, 5) + ** aSample[1] = (a, 10) + ** aSample[2] = (b, 5) + ** aSample[3] = (c, 100) + ** aSample[4] = (c, 105) + ** + ** Then the search space should ideally be the samples above and the + ** unique prefixes [a], [b] and [c]. But since that is hard to organize, + ** the code actually searches this set: + ** + ** 0: (a) + ** 1: (a, 5) + ** 2: (a, 10) + ** 3: (a, 10) + ** 4: (b) + ** 5: (b, 5) + ** 6: (c) + ** 7: (c, 100) + ** 8: (c, 105) + ** 9: (c, 105) + ** + ** For each sample in the aSample[] array, N samples are present in the + ** effective sample array. In the above, samples 0 and 1 are based on + ** sample aSample[0]. Samples 2 and 3 on aSample[1] etc. + ** + ** Often, sample i of each block of N effective samples has (i+1) fields. + ** Except, each sample may be extended to ensure that it is greater than or + ** equal to the previous sample in the array. For example, in the above, + ** sample 2 is the first sample of a block of N samples, so at first it + ** appears that it should be 1 field in size. However, that would make it + ** smaller than sample 1, so the binary search would not work. As a result, + ** it is extended to two fields. The duplicates that this creates do not + ** cause any problems. + */ + nField = pRec->nField; + iCol = 0; + iSample = pIdx->nSample * nField; + do{ + int iSamp; /* Index in aSample[] of test sample */ + int n; /* Number of fields in test sample */ + + iTest = (iMin+iSample)/2; + iSamp = iTest / nField; + if( iSamp>0 ){ + /* The proposed effective sample is a prefix of sample aSample[iSamp]. + ** Specifically, the shortest prefix of at least (1 + iTest%nField) + ** fields that is greater than the previous effective sample. */ + for(n=(iTest % nField) + 1; nnField = n; + res = sqlite3VdbeRecordCompare(aSample[iSamp].n, aSample[iSamp].p, pRec); + if( res<0 ){ + iLower = aSample[iSamp].anLt[n-1] + aSample[iSamp].anEq[n-1]; + iMin = iTest+1; + }else if( res==0 && ndb->mallocFailed==0 ){ + if( res==0 ){ + /* If (res==0) is true, then pRec must be equal to sample i. */ + assert( inSample ); + assert( iCol==nField-1 ); + pRec->nField = nField; + assert( 0==sqlite3VdbeRecordCompare(aSample[i].n, aSample[i].p, pRec) + || pParse->db->mallocFailed + ); + }else{ + /* Unless i==pIdx->nSample, indicating that pRec is larger than + ** all samples in the aSample[] array, pRec must be smaller than the + ** (iCol+1) field prefix of sample i. */ + assert( i<=pIdx->nSample && i>=0 ); + pRec->nField = iCol+1; + assert( i==pIdx->nSample + || sqlite3VdbeRecordCompare(aSample[i].n, aSample[i].p, pRec)>0 + || pParse->db->mallocFailed ); + + /* if i==0 and iCol==0, then record pRec is smaller than all samples + ** in the aSample[] array. Otherwise, if (iCol>0) then pRec must + ** be greater than or equal to the (iCol) field prefix of sample i. + ** If (i>0), then pRec must also be greater than sample (i-1). */ + if( iCol>0 ){ + pRec->nField = iCol; + assert( sqlite3VdbeRecordCompare(aSample[i].n, aSample[i].p, pRec)<=0 + || pParse->db->mallocFailed ); + } + if( i>0 ){ + pRec->nField = nField; + assert( sqlite3VdbeRecordCompare(aSample[i-1].n, aSample[i-1].p, pRec)<0 + || pParse->db->mallocFailed ); + } + } + } +#endif /* ifdef SQLITE_DEBUG */ + + if( res==0 ){ + /* Record pRec is equal to sample i */ + assert( iCol==nField-1 ); + aStat[0] = aSample[i].anLt[iCol]; + aStat[1] = aSample[i].anEq[iCol]; + }else{ + /* At this point, the (iCol+1) field prefix of aSample[i] is the first + ** sample that is greater than pRec. Or, if i==pIdx->nSample then pRec + ** is larger than all samples in the array. */ + tRowcnt iUpper, iGap; + if( i>=pIdx->nSample ){ + iUpper = sqlite3LogEstToInt(pIdx->aiRowLogEst[0]); + }else{ + iUpper = aSample[i].anLt[iCol]; + } + + if( iLower>=iUpper ){ + iGap = 0; + }else{ + iGap = iUpper - iLower; + } + if( roundUp ){ + iGap = (iGap*2)/3; + }else{ + iGap = iGap/3; + } + aStat[0] = iLower + iGap; + aStat[1] = pIdx->aAvgEq[nField-1]; + } + + /* Restore the pRec->nField value before returning. */ + pRec->nField = nField; + return i; +} +#endif /* SQLITE_ENABLE_STAT4 */ + +/* +** If it is not NULL, pTerm is a term that provides an upper or lower +** bound on a range scan. Without considering pTerm, it is estimated +** that the scan will visit nNew rows. This function returns the number +** estimated to be visited after taking pTerm into account. +** +** If the user explicitly specified a likelihood() value for this term, +** then the return value is the likelihood multiplied by the number of +** input rows. Otherwise, this function assumes that an "IS NOT NULL" term +** has a likelihood of 0.50, and any other term a likelihood of 0.25. +*/ +static LogEst whereRangeAdjust(WhereTerm *pTerm, LogEst nNew){ + LogEst nRet = nNew; + if( pTerm ){ + if( pTerm->truthProb<=0 ){ + nRet += pTerm->truthProb; + }else if( (pTerm->wtFlags & TERM_VNULL)==0 ){ + nRet -= 20; assert( 20==sqlite3LogEst(4) ); + } + } + return nRet; +} + + +#ifdef SQLITE_ENABLE_STAT4 +/* +** Return the affinity for a single column of an index. +*/ +SQLITE_PRIVATE char sqlite3IndexColumnAffinity(sqlite3 *db, Index *pIdx, int iCol){ + assert( iCol>=0 && iColnColumn ); + if( !pIdx->zColAff ){ + if( sqlite3IndexAffinityStr(db, pIdx)==0 ) return SQLITE_AFF_BLOB; + } + assert( pIdx->zColAff[iCol]!=0 ); + return pIdx->zColAff[iCol]; +} +#endif + + +#ifdef SQLITE_ENABLE_STAT4 +/* +** This function is called to estimate the number of rows visited by a +** range-scan on a skip-scan index. For example: +** +** CREATE INDEX i1 ON t1(a, b, c); +** SELECT * FROM t1 WHERE a=? AND c BETWEEN ? AND ?; +** +** Value pLoop->nOut is currently set to the estimated number of rows +** visited for scanning (a=? AND b=?). This function reduces that estimate +** by some factor to account for the (c BETWEEN ? AND ?) expression based +** on the stat4 data for the index. this scan will be peformed multiple +** times (once for each (a,b) combination that matches a=?) is dealt with +** by the caller. +** +** It does this by scanning through all stat4 samples, comparing values +** extracted from pLower and pUpper with the corresponding column in each +** sample. If L and U are the number of samples found to be less than or +** equal to the values extracted from pLower and pUpper respectively, and +** N is the total number of samples, the pLoop->nOut value is adjusted +** as follows: +** +** nOut = nOut * ( min(U - L, 1) / N ) +** +** If pLower is NULL, or a value cannot be extracted from the term, L is +** set to zero. If pUpper is NULL, or a value cannot be extracted from it, +** U is set to N. +** +** Normally, this function sets *pbDone to 1 before returning. However, +** if no value can be extracted from either pLower or pUpper (and so the +** estimate of the number of rows delivered remains unchanged), *pbDone +** is left as is. +** +** If an error occurs, an SQLite error code is returned. Otherwise, +** SQLITE_OK. +*/ +static int whereRangeSkipScanEst( + Parse *pParse, /* Parsing & code generating context */ + WhereTerm *pLower, /* Lower bound on the range. ex: "x>123" Might be NULL */ + WhereTerm *pUpper, /* Upper bound on the range. ex: "x<455" Might be NULL */ + WhereLoop *pLoop, /* Update the .nOut value of this loop */ + int *pbDone /* Set to true if at least one expr. value extracted */ +){ + Index *p = pLoop->u.btree.pIndex; + int nEq = pLoop->u.btree.nEq; + sqlite3 *db = pParse->db; + int nLower = -1; + int nUpper = p->nSample+1; + int rc = SQLITE_OK; + u8 aff = sqlite3IndexColumnAffinity(db, p, nEq); + CollSeq *pColl; + + sqlite3_value *p1 = 0; /* Value extracted from pLower */ + sqlite3_value *p2 = 0; /* Value extracted from pUpper */ + sqlite3_value *pVal = 0; /* Value extracted from record */ + + pColl = sqlite3LocateCollSeq(pParse, p->azColl[nEq]); + if( pLower ){ + rc = sqlite3Stat4ValueFromExpr(pParse, pLower->pExpr->pRight, aff, &p1); + nLower = 0; + } + if( pUpper && rc==SQLITE_OK ){ + rc = sqlite3Stat4ValueFromExpr(pParse, pUpper->pExpr->pRight, aff, &p2); + nUpper = p2 ? 0 : p->nSample; + } + + if( p1 || p2 ){ + int i; + int nDiff; + for(i=0; rc==SQLITE_OK && inSample; i++){ + rc = sqlite3Stat4Column(db, p->aSample[i].p, p->aSample[i].n, nEq, &pVal); + if( rc==SQLITE_OK && p1 ){ + int res = sqlite3MemCompare(p1, pVal, pColl); + if( res>=0 ) nLower++; + } + if( rc==SQLITE_OK && p2 ){ + int res = sqlite3MemCompare(p2, pVal, pColl); + if( res>=0 ) nUpper++; + } + } + nDiff = (nUpper - nLower); + if( nDiff<=0 ) nDiff = 1; + + /* If there is both an upper and lower bound specified, and the + ** comparisons indicate that they are close together, use the fallback + ** method (assume that the scan visits 1/64 of the rows) for estimating + ** the number of rows visited. Otherwise, estimate the number of rows + ** using the method described in the header comment for this function. */ + if( nDiff!=1 || pUpper==0 || pLower==0 ){ + int nAdjust = (sqlite3LogEst(p->nSample) - sqlite3LogEst(nDiff)); + pLoop->nOut -= nAdjust; + *pbDone = 1; + WHERETRACE(0x10, ("range skip-scan regions: %u..%u adjust=%d est=%d\n", + nLower, nUpper, nAdjust*-1, pLoop->nOut)); + } + + }else{ + assert( *pbDone==0 ); + } + + sqlite3ValueFree(p1); + sqlite3ValueFree(p2); + sqlite3ValueFree(pVal); + + return rc; +} +#endif /* SQLITE_ENABLE_STAT4 */ + +/* +** This function is used to estimate the number of rows that will be visited +** by scanning an index for a range of values. The range may have an upper +** bound, a lower bound, or both. The WHERE clause terms that set the upper +** and lower bounds are represented by pLower and pUpper respectively. For +** example, assuming that index p is on t1(a): +** +** ... FROM t1 WHERE a > ? AND a < ? ... +** |_____| |_____| +** | | +** pLower pUpper +** +** If either of the upper or lower bound is not present, then NULL is passed in +** place of the corresponding WhereTerm. +** +** The value in (pBuilder->pNew->u.btree.nEq) is the number of the index +** column subject to the range constraint. Or, equivalently, the number of +** equality constraints optimized by the proposed index scan. For example, +** assuming index p is on t1(a, b), and the SQL query is: +** +** ... FROM t1 WHERE a = ? AND b > ? AND b < ? ... +** +** then nEq is set to 1 (as the range restricted column, b, is the second +** left-most column of the index). Or, if the query is: +** +** ... FROM t1 WHERE a > ? AND a < ? ... +** +** then nEq is set to 0. +** +** When this function is called, *pnOut is set to the sqlite3LogEst() of the +** number of rows that the index scan is expected to visit without +** considering the range constraints. If nEq is 0, then *pnOut is the number of +** rows in the index. Assuming no error occurs, *pnOut is adjusted (reduced) +** to account for the range constraints pLower and pUpper. +** +** In the absence of sqlite_stat4 ANALYZE data, or if such data cannot be +** used, a single range inequality reduces the search space by a factor of 4. +** and a pair of constraints (x>? AND x123" Might be NULL */ + WhereTerm *pUpper, /* Upper bound on the range. ex: "x<455" Might be NULL */ + WhereLoop *pLoop /* Modify the .nOut and maybe .rRun fields */ +){ + int rc = SQLITE_OK; + int nOut = pLoop->nOut; + LogEst nNew; + +#ifdef SQLITE_ENABLE_STAT4 + Index *p = pLoop->u.btree.pIndex; + int nEq = pLoop->u.btree.nEq; + + if( p->nSample>0 && ALWAYS(nEqnSampleCol) + && OptimizationEnabled(pParse->db, SQLITE_Stat4) + ){ + if( nEq==pBuilder->nRecValid ){ + UnpackedRecord *pRec = pBuilder->pRec; + tRowcnt a[2]; + int nBtm = pLoop->u.btree.nBtm; + int nTop = pLoop->u.btree.nTop; + + /* Variable iLower will be set to the estimate of the number of rows in + ** the index that are less than the lower bound of the range query. The + ** lower bound being the concatenation of $P and $L, where $P is the + ** key-prefix formed by the nEq values matched against the nEq left-most + ** columns of the index, and $L is the value in pLower. + ** + ** Or, if pLower is NULL or $L cannot be extracted from it (because it + ** is not a simple variable or literal value), the lower bound of the + ** range is $P. Due to a quirk in the way whereKeyStats() works, even + ** if $L is available, whereKeyStats() is called for both ($P) and + ** ($P:$L) and the larger of the two returned values is used. + ** + ** Similarly, iUpper is to be set to the estimate of the number of rows + ** less than the upper bound of the range query. Where the upper bound + ** is either ($P) or ($P:$U). Again, even if $U is available, both values + ** of iUpper are requested of whereKeyStats() and the smaller used. + ** + ** The number of rows between the two bounds is then just iUpper-iLower. + */ + tRowcnt iLower; /* Rows less than the lower bound */ + tRowcnt iUpper; /* Rows less than the upper bound */ + int iLwrIdx = -2; /* aSample[] for the lower bound */ + int iUprIdx = -1; /* aSample[] for the upper bound */ + + if( pRec ){ + testcase( pRec->nField!=pBuilder->nRecValid ); + pRec->nField = pBuilder->nRecValid; + } + /* Determine iLower and iUpper using ($P) only. */ + if( nEq==0 ){ + iLower = 0; + iUpper = p->nRowEst0; + }else{ + /* Note: this call could be optimized away - since the same values must + ** have been requested when testing key $P in whereEqualScanEst(). */ + whereKeyStats(pParse, p, pRec, 0, a); + iLower = a[0]; + iUpper = a[0] + a[1]; + } + + assert( pLower==0 || (pLower->eOperator & (WO_GT|WO_GE))!=0 ); + assert( pUpper==0 || (pUpper->eOperator & (WO_LT|WO_LE))!=0 ); + assert( p->aSortOrder!=0 ); + if( p->aSortOrder[nEq] ){ + /* The roles of pLower and pUpper are swapped for a DESC index */ + SWAP(WhereTerm*, pLower, pUpper); + SWAP(int, nBtm, nTop); + } + + /* If possible, improve on the iLower estimate using ($P:$L). */ + if( pLower ){ + int n; /* Values extracted from pExpr */ + Expr *pExpr = pLower->pExpr->pRight; + rc = sqlite3Stat4ProbeSetValue(pParse, p, &pRec, pExpr, nBtm, nEq, &n); + if( rc==SQLITE_OK && n ){ + tRowcnt iNew; + u16 mask = WO_GT|WO_LE; + if( sqlite3ExprVectorSize(pExpr)>n ) mask = (WO_LE|WO_LT); + iLwrIdx = whereKeyStats(pParse, p, pRec, 0, a); + iNew = a[0] + ((pLower->eOperator & mask) ? a[1] : 0); + if( iNew>iLower ) iLower = iNew; + nOut--; + pLower = 0; + } + } + + /* If possible, improve on the iUpper estimate using ($P:$U). */ + if( pUpper ){ + int n; /* Values extracted from pExpr */ + Expr *pExpr = pUpper->pExpr->pRight; + rc = sqlite3Stat4ProbeSetValue(pParse, p, &pRec, pExpr, nTop, nEq, &n); + if( rc==SQLITE_OK && n ){ + tRowcnt iNew; + u16 mask = WO_GT|WO_LE; + if( sqlite3ExprVectorSize(pExpr)>n ) mask = (WO_LE|WO_LT); + iUprIdx = whereKeyStats(pParse, p, pRec, 1, a); + iNew = a[0] + ((pUpper->eOperator & mask) ? a[1] : 0); + if( iNewpRec = pRec; + if( rc==SQLITE_OK ){ + if( iUpper>iLower ){ + nNew = sqlite3LogEst(iUpper - iLower); + /* TUNING: If both iUpper and iLower are derived from the same + ** sample, then assume they are 4x more selective. This brings + ** the estimated selectivity more in line with what it would be + ** if estimated without the use of STAT4 tables. */ + if( iLwrIdx==iUprIdx ) nNew -= 20; assert( 20==sqlite3LogEst(4) ); + }else{ + nNew = 10; assert( 10==sqlite3LogEst(2) ); + } + if( nNewwtFlags & TERM_VNULL)==0 ); + nNew = whereRangeAdjust(pLower, nOut); + nNew = whereRangeAdjust(pUpper, nNew); + + /* TUNING: If there is both an upper and lower limit and neither limit + ** has an application-defined likelihood(), assume the range is + ** reduced by an additional 75%. This means that, by default, an open-ended + ** range query (e.g. col > ?) is assumed to match 1/4 of the rows in the + ** index. While a closed range (e.g. col BETWEEN ? AND ?) is estimated to + ** match 1/64 of the index. */ + if( pLower && pLower->truthProb>0 && pUpper && pUpper->truthProb>0 ){ + nNew -= 20; + } + + nOut -= (pLower!=0) + (pUpper!=0); + if( nNew<10 ) nNew = 10; + if( nNewnOut>nOut ){ + WHERETRACE(0x10,("Range scan lowers nOut from %d to %d\n", + pLoop->nOut, nOut)); + } +#endif + pLoop->nOut = (LogEst)nOut; + return rc; +} + +#ifdef SQLITE_ENABLE_STAT4 +/* +** Estimate the number of rows that will be returned based on +** an equality constraint x=VALUE and where that VALUE occurs in +** the histogram data. This only works when x is the left-most +** column of an index and sqlite_stat4 histogram data is available +** for that index. When pExpr==NULL that means the constraint is +** "x IS NULL" instead of "x=VALUE". +** +** Write the estimated row count into *pnRow and return SQLITE_OK. +** If unable to make an estimate, leave *pnRow unchanged and return +** non-zero. +** +** This routine can fail if it is unable to load a collating sequence +** required for string comparison, or if unable to allocate memory +** for a UTF conversion required for comparison. The error is stored +** in the pParse structure. +*/ +static int whereEqualScanEst( + Parse *pParse, /* Parsing & code generating context */ + WhereLoopBuilder *pBuilder, + Expr *pExpr, /* Expression for VALUE in the x=VALUE constraint */ + tRowcnt *pnRow /* Write the revised row estimate here */ +){ + Index *p = pBuilder->pNew->u.btree.pIndex; + int nEq = pBuilder->pNew->u.btree.nEq; + UnpackedRecord *pRec = pBuilder->pRec; + int rc; /* Subfunction return code */ + tRowcnt a[2]; /* Statistics */ + int bOk; + + assert( nEq>=1 ); + assert( nEq<=p->nColumn ); + assert( p->aSample!=0 ); + assert( p->nSample>0 ); + assert( pBuilder->nRecValidnRecValid<(nEq-1) ){ + return SQLITE_NOTFOUND; + } + + /* This is an optimization only. The call to sqlite3Stat4ProbeSetValue() + ** below would return the same value. */ + if( nEq>=p->nColumn ){ + *pnRow = 1; + return SQLITE_OK; + } + + rc = sqlite3Stat4ProbeSetValue(pParse, p, &pRec, pExpr, 1, nEq-1, &bOk); + pBuilder->pRec = pRec; + if( rc!=SQLITE_OK ) return rc; + if( bOk==0 ) return SQLITE_NOTFOUND; + pBuilder->nRecValid = nEq; + + whereKeyStats(pParse, p, pRec, 0, a); + WHERETRACE(0x10,("equality scan regions %s(%d): %d\n", + p->zName, nEq-1, (int)a[1])); + *pnRow = a[1]; + + return rc; +} +#endif /* SQLITE_ENABLE_STAT4 */ + +#ifdef SQLITE_ENABLE_STAT4 +/* +** Estimate the number of rows that will be returned based on +** an IN constraint where the right-hand side of the IN operator +** is a list of values. Example: +** +** WHERE x IN (1,2,3,4) +** +** Write the estimated row count into *pnRow and return SQLITE_OK. +** If unable to make an estimate, leave *pnRow unchanged and return +** non-zero. +** +** This routine can fail if it is unable to load a collating sequence +** required for string comparison, or if unable to allocate memory +** for a UTF conversion required for comparison. The error is stored +** in the pParse structure. +*/ +static int whereInScanEst( + Parse *pParse, /* Parsing & code generating context */ + WhereLoopBuilder *pBuilder, + ExprList *pList, /* The value list on the RHS of "x IN (v1,v2,v3,...)" */ + tRowcnt *pnRow /* Write the revised row estimate here */ +){ + Index *p = pBuilder->pNew->u.btree.pIndex; + i64 nRow0 = sqlite3LogEstToInt(p->aiRowLogEst[0]); + int nRecValid = pBuilder->nRecValid; + int rc = SQLITE_OK; /* Subfunction return code */ + tRowcnt nEst; /* Number of rows for a single term */ + tRowcnt nRowEst = 0; /* New estimate of the number of rows */ + int i; /* Loop counter */ + + assert( p->aSample!=0 ); + for(i=0; rc==SQLITE_OK && inExpr; i++){ + nEst = nRow0; + rc = whereEqualScanEst(pParse, pBuilder, pList->a[i].pExpr, &nEst); + nRowEst += nEst; + pBuilder->nRecValid = nRecValid; + } + + if( rc==SQLITE_OK ){ + if( nRowEst > nRow0 ) nRowEst = nRow0; + *pnRow = nRowEst; + WHERETRACE(0x10,("IN row estimate: est=%d\n", nRowEst)); + } + assert( pBuilder->nRecValid==nRecValid ); + return rc; +} +#endif /* SQLITE_ENABLE_STAT4 */ + + +#ifdef WHERETRACE_ENABLED +/* +** Print the content of a WhereTerm object +*/ +static void whereTermPrint(WhereTerm *pTerm, int iTerm){ + if( pTerm==0 ){ + sqlite3DebugPrintf("TERM-%-3d NULL\n", iTerm); + }else{ + char zType[4]; + char zLeft[50]; + memcpy(zType, "...", 4); + if( pTerm->wtFlags & TERM_VIRTUAL ) zType[0] = 'V'; + if( pTerm->eOperator & WO_EQUIV ) zType[1] = 'E'; + if( ExprHasProperty(pTerm->pExpr, EP_FromJoin) ) zType[2] = 'L'; + if( pTerm->eOperator & WO_SINGLE ){ + sqlite3_snprintf(sizeof(zLeft),zLeft,"left={%d:%d}", + pTerm->leftCursor, pTerm->u.leftColumn); + }else if( (pTerm->eOperator & WO_OR)!=0 && pTerm->u.pOrInfo!=0 ){ + sqlite3_snprintf(sizeof(zLeft),zLeft,"indexable=0x%lld", + pTerm->u.pOrInfo->indexable); + }else{ + sqlite3_snprintf(sizeof(zLeft),zLeft,"left=%d", pTerm->leftCursor); + } + sqlite3DebugPrintf( + "TERM-%-3d %p %s %-12s prob=%-3d op=0x%03x wtFlags=0x%04x", + iTerm, pTerm, zType, zLeft, pTerm->truthProb, + pTerm->eOperator, pTerm->wtFlags); + if( pTerm->iField ){ + sqlite3DebugPrintf(" iField=%d\n", pTerm->iField); + }else{ + sqlite3DebugPrintf("\n"); + } + sqlite3TreeViewExpr(0, pTerm->pExpr, 0); + } +} +#endif + +#ifdef WHERETRACE_ENABLED +/* +** Show the complete content of a WhereClause +*/ +SQLITE_PRIVATE void sqlite3WhereClausePrint(WhereClause *pWC){ + int i; + for(i=0; inTerm; i++){ + whereTermPrint(&pWC->a[i], i); + } +} +#endif + +#ifdef WHERETRACE_ENABLED +/* +** Print a WhereLoop object for debugging purposes +*/ +static void whereLoopPrint(WhereLoop *p, WhereClause *pWC){ + WhereInfo *pWInfo = pWC->pWInfo; + int nb = 1+(pWInfo->pTabList->nSrc+3)/4; + struct SrcList_item *pItem = pWInfo->pTabList->a + p->iTab; + Table *pTab = pItem->pTab; + Bitmask mAll = (((Bitmask)1)<<(nb*4)) - 1; + sqlite3DebugPrintf("%c%2d.%0*llx.%0*llx", p->cId, + p->iTab, nb, p->maskSelf, nb, p->prereq & mAll); + sqlite3DebugPrintf(" %12s", + pItem->zAlias ? pItem->zAlias : pTab->zName); + if( (p->wsFlags & WHERE_VIRTUALTABLE)==0 ){ + const char *zName; + if( p->u.btree.pIndex && (zName = p->u.btree.pIndex->zName)!=0 ){ + if( strncmp(zName, "sqlite_autoindex_", 17)==0 ){ + int i = sqlite3Strlen30(zName) - 1; + while( zName[i]!='_' ) i--; + zName += i; + } + sqlite3DebugPrintf(".%-16s %2d", zName, p->u.btree.nEq); + }else{ + sqlite3DebugPrintf("%20s",""); + } + }else{ + char *z; + if( p->u.vtab.idxStr ){ + z = sqlite3_mprintf("(%d,\"%s\",%x)", + p->u.vtab.idxNum, p->u.vtab.idxStr, p->u.vtab.omitMask); + }else{ + z = sqlite3_mprintf("(%d,%x)", p->u.vtab.idxNum, p->u.vtab.omitMask); + } + sqlite3DebugPrintf(" %-19s", z); + sqlite3_free(z); + } + if( p->wsFlags & WHERE_SKIPSCAN ){ + sqlite3DebugPrintf(" f %05x %d-%d", p->wsFlags, p->nLTerm,p->nSkip); + }else{ + sqlite3DebugPrintf(" f %05x N %d", p->wsFlags, p->nLTerm); + } + sqlite3DebugPrintf(" cost %d,%d,%d\n", p->rSetup, p->rRun, p->nOut); + if( p->nLTerm && (sqlite3WhereTrace & 0x100)!=0 ){ + int i; + for(i=0; inLTerm; i++){ + whereTermPrint(p->aLTerm[i], i); + } + } +} +#endif + +/* +** Convert bulk memory into a valid WhereLoop that can be passed +** to whereLoopClear harmlessly. +*/ +static void whereLoopInit(WhereLoop *p){ + p->aLTerm = p->aLTermSpace; + p->nLTerm = 0; + p->nLSlot = ArraySize(p->aLTermSpace); + p->wsFlags = 0; +} + +/* +** Clear the WhereLoop.u union. Leave WhereLoop.pLTerm intact. +*/ +static void whereLoopClearUnion(sqlite3 *db, WhereLoop *p){ + if( p->wsFlags & (WHERE_VIRTUALTABLE|WHERE_AUTO_INDEX) ){ + if( (p->wsFlags & WHERE_VIRTUALTABLE)!=0 && p->u.vtab.needFree ){ + sqlite3_free(p->u.vtab.idxStr); + p->u.vtab.needFree = 0; + p->u.vtab.idxStr = 0; + }else if( (p->wsFlags & WHERE_AUTO_INDEX)!=0 && p->u.btree.pIndex!=0 ){ + sqlite3DbFree(db, p->u.btree.pIndex->zColAff); + sqlite3DbFreeNN(db, p->u.btree.pIndex); + p->u.btree.pIndex = 0; + } + } +} + +/* +** Deallocate internal memory used by a WhereLoop object +*/ +static void whereLoopClear(sqlite3 *db, WhereLoop *p){ + if( p->aLTerm!=p->aLTermSpace ) sqlite3DbFreeNN(db, p->aLTerm); + whereLoopClearUnion(db, p); + whereLoopInit(p); +} + +/* +** Increase the memory allocation for pLoop->aLTerm[] to be at least n. +*/ +static int whereLoopResize(sqlite3 *db, WhereLoop *p, int n){ + WhereTerm **paNew; + if( p->nLSlot>=n ) return SQLITE_OK; + n = (n+7)&~7; + paNew = sqlite3DbMallocRawNN(db, sizeof(p->aLTerm[0])*n); + if( paNew==0 ) return SQLITE_NOMEM_BKPT; + memcpy(paNew, p->aLTerm, sizeof(p->aLTerm[0])*p->nLSlot); + if( p->aLTerm!=p->aLTermSpace ) sqlite3DbFreeNN(db, p->aLTerm); + p->aLTerm = paNew; + p->nLSlot = n; + return SQLITE_OK; +} + +/* +** Transfer content from the second pLoop into the first. +*/ +static int whereLoopXfer(sqlite3 *db, WhereLoop *pTo, WhereLoop *pFrom){ + whereLoopClearUnion(db, pTo); + if( whereLoopResize(db, pTo, pFrom->nLTerm) ){ + memset(&pTo->u, 0, sizeof(pTo->u)); + return SQLITE_NOMEM_BKPT; + } + memcpy(pTo, pFrom, WHERE_LOOP_XFER_SZ); + memcpy(pTo->aLTerm, pFrom->aLTerm, pTo->nLTerm*sizeof(pTo->aLTerm[0])); + if( pFrom->wsFlags & WHERE_VIRTUALTABLE ){ + pFrom->u.vtab.needFree = 0; + }else if( (pFrom->wsFlags & WHERE_AUTO_INDEX)!=0 ){ + pFrom->u.btree.pIndex = 0; + } + return SQLITE_OK; +} + +/* +** Delete a WhereLoop object +*/ +static void whereLoopDelete(sqlite3 *db, WhereLoop *p){ + whereLoopClear(db, p); + sqlite3DbFreeNN(db, p); +} + +/* +** Free a WhereInfo structure +*/ +static void whereInfoFree(sqlite3 *db, WhereInfo *pWInfo){ + int i; + assert( pWInfo!=0 ); + for(i=0; inLevel; i++){ + WhereLevel *pLevel = &pWInfo->a[i]; + if( pLevel->pWLoop && (pLevel->pWLoop->wsFlags & WHERE_IN_ABLE) ){ + sqlite3DbFree(db, pLevel->u.in.aInLoop); + } + } + sqlite3WhereClauseClear(&pWInfo->sWC); + while( pWInfo->pLoops ){ + WhereLoop *p = pWInfo->pLoops; + pWInfo->pLoops = p->pNextLoop; + whereLoopDelete(db, p); + } + sqlite3DbFreeNN(db, pWInfo); +} + +/* +** Return TRUE if all of the following are true: +** +** (1) X has the same or lower cost that Y +** (2) X uses fewer WHERE clause terms than Y +** (3) Every WHERE clause term used by X is also used by Y +** (4) X skips at least as many columns as Y +** (5) If X is a covering index, than Y is too +** +** Conditions (2) and (3) mean that X is a "proper subset" of Y. +** If X is a proper subset of Y then Y is a better choice and ought +** to have a lower cost. This routine returns TRUE when that cost +** relationship is inverted and needs to be adjusted. Constraint (4) +** was added because if X uses skip-scan less than Y it still might +** deserve a lower cost even if it is a proper subset of Y. Constraint (5) +** was added because a covering index probably deserves to have a lower cost +** than a non-covering index even if it is a proper subset. +*/ +static int whereLoopCheaperProperSubset( + const WhereLoop *pX, /* First WhereLoop to compare */ + const WhereLoop *pY /* Compare against this WhereLoop */ +){ + int i, j; + if( pX->nLTerm-pX->nSkip >= pY->nLTerm-pY->nSkip ){ + return 0; /* X is not a subset of Y */ + } + if( pY->nSkip > pX->nSkip ) return 0; + if( pX->rRun >= pY->rRun ){ + if( pX->rRun > pY->rRun ) return 0; /* X costs more than Y */ + if( pX->nOut > pY->nOut ) return 0; /* X costs more than Y */ + } + for(i=pX->nLTerm-1; i>=0; i--){ + if( pX->aLTerm[i]==0 ) continue; + for(j=pY->nLTerm-1; j>=0; j--){ + if( pY->aLTerm[j]==pX->aLTerm[i] ) break; + } + if( j<0 ) return 0; /* X not a subset of Y since term X[i] not used by Y */ + } + if( (pX->wsFlags&WHERE_IDX_ONLY)!=0 + && (pY->wsFlags&WHERE_IDX_ONLY)==0 ){ + return 0; /* Constraint (5) */ + } + return 1; /* All conditions meet */ +} + +/* +** Try to adjust the cost of WhereLoop pTemplate upwards or downwards so +** that: +** +** (1) pTemplate costs less than any other WhereLoops that are a proper +** subset of pTemplate +** +** (2) pTemplate costs more than any other WhereLoops for which pTemplate +** is a proper subset. +** +** To say "WhereLoop X is a proper subset of Y" means that X uses fewer +** WHERE clause terms than Y and that every WHERE clause term used by X is +** also used by Y. +*/ +static void whereLoopAdjustCost(const WhereLoop *p, WhereLoop *pTemplate){ + if( (pTemplate->wsFlags & WHERE_INDEXED)==0 ) return; + for(; p; p=p->pNextLoop){ + if( p->iTab!=pTemplate->iTab ) continue; + if( (p->wsFlags & WHERE_INDEXED)==0 ) continue; + if( whereLoopCheaperProperSubset(p, pTemplate) ){ + /* Adjust pTemplate cost downward so that it is cheaper than its + ** subset p. */ + WHERETRACE(0x80,("subset cost adjustment %d,%d to %d,%d\n", + pTemplate->rRun, pTemplate->nOut, p->rRun, p->nOut-1)); + pTemplate->rRun = p->rRun; + pTemplate->nOut = p->nOut - 1; + }else if( whereLoopCheaperProperSubset(pTemplate, p) ){ + /* Adjust pTemplate cost upward so that it is costlier than p since + ** pTemplate is a proper subset of p */ + WHERETRACE(0x80,("subset cost adjustment %d,%d to %d,%d\n", + pTemplate->rRun, pTemplate->nOut, p->rRun, p->nOut+1)); + pTemplate->rRun = p->rRun; + pTemplate->nOut = p->nOut + 1; + } + } +} + +/* +** Search the list of WhereLoops in *ppPrev looking for one that can be +** replaced by pTemplate. +** +** Return NULL if pTemplate does not belong on the WhereLoop list. +** In other words if pTemplate ought to be dropped from further consideration. +** +** If pX is a WhereLoop that pTemplate can replace, then return the +** link that points to pX. +** +** If pTemplate cannot replace any existing element of the list but needs +** to be added to the list as a new entry, then return a pointer to the +** tail of the list. +*/ +static WhereLoop **whereLoopFindLesser( + WhereLoop **ppPrev, + const WhereLoop *pTemplate +){ + WhereLoop *p; + for(p=(*ppPrev); p; ppPrev=&p->pNextLoop, p=*ppPrev){ + if( p->iTab!=pTemplate->iTab || p->iSortIdx!=pTemplate->iSortIdx ){ + /* If either the iTab or iSortIdx values for two WhereLoop are different + ** then those WhereLoops need to be considered separately. Neither is + ** a candidate to replace the other. */ + continue; + } + /* In the current implementation, the rSetup value is either zero + ** or the cost of building an automatic index (NlogN) and the NlogN + ** is the same for compatible WhereLoops. */ + assert( p->rSetup==0 || pTemplate->rSetup==0 + || p->rSetup==pTemplate->rSetup ); + + /* whereLoopAddBtree() always generates and inserts the automatic index + ** case first. Hence compatible candidate WhereLoops never have a larger + ** rSetup. Call this SETUP-INVARIANT */ + assert( p->rSetup>=pTemplate->rSetup ); + + /* Any loop using an appliation-defined index (or PRIMARY KEY or + ** UNIQUE constraint) with one or more == constraints is better + ** than an automatic index. Unless it is a skip-scan. */ + if( (p->wsFlags & WHERE_AUTO_INDEX)!=0 + && (pTemplate->nSkip)==0 + && (pTemplate->wsFlags & WHERE_INDEXED)!=0 + && (pTemplate->wsFlags & WHERE_COLUMN_EQ)!=0 + && (p->prereq & pTemplate->prereq)==pTemplate->prereq + ){ + break; + } + + /* If existing WhereLoop p is better than pTemplate, pTemplate can be + ** discarded. WhereLoop p is better if: + ** (1) p has no more dependencies than pTemplate, and + ** (2) p has an equal or lower cost than pTemplate + */ + if( (p->prereq & pTemplate->prereq)==p->prereq /* (1) */ + && p->rSetup<=pTemplate->rSetup /* (2a) */ + && p->rRun<=pTemplate->rRun /* (2b) */ + && p->nOut<=pTemplate->nOut /* (2c) */ + ){ + return 0; /* Discard pTemplate */ + } + + /* If pTemplate is always better than p, then cause p to be overwritten + ** with pTemplate. pTemplate is better than p if: + ** (1) pTemplate has no more dependences than p, and + ** (2) pTemplate has an equal or lower cost than p. + */ + if( (p->prereq & pTemplate->prereq)==pTemplate->prereq /* (1) */ + && p->rRun>=pTemplate->rRun /* (2a) */ + && p->nOut>=pTemplate->nOut /* (2b) */ + ){ + assert( p->rSetup>=pTemplate->rSetup ); /* SETUP-INVARIANT above */ + break; /* Cause p to be overwritten by pTemplate */ + } + } + return ppPrev; +} + +/* +** Insert or replace a WhereLoop entry using the template supplied. +** +** An existing WhereLoop entry might be overwritten if the new template +** is better and has fewer dependencies. Or the template will be ignored +** and no insert will occur if an existing WhereLoop is faster and has +** fewer dependencies than the template. Otherwise a new WhereLoop is +** added based on the template. +** +** If pBuilder->pOrSet is not NULL then we care about only the +** prerequisites and rRun and nOut costs of the N best loops. That +** information is gathered in the pBuilder->pOrSet object. This special +** processing mode is used only for OR clause processing. +** +** When accumulating multiple loops (when pBuilder->pOrSet is NULL) we +** still might overwrite similar loops with the new template if the +** new template is better. Loops may be overwritten if the following +** conditions are met: +** +** (1) They have the same iTab. +** (2) They have the same iSortIdx. +** (3) The template has same or fewer dependencies than the current loop +** (4) The template has the same or lower cost than the current loop +*/ +static int whereLoopInsert(WhereLoopBuilder *pBuilder, WhereLoop *pTemplate){ + WhereLoop **ppPrev, *p; + WhereInfo *pWInfo = pBuilder->pWInfo; + sqlite3 *db = pWInfo->pParse->db; + int rc; + + /* Stop the search once we hit the query planner search limit */ + if( pBuilder->iPlanLimit==0 ){ + WHERETRACE(0xffffffff,("=== query planner search limit reached ===\n")); + if( pBuilder->pOrSet ) pBuilder->pOrSet->n = 0; + return SQLITE_DONE; + } + pBuilder->iPlanLimit--; + + /* If pBuilder->pOrSet is defined, then only keep track of the costs + ** and prereqs. + */ + if( pBuilder->pOrSet!=0 ){ + if( pTemplate->nLTerm ){ +#if WHERETRACE_ENABLED + u16 n = pBuilder->pOrSet->n; + int x = +#endif + whereOrInsert(pBuilder->pOrSet, pTemplate->prereq, pTemplate->rRun, + pTemplate->nOut); +#if WHERETRACE_ENABLED /* 0x8 */ + if( sqlite3WhereTrace & 0x8 ){ + sqlite3DebugPrintf(x?" or-%d: ":" or-X: ", n); + whereLoopPrint(pTemplate, pBuilder->pWC); + } +#endif + } + return SQLITE_OK; + } + + /* Look for an existing WhereLoop to replace with pTemplate + */ + whereLoopAdjustCost(pWInfo->pLoops, pTemplate); + ppPrev = whereLoopFindLesser(&pWInfo->pLoops, pTemplate); + + if( ppPrev==0 ){ + /* There already exists a WhereLoop on the list that is better + ** than pTemplate, so just ignore pTemplate */ +#if WHERETRACE_ENABLED /* 0x8 */ + if( sqlite3WhereTrace & 0x8 ){ + sqlite3DebugPrintf(" skip: "); + whereLoopPrint(pTemplate, pBuilder->pWC); + } +#endif + return SQLITE_OK; + }else{ + p = *ppPrev; + } + + /* If we reach this point it means that either p[] should be overwritten + ** with pTemplate[] if p[] exists, or if p==NULL then allocate a new + ** WhereLoop and insert it. + */ +#if WHERETRACE_ENABLED /* 0x8 */ + if( sqlite3WhereTrace & 0x8 ){ + if( p!=0 ){ + sqlite3DebugPrintf("replace: "); + whereLoopPrint(p, pBuilder->pWC); + sqlite3DebugPrintf(" with: "); + }else{ + sqlite3DebugPrintf(" add: "); + } + whereLoopPrint(pTemplate, pBuilder->pWC); + } +#endif + if( p==0 ){ + /* Allocate a new WhereLoop to add to the end of the list */ + *ppPrev = p = sqlite3DbMallocRawNN(db, sizeof(WhereLoop)); + if( p==0 ) return SQLITE_NOMEM_BKPT; + whereLoopInit(p); + p->pNextLoop = 0; + }else{ + /* We will be overwriting WhereLoop p[]. But before we do, first + ** go through the rest of the list and delete any other entries besides + ** p[] that are also supplated by pTemplate */ + WhereLoop **ppTail = &p->pNextLoop; + WhereLoop *pToDel; + while( *ppTail ){ + ppTail = whereLoopFindLesser(ppTail, pTemplate); + if( ppTail==0 ) break; + pToDel = *ppTail; + if( pToDel==0 ) break; + *ppTail = pToDel->pNextLoop; +#if WHERETRACE_ENABLED /* 0x8 */ + if( sqlite3WhereTrace & 0x8 ){ + sqlite3DebugPrintf(" delete: "); + whereLoopPrint(pToDel, pBuilder->pWC); + } +#endif + whereLoopDelete(db, pToDel); + } + } + rc = whereLoopXfer(db, p, pTemplate); + if( (p->wsFlags & WHERE_VIRTUALTABLE)==0 ){ + Index *pIndex = p->u.btree.pIndex; + if( pIndex && pIndex->idxType==SQLITE_IDXTYPE_IPK ){ + p->u.btree.pIndex = 0; + } + } + return rc; +} + +/* +** Adjust the WhereLoop.nOut value downward to account for terms of the +** WHERE clause that reference the loop but which are not used by an +** index. +* +** For every WHERE clause term that is not used by the index +** and which has a truth probability assigned by one of the likelihood(), +** likely(), or unlikely() SQL functions, reduce the estimated number +** of output rows by the probability specified. +** +** TUNING: For every WHERE clause term that is not used by the index +** and which does not have an assigned truth probability, heuristics +** described below are used to try to estimate the truth probability. +** TODO --> Perhaps this is something that could be improved by better +** table statistics. +** +** Heuristic 1: Estimate the truth probability as 93.75%. The 93.75% +** value corresponds to -1 in LogEst notation, so this means decrement +** the WhereLoop.nOut field for every such WHERE clause term. +** +** Heuristic 2: If there exists one or more WHERE clause terms of the +** form "x==EXPR" and EXPR is not a constant 0 or 1, then make sure the +** final output row estimate is no greater than 1/4 of the total number +** of rows in the table. In other words, assume that x==EXPR will filter +** out at least 3 out of 4 rows. If EXPR is -1 or 0 or 1, then maybe the +** "x" column is boolean or else -1 or 0 or 1 is a common default value +** on the "x" column and so in that case only cap the output row estimate +** at 1/2 instead of 1/4. +*/ +static void whereLoopOutputAdjust( + WhereClause *pWC, /* The WHERE clause */ + WhereLoop *pLoop, /* The loop to adjust downward */ + LogEst nRow /* Number of rows in the entire table */ +){ + WhereTerm *pTerm, *pX; + Bitmask notAllowed = ~(pLoop->prereq|pLoop->maskSelf); + int i, j; + LogEst iReduce = 0; /* pLoop->nOut should not exceed nRow-iReduce */ + + assert( (pLoop->wsFlags & WHERE_AUTO_INDEX)==0 ); + for(i=pWC->nTerm, pTerm=pWC->a; i>0; i--, pTerm++){ + assert( pTerm!=0 ); + if( (pTerm->wtFlags & TERM_VIRTUAL)!=0 ) break; + if( (pTerm->prereqAll & pLoop->maskSelf)==0 ) continue; + if( (pTerm->prereqAll & notAllowed)!=0 ) continue; + for(j=pLoop->nLTerm-1; j>=0; j--){ + pX = pLoop->aLTerm[j]; + if( pX==0 ) continue; + if( pX==pTerm ) break; + if( pX->iParent>=0 && (&pWC->a[pX->iParent])==pTerm ) break; + } + if( j<0 ){ + if( pTerm->truthProb<=0 ){ + /* If a truth probability is specified using the likelihood() hints, + ** then use the probability provided by the application. */ + pLoop->nOut += pTerm->truthProb; + }else{ + /* In the absence of explicit truth probabilities, use heuristics to + ** guess a reasonable truth probability. */ + pLoop->nOut--; + if( pTerm->eOperator&(WO_EQ|WO_IS) ){ + Expr *pRight = pTerm->pExpr->pRight; + int k = 0; + testcase( pTerm->pExpr->op==TK_IS ); + if( sqlite3ExprIsInteger(pRight, &k) && k>=(-1) && k<=1 ){ + k = 10; + }else{ + k = 20; + } + if( iReducenOut > nRow-iReduce ) pLoop->nOut = nRow - iReduce; +} + +/* +** Term pTerm is a vector range comparison operation. The first comparison +** in the vector can be optimized using column nEq of the index. This +** function returns the total number of vector elements that can be used +** as part of the range comparison. +** +** For example, if the query is: +** +** WHERE a = ? AND (b, c, d) > (?, ?, ?) +** +** and the index: +** +** CREATE INDEX ... ON (a, b, c, d, e) +** +** then this function would be invoked with nEq=1. The value returned in +** this case is 3. +*/ +static int whereRangeVectorLen( + Parse *pParse, /* Parsing context */ + int iCur, /* Cursor open on pIdx */ + Index *pIdx, /* The index to be used for a inequality constraint */ + int nEq, /* Number of prior equality constraints on same index */ + WhereTerm *pTerm /* The vector inequality constraint */ +){ + int nCmp = sqlite3ExprVectorSize(pTerm->pExpr->pLeft); + int i; + + nCmp = MIN(nCmp, (pIdx->nColumn - nEq)); + for(i=1; ipExpr->pLeft->x.pList->a[i].pExpr; + Expr *pRhs = pTerm->pExpr->pRight; + if( pRhs->flags & EP_xIsSelect ){ + pRhs = pRhs->x.pSelect->pEList->a[i].pExpr; + }else{ + pRhs = pRhs->x.pList->a[i].pExpr; + } + + /* Check that the LHS of the comparison is a column reference to + ** the right column of the right source table. And that the sort + ** order of the index column is the same as the sort order of the + ** leftmost index column. */ + if( pLhs->op!=TK_COLUMN + || pLhs->iTable!=iCur + || pLhs->iColumn!=pIdx->aiColumn[i+nEq] + || pIdx->aSortOrder[i+nEq]!=pIdx->aSortOrder[nEq] + ){ + break; + } + + testcase( pLhs->iColumn==XN_ROWID ); + aff = sqlite3CompareAffinity(pRhs, sqlite3ExprAffinity(pLhs)); + idxaff = sqlite3TableColumnAffinity(pIdx->pTable, pLhs->iColumn); + if( aff!=idxaff ) break; + + pColl = sqlite3BinaryCompareCollSeq(pParse, pLhs, pRhs); + if( pColl==0 ) break; + if( sqlite3StrICmp(pColl->zName, pIdx->azColl[i+nEq]) ) break; + } + return i; +} + +/* +** Adjust the cost C by the costMult facter T. This only occurs if +** compiled with -DSQLITE_ENABLE_COSTMULT +*/ +#ifdef SQLITE_ENABLE_COSTMULT +# define ApplyCostMultiplier(C,T) C += T +#else +# define ApplyCostMultiplier(C,T) +#endif + +/* +** We have so far matched pBuilder->pNew->u.btree.nEq terms of the +** index pIndex. Try to match one more. +** +** When this function is called, pBuilder->pNew->nOut contains the +** number of rows expected to be visited by filtering using the nEq +** terms only. If it is modified, this value is restored before this +** function returns. +** +** If pProbe->idxType==SQLITE_IDXTYPE_IPK, that means pIndex is +** a fake index used for the INTEGER PRIMARY KEY. +*/ +static int whereLoopAddBtreeIndex( + WhereLoopBuilder *pBuilder, /* The WhereLoop factory */ + struct SrcList_item *pSrc, /* FROM clause term being analyzed */ + Index *pProbe, /* An index on pSrc */ + LogEst nInMul /* log(Number of iterations due to IN) */ +){ + WhereInfo *pWInfo = pBuilder->pWInfo; /* WHERE analyse context */ + Parse *pParse = pWInfo->pParse; /* Parsing context */ + sqlite3 *db = pParse->db; /* Database connection malloc context */ + WhereLoop *pNew; /* Template WhereLoop under construction */ + WhereTerm *pTerm; /* A WhereTerm under consideration */ + int opMask; /* Valid operators for constraints */ + WhereScan scan; /* Iterator for WHERE terms */ + Bitmask saved_prereq; /* Original value of pNew->prereq */ + u16 saved_nLTerm; /* Original value of pNew->nLTerm */ + u16 saved_nEq; /* Original value of pNew->u.btree.nEq */ + u16 saved_nBtm; /* Original value of pNew->u.btree.nBtm */ + u16 saved_nTop; /* Original value of pNew->u.btree.nTop */ + u16 saved_nSkip; /* Original value of pNew->nSkip */ + u32 saved_wsFlags; /* Original value of pNew->wsFlags */ + LogEst saved_nOut; /* Original value of pNew->nOut */ + int rc = SQLITE_OK; /* Return code */ + LogEst rSize; /* Number of rows in the table */ + LogEst rLogSize; /* Logarithm of table size */ + WhereTerm *pTop = 0, *pBtm = 0; /* Top and bottom range constraints */ + + pNew = pBuilder->pNew; + if( db->mallocFailed ) return SQLITE_NOMEM_BKPT; + WHERETRACE(0x800, ("BEGIN %s.addBtreeIdx(%s), nEq=%d\n", + pProbe->pTable->zName,pProbe->zName, pNew->u.btree.nEq)); + + assert( (pNew->wsFlags & WHERE_VIRTUALTABLE)==0 ); + assert( (pNew->wsFlags & WHERE_TOP_LIMIT)==0 ); + if( pNew->wsFlags & WHERE_BTM_LIMIT ){ + opMask = WO_LT|WO_LE; + }else{ + assert( pNew->u.btree.nBtm==0 ); + opMask = WO_EQ|WO_IN|WO_GT|WO_GE|WO_LT|WO_LE|WO_ISNULL|WO_IS; + } + if( pProbe->bUnordered ) opMask &= ~(WO_GT|WO_GE|WO_LT|WO_LE); + + assert( pNew->u.btree.nEqnColumn ); + + saved_nEq = pNew->u.btree.nEq; + saved_nBtm = pNew->u.btree.nBtm; + saved_nTop = pNew->u.btree.nTop; + saved_nSkip = pNew->nSkip; + saved_nLTerm = pNew->nLTerm; + saved_wsFlags = pNew->wsFlags; + saved_prereq = pNew->prereq; + saved_nOut = pNew->nOut; + pTerm = whereScanInit(&scan, pBuilder->pWC, pSrc->iCursor, saved_nEq, + opMask, pProbe); + pNew->rSetup = 0; + rSize = pProbe->aiRowLogEst[0]; + rLogSize = estLog(rSize); + for(; rc==SQLITE_OK && pTerm!=0; pTerm = whereScanNext(&scan)){ + u16 eOp = pTerm->eOperator; /* Shorthand for pTerm->eOperator */ + LogEst rCostIdx; + LogEst nOutUnadjusted; /* nOut before IN() and WHERE adjustments */ + int nIn = 0; +#ifdef SQLITE_ENABLE_STAT4 + int nRecValid = pBuilder->nRecValid; +#endif + if( (eOp==WO_ISNULL || (pTerm->wtFlags&TERM_VNULL)!=0) + && indexColumnNotNull(pProbe, saved_nEq) + ){ + continue; /* ignore IS [NOT] NULL constraints on NOT NULL columns */ + } + if( pTerm->prereqRight & pNew->maskSelf ) continue; + + /* Do not allow the upper bound of a LIKE optimization range constraint + ** to mix with a lower range bound from some other source */ + if( pTerm->wtFlags & TERM_LIKEOPT && pTerm->eOperator==WO_LT ) continue; + + /* Do not allow constraints from the WHERE clause to be used by the + ** right table of a LEFT JOIN. Only constraints in the ON clause are + ** allowed */ + if( (pSrc->fg.jointype & JT_LEFT)!=0 + && !ExprHasProperty(pTerm->pExpr, EP_FromJoin) + ){ + continue; + } + + if( IsUniqueIndex(pProbe) && saved_nEq==pProbe->nKeyCol-1 ){ + pBuilder->bldFlags |= SQLITE_BLDF_UNIQUE; + }else{ + pBuilder->bldFlags |= SQLITE_BLDF_INDEXED; + } + pNew->wsFlags = saved_wsFlags; + pNew->u.btree.nEq = saved_nEq; + pNew->u.btree.nBtm = saved_nBtm; + pNew->u.btree.nTop = saved_nTop; + pNew->nLTerm = saved_nLTerm; + if( whereLoopResize(db, pNew, pNew->nLTerm+1) ) break; /* OOM */ + pNew->aLTerm[pNew->nLTerm++] = pTerm; + pNew->prereq = (saved_prereq | pTerm->prereqRight) & ~pNew->maskSelf; + + assert( nInMul==0 + || (pNew->wsFlags & WHERE_COLUMN_NULL)!=0 + || (pNew->wsFlags & WHERE_COLUMN_IN)!=0 + || (pNew->wsFlags & WHERE_SKIPSCAN)!=0 + ); + + if( eOp & WO_IN ){ + Expr *pExpr = pTerm->pExpr; + if( ExprHasProperty(pExpr, EP_xIsSelect) ){ + /* "x IN (SELECT ...)": TUNING: the SELECT returns 25 rows */ + int i; + nIn = 46; assert( 46==sqlite3LogEst(25) ); + + /* The expression may actually be of the form (x, y) IN (SELECT...). + ** In this case there is a separate term for each of (x) and (y). + ** However, the nIn multiplier should only be applied once, not once + ** for each such term. The following loop checks that pTerm is the + ** first such term in use, and sets nIn back to 0 if it is not. */ + for(i=0; inLTerm-1; i++){ + if( pNew->aLTerm[i] && pNew->aLTerm[i]->pExpr==pExpr ) nIn = 0; + } + }else if( ALWAYS(pExpr->x.pList && pExpr->x.pList->nExpr) ){ + /* "x IN (value, value, ...)" */ + nIn = sqlite3LogEst(pExpr->x.pList->nExpr); + } + if( pProbe->hasStat1 ){ + LogEst M, logK, safetyMargin; + /* Let: + ** N = the total number of rows in the table + ** K = the number of entries on the RHS of the IN operator + ** M = the number of rows in the table that match terms to the + ** to the left in the same index. If the IN operator is on + ** the left-most index column, M==N. + ** + ** Given the definitions above, it is better to omit the IN operator + ** from the index lookup and instead do a scan of the M elements, + ** testing each scanned row against the IN operator separately, if: + ** + ** M*log(K) < K*log(N) + ** + ** Our estimates for M, K, and N might be inaccurate, so we build in + ** a safety margin of 2 (LogEst: 10) that favors using the IN operator + ** with the index, as using an index has better worst-case behavior. + ** If we do not have real sqlite_stat1 data, always prefer to use + ** the index. + */ + M = pProbe->aiRowLogEst[saved_nEq]; + logK = estLog(nIn); + safetyMargin = 10; /* TUNING: extra weight for indexed IN */ + if( M + logK + safetyMargin < nIn + rLogSize ){ + WHERETRACE(0x40, + ("Scan preferred over IN operator on column %d of \"%s\" (%d<%d)\n", + saved_nEq, pProbe->zName, M+logK+10, nIn+rLogSize)); + continue; + }else{ + WHERETRACE(0x40, + ("IN operator preferred on column %d of \"%s\" (%d>=%d)\n", + saved_nEq, pProbe->zName, M+logK+10, nIn+rLogSize)); + } + } + pNew->wsFlags |= WHERE_COLUMN_IN; + }else if( eOp & (WO_EQ|WO_IS) ){ + int iCol = pProbe->aiColumn[saved_nEq]; + pNew->wsFlags |= WHERE_COLUMN_EQ; + assert( saved_nEq==pNew->u.btree.nEq ); + if( iCol==XN_ROWID + || (iCol>=0 && nInMul==0 && saved_nEq==pProbe->nKeyCol-1) + ){ + if( iCol==XN_ROWID || pProbe->uniqNotNull + || (pProbe->nKeyCol==1 && pProbe->onError && eOp==WO_EQ) + ){ + pNew->wsFlags |= WHERE_ONEROW; + }else{ + pNew->wsFlags |= WHERE_UNQ_WANTED; + } + } + }else if( eOp & WO_ISNULL ){ + pNew->wsFlags |= WHERE_COLUMN_NULL; + }else if( eOp & (WO_GT|WO_GE) ){ + testcase( eOp & WO_GT ); + testcase( eOp & WO_GE ); + pNew->wsFlags |= WHERE_COLUMN_RANGE|WHERE_BTM_LIMIT; + pNew->u.btree.nBtm = whereRangeVectorLen( + pParse, pSrc->iCursor, pProbe, saved_nEq, pTerm + ); + pBtm = pTerm; + pTop = 0; + if( pTerm->wtFlags & TERM_LIKEOPT ){ + /* Range contraints that come from the LIKE optimization are + ** always used in pairs. */ + pTop = &pTerm[1]; + assert( (pTop-(pTerm->pWC->a))pWC->nTerm ); + assert( pTop->wtFlags & TERM_LIKEOPT ); + assert( pTop->eOperator==WO_LT ); + if( whereLoopResize(db, pNew, pNew->nLTerm+1) ) break; /* OOM */ + pNew->aLTerm[pNew->nLTerm++] = pTop; + pNew->wsFlags |= WHERE_TOP_LIMIT; + pNew->u.btree.nTop = 1; + } + }else{ + assert( eOp & (WO_LT|WO_LE) ); + testcase( eOp & WO_LT ); + testcase( eOp & WO_LE ); + pNew->wsFlags |= WHERE_COLUMN_RANGE|WHERE_TOP_LIMIT; + pNew->u.btree.nTop = whereRangeVectorLen( + pParse, pSrc->iCursor, pProbe, saved_nEq, pTerm + ); + pTop = pTerm; + pBtm = (pNew->wsFlags & WHERE_BTM_LIMIT)!=0 ? + pNew->aLTerm[pNew->nLTerm-2] : 0; + } + + /* At this point pNew->nOut is set to the number of rows expected to + ** be visited by the index scan before considering term pTerm, or the + ** values of nIn and nInMul. In other words, assuming that all + ** "x IN(...)" terms are replaced with "x = ?". This block updates + ** the value of pNew->nOut to account for pTerm (but not nIn/nInMul). */ + assert( pNew->nOut==saved_nOut ); + if( pNew->wsFlags & WHERE_COLUMN_RANGE ){ + /* Adjust nOut using stat4 data. Or, if there is no stat4 + ** data, using some other estimate. */ + whereRangeScanEst(pParse, pBuilder, pBtm, pTop, pNew); + }else{ + int nEq = ++pNew->u.btree.nEq; + assert( eOp & (WO_ISNULL|WO_EQ|WO_IN|WO_IS) ); + + assert( pNew->nOut==saved_nOut ); + if( pTerm->truthProb<=0 && pProbe->aiColumn[saved_nEq]>=0 ){ + assert( (eOp & WO_IN) || nIn==0 ); + testcase( eOp & WO_IN ); + pNew->nOut += pTerm->truthProb; + pNew->nOut -= nIn; + }else{ +#ifdef SQLITE_ENABLE_STAT4 + tRowcnt nOut = 0; + if( nInMul==0 + && pProbe->nSample + && pNew->u.btree.nEq<=pProbe->nSampleCol + && ((eOp & WO_IN)==0 || !ExprHasProperty(pTerm->pExpr, EP_xIsSelect)) + && OptimizationEnabled(db, SQLITE_Stat4) + ){ + Expr *pExpr = pTerm->pExpr; + if( (eOp & (WO_EQ|WO_ISNULL|WO_IS))!=0 ){ + testcase( eOp & WO_EQ ); + testcase( eOp & WO_IS ); + testcase( eOp & WO_ISNULL ); + rc = whereEqualScanEst(pParse, pBuilder, pExpr->pRight, &nOut); + }else{ + rc = whereInScanEst(pParse, pBuilder, pExpr->x.pList, &nOut); + } + if( rc==SQLITE_NOTFOUND ) rc = SQLITE_OK; + if( rc!=SQLITE_OK ) break; /* Jump out of the pTerm loop */ + if( nOut ){ + pNew->nOut = sqlite3LogEst(nOut); + if( pNew->nOut>saved_nOut ) pNew->nOut = saved_nOut; + pNew->nOut -= nIn; + } + } + if( nOut==0 ) +#endif + { + pNew->nOut += (pProbe->aiRowLogEst[nEq] - pProbe->aiRowLogEst[nEq-1]); + if( eOp & WO_ISNULL ){ + /* TUNING: If there is no likelihood() value, assume that a + ** "col IS NULL" expression matches twice as many rows + ** as (col=?). */ + pNew->nOut += 10; + } + } + } + } + + /* Set rCostIdx to the cost of visiting selected rows in index. Add + ** it to pNew->rRun, which is currently set to the cost of the index + ** seek only. Then, if this is a non-covering index, add the cost of + ** visiting the rows in the main table. */ + assert( pSrc->pTab->szTabRow>0 ); + rCostIdx = pNew->nOut + 1 + (15*pProbe->szIdxRow)/pSrc->pTab->szTabRow; + pNew->rRun = sqlite3LogEstAdd(rLogSize, rCostIdx); + if( (pNew->wsFlags & (WHERE_IDX_ONLY|WHERE_IPK))==0 ){ + pNew->rRun = sqlite3LogEstAdd(pNew->rRun, pNew->nOut + 16); + } + ApplyCostMultiplier(pNew->rRun, pProbe->pTable->costMult); + + nOutUnadjusted = pNew->nOut; + pNew->rRun += nInMul + nIn; + pNew->nOut += nInMul + nIn; + whereLoopOutputAdjust(pBuilder->pWC, pNew, rSize); + rc = whereLoopInsert(pBuilder, pNew); + + if( pNew->wsFlags & WHERE_COLUMN_RANGE ){ + pNew->nOut = saved_nOut; + }else{ + pNew->nOut = nOutUnadjusted; + } + + if( (pNew->wsFlags & WHERE_TOP_LIMIT)==0 + && pNew->u.btree.nEqnColumn + ){ + whereLoopAddBtreeIndex(pBuilder, pSrc, pProbe, nInMul+nIn); + } + pNew->nOut = saved_nOut; +#ifdef SQLITE_ENABLE_STAT4 + pBuilder->nRecValid = nRecValid; +#endif + } + pNew->prereq = saved_prereq; + pNew->u.btree.nEq = saved_nEq; + pNew->u.btree.nBtm = saved_nBtm; + pNew->u.btree.nTop = saved_nTop; + pNew->nSkip = saved_nSkip; + pNew->wsFlags = saved_wsFlags; + pNew->nOut = saved_nOut; + pNew->nLTerm = saved_nLTerm; + + /* Consider using a skip-scan if there are no WHERE clause constraints + ** available for the left-most terms of the index, and if the average + ** number of repeats in the left-most terms is at least 18. + ** + ** The magic number 18 is selected on the basis that scanning 17 rows + ** is almost always quicker than an index seek (even though if the index + ** contains fewer than 2^17 rows we assume otherwise in other parts of + ** the code). And, even if it is not, it should not be too much slower. + ** On the other hand, the extra seeks could end up being significantly + ** more expensive. */ + assert( 42==sqlite3LogEst(18) ); + if( saved_nEq==saved_nSkip + && saved_nEq+1nKeyCol + && pProbe->noSkipScan==0 + && OptimizationEnabled(db, SQLITE_SkipScan) + && pProbe->aiRowLogEst[saved_nEq+1]>=42 /* TUNING: Minimum for skip-scan */ + && (rc = whereLoopResize(db, pNew, pNew->nLTerm+1))==SQLITE_OK + ){ + LogEst nIter; + pNew->u.btree.nEq++; + pNew->nSkip++; + pNew->aLTerm[pNew->nLTerm++] = 0; + pNew->wsFlags |= WHERE_SKIPSCAN; + nIter = pProbe->aiRowLogEst[saved_nEq] - pProbe->aiRowLogEst[saved_nEq+1]; + pNew->nOut -= nIter; + /* TUNING: Because uncertainties in the estimates for skip-scan queries, + ** add a 1.375 fudge factor to make skip-scan slightly less likely. */ + nIter += 5; + whereLoopAddBtreeIndex(pBuilder, pSrc, pProbe, nIter + nInMul); + pNew->nOut = saved_nOut; + pNew->u.btree.nEq = saved_nEq; + pNew->nSkip = saved_nSkip; + pNew->wsFlags = saved_wsFlags; + } + + WHERETRACE(0x800, ("END %s.addBtreeIdx(%s), nEq=%d, rc=%d\n", + pProbe->pTable->zName, pProbe->zName, saved_nEq, rc)); + return rc; +} + +/* +** Return True if it is possible that pIndex might be useful in +** implementing the ORDER BY clause in pBuilder. +** +** Return False if pBuilder does not contain an ORDER BY clause or +** if there is no way for pIndex to be useful in implementing that +** ORDER BY clause. +*/ +static int indexMightHelpWithOrderBy( + WhereLoopBuilder *pBuilder, + Index *pIndex, + int iCursor +){ + ExprList *pOB; + ExprList *aColExpr; + int ii, jj; + + if( pIndex->bUnordered ) return 0; + if( (pOB = pBuilder->pWInfo->pOrderBy)==0 ) return 0; + for(ii=0; iinExpr; ii++){ + Expr *pExpr = sqlite3ExprSkipCollateAndLikely(pOB->a[ii].pExpr); + if( pExpr->op==TK_COLUMN && pExpr->iTable==iCursor ){ + if( pExpr->iColumn<0 ) return 1; + for(jj=0; jjnKeyCol; jj++){ + if( pExpr->iColumn==pIndex->aiColumn[jj] ) return 1; + } + }else if( (aColExpr = pIndex->aColExpr)!=0 ){ + for(jj=0; jjnKeyCol; jj++){ + if( pIndex->aiColumn[jj]!=XN_EXPR ) continue; + if( sqlite3ExprCompareSkip(pExpr,aColExpr->a[jj].pExpr,iCursor)==0 ){ + return 1; + } + } + } + } + return 0; +} + +/* Check to see if a partial index with pPartIndexWhere can be used +** in the current query. Return true if it can be and false if not. +*/ +static int whereUsablePartialIndex(int iTab, WhereClause *pWC, Expr *pWhere){ + int i; + WhereTerm *pTerm; + Parse *pParse = pWC->pWInfo->pParse; + while( pWhere->op==TK_AND ){ + if( !whereUsablePartialIndex(iTab,pWC,pWhere->pLeft) ) return 0; + pWhere = pWhere->pRight; + } + if( pParse->db->flags & SQLITE_EnableQPSG ) pParse = 0; + for(i=0, pTerm=pWC->a; inTerm; i++, pTerm++){ + Expr *pExpr; + if( pTerm->wtFlags & TERM_NOPARTIDX ) continue; + pExpr = pTerm->pExpr; + if( (!ExprHasProperty(pExpr, EP_FromJoin) || pExpr->iRightJoinTable==iTab) + && sqlite3ExprImpliesExpr(pParse, pExpr, pWhere, iTab) + ){ + return 1; + } + } + return 0; +} + +/* +** Add all WhereLoop objects for a single table of the join where the table +** is identified by pBuilder->pNew->iTab. That table is guaranteed to be +** a b-tree table, not a virtual table. +** +** The costs (WhereLoop.rRun) of the b-tree loops added by this function +** are calculated as follows: +** +** For a full scan, assuming the table (or index) contains nRow rows: +** +** cost = nRow * 3.0 // full-table scan +** cost = nRow * K // scan of covering index +** cost = nRow * (K+3.0) // scan of non-covering index +** +** where K is a value between 1.1 and 3.0 set based on the relative +** estimated average size of the index and table records. +** +** For an index scan, where nVisit is the number of index rows visited +** by the scan, and nSeek is the number of seek operations required on +** the index b-tree: +** +** cost = nSeek * (log(nRow) + K * nVisit) // covering index +** cost = nSeek * (log(nRow) + (K+3.0) * nVisit) // non-covering index +** +** Normally, nSeek is 1. nSeek values greater than 1 come about if the +** WHERE clause includes "x IN (....)" terms used in place of "x=?". Or when +** implicit "x IN (SELECT x FROM tbl)" terms are added for skip-scans. +** +** The estimated values (nRow, nVisit, nSeek) often contain a large amount +** of uncertainty. For this reason, scoring is designed to pick plans that +** "do the least harm" if the estimates are inaccurate. For example, a +** log(nRow) factor is omitted from a non-covering index scan in order to +** bias the scoring in favor of using an index, since the worst-case +** performance of using an index is far better than the worst-case performance +** of a full table scan. +*/ +static int whereLoopAddBtree( + WhereLoopBuilder *pBuilder, /* WHERE clause information */ + Bitmask mPrereq /* Extra prerequesites for using this table */ +){ + WhereInfo *pWInfo; /* WHERE analysis context */ + Index *pProbe; /* An index we are evaluating */ + Index sPk; /* A fake index object for the primary key */ + LogEst aiRowEstPk[2]; /* The aiRowLogEst[] value for the sPk index */ + i16 aiColumnPk = -1; /* The aColumn[] value for the sPk index */ + SrcList *pTabList; /* The FROM clause */ + struct SrcList_item *pSrc; /* The FROM clause btree term to add */ + WhereLoop *pNew; /* Template WhereLoop object */ + int rc = SQLITE_OK; /* Return code */ + int iSortIdx = 1; /* Index number */ + int b; /* A boolean value */ + LogEst rSize; /* number of rows in the table */ + LogEst rLogSize; /* Logarithm of the number of rows in the table */ + WhereClause *pWC; /* The parsed WHERE clause */ + Table *pTab; /* Table being queried */ + + pNew = pBuilder->pNew; + pWInfo = pBuilder->pWInfo; + pTabList = pWInfo->pTabList; + pSrc = pTabList->a + pNew->iTab; + pTab = pSrc->pTab; + pWC = pBuilder->pWC; + assert( !IsVirtual(pSrc->pTab) ); + + if( pSrc->pIBIndex ){ + /* An INDEXED BY clause specifies a particular index to use */ + pProbe = pSrc->pIBIndex; + }else if( !HasRowid(pTab) ){ + pProbe = pTab->pIndex; + }else{ + /* There is no INDEXED BY clause. Create a fake Index object in local + ** variable sPk to represent the rowid primary key index. Make this + ** fake index the first in a chain of Index objects with all of the real + ** indices to follow */ + Index *pFirst; /* First of real indices on the table */ + memset(&sPk, 0, sizeof(Index)); + sPk.nKeyCol = 1; + sPk.nColumn = 1; + sPk.aiColumn = &aiColumnPk; + sPk.aiRowLogEst = aiRowEstPk; + sPk.onError = OE_Replace; + sPk.pTable = pTab; + sPk.szIdxRow = pTab->szTabRow; + sPk.idxType = SQLITE_IDXTYPE_IPK; + aiRowEstPk[0] = pTab->nRowLogEst; + aiRowEstPk[1] = 0; + pFirst = pSrc->pTab->pIndex; + if( pSrc->fg.notIndexed==0 ){ + /* The real indices of the table are only considered if the + ** NOT INDEXED qualifier is omitted from the FROM clause */ + sPk.pNext = pFirst; + } + pProbe = &sPk; + } + rSize = pTab->nRowLogEst; + rLogSize = estLog(rSize); + +#ifndef SQLITE_OMIT_AUTOMATIC_INDEX + /* Automatic indexes */ + if( !pBuilder->pOrSet /* Not part of an OR optimization */ + && (pWInfo->wctrlFlags & WHERE_OR_SUBCLAUSE)==0 + && (pWInfo->pParse->db->flags & SQLITE_AutoIndex)!=0 + && pSrc->pIBIndex==0 /* Has no INDEXED BY clause */ + && !pSrc->fg.notIndexed /* Has no NOT INDEXED clause */ + && HasRowid(pTab) /* Not WITHOUT ROWID table. (FIXME: Why not?) */ + && !pSrc->fg.isCorrelated /* Not a correlated subquery */ + && !pSrc->fg.isRecursive /* Not a recursive common table expression. */ + ){ + /* Generate auto-index WhereLoops */ + WhereTerm *pTerm; + WhereTerm *pWCEnd = pWC->a + pWC->nTerm; + for(pTerm=pWC->a; rc==SQLITE_OK && pTermprereqRight & pNew->maskSelf ) continue; + if( termCanDriveIndex(pTerm, pSrc, 0) ){ + pNew->u.btree.nEq = 1; + pNew->nSkip = 0; + pNew->u.btree.pIndex = 0; + pNew->nLTerm = 1; + pNew->aLTerm[0] = pTerm; + /* TUNING: One-time cost for computing the automatic index is + ** estimated to be X*N*log2(N) where N is the number of rows in + ** the table being indexed and where X is 7 (LogEst=28) for normal + ** tables or 0.5 (LogEst=-10) for views and subqueries. The value + ** of X is smaller for views and subqueries so that the query planner + ** will be more aggressive about generating automatic indexes for + ** those objects, since there is no opportunity to add schema + ** indexes on subqueries and views. */ + pNew->rSetup = rLogSize + rSize; + if( pTab->pSelect==0 && (pTab->tabFlags & TF_Ephemeral)==0 ){ + pNew->rSetup += 28; + }else{ + pNew->rSetup -= 10; + } + ApplyCostMultiplier(pNew->rSetup, pTab->costMult); + if( pNew->rSetup<0 ) pNew->rSetup = 0; + /* TUNING: Each index lookup yields 20 rows in the table. This + ** is more than the usual guess of 10 rows, since we have no way + ** of knowing how selective the index will ultimately be. It would + ** not be unreasonable to make this value much larger. */ + pNew->nOut = 43; assert( 43==sqlite3LogEst(20) ); + pNew->rRun = sqlite3LogEstAdd(rLogSize,pNew->nOut); + pNew->wsFlags = WHERE_AUTO_INDEX; + pNew->prereq = mPrereq | pTerm->prereqRight; + rc = whereLoopInsert(pBuilder, pNew); + } + } + } +#endif /* SQLITE_OMIT_AUTOMATIC_INDEX */ + + /* Loop over all indices. If there was an INDEXED BY clause, then only + ** consider index pProbe. */ + for(; rc==SQLITE_OK && pProbe; + pProbe=(pSrc->pIBIndex ? 0 : pProbe->pNext), iSortIdx++ + ){ + if( pProbe->pPartIdxWhere!=0 + && !whereUsablePartialIndex(pSrc->iCursor, pWC, pProbe->pPartIdxWhere) ){ + testcase( pNew->iTab!=pSrc->iCursor ); /* See ticket [98d973b8f5] */ + continue; /* Partial index inappropriate for this query */ + } + if( pProbe->bNoQuery ) continue; + rSize = pProbe->aiRowLogEst[0]; + pNew->u.btree.nEq = 0; + pNew->u.btree.nBtm = 0; + pNew->u.btree.nTop = 0; + pNew->nSkip = 0; + pNew->nLTerm = 0; + pNew->iSortIdx = 0; + pNew->rSetup = 0; + pNew->prereq = mPrereq; + pNew->nOut = rSize; + pNew->u.btree.pIndex = pProbe; + b = indexMightHelpWithOrderBy(pBuilder, pProbe, pSrc->iCursor); + /* The ONEPASS_DESIRED flags never occurs together with ORDER BY */ + assert( (pWInfo->wctrlFlags & WHERE_ONEPASS_DESIRED)==0 || b==0 ); + if( pProbe->idxType==SQLITE_IDXTYPE_IPK ){ + /* Integer primary key index */ + pNew->wsFlags = WHERE_IPK; + + /* Full table scan */ + pNew->iSortIdx = b ? iSortIdx : 0; + /* TUNING: Cost of full table scan is (N*3.0). */ + pNew->rRun = rSize + 16; + ApplyCostMultiplier(pNew->rRun, pTab->costMult); + whereLoopOutputAdjust(pWC, pNew, rSize); + rc = whereLoopInsert(pBuilder, pNew); + pNew->nOut = rSize; + if( rc ) break; + }else{ + Bitmask m; + if( pProbe->isCovering ){ + pNew->wsFlags = WHERE_IDX_ONLY | WHERE_INDEXED; + m = 0; + }else{ + m = pSrc->colUsed & pProbe->colNotIdxed; + pNew->wsFlags = (m==0) ? (WHERE_IDX_ONLY|WHERE_INDEXED) : WHERE_INDEXED; + } + + /* Full scan via index */ + if( b + || !HasRowid(pTab) + || pProbe->pPartIdxWhere!=0 + || ( m==0 + && pProbe->bUnordered==0 + && (pProbe->szIdxRowszTabRow) + && (pWInfo->wctrlFlags & WHERE_ONEPASS_DESIRED)==0 + && sqlite3GlobalConfig.bUseCis + && OptimizationEnabled(pWInfo->pParse->db, SQLITE_CoverIdxScan) + ) + ){ + pNew->iSortIdx = b ? iSortIdx : 0; + + /* The cost of visiting the index rows is N*K, where K is + ** between 1.1 and 3.0, depending on the relative sizes of the + ** index and table rows. */ + pNew->rRun = rSize + 1 + (15*pProbe->szIdxRow)/pTab->szTabRow; + if( m!=0 ){ + /* If this is a non-covering index scan, add in the cost of + ** doing table lookups. The cost will be 3x the number of + ** lookups. Take into account WHERE clause terms that can be + ** satisfied using just the index, and that do not require a + ** table lookup. */ + LogEst nLookup = rSize + 16; /* Base cost: N*3 */ + int ii; + int iCur = pSrc->iCursor; + WhereClause *pWC2 = &pWInfo->sWC; + for(ii=0; iinTerm; ii++){ + WhereTerm *pTerm = &pWC2->a[ii]; + if( !sqlite3ExprCoveredByIndex(pTerm->pExpr, iCur, pProbe) ){ + break; + } + /* pTerm can be evaluated using just the index. So reduce + ** the expected number of table lookups accordingly */ + if( pTerm->truthProb<=0 ){ + nLookup += pTerm->truthProb; + }else{ + nLookup--; + if( pTerm->eOperator & (WO_EQ|WO_IS) ) nLookup -= 19; + } + } + + pNew->rRun = sqlite3LogEstAdd(pNew->rRun, nLookup); + } + ApplyCostMultiplier(pNew->rRun, pTab->costMult); + whereLoopOutputAdjust(pWC, pNew, rSize); + rc = whereLoopInsert(pBuilder, pNew); + pNew->nOut = rSize; + if( rc ) break; + } + } + + pBuilder->bldFlags = 0; + rc = whereLoopAddBtreeIndex(pBuilder, pSrc, pProbe, 0); + if( pBuilder->bldFlags==SQLITE_BLDF_INDEXED ){ + /* If a non-unique index is used, or if a prefix of the key for + ** unique index is used (making the index functionally non-unique) + ** then the sqlite_stat1 data becomes important for scoring the + ** plan */ + pTab->tabFlags |= TF_StatsUsed; + } +#ifdef SQLITE_ENABLE_STAT4 + sqlite3Stat4ProbeFree(pBuilder->pRec); + pBuilder->nRecValid = 0; + pBuilder->pRec = 0; +#endif + } + return rc; +} + +#ifndef SQLITE_OMIT_VIRTUALTABLE + +/* +** Argument pIdxInfo is already populated with all constraints that may +** be used by the virtual table identified by pBuilder->pNew->iTab. This +** function marks a subset of those constraints usable, invokes the +** xBestIndex method and adds the returned plan to pBuilder. +** +** A constraint is marked usable if: +** +** * Argument mUsable indicates that its prerequisites are available, and +** +** * It is not one of the operators specified in the mExclude mask passed +** as the fourth argument (which in practice is either WO_IN or 0). +** +** Argument mPrereq is a mask of tables that must be scanned before the +** virtual table in question. These are added to the plans prerequisites +** before it is added to pBuilder. +** +** Output parameter *pbIn is set to true if the plan added to pBuilder +** uses one or more WO_IN terms, or false otherwise. +*/ +static int whereLoopAddVirtualOne( + WhereLoopBuilder *pBuilder, + Bitmask mPrereq, /* Mask of tables that must be used. */ + Bitmask mUsable, /* Mask of usable tables */ + u16 mExclude, /* Exclude terms using these operators */ + sqlite3_index_info *pIdxInfo, /* Populated object for xBestIndex */ + u16 mNoOmit, /* Do not omit these constraints */ + int *pbIn /* OUT: True if plan uses an IN(...) op */ +){ + WhereClause *pWC = pBuilder->pWC; + struct sqlite3_index_constraint *pIdxCons; + struct sqlite3_index_constraint_usage *pUsage = pIdxInfo->aConstraintUsage; + int i; + int mxTerm; + int rc = SQLITE_OK; + WhereLoop *pNew = pBuilder->pNew; + Parse *pParse = pBuilder->pWInfo->pParse; + struct SrcList_item *pSrc = &pBuilder->pWInfo->pTabList->a[pNew->iTab]; + int nConstraint = pIdxInfo->nConstraint; + + assert( (mUsable & mPrereq)==mPrereq ); + *pbIn = 0; + pNew->prereq = mPrereq; + + /* Set the usable flag on the subset of constraints identified by + ** arguments mUsable and mExclude. */ + pIdxCons = *(struct sqlite3_index_constraint**)&pIdxInfo->aConstraint; + for(i=0; ia[pIdxCons->iTermOffset]; + pIdxCons->usable = 0; + if( (pTerm->prereqRight & mUsable)==pTerm->prereqRight + && (pTerm->eOperator & mExclude)==0 + ){ + pIdxCons->usable = 1; + } + } + + /* Initialize the output fields of the sqlite3_index_info structure */ + memset(pUsage, 0, sizeof(pUsage[0])*nConstraint); + assert( pIdxInfo->needToFreeIdxStr==0 ); + pIdxInfo->idxStr = 0; + pIdxInfo->idxNum = 0; + pIdxInfo->orderByConsumed = 0; + pIdxInfo->estimatedCost = SQLITE_BIG_DBL / (double)2; + pIdxInfo->estimatedRows = 25; + pIdxInfo->idxFlags = 0; + pIdxInfo->colUsed = (sqlite3_int64)pSrc->colUsed; + + /* Invoke the virtual table xBestIndex() method */ + rc = vtabBestIndex(pParse, pSrc->pTab, pIdxInfo); + if( rc ){ + if( rc==SQLITE_CONSTRAINT ){ + /* If the xBestIndex method returns SQLITE_CONSTRAINT, that means + ** that the particular combination of parameters provided is unusable. + ** Make no entries in the loop table. + */ + WHERETRACE(0xffff, (" ^^^^--- non-viable plan rejected!\n")); + return SQLITE_OK; + } + return rc; + } + + mxTerm = -1; + assert( pNew->nLSlot>=nConstraint ); + for(i=0; iaLTerm[i] = 0; + pNew->u.vtab.omitMask = 0; + pIdxCons = *(struct sqlite3_index_constraint**)&pIdxInfo->aConstraint; + for(i=0; i=0 ){ + WhereTerm *pTerm; + int j = pIdxCons->iTermOffset; + if( iTerm>=nConstraint + || j<0 + || j>=pWC->nTerm + || pNew->aLTerm[iTerm]!=0 + || pIdxCons->usable==0 + ){ + sqlite3ErrorMsg(pParse,"%s.xBestIndex malfunction",pSrc->pTab->zName); + testcase( pIdxInfo->needToFreeIdxStr ); + return SQLITE_ERROR; + } + testcase( iTerm==nConstraint-1 ); + testcase( j==0 ); + testcase( j==pWC->nTerm-1 ); + pTerm = &pWC->a[j]; + pNew->prereq |= pTerm->prereqRight; + assert( iTermnLSlot ); + pNew->aLTerm[iTerm] = pTerm; + if( iTerm>mxTerm ) mxTerm = iTerm; + testcase( iTerm==15 ); + testcase( iTerm==16 ); + if( iTerm<16 && pUsage[i].omit ) pNew->u.vtab.omitMask |= 1<eOperator & WO_IN)!=0 ){ + /* A virtual table that is constrained by an IN clause may not + ** consume the ORDER BY clause because (1) the order of IN terms + ** is not necessarily related to the order of output terms and + ** (2) Multiple outputs from a single IN value will not merge + ** together. */ + pIdxInfo->orderByConsumed = 0; + pIdxInfo->idxFlags &= ~SQLITE_INDEX_SCAN_UNIQUE; + *pbIn = 1; assert( (mExclude & WO_IN)==0 ); + } + } + } + pNew->u.vtab.omitMask &= ~mNoOmit; + + pNew->nLTerm = mxTerm+1; + for(i=0; i<=mxTerm; i++){ + if( pNew->aLTerm[i]==0 ){ + /* The non-zero argvIdx values must be contiguous. Raise an + ** error if they are not */ + sqlite3ErrorMsg(pParse,"%s.xBestIndex malfunction",pSrc->pTab->zName); + testcase( pIdxInfo->needToFreeIdxStr ); + return SQLITE_ERROR; + } + } + assert( pNew->nLTerm<=pNew->nLSlot ); + pNew->u.vtab.idxNum = pIdxInfo->idxNum; + pNew->u.vtab.needFree = pIdxInfo->needToFreeIdxStr; + pIdxInfo->needToFreeIdxStr = 0; + pNew->u.vtab.idxStr = pIdxInfo->idxStr; + pNew->u.vtab.isOrdered = (i8)(pIdxInfo->orderByConsumed ? + pIdxInfo->nOrderBy : 0); + pNew->rSetup = 0; + pNew->rRun = sqlite3LogEstFromDouble(pIdxInfo->estimatedCost); + pNew->nOut = sqlite3LogEst(pIdxInfo->estimatedRows); + + /* Set the WHERE_ONEROW flag if the xBestIndex() method indicated + ** that the scan will visit at most one row. Clear it otherwise. */ + if( pIdxInfo->idxFlags & SQLITE_INDEX_SCAN_UNIQUE ){ + pNew->wsFlags |= WHERE_ONEROW; + }else{ + pNew->wsFlags &= ~WHERE_ONEROW; + } + rc = whereLoopInsert(pBuilder, pNew); + if( pNew->u.vtab.needFree ){ + sqlite3_free(pNew->u.vtab.idxStr); + pNew->u.vtab.needFree = 0; + } + WHERETRACE(0xffff, (" bIn=%d prereqIn=%04llx prereqOut=%04llx\n", + *pbIn, (sqlite3_uint64)mPrereq, + (sqlite3_uint64)(pNew->prereq & ~mPrereq))); + + return rc; +} + +/* +** If this function is invoked from within an xBestIndex() callback, it +** returns a pointer to a buffer containing the name of the collation +** sequence associated with element iCons of the sqlite3_index_info.aConstraint +** array. Or, if iCons is out of range or there is no active xBestIndex +** call, return NULL. +*/ +SQLITE_API const char *sqlite3_vtab_collation(sqlite3_index_info *pIdxInfo, int iCons){ + HiddenIndexInfo *pHidden = (HiddenIndexInfo*)&pIdxInfo[1]; + const char *zRet = 0; + if( iCons>=0 && iConsnConstraint ){ + CollSeq *pC = 0; + int iTerm = pIdxInfo->aConstraint[iCons].iTermOffset; + Expr *pX = pHidden->pWC->a[iTerm].pExpr; + if( pX->pLeft ){ + pC = sqlite3BinaryCompareCollSeq(pHidden->pParse, pX->pLeft, pX->pRight); + } + zRet = (pC ? pC->zName : sqlite3StrBINARY); + } + return zRet; +} + +/* +** Add all WhereLoop objects for a table of the join identified by +** pBuilder->pNew->iTab. That table is guaranteed to be a virtual table. +** +** If there are no LEFT or CROSS JOIN joins in the query, both mPrereq and +** mUnusable are set to 0. Otherwise, mPrereq is a mask of all FROM clause +** entries that occur before the virtual table in the FROM clause and are +** separated from it by at least one LEFT or CROSS JOIN. Similarly, the +** mUnusable mask contains all FROM clause entries that occur after the +** virtual table and are separated from it by at least one LEFT or +** CROSS JOIN. +** +** For example, if the query were: +** +** ... FROM t1, t2 LEFT JOIN t3, t4, vt CROSS JOIN t5, t6; +** +** then mPrereq corresponds to (t1, t2) and mUnusable to (t5, t6). +** +** All the tables in mPrereq must be scanned before the current virtual +** table. So any terms for which all prerequisites are satisfied by +** mPrereq may be specified as "usable" in all calls to xBestIndex. +** Conversely, all tables in mUnusable must be scanned after the current +** virtual table, so any terms for which the prerequisites overlap with +** mUnusable should always be configured as "not-usable" for xBestIndex. +*/ +static int whereLoopAddVirtual( + WhereLoopBuilder *pBuilder, /* WHERE clause information */ + Bitmask mPrereq, /* Tables that must be scanned before this one */ + Bitmask mUnusable /* Tables that must be scanned after this one */ +){ + int rc = SQLITE_OK; /* Return code */ + WhereInfo *pWInfo; /* WHERE analysis context */ + Parse *pParse; /* The parsing context */ + WhereClause *pWC; /* The WHERE clause */ + struct SrcList_item *pSrc; /* The FROM clause term to search */ + sqlite3_index_info *p; /* Object to pass to xBestIndex() */ + int nConstraint; /* Number of constraints in p */ + int bIn; /* True if plan uses IN(...) operator */ + WhereLoop *pNew; + Bitmask mBest; /* Tables used by best possible plan */ + u16 mNoOmit; + + assert( (mPrereq & mUnusable)==0 ); + pWInfo = pBuilder->pWInfo; + pParse = pWInfo->pParse; + pWC = pBuilder->pWC; + pNew = pBuilder->pNew; + pSrc = &pWInfo->pTabList->a[pNew->iTab]; + assert( IsVirtual(pSrc->pTab) ); + p = allocateIndexInfo(pParse, pWC, mUnusable, pSrc, pBuilder->pOrderBy, + &mNoOmit); + if( p==0 ) return SQLITE_NOMEM_BKPT; + pNew->rSetup = 0; + pNew->wsFlags = WHERE_VIRTUALTABLE; + pNew->nLTerm = 0; + pNew->u.vtab.needFree = 0; + nConstraint = p->nConstraint; + if( whereLoopResize(pParse->db, pNew, nConstraint) ){ + sqlite3DbFree(pParse->db, p); + return SQLITE_NOMEM_BKPT; + } + + /* First call xBestIndex() with all constraints usable. */ + WHERETRACE(0x800, ("BEGIN %s.addVirtual()\n", pSrc->pTab->zName)); + WHERETRACE(0x40, (" VirtualOne: all usable\n")); + rc = whereLoopAddVirtualOne(pBuilder, mPrereq, ALLBITS, 0, p, mNoOmit, &bIn); + + /* If the call to xBestIndex() with all terms enabled produced a plan + ** that does not require any source tables (IOW: a plan with mBest==0) + ** and does not use an IN(...) operator, then there is no point in making + ** any further calls to xBestIndex() since they will all return the same + ** result (if the xBestIndex() implementation is sane). */ + if( rc==SQLITE_OK && ((mBest = (pNew->prereq & ~mPrereq))!=0 || bIn) ){ + int seenZero = 0; /* True if a plan with no prereqs seen */ + int seenZeroNoIN = 0; /* Plan with no prereqs and no IN(...) seen */ + Bitmask mPrev = 0; + Bitmask mBestNoIn = 0; + + /* If the plan produced by the earlier call uses an IN(...) term, call + ** xBestIndex again, this time with IN(...) terms disabled. */ + if( bIn ){ + WHERETRACE(0x40, (" VirtualOne: all usable w/o IN\n")); + rc = whereLoopAddVirtualOne( + pBuilder, mPrereq, ALLBITS, WO_IN, p, mNoOmit, &bIn); + assert( bIn==0 ); + mBestNoIn = pNew->prereq & ~mPrereq; + if( mBestNoIn==0 ){ + seenZero = 1; + seenZeroNoIN = 1; + } + } + + /* Call xBestIndex once for each distinct value of (prereqRight & ~mPrereq) + ** in the set of terms that apply to the current virtual table. */ + while( rc==SQLITE_OK ){ + int i; + Bitmask mNext = ALLBITS; + assert( mNext>0 ); + for(i=0; ia[p->aConstraint[i].iTermOffset].prereqRight & ~mPrereq + ); + if( mThis>mPrev && mThisprereq==mPrereq ){ + seenZero = 1; + if( bIn==0 ) seenZeroNoIN = 1; + } + } + + /* If the calls to xBestIndex() in the above loop did not find a plan + ** that requires no source tables at all (i.e. one guaranteed to be + ** usable), make a call here with all source tables disabled */ + if( rc==SQLITE_OK && seenZero==0 ){ + WHERETRACE(0x40, (" VirtualOne: all disabled\n")); + rc = whereLoopAddVirtualOne( + pBuilder, mPrereq, mPrereq, 0, p, mNoOmit, &bIn); + if( bIn==0 ) seenZeroNoIN = 1; + } + + /* If the calls to xBestIndex() have so far failed to find a plan + ** that requires no source tables at all and does not use an IN(...) + ** operator, make a final call to obtain one here. */ + if( rc==SQLITE_OK && seenZeroNoIN==0 ){ + WHERETRACE(0x40, (" VirtualOne: all disabled and w/o IN\n")); + rc = whereLoopAddVirtualOne( + pBuilder, mPrereq, mPrereq, WO_IN, p, mNoOmit, &bIn); + } + } + + if( p->needToFreeIdxStr ) sqlite3_free(p->idxStr); + sqlite3DbFreeNN(pParse->db, p); + WHERETRACE(0x800, ("END %s.addVirtual(), rc=%d\n", pSrc->pTab->zName, rc)); + return rc; +} +#endif /* SQLITE_OMIT_VIRTUALTABLE */ + +/* +** Add WhereLoop entries to handle OR terms. This works for either +** btrees or virtual tables. +*/ +static int whereLoopAddOr( + WhereLoopBuilder *pBuilder, + Bitmask mPrereq, + Bitmask mUnusable +){ + WhereInfo *pWInfo = pBuilder->pWInfo; + WhereClause *pWC; + WhereLoop *pNew; + WhereTerm *pTerm, *pWCEnd; + int rc = SQLITE_OK; + int iCur; + WhereClause tempWC; + WhereLoopBuilder sSubBuild; + WhereOrSet sSum, sCur; + struct SrcList_item *pItem; + + pWC = pBuilder->pWC; + pWCEnd = pWC->a + pWC->nTerm; + pNew = pBuilder->pNew; + memset(&sSum, 0, sizeof(sSum)); + pItem = pWInfo->pTabList->a + pNew->iTab; + iCur = pItem->iCursor; + + for(pTerm=pWC->a; pTermeOperator & WO_OR)!=0 + && (pTerm->u.pOrInfo->indexable & pNew->maskSelf)!=0 + ){ + WhereClause * const pOrWC = &pTerm->u.pOrInfo->wc; + WhereTerm * const pOrWCEnd = &pOrWC->a[pOrWC->nTerm]; + WhereTerm *pOrTerm; + int once = 1; + int i, j; + + sSubBuild = *pBuilder; + sSubBuild.pOrderBy = 0; + sSubBuild.pOrSet = &sCur; + + WHERETRACE(0x200, ("Begin processing OR-clause %p\n", pTerm)); + for(pOrTerm=pOrWC->a; pOrTermeOperator & WO_AND)!=0 ){ + sSubBuild.pWC = &pOrTerm->u.pAndInfo->wc; + }else if( pOrTerm->leftCursor==iCur ){ + tempWC.pWInfo = pWC->pWInfo; + tempWC.pOuter = pWC; + tempWC.op = TK_AND; + tempWC.nTerm = 1; + tempWC.a = pOrTerm; + sSubBuild.pWC = &tempWC; + }else{ + continue; + } + sCur.n = 0; +#ifdef WHERETRACE_ENABLED + WHERETRACE(0x200, ("OR-term %d of %p has %d subterms:\n", + (int)(pOrTerm-pOrWC->a), pTerm, sSubBuild.pWC->nTerm)); + if( sqlite3WhereTrace & 0x400 ){ + sqlite3WhereClausePrint(sSubBuild.pWC); + } +#endif +#ifndef SQLITE_OMIT_VIRTUALTABLE + if( IsVirtual(pItem->pTab) ){ + rc = whereLoopAddVirtual(&sSubBuild, mPrereq, mUnusable); + }else +#endif + { + rc = whereLoopAddBtree(&sSubBuild, mPrereq); + } + if( rc==SQLITE_OK ){ + rc = whereLoopAddOr(&sSubBuild, mPrereq, mUnusable); + } + assert( rc==SQLITE_OK || sCur.n==0 ); + if( sCur.n==0 ){ + sSum.n = 0; + break; + }else if( once ){ + whereOrMove(&sSum, &sCur); + once = 0; + }else{ + WhereOrSet sPrev; + whereOrMove(&sPrev, &sSum); + sSum.n = 0; + for(i=0; inLTerm = 1; + pNew->aLTerm[0] = pTerm; + pNew->wsFlags = WHERE_MULTI_OR; + pNew->rSetup = 0; + pNew->iSortIdx = 0; + memset(&pNew->u, 0, sizeof(pNew->u)); + for(i=0; rc==SQLITE_OK && irRun = sSum.a[i].rRun + 1; + pNew->nOut = sSum.a[i].nOut; + pNew->prereq = sSum.a[i].prereq; + rc = whereLoopInsert(pBuilder, pNew); + } + WHERETRACE(0x200, ("End processing OR-clause %p\n", pTerm)); + } + } + return rc; +} + +/* +** Add all WhereLoop objects for all tables +*/ +static int whereLoopAddAll(WhereLoopBuilder *pBuilder){ + WhereInfo *pWInfo = pBuilder->pWInfo; + Bitmask mPrereq = 0; + Bitmask mPrior = 0; + int iTab; + SrcList *pTabList = pWInfo->pTabList; + struct SrcList_item *pItem; + struct SrcList_item *pEnd = &pTabList->a[pWInfo->nLevel]; + sqlite3 *db = pWInfo->pParse->db; + int rc = SQLITE_OK; + WhereLoop *pNew; + u8 priorJointype = 0; + + /* Loop over the tables in the join, from left to right */ + pNew = pBuilder->pNew; + whereLoopInit(pNew); + pBuilder->iPlanLimit = SQLITE_QUERY_PLANNER_LIMIT; + for(iTab=0, pItem=pTabList->a; pItemiTab = iTab; + pBuilder->iPlanLimit += SQLITE_QUERY_PLANNER_LIMIT_INCR; + pNew->maskSelf = sqlite3WhereGetMask(&pWInfo->sMaskSet, pItem->iCursor); + if( ((pItem->fg.jointype|priorJointype) & (JT_LEFT|JT_CROSS))!=0 ){ + /* This condition is true when pItem is the FROM clause term on the + ** right-hand-side of a LEFT or CROSS JOIN. */ + mPrereq = mPrior; + } + priorJointype = pItem->fg.jointype; +#ifndef SQLITE_OMIT_VIRTUALTABLE + if( IsVirtual(pItem->pTab) ){ + struct SrcList_item *p; + for(p=&pItem[1]; pfg.jointype & (JT_LEFT|JT_CROSS)) ){ + mUnusable |= sqlite3WhereGetMask(&pWInfo->sMaskSet, p->iCursor); + } + } + rc = whereLoopAddVirtual(pBuilder, mPrereq, mUnusable); + }else +#endif /* SQLITE_OMIT_VIRTUALTABLE */ + { + rc = whereLoopAddBtree(pBuilder, mPrereq); + } + if( rc==SQLITE_OK && pBuilder->pWC->hasOr ){ + rc = whereLoopAddOr(pBuilder, mPrereq, mUnusable); + } + mPrior |= pNew->maskSelf; + if( rc || db->mallocFailed ){ + if( rc==SQLITE_DONE ){ + /* We hit the query planner search limit set by iPlanLimit */ + sqlite3_log(SQLITE_WARNING, "abbreviated query algorithm search"); + rc = SQLITE_OK; + }else{ + break; + } + } + } + + whereLoopClear(db, pNew); + return rc; +} + +/* +** Examine a WherePath (with the addition of the extra WhereLoop of the 6th +** parameters) to see if it outputs rows in the requested ORDER BY +** (or GROUP BY) without requiring a separate sort operation. Return N: +** +** N>0: N terms of the ORDER BY clause are satisfied +** N==0: No terms of the ORDER BY clause are satisfied +** N<0: Unknown yet how many terms of ORDER BY might be satisfied. +** +** Note that processing for WHERE_GROUPBY and WHERE_DISTINCTBY is not as +** strict. With GROUP BY and DISTINCT the only requirement is that +** equivalent rows appear immediately adjacent to one another. GROUP BY +** and DISTINCT do not require rows to appear in any particular order as long +** as equivalent rows are grouped together. Thus for GROUP BY and DISTINCT +** the pOrderBy terms can be matched in any order. With ORDER BY, the +** pOrderBy terms must be matched in strict left-to-right order. +*/ +static i8 wherePathSatisfiesOrderBy( + WhereInfo *pWInfo, /* The WHERE clause */ + ExprList *pOrderBy, /* ORDER BY or GROUP BY or DISTINCT clause to check */ + WherePath *pPath, /* The WherePath to check */ + u16 wctrlFlags, /* WHERE_GROUPBY or _DISTINCTBY or _ORDERBY_LIMIT */ + u16 nLoop, /* Number of entries in pPath->aLoop[] */ + WhereLoop *pLast, /* Add this WhereLoop to the end of pPath->aLoop[] */ + Bitmask *pRevMask /* OUT: Mask of WhereLoops to run in reverse order */ +){ + u8 revSet; /* True if rev is known */ + u8 rev; /* Composite sort order */ + u8 revIdx; /* Index sort order */ + u8 isOrderDistinct; /* All prior WhereLoops are order-distinct */ + u8 distinctColumns; /* True if the loop has UNIQUE NOT NULL columns */ + u8 isMatch; /* iColumn matches a term of the ORDER BY clause */ + u16 eqOpMask; /* Allowed equality operators */ + u16 nKeyCol; /* Number of key columns in pIndex */ + u16 nColumn; /* Total number of ordered columns in the index */ + u16 nOrderBy; /* Number terms in the ORDER BY clause */ + int iLoop; /* Index of WhereLoop in pPath being processed */ + int i, j; /* Loop counters */ + int iCur; /* Cursor number for current WhereLoop */ + int iColumn; /* A column number within table iCur */ + WhereLoop *pLoop = 0; /* Current WhereLoop being processed. */ + WhereTerm *pTerm; /* A single term of the WHERE clause */ + Expr *pOBExpr; /* An expression from the ORDER BY clause */ + CollSeq *pColl; /* COLLATE function from an ORDER BY clause term */ + Index *pIndex; /* The index associated with pLoop */ + sqlite3 *db = pWInfo->pParse->db; /* Database connection */ + Bitmask obSat = 0; /* Mask of ORDER BY terms satisfied so far */ + Bitmask obDone; /* Mask of all ORDER BY terms */ + Bitmask orderDistinctMask; /* Mask of all well-ordered loops */ + Bitmask ready; /* Mask of inner loops */ + + /* + ** We say the WhereLoop is "one-row" if it generates no more than one + ** row of output. A WhereLoop is one-row if all of the following are true: + ** (a) All index columns match with WHERE_COLUMN_EQ. + ** (b) The index is unique + ** Any WhereLoop with an WHERE_COLUMN_EQ constraint on the rowid is one-row. + ** Every one-row WhereLoop will have the WHERE_ONEROW bit set in wsFlags. + ** + ** We say the WhereLoop is "order-distinct" if the set of columns from + ** that WhereLoop that are in the ORDER BY clause are different for every + ** row of the WhereLoop. Every one-row WhereLoop is automatically + ** order-distinct. A WhereLoop that has no columns in the ORDER BY clause + ** is not order-distinct. To be order-distinct is not quite the same as being + ** UNIQUE since a UNIQUE column or index can have multiple rows that + ** are NULL and NULL values are equivalent for the purpose of order-distinct. + ** To be order-distinct, the columns must be UNIQUE and NOT NULL. + ** + ** The rowid for a table is always UNIQUE and NOT NULL so whenever the + ** rowid appears in the ORDER BY clause, the corresponding WhereLoop is + ** automatically order-distinct. + */ + + assert( pOrderBy!=0 ); + if( nLoop && OptimizationDisabled(db, SQLITE_OrderByIdxJoin) ) return 0; + + nOrderBy = pOrderBy->nExpr; + testcase( nOrderBy==BMS-1 ); + if( nOrderBy>BMS-1 ) return 0; /* Cannot optimize overly large ORDER BYs */ + isOrderDistinct = 1; + obDone = MASKBIT(nOrderBy)-1; + orderDistinctMask = 0; + ready = 0; + eqOpMask = WO_EQ | WO_IS | WO_ISNULL; + if( wctrlFlags & WHERE_ORDERBY_LIMIT ) eqOpMask |= WO_IN; + for(iLoop=0; isOrderDistinct && obSat0 ) ready |= pLoop->maskSelf; + if( iLoopaLoop[iLoop]; + if( wctrlFlags & WHERE_ORDERBY_LIMIT ) continue; + }else{ + pLoop = pLast; + } + if( pLoop->wsFlags & WHERE_VIRTUALTABLE ){ + if( pLoop->u.vtab.isOrdered ) obSat = obDone; + break; + }else if( wctrlFlags & WHERE_DISTINCTBY ){ + pLoop->u.btree.nDistinctCol = 0; + } + iCur = pWInfo->pTabList->a[pLoop->iTab].iCursor; + + /* Mark off any ORDER BY term X that is a column in the table of + ** the current loop for which there is term in the WHERE + ** clause of the form X IS NULL or X=? that reference only outer + ** loops. + */ + for(i=0; ia[i].pExpr); + if( pOBExpr->op!=TK_COLUMN ) continue; + if( pOBExpr->iTable!=iCur ) continue; + pTerm = sqlite3WhereFindTerm(&pWInfo->sWC, iCur, pOBExpr->iColumn, + ~ready, eqOpMask, 0); + if( pTerm==0 ) continue; + if( pTerm->eOperator==WO_IN ){ + /* IN terms are only valid for sorting in the ORDER BY LIMIT + ** optimization, and then only if they are actually used + ** by the query plan */ + assert( wctrlFlags & WHERE_ORDERBY_LIMIT ); + for(j=0; jnLTerm && pTerm!=pLoop->aLTerm[j]; j++){} + if( j>=pLoop->nLTerm ) continue; + } + if( (pTerm->eOperator&(WO_EQ|WO_IS))!=0 && pOBExpr->iColumn>=0 ){ + if( sqlite3ExprCollSeqMatch(pWInfo->pParse, + pOrderBy->a[i].pExpr, pTerm->pExpr)==0 ){ + continue; + } + testcase( pTerm->pExpr->op==TK_IS ); + } + obSat |= MASKBIT(i); + } + + if( (pLoop->wsFlags & WHERE_ONEROW)==0 ){ + if( pLoop->wsFlags & WHERE_IPK ){ + pIndex = 0; + nKeyCol = 0; + nColumn = 1; + }else if( (pIndex = pLoop->u.btree.pIndex)==0 || pIndex->bUnordered ){ + return 0; + }else{ + nKeyCol = pIndex->nKeyCol; + nColumn = pIndex->nColumn; + assert( nColumn==nKeyCol+1 || !HasRowid(pIndex->pTable) ); + assert( pIndex->aiColumn[nColumn-1]==XN_ROWID + || !HasRowid(pIndex->pTable)); + isOrderDistinct = IsUniqueIndex(pIndex) + && (pLoop->wsFlags & WHERE_SKIPSCAN)==0; + } + + /* Loop through all columns of the index and deal with the ones + ** that are not constrained by == or IN. + */ + rev = revSet = 0; + distinctColumns = 0; + for(j=0; j=pLoop->u.btree.nEq + || (pLoop->aLTerm[j]==0)==(jnSkip) + ); + if( ju.btree.nEq && j>=pLoop->nSkip ){ + u16 eOp = pLoop->aLTerm[j]->eOperator; + + /* Skip over == and IS and ISNULL terms. (Also skip IN terms when + ** doing WHERE_ORDERBY_LIMIT processing). Except, IS and ISNULL + ** terms imply that the index is not UNIQUE NOT NULL in which case + ** the loop need to be marked as not order-distinct because it can + ** have repeated NULL rows. + ** + ** If the current term is a column of an ((?,?) IN (SELECT...)) + ** expression for which the SELECT returns more than one column, + ** check that it is the only column used by this loop. Otherwise, + ** if it is one of two or more, none of the columns can be + ** considered to match an ORDER BY term. + */ + if( (eOp & eqOpMask)!=0 ){ + if( eOp & (WO_ISNULL|WO_IS) ){ + testcase( eOp & WO_ISNULL ); + testcase( eOp & WO_IS ); + testcase( isOrderDistinct ); + isOrderDistinct = 0; + } + continue; + }else if( ALWAYS(eOp & WO_IN) ){ + /* ALWAYS() justification: eOp is an equality operator due to the + ** ju.btree.nEq constraint above. Any equality other + ** than WO_IN is captured by the previous "if". So this one + ** always has to be WO_IN. */ + Expr *pX = pLoop->aLTerm[j]->pExpr; + for(i=j+1; iu.btree.nEq; i++){ + if( pLoop->aLTerm[i]->pExpr==pX ){ + assert( (pLoop->aLTerm[i]->eOperator & WO_IN) ); + bOnce = 0; + break; + } + } + } + } + + /* Get the column number in the table (iColumn) and sort order + ** (revIdx) for the j-th column of the index. + */ + if( pIndex ){ + iColumn = pIndex->aiColumn[j]; + revIdx = pIndex->aSortOrder[j] & KEYINFO_ORDER_DESC; + if( iColumn==pIndex->pTable->iPKey ) iColumn = XN_ROWID; + }else{ + iColumn = XN_ROWID; + revIdx = 0; + } + + /* An unconstrained column that might be NULL means that this + ** WhereLoop is not well-ordered + */ + if( isOrderDistinct + && iColumn>=0 + && j>=pLoop->u.btree.nEq + && pIndex->pTable->aCol[iColumn].notNull==0 + ){ + isOrderDistinct = 0; + } + + /* Find the ORDER BY term that corresponds to the j-th column + ** of the index and mark that ORDER BY term off + */ + isMatch = 0; + for(i=0; bOnce && ia[i].pExpr); + testcase( wctrlFlags & WHERE_GROUPBY ); + testcase( wctrlFlags & WHERE_DISTINCTBY ); + if( (wctrlFlags & (WHERE_GROUPBY|WHERE_DISTINCTBY))==0 ) bOnce = 0; + if( iColumn>=XN_ROWID ){ + if( pOBExpr->op!=TK_COLUMN ) continue; + if( pOBExpr->iTable!=iCur ) continue; + if( pOBExpr->iColumn!=iColumn ) continue; + }else{ + Expr *pIdxExpr = pIndex->aColExpr->a[j].pExpr; + if( sqlite3ExprCompareSkip(pOBExpr, pIdxExpr, iCur) ){ + continue; + } + } + if( iColumn!=XN_ROWID ){ + pColl = sqlite3ExprNNCollSeq(pWInfo->pParse, pOrderBy->a[i].pExpr); + if( sqlite3StrICmp(pColl->zName, pIndex->azColl[j])!=0 ) continue; + } + if( wctrlFlags & WHERE_DISTINCTBY ){ + pLoop->u.btree.nDistinctCol = j+1; + } + isMatch = 1; + break; + } + if( isMatch && (wctrlFlags & WHERE_GROUPBY)==0 ){ + /* Make sure the sort order is compatible in an ORDER BY clause. + ** Sort order is irrelevant for a GROUP BY clause. */ + if( revSet ){ + if( (rev ^ revIdx)!=(pOrderBy->a[i].sortFlags&KEYINFO_ORDER_DESC) ){ + isMatch = 0; + } + }else{ + rev = revIdx ^ (pOrderBy->a[i].sortFlags & KEYINFO_ORDER_DESC); + if( rev ) *pRevMask |= MASKBIT(iLoop); + revSet = 1; + } + } + if( isMatch && (pOrderBy->a[i].sortFlags & KEYINFO_ORDER_BIGNULL) ){ + if( j==pLoop->u.btree.nEq ){ + pLoop->wsFlags |= WHERE_BIGNULL_SORT; + }else{ + isMatch = 0; + } + } + if( isMatch ){ + if( iColumn==XN_ROWID ){ + testcase( distinctColumns==0 ); + distinctColumns = 1; + } + obSat |= MASKBIT(i); + }else{ + /* No match found */ + if( j==0 || jmaskSelf; + for(i=0; ia[i].pExpr; + mTerm = sqlite3WhereExprUsage(&pWInfo->sMaskSet,p); + if( mTerm==0 && !sqlite3ExprIsConstant(p) ) continue; + if( (mTerm&~orderDistinctMask)==0 ){ + obSat |= MASKBIT(i); + } + } + } + } /* End the loop over all WhereLoops from outer-most down to inner-most */ + if( obSat==obDone ) return (i8)nOrderBy; + if( !isOrderDistinct ){ + for(i=nOrderBy-1; i>0; i--){ + Bitmask m = MASKBIT(i) - 1; + if( (obSat&m)==m ) return i; + } + return 0; + } + return -1; +} + + +/* +** If the WHERE_GROUPBY flag is set in the mask passed to sqlite3WhereBegin(), +** the planner assumes that the specified pOrderBy list is actually a GROUP +** BY clause - and so any order that groups rows as required satisfies the +** request. +** +** Normally, in this case it is not possible for the caller to determine +** whether or not the rows are really being delivered in sorted order, or +** just in some other order that provides the required grouping. However, +** if the WHERE_SORTBYGROUP flag is also passed to sqlite3WhereBegin(), then +** this function may be called on the returned WhereInfo object. It returns +** true if the rows really will be sorted in the specified order, or false +** otherwise. +** +** For example, assuming: +** +** CREATE INDEX i1 ON t1(x, Y); +** +** then +** +** SELECT * FROM t1 GROUP BY x,y ORDER BY x,y; -- IsSorted()==1 +** SELECT * FROM t1 GROUP BY y,x ORDER BY y,x; -- IsSorted()==0 +*/ +SQLITE_PRIVATE int sqlite3WhereIsSorted(WhereInfo *pWInfo){ + assert( pWInfo->wctrlFlags & WHERE_GROUPBY ); + assert( pWInfo->wctrlFlags & WHERE_SORTBYGROUP ); + return pWInfo->sorted; +} + +#ifdef WHERETRACE_ENABLED +/* For debugging use only: */ +static const char *wherePathName(WherePath *pPath, int nLoop, WhereLoop *pLast){ + static char zName[65]; + int i; + for(i=0; iaLoop[i]->cId; } + if( pLast ) zName[i++] = pLast->cId; + zName[i] = 0; + return zName; +} +#endif + +/* +** Return the cost of sorting nRow rows, assuming that the keys have +** nOrderby columns and that the first nSorted columns are already in +** order. +*/ +static LogEst whereSortingCost( + WhereInfo *pWInfo, + LogEst nRow, + int nOrderBy, + int nSorted +){ + /* TUNING: Estimated cost of a full external sort, where N is + ** the number of rows to sort is: + ** + ** cost = (3.0 * N * log(N)). + ** + ** Or, if the order-by clause has X terms but only the last Y + ** terms are out of order, then block-sorting will reduce the + ** sorting cost to: + ** + ** cost = (3.0 * N * log(N)) * (Y/X) + ** + ** The (Y/X) term is implemented using stack variable rScale + ** below. */ + LogEst rScale, rSortCost; + assert( nOrderBy>0 && 66==sqlite3LogEst(100) ); + rScale = sqlite3LogEst((nOrderBy-nSorted)*100/nOrderBy) - 66; + rSortCost = nRow + rScale + 16; + + /* Multiple by log(M) where M is the number of output rows. + ** Use the LIMIT for M if it is smaller */ + if( (pWInfo->wctrlFlags & WHERE_USE_LIMIT)!=0 && pWInfo->iLimitiLimit; + } + rSortCost += estLog(nRow); + return rSortCost; +} + +/* +** Given the list of WhereLoop objects at pWInfo->pLoops, this routine +** attempts to find the lowest cost path that visits each WhereLoop +** once. This path is then loaded into the pWInfo->a[].pWLoop fields. +** +** Assume that the total number of output rows that will need to be sorted +** will be nRowEst (in the 10*log2 representation). Or, ignore sorting +** costs if nRowEst==0. +** +** Return SQLITE_OK on success or SQLITE_NOMEM of a memory allocation +** error occurs. +*/ +static int wherePathSolver(WhereInfo *pWInfo, LogEst nRowEst){ + int mxChoice; /* Maximum number of simultaneous paths tracked */ + int nLoop; /* Number of terms in the join */ + Parse *pParse; /* Parsing context */ + sqlite3 *db; /* The database connection */ + int iLoop; /* Loop counter over the terms of the join */ + int ii, jj; /* Loop counters */ + int mxI = 0; /* Index of next entry to replace */ + int nOrderBy; /* Number of ORDER BY clause terms */ + LogEst mxCost = 0; /* Maximum cost of a set of paths */ + LogEst mxUnsorted = 0; /* Maximum unsorted cost of a set of path */ + int nTo, nFrom; /* Number of valid entries in aTo[] and aFrom[] */ + WherePath *aFrom; /* All nFrom paths at the previous level */ + WherePath *aTo; /* The nTo best paths at the current level */ + WherePath *pFrom; /* An element of aFrom[] that we are working on */ + WherePath *pTo; /* An element of aTo[] that we are working on */ + WhereLoop *pWLoop; /* One of the WhereLoop objects */ + WhereLoop **pX; /* Used to divy up the pSpace memory */ + LogEst *aSortCost = 0; /* Sorting and partial sorting costs */ + char *pSpace; /* Temporary memory used by this routine */ + int nSpace; /* Bytes of space allocated at pSpace */ + + pParse = pWInfo->pParse; + db = pParse->db; + nLoop = pWInfo->nLevel; + /* TUNING: For simple queries, only the best path is tracked. + ** For 2-way joins, the 5 best paths are followed. + ** For joins of 3 or more tables, track the 10 best paths */ + mxChoice = (nLoop<=1) ? 1 : (nLoop==2 ? 5 : 10); + assert( nLoop<=pWInfo->pTabList->nSrc ); + WHERETRACE(0x002, ("---- begin solver. (nRowEst=%d)\n", nRowEst)); + + /* If nRowEst is zero and there is an ORDER BY clause, ignore it. In this + ** case the purpose of this call is to estimate the number of rows returned + ** by the overall query. Once this estimate has been obtained, the caller + ** will invoke this function a second time, passing the estimate as the + ** nRowEst parameter. */ + if( pWInfo->pOrderBy==0 || nRowEst==0 ){ + nOrderBy = 0; + }else{ + nOrderBy = pWInfo->pOrderBy->nExpr; + } + + /* Allocate and initialize space for aTo, aFrom and aSortCost[] */ + nSpace = (sizeof(WherePath)+sizeof(WhereLoop*)*nLoop)*mxChoice*2; + nSpace += sizeof(LogEst) * nOrderBy; + pSpace = sqlite3DbMallocRawNN(db, nSpace); + if( pSpace==0 ) return SQLITE_NOMEM_BKPT; + aTo = (WherePath*)pSpace; + aFrom = aTo+mxChoice; + memset(aFrom, 0, sizeof(aFrom[0])); + pX = (WhereLoop**)(aFrom+mxChoice); + for(ii=mxChoice*2, pFrom=aTo; ii>0; ii--, pFrom++, pX += nLoop){ + pFrom->aLoop = pX; + } + if( nOrderBy ){ + /* If there is an ORDER BY clause and it is not being ignored, set up + ** space for the aSortCost[] array. Each element of the aSortCost array + ** is either zero - meaning it has not yet been initialized - or the + ** cost of sorting nRowEst rows of data where the first X terms of + ** the ORDER BY clause are already in order, where X is the array + ** index. */ + aSortCost = (LogEst*)pX; + memset(aSortCost, 0, sizeof(LogEst) * nOrderBy); + } + assert( aSortCost==0 || &pSpace[nSpace]==(char*)&aSortCost[nOrderBy] ); + assert( aSortCost!=0 || &pSpace[nSpace]==(char*)pX ); + + /* Seed the search with a single WherePath containing zero WhereLoops. + ** + ** TUNING: Do not let the number of iterations go above 28. If the cost + ** of computing an automatic index is not paid back within the first 28 + ** rows, then do not use the automatic index. */ + aFrom[0].nRow = MIN(pParse->nQueryLoop, 48); assert( 48==sqlite3LogEst(28) ); + nFrom = 1; + assert( aFrom[0].isOrdered==0 ); + if( nOrderBy ){ + /* If nLoop is zero, then there are no FROM terms in the query. Since + ** in this case the query may return a maximum of one row, the results + ** are already in the requested order. Set isOrdered to nOrderBy to + ** indicate this. Or, if nLoop is greater than zero, set isOrdered to + ** -1, indicating that the result set may or may not be ordered, + ** depending on the loops added to the current plan. */ + aFrom[0].isOrdered = nLoop>0 ? -1 : nOrderBy; + } + + /* Compute successively longer WherePaths using the previous generation + ** of WherePaths as the basis for the next. Keep track of the mxChoice + ** best paths at each generation */ + for(iLoop=0; iLooppLoops; pWLoop; pWLoop=pWLoop->pNextLoop){ + LogEst nOut; /* Rows visited by (pFrom+pWLoop) */ + LogEst rCost; /* Cost of path (pFrom+pWLoop) */ + LogEst rUnsorted; /* Unsorted cost of (pFrom+pWLoop) */ + i8 isOrdered = pFrom->isOrdered; /* isOrdered for (pFrom+pWLoop) */ + Bitmask maskNew; /* Mask of src visited by (..) */ + Bitmask revMask = 0; /* Mask of rev-order loops for (..) */ + + if( (pWLoop->prereq & ~pFrom->maskLoop)!=0 ) continue; + if( (pWLoop->maskSelf & pFrom->maskLoop)!=0 ) continue; + if( (pWLoop->wsFlags & WHERE_AUTO_INDEX)!=0 && pFrom->nRow<3 ){ + /* Do not use an automatic index if the this loop is expected + ** to run less than 1.25 times. It is tempting to also exclude + ** automatic index usage on an outer loop, but sometimes an automatic + ** index is useful in the outer loop of a correlated subquery. */ + assert( 10==sqlite3LogEst(2) ); + continue; + } + + /* At this point, pWLoop is a candidate to be the next loop. + ** Compute its cost */ + rUnsorted = sqlite3LogEstAdd(pWLoop->rSetup,pWLoop->rRun + pFrom->nRow); + rUnsorted = sqlite3LogEstAdd(rUnsorted, pFrom->rUnsorted); + nOut = pFrom->nRow + pWLoop->nOut; + maskNew = pFrom->maskLoop | pWLoop->maskSelf; + if( isOrdered<0 ){ + isOrdered = wherePathSatisfiesOrderBy(pWInfo, + pWInfo->pOrderBy, pFrom, pWInfo->wctrlFlags, + iLoop, pWLoop, &revMask); + }else{ + revMask = pFrom->revLoop; + } + if( isOrdered>=0 && isOrderedisOrdered^isOrdered)&0x80)==0" is equivalent + ** to (pTo->isOrdered==(-1))==(isOrdered==(-1))" for the range + ** of legal values for isOrdered, -1..64. + */ + for(jj=0, pTo=aTo; jjmaskLoop==maskNew + && ((pTo->isOrdered^isOrdered)&0x80)==0 + ){ + testcase( jj==nTo-1 ); + break; + } + } + if( jj>=nTo ){ + /* None of the existing best-so-far paths match the candidate. */ + if( nTo>=mxChoice + && (rCost>mxCost || (rCost==mxCost && rUnsorted>=mxUnsorted)) + ){ + /* The current candidate is no better than any of the mxChoice + ** paths currently in the best-so-far buffer. So discard + ** this candidate as not viable. */ +#ifdef WHERETRACE_ENABLED /* 0x4 */ + if( sqlite3WhereTrace&0x4 ){ + sqlite3DebugPrintf("Skip %s cost=%-3d,%3d,%3d order=%c\n", + wherePathName(pFrom, iLoop, pWLoop), rCost, nOut, rUnsorted, + isOrdered>=0 ? isOrdered+'0' : '?'); + } +#endif + continue; + } + /* If we reach this points it means that the new candidate path + ** needs to be added to the set of best-so-far paths. */ + if( nTo=0 ? isOrdered+'0' : '?'); + } +#endif + }else{ + /* Control reaches here if best-so-far path pTo=aTo[jj] covers the + ** same set of loops and has the same isOrdered setting as the + ** candidate path. Check to see if the candidate should replace + ** pTo or if the candidate should be skipped. + ** + ** The conditional is an expanded vector comparison equivalent to: + ** (pTo->rCost,pTo->nRow,pTo->rUnsorted) <= (rCost,nOut,rUnsorted) + */ + if( pTo->rCostrCost==rCost + && (pTo->nRownRow==nOut && pTo->rUnsorted<=rUnsorted) + ) + ) + ){ +#ifdef WHERETRACE_ENABLED /* 0x4 */ + if( sqlite3WhereTrace&0x4 ){ + sqlite3DebugPrintf( + "Skip %s cost=%-3d,%3d,%3d order=%c", + wherePathName(pFrom, iLoop, pWLoop), rCost, nOut, rUnsorted, + isOrdered>=0 ? isOrdered+'0' : '?'); + sqlite3DebugPrintf(" vs %s cost=%-3d,%3d,%3d order=%c\n", + wherePathName(pTo, iLoop+1, 0), pTo->rCost, pTo->nRow, + pTo->rUnsorted, pTo->isOrdered>=0 ? pTo->isOrdered+'0' : '?'); + } +#endif + /* Discard the candidate path from further consideration */ + testcase( pTo->rCost==rCost ); + continue; + } + testcase( pTo->rCost==rCost+1 ); + /* Control reaches here if the candidate path is better than the + ** pTo path. Replace pTo with the candidate. */ +#ifdef WHERETRACE_ENABLED /* 0x4 */ + if( sqlite3WhereTrace&0x4 ){ + sqlite3DebugPrintf( + "Update %s cost=%-3d,%3d,%3d order=%c", + wherePathName(pFrom, iLoop, pWLoop), rCost, nOut, rUnsorted, + isOrdered>=0 ? isOrdered+'0' : '?'); + sqlite3DebugPrintf(" was %s cost=%-3d,%3d,%3d order=%c\n", + wherePathName(pTo, iLoop+1, 0), pTo->rCost, pTo->nRow, + pTo->rUnsorted, pTo->isOrdered>=0 ? pTo->isOrdered+'0' : '?'); + } +#endif + } + /* pWLoop is a winner. Add it to the set of best so far */ + pTo->maskLoop = pFrom->maskLoop | pWLoop->maskSelf; + pTo->revLoop = revMask; + pTo->nRow = nOut; + pTo->rCost = rCost; + pTo->rUnsorted = rUnsorted; + pTo->isOrdered = isOrdered; + memcpy(pTo->aLoop, pFrom->aLoop, sizeof(WhereLoop*)*iLoop); + pTo->aLoop[iLoop] = pWLoop; + if( nTo>=mxChoice ){ + mxI = 0; + mxCost = aTo[0].rCost; + mxUnsorted = aTo[0].nRow; + for(jj=1, pTo=&aTo[1]; jjrCost>mxCost + || (pTo->rCost==mxCost && pTo->rUnsorted>mxUnsorted) + ){ + mxCost = pTo->rCost; + mxUnsorted = pTo->rUnsorted; + mxI = jj; + } + } + } + } + } + +#ifdef WHERETRACE_ENABLED /* >=2 */ + if( sqlite3WhereTrace & 0x02 ){ + sqlite3DebugPrintf("---- after round %d ----\n", iLoop); + for(ii=0, pTo=aTo; iirCost, pTo->nRow, + pTo->isOrdered>=0 ? (pTo->isOrdered+'0') : '?'); + if( pTo->isOrdered>0 ){ + sqlite3DebugPrintf(" rev=0x%llx\n", pTo->revLoop); + }else{ + sqlite3DebugPrintf("\n"); + } + } + } +#endif + + /* Swap the roles of aFrom and aTo for the next generation */ + pFrom = aTo; + aTo = aFrom; + aFrom = pFrom; + nFrom = nTo; + } + + if( nFrom==0 ){ + sqlite3ErrorMsg(pParse, "no query solution"); + sqlite3DbFreeNN(db, pSpace); + return SQLITE_ERROR; + } + + /* Find the lowest cost path. pFrom will be left pointing to that path */ + pFrom = aFrom; + for(ii=1; iirCost>aFrom[ii].rCost ) pFrom = &aFrom[ii]; + } + assert( pWInfo->nLevel==nLoop ); + /* Load the lowest cost path into pWInfo */ + for(iLoop=0; iLoopa + iLoop; + pLevel->pWLoop = pWLoop = pFrom->aLoop[iLoop]; + pLevel->iFrom = pWLoop->iTab; + pLevel->iTabCur = pWInfo->pTabList->a[pLevel->iFrom].iCursor; + } + if( (pWInfo->wctrlFlags & WHERE_WANT_DISTINCT)!=0 + && (pWInfo->wctrlFlags & WHERE_DISTINCTBY)==0 + && pWInfo->eDistinct==WHERE_DISTINCT_NOOP + && nRowEst + ){ + Bitmask notUsed; + int rc = wherePathSatisfiesOrderBy(pWInfo, pWInfo->pResultSet, pFrom, + WHERE_DISTINCTBY, nLoop-1, pFrom->aLoop[nLoop-1], ¬Used); + if( rc==pWInfo->pResultSet->nExpr ){ + pWInfo->eDistinct = WHERE_DISTINCT_ORDERED; + } + } + pWInfo->bOrderedInnerLoop = 0; + if( pWInfo->pOrderBy ){ + if( pWInfo->wctrlFlags & WHERE_DISTINCTBY ){ + if( pFrom->isOrdered==pWInfo->pOrderBy->nExpr ){ + pWInfo->eDistinct = WHERE_DISTINCT_ORDERED; + } + }else{ + pWInfo->nOBSat = pFrom->isOrdered; + pWInfo->revMask = pFrom->revLoop; + if( pWInfo->nOBSat<=0 ){ + pWInfo->nOBSat = 0; + if( nLoop>0 ){ + u32 wsFlags = pFrom->aLoop[nLoop-1]->wsFlags; + if( (wsFlags & WHERE_ONEROW)==0 + && (wsFlags&(WHERE_IPK|WHERE_COLUMN_IN))!=(WHERE_IPK|WHERE_COLUMN_IN) + ){ + Bitmask m = 0; + int rc = wherePathSatisfiesOrderBy(pWInfo, pWInfo->pOrderBy, pFrom, + WHERE_ORDERBY_LIMIT, nLoop-1, pFrom->aLoop[nLoop-1], &m); + testcase( wsFlags & WHERE_IPK ); + testcase( wsFlags & WHERE_COLUMN_IN ); + if( rc==pWInfo->pOrderBy->nExpr ){ + pWInfo->bOrderedInnerLoop = 1; + pWInfo->revMask = m; + } + } + } + } + } + if( (pWInfo->wctrlFlags & WHERE_SORTBYGROUP) + && pWInfo->nOBSat==pWInfo->pOrderBy->nExpr && nLoop>0 + ){ + Bitmask revMask = 0; + int nOrder = wherePathSatisfiesOrderBy(pWInfo, pWInfo->pOrderBy, + pFrom, 0, nLoop-1, pFrom->aLoop[nLoop-1], &revMask + ); + assert( pWInfo->sorted==0 ); + if( nOrder==pWInfo->pOrderBy->nExpr ){ + pWInfo->sorted = 1; + pWInfo->revMask = revMask; + } + } + } + + + pWInfo->nRowOut = pFrom->nRow; + + /* Free temporary memory and return success */ + sqlite3DbFreeNN(db, pSpace); + return SQLITE_OK; +} + +/* +** Most queries use only a single table (they are not joins) and have +** simple == constraints against indexed fields. This routine attempts +** to plan those simple cases using much less ceremony than the +** general-purpose query planner, and thereby yield faster sqlite3_prepare() +** times for the common case. +** +** Return non-zero on success, if this query can be handled by this +** no-frills query planner. Return zero if this query needs the +** general-purpose query planner. +*/ +static int whereShortCut(WhereLoopBuilder *pBuilder){ + WhereInfo *pWInfo; + struct SrcList_item *pItem; + WhereClause *pWC; + WhereTerm *pTerm; + WhereLoop *pLoop; + int iCur; + int j; + Table *pTab; + Index *pIdx; + + pWInfo = pBuilder->pWInfo; + if( pWInfo->wctrlFlags & WHERE_OR_SUBCLAUSE ) return 0; + assert( pWInfo->pTabList->nSrc>=1 ); + pItem = pWInfo->pTabList->a; + pTab = pItem->pTab; + if( IsVirtual(pTab) ) return 0; + if( pItem->fg.isIndexedBy ) return 0; + iCur = pItem->iCursor; + pWC = &pWInfo->sWC; + pLoop = pBuilder->pNew; + pLoop->wsFlags = 0; + pLoop->nSkip = 0; + pTerm = sqlite3WhereFindTerm(pWC, iCur, -1, 0, WO_EQ|WO_IS, 0); + if( pTerm ){ + testcase( pTerm->eOperator & WO_IS ); + pLoop->wsFlags = WHERE_COLUMN_EQ|WHERE_IPK|WHERE_ONEROW; + pLoop->aLTerm[0] = pTerm; + pLoop->nLTerm = 1; + pLoop->u.btree.nEq = 1; + /* TUNING: Cost of a rowid lookup is 10 */ + pLoop->rRun = 33; /* 33==sqlite3LogEst(10) */ + }else{ + for(pIdx=pTab->pIndex; pIdx; pIdx=pIdx->pNext){ + int opMask; + assert( pLoop->aLTermSpace==pLoop->aLTerm ); + if( !IsUniqueIndex(pIdx) + || pIdx->pPartIdxWhere!=0 + || pIdx->nKeyCol>ArraySize(pLoop->aLTermSpace) + ) continue; + opMask = pIdx->uniqNotNull ? (WO_EQ|WO_IS) : WO_EQ; + for(j=0; jnKeyCol; j++){ + pTerm = sqlite3WhereFindTerm(pWC, iCur, j, 0, opMask, pIdx); + if( pTerm==0 ) break; + testcase( pTerm->eOperator & WO_IS ); + pLoop->aLTerm[j] = pTerm; + } + if( j!=pIdx->nKeyCol ) continue; + pLoop->wsFlags = WHERE_COLUMN_EQ|WHERE_ONEROW|WHERE_INDEXED; + if( pIdx->isCovering || (pItem->colUsed & pIdx->colNotIdxed)==0 ){ + pLoop->wsFlags |= WHERE_IDX_ONLY; + } + pLoop->nLTerm = j; + pLoop->u.btree.nEq = j; + pLoop->u.btree.pIndex = pIdx; + /* TUNING: Cost of a unique index lookup is 15 */ + pLoop->rRun = 39; /* 39==sqlite3LogEst(15) */ + break; + } + } + if( pLoop->wsFlags ){ + pLoop->nOut = (LogEst)1; + pWInfo->a[0].pWLoop = pLoop; + assert( pWInfo->sMaskSet.n==1 && iCur==pWInfo->sMaskSet.ix[0] ); + pLoop->maskSelf = 1; /* sqlite3WhereGetMask(&pWInfo->sMaskSet, iCur); */ + pWInfo->a[0].iTabCur = iCur; + pWInfo->nRowOut = 1; + if( pWInfo->pOrderBy ) pWInfo->nOBSat = pWInfo->pOrderBy->nExpr; + if( pWInfo->wctrlFlags & WHERE_WANT_DISTINCT ){ + pWInfo->eDistinct = WHERE_DISTINCT_UNIQUE; + } +#ifdef SQLITE_DEBUG + pLoop->cId = '0'; +#endif + return 1; + } + return 0; +} + +/* +** Helper function for exprIsDeterministic(). +*/ +static int exprNodeIsDeterministic(Walker *pWalker, Expr *pExpr){ + if( pExpr->op==TK_FUNCTION && ExprHasProperty(pExpr, EP_ConstFunc)==0 ){ + pWalker->eCode = 0; + return WRC_Abort; + } + return WRC_Continue; +} + +/* +** Return true if the expression contains no non-deterministic SQL +** functions. Do not consider non-deterministic SQL functions that are +** part of sub-select statements. +*/ +static int exprIsDeterministic(Expr *p){ + Walker w; + memset(&w, 0, sizeof(w)); + w.eCode = 1; + w.xExprCallback = exprNodeIsDeterministic; + w.xSelectCallback = sqlite3SelectWalkFail; + sqlite3WalkExpr(&w, p); + return w.eCode; +} + +/* +** Generate the beginning of the loop used for WHERE clause processing. +** The return value is a pointer to an opaque structure that contains +** information needed to terminate the loop. Later, the calling routine +** should invoke sqlite3WhereEnd() with the return value of this function +** in order to complete the WHERE clause processing. +** +** If an error occurs, this routine returns NULL. +** +** The basic idea is to do a nested loop, one loop for each table in +** the FROM clause of a select. (INSERT and UPDATE statements are the +** same as a SELECT with only a single table in the FROM clause.) For +** example, if the SQL is this: +** +** SELECT * FROM t1, t2, t3 WHERE ...; +** +** Then the code generated is conceptually like the following: +** +** foreach row1 in t1 do \ Code generated +** foreach row2 in t2 do |-- by sqlite3WhereBegin() +** foreach row3 in t3 do / +** ... +** end \ Code generated +** end |-- by sqlite3WhereEnd() +** end / +** +** Note that the loops might not be nested in the order in which they +** appear in the FROM clause if a different order is better able to make +** use of indices. Note also that when the IN operator appears in +** the WHERE clause, it might result in additional nested loops for +** scanning through all values on the right-hand side of the IN. +** +** There are Btree cursors associated with each table. t1 uses cursor +** number pTabList->a[0].iCursor. t2 uses the cursor pTabList->a[1].iCursor. +** And so forth. This routine generates code to open those VDBE cursors +** and sqlite3WhereEnd() generates the code to close them. +** +** The code that sqlite3WhereBegin() generates leaves the cursors named +** in pTabList pointing at their appropriate entries. The [...] code +** can use OP_Column and OP_Rowid opcodes on these cursors to extract +** data from the various tables of the loop. +** +** If the WHERE clause is empty, the foreach loops must each scan their +** entire tables. Thus a three-way join is an O(N^3) operation. But if +** the tables have indices and there are terms in the WHERE clause that +** refer to those indices, a complete table scan can be avoided and the +** code will run much faster. Most of the work of this routine is checking +** to see if there are indices that can be used to speed up the loop. +** +** Terms of the WHERE clause are also used to limit which rows actually +** make it to the "..." in the middle of the loop. After each "foreach", +** terms of the WHERE clause that use only terms in that loop and outer +** loops are evaluated and if false a jump is made around all subsequent +** inner loops (or around the "..." if the test occurs within the inner- +** most loop) +** +** OUTER JOINS +** +** An outer join of tables t1 and t2 is conceptally coded as follows: +** +** foreach row1 in t1 do +** flag = 0 +** foreach row2 in t2 do +** start: +** ... +** flag = 1 +** end +** if flag==0 then +** move the row2 cursor to a null row +** goto start +** fi +** end +** +** ORDER BY CLAUSE PROCESSING +** +** pOrderBy is a pointer to the ORDER BY clause (or the GROUP BY clause +** if the WHERE_GROUPBY flag is set in wctrlFlags) of a SELECT statement +** if there is one. If there is no ORDER BY clause or if this routine +** is called from an UPDATE or DELETE statement, then pOrderBy is NULL. +** +** The iIdxCur parameter is the cursor number of an index. If +** WHERE_OR_SUBCLAUSE is set, iIdxCur is the cursor number of an index +** to use for OR clause processing. The WHERE clause should use this +** specific cursor. If WHERE_ONEPASS_DESIRED is set, then iIdxCur is +** the first cursor in an array of cursors for all indices. iIdxCur should +** be used to compute the appropriate cursor depending on which index is +** used. +*/ +SQLITE_PRIVATE WhereInfo *sqlite3WhereBegin( + Parse *pParse, /* The parser context */ + SrcList *pTabList, /* FROM clause: A list of all tables to be scanned */ + Expr *pWhere, /* The WHERE clause */ + ExprList *pOrderBy, /* An ORDER BY (or GROUP BY) clause, or NULL */ + ExprList *pResultSet, /* Query result set. Req'd for DISTINCT */ + u16 wctrlFlags, /* The WHERE_* flags defined in sqliteInt.h */ + int iAuxArg /* If WHERE_OR_SUBCLAUSE is set, index cursor number + ** If WHERE_USE_LIMIT, then the limit amount */ +){ + int nByteWInfo; /* Num. bytes allocated for WhereInfo struct */ + int nTabList; /* Number of elements in pTabList */ + WhereInfo *pWInfo; /* Will become the return value of this function */ + Vdbe *v = pParse->pVdbe; /* The virtual database engine */ + Bitmask notReady; /* Cursors that are not yet positioned */ + WhereLoopBuilder sWLB; /* The WhereLoop builder */ + WhereMaskSet *pMaskSet; /* The expression mask set */ + WhereLevel *pLevel; /* A single level in pWInfo->a[] */ + WhereLoop *pLoop; /* Pointer to a single WhereLoop object */ + int ii; /* Loop counter */ + sqlite3 *db; /* Database connection */ + int rc; /* Return code */ + u8 bFordelete = 0; /* OPFLAG_FORDELETE or zero, as appropriate */ + + assert( (wctrlFlags & WHERE_ONEPASS_MULTIROW)==0 || ( + (wctrlFlags & WHERE_ONEPASS_DESIRED)!=0 + && (wctrlFlags & WHERE_OR_SUBCLAUSE)==0 + )); + + /* Only one of WHERE_OR_SUBCLAUSE or WHERE_USE_LIMIT */ + assert( (wctrlFlags & WHERE_OR_SUBCLAUSE)==0 + || (wctrlFlags & WHERE_USE_LIMIT)==0 ); + + /* Variable initialization */ + db = pParse->db; + memset(&sWLB, 0, sizeof(sWLB)); + + /* An ORDER/GROUP BY clause of more than 63 terms cannot be optimized */ + testcase( pOrderBy && pOrderBy->nExpr==BMS-1 ); + if( pOrderBy && pOrderBy->nExpr>=BMS ) pOrderBy = 0; + sWLB.pOrderBy = pOrderBy; + + /* Disable the DISTINCT optimization if SQLITE_DistinctOpt is set via + ** sqlite3_test_ctrl(SQLITE_TESTCTRL_OPTIMIZATIONS,...) */ + if( OptimizationDisabled(db, SQLITE_DistinctOpt) ){ + wctrlFlags &= ~WHERE_WANT_DISTINCT; + } + + /* The number of tables in the FROM clause is limited by the number of + ** bits in a Bitmask + */ + testcase( pTabList->nSrc==BMS ); + if( pTabList->nSrc>BMS ){ + sqlite3ErrorMsg(pParse, "at most %d tables in a join", BMS); + return 0; + } + + /* This function normally generates a nested loop for all tables in + ** pTabList. But if the WHERE_OR_SUBCLAUSE flag is set, then we should + ** only generate code for the first table in pTabList and assume that + ** any cursors associated with subsequent tables are uninitialized. + */ + nTabList = (wctrlFlags & WHERE_OR_SUBCLAUSE) ? 1 : pTabList->nSrc; + + /* Allocate and initialize the WhereInfo structure that will become the + ** return value. A single allocation is used to store the WhereInfo + ** struct, the contents of WhereInfo.a[], the WhereClause structure + ** and the WhereMaskSet structure. Since WhereClause contains an 8-byte + ** field (type Bitmask) it must be aligned on an 8-byte boundary on + ** some architectures. Hence the ROUND8() below. + */ + nByteWInfo = ROUND8(sizeof(WhereInfo)+(nTabList-1)*sizeof(WhereLevel)); + pWInfo = sqlite3DbMallocRawNN(db, nByteWInfo + sizeof(WhereLoop)); + if( db->mallocFailed ){ + sqlite3DbFree(db, pWInfo); + pWInfo = 0; + goto whereBeginError; + } + pWInfo->pParse = pParse; + pWInfo->pTabList = pTabList; + pWInfo->pOrderBy = pOrderBy; + pWInfo->pWhere = pWhere; + pWInfo->pResultSet = pResultSet; + pWInfo->aiCurOnePass[0] = pWInfo->aiCurOnePass[1] = -1; + pWInfo->nLevel = nTabList; + pWInfo->iBreak = pWInfo->iContinue = sqlite3VdbeMakeLabel(pParse); + pWInfo->wctrlFlags = wctrlFlags; + pWInfo->iLimit = iAuxArg; + pWInfo->savedNQueryLoop = pParse->nQueryLoop; + memset(&pWInfo->nOBSat, 0, + offsetof(WhereInfo,sWC) - offsetof(WhereInfo,nOBSat)); + memset(&pWInfo->a[0], 0, sizeof(WhereLoop)+nTabList*sizeof(WhereLevel)); + assert( pWInfo->eOnePass==ONEPASS_OFF ); /* ONEPASS defaults to OFF */ + pMaskSet = &pWInfo->sMaskSet; + sWLB.pWInfo = pWInfo; + sWLB.pWC = &pWInfo->sWC; + sWLB.pNew = (WhereLoop*)(((char*)pWInfo)+nByteWInfo); + assert( EIGHT_BYTE_ALIGNMENT(sWLB.pNew) ); + whereLoopInit(sWLB.pNew); +#ifdef SQLITE_DEBUG + sWLB.pNew->cId = '*'; +#endif + + /* Split the WHERE clause into separate subexpressions where each + ** subexpression is separated by an AND operator. + */ + initMaskSet(pMaskSet); + sqlite3WhereClauseInit(&pWInfo->sWC, pWInfo); + sqlite3WhereSplit(&pWInfo->sWC, pWhere, TK_AND); + + /* Special case: No FROM clause + */ + if( nTabList==0 ){ + if( pOrderBy ) pWInfo->nOBSat = pOrderBy->nExpr; + if( wctrlFlags & WHERE_WANT_DISTINCT ){ + pWInfo->eDistinct = WHERE_DISTINCT_UNIQUE; + } + ExplainQueryPlan((pParse, 0, "SCAN CONSTANT ROW")); + }else{ + /* Assign a bit from the bitmask to every term in the FROM clause. + ** + ** The N-th term of the FROM clause is assigned a bitmask of 1<nSrc tables in + ** pTabList, not just the first nTabList tables. nTabList is normally + ** equal to pTabList->nSrc but might be shortened to 1 if the + ** WHERE_OR_SUBCLAUSE flag is set. + */ + ii = 0; + do{ + createMask(pMaskSet, pTabList->a[ii].iCursor); + sqlite3WhereTabFuncArgs(pParse, &pTabList->a[ii], &pWInfo->sWC); + }while( (++ii)nSrc ); + #ifdef SQLITE_DEBUG + { + Bitmask mx = 0; + for(ii=0; iinSrc; ii++){ + Bitmask m = sqlite3WhereGetMask(pMaskSet, pTabList->a[ii].iCursor); + assert( m>=mx ); + mx = m; + } + } + #endif + } + + /* Analyze all of the subexpressions. */ + sqlite3WhereExprAnalyze(pTabList, &pWInfo->sWC); + if( db->mallocFailed ) goto whereBeginError; + + /* Special case: WHERE terms that do not refer to any tables in the join + ** (constant expressions). Evaluate each such term, and jump over all the + ** generated code if the result is not true. + ** + ** Do not do this if the expression contains non-deterministic functions + ** that are not within a sub-select. This is not strictly required, but + ** preserves SQLite's legacy behaviour in the following two cases: + ** + ** FROM ... WHERE random()>0; -- eval random() once per row + ** FROM ... WHERE (SELECT random())>0; -- eval random() once overall + */ + for(ii=0; iinTerm; ii++){ + WhereTerm *pT = &sWLB.pWC->a[ii]; + if( pT->wtFlags & TERM_VIRTUAL ) continue; + if( pT->prereqAll==0 && (nTabList==0 || exprIsDeterministic(pT->pExpr)) ){ + sqlite3ExprIfFalse(pParse, pT->pExpr, pWInfo->iBreak, SQLITE_JUMPIFNULL); + pT->wtFlags |= TERM_CODED; + } + } + + if( wctrlFlags & WHERE_WANT_DISTINCT ){ + if( isDistinctRedundant(pParse, pTabList, &pWInfo->sWC, pResultSet) ){ + /* The DISTINCT marking is pointless. Ignore it. */ + pWInfo->eDistinct = WHERE_DISTINCT_UNIQUE; + }else if( pOrderBy==0 ){ + /* Try to ORDER BY the result set to make distinct processing easier */ + pWInfo->wctrlFlags |= WHERE_DISTINCTBY; + pWInfo->pOrderBy = pResultSet; + } + } + + /* Construct the WhereLoop objects */ +#if defined(WHERETRACE_ENABLED) + if( sqlite3WhereTrace & 0xffff ){ + sqlite3DebugPrintf("*** Optimizer Start *** (wctrlFlags: 0x%x",wctrlFlags); + if( wctrlFlags & WHERE_USE_LIMIT ){ + sqlite3DebugPrintf(", limit: %d", iAuxArg); + } + sqlite3DebugPrintf(")\n"); + if( sqlite3WhereTrace & 0x100 ){ + Select sSelect; + memset(&sSelect, 0, sizeof(sSelect)); + sSelect.selFlags = SF_WhereBegin; + sSelect.pSrc = pTabList; + sSelect.pWhere = pWhere; + sSelect.pOrderBy = pOrderBy; + sSelect.pEList = pResultSet; + sqlite3TreeViewSelect(0, &sSelect, 0); + } + } + if( sqlite3WhereTrace & 0x100 ){ /* Display all terms of the WHERE clause */ + sqlite3WhereClausePrint(sWLB.pWC); + } +#endif + + if( nTabList!=1 || whereShortCut(&sWLB)==0 ){ + rc = whereLoopAddAll(&sWLB); + if( rc ) goto whereBeginError; + +#ifdef WHERETRACE_ENABLED + if( sqlite3WhereTrace ){ /* Display all of the WhereLoop objects */ + WhereLoop *p; + int i; + static const char zLabel[] = "0123456789abcdefghijklmnopqrstuvwyxz" + "ABCDEFGHIJKLMNOPQRSTUVWYXZ"; + for(p=pWInfo->pLoops, i=0; p; p=p->pNextLoop, i++){ + p->cId = zLabel[i%(sizeof(zLabel)-1)]; + whereLoopPrint(p, sWLB.pWC); + } + } +#endif + + wherePathSolver(pWInfo, 0); + if( db->mallocFailed ) goto whereBeginError; + if( pWInfo->pOrderBy ){ + wherePathSolver(pWInfo, pWInfo->nRowOut+1); + if( db->mallocFailed ) goto whereBeginError; + } + } + if( pWInfo->pOrderBy==0 && (db->flags & SQLITE_ReverseOrder)!=0 ){ + pWInfo->revMask = ALLBITS; + } + if( pParse->nErr || NEVER(db->mallocFailed) ){ + goto whereBeginError; + } +#ifdef WHERETRACE_ENABLED + if( sqlite3WhereTrace ){ + sqlite3DebugPrintf("---- Solution nRow=%d", pWInfo->nRowOut); + if( pWInfo->nOBSat>0 ){ + sqlite3DebugPrintf(" ORDERBY=%d,0x%llx", pWInfo->nOBSat, pWInfo->revMask); + } + switch( pWInfo->eDistinct ){ + case WHERE_DISTINCT_UNIQUE: { + sqlite3DebugPrintf(" DISTINCT=unique"); + break; + } + case WHERE_DISTINCT_ORDERED: { + sqlite3DebugPrintf(" DISTINCT=ordered"); + break; + } + case WHERE_DISTINCT_UNORDERED: { + sqlite3DebugPrintf(" DISTINCT=unordered"); + break; + } + } + sqlite3DebugPrintf("\n"); + for(ii=0; iinLevel; ii++){ + whereLoopPrint(pWInfo->a[ii].pWLoop, sWLB.pWC); + } + } +#endif + + /* Attempt to omit tables from the join that do not affect the result. + ** For a table to not affect the result, the following must be true: + ** + ** 1) The query must not be an aggregate. + ** 2) The table must be the RHS of a LEFT JOIN. + ** 3) Either the query must be DISTINCT, or else the ON or USING clause + ** must contain a constraint that limits the scan of the table to + ** at most a single row. + ** 4) The table must not be referenced by any part of the query apart + ** from its own USING or ON clause. + ** + ** For example, given: + ** + ** CREATE TABLE t1(ipk INTEGER PRIMARY KEY, v1); + ** CREATE TABLE t2(ipk INTEGER PRIMARY KEY, v2); + ** CREATE TABLE t3(ipk INTEGER PRIMARY KEY, v3); + ** + ** then table t2 can be omitted from the following: + ** + ** SELECT v1, v3 FROM t1 + ** LEFT JOIN t2 USING (t1.ipk=t2.ipk) + ** LEFT JOIN t3 USING (t1.ipk=t3.ipk) + ** + ** or from: + ** + ** SELECT DISTINCT v1, v3 FROM t1 + ** LEFT JOIN t2 + ** LEFT JOIN t3 USING (t1.ipk=t3.ipk) + */ + notReady = ~(Bitmask)0; + if( pWInfo->nLevel>=2 + && pResultSet!=0 /* guarantees condition (1) above */ + && OptimizationEnabled(db, SQLITE_OmitNoopJoin) + ){ + int i; + Bitmask tabUsed = sqlite3WhereExprListUsage(pMaskSet, pResultSet); + if( sWLB.pOrderBy ){ + tabUsed |= sqlite3WhereExprListUsage(pMaskSet, sWLB.pOrderBy); + } + for(i=pWInfo->nLevel-1; i>=1; i--){ + WhereTerm *pTerm, *pEnd; + struct SrcList_item *pItem; + pLoop = pWInfo->a[i].pWLoop; + pItem = &pWInfo->pTabList->a[pLoop->iTab]; + if( (pItem->fg.jointype & JT_LEFT)==0 ) continue; + if( (wctrlFlags & WHERE_WANT_DISTINCT)==0 + && (pLoop->wsFlags & WHERE_ONEROW)==0 + ){ + continue; + } + if( (tabUsed & pLoop->maskSelf)!=0 ) continue; + pEnd = sWLB.pWC->a + sWLB.pWC->nTerm; + for(pTerm=sWLB.pWC->a; pTermprereqAll & pLoop->maskSelf)!=0 ){ + if( !ExprHasProperty(pTerm->pExpr, EP_FromJoin) + || pTerm->pExpr->iRightJoinTable!=pItem->iCursor + ){ + break; + } + } + } + if( pTerm drop loop %c not used\n", pLoop->cId)); + notReady &= ~pLoop->maskSelf; + for(pTerm=sWLB.pWC->a; pTermprereqAll & pLoop->maskSelf)!=0 ){ + pTerm->wtFlags |= TERM_CODED; + } + } + if( i!=pWInfo->nLevel-1 ){ + int nByte = (pWInfo->nLevel-1-i) * sizeof(WhereLevel); + memmove(&pWInfo->a[i], &pWInfo->a[i+1], nByte); + } + pWInfo->nLevel--; + nTabList--; + } + } + WHERETRACE(0xffff,("*** Optimizer Finished ***\n")); + pWInfo->pParse->nQueryLoop += pWInfo->nRowOut; + + /* If the caller is an UPDATE or DELETE statement that is requesting + ** to use a one-pass algorithm, determine if this is appropriate. + ** + ** A one-pass approach can be used if the caller has requested one + ** and either (a) the scan visits at most one row or (b) each + ** of the following are true: + ** + ** * the caller has indicated that a one-pass approach can be used + ** with multiple rows (by setting WHERE_ONEPASS_MULTIROW), and + ** * the table is not a virtual table, and + ** * either the scan does not use the OR optimization or the caller + ** is a DELETE operation (WHERE_DUPLICATES_OK is only specified + ** for DELETE). + ** + ** The last qualification is because an UPDATE statement uses + ** WhereInfo.aiCurOnePass[1] to determine whether or not it really can + ** use a one-pass approach, and this is not set accurately for scans + ** that use the OR optimization. + */ + assert( (wctrlFlags & WHERE_ONEPASS_DESIRED)==0 || pWInfo->nLevel==1 ); + if( (wctrlFlags & WHERE_ONEPASS_DESIRED)!=0 ){ + int wsFlags = pWInfo->a[0].pWLoop->wsFlags; + int bOnerow = (wsFlags & WHERE_ONEROW)!=0; + assert( !(wsFlags & WHERE_VIRTUALTABLE) || IsVirtual(pTabList->a[0].pTab) ); + if( bOnerow || ( + 0!=(wctrlFlags & WHERE_ONEPASS_MULTIROW) + && !IsVirtual(pTabList->a[0].pTab) + && (0==(wsFlags & WHERE_MULTI_OR) || (wctrlFlags & WHERE_DUPLICATES_OK)) + )){ + pWInfo->eOnePass = bOnerow ? ONEPASS_SINGLE : ONEPASS_MULTI; + if( HasRowid(pTabList->a[0].pTab) && (wsFlags & WHERE_IDX_ONLY) ){ + if( wctrlFlags & WHERE_ONEPASS_MULTIROW ){ + bFordelete = OPFLAG_FORDELETE; + } + pWInfo->a[0].pWLoop->wsFlags = (wsFlags & ~WHERE_IDX_ONLY); + } + } + } + + /* Open all tables in the pTabList and any indices selected for + ** searching those tables. + */ + for(ii=0, pLevel=pWInfo->a; iia[pLevel->iFrom]; + pTab = pTabItem->pTab; + iDb = sqlite3SchemaToIndex(db, pTab->pSchema); + pLoop = pLevel->pWLoop; + if( (pTab->tabFlags & TF_Ephemeral)!=0 || pTab->pSelect ){ + /* Do nothing */ + }else +#ifndef SQLITE_OMIT_VIRTUALTABLE + if( (pLoop->wsFlags & WHERE_VIRTUALTABLE)!=0 ){ + const char *pVTab = (const char *)sqlite3GetVTable(db, pTab); + int iCur = pTabItem->iCursor; + sqlite3VdbeAddOp4(v, OP_VOpen, iCur, 0, 0, pVTab, P4_VTAB); + }else if( IsVirtual(pTab) ){ + /* noop */ + }else +#endif + if( (pLoop->wsFlags & WHERE_IDX_ONLY)==0 + && (wctrlFlags & WHERE_OR_SUBCLAUSE)==0 ){ + int op = OP_OpenRead; + if( pWInfo->eOnePass!=ONEPASS_OFF ){ + op = OP_OpenWrite; + pWInfo->aiCurOnePass[0] = pTabItem->iCursor; + }; + sqlite3OpenTable(pParse, pTabItem->iCursor, iDb, pTab, op); + assert( pTabItem->iCursor==pLevel->iTabCur ); + testcase( pWInfo->eOnePass==ONEPASS_OFF && pTab->nCol==BMS-1 ); + testcase( pWInfo->eOnePass==ONEPASS_OFF && pTab->nCol==BMS ); + if( pWInfo->eOnePass==ONEPASS_OFF && pTab->nColcolUsed; + int n = 0; + for(; b; b=b>>1, n++){} + sqlite3VdbeChangeP4(v, -1, SQLITE_INT_TO_PTR(n), P4_INT32); + assert( n<=pTab->nCol ); + } +#ifdef SQLITE_ENABLE_CURSOR_HINTS + if( pLoop->u.btree.pIndex!=0 ){ + sqlite3VdbeChangeP5(v, OPFLAG_SEEKEQ|bFordelete); + }else +#endif + { + sqlite3VdbeChangeP5(v, bFordelete); + } +#ifdef SQLITE_ENABLE_COLUMN_USED_MASK + sqlite3VdbeAddOp4Dup8(v, OP_ColumnsUsed, pTabItem->iCursor, 0, 0, + (const u8*)&pTabItem->colUsed, P4_INT64); +#endif + }else{ + sqlite3TableLock(pParse, iDb, pTab->tnum, 0, pTab->zName); + } + if( pLoop->wsFlags & WHERE_INDEXED ){ + Index *pIx = pLoop->u.btree.pIndex; + int iIndexCur; + int op = OP_OpenRead; + /* iAuxArg is always set to a positive value if ONEPASS is possible */ + assert( iAuxArg!=0 || (pWInfo->wctrlFlags & WHERE_ONEPASS_DESIRED)==0 ); + if( !HasRowid(pTab) && IsPrimaryKeyIndex(pIx) + && (wctrlFlags & WHERE_OR_SUBCLAUSE)!=0 + ){ + /* This is one term of an OR-optimization using the PRIMARY KEY of a + ** WITHOUT ROWID table. No need for a separate index */ + iIndexCur = pLevel->iTabCur; + op = 0; + }else if( pWInfo->eOnePass!=ONEPASS_OFF ){ + Index *pJ = pTabItem->pTab->pIndex; + iIndexCur = iAuxArg; + assert( wctrlFlags & WHERE_ONEPASS_DESIRED ); + while( ALWAYS(pJ) && pJ!=pIx ){ + iIndexCur++; + pJ = pJ->pNext; + } + op = OP_OpenWrite; + pWInfo->aiCurOnePass[1] = iIndexCur; + }else if( iAuxArg && (wctrlFlags & WHERE_OR_SUBCLAUSE)!=0 ){ + iIndexCur = iAuxArg; + op = OP_ReopenIdx; + }else{ + iIndexCur = pParse->nTab++; + } + pLevel->iIdxCur = iIndexCur; + assert( pIx->pSchema==pTab->pSchema ); + assert( iIndexCur>=0 ); + if( op ){ + sqlite3VdbeAddOp3(v, op, iIndexCur, pIx->tnum, iDb); + sqlite3VdbeSetP4KeyInfo(pParse, pIx); + if( (pLoop->wsFlags & WHERE_CONSTRAINT)!=0 + && (pLoop->wsFlags & (WHERE_COLUMN_RANGE|WHERE_SKIPSCAN))==0 + && (pLoop->wsFlags & WHERE_BIGNULL_SORT)==0 + && (pWInfo->wctrlFlags&WHERE_ORDERBY_MIN)==0 + && pWInfo->eDistinct!=WHERE_DISTINCT_ORDERED + ){ + sqlite3VdbeChangeP5(v, OPFLAG_SEEKEQ); /* Hint to COMDB2 */ + } + VdbeComment((v, "%s", pIx->zName)); +#ifdef SQLITE_ENABLE_COLUMN_USED_MASK + { + u64 colUsed = 0; + int ii, jj; + for(ii=0; iinColumn; ii++){ + jj = pIx->aiColumn[ii]; + if( jj<0 ) continue; + if( jj>63 ) jj = 63; + if( (pTabItem->colUsed & MASKBIT(jj))==0 ) continue; + colUsed |= ((u64)1)<<(ii<63 ? ii : 63); + } + sqlite3VdbeAddOp4Dup8(v, OP_ColumnsUsed, iIndexCur, 0, 0, + (u8*)&colUsed, P4_INT64); + } +#endif /* SQLITE_ENABLE_COLUMN_USED_MASK */ + } + } + if( iDb>=0 ) sqlite3CodeVerifySchema(pParse, iDb); + } + pWInfo->iTop = sqlite3VdbeCurrentAddr(v); + if( db->mallocFailed ) goto whereBeginError; + + /* Generate the code to do the search. Each iteration of the for + ** loop below generates code for a single nested loop of the VM + ** program. + */ + for(ii=0; iia[ii]; + wsFlags = pLevel->pWLoop->wsFlags; +#ifndef SQLITE_OMIT_AUTOMATIC_INDEX + if( (pLevel->pWLoop->wsFlags & WHERE_AUTO_INDEX)!=0 ){ + constructAutomaticIndex(pParse, &pWInfo->sWC, + &pTabList->a[pLevel->iFrom], notReady, pLevel); + if( db->mallocFailed ) goto whereBeginError; + } +#endif + addrExplain = sqlite3WhereExplainOneScan( + pParse, pTabList, pLevel, wctrlFlags + ); + pLevel->addrBody = sqlite3VdbeCurrentAddr(v); + notReady = sqlite3WhereCodeOneLoopStart(pParse,v,pWInfo,ii,pLevel,notReady); + pWInfo->iContinue = pLevel->addrCont; + if( (wsFlags&WHERE_MULTI_OR)==0 && (wctrlFlags&WHERE_OR_SUBCLAUSE)==0 ){ + sqlite3WhereAddScanStatus(v, pTabList, pLevel, addrExplain); + } + } + + /* Done. */ + VdbeModuleComment((v, "Begin WHERE-core")); + return pWInfo; + + /* Jump here if malloc fails */ +whereBeginError: + if( pWInfo ){ + pParse->nQueryLoop = pWInfo->savedNQueryLoop; + whereInfoFree(db, pWInfo); + } + return 0; +} + +/* +** Part of sqlite3WhereEnd() will rewrite opcodes to reference the +** index rather than the main table. In SQLITE_DEBUG mode, we want +** to trace those changes if PRAGMA vdbe_addoptrace=on. This routine +** does that. +*/ +#ifndef SQLITE_DEBUG +# define OpcodeRewriteTrace(D,K,P) /* no-op */ +#else +# define OpcodeRewriteTrace(D,K,P) sqlite3WhereOpcodeRewriteTrace(D,K,P) + static void sqlite3WhereOpcodeRewriteTrace( + sqlite3 *db, + int pc, + VdbeOp *pOp + ){ + if( (db->flags & SQLITE_VdbeAddopTrace)==0 ) return; + sqlite3VdbePrintOp(0, pc, pOp); + } +#endif + +/* +** Generate the end of the WHERE loop. See comments on +** sqlite3WhereBegin() for additional information. +*/ +SQLITE_PRIVATE void sqlite3WhereEnd(WhereInfo *pWInfo){ + Parse *pParse = pWInfo->pParse; + Vdbe *v = pParse->pVdbe; + int i; + WhereLevel *pLevel; + WhereLoop *pLoop; + SrcList *pTabList = pWInfo->pTabList; + sqlite3 *db = pParse->db; + + /* Generate loop termination code. + */ + VdbeModuleComment((v, "End WHERE-core")); + for(i=pWInfo->nLevel-1; i>=0; i--){ + int addr; + pLevel = &pWInfo->a[i]; + pLoop = pLevel->pWLoop; + if( pLevel->op!=OP_Noop ){ +#ifndef SQLITE_DISABLE_SKIPAHEAD_DISTINCT + int addrSeek = 0; + Index *pIdx; + int n; + if( pWInfo->eDistinct==WHERE_DISTINCT_ORDERED + && i==pWInfo->nLevel-1 /* Ticket [ef9318757b152e3] 2017-10-21 */ + && (pLoop->wsFlags & WHERE_INDEXED)!=0 + && (pIdx = pLoop->u.btree.pIndex)->hasStat1 + && (n = pLoop->u.btree.nDistinctCol)>0 + && pIdx->aiRowLogEst[n]>=36 + ){ + int r1 = pParse->nMem+1; + int j, op; + for(j=0; jiIdxCur, j, r1+j); + } + pParse->nMem += n+1; + op = pLevel->op==OP_Prev ? OP_SeekLT : OP_SeekGT; + addrSeek = sqlite3VdbeAddOp4Int(v, op, pLevel->iIdxCur, 0, r1, n); + VdbeCoverageIf(v, op==OP_SeekLT); + VdbeCoverageIf(v, op==OP_SeekGT); + sqlite3VdbeAddOp2(v, OP_Goto, 1, pLevel->p2); + } +#endif /* SQLITE_DISABLE_SKIPAHEAD_DISTINCT */ + /* The common case: Advance to the next row */ + sqlite3VdbeResolveLabel(v, pLevel->addrCont); + sqlite3VdbeAddOp3(v, pLevel->op, pLevel->p1, pLevel->p2, pLevel->p3); + sqlite3VdbeChangeP5(v, pLevel->p5); + VdbeCoverage(v); + VdbeCoverageIf(v, pLevel->op==OP_Next); + VdbeCoverageIf(v, pLevel->op==OP_Prev); + VdbeCoverageIf(v, pLevel->op==OP_VNext); + if( pLevel->regBignull ){ + sqlite3VdbeResolveLabel(v, pLevel->addrBignull); + sqlite3VdbeAddOp2(v, OP_DecrJumpZero, pLevel->regBignull, pLevel->p2-1); + VdbeCoverage(v); + } +#ifndef SQLITE_DISABLE_SKIPAHEAD_DISTINCT + if( addrSeek ) sqlite3VdbeJumpHere(v, addrSeek); +#endif + }else{ + sqlite3VdbeResolveLabel(v, pLevel->addrCont); + } + if( pLoop->wsFlags & WHERE_IN_ABLE && pLevel->u.in.nIn>0 ){ + struct InLoop *pIn; + int j; + sqlite3VdbeResolveLabel(v, pLevel->addrNxt); + for(j=pLevel->u.in.nIn, pIn=&pLevel->u.in.aInLoop[j-1]; j>0; j--, pIn--){ + sqlite3VdbeJumpHere(v, pIn->addrInTop+1); + if( pIn->eEndLoopOp!=OP_Noop ){ + if( pIn->nPrefix ){ + assert( pLoop->wsFlags & WHERE_IN_EARLYOUT ); + sqlite3VdbeAddOp4Int(v, OP_IfNoHope, pLevel->iIdxCur, + sqlite3VdbeCurrentAddr(v)+2, + pIn->iBase, pIn->nPrefix); + VdbeCoverage(v); + } + sqlite3VdbeAddOp2(v, pIn->eEndLoopOp, pIn->iCur, pIn->addrInTop); + VdbeCoverage(v); + VdbeCoverageIf(v, pIn->eEndLoopOp==OP_Prev); + VdbeCoverageIf(v, pIn->eEndLoopOp==OP_Next); + } + sqlite3VdbeJumpHere(v, pIn->addrInTop-1); + } + } + sqlite3VdbeResolveLabel(v, pLevel->addrBrk); + if( pLevel->addrSkip ){ + sqlite3VdbeGoto(v, pLevel->addrSkip); + VdbeComment((v, "next skip-scan on %s", pLoop->u.btree.pIndex->zName)); + sqlite3VdbeJumpHere(v, pLevel->addrSkip); + sqlite3VdbeJumpHere(v, pLevel->addrSkip-2); + } +#ifndef SQLITE_LIKE_DOESNT_MATCH_BLOBS + if( pLevel->addrLikeRep ){ + sqlite3VdbeAddOp2(v, OP_DecrJumpZero, (int)(pLevel->iLikeRepCntr>>1), + pLevel->addrLikeRep); + VdbeCoverage(v); + } +#endif + if( pLevel->iLeftJoin ){ + int ws = pLoop->wsFlags; + addr = sqlite3VdbeAddOp1(v, OP_IfPos, pLevel->iLeftJoin); VdbeCoverage(v); + assert( (ws & WHERE_IDX_ONLY)==0 || (ws & WHERE_INDEXED)!=0 ); + if( (ws & WHERE_IDX_ONLY)==0 ){ + assert( pLevel->iTabCur==pTabList->a[pLevel->iFrom].iCursor ); + sqlite3VdbeAddOp1(v, OP_NullRow, pLevel->iTabCur); + } + if( (ws & WHERE_INDEXED) + || ((ws & WHERE_MULTI_OR) && pLevel->u.pCovidx) + ){ + sqlite3VdbeAddOp1(v, OP_NullRow, pLevel->iIdxCur); + } + if( pLevel->op==OP_Return ){ + sqlite3VdbeAddOp2(v, OP_Gosub, pLevel->p1, pLevel->addrFirst); + }else{ + sqlite3VdbeGoto(v, pLevel->addrFirst); + } + sqlite3VdbeJumpHere(v, addr); + } + VdbeModuleComment((v, "End WHERE-loop%d: %s", i, + pWInfo->pTabList->a[pLevel->iFrom].pTab->zName)); + } + + /* The "break" point is here, just past the end of the outer loop. + ** Set it. + */ + sqlite3VdbeResolveLabel(v, pWInfo->iBreak); + + assert( pWInfo->nLevel<=pTabList->nSrc ); + for(i=0, pLevel=pWInfo->a; inLevel; i++, pLevel++){ + int k, last; + VdbeOp *pOp; + Index *pIdx = 0; + struct SrcList_item *pTabItem = &pTabList->a[pLevel->iFrom]; + Table *pTab = pTabItem->pTab; + assert( pTab!=0 ); + pLoop = pLevel->pWLoop; + + /* For a co-routine, change all OP_Column references to the table of + ** the co-routine into OP_Copy of result contained in a register. + ** OP_Rowid becomes OP_Null. + */ + if( pTabItem->fg.viaCoroutine ){ + testcase( pParse->db->mallocFailed ); + translateColumnToCopy(pParse, pLevel->addrBody, pLevel->iTabCur, + pTabItem->regResult, 0); + continue; + } + +#ifdef SQLITE_ENABLE_EARLY_CURSOR_CLOSE + /* Close all of the cursors that were opened by sqlite3WhereBegin. + ** Except, do not close cursors that will be reused by the OR optimization + ** (WHERE_OR_SUBCLAUSE). And do not close the OP_OpenWrite cursors + ** created for the ONEPASS optimization. + */ + if( (pTab->tabFlags & TF_Ephemeral)==0 + && pTab->pSelect==0 + && (pWInfo->wctrlFlags & WHERE_OR_SUBCLAUSE)==0 + ){ + int ws = pLoop->wsFlags; + if( pWInfo->eOnePass==ONEPASS_OFF && (ws & WHERE_IDX_ONLY)==0 ){ + sqlite3VdbeAddOp1(v, OP_Close, pTabItem->iCursor); + } + if( (ws & WHERE_INDEXED)!=0 + && (ws & (WHERE_IPK|WHERE_AUTO_INDEX))==0 + && pLevel->iIdxCur!=pWInfo->aiCurOnePass[1] + ){ + sqlite3VdbeAddOp1(v, OP_Close, pLevel->iIdxCur); + } + } +#endif + + /* If this scan uses an index, make VDBE code substitutions to read data + ** from the index instead of from the table where possible. In some cases + ** this optimization prevents the table from ever being read, which can + ** yield a significant performance boost. + ** + ** Calls to the code generator in between sqlite3WhereBegin and + ** sqlite3WhereEnd will have created code that references the table + ** directly. This loop scans all that code looking for opcodes + ** that reference the table and converts them into opcodes that + ** reference the index. + */ + if( pLoop->wsFlags & (WHERE_INDEXED|WHERE_IDX_ONLY) ){ + pIdx = pLoop->u.btree.pIndex; + }else if( pLoop->wsFlags & WHERE_MULTI_OR ){ + pIdx = pLevel->u.pCovidx; + } + if( pIdx + && (pWInfo->eOnePass==ONEPASS_OFF || !HasRowid(pIdx->pTable)) + && !db->mallocFailed + ){ + last = sqlite3VdbeCurrentAddr(v); + k = pLevel->addrBody; +#ifdef SQLITE_DEBUG + if( db->flags & SQLITE_VdbeAddopTrace ){ + printf("TRANSLATE opcodes in range %d..%d\n", k, last-1); + } +#endif + pOp = sqlite3VdbeGetOp(v, k); + for(; kp1!=pLevel->iTabCur ) continue; + if( pOp->opcode==OP_Column +#ifdef SQLITE_ENABLE_OFFSET_SQL_FUNC + || pOp->opcode==OP_Offset +#endif + ){ + int x = pOp->p2; + assert( pIdx->pTable==pTab ); + if( !HasRowid(pTab) ){ + Index *pPk = sqlite3PrimaryKeyIndex(pTab); + x = pPk->aiColumn[x]; + assert( x>=0 ); + } + x = sqlite3ColumnOfIndex(pIdx, x); + if( x>=0 ){ + pOp->p2 = x; + pOp->p1 = pLevel->iIdxCur; + OpcodeRewriteTrace(db, k, pOp); + } + assert( (pLoop->wsFlags & WHERE_IDX_ONLY)==0 || x>=0 + || pWInfo->eOnePass ); + }else if( pOp->opcode==OP_Rowid ){ + pOp->p1 = pLevel->iIdxCur; + pOp->opcode = OP_IdxRowid; + OpcodeRewriteTrace(db, k, pOp); + }else if( pOp->opcode==OP_IfNullRow ){ + pOp->p1 = pLevel->iIdxCur; + OpcodeRewriteTrace(db, k, pOp); + } + } +#ifdef SQLITE_DEBUG + if( db->flags & SQLITE_VdbeAddopTrace ) printf("TRANSLATE complete\n"); +#endif + } + } + + /* Final cleanup + */ + pParse->nQueryLoop = pWInfo->savedNQueryLoop; + whereInfoFree(db, pWInfo); + return; +} + +/************** End of where.c ***********************************************/ +/************** Begin file window.c ******************************************/ +/* +** 2018 May 08 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +*/ +/* #include "sqliteInt.h" */ + +#ifndef SQLITE_OMIT_WINDOWFUNC + +/* +** SELECT REWRITING +** +** Any SELECT statement that contains one or more window functions in +** either the select list or ORDER BY clause (the only two places window +** functions may be used) is transformed by function sqlite3WindowRewrite() +** in order to support window function processing. For example, with the +** schema: +** +** CREATE TABLE t1(a, b, c, d, e, f, g); +** +** the statement: +** +** SELECT a+1, max(b) OVER (PARTITION BY c ORDER BY d) FROM t1 ORDER BY e; +** +** is transformed to: +** +** SELECT a+1, max(b) OVER (PARTITION BY c ORDER BY d) FROM ( +** SELECT a, e, c, d, b FROM t1 ORDER BY c, d +** ) ORDER BY e; +** +** The flattening optimization is disabled when processing this transformed +** SELECT statement. This allows the implementation of the window function +** (in this case max()) to process rows sorted in order of (c, d), which +** makes things easier for obvious reasons. More generally: +** +** * FROM, WHERE, GROUP BY and HAVING clauses are all moved to +** the sub-query. +** +** * ORDER BY, LIMIT and OFFSET remain part of the parent query. +** +** * Terminals from each of the expression trees that make up the +** select-list and ORDER BY expressions in the parent query are +** selected by the sub-query. For the purposes of the transformation, +** terminals are column references and aggregate functions. +** +** If there is more than one window function in the SELECT that uses +** the same window declaration (the OVER bit), then a single scan may +** be used to process more than one window function. For example: +** +** SELECT max(b) OVER (PARTITION BY c ORDER BY d), +** min(e) OVER (PARTITION BY c ORDER BY d) +** FROM t1; +** +** is transformed in the same way as the example above. However: +** +** SELECT max(b) OVER (PARTITION BY c ORDER BY d), +** min(e) OVER (PARTITION BY a ORDER BY b) +** FROM t1; +** +** Must be transformed to: +** +** SELECT max(b) OVER (PARTITION BY c ORDER BY d) FROM ( +** SELECT e, min(e) OVER (PARTITION BY a ORDER BY b), c, d, b FROM +** SELECT a, e, c, d, b FROM t1 ORDER BY a, b +** ) ORDER BY c, d +** ) ORDER BY e; +** +** so that both min() and max() may process rows in the order defined by +** their respective window declarations. +** +** INTERFACE WITH SELECT.C +** +** When processing the rewritten SELECT statement, code in select.c calls +** sqlite3WhereBegin() to begin iterating through the results of the +** sub-query, which is always implemented as a co-routine. It then calls +** sqlite3WindowCodeStep() to process rows and finish the scan by calling +** sqlite3WhereEnd(). +** +** sqlite3WindowCodeStep() generates VM code so that, for each row returned +** by the sub-query a sub-routine (OP_Gosub) coded by select.c is invoked. +** When the sub-routine is invoked: +** +** * The results of all window-functions for the row are stored +** in the associated Window.regResult registers. +** +** * The required terminal values are stored in the current row of +** temp table Window.iEphCsr. +** +** In some cases, depending on the window frame and the specific window +** functions invoked, sqlite3WindowCodeStep() caches each entire partition +** in a temp table before returning any rows. In other cases it does not. +** This detail is encapsulated within this file, the code generated by +** select.c is the same in either case. +** +** BUILT-IN WINDOW FUNCTIONS +** +** This implementation features the following built-in window functions: +** +** row_number() +** rank() +** dense_rank() +** percent_rank() +** cume_dist() +** ntile(N) +** lead(expr [, offset [, default]]) +** lag(expr [, offset [, default]]) +** first_value(expr) +** last_value(expr) +** nth_value(expr, N) +** +** These are the same built-in window functions supported by Postgres. +** Although the behaviour of aggregate window functions (functions that +** can be used as either aggregates or window funtions) allows them to +** be implemented using an API, built-in window functions are much more +** esoteric. Additionally, some window functions (e.g. nth_value()) +** may only be implemented by caching the entire partition in memory. +** As such, some built-in window functions use the same API as aggregate +** window functions and some are implemented directly using VDBE +** instructions. Additionally, for those functions that use the API, the +** window frame is sometimes modified before the SELECT statement is +** rewritten. For example, regardless of the specified window frame, the +** row_number() function always uses: +** +** ROWS BETWEEN UNBOUNDED PRECEDING AND CURRENT ROW +** +** See sqlite3WindowUpdate() for details. +** +** As well as some of the built-in window functions, aggregate window +** functions min() and max() are implemented using VDBE instructions if +** the start of the window frame is declared as anything other than +** UNBOUNDED PRECEDING. +*/ + +/* +** Implementation of built-in window function row_number(). Assumes that the +** window frame has been coerced to: +** +** ROWS BETWEEN UNBOUNDED PRECEDING AND CURRENT ROW +*/ +static void row_numberStepFunc( + sqlite3_context *pCtx, + int nArg, + sqlite3_value **apArg +){ + i64 *p = (i64*)sqlite3_aggregate_context(pCtx, sizeof(*p)); + if( p ) (*p)++; + UNUSED_PARAMETER(nArg); + UNUSED_PARAMETER(apArg); +} +static void row_numberValueFunc(sqlite3_context *pCtx){ + i64 *p = (i64*)sqlite3_aggregate_context(pCtx, sizeof(*p)); + sqlite3_result_int64(pCtx, (p ? *p : 0)); +} + +/* +** Context object type used by rank(), dense_rank(), percent_rank() and +** cume_dist(). +*/ +struct CallCount { + i64 nValue; + i64 nStep; + i64 nTotal; +}; + +/* +** Implementation of built-in window function dense_rank(). Assumes that +** the window frame has been set to: +** +** RANGE BETWEEN UNBOUNDED PRECEDING AND CURRENT ROW +*/ +static void dense_rankStepFunc( + sqlite3_context *pCtx, + int nArg, + sqlite3_value **apArg +){ + struct CallCount *p; + p = (struct CallCount*)sqlite3_aggregate_context(pCtx, sizeof(*p)); + if( p ) p->nStep = 1; + UNUSED_PARAMETER(nArg); + UNUSED_PARAMETER(apArg); +} +static void dense_rankValueFunc(sqlite3_context *pCtx){ + struct CallCount *p; + p = (struct CallCount*)sqlite3_aggregate_context(pCtx, sizeof(*p)); + if( p ){ + if( p->nStep ){ + p->nValue++; + p->nStep = 0; + } + sqlite3_result_int64(pCtx, p->nValue); + } +} + +/* +** Implementation of built-in window function nth_value(). This +** implementation is used in "slow mode" only - when the EXCLUDE clause +** is not set to the default value "NO OTHERS". +*/ +struct NthValueCtx { + i64 nStep; + sqlite3_value *pValue; +}; +static void nth_valueStepFunc( + sqlite3_context *pCtx, + int nArg, + sqlite3_value **apArg +){ + struct NthValueCtx *p; + p = (struct NthValueCtx*)sqlite3_aggregate_context(pCtx, sizeof(*p)); + if( p ){ + i64 iVal; + switch( sqlite3_value_numeric_type(apArg[1]) ){ + case SQLITE_INTEGER: + iVal = sqlite3_value_int64(apArg[1]); + break; + case SQLITE_FLOAT: { + double fVal = sqlite3_value_double(apArg[1]); + if( ((i64)fVal)!=fVal ) goto error_out; + iVal = (i64)fVal; + break; + } + default: + goto error_out; + } + if( iVal<=0 ) goto error_out; + + p->nStep++; + if( iVal==p->nStep ){ + p->pValue = sqlite3_value_dup(apArg[0]); + if( !p->pValue ){ + sqlite3_result_error_nomem(pCtx); + } + } + } + UNUSED_PARAMETER(nArg); + UNUSED_PARAMETER(apArg); + return; + + error_out: + sqlite3_result_error( + pCtx, "second argument to nth_value must be a positive integer", -1 + ); +} +static void nth_valueFinalizeFunc(sqlite3_context *pCtx){ + struct NthValueCtx *p; + p = (struct NthValueCtx*)sqlite3_aggregate_context(pCtx, 0); + if( p && p->pValue ){ + sqlite3_result_value(pCtx, p->pValue); + sqlite3_value_free(p->pValue); + p->pValue = 0; + } +} +#define nth_valueInvFunc noopStepFunc +#define nth_valueValueFunc noopValueFunc + +static void first_valueStepFunc( + sqlite3_context *pCtx, + int nArg, + sqlite3_value **apArg +){ + struct NthValueCtx *p; + p = (struct NthValueCtx*)sqlite3_aggregate_context(pCtx, sizeof(*p)); + if( p && p->pValue==0 ){ + p->pValue = sqlite3_value_dup(apArg[0]); + if( !p->pValue ){ + sqlite3_result_error_nomem(pCtx); + } + } + UNUSED_PARAMETER(nArg); + UNUSED_PARAMETER(apArg); +} +static void first_valueFinalizeFunc(sqlite3_context *pCtx){ + struct NthValueCtx *p; + p = (struct NthValueCtx*)sqlite3_aggregate_context(pCtx, sizeof(*p)); + if( p && p->pValue ){ + sqlite3_result_value(pCtx, p->pValue); + sqlite3_value_free(p->pValue); + p->pValue = 0; + } +} +#define first_valueInvFunc noopStepFunc +#define first_valueValueFunc noopValueFunc + +/* +** Implementation of built-in window function rank(). Assumes that +** the window frame has been set to: +** +** RANGE BETWEEN UNBOUNDED PRECEDING AND CURRENT ROW +*/ +static void rankStepFunc( + sqlite3_context *pCtx, + int nArg, + sqlite3_value **apArg +){ + struct CallCount *p; + p = (struct CallCount*)sqlite3_aggregate_context(pCtx, sizeof(*p)); + if( p ){ + p->nStep++; + if( p->nValue==0 ){ + p->nValue = p->nStep; + } + } + UNUSED_PARAMETER(nArg); + UNUSED_PARAMETER(apArg); +} +static void rankValueFunc(sqlite3_context *pCtx){ + struct CallCount *p; + p = (struct CallCount*)sqlite3_aggregate_context(pCtx, sizeof(*p)); + if( p ){ + sqlite3_result_int64(pCtx, p->nValue); + p->nValue = 0; + } +} + +/* +** Implementation of built-in window function percent_rank(). Assumes that +** the window frame has been set to: +** +** GROUPS BETWEEN CURRENT ROW AND UNBOUNDED FOLLOWING +*/ +static void percent_rankStepFunc( + sqlite3_context *pCtx, + int nArg, + sqlite3_value **apArg +){ + struct CallCount *p; + UNUSED_PARAMETER(nArg); assert( nArg==0 ); + UNUSED_PARAMETER(apArg); + p = (struct CallCount*)sqlite3_aggregate_context(pCtx, sizeof(*p)); + if( p ){ + p->nTotal++; + } +} +static void percent_rankInvFunc( + sqlite3_context *pCtx, + int nArg, + sqlite3_value **apArg +){ + struct CallCount *p; + UNUSED_PARAMETER(nArg); assert( nArg==0 ); + UNUSED_PARAMETER(apArg); + p = (struct CallCount*)sqlite3_aggregate_context(pCtx, sizeof(*p)); + p->nStep++; +} +static void percent_rankValueFunc(sqlite3_context *pCtx){ + struct CallCount *p; + p = (struct CallCount*)sqlite3_aggregate_context(pCtx, sizeof(*p)); + if( p ){ + p->nValue = p->nStep; + if( p->nTotal>1 ){ + double r = (double)p->nValue / (double)(p->nTotal-1); + sqlite3_result_double(pCtx, r); + }else{ + sqlite3_result_double(pCtx, 0.0); + } + } +} +#define percent_rankFinalizeFunc percent_rankValueFunc + +/* +** Implementation of built-in window function cume_dist(). Assumes that +** the window frame has been set to: +** +** GROUPS BETWEEN 1 FOLLOWING AND UNBOUNDED FOLLOWING +*/ +static void cume_distStepFunc( + sqlite3_context *pCtx, + int nArg, + sqlite3_value **apArg +){ + struct CallCount *p; + UNUSED_PARAMETER(nArg); assert( nArg==0 ); + UNUSED_PARAMETER(apArg); + p = (struct CallCount*)sqlite3_aggregate_context(pCtx, sizeof(*p)); + if( p ){ + p->nTotal++; + } +} +static void cume_distInvFunc( + sqlite3_context *pCtx, + int nArg, + sqlite3_value **apArg +){ + struct CallCount *p; + UNUSED_PARAMETER(nArg); assert( nArg==0 ); + UNUSED_PARAMETER(apArg); + p = (struct CallCount*)sqlite3_aggregate_context(pCtx, sizeof(*p)); + p->nStep++; +} +static void cume_distValueFunc(sqlite3_context *pCtx){ + struct CallCount *p; + p = (struct CallCount*)sqlite3_aggregate_context(pCtx, 0); + if( p ){ + double r = (double)(p->nStep) / (double)(p->nTotal); + sqlite3_result_double(pCtx, r); + } +} +#define cume_distFinalizeFunc cume_distValueFunc + +/* +** Context object for ntile() window function. +*/ +struct NtileCtx { + i64 nTotal; /* Total rows in partition */ + i64 nParam; /* Parameter passed to ntile(N) */ + i64 iRow; /* Current row */ +}; + +/* +** Implementation of ntile(). This assumes that the window frame has +** been coerced to: +** +** ROWS CURRENT ROW AND UNBOUNDED FOLLOWING +*/ +static void ntileStepFunc( + sqlite3_context *pCtx, + int nArg, + sqlite3_value **apArg +){ + struct NtileCtx *p; + assert( nArg==1 ); UNUSED_PARAMETER(nArg); + p = (struct NtileCtx*)sqlite3_aggregate_context(pCtx, sizeof(*p)); + if( p ){ + if( p->nTotal==0 ){ + p->nParam = sqlite3_value_int64(apArg[0]); + if( p->nParam<=0 ){ + sqlite3_result_error( + pCtx, "argument of ntile must be a positive integer", -1 + ); + } + } + p->nTotal++; + } +} +static void ntileInvFunc( + sqlite3_context *pCtx, + int nArg, + sqlite3_value **apArg +){ + struct NtileCtx *p; + assert( nArg==1 ); UNUSED_PARAMETER(nArg); + UNUSED_PARAMETER(apArg); + p = (struct NtileCtx*)sqlite3_aggregate_context(pCtx, sizeof(*p)); + p->iRow++; +} +static void ntileValueFunc(sqlite3_context *pCtx){ + struct NtileCtx *p; + p = (struct NtileCtx*)sqlite3_aggregate_context(pCtx, sizeof(*p)); + if( p && p->nParam>0 ){ + int nSize = (p->nTotal / p->nParam); + if( nSize==0 ){ + sqlite3_result_int64(pCtx, p->iRow+1); + }else{ + i64 nLarge = p->nTotal - p->nParam*nSize; + i64 iSmall = nLarge*(nSize+1); + i64 iRow = p->iRow; + + assert( (nLarge*(nSize+1) + (p->nParam-nLarge)*nSize)==p->nTotal ); + + if( iRowpVal); + p->pVal = sqlite3_value_dup(apArg[0]); + if( p->pVal==0 ){ + sqlite3_result_error_nomem(pCtx); + }else{ + p->nVal++; + } + } +} +static void last_valueInvFunc( + sqlite3_context *pCtx, + int nArg, + sqlite3_value **apArg +){ + struct LastValueCtx *p; + UNUSED_PARAMETER(nArg); + UNUSED_PARAMETER(apArg); + p = (struct LastValueCtx*)sqlite3_aggregate_context(pCtx, sizeof(*p)); + if( ALWAYS(p) ){ + p->nVal--; + if( p->nVal==0 ){ + sqlite3_value_free(p->pVal); + p->pVal = 0; + } + } +} +static void last_valueValueFunc(sqlite3_context *pCtx){ + struct LastValueCtx *p; + p = (struct LastValueCtx*)sqlite3_aggregate_context(pCtx, 0); + if( p && p->pVal ){ + sqlite3_result_value(pCtx, p->pVal); + } +} +static void last_valueFinalizeFunc(sqlite3_context *pCtx){ + struct LastValueCtx *p; + p = (struct LastValueCtx*)sqlite3_aggregate_context(pCtx, sizeof(*p)); + if( p && p->pVal ){ + sqlite3_result_value(pCtx, p->pVal); + sqlite3_value_free(p->pVal); + p->pVal = 0; + } +} + +/* +** Static names for the built-in window function names. These static +** names are used, rather than string literals, so that FuncDef objects +** can be associated with a particular window function by direct +** comparison of the zName pointer. Example: +** +** if( pFuncDef->zName==row_valueName ){ ... } +*/ +static const char row_numberName[] = "row_number"; +static const char dense_rankName[] = "dense_rank"; +static const char rankName[] = "rank"; +static const char percent_rankName[] = "percent_rank"; +static const char cume_distName[] = "cume_dist"; +static const char ntileName[] = "ntile"; +static const char last_valueName[] = "last_value"; +static const char nth_valueName[] = "nth_value"; +static const char first_valueName[] = "first_value"; +static const char leadName[] = "lead"; +static const char lagName[] = "lag"; + +/* +** No-op implementations of xStep() and xFinalize(). Used as place-holders +** for built-in window functions that never call those interfaces. +** +** The noopValueFunc() is called but is expected to do nothing. The +** noopStepFunc() is never called, and so it is marked with NO_TEST to +** let the test coverage routine know not to expect this function to be +** invoked. +*/ +static void noopStepFunc( /*NO_TEST*/ + sqlite3_context *p, /*NO_TEST*/ + int n, /*NO_TEST*/ + sqlite3_value **a /*NO_TEST*/ +){ /*NO_TEST*/ + UNUSED_PARAMETER(p); /*NO_TEST*/ + UNUSED_PARAMETER(n); /*NO_TEST*/ + UNUSED_PARAMETER(a); /*NO_TEST*/ + assert(0); /*NO_TEST*/ +} /*NO_TEST*/ +static void noopValueFunc(sqlite3_context *p){ UNUSED_PARAMETER(p); /*no-op*/ } + +/* Window functions that use all window interfaces: xStep, xFinal, +** xValue, and xInverse */ +#define WINDOWFUNCALL(name,nArg,extra) { \ + nArg, (SQLITE_UTF8|SQLITE_FUNC_WINDOW|extra), 0, 0, \ + name ## StepFunc, name ## FinalizeFunc, name ## ValueFunc, \ + name ## InvFunc, name ## Name, {0} \ +} + +/* Window functions that are implemented using bytecode and thus have +** no-op routines for their methods */ +#define WINDOWFUNCNOOP(name,nArg,extra) { \ + nArg, (SQLITE_UTF8|SQLITE_FUNC_WINDOW|extra), 0, 0, \ + noopStepFunc, noopValueFunc, noopValueFunc, \ + noopStepFunc, name ## Name, {0} \ +} + +/* Window functions that use all window interfaces: xStep, the +** same routine for xFinalize and xValue and which never call +** xInverse. */ +#define WINDOWFUNCX(name,nArg,extra) { \ + nArg, (SQLITE_UTF8|SQLITE_FUNC_WINDOW|extra), 0, 0, \ + name ## StepFunc, name ## ValueFunc, name ## ValueFunc, \ + noopStepFunc, name ## Name, {0} \ +} + + +/* +** Register those built-in window functions that are not also aggregates. +*/ +SQLITE_PRIVATE void sqlite3WindowFunctions(void){ + static FuncDef aWindowFuncs[] = { + WINDOWFUNCX(row_number, 0, 0), + WINDOWFUNCX(dense_rank, 0, 0), + WINDOWFUNCX(rank, 0, 0), + WINDOWFUNCALL(percent_rank, 0, 0), + WINDOWFUNCALL(cume_dist, 0, 0), + WINDOWFUNCALL(ntile, 1, 0), + WINDOWFUNCALL(last_value, 1, 0), + WINDOWFUNCALL(nth_value, 2, 0), + WINDOWFUNCALL(first_value, 1, 0), + WINDOWFUNCNOOP(lead, 1, 0), + WINDOWFUNCNOOP(lead, 2, 0), + WINDOWFUNCNOOP(lead, 3, 0), + WINDOWFUNCNOOP(lag, 1, 0), + WINDOWFUNCNOOP(lag, 2, 0), + WINDOWFUNCNOOP(lag, 3, 0), + }; + sqlite3InsertBuiltinFuncs(aWindowFuncs, ArraySize(aWindowFuncs)); +} + +static Window *windowFind(Parse *pParse, Window *pList, const char *zName){ + Window *p; + for(p=pList; p; p=p->pNextWin){ + if( sqlite3StrICmp(p->zName, zName)==0 ) break; + } + if( p==0 ){ + sqlite3ErrorMsg(pParse, "no such window: %s", zName); + } + return p; +} + +/* +** This function is called immediately after resolving the function name +** for a window function within a SELECT statement. Argument pList is a +** linked list of WINDOW definitions for the current SELECT statement. +** Argument pFunc is the function definition just resolved and pWin +** is the Window object representing the associated OVER clause. This +** function updates the contents of pWin as follows: +** +** * If the OVER clause refered to a named window (as in "max(x) OVER win"), +** search list pList for a matching WINDOW definition, and update pWin +** accordingly. If no such WINDOW clause can be found, leave an error +** in pParse. +** +** * If the function is a built-in window function that requires the +** window to be coerced (see "BUILT-IN WINDOW FUNCTIONS" at the top +** of this file), pWin is updated here. +*/ +SQLITE_PRIVATE void sqlite3WindowUpdate( + Parse *pParse, + Window *pList, /* List of named windows for this SELECT */ + Window *pWin, /* Window frame to update */ + FuncDef *pFunc /* Window function definition */ +){ + if( pWin->zName && pWin->eFrmType==0 ){ + Window *p = windowFind(pParse, pList, pWin->zName); + if( p==0 ) return; + pWin->pPartition = sqlite3ExprListDup(pParse->db, p->pPartition, 0); + pWin->pOrderBy = sqlite3ExprListDup(pParse->db, p->pOrderBy, 0); + pWin->pStart = sqlite3ExprDup(pParse->db, p->pStart, 0); + pWin->pEnd = sqlite3ExprDup(pParse->db, p->pEnd, 0); + pWin->eStart = p->eStart; + pWin->eEnd = p->eEnd; + pWin->eFrmType = p->eFrmType; + pWin->eExclude = p->eExclude; + }else{ + sqlite3WindowChain(pParse, pWin, pList); + } + if( (pWin->eFrmType==TK_RANGE) + && (pWin->pStart || pWin->pEnd) + && (pWin->pOrderBy==0 || pWin->pOrderBy->nExpr!=1) + ){ + sqlite3ErrorMsg(pParse, + "RANGE with offset PRECEDING/FOLLOWING requires one ORDER BY expression" + ); + }else + if( pFunc->funcFlags & SQLITE_FUNC_WINDOW ){ + sqlite3 *db = pParse->db; + if( pWin->pFilter ){ + sqlite3ErrorMsg(pParse, + "FILTER clause may only be used with aggregate window functions" + ); + }else{ + struct WindowUpdate { + const char *zFunc; + int eFrmType; + int eStart; + int eEnd; + } aUp[] = { + { row_numberName, TK_ROWS, TK_UNBOUNDED, TK_CURRENT }, + { dense_rankName, TK_RANGE, TK_UNBOUNDED, TK_CURRENT }, + { rankName, TK_RANGE, TK_UNBOUNDED, TK_CURRENT }, + { percent_rankName, TK_GROUPS, TK_CURRENT, TK_UNBOUNDED }, + { cume_distName, TK_GROUPS, TK_FOLLOWING, TK_UNBOUNDED }, + { ntileName, TK_ROWS, TK_CURRENT, TK_UNBOUNDED }, + { leadName, TK_ROWS, TK_UNBOUNDED, TK_UNBOUNDED }, + { lagName, TK_ROWS, TK_UNBOUNDED, TK_CURRENT }, + }; + int i; + for(i=0; izName==aUp[i].zFunc ){ + sqlite3ExprDelete(db, pWin->pStart); + sqlite3ExprDelete(db, pWin->pEnd); + pWin->pEnd = pWin->pStart = 0; + pWin->eFrmType = aUp[i].eFrmType; + pWin->eStart = aUp[i].eStart; + pWin->eEnd = aUp[i].eEnd; + pWin->eExclude = 0; + if( pWin->eStart==TK_FOLLOWING ){ + pWin->pStart = sqlite3Expr(db, TK_INTEGER, "1"); + } + break; + } + } + } + } + pWin->pFunc = pFunc; +} + +/* +** Context object passed through sqlite3WalkExprList() to +** selectWindowRewriteExprCb() by selectWindowRewriteEList(). +*/ +typedef struct WindowRewrite WindowRewrite; +struct WindowRewrite { + Window *pWin; + SrcList *pSrc; + ExprList *pSub; + Table *pTab; + Select *pSubSelect; /* Current sub-select, if any */ +}; + +/* +** Callback function used by selectWindowRewriteEList(). If necessary, +** this function appends to the output expression-list and updates +** expression (*ppExpr) in place. +*/ +static int selectWindowRewriteExprCb(Walker *pWalker, Expr *pExpr){ + struct WindowRewrite *p = pWalker->u.pRewrite; + Parse *pParse = pWalker->pParse; + assert( p!=0 ); + assert( p->pWin!=0 ); + + /* If this function is being called from within a scalar sub-select + ** that used by the SELECT statement being processed, only process + ** TK_COLUMN expressions that refer to it (the outer SELECT). Do + ** not process aggregates or window functions at all, as they belong + ** to the scalar sub-select. */ + if( p->pSubSelect ){ + if( pExpr->op!=TK_COLUMN ){ + return WRC_Continue; + }else{ + int nSrc = p->pSrc->nSrc; + int i; + for(i=0; iiTable==p->pSrc->a[i].iCursor ) break; + } + if( i==nSrc ) return WRC_Continue; + } + } + + switch( pExpr->op ){ + + case TK_FUNCTION: + if( !ExprHasProperty(pExpr, EP_WinFunc) ){ + break; + }else{ + Window *pWin; + for(pWin=p->pWin; pWin; pWin=pWin->pNextWin){ + if( pExpr->y.pWin==pWin ){ + assert( pWin->pOwner==pExpr ); + return WRC_Prune; + } + } + } + /* Fall through. */ + + case TK_AGG_FUNCTION: + case TK_COLUMN: { + Expr *pDup = sqlite3ExprDup(pParse->db, pExpr, 0); + p->pSub = sqlite3ExprListAppend(pParse, p->pSub, pDup); + if( p->pSub ){ + assert( ExprHasProperty(pExpr, EP_Static)==0 ); + ExprSetProperty(pExpr, EP_Static); + sqlite3ExprDelete(pParse->db, pExpr); + ExprClearProperty(pExpr, EP_Static); + memset(pExpr, 0, sizeof(Expr)); + + pExpr->op = TK_COLUMN; + pExpr->iColumn = p->pSub->nExpr-1; + pExpr->iTable = p->pWin->iEphCsr; + pExpr->y.pTab = p->pTab; + } + + break; + } + + default: /* no-op */ + break; + } + + return WRC_Continue; +} +static int selectWindowRewriteSelectCb(Walker *pWalker, Select *pSelect){ + struct WindowRewrite *p = pWalker->u.pRewrite; + Select *pSave = p->pSubSelect; + if( pSave==pSelect ){ + return WRC_Continue; + }else{ + p->pSubSelect = pSelect; + sqlite3WalkSelect(pWalker, pSelect); + p->pSubSelect = pSave; + } + return WRC_Prune; +} + + +/* +** Iterate through each expression in expression-list pEList. For each: +** +** * TK_COLUMN, +** * aggregate function, or +** * window function with a Window object that is not a member of the +** Window list passed as the second argument (pWin). +** +** Append the node to output expression-list (*ppSub). And replace it +** with a TK_COLUMN that reads the (N-1)th element of table +** pWin->iEphCsr, where N is the number of elements in (*ppSub) after +** appending the new one. +*/ +static void selectWindowRewriteEList( + Parse *pParse, + Window *pWin, + SrcList *pSrc, + ExprList *pEList, /* Rewrite expressions in this list */ + Table *pTab, + ExprList **ppSub /* IN/OUT: Sub-select expression-list */ +){ + Walker sWalker; + WindowRewrite sRewrite; + + assert( pWin!=0 ); + memset(&sWalker, 0, sizeof(Walker)); + memset(&sRewrite, 0, sizeof(WindowRewrite)); + + sRewrite.pSub = *ppSub; + sRewrite.pWin = pWin; + sRewrite.pSrc = pSrc; + sRewrite.pTab = pTab; + + sWalker.pParse = pParse; + sWalker.xExprCallback = selectWindowRewriteExprCb; + sWalker.xSelectCallback = selectWindowRewriteSelectCb; + sWalker.u.pRewrite = &sRewrite; + + (void)sqlite3WalkExprList(&sWalker, pEList); + + *ppSub = sRewrite.pSub; +} + +/* +** Append a copy of each expression in expression-list pAppend to +** expression list pList. Return a pointer to the result list. +*/ +static ExprList *exprListAppendList( + Parse *pParse, /* Parsing context */ + ExprList *pList, /* List to which to append. Might be NULL */ + ExprList *pAppend, /* List of values to append. Might be NULL */ + int bIntToNull +){ + if( pAppend ){ + int i; + int nInit = pList ? pList->nExpr : 0; + for(i=0; inExpr; i++){ + Expr *pDup = sqlite3ExprDup(pParse->db, pAppend->a[i].pExpr, 0); + if( bIntToNull && pDup && pDup->op==TK_INTEGER ){ + pDup->op = TK_NULL; + pDup->flags &= ~(EP_IntValue|EP_IsTrue|EP_IsFalse); + } + pList = sqlite3ExprListAppend(pParse, pList, pDup); + if( pList ) pList->a[nInit+i].sortFlags = pAppend->a[i].sortFlags; + } + } + return pList; +} + +/* +** If the SELECT statement passed as the second argument does not invoke +** any SQL window functions, this function is a no-op. Otherwise, it +** rewrites the SELECT statement so that window function xStep functions +** are invoked in the correct order as described under "SELECT REWRITING" +** at the top of this file. +*/ +SQLITE_PRIVATE int sqlite3WindowRewrite(Parse *pParse, Select *p){ + int rc = SQLITE_OK; + if( p->pWin && p->pPrior==0 ){ + Vdbe *v = sqlite3GetVdbe(pParse); + sqlite3 *db = pParse->db; + Select *pSub = 0; /* The subquery */ + SrcList *pSrc = p->pSrc; + Expr *pWhere = p->pWhere; + ExprList *pGroupBy = p->pGroupBy; + Expr *pHaving = p->pHaving; + ExprList *pSort = 0; + + ExprList *pSublist = 0; /* Expression list for sub-query */ + Window *pMWin = p->pWin; /* Master window object */ + Window *pWin; /* Window object iterator */ + Table *pTab; + + pTab = sqlite3DbMallocZero(db, sizeof(Table)); + if( pTab==0 ){ + return SQLITE_NOMEM; + } + + p->pSrc = 0; + p->pWhere = 0; + p->pGroupBy = 0; + p->pHaving = 0; + p->selFlags &= ~SF_Aggregate; + + /* Create the ORDER BY clause for the sub-select. This is the concatenation + ** of the window PARTITION and ORDER BY clauses. Then, if this makes it + ** redundant, remove the ORDER BY from the parent SELECT. */ + pSort = sqlite3ExprListDup(db, pMWin->pPartition, 0); + pSort = exprListAppendList(pParse, pSort, pMWin->pOrderBy, 1); + if( pSort && p->pOrderBy && p->pOrderBy->nExpr<=pSort->nExpr ){ + int nSave = pSort->nExpr; + pSort->nExpr = p->pOrderBy->nExpr; + if( sqlite3ExprListCompare(pSort, p->pOrderBy, -1)==0 ){ + sqlite3ExprListDelete(db, p->pOrderBy); + p->pOrderBy = 0; + } + pSort->nExpr = nSave; + } + + /* Assign a cursor number for the ephemeral table used to buffer rows. + ** The OpenEphemeral instruction is coded later, after it is known how + ** many columns the table will have. */ + pMWin->iEphCsr = pParse->nTab++; + pParse->nTab += 3; + + selectWindowRewriteEList(pParse, pMWin, pSrc, p->pEList, pTab, &pSublist); + selectWindowRewriteEList(pParse, pMWin, pSrc, p->pOrderBy, pTab, &pSublist); + pMWin->nBufferCol = (pSublist ? pSublist->nExpr : 0); + + /* Append the PARTITION BY and ORDER BY expressions to the to the + ** sub-select expression list. They are required to figure out where + ** boundaries for partitions and sets of peer rows lie. */ + pSublist = exprListAppendList(pParse, pSublist, pMWin->pPartition, 0); + pSublist = exprListAppendList(pParse, pSublist, pMWin->pOrderBy, 0); + + /* Append the arguments passed to each window function to the + ** sub-select expression list. Also allocate two registers for each + ** window function - one for the accumulator, another for interim + ** results. */ + for(pWin=pMWin; pWin; pWin=pWin->pNextWin){ + ExprList *pArgs = pWin->pOwner->x.pList; + if( pWin->pFunc->funcFlags & SQLITE_FUNC_SUBTYPE ){ + selectWindowRewriteEList(pParse, pMWin, pSrc, pArgs, pTab, &pSublist); + pWin->iArgCol = (pSublist ? pSublist->nExpr : 0); + pWin->bExprArgs = 1; + }else{ + pWin->iArgCol = (pSublist ? pSublist->nExpr : 0); + pSublist = exprListAppendList(pParse, pSublist, pArgs, 0); + } + if( pWin->pFilter ){ + Expr *pFilter = sqlite3ExprDup(db, pWin->pFilter, 0); + pSublist = sqlite3ExprListAppend(pParse, pSublist, pFilter); + } + pWin->regAccum = ++pParse->nMem; + pWin->regResult = ++pParse->nMem; + sqlite3VdbeAddOp2(v, OP_Null, 0, pWin->regAccum); + } + + /* If there is no ORDER BY or PARTITION BY clause, and the window + ** function accepts zero arguments, and there are no other columns + ** selected (e.g. "SELECT row_number() OVER () FROM t1"), it is possible + ** that pSublist is still NULL here. Add a constant expression here to + ** keep everything legal in this case. + */ + if( pSublist==0 ){ + pSublist = sqlite3ExprListAppend(pParse, 0, + sqlite3Expr(db, TK_INTEGER, "0") + ); + } + + pSub = sqlite3SelectNew( + pParse, pSublist, pSrc, pWhere, pGroupBy, pHaving, pSort, 0, 0 + ); + p->pSrc = sqlite3SrcListAppend(pParse, 0, 0, 0); + if( p->pSrc ){ + Table *pTab2; + p->pSrc->a[0].pSelect = pSub; + sqlite3SrcListAssignCursors(pParse, p->pSrc); + pSub->selFlags |= SF_Expanded; + pTab2 = sqlite3ResultSetOfSelect(pParse, pSub, SQLITE_AFF_NONE); + if( pTab2==0 ){ + rc = SQLITE_NOMEM; + }else{ + memcpy(pTab, pTab2, sizeof(Table)); + pTab->tabFlags |= TF_Ephemeral; + p->pSrc->a[0].pTab = pTab; + pTab = pTab2; + } + sqlite3VdbeAddOp2(v, OP_OpenEphemeral, pMWin->iEphCsr, pSublist->nExpr); + sqlite3VdbeAddOp2(v, OP_OpenDup, pMWin->iEphCsr+1, pMWin->iEphCsr); + sqlite3VdbeAddOp2(v, OP_OpenDup, pMWin->iEphCsr+2, pMWin->iEphCsr); + sqlite3VdbeAddOp2(v, OP_OpenDup, pMWin->iEphCsr+3, pMWin->iEphCsr); + }else{ + sqlite3SelectDelete(db, pSub); + } + if( db->mallocFailed ) rc = SQLITE_NOMEM; + sqlite3DbFree(db, pTab); + } + + return rc; +} + +/* +** Unlink the Window object from the Select to which it is attached, +** if it is attached. +*/ +SQLITE_PRIVATE void sqlite3WindowUnlinkFromSelect(Window *p){ + if( p->ppThis ){ + *p->ppThis = p->pNextWin; + if( p->pNextWin ) p->pNextWin->ppThis = p->ppThis; + p->ppThis = 0; + } +} + +/* +** Free the Window object passed as the second argument. +*/ +SQLITE_PRIVATE void sqlite3WindowDelete(sqlite3 *db, Window *p){ + if( p ){ + sqlite3WindowUnlinkFromSelect(p); + sqlite3ExprDelete(db, p->pFilter); + sqlite3ExprListDelete(db, p->pPartition); + sqlite3ExprListDelete(db, p->pOrderBy); + sqlite3ExprDelete(db, p->pEnd); + sqlite3ExprDelete(db, p->pStart); + sqlite3DbFree(db, p->zName); + sqlite3DbFree(db, p->zBase); + sqlite3DbFree(db, p); + } +} + +/* +** Free the linked list of Window objects starting at the second argument. +*/ +SQLITE_PRIVATE void sqlite3WindowListDelete(sqlite3 *db, Window *p){ + while( p ){ + Window *pNext = p->pNextWin; + sqlite3WindowDelete(db, p); + p = pNext; + } +} + +/* +** The argument expression is an PRECEDING or FOLLOWING offset. The +** value should be a non-negative integer. If the value is not a +** constant, change it to NULL. The fact that it is then a non-negative +** integer will be caught later. But it is important not to leave +** variable values in the expression tree. +*/ +static Expr *sqlite3WindowOffsetExpr(Parse *pParse, Expr *pExpr){ + if( 0==sqlite3ExprIsConstant(pExpr) ){ + if( IN_RENAME_OBJECT ) sqlite3RenameExprUnmap(pParse, pExpr); + sqlite3ExprDelete(pParse->db, pExpr); + pExpr = sqlite3ExprAlloc(pParse->db, TK_NULL, 0, 0); + } + return pExpr; +} + +/* +** Allocate and return a new Window object describing a Window Definition. +*/ +SQLITE_PRIVATE Window *sqlite3WindowAlloc( + Parse *pParse, /* Parsing context */ + int eType, /* Frame type. TK_RANGE, TK_ROWS, TK_GROUPS, or 0 */ + int eStart, /* Start type: CURRENT, PRECEDING, FOLLOWING, UNBOUNDED */ + Expr *pStart, /* Start window size if TK_PRECEDING or FOLLOWING */ + int eEnd, /* End type: CURRENT, FOLLOWING, TK_UNBOUNDED, PRECEDING */ + Expr *pEnd, /* End window size if TK_FOLLOWING or PRECEDING */ + u8 eExclude /* EXCLUDE clause */ +){ + Window *pWin = 0; + int bImplicitFrame = 0; + + /* Parser assures the following: */ + assert( eType==0 || eType==TK_RANGE || eType==TK_ROWS || eType==TK_GROUPS ); + assert( eStart==TK_CURRENT || eStart==TK_PRECEDING + || eStart==TK_UNBOUNDED || eStart==TK_FOLLOWING ); + assert( eEnd==TK_CURRENT || eEnd==TK_FOLLOWING + || eEnd==TK_UNBOUNDED || eEnd==TK_PRECEDING ); + assert( (eStart==TK_PRECEDING || eStart==TK_FOLLOWING)==(pStart!=0) ); + assert( (eEnd==TK_FOLLOWING || eEnd==TK_PRECEDING)==(pEnd!=0) ); + + if( eType==0 ){ + bImplicitFrame = 1; + eType = TK_RANGE; + } + + /* Additionally, the + ** starting boundary type may not occur earlier in the following list than + ** the ending boundary type: + ** + ** UNBOUNDED PRECEDING + ** PRECEDING + ** CURRENT ROW + ** FOLLOWING + ** UNBOUNDED FOLLOWING + ** + ** The parser ensures that "UNBOUNDED PRECEDING" cannot be used as an ending + ** boundary, and than "UNBOUNDED FOLLOWING" cannot be used as a starting + ** frame boundary. + */ + if( (eStart==TK_CURRENT && eEnd==TK_PRECEDING) + || (eStart==TK_FOLLOWING && (eEnd==TK_PRECEDING || eEnd==TK_CURRENT)) + ){ + sqlite3ErrorMsg(pParse, "unsupported frame specification"); + goto windowAllocErr; + } + + pWin = (Window*)sqlite3DbMallocZero(pParse->db, sizeof(Window)); + if( pWin==0 ) goto windowAllocErr; + pWin->eFrmType = eType; + pWin->eStart = eStart; + pWin->eEnd = eEnd; + if( eExclude==0 && OptimizationDisabled(pParse->db, SQLITE_WindowFunc) ){ + eExclude = TK_NO; + } + pWin->eExclude = eExclude; + pWin->bImplicitFrame = bImplicitFrame; + pWin->pEnd = sqlite3WindowOffsetExpr(pParse, pEnd); + pWin->pStart = sqlite3WindowOffsetExpr(pParse, pStart); + return pWin; + +windowAllocErr: + sqlite3ExprDelete(pParse->db, pEnd); + sqlite3ExprDelete(pParse->db, pStart); + return 0; +} + +/* +** Attach PARTITION and ORDER BY clauses pPartition and pOrderBy to window +** pWin. Also, if parameter pBase is not NULL, set pWin->zBase to the +** equivalent nul-terminated string. +*/ +SQLITE_PRIVATE Window *sqlite3WindowAssemble( + Parse *pParse, + Window *pWin, + ExprList *pPartition, + ExprList *pOrderBy, + Token *pBase +){ + if( pWin ){ + pWin->pPartition = pPartition; + pWin->pOrderBy = pOrderBy; + if( pBase ){ + pWin->zBase = sqlite3DbStrNDup(pParse->db, pBase->z, pBase->n); + } + }else{ + sqlite3ExprListDelete(pParse->db, pPartition); + sqlite3ExprListDelete(pParse->db, pOrderBy); + } + return pWin; +} + +/* +** Window *pWin has just been created from a WINDOW clause. Tokne pBase +** is the base window. Earlier windows from the same WINDOW clause are +** stored in the linked list starting at pWin->pNextWin. This function +** either updates *pWin according to the base specification, or else +** leaves an error in pParse. +*/ +SQLITE_PRIVATE void sqlite3WindowChain(Parse *pParse, Window *pWin, Window *pList){ + if( pWin->zBase ){ + sqlite3 *db = pParse->db; + Window *pExist = windowFind(pParse, pList, pWin->zBase); + if( pExist ){ + const char *zErr = 0; + /* Check for errors */ + if( pWin->pPartition ){ + zErr = "PARTITION clause"; + }else if( pExist->pOrderBy && pWin->pOrderBy ){ + zErr = "ORDER BY clause"; + }else if( pExist->bImplicitFrame==0 ){ + zErr = "frame specification"; + } + if( zErr ){ + sqlite3ErrorMsg(pParse, + "cannot override %s of window: %s", zErr, pWin->zBase + ); + }else{ + pWin->pPartition = sqlite3ExprListDup(db, pExist->pPartition, 0); + if( pExist->pOrderBy ){ + assert( pWin->pOrderBy==0 ); + pWin->pOrderBy = sqlite3ExprListDup(db, pExist->pOrderBy, 0); + } + sqlite3DbFree(db, pWin->zBase); + pWin->zBase = 0; + } + } + } +} + +/* +** Attach window object pWin to expression p. +*/ +SQLITE_PRIVATE void sqlite3WindowAttach(Parse *pParse, Expr *p, Window *pWin){ + if( p ){ + assert( p->op==TK_FUNCTION ); + assert( pWin ); + p->y.pWin = pWin; + ExprSetProperty(p, EP_WinFunc); + pWin->pOwner = p; + if( (p->flags & EP_Distinct) && pWin->eFrmType!=TK_FILTER ){ + sqlite3ErrorMsg(pParse, + "DISTINCT is not supported for window functions" + ); + } + }else{ + sqlite3WindowDelete(pParse->db, pWin); + } +} + +/* +** Possibly link window pWin into the list at pSel->pWin (window functions +** to be processed as part of SELECT statement pSel). The window is linked +** in if either (a) there are no other windows already linked to this +** SELECT, or (b) the windows already linked use a compatible window frame. +*/ +SQLITE_PRIVATE void sqlite3WindowLink(Select *pSel, Window *pWin){ + if( 0==pSel->pWin + || 0==sqlite3WindowCompare(0, pSel->pWin, pWin, 0) + ){ + pWin->pNextWin = pSel->pWin; + if( pSel->pWin ){ + pSel->pWin->ppThis = &pWin->pNextWin; + } + pSel->pWin = pWin; + pWin->ppThis = &pSel->pWin; + } +} + +/* +** Return 0 if the two window objects are identical, or non-zero otherwise. +** Identical window objects can be processed in a single scan. +*/ +SQLITE_PRIVATE int sqlite3WindowCompare(Parse *pParse, Window *p1, Window *p2, int bFilter){ + if( p1->eFrmType!=p2->eFrmType ) return 1; + if( p1->eStart!=p2->eStart ) return 1; + if( p1->eEnd!=p2->eEnd ) return 1; + if( p1->eExclude!=p2->eExclude ) return 1; + if( sqlite3ExprCompare(pParse, p1->pStart, p2->pStart, -1) ) return 1; + if( sqlite3ExprCompare(pParse, p1->pEnd, p2->pEnd, -1) ) return 1; + if( sqlite3ExprListCompare(p1->pPartition, p2->pPartition, -1) ) return 1; + if( sqlite3ExprListCompare(p1->pOrderBy, p2->pOrderBy, -1) ) return 1; + if( bFilter ){ + if( sqlite3ExprCompare(pParse, p1->pFilter, p2->pFilter, -1) ) return 1; + } + return 0; +} + + +/* +** This is called by code in select.c before it calls sqlite3WhereBegin() +** to begin iterating through the sub-query results. It is used to allocate +** and initialize registers and cursors used by sqlite3WindowCodeStep(). +*/ +SQLITE_PRIVATE void sqlite3WindowCodeInit(Parse *pParse, Window *pMWin){ + Window *pWin; + Vdbe *v = sqlite3GetVdbe(pParse); + + /* Allocate registers to use for PARTITION BY values, if any. Initialize + ** said registers to NULL. */ + if( pMWin->pPartition ){ + int nExpr = pMWin->pPartition->nExpr; + pMWin->regPart = pParse->nMem+1; + pParse->nMem += nExpr; + sqlite3VdbeAddOp3(v, OP_Null, 0, pMWin->regPart, pMWin->regPart+nExpr-1); + } + + pMWin->regOne = ++pParse->nMem; + sqlite3VdbeAddOp2(v, OP_Integer, 1, pMWin->regOne); + + if( pMWin->eExclude ){ + pMWin->regStartRowid = ++pParse->nMem; + pMWin->regEndRowid = ++pParse->nMem; + pMWin->csrApp = pParse->nTab++; + sqlite3VdbeAddOp2(v, OP_Integer, 1, pMWin->regStartRowid); + sqlite3VdbeAddOp2(v, OP_Integer, 0, pMWin->regEndRowid); + sqlite3VdbeAddOp2(v, OP_OpenDup, pMWin->csrApp, pMWin->iEphCsr); + return; + } + + for(pWin=pMWin; pWin; pWin=pWin->pNextWin){ + FuncDef *p = pWin->pFunc; + if( (p->funcFlags & SQLITE_FUNC_MINMAX) && pWin->eStart!=TK_UNBOUNDED ){ + /* The inline versions of min() and max() require a single ephemeral + ** table and 3 registers. The registers are used as follows: + ** + ** regApp+0: slot to copy min()/max() argument to for MakeRecord + ** regApp+1: integer value used to ensure keys are unique + ** regApp+2: output of MakeRecord + */ + ExprList *pList = pWin->pOwner->x.pList; + KeyInfo *pKeyInfo = sqlite3KeyInfoFromExprList(pParse, pList, 0, 0); + pWin->csrApp = pParse->nTab++; + pWin->regApp = pParse->nMem+1; + pParse->nMem += 3; + if( pKeyInfo && pWin->pFunc->zName[1]=='i' ){ + assert( pKeyInfo->aSortFlags[0]==0 ); + pKeyInfo->aSortFlags[0] = KEYINFO_ORDER_DESC; + } + sqlite3VdbeAddOp2(v, OP_OpenEphemeral, pWin->csrApp, 2); + sqlite3VdbeAppendP4(v, pKeyInfo, P4_KEYINFO); + sqlite3VdbeAddOp2(v, OP_Integer, 0, pWin->regApp+1); + } + else if( p->zName==nth_valueName || p->zName==first_valueName ){ + /* Allocate two registers at pWin->regApp. These will be used to + ** store the start and end index of the current frame. */ + pWin->regApp = pParse->nMem+1; + pWin->csrApp = pParse->nTab++; + pParse->nMem += 2; + sqlite3VdbeAddOp2(v, OP_OpenDup, pWin->csrApp, pMWin->iEphCsr); + } + else if( p->zName==leadName || p->zName==lagName ){ + pWin->csrApp = pParse->nTab++; + sqlite3VdbeAddOp2(v, OP_OpenDup, pWin->csrApp, pMWin->iEphCsr); + } + } +} + +#define WINDOW_STARTING_INT 0 +#define WINDOW_ENDING_INT 1 +#define WINDOW_NTH_VALUE_INT 2 +#define WINDOW_STARTING_NUM 3 +#define WINDOW_ENDING_NUM 4 + +/* +** A "PRECEDING " (eCond==0) or "FOLLOWING " (eCond==1) or the +** value of the second argument to nth_value() (eCond==2) has just been +** evaluated and the result left in register reg. This function generates VM +** code to check that the value is a non-negative integer and throws an +** exception if it is not. +*/ +static void windowCheckValue(Parse *pParse, int reg, int eCond){ + static const char *azErr[] = { + "frame starting offset must be a non-negative integer", + "frame ending offset must be a non-negative integer", + "second argument to nth_value must be a positive integer", + "frame starting offset must be a non-negative number", + "frame ending offset must be a non-negative number", + }; + static int aOp[] = { OP_Ge, OP_Ge, OP_Gt, OP_Ge, OP_Ge }; + Vdbe *v = sqlite3GetVdbe(pParse); + int regZero = sqlite3GetTempReg(pParse); + assert( eCond>=0 && eCond=WINDOW_STARTING_NUM ){ + int regString = sqlite3GetTempReg(pParse); + sqlite3VdbeAddOp4(v, OP_String8, 0, regString, 0, "", P4_STATIC); + sqlite3VdbeAddOp3(v, OP_Ge, regString, sqlite3VdbeCurrentAddr(v)+2, reg); + sqlite3VdbeChangeP5(v, SQLITE_AFF_NUMERIC|SQLITE_JUMPIFNULL); + VdbeCoverage(v); + assert( eCond==3 || eCond==4 ); + VdbeCoverageIf(v, eCond==3); + VdbeCoverageIf(v, eCond==4); + }else{ + sqlite3VdbeAddOp2(v, OP_MustBeInt, reg, sqlite3VdbeCurrentAddr(v)+2); + VdbeCoverage(v); + assert( eCond==0 || eCond==1 || eCond==2 ); + VdbeCoverageIf(v, eCond==0); + VdbeCoverageIf(v, eCond==1); + VdbeCoverageIf(v, eCond==2); + } + sqlite3VdbeAddOp3(v, aOp[eCond], regZero, sqlite3VdbeCurrentAddr(v)+2, reg); + VdbeCoverageNeverNullIf(v, eCond==0); /* NULL case captured by */ + VdbeCoverageNeverNullIf(v, eCond==1); /* the OP_MustBeInt */ + VdbeCoverageNeverNullIf(v, eCond==2); + VdbeCoverageNeverNullIf(v, eCond==3); /* NULL case caught by */ + VdbeCoverageNeverNullIf(v, eCond==4); /* the OP_Ge */ + sqlite3MayAbort(pParse); + sqlite3VdbeAddOp2(v, OP_Halt, SQLITE_ERROR, OE_Abort); + sqlite3VdbeAppendP4(v, (void*)azErr[eCond], P4_STATIC); + sqlite3ReleaseTempReg(pParse, regZero); +} + +/* +** Return the number of arguments passed to the window-function associated +** with the object passed as the only argument to this function. +*/ +static int windowArgCount(Window *pWin){ + ExprList *pList = pWin->pOwner->x.pList; + return (pList ? pList->nExpr : 0); +} + +typedef struct WindowCodeArg WindowCodeArg; +typedef struct WindowCsrAndReg WindowCsrAndReg; + +/* +** See comments above struct WindowCodeArg. +*/ +struct WindowCsrAndReg { + int csr; /* Cursor number */ + int reg; /* First in array of peer values */ +}; + +/* +** A single instance of this structure is allocated on the stack by +** sqlite3WindowCodeStep() and a pointer to it passed to the various helper +** routines. This is to reduce the number of arguments required by each +** helper function. +** +** regArg: +** Each window function requires an accumulator register (just as an +** ordinary aggregate function does). This variable is set to the first +** in an array of accumulator registers - one for each window function +** in the WindowCodeArg.pMWin list. +** +** eDelete: +** The window functions implementation sometimes caches the input rows +** that it processes in a temporary table. If it is not zero, this +** variable indicates when rows may be removed from the temp table (in +** order to reduce memory requirements - it would always be safe just +** to leave them there). Possible values for eDelete are: +** +** WINDOW_RETURN_ROW: +** An input row can be discarded after it is returned to the caller. +** +** WINDOW_AGGINVERSE: +** An input row can be discarded after the window functions xInverse() +** callbacks have been invoked in it. +** +** WINDOW_AGGSTEP: +** An input row can be discarded after the window functions xStep() +** callbacks have been invoked in it. +** +** start,current,end +** Consider a window-frame similar to the following: +** +** (ORDER BY a, b GROUPS BETWEEN 2 PRECEDING AND 2 FOLLOWING) +** +** The windows functions implmentation caches the input rows in a temp +** table, sorted by "a, b" (it actually populates the cache lazily, and +** aggressively removes rows once they are no longer required, but that's +** a mere detail). It keeps three cursors open on the temp table. One +** (current) that points to the next row to return to the query engine +** once its window function values have been calculated. Another (end) +** points to the next row to call the xStep() method of each window function +** on (so that it is 2 groups ahead of current). And a third (start) that +** points to the next row to call the xInverse() method of each window +** function on. +** +** Each cursor (start, current and end) consists of a VDBE cursor +** (WindowCsrAndReg.csr) and an array of registers (starting at +** WindowCodeArg.reg) that always contains a copy of the peer values +** read from the corresponding cursor. +** +** Depending on the window-frame in question, all three cursors may not +** be required. In this case both WindowCodeArg.csr and reg are set to +** 0. +*/ +struct WindowCodeArg { + Parse *pParse; /* Parse context */ + Window *pMWin; /* First in list of functions being processed */ + Vdbe *pVdbe; /* VDBE object */ + int addrGosub; /* OP_Gosub to this address to return one row */ + int regGosub; /* Register used with OP_Gosub(addrGosub) */ + int regArg; /* First in array of accumulator registers */ + int eDelete; /* See above */ + + WindowCsrAndReg start; + WindowCsrAndReg current; + WindowCsrAndReg end; +}; + +/* +** Generate VM code to read the window frames peer values from cursor csr into +** an array of registers starting at reg. +*/ +static void windowReadPeerValues( + WindowCodeArg *p, + int csr, + int reg +){ + Window *pMWin = p->pMWin; + ExprList *pOrderBy = pMWin->pOrderBy; + if( pOrderBy ){ + Vdbe *v = sqlite3GetVdbe(p->pParse); + ExprList *pPart = pMWin->pPartition; + int iColOff = pMWin->nBufferCol + (pPart ? pPart->nExpr : 0); + int i; + for(i=0; inExpr; i++){ + sqlite3VdbeAddOp3(v, OP_Column, csr, iColOff+i, reg+i); + } + } +} + +/* +** Generate VM code to invoke either xStep() (if bInverse is 0) or +** xInverse (if bInverse is non-zero) for each window function in the +** linked list starting at pMWin. Or, for built-in window functions +** that do not use the standard function API, generate the required +** inline VM code. +** +** If argument csr is greater than or equal to 0, then argument reg is +** the first register in an array of registers guaranteed to be large +** enough to hold the array of arguments for each function. In this case +** the arguments are extracted from the current row of csr into the +** array of registers before invoking OP_AggStep or OP_AggInverse +** +** Or, if csr is less than zero, then the array of registers at reg is +** already populated with all columns from the current row of the sub-query. +** +** If argument regPartSize is non-zero, then it is a register containing the +** number of rows in the current partition. +*/ +static void windowAggStep( + WindowCodeArg *p, + Window *pMWin, /* Linked list of window functions */ + int csr, /* Read arguments from this cursor */ + int bInverse, /* True to invoke xInverse instead of xStep */ + int reg /* Array of registers */ +){ + Parse *pParse = p->pParse; + Vdbe *v = sqlite3GetVdbe(pParse); + Window *pWin; + for(pWin=pMWin; pWin; pWin=pWin->pNextWin){ + FuncDef *pFunc = pWin->pFunc; + int regArg; + int nArg = pWin->bExprArgs ? 0 : windowArgCount(pWin); + int i; + + assert( bInverse==0 || pWin->eStart!=TK_UNBOUNDED ); + + /* All OVER clauses in the same window function aggregate step must + ** be the same. */ + assert( pWin==pMWin || sqlite3WindowCompare(pParse,pWin,pMWin,0)==0 ); + + for(i=0; izName!=nth_valueName ){ + sqlite3VdbeAddOp3(v, OP_Column, csr, pWin->iArgCol+i, reg+i); + }else{ + sqlite3VdbeAddOp3(v, OP_Column, pMWin->iEphCsr, pWin->iArgCol+i, reg+i); + } + } + regArg = reg; + + if( pMWin->regStartRowid==0 + && (pFunc->funcFlags & SQLITE_FUNC_MINMAX) + && (pWin->eStart!=TK_UNBOUNDED) + ){ + int addrIsNull = sqlite3VdbeAddOp1(v, OP_IsNull, regArg); + VdbeCoverage(v); + if( bInverse==0 ){ + sqlite3VdbeAddOp2(v, OP_AddImm, pWin->regApp+1, 1); + sqlite3VdbeAddOp2(v, OP_SCopy, regArg, pWin->regApp); + sqlite3VdbeAddOp3(v, OP_MakeRecord, pWin->regApp, 2, pWin->regApp+2); + sqlite3VdbeAddOp2(v, OP_IdxInsert, pWin->csrApp, pWin->regApp+2); + }else{ + sqlite3VdbeAddOp4Int(v, OP_SeekGE, pWin->csrApp, 0, regArg, 1); + VdbeCoverageNeverTaken(v); + sqlite3VdbeAddOp1(v, OP_Delete, pWin->csrApp); + sqlite3VdbeJumpHere(v, sqlite3VdbeCurrentAddr(v)-2); + } + sqlite3VdbeJumpHere(v, addrIsNull); + }else if( pWin->regApp ){ + assert( pFunc->zName==nth_valueName + || pFunc->zName==first_valueName + ); + assert( bInverse==0 || bInverse==1 ); + sqlite3VdbeAddOp2(v, OP_AddImm, pWin->regApp+1-bInverse, 1); + }else if( pFunc->xSFunc!=noopStepFunc ){ + int addrIf = 0; + if( pWin->pFilter ){ + int regTmp; + assert( pWin->bExprArgs || !nArg ||nArg==pWin->pOwner->x.pList->nExpr ); + assert( pWin->bExprArgs || nArg ||pWin->pOwner->x.pList==0 ); + regTmp = sqlite3GetTempReg(pParse); + sqlite3VdbeAddOp3(v, OP_Column, csr, pWin->iArgCol+nArg,regTmp); + addrIf = sqlite3VdbeAddOp3(v, OP_IfNot, regTmp, 0, 1); + VdbeCoverage(v); + sqlite3ReleaseTempReg(pParse, regTmp); + } + + if( pWin->bExprArgs ){ + int iStart = sqlite3VdbeCurrentAddr(v); + VdbeOp *pOp, *pEnd; + + nArg = pWin->pOwner->x.pList->nExpr; + regArg = sqlite3GetTempRange(pParse, nArg); + sqlite3ExprCodeExprList(pParse, pWin->pOwner->x.pList, regArg, 0, 0); + + pEnd = sqlite3VdbeGetOp(v, -1); + for(pOp=sqlite3VdbeGetOp(v, iStart); pOp<=pEnd; pOp++){ + if( pOp->opcode==OP_Column && pOp->p1==pWin->iEphCsr ){ + pOp->p1 = csr; + } + } + } + if( pFunc->funcFlags & SQLITE_FUNC_NEEDCOLL ){ + CollSeq *pColl; + assert( nArg>0 ); + pColl = sqlite3ExprNNCollSeq(pParse, pWin->pOwner->x.pList->a[0].pExpr); + sqlite3VdbeAddOp4(v, OP_CollSeq, 0,0,0, (const char*)pColl, P4_COLLSEQ); + } + sqlite3VdbeAddOp3(v, bInverse? OP_AggInverse : OP_AggStep, + bInverse, regArg, pWin->regAccum); + sqlite3VdbeAppendP4(v, pFunc, P4_FUNCDEF); + sqlite3VdbeChangeP5(v, (u8)nArg); + if( pWin->bExprArgs ){ + sqlite3ReleaseTempRange(pParse, regArg, nArg); + } + if( addrIf ) sqlite3VdbeJumpHere(v, addrIf); + } + } +} + +/* +** Values that may be passed as the second argument to windowCodeOp(). +*/ +#define WINDOW_RETURN_ROW 1 +#define WINDOW_AGGINVERSE 2 +#define WINDOW_AGGSTEP 3 + +/* +** Generate VM code to invoke either xValue() (bFin==0) or xFinalize() +** (bFin==1) for each window function in the linked list starting at +** pMWin. Or, for built-in window-functions that do not use the standard +** API, generate the equivalent VM code. +*/ +static void windowAggFinal(WindowCodeArg *p, int bFin){ + Parse *pParse = p->pParse; + Window *pMWin = p->pMWin; + Vdbe *v = sqlite3GetVdbe(pParse); + Window *pWin; + + for(pWin=pMWin; pWin; pWin=pWin->pNextWin){ + if( pMWin->regStartRowid==0 + && (pWin->pFunc->funcFlags & SQLITE_FUNC_MINMAX) + && (pWin->eStart!=TK_UNBOUNDED) + ){ + sqlite3VdbeAddOp2(v, OP_Null, 0, pWin->regResult); + sqlite3VdbeAddOp1(v, OP_Last, pWin->csrApp); + VdbeCoverage(v); + sqlite3VdbeAddOp3(v, OP_Column, pWin->csrApp, 0, pWin->regResult); + sqlite3VdbeJumpHere(v, sqlite3VdbeCurrentAddr(v)-2); + }else if( pWin->regApp ){ + assert( pMWin->regStartRowid==0 ); + }else{ + int nArg = windowArgCount(pWin); + if( bFin ){ + sqlite3VdbeAddOp2(v, OP_AggFinal, pWin->regAccum, nArg); + sqlite3VdbeAppendP4(v, pWin->pFunc, P4_FUNCDEF); + sqlite3VdbeAddOp2(v, OP_Copy, pWin->regAccum, pWin->regResult); + sqlite3VdbeAddOp2(v, OP_Null, 0, pWin->regAccum); + }else{ + sqlite3VdbeAddOp3(v, OP_AggValue,pWin->regAccum,nArg,pWin->regResult); + sqlite3VdbeAppendP4(v, pWin->pFunc, P4_FUNCDEF); + } + } + } +} + +/* +** Generate code to calculate the current values of all window functions in the +** p->pMWin list by doing a full scan of the current window frame. Store the +** results in the Window.regResult registers, ready to return the upper +** layer. +*/ +static void windowFullScan(WindowCodeArg *p){ + Window *pWin; + Parse *pParse = p->pParse; + Window *pMWin = p->pMWin; + Vdbe *v = p->pVdbe; + + int regCRowid = 0; /* Current rowid value */ + int regCPeer = 0; /* Current peer values */ + int regRowid = 0; /* AggStep rowid value */ + int regPeer = 0; /* AggStep peer values */ + + int nPeer; + int lblNext; + int lblBrk; + int addrNext; + int csr; + + VdbeModuleComment((v, "windowFullScan begin")); + + assert( pMWin!=0 ); + csr = pMWin->csrApp; + nPeer = (pMWin->pOrderBy ? pMWin->pOrderBy->nExpr : 0); + + lblNext = sqlite3VdbeMakeLabel(pParse); + lblBrk = sqlite3VdbeMakeLabel(pParse); + + regCRowid = sqlite3GetTempReg(pParse); + regRowid = sqlite3GetTempReg(pParse); + if( nPeer ){ + regCPeer = sqlite3GetTempRange(pParse, nPeer); + regPeer = sqlite3GetTempRange(pParse, nPeer); + } + + sqlite3VdbeAddOp2(v, OP_Rowid, pMWin->iEphCsr, regCRowid); + windowReadPeerValues(p, pMWin->iEphCsr, regCPeer); + + for(pWin=pMWin; pWin; pWin=pWin->pNextWin){ + sqlite3VdbeAddOp2(v, OP_Null, 0, pWin->regAccum); + } + + sqlite3VdbeAddOp3(v, OP_SeekGE, csr, lblBrk, pMWin->regStartRowid); + VdbeCoverage(v); + addrNext = sqlite3VdbeCurrentAddr(v); + sqlite3VdbeAddOp2(v, OP_Rowid, csr, regRowid); + sqlite3VdbeAddOp3(v, OP_Gt, pMWin->regEndRowid, lblBrk, regRowid); + VdbeCoverageNeverNull(v); + + if( pMWin->eExclude==TK_CURRENT ){ + sqlite3VdbeAddOp3(v, OP_Eq, regCRowid, lblNext, regRowid); + VdbeCoverageNeverNull(v); + }else if( pMWin->eExclude!=TK_NO ){ + int addr; + int addrEq = 0; + KeyInfo *pKeyInfo = 0; + + if( pMWin->pOrderBy ){ + pKeyInfo = sqlite3KeyInfoFromExprList(pParse, pMWin->pOrderBy, 0, 0); + } + if( pMWin->eExclude==TK_TIES ){ + addrEq = sqlite3VdbeAddOp3(v, OP_Eq, regCRowid, 0, regRowid); + VdbeCoverageNeverNull(v); + } + if( pKeyInfo ){ + windowReadPeerValues(p, csr, regPeer); + sqlite3VdbeAddOp3(v, OP_Compare, regPeer, regCPeer, nPeer); + sqlite3VdbeAppendP4(v, (void*)pKeyInfo, P4_KEYINFO); + addr = sqlite3VdbeCurrentAddr(v)+1; + sqlite3VdbeAddOp3(v, OP_Jump, addr, lblNext, addr); + VdbeCoverageEqNe(v); + }else{ + sqlite3VdbeAddOp2(v, OP_Goto, 0, lblNext); + } + if( addrEq ) sqlite3VdbeJumpHere(v, addrEq); + } + + windowAggStep(p, pMWin, csr, 0, p->regArg); + + sqlite3VdbeResolveLabel(v, lblNext); + sqlite3VdbeAddOp2(v, OP_Next, csr, addrNext); + VdbeCoverage(v); + sqlite3VdbeJumpHere(v, addrNext-1); + sqlite3VdbeJumpHere(v, addrNext+1); + sqlite3ReleaseTempReg(pParse, regRowid); + sqlite3ReleaseTempReg(pParse, regCRowid); + if( nPeer ){ + sqlite3ReleaseTempRange(pParse, regPeer, nPeer); + sqlite3ReleaseTempRange(pParse, regCPeer, nPeer); + } + + windowAggFinal(p, 1); + VdbeModuleComment((v, "windowFullScan end")); +} + +/* +** Invoke the sub-routine at regGosub (generated by code in select.c) to +** return the current row of Window.iEphCsr. If all window functions are +** aggregate window functions that use the standard API, a single +** OP_Gosub instruction is all that this routine generates. Extra VM code +** for per-row processing is only generated for the following built-in window +** functions: +** +** nth_value() +** first_value() +** lag() +** lead() +*/ +static void windowReturnOneRow(WindowCodeArg *p){ + Window *pMWin = p->pMWin; + Vdbe *v = p->pVdbe; + + if( pMWin->regStartRowid ){ + windowFullScan(p); + }else{ + Parse *pParse = p->pParse; + Window *pWin; + + for(pWin=pMWin; pWin; pWin=pWin->pNextWin){ + FuncDef *pFunc = pWin->pFunc; + if( pFunc->zName==nth_valueName + || pFunc->zName==first_valueName + ){ + int csr = pWin->csrApp; + int lbl = sqlite3VdbeMakeLabel(pParse); + int tmpReg = sqlite3GetTempReg(pParse); + sqlite3VdbeAddOp2(v, OP_Null, 0, pWin->regResult); + + if( pFunc->zName==nth_valueName ){ + sqlite3VdbeAddOp3(v, OP_Column,pMWin->iEphCsr,pWin->iArgCol+1,tmpReg); + windowCheckValue(pParse, tmpReg, 2); + }else{ + sqlite3VdbeAddOp2(v, OP_Integer, 1, tmpReg); + } + sqlite3VdbeAddOp3(v, OP_Add, tmpReg, pWin->regApp, tmpReg); + sqlite3VdbeAddOp3(v, OP_Gt, pWin->regApp+1, lbl, tmpReg); + VdbeCoverageNeverNull(v); + sqlite3VdbeAddOp3(v, OP_SeekRowid, csr, 0, tmpReg); + VdbeCoverageNeverTaken(v); + sqlite3VdbeAddOp3(v, OP_Column, csr, pWin->iArgCol, pWin->regResult); + sqlite3VdbeResolveLabel(v, lbl); + sqlite3ReleaseTempReg(pParse, tmpReg); + } + else if( pFunc->zName==leadName || pFunc->zName==lagName ){ + int nArg = pWin->pOwner->x.pList->nExpr; + int csr = pWin->csrApp; + int lbl = sqlite3VdbeMakeLabel(pParse); + int tmpReg = sqlite3GetTempReg(pParse); + int iEph = pMWin->iEphCsr; + + if( nArg<3 ){ + sqlite3VdbeAddOp2(v, OP_Null, 0, pWin->regResult); + }else{ + sqlite3VdbeAddOp3(v, OP_Column, iEph,pWin->iArgCol+2,pWin->regResult); + } + sqlite3VdbeAddOp2(v, OP_Rowid, iEph, tmpReg); + if( nArg<2 ){ + int val = (pFunc->zName==leadName ? 1 : -1); + sqlite3VdbeAddOp2(v, OP_AddImm, tmpReg, val); + }else{ + int op = (pFunc->zName==leadName ? OP_Add : OP_Subtract); + int tmpReg2 = sqlite3GetTempReg(pParse); + sqlite3VdbeAddOp3(v, OP_Column, iEph, pWin->iArgCol+1, tmpReg2); + sqlite3VdbeAddOp3(v, op, tmpReg2, tmpReg, tmpReg); + sqlite3ReleaseTempReg(pParse, tmpReg2); + } + + sqlite3VdbeAddOp3(v, OP_SeekRowid, csr, lbl, tmpReg); + VdbeCoverage(v); + sqlite3VdbeAddOp3(v, OP_Column, csr, pWin->iArgCol, pWin->regResult); + sqlite3VdbeResolveLabel(v, lbl); + sqlite3ReleaseTempReg(pParse, tmpReg); + } + } + } + sqlite3VdbeAddOp2(v, OP_Gosub, p->regGosub, p->addrGosub); +} + +/* +** Generate code to set the accumulator register for each window function +** in the linked list passed as the second argument to NULL. And perform +** any equivalent initialization required by any built-in window functions +** in the list. +*/ +static int windowInitAccum(Parse *pParse, Window *pMWin){ + Vdbe *v = sqlite3GetVdbe(pParse); + int regArg; + int nArg = 0; + Window *pWin; + for(pWin=pMWin; pWin; pWin=pWin->pNextWin){ + FuncDef *pFunc = pWin->pFunc; + sqlite3VdbeAddOp2(v, OP_Null, 0, pWin->regAccum); + nArg = MAX(nArg, windowArgCount(pWin)); + if( pMWin->regStartRowid==0 ){ + if( pFunc->zName==nth_valueName || pFunc->zName==first_valueName ){ + sqlite3VdbeAddOp2(v, OP_Integer, 0, pWin->regApp); + sqlite3VdbeAddOp2(v, OP_Integer, 0, pWin->regApp+1); + } + + if( (pFunc->funcFlags & SQLITE_FUNC_MINMAX) && pWin->csrApp ){ + assert( pWin->eStart!=TK_UNBOUNDED ); + sqlite3VdbeAddOp1(v, OP_ResetSorter, pWin->csrApp); + sqlite3VdbeAddOp2(v, OP_Integer, 0, pWin->regApp+1); + } + } + } + regArg = pParse->nMem+1; + pParse->nMem += nArg; + return regArg; +} + +/* +** Return true if the current frame should be cached in the ephemeral table, +** even if there are no xInverse() calls required. +*/ +static int windowCacheFrame(Window *pMWin){ + Window *pWin; + if( pMWin->regStartRowid ) return 1; + for(pWin=pMWin; pWin; pWin=pWin->pNextWin){ + FuncDef *pFunc = pWin->pFunc; + if( (pFunc->zName==nth_valueName) + || (pFunc->zName==first_valueName) + || (pFunc->zName==leadName) + || (pFunc->zName==lagName) + ){ + return 1; + } + } + return 0; +} + +/* +** regOld and regNew are each the first register in an array of size +** pOrderBy->nExpr. This function generates code to compare the two +** arrays of registers using the collation sequences and other comparison +** parameters specified by pOrderBy. +** +** If the two arrays are not equal, the contents of regNew is copied to +** regOld and control falls through. Otherwise, if the contents of the arrays +** are equal, an OP_Goto is executed. The address of the OP_Goto is returned. +*/ +static void windowIfNewPeer( + Parse *pParse, + ExprList *pOrderBy, + int regNew, /* First in array of new values */ + int regOld, /* First in array of old values */ + int addr /* Jump here */ +){ + Vdbe *v = sqlite3GetVdbe(pParse); + if( pOrderBy ){ + int nVal = pOrderBy->nExpr; + KeyInfo *pKeyInfo = sqlite3KeyInfoFromExprList(pParse, pOrderBy, 0, 0); + sqlite3VdbeAddOp3(v, OP_Compare, regOld, regNew, nVal); + sqlite3VdbeAppendP4(v, (void*)pKeyInfo, P4_KEYINFO); + sqlite3VdbeAddOp3(v, OP_Jump, + sqlite3VdbeCurrentAddr(v)+1, addr, sqlite3VdbeCurrentAddr(v)+1 + ); + VdbeCoverageEqNe(v); + sqlite3VdbeAddOp3(v, OP_Copy, regNew, regOld, nVal-1); + }else{ + sqlite3VdbeAddOp2(v, OP_Goto, 0, addr); + } +} + +/* +** This function is called as part of generating VM programs for RANGE +** offset PRECEDING/FOLLOWING frame boundaries. Assuming "ASC" order for +** the ORDER BY term in the window, and that argument op is OP_Ge, it generates +** code equivalent to: +** +** if( csr1.peerVal + regVal >= csr2.peerVal ) goto lbl; +** +** The value of parameter op may also be OP_Gt or OP_Le. In these cases the +** operator in the above pseudo-code is replaced with ">" or "<=", respectively. +** +** If the sort-order for the ORDER BY term in the window is DESC, then the +** comparison is reversed. Instead of adding regVal to csr1.peerVal, it is +** subtracted. And the comparison operator is inverted to - ">=" becomes "<=", +** ">" becomes "<", and so on. So, with DESC sort order, if the argument op +** is OP_Ge, the generated code is equivalent to: +** +** if( csr1.peerVal - regVal <= csr2.peerVal ) goto lbl; +** +** A special type of arithmetic is used such that if csr1.peerVal is not +** a numeric type (real or integer), then the result of the addition addition +** or subtraction is a a copy of csr1.peerVal. +*/ +static void windowCodeRangeTest( + WindowCodeArg *p, + int op, /* OP_Ge, OP_Gt, or OP_Le */ + int csr1, /* Cursor number for cursor 1 */ + int regVal, /* Register containing non-negative number */ + int csr2, /* Cursor number for cursor 2 */ + int lbl /* Jump destination if condition is true */ +){ + Parse *pParse = p->pParse; + Vdbe *v = sqlite3GetVdbe(pParse); + ExprList *pOrderBy = p->pMWin->pOrderBy; /* ORDER BY clause for window */ + int reg1 = sqlite3GetTempReg(pParse); /* Reg. for csr1.peerVal+regVal */ + int reg2 = sqlite3GetTempReg(pParse); /* Reg. for csr2.peerVal */ + int regString = ++pParse->nMem; /* Reg. for constant value '' */ + int arith = OP_Add; /* OP_Add or OP_Subtract */ + int addrGe; /* Jump destination */ + + assert( op==OP_Ge || op==OP_Gt || op==OP_Le ); + assert( pOrderBy && pOrderBy->nExpr==1 ); + if( pOrderBy->a[0].sortFlags & KEYINFO_ORDER_DESC ){ + switch( op ){ + case OP_Ge: op = OP_Le; break; + case OP_Gt: op = OP_Lt; break; + default: assert( op==OP_Le ); op = OP_Ge; break; + } + arith = OP_Subtract; + } + + /* Read the peer-value from each cursor into a register */ + windowReadPeerValues(p, csr1, reg1); + windowReadPeerValues(p, csr2, reg2); + + VdbeModuleComment((v, "CodeRangeTest: if( R%d %s R%d %s R%d ) goto lbl", + reg1, (arith==OP_Add ? "+" : "-"), regVal, + ((op==OP_Ge) ? ">=" : (op==OP_Le) ? "<=" : (op==OP_Gt) ? ">" : "<"), reg2 + )); + + /* Register reg1 currently contains csr1.peerVal (the peer-value from csr1). + ** This block adds (or subtracts for DESC) the numeric value in regVal + ** from it. Or, if reg1 is not numeric (it is a NULL, a text value or a blob), + ** then leave reg1 as it is. In pseudo-code, this is implemented as: + ** + ** if( reg1>='' ) goto addrGe; + ** reg1 = reg1 +/- regVal + ** addrGe: + ** + ** Since all strings and blobs are greater-than-or-equal-to an empty string, + ** the add/subtract is skipped for these, as required. If reg1 is a NULL, + ** then the arithmetic is performed, but since adding or subtracting from + ** NULL is always NULL anyway, this case is handled as required too. */ + sqlite3VdbeAddOp4(v, OP_String8, 0, regString, 0, "", P4_STATIC); + addrGe = sqlite3VdbeAddOp3(v, OP_Ge, regString, 0, reg1); + VdbeCoverage(v); + sqlite3VdbeAddOp3(v, arith, regVal, reg1, reg1); + sqlite3VdbeJumpHere(v, addrGe); + + /* If the BIGNULL flag is set for the ORDER BY, then it is required to + ** consider NULL values to be larger than all other values, instead of + ** the usual smaller. The VDBE opcodes OP_Ge and so on do not handle this + ** (and adding that capability causes a performance regression), so + ** instead if the BIGNULL flag is set then cases where either reg1 or + ** reg2 are NULL are handled separately in the following block. The code + ** generated is equivalent to: + ** + ** if( reg1 IS NULL ){ + ** if( op==OP_Ge ) goto lbl; + ** if( op==OP_Gt && reg2 IS NOT NULL ) goto lbl; + ** if( op==OP_Le && reg2 IS NULL ) goto lbl; + ** }else if( reg2 IS NULL ){ + ** if( op==OP_Le ) goto lbl; + ** } + ** + ** Additionally, if either reg1 or reg2 are NULL but the jump to lbl is + ** not taken, control jumps over the comparison operator coded below this + ** block. */ + if( pOrderBy->a[0].sortFlags & KEYINFO_ORDER_BIGNULL ){ + /* This block runs if reg1 contains a NULL. */ + int addr = sqlite3VdbeAddOp1(v, OP_NotNull, reg1); VdbeCoverage(v); + switch( op ){ + case OP_Ge: + sqlite3VdbeAddOp2(v, OP_Goto, 0, lbl); + break; + case OP_Gt: + sqlite3VdbeAddOp2(v, OP_NotNull, reg2, lbl); + VdbeCoverage(v); + break; + case OP_Le: + sqlite3VdbeAddOp2(v, OP_IsNull, reg2, lbl); + VdbeCoverage(v); + break; + default: assert( op==OP_Lt ); /* no-op */ break; + } + sqlite3VdbeAddOp2(v, OP_Goto, 0, sqlite3VdbeCurrentAddr(v)+3); + + /* This block runs if reg1 is not NULL, but reg2 is. */ + sqlite3VdbeJumpHere(v, addr); + sqlite3VdbeAddOp2(v, OP_IsNull, reg2, lbl); VdbeCoverage(v); + if( op==OP_Gt || op==OP_Ge ){ + sqlite3VdbeChangeP2(v, -1, sqlite3VdbeCurrentAddr(v)+1); + } + } + + /* Compare registers reg2 and reg1, taking the jump if required. Note that + ** control skips over this test if the BIGNULL flag is set and either + ** reg1 or reg2 contain a NULL value. */ + sqlite3VdbeAddOp3(v, op, reg2, lbl, reg1); VdbeCoverage(v); + sqlite3VdbeChangeP5(v, SQLITE_NULLEQ); + + assert( op==OP_Ge || op==OP_Gt || op==OP_Lt || op==OP_Le ); + testcase(op==OP_Ge); VdbeCoverageIf(v, op==OP_Ge); + testcase(op==OP_Lt); VdbeCoverageIf(v, op==OP_Lt); + testcase(op==OP_Le); VdbeCoverageIf(v, op==OP_Le); + testcase(op==OP_Gt); VdbeCoverageIf(v, op==OP_Gt); + sqlite3ReleaseTempReg(pParse, reg1); + sqlite3ReleaseTempReg(pParse, reg2); + + VdbeModuleComment((v, "CodeRangeTest: end")); +} + +/* +** Helper function for sqlite3WindowCodeStep(). Each call to this function +** generates VM code for a single RETURN_ROW, AGGSTEP or AGGINVERSE +** operation. Refer to the header comment for sqlite3WindowCodeStep() for +** details. +*/ +static int windowCodeOp( + WindowCodeArg *p, /* Context object */ + int op, /* WINDOW_RETURN_ROW, AGGSTEP or AGGINVERSE */ + int regCountdown, /* Register for OP_IfPos countdown */ + int jumpOnEof /* Jump here if stepped cursor reaches EOF */ +){ + int csr, reg; + Parse *pParse = p->pParse; + Window *pMWin = p->pMWin; + int ret = 0; + Vdbe *v = p->pVdbe; + int addrContinue = 0; + int bPeer = (pMWin->eFrmType!=TK_ROWS); + + int lblDone = sqlite3VdbeMakeLabel(pParse); + int addrNextRange = 0; + + /* Special case - WINDOW_AGGINVERSE is always a no-op if the frame + ** starts with UNBOUNDED PRECEDING. */ + if( op==WINDOW_AGGINVERSE && pMWin->eStart==TK_UNBOUNDED ){ + assert( regCountdown==0 && jumpOnEof==0 ); + return 0; + } + + if( regCountdown>0 ){ + if( pMWin->eFrmType==TK_RANGE ){ + addrNextRange = sqlite3VdbeCurrentAddr(v); + assert( op==WINDOW_AGGINVERSE || op==WINDOW_AGGSTEP ); + if( op==WINDOW_AGGINVERSE ){ + if( pMWin->eStart==TK_FOLLOWING ){ + windowCodeRangeTest( + p, OP_Le, p->current.csr, regCountdown, p->start.csr, lblDone + ); + }else{ + windowCodeRangeTest( + p, OP_Ge, p->start.csr, regCountdown, p->current.csr, lblDone + ); + } + }else{ + windowCodeRangeTest( + p, OP_Gt, p->end.csr, regCountdown, p->current.csr, lblDone + ); + } + }else{ + sqlite3VdbeAddOp3(v, OP_IfPos, regCountdown, lblDone, 1); + VdbeCoverage(v); + } + } + + if( op==WINDOW_RETURN_ROW && pMWin->regStartRowid==0 ){ + windowAggFinal(p, 0); + } + addrContinue = sqlite3VdbeCurrentAddr(v); + + /* If this is a (RANGE BETWEEN a FOLLOWING AND b FOLLOWING) or + ** (RANGE BETWEEN b PRECEDING AND a PRECEDING) frame, ensure the + ** start cursor does not advance past the end cursor within the + ** temporary table. It otherwise might, if (a>b). */ + if( pMWin->eStart==pMWin->eEnd && regCountdown + && pMWin->eFrmType==TK_RANGE && op==WINDOW_AGGINVERSE + ){ + int regRowid1 = sqlite3GetTempReg(pParse); + int regRowid2 = sqlite3GetTempReg(pParse); + sqlite3VdbeAddOp2(v, OP_Rowid, p->start.csr, regRowid1); + sqlite3VdbeAddOp2(v, OP_Rowid, p->end.csr, regRowid2); + sqlite3VdbeAddOp3(v, OP_Ge, regRowid2, lblDone, regRowid1); + VdbeCoverage(v); + sqlite3ReleaseTempReg(pParse, regRowid1); + sqlite3ReleaseTempReg(pParse, regRowid2); + assert( pMWin->eStart==TK_PRECEDING || pMWin->eStart==TK_FOLLOWING ); + } + + switch( op ){ + case WINDOW_RETURN_ROW: + csr = p->current.csr; + reg = p->current.reg; + windowReturnOneRow(p); + break; + + case WINDOW_AGGINVERSE: + csr = p->start.csr; + reg = p->start.reg; + if( pMWin->regStartRowid ){ + assert( pMWin->regEndRowid ); + sqlite3VdbeAddOp2(v, OP_AddImm, pMWin->regStartRowid, 1); + }else{ + windowAggStep(p, pMWin, csr, 1, p->regArg); + } + break; + + default: + assert( op==WINDOW_AGGSTEP ); + csr = p->end.csr; + reg = p->end.reg; + if( pMWin->regStartRowid ){ + assert( pMWin->regEndRowid ); + sqlite3VdbeAddOp2(v, OP_AddImm, pMWin->regEndRowid, 1); + }else{ + windowAggStep(p, pMWin, csr, 0, p->regArg); + } + break; + } + + if( op==p->eDelete ){ + sqlite3VdbeAddOp1(v, OP_Delete, csr); + sqlite3VdbeChangeP5(v, OPFLAG_SAVEPOSITION); + } + + if( jumpOnEof ){ + sqlite3VdbeAddOp2(v, OP_Next, csr, sqlite3VdbeCurrentAddr(v)+2); + VdbeCoverage(v); + ret = sqlite3VdbeAddOp0(v, OP_Goto); + }else{ + sqlite3VdbeAddOp2(v, OP_Next, csr, sqlite3VdbeCurrentAddr(v)+1+bPeer); + VdbeCoverage(v); + if( bPeer ){ + sqlite3VdbeAddOp2(v, OP_Goto, 0, lblDone); + } + } + + if( bPeer ){ + int nReg = (pMWin->pOrderBy ? pMWin->pOrderBy->nExpr : 0); + int regTmp = (nReg ? sqlite3GetTempRange(pParse, nReg) : 0); + windowReadPeerValues(p, csr, regTmp); + windowIfNewPeer(pParse, pMWin->pOrderBy, regTmp, reg, addrContinue); + sqlite3ReleaseTempRange(pParse, regTmp, nReg); + } + + if( addrNextRange ){ + sqlite3VdbeAddOp2(v, OP_Goto, 0, addrNextRange); + } + sqlite3VdbeResolveLabel(v, lblDone); + return ret; +} + + +/* +** Allocate and return a duplicate of the Window object indicated by the +** third argument. Set the Window.pOwner field of the new object to +** pOwner. +*/ +SQLITE_PRIVATE Window *sqlite3WindowDup(sqlite3 *db, Expr *pOwner, Window *p){ + Window *pNew = 0; + if( ALWAYS(p) ){ + pNew = sqlite3DbMallocZero(db, sizeof(Window)); + if( pNew ){ + pNew->zName = sqlite3DbStrDup(db, p->zName); + pNew->zBase = sqlite3DbStrDup(db, p->zBase); + pNew->pFilter = sqlite3ExprDup(db, p->pFilter, 0); + pNew->pFunc = p->pFunc; + pNew->pPartition = sqlite3ExprListDup(db, p->pPartition, 0); + pNew->pOrderBy = sqlite3ExprListDup(db, p->pOrderBy, 0); + pNew->eFrmType = p->eFrmType; + pNew->eEnd = p->eEnd; + pNew->eStart = p->eStart; + pNew->eExclude = p->eExclude; + pNew->regResult = p->regResult; + pNew->pStart = sqlite3ExprDup(db, p->pStart, 0); + pNew->pEnd = sqlite3ExprDup(db, p->pEnd, 0); + pNew->pOwner = pOwner; + pNew->bImplicitFrame = p->bImplicitFrame; + } + } + return pNew; +} + +/* +** Return a copy of the linked list of Window objects passed as the +** second argument. +*/ +SQLITE_PRIVATE Window *sqlite3WindowListDup(sqlite3 *db, Window *p){ + Window *pWin; + Window *pRet = 0; + Window **pp = &pRet; + + for(pWin=p; pWin; pWin=pWin->pNextWin){ + *pp = sqlite3WindowDup(db, 0, pWin); + if( *pp==0 ) break; + pp = &((*pp)->pNextWin); + } + + return pRet; +} + +/* +** Return true if it can be determined at compile time that expression +** pExpr evaluates to a value that, when cast to an integer, is greater +** than zero. False otherwise. +** +** If an OOM error occurs, this function sets the Parse.db.mallocFailed +** flag and returns zero. +*/ +static int windowExprGtZero(Parse *pParse, Expr *pExpr){ + int ret = 0; + sqlite3 *db = pParse->db; + sqlite3_value *pVal = 0; + sqlite3ValueFromExpr(db, pExpr, db->enc, SQLITE_AFF_NUMERIC, &pVal); + if( pVal && sqlite3_value_int(pVal)>0 ){ + ret = 1; + } + sqlite3ValueFree(pVal); + return ret; +} + +/* +** sqlite3WhereBegin() has already been called for the SELECT statement +** passed as the second argument when this function is invoked. It generates +** code to populate the Window.regResult register for each window function +** and invoke the sub-routine at instruction addrGosub once for each row. +** sqlite3WhereEnd() is always called before returning. +** +** This function handles several different types of window frames, which +** require slightly different processing. The following pseudo code is +** used to implement window frames of the form: +** +** ROWS BETWEEN PRECEDING AND FOLLOWING +** +** Other window frame types use variants of the following: +** +** ... loop started by sqlite3WhereBegin() ... +** if( new partition ){ +** Gosub flush +** } +** Insert new row into eph table. +** +** if( first row of partition ){ +** // Rewind three cursors, all open on the eph table. +** Rewind(csrEnd); +** Rewind(csrStart); +** Rewind(csrCurrent); +** +** regEnd = // FOLLOWING expression +** regStart = // PRECEDING expression +** }else{ +** // First time this branch is taken, the eph table contains two +** // rows. The first row in the partition, which all three cursors +** // currently point to, and the following row. +** AGGSTEP +** if( (regEnd--)<=0 ){ +** RETURN_ROW +** if( (regStart--)<=0 ){ +** AGGINVERSE +** } +** } +** } +** } +** flush: +** AGGSTEP +** while( 1 ){ +** RETURN ROW +** if( csrCurrent is EOF ) break; +** if( (regStart--)<=0 ){ +** AggInverse(csrStart) +** Next(csrStart) +** } +** } +** +** The pseudo-code above uses the following shorthand: +** +** AGGSTEP: invoke the aggregate xStep() function for each window function +** with arguments read from the current row of cursor csrEnd, then +** step cursor csrEnd forward one row (i.e. sqlite3BtreeNext()). +** +** RETURN_ROW: return a row to the caller based on the contents of the +** current row of csrCurrent and the current state of all +** aggregates. Then step cursor csrCurrent forward one row. +** +** AGGINVERSE: invoke the aggregate xInverse() function for each window +** functions with arguments read from the current row of cursor +** csrStart. Then step csrStart forward one row. +** +** There are two other ROWS window frames that are handled significantly +** differently from the above - "BETWEEN PRECEDING AND PRECEDING" +** and "BETWEEN FOLLOWING AND FOLLOWING". These are special +** cases because they change the order in which the three cursors (csrStart, +** csrCurrent and csrEnd) iterate through the ephemeral table. Cases that +** use UNBOUNDED or CURRENT ROW are much simpler variations on one of these +** three. +** +** ROWS BETWEEN PRECEDING AND PRECEDING +** +** ... loop started by sqlite3WhereBegin() ... +** if( new partition ){ +** Gosub flush +** } +** Insert new row into eph table. +** if( first row of partition ){ +** Rewind(csrEnd) ; Rewind(csrStart) ; Rewind(csrCurrent) +** regEnd = +** regStart = +** }else{ +** if( (regEnd--)<=0 ){ +** AGGSTEP +** } +** RETURN_ROW +** if( (regStart--)<=0 ){ +** AGGINVERSE +** } +** } +** } +** flush: +** if( (regEnd--)<=0 ){ +** AGGSTEP +** } +** RETURN_ROW +** +** +** ROWS BETWEEN FOLLOWING AND FOLLOWING +** +** ... loop started by sqlite3WhereBegin() ... +** if( new partition ){ +** Gosub flush +** } +** Insert new row into eph table. +** if( first row of partition ){ +** Rewind(csrEnd) ; Rewind(csrStart) ; Rewind(csrCurrent) +** regEnd = +** regStart = regEnd - +** }else{ +** AGGSTEP +** if( (regEnd--)<=0 ){ +** RETURN_ROW +** } +** if( (regStart--)<=0 ){ +** AGGINVERSE +** } +** } +** } +** flush: +** AGGSTEP +** while( 1 ){ +** if( (regEnd--)<=0 ){ +** RETURN_ROW +** if( eof ) break; +** } +** if( (regStart--)<=0 ){ +** AGGINVERSE +** if( eof ) break +** } +** } +** while( !eof csrCurrent ){ +** RETURN_ROW +** } +** +** For the most part, the patterns above are adapted to support UNBOUNDED by +** assuming that it is equivalent to "infinity PRECEDING/FOLLOWING" and +** CURRENT ROW by assuming that it is equivilent to "0 PRECEDING/FOLLOWING". +** This is optimized of course - branches that will never be taken and +** conditions that are always true are omitted from the VM code. The only +** exceptional case is: +** +** ROWS BETWEEN FOLLOWING AND UNBOUNDED FOLLOWING +** +** ... loop started by sqlite3WhereBegin() ... +** if( new partition ){ +** Gosub flush +** } +** Insert new row into eph table. +** if( first row of partition ){ +** Rewind(csrEnd) ; Rewind(csrStart) ; Rewind(csrCurrent) +** regStart = +** }else{ +** AGGSTEP +** } +** } +** flush: +** AGGSTEP +** while( 1 ){ +** if( (regStart--)<=0 ){ +** AGGINVERSE +** if( eof ) break +** } +** RETURN_ROW +** } +** while( !eof csrCurrent ){ +** RETURN_ROW +** } +** +** Also requiring special handling are the cases: +** +** ROWS BETWEEN PRECEDING AND PRECEDING +** ROWS BETWEEN FOLLOWING AND FOLLOWING +** +** when (expr1 < expr2). This is detected at runtime, not by this function. +** To handle this case, the pseudo-code programs depicted above are modified +** slightly to be: +** +** ... loop started by sqlite3WhereBegin() ... +** if( new partition ){ +** Gosub flush +** } +** Insert new row into eph table. +** if( first row of partition ){ +** Rewind(csrEnd) ; Rewind(csrStart) ; Rewind(csrCurrent) +** regEnd = +** regStart = +** if( regEnd < regStart ){ +** RETURN_ROW +** delete eph table contents +** continue +** } +** ... +** +** The new "continue" statement in the above jumps to the next iteration +** of the outer loop - the one started by sqlite3WhereBegin(). +** +** The various GROUPS cases are implemented using the same patterns as +** ROWS. The VM code is modified slightly so that: +** +** 1. The else branch in the main loop is only taken if the row just +** added to the ephemeral table is the start of a new group. In +** other words, it becomes: +** +** ... loop started by sqlite3WhereBegin() ... +** if( new partition ){ +** Gosub flush +** } +** Insert new row into eph table. +** if( first row of partition ){ +** Rewind(csrEnd) ; Rewind(csrStart) ; Rewind(csrCurrent) +** regEnd = +** regStart = +** }else if( new group ){ +** ... +** } +** } +** +** 2. Instead of processing a single row, each RETURN_ROW, AGGSTEP or +** AGGINVERSE step processes the current row of the relevant cursor and +** all subsequent rows belonging to the same group. +** +** RANGE window frames are a little different again. As for GROUPS, the +** main loop runs once per group only. And RETURN_ROW, AGGSTEP and AGGINVERSE +** deal in groups instead of rows. As for ROWS and GROUPS, there are three +** basic cases: +** +** RANGE BETWEEN PRECEDING AND FOLLOWING +** +** ... loop started by sqlite3WhereBegin() ... +** if( new partition ){ +** Gosub flush +** } +** Insert new row into eph table. +** if( first row of partition ){ +** Rewind(csrEnd) ; Rewind(csrStart) ; Rewind(csrCurrent) +** regEnd = +** regStart = +** }else{ +** AGGSTEP +** while( (csrCurrent.key + regEnd) < csrEnd.key ){ +** RETURN_ROW +** while( csrStart.key + regStart) < csrCurrent.key ){ +** AGGINVERSE +** } +** } +** } +** } +** flush: +** AGGSTEP +** while( 1 ){ +** RETURN ROW +** if( csrCurrent is EOF ) break; +** while( csrStart.key + regStart) < csrCurrent.key ){ +** AGGINVERSE +** } +** } +** } +** +** In the above notation, "csr.key" means the current value of the ORDER BY +** expression (there is only ever 1 for a RANGE that uses an FOLLOWING +** or PRECEDING AND PRECEDING +** +** ... loop started by sqlite3WhereBegin() ... +** if( new partition ){ +** Gosub flush +** } +** Insert new row into eph table. +** if( first row of partition ){ +** Rewind(csrEnd) ; Rewind(csrStart) ; Rewind(csrCurrent) +** regEnd = +** regStart = +** }else{ +** while( (csrEnd.key + regEnd) <= csrCurrent.key ){ +** AGGSTEP +** } +** while( (csrStart.key + regStart) < csrCurrent.key ){ +** AGGINVERSE +** } +** RETURN_ROW +** } +** } +** flush: +** while( (csrEnd.key + regEnd) <= csrCurrent.key ){ +** AGGSTEP +** } +** while( (csrStart.key + regStart) < csrCurrent.key ){ +** AGGINVERSE +** } +** RETURN_ROW +** +** RANGE BETWEEN FOLLOWING AND FOLLOWING +** +** ... loop started by sqlite3WhereBegin() ... +** if( new partition ){ +** Gosub flush +** } +** Insert new row into eph table. +** if( first row of partition ){ +** Rewind(csrEnd) ; Rewind(csrStart) ; Rewind(csrCurrent) +** regEnd = +** regStart = +** }else{ +** AGGSTEP +** while( (csrCurrent.key + regEnd) < csrEnd.key ){ +** while( (csrCurrent.key + regStart) > csrStart.key ){ +** AGGINVERSE +** } +** RETURN_ROW +** } +** } +** } +** flush: +** AGGSTEP +** while( 1 ){ +** while( (csrCurrent.key + regStart) > csrStart.key ){ +** AGGINVERSE +** if( eof ) break "while( 1 )" loop. +** } +** RETURN_ROW +** } +** while( !eof csrCurrent ){ +** RETURN_ROW +** } +** +** The text above leaves out many details. Refer to the code and comments +** below for a more complete picture. +*/ +SQLITE_PRIVATE void sqlite3WindowCodeStep( + Parse *pParse, /* Parse context */ + Select *p, /* Rewritten SELECT statement */ + WhereInfo *pWInfo, /* Context returned by sqlite3WhereBegin() */ + int regGosub, /* Register for OP_Gosub */ + int addrGosub /* OP_Gosub here to return each row */ +){ + Window *pMWin = p->pWin; + ExprList *pOrderBy = pMWin->pOrderBy; + Vdbe *v = sqlite3GetVdbe(pParse); + int csrWrite; /* Cursor used to write to eph. table */ + int csrInput = p->pSrc->a[0].iCursor; /* Cursor of sub-select */ + int nInput = p->pSrc->a[0].pTab->nCol; /* Number of cols returned by sub */ + int iInput; /* To iterate through sub cols */ + int addrNe; /* Address of OP_Ne */ + int addrGosubFlush = 0; /* Address of OP_Gosub to flush: */ + int addrInteger = 0; /* Address of OP_Integer */ + int addrEmpty; /* Address of OP_Rewind in flush: */ + int regNew; /* Array of registers holding new input row */ + int regRecord; /* regNew array in record form */ + int regRowid; /* Rowid for regRecord in eph table */ + int regNewPeer = 0; /* Peer values for new row (part of regNew) */ + int regPeer = 0; /* Peer values for current row */ + int regFlushPart = 0; /* Register for "Gosub flush_partition" */ + WindowCodeArg s; /* Context object for sub-routines */ + int lblWhereEnd; /* Label just before sqlite3WhereEnd() code */ + int regStart = 0; /* Value of PRECEDING */ + int regEnd = 0; /* Value of FOLLOWING */ + + assert( pMWin->eStart==TK_PRECEDING || pMWin->eStart==TK_CURRENT + || pMWin->eStart==TK_FOLLOWING || pMWin->eStart==TK_UNBOUNDED + ); + assert( pMWin->eEnd==TK_FOLLOWING || pMWin->eEnd==TK_CURRENT + || pMWin->eEnd==TK_UNBOUNDED || pMWin->eEnd==TK_PRECEDING + ); + assert( pMWin->eExclude==0 || pMWin->eExclude==TK_CURRENT + || pMWin->eExclude==TK_GROUP || pMWin->eExclude==TK_TIES + || pMWin->eExclude==TK_NO + ); + + lblWhereEnd = sqlite3VdbeMakeLabel(pParse); + + /* Fill in the context object */ + memset(&s, 0, sizeof(WindowCodeArg)); + s.pParse = pParse; + s.pMWin = pMWin; + s.pVdbe = v; + s.regGosub = regGosub; + s.addrGosub = addrGosub; + s.current.csr = pMWin->iEphCsr; + csrWrite = s.current.csr+1; + s.start.csr = s.current.csr+2; + s.end.csr = s.current.csr+3; + + /* Figure out when rows may be deleted from the ephemeral table. There + ** are four options - they may never be deleted (eDelete==0), they may + ** be deleted as soon as they are no longer part of the window frame + ** (eDelete==WINDOW_AGGINVERSE), they may be deleted as after the row + ** has been returned to the caller (WINDOW_RETURN_ROW), or they may + ** be deleted after they enter the frame (WINDOW_AGGSTEP). */ + switch( pMWin->eStart ){ + case TK_FOLLOWING: + if( pMWin->eFrmType!=TK_RANGE + && windowExprGtZero(pParse, pMWin->pStart) + ){ + s.eDelete = WINDOW_RETURN_ROW; + } + break; + case TK_UNBOUNDED: + if( windowCacheFrame(pMWin)==0 ){ + if( pMWin->eEnd==TK_PRECEDING ){ + if( pMWin->eFrmType!=TK_RANGE + && windowExprGtZero(pParse, pMWin->pEnd) + ){ + s.eDelete = WINDOW_AGGSTEP; + } + }else{ + s.eDelete = WINDOW_RETURN_ROW; + } + } + break; + default: + s.eDelete = WINDOW_AGGINVERSE; + break; + } + + /* Allocate registers for the array of values from the sub-query, the + ** samve values in record form, and the rowid used to insert said record + ** into the ephemeral table. */ + regNew = pParse->nMem+1; + pParse->nMem += nInput; + regRecord = ++pParse->nMem; + regRowid = ++pParse->nMem; + + /* If the window frame contains an " PRECEDING" or " FOLLOWING" + ** clause, allocate registers to store the results of evaluating each + ** . */ + if( pMWin->eStart==TK_PRECEDING || pMWin->eStart==TK_FOLLOWING ){ + regStart = ++pParse->nMem; + } + if( pMWin->eEnd==TK_PRECEDING || pMWin->eEnd==TK_FOLLOWING ){ + regEnd = ++pParse->nMem; + } + + /* If this is not a "ROWS BETWEEN ..." frame, then allocate arrays of + ** registers to store copies of the ORDER BY expressions (peer values) + ** for the main loop, and for each cursor (start, current and end). */ + if( pMWin->eFrmType!=TK_ROWS ){ + int nPeer = (pOrderBy ? pOrderBy->nExpr : 0); + regNewPeer = regNew + pMWin->nBufferCol; + if( pMWin->pPartition ) regNewPeer += pMWin->pPartition->nExpr; + regPeer = pParse->nMem+1; pParse->nMem += nPeer; + s.start.reg = pParse->nMem+1; pParse->nMem += nPeer; + s.current.reg = pParse->nMem+1; pParse->nMem += nPeer; + s.end.reg = pParse->nMem+1; pParse->nMem += nPeer; + } + + /* Load the column values for the row returned by the sub-select + ** into an array of registers starting at regNew. Assemble them into + ** a record in register regRecord. */ + for(iInput=0; iInputpPartition ){ + int addr; + ExprList *pPart = pMWin->pPartition; + int nPart = pPart->nExpr; + int regNewPart = regNew + pMWin->nBufferCol; + KeyInfo *pKeyInfo = sqlite3KeyInfoFromExprList(pParse, pPart, 0, 0); + + regFlushPart = ++pParse->nMem; + addr = sqlite3VdbeAddOp3(v, OP_Compare, regNewPart, pMWin->regPart, nPart); + sqlite3VdbeAppendP4(v, (void*)pKeyInfo, P4_KEYINFO); + sqlite3VdbeAddOp3(v, OP_Jump, addr+2, addr+4, addr+2); + VdbeCoverageEqNe(v); + addrGosubFlush = sqlite3VdbeAddOp1(v, OP_Gosub, regFlushPart); + VdbeComment((v, "call flush_partition")); + sqlite3VdbeAddOp3(v, OP_Copy, regNewPart, pMWin->regPart, nPart-1); + } + + /* Insert the new row into the ephemeral table */ + sqlite3VdbeAddOp2(v, OP_NewRowid, csrWrite, regRowid); + sqlite3VdbeAddOp3(v, OP_Insert, csrWrite, regRecord, regRowid); + addrNe = sqlite3VdbeAddOp3(v, OP_Ne, pMWin->regOne, 0, regRowid); + VdbeCoverageNeverNull(v); + + /* This block is run for the first row of each partition */ + s.regArg = windowInitAccum(pParse, pMWin); + + if( regStart ){ + sqlite3ExprCode(pParse, pMWin->pStart, regStart); + windowCheckValue(pParse, regStart, 0 + (pMWin->eFrmType==TK_RANGE?3:0)); + } + if( regEnd ){ + sqlite3ExprCode(pParse, pMWin->pEnd, regEnd); + windowCheckValue(pParse, regEnd, 1 + (pMWin->eFrmType==TK_RANGE?3:0)); + } + + if( pMWin->eFrmType!=TK_RANGE && pMWin->eStart==pMWin->eEnd && regStart ){ + int op = ((pMWin->eStart==TK_FOLLOWING) ? OP_Ge : OP_Le); + int addrGe = sqlite3VdbeAddOp3(v, op, regStart, 0, regEnd); + VdbeCoverageNeverNullIf(v, op==OP_Ge); /* NeverNull because bound */ + VdbeCoverageNeverNullIf(v, op==OP_Le); /* values previously checked */ + windowAggFinal(&s, 0); + sqlite3VdbeAddOp2(v, OP_Rewind, s.current.csr, 1); + VdbeCoverageNeverTaken(v); + windowReturnOneRow(&s); + sqlite3VdbeAddOp1(v, OP_ResetSorter, s.current.csr); + sqlite3VdbeAddOp2(v, OP_Goto, 0, lblWhereEnd); + sqlite3VdbeJumpHere(v, addrGe); + } + if( pMWin->eStart==TK_FOLLOWING && pMWin->eFrmType!=TK_RANGE && regEnd ){ + assert( pMWin->eEnd==TK_FOLLOWING ); + sqlite3VdbeAddOp3(v, OP_Subtract, regStart, regEnd, regStart); + } + + if( pMWin->eStart!=TK_UNBOUNDED ){ + sqlite3VdbeAddOp2(v, OP_Rewind, s.start.csr, 1); + VdbeCoverageNeverTaken(v); + } + sqlite3VdbeAddOp2(v, OP_Rewind, s.current.csr, 1); + VdbeCoverageNeverTaken(v); + sqlite3VdbeAddOp2(v, OP_Rewind, s.end.csr, 1); + VdbeCoverageNeverTaken(v); + if( regPeer && pOrderBy ){ + sqlite3VdbeAddOp3(v, OP_Copy, regNewPeer, regPeer, pOrderBy->nExpr-1); + sqlite3VdbeAddOp3(v, OP_Copy, regPeer, s.start.reg, pOrderBy->nExpr-1); + sqlite3VdbeAddOp3(v, OP_Copy, regPeer, s.current.reg, pOrderBy->nExpr-1); + sqlite3VdbeAddOp3(v, OP_Copy, regPeer, s.end.reg, pOrderBy->nExpr-1); + } + + sqlite3VdbeAddOp2(v, OP_Goto, 0, lblWhereEnd); + + sqlite3VdbeJumpHere(v, addrNe); + + /* Beginning of the block executed for the second and subsequent rows. */ + if( regPeer ){ + windowIfNewPeer(pParse, pOrderBy, regNewPeer, regPeer, lblWhereEnd); + } + if( pMWin->eStart==TK_FOLLOWING ){ + windowCodeOp(&s, WINDOW_AGGSTEP, 0, 0); + if( pMWin->eEnd!=TK_UNBOUNDED ){ + if( pMWin->eFrmType==TK_RANGE ){ + int lbl = sqlite3VdbeMakeLabel(pParse); + int addrNext = sqlite3VdbeCurrentAddr(v); + windowCodeRangeTest(&s, OP_Ge, s.current.csr, regEnd, s.end.csr, lbl); + windowCodeOp(&s, WINDOW_AGGINVERSE, regStart, 0); + windowCodeOp(&s, WINDOW_RETURN_ROW, 0, 0); + sqlite3VdbeAddOp2(v, OP_Goto, 0, addrNext); + sqlite3VdbeResolveLabel(v, lbl); + }else{ + windowCodeOp(&s, WINDOW_RETURN_ROW, regEnd, 0); + windowCodeOp(&s, WINDOW_AGGINVERSE, regStart, 0); + } + } + }else + if( pMWin->eEnd==TK_PRECEDING ){ + int bRPS = (pMWin->eStart==TK_PRECEDING && pMWin->eFrmType==TK_RANGE); + windowCodeOp(&s, WINDOW_AGGSTEP, regEnd, 0); + if( bRPS ) windowCodeOp(&s, WINDOW_AGGINVERSE, regStart, 0); + windowCodeOp(&s, WINDOW_RETURN_ROW, 0, 0); + if( !bRPS ) windowCodeOp(&s, WINDOW_AGGINVERSE, regStart, 0); + }else{ + int addr = 0; + windowCodeOp(&s, WINDOW_AGGSTEP, 0, 0); + if( pMWin->eEnd!=TK_UNBOUNDED ){ + if( pMWin->eFrmType==TK_RANGE ){ + int lbl = 0; + addr = sqlite3VdbeCurrentAddr(v); + if( regEnd ){ + lbl = sqlite3VdbeMakeLabel(pParse); + windowCodeRangeTest(&s, OP_Ge, s.current.csr, regEnd, s.end.csr, lbl); + } + windowCodeOp(&s, WINDOW_RETURN_ROW, 0, 0); + windowCodeOp(&s, WINDOW_AGGINVERSE, regStart, 0); + if( regEnd ){ + sqlite3VdbeAddOp2(v, OP_Goto, 0, addr); + sqlite3VdbeResolveLabel(v, lbl); + } + }else{ + if( regEnd ){ + addr = sqlite3VdbeAddOp3(v, OP_IfPos, regEnd, 0, 1); + VdbeCoverage(v); + } + windowCodeOp(&s, WINDOW_RETURN_ROW, 0, 0); + windowCodeOp(&s, WINDOW_AGGINVERSE, regStart, 0); + if( regEnd ) sqlite3VdbeJumpHere(v, addr); + } + } + } + + /* End of the main input loop */ + sqlite3VdbeResolveLabel(v, lblWhereEnd); + sqlite3WhereEnd(pWInfo); + + /* Fall through */ + if( pMWin->pPartition ){ + addrInteger = sqlite3VdbeAddOp2(v, OP_Integer, 0, regFlushPart); + sqlite3VdbeJumpHere(v, addrGosubFlush); + } + + addrEmpty = sqlite3VdbeAddOp1(v, OP_Rewind, csrWrite); + VdbeCoverage(v); + if( pMWin->eEnd==TK_PRECEDING ){ + int bRPS = (pMWin->eStart==TK_PRECEDING && pMWin->eFrmType==TK_RANGE); + windowCodeOp(&s, WINDOW_AGGSTEP, regEnd, 0); + if( bRPS ) windowCodeOp(&s, WINDOW_AGGINVERSE, regStart, 0); + windowCodeOp(&s, WINDOW_RETURN_ROW, 0, 0); + }else if( pMWin->eStart==TK_FOLLOWING ){ + int addrStart; + int addrBreak1; + int addrBreak2; + int addrBreak3; + windowCodeOp(&s, WINDOW_AGGSTEP, 0, 0); + if( pMWin->eFrmType==TK_RANGE ){ + addrStart = sqlite3VdbeCurrentAddr(v); + addrBreak2 = windowCodeOp(&s, WINDOW_AGGINVERSE, regStart, 1); + addrBreak1 = windowCodeOp(&s, WINDOW_RETURN_ROW, 0, 1); + }else + if( pMWin->eEnd==TK_UNBOUNDED ){ + addrStart = sqlite3VdbeCurrentAddr(v); + addrBreak1 = windowCodeOp(&s, WINDOW_RETURN_ROW, regStart, 1); + addrBreak2 = windowCodeOp(&s, WINDOW_AGGINVERSE, 0, 1); + }else{ + assert( pMWin->eEnd==TK_FOLLOWING ); + addrStart = sqlite3VdbeCurrentAddr(v); + addrBreak1 = windowCodeOp(&s, WINDOW_RETURN_ROW, regEnd, 1); + addrBreak2 = windowCodeOp(&s, WINDOW_AGGINVERSE, regStart, 1); + } + sqlite3VdbeAddOp2(v, OP_Goto, 0, addrStart); + sqlite3VdbeJumpHere(v, addrBreak2); + addrStart = sqlite3VdbeCurrentAddr(v); + addrBreak3 = windowCodeOp(&s, WINDOW_RETURN_ROW, 0, 1); + sqlite3VdbeAddOp2(v, OP_Goto, 0, addrStart); + sqlite3VdbeJumpHere(v, addrBreak1); + sqlite3VdbeJumpHere(v, addrBreak3); + }else{ + int addrBreak; + int addrStart; + windowCodeOp(&s, WINDOW_AGGSTEP, 0, 0); + addrStart = sqlite3VdbeCurrentAddr(v); + addrBreak = windowCodeOp(&s, WINDOW_RETURN_ROW, 0, 1); + windowCodeOp(&s, WINDOW_AGGINVERSE, regStart, 0); + sqlite3VdbeAddOp2(v, OP_Goto, 0, addrStart); + sqlite3VdbeJumpHere(v, addrBreak); + } + sqlite3VdbeJumpHere(v, addrEmpty); + + sqlite3VdbeAddOp1(v, OP_ResetSorter, s.current.csr); + if( pMWin->pPartition ){ + if( pMWin->regStartRowid ){ + sqlite3VdbeAddOp2(v, OP_Integer, 1, pMWin->regStartRowid); + sqlite3VdbeAddOp2(v, OP_Integer, 0, pMWin->regEndRowid); + } + sqlite3VdbeChangeP1(v, addrInteger, sqlite3VdbeCurrentAddr(v)); + sqlite3VdbeAddOp1(v, OP_Return, regFlushPart); + } +} + +#endif /* SQLITE_OMIT_WINDOWFUNC */ + +/************** End of window.c **********************************************/ +/************** Begin file parse.c *******************************************/ +/* +** 2000-05-29 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** Driver template for the LEMON parser generator. +** +** The "lemon" program processes an LALR(1) input grammar file, then uses +** this template to construct a parser. The "lemon" program inserts text +** at each "%%" line. Also, any "P-a-r-s-e" identifer prefix (without the +** interstitial "-" characters) contained in this template is changed into +** the value of the %name directive from the grammar. Otherwise, the content +** of this template is copied straight through into the generate parser +** source file. +** +** The following is the concatenation of all %include directives from the +** input grammar file: +*/ +/* #include */ +/* #include */ +/************ Begin %include sections from the grammar ************************/ + +/* #include "sqliteInt.h" */ + +/* +** Disable all error recovery processing in the parser push-down +** automaton. +*/ +#define YYNOERRORRECOVERY 1 + +/* +** Make yytestcase() the same as testcase() +*/ +#define yytestcase(X) testcase(X) + +/* +** Indicate that sqlite3ParserFree() will never be called with a null +** pointer. +*/ +#define YYPARSEFREENEVERNULL 1 + +/* +** In the amalgamation, the parse.c file generated by lemon and the +** tokenize.c file are concatenated. In that case, sqlite3RunParser() +** has access to the the size of the yyParser object and so the parser +** engine can be allocated from stack. In that case, only the +** sqlite3ParserInit() and sqlite3ParserFinalize() routines are invoked +** and the sqlite3ParserAlloc() and sqlite3ParserFree() routines can be +** omitted. +*/ +#ifdef SQLITE_AMALGAMATION +# define sqlite3Parser_ENGINEALWAYSONSTACK 1 +#endif + +/* +** Alternative datatype for the argument to the malloc() routine passed +** into sqlite3ParserAlloc(). The default is size_t. +*/ +#define YYMALLOCARGTYPE u64 + +/* +** An instance of the following structure describes the event of a +** TRIGGER. "a" is the event type, one of TK_UPDATE, TK_INSERT, +** TK_DELETE, or TK_INSTEAD. If the event is of the form +** +** UPDATE ON (a,b,c) +** +** Then the "b" IdList records the list "a,b,c". +*/ +struct TrigEvent { int a; IdList * b; }; + +struct FrameBound { int eType; Expr *pExpr; }; + +/* +** Disable lookaside memory allocation for objects that might be +** shared across database connections. +*/ +static void disableLookaside(Parse *pParse){ + pParse->disableLookaside++; + pParse->db->lookaside.bDisable++; +} + + + /* + ** For a compound SELECT statement, make sure p->pPrior->pNext==p for + ** all elements in the list. And make sure list length does not exceed + ** SQLITE_LIMIT_COMPOUND_SELECT. + */ + static void parserDoubleLinkSelect(Parse *pParse, Select *p){ + assert( p!=0 ); + if( p->pPrior ){ + Select *pNext = 0, *pLoop; + int mxSelect, cnt = 0; + for(pLoop=p; pLoop; pNext=pLoop, pLoop=pLoop->pPrior, cnt++){ + pLoop->pNext = pNext; + pLoop->selFlags |= SF_Compound; + } + if( (p->selFlags & SF_MultiValue)==0 && + (mxSelect = pParse->db->aLimit[SQLITE_LIMIT_COMPOUND_SELECT])>0 && + cnt>mxSelect + ){ + sqlite3ErrorMsg(pParse, "too many terms in compound SELECT"); + } + } + } + + + /* Construct a new Expr object from a single identifier. Use the + ** new Expr to populate pOut. Set the span of pOut to be the identifier + ** that created the expression. + */ + static Expr *tokenExpr(Parse *pParse, int op, Token t){ + Expr *p = sqlite3DbMallocRawNN(pParse->db, sizeof(Expr)+t.n+1); + if( p ){ + /* memset(p, 0, sizeof(Expr)); */ + p->op = (u8)op; + p->affExpr = 0; + p->flags = EP_Leaf; + p->iAgg = -1; + p->pLeft = p->pRight = 0; + p->x.pList = 0; + p->pAggInfo = 0; + p->y.pTab = 0; + p->op2 = 0; + p->iTable = 0; + p->iColumn = 0; + p->u.zToken = (char*)&p[1]; + memcpy(p->u.zToken, t.z, t.n); + p->u.zToken[t.n] = 0; + if( sqlite3Isquote(p->u.zToken[0]) ){ + sqlite3DequoteExpr(p); + } +#if SQLITE_MAX_EXPR_DEPTH>0 + p->nHeight = 1; +#endif + if( IN_RENAME_OBJECT ){ + return (Expr*)sqlite3RenameTokenMap(pParse, (void*)p, &t); + } + } + return p; + } + + + /* A routine to convert a binary TK_IS or TK_ISNOT expression into a + ** unary TK_ISNULL or TK_NOTNULL expression. */ + static void binaryToUnaryIfNull(Parse *pParse, Expr *pY, Expr *pA, int op){ + sqlite3 *db = pParse->db; + if( pA && pY && pY->op==TK_NULL && !IN_RENAME_OBJECT ){ + pA->op = (u8)op; + sqlite3ExprDelete(db, pA->pRight); + pA->pRight = 0; + } + } + + /* Add a single new term to an ExprList that is used to store a + ** list of identifiers. Report an error if the ID list contains + ** a COLLATE clause or an ASC or DESC keyword, except ignore the + ** error while parsing a legacy schema. + */ + static ExprList *parserAddExprIdListTerm( + Parse *pParse, + ExprList *pPrior, + Token *pIdToken, + int hasCollate, + int sortOrder + ){ + ExprList *p = sqlite3ExprListAppend(pParse, pPrior, 0); + if( (hasCollate || sortOrder!=SQLITE_SO_UNDEFINED) + && pParse->db->init.busy==0 + ){ + sqlite3ErrorMsg(pParse, "syntax error after column name \"%.*s\"", + pIdToken->n, pIdToken->z); + } + sqlite3ExprListSetName(pParse, p, pIdToken, 1); + return p; + } + +#if TK_SPAN>255 +# error too many tokens in the grammar +#endif +/**************** End of %include directives **********************************/ +/* These constants specify the various numeric values for terminal symbols +** in a format understandable to "makeheaders". This section is blank unless +** "lemon" is run with the "-m" command-line option. +***************** Begin makeheaders token definitions *************************/ +/**************** End makeheaders token definitions ***************************/ + +/* The next sections is a series of control #defines. +** various aspects of the generated parser. +** YYCODETYPE is the data type used to store the integer codes +** that represent terminal and non-terminal symbols. +** "unsigned char" is used if there are fewer than +** 256 symbols. Larger types otherwise. +** YYNOCODE is a number of type YYCODETYPE that is not used for +** any terminal or nonterminal symbol. +** YYFALLBACK If defined, this indicates that one or more tokens +** (also known as: "terminal symbols") have fall-back +** values which should be used if the original symbol +** would not parse. This permits keywords to sometimes +** be used as identifiers, for example. +** YYACTIONTYPE is the data type used for "action codes" - numbers +** that indicate what to do in response to the next +** token. +** sqlite3ParserTOKENTYPE is the data type used for minor type for terminal +** symbols. Background: A "minor type" is a semantic +** value associated with a terminal or non-terminal +** symbols. For example, for an "ID" terminal symbol, +** the minor type might be the name of the identifier. +** Each non-terminal can have a different minor type. +** Terminal symbols all have the same minor type, though. +** This macros defines the minor type for terminal +** symbols. +** YYMINORTYPE is the data type used for all minor types. +** This is typically a union of many types, one of +** which is sqlite3ParserTOKENTYPE. The entry in the union +** for terminal symbols is called "yy0". +** YYSTACKDEPTH is the maximum depth of the parser's stack. If +** zero the stack is dynamically sized using realloc() +** sqlite3ParserARG_SDECL A static variable declaration for the %extra_argument +** sqlite3ParserARG_PDECL A parameter declaration for the %extra_argument +** sqlite3ParserARG_PARAM Code to pass %extra_argument as a subroutine parameter +** sqlite3ParserARG_STORE Code to store %extra_argument into yypParser +** sqlite3ParserARG_FETCH Code to extract %extra_argument from yypParser +** sqlite3ParserCTX_* As sqlite3ParserARG_ except for %extra_context +** YYERRORSYMBOL is the code number of the error symbol. If not +** defined, then do no error processing. +** YYNSTATE the combined number of states. +** YYNRULE the number of rules in the grammar +** YYNTOKEN Number of terminal symbols +** YY_MAX_SHIFT Maximum value for shift actions +** YY_MIN_SHIFTREDUCE Minimum value for shift-reduce actions +** YY_MAX_SHIFTREDUCE Maximum value for shift-reduce actions +** YY_ERROR_ACTION The yy_action[] code for syntax error +** YY_ACCEPT_ACTION The yy_action[] code for accept +** YY_NO_ACTION The yy_action[] code for no-op +** YY_MIN_REDUCE Minimum value for reduce actions +** YY_MAX_REDUCE Maximum value for reduce actions +*/ +#ifndef INTERFACE +# define INTERFACE 1 +#endif +/************* Begin control #defines *****************************************/ +#define YYCODETYPE unsigned short int +#define YYNOCODE 307 +#define YYACTIONTYPE unsigned short int +#define YYWILDCARD 98 +#define sqlite3ParserTOKENTYPE Token +typedef union { + int yyinit; + sqlite3ParserTOKENTYPE yy0; + const char* yy8; + Select* yy25; + int yy32; + Expr* yy46; + struct FrameBound yy57; + u8 yy118; + ExprList* yy138; + Upsert* yy288; + With* yy297; + IdList* yy406; + Window* yy455; + struct {int value; int mask;} yy495; + TriggerStep* yy527; + struct TrigEvent yy572; + SrcList* yy609; +} YYMINORTYPE; +#ifndef YYSTACKDEPTH +#define YYSTACKDEPTH 100 +#endif +#define sqlite3ParserARG_SDECL +#define sqlite3ParserARG_PDECL +#define sqlite3ParserARG_PARAM +#define sqlite3ParserARG_FETCH +#define sqlite3ParserARG_STORE +#define sqlite3ParserCTX_SDECL Parse *pParse; +#define sqlite3ParserCTX_PDECL ,Parse *pParse +#define sqlite3ParserCTX_PARAM ,pParse +#define sqlite3ParserCTX_FETCH Parse *pParse=yypParser->pParse; +#define sqlite3ParserCTX_STORE yypParser->pParse=pParse; +#define YYFALLBACK 1 +#define YYNSTATE 543 +#define YYNRULE 381 +#define YYNTOKEN 179 +#define YY_MAX_SHIFT 542 +#define YY_MIN_SHIFTREDUCE 790 +#define YY_MAX_SHIFTREDUCE 1170 +#define YY_ERROR_ACTION 1171 +#define YY_ACCEPT_ACTION 1172 +#define YY_NO_ACTION 1173 +#define YY_MIN_REDUCE 1174 +#define YY_MAX_REDUCE 1554 +/************* End control #defines *******************************************/ +#define YY_NLOOKAHEAD ((int)(sizeof(yy_lookahead)/sizeof(yy_lookahead[0]))) + +/* Define the yytestcase() macro to be a no-op if is not already defined +** otherwise. +** +** Applications can choose to define yytestcase() in the %include section +** to a macro that can assist in verifying code coverage. For production +** code the yytestcase() macro should be turned off. But it is useful +** for testing. +*/ +#ifndef yytestcase +# define yytestcase(X) +#endif + + +/* Next are the tables used to determine what action to take based on the +** current state and lookahead token. These tables are used to implement +** functions that take a state number and lookahead value and return an +** action integer. +** +** Suppose the action integer is N. Then the action is determined as +** follows +** +** 0 <= N <= YY_MAX_SHIFT Shift N. That is, push the lookahead +** token onto the stack and goto state N. +** +** N between YY_MIN_SHIFTREDUCE Shift to an arbitrary state then +** and YY_MAX_SHIFTREDUCE reduce by rule N-YY_MIN_SHIFTREDUCE. +** +** N == YY_ERROR_ACTION A syntax error has occurred. +** +** N == YY_ACCEPT_ACTION The parser accepts its input. +** +** N == YY_NO_ACTION No such action. Denotes unused +** slots in the yy_action[] table. +** +** N between YY_MIN_REDUCE Reduce by rule N-YY_MIN_REDUCE +** and YY_MAX_REDUCE +** +** The action table is constructed as a single large table named yy_action[]. +** Given state S and lookahead X, the action is computed as either: +** +** (A) N = yy_action[ yy_shift_ofst[S] + X ] +** (B) N = yy_default[S] +** +** The (A) formula is preferred. The B formula is used instead if +** yy_lookahead[yy_shift_ofst[S]+X] is not equal to X. +** +** The formulas above are for computing the action when the lookahead is +** a terminal symbol. If the lookahead is a non-terminal (as occurs after +** a reduce action) then the yy_reduce_ofst[] array is used in place of +** the yy_shift_ofst[] array. +** +** The following are the tables generated in this section: +** +** yy_action[] A single table containing all actions. +** yy_lookahead[] A table containing the lookahead for each entry in +** yy_action. Used to detect hash collisions. +** yy_shift_ofst[] For each state, the offset into yy_action for +** shifting terminals. +** yy_reduce_ofst[] For each state, the offset into yy_action for +** shifting non-terminals after a reduce. +** yy_default[] Default action for each state. +** +*********** Begin parsing tables **********************************************/ +#define YY_ACTTAB_COUNT (1913) +static const YYACTIONTYPE yy_action[] = { + /* 0 */ 537, 339, 537, 1241, 1220, 537, 12, 537, 112, 109, + /* 10 */ 209, 537, 1241, 537, 1205, 462, 112, 109, 209, 386, + /* 20 */ 338, 462, 42, 42, 42, 42, 445, 42, 42, 70, + /* 30 */ 70, 922, 1208, 70, 70, 70, 70, 1443, 403, 923, + /* 40 */ 531, 531, 531, 119, 120, 110, 1148, 1148, 991, 994, + /* 50 */ 984, 984, 117, 117, 118, 118, 118, 118, 425, 386, + /* 60 */ 1498, 542, 2, 1176, 1442, 519, 141, 1518, 289, 519, + /* 70 */ 134, 519, 95, 259, 495, 1215, 189, 1254, 518, 494, + /* 80 */ 484, 437, 296, 119, 120, 110, 1148, 1148, 991, 994, + /* 90 */ 984, 984, 117, 117, 118, 118, 118, 118, 270, 116, + /* 100 */ 116, 116, 116, 115, 115, 114, 114, 114, 113, 418, + /* 110 */ 264, 264, 264, 264, 423, 1479, 352, 1481, 123, 351, + /* 120 */ 1479, 508, 1094, 534, 1034, 534, 1099, 386, 1099, 239, + /* 130 */ 206, 112, 109, 209, 96, 1094, 376, 219, 1094, 116, + /* 140 */ 116, 116, 116, 115, 115, 114, 114, 114, 113, 418, + /* 150 */ 480, 119, 120, 110, 1148, 1148, 991, 994, 984, 984, + /* 160 */ 117, 117, 118, 118, 118, 118, 353, 422, 1407, 264, + /* 170 */ 264, 114, 114, 114, 113, 418, 883, 121, 416, 416, + /* 180 */ 416, 882, 534, 116, 116, 116, 116, 115, 115, 114, + /* 190 */ 114, 114, 113, 418, 212, 415, 414, 386, 443, 383, + /* 200 */ 382, 118, 118, 118, 118, 111, 177, 116, 116, 116, + /* 210 */ 116, 115, 115, 114, 114, 114, 113, 418, 112, 109, + /* 220 */ 209, 119, 120, 110, 1148, 1148, 991, 994, 984, 984, + /* 230 */ 117, 117, 118, 118, 118, 118, 386, 438, 312, 1163, + /* 240 */ 1155, 80, 1155, 1127, 514, 79, 116, 116, 116, 116, + /* 250 */ 115, 115, 114, 114, 114, 113, 418, 514, 428, 418, + /* 260 */ 119, 120, 110, 1148, 1148, 991, 994, 984, 984, 117, + /* 270 */ 117, 118, 118, 118, 118, 428, 427, 116, 116, 116, + /* 280 */ 116, 115, 115, 114, 114, 114, 113, 418, 115, 115, + /* 290 */ 114, 114, 114, 113, 418, 1127, 1127, 1128, 1129, 1094, + /* 300 */ 258, 258, 192, 386, 408, 371, 1168, 326, 118, 118, + /* 310 */ 118, 118, 1094, 534, 374, 1094, 116, 116, 116, 116, + /* 320 */ 115, 115, 114, 114, 114, 113, 418, 119, 120, 110, + /* 330 */ 1148, 1148, 991, 994, 984, 984, 117, 117, 118, 118, + /* 340 */ 118, 118, 386, 354, 445, 428, 829, 238, 1127, 1128, + /* 350 */ 1129, 515, 1466, 116, 116, 116, 116, 115, 115, 114, + /* 360 */ 114, 114, 113, 418, 1127, 1467, 119, 120, 110, 1148, + /* 370 */ 1148, 991, 994, 984, 984, 117, 117, 118, 118, 118, + /* 380 */ 118, 1169, 82, 116, 116, 116, 116, 115, 115, 114, + /* 390 */ 114, 114, 113, 418, 405, 112, 109, 209, 161, 445, + /* 400 */ 250, 267, 336, 478, 331, 477, 236, 951, 1127, 386, + /* 410 */ 888, 1521, 329, 822, 852, 162, 274, 1127, 1128, 1129, + /* 420 */ 338, 169, 116, 116, 116, 116, 115, 115, 114, 114, + /* 430 */ 114, 113, 418, 119, 120, 110, 1148, 1148, 991, 994, + /* 440 */ 984, 984, 117, 117, 118, 118, 118, 118, 386, 438, + /* 450 */ 312, 1502, 1112, 1176, 161, 288, 528, 311, 289, 883, + /* 460 */ 134, 1127, 1128, 1129, 882, 537, 143, 1254, 288, 528, + /* 470 */ 297, 275, 119, 120, 110, 1148, 1148, 991, 994, 984, + /* 480 */ 984, 117, 117, 118, 118, 118, 118, 70, 70, 116, + /* 490 */ 116, 116, 116, 115, 115, 114, 114, 114, 113, 418, + /* 500 */ 264, 264, 12, 264, 264, 395, 1127, 483, 1473, 1094, + /* 510 */ 204, 482, 6, 534, 1258, 386, 534, 1474, 825, 972, + /* 520 */ 504, 6, 1094, 500, 95, 1094, 534, 219, 116, 116, + /* 530 */ 116, 116, 115, 115, 114, 114, 114, 113, 418, 119, + /* 540 */ 120, 110, 1148, 1148, 991, 994, 984, 984, 117, 117, + /* 550 */ 118, 118, 118, 118, 386, 1339, 971, 422, 956, 1127, + /* 560 */ 1128, 1129, 231, 512, 1473, 475, 472, 471, 6, 113, + /* 570 */ 418, 825, 962, 298, 503, 470, 961, 452, 119, 120, + /* 580 */ 110, 1148, 1148, 991, 994, 984, 984, 117, 117, 118, + /* 590 */ 118, 118, 118, 395, 537, 116, 116, 116, 116, 115, + /* 600 */ 115, 114, 114, 114, 113, 418, 202, 961, 961, 963, + /* 610 */ 231, 971, 1127, 475, 472, 471, 13, 13, 951, 1127, + /* 620 */ 834, 386, 1207, 470, 399, 183, 447, 962, 462, 162, + /* 630 */ 397, 961, 1246, 1246, 116, 116, 116, 116, 115, 115, + /* 640 */ 114, 114, 114, 113, 418, 119, 120, 110, 1148, 1148, + /* 650 */ 991, 994, 984, 984, 117, 117, 118, 118, 118, 118, + /* 660 */ 386, 271, 961, 961, 963, 1127, 1128, 1129, 311, 433, + /* 670 */ 299, 1406, 1127, 1128, 1129, 178, 1471, 138, 162, 32, + /* 680 */ 6, 1127, 288, 528, 119, 120, 110, 1148, 1148, 991, + /* 690 */ 994, 984, 984, 117, 117, 118, 118, 118, 118, 909, + /* 700 */ 390, 116, 116, 116, 116, 115, 115, 114, 114, 114, + /* 710 */ 113, 418, 1127, 429, 817, 537, 1127, 265, 265, 981, + /* 720 */ 981, 992, 995, 324, 1055, 93, 520, 5, 338, 537, + /* 730 */ 534, 288, 528, 1522, 1127, 1128, 1129, 70, 70, 1056, + /* 740 */ 116, 116, 116, 116, 115, 115, 114, 114, 114, 113, + /* 750 */ 418, 70, 70, 1495, 1057, 537, 98, 1244, 1244, 264, + /* 760 */ 264, 908, 371, 1076, 1127, 1127, 1128, 1129, 817, 1127, + /* 770 */ 1128, 1129, 534, 519, 140, 863, 386, 13, 13, 456, + /* 780 */ 192, 193, 521, 453, 319, 864, 322, 284, 365, 430, + /* 790 */ 985, 402, 379, 1077, 1548, 101, 386, 1548, 3, 395, + /* 800 */ 119, 120, 110, 1148, 1148, 991, 994, 984, 984, 117, + /* 810 */ 117, 118, 118, 118, 118, 386, 451, 1127, 1128, 1129, + /* 820 */ 119, 120, 110, 1148, 1148, 991, 994, 984, 984, 117, + /* 830 */ 117, 118, 118, 118, 118, 1127, 1354, 1412, 1169, 119, + /* 840 */ 108, 110, 1148, 1148, 991, 994, 984, 984, 117, 117, + /* 850 */ 118, 118, 118, 118, 1412, 1414, 116, 116, 116, 116, + /* 860 */ 115, 115, 114, 114, 114, 113, 418, 272, 535, 1075, + /* 870 */ 877, 877, 337, 1492, 309, 462, 116, 116, 116, 116, + /* 880 */ 115, 115, 114, 114, 114, 113, 418, 537, 1127, 1128, + /* 890 */ 1129, 537, 360, 537, 356, 116, 116, 116, 116, 115, + /* 900 */ 115, 114, 114, 114, 113, 418, 386, 264, 264, 13, + /* 910 */ 13, 273, 1127, 13, 13, 13, 13, 304, 1253, 386, + /* 920 */ 534, 1077, 1549, 404, 1412, 1549, 496, 277, 451, 186, + /* 930 */ 1252, 120, 110, 1148, 1148, 991, 994, 984, 984, 117, + /* 940 */ 117, 118, 118, 118, 118, 110, 1148, 1148, 991, 994, + /* 950 */ 984, 984, 117, 117, 118, 118, 118, 118, 105, 529, + /* 960 */ 537, 4, 1339, 264, 264, 1127, 1128, 1129, 1039, 1039, + /* 970 */ 459, 795, 796, 797, 536, 532, 534, 242, 301, 807, + /* 980 */ 303, 462, 69, 69, 451, 1353, 116, 116, 116, 116, + /* 990 */ 115, 115, 114, 114, 114, 113, 418, 1075, 419, 116, + /* 1000 */ 116, 116, 116, 115, 115, 114, 114, 114, 113, 418, + /* 1010 */ 526, 537, 1146, 192, 350, 105, 529, 537, 4, 497, + /* 1020 */ 162, 337, 1492, 310, 1249, 385, 1550, 372, 9, 462, + /* 1030 */ 242, 400, 532, 13, 13, 499, 971, 843, 436, 70, + /* 1040 */ 70, 359, 103, 103, 8, 339, 278, 187, 278, 104, + /* 1050 */ 1127, 419, 539, 538, 1339, 419, 961, 302, 1339, 1172, + /* 1060 */ 1, 1, 542, 2, 1176, 1146, 1146, 526, 476, 289, + /* 1070 */ 30, 134, 317, 288, 528, 285, 844, 1014, 1254, 276, + /* 1080 */ 1472, 506, 410, 1194, 6, 207, 505, 961, 961, 963, + /* 1090 */ 964, 27, 449, 971, 415, 414, 234, 233, 232, 103, + /* 1100 */ 103, 31, 1152, 1127, 1128, 1129, 104, 1154, 419, 539, + /* 1110 */ 538, 264, 264, 961, 1399, 1153, 264, 264, 1470, 1146, + /* 1120 */ 537, 216, 6, 401, 534, 1197, 392, 458, 406, 534, + /* 1130 */ 537, 485, 358, 537, 261, 537, 1339, 907, 219, 1155, + /* 1140 */ 467, 1155, 50, 50, 961, 961, 963, 964, 27, 1497, + /* 1150 */ 1116, 421, 70, 70, 268, 70, 70, 13, 13, 369, + /* 1160 */ 369, 368, 253, 366, 264, 264, 804, 235, 422, 105, + /* 1170 */ 529, 516, 4, 287, 487, 510, 493, 534, 486, 213, + /* 1180 */ 1055, 294, 490, 384, 1127, 450, 532, 338, 413, 293, + /* 1190 */ 522, 417, 335, 1036, 509, 1056, 107, 1036, 16, 16, + /* 1200 */ 1469, 1094, 334, 1105, 6, 411, 1145, 264, 264, 419, + /* 1210 */ 1057, 102, 511, 100, 1094, 264, 264, 1094, 922, 215, + /* 1220 */ 534, 526, 907, 264, 264, 208, 923, 154, 534, 457, + /* 1230 */ 156, 525, 391, 142, 218, 506, 534, 1127, 1128, 1129, + /* 1240 */ 507, 139, 1131, 38, 214, 530, 392, 971, 329, 1454, + /* 1250 */ 907, 1105, 537, 103, 103, 105, 529, 537, 4, 537, + /* 1260 */ 104, 424, 419, 539, 538, 537, 502, 961, 517, 537, + /* 1270 */ 1072, 537, 532, 373, 54, 54, 288, 528, 387, 55, + /* 1280 */ 55, 15, 15, 288, 528, 17, 136, 44, 44, 1451, + /* 1290 */ 537, 56, 56, 57, 57, 419, 1131, 291, 961, 961, + /* 1300 */ 963, 964, 27, 393, 163, 537, 426, 526, 263, 206, + /* 1310 */ 208, 517, 58, 58, 235, 440, 842, 841, 197, 105, + /* 1320 */ 529, 506, 4, 1033, 439, 1033, 505, 59, 59, 308, + /* 1330 */ 849, 850, 95, 971, 537, 907, 532, 948, 832, 103, + /* 1340 */ 103, 105, 529, 537, 4, 1021, 104, 537, 419, 539, + /* 1350 */ 538, 1116, 421, 961, 537, 268, 60, 60, 532, 419, + /* 1360 */ 369, 369, 368, 253, 366, 61, 61, 804, 965, 45, + /* 1370 */ 45, 526, 537, 1032, 1277, 1032, 46, 46, 537, 391, + /* 1380 */ 213, 419, 294, 266, 961, 961, 963, 964, 27, 292, + /* 1390 */ 293, 295, 832, 526, 48, 48, 1290, 971, 1289, 1021, + /* 1400 */ 49, 49, 432, 103, 103, 887, 953, 537, 1457, 241, + /* 1410 */ 104, 305, 419, 539, 538, 925, 926, 961, 444, 971, + /* 1420 */ 215, 241, 965, 1224, 537, 103, 103, 1431, 154, 62, + /* 1430 */ 62, 156, 104, 1430, 419, 539, 538, 97, 529, 961, + /* 1440 */ 4, 537, 454, 537, 314, 214, 63, 63, 961, 961, + /* 1450 */ 963, 964, 27, 537, 532, 446, 1286, 318, 241, 537, + /* 1460 */ 321, 323, 325, 64, 64, 14, 14, 1237, 537, 1223, + /* 1470 */ 961, 961, 963, 964, 27, 65, 65, 419, 537, 387, + /* 1480 */ 537, 125, 125, 537, 288, 528, 537, 1486, 537, 526, + /* 1490 */ 66, 66, 313, 524, 537, 95, 468, 1221, 1511, 237, + /* 1500 */ 51, 51, 67, 67, 330, 68, 68, 426, 52, 52, + /* 1510 */ 149, 149, 1222, 340, 341, 971, 150, 150, 1298, 463, + /* 1520 */ 327, 103, 103, 95, 537, 1338, 1273, 537, 104, 537, + /* 1530 */ 419, 539, 538, 1284, 537, 961, 268, 283, 523, 1344, + /* 1540 */ 1204, 369, 369, 368, 253, 366, 75, 75, 804, 53, + /* 1550 */ 53, 71, 71, 537, 1196, 537, 126, 126, 537, 1017, + /* 1560 */ 537, 213, 237, 294, 537, 1185, 961, 961, 963, 964, + /* 1570 */ 27, 293, 537, 1184, 537, 72, 72, 127, 127, 1186, + /* 1580 */ 128, 128, 124, 124, 1505, 537, 148, 148, 537, 256, + /* 1590 */ 195, 537, 1270, 537, 147, 147, 132, 132, 537, 11, + /* 1600 */ 537, 215, 537, 199, 343, 345, 347, 131, 131, 154, + /* 1610 */ 129, 129, 156, 130, 130, 74, 74, 537, 370, 1323, + /* 1620 */ 76, 76, 73, 73, 43, 43, 214, 431, 211, 1331, + /* 1630 */ 300, 916, 880, 815, 241, 107, 137, 307, 881, 47, + /* 1640 */ 47, 107, 473, 378, 203, 448, 333, 1403, 1220, 1402, + /* 1650 */ 349, 190, 527, 191, 363, 198, 1508, 1163, 245, 165, + /* 1660 */ 387, 1450, 1448, 1160, 78, 288, 528, 1408, 81, 394, + /* 1670 */ 82, 442, 175, 159, 167, 93, 1328, 35, 1320, 434, + /* 1680 */ 170, 171, 172, 173, 435, 466, 221, 375, 426, 377, + /* 1690 */ 1334, 179, 455, 441, 1397, 225, 87, 36, 461, 1419, + /* 1700 */ 316, 257, 227, 184, 320, 464, 228, 479, 1187, 229, + /* 1710 */ 380, 1240, 1239, 407, 1238, 1212, 834, 332, 1231, 381, + /* 1720 */ 409, 1211, 204, 1210, 1491, 498, 1520, 1281, 92, 281, + /* 1730 */ 1230, 489, 282, 492, 342, 243, 1282, 344, 244, 1280, + /* 1740 */ 346, 412, 1279, 1477, 348, 122, 1476, 517, 10, 357, + /* 1750 */ 286, 1305, 1304, 99, 1383, 94, 501, 251, 1193, 34, + /* 1760 */ 1263, 355, 540, 194, 1262, 361, 362, 1122, 252, 254, + /* 1770 */ 255, 388, 541, 1182, 1177, 151, 1435, 389, 1436, 1434, + /* 1780 */ 1433, 791, 152, 135, 279, 200, 201, 420, 196, 77, + /* 1790 */ 153, 290, 269, 210, 1031, 133, 1029, 945, 166, 155, + /* 1800 */ 217, 168, 866, 306, 220, 1045, 174, 949, 157, 396, + /* 1810 */ 83, 398, 176, 84, 85, 164, 86, 158, 1048, 222, + /* 1820 */ 223, 1044, 144, 18, 224, 315, 1037, 180, 241, 460, + /* 1830 */ 1157, 226, 181, 37, 806, 465, 334, 230, 328, 469, + /* 1840 */ 182, 88, 474, 19, 20, 160, 89, 280, 145, 90, + /* 1850 */ 481, 845, 1110, 146, 997, 205, 1080, 39, 91, 40, + /* 1860 */ 488, 1081, 915, 491, 260, 262, 185, 910, 240, 107, + /* 1870 */ 1100, 1096, 1098, 1104, 21, 1084, 33, 513, 247, 22, + /* 1880 */ 23, 24, 1103, 25, 188, 95, 1012, 998, 996, 26, + /* 1890 */ 1000, 1054, 7, 1053, 1001, 246, 28, 41, 533, 966, + /* 1900 */ 816, 106, 29, 367, 248, 249, 1513, 1512, 364, 1117, + /* 1910 */ 1173, 1173, 876, +}; +static const YYCODETYPE yy_lookahead[] = { + /* 0 */ 187, 187, 187, 216, 217, 187, 206, 187, 264, 265, + /* 10 */ 266, 187, 225, 187, 209, 187, 264, 265, 266, 19, + /* 20 */ 187, 187, 209, 210, 209, 210, 187, 209, 210, 209, + /* 30 */ 210, 31, 209, 209, 210, 209, 210, 285, 224, 39, + /* 40 */ 203, 204, 205, 43, 44, 45, 46, 47, 48, 49, + /* 50 */ 50, 51, 52, 53, 54, 55, 56, 57, 230, 19, + /* 60 */ 181, 182, 183, 184, 230, 245, 233, 208, 189, 245, + /* 70 */ 191, 245, 26, 206, 254, 216, 276, 198, 254, 198, + /* 80 */ 254, 281, 187, 43, 44, 45, 46, 47, 48, 49, + /* 90 */ 50, 51, 52, 53, 54, 55, 56, 57, 259, 99, + /* 100 */ 100, 101, 102, 103, 104, 105, 106, 107, 108, 109, + /* 110 */ 231, 232, 231, 232, 286, 302, 303, 302, 22, 304, + /* 120 */ 302, 303, 76, 244, 11, 244, 86, 19, 88, 248, + /* 130 */ 249, 264, 265, 266, 26, 89, 198, 258, 92, 99, + /* 140 */ 100, 101, 102, 103, 104, 105, 106, 107, 108, 109, + /* 150 */ 105, 43, 44, 45, 46, 47, 48, 49, 50, 51, + /* 160 */ 52, 53, 54, 55, 56, 57, 212, 288, 273, 231, + /* 170 */ 232, 105, 106, 107, 108, 109, 131, 69, 203, 204, + /* 180 */ 205, 136, 244, 99, 100, 101, 102, 103, 104, 105, + /* 190 */ 106, 107, 108, 109, 15, 103, 104, 19, 260, 103, + /* 200 */ 104, 54, 55, 56, 57, 58, 22, 99, 100, 101, + /* 210 */ 102, 103, 104, 105, 106, 107, 108, 109, 264, 265, + /* 220 */ 266, 43, 44, 45, 46, 47, 48, 49, 50, 51, + /* 230 */ 52, 53, 54, 55, 56, 57, 19, 124, 125, 60, + /* 240 */ 148, 24, 150, 59, 187, 67, 99, 100, 101, 102, + /* 250 */ 103, 104, 105, 106, 107, 108, 109, 187, 187, 109, + /* 260 */ 43, 44, 45, 46, 47, 48, 49, 50, 51, 52, + /* 270 */ 53, 54, 55, 56, 57, 204, 205, 99, 100, 101, + /* 280 */ 102, 103, 104, 105, 106, 107, 108, 109, 103, 104, + /* 290 */ 105, 106, 107, 108, 109, 59, 112, 113, 114, 76, + /* 300 */ 231, 232, 187, 19, 19, 22, 23, 23, 54, 55, + /* 310 */ 56, 57, 89, 244, 199, 92, 99, 100, 101, 102, + /* 320 */ 103, 104, 105, 106, 107, 108, 109, 43, 44, 45, + /* 330 */ 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, + /* 340 */ 56, 57, 19, 212, 187, 274, 23, 26, 112, 113, + /* 350 */ 114, 294, 295, 99, 100, 101, 102, 103, 104, 105, + /* 360 */ 106, 107, 108, 109, 59, 295, 43, 44, 45, 46, + /* 370 */ 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, + /* 380 */ 57, 98, 146, 99, 100, 101, 102, 103, 104, 105, + /* 390 */ 106, 107, 108, 109, 109, 264, 265, 266, 187, 187, + /* 400 */ 115, 116, 117, 118, 119, 120, 121, 73, 59, 19, + /* 410 */ 105, 23, 127, 23, 26, 81, 259, 112, 113, 114, + /* 420 */ 187, 72, 99, 100, 101, 102, 103, 104, 105, 106, + /* 430 */ 107, 108, 109, 43, 44, 45, 46, 47, 48, 49, + /* 440 */ 50, 51, 52, 53, 54, 55, 56, 57, 19, 124, + /* 450 */ 125, 182, 23, 184, 187, 134, 135, 123, 189, 131, + /* 460 */ 191, 112, 113, 114, 136, 187, 233, 198, 134, 135, + /* 470 */ 198, 259, 43, 44, 45, 46, 47, 48, 49, 50, + /* 480 */ 51, 52, 53, 54, 55, 56, 57, 209, 210, 99, + /* 490 */ 100, 101, 102, 103, 104, 105, 106, 107, 108, 109, + /* 500 */ 231, 232, 206, 231, 232, 187, 59, 296, 297, 76, + /* 510 */ 160, 161, 301, 244, 232, 19, 244, 297, 59, 23, + /* 520 */ 87, 301, 89, 245, 26, 92, 244, 258, 99, 100, + /* 530 */ 101, 102, 103, 104, 105, 106, 107, 108, 109, 43, + /* 540 */ 44, 45, 46, 47, 48, 49, 50, 51, 52, 53, + /* 550 */ 54, 55, 56, 57, 19, 187, 97, 288, 23, 112, + /* 560 */ 113, 114, 115, 296, 297, 118, 119, 120, 301, 108, + /* 570 */ 109, 112, 113, 255, 141, 128, 117, 281, 43, 44, + /* 580 */ 45, 46, 47, 48, 49, 50, 51, 52, 53, 54, + /* 590 */ 55, 56, 57, 187, 187, 99, 100, 101, 102, 103, + /* 600 */ 104, 105, 106, 107, 108, 109, 26, 148, 149, 150, + /* 610 */ 115, 97, 59, 118, 119, 120, 209, 210, 73, 59, + /* 620 */ 122, 19, 209, 128, 256, 72, 187, 113, 187, 81, + /* 630 */ 223, 117, 227, 228, 99, 100, 101, 102, 103, 104, + /* 640 */ 105, 106, 107, 108, 109, 43, 44, 45, 46, 47, + /* 650 */ 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, + /* 660 */ 19, 255, 148, 149, 150, 112, 113, 114, 123, 124, + /* 670 */ 125, 230, 112, 113, 114, 22, 297, 22, 81, 22, + /* 680 */ 301, 59, 134, 135, 43, 44, 45, 46, 47, 48, + /* 690 */ 49, 50, 51, 52, 53, 54, 55, 56, 57, 139, + /* 700 */ 192, 99, 100, 101, 102, 103, 104, 105, 106, 107, + /* 710 */ 108, 109, 59, 116, 59, 187, 59, 231, 232, 46, + /* 720 */ 47, 48, 49, 16, 12, 145, 198, 22, 187, 187, + /* 730 */ 244, 134, 135, 222, 112, 113, 114, 209, 210, 27, + /* 740 */ 99, 100, 101, 102, 103, 104, 105, 106, 107, 108, + /* 750 */ 109, 209, 210, 187, 42, 187, 154, 227, 228, 231, + /* 760 */ 232, 139, 22, 23, 59, 112, 113, 114, 113, 112, + /* 770 */ 113, 114, 244, 245, 233, 63, 19, 209, 210, 271, + /* 780 */ 187, 24, 254, 275, 77, 73, 79, 245, 195, 260, + /* 790 */ 117, 223, 199, 22, 23, 154, 19, 26, 22, 187, + /* 800 */ 43, 44, 45, 46, 47, 48, 49, 50, 51, 52, + /* 810 */ 53, 54, 55, 56, 57, 19, 187, 112, 113, 114, + /* 820 */ 43, 44, 45, 46, 47, 48, 49, 50, 51, 52, + /* 830 */ 53, 54, 55, 56, 57, 59, 263, 187, 98, 43, + /* 840 */ 44, 45, 46, 47, 48, 49, 50, 51, 52, 53, + /* 850 */ 54, 55, 56, 57, 204, 205, 99, 100, 101, 102, + /* 860 */ 103, 104, 105, 106, 107, 108, 109, 255, 130, 98, + /* 870 */ 132, 133, 299, 300, 198, 187, 99, 100, 101, 102, + /* 880 */ 103, 104, 105, 106, 107, 108, 109, 187, 112, 113, + /* 890 */ 114, 187, 241, 187, 243, 99, 100, 101, 102, 103, + /* 900 */ 104, 105, 106, 107, 108, 109, 19, 231, 232, 209, + /* 910 */ 210, 282, 59, 209, 210, 209, 210, 16, 230, 19, + /* 920 */ 244, 22, 23, 223, 274, 26, 19, 223, 187, 223, + /* 930 */ 198, 44, 45, 46, 47, 48, 49, 50, 51, 52, + /* 940 */ 53, 54, 55, 56, 57, 45, 46, 47, 48, 49, + /* 950 */ 50, 51, 52, 53, 54, 55, 56, 57, 19, 20, + /* 960 */ 187, 22, 187, 231, 232, 112, 113, 114, 123, 124, + /* 970 */ 125, 7, 8, 9, 187, 36, 244, 24, 77, 21, + /* 980 */ 79, 187, 209, 210, 187, 263, 99, 100, 101, 102, + /* 990 */ 103, 104, 105, 106, 107, 108, 109, 98, 59, 99, + /* 1000 */ 100, 101, 102, 103, 104, 105, 106, 107, 108, 109, + /* 1010 */ 71, 187, 59, 187, 187, 19, 20, 187, 22, 112, + /* 1020 */ 81, 299, 300, 282, 230, 199, 291, 292, 22, 187, + /* 1030 */ 24, 256, 36, 209, 210, 187, 97, 35, 80, 209, + /* 1040 */ 210, 268, 103, 104, 48, 187, 220, 223, 222, 110, + /* 1050 */ 59, 112, 113, 114, 187, 59, 117, 156, 187, 179, + /* 1060 */ 180, 181, 182, 183, 184, 59, 113, 71, 66, 189, + /* 1070 */ 22, 191, 230, 134, 135, 245, 74, 119, 198, 282, + /* 1080 */ 297, 85, 224, 198, 301, 187, 90, 148, 149, 150, + /* 1090 */ 151, 152, 19, 97, 103, 104, 123, 124, 125, 103, + /* 1100 */ 104, 53, 111, 112, 113, 114, 110, 116, 112, 113, + /* 1110 */ 114, 231, 232, 117, 156, 124, 231, 232, 297, 113, + /* 1120 */ 187, 24, 301, 256, 244, 201, 202, 256, 126, 244, + /* 1130 */ 187, 198, 187, 187, 23, 187, 187, 26, 258, 148, + /* 1140 */ 19, 150, 209, 210, 148, 149, 150, 151, 152, 0, + /* 1150 */ 1, 2, 209, 210, 5, 209, 210, 209, 210, 10, + /* 1160 */ 11, 12, 13, 14, 231, 232, 17, 46, 288, 19, + /* 1170 */ 20, 223, 22, 236, 198, 66, 187, 244, 245, 30, + /* 1180 */ 12, 32, 198, 246, 59, 112, 36, 187, 245, 40, + /* 1190 */ 198, 245, 117, 29, 85, 27, 26, 33, 209, 210, + /* 1200 */ 297, 76, 127, 94, 301, 256, 26, 231, 232, 59, + /* 1210 */ 42, 153, 87, 155, 89, 231, 232, 92, 31, 70, + /* 1220 */ 244, 71, 26, 231, 232, 114, 39, 78, 244, 65, + /* 1230 */ 81, 63, 111, 233, 137, 85, 244, 112, 113, 114, + /* 1240 */ 90, 22, 59, 24, 95, 201, 202, 97, 127, 187, + /* 1250 */ 139, 142, 187, 103, 104, 19, 20, 187, 22, 187, + /* 1260 */ 110, 187, 112, 113, 114, 187, 141, 117, 141, 187, + /* 1270 */ 23, 187, 36, 26, 209, 210, 134, 135, 129, 209, + /* 1280 */ 210, 209, 210, 134, 135, 22, 159, 209, 210, 187, + /* 1290 */ 187, 209, 210, 209, 210, 59, 113, 187, 148, 149, + /* 1300 */ 150, 151, 152, 289, 290, 187, 157, 71, 248, 249, + /* 1310 */ 114, 141, 209, 210, 46, 125, 116, 117, 138, 19, + /* 1320 */ 20, 85, 22, 148, 61, 150, 90, 209, 210, 23, + /* 1330 */ 7, 8, 26, 97, 187, 139, 36, 147, 59, 103, + /* 1340 */ 104, 19, 20, 187, 22, 59, 110, 187, 112, 113, + /* 1350 */ 114, 1, 2, 117, 187, 5, 209, 210, 36, 59, + /* 1360 */ 10, 11, 12, 13, 14, 209, 210, 17, 59, 209, + /* 1370 */ 210, 71, 187, 148, 250, 150, 209, 210, 187, 111, + /* 1380 */ 30, 59, 32, 22, 148, 149, 150, 151, 152, 187, + /* 1390 */ 40, 187, 113, 71, 209, 210, 187, 97, 187, 113, + /* 1400 */ 209, 210, 187, 103, 104, 105, 23, 187, 187, 26, + /* 1410 */ 110, 187, 112, 113, 114, 83, 84, 117, 23, 97, + /* 1420 */ 70, 26, 113, 218, 187, 103, 104, 187, 78, 209, + /* 1430 */ 210, 81, 110, 187, 112, 113, 114, 19, 20, 117, + /* 1440 */ 22, 187, 187, 187, 187, 95, 209, 210, 148, 149, + /* 1450 */ 150, 151, 152, 187, 36, 23, 187, 187, 26, 187, + /* 1460 */ 187, 187, 187, 209, 210, 209, 210, 187, 187, 218, + /* 1470 */ 148, 149, 150, 151, 152, 209, 210, 59, 187, 129, + /* 1480 */ 187, 209, 210, 187, 134, 135, 187, 306, 187, 71, + /* 1490 */ 209, 210, 23, 228, 187, 26, 23, 187, 137, 26, + /* 1500 */ 209, 210, 209, 210, 187, 209, 210, 157, 209, 210, + /* 1510 */ 209, 210, 218, 187, 187, 97, 209, 210, 187, 278, + /* 1520 */ 23, 103, 104, 26, 187, 187, 187, 187, 110, 187, + /* 1530 */ 112, 113, 114, 187, 187, 117, 5, 247, 187, 187, + /* 1540 */ 187, 10, 11, 12, 13, 14, 209, 210, 17, 209, + /* 1550 */ 210, 209, 210, 187, 187, 187, 209, 210, 187, 23, + /* 1560 */ 187, 30, 26, 32, 187, 187, 148, 149, 150, 151, + /* 1570 */ 152, 40, 187, 187, 187, 209, 210, 209, 210, 187, + /* 1580 */ 209, 210, 209, 210, 187, 187, 209, 210, 187, 277, + /* 1590 */ 234, 187, 247, 187, 209, 210, 209, 210, 187, 235, + /* 1600 */ 187, 70, 187, 207, 247, 247, 247, 209, 210, 78, + /* 1610 */ 209, 210, 81, 209, 210, 209, 210, 187, 185, 238, + /* 1620 */ 209, 210, 209, 210, 209, 210, 95, 251, 287, 238, + /* 1630 */ 251, 23, 23, 23, 26, 26, 26, 283, 23, 209, + /* 1640 */ 210, 26, 213, 238, 221, 283, 212, 212, 217, 212, + /* 1650 */ 251, 241, 270, 241, 237, 235, 190, 60, 137, 287, + /* 1660 */ 129, 194, 194, 38, 284, 134, 135, 273, 284, 194, + /* 1670 */ 146, 111, 22, 43, 226, 145, 262, 261, 238, 18, + /* 1680 */ 229, 229, 229, 229, 194, 18, 193, 238, 157, 262, + /* 1690 */ 226, 226, 194, 238, 238, 193, 153, 261, 62, 280, + /* 1700 */ 279, 194, 193, 22, 194, 214, 193, 111, 194, 193, + /* 1710 */ 214, 211, 211, 64, 211, 211, 122, 211, 219, 214, + /* 1720 */ 109, 213, 160, 211, 300, 140, 211, 253, 111, 272, + /* 1730 */ 219, 214, 272, 214, 252, 194, 253, 252, 91, 253, + /* 1740 */ 252, 82, 253, 305, 252, 144, 305, 141, 22, 194, + /* 1750 */ 269, 257, 257, 153, 267, 143, 142, 25, 197, 26, + /* 1760 */ 242, 241, 196, 240, 242, 239, 238, 13, 188, 188, + /* 1770 */ 6, 293, 186, 186, 186, 200, 206, 293, 206, 206, + /* 1780 */ 206, 4, 200, 215, 215, 207, 207, 3, 22, 206, + /* 1790 */ 200, 158, 96, 15, 23, 16, 23, 135, 146, 126, + /* 1800 */ 24, 138, 20, 16, 140, 1, 138, 147, 126, 61, + /* 1810 */ 53, 37, 146, 53, 53, 290, 53, 126, 112, 34, + /* 1820 */ 137, 1, 5, 22, 111, 156, 68, 68, 26, 41, + /* 1830 */ 75, 137, 111, 24, 20, 19, 127, 121, 23, 67, + /* 1840 */ 22, 22, 67, 22, 22, 37, 22, 67, 23, 145, + /* 1850 */ 22, 28, 23, 23, 23, 137, 23, 22, 26, 22, + /* 1860 */ 24, 23, 112, 24, 23, 23, 22, 139, 34, 26, + /* 1870 */ 75, 88, 86, 75, 34, 23, 22, 24, 22, 34, + /* 1880 */ 34, 34, 93, 34, 26, 26, 23, 23, 23, 34, + /* 1890 */ 23, 23, 44, 23, 11, 26, 22, 22, 26, 23, + /* 1900 */ 23, 22, 22, 15, 137, 137, 137, 137, 23, 1, + /* 1910 */ 307, 307, 131, 307, 307, 307, 307, 307, 307, 307, + /* 1920 */ 307, 307, 307, 307, 307, 307, 307, 307, 307, 307, + /* 1930 */ 307, 307, 307, 307, 307, 307, 307, 307, 307, 307, + /* 1940 */ 307, 307, 307, 307, 307, 307, 307, 307, 307, 307, + /* 1950 */ 307, 307, 307, 307, 307, 307, 307, 307, 307, 307, + /* 1960 */ 307, 307, 307, 307, 307, 307, 307, 307, 307, 307, + /* 1970 */ 307, 307, 307, 307, 307, 307, 307, 307, 307, 307, + /* 1980 */ 307, 307, 307, 307, 307, 307, 307, 307, 307, 307, + /* 1990 */ 307, 307, 307, 307, 307, 307, 307, 307, 307, 307, + /* 2000 */ 307, 307, 307, 307, 307, 307, 307, 307, 307, 307, + /* 2010 */ 307, 307, 307, 307, 307, 307, 307, 307, 307, 307, + /* 2020 */ 307, 307, 307, 307, 307, 307, 307, 307, 307, 307, + /* 2030 */ 307, 307, 307, 307, 307, 307, 307, 307, 307, 307, + /* 2040 */ 307, 307, 307, 307, 307, 307, 307, 307, 307, 307, + /* 2050 */ 307, 307, 307, 307, 307, 307, 307, 307, 307, 307, + /* 2060 */ 307, 307, 307, 307, 307, 307, 307, 307, 307, 307, + /* 2070 */ 307, 307, 307, 307, 307, 307, 307, 307, 307, 307, + /* 2080 */ 307, 307, 307, 307, 307, 307, 307, 307, 307, 307, + /* 2090 */ 307, 307, +}; +#define YY_SHIFT_COUNT (542) +#define YY_SHIFT_MIN (0) +#define YY_SHIFT_MAX (1908) +static const unsigned short int yy_shift_ofst[] = { + /* 0 */ 1350, 1149, 1531, 939, 939, 548, 996, 1150, 1236, 1322, + /* 10 */ 1322, 1322, 334, 0, 0, 178, 777, 1322, 1322, 1322, + /* 20 */ 1322, 1322, 1322, 1322, 1322, 1322, 1322, 1322, 1322, 1322, + /* 30 */ 991, 991, 1125, 1125, 447, 597, 548, 548, 548, 548, + /* 40 */ 548, 548, 40, 108, 217, 284, 323, 390, 429, 496, + /* 50 */ 535, 602, 641, 757, 777, 777, 777, 777, 777, 777, + /* 60 */ 777, 777, 777, 777, 777, 777, 777, 777, 777, 777, + /* 70 */ 777, 777, 796, 777, 887, 900, 900, 1300, 1322, 1322, + /* 80 */ 1322, 1322, 1322, 1322, 1322, 1322, 1322, 1322, 1322, 1322, + /* 90 */ 1322, 1322, 1322, 1322, 1322, 1322, 1322, 1322, 1322, 1322, + /* 100 */ 1322, 1322, 1322, 1322, 1322, 1322, 1322, 1322, 1322, 1322, + /* 110 */ 1418, 1322, 1322, 1322, 1322, 1322, 1322, 1322, 1322, 1322, + /* 120 */ 1322, 1322, 1322, 1322, 147, 254, 254, 254, 254, 254, + /* 130 */ 84, 185, 66, 853, 958, 1121, 853, 92, 92, 853, + /* 140 */ 321, 321, 321, 321, 325, 350, 350, 461, 150, 1913, + /* 150 */ 1913, 285, 285, 285, 236, 184, 349, 184, 184, 712, + /* 160 */ 712, 433, 553, 771, 899, 853, 853, 853, 853, 853, + /* 170 */ 853, 853, 853, 853, 853, 853, 853, 853, 853, 853, + /* 180 */ 853, 853, 853, 853, 853, 853, 46, 46, 853, 113, + /* 190 */ 223, 223, 1183, 1183, 1127, 1142, 1913, 1913, 1913, 459, + /* 200 */ 514, 514, 653, 495, 657, 305, 705, 560, 622, 776, + /* 210 */ 853, 853, 853, 853, 853, 853, 853, 853, 853, 545, + /* 220 */ 853, 853, 853, 853, 853, 853, 853, 853, 853, 853, + /* 230 */ 853, 853, 1002, 1002, 1002, 853, 853, 853, 853, 1111, + /* 240 */ 853, 853, 853, 1006, 1109, 853, 853, 1168, 853, 853, + /* 250 */ 853, 853, 853, 853, 853, 853, 845, 1164, 738, 953, + /* 260 */ 953, 953, 953, 1196, 738, 738, 45, 96, 964, 179, + /* 270 */ 580, 907, 907, 1073, 580, 580, 1073, 498, 388, 1268, + /* 280 */ 1187, 1187, 1187, 907, 1170, 1170, 1058, 1180, 328, 1219, + /* 290 */ 1597, 1521, 1521, 1625, 1625, 1521, 1524, 1560, 1650, 1630, + /* 300 */ 1530, 1661, 1661, 1661, 1661, 1521, 1667, 1530, 1530, 1560, + /* 310 */ 1650, 1630, 1630, 1530, 1521, 1667, 1543, 1636, 1521, 1667, + /* 320 */ 1681, 1521, 1667, 1521, 1667, 1681, 1596, 1596, 1596, 1649, + /* 330 */ 1681, 1596, 1594, 1596, 1649, 1596, 1596, 1562, 1681, 1611, + /* 340 */ 1611, 1681, 1585, 1617, 1585, 1617, 1585, 1617, 1585, 1617, + /* 350 */ 1521, 1647, 1647, 1659, 1659, 1601, 1606, 1726, 1521, 1600, + /* 360 */ 1601, 1612, 1614, 1530, 1732, 1733, 1754, 1754, 1764, 1764, + /* 370 */ 1764, 1913, 1913, 1913, 1913, 1913, 1913, 1913, 1913, 1913, + /* 380 */ 1913, 1913, 1913, 1913, 1913, 1913, 673, 901, 283, 740, + /* 390 */ 707, 973, 655, 1247, 1048, 1097, 1190, 1306, 1263, 1383, + /* 400 */ 1395, 1432, 1469, 1473, 1497, 1279, 1200, 1323, 1075, 1286, + /* 410 */ 1536, 1608, 1332, 1609, 1175, 1225, 1610, 1615, 1309, 1361, + /* 420 */ 1777, 1784, 1766, 1633, 1778, 1696, 1779, 1771, 1773, 1662, + /* 430 */ 1652, 1673, 1776, 1663, 1782, 1664, 1787, 1804, 1668, 1660, + /* 440 */ 1682, 1748, 1774, 1666, 1757, 1760, 1761, 1763, 1691, 1706, + /* 450 */ 1785, 1683, 1820, 1817, 1801, 1713, 1669, 1758, 1802, 1759, + /* 460 */ 1755, 1788, 1694, 1721, 1809, 1814, 1816, 1709, 1716, 1818, + /* 470 */ 1772, 1819, 1821, 1815, 1822, 1775, 1823, 1824, 1780, 1808, + /* 480 */ 1825, 1704, 1828, 1829, 1830, 1831, 1832, 1833, 1835, 1836, + /* 490 */ 1838, 1837, 1839, 1718, 1841, 1842, 1750, 1834, 1844, 1728, + /* 500 */ 1843, 1840, 1845, 1846, 1847, 1783, 1795, 1786, 1848, 1798, + /* 510 */ 1789, 1849, 1852, 1854, 1853, 1858, 1859, 1855, 1863, 1843, + /* 520 */ 1864, 1865, 1867, 1868, 1869, 1870, 1856, 1883, 1874, 1875, + /* 530 */ 1876, 1877, 1879, 1880, 1872, 1781, 1767, 1768, 1769, 1770, + /* 540 */ 1885, 1888, 1908, +}; +#define YY_REDUCE_COUNT (385) +#define YY_REDUCE_MIN (-256) +#define YY_REDUCE_MAX (1590) +static const short yy_reduce_ofst[] = { + /* 0 */ 880, -121, 269, 528, 933, -119, -187, -185, -182, -180, + /* 10 */ -176, -174, -62, -46, 131, -248, -133, 407, 568, 700, + /* 20 */ 704, 278, 706, 824, 542, 830, 948, 773, 943, 946, + /* 30 */ 71, 650, 211, 267, 826, 272, 676, 732, 885, 976, + /* 40 */ 984, 992, -256, -256, -256, -256, -256, -256, -256, -256, + /* 50 */ -256, -256, -256, -256, -256, -256, -256, -256, -256, -256, + /* 60 */ -256, -256, -256, -256, -256, -256, -256, -256, -256, -256, + /* 70 */ -256, -256, -256, -256, -256, -256, -256, 989, 1065, 1070, + /* 80 */ 1072, 1078, 1082, 1084, 1103, 1118, 1147, 1156, 1160, 1167, + /* 90 */ 1185, 1191, 1220, 1237, 1254, 1256, 1266, 1272, 1281, 1291, + /* 100 */ 1293, 1296, 1299, 1301, 1307, 1337, 1340, 1342, 1347, 1366, + /* 110 */ 1368, 1371, 1373, 1377, 1385, 1387, 1398, 1401, 1404, 1406, + /* 120 */ 1411, 1413, 1415, 1430, -256, -256, -256, -256, -256, -256, + /* 130 */ -256, -256, -256, -172, 508, -213, 57, -163, -25, 593, + /* 140 */ 69, 486, 69, 486, -200, 573, 722, -256, -256, -256, + /* 150 */ -256, -141, -141, -141, -105, -161, -167, 157, 212, 405, + /* 160 */ 530, 220, 233, 735, 735, 115, 318, 406, 612, 541, + /* 170 */ -166, 441, 688, 794, 629, 368, 741, 775, 867, 797, + /* 180 */ 871, 842, -186, 1000, 858, 949, 379, 783, 70, 296, + /* 190 */ 821, 903, 924, 1044, 651, 282, 1014, 1060, 937, -195, + /* 200 */ -177, 413, 439, 511, 566, 787, 827, 848, 898, 945, + /* 210 */ 1062, 1074, 1102, 1110, 1202, 1204, 1209, 1211, 1215, 529, + /* 220 */ 1221, 1224, 1240, 1246, 1255, 1257, 1269, 1270, 1273, 1274, + /* 230 */ 1275, 1280, 1205, 1251, 1294, 1310, 1317, 1326, 1327, 1124, + /* 240 */ 1331, 1338, 1339, 1290, 1181, 1346, 1351, 1265, 1352, 787, + /* 250 */ 1353, 1367, 1378, 1386, 1392, 1397, 1241, 1312, 1356, 1345, + /* 260 */ 1357, 1358, 1359, 1124, 1356, 1356, 1364, 1396, 1433, 1341, + /* 270 */ 1381, 1376, 1379, 1354, 1391, 1405, 1362, 1429, 1423, 1431, + /* 280 */ 1434, 1435, 1437, 1399, 1410, 1412, 1382, 1417, 1420, 1466, + /* 290 */ 1372, 1467, 1468, 1380, 1384, 1475, 1394, 1414, 1416, 1448, + /* 300 */ 1440, 1451, 1452, 1453, 1454, 1490, 1493, 1449, 1455, 1427, + /* 310 */ 1436, 1464, 1465, 1456, 1498, 1502, 1419, 1421, 1507, 1509, + /* 320 */ 1491, 1510, 1513, 1514, 1516, 1496, 1500, 1501, 1503, 1499, + /* 330 */ 1505, 1504, 1508, 1506, 1511, 1512, 1515, 1424, 1517, 1457, + /* 340 */ 1460, 1519, 1474, 1482, 1483, 1485, 1486, 1488, 1489, 1492, + /* 350 */ 1541, 1438, 1441, 1494, 1495, 1518, 1520, 1487, 1555, 1481, + /* 360 */ 1522, 1523, 1526, 1528, 1561, 1566, 1580, 1581, 1586, 1587, + /* 370 */ 1588, 1478, 1484, 1525, 1575, 1570, 1572, 1573, 1574, 1582, + /* 380 */ 1568, 1569, 1578, 1579, 1583, 1590, +}; +static const YYACTIONTYPE yy_default[] = { + /* 0 */ 1554, 1554, 1554, 1392, 1171, 1278, 1171, 1171, 1171, 1392, + /* 10 */ 1392, 1392, 1171, 1308, 1308, 1445, 1202, 1171, 1171, 1171, + /* 20 */ 1171, 1171, 1171, 1171, 1171, 1171, 1171, 1391, 1171, 1171, + /* 30 */ 1171, 1171, 1475, 1475, 1171, 1171, 1171, 1171, 1171, 1171, + /* 40 */ 1171, 1171, 1171, 1317, 1171, 1171, 1171, 1171, 1171, 1393, + /* 50 */ 1394, 1171, 1171, 1171, 1444, 1446, 1409, 1327, 1326, 1325, + /* 60 */ 1324, 1427, 1295, 1322, 1315, 1319, 1387, 1388, 1386, 1390, + /* 70 */ 1394, 1393, 1171, 1318, 1358, 1372, 1357, 1171, 1171, 1171, + /* 80 */ 1171, 1171, 1171, 1171, 1171, 1171, 1171, 1171, 1171, 1171, + /* 90 */ 1171, 1171, 1171, 1171, 1171, 1171, 1171, 1171, 1171, 1171, + /* 100 */ 1171, 1171, 1171, 1171, 1171, 1171, 1171, 1171, 1171, 1171, + /* 110 */ 1171, 1171, 1171, 1171, 1171, 1171, 1171, 1171, 1171, 1171, + /* 120 */ 1171, 1171, 1171, 1171, 1366, 1371, 1377, 1370, 1367, 1360, + /* 130 */ 1359, 1361, 1362, 1171, 1192, 1242, 1171, 1171, 1171, 1171, + /* 140 */ 1463, 1462, 1171, 1171, 1202, 1352, 1351, 1363, 1364, 1374, + /* 150 */ 1373, 1452, 1510, 1509, 1410, 1171, 1171, 1171, 1171, 1171, + /* 160 */ 1171, 1475, 1171, 1171, 1171, 1171, 1171, 1171, 1171, 1171, + /* 170 */ 1171, 1171, 1171, 1171, 1171, 1171, 1171, 1171, 1171, 1171, + /* 180 */ 1171, 1171, 1171, 1171, 1171, 1171, 1475, 1475, 1171, 1202, + /* 190 */ 1475, 1475, 1198, 1198, 1302, 1171, 1458, 1278, 1269, 1171, + /* 200 */ 1171, 1171, 1171, 1171, 1171, 1171, 1171, 1171, 1171, 1171, + /* 210 */ 1171, 1171, 1171, 1449, 1447, 1171, 1171, 1171, 1171, 1171, + /* 220 */ 1171, 1171, 1171, 1171, 1171, 1171, 1171, 1171, 1171, 1171, + /* 230 */ 1171, 1171, 1171, 1171, 1171, 1171, 1171, 1171, 1171, 1171, + /* 240 */ 1171, 1171, 1171, 1274, 1171, 1171, 1171, 1171, 1171, 1171, + /* 250 */ 1171, 1171, 1171, 1171, 1171, 1504, 1171, 1422, 1256, 1274, + /* 260 */ 1274, 1274, 1274, 1276, 1257, 1255, 1268, 1203, 1178, 1546, + /* 270 */ 1321, 1297, 1297, 1543, 1321, 1321, 1543, 1217, 1524, 1214, + /* 280 */ 1308, 1308, 1308, 1297, 1302, 1302, 1389, 1275, 1268, 1171, + /* 290 */ 1546, 1283, 1283, 1545, 1545, 1283, 1410, 1330, 1336, 1245, + /* 300 */ 1321, 1251, 1251, 1251, 1251, 1283, 1189, 1321, 1321, 1330, + /* 310 */ 1336, 1245, 1245, 1321, 1283, 1189, 1426, 1540, 1283, 1189, + /* 320 */ 1400, 1283, 1189, 1283, 1189, 1400, 1243, 1243, 1243, 1232, + /* 330 */ 1400, 1243, 1217, 1243, 1232, 1243, 1243, 1493, 1400, 1404, + /* 340 */ 1404, 1400, 1301, 1296, 1301, 1296, 1301, 1296, 1301, 1296, + /* 350 */ 1283, 1485, 1485, 1311, 1311, 1316, 1302, 1395, 1283, 1171, + /* 360 */ 1316, 1314, 1312, 1321, 1195, 1235, 1507, 1507, 1503, 1503, + /* 370 */ 1503, 1551, 1551, 1458, 1519, 1202, 1202, 1202, 1202, 1519, + /* 380 */ 1219, 1219, 1203, 1203, 1202, 1519, 1171, 1171, 1171, 1171, + /* 390 */ 1171, 1171, 1514, 1171, 1411, 1287, 1171, 1171, 1171, 1171, + /* 400 */ 1171, 1171, 1171, 1171, 1171, 1171, 1171, 1171, 1171, 1171, + /* 410 */ 1171, 1171, 1171, 1171, 1171, 1171, 1171, 1171, 1171, 1341, + /* 420 */ 1171, 1174, 1455, 1171, 1171, 1453, 1171, 1171, 1171, 1171, + /* 430 */ 1171, 1171, 1288, 1171, 1171, 1171, 1171, 1171, 1171, 1171, + /* 440 */ 1171, 1171, 1171, 1171, 1171, 1171, 1171, 1171, 1171, 1171, + /* 450 */ 1171, 1542, 1171, 1171, 1171, 1171, 1171, 1171, 1425, 1424, + /* 460 */ 1171, 1171, 1285, 1171, 1171, 1171, 1171, 1171, 1171, 1171, + /* 470 */ 1171, 1171, 1171, 1171, 1171, 1171, 1171, 1171, 1171, 1171, + /* 480 */ 1171, 1171, 1171, 1171, 1171, 1171, 1171, 1171, 1171, 1171, + /* 490 */ 1171, 1171, 1171, 1171, 1171, 1171, 1171, 1171, 1171, 1171, + /* 500 */ 1313, 1171, 1171, 1171, 1171, 1171, 1171, 1171, 1171, 1171, + /* 510 */ 1171, 1171, 1171, 1171, 1171, 1490, 1303, 1171, 1171, 1533, + /* 520 */ 1171, 1171, 1171, 1171, 1171, 1171, 1171, 1171, 1171, 1171, + /* 530 */ 1171, 1171, 1171, 1171, 1528, 1259, 1343, 1171, 1342, 1346, + /* 540 */ 1171, 1183, 1171, +}; +/********** End of lemon-generated parsing tables *****************************/ + +/* The next table maps tokens (terminal symbols) into fallback tokens. +** If a construct like the following: +** +** %fallback ID X Y Z. +** +** appears in the grammar, then ID becomes a fallback token for X, Y, +** and Z. Whenever one of the tokens X, Y, or Z is input to the parser +** but it does not parse, the type of the token is changed to ID and +** the parse is retried before an error is thrown. +** +** This feature can be used, for example, to cause some keywords in a language +** to revert to identifiers if they keyword does not apply in the context where +** it appears. +*/ +#ifdef YYFALLBACK +static const YYCODETYPE yyFallback[] = { + 0, /* $ => nothing */ + 0, /* SEMI => nothing */ + 59, /* EXPLAIN => ID */ + 59, /* QUERY => ID */ + 59, /* PLAN => ID */ + 59, /* BEGIN => ID */ + 0, /* TRANSACTION => nothing */ + 59, /* DEFERRED => ID */ + 59, /* IMMEDIATE => ID */ + 59, /* EXCLUSIVE => ID */ + 0, /* COMMIT => nothing */ + 59, /* END => ID */ + 59, /* ROLLBACK => ID */ + 59, /* SAVEPOINT => ID */ + 59, /* RELEASE => ID */ + 0, /* TO => nothing */ + 0, /* TABLE => nothing */ + 0, /* CREATE => nothing */ + 59, /* IF => ID */ + 0, /* NOT => nothing */ + 0, /* EXISTS => nothing */ + 59, /* TEMP => ID */ + 0, /* LP => nothing */ + 0, /* RP => nothing */ + 0, /* AS => nothing */ + 59, /* WITHOUT => ID */ + 0, /* COMMA => nothing */ + 59, /* ABORT => ID */ + 59, /* ACTION => ID */ + 59, /* AFTER => ID */ + 59, /* ANALYZE => ID */ + 59, /* ASC => ID */ + 59, /* ATTACH => ID */ + 59, /* BEFORE => ID */ + 59, /* BY => ID */ + 59, /* CASCADE => ID */ + 59, /* CAST => ID */ + 59, /* CONFLICT => ID */ + 59, /* DATABASE => ID */ + 59, /* DESC => ID */ + 59, /* DETACH => ID */ + 59, /* EACH => ID */ + 59, /* FAIL => ID */ + 0, /* OR => nothing */ + 0, /* AND => nothing */ + 0, /* IS => nothing */ + 59, /* MATCH => ID */ + 59, /* LIKE_KW => ID */ + 0, /* BETWEEN => nothing */ + 0, /* IN => nothing */ + 0, /* ISNULL => nothing */ + 0, /* NOTNULL => nothing */ + 0, /* NE => nothing */ + 0, /* EQ => nothing */ + 0, /* GT => nothing */ + 0, /* LE => nothing */ + 0, /* LT => nothing */ + 0, /* GE => nothing */ + 0, /* ESCAPE => nothing */ + 0, /* ID => nothing */ + 59, /* COLUMNKW => ID */ + 59, /* DO => ID */ + 59, /* FOR => ID */ + 59, /* IGNORE => ID */ + 59, /* INITIALLY => ID */ + 59, /* INSTEAD => ID */ + 59, /* NO => ID */ + 59, /* KEY => ID */ + 59, /* OF => ID */ + 59, /* OFFSET => ID */ + 59, /* PRAGMA => ID */ + 59, /* RAISE => ID */ + 59, /* RECURSIVE => ID */ + 59, /* REPLACE => ID */ + 59, /* RESTRICT => ID */ + 59, /* ROW => ID */ + 59, /* ROWS => ID */ + 59, /* TRIGGER => ID */ + 59, /* VACUUM => ID */ + 59, /* VIEW => ID */ + 59, /* VIRTUAL => ID */ + 59, /* WITH => ID */ + 59, /* NULLS => ID */ + 59, /* FIRST => ID */ + 59, /* LAST => ID */ + 59, /* CURRENT => ID */ + 59, /* FOLLOWING => ID */ + 59, /* PARTITION => ID */ + 59, /* PRECEDING => ID */ + 59, /* RANGE => ID */ + 59, /* UNBOUNDED => ID */ + 59, /* EXCLUDE => ID */ + 59, /* GROUPS => ID */ + 59, /* OTHERS => ID */ + 59, /* TIES => ID */ + 59, /* REINDEX => ID */ + 59, /* RENAME => ID */ + 59, /* CTIME_KW => ID */ + 0, /* ANY => nothing */ + 0, /* BITAND => nothing */ + 0, /* BITOR => nothing */ + 0, /* LSHIFT => nothing */ + 0, /* RSHIFT => nothing */ + 0, /* PLUS => nothing */ + 0, /* MINUS => nothing */ + 0, /* STAR => nothing */ + 0, /* SLASH => nothing */ + 0, /* REM => nothing */ + 0, /* CONCAT => nothing */ + 0, /* COLLATE => nothing */ + 0, /* BITNOT => nothing */ + 0, /* ON => nothing */ + 0, /* INDEXED => nothing */ + 0, /* STRING => nothing */ + 0, /* JOIN_KW => nothing */ + 0, /* CONSTRAINT => nothing */ + 0, /* DEFAULT => nothing */ + 0, /* NULL => nothing */ + 0, /* PRIMARY => nothing */ + 0, /* UNIQUE => nothing */ + 0, /* CHECK => nothing */ + 0, /* REFERENCES => nothing */ + 0, /* AUTOINCR => nothing */ + 0, /* INSERT => nothing */ + 0, /* DELETE => nothing */ + 0, /* UPDATE => nothing */ + 0, /* SET => nothing */ + 0, /* DEFERRABLE => nothing */ + 0, /* FOREIGN => nothing */ + 0, /* DROP => nothing */ + 0, /* UNION => nothing */ + 0, /* ALL => nothing */ + 0, /* EXCEPT => nothing */ + 0, /* INTERSECT => nothing */ + 0, /* SELECT => nothing */ + 0, /* VALUES => nothing */ + 0, /* DISTINCT => nothing */ + 0, /* DOT => nothing */ + 0, /* FROM => nothing */ + 0, /* JOIN => nothing */ + 0, /* USING => nothing */ + 0, /* ORDER => nothing */ + 0, /* GROUP => nothing */ + 0, /* HAVING => nothing */ + 0, /* LIMIT => nothing */ + 0, /* WHERE => nothing */ + 0, /* INTO => nothing */ + 0, /* NOTHING => nothing */ + 0, /* FLOAT => nothing */ + 0, /* BLOB => nothing */ + 0, /* INTEGER => nothing */ + 0, /* VARIABLE => nothing */ + 0, /* CASE => nothing */ + 0, /* WHEN => nothing */ + 0, /* THEN => nothing */ + 0, /* ELSE => nothing */ + 0, /* INDEX => nothing */ + 0, /* ALTER => nothing */ + 0, /* ADD => nothing */ + 0, /* WINDOW => nothing */ + 0, /* OVER => nothing */ + 0, /* FILTER => nothing */ + 0, /* COLUMN => nothing */ + 0, /* AGG_FUNCTION => nothing */ + 0, /* AGG_COLUMN => nothing */ + 0, /* TRUEFALSE => nothing */ + 0, /* ISNOT => nothing */ + 0, /* FUNCTION => nothing */ + 0, /* UMINUS => nothing */ + 0, /* UPLUS => nothing */ + 0, /* TRUTH => nothing */ + 0, /* REGISTER => nothing */ + 0, /* VECTOR => nothing */ + 0, /* SELECT_COLUMN => nothing */ + 0, /* IF_NULL_ROW => nothing */ + 0, /* ASTERISK => nothing */ + 0, /* SPAN => nothing */ + 0, /* SPACE => nothing */ + 0, /* ILLEGAL => nothing */ +}; +#endif /* YYFALLBACK */ + +/* The following structure represents a single element of the +** parser's stack. Information stored includes: +** +** + The state number for the parser at this level of the stack. +** +** + The value of the token stored at this level of the stack. +** (In other words, the "major" token.) +** +** + The semantic value stored at this level of the stack. This is +** the information used by the action routines in the grammar. +** It is sometimes called the "minor" token. +** +** After the "shift" half of a SHIFTREDUCE action, the stateno field +** actually contains the reduce action for the second half of the +** SHIFTREDUCE. +*/ +struct yyStackEntry { + YYACTIONTYPE stateno; /* The state-number, or reduce action in SHIFTREDUCE */ + YYCODETYPE major; /* The major token value. This is the code + ** number for the token at this stack level */ + YYMINORTYPE minor; /* The user-supplied minor token value. This + ** is the value of the token */ +}; +typedef struct yyStackEntry yyStackEntry; + +/* The state of the parser is completely contained in an instance of +** the following structure */ +struct yyParser { + yyStackEntry *yytos; /* Pointer to top element of the stack */ +#ifdef YYTRACKMAXSTACKDEPTH + int yyhwm; /* High-water mark of the stack */ +#endif +#ifndef YYNOERRORRECOVERY + int yyerrcnt; /* Shifts left before out of the error */ +#endif + sqlite3ParserARG_SDECL /* A place to hold %extra_argument */ + sqlite3ParserCTX_SDECL /* A place to hold %extra_context */ +#if YYSTACKDEPTH<=0 + int yystksz; /* Current side of the stack */ + yyStackEntry *yystack; /* The parser's stack */ + yyStackEntry yystk0; /* First stack entry */ +#else + yyStackEntry yystack[YYSTACKDEPTH]; /* The parser's stack */ + yyStackEntry *yystackEnd; /* Last entry in the stack */ +#endif +}; +typedef struct yyParser yyParser; + +#ifndef NDEBUG +/* #include */ +static FILE *yyTraceFILE = 0; +static char *yyTracePrompt = 0; +#endif /* NDEBUG */ + +#ifndef NDEBUG +/* +** Turn parser tracing on by giving a stream to which to write the trace +** and a prompt to preface each trace message. Tracing is turned off +** by making either argument NULL +** +** Inputs: +**
      +**
    • A FILE* to which trace output should be written. +** If NULL, then tracing is turned off. +**
    • A prefix string written at the beginning of every +** line of trace output. If NULL, then tracing is +** turned off. +**
    +** +** Outputs: +** None. +*/ +SQLITE_PRIVATE void sqlite3ParserTrace(FILE *TraceFILE, char *zTracePrompt){ + yyTraceFILE = TraceFILE; + yyTracePrompt = zTracePrompt; + if( yyTraceFILE==0 ) yyTracePrompt = 0; + else if( yyTracePrompt==0 ) yyTraceFILE = 0; +} +#endif /* NDEBUG */ + +#if defined(YYCOVERAGE) || !defined(NDEBUG) +/* For tracing shifts, the names of all terminals and nonterminals +** are required. The following table supplies these names */ +static const char *const yyTokenName[] = { + /* 0 */ "$", + /* 1 */ "SEMI", + /* 2 */ "EXPLAIN", + /* 3 */ "QUERY", + /* 4 */ "PLAN", + /* 5 */ "BEGIN", + /* 6 */ "TRANSACTION", + /* 7 */ "DEFERRED", + /* 8 */ "IMMEDIATE", + /* 9 */ "EXCLUSIVE", + /* 10 */ "COMMIT", + /* 11 */ "END", + /* 12 */ "ROLLBACK", + /* 13 */ "SAVEPOINT", + /* 14 */ "RELEASE", + /* 15 */ "TO", + /* 16 */ "TABLE", + /* 17 */ "CREATE", + /* 18 */ "IF", + /* 19 */ "NOT", + /* 20 */ "EXISTS", + /* 21 */ "TEMP", + /* 22 */ "LP", + /* 23 */ "RP", + /* 24 */ "AS", + /* 25 */ "WITHOUT", + /* 26 */ "COMMA", + /* 27 */ "ABORT", + /* 28 */ "ACTION", + /* 29 */ "AFTER", + /* 30 */ "ANALYZE", + /* 31 */ "ASC", + /* 32 */ "ATTACH", + /* 33 */ "BEFORE", + /* 34 */ "BY", + /* 35 */ "CASCADE", + /* 36 */ "CAST", + /* 37 */ "CONFLICT", + /* 38 */ "DATABASE", + /* 39 */ "DESC", + /* 40 */ "DETACH", + /* 41 */ "EACH", + /* 42 */ "FAIL", + /* 43 */ "OR", + /* 44 */ "AND", + /* 45 */ "IS", + /* 46 */ "MATCH", + /* 47 */ "LIKE_KW", + /* 48 */ "BETWEEN", + /* 49 */ "IN", + /* 50 */ "ISNULL", + /* 51 */ "NOTNULL", + /* 52 */ "NE", + /* 53 */ "EQ", + /* 54 */ "GT", + /* 55 */ "LE", + /* 56 */ "LT", + /* 57 */ "GE", + /* 58 */ "ESCAPE", + /* 59 */ "ID", + /* 60 */ "COLUMNKW", + /* 61 */ "DO", + /* 62 */ "FOR", + /* 63 */ "IGNORE", + /* 64 */ "INITIALLY", + /* 65 */ "INSTEAD", + /* 66 */ "NO", + /* 67 */ "KEY", + /* 68 */ "OF", + /* 69 */ "OFFSET", + /* 70 */ "PRAGMA", + /* 71 */ "RAISE", + /* 72 */ "RECURSIVE", + /* 73 */ "REPLACE", + /* 74 */ "RESTRICT", + /* 75 */ "ROW", + /* 76 */ "ROWS", + /* 77 */ "TRIGGER", + /* 78 */ "VACUUM", + /* 79 */ "VIEW", + /* 80 */ "VIRTUAL", + /* 81 */ "WITH", + /* 82 */ "NULLS", + /* 83 */ "FIRST", + /* 84 */ "LAST", + /* 85 */ "CURRENT", + /* 86 */ "FOLLOWING", + /* 87 */ "PARTITION", + /* 88 */ "PRECEDING", + /* 89 */ "RANGE", + /* 90 */ "UNBOUNDED", + /* 91 */ "EXCLUDE", + /* 92 */ "GROUPS", + /* 93 */ "OTHERS", + /* 94 */ "TIES", + /* 95 */ "REINDEX", + /* 96 */ "RENAME", + /* 97 */ "CTIME_KW", + /* 98 */ "ANY", + /* 99 */ "BITAND", + /* 100 */ "BITOR", + /* 101 */ "LSHIFT", + /* 102 */ "RSHIFT", + /* 103 */ "PLUS", + /* 104 */ "MINUS", + /* 105 */ "STAR", + /* 106 */ "SLASH", + /* 107 */ "REM", + /* 108 */ "CONCAT", + /* 109 */ "COLLATE", + /* 110 */ "BITNOT", + /* 111 */ "ON", + /* 112 */ "INDEXED", + /* 113 */ "STRING", + /* 114 */ "JOIN_KW", + /* 115 */ "CONSTRAINT", + /* 116 */ "DEFAULT", + /* 117 */ "NULL", + /* 118 */ "PRIMARY", + /* 119 */ "UNIQUE", + /* 120 */ "CHECK", + /* 121 */ "REFERENCES", + /* 122 */ "AUTOINCR", + /* 123 */ "INSERT", + /* 124 */ "DELETE", + /* 125 */ "UPDATE", + /* 126 */ "SET", + /* 127 */ "DEFERRABLE", + /* 128 */ "FOREIGN", + /* 129 */ "DROP", + /* 130 */ "UNION", + /* 131 */ "ALL", + /* 132 */ "EXCEPT", + /* 133 */ "INTERSECT", + /* 134 */ "SELECT", + /* 135 */ "VALUES", + /* 136 */ "DISTINCT", + /* 137 */ "DOT", + /* 138 */ "FROM", + /* 139 */ "JOIN", + /* 140 */ "USING", + /* 141 */ "ORDER", + /* 142 */ "GROUP", + /* 143 */ "HAVING", + /* 144 */ "LIMIT", + /* 145 */ "WHERE", + /* 146 */ "INTO", + /* 147 */ "NOTHING", + /* 148 */ "FLOAT", + /* 149 */ "BLOB", + /* 150 */ "INTEGER", + /* 151 */ "VARIABLE", + /* 152 */ "CASE", + /* 153 */ "WHEN", + /* 154 */ "THEN", + /* 155 */ "ELSE", + /* 156 */ "INDEX", + /* 157 */ "ALTER", + /* 158 */ "ADD", + /* 159 */ "WINDOW", + /* 160 */ "OVER", + /* 161 */ "FILTER", + /* 162 */ "COLUMN", + /* 163 */ "AGG_FUNCTION", + /* 164 */ "AGG_COLUMN", + /* 165 */ "TRUEFALSE", + /* 166 */ "ISNOT", + /* 167 */ "FUNCTION", + /* 168 */ "UMINUS", + /* 169 */ "UPLUS", + /* 170 */ "TRUTH", + /* 171 */ "REGISTER", + /* 172 */ "VECTOR", + /* 173 */ "SELECT_COLUMN", + /* 174 */ "IF_NULL_ROW", + /* 175 */ "ASTERISK", + /* 176 */ "SPAN", + /* 177 */ "SPACE", + /* 178 */ "ILLEGAL", + /* 179 */ "input", + /* 180 */ "cmdlist", + /* 181 */ "ecmd", + /* 182 */ "cmdx", + /* 183 */ "explain", + /* 184 */ "cmd", + /* 185 */ "transtype", + /* 186 */ "trans_opt", + /* 187 */ "nm", + /* 188 */ "savepoint_opt", + /* 189 */ "create_table", + /* 190 */ "create_table_args", + /* 191 */ "createkw", + /* 192 */ "temp", + /* 193 */ "ifnotexists", + /* 194 */ "dbnm", + /* 195 */ "columnlist", + /* 196 */ "conslist_opt", + /* 197 */ "table_options", + /* 198 */ "select", + /* 199 */ "columnname", + /* 200 */ "carglist", + /* 201 */ "typetoken", + /* 202 */ "typename", + /* 203 */ "signed", + /* 204 */ "plus_num", + /* 205 */ "minus_num", + /* 206 */ "scanpt", + /* 207 */ "scantok", + /* 208 */ "ccons", + /* 209 */ "term", + /* 210 */ "expr", + /* 211 */ "onconf", + /* 212 */ "sortorder", + /* 213 */ "autoinc", + /* 214 */ "eidlist_opt", + /* 215 */ "refargs", + /* 216 */ "defer_subclause", + /* 217 */ "refarg", + /* 218 */ "refact", + /* 219 */ "init_deferred_pred_opt", + /* 220 */ "conslist", + /* 221 */ "tconscomma", + /* 222 */ "tcons", + /* 223 */ "sortlist", + /* 224 */ "eidlist", + /* 225 */ "defer_subclause_opt", + /* 226 */ "orconf", + /* 227 */ "resolvetype", + /* 228 */ "raisetype", + /* 229 */ "ifexists", + /* 230 */ "fullname", + /* 231 */ "selectnowith", + /* 232 */ "oneselect", + /* 233 */ "wqlist", + /* 234 */ "multiselect_op", + /* 235 */ "distinct", + /* 236 */ "selcollist", + /* 237 */ "from", + /* 238 */ "where_opt", + /* 239 */ "groupby_opt", + /* 240 */ "having_opt", + /* 241 */ "orderby_opt", + /* 242 */ "limit_opt", + /* 243 */ "window_clause", + /* 244 */ "values", + /* 245 */ "nexprlist", + /* 246 */ "sclp", + /* 247 */ "as", + /* 248 */ "seltablist", + /* 249 */ "stl_prefix", + /* 250 */ "joinop", + /* 251 */ "indexed_opt", + /* 252 */ "on_opt", + /* 253 */ "using_opt", + /* 254 */ "exprlist", + /* 255 */ "xfullname", + /* 256 */ "idlist", + /* 257 */ "nulls", + /* 258 */ "with", + /* 259 */ "setlist", + /* 260 */ "insert_cmd", + /* 261 */ "idlist_opt", + /* 262 */ "upsert", + /* 263 */ "filter_over", + /* 264 */ "likeop", + /* 265 */ "between_op", + /* 266 */ "in_op", + /* 267 */ "paren_exprlist", + /* 268 */ "case_operand", + /* 269 */ "case_exprlist", + /* 270 */ "case_else", + /* 271 */ "uniqueflag", + /* 272 */ "collate", + /* 273 */ "vinto", + /* 274 */ "nmnum", + /* 275 */ "trigger_decl", + /* 276 */ "trigger_cmd_list", + /* 277 */ "trigger_time", + /* 278 */ "trigger_event", + /* 279 */ "foreach_clause", + /* 280 */ "when_clause", + /* 281 */ "trigger_cmd", + /* 282 */ "trnm", + /* 283 */ "tridxby", + /* 284 */ "database_kw_opt", + /* 285 */ "key_opt", + /* 286 */ "add_column_fullname", + /* 287 */ "kwcolumn_opt", + /* 288 */ "create_vtab", + /* 289 */ "vtabarglist", + /* 290 */ "vtabarg", + /* 291 */ "vtabargtoken", + /* 292 */ "lp", + /* 293 */ "anylist", + /* 294 */ "windowdefn_list", + /* 295 */ "windowdefn", + /* 296 */ "window", + /* 297 */ "frame_opt", + /* 298 */ "part_opt", + /* 299 */ "filter_clause", + /* 300 */ "over_clause", + /* 301 */ "range_or_rows", + /* 302 */ "frame_bound", + /* 303 */ "frame_bound_s", + /* 304 */ "frame_bound_e", + /* 305 */ "frame_exclude_opt", + /* 306 */ "frame_exclude", +}; +#endif /* defined(YYCOVERAGE) || !defined(NDEBUG) */ + +#ifndef NDEBUG +/* For tracing reduce actions, the names of all rules are required. +*/ +static const char *const yyRuleName[] = { + /* 0 */ "explain ::= EXPLAIN", + /* 1 */ "explain ::= EXPLAIN QUERY PLAN", + /* 2 */ "cmdx ::= cmd", + /* 3 */ "cmd ::= BEGIN transtype trans_opt", + /* 4 */ "transtype ::=", + /* 5 */ "transtype ::= DEFERRED", + /* 6 */ "transtype ::= IMMEDIATE", + /* 7 */ "transtype ::= EXCLUSIVE", + /* 8 */ "cmd ::= COMMIT|END trans_opt", + /* 9 */ "cmd ::= ROLLBACK trans_opt", + /* 10 */ "cmd ::= SAVEPOINT nm", + /* 11 */ "cmd ::= RELEASE savepoint_opt nm", + /* 12 */ "cmd ::= ROLLBACK trans_opt TO savepoint_opt nm", + /* 13 */ "create_table ::= createkw temp TABLE ifnotexists nm dbnm", + /* 14 */ "createkw ::= CREATE", + /* 15 */ "ifnotexists ::=", + /* 16 */ "ifnotexists ::= IF NOT EXISTS", + /* 17 */ "temp ::= TEMP", + /* 18 */ "temp ::=", + /* 19 */ "create_table_args ::= LP columnlist conslist_opt RP table_options", + /* 20 */ "create_table_args ::= AS select", + /* 21 */ "table_options ::=", + /* 22 */ "table_options ::= WITHOUT nm", + /* 23 */ "columnname ::= nm typetoken", + /* 24 */ "typetoken ::=", + /* 25 */ "typetoken ::= typename LP signed RP", + /* 26 */ "typetoken ::= typename LP signed COMMA signed RP", + /* 27 */ "typename ::= typename ID|STRING", + /* 28 */ "scanpt ::=", + /* 29 */ "scantok ::=", + /* 30 */ "ccons ::= CONSTRAINT nm", + /* 31 */ "ccons ::= DEFAULT scantok term", + /* 32 */ "ccons ::= DEFAULT LP expr RP", + /* 33 */ "ccons ::= DEFAULT PLUS scantok term", + /* 34 */ "ccons ::= DEFAULT MINUS scantok term", + /* 35 */ "ccons ::= DEFAULT scantok ID|INDEXED", + /* 36 */ "ccons ::= NOT NULL onconf", + /* 37 */ "ccons ::= PRIMARY KEY sortorder onconf autoinc", + /* 38 */ "ccons ::= UNIQUE onconf", + /* 39 */ "ccons ::= CHECK LP expr RP", + /* 40 */ "ccons ::= REFERENCES nm eidlist_opt refargs", + /* 41 */ "ccons ::= defer_subclause", + /* 42 */ "ccons ::= COLLATE ID|STRING", + /* 43 */ "autoinc ::=", + /* 44 */ "autoinc ::= AUTOINCR", + /* 45 */ "refargs ::=", + /* 46 */ "refargs ::= refargs refarg", + /* 47 */ "refarg ::= MATCH nm", + /* 48 */ "refarg ::= ON INSERT refact", + /* 49 */ "refarg ::= ON DELETE refact", + /* 50 */ "refarg ::= ON UPDATE refact", + /* 51 */ "refact ::= SET NULL", + /* 52 */ "refact ::= SET DEFAULT", + /* 53 */ "refact ::= CASCADE", + /* 54 */ "refact ::= RESTRICT", + /* 55 */ "refact ::= NO ACTION", + /* 56 */ "defer_subclause ::= NOT DEFERRABLE init_deferred_pred_opt", + /* 57 */ "defer_subclause ::= DEFERRABLE init_deferred_pred_opt", + /* 58 */ "init_deferred_pred_opt ::=", + /* 59 */ "init_deferred_pred_opt ::= INITIALLY DEFERRED", + /* 60 */ "init_deferred_pred_opt ::= INITIALLY IMMEDIATE", + /* 61 */ "conslist_opt ::=", + /* 62 */ "tconscomma ::= COMMA", + /* 63 */ "tcons ::= CONSTRAINT nm", + /* 64 */ "tcons ::= PRIMARY KEY LP sortlist autoinc RP onconf", + /* 65 */ "tcons ::= UNIQUE LP sortlist RP onconf", + /* 66 */ "tcons ::= CHECK LP expr RP onconf", + /* 67 */ "tcons ::= FOREIGN KEY LP eidlist RP REFERENCES nm eidlist_opt refargs defer_subclause_opt", + /* 68 */ "defer_subclause_opt ::=", + /* 69 */ "onconf ::=", + /* 70 */ "onconf ::= ON CONFLICT resolvetype", + /* 71 */ "orconf ::=", + /* 72 */ "orconf ::= OR resolvetype", + /* 73 */ "resolvetype ::= IGNORE", + /* 74 */ "resolvetype ::= REPLACE", + /* 75 */ "cmd ::= DROP TABLE ifexists fullname", + /* 76 */ "ifexists ::= IF EXISTS", + /* 77 */ "ifexists ::=", + /* 78 */ "cmd ::= createkw temp VIEW ifnotexists nm dbnm eidlist_opt AS select", + /* 79 */ "cmd ::= DROP VIEW ifexists fullname", + /* 80 */ "cmd ::= select", + /* 81 */ "select ::= WITH wqlist selectnowith", + /* 82 */ "select ::= WITH RECURSIVE wqlist selectnowith", + /* 83 */ "select ::= selectnowith", + /* 84 */ "selectnowith ::= selectnowith multiselect_op oneselect", + /* 85 */ "multiselect_op ::= UNION", + /* 86 */ "multiselect_op ::= UNION ALL", + /* 87 */ "multiselect_op ::= EXCEPT|INTERSECT", + /* 88 */ "oneselect ::= SELECT distinct selcollist from where_opt groupby_opt having_opt orderby_opt limit_opt", + /* 89 */ "oneselect ::= SELECT distinct selcollist from where_opt groupby_opt having_opt window_clause orderby_opt limit_opt", + /* 90 */ "values ::= VALUES LP nexprlist RP", + /* 91 */ "values ::= values COMMA LP nexprlist RP", + /* 92 */ "distinct ::= DISTINCT", + /* 93 */ "distinct ::= ALL", + /* 94 */ "distinct ::=", + /* 95 */ "sclp ::=", + /* 96 */ "selcollist ::= sclp scanpt expr scanpt as", + /* 97 */ "selcollist ::= sclp scanpt STAR", + /* 98 */ "selcollist ::= sclp scanpt nm DOT STAR", + /* 99 */ "as ::= AS nm", + /* 100 */ "as ::=", + /* 101 */ "from ::=", + /* 102 */ "from ::= FROM seltablist", + /* 103 */ "stl_prefix ::= seltablist joinop", + /* 104 */ "stl_prefix ::=", + /* 105 */ "seltablist ::= stl_prefix nm dbnm as indexed_opt on_opt using_opt", + /* 106 */ "seltablist ::= stl_prefix nm dbnm LP exprlist RP as on_opt using_opt", + /* 107 */ "seltablist ::= stl_prefix LP select RP as on_opt using_opt", + /* 108 */ "seltablist ::= stl_prefix LP seltablist RP as on_opt using_opt", + /* 109 */ "dbnm ::=", + /* 110 */ "dbnm ::= DOT nm", + /* 111 */ "fullname ::= nm", + /* 112 */ "fullname ::= nm DOT nm", + /* 113 */ "xfullname ::= nm", + /* 114 */ "xfullname ::= nm DOT nm", + /* 115 */ "xfullname ::= nm DOT nm AS nm", + /* 116 */ "xfullname ::= nm AS nm", + /* 117 */ "joinop ::= COMMA|JOIN", + /* 118 */ "joinop ::= JOIN_KW JOIN", + /* 119 */ "joinop ::= JOIN_KW nm JOIN", + /* 120 */ "joinop ::= JOIN_KW nm nm JOIN", + /* 121 */ "on_opt ::= ON expr", + /* 122 */ "on_opt ::=", + /* 123 */ "indexed_opt ::=", + /* 124 */ "indexed_opt ::= INDEXED BY nm", + /* 125 */ "indexed_opt ::= NOT INDEXED", + /* 126 */ "using_opt ::= USING LP idlist RP", + /* 127 */ "using_opt ::=", + /* 128 */ "orderby_opt ::=", + /* 129 */ "orderby_opt ::= ORDER BY sortlist", + /* 130 */ "sortlist ::= sortlist COMMA expr sortorder nulls", + /* 131 */ "sortlist ::= expr sortorder nulls", + /* 132 */ "sortorder ::= ASC", + /* 133 */ "sortorder ::= DESC", + /* 134 */ "sortorder ::=", + /* 135 */ "nulls ::= NULLS FIRST", + /* 136 */ "nulls ::= NULLS LAST", + /* 137 */ "nulls ::=", + /* 138 */ "groupby_opt ::=", + /* 139 */ "groupby_opt ::= GROUP BY nexprlist", + /* 140 */ "having_opt ::=", + /* 141 */ "having_opt ::= HAVING expr", + /* 142 */ "limit_opt ::=", + /* 143 */ "limit_opt ::= LIMIT expr", + /* 144 */ "limit_opt ::= LIMIT expr OFFSET expr", + /* 145 */ "limit_opt ::= LIMIT expr COMMA expr", + /* 146 */ "cmd ::= with DELETE FROM xfullname indexed_opt where_opt", + /* 147 */ "where_opt ::=", + /* 148 */ "where_opt ::= WHERE expr", + /* 149 */ "cmd ::= with UPDATE orconf xfullname indexed_opt SET setlist where_opt", + /* 150 */ "setlist ::= setlist COMMA nm EQ expr", + /* 151 */ "setlist ::= setlist COMMA LP idlist RP EQ expr", + /* 152 */ "setlist ::= nm EQ expr", + /* 153 */ "setlist ::= LP idlist RP EQ expr", + /* 154 */ "cmd ::= with insert_cmd INTO xfullname idlist_opt select upsert", + /* 155 */ "cmd ::= with insert_cmd INTO xfullname idlist_opt DEFAULT VALUES", + /* 156 */ "upsert ::=", + /* 157 */ "upsert ::= ON CONFLICT LP sortlist RP where_opt DO UPDATE SET setlist where_opt", + /* 158 */ "upsert ::= ON CONFLICT LP sortlist RP where_opt DO NOTHING", + /* 159 */ "upsert ::= ON CONFLICT DO NOTHING", + /* 160 */ "insert_cmd ::= INSERT orconf", + /* 161 */ "insert_cmd ::= REPLACE", + /* 162 */ "idlist_opt ::=", + /* 163 */ "idlist_opt ::= LP idlist RP", + /* 164 */ "idlist ::= idlist COMMA nm", + /* 165 */ "idlist ::= nm", + /* 166 */ "expr ::= LP expr RP", + /* 167 */ "expr ::= ID|INDEXED", + /* 168 */ "expr ::= JOIN_KW", + /* 169 */ "expr ::= nm DOT nm", + /* 170 */ "expr ::= nm DOT nm DOT nm", + /* 171 */ "term ::= NULL|FLOAT|BLOB", + /* 172 */ "term ::= STRING", + /* 173 */ "term ::= INTEGER", + /* 174 */ "expr ::= VARIABLE", + /* 175 */ "expr ::= expr COLLATE ID|STRING", + /* 176 */ "expr ::= CAST LP expr AS typetoken RP", + /* 177 */ "expr ::= ID|INDEXED LP distinct exprlist RP", + /* 178 */ "expr ::= ID|INDEXED LP STAR RP", + /* 179 */ "expr ::= ID|INDEXED LP distinct exprlist RP filter_over", + /* 180 */ "expr ::= ID|INDEXED LP STAR RP filter_over", + /* 181 */ "term ::= CTIME_KW", + /* 182 */ "expr ::= LP nexprlist COMMA expr RP", + /* 183 */ "expr ::= expr AND expr", + /* 184 */ "expr ::= expr OR expr", + /* 185 */ "expr ::= expr LT|GT|GE|LE expr", + /* 186 */ "expr ::= expr EQ|NE expr", + /* 187 */ "expr ::= expr BITAND|BITOR|LSHIFT|RSHIFT expr", + /* 188 */ "expr ::= expr PLUS|MINUS expr", + /* 189 */ "expr ::= expr STAR|SLASH|REM expr", + /* 190 */ "expr ::= expr CONCAT expr", + /* 191 */ "likeop ::= NOT LIKE_KW|MATCH", + /* 192 */ "expr ::= expr likeop expr", + /* 193 */ "expr ::= expr likeop expr ESCAPE expr", + /* 194 */ "expr ::= expr ISNULL|NOTNULL", + /* 195 */ "expr ::= expr NOT NULL", + /* 196 */ "expr ::= expr IS expr", + /* 197 */ "expr ::= expr IS NOT expr", + /* 198 */ "expr ::= NOT expr", + /* 199 */ "expr ::= BITNOT expr", + /* 200 */ "expr ::= PLUS|MINUS expr", + /* 201 */ "between_op ::= BETWEEN", + /* 202 */ "between_op ::= NOT BETWEEN", + /* 203 */ "expr ::= expr between_op expr AND expr", + /* 204 */ "in_op ::= IN", + /* 205 */ "in_op ::= NOT IN", + /* 206 */ "expr ::= expr in_op LP exprlist RP", + /* 207 */ "expr ::= LP select RP", + /* 208 */ "expr ::= expr in_op LP select RP", + /* 209 */ "expr ::= expr in_op nm dbnm paren_exprlist", + /* 210 */ "expr ::= EXISTS LP select RP", + /* 211 */ "expr ::= CASE case_operand case_exprlist case_else END", + /* 212 */ "case_exprlist ::= case_exprlist WHEN expr THEN expr", + /* 213 */ "case_exprlist ::= WHEN expr THEN expr", + /* 214 */ "case_else ::= ELSE expr", + /* 215 */ "case_else ::=", + /* 216 */ "case_operand ::= expr", + /* 217 */ "case_operand ::=", + /* 218 */ "exprlist ::=", + /* 219 */ "nexprlist ::= nexprlist COMMA expr", + /* 220 */ "nexprlist ::= expr", + /* 221 */ "paren_exprlist ::=", + /* 222 */ "paren_exprlist ::= LP exprlist RP", + /* 223 */ "cmd ::= createkw uniqueflag INDEX ifnotexists nm dbnm ON nm LP sortlist RP where_opt", + /* 224 */ "uniqueflag ::= UNIQUE", + /* 225 */ "uniqueflag ::=", + /* 226 */ "eidlist_opt ::=", + /* 227 */ "eidlist_opt ::= LP eidlist RP", + /* 228 */ "eidlist ::= eidlist COMMA nm collate sortorder", + /* 229 */ "eidlist ::= nm collate sortorder", + /* 230 */ "collate ::=", + /* 231 */ "collate ::= COLLATE ID|STRING", + /* 232 */ "cmd ::= DROP INDEX ifexists fullname", + /* 233 */ "cmd ::= VACUUM vinto", + /* 234 */ "cmd ::= VACUUM nm vinto", + /* 235 */ "vinto ::= INTO expr", + /* 236 */ "vinto ::=", + /* 237 */ "cmd ::= PRAGMA nm dbnm", + /* 238 */ "cmd ::= PRAGMA nm dbnm EQ nmnum", + /* 239 */ "cmd ::= PRAGMA nm dbnm LP nmnum RP", + /* 240 */ "cmd ::= PRAGMA nm dbnm EQ minus_num", + /* 241 */ "cmd ::= PRAGMA nm dbnm LP minus_num RP", + /* 242 */ "plus_num ::= PLUS INTEGER|FLOAT", + /* 243 */ "minus_num ::= MINUS INTEGER|FLOAT", + /* 244 */ "cmd ::= createkw trigger_decl BEGIN trigger_cmd_list END", + /* 245 */ "trigger_decl ::= temp TRIGGER ifnotexists nm dbnm trigger_time trigger_event ON fullname foreach_clause when_clause", + /* 246 */ "trigger_time ::= BEFORE|AFTER", + /* 247 */ "trigger_time ::= INSTEAD OF", + /* 248 */ "trigger_time ::=", + /* 249 */ "trigger_event ::= DELETE|INSERT", + /* 250 */ "trigger_event ::= UPDATE", + /* 251 */ "trigger_event ::= UPDATE OF idlist", + /* 252 */ "when_clause ::=", + /* 253 */ "when_clause ::= WHEN expr", + /* 254 */ "trigger_cmd_list ::= trigger_cmd_list trigger_cmd SEMI", + /* 255 */ "trigger_cmd_list ::= trigger_cmd SEMI", + /* 256 */ "trnm ::= nm DOT nm", + /* 257 */ "tridxby ::= INDEXED BY nm", + /* 258 */ "tridxby ::= NOT INDEXED", + /* 259 */ "trigger_cmd ::= UPDATE orconf trnm tridxby SET setlist where_opt scanpt", + /* 260 */ "trigger_cmd ::= scanpt insert_cmd INTO trnm idlist_opt select upsert scanpt", + /* 261 */ "trigger_cmd ::= DELETE FROM trnm tridxby where_opt scanpt", + /* 262 */ "trigger_cmd ::= scanpt select scanpt", + /* 263 */ "expr ::= RAISE LP IGNORE RP", + /* 264 */ "expr ::= RAISE LP raisetype COMMA nm RP", + /* 265 */ "raisetype ::= ROLLBACK", + /* 266 */ "raisetype ::= ABORT", + /* 267 */ "raisetype ::= FAIL", + /* 268 */ "cmd ::= DROP TRIGGER ifexists fullname", + /* 269 */ "cmd ::= ATTACH database_kw_opt expr AS expr key_opt", + /* 270 */ "cmd ::= DETACH database_kw_opt expr", + /* 271 */ "key_opt ::=", + /* 272 */ "key_opt ::= KEY expr", + /* 273 */ "cmd ::= REINDEX", + /* 274 */ "cmd ::= REINDEX nm dbnm", + /* 275 */ "cmd ::= ANALYZE", + /* 276 */ "cmd ::= ANALYZE nm dbnm", + /* 277 */ "cmd ::= ALTER TABLE fullname RENAME TO nm", + /* 278 */ "cmd ::= ALTER TABLE add_column_fullname ADD kwcolumn_opt columnname carglist", + /* 279 */ "add_column_fullname ::= fullname", + /* 280 */ "cmd ::= ALTER TABLE fullname RENAME kwcolumn_opt nm TO nm", + /* 281 */ "cmd ::= create_vtab", + /* 282 */ "cmd ::= create_vtab LP vtabarglist RP", + /* 283 */ "create_vtab ::= createkw VIRTUAL TABLE ifnotexists nm dbnm USING nm", + /* 284 */ "vtabarg ::=", + /* 285 */ "vtabargtoken ::= ANY", + /* 286 */ "vtabargtoken ::= lp anylist RP", + /* 287 */ "lp ::= LP", + /* 288 */ "with ::= WITH wqlist", + /* 289 */ "with ::= WITH RECURSIVE wqlist", + /* 290 */ "wqlist ::= nm eidlist_opt AS LP select RP", + /* 291 */ "wqlist ::= wqlist COMMA nm eidlist_opt AS LP select RP", + /* 292 */ "windowdefn_list ::= windowdefn", + /* 293 */ "windowdefn_list ::= windowdefn_list COMMA windowdefn", + /* 294 */ "windowdefn ::= nm AS LP window RP", + /* 295 */ "window ::= PARTITION BY nexprlist orderby_opt frame_opt", + /* 296 */ "window ::= nm PARTITION BY nexprlist orderby_opt frame_opt", + /* 297 */ "window ::= ORDER BY sortlist frame_opt", + /* 298 */ "window ::= nm ORDER BY sortlist frame_opt", + /* 299 */ "window ::= frame_opt", + /* 300 */ "window ::= nm frame_opt", + /* 301 */ "frame_opt ::=", + /* 302 */ "frame_opt ::= range_or_rows frame_bound_s frame_exclude_opt", + /* 303 */ "frame_opt ::= range_or_rows BETWEEN frame_bound_s AND frame_bound_e frame_exclude_opt", + /* 304 */ "range_or_rows ::= RANGE|ROWS|GROUPS", + /* 305 */ "frame_bound_s ::= frame_bound", + /* 306 */ "frame_bound_s ::= UNBOUNDED PRECEDING", + /* 307 */ "frame_bound_e ::= frame_bound", + /* 308 */ "frame_bound_e ::= UNBOUNDED FOLLOWING", + /* 309 */ "frame_bound ::= expr PRECEDING|FOLLOWING", + /* 310 */ "frame_bound ::= CURRENT ROW", + /* 311 */ "frame_exclude_opt ::=", + /* 312 */ "frame_exclude_opt ::= EXCLUDE frame_exclude", + /* 313 */ "frame_exclude ::= NO OTHERS", + /* 314 */ "frame_exclude ::= CURRENT ROW", + /* 315 */ "frame_exclude ::= GROUP|TIES", + /* 316 */ "window_clause ::= WINDOW windowdefn_list", + /* 317 */ "filter_over ::= filter_clause over_clause", + /* 318 */ "filter_over ::= over_clause", + /* 319 */ "filter_over ::= filter_clause", + /* 320 */ "over_clause ::= OVER LP window RP", + /* 321 */ "over_clause ::= OVER nm", + /* 322 */ "filter_clause ::= FILTER LP WHERE expr RP", + /* 323 */ "input ::= cmdlist", + /* 324 */ "cmdlist ::= cmdlist ecmd", + /* 325 */ "cmdlist ::= ecmd", + /* 326 */ "ecmd ::= SEMI", + /* 327 */ "ecmd ::= cmdx SEMI", + /* 328 */ "ecmd ::= explain cmdx", + /* 329 */ "trans_opt ::=", + /* 330 */ "trans_opt ::= TRANSACTION", + /* 331 */ "trans_opt ::= TRANSACTION nm", + /* 332 */ "savepoint_opt ::= SAVEPOINT", + /* 333 */ "savepoint_opt ::=", + /* 334 */ "cmd ::= create_table create_table_args", + /* 335 */ "columnlist ::= columnlist COMMA columnname carglist", + /* 336 */ "columnlist ::= columnname carglist", + /* 337 */ "nm ::= ID|INDEXED", + /* 338 */ "nm ::= STRING", + /* 339 */ "nm ::= JOIN_KW", + /* 340 */ "typetoken ::= typename", + /* 341 */ "typename ::= ID|STRING", + /* 342 */ "signed ::= plus_num", + /* 343 */ "signed ::= minus_num", + /* 344 */ "carglist ::= carglist ccons", + /* 345 */ "carglist ::=", + /* 346 */ "ccons ::= NULL onconf", + /* 347 */ "conslist_opt ::= COMMA conslist", + /* 348 */ "conslist ::= conslist tconscomma tcons", + /* 349 */ "conslist ::= tcons", + /* 350 */ "tconscomma ::=", + /* 351 */ "defer_subclause_opt ::= defer_subclause", + /* 352 */ "resolvetype ::= raisetype", + /* 353 */ "selectnowith ::= oneselect", + /* 354 */ "oneselect ::= values", + /* 355 */ "sclp ::= selcollist COMMA", + /* 356 */ "as ::= ID|STRING", + /* 357 */ "expr ::= term", + /* 358 */ "likeop ::= LIKE_KW|MATCH", + /* 359 */ "exprlist ::= nexprlist", + /* 360 */ "nmnum ::= plus_num", + /* 361 */ "nmnum ::= nm", + /* 362 */ "nmnum ::= ON", + /* 363 */ "nmnum ::= DELETE", + /* 364 */ "nmnum ::= DEFAULT", + /* 365 */ "plus_num ::= INTEGER|FLOAT", + /* 366 */ "foreach_clause ::=", + /* 367 */ "foreach_clause ::= FOR EACH ROW", + /* 368 */ "trnm ::= nm", + /* 369 */ "tridxby ::=", + /* 370 */ "database_kw_opt ::= DATABASE", + /* 371 */ "database_kw_opt ::=", + /* 372 */ "kwcolumn_opt ::=", + /* 373 */ "kwcolumn_opt ::= COLUMNKW", + /* 374 */ "vtabarglist ::= vtabarg", + /* 375 */ "vtabarglist ::= vtabarglist COMMA vtabarg", + /* 376 */ "vtabarg ::= vtabarg vtabargtoken", + /* 377 */ "anylist ::=", + /* 378 */ "anylist ::= anylist LP anylist RP", + /* 379 */ "anylist ::= anylist ANY", + /* 380 */ "with ::=", +}; +#endif /* NDEBUG */ + + +#if YYSTACKDEPTH<=0 +/* +** Try to increase the size of the parser stack. Return the number +** of errors. Return 0 on success. +*/ +static int yyGrowStack(yyParser *p){ + int newSize; + int idx; + yyStackEntry *pNew; + + newSize = p->yystksz*2 + 100; + idx = p->yytos ? (int)(p->yytos - p->yystack) : 0; + if( p->yystack==&p->yystk0 ){ + pNew = malloc(newSize*sizeof(pNew[0])); + if( pNew ) pNew[0] = p->yystk0; + }else{ + pNew = realloc(p->yystack, newSize*sizeof(pNew[0])); + } + if( pNew ){ + p->yystack = pNew; + p->yytos = &p->yystack[idx]; +#ifndef NDEBUG + if( yyTraceFILE ){ + fprintf(yyTraceFILE,"%sStack grows from %d to %d entries.\n", + yyTracePrompt, p->yystksz, newSize); + } +#endif + p->yystksz = newSize; + } + return pNew==0; +} +#endif + +/* Datatype of the argument to the memory allocated passed as the +** second argument to sqlite3ParserAlloc() below. This can be changed by +** putting an appropriate #define in the %include section of the input +** grammar. +*/ +#ifndef YYMALLOCARGTYPE +# define YYMALLOCARGTYPE size_t +#endif + +/* Initialize a new parser that has already been allocated. +*/ +SQLITE_PRIVATE void sqlite3ParserInit(void *yypRawParser sqlite3ParserCTX_PDECL){ + yyParser *yypParser = (yyParser*)yypRawParser; + sqlite3ParserCTX_STORE +#ifdef YYTRACKMAXSTACKDEPTH + yypParser->yyhwm = 0; +#endif +#if YYSTACKDEPTH<=0 + yypParser->yytos = NULL; + yypParser->yystack = NULL; + yypParser->yystksz = 0; + if( yyGrowStack(yypParser) ){ + yypParser->yystack = &yypParser->yystk0; + yypParser->yystksz = 1; + } +#endif +#ifndef YYNOERRORRECOVERY + yypParser->yyerrcnt = -1; +#endif + yypParser->yytos = yypParser->yystack; + yypParser->yystack[0].stateno = 0; + yypParser->yystack[0].major = 0; +#if YYSTACKDEPTH>0 + yypParser->yystackEnd = &yypParser->yystack[YYSTACKDEPTH-1]; +#endif +} + +#ifndef sqlite3Parser_ENGINEALWAYSONSTACK +/* +** This function allocates a new parser. +** The only argument is a pointer to a function which works like +** malloc. +** +** Inputs: +** A pointer to the function used to allocate memory. +** +** Outputs: +** A pointer to a parser. This pointer is used in subsequent calls +** to sqlite3Parser and sqlite3ParserFree. +*/ +SQLITE_PRIVATE void *sqlite3ParserAlloc(void *(*mallocProc)(YYMALLOCARGTYPE) sqlite3ParserCTX_PDECL){ + yyParser *yypParser; + yypParser = (yyParser*)(*mallocProc)( (YYMALLOCARGTYPE)sizeof(yyParser) ); + if( yypParser ){ + sqlite3ParserCTX_STORE + sqlite3ParserInit(yypParser sqlite3ParserCTX_PARAM); + } + return (void*)yypParser; +} +#endif /* sqlite3Parser_ENGINEALWAYSONSTACK */ + + +/* The following function deletes the "minor type" or semantic value +** associated with a symbol. The symbol can be either a terminal +** or nonterminal. "yymajor" is the symbol code, and "yypminor" is +** a pointer to the value to be deleted. The code used to do the +** deletions is derived from the %destructor and/or %token_destructor +** directives of the input grammar. +*/ +static void yy_destructor( + yyParser *yypParser, /* The parser */ + YYCODETYPE yymajor, /* Type code for object to destroy */ + YYMINORTYPE *yypminor /* The object to be destroyed */ +){ + sqlite3ParserARG_FETCH + sqlite3ParserCTX_FETCH + switch( yymajor ){ + /* Here is inserted the actions which take place when a + ** terminal or non-terminal is destroyed. This can happen + ** when the symbol is popped from the stack during a + ** reduce or during error processing or when a parser is + ** being destroyed before it is finished parsing. + ** + ** Note: during a reduce, the only symbols destroyed are those + ** which appear on the RHS of the rule, but which are *not* used + ** inside the C code. + */ +/********* Begin destructor definitions ***************************************/ + case 198: /* select */ + case 231: /* selectnowith */ + case 232: /* oneselect */ + case 244: /* values */ +{ +sqlite3SelectDelete(pParse->db, (yypminor->yy25)); +} + break; + case 209: /* term */ + case 210: /* expr */ + case 238: /* where_opt */ + case 240: /* having_opt */ + case 252: /* on_opt */ + case 268: /* case_operand */ + case 270: /* case_else */ + case 273: /* vinto */ + case 280: /* when_clause */ + case 285: /* key_opt */ + case 299: /* filter_clause */ +{ +sqlite3ExprDelete(pParse->db, (yypminor->yy46)); +} + break; + case 214: /* eidlist_opt */ + case 223: /* sortlist */ + case 224: /* eidlist */ + case 236: /* selcollist */ + case 239: /* groupby_opt */ + case 241: /* orderby_opt */ + case 245: /* nexprlist */ + case 246: /* sclp */ + case 254: /* exprlist */ + case 259: /* setlist */ + case 267: /* paren_exprlist */ + case 269: /* case_exprlist */ + case 298: /* part_opt */ +{ +sqlite3ExprListDelete(pParse->db, (yypminor->yy138)); +} + break; + case 230: /* fullname */ + case 237: /* from */ + case 248: /* seltablist */ + case 249: /* stl_prefix */ + case 255: /* xfullname */ +{ +sqlite3SrcListDelete(pParse->db, (yypminor->yy609)); +} + break; + case 233: /* wqlist */ +{ +sqlite3WithDelete(pParse->db, (yypminor->yy297)); +} + break; + case 243: /* window_clause */ + case 294: /* windowdefn_list */ +{ +sqlite3WindowListDelete(pParse->db, (yypminor->yy455)); +} + break; + case 253: /* using_opt */ + case 256: /* idlist */ + case 261: /* idlist_opt */ +{ +sqlite3IdListDelete(pParse->db, (yypminor->yy406)); +} + break; + case 263: /* filter_over */ + case 295: /* windowdefn */ + case 296: /* window */ + case 297: /* frame_opt */ + case 300: /* over_clause */ +{ +sqlite3WindowDelete(pParse->db, (yypminor->yy455)); +} + break; + case 276: /* trigger_cmd_list */ + case 281: /* trigger_cmd */ +{ +sqlite3DeleteTriggerStep(pParse->db, (yypminor->yy527)); +} + break; + case 278: /* trigger_event */ +{ +sqlite3IdListDelete(pParse->db, (yypminor->yy572).b); +} + break; + case 302: /* frame_bound */ + case 303: /* frame_bound_s */ + case 304: /* frame_bound_e */ +{ +sqlite3ExprDelete(pParse->db, (yypminor->yy57).pExpr); +} + break; +/********* End destructor definitions *****************************************/ + default: break; /* If no destructor action specified: do nothing */ + } +} + +/* +** Pop the parser's stack once. +** +** If there is a destructor routine associated with the token which +** is popped from the stack, then call it. +*/ +static void yy_pop_parser_stack(yyParser *pParser){ + yyStackEntry *yytos; + assert( pParser->yytos!=0 ); + assert( pParser->yytos > pParser->yystack ); + yytos = pParser->yytos--; +#ifndef NDEBUG + if( yyTraceFILE ){ + fprintf(yyTraceFILE,"%sPopping %s\n", + yyTracePrompt, + yyTokenName[yytos->major]); + } +#endif + yy_destructor(pParser, yytos->major, &yytos->minor); +} + +/* +** Clear all secondary memory allocations from the parser +*/ +SQLITE_PRIVATE void sqlite3ParserFinalize(void *p){ + yyParser *pParser = (yyParser*)p; + while( pParser->yytos>pParser->yystack ) yy_pop_parser_stack(pParser); +#if YYSTACKDEPTH<=0 + if( pParser->yystack!=&pParser->yystk0 ) free(pParser->yystack); +#endif +} + +#ifndef sqlite3Parser_ENGINEALWAYSONSTACK +/* +** Deallocate and destroy a parser. Destructors are called for +** all stack elements before shutting the parser down. +** +** If the YYPARSEFREENEVERNULL macro exists (for example because it +** is defined in a %include section of the input grammar) then it is +** assumed that the input pointer is never NULL. +*/ +SQLITE_PRIVATE void sqlite3ParserFree( + void *p, /* The parser to be deleted */ + void (*freeProc)(void*) /* Function used to reclaim memory */ +){ +#ifndef YYPARSEFREENEVERNULL + if( p==0 ) return; +#endif + sqlite3ParserFinalize(p); + (*freeProc)(p); +} +#endif /* sqlite3Parser_ENGINEALWAYSONSTACK */ + +/* +** Return the peak depth of the stack for a parser. +*/ +#ifdef YYTRACKMAXSTACKDEPTH +SQLITE_PRIVATE int sqlite3ParserStackPeak(void *p){ + yyParser *pParser = (yyParser*)p; + return pParser->yyhwm; +} +#endif + +/* This array of booleans keeps track of the parser statement +** coverage. The element yycoverage[X][Y] is set when the parser +** is in state X and has a lookahead token Y. In a well-tested +** systems, every element of this matrix should end up being set. +*/ +#if defined(YYCOVERAGE) +static unsigned char yycoverage[YYNSTATE][YYNTOKEN]; +#endif + +/* +** Write into out a description of every state/lookahead combination that +** +** (1) has not been used by the parser, and +** (2) is not a syntax error. +** +** Return the number of missed state/lookahead combinations. +*/ +#if defined(YYCOVERAGE) +SQLITE_PRIVATE int sqlite3ParserCoverage(FILE *out){ + int stateno, iLookAhead, i; + int nMissed = 0; + for(stateno=0; statenoYY_MAX_SHIFT ) return stateno; + assert( stateno <= YY_SHIFT_COUNT ); +#if defined(YYCOVERAGE) + yycoverage[stateno][iLookAhead] = 1; +#endif + do{ + i = yy_shift_ofst[stateno]; + assert( i>=0 ); + assert( i<=YY_ACTTAB_COUNT ); + assert( i+YYNTOKEN<=(int)YY_NLOOKAHEAD ); + assert( iLookAhead!=YYNOCODE ); + assert( iLookAhead < YYNTOKEN ); + i += iLookAhead; + assert( i<(int)YY_NLOOKAHEAD ); + if( yy_lookahead[i]!=iLookAhead ){ +#ifdef YYFALLBACK + YYCODETYPE iFallback; /* Fallback token */ + assert( iLookAhead %s\n", + yyTracePrompt, yyTokenName[iLookAhead], yyTokenName[iFallback]); + } +#endif + assert( yyFallback[iFallback]==0 ); /* Fallback loop must terminate */ + iLookAhead = iFallback; + continue; + } +#endif +#ifdef YYWILDCARD + { + int j = i - iLookAhead + YYWILDCARD; + assert( j<(int)(sizeof(yy_lookahead)/sizeof(yy_lookahead[0])) ); + if( yy_lookahead[j]==YYWILDCARD && iLookAhead>0 ){ +#ifndef NDEBUG + if( yyTraceFILE ){ + fprintf(yyTraceFILE, "%sWILDCARD %s => %s\n", + yyTracePrompt, yyTokenName[iLookAhead], + yyTokenName[YYWILDCARD]); + } +#endif /* NDEBUG */ + return yy_action[j]; + } + } +#endif /* YYWILDCARD */ + return yy_default[stateno]; + }else{ + assert( i>=0 && iYY_REDUCE_COUNT ){ + return yy_default[stateno]; + } +#else + assert( stateno<=YY_REDUCE_COUNT ); +#endif + i = yy_reduce_ofst[stateno]; + assert( iLookAhead!=YYNOCODE ); + i += iLookAhead; +#ifdef YYERRORSYMBOL + if( i<0 || i>=YY_ACTTAB_COUNT || yy_lookahead[i]!=iLookAhead ){ + return yy_default[stateno]; + } +#else + assert( i>=0 && iyytos>yypParser->yystack ) yy_pop_parser_stack(yypParser); + /* Here code is inserted which will execute if the parser + ** stack every overflows */ +/******** Begin %stack_overflow code ******************************************/ + + sqlite3ErrorMsg(pParse, "parser stack overflow"); +/******** End %stack_overflow code ********************************************/ + sqlite3ParserARG_STORE /* Suppress warning about unused %extra_argument var */ + sqlite3ParserCTX_STORE +} + +/* +** Print tracing information for a SHIFT action +*/ +#ifndef NDEBUG +static void yyTraceShift(yyParser *yypParser, int yyNewState, const char *zTag){ + if( yyTraceFILE ){ + if( yyNewStateyytos->major], + yyNewState); + }else{ + fprintf(yyTraceFILE,"%s%s '%s', pending reduce %d\n", + yyTracePrompt, zTag, yyTokenName[yypParser->yytos->major], + yyNewState - YY_MIN_REDUCE); + } + } +} +#else +# define yyTraceShift(X,Y,Z) +#endif + +/* +** Perform a shift action. +*/ +static void yy_shift( + yyParser *yypParser, /* The parser to be shifted */ + YYACTIONTYPE yyNewState, /* The new state to shift in */ + YYCODETYPE yyMajor, /* The major token to shift in */ + sqlite3ParserTOKENTYPE yyMinor /* The minor token to shift in */ +){ + yyStackEntry *yytos; + yypParser->yytos++; +#ifdef YYTRACKMAXSTACKDEPTH + if( (int)(yypParser->yytos - yypParser->yystack)>yypParser->yyhwm ){ + yypParser->yyhwm++; + assert( yypParser->yyhwm == (int)(yypParser->yytos - yypParser->yystack) ); + } +#endif +#if YYSTACKDEPTH>0 + if( yypParser->yytos>yypParser->yystackEnd ){ + yypParser->yytos--; + yyStackOverflow(yypParser); + return; + } +#else + if( yypParser->yytos>=&yypParser->yystack[yypParser->yystksz] ){ + if( yyGrowStack(yypParser) ){ + yypParser->yytos--; + yyStackOverflow(yypParser); + return; + } + } +#endif + if( yyNewState > YY_MAX_SHIFT ){ + yyNewState += YY_MIN_REDUCE - YY_MIN_SHIFTREDUCE; + } + yytos = yypParser->yytos; + yytos->stateno = yyNewState; + yytos->major = yyMajor; + yytos->minor.yy0 = yyMinor; + yyTraceShift(yypParser, yyNewState, "Shift"); +} + +/* For rule J, yyRuleInfoLhs[J] contains the symbol on the left-hand side +** of that rule */ +static const YYCODETYPE yyRuleInfoLhs[] = { + 183, /* (0) explain ::= EXPLAIN */ + 183, /* (1) explain ::= EXPLAIN QUERY PLAN */ + 182, /* (2) cmdx ::= cmd */ + 184, /* (3) cmd ::= BEGIN transtype trans_opt */ + 185, /* (4) transtype ::= */ + 185, /* (5) transtype ::= DEFERRED */ + 185, /* (6) transtype ::= IMMEDIATE */ + 185, /* (7) transtype ::= EXCLUSIVE */ + 184, /* (8) cmd ::= COMMIT|END trans_opt */ + 184, /* (9) cmd ::= ROLLBACK trans_opt */ + 184, /* (10) cmd ::= SAVEPOINT nm */ + 184, /* (11) cmd ::= RELEASE savepoint_opt nm */ + 184, /* (12) cmd ::= ROLLBACK trans_opt TO savepoint_opt nm */ + 189, /* (13) create_table ::= createkw temp TABLE ifnotexists nm dbnm */ + 191, /* (14) createkw ::= CREATE */ + 193, /* (15) ifnotexists ::= */ + 193, /* (16) ifnotexists ::= IF NOT EXISTS */ + 192, /* (17) temp ::= TEMP */ + 192, /* (18) temp ::= */ + 190, /* (19) create_table_args ::= LP columnlist conslist_opt RP table_options */ + 190, /* (20) create_table_args ::= AS select */ + 197, /* (21) table_options ::= */ + 197, /* (22) table_options ::= WITHOUT nm */ + 199, /* (23) columnname ::= nm typetoken */ + 201, /* (24) typetoken ::= */ + 201, /* (25) typetoken ::= typename LP signed RP */ + 201, /* (26) typetoken ::= typename LP signed COMMA signed RP */ + 202, /* (27) typename ::= typename ID|STRING */ + 206, /* (28) scanpt ::= */ + 207, /* (29) scantok ::= */ + 208, /* (30) ccons ::= CONSTRAINT nm */ + 208, /* (31) ccons ::= DEFAULT scantok term */ + 208, /* (32) ccons ::= DEFAULT LP expr RP */ + 208, /* (33) ccons ::= DEFAULT PLUS scantok term */ + 208, /* (34) ccons ::= DEFAULT MINUS scantok term */ + 208, /* (35) ccons ::= DEFAULT scantok ID|INDEXED */ + 208, /* (36) ccons ::= NOT NULL onconf */ + 208, /* (37) ccons ::= PRIMARY KEY sortorder onconf autoinc */ + 208, /* (38) ccons ::= UNIQUE onconf */ + 208, /* (39) ccons ::= CHECK LP expr RP */ + 208, /* (40) ccons ::= REFERENCES nm eidlist_opt refargs */ + 208, /* (41) ccons ::= defer_subclause */ + 208, /* (42) ccons ::= COLLATE ID|STRING */ + 213, /* (43) autoinc ::= */ + 213, /* (44) autoinc ::= AUTOINCR */ + 215, /* (45) refargs ::= */ + 215, /* (46) refargs ::= refargs refarg */ + 217, /* (47) refarg ::= MATCH nm */ + 217, /* (48) refarg ::= ON INSERT refact */ + 217, /* (49) refarg ::= ON DELETE refact */ + 217, /* (50) refarg ::= ON UPDATE refact */ + 218, /* (51) refact ::= SET NULL */ + 218, /* (52) refact ::= SET DEFAULT */ + 218, /* (53) refact ::= CASCADE */ + 218, /* (54) refact ::= RESTRICT */ + 218, /* (55) refact ::= NO ACTION */ + 216, /* (56) defer_subclause ::= NOT DEFERRABLE init_deferred_pred_opt */ + 216, /* (57) defer_subclause ::= DEFERRABLE init_deferred_pred_opt */ + 219, /* (58) init_deferred_pred_opt ::= */ + 219, /* (59) init_deferred_pred_opt ::= INITIALLY DEFERRED */ + 219, /* (60) init_deferred_pred_opt ::= INITIALLY IMMEDIATE */ + 196, /* (61) conslist_opt ::= */ + 221, /* (62) tconscomma ::= COMMA */ + 222, /* (63) tcons ::= CONSTRAINT nm */ + 222, /* (64) tcons ::= PRIMARY KEY LP sortlist autoinc RP onconf */ + 222, /* (65) tcons ::= UNIQUE LP sortlist RP onconf */ + 222, /* (66) tcons ::= CHECK LP expr RP onconf */ + 222, /* (67) tcons ::= FOREIGN KEY LP eidlist RP REFERENCES nm eidlist_opt refargs defer_subclause_opt */ + 225, /* (68) defer_subclause_opt ::= */ + 211, /* (69) onconf ::= */ + 211, /* (70) onconf ::= ON CONFLICT resolvetype */ + 226, /* (71) orconf ::= */ + 226, /* (72) orconf ::= OR resolvetype */ + 227, /* (73) resolvetype ::= IGNORE */ + 227, /* (74) resolvetype ::= REPLACE */ + 184, /* (75) cmd ::= DROP TABLE ifexists fullname */ + 229, /* (76) ifexists ::= IF EXISTS */ + 229, /* (77) ifexists ::= */ + 184, /* (78) cmd ::= createkw temp VIEW ifnotexists nm dbnm eidlist_opt AS select */ + 184, /* (79) cmd ::= DROP VIEW ifexists fullname */ + 184, /* (80) cmd ::= select */ + 198, /* (81) select ::= WITH wqlist selectnowith */ + 198, /* (82) select ::= WITH RECURSIVE wqlist selectnowith */ + 198, /* (83) select ::= selectnowith */ + 231, /* (84) selectnowith ::= selectnowith multiselect_op oneselect */ + 234, /* (85) multiselect_op ::= UNION */ + 234, /* (86) multiselect_op ::= UNION ALL */ + 234, /* (87) multiselect_op ::= EXCEPT|INTERSECT */ + 232, /* (88) oneselect ::= SELECT distinct selcollist from where_opt groupby_opt having_opt orderby_opt limit_opt */ + 232, /* (89) oneselect ::= SELECT distinct selcollist from where_opt groupby_opt having_opt window_clause orderby_opt limit_opt */ + 244, /* (90) values ::= VALUES LP nexprlist RP */ + 244, /* (91) values ::= values COMMA LP nexprlist RP */ + 235, /* (92) distinct ::= DISTINCT */ + 235, /* (93) distinct ::= ALL */ + 235, /* (94) distinct ::= */ + 246, /* (95) sclp ::= */ + 236, /* (96) selcollist ::= sclp scanpt expr scanpt as */ + 236, /* (97) selcollist ::= sclp scanpt STAR */ + 236, /* (98) selcollist ::= sclp scanpt nm DOT STAR */ + 247, /* (99) as ::= AS nm */ + 247, /* (100) as ::= */ + 237, /* (101) from ::= */ + 237, /* (102) from ::= FROM seltablist */ + 249, /* (103) stl_prefix ::= seltablist joinop */ + 249, /* (104) stl_prefix ::= */ + 248, /* (105) seltablist ::= stl_prefix nm dbnm as indexed_opt on_opt using_opt */ + 248, /* (106) seltablist ::= stl_prefix nm dbnm LP exprlist RP as on_opt using_opt */ + 248, /* (107) seltablist ::= stl_prefix LP select RP as on_opt using_opt */ + 248, /* (108) seltablist ::= stl_prefix LP seltablist RP as on_opt using_opt */ + 194, /* (109) dbnm ::= */ + 194, /* (110) dbnm ::= DOT nm */ + 230, /* (111) fullname ::= nm */ + 230, /* (112) fullname ::= nm DOT nm */ + 255, /* (113) xfullname ::= nm */ + 255, /* (114) xfullname ::= nm DOT nm */ + 255, /* (115) xfullname ::= nm DOT nm AS nm */ + 255, /* (116) xfullname ::= nm AS nm */ + 250, /* (117) joinop ::= COMMA|JOIN */ + 250, /* (118) joinop ::= JOIN_KW JOIN */ + 250, /* (119) joinop ::= JOIN_KW nm JOIN */ + 250, /* (120) joinop ::= JOIN_KW nm nm JOIN */ + 252, /* (121) on_opt ::= ON expr */ + 252, /* (122) on_opt ::= */ + 251, /* (123) indexed_opt ::= */ + 251, /* (124) indexed_opt ::= INDEXED BY nm */ + 251, /* (125) indexed_opt ::= NOT INDEXED */ + 253, /* (126) using_opt ::= USING LP idlist RP */ + 253, /* (127) using_opt ::= */ + 241, /* (128) orderby_opt ::= */ + 241, /* (129) orderby_opt ::= ORDER BY sortlist */ + 223, /* (130) sortlist ::= sortlist COMMA expr sortorder nulls */ + 223, /* (131) sortlist ::= expr sortorder nulls */ + 212, /* (132) sortorder ::= ASC */ + 212, /* (133) sortorder ::= DESC */ + 212, /* (134) sortorder ::= */ + 257, /* (135) nulls ::= NULLS FIRST */ + 257, /* (136) nulls ::= NULLS LAST */ + 257, /* (137) nulls ::= */ + 239, /* (138) groupby_opt ::= */ + 239, /* (139) groupby_opt ::= GROUP BY nexprlist */ + 240, /* (140) having_opt ::= */ + 240, /* (141) having_opt ::= HAVING expr */ + 242, /* (142) limit_opt ::= */ + 242, /* (143) limit_opt ::= LIMIT expr */ + 242, /* (144) limit_opt ::= LIMIT expr OFFSET expr */ + 242, /* (145) limit_opt ::= LIMIT expr COMMA expr */ + 184, /* (146) cmd ::= with DELETE FROM xfullname indexed_opt where_opt */ + 238, /* (147) where_opt ::= */ + 238, /* (148) where_opt ::= WHERE expr */ + 184, /* (149) cmd ::= with UPDATE orconf xfullname indexed_opt SET setlist where_opt */ + 259, /* (150) setlist ::= setlist COMMA nm EQ expr */ + 259, /* (151) setlist ::= setlist COMMA LP idlist RP EQ expr */ + 259, /* (152) setlist ::= nm EQ expr */ + 259, /* (153) setlist ::= LP idlist RP EQ expr */ + 184, /* (154) cmd ::= with insert_cmd INTO xfullname idlist_opt select upsert */ + 184, /* (155) cmd ::= with insert_cmd INTO xfullname idlist_opt DEFAULT VALUES */ + 262, /* (156) upsert ::= */ + 262, /* (157) upsert ::= ON CONFLICT LP sortlist RP where_opt DO UPDATE SET setlist where_opt */ + 262, /* (158) upsert ::= ON CONFLICT LP sortlist RP where_opt DO NOTHING */ + 262, /* (159) upsert ::= ON CONFLICT DO NOTHING */ + 260, /* (160) insert_cmd ::= INSERT orconf */ + 260, /* (161) insert_cmd ::= REPLACE */ + 261, /* (162) idlist_opt ::= */ + 261, /* (163) idlist_opt ::= LP idlist RP */ + 256, /* (164) idlist ::= idlist COMMA nm */ + 256, /* (165) idlist ::= nm */ + 210, /* (166) expr ::= LP expr RP */ + 210, /* (167) expr ::= ID|INDEXED */ + 210, /* (168) expr ::= JOIN_KW */ + 210, /* (169) expr ::= nm DOT nm */ + 210, /* (170) expr ::= nm DOT nm DOT nm */ + 209, /* (171) term ::= NULL|FLOAT|BLOB */ + 209, /* (172) term ::= STRING */ + 209, /* (173) term ::= INTEGER */ + 210, /* (174) expr ::= VARIABLE */ + 210, /* (175) expr ::= expr COLLATE ID|STRING */ + 210, /* (176) expr ::= CAST LP expr AS typetoken RP */ + 210, /* (177) expr ::= ID|INDEXED LP distinct exprlist RP */ + 210, /* (178) expr ::= ID|INDEXED LP STAR RP */ + 210, /* (179) expr ::= ID|INDEXED LP distinct exprlist RP filter_over */ + 210, /* (180) expr ::= ID|INDEXED LP STAR RP filter_over */ + 209, /* (181) term ::= CTIME_KW */ + 210, /* (182) expr ::= LP nexprlist COMMA expr RP */ + 210, /* (183) expr ::= expr AND expr */ + 210, /* (184) expr ::= expr OR expr */ + 210, /* (185) expr ::= expr LT|GT|GE|LE expr */ + 210, /* (186) expr ::= expr EQ|NE expr */ + 210, /* (187) expr ::= expr BITAND|BITOR|LSHIFT|RSHIFT expr */ + 210, /* (188) expr ::= expr PLUS|MINUS expr */ + 210, /* (189) expr ::= expr STAR|SLASH|REM expr */ + 210, /* (190) expr ::= expr CONCAT expr */ + 264, /* (191) likeop ::= NOT LIKE_KW|MATCH */ + 210, /* (192) expr ::= expr likeop expr */ + 210, /* (193) expr ::= expr likeop expr ESCAPE expr */ + 210, /* (194) expr ::= expr ISNULL|NOTNULL */ + 210, /* (195) expr ::= expr NOT NULL */ + 210, /* (196) expr ::= expr IS expr */ + 210, /* (197) expr ::= expr IS NOT expr */ + 210, /* (198) expr ::= NOT expr */ + 210, /* (199) expr ::= BITNOT expr */ + 210, /* (200) expr ::= PLUS|MINUS expr */ + 265, /* (201) between_op ::= BETWEEN */ + 265, /* (202) between_op ::= NOT BETWEEN */ + 210, /* (203) expr ::= expr between_op expr AND expr */ + 266, /* (204) in_op ::= IN */ + 266, /* (205) in_op ::= NOT IN */ + 210, /* (206) expr ::= expr in_op LP exprlist RP */ + 210, /* (207) expr ::= LP select RP */ + 210, /* (208) expr ::= expr in_op LP select RP */ + 210, /* (209) expr ::= expr in_op nm dbnm paren_exprlist */ + 210, /* (210) expr ::= EXISTS LP select RP */ + 210, /* (211) expr ::= CASE case_operand case_exprlist case_else END */ + 269, /* (212) case_exprlist ::= case_exprlist WHEN expr THEN expr */ + 269, /* (213) case_exprlist ::= WHEN expr THEN expr */ + 270, /* (214) case_else ::= ELSE expr */ + 270, /* (215) case_else ::= */ + 268, /* (216) case_operand ::= expr */ + 268, /* (217) case_operand ::= */ + 254, /* (218) exprlist ::= */ + 245, /* (219) nexprlist ::= nexprlist COMMA expr */ + 245, /* (220) nexprlist ::= expr */ + 267, /* (221) paren_exprlist ::= */ + 267, /* (222) paren_exprlist ::= LP exprlist RP */ + 184, /* (223) cmd ::= createkw uniqueflag INDEX ifnotexists nm dbnm ON nm LP sortlist RP where_opt */ + 271, /* (224) uniqueflag ::= UNIQUE */ + 271, /* (225) uniqueflag ::= */ + 214, /* (226) eidlist_opt ::= */ + 214, /* (227) eidlist_opt ::= LP eidlist RP */ + 224, /* (228) eidlist ::= eidlist COMMA nm collate sortorder */ + 224, /* (229) eidlist ::= nm collate sortorder */ + 272, /* (230) collate ::= */ + 272, /* (231) collate ::= COLLATE ID|STRING */ + 184, /* (232) cmd ::= DROP INDEX ifexists fullname */ + 184, /* (233) cmd ::= VACUUM vinto */ + 184, /* (234) cmd ::= VACUUM nm vinto */ + 273, /* (235) vinto ::= INTO expr */ + 273, /* (236) vinto ::= */ + 184, /* (237) cmd ::= PRAGMA nm dbnm */ + 184, /* (238) cmd ::= PRAGMA nm dbnm EQ nmnum */ + 184, /* (239) cmd ::= PRAGMA nm dbnm LP nmnum RP */ + 184, /* (240) cmd ::= PRAGMA nm dbnm EQ minus_num */ + 184, /* (241) cmd ::= PRAGMA nm dbnm LP minus_num RP */ + 204, /* (242) plus_num ::= PLUS INTEGER|FLOAT */ + 205, /* (243) minus_num ::= MINUS INTEGER|FLOAT */ + 184, /* (244) cmd ::= createkw trigger_decl BEGIN trigger_cmd_list END */ + 275, /* (245) trigger_decl ::= temp TRIGGER ifnotexists nm dbnm trigger_time trigger_event ON fullname foreach_clause when_clause */ + 277, /* (246) trigger_time ::= BEFORE|AFTER */ + 277, /* (247) trigger_time ::= INSTEAD OF */ + 277, /* (248) trigger_time ::= */ + 278, /* (249) trigger_event ::= DELETE|INSERT */ + 278, /* (250) trigger_event ::= UPDATE */ + 278, /* (251) trigger_event ::= UPDATE OF idlist */ + 280, /* (252) when_clause ::= */ + 280, /* (253) when_clause ::= WHEN expr */ + 276, /* (254) trigger_cmd_list ::= trigger_cmd_list trigger_cmd SEMI */ + 276, /* (255) trigger_cmd_list ::= trigger_cmd SEMI */ + 282, /* (256) trnm ::= nm DOT nm */ + 283, /* (257) tridxby ::= INDEXED BY nm */ + 283, /* (258) tridxby ::= NOT INDEXED */ + 281, /* (259) trigger_cmd ::= UPDATE orconf trnm tridxby SET setlist where_opt scanpt */ + 281, /* (260) trigger_cmd ::= scanpt insert_cmd INTO trnm idlist_opt select upsert scanpt */ + 281, /* (261) trigger_cmd ::= DELETE FROM trnm tridxby where_opt scanpt */ + 281, /* (262) trigger_cmd ::= scanpt select scanpt */ + 210, /* (263) expr ::= RAISE LP IGNORE RP */ + 210, /* (264) expr ::= RAISE LP raisetype COMMA nm RP */ + 228, /* (265) raisetype ::= ROLLBACK */ + 228, /* (266) raisetype ::= ABORT */ + 228, /* (267) raisetype ::= FAIL */ + 184, /* (268) cmd ::= DROP TRIGGER ifexists fullname */ + 184, /* (269) cmd ::= ATTACH database_kw_opt expr AS expr key_opt */ + 184, /* (270) cmd ::= DETACH database_kw_opt expr */ + 285, /* (271) key_opt ::= */ + 285, /* (272) key_opt ::= KEY expr */ + 184, /* (273) cmd ::= REINDEX */ + 184, /* (274) cmd ::= REINDEX nm dbnm */ + 184, /* (275) cmd ::= ANALYZE */ + 184, /* (276) cmd ::= ANALYZE nm dbnm */ + 184, /* (277) cmd ::= ALTER TABLE fullname RENAME TO nm */ + 184, /* (278) cmd ::= ALTER TABLE add_column_fullname ADD kwcolumn_opt columnname carglist */ + 286, /* (279) add_column_fullname ::= fullname */ + 184, /* (280) cmd ::= ALTER TABLE fullname RENAME kwcolumn_opt nm TO nm */ + 184, /* (281) cmd ::= create_vtab */ + 184, /* (282) cmd ::= create_vtab LP vtabarglist RP */ + 288, /* (283) create_vtab ::= createkw VIRTUAL TABLE ifnotexists nm dbnm USING nm */ + 290, /* (284) vtabarg ::= */ + 291, /* (285) vtabargtoken ::= ANY */ + 291, /* (286) vtabargtoken ::= lp anylist RP */ + 292, /* (287) lp ::= LP */ + 258, /* (288) with ::= WITH wqlist */ + 258, /* (289) with ::= WITH RECURSIVE wqlist */ + 233, /* (290) wqlist ::= nm eidlist_opt AS LP select RP */ + 233, /* (291) wqlist ::= wqlist COMMA nm eidlist_opt AS LP select RP */ + 294, /* (292) windowdefn_list ::= windowdefn */ + 294, /* (293) windowdefn_list ::= windowdefn_list COMMA windowdefn */ + 295, /* (294) windowdefn ::= nm AS LP window RP */ + 296, /* (295) window ::= PARTITION BY nexprlist orderby_opt frame_opt */ + 296, /* (296) window ::= nm PARTITION BY nexprlist orderby_opt frame_opt */ + 296, /* (297) window ::= ORDER BY sortlist frame_opt */ + 296, /* (298) window ::= nm ORDER BY sortlist frame_opt */ + 296, /* (299) window ::= frame_opt */ + 296, /* (300) window ::= nm frame_opt */ + 297, /* (301) frame_opt ::= */ + 297, /* (302) frame_opt ::= range_or_rows frame_bound_s frame_exclude_opt */ + 297, /* (303) frame_opt ::= range_or_rows BETWEEN frame_bound_s AND frame_bound_e frame_exclude_opt */ + 301, /* (304) range_or_rows ::= RANGE|ROWS|GROUPS */ + 303, /* (305) frame_bound_s ::= frame_bound */ + 303, /* (306) frame_bound_s ::= UNBOUNDED PRECEDING */ + 304, /* (307) frame_bound_e ::= frame_bound */ + 304, /* (308) frame_bound_e ::= UNBOUNDED FOLLOWING */ + 302, /* (309) frame_bound ::= expr PRECEDING|FOLLOWING */ + 302, /* (310) frame_bound ::= CURRENT ROW */ + 305, /* (311) frame_exclude_opt ::= */ + 305, /* (312) frame_exclude_opt ::= EXCLUDE frame_exclude */ + 306, /* (313) frame_exclude ::= NO OTHERS */ + 306, /* (314) frame_exclude ::= CURRENT ROW */ + 306, /* (315) frame_exclude ::= GROUP|TIES */ + 243, /* (316) window_clause ::= WINDOW windowdefn_list */ + 263, /* (317) filter_over ::= filter_clause over_clause */ + 263, /* (318) filter_over ::= over_clause */ + 263, /* (319) filter_over ::= filter_clause */ + 300, /* (320) over_clause ::= OVER LP window RP */ + 300, /* (321) over_clause ::= OVER nm */ + 299, /* (322) filter_clause ::= FILTER LP WHERE expr RP */ + 179, /* (323) input ::= cmdlist */ + 180, /* (324) cmdlist ::= cmdlist ecmd */ + 180, /* (325) cmdlist ::= ecmd */ + 181, /* (326) ecmd ::= SEMI */ + 181, /* (327) ecmd ::= cmdx SEMI */ + 181, /* (328) ecmd ::= explain cmdx */ + 186, /* (329) trans_opt ::= */ + 186, /* (330) trans_opt ::= TRANSACTION */ + 186, /* (331) trans_opt ::= TRANSACTION nm */ + 188, /* (332) savepoint_opt ::= SAVEPOINT */ + 188, /* (333) savepoint_opt ::= */ + 184, /* (334) cmd ::= create_table create_table_args */ + 195, /* (335) columnlist ::= columnlist COMMA columnname carglist */ + 195, /* (336) columnlist ::= columnname carglist */ + 187, /* (337) nm ::= ID|INDEXED */ + 187, /* (338) nm ::= STRING */ + 187, /* (339) nm ::= JOIN_KW */ + 201, /* (340) typetoken ::= typename */ + 202, /* (341) typename ::= ID|STRING */ + 203, /* (342) signed ::= plus_num */ + 203, /* (343) signed ::= minus_num */ + 200, /* (344) carglist ::= carglist ccons */ + 200, /* (345) carglist ::= */ + 208, /* (346) ccons ::= NULL onconf */ + 196, /* (347) conslist_opt ::= COMMA conslist */ + 220, /* (348) conslist ::= conslist tconscomma tcons */ + 220, /* (349) conslist ::= tcons */ + 221, /* (350) tconscomma ::= */ + 225, /* (351) defer_subclause_opt ::= defer_subclause */ + 227, /* (352) resolvetype ::= raisetype */ + 231, /* (353) selectnowith ::= oneselect */ + 232, /* (354) oneselect ::= values */ + 246, /* (355) sclp ::= selcollist COMMA */ + 247, /* (356) as ::= ID|STRING */ + 210, /* (357) expr ::= term */ + 264, /* (358) likeop ::= LIKE_KW|MATCH */ + 254, /* (359) exprlist ::= nexprlist */ + 274, /* (360) nmnum ::= plus_num */ + 274, /* (361) nmnum ::= nm */ + 274, /* (362) nmnum ::= ON */ + 274, /* (363) nmnum ::= DELETE */ + 274, /* (364) nmnum ::= DEFAULT */ + 204, /* (365) plus_num ::= INTEGER|FLOAT */ + 279, /* (366) foreach_clause ::= */ + 279, /* (367) foreach_clause ::= FOR EACH ROW */ + 282, /* (368) trnm ::= nm */ + 283, /* (369) tridxby ::= */ + 284, /* (370) database_kw_opt ::= DATABASE */ + 284, /* (371) database_kw_opt ::= */ + 287, /* (372) kwcolumn_opt ::= */ + 287, /* (373) kwcolumn_opt ::= COLUMNKW */ + 289, /* (374) vtabarglist ::= vtabarg */ + 289, /* (375) vtabarglist ::= vtabarglist COMMA vtabarg */ + 290, /* (376) vtabarg ::= vtabarg vtabargtoken */ + 293, /* (377) anylist ::= */ + 293, /* (378) anylist ::= anylist LP anylist RP */ + 293, /* (379) anylist ::= anylist ANY */ + 258, /* (380) with ::= */ +}; + +/* For rule J, yyRuleInfoNRhs[J] contains the negative of the number +** of symbols on the right-hand side of that rule. */ +static const signed char yyRuleInfoNRhs[] = { + -1, /* (0) explain ::= EXPLAIN */ + -3, /* (1) explain ::= EXPLAIN QUERY PLAN */ + -1, /* (2) cmdx ::= cmd */ + -3, /* (3) cmd ::= BEGIN transtype trans_opt */ + 0, /* (4) transtype ::= */ + -1, /* (5) transtype ::= DEFERRED */ + -1, /* (6) transtype ::= IMMEDIATE */ + -1, /* (7) transtype ::= EXCLUSIVE */ + -2, /* (8) cmd ::= COMMIT|END trans_opt */ + -2, /* (9) cmd ::= ROLLBACK trans_opt */ + -2, /* (10) cmd ::= SAVEPOINT nm */ + -3, /* (11) cmd ::= RELEASE savepoint_opt nm */ + -5, /* (12) cmd ::= ROLLBACK trans_opt TO savepoint_opt nm */ + -6, /* (13) create_table ::= createkw temp TABLE ifnotexists nm dbnm */ + -1, /* (14) createkw ::= CREATE */ + 0, /* (15) ifnotexists ::= */ + -3, /* (16) ifnotexists ::= IF NOT EXISTS */ + -1, /* (17) temp ::= TEMP */ + 0, /* (18) temp ::= */ + -5, /* (19) create_table_args ::= LP columnlist conslist_opt RP table_options */ + -2, /* (20) create_table_args ::= AS select */ + 0, /* (21) table_options ::= */ + -2, /* (22) table_options ::= WITHOUT nm */ + -2, /* (23) columnname ::= nm typetoken */ + 0, /* (24) typetoken ::= */ + -4, /* (25) typetoken ::= typename LP signed RP */ + -6, /* (26) typetoken ::= typename LP signed COMMA signed RP */ + -2, /* (27) typename ::= typename ID|STRING */ + 0, /* (28) scanpt ::= */ + 0, /* (29) scantok ::= */ + -2, /* (30) ccons ::= CONSTRAINT nm */ + -3, /* (31) ccons ::= DEFAULT scantok term */ + -4, /* (32) ccons ::= DEFAULT LP expr RP */ + -4, /* (33) ccons ::= DEFAULT PLUS scantok term */ + -4, /* (34) ccons ::= DEFAULT MINUS scantok term */ + -3, /* (35) ccons ::= DEFAULT scantok ID|INDEXED */ + -3, /* (36) ccons ::= NOT NULL onconf */ + -5, /* (37) ccons ::= PRIMARY KEY sortorder onconf autoinc */ + -2, /* (38) ccons ::= UNIQUE onconf */ + -4, /* (39) ccons ::= CHECK LP expr RP */ + -4, /* (40) ccons ::= REFERENCES nm eidlist_opt refargs */ + -1, /* (41) ccons ::= defer_subclause */ + -2, /* (42) ccons ::= COLLATE ID|STRING */ + 0, /* (43) autoinc ::= */ + -1, /* (44) autoinc ::= AUTOINCR */ + 0, /* (45) refargs ::= */ + -2, /* (46) refargs ::= refargs refarg */ + -2, /* (47) refarg ::= MATCH nm */ + -3, /* (48) refarg ::= ON INSERT refact */ + -3, /* (49) refarg ::= ON DELETE refact */ + -3, /* (50) refarg ::= ON UPDATE refact */ + -2, /* (51) refact ::= SET NULL */ + -2, /* (52) refact ::= SET DEFAULT */ + -1, /* (53) refact ::= CASCADE */ + -1, /* (54) refact ::= RESTRICT */ + -2, /* (55) refact ::= NO ACTION */ + -3, /* (56) defer_subclause ::= NOT DEFERRABLE init_deferred_pred_opt */ + -2, /* (57) defer_subclause ::= DEFERRABLE init_deferred_pred_opt */ + 0, /* (58) init_deferred_pred_opt ::= */ + -2, /* (59) init_deferred_pred_opt ::= INITIALLY DEFERRED */ + -2, /* (60) init_deferred_pred_opt ::= INITIALLY IMMEDIATE */ + 0, /* (61) conslist_opt ::= */ + -1, /* (62) tconscomma ::= COMMA */ + -2, /* (63) tcons ::= CONSTRAINT nm */ + -7, /* (64) tcons ::= PRIMARY KEY LP sortlist autoinc RP onconf */ + -5, /* (65) tcons ::= UNIQUE LP sortlist RP onconf */ + -5, /* (66) tcons ::= CHECK LP expr RP onconf */ + -10, /* (67) tcons ::= FOREIGN KEY LP eidlist RP REFERENCES nm eidlist_opt refargs defer_subclause_opt */ + 0, /* (68) defer_subclause_opt ::= */ + 0, /* (69) onconf ::= */ + -3, /* (70) onconf ::= ON CONFLICT resolvetype */ + 0, /* (71) orconf ::= */ + -2, /* (72) orconf ::= OR resolvetype */ + -1, /* (73) resolvetype ::= IGNORE */ + -1, /* (74) resolvetype ::= REPLACE */ + -4, /* (75) cmd ::= DROP TABLE ifexists fullname */ + -2, /* (76) ifexists ::= IF EXISTS */ + 0, /* (77) ifexists ::= */ + -9, /* (78) cmd ::= createkw temp VIEW ifnotexists nm dbnm eidlist_opt AS select */ + -4, /* (79) cmd ::= DROP VIEW ifexists fullname */ + -1, /* (80) cmd ::= select */ + -3, /* (81) select ::= WITH wqlist selectnowith */ + -4, /* (82) select ::= WITH RECURSIVE wqlist selectnowith */ + -1, /* (83) select ::= selectnowith */ + -3, /* (84) selectnowith ::= selectnowith multiselect_op oneselect */ + -1, /* (85) multiselect_op ::= UNION */ + -2, /* (86) multiselect_op ::= UNION ALL */ + -1, /* (87) multiselect_op ::= EXCEPT|INTERSECT */ + -9, /* (88) oneselect ::= SELECT distinct selcollist from where_opt groupby_opt having_opt orderby_opt limit_opt */ + -10, /* (89) oneselect ::= SELECT distinct selcollist from where_opt groupby_opt having_opt window_clause orderby_opt limit_opt */ + -4, /* (90) values ::= VALUES LP nexprlist RP */ + -5, /* (91) values ::= values COMMA LP nexprlist RP */ + -1, /* (92) distinct ::= DISTINCT */ + -1, /* (93) distinct ::= ALL */ + 0, /* (94) distinct ::= */ + 0, /* (95) sclp ::= */ + -5, /* (96) selcollist ::= sclp scanpt expr scanpt as */ + -3, /* (97) selcollist ::= sclp scanpt STAR */ + -5, /* (98) selcollist ::= sclp scanpt nm DOT STAR */ + -2, /* (99) as ::= AS nm */ + 0, /* (100) as ::= */ + 0, /* (101) from ::= */ + -2, /* (102) from ::= FROM seltablist */ + -2, /* (103) stl_prefix ::= seltablist joinop */ + 0, /* (104) stl_prefix ::= */ + -7, /* (105) seltablist ::= stl_prefix nm dbnm as indexed_opt on_opt using_opt */ + -9, /* (106) seltablist ::= stl_prefix nm dbnm LP exprlist RP as on_opt using_opt */ + -7, /* (107) seltablist ::= stl_prefix LP select RP as on_opt using_opt */ + -7, /* (108) seltablist ::= stl_prefix LP seltablist RP as on_opt using_opt */ + 0, /* (109) dbnm ::= */ + -2, /* (110) dbnm ::= DOT nm */ + -1, /* (111) fullname ::= nm */ + -3, /* (112) fullname ::= nm DOT nm */ + -1, /* (113) xfullname ::= nm */ + -3, /* (114) xfullname ::= nm DOT nm */ + -5, /* (115) xfullname ::= nm DOT nm AS nm */ + -3, /* (116) xfullname ::= nm AS nm */ + -1, /* (117) joinop ::= COMMA|JOIN */ + -2, /* (118) joinop ::= JOIN_KW JOIN */ + -3, /* (119) joinop ::= JOIN_KW nm JOIN */ + -4, /* (120) joinop ::= JOIN_KW nm nm JOIN */ + -2, /* (121) on_opt ::= ON expr */ + 0, /* (122) on_opt ::= */ + 0, /* (123) indexed_opt ::= */ + -3, /* (124) indexed_opt ::= INDEXED BY nm */ + -2, /* (125) indexed_opt ::= NOT INDEXED */ + -4, /* (126) using_opt ::= USING LP idlist RP */ + 0, /* (127) using_opt ::= */ + 0, /* (128) orderby_opt ::= */ + -3, /* (129) orderby_opt ::= ORDER BY sortlist */ + -5, /* (130) sortlist ::= sortlist COMMA expr sortorder nulls */ + -3, /* (131) sortlist ::= expr sortorder nulls */ + -1, /* (132) sortorder ::= ASC */ + -1, /* (133) sortorder ::= DESC */ + 0, /* (134) sortorder ::= */ + -2, /* (135) nulls ::= NULLS FIRST */ + -2, /* (136) nulls ::= NULLS LAST */ + 0, /* (137) nulls ::= */ + 0, /* (138) groupby_opt ::= */ + -3, /* (139) groupby_opt ::= GROUP BY nexprlist */ + 0, /* (140) having_opt ::= */ + -2, /* (141) having_opt ::= HAVING expr */ + 0, /* (142) limit_opt ::= */ + -2, /* (143) limit_opt ::= LIMIT expr */ + -4, /* (144) limit_opt ::= LIMIT expr OFFSET expr */ + -4, /* (145) limit_opt ::= LIMIT expr COMMA expr */ + -6, /* (146) cmd ::= with DELETE FROM xfullname indexed_opt where_opt */ + 0, /* (147) where_opt ::= */ + -2, /* (148) where_opt ::= WHERE expr */ + -8, /* (149) cmd ::= with UPDATE orconf xfullname indexed_opt SET setlist where_opt */ + -5, /* (150) setlist ::= setlist COMMA nm EQ expr */ + -7, /* (151) setlist ::= setlist COMMA LP idlist RP EQ expr */ + -3, /* (152) setlist ::= nm EQ expr */ + -5, /* (153) setlist ::= LP idlist RP EQ expr */ + -7, /* (154) cmd ::= with insert_cmd INTO xfullname idlist_opt select upsert */ + -7, /* (155) cmd ::= with insert_cmd INTO xfullname idlist_opt DEFAULT VALUES */ + 0, /* (156) upsert ::= */ + -11, /* (157) upsert ::= ON CONFLICT LP sortlist RP where_opt DO UPDATE SET setlist where_opt */ + -8, /* (158) upsert ::= ON CONFLICT LP sortlist RP where_opt DO NOTHING */ + -4, /* (159) upsert ::= ON CONFLICT DO NOTHING */ + -2, /* (160) insert_cmd ::= INSERT orconf */ + -1, /* (161) insert_cmd ::= REPLACE */ + 0, /* (162) idlist_opt ::= */ + -3, /* (163) idlist_opt ::= LP idlist RP */ + -3, /* (164) idlist ::= idlist COMMA nm */ + -1, /* (165) idlist ::= nm */ + -3, /* (166) expr ::= LP expr RP */ + -1, /* (167) expr ::= ID|INDEXED */ + -1, /* (168) expr ::= JOIN_KW */ + -3, /* (169) expr ::= nm DOT nm */ + -5, /* (170) expr ::= nm DOT nm DOT nm */ + -1, /* (171) term ::= NULL|FLOAT|BLOB */ + -1, /* (172) term ::= STRING */ + -1, /* (173) term ::= INTEGER */ + -1, /* (174) expr ::= VARIABLE */ + -3, /* (175) expr ::= expr COLLATE ID|STRING */ + -6, /* (176) expr ::= CAST LP expr AS typetoken RP */ + -5, /* (177) expr ::= ID|INDEXED LP distinct exprlist RP */ + -4, /* (178) expr ::= ID|INDEXED LP STAR RP */ + -6, /* (179) expr ::= ID|INDEXED LP distinct exprlist RP filter_over */ + -5, /* (180) expr ::= ID|INDEXED LP STAR RP filter_over */ + -1, /* (181) term ::= CTIME_KW */ + -5, /* (182) expr ::= LP nexprlist COMMA expr RP */ + -3, /* (183) expr ::= expr AND expr */ + -3, /* (184) expr ::= expr OR expr */ + -3, /* (185) expr ::= expr LT|GT|GE|LE expr */ + -3, /* (186) expr ::= expr EQ|NE expr */ + -3, /* (187) expr ::= expr BITAND|BITOR|LSHIFT|RSHIFT expr */ + -3, /* (188) expr ::= expr PLUS|MINUS expr */ + -3, /* (189) expr ::= expr STAR|SLASH|REM expr */ + -3, /* (190) expr ::= expr CONCAT expr */ + -2, /* (191) likeop ::= NOT LIKE_KW|MATCH */ + -3, /* (192) expr ::= expr likeop expr */ + -5, /* (193) expr ::= expr likeop expr ESCAPE expr */ + -2, /* (194) expr ::= expr ISNULL|NOTNULL */ + -3, /* (195) expr ::= expr NOT NULL */ + -3, /* (196) expr ::= expr IS expr */ + -4, /* (197) expr ::= expr IS NOT expr */ + -2, /* (198) expr ::= NOT expr */ + -2, /* (199) expr ::= BITNOT expr */ + -2, /* (200) expr ::= PLUS|MINUS expr */ + -1, /* (201) between_op ::= BETWEEN */ + -2, /* (202) between_op ::= NOT BETWEEN */ + -5, /* (203) expr ::= expr between_op expr AND expr */ + -1, /* (204) in_op ::= IN */ + -2, /* (205) in_op ::= NOT IN */ + -5, /* (206) expr ::= expr in_op LP exprlist RP */ + -3, /* (207) expr ::= LP select RP */ + -5, /* (208) expr ::= expr in_op LP select RP */ + -5, /* (209) expr ::= expr in_op nm dbnm paren_exprlist */ + -4, /* (210) expr ::= EXISTS LP select RP */ + -5, /* (211) expr ::= CASE case_operand case_exprlist case_else END */ + -5, /* (212) case_exprlist ::= case_exprlist WHEN expr THEN expr */ + -4, /* (213) case_exprlist ::= WHEN expr THEN expr */ + -2, /* (214) case_else ::= ELSE expr */ + 0, /* (215) case_else ::= */ + -1, /* (216) case_operand ::= expr */ + 0, /* (217) case_operand ::= */ + 0, /* (218) exprlist ::= */ + -3, /* (219) nexprlist ::= nexprlist COMMA expr */ + -1, /* (220) nexprlist ::= expr */ + 0, /* (221) paren_exprlist ::= */ + -3, /* (222) paren_exprlist ::= LP exprlist RP */ + -12, /* (223) cmd ::= createkw uniqueflag INDEX ifnotexists nm dbnm ON nm LP sortlist RP where_opt */ + -1, /* (224) uniqueflag ::= UNIQUE */ + 0, /* (225) uniqueflag ::= */ + 0, /* (226) eidlist_opt ::= */ + -3, /* (227) eidlist_opt ::= LP eidlist RP */ + -5, /* (228) eidlist ::= eidlist COMMA nm collate sortorder */ + -3, /* (229) eidlist ::= nm collate sortorder */ + 0, /* (230) collate ::= */ + -2, /* (231) collate ::= COLLATE ID|STRING */ + -4, /* (232) cmd ::= DROP INDEX ifexists fullname */ + -2, /* (233) cmd ::= VACUUM vinto */ + -3, /* (234) cmd ::= VACUUM nm vinto */ + -2, /* (235) vinto ::= INTO expr */ + 0, /* (236) vinto ::= */ + -3, /* (237) cmd ::= PRAGMA nm dbnm */ + -5, /* (238) cmd ::= PRAGMA nm dbnm EQ nmnum */ + -6, /* (239) cmd ::= PRAGMA nm dbnm LP nmnum RP */ + -5, /* (240) cmd ::= PRAGMA nm dbnm EQ minus_num */ + -6, /* (241) cmd ::= PRAGMA nm dbnm LP minus_num RP */ + -2, /* (242) plus_num ::= PLUS INTEGER|FLOAT */ + -2, /* (243) minus_num ::= MINUS INTEGER|FLOAT */ + -5, /* (244) cmd ::= createkw trigger_decl BEGIN trigger_cmd_list END */ + -11, /* (245) trigger_decl ::= temp TRIGGER ifnotexists nm dbnm trigger_time trigger_event ON fullname foreach_clause when_clause */ + -1, /* (246) trigger_time ::= BEFORE|AFTER */ + -2, /* (247) trigger_time ::= INSTEAD OF */ + 0, /* (248) trigger_time ::= */ + -1, /* (249) trigger_event ::= DELETE|INSERT */ + -1, /* (250) trigger_event ::= UPDATE */ + -3, /* (251) trigger_event ::= UPDATE OF idlist */ + 0, /* (252) when_clause ::= */ + -2, /* (253) when_clause ::= WHEN expr */ + -3, /* (254) trigger_cmd_list ::= trigger_cmd_list trigger_cmd SEMI */ + -2, /* (255) trigger_cmd_list ::= trigger_cmd SEMI */ + -3, /* (256) trnm ::= nm DOT nm */ + -3, /* (257) tridxby ::= INDEXED BY nm */ + -2, /* (258) tridxby ::= NOT INDEXED */ + -8, /* (259) trigger_cmd ::= UPDATE orconf trnm tridxby SET setlist where_opt scanpt */ + -8, /* (260) trigger_cmd ::= scanpt insert_cmd INTO trnm idlist_opt select upsert scanpt */ + -6, /* (261) trigger_cmd ::= DELETE FROM trnm tridxby where_opt scanpt */ + -3, /* (262) trigger_cmd ::= scanpt select scanpt */ + -4, /* (263) expr ::= RAISE LP IGNORE RP */ + -6, /* (264) expr ::= RAISE LP raisetype COMMA nm RP */ + -1, /* (265) raisetype ::= ROLLBACK */ + -1, /* (266) raisetype ::= ABORT */ + -1, /* (267) raisetype ::= FAIL */ + -4, /* (268) cmd ::= DROP TRIGGER ifexists fullname */ + -6, /* (269) cmd ::= ATTACH database_kw_opt expr AS expr key_opt */ + -3, /* (270) cmd ::= DETACH database_kw_opt expr */ + 0, /* (271) key_opt ::= */ + -2, /* (272) key_opt ::= KEY expr */ + -1, /* (273) cmd ::= REINDEX */ + -3, /* (274) cmd ::= REINDEX nm dbnm */ + -1, /* (275) cmd ::= ANALYZE */ + -3, /* (276) cmd ::= ANALYZE nm dbnm */ + -6, /* (277) cmd ::= ALTER TABLE fullname RENAME TO nm */ + -7, /* (278) cmd ::= ALTER TABLE add_column_fullname ADD kwcolumn_opt columnname carglist */ + -1, /* (279) add_column_fullname ::= fullname */ + -8, /* (280) cmd ::= ALTER TABLE fullname RENAME kwcolumn_opt nm TO nm */ + -1, /* (281) cmd ::= create_vtab */ + -4, /* (282) cmd ::= create_vtab LP vtabarglist RP */ + -8, /* (283) create_vtab ::= createkw VIRTUAL TABLE ifnotexists nm dbnm USING nm */ + 0, /* (284) vtabarg ::= */ + -1, /* (285) vtabargtoken ::= ANY */ + -3, /* (286) vtabargtoken ::= lp anylist RP */ + -1, /* (287) lp ::= LP */ + -2, /* (288) with ::= WITH wqlist */ + -3, /* (289) with ::= WITH RECURSIVE wqlist */ + -6, /* (290) wqlist ::= nm eidlist_opt AS LP select RP */ + -8, /* (291) wqlist ::= wqlist COMMA nm eidlist_opt AS LP select RP */ + -1, /* (292) windowdefn_list ::= windowdefn */ + -3, /* (293) windowdefn_list ::= windowdefn_list COMMA windowdefn */ + -5, /* (294) windowdefn ::= nm AS LP window RP */ + -5, /* (295) window ::= PARTITION BY nexprlist orderby_opt frame_opt */ + -6, /* (296) window ::= nm PARTITION BY nexprlist orderby_opt frame_opt */ + -4, /* (297) window ::= ORDER BY sortlist frame_opt */ + -5, /* (298) window ::= nm ORDER BY sortlist frame_opt */ + -1, /* (299) window ::= frame_opt */ + -2, /* (300) window ::= nm frame_opt */ + 0, /* (301) frame_opt ::= */ + -3, /* (302) frame_opt ::= range_or_rows frame_bound_s frame_exclude_opt */ + -6, /* (303) frame_opt ::= range_or_rows BETWEEN frame_bound_s AND frame_bound_e frame_exclude_opt */ + -1, /* (304) range_or_rows ::= RANGE|ROWS|GROUPS */ + -1, /* (305) frame_bound_s ::= frame_bound */ + -2, /* (306) frame_bound_s ::= UNBOUNDED PRECEDING */ + -1, /* (307) frame_bound_e ::= frame_bound */ + -2, /* (308) frame_bound_e ::= UNBOUNDED FOLLOWING */ + -2, /* (309) frame_bound ::= expr PRECEDING|FOLLOWING */ + -2, /* (310) frame_bound ::= CURRENT ROW */ + 0, /* (311) frame_exclude_opt ::= */ + -2, /* (312) frame_exclude_opt ::= EXCLUDE frame_exclude */ + -2, /* (313) frame_exclude ::= NO OTHERS */ + -2, /* (314) frame_exclude ::= CURRENT ROW */ + -1, /* (315) frame_exclude ::= GROUP|TIES */ + -2, /* (316) window_clause ::= WINDOW windowdefn_list */ + -2, /* (317) filter_over ::= filter_clause over_clause */ + -1, /* (318) filter_over ::= over_clause */ + -1, /* (319) filter_over ::= filter_clause */ + -4, /* (320) over_clause ::= OVER LP window RP */ + -2, /* (321) over_clause ::= OVER nm */ + -5, /* (322) filter_clause ::= FILTER LP WHERE expr RP */ + -1, /* (323) input ::= cmdlist */ + -2, /* (324) cmdlist ::= cmdlist ecmd */ + -1, /* (325) cmdlist ::= ecmd */ + -1, /* (326) ecmd ::= SEMI */ + -2, /* (327) ecmd ::= cmdx SEMI */ + -2, /* (328) ecmd ::= explain cmdx */ + 0, /* (329) trans_opt ::= */ + -1, /* (330) trans_opt ::= TRANSACTION */ + -2, /* (331) trans_opt ::= TRANSACTION nm */ + -1, /* (332) savepoint_opt ::= SAVEPOINT */ + 0, /* (333) savepoint_opt ::= */ + -2, /* (334) cmd ::= create_table create_table_args */ + -4, /* (335) columnlist ::= columnlist COMMA columnname carglist */ + -2, /* (336) columnlist ::= columnname carglist */ + -1, /* (337) nm ::= ID|INDEXED */ + -1, /* (338) nm ::= STRING */ + -1, /* (339) nm ::= JOIN_KW */ + -1, /* (340) typetoken ::= typename */ + -1, /* (341) typename ::= ID|STRING */ + -1, /* (342) signed ::= plus_num */ + -1, /* (343) signed ::= minus_num */ + -2, /* (344) carglist ::= carglist ccons */ + 0, /* (345) carglist ::= */ + -2, /* (346) ccons ::= NULL onconf */ + -2, /* (347) conslist_opt ::= COMMA conslist */ + -3, /* (348) conslist ::= conslist tconscomma tcons */ + -1, /* (349) conslist ::= tcons */ + 0, /* (350) tconscomma ::= */ + -1, /* (351) defer_subclause_opt ::= defer_subclause */ + -1, /* (352) resolvetype ::= raisetype */ + -1, /* (353) selectnowith ::= oneselect */ + -1, /* (354) oneselect ::= values */ + -2, /* (355) sclp ::= selcollist COMMA */ + -1, /* (356) as ::= ID|STRING */ + -1, /* (357) expr ::= term */ + -1, /* (358) likeop ::= LIKE_KW|MATCH */ + -1, /* (359) exprlist ::= nexprlist */ + -1, /* (360) nmnum ::= plus_num */ + -1, /* (361) nmnum ::= nm */ + -1, /* (362) nmnum ::= ON */ + -1, /* (363) nmnum ::= DELETE */ + -1, /* (364) nmnum ::= DEFAULT */ + -1, /* (365) plus_num ::= INTEGER|FLOAT */ + 0, /* (366) foreach_clause ::= */ + -3, /* (367) foreach_clause ::= FOR EACH ROW */ + -1, /* (368) trnm ::= nm */ + 0, /* (369) tridxby ::= */ + -1, /* (370) database_kw_opt ::= DATABASE */ + 0, /* (371) database_kw_opt ::= */ + 0, /* (372) kwcolumn_opt ::= */ + -1, /* (373) kwcolumn_opt ::= COLUMNKW */ + -1, /* (374) vtabarglist ::= vtabarg */ + -3, /* (375) vtabarglist ::= vtabarglist COMMA vtabarg */ + -2, /* (376) vtabarg ::= vtabarg vtabargtoken */ + 0, /* (377) anylist ::= */ + -4, /* (378) anylist ::= anylist LP anylist RP */ + -2, /* (379) anylist ::= anylist ANY */ + 0, /* (380) with ::= */ +}; + +static void yy_accept(yyParser*); /* Forward Declaration */ + +/* +** Perform a reduce action and the shift that must immediately +** follow the reduce. +** +** The yyLookahead and yyLookaheadToken parameters provide reduce actions +** access to the lookahead token (if any). The yyLookahead will be YYNOCODE +** if the lookahead token has already been consumed. As this procedure is +** only called from one place, optimizing compilers will in-line it, which +** means that the extra parameters have no performance impact. +*/ +static YYACTIONTYPE yy_reduce( + yyParser *yypParser, /* The parser */ + unsigned int yyruleno, /* Number of the rule by which to reduce */ + int yyLookahead, /* Lookahead token, or YYNOCODE if none */ + sqlite3ParserTOKENTYPE yyLookaheadToken /* Value of the lookahead token */ + sqlite3ParserCTX_PDECL /* %extra_context */ +){ + int yygoto; /* The next state */ + YYACTIONTYPE yyact; /* The next action */ + yyStackEntry *yymsp; /* The top of the parser's stack */ + int yysize; /* Amount to pop the stack */ + sqlite3ParserARG_FETCH + (void)yyLookahead; + (void)yyLookaheadToken; + yymsp = yypParser->yytos; +#ifndef NDEBUG + if( yyTraceFILE && yyruleno<(int)(sizeof(yyRuleName)/sizeof(yyRuleName[0])) ){ + yysize = yyRuleInfoNRhs[yyruleno]; + if( yysize ){ + fprintf(yyTraceFILE, "%sReduce %d [%s], go to state %d.\n", + yyTracePrompt, + yyruleno, yyRuleName[yyruleno], yymsp[yysize].stateno); + }else{ + fprintf(yyTraceFILE, "%sReduce %d [%s].\n", + yyTracePrompt, yyruleno, yyRuleName[yyruleno]); + } + } +#endif /* NDEBUG */ + + /* Check that the stack is large enough to grow by a single entry + ** if the RHS of the rule is empty. This ensures that there is room + ** enough on the stack to push the LHS value */ + if( yyRuleInfoNRhs[yyruleno]==0 ){ +#ifdef YYTRACKMAXSTACKDEPTH + if( (int)(yypParser->yytos - yypParser->yystack)>yypParser->yyhwm ){ + yypParser->yyhwm++; + assert( yypParser->yyhwm == (int)(yypParser->yytos - yypParser->yystack)); + } +#endif +#if YYSTACKDEPTH>0 + if( yypParser->yytos>=yypParser->yystackEnd ){ + yyStackOverflow(yypParser); + /* The call to yyStackOverflow() above pops the stack until it is + ** empty, causing the main parser loop to exit. So the return value + ** is never used and does not matter. */ + return 0; + } +#else + if( yypParser->yytos>=&yypParser->yystack[yypParser->yystksz-1] ){ + if( yyGrowStack(yypParser) ){ + yyStackOverflow(yypParser); + /* The call to yyStackOverflow() above pops the stack until it is + ** empty, causing the main parser loop to exit. So the return value + ** is never used and does not matter. */ + return 0; + } + yymsp = yypParser->yytos; + } +#endif + } + + switch( yyruleno ){ + /* Beginning here are the reduction cases. A typical example + ** follows: + ** case 0: + ** #line + ** { ... } // User supplied code + ** #line + ** break; + */ +/********** Begin reduce actions **********************************************/ + YYMINORTYPE yylhsminor; + case 0: /* explain ::= EXPLAIN */ +{ pParse->explain = 1; } + break; + case 1: /* explain ::= EXPLAIN QUERY PLAN */ +{ pParse->explain = 2; } + break; + case 2: /* cmdx ::= cmd */ +{ sqlite3FinishCoding(pParse); } + break; + case 3: /* cmd ::= BEGIN transtype trans_opt */ +{sqlite3BeginTransaction(pParse, yymsp[-1].minor.yy32);} + break; + case 4: /* transtype ::= */ +{yymsp[1].minor.yy32 = TK_DEFERRED;} + break; + case 5: /* transtype ::= DEFERRED */ + case 6: /* transtype ::= IMMEDIATE */ yytestcase(yyruleno==6); + case 7: /* transtype ::= EXCLUSIVE */ yytestcase(yyruleno==7); + case 304: /* range_or_rows ::= RANGE|ROWS|GROUPS */ yytestcase(yyruleno==304); +{yymsp[0].minor.yy32 = yymsp[0].major; /*A-overwrites-X*/} + break; + case 8: /* cmd ::= COMMIT|END trans_opt */ + case 9: /* cmd ::= ROLLBACK trans_opt */ yytestcase(yyruleno==9); +{sqlite3EndTransaction(pParse,yymsp[-1].major);} + break; + case 10: /* cmd ::= SAVEPOINT nm */ +{ + sqlite3Savepoint(pParse, SAVEPOINT_BEGIN, &yymsp[0].minor.yy0); +} + break; + case 11: /* cmd ::= RELEASE savepoint_opt nm */ +{ + sqlite3Savepoint(pParse, SAVEPOINT_RELEASE, &yymsp[0].minor.yy0); +} + break; + case 12: /* cmd ::= ROLLBACK trans_opt TO savepoint_opt nm */ +{ + sqlite3Savepoint(pParse, SAVEPOINT_ROLLBACK, &yymsp[0].minor.yy0); +} + break; + case 13: /* create_table ::= createkw temp TABLE ifnotexists nm dbnm */ +{ + sqlite3StartTable(pParse,&yymsp[-1].minor.yy0,&yymsp[0].minor.yy0,yymsp[-4].minor.yy32,0,0,yymsp[-2].minor.yy32); +} + break; + case 14: /* createkw ::= CREATE */ +{disableLookaside(pParse);} + break; + case 15: /* ifnotexists ::= */ + case 18: /* temp ::= */ yytestcase(yyruleno==18); + case 21: /* table_options ::= */ yytestcase(yyruleno==21); + case 43: /* autoinc ::= */ yytestcase(yyruleno==43); + case 58: /* init_deferred_pred_opt ::= */ yytestcase(yyruleno==58); + case 68: /* defer_subclause_opt ::= */ yytestcase(yyruleno==68); + case 77: /* ifexists ::= */ yytestcase(yyruleno==77); + case 94: /* distinct ::= */ yytestcase(yyruleno==94); + case 230: /* collate ::= */ yytestcase(yyruleno==230); +{yymsp[1].minor.yy32 = 0;} + break; + case 16: /* ifnotexists ::= IF NOT EXISTS */ +{yymsp[-2].minor.yy32 = 1;} + break; + case 17: /* temp ::= TEMP */ + case 44: /* autoinc ::= AUTOINCR */ yytestcase(yyruleno==44); +{yymsp[0].minor.yy32 = 1;} + break; + case 19: /* create_table_args ::= LP columnlist conslist_opt RP table_options */ +{ + sqlite3EndTable(pParse,&yymsp[-2].minor.yy0,&yymsp[-1].minor.yy0,yymsp[0].minor.yy32,0); +} + break; + case 20: /* create_table_args ::= AS select */ +{ + sqlite3EndTable(pParse,0,0,0,yymsp[0].minor.yy25); + sqlite3SelectDelete(pParse->db, yymsp[0].minor.yy25); +} + break; + case 22: /* table_options ::= WITHOUT nm */ +{ + if( yymsp[0].minor.yy0.n==5 && sqlite3_strnicmp(yymsp[0].minor.yy0.z,"rowid",5)==0 ){ + yymsp[-1].minor.yy32 = TF_WithoutRowid | TF_NoVisibleRowid; + }else{ + yymsp[-1].minor.yy32 = 0; + sqlite3ErrorMsg(pParse, "unknown table option: %.*s", yymsp[0].minor.yy0.n, yymsp[0].minor.yy0.z); + } +} + break; + case 23: /* columnname ::= nm typetoken */ +{sqlite3AddColumn(pParse,&yymsp[-1].minor.yy0,&yymsp[0].minor.yy0);} + break; + case 24: /* typetoken ::= */ + case 61: /* conslist_opt ::= */ yytestcase(yyruleno==61); + case 100: /* as ::= */ yytestcase(yyruleno==100); +{yymsp[1].minor.yy0.n = 0; yymsp[1].minor.yy0.z = 0;} + break; + case 25: /* typetoken ::= typename LP signed RP */ +{ + yymsp[-3].minor.yy0.n = (int)(&yymsp[0].minor.yy0.z[yymsp[0].minor.yy0.n] - yymsp[-3].minor.yy0.z); +} + break; + case 26: /* typetoken ::= typename LP signed COMMA signed RP */ +{ + yymsp[-5].minor.yy0.n = (int)(&yymsp[0].minor.yy0.z[yymsp[0].minor.yy0.n] - yymsp[-5].minor.yy0.z); +} + break; + case 27: /* typename ::= typename ID|STRING */ +{yymsp[-1].minor.yy0.n=yymsp[0].minor.yy0.n+(int)(yymsp[0].minor.yy0.z-yymsp[-1].minor.yy0.z);} + break; + case 28: /* scanpt ::= */ +{ + assert( yyLookahead!=YYNOCODE ); + yymsp[1].minor.yy8 = yyLookaheadToken.z; +} + break; + case 29: /* scantok ::= */ +{ + assert( yyLookahead!=YYNOCODE ); + yymsp[1].minor.yy0 = yyLookaheadToken; +} + break; + case 30: /* ccons ::= CONSTRAINT nm */ + case 63: /* tcons ::= CONSTRAINT nm */ yytestcase(yyruleno==63); +{pParse->constraintName = yymsp[0].minor.yy0;} + break; + case 31: /* ccons ::= DEFAULT scantok term */ +{sqlite3AddDefaultValue(pParse,yymsp[0].minor.yy46,yymsp[-1].minor.yy0.z,&yymsp[-1].minor.yy0.z[yymsp[-1].minor.yy0.n]);} + break; + case 32: /* ccons ::= DEFAULT LP expr RP */ +{sqlite3AddDefaultValue(pParse,yymsp[-1].minor.yy46,yymsp[-2].minor.yy0.z+1,yymsp[0].minor.yy0.z);} + break; + case 33: /* ccons ::= DEFAULT PLUS scantok term */ +{sqlite3AddDefaultValue(pParse,yymsp[0].minor.yy46,yymsp[-2].minor.yy0.z,&yymsp[-1].minor.yy0.z[yymsp[-1].minor.yy0.n]);} + break; + case 34: /* ccons ::= DEFAULT MINUS scantok term */ +{ + Expr *p = sqlite3PExpr(pParse, TK_UMINUS, yymsp[0].minor.yy46, 0); + sqlite3AddDefaultValue(pParse,p,yymsp[-2].minor.yy0.z,&yymsp[-1].minor.yy0.z[yymsp[-1].minor.yy0.n]); +} + break; + case 35: /* ccons ::= DEFAULT scantok ID|INDEXED */ +{ + Expr *p = tokenExpr(pParse, TK_STRING, yymsp[0].minor.yy0); + if( p ){ + sqlite3ExprIdToTrueFalse(p); + testcase( p->op==TK_TRUEFALSE && sqlite3ExprTruthValue(p) ); + } + sqlite3AddDefaultValue(pParse,p,yymsp[0].minor.yy0.z,yymsp[0].minor.yy0.z+yymsp[0].minor.yy0.n); +} + break; + case 36: /* ccons ::= NOT NULL onconf */ +{sqlite3AddNotNull(pParse, yymsp[0].minor.yy32);} + break; + case 37: /* ccons ::= PRIMARY KEY sortorder onconf autoinc */ +{sqlite3AddPrimaryKey(pParse,0,yymsp[-1].minor.yy32,yymsp[0].minor.yy32,yymsp[-2].minor.yy32);} + break; + case 38: /* ccons ::= UNIQUE onconf */ +{sqlite3CreateIndex(pParse,0,0,0,0,yymsp[0].minor.yy32,0,0,0,0, + SQLITE_IDXTYPE_UNIQUE);} + break; + case 39: /* ccons ::= CHECK LP expr RP */ +{sqlite3AddCheckConstraint(pParse,yymsp[-1].minor.yy46);} + break; + case 40: /* ccons ::= REFERENCES nm eidlist_opt refargs */ +{sqlite3CreateForeignKey(pParse,0,&yymsp[-2].minor.yy0,yymsp[-1].minor.yy138,yymsp[0].minor.yy32);} + break; + case 41: /* ccons ::= defer_subclause */ +{sqlite3DeferForeignKey(pParse,yymsp[0].minor.yy32);} + break; + case 42: /* ccons ::= COLLATE ID|STRING */ +{sqlite3AddCollateType(pParse, &yymsp[0].minor.yy0);} + break; + case 45: /* refargs ::= */ +{ yymsp[1].minor.yy32 = OE_None*0x0101; /* EV: R-19803-45884 */} + break; + case 46: /* refargs ::= refargs refarg */ +{ yymsp[-1].minor.yy32 = (yymsp[-1].minor.yy32 & ~yymsp[0].minor.yy495.mask) | yymsp[0].minor.yy495.value; } + break; + case 47: /* refarg ::= MATCH nm */ +{ yymsp[-1].minor.yy495.value = 0; yymsp[-1].minor.yy495.mask = 0x000000; } + break; + case 48: /* refarg ::= ON INSERT refact */ +{ yymsp[-2].minor.yy495.value = 0; yymsp[-2].minor.yy495.mask = 0x000000; } + break; + case 49: /* refarg ::= ON DELETE refact */ +{ yymsp[-2].minor.yy495.value = yymsp[0].minor.yy32; yymsp[-2].minor.yy495.mask = 0x0000ff; } + break; + case 50: /* refarg ::= ON UPDATE refact */ +{ yymsp[-2].minor.yy495.value = yymsp[0].minor.yy32<<8; yymsp[-2].minor.yy495.mask = 0x00ff00; } + break; + case 51: /* refact ::= SET NULL */ +{ yymsp[-1].minor.yy32 = OE_SetNull; /* EV: R-33326-45252 */} + break; + case 52: /* refact ::= SET DEFAULT */ +{ yymsp[-1].minor.yy32 = OE_SetDflt; /* EV: R-33326-45252 */} + break; + case 53: /* refact ::= CASCADE */ +{ yymsp[0].minor.yy32 = OE_Cascade; /* EV: R-33326-45252 */} + break; + case 54: /* refact ::= RESTRICT */ +{ yymsp[0].minor.yy32 = OE_Restrict; /* EV: R-33326-45252 */} + break; + case 55: /* refact ::= NO ACTION */ +{ yymsp[-1].minor.yy32 = OE_None; /* EV: R-33326-45252 */} + break; + case 56: /* defer_subclause ::= NOT DEFERRABLE init_deferred_pred_opt */ +{yymsp[-2].minor.yy32 = 0;} + break; + case 57: /* defer_subclause ::= DEFERRABLE init_deferred_pred_opt */ + case 72: /* orconf ::= OR resolvetype */ yytestcase(yyruleno==72); + case 160: /* insert_cmd ::= INSERT orconf */ yytestcase(yyruleno==160); +{yymsp[-1].minor.yy32 = yymsp[0].minor.yy32;} + break; + case 59: /* init_deferred_pred_opt ::= INITIALLY DEFERRED */ + case 76: /* ifexists ::= IF EXISTS */ yytestcase(yyruleno==76); + case 202: /* between_op ::= NOT BETWEEN */ yytestcase(yyruleno==202); + case 205: /* in_op ::= NOT IN */ yytestcase(yyruleno==205); + case 231: /* collate ::= COLLATE ID|STRING */ yytestcase(yyruleno==231); +{yymsp[-1].minor.yy32 = 1;} + break; + case 60: /* init_deferred_pred_opt ::= INITIALLY IMMEDIATE */ +{yymsp[-1].minor.yy32 = 0;} + break; + case 62: /* tconscomma ::= COMMA */ +{pParse->constraintName.n = 0;} + break; + case 64: /* tcons ::= PRIMARY KEY LP sortlist autoinc RP onconf */ +{sqlite3AddPrimaryKey(pParse,yymsp[-3].minor.yy138,yymsp[0].minor.yy32,yymsp[-2].minor.yy32,0);} + break; + case 65: /* tcons ::= UNIQUE LP sortlist RP onconf */ +{sqlite3CreateIndex(pParse,0,0,0,yymsp[-2].minor.yy138,yymsp[0].minor.yy32,0,0,0,0, + SQLITE_IDXTYPE_UNIQUE);} + break; + case 66: /* tcons ::= CHECK LP expr RP onconf */ +{sqlite3AddCheckConstraint(pParse,yymsp[-2].minor.yy46);} + break; + case 67: /* tcons ::= FOREIGN KEY LP eidlist RP REFERENCES nm eidlist_opt refargs defer_subclause_opt */ +{ + sqlite3CreateForeignKey(pParse, yymsp[-6].minor.yy138, &yymsp[-3].minor.yy0, yymsp[-2].minor.yy138, yymsp[-1].minor.yy32); + sqlite3DeferForeignKey(pParse, yymsp[0].minor.yy32); +} + break; + case 69: /* onconf ::= */ + case 71: /* orconf ::= */ yytestcase(yyruleno==71); +{yymsp[1].minor.yy32 = OE_Default;} + break; + case 70: /* onconf ::= ON CONFLICT resolvetype */ +{yymsp[-2].minor.yy32 = yymsp[0].minor.yy32;} + break; + case 73: /* resolvetype ::= IGNORE */ +{yymsp[0].minor.yy32 = OE_Ignore;} + break; + case 74: /* resolvetype ::= REPLACE */ + case 161: /* insert_cmd ::= REPLACE */ yytestcase(yyruleno==161); +{yymsp[0].minor.yy32 = OE_Replace;} + break; + case 75: /* cmd ::= DROP TABLE ifexists fullname */ +{ + sqlite3DropTable(pParse, yymsp[0].minor.yy609, 0, yymsp[-1].minor.yy32); +} + break; + case 78: /* cmd ::= createkw temp VIEW ifnotexists nm dbnm eidlist_opt AS select */ +{ + sqlite3CreateView(pParse, &yymsp[-8].minor.yy0, &yymsp[-4].minor.yy0, &yymsp[-3].minor.yy0, yymsp[-2].minor.yy138, yymsp[0].minor.yy25, yymsp[-7].minor.yy32, yymsp[-5].minor.yy32); +} + break; + case 79: /* cmd ::= DROP VIEW ifexists fullname */ +{ + sqlite3DropTable(pParse, yymsp[0].minor.yy609, 1, yymsp[-1].minor.yy32); +} + break; + case 80: /* cmd ::= select */ +{ + SelectDest dest = {SRT_Output, 0, 0, 0, 0, 0}; + sqlite3Select(pParse, yymsp[0].minor.yy25, &dest); + sqlite3SelectDelete(pParse->db, yymsp[0].minor.yy25); +} + break; + case 81: /* select ::= WITH wqlist selectnowith */ +{ + Select *p = yymsp[0].minor.yy25; + if( p ){ + p->pWith = yymsp[-1].minor.yy297; + parserDoubleLinkSelect(pParse, p); + }else{ + sqlite3WithDelete(pParse->db, yymsp[-1].minor.yy297); + } + yymsp[-2].minor.yy25 = p; +} + break; + case 82: /* select ::= WITH RECURSIVE wqlist selectnowith */ +{ + Select *p = yymsp[0].minor.yy25; + if( p ){ + p->pWith = yymsp[-1].minor.yy297; + parserDoubleLinkSelect(pParse, p); + }else{ + sqlite3WithDelete(pParse->db, yymsp[-1].minor.yy297); + } + yymsp[-3].minor.yy25 = p; +} + break; + case 83: /* select ::= selectnowith */ +{ + Select *p = yymsp[0].minor.yy25; + if( p ){ + parserDoubleLinkSelect(pParse, p); + } + yymsp[0].minor.yy25 = p; /*A-overwrites-X*/ +} + break; + case 84: /* selectnowith ::= selectnowith multiselect_op oneselect */ +{ + Select *pRhs = yymsp[0].minor.yy25; + Select *pLhs = yymsp[-2].minor.yy25; + if( pRhs && pRhs->pPrior ){ + SrcList *pFrom; + Token x; + x.n = 0; + parserDoubleLinkSelect(pParse, pRhs); + pFrom = sqlite3SrcListAppendFromTerm(pParse,0,0,0,&x,pRhs,0,0); + pRhs = sqlite3SelectNew(pParse,0,pFrom,0,0,0,0,0,0); + } + if( pRhs ){ + pRhs->op = (u8)yymsp[-1].minor.yy32; + pRhs->pPrior = pLhs; + if( ALWAYS(pLhs) ) pLhs->selFlags &= ~SF_MultiValue; + pRhs->selFlags &= ~SF_MultiValue; + if( yymsp[-1].minor.yy32!=TK_ALL ) pParse->hasCompound = 1; + }else{ + sqlite3SelectDelete(pParse->db, pLhs); + } + yymsp[-2].minor.yy25 = pRhs; +} + break; + case 85: /* multiselect_op ::= UNION */ + case 87: /* multiselect_op ::= EXCEPT|INTERSECT */ yytestcase(yyruleno==87); +{yymsp[0].minor.yy32 = yymsp[0].major; /*A-overwrites-OP*/} + break; + case 86: /* multiselect_op ::= UNION ALL */ +{yymsp[-1].minor.yy32 = TK_ALL;} + break; + case 88: /* oneselect ::= SELECT distinct selcollist from where_opt groupby_opt having_opt orderby_opt limit_opt */ +{ + yymsp[-8].minor.yy25 = sqlite3SelectNew(pParse,yymsp[-6].minor.yy138,yymsp[-5].minor.yy609,yymsp[-4].minor.yy46,yymsp[-3].minor.yy138,yymsp[-2].minor.yy46,yymsp[-1].minor.yy138,yymsp[-7].minor.yy32,yymsp[0].minor.yy46); +} + break; + case 89: /* oneselect ::= SELECT distinct selcollist from where_opt groupby_opt having_opt window_clause orderby_opt limit_opt */ +{ + yymsp[-9].minor.yy25 = sqlite3SelectNew(pParse,yymsp[-7].minor.yy138,yymsp[-6].minor.yy609,yymsp[-5].minor.yy46,yymsp[-4].minor.yy138,yymsp[-3].minor.yy46,yymsp[-1].minor.yy138,yymsp[-8].minor.yy32,yymsp[0].minor.yy46); + if( yymsp[-9].minor.yy25 ){ + yymsp[-9].minor.yy25->pWinDefn = yymsp[-2].minor.yy455; + }else{ + sqlite3WindowListDelete(pParse->db, yymsp[-2].minor.yy455); + } +} + break; + case 90: /* values ::= VALUES LP nexprlist RP */ +{ + yymsp[-3].minor.yy25 = sqlite3SelectNew(pParse,yymsp[-1].minor.yy138,0,0,0,0,0,SF_Values,0); +} + break; + case 91: /* values ::= values COMMA LP nexprlist RP */ +{ + Select *pRight, *pLeft = yymsp[-4].minor.yy25; + pRight = sqlite3SelectNew(pParse,yymsp[-1].minor.yy138,0,0,0,0,0,SF_Values|SF_MultiValue,0); + if( ALWAYS(pLeft) ) pLeft->selFlags &= ~SF_MultiValue; + if( pRight ){ + pRight->op = TK_ALL; + pRight->pPrior = pLeft; + yymsp[-4].minor.yy25 = pRight; + }else{ + yymsp[-4].minor.yy25 = pLeft; + } +} + break; + case 92: /* distinct ::= DISTINCT */ +{yymsp[0].minor.yy32 = SF_Distinct;} + break; + case 93: /* distinct ::= ALL */ +{yymsp[0].minor.yy32 = SF_All;} + break; + case 95: /* sclp ::= */ + case 128: /* orderby_opt ::= */ yytestcase(yyruleno==128); + case 138: /* groupby_opt ::= */ yytestcase(yyruleno==138); + case 218: /* exprlist ::= */ yytestcase(yyruleno==218); + case 221: /* paren_exprlist ::= */ yytestcase(yyruleno==221); + case 226: /* eidlist_opt ::= */ yytestcase(yyruleno==226); +{yymsp[1].minor.yy138 = 0;} + break; + case 96: /* selcollist ::= sclp scanpt expr scanpt as */ +{ + yymsp[-4].minor.yy138 = sqlite3ExprListAppend(pParse, yymsp[-4].minor.yy138, yymsp[-2].minor.yy46); + if( yymsp[0].minor.yy0.n>0 ) sqlite3ExprListSetName(pParse, yymsp[-4].minor.yy138, &yymsp[0].minor.yy0, 1); + sqlite3ExprListSetSpan(pParse,yymsp[-4].minor.yy138,yymsp[-3].minor.yy8,yymsp[-1].minor.yy8); +} + break; + case 97: /* selcollist ::= sclp scanpt STAR */ +{ + Expr *p = sqlite3Expr(pParse->db, TK_ASTERISK, 0); + yymsp[-2].minor.yy138 = sqlite3ExprListAppend(pParse, yymsp[-2].minor.yy138, p); +} + break; + case 98: /* selcollist ::= sclp scanpt nm DOT STAR */ +{ + Expr *pRight = sqlite3PExpr(pParse, TK_ASTERISK, 0, 0); + Expr *pLeft = sqlite3ExprAlloc(pParse->db, TK_ID, &yymsp[-2].minor.yy0, 1); + Expr *pDot = sqlite3PExpr(pParse, TK_DOT, pLeft, pRight); + yymsp[-4].minor.yy138 = sqlite3ExprListAppend(pParse,yymsp[-4].minor.yy138, pDot); +} + break; + case 99: /* as ::= AS nm */ + case 110: /* dbnm ::= DOT nm */ yytestcase(yyruleno==110); + case 242: /* plus_num ::= PLUS INTEGER|FLOAT */ yytestcase(yyruleno==242); + case 243: /* minus_num ::= MINUS INTEGER|FLOAT */ yytestcase(yyruleno==243); +{yymsp[-1].minor.yy0 = yymsp[0].minor.yy0;} + break; + case 101: /* from ::= */ +{yymsp[1].minor.yy609 = sqlite3DbMallocZero(pParse->db, sizeof(*yymsp[1].minor.yy609));} + break; + case 102: /* from ::= FROM seltablist */ +{ + yymsp[-1].minor.yy609 = yymsp[0].minor.yy609; + sqlite3SrcListShiftJoinType(yymsp[-1].minor.yy609); +} + break; + case 103: /* stl_prefix ::= seltablist joinop */ +{ + if( ALWAYS(yymsp[-1].minor.yy609 && yymsp[-1].minor.yy609->nSrc>0) ) yymsp[-1].minor.yy609->a[yymsp[-1].minor.yy609->nSrc-1].fg.jointype = (u8)yymsp[0].minor.yy32; +} + break; + case 104: /* stl_prefix ::= */ +{yymsp[1].minor.yy609 = 0;} + break; + case 105: /* seltablist ::= stl_prefix nm dbnm as indexed_opt on_opt using_opt */ +{ + yymsp[-6].minor.yy609 = sqlite3SrcListAppendFromTerm(pParse,yymsp[-6].minor.yy609,&yymsp[-5].minor.yy0,&yymsp[-4].minor.yy0,&yymsp[-3].minor.yy0,0,yymsp[-1].minor.yy46,yymsp[0].minor.yy406); + sqlite3SrcListIndexedBy(pParse, yymsp[-6].minor.yy609, &yymsp[-2].minor.yy0); +} + break; + case 106: /* seltablist ::= stl_prefix nm dbnm LP exprlist RP as on_opt using_opt */ +{ + yymsp[-8].minor.yy609 = sqlite3SrcListAppendFromTerm(pParse,yymsp[-8].minor.yy609,&yymsp[-7].minor.yy0,&yymsp[-6].minor.yy0,&yymsp[-2].minor.yy0,0,yymsp[-1].minor.yy46,yymsp[0].minor.yy406); + sqlite3SrcListFuncArgs(pParse, yymsp[-8].minor.yy609, yymsp[-4].minor.yy138); +} + break; + case 107: /* seltablist ::= stl_prefix LP select RP as on_opt using_opt */ +{ + yymsp[-6].minor.yy609 = sqlite3SrcListAppendFromTerm(pParse,yymsp[-6].minor.yy609,0,0,&yymsp[-2].minor.yy0,yymsp[-4].minor.yy25,yymsp[-1].minor.yy46,yymsp[0].minor.yy406); + } + break; + case 108: /* seltablist ::= stl_prefix LP seltablist RP as on_opt using_opt */ +{ + if( yymsp[-6].minor.yy609==0 && yymsp[-2].minor.yy0.n==0 && yymsp[-1].minor.yy46==0 && yymsp[0].minor.yy406==0 ){ + yymsp[-6].minor.yy609 = yymsp[-4].minor.yy609; + }else if( yymsp[-4].minor.yy609->nSrc==1 ){ + yymsp[-6].minor.yy609 = sqlite3SrcListAppendFromTerm(pParse,yymsp[-6].minor.yy609,0,0,&yymsp[-2].minor.yy0,0,yymsp[-1].minor.yy46,yymsp[0].minor.yy406); + if( yymsp[-6].minor.yy609 ){ + struct SrcList_item *pNew = &yymsp[-6].minor.yy609->a[yymsp[-6].minor.yy609->nSrc-1]; + struct SrcList_item *pOld = yymsp[-4].minor.yy609->a; + pNew->zName = pOld->zName; + pNew->zDatabase = pOld->zDatabase; + pNew->pSelect = pOld->pSelect; + if( pOld->fg.isTabFunc ){ + pNew->u1.pFuncArg = pOld->u1.pFuncArg; + pOld->u1.pFuncArg = 0; + pOld->fg.isTabFunc = 0; + pNew->fg.isTabFunc = 1; + } + pOld->zName = pOld->zDatabase = 0; + pOld->pSelect = 0; + } + sqlite3SrcListDelete(pParse->db, yymsp[-4].minor.yy609); + }else{ + Select *pSubquery; + sqlite3SrcListShiftJoinType(yymsp[-4].minor.yy609); + pSubquery = sqlite3SelectNew(pParse,0,yymsp[-4].minor.yy609,0,0,0,0,SF_NestedFrom,0); + yymsp[-6].minor.yy609 = sqlite3SrcListAppendFromTerm(pParse,yymsp[-6].minor.yy609,0,0,&yymsp[-2].minor.yy0,pSubquery,yymsp[-1].minor.yy46,yymsp[0].minor.yy406); + } + } + break; + case 109: /* dbnm ::= */ + case 123: /* indexed_opt ::= */ yytestcase(yyruleno==123); +{yymsp[1].minor.yy0.z=0; yymsp[1].minor.yy0.n=0;} + break; + case 111: /* fullname ::= nm */ +{ + yylhsminor.yy609 = sqlite3SrcListAppend(pParse,0,&yymsp[0].minor.yy0,0); + if( IN_RENAME_OBJECT && yylhsminor.yy609 ) sqlite3RenameTokenMap(pParse, yylhsminor.yy609->a[0].zName, &yymsp[0].minor.yy0); +} + yymsp[0].minor.yy609 = yylhsminor.yy609; + break; + case 112: /* fullname ::= nm DOT nm */ +{ + yylhsminor.yy609 = sqlite3SrcListAppend(pParse,0,&yymsp[-2].minor.yy0,&yymsp[0].minor.yy0); + if( IN_RENAME_OBJECT && yylhsminor.yy609 ) sqlite3RenameTokenMap(pParse, yylhsminor.yy609->a[0].zName, &yymsp[0].minor.yy0); +} + yymsp[-2].minor.yy609 = yylhsminor.yy609; + break; + case 113: /* xfullname ::= nm */ +{yymsp[0].minor.yy609 = sqlite3SrcListAppend(pParse,0,&yymsp[0].minor.yy0,0); /*A-overwrites-X*/} + break; + case 114: /* xfullname ::= nm DOT nm */ +{yymsp[-2].minor.yy609 = sqlite3SrcListAppend(pParse,0,&yymsp[-2].minor.yy0,&yymsp[0].minor.yy0); /*A-overwrites-X*/} + break; + case 115: /* xfullname ::= nm DOT nm AS nm */ +{ + yymsp[-4].minor.yy609 = sqlite3SrcListAppend(pParse,0,&yymsp[-4].minor.yy0,&yymsp[-2].minor.yy0); /*A-overwrites-X*/ + if( yymsp[-4].minor.yy609 ) yymsp[-4].minor.yy609->a[0].zAlias = sqlite3NameFromToken(pParse->db, &yymsp[0].minor.yy0); +} + break; + case 116: /* xfullname ::= nm AS nm */ +{ + yymsp[-2].minor.yy609 = sqlite3SrcListAppend(pParse,0,&yymsp[-2].minor.yy0,0); /*A-overwrites-X*/ + if( yymsp[-2].minor.yy609 ) yymsp[-2].minor.yy609->a[0].zAlias = sqlite3NameFromToken(pParse->db, &yymsp[0].minor.yy0); +} + break; + case 117: /* joinop ::= COMMA|JOIN */ +{ yymsp[0].minor.yy32 = JT_INNER; } + break; + case 118: /* joinop ::= JOIN_KW JOIN */ +{yymsp[-1].minor.yy32 = sqlite3JoinType(pParse,&yymsp[-1].minor.yy0,0,0); /*X-overwrites-A*/} + break; + case 119: /* joinop ::= JOIN_KW nm JOIN */ +{yymsp[-2].minor.yy32 = sqlite3JoinType(pParse,&yymsp[-2].minor.yy0,&yymsp[-1].minor.yy0,0); /*X-overwrites-A*/} + break; + case 120: /* joinop ::= JOIN_KW nm nm JOIN */ +{yymsp[-3].minor.yy32 = sqlite3JoinType(pParse,&yymsp[-3].minor.yy0,&yymsp[-2].minor.yy0,&yymsp[-1].minor.yy0);/*X-overwrites-A*/} + break; + case 121: /* on_opt ::= ON expr */ + case 141: /* having_opt ::= HAVING expr */ yytestcase(yyruleno==141); + case 148: /* where_opt ::= WHERE expr */ yytestcase(yyruleno==148); + case 214: /* case_else ::= ELSE expr */ yytestcase(yyruleno==214); + case 235: /* vinto ::= INTO expr */ yytestcase(yyruleno==235); +{yymsp[-1].minor.yy46 = yymsp[0].minor.yy46;} + break; + case 122: /* on_opt ::= */ + case 140: /* having_opt ::= */ yytestcase(yyruleno==140); + case 142: /* limit_opt ::= */ yytestcase(yyruleno==142); + case 147: /* where_opt ::= */ yytestcase(yyruleno==147); + case 215: /* case_else ::= */ yytestcase(yyruleno==215); + case 217: /* case_operand ::= */ yytestcase(yyruleno==217); + case 236: /* vinto ::= */ yytestcase(yyruleno==236); +{yymsp[1].minor.yy46 = 0;} + break; + case 124: /* indexed_opt ::= INDEXED BY nm */ +{yymsp[-2].minor.yy0 = yymsp[0].minor.yy0;} + break; + case 125: /* indexed_opt ::= NOT INDEXED */ +{yymsp[-1].minor.yy0.z=0; yymsp[-1].minor.yy0.n=1;} + break; + case 126: /* using_opt ::= USING LP idlist RP */ +{yymsp[-3].minor.yy406 = yymsp[-1].minor.yy406;} + break; + case 127: /* using_opt ::= */ + case 162: /* idlist_opt ::= */ yytestcase(yyruleno==162); +{yymsp[1].minor.yy406 = 0;} + break; + case 129: /* orderby_opt ::= ORDER BY sortlist */ + case 139: /* groupby_opt ::= GROUP BY nexprlist */ yytestcase(yyruleno==139); +{yymsp[-2].minor.yy138 = yymsp[0].minor.yy138;} + break; + case 130: /* sortlist ::= sortlist COMMA expr sortorder nulls */ +{ + yymsp[-4].minor.yy138 = sqlite3ExprListAppend(pParse,yymsp[-4].minor.yy138,yymsp[-2].minor.yy46); + sqlite3ExprListSetSortOrder(yymsp[-4].minor.yy138,yymsp[-1].minor.yy32,yymsp[0].minor.yy32); +} + break; + case 131: /* sortlist ::= expr sortorder nulls */ +{ + yymsp[-2].minor.yy138 = sqlite3ExprListAppend(pParse,0,yymsp[-2].minor.yy46); /*A-overwrites-Y*/ + sqlite3ExprListSetSortOrder(yymsp[-2].minor.yy138,yymsp[-1].minor.yy32,yymsp[0].minor.yy32); +} + break; + case 132: /* sortorder ::= ASC */ +{yymsp[0].minor.yy32 = SQLITE_SO_ASC;} + break; + case 133: /* sortorder ::= DESC */ +{yymsp[0].minor.yy32 = SQLITE_SO_DESC;} + break; + case 134: /* sortorder ::= */ + case 137: /* nulls ::= */ yytestcase(yyruleno==137); +{yymsp[1].minor.yy32 = SQLITE_SO_UNDEFINED;} + break; + case 135: /* nulls ::= NULLS FIRST */ +{yymsp[-1].minor.yy32 = SQLITE_SO_ASC;} + break; + case 136: /* nulls ::= NULLS LAST */ +{yymsp[-1].minor.yy32 = SQLITE_SO_DESC;} + break; + case 143: /* limit_opt ::= LIMIT expr */ +{yymsp[-1].minor.yy46 = sqlite3PExpr(pParse,TK_LIMIT,yymsp[0].minor.yy46,0);} + break; + case 144: /* limit_opt ::= LIMIT expr OFFSET expr */ +{yymsp[-3].minor.yy46 = sqlite3PExpr(pParse,TK_LIMIT,yymsp[-2].minor.yy46,yymsp[0].minor.yy46);} + break; + case 145: /* limit_opt ::= LIMIT expr COMMA expr */ +{yymsp[-3].minor.yy46 = sqlite3PExpr(pParse,TK_LIMIT,yymsp[0].minor.yy46,yymsp[-2].minor.yy46);} + break; + case 146: /* cmd ::= with DELETE FROM xfullname indexed_opt where_opt */ +{ + sqlite3SrcListIndexedBy(pParse, yymsp[-2].minor.yy609, &yymsp[-1].minor.yy0); + sqlite3DeleteFrom(pParse,yymsp[-2].minor.yy609,yymsp[0].minor.yy46,0,0); +} + break; + case 149: /* cmd ::= with UPDATE orconf xfullname indexed_opt SET setlist where_opt */ +{ + sqlite3SrcListIndexedBy(pParse, yymsp[-4].minor.yy609, &yymsp[-3].minor.yy0); + sqlite3ExprListCheckLength(pParse,yymsp[-1].minor.yy138,"set list"); + sqlite3Update(pParse,yymsp[-4].minor.yy609,yymsp[-1].minor.yy138,yymsp[0].minor.yy46,yymsp[-5].minor.yy32,0,0,0); +} + break; + case 150: /* setlist ::= setlist COMMA nm EQ expr */ +{ + yymsp[-4].minor.yy138 = sqlite3ExprListAppend(pParse, yymsp[-4].minor.yy138, yymsp[0].minor.yy46); + sqlite3ExprListSetName(pParse, yymsp[-4].minor.yy138, &yymsp[-2].minor.yy0, 1); +} + break; + case 151: /* setlist ::= setlist COMMA LP idlist RP EQ expr */ +{ + yymsp[-6].minor.yy138 = sqlite3ExprListAppendVector(pParse, yymsp[-6].minor.yy138, yymsp[-3].minor.yy406, yymsp[0].minor.yy46); +} + break; + case 152: /* setlist ::= nm EQ expr */ +{ + yylhsminor.yy138 = sqlite3ExprListAppend(pParse, 0, yymsp[0].minor.yy46); + sqlite3ExprListSetName(pParse, yylhsminor.yy138, &yymsp[-2].minor.yy0, 1); +} + yymsp[-2].minor.yy138 = yylhsminor.yy138; + break; + case 153: /* setlist ::= LP idlist RP EQ expr */ +{ + yymsp[-4].minor.yy138 = sqlite3ExprListAppendVector(pParse, 0, yymsp[-3].minor.yy406, yymsp[0].minor.yy46); +} + break; + case 154: /* cmd ::= with insert_cmd INTO xfullname idlist_opt select upsert */ +{ + sqlite3Insert(pParse, yymsp[-3].minor.yy609, yymsp[-1].minor.yy25, yymsp[-2].minor.yy406, yymsp[-5].minor.yy32, yymsp[0].minor.yy288); +} + break; + case 155: /* cmd ::= with insert_cmd INTO xfullname idlist_opt DEFAULT VALUES */ +{ + sqlite3Insert(pParse, yymsp[-3].minor.yy609, 0, yymsp[-2].minor.yy406, yymsp[-5].minor.yy32, 0); +} + break; + case 156: /* upsert ::= */ +{ yymsp[1].minor.yy288 = 0; } + break; + case 157: /* upsert ::= ON CONFLICT LP sortlist RP where_opt DO UPDATE SET setlist where_opt */ +{ yymsp[-10].minor.yy288 = sqlite3UpsertNew(pParse->db,yymsp[-7].minor.yy138,yymsp[-5].minor.yy46,yymsp[-1].minor.yy138,yymsp[0].minor.yy46);} + break; + case 158: /* upsert ::= ON CONFLICT LP sortlist RP where_opt DO NOTHING */ +{ yymsp[-7].minor.yy288 = sqlite3UpsertNew(pParse->db,yymsp[-4].minor.yy138,yymsp[-2].minor.yy46,0,0); } + break; + case 159: /* upsert ::= ON CONFLICT DO NOTHING */ +{ yymsp[-3].minor.yy288 = sqlite3UpsertNew(pParse->db,0,0,0,0); } + break; + case 163: /* idlist_opt ::= LP idlist RP */ +{yymsp[-2].minor.yy406 = yymsp[-1].minor.yy406;} + break; + case 164: /* idlist ::= idlist COMMA nm */ +{yymsp[-2].minor.yy406 = sqlite3IdListAppend(pParse,yymsp[-2].minor.yy406,&yymsp[0].minor.yy0);} + break; + case 165: /* idlist ::= nm */ +{yymsp[0].minor.yy406 = sqlite3IdListAppend(pParse,0,&yymsp[0].minor.yy0); /*A-overwrites-Y*/} + break; + case 166: /* expr ::= LP expr RP */ +{yymsp[-2].minor.yy46 = yymsp[-1].minor.yy46;} + break; + case 167: /* expr ::= ID|INDEXED */ + case 168: /* expr ::= JOIN_KW */ yytestcase(yyruleno==168); +{yymsp[0].minor.yy46=tokenExpr(pParse,TK_ID,yymsp[0].minor.yy0); /*A-overwrites-X*/} + break; + case 169: /* expr ::= nm DOT nm */ +{ + Expr *temp1 = sqlite3ExprAlloc(pParse->db, TK_ID, &yymsp[-2].minor.yy0, 1); + Expr *temp2 = sqlite3ExprAlloc(pParse->db, TK_ID, &yymsp[0].minor.yy0, 1); + if( IN_RENAME_OBJECT ){ + sqlite3RenameTokenMap(pParse, (void*)temp2, &yymsp[0].minor.yy0); + sqlite3RenameTokenMap(pParse, (void*)temp1, &yymsp[-2].minor.yy0); + } + yylhsminor.yy46 = sqlite3PExpr(pParse, TK_DOT, temp1, temp2); +} + yymsp[-2].minor.yy46 = yylhsminor.yy46; + break; + case 170: /* expr ::= nm DOT nm DOT nm */ +{ + Expr *temp1 = sqlite3ExprAlloc(pParse->db, TK_ID, &yymsp[-4].minor.yy0, 1); + Expr *temp2 = sqlite3ExprAlloc(pParse->db, TK_ID, &yymsp[-2].minor.yy0, 1); + Expr *temp3 = sqlite3ExprAlloc(pParse->db, TK_ID, &yymsp[0].minor.yy0, 1); + Expr *temp4 = sqlite3PExpr(pParse, TK_DOT, temp2, temp3); + if( IN_RENAME_OBJECT ){ + sqlite3RenameTokenMap(pParse, (void*)temp3, &yymsp[0].minor.yy0); + sqlite3RenameTokenMap(pParse, (void*)temp2, &yymsp[-2].minor.yy0); + } + yylhsminor.yy46 = sqlite3PExpr(pParse, TK_DOT, temp1, temp4); +} + yymsp[-4].minor.yy46 = yylhsminor.yy46; + break; + case 171: /* term ::= NULL|FLOAT|BLOB */ + case 172: /* term ::= STRING */ yytestcase(yyruleno==172); +{yymsp[0].minor.yy46=tokenExpr(pParse,yymsp[0].major,yymsp[0].minor.yy0); /*A-overwrites-X*/} + break; + case 173: /* term ::= INTEGER */ +{ + yylhsminor.yy46 = sqlite3ExprAlloc(pParse->db, TK_INTEGER, &yymsp[0].minor.yy0, 1); +} + yymsp[0].minor.yy46 = yylhsminor.yy46; + break; + case 174: /* expr ::= VARIABLE */ +{ + if( !(yymsp[0].minor.yy0.z[0]=='#' && sqlite3Isdigit(yymsp[0].minor.yy0.z[1])) ){ + u32 n = yymsp[0].minor.yy0.n; + yymsp[0].minor.yy46 = tokenExpr(pParse, TK_VARIABLE, yymsp[0].minor.yy0); + sqlite3ExprAssignVarNumber(pParse, yymsp[0].minor.yy46, n); + }else{ + /* When doing a nested parse, one can include terms in an expression + ** that look like this: #1 #2 ... These terms refer to registers + ** in the virtual machine. #N is the N-th register. */ + Token t = yymsp[0].minor.yy0; /*A-overwrites-X*/ + assert( t.n>=2 ); + if( pParse->nested==0 ){ + sqlite3ErrorMsg(pParse, "near \"%T\": syntax error", &t); + yymsp[0].minor.yy46 = 0; + }else{ + yymsp[0].minor.yy46 = sqlite3PExpr(pParse, TK_REGISTER, 0, 0); + if( yymsp[0].minor.yy46 ) sqlite3GetInt32(&t.z[1], &yymsp[0].minor.yy46->iTable); + } + } +} + break; + case 175: /* expr ::= expr COLLATE ID|STRING */ +{ + yymsp[-2].minor.yy46 = sqlite3ExprAddCollateToken(pParse, yymsp[-2].minor.yy46, &yymsp[0].minor.yy0, 1); +} + break; + case 176: /* expr ::= CAST LP expr AS typetoken RP */ +{ + yymsp[-5].minor.yy46 = sqlite3ExprAlloc(pParse->db, TK_CAST, &yymsp[-1].minor.yy0, 1); + sqlite3ExprAttachSubtrees(pParse->db, yymsp[-5].minor.yy46, yymsp[-3].minor.yy46, 0); +} + break; + case 177: /* expr ::= ID|INDEXED LP distinct exprlist RP */ +{ + yylhsminor.yy46 = sqlite3ExprFunction(pParse, yymsp[-1].minor.yy138, &yymsp[-4].minor.yy0, yymsp[-2].minor.yy32); +} + yymsp[-4].minor.yy46 = yylhsminor.yy46; + break; + case 178: /* expr ::= ID|INDEXED LP STAR RP */ +{ + yylhsminor.yy46 = sqlite3ExprFunction(pParse, 0, &yymsp[-3].minor.yy0, 0); +} + yymsp[-3].minor.yy46 = yylhsminor.yy46; + break; + case 179: /* expr ::= ID|INDEXED LP distinct exprlist RP filter_over */ +{ + yylhsminor.yy46 = sqlite3ExprFunction(pParse, yymsp[-2].minor.yy138, &yymsp[-5].minor.yy0, yymsp[-3].minor.yy32); + sqlite3WindowAttach(pParse, yylhsminor.yy46, yymsp[0].minor.yy455); +} + yymsp[-5].minor.yy46 = yylhsminor.yy46; + break; + case 180: /* expr ::= ID|INDEXED LP STAR RP filter_over */ +{ + yylhsminor.yy46 = sqlite3ExprFunction(pParse, 0, &yymsp[-4].minor.yy0, 0); + sqlite3WindowAttach(pParse, yylhsminor.yy46, yymsp[0].minor.yy455); +} + yymsp[-4].minor.yy46 = yylhsminor.yy46; + break; + case 181: /* term ::= CTIME_KW */ +{ + yylhsminor.yy46 = sqlite3ExprFunction(pParse, 0, &yymsp[0].minor.yy0, 0); +} + yymsp[0].minor.yy46 = yylhsminor.yy46; + break; + case 182: /* expr ::= LP nexprlist COMMA expr RP */ +{ + ExprList *pList = sqlite3ExprListAppend(pParse, yymsp[-3].minor.yy138, yymsp[-1].minor.yy46); + yymsp[-4].minor.yy46 = sqlite3PExpr(pParse, TK_VECTOR, 0, 0); + if( yymsp[-4].minor.yy46 ){ + yymsp[-4].minor.yy46->x.pList = pList; + }else{ + sqlite3ExprListDelete(pParse->db, pList); + } +} + break; + case 183: /* expr ::= expr AND expr */ +{yymsp[-2].minor.yy46=sqlite3ExprAnd(pParse,yymsp[-2].minor.yy46,yymsp[0].minor.yy46);} + break; + case 184: /* expr ::= expr OR expr */ + case 185: /* expr ::= expr LT|GT|GE|LE expr */ yytestcase(yyruleno==185); + case 186: /* expr ::= expr EQ|NE expr */ yytestcase(yyruleno==186); + case 187: /* expr ::= expr BITAND|BITOR|LSHIFT|RSHIFT expr */ yytestcase(yyruleno==187); + case 188: /* expr ::= expr PLUS|MINUS expr */ yytestcase(yyruleno==188); + case 189: /* expr ::= expr STAR|SLASH|REM expr */ yytestcase(yyruleno==189); + case 190: /* expr ::= expr CONCAT expr */ yytestcase(yyruleno==190); +{yymsp[-2].minor.yy46=sqlite3PExpr(pParse,yymsp[-1].major,yymsp[-2].minor.yy46,yymsp[0].minor.yy46);} + break; + case 191: /* likeop ::= NOT LIKE_KW|MATCH */ +{yymsp[-1].minor.yy0=yymsp[0].minor.yy0; yymsp[-1].minor.yy0.n|=0x80000000; /*yymsp[-1].minor.yy0-overwrite-yymsp[0].minor.yy0*/} + break; + case 192: /* expr ::= expr likeop expr */ +{ + ExprList *pList; + int bNot = yymsp[-1].minor.yy0.n & 0x80000000; + yymsp[-1].minor.yy0.n &= 0x7fffffff; + pList = sqlite3ExprListAppend(pParse,0, yymsp[0].minor.yy46); + pList = sqlite3ExprListAppend(pParse,pList, yymsp[-2].minor.yy46); + yymsp[-2].minor.yy46 = sqlite3ExprFunction(pParse, pList, &yymsp[-1].minor.yy0, 0); + if( bNot ) yymsp[-2].minor.yy46 = sqlite3PExpr(pParse, TK_NOT, yymsp[-2].minor.yy46, 0); + if( yymsp[-2].minor.yy46 ) yymsp[-2].minor.yy46->flags |= EP_InfixFunc; +} + break; + case 193: /* expr ::= expr likeop expr ESCAPE expr */ +{ + ExprList *pList; + int bNot = yymsp[-3].minor.yy0.n & 0x80000000; + yymsp[-3].minor.yy0.n &= 0x7fffffff; + pList = sqlite3ExprListAppend(pParse,0, yymsp[-2].minor.yy46); + pList = sqlite3ExprListAppend(pParse,pList, yymsp[-4].minor.yy46); + pList = sqlite3ExprListAppend(pParse,pList, yymsp[0].minor.yy46); + yymsp[-4].minor.yy46 = sqlite3ExprFunction(pParse, pList, &yymsp[-3].minor.yy0, 0); + if( bNot ) yymsp[-4].minor.yy46 = sqlite3PExpr(pParse, TK_NOT, yymsp[-4].minor.yy46, 0); + if( yymsp[-4].minor.yy46 ) yymsp[-4].minor.yy46->flags |= EP_InfixFunc; +} + break; + case 194: /* expr ::= expr ISNULL|NOTNULL */ +{yymsp[-1].minor.yy46 = sqlite3PExpr(pParse,yymsp[0].major,yymsp[-1].minor.yy46,0);} + break; + case 195: /* expr ::= expr NOT NULL */ +{yymsp[-2].minor.yy46 = sqlite3PExpr(pParse,TK_NOTNULL,yymsp[-2].minor.yy46,0);} + break; + case 196: /* expr ::= expr IS expr */ +{ + yymsp[-2].minor.yy46 = sqlite3PExpr(pParse,TK_IS,yymsp[-2].minor.yy46,yymsp[0].minor.yy46); + binaryToUnaryIfNull(pParse, yymsp[0].minor.yy46, yymsp[-2].minor.yy46, TK_ISNULL); +} + break; + case 197: /* expr ::= expr IS NOT expr */ +{ + yymsp[-3].minor.yy46 = sqlite3PExpr(pParse,TK_ISNOT,yymsp[-3].minor.yy46,yymsp[0].minor.yy46); + binaryToUnaryIfNull(pParse, yymsp[0].minor.yy46, yymsp[-3].minor.yy46, TK_NOTNULL); +} + break; + case 198: /* expr ::= NOT expr */ + case 199: /* expr ::= BITNOT expr */ yytestcase(yyruleno==199); +{yymsp[-1].minor.yy46 = sqlite3PExpr(pParse, yymsp[-1].major, yymsp[0].minor.yy46, 0);/*A-overwrites-B*/} + break; + case 200: /* expr ::= PLUS|MINUS expr */ +{ + yymsp[-1].minor.yy46 = sqlite3PExpr(pParse, yymsp[-1].major==TK_PLUS ? TK_UPLUS : TK_UMINUS, yymsp[0].minor.yy46, 0); + /*A-overwrites-B*/ +} + break; + case 201: /* between_op ::= BETWEEN */ + case 204: /* in_op ::= IN */ yytestcase(yyruleno==204); +{yymsp[0].minor.yy32 = 0;} + break; + case 203: /* expr ::= expr between_op expr AND expr */ +{ + ExprList *pList = sqlite3ExprListAppend(pParse,0, yymsp[-2].minor.yy46); + pList = sqlite3ExprListAppend(pParse,pList, yymsp[0].minor.yy46); + yymsp[-4].minor.yy46 = sqlite3PExpr(pParse, TK_BETWEEN, yymsp[-4].minor.yy46, 0); + if( yymsp[-4].minor.yy46 ){ + yymsp[-4].minor.yy46->x.pList = pList; + }else{ + sqlite3ExprListDelete(pParse->db, pList); + } + if( yymsp[-3].minor.yy32 ) yymsp[-4].minor.yy46 = sqlite3PExpr(pParse, TK_NOT, yymsp[-4].minor.yy46, 0); +} + break; + case 206: /* expr ::= expr in_op LP exprlist RP */ +{ + if( yymsp[-1].minor.yy138==0 ){ + /* Expressions of the form + ** + ** expr1 IN () + ** expr1 NOT IN () + ** + ** simplify to constants 0 (false) and 1 (true), respectively, + ** regardless of the value of expr1. + */ + sqlite3ExprUnmapAndDelete(pParse, yymsp[-4].minor.yy46); + yymsp[-4].minor.yy46 = sqlite3Expr(pParse->db, TK_INTEGER, yymsp[-3].minor.yy32 ? "1" : "0"); + }else{ + yymsp[-4].minor.yy46 = sqlite3PExpr(pParse, TK_IN, yymsp[-4].minor.yy46, 0); + if( yymsp[-4].minor.yy46 ){ + yymsp[-4].minor.yy46->x.pList = yymsp[-1].minor.yy138; + sqlite3ExprSetHeightAndFlags(pParse, yymsp[-4].minor.yy46); + }else{ + sqlite3ExprListDelete(pParse->db, yymsp[-1].minor.yy138); + } + if( yymsp[-3].minor.yy32 ) yymsp[-4].minor.yy46 = sqlite3PExpr(pParse, TK_NOT, yymsp[-4].minor.yy46, 0); + } + } + break; + case 207: /* expr ::= LP select RP */ +{ + yymsp[-2].minor.yy46 = sqlite3PExpr(pParse, TK_SELECT, 0, 0); + sqlite3PExprAddSelect(pParse, yymsp[-2].minor.yy46, yymsp[-1].minor.yy25); + } + break; + case 208: /* expr ::= expr in_op LP select RP */ +{ + yymsp[-4].minor.yy46 = sqlite3PExpr(pParse, TK_IN, yymsp[-4].minor.yy46, 0); + sqlite3PExprAddSelect(pParse, yymsp[-4].minor.yy46, yymsp[-1].minor.yy25); + if( yymsp[-3].minor.yy32 ) yymsp[-4].minor.yy46 = sqlite3PExpr(pParse, TK_NOT, yymsp[-4].minor.yy46, 0); + } + break; + case 209: /* expr ::= expr in_op nm dbnm paren_exprlist */ +{ + SrcList *pSrc = sqlite3SrcListAppend(pParse, 0,&yymsp[-2].minor.yy0,&yymsp[-1].minor.yy0); + Select *pSelect = sqlite3SelectNew(pParse, 0,pSrc,0,0,0,0,0,0); + if( yymsp[0].minor.yy138 ) sqlite3SrcListFuncArgs(pParse, pSelect ? pSrc : 0, yymsp[0].minor.yy138); + yymsp[-4].minor.yy46 = sqlite3PExpr(pParse, TK_IN, yymsp[-4].minor.yy46, 0); + sqlite3PExprAddSelect(pParse, yymsp[-4].minor.yy46, pSelect); + if( yymsp[-3].minor.yy32 ) yymsp[-4].minor.yy46 = sqlite3PExpr(pParse, TK_NOT, yymsp[-4].minor.yy46, 0); + } + break; + case 210: /* expr ::= EXISTS LP select RP */ +{ + Expr *p; + p = yymsp[-3].minor.yy46 = sqlite3PExpr(pParse, TK_EXISTS, 0, 0); + sqlite3PExprAddSelect(pParse, p, yymsp[-1].minor.yy25); + } + break; + case 211: /* expr ::= CASE case_operand case_exprlist case_else END */ +{ + yymsp[-4].minor.yy46 = sqlite3PExpr(pParse, TK_CASE, yymsp[-3].minor.yy46, 0); + if( yymsp[-4].minor.yy46 ){ + yymsp[-4].minor.yy46->x.pList = yymsp[-1].minor.yy46 ? sqlite3ExprListAppend(pParse,yymsp[-2].minor.yy138,yymsp[-1].minor.yy46) : yymsp[-2].minor.yy138; + sqlite3ExprSetHeightAndFlags(pParse, yymsp[-4].minor.yy46); + }else{ + sqlite3ExprListDelete(pParse->db, yymsp[-2].minor.yy138); + sqlite3ExprDelete(pParse->db, yymsp[-1].minor.yy46); + } +} + break; + case 212: /* case_exprlist ::= case_exprlist WHEN expr THEN expr */ +{ + yymsp[-4].minor.yy138 = sqlite3ExprListAppend(pParse,yymsp[-4].minor.yy138, yymsp[-2].minor.yy46); + yymsp[-4].minor.yy138 = sqlite3ExprListAppend(pParse,yymsp[-4].minor.yy138, yymsp[0].minor.yy46); +} + break; + case 213: /* case_exprlist ::= WHEN expr THEN expr */ +{ + yymsp[-3].minor.yy138 = sqlite3ExprListAppend(pParse,0, yymsp[-2].minor.yy46); + yymsp[-3].minor.yy138 = sqlite3ExprListAppend(pParse,yymsp[-3].minor.yy138, yymsp[0].minor.yy46); +} + break; + case 216: /* case_operand ::= expr */ +{yymsp[0].minor.yy46 = yymsp[0].minor.yy46; /*A-overwrites-X*/} + break; + case 219: /* nexprlist ::= nexprlist COMMA expr */ +{yymsp[-2].minor.yy138 = sqlite3ExprListAppend(pParse,yymsp[-2].minor.yy138,yymsp[0].minor.yy46);} + break; + case 220: /* nexprlist ::= expr */ +{yymsp[0].minor.yy138 = sqlite3ExprListAppend(pParse,0,yymsp[0].minor.yy46); /*A-overwrites-Y*/} + break; + case 222: /* paren_exprlist ::= LP exprlist RP */ + case 227: /* eidlist_opt ::= LP eidlist RP */ yytestcase(yyruleno==227); +{yymsp[-2].minor.yy138 = yymsp[-1].minor.yy138;} + break; + case 223: /* cmd ::= createkw uniqueflag INDEX ifnotexists nm dbnm ON nm LP sortlist RP where_opt */ +{ + sqlite3CreateIndex(pParse, &yymsp[-7].minor.yy0, &yymsp[-6].minor.yy0, + sqlite3SrcListAppend(pParse,0,&yymsp[-4].minor.yy0,0), yymsp[-2].minor.yy138, yymsp[-10].minor.yy32, + &yymsp[-11].minor.yy0, yymsp[0].minor.yy46, SQLITE_SO_ASC, yymsp[-8].minor.yy32, SQLITE_IDXTYPE_APPDEF); + if( IN_RENAME_OBJECT && pParse->pNewIndex ){ + sqlite3RenameTokenMap(pParse, pParse->pNewIndex->zName, &yymsp[-4].minor.yy0); + } +} + break; + case 224: /* uniqueflag ::= UNIQUE */ + case 266: /* raisetype ::= ABORT */ yytestcase(yyruleno==266); +{yymsp[0].minor.yy32 = OE_Abort;} + break; + case 225: /* uniqueflag ::= */ +{yymsp[1].minor.yy32 = OE_None;} + break; + case 228: /* eidlist ::= eidlist COMMA nm collate sortorder */ +{ + yymsp[-4].minor.yy138 = parserAddExprIdListTerm(pParse, yymsp[-4].minor.yy138, &yymsp[-2].minor.yy0, yymsp[-1].minor.yy32, yymsp[0].minor.yy32); +} + break; + case 229: /* eidlist ::= nm collate sortorder */ +{ + yymsp[-2].minor.yy138 = parserAddExprIdListTerm(pParse, 0, &yymsp[-2].minor.yy0, yymsp[-1].minor.yy32, yymsp[0].minor.yy32); /*A-overwrites-Y*/ +} + break; + case 232: /* cmd ::= DROP INDEX ifexists fullname */ +{sqlite3DropIndex(pParse, yymsp[0].minor.yy609, yymsp[-1].minor.yy32);} + break; + case 233: /* cmd ::= VACUUM vinto */ +{sqlite3Vacuum(pParse,0,yymsp[0].minor.yy46);} + break; + case 234: /* cmd ::= VACUUM nm vinto */ +{sqlite3Vacuum(pParse,&yymsp[-1].minor.yy0,yymsp[0].minor.yy46);} + break; + case 237: /* cmd ::= PRAGMA nm dbnm */ +{sqlite3Pragma(pParse,&yymsp[-1].minor.yy0,&yymsp[0].minor.yy0,0,0);} + break; + case 238: /* cmd ::= PRAGMA nm dbnm EQ nmnum */ +{sqlite3Pragma(pParse,&yymsp[-3].minor.yy0,&yymsp[-2].minor.yy0,&yymsp[0].minor.yy0,0);} + break; + case 239: /* cmd ::= PRAGMA nm dbnm LP nmnum RP */ +{sqlite3Pragma(pParse,&yymsp[-4].minor.yy0,&yymsp[-3].minor.yy0,&yymsp[-1].minor.yy0,0);} + break; + case 240: /* cmd ::= PRAGMA nm dbnm EQ minus_num */ +{sqlite3Pragma(pParse,&yymsp[-3].minor.yy0,&yymsp[-2].minor.yy0,&yymsp[0].minor.yy0,1);} + break; + case 241: /* cmd ::= PRAGMA nm dbnm LP minus_num RP */ +{sqlite3Pragma(pParse,&yymsp[-4].minor.yy0,&yymsp[-3].minor.yy0,&yymsp[-1].minor.yy0,1);} + break; + case 244: /* cmd ::= createkw trigger_decl BEGIN trigger_cmd_list END */ +{ + Token all; + all.z = yymsp[-3].minor.yy0.z; + all.n = (int)(yymsp[0].minor.yy0.z - yymsp[-3].minor.yy0.z) + yymsp[0].minor.yy0.n; + sqlite3FinishTrigger(pParse, yymsp[-1].minor.yy527, &all); +} + break; + case 245: /* trigger_decl ::= temp TRIGGER ifnotexists nm dbnm trigger_time trigger_event ON fullname foreach_clause when_clause */ +{ + sqlite3BeginTrigger(pParse, &yymsp[-7].minor.yy0, &yymsp[-6].minor.yy0, yymsp[-5].minor.yy32, yymsp[-4].minor.yy572.a, yymsp[-4].minor.yy572.b, yymsp[-2].minor.yy609, yymsp[0].minor.yy46, yymsp[-10].minor.yy32, yymsp[-8].minor.yy32); + yymsp[-10].minor.yy0 = (yymsp[-6].minor.yy0.n==0?yymsp[-7].minor.yy0:yymsp[-6].minor.yy0); /*A-overwrites-T*/ +} + break; + case 246: /* trigger_time ::= BEFORE|AFTER */ +{ yymsp[0].minor.yy32 = yymsp[0].major; /*A-overwrites-X*/ } + break; + case 247: /* trigger_time ::= INSTEAD OF */ +{ yymsp[-1].minor.yy32 = TK_INSTEAD;} + break; + case 248: /* trigger_time ::= */ +{ yymsp[1].minor.yy32 = TK_BEFORE; } + break; + case 249: /* trigger_event ::= DELETE|INSERT */ + case 250: /* trigger_event ::= UPDATE */ yytestcase(yyruleno==250); +{yymsp[0].minor.yy572.a = yymsp[0].major; /*A-overwrites-X*/ yymsp[0].minor.yy572.b = 0;} + break; + case 251: /* trigger_event ::= UPDATE OF idlist */ +{yymsp[-2].minor.yy572.a = TK_UPDATE; yymsp[-2].minor.yy572.b = yymsp[0].minor.yy406;} + break; + case 252: /* when_clause ::= */ + case 271: /* key_opt ::= */ yytestcase(yyruleno==271); +{ yymsp[1].minor.yy46 = 0; } + break; + case 253: /* when_clause ::= WHEN expr */ + case 272: /* key_opt ::= KEY expr */ yytestcase(yyruleno==272); +{ yymsp[-1].minor.yy46 = yymsp[0].minor.yy46; } + break; + case 254: /* trigger_cmd_list ::= trigger_cmd_list trigger_cmd SEMI */ +{ + assert( yymsp[-2].minor.yy527!=0 ); + yymsp[-2].minor.yy527->pLast->pNext = yymsp[-1].minor.yy527; + yymsp[-2].minor.yy527->pLast = yymsp[-1].minor.yy527; +} + break; + case 255: /* trigger_cmd_list ::= trigger_cmd SEMI */ +{ + assert( yymsp[-1].minor.yy527!=0 ); + yymsp[-1].minor.yy527->pLast = yymsp[-1].minor.yy527; +} + break; + case 256: /* trnm ::= nm DOT nm */ +{ + yymsp[-2].minor.yy0 = yymsp[0].minor.yy0; + sqlite3ErrorMsg(pParse, + "qualified table names are not allowed on INSERT, UPDATE, and DELETE " + "statements within triggers"); +} + break; + case 257: /* tridxby ::= INDEXED BY nm */ +{ + sqlite3ErrorMsg(pParse, + "the INDEXED BY clause is not allowed on UPDATE or DELETE statements " + "within triggers"); +} + break; + case 258: /* tridxby ::= NOT INDEXED */ +{ + sqlite3ErrorMsg(pParse, + "the NOT INDEXED clause is not allowed on UPDATE or DELETE statements " + "within triggers"); +} + break; + case 259: /* trigger_cmd ::= UPDATE orconf trnm tridxby SET setlist where_opt scanpt */ +{yylhsminor.yy527 = sqlite3TriggerUpdateStep(pParse, &yymsp[-5].minor.yy0, yymsp[-2].minor.yy138, yymsp[-1].minor.yy46, yymsp[-6].minor.yy32, yymsp[-7].minor.yy0.z, yymsp[0].minor.yy8);} + yymsp[-7].minor.yy527 = yylhsminor.yy527; + break; + case 260: /* trigger_cmd ::= scanpt insert_cmd INTO trnm idlist_opt select upsert scanpt */ +{ + yylhsminor.yy527 = sqlite3TriggerInsertStep(pParse,&yymsp[-4].minor.yy0,yymsp[-3].minor.yy406,yymsp[-2].minor.yy25,yymsp[-6].minor.yy32,yymsp[-1].minor.yy288,yymsp[-7].minor.yy8,yymsp[0].minor.yy8);/*yylhsminor.yy527-overwrites-yymsp[-6].minor.yy32*/ +} + yymsp[-7].minor.yy527 = yylhsminor.yy527; + break; + case 261: /* trigger_cmd ::= DELETE FROM trnm tridxby where_opt scanpt */ +{yylhsminor.yy527 = sqlite3TriggerDeleteStep(pParse, &yymsp[-3].minor.yy0, yymsp[-1].minor.yy46, yymsp[-5].minor.yy0.z, yymsp[0].minor.yy8);} + yymsp[-5].minor.yy527 = yylhsminor.yy527; + break; + case 262: /* trigger_cmd ::= scanpt select scanpt */ +{yylhsminor.yy527 = sqlite3TriggerSelectStep(pParse->db, yymsp[-1].minor.yy25, yymsp[-2].minor.yy8, yymsp[0].minor.yy8); /*yylhsminor.yy527-overwrites-yymsp[-1].minor.yy25*/} + yymsp[-2].minor.yy527 = yylhsminor.yy527; + break; + case 263: /* expr ::= RAISE LP IGNORE RP */ +{ + yymsp[-3].minor.yy46 = sqlite3PExpr(pParse, TK_RAISE, 0, 0); + if( yymsp[-3].minor.yy46 ){ + yymsp[-3].minor.yy46->affExpr = OE_Ignore; + } +} + break; + case 264: /* expr ::= RAISE LP raisetype COMMA nm RP */ +{ + yymsp[-5].minor.yy46 = sqlite3ExprAlloc(pParse->db, TK_RAISE, &yymsp[-1].minor.yy0, 1); + if( yymsp[-5].minor.yy46 ) { + yymsp[-5].minor.yy46->affExpr = (char)yymsp[-3].minor.yy32; + } +} + break; + case 265: /* raisetype ::= ROLLBACK */ +{yymsp[0].minor.yy32 = OE_Rollback;} + break; + case 267: /* raisetype ::= FAIL */ +{yymsp[0].minor.yy32 = OE_Fail;} + break; + case 268: /* cmd ::= DROP TRIGGER ifexists fullname */ +{ + sqlite3DropTrigger(pParse,yymsp[0].minor.yy609,yymsp[-1].minor.yy32); +} + break; + case 269: /* cmd ::= ATTACH database_kw_opt expr AS expr key_opt */ +{ + sqlite3Attach(pParse, yymsp[-3].minor.yy46, yymsp[-1].minor.yy46, yymsp[0].minor.yy46); +} + break; + case 270: /* cmd ::= DETACH database_kw_opt expr */ +{ + sqlite3Detach(pParse, yymsp[0].minor.yy46); +} + break; + case 273: /* cmd ::= REINDEX */ +{sqlite3Reindex(pParse, 0, 0);} + break; + case 274: /* cmd ::= REINDEX nm dbnm */ +{sqlite3Reindex(pParse, &yymsp[-1].minor.yy0, &yymsp[0].minor.yy0);} + break; + case 275: /* cmd ::= ANALYZE */ +{sqlite3Analyze(pParse, 0, 0);} + break; + case 276: /* cmd ::= ANALYZE nm dbnm */ +{sqlite3Analyze(pParse, &yymsp[-1].minor.yy0, &yymsp[0].minor.yy0);} + break; + case 277: /* cmd ::= ALTER TABLE fullname RENAME TO nm */ +{ + sqlite3AlterRenameTable(pParse,yymsp[-3].minor.yy609,&yymsp[0].minor.yy0); +} + break; + case 278: /* cmd ::= ALTER TABLE add_column_fullname ADD kwcolumn_opt columnname carglist */ +{ + yymsp[-1].minor.yy0.n = (int)(pParse->sLastToken.z-yymsp[-1].minor.yy0.z) + pParse->sLastToken.n; + sqlite3AlterFinishAddColumn(pParse, &yymsp[-1].minor.yy0); +} + break; + case 279: /* add_column_fullname ::= fullname */ +{ + disableLookaside(pParse); + sqlite3AlterBeginAddColumn(pParse, yymsp[0].minor.yy609); +} + break; + case 280: /* cmd ::= ALTER TABLE fullname RENAME kwcolumn_opt nm TO nm */ +{ + sqlite3AlterRenameColumn(pParse, yymsp[-5].minor.yy609, &yymsp[-2].minor.yy0, &yymsp[0].minor.yy0); +} + break; + case 281: /* cmd ::= create_vtab */ +{sqlite3VtabFinishParse(pParse,0);} + break; + case 282: /* cmd ::= create_vtab LP vtabarglist RP */ +{sqlite3VtabFinishParse(pParse,&yymsp[0].minor.yy0);} + break; + case 283: /* create_vtab ::= createkw VIRTUAL TABLE ifnotexists nm dbnm USING nm */ +{ + sqlite3VtabBeginParse(pParse, &yymsp[-3].minor.yy0, &yymsp[-2].minor.yy0, &yymsp[0].minor.yy0, yymsp[-4].minor.yy32); +} + break; + case 284: /* vtabarg ::= */ +{sqlite3VtabArgInit(pParse);} + break; + case 285: /* vtabargtoken ::= ANY */ + case 286: /* vtabargtoken ::= lp anylist RP */ yytestcase(yyruleno==286); + case 287: /* lp ::= LP */ yytestcase(yyruleno==287); +{sqlite3VtabArgExtend(pParse,&yymsp[0].minor.yy0);} + break; + case 288: /* with ::= WITH wqlist */ + case 289: /* with ::= WITH RECURSIVE wqlist */ yytestcase(yyruleno==289); +{ sqlite3WithPush(pParse, yymsp[0].minor.yy297, 1); } + break; + case 290: /* wqlist ::= nm eidlist_opt AS LP select RP */ +{ + yymsp[-5].minor.yy297 = sqlite3WithAdd(pParse, 0, &yymsp[-5].minor.yy0, yymsp[-4].minor.yy138, yymsp[-1].minor.yy25); /*A-overwrites-X*/ +} + break; + case 291: /* wqlist ::= wqlist COMMA nm eidlist_opt AS LP select RP */ +{ + yymsp[-7].minor.yy297 = sqlite3WithAdd(pParse, yymsp[-7].minor.yy297, &yymsp[-5].minor.yy0, yymsp[-4].minor.yy138, yymsp[-1].minor.yy25); +} + break; + case 292: /* windowdefn_list ::= windowdefn */ +{ yylhsminor.yy455 = yymsp[0].minor.yy455; } + yymsp[0].minor.yy455 = yylhsminor.yy455; + break; + case 293: /* windowdefn_list ::= windowdefn_list COMMA windowdefn */ +{ + assert( yymsp[0].minor.yy455!=0 ); + sqlite3WindowChain(pParse, yymsp[0].minor.yy455, yymsp[-2].minor.yy455); + yymsp[0].minor.yy455->pNextWin = yymsp[-2].minor.yy455; + yylhsminor.yy455 = yymsp[0].minor.yy455; +} + yymsp[-2].minor.yy455 = yylhsminor.yy455; + break; + case 294: /* windowdefn ::= nm AS LP window RP */ +{ + if( ALWAYS(yymsp[-1].minor.yy455) ){ + yymsp[-1].minor.yy455->zName = sqlite3DbStrNDup(pParse->db, yymsp[-4].minor.yy0.z, yymsp[-4].minor.yy0.n); + } + yylhsminor.yy455 = yymsp[-1].minor.yy455; +} + yymsp[-4].minor.yy455 = yylhsminor.yy455; + break; + case 295: /* window ::= PARTITION BY nexprlist orderby_opt frame_opt */ +{ + yymsp[-4].minor.yy455 = sqlite3WindowAssemble(pParse, yymsp[0].minor.yy455, yymsp[-2].minor.yy138, yymsp[-1].minor.yy138, 0); +} + break; + case 296: /* window ::= nm PARTITION BY nexprlist orderby_opt frame_opt */ +{ + yylhsminor.yy455 = sqlite3WindowAssemble(pParse, yymsp[0].minor.yy455, yymsp[-2].minor.yy138, yymsp[-1].minor.yy138, &yymsp[-5].minor.yy0); +} + yymsp[-5].minor.yy455 = yylhsminor.yy455; + break; + case 297: /* window ::= ORDER BY sortlist frame_opt */ +{ + yymsp[-3].minor.yy455 = sqlite3WindowAssemble(pParse, yymsp[0].minor.yy455, 0, yymsp[-1].minor.yy138, 0); +} + break; + case 298: /* window ::= nm ORDER BY sortlist frame_opt */ +{ + yylhsminor.yy455 = sqlite3WindowAssemble(pParse, yymsp[0].minor.yy455, 0, yymsp[-1].minor.yy138, &yymsp[-4].minor.yy0); +} + yymsp[-4].minor.yy455 = yylhsminor.yy455; + break; + case 299: /* window ::= frame_opt */ + case 318: /* filter_over ::= over_clause */ yytestcase(yyruleno==318); +{ + yylhsminor.yy455 = yymsp[0].minor.yy455; +} + yymsp[0].minor.yy455 = yylhsminor.yy455; + break; + case 300: /* window ::= nm frame_opt */ +{ + yylhsminor.yy455 = sqlite3WindowAssemble(pParse, yymsp[0].minor.yy455, 0, 0, &yymsp[-1].minor.yy0); +} + yymsp[-1].minor.yy455 = yylhsminor.yy455; + break; + case 301: /* frame_opt ::= */ +{ + yymsp[1].minor.yy455 = sqlite3WindowAlloc(pParse, 0, TK_UNBOUNDED, 0, TK_CURRENT, 0, 0); +} + break; + case 302: /* frame_opt ::= range_or_rows frame_bound_s frame_exclude_opt */ +{ + yylhsminor.yy455 = sqlite3WindowAlloc(pParse, yymsp[-2].minor.yy32, yymsp[-1].minor.yy57.eType, yymsp[-1].minor.yy57.pExpr, TK_CURRENT, 0, yymsp[0].minor.yy118); +} + yymsp[-2].minor.yy455 = yylhsminor.yy455; + break; + case 303: /* frame_opt ::= range_or_rows BETWEEN frame_bound_s AND frame_bound_e frame_exclude_opt */ +{ + yylhsminor.yy455 = sqlite3WindowAlloc(pParse, yymsp[-5].minor.yy32, yymsp[-3].minor.yy57.eType, yymsp[-3].minor.yy57.pExpr, yymsp[-1].minor.yy57.eType, yymsp[-1].minor.yy57.pExpr, yymsp[0].minor.yy118); +} + yymsp[-5].minor.yy455 = yylhsminor.yy455; + break; + case 305: /* frame_bound_s ::= frame_bound */ + case 307: /* frame_bound_e ::= frame_bound */ yytestcase(yyruleno==307); +{yylhsminor.yy57 = yymsp[0].minor.yy57;} + yymsp[0].minor.yy57 = yylhsminor.yy57; + break; + case 306: /* frame_bound_s ::= UNBOUNDED PRECEDING */ + case 308: /* frame_bound_e ::= UNBOUNDED FOLLOWING */ yytestcase(yyruleno==308); + case 310: /* frame_bound ::= CURRENT ROW */ yytestcase(yyruleno==310); +{yylhsminor.yy57.eType = yymsp[-1].major; yylhsminor.yy57.pExpr = 0;} + yymsp[-1].minor.yy57 = yylhsminor.yy57; + break; + case 309: /* frame_bound ::= expr PRECEDING|FOLLOWING */ +{yylhsminor.yy57.eType = yymsp[0].major; yylhsminor.yy57.pExpr = yymsp[-1].minor.yy46;} + yymsp[-1].minor.yy57 = yylhsminor.yy57; + break; + case 311: /* frame_exclude_opt ::= */ +{yymsp[1].minor.yy118 = 0;} + break; + case 312: /* frame_exclude_opt ::= EXCLUDE frame_exclude */ +{yymsp[-1].minor.yy118 = yymsp[0].minor.yy118;} + break; + case 313: /* frame_exclude ::= NO OTHERS */ + case 314: /* frame_exclude ::= CURRENT ROW */ yytestcase(yyruleno==314); +{yymsp[-1].minor.yy118 = yymsp[-1].major; /*A-overwrites-X*/} + break; + case 315: /* frame_exclude ::= GROUP|TIES */ +{yymsp[0].minor.yy118 = yymsp[0].major; /*A-overwrites-X*/} + break; + case 316: /* window_clause ::= WINDOW windowdefn_list */ +{ yymsp[-1].minor.yy455 = yymsp[0].minor.yy455; } + break; + case 317: /* filter_over ::= filter_clause over_clause */ +{ + yymsp[0].minor.yy455->pFilter = yymsp[-1].minor.yy46; + yylhsminor.yy455 = yymsp[0].minor.yy455; +} + yymsp[-1].minor.yy455 = yylhsminor.yy455; + break; + case 319: /* filter_over ::= filter_clause */ +{ + yylhsminor.yy455 = (Window*)sqlite3DbMallocZero(pParse->db, sizeof(Window)); + if( yylhsminor.yy455 ){ + yylhsminor.yy455->eFrmType = TK_FILTER; + yylhsminor.yy455->pFilter = yymsp[0].minor.yy46; + }else{ + sqlite3ExprDelete(pParse->db, yymsp[0].minor.yy46); + } +} + yymsp[0].minor.yy455 = yylhsminor.yy455; + break; + case 320: /* over_clause ::= OVER LP window RP */ +{ + yymsp[-3].minor.yy455 = yymsp[-1].minor.yy455; + assert( yymsp[-3].minor.yy455!=0 ); +} + break; + case 321: /* over_clause ::= OVER nm */ +{ + yymsp[-1].minor.yy455 = (Window*)sqlite3DbMallocZero(pParse->db, sizeof(Window)); + if( yymsp[-1].minor.yy455 ){ + yymsp[-1].minor.yy455->zName = sqlite3DbStrNDup(pParse->db, yymsp[0].minor.yy0.z, yymsp[0].minor.yy0.n); + } +} + break; + case 322: /* filter_clause ::= FILTER LP WHERE expr RP */ +{ yymsp[-4].minor.yy46 = yymsp[-1].minor.yy46; } + break; + default: + /* (323) input ::= cmdlist */ yytestcase(yyruleno==323); + /* (324) cmdlist ::= cmdlist ecmd */ yytestcase(yyruleno==324); + /* (325) cmdlist ::= ecmd (OPTIMIZED OUT) */ assert(yyruleno!=325); + /* (326) ecmd ::= SEMI */ yytestcase(yyruleno==326); + /* (327) ecmd ::= cmdx SEMI */ yytestcase(yyruleno==327); + /* (328) ecmd ::= explain cmdx */ yytestcase(yyruleno==328); + /* (329) trans_opt ::= */ yytestcase(yyruleno==329); + /* (330) trans_opt ::= TRANSACTION */ yytestcase(yyruleno==330); + /* (331) trans_opt ::= TRANSACTION nm */ yytestcase(yyruleno==331); + /* (332) savepoint_opt ::= SAVEPOINT */ yytestcase(yyruleno==332); + /* (333) savepoint_opt ::= */ yytestcase(yyruleno==333); + /* (334) cmd ::= create_table create_table_args */ yytestcase(yyruleno==334); + /* (335) columnlist ::= columnlist COMMA columnname carglist */ yytestcase(yyruleno==335); + /* (336) columnlist ::= columnname carglist */ yytestcase(yyruleno==336); + /* (337) nm ::= ID|INDEXED */ yytestcase(yyruleno==337); + /* (338) nm ::= STRING */ yytestcase(yyruleno==338); + /* (339) nm ::= JOIN_KW */ yytestcase(yyruleno==339); + /* (340) typetoken ::= typename */ yytestcase(yyruleno==340); + /* (341) typename ::= ID|STRING */ yytestcase(yyruleno==341); + /* (342) signed ::= plus_num (OPTIMIZED OUT) */ assert(yyruleno!=342); + /* (343) signed ::= minus_num (OPTIMIZED OUT) */ assert(yyruleno!=343); + /* (344) carglist ::= carglist ccons */ yytestcase(yyruleno==344); + /* (345) carglist ::= */ yytestcase(yyruleno==345); + /* (346) ccons ::= NULL onconf */ yytestcase(yyruleno==346); + /* (347) conslist_opt ::= COMMA conslist */ yytestcase(yyruleno==347); + /* (348) conslist ::= conslist tconscomma tcons */ yytestcase(yyruleno==348); + /* (349) conslist ::= tcons (OPTIMIZED OUT) */ assert(yyruleno!=349); + /* (350) tconscomma ::= */ yytestcase(yyruleno==350); + /* (351) defer_subclause_opt ::= defer_subclause (OPTIMIZED OUT) */ assert(yyruleno!=351); + /* (352) resolvetype ::= raisetype (OPTIMIZED OUT) */ assert(yyruleno!=352); + /* (353) selectnowith ::= oneselect (OPTIMIZED OUT) */ assert(yyruleno!=353); + /* (354) oneselect ::= values */ yytestcase(yyruleno==354); + /* (355) sclp ::= selcollist COMMA */ yytestcase(yyruleno==355); + /* (356) as ::= ID|STRING */ yytestcase(yyruleno==356); + /* (357) expr ::= term (OPTIMIZED OUT) */ assert(yyruleno!=357); + /* (358) likeop ::= LIKE_KW|MATCH */ yytestcase(yyruleno==358); + /* (359) exprlist ::= nexprlist */ yytestcase(yyruleno==359); + /* (360) nmnum ::= plus_num (OPTIMIZED OUT) */ assert(yyruleno!=360); + /* (361) nmnum ::= nm (OPTIMIZED OUT) */ assert(yyruleno!=361); + /* (362) nmnum ::= ON */ yytestcase(yyruleno==362); + /* (363) nmnum ::= DELETE */ yytestcase(yyruleno==363); + /* (364) nmnum ::= DEFAULT */ yytestcase(yyruleno==364); + /* (365) plus_num ::= INTEGER|FLOAT */ yytestcase(yyruleno==365); + /* (366) foreach_clause ::= */ yytestcase(yyruleno==366); + /* (367) foreach_clause ::= FOR EACH ROW */ yytestcase(yyruleno==367); + /* (368) trnm ::= nm */ yytestcase(yyruleno==368); + /* (369) tridxby ::= */ yytestcase(yyruleno==369); + /* (370) database_kw_opt ::= DATABASE */ yytestcase(yyruleno==370); + /* (371) database_kw_opt ::= */ yytestcase(yyruleno==371); + /* (372) kwcolumn_opt ::= */ yytestcase(yyruleno==372); + /* (373) kwcolumn_opt ::= COLUMNKW */ yytestcase(yyruleno==373); + /* (374) vtabarglist ::= vtabarg */ yytestcase(yyruleno==374); + /* (375) vtabarglist ::= vtabarglist COMMA vtabarg */ yytestcase(yyruleno==375); + /* (376) vtabarg ::= vtabarg vtabargtoken */ yytestcase(yyruleno==376); + /* (377) anylist ::= */ yytestcase(yyruleno==377); + /* (378) anylist ::= anylist LP anylist RP */ yytestcase(yyruleno==378); + /* (379) anylist ::= anylist ANY */ yytestcase(yyruleno==379); + /* (380) with ::= */ yytestcase(yyruleno==380); + break; +/********** End reduce actions ************************************************/ + }; + assert( yyrulenoYY_MAX_SHIFT && yyact<=YY_MAX_SHIFTREDUCE) ); + + /* It is not possible for a REDUCE to be followed by an error */ + assert( yyact!=YY_ERROR_ACTION ); + + yymsp += yysize+1; + yypParser->yytos = yymsp; + yymsp->stateno = (YYACTIONTYPE)yyact; + yymsp->major = (YYCODETYPE)yygoto; + yyTraceShift(yypParser, yyact, "... then shift"); + return yyact; +} + +/* +** The following code executes when the parse fails +*/ +#ifndef YYNOERRORRECOVERY +static void yy_parse_failed( + yyParser *yypParser /* The parser */ +){ + sqlite3ParserARG_FETCH + sqlite3ParserCTX_FETCH +#ifndef NDEBUG + if( yyTraceFILE ){ + fprintf(yyTraceFILE,"%sFail!\n",yyTracePrompt); + } +#endif + while( yypParser->yytos>yypParser->yystack ) yy_pop_parser_stack(yypParser); + /* Here code is inserted which will be executed whenever the + ** parser fails */ +/************ Begin %parse_failure code ***************************************/ +/************ End %parse_failure code *****************************************/ + sqlite3ParserARG_STORE /* Suppress warning about unused %extra_argument variable */ + sqlite3ParserCTX_STORE +} +#endif /* YYNOERRORRECOVERY */ + +/* +** The following code executes when a syntax error first occurs. +*/ +static void yy_syntax_error( + yyParser *yypParser, /* The parser */ + int yymajor, /* The major type of the error token */ + sqlite3ParserTOKENTYPE yyminor /* The minor type of the error token */ +){ + sqlite3ParserARG_FETCH + sqlite3ParserCTX_FETCH +#define TOKEN yyminor +/************ Begin %syntax_error code ****************************************/ + + UNUSED_PARAMETER(yymajor); /* Silence some compiler warnings */ + if( TOKEN.z[0] ){ + sqlite3ErrorMsg(pParse, "near \"%T\": syntax error", &TOKEN); + }else{ + sqlite3ErrorMsg(pParse, "incomplete input"); + } +/************ End %syntax_error code ******************************************/ + sqlite3ParserARG_STORE /* Suppress warning about unused %extra_argument variable */ + sqlite3ParserCTX_STORE +} + +/* +** The following is executed when the parser accepts +*/ +static void yy_accept( + yyParser *yypParser /* The parser */ +){ + sqlite3ParserARG_FETCH + sqlite3ParserCTX_FETCH +#ifndef NDEBUG + if( yyTraceFILE ){ + fprintf(yyTraceFILE,"%sAccept!\n",yyTracePrompt); + } +#endif +#ifndef YYNOERRORRECOVERY + yypParser->yyerrcnt = -1; +#endif + assert( yypParser->yytos==yypParser->yystack ); + /* Here code is inserted which will be executed whenever the + ** parser accepts */ +/*********** Begin %parse_accept code *****************************************/ +/*********** End %parse_accept code *******************************************/ + sqlite3ParserARG_STORE /* Suppress warning about unused %extra_argument variable */ + sqlite3ParserCTX_STORE +} + +/* The main parser program. +** The first argument is a pointer to a structure obtained from +** "sqlite3ParserAlloc" which describes the current state of the parser. +** The second argument is the major token number. The third is +** the minor token. The fourth optional argument is whatever the +** user wants (and specified in the grammar) and is available for +** use by the action routines. +** +** Inputs: +**
      +**
    • A pointer to the parser (an opaque structure.) +**
    • The major token number. +**
    • The minor token number. +**
    • An option argument of a grammar-specified type. +**
    +** +** Outputs: +** None. +*/ +SQLITE_PRIVATE void sqlite3Parser( + void *yyp, /* The parser */ + int yymajor, /* The major token code number */ + sqlite3ParserTOKENTYPE yyminor /* The value for the token */ + sqlite3ParserARG_PDECL /* Optional %extra_argument parameter */ +){ + YYMINORTYPE yyminorunion; + YYACTIONTYPE yyact; /* The parser action. */ +#if !defined(YYERRORSYMBOL) && !defined(YYNOERRORRECOVERY) + int yyendofinput; /* True if we are at the end of input */ +#endif +#ifdef YYERRORSYMBOL + int yyerrorhit = 0; /* True if yymajor has invoked an error */ +#endif + yyParser *yypParser = (yyParser*)yyp; /* The parser */ + sqlite3ParserCTX_FETCH + sqlite3ParserARG_STORE + + assert( yypParser->yytos!=0 ); +#if !defined(YYERRORSYMBOL) && !defined(YYNOERRORRECOVERY) + yyendofinput = (yymajor==0); +#endif + + yyact = yypParser->yytos->stateno; +#ifndef NDEBUG + if( yyTraceFILE ){ + if( yyact < YY_MIN_REDUCE ){ + fprintf(yyTraceFILE,"%sInput '%s' in state %d\n", + yyTracePrompt,yyTokenName[yymajor],yyact); + }else{ + fprintf(yyTraceFILE,"%sInput '%s' with pending reduce %d\n", + yyTracePrompt,yyTokenName[yymajor],yyact-YY_MIN_REDUCE); + } + } +#endif + + do{ + assert( yyact==yypParser->yytos->stateno ); + yyact = yy_find_shift_action((YYCODETYPE)yymajor,yyact); + if( yyact >= YY_MIN_REDUCE ){ + yyact = yy_reduce(yypParser,yyact-YY_MIN_REDUCE,yymajor, + yyminor sqlite3ParserCTX_PARAM); + }else if( yyact <= YY_MAX_SHIFTREDUCE ){ + yy_shift(yypParser,yyact,(YYCODETYPE)yymajor,yyminor); +#ifndef YYNOERRORRECOVERY + yypParser->yyerrcnt--; +#endif + break; + }else if( yyact==YY_ACCEPT_ACTION ){ + yypParser->yytos--; + yy_accept(yypParser); + return; + }else{ + assert( yyact == YY_ERROR_ACTION ); + yyminorunion.yy0 = yyminor; +#ifdef YYERRORSYMBOL + int yymx; +#endif +#ifndef NDEBUG + if( yyTraceFILE ){ + fprintf(yyTraceFILE,"%sSyntax Error!\n",yyTracePrompt); + } +#endif +#ifdef YYERRORSYMBOL + /* A syntax error has occurred. + ** The response to an error depends upon whether or not the + ** grammar defines an error token "ERROR". + ** + ** This is what we do if the grammar does define ERROR: + ** + ** * Call the %syntax_error function. + ** + ** * Begin popping the stack until we enter a state where + ** it is legal to shift the error symbol, then shift + ** the error symbol. + ** + ** * Set the error count to three. + ** + ** * Begin accepting and shifting new tokens. No new error + ** processing will occur until three tokens have been + ** shifted successfully. + ** + */ + if( yypParser->yyerrcnt<0 ){ + yy_syntax_error(yypParser,yymajor,yyminor); + } + yymx = yypParser->yytos->major; + if( yymx==YYERRORSYMBOL || yyerrorhit ){ +#ifndef NDEBUG + if( yyTraceFILE ){ + fprintf(yyTraceFILE,"%sDiscard input token %s\n", + yyTracePrompt,yyTokenName[yymajor]); + } +#endif + yy_destructor(yypParser, (YYCODETYPE)yymajor, &yyminorunion); + yymajor = YYNOCODE; + }else{ + while( yypParser->yytos >= yypParser->yystack + && (yyact = yy_find_reduce_action( + yypParser->yytos->stateno, + YYERRORSYMBOL)) > YY_MAX_SHIFTREDUCE + ){ + yy_pop_parser_stack(yypParser); + } + if( yypParser->yytos < yypParser->yystack || yymajor==0 ){ + yy_destructor(yypParser,(YYCODETYPE)yymajor,&yyminorunion); + yy_parse_failed(yypParser); +#ifndef YYNOERRORRECOVERY + yypParser->yyerrcnt = -1; +#endif + yymajor = YYNOCODE; + }else if( yymx!=YYERRORSYMBOL ){ + yy_shift(yypParser,yyact,YYERRORSYMBOL,yyminor); + } + } + yypParser->yyerrcnt = 3; + yyerrorhit = 1; + if( yymajor==YYNOCODE ) break; + yyact = yypParser->yytos->stateno; +#elif defined(YYNOERRORRECOVERY) + /* If the YYNOERRORRECOVERY macro is defined, then do not attempt to + ** do any kind of error recovery. Instead, simply invoke the syntax + ** error routine and continue going as if nothing had happened. + ** + ** Applications can set this macro (for example inside %include) if + ** they intend to abandon the parse upon the first syntax error seen. + */ + yy_syntax_error(yypParser,yymajor, yyminor); + yy_destructor(yypParser,(YYCODETYPE)yymajor,&yyminorunion); + break; +#else /* YYERRORSYMBOL is not defined */ + /* This is what we do if the grammar does not define ERROR: + ** + ** * Report an error message, and throw away the input token. + ** + ** * If the input token is $, then fail the parse. + ** + ** As before, subsequent error messages are suppressed until + ** three input tokens have been successfully shifted. + */ + if( yypParser->yyerrcnt<=0 ){ + yy_syntax_error(yypParser,yymajor, yyminor); + } + yypParser->yyerrcnt = 3; + yy_destructor(yypParser,(YYCODETYPE)yymajor,&yyminorunion); + if( yyendofinput ){ + yy_parse_failed(yypParser); +#ifndef YYNOERRORRECOVERY + yypParser->yyerrcnt = -1; +#endif + } + break; +#endif + } + }while( yypParser->yytos>yypParser->yystack ); +#ifndef NDEBUG + if( yyTraceFILE ){ + yyStackEntry *i; + char cDiv = '['; + fprintf(yyTraceFILE,"%sReturn. Stack=",yyTracePrompt); + for(i=&yypParser->yystack[1]; i<=yypParser->yytos; i++){ + fprintf(yyTraceFILE,"%c%s", cDiv, yyTokenName[i->major]); + cDiv = ' '; + } + fprintf(yyTraceFILE,"]\n"); + } +#endif + return; +} + +/* +** Return the fallback token corresponding to canonical token iToken, or +** 0 if iToken has no fallback. +*/ +SQLITE_PRIVATE int sqlite3ParserFallback(int iToken){ +#ifdef YYFALLBACK + assert( iToken<(int)(sizeof(yyFallback)/sizeof(yyFallback[0])) ); + return yyFallback[iToken]; +#else + (void)iToken; +#endif + return 0; +} + +/************** End of parse.c ***********************************************/ +/************** Begin file tokenize.c ****************************************/ +/* +** 2001 September 15 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** An tokenizer for SQL +** +** This file contains C code that splits an SQL input string up into +** individual tokens and sends those tokens one-by-one over to the +** parser for analysis. +*/ +/* #include "sqliteInt.h" */ +/* #include */ + +/* Character classes for tokenizing +** +** In the sqlite3GetToken() function, a switch() on aiClass[c] is implemented +** using a lookup table, whereas a switch() directly on c uses a binary search. +** The lookup table is much faster. To maximize speed, and to ensure that +** a lookup table is used, all of the classes need to be small integers and +** all of them need to be used within the switch. +*/ +#define CC_X 0 /* The letter 'x', or start of BLOB literal */ +#define CC_KYWD 1 /* Alphabetics or '_'. Usable in a keyword */ +#define CC_ID 2 /* unicode characters usable in IDs */ +#define CC_DIGIT 3 /* Digits */ +#define CC_DOLLAR 4 /* '$' */ +#define CC_VARALPHA 5 /* '@', '#', ':'. Alphabetic SQL variables */ +#define CC_VARNUM 6 /* '?'. Numeric SQL variables */ +#define CC_SPACE 7 /* Space characters */ +#define CC_QUOTE 8 /* '"', '\'', or '`'. String literals, quoted ids */ +#define CC_QUOTE2 9 /* '['. [...] style quoted ids */ +#define CC_PIPE 10 /* '|'. Bitwise OR or concatenate */ +#define CC_MINUS 11 /* '-'. Minus or SQL-style comment */ +#define CC_LT 12 /* '<'. Part of < or <= or <> */ +#define CC_GT 13 /* '>'. Part of > or >= */ +#define CC_EQ 14 /* '='. Part of = or == */ +#define CC_BANG 15 /* '!'. Part of != */ +#define CC_SLASH 16 /* '/'. / or c-style comment */ +#define CC_LP 17 /* '(' */ +#define CC_RP 18 /* ')' */ +#define CC_SEMI 19 /* ';' */ +#define CC_PLUS 20 /* '+' */ +#define CC_STAR 21 /* '*' */ +#define CC_PERCENT 22 /* '%' */ +#define CC_COMMA 23 /* ',' */ +#define CC_AND 24 /* '&' */ +#define CC_TILDA 25 /* '~' */ +#define CC_DOT 26 /* '.' */ +#define CC_ILLEGAL 27 /* Illegal character */ +#define CC_NUL 28 /* 0x00 */ + +static const unsigned char aiClass[] = { +#ifdef SQLITE_ASCII +/* x0 x1 x2 x3 x4 x5 x6 x7 x8 x9 xa xb xc xd xe xf */ +/* 0x */ 28, 27, 27, 27, 27, 27, 27, 27, 27, 7, 7, 27, 7, 7, 27, 27, +/* 1x */ 27, 27, 27, 27, 27, 27, 27, 27, 27, 27, 27, 27, 27, 27, 27, 27, +/* 2x */ 7, 15, 8, 5, 4, 22, 24, 8, 17, 18, 21, 20, 23, 11, 26, 16, +/* 3x */ 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 5, 19, 12, 14, 13, 6, +/* 4x */ 5, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, +/* 5x */ 1, 1, 1, 1, 1, 1, 1, 1, 0, 1, 1, 9, 27, 27, 27, 1, +/* 6x */ 8, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, +/* 7x */ 1, 1, 1, 1, 1, 1, 1, 1, 0, 1, 1, 27, 10, 27, 25, 27, +/* 8x */ 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, +/* 9x */ 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, +/* Ax */ 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, +/* Bx */ 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, +/* Cx */ 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, +/* Dx */ 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, +/* Ex */ 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, +/* Fx */ 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2 +#endif +#ifdef SQLITE_EBCDIC +/* x0 x1 x2 x3 x4 x5 x6 x7 x8 x9 xa xb xc xd xe xf */ +/* 0x */ 27, 27, 27, 27, 27, 7, 27, 27, 27, 27, 27, 27, 7, 7, 27, 27, +/* 1x */ 27, 27, 27, 27, 27, 27, 27, 27, 27, 27, 27, 27, 27, 27, 27, 27, +/* 2x */ 27, 27, 27, 27, 27, 7, 27, 27, 27, 27, 27, 27, 27, 27, 27, 27, +/* 3x */ 27, 27, 27, 27, 27, 27, 27, 27, 27, 27, 27, 27, 27, 27, 27, 27, +/* 4x */ 7, 27, 27, 27, 27, 27, 27, 27, 27, 27, 27, 26, 12, 17, 20, 10, +/* 5x */ 24, 27, 27, 27, 27, 27, 27, 27, 27, 27, 15, 4, 21, 18, 19, 27, +/* 6x */ 11, 16, 27, 27, 27, 27, 27, 27, 27, 27, 27, 23, 22, 1, 13, 6, +/* 7x */ 27, 27, 27, 27, 27, 27, 27, 27, 27, 8, 5, 5, 5, 8, 14, 8, +/* 8x */ 27, 1, 1, 1, 1, 1, 1, 1, 1, 1, 27, 27, 27, 27, 27, 27, +/* 9x */ 27, 1, 1, 1, 1, 1, 1, 1, 1, 1, 27, 27, 27, 27, 27, 27, +/* Ax */ 27, 25, 1, 1, 1, 1, 1, 0, 1, 1, 27, 27, 27, 27, 27, 27, +/* Bx */ 27, 27, 27, 27, 27, 27, 27, 27, 27, 27, 9, 27, 27, 27, 27, 27, +/* Cx */ 27, 1, 1, 1, 1, 1, 1, 1, 1, 1, 27, 27, 27, 27, 27, 27, +/* Dx */ 27, 1, 1, 1, 1, 1, 1, 1, 1, 1, 27, 27, 27, 27, 27, 27, +/* Ex */ 27, 27, 1, 1, 1, 1, 1, 0, 1, 1, 27, 27, 27, 27, 27, 27, +/* Fx */ 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 27, 27, 27, 27, 27, 27, +#endif +}; + +/* +** The charMap() macro maps alphabetic characters (only) into their +** lower-case ASCII equivalent. On ASCII machines, this is just +** an upper-to-lower case map. On EBCDIC machines we also need +** to adjust the encoding. The mapping is only valid for alphabetics +** which are the only characters for which this feature is used. +** +** Used by keywordhash.h +*/ +#ifdef SQLITE_ASCII +# define charMap(X) sqlite3UpperToLower[(unsigned char)X] +#endif +#ifdef SQLITE_EBCDIC +# define charMap(X) ebcdicToAscii[(unsigned char)X] +const unsigned char ebcdicToAscii[] = { +/* 0 1 2 3 4 5 6 7 8 9 A B C D E F */ + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, /* 0x */ + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, /* 1x */ + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, /* 2x */ + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, /* 3x */ + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, /* 4x */ + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, /* 5x */ + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 95, 0, 0, /* 6x */ + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, /* 7x */ + 0, 97, 98, 99,100,101,102,103,104,105, 0, 0, 0, 0, 0, 0, /* 8x */ + 0,106,107,108,109,110,111,112,113,114, 0, 0, 0, 0, 0, 0, /* 9x */ + 0, 0,115,116,117,118,119,120,121,122, 0, 0, 0, 0, 0, 0, /* Ax */ + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, /* Bx */ + 0, 97, 98, 99,100,101,102,103,104,105, 0, 0, 0, 0, 0, 0, /* Cx */ + 0,106,107,108,109,110,111,112,113,114, 0, 0, 0, 0, 0, 0, /* Dx */ + 0, 0,115,116,117,118,119,120,121,122, 0, 0, 0, 0, 0, 0, /* Ex */ + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, /* Fx */ +}; +#endif + +/* +** The sqlite3KeywordCode function looks up an identifier to determine if +** it is a keyword. If it is a keyword, the token code of that keyword is +** returned. If the input is not a keyword, TK_ID is returned. +** +** The implementation of this routine was generated by a program, +** mkkeywordhash.c, located in the tool subdirectory of the distribution. +** The output of the mkkeywordhash.c program is written into a file +** named keywordhash.h and then included into this source file by +** the #include below. +*/ +/************** Include keywordhash.h in the middle of tokenize.c ************/ +/************** Begin file keywordhash.h *************************************/ +/***** This file contains automatically generated code ****** +** +** The code in this file has been automatically generated by +** +** sqlite/tool/mkkeywordhash.c +** +** The code in this file implements a function that determines whether +** or not a given identifier is really an SQL keyword. The same thing +** might be implemented more directly using a hand-written hash table. +** But by using this automatically generated code, the size of the code +** is substantially reduced. This is important for embedded applications +** on platforms with limited memory. +*/ +/* Hash score: 221 */ +/* zKWText[] encodes 967 bytes of keyword text in 638 bytes */ +/* REINDEXEDESCAPEACHECKEYBEFOREIGNOREGEXPLAINSTEADDATABASELECT */ +/* ABLEFTHENDEFERRABLELSEXCLUDELETEMPORARYISNULLSAVEPOINTERSECT */ +/* IESNOTNULLIKEXCEPTRANSACTIONATURALTERAISEXCLUSIVEXISTS */ +/* CONSTRAINTOFFSETRIGGEREFERENCESUNIQUERYWITHOUTERELEASEATTACH */ +/* AVINGLOBEGINNERANGEBETWEENOTHINGROUPSCASCADETACHCASECOLLATE */ +/* CREATECURRENT_DATEIMMEDIATEJOINSERTMATCHPLANALYZEPRAGMABORT */ +/* UPDATEVALUESVIRTUALASTWHENWHERECURSIVEAFTERENAMEANDEFAULT */ +/* AUTOINCREMENTCASTCOLUMNCOMMITCONFLICTCROSSCURRENT_TIMESTAMP */ +/* ARTITIONDEFERREDISTINCTDROPRECEDINGFAILIMITFILTEREPLACEFIRST */ +/* FOLLOWINGFROMFULLIFORDERESTRICTOTHERSOVERIGHTROLLBACKROWS */ +/* UNBOUNDEDUNIONUSINGVACUUMVIEWINDOWBYINITIALLYPRIMARY */ +static const char zKWText[637] = { + 'R','E','I','N','D','E','X','E','D','E','S','C','A','P','E','A','C','H', + 'E','C','K','E','Y','B','E','F','O','R','E','I','G','N','O','R','E','G', + 'E','X','P','L','A','I','N','S','T','E','A','D','D','A','T','A','B','A', + 'S','E','L','E','C','T','A','B','L','E','F','T','H','E','N','D','E','F', + 'E','R','R','A','B','L','E','L','S','E','X','C','L','U','D','E','L','E', + 'T','E','M','P','O','R','A','R','Y','I','S','N','U','L','L','S','A','V', + 'E','P','O','I','N','T','E','R','S','E','C','T','I','E','S','N','O','T', + 'N','U','L','L','I','K','E','X','C','E','P','T','R','A','N','S','A','C', + 'T','I','O','N','A','T','U','R','A','L','T','E','R','A','I','S','E','X', + 'C','L','U','S','I','V','E','X','I','S','T','S','C','O','N','S','T','R', + 'A','I','N','T','O','F','F','S','E','T','R','I','G','G','E','R','E','F', + 'E','R','E','N','C','E','S','U','N','I','Q','U','E','R','Y','W','I','T', + 'H','O','U','T','E','R','E','L','E','A','S','E','A','T','T','A','C','H', + 'A','V','I','N','G','L','O','B','E','G','I','N','N','E','R','A','N','G', + 'E','B','E','T','W','E','E','N','O','T','H','I','N','G','R','O','U','P', + 'S','C','A','S','C','A','D','E','T','A','C','H','C','A','S','E','C','O', + 'L','L','A','T','E','C','R','E','A','T','E','C','U','R','R','E','N','T', + '_','D','A','T','E','I','M','M','E','D','I','A','T','E','J','O','I','N', + 'S','E','R','T','M','A','T','C','H','P','L','A','N','A','L','Y','Z','E', + 'P','R','A','G','M','A','B','O','R','T','U','P','D','A','T','E','V','A', + 'L','U','E','S','V','I','R','T','U','A','L','A','S','T','W','H','E','N', + 'W','H','E','R','E','C','U','R','S','I','V','E','A','F','T','E','R','E', + 'N','A','M','E','A','N','D','E','F','A','U','L','T','A','U','T','O','I', + 'N','C','R','E','M','E','N','T','C','A','S','T','C','O','L','U','M','N', + 'C','O','M','M','I','T','C','O','N','F','L','I','C','T','C','R','O','S', + 'S','C','U','R','R','E','N','T','_','T','I','M','E','S','T','A','M','P', + 'A','R','T','I','T','I','O','N','D','E','F','E','R','R','E','D','I','S', + 'T','I','N','C','T','D','R','O','P','R','E','C','E','D','I','N','G','F', + 'A','I','L','I','M','I','T','F','I','L','T','E','R','E','P','L','A','C', + 'E','F','I','R','S','T','F','O','L','L','O','W','I','N','G','F','R','O', + 'M','F','U','L','L','I','F','O','R','D','E','R','E','S','T','R','I','C', + 'T','O','T','H','E','R','S','O','V','E','R','I','G','H','T','R','O','L', + 'L','B','A','C','K','R','O','W','S','U','N','B','O','U','N','D','E','D', + 'U','N','I','O','N','U','S','I','N','G','V','A','C','U','U','M','V','I', + 'E','W','I','N','D','O','W','B','Y','I','N','I','T','I','A','L','L','Y', + 'P','R','I','M','A','R','Y', +}; +/* aKWHash[i] is the hash value for the i-th keyword */ +static const unsigned char aKWHash[127] = { + 82, 113, 130, 80, 110, 29, 0, 0, 89, 0, 83, 70, 0, + 53, 35, 84, 15, 0, 129, 92, 64, 124, 131, 19, 0, 0, + 136, 0, 134, 126, 0, 22, 100, 0, 9, 0, 0, 121, 78, + 0, 76, 6, 0, 58, 97, 143, 0, 132, 108, 0, 0, 48, + 0, 111, 24, 0, 17, 0, 137, 63, 23, 26, 5, 65, 138, + 103, 120, 0, 142, 114, 69, 141, 66, 118, 72, 0, 98, 0, + 107, 41, 0, 106, 0, 0, 0, 102, 99, 104, 109, 123, 14, + 50, 122, 0, 87, 0, 139, 119, 140, 68, 127, 135, 86, 81, + 37, 91, 117, 0, 0, 101, 51, 128, 125, 0, 133, 0, 0, + 44, 0, 93, 67, 39, 0, 20, 45, 115, 88, +}; +/* aKWNext[] forms the hash collision chain. If aKWHash[i]==0 +** then the i-th keyword has no more hash collisions. Otherwise, +** the next keyword with the same hash is aKWHash[i]-1. */ +static const unsigned char aKWNext[143] = { + 0, 0, 0, 0, 4, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 2, 0, 0, 0, 0, 0, 0, 13, 0, 0, 0, 0, + 0, 0, 0, 21, 0, 0, 0, 0, 12, 0, 0, 0, 0, + 0, 0, 0, 7, 0, 36, 0, 0, 28, 0, 0, 0, 31, + 0, 0, 0, 40, 0, 0, 0, 0, 0, 60, 0, 54, 0, + 0, 38, 47, 0, 0, 0, 3, 0, 0, 74, 1, 73, 0, + 0, 0, 52, 0, 0, 0, 0, 0, 0, 57, 59, 56, 30, + 0, 0, 0, 46, 0, 16, 49, 10, 0, 0, 0, 0, 0, + 0, 0, 11, 79, 95, 0, 0, 8, 0, 112, 0, 105, 0, + 43, 62, 0, 77, 0, 116, 0, 61, 0, 0, 94, 42, 55, + 0, 75, 34, 90, 32, 33, 27, 25, 18, 96, 0, 71, 85, +}; +/* aKWLen[i] is the length (in bytes) of the i-th keyword */ +static const unsigned char aKWLen[143] = { + 7, 7, 5, 4, 6, 4, 5, 3, 6, 7, 3, 6, 6, + 7, 7, 3, 8, 2, 6, 5, 4, 4, 3, 10, 4, 7, + 6, 9, 4, 2, 6, 5, 9, 9, 4, 7, 3, 2, 4, + 4, 6, 11, 6, 2, 7, 5, 5, 9, 6, 10, 4, 6, + 2, 3, 7, 10, 6, 5, 7, 4, 5, 7, 6, 6, 4, + 5, 5, 5, 7, 7, 6, 5, 7, 3, 6, 4, 7, 6, + 12, 9, 4, 6, 5, 4, 7, 6, 5, 6, 6, 7, 4, + 4, 5, 9, 5, 6, 3, 7, 13, 2, 2, 4, 6, 6, + 8, 5, 17, 12, 7, 9, 8, 8, 2, 4, 9, 4, 5, + 6, 7, 5, 9, 4, 4, 2, 5, 8, 6, 4, 5, 8, + 4, 3, 9, 5, 5, 6, 4, 6, 2, 2, 9, 3, 7, +}; +/* aKWOffset[i] is the index into zKWText[] of the start of +** the text for the i-th keyword. */ +static const unsigned short int aKWOffset[143] = { + 0, 2, 2, 8, 9, 14, 16, 20, 23, 25, 25, 29, 33, + 36, 41, 46, 48, 53, 54, 59, 62, 65, 67, 69, 78, 81, + 86, 90, 90, 94, 99, 101, 105, 111, 119, 123, 123, 123, 126, + 129, 132, 137, 142, 146, 147, 152, 156, 160, 168, 174, 181, 184, + 184, 187, 189, 195, 205, 208, 213, 213, 217, 221, 228, 233, 238, + 241, 244, 248, 253, 259, 265, 265, 271, 272, 276, 282, 286, 293, + 299, 311, 320, 322, 328, 333, 335, 342, 347, 352, 358, 364, 370, + 374, 378, 381, 390, 394, 400, 402, 409, 411, 413, 422, 426, 432, + 438, 446, 451, 451, 451, 467, 476, 483, 484, 491, 494, 503, 506, + 511, 516, 523, 528, 537, 541, 545, 547, 551, 559, 565, 568, 573, + 581, 581, 585, 594, 599, 604, 610, 613, 616, 619, 621, 626, 630, +}; +/* aKWCode[i] is the parser symbol code for the i-th keyword */ +static const unsigned char aKWCode[143] = { + TK_REINDEX, TK_INDEXED, TK_INDEX, TK_DESC, TK_ESCAPE, + TK_EACH, TK_CHECK, TK_KEY, TK_BEFORE, TK_FOREIGN, + TK_FOR, TK_IGNORE, TK_LIKE_KW, TK_EXPLAIN, TK_INSTEAD, + TK_ADD, TK_DATABASE, TK_AS, TK_SELECT, TK_TABLE, + TK_JOIN_KW, TK_THEN, TK_END, TK_DEFERRABLE, TK_ELSE, + TK_EXCLUDE, TK_DELETE, TK_TEMP, TK_TEMP, TK_OR, + TK_ISNULL, TK_NULLS, TK_SAVEPOINT, TK_INTERSECT, TK_TIES, + TK_NOTNULL, TK_NOT, TK_NO, TK_NULL, TK_LIKE_KW, + TK_EXCEPT, TK_TRANSACTION,TK_ACTION, TK_ON, TK_JOIN_KW, + TK_ALTER, TK_RAISE, TK_EXCLUSIVE, TK_EXISTS, TK_CONSTRAINT, + TK_INTO, TK_OFFSET, TK_OF, TK_SET, TK_TRIGGER, + TK_REFERENCES, TK_UNIQUE, TK_QUERY, TK_WITHOUT, TK_WITH, + TK_JOIN_KW, TK_RELEASE, TK_ATTACH, TK_HAVING, TK_LIKE_KW, + TK_BEGIN, TK_JOIN_KW, TK_RANGE, TK_BETWEEN, TK_NOTHING, + TK_GROUPS, TK_GROUP, TK_CASCADE, TK_ASC, TK_DETACH, + TK_CASE, TK_COLLATE, TK_CREATE, TK_CTIME_KW, TK_IMMEDIATE, + TK_JOIN, TK_INSERT, TK_MATCH, TK_PLAN, TK_ANALYZE, + TK_PRAGMA, TK_ABORT, TK_UPDATE, TK_VALUES, TK_VIRTUAL, + TK_LAST, TK_WHEN, TK_WHERE, TK_RECURSIVE, TK_AFTER, + TK_RENAME, TK_AND, TK_DEFAULT, TK_AUTOINCR, TK_TO, + TK_IN, TK_CAST, TK_COLUMNKW, TK_COMMIT, TK_CONFLICT, + TK_JOIN_KW, TK_CTIME_KW, TK_CTIME_KW, TK_CURRENT, TK_PARTITION, + TK_DEFERRED, TK_DISTINCT, TK_IS, TK_DROP, TK_PRECEDING, + TK_FAIL, TK_LIMIT, TK_FILTER, TK_REPLACE, TK_FIRST, + TK_FOLLOWING, TK_FROM, TK_JOIN_KW, TK_IF, TK_ORDER, + TK_RESTRICT, TK_OTHERS, TK_OVER, TK_JOIN_KW, TK_ROLLBACK, + TK_ROWS, TK_ROW, TK_UNBOUNDED, TK_UNION, TK_USING, + TK_VACUUM, TK_VIEW, TK_WINDOW, TK_DO, TK_BY, + TK_INITIALLY, TK_ALL, TK_PRIMARY, +}; +/* Check to see if z[0..n-1] is a keyword. If it is, write the +** parser symbol code for that keyword into *pType. Always +** return the integer n (the length of the token). */ +static int keywordCode(const char *z, int n, int *pType){ + int i, j; + const char *zKW; + if( n>=2 ){ + i = ((charMap(z[0])*4) ^ (charMap(z[n-1])*3) ^ n) % 127; + for(i=((int)aKWHash[i])-1; i>=0; i=((int)aKWNext[i])-1){ + if( aKWLen[i]!=n ) continue; + j = 0; + zKW = &zKWText[aKWOffset[i]]; +#ifdef SQLITE_ASCII + while( j=SQLITE_N_KEYWORD ) return SQLITE_ERROR; + *pzName = zKWText + aKWOffset[i]; + *pnName = aKWLen[i]; + return SQLITE_OK; +} +SQLITE_API int sqlite3_keyword_count(void){ return SQLITE_N_KEYWORD; } +SQLITE_API int sqlite3_keyword_check(const char *zName, int nName){ + return TK_ID!=sqlite3KeywordCode((const u8*)zName, nName); +} + +/************** End of keywordhash.h *****************************************/ +/************** Continuing where we left off in tokenize.c *******************/ + + +/* +** If X is a character that can be used in an identifier then +** IdChar(X) will be true. Otherwise it is false. +** +** For ASCII, any character with the high-order bit set is +** allowed in an identifier. For 7-bit characters, +** sqlite3IsIdChar[X] must be 1. +** +** For EBCDIC, the rules are more complex but have the same +** end result. +** +** Ticket #1066. the SQL standard does not allow '$' in the +** middle of identifiers. But many SQL implementations do. +** SQLite will allow '$' in identifiers for compatibility. +** But the feature is undocumented. +*/ +#ifdef SQLITE_ASCII +#define IdChar(C) ((sqlite3CtypeMap[(unsigned char)C]&0x46)!=0) +#endif +#ifdef SQLITE_EBCDIC +SQLITE_PRIVATE const char sqlite3IsEbcdicIdChar[] = { +/* x0 x1 x2 x3 x4 x5 x6 x7 x8 x9 xA xB xC xD xE xF */ + 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, /* 4x */ + 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 1, 0, 0, 0, 0, /* 5x */ + 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 1, 0, 0, /* 6x */ + 0, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, /* 7x */ + 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 1, 1, 1, 0, /* 8x */ + 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 1, 0, 1, 0, /* 9x */ + 1, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 1, 1, 1, 0, /* Ax */ + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, /* Bx */ + 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 1, 1, 1, 1, 1, /* Cx */ + 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 1, 1, 1, 1, 1, /* Dx */ + 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 0, 1, 1, 1, 1, 1, /* Ex */ + 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 1, 1, 1, 1, 0, /* Fx */ +}; +#define IdChar(C) (((c=C)>=0x42 && sqlite3IsEbcdicIdChar[c-0x40])) +#endif + +/* Make the IdChar function accessible from ctime.c and alter.c */ +SQLITE_PRIVATE int sqlite3IsIdChar(u8 c){ return IdChar(c); } + +#ifndef SQLITE_OMIT_WINDOWFUNC +/* +** Return the id of the next token in string (*pz). Before returning, set +** (*pz) to point to the byte following the parsed token. +*/ +static int getToken(const unsigned char **pz){ + const unsigned char *z = *pz; + int t; /* Token type to return */ + do { + z += sqlite3GetToken(z, &t); + }while( t==TK_SPACE ); + if( t==TK_ID + || t==TK_STRING + || t==TK_JOIN_KW + || t==TK_WINDOW + || t==TK_OVER + || sqlite3ParserFallback(t)==TK_ID + ){ + t = TK_ID; + } + *pz = z; + return t; +} + +/* +** The following three functions are called immediately after the tokenizer +** reads the keywords WINDOW, OVER and FILTER, respectively, to determine +** whether the token should be treated as a keyword or an SQL identifier. +** This cannot be handled by the usual lemon %fallback method, due to +** the ambiguity in some constructions. e.g. +** +** SELECT sum(x) OVER ... +** +** In the above, "OVER" might be a keyword, or it might be an alias for the +** sum(x) expression. If a "%fallback ID OVER" directive were added to +** grammar, then SQLite would always treat "OVER" as an alias, making it +** impossible to call a window-function without a FILTER clause. +** +** WINDOW is treated as a keyword if: +** +** * the following token is an identifier, or a keyword that can fallback +** to being an identifier, and +** * the token after than one is TK_AS. +** +** OVER is a keyword if: +** +** * the previous token was TK_RP, and +** * the next token is either TK_LP or an identifier. +** +** FILTER is a keyword if: +** +** * the previous token was TK_RP, and +** * the next token is TK_LP. +*/ +static int analyzeWindowKeyword(const unsigned char *z){ + int t; + t = getToken(&z); + if( t!=TK_ID ) return TK_ID; + t = getToken(&z); + if( t!=TK_AS ) return TK_ID; + return TK_WINDOW; +} +static int analyzeOverKeyword(const unsigned char *z, int lastToken){ + if( lastToken==TK_RP ){ + int t = getToken(&z); + if( t==TK_LP || t==TK_ID ) return TK_OVER; + } + return TK_ID; +} +static int analyzeFilterKeyword(const unsigned char *z, int lastToken){ + if( lastToken==TK_RP && getToken(&z)==TK_LP ){ + return TK_FILTER; + } + return TK_ID; +} +#endif /* SQLITE_OMIT_WINDOWFUNC */ + +/* +** Return the length (in bytes) of the token that begins at z[0]. +** Store the token type in *tokenType before returning. +*/ +SQLITE_PRIVATE int sqlite3GetToken(const unsigned char *z, int *tokenType){ + int i, c; + switch( aiClass[*z] ){ /* Switch on the character-class of the first byte + ** of the token. See the comment on the CC_ defines + ** above. */ + case CC_SPACE: { + testcase( z[0]==' ' ); + testcase( z[0]=='\t' ); + testcase( z[0]=='\n' ); + testcase( z[0]=='\f' ); + testcase( z[0]=='\r' ); + for(i=1; sqlite3Isspace(z[i]); i++){} + *tokenType = TK_SPACE; + return i; + } + case CC_MINUS: { + if( z[1]=='-' ){ + for(i=2; (c=z[i])!=0 && c!='\n'; i++){} + *tokenType = TK_SPACE; /* IMP: R-22934-25134 */ + return i; + } + *tokenType = TK_MINUS; + return 1; + } + case CC_LP: { + *tokenType = TK_LP; + return 1; + } + case CC_RP: { + *tokenType = TK_RP; + return 1; + } + case CC_SEMI: { + *tokenType = TK_SEMI; + return 1; + } + case CC_PLUS: { + *tokenType = TK_PLUS; + return 1; + } + case CC_STAR: { + *tokenType = TK_STAR; + return 1; + } + case CC_SLASH: { + if( z[1]!='*' || z[2]==0 ){ + *tokenType = TK_SLASH; + return 1; + } + for(i=3, c=z[2]; (c!='*' || z[i]!='/') && (c=z[i])!=0; i++){} + if( c ) i++; + *tokenType = TK_SPACE; /* IMP: R-22934-25134 */ + return i; + } + case CC_PERCENT: { + *tokenType = TK_REM; + return 1; + } + case CC_EQ: { + *tokenType = TK_EQ; + return 1 + (z[1]=='='); + } + case CC_LT: { + if( (c=z[1])=='=' ){ + *tokenType = TK_LE; + return 2; + }else if( c=='>' ){ + *tokenType = TK_NE; + return 2; + }else if( c=='<' ){ + *tokenType = TK_LSHIFT; + return 2; + }else{ + *tokenType = TK_LT; + return 1; + } + } + case CC_GT: { + if( (c=z[1])=='=' ){ + *tokenType = TK_GE; + return 2; + }else if( c=='>' ){ + *tokenType = TK_RSHIFT; + return 2; + }else{ + *tokenType = TK_GT; + return 1; + } + } + case CC_BANG: { + if( z[1]!='=' ){ + *tokenType = TK_ILLEGAL; + return 1; + }else{ + *tokenType = TK_NE; + return 2; + } + } + case CC_PIPE: { + if( z[1]!='|' ){ + *tokenType = TK_BITOR; + return 1; + }else{ + *tokenType = TK_CONCAT; + return 2; + } + } + case CC_COMMA: { + *tokenType = TK_COMMA; + return 1; + } + case CC_AND: { + *tokenType = TK_BITAND; + return 1; + } + case CC_TILDA: { + *tokenType = TK_BITNOT; + return 1; + } + case CC_QUOTE: { + int delim = z[0]; + testcase( delim=='`' ); + testcase( delim=='\'' ); + testcase( delim=='"' ); + for(i=1; (c=z[i])!=0; i++){ + if( c==delim ){ + if( z[i+1]==delim ){ + i++; + }else{ + break; + } + } + } + if( c=='\'' ){ + *tokenType = TK_STRING; + return i+1; + }else if( c!=0 ){ + *tokenType = TK_ID; + return i+1; + }else{ + *tokenType = TK_ILLEGAL; + return i; + } + } + case CC_DOT: { +#ifndef SQLITE_OMIT_FLOATING_POINT + if( !sqlite3Isdigit(z[1]) ) +#endif + { + *tokenType = TK_DOT; + return 1; + } + /* If the next character is a digit, this is a floating point + ** number that begins with ".". Fall thru into the next case */ + } + case CC_DIGIT: { + testcase( z[0]=='0' ); testcase( z[0]=='1' ); testcase( z[0]=='2' ); + testcase( z[0]=='3' ); testcase( z[0]=='4' ); testcase( z[0]=='5' ); + testcase( z[0]=='6' ); testcase( z[0]=='7' ); testcase( z[0]=='8' ); + testcase( z[0]=='9' ); + *tokenType = TK_INTEGER; +#ifndef SQLITE_OMIT_HEX_INTEGER + if( z[0]=='0' && (z[1]=='x' || z[1]=='X') && sqlite3Isxdigit(z[2]) ){ + for(i=3; sqlite3Isxdigit(z[i]); i++){} + return i; + } +#endif + for(i=0; sqlite3Isdigit(z[i]); i++){} +#ifndef SQLITE_OMIT_FLOATING_POINT + if( z[i]=='.' ){ + i++; + while( sqlite3Isdigit(z[i]) ){ i++; } + *tokenType = TK_FLOAT; + } + if( (z[i]=='e' || z[i]=='E') && + ( sqlite3Isdigit(z[i+1]) + || ((z[i+1]=='+' || z[i+1]=='-') && sqlite3Isdigit(z[i+2])) + ) + ){ + i += 2; + while( sqlite3Isdigit(z[i]) ){ i++; } + *tokenType = TK_FLOAT; + } +#endif + while( IdChar(z[i]) ){ + *tokenType = TK_ILLEGAL; + i++; + } + return i; + } + case CC_QUOTE2: { + for(i=1, c=z[0]; c!=']' && (c=z[i])!=0; i++){} + *tokenType = c==']' ? TK_ID : TK_ILLEGAL; + return i; + } + case CC_VARNUM: { + *tokenType = TK_VARIABLE; + for(i=1; sqlite3Isdigit(z[i]); i++){} + return i; + } + case CC_DOLLAR: + case CC_VARALPHA: { + int n = 0; + testcase( z[0]=='$' ); testcase( z[0]=='@' ); + testcase( z[0]==':' ); testcase( z[0]=='#' ); + *tokenType = TK_VARIABLE; + for(i=1; (c=z[i])!=0; i++){ + if( IdChar(c) ){ + n++; +#ifndef SQLITE_OMIT_TCL_VARIABLE + }else if( c=='(' && n>0 ){ + do{ + i++; + }while( (c=z[i])!=0 && !sqlite3Isspace(c) && c!=')' ); + if( c==')' ){ + i++; + }else{ + *tokenType = TK_ILLEGAL; + } + break; + }else if( c==':' && z[i+1]==':' ){ + i++; +#endif + }else{ + break; + } + } + if( n==0 ) *tokenType = TK_ILLEGAL; + return i; + } + case CC_KYWD: { + for(i=1; aiClass[z[i]]<=CC_KYWD; i++){} + if( IdChar(z[i]) ){ + /* This token started out using characters that can appear in keywords, + ** but z[i] is a character not allowed within keywords, so this must + ** be an identifier instead */ + i++; + break; + } + *tokenType = TK_ID; + return keywordCode((char*)z, i, tokenType); + } + case CC_X: { +#ifndef SQLITE_OMIT_BLOB_LITERAL + testcase( z[0]=='x' ); testcase( z[0]=='X' ); + if( z[1]=='\'' ){ + *tokenType = TK_BLOB; + for(i=2; sqlite3Isxdigit(z[i]); i++){} + if( z[i]!='\'' || i%2 ){ + *tokenType = TK_ILLEGAL; + while( z[i] && z[i]!='\'' ){ i++; } + } + if( z[i] ) i++; + return i; + } +#endif + /* If it is not a BLOB literal, then it must be an ID, since no + ** SQL keywords start with the letter 'x'. Fall through */ + } + case CC_ID: { + i = 1; + break; + } + case CC_NUL: { + *tokenType = TK_ILLEGAL; + return 0; + } + default: { + *tokenType = TK_ILLEGAL; + return 1; + } + } + while( IdChar(z[i]) ){ i++; } + *tokenType = TK_ID; + return i; +} + +/* +** Run the parser on the given SQL string. The parser structure is +** passed in. An SQLITE_ status code is returned. If an error occurs +** then an and attempt is made to write an error message into +** memory obtained from sqlite3_malloc() and to make *pzErrMsg point to that +** error message. +*/ +SQLITE_PRIVATE int sqlite3RunParser(Parse *pParse, const char *zSql, char **pzErrMsg){ + int nErr = 0; /* Number of errors encountered */ + void *pEngine; /* The LEMON-generated LALR(1) parser */ + int n = 0; /* Length of the next token token */ + int tokenType; /* type of the next token */ + int lastTokenParsed = -1; /* type of the previous token */ + sqlite3 *db = pParse->db; /* The database connection */ + int mxSqlLen; /* Max length of an SQL string */ +#ifdef sqlite3Parser_ENGINEALWAYSONSTACK + yyParser sEngine; /* Space to hold the Lemon-generated Parser object */ +#endif + VVA_ONLY( u8 startedWithOom = db->mallocFailed ); + + assert( zSql!=0 ); + mxSqlLen = db->aLimit[SQLITE_LIMIT_SQL_LENGTH]; + if( db->nVdbeActive==0 ){ + db->u1.isInterrupted = 0; + } + pParse->rc = SQLITE_OK; + pParse->zTail = zSql; + assert( pzErrMsg!=0 ); +#ifdef SQLITE_DEBUG + if( db->flags & SQLITE_ParserTrace ){ + printf("parser: [[[%s]]]\n", zSql); + sqlite3ParserTrace(stdout, "parser: "); + }else{ + sqlite3ParserTrace(0, 0); + } +#endif +#ifdef sqlite3Parser_ENGINEALWAYSONSTACK + pEngine = &sEngine; + sqlite3ParserInit(pEngine, pParse); +#else + pEngine = sqlite3ParserAlloc(sqlite3Malloc, pParse); + if( pEngine==0 ){ + sqlite3OomFault(db); + return SQLITE_NOMEM_BKPT; + } +#endif + assert( pParse->pNewTable==0 ); + assert( pParse->pNewTrigger==0 ); + assert( pParse->nVar==0 ); + assert( pParse->pVList==0 ); + pParse->pParentParse = db->pParse; + db->pParse = pParse; + while( 1 ){ + n = sqlite3GetToken((u8*)zSql, &tokenType); + mxSqlLen -= n; + if( mxSqlLen<0 ){ + pParse->rc = SQLITE_TOOBIG; + break; + } +#ifndef SQLITE_OMIT_WINDOWFUNC + if( tokenType>=TK_WINDOW ){ + assert( tokenType==TK_SPACE || tokenType==TK_OVER || tokenType==TK_FILTER + || tokenType==TK_ILLEGAL || tokenType==TK_WINDOW + ); +#else + if( tokenType>=TK_SPACE ){ + assert( tokenType==TK_SPACE || tokenType==TK_ILLEGAL ); +#endif /* SQLITE_OMIT_WINDOWFUNC */ + if( db->u1.isInterrupted ){ + pParse->rc = SQLITE_INTERRUPT; + break; + } + if( tokenType==TK_SPACE ){ + zSql += n; + continue; + } + if( zSql[0]==0 ){ + /* Upon reaching the end of input, call the parser two more times + ** with tokens TK_SEMI and 0, in that order. */ + if( lastTokenParsed==TK_SEMI ){ + tokenType = 0; + }else if( lastTokenParsed==0 ){ + break; + }else{ + tokenType = TK_SEMI; + } + n = 0; +#ifndef SQLITE_OMIT_WINDOWFUNC + }else if( tokenType==TK_WINDOW ){ + assert( n==6 ); + tokenType = analyzeWindowKeyword((const u8*)&zSql[6]); + }else if( tokenType==TK_OVER ){ + assert( n==4 ); + tokenType = analyzeOverKeyword((const u8*)&zSql[4], lastTokenParsed); + }else if( tokenType==TK_FILTER ){ + assert( n==6 ); + tokenType = analyzeFilterKeyword((const u8*)&zSql[6], lastTokenParsed); +#endif /* SQLITE_OMIT_WINDOWFUNC */ + }else{ + sqlite3ErrorMsg(pParse, "unrecognized token: \"%.*s\"", n, zSql); + break; + } + } + pParse->sLastToken.z = zSql; + pParse->sLastToken.n = n; + sqlite3Parser(pEngine, tokenType, pParse->sLastToken); + lastTokenParsed = tokenType; + zSql += n; + assert( db->mallocFailed==0 || pParse->rc!=SQLITE_OK || startedWithOom ); + if( pParse->rc!=SQLITE_OK ) break; + } + assert( nErr==0 ); +#ifdef YYTRACKMAXSTACKDEPTH + sqlite3_mutex_enter(sqlite3MallocMutex()); + sqlite3StatusHighwater(SQLITE_STATUS_PARSER_STACK, + sqlite3ParserStackPeak(pEngine) + ); + sqlite3_mutex_leave(sqlite3MallocMutex()); +#endif /* YYDEBUG */ +#ifdef sqlite3Parser_ENGINEALWAYSONSTACK + sqlite3ParserFinalize(pEngine); +#else + sqlite3ParserFree(pEngine, sqlite3_free); +#endif + if( db->mallocFailed ){ + pParse->rc = SQLITE_NOMEM_BKPT; + } + if( pParse->rc!=SQLITE_OK && pParse->rc!=SQLITE_DONE && pParse->zErrMsg==0 ){ + pParse->zErrMsg = sqlite3MPrintf(db, "%s", sqlite3ErrStr(pParse->rc)); + } + assert( pzErrMsg!=0 ); + if( pParse->zErrMsg ){ + *pzErrMsg = pParse->zErrMsg; + sqlite3_log(pParse->rc, "%s in \"%s\"", + *pzErrMsg, pParse->zTail); + pParse->zErrMsg = 0; + nErr++; + } + pParse->zTail = zSql; + if( pParse->pVdbe && pParse->nErr>0 && pParse->nested==0 ){ + sqlite3VdbeDelete(pParse->pVdbe); + pParse->pVdbe = 0; + } +#ifndef SQLITE_OMIT_SHARED_CACHE + if( pParse->nested==0 ){ + sqlite3DbFree(db, pParse->aTableLock); + pParse->aTableLock = 0; + pParse->nTableLock = 0; + } +#endif +#ifndef SQLITE_OMIT_VIRTUALTABLE + sqlite3_free(pParse->apVtabLock); +#endif + + if( !IN_SPECIAL_PARSE ){ + /* If the pParse->declareVtab flag is set, do not delete any table + ** structure built up in pParse->pNewTable. The calling code (see vtab.c) + ** will take responsibility for freeing the Table structure. + */ + sqlite3DeleteTable(db, pParse->pNewTable); + } + if( !IN_RENAME_OBJECT ){ + sqlite3DeleteTrigger(db, pParse->pNewTrigger); + } + + if( pParse->pWithToFree ) sqlite3WithDelete(db, pParse->pWithToFree); + sqlite3DbFree(db, pParse->pVList); + while( pParse->pAinc ){ + AutoincInfo *p = pParse->pAinc; + pParse->pAinc = p->pNext; + sqlite3DbFreeNN(db, p); + } + while( pParse->pZombieTab ){ + Table *p = pParse->pZombieTab; + pParse->pZombieTab = p->pNextZombie; + sqlite3DeleteTable(db, p); + } + db->pParse = pParse->pParentParse; + pParse->pParentParse = 0; + assert( nErr==0 || pParse->rc!=SQLITE_OK ); + return nErr; +} + + +#ifdef SQLITE_ENABLE_NORMALIZE +/* +** Insert a single space character into pStr if the current string +** ends with an identifier +*/ +static void addSpaceSeparator(sqlite3_str *pStr){ + if( pStr->nChar && sqlite3IsIdChar(pStr->zText[pStr->nChar-1]) ){ + sqlite3_str_append(pStr, " ", 1); + } +} + +/* +** Compute a normalization of the SQL given by zSql[0..nSql-1]. Return +** the normalization in space obtained from sqlite3DbMalloc(). Or return +** NULL if anything goes wrong or if zSql is NULL. +*/ +SQLITE_PRIVATE char *sqlite3Normalize( + Vdbe *pVdbe, /* VM being reprepared */ + const char *zSql /* The original SQL string */ +){ + sqlite3 *db; /* The database connection */ + int i; /* Next unread byte of zSql[] */ + int n; /* length of current token */ + int tokenType; /* type of current token */ + int prevType = 0; /* Previous non-whitespace token */ + int nParen; /* Number of nested levels of parentheses */ + int iStartIN; /* Start of RHS of IN operator in z[] */ + int nParenAtIN; /* Value of nParent at start of RHS of IN operator */ + int j; /* Bytes of normalized SQL generated so far */ + sqlite3_str *pStr; /* The normalized SQL string under construction */ + + db = sqlite3VdbeDb(pVdbe); + tokenType = -1; + nParen = iStartIN = nParenAtIN = 0; + pStr = sqlite3_str_new(db); + assert( pStr!=0 ); /* sqlite3_str_new() never returns NULL */ + for(i=0; zSql[i] && pStr->accError==0; i+=n){ + if( tokenType!=TK_SPACE ){ + prevType = tokenType; + } + n = sqlite3GetToken((unsigned char*)zSql+i, &tokenType); + if( NEVER(n<=0) ) break; + switch( tokenType ){ + case TK_SPACE: { + break; + } + case TK_NULL: { + if( prevType==TK_IS || prevType==TK_NOT ){ + sqlite3_str_append(pStr, " NULL", 5); + break; + } + /* Fall through */ + } + case TK_STRING: + case TK_INTEGER: + case TK_FLOAT: + case TK_VARIABLE: + case TK_BLOB: { + sqlite3_str_append(pStr, "?", 1); + break; + } + case TK_LP: { + nParen++; + if( prevType==TK_IN ){ + iStartIN = pStr->nChar; + nParenAtIN = nParen; + } + sqlite3_str_append(pStr, "(", 1); + break; + } + case TK_RP: { + if( iStartIN>0 && nParen==nParenAtIN ){ + assert( pStr->nChar>=iStartIN ); + pStr->nChar = iStartIN+1; + sqlite3_str_append(pStr, "?,?,?", 5); + iStartIN = 0; + } + nParen--; + sqlite3_str_append(pStr, ")", 1); + break; + } + case TK_ID: { + iStartIN = 0; + j = pStr->nChar; + if( sqlite3Isquote(zSql[i]) ){ + char *zId = sqlite3DbStrNDup(db, zSql+i, n); + int nId; + int eType = 0; + if( zId==0 ) break; + sqlite3Dequote(zId); + if( zSql[i]=='"' && sqlite3VdbeUsesDoubleQuotedString(pVdbe, zId) ){ + sqlite3_str_append(pStr, "?", 1); + sqlite3DbFree(db, zId); + break; + } + nId = sqlite3Strlen30(zId); + if( sqlite3GetToken((u8*)zId, &eType)==nId && eType==TK_ID ){ + addSpaceSeparator(pStr); + sqlite3_str_append(pStr, zId, nId); + }else{ + sqlite3_str_appendf(pStr, "\"%w\"", zId); + } + sqlite3DbFree(db, zId); + }else{ + addSpaceSeparator(pStr); + sqlite3_str_append(pStr, zSql+i, n); + } + while( jnChar ){ + pStr->zText[j] = sqlite3Tolower(pStr->zText[j]); + j++; + } + break; + } + case TK_SELECT: { + iStartIN = 0; + /* fall through */ + } + default: { + if( sqlite3IsIdChar(zSql[i]) ) addSpaceSeparator(pStr); + j = pStr->nChar; + sqlite3_str_append(pStr, zSql+i, n); + while( jnChar ){ + pStr->zText[j] = sqlite3Toupper(pStr->zText[j]); + j++; + } + break; + } + } + } + if( tokenType!=TK_SEMI ) sqlite3_str_append(pStr, ";", 1); + return sqlite3_str_finish(pStr); +} +#endif /* SQLITE_ENABLE_NORMALIZE */ + +/************** End of tokenize.c ********************************************/ +/************** Begin file complete.c ****************************************/ +/* +** 2001 September 15 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** An tokenizer for SQL +** +** This file contains C code that implements the sqlite3_complete() API. +** This code used to be part of the tokenizer.c source file. But by +** separating it out, the code will be automatically omitted from +** static links that do not use it. +*/ +/* #include "sqliteInt.h" */ +#ifndef SQLITE_OMIT_COMPLETE + +/* +** This is defined in tokenize.c. We just have to import the definition. +*/ +#ifndef SQLITE_AMALGAMATION +#ifdef SQLITE_ASCII +#define IdChar(C) ((sqlite3CtypeMap[(unsigned char)C]&0x46)!=0) +#endif +#ifdef SQLITE_EBCDIC +SQLITE_PRIVATE const char sqlite3IsEbcdicIdChar[]; +#define IdChar(C) (((c=C)>=0x42 && sqlite3IsEbcdicIdChar[c-0x40])) +#endif +#endif /* SQLITE_AMALGAMATION */ + + +/* +** Token types used by the sqlite3_complete() routine. See the header +** comments on that procedure for additional information. +*/ +#define tkSEMI 0 +#define tkWS 1 +#define tkOTHER 2 +#ifndef SQLITE_OMIT_TRIGGER +#define tkEXPLAIN 3 +#define tkCREATE 4 +#define tkTEMP 5 +#define tkTRIGGER 6 +#define tkEND 7 +#endif + +/* +** Return TRUE if the given SQL string ends in a semicolon. +** +** Special handling is require for CREATE TRIGGER statements. +** Whenever the CREATE TRIGGER keywords are seen, the statement +** must end with ";END;". +** +** This implementation uses a state machine with 8 states: +** +** (0) INVALID We have not yet seen a non-whitespace character. +** +** (1) START At the beginning or end of an SQL statement. This routine +** returns 1 if it ends in the START state and 0 if it ends +** in any other state. +** +** (2) NORMAL We are in the middle of statement which ends with a single +** semicolon. +** +** (3) EXPLAIN The keyword EXPLAIN has been seen at the beginning of +** a statement. +** +** (4) CREATE The keyword CREATE has been seen at the beginning of a +** statement, possibly preceded by EXPLAIN and/or followed by +** TEMP or TEMPORARY +** +** (5) TRIGGER We are in the middle of a trigger definition that must be +** ended by a semicolon, the keyword END, and another semicolon. +** +** (6) SEMI We've seen the first semicolon in the ";END;" that occurs at +** the end of a trigger definition. +** +** (7) END We've seen the ";END" of the ";END;" that occurs at the end +** of a trigger definition. +** +** Transitions between states above are determined by tokens extracted +** from the input. The following tokens are significant: +** +** (0) tkSEMI A semicolon. +** (1) tkWS Whitespace. +** (2) tkOTHER Any other SQL token. +** (3) tkEXPLAIN The "explain" keyword. +** (4) tkCREATE The "create" keyword. +** (5) tkTEMP The "temp" or "temporary" keyword. +** (6) tkTRIGGER The "trigger" keyword. +** (7) tkEND The "end" keyword. +** +** Whitespace never causes a state transition and is always ignored. +** This means that a SQL string of all whitespace is invalid. +** +** If we compile with SQLITE_OMIT_TRIGGER, all of the computation needed +** to recognize the end of a trigger can be omitted. All we have to do +** is look for a semicolon that is not part of an string or comment. +*/ +SQLITE_API int sqlite3_complete(const char *zSql){ + u8 state = 0; /* Current state, using numbers defined in header comment */ + u8 token; /* Value of the next token */ + +#ifndef SQLITE_OMIT_TRIGGER + /* A complex statement machine used to detect the end of a CREATE TRIGGER + ** statement. This is the normal case. + */ + static const u8 trans[8][8] = { + /* Token: */ + /* State: ** SEMI WS OTHER EXPLAIN CREATE TEMP TRIGGER END */ + /* 0 INVALID: */ { 1, 0, 2, 3, 4, 2, 2, 2, }, + /* 1 START: */ { 1, 1, 2, 3, 4, 2, 2, 2, }, + /* 2 NORMAL: */ { 1, 2, 2, 2, 2, 2, 2, 2, }, + /* 3 EXPLAIN: */ { 1, 3, 3, 2, 4, 2, 2, 2, }, + /* 4 CREATE: */ { 1, 4, 2, 2, 2, 4, 5, 2, }, + /* 5 TRIGGER: */ { 6, 5, 5, 5, 5, 5, 5, 5, }, + /* 6 SEMI: */ { 6, 6, 5, 5, 5, 5, 5, 7, }, + /* 7 END: */ { 1, 7, 5, 5, 5, 5, 5, 5, }, + }; +#else + /* If triggers are not supported by this compile then the statement machine + ** used to detect the end of a statement is much simpler + */ + static const u8 trans[3][3] = { + /* Token: */ + /* State: ** SEMI WS OTHER */ + /* 0 INVALID: */ { 1, 0, 2, }, + /* 1 START: */ { 1, 1, 2, }, + /* 2 NORMAL: */ { 1, 2, 2, }, + }; +#endif /* SQLITE_OMIT_TRIGGER */ + +#ifdef SQLITE_ENABLE_API_ARMOR + if( zSql==0 ){ + (void)SQLITE_MISUSE_BKPT; + return 0; + } +#endif + + while( *zSql ){ + switch( *zSql ){ + case ';': { /* A semicolon */ + token = tkSEMI; + break; + } + case ' ': + case '\r': + case '\t': + case '\n': + case '\f': { /* White space is ignored */ + token = tkWS; + break; + } + case '/': { /* C-style comments */ + if( zSql[1]!='*' ){ + token = tkOTHER; + break; + } + zSql += 2; + while( zSql[0] && (zSql[0]!='*' || zSql[1]!='/') ){ zSql++; } + if( zSql[0]==0 ) return 0; + zSql++; + token = tkWS; + break; + } + case '-': { /* SQL-style comments from "--" to end of line */ + if( zSql[1]!='-' ){ + token = tkOTHER; + break; + } + while( *zSql && *zSql!='\n' ){ zSql++; } + if( *zSql==0 ) return state==1; + token = tkWS; + break; + } + case '[': { /* Microsoft-style identifiers in [...] */ + zSql++; + while( *zSql && *zSql!=']' ){ zSql++; } + if( *zSql==0 ) return 0; + token = tkOTHER; + break; + } + case '`': /* Grave-accent quoted symbols used by MySQL */ + case '"': /* single- and double-quoted strings */ + case '\'': { + int c = *zSql; + zSql++; + while( *zSql && *zSql!=c ){ zSql++; } + if( *zSql==0 ) return 0; + token = tkOTHER; + break; + } + default: { +#ifdef SQLITE_EBCDIC + unsigned char c; +#endif + if( IdChar((u8)*zSql) ){ + /* Keywords and unquoted identifiers */ + int nId; + for(nId=1; IdChar(zSql[nId]); nId++){} +#ifdef SQLITE_OMIT_TRIGGER + token = tkOTHER; +#else + switch( *zSql ){ + case 'c': case 'C': { + if( nId==6 && sqlite3StrNICmp(zSql, "create", 6)==0 ){ + token = tkCREATE; + }else{ + token = tkOTHER; + } + break; + } + case 't': case 'T': { + if( nId==7 && sqlite3StrNICmp(zSql, "trigger", 7)==0 ){ + token = tkTRIGGER; + }else if( nId==4 && sqlite3StrNICmp(zSql, "temp", 4)==0 ){ + token = tkTEMP; + }else if( nId==9 && sqlite3StrNICmp(zSql, "temporary", 9)==0 ){ + token = tkTEMP; + }else{ + token = tkOTHER; + } + break; + } + case 'e': case 'E': { + if( nId==3 && sqlite3StrNICmp(zSql, "end", 3)==0 ){ + token = tkEND; + }else +#ifndef SQLITE_OMIT_EXPLAIN + if( nId==7 && sqlite3StrNICmp(zSql, "explain", 7)==0 ){ + token = tkEXPLAIN; + }else +#endif + { + token = tkOTHER; + } + break; + } + default: { + token = tkOTHER; + break; + } + } +#endif /* SQLITE_OMIT_TRIGGER */ + zSql += nId-1; + }else{ + /* Operators and special symbols */ + token = tkOTHER; + } + break; + } + } + state = trans[state][token]; + zSql++; + } + return state==1; +} + +#ifndef SQLITE_OMIT_UTF16 +/* +** This routine is the same as the sqlite3_complete() routine described +** above, except that the parameter is required to be UTF-16 encoded, not +** UTF-8. +*/ +SQLITE_API int sqlite3_complete16(const void *zSql){ + sqlite3_value *pVal; + char const *zSql8; + int rc; + +#ifndef SQLITE_OMIT_AUTOINIT + rc = sqlite3_initialize(); + if( rc ) return rc; +#endif + pVal = sqlite3ValueNew(0); + sqlite3ValueSetStr(pVal, -1, zSql, SQLITE_UTF16NATIVE, SQLITE_STATIC); + zSql8 = sqlite3ValueText(pVal, SQLITE_UTF8); + if( zSql8 ){ + rc = sqlite3_complete(zSql8); + }else{ + rc = SQLITE_NOMEM_BKPT; + } + sqlite3ValueFree(pVal); + return rc & 0xff; +} +#endif /* SQLITE_OMIT_UTF16 */ +#endif /* SQLITE_OMIT_COMPLETE */ + +/************** End of complete.c ********************************************/ +/************** Begin file main.c ********************************************/ +/* +** 2001 September 15 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** Main file for the SQLite library. The routines in this file +** implement the programmer interface to the library. Routines in +** other files are for internal use by SQLite and should not be +** accessed by users of the library. +*/ +/* #include "sqliteInt.h" */ + +#ifdef SQLITE_ENABLE_FTS3 +/************** Include fts3.h in the middle of main.c ***********************/ +/************** Begin file fts3.h ********************************************/ +/* +** 2006 Oct 10 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +****************************************************************************** +** +** This header file is used by programs that want to link against the +** FTS3 library. All it does is declare the sqlite3Fts3Init() interface. +*/ +/* #include "sqlite3.h" */ + +#if 0 +extern "C" { +#endif /* __cplusplus */ + +SQLITE_PRIVATE int sqlite3Fts3Init(sqlite3 *db); + +#if 0 +} /* extern "C" */ +#endif /* __cplusplus */ + +/************** End of fts3.h ************************************************/ +/************** Continuing where we left off in main.c ***********************/ +#endif +#ifdef SQLITE_ENABLE_RTREE +/************** Include rtree.h in the middle of main.c **********************/ +/************** Begin file rtree.h *******************************************/ +/* +** 2008 May 26 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +****************************************************************************** +** +** This header file is used by programs that want to link against the +** RTREE library. All it does is declare the sqlite3RtreeInit() interface. +*/ +/* #include "sqlite3.h" */ + +#ifdef SQLITE_OMIT_VIRTUALTABLE +# undef SQLITE_ENABLE_RTREE +#endif + +#if 0 +extern "C" { +#endif /* __cplusplus */ + +SQLITE_PRIVATE int sqlite3RtreeInit(sqlite3 *db); + +#if 0 +} /* extern "C" */ +#endif /* __cplusplus */ + +/************** End of rtree.h ***********************************************/ +/************** Continuing where we left off in main.c ***********************/ +#endif +#if defined(SQLITE_ENABLE_ICU) || defined(SQLITE_ENABLE_ICU_COLLATIONS) +/************** Include sqliteicu.h in the middle of main.c ******************/ +/************** Begin file sqliteicu.h ***************************************/ +/* +** 2008 May 26 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +****************************************************************************** +** +** This header file is used by programs that want to link against the +** ICU extension. All it does is declare the sqlite3IcuInit() interface. +*/ +/* #include "sqlite3.h" */ + +#if 0 +extern "C" { +#endif /* __cplusplus */ + +SQLITE_PRIVATE int sqlite3IcuInit(sqlite3 *db); + +#if 0 +} /* extern "C" */ +#endif /* __cplusplus */ + + +/************** End of sqliteicu.h *******************************************/ +/************** Continuing where we left off in main.c ***********************/ +#endif +#ifdef SQLITE_ENABLE_JSON1 +SQLITE_PRIVATE int sqlite3Json1Init(sqlite3*); +#endif +#ifdef SQLITE_ENABLE_STMTVTAB +SQLITE_PRIVATE int sqlite3StmtVtabInit(sqlite3*); +#endif +#ifdef SQLITE_ENABLE_FTS5 +SQLITE_PRIVATE int sqlite3Fts5Init(sqlite3*); +#endif + +#ifndef SQLITE_AMALGAMATION +/* IMPLEMENTATION-OF: R-46656-45156 The sqlite3_version[] string constant +** contains the text of SQLITE_VERSION macro. +*/ +SQLITE_API const char sqlite3_version[] = SQLITE_VERSION; +#endif + +/* IMPLEMENTATION-OF: R-53536-42575 The sqlite3_libversion() function returns +** a pointer to the to the sqlite3_version[] string constant. +*/ +SQLITE_API const char *sqlite3_libversion(void){ return sqlite3_version; } + +/* IMPLEMENTATION-OF: R-25063-23286 The sqlite3_sourceid() function returns a +** pointer to a string constant whose value is the same as the +** SQLITE_SOURCE_ID C preprocessor macro. Except if SQLite is built using +** an edited copy of the amalgamation, then the last four characters of +** the hash might be different from SQLITE_SOURCE_ID. +*/ +/* SQLITE_API const char *sqlite3_sourceid(void){ return SQLITE_SOURCE_ID; } */ + +/* IMPLEMENTATION-OF: R-35210-63508 The sqlite3_libversion_number() function +** returns an integer equal to SQLITE_VERSION_NUMBER. +*/ +SQLITE_API int sqlite3_libversion_number(void){ return SQLITE_VERSION_NUMBER; } + +/* IMPLEMENTATION-OF: R-20790-14025 The sqlite3_threadsafe() function returns +** zero if and only if SQLite was compiled with mutexing code omitted due to +** the SQLITE_THREADSAFE compile-time option being set to 0. +*/ +SQLITE_API int sqlite3_threadsafe(void){ return SQLITE_THREADSAFE; } + +/* +** When compiling the test fixture or with debugging enabled (on Win32), +** this variable being set to non-zero will cause OSTRACE macros to emit +** extra diagnostic information. +*/ +#ifdef SQLITE_HAVE_OS_TRACE +# ifndef SQLITE_DEBUG_OS_TRACE +# define SQLITE_DEBUG_OS_TRACE 0 +# endif + int sqlite3OSTrace = SQLITE_DEBUG_OS_TRACE; +#endif + +#if !defined(SQLITE_OMIT_TRACE) && defined(SQLITE_ENABLE_IOTRACE) +/* +** If the following function pointer is not NULL and if +** SQLITE_ENABLE_IOTRACE is enabled, then messages describing +** I/O active are written using this function. These messages +** are intended for debugging activity only. +*/ +SQLITE_API void (SQLITE_CDECL *sqlite3IoTrace)(const char*, ...) = 0; +#endif + +/* +** If the following global variable points to a string which is the +** name of a directory, then that directory will be used to store +** temporary files. +** +** See also the "PRAGMA temp_store_directory" SQL command. +*/ +SQLITE_API char *sqlite3_temp_directory = 0; + +/* +** If the following global variable points to a string which is the +** name of a directory, then that directory will be used to store +** all database files specified with a relative pathname. +** +** See also the "PRAGMA data_store_directory" SQL command. +*/ +SQLITE_API char *sqlite3_data_directory = 0; + +/* +** Initialize SQLite. +** +** This routine must be called to initialize the memory allocation, +** VFS, and mutex subsystems prior to doing any serious work with +** SQLite. But as long as you do not compile with SQLITE_OMIT_AUTOINIT +** this routine will be called automatically by key routines such as +** sqlite3_open(). +** +** This routine is a no-op except on its very first call for the process, +** or for the first call after a call to sqlite3_shutdown. +** +** The first thread to call this routine runs the initialization to +** completion. If subsequent threads call this routine before the first +** thread has finished the initialization process, then the subsequent +** threads must block until the first thread finishes with the initialization. +** +** The first thread might call this routine recursively. Recursive +** calls to this routine should not block, of course. Otherwise the +** initialization process would never complete. +** +** Let X be the first thread to enter this routine. Let Y be some other +** thread. Then while the initial invocation of this routine by X is +** incomplete, it is required that: +** +** * Calls to this routine from Y must block until the outer-most +** call by X completes. +** +** * Recursive calls to this routine from thread X return immediately +** without blocking. +*/ +SQLITE_API int sqlite3_initialize(void){ + MUTEX_LOGIC( sqlite3_mutex *pMaster; ) /* The main static mutex */ + int rc; /* Result code */ +#ifdef SQLITE_EXTRA_INIT + int bRunExtraInit = 0; /* Extra initialization needed */ +#endif + +#ifdef SQLITE_OMIT_WSD + rc = sqlite3_wsd_init(4096, 24); + if( rc!=SQLITE_OK ){ + return rc; + } +#endif + + /* If the following assert() fails on some obscure processor/compiler + ** combination, the work-around is to set the correct pointer + ** size at compile-time using -DSQLITE_PTRSIZE=n compile-time option */ + assert( SQLITE_PTRSIZE==sizeof(char*) ); + + /* If SQLite is already completely initialized, then this call + ** to sqlite3_initialize() should be a no-op. But the initialization + ** must be complete. So isInit must not be set until the very end + ** of this routine. + */ + if( sqlite3GlobalConfig.isInit ) return SQLITE_OK; + + /* Make sure the mutex subsystem is initialized. If unable to + ** initialize the mutex subsystem, return early with the error. + ** If the system is so sick that we are unable to allocate a mutex, + ** there is not much SQLite is going to be able to do. + ** + ** The mutex subsystem must take care of serializing its own + ** initialization. + */ + rc = sqlite3MutexInit(); + if( rc ) return rc; + + /* Initialize the malloc() system and the recursive pInitMutex mutex. + ** This operation is protected by the STATIC_MASTER mutex. Note that + ** MutexAlloc() is called for a static mutex prior to initializing the + ** malloc subsystem - this implies that the allocation of a static + ** mutex must not require support from the malloc subsystem. + */ + MUTEX_LOGIC( pMaster = sqlite3MutexAlloc(SQLITE_MUTEX_STATIC_MASTER); ) + sqlite3_mutex_enter(pMaster); + sqlite3GlobalConfig.isMutexInit = 1; + if( !sqlite3GlobalConfig.isMallocInit ){ + rc = sqlite3MallocInit(); + } + if( rc==SQLITE_OK ){ + sqlite3GlobalConfig.isMallocInit = 1; + if( !sqlite3GlobalConfig.pInitMutex ){ + sqlite3GlobalConfig.pInitMutex = + sqlite3MutexAlloc(SQLITE_MUTEX_RECURSIVE); + if( sqlite3GlobalConfig.bCoreMutex && !sqlite3GlobalConfig.pInitMutex ){ + rc = SQLITE_NOMEM_BKPT; + } + } + } + if( rc==SQLITE_OK ){ + sqlite3GlobalConfig.nRefInitMutex++; + } + sqlite3_mutex_leave(pMaster); + + /* If rc is not SQLITE_OK at this point, then either the malloc + ** subsystem could not be initialized or the system failed to allocate + ** the pInitMutex mutex. Return an error in either case. */ + if( rc!=SQLITE_OK ){ + return rc; + } + + /* Do the rest of the initialization under the recursive mutex so + ** that we will be able to handle recursive calls into + ** sqlite3_initialize(). The recursive calls normally come through + ** sqlite3_os_init() when it invokes sqlite3_vfs_register(), but other + ** recursive calls might also be possible. + ** + ** IMPLEMENTATION-OF: R-00140-37445 SQLite automatically serializes calls + ** to the xInit method, so the xInit method need not be threadsafe. + ** + ** The following mutex is what serializes access to the appdef pcache xInit + ** methods. The sqlite3_pcache_methods.xInit() all is embedded in the + ** call to sqlite3PcacheInitialize(). + */ + sqlite3_mutex_enter(sqlite3GlobalConfig.pInitMutex); + if( sqlite3GlobalConfig.isInit==0 && sqlite3GlobalConfig.inProgress==0 ){ + sqlite3GlobalConfig.inProgress = 1; +#ifdef SQLITE_ENABLE_SQLLOG + { + extern void sqlite3_init_sqllog(void); + sqlite3_init_sqllog(); + } +#endif + memset(&sqlite3BuiltinFunctions, 0, sizeof(sqlite3BuiltinFunctions)); + sqlite3RegisterBuiltinFunctions(); + if( sqlite3GlobalConfig.isPCacheInit==0 ){ + rc = sqlite3PcacheInitialize(); + } + if( rc==SQLITE_OK ){ + sqlite3GlobalConfig.isPCacheInit = 1; + rc = sqlite3OsInit(); + } +#ifdef SQLITE_ENABLE_DESERIALIZE + if( rc==SQLITE_OK ){ + rc = sqlite3MemdbInit(); + } +#endif + if( rc==SQLITE_OK ){ + sqlite3PCacheBufferSetup( sqlite3GlobalConfig.pPage, + sqlite3GlobalConfig.szPage, sqlite3GlobalConfig.nPage); + sqlite3GlobalConfig.isInit = 1; +#ifdef SQLITE_EXTRA_INIT + bRunExtraInit = 1; +#endif + } + sqlite3GlobalConfig.inProgress = 0; + } + sqlite3_mutex_leave(sqlite3GlobalConfig.pInitMutex); + + /* Go back under the static mutex and clean up the recursive + ** mutex to prevent a resource leak. + */ + sqlite3_mutex_enter(pMaster); + sqlite3GlobalConfig.nRefInitMutex--; + if( sqlite3GlobalConfig.nRefInitMutex<=0 ){ + assert( sqlite3GlobalConfig.nRefInitMutex==0 ); + sqlite3_mutex_free(sqlite3GlobalConfig.pInitMutex); + sqlite3GlobalConfig.pInitMutex = 0; + } + sqlite3_mutex_leave(pMaster); + + /* The following is just a sanity check to make sure SQLite has + ** been compiled correctly. It is important to run this code, but + ** we don't want to run it too often and soak up CPU cycles for no + ** reason. So we run it once during initialization. + */ +#ifndef NDEBUG +#ifndef SQLITE_OMIT_FLOATING_POINT + /* This section of code's only "output" is via assert() statements. */ + if( rc==SQLITE_OK ){ + u64 x = (((u64)1)<<63)-1; + double y; + assert(sizeof(x)==8); + assert(sizeof(x)==sizeof(y)); + memcpy(&y, &x, 8); + assert( sqlite3IsNaN(y) ); + } +#endif +#endif + + /* Do extra initialization steps requested by the SQLITE_EXTRA_INIT + ** compile-time option. + */ +#ifdef SQLITE_EXTRA_INIT + if( bRunExtraInit ){ + int SQLITE_EXTRA_INIT(const char*); + rc = SQLITE_EXTRA_INIT(0); + } +#endif + + return rc; +} + +/* +** Undo the effects of sqlite3_initialize(). Must not be called while +** there are outstanding database connections or memory allocations or +** while any part of SQLite is otherwise in use in any thread. This +** routine is not threadsafe. But it is safe to invoke this routine +** on when SQLite is already shut down. If SQLite is already shut down +** when this routine is invoked, then this routine is a harmless no-op. +*/ +SQLITE_API int sqlite3_shutdown(void){ +#ifdef SQLITE_OMIT_WSD + int rc = sqlite3_wsd_init(4096, 24); + if( rc!=SQLITE_OK ){ + return rc; + } +#endif + + if( sqlite3GlobalConfig.isInit ){ +#ifdef SQLITE_EXTRA_SHUTDOWN + void SQLITE_EXTRA_SHUTDOWN(void); + SQLITE_EXTRA_SHUTDOWN(); +#endif + sqlite3_os_end(); + sqlite3_reset_auto_extension(); + sqlite3GlobalConfig.isInit = 0; + } + if( sqlite3GlobalConfig.isPCacheInit ){ + sqlite3PcacheShutdown(); + sqlite3GlobalConfig.isPCacheInit = 0; + } + if( sqlite3GlobalConfig.isMallocInit ){ + sqlite3MallocEnd(); + sqlite3GlobalConfig.isMallocInit = 0; + +#ifndef SQLITE_OMIT_SHUTDOWN_DIRECTORIES + /* The heap subsystem has now been shutdown and these values are supposed + ** to be NULL or point to memory that was obtained from sqlite3_malloc(), + ** which would rely on that heap subsystem; therefore, make sure these + ** values cannot refer to heap memory that was just invalidated when the + ** heap subsystem was shutdown. This is only done if the current call to + ** this function resulted in the heap subsystem actually being shutdown. + */ + sqlite3_data_directory = 0; + sqlite3_temp_directory = 0; +#endif + } + if( sqlite3GlobalConfig.isMutexInit ){ + sqlite3MutexEnd(); + sqlite3GlobalConfig.isMutexInit = 0; + } + + return SQLITE_OK; +} + +/* +** This API allows applications to modify the global configuration of +** the SQLite library at run-time. +** +** This routine should only be called when there are no outstanding +** database connections or memory allocations. This routine is not +** threadsafe. Failure to heed these warnings can lead to unpredictable +** behavior. +*/ +SQLITE_API int sqlite3_config(int op, ...){ + va_list ap; + int rc = SQLITE_OK; + + /* sqlite3_config() shall return SQLITE_MISUSE if it is invoked while + ** the SQLite library is in use. */ + if( sqlite3GlobalConfig.isInit ) return SQLITE_MISUSE_BKPT; + + va_start(ap, op); + switch( op ){ + + /* Mutex configuration options are only available in a threadsafe + ** compile. + */ +#if defined(SQLITE_THREADSAFE) && SQLITE_THREADSAFE>0 /* IMP: R-54466-46756 */ + case SQLITE_CONFIG_SINGLETHREAD: { + /* EVIDENCE-OF: R-02748-19096 This option sets the threading mode to + ** Single-thread. */ + sqlite3GlobalConfig.bCoreMutex = 0; /* Disable mutex on core */ + sqlite3GlobalConfig.bFullMutex = 0; /* Disable mutex on connections */ + break; + } +#endif +#if defined(SQLITE_THREADSAFE) && SQLITE_THREADSAFE>0 /* IMP: R-20520-54086 */ + case SQLITE_CONFIG_MULTITHREAD: { + /* EVIDENCE-OF: R-14374-42468 This option sets the threading mode to + ** Multi-thread. */ + sqlite3GlobalConfig.bCoreMutex = 1; /* Enable mutex on core */ + sqlite3GlobalConfig.bFullMutex = 0; /* Disable mutex on connections */ + break; + } +#endif +#if defined(SQLITE_THREADSAFE) && SQLITE_THREADSAFE>0 /* IMP: R-59593-21810 */ + case SQLITE_CONFIG_SERIALIZED: { + /* EVIDENCE-OF: R-41220-51800 This option sets the threading mode to + ** Serialized. */ + sqlite3GlobalConfig.bCoreMutex = 1; /* Enable mutex on core */ + sqlite3GlobalConfig.bFullMutex = 1; /* Enable mutex on connections */ + break; + } +#endif +#if defined(SQLITE_THREADSAFE) && SQLITE_THREADSAFE>0 /* IMP: R-63666-48755 */ + case SQLITE_CONFIG_MUTEX: { + /* Specify an alternative mutex implementation */ + sqlite3GlobalConfig.mutex = *va_arg(ap, sqlite3_mutex_methods*); + break; + } +#endif +#if defined(SQLITE_THREADSAFE) && SQLITE_THREADSAFE>0 /* IMP: R-14450-37597 */ + case SQLITE_CONFIG_GETMUTEX: { + /* Retrieve the current mutex implementation */ + *va_arg(ap, sqlite3_mutex_methods*) = sqlite3GlobalConfig.mutex; + break; + } +#endif + + case SQLITE_CONFIG_MALLOC: { + /* EVIDENCE-OF: R-55594-21030 The SQLITE_CONFIG_MALLOC option takes a + ** single argument which is a pointer to an instance of the + ** sqlite3_mem_methods structure. The argument specifies alternative + ** low-level memory allocation routines to be used in place of the memory + ** allocation routines built into SQLite. */ + sqlite3GlobalConfig.m = *va_arg(ap, sqlite3_mem_methods*); + break; + } + case SQLITE_CONFIG_GETMALLOC: { + /* EVIDENCE-OF: R-51213-46414 The SQLITE_CONFIG_GETMALLOC option takes a + ** single argument which is a pointer to an instance of the + ** sqlite3_mem_methods structure. The sqlite3_mem_methods structure is + ** filled with the currently defined memory allocation routines. */ + if( sqlite3GlobalConfig.m.xMalloc==0 ) sqlite3MemSetDefault(); + *va_arg(ap, sqlite3_mem_methods*) = sqlite3GlobalConfig.m; + break; + } + case SQLITE_CONFIG_MEMSTATUS: { + /* EVIDENCE-OF: R-61275-35157 The SQLITE_CONFIG_MEMSTATUS option takes + ** single argument of type int, interpreted as a boolean, which enables + ** or disables the collection of memory allocation statistics. */ + sqlite3GlobalConfig.bMemstat = va_arg(ap, int); + break; + } + case SQLITE_CONFIG_SMALL_MALLOC: { + sqlite3GlobalConfig.bSmallMalloc = va_arg(ap, int); + break; + } + case SQLITE_CONFIG_PAGECACHE: { + /* EVIDENCE-OF: R-18761-36601 There are three arguments to + ** SQLITE_CONFIG_PAGECACHE: A pointer to 8-byte aligned memory (pMem), + ** the size of each page cache line (sz), and the number of cache lines + ** (N). */ + sqlite3GlobalConfig.pPage = va_arg(ap, void*); + sqlite3GlobalConfig.szPage = va_arg(ap, int); + sqlite3GlobalConfig.nPage = va_arg(ap, int); + break; + } + case SQLITE_CONFIG_PCACHE_HDRSZ: { + /* EVIDENCE-OF: R-39100-27317 The SQLITE_CONFIG_PCACHE_HDRSZ option takes + ** a single parameter which is a pointer to an integer and writes into + ** that integer the number of extra bytes per page required for each page + ** in SQLITE_CONFIG_PAGECACHE. */ + *va_arg(ap, int*) = + sqlite3HeaderSizeBtree() + + sqlite3HeaderSizePcache() + + sqlite3HeaderSizePcache1(); + break; + } + + case SQLITE_CONFIG_PCACHE: { + /* no-op */ + break; + } + case SQLITE_CONFIG_GETPCACHE: { + /* now an error */ + rc = SQLITE_ERROR; + break; + } + + case SQLITE_CONFIG_PCACHE2: { + /* EVIDENCE-OF: R-63325-48378 The SQLITE_CONFIG_PCACHE2 option takes a + ** single argument which is a pointer to an sqlite3_pcache_methods2 + ** object. This object specifies the interface to a custom page cache + ** implementation. */ + sqlite3GlobalConfig.pcache2 = *va_arg(ap, sqlite3_pcache_methods2*); + break; + } + case SQLITE_CONFIG_GETPCACHE2: { + /* EVIDENCE-OF: R-22035-46182 The SQLITE_CONFIG_GETPCACHE2 option takes a + ** single argument which is a pointer to an sqlite3_pcache_methods2 + ** object. SQLite copies of the current page cache implementation into + ** that object. */ + if( sqlite3GlobalConfig.pcache2.xInit==0 ){ + sqlite3PCacheSetDefault(); + } + *va_arg(ap, sqlite3_pcache_methods2*) = sqlite3GlobalConfig.pcache2; + break; + } + +/* EVIDENCE-OF: R-06626-12911 The SQLITE_CONFIG_HEAP option is only +** available if SQLite is compiled with either SQLITE_ENABLE_MEMSYS3 or +** SQLITE_ENABLE_MEMSYS5 and returns SQLITE_ERROR if invoked otherwise. */ +#if defined(SQLITE_ENABLE_MEMSYS3) || defined(SQLITE_ENABLE_MEMSYS5) + case SQLITE_CONFIG_HEAP: { + /* EVIDENCE-OF: R-19854-42126 There are three arguments to + ** SQLITE_CONFIG_HEAP: An 8-byte aligned pointer to the memory, the + ** number of bytes in the memory buffer, and the minimum allocation size. + */ + sqlite3GlobalConfig.pHeap = va_arg(ap, void*); + sqlite3GlobalConfig.nHeap = va_arg(ap, int); + sqlite3GlobalConfig.mnReq = va_arg(ap, int); + + if( sqlite3GlobalConfig.mnReq<1 ){ + sqlite3GlobalConfig.mnReq = 1; + }else if( sqlite3GlobalConfig.mnReq>(1<<12) ){ + /* cap min request size at 2^12 */ + sqlite3GlobalConfig.mnReq = (1<<12); + } + + if( sqlite3GlobalConfig.pHeap==0 ){ + /* EVIDENCE-OF: R-49920-60189 If the first pointer (the memory pointer) + ** is NULL, then SQLite reverts to using its default memory allocator + ** (the system malloc() implementation), undoing any prior invocation of + ** SQLITE_CONFIG_MALLOC. + ** + ** Setting sqlite3GlobalConfig.m to all zeros will cause malloc to + ** revert to its default implementation when sqlite3_initialize() is run + */ + memset(&sqlite3GlobalConfig.m, 0, sizeof(sqlite3GlobalConfig.m)); + }else{ + /* EVIDENCE-OF: R-61006-08918 If the memory pointer is not NULL then the + ** alternative memory allocator is engaged to handle all of SQLites + ** memory allocation needs. */ +#ifdef SQLITE_ENABLE_MEMSYS3 + sqlite3GlobalConfig.m = *sqlite3MemGetMemsys3(); +#endif +#ifdef SQLITE_ENABLE_MEMSYS5 + sqlite3GlobalConfig.m = *sqlite3MemGetMemsys5(); +#endif + } + break; + } +#endif + + case SQLITE_CONFIG_LOOKASIDE: { + sqlite3GlobalConfig.szLookaside = va_arg(ap, int); + sqlite3GlobalConfig.nLookaside = va_arg(ap, int); + break; + } + + /* Record a pointer to the logger function and its first argument. + ** The default is NULL. Logging is disabled if the function pointer is + ** NULL. + */ + case SQLITE_CONFIG_LOG: { + /* MSVC is picky about pulling func ptrs from va lists. + ** http://support.microsoft.com/kb/47961 + ** sqlite3GlobalConfig.xLog = va_arg(ap, void(*)(void*,int,const char*)); + */ + typedef void(*LOGFUNC_t)(void*,int,const char*); + sqlite3GlobalConfig.xLog = va_arg(ap, LOGFUNC_t); + sqlite3GlobalConfig.pLogArg = va_arg(ap, void*); + break; + } + + /* EVIDENCE-OF: R-55548-33817 The compile-time setting for URI filenames + ** can be changed at start-time using the + ** sqlite3_config(SQLITE_CONFIG_URI,1) or + ** sqlite3_config(SQLITE_CONFIG_URI,0) configuration calls. + */ + case SQLITE_CONFIG_URI: { + /* EVIDENCE-OF: R-25451-61125 The SQLITE_CONFIG_URI option takes a single + ** argument of type int. If non-zero, then URI handling is globally + ** enabled. If the parameter is zero, then URI handling is globally + ** disabled. */ + sqlite3GlobalConfig.bOpenUri = va_arg(ap, int); + break; + } + + case SQLITE_CONFIG_COVERING_INDEX_SCAN: { + /* EVIDENCE-OF: R-36592-02772 The SQLITE_CONFIG_COVERING_INDEX_SCAN + ** option takes a single integer argument which is interpreted as a + ** boolean in order to enable or disable the use of covering indices for + ** full table scans in the query optimizer. */ + sqlite3GlobalConfig.bUseCis = va_arg(ap, int); + break; + } + +#ifdef SQLITE_ENABLE_SQLLOG + case SQLITE_CONFIG_SQLLOG: { + typedef void(*SQLLOGFUNC_t)(void*, sqlite3*, const char*, int); + sqlite3GlobalConfig.xSqllog = va_arg(ap, SQLLOGFUNC_t); + sqlite3GlobalConfig.pSqllogArg = va_arg(ap, void *); + break; + } +#endif + + case SQLITE_CONFIG_MMAP_SIZE: { + /* EVIDENCE-OF: R-58063-38258 SQLITE_CONFIG_MMAP_SIZE takes two 64-bit + ** integer (sqlite3_int64) values that are the default mmap size limit + ** (the default setting for PRAGMA mmap_size) and the maximum allowed + ** mmap size limit. */ + sqlite3_int64 szMmap = va_arg(ap, sqlite3_int64); + sqlite3_int64 mxMmap = va_arg(ap, sqlite3_int64); + /* EVIDENCE-OF: R-53367-43190 If either argument to this option is + ** negative, then that argument is changed to its compile-time default. + ** + ** EVIDENCE-OF: R-34993-45031 The maximum allowed mmap size will be + ** silently truncated if necessary so that it does not exceed the + ** compile-time maximum mmap size set by the SQLITE_MAX_MMAP_SIZE + ** compile-time option. + */ + if( mxMmap<0 || mxMmap>SQLITE_MAX_MMAP_SIZE ){ + mxMmap = SQLITE_MAX_MMAP_SIZE; + } + if( szMmap<0 ) szMmap = SQLITE_DEFAULT_MMAP_SIZE; + if( szMmap>mxMmap) szMmap = mxMmap; + sqlite3GlobalConfig.mxMmap = mxMmap; + sqlite3GlobalConfig.szMmap = szMmap; + break; + } + +#if SQLITE_OS_WIN && defined(SQLITE_WIN32_MALLOC) /* IMP: R-04780-55815 */ + case SQLITE_CONFIG_WIN32_HEAPSIZE: { + /* EVIDENCE-OF: R-34926-03360 SQLITE_CONFIG_WIN32_HEAPSIZE takes a 32-bit + ** unsigned integer value that specifies the maximum size of the created + ** heap. */ + sqlite3GlobalConfig.nHeap = va_arg(ap, int); + break; + } +#endif + + case SQLITE_CONFIG_PMASZ: { + sqlite3GlobalConfig.szPma = va_arg(ap, unsigned int); + break; + } + + case SQLITE_CONFIG_STMTJRNL_SPILL: { + sqlite3GlobalConfig.nStmtSpill = va_arg(ap, int); + break; + } + +#ifdef SQLITE_ENABLE_SORTER_REFERENCES + case SQLITE_CONFIG_SORTERREF_SIZE: { + int iVal = va_arg(ap, int); + if( iVal<0 ){ + iVal = SQLITE_DEFAULT_SORTERREF_SIZE; + } + sqlite3GlobalConfig.szSorterRef = (u32)iVal; + break; + } +#endif /* SQLITE_ENABLE_SORTER_REFERENCES */ + +#ifdef SQLITE_ENABLE_DESERIALIZE + case SQLITE_CONFIG_MEMDB_MAXSIZE: { + sqlite3GlobalConfig.mxMemdbSize = va_arg(ap, sqlite3_int64); + break; + } +#endif /* SQLITE_ENABLE_DESERIALIZE */ + + default: { + rc = SQLITE_ERROR; + break; + } + } + va_end(ap); + return rc; +} + +/* +** Set up the lookaside buffers for a database connection. +** Return SQLITE_OK on success. +** If lookaside is already active, return SQLITE_BUSY. +** +** The sz parameter is the number of bytes in each lookaside slot. +** The cnt parameter is the number of slots. If pStart is NULL the +** space for the lookaside memory is obtained from sqlite3_malloc(). +** If pStart is not NULL then it is sz*cnt bytes of memory to use for +** the lookaside memory. +*/ +static int setupLookaside(sqlite3 *db, void *pBuf, int sz, int cnt){ +#ifndef SQLITE_OMIT_LOOKASIDE + void *pStart; + + if( sqlite3LookasideUsed(db,0)>0 ){ + return SQLITE_BUSY; + } + /* Free any existing lookaside buffer for this handle before + ** allocating a new one so we don't have to have space for + ** both at the same time. + */ + if( db->lookaside.bMalloced ){ + sqlite3_free(db->lookaside.pStart); + } + /* The size of a lookaside slot after ROUNDDOWN8 needs to be larger + ** than a pointer to be useful. + */ + sz = ROUNDDOWN8(sz); /* IMP: R-33038-09382 */ + if( sz<=(int)sizeof(LookasideSlot*) ) sz = 0; + if( cnt<0 ) cnt = 0; + if( sz==0 || cnt==0 ){ + sz = 0; + pStart = 0; + }else if( pBuf==0 ){ + sqlite3BeginBenignMalloc(); + pStart = sqlite3Malloc( sz*(sqlite3_int64)cnt ); /* IMP: R-61949-35727 */ + sqlite3EndBenignMalloc(); + if( pStart ) cnt = sqlite3MallocSize(pStart)/sz; + }else{ + pStart = pBuf; + } + db->lookaside.pStart = pStart; + db->lookaside.pInit = 0; + db->lookaside.pFree = 0; + db->lookaside.sz = (u16)sz; + if( pStart ){ + int i; + LookasideSlot *p; + assert( sz > (int)sizeof(LookasideSlot*) ); + db->lookaside.nSlot = cnt; + p = (LookasideSlot*)pStart; + for(i=cnt-1; i>=0; i--){ + p->pNext = db->lookaside.pInit; + db->lookaside.pInit = p; + p = (LookasideSlot*)&((u8*)p)[sz]; + } + db->lookaside.pEnd = p; + db->lookaside.bDisable = 0; + db->lookaside.bMalloced = pBuf==0 ?1:0; + }else{ + db->lookaside.pStart = db; + db->lookaside.pEnd = db; + db->lookaside.bDisable = 1; + db->lookaside.bMalloced = 0; + db->lookaside.nSlot = 0; + } +#endif /* SQLITE_OMIT_LOOKASIDE */ + return SQLITE_OK; +} + +/* +** Return the mutex associated with a database connection. +*/ +SQLITE_API sqlite3_mutex *sqlite3_db_mutex(sqlite3 *db){ +#ifdef SQLITE_ENABLE_API_ARMOR + if( !sqlite3SafetyCheckOk(db) ){ + (void)SQLITE_MISUSE_BKPT; + return 0; + } +#endif + return db->mutex; +} + +/* +** Free up as much memory as we can from the given database +** connection. +*/ +SQLITE_API int sqlite3_db_release_memory(sqlite3 *db){ + int i; + +#ifdef SQLITE_ENABLE_API_ARMOR + if( !sqlite3SafetyCheckOk(db) ) return SQLITE_MISUSE_BKPT; +#endif + sqlite3_mutex_enter(db->mutex); + sqlite3BtreeEnterAll(db); + for(i=0; inDb; i++){ + Btree *pBt = db->aDb[i].pBt; + if( pBt ){ + Pager *pPager = sqlite3BtreePager(pBt); + sqlite3PagerShrink(pPager); + } + } + sqlite3BtreeLeaveAll(db); + sqlite3_mutex_leave(db->mutex); + return SQLITE_OK; +} + +/* +** Flush any dirty pages in the pager-cache for any attached database +** to disk. +*/ +SQLITE_API int sqlite3_db_cacheflush(sqlite3 *db){ + int i; + int rc = SQLITE_OK; + int bSeenBusy = 0; + +#ifdef SQLITE_ENABLE_API_ARMOR + if( !sqlite3SafetyCheckOk(db) ) return SQLITE_MISUSE_BKPT; +#endif + sqlite3_mutex_enter(db->mutex); + sqlite3BtreeEnterAll(db); + for(i=0; rc==SQLITE_OK && inDb; i++){ + Btree *pBt = db->aDb[i].pBt; + if( pBt && sqlite3BtreeIsInTrans(pBt) ){ + Pager *pPager = sqlite3BtreePager(pBt); + rc = sqlite3PagerFlush(pPager); + if( rc==SQLITE_BUSY ){ + bSeenBusy = 1; + rc = SQLITE_OK; + } + } + } + sqlite3BtreeLeaveAll(db); + sqlite3_mutex_leave(db->mutex); + return ((rc==SQLITE_OK && bSeenBusy) ? SQLITE_BUSY : rc); +} + +/* +** Configuration settings for an individual database connection +*/ +SQLITE_API int sqlite3_db_config(sqlite3 *db, int op, ...){ + va_list ap; + int rc; + va_start(ap, op); + switch( op ){ + case SQLITE_DBCONFIG_MAINDBNAME: { + /* IMP: R-06824-28531 */ + /* IMP: R-36257-52125 */ + db->aDb[0].zDbSName = va_arg(ap,char*); + rc = SQLITE_OK; + break; + } + case SQLITE_DBCONFIG_LOOKASIDE: { + void *pBuf = va_arg(ap, void*); /* IMP: R-26835-10964 */ + int sz = va_arg(ap, int); /* IMP: R-47871-25994 */ + int cnt = va_arg(ap, int); /* IMP: R-04460-53386 */ + rc = setupLookaside(db, pBuf, sz, cnt); + break; + } + default: { + static const struct { + int op; /* The opcode */ + u32 mask; /* Mask of the bit in sqlite3.flags to set/clear */ + } aFlagOp[] = { + { SQLITE_DBCONFIG_ENABLE_FKEY, SQLITE_ForeignKeys }, + { SQLITE_DBCONFIG_ENABLE_TRIGGER, SQLITE_EnableTrigger }, + { SQLITE_DBCONFIG_ENABLE_VIEW, SQLITE_EnableView }, + { SQLITE_DBCONFIG_ENABLE_FTS3_TOKENIZER, SQLITE_Fts3Tokenizer }, + { SQLITE_DBCONFIG_ENABLE_LOAD_EXTENSION, SQLITE_LoadExtension }, + { SQLITE_DBCONFIG_NO_CKPT_ON_CLOSE, SQLITE_NoCkptOnClose }, + { SQLITE_DBCONFIG_ENABLE_QPSG, SQLITE_EnableQPSG }, + { SQLITE_DBCONFIG_TRIGGER_EQP, SQLITE_TriggerEQP }, + { SQLITE_DBCONFIG_RESET_DATABASE, SQLITE_ResetDatabase }, + { SQLITE_DBCONFIG_DEFENSIVE, SQLITE_Defensive }, + { SQLITE_DBCONFIG_WRITABLE_SCHEMA, SQLITE_WriteSchema| + SQLITE_NoSchemaError }, + { SQLITE_DBCONFIG_LEGACY_ALTER_TABLE, SQLITE_LegacyAlter }, + { SQLITE_DBCONFIG_DQS_DDL, SQLITE_DqsDDL }, + { SQLITE_DBCONFIG_DQS_DML, SQLITE_DqsDML }, + }; + unsigned int i; + rc = SQLITE_ERROR; /* IMP: R-42790-23372 */ + for(i=0; iflags; + if( onoff>0 ){ + db->flags |= aFlagOp[i].mask; + }else if( onoff==0 ){ + db->flags &= ~(u64)aFlagOp[i].mask; + } + if( oldFlags!=db->flags ){ + sqlite3ExpirePreparedStatements(db, 0); + } + if( pRes ){ + *pRes = (db->flags & aFlagOp[i].mask)!=0; + } + rc = SQLITE_OK; + break; + } + } + break; + } + } + va_end(ap); + return rc; +} + +/* +** This is the default collating function named "BINARY" which is always +** available. +*/ +static int binCollFunc( + void *NotUsed, + int nKey1, const void *pKey1, + int nKey2, const void *pKey2 +){ + int rc, n; + UNUSED_PARAMETER(NotUsed); + n = nKey1xCmp!=binCollFunc || strcmp(p->zName,"BINARY")==0 ); + return p==0 || p->xCmp==binCollFunc; +} + +/* +** Another built-in collating sequence: NOCASE. +** +** This collating sequence is intended to be used for "case independent +** comparison". SQLite's knowledge of upper and lower case equivalents +** extends only to the 26 characters used in the English language. +** +** At the moment there is only a UTF-8 implementation. +*/ +static int nocaseCollatingFunc( + void *NotUsed, + int nKey1, const void *pKey1, + int nKey2, const void *pKey2 +){ + int r = sqlite3StrNICmp( + (const char *)pKey1, (const char *)pKey2, (nKey1lastRowid; +} + +/* +** Set the value returned by the sqlite3_last_insert_rowid() API function. +*/ +SQLITE_API void sqlite3_set_last_insert_rowid(sqlite3 *db, sqlite3_int64 iRowid){ +#ifdef SQLITE_ENABLE_API_ARMOR + if( !sqlite3SafetyCheckOk(db) ){ + (void)SQLITE_MISUSE_BKPT; + return; + } +#endif + sqlite3_mutex_enter(db->mutex); + db->lastRowid = iRowid; + sqlite3_mutex_leave(db->mutex); +} + +/* +** Return the number of changes in the most recent call to sqlite3_exec(). +*/ +SQLITE_API int sqlite3_changes(sqlite3 *db){ +#ifdef SQLITE_ENABLE_API_ARMOR + if( !sqlite3SafetyCheckOk(db) ){ + (void)SQLITE_MISUSE_BKPT; + return 0; + } +#endif + return db->nChange; +} + +/* +** Return the number of changes since the database handle was opened. +*/ +SQLITE_API int sqlite3_total_changes(sqlite3 *db){ +#ifdef SQLITE_ENABLE_API_ARMOR + if( !sqlite3SafetyCheckOk(db) ){ + (void)SQLITE_MISUSE_BKPT; + return 0; + } +#endif + return db->nTotalChange; +} + +/* +** Close all open savepoints. This function only manipulates fields of the +** database handle object, it does not close any savepoints that may be open +** at the b-tree/pager level. +*/ +SQLITE_PRIVATE void sqlite3CloseSavepoints(sqlite3 *db){ + while( db->pSavepoint ){ + Savepoint *pTmp = db->pSavepoint; + db->pSavepoint = pTmp->pNext; + sqlite3DbFree(db, pTmp); + } + db->nSavepoint = 0; + db->nStatement = 0; + db->isTransactionSavepoint = 0; +} + +/* +** Invoke the destructor function associated with FuncDef p, if any. Except, +** if this is not the last copy of the function, do not invoke it. Multiple +** copies of a single function are created when create_function() is called +** with SQLITE_ANY as the encoding. +*/ +static void functionDestroy(sqlite3 *db, FuncDef *p){ + FuncDestructor *pDestructor = p->u.pDestructor; + if( pDestructor ){ + pDestructor->nRef--; + if( pDestructor->nRef==0 ){ + pDestructor->xDestroy(pDestructor->pUserData); + sqlite3DbFree(db, pDestructor); + } + } +} + +/* +** Disconnect all sqlite3_vtab objects that belong to database connection +** db. This is called when db is being closed. +*/ +static void disconnectAllVtab(sqlite3 *db){ +#ifndef SQLITE_OMIT_VIRTUALTABLE + int i; + HashElem *p; + sqlite3BtreeEnterAll(db); + for(i=0; inDb; i++){ + Schema *pSchema = db->aDb[i].pSchema; + if( pSchema ){ + for(p=sqliteHashFirst(&pSchema->tblHash); p; p=sqliteHashNext(p)){ + Table *pTab = (Table *)sqliteHashData(p); + if( IsVirtual(pTab) ) sqlite3VtabDisconnect(db, pTab); + } + } + } + for(p=sqliteHashFirst(&db->aModule); p; p=sqliteHashNext(p)){ + Module *pMod = (Module *)sqliteHashData(p); + if( pMod->pEpoTab ){ + sqlite3VtabDisconnect(db, pMod->pEpoTab); + } + } + sqlite3VtabUnlockList(db); + sqlite3BtreeLeaveAll(db); +#else + UNUSED_PARAMETER(db); +#endif +} + +/* +** Return TRUE if database connection db has unfinalized prepared +** statements or unfinished sqlite3_backup objects. +*/ +static int connectionIsBusy(sqlite3 *db){ + int j; + assert( sqlite3_mutex_held(db->mutex) ); + if( db->pVdbe ) return 1; + for(j=0; jnDb; j++){ + Btree *pBt = db->aDb[j].pBt; + if( pBt && sqlite3BtreeIsInBackup(pBt) ) return 1; + } + return 0; +} + +/* +** Close an existing SQLite database +*/ +static int sqlite3Close(sqlite3 *db, int forceZombie){ + if( !db ){ + /* EVIDENCE-OF: R-63257-11740 Calling sqlite3_close() or + ** sqlite3_close_v2() with a NULL pointer argument is a harmless no-op. */ + return SQLITE_OK; + } + if( !sqlite3SafetyCheckSickOrOk(db) ){ + return SQLITE_MISUSE_BKPT; + } + sqlite3_mutex_enter(db->mutex); + if( db->mTrace & SQLITE_TRACE_CLOSE ){ + db->xTrace(SQLITE_TRACE_CLOSE, db->pTraceArg, db, 0); + } + + /* Force xDisconnect calls on all virtual tables */ + disconnectAllVtab(db); + + /* If a transaction is open, the disconnectAllVtab() call above + ** will not have called the xDisconnect() method on any virtual + ** tables in the db->aVTrans[] array. The following sqlite3VtabRollback() + ** call will do so. We need to do this before the check for active + ** SQL statements below, as the v-table implementation may be storing + ** some prepared statements internally. + */ + sqlite3VtabRollback(db); + + /* Legacy behavior (sqlite3_close() behavior) is to return + ** SQLITE_BUSY if the connection can not be closed immediately. + */ + if( !forceZombie && connectionIsBusy(db) ){ + sqlite3ErrorWithMsg(db, SQLITE_BUSY, "unable to close due to unfinalized " + "statements or unfinished backups"); + sqlite3_mutex_leave(db->mutex); + return SQLITE_BUSY; + } + +#ifdef SQLITE_ENABLE_SQLLOG + if( sqlite3GlobalConfig.xSqllog ){ + /* Closing the handle. Fourth parameter is passed the value 2. */ + sqlite3GlobalConfig.xSqllog(sqlite3GlobalConfig.pSqllogArg, db, 0, 2); + } +#endif + + /* Convert the connection into a zombie and then close it. + */ + db->magic = SQLITE_MAGIC_ZOMBIE; + sqlite3LeaveMutexAndCloseZombie(db); + return SQLITE_OK; +} + +/* +** Two variations on the public interface for closing a database +** connection. The sqlite3_close() version returns SQLITE_BUSY and +** leaves the connection option if there are unfinalized prepared +** statements or unfinished sqlite3_backups. The sqlite3_close_v2() +** version forces the connection to become a zombie if there are +** unclosed resources, and arranges for deallocation when the last +** prepare statement or sqlite3_backup closes. +*/ +SQLITE_API int sqlite3_close(sqlite3 *db){ return sqlite3Close(db,0); } +SQLITE_API int sqlite3_close_v2(sqlite3 *db){ return sqlite3Close(db,1); } + + +/* +** Close the mutex on database connection db. +** +** Furthermore, if database connection db is a zombie (meaning that there +** has been a prior call to sqlite3_close(db) or sqlite3_close_v2(db)) and +** every sqlite3_stmt has now been finalized and every sqlite3_backup has +** finished, then free all resources. +*/ +SQLITE_PRIVATE void sqlite3LeaveMutexAndCloseZombie(sqlite3 *db){ + HashElem *i; /* Hash table iterator */ + int j; + + /* If there are outstanding sqlite3_stmt or sqlite3_backup objects + ** or if the connection has not yet been closed by sqlite3_close_v2(), + ** then just leave the mutex and return. + */ + if( db->magic!=SQLITE_MAGIC_ZOMBIE || connectionIsBusy(db) ){ + sqlite3_mutex_leave(db->mutex); + return; + } + + /* If we reach this point, it means that the database connection has + ** closed all sqlite3_stmt and sqlite3_backup objects and has been + ** passed to sqlite3_close (meaning that it is a zombie). Therefore, + ** go ahead and free all resources. + */ + + /* If a transaction is open, roll it back. This also ensures that if + ** any database schemas have been modified by an uncommitted transaction + ** they are reset. And that the required b-tree mutex is held to make + ** the pager rollback and schema reset an atomic operation. */ + sqlite3RollbackAll(db, SQLITE_OK); + + /* Free any outstanding Savepoint structures. */ + sqlite3CloseSavepoints(db); + + /* Close all database connections */ + for(j=0; jnDb; j++){ + struct Db *pDb = &db->aDb[j]; + if( pDb->pBt ){ + sqlite3BtreeClose(pDb->pBt); + pDb->pBt = 0; + if( j!=1 ){ + pDb->pSchema = 0; + } + } + } + /* Clear the TEMP schema separately and last */ + if( db->aDb[1].pSchema ){ + sqlite3SchemaClear(db->aDb[1].pSchema); + } + sqlite3VtabUnlockList(db); + + /* Free up the array of auxiliary databases */ + sqlite3CollapseDatabaseArray(db); + assert( db->nDb<=2 ); + assert( db->aDb==db->aDbStatic ); + + /* Tell the code in notify.c that the connection no longer holds any + ** locks and does not require any further unlock-notify callbacks. + */ + sqlite3ConnectionClosed(db); + + for(i=sqliteHashFirst(&db->aFunc); i; i=sqliteHashNext(i)){ + FuncDef *pNext, *p; + p = sqliteHashData(i); + do{ + functionDestroy(db, p); + pNext = p->pNext; + sqlite3DbFree(db, p); + p = pNext; + }while( p ); + } + sqlite3HashClear(&db->aFunc); + for(i=sqliteHashFirst(&db->aCollSeq); i; i=sqliteHashNext(i)){ + CollSeq *pColl = (CollSeq *)sqliteHashData(i); + /* Invoke any destructors registered for collation sequence user data. */ + for(j=0; j<3; j++){ + if( pColl[j].xDel ){ + pColl[j].xDel(pColl[j].pUser); + } + } + sqlite3DbFree(db, pColl); + } + sqlite3HashClear(&db->aCollSeq); +#ifndef SQLITE_OMIT_VIRTUALTABLE + for(i=sqliteHashFirst(&db->aModule); i; i=sqliteHashNext(i)){ + Module *pMod = (Module *)sqliteHashData(i); + sqlite3VtabEponymousTableClear(db, pMod); + sqlite3VtabModuleUnref(db, pMod); + } + sqlite3HashClear(&db->aModule); +#endif + + sqlite3Error(db, SQLITE_OK); /* Deallocates any cached error strings. */ + sqlite3ValueFree(db->pErr); + sqlite3CloseExtensions(db); +#if SQLITE_USER_AUTHENTICATION + sqlite3_free(db->auth.zAuthUser); + sqlite3_free(db->auth.zAuthPW); +#endif + + db->magic = SQLITE_MAGIC_ERROR; + + /* The temp-database schema is allocated differently from the other schema + ** objects (using sqliteMalloc() directly, instead of sqlite3BtreeSchema()). + ** So it needs to be freed here. Todo: Why not roll the temp schema into + ** the same sqliteMalloc() as the one that allocates the database + ** structure? + */ + sqlite3DbFree(db, db->aDb[1].pSchema); + sqlite3_mutex_leave(db->mutex); + db->magic = SQLITE_MAGIC_CLOSED; + sqlite3_mutex_free(db->mutex); + assert( sqlite3LookasideUsed(db,0)==0 ); + if( db->lookaside.bMalloced ){ + sqlite3_free(db->lookaside.pStart); + } + sqlite3_free(db); +} + +/* +** Rollback all database files. If tripCode is not SQLITE_OK, then +** any write cursors are invalidated ("tripped" - as in "tripping a circuit +** breaker") and made to return tripCode if there are any further +** attempts to use that cursor. Read cursors remain open and valid +** but are "saved" in case the table pages are moved around. +*/ +SQLITE_PRIVATE void sqlite3RollbackAll(sqlite3 *db, int tripCode){ + int i; + int inTrans = 0; + int schemaChange; + assert( sqlite3_mutex_held(db->mutex) ); + sqlite3BeginBenignMalloc(); + + /* Obtain all b-tree mutexes before making any calls to BtreeRollback(). + ** This is important in case the transaction being rolled back has + ** modified the database schema. If the b-tree mutexes are not taken + ** here, then another shared-cache connection might sneak in between + ** the database rollback and schema reset, which can cause false + ** corruption reports in some cases. */ + sqlite3BtreeEnterAll(db); + schemaChange = (db->mDbFlags & DBFLAG_SchemaChange)!=0 && db->init.busy==0; + + for(i=0; inDb; i++){ + Btree *p = db->aDb[i].pBt; + if( p ){ + if( sqlite3BtreeIsInTrans(p) ){ + inTrans = 1; + } + sqlite3BtreeRollback(p, tripCode, !schemaChange); + } + } + sqlite3VtabRollback(db); + sqlite3EndBenignMalloc(); + + if( schemaChange ){ + sqlite3ExpirePreparedStatements(db, 0); + sqlite3ResetAllSchemasOfConnection(db); + } + sqlite3BtreeLeaveAll(db); + + /* Any deferred constraint violations have now been resolved. */ + db->nDeferredCons = 0; + db->nDeferredImmCons = 0; + db->flags &= ~(u64)SQLITE_DeferFKs; + + /* If one has been configured, invoke the rollback-hook callback */ + if( db->xRollbackCallback && (inTrans || !db->autoCommit) ){ + db->xRollbackCallback(db->pRollbackArg); + } +} + +/* +** Return a static string containing the name corresponding to the error code +** specified in the argument. +*/ +#if defined(SQLITE_NEED_ERR_NAME) +SQLITE_PRIVATE const char *sqlite3ErrName(int rc){ + const char *zName = 0; + int i, origRc = rc; + for(i=0; i<2 && zName==0; i++, rc &= 0xff){ + switch( rc ){ + case SQLITE_OK: zName = "SQLITE_OK"; break; + case SQLITE_ERROR: zName = "SQLITE_ERROR"; break; + case SQLITE_ERROR_SNAPSHOT: zName = "SQLITE_ERROR_SNAPSHOT"; break; + case SQLITE_INTERNAL: zName = "SQLITE_INTERNAL"; break; + case SQLITE_PERM: zName = "SQLITE_PERM"; break; + case SQLITE_ABORT: zName = "SQLITE_ABORT"; break; + case SQLITE_ABORT_ROLLBACK: zName = "SQLITE_ABORT_ROLLBACK"; break; + case SQLITE_BUSY: zName = "SQLITE_BUSY"; break; + case SQLITE_BUSY_RECOVERY: zName = "SQLITE_BUSY_RECOVERY"; break; + case SQLITE_BUSY_SNAPSHOT: zName = "SQLITE_BUSY_SNAPSHOT"; break; + case SQLITE_LOCKED: zName = "SQLITE_LOCKED"; break; + case SQLITE_LOCKED_SHAREDCACHE: zName = "SQLITE_LOCKED_SHAREDCACHE";break; + case SQLITE_NOMEM: zName = "SQLITE_NOMEM"; break; + case SQLITE_READONLY: zName = "SQLITE_READONLY"; break; + case SQLITE_READONLY_RECOVERY: zName = "SQLITE_READONLY_RECOVERY"; break; + case SQLITE_READONLY_CANTINIT: zName = "SQLITE_READONLY_CANTINIT"; break; + case SQLITE_READONLY_ROLLBACK: zName = "SQLITE_READONLY_ROLLBACK"; break; + case SQLITE_READONLY_DBMOVED: zName = "SQLITE_READONLY_DBMOVED"; break; + case SQLITE_READONLY_DIRECTORY: zName = "SQLITE_READONLY_DIRECTORY";break; + case SQLITE_INTERRUPT: zName = "SQLITE_INTERRUPT"; break; + case SQLITE_IOERR: zName = "SQLITE_IOERR"; break; + case SQLITE_IOERR_READ: zName = "SQLITE_IOERR_READ"; break; + case SQLITE_IOERR_SHORT_READ: zName = "SQLITE_IOERR_SHORT_READ"; break; + case SQLITE_IOERR_WRITE: zName = "SQLITE_IOERR_WRITE"; break; + case SQLITE_IOERR_FSYNC: zName = "SQLITE_IOERR_FSYNC"; break; + case SQLITE_IOERR_DIR_FSYNC: zName = "SQLITE_IOERR_DIR_FSYNC"; break; + case SQLITE_IOERR_TRUNCATE: zName = "SQLITE_IOERR_TRUNCATE"; break; + case SQLITE_IOERR_FSTAT: zName = "SQLITE_IOERR_FSTAT"; break; + case SQLITE_IOERR_UNLOCK: zName = "SQLITE_IOERR_UNLOCK"; break; + case SQLITE_IOERR_RDLOCK: zName = "SQLITE_IOERR_RDLOCK"; break; + case SQLITE_IOERR_DELETE: zName = "SQLITE_IOERR_DELETE"; break; + case SQLITE_IOERR_NOMEM: zName = "SQLITE_IOERR_NOMEM"; break; + case SQLITE_IOERR_ACCESS: zName = "SQLITE_IOERR_ACCESS"; break; + case SQLITE_IOERR_CHECKRESERVEDLOCK: + zName = "SQLITE_IOERR_CHECKRESERVEDLOCK"; break; + case SQLITE_IOERR_LOCK: zName = "SQLITE_IOERR_LOCK"; break; + case SQLITE_IOERR_CLOSE: zName = "SQLITE_IOERR_CLOSE"; break; + case SQLITE_IOERR_DIR_CLOSE: zName = "SQLITE_IOERR_DIR_CLOSE"; break; + case SQLITE_IOERR_SHMOPEN: zName = "SQLITE_IOERR_SHMOPEN"; break; + case SQLITE_IOERR_SHMSIZE: zName = "SQLITE_IOERR_SHMSIZE"; break; + case SQLITE_IOERR_SHMLOCK: zName = "SQLITE_IOERR_SHMLOCK"; break; + case SQLITE_IOERR_SHMMAP: zName = "SQLITE_IOERR_SHMMAP"; break; + case SQLITE_IOERR_SEEK: zName = "SQLITE_IOERR_SEEK"; break; + case SQLITE_IOERR_DELETE_NOENT: zName = "SQLITE_IOERR_DELETE_NOENT";break; + case SQLITE_IOERR_MMAP: zName = "SQLITE_IOERR_MMAP"; break; + case SQLITE_IOERR_GETTEMPPATH: zName = "SQLITE_IOERR_GETTEMPPATH"; break; + case SQLITE_IOERR_CONVPATH: zName = "SQLITE_IOERR_CONVPATH"; break; + case SQLITE_CORRUPT: zName = "SQLITE_CORRUPT"; break; + case SQLITE_CORRUPT_VTAB: zName = "SQLITE_CORRUPT_VTAB"; break; + case SQLITE_NOTFOUND: zName = "SQLITE_NOTFOUND"; break; + case SQLITE_FULL: zName = "SQLITE_FULL"; break; + case SQLITE_CANTOPEN: zName = "SQLITE_CANTOPEN"; break; + case SQLITE_CANTOPEN_NOTEMPDIR: zName = "SQLITE_CANTOPEN_NOTEMPDIR";break; + case SQLITE_CANTOPEN_ISDIR: zName = "SQLITE_CANTOPEN_ISDIR"; break; + case SQLITE_CANTOPEN_FULLPATH: zName = "SQLITE_CANTOPEN_FULLPATH"; break; + case SQLITE_CANTOPEN_CONVPATH: zName = "SQLITE_CANTOPEN_CONVPATH"; break; + case SQLITE_PROTOCOL: zName = "SQLITE_PROTOCOL"; break; + case SQLITE_EMPTY: zName = "SQLITE_EMPTY"; break; + case SQLITE_SCHEMA: zName = "SQLITE_SCHEMA"; break; + case SQLITE_TOOBIG: zName = "SQLITE_TOOBIG"; break; + case SQLITE_CONSTRAINT: zName = "SQLITE_CONSTRAINT"; break; + case SQLITE_CONSTRAINT_UNIQUE: zName = "SQLITE_CONSTRAINT_UNIQUE"; break; + case SQLITE_CONSTRAINT_TRIGGER: zName = "SQLITE_CONSTRAINT_TRIGGER";break; + case SQLITE_CONSTRAINT_FOREIGNKEY: + zName = "SQLITE_CONSTRAINT_FOREIGNKEY"; break; + case SQLITE_CONSTRAINT_CHECK: zName = "SQLITE_CONSTRAINT_CHECK"; break; + case SQLITE_CONSTRAINT_PRIMARYKEY: + zName = "SQLITE_CONSTRAINT_PRIMARYKEY"; break; + case SQLITE_CONSTRAINT_NOTNULL: zName = "SQLITE_CONSTRAINT_NOTNULL";break; + case SQLITE_CONSTRAINT_COMMITHOOK: + zName = "SQLITE_CONSTRAINT_COMMITHOOK"; break; + case SQLITE_CONSTRAINT_VTAB: zName = "SQLITE_CONSTRAINT_VTAB"; break; + case SQLITE_CONSTRAINT_FUNCTION: + zName = "SQLITE_CONSTRAINT_FUNCTION"; break; + case SQLITE_CONSTRAINT_ROWID: zName = "SQLITE_CONSTRAINT_ROWID"; break; + case SQLITE_MISMATCH: zName = "SQLITE_MISMATCH"; break; + case SQLITE_MISUSE: zName = "SQLITE_MISUSE"; break; + case SQLITE_NOLFS: zName = "SQLITE_NOLFS"; break; + case SQLITE_AUTH: zName = "SQLITE_AUTH"; break; + case SQLITE_FORMAT: zName = "SQLITE_FORMAT"; break; + case SQLITE_RANGE: zName = "SQLITE_RANGE"; break; + case SQLITE_NOTADB: zName = "SQLITE_NOTADB"; break; + case SQLITE_ROW: zName = "SQLITE_ROW"; break; + case SQLITE_NOTICE: zName = "SQLITE_NOTICE"; break; + case SQLITE_NOTICE_RECOVER_WAL: zName = "SQLITE_NOTICE_RECOVER_WAL";break; + case SQLITE_NOTICE_RECOVER_ROLLBACK: + zName = "SQLITE_NOTICE_RECOVER_ROLLBACK"; break; + case SQLITE_WARNING: zName = "SQLITE_WARNING"; break; + case SQLITE_WARNING_AUTOINDEX: zName = "SQLITE_WARNING_AUTOINDEX"; break; + case SQLITE_DONE: zName = "SQLITE_DONE"; break; + } + } + if( zName==0 ){ + static char zBuf[50]; + sqlite3_snprintf(sizeof(zBuf), zBuf, "SQLITE_UNKNOWN(%d)", origRc); + zName = zBuf; + } + return zName; +} +#endif + +/* +** Return a static string that describes the kind of error specified in the +** argument. +*/ +SQLITE_PRIVATE const char *sqlite3ErrStr(int rc){ + static const char* const aMsg[] = { + /* SQLITE_OK */ "not an error", + /* SQLITE_ERROR */ "SQL logic error", + /* SQLITE_INTERNAL */ 0, + /* SQLITE_PERM */ "access permission denied", + /* SQLITE_ABORT */ "query aborted", + /* SQLITE_BUSY */ "database is locked", + /* SQLITE_LOCKED */ "database table is locked", + /* SQLITE_NOMEM */ "out of memory", + /* SQLITE_READONLY */ "attempt to write a readonly database", + /* SQLITE_INTERRUPT */ "interrupted", + /* SQLITE_IOERR */ "disk I/O error", + /* SQLITE_CORRUPT */ "database disk image is malformed", + /* SQLITE_NOTFOUND */ "unknown operation", + /* SQLITE_FULL */ "database or disk is full", + /* SQLITE_CANTOPEN */ "unable to open database file", + /* SQLITE_PROTOCOL */ "locking protocol", + /* SQLITE_EMPTY */ 0, + /* SQLITE_SCHEMA */ "database schema has changed", + /* SQLITE_TOOBIG */ "string or blob too big", + /* SQLITE_CONSTRAINT */ "constraint failed", + /* SQLITE_MISMATCH */ "datatype mismatch", + /* SQLITE_MISUSE */ "bad parameter or other API misuse", +#ifdef SQLITE_DISABLE_LFS + /* SQLITE_NOLFS */ "large file support is disabled", +#else + /* SQLITE_NOLFS */ 0, +#endif + /* SQLITE_AUTH */ "authorization denied", + /* SQLITE_FORMAT */ 0, + /* SQLITE_RANGE */ "column index out of range", + /* SQLITE_NOTADB */ "file is not a database", + /* SQLITE_NOTICE */ "notification message", + /* SQLITE_WARNING */ "warning message", + }; + const char *zErr = "unknown error"; + switch( rc ){ + case SQLITE_ABORT_ROLLBACK: { + zErr = "abort due to ROLLBACK"; + break; + } + case SQLITE_ROW: { + zErr = "another row available"; + break; + } + case SQLITE_DONE: { + zErr = "no more rows available"; + break; + } + default: { + rc &= 0xff; + if( ALWAYS(rc>=0) && rcbusyTimeout; + int delay, prior; + +#ifdef SQLITE_ENABLE_SETLK_TIMEOUT + if( sqlite3OsFileControl(pFile,SQLITE_FCNTL_LOCK_TIMEOUT,&tmout)==SQLITE_OK ){ + if( count ){ + tmout = 0; + sqlite3OsFileControl(pFile, SQLITE_FCNTL_LOCK_TIMEOUT, &tmout); + return 0; + }else{ + return 1; + } + } +#else + UNUSED_PARAMETER(pFile); +#endif + assert( count>=0 ); + if( count < NDELAY ){ + delay = delays[count]; + prior = totals[count]; + }else{ + delay = delays[NDELAY-1]; + prior = totals[NDELAY-1] + delay*(count-(NDELAY-1)); + } + if( prior + delay > tmout ){ + delay = tmout - prior; + if( delay<=0 ) return 0; + } + sqlite3OsSleep(db->pVfs, delay*1000); + return 1; +#else + /* This case for unix systems that lack usleep() support. Sleeping + ** must be done in increments of whole seconds */ + sqlite3 *db = (sqlite3 *)ptr; + int tmout = ((sqlite3 *)ptr)->busyTimeout; + UNUSED_PARAMETER(pFile); + if( (count+1)*1000 > tmout ){ + return 0; + } + sqlite3OsSleep(db->pVfs, 1000000); + return 1; +#endif +} + +/* +** Invoke the given busy handler. +** +** This routine is called when an operation failed to acquire a +** lock on VFS file pFile. +** +** If this routine returns non-zero, the lock is retried. If it +** returns 0, the operation aborts with an SQLITE_BUSY error. +*/ +SQLITE_PRIVATE int sqlite3InvokeBusyHandler(BusyHandler *p, sqlite3_file *pFile){ + int rc; + if( p->xBusyHandler==0 || p->nBusy<0 ) return 0; + if( p->bExtraFileArg ){ + /* Add an extra parameter with the pFile pointer to the end of the + ** callback argument list */ + int (*xTra)(void*,int,sqlite3_file*); + xTra = (int(*)(void*,int,sqlite3_file*))p->xBusyHandler; + rc = xTra(p->pBusyArg, p->nBusy, pFile); + }else{ + /* Legacy style busy handler callback */ + rc = p->xBusyHandler(p->pBusyArg, p->nBusy); + } + if( rc==0 ){ + p->nBusy = -1; + }else{ + p->nBusy++; + } + return rc; +} + +/* +** This routine sets the busy callback for an Sqlite database to the +** given callback function with the given argument. +*/ +SQLITE_API int sqlite3_busy_handler( + sqlite3 *db, + int (*xBusy)(void*,int), + void *pArg +){ +#ifdef SQLITE_ENABLE_API_ARMOR + if( !sqlite3SafetyCheckOk(db) ) return SQLITE_MISUSE_BKPT; +#endif + sqlite3_mutex_enter(db->mutex); + db->busyHandler.xBusyHandler = xBusy; + db->busyHandler.pBusyArg = pArg; + db->busyHandler.nBusy = 0; + db->busyHandler.bExtraFileArg = 0; + db->busyTimeout = 0; + sqlite3_mutex_leave(db->mutex); + return SQLITE_OK; +} + +#ifndef SQLITE_OMIT_PROGRESS_CALLBACK +/* +** This routine sets the progress callback for an Sqlite database to the +** given callback function with the given argument. The progress callback will +** be invoked every nOps opcodes. +*/ +SQLITE_API void sqlite3_progress_handler( + sqlite3 *db, + int nOps, + int (*xProgress)(void*), + void *pArg +){ +#ifdef SQLITE_ENABLE_API_ARMOR + if( !sqlite3SafetyCheckOk(db) ){ + (void)SQLITE_MISUSE_BKPT; + return; + } +#endif + sqlite3_mutex_enter(db->mutex); + if( nOps>0 ){ + db->xProgress = xProgress; + db->nProgressOps = (unsigned)nOps; + db->pProgressArg = pArg; + }else{ + db->xProgress = 0; + db->nProgressOps = 0; + db->pProgressArg = 0; + } + sqlite3_mutex_leave(db->mutex); +} +#endif + + +/* +** This routine installs a default busy handler that waits for the +** specified number of milliseconds before returning 0. +*/ +SQLITE_API int sqlite3_busy_timeout(sqlite3 *db, int ms){ +#ifdef SQLITE_ENABLE_API_ARMOR + if( !sqlite3SafetyCheckOk(db) ) return SQLITE_MISUSE_BKPT; +#endif + if( ms>0 ){ + sqlite3_busy_handler(db, (int(*)(void*,int))sqliteDefaultBusyCallback, + (void*)db); + db->busyTimeout = ms; + db->busyHandler.bExtraFileArg = 1; + }else{ + sqlite3_busy_handler(db, 0, 0); + } + return SQLITE_OK; +} + +/* +** Cause any pending operation to stop at its earliest opportunity. +*/ +SQLITE_API void sqlite3_interrupt(sqlite3 *db){ +#ifdef SQLITE_ENABLE_API_ARMOR + if( !sqlite3SafetyCheckOk(db) && (db==0 || db->magic!=SQLITE_MAGIC_ZOMBIE) ){ + (void)SQLITE_MISUSE_BKPT; + return; + } +#endif + db->u1.isInterrupted = 1; +} + + +/* +** This function is exactly the same as sqlite3_create_function(), except +** that it is designed to be called by internal code. The difference is +** that if a malloc() fails in sqlite3_create_function(), an error code +** is returned and the mallocFailed flag cleared. +*/ +SQLITE_PRIVATE int sqlite3CreateFunc( + sqlite3 *db, + const char *zFunctionName, + int nArg, + int enc, + void *pUserData, + void (*xSFunc)(sqlite3_context*,int,sqlite3_value **), + void (*xStep)(sqlite3_context*,int,sqlite3_value **), + void (*xFinal)(sqlite3_context*), + void (*xValue)(sqlite3_context*), + void (*xInverse)(sqlite3_context*,int,sqlite3_value **), + FuncDestructor *pDestructor +){ + FuncDef *p; + int nName; + int extraFlags; + + assert( sqlite3_mutex_held(db->mutex) ); + assert( xValue==0 || xSFunc==0 ); + if( zFunctionName==0 /* Must have a valid name */ + || (xSFunc!=0 && xFinal!=0) /* Not both xSFunc and xFinal */ + || ((xFinal==0)!=(xStep==0)) /* Both or neither of xFinal and xStep */ + || ((xValue==0)!=(xInverse==0)) /* Both or neither of xValue, xInverse */ + || (nArg<-1 || nArg>SQLITE_MAX_FUNCTION_ARG) + || (255<(nName = sqlite3Strlen30( zFunctionName))) + ){ + return SQLITE_MISUSE_BKPT; + } + + assert( SQLITE_FUNC_CONSTANT==SQLITE_DETERMINISTIC ); + assert( SQLITE_FUNC_DIRECT==SQLITE_DIRECTONLY ); + extraFlags = enc & (SQLITE_DETERMINISTIC|SQLITE_DIRECTONLY|SQLITE_SUBTYPE); + enc &= (SQLITE_FUNC_ENCMASK|SQLITE_ANY); + +#ifndef SQLITE_OMIT_UTF16 + /* If SQLITE_UTF16 is specified as the encoding type, transform this + ** to one of SQLITE_UTF16LE or SQLITE_UTF16BE using the + ** SQLITE_UTF16NATIVE macro. SQLITE_UTF16 is not used internally. + ** + ** If SQLITE_ANY is specified, add three versions of the function + ** to the hash table. + */ + if( enc==SQLITE_UTF16 ){ + enc = SQLITE_UTF16NATIVE; + }else if( enc==SQLITE_ANY ){ + int rc; + rc = sqlite3CreateFunc(db, zFunctionName, nArg, SQLITE_UTF8|extraFlags, + pUserData, xSFunc, xStep, xFinal, xValue, xInverse, pDestructor); + if( rc==SQLITE_OK ){ + rc = sqlite3CreateFunc(db, zFunctionName, nArg, SQLITE_UTF16LE|extraFlags, + pUserData, xSFunc, xStep, xFinal, xValue, xInverse, pDestructor); + } + if( rc!=SQLITE_OK ){ + return rc; + } + enc = SQLITE_UTF16BE; + } +#else + enc = SQLITE_UTF8; +#endif + + /* Check if an existing function is being overridden or deleted. If so, + ** and there are active VMs, then return SQLITE_BUSY. If a function + ** is being overridden/deleted but there are no active VMs, allow the + ** operation to continue but invalidate all precompiled statements. + */ + p = sqlite3FindFunction(db, zFunctionName, nArg, (u8)enc, 0); + if( p && (p->funcFlags & SQLITE_FUNC_ENCMASK)==(u32)enc && p->nArg==nArg ){ + if( db->nVdbeActive ){ + sqlite3ErrorWithMsg(db, SQLITE_BUSY, + "unable to delete/modify user-function due to active statements"); + assert( !db->mallocFailed ); + return SQLITE_BUSY; + }else{ + sqlite3ExpirePreparedStatements(db, 0); + } + } + + p = sqlite3FindFunction(db, zFunctionName, nArg, (u8)enc, 1); + assert(p || db->mallocFailed); + if( !p ){ + return SQLITE_NOMEM_BKPT; + } + + /* If an older version of the function with a configured destructor is + ** being replaced invoke the destructor function here. */ + functionDestroy(db, p); + + if( pDestructor ){ + pDestructor->nRef++; + } + p->u.pDestructor = pDestructor; + p->funcFlags = (p->funcFlags & SQLITE_FUNC_ENCMASK) | extraFlags; + testcase( p->funcFlags & SQLITE_DETERMINISTIC ); + testcase( p->funcFlags & SQLITE_DIRECTONLY ); + p->xSFunc = xSFunc ? xSFunc : xStep; + p->xFinalize = xFinal; + p->xValue = xValue; + p->xInverse = xInverse; + p->pUserData = pUserData; + p->nArg = (u16)nArg; + return SQLITE_OK; +} + +/* +** Worker function used by utf-8 APIs that create new functions: +** +** sqlite3_create_function() +** sqlite3_create_function_v2() +** sqlite3_create_window_function() +*/ +static int createFunctionApi( + sqlite3 *db, + const char *zFunc, + int nArg, + int enc, + void *p, + void (*xSFunc)(sqlite3_context*,int,sqlite3_value**), + void (*xStep)(sqlite3_context*,int,sqlite3_value**), + void (*xFinal)(sqlite3_context*), + void (*xValue)(sqlite3_context*), + void (*xInverse)(sqlite3_context*,int,sqlite3_value**), + void(*xDestroy)(void*) +){ + int rc = SQLITE_ERROR; + FuncDestructor *pArg = 0; + +#ifdef SQLITE_ENABLE_API_ARMOR + if( !sqlite3SafetyCheckOk(db) ){ + return SQLITE_MISUSE_BKPT; + } +#endif + sqlite3_mutex_enter(db->mutex); + if( xDestroy ){ + pArg = (FuncDestructor *)sqlite3Malloc(sizeof(FuncDestructor)); + if( !pArg ){ + sqlite3OomFault(db); + xDestroy(p); + goto out; + } + pArg->nRef = 0; + pArg->xDestroy = xDestroy; + pArg->pUserData = p; + } + rc = sqlite3CreateFunc(db, zFunc, nArg, enc, p, + xSFunc, xStep, xFinal, xValue, xInverse, pArg + ); + if( pArg && pArg->nRef==0 ){ + assert( rc!=SQLITE_OK ); + xDestroy(p); + sqlite3_free(pArg); + } + + out: + rc = sqlite3ApiExit(db, rc); + sqlite3_mutex_leave(db->mutex); + return rc; +} + +/* +** Create new user functions. +*/ +SQLITE_API int sqlite3_create_function( + sqlite3 *db, + const char *zFunc, + int nArg, + int enc, + void *p, + void (*xSFunc)(sqlite3_context*,int,sqlite3_value **), + void (*xStep)(sqlite3_context*,int,sqlite3_value **), + void (*xFinal)(sqlite3_context*) +){ + return createFunctionApi(db, zFunc, nArg, enc, p, xSFunc, xStep, + xFinal, 0, 0, 0); +} +SQLITE_API int sqlite3_create_function_v2( + sqlite3 *db, + const char *zFunc, + int nArg, + int enc, + void *p, + void (*xSFunc)(sqlite3_context*,int,sqlite3_value **), + void (*xStep)(sqlite3_context*,int,sqlite3_value **), + void (*xFinal)(sqlite3_context*), + void (*xDestroy)(void *) +){ + return createFunctionApi(db, zFunc, nArg, enc, p, xSFunc, xStep, + xFinal, 0, 0, xDestroy); +} +SQLITE_API int sqlite3_create_window_function( + sqlite3 *db, + const char *zFunc, + int nArg, + int enc, + void *p, + void (*xStep)(sqlite3_context*,int,sqlite3_value **), + void (*xFinal)(sqlite3_context*), + void (*xValue)(sqlite3_context*), + void (*xInverse)(sqlite3_context*,int,sqlite3_value **), + void (*xDestroy)(void *) +){ + return createFunctionApi(db, zFunc, nArg, enc, p, 0, xStep, + xFinal, xValue, xInverse, xDestroy); +} + +#ifndef SQLITE_OMIT_UTF16 +SQLITE_API int sqlite3_create_function16( + sqlite3 *db, + const void *zFunctionName, + int nArg, + int eTextRep, + void *p, + void (*xSFunc)(sqlite3_context*,int,sqlite3_value**), + void (*xStep)(sqlite3_context*,int,sqlite3_value**), + void (*xFinal)(sqlite3_context*) +){ + int rc; + char *zFunc8; + +#ifdef SQLITE_ENABLE_API_ARMOR + if( !sqlite3SafetyCheckOk(db) || zFunctionName==0 ) return SQLITE_MISUSE_BKPT; +#endif + sqlite3_mutex_enter(db->mutex); + assert( !db->mallocFailed ); + zFunc8 = sqlite3Utf16to8(db, zFunctionName, -1, SQLITE_UTF16NATIVE); + rc = sqlite3CreateFunc(db, zFunc8, nArg, eTextRep, p, xSFunc,xStep,xFinal,0,0,0); + sqlite3DbFree(db, zFunc8); + rc = sqlite3ApiExit(db, rc); + sqlite3_mutex_leave(db->mutex); + return rc; +} +#endif + + +/* +** The following is the implementation of an SQL function that always +** fails with an error message stating that the function is used in the +** wrong context. The sqlite3_overload_function() API might construct +** SQL function that use this routine so that the functions will exist +** for name resolution but are actually overloaded by the xFindFunction +** method of virtual tables. +*/ +static void sqlite3InvalidFunction( + sqlite3_context *context, /* The function calling context */ + int NotUsed, /* Number of arguments to the function */ + sqlite3_value **NotUsed2 /* Value of each argument */ +){ + const char *zName = (const char*)sqlite3_user_data(context); + char *zErr; + UNUSED_PARAMETER2(NotUsed, NotUsed2); + zErr = sqlite3_mprintf( + "unable to use function %s in the requested context", zName); + sqlite3_result_error(context, zErr, -1); + sqlite3_free(zErr); +} + +/* +** Declare that a function has been overloaded by a virtual table. +** +** If the function already exists as a regular global function, then +** this routine is a no-op. If the function does not exist, then create +** a new one that always throws a run-time error. +** +** When virtual tables intend to provide an overloaded function, they +** should call this routine to make sure the global function exists. +** A global function must exist in order for name resolution to work +** properly. +*/ +SQLITE_API int sqlite3_overload_function( + sqlite3 *db, + const char *zName, + int nArg +){ + int rc; + char *zCopy; + +#ifdef SQLITE_ENABLE_API_ARMOR + if( !sqlite3SafetyCheckOk(db) || zName==0 || nArg<-2 ){ + return SQLITE_MISUSE_BKPT; + } +#endif + sqlite3_mutex_enter(db->mutex); + rc = sqlite3FindFunction(db, zName, nArg, SQLITE_UTF8, 0)!=0; + sqlite3_mutex_leave(db->mutex); + if( rc ) return SQLITE_OK; + zCopy = sqlite3_mprintf(zName); + if( zCopy==0 ) return SQLITE_NOMEM; + return sqlite3_create_function_v2(db, zName, nArg, SQLITE_UTF8, + zCopy, sqlite3InvalidFunction, 0, 0, sqlite3_free); +} + +#ifndef SQLITE_OMIT_TRACE +/* +** Register a trace function. The pArg from the previously registered trace +** is returned. +** +** A NULL trace function means that no tracing is executes. A non-NULL +** trace is a pointer to a function that is invoked at the start of each +** SQL statement. +*/ +#ifndef SQLITE_OMIT_DEPRECATED +SQLITE_API void *sqlite3_trace(sqlite3 *db, void(*xTrace)(void*,const char*), void *pArg){ + void *pOld; + +#ifdef SQLITE_ENABLE_API_ARMOR + if( !sqlite3SafetyCheckOk(db) ){ + (void)SQLITE_MISUSE_BKPT; + return 0; + } +#endif + sqlite3_mutex_enter(db->mutex); + pOld = db->pTraceArg; + db->mTrace = xTrace ? SQLITE_TRACE_LEGACY : 0; + db->xTrace = (int(*)(u32,void*,void*,void*))xTrace; + db->pTraceArg = pArg; + sqlite3_mutex_leave(db->mutex); + return pOld; +} +#endif /* SQLITE_OMIT_DEPRECATED */ + +/* Register a trace callback using the version-2 interface. +*/ +SQLITE_API int sqlite3_trace_v2( + sqlite3 *db, /* Trace this connection */ + unsigned mTrace, /* Mask of events to be traced */ + int(*xTrace)(unsigned,void*,void*,void*), /* Callback to invoke */ + void *pArg /* Context */ +){ +#ifdef SQLITE_ENABLE_API_ARMOR + if( !sqlite3SafetyCheckOk(db) ){ + return SQLITE_MISUSE_BKPT; + } +#endif + sqlite3_mutex_enter(db->mutex); + if( mTrace==0 ) xTrace = 0; + if( xTrace==0 ) mTrace = 0; + db->mTrace = mTrace; + db->xTrace = xTrace; + db->pTraceArg = pArg; + sqlite3_mutex_leave(db->mutex); + return SQLITE_OK; +} + +#ifndef SQLITE_OMIT_DEPRECATED +/* +** Register a profile function. The pArg from the previously registered +** profile function is returned. +** +** A NULL profile function means that no profiling is executes. A non-NULL +** profile is a pointer to a function that is invoked at the conclusion of +** each SQL statement that is run. +*/ +SQLITE_API void *sqlite3_profile( + sqlite3 *db, + void (*xProfile)(void*,const char*,sqlite_uint64), + void *pArg +){ + void *pOld; + +#ifdef SQLITE_ENABLE_API_ARMOR + if( !sqlite3SafetyCheckOk(db) ){ + (void)SQLITE_MISUSE_BKPT; + return 0; + } +#endif + sqlite3_mutex_enter(db->mutex); + pOld = db->pProfileArg; + db->xProfile = xProfile; + db->pProfileArg = pArg; + db->mTrace &= SQLITE_TRACE_NONLEGACY_MASK; + if( db->xProfile ) db->mTrace |= SQLITE_TRACE_XPROFILE; + sqlite3_mutex_leave(db->mutex); + return pOld; +} +#endif /* SQLITE_OMIT_DEPRECATED */ +#endif /* SQLITE_OMIT_TRACE */ + +/* +** Register a function to be invoked when a transaction commits. +** If the invoked function returns non-zero, then the commit becomes a +** rollback. +*/ +SQLITE_API void *sqlite3_commit_hook( + sqlite3 *db, /* Attach the hook to this database */ + int (*xCallback)(void*), /* Function to invoke on each commit */ + void *pArg /* Argument to the function */ +){ + void *pOld; + +#ifdef SQLITE_ENABLE_API_ARMOR + if( !sqlite3SafetyCheckOk(db) ){ + (void)SQLITE_MISUSE_BKPT; + return 0; + } +#endif + sqlite3_mutex_enter(db->mutex); + pOld = db->pCommitArg; + db->xCommitCallback = xCallback; + db->pCommitArg = pArg; + sqlite3_mutex_leave(db->mutex); + return pOld; +} + +/* +** Register a callback to be invoked each time a row is updated, +** inserted or deleted using this database connection. +*/ +SQLITE_API void *sqlite3_update_hook( + sqlite3 *db, /* Attach the hook to this database */ + void (*xCallback)(void*,int,char const *,char const *,sqlite_int64), + void *pArg /* Argument to the function */ +){ + void *pRet; + +#ifdef SQLITE_ENABLE_API_ARMOR + if( !sqlite3SafetyCheckOk(db) ){ + (void)SQLITE_MISUSE_BKPT; + return 0; + } +#endif + sqlite3_mutex_enter(db->mutex); + pRet = db->pUpdateArg; + db->xUpdateCallback = xCallback; + db->pUpdateArg = pArg; + sqlite3_mutex_leave(db->mutex); + return pRet; +} + +/* +** Register a callback to be invoked each time a transaction is rolled +** back by this database connection. +*/ +SQLITE_API void *sqlite3_rollback_hook( + sqlite3 *db, /* Attach the hook to this database */ + void (*xCallback)(void*), /* Callback function */ + void *pArg /* Argument to the function */ +){ + void *pRet; + +#ifdef SQLITE_ENABLE_API_ARMOR + if( !sqlite3SafetyCheckOk(db) ){ + (void)SQLITE_MISUSE_BKPT; + return 0; + } +#endif + sqlite3_mutex_enter(db->mutex); + pRet = db->pRollbackArg; + db->xRollbackCallback = xCallback; + db->pRollbackArg = pArg; + sqlite3_mutex_leave(db->mutex); + return pRet; +} + +#ifdef SQLITE_ENABLE_PREUPDATE_HOOK +/* +** Register a callback to be invoked each time a row is updated, +** inserted or deleted using this database connection. +*/ +SQLITE_API void *sqlite3_preupdate_hook( + sqlite3 *db, /* Attach the hook to this database */ + void(*xCallback)( /* Callback function */ + void*,sqlite3*,int,char const*,char const*,sqlite3_int64,sqlite3_int64), + void *pArg /* First callback argument */ +){ + void *pRet; + sqlite3_mutex_enter(db->mutex); + pRet = db->pPreUpdateArg; + db->xPreUpdateCallback = xCallback; + db->pPreUpdateArg = pArg; + sqlite3_mutex_leave(db->mutex); + return pRet; +} +#endif /* SQLITE_ENABLE_PREUPDATE_HOOK */ + +#ifndef SQLITE_OMIT_WAL +/* +** The sqlite3_wal_hook() callback registered by sqlite3_wal_autocheckpoint(). +** Invoke sqlite3_wal_checkpoint if the number of frames in the log file +** is greater than sqlite3.pWalArg cast to an integer (the value configured by +** wal_autocheckpoint()). +*/ +SQLITE_PRIVATE int sqlite3WalDefaultHook( + void *pClientData, /* Argument */ + sqlite3 *db, /* Connection */ + const char *zDb, /* Database */ + int nFrame /* Size of WAL */ +){ + if( nFrame>=SQLITE_PTR_TO_INT(pClientData) ){ + sqlite3BeginBenignMalloc(); + sqlite3_wal_checkpoint(db, zDb); + sqlite3EndBenignMalloc(); + } + return SQLITE_OK; +} +#endif /* SQLITE_OMIT_WAL */ + +/* +** Configure an sqlite3_wal_hook() callback to automatically checkpoint +** a database after committing a transaction if there are nFrame or +** more frames in the log file. Passing zero or a negative value as the +** nFrame parameter disables automatic checkpoints entirely. +** +** The callback registered by this function replaces any existing callback +** registered using sqlite3_wal_hook(). Likewise, registering a callback +** using sqlite3_wal_hook() disables the automatic checkpoint mechanism +** configured by this function. +*/ +SQLITE_API int sqlite3_wal_autocheckpoint(sqlite3 *db, int nFrame){ +#ifdef SQLITE_OMIT_WAL + UNUSED_PARAMETER(db); + UNUSED_PARAMETER(nFrame); +#else +#ifdef SQLITE_ENABLE_API_ARMOR + if( !sqlite3SafetyCheckOk(db) ) return SQLITE_MISUSE_BKPT; +#endif + if( nFrame>0 ){ + sqlite3_wal_hook(db, sqlite3WalDefaultHook, SQLITE_INT_TO_PTR(nFrame)); + }else{ + sqlite3_wal_hook(db, 0, 0); + } +#endif + return SQLITE_OK; +} + +/* +** Register a callback to be invoked each time a transaction is written +** into the write-ahead-log by this database connection. +*/ +SQLITE_API void *sqlite3_wal_hook( + sqlite3 *db, /* Attach the hook to this db handle */ + int(*xCallback)(void *, sqlite3*, const char*, int), + void *pArg /* First argument passed to xCallback() */ +){ +#ifndef SQLITE_OMIT_WAL + void *pRet; +#ifdef SQLITE_ENABLE_API_ARMOR + if( !sqlite3SafetyCheckOk(db) ){ + (void)SQLITE_MISUSE_BKPT; + return 0; + } +#endif + sqlite3_mutex_enter(db->mutex); + pRet = db->pWalArg; + db->xWalCallback = xCallback; + db->pWalArg = pArg; + sqlite3_mutex_leave(db->mutex); + return pRet; +#else + return 0; +#endif +} + +/* +** Checkpoint database zDb. +*/ +SQLITE_API int sqlite3_wal_checkpoint_v2( + sqlite3 *db, /* Database handle */ + const char *zDb, /* Name of attached database (or NULL) */ + int eMode, /* SQLITE_CHECKPOINT_* value */ + int *pnLog, /* OUT: Size of WAL log in frames */ + int *pnCkpt /* OUT: Total number of frames checkpointed */ +){ +#ifdef SQLITE_OMIT_WAL + return SQLITE_OK; +#else + int rc; /* Return code */ + int iDb = SQLITE_MAX_ATTACHED; /* sqlite3.aDb[] index of db to checkpoint */ + +#ifdef SQLITE_ENABLE_API_ARMOR + if( !sqlite3SafetyCheckOk(db) ) return SQLITE_MISUSE_BKPT; +#endif + + /* Initialize the output variables to -1 in case an error occurs. */ + if( pnLog ) *pnLog = -1; + if( pnCkpt ) *pnCkpt = -1; + + assert( SQLITE_CHECKPOINT_PASSIVE==0 ); + assert( SQLITE_CHECKPOINT_FULL==1 ); + assert( SQLITE_CHECKPOINT_RESTART==2 ); + assert( SQLITE_CHECKPOINT_TRUNCATE==3 ); + if( eModeSQLITE_CHECKPOINT_TRUNCATE ){ + /* EVIDENCE-OF: R-03996-12088 The M parameter must be a valid checkpoint + ** mode: */ + return SQLITE_MISUSE; + } + + sqlite3_mutex_enter(db->mutex); + if( zDb && zDb[0] ){ + iDb = sqlite3FindDbName(db, zDb); + } + if( iDb<0 ){ + rc = SQLITE_ERROR; + sqlite3ErrorWithMsg(db, SQLITE_ERROR, "unknown database: %s", zDb); + }else{ + db->busyHandler.nBusy = 0; + rc = sqlite3Checkpoint(db, iDb, eMode, pnLog, pnCkpt); + sqlite3Error(db, rc); + } + rc = sqlite3ApiExit(db, rc); + + /* If there are no active statements, clear the interrupt flag at this + ** point. */ + if( db->nVdbeActive==0 ){ + db->u1.isInterrupted = 0; + } + + sqlite3_mutex_leave(db->mutex); + return rc; +#endif +} + + +/* +** Checkpoint database zDb. If zDb is NULL, or if the buffer zDb points +** to contains a zero-length string, all attached databases are +** checkpointed. +*/ +SQLITE_API int sqlite3_wal_checkpoint(sqlite3 *db, const char *zDb){ + /* EVIDENCE-OF: R-41613-20553 The sqlite3_wal_checkpoint(D,X) is equivalent to + ** sqlite3_wal_checkpoint_v2(D,X,SQLITE_CHECKPOINT_PASSIVE,0,0). */ + return sqlite3_wal_checkpoint_v2(db,zDb,SQLITE_CHECKPOINT_PASSIVE,0,0); +} + +#ifndef SQLITE_OMIT_WAL +/* +** Run a checkpoint on database iDb. This is a no-op if database iDb is +** not currently open in WAL mode. +** +** If a transaction is open on the database being checkpointed, this +** function returns SQLITE_LOCKED and a checkpoint is not attempted. If +** an error occurs while running the checkpoint, an SQLite error code is +** returned (i.e. SQLITE_IOERR). Otherwise, SQLITE_OK. +** +** The mutex on database handle db should be held by the caller. The mutex +** associated with the specific b-tree being checkpointed is taken by +** this function while the checkpoint is running. +** +** If iDb is passed SQLITE_MAX_ATTACHED, then all attached databases are +** checkpointed. If an error is encountered it is returned immediately - +** no attempt is made to checkpoint any remaining databases. +** +** Parameter eMode is one of SQLITE_CHECKPOINT_PASSIVE, FULL, RESTART +** or TRUNCATE. +*/ +SQLITE_PRIVATE int sqlite3Checkpoint(sqlite3 *db, int iDb, int eMode, int *pnLog, int *pnCkpt){ + int rc = SQLITE_OK; /* Return code */ + int i; /* Used to iterate through attached dbs */ + int bBusy = 0; /* True if SQLITE_BUSY has been encountered */ + + assert( sqlite3_mutex_held(db->mutex) ); + assert( !pnLog || *pnLog==-1 ); + assert( !pnCkpt || *pnCkpt==-1 ); + + for(i=0; inDb && rc==SQLITE_OK; i++){ + if( i==iDb || iDb==SQLITE_MAX_ATTACHED ){ + rc = sqlite3BtreeCheckpoint(db->aDb[i].pBt, eMode, pnLog, pnCkpt); + pnLog = 0; + pnCkpt = 0; + if( rc==SQLITE_BUSY ){ + bBusy = 1; + rc = SQLITE_OK; + } + } + } + + return (rc==SQLITE_OK && bBusy) ? SQLITE_BUSY : rc; +} +#endif /* SQLITE_OMIT_WAL */ + +/* +** This function returns true if main-memory should be used instead of +** a temporary file for transient pager files and statement journals. +** The value returned depends on the value of db->temp_store (runtime +** parameter) and the compile time value of SQLITE_TEMP_STORE. The +** following table describes the relationship between these two values +** and this functions return value. +** +** SQLITE_TEMP_STORE db->temp_store Location of temporary database +** ----------------- -------------- ------------------------------ +** 0 any file (return 0) +** 1 1 file (return 0) +** 1 2 memory (return 1) +** 1 0 file (return 0) +** 2 1 file (return 0) +** 2 2 memory (return 1) +** 2 0 memory (return 1) +** 3 any memory (return 1) +*/ +SQLITE_PRIVATE int sqlite3TempInMemory(const sqlite3 *db){ +#if SQLITE_TEMP_STORE==1 + return ( db->temp_store==2 ); +#endif +#if SQLITE_TEMP_STORE==2 + return ( db->temp_store!=1 ); +#endif +#if SQLITE_TEMP_STORE==3 + UNUSED_PARAMETER(db); + return 1; +#endif +#if SQLITE_TEMP_STORE<1 || SQLITE_TEMP_STORE>3 + UNUSED_PARAMETER(db); + return 0; +#endif +} + +/* +** Return UTF-8 encoded English language explanation of the most recent +** error. +*/ +SQLITE_API const char *sqlite3_errmsg(sqlite3 *db){ + const char *z; + if( !db ){ + return sqlite3ErrStr(SQLITE_NOMEM_BKPT); + } + if( !sqlite3SafetyCheckSickOrOk(db) ){ + return sqlite3ErrStr(SQLITE_MISUSE_BKPT); + } + sqlite3_mutex_enter(db->mutex); + if( db->mallocFailed ){ + z = sqlite3ErrStr(SQLITE_NOMEM_BKPT); + }else{ + testcase( db->pErr==0 ); + z = db->errCode ? (char*)sqlite3_value_text(db->pErr) : 0; + assert( !db->mallocFailed ); + if( z==0 ){ + z = sqlite3ErrStr(db->errCode); + } + } + sqlite3_mutex_leave(db->mutex); + return z; +} + +#ifndef SQLITE_OMIT_UTF16 +/* +** Return UTF-16 encoded English language explanation of the most recent +** error. +*/ +SQLITE_API const void *sqlite3_errmsg16(sqlite3 *db){ + static const u16 outOfMem[] = { + 'o', 'u', 't', ' ', 'o', 'f', ' ', 'm', 'e', 'm', 'o', 'r', 'y', 0 + }; + static const u16 misuse[] = { + 'b', 'a', 'd', ' ', 'p', 'a', 'r', 'a', 'm', 'e', 't', 'e', 'r', ' ', + 'o', 'r', ' ', 'o', 't', 'h', 'e', 'r', ' ', 'A', 'P', 'I', ' ', + 'm', 'i', 's', 'u', 's', 'e', 0 + }; + + const void *z; + if( !db ){ + return (void *)outOfMem; + } + if( !sqlite3SafetyCheckSickOrOk(db) ){ + return (void *)misuse; + } + sqlite3_mutex_enter(db->mutex); + if( db->mallocFailed ){ + z = (void *)outOfMem; + }else{ + z = sqlite3_value_text16(db->pErr); + if( z==0 ){ + sqlite3ErrorWithMsg(db, db->errCode, sqlite3ErrStr(db->errCode)); + z = sqlite3_value_text16(db->pErr); + } + /* A malloc() may have failed within the call to sqlite3_value_text16() + ** above. If this is the case, then the db->mallocFailed flag needs to + ** be cleared before returning. Do this directly, instead of via + ** sqlite3ApiExit(), to avoid setting the database handle error message. + */ + sqlite3OomClear(db); + } + sqlite3_mutex_leave(db->mutex); + return z; +} +#endif /* SQLITE_OMIT_UTF16 */ + +/* +** Return the most recent error code generated by an SQLite routine. If NULL is +** passed to this function, we assume a malloc() failed during sqlite3_open(). +*/ +SQLITE_API int sqlite3_errcode(sqlite3 *db){ + if( db && !sqlite3SafetyCheckSickOrOk(db) ){ + return SQLITE_MISUSE_BKPT; + } + if( !db || db->mallocFailed ){ + return SQLITE_NOMEM_BKPT; + } + return db->errCode & db->errMask; +} +SQLITE_API int sqlite3_extended_errcode(sqlite3 *db){ + if( db && !sqlite3SafetyCheckSickOrOk(db) ){ + return SQLITE_MISUSE_BKPT; + } + if( !db || db->mallocFailed ){ + return SQLITE_NOMEM_BKPT; + } + return db->errCode; +} +SQLITE_API int sqlite3_system_errno(sqlite3 *db){ + return db ? db->iSysErrno : 0; +} + +/* +** Return a string that describes the kind of error specified in the +** argument. For now, this simply calls the internal sqlite3ErrStr() +** function. +*/ +SQLITE_API const char *sqlite3_errstr(int rc){ + return sqlite3ErrStr(rc); +} + +/* +** Create a new collating function for database "db". The name is zName +** and the encoding is enc. +*/ +static int createCollation( + sqlite3* db, + const char *zName, + u8 enc, + void* pCtx, + int(*xCompare)(void*,int,const void*,int,const void*), + void(*xDel)(void*) +){ + CollSeq *pColl; + int enc2; + + assert( sqlite3_mutex_held(db->mutex) ); + + /* If SQLITE_UTF16 is specified as the encoding type, transform this + ** to one of SQLITE_UTF16LE or SQLITE_UTF16BE using the + ** SQLITE_UTF16NATIVE macro. SQLITE_UTF16 is not used internally. + */ + enc2 = enc; + testcase( enc2==SQLITE_UTF16 ); + testcase( enc2==SQLITE_UTF16_ALIGNED ); + if( enc2==SQLITE_UTF16 || enc2==SQLITE_UTF16_ALIGNED ){ + enc2 = SQLITE_UTF16NATIVE; + } + if( enc2SQLITE_UTF16BE ){ + return SQLITE_MISUSE_BKPT; + } + + /* Check if this call is removing or replacing an existing collation + ** sequence. If so, and there are active VMs, return busy. If there + ** are no active VMs, invalidate any pre-compiled statements. + */ + pColl = sqlite3FindCollSeq(db, (u8)enc2, zName, 0); + if( pColl && pColl->xCmp ){ + if( db->nVdbeActive ){ + sqlite3ErrorWithMsg(db, SQLITE_BUSY, + "unable to delete/modify collation sequence due to active statements"); + return SQLITE_BUSY; + } + sqlite3ExpirePreparedStatements(db, 0); + + /* If collation sequence pColl was created directly by a call to + ** sqlite3_create_collation, and not generated by synthCollSeq(), + ** then any copies made by synthCollSeq() need to be invalidated. + ** Also, collation destructor - CollSeq.xDel() - function may need + ** to be called. + */ + if( (pColl->enc & ~SQLITE_UTF16_ALIGNED)==enc2 ){ + CollSeq *aColl = sqlite3HashFind(&db->aCollSeq, zName); + int j; + for(j=0; j<3; j++){ + CollSeq *p = &aColl[j]; + if( p->enc==pColl->enc ){ + if( p->xDel ){ + p->xDel(p->pUser); + } + p->xCmp = 0; + } + } + } + } + + pColl = sqlite3FindCollSeq(db, (u8)enc2, zName, 1); + if( pColl==0 ) return SQLITE_NOMEM_BKPT; + pColl->xCmp = xCompare; + pColl->pUser = pCtx; + pColl->xDel = xDel; + pColl->enc = (u8)(enc2 | (enc & SQLITE_UTF16_ALIGNED)); + sqlite3Error(db, SQLITE_OK); + return SQLITE_OK; +} + + +/* +** This array defines hard upper bounds on limit values. The +** initializer must be kept in sync with the SQLITE_LIMIT_* +** #defines in sqlite3.h. +*/ +static const int aHardLimit[] = { + SQLITE_MAX_LENGTH, + SQLITE_MAX_SQL_LENGTH, + SQLITE_MAX_COLUMN, + SQLITE_MAX_EXPR_DEPTH, + SQLITE_MAX_COMPOUND_SELECT, + SQLITE_MAX_VDBE_OP, + SQLITE_MAX_FUNCTION_ARG, + SQLITE_MAX_ATTACHED, + SQLITE_MAX_LIKE_PATTERN_LENGTH, + SQLITE_MAX_VARIABLE_NUMBER, /* IMP: R-38091-32352 */ + SQLITE_MAX_TRIGGER_DEPTH, + SQLITE_MAX_WORKER_THREADS, +}; + +/* +** Make sure the hard limits are set to reasonable values +*/ +#if SQLITE_MAX_LENGTH<100 +# error SQLITE_MAX_LENGTH must be at least 100 +#endif +#if SQLITE_MAX_SQL_LENGTH<100 +# error SQLITE_MAX_SQL_LENGTH must be at least 100 +#endif +#if SQLITE_MAX_SQL_LENGTH>SQLITE_MAX_LENGTH +# error SQLITE_MAX_SQL_LENGTH must not be greater than SQLITE_MAX_LENGTH +#endif +#if SQLITE_MAX_COMPOUND_SELECT<2 +# error SQLITE_MAX_COMPOUND_SELECT must be at least 2 +#endif +#if SQLITE_MAX_VDBE_OP<40 +# error SQLITE_MAX_VDBE_OP must be at least 40 +#endif +#if SQLITE_MAX_FUNCTION_ARG<0 || SQLITE_MAX_FUNCTION_ARG>127 +# error SQLITE_MAX_FUNCTION_ARG must be between 0 and 127 +#endif +#if SQLITE_MAX_ATTACHED<0 || SQLITE_MAX_ATTACHED>125 +# error SQLITE_MAX_ATTACHED must be between 0 and 125 +#endif +#if SQLITE_MAX_LIKE_PATTERN_LENGTH<1 +# error SQLITE_MAX_LIKE_PATTERN_LENGTH must be at least 1 +#endif +#if SQLITE_MAX_COLUMN>32767 +# error SQLITE_MAX_COLUMN must not exceed 32767 +#endif +#if SQLITE_MAX_TRIGGER_DEPTH<1 +# error SQLITE_MAX_TRIGGER_DEPTH must be at least 1 +#endif +#if SQLITE_MAX_WORKER_THREADS<0 || SQLITE_MAX_WORKER_THREADS>50 +# error SQLITE_MAX_WORKER_THREADS must be between 0 and 50 +#endif + + +/* +** Change the value of a limit. Report the old value. +** If an invalid limit index is supplied, report -1. +** Make no changes but still report the old value if the +** new limit is negative. +** +** A new lower limit does not shrink existing constructs. +** It merely prevents new constructs that exceed the limit +** from forming. +*/ +SQLITE_API int sqlite3_limit(sqlite3 *db, int limitId, int newLimit){ + int oldLimit; + +#ifdef SQLITE_ENABLE_API_ARMOR + if( !sqlite3SafetyCheckOk(db) ){ + (void)SQLITE_MISUSE_BKPT; + return -1; + } +#endif + + /* EVIDENCE-OF: R-30189-54097 For each limit category SQLITE_LIMIT_NAME + ** there is a hard upper bound set at compile-time by a C preprocessor + ** macro called SQLITE_MAX_NAME. (The "_LIMIT_" in the name is changed to + ** "_MAX_".) + */ + assert( aHardLimit[SQLITE_LIMIT_LENGTH]==SQLITE_MAX_LENGTH ); + assert( aHardLimit[SQLITE_LIMIT_SQL_LENGTH]==SQLITE_MAX_SQL_LENGTH ); + assert( aHardLimit[SQLITE_LIMIT_COLUMN]==SQLITE_MAX_COLUMN ); + assert( aHardLimit[SQLITE_LIMIT_EXPR_DEPTH]==SQLITE_MAX_EXPR_DEPTH ); + assert( aHardLimit[SQLITE_LIMIT_COMPOUND_SELECT]==SQLITE_MAX_COMPOUND_SELECT); + assert( aHardLimit[SQLITE_LIMIT_VDBE_OP]==SQLITE_MAX_VDBE_OP ); + assert( aHardLimit[SQLITE_LIMIT_FUNCTION_ARG]==SQLITE_MAX_FUNCTION_ARG ); + assert( aHardLimit[SQLITE_LIMIT_ATTACHED]==SQLITE_MAX_ATTACHED ); + assert( aHardLimit[SQLITE_LIMIT_LIKE_PATTERN_LENGTH]== + SQLITE_MAX_LIKE_PATTERN_LENGTH ); + assert( aHardLimit[SQLITE_LIMIT_VARIABLE_NUMBER]==SQLITE_MAX_VARIABLE_NUMBER); + assert( aHardLimit[SQLITE_LIMIT_TRIGGER_DEPTH]==SQLITE_MAX_TRIGGER_DEPTH ); + assert( aHardLimit[SQLITE_LIMIT_WORKER_THREADS]==SQLITE_MAX_WORKER_THREADS ); + assert( SQLITE_LIMIT_WORKER_THREADS==(SQLITE_N_LIMIT-1) ); + + + if( limitId<0 || limitId>=SQLITE_N_LIMIT ){ + return -1; + } + oldLimit = db->aLimit[limitId]; + if( newLimit>=0 ){ /* IMP: R-52476-28732 */ + if( newLimit>aHardLimit[limitId] ){ + newLimit = aHardLimit[limitId]; /* IMP: R-51463-25634 */ + } + db->aLimit[limitId] = newLimit; + } + return oldLimit; /* IMP: R-53341-35419 */ +} + +/* +** This function is used to parse both URIs and non-URI filenames passed by the +** user to API functions sqlite3_open() or sqlite3_open_v2(), and for database +** URIs specified as part of ATTACH statements. +** +** The first argument to this function is the name of the VFS to use (or +** a NULL to signify the default VFS) if the URI does not contain a "vfs=xxx" +** query parameter. The second argument contains the URI (or non-URI filename) +** itself. When this function is called the *pFlags variable should contain +** the default flags to open the database handle with. The value stored in +** *pFlags may be updated before returning if the URI filename contains +** "cache=xxx" or "mode=xxx" query parameters. +** +** If successful, SQLITE_OK is returned. In this case *ppVfs is set to point to +** the VFS that should be used to open the database file. *pzFile is set to +** point to a buffer containing the name of the file to open. It is the +** responsibility of the caller to eventually call sqlite3_free() to release +** this buffer. +** +** If an error occurs, then an SQLite error code is returned and *pzErrMsg +** may be set to point to a buffer containing an English language error +** message. It is the responsibility of the caller to eventually release +** this buffer by calling sqlite3_free(). +*/ +SQLITE_PRIVATE int sqlite3ParseUri( + const char *zDefaultVfs, /* VFS to use if no "vfs=xxx" query option */ + const char *zUri, /* Nul-terminated URI to parse */ + unsigned int *pFlags, /* IN/OUT: SQLITE_OPEN_XXX flags */ + sqlite3_vfs **ppVfs, /* OUT: VFS to use */ + char **pzFile, /* OUT: Filename component of URI */ + char **pzErrMsg /* OUT: Error message (if rc!=SQLITE_OK) */ +){ + int rc = SQLITE_OK; + unsigned int flags = *pFlags; + const char *zVfs = zDefaultVfs; + char *zFile; + char c; + int nUri = sqlite3Strlen30(zUri); + + assert( *pzErrMsg==0 ); + + if( ((flags & SQLITE_OPEN_URI) /* IMP: R-48725-32206 */ + || sqlite3GlobalConfig.bOpenUri) /* IMP: R-51689-46548 */ + && nUri>=5 && memcmp(zUri, "file:", 5)==0 /* IMP: R-57884-37496 */ + ){ + char *zOpt; + int eState; /* Parser state when parsing URI */ + int iIn; /* Input character index */ + int iOut = 0; /* Output character index */ + u64 nByte = nUri+2; /* Bytes of space to allocate */ + + /* Make sure the SQLITE_OPEN_URI flag is set to indicate to the VFS xOpen + ** method that there may be extra parameters following the file-name. */ + flags |= SQLITE_OPEN_URI; + + for(iIn=0; iIn=0 && octet<256 ); + if( octet==0 ){ +#ifndef SQLITE_ENABLE_URI_00_ERROR + /* This branch is taken when "%00" appears within the URI. In this + ** case we ignore all text in the remainder of the path, name or + ** value currently being parsed. So ignore the current character + ** and skip to the next "?", "=" or "&", as appropriate. */ + while( (c = zUri[iIn])!=0 && c!='#' + && (eState!=0 || c!='?') + && (eState!=1 || (c!='=' && c!='&')) + && (eState!=2 || c!='&') + ){ + iIn++; + } + continue; +#else + /* If ENABLE_URI_00_ERROR is defined, "%00" in a URI is an error. */ + *pzErrMsg = sqlite3_mprintf("unexpected %%00 in uri"); + rc = SQLITE_ERROR; + goto parse_uri_out; +#endif + } + c = octet; + }else if( eState==1 && (c=='&' || c=='=') ){ + if( zFile[iOut-1]==0 ){ + /* An empty option name. Ignore this option altogether. */ + while( zUri[iIn] && zUri[iIn]!='#' && zUri[iIn-1]!='&' ) iIn++; + continue; + } + if( c=='&' ){ + zFile[iOut++] = '\0'; + }else{ + eState = 2; + } + c = 0; + }else if( (eState==0 && c=='?') || (eState==2 && c=='&') ){ + c = 0; + eState = 1; + } + zFile[iOut++] = c; + } + if( eState==1 ) zFile[iOut++] = '\0'; + zFile[iOut++] = '\0'; + zFile[iOut++] = '\0'; + + /* Check if there were any options specified that should be interpreted + ** here. Options that are interpreted here include "vfs" and those that + ** correspond to flags that may be passed to the sqlite3_open_v2() + ** method. */ + zOpt = &zFile[sqlite3Strlen30(zFile)+1]; + while( zOpt[0] ){ + int nOpt = sqlite3Strlen30(zOpt); + char *zVal = &zOpt[nOpt+1]; + int nVal = sqlite3Strlen30(zVal); + + if( nOpt==3 && memcmp("vfs", zOpt, 3)==0 ){ + zVfs = zVal; + }else{ + struct OpenMode { + const char *z; + int mode; + } *aMode = 0; + char *zModeType = 0; + int mask = 0; + int limit = 0; + + if( nOpt==5 && memcmp("cache", zOpt, 5)==0 ){ + static struct OpenMode aCacheMode[] = { + { "shared", SQLITE_OPEN_SHAREDCACHE }, + { "private", SQLITE_OPEN_PRIVATECACHE }, + { 0, 0 } + }; + + mask = SQLITE_OPEN_SHAREDCACHE|SQLITE_OPEN_PRIVATECACHE; + aMode = aCacheMode; + limit = mask; + zModeType = "cache"; + } + if( nOpt==4 && memcmp("mode", zOpt, 4)==0 ){ + static struct OpenMode aOpenMode[] = { + { "ro", SQLITE_OPEN_READONLY }, + { "rw", SQLITE_OPEN_READWRITE }, + { "rwc", SQLITE_OPEN_READWRITE | SQLITE_OPEN_CREATE }, + { "memory", SQLITE_OPEN_MEMORY }, + { 0, 0 } + }; + + mask = SQLITE_OPEN_READONLY | SQLITE_OPEN_READWRITE + | SQLITE_OPEN_CREATE | SQLITE_OPEN_MEMORY; + aMode = aOpenMode; + limit = mask & flags; + zModeType = "access"; + } + + if( aMode ){ + int i; + int mode = 0; + for(i=0; aMode[i].z; i++){ + const char *z = aMode[i].z; + if( nVal==sqlite3Strlen30(z) && 0==memcmp(zVal, z, nVal) ){ + mode = aMode[i].mode; + break; + } + } + if( mode==0 ){ + *pzErrMsg = sqlite3_mprintf("no such %s mode: %s", zModeType, zVal); + rc = SQLITE_ERROR; + goto parse_uri_out; + } + if( (mode & ~SQLITE_OPEN_MEMORY)>limit ){ + *pzErrMsg = sqlite3_mprintf("%s mode not allowed: %s", + zModeType, zVal); + rc = SQLITE_PERM; + goto parse_uri_out; + } + flags = (flags & ~mask) | mode; + } + } + + zOpt = &zVal[nVal+1]; + } + + }else{ + zFile = sqlite3_malloc64(nUri+2); + if( !zFile ) return SQLITE_NOMEM_BKPT; + if( nUri ){ + memcpy(zFile, zUri, nUri); + } + zFile[nUri] = '\0'; + zFile[nUri+1] = '\0'; + flags &= ~SQLITE_OPEN_URI; + } + + *ppVfs = sqlite3_vfs_find(zVfs); + if( *ppVfs==0 ){ + *pzErrMsg = sqlite3_mprintf("no such vfs: %s", zVfs); + rc = SQLITE_ERROR; + } + parse_uri_out: + if( rc!=SQLITE_OK ){ + sqlite3_free(zFile); + zFile = 0; + } + *pFlags = flags; + *pzFile = zFile; + return rc; +} + +#if defined(SQLITE_HAS_CODEC) +/* +** Process URI filename query parameters relevant to the SQLite Encryption +** Extension. Return true if any of the relevant query parameters are +** seen and return false if not. +*/ +SQLITE_PRIVATE int sqlite3CodecQueryParameters( + sqlite3 *db, /* Database connection */ + const char *zDb, /* Which schema is being created/attached */ + const char *zUri /* URI filename */ +){ + const char *zKey; + if( (zKey = sqlite3_uri_parameter(zUri, "hexkey"))!=0 && zKey[0] ){ + u8 iByte; + int i; + char zDecoded[40]; + for(i=0, iByte=0; imutex = sqlite3MutexAlloc(SQLITE_MUTEX_RECURSIVE); + if( db->mutex==0 ){ + sqlite3_free(db); + db = 0; + goto opendb_out; + } + if( isThreadsafe==0 ){ + sqlite3MutexWarnOnContention(db->mutex); + } + } + sqlite3_mutex_enter(db->mutex); + db->errMask = 0xff; + db->nDb = 2; + db->magic = SQLITE_MAGIC_BUSY; + db->aDb = db->aDbStatic; + db->lookaside.bDisable = 1; + + assert( sizeof(db->aLimit)==sizeof(aHardLimit) ); + memcpy(db->aLimit, aHardLimit, sizeof(db->aLimit)); + db->aLimit[SQLITE_LIMIT_WORKER_THREADS] = SQLITE_DEFAULT_WORKER_THREADS; + db->autoCommit = 1; + db->nextAutovac = -1; + db->szMmap = sqlite3GlobalConfig.szMmap; + db->nextPagesize = 0; + db->nMaxSorterMmap = 0x7FFFFFFF; + db->flags |= SQLITE_ShortColNames + | SQLITE_EnableTrigger + | SQLITE_EnableView + | SQLITE_CacheSpill + +/* The SQLITE_DQS compile-time option determines the default settings +** for SQLITE_DBCONFIG_DQS_DDL and SQLITE_DBCONFIG_DQS_DML. +** +** SQLITE_DQS SQLITE_DBCONFIG_DQS_DDL SQLITE_DBCONFIG_DQS_DML +** ---------- ----------------------- ----------------------- +** undefined on on +** 3 on on +** 2 on off +** 1 off on +** 0 off off +** +** Legacy behavior is 3 (double-quoted string literals are allowed anywhere) +** and so that is the default. But developers are encouranged to use +** -DSQLITE_DQS=0 (best) or -DSQLITE_DQS=1 (second choice) if possible. +*/ +#if !defined(SQLITE_DQS) +# define SQLITE_DQS 3 +#endif +#if (SQLITE_DQS&1)==1 + | SQLITE_DqsDML +#endif +#if (SQLITE_DQS&2)==2 + | SQLITE_DqsDDL +#endif + +#if !defined(SQLITE_DEFAULT_AUTOMATIC_INDEX) || SQLITE_DEFAULT_AUTOMATIC_INDEX + | SQLITE_AutoIndex +#endif +#if SQLITE_DEFAULT_CKPTFULLFSYNC + | SQLITE_CkptFullFSync +#endif +#if SQLITE_DEFAULT_FILE_FORMAT<4 + | SQLITE_LegacyFileFmt +#endif +#ifdef SQLITE_ENABLE_LOAD_EXTENSION + | SQLITE_LoadExtension +#endif +#if SQLITE_DEFAULT_RECURSIVE_TRIGGERS + | SQLITE_RecTriggers +#endif +#if defined(SQLITE_DEFAULT_FOREIGN_KEYS) && SQLITE_DEFAULT_FOREIGN_KEYS + | SQLITE_ForeignKeys +#endif +#if defined(SQLITE_REVERSE_UNORDERED_SELECTS) + | SQLITE_ReverseOrder +#endif +#if defined(SQLITE_ENABLE_OVERSIZE_CELL_CHECK) + | SQLITE_CellSizeCk +#endif +#if defined(SQLITE_ENABLE_FTS3_TOKENIZER) + | SQLITE_Fts3Tokenizer +#endif +#if defined(SQLITE_ENABLE_QPSG) + | SQLITE_EnableQPSG +#endif +#if defined(SQLITE_DEFAULT_DEFENSIVE) + | SQLITE_Defensive +#endif + ; + sqlite3HashInit(&db->aCollSeq); +#ifndef SQLITE_OMIT_VIRTUALTABLE + sqlite3HashInit(&db->aModule); +#endif + + /* Add the default collation sequence BINARY. BINARY works for both UTF-8 + ** and UTF-16, so add a version for each to avoid any unnecessary + ** conversions. The only error that can occur here is a malloc() failure. + ** + ** EVIDENCE-OF: R-52786-44878 SQLite defines three built-in collating + ** functions: + */ + createCollation(db, sqlite3StrBINARY, SQLITE_UTF8, 0, binCollFunc, 0); + createCollation(db, sqlite3StrBINARY, SQLITE_UTF16BE, 0, binCollFunc, 0); + createCollation(db, sqlite3StrBINARY, SQLITE_UTF16LE, 0, binCollFunc, 0); + createCollation(db, "NOCASE", SQLITE_UTF8, 0, nocaseCollatingFunc, 0); + createCollation(db, "RTRIM", SQLITE_UTF8, 0, rtrimCollFunc, 0); + if( db->mallocFailed ){ + goto opendb_out; + } + /* EVIDENCE-OF: R-08308-17224 The default collating function for all + ** strings is BINARY. + */ + db->pDfltColl = sqlite3FindCollSeq(db, SQLITE_UTF8, sqlite3StrBINARY, 0); + assert( db->pDfltColl!=0 ); + + /* Parse the filename/URI argument + ** + ** Only allow sensible combinations of bits in the flags argument. + ** Throw an error if any non-sense combination is used. If we + ** do not block illegal combinations here, it could trigger + ** assert() statements in deeper layers. Sensible combinations + ** are: + ** + ** 1: SQLITE_OPEN_READONLY + ** 2: SQLITE_OPEN_READWRITE + ** 6: SQLITE_OPEN_READWRITE | SQLITE_OPEN_CREATE + */ + db->openFlags = flags; + assert( SQLITE_OPEN_READONLY == 0x01 ); + assert( SQLITE_OPEN_READWRITE == 0x02 ); + assert( SQLITE_OPEN_CREATE == 0x04 ); + testcase( (1<<(flags&7))==0x02 ); /* READONLY */ + testcase( (1<<(flags&7))==0x04 ); /* READWRITE */ + testcase( (1<<(flags&7))==0x40 ); /* READWRITE | CREATE */ + if( ((1<<(flags&7)) & 0x46)==0 ){ + rc = SQLITE_MISUSE_BKPT; /* IMP: R-65497-44594 */ + }else{ + rc = sqlite3ParseUri(zVfs, zFilename, &flags, &db->pVfs, &zOpen, &zErrMsg); + } + if( rc!=SQLITE_OK ){ + if( rc==SQLITE_NOMEM ) sqlite3OomFault(db); + sqlite3ErrorWithMsg(db, rc, zErrMsg ? "%s" : 0, zErrMsg); + sqlite3_free(zErrMsg); + goto opendb_out; + } + + /* Open the backend database driver */ + rc = sqlite3BtreeOpen(db->pVfs, zOpen, db, &db->aDb[0].pBt, 0, + flags | SQLITE_OPEN_MAIN_DB); + if( rc!=SQLITE_OK ){ + if( rc==SQLITE_IOERR_NOMEM ){ + rc = SQLITE_NOMEM_BKPT; + } + sqlite3Error(db, rc); + goto opendb_out; + } + sqlite3BtreeEnter(db->aDb[0].pBt); + db->aDb[0].pSchema = sqlite3SchemaGet(db, db->aDb[0].pBt); + if( !db->mallocFailed ) ENC(db) = SCHEMA_ENC(db); + sqlite3BtreeLeave(db->aDb[0].pBt); + db->aDb[1].pSchema = sqlite3SchemaGet(db, 0); + + /* The default safety_level for the main database is FULL; for the temp + ** database it is OFF. This matches the pager layer defaults. + */ + db->aDb[0].zDbSName = "main"; + db->aDb[0].safety_level = SQLITE_DEFAULT_SYNCHRONOUS+1; + db->aDb[1].zDbSName = "temp"; + db->aDb[1].safety_level = PAGER_SYNCHRONOUS_OFF; + + db->magic = SQLITE_MAGIC_OPEN; + if( db->mallocFailed ){ + goto opendb_out; + } + + /* Register all built-in functions, but do not attempt to read the + ** database schema yet. This is delayed until the first time the database + ** is accessed. + */ + sqlite3Error(db, SQLITE_OK); + sqlite3RegisterPerConnectionBuiltinFunctions(db); + rc = sqlite3_errcode(db); + +#ifdef SQLITE_ENABLE_FTS5 + /* Register any built-in FTS5 module before loading the automatic + ** extensions. This allows automatic extensions to register FTS5 + ** tokenizers and auxiliary functions. */ + if( !db->mallocFailed && rc==SQLITE_OK ){ + rc = sqlite3Fts5Init(db); + } +#endif + + /* Load automatic extensions - extensions that have been registered + ** using the sqlite3_automatic_extension() API. + */ + if( rc==SQLITE_OK ){ + sqlite3AutoLoadExtensions(db); + rc = sqlite3_errcode(db); + if( rc!=SQLITE_OK ){ + goto opendb_out; + } + } + +#ifdef SQLITE_ENABLE_FTS1 + if( !db->mallocFailed ){ + extern int sqlite3Fts1Init(sqlite3*); + rc = sqlite3Fts1Init(db); + } +#endif + +#ifdef SQLITE_ENABLE_FTS2 + if( !db->mallocFailed && rc==SQLITE_OK ){ + extern int sqlite3Fts2Init(sqlite3*); + rc = sqlite3Fts2Init(db); + } +#endif + +#ifdef SQLITE_ENABLE_FTS3 /* automatically defined by SQLITE_ENABLE_FTS4 */ + if( !db->mallocFailed && rc==SQLITE_OK ){ + rc = sqlite3Fts3Init(db); + } +#endif + +#if defined(SQLITE_ENABLE_ICU) || defined(SQLITE_ENABLE_ICU_COLLATIONS) + if( !db->mallocFailed && rc==SQLITE_OK ){ + rc = sqlite3IcuInit(db); + } +#endif + +#ifdef SQLITE_ENABLE_RTREE + if( !db->mallocFailed && rc==SQLITE_OK){ + rc = sqlite3RtreeInit(db); + } +#endif + +#ifdef SQLITE_ENABLE_DBPAGE_VTAB + if( !db->mallocFailed && rc==SQLITE_OK){ + rc = sqlite3DbpageRegister(db); + } +#endif + +#ifdef SQLITE_ENABLE_DBSTAT_VTAB + if( !db->mallocFailed && rc==SQLITE_OK){ + rc = sqlite3DbstatRegister(db); + } +#endif + +#ifdef SQLITE_ENABLE_JSON1 + if( !db->mallocFailed && rc==SQLITE_OK){ + rc = sqlite3Json1Init(db); + } +#endif + +#ifdef SQLITE_ENABLE_STMTVTAB + if( !db->mallocFailed && rc==SQLITE_OK){ + rc = sqlite3StmtVtabInit(db); + } +#endif + + /* -DSQLITE_DEFAULT_LOCKING_MODE=1 makes EXCLUSIVE the default locking + ** mode. -DSQLITE_DEFAULT_LOCKING_MODE=0 make NORMAL the default locking + ** mode. Doing nothing at all also makes NORMAL the default. + */ +#ifdef SQLITE_DEFAULT_LOCKING_MODE + db->dfltLockMode = SQLITE_DEFAULT_LOCKING_MODE; + sqlite3PagerLockingMode(sqlite3BtreePager(db->aDb[0].pBt), + SQLITE_DEFAULT_LOCKING_MODE); +#endif + + if( rc ) sqlite3Error(db, rc); + + /* Enable the lookaside-malloc subsystem */ + setupLookaside(db, 0, sqlite3GlobalConfig.szLookaside, + sqlite3GlobalConfig.nLookaside); + + sqlite3_wal_autocheckpoint(db, SQLITE_DEFAULT_WAL_AUTOCHECKPOINT); + +opendb_out: + if( db ){ + assert( db->mutex!=0 || isThreadsafe==0 + || sqlite3GlobalConfig.bFullMutex==0 ); + sqlite3_mutex_leave(db->mutex); + } + rc = sqlite3_errcode(db); + assert( db!=0 || rc==SQLITE_NOMEM ); + if( rc==SQLITE_NOMEM ){ + sqlite3_close(db); + db = 0; + }else if( rc!=SQLITE_OK ){ + db->magic = SQLITE_MAGIC_SICK; + } + *ppDb = db; +#ifdef SQLITE_ENABLE_SQLLOG + if( sqlite3GlobalConfig.xSqllog ){ + /* Opening a db handle. Fourth parameter is passed 0. */ + void *pArg = sqlite3GlobalConfig.pSqllogArg; + sqlite3GlobalConfig.xSqllog(pArg, db, zFilename, 0); + } +#endif +#if defined(SQLITE_HAS_CODEC) + if( rc==SQLITE_OK ) sqlite3CodecQueryParameters(db, 0, zOpen); +#endif + sqlite3_free(zOpen); + return rc & 0xff; +} + + +/* +** Open a new database handle. +*/ +SQLITE_API int sqlite3_open( + const char *zFilename, + sqlite3 **ppDb +){ + return openDatabase(zFilename, ppDb, + SQLITE_OPEN_READWRITE | SQLITE_OPEN_CREATE, 0); +} +SQLITE_API int sqlite3_open_v2( + const char *filename, /* Database filename (UTF-8) */ + sqlite3 **ppDb, /* OUT: SQLite db handle */ + int flags, /* Flags */ + const char *zVfs /* Name of VFS module to use */ +){ + return openDatabase(filename, ppDb, (unsigned int)flags, zVfs); +} + +#ifndef SQLITE_OMIT_UTF16 +/* +** Open a new database handle. +*/ +SQLITE_API int sqlite3_open16( + const void *zFilename, + sqlite3 **ppDb +){ + char const *zFilename8; /* zFilename encoded in UTF-8 instead of UTF-16 */ + sqlite3_value *pVal; + int rc; + +#ifdef SQLITE_ENABLE_API_ARMOR + if( ppDb==0 ) return SQLITE_MISUSE_BKPT; +#endif + *ppDb = 0; +#ifndef SQLITE_OMIT_AUTOINIT + rc = sqlite3_initialize(); + if( rc ) return rc; +#endif + if( zFilename==0 ) zFilename = "\000\000"; + pVal = sqlite3ValueNew(0); + sqlite3ValueSetStr(pVal, -1, zFilename, SQLITE_UTF16NATIVE, SQLITE_STATIC); + zFilename8 = sqlite3ValueText(pVal, SQLITE_UTF8); + if( zFilename8 ){ + rc = openDatabase(zFilename8, ppDb, + SQLITE_OPEN_READWRITE | SQLITE_OPEN_CREATE, 0); + assert( *ppDb || rc==SQLITE_NOMEM ); + if( rc==SQLITE_OK && !DbHasProperty(*ppDb, 0, DB_SchemaLoaded) ){ + SCHEMA_ENC(*ppDb) = ENC(*ppDb) = SQLITE_UTF16NATIVE; + } + }else{ + rc = SQLITE_NOMEM_BKPT; + } + sqlite3ValueFree(pVal); + + return rc & 0xff; +} +#endif /* SQLITE_OMIT_UTF16 */ + +/* +** Register a new collation sequence with the database handle db. +*/ +SQLITE_API int sqlite3_create_collation( + sqlite3* db, + const char *zName, + int enc, + void* pCtx, + int(*xCompare)(void*,int,const void*,int,const void*) +){ + return sqlite3_create_collation_v2(db, zName, enc, pCtx, xCompare, 0); +} + +/* +** Register a new collation sequence with the database handle db. +*/ +SQLITE_API int sqlite3_create_collation_v2( + sqlite3* db, + const char *zName, + int enc, + void* pCtx, + int(*xCompare)(void*,int,const void*,int,const void*), + void(*xDel)(void*) +){ + int rc; + +#ifdef SQLITE_ENABLE_API_ARMOR + if( !sqlite3SafetyCheckOk(db) || zName==0 ) return SQLITE_MISUSE_BKPT; +#endif + sqlite3_mutex_enter(db->mutex); + assert( !db->mallocFailed ); + rc = createCollation(db, zName, (u8)enc, pCtx, xCompare, xDel); + rc = sqlite3ApiExit(db, rc); + sqlite3_mutex_leave(db->mutex); + return rc; +} + +#ifndef SQLITE_OMIT_UTF16 +/* +** Register a new collation sequence with the database handle db. +*/ +SQLITE_API int sqlite3_create_collation16( + sqlite3* db, + const void *zName, + int enc, + void* pCtx, + int(*xCompare)(void*,int,const void*,int,const void*) +){ + int rc = SQLITE_OK; + char *zName8; + +#ifdef SQLITE_ENABLE_API_ARMOR + if( !sqlite3SafetyCheckOk(db) || zName==0 ) return SQLITE_MISUSE_BKPT; +#endif + sqlite3_mutex_enter(db->mutex); + assert( !db->mallocFailed ); + zName8 = sqlite3Utf16to8(db, zName, -1, SQLITE_UTF16NATIVE); + if( zName8 ){ + rc = createCollation(db, zName8, (u8)enc, pCtx, xCompare, 0); + sqlite3DbFree(db, zName8); + } + rc = sqlite3ApiExit(db, rc); + sqlite3_mutex_leave(db->mutex); + return rc; +} +#endif /* SQLITE_OMIT_UTF16 */ + +/* +** Register a collation sequence factory callback with the database handle +** db. Replace any previously installed collation sequence factory. +*/ +SQLITE_API int sqlite3_collation_needed( + sqlite3 *db, + void *pCollNeededArg, + void(*xCollNeeded)(void*,sqlite3*,int eTextRep,const char*) +){ +#ifdef SQLITE_ENABLE_API_ARMOR + if( !sqlite3SafetyCheckOk(db) ) return SQLITE_MISUSE_BKPT; +#endif + sqlite3_mutex_enter(db->mutex); + db->xCollNeeded = xCollNeeded; + db->xCollNeeded16 = 0; + db->pCollNeededArg = pCollNeededArg; + sqlite3_mutex_leave(db->mutex); + return SQLITE_OK; +} + +#ifndef SQLITE_OMIT_UTF16 +/* +** Register a collation sequence factory callback with the database handle +** db. Replace any previously installed collation sequence factory. +*/ +SQLITE_API int sqlite3_collation_needed16( + sqlite3 *db, + void *pCollNeededArg, + void(*xCollNeeded16)(void*,sqlite3*,int eTextRep,const void*) +){ +#ifdef SQLITE_ENABLE_API_ARMOR + if( !sqlite3SafetyCheckOk(db) ) return SQLITE_MISUSE_BKPT; +#endif + sqlite3_mutex_enter(db->mutex); + db->xCollNeeded = 0; + db->xCollNeeded16 = xCollNeeded16; + db->pCollNeededArg = pCollNeededArg; + sqlite3_mutex_leave(db->mutex); + return SQLITE_OK; +} +#endif /* SQLITE_OMIT_UTF16 */ + +#ifndef SQLITE_OMIT_DEPRECATED +/* +** This function is now an anachronism. It used to be used to recover from a +** malloc() failure, but SQLite now does this automatically. +*/ +SQLITE_API int sqlite3_global_recover(void){ + return SQLITE_OK; +} +#endif + +/* +** Test to see whether or not the database connection is in autocommit +** mode. Return TRUE if it is and FALSE if not. Autocommit mode is on +** by default. Autocommit is disabled by a BEGIN statement and reenabled +** by the next COMMIT or ROLLBACK. +*/ +SQLITE_API int sqlite3_get_autocommit(sqlite3 *db){ +#ifdef SQLITE_ENABLE_API_ARMOR + if( !sqlite3SafetyCheckOk(db) ){ + (void)SQLITE_MISUSE_BKPT; + return 0; + } +#endif + return db->autoCommit; +} + +/* +** The following routines are substitutes for constants SQLITE_CORRUPT, +** SQLITE_MISUSE, SQLITE_CANTOPEN, SQLITE_NOMEM and possibly other error +** constants. They serve two purposes: +** +** 1. Serve as a convenient place to set a breakpoint in a debugger +** to detect when version error conditions occurs. +** +** 2. Invoke sqlite3_log() to provide the source code location where +** a low-level error is first detected. +*/ +SQLITE_PRIVATE int sqlite3ReportError(int iErr, int lineno, const char *zType){ + sqlite3_log(iErr, "%s at line %d of [%.10s]", + zType, lineno, 20+sqlite3_sourceid()); + return iErr; +} +SQLITE_PRIVATE int sqlite3CorruptError(int lineno){ + testcase( sqlite3GlobalConfig.xLog!=0 ); + return sqlite3ReportError(SQLITE_CORRUPT, lineno, "database corruption"); +} +SQLITE_PRIVATE int sqlite3MisuseError(int lineno){ + testcase( sqlite3GlobalConfig.xLog!=0 ); + return sqlite3ReportError(SQLITE_MISUSE, lineno, "misuse"); +} +SQLITE_PRIVATE int sqlite3CantopenError(int lineno){ + testcase( sqlite3GlobalConfig.xLog!=0 ); + return sqlite3ReportError(SQLITE_CANTOPEN, lineno, "cannot open file"); +} +#ifdef SQLITE_DEBUG +SQLITE_PRIVATE int sqlite3CorruptPgnoError(int lineno, Pgno pgno){ + char zMsg[100]; + sqlite3_snprintf(sizeof(zMsg), zMsg, "database corruption page %d", pgno); + testcase( sqlite3GlobalConfig.xLog!=0 ); + return sqlite3ReportError(SQLITE_CORRUPT, lineno, zMsg); +} +SQLITE_PRIVATE int sqlite3NomemError(int lineno){ + testcase( sqlite3GlobalConfig.xLog!=0 ); + return sqlite3ReportError(SQLITE_NOMEM, lineno, "OOM"); +} +SQLITE_PRIVATE int sqlite3IoerrnomemError(int lineno){ + testcase( sqlite3GlobalConfig.xLog!=0 ); + return sqlite3ReportError(SQLITE_IOERR_NOMEM, lineno, "I/O OOM error"); +} +#endif + +#ifndef SQLITE_OMIT_DEPRECATED +/* +** This is a convenience routine that makes sure that all thread-specific +** data for this thread has been deallocated. +** +** SQLite no longer uses thread-specific data so this routine is now a +** no-op. It is retained for historical compatibility. +*/ +SQLITE_API void sqlite3_thread_cleanup(void){ +} +#endif + +/* +** Return meta information about a specific column of a database table. +** See comment in sqlite3.h (sqlite.h.in) for details. +*/ +SQLITE_API int sqlite3_table_column_metadata( + sqlite3 *db, /* Connection handle */ + const char *zDbName, /* Database name or NULL */ + const char *zTableName, /* Table name */ + const char *zColumnName, /* Column name */ + char const **pzDataType, /* OUTPUT: Declared data type */ + char const **pzCollSeq, /* OUTPUT: Collation sequence name */ + int *pNotNull, /* OUTPUT: True if NOT NULL constraint exists */ + int *pPrimaryKey, /* OUTPUT: True if column part of PK */ + int *pAutoinc /* OUTPUT: True if column is auto-increment */ +){ + int rc; + char *zErrMsg = 0; + Table *pTab = 0; + Column *pCol = 0; + int iCol = 0; + char const *zDataType = 0; + char const *zCollSeq = 0; + int notnull = 0; + int primarykey = 0; + int autoinc = 0; + + +#ifdef SQLITE_ENABLE_API_ARMOR + if( !sqlite3SafetyCheckOk(db) || zTableName==0 ){ + return SQLITE_MISUSE_BKPT; + } +#endif + + /* Ensure the database schema has been loaded */ + sqlite3_mutex_enter(db->mutex); + sqlite3BtreeEnterAll(db); + rc = sqlite3Init(db, &zErrMsg); + if( SQLITE_OK!=rc ){ + goto error_out; + } + + /* Locate the table in question */ + pTab = sqlite3FindTable(db, zTableName, zDbName); + if( !pTab || pTab->pSelect ){ + pTab = 0; + goto error_out; + } + + /* Find the column for which info is requested */ + if( zColumnName==0 ){ + /* Query for existance of table only */ + }else{ + for(iCol=0; iColnCol; iCol++){ + pCol = &pTab->aCol[iCol]; + if( 0==sqlite3StrICmp(pCol->zName, zColumnName) ){ + break; + } + } + if( iCol==pTab->nCol ){ + if( HasRowid(pTab) && sqlite3IsRowid(zColumnName) ){ + iCol = pTab->iPKey; + pCol = iCol>=0 ? &pTab->aCol[iCol] : 0; + }else{ + pTab = 0; + goto error_out; + } + } + } + + /* The following block stores the meta information that will be returned + ** to the caller in local variables zDataType, zCollSeq, notnull, primarykey + ** and autoinc. At this point there are two possibilities: + ** + ** 1. The specified column name was rowid", "oid" or "_rowid_" + ** and there is no explicitly declared IPK column. + ** + ** 2. The table is not a view and the column name identified an + ** explicitly declared column. Copy meta information from *pCol. + */ + if( pCol ){ + zDataType = sqlite3ColumnType(pCol,0); + zCollSeq = pCol->zColl; + notnull = pCol->notNull!=0; + primarykey = (pCol->colFlags & COLFLAG_PRIMKEY)!=0; + autoinc = pTab->iPKey==iCol && (pTab->tabFlags & TF_Autoincrement)!=0; + }else{ + zDataType = "INTEGER"; + primarykey = 1; + } + if( !zCollSeq ){ + zCollSeq = sqlite3StrBINARY; + } + +error_out: + sqlite3BtreeLeaveAll(db); + + /* Whether the function call succeeded or failed, set the output parameters + ** to whatever their local counterparts contain. If an error did occur, + ** this has the effect of zeroing all output parameters. + */ + if( pzDataType ) *pzDataType = zDataType; + if( pzCollSeq ) *pzCollSeq = zCollSeq; + if( pNotNull ) *pNotNull = notnull; + if( pPrimaryKey ) *pPrimaryKey = primarykey; + if( pAutoinc ) *pAutoinc = autoinc; + + if( SQLITE_OK==rc && !pTab ){ + sqlite3DbFree(db, zErrMsg); + zErrMsg = sqlite3MPrintf(db, "no such table column: %s.%s", zTableName, + zColumnName); + rc = SQLITE_ERROR; + } + sqlite3ErrorWithMsg(db, rc, (zErrMsg?"%s":0), zErrMsg); + sqlite3DbFree(db, zErrMsg); + rc = sqlite3ApiExit(db, rc); + sqlite3_mutex_leave(db->mutex); + return rc; +} + +/* +** Sleep for a little while. Return the amount of time slept. +*/ +SQLITE_API int sqlite3_sleep(int ms){ + sqlite3_vfs *pVfs; + int rc; + pVfs = sqlite3_vfs_find(0); + if( pVfs==0 ) return 0; + + /* This function works in milliseconds, but the underlying OsSleep() + ** API uses microseconds. Hence the 1000's. + */ + rc = (sqlite3OsSleep(pVfs, 1000*ms)/1000); + return rc; +} + +/* +** Enable or disable the extended result codes. +*/ +SQLITE_API int sqlite3_extended_result_codes(sqlite3 *db, int onoff){ +#ifdef SQLITE_ENABLE_API_ARMOR + if( !sqlite3SafetyCheckOk(db) ) return SQLITE_MISUSE_BKPT; +#endif + sqlite3_mutex_enter(db->mutex); + db->errMask = onoff ? 0xffffffff : 0xff; + sqlite3_mutex_leave(db->mutex); + return SQLITE_OK; +} + +/* +** Invoke the xFileControl method on a particular database. +*/ +SQLITE_API int sqlite3_file_control(sqlite3 *db, const char *zDbName, int op, void *pArg){ + int rc = SQLITE_ERROR; + Btree *pBtree; + +#ifdef SQLITE_ENABLE_API_ARMOR + if( !sqlite3SafetyCheckOk(db) ) return SQLITE_MISUSE_BKPT; +#endif + sqlite3_mutex_enter(db->mutex); + pBtree = sqlite3DbNameToBtree(db, zDbName); + if( pBtree ){ + Pager *pPager; + sqlite3_file *fd; + sqlite3BtreeEnter(pBtree); + pPager = sqlite3BtreePager(pBtree); + assert( pPager!=0 ); + fd = sqlite3PagerFile(pPager); + assert( fd!=0 ); + if( op==SQLITE_FCNTL_FILE_POINTER ){ + *(sqlite3_file**)pArg = fd; + rc = SQLITE_OK; + }else if( op==SQLITE_FCNTL_VFS_POINTER ){ + *(sqlite3_vfs**)pArg = sqlite3PagerVfs(pPager); + rc = SQLITE_OK; + }else if( op==SQLITE_FCNTL_JOURNAL_POINTER ){ + *(sqlite3_file**)pArg = sqlite3PagerJrnlFile(pPager); + rc = SQLITE_OK; + }else if( op==SQLITE_FCNTL_DATA_VERSION ){ + *(unsigned int*)pArg = sqlite3PagerDataVersion(pPager); + rc = SQLITE_OK; + }else{ + rc = sqlite3OsFileControl(fd, op, pArg); + } + sqlite3BtreeLeave(pBtree); + } + sqlite3_mutex_leave(db->mutex); + return rc; +} + +/* +** Interface to the testing logic. +*/ +SQLITE_API int sqlite3_test_control(int op, ...){ + int rc = 0; +#ifdef SQLITE_UNTESTABLE + UNUSED_PARAMETER(op); +#else + va_list ap; + va_start(ap, op); + switch( op ){ + + /* + ** Save the current state of the PRNG. + */ + case SQLITE_TESTCTRL_PRNG_SAVE: { + sqlite3PrngSaveState(); + break; + } + + /* + ** Restore the state of the PRNG to the last state saved using + ** PRNG_SAVE. If PRNG_SAVE has never before been called, then + ** this verb acts like PRNG_RESET. + */ + case SQLITE_TESTCTRL_PRNG_RESTORE: { + sqlite3PrngRestoreState(); + break; + } + + /* sqlite3_test_control(SQLITE_TESTCTRL_PRNG_SEED, int x, sqlite3 *db); + ** + ** Control the seed for the pseudo-random number generator (PRNG) that + ** is built into SQLite. Cases: + ** + ** x!=0 && db!=0 Seed the PRNG to the current value of the + ** schema cookie in the main database for db, or + ** x if the schema cookie is zero. This case + ** is convenient to use with database fuzzers + ** as it allows the fuzzer some control over the + ** the PRNG seed. + ** + ** x!=0 && db==0 Seed the PRNG to the value of x. + ** + ** x==0 && db==0 Revert to default behavior of using the + ** xRandomness method on the primary VFS. + ** + ** This test-control also resets the PRNG so that the new seed will + ** be used for the next call to sqlite3_randomness(). + */ + case SQLITE_TESTCTRL_PRNG_SEED: { + int x = va_arg(ap, int); + int y; + sqlite3 *db = va_arg(ap, sqlite3*); + assert( db==0 || db->aDb[0].pSchema!=0 ); + if( db && (y = db->aDb[0].pSchema->schema_cookie)!=0 ){ x = y; } + sqlite3Config.iPrngSeed = x; + sqlite3_randomness(0,0); + break; + } + + /* + ** sqlite3_test_control(BITVEC_TEST, size, program) + ** + ** Run a test against a Bitvec object of size. The program argument + ** is an array of integers that defines the test. Return -1 on a + ** memory allocation error, 0 on success, or non-zero for an error. + ** See the sqlite3BitvecBuiltinTest() for additional information. + */ + case SQLITE_TESTCTRL_BITVEC_TEST: { + int sz = va_arg(ap, int); + int *aProg = va_arg(ap, int*); + rc = sqlite3BitvecBuiltinTest(sz, aProg); + break; + } + + /* + ** sqlite3_test_control(FAULT_INSTALL, xCallback) + ** + ** Arrange to invoke xCallback() whenever sqlite3FaultSim() is called, + ** if xCallback is not NULL. + ** + ** As a test of the fault simulator mechanism itself, sqlite3FaultSim(0) + ** is called immediately after installing the new callback and the return + ** value from sqlite3FaultSim(0) becomes the return from + ** sqlite3_test_control(). + */ + case SQLITE_TESTCTRL_FAULT_INSTALL: { + /* MSVC is picky about pulling func ptrs from va lists. + ** http://support.microsoft.com/kb/47961 + ** sqlite3GlobalConfig.xTestCallback = va_arg(ap, int(*)(int)); + */ + typedef int(*TESTCALLBACKFUNC_t)(int); + sqlite3GlobalConfig.xTestCallback = va_arg(ap, TESTCALLBACKFUNC_t); + rc = sqlite3FaultSim(0); + break; + } + + /* + ** sqlite3_test_control(BENIGN_MALLOC_HOOKS, xBegin, xEnd) + ** + ** Register hooks to call to indicate which malloc() failures + ** are benign. + */ + case SQLITE_TESTCTRL_BENIGN_MALLOC_HOOKS: { + typedef void (*void_function)(void); + void_function xBenignBegin; + void_function xBenignEnd; + xBenignBegin = va_arg(ap, void_function); + xBenignEnd = va_arg(ap, void_function); + sqlite3BenignMallocHooks(xBenignBegin, xBenignEnd); + break; + } + + /* + ** sqlite3_test_control(SQLITE_TESTCTRL_PENDING_BYTE, unsigned int X) + ** + ** Set the PENDING byte to the value in the argument, if X>0. + ** Make no changes if X==0. Return the value of the pending byte + ** as it existing before this routine was called. + ** + ** IMPORTANT: Changing the PENDING byte from 0x40000000 results in + ** an incompatible database file format. Changing the PENDING byte + ** while any database connection is open results in undefined and + ** deleterious behavior. + */ + case SQLITE_TESTCTRL_PENDING_BYTE: { + rc = PENDING_BYTE; +#ifndef SQLITE_OMIT_WSD + { + unsigned int newVal = va_arg(ap, unsigned int); + if( newVal ) sqlite3PendingByte = newVal; + } +#endif + break; + } + + /* + ** sqlite3_test_control(SQLITE_TESTCTRL_ASSERT, int X) + ** + ** This action provides a run-time test to see whether or not + ** assert() was enabled at compile-time. If X is true and assert() + ** is enabled, then the return value is true. If X is true and + ** assert() is disabled, then the return value is zero. If X is + ** false and assert() is enabled, then the assertion fires and the + ** process aborts. If X is false and assert() is disabled, then the + ** return value is zero. + */ + case SQLITE_TESTCTRL_ASSERT: { + volatile int x = 0; + assert( /*side-effects-ok*/ (x = va_arg(ap,int))!=0 ); + rc = x; + break; + } + + + /* + ** sqlite3_test_control(SQLITE_TESTCTRL_ALWAYS, int X) + ** + ** This action provides a run-time test to see how the ALWAYS and + ** NEVER macros were defined at compile-time. + ** + ** The return value is ALWAYS(X) if X is true, or 0 if X is false. + ** + ** The recommended test is X==2. If the return value is 2, that means + ** ALWAYS() and NEVER() are both no-op pass-through macros, which is the + ** default setting. If the return value is 1, then ALWAYS() is either + ** hard-coded to true or else it asserts if its argument is false. + ** The first behavior (hard-coded to true) is the case if + ** SQLITE_TESTCTRL_ASSERT shows that assert() is disabled and the second + ** behavior (assert if the argument to ALWAYS() is false) is the case if + ** SQLITE_TESTCTRL_ASSERT shows that assert() is enabled. + ** + ** The run-time test procedure might look something like this: + ** + ** if( sqlite3_test_control(SQLITE_TESTCTRL_ALWAYS, 2)==2 ){ + ** // ALWAYS() and NEVER() are no-op pass-through macros + ** }else if( sqlite3_test_control(SQLITE_TESTCTRL_ASSERT, 1) ){ + ** // ALWAYS(x) asserts that x is true. NEVER(x) asserts x is false. + ** }else{ + ** // ALWAYS(x) is a constant 1. NEVER(x) is a constant 0. + ** } + */ + case SQLITE_TESTCTRL_ALWAYS: { + int x = va_arg(ap,int); + rc = x ? ALWAYS(x) : 0; + break; + } + + /* + ** sqlite3_test_control(SQLITE_TESTCTRL_BYTEORDER); + ** + ** The integer returned reveals the byte-order of the computer on which + ** SQLite is running: + ** + ** 1 big-endian, determined at run-time + ** 10 little-endian, determined at run-time + ** 432101 big-endian, determined at compile-time + ** 123410 little-endian, determined at compile-time + */ + case SQLITE_TESTCTRL_BYTEORDER: { + rc = SQLITE_BYTEORDER*100 + SQLITE_LITTLEENDIAN*10 + SQLITE_BIGENDIAN; + break; + } + + /* sqlite3_test_control(SQLITE_TESTCTRL_RESERVE, sqlite3 *db, int N) + ** + ** Set the nReserve size to N for the main database on the database + ** connection db. + */ + case SQLITE_TESTCTRL_RESERVE: { + sqlite3 *db = va_arg(ap, sqlite3*); + int x = va_arg(ap,int); + sqlite3_mutex_enter(db->mutex); + sqlite3BtreeSetPageSize(db->aDb[0].pBt, 0, x, 0); + sqlite3_mutex_leave(db->mutex); + break; + } + + /* sqlite3_test_control(SQLITE_TESTCTRL_OPTIMIZATIONS, sqlite3 *db, int N) + ** + ** Enable or disable various optimizations for testing purposes. The + ** argument N is a bitmask of optimizations to be disabled. For normal + ** operation N should be 0. The idea is that a test program (like the + ** SQL Logic Test or SLT test module) can run the same SQL multiple times + ** with various optimizations disabled to verify that the same answer + ** is obtained in every case. + */ + case SQLITE_TESTCTRL_OPTIMIZATIONS: { + sqlite3 *db = va_arg(ap, sqlite3*); + db->dbOptFlags = (u16)(va_arg(ap, int) & 0xffff); + break; + } + + /* sqlite3_test_control(SQLITE_TESTCTRL_LOCALTIME_FAULT, int onoff); + ** + ** If parameter onoff is non-zero, subsequent calls to localtime() + ** and its variants fail. If onoff is zero, undo this setting. + */ + case SQLITE_TESTCTRL_LOCALTIME_FAULT: { + sqlite3GlobalConfig.bLocaltimeFault = va_arg(ap, int); + break; + } + + /* sqlite3_test_control(SQLITE_TESTCTRL_INTERNAL_FUNCS, int onoff); + ** + ** If parameter onoff is non-zero, internal-use-only SQL functions + ** are visible to ordinary SQL. This is useful for testing but is + ** unsafe because invalid parameters to those internal-use-only functions + ** can result in crashes or segfaults. + */ + case SQLITE_TESTCTRL_INTERNAL_FUNCTIONS: { + sqlite3GlobalConfig.bInternalFunctions = va_arg(ap, int); + break; + } + + /* sqlite3_test_control(SQLITE_TESTCTRL_NEVER_CORRUPT, int); + ** + ** Set or clear a flag that indicates that the database file is always well- + ** formed and never corrupt. This flag is clear by default, indicating that + ** database files might have arbitrary corruption. Setting the flag during + ** testing causes certain assert() statements in the code to be activated + ** that demonstrat invariants on well-formed database files. + */ + case SQLITE_TESTCTRL_NEVER_CORRUPT: { + sqlite3GlobalConfig.neverCorrupt = va_arg(ap, int); + break; + } + + /* sqlite3_test_control(SQLITE_TESTCTRL_EXTRA_SCHEMA_CHECKS, int); + ** + ** Set or clear a flag that causes SQLite to verify that type, name, + ** and tbl_name fields of the sqlite_master table. This is normally + ** on, but it is sometimes useful to turn it off for testing. + */ + case SQLITE_TESTCTRL_EXTRA_SCHEMA_CHECKS: { + sqlite3GlobalConfig.bExtraSchemaChecks = va_arg(ap, int); + break; + } + + /* Set the threshold at which OP_Once counters reset back to zero. + ** By default this is 0x7ffffffe (over 2 billion), but that value is + ** too big to test in a reasonable amount of time, so this control is + ** provided to set a small and easily reachable reset value. + */ + case SQLITE_TESTCTRL_ONCE_RESET_THRESHOLD: { + sqlite3GlobalConfig.iOnceResetThreshold = va_arg(ap, int); + break; + } + + /* sqlite3_test_control(SQLITE_TESTCTRL_VDBE_COVERAGE, xCallback, ptr); + ** + ** Set the VDBE coverage callback function to xCallback with context + ** pointer ptr. + */ + case SQLITE_TESTCTRL_VDBE_COVERAGE: { +#ifdef SQLITE_VDBE_COVERAGE + typedef void (*branch_callback)(void*,unsigned int, + unsigned char,unsigned char); + sqlite3GlobalConfig.xVdbeBranch = va_arg(ap,branch_callback); + sqlite3GlobalConfig.pVdbeBranchArg = va_arg(ap,void*); +#endif + break; + } + + /* sqlite3_test_control(SQLITE_TESTCTRL_SORTER_MMAP, db, nMax); */ + case SQLITE_TESTCTRL_SORTER_MMAP: { + sqlite3 *db = va_arg(ap, sqlite3*); + db->nMaxSorterMmap = va_arg(ap, int); + break; + } + + /* sqlite3_test_control(SQLITE_TESTCTRL_ISINIT); + ** + ** Return SQLITE_OK if SQLite has been initialized and SQLITE_ERROR if + ** not. + */ + case SQLITE_TESTCTRL_ISINIT: { + if( sqlite3GlobalConfig.isInit==0 ) rc = SQLITE_ERROR; + break; + } + + /* sqlite3_test_control(SQLITE_TESTCTRL_IMPOSTER, db, dbName, onOff, tnum); + ** + ** This test control is used to create imposter tables. "db" is a pointer + ** to the database connection. dbName is the database name (ex: "main" or + ** "temp") which will receive the imposter. "onOff" turns imposter mode on + ** or off. "tnum" is the root page of the b-tree to which the imposter + ** table should connect. + ** + ** Enable imposter mode only when the schema has already been parsed. Then + ** run a single CREATE TABLE statement to construct the imposter table in + ** the parsed schema. Then turn imposter mode back off again. + ** + ** If onOff==0 and tnum>0 then reset the schema for all databases, causing + ** the schema to be reparsed the next time it is needed. This has the + ** effect of erasing all imposter tables. + */ + case SQLITE_TESTCTRL_IMPOSTER: { + sqlite3 *db = va_arg(ap, sqlite3*); + sqlite3_mutex_enter(db->mutex); + db->init.iDb = sqlite3FindDbName(db, va_arg(ap,const char*)); + db->init.busy = db->init.imposterTable = va_arg(ap,int); + db->init.newTnum = va_arg(ap,int); + if( db->init.busy==0 && db->init.newTnum>0 ){ + sqlite3ResetAllSchemasOfConnection(db); + } + sqlite3_mutex_leave(db->mutex); + break; + } + +#if defined(YYCOVERAGE) + /* sqlite3_test_control(SQLITE_TESTCTRL_PARSER_COVERAGE, FILE *out) + ** + ** This test control (only available when SQLite is compiled with + ** -DYYCOVERAGE) writes a report onto "out" that shows all + ** state/lookahead combinations in the parser state machine + ** which are never exercised. If any state is missed, make the + ** return code SQLITE_ERROR. + */ + case SQLITE_TESTCTRL_PARSER_COVERAGE: { + FILE *out = va_arg(ap, FILE*); + if( sqlite3ParserCoverage(out) ) rc = SQLITE_ERROR; + break; + } +#endif /* defined(YYCOVERAGE) */ + + /* sqlite3_test_control(SQLITE_TESTCTRL_RESULT_INTREAL, sqlite3_context*); + ** + ** This test-control causes the most recent sqlite3_result_int64() value + ** to be interpreted as a MEM_IntReal instead of as an MEM_Int. Normally, + ** MEM_IntReal values only arise during an INSERT operation of integer + ** values into a REAL column, so they can be challenging to test. This + ** test-control enables us to write an intreal() SQL function that can + ** inject an intreal() value at arbitrary places in an SQL statement, + ** for testing purposes. + */ + case SQLITE_TESTCTRL_RESULT_INTREAL: { + sqlite3_context *pCtx = va_arg(ap, sqlite3_context*); + sqlite3ResultIntReal(pCtx); + break; + } + } + va_end(ap); +#endif /* SQLITE_UNTESTABLE */ + return rc; +} + +/* +** This is a utility routine, useful to VFS implementations, that checks +** to see if a database file was a URI that contained a specific query +** parameter, and if so obtains the value of the query parameter. +** +** The zFilename argument is the filename pointer passed into the xOpen() +** method of a VFS implementation. The zParam argument is the name of the +** query parameter we seek. This routine returns the value of the zParam +** parameter if it exists. If the parameter does not exist, this routine +** returns a NULL pointer. +*/ +SQLITE_API const char *sqlite3_uri_parameter(const char *zFilename, const char *zParam){ + if( zFilename==0 || zParam==0 ) return 0; + zFilename += sqlite3Strlen30(zFilename) + 1; + while( zFilename[0] ){ + int x = strcmp(zFilename, zParam); + zFilename += sqlite3Strlen30(zFilename) + 1; + if( x==0 ) return zFilename; + zFilename += sqlite3Strlen30(zFilename) + 1; + } + return 0; +} + +/* +** Return a boolean value for a query parameter. +*/ +SQLITE_API int sqlite3_uri_boolean(const char *zFilename, const char *zParam, int bDflt){ + const char *z = sqlite3_uri_parameter(zFilename, zParam); + bDflt = bDflt!=0; + return z ? sqlite3GetBoolean(z, bDflt) : bDflt; +} + +/* +** Return a 64-bit integer value for a query parameter. +*/ +SQLITE_API sqlite3_int64 sqlite3_uri_int64( + const char *zFilename, /* Filename as passed to xOpen */ + const char *zParam, /* URI parameter sought */ + sqlite3_int64 bDflt /* return if parameter is missing */ +){ + const char *z = sqlite3_uri_parameter(zFilename, zParam); + sqlite3_int64 v; + if( z && sqlite3DecOrHexToI64(z, &v)==0 ){ + bDflt = v; + } + return bDflt; +} + +/* +** Return the Btree pointer identified by zDbName. Return NULL if not found. +*/ +SQLITE_PRIVATE Btree *sqlite3DbNameToBtree(sqlite3 *db, const char *zDbName){ + int iDb = zDbName ? sqlite3FindDbName(db, zDbName) : 0; + return iDb<0 ? 0 : db->aDb[iDb].pBt; +} + +/* +** Return the filename of the database associated with a database +** connection. +*/ +SQLITE_API const char *sqlite3_db_filename(sqlite3 *db, const char *zDbName){ + Btree *pBt; +#ifdef SQLITE_ENABLE_API_ARMOR + if( !sqlite3SafetyCheckOk(db) ){ + (void)SQLITE_MISUSE_BKPT; + return 0; + } +#endif + pBt = sqlite3DbNameToBtree(db, zDbName); + return pBt ? sqlite3BtreeGetFilename(pBt) : 0; +} + +/* +** Return 1 if database is read-only or 0 if read/write. Return -1 if +** no such database exists. +*/ +SQLITE_API int sqlite3_db_readonly(sqlite3 *db, const char *zDbName){ + Btree *pBt; +#ifdef SQLITE_ENABLE_API_ARMOR + if( !sqlite3SafetyCheckOk(db) ){ + (void)SQLITE_MISUSE_BKPT; + return -1; + } +#endif + pBt = sqlite3DbNameToBtree(db, zDbName); + return pBt ? sqlite3BtreeIsReadonly(pBt) : -1; +} + +#ifdef SQLITE_ENABLE_SNAPSHOT +/* +** Obtain a snapshot handle for the snapshot of database zDb currently +** being read by handle db. +*/ +SQLITE_API int sqlite3_snapshot_get( + sqlite3 *db, + const char *zDb, + sqlite3_snapshot **ppSnapshot +){ + int rc = SQLITE_ERROR; +#ifndef SQLITE_OMIT_WAL + +#ifdef SQLITE_ENABLE_API_ARMOR + if( !sqlite3SafetyCheckOk(db) ){ + return SQLITE_MISUSE_BKPT; + } +#endif + sqlite3_mutex_enter(db->mutex); + + if( db->autoCommit==0 ){ + int iDb = sqlite3FindDbName(db, zDb); + if( iDb==0 || iDb>1 ){ + Btree *pBt = db->aDb[iDb].pBt; + if( 0==sqlite3BtreeIsInTrans(pBt) ){ + rc = sqlite3BtreeBeginTrans(pBt, 0, 0); + if( rc==SQLITE_OK ){ + rc = sqlite3PagerSnapshotGet(sqlite3BtreePager(pBt), ppSnapshot); + } + } + } + } + + sqlite3_mutex_leave(db->mutex); +#endif /* SQLITE_OMIT_WAL */ + return rc; +} + +/* +** Open a read-transaction on the snapshot idendified by pSnapshot. +*/ +SQLITE_API int sqlite3_snapshot_open( + sqlite3 *db, + const char *zDb, + sqlite3_snapshot *pSnapshot +){ + int rc = SQLITE_ERROR; +#ifndef SQLITE_OMIT_WAL + +#ifdef SQLITE_ENABLE_API_ARMOR + if( !sqlite3SafetyCheckOk(db) ){ + return SQLITE_MISUSE_BKPT; + } +#endif + sqlite3_mutex_enter(db->mutex); + if( db->autoCommit==0 ){ + int iDb; + iDb = sqlite3FindDbName(db, zDb); + if( iDb==0 || iDb>1 ){ + Btree *pBt = db->aDb[iDb].pBt; + if( sqlite3BtreeIsInTrans(pBt)==0 ){ + Pager *pPager = sqlite3BtreePager(pBt); + int bUnlock = 0; + if( sqlite3BtreeIsInReadTrans(pBt) ){ + if( db->nVdbeActive==0 ){ + rc = sqlite3PagerSnapshotCheck(pPager, pSnapshot); + if( rc==SQLITE_OK ){ + bUnlock = 1; + rc = sqlite3BtreeCommit(pBt); + } + } + }else{ + rc = SQLITE_OK; + } + if( rc==SQLITE_OK ){ + rc = sqlite3PagerSnapshotOpen(pPager, pSnapshot); + } + if( rc==SQLITE_OK ){ + rc = sqlite3BtreeBeginTrans(pBt, 0, 0); + sqlite3PagerSnapshotOpen(pPager, 0); + } + if( bUnlock ){ + sqlite3PagerSnapshotUnlock(pPager); + } + } + } + } + + sqlite3_mutex_leave(db->mutex); +#endif /* SQLITE_OMIT_WAL */ + return rc; +} + +/* +** Recover as many snapshots as possible from the wal file associated with +** schema zDb of database db. +*/ +SQLITE_API int sqlite3_snapshot_recover(sqlite3 *db, const char *zDb){ + int rc = SQLITE_ERROR; + int iDb; +#ifndef SQLITE_OMIT_WAL + +#ifdef SQLITE_ENABLE_API_ARMOR + if( !sqlite3SafetyCheckOk(db) ){ + return SQLITE_MISUSE_BKPT; + } +#endif + + sqlite3_mutex_enter(db->mutex); + iDb = sqlite3FindDbName(db, zDb); + if( iDb==0 || iDb>1 ){ + Btree *pBt = db->aDb[iDb].pBt; + if( 0==sqlite3BtreeIsInReadTrans(pBt) ){ + rc = sqlite3BtreeBeginTrans(pBt, 0, 0); + if( rc==SQLITE_OK ){ + rc = sqlite3PagerSnapshotRecover(sqlite3BtreePager(pBt)); + sqlite3BtreeCommit(pBt); + } + } + } + sqlite3_mutex_leave(db->mutex); +#endif /* SQLITE_OMIT_WAL */ + return rc; +} + +/* +** Free a snapshot handle obtained from sqlite3_snapshot_get(). +*/ +SQLITE_API void sqlite3_snapshot_free(sqlite3_snapshot *pSnapshot){ + sqlite3_free(pSnapshot); +} +#endif /* SQLITE_ENABLE_SNAPSHOT */ + +#ifndef SQLITE_OMIT_COMPILEOPTION_DIAGS +/* +** Given the name of a compile-time option, return true if that option +** was used and false if not. +** +** The name can optionally begin with "SQLITE_" but the "SQLITE_" prefix +** is not required for a match. +*/ +SQLITE_API int sqlite3_compileoption_used(const char *zOptName){ + int i, n; + int nOpt; + const char **azCompileOpt; + +#if SQLITE_ENABLE_API_ARMOR + if( zOptName==0 ){ + (void)SQLITE_MISUSE_BKPT; + return 0; + } +#endif + + azCompileOpt = sqlite3CompileOptions(&nOpt); + + if( sqlite3StrNICmp(zOptName, "SQLITE_", 7)==0 ) zOptName += 7; + n = sqlite3Strlen30(zOptName); + + /* Since nOpt is normally in single digits, a linear search is + ** adequate. No need for a binary search. */ + for(i=0; i=0 && NpNextBlocked){ + int seen = 0; + sqlite3 *p2; + + /* Verify property (1) */ + assert( p->pUnlockConnection || p->pBlockingConnection ); + + /* Verify property (2) */ + for(p2=sqlite3BlockedList; p2!=p; p2=p2->pNextBlocked){ + if( p2->xUnlockNotify==p->xUnlockNotify ) seen = 1; + assert( p2->xUnlockNotify==p->xUnlockNotify || !seen ); + assert( db==0 || p->pUnlockConnection!=db ); + assert( db==0 || p->pBlockingConnection!=db ); + } + } +} +#else +# define checkListProperties(x) +#endif + +/* +** Remove connection db from the blocked connections list. If connection +** db is not currently a part of the list, this function is a no-op. +*/ +static void removeFromBlockedList(sqlite3 *db){ + sqlite3 **pp; + assertMutexHeld(); + for(pp=&sqlite3BlockedList; *pp; pp = &(*pp)->pNextBlocked){ + if( *pp==db ){ + *pp = (*pp)->pNextBlocked; + break; + } + } +} + +/* +** Add connection db to the blocked connections list. It is assumed +** that it is not already a part of the list. +*/ +static void addToBlockedList(sqlite3 *db){ + sqlite3 **pp; + assertMutexHeld(); + for( + pp=&sqlite3BlockedList; + *pp && (*pp)->xUnlockNotify!=db->xUnlockNotify; + pp=&(*pp)->pNextBlocked + ); + db->pNextBlocked = *pp; + *pp = db; +} + +/* +** Obtain the STATIC_MASTER mutex. +*/ +static void enterMutex(void){ + sqlite3_mutex_enter(sqlite3MutexAlloc(SQLITE_MUTEX_STATIC_MASTER)); + checkListProperties(0); +} + +/* +** Release the STATIC_MASTER mutex. +*/ +static void leaveMutex(void){ + assertMutexHeld(); + checkListProperties(0); + sqlite3_mutex_leave(sqlite3MutexAlloc(SQLITE_MUTEX_STATIC_MASTER)); +} + +/* +** Register an unlock-notify callback. +** +** This is called after connection "db" has attempted some operation +** but has received an SQLITE_LOCKED error because another connection +** (call it pOther) in the same process was busy using the same shared +** cache. pOther is found by looking at db->pBlockingConnection. +** +** If there is no blocking connection, the callback is invoked immediately, +** before this routine returns. +** +** If pOther is already blocked on db, then report SQLITE_LOCKED, to indicate +** a deadlock. +** +** Otherwise, make arrangements to invoke xNotify when pOther drops +** its locks. +** +** Each call to this routine overrides any prior callbacks registered +** on the same "db". If xNotify==0 then any prior callbacks are immediately +** cancelled. +*/ +SQLITE_API int sqlite3_unlock_notify( + sqlite3 *db, + void (*xNotify)(void **, int), + void *pArg +){ + int rc = SQLITE_OK; + + sqlite3_mutex_enter(db->mutex); + enterMutex(); + + if( xNotify==0 ){ + removeFromBlockedList(db); + db->pBlockingConnection = 0; + db->pUnlockConnection = 0; + db->xUnlockNotify = 0; + db->pUnlockArg = 0; + }else if( 0==db->pBlockingConnection ){ + /* The blocking transaction has been concluded. Or there never was a + ** blocking transaction. In either case, invoke the notify callback + ** immediately. + */ + xNotify(&pArg, 1); + }else{ + sqlite3 *p; + + for(p=db->pBlockingConnection; p && p!=db; p=p->pUnlockConnection){} + if( p ){ + rc = SQLITE_LOCKED; /* Deadlock detected. */ + }else{ + db->pUnlockConnection = db->pBlockingConnection; + db->xUnlockNotify = xNotify; + db->pUnlockArg = pArg; + removeFromBlockedList(db); + addToBlockedList(db); + } + } + + leaveMutex(); + assert( !db->mallocFailed ); + sqlite3ErrorWithMsg(db, rc, (rc?"database is deadlocked":0)); + sqlite3_mutex_leave(db->mutex); + return rc; +} + +/* +** This function is called while stepping or preparing a statement +** associated with connection db. The operation will return SQLITE_LOCKED +** to the user because it requires a lock that will not be available +** until connection pBlocker concludes its current transaction. +*/ +SQLITE_PRIVATE void sqlite3ConnectionBlocked(sqlite3 *db, sqlite3 *pBlocker){ + enterMutex(); + if( db->pBlockingConnection==0 && db->pUnlockConnection==0 ){ + addToBlockedList(db); + } + db->pBlockingConnection = pBlocker; + leaveMutex(); +} + +/* +** This function is called when +** the transaction opened by database db has just finished. Locks held +** by database connection db have been released. +** +** This function loops through each entry in the blocked connections +** list and does the following: +** +** 1) If the sqlite3.pBlockingConnection member of a list entry is +** set to db, then set pBlockingConnection=0. +** +** 2) If the sqlite3.pUnlockConnection member of a list entry is +** set to db, then invoke the configured unlock-notify callback and +** set pUnlockConnection=0. +** +** 3) If the two steps above mean that pBlockingConnection==0 and +** pUnlockConnection==0, remove the entry from the blocked connections +** list. +*/ +SQLITE_PRIVATE void sqlite3ConnectionUnlocked(sqlite3 *db){ + void (*xUnlockNotify)(void **, int) = 0; /* Unlock-notify cb to invoke */ + int nArg = 0; /* Number of entries in aArg[] */ + sqlite3 **pp; /* Iterator variable */ + void **aArg; /* Arguments to the unlock callback */ + void **aDyn = 0; /* Dynamically allocated space for aArg[] */ + void *aStatic[16]; /* Starter space for aArg[]. No malloc required */ + + aArg = aStatic; + enterMutex(); /* Enter STATIC_MASTER mutex */ + + /* This loop runs once for each entry in the blocked-connections list. */ + for(pp=&sqlite3BlockedList; *pp; /* no-op */ ){ + sqlite3 *p = *pp; + + /* Step 1. */ + if( p->pBlockingConnection==db ){ + p->pBlockingConnection = 0; + } + + /* Step 2. */ + if( p->pUnlockConnection==db ){ + assert( p->xUnlockNotify ); + if( p->xUnlockNotify!=xUnlockNotify && nArg!=0 ){ + xUnlockNotify(aArg, nArg); + nArg = 0; + } + + sqlite3BeginBenignMalloc(); + assert( aArg==aDyn || (aDyn==0 && aArg==aStatic) ); + assert( nArg<=(int)ArraySize(aStatic) || aArg==aDyn ); + if( (!aDyn && nArg==(int)ArraySize(aStatic)) + || (aDyn && nArg==(int)(sqlite3MallocSize(aDyn)/sizeof(void*))) + ){ + /* The aArg[] array needs to grow. */ + void **pNew = (void **)sqlite3Malloc(nArg*sizeof(void *)*2); + if( pNew ){ + memcpy(pNew, aArg, nArg*sizeof(void *)); + sqlite3_free(aDyn); + aDyn = aArg = pNew; + }else{ + /* This occurs when the array of context pointers that need to + ** be passed to the unlock-notify callback is larger than the + ** aStatic[] array allocated on the stack and the attempt to + ** allocate a larger array from the heap has failed. + ** + ** This is a difficult situation to handle. Returning an error + ** code to the caller is insufficient, as even if an error code + ** is returned the transaction on connection db will still be + ** closed and the unlock-notify callbacks on blocked connections + ** will go unissued. This might cause the application to wait + ** indefinitely for an unlock-notify callback that will never + ** arrive. + ** + ** Instead, invoke the unlock-notify callback with the context + ** array already accumulated. We can then clear the array and + ** begin accumulating any further context pointers without + ** requiring any dynamic allocation. This is sub-optimal because + ** it means that instead of one callback with a large array of + ** context pointers the application will receive two or more + ** callbacks with smaller arrays of context pointers, which will + ** reduce the applications ability to prioritize multiple + ** connections. But it is the best that can be done under the + ** circumstances. + */ + xUnlockNotify(aArg, nArg); + nArg = 0; + } + } + sqlite3EndBenignMalloc(); + + aArg[nArg++] = p->pUnlockArg; + xUnlockNotify = p->xUnlockNotify; + p->pUnlockConnection = 0; + p->xUnlockNotify = 0; + p->pUnlockArg = 0; + } + + /* Step 3. */ + if( p->pBlockingConnection==0 && p->pUnlockConnection==0 ){ + /* Remove connection p from the blocked connections list. */ + *pp = p->pNextBlocked; + p->pNextBlocked = 0; + }else{ + pp = &p->pNextBlocked; + } + } + + if( nArg!=0 ){ + xUnlockNotify(aArg, nArg); + } + sqlite3_free(aDyn); + leaveMutex(); /* Leave STATIC_MASTER mutex */ +} + +/* +** This is called when the database connection passed as an argument is +** being closed. The connection is removed from the blocked list. +*/ +SQLITE_PRIVATE void sqlite3ConnectionClosed(sqlite3 *db){ + sqlite3ConnectionUnlocked(db); + enterMutex(); + removeFromBlockedList(db); + checkListProperties(db); + leaveMutex(); +} +#endif + +/************** End of notify.c **********************************************/ +/************** Begin file fts3.c ********************************************/ +/* +** 2006 Oct 10 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +****************************************************************************** +** +** This is an SQLite module implementing full-text search. +*/ + +/* +** The code in this file is only compiled if: +** +** * The FTS3 module is being built as an extension +** (in which case SQLITE_CORE is not defined), or +** +** * The FTS3 module is being built into the core of +** SQLite (in which case SQLITE_ENABLE_FTS3 is defined). +*/ + +/* The full-text index is stored in a series of b+tree (-like) +** structures called segments which map terms to doclists. The +** structures are like b+trees in layout, but are constructed from the +** bottom up in optimal fashion and are not updatable. Since trees +** are built from the bottom up, things will be described from the +** bottom up. +** +** +**** Varints **** +** The basic unit of encoding is a variable-length integer called a +** varint. We encode variable-length integers in little-endian order +** using seven bits * per byte as follows: +** +** KEY: +** A = 0xxxxxxx 7 bits of data and one flag bit +** B = 1xxxxxxx 7 bits of data and one flag bit +** +** 7 bits - A +** 14 bits - BA +** 21 bits - BBA +** and so on. +** +** This is similar in concept to how sqlite encodes "varints" but +** the encoding is not the same. SQLite varints are big-endian +** are are limited to 9 bytes in length whereas FTS3 varints are +** little-endian and can be up to 10 bytes in length (in theory). +** +** Example encodings: +** +** 1: 0x01 +** 127: 0x7f +** 128: 0x81 0x00 +** +** +**** Document lists **** +** A doclist (document list) holds a docid-sorted list of hits for a +** given term. Doclists hold docids and associated token positions. +** A docid is the unique integer identifier for a single document. +** A position is the index of a word within the document. The first +** word of the document has a position of 0. +** +** FTS3 used to optionally store character offsets using a compile-time +** option. But that functionality is no longer supported. +** +** A doclist is stored like this: +** +** array { +** varint docid; (delta from previous doclist) +** array { (position list for column 0) +** varint position; (2 more than the delta from previous position) +** } +** array { +** varint POS_COLUMN; (marks start of position list for new column) +** varint column; (index of new column) +** array { +** varint position; (2 more than the delta from previous position) +** } +** } +** varint POS_END; (marks end of positions for this document. +** } +** +** Here, array { X } means zero or more occurrences of X, adjacent in +** memory. A "position" is an index of a token in the token stream +** generated by the tokenizer. Note that POS_END and POS_COLUMN occur +** in the same logical place as the position element, and act as sentinals +** ending a position list array. POS_END is 0. POS_COLUMN is 1. +** The positions numbers are not stored literally but rather as two more +** than the difference from the prior position, or the just the position plus +** 2 for the first position. Example: +** +** label: A B C D E F G H I J K +** value: 123 5 9 1 1 14 35 0 234 72 0 +** +** The 123 value is the first docid. For column zero in this document +** there are two matches at positions 3 and 10 (5-2 and 9-2+3). The 1 +** at D signals the start of a new column; the 1 at E indicates that the +** new column is column number 1. There are two positions at 12 and 45 +** (14-2 and 35-2+12). The 0 at H indicate the end-of-document. The +** 234 at I is the delta to next docid (357). It has one position 70 +** (72-2) and then terminates with the 0 at K. +** +** A "position-list" is the list of positions for multiple columns for +** a single docid. A "column-list" is the set of positions for a single +** column. Hence, a position-list consists of one or more column-lists, +** a document record consists of a docid followed by a position-list and +** a doclist consists of one or more document records. +** +** A bare doclist omits the position information, becoming an +** array of varint-encoded docids. +** +**** Segment leaf nodes **** +** Segment leaf nodes store terms and doclists, ordered by term. Leaf +** nodes are written using LeafWriter, and read using LeafReader (to +** iterate through a single leaf node's data) and LeavesReader (to +** iterate through a segment's entire leaf layer). Leaf nodes have +** the format: +** +** varint iHeight; (height from leaf level, always 0) +** varint nTerm; (length of first term) +** char pTerm[nTerm]; (content of first term) +** varint nDoclist; (length of term's associated doclist) +** char pDoclist[nDoclist]; (content of doclist) +** array { +** (further terms are delta-encoded) +** varint nPrefix; (length of prefix shared with previous term) +** varint nSuffix; (length of unshared suffix) +** char pTermSuffix[nSuffix];(unshared suffix of next term) +** varint nDoclist; (length of term's associated doclist) +** char pDoclist[nDoclist]; (content of doclist) +** } +** +** Here, array { X } means zero or more occurrences of X, adjacent in +** memory. +** +** Leaf nodes are broken into blocks which are stored contiguously in +** the %_segments table in sorted order. This means that when the end +** of a node is reached, the next term is in the node with the next +** greater node id. +** +** New data is spilled to a new leaf node when the current node +** exceeds LEAF_MAX bytes (default 2048). New data which itself is +** larger than STANDALONE_MIN (default 1024) is placed in a standalone +** node (a leaf node with a single term and doclist). The goal of +** these settings is to pack together groups of small doclists while +** making it efficient to directly access large doclists. The +** assumption is that large doclists represent terms which are more +** likely to be query targets. +** +** TODO(shess) It may be useful for blocking decisions to be more +** dynamic. For instance, it may make more sense to have a 2.5k leaf +** node rather than splitting into 2k and .5k nodes. My intuition is +** that this might extend through 2x or 4x the pagesize. +** +** +**** Segment interior nodes **** +** Segment interior nodes store blockids for subtree nodes and terms +** to describe what data is stored by the each subtree. Interior +** nodes are written using InteriorWriter, and read using +** InteriorReader. InteriorWriters are created as needed when +** SegmentWriter creates new leaf nodes, or when an interior node +** itself grows too big and must be split. The format of interior +** nodes: +** +** varint iHeight; (height from leaf level, always >0) +** varint iBlockid; (block id of node's leftmost subtree) +** optional { +** varint nTerm; (length of first term) +** char pTerm[nTerm]; (content of first term) +** array { +** (further terms are delta-encoded) +** varint nPrefix; (length of shared prefix with previous term) +** varint nSuffix; (length of unshared suffix) +** char pTermSuffix[nSuffix]; (unshared suffix of next term) +** } +** } +** +** Here, optional { X } means an optional element, while array { X } +** means zero or more occurrences of X, adjacent in memory. +** +** An interior node encodes n terms separating n+1 subtrees. The +** subtree blocks are contiguous, so only the first subtree's blockid +** is encoded. The subtree at iBlockid will contain all terms less +** than the first term encoded (or all terms if no term is encoded). +** Otherwise, for terms greater than or equal to pTerm[i] but less +** than pTerm[i+1], the subtree for that term will be rooted at +** iBlockid+i. Interior nodes only store enough term data to +** distinguish adjacent children (if the rightmost term of the left +** child is "something", and the leftmost term of the right child is +** "wicked", only "w" is stored). +** +** New data is spilled to a new interior node at the same height when +** the current node exceeds INTERIOR_MAX bytes (default 2048). +** INTERIOR_MIN_TERMS (default 7) keeps large terms from monopolizing +** interior nodes and making the tree too skinny. The interior nodes +** at a given height are naturally tracked by interior nodes at +** height+1, and so on. +** +** +**** Segment directory **** +** The segment directory in table %_segdir stores meta-information for +** merging and deleting segments, and also the root node of the +** segment's tree. +** +** The root node is the top node of the segment's tree after encoding +** the entire segment, restricted to ROOT_MAX bytes (default 1024). +** This could be either a leaf node or an interior node. If the top +** node requires more than ROOT_MAX bytes, it is flushed to %_segments +** and a new root interior node is generated (which should always fit +** within ROOT_MAX because it only needs space for 2 varints, the +** height and the blockid of the previous root). +** +** The meta-information in the segment directory is: +** level - segment level (see below) +** idx - index within level +** - (level,idx uniquely identify a segment) +** start_block - first leaf node +** leaves_end_block - last leaf node +** end_block - last block (including interior nodes) +** root - contents of root node +** +** If the root node is a leaf node, then start_block, +** leaves_end_block, and end_block are all 0. +** +** +**** Segment merging **** +** To amortize update costs, segments are grouped into levels and +** merged in batches. Each increase in level represents exponentially +** more documents. +** +** New documents (actually, document updates) are tokenized and +** written individually (using LeafWriter) to a level 0 segment, with +** incrementing idx. When idx reaches MERGE_COUNT (default 16), all +** level 0 segments are merged into a single level 1 segment. Level 1 +** is populated like level 0, and eventually MERGE_COUNT level 1 +** segments are merged to a single level 2 segment (representing +** MERGE_COUNT^2 updates), and so on. +** +** A segment merge traverses all segments at a given level in +** parallel, performing a straightforward sorted merge. Since segment +** leaf nodes are written in to the %_segments table in order, this +** merge traverses the underlying sqlite disk structures efficiently. +** After the merge, all segment blocks from the merged level are +** deleted. +** +** MERGE_COUNT controls how often we merge segments. 16 seems to be +** somewhat of a sweet spot for insertion performance. 32 and 64 show +** very similar performance numbers to 16 on insertion, though they're +** a tiny bit slower (perhaps due to more overhead in merge-time +** sorting). 8 is about 20% slower than 16, 4 about 50% slower than +** 16, 2 about 66% slower than 16. +** +** At query time, high MERGE_COUNT increases the number of segments +** which need to be scanned and merged. For instance, with 100k docs +** inserted: +** +** MERGE_COUNT segments +** 16 25 +** 8 12 +** 4 10 +** 2 6 +** +** This appears to have only a moderate impact on queries for very +** frequent terms (which are somewhat dominated by segment merge +** costs), and infrequent and non-existent terms still seem to be fast +** even with many segments. +** +** TODO(shess) That said, it would be nice to have a better query-side +** argument for MERGE_COUNT of 16. Also, it is possible/likely that +** optimizations to things like doclist merging will swing the sweet +** spot around. +** +** +** +**** Handling of deletions and updates **** +** Since we're using a segmented structure, with no docid-oriented +** index into the term index, we clearly cannot simply update the term +** index when a document is deleted or updated. For deletions, we +** write an empty doclist (varint(docid) varint(POS_END)), for updates +** we simply write the new doclist. Segment merges overwrite older +** data for a particular docid with newer data, so deletes or updates +** will eventually overtake the earlier data and knock it out. The +** query logic likewise merges doclists so that newer data knocks out +** older data. +*/ + +/************** Include fts3Int.h in the middle of fts3.c ********************/ +/************** Begin file fts3Int.h *****************************************/ +/* +** 2009 Nov 12 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +****************************************************************************** +** +*/ +#ifndef _FTSINT_H +#define _FTSINT_H + +#if !defined(NDEBUG) && !defined(SQLITE_DEBUG) +# define NDEBUG 1 +#endif + +/* FTS3/FTS4 require virtual tables */ +#ifdef SQLITE_OMIT_VIRTUALTABLE +# undef SQLITE_ENABLE_FTS3 +# undef SQLITE_ENABLE_FTS4 +#endif + +/* +** FTS4 is really an extension for FTS3. It is enabled using the +** SQLITE_ENABLE_FTS3 macro. But to avoid confusion we also all +** the SQLITE_ENABLE_FTS4 macro to serve as an alisse for SQLITE_ENABLE_FTS3. +*/ +#if defined(SQLITE_ENABLE_FTS4) && !defined(SQLITE_ENABLE_FTS3) +# define SQLITE_ENABLE_FTS3 +#endif + +#if !defined(SQLITE_CORE) || defined(SQLITE_ENABLE_FTS3) + +/* If not building as part of the core, include sqlite3ext.h. */ +#ifndef SQLITE_CORE +/* # include "sqlite3ext.h" */ +SQLITE_EXTENSION_INIT3 +#endif + +/* #include "sqlite3.h" */ +/************** Include fts3_tokenizer.h in the middle of fts3Int.h **********/ +/************** Begin file fts3_tokenizer.h **********************************/ +/* +** 2006 July 10 +** +** The author disclaims copyright to this source code. +** +************************************************************************* +** Defines the interface to tokenizers used by fulltext-search. There +** are three basic components: +** +** sqlite3_tokenizer_module is a singleton defining the tokenizer +** interface functions. This is essentially the class structure for +** tokenizers. +** +** sqlite3_tokenizer is used to define a particular tokenizer, perhaps +** including customization information defined at creation time. +** +** sqlite3_tokenizer_cursor is generated by a tokenizer to generate +** tokens from a particular input. +*/ +#ifndef _FTS3_TOKENIZER_H_ +#define _FTS3_TOKENIZER_H_ + +/* TODO(shess) Only used for SQLITE_OK and SQLITE_DONE at this time. +** If tokenizers are to be allowed to call sqlite3_*() functions, then +** we will need a way to register the API consistently. +*/ +/* #include "sqlite3.h" */ + +/* +** Structures used by the tokenizer interface. When a new tokenizer +** implementation is registered, the caller provides a pointer to +** an sqlite3_tokenizer_module containing pointers to the callback +** functions that make up an implementation. +** +** When an fts3 table is created, it passes any arguments passed to +** the tokenizer clause of the CREATE VIRTUAL TABLE statement to the +** sqlite3_tokenizer_module.xCreate() function of the requested tokenizer +** implementation. The xCreate() function in turn returns an +** sqlite3_tokenizer structure representing the specific tokenizer to +** be used for the fts3 table (customized by the tokenizer clause arguments). +** +** To tokenize an input buffer, the sqlite3_tokenizer_module.xOpen() +** method is called. It returns an sqlite3_tokenizer_cursor object +** that may be used to tokenize a specific input buffer based on +** the tokenization rules supplied by a specific sqlite3_tokenizer +** object. +*/ +typedef struct sqlite3_tokenizer_module sqlite3_tokenizer_module; +typedef struct sqlite3_tokenizer sqlite3_tokenizer; +typedef struct sqlite3_tokenizer_cursor sqlite3_tokenizer_cursor; + +struct sqlite3_tokenizer_module { + + /* + ** Structure version. Should always be set to 0 or 1. + */ + int iVersion; + + /* + ** Create a new tokenizer. The values in the argv[] array are the + ** arguments passed to the "tokenizer" clause of the CREATE VIRTUAL + ** TABLE statement that created the fts3 table. For example, if + ** the following SQL is executed: + ** + ** CREATE .. USING fts3( ... , tokenizer arg1 arg2) + ** + ** then argc is set to 2, and the argv[] array contains pointers + ** to the strings "arg1" and "arg2". + ** + ** This method should return either SQLITE_OK (0), or an SQLite error + ** code. If SQLITE_OK is returned, then *ppTokenizer should be set + ** to point at the newly created tokenizer structure. The generic + ** sqlite3_tokenizer.pModule variable should not be initialized by + ** this callback. The caller will do so. + */ + int (*xCreate)( + int argc, /* Size of argv array */ + const char *const*argv, /* Tokenizer argument strings */ + sqlite3_tokenizer **ppTokenizer /* OUT: Created tokenizer */ + ); + + /* + ** Destroy an existing tokenizer. The fts3 module calls this method + ** exactly once for each successful call to xCreate(). + */ + int (*xDestroy)(sqlite3_tokenizer *pTokenizer); + + /* + ** Create a tokenizer cursor to tokenize an input buffer. The caller + ** is responsible for ensuring that the input buffer remains valid + ** until the cursor is closed (using the xClose() method). + */ + int (*xOpen)( + sqlite3_tokenizer *pTokenizer, /* Tokenizer object */ + const char *pInput, int nBytes, /* Input buffer */ + sqlite3_tokenizer_cursor **ppCursor /* OUT: Created tokenizer cursor */ + ); + + /* + ** Destroy an existing tokenizer cursor. The fts3 module calls this + ** method exactly once for each successful call to xOpen(). + */ + int (*xClose)(sqlite3_tokenizer_cursor *pCursor); + + /* + ** Retrieve the next token from the tokenizer cursor pCursor. This + ** method should either return SQLITE_OK and set the values of the + ** "OUT" variables identified below, or SQLITE_DONE to indicate that + ** the end of the buffer has been reached, or an SQLite error code. + ** + ** *ppToken should be set to point at a buffer containing the + ** normalized version of the token (i.e. after any case-folding and/or + ** stemming has been performed). *pnBytes should be set to the length + ** of this buffer in bytes. The input text that generated the token is + ** identified by the byte offsets returned in *piStartOffset and + ** *piEndOffset. *piStartOffset should be set to the index of the first + ** byte of the token in the input buffer. *piEndOffset should be set + ** to the index of the first byte just past the end of the token in + ** the input buffer. + ** + ** The buffer *ppToken is set to point at is managed by the tokenizer + ** implementation. It is only required to be valid until the next call + ** to xNext() or xClose(). + */ + /* TODO(shess) current implementation requires pInput to be + ** nul-terminated. This should either be fixed, or pInput/nBytes + ** should be converted to zInput. + */ + int (*xNext)( + sqlite3_tokenizer_cursor *pCursor, /* Tokenizer cursor */ + const char **ppToken, int *pnBytes, /* OUT: Normalized text for token */ + int *piStartOffset, /* OUT: Byte offset of token in input buffer */ + int *piEndOffset, /* OUT: Byte offset of end of token in input buffer */ + int *piPosition /* OUT: Number of tokens returned before this one */ + ); + + /*********************************************************************** + ** Methods below this point are only available if iVersion>=1. + */ + + /* + ** Configure the language id of a tokenizer cursor. + */ + int (*xLanguageid)(sqlite3_tokenizer_cursor *pCsr, int iLangid); +}; + +struct sqlite3_tokenizer { + const sqlite3_tokenizer_module *pModule; /* The module for this tokenizer */ + /* Tokenizer implementations will typically add additional fields */ +}; + +struct sqlite3_tokenizer_cursor { + sqlite3_tokenizer *pTokenizer; /* Tokenizer for this cursor. */ + /* Tokenizer implementations will typically add additional fields */ +}; + +int fts3_global_term_cnt(int iTerm, int iCol); +int fts3_term_cnt(int iTerm, int iCol); + + +#endif /* _FTS3_TOKENIZER_H_ */ + +/************** End of fts3_tokenizer.h **************************************/ +/************** Continuing where we left off in fts3Int.h ********************/ +/************** Include fts3_hash.h in the middle of fts3Int.h ***************/ +/************** Begin file fts3_hash.h ***************************************/ +/* +** 2001 September 22 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** This is the header file for the generic hash-table implementation +** used in SQLite. We've modified it slightly to serve as a standalone +** hash table implementation for the full-text indexing module. +** +*/ +#ifndef _FTS3_HASH_H_ +#define _FTS3_HASH_H_ + +/* Forward declarations of structures. */ +typedef struct Fts3Hash Fts3Hash; +typedef struct Fts3HashElem Fts3HashElem; + +/* A complete hash table is an instance of the following structure. +** The internals of this structure are intended to be opaque -- client +** code should not attempt to access or modify the fields of this structure +** directly. Change this structure only by using the routines below. +** However, many of the "procedures" and "functions" for modifying and +** accessing this structure are really macros, so we can't really make +** this structure opaque. +*/ +struct Fts3Hash { + char keyClass; /* HASH_INT, _POINTER, _STRING, _BINARY */ + char copyKey; /* True if copy of key made on insert */ + int count; /* Number of entries in this table */ + Fts3HashElem *first; /* The first element of the array */ + int htsize; /* Number of buckets in the hash table */ + struct _fts3ht { /* the hash table */ + int count; /* Number of entries with this hash */ + Fts3HashElem *chain; /* Pointer to first entry with this hash */ + } *ht; +}; + +/* Each element in the hash table is an instance of the following +** structure. All elements are stored on a single doubly-linked list. +** +** Again, this structure is intended to be opaque, but it can't really +** be opaque because it is used by macros. +*/ +struct Fts3HashElem { + Fts3HashElem *next, *prev; /* Next and previous elements in the table */ + void *data; /* Data associated with this element */ + void *pKey; int nKey; /* Key associated with this element */ +}; + +/* +** There are 2 different modes of operation for a hash table: +** +** FTS3_HASH_STRING pKey points to a string that is nKey bytes long +** (including the null-terminator, if any). Case +** is respected in comparisons. +** +** FTS3_HASH_BINARY pKey points to binary data nKey bytes long. +** memcmp() is used to compare keys. +** +** A copy of the key is made if the copyKey parameter to fts3HashInit is 1. +*/ +#define FTS3_HASH_STRING 1 +#define FTS3_HASH_BINARY 2 + +/* +** Access routines. To delete, insert a NULL pointer. +*/ +SQLITE_PRIVATE void sqlite3Fts3HashInit(Fts3Hash *pNew, char keyClass, char copyKey); +SQLITE_PRIVATE void *sqlite3Fts3HashInsert(Fts3Hash*, const void *pKey, int nKey, void *pData); +SQLITE_PRIVATE void *sqlite3Fts3HashFind(const Fts3Hash*, const void *pKey, int nKey); +SQLITE_PRIVATE void sqlite3Fts3HashClear(Fts3Hash*); +SQLITE_PRIVATE Fts3HashElem *sqlite3Fts3HashFindElem(const Fts3Hash *, const void *, int); + +/* +** Shorthand for the functions above +*/ +#define fts3HashInit sqlite3Fts3HashInit +#define fts3HashInsert sqlite3Fts3HashInsert +#define fts3HashFind sqlite3Fts3HashFind +#define fts3HashClear sqlite3Fts3HashClear +#define fts3HashFindElem sqlite3Fts3HashFindElem + +/* +** Macros for looping over all elements of a hash table. The idiom is +** like this: +** +** Fts3Hash h; +** Fts3HashElem *p; +** ... +** for(p=fts3HashFirst(&h); p; p=fts3HashNext(p)){ +** SomeStructure *pData = fts3HashData(p); +** // do something with pData +** } +*/ +#define fts3HashFirst(H) ((H)->first) +#define fts3HashNext(E) ((E)->next) +#define fts3HashData(E) ((E)->data) +#define fts3HashKey(E) ((E)->pKey) +#define fts3HashKeysize(E) ((E)->nKey) + +/* +** Number of entries in a hash table +*/ +#define fts3HashCount(H) ((H)->count) + +#endif /* _FTS3_HASH_H_ */ + +/************** End of fts3_hash.h *******************************************/ +/************** Continuing where we left off in fts3Int.h ********************/ + +/* +** This constant determines the maximum depth of an FTS expression tree +** that the library will create and use. FTS uses recursion to perform +** various operations on the query tree, so the disadvantage of a large +** limit is that it may allow very large queries to use large amounts +** of stack space (perhaps causing a stack overflow). +*/ +#ifndef SQLITE_FTS3_MAX_EXPR_DEPTH +# define SQLITE_FTS3_MAX_EXPR_DEPTH 12 +#endif + + +/* +** This constant controls how often segments are merged. Once there are +** FTS3_MERGE_COUNT segments of level N, they are merged into a single +** segment of level N+1. +*/ +#define FTS3_MERGE_COUNT 16 + +/* +** This is the maximum amount of data (in bytes) to store in the +** Fts3Table.pendingTerms hash table. Normally, the hash table is +** populated as documents are inserted/updated/deleted in a transaction +** and used to create a new segment when the transaction is committed. +** However if this limit is reached midway through a transaction, a new +** segment is created and the hash table cleared immediately. +*/ +#define FTS3_MAX_PENDING_DATA (1*1024*1024) + +/* +** Macro to return the number of elements in an array. SQLite has a +** similar macro called ArraySize(). Use a different name to avoid +** a collision when building an amalgamation with built-in FTS3. +*/ +#define SizeofArray(X) ((int)(sizeof(X)/sizeof(X[0]))) + + +#ifndef MIN +# define MIN(x,y) ((x)<(y)?(x):(y)) +#endif +#ifndef MAX +# define MAX(x,y) ((x)>(y)?(x):(y)) +#endif + +/* +** Maximum length of a varint encoded integer. The varint format is different +** from that used by SQLite, so the maximum length is 10, not 9. +*/ +#define FTS3_VARINT_MAX 10 + +#define FTS3_BUFFER_PADDING 8 + +/* +** FTS4 virtual tables may maintain multiple indexes - one index of all terms +** in the document set and zero or more prefix indexes. All indexes are stored +** as one or more b+-trees in the %_segments and %_segdir tables. +** +** It is possible to determine which index a b+-tree belongs to based on the +** value stored in the "%_segdir.level" column. Given this value L, the index +** that the b+-tree belongs to is (L<<10). In other words, all b+-trees with +** level values between 0 and 1023 (inclusive) belong to index 0, all levels +** between 1024 and 2047 to index 1, and so on. +** +** It is considered impossible for an index to use more than 1024 levels. In +** theory though this may happen, but only after at least +** (FTS3_MERGE_COUNT^1024) separate flushes of the pending-terms tables. +*/ +#define FTS3_SEGDIR_MAXLEVEL 1024 +#define FTS3_SEGDIR_MAXLEVEL_STR "1024" + +/* +** The testcase() macro is only used by the amalgamation. If undefined, +** make it a no-op. +*/ +#ifndef testcase +# define testcase(X) +#endif + +/* +** Terminator values for position-lists and column-lists. +*/ +#define POS_COLUMN (1) /* Column-list terminator */ +#define POS_END (0) /* Position-list terminator */ + +/* +** The assert_fts3_nc() macro is similar to the assert() macro, except that it +** is used for assert() conditions that are true only if it can be +** guranteed that the database is not corrupt. +*/ +#if defined(SQLITE_DEBUG) || defined(SQLITE_TEST) +SQLITE_API extern int sqlite3_fts3_may_be_corrupt; +# define assert_fts3_nc(x) assert(sqlite3_fts3_may_be_corrupt || (x)) +#else +# define assert_fts3_nc(x) assert(x) +#endif + +/* +** This section provides definitions to allow the +** FTS3 extension to be compiled outside of the +** amalgamation. +*/ +#ifndef SQLITE_AMALGAMATION +/* +** Macros indicating that conditional expressions are always true or +** false. +*/ +#ifdef SQLITE_COVERAGE_TEST +# define ALWAYS(x) (1) +# define NEVER(X) (0) +#elif defined(SQLITE_DEBUG) +# define ALWAYS(x) sqlite3Fts3Always((x)!=0) +# define NEVER(x) sqlite3Fts3Never((x)!=0) +SQLITE_PRIVATE int sqlite3Fts3Always(int b); +SQLITE_PRIVATE int sqlite3Fts3Never(int b); +#else +# define ALWAYS(x) (x) +# define NEVER(x) (x) +#endif + +/* +** Internal types used by SQLite. +*/ +typedef unsigned char u8; /* 1-byte (or larger) unsigned integer */ +typedef short int i16; /* 2-byte (or larger) signed integer */ +typedef unsigned int u32; /* 4-byte unsigned integer */ +typedef sqlite3_uint64 u64; /* 8-byte unsigned integer */ +typedef sqlite3_int64 i64; /* 8-byte signed integer */ + +/* +** Macro used to suppress compiler warnings for unused parameters. +*/ +#define UNUSED_PARAMETER(x) (void)(x) + +/* +** Activate assert() only if SQLITE_TEST is enabled. +*/ +#if !defined(NDEBUG) && !defined(SQLITE_DEBUG) +# define NDEBUG 1 +#endif + +/* +** The TESTONLY macro is used to enclose variable declarations or +** other bits of code that are needed to support the arguments +** within testcase() and assert() macros. +*/ +#if defined(SQLITE_DEBUG) || defined(SQLITE_COVERAGE_TEST) +# define TESTONLY(X) X +#else +# define TESTONLY(X) +#endif + +#endif /* SQLITE_AMALGAMATION */ + +#ifdef SQLITE_DEBUG +SQLITE_PRIVATE int sqlite3Fts3Corrupt(void); +# define FTS_CORRUPT_VTAB sqlite3Fts3Corrupt() +#else +# define FTS_CORRUPT_VTAB SQLITE_CORRUPT_VTAB +#endif + +typedef struct Fts3Table Fts3Table; +typedef struct Fts3Cursor Fts3Cursor; +typedef struct Fts3Expr Fts3Expr; +typedef struct Fts3Phrase Fts3Phrase; +typedef struct Fts3PhraseToken Fts3PhraseToken; + +typedef struct Fts3Doclist Fts3Doclist; +typedef struct Fts3SegFilter Fts3SegFilter; +typedef struct Fts3DeferredToken Fts3DeferredToken; +typedef struct Fts3SegReader Fts3SegReader; +typedef struct Fts3MultiSegReader Fts3MultiSegReader; + +typedef struct MatchinfoBuffer MatchinfoBuffer; + +/* +** A connection to a fulltext index is an instance of the following +** structure. The xCreate and xConnect methods create an instance +** of this structure and xDestroy and xDisconnect free that instance. +** All other methods receive a pointer to the structure as one of their +** arguments. +*/ +struct Fts3Table { + sqlite3_vtab base; /* Base class used by SQLite core */ + sqlite3 *db; /* The database connection */ + const char *zDb; /* logical database name */ + const char *zName; /* virtual table name */ + int nColumn; /* number of named columns in virtual table */ + char **azColumn; /* column names. malloced */ + u8 *abNotindexed; /* True for 'notindexed' columns */ + sqlite3_tokenizer *pTokenizer; /* tokenizer for inserts and queries */ + char *zContentTbl; /* content=xxx option, or NULL */ + char *zLanguageid; /* languageid=xxx option, or NULL */ + int nAutoincrmerge; /* Value configured by 'automerge' */ + u32 nLeafAdd; /* Number of leaf blocks added this trans */ + + /* Precompiled statements used by the implementation. Each of these + ** statements is run and reset within a single virtual table API call. + */ + sqlite3_stmt *aStmt[40]; + sqlite3_stmt *pSeekStmt; /* Cache for fts3CursorSeekStmt() */ + + char *zReadExprlist; + char *zWriteExprlist; + + int nNodeSize; /* Soft limit for node size */ + u8 bFts4; /* True for FTS4, false for FTS3 */ + u8 bHasStat; /* True if %_stat table exists (2==unknown) */ + u8 bHasDocsize; /* True if %_docsize table exists */ + u8 bDescIdx; /* True if doclists are in reverse order */ + u8 bIgnoreSavepoint; /* True to ignore xSavepoint invocations */ + int nPgsz; /* Page size for host database */ + char *zSegmentsTbl; /* Name of %_segments table */ + sqlite3_blob *pSegments; /* Blob handle open on %_segments table */ + + /* + ** The following array of hash tables is used to buffer pending index + ** updates during transactions. All pending updates buffered at any one + ** time must share a common language-id (see the FTS4 langid= feature). + ** The current language id is stored in variable iPrevLangid. + ** + ** A single FTS4 table may have multiple full-text indexes. For each index + ** there is an entry in the aIndex[] array. Index 0 is an index of all the + ** terms that appear in the document set. Each subsequent index in aIndex[] + ** is an index of prefixes of a specific length. + ** + ** Variable nPendingData contains an estimate the memory consumed by the + ** pending data structures, including hash table overhead, but not including + ** malloc overhead. When nPendingData exceeds nMaxPendingData, all hash + ** tables are flushed to disk. Variable iPrevDocid is the docid of the most + ** recently inserted record. + */ + int nIndex; /* Size of aIndex[] */ + struct Fts3Index { + int nPrefix; /* Prefix length (0 for main terms index) */ + Fts3Hash hPending; /* Pending terms table for this index */ + } *aIndex; + int nMaxPendingData; /* Max pending data before flush to disk */ + int nPendingData; /* Current bytes of pending data */ + sqlite_int64 iPrevDocid; /* Docid of most recently inserted document */ + int iPrevLangid; /* Langid of recently inserted document */ + int bPrevDelete; /* True if last operation was a delete */ + +#if defined(SQLITE_DEBUG) || defined(SQLITE_COVERAGE_TEST) + /* State variables used for validating that the transaction control + ** methods of the virtual table are called at appropriate times. These + ** values do not contribute to FTS functionality; they are used for + ** verifying the operation of the SQLite core. + */ + int inTransaction; /* True after xBegin but before xCommit/xRollback */ + int mxSavepoint; /* Largest valid xSavepoint integer */ +#endif + +#ifdef SQLITE_TEST + /* True to disable the incremental doclist optimization. This is controled + ** by special insert command 'test-no-incr-doclist'. */ + int bNoIncrDoclist; +#endif +}; + +/* +** When the core wants to read from the virtual table, it creates a +** virtual table cursor (an instance of the following structure) using +** the xOpen method. Cursors are destroyed using the xClose method. +*/ +struct Fts3Cursor { + sqlite3_vtab_cursor base; /* Base class used by SQLite core */ + i16 eSearch; /* Search strategy (see below) */ + u8 isEof; /* True if at End Of Results */ + u8 isRequireSeek; /* True if must seek pStmt to %_content row */ + u8 bSeekStmt; /* True if pStmt is a seek */ + sqlite3_stmt *pStmt; /* Prepared statement in use by the cursor */ + Fts3Expr *pExpr; /* Parsed MATCH query string */ + int iLangid; /* Language being queried for */ + int nPhrase; /* Number of matchable phrases in query */ + Fts3DeferredToken *pDeferred; /* Deferred search tokens, if any */ + sqlite3_int64 iPrevId; /* Previous id read from aDoclist */ + char *pNextId; /* Pointer into the body of aDoclist */ + char *aDoclist; /* List of docids for full-text queries */ + int nDoclist; /* Size of buffer at aDoclist */ + u8 bDesc; /* True to sort in descending order */ + int eEvalmode; /* An FTS3_EVAL_XX constant */ + int nRowAvg; /* Average size of database rows, in pages */ + sqlite3_int64 nDoc; /* Documents in table */ + i64 iMinDocid; /* Minimum docid to return */ + i64 iMaxDocid; /* Maximum docid to return */ + int isMatchinfoNeeded; /* True when aMatchinfo[] needs filling in */ + MatchinfoBuffer *pMIBuffer; /* Buffer for matchinfo data */ +}; + +#define FTS3_EVAL_FILTER 0 +#define FTS3_EVAL_NEXT 1 +#define FTS3_EVAL_MATCHINFO 2 + +/* +** The Fts3Cursor.eSearch member is always set to one of the following. +** Actualy, Fts3Cursor.eSearch can be greater than or equal to +** FTS3_FULLTEXT_SEARCH. If so, then Fts3Cursor.eSearch - 2 is the index +** of the column to be searched. For example, in +** +** CREATE VIRTUAL TABLE ex1 USING fts3(a,b,c,d); +** SELECT docid FROM ex1 WHERE b MATCH 'one two three'; +** +** Because the LHS of the MATCH operator is 2nd column "b", +** Fts3Cursor.eSearch will be set to FTS3_FULLTEXT_SEARCH+1. (+0 for a, +** +1 for b, +2 for c, +3 for d.) If the LHS of MATCH were "ex1" +** indicating that all columns should be searched, +** then eSearch would be set to FTS3_FULLTEXT_SEARCH+4. +*/ +#define FTS3_FULLSCAN_SEARCH 0 /* Linear scan of %_content table */ +#define FTS3_DOCID_SEARCH 1 /* Lookup by rowid on %_content table */ +#define FTS3_FULLTEXT_SEARCH 2 /* Full-text index search */ + +/* +** The lower 16-bits of the sqlite3_index_info.idxNum value set by +** the xBestIndex() method contains the Fts3Cursor.eSearch value described +** above. The upper 16-bits contain a combination of the following +** bits, used to describe extra constraints on full-text searches. +*/ +#define FTS3_HAVE_LANGID 0x00010000 /* languageid=? */ +#define FTS3_HAVE_DOCID_GE 0x00020000 /* docid>=? */ +#define FTS3_HAVE_DOCID_LE 0x00040000 /* docid<=? */ + +struct Fts3Doclist { + char *aAll; /* Array containing doclist (or NULL) */ + int nAll; /* Size of a[] in bytes */ + char *pNextDocid; /* Pointer to next docid */ + + sqlite3_int64 iDocid; /* Current docid (if pList!=0) */ + int bFreeList; /* True if pList should be sqlite3_free()d */ + char *pList; /* Pointer to position list following iDocid */ + int nList; /* Length of position list */ +}; + +/* +** A "phrase" is a sequence of one or more tokens that must match in +** sequence. A single token is the base case and the most common case. +** For a sequence of tokens contained in double-quotes (i.e. "one two three") +** nToken will be the number of tokens in the string. +*/ +struct Fts3PhraseToken { + char *z; /* Text of the token */ + int n; /* Number of bytes in buffer z */ + int isPrefix; /* True if token ends with a "*" character */ + int bFirst; /* True if token must appear at position 0 */ + + /* Variables above this point are populated when the expression is + ** parsed (by code in fts3_expr.c). Below this point the variables are + ** used when evaluating the expression. */ + Fts3DeferredToken *pDeferred; /* Deferred token object for this token */ + Fts3MultiSegReader *pSegcsr; /* Segment-reader for this token */ +}; + +struct Fts3Phrase { + /* Cache of doclist for this phrase. */ + Fts3Doclist doclist; + int bIncr; /* True if doclist is loaded incrementally */ + int iDoclistToken; + + /* Used by sqlite3Fts3EvalPhrasePoslist() if this is a descendent of an + ** OR condition. */ + char *pOrPoslist; + i64 iOrDocid; + + /* Variables below this point are populated by fts3_expr.c when parsing + ** a MATCH expression. Everything above is part of the evaluation phase. + */ + int nToken; /* Number of tokens in the phrase */ + int iColumn; /* Index of column this phrase must match */ + Fts3PhraseToken aToken[1]; /* One entry for each token in the phrase */ +}; + +/* +** A tree of these objects forms the RHS of a MATCH operator. +** +** If Fts3Expr.eType is FTSQUERY_PHRASE and isLoaded is true, then aDoclist +** points to a malloced buffer, size nDoclist bytes, containing the results +** of this phrase query in FTS3 doclist format. As usual, the initial +** "Length" field found in doclists stored on disk is omitted from this +** buffer. +** +** Variable aMI is used only for FTSQUERY_NEAR nodes to store the global +** matchinfo data. If it is not NULL, it points to an array of size nCol*3, +** where nCol is the number of columns in the queried FTS table. The array +** is populated as follows: +** +** aMI[iCol*3 + 0] = Undefined +** aMI[iCol*3 + 1] = Number of occurrences +** aMI[iCol*3 + 2] = Number of rows containing at least one instance +** +** The aMI array is allocated using sqlite3_malloc(). It should be freed +** when the expression node is. +*/ +struct Fts3Expr { + int eType; /* One of the FTSQUERY_XXX values defined below */ + int nNear; /* Valid if eType==FTSQUERY_NEAR */ + Fts3Expr *pParent; /* pParent->pLeft==this or pParent->pRight==this */ + Fts3Expr *pLeft; /* Left operand */ + Fts3Expr *pRight; /* Right operand */ + Fts3Phrase *pPhrase; /* Valid if eType==FTSQUERY_PHRASE */ + + /* The following are used by the fts3_eval.c module. */ + sqlite3_int64 iDocid; /* Current docid */ + u8 bEof; /* True this expression is at EOF already */ + u8 bStart; /* True if iDocid is valid */ + u8 bDeferred; /* True if this expression is entirely deferred */ + + /* The following are used by the fts3_snippet.c module. */ + int iPhrase; /* Index of this phrase in matchinfo() results */ + u32 *aMI; /* See above */ +}; + +/* +** Candidate values for Fts3Query.eType. Note that the order of the first +** four values is in order of precedence when parsing expressions. For +** example, the following: +** +** "a OR b AND c NOT d NEAR e" +** +** is equivalent to: +** +** "a OR (b AND (c NOT (d NEAR e)))" +*/ +#define FTSQUERY_NEAR 1 +#define FTSQUERY_NOT 2 +#define FTSQUERY_AND 3 +#define FTSQUERY_OR 4 +#define FTSQUERY_PHRASE 5 + + +/* fts3_write.c */ +SQLITE_PRIVATE int sqlite3Fts3UpdateMethod(sqlite3_vtab*,int,sqlite3_value**,sqlite3_int64*); +SQLITE_PRIVATE int sqlite3Fts3PendingTermsFlush(Fts3Table *); +SQLITE_PRIVATE void sqlite3Fts3PendingTermsClear(Fts3Table *); +SQLITE_PRIVATE int sqlite3Fts3Optimize(Fts3Table *); +SQLITE_PRIVATE int sqlite3Fts3SegReaderNew(int, int, sqlite3_int64, + sqlite3_int64, sqlite3_int64, const char *, int, Fts3SegReader**); +SQLITE_PRIVATE int sqlite3Fts3SegReaderPending( + Fts3Table*,int,const char*,int,int,Fts3SegReader**); +SQLITE_PRIVATE void sqlite3Fts3SegReaderFree(Fts3SegReader *); +SQLITE_PRIVATE int sqlite3Fts3AllSegdirs(Fts3Table*, int, int, int, sqlite3_stmt **); +SQLITE_PRIVATE int sqlite3Fts3ReadBlock(Fts3Table*, sqlite3_int64, char **, int*, int*); + +SQLITE_PRIVATE int sqlite3Fts3SelectDoctotal(Fts3Table *, sqlite3_stmt **); +SQLITE_PRIVATE int sqlite3Fts3SelectDocsize(Fts3Table *, sqlite3_int64, sqlite3_stmt **); + +#ifndef SQLITE_DISABLE_FTS4_DEFERRED +SQLITE_PRIVATE void sqlite3Fts3FreeDeferredTokens(Fts3Cursor *); +SQLITE_PRIVATE int sqlite3Fts3DeferToken(Fts3Cursor *, Fts3PhraseToken *, int); +SQLITE_PRIVATE int sqlite3Fts3CacheDeferredDoclists(Fts3Cursor *); +SQLITE_PRIVATE void sqlite3Fts3FreeDeferredDoclists(Fts3Cursor *); +SQLITE_PRIVATE int sqlite3Fts3DeferredTokenList(Fts3DeferredToken *, char **, int *); +#else +# define sqlite3Fts3FreeDeferredTokens(x) +# define sqlite3Fts3DeferToken(x,y,z) SQLITE_OK +# define sqlite3Fts3CacheDeferredDoclists(x) SQLITE_OK +# define sqlite3Fts3FreeDeferredDoclists(x) +# define sqlite3Fts3DeferredTokenList(x,y,z) SQLITE_OK +#endif + +SQLITE_PRIVATE void sqlite3Fts3SegmentsClose(Fts3Table *); +SQLITE_PRIVATE int sqlite3Fts3MaxLevel(Fts3Table *, int *); + +/* Special values interpreted by sqlite3SegReaderCursor() */ +#define FTS3_SEGCURSOR_PENDING -1 +#define FTS3_SEGCURSOR_ALL -2 + +SQLITE_PRIVATE int sqlite3Fts3SegReaderStart(Fts3Table*, Fts3MultiSegReader*, Fts3SegFilter*); +SQLITE_PRIVATE int sqlite3Fts3SegReaderStep(Fts3Table *, Fts3MultiSegReader *); +SQLITE_PRIVATE void sqlite3Fts3SegReaderFinish(Fts3MultiSegReader *); + +SQLITE_PRIVATE int sqlite3Fts3SegReaderCursor(Fts3Table *, + int, int, int, const char *, int, int, int, Fts3MultiSegReader *); + +/* Flags allowed as part of the 4th argument to SegmentReaderIterate() */ +#define FTS3_SEGMENT_REQUIRE_POS 0x00000001 +#define FTS3_SEGMENT_IGNORE_EMPTY 0x00000002 +#define FTS3_SEGMENT_COLUMN_FILTER 0x00000004 +#define FTS3_SEGMENT_PREFIX 0x00000008 +#define FTS3_SEGMENT_SCAN 0x00000010 +#define FTS3_SEGMENT_FIRST 0x00000020 + +/* Type passed as 4th argument to SegmentReaderIterate() */ +struct Fts3SegFilter { + const char *zTerm; + int nTerm; + int iCol; + int flags; +}; + +struct Fts3MultiSegReader { + /* Used internally by sqlite3Fts3SegReaderXXX() calls */ + Fts3SegReader **apSegment; /* Array of Fts3SegReader objects */ + int nSegment; /* Size of apSegment array */ + int nAdvance; /* How many seg-readers to advance */ + Fts3SegFilter *pFilter; /* Pointer to filter object */ + char *aBuffer; /* Buffer to merge doclists in */ + int nBuffer; /* Allocated size of aBuffer[] in bytes */ + + int iColFilter; /* If >=0, filter for this column */ + int bRestart; + + /* Used by fts3.c only. */ + int nCost; /* Cost of running iterator */ + int bLookup; /* True if a lookup of a single entry. */ + + /* Output values. Valid only after Fts3SegReaderStep() returns SQLITE_ROW. */ + char *zTerm; /* Pointer to term buffer */ + int nTerm; /* Size of zTerm in bytes */ + char *aDoclist; /* Pointer to doclist buffer */ + int nDoclist; /* Size of aDoclist[] in bytes */ +}; + +SQLITE_PRIVATE int sqlite3Fts3Incrmerge(Fts3Table*,int,int); + +#define fts3GetVarint32(p, piVal) ( \ + (*(u8*)(p)&0x80) ? sqlite3Fts3GetVarint32(p, piVal) : (*piVal=*(u8*)(p), 1) \ +) + +/* fts3.c */ +SQLITE_PRIVATE void sqlite3Fts3ErrMsg(char**,const char*,...); +SQLITE_PRIVATE int sqlite3Fts3PutVarint(char *, sqlite3_int64); +SQLITE_PRIVATE int sqlite3Fts3GetVarint(const char *, sqlite_int64 *); +SQLITE_PRIVATE int sqlite3Fts3GetVarint32(const char *, int *); +SQLITE_PRIVATE int sqlite3Fts3VarintLen(sqlite3_uint64); +SQLITE_PRIVATE void sqlite3Fts3Dequote(char *); +SQLITE_PRIVATE void sqlite3Fts3DoclistPrev(int,char*,int,char**,sqlite3_int64*,int*,u8*); +SQLITE_PRIVATE int sqlite3Fts3EvalPhraseStats(Fts3Cursor *, Fts3Expr *, u32 *); +SQLITE_PRIVATE int sqlite3Fts3FirstFilter(sqlite3_int64, char *, int, char *); +SQLITE_PRIVATE void sqlite3Fts3CreateStatTable(int*, Fts3Table*); +SQLITE_PRIVATE int sqlite3Fts3EvalTestDeferred(Fts3Cursor *pCsr, int *pRc); + +/* fts3_tokenizer.c */ +SQLITE_PRIVATE const char *sqlite3Fts3NextToken(const char *, int *); +SQLITE_PRIVATE int sqlite3Fts3InitHashTable(sqlite3 *, Fts3Hash *, const char *); +SQLITE_PRIVATE int sqlite3Fts3InitTokenizer(Fts3Hash *pHash, const char *, + sqlite3_tokenizer **, char ** +); +SQLITE_PRIVATE int sqlite3Fts3IsIdChar(char); + +/* fts3_snippet.c */ +SQLITE_PRIVATE void sqlite3Fts3Offsets(sqlite3_context*, Fts3Cursor*); +SQLITE_PRIVATE void sqlite3Fts3Snippet(sqlite3_context *, Fts3Cursor *, const char *, + const char *, const char *, int, int +); +SQLITE_PRIVATE void sqlite3Fts3Matchinfo(sqlite3_context *, Fts3Cursor *, const char *); +SQLITE_PRIVATE void sqlite3Fts3MIBufferFree(MatchinfoBuffer *p); + +/* fts3_expr.c */ +SQLITE_PRIVATE int sqlite3Fts3ExprParse(sqlite3_tokenizer *, int, + char **, int, int, int, const char *, int, Fts3Expr **, char ** +); +SQLITE_PRIVATE void sqlite3Fts3ExprFree(Fts3Expr *); +#ifdef SQLITE_TEST +SQLITE_PRIVATE int sqlite3Fts3ExprInitTestInterface(sqlite3 *db, Fts3Hash*); +SQLITE_PRIVATE int sqlite3Fts3InitTerm(sqlite3 *db); +#endif + +SQLITE_PRIVATE int sqlite3Fts3OpenTokenizer(sqlite3_tokenizer *, int, const char *, int, + sqlite3_tokenizer_cursor ** +); + +/* fts3_aux.c */ +SQLITE_PRIVATE int sqlite3Fts3InitAux(sqlite3 *db); + +SQLITE_PRIVATE void sqlite3Fts3EvalPhraseCleanup(Fts3Phrase *); + +SQLITE_PRIVATE int sqlite3Fts3MsrIncrStart( + Fts3Table*, Fts3MultiSegReader*, int, const char*, int); +SQLITE_PRIVATE int sqlite3Fts3MsrIncrNext( + Fts3Table *, Fts3MultiSegReader *, sqlite3_int64 *, char **, int *); +SQLITE_PRIVATE int sqlite3Fts3EvalPhrasePoslist(Fts3Cursor *, Fts3Expr *, int iCol, char **); +SQLITE_PRIVATE int sqlite3Fts3MsrOvfl(Fts3Cursor *, Fts3MultiSegReader *, int *); +SQLITE_PRIVATE int sqlite3Fts3MsrIncrRestart(Fts3MultiSegReader *pCsr); + +/* fts3_tokenize_vtab.c */ +SQLITE_PRIVATE int sqlite3Fts3InitTok(sqlite3*, Fts3Hash *); + +/* fts3_unicode2.c (functions generated by parsing unicode text files) */ +#ifndef SQLITE_DISABLE_FTS3_UNICODE +SQLITE_PRIVATE int sqlite3FtsUnicodeFold(int, int); +SQLITE_PRIVATE int sqlite3FtsUnicodeIsalnum(int); +SQLITE_PRIVATE int sqlite3FtsUnicodeIsdiacritic(int); +#endif + +#endif /* !SQLITE_CORE || SQLITE_ENABLE_FTS3 */ +#endif /* _FTSINT_H */ + +/************** End of fts3Int.h *********************************************/ +/************** Continuing where we left off in fts3.c ***********************/ +#if !defined(SQLITE_CORE) || defined(SQLITE_ENABLE_FTS3) + +#if defined(SQLITE_ENABLE_FTS3) && !defined(SQLITE_CORE) +# define SQLITE_CORE 1 +#endif + +/* #include */ +/* #include */ +/* #include */ +/* #include */ +/* #include */ +/* #include */ + +/* #include "fts3.h" */ +#ifndef SQLITE_CORE +/* # include "sqlite3ext.h" */ + SQLITE_EXTENSION_INIT1 +#endif + +/* +** The following are copied from sqliteInt.h. +** +** Constants for the largest and smallest possible 64-bit signed integers. +** These macros are designed to work correctly on both 32-bit and 64-bit +** compilers. +*/ +#ifndef SQLITE_AMALGAMATION +# define LARGEST_INT64 (0xffffffff|(((sqlite3_int64)0x7fffffff)<<32)) +# define SMALLEST_INT64 (((sqlite3_int64)-1) - LARGEST_INT64) +#endif + +static int fts3EvalNext(Fts3Cursor *pCsr); +static int fts3EvalStart(Fts3Cursor *pCsr); +static int fts3TermSegReaderCursor( + Fts3Cursor *, const char *, int, int, Fts3MultiSegReader **); + +#ifndef SQLITE_AMALGAMATION +# if defined(SQLITE_DEBUG) +SQLITE_PRIVATE int sqlite3Fts3Always(int b) { assert( b ); return b; } +SQLITE_PRIVATE int sqlite3Fts3Never(int b) { assert( !b ); return b; } +# endif +#endif + +/* +** This variable is set to false when running tests for which the on disk +** structures should not be corrupt. Otherwise, true. If it is false, extra +** assert() conditions in the fts3 code are activated - conditions that are +** only true if it is guaranteed that the fts3 database is not corrupt. +*/ +SQLITE_API int sqlite3_fts3_may_be_corrupt = 1; + +/* +** Write a 64-bit variable-length integer to memory starting at p[0]. +** The length of data written will be between 1 and FTS3_VARINT_MAX bytes. +** The number of bytes written is returned. +*/ +SQLITE_PRIVATE int sqlite3Fts3PutVarint(char *p, sqlite_int64 v){ + unsigned char *q = (unsigned char *) p; + sqlite_uint64 vu = v; + do{ + *q++ = (unsigned char) ((vu & 0x7f) | 0x80); + vu >>= 7; + }while( vu!=0 ); + q[-1] &= 0x7f; /* turn off high bit in final byte */ + assert( q - (unsigned char *)p <= FTS3_VARINT_MAX ); + return (int) (q - (unsigned char *)p); +} + +#define GETVARINT_STEP(v, ptr, shift, mask1, mask2, var, ret) \ + v = (v & mask1) | ( (*(const unsigned char*)(ptr++)) << shift ); \ + if( (v & mask2)==0 ){ var = v; return ret; } +#define GETVARINT_INIT(v, ptr, shift, mask1, mask2, var, ret) \ + v = (*ptr++); \ + if( (v & mask2)==0 ){ var = v; return ret; } + +/* +** Read a 64-bit variable-length integer from memory starting at p[0]. +** Return the number of bytes read, or 0 on error. +** The value is stored in *v. +*/ +SQLITE_PRIVATE int sqlite3Fts3GetVarint(const char *pBuf, sqlite_int64 *v){ + const unsigned char *p = (const unsigned char*)pBuf; + const unsigned char *pStart = p; + u32 a; + u64 b; + int shift; + + GETVARINT_INIT(a, p, 0, 0x00, 0x80, *v, 1); + GETVARINT_STEP(a, p, 7, 0x7F, 0x4000, *v, 2); + GETVARINT_STEP(a, p, 14, 0x3FFF, 0x200000, *v, 3); + GETVARINT_STEP(a, p, 21, 0x1FFFFF, 0x10000000, *v, 4); + b = (a & 0x0FFFFFFF ); + + for(shift=28; shift<=63; shift+=7){ + u64 c = *p++; + b += (c&0x7F) << shift; + if( (c & 0x80)==0 ) break; + } + *v = b; + return (int)(p - pStart); +} + +/* +** Similar to sqlite3Fts3GetVarint(), except that the output is truncated to +** a non-negative 32-bit integer before it is returned. +*/ +SQLITE_PRIVATE int sqlite3Fts3GetVarint32(const char *p, int *pi){ + const unsigned char *ptr = (const unsigned char*)p; + u32 a; + +#ifndef fts3GetVarint32 + GETVARINT_INIT(a, ptr, 0, 0x00, 0x80, *pi, 1); +#else + a = (*ptr++); + assert( a & 0x80 ); +#endif + + GETVARINT_STEP(a, ptr, 7, 0x7F, 0x4000, *pi, 2); + GETVARINT_STEP(a, ptr, 14, 0x3FFF, 0x200000, *pi, 3); + GETVARINT_STEP(a, ptr, 21, 0x1FFFFF, 0x10000000, *pi, 4); + a = (a & 0x0FFFFFFF ); + *pi = (int)(a | ((u32)(*ptr & 0x07) << 28)); + assert( 0==(a & 0x80000000) ); + assert( *pi>=0 ); + return 5; +} + +/* +** Return the number of bytes required to encode v as a varint +*/ +SQLITE_PRIVATE int sqlite3Fts3VarintLen(sqlite3_uint64 v){ + int i = 0; + do{ + i++; + v >>= 7; + }while( v!=0 ); + return i; +} + +/* +** Convert an SQL-style quoted string into a normal string by removing +** the quote characters. The conversion is done in-place. If the +** input does not begin with a quote character, then this routine +** is a no-op. +** +** Examples: +** +** "abc" becomes abc +** 'xyz' becomes xyz +** [pqr] becomes pqr +** `mno` becomes mno +** +*/ +SQLITE_PRIVATE void sqlite3Fts3Dequote(char *z){ + char quote; /* Quote character (if any ) */ + + quote = z[0]; + if( quote=='[' || quote=='\'' || quote=='"' || quote=='`' ){ + int iIn = 1; /* Index of next byte to read from input */ + int iOut = 0; /* Index of next byte to write to output */ + + /* If the first byte was a '[', then the close-quote character is a ']' */ + if( quote=='[' ) quote = ']'; + + while( z[iIn] ){ + if( z[iIn]==quote ){ + if( z[iIn+1]!=quote ) break; + z[iOut++] = quote; + iIn += 2; + }else{ + z[iOut++] = z[iIn++]; + } + } + z[iOut] = '\0'; + } +} + +/* +** Read a single varint from the doclist at *pp and advance *pp to point +** to the first byte past the end of the varint. Add the value of the varint +** to *pVal. +*/ +static void fts3GetDeltaVarint(char **pp, sqlite3_int64 *pVal){ + sqlite3_int64 iVal; + *pp += sqlite3Fts3GetVarint(*pp, &iVal); + *pVal += iVal; +} + +/* +** When this function is called, *pp points to the first byte following a +** varint that is part of a doclist (or position-list, or any other list +** of varints). This function moves *pp to point to the start of that varint, +** and sets *pVal by the varint value. +** +** Argument pStart points to the first byte of the doclist that the +** varint is part of. +*/ +static void fts3GetReverseVarint( + char **pp, + char *pStart, + sqlite3_int64 *pVal +){ + sqlite3_int64 iVal; + char *p; + + /* Pointer p now points at the first byte past the varint we are + ** interested in. So, unless the doclist is corrupt, the 0x80 bit is + ** clear on character p[-1]. */ + for(p = (*pp)-2; p>=pStart && *p&0x80; p--); + p++; + *pp = p; + + sqlite3Fts3GetVarint(p, &iVal); + *pVal = iVal; +} + +/* +** The xDisconnect() virtual table method. +*/ +static int fts3DisconnectMethod(sqlite3_vtab *pVtab){ + Fts3Table *p = (Fts3Table *)pVtab; + int i; + + assert( p->nPendingData==0 ); + assert( p->pSegments==0 ); + + /* Free any prepared statements held */ + sqlite3_finalize(p->pSeekStmt); + for(i=0; iaStmt); i++){ + sqlite3_finalize(p->aStmt[i]); + } + sqlite3_free(p->zSegmentsTbl); + sqlite3_free(p->zReadExprlist); + sqlite3_free(p->zWriteExprlist); + sqlite3_free(p->zContentTbl); + sqlite3_free(p->zLanguageid); + + /* Invoke the tokenizer destructor to free the tokenizer. */ + p->pTokenizer->pModule->xDestroy(p->pTokenizer); + + sqlite3_free(p); + return SQLITE_OK; +} + +/* +** Write an error message into *pzErr +*/ +SQLITE_PRIVATE void sqlite3Fts3ErrMsg(char **pzErr, const char *zFormat, ...){ + va_list ap; + sqlite3_free(*pzErr); + va_start(ap, zFormat); + *pzErr = sqlite3_vmprintf(zFormat, ap); + va_end(ap); +} + +/* +** Construct one or more SQL statements from the format string given +** and then evaluate those statements. The success code is written +** into *pRc. +** +** If *pRc is initially non-zero then this routine is a no-op. +*/ +static void fts3DbExec( + int *pRc, /* Success code */ + sqlite3 *db, /* Database in which to run SQL */ + const char *zFormat, /* Format string for SQL */ + ... /* Arguments to the format string */ +){ + va_list ap; + char *zSql; + if( *pRc ) return; + va_start(ap, zFormat); + zSql = sqlite3_vmprintf(zFormat, ap); + va_end(ap); + if( zSql==0 ){ + *pRc = SQLITE_NOMEM; + }else{ + *pRc = sqlite3_exec(db, zSql, 0, 0, 0); + sqlite3_free(zSql); + } +} + +/* +** The xDestroy() virtual table method. +*/ +static int fts3DestroyMethod(sqlite3_vtab *pVtab){ + Fts3Table *p = (Fts3Table *)pVtab; + int rc = SQLITE_OK; /* Return code */ + const char *zDb = p->zDb; /* Name of database (e.g. "main", "temp") */ + sqlite3 *db = p->db; /* Database handle */ + + /* Drop the shadow tables */ + fts3DbExec(&rc, db, + "DROP TABLE IF EXISTS %Q.'%q_segments';" + "DROP TABLE IF EXISTS %Q.'%q_segdir';" + "DROP TABLE IF EXISTS %Q.'%q_docsize';" + "DROP TABLE IF EXISTS %Q.'%q_stat';" + "%s DROP TABLE IF EXISTS %Q.'%q_content';", + zDb, p->zName, + zDb, p->zName, + zDb, p->zName, + zDb, p->zName, + (p->zContentTbl ? "--" : ""), zDb,p->zName + ); + + /* If everything has worked, invoke fts3DisconnectMethod() to free the + ** memory associated with the Fts3Table structure and return SQLITE_OK. + ** Otherwise, return an SQLite error code. + */ + return (rc==SQLITE_OK ? fts3DisconnectMethod(pVtab) : rc); +} + + +/* +** Invoke sqlite3_declare_vtab() to declare the schema for the FTS3 table +** passed as the first argument. This is done as part of the xConnect() +** and xCreate() methods. +** +** If *pRc is non-zero when this function is called, it is a no-op. +** Otherwise, if an error occurs, an SQLite error code is stored in *pRc +** before returning. +*/ +static void fts3DeclareVtab(int *pRc, Fts3Table *p){ + if( *pRc==SQLITE_OK ){ + int i; /* Iterator variable */ + int rc; /* Return code */ + char *zSql; /* SQL statement passed to declare_vtab() */ + char *zCols; /* List of user defined columns */ + const char *zLanguageid; + + zLanguageid = (p->zLanguageid ? p->zLanguageid : "__langid"); + sqlite3_vtab_config(p->db, SQLITE_VTAB_CONSTRAINT_SUPPORT, 1); + + /* Create a list of user columns for the virtual table */ + zCols = sqlite3_mprintf("%Q, ", p->azColumn[0]); + for(i=1; zCols && inColumn; i++){ + zCols = sqlite3_mprintf("%z%Q, ", zCols, p->azColumn[i]); + } + + /* Create the whole "CREATE TABLE" statement to pass to SQLite */ + zSql = sqlite3_mprintf( + "CREATE TABLE x(%s %Q HIDDEN, docid HIDDEN, %Q HIDDEN)", + zCols, p->zName, zLanguageid + ); + if( !zCols || !zSql ){ + rc = SQLITE_NOMEM; + }else{ + rc = sqlite3_declare_vtab(p->db, zSql); + } + + sqlite3_free(zSql); + sqlite3_free(zCols); + *pRc = rc; + } +} + +/* +** Create the %_stat table if it does not already exist. +*/ +SQLITE_PRIVATE void sqlite3Fts3CreateStatTable(int *pRc, Fts3Table *p){ + fts3DbExec(pRc, p->db, + "CREATE TABLE IF NOT EXISTS %Q.'%q_stat'" + "(id INTEGER PRIMARY KEY, value BLOB);", + p->zDb, p->zName + ); + if( (*pRc)==SQLITE_OK ) p->bHasStat = 1; +} + +/* +** Create the backing store tables (%_content, %_segments and %_segdir) +** required by the FTS3 table passed as the only argument. This is done +** as part of the vtab xCreate() method. +** +** If the p->bHasDocsize boolean is true (indicating that this is an +** FTS4 table, not an FTS3 table) then also create the %_docsize and +** %_stat tables required by FTS4. +*/ +static int fts3CreateTables(Fts3Table *p){ + int rc = SQLITE_OK; /* Return code */ + int i; /* Iterator variable */ + sqlite3 *db = p->db; /* The database connection */ + + if( p->zContentTbl==0 ){ + const char *zLanguageid = p->zLanguageid; + char *zContentCols; /* Columns of %_content table */ + + /* Create a list of user columns for the content table */ + zContentCols = sqlite3_mprintf("docid INTEGER PRIMARY KEY"); + for(i=0; zContentCols && inColumn; i++){ + char *z = p->azColumn[i]; + zContentCols = sqlite3_mprintf("%z, 'c%d%q'", zContentCols, i, z); + } + if( zLanguageid && zContentCols ){ + zContentCols = sqlite3_mprintf("%z, langid", zContentCols, zLanguageid); + } + if( zContentCols==0 ) rc = SQLITE_NOMEM; + + /* Create the content table */ + fts3DbExec(&rc, db, + "CREATE TABLE %Q.'%q_content'(%s)", + p->zDb, p->zName, zContentCols + ); + sqlite3_free(zContentCols); + } + + /* Create other tables */ + fts3DbExec(&rc, db, + "CREATE TABLE %Q.'%q_segments'(blockid INTEGER PRIMARY KEY, block BLOB);", + p->zDb, p->zName + ); + fts3DbExec(&rc, db, + "CREATE TABLE %Q.'%q_segdir'(" + "level INTEGER," + "idx INTEGER," + "start_block INTEGER," + "leaves_end_block INTEGER," + "end_block INTEGER," + "root BLOB," + "PRIMARY KEY(level, idx)" + ");", + p->zDb, p->zName + ); + if( p->bHasDocsize ){ + fts3DbExec(&rc, db, + "CREATE TABLE %Q.'%q_docsize'(docid INTEGER PRIMARY KEY, size BLOB);", + p->zDb, p->zName + ); + } + assert( p->bHasStat==p->bFts4 ); + if( p->bHasStat ){ + sqlite3Fts3CreateStatTable(&rc, p); + } + return rc; +} + +/* +** Store the current database page-size in bytes in p->nPgsz. +** +** If *pRc is non-zero when this function is called, it is a no-op. +** Otherwise, if an error occurs, an SQLite error code is stored in *pRc +** before returning. +*/ +static void fts3DatabasePageSize(int *pRc, Fts3Table *p){ + if( *pRc==SQLITE_OK ){ + int rc; /* Return code */ + char *zSql; /* SQL text "PRAGMA %Q.page_size" */ + sqlite3_stmt *pStmt; /* Compiled "PRAGMA %Q.page_size" statement */ + + zSql = sqlite3_mprintf("PRAGMA %Q.page_size", p->zDb); + if( !zSql ){ + rc = SQLITE_NOMEM; + }else{ + rc = sqlite3_prepare(p->db, zSql, -1, &pStmt, 0); + if( rc==SQLITE_OK ){ + sqlite3_step(pStmt); + p->nPgsz = sqlite3_column_int(pStmt, 0); + rc = sqlite3_finalize(pStmt); + }else if( rc==SQLITE_AUTH ){ + p->nPgsz = 1024; + rc = SQLITE_OK; + } + } + assert( p->nPgsz>0 || rc!=SQLITE_OK ); + sqlite3_free(zSql); + *pRc = rc; + } +} + +/* +** "Special" FTS4 arguments are column specifications of the following form: +** +** = +** +** There may not be whitespace surrounding the "=" character. The +** term may be quoted, but the may not. +*/ +static int fts3IsSpecialColumn( + const char *z, + int *pnKey, + char **pzValue +){ + char *zValue; + const char *zCsr = z; + + while( *zCsr!='=' ){ + if( *zCsr=='\0' ) return 0; + zCsr++; + } + + *pnKey = (int)(zCsr-z); + zValue = sqlite3_mprintf("%s", &zCsr[1]); + if( zValue ){ + sqlite3Fts3Dequote(zValue); + } + *pzValue = zValue; + return 1; +} + +/* +** Append the output of a printf() style formatting to an existing string. +*/ +static void fts3Appendf( + int *pRc, /* IN/OUT: Error code */ + char **pz, /* IN/OUT: Pointer to string buffer */ + const char *zFormat, /* Printf format string to append */ + ... /* Arguments for printf format string */ +){ + if( *pRc==SQLITE_OK ){ + va_list ap; + char *z; + va_start(ap, zFormat); + z = sqlite3_vmprintf(zFormat, ap); + va_end(ap); + if( z && *pz ){ + char *z2 = sqlite3_mprintf("%s%s", *pz, z); + sqlite3_free(z); + z = z2; + } + if( z==0 ) *pRc = SQLITE_NOMEM; + sqlite3_free(*pz); + *pz = z; + } +} + +/* +** Return a copy of input string zInput enclosed in double-quotes (") and +** with all double quote characters escaped. For example: +** +** fts3QuoteId("un \"zip\"") -> "un \"\"zip\"\"" +** +** The pointer returned points to memory obtained from sqlite3_malloc(). It +** is the callers responsibility to call sqlite3_free() to release this +** memory. +*/ +static char *fts3QuoteId(char const *zInput){ + sqlite3_int64 nRet; + char *zRet; + nRet = 2 + (int)strlen(zInput)*2 + 1; + zRet = sqlite3_malloc64(nRet); + if( zRet ){ + int i; + char *z = zRet; + *(z++) = '"'; + for(i=0; zInput[i]; i++){ + if( zInput[i]=='"' ) *(z++) = '"'; + *(z++) = zInput[i]; + } + *(z++) = '"'; + *(z++) = '\0'; + } + return zRet; +} + +/* +** Return a list of comma separated SQL expressions and a FROM clause that +** could be used in a SELECT statement such as the following: +** +** SELECT FROM %_content AS x ... +** +** to return the docid, followed by each column of text data in order +** from left to write. If parameter zFunc is not NULL, then instead of +** being returned directly each column of text data is passed to an SQL +** function named zFunc first. For example, if zFunc is "unzip" and the +** table has the three user-defined columns "a", "b", and "c", the following +** string is returned: +** +** "docid, unzip(x.'a'), unzip(x.'b'), unzip(x.'c') FROM %_content AS x" +** +** The pointer returned points to a buffer allocated by sqlite3_malloc(). It +** is the responsibility of the caller to eventually free it. +** +** If *pRc is not SQLITE_OK when this function is called, it is a no-op (and +** a NULL pointer is returned). Otherwise, if an OOM error is encountered +** by this function, NULL is returned and *pRc is set to SQLITE_NOMEM. If +** no error occurs, *pRc is left unmodified. +*/ +static char *fts3ReadExprList(Fts3Table *p, const char *zFunc, int *pRc){ + char *zRet = 0; + char *zFree = 0; + char *zFunction; + int i; + + if( p->zContentTbl==0 ){ + if( !zFunc ){ + zFunction = ""; + }else{ + zFree = zFunction = fts3QuoteId(zFunc); + } + fts3Appendf(pRc, &zRet, "docid"); + for(i=0; inColumn; i++){ + fts3Appendf(pRc, &zRet, ",%s(x.'c%d%q')", zFunction, i, p->azColumn[i]); + } + if( p->zLanguageid ){ + fts3Appendf(pRc, &zRet, ", x.%Q", "langid"); + } + sqlite3_free(zFree); + }else{ + fts3Appendf(pRc, &zRet, "rowid"); + for(i=0; inColumn; i++){ + fts3Appendf(pRc, &zRet, ", x.'%q'", p->azColumn[i]); + } + if( p->zLanguageid ){ + fts3Appendf(pRc, &zRet, ", x.%Q", p->zLanguageid); + } + } + fts3Appendf(pRc, &zRet, " FROM '%q'.'%q%s' AS x", + p->zDb, + (p->zContentTbl ? p->zContentTbl : p->zName), + (p->zContentTbl ? "" : "_content") + ); + return zRet; +} + +/* +** Return a list of N comma separated question marks, where N is the number +** of columns in the %_content table (one for the docid plus one for each +** user-defined text column). +** +** If argument zFunc is not NULL, then all but the first question mark +** is preceded by zFunc and an open bracket, and followed by a closed +** bracket. For example, if zFunc is "zip" and the FTS3 table has three +** user-defined text columns, the following string is returned: +** +** "?, zip(?), zip(?), zip(?)" +** +** The pointer returned points to a buffer allocated by sqlite3_malloc(). It +** is the responsibility of the caller to eventually free it. +** +** If *pRc is not SQLITE_OK when this function is called, it is a no-op (and +** a NULL pointer is returned). Otherwise, if an OOM error is encountered +** by this function, NULL is returned and *pRc is set to SQLITE_NOMEM. If +** no error occurs, *pRc is left unmodified. +*/ +static char *fts3WriteExprList(Fts3Table *p, const char *zFunc, int *pRc){ + char *zRet = 0; + char *zFree = 0; + char *zFunction; + int i; + + if( !zFunc ){ + zFunction = ""; + }else{ + zFree = zFunction = fts3QuoteId(zFunc); + } + fts3Appendf(pRc, &zRet, "?"); + for(i=0; inColumn; i++){ + fts3Appendf(pRc, &zRet, ",%s(?)", zFunction); + } + if( p->zLanguageid ){ + fts3Appendf(pRc, &zRet, ", ?"); + } + sqlite3_free(zFree); + return zRet; +} + +/* +** This function interprets the string at (*pp) as a non-negative integer +** value. It reads the integer and sets *pnOut to the value read, then +** sets *pp to point to the byte immediately following the last byte of +** the integer value. +** +** Only decimal digits ('0'..'9') may be part of an integer value. +** +** If *pp does not being with a decimal digit SQLITE_ERROR is returned and +** the output value undefined. Otherwise SQLITE_OK is returned. +** +** This function is used when parsing the "prefix=" FTS4 parameter. +*/ +static int fts3GobbleInt(const char **pp, int *pnOut){ + const int MAX_NPREFIX = 10000000; + const char *p; /* Iterator pointer */ + int nInt = 0; /* Output value */ + + for(p=*pp; p[0]>='0' && p[0]<='9'; p++){ + nInt = nInt * 10 + (p[0] - '0'); + if( nInt>MAX_NPREFIX ){ + nInt = 0; + break; + } + } + if( p==*pp ) return SQLITE_ERROR; + *pnOut = nInt; + *pp = p; + return SQLITE_OK; +} + +/* +** This function is called to allocate an array of Fts3Index structures +** representing the indexes maintained by the current FTS table. FTS tables +** always maintain the main "terms" index, but may also maintain one or +** more "prefix" indexes, depending on the value of the "prefix=" parameter +** (if any) specified as part of the CREATE VIRTUAL TABLE statement. +** +** Argument zParam is passed the value of the "prefix=" option if one was +** specified, or NULL otherwise. +** +** If no error occurs, SQLITE_OK is returned and *apIndex set to point to +** the allocated array. *pnIndex is set to the number of elements in the +** array. If an error does occur, an SQLite error code is returned. +** +** Regardless of whether or not an error is returned, it is the responsibility +** of the caller to call sqlite3_free() on the output array to free it. +*/ +static int fts3PrefixParameter( + const char *zParam, /* ABC in prefix=ABC parameter to parse */ + int *pnIndex, /* OUT: size of *apIndex[] array */ + struct Fts3Index **apIndex /* OUT: Array of indexes for this table */ +){ + struct Fts3Index *aIndex; /* Allocated array */ + int nIndex = 1; /* Number of entries in array */ + + if( zParam && zParam[0] ){ + const char *p; + nIndex++; + for(p=zParam; *p; p++){ + if( *p==',' ) nIndex++; + } + } + + aIndex = sqlite3_malloc64(sizeof(struct Fts3Index) * nIndex); + *apIndex = aIndex; + if( !aIndex ){ + return SQLITE_NOMEM; + } + + memset(aIndex, 0, sizeof(struct Fts3Index) * nIndex); + if( zParam ){ + const char *p = zParam; + int i; + for(i=1; i=0 ); + if( nPrefix==0 ){ + nIndex--; + i--; + }else{ + aIndex[i].nPrefix = nPrefix; + } + p++; + } + } + + *pnIndex = nIndex; + return SQLITE_OK; +} + +/* +** This function is called when initializing an FTS4 table that uses the +** content=xxx option. It determines the number of and names of the columns +** of the new FTS4 table. +** +** The third argument passed to this function is the value passed to the +** config=xxx option (i.e. "xxx"). This function queries the database for +** a table of that name. If found, the output variables are populated +** as follows: +** +** *pnCol: Set to the number of columns table xxx has, +** +** *pnStr: Set to the total amount of space required to store a copy +** of each columns name, including the nul-terminator. +** +** *pazCol: Set to point to an array of *pnCol strings. Each string is +** the name of the corresponding column in table xxx. The array +** and its contents are allocated using a single allocation. It +** is the responsibility of the caller to free this allocation +** by eventually passing the *pazCol value to sqlite3_free(). +** +** If the table cannot be found, an error code is returned and the output +** variables are undefined. Or, if an OOM is encountered, SQLITE_NOMEM is +** returned (and the output variables are undefined). +*/ +static int fts3ContentColumns( + sqlite3 *db, /* Database handle */ + const char *zDb, /* Name of db (i.e. "main", "temp" etc.) */ + const char *zTbl, /* Name of content table */ + const char ***pazCol, /* OUT: Malloc'd array of column names */ + int *pnCol, /* OUT: Size of array *pazCol */ + int *pnStr, /* OUT: Bytes of string content */ + char **pzErr /* OUT: error message */ +){ + int rc = SQLITE_OK; /* Return code */ + char *zSql; /* "SELECT *" statement on zTbl */ + sqlite3_stmt *pStmt = 0; /* Compiled version of zSql */ + + zSql = sqlite3_mprintf("SELECT * FROM %Q.%Q", zDb, zTbl); + if( !zSql ){ + rc = SQLITE_NOMEM; + }else{ + rc = sqlite3_prepare(db, zSql, -1, &pStmt, 0); + if( rc!=SQLITE_OK ){ + sqlite3Fts3ErrMsg(pzErr, "%s", sqlite3_errmsg(db)); + } + } + sqlite3_free(zSql); + + if( rc==SQLITE_OK ){ + const char **azCol; /* Output array */ + sqlite3_int64 nStr = 0; /* Size of all column names (incl. 0x00) */ + int nCol; /* Number of table columns */ + int i; /* Used to iterate through columns */ + + /* Loop through the returned columns. Set nStr to the number of bytes of + ** space required to store a copy of each column name, including the + ** nul-terminator byte. */ + nCol = sqlite3_column_count(pStmt); + for(i=0; i module name ("fts3" or "fts4") +** argv[1] -> database name +** argv[2] -> table name +** argv[...] -> "column name" and other module argument fields. +*/ +static int fts3InitVtab( + int isCreate, /* True for xCreate, false for xConnect */ + sqlite3 *db, /* The SQLite database connection */ + void *pAux, /* Hash table containing tokenizers */ + int argc, /* Number of elements in argv array */ + const char * const *argv, /* xCreate/xConnect argument array */ + sqlite3_vtab **ppVTab, /* Write the resulting vtab structure here */ + char **pzErr /* Write any error message here */ +){ + Fts3Hash *pHash = (Fts3Hash *)pAux; + Fts3Table *p = 0; /* Pointer to allocated vtab */ + int rc = SQLITE_OK; /* Return code */ + int i; /* Iterator variable */ + sqlite3_int64 nByte; /* Size of allocation used for *p */ + int iCol; /* Column index */ + int nString = 0; /* Bytes required to hold all column names */ + int nCol = 0; /* Number of columns in the FTS table */ + char *zCsr; /* Space for holding column names */ + int nDb; /* Bytes required to hold database name */ + int nName; /* Bytes required to hold table name */ + int isFts4 = (argv[0][3]=='4'); /* True for FTS4, false for FTS3 */ + const char **aCol; /* Array of column names */ + sqlite3_tokenizer *pTokenizer = 0; /* Tokenizer for this table */ + + int nIndex = 0; /* Size of aIndex[] array */ + struct Fts3Index *aIndex = 0; /* Array of indexes for this table */ + + /* The results of parsing supported FTS4 key=value options: */ + int bNoDocsize = 0; /* True to omit %_docsize table */ + int bDescIdx = 0; /* True to store descending indexes */ + char *zPrefix = 0; /* Prefix parameter value (or NULL) */ + char *zCompress = 0; /* compress=? parameter (or NULL) */ + char *zUncompress = 0; /* uncompress=? parameter (or NULL) */ + char *zContent = 0; /* content=? parameter (or NULL) */ + char *zLanguageid = 0; /* languageid=? parameter (or NULL) */ + char **azNotindexed = 0; /* The set of notindexed= columns */ + int nNotindexed = 0; /* Size of azNotindexed[] array */ + + assert( strlen(argv[0])==4 ); + assert( (sqlite3_strnicmp(argv[0], "fts4", 4)==0 && isFts4) + || (sqlite3_strnicmp(argv[0], "fts3", 4)==0 && !isFts4) + ); + + nDb = (int)strlen(argv[1]) + 1; + nName = (int)strlen(argv[2]) + 1; + + nByte = sizeof(const char *) * (argc-2); + aCol = (const char **)sqlite3_malloc64(nByte); + if( aCol ){ + memset((void*)aCol, 0, nByte); + azNotindexed = (char **)sqlite3_malloc64(nByte); + } + if( azNotindexed ){ + memset(azNotindexed, 0, nByte); + } + if( !aCol || !azNotindexed ){ + rc = SQLITE_NOMEM; + goto fts3_init_out; + } + + /* Loop through all of the arguments passed by the user to the FTS3/4 + ** module (i.e. all the column names and special arguments). This loop + ** does the following: + ** + ** + Figures out the number of columns the FTSX table will have, and + ** the number of bytes of space that must be allocated to store copies + ** of the column names. + ** + ** + If there is a tokenizer specification included in the arguments, + ** initializes the tokenizer pTokenizer. + */ + for(i=3; rc==SQLITE_OK && i8 + && 0==sqlite3_strnicmp(z, "tokenize", 8) + && 0==sqlite3Fts3IsIdChar(z[8]) + ){ + rc = sqlite3Fts3InitTokenizer(pHash, &z[9], &pTokenizer, pzErr); + } + + /* Check if it is an FTS4 special argument. */ + else if( isFts4 && fts3IsSpecialColumn(z, &nKey, &zVal) ){ + struct Fts4Option { + const char *zOpt; + int nOpt; + } aFts4Opt[] = { + { "matchinfo", 9 }, /* 0 -> MATCHINFO */ + { "prefix", 6 }, /* 1 -> PREFIX */ + { "compress", 8 }, /* 2 -> COMPRESS */ + { "uncompress", 10 }, /* 3 -> UNCOMPRESS */ + { "order", 5 }, /* 4 -> ORDER */ + { "content", 7 }, /* 5 -> CONTENT */ + { "languageid", 10 }, /* 6 -> LANGUAGEID */ + { "notindexed", 10 } /* 7 -> NOTINDEXED */ + }; + + int iOpt; + if( !zVal ){ + rc = SQLITE_NOMEM; + }else{ + for(iOpt=0; iOptnOpt && !sqlite3_strnicmp(z, pOp->zOpt, pOp->nOpt) ){ + break; + } + } + switch( iOpt ){ + case 0: /* MATCHINFO */ + if( strlen(zVal)!=4 || sqlite3_strnicmp(zVal, "fts3", 4) ){ + sqlite3Fts3ErrMsg(pzErr, "unrecognized matchinfo: %s", zVal); + rc = SQLITE_ERROR; + } + bNoDocsize = 1; + break; + + case 1: /* PREFIX */ + sqlite3_free(zPrefix); + zPrefix = zVal; + zVal = 0; + break; + + case 2: /* COMPRESS */ + sqlite3_free(zCompress); + zCompress = zVal; + zVal = 0; + break; + + case 3: /* UNCOMPRESS */ + sqlite3_free(zUncompress); + zUncompress = zVal; + zVal = 0; + break; + + case 4: /* ORDER */ + if( (strlen(zVal)!=3 || sqlite3_strnicmp(zVal, "asc", 3)) + && (strlen(zVal)!=4 || sqlite3_strnicmp(zVal, "desc", 4)) + ){ + sqlite3Fts3ErrMsg(pzErr, "unrecognized order: %s", zVal); + rc = SQLITE_ERROR; + } + bDescIdx = (zVal[0]=='d' || zVal[0]=='D'); + break; + + case 5: /* CONTENT */ + sqlite3_free(zContent); + zContent = zVal; + zVal = 0; + break; + + case 6: /* LANGUAGEID */ + assert( iOpt==6 ); + sqlite3_free(zLanguageid); + zLanguageid = zVal; + zVal = 0; + break; + + case 7: /* NOTINDEXED */ + azNotindexed[nNotindexed++] = zVal; + zVal = 0; + break; + + default: + assert( iOpt==SizeofArray(aFts4Opt) ); + sqlite3Fts3ErrMsg(pzErr, "unrecognized parameter: %s", z); + rc = SQLITE_ERROR; + break; + } + sqlite3_free(zVal); + } + } + + /* Otherwise, the argument is a column name. */ + else { + nString += (int)(strlen(z) + 1); + aCol[nCol++] = z; + } + } + + /* If a content=xxx option was specified, the following: + ** + ** 1. Ignore any compress= and uncompress= options. + ** + ** 2. If no column names were specified as part of the CREATE VIRTUAL + ** TABLE statement, use all columns from the content table. + */ + if( rc==SQLITE_OK && zContent ){ + sqlite3_free(zCompress); + sqlite3_free(zUncompress); + zCompress = 0; + zUncompress = 0; + if( nCol==0 ){ + sqlite3_free((void*)aCol); + aCol = 0; + rc = fts3ContentColumns(db, argv[1], zContent,&aCol,&nCol,&nString,pzErr); + + /* If a languageid= option was specified, remove the language id + ** column from the aCol[] array. */ + if( rc==SQLITE_OK && zLanguageid ){ + int j; + for(j=0; jdb = db; + p->nColumn = nCol; + p->nPendingData = 0; + p->azColumn = (char **)&p[1]; + p->pTokenizer = pTokenizer; + p->nMaxPendingData = FTS3_MAX_PENDING_DATA; + p->bHasDocsize = (isFts4 && bNoDocsize==0); + p->bHasStat = (u8)isFts4; + p->bFts4 = (u8)isFts4; + p->bDescIdx = (u8)bDescIdx; + p->nAutoincrmerge = 0xff; /* 0xff means setting unknown */ + p->zContentTbl = zContent; + p->zLanguageid = zLanguageid; + zContent = 0; + zLanguageid = 0; + TESTONLY( p->inTransaction = -1 ); + TESTONLY( p->mxSavepoint = -1 ); + + p->aIndex = (struct Fts3Index *)&p->azColumn[nCol]; + memcpy(p->aIndex, aIndex, sizeof(struct Fts3Index) * nIndex); + p->nIndex = nIndex; + for(i=0; iaIndex[i].hPending, FTS3_HASH_STRING, 1); + } + p->abNotindexed = (u8 *)&p->aIndex[nIndex]; + + /* Fill in the zName and zDb fields of the vtab structure. */ + zCsr = (char *)&p->abNotindexed[nCol]; + p->zName = zCsr; + memcpy(zCsr, argv[2], nName); + zCsr += nName; + p->zDb = zCsr; + memcpy(zCsr, argv[1], nDb); + zCsr += nDb; + + /* Fill in the azColumn array */ + for(iCol=0; iCol0 ){ + memcpy(zCsr, z, n); + } + zCsr[n] = '\0'; + sqlite3Fts3Dequote(zCsr); + p->azColumn[iCol] = zCsr; + zCsr += n+1; + assert( zCsr <= &((char *)p)[nByte] ); + } + + /* Fill in the abNotindexed array */ + for(iCol=0; iColazColumn[iCol]); + for(i=0; iazColumn[iCol], zNot, n) + ){ + p->abNotindexed[iCol] = 1; + sqlite3_free(zNot); + azNotindexed[i] = 0; + } + } + } + for(i=0; izReadExprlist = fts3ReadExprList(p, zUncompress, &rc); + p->zWriteExprlist = fts3WriteExprList(p, zCompress, &rc); + if( rc!=SQLITE_OK ) goto fts3_init_out; + + /* If this is an xCreate call, create the underlying tables in the + ** database. TODO: For xConnect(), it could verify that said tables exist. + */ + if( isCreate ){ + rc = fts3CreateTables(p); + } + + /* Check to see if a legacy fts3 table has been "upgraded" by the + ** addition of a %_stat table so that it can use incremental merge. + */ + if( !isFts4 && !isCreate ){ + p->bHasStat = 2; + } + + /* Figure out the page-size for the database. This is required in order to + ** estimate the cost of loading large doclists from the database. */ + fts3DatabasePageSize(&rc, p); + p->nNodeSize = p->nPgsz-35; + + /* Declare the table schema to SQLite. */ + fts3DeclareVtab(&rc, p); + +fts3_init_out: + sqlite3_free(zPrefix); + sqlite3_free(aIndex); + sqlite3_free(zCompress); + sqlite3_free(zUncompress); + sqlite3_free(zContent); + sqlite3_free(zLanguageid); + for(i=0; ipModule->xDestroy(pTokenizer); + } + }else{ + assert( p->pSegments==0 ); + *ppVTab = &p->base; + } + return rc; +} + +/* +** The xConnect() and xCreate() methods for the virtual table. All the +** work is done in function fts3InitVtab(). +*/ +static int fts3ConnectMethod( + sqlite3 *db, /* Database connection */ + void *pAux, /* Pointer to tokenizer hash table */ + int argc, /* Number of elements in argv array */ + const char * const *argv, /* xCreate/xConnect argument array */ + sqlite3_vtab **ppVtab, /* OUT: New sqlite3_vtab object */ + char **pzErr /* OUT: sqlite3_malloc'd error message */ +){ + return fts3InitVtab(0, db, pAux, argc, argv, ppVtab, pzErr); +} +static int fts3CreateMethod( + sqlite3 *db, /* Database connection */ + void *pAux, /* Pointer to tokenizer hash table */ + int argc, /* Number of elements in argv array */ + const char * const *argv, /* xCreate/xConnect argument array */ + sqlite3_vtab **ppVtab, /* OUT: New sqlite3_vtab object */ + char **pzErr /* OUT: sqlite3_malloc'd error message */ +){ + return fts3InitVtab(1, db, pAux, argc, argv, ppVtab, pzErr); +} + +/* +** Set the pIdxInfo->estimatedRows variable to nRow. Unless this +** extension is currently being used by a version of SQLite too old to +** support estimatedRows. In that case this function is a no-op. +*/ +static void fts3SetEstimatedRows(sqlite3_index_info *pIdxInfo, i64 nRow){ +#if SQLITE_VERSION_NUMBER>=3008002 + if( sqlite3_libversion_number()>=3008002 ){ + pIdxInfo->estimatedRows = nRow; + } +#endif +} + +/* +** Set the SQLITE_INDEX_SCAN_UNIQUE flag in pIdxInfo->flags. Unless this +** extension is currently being used by a version of SQLite too old to +** support index-info flags. In that case this function is a no-op. +*/ +static void fts3SetUniqueFlag(sqlite3_index_info *pIdxInfo){ +#if SQLITE_VERSION_NUMBER>=3008012 + if( sqlite3_libversion_number()>=3008012 ){ + pIdxInfo->idxFlags |= SQLITE_INDEX_SCAN_UNIQUE; + } +#endif +} + +/* +** Implementation of the xBestIndex method for FTS3 tables. There +** are three possible strategies, in order of preference: +** +** 1. Direct lookup by rowid or docid. +** 2. Full-text search using a MATCH operator on a non-docid column. +** 3. Linear scan of %_content table. +*/ +static int fts3BestIndexMethod(sqlite3_vtab *pVTab, sqlite3_index_info *pInfo){ + Fts3Table *p = (Fts3Table *)pVTab; + int i; /* Iterator variable */ + int iCons = -1; /* Index of constraint to use */ + + int iLangidCons = -1; /* Index of langid=x constraint, if present */ + int iDocidGe = -1; /* Index of docid>=x constraint, if present */ + int iDocidLe = -1; /* Index of docid<=x constraint, if present */ + int iIdx; + + /* By default use a full table scan. This is an expensive option, + ** so search through the constraints to see if a more efficient + ** strategy is possible. + */ + pInfo->idxNum = FTS3_FULLSCAN_SEARCH; + pInfo->estimatedCost = 5000000; + for(i=0; inConstraint; i++){ + int bDocid; /* True if this constraint is on docid */ + struct sqlite3_index_constraint *pCons = &pInfo->aConstraint[i]; + if( pCons->usable==0 ){ + if( pCons->op==SQLITE_INDEX_CONSTRAINT_MATCH ){ + /* There exists an unusable MATCH constraint. This means that if + ** the planner does elect to use the results of this call as part + ** of the overall query plan the user will see an "unable to use + ** function MATCH in the requested context" error. To discourage + ** this, return a very high cost here. */ + pInfo->idxNum = FTS3_FULLSCAN_SEARCH; + pInfo->estimatedCost = 1e50; + fts3SetEstimatedRows(pInfo, ((sqlite3_int64)1) << 50); + return SQLITE_OK; + } + continue; + } + + bDocid = (pCons->iColumn<0 || pCons->iColumn==p->nColumn+1); + + /* A direct lookup on the rowid or docid column. Assign a cost of 1.0. */ + if( iCons<0 && pCons->op==SQLITE_INDEX_CONSTRAINT_EQ && bDocid ){ + pInfo->idxNum = FTS3_DOCID_SEARCH; + pInfo->estimatedCost = 1.0; + iCons = i; + } + + /* A MATCH constraint. Use a full-text search. + ** + ** If there is more than one MATCH constraint available, use the first + ** one encountered. If there is both a MATCH constraint and a direct + ** rowid/docid lookup, prefer the MATCH strategy. This is done even + ** though the rowid/docid lookup is faster than a MATCH query, selecting + ** it would lead to an "unable to use function MATCH in the requested + ** context" error. + */ + if( pCons->op==SQLITE_INDEX_CONSTRAINT_MATCH + && pCons->iColumn>=0 && pCons->iColumn<=p->nColumn + ){ + pInfo->idxNum = FTS3_FULLTEXT_SEARCH + pCons->iColumn; + pInfo->estimatedCost = 2.0; + iCons = i; + } + + /* Equality constraint on the langid column */ + if( pCons->op==SQLITE_INDEX_CONSTRAINT_EQ + && pCons->iColumn==p->nColumn + 2 + ){ + iLangidCons = i; + } + + if( bDocid ){ + switch( pCons->op ){ + case SQLITE_INDEX_CONSTRAINT_GE: + case SQLITE_INDEX_CONSTRAINT_GT: + iDocidGe = i; + break; + + case SQLITE_INDEX_CONSTRAINT_LE: + case SQLITE_INDEX_CONSTRAINT_LT: + iDocidLe = i; + break; + } + } + } + + /* If using a docid=? or rowid=? strategy, set the UNIQUE flag. */ + if( pInfo->idxNum==FTS3_DOCID_SEARCH ) fts3SetUniqueFlag(pInfo); + + iIdx = 1; + if( iCons>=0 ){ + pInfo->aConstraintUsage[iCons].argvIndex = iIdx++; + pInfo->aConstraintUsage[iCons].omit = 1; + } + if( iLangidCons>=0 ){ + pInfo->idxNum |= FTS3_HAVE_LANGID; + pInfo->aConstraintUsage[iLangidCons].argvIndex = iIdx++; + } + if( iDocidGe>=0 ){ + pInfo->idxNum |= FTS3_HAVE_DOCID_GE; + pInfo->aConstraintUsage[iDocidGe].argvIndex = iIdx++; + } + if( iDocidLe>=0 ){ + pInfo->idxNum |= FTS3_HAVE_DOCID_LE; + pInfo->aConstraintUsage[iDocidLe].argvIndex = iIdx++; + } + + /* Regardless of the strategy selected, FTS can deliver rows in rowid (or + ** docid) order. Both ascending and descending are possible. + */ + if( pInfo->nOrderBy==1 ){ + struct sqlite3_index_orderby *pOrder = &pInfo->aOrderBy[0]; + if( pOrder->iColumn<0 || pOrder->iColumn==p->nColumn+1 ){ + if( pOrder->desc ){ + pInfo->idxStr = "DESC"; + }else{ + pInfo->idxStr = "ASC"; + } + pInfo->orderByConsumed = 1; + } + } + + assert( p->pSegments==0 ); + return SQLITE_OK; +} + +/* +** Implementation of xOpen method. +*/ +static int fts3OpenMethod(sqlite3_vtab *pVTab, sqlite3_vtab_cursor **ppCsr){ + sqlite3_vtab_cursor *pCsr; /* Allocated cursor */ + + UNUSED_PARAMETER(pVTab); + + /* Allocate a buffer large enough for an Fts3Cursor structure. If the + ** allocation succeeds, zero it and return SQLITE_OK. Otherwise, + ** if the allocation fails, return SQLITE_NOMEM. + */ + *ppCsr = pCsr = (sqlite3_vtab_cursor *)sqlite3_malloc(sizeof(Fts3Cursor)); + if( !pCsr ){ + return SQLITE_NOMEM; + } + memset(pCsr, 0, sizeof(Fts3Cursor)); + return SQLITE_OK; +} + +/* +** Finalize the statement handle at pCsr->pStmt. +** +** Or, if that statement handle is one created by fts3CursorSeekStmt(), +** and the Fts3Table.pSeekStmt slot is currently NULL, save the statement +** pointer there instead of finalizing it. +*/ +static void fts3CursorFinalizeStmt(Fts3Cursor *pCsr){ + if( pCsr->bSeekStmt ){ + Fts3Table *p = (Fts3Table *)pCsr->base.pVtab; + if( p->pSeekStmt==0 ){ + p->pSeekStmt = pCsr->pStmt; + sqlite3_reset(pCsr->pStmt); + pCsr->pStmt = 0; + } + pCsr->bSeekStmt = 0; + } + sqlite3_finalize(pCsr->pStmt); +} + +/* +** Free all resources currently held by the cursor passed as the only +** argument. +*/ +static void fts3ClearCursor(Fts3Cursor *pCsr){ + fts3CursorFinalizeStmt(pCsr); + sqlite3Fts3FreeDeferredTokens(pCsr); + sqlite3_free(pCsr->aDoclist); + sqlite3Fts3MIBufferFree(pCsr->pMIBuffer); + sqlite3Fts3ExprFree(pCsr->pExpr); + memset(&(&pCsr->base)[1], 0, sizeof(Fts3Cursor)-sizeof(sqlite3_vtab_cursor)); +} + +/* +** Close the cursor. For additional information see the documentation +** on the xClose method of the virtual table interface. +*/ +static int fts3CloseMethod(sqlite3_vtab_cursor *pCursor){ + Fts3Cursor *pCsr = (Fts3Cursor *)pCursor; + assert( ((Fts3Table *)pCsr->base.pVtab)->pSegments==0 ); + fts3ClearCursor(pCsr); + assert( ((Fts3Table *)pCsr->base.pVtab)->pSegments==0 ); + sqlite3_free(pCsr); + return SQLITE_OK; +} + +/* +** If pCsr->pStmt has not been prepared (i.e. if pCsr->pStmt==0), then +** compose and prepare an SQL statement of the form: +** +** "SELECT FROM %_content WHERE rowid = ?" +** +** (or the equivalent for a content=xxx table) and set pCsr->pStmt to +** it. If an error occurs, return an SQLite error code. +*/ +static int fts3CursorSeekStmt(Fts3Cursor *pCsr){ + int rc = SQLITE_OK; + if( pCsr->pStmt==0 ){ + Fts3Table *p = (Fts3Table *)pCsr->base.pVtab; + char *zSql; + if( p->pSeekStmt ){ + pCsr->pStmt = p->pSeekStmt; + p->pSeekStmt = 0; + }else{ + zSql = sqlite3_mprintf("SELECT %s WHERE rowid = ?", p->zReadExprlist); + if( !zSql ) return SQLITE_NOMEM; + rc = sqlite3_prepare_v3(p->db, zSql,-1,SQLITE_PREPARE_PERSISTENT,&pCsr->pStmt,0); + sqlite3_free(zSql); + } + if( rc==SQLITE_OK ) pCsr->bSeekStmt = 1; + } + return rc; +} + +/* +** Position the pCsr->pStmt statement so that it is on the row +** of the %_content table that contains the last match. Return +** SQLITE_OK on success. +*/ +static int fts3CursorSeek(sqlite3_context *pContext, Fts3Cursor *pCsr){ + int rc = SQLITE_OK; + if( pCsr->isRequireSeek ){ + rc = fts3CursorSeekStmt(pCsr); + if( rc==SQLITE_OK ){ + sqlite3_bind_int64(pCsr->pStmt, 1, pCsr->iPrevId); + pCsr->isRequireSeek = 0; + if( SQLITE_ROW==sqlite3_step(pCsr->pStmt) ){ + return SQLITE_OK; + }else{ + rc = sqlite3_reset(pCsr->pStmt); + if( rc==SQLITE_OK && ((Fts3Table *)pCsr->base.pVtab)->zContentTbl==0 ){ + /* If no row was found and no error has occurred, then the %_content + ** table is missing a row that is present in the full-text index. + ** The data structures are corrupt. */ + rc = FTS_CORRUPT_VTAB; + pCsr->isEof = 1; + } + } + } + } + + if( rc!=SQLITE_OK && pContext ){ + sqlite3_result_error_code(pContext, rc); + } + return rc; +} + +/* +** This function is used to process a single interior node when searching +** a b-tree for a term or term prefix. The node data is passed to this +** function via the zNode/nNode parameters. The term to search for is +** passed in zTerm/nTerm. +** +** If piFirst is not NULL, then this function sets *piFirst to the blockid +** of the child node that heads the sub-tree that may contain the term. +** +** If piLast is not NULL, then *piLast is set to the right-most child node +** that heads a sub-tree that may contain a term for which zTerm/nTerm is +** a prefix. +** +** If an OOM error occurs, SQLITE_NOMEM is returned. Otherwise, SQLITE_OK. +*/ +static int fts3ScanInteriorNode( + const char *zTerm, /* Term to select leaves for */ + int nTerm, /* Size of term zTerm in bytes */ + const char *zNode, /* Buffer containing segment interior node */ + int nNode, /* Size of buffer at zNode */ + sqlite3_int64 *piFirst, /* OUT: Selected child node */ + sqlite3_int64 *piLast /* OUT: Selected child node */ +){ + int rc = SQLITE_OK; /* Return code */ + const char *zCsr = zNode; /* Cursor to iterate through node */ + const char *zEnd = &zCsr[nNode];/* End of interior node buffer */ + char *zBuffer = 0; /* Buffer to load terms into */ + i64 nAlloc = 0; /* Size of allocated buffer */ + int isFirstTerm = 1; /* True when processing first term on page */ + sqlite3_int64 iChild; /* Block id of child node to descend to */ + + /* Skip over the 'height' varint that occurs at the start of every + ** interior node. Then load the blockid of the left-child of the b-tree + ** node into variable iChild. + ** + ** Even if the data structure on disk is corrupted, this (reading two + ** varints from the buffer) does not risk an overread. If zNode is a + ** root node, then the buffer comes from a SELECT statement. SQLite does + ** not make this guarantee explicitly, but in practice there are always + ** either more than 20 bytes of allocated space following the nNode bytes of + ** contents, or two zero bytes. Or, if the node is read from the %_segments + ** table, then there are always 20 bytes of zeroed padding following the + ** nNode bytes of content (see sqlite3Fts3ReadBlock() for details). + */ + zCsr += sqlite3Fts3GetVarint(zCsr, &iChild); + zCsr += sqlite3Fts3GetVarint(zCsr, &iChild); + if( zCsr>zEnd ){ + return FTS_CORRUPT_VTAB; + } + + while( zCsr=0 && nSuffix>=0 ); + if( nPrefix>zCsr-zNode || nSuffix>zEnd-zCsr || nSuffix==0 ){ + rc = FTS_CORRUPT_VTAB; + goto finish_scan; + } + if( (i64)nPrefix+nSuffix>nAlloc ){ + char *zNew; + nAlloc = ((i64)nPrefix+nSuffix) * 2; + zNew = (char *)sqlite3_realloc64(zBuffer, nAlloc); + if( !zNew ){ + rc = SQLITE_NOMEM; + goto finish_scan; + } + zBuffer = zNew; + } + assert( zBuffer ); + memcpy(&zBuffer[nPrefix], zCsr, nSuffix); + nBuffer = nPrefix + nSuffix; + zCsr += nSuffix; + + /* Compare the term we are searching for with the term just loaded from + ** the interior node. If the specified term is greater than or equal + ** to the term from the interior node, then all terms on the sub-tree + ** headed by node iChild are smaller than zTerm. No need to search + ** iChild. + ** + ** If the interior node term is larger than the specified term, then + ** the tree headed by iChild may contain the specified term. + */ + cmp = memcmp(zTerm, zBuffer, (nBuffer>nTerm ? nTerm : nBuffer)); + if( piFirst && (cmp<0 || (cmp==0 && nBuffer>nTerm)) ){ + *piFirst = iChild; + piFirst = 0; + } + + if( piLast && cmp<0 ){ + *piLast = iChild; + piLast = 0; + } + + iChild++; + }; + + if( piFirst ) *piFirst = iChild; + if( piLast ) *piLast = iChild; + + finish_scan: + sqlite3_free(zBuffer); + return rc; +} + + +/* +** The buffer pointed to by argument zNode (size nNode bytes) contains an +** interior node of a b-tree segment. The zTerm buffer (size nTerm bytes) +** contains a term. This function searches the sub-tree headed by the zNode +** node for the range of leaf nodes that may contain the specified term +** or terms for which the specified term is a prefix. +** +** If piLeaf is not NULL, then *piLeaf is set to the blockid of the +** left-most leaf node in the tree that may contain the specified term. +** If piLeaf2 is not NULL, then *piLeaf2 is set to the blockid of the +** right-most leaf node that may contain a term for which the specified +** term is a prefix. +** +** It is possible that the range of returned leaf nodes does not contain +** the specified term or any terms for which it is a prefix. However, if the +** segment does contain any such terms, they are stored within the identified +** range. Because this function only inspects interior segment nodes (and +** never loads leaf nodes into memory), it is not possible to be sure. +** +** If an error occurs, an error code other than SQLITE_OK is returned. +*/ +static int fts3SelectLeaf( + Fts3Table *p, /* Virtual table handle */ + const char *zTerm, /* Term to select leaves for */ + int nTerm, /* Size of term zTerm in bytes */ + const char *zNode, /* Buffer containing segment interior node */ + int nNode, /* Size of buffer at zNode */ + sqlite3_int64 *piLeaf, /* Selected leaf node */ + sqlite3_int64 *piLeaf2 /* Selected leaf node */ +){ + int rc = SQLITE_OK; /* Return code */ + int iHeight; /* Height of this node in tree */ + + assert( piLeaf || piLeaf2 ); + + fts3GetVarint32(zNode, &iHeight); + rc = fts3ScanInteriorNode(zTerm, nTerm, zNode, nNode, piLeaf, piLeaf2); + assert( !piLeaf2 || !piLeaf || rc!=SQLITE_OK || (*piLeaf<=*piLeaf2) ); + + if( rc==SQLITE_OK && iHeight>1 ){ + char *zBlob = 0; /* Blob read from %_segments table */ + int nBlob = 0; /* Size of zBlob in bytes */ + + if( piLeaf && piLeaf2 && (*piLeaf!=*piLeaf2) ){ + rc = sqlite3Fts3ReadBlock(p, *piLeaf, &zBlob, &nBlob, 0); + if( rc==SQLITE_OK ){ + rc = fts3SelectLeaf(p, zTerm, nTerm, zBlob, nBlob, piLeaf, 0); + } + sqlite3_free(zBlob); + piLeaf = 0; + zBlob = 0; + } + + if( rc==SQLITE_OK ){ + rc = sqlite3Fts3ReadBlock(p, piLeaf?*piLeaf:*piLeaf2, &zBlob, &nBlob, 0); + } + if( rc==SQLITE_OK ){ + rc = fts3SelectLeaf(p, zTerm, nTerm, zBlob, nBlob, piLeaf, piLeaf2); + } + sqlite3_free(zBlob); + } + + return rc; +} + +/* +** This function is used to create delta-encoded serialized lists of FTS3 +** varints. Each call to this function appends a single varint to a list. +*/ +static void fts3PutDeltaVarint( + char **pp, /* IN/OUT: Output pointer */ + sqlite3_int64 *piPrev, /* IN/OUT: Previous value written to list */ + sqlite3_int64 iVal /* Write this value to the list */ +){ + assert( iVal-*piPrev > 0 || (*piPrev==0 && iVal==0) ); + *pp += sqlite3Fts3PutVarint(*pp, iVal-*piPrev); + *piPrev = iVal; +} + +/* +** When this function is called, *ppPoslist is assumed to point to the +** start of a position-list. After it returns, *ppPoslist points to the +** first byte after the position-list. +** +** A position list is list of positions (delta encoded) and columns for +** a single document record of a doclist. So, in other words, this +** routine advances *ppPoslist so that it points to the next docid in +** the doclist, or to the first byte past the end of the doclist. +** +** If pp is not NULL, then the contents of the position list are copied +** to *pp. *pp is set to point to the first byte past the last byte copied +** before this function returns. +*/ +static void fts3PoslistCopy(char **pp, char **ppPoslist){ + char *pEnd = *ppPoslist; + char c = 0; + + /* The end of a position list is marked by a zero encoded as an FTS3 + ** varint. A single POS_END (0) byte. Except, if the 0 byte is preceded by + ** a byte with the 0x80 bit set, then it is not a varint 0, but the tail + ** of some other, multi-byte, value. + ** + ** The following while-loop moves pEnd to point to the first byte that is not + ** immediately preceded by a byte with the 0x80 bit set. Then increments + ** pEnd once more so that it points to the byte immediately following the + ** last byte in the position-list. + */ + while( *pEnd | c ){ + c = *pEnd++ & 0x80; + testcase( c!=0 && (*pEnd)==0 ); + } + pEnd++; /* Advance past the POS_END terminator byte */ + + if( pp ){ + int n = (int)(pEnd - *ppPoslist); + char *p = *pp; + memcpy(p, *ppPoslist, n); + p += n; + *pp = p; + } + *ppPoslist = pEnd; +} + +/* +** When this function is called, *ppPoslist is assumed to point to the +** start of a column-list. After it returns, *ppPoslist points to the +** to the terminator (POS_COLUMN or POS_END) byte of the column-list. +** +** A column-list is list of delta-encoded positions for a single column +** within a single document within a doclist. +** +** The column-list is terminated either by a POS_COLUMN varint (1) or +** a POS_END varint (0). This routine leaves *ppPoslist pointing to +** the POS_COLUMN or POS_END that terminates the column-list. +** +** If pp is not NULL, then the contents of the column-list are copied +** to *pp. *pp is set to point to the first byte past the last byte copied +** before this function returns. The POS_COLUMN or POS_END terminator +** is not copied into *pp. +*/ +static void fts3ColumnlistCopy(char **pp, char **ppPoslist){ + char *pEnd = *ppPoslist; + char c = 0; + + /* A column-list is terminated by either a 0x01 or 0x00 byte that is + ** not part of a multi-byte varint. + */ + while( 0xFE & (*pEnd | c) ){ + c = *pEnd++ & 0x80; + testcase( c!=0 && ((*pEnd)&0xfe)==0 ); + } + if( pp ){ + int n = (int)(pEnd - *ppPoslist); + char *p = *pp; + memcpy(p, *ppPoslist, n); + p += n; + *pp = p; + } + *ppPoslist = pEnd; +} + +/* +** Value used to signify the end of an position-list. This must be +** as large or larger than any value that might appear on the +** position-list, even a position list that has been corrupted. +*/ +#define POSITION_LIST_END LARGEST_INT64 + +/* +** This function is used to help parse position-lists. When this function is +** called, *pp may point to the start of the next varint in the position-list +** being parsed, or it may point to 1 byte past the end of the position-list +** (in which case **pp will be a terminator bytes POS_END (0) or +** (1)). +** +** If *pp points past the end of the current position-list, set *pi to +** POSITION_LIST_END and return. Otherwise, read the next varint from *pp, +** increment the current value of *pi by the value read, and set *pp to +** point to the next value before returning. +** +** Before calling this routine *pi must be initialized to the value of +** the previous position, or zero if we are reading the first position +** in the position-list. Because positions are delta-encoded, the value +** of the previous position is needed in order to compute the value of +** the next position. +*/ +static void fts3ReadNextPos( + char **pp, /* IN/OUT: Pointer into position-list buffer */ + sqlite3_int64 *pi /* IN/OUT: Value read from position-list */ +){ + if( (**pp)&0xFE ){ + fts3GetDeltaVarint(pp, pi); + *pi -= 2; + }else{ + *pi = POSITION_LIST_END; + } +} + +/* +** If parameter iCol is not 0, write an POS_COLUMN (1) byte followed by +** the value of iCol encoded as a varint to *pp. This will start a new +** column list. +** +** Set *pp to point to the byte just after the last byte written before +** returning (do not modify it if iCol==0). Return the total number of bytes +** written (0 if iCol==0). +*/ +static int fts3PutColNumber(char **pp, int iCol){ + int n = 0; /* Number of bytes written */ + if( iCol ){ + char *p = *pp; /* Output pointer */ + n = 1 + sqlite3Fts3PutVarint(&p[1], iCol); + *p = 0x01; + *pp = &p[n]; + } + return n; +} + +/* +** Compute the union of two position lists. The output written +** into *pp contains all positions of both *pp1 and *pp2 in sorted +** order and with any duplicates removed. All pointers are +** updated appropriately. The caller is responsible for insuring +** that there is enough space in *pp to hold the complete output. +*/ +static int fts3PoslistMerge( + char **pp, /* Output buffer */ + char **pp1, /* Left input list */ + char **pp2 /* Right input list */ +){ + char *p = *pp; + char *p1 = *pp1; + char *p2 = *pp2; + + while( *p1 || *p2 ){ + int iCol1; /* The current column index in pp1 */ + int iCol2; /* The current column index in pp2 */ + + if( *p1==POS_COLUMN ){ + fts3GetVarint32(&p1[1], &iCol1); + if( iCol1==0 ) return FTS_CORRUPT_VTAB; + } + else if( *p1==POS_END ) iCol1 = 0x7fffffff; + else iCol1 = 0; + + if( *p2==POS_COLUMN ){ + fts3GetVarint32(&p2[1], &iCol2); + if( iCol2==0 ) return FTS_CORRUPT_VTAB; + } + else if( *p2==POS_END ) iCol2 = 0x7fffffff; + else iCol2 = 0; + + if( iCol1==iCol2 ){ + sqlite3_int64 i1 = 0; /* Last position from pp1 */ + sqlite3_int64 i2 = 0; /* Last position from pp2 */ + sqlite3_int64 iPrev = 0; + int n = fts3PutColNumber(&p, iCol1); + p1 += n; + p2 += n; + + /* At this point, both p1 and p2 point to the start of column-lists + ** for the same column (the column with index iCol1 and iCol2). + ** A column-list is a list of non-negative delta-encoded varints, each + ** incremented by 2 before being stored. Each list is terminated by a + ** POS_END (0) or POS_COLUMN (1). The following block merges the two lists + ** and writes the results to buffer p. p is left pointing to the byte + ** after the list written. No terminator (POS_END or POS_COLUMN) is + ** written to the output. + */ + fts3GetDeltaVarint(&p1, &i1); + fts3GetDeltaVarint(&p2, &i2); + do { + fts3PutDeltaVarint(&p, &iPrev, (i1pos(*pp1) && pos(*pp2)-pos(*pp1)<=nToken). i.e. +** when the *pp1 token appears before the *pp2 token, but not more than nToken +** slots before it. +** +** e.g. nToken==1 searches for adjacent positions. +*/ +static int fts3PoslistPhraseMerge( + char **pp, /* IN/OUT: Preallocated output buffer */ + int nToken, /* Maximum difference in token positions */ + int isSaveLeft, /* Save the left position */ + int isExact, /* If *pp1 is exactly nTokens before *pp2 */ + char **pp1, /* IN/OUT: Left input list */ + char **pp2 /* IN/OUT: Right input list */ +){ + char *p = *pp; + char *p1 = *pp1; + char *p2 = *pp2; + int iCol1 = 0; + int iCol2 = 0; + + /* Never set both isSaveLeft and isExact for the same invocation. */ + assert( isSaveLeft==0 || isExact==0 ); + + assert( p!=0 && *p1!=0 && *p2!=0 ); + if( *p1==POS_COLUMN ){ + p1++; + p1 += fts3GetVarint32(p1, &iCol1); + } + if( *p2==POS_COLUMN ){ + p2++; + p2 += fts3GetVarint32(p2, &iCol2); + } + + while( 1 ){ + if( iCol1==iCol2 ){ + char *pSave = p; + sqlite3_int64 iPrev = 0; + sqlite3_int64 iPos1 = 0; + sqlite3_int64 iPos2 = 0; + + if( iCol1 ){ + *p++ = POS_COLUMN; + p += sqlite3Fts3PutVarint(p, iCol1); + } + + fts3GetDeltaVarint(&p1, &iPos1); iPos1 -= 2; + fts3GetDeltaVarint(&p2, &iPos2); iPos2 -= 2; + if( iPos1<0 || iPos2<0 ) break; + + while( 1 ){ + if( iPos2==iPos1+nToken + || (isExact==0 && iPos2>iPos1 && iPos2<=iPos1+nToken) + ){ + sqlite3_int64 iSave; + iSave = isSaveLeft ? iPos1 : iPos2; + fts3PutDeltaVarint(&p, &iPrev, iSave+2); iPrev -= 2; + pSave = 0; + assert( p ); + } + if( (!isSaveLeft && iPos2<=(iPos1+nToken)) || iPos2<=iPos1 ){ + if( (*p2&0xFE)==0 ) break; + fts3GetDeltaVarint(&p2, &iPos2); iPos2 -= 2; + }else{ + if( (*p1&0xFE)==0 ) break; + fts3GetDeltaVarint(&p1, &iPos1); iPos1 -= 2; + } + } + + if( pSave ){ + assert( pp && p ); + p = pSave; + } + + fts3ColumnlistCopy(0, &p1); + fts3ColumnlistCopy(0, &p2); + assert( (*p1&0xFE)==0 && (*p2&0xFE)==0 ); + if( 0==*p1 || 0==*p2 ) break; + + p1++; + p1 += fts3GetVarint32(p1, &iCol1); + p2++; + p2 += fts3GetVarint32(p2, &iCol2); + } + + /* Advance pointer p1 or p2 (whichever corresponds to the smaller of + ** iCol1 and iCol2) so that it points to either the 0x00 that marks the + ** end of the position list, or the 0x01 that precedes the next + ** column-number in the position list. + */ + else if( iCol1=pEnd ){ + *pp = 0; + }else{ + sqlite3_int64 iVal; + *pp += sqlite3Fts3GetVarint(*pp, &iVal); + if( bDescIdx ){ + *pVal -= iVal; + }else{ + *pVal += iVal; + } + } +} + +/* +** This function is used to write a single varint to a buffer. The varint +** is written to *pp. Before returning, *pp is set to point 1 byte past the +** end of the value written. +** +** If *pbFirst is zero when this function is called, the value written to +** the buffer is that of parameter iVal. +** +** If *pbFirst is non-zero when this function is called, then the value +** written is either (iVal-*piPrev) (if bDescIdx is zero) or (*piPrev-iVal) +** (if bDescIdx is non-zero). +** +** Before returning, this function always sets *pbFirst to 1 and *piPrev +** to the value of parameter iVal. +*/ +static void fts3PutDeltaVarint3( + char **pp, /* IN/OUT: Output pointer */ + int bDescIdx, /* True for descending docids */ + sqlite3_int64 *piPrev, /* IN/OUT: Previous value written to list */ + int *pbFirst, /* IN/OUT: True after first int written */ + sqlite3_int64 iVal /* Write this value to the list */ +){ + sqlite3_int64 iWrite; + if( bDescIdx==0 || *pbFirst==0 ){ + iWrite = iVal - *piPrev; + }else{ + iWrite = *piPrev - iVal; + } + assert( *pbFirst || *piPrev==0 ); + assert_fts3_nc( *pbFirst==0 || iWrite>0 ); + assert( *pbFirst==0 || iWrite>=0 ); + *pp += sqlite3Fts3PutVarint(*pp, iWrite); + *piPrev = iVal; + *pbFirst = 1; +} + + +/* +** This macro is used by various functions that merge doclists. The two +** arguments are 64-bit docid values. If the value of the stack variable +** bDescDoclist is 0 when this macro is invoked, then it returns (i1-i2). +** Otherwise, (i2-i1). +** +** Using this makes it easier to write code that can merge doclists that are +** sorted in either ascending or descending order. +*/ +#define DOCID_CMP(i1, i2) ((bDescDoclist?-1:1) * (i1-i2)) + +/* +** This function does an "OR" merge of two doclists (output contains all +** positions contained in either argument doclist). If the docids in the +** input doclists are sorted in ascending order, parameter bDescDoclist +** should be false. If they are sorted in ascending order, it should be +** passed a non-zero value. +** +** If no error occurs, *paOut is set to point at an sqlite3_malloc'd buffer +** containing the output doclist and SQLITE_OK is returned. In this case +** *pnOut is set to the number of bytes in the output doclist. +** +** If an error occurs, an SQLite error code is returned. The output values +** are undefined in this case. +*/ +static int fts3DoclistOrMerge( + int bDescDoclist, /* True if arguments are desc */ + char *a1, int n1, /* First doclist */ + char *a2, int n2, /* Second doclist */ + char **paOut, int *pnOut /* OUT: Malloc'd doclist */ +){ + int rc = SQLITE_OK; + sqlite3_int64 i1 = 0; + sqlite3_int64 i2 = 0; + sqlite3_int64 iPrev = 0; + char *pEnd1 = &a1[n1]; + char *pEnd2 = &a2[n2]; + char *p1 = a1; + char *p2 = a2; + char *p; + char *aOut; + int bFirstOut = 0; + + *paOut = 0; + *pnOut = 0; + + /* Allocate space for the output. Both the input and output doclists + ** are delta encoded. If they are in ascending order (bDescDoclist==0), + ** then the first docid in each list is simply encoded as a varint. For + ** each subsequent docid, the varint stored is the difference between the + ** current and previous docid (a positive number - since the list is in + ** ascending order). + ** + ** The first docid written to the output is therefore encoded using the + ** same number of bytes as it is in whichever of the input lists it is + ** read from. And each subsequent docid read from the same input list + ** consumes either the same or less bytes as it did in the input (since + ** the difference between it and the previous value in the output must + ** be a positive value less than or equal to the delta value read from + ** the input list). The same argument applies to all but the first docid + ** read from the 'other' list. And to the contents of all position lists + ** that will be copied and merged from the input to the output. + ** + ** However, if the first docid copied to the output is a negative number, + ** then the encoding of the first docid from the 'other' input list may + ** be larger in the output than it was in the input (since the delta value + ** may be a larger positive integer than the actual docid). + ** + ** The space required to store the output is therefore the sum of the + ** sizes of the two inputs, plus enough space for exactly one of the input + ** docids to grow. + ** + ** A symetric argument may be made if the doclists are in descending + ** order. + */ + aOut = sqlite3_malloc64((i64)n1+n2+FTS3_VARINT_MAX-1+FTS3_BUFFER_PADDING); + if( !aOut ) return SQLITE_NOMEM; + + p = aOut; + fts3GetDeltaVarint3(&p1, pEnd1, 0, &i1); + fts3GetDeltaVarint3(&p2, pEnd2, 0, &i2); + while( p1 || p2 ){ + sqlite3_int64 iDiff = DOCID_CMP(i1, i2); + + if( p2 && p1 && iDiff==0 ){ + fts3PutDeltaVarint3(&p, bDescDoclist, &iPrev, &bFirstOut, i1); + rc = fts3PoslistMerge(&p, &p1, &p2); + if( rc ) break; + fts3GetDeltaVarint3(&p1, pEnd1, bDescDoclist, &i1); + fts3GetDeltaVarint3(&p2, pEnd2, bDescDoclist, &i2); + }else if( !p2 || (p1 && iDiff<0) ){ + fts3PutDeltaVarint3(&p, bDescDoclist, &iPrev, &bFirstOut, i1); + fts3PoslistCopy(&p, &p1); + fts3GetDeltaVarint3(&p1, pEnd1, bDescDoclist, &i1); + }else{ + fts3PutDeltaVarint3(&p, bDescDoclist, &iPrev, &bFirstOut, i2); + fts3PoslistCopy(&p, &p2); + fts3GetDeltaVarint3(&p2, pEnd2, bDescDoclist, &i2); + } + + assert( (p-aOut)<=((p1?(p1-a1):n1)+(p2?(p2-a2):n2)+FTS3_VARINT_MAX-1) ); + } + + if( rc!=SQLITE_OK ){ + sqlite3_free(aOut); + p = aOut = 0; + }else{ + assert( (p-aOut)<=n1+n2+FTS3_VARINT_MAX-1 ); + memset(&aOut[(p-aOut)], 0, FTS3_BUFFER_PADDING); + } + *paOut = aOut; + *pnOut = (int)(p-aOut); + return rc; +} + +/* +** This function does a "phrase" merge of two doclists. In a phrase merge, +** the output contains a copy of each position from the right-hand input +** doclist for which there is a position in the left-hand input doclist +** exactly nDist tokens before it. +** +** If the docids in the input doclists are sorted in ascending order, +** parameter bDescDoclist should be false. If they are sorted in ascending +** order, it should be passed a non-zero value. +** +** The right-hand input doclist is overwritten by this function. +*/ +static int fts3DoclistPhraseMerge( + int bDescDoclist, /* True if arguments are desc */ + int nDist, /* Distance from left to right (1=adjacent) */ + char *aLeft, int nLeft, /* Left doclist */ + char **paRight, int *pnRight /* IN/OUT: Right/output doclist */ +){ + sqlite3_int64 i1 = 0; + sqlite3_int64 i2 = 0; + sqlite3_int64 iPrev = 0; + char *aRight = *paRight; + char *pEnd1 = &aLeft[nLeft]; + char *pEnd2 = &aRight[*pnRight]; + char *p1 = aLeft; + char *p2 = aRight; + char *p; + int bFirstOut = 0; + char *aOut; + + assert( nDist>0 ); + if( bDescDoclist ){ + aOut = sqlite3_malloc64((sqlite3_int64)*pnRight + FTS3_VARINT_MAX); + if( aOut==0 ) return SQLITE_NOMEM; + }else{ + aOut = aRight; + } + p = aOut; + + fts3GetDeltaVarint3(&p1, pEnd1, 0, &i1); + fts3GetDeltaVarint3(&p2, pEnd2, 0, &i2); + + while( p1 && p2 ){ + sqlite3_int64 iDiff = DOCID_CMP(i1, i2); + if( iDiff==0 ){ + char *pSave = p; + sqlite3_int64 iPrevSave = iPrev; + int bFirstOutSave = bFirstOut; + + fts3PutDeltaVarint3(&p, bDescDoclist, &iPrev, &bFirstOut, i1); + if( 0==fts3PoslistPhraseMerge(&p, nDist, 0, 1, &p1, &p2) ){ + p = pSave; + iPrev = iPrevSave; + bFirstOut = bFirstOutSave; + } + fts3GetDeltaVarint3(&p1, pEnd1, bDescDoclist, &i1); + fts3GetDeltaVarint3(&p2, pEnd2, bDescDoclist, &i2); + }else if( iDiff<0 ){ + fts3PoslistCopy(0, &p1); + fts3GetDeltaVarint3(&p1, pEnd1, bDescDoclist, &i1); + }else{ + fts3PoslistCopy(0, &p2); + fts3GetDeltaVarint3(&p2, pEnd2, bDescDoclist, &i2); + } + } + + *pnRight = (int)(p - aOut); + if( bDescDoclist ){ + sqlite3_free(aRight); + *paRight = aOut; + } + + return SQLITE_OK; +} + +/* +** Argument pList points to a position list nList bytes in size. This +** function checks to see if the position list contains any entries for +** a token in position 0 (of any column). If so, it writes argument iDelta +** to the output buffer pOut, followed by a position list consisting only +** of the entries from pList at position 0, and terminated by an 0x00 byte. +** The value returned is the number of bytes written to pOut (if any). +*/ +SQLITE_PRIVATE int sqlite3Fts3FirstFilter( + sqlite3_int64 iDelta, /* Varint that may be written to pOut */ + char *pList, /* Position list (no 0x00 term) */ + int nList, /* Size of pList in bytes */ + char *pOut /* Write output here */ +){ + int nOut = 0; + int bWritten = 0; /* True once iDelta has been written */ + char *p = pList; + char *pEnd = &pList[nList]; + + if( *p!=0x01 ){ + if( *p==0x02 ){ + nOut += sqlite3Fts3PutVarint(&pOut[nOut], iDelta); + pOut[nOut++] = 0x02; + bWritten = 1; + } + fts3ColumnlistCopy(0, &p); + } + + while( paaOutput); i++){ + if( pTS->aaOutput[i] ){ + if( !aOut ){ + aOut = pTS->aaOutput[i]; + nOut = pTS->anOutput[i]; + pTS->aaOutput[i] = 0; + }else{ + int nNew; + char *aNew; + + int rc = fts3DoclistOrMerge(p->bDescIdx, + pTS->aaOutput[i], pTS->anOutput[i], aOut, nOut, &aNew, &nNew + ); + if( rc!=SQLITE_OK ){ + sqlite3_free(aOut); + return rc; + } + + sqlite3_free(pTS->aaOutput[i]); + sqlite3_free(aOut); + pTS->aaOutput[i] = 0; + aOut = aNew; + nOut = nNew; + } + } + } + + pTS->aaOutput[0] = aOut; + pTS->anOutput[0] = nOut; + return SQLITE_OK; +} + +/* +** Merge the doclist aDoclist/nDoclist into the TermSelect object passed +** as the first argument. The merge is an "OR" merge (see function +** fts3DoclistOrMerge() for details). +** +** This function is called with the doclist for each term that matches +** a queried prefix. It merges all these doclists into one, the doclist +** for the specified prefix. Since there can be a very large number of +** doclists to merge, the merging is done pair-wise using the TermSelect +** object. +** +** This function returns SQLITE_OK if the merge is successful, or an +** SQLite error code (SQLITE_NOMEM) if an error occurs. +*/ +static int fts3TermSelectMerge( + Fts3Table *p, /* FTS table handle */ + TermSelect *pTS, /* TermSelect object to merge into */ + char *aDoclist, /* Pointer to doclist */ + int nDoclist /* Size of aDoclist in bytes */ +){ + if( pTS->aaOutput[0]==0 ){ + /* If this is the first term selected, copy the doclist to the output + ** buffer using memcpy(). + ** + ** Add FTS3_VARINT_MAX bytes of unused space to the end of the + ** allocation. This is so as to ensure that the buffer is big enough + ** to hold the current doclist AND'd with any other doclist. If the + ** doclists are stored in order=ASC order, this padding would not be + ** required (since the size of [doclistA AND doclistB] is always less + ** than or equal to the size of [doclistA] in that case). But this is + ** not true for order=DESC. For example, a doclist containing (1, -1) + ** may be smaller than (-1), as in the first example the -1 may be stored + ** as a single-byte delta, whereas in the second it must be stored as a + ** FTS3_VARINT_MAX byte varint. + ** + ** Similar padding is added in the fts3DoclistOrMerge() function. + */ + pTS->aaOutput[0] = sqlite3_malloc(nDoclist + FTS3_VARINT_MAX + 1); + pTS->anOutput[0] = nDoclist; + if( pTS->aaOutput[0] ){ + memcpy(pTS->aaOutput[0], aDoclist, nDoclist); + memset(&pTS->aaOutput[0][nDoclist], 0, FTS3_VARINT_MAX); + }else{ + return SQLITE_NOMEM; + } + }else{ + char *aMerge = aDoclist; + int nMerge = nDoclist; + int iOut; + + for(iOut=0; iOutaaOutput); iOut++){ + if( pTS->aaOutput[iOut]==0 ){ + assert( iOut>0 ); + pTS->aaOutput[iOut] = aMerge; + pTS->anOutput[iOut] = nMerge; + break; + }else{ + char *aNew; + int nNew; + + int rc = fts3DoclistOrMerge(p->bDescIdx, aMerge, nMerge, + pTS->aaOutput[iOut], pTS->anOutput[iOut], &aNew, &nNew + ); + if( rc!=SQLITE_OK ){ + if( aMerge!=aDoclist ) sqlite3_free(aMerge); + return rc; + } + + if( aMerge!=aDoclist ) sqlite3_free(aMerge); + sqlite3_free(pTS->aaOutput[iOut]); + pTS->aaOutput[iOut] = 0; + + aMerge = aNew; + nMerge = nNew; + if( (iOut+1)==SizeofArray(pTS->aaOutput) ){ + pTS->aaOutput[iOut] = aMerge; + pTS->anOutput[iOut] = nMerge; + } + } + } + } + return SQLITE_OK; +} + +/* +** Append SegReader object pNew to the end of the pCsr->apSegment[] array. +*/ +static int fts3SegReaderCursorAppend( + Fts3MultiSegReader *pCsr, + Fts3SegReader *pNew +){ + if( (pCsr->nSegment%16)==0 ){ + Fts3SegReader **apNew; + sqlite3_int64 nByte = (pCsr->nSegment + 16)*sizeof(Fts3SegReader*); + apNew = (Fts3SegReader **)sqlite3_realloc64(pCsr->apSegment, nByte); + if( !apNew ){ + sqlite3Fts3SegReaderFree(pNew); + return SQLITE_NOMEM; + } + pCsr->apSegment = apNew; + } + pCsr->apSegment[pCsr->nSegment++] = pNew; + return SQLITE_OK; +} + +/* +** Add seg-reader objects to the Fts3MultiSegReader object passed as the +** 8th argument. +** +** This function returns SQLITE_OK if successful, or an SQLite error code +** otherwise. +*/ +static int fts3SegReaderCursor( + Fts3Table *p, /* FTS3 table handle */ + int iLangid, /* Language id */ + int iIndex, /* Index to search (from 0 to p->nIndex-1) */ + int iLevel, /* Level of segments to scan */ + const char *zTerm, /* Term to query for */ + int nTerm, /* Size of zTerm in bytes */ + int isPrefix, /* True for a prefix search */ + int isScan, /* True to scan from zTerm to EOF */ + Fts3MultiSegReader *pCsr /* Cursor object to populate */ +){ + int rc = SQLITE_OK; /* Error code */ + sqlite3_stmt *pStmt = 0; /* Statement to iterate through segments */ + int rc2; /* Result of sqlite3_reset() */ + + /* If iLevel is less than 0 and this is not a scan, include a seg-reader + ** for the pending-terms. If this is a scan, then this call must be being + ** made by an fts4aux module, not an FTS table. In this case calling + ** Fts3SegReaderPending might segfault, as the data structures used by + ** fts4aux are not completely populated. So it's easiest to filter these + ** calls out here. */ + if( iLevel<0 && p->aIndex ){ + Fts3SegReader *pSeg = 0; + rc = sqlite3Fts3SegReaderPending(p, iIndex, zTerm, nTerm, isPrefix||isScan, &pSeg); + if( rc==SQLITE_OK && pSeg ){ + rc = fts3SegReaderCursorAppend(pCsr, pSeg); + } + } + + if( iLevel!=FTS3_SEGCURSOR_PENDING ){ + if( rc==SQLITE_OK ){ + rc = sqlite3Fts3AllSegdirs(p, iLangid, iIndex, iLevel, &pStmt); + } + + while( rc==SQLITE_OK && SQLITE_ROW==(rc = sqlite3_step(pStmt)) ){ + Fts3SegReader *pSeg = 0; + + /* Read the values returned by the SELECT into local variables. */ + sqlite3_int64 iStartBlock = sqlite3_column_int64(pStmt, 1); + sqlite3_int64 iLeavesEndBlock = sqlite3_column_int64(pStmt, 2); + sqlite3_int64 iEndBlock = sqlite3_column_int64(pStmt, 3); + int nRoot = sqlite3_column_bytes(pStmt, 4); + char const *zRoot = sqlite3_column_blob(pStmt, 4); + + /* If zTerm is not NULL, and this segment is not stored entirely on its + ** root node, the range of leaves scanned can be reduced. Do this. */ + if( iStartBlock && zTerm && zRoot ){ + sqlite3_int64 *pi = (isPrefix ? &iLeavesEndBlock : 0); + rc = fts3SelectLeaf(p, zTerm, nTerm, zRoot, nRoot, &iStartBlock, pi); + if( rc!=SQLITE_OK ) goto finished; + if( isPrefix==0 && isScan==0 ) iLeavesEndBlock = iStartBlock; + } + + rc = sqlite3Fts3SegReaderNew(pCsr->nSegment+1, + (isPrefix==0 && isScan==0), + iStartBlock, iLeavesEndBlock, + iEndBlock, zRoot, nRoot, &pSeg + ); + if( rc!=SQLITE_OK ) goto finished; + rc = fts3SegReaderCursorAppend(pCsr, pSeg); + } + } + + finished: + rc2 = sqlite3_reset(pStmt); + if( rc==SQLITE_DONE ) rc = rc2; + + return rc; +} + +/* +** Set up a cursor object for iterating through a full-text index or a +** single level therein. +*/ +SQLITE_PRIVATE int sqlite3Fts3SegReaderCursor( + Fts3Table *p, /* FTS3 table handle */ + int iLangid, /* Language-id to search */ + int iIndex, /* Index to search (from 0 to p->nIndex-1) */ + int iLevel, /* Level of segments to scan */ + const char *zTerm, /* Term to query for */ + int nTerm, /* Size of zTerm in bytes */ + int isPrefix, /* True for a prefix search */ + int isScan, /* True to scan from zTerm to EOF */ + Fts3MultiSegReader *pCsr /* Cursor object to populate */ +){ + assert( iIndex>=0 && iIndexnIndex ); + assert( iLevel==FTS3_SEGCURSOR_ALL + || iLevel==FTS3_SEGCURSOR_PENDING + || iLevel>=0 + ); + assert( iLevelbase.pVtab; + + if( isPrefix ){ + for(i=1; bFound==0 && inIndex; i++){ + if( p->aIndex[i].nPrefix==nTerm ){ + bFound = 1; + rc = sqlite3Fts3SegReaderCursor(p, pCsr->iLangid, + i, FTS3_SEGCURSOR_ALL, zTerm, nTerm, 0, 0, pSegcsr + ); + pSegcsr->bLookup = 1; + } + } + + for(i=1; bFound==0 && inIndex; i++){ + if( p->aIndex[i].nPrefix==nTerm+1 ){ + bFound = 1; + rc = sqlite3Fts3SegReaderCursor(p, pCsr->iLangid, + i, FTS3_SEGCURSOR_ALL, zTerm, nTerm, 1, 0, pSegcsr + ); + if( rc==SQLITE_OK ){ + rc = fts3SegReaderCursorAddZero( + p, pCsr->iLangid, zTerm, nTerm, pSegcsr + ); + } + } + } + } + + if( bFound==0 ){ + rc = sqlite3Fts3SegReaderCursor(p, pCsr->iLangid, + 0, FTS3_SEGCURSOR_ALL, zTerm, nTerm, isPrefix, 0, pSegcsr + ); + pSegcsr->bLookup = !isPrefix; + } + } + + *ppSegcsr = pSegcsr; + return rc; +} + +/* +** Free an Fts3MultiSegReader allocated by fts3TermSegReaderCursor(). +*/ +static void fts3SegReaderCursorFree(Fts3MultiSegReader *pSegcsr){ + sqlite3Fts3SegReaderFinish(pSegcsr); + sqlite3_free(pSegcsr); +} + +/* +** This function retrieves the doclist for the specified term (or term +** prefix) from the database. +*/ +static int fts3TermSelect( + Fts3Table *p, /* Virtual table handle */ + Fts3PhraseToken *pTok, /* Token to query for */ + int iColumn, /* Column to query (or -ve for all columns) */ + int *pnOut, /* OUT: Size of buffer at *ppOut */ + char **ppOut /* OUT: Malloced result buffer */ +){ + int rc; /* Return code */ + Fts3MultiSegReader *pSegcsr; /* Seg-reader cursor for this term */ + TermSelect tsc; /* Object for pair-wise doclist merging */ + Fts3SegFilter filter; /* Segment term filter configuration */ + + pSegcsr = pTok->pSegcsr; + memset(&tsc, 0, sizeof(TermSelect)); + + filter.flags = FTS3_SEGMENT_IGNORE_EMPTY | FTS3_SEGMENT_REQUIRE_POS + | (pTok->isPrefix ? FTS3_SEGMENT_PREFIX : 0) + | (pTok->bFirst ? FTS3_SEGMENT_FIRST : 0) + | (iColumnnColumn ? FTS3_SEGMENT_COLUMN_FILTER : 0); + filter.iCol = iColumn; + filter.zTerm = pTok->z; + filter.nTerm = pTok->n; + + rc = sqlite3Fts3SegReaderStart(p, pSegcsr, &filter); + while( SQLITE_OK==rc + && SQLITE_ROW==(rc = sqlite3Fts3SegReaderStep(p, pSegcsr)) + ){ + rc = fts3TermSelectMerge(p, &tsc, pSegcsr->aDoclist, pSegcsr->nDoclist); + } + + if( rc==SQLITE_OK ){ + rc = fts3TermSelectFinishMerge(p, &tsc); + } + if( rc==SQLITE_OK ){ + *ppOut = tsc.aaOutput[0]; + *pnOut = tsc.anOutput[0]; + }else{ + int i; + for(i=0; ipSegcsr = 0; + return rc; +} + +/* +** This function counts the total number of docids in the doclist stored +** in buffer aList[], size nList bytes. +** +** If the isPoslist argument is true, then it is assumed that the doclist +** contains a position-list following each docid. Otherwise, it is assumed +** that the doclist is simply a list of docids stored as delta encoded +** varints. +*/ +static int fts3DoclistCountDocids(char *aList, int nList){ + int nDoc = 0; /* Return value */ + if( aList ){ + char *aEnd = &aList[nList]; /* Pointer to one byte after EOF */ + char *p = aList; /* Cursor */ + while( peSearch==FTS3_DOCID_SEARCH || pCsr->eSearch==FTS3_FULLSCAN_SEARCH ){ + if( SQLITE_ROW!=sqlite3_step(pCsr->pStmt) ){ + pCsr->isEof = 1; + rc = sqlite3_reset(pCsr->pStmt); + }else{ + pCsr->iPrevId = sqlite3_column_int64(pCsr->pStmt, 0); + rc = SQLITE_OK; + } + }else{ + rc = fts3EvalNext((Fts3Cursor *)pCursor); + } + assert( ((Fts3Table *)pCsr->base.pVtab)->pSegments==0 ); + return rc; +} + +/* +** If the numeric type of argument pVal is "integer", then return it +** converted to a 64-bit signed integer. Otherwise, return a copy of +** the second parameter, iDefault. +*/ +static sqlite3_int64 fts3DocidRange(sqlite3_value *pVal, i64 iDefault){ + if( pVal ){ + int eType = sqlite3_value_numeric_type(pVal); + if( eType==SQLITE_INTEGER ){ + return sqlite3_value_int64(pVal); + } + } + return iDefault; +} + +/* +** This is the xFilter interface for the virtual table. See +** the virtual table xFilter method documentation for additional +** information. +** +** If idxNum==FTS3_FULLSCAN_SEARCH then do a full table scan against +** the %_content table. +** +** If idxNum==FTS3_DOCID_SEARCH then do a docid lookup for a single entry +** in the %_content table. +** +** If idxNum>=FTS3_FULLTEXT_SEARCH then use the full text index. The +** column on the left-hand side of the MATCH operator is column +** number idxNum-FTS3_FULLTEXT_SEARCH, 0 indexed. argv[0] is the right-hand +** side of the MATCH operator. +*/ +static int fts3FilterMethod( + sqlite3_vtab_cursor *pCursor, /* The cursor used for this query */ + int idxNum, /* Strategy index */ + const char *idxStr, /* Unused */ + int nVal, /* Number of elements in apVal */ + sqlite3_value **apVal /* Arguments for the indexing scheme */ +){ + int rc = SQLITE_OK; + char *zSql; /* SQL statement used to access %_content */ + int eSearch; + Fts3Table *p = (Fts3Table *)pCursor->pVtab; + Fts3Cursor *pCsr = (Fts3Cursor *)pCursor; + + sqlite3_value *pCons = 0; /* The MATCH or rowid constraint, if any */ + sqlite3_value *pLangid = 0; /* The "langid = ?" constraint, if any */ + sqlite3_value *pDocidGe = 0; /* The "docid >= ?" constraint, if any */ + sqlite3_value *pDocidLe = 0; /* The "docid <= ?" constraint, if any */ + int iIdx; + + UNUSED_PARAMETER(idxStr); + UNUSED_PARAMETER(nVal); + + eSearch = (idxNum & 0x0000FFFF); + assert( eSearch>=0 && eSearch<=(FTS3_FULLTEXT_SEARCH+p->nColumn) ); + assert( p->pSegments==0 ); + + /* Collect arguments into local variables */ + iIdx = 0; + if( eSearch!=FTS3_FULLSCAN_SEARCH ) pCons = apVal[iIdx++]; + if( idxNum & FTS3_HAVE_LANGID ) pLangid = apVal[iIdx++]; + if( idxNum & FTS3_HAVE_DOCID_GE ) pDocidGe = apVal[iIdx++]; + if( idxNum & FTS3_HAVE_DOCID_LE ) pDocidLe = apVal[iIdx++]; + assert( iIdx==nVal ); + + /* In case the cursor has been used before, clear it now. */ + fts3ClearCursor(pCsr); + + /* Set the lower and upper bounds on docids to return */ + pCsr->iMinDocid = fts3DocidRange(pDocidGe, SMALLEST_INT64); + pCsr->iMaxDocid = fts3DocidRange(pDocidLe, LARGEST_INT64); + + if( idxStr ){ + pCsr->bDesc = (idxStr[0]=='D'); + }else{ + pCsr->bDesc = p->bDescIdx; + } + pCsr->eSearch = (i16)eSearch; + + if( eSearch!=FTS3_DOCID_SEARCH && eSearch!=FTS3_FULLSCAN_SEARCH ){ + int iCol = eSearch-FTS3_FULLTEXT_SEARCH; + const char *zQuery = (const char *)sqlite3_value_text(pCons); + + if( zQuery==0 && sqlite3_value_type(pCons)!=SQLITE_NULL ){ + return SQLITE_NOMEM; + } + + pCsr->iLangid = 0; + if( pLangid ) pCsr->iLangid = sqlite3_value_int(pLangid); + + assert( p->base.zErrMsg==0 ); + rc = sqlite3Fts3ExprParse(p->pTokenizer, pCsr->iLangid, + p->azColumn, p->bFts4, p->nColumn, iCol, zQuery, -1, &pCsr->pExpr, + &p->base.zErrMsg + ); + if( rc!=SQLITE_OK ){ + return rc; + } + + rc = fts3EvalStart(pCsr); + sqlite3Fts3SegmentsClose(p); + if( rc!=SQLITE_OK ) return rc; + pCsr->pNextId = pCsr->aDoclist; + pCsr->iPrevId = 0; + } + + /* Compile a SELECT statement for this cursor. For a full-table-scan, the + ** statement loops through all rows of the %_content table. For a + ** full-text query or docid lookup, the statement retrieves a single + ** row by docid. + */ + if( eSearch==FTS3_FULLSCAN_SEARCH ){ + if( pDocidGe || pDocidLe ){ + zSql = sqlite3_mprintf( + "SELECT %s WHERE rowid BETWEEN %lld AND %lld ORDER BY rowid %s", + p->zReadExprlist, pCsr->iMinDocid, pCsr->iMaxDocid, + (pCsr->bDesc ? "DESC" : "ASC") + ); + }else{ + zSql = sqlite3_mprintf("SELECT %s ORDER BY rowid %s", + p->zReadExprlist, (pCsr->bDesc ? "DESC" : "ASC") + ); + } + if( zSql ){ + rc = sqlite3_prepare_v3(p->db,zSql,-1,SQLITE_PREPARE_PERSISTENT,&pCsr->pStmt,0); + sqlite3_free(zSql); + }else{ + rc = SQLITE_NOMEM; + } + }else if( eSearch==FTS3_DOCID_SEARCH ){ + rc = fts3CursorSeekStmt(pCsr); + if( rc==SQLITE_OK ){ + rc = sqlite3_bind_value(pCsr->pStmt, 1, pCons); + } + } + if( rc!=SQLITE_OK ) return rc; + + return fts3NextMethod(pCursor); +} + +/* +** This is the xEof method of the virtual table. SQLite calls this +** routine to find out if it has reached the end of a result set. +*/ +static int fts3EofMethod(sqlite3_vtab_cursor *pCursor){ + Fts3Cursor *pCsr = (Fts3Cursor*)pCursor; + if( pCsr->isEof ){ + fts3ClearCursor(pCsr); + pCsr->isEof = 1; + } + return pCsr->isEof; +} + +/* +** This is the xRowid method. The SQLite core calls this routine to +** retrieve the rowid for the current row of the result set. fts3 +** exposes %_content.docid as the rowid for the virtual table. The +** rowid should be written to *pRowid. +*/ +static int fts3RowidMethod(sqlite3_vtab_cursor *pCursor, sqlite_int64 *pRowid){ + Fts3Cursor *pCsr = (Fts3Cursor *) pCursor; + *pRowid = pCsr->iPrevId; + return SQLITE_OK; +} + +/* +** This is the xColumn method, called by SQLite to request a value from +** the row that the supplied cursor currently points to. +** +** If: +** +** (iCol < p->nColumn) -> The value of the iCol'th user column. +** (iCol == p->nColumn) -> Magic column with the same name as the table. +** (iCol == p->nColumn+1) -> Docid column +** (iCol == p->nColumn+2) -> Langid column +*/ +static int fts3ColumnMethod( + sqlite3_vtab_cursor *pCursor, /* Cursor to retrieve value from */ + sqlite3_context *pCtx, /* Context for sqlite3_result_xxx() calls */ + int iCol /* Index of column to read value from */ +){ + int rc = SQLITE_OK; /* Return Code */ + Fts3Cursor *pCsr = (Fts3Cursor *) pCursor; + Fts3Table *p = (Fts3Table *)pCursor->pVtab; + + /* The column value supplied by SQLite must be in range. */ + assert( iCol>=0 && iCol<=p->nColumn+2 ); + + switch( iCol-p->nColumn ){ + case 0: + /* The special 'table-name' column */ + sqlite3_result_pointer(pCtx, pCsr, "fts3cursor", 0); + break; + + case 1: + /* The docid column */ + sqlite3_result_int64(pCtx, pCsr->iPrevId); + break; + + case 2: + if( pCsr->pExpr ){ + sqlite3_result_int64(pCtx, pCsr->iLangid); + break; + }else if( p->zLanguageid==0 ){ + sqlite3_result_int(pCtx, 0); + break; + }else{ + iCol = p->nColumn; + /* fall-through */ + } + + default: + /* A user column. Or, if this is a full-table scan, possibly the + ** language-id column. Seek the cursor. */ + rc = fts3CursorSeek(0, pCsr); + if( rc==SQLITE_OK && sqlite3_data_count(pCsr->pStmt)-1>iCol ){ + sqlite3_result_value(pCtx, sqlite3_column_value(pCsr->pStmt, iCol+1)); + } + break; + } + + assert( ((Fts3Table *)pCsr->base.pVtab)->pSegments==0 ); + return rc; +} + +/* +** This function is the implementation of the xUpdate callback used by +** FTS3 virtual tables. It is invoked by SQLite each time a row is to be +** inserted, updated or deleted. +*/ +static int fts3UpdateMethod( + sqlite3_vtab *pVtab, /* Virtual table handle */ + int nArg, /* Size of argument array */ + sqlite3_value **apVal, /* Array of arguments */ + sqlite_int64 *pRowid /* OUT: The affected (or effected) rowid */ +){ + return sqlite3Fts3UpdateMethod(pVtab, nArg, apVal, pRowid); +} + +/* +** Implementation of xSync() method. Flush the contents of the pending-terms +** hash-table to the database. +*/ +static int fts3SyncMethod(sqlite3_vtab *pVtab){ + + /* Following an incremental-merge operation, assuming that the input + ** segments are not completely consumed (the usual case), they are updated + ** in place to remove the entries that have already been merged. This + ** involves updating the leaf block that contains the smallest unmerged + ** entry and each block (if any) between the leaf and the root node. So + ** if the height of the input segment b-trees is N, and input segments + ** are merged eight at a time, updating the input segments at the end + ** of an incremental-merge requires writing (8*(1+N)) blocks. N is usually + ** small - often between 0 and 2. So the overhead of the incremental + ** merge is somewhere between 8 and 24 blocks. To avoid this overhead + ** dwarfing the actual productive work accomplished, the incremental merge + ** is only attempted if it will write at least 64 leaf blocks. Hence + ** nMinMerge. + ** + ** Of course, updating the input segments also involves deleting a bunch + ** of blocks from the segments table. But this is not considered overhead + ** as it would also be required by a crisis-merge that used the same input + ** segments. + */ + const u32 nMinMerge = 64; /* Minimum amount of incr-merge work to do */ + + Fts3Table *p = (Fts3Table*)pVtab; + int rc; + i64 iLastRowid = sqlite3_last_insert_rowid(p->db); + + rc = sqlite3Fts3PendingTermsFlush(p); + if( rc==SQLITE_OK + && p->nLeafAdd>(nMinMerge/16) + && p->nAutoincrmerge && p->nAutoincrmerge!=0xff + ){ + int mxLevel = 0; /* Maximum relative level value in db */ + int A; /* Incr-merge parameter A */ + + rc = sqlite3Fts3MaxLevel(p, &mxLevel); + assert( rc==SQLITE_OK || mxLevel==0 ); + A = p->nLeafAdd * mxLevel; + A += (A/2); + if( A>(int)nMinMerge ) rc = sqlite3Fts3Incrmerge(p, A, p->nAutoincrmerge); + } + sqlite3Fts3SegmentsClose(p); + sqlite3_set_last_insert_rowid(p->db, iLastRowid); + return rc; +} + +/* +** If it is currently unknown whether or not the FTS table has an %_stat +** table (if p->bHasStat==2), attempt to determine this (set p->bHasStat +** to 0 or 1). Return SQLITE_OK if successful, or an SQLite error code +** if an error occurs. +*/ +static int fts3SetHasStat(Fts3Table *p){ + int rc = SQLITE_OK; + if( p->bHasStat==2 ){ + char *zTbl = sqlite3_mprintf("%s_stat", p->zName); + if( zTbl ){ + int res = sqlite3_table_column_metadata(p->db, p->zDb, zTbl, 0,0,0,0,0,0); + sqlite3_free(zTbl); + p->bHasStat = (res==SQLITE_OK); + }else{ + rc = SQLITE_NOMEM; + } + } + return rc; +} + +/* +** Implementation of xBegin() method. +*/ +static int fts3BeginMethod(sqlite3_vtab *pVtab){ + Fts3Table *p = (Fts3Table*)pVtab; + UNUSED_PARAMETER(pVtab); + assert( p->pSegments==0 ); + assert( p->nPendingData==0 ); + assert( p->inTransaction!=1 ); + TESTONLY( p->inTransaction = 1 ); + TESTONLY( p->mxSavepoint = -1; ); + p->nLeafAdd = 0; + return fts3SetHasStat(p); +} + +/* +** Implementation of xCommit() method. This is a no-op. The contents of +** the pending-terms hash-table have already been flushed into the database +** by fts3SyncMethod(). +*/ +static int fts3CommitMethod(sqlite3_vtab *pVtab){ + TESTONLY( Fts3Table *p = (Fts3Table*)pVtab ); + UNUSED_PARAMETER(pVtab); + assert( p->nPendingData==0 ); + assert( p->inTransaction!=0 ); + assert( p->pSegments==0 ); + TESTONLY( p->inTransaction = 0 ); + TESTONLY( p->mxSavepoint = -1; ); + return SQLITE_OK; +} + +/* +** Implementation of xRollback(). Discard the contents of the pending-terms +** hash-table. Any changes made to the database are reverted by SQLite. +*/ +static int fts3RollbackMethod(sqlite3_vtab *pVtab){ + Fts3Table *p = (Fts3Table*)pVtab; + sqlite3Fts3PendingTermsClear(p); + assert( p->inTransaction!=0 ); + TESTONLY( p->inTransaction = 0 ); + TESTONLY( p->mxSavepoint = -1; ); + return SQLITE_OK; +} + +/* +** When called, *ppPoslist must point to the byte immediately following the +** end of a position-list. i.e. ( (*ppPoslist)[-1]==POS_END ). This function +** moves *ppPoslist so that it instead points to the first byte of the +** same position list. +*/ +static void fts3ReversePoslist(char *pStart, char **ppPoslist){ + char *p = &(*ppPoslist)[-2]; + char c = 0; + + /* Skip backwards passed any trailing 0x00 bytes added by NearTrim() */ + while( p>pStart && (c=*p--)==0 ); + + /* Search backwards for a varint with value zero (the end of the previous + ** poslist). This is an 0x00 byte preceded by some byte that does not + ** have the 0x80 bit set. */ + while( p>pStart && (*p & 0x80) | c ){ + c = *p--; + } + assert( p==pStart || c==0 ); + + /* At this point p points to that preceding byte without the 0x80 bit + ** set. So to find the start of the poslist, skip forward 2 bytes then + ** over a varint. + ** + ** Normally. The other case is that p==pStart and the poslist to return + ** is the first in the doclist. In this case do not skip forward 2 bytes. + ** The second part of the if condition (c==0 && *ppPoslist>&p[2]) + ** is required for cases where the first byte of a doclist and the + ** doclist is empty. For example, if the first docid is 10, a doclist + ** that begins with: + ** + ** 0x0A 0x00 + */ + if( p>pStart || (c==0 && *ppPoslist>&p[2]) ){ p = &p[2]; } + while( *p++&0x80 ); + *ppPoslist = p; +} + +/* +** Helper function used by the implementation of the overloaded snippet(), +** offsets() and optimize() SQL functions. +** +** If the value passed as the third argument is a blob of size +** sizeof(Fts3Cursor*), then the blob contents are copied to the +** output variable *ppCsr and SQLITE_OK is returned. Otherwise, an error +** message is written to context pContext and SQLITE_ERROR returned. The +** string passed via zFunc is used as part of the error message. +*/ +static int fts3FunctionArg( + sqlite3_context *pContext, /* SQL function call context */ + const char *zFunc, /* Function name */ + sqlite3_value *pVal, /* argv[0] passed to function */ + Fts3Cursor **ppCsr /* OUT: Store cursor handle here */ +){ + int rc; + *ppCsr = (Fts3Cursor*)sqlite3_value_pointer(pVal, "fts3cursor"); + if( (*ppCsr)!=0 ){ + rc = SQLITE_OK; + }else{ + char *zErr = sqlite3_mprintf("illegal first argument to %s", zFunc); + sqlite3_result_error(pContext, zErr, -1); + sqlite3_free(zErr); + rc = SQLITE_ERROR; + } + return rc; +} + +/* +** Implementation of the snippet() function for FTS3 +*/ +static void fts3SnippetFunc( + sqlite3_context *pContext, /* SQLite function call context */ + int nVal, /* Size of apVal[] array */ + sqlite3_value **apVal /* Array of arguments */ +){ + Fts3Cursor *pCsr; /* Cursor handle passed through apVal[0] */ + const char *zStart = ""; + const char *zEnd = ""; + const char *zEllipsis = "..."; + int iCol = -1; + int nToken = 15; /* Default number of tokens in snippet */ + + /* There must be at least one argument passed to this function (otherwise + ** the non-overloaded version would have been called instead of this one). + */ + assert( nVal>=1 ); + + if( nVal>6 ){ + sqlite3_result_error(pContext, + "wrong number of arguments to function snippet()", -1); + return; + } + if( fts3FunctionArg(pContext, "snippet", apVal[0], &pCsr) ) return; + + switch( nVal ){ + case 6: nToken = sqlite3_value_int(apVal[5]); + case 5: iCol = sqlite3_value_int(apVal[4]); + case 4: zEllipsis = (const char*)sqlite3_value_text(apVal[3]); + case 3: zEnd = (const char*)sqlite3_value_text(apVal[2]); + case 2: zStart = (const char*)sqlite3_value_text(apVal[1]); + } + if( !zEllipsis || !zEnd || !zStart ){ + sqlite3_result_error_nomem(pContext); + }else if( nToken==0 ){ + sqlite3_result_text(pContext, "", -1, SQLITE_STATIC); + }else if( SQLITE_OK==fts3CursorSeek(pContext, pCsr) ){ + sqlite3Fts3Snippet(pContext, pCsr, zStart, zEnd, zEllipsis, iCol, nToken); + } +} + +/* +** Implementation of the offsets() function for FTS3 +*/ +static void fts3OffsetsFunc( + sqlite3_context *pContext, /* SQLite function call context */ + int nVal, /* Size of argument array */ + sqlite3_value **apVal /* Array of arguments */ +){ + Fts3Cursor *pCsr; /* Cursor handle passed through apVal[0] */ + + UNUSED_PARAMETER(nVal); + + assert( nVal==1 ); + if( fts3FunctionArg(pContext, "offsets", apVal[0], &pCsr) ) return; + assert( pCsr ); + if( SQLITE_OK==fts3CursorSeek(pContext, pCsr) ){ + sqlite3Fts3Offsets(pContext, pCsr); + } +} + +/* +** Implementation of the special optimize() function for FTS3. This +** function merges all segments in the database to a single segment. +** Example usage is: +** +** SELECT optimize(t) FROM t LIMIT 1; +** +** where 't' is the name of an FTS3 table. +*/ +static void fts3OptimizeFunc( + sqlite3_context *pContext, /* SQLite function call context */ + int nVal, /* Size of argument array */ + sqlite3_value **apVal /* Array of arguments */ +){ + int rc; /* Return code */ + Fts3Table *p; /* Virtual table handle */ + Fts3Cursor *pCursor; /* Cursor handle passed through apVal[0] */ + + UNUSED_PARAMETER(nVal); + + assert( nVal==1 ); + if( fts3FunctionArg(pContext, "optimize", apVal[0], &pCursor) ) return; + p = (Fts3Table *)pCursor->base.pVtab; + assert( p ); + + rc = sqlite3Fts3Optimize(p); + + switch( rc ){ + case SQLITE_OK: + sqlite3_result_text(pContext, "Index optimized", -1, SQLITE_STATIC); + break; + case SQLITE_DONE: + sqlite3_result_text(pContext, "Index already optimal", -1, SQLITE_STATIC); + break; + default: + sqlite3_result_error_code(pContext, rc); + break; + } +} + +/* +** Implementation of the matchinfo() function for FTS3 +*/ +static void fts3MatchinfoFunc( + sqlite3_context *pContext, /* SQLite function call context */ + int nVal, /* Size of argument array */ + sqlite3_value **apVal /* Array of arguments */ +){ + Fts3Cursor *pCsr; /* Cursor handle passed through apVal[0] */ + assert( nVal==1 || nVal==2 ); + if( SQLITE_OK==fts3FunctionArg(pContext, "matchinfo", apVal[0], &pCsr) ){ + const char *zArg = 0; + if( nVal>1 ){ + zArg = (const char *)sqlite3_value_text(apVal[1]); + } + sqlite3Fts3Matchinfo(pContext, pCsr, zArg); + } +} + +/* +** This routine implements the xFindFunction method for the FTS3 +** virtual table. +*/ +static int fts3FindFunctionMethod( + sqlite3_vtab *pVtab, /* Virtual table handle */ + int nArg, /* Number of SQL function arguments */ + const char *zName, /* Name of SQL function */ + void (**pxFunc)(sqlite3_context*,int,sqlite3_value**), /* OUT: Result */ + void **ppArg /* Unused */ +){ + struct Overloaded { + const char *zName; + void (*xFunc)(sqlite3_context*,int,sqlite3_value**); + } aOverload[] = { + { "snippet", fts3SnippetFunc }, + { "offsets", fts3OffsetsFunc }, + { "optimize", fts3OptimizeFunc }, + { "matchinfo", fts3MatchinfoFunc }, + }; + int i; /* Iterator variable */ + + UNUSED_PARAMETER(pVtab); + UNUSED_PARAMETER(nArg); + UNUSED_PARAMETER(ppArg); + + for(i=0; idb; /* Database connection */ + int rc; /* Return Code */ + + /* At this point it must be known if the %_stat table exists or not. + ** So bHasStat may not be 2. */ + rc = fts3SetHasStat(p); + + /* As it happens, the pending terms table is always empty here. This is + ** because an "ALTER TABLE RENAME TABLE" statement inside a transaction + ** always opens a savepoint transaction. And the xSavepoint() method + ** flushes the pending terms table. But leave the (no-op) call to + ** PendingTermsFlush() in in case that changes. + */ + assert( p->nPendingData==0 ); + if( rc==SQLITE_OK ){ + rc = sqlite3Fts3PendingTermsFlush(p); + } + + if( p->zContentTbl==0 ){ + fts3DbExec(&rc, db, + "ALTER TABLE %Q.'%q_content' RENAME TO '%q_content';", + p->zDb, p->zName, zName + ); + } + + if( p->bHasDocsize ){ + fts3DbExec(&rc, db, + "ALTER TABLE %Q.'%q_docsize' RENAME TO '%q_docsize';", + p->zDb, p->zName, zName + ); + } + if( p->bHasStat ){ + fts3DbExec(&rc, db, + "ALTER TABLE %Q.'%q_stat' RENAME TO '%q_stat';", + p->zDb, p->zName, zName + ); + } + fts3DbExec(&rc, db, + "ALTER TABLE %Q.'%q_segments' RENAME TO '%q_segments';", + p->zDb, p->zName, zName + ); + fts3DbExec(&rc, db, + "ALTER TABLE %Q.'%q_segdir' RENAME TO '%q_segdir';", + p->zDb, p->zName, zName + ); + return rc; +} + +/* +** The xSavepoint() method. +** +** Flush the contents of the pending-terms table to disk. +*/ +static int fts3SavepointMethod(sqlite3_vtab *pVtab, int iSavepoint){ + int rc = SQLITE_OK; + UNUSED_PARAMETER(iSavepoint); + assert( ((Fts3Table *)pVtab)->inTransaction ); + assert( ((Fts3Table *)pVtab)->mxSavepoint <= iSavepoint ); + TESTONLY( ((Fts3Table *)pVtab)->mxSavepoint = iSavepoint ); + if( ((Fts3Table *)pVtab)->bIgnoreSavepoint==0 ){ + rc = fts3SyncMethod(pVtab); + } + return rc; +} + +/* +** The xRelease() method. +** +** This is a no-op. +*/ +static int fts3ReleaseMethod(sqlite3_vtab *pVtab, int iSavepoint){ + TESTONLY( Fts3Table *p = (Fts3Table*)pVtab ); + UNUSED_PARAMETER(iSavepoint); + UNUSED_PARAMETER(pVtab); + assert( p->inTransaction ); + assert( p->mxSavepoint >= iSavepoint ); + TESTONLY( p->mxSavepoint = iSavepoint-1 ); + return SQLITE_OK; +} + +/* +** The xRollbackTo() method. +** +** Discard the contents of the pending terms table. +*/ +static int fts3RollbackToMethod(sqlite3_vtab *pVtab, int iSavepoint){ + Fts3Table *p = (Fts3Table*)pVtab; + UNUSED_PARAMETER(iSavepoint); + assert( p->inTransaction ); + TESTONLY( p->mxSavepoint = iSavepoint ); + sqlite3Fts3PendingTermsClear(p); + return SQLITE_OK; +} + +/* +** Return true if zName is the extension on one of the shadow tables used +** by this module. +*/ +static int fts3ShadowName(const char *zName){ + static const char *azName[] = { + "content", "docsize", "segdir", "segments", "stat", + }; + unsigned int i; + for(i=0; ieType==FTSQUERY_PHRASE ){ + int i; + int nToken = pExpr->pPhrase->nToken; + *pnToken += nToken; + for(i=0; ipPhrase->aToken[i]; + int rc = fts3TermSegReaderCursor(pCsr, + pToken->z, pToken->n, pToken->isPrefix, &pToken->pSegcsr + ); + if( rc!=SQLITE_OK ){ + *pRc = rc; + return; + } + } + assert( pExpr->pPhrase->iDoclistToken==0 ); + pExpr->pPhrase->iDoclistToken = -1; + }else{ + *pnOr += (pExpr->eType==FTSQUERY_OR); + fts3EvalAllocateReaders(pCsr, pExpr->pLeft, pnToken, pnOr, pRc); + fts3EvalAllocateReaders(pCsr, pExpr->pRight, pnToken, pnOr, pRc); + } + } +} + +/* +** Arguments pList/nList contain the doclist for token iToken of phrase p. +** It is merged into the main doclist stored in p->doclist.aAll/nAll. +** +** This function assumes that pList points to a buffer allocated using +** sqlite3_malloc(). This function takes responsibility for eventually +** freeing the buffer. +** +** SQLITE_OK is returned if successful, or SQLITE_NOMEM if an error occurs. +*/ +static int fts3EvalPhraseMergeToken( + Fts3Table *pTab, /* FTS Table pointer */ + Fts3Phrase *p, /* Phrase to merge pList/nList into */ + int iToken, /* Token pList/nList corresponds to */ + char *pList, /* Pointer to doclist */ + int nList /* Number of bytes in pList */ +){ + int rc = SQLITE_OK; + assert( iToken!=p->iDoclistToken ); + + if( pList==0 ){ + sqlite3_free(p->doclist.aAll); + p->doclist.aAll = 0; + p->doclist.nAll = 0; + } + + else if( p->iDoclistToken<0 ){ + p->doclist.aAll = pList; + p->doclist.nAll = nList; + } + + else if( p->doclist.aAll==0 ){ + sqlite3_free(pList); + } + + else { + char *pLeft; + char *pRight; + int nLeft; + int nRight; + int nDiff; + + if( p->iDoclistTokendoclist.aAll; + nLeft = p->doclist.nAll; + pRight = pList; + nRight = nList; + nDiff = iToken - p->iDoclistToken; + }else{ + pRight = p->doclist.aAll; + nRight = p->doclist.nAll; + pLeft = pList; + nLeft = nList; + nDiff = p->iDoclistToken - iToken; + } + + rc = fts3DoclistPhraseMerge( + pTab->bDescIdx, nDiff, pLeft, nLeft, &pRight, &nRight + ); + sqlite3_free(pLeft); + p->doclist.aAll = pRight; + p->doclist.nAll = nRight; + } + + if( iToken>p->iDoclistToken ) p->iDoclistToken = iToken; + return rc; +} + +/* +** Load the doclist for phrase p into p->doclist.aAll/nAll. The loaded doclist +** does not take deferred tokens into account. +** +** SQLITE_OK is returned if no error occurs, otherwise an SQLite error code. +*/ +static int fts3EvalPhraseLoad( + Fts3Cursor *pCsr, /* FTS Cursor handle */ + Fts3Phrase *p /* Phrase object */ +){ + Fts3Table *pTab = (Fts3Table *)pCsr->base.pVtab; + int iToken; + int rc = SQLITE_OK; + + for(iToken=0; rc==SQLITE_OK && iTokennToken; iToken++){ + Fts3PhraseToken *pToken = &p->aToken[iToken]; + assert( pToken->pDeferred==0 || pToken->pSegcsr==0 ); + + if( pToken->pSegcsr ){ + int nThis = 0; + char *pThis = 0; + rc = fts3TermSelect(pTab, pToken, p->iColumn, &nThis, &pThis); + if( rc==SQLITE_OK ){ + rc = fts3EvalPhraseMergeToken(pTab, p, iToken, pThis, nThis); + } + } + assert( pToken->pSegcsr==0 ); + } + + return rc; +} + +#ifndef SQLITE_DISABLE_FTS4_DEFERRED +/* +** This function is called on each phrase after the position lists for +** any deferred tokens have been loaded into memory. It updates the phrases +** current position list to include only those positions that are really +** instances of the phrase (after considering deferred tokens). If this +** means that the phrase does not appear in the current row, doclist.pList +** and doclist.nList are both zeroed. +** +** SQLITE_OK is returned if no error occurs, otherwise an SQLite error code. +*/ +static int fts3EvalDeferredPhrase(Fts3Cursor *pCsr, Fts3Phrase *pPhrase){ + int iToken; /* Used to iterate through phrase tokens */ + char *aPoslist = 0; /* Position list for deferred tokens */ + int nPoslist = 0; /* Number of bytes in aPoslist */ + int iPrev = -1; /* Token number of previous deferred token */ + + assert( pPhrase->doclist.bFreeList==0 ); + + for(iToken=0; iTokennToken; iToken++){ + Fts3PhraseToken *pToken = &pPhrase->aToken[iToken]; + Fts3DeferredToken *pDeferred = pToken->pDeferred; + + if( pDeferred ){ + char *pList; + int nList; + int rc = sqlite3Fts3DeferredTokenList(pDeferred, &pList, &nList); + if( rc!=SQLITE_OK ) return rc; + + if( pList==0 ){ + sqlite3_free(aPoslist); + pPhrase->doclist.pList = 0; + pPhrase->doclist.nList = 0; + return SQLITE_OK; + + }else if( aPoslist==0 ){ + aPoslist = pList; + nPoslist = nList; + + }else{ + char *aOut = pList; + char *p1 = aPoslist; + char *p2 = aOut; + + assert( iPrev>=0 ); + fts3PoslistPhraseMerge(&aOut, iToken-iPrev, 0, 1, &p1, &p2); + sqlite3_free(aPoslist); + aPoslist = pList; + nPoslist = (int)(aOut - aPoslist); + if( nPoslist==0 ){ + sqlite3_free(aPoslist); + pPhrase->doclist.pList = 0; + pPhrase->doclist.nList = 0; + return SQLITE_OK; + } + } + iPrev = iToken; + } + } + + if( iPrev>=0 ){ + int nMaxUndeferred = pPhrase->iDoclistToken; + if( nMaxUndeferred<0 ){ + pPhrase->doclist.pList = aPoslist; + pPhrase->doclist.nList = nPoslist; + pPhrase->doclist.iDocid = pCsr->iPrevId; + pPhrase->doclist.bFreeList = 1; + }else{ + int nDistance; + char *p1; + char *p2; + char *aOut; + + if( nMaxUndeferred>iPrev ){ + p1 = aPoslist; + p2 = pPhrase->doclist.pList; + nDistance = nMaxUndeferred - iPrev; + }else{ + p1 = pPhrase->doclist.pList; + p2 = aPoslist; + nDistance = iPrev - nMaxUndeferred; + } + + aOut = (char *)sqlite3_malloc(nPoslist+8); + if( !aOut ){ + sqlite3_free(aPoslist); + return SQLITE_NOMEM; + } + + pPhrase->doclist.pList = aOut; + if( fts3PoslistPhraseMerge(&aOut, nDistance, 0, 1, &p1, &p2) ){ + pPhrase->doclist.bFreeList = 1; + pPhrase->doclist.nList = (int)(aOut - pPhrase->doclist.pList); + }else{ + sqlite3_free(aOut); + pPhrase->doclist.pList = 0; + pPhrase->doclist.nList = 0; + } + sqlite3_free(aPoslist); + } + } + + return SQLITE_OK; +} +#endif /* SQLITE_DISABLE_FTS4_DEFERRED */ + +/* +** Maximum number of tokens a phrase may have to be considered for the +** incremental doclists strategy. +*/ +#define MAX_INCR_PHRASE_TOKENS 4 + +/* +** This function is called for each Fts3Phrase in a full-text query +** expression to initialize the mechanism for returning rows. Once this +** function has been called successfully on an Fts3Phrase, it may be +** used with fts3EvalPhraseNext() to iterate through the matching docids. +** +** If parameter bOptOk is true, then the phrase may (or may not) use the +** incremental loading strategy. Otherwise, the entire doclist is loaded into +** memory within this call. +** +** SQLITE_OK is returned if no error occurs, otherwise an SQLite error code. +*/ +static int fts3EvalPhraseStart(Fts3Cursor *pCsr, int bOptOk, Fts3Phrase *p){ + Fts3Table *pTab = (Fts3Table *)pCsr->base.pVtab; + int rc = SQLITE_OK; /* Error code */ + int i; + + /* Determine if doclists may be loaded from disk incrementally. This is + ** possible if the bOptOk argument is true, the FTS doclists will be + ** scanned in forward order, and the phrase consists of + ** MAX_INCR_PHRASE_TOKENS or fewer tokens, none of which are are "^first" + ** tokens or prefix tokens that cannot use a prefix-index. */ + int bHaveIncr = 0; + int bIncrOk = (bOptOk + && pCsr->bDesc==pTab->bDescIdx + && p->nToken<=MAX_INCR_PHRASE_TOKENS && p->nToken>0 +#ifdef SQLITE_TEST + && pTab->bNoIncrDoclist==0 +#endif + ); + for(i=0; bIncrOk==1 && inToken; i++){ + Fts3PhraseToken *pToken = &p->aToken[i]; + if( pToken->bFirst || (pToken->pSegcsr!=0 && !pToken->pSegcsr->bLookup) ){ + bIncrOk = 0; + } + if( pToken->pSegcsr ) bHaveIncr = 1; + } + + if( bIncrOk && bHaveIncr ){ + /* Use the incremental approach. */ + int iCol = (p->iColumn >= pTab->nColumn ? -1 : p->iColumn); + for(i=0; rc==SQLITE_OK && inToken; i++){ + Fts3PhraseToken *pToken = &p->aToken[i]; + Fts3MultiSegReader *pSegcsr = pToken->pSegcsr; + if( pSegcsr ){ + rc = sqlite3Fts3MsrIncrStart(pTab, pSegcsr, iCol, pToken->z, pToken->n); + } + } + p->bIncr = 1; + }else{ + /* Load the full doclist for the phrase into memory. */ + rc = fts3EvalPhraseLoad(pCsr, p); + p->bIncr = 0; + } + + assert( rc!=SQLITE_OK || p->nToken<1 || p->aToken[0].pSegcsr==0 || p->bIncr ); + return rc; +} + +/* +** This function is used to iterate backwards (from the end to start) +** through doclists. It is used by this module to iterate through phrase +** doclists in reverse and by the fts3_write.c module to iterate through +** pending-terms lists when writing to databases with "order=desc". +** +** The doclist may be sorted in ascending (parameter bDescIdx==0) or +** descending (parameter bDescIdx==1) order of docid. Regardless, this +** function iterates from the end of the doclist to the beginning. +*/ +SQLITE_PRIVATE void sqlite3Fts3DoclistPrev( + int bDescIdx, /* True if the doclist is desc */ + char *aDoclist, /* Pointer to entire doclist */ + int nDoclist, /* Length of aDoclist in bytes */ + char **ppIter, /* IN/OUT: Iterator pointer */ + sqlite3_int64 *piDocid, /* IN/OUT: Docid pointer */ + int *pnList, /* OUT: List length pointer */ + u8 *pbEof /* OUT: End-of-file flag */ +){ + char *p = *ppIter; + + assert( nDoclist>0 ); + assert( *pbEof==0 ); + assert( p || *piDocid==0 ); + assert( !p || (p>aDoclist && p<&aDoclist[nDoclist]) ); + + if( p==0 ){ + sqlite3_int64 iDocid = 0; + char *pNext = 0; + char *pDocid = aDoclist; + char *pEnd = &aDoclist[nDoclist]; + int iMul = 1; + + while( pDocid0 ); + assert( *pbEof==0 ); + assert( p || *piDocid==0 ); + assert( !p || (p>=aDoclist && p<=&aDoclist[nDoclist]) ); + + if( p==0 ){ + p = aDoclist; + p += sqlite3Fts3GetVarint(p, piDocid); + }else{ + fts3PoslistCopy(0, &p); + while( p<&aDoclist[nDoclist] && *p==0 ) p++; + if( p>=&aDoclist[nDoclist] ){ + *pbEof = 1; + }else{ + sqlite3_int64 iVar; + p += sqlite3Fts3GetVarint(p, &iVar); + *piDocid += ((bDescIdx ? -1 : 1) * iVar); + } + } + + *ppIter = p; +} + +/* +** Advance the iterator pDL to the next entry in pDL->aAll/nAll. Set *pbEof +** to true if EOF is reached. +*/ +static void fts3EvalDlPhraseNext( + Fts3Table *pTab, + Fts3Doclist *pDL, + u8 *pbEof +){ + char *pIter; /* Used to iterate through aAll */ + char *pEnd = &pDL->aAll[pDL->nAll]; /* 1 byte past end of aAll */ + + if( pDL->pNextDocid ){ + pIter = pDL->pNextDocid; + }else{ + pIter = pDL->aAll; + } + + if( pIter>=pEnd ){ + /* We have already reached the end of this doclist. EOF. */ + *pbEof = 1; + }else{ + sqlite3_int64 iDelta; + pIter += sqlite3Fts3GetVarint(pIter, &iDelta); + if( pTab->bDescIdx==0 || pDL->pNextDocid==0 ){ + pDL->iDocid += iDelta; + }else{ + pDL->iDocid -= iDelta; + } + pDL->pList = pIter; + fts3PoslistCopy(0, &pIter); + pDL->nList = (int)(pIter - pDL->pList); + + /* pIter now points just past the 0x00 that terminates the position- + ** list for document pDL->iDocid. However, if this position-list was + ** edited in place by fts3EvalNearTrim(), then pIter may not actually + ** point to the start of the next docid value. The following line deals + ** with this case by advancing pIter past the zero-padding added by + ** fts3EvalNearTrim(). */ + while( pIterpNextDocid = pIter; + assert( pIter>=&pDL->aAll[pDL->nAll] || *pIter ); + *pbEof = 0; + } +} + +/* +** Helper type used by fts3EvalIncrPhraseNext() and incrPhraseTokenNext(). +*/ +typedef struct TokenDoclist TokenDoclist; +struct TokenDoclist { + int bIgnore; + sqlite3_int64 iDocid; + char *pList; + int nList; +}; + +/* +** Token pToken is an incrementally loaded token that is part of a +** multi-token phrase. Advance it to the next matching document in the +** database and populate output variable *p with the details of the new +** entry. Or, if the iterator has reached EOF, set *pbEof to true. +** +** If an error occurs, return an SQLite error code. Otherwise, return +** SQLITE_OK. +*/ +static int incrPhraseTokenNext( + Fts3Table *pTab, /* Virtual table handle */ + Fts3Phrase *pPhrase, /* Phrase to advance token of */ + int iToken, /* Specific token to advance */ + TokenDoclist *p, /* OUT: Docid and doclist for new entry */ + u8 *pbEof /* OUT: True if iterator is at EOF */ +){ + int rc = SQLITE_OK; + + if( pPhrase->iDoclistToken==iToken ){ + assert( p->bIgnore==0 ); + assert( pPhrase->aToken[iToken].pSegcsr==0 ); + fts3EvalDlPhraseNext(pTab, &pPhrase->doclist, pbEof); + p->pList = pPhrase->doclist.pList; + p->nList = pPhrase->doclist.nList; + p->iDocid = pPhrase->doclist.iDocid; + }else{ + Fts3PhraseToken *pToken = &pPhrase->aToken[iToken]; + assert( pToken->pDeferred==0 ); + assert( pToken->pSegcsr || pPhrase->iDoclistToken>=0 ); + if( pToken->pSegcsr ){ + assert( p->bIgnore==0 ); + rc = sqlite3Fts3MsrIncrNext( + pTab, pToken->pSegcsr, &p->iDocid, &p->pList, &p->nList + ); + if( p->pList==0 ) *pbEof = 1; + }else{ + p->bIgnore = 1; + } + } + + return rc; +} + + +/* +** The phrase iterator passed as the second argument: +** +** * features at least one token that uses an incremental doclist, and +** +** * does not contain any deferred tokens. +** +** Advance it to the next matching documnent in the database and populate +** the Fts3Doclist.pList and nList fields. +** +** If there is no "next" entry and no error occurs, then *pbEof is set to +** 1 before returning. Otherwise, if no error occurs and the iterator is +** successfully advanced, *pbEof is set to 0. +** +** If an error occurs, return an SQLite error code. Otherwise, return +** SQLITE_OK. +*/ +static int fts3EvalIncrPhraseNext( + Fts3Cursor *pCsr, /* FTS Cursor handle */ + Fts3Phrase *p, /* Phrase object to advance to next docid */ + u8 *pbEof /* OUT: Set to 1 if EOF */ +){ + int rc = SQLITE_OK; + Fts3Doclist *pDL = &p->doclist; + Fts3Table *pTab = (Fts3Table *)pCsr->base.pVtab; + u8 bEof = 0; + + /* This is only called if it is guaranteed that the phrase has at least + ** one incremental token. In which case the bIncr flag is set. */ + assert( p->bIncr==1 ); + + if( p->nToken==1 ){ + rc = sqlite3Fts3MsrIncrNext(pTab, p->aToken[0].pSegcsr, + &pDL->iDocid, &pDL->pList, &pDL->nList + ); + if( pDL->pList==0 ) bEof = 1; + }else{ + int bDescDoclist = pCsr->bDesc; + struct TokenDoclist a[MAX_INCR_PHRASE_TOKENS]; + + memset(a, 0, sizeof(a)); + assert( p->nToken<=MAX_INCR_PHRASE_TOKENS ); + assert( p->iDoclistTokennToken && bEof==0; i++){ + rc = incrPhraseTokenNext(pTab, p, i, &a[i], &bEof); + if( a[i].bIgnore==0 && (bMaxSet==0 || DOCID_CMP(iMax, a[i].iDocid)<0) ){ + iMax = a[i].iDocid; + bMaxSet = 1; + } + } + assert( rc!=SQLITE_OK || (p->nToken>=1 && a[p->nToken-1].bIgnore==0) ); + assert( rc!=SQLITE_OK || bMaxSet ); + + /* Keep advancing iterators until they all point to the same document */ + for(i=0; inToken; i++){ + while( rc==SQLITE_OK && bEof==0 + && a[i].bIgnore==0 && DOCID_CMP(a[i].iDocid, iMax)<0 + ){ + rc = incrPhraseTokenNext(pTab, p, i, &a[i], &bEof); + if( DOCID_CMP(a[i].iDocid, iMax)>0 ){ + iMax = a[i].iDocid; + i = 0; + } + } + } + + /* Check if the current entries really are a phrase match */ + if( bEof==0 ){ + int nList = 0; + int nByte = a[p->nToken-1].nList; + char *aDoclist = sqlite3_malloc(nByte+FTS3_BUFFER_PADDING); + if( !aDoclist ) return SQLITE_NOMEM; + memcpy(aDoclist, a[p->nToken-1].pList, nByte+1); + memset(&aDoclist[nByte], 0, FTS3_BUFFER_PADDING); + + for(i=0; i<(p->nToken-1); i++){ + if( a[i].bIgnore==0 ){ + char *pL = a[i].pList; + char *pR = aDoclist; + char *pOut = aDoclist; + int nDist = p->nToken-1-i; + int res = fts3PoslistPhraseMerge(&pOut, nDist, 0, 1, &pL, &pR); + if( res==0 ) break; + nList = (int)(pOut - aDoclist); + } + } + if( i==(p->nToken-1) ){ + pDL->iDocid = iMax; + pDL->pList = aDoclist; + pDL->nList = nList; + pDL->bFreeList = 1; + break; + } + sqlite3_free(aDoclist); + } + } + } + + *pbEof = bEof; + return rc; +} + +/* +** Attempt to move the phrase iterator to point to the next matching docid. +** If an error occurs, return an SQLite error code. Otherwise, return +** SQLITE_OK. +** +** If there is no "next" entry and no error occurs, then *pbEof is set to +** 1 before returning. Otherwise, if no error occurs and the iterator is +** successfully advanced, *pbEof is set to 0. +*/ +static int fts3EvalPhraseNext( + Fts3Cursor *pCsr, /* FTS Cursor handle */ + Fts3Phrase *p, /* Phrase object to advance to next docid */ + u8 *pbEof /* OUT: Set to 1 if EOF */ +){ + int rc = SQLITE_OK; + Fts3Doclist *pDL = &p->doclist; + Fts3Table *pTab = (Fts3Table *)pCsr->base.pVtab; + + if( p->bIncr ){ + rc = fts3EvalIncrPhraseNext(pCsr, p, pbEof); + }else if( pCsr->bDesc!=pTab->bDescIdx && pDL->nAll ){ + sqlite3Fts3DoclistPrev(pTab->bDescIdx, pDL->aAll, pDL->nAll, + &pDL->pNextDocid, &pDL->iDocid, &pDL->nList, pbEof + ); + pDL->pList = pDL->pNextDocid; + }else{ + fts3EvalDlPhraseNext(pTab, pDL, pbEof); + } + + return rc; +} + +/* +** +** If *pRc is not SQLITE_OK when this function is called, it is a no-op. +** Otherwise, fts3EvalPhraseStart() is called on all phrases within the +** expression. Also the Fts3Expr.bDeferred variable is set to true for any +** expressions for which all descendent tokens are deferred. +** +** If parameter bOptOk is zero, then it is guaranteed that the +** Fts3Phrase.doclist.aAll/nAll variables contain the entire doclist for +** each phrase in the expression (subject to deferred token processing). +** Or, if bOptOk is non-zero, then one or more tokens within the expression +** may be loaded incrementally, meaning doclist.aAll/nAll is not available. +** +** If an error occurs within this function, *pRc is set to an SQLite error +** code before returning. +*/ +static void fts3EvalStartReaders( + Fts3Cursor *pCsr, /* FTS Cursor handle */ + Fts3Expr *pExpr, /* Expression to initialize phrases in */ + int *pRc /* IN/OUT: Error code */ +){ + if( pExpr && SQLITE_OK==*pRc ){ + if( pExpr->eType==FTSQUERY_PHRASE ){ + int nToken = pExpr->pPhrase->nToken; + if( nToken ){ + int i; + for(i=0; ipPhrase->aToken[i].pDeferred==0 ) break; + } + pExpr->bDeferred = (i==nToken); + } + *pRc = fts3EvalPhraseStart(pCsr, 1, pExpr->pPhrase); + }else{ + fts3EvalStartReaders(pCsr, pExpr->pLeft, pRc); + fts3EvalStartReaders(pCsr, pExpr->pRight, pRc); + pExpr->bDeferred = (pExpr->pLeft->bDeferred && pExpr->pRight->bDeferred); + } + } +} + +/* +** An array of the following structures is assembled as part of the process +** of selecting tokens to defer before the query starts executing (as part +** of the xFilter() method). There is one element in the array for each +** token in the FTS expression. +** +** Tokens are divided into AND/NEAR clusters. All tokens in a cluster belong +** to phrases that are connected only by AND and NEAR operators (not OR or +** NOT). When determining tokens to defer, each AND/NEAR cluster is considered +** separately. The root of a tokens AND/NEAR cluster is stored in +** Fts3TokenAndCost.pRoot. +*/ +typedef struct Fts3TokenAndCost Fts3TokenAndCost; +struct Fts3TokenAndCost { + Fts3Phrase *pPhrase; /* The phrase the token belongs to */ + int iToken; /* Position of token in phrase */ + Fts3PhraseToken *pToken; /* The token itself */ + Fts3Expr *pRoot; /* Root of NEAR/AND cluster */ + int nOvfl; /* Number of overflow pages to load doclist */ + int iCol; /* The column the token must match */ +}; + +/* +** This function is used to populate an allocated Fts3TokenAndCost array. +** +** If *pRc is not SQLITE_OK when this function is called, it is a no-op. +** Otherwise, if an error occurs during execution, *pRc is set to an +** SQLite error code. +*/ +static void fts3EvalTokenCosts( + Fts3Cursor *pCsr, /* FTS Cursor handle */ + Fts3Expr *pRoot, /* Root of current AND/NEAR cluster */ + Fts3Expr *pExpr, /* Expression to consider */ + Fts3TokenAndCost **ppTC, /* Write new entries to *(*ppTC)++ */ + Fts3Expr ***ppOr, /* Write new OR root to *(*ppOr)++ */ + int *pRc /* IN/OUT: Error code */ +){ + if( *pRc==SQLITE_OK ){ + if( pExpr->eType==FTSQUERY_PHRASE ){ + Fts3Phrase *pPhrase = pExpr->pPhrase; + int i; + for(i=0; *pRc==SQLITE_OK && inToken; i++){ + Fts3TokenAndCost *pTC = (*ppTC)++; + pTC->pPhrase = pPhrase; + pTC->iToken = i; + pTC->pRoot = pRoot; + pTC->pToken = &pPhrase->aToken[i]; + pTC->iCol = pPhrase->iColumn; + *pRc = sqlite3Fts3MsrOvfl(pCsr, pTC->pToken->pSegcsr, &pTC->nOvfl); + } + }else if( pExpr->eType!=FTSQUERY_NOT ){ + assert( pExpr->eType==FTSQUERY_OR + || pExpr->eType==FTSQUERY_AND + || pExpr->eType==FTSQUERY_NEAR + ); + assert( pExpr->pLeft && pExpr->pRight ); + if( pExpr->eType==FTSQUERY_OR ){ + pRoot = pExpr->pLeft; + **ppOr = pRoot; + (*ppOr)++; + } + fts3EvalTokenCosts(pCsr, pRoot, pExpr->pLeft, ppTC, ppOr, pRc); + if( pExpr->eType==FTSQUERY_OR ){ + pRoot = pExpr->pRight; + **ppOr = pRoot; + (*ppOr)++; + } + fts3EvalTokenCosts(pCsr, pRoot, pExpr->pRight, ppTC, ppOr, pRc); + } + } +} + +/* +** Determine the average document (row) size in pages. If successful, +** write this value to *pnPage and return SQLITE_OK. Otherwise, return +** an SQLite error code. +** +** The average document size in pages is calculated by first calculating +** determining the average size in bytes, B. If B is less than the amount +** of data that will fit on a single leaf page of an intkey table in +** this database, then the average docsize is 1. Otherwise, it is 1 plus +** the number of overflow pages consumed by a record B bytes in size. +*/ +static int fts3EvalAverageDocsize(Fts3Cursor *pCsr, int *pnPage){ + int rc = SQLITE_OK; + if( pCsr->nRowAvg==0 ){ + /* The average document size, which is required to calculate the cost + ** of each doclist, has not yet been determined. Read the required + ** data from the %_stat table to calculate it. + ** + ** Entry 0 of the %_stat table is a blob containing (nCol+1) FTS3 + ** varints, where nCol is the number of columns in the FTS3 table. + ** The first varint is the number of documents currently stored in + ** the table. The following nCol varints contain the total amount of + ** data stored in all rows of each column of the table, from left + ** to right. + */ + Fts3Table *p = (Fts3Table*)pCsr->base.pVtab; + sqlite3_stmt *pStmt; + sqlite3_int64 nDoc = 0; + sqlite3_int64 nByte = 0; + const char *pEnd; + const char *a; + + rc = sqlite3Fts3SelectDoctotal(p, &pStmt); + if( rc!=SQLITE_OK ) return rc; + a = sqlite3_column_blob(pStmt, 0); + assert( a ); + + pEnd = &a[sqlite3_column_bytes(pStmt, 0)]; + a += sqlite3Fts3GetVarint(a, &nDoc); + while( anDoc = nDoc; + pCsr->nRowAvg = (int)(((nByte / nDoc) + p->nPgsz) / p->nPgsz); + assert( pCsr->nRowAvg>0 ); + rc = sqlite3_reset(pStmt); + } + + *pnPage = pCsr->nRowAvg; + return rc; +} + +/* +** This function is called to select the tokens (if any) that will be +** deferred. The array aTC[] has already been populated when this is +** called. +** +** This function is called once for each AND/NEAR cluster in the +** expression. Each invocation determines which tokens to defer within +** the cluster with root node pRoot. See comments above the definition +** of struct Fts3TokenAndCost for more details. +** +** If no error occurs, SQLITE_OK is returned and sqlite3Fts3DeferToken() +** called on each token to defer. Otherwise, an SQLite error code is +** returned. +*/ +static int fts3EvalSelectDeferred( + Fts3Cursor *pCsr, /* FTS Cursor handle */ + Fts3Expr *pRoot, /* Consider tokens with this root node */ + Fts3TokenAndCost *aTC, /* Array of expression tokens and costs */ + int nTC /* Number of entries in aTC[] */ +){ + Fts3Table *pTab = (Fts3Table *)pCsr->base.pVtab; + int nDocSize = 0; /* Number of pages per doc loaded */ + int rc = SQLITE_OK; /* Return code */ + int ii; /* Iterator variable for various purposes */ + int nOvfl = 0; /* Total overflow pages used by doclists */ + int nToken = 0; /* Total number of tokens in cluster */ + + int nMinEst = 0; /* The minimum count for any phrase so far. */ + int nLoad4 = 1; /* (Phrases that will be loaded)^4. */ + + /* Tokens are never deferred for FTS tables created using the content=xxx + ** option. The reason being that it is not guaranteed that the content + ** table actually contains the same data as the index. To prevent this from + ** causing any problems, the deferred token optimization is completely + ** disabled for content=xxx tables. */ + if( pTab->zContentTbl ){ + return SQLITE_OK; + } + + /* Count the tokens in this AND/NEAR cluster. If none of the doclists + ** associated with the tokens spill onto overflow pages, or if there is + ** only 1 token, exit early. No tokens to defer in this case. */ + for(ii=0; ii0 ); + + + /* Iterate through all tokens in this AND/NEAR cluster, in ascending order + ** of the number of overflow pages that will be loaded by the pager layer + ** to retrieve the entire doclist for the token from the full-text index. + ** Load the doclists for tokens that are either: + ** + ** a. The cheapest token in the entire query (i.e. the one visited by the + ** first iteration of this loop), or + ** + ** b. Part of a multi-token phrase. + ** + ** After each token doclist is loaded, merge it with the others from the + ** same phrase and count the number of documents that the merged doclist + ** contains. Set variable "nMinEst" to the smallest number of documents in + ** any phrase doclist for which 1 or more token doclists have been loaded. + ** Let nOther be the number of other phrases for which it is certain that + ** one or more tokens will not be deferred. + ** + ** Then, for each token, defer it if loading the doclist would result in + ** loading N or more overflow pages into memory, where N is computed as: + ** + ** (nMinEst + 4^nOther - 1) / (4^nOther) + */ + for(ii=0; iinOvfl) + ){ + pTC = &aTC[iTC]; + } + } + assert( pTC ); + + if( ii && pTC->nOvfl>=((nMinEst+(nLoad4/4)-1)/(nLoad4/4))*nDocSize ){ + /* The number of overflow pages to load for this (and therefore all + ** subsequent) tokens is greater than the estimated number of pages + ** that will be loaded if all subsequent tokens are deferred. + */ + Fts3PhraseToken *pToken = pTC->pToken; + rc = sqlite3Fts3DeferToken(pCsr, pToken, pTC->iCol); + fts3SegReaderCursorFree(pToken->pSegcsr); + pToken->pSegcsr = 0; + }else{ + /* Set nLoad4 to the value of (4^nOther) for the next iteration of the + ** for-loop. Except, limit the value to 2^24 to prevent it from + ** overflowing the 32-bit integer it is stored in. */ + if( ii<12 ) nLoad4 = nLoad4*4; + + if( ii==0 || (pTC->pPhrase->nToken>1 && ii!=nToken-1) ){ + /* Either this is the cheapest token in the entire query, or it is + ** part of a multi-token phrase. Either way, the entire doclist will + ** (eventually) be loaded into memory. It may as well be now. */ + Fts3PhraseToken *pToken = pTC->pToken; + int nList = 0; + char *pList = 0; + rc = fts3TermSelect(pTab, pToken, pTC->iCol, &nList, &pList); + assert( rc==SQLITE_OK || pList==0 ); + if( rc==SQLITE_OK ){ + rc = fts3EvalPhraseMergeToken( + pTab, pTC->pPhrase, pTC->iToken,pList,nList + ); + } + if( rc==SQLITE_OK ){ + int nCount; + nCount = fts3DoclistCountDocids( + pTC->pPhrase->doclist.aAll, pTC->pPhrase->doclist.nAll + ); + if( ii==0 || nCountpToken = 0; + } + + return rc; +} + +/* +** This function is called from within the xFilter method. It initializes +** the full-text query currently stored in pCsr->pExpr. To iterate through +** the results of a query, the caller does: +** +** fts3EvalStart(pCsr); +** while( 1 ){ +** fts3EvalNext(pCsr); +** if( pCsr->bEof ) break; +** ... return row pCsr->iPrevId to the caller ... +** } +*/ +static int fts3EvalStart(Fts3Cursor *pCsr){ + Fts3Table *pTab = (Fts3Table *)pCsr->base.pVtab; + int rc = SQLITE_OK; + int nToken = 0; + int nOr = 0; + + /* Allocate a MultiSegReader for each token in the expression. */ + fts3EvalAllocateReaders(pCsr, pCsr->pExpr, &nToken, &nOr, &rc); + + /* Determine which, if any, tokens in the expression should be deferred. */ +#ifndef SQLITE_DISABLE_FTS4_DEFERRED + if( rc==SQLITE_OK && nToken>1 && pTab->bFts4 ){ + Fts3TokenAndCost *aTC; + Fts3Expr **apOr; + aTC = (Fts3TokenAndCost *)sqlite3_malloc64( + sizeof(Fts3TokenAndCost) * nToken + + sizeof(Fts3Expr *) * nOr * 2 + ); + apOr = (Fts3Expr **)&aTC[nToken]; + + if( !aTC ){ + rc = SQLITE_NOMEM; + }else{ + int ii; + Fts3TokenAndCost *pTC = aTC; + Fts3Expr **ppOr = apOr; + + fts3EvalTokenCosts(pCsr, 0, pCsr->pExpr, &pTC, &ppOr, &rc); + nToken = (int)(pTC-aTC); + nOr = (int)(ppOr-apOr); + + if( rc==SQLITE_OK ){ + rc = fts3EvalSelectDeferred(pCsr, 0, aTC, nToken); + for(ii=0; rc==SQLITE_OK && iipExpr, &rc); + return rc; +} + +/* +** Invalidate the current position list for phrase pPhrase. +*/ +static void fts3EvalInvalidatePoslist(Fts3Phrase *pPhrase){ + if( pPhrase->doclist.bFreeList ){ + sqlite3_free(pPhrase->doclist.pList); + } + pPhrase->doclist.pList = 0; + pPhrase->doclist.nList = 0; + pPhrase->doclist.bFreeList = 0; +} + +/* +** This function is called to edit the position list associated with +** the phrase object passed as the fifth argument according to a NEAR +** condition. For example: +** +** abc NEAR/5 "def ghi" +** +** Parameter nNear is passed the NEAR distance of the expression (5 in +** the example above). When this function is called, *paPoslist points to +** the position list, and *pnToken is the number of phrase tokens in, the +** phrase on the other side of the NEAR operator to pPhrase. For example, +** if pPhrase refers to the "def ghi" phrase, then *paPoslist points to +** the position list associated with phrase "abc". +** +** All positions in the pPhrase position list that are not sufficiently +** close to a position in the *paPoslist position list are removed. If this +** leaves 0 positions, zero is returned. Otherwise, non-zero. +** +** Before returning, *paPoslist is set to point to the position lsit +** associated with pPhrase. And *pnToken is set to the number of tokens in +** pPhrase. +*/ +static int fts3EvalNearTrim( + int nNear, /* NEAR distance. As in "NEAR/nNear". */ + char *aTmp, /* Temporary space to use */ + char **paPoslist, /* IN/OUT: Position list */ + int *pnToken, /* IN/OUT: Tokens in phrase of *paPoslist */ + Fts3Phrase *pPhrase /* The phrase object to trim the doclist of */ +){ + int nParam1 = nNear + pPhrase->nToken; + int nParam2 = nNear + *pnToken; + int nNew; + char *p2; + char *pOut; + int res; + + assert( pPhrase->doclist.pList ); + + p2 = pOut = pPhrase->doclist.pList; + res = fts3PoslistNearMerge( + &pOut, aTmp, nParam1, nParam2, paPoslist, &p2 + ); + if( res ){ + nNew = (int)(pOut - pPhrase->doclist.pList) - 1; + assert( pPhrase->doclist.pList[nNew]=='\0' ); + assert( nNew<=pPhrase->doclist.nList && nNew>0 ); + memset(&pPhrase->doclist.pList[nNew], 0, pPhrase->doclist.nList - nNew); + pPhrase->doclist.nList = nNew; + *paPoslist = pPhrase->doclist.pList; + *pnToken = pPhrase->nToken; + } + + return res; +} + +/* +** This function is a no-op if *pRc is other than SQLITE_OK when it is called. +** Otherwise, it advances the expression passed as the second argument to +** point to the next matching row in the database. Expressions iterate through +** matching rows in docid order. Ascending order if Fts3Cursor.bDesc is zero, +** or descending if it is non-zero. +** +** If an error occurs, *pRc is set to an SQLite error code. Otherwise, if +** successful, the following variables in pExpr are set: +** +** Fts3Expr.bEof (non-zero if EOF - there is no next row) +** Fts3Expr.iDocid (valid if bEof==0. The docid of the next row) +** +** If the expression is of type FTSQUERY_PHRASE, and the expression is not +** at EOF, then the following variables are populated with the position list +** for the phrase for the visited row: +** +** FTs3Expr.pPhrase->doclist.nList (length of pList in bytes) +** FTs3Expr.pPhrase->doclist.pList (pointer to position list) +** +** It says above that this function advances the expression to the next +** matching row. This is usually true, but there are the following exceptions: +** +** 1. Deferred tokens are not taken into account. If a phrase consists +** entirely of deferred tokens, it is assumed to match every row in +** the db. In this case the position-list is not populated at all. +** +** Or, if a phrase contains one or more deferred tokens and one or +** more non-deferred tokens, then the expression is advanced to the +** next possible match, considering only non-deferred tokens. In other +** words, if the phrase is "A B C", and "B" is deferred, the expression +** is advanced to the next row that contains an instance of "A * C", +** where "*" may match any single token. The position list in this case +** is populated as for "A * C" before returning. +** +** 2. NEAR is treated as AND. If the expression is "x NEAR y", it is +** advanced to point to the next row that matches "x AND y". +** +** See sqlite3Fts3EvalTestDeferred() for details on testing if a row is +** really a match, taking into account deferred tokens and NEAR operators. +*/ +static void fts3EvalNextRow( + Fts3Cursor *pCsr, /* FTS Cursor handle */ + Fts3Expr *pExpr, /* Expr. to advance to next matching row */ + int *pRc /* IN/OUT: Error code */ +){ + if( *pRc==SQLITE_OK ){ + int bDescDoclist = pCsr->bDesc; /* Used by DOCID_CMP() macro */ + assert( pExpr->bEof==0 ); + pExpr->bStart = 1; + + switch( pExpr->eType ){ + case FTSQUERY_NEAR: + case FTSQUERY_AND: { + Fts3Expr *pLeft = pExpr->pLeft; + Fts3Expr *pRight = pExpr->pRight; + assert( !pLeft->bDeferred || !pRight->bDeferred ); + + if( pLeft->bDeferred ){ + /* LHS is entirely deferred. So we assume it matches every row. + ** Advance the RHS iterator to find the next row visited. */ + fts3EvalNextRow(pCsr, pRight, pRc); + pExpr->iDocid = pRight->iDocid; + pExpr->bEof = pRight->bEof; + }else if( pRight->bDeferred ){ + /* RHS is entirely deferred. So we assume it matches every row. + ** Advance the LHS iterator to find the next row visited. */ + fts3EvalNextRow(pCsr, pLeft, pRc); + pExpr->iDocid = pLeft->iDocid; + pExpr->bEof = pLeft->bEof; + }else{ + /* Neither the RHS or LHS are deferred. */ + fts3EvalNextRow(pCsr, pLeft, pRc); + fts3EvalNextRow(pCsr, pRight, pRc); + while( !pLeft->bEof && !pRight->bEof && *pRc==SQLITE_OK ){ + sqlite3_int64 iDiff = DOCID_CMP(pLeft->iDocid, pRight->iDocid); + if( iDiff==0 ) break; + if( iDiff<0 ){ + fts3EvalNextRow(pCsr, pLeft, pRc); + }else{ + fts3EvalNextRow(pCsr, pRight, pRc); + } + } + pExpr->iDocid = pLeft->iDocid; + pExpr->bEof = (pLeft->bEof || pRight->bEof); + if( pExpr->eType==FTSQUERY_NEAR && pExpr->bEof ){ + assert( pRight->eType==FTSQUERY_PHRASE ); + if( pRight->pPhrase->doclist.aAll ){ + Fts3Doclist *pDl = &pRight->pPhrase->doclist; + while( *pRc==SQLITE_OK && pRight->bEof==0 ){ + memset(pDl->pList, 0, pDl->nList); + fts3EvalNextRow(pCsr, pRight, pRc); + } + } + if( pLeft->pPhrase && pLeft->pPhrase->doclist.aAll ){ + Fts3Doclist *pDl = &pLeft->pPhrase->doclist; + while( *pRc==SQLITE_OK && pLeft->bEof==0 ){ + memset(pDl->pList, 0, pDl->nList); + fts3EvalNextRow(pCsr, pLeft, pRc); + } + } + } + } + break; + } + + case FTSQUERY_OR: { + Fts3Expr *pLeft = pExpr->pLeft; + Fts3Expr *pRight = pExpr->pRight; + sqlite3_int64 iCmp = DOCID_CMP(pLeft->iDocid, pRight->iDocid); + + assert( pLeft->bStart || pLeft->iDocid==pRight->iDocid ); + assert( pRight->bStart || pLeft->iDocid==pRight->iDocid ); + + if( pRight->bEof || (pLeft->bEof==0 && iCmp<0) ){ + fts3EvalNextRow(pCsr, pLeft, pRc); + }else if( pLeft->bEof || iCmp>0 ){ + fts3EvalNextRow(pCsr, pRight, pRc); + }else{ + fts3EvalNextRow(pCsr, pLeft, pRc); + fts3EvalNextRow(pCsr, pRight, pRc); + } + + pExpr->bEof = (pLeft->bEof && pRight->bEof); + iCmp = DOCID_CMP(pLeft->iDocid, pRight->iDocid); + if( pRight->bEof || (pLeft->bEof==0 && iCmp<0) ){ + pExpr->iDocid = pLeft->iDocid; + }else{ + pExpr->iDocid = pRight->iDocid; + } + + break; + } + + case FTSQUERY_NOT: { + Fts3Expr *pLeft = pExpr->pLeft; + Fts3Expr *pRight = pExpr->pRight; + + if( pRight->bStart==0 ){ + fts3EvalNextRow(pCsr, pRight, pRc); + assert( *pRc!=SQLITE_OK || pRight->bStart ); + } + + fts3EvalNextRow(pCsr, pLeft, pRc); + if( pLeft->bEof==0 ){ + while( !*pRc + && !pRight->bEof + && DOCID_CMP(pLeft->iDocid, pRight->iDocid)>0 + ){ + fts3EvalNextRow(pCsr, pRight, pRc); + } + } + pExpr->iDocid = pLeft->iDocid; + pExpr->bEof = pLeft->bEof; + break; + } + + default: { + Fts3Phrase *pPhrase = pExpr->pPhrase; + fts3EvalInvalidatePoslist(pPhrase); + *pRc = fts3EvalPhraseNext(pCsr, pPhrase, &pExpr->bEof); + pExpr->iDocid = pPhrase->doclist.iDocid; + break; + } + } + } +} + +/* +** If *pRc is not SQLITE_OK, or if pExpr is not the root node of a NEAR +** cluster, then this function returns 1 immediately. +** +** Otherwise, it checks if the current row really does match the NEAR +** expression, using the data currently stored in the position lists +** (Fts3Expr->pPhrase.doclist.pList/nList) for each phrase in the expression. +** +** If the current row is a match, the position list associated with each +** phrase in the NEAR expression is edited in place to contain only those +** phrase instances sufficiently close to their peers to satisfy all NEAR +** constraints. In this case it returns 1. If the NEAR expression does not +** match the current row, 0 is returned. The position lists may or may not +** be edited if 0 is returned. +*/ +static int fts3EvalNearTest(Fts3Expr *pExpr, int *pRc){ + int res = 1; + + /* The following block runs if pExpr is the root of a NEAR query. + ** For example, the query: + ** + ** "w" NEAR "x" NEAR "y" NEAR "z" + ** + ** which is represented in tree form as: + ** + ** | + ** +--NEAR--+ <-- root of NEAR query + ** | | + ** +--NEAR--+ "z" + ** | | + ** +--NEAR--+ "y" + ** | | + ** "w" "x" + ** + ** The right-hand child of a NEAR node is always a phrase. The + ** left-hand child may be either a phrase or a NEAR node. There are + ** no exceptions to this - it's the way the parser in fts3_expr.c works. + */ + if( *pRc==SQLITE_OK + && pExpr->eType==FTSQUERY_NEAR + && (pExpr->pParent==0 || pExpr->pParent->eType!=FTSQUERY_NEAR) + ){ + Fts3Expr *p; + sqlite3_int64 nTmp = 0; /* Bytes of temp space */ + char *aTmp; /* Temp space for PoslistNearMerge() */ + + /* Allocate temporary working space. */ + for(p=pExpr; p->pLeft; p=p->pLeft){ + assert( p->pRight->pPhrase->doclist.nList>0 ); + nTmp += p->pRight->pPhrase->doclist.nList; + } + nTmp += p->pPhrase->doclist.nList; + aTmp = sqlite3_malloc64(nTmp*2); + if( !aTmp ){ + *pRc = SQLITE_NOMEM; + res = 0; + }else{ + char *aPoslist = p->pPhrase->doclist.pList; + int nToken = p->pPhrase->nToken; + + for(p=p->pParent;res && p && p->eType==FTSQUERY_NEAR; p=p->pParent){ + Fts3Phrase *pPhrase = p->pRight->pPhrase; + int nNear = p->nNear; + res = fts3EvalNearTrim(nNear, aTmp, &aPoslist, &nToken, pPhrase); + } + + aPoslist = pExpr->pRight->pPhrase->doclist.pList; + nToken = pExpr->pRight->pPhrase->nToken; + for(p=pExpr->pLeft; p && res; p=p->pLeft){ + int nNear; + Fts3Phrase *pPhrase; + assert( p->pParent && p->pParent->pLeft==p ); + nNear = p->pParent->nNear; + pPhrase = ( + p->eType==FTSQUERY_NEAR ? p->pRight->pPhrase : p->pPhrase + ); + res = fts3EvalNearTrim(nNear, aTmp, &aPoslist, &nToken, pPhrase); + } + } + + sqlite3_free(aTmp); + } + + return res; +} + +/* +** This function is a helper function for sqlite3Fts3EvalTestDeferred(). +** Assuming no error occurs or has occurred, It returns non-zero if the +** expression passed as the second argument matches the row that pCsr +** currently points to, or zero if it does not. +** +** If *pRc is not SQLITE_OK when this function is called, it is a no-op. +** If an error occurs during execution of this function, *pRc is set to +** the appropriate SQLite error code. In this case the returned value is +** undefined. +*/ +static int fts3EvalTestExpr( + Fts3Cursor *pCsr, /* FTS cursor handle */ + Fts3Expr *pExpr, /* Expr to test. May or may not be root. */ + int *pRc /* IN/OUT: Error code */ +){ + int bHit = 1; /* Return value */ + if( *pRc==SQLITE_OK ){ + switch( pExpr->eType ){ + case FTSQUERY_NEAR: + case FTSQUERY_AND: + bHit = ( + fts3EvalTestExpr(pCsr, pExpr->pLeft, pRc) + && fts3EvalTestExpr(pCsr, pExpr->pRight, pRc) + && fts3EvalNearTest(pExpr, pRc) + ); + + /* If the NEAR expression does not match any rows, zero the doclist for + ** all phrases involved in the NEAR. This is because the snippet(), + ** offsets() and matchinfo() functions are not supposed to recognize + ** any instances of phrases that are part of unmatched NEAR queries. + ** For example if this expression: + ** + ** ... MATCH 'a OR (b NEAR c)' + ** + ** is matched against a row containing: + ** + ** 'a b d e' + ** + ** then any snippet() should ony highlight the "a" term, not the "b" + ** (as "b" is part of a non-matching NEAR clause). + */ + if( bHit==0 + && pExpr->eType==FTSQUERY_NEAR + && (pExpr->pParent==0 || pExpr->pParent->eType!=FTSQUERY_NEAR) + ){ + Fts3Expr *p; + for(p=pExpr; p->pPhrase==0; p=p->pLeft){ + if( p->pRight->iDocid==pCsr->iPrevId ){ + fts3EvalInvalidatePoslist(p->pRight->pPhrase); + } + } + if( p->iDocid==pCsr->iPrevId ){ + fts3EvalInvalidatePoslist(p->pPhrase); + } + } + + break; + + case FTSQUERY_OR: { + int bHit1 = fts3EvalTestExpr(pCsr, pExpr->pLeft, pRc); + int bHit2 = fts3EvalTestExpr(pCsr, pExpr->pRight, pRc); + bHit = bHit1 || bHit2; + break; + } + + case FTSQUERY_NOT: + bHit = ( + fts3EvalTestExpr(pCsr, pExpr->pLeft, pRc) + && !fts3EvalTestExpr(pCsr, pExpr->pRight, pRc) + ); + break; + + default: { +#ifndef SQLITE_DISABLE_FTS4_DEFERRED + if( pCsr->pDeferred + && (pExpr->iDocid==pCsr->iPrevId || pExpr->bDeferred) + ){ + Fts3Phrase *pPhrase = pExpr->pPhrase; + assert( pExpr->bDeferred || pPhrase->doclist.bFreeList==0 ); + if( pExpr->bDeferred ){ + fts3EvalInvalidatePoslist(pPhrase); + } + *pRc = fts3EvalDeferredPhrase(pCsr, pPhrase); + bHit = (pPhrase->doclist.pList!=0); + pExpr->iDocid = pCsr->iPrevId; + }else +#endif + { + bHit = (pExpr->bEof==0 && pExpr->iDocid==pCsr->iPrevId); + } + break; + } + } + } + return bHit; +} + +/* +** This function is called as the second part of each xNext operation when +** iterating through the results of a full-text query. At this point the +** cursor points to a row that matches the query expression, with the +** following caveats: +** +** * Up until this point, "NEAR" operators in the expression have been +** treated as "AND". +** +** * Deferred tokens have not yet been considered. +** +** If *pRc is not SQLITE_OK when this function is called, it immediately +** returns 0. Otherwise, it tests whether or not after considering NEAR +** operators and deferred tokens the current row is still a match for the +** expression. It returns 1 if both of the following are true: +** +** 1. *pRc is SQLITE_OK when this function returns, and +** +** 2. After scanning the current FTS table row for the deferred tokens, +** it is determined that the row does *not* match the query. +** +** Or, if no error occurs and it seems the current row does match the FTS +** query, return 0. +*/ +SQLITE_PRIVATE int sqlite3Fts3EvalTestDeferred(Fts3Cursor *pCsr, int *pRc){ + int rc = *pRc; + int bMiss = 0; + if( rc==SQLITE_OK ){ + + /* If there are one or more deferred tokens, load the current row into + ** memory and scan it to determine the position list for each deferred + ** token. Then, see if this row is really a match, considering deferred + ** tokens and NEAR operators (neither of which were taken into account + ** earlier, by fts3EvalNextRow()). + */ + if( pCsr->pDeferred ){ + rc = fts3CursorSeek(0, pCsr); + if( rc==SQLITE_OK ){ + rc = sqlite3Fts3CacheDeferredDoclists(pCsr); + } + } + bMiss = (0==fts3EvalTestExpr(pCsr, pCsr->pExpr, &rc)); + + /* Free the position-lists accumulated for each deferred token above. */ + sqlite3Fts3FreeDeferredDoclists(pCsr); + *pRc = rc; + } + return (rc==SQLITE_OK && bMiss); +} + +/* +** Advance to the next document that matches the FTS expression in +** Fts3Cursor.pExpr. +*/ +static int fts3EvalNext(Fts3Cursor *pCsr){ + int rc = SQLITE_OK; /* Return Code */ + Fts3Expr *pExpr = pCsr->pExpr; + assert( pCsr->isEof==0 ); + if( pExpr==0 ){ + pCsr->isEof = 1; + }else{ + do { + if( pCsr->isRequireSeek==0 ){ + sqlite3_reset(pCsr->pStmt); + } + assert( sqlite3_data_count(pCsr->pStmt)==0 ); + fts3EvalNextRow(pCsr, pExpr, &rc); + pCsr->isEof = pExpr->bEof; + pCsr->isRequireSeek = 1; + pCsr->isMatchinfoNeeded = 1; + pCsr->iPrevId = pExpr->iDocid; + }while( pCsr->isEof==0 && sqlite3Fts3EvalTestDeferred(pCsr, &rc) ); + } + + /* Check if the cursor is past the end of the docid range specified + ** by Fts3Cursor.iMinDocid/iMaxDocid. If so, set the EOF flag. */ + if( rc==SQLITE_OK && ( + (pCsr->bDesc==0 && pCsr->iPrevId>pCsr->iMaxDocid) + || (pCsr->bDesc!=0 && pCsr->iPrevIdiMinDocid) + )){ + pCsr->isEof = 1; + } + + return rc; +} + +/* +** Restart interation for expression pExpr so that the next call to +** fts3EvalNext() visits the first row. Do not allow incremental +** loading or merging of phrase doclists for this iteration. +** +** If *pRc is other than SQLITE_OK when this function is called, it is +** a no-op. If an error occurs within this function, *pRc is set to an +** SQLite error code before returning. +*/ +static void fts3EvalRestart( + Fts3Cursor *pCsr, + Fts3Expr *pExpr, + int *pRc +){ + if( pExpr && *pRc==SQLITE_OK ){ + Fts3Phrase *pPhrase = pExpr->pPhrase; + + if( pPhrase ){ + fts3EvalInvalidatePoslist(pPhrase); + if( pPhrase->bIncr ){ + int i; + for(i=0; inToken; i++){ + Fts3PhraseToken *pToken = &pPhrase->aToken[i]; + assert( pToken->pDeferred==0 ); + if( pToken->pSegcsr ){ + sqlite3Fts3MsrIncrRestart(pToken->pSegcsr); + } + } + *pRc = fts3EvalPhraseStart(pCsr, 0, pPhrase); + } + pPhrase->doclist.pNextDocid = 0; + pPhrase->doclist.iDocid = 0; + pPhrase->pOrPoslist = 0; + } + + pExpr->iDocid = 0; + pExpr->bEof = 0; + pExpr->bStart = 0; + + fts3EvalRestart(pCsr, pExpr->pLeft, pRc); + fts3EvalRestart(pCsr, pExpr->pRight, pRc); + } +} + +/* +** After allocating the Fts3Expr.aMI[] array for each phrase in the +** expression rooted at pExpr, the cursor iterates through all rows matched +** by pExpr, calling this function for each row. This function increments +** the values in Fts3Expr.aMI[] according to the position-list currently +** found in Fts3Expr.pPhrase->doclist.pList for each of the phrase +** expression nodes. +*/ +static void fts3EvalUpdateCounts(Fts3Expr *pExpr, int nCol){ + if( pExpr ){ + Fts3Phrase *pPhrase = pExpr->pPhrase; + if( pPhrase && pPhrase->doclist.pList ){ + int iCol = 0; + char *p = pPhrase->doclist.pList; + + do{ + u8 c = 0; + int iCnt = 0; + while( 0xFE & (*p | c) ){ + if( (c&0x80)==0 ) iCnt++; + c = *p++ & 0x80; + } + + /* aMI[iCol*3 + 1] = Number of occurrences + ** aMI[iCol*3 + 2] = Number of rows containing at least one instance + */ + pExpr->aMI[iCol*3 + 1] += iCnt; + pExpr->aMI[iCol*3 + 2] += (iCnt>0); + if( *p==0x00 ) break; + p++; + p += fts3GetVarint32(p, &iCol); + }while( iColpLeft, nCol); + fts3EvalUpdateCounts(pExpr->pRight, nCol); + } +} + +/* +** Expression pExpr must be of type FTSQUERY_PHRASE. +** +** If it is not already allocated and populated, this function allocates and +** populates the Fts3Expr.aMI[] array for expression pExpr. If pExpr is part +** of a NEAR expression, then it also allocates and populates the same array +** for all other phrases that are part of the NEAR expression. +** +** SQLITE_OK is returned if the aMI[] array is successfully allocated and +** populated. Otherwise, if an error occurs, an SQLite error code is returned. +*/ +static int fts3EvalGatherStats( + Fts3Cursor *pCsr, /* Cursor object */ + Fts3Expr *pExpr /* FTSQUERY_PHRASE expression */ +){ + int rc = SQLITE_OK; /* Return code */ + + assert( pExpr->eType==FTSQUERY_PHRASE ); + if( pExpr->aMI==0 ){ + Fts3Table *pTab = (Fts3Table *)pCsr->base.pVtab; + Fts3Expr *pRoot; /* Root of NEAR expression */ + Fts3Expr *p; /* Iterator used for several purposes */ + + sqlite3_int64 iPrevId = pCsr->iPrevId; + sqlite3_int64 iDocid; + u8 bEof; + + /* Find the root of the NEAR expression */ + pRoot = pExpr; + while( pRoot->pParent && pRoot->pParent->eType==FTSQUERY_NEAR ){ + pRoot = pRoot->pParent; + } + iDocid = pRoot->iDocid; + bEof = pRoot->bEof; + assert( pRoot->bStart ); + + /* Allocate space for the aMSI[] array of each FTSQUERY_PHRASE node */ + for(p=pRoot; p; p=p->pLeft){ + Fts3Expr *pE = (p->eType==FTSQUERY_PHRASE?p:p->pRight); + assert( pE->aMI==0 ); + pE->aMI = (u32 *)sqlite3_malloc64(pTab->nColumn * 3 * sizeof(u32)); + if( !pE->aMI ) return SQLITE_NOMEM; + memset(pE->aMI, 0, pTab->nColumn * 3 * sizeof(u32)); + } + + fts3EvalRestart(pCsr, pRoot, &rc); + + while( pCsr->isEof==0 && rc==SQLITE_OK ){ + + do { + /* Ensure the %_content statement is reset. */ + if( pCsr->isRequireSeek==0 ) sqlite3_reset(pCsr->pStmt); + assert( sqlite3_data_count(pCsr->pStmt)==0 ); + + /* Advance to the next document */ + fts3EvalNextRow(pCsr, pRoot, &rc); + pCsr->isEof = pRoot->bEof; + pCsr->isRequireSeek = 1; + pCsr->isMatchinfoNeeded = 1; + pCsr->iPrevId = pRoot->iDocid; + }while( pCsr->isEof==0 + && pRoot->eType==FTSQUERY_NEAR + && sqlite3Fts3EvalTestDeferred(pCsr, &rc) + ); + + if( rc==SQLITE_OK && pCsr->isEof==0 ){ + fts3EvalUpdateCounts(pRoot, pTab->nColumn); + } + } + + pCsr->isEof = 0; + pCsr->iPrevId = iPrevId; + + if( bEof ){ + pRoot->bEof = bEof; + }else{ + /* Caution: pRoot may iterate through docids in ascending or descending + ** order. For this reason, even though it seems more defensive, the + ** do loop can not be written: + ** + ** do {...} while( pRoot->iDocidbEof==0 ); + }while( pRoot->iDocid!=iDocid && rc==SQLITE_OK ); + } + } + return rc; +} + +/* +** This function is used by the matchinfo() module to query a phrase +** expression node for the following information: +** +** 1. The total number of occurrences of the phrase in each column of +** the FTS table (considering all rows), and +** +** 2. For each column, the number of rows in the table for which the +** column contains at least one instance of the phrase. +** +** If no error occurs, SQLITE_OK is returned and the values for each column +** written into the array aiOut as follows: +** +** aiOut[iCol*3 + 1] = Number of occurrences +** aiOut[iCol*3 + 2] = Number of rows containing at least one instance +** +** Caveats: +** +** * If a phrase consists entirely of deferred tokens, then all output +** values are set to the number of documents in the table. In other +** words we assume that very common tokens occur exactly once in each +** column of each row of the table. +** +** * If a phrase contains some deferred tokens (and some non-deferred +** tokens), count the potential occurrence identified by considering +** the non-deferred tokens instead of actual phrase occurrences. +** +** * If the phrase is part of a NEAR expression, then only phrase instances +** that meet the NEAR constraint are included in the counts. +*/ +SQLITE_PRIVATE int sqlite3Fts3EvalPhraseStats( + Fts3Cursor *pCsr, /* FTS cursor handle */ + Fts3Expr *pExpr, /* Phrase expression */ + u32 *aiOut /* Array to write results into (see above) */ +){ + Fts3Table *pTab = (Fts3Table *)pCsr->base.pVtab; + int rc = SQLITE_OK; + int iCol; + + if( pExpr->bDeferred && pExpr->pParent->eType!=FTSQUERY_NEAR ){ + assert( pCsr->nDoc>0 ); + for(iCol=0; iColnColumn; iCol++){ + aiOut[iCol*3 + 1] = (u32)pCsr->nDoc; + aiOut[iCol*3 + 2] = (u32)pCsr->nDoc; + } + }else{ + rc = fts3EvalGatherStats(pCsr, pExpr); + if( rc==SQLITE_OK ){ + assert( pExpr->aMI ); + for(iCol=0; iColnColumn; iCol++){ + aiOut[iCol*3 + 1] = pExpr->aMI[iCol*3 + 1]; + aiOut[iCol*3 + 2] = pExpr->aMI[iCol*3 + 2]; + } + } + } + + return rc; +} + +/* +** The expression pExpr passed as the second argument to this function +** must be of type FTSQUERY_PHRASE. +** +** The returned value is either NULL or a pointer to a buffer containing +** a position-list indicating the occurrences of the phrase in column iCol +** of the current row. +** +** More specifically, the returned buffer contains 1 varint for each +** occurrence of the phrase in the column, stored using the normal (delta+2) +** compression and is terminated by either an 0x01 or 0x00 byte. For example, +** if the requested column contains "a b X c d X X" and the position-list +** for 'X' is requested, the buffer returned may contain: +** +** 0x04 0x05 0x03 0x01 or 0x04 0x05 0x03 0x00 +** +** This function works regardless of whether or not the phrase is deferred, +** incremental, or neither. +*/ +SQLITE_PRIVATE int sqlite3Fts3EvalPhrasePoslist( + Fts3Cursor *pCsr, /* FTS3 cursor object */ + Fts3Expr *pExpr, /* Phrase to return doclist for */ + int iCol, /* Column to return position list for */ + char **ppOut /* OUT: Pointer to position list */ +){ + Fts3Phrase *pPhrase = pExpr->pPhrase; + Fts3Table *pTab = (Fts3Table *)pCsr->base.pVtab; + char *pIter; + int iThis; + sqlite3_int64 iDocid; + + /* If this phrase is applies specifically to some column other than + ** column iCol, return a NULL pointer. */ + *ppOut = 0; + assert( iCol>=0 && iColnColumn ); + if( (pPhrase->iColumnnColumn && pPhrase->iColumn!=iCol) ){ + return SQLITE_OK; + } + + iDocid = pExpr->iDocid; + pIter = pPhrase->doclist.pList; + if( iDocid!=pCsr->iPrevId || pExpr->bEof ){ + int rc = SQLITE_OK; + int bDescDoclist = pTab->bDescIdx; /* For DOCID_CMP macro */ + int bOr = 0; + u8 bTreeEof = 0; + Fts3Expr *p; /* Used to iterate from pExpr to root */ + Fts3Expr *pNear; /* Most senior NEAR ancestor (or pExpr) */ + int bMatch; + + /* Check if this phrase descends from an OR expression node. If not, + ** return NULL. Otherwise, the entry that corresponds to docid + ** pCsr->iPrevId may lie earlier in the doclist buffer. Or, if the + ** tree that the node is part of has been marked as EOF, but the node + ** itself is not EOF, then it may point to an earlier entry. */ + pNear = pExpr; + for(p=pExpr->pParent; p; p=p->pParent){ + if( p->eType==FTSQUERY_OR ) bOr = 1; + if( p->eType==FTSQUERY_NEAR ) pNear = p; + if( p->bEof ) bTreeEof = 1; + } + if( bOr==0 ) return SQLITE_OK; + + /* This is the descendent of an OR node. In this case we cannot use + ** an incremental phrase. Load the entire doclist for the phrase + ** into memory in this case. */ + if( pPhrase->bIncr ){ + int bEofSave = pNear->bEof; + fts3EvalRestart(pCsr, pNear, &rc); + while( rc==SQLITE_OK && !pNear->bEof ){ + fts3EvalNextRow(pCsr, pNear, &rc); + if( bEofSave==0 && pNear->iDocid==iDocid ) break; + } + assert( rc!=SQLITE_OK || pPhrase->bIncr==0 ); + } + if( bTreeEof ){ + while( rc==SQLITE_OK && !pNear->bEof ){ + fts3EvalNextRow(pCsr, pNear, &rc); + } + } + if( rc!=SQLITE_OK ) return rc; + + bMatch = 1; + for(p=pNear; p; p=p->pLeft){ + u8 bEof = 0; + Fts3Expr *pTest = p; + Fts3Phrase *pPh; + assert( pTest->eType==FTSQUERY_NEAR || pTest->eType==FTSQUERY_PHRASE ); + if( pTest->eType==FTSQUERY_NEAR ) pTest = pTest->pRight; + assert( pTest->eType==FTSQUERY_PHRASE ); + pPh = pTest->pPhrase; + + pIter = pPh->pOrPoslist; + iDocid = pPh->iOrDocid; + if( pCsr->bDesc==bDescDoclist ){ + bEof = !pPh->doclist.nAll || + (pIter >= (pPh->doclist.aAll + pPh->doclist.nAll)); + while( (pIter==0 || DOCID_CMP(iDocid, pCsr->iPrevId)<0 ) && bEof==0 ){ + sqlite3Fts3DoclistNext( + bDescDoclist, pPh->doclist.aAll, pPh->doclist.nAll, + &pIter, &iDocid, &bEof + ); + } + }else{ + bEof = !pPh->doclist.nAll || (pIter && pIter<=pPh->doclist.aAll); + while( (pIter==0 || DOCID_CMP(iDocid, pCsr->iPrevId)>0 ) && bEof==0 ){ + int dummy; + sqlite3Fts3DoclistPrev( + bDescDoclist, pPh->doclist.aAll, pPh->doclist.nAll, + &pIter, &iDocid, &dummy, &bEof + ); + } + } + pPh->pOrPoslist = pIter; + pPh->iOrDocid = iDocid; + if( bEof || iDocid!=pCsr->iPrevId ) bMatch = 0; + } + + if( bMatch ){ + pIter = pPhrase->pOrPoslist; + }else{ + pIter = 0; + } + } + if( pIter==0 ) return SQLITE_OK; + + if( *pIter==0x01 ){ + pIter++; + pIter += fts3GetVarint32(pIter, &iThis); + }else{ + iThis = 0; + } + while( iThisdoclist, and +** * any Fts3MultiSegReader objects held by phrase tokens. +*/ +SQLITE_PRIVATE void sqlite3Fts3EvalPhraseCleanup(Fts3Phrase *pPhrase){ + if( pPhrase ){ + int i; + sqlite3_free(pPhrase->doclist.aAll); + fts3EvalInvalidatePoslist(pPhrase); + memset(&pPhrase->doclist, 0, sizeof(Fts3Doclist)); + for(i=0; inToken; i++){ + fts3SegReaderCursorFree(pPhrase->aToken[i].pSegcsr); + pPhrase->aToken[i].pSegcsr = 0; + } + } +} + + +/* +** Return SQLITE_CORRUPT_VTAB. +*/ +#ifdef SQLITE_DEBUG +SQLITE_PRIVATE int sqlite3Fts3Corrupt(){ + return SQLITE_CORRUPT_VTAB; +} +#endif + +#if !SQLITE_CORE +/* +** Initialize API pointer table, if required. +*/ +#ifdef _WIN32 +__declspec(dllexport) +#endif +SQLITE_API int sqlite3_fts3_init( + sqlite3 *db, + char **pzErrMsg, + const sqlite3_api_routines *pApi +){ + SQLITE_EXTENSION_INIT2(pApi) + return sqlite3Fts3Init(db); +} +#endif + +#endif + +/************** End of fts3.c ************************************************/ +/************** Begin file fts3_aux.c ****************************************/ +/* +** 2011 Jan 27 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +****************************************************************************** +** +*/ +/* #include "fts3Int.h" */ +#if !defined(SQLITE_CORE) || defined(SQLITE_ENABLE_FTS3) + +/* #include */ +/* #include */ + +typedef struct Fts3auxTable Fts3auxTable; +typedef struct Fts3auxCursor Fts3auxCursor; + +struct Fts3auxTable { + sqlite3_vtab base; /* Base class used by SQLite core */ + Fts3Table *pFts3Tab; +}; + +struct Fts3auxCursor { + sqlite3_vtab_cursor base; /* Base class used by SQLite core */ + Fts3MultiSegReader csr; /* Must be right after "base" */ + Fts3SegFilter filter; + char *zStop; + int nStop; /* Byte-length of string zStop */ + int iLangid; /* Language id to query */ + int isEof; /* True if cursor is at EOF */ + sqlite3_int64 iRowid; /* Current rowid */ + + int iCol; /* Current value of 'col' column */ + int nStat; /* Size of aStat[] array */ + struct Fts3auxColstats { + sqlite3_int64 nDoc; /* 'documents' values for current csr row */ + sqlite3_int64 nOcc; /* 'occurrences' values for current csr row */ + } *aStat; +}; + +/* +** Schema of the terms table. +*/ +#define FTS3_AUX_SCHEMA \ + "CREATE TABLE x(term, col, documents, occurrences, languageid HIDDEN)" + +/* +** This function does all the work for both the xConnect and xCreate methods. +** These tables have no persistent representation of their own, so xConnect +** and xCreate are identical operations. +*/ +static int fts3auxConnectMethod( + sqlite3 *db, /* Database connection */ + void *pUnused, /* Unused */ + int argc, /* Number of elements in argv array */ + const char * const *argv, /* xCreate/xConnect argument array */ + sqlite3_vtab **ppVtab, /* OUT: New sqlite3_vtab object */ + char **pzErr /* OUT: sqlite3_malloc'd error message */ +){ + char const *zDb; /* Name of database (e.g. "main") */ + char const *zFts3; /* Name of fts3 table */ + int nDb; /* Result of strlen(zDb) */ + int nFts3; /* Result of strlen(zFts3) */ + sqlite3_int64 nByte; /* Bytes of space to allocate here */ + int rc; /* value returned by declare_vtab() */ + Fts3auxTable *p; /* Virtual table object to return */ + + UNUSED_PARAMETER(pUnused); + + /* The user should invoke this in one of two forms: + ** + ** CREATE VIRTUAL TABLE xxx USING fts4aux(fts4-table); + ** CREATE VIRTUAL TABLE xxx USING fts4aux(fts4-table-db, fts4-table); + */ + if( argc!=4 && argc!=5 ) goto bad_args; + + zDb = argv[1]; + nDb = (int)strlen(zDb); + if( argc==5 ){ + if( nDb==4 && 0==sqlite3_strnicmp("temp", zDb, 4) ){ + zDb = argv[3]; + nDb = (int)strlen(zDb); + zFts3 = argv[4]; + }else{ + goto bad_args; + } + }else{ + zFts3 = argv[3]; + } + nFts3 = (int)strlen(zFts3); + + rc = sqlite3_declare_vtab(db, FTS3_AUX_SCHEMA); + if( rc!=SQLITE_OK ) return rc; + + nByte = sizeof(Fts3auxTable) + sizeof(Fts3Table) + nDb + nFts3 + 2; + p = (Fts3auxTable *)sqlite3_malloc64(nByte); + if( !p ) return SQLITE_NOMEM; + memset(p, 0, nByte); + + p->pFts3Tab = (Fts3Table *)&p[1]; + p->pFts3Tab->zDb = (char *)&p->pFts3Tab[1]; + p->pFts3Tab->zName = &p->pFts3Tab->zDb[nDb+1]; + p->pFts3Tab->db = db; + p->pFts3Tab->nIndex = 1; + + memcpy((char *)p->pFts3Tab->zDb, zDb, nDb); + memcpy((char *)p->pFts3Tab->zName, zFts3, nFts3); + sqlite3Fts3Dequote((char *)p->pFts3Tab->zName); + + *ppVtab = (sqlite3_vtab *)p; + return SQLITE_OK; + + bad_args: + sqlite3Fts3ErrMsg(pzErr, "invalid arguments to fts4aux constructor"); + return SQLITE_ERROR; +} + +/* +** This function does the work for both the xDisconnect and xDestroy methods. +** These tables have no persistent representation of their own, so xDisconnect +** and xDestroy are identical operations. +*/ +static int fts3auxDisconnectMethod(sqlite3_vtab *pVtab){ + Fts3auxTable *p = (Fts3auxTable *)pVtab; + Fts3Table *pFts3 = p->pFts3Tab; + int i; + + /* Free any prepared statements held */ + for(i=0; iaStmt); i++){ + sqlite3_finalize(pFts3->aStmt[i]); + } + sqlite3_free(pFts3->zSegmentsTbl); + sqlite3_free(p); + return SQLITE_OK; +} + +#define FTS4AUX_EQ_CONSTRAINT 1 +#define FTS4AUX_GE_CONSTRAINT 2 +#define FTS4AUX_LE_CONSTRAINT 4 + +/* +** xBestIndex - Analyze a WHERE and ORDER BY clause. +*/ +static int fts3auxBestIndexMethod( + sqlite3_vtab *pVTab, + sqlite3_index_info *pInfo +){ + int i; + int iEq = -1; + int iGe = -1; + int iLe = -1; + int iLangid = -1; + int iNext = 1; /* Next free argvIndex value */ + + UNUSED_PARAMETER(pVTab); + + /* This vtab delivers always results in "ORDER BY term ASC" order. */ + if( pInfo->nOrderBy==1 + && pInfo->aOrderBy[0].iColumn==0 + && pInfo->aOrderBy[0].desc==0 + ){ + pInfo->orderByConsumed = 1; + } + + /* Search for equality and range constraints on the "term" column. + ** And equality constraints on the hidden "languageid" column. */ + for(i=0; inConstraint; i++){ + if( pInfo->aConstraint[i].usable ){ + int op = pInfo->aConstraint[i].op; + int iCol = pInfo->aConstraint[i].iColumn; + + if( iCol==0 ){ + if( op==SQLITE_INDEX_CONSTRAINT_EQ ) iEq = i; + if( op==SQLITE_INDEX_CONSTRAINT_LT ) iLe = i; + if( op==SQLITE_INDEX_CONSTRAINT_LE ) iLe = i; + if( op==SQLITE_INDEX_CONSTRAINT_GT ) iGe = i; + if( op==SQLITE_INDEX_CONSTRAINT_GE ) iGe = i; + } + if( iCol==4 ){ + if( op==SQLITE_INDEX_CONSTRAINT_EQ ) iLangid = i; + } + } + } + + if( iEq>=0 ){ + pInfo->idxNum = FTS4AUX_EQ_CONSTRAINT; + pInfo->aConstraintUsage[iEq].argvIndex = iNext++; + pInfo->estimatedCost = 5; + }else{ + pInfo->idxNum = 0; + pInfo->estimatedCost = 20000; + if( iGe>=0 ){ + pInfo->idxNum += FTS4AUX_GE_CONSTRAINT; + pInfo->aConstraintUsage[iGe].argvIndex = iNext++; + pInfo->estimatedCost /= 2; + } + if( iLe>=0 ){ + pInfo->idxNum += FTS4AUX_LE_CONSTRAINT; + pInfo->aConstraintUsage[iLe].argvIndex = iNext++; + pInfo->estimatedCost /= 2; + } + } + if( iLangid>=0 ){ + pInfo->aConstraintUsage[iLangid].argvIndex = iNext++; + pInfo->estimatedCost--; + } + + return SQLITE_OK; +} + +/* +** xOpen - Open a cursor. +*/ +static int fts3auxOpenMethod(sqlite3_vtab *pVTab, sqlite3_vtab_cursor **ppCsr){ + Fts3auxCursor *pCsr; /* Pointer to cursor object to return */ + + UNUSED_PARAMETER(pVTab); + + pCsr = (Fts3auxCursor *)sqlite3_malloc(sizeof(Fts3auxCursor)); + if( !pCsr ) return SQLITE_NOMEM; + memset(pCsr, 0, sizeof(Fts3auxCursor)); + + *ppCsr = (sqlite3_vtab_cursor *)pCsr; + return SQLITE_OK; +} + +/* +** xClose - Close a cursor. +*/ +static int fts3auxCloseMethod(sqlite3_vtab_cursor *pCursor){ + Fts3Table *pFts3 = ((Fts3auxTable *)pCursor->pVtab)->pFts3Tab; + Fts3auxCursor *pCsr = (Fts3auxCursor *)pCursor; + + sqlite3Fts3SegmentsClose(pFts3); + sqlite3Fts3SegReaderFinish(&pCsr->csr); + sqlite3_free((void *)pCsr->filter.zTerm); + sqlite3_free(pCsr->zStop); + sqlite3_free(pCsr->aStat); + sqlite3_free(pCsr); + return SQLITE_OK; +} + +static int fts3auxGrowStatArray(Fts3auxCursor *pCsr, int nSize){ + if( nSize>pCsr->nStat ){ + struct Fts3auxColstats *aNew; + aNew = (struct Fts3auxColstats *)sqlite3_realloc64(pCsr->aStat, + sizeof(struct Fts3auxColstats) * nSize + ); + if( aNew==0 ) return SQLITE_NOMEM; + memset(&aNew[pCsr->nStat], 0, + sizeof(struct Fts3auxColstats) * (nSize - pCsr->nStat) + ); + pCsr->aStat = aNew; + pCsr->nStat = nSize; + } + return SQLITE_OK; +} + +/* +** xNext - Advance the cursor to the next row, if any. +*/ +static int fts3auxNextMethod(sqlite3_vtab_cursor *pCursor){ + Fts3auxCursor *pCsr = (Fts3auxCursor *)pCursor; + Fts3Table *pFts3 = ((Fts3auxTable *)pCursor->pVtab)->pFts3Tab; + int rc; + + /* Increment our pretend rowid value. */ + pCsr->iRowid++; + + for(pCsr->iCol++; pCsr->iColnStat; pCsr->iCol++){ + if( pCsr->aStat[pCsr->iCol].nDoc>0 ) return SQLITE_OK; + } + + rc = sqlite3Fts3SegReaderStep(pFts3, &pCsr->csr); + if( rc==SQLITE_ROW ){ + int i = 0; + int nDoclist = pCsr->csr.nDoclist; + char *aDoclist = pCsr->csr.aDoclist; + int iCol; + + int eState = 0; + + if( pCsr->zStop ){ + int n = (pCsr->nStopcsr.nTerm) ? pCsr->nStop : pCsr->csr.nTerm; + int mc = memcmp(pCsr->zStop, pCsr->csr.zTerm, n); + if( mc<0 || (mc==0 && pCsr->csr.nTerm>pCsr->nStop) ){ + pCsr->isEof = 1; + return SQLITE_OK; + } + } + + if( fts3auxGrowStatArray(pCsr, 2) ) return SQLITE_NOMEM; + memset(pCsr->aStat, 0, sizeof(struct Fts3auxColstats) * pCsr->nStat); + iCol = 0; + + while( iaStat[0].nDoc++; + eState = 1; + iCol = 0; + break; + + /* State 1. In this state we are expecting either a 1, indicating + ** that the following integer will be a column number, or the + ** start of a position list for column 0. + ** + ** The only difference between state 1 and state 2 is that if the + ** integer encountered in state 1 is not 0 or 1, then we need to + ** increment the column 0 "nDoc" count for this term. + */ + case 1: + assert( iCol==0 ); + if( v>1 ){ + pCsr->aStat[1].nDoc++; + } + eState = 2; + /* fall through */ + + case 2: + if( v==0 ){ /* 0x00. Next integer will be a docid. */ + eState = 0; + }else if( v==1 ){ /* 0x01. Next integer will be a column number. */ + eState = 3; + }else{ /* 2 or greater. A position. */ + pCsr->aStat[iCol+1].nOcc++; + pCsr->aStat[0].nOcc++; + } + break; + + /* State 3. The integer just read is a column number. */ + default: assert( eState==3 ); + iCol = (int)v; + if( fts3auxGrowStatArray(pCsr, iCol+2) ) return SQLITE_NOMEM; + pCsr->aStat[iCol+1].nDoc++; + eState = 2; + break; + } + } + + pCsr->iCol = 0; + rc = SQLITE_OK; + }else{ + pCsr->isEof = 1; + } + return rc; +} + +/* +** xFilter - Initialize a cursor to point at the start of its data. +*/ +static int fts3auxFilterMethod( + sqlite3_vtab_cursor *pCursor, /* The cursor used for this query */ + int idxNum, /* Strategy index */ + const char *idxStr, /* Unused */ + int nVal, /* Number of elements in apVal */ + sqlite3_value **apVal /* Arguments for the indexing scheme */ +){ + Fts3auxCursor *pCsr = (Fts3auxCursor *)pCursor; + Fts3Table *pFts3 = ((Fts3auxTable *)pCursor->pVtab)->pFts3Tab; + int rc; + int isScan = 0; + int iLangVal = 0; /* Language id to query */ + + int iEq = -1; /* Index of term=? value in apVal */ + int iGe = -1; /* Index of term>=? value in apVal */ + int iLe = -1; /* Index of term<=? value in apVal */ + int iLangid = -1; /* Index of languageid=? value in apVal */ + int iNext = 0; + + UNUSED_PARAMETER(nVal); + UNUSED_PARAMETER(idxStr); + + assert( idxStr==0 ); + assert( idxNum==FTS4AUX_EQ_CONSTRAINT || idxNum==0 + || idxNum==FTS4AUX_LE_CONSTRAINT || idxNum==FTS4AUX_GE_CONSTRAINT + || idxNum==(FTS4AUX_LE_CONSTRAINT|FTS4AUX_GE_CONSTRAINT) + ); + + if( idxNum==FTS4AUX_EQ_CONSTRAINT ){ + iEq = iNext++; + }else{ + isScan = 1; + if( idxNum & FTS4AUX_GE_CONSTRAINT ){ + iGe = iNext++; + } + if( idxNum & FTS4AUX_LE_CONSTRAINT ){ + iLe = iNext++; + } + } + if( iNextfilter.zTerm); + sqlite3Fts3SegReaderFinish(&pCsr->csr); + sqlite3_free((void *)pCsr->filter.zTerm); + sqlite3_free(pCsr->aStat); + memset(&pCsr->csr, 0, ((u8*)&pCsr[1]) - (u8*)&pCsr->csr); + + pCsr->filter.flags = FTS3_SEGMENT_REQUIRE_POS|FTS3_SEGMENT_IGNORE_EMPTY; + if( isScan ) pCsr->filter.flags |= FTS3_SEGMENT_SCAN; + + if( iEq>=0 || iGe>=0 ){ + const unsigned char *zStr = sqlite3_value_text(apVal[0]); + assert( (iEq==0 && iGe==-1) || (iEq==-1 && iGe==0) ); + if( zStr ){ + pCsr->filter.zTerm = sqlite3_mprintf("%s", zStr); + if( pCsr->filter.zTerm==0 ) return SQLITE_NOMEM; + pCsr->filter.nTerm = (int)strlen(pCsr->filter.zTerm); + } + } + + if( iLe>=0 ){ + pCsr->zStop = sqlite3_mprintf("%s", sqlite3_value_text(apVal[iLe])); + if( pCsr->zStop==0 ) return SQLITE_NOMEM; + pCsr->nStop = (int)strlen(pCsr->zStop); + } + + if( iLangid>=0 ){ + iLangVal = sqlite3_value_int(apVal[iLangid]); + + /* If the user specified a negative value for the languageid, use zero + ** instead. This works, as the "languageid=?" constraint will also + ** be tested by the VDBE layer. The test will always be false (since + ** this module will not return a row with a negative languageid), and + ** so the overall query will return zero rows. */ + if( iLangVal<0 ) iLangVal = 0; + } + pCsr->iLangid = iLangVal; + + rc = sqlite3Fts3SegReaderCursor(pFts3, iLangVal, 0, FTS3_SEGCURSOR_ALL, + pCsr->filter.zTerm, pCsr->filter.nTerm, 0, isScan, &pCsr->csr + ); + if( rc==SQLITE_OK ){ + rc = sqlite3Fts3SegReaderStart(pFts3, &pCsr->csr, &pCsr->filter); + } + + if( rc==SQLITE_OK ) rc = fts3auxNextMethod(pCursor); + return rc; +} + +/* +** xEof - Return true if the cursor is at EOF, or false otherwise. +*/ +static int fts3auxEofMethod(sqlite3_vtab_cursor *pCursor){ + Fts3auxCursor *pCsr = (Fts3auxCursor *)pCursor; + return pCsr->isEof; +} + +/* +** xColumn - Return a column value. +*/ +static int fts3auxColumnMethod( + sqlite3_vtab_cursor *pCursor, /* Cursor to retrieve value from */ + sqlite3_context *pCtx, /* Context for sqlite3_result_xxx() calls */ + int iCol /* Index of column to read value from */ +){ + Fts3auxCursor *p = (Fts3auxCursor *)pCursor; + + assert( p->isEof==0 ); + switch( iCol ){ + case 0: /* term */ + sqlite3_result_text(pCtx, p->csr.zTerm, p->csr.nTerm, SQLITE_TRANSIENT); + break; + + case 1: /* col */ + if( p->iCol ){ + sqlite3_result_int(pCtx, p->iCol-1); + }else{ + sqlite3_result_text(pCtx, "*", -1, SQLITE_STATIC); + } + break; + + case 2: /* documents */ + sqlite3_result_int64(pCtx, p->aStat[p->iCol].nDoc); + break; + + case 3: /* occurrences */ + sqlite3_result_int64(pCtx, p->aStat[p->iCol].nOcc); + break; + + default: /* languageid */ + assert( iCol==4 ); + sqlite3_result_int(pCtx, p->iLangid); + break; + } + + return SQLITE_OK; +} + +/* +** xRowid - Return the current rowid for the cursor. +*/ +static int fts3auxRowidMethod( + sqlite3_vtab_cursor *pCursor, /* Cursor to retrieve value from */ + sqlite_int64 *pRowid /* OUT: Rowid value */ +){ + Fts3auxCursor *pCsr = (Fts3auxCursor *)pCursor; + *pRowid = pCsr->iRowid; + return SQLITE_OK; +} + +/* +** Register the fts3aux module with database connection db. Return SQLITE_OK +** if successful or an error code if sqlite3_create_module() fails. +*/ +SQLITE_PRIVATE int sqlite3Fts3InitAux(sqlite3 *db){ + static const sqlite3_module fts3aux_module = { + 0, /* iVersion */ + fts3auxConnectMethod, /* xCreate */ + fts3auxConnectMethod, /* xConnect */ + fts3auxBestIndexMethod, /* xBestIndex */ + fts3auxDisconnectMethod, /* xDisconnect */ + fts3auxDisconnectMethod, /* xDestroy */ + fts3auxOpenMethod, /* xOpen */ + fts3auxCloseMethod, /* xClose */ + fts3auxFilterMethod, /* xFilter */ + fts3auxNextMethod, /* xNext */ + fts3auxEofMethod, /* xEof */ + fts3auxColumnMethod, /* xColumn */ + fts3auxRowidMethod, /* xRowid */ + 0, /* xUpdate */ + 0, /* xBegin */ + 0, /* xSync */ + 0, /* xCommit */ + 0, /* xRollback */ + 0, /* xFindFunction */ + 0, /* xRename */ + 0, /* xSavepoint */ + 0, /* xRelease */ + 0, /* xRollbackTo */ + 0 /* xShadowName */ + }; + int rc; /* Return code */ + + rc = sqlite3_create_module(db, "fts4aux", &fts3aux_module, 0); + return rc; +} + +#endif /* !defined(SQLITE_CORE) || defined(SQLITE_ENABLE_FTS3) */ + +/************** End of fts3_aux.c ********************************************/ +/************** Begin file fts3_expr.c ***************************************/ +/* +** 2008 Nov 28 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +****************************************************************************** +** +** This module contains code that implements a parser for fts3 query strings +** (the right-hand argument to the MATCH operator). Because the supported +** syntax is relatively simple, the whole tokenizer/parser system is +** hand-coded. +*/ +/* #include "fts3Int.h" */ +#if !defined(SQLITE_CORE) || defined(SQLITE_ENABLE_FTS3) + +/* +** By default, this module parses the legacy syntax that has been +** traditionally used by fts3. Or, if SQLITE_ENABLE_FTS3_PARENTHESIS +** is defined, then it uses the new syntax. The differences between +** the new and the old syntaxes are: +** +** a) The new syntax supports parenthesis. The old does not. +** +** b) The new syntax supports the AND and NOT operators. The old does not. +** +** c) The old syntax supports the "-" token qualifier. This is not +** supported by the new syntax (it is replaced by the NOT operator). +** +** d) When using the old syntax, the OR operator has a greater precedence +** than an implicit AND. When using the new, both implicity and explicit +** AND operators have a higher precedence than OR. +** +** If compiled with SQLITE_TEST defined, then this module exports the +** symbol "int sqlite3_fts3_enable_parentheses". Setting this variable +** to zero causes the module to use the old syntax. If it is set to +** non-zero the new syntax is activated. This is so both syntaxes can +** be tested using a single build of testfixture. +** +** The following describes the syntax supported by the fts3 MATCH +** operator in a similar format to that used by the lemon parser +** generator. This module does not use actually lemon, it uses a +** custom parser. +** +** query ::= andexpr (OR andexpr)*. +** +** andexpr ::= notexpr (AND? notexpr)*. +** +** notexpr ::= nearexpr (NOT nearexpr|-TOKEN)*. +** notexpr ::= LP query RP. +** +** nearexpr ::= phrase (NEAR distance_opt nearexpr)*. +** +** distance_opt ::= . +** distance_opt ::= / INTEGER. +** +** phrase ::= TOKEN. +** phrase ::= COLUMN:TOKEN. +** phrase ::= "TOKEN TOKEN TOKEN...". +*/ + +#ifdef SQLITE_TEST +SQLITE_API int sqlite3_fts3_enable_parentheses = 0; +#else +# ifdef SQLITE_ENABLE_FTS3_PARENTHESIS +# define sqlite3_fts3_enable_parentheses 1 +# else +# define sqlite3_fts3_enable_parentheses 0 +# endif +#endif + +/* +** Default span for NEAR operators. +*/ +#define SQLITE_FTS3_DEFAULT_NEAR_PARAM 10 + +/* #include */ +/* #include */ + +/* +** isNot: +** This variable is used by function getNextNode(). When getNextNode() is +** called, it sets ParseContext.isNot to true if the 'next node' is a +** FTSQUERY_PHRASE with a unary "-" attached to it. i.e. "mysql" in the +** FTS3 query "sqlite -mysql". Otherwise, ParseContext.isNot is set to +** zero. +*/ +typedef struct ParseContext ParseContext; +struct ParseContext { + sqlite3_tokenizer *pTokenizer; /* Tokenizer module */ + int iLangid; /* Language id used with tokenizer */ + const char **azCol; /* Array of column names for fts3 table */ + int bFts4; /* True to allow FTS4-only syntax */ + int nCol; /* Number of entries in azCol[] */ + int iDefaultCol; /* Default column to query */ + int isNot; /* True if getNextNode() sees a unary - */ + sqlite3_context *pCtx; /* Write error message here */ + int nNest; /* Number of nested brackets */ +}; + +/* +** This function is equivalent to the standard isspace() function. +** +** The standard isspace() can be awkward to use safely, because although it +** is defined to accept an argument of type int, its behavior when passed +** an integer that falls outside of the range of the unsigned char type +** is undefined (and sometimes, "undefined" means segfault). This wrapper +** is defined to accept an argument of type char, and always returns 0 for +** any values that fall outside of the range of the unsigned char type (i.e. +** negative values). +*/ +static int fts3isspace(char c){ + return c==' ' || c=='\t' || c=='\n' || c=='\r' || c=='\v' || c=='\f'; +} + +/* +** Allocate nByte bytes of memory using sqlite3_malloc(). If successful, +** zero the memory before returning a pointer to it. If unsuccessful, +** return NULL. +*/ +static void *fts3MallocZero(sqlite3_int64 nByte){ + void *pRet = sqlite3_malloc64(nByte); + if( pRet ) memset(pRet, 0, nByte); + return pRet; +} + +SQLITE_PRIVATE int sqlite3Fts3OpenTokenizer( + sqlite3_tokenizer *pTokenizer, + int iLangid, + const char *z, + int n, + sqlite3_tokenizer_cursor **ppCsr +){ + sqlite3_tokenizer_module const *pModule = pTokenizer->pModule; + sqlite3_tokenizer_cursor *pCsr = 0; + int rc; + + rc = pModule->xOpen(pTokenizer, z, n, &pCsr); + assert( rc==SQLITE_OK || pCsr==0 ); + if( rc==SQLITE_OK ){ + pCsr->pTokenizer = pTokenizer; + if( pModule->iVersion>=1 ){ + rc = pModule->xLanguageid(pCsr, iLangid); + if( rc!=SQLITE_OK ){ + pModule->xClose(pCsr); + pCsr = 0; + } + } + } + *ppCsr = pCsr; + return rc; +} + +/* +** Function getNextNode(), which is called by fts3ExprParse(), may itself +** call fts3ExprParse(). So this forward declaration is required. +*/ +static int fts3ExprParse(ParseContext *, const char *, int, Fts3Expr **, int *); + +/* +** Extract the next token from buffer z (length n) using the tokenizer +** and other information (column names etc.) in pParse. Create an Fts3Expr +** structure of type FTSQUERY_PHRASE containing a phrase consisting of this +** single token and set *ppExpr to point to it. If the end of the buffer is +** reached before a token is found, set *ppExpr to zero. It is the +** responsibility of the caller to eventually deallocate the allocated +** Fts3Expr structure (if any) by passing it to sqlite3_free(). +** +** Return SQLITE_OK if successful, or SQLITE_NOMEM if a memory allocation +** fails. +*/ +static int getNextToken( + ParseContext *pParse, /* fts3 query parse context */ + int iCol, /* Value for Fts3Phrase.iColumn */ + const char *z, int n, /* Input string */ + Fts3Expr **ppExpr, /* OUT: expression */ + int *pnConsumed /* OUT: Number of bytes consumed */ +){ + sqlite3_tokenizer *pTokenizer = pParse->pTokenizer; + sqlite3_tokenizer_module const *pModule = pTokenizer->pModule; + int rc; + sqlite3_tokenizer_cursor *pCursor; + Fts3Expr *pRet = 0; + int i = 0; + + /* Set variable i to the maximum number of bytes of input to tokenize. */ + for(i=0; iiLangid, z, i, &pCursor); + if( rc==SQLITE_OK ){ + const char *zToken; + int nToken = 0, iStart = 0, iEnd = 0, iPosition = 0; + sqlite3_int64 nByte; /* total space to allocate */ + + rc = pModule->xNext(pCursor, &zToken, &nToken, &iStart, &iEnd, &iPosition); + if( rc==SQLITE_OK ){ + nByte = sizeof(Fts3Expr) + sizeof(Fts3Phrase) + nToken; + pRet = (Fts3Expr *)fts3MallocZero(nByte); + if( !pRet ){ + rc = SQLITE_NOMEM; + }else{ + pRet->eType = FTSQUERY_PHRASE; + pRet->pPhrase = (Fts3Phrase *)&pRet[1]; + pRet->pPhrase->nToken = 1; + pRet->pPhrase->iColumn = iCol; + pRet->pPhrase->aToken[0].n = nToken; + pRet->pPhrase->aToken[0].z = (char *)&pRet->pPhrase[1]; + memcpy(pRet->pPhrase->aToken[0].z, zToken, nToken); + + if( iEndpPhrase->aToken[0].isPrefix = 1; + iEnd++; + } + + while( 1 ){ + if( !sqlite3_fts3_enable_parentheses + && iStart>0 && z[iStart-1]=='-' + ){ + pParse->isNot = 1; + iStart--; + }else if( pParse->bFts4 && iStart>0 && z[iStart-1]=='^' ){ + pRet->pPhrase->aToken[0].bFirst = 1; + iStart--; + }else{ + break; + } + } + + } + *pnConsumed = iEnd; + }else if( i && rc==SQLITE_DONE ){ + rc = SQLITE_OK; + } + + pModule->xClose(pCursor); + } + + *ppExpr = pRet; + return rc; +} + + +/* +** Enlarge a memory allocation. If an out-of-memory allocation occurs, +** then free the old allocation. +*/ +static void *fts3ReallocOrFree(void *pOrig, sqlite3_int64 nNew){ + void *pRet = sqlite3_realloc64(pOrig, nNew); + if( !pRet ){ + sqlite3_free(pOrig); + } + return pRet; +} + +/* +** Buffer zInput, length nInput, contains the contents of a quoted string +** that appeared as part of an fts3 query expression. Neither quote character +** is included in the buffer. This function attempts to tokenize the entire +** input buffer and create an Fts3Expr structure of type FTSQUERY_PHRASE +** containing the results. +** +** If successful, SQLITE_OK is returned and *ppExpr set to point at the +** allocated Fts3Expr structure. Otherwise, either SQLITE_NOMEM (out of memory +** error) or SQLITE_ERROR (tokenization error) is returned and *ppExpr set +** to 0. +*/ +static int getNextString( + ParseContext *pParse, /* fts3 query parse context */ + const char *zInput, int nInput, /* Input string */ + Fts3Expr **ppExpr /* OUT: expression */ +){ + sqlite3_tokenizer *pTokenizer = pParse->pTokenizer; + sqlite3_tokenizer_module const *pModule = pTokenizer->pModule; + int rc; + Fts3Expr *p = 0; + sqlite3_tokenizer_cursor *pCursor = 0; + char *zTemp = 0; + int nTemp = 0; + + const int nSpace = sizeof(Fts3Expr) + sizeof(Fts3Phrase); + int nToken = 0; + + /* The final Fts3Expr data structure, including the Fts3Phrase, + ** Fts3PhraseToken structures token buffers are all stored as a single + ** allocation so that the expression can be freed with a single call to + ** sqlite3_free(). Setting this up requires a two pass approach. + ** + ** The first pass, in the block below, uses a tokenizer cursor to iterate + ** through the tokens in the expression. This pass uses fts3ReallocOrFree() + ** to assemble data in two dynamic buffers: + ** + ** Buffer p: Points to the Fts3Expr structure, followed by the Fts3Phrase + ** structure, followed by the array of Fts3PhraseToken + ** structures. This pass only populates the Fts3PhraseToken array. + ** + ** Buffer zTemp: Contains copies of all tokens. + ** + ** The second pass, in the block that begins "if( rc==SQLITE_DONE )" below, + ** appends buffer zTemp to buffer p, and fills in the Fts3Expr and Fts3Phrase + ** structures. + */ + rc = sqlite3Fts3OpenTokenizer( + pTokenizer, pParse->iLangid, zInput, nInput, &pCursor); + if( rc==SQLITE_OK ){ + int ii; + for(ii=0; rc==SQLITE_OK; ii++){ + const char *zByte; + int nByte = 0, iBegin = 0, iEnd = 0, iPos = 0; + rc = pModule->xNext(pCursor, &zByte, &nByte, &iBegin, &iEnd, &iPos); + if( rc==SQLITE_OK ){ + Fts3PhraseToken *pToken; + + p = fts3ReallocOrFree(p, nSpace + ii*sizeof(Fts3PhraseToken)); + if( !p ) goto no_mem; + + zTemp = fts3ReallocOrFree(zTemp, nTemp + nByte); + if( !zTemp ) goto no_mem; + + assert( nToken==ii ); + pToken = &((Fts3Phrase *)(&p[1]))->aToken[ii]; + memset(pToken, 0, sizeof(Fts3PhraseToken)); + + memcpy(&zTemp[nTemp], zByte, nByte); + nTemp += nByte; + + pToken->n = nByte; + pToken->isPrefix = (iEndbFirst = (iBegin>0 && zInput[iBegin-1]=='^'); + nToken = ii+1; + } + } + + pModule->xClose(pCursor); + pCursor = 0; + } + + if( rc==SQLITE_DONE ){ + int jj; + char *zBuf = 0; + + p = fts3ReallocOrFree(p, nSpace + nToken*sizeof(Fts3PhraseToken) + nTemp); + if( !p ) goto no_mem; + memset(p, 0, (char *)&(((Fts3Phrase *)&p[1])->aToken[0])-(char *)p); + p->eType = FTSQUERY_PHRASE; + p->pPhrase = (Fts3Phrase *)&p[1]; + p->pPhrase->iColumn = pParse->iDefaultCol; + p->pPhrase->nToken = nToken; + + zBuf = (char *)&p->pPhrase->aToken[nToken]; + if( zTemp ){ + memcpy(zBuf, zTemp, nTemp); + sqlite3_free(zTemp); + }else{ + assert( nTemp==0 ); + } + + for(jj=0; jjpPhrase->nToken; jj++){ + p->pPhrase->aToken[jj].z = zBuf; + zBuf += p->pPhrase->aToken[jj].n; + } + rc = SQLITE_OK; + } + + *ppExpr = p; + return rc; +no_mem: + + if( pCursor ){ + pModule->xClose(pCursor); + } + sqlite3_free(zTemp); + sqlite3_free(p); + *ppExpr = 0; + return SQLITE_NOMEM; +} + +/* +** The output variable *ppExpr is populated with an allocated Fts3Expr +** structure, or set to 0 if the end of the input buffer is reached. +** +** Returns an SQLite error code. SQLITE_OK if everything works, SQLITE_NOMEM +** if a malloc failure occurs, or SQLITE_ERROR if a parse error is encountered. +** If SQLITE_ERROR is returned, pContext is populated with an error message. +*/ +static int getNextNode( + ParseContext *pParse, /* fts3 query parse context */ + const char *z, int n, /* Input string */ + Fts3Expr **ppExpr, /* OUT: expression */ + int *pnConsumed /* OUT: Number of bytes consumed */ +){ + static const struct Fts3Keyword { + char *z; /* Keyword text */ + unsigned char n; /* Length of the keyword */ + unsigned char parenOnly; /* Only valid in paren mode */ + unsigned char eType; /* Keyword code */ + } aKeyword[] = { + { "OR" , 2, 0, FTSQUERY_OR }, + { "AND", 3, 1, FTSQUERY_AND }, + { "NOT", 3, 1, FTSQUERY_NOT }, + { "NEAR", 4, 0, FTSQUERY_NEAR } + }; + int ii; + int iCol; + int iColLen; + int rc; + Fts3Expr *pRet = 0; + + const char *zInput = z; + int nInput = n; + + pParse->isNot = 0; + + /* Skip over any whitespace before checking for a keyword, an open or + ** close bracket, or a quoted string. + */ + while( nInput>0 && fts3isspace(*zInput) ){ + nInput--; + zInput++; + } + if( nInput==0 ){ + return SQLITE_DONE; + } + + /* See if we are dealing with a keyword. */ + for(ii=0; ii<(int)(sizeof(aKeyword)/sizeof(struct Fts3Keyword)); ii++){ + const struct Fts3Keyword *pKey = &aKeyword[ii]; + + if( (pKey->parenOnly & ~sqlite3_fts3_enable_parentheses)!=0 ){ + continue; + } + + if( nInput>=pKey->n && 0==memcmp(zInput, pKey->z, pKey->n) ){ + int nNear = SQLITE_FTS3_DEFAULT_NEAR_PARAM; + int nKey = pKey->n; + char cNext; + + /* If this is a "NEAR" keyword, check for an explicit nearness. */ + if( pKey->eType==FTSQUERY_NEAR ){ + assert( nKey==4 ); + if( zInput[4]=='/' && zInput[5]>='0' && zInput[5]<='9' ){ + nNear = 0; + for(nKey=5; zInput[nKey]>='0' && zInput[nKey]<='9'; nKey++){ + nNear = nNear * 10 + (zInput[nKey] - '0'); + } + } + } + + /* At this point this is probably a keyword. But for that to be true, + ** the next byte must contain either whitespace, an open or close + ** parenthesis, a quote character, or EOF. + */ + cNext = zInput[nKey]; + if( fts3isspace(cNext) + || cNext=='"' || cNext=='(' || cNext==')' || cNext==0 + ){ + pRet = (Fts3Expr *)fts3MallocZero(sizeof(Fts3Expr)); + if( !pRet ){ + return SQLITE_NOMEM; + } + pRet->eType = pKey->eType; + pRet->nNear = nNear; + *ppExpr = pRet; + *pnConsumed = (int)((zInput - z) + nKey); + return SQLITE_OK; + } + + /* Turns out that wasn't a keyword after all. This happens if the + ** user has supplied a token such as "ORacle". Continue. + */ + } + } + + /* See if we are dealing with a quoted phrase. If this is the case, then + ** search for the closing quote and pass the whole string to getNextString() + ** for processing. This is easy to do, as fts3 has no syntax for escaping + ** a quote character embedded in a string. + */ + if( *zInput=='"' ){ + for(ii=1; iinNest++; + rc = fts3ExprParse(pParse, zInput+1, nInput-1, ppExpr, &nConsumed); + *pnConsumed = (int)(zInput - z) + 1 + nConsumed; + return rc; + }else if( *zInput==')' ){ + pParse->nNest--; + *pnConsumed = (int)((zInput - z) + 1); + *ppExpr = 0; + return SQLITE_DONE; + } + } + + /* If control flows to this point, this must be a regular token, or + ** the end of the input. Read a regular token using the sqlite3_tokenizer + ** interface. Before doing so, figure out if there is an explicit + ** column specifier for the token. + ** + ** TODO: Strangely, it is not possible to associate a column specifier + ** with a quoted phrase, only with a single token. Not sure if this was + ** an implementation artifact or an intentional decision when fts3 was + ** first implemented. Whichever it was, this module duplicates the + ** limitation. + */ + iCol = pParse->iDefaultCol; + iColLen = 0; + for(ii=0; iinCol; ii++){ + const char *zStr = pParse->azCol[ii]; + int nStr = (int)strlen(zStr); + if( nInput>nStr && zInput[nStr]==':' + && sqlite3_strnicmp(zStr, zInput, nStr)==0 + ){ + iCol = ii; + iColLen = (int)((zInput - z) + nStr + 1); + break; + } + } + rc = getNextToken(pParse, iCol, &z[iColLen], n-iColLen, ppExpr, pnConsumed); + *pnConsumed += iColLen; + return rc; +} + +/* +** The argument is an Fts3Expr structure for a binary operator (any type +** except an FTSQUERY_PHRASE). Return an integer value representing the +** precedence of the operator. Lower values have a higher precedence (i.e. +** group more tightly). For example, in the C language, the == operator +** groups more tightly than ||, and would therefore have a higher precedence. +** +** When using the new fts3 query syntax (when SQLITE_ENABLE_FTS3_PARENTHESIS +** is defined), the order of the operators in precedence from highest to +** lowest is: +** +** NEAR +** NOT +** AND (including implicit ANDs) +** OR +** +** Note that when using the old query syntax, the OR operator has a higher +** precedence than the AND operator. +*/ +static int opPrecedence(Fts3Expr *p){ + assert( p->eType!=FTSQUERY_PHRASE ); + if( sqlite3_fts3_enable_parentheses ){ + return p->eType; + }else if( p->eType==FTSQUERY_NEAR ){ + return 1; + }else if( p->eType==FTSQUERY_OR ){ + return 2; + } + assert( p->eType==FTSQUERY_AND ); + return 3; +} + +/* +** Argument ppHead contains a pointer to the current head of a query +** expression tree being parsed. pPrev is the expression node most recently +** inserted into the tree. This function adds pNew, which is always a binary +** operator node, into the expression tree based on the relative precedence +** of pNew and the existing nodes of the tree. This may result in the head +** of the tree changing, in which case *ppHead is set to the new root node. +*/ +static void insertBinaryOperator( + Fts3Expr **ppHead, /* Pointer to the root node of a tree */ + Fts3Expr *pPrev, /* Node most recently inserted into the tree */ + Fts3Expr *pNew /* New binary node to insert into expression tree */ +){ + Fts3Expr *pSplit = pPrev; + while( pSplit->pParent && opPrecedence(pSplit->pParent)<=opPrecedence(pNew) ){ + pSplit = pSplit->pParent; + } + + if( pSplit->pParent ){ + assert( pSplit->pParent->pRight==pSplit ); + pSplit->pParent->pRight = pNew; + pNew->pParent = pSplit->pParent; + }else{ + *ppHead = pNew; + } + pNew->pLeft = pSplit; + pSplit->pParent = pNew; +} + +/* +** Parse the fts3 query expression found in buffer z, length n. This function +** returns either when the end of the buffer is reached or an unmatched +** closing bracket - ')' - is encountered. +** +** If successful, SQLITE_OK is returned, *ppExpr is set to point to the +** parsed form of the expression and *pnConsumed is set to the number of +** bytes read from buffer z. Otherwise, *ppExpr is set to 0 and SQLITE_NOMEM +** (out of memory error) or SQLITE_ERROR (parse error) is returned. +*/ +static int fts3ExprParse( + ParseContext *pParse, /* fts3 query parse context */ + const char *z, int n, /* Text of MATCH query */ + Fts3Expr **ppExpr, /* OUT: Parsed query structure */ + int *pnConsumed /* OUT: Number of bytes consumed */ +){ + Fts3Expr *pRet = 0; + Fts3Expr *pPrev = 0; + Fts3Expr *pNotBranch = 0; /* Only used in legacy parse mode */ + int nIn = n; + const char *zIn = z; + int rc = SQLITE_OK; + int isRequirePhrase = 1; + + while( rc==SQLITE_OK ){ + Fts3Expr *p = 0; + int nByte = 0; + + rc = getNextNode(pParse, zIn, nIn, &p, &nByte); + assert( nByte>0 || (rc!=SQLITE_OK && p==0) ); + if( rc==SQLITE_OK ){ + if( p ){ + int isPhrase; + + if( !sqlite3_fts3_enable_parentheses + && p->eType==FTSQUERY_PHRASE && pParse->isNot + ){ + /* Create an implicit NOT operator. */ + Fts3Expr *pNot = fts3MallocZero(sizeof(Fts3Expr)); + if( !pNot ){ + sqlite3Fts3ExprFree(p); + rc = SQLITE_NOMEM; + goto exprparse_out; + } + pNot->eType = FTSQUERY_NOT; + pNot->pRight = p; + p->pParent = pNot; + if( pNotBranch ){ + pNot->pLeft = pNotBranch; + pNotBranch->pParent = pNot; + } + pNotBranch = pNot; + p = pPrev; + }else{ + int eType = p->eType; + isPhrase = (eType==FTSQUERY_PHRASE || p->pLeft); + + /* The isRequirePhrase variable is set to true if a phrase or + ** an expression contained in parenthesis is required. If a + ** binary operator (AND, OR, NOT or NEAR) is encounted when + ** isRequirePhrase is set, this is a syntax error. + */ + if( !isPhrase && isRequirePhrase ){ + sqlite3Fts3ExprFree(p); + rc = SQLITE_ERROR; + goto exprparse_out; + } + + if( isPhrase && !isRequirePhrase ){ + /* Insert an implicit AND operator. */ + Fts3Expr *pAnd; + assert( pRet && pPrev ); + pAnd = fts3MallocZero(sizeof(Fts3Expr)); + if( !pAnd ){ + sqlite3Fts3ExprFree(p); + rc = SQLITE_NOMEM; + goto exprparse_out; + } + pAnd->eType = FTSQUERY_AND; + insertBinaryOperator(&pRet, pPrev, pAnd); + pPrev = pAnd; + } + + /* This test catches attempts to make either operand of a NEAR + ** operator something other than a phrase. For example, either of + ** the following: + ** + ** (bracketed expression) NEAR phrase + ** phrase NEAR (bracketed expression) + ** + ** Return an error in either case. + */ + if( pPrev && ( + (eType==FTSQUERY_NEAR && !isPhrase && pPrev->eType!=FTSQUERY_PHRASE) + || (eType!=FTSQUERY_PHRASE && isPhrase && pPrev->eType==FTSQUERY_NEAR) + )){ + sqlite3Fts3ExprFree(p); + rc = SQLITE_ERROR; + goto exprparse_out; + } + + if( isPhrase ){ + if( pRet ){ + assert( pPrev && pPrev->pLeft && pPrev->pRight==0 ); + pPrev->pRight = p; + p->pParent = pPrev; + }else{ + pRet = p; + } + }else{ + insertBinaryOperator(&pRet, pPrev, p); + } + isRequirePhrase = !isPhrase; + } + pPrev = p; + } + assert( nByte>0 ); + } + assert( rc!=SQLITE_OK || (nByte>0 && nByte<=nIn) ); + nIn -= nByte; + zIn += nByte; + } + + if( rc==SQLITE_DONE && pRet && isRequirePhrase ){ + rc = SQLITE_ERROR; + } + + if( rc==SQLITE_DONE ){ + rc = SQLITE_OK; + if( !sqlite3_fts3_enable_parentheses && pNotBranch ){ + if( !pRet ){ + rc = SQLITE_ERROR; + }else{ + Fts3Expr *pIter = pNotBranch; + while( pIter->pLeft ){ + pIter = pIter->pLeft; + } + pIter->pLeft = pRet; + pRet->pParent = pIter; + pRet = pNotBranch; + } + } + } + *pnConsumed = n - nIn; + +exprparse_out: + if( rc!=SQLITE_OK ){ + sqlite3Fts3ExprFree(pRet); + sqlite3Fts3ExprFree(pNotBranch); + pRet = 0; + } + *ppExpr = pRet; + return rc; +} + +/* +** Return SQLITE_ERROR if the maximum depth of the expression tree passed +** as the only argument is more than nMaxDepth. +*/ +static int fts3ExprCheckDepth(Fts3Expr *p, int nMaxDepth){ + int rc = SQLITE_OK; + if( p ){ + if( nMaxDepth<0 ){ + rc = SQLITE_TOOBIG; + }else{ + rc = fts3ExprCheckDepth(p->pLeft, nMaxDepth-1); + if( rc==SQLITE_OK ){ + rc = fts3ExprCheckDepth(p->pRight, nMaxDepth-1); + } + } + } + return rc; +} + +/* +** This function attempts to transform the expression tree at (*pp) to +** an equivalent but more balanced form. The tree is modified in place. +** If successful, SQLITE_OK is returned and (*pp) set to point to the +** new root expression node. +** +** nMaxDepth is the maximum allowable depth of the balanced sub-tree. +** +** Otherwise, if an error occurs, an SQLite error code is returned and +** expression (*pp) freed. +*/ +static int fts3ExprBalance(Fts3Expr **pp, int nMaxDepth){ + int rc = SQLITE_OK; /* Return code */ + Fts3Expr *pRoot = *pp; /* Initial root node */ + Fts3Expr *pFree = 0; /* List of free nodes. Linked by pParent. */ + int eType = pRoot->eType; /* Type of node in this tree */ + + if( nMaxDepth==0 ){ + rc = SQLITE_ERROR; + } + + if( rc==SQLITE_OK ){ + if( (eType==FTSQUERY_AND || eType==FTSQUERY_OR) ){ + Fts3Expr **apLeaf; + apLeaf = (Fts3Expr **)sqlite3_malloc64(sizeof(Fts3Expr *) * nMaxDepth); + if( 0==apLeaf ){ + rc = SQLITE_NOMEM; + }else{ + memset(apLeaf, 0, sizeof(Fts3Expr *) * nMaxDepth); + } + + if( rc==SQLITE_OK ){ + int i; + Fts3Expr *p; + + /* Set $p to point to the left-most leaf in the tree of eType nodes. */ + for(p=pRoot; p->eType==eType; p=p->pLeft){ + assert( p->pParent==0 || p->pParent->pLeft==p ); + assert( p->pLeft && p->pRight ); + } + + /* This loop runs once for each leaf in the tree of eType nodes. */ + while( 1 ){ + int iLvl; + Fts3Expr *pParent = p->pParent; /* Current parent of p */ + + assert( pParent==0 || pParent->pLeft==p ); + p->pParent = 0; + if( pParent ){ + pParent->pLeft = 0; + }else{ + pRoot = 0; + } + rc = fts3ExprBalance(&p, nMaxDepth-1); + if( rc!=SQLITE_OK ) break; + + for(iLvl=0; p && iLvlpLeft = apLeaf[iLvl]; + pFree->pRight = p; + pFree->pLeft->pParent = pFree; + pFree->pRight->pParent = pFree; + + p = pFree; + pFree = pFree->pParent; + p->pParent = 0; + apLeaf[iLvl] = 0; + } + } + if( p ){ + sqlite3Fts3ExprFree(p); + rc = SQLITE_TOOBIG; + break; + } + + /* If that was the last leaf node, break out of the loop */ + if( pParent==0 ) break; + + /* Set $p to point to the next leaf in the tree of eType nodes */ + for(p=pParent->pRight; p->eType==eType; p=p->pLeft); + + /* Remove pParent from the original tree. */ + assert( pParent->pParent==0 || pParent->pParent->pLeft==pParent ); + pParent->pRight->pParent = pParent->pParent; + if( pParent->pParent ){ + pParent->pParent->pLeft = pParent->pRight; + }else{ + assert( pParent==pRoot ); + pRoot = pParent->pRight; + } + + /* Link pParent into the free node list. It will be used as an + ** internal node of the new tree. */ + pParent->pParent = pFree; + pFree = pParent; + } + + if( rc==SQLITE_OK ){ + p = 0; + for(i=0; ipParent = 0; + }else{ + assert( pFree!=0 ); + pFree->pRight = p; + pFree->pLeft = apLeaf[i]; + pFree->pLeft->pParent = pFree; + pFree->pRight->pParent = pFree; + + p = pFree; + pFree = pFree->pParent; + p->pParent = 0; + } + } + } + pRoot = p; + }else{ + /* An error occurred. Delete the contents of the apLeaf[] array + ** and pFree list. Everything else is cleaned up by the call to + ** sqlite3Fts3ExprFree(pRoot) below. */ + Fts3Expr *pDel; + for(i=0; ipParent; + sqlite3_free(pDel); + } + } + + assert( pFree==0 ); + sqlite3_free( apLeaf ); + } + }else if( eType==FTSQUERY_NOT ){ + Fts3Expr *pLeft = pRoot->pLeft; + Fts3Expr *pRight = pRoot->pRight; + + pRoot->pLeft = 0; + pRoot->pRight = 0; + pLeft->pParent = 0; + pRight->pParent = 0; + + rc = fts3ExprBalance(&pLeft, nMaxDepth-1); + if( rc==SQLITE_OK ){ + rc = fts3ExprBalance(&pRight, nMaxDepth-1); + } + + if( rc!=SQLITE_OK ){ + sqlite3Fts3ExprFree(pRight); + sqlite3Fts3ExprFree(pLeft); + }else{ + assert( pLeft && pRight ); + pRoot->pLeft = pLeft; + pLeft->pParent = pRoot; + pRoot->pRight = pRight; + pRight->pParent = pRoot; + } + } + } + + if( rc!=SQLITE_OK ){ + sqlite3Fts3ExprFree(pRoot); + pRoot = 0; + } + *pp = pRoot; + return rc; +} + +/* +** This function is similar to sqlite3Fts3ExprParse(), with the following +** differences: +** +** 1. It does not do expression rebalancing. +** 2. It does not check that the expression does not exceed the +** maximum allowable depth. +** 3. Even if it fails, *ppExpr may still be set to point to an +** expression tree. It should be deleted using sqlite3Fts3ExprFree() +** in this case. +*/ +static int fts3ExprParseUnbalanced( + sqlite3_tokenizer *pTokenizer, /* Tokenizer module */ + int iLangid, /* Language id for tokenizer */ + char **azCol, /* Array of column names for fts3 table */ + int bFts4, /* True to allow FTS4-only syntax */ + int nCol, /* Number of entries in azCol[] */ + int iDefaultCol, /* Default column to query */ + const char *z, int n, /* Text of MATCH query */ + Fts3Expr **ppExpr /* OUT: Parsed query structure */ +){ + int nParsed; + int rc; + ParseContext sParse; + + memset(&sParse, 0, sizeof(ParseContext)); + sParse.pTokenizer = pTokenizer; + sParse.iLangid = iLangid; + sParse.azCol = (const char **)azCol; + sParse.nCol = nCol; + sParse.iDefaultCol = iDefaultCol; + sParse.bFts4 = bFts4; + if( z==0 ){ + *ppExpr = 0; + return SQLITE_OK; + } + if( n<0 ){ + n = (int)strlen(z); + } + rc = fts3ExprParse(&sParse, z, n, ppExpr, &nParsed); + assert( rc==SQLITE_OK || *ppExpr==0 ); + + /* Check for mismatched parenthesis */ + if( rc==SQLITE_OK && sParse.nNest ){ + rc = SQLITE_ERROR; + } + + return rc; +} + +/* +** Parameters z and n contain a pointer to and length of a buffer containing +** an fts3 query expression, respectively. This function attempts to parse the +** query expression and create a tree of Fts3Expr structures representing the +** parsed expression. If successful, *ppExpr is set to point to the head +** of the parsed expression tree and SQLITE_OK is returned. If an error +** occurs, either SQLITE_NOMEM (out-of-memory error) or SQLITE_ERROR (parse +** error) is returned and *ppExpr is set to 0. +** +** If parameter n is a negative number, then z is assumed to point to a +** nul-terminated string and the length is determined using strlen(). +** +** The first parameter, pTokenizer, is passed the fts3 tokenizer module to +** use to normalize query tokens while parsing the expression. The azCol[] +** array, which is assumed to contain nCol entries, should contain the names +** of each column in the target fts3 table, in order from left to right. +** Column names must be nul-terminated strings. +** +** The iDefaultCol parameter should be passed the index of the table column +** that appears on the left-hand-side of the MATCH operator (the default +** column to match against for tokens for which a column name is not explicitly +** specified as part of the query string), or -1 if tokens may by default +** match any table column. +*/ +SQLITE_PRIVATE int sqlite3Fts3ExprParse( + sqlite3_tokenizer *pTokenizer, /* Tokenizer module */ + int iLangid, /* Language id for tokenizer */ + char **azCol, /* Array of column names for fts3 table */ + int bFts4, /* True to allow FTS4-only syntax */ + int nCol, /* Number of entries in azCol[] */ + int iDefaultCol, /* Default column to query */ + const char *z, int n, /* Text of MATCH query */ + Fts3Expr **ppExpr, /* OUT: Parsed query structure */ + char **pzErr /* OUT: Error message (sqlite3_malloc) */ +){ + int rc = fts3ExprParseUnbalanced( + pTokenizer, iLangid, azCol, bFts4, nCol, iDefaultCol, z, n, ppExpr + ); + + /* Rebalance the expression. And check that its depth does not exceed + ** SQLITE_FTS3_MAX_EXPR_DEPTH. */ + if( rc==SQLITE_OK && *ppExpr ){ + rc = fts3ExprBalance(ppExpr, SQLITE_FTS3_MAX_EXPR_DEPTH); + if( rc==SQLITE_OK ){ + rc = fts3ExprCheckDepth(*ppExpr, SQLITE_FTS3_MAX_EXPR_DEPTH); + } + } + + if( rc!=SQLITE_OK ){ + sqlite3Fts3ExprFree(*ppExpr); + *ppExpr = 0; + if( rc==SQLITE_TOOBIG ){ + sqlite3Fts3ErrMsg(pzErr, + "FTS expression tree is too large (maximum depth %d)", + SQLITE_FTS3_MAX_EXPR_DEPTH + ); + rc = SQLITE_ERROR; + }else if( rc==SQLITE_ERROR ){ + sqlite3Fts3ErrMsg(pzErr, "malformed MATCH expression: [%s]", z); + } + } + + return rc; +} + +/* +** Free a single node of an expression tree. +*/ +static void fts3FreeExprNode(Fts3Expr *p){ + assert( p->eType==FTSQUERY_PHRASE || p->pPhrase==0 ); + sqlite3Fts3EvalPhraseCleanup(p->pPhrase); + sqlite3_free(p->aMI); + sqlite3_free(p); +} + +/* +** Free a parsed fts3 query expression allocated by sqlite3Fts3ExprParse(). +** +** This function would be simpler if it recursively called itself. But +** that would mean passing a sufficiently large expression to ExprParse() +** could cause a stack overflow. +*/ +SQLITE_PRIVATE void sqlite3Fts3ExprFree(Fts3Expr *pDel){ + Fts3Expr *p; + assert( pDel==0 || pDel->pParent==0 ); + for(p=pDel; p && (p->pLeft||p->pRight); p=(p->pLeft ? p->pLeft : p->pRight)){ + assert( p->pParent==0 || p==p->pParent->pRight || p==p->pParent->pLeft ); + } + while( p ){ + Fts3Expr *pParent = p->pParent; + fts3FreeExprNode(p); + if( pParent && p==pParent->pLeft && pParent->pRight ){ + p = pParent->pRight; + while( p && (p->pLeft || p->pRight) ){ + assert( p==p->pParent->pRight || p==p->pParent->pLeft ); + p = (p->pLeft ? p->pLeft : p->pRight); + } + }else{ + p = pParent; + } + } +} + +/**************************************************************************** +***************************************************************************** +** Everything after this point is just test code. +*/ + +#ifdef SQLITE_TEST + +/* #include */ + +/* +** Return a pointer to a buffer containing a text representation of the +** expression passed as the first argument. The buffer is obtained from +** sqlite3_malloc(). It is the responsibility of the caller to use +** sqlite3_free() to release the memory. If an OOM condition is encountered, +** NULL is returned. +** +** If the second argument is not NULL, then its contents are prepended to +** the returned expression text and then freed using sqlite3_free(). +*/ +static char *exprToString(Fts3Expr *pExpr, char *zBuf){ + if( pExpr==0 ){ + return sqlite3_mprintf(""); + } + switch( pExpr->eType ){ + case FTSQUERY_PHRASE: { + Fts3Phrase *pPhrase = pExpr->pPhrase; + int i; + zBuf = sqlite3_mprintf( + "%zPHRASE %d 0", zBuf, pPhrase->iColumn); + for(i=0; zBuf && inToken; i++){ + zBuf = sqlite3_mprintf("%z %.*s%s", zBuf, + pPhrase->aToken[i].n, pPhrase->aToken[i].z, + (pPhrase->aToken[i].isPrefix?"+":"") + ); + } + return zBuf; + } + + case FTSQUERY_NEAR: + zBuf = sqlite3_mprintf("%zNEAR/%d ", zBuf, pExpr->nNear); + break; + case FTSQUERY_NOT: + zBuf = sqlite3_mprintf("%zNOT ", zBuf); + break; + case FTSQUERY_AND: + zBuf = sqlite3_mprintf("%zAND ", zBuf); + break; + case FTSQUERY_OR: + zBuf = sqlite3_mprintf("%zOR ", zBuf); + break; + } + + if( zBuf ) zBuf = sqlite3_mprintf("%z{", zBuf); + if( zBuf ) zBuf = exprToString(pExpr->pLeft, zBuf); + if( zBuf ) zBuf = sqlite3_mprintf("%z} {", zBuf); + + if( zBuf ) zBuf = exprToString(pExpr->pRight, zBuf); + if( zBuf ) zBuf = sqlite3_mprintf("%z}", zBuf); + + return zBuf; +} + +/* +** This is the implementation of a scalar SQL function used to test the +** expression parser. It should be called as follows: +** +** fts3_exprtest(, , , ...); +** +** The first argument, , is the name of the fts3 tokenizer used +** to parse the query expression (see README.tokenizers). The second argument +** is the query expression to parse. Each subsequent argument is the name +** of a column of the fts3 table that the query expression may refer to. +** For example: +** +** SELECT fts3_exprtest('simple', 'Bill col2:Bloggs', 'col1', 'col2'); +*/ +static void fts3ExprTestCommon( + int bRebalance, + sqlite3_context *context, + int argc, + sqlite3_value **argv +){ + sqlite3_tokenizer *pTokenizer = 0; + int rc; + char **azCol = 0; + const char *zExpr; + int nExpr; + int nCol; + int ii; + Fts3Expr *pExpr; + char *zBuf = 0; + Fts3Hash *pHash = (Fts3Hash*)sqlite3_user_data(context); + const char *zTokenizer = 0; + char *zErr = 0; + + if( argc<3 ){ + sqlite3_result_error(context, + "Usage: fts3_exprtest(tokenizer, expr, col1, ...", -1 + ); + return; + } + + zTokenizer = (const char*)sqlite3_value_text(argv[0]); + rc = sqlite3Fts3InitTokenizer(pHash, zTokenizer, &pTokenizer, &zErr); + if( rc!=SQLITE_OK ){ + if( rc==SQLITE_NOMEM ){ + sqlite3_result_error_nomem(context); + }else{ + sqlite3_result_error(context, zErr, -1); + } + sqlite3_free(zErr); + return; + } + + zExpr = (const char *)sqlite3_value_text(argv[1]); + nExpr = sqlite3_value_bytes(argv[1]); + nCol = argc-2; + azCol = (char **)sqlite3_malloc64(nCol*sizeof(char *)); + if( !azCol ){ + sqlite3_result_error_nomem(context); + goto exprtest_out; + } + for(ii=0; iipModule->xDestroy(pTokenizer); + } + sqlite3_free(azCol); +} + +static void fts3ExprTest( + sqlite3_context *context, + int argc, + sqlite3_value **argv +){ + fts3ExprTestCommon(0, context, argc, argv); +} +static void fts3ExprTestRebalance( + sqlite3_context *context, + int argc, + sqlite3_value **argv +){ + fts3ExprTestCommon(1, context, argc, argv); +} + +/* +** Register the query expression parser test function fts3_exprtest() +** with database connection db. +*/ +SQLITE_PRIVATE int sqlite3Fts3ExprInitTestInterface(sqlite3 *db, Fts3Hash *pHash){ + int rc = sqlite3_create_function( + db, "fts3_exprtest", -1, SQLITE_UTF8, (void*)pHash, fts3ExprTest, 0, 0 + ); + if( rc==SQLITE_OK ){ + rc = sqlite3_create_function(db, "fts3_exprtest_rebalance", + -1, SQLITE_UTF8, (void*)pHash, fts3ExprTestRebalance, 0, 0 + ); + } + return rc; +} + +#endif +#endif /* !defined(SQLITE_CORE) || defined(SQLITE_ENABLE_FTS3) */ + +/************** End of fts3_expr.c *******************************************/ +/************** Begin file fts3_hash.c ***************************************/ +/* +** 2001 September 22 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** This is the implementation of generic hash-tables used in SQLite. +** We've modified it slightly to serve as a standalone hash table +** implementation for the full-text indexing module. +*/ + +/* +** The code in this file is only compiled if: +** +** * The FTS3 module is being built as an extension +** (in which case SQLITE_CORE is not defined), or +** +** * The FTS3 module is being built into the core of +** SQLite (in which case SQLITE_ENABLE_FTS3 is defined). +*/ +/* #include "fts3Int.h" */ +#if !defined(SQLITE_CORE) || defined(SQLITE_ENABLE_FTS3) + +/* #include */ +/* #include */ +/* #include */ + +/* #include "fts3_hash.h" */ + +/* +** Malloc and Free functions +*/ +static void *fts3HashMalloc(sqlite3_int64 n){ + void *p = sqlite3_malloc64(n); + if( p ){ + memset(p, 0, n); + } + return p; +} +static void fts3HashFree(void *p){ + sqlite3_free(p); +} + +/* Turn bulk memory into a hash table object by initializing the +** fields of the Hash structure. +** +** "pNew" is a pointer to the hash table that is to be initialized. +** keyClass is one of the constants +** FTS3_HASH_BINARY or FTS3_HASH_STRING. The value of keyClass +** determines what kind of key the hash table will use. "copyKey" is +** true if the hash table should make its own private copy of keys and +** false if it should just use the supplied pointer. +*/ +SQLITE_PRIVATE void sqlite3Fts3HashInit(Fts3Hash *pNew, char keyClass, char copyKey){ + assert( pNew!=0 ); + assert( keyClass>=FTS3_HASH_STRING && keyClass<=FTS3_HASH_BINARY ); + pNew->keyClass = keyClass; + pNew->copyKey = copyKey; + pNew->first = 0; + pNew->count = 0; + pNew->htsize = 0; + pNew->ht = 0; +} + +/* Remove all entries from a hash table. Reclaim all memory. +** Call this routine to delete a hash table or to reset a hash table +** to the empty state. +*/ +SQLITE_PRIVATE void sqlite3Fts3HashClear(Fts3Hash *pH){ + Fts3HashElem *elem; /* For looping over all elements of the table */ + + assert( pH!=0 ); + elem = pH->first; + pH->first = 0; + fts3HashFree(pH->ht); + pH->ht = 0; + pH->htsize = 0; + while( elem ){ + Fts3HashElem *next_elem = elem->next; + if( pH->copyKey && elem->pKey ){ + fts3HashFree(elem->pKey); + } + fts3HashFree(elem); + elem = next_elem; + } + pH->count = 0; +} + +/* +** Hash and comparison functions when the mode is FTS3_HASH_STRING +*/ +static int fts3StrHash(const void *pKey, int nKey){ + const char *z = (const char *)pKey; + unsigned h = 0; + if( nKey<=0 ) nKey = (int) strlen(z); + while( nKey > 0 ){ + h = (h<<3) ^ h ^ *z++; + nKey--; + } + return (int)(h & 0x7fffffff); +} +static int fts3StrCompare(const void *pKey1, int n1, const void *pKey2, int n2){ + if( n1!=n2 ) return 1; + return strncmp((const char*)pKey1,(const char*)pKey2,n1); +} + +/* +** Hash and comparison functions when the mode is FTS3_HASH_BINARY +*/ +static int fts3BinHash(const void *pKey, int nKey){ + int h = 0; + const char *z = (const char *)pKey; + while( nKey-- > 0 ){ + h = (h<<3) ^ h ^ *(z++); + } + return h & 0x7fffffff; +} +static int fts3BinCompare(const void *pKey1, int n1, const void *pKey2, int n2){ + if( n1!=n2 ) return 1; + return memcmp(pKey1,pKey2,n1); +} + +/* +** Return a pointer to the appropriate hash function given the key class. +** +** The C syntax in this function definition may be unfamilar to some +** programmers, so we provide the following additional explanation: +** +** The name of the function is "ftsHashFunction". The function takes a +** single parameter "keyClass". The return value of ftsHashFunction() +** is a pointer to another function. Specifically, the return value +** of ftsHashFunction() is a pointer to a function that takes two parameters +** with types "const void*" and "int" and returns an "int". +*/ +static int (*ftsHashFunction(int keyClass))(const void*,int){ + if( keyClass==FTS3_HASH_STRING ){ + return &fts3StrHash; + }else{ + assert( keyClass==FTS3_HASH_BINARY ); + return &fts3BinHash; + } +} + +/* +** Return a pointer to the appropriate hash function given the key class. +** +** For help in interpreted the obscure C code in the function definition, +** see the header comment on the previous function. +*/ +static int (*ftsCompareFunction(int keyClass))(const void*,int,const void*,int){ + if( keyClass==FTS3_HASH_STRING ){ + return &fts3StrCompare; + }else{ + assert( keyClass==FTS3_HASH_BINARY ); + return &fts3BinCompare; + } +} + +/* Link an element into the hash table +*/ +static void fts3HashInsertElement( + Fts3Hash *pH, /* The complete hash table */ + struct _fts3ht *pEntry, /* The entry into which pNew is inserted */ + Fts3HashElem *pNew /* The element to be inserted */ +){ + Fts3HashElem *pHead; /* First element already in pEntry */ + pHead = pEntry->chain; + if( pHead ){ + pNew->next = pHead; + pNew->prev = pHead->prev; + if( pHead->prev ){ pHead->prev->next = pNew; } + else { pH->first = pNew; } + pHead->prev = pNew; + }else{ + pNew->next = pH->first; + if( pH->first ){ pH->first->prev = pNew; } + pNew->prev = 0; + pH->first = pNew; + } + pEntry->count++; + pEntry->chain = pNew; +} + + +/* Resize the hash table so that it cantains "new_size" buckets. +** "new_size" must be a power of 2. The hash table might fail +** to resize if sqliteMalloc() fails. +** +** Return non-zero if a memory allocation error occurs. +*/ +static int fts3Rehash(Fts3Hash *pH, int new_size){ + struct _fts3ht *new_ht; /* The new hash table */ + Fts3HashElem *elem, *next_elem; /* For looping over existing elements */ + int (*xHash)(const void*,int); /* The hash function */ + + assert( (new_size & (new_size-1))==0 ); + new_ht = (struct _fts3ht *)fts3HashMalloc( new_size*sizeof(struct _fts3ht) ); + if( new_ht==0 ) return 1; + fts3HashFree(pH->ht); + pH->ht = new_ht; + pH->htsize = new_size; + xHash = ftsHashFunction(pH->keyClass); + for(elem=pH->first, pH->first=0; elem; elem = next_elem){ + int h = (*xHash)(elem->pKey, elem->nKey) & (new_size-1); + next_elem = elem->next; + fts3HashInsertElement(pH, &new_ht[h], elem); + } + return 0; +} + +/* This function (for internal use only) locates an element in an +** hash table that matches the given key. The hash for this key has +** already been computed and is passed as the 4th parameter. +*/ +static Fts3HashElem *fts3FindElementByHash( + const Fts3Hash *pH, /* The pH to be searched */ + const void *pKey, /* The key we are searching for */ + int nKey, + int h /* The hash for this key. */ +){ + Fts3HashElem *elem; /* Used to loop thru the element list */ + int count; /* Number of elements left to test */ + int (*xCompare)(const void*,int,const void*,int); /* comparison function */ + + if( pH->ht ){ + struct _fts3ht *pEntry = &pH->ht[h]; + elem = pEntry->chain; + count = pEntry->count; + xCompare = ftsCompareFunction(pH->keyClass); + while( count-- && elem ){ + if( (*xCompare)(elem->pKey,elem->nKey,pKey,nKey)==0 ){ + return elem; + } + elem = elem->next; + } + } + return 0; +} + +/* Remove a single entry from the hash table given a pointer to that +** element and a hash on the element's key. +*/ +static void fts3RemoveElementByHash( + Fts3Hash *pH, /* The pH containing "elem" */ + Fts3HashElem* elem, /* The element to be removed from the pH */ + int h /* Hash value for the element */ +){ + struct _fts3ht *pEntry; + if( elem->prev ){ + elem->prev->next = elem->next; + }else{ + pH->first = elem->next; + } + if( elem->next ){ + elem->next->prev = elem->prev; + } + pEntry = &pH->ht[h]; + if( pEntry->chain==elem ){ + pEntry->chain = elem->next; + } + pEntry->count--; + if( pEntry->count<=0 ){ + pEntry->chain = 0; + } + if( pH->copyKey && elem->pKey ){ + fts3HashFree(elem->pKey); + } + fts3HashFree( elem ); + pH->count--; + if( pH->count<=0 ){ + assert( pH->first==0 ); + assert( pH->count==0 ); + fts3HashClear(pH); + } +} + +SQLITE_PRIVATE Fts3HashElem *sqlite3Fts3HashFindElem( + const Fts3Hash *pH, + const void *pKey, + int nKey +){ + int h; /* A hash on key */ + int (*xHash)(const void*,int); /* The hash function */ + + if( pH==0 || pH->ht==0 ) return 0; + xHash = ftsHashFunction(pH->keyClass); + assert( xHash!=0 ); + h = (*xHash)(pKey,nKey); + assert( (pH->htsize & (pH->htsize-1))==0 ); + return fts3FindElementByHash(pH,pKey,nKey, h & (pH->htsize-1)); +} + +/* +** Attempt to locate an element of the hash table pH with a key +** that matches pKey,nKey. Return the data for this element if it is +** found, or NULL if there is no match. +*/ +SQLITE_PRIVATE void *sqlite3Fts3HashFind(const Fts3Hash *pH, const void *pKey, int nKey){ + Fts3HashElem *pElem; /* The element that matches key (if any) */ + + pElem = sqlite3Fts3HashFindElem(pH, pKey, nKey); + return pElem ? pElem->data : 0; +} + +/* Insert an element into the hash table pH. The key is pKey,nKey +** and the data is "data". +** +** If no element exists with a matching key, then a new +** element is created. A copy of the key is made if the copyKey +** flag is set. NULL is returned. +** +** If another element already exists with the same key, then the +** new data replaces the old data and the old data is returned. +** The key is not copied in this instance. If a malloc fails, then +** the new data is returned and the hash table is unchanged. +** +** If the "data" parameter to this function is NULL, then the +** element corresponding to "key" is removed from the hash table. +*/ +SQLITE_PRIVATE void *sqlite3Fts3HashInsert( + Fts3Hash *pH, /* The hash table to insert into */ + const void *pKey, /* The key */ + int nKey, /* Number of bytes in the key */ + void *data /* The data */ +){ + int hraw; /* Raw hash value of the key */ + int h; /* the hash of the key modulo hash table size */ + Fts3HashElem *elem; /* Used to loop thru the element list */ + Fts3HashElem *new_elem; /* New element added to the pH */ + int (*xHash)(const void*,int); /* The hash function */ + + assert( pH!=0 ); + xHash = ftsHashFunction(pH->keyClass); + assert( xHash!=0 ); + hraw = (*xHash)(pKey, nKey); + assert( (pH->htsize & (pH->htsize-1))==0 ); + h = hraw & (pH->htsize-1); + elem = fts3FindElementByHash(pH,pKey,nKey,h); + if( elem ){ + void *old_data = elem->data; + if( data==0 ){ + fts3RemoveElementByHash(pH,elem,h); + }else{ + elem->data = data; + } + return old_data; + } + if( data==0 ) return 0; + if( (pH->htsize==0 && fts3Rehash(pH,8)) + || (pH->count>=pH->htsize && fts3Rehash(pH, pH->htsize*2)) + ){ + pH->count = 0; + return data; + } + assert( pH->htsize>0 ); + new_elem = (Fts3HashElem*)fts3HashMalloc( sizeof(Fts3HashElem) ); + if( new_elem==0 ) return data; + if( pH->copyKey && pKey!=0 ){ + new_elem->pKey = fts3HashMalloc( nKey ); + if( new_elem->pKey==0 ){ + fts3HashFree(new_elem); + return data; + } + memcpy((void*)new_elem->pKey, pKey, nKey); + }else{ + new_elem->pKey = (void*)pKey; + } + new_elem->nKey = nKey; + pH->count++; + assert( pH->htsize>0 ); + assert( (pH->htsize & (pH->htsize-1))==0 ); + h = hraw & (pH->htsize-1); + fts3HashInsertElement(pH, &pH->ht[h], new_elem); + new_elem->data = data; + return 0; +} + +#endif /* !defined(SQLITE_CORE) || defined(SQLITE_ENABLE_FTS3) */ + +/************** End of fts3_hash.c *******************************************/ +/************** Begin file fts3_porter.c *************************************/ +/* +** 2006 September 30 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** Implementation of the full-text-search tokenizer that implements +** a Porter stemmer. +*/ + +/* +** The code in this file is only compiled if: +** +** * The FTS3 module is being built as an extension +** (in which case SQLITE_CORE is not defined), or +** +** * The FTS3 module is being built into the core of +** SQLite (in which case SQLITE_ENABLE_FTS3 is defined). +*/ +/* #include "fts3Int.h" */ +#if !defined(SQLITE_CORE) || defined(SQLITE_ENABLE_FTS3) + +/* #include */ +/* #include */ +/* #include */ +/* #include */ + +/* #include "fts3_tokenizer.h" */ + +/* +** Class derived from sqlite3_tokenizer +*/ +typedef struct porter_tokenizer { + sqlite3_tokenizer base; /* Base class */ +} porter_tokenizer; + +/* +** Class derived from sqlite3_tokenizer_cursor +*/ +typedef struct porter_tokenizer_cursor { + sqlite3_tokenizer_cursor base; + const char *zInput; /* input we are tokenizing */ + int nInput; /* size of the input */ + int iOffset; /* current position in zInput */ + int iToken; /* index of next token to be returned */ + char *zToken; /* storage for current token */ + int nAllocated; /* space allocated to zToken buffer */ +} porter_tokenizer_cursor; + + +/* +** Create a new tokenizer instance. +*/ +static int porterCreate( + int argc, const char * const *argv, + sqlite3_tokenizer **ppTokenizer +){ + porter_tokenizer *t; + + UNUSED_PARAMETER(argc); + UNUSED_PARAMETER(argv); + + t = (porter_tokenizer *) sqlite3_malloc(sizeof(*t)); + if( t==NULL ) return SQLITE_NOMEM; + memset(t, 0, sizeof(*t)); + *ppTokenizer = &t->base; + return SQLITE_OK; +} + +/* +** Destroy a tokenizer +*/ +static int porterDestroy(sqlite3_tokenizer *pTokenizer){ + sqlite3_free(pTokenizer); + return SQLITE_OK; +} + +/* +** Prepare to begin tokenizing a particular string. The input +** string to be tokenized is zInput[0..nInput-1]. A cursor +** used to incrementally tokenize this string is returned in +** *ppCursor. +*/ +static int porterOpen( + sqlite3_tokenizer *pTokenizer, /* The tokenizer */ + const char *zInput, int nInput, /* String to be tokenized */ + sqlite3_tokenizer_cursor **ppCursor /* OUT: Tokenization cursor */ +){ + porter_tokenizer_cursor *c; + + UNUSED_PARAMETER(pTokenizer); + + c = (porter_tokenizer_cursor *) sqlite3_malloc(sizeof(*c)); + if( c==NULL ) return SQLITE_NOMEM; + + c->zInput = zInput; + if( zInput==0 ){ + c->nInput = 0; + }else if( nInput<0 ){ + c->nInput = (int)strlen(zInput); + }else{ + c->nInput = nInput; + } + c->iOffset = 0; /* start tokenizing at the beginning */ + c->iToken = 0; + c->zToken = NULL; /* no space allocated, yet. */ + c->nAllocated = 0; + + *ppCursor = &c->base; + return SQLITE_OK; +} + +/* +** Close a tokenization cursor previously opened by a call to +** porterOpen() above. +*/ +static int porterClose(sqlite3_tokenizer_cursor *pCursor){ + porter_tokenizer_cursor *c = (porter_tokenizer_cursor *) pCursor; + sqlite3_free(c->zToken); + sqlite3_free(c); + return SQLITE_OK; +} +/* +** Vowel or consonant +*/ +static const char cType[] = { + 0, 1, 1, 1, 0, 1, 1, 1, 0, 1, 1, 1, 1, 1, 0, 1, 1, 1, 1, 1, 0, + 1, 1, 1, 2, 1 +}; + +/* +** isConsonant() and isVowel() determine if their first character in +** the string they point to is a consonant or a vowel, according +** to Porter ruls. +** +** A consonate is any letter other than 'a', 'e', 'i', 'o', or 'u'. +** 'Y' is a consonant unless it follows another consonant, +** in which case it is a vowel. +** +** In these routine, the letters are in reverse order. So the 'y' rule +** is that 'y' is a consonant unless it is followed by another +** consonent. +*/ +static int isVowel(const char*); +static int isConsonant(const char *z){ + int j; + char x = *z; + if( x==0 ) return 0; + assert( x>='a' && x<='z' ); + j = cType[x-'a']; + if( j<2 ) return j; + return z[1]==0 || isVowel(z + 1); +} +static int isVowel(const char *z){ + int j; + char x = *z; + if( x==0 ) return 0; + assert( x>='a' && x<='z' ); + j = cType[x-'a']; + if( j<2 ) return 1-j; + return isConsonant(z + 1); +} + +/* +** Let any sequence of one or more vowels be represented by V and let +** C be sequence of one or more consonants. Then every word can be +** represented as: +** +** [C] (VC){m} [V] +** +** In prose: A word is an optional consonant followed by zero or +** vowel-consonant pairs followed by an optional vowel. "m" is the +** number of vowel consonant pairs. This routine computes the value +** of m for the first i bytes of a word. +** +** Return true if the m-value for z is 1 or more. In other words, +** return true if z contains at least one vowel that is followed +** by a consonant. +** +** In this routine z[] is in reverse order. So we are really looking +** for an instance of a consonant followed by a vowel. +*/ +static int m_gt_0(const char *z){ + while( isVowel(z) ){ z++; } + if( *z==0 ) return 0; + while( isConsonant(z) ){ z++; } + return *z!=0; +} + +/* Like mgt0 above except we are looking for a value of m which is +** exactly 1 +*/ +static int m_eq_1(const char *z){ + while( isVowel(z) ){ z++; } + if( *z==0 ) return 0; + while( isConsonant(z) ){ z++; } + if( *z==0 ) return 0; + while( isVowel(z) ){ z++; } + if( *z==0 ) return 1; + while( isConsonant(z) ){ z++; } + return *z==0; +} + +/* Like mgt0 above except we are looking for a value of m>1 instead +** or m>0 +*/ +static int m_gt_1(const char *z){ + while( isVowel(z) ){ z++; } + if( *z==0 ) return 0; + while( isConsonant(z) ){ z++; } + if( *z==0 ) return 0; + while( isVowel(z) ){ z++; } + if( *z==0 ) return 0; + while( isConsonant(z) ){ z++; } + return *z!=0; +} + +/* +** Return TRUE if there is a vowel anywhere within z[0..n-1] +*/ +static int hasVowel(const char *z){ + while( isConsonant(z) ){ z++; } + return *z!=0; +} + +/* +** Return TRUE if the word ends in a double consonant. +** +** The text is reversed here. So we are really looking at +** the first two characters of z[]. +*/ +static int doubleConsonant(const char *z){ + return isConsonant(z) && z[0]==z[1]; +} + +/* +** Return TRUE if the word ends with three letters which +** are consonant-vowel-consonent and where the final consonant +** is not 'w', 'x', or 'y'. +** +** The word is reversed here. So we are really checking the +** first three letters and the first one cannot be in [wxy]. +*/ +static int star_oh(const char *z){ + return + isConsonant(z) && + z[0]!='w' && z[0]!='x' && z[0]!='y' && + isVowel(z+1) && + isConsonant(z+2); +} + +/* +** If the word ends with zFrom and xCond() is true for the stem +** of the word that preceeds the zFrom ending, then change the +** ending to zTo. +** +** The input word *pz and zFrom are both in reverse order. zTo +** is in normal order. +** +** Return TRUE if zFrom matches. Return FALSE if zFrom does not +** match. Not that TRUE is returned even if xCond() fails and +** no substitution occurs. +*/ +static int stem( + char **pz, /* The word being stemmed (Reversed) */ + const char *zFrom, /* If the ending matches this... (Reversed) */ + const char *zTo, /* ... change the ending to this (not reversed) */ + int (*xCond)(const char*) /* Condition that must be true */ +){ + char *z = *pz; + while( *zFrom && *zFrom==*z ){ z++; zFrom++; } + if( *zFrom!=0 ) return 0; + if( xCond && !xCond(z) ) return 1; + while( *zTo ){ + *(--z) = *(zTo++); + } + *pz = z; + return 1; +} + +/* +** This is the fallback stemmer used when the porter stemmer is +** inappropriate. The input word is copied into the output with +** US-ASCII case folding. If the input word is too long (more +** than 20 bytes if it contains no digits or more than 6 bytes if +** it contains digits) then word is truncated to 20 or 6 bytes +** by taking 10 or 3 bytes from the beginning and end. +*/ +static void copy_stemmer(const char *zIn, int nIn, char *zOut, int *pnOut){ + int i, mx, j; + int hasDigit = 0; + for(i=0; i='A' && c<='Z' ){ + zOut[i] = c - 'A' + 'a'; + }else{ + if( c>='0' && c<='9' ) hasDigit = 1; + zOut[i] = c; + } + } + mx = hasDigit ? 3 : 10; + if( nIn>mx*2 ){ + for(j=mx, i=nIn-mx; i=(int)sizeof(zReverse)-7 ){ + /* The word is too big or too small for the porter stemmer. + ** Fallback to the copy stemmer */ + copy_stemmer(zIn, nIn, zOut, pnOut); + return; + } + for(i=0, j=sizeof(zReverse)-6; i='A' && c<='Z' ){ + zReverse[j] = c + 'a' - 'A'; + }else if( c>='a' && c<='z' ){ + zReverse[j] = c; + }else{ + /* The use of a character not in [a-zA-Z] means that we fallback + ** to the copy stemmer */ + copy_stemmer(zIn, nIn, zOut, pnOut); + return; + } + } + memset(&zReverse[sizeof(zReverse)-5], 0, 5); + z = &zReverse[j+1]; + + + /* Step 1a */ + if( z[0]=='s' ){ + if( + !stem(&z, "sess", "ss", 0) && + !stem(&z, "sei", "i", 0) && + !stem(&z, "ss", "ss", 0) + ){ + z++; + } + } + + /* Step 1b */ + z2 = z; + if( stem(&z, "dee", "ee", m_gt_0) ){ + /* Do nothing. The work was all in the test */ + }else if( + (stem(&z, "gni", "", hasVowel) || stem(&z, "de", "", hasVowel)) + && z!=z2 + ){ + if( stem(&z, "ta", "ate", 0) || + stem(&z, "lb", "ble", 0) || + stem(&z, "zi", "ize", 0) ){ + /* Do nothing. The work was all in the test */ + }else if( doubleConsonant(z) && (*z!='l' && *z!='s' && *z!='z') ){ + z++; + }else if( m_eq_1(z) && star_oh(z) ){ + *(--z) = 'e'; + } + } + + /* Step 1c */ + if( z[0]=='y' && hasVowel(z+1) ){ + z[0] = 'i'; + } + + /* Step 2 */ + switch( z[1] ){ + case 'a': + if( !stem(&z, "lanoita", "ate", m_gt_0) ){ + stem(&z, "lanoit", "tion", m_gt_0); + } + break; + case 'c': + if( !stem(&z, "icne", "ence", m_gt_0) ){ + stem(&z, "icna", "ance", m_gt_0); + } + break; + case 'e': + stem(&z, "rezi", "ize", m_gt_0); + break; + case 'g': + stem(&z, "igol", "log", m_gt_0); + break; + case 'l': + if( !stem(&z, "ilb", "ble", m_gt_0) + && !stem(&z, "illa", "al", m_gt_0) + && !stem(&z, "iltne", "ent", m_gt_0) + && !stem(&z, "ile", "e", m_gt_0) + ){ + stem(&z, "ilsuo", "ous", m_gt_0); + } + break; + case 'o': + if( !stem(&z, "noitazi", "ize", m_gt_0) + && !stem(&z, "noita", "ate", m_gt_0) + ){ + stem(&z, "rota", "ate", m_gt_0); + } + break; + case 's': + if( !stem(&z, "msila", "al", m_gt_0) + && !stem(&z, "ssenevi", "ive", m_gt_0) + && !stem(&z, "ssenluf", "ful", m_gt_0) + ){ + stem(&z, "ssensuo", "ous", m_gt_0); + } + break; + case 't': + if( !stem(&z, "itila", "al", m_gt_0) + && !stem(&z, "itivi", "ive", m_gt_0) + ){ + stem(&z, "itilib", "ble", m_gt_0); + } + break; + } + + /* Step 3 */ + switch( z[0] ){ + case 'e': + if( !stem(&z, "etaci", "ic", m_gt_0) + && !stem(&z, "evita", "", m_gt_0) + ){ + stem(&z, "ezila", "al", m_gt_0); + } + break; + case 'i': + stem(&z, "itici", "ic", m_gt_0); + break; + case 'l': + if( !stem(&z, "laci", "ic", m_gt_0) ){ + stem(&z, "luf", "", m_gt_0); + } + break; + case 's': + stem(&z, "ssen", "", m_gt_0); + break; + } + + /* Step 4 */ + switch( z[1] ){ + case 'a': + if( z[0]=='l' && m_gt_1(z+2) ){ + z += 2; + } + break; + case 'c': + if( z[0]=='e' && z[2]=='n' && (z[3]=='a' || z[3]=='e') && m_gt_1(z+4) ){ + z += 4; + } + break; + case 'e': + if( z[0]=='r' && m_gt_1(z+2) ){ + z += 2; + } + break; + case 'i': + if( z[0]=='c' && m_gt_1(z+2) ){ + z += 2; + } + break; + case 'l': + if( z[0]=='e' && z[2]=='b' && (z[3]=='a' || z[3]=='i') && m_gt_1(z+4) ){ + z += 4; + } + break; + case 'n': + if( z[0]=='t' ){ + if( z[2]=='a' ){ + if( m_gt_1(z+3) ){ + z += 3; + } + }else if( z[2]=='e' ){ + if( !stem(&z, "tneme", "", m_gt_1) + && !stem(&z, "tnem", "", m_gt_1) + ){ + stem(&z, "tne", "", m_gt_1); + } + } + } + break; + case 'o': + if( z[0]=='u' ){ + if( m_gt_1(z+2) ){ + z += 2; + } + }else if( z[3]=='s' || z[3]=='t' ){ + stem(&z, "noi", "", m_gt_1); + } + break; + case 's': + if( z[0]=='m' && z[2]=='i' && m_gt_1(z+3) ){ + z += 3; + } + break; + case 't': + if( !stem(&z, "eta", "", m_gt_1) ){ + stem(&z, "iti", "", m_gt_1); + } + break; + case 'u': + if( z[0]=='s' && z[2]=='o' && m_gt_1(z+3) ){ + z += 3; + } + break; + case 'v': + case 'z': + if( z[0]=='e' && z[2]=='i' && m_gt_1(z+3) ){ + z += 3; + } + break; + } + + /* Step 5a */ + if( z[0]=='e' ){ + if( m_gt_1(z+1) ){ + z++; + }else if( m_eq_1(z+1) && !star_oh(z+1) ){ + z++; + } + } + + /* Step 5b */ + if( m_gt_1(z) && z[0]=='l' && z[1]=='l' ){ + z++; + } + + /* z[] is now the stemmed word in reverse order. Flip it back + ** around into forward order and return. + */ + *pnOut = i = (int)strlen(z); + zOut[i] = 0; + while( *z ){ + zOut[--i] = *(z++); + } +} + +/* +** Characters that can be part of a token. We assume any character +** whose value is greater than 0x80 (any UTF character) can be +** part of a token. In other words, delimiters all must have +** values of 0x7f or lower. +*/ +static const char porterIdChar[] = { +/* x0 x1 x2 x3 x4 x5 x6 x7 x8 x9 xA xB xC xD xE xF */ + 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, /* 3x */ + 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, /* 4x */ + 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 1, /* 5x */ + 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, /* 6x */ + 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, /* 7x */ +}; +#define isDelim(C) (((ch=C)&0x80)==0 && (ch<0x30 || !porterIdChar[ch-0x30])) + +/* +** Extract the next token from a tokenization cursor. The cursor must +** have been opened by a prior call to porterOpen(). +*/ +static int porterNext( + sqlite3_tokenizer_cursor *pCursor, /* Cursor returned by porterOpen */ + const char **pzToken, /* OUT: *pzToken is the token text */ + int *pnBytes, /* OUT: Number of bytes in token */ + int *piStartOffset, /* OUT: Starting offset of token */ + int *piEndOffset, /* OUT: Ending offset of token */ + int *piPosition /* OUT: Position integer of token */ +){ + porter_tokenizer_cursor *c = (porter_tokenizer_cursor *) pCursor; + const char *z = c->zInput; + + while( c->iOffsetnInput ){ + int iStartOffset, ch; + + /* Scan past delimiter characters */ + while( c->iOffsetnInput && isDelim(z[c->iOffset]) ){ + c->iOffset++; + } + + /* Count non-delimiter characters. */ + iStartOffset = c->iOffset; + while( c->iOffsetnInput && !isDelim(z[c->iOffset]) ){ + c->iOffset++; + } + + if( c->iOffset>iStartOffset ){ + int n = c->iOffset-iStartOffset; + if( n>c->nAllocated ){ + char *pNew; + c->nAllocated = n+20; + pNew = sqlite3_realloc(c->zToken, c->nAllocated); + if( !pNew ) return SQLITE_NOMEM; + c->zToken = pNew; + } + porter_stemmer(&z[iStartOffset], n, c->zToken, pnBytes); + *pzToken = c->zToken; + *piStartOffset = iStartOffset; + *piEndOffset = c->iOffset; + *piPosition = c->iToken++; + return SQLITE_OK; + } + } + return SQLITE_DONE; +} + +/* +** The set of routines that implement the porter-stemmer tokenizer +*/ +static const sqlite3_tokenizer_module porterTokenizerModule = { + 0, + porterCreate, + porterDestroy, + porterOpen, + porterClose, + porterNext, + 0 +}; + +/* +** Allocate a new porter tokenizer. Return a pointer to the new +** tokenizer in *ppModule +*/ +SQLITE_PRIVATE void sqlite3Fts3PorterTokenizerModule( + sqlite3_tokenizer_module const**ppModule +){ + *ppModule = &porterTokenizerModule; +} + +#endif /* !defined(SQLITE_CORE) || defined(SQLITE_ENABLE_FTS3) */ + +/************** End of fts3_porter.c *****************************************/ +/************** Begin file fts3_tokenizer.c **********************************/ +/* +** 2007 June 22 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +****************************************************************************** +** +** This is part of an SQLite module implementing full-text search. +** This particular file implements the generic tokenizer interface. +*/ + +/* +** The code in this file is only compiled if: +** +** * The FTS3 module is being built as an extension +** (in which case SQLITE_CORE is not defined), or +** +** * The FTS3 module is being built into the core of +** SQLite (in which case SQLITE_ENABLE_FTS3 is defined). +*/ +/* #include "fts3Int.h" */ +#if !defined(SQLITE_CORE) || defined(SQLITE_ENABLE_FTS3) + +/* #include */ +/* #include */ + +/* +** Return true if the two-argument version of fts3_tokenizer() +** has been activated via a prior call to sqlite3_db_config(db, +** SQLITE_DBCONFIG_ENABLE_FTS3_TOKENIZER, 1, 0); +*/ +static int fts3TokenizerEnabled(sqlite3_context *context){ + sqlite3 *db = sqlite3_context_db_handle(context); + int isEnabled = 0; + sqlite3_db_config(db,SQLITE_DBCONFIG_ENABLE_FTS3_TOKENIZER,-1,&isEnabled); + return isEnabled; +} + +/* +** Implementation of the SQL scalar function for accessing the underlying +** hash table. This function may be called as follows: +** +** SELECT (); +** SELECT (, ); +** +** where is the name passed as the second argument +** to the sqlite3Fts3InitHashTable() function (e.g. 'fts3_tokenizer'). +** +** If the argument is specified, it must be a blob value +** containing a pointer to be stored as the hash data corresponding +** to the string . If is not specified, then +** the string must already exist in the has table. Otherwise, +** an error is returned. +** +** Whether or not the argument is specified, the value returned +** is a blob containing the pointer stored as the hash data corresponding +** to string (after the hash-table is updated, if applicable). +*/ +static void fts3TokenizerFunc( + sqlite3_context *context, + int argc, + sqlite3_value **argv +){ + Fts3Hash *pHash; + void *pPtr = 0; + const unsigned char *zName; + int nName; + + assert( argc==1 || argc==2 ); + + pHash = (Fts3Hash *)sqlite3_user_data(context); + + zName = sqlite3_value_text(argv[0]); + nName = sqlite3_value_bytes(argv[0])+1; + + if( argc==2 ){ + if( fts3TokenizerEnabled(context) || sqlite3_value_frombind(argv[1]) ){ + void *pOld; + int n = sqlite3_value_bytes(argv[1]); + if( zName==0 || n!=sizeof(pPtr) ){ + sqlite3_result_error(context, "argument type mismatch", -1); + return; + } + pPtr = *(void **)sqlite3_value_blob(argv[1]); + pOld = sqlite3Fts3HashInsert(pHash, (void *)zName, nName, pPtr); + if( pOld==pPtr ){ + sqlite3_result_error(context, "out of memory", -1); + } + }else{ + sqlite3_result_error(context, "fts3tokenize disabled", -1); + return; + } + }else{ + if( zName ){ + pPtr = sqlite3Fts3HashFind(pHash, zName, nName); + } + if( !pPtr ){ + char *zErr = sqlite3_mprintf("unknown tokenizer: %s", zName); + sqlite3_result_error(context, zErr, -1); + sqlite3_free(zErr); + return; + } + } + if( fts3TokenizerEnabled(context) || sqlite3_value_frombind(argv[0]) ){ + sqlite3_result_blob(context, (void *)&pPtr, sizeof(pPtr), SQLITE_TRANSIENT); + } +} + +SQLITE_PRIVATE int sqlite3Fts3IsIdChar(char c){ + static const char isFtsIdChar[] = { + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, /* 0x */ + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, /* 1x */ + 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, /* 2x */ + 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, /* 3x */ + 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, /* 4x */ + 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 1, /* 5x */ + 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, /* 6x */ + 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, /* 7x */ + }; + return (c&0x80 || isFtsIdChar[(int)(c)]); +} + +SQLITE_PRIVATE const char *sqlite3Fts3NextToken(const char *zStr, int *pn){ + const char *z1; + const char *z2 = 0; + + /* Find the start of the next token. */ + z1 = zStr; + while( z2==0 ){ + char c = *z1; + switch( c ){ + case '\0': return 0; /* No more tokens here */ + case '\'': + case '"': + case '`': { + z2 = z1; + while( *++z2 && (*z2!=c || *++z2==c) ); + break; + } + case '[': + z2 = &z1[1]; + while( *z2 && z2[0]!=']' ) z2++; + if( *z2 ) z2++; + break; + + default: + if( sqlite3Fts3IsIdChar(*z1) ){ + z2 = &z1[1]; + while( sqlite3Fts3IsIdChar(*z2) ) z2++; + }else{ + z1++; + } + } + } + + *pn = (int)(z2-z1); + return z1; +} + +SQLITE_PRIVATE int sqlite3Fts3InitTokenizer( + Fts3Hash *pHash, /* Tokenizer hash table */ + const char *zArg, /* Tokenizer name */ + sqlite3_tokenizer **ppTok, /* OUT: Tokenizer (if applicable) */ + char **pzErr /* OUT: Set to malloced error message */ +){ + int rc; + char *z = (char *)zArg; + int n = 0; + char *zCopy; + char *zEnd; /* Pointer to nul-term of zCopy */ + sqlite3_tokenizer_module *m; + + zCopy = sqlite3_mprintf("%s", zArg); + if( !zCopy ) return SQLITE_NOMEM; + zEnd = &zCopy[strlen(zCopy)]; + + z = (char *)sqlite3Fts3NextToken(zCopy, &n); + if( z==0 ){ + assert( n==0 ); + z = zCopy; + } + z[n] = '\0'; + sqlite3Fts3Dequote(z); + + m = (sqlite3_tokenizer_module *)sqlite3Fts3HashFind(pHash,z,(int)strlen(z)+1); + if( !m ){ + sqlite3Fts3ErrMsg(pzErr, "unknown tokenizer: %s", z); + rc = SQLITE_ERROR; + }else{ + char const **aArg = 0; + int iArg = 0; + z = &z[n+1]; + while( zxCreate(iArg, aArg, ppTok); + assert( rc!=SQLITE_OK || *ppTok ); + if( rc!=SQLITE_OK ){ + sqlite3Fts3ErrMsg(pzErr, "unknown tokenizer"); + }else{ + (*ppTok)->pModule = m; + } + sqlite3_free((void *)aArg); + } + + sqlite3_free(zCopy); + return rc; +} + + +#ifdef SQLITE_TEST + +#if defined(INCLUDE_SQLITE_TCL_H) +# include "sqlite_tcl.h" +#else +# include "tcl.h" +#endif +/* #include */ + +/* +** Implementation of a special SQL scalar function for testing tokenizers +** designed to be used in concert with the Tcl testing framework. This +** function must be called with two or more arguments: +** +** SELECT (, ..., ); +** +** where is the name passed as the second argument +** to the sqlite3Fts3InitHashTable() function (e.g. 'fts3_tokenizer') +** concatenated with the string '_test' (e.g. 'fts3_tokenizer_test'). +** +** The return value is a string that may be interpreted as a Tcl +** list. For each token in the , three elements are +** added to the returned list. The first is the token position, the +** second is the token text (folded, stemmed, etc.) and the third is the +** substring of associated with the token. For example, +** using the built-in "simple" tokenizer: +** +** SELECT fts_tokenizer_test('simple', 'I don't see how'); +** +** will return the string: +** +** "{0 i I 1 dont don't 2 see see 3 how how}" +** +*/ +static void testFunc( + sqlite3_context *context, + int argc, + sqlite3_value **argv +){ + Fts3Hash *pHash; + sqlite3_tokenizer_module *p; + sqlite3_tokenizer *pTokenizer = 0; + sqlite3_tokenizer_cursor *pCsr = 0; + + const char *zErr = 0; + + const char *zName; + int nName; + const char *zInput; + int nInput; + + const char *azArg[64]; + + const char *zToken; + int nToken = 0; + int iStart = 0; + int iEnd = 0; + int iPos = 0; + int i; + + Tcl_Obj *pRet; + + if( argc<2 ){ + sqlite3_result_error(context, "insufficient arguments", -1); + return; + } + + nName = sqlite3_value_bytes(argv[0]); + zName = (const char *)sqlite3_value_text(argv[0]); + nInput = sqlite3_value_bytes(argv[argc-1]); + zInput = (const char *)sqlite3_value_text(argv[argc-1]); + + pHash = (Fts3Hash *)sqlite3_user_data(context); + p = (sqlite3_tokenizer_module *)sqlite3Fts3HashFind(pHash, zName, nName+1); + + if( !p ){ + char *zErr2 = sqlite3_mprintf("unknown tokenizer: %s", zName); + sqlite3_result_error(context, zErr2, -1); + sqlite3_free(zErr2); + return; + } + + pRet = Tcl_NewObj(); + Tcl_IncrRefCount(pRet); + + for(i=1; ixCreate(argc-2, azArg, &pTokenizer) ){ + zErr = "error in xCreate()"; + goto finish; + } + pTokenizer->pModule = p; + if( sqlite3Fts3OpenTokenizer(pTokenizer, 0, zInput, nInput, &pCsr) ){ + zErr = "error in xOpen()"; + goto finish; + } + + while( SQLITE_OK==p->xNext(pCsr, &zToken, &nToken, &iStart, &iEnd, &iPos) ){ + Tcl_ListObjAppendElement(0, pRet, Tcl_NewIntObj(iPos)); + Tcl_ListObjAppendElement(0, pRet, Tcl_NewStringObj(zToken, nToken)); + zToken = &zInput[iStart]; + nToken = iEnd-iStart; + Tcl_ListObjAppendElement(0, pRet, Tcl_NewStringObj(zToken, nToken)); + } + + if( SQLITE_OK!=p->xClose(pCsr) ){ + zErr = "error in xClose()"; + goto finish; + } + if( SQLITE_OK!=p->xDestroy(pTokenizer) ){ + zErr = "error in xDestroy()"; + goto finish; + } + +finish: + if( zErr ){ + sqlite3_result_error(context, zErr, -1); + }else{ + sqlite3_result_text(context, Tcl_GetString(pRet), -1, SQLITE_TRANSIENT); + } + Tcl_DecrRefCount(pRet); +} + +static +int registerTokenizer( + sqlite3 *db, + char *zName, + const sqlite3_tokenizer_module *p +){ + int rc; + sqlite3_stmt *pStmt; + const char zSql[] = "SELECT fts3_tokenizer(?, ?)"; + + rc = sqlite3_prepare_v2(db, zSql, -1, &pStmt, 0); + if( rc!=SQLITE_OK ){ + return rc; + } + + sqlite3_bind_text(pStmt, 1, zName, -1, SQLITE_STATIC); + sqlite3_bind_blob(pStmt, 2, &p, sizeof(p), SQLITE_STATIC); + sqlite3_step(pStmt); + + return sqlite3_finalize(pStmt); +} + + +static +int queryTokenizer( + sqlite3 *db, + char *zName, + const sqlite3_tokenizer_module **pp +){ + int rc; + sqlite3_stmt *pStmt; + const char zSql[] = "SELECT fts3_tokenizer(?)"; + + *pp = 0; + rc = sqlite3_prepare_v2(db, zSql, -1, &pStmt, 0); + if( rc!=SQLITE_OK ){ + return rc; + } + + sqlite3_bind_text(pStmt, 1, zName, -1, SQLITE_STATIC); + if( SQLITE_ROW==sqlite3_step(pStmt) ){ + if( sqlite3_column_type(pStmt, 0)==SQLITE_BLOB ){ + memcpy((void *)pp, sqlite3_column_blob(pStmt, 0), sizeof(*pp)); + } + } + + return sqlite3_finalize(pStmt); +} + +SQLITE_PRIVATE void sqlite3Fts3SimpleTokenizerModule(sqlite3_tokenizer_module const**ppModule); + +/* +** Implementation of the scalar function fts3_tokenizer_internal_test(). +** This function is used for testing only, it is not included in the +** build unless SQLITE_TEST is defined. +** +** The purpose of this is to test that the fts3_tokenizer() function +** can be used as designed by the C-code in the queryTokenizer and +** registerTokenizer() functions above. These two functions are repeated +** in the README.tokenizer file as an example, so it is important to +** test them. +** +** To run the tests, evaluate the fts3_tokenizer_internal_test() scalar +** function with no arguments. An assert() will fail if a problem is +** detected. i.e.: +** +** SELECT fts3_tokenizer_internal_test(); +** +*/ +static void intTestFunc( + sqlite3_context *context, + int argc, + sqlite3_value **argv +){ + int rc; + const sqlite3_tokenizer_module *p1; + const sqlite3_tokenizer_module *p2; + sqlite3 *db = (sqlite3 *)sqlite3_user_data(context); + + UNUSED_PARAMETER(argc); + UNUSED_PARAMETER(argv); + + /* Test the query function */ + sqlite3Fts3SimpleTokenizerModule(&p1); + rc = queryTokenizer(db, "simple", &p2); + assert( rc==SQLITE_OK ); + assert( p1==p2 ); + rc = queryTokenizer(db, "nosuchtokenizer", &p2); + assert( rc==SQLITE_ERROR ); + assert( p2==0 ); + assert( 0==strcmp(sqlite3_errmsg(db), "unknown tokenizer: nosuchtokenizer") ); + + /* Test the storage function */ + if( fts3TokenizerEnabled(context) ){ + rc = registerTokenizer(db, "nosuchtokenizer", p1); + assert( rc==SQLITE_OK ); + rc = queryTokenizer(db, "nosuchtokenizer", &p2); + assert( rc==SQLITE_OK ); + assert( p2==p1 ); + } + + sqlite3_result_text(context, "ok", -1, SQLITE_STATIC); +} + +#endif + +/* +** Set up SQL objects in database db used to access the contents of +** the hash table pointed to by argument pHash. The hash table must +** been initialized to use string keys, and to take a private copy +** of the key when a value is inserted. i.e. by a call similar to: +** +** sqlite3Fts3HashInit(pHash, FTS3_HASH_STRING, 1); +** +** This function adds a scalar function (see header comment above +** fts3TokenizerFunc() in this file for details) and, if ENABLE_TABLE is +** defined at compilation time, a temporary virtual table (see header +** comment above struct HashTableVtab) to the database schema. Both +** provide read/write access to the contents of *pHash. +** +** The third argument to this function, zName, is used as the name +** of both the scalar and, if created, the virtual table. +*/ +SQLITE_PRIVATE int sqlite3Fts3InitHashTable( + sqlite3 *db, + Fts3Hash *pHash, + const char *zName +){ + int rc = SQLITE_OK; + void *p = (void *)pHash; + const int any = SQLITE_ANY; + +#ifdef SQLITE_TEST + char *zTest = 0; + char *zTest2 = 0; + void *pdb = (void *)db; + zTest = sqlite3_mprintf("%s_test", zName); + zTest2 = sqlite3_mprintf("%s_internal_test", zName); + if( !zTest || !zTest2 ){ + rc = SQLITE_NOMEM; + } +#endif + + if( SQLITE_OK==rc ){ + rc = sqlite3_create_function(db, zName, 1, any, p, fts3TokenizerFunc, 0, 0); + } + if( SQLITE_OK==rc ){ + rc = sqlite3_create_function(db, zName, 2, any, p, fts3TokenizerFunc, 0, 0); + } +#ifdef SQLITE_TEST + if( SQLITE_OK==rc ){ + rc = sqlite3_create_function(db, zTest, -1, any, p, testFunc, 0, 0); + } + if( SQLITE_OK==rc ){ + rc = sqlite3_create_function(db, zTest2, 0, any, pdb, intTestFunc, 0, 0); + } +#endif + +#ifdef SQLITE_TEST + sqlite3_free(zTest); + sqlite3_free(zTest2); +#endif + + return rc; +} + +#endif /* !defined(SQLITE_CORE) || defined(SQLITE_ENABLE_FTS3) */ + +/************** End of fts3_tokenizer.c **************************************/ +/************** Begin file fts3_tokenizer1.c *********************************/ +/* +** 2006 Oct 10 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +****************************************************************************** +** +** Implementation of the "simple" full-text-search tokenizer. +*/ + +/* +** The code in this file is only compiled if: +** +** * The FTS3 module is being built as an extension +** (in which case SQLITE_CORE is not defined), or +** +** * The FTS3 module is being built into the core of +** SQLite (in which case SQLITE_ENABLE_FTS3 is defined). +*/ +/* #include "fts3Int.h" */ +#if !defined(SQLITE_CORE) || defined(SQLITE_ENABLE_FTS3) + +/* #include */ +/* #include */ +/* #include */ +/* #include */ + +/* #include "fts3_tokenizer.h" */ + +typedef struct simple_tokenizer { + sqlite3_tokenizer base; + char delim[128]; /* flag ASCII delimiters */ +} simple_tokenizer; + +typedef struct simple_tokenizer_cursor { + sqlite3_tokenizer_cursor base; + const char *pInput; /* input we are tokenizing */ + int nBytes; /* size of the input */ + int iOffset; /* current position in pInput */ + int iToken; /* index of next token to be returned */ + char *pToken; /* storage for current token */ + int nTokenAllocated; /* space allocated to zToken buffer */ +} simple_tokenizer_cursor; + + +static int simpleDelim(simple_tokenizer *t, unsigned char c){ + return c<0x80 && t->delim[c]; +} +static int fts3_isalnum(int x){ + return (x>='0' && x<='9') || (x>='A' && x<='Z') || (x>='a' && x<='z'); +} + +/* +** Create a new tokenizer instance. +*/ +static int simpleCreate( + int argc, const char * const *argv, + sqlite3_tokenizer **ppTokenizer +){ + simple_tokenizer *t; + + t = (simple_tokenizer *) sqlite3_malloc(sizeof(*t)); + if( t==NULL ) return SQLITE_NOMEM; + memset(t, 0, sizeof(*t)); + + /* TODO(shess) Delimiters need to remain the same from run to run, + ** else we need to reindex. One solution would be a meta-table to + ** track such information in the database, then we'd only want this + ** information on the initial create. + */ + if( argc>1 ){ + int i, n = (int)strlen(argv[1]); + for(i=0; i=0x80 ){ + sqlite3_free(t); + return SQLITE_ERROR; + } + t->delim[ch] = 1; + } + } else { + /* Mark non-alphanumeric ASCII characters as delimiters */ + int i; + for(i=1; i<0x80; i++){ + t->delim[i] = !fts3_isalnum(i) ? -1 : 0; + } + } + + *ppTokenizer = &t->base; + return SQLITE_OK; +} + +/* +** Destroy a tokenizer +*/ +static int simpleDestroy(sqlite3_tokenizer *pTokenizer){ + sqlite3_free(pTokenizer); + return SQLITE_OK; +} + +/* +** Prepare to begin tokenizing a particular string. The input +** string to be tokenized is pInput[0..nBytes-1]. A cursor +** used to incrementally tokenize this string is returned in +** *ppCursor. +*/ +static int simpleOpen( + sqlite3_tokenizer *pTokenizer, /* The tokenizer */ + const char *pInput, int nBytes, /* String to be tokenized */ + sqlite3_tokenizer_cursor **ppCursor /* OUT: Tokenization cursor */ +){ + simple_tokenizer_cursor *c; + + UNUSED_PARAMETER(pTokenizer); + + c = (simple_tokenizer_cursor *) sqlite3_malloc(sizeof(*c)); + if( c==NULL ) return SQLITE_NOMEM; + + c->pInput = pInput; + if( pInput==0 ){ + c->nBytes = 0; + }else if( nBytes<0 ){ + c->nBytes = (int)strlen(pInput); + }else{ + c->nBytes = nBytes; + } + c->iOffset = 0; /* start tokenizing at the beginning */ + c->iToken = 0; + c->pToken = NULL; /* no space allocated, yet. */ + c->nTokenAllocated = 0; + + *ppCursor = &c->base; + return SQLITE_OK; +} + +/* +** Close a tokenization cursor previously opened by a call to +** simpleOpen() above. +*/ +static int simpleClose(sqlite3_tokenizer_cursor *pCursor){ + simple_tokenizer_cursor *c = (simple_tokenizer_cursor *) pCursor; + sqlite3_free(c->pToken); + sqlite3_free(c); + return SQLITE_OK; +} + +/* +** Extract the next token from a tokenization cursor. The cursor must +** have been opened by a prior call to simpleOpen(). +*/ +static int simpleNext( + sqlite3_tokenizer_cursor *pCursor, /* Cursor returned by simpleOpen */ + const char **ppToken, /* OUT: *ppToken is the token text */ + int *pnBytes, /* OUT: Number of bytes in token */ + int *piStartOffset, /* OUT: Starting offset of token */ + int *piEndOffset, /* OUT: Ending offset of token */ + int *piPosition /* OUT: Position integer of token */ +){ + simple_tokenizer_cursor *c = (simple_tokenizer_cursor *) pCursor; + simple_tokenizer *t = (simple_tokenizer *) pCursor->pTokenizer; + unsigned char *p = (unsigned char *)c->pInput; + + while( c->iOffsetnBytes ){ + int iStartOffset; + + /* Scan past delimiter characters */ + while( c->iOffsetnBytes && simpleDelim(t, p[c->iOffset]) ){ + c->iOffset++; + } + + /* Count non-delimiter characters. */ + iStartOffset = c->iOffset; + while( c->iOffsetnBytes && !simpleDelim(t, p[c->iOffset]) ){ + c->iOffset++; + } + + if( c->iOffset>iStartOffset ){ + int i, n = c->iOffset-iStartOffset; + if( n>c->nTokenAllocated ){ + char *pNew; + c->nTokenAllocated = n+20; + pNew = sqlite3_realloc(c->pToken, c->nTokenAllocated); + if( !pNew ) return SQLITE_NOMEM; + c->pToken = pNew; + } + for(i=0; ipToken[i] = (char)((ch>='A' && ch<='Z') ? ch-'A'+'a' : ch); + } + *ppToken = c->pToken; + *pnBytes = n; + *piStartOffset = iStartOffset; + *piEndOffset = c->iOffset; + *piPosition = c->iToken++; + + return SQLITE_OK; + } + } + return SQLITE_DONE; +} + +/* +** The set of routines that implement the simple tokenizer +*/ +static const sqlite3_tokenizer_module simpleTokenizerModule = { + 0, + simpleCreate, + simpleDestroy, + simpleOpen, + simpleClose, + simpleNext, + 0, +}; + +/* +** Allocate a new simple tokenizer. Return a pointer to the new +** tokenizer in *ppModule +*/ +SQLITE_PRIVATE void sqlite3Fts3SimpleTokenizerModule( + sqlite3_tokenizer_module const**ppModule +){ + *ppModule = &simpleTokenizerModule; +} + +#endif /* !defined(SQLITE_CORE) || defined(SQLITE_ENABLE_FTS3) */ + +/************** End of fts3_tokenizer1.c *************************************/ +/************** Begin file fts3_tokenize_vtab.c ******************************/ +/* +** 2013 Apr 22 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +****************************************************************************** +** +** This file contains code for the "fts3tokenize" virtual table module. +** An fts3tokenize virtual table is created as follows: +** +** CREATE VIRTUAL TABLE USING fts3tokenize( +** , , ... +** ); +** +** The table created has the following schema: +** +** CREATE TABLE (input, token, start, end, position) +** +** When queried, the query must include a WHERE clause of type: +** +** input = +** +** The virtual table module tokenizes this , using the FTS3 +** tokenizer specified by the arguments to the CREATE VIRTUAL TABLE +** statement and returns one row for each token in the result. With +** fields set as follows: +** +** input: Always set to a copy of +** token: A token from the input. +** start: Byte offset of the token within the input . +** end: Byte offset of the byte immediately following the end of the +** token within the input string. +** pos: Token offset of token within input. +** +*/ +/* #include "fts3Int.h" */ +#if !defined(SQLITE_CORE) || defined(SQLITE_ENABLE_FTS3) + +/* #include */ +/* #include */ + +typedef struct Fts3tokTable Fts3tokTable; +typedef struct Fts3tokCursor Fts3tokCursor; + +/* +** Virtual table structure. +*/ +struct Fts3tokTable { + sqlite3_vtab base; /* Base class used by SQLite core */ + const sqlite3_tokenizer_module *pMod; + sqlite3_tokenizer *pTok; +}; + +/* +** Virtual table cursor structure. +*/ +struct Fts3tokCursor { + sqlite3_vtab_cursor base; /* Base class used by SQLite core */ + char *zInput; /* Input string */ + sqlite3_tokenizer_cursor *pCsr; /* Cursor to iterate through zInput */ + int iRowid; /* Current 'rowid' value */ + const char *zToken; /* Current 'token' value */ + int nToken; /* Size of zToken in bytes */ + int iStart; /* Current 'start' value */ + int iEnd; /* Current 'end' value */ + int iPos; /* Current 'pos' value */ +}; + +/* +** Query FTS for the tokenizer implementation named zName. +*/ +static int fts3tokQueryTokenizer( + Fts3Hash *pHash, + const char *zName, + const sqlite3_tokenizer_module **pp, + char **pzErr +){ + sqlite3_tokenizer_module *p; + int nName = (int)strlen(zName); + + p = (sqlite3_tokenizer_module *)sqlite3Fts3HashFind(pHash, zName, nName+1); + if( !p ){ + sqlite3Fts3ErrMsg(pzErr, "unknown tokenizer: %s", zName); + return SQLITE_ERROR; + } + + *pp = p; + return SQLITE_OK; +} + +/* +** The second argument, argv[], is an array of pointers to nul-terminated +** strings. This function makes a copy of the array and strings into a +** single block of memory. It then dequotes any of the strings that appear +** to be quoted. +** +** If successful, output parameter *pazDequote is set to point at the +** array of dequoted strings and SQLITE_OK is returned. The caller is +** responsible for eventually calling sqlite3_free() to free the array +** in this case. Or, if an error occurs, an SQLite error code is returned. +** The final value of *pazDequote is undefined in this case. +*/ +static int fts3tokDequoteArray( + int argc, /* Number of elements in argv[] */ + const char * const *argv, /* Input array */ + char ***pazDequote /* Output array */ +){ + int rc = SQLITE_OK; /* Return code */ + if( argc==0 ){ + *pazDequote = 0; + }else{ + int i; + int nByte = 0; + char **azDequote; + + for(i=0; ixCreate((nDequote>1 ? nDequote-1 : 0), azArg, &pTok); + } + + if( rc==SQLITE_OK ){ + pTab = (Fts3tokTable *)sqlite3_malloc(sizeof(Fts3tokTable)); + if( pTab==0 ){ + rc = SQLITE_NOMEM; + } + } + + if( rc==SQLITE_OK ){ + memset(pTab, 0, sizeof(Fts3tokTable)); + pTab->pMod = pMod; + pTab->pTok = pTok; + *ppVtab = &pTab->base; + }else{ + if( pTok ){ + pMod->xDestroy(pTok); + } + } + + sqlite3_free(azDequote); + return rc; +} + +/* +** This function does the work for both the xDisconnect and xDestroy methods. +** These tables have no persistent representation of their own, so xDisconnect +** and xDestroy are identical operations. +*/ +static int fts3tokDisconnectMethod(sqlite3_vtab *pVtab){ + Fts3tokTable *pTab = (Fts3tokTable *)pVtab; + + pTab->pMod->xDestroy(pTab->pTok); + sqlite3_free(pTab); + return SQLITE_OK; +} + +/* +** xBestIndex - Analyze a WHERE and ORDER BY clause. +*/ +static int fts3tokBestIndexMethod( + sqlite3_vtab *pVTab, + sqlite3_index_info *pInfo +){ + int i; + UNUSED_PARAMETER(pVTab); + + for(i=0; inConstraint; i++){ + if( pInfo->aConstraint[i].usable + && pInfo->aConstraint[i].iColumn==0 + && pInfo->aConstraint[i].op==SQLITE_INDEX_CONSTRAINT_EQ + ){ + pInfo->idxNum = 1; + pInfo->aConstraintUsage[i].argvIndex = 1; + pInfo->aConstraintUsage[i].omit = 1; + pInfo->estimatedCost = 1; + return SQLITE_OK; + } + } + + pInfo->idxNum = 0; + assert( pInfo->estimatedCost>1000000.0 ); + + return SQLITE_OK; +} + +/* +** xOpen - Open a cursor. +*/ +static int fts3tokOpenMethod(sqlite3_vtab *pVTab, sqlite3_vtab_cursor **ppCsr){ + Fts3tokCursor *pCsr; + UNUSED_PARAMETER(pVTab); + + pCsr = (Fts3tokCursor *)sqlite3_malloc(sizeof(Fts3tokCursor)); + if( pCsr==0 ){ + return SQLITE_NOMEM; + } + memset(pCsr, 0, sizeof(Fts3tokCursor)); + + *ppCsr = (sqlite3_vtab_cursor *)pCsr; + return SQLITE_OK; +} + +/* +** Reset the tokenizer cursor passed as the only argument. As if it had +** just been returned by fts3tokOpenMethod(). +*/ +static void fts3tokResetCursor(Fts3tokCursor *pCsr){ + if( pCsr->pCsr ){ + Fts3tokTable *pTab = (Fts3tokTable *)(pCsr->base.pVtab); + pTab->pMod->xClose(pCsr->pCsr); + pCsr->pCsr = 0; + } + sqlite3_free(pCsr->zInput); + pCsr->zInput = 0; + pCsr->zToken = 0; + pCsr->nToken = 0; + pCsr->iStart = 0; + pCsr->iEnd = 0; + pCsr->iPos = 0; + pCsr->iRowid = 0; +} + +/* +** xClose - Close a cursor. +*/ +static int fts3tokCloseMethod(sqlite3_vtab_cursor *pCursor){ + Fts3tokCursor *pCsr = (Fts3tokCursor *)pCursor; + + fts3tokResetCursor(pCsr); + sqlite3_free(pCsr); + return SQLITE_OK; +} + +/* +** xNext - Advance the cursor to the next row, if any. +*/ +static int fts3tokNextMethod(sqlite3_vtab_cursor *pCursor){ + Fts3tokCursor *pCsr = (Fts3tokCursor *)pCursor; + Fts3tokTable *pTab = (Fts3tokTable *)(pCursor->pVtab); + int rc; /* Return code */ + + pCsr->iRowid++; + rc = pTab->pMod->xNext(pCsr->pCsr, + &pCsr->zToken, &pCsr->nToken, + &pCsr->iStart, &pCsr->iEnd, &pCsr->iPos + ); + + if( rc!=SQLITE_OK ){ + fts3tokResetCursor(pCsr); + if( rc==SQLITE_DONE ) rc = SQLITE_OK; + } + + return rc; +} + +/* +** xFilter - Initialize a cursor to point at the start of its data. +*/ +static int fts3tokFilterMethod( + sqlite3_vtab_cursor *pCursor, /* The cursor used for this query */ + int idxNum, /* Strategy index */ + const char *idxStr, /* Unused */ + int nVal, /* Number of elements in apVal */ + sqlite3_value **apVal /* Arguments for the indexing scheme */ +){ + int rc = SQLITE_ERROR; + Fts3tokCursor *pCsr = (Fts3tokCursor *)pCursor; + Fts3tokTable *pTab = (Fts3tokTable *)(pCursor->pVtab); + UNUSED_PARAMETER(idxStr); + UNUSED_PARAMETER(nVal); + + fts3tokResetCursor(pCsr); + if( idxNum==1 ){ + const char *zByte = (const char *)sqlite3_value_text(apVal[0]); + int nByte = sqlite3_value_bytes(apVal[0]); + pCsr->zInput = sqlite3_malloc64(nByte+1); + if( pCsr->zInput==0 ){ + rc = SQLITE_NOMEM; + }else{ + memcpy(pCsr->zInput, zByte, nByte); + pCsr->zInput[nByte] = 0; + rc = pTab->pMod->xOpen(pTab->pTok, pCsr->zInput, nByte, &pCsr->pCsr); + if( rc==SQLITE_OK ){ + pCsr->pCsr->pTokenizer = pTab->pTok; + } + } + } + + if( rc!=SQLITE_OK ) return rc; + return fts3tokNextMethod(pCursor); +} + +/* +** xEof - Return true if the cursor is at EOF, or false otherwise. +*/ +static int fts3tokEofMethod(sqlite3_vtab_cursor *pCursor){ + Fts3tokCursor *pCsr = (Fts3tokCursor *)pCursor; + return (pCsr->zToken==0); +} + +/* +** xColumn - Return a column value. +*/ +static int fts3tokColumnMethod( + sqlite3_vtab_cursor *pCursor, /* Cursor to retrieve value from */ + sqlite3_context *pCtx, /* Context for sqlite3_result_xxx() calls */ + int iCol /* Index of column to read value from */ +){ + Fts3tokCursor *pCsr = (Fts3tokCursor *)pCursor; + + /* CREATE TABLE x(input, token, start, end, position) */ + switch( iCol ){ + case 0: + sqlite3_result_text(pCtx, pCsr->zInput, -1, SQLITE_TRANSIENT); + break; + case 1: + sqlite3_result_text(pCtx, pCsr->zToken, pCsr->nToken, SQLITE_TRANSIENT); + break; + case 2: + sqlite3_result_int(pCtx, pCsr->iStart); + break; + case 3: + sqlite3_result_int(pCtx, pCsr->iEnd); + break; + default: + assert( iCol==4 ); + sqlite3_result_int(pCtx, pCsr->iPos); + break; + } + return SQLITE_OK; +} + +/* +** xRowid - Return the current rowid for the cursor. +*/ +static int fts3tokRowidMethod( + sqlite3_vtab_cursor *pCursor, /* Cursor to retrieve value from */ + sqlite_int64 *pRowid /* OUT: Rowid value */ +){ + Fts3tokCursor *pCsr = (Fts3tokCursor *)pCursor; + *pRowid = (sqlite3_int64)pCsr->iRowid; + return SQLITE_OK; +} + +/* +** Register the fts3tok module with database connection db. Return SQLITE_OK +** if successful or an error code if sqlite3_create_module() fails. +*/ +SQLITE_PRIVATE int sqlite3Fts3InitTok(sqlite3 *db, Fts3Hash *pHash){ + static const sqlite3_module fts3tok_module = { + 0, /* iVersion */ + fts3tokConnectMethod, /* xCreate */ + fts3tokConnectMethod, /* xConnect */ + fts3tokBestIndexMethod, /* xBestIndex */ + fts3tokDisconnectMethod, /* xDisconnect */ + fts3tokDisconnectMethod, /* xDestroy */ + fts3tokOpenMethod, /* xOpen */ + fts3tokCloseMethod, /* xClose */ + fts3tokFilterMethod, /* xFilter */ + fts3tokNextMethod, /* xNext */ + fts3tokEofMethod, /* xEof */ + fts3tokColumnMethod, /* xColumn */ + fts3tokRowidMethod, /* xRowid */ + 0, /* xUpdate */ + 0, /* xBegin */ + 0, /* xSync */ + 0, /* xCommit */ + 0, /* xRollback */ + 0, /* xFindFunction */ + 0, /* xRename */ + 0, /* xSavepoint */ + 0, /* xRelease */ + 0, /* xRollbackTo */ + 0 /* xShadowName */ + }; + int rc; /* Return code */ + + rc = sqlite3_create_module(db, "fts3tokenize", &fts3tok_module, (void*)pHash); + return rc; +} + +#endif /* !defined(SQLITE_CORE) || defined(SQLITE_ENABLE_FTS3) */ + +/************** End of fts3_tokenize_vtab.c **********************************/ +/************** Begin file fts3_write.c **************************************/ +/* +** 2009 Oct 23 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +****************************************************************************** +** +** This file is part of the SQLite FTS3 extension module. Specifically, +** this file contains code to insert, update and delete rows from FTS3 +** tables. It also contains code to merge FTS3 b-tree segments. Some +** of the sub-routines used to merge segments are also used by the query +** code in fts3.c. +*/ + +/* #include "fts3Int.h" */ +#if !defined(SQLITE_CORE) || defined(SQLITE_ENABLE_FTS3) + +/* #include */ +/* #include */ +/* #include */ + + +#define FTS_MAX_APPENDABLE_HEIGHT 16 + +/* +** When full-text index nodes are loaded from disk, the buffer that they +** are loaded into has the following number of bytes of padding at the end +** of it. i.e. if a full-text index node is 900 bytes in size, then a buffer +** of 920 bytes is allocated for it. +** +** This means that if we have a pointer into a buffer containing node data, +** it is always safe to read up to two varints from it without risking an +** overread, even if the node data is corrupted. +*/ +#define FTS3_NODE_PADDING (FTS3_VARINT_MAX*2) + +/* +** Under certain circumstances, b-tree nodes (doclists) can be loaded into +** memory incrementally instead of all at once. This can be a big performance +** win (reduced IO and CPU) if SQLite stops calling the virtual table xNext() +** method before retrieving all query results (as may happen, for example, +** if a query has a LIMIT clause). +** +** Incremental loading is used for b-tree nodes FTS3_NODE_CHUNK_THRESHOLD +** bytes and larger. Nodes are loaded in chunks of FTS3_NODE_CHUNKSIZE bytes. +** The code is written so that the hard lower-limit for each of these values +** is 1. Clearly such small values would be inefficient, but can be useful +** for testing purposes. +** +** If this module is built with SQLITE_TEST defined, these constants may +** be overridden at runtime for testing purposes. File fts3_test.c contains +** a Tcl interface to read and write the values. +*/ +#ifdef SQLITE_TEST +int test_fts3_node_chunksize = (4*1024); +int test_fts3_node_chunk_threshold = (4*1024)*4; +# define FTS3_NODE_CHUNKSIZE test_fts3_node_chunksize +# define FTS3_NODE_CHUNK_THRESHOLD test_fts3_node_chunk_threshold +#else +# define FTS3_NODE_CHUNKSIZE (4*1024) +# define FTS3_NODE_CHUNK_THRESHOLD (FTS3_NODE_CHUNKSIZE*4) +#endif + +/* +** The two values that may be meaningfully bound to the :1 parameter in +** statements SQL_REPLACE_STAT and SQL_SELECT_STAT. +*/ +#define FTS_STAT_DOCTOTAL 0 +#define FTS_STAT_INCRMERGEHINT 1 +#define FTS_STAT_AUTOINCRMERGE 2 + +/* +** If FTS_LOG_MERGES is defined, call sqlite3_log() to report each automatic +** and incremental merge operation that takes place. This is used for +** debugging FTS only, it should not usually be turned on in production +** systems. +*/ +#ifdef FTS3_LOG_MERGES +static void fts3LogMerge(int nMerge, sqlite3_int64 iAbsLevel){ + sqlite3_log(SQLITE_OK, "%d-way merge from level %d", nMerge, (int)iAbsLevel); +} +#else +#define fts3LogMerge(x, y) +#endif + + +typedef struct PendingList PendingList; +typedef struct SegmentNode SegmentNode; +typedef struct SegmentWriter SegmentWriter; + +/* +** An instance of the following data structure is used to build doclists +** incrementally. See function fts3PendingListAppend() for details. +*/ +struct PendingList { + int nData; + char *aData; + int nSpace; + sqlite3_int64 iLastDocid; + sqlite3_int64 iLastCol; + sqlite3_int64 iLastPos; +}; + + +/* +** Each cursor has a (possibly empty) linked list of the following objects. +*/ +struct Fts3DeferredToken { + Fts3PhraseToken *pToken; /* Pointer to corresponding expr token */ + int iCol; /* Column token must occur in */ + Fts3DeferredToken *pNext; /* Next in list of deferred tokens */ + PendingList *pList; /* Doclist is assembled here */ +}; + +/* +** An instance of this structure is used to iterate through the terms on +** a contiguous set of segment b-tree leaf nodes. Although the details of +** this structure are only manipulated by code in this file, opaque handles +** of type Fts3SegReader* are also used by code in fts3.c to iterate through +** terms when querying the full-text index. See functions: +** +** sqlite3Fts3SegReaderNew() +** sqlite3Fts3SegReaderFree() +** sqlite3Fts3SegReaderIterate() +** +** Methods used to manipulate Fts3SegReader structures: +** +** fts3SegReaderNext() +** fts3SegReaderFirstDocid() +** fts3SegReaderNextDocid() +*/ +struct Fts3SegReader { + int iIdx; /* Index within level, or 0x7FFFFFFF for PT */ + u8 bLookup; /* True for a lookup only */ + u8 rootOnly; /* True for a root-only reader */ + + sqlite3_int64 iStartBlock; /* Rowid of first leaf block to traverse */ + sqlite3_int64 iLeafEndBlock; /* Rowid of final leaf block to traverse */ + sqlite3_int64 iEndBlock; /* Rowid of final block in segment (or 0) */ + sqlite3_int64 iCurrentBlock; /* Current leaf block (or 0) */ + + char *aNode; /* Pointer to node data (or NULL) */ + int nNode; /* Size of buffer at aNode (or 0) */ + int nPopulate; /* If >0, bytes of buffer aNode[] loaded */ + sqlite3_blob *pBlob; /* If not NULL, blob handle to read node */ + + Fts3HashElem **ppNextElem; + + /* Variables set by fts3SegReaderNext(). These may be read directly + ** by the caller. They are valid from the time SegmentReaderNew() returns + ** until SegmentReaderNext() returns something other than SQLITE_OK + ** (i.e. SQLITE_DONE). + */ + int nTerm; /* Number of bytes in current term */ + char *zTerm; /* Pointer to current term */ + int nTermAlloc; /* Allocated size of zTerm buffer */ + char *aDoclist; /* Pointer to doclist of current entry */ + int nDoclist; /* Size of doclist in current entry */ + + /* The following variables are used by fts3SegReaderNextDocid() to iterate + ** through the current doclist (aDoclist/nDoclist). + */ + char *pOffsetList; + int nOffsetList; /* For descending pending seg-readers only */ + sqlite3_int64 iDocid; +}; + +#define fts3SegReaderIsPending(p) ((p)->ppNextElem!=0) +#define fts3SegReaderIsRootOnly(p) ((p)->rootOnly!=0) + +/* +** An instance of this structure is used to create a segment b-tree in the +** database. The internal details of this type are only accessed by the +** following functions: +** +** fts3SegWriterAdd() +** fts3SegWriterFlush() +** fts3SegWriterFree() +*/ +struct SegmentWriter { + SegmentNode *pTree; /* Pointer to interior tree structure */ + sqlite3_int64 iFirst; /* First slot in %_segments written */ + sqlite3_int64 iFree; /* Next free slot in %_segments */ + char *zTerm; /* Pointer to previous term buffer */ + int nTerm; /* Number of bytes in zTerm */ + int nMalloc; /* Size of malloc'd buffer at zMalloc */ + char *zMalloc; /* Malloc'd space (possibly) used for zTerm */ + int nSize; /* Size of allocation at aData */ + int nData; /* Bytes of data in aData */ + char *aData; /* Pointer to block from malloc() */ + i64 nLeafData; /* Number of bytes of leaf data written */ +}; + +/* +** Type SegmentNode is used by the following three functions to create +** the interior part of the segment b+-tree structures (everything except +** the leaf nodes). These functions and type are only ever used by code +** within the fts3SegWriterXXX() family of functions described above. +** +** fts3NodeAddTerm() +** fts3NodeWrite() +** fts3NodeFree() +** +** When a b+tree is written to the database (either as a result of a merge +** or the pending-terms table being flushed), leaves are written into the +** database file as soon as they are completely populated. The interior of +** the tree is assembled in memory and written out only once all leaves have +** been populated and stored. This is Ok, as the b+-tree fanout is usually +** very large, meaning that the interior of the tree consumes relatively +** little memory. +*/ +struct SegmentNode { + SegmentNode *pParent; /* Parent node (or NULL for root node) */ + SegmentNode *pRight; /* Pointer to right-sibling */ + SegmentNode *pLeftmost; /* Pointer to left-most node of this depth */ + int nEntry; /* Number of terms written to node so far */ + char *zTerm; /* Pointer to previous term buffer */ + int nTerm; /* Number of bytes in zTerm */ + int nMalloc; /* Size of malloc'd buffer at zMalloc */ + char *zMalloc; /* Malloc'd space (possibly) used for zTerm */ + int nData; /* Bytes of valid data so far */ + char *aData; /* Node data */ +}; + +/* +** Valid values for the second argument to fts3SqlStmt(). +*/ +#define SQL_DELETE_CONTENT 0 +#define SQL_IS_EMPTY 1 +#define SQL_DELETE_ALL_CONTENT 2 +#define SQL_DELETE_ALL_SEGMENTS 3 +#define SQL_DELETE_ALL_SEGDIR 4 +#define SQL_DELETE_ALL_DOCSIZE 5 +#define SQL_DELETE_ALL_STAT 6 +#define SQL_SELECT_CONTENT_BY_ROWID 7 +#define SQL_NEXT_SEGMENT_INDEX 8 +#define SQL_INSERT_SEGMENTS 9 +#define SQL_NEXT_SEGMENTS_ID 10 +#define SQL_INSERT_SEGDIR 11 +#define SQL_SELECT_LEVEL 12 +#define SQL_SELECT_LEVEL_RANGE 13 +#define SQL_SELECT_LEVEL_COUNT 14 +#define SQL_SELECT_SEGDIR_MAX_LEVEL 15 +#define SQL_DELETE_SEGDIR_LEVEL 16 +#define SQL_DELETE_SEGMENTS_RANGE 17 +#define SQL_CONTENT_INSERT 18 +#define SQL_DELETE_DOCSIZE 19 +#define SQL_REPLACE_DOCSIZE 20 +#define SQL_SELECT_DOCSIZE 21 +#define SQL_SELECT_STAT 22 +#define SQL_REPLACE_STAT 23 + +#define SQL_SELECT_ALL_PREFIX_LEVEL 24 +#define SQL_DELETE_ALL_TERMS_SEGDIR 25 +#define SQL_DELETE_SEGDIR_RANGE 26 +#define SQL_SELECT_ALL_LANGID 27 +#define SQL_FIND_MERGE_LEVEL 28 +#define SQL_MAX_LEAF_NODE_ESTIMATE 29 +#define SQL_DELETE_SEGDIR_ENTRY 30 +#define SQL_SHIFT_SEGDIR_ENTRY 31 +#define SQL_SELECT_SEGDIR 32 +#define SQL_CHOMP_SEGDIR 33 +#define SQL_SEGMENT_IS_APPENDABLE 34 +#define SQL_SELECT_INDEXES 35 +#define SQL_SELECT_MXLEVEL 36 + +#define SQL_SELECT_LEVEL_RANGE2 37 +#define SQL_UPDATE_LEVEL_IDX 38 +#define SQL_UPDATE_LEVEL 39 + +/* +** This function is used to obtain an SQLite prepared statement handle +** for the statement identified by the second argument. If successful, +** *pp is set to the requested statement handle and SQLITE_OK returned. +** Otherwise, an SQLite error code is returned and *pp is set to 0. +** +** If argument apVal is not NULL, then it must point to an array with +** at least as many entries as the requested statement has bound +** parameters. The values are bound to the statements parameters before +** returning. +*/ +static int fts3SqlStmt( + Fts3Table *p, /* Virtual table handle */ + int eStmt, /* One of the SQL_XXX constants above */ + sqlite3_stmt **pp, /* OUT: Statement handle */ + sqlite3_value **apVal /* Values to bind to statement */ +){ + const char *azSql[] = { +/* 0 */ "DELETE FROM %Q.'%q_content' WHERE rowid = ?", +/* 1 */ "SELECT NOT EXISTS(SELECT docid FROM %Q.'%q_content' WHERE rowid!=?)", +/* 2 */ "DELETE FROM %Q.'%q_content'", +/* 3 */ "DELETE FROM %Q.'%q_segments'", +/* 4 */ "DELETE FROM %Q.'%q_segdir'", +/* 5 */ "DELETE FROM %Q.'%q_docsize'", +/* 6 */ "DELETE FROM %Q.'%q_stat'", +/* 7 */ "SELECT %s WHERE rowid=?", +/* 8 */ "SELECT (SELECT max(idx) FROM %Q.'%q_segdir' WHERE level = ?) + 1", +/* 9 */ "REPLACE INTO %Q.'%q_segments'(blockid, block) VALUES(?, ?)", +/* 10 */ "SELECT coalesce((SELECT max(blockid) FROM %Q.'%q_segments') + 1, 1)", +/* 11 */ "REPLACE INTO %Q.'%q_segdir' VALUES(?,?,?,?,?,?)", + + /* Return segments in order from oldest to newest.*/ +/* 12 */ "SELECT idx, start_block, leaves_end_block, end_block, root " + "FROM %Q.'%q_segdir' WHERE level = ? ORDER BY idx ASC", +/* 13 */ "SELECT idx, start_block, leaves_end_block, end_block, root " + "FROM %Q.'%q_segdir' WHERE level BETWEEN ? AND ?" + "ORDER BY level DESC, idx ASC", + +/* 14 */ "SELECT count(*) FROM %Q.'%q_segdir' WHERE level = ?", +/* 15 */ "SELECT max(level) FROM %Q.'%q_segdir' WHERE level BETWEEN ? AND ?", + +/* 16 */ "DELETE FROM %Q.'%q_segdir' WHERE level = ?", +/* 17 */ "DELETE FROM %Q.'%q_segments' WHERE blockid BETWEEN ? AND ?", +/* 18 */ "INSERT INTO %Q.'%q_content' VALUES(%s)", +/* 19 */ "DELETE FROM %Q.'%q_docsize' WHERE docid = ?", +/* 20 */ "REPLACE INTO %Q.'%q_docsize' VALUES(?,?)", +/* 21 */ "SELECT size FROM %Q.'%q_docsize' WHERE docid=?", +/* 22 */ "SELECT value FROM %Q.'%q_stat' WHERE id=?", +/* 23 */ "REPLACE INTO %Q.'%q_stat' VALUES(?,?)", +/* 24 */ "", +/* 25 */ "", + +/* 26 */ "DELETE FROM %Q.'%q_segdir' WHERE level BETWEEN ? AND ?", +/* 27 */ "SELECT ? UNION SELECT level / (1024 * ?) FROM %Q.'%q_segdir'", + +/* This statement is used to determine which level to read the input from +** when performing an incremental merge. It returns the absolute level number +** of the oldest level in the db that contains at least ? segments. Or, +** if no level in the FTS index contains more than ? segments, the statement +** returns zero rows. */ +/* 28 */ "SELECT level, count(*) AS cnt FROM %Q.'%q_segdir' " + " GROUP BY level HAVING cnt>=?" + " ORDER BY (level %% 1024) ASC LIMIT 1", + +/* Estimate the upper limit on the number of leaf nodes in a new segment +** created by merging the oldest :2 segments from absolute level :1. See +** function sqlite3Fts3Incrmerge() for details. */ +/* 29 */ "SELECT 2 * total(1 + leaves_end_block - start_block) " + " FROM %Q.'%q_segdir' WHERE level = ? AND idx < ?", + +/* SQL_DELETE_SEGDIR_ENTRY +** Delete the %_segdir entry on absolute level :1 with index :2. */ +/* 30 */ "DELETE FROM %Q.'%q_segdir' WHERE level = ? AND idx = ?", + +/* SQL_SHIFT_SEGDIR_ENTRY +** Modify the idx value for the segment with idx=:3 on absolute level :2 +** to :1. */ +/* 31 */ "UPDATE %Q.'%q_segdir' SET idx = ? WHERE level=? AND idx=?", + +/* SQL_SELECT_SEGDIR +** Read a single entry from the %_segdir table. The entry from absolute +** level :1 with index value :2. */ +/* 32 */ "SELECT idx, start_block, leaves_end_block, end_block, root " + "FROM %Q.'%q_segdir' WHERE level = ? AND idx = ?", + +/* SQL_CHOMP_SEGDIR +** Update the start_block (:1) and root (:2) fields of the %_segdir +** entry located on absolute level :3 with index :4. */ +/* 33 */ "UPDATE %Q.'%q_segdir' SET start_block = ?, root = ?" + "WHERE level = ? AND idx = ?", + +/* SQL_SEGMENT_IS_APPENDABLE +** Return a single row if the segment with end_block=? is appendable. Or +** no rows otherwise. */ +/* 34 */ "SELECT 1 FROM %Q.'%q_segments' WHERE blockid=? AND block IS NULL", + +/* SQL_SELECT_INDEXES +** Return the list of valid segment indexes for absolute level ? */ +/* 35 */ "SELECT idx FROM %Q.'%q_segdir' WHERE level=? ORDER BY 1 ASC", + +/* SQL_SELECT_MXLEVEL +** Return the largest relative level in the FTS index or indexes. */ +/* 36 */ "SELECT max( level %% 1024 ) FROM %Q.'%q_segdir'", + + /* Return segments in order from oldest to newest.*/ +/* 37 */ "SELECT level, idx, end_block " + "FROM %Q.'%q_segdir' WHERE level BETWEEN ? AND ? " + "ORDER BY level DESC, idx ASC", + + /* Update statements used while promoting segments */ +/* 38 */ "UPDATE OR FAIL %Q.'%q_segdir' SET level=-1,idx=? " + "WHERE level=? AND idx=?", +/* 39 */ "UPDATE OR FAIL %Q.'%q_segdir' SET level=? WHERE level=-1" + + }; + int rc = SQLITE_OK; + sqlite3_stmt *pStmt; + + assert( SizeofArray(azSql)==SizeofArray(p->aStmt) ); + assert( eStmt=0 ); + + pStmt = p->aStmt[eStmt]; + if( !pStmt ){ + int f = SQLITE_PREPARE_PERSISTENT|SQLITE_PREPARE_NO_VTAB; + char *zSql; + if( eStmt==SQL_CONTENT_INSERT ){ + zSql = sqlite3_mprintf(azSql[eStmt], p->zDb, p->zName, p->zWriteExprlist); + }else if( eStmt==SQL_SELECT_CONTENT_BY_ROWID ){ + f &= ~SQLITE_PREPARE_NO_VTAB; + zSql = sqlite3_mprintf(azSql[eStmt], p->zReadExprlist); + }else{ + zSql = sqlite3_mprintf(azSql[eStmt], p->zDb, p->zName); + } + if( !zSql ){ + rc = SQLITE_NOMEM; + }else{ + rc = sqlite3_prepare_v3(p->db, zSql, -1, f, &pStmt, NULL); + sqlite3_free(zSql); + assert( rc==SQLITE_OK || pStmt==0 ); + p->aStmt[eStmt] = pStmt; + } + } + if( apVal ){ + int i; + int nParam = sqlite3_bind_parameter_count(pStmt); + for(i=0; rc==SQLITE_OK && inPendingData==0 ){ + sqlite3_stmt *pStmt; + rc = fts3SqlStmt(p, SQL_DELETE_SEGDIR_LEVEL, &pStmt, 0); + if( rc==SQLITE_OK ){ + sqlite3_bind_null(pStmt, 1); + sqlite3_step(pStmt); + rc = sqlite3_reset(pStmt); + } + } + + return rc; +} + +/* +** FTS maintains a separate indexes for each language-id (a 32-bit integer). +** Within each language id, a separate index is maintained to store the +** document terms, and each configured prefix size (configured the FTS +** "prefix=" option). And each index consists of multiple levels ("relative +** levels"). +** +** All three of these values (the language id, the specific index and the +** level within the index) are encoded in 64-bit integer values stored +** in the %_segdir table on disk. This function is used to convert three +** separate component values into the single 64-bit integer value that +** can be used to query the %_segdir table. +** +** Specifically, each language-id/index combination is allocated 1024 +** 64-bit integer level values ("absolute levels"). The main terms index +** for language-id 0 is allocate values 0-1023. The first prefix index +** (if any) for language-id 0 is allocated values 1024-2047. And so on. +** Language 1 indexes are allocated immediately following language 0. +** +** So, for a system with nPrefix prefix indexes configured, the block of +** absolute levels that corresponds to language-id iLangid and index +** iIndex starts at absolute level ((iLangid * (nPrefix+1) + iIndex) * 1024). +*/ +static sqlite3_int64 getAbsoluteLevel( + Fts3Table *p, /* FTS3 table handle */ + int iLangid, /* Language id */ + int iIndex, /* Index in p->aIndex[] */ + int iLevel /* Level of segments */ +){ + sqlite3_int64 iBase; /* First absolute level for iLangid/iIndex */ + assert_fts3_nc( iLangid>=0 ); + assert( p->nIndex>0 ); + assert( iIndex>=0 && iIndexnIndex ); + + iBase = ((sqlite3_int64)iLangid * p->nIndex + iIndex) * FTS3_SEGDIR_MAXLEVEL; + return iBase + iLevel; +} + +/* +** Set *ppStmt to a statement handle that may be used to iterate through +** all rows in the %_segdir table, from oldest to newest. If successful, +** return SQLITE_OK. If an error occurs while preparing the statement, +** return an SQLite error code. +** +** There is only ever one instance of this SQL statement compiled for +** each FTS3 table. +** +** The statement returns the following columns from the %_segdir table: +** +** 0: idx +** 1: start_block +** 2: leaves_end_block +** 3: end_block +** 4: root +*/ +SQLITE_PRIVATE int sqlite3Fts3AllSegdirs( + Fts3Table *p, /* FTS3 table */ + int iLangid, /* Language being queried */ + int iIndex, /* Index for p->aIndex[] */ + int iLevel, /* Level to select (relative level) */ + sqlite3_stmt **ppStmt /* OUT: Compiled statement */ +){ + int rc; + sqlite3_stmt *pStmt = 0; + + assert( iLevel==FTS3_SEGCURSOR_ALL || iLevel>=0 ); + assert( iLevel=0 && iIndexnIndex ); + + if( iLevel<0 ){ + /* "SELECT * FROM %_segdir WHERE level BETWEEN ? AND ? ORDER BY ..." */ + rc = fts3SqlStmt(p, SQL_SELECT_LEVEL_RANGE, &pStmt, 0); + if( rc==SQLITE_OK ){ + sqlite3_bind_int64(pStmt, 1, getAbsoluteLevel(p, iLangid, iIndex, 0)); + sqlite3_bind_int64(pStmt, 2, + getAbsoluteLevel(p, iLangid, iIndex, FTS3_SEGDIR_MAXLEVEL-1) + ); + } + }else{ + /* "SELECT * FROM %_segdir WHERE level = ? ORDER BY ..." */ + rc = fts3SqlStmt(p, SQL_SELECT_LEVEL, &pStmt, 0); + if( rc==SQLITE_OK ){ + sqlite3_bind_int64(pStmt, 1, getAbsoluteLevel(p, iLangid, iIndex,iLevel)); + } + } + *ppStmt = pStmt; + return rc; +} + + +/* +** Append a single varint to a PendingList buffer. SQLITE_OK is returned +** if successful, or an SQLite error code otherwise. +** +** This function also serves to allocate the PendingList structure itself. +** For example, to create a new PendingList structure containing two +** varints: +** +** PendingList *p = 0; +** fts3PendingListAppendVarint(&p, 1); +** fts3PendingListAppendVarint(&p, 2); +*/ +static int fts3PendingListAppendVarint( + PendingList **pp, /* IN/OUT: Pointer to PendingList struct */ + sqlite3_int64 i /* Value to append to data */ +){ + PendingList *p = *pp; + + /* Allocate or grow the PendingList as required. */ + if( !p ){ + p = sqlite3_malloc(sizeof(*p) + 100); + if( !p ){ + return SQLITE_NOMEM; + } + p->nSpace = 100; + p->aData = (char *)&p[1]; + p->nData = 0; + } + else if( p->nData+FTS3_VARINT_MAX+1>p->nSpace ){ + int nNew = p->nSpace * 2; + p = sqlite3_realloc(p, sizeof(*p) + nNew); + if( !p ){ + sqlite3_free(*pp); + *pp = 0; + return SQLITE_NOMEM; + } + p->nSpace = nNew; + p->aData = (char *)&p[1]; + } + + /* Append the new serialized varint to the end of the list. */ + p->nData += sqlite3Fts3PutVarint(&p->aData[p->nData], i); + p->aData[p->nData] = '\0'; + *pp = p; + return SQLITE_OK; +} + +/* +** Add a docid/column/position entry to a PendingList structure. Non-zero +** is returned if the structure is sqlite3_realloced as part of adding +** the entry. Otherwise, zero. +** +** If an OOM error occurs, *pRc is set to SQLITE_NOMEM before returning. +** Zero is always returned in this case. Otherwise, if no OOM error occurs, +** it is set to SQLITE_OK. +*/ +static int fts3PendingListAppend( + PendingList **pp, /* IN/OUT: PendingList structure */ + sqlite3_int64 iDocid, /* Docid for entry to add */ + sqlite3_int64 iCol, /* Column for entry to add */ + sqlite3_int64 iPos, /* Position of term for entry to add */ + int *pRc /* OUT: Return code */ +){ + PendingList *p = *pp; + int rc = SQLITE_OK; + + assert( !p || p->iLastDocid<=iDocid ); + + if( !p || p->iLastDocid!=iDocid ){ + sqlite3_int64 iDelta = iDocid - (p ? p->iLastDocid : 0); + if( p ){ + assert( p->nDatanSpace ); + assert( p->aData[p->nData]==0 ); + p->nData++; + } + if( SQLITE_OK!=(rc = fts3PendingListAppendVarint(&p, iDelta)) ){ + goto pendinglistappend_out; + } + p->iLastCol = -1; + p->iLastPos = 0; + p->iLastDocid = iDocid; + } + if( iCol>0 && p->iLastCol!=iCol ){ + if( SQLITE_OK!=(rc = fts3PendingListAppendVarint(&p, 1)) + || SQLITE_OK!=(rc = fts3PendingListAppendVarint(&p, iCol)) + ){ + goto pendinglistappend_out; + } + p->iLastCol = iCol; + p->iLastPos = 0; + } + if( iCol>=0 ){ + assert( iPos>p->iLastPos || (iPos==0 && p->iLastPos==0) ); + rc = fts3PendingListAppendVarint(&p, 2+iPos-p->iLastPos); + if( rc==SQLITE_OK ){ + p->iLastPos = iPos; + } + } + + pendinglistappend_out: + *pRc = rc; + if( p!=*pp ){ + *pp = p; + return 1; + } + return 0; +} + +/* +** Free a PendingList object allocated by fts3PendingListAppend(). +*/ +static void fts3PendingListDelete(PendingList *pList){ + sqlite3_free(pList); +} + +/* +** Add an entry to one of the pending-terms hash tables. +*/ +static int fts3PendingTermsAddOne( + Fts3Table *p, + int iCol, + int iPos, + Fts3Hash *pHash, /* Pending terms hash table to add entry to */ + const char *zToken, + int nToken +){ + PendingList *pList; + int rc = SQLITE_OK; + + pList = (PendingList *)fts3HashFind(pHash, zToken, nToken); + if( pList ){ + p->nPendingData -= (pList->nData + nToken + sizeof(Fts3HashElem)); + } + if( fts3PendingListAppend(&pList, p->iPrevDocid, iCol, iPos, &rc) ){ + if( pList==fts3HashInsert(pHash, zToken, nToken, pList) ){ + /* Malloc failed while inserting the new entry. This can only + ** happen if there was no previous entry for this token. + */ + assert( 0==fts3HashFind(pHash, zToken, nToken) ); + sqlite3_free(pList); + rc = SQLITE_NOMEM; + } + } + if( rc==SQLITE_OK ){ + p->nPendingData += (pList->nData + nToken + sizeof(Fts3HashElem)); + } + return rc; +} + +/* +** Tokenize the nul-terminated string zText and add all tokens to the +** pending-terms hash-table. The docid used is that currently stored in +** p->iPrevDocid, and the column is specified by argument iCol. +** +** If successful, SQLITE_OK is returned. Otherwise, an SQLite error code. +*/ +static int fts3PendingTermsAdd( + Fts3Table *p, /* Table into which text will be inserted */ + int iLangid, /* Language id to use */ + const char *zText, /* Text of document to be inserted */ + int iCol, /* Column into which text is being inserted */ + u32 *pnWord /* IN/OUT: Incr. by number tokens inserted */ +){ + int rc; + int iStart = 0; + int iEnd = 0; + int iPos = 0; + int nWord = 0; + + char const *zToken; + int nToken = 0; + + sqlite3_tokenizer *pTokenizer = p->pTokenizer; + sqlite3_tokenizer_module const *pModule = pTokenizer->pModule; + sqlite3_tokenizer_cursor *pCsr; + int (*xNext)(sqlite3_tokenizer_cursor *pCursor, + const char**,int*,int*,int*,int*); + + assert( pTokenizer && pModule ); + + /* If the user has inserted a NULL value, this function may be called with + ** zText==0. In this case, add zero token entries to the hash table and + ** return early. */ + if( zText==0 ){ + *pnWord = 0; + return SQLITE_OK; + } + + rc = sqlite3Fts3OpenTokenizer(pTokenizer, iLangid, zText, -1, &pCsr); + if( rc!=SQLITE_OK ){ + return rc; + } + + xNext = pModule->xNext; + while( SQLITE_OK==rc + && SQLITE_OK==(rc = xNext(pCsr, &zToken, &nToken, &iStart, &iEnd, &iPos)) + ){ + int i; + if( iPos>=nWord ) nWord = iPos+1; + + /* Positions cannot be negative; we use -1 as a terminator internally. + ** Tokens must have a non-zero length. + */ + if( iPos<0 || !zToken || nToken<=0 ){ + rc = SQLITE_ERROR; + break; + } + + /* Add the term to the terms index */ + rc = fts3PendingTermsAddOne( + p, iCol, iPos, &p->aIndex[0].hPending, zToken, nToken + ); + + /* Add the term to each of the prefix indexes that it is not too + ** short for. */ + for(i=1; rc==SQLITE_OK && inIndex; i++){ + struct Fts3Index *pIndex = &p->aIndex[i]; + if( nTokennPrefix ) continue; + rc = fts3PendingTermsAddOne( + p, iCol, iPos, &pIndex->hPending, zToken, pIndex->nPrefix + ); + } + } + + pModule->xClose(pCsr); + *pnWord += nWord; + return (rc==SQLITE_DONE ? SQLITE_OK : rc); +} + +/* +** Calling this function indicates that subsequent calls to +** fts3PendingTermsAdd() are to add term/position-list pairs for the +** contents of the document with docid iDocid. +*/ +static int fts3PendingTermsDocid( + Fts3Table *p, /* Full-text table handle */ + int bDelete, /* True if this op is a delete */ + int iLangid, /* Language id of row being written */ + sqlite_int64 iDocid /* Docid of row being written */ +){ + assert( iLangid>=0 ); + assert( bDelete==1 || bDelete==0 ); + + /* TODO(shess) Explore whether partially flushing the buffer on + ** forced-flush would provide better performance. I suspect that if + ** we ordered the doclists by size and flushed the largest until the + ** buffer was half empty, that would let the less frequent terms + ** generate longer doclists. + */ + if( iDocidiPrevDocid + || (iDocid==p->iPrevDocid && p->bPrevDelete==0) + || p->iPrevLangid!=iLangid + || p->nPendingData>p->nMaxPendingData + ){ + int rc = sqlite3Fts3PendingTermsFlush(p); + if( rc!=SQLITE_OK ) return rc; + } + p->iPrevDocid = iDocid; + p->iPrevLangid = iLangid; + p->bPrevDelete = bDelete; + return SQLITE_OK; +} + +/* +** Discard the contents of the pending-terms hash tables. +*/ +SQLITE_PRIVATE void sqlite3Fts3PendingTermsClear(Fts3Table *p){ + int i; + for(i=0; inIndex; i++){ + Fts3HashElem *pElem; + Fts3Hash *pHash = &p->aIndex[i].hPending; + for(pElem=fts3HashFirst(pHash); pElem; pElem=fts3HashNext(pElem)){ + PendingList *pList = (PendingList *)fts3HashData(pElem); + fts3PendingListDelete(pList); + } + fts3HashClear(pHash); + } + p->nPendingData = 0; +} + +/* +** This function is called by the xUpdate() method as part of an INSERT +** operation. It adds entries for each term in the new record to the +** pendingTerms hash table. +** +** Argument apVal is the same as the similarly named argument passed to +** fts3InsertData(). Parameter iDocid is the docid of the new row. +*/ +static int fts3InsertTerms( + Fts3Table *p, + int iLangid, + sqlite3_value **apVal, + u32 *aSz +){ + int i; /* Iterator variable */ + for(i=2; inColumn+2; i++){ + int iCol = i-2; + if( p->abNotindexed[iCol]==0 ){ + const char *zText = (const char *)sqlite3_value_text(apVal[i]); + int rc = fts3PendingTermsAdd(p, iLangid, zText, iCol, &aSz[iCol]); + if( rc!=SQLITE_OK ){ + return rc; + } + aSz[p->nColumn] += sqlite3_value_bytes(apVal[i]); + } + } + return SQLITE_OK; +} + +/* +** This function is called by the xUpdate() method for an INSERT operation. +** The apVal parameter is passed a copy of the apVal argument passed by +** SQLite to the xUpdate() method. i.e: +** +** apVal[0] Not used for INSERT. +** apVal[1] rowid +** apVal[2] Left-most user-defined column +** ... +** apVal[p->nColumn+1] Right-most user-defined column +** apVal[p->nColumn+2] Hidden column with same name as table +** apVal[p->nColumn+3] Hidden "docid" column (alias for rowid) +** apVal[p->nColumn+4] Hidden languageid column +*/ +static int fts3InsertData( + Fts3Table *p, /* Full-text table */ + sqlite3_value **apVal, /* Array of values to insert */ + sqlite3_int64 *piDocid /* OUT: Docid for row just inserted */ +){ + int rc; /* Return code */ + sqlite3_stmt *pContentInsert; /* INSERT INTO %_content VALUES(...) */ + + if( p->zContentTbl ){ + sqlite3_value *pRowid = apVal[p->nColumn+3]; + if( sqlite3_value_type(pRowid)==SQLITE_NULL ){ + pRowid = apVal[1]; + } + if( sqlite3_value_type(pRowid)!=SQLITE_INTEGER ){ + return SQLITE_CONSTRAINT; + } + *piDocid = sqlite3_value_int64(pRowid); + return SQLITE_OK; + } + + /* Locate the statement handle used to insert data into the %_content + ** table. The SQL for this statement is: + ** + ** INSERT INTO %_content VALUES(?, ?, ?, ...) + ** + ** The statement features N '?' variables, where N is the number of user + ** defined columns in the FTS3 table, plus one for the docid field. + */ + rc = fts3SqlStmt(p, SQL_CONTENT_INSERT, &pContentInsert, &apVal[1]); + if( rc==SQLITE_OK && p->zLanguageid ){ + rc = sqlite3_bind_int( + pContentInsert, p->nColumn+2, + sqlite3_value_int(apVal[p->nColumn+4]) + ); + } + if( rc!=SQLITE_OK ) return rc; + + /* There is a quirk here. The users INSERT statement may have specified + ** a value for the "rowid" field, for the "docid" field, or for both. + ** Which is a problem, since "rowid" and "docid" are aliases for the + ** same value. For example: + ** + ** INSERT INTO fts3tbl(rowid, docid) VALUES(1, 2); + ** + ** In FTS3, this is an error. It is an error to specify non-NULL values + ** for both docid and some other rowid alias. + */ + if( SQLITE_NULL!=sqlite3_value_type(apVal[3+p->nColumn]) ){ + if( SQLITE_NULL==sqlite3_value_type(apVal[0]) + && SQLITE_NULL!=sqlite3_value_type(apVal[1]) + ){ + /* A rowid/docid conflict. */ + return SQLITE_ERROR; + } + rc = sqlite3_bind_value(pContentInsert, 1, apVal[3+p->nColumn]); + if( rc!=SQLITE_OK ) return rc; + } + + /* Execute the statement to insert the record. Set *piDocid to the + ** new docid value. + */ + sqlite3_step(pContentInsert); + rc = sqlite3_reset(pContentInsert); + + *piDocid = sqlite3_last_insert_rowid(p->db); + return rc; +} + + + +/* +** Remove all data from the FTS3 table. Clear the hash table containing +** pending terms. +*/ +static int fts3DeleteAll(Fts3Table *p, int bContent){ + int rc = SQLITE_OK; /* Return code */ + + /* Discard the contents of the pending-terms hash table. */ + sqlite3Fts3PendingTermsClear(p); + + /* Delete everything from the shadow tables. Except, leave %_content as + ** is if bContent is false. */ + assert( p->zContentTbl==0 || bContent==0 ); + if( bContent ) fts3SqlExec(&rc, p, SQL_DELETE_ALL_CONTENT, 0); + fts3SqlExec(&rc, p, SQL_DELETE_ALL_SEGMENTS, 0); + fts3SqlExec(&rc, p, SQL_DELETE_ALL_SEGDIR, 0); + if( p->bHasDocsize ){ + fts3SqlExec(&rc, p, SQL_DELETE_ALL_DOCSIZE, 0); + } + if( p->bHasStat ){ + fts3SqlExec(&rc, p, SQL_DELETE_ALL_STAT, 0); + } + return rc; +} + +/* +** +*/ +static int langidFromSelect(Fts3Table *p, sqlite3_stmt *pSelect){ + int iLangid = 0; + if( p->zLanguageid ) iLangid = sqlite3_column_int(pSelect, p->nColumn+1); + return iLangid; +} + +/* +** The first element in the apVal[] array is assumed to contain the docid +** (an integer) of a row about to be deleted. Remove all terms from the +** full-text index. +*/ +static void fts3DeleteTerms( + int *pRC, /* Result code */ + Fts3Table *p, /* The FTS table to delete from */ + sqlite3_value *pRowid, /* The docid to be deleted */ + u32 *aSz, /* Sizes of deleted document written here */ + int *pbFound /* OUT: Set to true if row really does exist */ +){ + int rc; + sqlite3_stmt *pSelect; + + assert( *pbFound==0 ); + if( *pRC ) return; + rc = fts3SqlStmt(p, SQL_SELECT_CONTENT_BY_ROWID, &pSelect, &pRowid); + if( rc==SQLITE_OK ){ + if( SQLITE_ROW==sqlite3_step(pSelect) ){ + int i; + int iLangid = langidFromSelect(p, pSelect); + i64 iDocid = sqlite3_column_int64(pSelect, 0); + rc = fts3PendingTermsDocid(p, 1, iLangid, iDocid); + for(i=1; rc==SQLITE_OK && i<=p->nColumn; i++){ + int iCol = i-1; + if( p->abNotindexed[iCol]==0 ){ + const char *zText = (const char *)sqlite3_column_text(pSelect, i); + rc = fts3PendingTermsAdd(p, iLangid, zText, -1, &aSz[iCol]); + aSz[p->nColumn] += sqlite3_column_bytes(pSelect, i); + } + } + if( rc!=SQLITE_OK ){ + sqlite3_reset(pSelect); + *pRC = rc; + return; + } + *pbFound = 1; + } + rc = sqlite3_reset(pSelect); + }else{ + sqlite3_reset(pSelect); + } + *pRC = rc; +} + +/* +** Forward declaration to account for the circular dependency between +** functions fts3SegmentMerge() and fts3AllocateSegdirIdx(). +*/ +static int fts3SegmentMerge(Fts3Table *, int, int, int); + +/* +** This function allocates a new level iLevel index in the segdir table. +** Usually, indexes are allocated within a level sequentially starting +** with 0, so the allocated index is one greater than the value returned +** by: +** +** SELECT max(idx) FROM %_segdir WHERE level = :iLevel +** +** However, if there are already FTS3_MERGE_COUNT indexes at the requested +** level, they are merged into a single level (iLevel+1) segment and the +** allocated index is 0. +** +** If successful, *piIdx is set to the allocated index slot and SQLITE_OK +** returned. Otherwise, an SQLite error code is returned. +*/ +static int fts3AllocateSegdirIdx( + Fts3Table *p, + int iLangid, /* Language id */ + int iIndex, /* Index for p->aIndex */ + int iLevel, + int *piIdx +){ + int rc; /* Return Code */ + sqlite3_stmt *pNextIdx; /* Query for next idx at level iLevel */ + int iNext = 0; /* Result of query pNextIdx */ + + assert( iLangid>=0 ); + assert( p->nIndex>=1 ); + + /* Set variable iNext to the next available segdir index at level iLevel. */ + rc = fts3SqlStmt(p, SQL_NEXT_SEGMENT_INDEX, &pNextIdx, 0); + if( rc==SQLITE_OK ){ + sqlite3_bind_int64( + pNextIdx, 1, getAbsoluteLevel(p, iLangid, iIndex, iLevel) + ); + if( SQLITE_ROW==sqlite3_step(pNextIdx) ){ + iNext = sqlite3_column_int(pNextIdx, 0); + } + rc = sqlite3_reset(pNextIdx); + } + + if( rc==SQLITE_OK ){ + /* If iNext is FTS3_MERGE_COUNT, indicating that level iLevel is already + ** full, merge all segments in level iLevel into a single iLevel+1 + ** segment and allocate (newly freed) index 0 at level iLevel. Otherwise, + ** if iNext is less than FTS3_MERGE_COUNT, allocate index iNext. + */ + if( iNext>=FTS3_MERGE_COUNT ){ + fts3LogMerge(16, getAbsoluteLevel(p, iLangid, iIndex, iLevel)); + rc = fts3SegmentMerge(p, iLangid, iIndex, iLevel); + *piIdx = 0; + }else{ + *piIdx = iNext; + } + } + + return rc; +} + +/* +** The %_segments table is declared as follows: +** +** CREATE TABLE %_segments(blockid INTEGER PRIMARY KEY, block BLOB) +** +** This function reads data from a single row of the %_segments table. The +** specific row is identified by the iBlockid parameter. If paBlob is not +** NULL, then a buffer is allocated using sqlite3_malloc() and populated +** with the contents of the blob stored in the "block" column of the +** identified table row is. Whether or not paBlob is NULL, *pnBlob is set +** to the size of the blob in bytes before returning. +** +** If an error occurs, or the table does not contain the specified row, +** an SQLite error code is returned. Otherwise, SQLITE_OK is returned. If +** paBlob is non-NULL, then it is the responsibility of the caller to +** eventually free the returned buffer. +** +** This function may leave an open sqlite3_blob* handle in the +** Fts3Table.pSegments variable. This handle is reused by subsequent calls +** to this function. The handle may be closed by calling the +** sqlite3Fts3SegmentsClose() function. Reusing a blob handle is a handy +** performance improvement, but the blob handle should always be closed +** before control is returned to the user (to prevent a lock being held +** on the database file for longer than necessary). Thus, any virtual table +** method (xFilter etc.) that may directly or indirectly call this function +** must call sqlite3Fts3SegmentsClose() before returning. +*/ +SQLITE_PRIVATE int sqlite3Fts3ReadBlock( + Fts3Table *p, /* FTS3 table handle */ + sqlite3_int64 iBlockid, /* Access the row with blockid=$iBlockid */ + char **paBlob, /* OUT: Blob data in malloc'd buffer */ + int *pnBlob, /* OUT: Size of blob data */ + int *pnLoad /* OUT: Bytes actually loaded */ +){ + int rc; /* Return code */ + + /* pnBlob must be non-NULL. paBlob may be NULL or non-NULL. */ + assert( pnBlob ); + + if( p->pSegments ){ + rc = sqlite3_blob_reopen(p->pSegments, iBlockid); + }else{ + if( 0==p->zSegmentsTbl ){ + p->zSegmentsTbl = sqlite3_mprintf("%s_segments", p->zName); + if( 0==p->zSegmentsTbl ) return SQLITE_NOMEM; + } + rc = sqlite3_blob_open( + p->db, p->zDb, p->zSegmentsTbl, "block", iBlockid, 0, &p->pSegments + ); + } + + if( rc==SQLITE_OK ){ + int nByte = sqlite3_blob_bytes(p->pSegments); + *pnBlob = nByte; + if( paBlob ){ + char *aByte = sqlite3_malloc(nByte + FTS3_NODE_PADDING); + if( !aByte ){ + rc = SQLITE_NOMEM; + }else{ + if( pnLoad && nByte>(FTS3_NODE_CHUNK_THRESHOLD) ){ + nByte = FTS3_NODE_CHUNKSIZE; + *pnLoad = nByte; + } + rc = sqlite3_blob_read(p->pSegments, aByte, nByte, 0); + memset(&aByte[nByte], 0, FTS3_NODE_PADDING); + if( rc!=SQLITE_OK ){ + sqlite3_free(aByte); + aByte = 0; + } + } + *paBlob = aByte; + } + } + + return rc; +} + +/* +** Close the blob handle at p->pSegments, if it is open. See comments above +** the sqlite3Fts3ReadBlock() function for details. +*/ +SQLITE_PRIVATE void sqlite3Fts3SegmentsClose(Fts3Table *p){ + sqlite3_blob_close(p->pSegments); + p->pSegments = 0; +} + +static int fts3SegReaderIncrRead(Fts3SegReader *pReader){ + int nRead; /* Number of bytes to read */ + int rc; /* Return code */ + + nRead = MIN(pReader->nNode - pReader->nPopulate, FTS3_NODE_CHUNKSIZE); + rc = sqlite3_blob_read( + pReader->pBlob, + &pReader->aNode[pReader->nPopulate], + nRead, + pReader->nPopulate + ); + + if( rc==SQLITE_OK ){ + pReader->nPopulate += nRead; + memset(&pReader->aNode[pReader->nPopulate], 0, FTS3_NODE_PADDING); + if( pReader->nPopulate==pReader->nNode ){ + sqlite3_blob_close(pReader->pBlob); + pReader->pBlob = 0; + pReader->nPopulate = 0; + } + } + return rc; +} + +static int fts3SegReaderRequire(Fts3SegReader *pReader, char *pFrom, int nByte){ + int rc = SQLITE_OK; + assert( !pReader->pBlob + || (pFrom>=pReader->aNode && pFrom<&pReader->aNode[pReader->nNode]) + ); + while( pReader->pBlob && rc==SQLITE_OK + && (pFrom - pReader->aNode + nByte)>pReader->nPopulate + ){ + rc = fts3SegReaderIncrRead(pReader); + } + return rc; +} + +/* +** Set an Fts3SegReader cursor to point at EOF. +*/ +static void fts3SegReaderSetEof(Fts3SegReader *pSeg){ + if( !fts3SegReaderIsRootOnly(pSeg) ){ + sqlite3_free(pSeg->aNode); + sqlite3_blob_close(pSeg->pBlob); + pSeg->pBlob = 0; + } + pSeg->aNode = 0; +} + +/* +** Move the iterator passed as the first argument to the next term in the +** segment. If successful, SQLITE_OK is returned. If there is no next term, +** SQLITE_DONE. Otherwise, an SQLite error code. +*/ +static int fts3SegReaderNext( + Fts3Table *p, + Fts3SegReader *pReader, + int bIncr +){ + int rc; /* Return code of various sub-routines */ + char *pNext; /* Cursor variable */ + int nPrefix; /* Number of bytes in term prefix */ + int nSuffix; /* Number of bytes in term suffix */ + + if( !pReader->aDoclist ){ + pNext = pReader->aNode; + }else{ + pNext = &pReader->aDoclist[pReader->nDoclist]; + } + + if( !pNext || pNext>=&pReader->aNode[pReader->nNode] ){ + + if( fts3SegReaderIsPending(pReader) ){ + Fts3HashElem *pElem = *(pReader->ppNextElem); + sqlite3_free(pReader->aNode); + pReader->aNode = 0; + if( pElem ){ + char *aCopy; + PendingList *pList = (PendingList *)fts3HashData(pElem); + int nCopy = pList->nData+1; + pReader->zTerm = (char *)fts3HashKey(pElem); + pReader->nTerm = fts3HashKeysize(pElem); + aCopy = (char*)sqlite3_malloc(nCopy); + if( !aCopy ) return SQLITE_NOMEM; + memcpy(aCopy, pList->aData, nCopy); + pReader->nNode = pReader->nDoclist = nCopy; + pReader->aNode = pReader->aDoclist = aCopy; + pReader->ppNextElem++; + assert( pReader->aNode ); + } + return SQLITE_OK; + } + + fts3SegReaderSetEof(pReader); + + /* If iCurrentBlock>=iLeafEndBlock, this is an EOF condition. All leaf + ** blocks have already been traversed. */ +#ifdef CORRUPT_DB + assert( pReader->iCurrentBlock<=pReader->iLeafEndBlock || CORRUPT_DB ); +#endif + if( pReader->iCurrentBlock>=pReader->iLeafEndBlock ){ + return SQLITE_OK; + } + + rc = sqlite3Fts3ReadBlock( + p, ++pReader->iCurrentBlock, &pReader->aNode, &pReader->nNode, + (bIncr ? &pReader->nPopulate : 0) + ); + if( rc!=SQLITE_OK ) return rc; + assert( pReader->pBlob==0 ); + if( bIncr && pReader->nPopulatenNode ){ + pReader->pBlob = p->pSegments; + p->pSegments = 0; + } + pNext = pReader->aNode; + } + + assert( !fts3SegReaderIsPending(pReader) ); + + rc = fts3SegReaderRequire(pReader, pNext, FTS3_VARINT_MAX*2); + if( rc!=SQLITE_OK ) return rc; + + /* Because of the FTS3_NODE_PADDING bytes of padding, the following is + ** safe (no risk of overread) even if the node data is corrupted. */ + pNext += fts3GetVarint32(pNext, &nPrefix); + pNext += fts3GetVarint32(pNext, &nSuffix); + if( nSuffix<=0 + || (&pReader->aNode[pReader->nNode] - pNext)pReader->nTermAlloc + ){ + return FTS_CORRUPT_VTAB; + } + + /* Both nPrefix and nSuffix were read by fts3GetVarint32() and so are + ** between 0 and 0x7FFFFFFF. But the sum of the two may cause integer + ** overflow - hence the (i64) casts. */ + if( (i64)nPrefix+nSuffix>(i64)pReader->nTermAlloc ){ + i64 nNew = ((i64)nPrefix+nSuffix)*2; + char *zNew = sqlite3_realloc64(pReader->zTerm, nNew); + if( !zNew ){ + return SQLITE_NOMEM; + } + pReader->zTerm = zNew; + pReader->nTermAlloc = nNew; + } + + rc = fts3SegReaderRequire(pReader, pNext, nSuffix+FTS3_VARINT_MAX); + if( rc!=SQLITE_OK ) return rc; + + memcpy(&pReader->zTerm[nPrefix], pNext, nSuffix); + pReader->nTerm = nPrefix+nSuffix; + pNext += nSuffix; + pNext += fts3GetVarint32(pNext, &pReader->nDoclist); + pReader->aDoclist = pNext; + pReader->pOffsetList = 0; + + /* Check that the doclist does not appear to extend past the end of the + ** b-tree node. And that the final byte of the doclist is 0x00. If either + ** of these statements is untrue, then the data structure is corrupt. + */ + if( pReader->nDoclist > pReader->nNode-(pReader->aDoclist-pReader->aNode) + || (pReader->nPopulate==0 && pReader->aDoclist[pReader->nDoclist-1]) + ){ + return FTS_CORRUPT_VTAB; + } + return SQLITE_OK; +} + +/* +** Set the SegReader to point to the first docid in the doclist associated +** with the current term. +*/ +static int fts3SegReaderFirstDocid(Fts3Table *pTab, Fts3SegReader *pReader){ + int rc = SQLITE_OK; + assert( pReader->aDoclist ); + assert( !pReader->pOffsetList ); + if( pTab->bDescIdx && fts3SegReaderIsPending(pReader) ){ + u8 bEof = 0; + pReader->iDocid = 0; + pReader->nOffsetList = 0; + sqlite3Fts3DoclistPrev(0, + pReader->aDoclist, pReader->nDoclist, &pReader->pOffsetList, + &pReader->iDocid, &pReader->nOffsetList, &bEof + ); + }else{ + rc = fts3SegReaderRequire(pReader, pReader->aDoclist, FTS3_VARINT_MAX); + if( rc==SQLITE_OK ){ + int n = sqlite3Fts3GetVarint(pReader->aDoclist, &pReader->iDocid); + pReader->pOffsetList = &pReader->aDoclist[n]; + } + } + return rc; +} + +/* +** Advance the SegReader to point to the next docid in the doclist +** associated with the current term. +** +** If arguments ppOffsetList and pnOffsetList are not NULL, then +** *ppOffsetList is set to point to the first column-offset list +** in the doclist entry (i.e. immediately past the docid varint). +** *pnOffsetList is set to the length of the set of column-offset +** lists, not including the nul-terminator byte. For example: +*/ +static int fts3SegReaderNextDocid( + Fts3Table *pTab, + Fts3SegReader *pReader, /* Reader to advance to next docid */ + char **ppOffsetList, /* OUT: Pointer to current position-list */ + int *pnOffsetList /* OUT: Length of *ppOffsetList in bytes */ +){ + int rc = SQLITE_OK; + char *p = pReader->pOffsetList; + char c = 0; + + assert( p ); + + if( pTab->bDescIdx && fts3SegReaderIsPending(pReader) ){ + /* A pending-terms seg-reader for an FTS4 table that uses order=desc. + ** Pending-terms doclists are always built up in ascending order, so + ** we have to iterate through them backwards here. */ + u8 bEof = 0; + if( ppOffsetList ){ + *ppOffsetList = pReader->pOffsetList; + *pnOffsetList = pReader->nOffsetList - 1; + } + sqlite3Fts3DoclistPrev(0, + pReader->aDoclist, pReader->nDoclist, &p, &pReader->iDocid, + &pReader->nOffsetList, &bEof + ); + if( bEof ){ + pReader->pOffsetList = 0; + }else{ + pReader->pOffsetList = p; + } + }else{ + char *pEnd = &pReader->aDoclist[pReader->nDoclist]; + + /* Pointer p currently points at the first byte of an offset list. The + ** following block advances it to point one byte past the end of + ** the same offset list. */ + while( 1 ){ + + /* The following line of code (and the "p++" below the while() loop) is + ** normally all that is required to move pointer p to the desired + ** position. The exception is if this node is being loaded from disk + ** incrementally and pointer "p" now points to the first byte past + ** the populated part of pReader->aNode[]. + */ + while( *p | c ) c = *p++ & 0x80; + assert( *p==0 ); + + if( pReader->pBlob==0 || p<&pReader->aNode[pReader->nPopulate] ) break; + rc = fts3SegReaderIncrRead(pReader); + if( rc!=SQLITE_OK ) return rc; + } + p++; + + /* If required, populate the output variables with a pointer to and the + ** size of the previous offset-list. + */ + if( ppOffsetList ){ + *ppOffsetList = pReader->pOffsetList; + *pnOffsetList = (int)(p - pReader->pOffsetList - 1); + } + + /* List may have been edited in place by fts3EvalNearTrim() */ + while( p=pEnd ){ + pReader->pOffsetList = 0; + }else{ + rc = fts3SegReaderRequire(pReader, p, FTS3_VARINT_MAX); + if( rc==SQLITE_OK ){ + sqlite3_int64 iDelta; + pReader->pOffsetList = p + sqlite3Fts3GetVarint(p, &iDelta); + if( pTab->bDescIdx ){ + pReader->iDocid -= iDelta; + }else{ + pReader->iDocid += iDelta; + } + } + } + } + + return SQLITE_OK; +} + + +SQLITE_PRIVATE int sqlite3Fts3MsrOvfl( + Fts3Cursor *pCsr, + Fts3MultiSegReader *pMsr, + int *pnOvfl +){ + Fts3Table *p = (Fts3Table*)pCsr->base.pVtab; + int nOvfl = 0; + int ii; + int rc = SQLITE_OK; + int pgsz = p->nPgsz; + + assert( p->bFts4 ); + assert( pgsz>0 ); + + for(ii=0; rc==SQLITE_OK && iinSegment; ii++){ + Fts3SegReader *pReader = pMsr->apSegment[ii]; + if( !fts3SegReaderIsPending(pReader) + && !fts3SegReaderIsRootOnly(pReader) + ){ + sqlite3_int64 jj; + for(jj=pReader->iStartBlock; jj<=pReader->iLeafEndBlock; jj++){ + int nBlob; + rc = sqlite3Fts3ReadBlock(p, jj, 0, &nBlob, 0); + if( rc!=SQLITE_OK ) break; + if( (nBlob+35)>pgsz ){ + nOvfl += (nBlob + 34)/pgsz; + } + } + } + } + *pnOvfl = nOvfl; + return rc; +} + +/* +** Free all allocations associated with the iterator passed as the +** second argument. +*/ +SQLITE_PRIVATE void sqlite3Fts3SegReaderFree(Fts3SegReader *pReader){ + if( pReader ){ + if( !fts3SegReaderIsPending(pReader) ){ + sqlite3_free(pReader->zTerm); + } + if( !fts3SegReaderIsRootOnly(pReader) ){ + sqlite3_free(pReader->aNode); + } + sqlite3_blob_close(pReader->pBlob); + } + sqlite3_free(pReader); +} + +/* +** Allocate a new SegReader object. +*/ +SQLITE_PRIVATE int sqlite3Fts3SegReaderNew( + int iAge, /* Segment "age". */ + int bLookup, /* True for a lookup only */ + sqlite3_int64 iStartLeaf, /* First leaf to traverse */ + sqlite3_int64 iEndLeaf, /* Final leaf to traverse */ + sqlite3_int64 iEndBlock, /* Final block of segment */ + const char *zRoot, /* Buffer containing root node */ + int nRoot, /* Size of buffer containing root node */ + Fts3SegReader **ppReader /* OUT: Allocated Fts3SegReader */ +){ + Fts3SegReader *pReader; /* Newly allocated SegReader object */ + int nExtra = 0; /* Bytes to allocate segment root node */ + + assert( zRoot!=0 || nRoot==0 ); +#ifdef CORRUPT_DB + assert( zRoot!=0 || CORRUPT_DB ); +#endif + + if( iStartLeaf==0 ){ + if( iEndLeaf!=0 ) return FTS_CORRUPT_VTAB; + nExtra = nRoot + FTS3_NODE_PADDING; + } + + pReader = (Fts3SegReader *)sqlite3_malloc(sizeof(Fts3SegReader) + nExtra); + if( !pReader ){ + return SQLITE_NOMEM; + } + memset(pReader, 0, sizeof(Fts3SegReader)); + pReader->iIdx = iAge; + pReader->bLookup = bLookup!=0; + pReader->iStartBlock = iStartLeaf; + pReader->iLeafEndBlock = iEndLeaf; + pReader->iEndBlock = iEndBlock; + + if( nExtra ){ + /* The entire segment is stored in the root node. */ + pReader->aNode = (char *)&pReader[1]; + pReader->rootOnly = 1; + pReader->nNode = nRoot; + if( nRoot ) memcpy(pReader->aNode, zRoot, nRoot); + memset(&pReader->aNode[nRoot], 0, FTS3_NODE_PADDING); + }else{ + pReader->iCurrentBlock = iStartLeaf-1; + } + *ppReader = pReader; + return SQLITE_OK; +} + +/* +** This is a comparison function used as a qsort() callback when sorting +** an array of pending terms by term. This occurs as part of flushing +** the contents of the pending-terms hash table to the database. +*/ +static int SQLITE_CDECL fts3CompareElemByTerm( + const void *lhs, + const void *rhs +){ + char *z1 = fts3HashKey(*(Fts3HashElem **)lhs); + char *z2 = fts3HashKey(*(Fts3HashElem **)rhs); + int n1 = fts3HashKeysize(*(Fts3HashElem **)lhs); + int n2 = fts3HashKeysize(*(Fts3HashElem **)rhs); + + int n = (n1aIndex */ + const char *zTerm, /* Term to search for */ + int nTerm, /* Size of buffer zTerm */ + int bPrefix, /* True for a prefix iterator */ + Fts3SegReader **ppReader /* OUT: SegReader for pending-terms */ +){ + Fts3SegReader *pReader = 0; /* Fts3SegReader object to return */ + Fts3HashElem *pE; /* Iterator variable */ + Fts3HashElem **aElem = 0; /* Array of term hash entries to scan */ + int nElem = 0; /* Size of array at aElem */ + int rc = SQLITE_OK; /* Return Code */ + Fts3Hash *pHash; + + pHash = &p->aIndex[iIndex].hPending; + if( bPrefix ){ + int nAlloc = 0; /* Size of allocated array at aElem */ + + for(pE=fts3HashFirst(pHash); pE; pE=fts3HashNext(pE)){ + char *zKey = (char *)fts3HashKey(pE); + int nKey = fts3HashKeysize(pE); + if( nTerm==0 || (nKey>=nTerm && 0==memcmp(zKey, zTerm, nTerm)) ){ + if( nElem==nAlloc ){ + Fts3HashElem **aElem2; + nAlloc += 16; + aElem2 = (Fts3HashElem **)sqlite3_realloc( + aElem, nAlloc*sizeof(Fts3HashElem *) + ); + if( !aElem2 ){ + rc = SQLITE_NOMEM; + nElem = 0; + break; + } + aElem = aElem2; + } + + aElem[nElem++] = pE; + } + } + + /* If more than one term matches the prefix, sort the Fts3HashElem + ** objects in term order using qsort(). This uses the same comparison + ** callback as is used when flushing terms to disk. + */ + if( nElem>1 ){ + qsort(aElem, nElem, sizeof(Fts3HashElem *), fts3CompareElemByTerm); + } + + }else{ + /* The query is a simple term lookup that matches at most one term in + ** the index. All that is required is a straight hash-lookup. + ** + ** Because the stack address of pE may be accessed via the aElem pointer + ** below, the "Fts3HashElem *pE" must be declared so that it is valid + ** within this entire function, not just this "else{...}" block. + */ + pE = fts3HashFindElem(pHash, zTerm, nTerm); + if( pE ){ + aElem = &pE; + nElem = 1; + } + } + + if( nElem>0 ){ + sqlite3_int64 nByte; + nByte = sizeof(Fts3SegReader) + (nElem+1)*sizeof(Fts3HashElem *); + pReader = (Fts3SegReader *)sqlite3_malloc64(nByte); + if( !pReader ){ + rc = SQLITE_NOMEM; + }else{ + memset(pReader, 0, nByte); + pReader->iIdx = 0x7FFFFFFF; + pReader->ppNextElem = (Fts3HashElem **)&pReader[1]; + memcpy(pReader->ppNextElem, aElem, nElem*sizeof(Fts3HashElem *)); + } + } + + if( bPrefix ){ + sqlite3_free(aElem); + } + *ppReader = pReader; + return rc; +} + +/* +** Compare the entries pointed to by two Fts3SegReader structures. +** Comparison is as follows: +** +** 1) EOF is greater than not EOF. +** +** 2) The current terms (if any) are compared using memcmp(). If one +** term is a prefix of another, the longer term is considered the +** larger. +** +** 3) By segment age. An older segment is considered larger. +*/ +static int fts3SegReaderCmp(Fts3SegReader *pLhs, Fts3SegReader *pRhs){ + int rc; + if( pLhs->aNode && pRhs->aNode ){ + int rc2 = pLhs->nTerm - pRhs->nTerm; + if( rc2<0 ){ + rc = memcmp(pLhs->zTerm, pRhs->zTerm, pLhs->nTerm); + }else{ + rc = memcmp(pLhs->zTerm, pRhs->zTerm, pRhs->nTerm); + } + if( rc==0 ){ + rc = rc2; + } + }else{ + rc = (pLhs->aNode==0) - (pRhs->aNode==0); + } + if( rc==0 ){ + rc = pRhs->iIdx - pLhs->iIdx; + } + assert( rc!=0 ); + return rc; +} + +/* +** A different comparison function for SegReader structures. In this +** version, it is assumed that each SegReader points to an entry in +** a doclist for identical terms. Comparison is made as follows: +** +** 1) EOF (end of doclist in this case) is greater than not EOF. +** +** 2) By current docid. +** +** 3) By segment age. An older segment is considered larger. +*/ +static int fts3SegReaderDoclistCmp(Fts3SegReader *pLhs, Fts3SegReader *pRhs){ + int rc = (pLhs->pOffsetList==0)-(pRhs->pOffsetList==0); + if( rc==0 ){ + if( pLhs->iDocid==pRhs->iDocid ){ + rc = pRhs->iIdx - pLhs->iIdx; + }else{ + rc = (pLhs->iDocid > pRhs->iDocid) ? 1 : -1; + } + } + assert( pLhs->aNode && pRhs->aNode ); + return rc; +} +static int fts3SegReaderDoclistCmpRev(Fts3SegReader *pLhs, Fts3SegReader *pRhs){ + int rc = (pLhs->pOffsetList==0)-(pRhs->pOffsetList==0); + if( rc==0 ){ + if( pLhs->iDocid==pRhs->iDocid ){ + rc = pRhs->iIdx - pLhs->iIdx; + }else{ + rc = (pLhs->iDocid < pRhs->iDocid) ? 1 : -1; + } + } + assert( pLhs->aNode && pRhs->aNode ); + return rc; +} + +/* +** Compare the term that the Fts3SegReader object passed as the first argument +** points to with the term specified by arguments zTerm and nTerm. +** +** If the pSeg iterator is already at EOF, return 0. Otherwise, return +** -ve if the pSeg term is less than zTerm/nTerm, 0 if the two terms are +** equal, or +ve if the pSeg term is greater than zTerm/nTerm. +*/ +static int fts3SegReaderTermCmp( + Fts3SegReader *pSeg, /* Segment reader object */ + const char *zTerm, /* Term to compare to */ + int nTerm /* Size of term zTerm in bytes */ +){ + int res = 0; + if( pSeg->aNode ){ + if( pSeg->nTerm>nTerm ){ + res = memcmp(pSeg->zTerm, zTerm, nTerm); + }else{ + res = memcmp(pSeg->zTerm, zTerm, pSeg->nTerm); + } + if( res==0 ){ + res = pSeg->nTerm-nTerm; + } + } + return res; +} + +/* +** Argument apSegment is an array of nSegment elements. It is known that +** the final (nSegment-nSuspect) members are already in sorted order +** (according to the comparison function provided). This function shuffles +** the array around until all entries are in sorted order. +*/ +static void fts3SegReaderSort( + Fts3SegReader **apSegment, /* Array to sort entries of */ + int nSegment, /* Size of apSegment array */ + int nSuspect, /* Unsorted entry count */ + int (*xCmp)(Fts3SegReader *, Fts3SegReader *) /* Comparison function */ +){ + int i; /* Iterator variable */ + + assert( nSuspect<=nSegment ); + + if( nSuspect==nSegment ) nSuspect--; + for(i=nSuspect-1; i>=0; i--){ + int j; + for(j=i; j<(nSegment-1); j++){ + Fts3SegReader *pTmp; + if( xCmp(apSegment[j], apSegment[j+1])<0 ) break; + pTmp = apSegment[j+1]; + apSegment[j+1] = apSegment[j]; + apSegment[j] = pTmp; + } + } + +#ifndef NDEBUG + /* Check that the list really is sorted now. */ + for(i=0; i<(nSuspect-1); i++){ + assert( xCmp(apSegment[i], apSegment[i+1])<0 ); + } +#endif +} + +/* +** Insert a record into the %_segments table. +*/ +static int fts3WriteSegment( + Fts3Table *p, /* Virtual table handle */ + sqlite3_int64 iBlock, /* Block id for new block */ + char *z, /* Pointer to buffer containing block data */ + int n /* Size of buffer z in bytes */ +){ + sqlite3_stmt *pStmt; + int rc = fts3SqlStmt(p, SQL_INSERT_SEGMENTS, &pStmt, 0); + if( rc==SQLITE_OK ){ + sqlite3_bind_int64(pStmt, 1, iBlock); + sqlite3_bind_blob(pStmt, 2, z, n, SQLITE_STATIC); + sqlite3_step(pStmt); + rc = sqlite3_reset(pStmt); + sqlite3_bind_null(pStmt, 2); + } + return rc; +} + +/* +** Find the largest relative level number in the table. If successful, set +** *pnMax to this value and return SQLITE_OK. Otherwise, if an error occurs, +** set *pnMax to zero and return an SQLite error code. +*/ +SQLITE_PRIVATE int sqlite3Fts3MaxLevel(Fts3Table *p, int *pnMax){ + int rc; + int mxLevel = 0; + sqlite3_stmt *pStmt = 0; + + rc = fts3SqlStmt(p, SQL_SELECT_MXLEVEL, &pStmt, 0); + if( rc==SQLITE_OK ){ + if( SQLITE_ROW==sqlite3_step(pStmt) ){ + mxLevel = sqlite3_column_int(pStmt, 0); + } + rc = sqlite3_reset(pStmt); + } + *pnMax = mxLevel; + return rc; +} + +/* +** Insert a record into the %_segdir table. +*/ +static int fts3WriteSegdir( + Fts3Table *p, /* Virtual table handle */ + sqlite3_int64 iLevel, /* Value for "level" field (absolute level) */ + int iIdx, /* Value for "idx" field */ + sqlite3_int64 iStartBlock, /* Value for "start_block" field */ + sqlite3_int64 iLeafEndBlock, /* Value for "leaves_end_block" field */ + sqlite3_int64 iEndBlock, /* Value for "end_block" field */ + sqlite3_int64 nLeafData, /* Bytes of leaf data in segment */ + char *zRoot, /* Blob value for "root" field */ + int nRoot /* Number of bytes in buffer zRoot */ +){ + sqlite3_stmt *pStmt; + int rc = fts3SqlStmt(p, SQL_INSERT_SEGDIR, &pStmt, 0); + if( rc==SQLITE_OK ){ + sqlite3_bind_int64(pStmt, 1, iLevel); + sqlite3_bind_int(pStmt, 2, iIdx); + sqlite3_bind_int64(pStmt, 3, iStartBlock); + sqlite3_bind_int64(pStmt, 4, iLeafEndBlock); + if( nLeafData==0 ){ + sqlite3_bind_int64(pStmt, 5, iEndBlock); + }else{ + char *zEnd = sqlite3_mprintf("%lld %lld", iEndBlock, nLeafData); + if( !zEnd ) return SQLITE_NOMEM; + sqlite3_bind_text(pStmt, 5, zEnd, -1, sqlite3_free); + } + sqlite3_bind_blob(pStmt, 6, zRoot, nRoot, SQLITE_STATIC); + sqlite3_step(pStmt); + rc = sqlite3_reset(pStmt); + sqlite3_bind_null(pStmt, 6); + } + return rc; +} + +/* +** Return the size of the common prefix (if any) shared by zPrev and +** zNext, in bytes. For example, +** +** fts3PrefixCompress("abc", 3, "abcdef", 6) // returns 3 +** fts3PrefixCompress("abX", 3, "abcdef", 6) // returns 2 +** fts3PrefixCompress("abX", 3, "Xbcdef", 6) // returns 0 +*/ +static int fts3PrefixCompress( + const char *zPrev, /* Buffer containing previous term */ + int nPrev, /* Size of buffer zPrev in bytes */ + const char *zNext, /* Buffer containing next term */ + int nNext /* Size of buffer zNext in bytes */ +){ + int n; + UNUSED_PARAMETER(nNext); + for(n=0; nnData; /* Current size of node in bytes */ + int nReq = nData; /* Required space after adding zTerm */ + int nPrefix; /* Number of bytes of prefix compression */ + int nSuffix; /* Suffix length */ + + nPrefix = fts3PrefixCompress(pTree->zTerm, pTree->nTerm, zTerm, nTerm); + nSuffix = nTerm-nPrefix; + + nReq += sqlite3Fts3VarintLen(nPrefix)+sqlite3Fts3VarintLen(nSuffix)+nSuffix; + if( nReq<=p->nNodeSize || !pTree->zTerm ){ + + if( nReq>p->nNodeSize ){ + /* An unusual case: this is the first term to be added to the node + ** and the static node buffer (p->nNodeSize bytes) is not large + ** enough. Use a separately malloced buffer instead This wastes + ** p->nNodeSize bytes, but since this scenario only comes about when + ** the database contain two terms that share a prefix of almost 2KB, + ** this is not expected to be a serious problem. + */ + assert( pTree->aData==(char *)&pTree[1] ); + pTree->aData = (char *)sqlite3_malloc(nReq); + if( !pTree->aData ){ + return SQLITE_NOMEM; + } + } + + if( pTree->zTerm ){ + /* There is no prefix-length field for first term in a node */ + nData += sqlite3Fts3PutVarint(&pTree->aData[nData], nPrefix); + } + + nData += sqlite3Fts3PutVarint(&pTree->aData[nData], nSuffix); + memcpy(&pTree->aData[nData], &zTerm[nPrefix], nSuffix); + pTree->nData = nData + nSuffix; + pTree->nEntry++; + + if( isCopyTerm ){ + if( pTree->nMalloczMalloc, nTerm*2); + if( !zNew ){ + return SQLITE_NOMEM; + } + pTree->nMalloc = nTerm*2; + pTree->zMalloc = zNew; + } + pTree->zTerm = pTree->zMalloc; + memcpy(pTree->zTerm, zTerm, nTerm); + pTree->nTerm = nTerm; + }else{ + pTree->zTerm = (char *)zTerm; + pTree->nTerm = nTerm; + } + return SQLITE_OK; + } + } + + /* If control flows to here, it was not possible to append zTerm to the + ** current node. Create a new node (a right-sibling of the current node). + ** If this is the first node in the tree, the term is added to it. + ** + ** Otherwise, the term is not added to the new node, it is left empty for + ** now. Instead, the term is inserted into the parent of pTree. If pTree + ** has no parent, one is created here. + */ + pNew = (SegmentNode *)sqlite3_malloc(sizeof(SegmentNode) + p->nNodeSize); + if( !pNew ){ + return SQLITE_NOMEM; + } + memset(pNew, 0, sizeof(SegmentNode)); + pNew->nData = 1 + FTS3_VARINT_MAX; + pNew->aData = (char *)&pNew[1]; + + if( pTree ){ + SegmentNode *pParent = pTree->pParent; + rc = fts3NodeAddTerm(p, &pParent, isCopyTerm, zTerm, nTerm); + if( pTree->pParent==0 ){ + pTree->pParent = pParent; + } + pTree->pRight = pNew; + pNew->pLeftmost = pTree->pLeftmost; + pNew->pParent = pParent; + pNew->zMalloc = pTree->zMalloc; + pNew->nMalloc = pTree->nMalloc; + pTree->zMalloc = 0; + }else{ + pNew->pLeftmost = pNew; + rc = fts3NodeAddTerm(p, &pNew, isCopyTerm, zTerm, nTerm); + } + + *ppTree = pNew; + return rc; +} + +/* +** Helper function for fts3NodeWrite(). +*/ +static int fts3TreeFinishNode( + SegmentNode *pTree, + int iHeight, + sqlite3_int64 iLeftChild +){ + int nStart; + assert( iHeight>=1 && iHeight<128 ); + nStart = FTS3_VARINT_MAX - sqlite3Fts3VarintLen(iLeftChild); + pTree->aData[nStart] = (char)iHeight; + sqlite3Fts3PutVarint(&pTree->aData[nStart+1], iLeftChild); + return nStart; +} + +/* +** Write the buffer for the segment node pTree and all of its peers to the +** database. Then call this function recursively to write the parent of +** pTree and its peers to the database. +** +** Except, if pTree is a root node, do not write it to the database. Instead, +** set output variables *paRoot and *pnRoot to contain the root node. +** +** If successful, SQLITE_OK is returned and output variable *piLast is +** set to the largest blockid written to the database (or zero if no +** blocks were written to the db). Otherwise, an SQLite error code is +** returned. +*/ +static int fts3NodeWrite( + Fts3Table *p, /* Virtual table handle */ + SegmentNode *pTree, /* SegmentNode handle */ + int iHeight, /* Height of this node in tree */ + sqlite3_int64 iLeaf, /* Block id of first leaf node */ + sqlite3_int64 iFree, /* Block id of next free slot in %_segments */ + sqlite3_int64 *piLast, /* OUT: Block id of last entry written */ + char **paRoot, /* OUT: Data for root node */ + int *pnRoot /* OUT: Size of root node in bytes */ +){ + int rc = SQLITE_OK; + + if( !pTree->pParent ){ + /* Root node of the tree. */ + int nStart = fts3TreeFinishNode(pTree, iHeight, iLeaf); + *piLast = iFree-1; + *pnRoot = pTree->nData - nStart; + *paRoot = &pTree->aData[nStart]; + }else{ + SegmentNode *pIter; + sqlite3_int64 iNextFree = iFree; + sqlite3_int64 iNextLeaf = iLeaf; + for(pIter=pTree->pLeftmost; pIter && rc==SQLITE_OK; pIter=pIter->pRight){ + int nStart = fts3TreeFinishNode(pIter, iHeight, iNextLeaf); + int nWrite = pIter->nData - nStart; + + rc = fts3WriteSegment(p, iNextFree, &pIter->aData[nStart], nWrite); + iNextFree++; + iNextLeaf += (pIter->nEntry+1); + } + if( rc==SQLITE_OK ){ + assert( iNextLeaf==iFree ); + rc = fts3NodeWrite( + p, pTree->pParent, iHeight+1, iFree, iNextFree, piLast, paRoot, pnRoot + ); + } + } + + return rc; +} + +/* +** Free all memory allocations associated with the tree pTree. +*/ +static void fts3NodeFree(SegmentNode *pTree){ + if( pTree ){ + SegmentNode *p = pTree->pLeftmost; + fts3NodeFree(p->pParent); + while( p ){ + SegmentNode *pRight = p->pRight; + if( p->aData!=(char *)&p[1] ){ + sqlite3_free(p->aData); + } + assert( pRight==0 || p->zMalloc==0 ); + sqlite3_free(p->zMalloc); + sqlite3_free(p); + p = pRight; + } + } +} + +/* +** Add a term to the segment being constructed by the SegmentWriter object +** *ppWriter. When adding the first term to a segment, *ppWriter should +** be passed NULL. This function will allocate a new SegmentWriter object +** and return it via the input/output variable *ppWriter in this case. +** +** If successful, SQLITE_OK is returned. Otherwise, an SQLite error code. +*/ +static int fts3SegWriterAdd( + Fts3Table *p, /* Virtual table handle */ + SegmentWriter **ppWriter, /* IN/OUT: SegmentWriter handle */ + int isCopyTerm, /* True if buffer zTerm must be copied */ + const char *zTerm, /* Pointer to buffer containing term */ + int nTerm, /* Size of term in bytes */ + const char *aDoclist, /* Pointer to buffer containing doclist */ + int nDoclist /* Size of doclist in bytes */ +){ + int nPrefix; /* Size of term prefix in bytes */ + int nSuffix; /* Size of term suffix in bytes */ + int nReq; /* Number of bytes required on leaf page */ + int nData; + SegmentWriter *pWriter = *ppWriter; + + if( !pWriter ){ + int rc; + sqlite3_stmt *pStmt; + + /* Allocate the SegmentWriter structure */ + pWriter = (SegmentWriter *)sqlite3_malloc(sizeof(SegmentWriter)); + if( !pWriter ) return SQLITE_NOMEM; + memset(pWriter, 0, sizeof(SegmentWriter)); + *ppWriter = pWriter; + + /* Allocate a buffer in which to accumulate data */ + pWriter->aData = (char *)sqlite3_malloc(p->nNodeSize); + if( !pWriter->aData ) return SQLITE_NOMEM; + pWriter->nSize = p->nNodeSize; + + /* Find the next free blockid in the %_segments table */ + rc = fts3SqlStmt(p, SQL_NEXT_SEGMENTS_ID, &pStmt, 0); + if( rc!=SQLITE_OK ) return rc; + if( SQLITE_ROW==sqlite3_step(pStmt) ){ + pWriter->iFree = sqlite3_column_int64(pStmt, 0); + pWriter->iFirst = pWriter->iFree; + } + rc = sqlite3_reset(pStmt); + if( rc!=SQLITE_OK ) return rc; + } + nData = pWriter->nData; + + nPrefix = fts3PrefixCompress(pWriter->zTerm, pWriter->nTerm, zTerm, nTerm); + nSuffix = nTerm-nPrefix; + + /* If nSuffix is zero or less, then zTerm/nTerm must be a prefix of + ** pWriter->zTerm/pWriter->nTerm. i.e. must be equal to or less than when + ** compared with BINARY collation. This indicates corruption. */ + if( nSuffix<=0 ) return FTS_CORRUPT_VTAB; + + /* Figure out how many bytes are required by this new entry */ + nReq = sqlite3Fts3VarintLen(nPrefix) + /* varint containing prefix size */ + sqlite3Fts3VarintLen(nSuffix) + /* varint containing suffix size */ + nSuffix + /* Term suffix */ + sqlite3Fts3VarintLen(nDoclist) + /* Size of doclist */ + nDoclist; /* Doclist data */ + + if( nData>0 && nData+nReq>p->nNodeSize ){ + int rc; + + /* The current leaf node is full. Write it out to the database. */ + rc = fts3WriteSegment(p, pWriter->iFree++, pWriter->aData, nData); + if( rc!=SQLITE_OK ) return rc; + p->nLeafAdd++; + + /* Add the current term to the interior node tree. The term added to + ** the interior tree must: + ** + ** a) be greater than the largest term on the leaf node just written + ** to the database (still available in pWriter->zTerm), and + ** + ** b) be less than or equal to the term about to be added to the new + ** leaf node (zTerm/nTerm). + ** + ** In other words, it must be the prefix of zTerm 1 byte longer than + ** the common prefix (if any) of zTerm and pWriter->zTerm. + */ + assert( nPrefixpTree, isCopyTerm, zTerm, nPrefix+1); + if( rc!=SQLITE_OK ) return rc; + + nData = 0; + pWriter->nTerm = 0; + + nPrefix = 0; + nSuffix = nTerm; + nReq = 1 + /* varint containing prefix size */ + sqlite3Fts3VarintLen(nTerm) + /* varint containing suffix size */ + nTerm + /* Term suffix */ + sqlite3Fts3VarintLen(nDoclist) + /* Size of doclist */ + nDoclist; /* Doclist data */ + } + + /* Increase the total number of bytes written to account for the new entry. */ + pWriter->nLeafData += nReq; + + /* If the buffer currently allocated is too small for this entry, realloc + ** the buffer to make it large enough. + */ + if( nReq>pWriter->nSize ){ + char *aNew = sqlite3_realloc(pWriter->aData, nReq); + if( !aNew ) return SQLITE_NOMEM; + pWriter->aData = aNew; + pWriter->nSize = nReq; + } + assert( nData+nReq<=pWriter->nSize ); + + /* Append the prefix-compressed term and doclist to the buffer. */ + nData += sqlite3Fts3PutVarint(&pWriter->aData[nData], nPrefix); + nData += sqlite3Fts3PutVarint(&pWriter->aData[nData], nSuffix); + memcpy(&pWriter->aData[nData], &zTerm[nPrefix], nSuffix); + nData += nSuffix; + nData += sqlite3Fts3PutVarint(&pWriter->aData[nData], nDoclist); + memcpy(&pWriter->aData[nData], aDoclist, nDoclist); + pWriter->nData = nData + nDoclist; + + /* Save the current term so that it can be used to prefix-compress the next. + ** If the isCopyTerm parameter is true, then the buffer pointed to by + ** zTerm is transient, so take a copy of the term data. Otherwise, just + ** store a copy of the pointer. + */ + if( isCopyTerm ){ + if( nTerm>pWriter->nMalloc ){ + char *zNew = sqlite3_realloc(pWriter->zMalloc, nTerm*2); + if( !zNew ){ + return SQLITE_NOMEM; + } + pWriter->nMalloc = nTerm*2; + pWriter->zMalloc = zNew; + pWriter->zTerm = zNew; + } + assert( pWriter->zTerm==pWriter->zMalloc ); + memcpy(pWriter->zTerm, zTerm, nTerm); + }else{ + pWriter->zTerm = (char *)zTerm; + } + pWriter->nTerm = nTerm; + + return SQLITE_OK; +} + +/* +** Flush all data associated with the SegmentWriter object pWriter to the +** database. This function must be called after all terms have been added +** to the segment using fts3SegWriterAdd(). If successful, SQLITE_OK is +** returned. Otherwise, an SQLite error code. +*/ +static int fts3SegWriterFlush( + Fts3Table *p, /* Virtual table handle */ + SegmentWriter *pWriter, /* SegmentWriter to flush to the db */ + sqlite3_int64 iLevel, /* Value for 'level' column of %_segdir */ + int iIdx /* Value for 'idx' column of %_segdir */ +){ + int rc; /* Return code */ + if( pWriter->pTree ){ + sqlite3_int64 iLast = 0; /* Largest block id written to database */ + sqlite3_int64 iLastLeaf; /* Largest leaf block id written to db */ + char *zRoot = NULL; /* Pointer to buffer containing root node */ + int nRoot = 0; /* Size of buffer zRoot */ + + iLastLeaf = pWriter->iFree; + rc = fts3WriteSegment(p, pWriter->iFree++, pWriter->aData, pWriter->nData); + if( rc==SQLITE_OK ){ + rc = fts3NodeWrite(p, pWriter->pTree, 1, + pWriter->iFirst, pWriter->iFree, &iLast, &zRoot, &nRoot); + } + if( rc==SQLITE_OK ){ + rc = fts3WriteSegdir(p, iLevel, iIdx, + pWriter->iFirst, iLastLeaf, iLast, pWriter->nLeafData, zRoot, nRoot); + } + }else{ + /* The entire tree fits on the root node. Write it to the segdir table. */ + rc = fts3WriteSegdir(p, iLevel, iIdx, + 0, 0, 0, pWriter->nLeafData, pWriter->aData, pWriter->nData); + } + p->nLeafAdd++; + return rc; +} + +/* +** Release all memory held by the SegmentWriter object passed as the +** first argument. +*/ +static void fts3SegWriterFree(SegmentWriter *pWriter){ + if( pWriter ){ + sqlite3_free(pWriter->aData); + sqlite3_free(pWriter->zMalloc); + fts3NodeFree(pWriter->pTree); + sqlite3_free(pWriter); + } +} + +/* +** The first value in the apVal[] array is assumed to contain an integer. +** This function tests if there exist any documents with docid values that +** are different from that integer. i.e. if deleting the document with docid +** pRowid would mean the FTS3 table were empty. +** +** If successful, *pisEmpty is set to true if the table is empty except for +** document pRowid, or false otherwise, and SQLITE_OK is returned. If an +** error occurs, an SQLite error code is returned. +*/ +static int fts3IsEmpty(Fts3Table *p, sqlite3_value *pRowid, int *pisEmpty){ + sqlite3_stmt *pStmt; + int rc; + if( p->zContentTbl ){ + /* If using the content=xxx option, assume the table is never empty */ + *pisEmpty = 0; + rc = SQLITE_OK; + }else{ + rc = fts3SqlStmt(p, SQL_IS_EMPTY, &pStmt, &pRowid); + if( rc==SQLITE_OK ){ + if( SQLITE_ROW==sqlite3_step(pStmt) ){ + *pisEmpty = sqlite3_column_int(pStmt, 0); + } + rc = sqlite3_reset(pStmt); + } + } + return rc; +} + +/* +** Set *pnMax to the largest segment level in the database for the index +** iIndex. +** +** Segment levels are stored in the 'level' column of the %_segdir table. +** +** Return SQLITE_OK if successful, or an SQLite error code if not. +*/ +static int fts3SegmentMaxLevel( + Fts3Table *p, + int iLangid, + int iIndex, + sqlite3_int64 *pnMax +){ + sqlite3_stmt *pStmt; + int rc; + assert( iIndex>=0 && iIndexnIndex ); + + /* Set pStmt to the compiled version of: + ** + ** SELECT max(level) FROM %Q.'%q_segdir' WHERE level BETWEEN ? AND ? + ** + ** (1024 is actually the value of macro FTS3_SEGDIR_PREFIXLEVEL_STR). + */ + rc = fts3SqlStmt(p, SQL_SELECT_SEGDIR_MAX_LEVEL, &pStmt, 0); + if( rc!=SQLITE_OK ) return rc; + sqlite3_bind_int64(pStmt, 1, getAbsoluteLevel(p, iLangid, iIndex, 0)); + sqlite3_bind_int64(pStmt, 2, + getAbsoluteLevel(p, iLangid, iIndex, FTS3_SEGDIR_MAXLEVEL-1) + ); + if( SQLITE_ROW==sqlite3_step(pStmt) ){ + *pnMax = sqlite3_column_int64(pStmt, 0); + } + return sqlite3_reset(pStmt); +} + +/* +** iAbsLevel is an absolute level that may be assumed to exist within +** the database. This function checks if it is the largest level number +** within its index. Assuming no error occurs, *pbMax is set to 1 if +** iAbsLevel is indeed the largest level, or 0 otherwise, and SQLITE_OK +** is returned. If an error occurs, an error code is returned and the +** final value of *pbMax is undefined. +*/ +static int fts3SegmentIsMaxLevel(Fts3Table *p, i64 iAbsLevel, int *pbMax){ + + /* Set pStmt to the compiled version of: + ** + ** SELECT max(level) FROM %Q.'%q_segdir' WHERE level BETWEEN ? AND ? + ** + ** (1024 is actually the value of macro FTS3_SEGDIR_PREFIXLEVEL_STR). + */ + sqlite3_stmt *pStmt; + int rc = fts3SqlStmt(p, SQL_SELECT_SEGDIR_MAX_LEVEL, &pStmt, 0); + if( rc!=SQLITE_OK ) return rc; + sqlite3_bind_int64(pStmt, 1, iAbsLevel+1); + sqlite3_bind_int64(pStmt, 2, + ((iAbsLevel/FTS3_SEGDIR_MAXLEVEL)+1) * FTS3_SEGDIR_MAXLEVEL + ); + + *pbMax = 0; + if( SQLITE_ROW==sqlite3_step(pStmt) ){ + *pbMax = sqlite3_column_type(pStmt, 0)==SQLITE_NULL; + } + return sqlite3_reset(pStmt); +} + +/* +** Delete all entries in the %_segments table associated with the segment +** opened with seg-reader pSeg. This function does not affect the contents +** of the %_segdir table. +*/ +static int fts3DeleteSegment( + Fts3Table *p, /* FTS table handle */ + Fts3SegReader *pSeg /* Segment to delete */ +){ + int rc = SQLITE_OK; /* Return code */ + if( pSeg->iStartBlock ){ + sqlite3_stmt *pDelete; /* SQL statement to delete rows */ + rc = fts3SqlStmt(p, SQL_DELETE_SEGMENTS_RANGE, &pDelete, 0); + if( rc==SQLITE_OK ){ + sqlite3_bind_int64(pDelete, 1, pSeg->iStartBlock); + sqlite3_bind_int64(pDelete, 2, pSeg->iEndBlock); + sqlite3_step(pDelete); + rc = sqlite3_reset(pDelete); + } + } + return rc; +} + +/* +** This function is used after merging multiple segments into a single large +** segment to delete the old, now redundant, segment b-trees. Specifically, +** it: +** +** 1) Deletes all %_segments entries for the segments associated with +** each of the SegReader objects in the array passed as the third +** argument, and +** +** 2) deletes all %_segdir entries with level iLevel, or all %_segdir +** entries regardless of level if (iLevel<0). +** +** SQLITE_OK is returned if successful, otherwise an SQLite error code. +*/ +static int fts3DeleteSegdir( + Fts3Table *p, /* Virtual table handle */ + int iLangid, /* Language id */ + int iIndex, /* Index for p->aIndex */ + int iLevel, /* Level of %_segdir entries to delete */ + Fts3SegReader **apSegment, /* Array of SegReader objects */ + int nReader /* Size of array apSegment */ +){ + int rc = SQLITE_OK; /* Return Code */ + int i; /* Iterator variable */ + sqlite3_stmt *pDelete = 0; /* SQL statement to delete rows */ + + for(i=0; rc==SQLITE_OK && i=0 || iLevel==FTS3_SEGCURSOR_ALL ); + if( iLevel==FTS3_SEGCURSOR_ALL ){ + rc = fts3SqlStmt(p, SQL_DELETE_SEGDIR_RANGE, &pDelete, 0); + if( rc==SQLITE_OK ){ + sqlite3_bind_int64(pDelete, 1, getAbsoluteLevel(p, iLangid, iIndex, 0)); + sqlite3_bind_int64(pDelete, 2, + getAbsoluteLevel(p, iLangid, iIndex, FTS3_SEGDIR_MAXLEVEL-1) + ); + } + }else{ + rc = fts3SqlStmt(p, SQL_DELETE_SEGDIR_LEVEL, &pDelete, 0); + if( rc==SQLITE_OK ){ + sqlite3_bind_int64( + pDelete, 1, getAbsoluteLevel(p, iLangid, iIndex, iLevel) + ); + } + } + + if( rc==SQLITE_OK ){ + sqlite3_step(pDelete); + rc = sqlite3_reset(pDelete); + } + + return rc; +} + +/* +** When this function is called, buffer *ppList (size *pnList bytes) contains +** a position list that may (or may not) feature multiple columns. This +** function adjusts the pointer *ppList and the length *pnList so that they +** identify the subset of the position list that corresponds to column iCol. +** +** If there are no entries in the input position list for column iCol, then +** *pnList is set to zero before returning. +** +** If parameter bZero is non-zero, then any part of the input list following +** the end of the output list is zeroed before returning. +*/ +static void fts3ColumnFilter( + int iCol, /* Column to filter on */ + int bZero, /* Zero out anything following *ppList */ + char **ppList, /* IN/OUT: Pointer to position list */ + int *pnList /* IN/OUT: Size of buffer *ppList in bytes */ +){ + char *pList = *ppList; + int nList = *pnList; + char *pEnd = &pList[nList]; + int iCurrent = 0; + char *p = pList; + + assert( iCol>=0 ); + while( 1 ){ + char c = 0; + while( p0){ + memset(&pList[nList], 0, pEnd - &pList[nList]); + } + *ppList = pList; + *pnList = nList; +} + +/* +** Cache data in the Fts3MultiSegReader.aBuffer[] buffer (overwriting any +** existing data). Grow the buffer if required. +** +** If successful, return SQLITE_OK. Otherwise, if an OOM error is encountered +** trying to resize the buffer, return SQLITE_NOMEM. +*/ +static int fts3MsrBufferData( + Fts3MultiSegReader *pMsr, /* Multi-segment-reader handle */ + char *pList, + int nList +){ + if( nList>pMsr->nBuffer ){ + char *pNew; + pMsr->nBuffer = nList*2; + pNew = (char *)sqlite3_realloc(pMsr->aBuffer, pMsr->nBuffer); + if( !pNew ) return SQLITE_NOMEM; + pMsr->aBuffer = pNew; + } + + memcpy(pMsr->aBuffer, pList, nList); + return SQLITE_OK; +} + +SQLITE_PRIVATE int sqlite3Fts3MsrIncrNext( + Fts3Table *p, /* Virtual table handle */ + Fts3MultiSegReader *pMsr, /* Multi-segment-reader handle */ + sqlite3_int64 *piDocid, /* OUT: Docid value */ + char **paPoslist, /* OUT: Pointer to position list */ + int *pnPoslist /* OUT: Size of position list in bytes */ +){ + int nMerge = pMsr->nAdvance; + Fts3SegReader **apSegment = pMsr->apSegment; + int (*xCmp)(Fts3SegReader *, Fts3SegReader *) = ( + p->bDescIdx ? fts3SegReaderDoclistCmpRev : fts3SegReaderDoclistCmp + ); + + if( nMerge==0 ){ + *paPoslist = 0; + return SQLITE_OK; + } + + while( 1 ){ + Fts3SegReader *pSeg; + pSeg = pMsr->apSegment[0]; + + if( pSeg->pOffsetList==0 ){ + *paPoslist = 0; + break; + }else{ + int rc; + char *pList; + int nList; + int j; + sqlite3_int64 iDocid = apSegment[0]->iDocid; + + rc = fts3SegReaderNextDocid(p, apSegment[0], &pList, &nList); + j = 1; + while( rc==SQLITE_OK + && jpOffsetList + && apSegment[j]->iDocid==iDocid + ){ + rc = fts3SegReaderNextDocid(p, apSegment[j], 0, 0); + j++; + } + if( rc!=SQLITE_OK ) return rc; + fts3SegReaderSort(pMsr->apSegment, nMerge, j, xCmp); + + if( nList>0 && fts3SegReaderIsPending(apSegment[0]) ){ + rc = fts3MsrBufferData(pMsr, pList, nList+1); + if( rc!=SQLITE_OK ) return rc; + assert( (pMsr->aBuffer[nList] & 0xFE)==0x00 ); + pList = pMsr->aBuffer; + } + + if( pMsr->iColFilter>=0 ){ + fts3ColumnFilter(pMsr->iColFilter, 1, &pList, &nList); + } + + if( nList>0 ){ + *paPoslist = pList; + *piDocid = iDocid; + *pnPoslist = nList; + break; + } + } + } + + return SQLITE_OK; +} + +static int fts3SegReaderStart( + Fts3Table *p, /* Virtual table handle */ + Fts3MultiSegReader *pCsr, /* Cursor object */ + const char *zTerm, /* Term searched for (or NULL) */ + int nTerm /* Length of zTerm in bytes */ +){ + int i; + int nSeg = pCsr->nSegment; + + /* If the Fts3SegFilter defines a specific term (or term prefix) to search + ** for, then advance each segment iterator until it points to a term of + ** equal or greater value than the specified term. This prevents many + ** unnecessary merge/sort operations for the case where single segment + ** b-tree leaf nodes contain more than one term. + */ + for(i=0; pCsr->bRestart==0 && inSegment; i++){ + int res = 0; + Fts3SegReader *pSeg = pCsr->apSegment[i]; + do { + int rc = fts3SegReaderNext(p, pSeg, 0); + if( rc!=SQLITE_OK ) return rc; + }while( zTerm && (res = fts3SegReaderTermCmp(pSeg, zTerm, nTerm))<0 ); + + if( pSeg->bLookup && res!=0 ){ + fts3SegReaderSetEof(pSeg); + } + } + fts3SegReaderSort(pCsr->apSegment, nSeg, nSeg, fts3SegReaderCmp); + + return SQLITE_OK; +} + +SQLITE_PRIVATE int sqlite3Fts3SegReaderStart( + Fts3Table *p, /* Virtual table handle */ + Fts3MultiSegReader *pCsr, /* Cursor object */ + Fts3SegFilter *pFilter /* Restrictions on range of iteration */ +){ + pCsr->pFilter = pFilter; + return fts3SegReaderStart(p, pCsr, pFilter->zTerm, pFilter->nTerm); +} + +SQLITE_PRIVATE int sqlite3Fts3MsrIncrStart( + Fts3Table *p, /* Virtual table handle */ + Fts3MultiSegReader *pCsr, /* Cursor object */ + int iCol, /* Column to match on. */ + const char *zTerm, /* Term to iterate through a doclist for */ + int nTerm /* Number of bytes in zTerm */ +){ + int i; + int rc; + int nSegment = pCsr->nSegment; + int (*xCmp)(Fts3SegReader *, Fts3SegReader *) = ( + p->bDescIdx ? fts3SegReaderDoclistCmpRev : fts3SegReaderDoclistCmp + ); + + assert( pCsr->pFilter==0 ); + assert( zTerm && nTerm>0 ); + + /* Advance each segment iterator until it points to the term zTerm/nTerm. */ + rc = fts3SegReaderStart(p, pCsr, zTerm, nTerm); + if( rc!=SQLITE_OK ) return rc; + + /* Determine how many of the segments actually point to zTerm/nTerm. */ + for(i=0; iapSegment[i]; + if( !pSeg->aNode || fts3SegReaderTermCmp(pSeg, zTerm, nTerm) ){ + break; + } + } + pCsr->nAdvance = i; + + /* Advance each of the segments to point to the first docid. */ + for(i=0; inAdvance; i++){ + rc = fts3SegReaderFirstDocid(p, pCsr->apSegment[i]); + if( rc!=SQLITE_OK ) return rc; + } + fts3SegReaderSort(pCsr->apSegment, i, i, xCmp); + + assert( iCol<0 || iColnColumn ); + pCsr->iColFilter = iCol; + + return SQLITE_OK; +} + +/* +** This function is called on a MultiSegReader that has been started using +** sqlite3Fts3MsrIncrStart(). One or more calls to MsrIncrNext() may also +** have been made. Calling this function puts the MultiSegReader in such +** a state that if the next two calls are: +** +** sqlite3Fts3SegReaderStart() +** sqlite3Fts3SegReaderStep() +** +** then the entire doclist for the term is available in +** MultiSegReader.aDoclist/nDoclist. +*/ +SQLITE_PRIVATE int sqlite3Fts3MsrIncrRestart(Fts3MultiSegReader *pCsr){ + int i; /* Used to iterate through segment-readers */ + + assert( pCsr->zTerm==0 ); + assert( pCsr->nTerm==0 ); + assert( pCsr->aDoclist==0 ); + assert( pCsr->nDoclist==0 ); + + pCsr->nAdvance = 0; + pCsr->bRestart = 1; + for(i=0; inSegment; i++){ + pCsr->apSegment[i]->pOffsetList = 0; + pCsr->apSegment[i]->nOffsetList = 0; + pCsr->apSegment[i]->iDocid = 0; + } + + return SQLITE_OK; +} + + +SQLITE_PRIVATE int sqlite3Fts3SegReaderStep( + Fts3Table *p, /* Virtual table handle */ + Fts3MultiSegReader *pCsr /* Cursor object */ +){ + int rc = SQLITE_OK; + + int isIgnoreEmpty = (pCsr->pFilter->flags & FTS3_SEGMENT_IGNORE_EMPTY); + int isRequirePos = (pCsr->pFilter->flags & FTS3_SEGMENT_REQUIRE_POS); + int isColFilter = (pCsr->pFilter->flags & FTS3_SEGMENT_COLUMN_FILTER); + int isPrefix = (pCsr->pFilter->flags & FTS3_SEGMENT_PREFIX); + int isScan = (pCsr->pFilter->flags & FTS3_SEGMENT_SCAN); + int isFirst = (pCsr->pFilter->flags & FTS3_SEGMENT_FIRST); + + Fts3SegReader **apSegment = pCsr->apSegment; + int nSegment = pCsr->nSegment; + Fts3SegFilter *pFilter = pCsr->pFilter; + int (*xCmp)(Fts3SegReader *, Fts3SegReader *) = ( + p->bDescIdx ? fts3SegReaderDoclistCmpRev : fts3SegReaderDoclistCmp + ); + + if( pCsr->nSegment==0 ) return SQLITE_OK; + + do { + int nMerge; + int i; + + /* Advance the first pCsr->nAdvance entries in the apSegment[] array + ** forward. Then sort the list in order of current term again. + */ + for(i=0; inAdvance; i++){ + Fts3SegReader *pSeg = apSegment[i]; + if( pSeg->bLookup ){ + fts3SegReaderSetEof(pSeg); + }else{ + rc = fts3SegReaderNext(p, pSeg, 0); + } + if( rc!=SQLITE_OK ) return rc; + } + fts3SegReaderSort(apSegment, nSegment, pCsr->nAdvance, fts3SegReaderCmp); + pCsr->nAdvance = 0; + + /* If all the seg-readers are at EOF, we're finished. return SQLITE_OK. */ + assert( rc==SQLITE_OK ); + if( apSegment[0]->aNode==0 ) break; + + pCsr->nTerm = apSegment[0]->nTerm; + pCsr->zTerm = apSegment[0]->zTerm; + + /* If this is a prefix-search, and if the term that apSegment[0] points + ** to does not share a suffix with pFilter->zTerm/nTerm, then all + ** required callbacks have been made. In this case exit early. + ** + ** Similarly, if this is a search for an exact match, and the first term + ** of segment apSegment[0] is not a match, exit early. + */ + if( pFilter->zTerm && !isScan ){ + if( pCsr->nTermnTerm + || (!isPrefix && pCsr->nTerm>pFilter->nTerm) + || memcmp(pCsr->zTerm, pFilter->zTerm, pFilter->nTerm) + ){ + break; + } + } + + nMerge = 1; + while( nMergeaNode + && apSegment[nMerge]->nTerm==pCsr->nTerm + && 0==memcmp(pCsr->zTerm, apSegment[nMerge]->zTerm, pCsr->nTerm) + ){ + nMerge++; + } + + assert( isIgnoreEmpty || (isRequirePos && !isColFilter) ); + if( nMerge==1 + && !isIgnoreEmpty + && !isFirst + && (p->bDescIdx==0 || fts3SegReaderIsPending(apSegment[0])==0) + ){ + pCsr->nDoclist = apSegment[0]->nDoclist; + if( fts3SegReaderIsPending(apSegment[0]) ){ + rc = fts3MsrBufferData(pCsr, apSegment[0]->aDoclist, pCsr->nDoclist); + pCsr->aDoclist = pCsr->aBuffer; + }else{ + pCsr->aDoclist = apSegment[0]->aDoclist; + } + if( rc==SQLITE_OK ) rc = SQLITE_ROW; + }else{ + int nDoclist = 0; /* Size of doclist */ + sqlite3_int64 iPrev = 0; /* Previous docid stored in doclist */ + + /* The current term of the first nMerge entries in the array + ** of Fts3SegReader objects is the same. The doclists must be merged + ** and a single term returned with the merged doclist. + */ + for(i=0; ipOffsetList ){ + int j; /* Number of segments that share a docid */ + char *pList = 0; + int nList = 0; + int nByte; + sqlite3_int64 iDocid = apSegment[0]->iDocid; + fts3SegReaderNextDocid(p, apSegment[0], &pList, &nList); + j = 1; + while( jpOffsetList + && apSegment[j]->iDocid==iDocid + ){ + fts3SegReaderNextDocid(p, apSegment[j], 0, 0); + j++; + } + + if( isColFilter ){ + fts3ColumnFilter(pFilter->iCol, 0, &pList, &nList); + } + + if( !isIgnoreEmpty || nList>0 ){ + + /* Calculate the 'docid' delta value to write into the merged + ** doclist. */ + sqlite3_int64 iDelta; + if( p->bDescIdx && nDoclist>0 ){ + iDelta = iPrev - iDocid; + }else{ + iDelta = iDocid - iPrev; + } + if( iDelta<=0 && (nDoclist>0 || iDelta!=iDocid) ){ + return FTS_CORRUPT_VTAB; + } + assert( nDoclist>0 || iDelta==iDocid ); + + nByte = sqlite3Fts3VarintLen(iDelta) + (isRequirePos?nList+1:0); + if( nDoclist+nByte>pCsr->nBuffer ){ + char *aNew; + pCsr->nBuffer = (nDoclist+nByte)*2; + aNew = sqlite3_realloc(pCsr->aBuffer, pCsr->nBuffer); + if( !aNew ){ + return SQLITE_NOMEM; + } + pCsr->aBuffer = aNew; + } + + if( isFirst ){ + char *a = &pCsr->aBuffer[nDoclist]; + int nWrite; + + nWrite = sqlite3Fts3FirstFilter(iDelta, pList, nList, a); + if( nWrite ){ + iPrev = iDocid; + nDoclist += nWrite; + } + }else{ + nDoclist += sqlite3Fts3PutVarint(&pCsr->aBuffer[nDoclist], iDelta); + iPrev = iDocid; + if( isRequirePos ){ + memcpy(&pCsr->aBuffer[nDoclist], pList, nList); + nDoclist += nList; + pCsr->aBuffer[nDoclist++] = '\0'; + } + } + } + + fts3SegReaderSort(apSegment, nMerge, j, xCmp); + } + if( nDoclist>0 ){ + pCsr->aDoclist = pCsr->aBuffer; + pCsr->nDoclist = nDoclist; + rc = SQLITE_ROW; + } + } + pCsr->nAdvance = nMerge; + }while( rc==SQLITE_OK ); + + return rc; +} + + +SQLITE_PRIVATE void sqlite3Fts3SegReaderFinish( + Fts3MultiSegReader *pCsr /* Cursor object */ +){ + if( pCsr ){ + int i; + for(i=0; inSegment; i++){ + sqlite3Fts3SegReaderFree(pCsr->apSegment[i]); + } + sqlite3_free(pCsr->apSegment); + sqlite3_free(pCsr->aBuffer); + + pCsr->nSegment = 0; + pCsr->apSegment = 0; + pCsr->aBuffer = 0; + } +} + +/* +** Decode the "end_block" field, selected by column iCol of the SELECT +** statement passed as the first argument. +** +** The "end_block" field may contain either an integer, or a text field +** containing the text representation of two non-negative integers separated +** by one or more space (0x20) characters. In the first case, set *piEndBlock +** to the integer value and *pnByte to zero before returning. In the second, +** set *piEndBlock to the first value and *pnByte to the second. +*/ +static void fts3ReadEndBlockField( + sqlite3_stmt *pStmt, + int iCol, + i64 *piEndBlock, + i64 *pnByte +){ + const unsigned char *zText = sqlite3_column_text(pStmt, iCol); + if( zText ){ + int i; + int iMul = 1; + i64 iVal = 0; + for(i=0; zText[i]>='0' && zText[i]<='9'; i++){ + iVal = iVal*10 + (zText[i] - '0'); + } + *piEndBlock = iVal; + while( zText[i]==' ' ) i++; + iVal = 0; + if( zText[i]=='-' ){ + i++; + iMul = -1; + } + for(/* no-op */; zText[i]>='0' && zText[i]<='9'; i++){ + iVal = iVal*10 + (zText[i] - '0'); + } + *pnByte = (iVal * (i64)iMul); + } +} + + +/* +** A segment of size nByte bytes has just been written to absolute level +** iAbsLevel. Promote any segments that should be promoted as a result. +*/ +static int fts3PromoteSegments( + Fts3Table *p, /* FTS table handle */ + sqlite3_int64 iAbsLevel, /* Absolute level just updated */ + sqlite3_int64 nByte /* Size of new segment at iAbsLevel */ +){ + int rc = SQLITE_OK; + sqlite3_stmt *pRange; + + rc = fts3SqlStmt(p, SQL_SELECT_LEVEL_RANGE2, &pRange, 0); + + if( rc==SQLITE_OK ){ + int bOk = 0; + i64 iLast = (iAbsLevel/FTS3_SEGDIR_MAXLEVEL + 1) * FTS3_SEGDIR_MAXLEVEL - 1; + i64 nLimit = (nByte*3)/2; + + /* Loop through all entries in the %_segdir table corresponding to + ** segments in this index on levels greater than iAbsLevel. If there is + ** at least one such segment, and it is possible to determine that all + ** such segments are smaller than nLimit bytes in size, they will be + ** promoted to level iAbsLevel. */ + sqlite3_bind_int64(pRange, 1, iAbsLevel+1); + sqlite3_bind_int64(pRange, 2, iLast); + while( SQLITE_ROW==sqlite3_step(pRange) ){ + i64 nSize = 0, dummy; + fts3ReadEndBlockField(pRange, 2, &dummy, &nSize); + if( nSize<=0 || nSize>nLimit ){ + /* If nSize==0, then the %_segdir.end_block field does not not + ** contain a size value. This happens if it was written by an + ** old version of FTS. In this case it is not possible to determine + ** the size of the segment, and so segment promotion does not + ** take place. */ + bOk = 0; + break; + } + bOk = 1; + } + rc = sqlite3_reset(pRange); + + if( bOk ){ + int iIdx = 0; + sqlite3_stmt *pUpdate1 = 0; + sqlite3_stmt *pUpdate2 = 0; + + if( rc==SQLITE_OK ){ + rc = fts3SqlStmt(p, SQL_UPDATE_LEVEL_IDX, &pUpdate1, 0); + } + if( rc==SQLITE_OK ){ + rc = fts3SqlStmt(p, SQL_UPDATE_LEVEL, &pUpdate2, 0); + } + + if( rc==SQLITE_OK ){ + + /* Loop through all %_segdir entries for segments in this index with + ** levels equal to or greater than iAbsLevel. As each entry is visited, + ** updated it to set (level = -1) and (idx = N), where N is 0 for the + ** oldest segment in the range, 1 for the next oldest, and so on. + ** + ** In other words, move all segments being promoted to level -1, + ** setting the "idx" fields as appropriate to keep them in the same + ** order. The contents of level -1 (which is never used, except + ** transiently here), will be moved back to level iAbsLevel below. */ + sqlite3_bind_int64(pRange, 1, iAbsLevel); + while( SQLITE_ROW==sqlite3_step(pRange) ){ + sqlite3_bind_int(pUpdate1, 1, iIdx++); + sqlite3_bind_int(pUpdate1, 2, sqlite3_column_int(pRange, 0)); + sqlite3_bind_int(pUpdate1, 3, sqlite3_column_int(pRange, 1)); + sqlite3_step(pUpdate1); + rc = sqlite3_reset(pUpdate1); + if( rc!=SQLITE_OK ){ + sqlite3_reset(pRange); + break; + } + } + } + if( rc==SQLITE_OK ){ + rc = sqlite3_reset(pRange); + } + + /* Move level -1 to level iAbsLevel */ + if( rc==SQLITE_OK ){ + sqlite3_bind_int64(pUpdate2, 1, iAbsLevel); + sqlite3_step(pUpdate2); + rc = sqlite3_reset(pUpdate2); + } + } + } + + + return rc; +} + +/* +** Merge all level iLevel segments in the database into a single +** iLevel+1 segment. Or, if iLevel<0, merge all segments into a +** single segment with a level equal to the numerically largest level +** currently present in the database. +** +** If this function is called with iLevel<0, but there is only one +** segment in the database, SQLITE_DONE is returned immediately. +** Otherwise, if successful, SQLITE_OK is returned. If an error occurs, +** an SQLite error code is returned. +*/ +static int fts3SegmentMerge( + Fts3Table *p, + int iLangid, /* Language id to merge */ + int iIndex, /* Index in p->aIndex[] to merge */ + int iLevel /* Level to merge */ +){ + int rc; /* Return code */ + int iIdx = 0; /* Index of new segment */ + sqlite3_int64 iNewLevel = 0; /* Level/index to create new segment at */ + SegmentWriter *pWriter = 0; /* Used to write the new, merged, segment */ + Fts3SegFilter filter; /* Segment term filter condition */ + Fts3MultiSegReader csr; /* Cursor to iterate through level(s) */ + int bIgnoreEmpty = 0; /* True to ignore empty segments */ + i64 iMaxLevel = 0; /* Max level number for this index/langid */ + + assert( iLevel==FTS3_SEGCURSOR_ALL + || iLevel==FTS3_SEGCURSOR_PENDING + || iLevel>=0 + ); + assert( iLevel=0 && iIndexnIndex ); + + rc = sqlite3Fts3SegReaderCursor(p, iLangid, iIndex, iLevel, 0, 0, 1, 0, &csr); + if( rc!=SQLITE_OK || csr.nSegment==0 ) goto finished; + + if( iLevel!=FTS3_SEGCURSOR_PENDING ){ + rc = fts3SegmentMaxLevel(p, iLangid, iIndex, &iMaxLevel); + if( rc!=SQLITE_OK ) goto finished; + } + + if( iLevel==FTS3_SEGCURSOR_ALL ){ + /* This call is to merge all segments in the database to a single + ** segment. The level of the new segment is equal to the numerically + ** greatest segment level currently present in the database for this + ** index. The idx of the new segment is always 0. */ + if( csr.nSegment==1 && 0==fts3SegReaderIsPending(csr.apSegment[0]) ){ + rc = SQLITE_DONE; + goto finished; + } + iNewLevel = iMaxLevel; + bIgnoreEmpty = 1; + + }else{ + /* This call is to merge all segments at level iLevel. find the next + ** available segment index at level iLevel+1. The call to + ** fts3AllocateSegdirIdx() will merge the segments at level iLevel+1 to + ** a single iLevel+2 segment if necessary. */ + assert( FTS3_SEGCURSOR_PENDING==-1 ); + iNewLevel = getAbsoluteLevel(p, iLangid, iIndex, iLevel+1); + rc = fts3AllocateSegdirIdx(p, iLangid, iIndex, iLevel+1, &iIdx); + bIgnoreEmpty = (iLevel!=FTS3_SEGCURSOR_PENDING) && (iNewLevel>iMaxLevel); + } + if( rc!=SQLITE_OK ) goto finished; + + assert( csr.nSegment>0 ); + assert_fts3_nc( iNewLevel>=getAbsoluteLevel(p, iLangid, iIndex, 0) ); + assert_fts3_nc( + iNewLevelnLeafData); + } + } + } + + finished: + fts3SegWriterFree(pWriter); + sqlite3Fts3SegReaderFinish(&csr); + return rc; +} + + +/* +** Flush the contents of pendingTerms to level 0 segments. +*/ +SQLITE_PRIVATE int sqlite3Fts3PendingTermsFlush(Fts3Table *p){ + int rc = SQLITE_OK; + int i; + + for(i=0; rc==SQLITE_OK && inIndex; i++){ + rc = fts3SegmentMerge(p, p->iPrevLangid, i, FTS3_SEGCURSOR_PENDING); + if( rc==SQLITE_DONE ) rc = SQLITE_OK; + } + sqlite3Fts3PendingTermsClear(p); + + /* Determine the auto-incr-merge setting if unknown. If enabled, + ** estimate the number of leaf blocks of content to be written + */ + if( rc==SQLITE_OK && p->bHasStat + && p->nAutoincrmerge==0xff && p->nLeafAdd>0 + ){ + sqlite3_stmt *pStmt = 0; + rc = fts3SqlStmt(p, SQL_SELECT_STAT, &pStmt, 0); + if( rc==SQLITE_OK ){ + sqlite3_bind_int(pStmt, 1, FTS_STAT_AUTOINCRMERGE); + rc = sqlite3_step(pStmt); + if( rc==SQLITE_ROW ){ + p->nAutoincrmerge = sqlite3_column_int(pStmt, 0); + if( p->nAutoincrmerge==1 ) p->nAutoincrmerge = 8; + }else if( rc==SQLITE_DONE ){ + p->nAutoincrmerge = 0; + } + rc = sqlite3_reset(pStmt); + } + } + return rc; +} + +/* +** Encode N integers as varints into a blob. +*/ +static void fts3EncodeIntArray( + int N, /* The number of integers to encode */ + u32 *a, /* The integer values */ + char *zBuf, /* Write the BLOB here */ + int *pNBuf /* Write number of bytes if zBuf[] used here */ +){ + int i, j; + for(i=j=0; iiPrevDocid. The sizes are encoded as +** a blob of varints. +*/ +static void fts3InsertDocsize( + int *pRC, /* Result code */ + Fts3Table *p, /* Table into which to insert */ + u32 *aSz /* Sizes of each column, in tokens */ +){ + char *pBlob; /* The BLOB encoding of the document size */ + int nBlob; /* Number of bytes in the BLOB */ + sqlite3_stmt *pStmt; /* Statement used to insert the encoding */ + int rc; /* Result code from subfunctions */ + + if( *pRC ) return; + pBlob = sqlite3_malloc64( 10*(sqlite3_int64)p->nColumn ); + if( pBlob==0 ){ + *pRC = SQLITE_NOMEM; + return; + } + fts3EncodeIntArray(p->nColumn, aSz, pBlob, &nBlob); + rc = fts3SqlStmt(p, SQL_REPLACE_DOCSIZE, &pStmt, 0); + if( rc ){ + sqlite3_free(pBlob); + *pRC = rc; + return; + } + sqlite3_bind_int64(pStmt, 1, p->iPrevDocid); + sqlite3_bind_blob(pStmt, 2, pBlob, nBlob, sqlite3_free); + sqlite3_step(pStmt); + *pRC = sqlite3_reset(pStmt); +} + +/* +** Record 0 of the %_stat table contains a blob consisting of N varints, +** where N is the number of user defined columns in the fts3 table plus +** two. If nCol is the number of user defined columns, then values of the +** varints are set as follows: +** +** Varint 0: Total number of rows in the table. +** +** Varint 1..nCol: For each column, the total number of tokens stored in +** the column for all rows of the table. +** +** Varint 1+nCol: The total size, in bytes, of all text values in all +** columns of all rows of the table. +** +*/ +static void fts3UpdateDocTotals( + int *pRC, /* The result code */ + Fts3Table *p, /* Table being updated */ + u32 *aSzIns, /* Size increases */ + u32 *aSzDel, /* Size decreases */ + int nChng /* Change in the number of documents */ +){ + char *pBlob; /* Storage for BLOB written into %_stat */ + int nBlob; /* Size of BLOB written into %_stat */ + u32 *a; /* Array of integers that becomes the BLOB */ + sqlite3_stmt *pStmt; /* Statement for reading and writing */ + int i; /* Loop counter */ + int rc; /* Result code from subfunctions */ + + const int nStat = p->nColumn+2; + + if( *pRC ) return; + a = sqlite3_malloc64( (sizeof(u32)+10)*(sqlite3_int64)nStat ); + if( a==0 ){ + *pRC = SQLITE_NOMEM; + return; + } + pBlob = (char*)&a[nStat]; + rc = fts3SqlStmt(p, SQL_SELECT_STAT, &pStmt, 0); + if( rc ){ + sqlite3_free(a); + *pRC = rc; + return; + } + sqlite3_bind_int(pStmt, 1, FTS_STAT_DOCTOTAL); + if( sqlite3_step(pStmt)==SQLITE_ROW ){ + fts3DecodeIntArray(nStat, a, + sqlite3_column_blob(pStmt, 0), + sqlite3_column_bytes(pStmt, 0)); + }else{ + memset(a, 0, sizeof(u32)*(nStat) ); + } + rc = sqlite3_reset(pStmt); + if( rc!=SQLITE_OK ){ + sqlite3_free(a); + *pRC = rc; + return; + } + if( nChng<0 && a[0]<(u32)(-nChng) ){ + a[0] = 0; + }else{ + a[0] += nChng; + } + for(i=0; inColumn+1; i++){ + u32 x = a[i+1]; + if( x+aSzIns[i] < aSzDel[i] ){ + x = 0; + }else{ + x = x + aSzIns[i] - aSzDel[i]; + } + a[i+1] = x; + } + fts3EncodeIntArray(nStat, a, pBlob, &nBlob); + rc = fts3SqlStmt(p, SQL_REPLACE_STAT, &pStmt, 0); + if( rc ){ + sqlite3_free(a); + *pRC = rc; + return; + } + sqlite3_bind_int(pStmt, 1, FTS_STAT_DOCTOTAL); + sqlite3_bind_blob(pStmt, 2, pBlob, nBlob, SQLITE_STATIC); + sqlite3_step(pStmt); + *pRC = sqlite3_reset(pStmt); + sqlite3_bind_null(pStmt, 2); + sqlite3_free(a); +} + +/* +** Merge the entire database so that there is one segment for each +** iIndex/iLangid combination. +*/ +static int fts3DoOptimize(Fts3Table *p, int bReturnDone){ + int bSeenDone = 0; + int rc; + sqlite3_stmt *pAllLangid = 0; + + rc = fts3SqlStmt(p, SQL_SELECT_ALL_LANGID, &pAllLangid, 0); + if( rc==SQLITE_OK ){ + int rc2; + sqlite3_bind_int(pAllLangid, 1, p->iPrevLangid); + sqlite3_bind_int(pAllLangid, 2, p->nIndex); + while( sqlite3_step(pAllLangid)==SQLITE_ROW ){ + int i; + int iLangid = sqlite3_column_int(pAllLangid, 0); + for(i=0; rc==SQLITE_OK && inIndex; i++){ + rc = fts3SegmentMerge(p, iLangid, i, FTS3_SEGCURSOR_ALL); + if( rc==SQLITE_DONE ){ + bSeenDone = 1; + rc = SQLITE_OK; + } + } + } + rc2 = sqlite3_reset(pAllLangid); + if( rc==SQLITE_OK ) rc = rc2; + } + + sqlite3Fts3SegmentsClose(p); + sqlite3Fts3PendingTermsClear(p); + + return (rc==SQLITE_OK && bReturnDone && bSeenDone) ? SQLITE_DONE : rc; +} + +/* +** This function is called when the user executes the following statement: +** +** INSERT INTO () VALUES('rebuild'); +** +** The entire FTS index is discarded and rebuilt. If the table is one +** created using the content=xxx option, then the new index is based on +** the current contents of the xxx table. Otherwise, it is rebuilt based +** on the contents of the %_content table. +*/ +static int fts3DoRebuild(Fts3Table *p){ + int rc; /* Return Code */ + + rc = fts3DeleteAll(p, 0); + if( rc==SQLITE_OK ){ + u32 *aSz = 0; + u32 *aSzIns = 0; + u32 *aSzDel = 0; + sqlite3_stmt *pStmt = 0; + int nEntry = 0; + + /* Compose and prepare an SQL statement to loop through the content table */ + char *zSql = sqlite3_mprintf("SELECT %s" , p->zReadExprlist); + if( !zSql ){ + rc = SQLITE_NOMEM; + }else{ + rc = sqlite3_prepare_v2(p->db, zSql, -1, &pStmt, 0); + sqlite3_free(zSql); + } + + if( rc==SQLITE_OK ){ + sqlite3_int64 nByte = sizeof(u32) * ((sqlite3_int64)p->nColumn+1)*3; + aSz = (u32 *)sqlite3_malloc64(nByte); + if( aSz==0 ){ + rc = SQLITE_NOMEM; + }else{ + memset(aSz, 0, nByte); + aSzIns = &aSz[p->nColumn+1]; + aSzDel = &aSzIns[p->nColumn+1]; + } + } + + while( rc==SQLITE_OK && SQLITE_ROW==sqlite3_step(pStmt) ){ + int iCol; + int iLangid = langidFromSelect(p, pStmt); + rc = fts3PendingTermsDocid(p, 0, iLangid, sqlite3_column_int64(pStmt, 0)); + memset(aSz, 0, sizeof(aSz[0]) * (p->nColumn+1)); + for(iCol=0; rc==SQLITE_OK && iColnColumn; iCol++){ + if( p->abNotindexed[iCol]==0 ){ + const char *z = (const char *) sqlite3_column_text(pStmt, iCol+1); + rc = fts3PendingTermsAdd(p, iLangid, z, iCol, &aSz[iCol]); + aSz[p->nColumn] += sqlite3_column_bytes(pStmt, iCol+1); + } + } + if( p->bHasDocsize ){ + fts3InsertDocsize(&rc, p, aSz); + } + if( rc!=SQLITE_OK ){ + sqlite3_finalize(pStmt); + pStmt = 0; + }else{ + nEntry++; + for(iCol=0; iCol<=p->nColumn; iCol++){ + aSzIns[iCol] += aSz[iCol]; + } + } + } + if( p->bFts4 ){ + fts3UpdateDocTotals(&rc, p, aSzIns, aSzDel, nEntry); + } + sqlite3_free(aSz); + + if( pStmt ){ + int rc2 = sqlite3_finalize(pStmt); + if( rc==SQLITE_OK ){ + rc = rc2; + } + } + } + + return rc; +} + + +/* +** This function opens a cursor used to read the input data for an +** incremental merge operation. Specifically, it opens a cursor to scan +** the oldest nSeg segments (idx=0 through idx=(nSeg-1)) in absolute +** level iAbsLevel. +*/ +static int fts3IncrmergeCsr( + Fts3Table *p, /* FTS3 table handle */ + sqlite3_int64 iAbsLevel, /* Absolute level to open */ + int nSeg, /* Number of segments to merge */ + Fts3MultiSegReader *pCsr /* Cursor object to populate */ +){ + int rc; /* Return Code */ + sqlite3_stmt *pStmt = 0; /* Statement used to read %_segdir entry */ + sqlite3_int64 nByte; /* Bytes allocated at pCsr->apSegment[] */ + + /* Allocate space for the Fts3MultiSegReader.aCsr[] array */ + memset(pCsr, 0, sizeof(*pCsr)); + nByte = sizeof(Fts3SegReader *) * nSeg; + pCsr->apSegment = (Fts3SegReader **)sqlite3_malloc64(nByte); + + if( pCsr->apSegment==0 ){ + rc = SQLITE_NOMEM; + }else{ + memset(pCsr->apSegment, 0, nByte); + rc = fts3SqlStmt(p, SQL_SELECT_LEVEL, &pStmt, 0); + } + if( rc==SQLITE_OK ){ + int i; + int rc2; + sqlite3_bind_int64(pStmt, 1, iAbsLevel); + assert( pCsr->nSegment==0 ); + for(i=0; rc==SQLITE_OK && sqlite3_step(pStmt)==SQLITE_ROW && iapSegment[i] + ); + pCsr->nSegment++; + } + rc2 = sqlite3_reset(pStmt); + if( rc==SQLITE_OK ) rc = rc2; + } + + return rc; +} + +typedef struct IncrmergeWriter IncrmergeWriter; +typedef struct NodeWriter NodeWriter; +typedef struct Blob Blob; +typedef struct NodeReader NodeReader; + +/* +** An instance of the following structure is used as a dynamic buffer +** to build up nodes or other blobs of data in. +** +** The function blobGrowBuffer() is used to extend the allocation. +*/ +struct Blob { + char *a; /* Pointer to allocation */ + int n; /* Number of valid bytes of data in a[] */ + int nAlloc; /* Allocated size of a[] (nAlloc>=n) */ +}; + +/* +** This structure is used to build up buffers containing segment b-tree +** nodes (blocks). +*/ +struct NodeWriter { + sqlite3_int64 iBlock; /* Current block id */ + Blob key; /* Last key written to the current block */ + Blob block; /* Current block image */ +}; + +/* +** An object of this type contains the state required to create or append +** to an appendable b-tree segment. +*/ +struct IncrmergeWriter { + int nLeafEst; /* Space allocated for leaf blocks */ + int nWork; /* Number of leaf pages flushed */ + sqlite3_int64 iAbsLevel; /* Absolute level of input segments */ + int iIdx; /* Index of *output* segment in iAbsLevel+1 */ + sqlite3_int64 iStart; /* Block number of first allocated block */ + sqlite3_int64 iEnd; /* Block number of last allocated block */ + sqlite3_int64 nLeafData; /* Bytes of leaf page data so far */ + u8 bNoLeafData; /* If true, store 0 for segment size */ + NodeWriter aNodeWriter[FTS_MAX_APPENDABLE_HEIGHT]; +}; + +/* +** An object of the following type is used to read data from a single +** FTS segment node. See the following functions: +** +** nodeReaderInit() +** nodeReaderNext() +** nodeReaderRelease() +*/ +struct NodeReader { + const char *aNode; + int nNode; + int iOff; /* Current offset within aNode[] */ + + /* Output variables. Containing the current node entry. */ + sqlite3_int64 iChild; /* Pointer to child node */ + Blob term; /* Current term */ + const char *aDoclist; /* Pointer to doclist */ + int nDoclist; /* Size of doclist in bytes */ +}; + +/* +** If *pRc is not SQLITE_OK when this function is called, it is a no-op. +** Otherwise, if the allocation at pBlob->a is not already at least nMin +** bytes in size, extend (realloc) it to be so. +** +** If an OOM error occurs, set *pRc to SQLITE_NOMEM and leave pBlob->a +** unmodified. Otherwise, if the allocation succeeds, update pBlob->nAlloc +** to reflect the new size of the pBlob->a[] buffer. +*/ +static void blobGrowBuffer(Blob *pBlob, int nMin, int *pRc){ + if( *pRc==SQLITE_OK && nMin>pBlob->nAlloc ){ + int nAlloc = nMin; + char *a = (char *)sqlite3_realloc(pBlob->a, nAlloc); + if( a ){ + pBlob->nAlloc = nAlloc; + pBlob->a = a; + }else{ + *pRc = SQLITE_NOMEM; + } + } +} + +/* +** Attempt to advance the node-reader object passed as the first argument to +** the next entry on the node. +** +** Return an error code if an error occurs (SQLITE_NOMEM is possible). +** Otherwise return SQLITE_OK. If there is no next entry on the node +** (e.g. because the current entry is the last) set NodeReader->aNode to +** NULL to indicate EOF. Otherwise, populate the NodeReader structure output +** variables for the new entry. +*/ +static int nodeReaderNext(NodeReader *p){ + int bFirst = (p->term.n==0); /* True for first term on the node */ + int nPrefix = 0; /* Bytes to copy from previous term */ + int nSuffix = 0; /* Bytes to append to the prefix */ + int rc = SQLITE_OK; /* Return code */ + + assert( p->aNode ); + if( p->iChild && bFirst==0 ) p->iChild++; + if( p->iOff>=p->nNode ){ + /* EOF */ + p->aNode = 0; + }else{ + if( bFirst==0 ){ + p->iOff += fts3GetVarint32(&p->aNode[p->iOff], &nPrefix); + } + p->iOff += fts3GetVarint32(&p->aNode[p->iOff], &nSuffix); + + if( nPrefix>p->term.n || nSuffix>p->nNode-p->iOff || nSuffix==0 ){ + return FTS_CORRUPT_VTAB; + } + blobGrowBuffer(&p->term, nPrefix+nSuffix, &rc); + if( rc==SQLITE_OK ){ + memcpy(&p->term.a[nPrefix], &p->aNode[p->iOff], nSuffix); + p->term.n = nPrefix+nSuffix; + p->iOff += nSuffix; + if( p->iChild==0 ){ + p->iOff += fts3GetVarint32(&p->aNode[p->iOff], &p->nDoclist); + if( (p->nNode-p->iOff)nDoclist ){ + return FTS_CORRUPT_VTAB; + } + p->aDoclist = &p->aNode[p->iOff]; + p->iOff += p->nDoclist; + } + } + } + + assert_fts3_nc( p->iOff<=p->nNode ); + return rc; +} + +/* +** Release all dynamic resources held by node-reader object *p. +*/ +static void nodeReaderRelease(NodeReader *p){ + sqlite3_free(p->term.a); +} + +/* +** Initialize a node-reader object to read the node in buffer aNode/nNode. +** +** If successful, SQLITE_OK is returned and the NodeReader object set to +** point to the first entry on the node (if any). Otherwise, an SQLite +** error code is returned. +*/ +static int nodeReaderInit(NodeReader *p, const char *aNode, int nNode){ + memset(p, 0, sizeof(NodeReader)); + p->aNode = aNode; + p->nNode = nNode; + + /* Figure out if this is a leaf or an internal node. */ + if( aNode && aNode[0] ){ + /* An internal node. */ + p->iOff = 1 + sqlite3Fts3GetVarint(&p->aNode[1], &p->iChild); + }else{ + p->iOff = 1; + } + + return aNode ? nodeReaderNext(p) : SQLITE_OK; +} + +/* +** This function is called while writing an FTS segment each time a leaf o +** node is finished and written to disk. The key (zTerm/nTerm) is guaranteed +** to be greater than the largest key on the node just written, but smaller +** than or equal to the first key that will be written to the next leaf +** node. +** +** The block id of the leaf node just written to disk may be found in +** (pWriter->aNodeWriter[0].iBlock) when this function is called. +*/ +static int fts3IncrmergePush( + Fts3Table *p, /* Fts3 table handle */ + IncrmergeWriter *pWriter, /* Writer object */ + const char *zTerm, /* Term to write to internal node */ + int nTerm /* Bytes at zTerm */ +){ + sqlite3_int64 iPtr = pWriter->aNodeWriter[0].iBlock; + int iLayer; + + assert( nTerm>0 ); + for(iLayer=1; ALWAYS(iLayeraNodeWriter[iLayer]; + int rc = SQLITE_OK; + int nPrefix; + int nSuffix; + int nSpace; + + /* Figure out how much space the key will consume if it is written to + ** the current node of layer iLayer. Due to the prefix compression, + ** the space required changes depending on which node the key is to + ** be added to. */ + nPrefix = fts3PrefixCompress(pNode->key.a, pNode->key.n, zTerm, nTerm); + nSuffix = nTerm - nPrefix; + nSpace = sqlite3Fts3VarintLen(nPrefix); + nSpace += sqlite3Fts3VarintLen(nSuffix) + nSuffix; + + if( pNode->key.n==0 || (pNode->block.n + nSpace)<=p->nNodeSize ){ + /* If the current node of layer iLayer contains zero keys, or if adding + ** the key to it will not cause it to grow to larger than nNodeSize + ** bytes in size, write the key here. */ + + Blob *pBlk = &pNode->block; + if( pBlk->n==0 ){ + blobGrowBuffer(pBlk, p->nNodeSize, &rc); + if( rc==SQLITE_OK ){ + pBlk->a[0] = (char)iLayer; + pBlk->n = 1 + sqlite3Fts3PutVarint(&pBlk->a[1], iPtr); + } + } + blobGrowBuffer(pBlk, pBlk->n + nSpace, &rc); + blobGrowBuffer(&pNode->key, nTerm, &rc); + + if( rc==SQLITE_OK ){ + if( pNode->key.n ){ + pBlk->n += sqlite3Fts3PutVarint(&pBlk->a[pBlk->n], nPrefix); + } + pBlk->n += sqlite3Fts3PutVarint(&pBlk->a[pBlk->n], nSuffix); + memcpy(&pBlk->a[pBlk->n], &zTerm[nPrefix], nSuffix); + pBlk->n += nSuffix; + + memcpy(pNode->key.a, zTerm, nTerm); + pNode->key.n = nTerm; + } + }else{ + /* Otherwise, flush the current node of layer iLayer to disk. + ** Then allocate a new, empty sibling node. The key will be written + ** into the parent of this node. */ + rc = fts3WriteSegment(p, pNode->iBlock, pNode->block.a, pNode->block.n); + + assert( pNode->block.nAlloc>=p->nNodeSize ); + pNode->block.a[0] = (char)iLayer; + pNode->block.n = 1 + sqlite3Fts3PutVarint(&pNode->block.a[1], iPtr+1); + + iNextPtr = pNode->iBlock; + pNode->iBlock++; + pNode->key.n = 0; + } + + if( rc!=SQLITE_OK || iNextPtr==0 ) return rc; + iPtr = iNextPtr; + } + + assert( 0 ); + return 0; +} + +/* +** Append a term and (optionally) doclist to the FTS segment node currently +** stored in blob *pNode. The node need not contain any terms, but the +** header must be written before this function is called. +** +** A node header is a single 0x00 byte for a leaf node, or a height varint +** followed by the left-hand-child varint for an internal node. +** +** The term to be appended is passed via arguments zTerm/nTerm. For a +** leaf node, the doclist is passed as aDoclist/nDoclist. For an internal +** node, both aDoclist and nDoclist must be passed 0. +** +** If the size of the value in blob pPrev is zero, then this is the first +** term written to the node. Otherwise, pPrev contains a copy of the +** previous term. Before this function returns, it is updated to contain a +** copy of zTerm/nTerm. +** +** It is assumed that the buffer associated with pNode is already large +** enough to accommodate the new entry. The buffer associated with pPrev +** is extended by this function if requrired. +** +** If an error (i.e. OOM condition) occurs, an SQLite error code is +** returned. Otherwise, SQLITE_OK. +*/ +static int fts3AppendToNode( + Blob *pNode, /* Current node image to append to */ + Blob *pPrev, /* Buffer containing previous term written */ + const char *zTerm, /* New term to write */ + int nTerm, /* Size of zTerm in bytes */ + const char *aDoclist, /* Doclist (or NULL) to write */ + int nDoclist /* Size of aDoclist in bytes */ +){ + int rc = SQLITE_OK; /* Return code */ + int bFirst = (pPrev->n==0); /* True if this is the first term written */ + int nPrefix; /* Size of term prefix in bytes */ + int nSuffix; /* Size of term suffix in bytes */ + + /* Node must have already been started. There must be a doclist for a + ** leaf node, and there must not be a doclist for an internal node. */ + assert( pNode->n>0 ); + assert_fts3_nc( (pNode->a[0]=='\0')==(aDoclist!=0) ); + + blobGrowBuffer(pPrev, nTerm, &rc); + if( rc!=SQLITE_OK ) return rc; + + nPrefix = fts3PrefixCompress(pPrev->a, pPrev->n, zTerm, nTerm); + nSuffix = nTerm - nPrefix; + if( nSuffix<=0 ) return FTS_CORRUPT_VTAB; + memcpy(pPrev->a, zTerm, nTerm); + pPrev->n = nTerm; + + if( bFirst==0 ){ + pNode->n += sqlite3Fts3PutVarint(&pNode->a[pNode->n], nPrefix); + } + pNode->n += sqlite3Fts3PutVarint(&pNode->a[pNode->n], nSuffix); + memcpy(&pNode->a[pNode->n], &zTerm[nPrefix], nSuffix); + pNode->n += nSuffix; + + if( aDoclist ){ + pNode->n += sqlite3Fts3PutVarint(&pNode->a[pNode->n], nDoclist); + memcpy(&pNode->a[pNode->n], aDoclist, nDoclist); + pNode->n += nDoclist; + } + + assert( pNode->n<=pNode->nAlloc ); + + return SQLITE_OK; +} + +/* +** Append the current term and doclist pointed to by cursor pCsr to the +** appendable b-tree segment opened for writing by pWriter. +** +** Return SQLITE_OK if successful, or an SQLite error code otherwise. +*/ +static int fts3IncrmergeAppend( + Fts3Table *p, /* Fts3 table handle */ + IncrmergeWriter *pWriter, /* Writer object */ + Fts3MultiSegReader *pCsr /* Cursor containing term and doclist */ +){ + const char *zTerm = pCsr->zTerm; + int nTerm = pCsr->nTerm; + const char *aDoclist = pCsr->aDoclist; + int nDoclist = pCsr->nDoclist; + int rc = SQLITE_OK; /* Return code */ + int nSpace; /* Total space in bytes required on leaf */ + int nPrefix; /* Size of prefix shared with previous term */ + int nSuffix; /* Size of suffix (nTerm - nPrefix) */ + NodeWriter *pLeaf; /* Object used to write leaf nodes */ + + pLeaf = &pWriter->aNodeWriter[0]; + nPrefix = fts3PrefixCompress(pLeaf->key.a, pLeaf->key.n, zTerm, nTerm); + nSuffix = nTerm - nPrefix; + + nSpace = sqlite3Fts3VarintLen(nPrefix); + nSpace += sqlite3Fts3VarintLen(nSuffix) + nSuffix; + nSpace += sqlite3Fts3VarintLen(nDoclist) + nDoclist; + + /* If the current block is not empty, and if adding this term/doclist + ** to the current block would make it larger than Fts3Table.nNodeSize + ** bytes, write this block out to the database. */ + if( pLeaf->block.n>0 && (pLeaf->block.n + nSpace)>p->nNodeSize ){ + rc = fts3WriteSegment(p, pLeaf->iBlock, pLeaf->block.a, pLeaf->block.n); + pWriter->nWork++; + + /* Add the current term to the parent node. The term added to the + ** parent must: + ** + ** a) be greater than the largest term on the leaf node just written + ** to the database (still available in pLeaf->key), and + ** + ** b) be less than or equal to the term about to be added to the new + ** leaf node (zTerm/nTerm). + ** + ** In other words, it must be the prefix of zTerm 1 byte longer than + ** the common prefix (if any) of zTerm and pWriter->zTerm. + */ + if( rc==SQLITE_OK ){ + rc = fts3IncrmergePush(p, pWriter, zTerm, nPrefix+1); + } + + /* Advance to the next output block */ + pLeaf->iBlock++; + pLeaf->key.n = 0; + pLeaf->block.n = 0; + + nSuffix = nTerm; + nSpace = 1; + nSpace += sqlite3Fts3VarintLen(nSuffix) + nSuffix; + nSpace += sqlite3Fts3VarintLen(nDoclist) + nDoclist; + } + + pWriter->nLeafData += nSpace; + blobGrowBuffer(&pLeaf->block, pLeaf->block.n + nSpace, &rc); + if( rc==SQLITE_OK ){ + if( pLeaf->block.n==0 ){ + pLeaf->block.n = 1; + pLeaf->block.a[0] = '\0'; + } + rc = fts3AppendToNode( + &pLeaf->block, &pLeaf->key, zTerm, nTerm, aDoclist, nDoclist + ); + } + + return rc; +} + +/* +** This function is called to release all dynamic resources held by the +** merge-writer object pWriter, and if no error has occurred, to flush +** all outstanding node buffers held by pWriter to disk. +** +** If *pRc is not SQLITE_OK when this function is called, then no attempt +** is made to write any data to disk. Instead, this function serves only +** to release outstanding resources. +** +** Otherwise, if *pRc is initially SQLITE_OK and an error occurs while +** flushing buffers to disk, *pRc is set to an SQLite error code before +** returning. +*/ +static void fts3IncrmergeRelease( + Fts3Table *p, /* FTS3 table handle */ + IncrmergeWriter *pWriter, /* Merge-writer object */ + int *pRc /* IN/OUT: Error code */ +){ + int i; /* Used to iterate through non-root layers */ + int iRoot; /* Index of root in pWriter->aNodeWriter */ + NodeWriter *pRoot; /* NodeWriter for root node */ + int rc = *pRc; /* Error code */ + + /* Set iRoot to the index in pWriter->aNodeWriter[] of the output segment + ** root node. If the segment fits entirely on a single leaf node, iRoot + ** will be set to 0. If the root node is the parent of the leaves, iRoot + ** will be 1. And so on. */ + for(iRoot=FTS_MAX_APPENDABLE_HEIGHT-1; iRoot>=0; iRoot--){ + NodeWriter *pNode = &pWriter->aNodeWriter[iRoot]; + if( pNode->block.n>0 ) break; + assert( *pRc || pNode->block.nAlloc==0 ); + assert( *pRc || pNode->key.nAlloc==0 ); + sqlite3_free(pNode->block.a); + sqlite3_free(pNode->key.a); + } + + /* Empty output segment. This is a no-op. */ + if( iRoot<0 ) return; + + /* The entire output segment fits on a single node. Normally, this means + ** the node would be stored as a blob in the "root" column of the %_segdir + ** table. However, this is not permitted in this case. The problem is that + ** space has already been reserved in the %_segments table, and so the + ** start_block and end_block fields of the %_segdir table must be populated. + ** And, by design or by accident, released versions of FTS cannot handle + ** segments that fit entirely on the root node with start_block!=0. + ** + ** Instead, create a synthetic root node that contains nothing but a + ** pointer to the single content node. So that the segment consists of a + ** single leaf and a single interior (root) node. + ** + ** Todo: Better might be to defer allocating space in the %_segments + ** table until we are sure it is needed. + */ + if( iRoot==0 ){ + Blob *pBlock = &pWriter->aNodeWriter[1].block; + blobGrowBuffer(pBlock, 1 + FTS3_VARINT_MAX, &rc); + if( rc==SQLITE_OK ){ + pBlock->a[0] = 0x01; + pBlock->n = 1 + sqlite3Fts3PutVarint( + &pBlock->a[1], pWriter->aNodeWriter[0].iBlock + ); + } + iRoot = 1; + } + pRoot = &pWriter->aNodeWriter[iRoot]; + + /* Flush all currently outstanding nodes to disk. */ + for(i=0; iaNodeWriter[i]; + if( pNode->block.n>0 && rc==SQLITE_OK ){ + rc = fts3WriteSegment(p, pNode->iBlock, pNode->block.a, pNode->block.n); + } + sqlite3_free(pNode->block.a); + sqlite3_free(pNode->key.a); + } + + /* Write the %_segdir record. */ + if( rc==SQLITE_OK ){ + rc = fts3WriteSegdir(p, + pWriter->iAbsLevel+1, /* level */ + pWriter->iIdx, /* idx */ + pWriter->iStart, /* start_block */ + pWriter->aNodeWriter[0].iBlock, /* leaves_end_block */ + pWriter->iEnd, /* end_block */ + (pWriter->bNoLeafData==0 ? pWriter->nLeafData : 0), /* end_block */ + pRoot->block.a, pRoot->block.n /* root */ + ); + } + sqlite3_free(pRoot->block.a); + sqlite3_free(pRoot->key.a); + + *pRc = rc; +} + +/* +** Compare the term in buffer zLhs (size in bytes nLhs) with that in +** zRhs (size in bytes nRhs) using memcmp. If one term is a prefix of +** the other, it is considered to be smaller than the other. +** +** Return -ve if zLhs is smaller than zRhs, 0 if it is equal, or +ve +** if it is greater. +*/ +static int fts3TermCmp( + const char *zLhs, int nLhs, /* LHS of comparison */ + const char *zRhs, int nRhs /* RHS of comparison */ +){ + int nCmp = MIN(nLhs, nRhs); + int res; + + res = (nCmp ? memcmp(zLhs, zRhs, nCmp) : 0); + if( res==0 ) res = nLhs - nRhs; + + return res; +} + + +/* +** Query to see if the entry in the %_segments table with blockid iEnd is +** NULL. If no error occurs and the entry is NULL, set *pbRes 1 before +** returning. Otherwise, set *pbRes to 0. +** +** Or, if an error occurs while querying the database, return an SQLite +** error code. The final value of *pbRes is undefined in this case. +** +** This is used to test if a segment is an "appendable" segment. If it +** is, then a NULL entry has been inserted into the %_segments table +** with blockid %_segdir.end_block. +*/ +static int fts3IsAppendable(Fts3Table *p, sqlite3_int64 iEnd, int *pbRes){ + int bRes = 0; /* Result to set *pbRes to */ + sqlite3_stmt *pCheck = 0; /* Statement to query database with */ + int rc; /* Return code */ + + rc = fts3SqlStmt(p, SQL_SEGMENT_IS_APPENDABLE, &pCheck, 0); + if( rc==SQLITE_OK ){ + sqlite3_bind_int64(pCheck, 1, iEnd); + if( SQLITE_ROW==sqlite3_step(pCheck) ) bRes = 1; + rc = sqlite3_reset(pCheck); + } + + *pbRes = bRes; + return rc; +} + +/* +** This function is called when initializing an incremental-merge operation. +** It checks if the existing segment with index value iIdx at absolute level +** (iAbsLevel+1) can be appended to by the incremental merge. If it can, the +** merge-writer object *pWriter is initialized to write to it. +** +** An existing segment can be appended to by an incremental merge if: +** +** * It was initially created as an appendable segment (with all required +** space pre-allocated), and +** +** * The first key read from the input (arguments zKey and nKey) is +** greater than the largest key currently stored in the potential +** output segment. +*/ +static int fts3IncrmergeLoad( + Fts3Table *p, /* Fts3 table handle */ + sqlite3_int64 iAbsLevel, /* Absolute level of input segments */ + int iIdx, /* Index of candidate output segment */ + const char *zKey, /* First key to write */ + int nKey, /* Number of bytes in nKey */ + IncrmergeWriter *pWriter /* Populate this object */ +){ + int rc; /* Return code */ + sqlite3_stmt *pSelect = 0; /* SELECT to read %_segdir entry */ + + rc = fts3SqlStmt(p, SQL_SELECT_SEGDIR, &pSelect, 0); + if( rc==SQLITE_OK ){ + sqlite3_int64 iStart = 0; /* Value of %_segdir.start_block */ + sqlite3_int64 iLeafEnd = 0; /* Value of %_segdir.leaves_end_block */ + sqlite3_int64 iEnd = 0; /* Value of %_segdir.end_block */ + const char *aRoot = 0; /* Pointer to %_segdir.root buffer */ + int nRoot = 0; /* Size of aRoot[] in bytes */ + int rc2; /* Return code from sqlite3_reset() */ + int bAppendable = 0; /* Set to true if segment is appendable */ + + /* Read the %_segdir entry for index iIdx absolute level (iAbsLevel+1) */ + sqlite3_bind_int64(pSelect, 1, iAbsLevel+1); + sqlite3_bind_int(pSelect, 2, iIdx); + if( sqlite3_step(pSelect)==SQLITE_ROW ){ + iStart = sqlite3_column_int64(pSelect, 1); + iLeafEnd = sqlite3_column_int64(pSelect, 2); + fts3ReadEndBlockField(pSelect, 3, &iEnd, &pWriter->nLeafData); + if( pWriter->nLeafData<0 ){ + pWriter->nLeafData = pWriter->nLeafData * -1; + } + pWriter->bNoLeafData = (pWriter->nLeafData==0); + nRoot = sqlite3_column_bytes(pSelect, 4); + aRoot = sqlite3_column_blob(pSelect, 4); + }else{ + return sqlite3_reset(pSelect); + } + + /* Check for the zero-length marker in the %_segments table */ + rc = fts3IsAppendable(p, iEnd, &bAppendable); + + /* Check that zKey/nKey is larger than the largest key the candidate */ + if( rc==SQLITE_OK && bAppendable ){ + char *aLeaf = 0; + int nLeaf = 0; + + rc = sqlite3Fts3ReadBlock(p, iLeafEnd, &aLeaf, &nLeaf, 0); + if( rc==SQLITE_OK ){ + NodeReader reader; + for(rc = nodeReaderInit(&reader, aLeaf, nLeaf); + rc==SQLITE_OK && reader.aNode; + rc = nodeReaderNext(&reader) + ){ + assert( reader.aNode ); + } + if( fts3TermCmp(zKey, nKey, reader.term.a, reader.term.n)<=0 ){ + bAppendable = 0; + } + nodeReaderRelease(&reader); + } + sqlite3_free(aLeaf); + } + + if( rc==SQLITE_OK && bAppendable ){ + /* It is possible to append to this segment. Set up the IncrmergeWriter + ** object to do so. */ + int i; + int nHeight = (int)aRoot[0]; + NodeWriter *pNode; + + pWriter->nLeafEst = (int)((iEnd - iStart) + 1)/FTS_MAX_APPENDABLE_HEIGHT; + pWriter->iStart = iStart; + pWriter->iEnd = iEnd; + pWriter->iAbsLevel = iAbsLevel; + pWriter->iIdx = iIdx; + + for(i=nHeight+1; iaNodeWriter[i].iBlock = pWriter->iStart + i*pWriter->nLeafEst; + } + + pNode = &pWriter->aNodeWriter[nHeight]; + pNode->iBlock = pWriter->iStart + pWriter->nLeafEst*nHeight; + blobGrowBuffer(&pNode->block, + MAX(nRoot, p->nNodeSize)+FTS3_NODE_PADDING, &rc + ); + if( rc==SQLITE_OK ){ + memcpy(pNode->block.a, aRoot, nRoot); + pNode->block.n = nRoot; + memset(&pNode->block.a[nRoot], 0, FTS3_NODE_PADDING); + } + + for(i=nHeight; i>=0 && rc==SQLITE_OK; i--){ + NodeReader reader; + pNode = &pWriter->aNodeWriter[i]; + + if( pNode->block.a){ + rc = nodeReaderInit(&reader, pNode->block.a, pNode->block.n); + while( reader.aNode && rc==SQLITE_OK ) rc = nodeReaderNext(&reader); + blobGrowBuffer(&pNode->key, reader.term.n, &rc); + if( rc==SQLITE_OK ){ + memcpy(pNode->key.a, reader.term.a, reader.term.n); + pNode->key.n = reader.term.n; + if( i>0 ){ + char *aBlock = 0; + int nBlock = 0; + pNode = &pWriter->aNodeWriter[i-1]; + pNode->iBlock = reader.iChild; + rc = sqlite3Fts3ReadBlock(p, reader.iChild, &aBlock, &nBlock, 0); + blobGrowBuffer(&pNode->block, + MAX(nBlock, p->nNodeSize)+FTS3_NODE_PADDING, &rc + ); + if( rc==SQLITE_OK ){ + memcpy(pNode->block.a, aBlock, nBlock); + pNode->block.n = nBlock; + memset(&pNode->block.a[nBlock], 0, FTS3_NODE_PADDING); + } + sqlite3_free(aBlock); + } + } + } + nodeReaderRelease(&reader); + } + } + + rc2 = sqlite3_reset(pSelect); + if( rc==SQLITE_OK ) rc = rc2; + } + + return rc; +} + +/* +** Determine the largest segment index value that exists within absolute +** level iAbsLevel+1. If no error occurs, set *piIdx to this value plus +** one before returning SQLITE_OK. Or, if there are no segments at all +** within level iAbsLevel, set *piIdx to zero. +** +** If an error occurs, return an SQLite error code. The final value of +** *piIdx is undefined in this case. +*/ +static int fts3IncrmergeOutputIdx( + Fts3Table *p, /* FTS Table handle */ + sqlite3_int64 iAbsLevel, /* Absolute index of input segments */ + int *piIdx /* OUT: Next free index at iAbsLevel+1 */ +){ + int rc; + sqlite3_stmt *pOutputIdx = 0; /* SQL used to find output index */ + + rc = fts3SqlStmt(p, SQL_NEXT_SEGMENT_INDEX, &pOutputIdx, 0); + if( rc==SQLITE_OK ){ + sqlite3_bind_int64(pOutputIdx, 1, iAbsLevel+1); + sqlite3_step(pOutputIdx); + *piIdx = sqlite3_column_int(pOutputIdx, 0); + rc = sqlite3_reset(pOutputIdx); + } + + return rc; +} + +/* +** Allocate an appendable output segment on absolute level iAbsLevel+1 +** with idx value iIdx. +** +** In the %_segdir table, a segment is defined by the values in three +** columns: +** +** start_block +** leaves_end_block +** end_block +** +** When an appendable segment is allocated, it is estimated that the +** maximum number of leaf blocks that may be required is the sum of the +** number of leaf blocks consumed by the input segments, plus the number +** of input segments, multiplied by two. This value is stored in stack +** variable nLeafEst. +** +** A total of 16*nLeafEst blocks are allocated when an appendable segment +** is created ((1 + end_block - start_block)==16*nLeafEst). The contiguous +** array of leaf nodes starts at the first block allocated. The array +** of interior nodes that are parents of the leaf nodes start at block +** (start_block + (1 + end_block - start_block) / 16). And so on. +** +** In the actual code below, the value "16" is replaced with the +** pre-processor macro FTS_MAX_APPENDABLE_HEIGHT. +*/ +static int fts3IncrmergeWriter( + Fts3Table *p, /* Fts3 table handle */ + sqlite3_int64 iAbsLevel, /* Absolute level of input segments */ + int iIdx, /* Index of new output segment */ + Fts3MultiSegReader *pCsr, /* Cursor that data will be read from */ + IncrmergeWriter *pWriter /* Populate this object */ +){ + int rc; /* Return Code */ + int i; /* Iterator variable */ + int nLeafEst = 0; /* Blocks allocated for leaf nodes */ + sqlite3_stmt *pLeafEst = 0; /* SQL used to determine nLeafEst */ + sqlite3_stmt *pFirstBlock = 0; /* SQL used to determine first block */ + + /* Calculate nLeafEst. */ + rc = fts3SqlStmt(p, SQL_MAX_LEAF_NODE_ESTIMATE, &pLeafEst, 0); + if( rc==SQLITE_OK ){ + sqlite3_bind_int64(pLeafEst, 1, iAbsLevel); + sqlite3_bind_int64(pLeafEst, 2, pCsr->nSegment); + if( SQLITE_ROW==sqlite3_step(pLeafEst) ){ + nLeafEst = sqlite3_column_int(pLeafEst, 0); + } + rc = sqlite3_reset(pLeafEst); + } + if( rc!=SQLITE_OK ) return rc; + + /* Calculate the first block to use in the output segment */ + rc = fts3SqlStmt(p, SQL_NEXT_SEGMENTS_ID, &pFirstBlock, 0); + if( rc==SQLITE_OK ){ + if( SQLITE_ROW==sqlite3_step(pFirstBlock) ){ + pWriter->iStart = sqlite3_column_int64(pFirstBlock, 0); + pWriter->iEnd = pWriter->iStart - 1; + pWriter->iEnd += nLeafEst * FTS_MAX_APPENDABLE_HEIGHT; + } + rc = sqlite3_reset(pFirstBlock); + } + if( rc!=SQLITE_OK ) return rc; + + /* Insert the marker in the %_segments table to make sure nobody tries + ** to steal the space just allocated. This is also used to identify + ** appendable segments. */ + rc = fts3WriteSegment(p, pWriter->iEnd, 0, 0); + if( rc!=SQLITE_OK ) return rc; + + pWriter->iAbsLevel = iAbsLevel; + pWriter->nLeafEst = nLeafEst; + pWriter->iIdx = iIdx; + + /* Set up the array of NodeWriter objects */ + for(i=0; iaNodeWriter[i].iBlock = pWriter->iStart + i*pWriter->nLeafEst; + } + return SQLITE_OK; +} + +/* +** Remove an entry from the %_segdir table. This involves running the +** following two statements: +** +** DELETE FROM %_segdir WHERE level = :iAbsLevel AND idx = :iIdx +** UPDATE %_segdir SET idx = idx - 1 WHERE level = :iAbsLevel AND idx > :iIdx +** +** The DELETE statement removes the specific %_segdir level. The UPDATE +** statement ensures that the remaining segments have contiguously allocated +** idx values. +*/ +static int fts3RemoveSegdirEntry( + Fts3Table *p, /* FTS3 table handle */ + sqlite3_int64 iAbsLevel, /* Absolute level to delete from */ + int iIdx /* Index of %_segdir entry to delete */ +){ + int rc; /* Return code */ + sqlite3_stmt *pDelete = 0; /* DELETE statement */ + + rc = fts3SqlStmt(p, SQL_DELETE_SEGDIR_ENTRY, &pDelete, 0); + if( rc==SQLITE_OK ){ + sqlite3_bind_int64(pDelete, 1, iAbsLevel); + sqlite3_bind_int(pDelete, 2, iIdx); + sqlite3_step(pDelete); + rc = sqlite3_reset(pDelete); + } + + return rc; +} + +/* +** One or more segments have just been removed from absolute level iAbsLevel. +** Update the 'idx' values of the remaining segments in the level so that +** the idx values are a contiguous sequence starting from 0. +*/ +static int fts3RepackSegdirLevel( + Fts3Table *p, /* FTS3 table handle */ + sqlite3_int64 iAbsLevel /* Absolute level to repack */ +){ + int rc; /* Return code */ + int *aIdx = 0; /* Array of remaining idx values */ + int nIdx = 0; /* Valid entries in aIdx[] */ + int nAlloc = 0; /* Allocated size of aIdx[] */ + int i; /* Iterator variable */ + sqlite3_stmt *pSelect = 0; /* Select statement to read idx values */ + sqlite3_stmt *pUpdate = 0; /* Update statement to modify idx values */ + + rc = fts3SqlStmt(p, SQL_SELECT_INDEXES, &pSelect, 0); + if( rc==SQLITE_OK ){ + int rc2; + sqlite3_bind_int64(pSelect, 1, iAbsLevel); + while( SQLITE_ROW==sqlite3_step(pSelect) ){ + if( nIdx>=nAlloc ){ + int *aNew; + nAlloc += 16; + aNew = sqlite3_realloc(aIdx, nAlloc*sizeof(int)); + if( !aNew ){ + rc = SQLITE_NOMEM; + break; + } + aIdx = aNew; + } + aIdx[nIdx++] = sqlite3_column_int(pSelect, 0); + } + rc2 = sqlite3_reset(pSelect); + if( rc==SQLITE_OK ) rc = rc2; + } + + if( rc==SQLITE_OK ){ + rc = fts3SqlStmt(p, SQL_SHIFT_SEGDIR_ENTRY, &pUpdate, 0); + } + if( rc==SQLITE_OK ){ + sqlite3_bind_int64(pUpdate, 2, iAbsLevel); + } + + assert( p->bIgnoreSavepoint==0 ); + p->bIgnoreSavepoint = 1; + for(i=0; rc==SQLITE_OK && ibIgnoreSavepoint = 0; + + sqlite3_free(aIdx); + return rc; +} + +static void fts3StartNode(Blob *pNode, int iHeight, sqlite3_int64 iChild){ + pNode->a[0] = (char)iHeight; + if( iChild ){ + assert( pNode->nAlloc>=1+sqlite3Fts3VarintLen(iChild) ); + pNode->n = 1 + sqlite3Fts3PutVarint(&pNode->a[1], iChild); + }else{ + assert( pNode->nAlloc>=1 ); + pNode->n = 1; + } +} + +/* +** The first two arguments are a pointer to and the size of a segment b-tree +** node. The node may be a leaf or an internal node. +** +** This function creates a new node image in blob object *pNew by copying +** all terms that are greater than or equal to zTerm/nTerm (for leaf nodes) +** or greater than zTerm/nTerm (for internal nodes) from aNode/nNode. +*/ +static int fts3TruncateNode( + const char *aNode, /* Current node image */ + int nNode, /* Size of aNode in bytes */ + Blob *pNew, /* OUT: Write new node image here */ + const char *zTerm, /* Omit all terms smaller than this */ + int nTerm, /* Size of zTerm in bytes */ + sqlite3_int64 *piBlock /* OUT: Block number in next layer down */ +){ + NodeReader reader; /* Reader object */ + Blob prev = {0, 0, 0}; /* Previous term written to new node */ + int rc = SQLITE_OK; /* Return code */ + int bLeaf; /* True for a leaf node */ + + if( nNode<1 ) return FTS_CORRUPT_VTAB; + bLeaf = aNode[0]=='\0'; + + /* Allocate required output space */ + blobGrowBuffer(pNew, nNode, &rc); + if( rc!=SQLITE_OK ) return rc; + pNew->n = 0; + + /* Populate new node buffer */ + for(rc = nodeReaderInit(&reader, aNode, nNode); + rc==SQLITE_OK && reader.aNode; + rc = nodeReaderNext(&reader) + ){ + if( pNew->n==0 ){ + int res = fts3TermCmp(reader.term.a, reader.term.n, zTerm, nTerm); + if( res<0 || (bLeaf==0 && res==0) ) continue; + fts3StartNode(pNew, (int)aNode[0], reader.iChild); + *piBlock = reader.iChild; + } + rc = fts3AppendToNode( + pNew, &prev, reader.term.a, reader.term.n, + reader.aDoclist, reader.nDoclist + ); + if( rc!=SQLITE_OK ) break; + } + if( pNew->n==0 ){ + fts3StartNode(pNew, (int)aNode[0], reader.iChild); + *piBlock = reader.iChild; + } + assert( pNew->n<=pNew->nAlloc ); + + nodeReaderRelease(&reader); + sqlite3_free(prev.a); + return rc; +} + +/* +** Remove all terms smaller than zTerm/nTerm from segment iIdx in absolute +** level iAbsLevel. This may involve deleting entries from the %_segments +** table, and modifying existing entries in both the %_segments and %_segdir +** tables. +** +** SQLITE_OK is returned if the segment is updated successfully. Or an +** SQLite error code otherwise. +*/ +static int fts3TruncateSegment( + Fts3Table *p, /* FTS3 table handle */ + sqlite3_int64 iAbsLevel, /* Absolute level of segment to modify */ + int iIdx, /* Index within level of segment to modify */ + const char *zTerm, /* Remove terms smaller than this */ + int nTerm /* Number of bytes in buffer zTerm */ +){ + int rc = SQLITE_OK; /* Return code */ + Blob root = {0,0,0}; /* New root page image */ + Blob block = {0,0,0}; /* Buffer used for any other block */ + sqlite3_int64 iBlock = 0; /* Block id */ + sqlite3_int64 iNewStart = 0; /* New value for iStartBlock */ + sqlite3_int64 iOldStart = 0; /* Old value for iStartBlock */ + sqlite3_stmt *pFetch = 0; /* Statement used to fetch segdir */ + + rc = fts3SqlStmt(p, SQL_SELECT_SEGDIR, &pFetch, 0); + if( rc==SQLITE_OK ){ + int rc2; /* sqlite3_reset() return code */ + sqlite3_bind_int64(pFetch, 1, iAbsLevel); + sqlite3_bind_int(pFetch, 2, iIdx); + if( SQLITE_ROW==sqlite3_step(pFetch) ){ + const char *aRoot = sqlite3_column_blob(pFetch, 4); + int nRoot = sqlite3_column_bytes(pFetch, 4); + iOldStart = sqlite3_column_int64(pFetch, 1); + rc = fts3TruncateNode(aRoot, nRoot, &root, zTerm, nTerm, &iBlock); + } + rc2 = sqlite3_reset(pFetch); + if( rc==SQLITE_OK ) rc = rc2; + } + + while( rc==SQLITE_OK && iBlock ){ + char *aBlock = 0; + int nBlock = 0; + iNewStart = iBlock; + + rc = sqlite3Fts3ReadBlock(p, iBlock, &aBlock, &nBlock, 0); + if( rc==SQLITE_OK ){ + rc = fts3TruncateNode(aBlock, nBlock, &block, zTerm, nTerm, &iBlock); + } + if( rc==SQLITE_OK ){ + rc = fts3WriteSegment(p, iNewStart, block.a, block.n); + } + sqlite3_free(aBlock); + } + + /* Variable iNewStart now contains the first valid leaf node. */ + if( rc==SQLITE_OK && iNewStart ){ + sqlite3_stmt *pDel = 0; + rc = fts3SqlStmt(p, SQL_DELETE_SEGMENTS_RANGE, &pDel, 0); + if( rc==SQLITE_OK ){ + sqlite3_bind_int64(pDel, 1, iOldStart); + sqlite3_bind_int64(pDel, 2, iNewStart-1); + sqlite3_step(pDel); + rc = sqlite3_reset(pDel); + } + } + + if( rc==SQLITE_OK ){ + sqlite3_stmt *pChomp = 0; + rc = fts3SqlStmt(p, SQL_CHOMP_SEGDIR, &pChomp, 0); + if( rc==SQLITE_OK ){ + sqlite3_bind_int64(pChomp, 1, iNewStart); + sqlite3_bind_blob(pChomp, 2, root.a, root.n, SQLITE_STATIC); + sqlite3_bind_int64(pChomp, 3, iAbsLevel); + sqlite3_bind_int(pChomp, 4, iIdx); + sqlite3_step(pChomp); + rc = sqlite3_reset(pChomp); + sqlite3_bind_null(pChomp, 2); + } + } + + sqlite3_free(root.a); + sqlite3_free(block.a); + return rc; +} + +/* +** This function is called after an incrmental-merge operation has run to +** merge (or partially merge) two or more segments from absolute level +** iAbsLevel. +** +** Each input segment is either removed from the db completely (if all of +** its data was copied to the output segment by the incrmerge operation) +** or modified in place so that it no longer contains those entries that +** have been duplicated in the output segment. +*/ +static int fts3IncrmergeChomp( + Fts3Table *p, /* FTS table handle */ + sqlite3_int64 iAbsLevel, /* Absolute level containing segments */ + Fts3MultiSegReader *pCsr, /* Chomp all segments opened by this cursor */ + int *pnRem /* Number of segments not deleted */ +){ + int i; + int nRem = 0; + int rc = SQLITE_OK; + + for(i=pCsr->nSegment-1; i>=0 && rc==SQLITE_OK; i--){ + Fts3SegReader *pSeg = 0; + int j; + + /* Find the Fts3SegReader object with Fts3SegReader.iIdx==i. It is hiding + ** somewhere in the pCsr->apSegment[] array. */ + for(j=0; ALWAYS(jnSegment); j++){ + pSeg = pCsr->apSegment[j]; + if( pSeg->iIdx==i ) break; + } + assert( jnSegment && pSeg->iIdx==i ); + + if( pSeg->aNode==0 ){ + /* Seg-reader is at EOF. Remove the entire input segment. */ + rc = fts3DeleteSegment(p, pSeg); + if( rc==SQLITE_OK ){ + rc = fts3RemoveSegdirEntry(p, iAbsLevel, pSeg->iIdx); + } + *pnRem = 0; + }else{ + /* The incremental merge did not copy all the data from this + ** segment to the upper level. The segment is modified in place + ** so that it contains no keys smaller than zTerm/nTerm. */ + const char *zTerm = pSeg->zTerm; + int nTerm = pSeg->nTerm; + rc = fts3TruncateSegment(p, iAbsLevel, pSeg->iIdx, zTerm, nTerm); + nRem++; + } + } + + if( rc==SQLITE_OK && nRem!=pCsr->nSegment ){ + rc = fts3RepackSegdirLevel(p, iAbsLevel); + } + + *pnRem = nRem; + return rc; +} + +/* +** Store an incr-merge hint in the database. +*/ +static int fts3IncrmergeHintStore(Fts3Table *p, Blob *pHint){ + sqlite3_stmt *pReplace = 0; + int rc; /* Return code */ + + rc = fts3SqlStmt(p, SQL_REPLACE_STAT, &pReplace, 0); + if( rc==SQLITE_OK ){ + sqlite3_bind_int(pReplace, 1, FTS_STAT_INCRMERGEHINT); + sqlite3_bind_blob(pReplace, 2, pHint->a, pHint->n, SQLITE_STATIC); + sqlite3_step(pReplace); + rc = sqlite3_reset(pReplace); + sqlite3_bind_null(pReplace, 2); + } + + return rc; +} + +/* +** Load an incr-merge hint from the database. The incr-merge hint, if one +** exists, is stored in the rowid==1 row of the %_stat table. +** +** If successful, populate blob *pHint with the value read from the %_stat +** table and return SQLITE_OK. Otherwise, if an error occurs, return an +** SQLite error code. +*/ +static int fts3IncrmergeHintLoad(Fts3Table *p, Blob *pHint){ + sqlite3_stmt *pSelect = 0; + int rc; + + pHint->n = 0; + rc = fts3SqlStmt(p, SQL_SELECT_STAT, &pSelect, 0); + if( rc==SQLITE_OK ){ + int rc2; + sqlite3_bind_int(pSelect, 1, FTS_STAT_INCRMERGEHINT); + if( SQLITE_ROW==sqlite3_step(pSelect) ){ + const char *aHint = sqlite3_column_blob(pSelect, 0); + int nHint = sqlite3_column_bytes(pSelect, 0); + if( aHint ){ + blobGrowBuffer(pHint, nHint, &rc); + if( rc==SQLITE_OK ){ + memcpy(pHint->a, aHint, nHint); + pHint->n = nHint; + } + } + } + rc2 = sqlite3_reset(pSelect); + if( rc==SQLITE_OK ) rc = rc2; + } + + return rc; +} + +/* +** If *pRc is not SQLITE_OK when this function is called, it is a no-op. +** Otherwise, append an entry to the hint stored in blob *pHint. Each entry +** consists of two varints, the absolute level number of the input segments +** and the number of input segments. +** +** If successful, leave *pRc set to SQLITE_OK and return. If an error occurs, +** set *pRc to an SQLite error code before returning. +*/ +static void fts3IncrmergeHintPush( + Blob *pHint, /* Hint blob to append to */ + i64 iAbsLevel, /* First varint to store in hint */ + int nInput, /* Second varint to store in hint */ + int *pRc /* IN/OUT: Error code */ +){ + blobGrowBuffer(pHint, pHint->n + 2*FTS3_VARINT_MAX, pRc); + if( *pRc==SQLITE_OK ){ + pHint->n += sqlite3Fts3PutVarint(&pHint->a[pHint->n], iAbsLevel); + pHint->n += sqlite3Fts3PutVarint(&pHint->a[pHint->n], (i64)nInput); + } +} + +/* +** Read the last entry (most recently pushed) from the hint blob *pHint +** and then remove the entry. Write the two values read to *piAbsLevel and +** *pnInput before returning. +** +** If no error occurs, return SQLITE_OK. If the hint blob in *pHint does +** not contain at least two valid varints, return SQLITE_CORRUPT_VTAB. +*/ +static int fts3IncrmergeHintPop(Blob *pHint, i64 *piAbsLevel, int *pnInput){ + const int nHint = pHint->n; + int i; + + i = pHint->n-2; + while( i>0 && (pHint->a[i-1] & 0x80) ) i--; + while( i>0 && (pHint->a[i-1] & 0x80) ) i--; + + pHint->n = i; + i += sqlite3Fts3GetVarint(&pHint->a[i], piAbsLevel); + i += fts3GetVarint32(&pHint->a[i], pnInput); + if( i!=nHint ) return FTS_CORRUPT_VTAB; + + return SQLITE_OK; +} + + +/* +** Attempt an incremental merge that writes nMerge leaf blocks. +** +** Incremental merges happen nMin segments at a time. The segments +** to be merged are the nMin oldest segments (the ones with the smallest +** values for the _segdir.idx field) in the highest level that contains +** at least nMin segments. Multiple merges might occur in an attempt to +** write the quota of nMerge leaf blocks. +*/ +SQLITE_PRIVATE int sqlite3Fts3Incrmerge(Fts3Table *p, int nMerge, int nMin){ + int rc; /* Return code */ + int nRem = nMerge; /* Number of leaf pages yet to be written */ + Fts3MultiSegReader *pCsr; /* Cursor used to read input data */ + Fts3SegFilter *pFilter; /* Filter used with cursor pCsr */ + IncrmergeWriter *pWriter; /* Writer object */ + int nSeg = 0; /* Number of input segments */ + sqlite3_int64 iAbsLevel = 0; /* Absolute level number to work on */ + Blob hint = {0, 0, 0}; /* Hint read from %_stat table */ + int bDirtyHint = 0; /* True if blob 'hint' has been modified */ + + /* Allocate space for the cursor, filter and writer objects */ + const int nAlloc = sizeof(*pCsr) + sizeof(*pFilter) + sizeof(*pWriter); + pWriter = (IncrmergeWriter *)sqlite3_malloc(nAlloc); + if( !pWriter ) return SQLITE_NOMEM; + pFilter = (Fts3SegFilter *)&pWriter[1]; + pCsr = (Fts3MultiSegReader *)&pFilter[1]; + + rc = fts3IncrmergeHintLoad(p, &hint); + while( rc==SQLITE_OK && nRem>0 ){ + const i64 nMod = FTS3_SEGDIR_MAXLEVEL * p->nIndex; + sqlite3_stmt *pFindLevel = 0; /* SQL used to determine iAbsLevel */ + int bUseHint = 0; /* True if attempting to append */ + int iIdx = 0; /* Largest idx in level (iAbsLevel+1) */ + + /* Search the %_segdir table for the absolute level with the smallest + ** relative level number that contains at least nMin segments, if any. + ** If one is found, set iAbsLevel to the absolute level number and + ** nSeg to nMin. If no level with at least nMin segments can be found, + ** set nSeg to -1. + */ + rc = fts3SqlStmt(p, SQL_FIND_MERGE_LEVEL, &pFindLevel, 0); + sqlite3_bind_int(pFindLevel, 1, MAX(2, nMin)); + if( sqlite3_step(pFindLevel)==SQLITE_ROW ){ + iAbsLevel = sqlite3_column_int64(pFindLevel, 0); + nSeg = sqlite3_column_int(pFindLevel, 1); + assert( nSeg>=2 ); + }else{ + nSeg = -1; + } + rc = sqlite3_reset(pFindLevel); + + /* If the hint read from the %_stat table is not empty, check if the + ** last entry in it specifies a relative level smaller than or equal + ** to the level identified by the block above (if any). If so, this + ** iteration of the loop will work on merging at the hinted level. + */ + if( rc==SQLITE_OK && hint.n ){ + int nHint = hint.n; + sqlite3_int64 iHintAbsLevel = 0; /* Hint level */ + int nHintSeg = 0; /* Hint number of segments */ + + rc = fts3IncrmergeHintPop(&hint, &iHintAbsLevel, &nHintSeg); + if( nSeg<0 || (iAbsLevel % nMod) >= (iHintAbsLevel % nMod) ){ + iAbsLevel = iHintAbsLevel; + nSeg = nHintSeg; + bUseHint = 1; + bDirtyHint = 1; + }else{ + /* This undoes the effect of the HintPop() above - so that no entry + ** is removed from the hint blob. */ + hint.n = nHint; + } + } + + /* If nSeg is less that zero, then there is no level with at least + ** nMin segments and no hint in the %_stat table. No work to do. + ** Exit early in this case. */ + if( nSeg<0 ) break; + + /* Open a cursor to iterate through the contents of the oldest nSeg + ** indexes of absolute level iAbsLevel. If this cursor is opened using + ** the 'hint' parameters, it is possible that there are less than nSeg + ** segments available in level iAbsLevel. In this case, no work is + ** done on iAbsLevel - fall through to the next iteration of the loop + ** to start work on some other level. */ + memset(pWriter, 0, nAlloc); + pFilter->flags = FTS3_SEGMENT_REQUIRE_POS; + + if( rc==SQLITE_OK ){ + rc = fts3IncrmergeOutputIdx(p, iAbsLevel, &iIdx); + assert( bUseHint==1 || bUseHint==0 ); + if( iIdx==0 || (bUseHint && iIdx==1) ){ + int bIgnore = 0; + rc = fts3SegmentIsMaxLevel(p, iAbsLevel+1, &bIgnore); + if( bIgnore ){ + pFilter->flags |= FTS3_SEGMENT_IGNORE_EMPTY; + } + } + } + + if( rc==SQLITE_OK ){ + rc = fts3IncrmergeCsr(p, iAbsLevel, nSeg, pCsr); + } + if( SQLITE_OK==rc && pCsr->nSegment==nSeg + && SQLITE_OK==(rc = sqlite3Fts3SegReaderStart(p, pCsr, pFilter)) + && SQLITE_ROW==(rc = sqlite3Fts3SegReaderStep(p, pCsr)) + ){ + if( bUseHint && iIdx>0 ){ + const char *zKey = pCsr->zTerm; + int nKey = pCsr->nTerm; + rc = fts3IncrmergeLoad(p, iAbsLevel, iIdx-1, zKey, nKey, pWriter); + }else{ + rc = fts3IncrmergeWriter(p, iAbsLevel, iIdx, pCsr, pWriter); + } + + if( rc==SQLITE_OK && pWriter->nLeafEst ){ + fts3LogMerge(nSeg, iAbsLevel); + do { + rc = fts3IncrmergeAppend(p, pWriter, pCsr); + if( rc==SQLITE_OK ) rc = sqlite3Fts3SegReaderStep(p, pCsr); + if( pWriter->nWork>=nRem && rc==SQLITE_ROW ) rc = SQLITE_OK; + }while( rc==SQLITE_ROW ); + + /* Update or delete the input segments */ + if( rc==SQLITE_OK ){ + nRem -= (1 + pWriter->nWork); + rc = fts3IncrmergeChomp(p, iAbsLevel, pCsr, &nSeg); + if( nSeg!=0 ){ + bDirtyHint = 1; + fts3IncrmergeHintPush(&hint, iAbsLevel, nSeg, &rc); + } + } + } + + if( nSeg!=0 ){ + pWriter->nLeafData = pWriter->nLeafData * -1; + } + fts3IncrmergeRelease(p, pWriter, &rc); + if( nSeg==0 && pWriter->bNoLeafData==0 ){ + fts3PromoteSegments(p, iAbsLevel+1, pWriter->nLeafData); + } + } + + sqlite3Fts3SegReaderFinish(pCsr); + } + + /* Write the hint values into the %_stat table for the next incr-merger */ + if( bDirtyHint && rc==SQLITE_OK ){ + rc = fts3IncrmergeHintStore(p, &hint); + } + + sqlite3_free(pWriter); + sqlite3_free(hint.a); + return rc; +} + +/* +** Convert the text beginning at *pz into an integer and return +** its value. Advance *pz to point to the first character past +** the integer. +** +** This function used for parameters to merge= and incrmerge= +** commands. +*/ +static int fts3Getint(const char **pz){ + const char *z = *pz; + int i = 0; + while( (*z)>='0' && (*z)<='9' && i<214748363 ) i = 10*i + *(z++) - '0'; + *pz = z; + return i; +} + +/* +** Process statements of the form: +** +** INSERT INTO table(table) VALUES('merge=A,B'); +** +** A and B are integers that decode to be the number of leaf pages +** written for the merge, and the minimum number of segments on a level +** before it will be selected for a merge, respectively. +*/ +static int fts3DoIncrmerge( + Fts3Table *p, /* FTS3 table handle */ + const char *zParam /* Nul-terminated string containing "A,B" */ +){ + int rc; + int nMin = (FTS3_MERGE_COUNT / 2); + int nMerge = 0; + const char *z = zParam; + + /* Read the first integer value */ + nMerge = fts3Getint(&z); + + /* If the first integer value is followed by a ',', read the second + ** integer value. */ + if( z[0]==',' && z[1]!='\0' ){ + z++; + nMin = fts3Getint(&z); + } + + if( z[0]!='\0' || nMin<2 ){ + rc = SQLITE_ERROR; + }else{ + rc = SQLITE_OK; + if( !p->bHasStat ){ + assert( p->bFts4==0 ); + sqlite3Fts3CreateStatTable(&rc, p); + } + if( rc==SQLITE_OK ){ + rc = sqlite3Fts3Incrmerge(p, nMerge, nMin); + } + sqlite3Fts3SegmentsClose(p); + } + return rc; +} + +/* +** Process statements of the form: +** +** INSERT INTO table(table) VALUES('automerge=X'); +** +** where X is an integer. X==0 means to turn automerge off. X!=0 means +** turn it on. The setting is persistent. +*/ +static int fts3DoAutoincrmerge( + Fts3Table *p, /* FTS3 table handle */ + const char *zParam /* Nul-terminated string containing boolean */ +){ + int rc = SQLITE_OK; + sqlite3_stmt *pStmt = 0; + p->nAutoincrmerge = fts3Getint(&zParam); + if( p->nAutoincrmerge==1 || p->nAutoincrmerge>FTS3_MERGE_COUNT ){ + p->nAutoincrmerge = 8; + } + if( !p->bHasStat ){ + assert( p->bFts4==0 ); + sqlite3Fts3CreateStatTable(&rc, p); + if( rc ) return rc; + } + rc = fts3SqlStmt(p, SQL_REPLACE_STAT, &pStmt, 0); + if( rc ) return rc; + sqlite3_bind_int(pStmt, 1, FTS_STAT_AUTOINCRMERGE); + sqlite3_bind_int(pStmt, 2, p->nAutoincrmerge); + sqlite3_step(pStmt); + rc = sqlite3_reset(pStmt); + return rc; +} + +/* +** Return a 64-bit checksum for the FTS index entry specified by the +** arguments to this function. +*/ +static u64 fts3ChecksumEntry( + const char *zTerm, /* Pointer to buffer containing term */ + int nTerm, /* Size of zTerm in bytes */ + int iLangid, /* Language id for current row */ + int iIndex, /* Index (0..Fts3Table.nIndex-1) */ + i64 iDocid, /* Docid for current row. */ + int iCol, /* Column number */ + int iPos /* Position */ +){ + int i; + u64 ret = (u64)iDocid; + + ret += (ret<<3) + iLangid; + ret += (ret<<3) + iIndex; + ret += (ret<<3) + iCol; + ret += (ret<<3) + iPos; + for(i=0; inIndex-1) */ + int *pRc /* OUT: Return code */ +){ + Fts3SegFilter filter; + Fts3MultiSegReader csr; + int rc; + u64 cksum = 0; + + assert( *pRc==SQLITE_OK ); + + memset(&filter, 0, sizeof(filter)); + memset(&csr, 0, sizeof(csr)); + filter.flags = FTS3_SEGMENT_REQUIRE_POS|FTS3_SEGMENT_IGNORE_EMPTY; + filter.flags |= FTS3_SEGMENT_SCAN; + + rc = sqlite3Fts3SegReaderCursor( + p, iLangid, iIndex, FTS3_SEGCURSOR_ALL, 0, 0, 0, 1,&csr + ); + if( rc==SQLITE_OK ){ + rc = sqlite3Fts3SegReaderStart(p, &csr, &filter); + } + + if( rc==SQLITE_OK ){ + while( SQLITE_ROW==(rc = sqlite3Fts3SegReaderStep(p, &csr)) ){ + char *pCsr = csr.aDoclist; + char *pEnd = &pCsr[csr.nDoclist]; + + i64 iDocid = 0; + i64 iCol = 0; + i64 iPos = 0; + + pCsr += sqlite3Fts3GetVarint(pCsr, &iDocid); + while( pCsriPrevLangid); + sqlite3_bind_int(pAllLangid, 2, p->nIndex); + while( rc==SQLITE_OK && sqlite3_step(pAllLangid)==SQLITE_ROW ){ + int iLangid = sqlite3_column_int(pAllLangid, 0); + int i; + for(i=0; inIndex; i++){ + cksum1 = cksum1 ^ fts3ChecksumIndex(p, iLangid, i, &rc); + } + } + rc2 = sqlite3_reset(pAllLangid); + if( rc==SQLITE_OK ) rc = rc2; + } + + /* This block calculates the checksum according to the %_content table */ + if( rc==SQLITE_OK ){ + sqlite3_tokenizer_module const *pModule = p->pTokenizer->pModule; + sqlite3_stmt *pStmt = 0; + char *zSql; + + zSql = sqlite3_mprintf("SELECT %s" , p->zReadExprlist); + if( !zSql ){ + rc = SQLITE_NOMEM; + }else{ + rc = sqlite3_prepare_v2(p->db, zSql, -1, &pStmt, 0); + sqlite3_free(zSql); + } + + while( rc==SQLITE_OK && SQLITE_ROW==sqlite3_step(pStmt) ){ + i64 iDocid = sqlite3_column_int64(pStmt, 0); + int iLang = langidFromSelect(p, pStmt); + int iCol; + + for(iCol=0; rc==SQLITE_OK && iColnColumn; iCol++){ + if( p->abNotindexed[iCol]==0 ){ + const char *zText = (const char *)sqlite3_column_text(pStmt, iCol+1); + int nText = sqlite3_column_bytes(pStmt, iCol+1); + sqlite3_tokenizer_cursor *pT = 0; + + rc = sqlite3Fts3OpenTokenizer(p->pTokenizer, iLang, zText, nText,&pT); + while( rc==SQLITE_OK ){ + char const *zToken; /* Buffer containing token */ + int nToken = 0; /* Number of bytes in token */ + int iDum1 = 0, iDum2 = 0; /* Dummy variables */ + int iPos = 0; /* Position of token in zText */ + + rc = pModule->xNext(pT, &zToken, &nToken, &iDum1, &iDum2, &iPos); + if( rc==SQLITE_OK ){ + int i; + cksum2 = cksum2 ^ fts3ChecksumEntry( + zToken, nToken, iLang, 0, iDocid, iCol, iPos + ); + for(i=1; inIndex; i++){ + if( p->aIndex[i].nPrefix<=nToken ){ + cksum2 = cksum2 ^ fts3ChecksumEntry( + zToken, p->aIndex[i].nPrefix, iLang, i, iDocid, iCol, iPos + ); + } + } + } + } + if( pT ) pModule->xClose(pT); + if( rc==SQLITE_DONE ) rc = SQLITE_OK; + } + } + } + + sqlite3_finalize(pStmt); + } + + *pbOk = (cksum1==cksum2); + return rc; +} + +/* +** Run the integrity-check. If no error occurs and the current contents of +** the FTS index are correct, return SQLITE_OK. Or, if the contents of the +** FTS index are incorrect, return SQLITE_CORRUPT_VTAB. +** +** Or, if an error (e.g. an OOM or IO error) occurs, return an SQLite +** error code. +** +** The integrity-check works as follows. For each token and indexed token +** prefix in the document set, a 64-bit checksum is calculated (by code +** in fts3ChecksumEntry()) based on the following: +** +** + The index number (0 for the main index, 1 for the first prefix +** index etc.), +** + The token (or token prefix) text itself, +** + The language-id of the row it appears in, +** + The docid of the row it appears in, +** + The column it appears in, and +** + The tokens position within that column. +** +** The checksums for all entries in the index are XORed together to create +** a single checksum for the entire index. +** +** The integrity-check code calculates the same checksum in two ways: +** +** 1. By scanning the contents of the FTS index, and +** 2. By scanning and tokenizing the content table. +** +** If the two checksums are identical, the integrity-check is deemed to have +** passed. +*/ +static int fts3DoIntegrityCheck( + Fts3Table *p /* FTS3 table handle */ +){ + int rc; + int bOk = 0; + rc = fts3IntegrityCheck(p, &bOk); + if( rc==SQLITE_OK && bOk==0 ) rc = FTS_CORRUPT_VTAB; + return rc; +} + +/* +** Handle a 'special' INSERT of the form: +** +** "INSERT INTO tbl(tbl) VALUES()" +** +** Argument pVal contains the result of . Currently the only +** meaningful value to insert is the text 'optimize'. +*/ +static int fts3SpecialInsert(Fts3Table *p, sqlite3_value *pVal){ + int rc; /* Return Code */ + const char *zVal = (const char *)sqlite3_value_text(pVal); + int nVal = sqlite3_value_bytes(pVal); + + if( !zVal ){ + return SQLITE_NOMEM; + }else if( nVal==8 && 0==sqlite3_strnicmp(zVal, "optimize", 8) ){ + rc = fts3DoOptimize(p, 0); + }else if( nVal==7 && 0==sqlite3_strnicmp(zVal, "rebuild", 7) ){ + rc = fts3DoRebuild(p); + }else if( nVal==15 && 0==sqlite3_strnicmp(zVal, "integrity-check", 15) ){ + rc = fts3DoIntegrityCheck(p); + }else if( nVal>6 && 0==sqlite3_strnicmp(zVal, "merge=", 6) ){ + rc = fts3DoIncrmerge(p, &zVal[6]); + }else if( nVal>10 && 0==sqlite3_strnicmp(zVal, "automerge=", 10) ){ + rc = fts3DoAutoincrmerge(p, &zVal[10]); +#ifdef SQLITE_TEST + }else if( nVal>9 && 0==sqlite3_strnicmp(zVal, "nodesize=", 9) ){ + p->nNodeSize = atoi(&zVal[9]); + rc = SQLITE_OK; + }else if( nVal>11 && 0==sqlite3_strnicmp(zVal, "maxpending=", 9) ){ + p->nMaxPendingData = atoi(&zVal[11]); + rc = SQLITE_OK; + }else if( nVal>21 && 0==sqlite3_strnicmp(zVal, "test-no-incr-doclist=", 21) ){ + p->bNoIncrDoclist = atoi(&zVal[21]); + rc = SQLITE_OK; +#endif + }else{ + rc = SQLITE_ERROR; + } + + return rc; +} + +#ifndef SQLITE_DISABLE_FTS4_DEFERRED +/* +** Delete all cached deferred doclists. Deferred doclists are cached +** (allocated) by the sqlite3Fts3CacheDeferredDoclists() function. +*/ +SQLITE_PRIVATE void sqlite3Fts3FreeDeferredDoclists(Fts3Cursor *pCsr){ + Fts3DeferredToken *pDef; + for(pDef=pCsr->pDeferred; pDef; pDef=pDef->pNext){ + fts3PendingListDelete(pDef->pList); + pDef->pList = 0; + } +} + +/* +** Free all entries in the pCsr->pDeffered list. Entries are added to +** this list using sqlite3Fts3DeferToken(). +*/ +SQLITE_PRIVATE void sqlite3Fts3FreeDeferredTokens(Fts3Cursor *pCsr){ + Fts3DeferredToken *pDef; + Fts3DeferredToken *pNext; + for(pDef=pCsr->pDeferred; pDef; pDef=pNext){ + pNext = pDef->pNext; + fts3PendingListDelete(pDef->pList); + sqlite3_free(pDef); + } + pCsr->pDeferred = 0; +} + +/* +** Generate deferred-doclists for all tokens in the pCsr->pDeferred list +** based on the row that pCsr currently points to. +** +** A deferred-doclist is like any other doclist with position information +** included, except that it only contains entries for a single row of the +** table, not for all rows. +*/ +SQLITE_PRIVATE int sqlite3Fts3CacheDeferredDoclists(Fts3Cursor *pCsr){ + int rc = SQLITE_OK; /* Return code */ + if( pCsr->pDeferred ){ + int i; /* Used to iterate through table columns */ + sqlite3_int64 iDocid; /* Docid of the row pCsr points to */ + Fts3DeferredToken *pDef; /* Used to iterate through deferred tokens */ + + Fts3Table *p = (Fts3Table *)pCsr->base.pVtab; + sqlite3_tokenizer *pT = p->pTokenizer; + sqlite3_tokenizer_module const *pModule = pT->pModule; + + assert( pCsr->isRequireSeek==0 ); + iDocid = sqlite3_column_int64(pCsr->pStmt, 0); + + for(i=0; inColumn && rc==SQLITE_OK; i++){ + if( p->abNotindexed[i]==0 ){ + const char *zText = (const char *)sqlite3_column_text(pCsr->pStmt, i+1); + sqlite3_tokenizer_cursor *pTC = 0; + + rc = sqlite3Fts3OpenTokenizer(pT, pCsr->iLangid, zText, -1, &pTC); + while( rc==SQLITE_OK ){ + char const *zToken; /* Buffer containing token */ + int nToken = 0; /* Number of bytes in token */ + int iDum1 = 0, iDum2 = 0; /* Dummy variables */ + int iPos = 0; /* Position of token in zText */ + + rc = pModule->xNext(pTC, &zToken, &nToken, &iDum1, &iDum2, &iPos); + for(pDef=pCsr->pDeferred; pDef && rc==SQLITE_OK; pDef=pDef->pNext){ + Fts3PhraseToken *pPT = pDef->pToken; + if( (pDef->iCol>=p->nColumn || pDef->iCol==i) + && (pPT->bFirst==0 || iPos==0) + && (pPT->n==nToken || (pPT->isPrefix && pPT->nz, pPT->n)) + ){ + fts3PendingListAppend(&pDef->pList, iDocid, i, iPos, &rc); + } + } + } + if( pTC ) pModule->xClose(pTC); + if( rc==SQLITE_DONE ) rc = SQLITE_OK; + } + } + + for(pDef=pCsr->pDeferred; pDef && rc==SQLITE_OK; pDef=pDef->pNext){ + if( pDef->pList ){ + rc = fts3PendingListAppendVarint(&pDef->pList, 0); + } + } + } + + return rc; +} + +SQLITE_PRIVATE int sqlite3Fts3DeferredTokenList( + Fts3DeferredToken *p, + char **ppData, + int *pnData +){ + char *pRet; + int nSkip; + sqlite3_int64 dummy; + + *ppData = 0; + *pnData = 0; + + if( p->pList==0 ){ + return SQLITE_OK; + } + + pRet = (char *)sqlite3_malloc(p->pList->nData); + if( !pRet ) return SQLITE_NOMEM; + + nSkip = sqlite3Fts3GetVarint(p->pList->aData, &dummy); + *pnData = p->pList->nData - nSkip; + *ppData = pRet; + + memcpy(pRet, &p->pList->aData[nSkip], *pnData); + return SQLITE_OK; +} + +/* +** Add an entry for token pToken to the pCsr->pDeferred list. +*/ +SQLITE_PRIVATE int sqlite3Fts3DeferToken( + Fts3Cursor *pCsr, /* Fts3 table cursor */ + Fts3PhraseToken *pToken, /* Token to defer */ + int iCol /* Column that token must appear in (or -1) */ +){ + Fts3DeferredToken *pDeferred; + pDeferred = sqlite3_malloc(sizeof(*pDeferred)); + if( !pDeferred ){ + return SQLITE_NOMEM; + } + memset(pDeferred, 0, sizeof(*pDeferred)); + pDeferred->pToken = pToken; + pDeferred->pNext = pCsr->pDeferred; + pDeferred->iCol = iCol; + pCsr->pDeferred = pDeferred; + + assert( pToken->pDeferred==0 ); + pToken->pDeferred = pDeferred; + + return SQLITE_OK; +} +#endif + +/* +** SQLite value pRowid contains the rowid of a row that may or may not be +** present in the FTS3 table. If it is, delete it and adjust the contents +** of subsiduary data structures accordingly. +*/ +static int fts3DeleteByRowid( + Fts3Table *p, + sqlite3_value *pRowid, + int *pnChng, /* IN/OUT: Decrement if row is deleted */ + u32 *aSzDel +){ + int rc = SQLITE_OK; /* Return code */ + int bFound = 0; /* True if *pRowid really is in the table */ + + fts3DeleteTerms(&rc, p, pRowid, aSzDel, &bFound); + if( bFound && rc==SQLITE_OK ){ + int isEmpty = 0; /* Deleting *pRowid leaves the table empty */ + rc = fts3IsEmpty(p, pRowid, &isEmpty); + if( rc==SQLITE_OK ){ + if( isEmpty ){ + /* Deleting this row means the whole table is empty. In this case + ** delete the contents of all three tables and throw away any + ** data in the pendingTerms hash table. */ + rc = fts3DeleteAll(p, 1); + *pnChng = 0; + memset(aSzDel, 0, sizeof(u32) * (p->nColumn+1) * 2); + }else{ + *pnChng = *pnChng - 1; + if( p->zContentTbl==0 ){ + fts3SqlExec(&rc, p, SQL_DELETE_CONTENT, &pRowid); + } + if( p->bHasDocsize ){ + fts3SqlExec(&rc, p, SQL_DELETE_DOCSIZE, &pRowid); + } + } + } + } + + return rc; +} + +/* +** This function does the work for the xUpdate method of FTS3 virtual +** tables. The schema of the virtual table being: +** +** CREATE TABLE
    ( +** , +**
    HIDDEN, +** docid HIDDEN, +** HIDDEN +** ); +** +** +*/ +SQLITE_PRIVATE int sqlite3Fts3UpdateMethod( + sqlite3_vtab *pVtab, /* FTS3 vtab object */ + int nArg, /* Size of argument array */ + sqlite3_value **apVal, /* Array of arguments */ + sqlite_int64 *pRowid /* OUT: The affected (or effected) rowid */ +){ + Fts3Table *p = (Fts3Table *)pVtab; + int rc = SQLITE_OK; /* Return Code */ + u32 *aSzIns = 0; /* Sizes of inserted documents */ + u32 *aSzDel = 0; /* Sizes of deleted documents */ + int nChng = 0; /* Net change in number of documents */ + int bInsertDone = 0; + + /* At this point it must be known if the %_stat table exists or not. + ** So bHasStat may not be 2. */ + assert( p->bHasStat==0 || p->bHasStat==1 ); + + assert( p->pSegments==0 ); + assert( + nArg==1 /* DELETE operations */ + || nArg==(2 + p->nColumn + 3) /* INSERT or UPDATE operations */ + ); + + /* Check for a "special" INSERT operation. One of the form: + ** + ** INSERT INTO xyz(xyz) VALUES('command'); + */ + if( nArg>1 + && sqlite3_value_type(apVal[0])==SQLITE_NULL + && sqlite3_value_type(apVal[p->nColumn+2])!=SQLITE_NULL + ){ + rc = fts3SpecialInsert(p, apVal[p->nColumn+2]); + goto update_out; + } + + if( nArg>1 && sqlite3_value_int(apVal[2 + p->nColumn + 2])<0 ){ + rc = SQLITE_CONSTRAINT; + goto update_out; + } + + /* Allocate space to hold the change in document sizes */ + aSzDel = sqlite3_malloc64(sizeof(aSzDel[0])*((sqlite3_int64)p->nColumn+1)*2); + if( aSzDel==0 ){ + rc = SQLITE_NOMEM; + goto update_out; + } + aSzIns = &aSzDel[p->nColumn+1]; + memset(aSzDel, 0, sizeof(aSzDel[0])*(p->nColumn+1)*2); + + rc = fts3Writelock(p); + if( rc!=SQLITE_OK ) goto update_out; + + /* If this is an INSERT operation, or an UPDATE that modifies the rowid + ** value, then this operation requires constraint handling. + ** + ** If the on-conflict mode is REPLACE, this means that the existing row + ** should be deleted from the database before inserting the new row. Or, + ** if the on-conflict mode is other than REPLACE, then this method must + ** detect the conflict and return SQLITE_CONSTRAINT before beginning to + ** modify the database file. + */ + if( nArg>1 && p->zContentTbl==0 ){ + /* Find the value object that holds the new rowid value. */ + sqlite3_value *pNewRowid = apVal[3+p->nColumn]; + if( sqlite3_value_type(pNewRowid)==SQLITE_NULL ){ + pNewRowid = apVal[1]; + } + + if( sqlite3_value_type(pNewRowid)!=SQLITE_NULL && ( + sqlite3_value_type(apVal[0])==SQLITE_NULL + || sqlite3_value_int64(apVal[0])!=sqlite3_value_int64(pNewRowid) + )){ + /* The new rowid is not NULL (in this case the rowid will be + ** automatically assigned and there is no chance of a conflict), and + ** the statement is either an INSERT or an UPDATE that modifies the + ** rowid column. So if the conflict mode is REPLACE, then delete any + ** existing row with rowid=pNewRowid. + ** + ** Or, if the conflict mode is not REPLACE, insert the new record into + ** the %_content table. If we hit the duplicate rowid constraint (or any + ** other error) while doing so, return immediately. + ** + ** This branch may also run if pNewRowid contains a value that cannot + ** be losslessly converted to an integer. In this case, the eventual + ** call to fts3InsertData() (either just below or further on in this + ** function) will return SQLITE_MISMATCH. If fts3DeleteByRowid is + ** invoked, it will delete zero rows (since no row will have + ** docid=$pNewRowid if $pNewRowid is not an integer value). + */ + if( sqlite3_vtab_on_conflict(p->db)==SQLITE_REPLACE ){ + rc = fts3DeleteByRowid(p, pNewRowid, &nChng, aSzDel); + }else{ + rc = fts3InsertData(p, apVal, pRowid); + bInsertDone = 1; + } + } + } + if( rc!=SQLITE_OK ){ + goto update_out; + } + + /* If this is a DELETE or UPDATE operation, remove the old record. */ + if( sqlite3_value_type(apVal[0])!=SQLITE_NULL ){ + assert( sqlite3_value_type(apVal[0])==SQLITE_INTEGER ); + rc = fts3DeleteByRowid(p, apVal[0], &nChng, aSzDel); + } + + /* If this is an INSERT or UPDATE operation, insert the new record. */ + if( nArg>1 && rc==SQLITE_OK ){ + int iLangid = sqlite3_value_int(apVal[2 + p->nColumn + 2]); + if( bInsertDone==0 ){ + rc = fts3InsertData(p, apVal, pRowid); + if( rc==SQLITE_CONSTRAINT && p->zContentTbl==0 ){ + rc = FTS_CORRUPT_VTAB; + } + } + if( rc==SQLITE_OK ){ + rc = fts3PendingTermsDocid(p, 0, iLangid, *pRowid); + } + if( rc==SQLITE_OK ){ + assert( p->iPrevDocid==*pRowid ); + rc = fts3InsertTerms(p, iLangid, apVal, aSzIns); + } + if( p->bHasDocsize ){ + fts3InsertDocsize(&rc, p, aSzIns); + } + nChng++; + } + + if( p->bFts4 ){ + fts3UpdateDocTotals(&rc, p, aSzIns, aSzDel, nChng); + } + + update_out: + sqlite3_free(aSzDel); + sqlite3Fts3SegmentsClose(p); + return rc; +} + +/* +** Flush any data in the pending-terms hash table to disk. If successful, +** merge all segments in the database (including the new segment, if +** there was any data to flush) into a single segment. +*/ +SQLITE_PRIVATE int sqlite3Fts3Optimize(Fts3Table *p){ + int rc; + rc = sqlite3_exec(p->db, "SAVEPOINT fts3", 0, 0, 0); + if( rc==SQLITE_OK ){ + rc = fts3DoOptimize(p, 1); + if( rc==SQLITE_OK || rc==SQLITE_DONE ){ + int rc2 = sqlite3_exec(p->db, "RELEASE fts3", 0, 0, 0); + if( rc2!=SQLITE_OK ) rc = rc2; + }else{ + sqlite3_exec(p->db, "ROLLBACK TO fts3", 0, 0, 0); + sqlite3_exec(p->db, "RELEASE fts3", 0, 0, 0); + } + } + sqlite3Fts3SegmentsClose(p); + return rc; +} + +#endif + +/************** End of fts3_write.c ******************************************/ +/************** Begin file fts3_snippet.c ************************************/ +/* +** 2009 Oct 23 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +****************************************************************************** +*/ + +/* #include "fts3Int.h" */ +#if !defined(SQLITE_CORE) || defined(SQLITE_ENABLE_FTS3) + +/* #include */ +/* #include */ + +/* +** Characters that may appear in the second argument to matchinfo(). +*/ +#define FTS3_MATCHINFO_NPHRASE 'p' /* 1 value */ +#define FTS3_MATCHINFO_NCOL 'c' /* 1 value */ +#define FTS3_MATCHINFO_NDOC 'n' /* 1 value */ +#define FTS3_MATCHINFO_AVGLENGTH 'a' /* nCol values */ +#define FTS3_MATCHINFO_LENGTH 'l' /* nCol values */ +#define FTS3_MATCHINFO_LCS 's' /* nCol values */ +#define FTS3_MATCHINFO_HITS 'x' /* 3*nCol*nPhrase values */ +#define FTS3_MATCHINFO_LHITS 'y' /* nCol*nPhrase values */ +#define FTS3_MATCHINFO_LHITS_BM 'b' /* nCol*nPhrase values */ + +/* +** The default value for the second argument to matchinfo(). +*/ +#define FTS3_MATCHINFO_DEFAULT "pcx" + + +/* +** Used as an fts3ExprIterate() context when loading phrase doclists to +** Fts3Expr.aDoclist[]/nDoclist. +*/ +typedef struct LoadDoclistCtx LoadDoclistCtx; +struct LoadDoclistCtx { + Fts3Cursor *pCsr; /* FTS3 Cursor */ + int nPhrase; /* Number of phrases seen so far */ + int nToken; /* Number of tokens seen so far */ +}; + +/* +** The following types are used as part of the implementation of the +** fts3BestSnippet() routine. +*/ +typedef struct SnippetIter SnippetIter; +typedef struct SnippetPhrase SnippetPhrase; +typedef struct SnippetFragment SnippetFragment; + +struct SnippetIter { + Fts3Cursor *pCsr; /* Cursor snippet is being generated from */ + int iCol; /* Extract snippet from this column */ + int nSnippet; /* Requested snippet length (in tokens) */ + int nPhrase; /* Number of phrases in query */ + SnippetPhrase *aPhrase; /* Array of size nPhrase */ + int iCurrent; /* First token of current snippet */ +}; + +struct SnippetPhrase { + int nToken; /* Number of tokens in phrase */ + char *pList; /* Pointer to start of phrase position list */ + int iHead; /* Next value in position list */ + char *pHead; /* Position list data following iHead */ + int iTail; /* Next value in trailing position list */ + char *pTail; /* Position list data following iTail */ +}; + +struct SnippetFragment { + int iCol; /* Column snippet is extracted from */ + int iPos; /* Index of first token in snippet */ + u64 covered; /* Mask of query phrases covered */ + u64 hlmask; /* Mask of snippet terms to highlight */ +}; + +/* +** This type is used as an fts3ExprIterate() context object while +** accumulating the data returned by the matchinfo() function. +*/ +typedef struct MatchInfo MatchInfo; +struct MatchInfo { + Fts3Cursor *pCursor; /* FTS3 Cursor */ + int nCol; /* Number of columns in table */ + int nPhrase; /* Number of matchable phrases in query */ + sqlite3_int64 nDoc; /* Number of docs in database */ + char flag; + u32 *aMatchinfo; /* Pre-allocated buffer */ +}; + +/* +** An instance of this structure is used to manage a pair of buffers, each +** (nElem * sizeof(u32)) bytes in size. See the MatchinfoBuffer code below +** for details. +*/ +struct MatchinfoBuffer { + u8 aRef[3]; + int nElem; + int bGlobal; /* Set if global data is loaded */ + char *zMatchinfo; + u32 aMatchinfo[1]; +}; + + +/* +** The snippet() and offsets() functions both return text values. An instance +** of the following structure is used to accumulate those values while the +** functions are running. See fts3StringAppend() for details. +*/ +typedef struct StrBuffer StrBuffer; +struct StrBuffer { + char *z; /* Pointer to buffer containing string */ + int n; /* Length of z in bytes (excl. nul-term) */ + int nAlloc; /* Allocated size of buffer z in bytes */ +}; + + +/************************************************************************* +** Start of MatchinfoBuffer code. +*/ + +/* +** Allocate a two-slot MatchinfoBuffer object. +*/ +static MatchinfoBuffer *fts3MIBufferNew(size_t nElem, const char *zMatchinfo){ + MatchinfoBuffer *pRet; + sqlite3_int64 nByte = sizeof(u32) * (2*(sqlite3_int64)nElem + 1) + + sizeof(MatchinfoBuffer); + sqlite3_int64 nStr = strlen(zMatchinfo); + + pRet = sqlite3_malloc64(nByte + nStr+1); + if( pRet ){ + memset(pRet, 0, nByte); + pRet->aMatchinfo[0] = (u8*)(&pRet->aMatchinfo[1]) - (u8*)pRet; + pRet->aMatchinfo[1+nElem] = pRet->aMatchinfo[0] + + sizeof(u32)*((int)nElem+1); + pRet->nElem = (int)nElem; + pRet->zMatchinfo = ((char*)pRet) + nByte; + memcpy(pRet->zMatchinfo, zMatchinfo, nStr+1); + pRet->aRef[0] = 1; + } + + return pRet; +} + +static void fts3MIBufferFree(void *p){ + MatchinfoBuffer *pBuf = (MatchinfoBuffer*)((u8*)p - ((u32*)p)[-1]); + + assert( (u32*)p==&pBuf->aMatchinfo[1] + || (u32*)p==&pBuf->aMatchinfo[pBuf->nElem+2] + ); + if( (u32*)p==&pBuf->aMatchinfo[1] ){ + pBuf->aRef[1] = 0; + }else{ + pBuf->aRef[2] = 0; + } + + if( pBuf->aRef[0]==0 && pBuf->aRef[1]==0 && pBuf->aRef[2]==0 ){ + sqlite3_free(pBuf); + } +} + +static void (*fts3MIBufferAlloc(MatchinfoBuffer *p, u32 **paOut))(void*){ + void (*xRet)(void*) = 0; + u32 *aOut = 0; + + if( p->aRef[1]==0 ){ + p->aRef[1] = 1; + aOut = &p->aMatchinfo[1]; + xRet = fts3MIBufferFree; + } + else if( p->aRef[2]==0 ){ + p->aRef[2] = 1; + aOut = &p->aMatchinfo[p->nElem+2]; + xRet = fts3MIBufferFree; + }else{ + aOut = (u32*)sqlite3_malloc64(p->nElem * sizeof(u32)); + if( aOut ){ + xRet = sqlite3_free; + if( p->bGlobal ) memcpy(aOut, &p->aMatchinfo[1], p->nElem*sizeof(u32)); + } + } + + *paOut = aOut; + return xRet; +} + +static void fts3MIBufferSetGlobal(MatchinfoBuffer *p){ + p->bGlobal = 1; + memcpy(&p->aMatchinfo[2+p->nElem], &p->aMatchinfo[1], p->nElem*sizeof(u32)); +} + +/* +** Free a MatchinfoBuffer object allocated using fts3MIBufferNew() +*/ +SQLITE_PRIVATE void sqlite3Fts3MIBufferFree(MatchinfoBuffer *p){ + if( p ){ + assert( p->aRef[0]==1 ); + p->aRef[0] = 0; + if( p->aRef[0]==0 && p->aRef[1]==0 && p->aRef[2]==0 ){ + sqlite3_free(p); + } + } +} + +/* +** End of MatchinfoBuffer code. +*************************************************************************/ + + +/* +** This function is used to help iterate through a position-list. A position +** list is a list of unique integers, sorted from smallest to largest. Each +** element of the list is represented by an FTS3 varint that takes the value +** of the difference between the current element and the previous one plus +** two. For example, to store the position-list: +** +** 4 9 113 +** +** the three varints: +** +** 6 7 106 +** +** are encoded. +** +** When this function is called, *pp points to the start of an element of +** the list. *piPos contains the value of the previous entry in the list. +** After it returns, *piPos contains the value of the next element of the +** list and *pp is advanced to the following varint. +*/ +static void fts3GetDeltaPosition(char **pp, int *piPos){ + int iVal; + *pp += fts3GetVarint32(*pp, &iVal); + *piPos += (iVal-2); +} + +/* +** Helper function for fts3ExprIterate() (see below). +*/ +static int fts3ExprIterate2( + Fts3Expr *pExpr, /* Expression to iterate phrases of */ + int *piPhrase, /* Pointer to phrase counter */ + int (*x)(Fts3Expr*,int,void*), /* Callback function to invoke for phrases */ + void *pCtx /* Second argument to pass to callback */ +){ + int rc; /* Return code */ + int eType = pExpr->eType; /* Type of expression node pExpr */ + + if( eType!=FTSQUERY_PHRASE ){ + assert( pExpr->pLeft && pExpr->pRight ); + rc = fts3ExprIterate2(pExpr->pLeft, piPhrase, x, pCtx); + if( rc==SQLITE_OK && eType!=FTSQUERY_NOT ){ + rc = fts3ExprIterate2(pExpr->pRight, piPhrase, x, pCtx); + } + }else{ + rc = x(pExpr, *piPhrase, pCtx); + (*piPhrase)++; + } + return rc; +} + +/* +** Iterate through all phrase nodes in an FTS3 query, except those that +** are part of a sub-tree that is the right-hand-side of a NOT operator. +** For each phrase node found, the supplied callback function is invoked. +** +** If the callback function returns anything other than SQLITE_OK, +** the iteration is abandoned and the error code returned immediately. +** Otherwise, SQLITE_OK is returned after a callback has been made for +** all eligible phrase nodes. +*/ +static int fts3ExprIterate( + Fts3Expr *pExpr, /* Expression to iterate phrases of */ + int (*x)(Fts3Expr*,int,void*), /* Callback function to invoke for phrases */ + void *pCtx /* Second argument to pass to callback */ +){ + int iPhrase = 0; /* Variable used as the phrase counter */ + return fts3ExprIterate2(pExpr, &iPhrase, x, pCtx); +} + + +/* +** This is an fts3ExprIterate() callback used while loading the doclists +** for each phrase into Fts3Expr.aDoclist[]/nDoclist. See also +** fts3ExprLoadDoclists(). +*/ +static int fts3ExprLoadDoclistsCb(Fts3Expr *pExpr, int iPhrase, void *ctx){ + int rc = SQLITE_OK; + Fts3Phrase *pPhrase = pExpr->pPhrase; + LoadDoclistCtx *p = (LoadDoclistCtx *)ctx; + + UNUSED_PARAMETER(iPhrase); + + p->nPhrase++; + p->nToken += pPhrase->nToken; + + return rc; +} + +/* +** Load the doclists for each phrase in the query associated with FTS3 cursor +** pCsr. +** +** If pnPhrase is not NULL, then *pnPhrase is set to the number of matchable +** phrases in the expression (all phrases except those directly or +** indirectly descended from the right-hand-side of a NOT operator). If +** pnToken is not NULL, then it is set to the number of tokens in all +** matchable phrases of the expression. +*/ +static int fts3ExprLoadDoclists( + Fts3Cursor *pCsr, /* Fts3 cursor for current query */ + int *pnPhrase, /* OUT: Number of phrases in query */ + int *pnToken /* OUT: Number of tokens in query */ +){ + int rc; /* Return Code */ + LoadDoclistCtx sCtx = {0,0,0}; /* Context for fts3ExprIterate() */ + sCtx.pCsr = pCsr; + rc = fts3ExprIterate(pCsr->pExpr, fts3ExprLoadDoclistsCb, (void *)&sCtx); + if( pnPhrase ) *pnPhrase = sCtx.nPhrase; + if( pnToken ) *pnToken = sCtx.nToken; + return rc; +} + +static int fts3ExprPhraseCountCb(Fts3Expr *pExpr, int iPhrase, void *ctx){ + (*(int *)ctx)++; + pExpr->iPhrase = iPhrase; + return SQLITE_OK; +} +static int fts3ExprPhraseCount(Fts3Expr *pExpr){ + int nPhrase = 0; + (void)fts3ExprIterate(pExpr, fts3ExprPhraseCountCb, (void *)&nPhrase); + return nPhrase; +} + +/* +** Advance the position list iterator specified by the first two +** arguments so that it points to the first element with a value greater +** than or equal to parameter iNext. +*/ +static void fts3SnippetAdvance(char **ppIter, int *piIter, int iNext){ + char *pIter = *ppIter; + if( pIter ){ + int iIter = *piIter; + + while( iIteriCurrent<0 ){ + /* The SnippetIter object has just been initialized. The first snippet + ** candidate always starts at offset 0 (even if this candidate has a + ** score of 0.0). + */ + pIter->iCurrent = 0; + + /* Advance the 'head' iterator of each phrase to the first offset that + ** is greater than or equal to (iNext+nSnippet). + */ + for(i=0; inPhrase; i++){ + SnippetPhrase *pPhrase = &pIter->aPhrase[i]; + fts3SnippetAdvance(&pPhrase->pHead, &pPhrase->iHead, pIter->nSnippet); + } + }else{ + int iStart; + int iEnd = 0x7FFFFFFF; + + for(i=0; inPhrase; i++){ + SnippetPhrase *pPhrase = &pIter->aPhrase[i]; + if( pPhrase->pHead && pPhrase->iHeadiHead; + } + } + if( iEnd==0x7FFFFFFF ){ + return 1; + } + + pIter->iCurrent = iStart = iEnd - pIter->nSnippet + 1; + for(i=0; inPhrase; i++){ + SnippetPhrase *pPhrase = &pIter->aPhrase[i]; + fts3SnippetAdvance(&pPhrase->pHead, &pPhrase->iHead, iEnd+1); + fts3SnippetAdvance(&pPhrase->pTail, &pPhrase->iTail, iStart); + } + } + + return 0; +} + +/* +** Retrieve information about the current candidate snippet of snippet +** iterator pIter. +*/ +static void fts3SnippetDetails( + SnippetIter *pIter, /* Snippet iterator */ + u64 mCovered, /* Bitmask of phrases already covered */ + int *piToken, /* OUT: First token of proposed snippet */ + int *piScore, /* OUT: "Score" for this snippet */ + u64 *pmCover, /* OUT: Bitmask of phrases covered */ + u64 *pmHighlight /* OUT: Bitmask of terms to highlight */ +){ + int iStart = pIter->iCurrent; /* First token of snippet */ + int iScore = 0; /* Score of this snippet */ + int i; /* Loop counter */ + u64 mCover = 0; /* Mask of phrases covered by this snippet */ + u64 mHighlight = 0; /* Mask of tokens to highlight in snippet */ + + for(i=0; inPhrase; i++){ + SnippetPhrase *pPhrase = &pIter->aPhrase[i]; + if( pPhrase->pTail ){ + char *pCsr = pPhrase->pTail; + int iCsr = pPhrase->iTail; + + while( iCsr<(iStart+pIter->nSnippet) && iCsr>=iStart ){ + int j; + u64 mPhrase = (u64)1 << (i%64); + u64 mPos = (u64)1 << (iCsr - iStart); + assert( iCsr>=iStart && (iCsr - iStart)<=64 ); + assert( i>=0 ); + if( (mCover|mCovered)&mPhrase ){ + iScore++; + }else{ + iScore += 1000; + } + mCover |= mPhrase; + + for(j=0; jnToken; j++){ + mHighlight |= (mPos>>j); + } + + if( 0==(*pCsr & 0x0FE) ) break; + fts3GetDeltaPosition(&pCsr, &iCsr); + } + } + } + + /* Set the output variables before returning. */ + *piToken = iStart; + *piScore = iScore; + *pmCover = mCover; + *pmHighlight = mHighlight; +} + +/* +** This function is an fts3ExprIterate() callback used by fts3BestSnippet(). +** Each invocation populates an element of the SnippetIter.aPhrase[] array. +*/ +static int fts3SnippetFindPositions(Fts3Expr *pExpr, int iPhrase, void *ctx){ + SnippetIter *p = (SnippetIter *)ctx; + SnippetPhrase *pPhrase = &p->aPhrase[iPhrase]; + char *pCsr; + int rc; + + pPhrase->nToken = pExpr->pPhrase->nToken; + rc = sqlite3Fts3EvalPhrasePoslist(p->pCsr, pExpr, p->iCol, &pCsr); + assert( rc==SQLITE_OK || pCsr==0 ); + if( pCsr ){ + int iFirst = 0; + pPhrase->pList = pCsr; + fts3GetDeltaPosition(&pCsr, &iFirst); + if( iFirst<0 ){ + rc = FTS_CORRUPT_VTAB; + }else{ + pPhrase->pHead = pCsr; + pPhrase->pTail = pCsr; + pPhrase->iHead = iFirst; + pPhrase->iTail = iFirst; + } + }else{ + assert( rc!=SQLITE_OK || ( + pPhrase->pList==0 && pPhrase->pHead==0 && pPhrase->pTail==0 + )); + } + + return rc; +} + +/* +** Select the fragment of text consisting of nFragment contiguous tokens +** from column iCol that represent the "best" snippet. The best snippet +** is the snippet with the highest score, where scores are calculated +** by adding: +** +** (a) +1 point for each occurrence of a matchable phrase in the snippet. +** +** (b) +1000 points for the first occurrence of each matchable phrase in +** the snippet for which the corresponding mCovered bit is not set. +** +** The selected snippet parameters are stored in structure *pFragment before +** returning. The score of the selected snippet is stored in *piScore +** before returning. +*/ +static int fts3BestSnippet( + int nSnippet, /* Desired snippet length */ + Fts3Cursor *pCsr, /* Cursor to create snippet for */ + int iCol, /* Index of column to create snippet from */ + u64 mCovered, /* Mask of phrases already covered */ + u64 *pmSeen, /* IN/OUT: Mask of phrases seen */ + SnippetFragment *pFragment, /* OUT: Best snippet found */ + int *piScore /* OUT: Score of snippet pFragment */ +){ + int rc; /* Return Code */ + int nList; /* Number of phrases in expression */ + SnippetIter sIter; /* Iterates through snippet candidates */ + sqlite3_int64 nByte; /* Number of bytes of space to allocate */ + int iBestScore = -1; /* Best snippet score found so far */ + int i; /* Loop counter */ + + memset(&sIter, 0, sizeof(sIter)); + + /* Iterate through the phrases in the expression to count them. The same + ** callback makes sure the doclists are loaded for each phrase. + */ + rc = fts3ExprLoadDoclists(pCsr, &nList, 0); + if( rc!=SQLITE_OK ){ + return rc; + } + + /* Now that it is known how many phrases there are, allocate and zero + ** the required space using malloc(). + */ + nByte = sizeof(SnippetPhrase) * nList; + sIter.aPhrase = (SnippetPhrase *)sqlite3_malloc64(nByte); + if( !sIter.aPhrase ){ + return SQLITE_NOMEM; + } + memset(sIter.aPhrase, 0, nByte); + + /* Initialize the contents of the SnippetIter object. Then iterate through + ** the set of phrases in the expression to populate the aPhrase[] array. + */ + sIter.pCsr = pCsr; + sIter.iCol = iCol; + sIter.nSnippet = nSnippet; + sIter.nPhrase = nList; + sIter.iCurrent = -1; + rc = fts3ExprIterate(pCsr->pExpr, fts3SnippetFindPositions, (void*)&sIter); + if( rc==SQLITE_OK ){ + + /* Set the *pmSeen output variable. */ + for(i=0; iiCol = iCol; + while( !fts3SnippetNextCandidate(&sIter) ){ + int iPos; + int iScore; + u64 mCover; + u64 mHighlite; + fts3SnippetDetails(&sIter, mCovered, &iPos, &iScore, &mCover,&mHighlite); + assert( iScore>=0 ); + if( iScore>iBestScore ){ + pFragment->iPos = iPos; + pFragment->hlmask = mHighlite; + pFragment->covered = mCover; + iBestScore = iScore; + } + } + + *piScore = iBestScore; + } + sqlite3_free(sIter.aPhrase); + return rc; +} + + +/* +** Append a string to the string-buffer passed as the first argument. +** +** If nAppend is negative, then the length of the string zAppend is +** determined using strlen(). +*/ +static int fts3StringAppend( + StrBuffer *pStr, /* Buffer to append to */ + const char *zAppend, /* Pointer to data to append to buffer */ + int nAppend /* Size of zAppend in bytes (or -1) */ +){ + if( nAppend<0 ){ + nAppend = (int)strlen(zAppend); + } + + /* If there is insufficient space allocated at StrBuffer.z, use realloc() + ** to grow the buffer until so that it is big enough to accomadate the + ** appended data. + */ + if( pStr->n+nAppend+1>=pStr->nAlloc ){ + sqlite3_int64 nAlloc = pStr->nAlloc+(sqlite3_int64)nAppend+100; + char *zNew = sqlite3_realloc64(pStr->z, nAlloc); + if( !zNew ){ + return SQLITE_NOMEM; + } + pStr->z = zNew; + pStr->nAlloc = nAlloc; + } + assert( pStr->z!=0 && (pStr->nAlloc >= pStr->n+nAppend+1) ); + + /* Append the data to the string buffer. */ + memcpy(&pStr->z[pStr->n], zAppend, nAppend); + pStr->n += nAppend; + pStr->z[pStr->n] = '\0'; + + return SQLITE_OK; +} + +/* +** The fts3BestSnippet() function often selects snippets that end with a +** query term. That is, the final term of the snippet is always a term +** that requires highlighting. For example, if 'X' is a highlighted term +** and '.' is a non-highlighted term, BestSnippet() may select: +** +** ........X.....X +** +** This function "shifts" the beginning of the snippet forward in the +** document so that there are approximately the same number of +** non-highlighted terms to the right of the final highlighted term as there +** are to the left of the first highlighted term. For example, to this: +** +** ....X.....X.... +** +** This is done as part of extracting the snippet text, not when selecting +** the snippet. Snippet selection is done based on doclists only, so there +** is no way for fts3BestSnippet() to know whether or not the document +** actually contains terms that follow the final highlighted term. +*/ +static int fts3SnippetShift( + Fts3Table *pTab, /* FTS3 table snippet comes from */ + int iLangid, /* Language id to use in tokenizing */ + int nSnippet, /* Number of tokens desired for snippet */ + const char *zDoc, /* Document text to extract snippet from */ + int nDoc, /* Size of buffer zDoc in bytes */ + int *piPos, /* IN/OUT: First token of snippet */ + u64 *pHlmask /* IN/OUT: Mask of tokens to highlight */ +){ + u64 hlmask = *pHlmask; /* Local copy of initial highlight-mask */ + + if( hlmask ){ + int nLeft; /* Tokens to the left of first highlight */ + int nRight; /* Tokens to the right of last highlight */ + int nDesired; /* Ideal number of tokens to shift forward */ + + for(nLeft=0; !(hlmask & ((u64)1 << nLeft)); nLeft++); + for(nRight=0; !(hlmask & ((u64)1 << (nSnippet-1-nRight))); nRight++); + assert( (nSnippet-1-nRight)<=63 && (nSnippet-1-nRight)>=0 ); + nDesired = (nLeft-nRight)/2; + + /* Ideally, the start of the snippet should be pushed forward in the + ** document nDesired tokens. This block checks if there are actually + ** nDesired tokens to the right of the snippet. If so, *piPos and + ** *pHlMask are updated to shift the snippet nDesired tokens to the + ** right. Otherwise, the snippet is shifted by the number of tokens + ** available. + */ + if( nDesired>0 ){ + int nShift; /* Number of tokens to shift snippet by */ + int iCurrent = 0; /* Token counter */ + int rc; /* Return Code */ + sqlite3_tokenizer_module *pMod; + sqlite3_tokenizer_cursor *pC; + pMod = (sqlite3_tokenizer_module *)pTab->pTokenizer->pModule; + + /* Open a cursor on zDoc/nDoc. Check if there are (nSnippet+nDesired) + ** or more tokens in zDoc/nDoc. + */ + rc = sqlite3Fts3OpenTokenizer(pTab->pTokenizer, iLangid, zDoc, nDoc, &pC); + if( rc!=SQLITE_OK ){ + return rc; + } + while( rc==SQLITE_OK && iCurrent<(nSnippet+nDesired) ){ + const char *ZDUMMY; int DUMMY1 = 0, DUMMY2 = 0, DUMMY3 = 0; + rc = pMod->xNext(pC, &ZDUMMY, &DUMMY1, &DUMMY2, &DUMMY3, &iCurrent); + } + pMod->xClose(pC); + if( rc!=SQLITE_OK && rc!=SQLITE_DONE ){ return rc; } + + nShift = (rc==SQLITE_DONE)+iCurrent-nSnippet; + assert( nShift<=nDesired ); + if( nShift>0 ){ + *piPos += nShift; + *pHlmask = hlmask >> nShift; + } + } + } + return SQLITE_OK; +} + +/* +** Extract the snippet text for fragment pFragment from cursor pCsr and +** append it to string buffer pOut. +*/ +static int fts3SnippetText( + Fts3Cursor *pCsr, /* FTS3 Cursor */ + SnippetFragment *pFragment, /* Snippet to extract */ + int iFragment, /* Fragment number */ + int isLast, /* True for final fragment in snippet */ + int nSnippet, /* Number of tokens in extracted snippet */ + const char *zOpen, /* String inserted before highlighted term */ + const char *zClose, /* String inserted after highlighted term */ + const char *zEllipsis, /* String inserted between snippets */ + StrBuffer *pOut /* Write output here */ +){ + Fts3Table *pTab = (Fts3Table *)pCsr->base.pVtab; + int rc; /* Return code */ + const char *zDoc; /* Document text to extract snippet from */ + int nDoc; /* Size of zDoc in bytes */ + int iCurrent = 0; /* Current token number of document */ + int iEnd = 0; /* Byte offset of end of current token */ + int isShiftDone = 0; /* True after snippet is shifted */ + int iPos = pFragment->iPos; /* First token of snippet */ + u64 hlmask = pFragment->hlmask; /* Highlight-mask for snippet */ + int iCol = pFragment->iCol+1; /* Query column to extract text from */ + sqlite3_tokenizer_module *pMod; /* Tokenizer module methods object */ + sqlite3_tokenizer_cursor *pC; /* Tokenizer cursor open on zDoc/nDoc */ + + zDoc = (const char *)sqlite3_column_text(pCsr->pStmt, iCol); + if( zDoc==0 ){ + if( sqlite3_column_type(pCsr->pStmt, iCol)!=SQLITE_NULL ){ + return SQLITE_NOMEM; + } + return SQLITE_OK; + } + nDoc = sqlite3_column_bytes(pCsr->pStmt, iCol); + + /* Open a token cursor on the document. */ + pMod = (sqlite3_tokenizer_module *)pTab->pTokenizer->pModule; + rc = sqlite3Fts3OpenTokenizer(pTab->pTokenizer, pCsr->iLangid, zDoc,nDoc,&pC); + if( rc!=SQLITE_OK ){ + return rc; + } + + while( rc==SQLITE_OK ){ + const char *ZDUMMY; /* Dummy argument used with tokenizer */ + int DUMMY1 = -1; /* Dummy argument used with tokenizer */ + int iBegin = 0; /* Offset in zDoc of start of token */ + int iFin = 0; /* Offset in zDoc of end of token */ + int isHighlight = 0; /* True for highlighted terms */ + + /* Variable DUMMY1 is initialized to a negative value above. Elsewhere + ** in the FTS code the variable that the third argument to xNext points to + ** is initialized to zero before the first (*but not necessarily + ** subsequent*) call to xNext(). This is done for a particular application + ** that needs to know whether or not the tokenizer is being used for + ** snippet generation or for some other purpose. + ** + ** Extreme care is required when writing code to depend on this + ** initialization. It is not a documented part of the tokenizer interface. + ** If a tokenizer is used directly by any code outside of FTS, this + ** convention might not be respected. */ + rc = pMod->xNext(pC, &ZDUMMY, &DUMMY1, &iBegin, &iFin, &iCurrent); + if( rc!=SQLITE_OK ){ + if( rc==SQLITE_DONE ){ + /* Special case - the last token of the snippet is also the last token + ** of the column. Append any punctuation that occurred between the end + ** of the previous token and the end of the document to the output. + ** Then break out of the loop. */ + rc = fts3StringAppend(pOut, &zDoc[iEnd], -1); + } + break; + } + if( iCurrentiLangid, nSnippet, &zDoc[iBegin], n, &iPos, &hlmask + ); + isShiftDone = 1; + + /* Now that the shift has been done, check if the initial "..." are + ** required. They are required if (a) this is not the first fragment, + ** or (b) this fragment does not begin at position 0 of its column. + */ + if( rc==SQLITE_OK ){ + if( iPos>0 || iFragment>0 ){ + rc = fts3StringAppend(pOut, zEllipsis, -1); + }else if( iBegin ){ + rc = fts3StringAppend(pOut, zDoc, iBegin); + } + } + if( rc!=SQLITE_OK || iCurrent=(iPos+nSnippet) ){ + if( isLast ){ + rc = fts3StringAppend(pOut, zEllipsis, -1); + } + break; + } + + /* Set isHighlight to true if this term should be highlighted. */ + isHighlight = (hlmask & ((u64)1 << (iCurrent-iPos)))!=0; + + if( iCurrent>iPos ) rc = fts3StringAppend(pOut, &zDoc[iEnd], iBegin-iEnd); + if( rc==SQLITE_OK && isHighlight ) rc = fts3StringAppend(pOut, zOpen, -1); + if( rc==SQLITE_OK ) rc = fts3StringAppend(pOut, &zDoc[iBegin], iFin-iBegin); + if( rc==SQLITE_OK && isHighlight ) rc = fts3StringAppend(pOut, zClose, -1); + + iEnd = iFin; + } + + pMod->xClose(pC); + return rc; +} + + +/* +** This function is used to count the entries in a column-list (a +** delta-encoded list of term offsets within a single column of a single +** row). When this function is called, *ppCollist should point to the +** beginning of the first varint in the column-list (the varint that +** contains the position of the first matching term in the column data). +** Before returning, *ppCollist is set to point to the first byte after +** the last varint in the column-list (either the 0x00 signifying the end +** of the position-list, or the 0x01 that precedes the column number of +** the next column in the position-list). +** +** The number of elements in the column-list is returned. +*/ +static int fts3ColumnlistCount(char **ppCollist){ + char *pEnd = *ppCollist; + char c = 0; + int nEntry = 0; + + /* A column-list is terminated by either a 0x01 or 0x00. */ + while( 0xFE & (*pEnd | c) ){ + c = *pEnd++ & 0x80; + if( !c ) nEntry++; + } + + *ppCollist = pEnd; + return nEntry; +} + +/* +** This function gathers 'y' or 'b' data for a single phrase. +*/ +static int fts3ExprLHits( + Fts3Expr *pExpr, /* Phrase expression node */ + MatchInfo *p /* Matchinfo context */ +){ + Fts3Table *pTab = (Fts3Table *)p->pCursor->base.pVtab; + int iStart; + Fts3Phrase *pPhrase = pExpr->pPhrase; + char *pIter = pPhrase->doclist.pList; + int iCol = 0; + + assert( p->flag==FTS3_MATCHINFO_LHITS_BM || p->flag==FTS3_MATCHINFO_LHITS ); + if( p->flag==FTS3_MATCHINFO_LHITS ){ + iStart = pExpr->iPhrase * p->nCol; + }else{ + iStart = pExpr->iPhrase * ((p->nCol + 31) / 32); + } + + while( 1 ){ + int nHit = fts3ColumnlistCount(&pIter); + if( (pPhrase->iColumn>=pTab->nColumn || pPhrase->iColumn==iCol) ){ + if( p->flag==FTS3_MATCHINFO_LHITS ){ + p->aMatchinfo[iStart + iCol] = (u32)nHit; + }else if( nHit ){ + p->aMatchinfo[iStart + (iCol+1)/32] |= (1 << (iCol&0x1F)); + } + } + assert( *pIter==0x00 || *pIter==0x01 ); + if( *pIter!=0x01 ) break; + pIter++; + pIter += fts3GetVarint32(pIter, &iCol); + if( iCol>=p->nCol ) return FTS_CORRUPT_VTAB; + } + return SQLITE_OK; +} + +/* +** Gather the results for matchinfo directives 'y' and 'b'. +*/ +static int fts3ExprLHitGather( + Fts3Expr *pExpr, + MatchInfo *p +){ + int rc = SQLITE_OK; + assert( (pExpr->pLeft==0)==(pExpr->pRight==0) ); + if( pExpr->bEof==0 && pExpr->iDocid==p->pCursor->iPrevId ){ + if( pExpr->pLeft ){ + rc = fts3ExprLHitGather(pExpr->pLeft, p); + if( rc==SQLITE_OK ) rc = fts3ExprLHitGather(pExpr->pRight, p); + }else{ + rc = fts3ExprLHits(pExpr, p); + } + } + return rc; +} + +/* +** fts3ExprIterate() callback used to collect the "global" matchinfo stats +** for a single query. +** +** fts3ExprIterate() callback to load the 'global' elements of a +** FTS3_MATCHINFO_HITS matchinfo array. The global stats are those elements +** of the matchinfo array that are constant for all rows returned by the +** current query. +** +** Argument pCtx is actually a pointer to a struct of type MatchInfo. This +** function populates Matchinfo.aMatchinfo[] as follows: +** +** for(iCol=0; iColpCursor, pExpr, &p->aMatchinfo[3*iPhrase*p->nCol] + ); +} + +/* +** fts3ExprIterate() callback used to collect the "local" part of the +** FTS3_MATCHINFO_HITS array. The local stats are those elements of the +** array that are different for each row returned by the query. +*/ +static int fts3ExprLocalHitsCb( + Fts3Expr *pExpr, /* Phrase expression node */ + int iPhrase, /* Phrase number */ + void *pCtx /* Pointer to MatchInfo structure */ +){ + int rc = SQLITE_OK; + MatchInfo *p = (MatchInfo *)pCtx; + int iStart = iPhrase * p->nCol * 3; + int i; + + for(i=0; inCol && rc==SQLITE_OK; i++){ + char *pCsr; + rc = sqlite3Fts3EvalPhrasePoslist(p->pCursor, pExpr, i, &pCsr); + if( pCsr ){ + p->aMatchinfo[iStart+i*3] = fts3ColumnlistCount(&pCsr); + }else{ + p->aMatchinfo[iStart+i*3] = 0; + } + } + + return rc; +} + +static int fts3MatchinfoCheck( + Fts3Table *pTab, + char cArg, + char **pzErr +){ + if( (cArg==FTS3_MATCHINFO_NPHRASE) + || (cArg==FTS3_MATCHINFO_NCOL) + || (cArg==FTS3_MATCHINFO_NDOC && pTab->bFts4) + || (cArg==FTS3_MATCHINFO_AVGLENGTH && pTab->bFts4) + || (cArg==FTS3_MATCHINFO_LENGTH && pTab->bHasDocsize) + || (cArg==FTS3_MATCHINFO_LCS) + || (cArg==FTS3_MATCHINFO_HITS) + || (cArg==FTS3_MATCHINFO_LHITS) + || (cArg==FTS3_MATCHINFO_LHITS_BM) + ){ + return SQLITE_OK; + } + sqlite3Fts3ErrMsg(pzErr, "unrecognized matchinfo request: %c", cArg); + return SQLITE_ERROR; +} + +static size_t fts3MatchinfoSize(MatchInfo *pInfo, char cArg){ + size_t nVal; /* Number of integers output by cArg */ + + switch( cArg ){ + case FTS3_MATCHINFO_NDOC: + case FTS3_MATCHINFO_NPHRASE: + case FTS3_MATCHINFO_NCOL: + nVal = 1; + break; + + case FTS3_MATCHINFO_AVGLENGTH: + case FTS3_MATCHINFO_LENGTH: + case FTS3_MATCHINFO_LCS: + nVal = pInfo->nCol; + break; + + case FTS3_MATCHINFO_LHITS: + nVal = pInfo->nCol * pInfo->nPhrase; + break; + + case FTS3_MATCHINFO_LHITS_BM: + nVal = pInfo->nPhrase * ((pInfo->nCol + 31) / 32); + break; + + default: + assert( cArg==FTS3_MATCHINFO_HITS ); + nVal = pInfo->nCol * pInfo->nPhrase * 3; + break; + } + + return nVal; +} + +static int fts3MatchinfoSelectDoctotal( + Fts3Table *pTab, + sqlite3_stmt **ppStmt, + sqlite3_int64 *pnDoc, + const char **paLen +){ + sqlite3_stmt *pStmt; + const char *a; + sqlite3_int64 nDoc; + + if( !*ppStmt ){ + int rc = sqlite3Fts3SelectDoctotal(pTab, ppStmt); + if( rc!=SQLITE_OK ) return rc; + } + pStmt = *ppStmt; + assert( sqlite3_data_count(pStmt)==1 ); + + a = sqlite3_column_blob(pStmt, 0); + a += sqlite3Fts3GetVarint(a, &nDoc); + if( nDoc==0 ) return FTS_CORRUPT_VTAB; + *pnDoc = (u32)nDoc; + + if( paLen ) *paLen = a; + return SQLITE_OK; +} + +/* +** An instance of the following structure is used to store state while +** iterating through a multi-column position-list corresponding to the +** hits for a single phrase on a single row in order to calculate the +** values for a matchinfo() FTS3_MATCHINFO_LCS request. +*/ +typedef struct LcsIterator LcsIterator; +struct LcsIterator { + Fts3Expr *pExpr; /* Pointer to phrase expression */ + int iPosOffset; /* Tokens count up to end of this phrase */ + char *pRead; /* Cursor used to iterate through aDoclist */ + int iPos; /* Current position */ +}; + +/* +** If LcsIterator.iCol is set to the following value, the iterator has +** finished iterating through all offsets for all columns. +*/ +#define LCS_ITERATOR_FINISHED 0x7FFFFFFF; + +static int fts3MatchinfoLcsCb( + Fts3Expr *pExpr, /* Phrase expression node */ + int iPhrase, /* Phrase number (numbered from zero) */ + void *pCtx /* Pointer to MatchInfo structure */ +){ + LcsIterator *aIter = (LcsIterator *)pCtx; + aIter[iPhrase].pExpr = pExpr; + return SQLITE_OK; +} + +/* +** Advance the iterator passed as an argument to the next position. Return +** 1 if the iterator is at EOF or if it now points to the start of the +** position list for the next column. +*/ +static int fts3LcsIteratorAdvance(LcsIterator *pIter){ + char *pRead = pIter->pRead; + sqlite3_int64 iRead; + int rc = 0; + + pRead += sqlite3Fts3GetVarint(pRead, &iRead); + if( iRead==0 || iRead==1 ){ + pRead = 0; + rc = 1; + }else{ + pIter->iPos += (int)(iRead-2); + } + + pIter->pRead = pRead; + return rc; +} + +/* +** This function implements the FTS3_MATCHINFO_LCS matchinfo() flag. +** +** If the call is successful, the longest-common-substring lengths for each +** column are written into the first nCol elements of the pInfo->aMatchinfo[] +** array before returning. SQLITE_OK is returned in this case. +** +** Otherwise, if an error occurs, an SQLite error code is returned and the +** data written to the first nCol elements of pInfo->aMatchinfo[] is +** undefined. +*/ +static int fts3MatchinfoLcs(Fts3Cursor *pCsr, MatchInfo *pInfo){ + LcsIterator *aIter; + int i; + int iCol; + int nToken = 0; + int rc = SQLITE_OK; + + /* Allocate and populate the array of LcsIterator objects. The array + ** contains one element for each matchable phrase in the query. + **/ + aIter = sqlite3_malloc64(sizeof(LcsIterator) * pCsr->nPhrase); + if( !aIter ) return SQLITE_NOMEM; + memset(aIter, 0, sizeof(LcsIterator) * pCsr->nPhrase); + (void)fts3ExprIterate(pCsr->pExpr, fts3MatchinfoLcsCb, (void*)aIter); + + for(i=0; inPhrase; i++){ + LcsIterator *pIter = &aIter[i]; + nToken -= pIter->pExpr->pPhrase->nToken; + pIter->iPosOffset = nToken; + } + + for(iCol=0; iColnCol; iCol++){ + int nLcs = 0; /* LCS value for this column */ + int nLive = 0; /* Number of iterators in aIter not at EOF */ + + for(i=0; inPhrase; i++){ + LcsIterator *pIt = &aIter[i]; + rc = sqlite3Fts3EvalPhrasePoslist(pCsr, pIt->pExpr, iCol, &pIt->pRead); + if( rc!=SQLITE_OK ) goto matchinfo_lcs_out; + if( pIt->pRead ){ + pIt->iPos = pIt->iPosOffset; + fts3LcsIteratorAdvance(pIt); + if( pIt->pRead==0 ){ + rc = FTS_CORRUPT_VTAB; + goto matchinfo_lcs_out; + } + nLive++; + } + } + + while( nLive>0 ){ + LcsIterator *pAdv = 0; /* The iterator to advance by one position */ + int nThisLcs = 0; /* LCS for the current iterator positions */ + + for(i=0; inPhrase; i++){ + LcsIterator *pIter = &aIter[i]; + if( pIter->pRead==0 ){ + /* This iterator is already at EOF for this column. */ + nThisLcs = 0; + }else{ + if( pAdv==0 || pIter->iPosiPos ){ + pAdv = pIter; + } + if( nThisLcs==0 || pIter->iPos==pIter[-1].iPos ){ + nThisLcs++; + }else{ + nThisLcs = 1; + } + if( nThisLcs>nLcs ) nLcs = nThisLcs; + } + } + if( fts3LcsIteratorAdvance(pAdv) ) nLive--; + } + + pInfo->aMatchinfo[iCol] = nLcs; + } + + matchinfo_lcs_out: + sqlite3_free(aIter); + return rc; +} + +/* +** Populate the buffer pInfo->aMatchinfo[] with an array of integers to +** be returned by the matchinfo() function. Argument zArg contains the +** format string passed as the second argument to matchinfo (or the +** default value "pcx" if no second argument was specified). The format +** string has already been validated and the pInfo->aMatchinfo[] array +** is guaranteed to be large enough for the output. +** +** If bGlobal is true, then populate all fields of the matchinfo() output. +** If it is false, then assume that those fields that do not change between +** rows (i.e. FTS3_MATCHINFO_NPHRASE, NCOL, NDOC, AVGLENGTH and part of HITS) +** have already been populated. +** +** Return SQLITE_OK if successful, or an SQLite error code if an error +** occurs. If a value other than SQLITE_OK is returned, the state the +** pInfo->aMatchinfo[] buffer is left in is undefined. +*/ +static int fts3MatchinfoValues( + Fts3Cursor *pCsr, /* FTS3 cursor object */ + int bGlobal, /* True to grab the global stats */ + MatchInfo *pInfo, /* Matchinfo context object */ + const char *zArg /* Matchinfo format string */ +){ + int rc = SQLITE_OK; + int i; + Fts3Table *pTab = (Fts3Table *)pCsr->base.pVtab; + sqlite3_stmt *pSelect = 0; + + for(i=0; rc==SQLITE_OK && zArg[i]; i++){ + pInfo->flag = zArg[i]; + switch( zArg[i] ){ + case FTS3_MATCHINFO_NPHRASE: + if( bGlobal ) pInfo->aMatchinfo[0] = pInfo->nPhrase; + break; + + case FTS3_MATCHINFO_NCOL: + if( bGlobal ) pInfo->aMatchinfo[0] = pInfo->nCol; + break; + + case FTS3_MATCHINFO_NDOC: + if( bGlobal ){ + sqlite3_int64 nDoc = 0; + rc = fts3MatchinfoSelectDoctotal(pTab, &pSelect, &nDoc, 0); + pInfo->aMatchinfo[0] = (u32)nDoc; + } + break; + + case FTS3_MATCHINFO_AVGLENGTH: + if( bGlobal ){ + sqlite3_int64 nDoc; /* Number of rows in table */ + const char *a; /* Aggregate column length array */ + + rc = fts3MatchinfoSelectDoctotal(pTab, &pSelect, &nDoc, &a); + if( rc==SQLITE_OK ){ + int iCol; + for(iCol=0; iColnCol; iCol++){ + u32 iVal; + sqlite3_int64 nToken; + a += sqlite3Fts3GetVarint(a, &nToken); + iVal = (u32)(((u32)(nToken&0xffffffff)+nDoc/2)/nDoc); + pInfo->aMatchinfo[iCol] = iVal; + } + } + } + break; + + case FTS3_MATCHINFO_LENGTH: { + sqlite3_stmt *pSelectDocsize = 0; + rc = sqlite3Fts3SelectDocsize(pTab, pCsr->iPrevId, &pSelectDocsize); + if( rc==SQLITE_OK ){ + int iCol; + const char *a = sqlite3_column_blob(pSelectDocsize, 0); + for(iCol=0; iColnCol; iCol++){ + sqlite3_int64 nToken; + a += sqlite3Fts3GetVarint(a, &nToken); + pInfo->aMatchinfo[iCol] = (u32)nToken; + } + } + sqlite3_reset(pSelectDocsize); + break; + } + + case FTS3_MATCHINFO_LCS: + rc = fts3ExprLoadDoclists(pCsr, 0, 0); + if( rc==SQLITE_OK ){ + rc = fts3MatchinfoLcs(pCsr, pInfo); + } + break; + + case FTS3_MATCHINFO_LHITS_BM: + case FTS3_MATCHINFO_LHITS: { + size_t nZero = fts3MatchinfoSize(pInfo, zArg[i]) * sizeof(u32); + memset(pInfo->aMatchinfo, 0, nZero); + rc = fts3ExprLHitGather(pCsr->pExpr, pInfo); + break; + } + + default: { + Fts3Expr *pExpr; + assert( zArg[i]==FTS3_MATCHINFO_HITS ); + pExpr = pCsr->pExpr; + rc = fts3ExprLoadDoclists(pCsr, 0, 0); + if( rc!=SQLITE_OK ) break; + if( bGlobal ){ + if( pCsr->pDeferred ){ + rc = fts3MatchinfoSelectDoctotal(pTab, &pSelect, &pInfo->nDoc, 0); + if( rc!=SQLITE_OK ) break; + } + rc = fts3ExprIterate(pExpr, fts3ExprGlobalHitsCb,(void*)pInfo); + sqlite3Fts3EvalTestDeferred(pCsr, &rc); + if( rc!=SQLITE_OK ) break; + } + (void)fts3ExprIterate(pExpr, fts3ExprLocalHitsCb,(void*)pInfo); + break; + } + } + + pInfo->aMatchinfo += fts3MatchinfoSize(pInfo, zArg[i]); + } + + sqlite3_reset(pSelect); + return rc; +} + + +/* +** Populate pCsr->aMatchinfo[] with data for the current row. The +** 'matchinfo' data is an array of 32-bit unsigned integers (C type u32). +*/ +static void fts3GetMatchinfo( + sqlite3_context *pCtx, /* Return results here */ + Fts3Cursor *pCsr, /* FTS3 Cursor object */ + const char *zArg /* Second argument to matchinfo() function */ +){ + MatchInfo sInfo; + Fts3Table *pTab = (Fts3Table *)pCsr->base.pVtab; + int rc = SQLITE_OK; + int bGlobal = 0; /* Collect 'global' stats as well as local */ + + u32 *aOut = 0; + void (*xDestroyOut)(void*) = 0; + + memset(&sInfo, 0, sizeof(MatchInfo)); + sInfo.pCursor = pCsr; + sInfo.nCol = pTab->nColumn; + + /* If there is cached matchinfo() data, but the format string for the + ** cache does not match the format string for this request, discard + ** the cached data. */ + if( pCsr->pMIBuffer && strcmp(pCsr->pMIBuffer->zMatchinfo, zArg) ){ + sqlite3Fts3MIBufferFree(pCsr->pMIBuffer); + pCsr->pMIBuffer = 0; + } + + /* If Fts3Cursor.pMIBuffer is NULL, then this is the first time the + ** matchinfo function has been called for this query. In this case + ** allocate the array used to accumulate the matchinfo data and + ** initialize those elements that are constant for every row. + */ + if( pCsr->pMIBuffer==0 ){ + size_t nMatchinfo = 0; /* Number of u32 elements in match-info */ + int i; /* Used to iterate through zArg */ + + /* Determine the number of phrases in the query */ + pCsr->nPhrase = fts3ExprPhraseCount(pCsr->pExpr); + sInfo.nPhrase = pCsr->nPhrase; + + /* Determine the number of integers in the buffer returned by this call. */ + for(i=0; zArg[i]; i++){ + char *zErr = 0; + if( fts3MatchinfoCheck(pTab, zArg[i], &zErr) ){ + sqlite3_result_error(pCtx, zErr, -1); + sqlite3_free(zErr); + return; + } + nMatchinfo += fts3MatchinfoSize(&sInfo, zArg[i]); + } + + /* Allocate space for Fts3Cursor.aMatchinfo[] and Fts3Cursor.zMatchinfo. */ + pCsr->pMIBuffer = fts3MIBufferNew(nMatchinfo, zArg); + if( !pCsr->pMIBuffer ) rc = SQLITE_NOMEM; + + pCsr->isMatchinfoNeeded = 1; + bGlobal = 1; + } + + if( rc==SQLITE_OK ){ + xDestroyOut = fts3MIBufferAlloc(pCsr->pMIBuffer, &aOut); + if( xDestroyOut==0 ){ + rc = SQLITE_NOMEM; + } + } + + if( rc==SQLITE_OK ){ + sInfo.aMatchinfo = aOut; + sInfo.nPhrase = pCsr->nPhrase; + rc = fts3MatchinfoValues(pCsr, bGlobal, &sInfo, zArg); + if( bGlobal ){ + fts3MIBufferSetGlobal(pCsr->pMIBuffer); + } + } + + if( rc!=SQLITE_OK ){ + sqlite3_result_error_code(pCtx, rc); + if( xDestroyOut ) xDestroyOut(aOut); + }else{ + int n = pCsr->pMIBuffer->nElem * sizeof(u32); + sqlite3_result_blob(pCtx, aOut, n, xDestroyOut); + } +} + +/* +** Implementation of snippet() function. +*/ +SQLITE_PRIVATE void sqlite3Fts3Snippet( + sqlite3_context *pCtx, /* SQLite function call context */ + Fts3Cursor *pCsr, /* Cursor object */ + const char *zStart, /* Snippet start text - "" */ + const char *zEnd, /* Snippet end text - "" */ + const char *zEllipsis, /* Snippet ellipsis text - "..." */ + int iCol, /* Extract snippet from this column */ + int nToken /* Approximate number of tokens in snippet */ +){ + Fts3Table *pTab = (Fts3Table *)pCsr->base.pVtab; + int rc = SQLITE_OK; + int i; + StrBuffer res = {0, 0, 0}; + + /* The returned text includes up to four fragments of text extracted from + ** the data in the current row. The first iteration of the for(...) loop + ** below attempts to locate a single fragment of text nToken tokens in + ** size that contains at least one instance of all phrases in the query + ** expression that appear in the current row. If such a fragment of text + ** cannot be found, the second iteration of the loop attempts to locate + ** a pair of fragments, and so on. + */ + int nSnippet = 0; /* Number of fragments in this snippet */ + SnippetFragment aSnippet[4]; /* Maximum of 4 fragments per snippet */ + int nFToken = -1; /* Number of tokens in each fragment */ + + if( !pCsr->pExpr ){ + sqlite3_result_text(pCtx, "", 0, SQLITE_STATIC); + return; + } + + /* Limit the snippet length to 64 tokens. */ + if( nToken<-64 ) nToken = -64; + if( nToken>+64 ) nToken = +64; + + for(nSnippet=1; 1; nSnippet++){ + + int iSnip; /* Loop counter 0..nSnippet-1 */ + u64 mCovered = 0; /* Bitmask of phrases covered by snippet */ + u64 mSeen = 0; /* Bitmask of phrases seen by BestSnippet() */ + + if( nToken>=0 ){ + nFToken = (nToken+nSnippet-1) / nSnippet; + }else{ + nFToken = -1 * nToken; + } + + for(iSnip=0; iSnipnColumn; iRead++){ + SnippetFragment sF = {0, 0, 0, 0}; + int iS = 0; + if( iCol>=0 && iRead!=iCol ) continue; + + /* Find the best snippet of nFToken tokens in column iRead. */ + rc = fts3BestSnippet(nFToken, pCsr, iRead, mCovered, &mSeen, &sF, &iS); + if( rc!=SQLITE_OK ){ + goto snippet_out; + } + if( iS>iBestScore ){ + *pFragment = sF; + iBestScore = iS; + } + } + + mCovered |= pFragment->covered; + } + + /* If all query phrases seen by fts3BestSnippet() are present in at least + ** one of the nSnippet snippet fragments, break out of the loop. + */ + assert( (mCovered&mSeen)==mCovered ); + if( mSeen==mCovered || nSnippet==SizeofArray(aSnippet) ) break; + } + + assert( nFToken>0 ); + + for(i=0; ipCsr, pExpr, p->iCol, &pList); + nTerm = pExpr->pPhrase->nToken; + if( pList ){ + fts3GetDeltaPosition(&pList, &iPos); + assert_fts3_nc( iPos>=0 ); + } + + for(iTerm=0; iTermaTerm[p->iTerm++]; + pT->iOff = nTerm-iTerm-1; + pT->pList = pList; + pT->iPos = iPos; + } + + return rc; +} + +/* +** Implementation of offsets() function. +*/ +SQLITE_PRIVATE void sqlite3Fts3Offsets( + sqlite3_context *pCtx, /* SQLite function call context */ + Fts3Cursor *pCsr /* Cursor object */ +){ + Fts3Table *pTab = (Fts3Table *)pCsr->base.pVtab; + sqlite3_tokenizer_module const *pMod = pTab->pTokenizer->pModule; + int rc; /* Return Code */ + int nToken; /* Number of tokens in query */ + int iCol; /* Column currently being processed */ + StrBuffer res = {0, 0, 0}; /* Result string */ + TermOffsetCtx sCtx; /* Context for fts3ExprTermOffsetInit() */ + + if( !pCsr->pExpr ){ + sqlite3_result_text(pCtx, "", 0, SQLITE_STATIC); + return; + } + + memset(&sCtx, 0, sizeof(sCtx)); + assert( pCsr->isRequireSeek==0 ); + + /* Count the number of terms in the query */ + rc = fts3ExprLoadDoclists(pCsr, 0, &nToken); + if( rc!=SQLITE_OK ) goto offsets_out; + + /* Allocate the array of TermOffset iterators. */ + sCtx.aTerm = (TermOffset *)sqlite3_malloc64(sizeof(TermOffset)*nToken); + if( 0==sCtx.aTerm ){ + rc = SQLITE_NOMEM; + goto offsets_out; + } + sCtx.iDocid = pCsr->iPrevId; + sCtx.pCsr = pCsr; + + /* Loop through the table columns, appending offset information to + ** string-buffer res for each column. + */ + for(iCol=0; iColnColumn; iCol++){ + sqlite3_tokenizer_cursor *pC; /* Tokenizer cursor */ + const char *ZDUMMY; /* Dummy argument used with xNext() */ + int NDUMMY = 0; /* Dummy argument used with xNext() */ + int iStart = 0; + int iEnd = 0; + int iCurrent = 0; + const char *zDoc; + int nDoc; + + /* Initialize the contents of sCtx.aTerm[] for column iCol. There is + ** no way that this operation can fail, so the return code from + ** fts3ExprIterate() can be discarded. + */ + sCtx.iCol = iCol; + sCtx.iTerm = 0; + (void)fts3ExprIterate(pCsr->pExpr, fts3ExprTermOffsetInit, (void*)&sCtx); + + /* Retreive the text stored in column iCol. If an SQL NULL is stored + ** in column iCol, jump immediately to the next iteration of the loop. + ** If an OOM occurs while retrieving the data (this can happen if SQLite + ** needs to transform the data from utf-16 to utf-8), return SQLITE_NOMEM + ** to the caller. + */ + zDoc = (const char *)sqlite3_column_text(pCsr->pStmt, iCol+1); + nDoc = sqlite3_column_bytes(pCsr->pStmt, iCol+1); + if( zDoc==0 ){ + if( sqlite3_column_type(pCsr->pStmt, iCol+1)==SQLITE_NULL ){ + continue; + } + rc = SQLITE_NOMEM; + goto offsets_out; + } + + /* Initialize a tokenizer iterator to iterate through column iCol. */ + rc = sqlite3Fts3OpenTokenizer(pTab->pTokenizer, pCsr->iLangid, + zDoc, nDoc, &pC + ); + if( rc!=SQLITE_OK ) goto offsets_out; + + rc = pMod->xNext(pC, &ZDUMMY, &NDUMMY, &iStart, &iEnd, &iCurrent); + while( rc==SQLITE_OK ){ + int i; /* Used to loop through terms */ + int iMinPos = 0x7FFFFFFF; /* Position of next token */ + TermOffset *pTerm = 0; /* TermOffset associated with next token */ + + for(i=0; ipList && (pT->iPos-pT->iOff)iPos-pT->iOff; + pTerm = pT; + } + } + + if( !pTerm ){ + /* All offsets for this column have been gathered. */ + rc = SQLITE_DONE; + }else{ + assert_fts3_nc( iCurrent<=iMinPos ); + if( 0==(0xFE&*pTerm->pList) ){ + pTerm->pList = 0; + }else{ + fts3GetDeltaPosition(&pTerm->pList, &pTerm->iPos); + } + while( rc==SQLITE_OK && iCurrentxNext(pC, &ZDUMMY, &NDUMMY, &iStart, &iEnd, &iCurrent); + } + if( rc==SQLITE_OK ){ + char aBuffer[64]; + sqlite3_snprintf(sizeof(aBuffer), aBuffer, + "%d %d %d %d ", iCol, pTerm-sCtx.aTerm, iStart, iEnd-iStart + ); + rc = fts3StringAppend(&res, aBuffer, -1); + }else if( rc==SQLITE_DONE && pTab->zContentTbl==0 ){ + rc = FTS_CORRUPT_VTAB; + } + } + } + if( rc==SQLITE_DONE ){ + rc = SQLITE_OK; + } + + pMod->xClose(pC); + if( rc!=SQLITE_OK ) goto offsets_out; + } + + offsets_out: + sqlite3_free(sCtx.aTerm); + assert( rc!=SQLITE_DONE ); + sqlite3Fts3SegmentsClose(pTab); + if( rc!=SQLITE_OK ){ + sqlite3_result_error_code(pCtx, rc); + sqlite3_free(res.z); + }else{ + sqlite3_result_text(pCtx, res.z, res.n-1, sqlite3_free); + } + return; +} + +/* +** Implementation of matchinfo() function. +*/ +SQLITE_PRIVATE void sqlite3Fts3Matchinfo( + sqlite3_context *pContext, /* Function call context */ + Fts3Cursor *pCsr, /* FTS3 table cursor */ + const char *zArg /* Second arg to matchinfo() function */ +){ + Fts3Table *pTab = (Fts3Table *)pCsr->base.pVtab; + const char *zFormat; + + if( zArg ){ + zFormat = zArg; + }else{ + zFormat = FTS3_MATCHINFO_DEFAULT; + } + + if( !pCsr->pExpr ){ + sqlite3_result_blob(pContext, "", 0, SQLITE_STATIC); + return; + }else{ + /* Retrieve matchinfo() data. */ + fts3GetMatchinfo(pContext, pCsr, zFormat); + sqlite3Fts3SegmentsClose(pTab); + } +} + +#endif + +/************** End of fts3_snippet.c ****************************************/ +/************** Begin file fts3_unicode.c ************************************/ +/* +** 2012 May 24 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +****************************************************************************** +** +** Implementation of the "unicode" full-text-search tokenizer. +*/ + +#ifndef SQLITE_DISABLE_FTS3_UNICODE + +/* #include "fts3Int.h" */ +#if !defined(SQLITE_CORE) || defined(SQLITE_ENABLE_FTS3) + +/* #include */ +/* #include */ +/* #include */ +/* #include */ + +/* #include "fts3_tokenizer.h" */ + +/* +** The following two macros - READ_UTF8 and WRITE_UTF8 - have been copied +** from the sqlite3 source file utf.c. If this file is compiled as part +** of the amalgamation, they are not required. +*/ +#ifndef SQLITE_AMALGAMATION + +static const unsigned char sqlite3Utf8Trans1[] = { + 0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07, + 0x08, 0x09, 0x0a, 0x0b, 0x0c, 0x0d, 0x0e, 0x0f, + 0x10, 0x11, 0x12, 0x13, 0x14, 0x15, 0x16, 0x17, + 0x18, 0x19, 0x1a, 0x1b, 0x1c, 0x1d, 0x1e, 0x1f, + 0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07, + 0x08, 0x09, 0x0a, 0x0b, 0x0c, 0x0d, 0x0e, 0x0f, + 0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07, + 0x00, 0x01, 0x02, 0x03, 0x00, 0x01, 0x00, 0x00, +}; + +#define READ_UTF8(zIn, zTerm, c) \ + c = *(zIn++); \ + if( c>=0xc0 ){ \ + c = sqlite3Utf8Trans1[c-0xc0]; \ + while( zIn!=zTerm && (*zIn & 0xc0)==0x80 ){ \ + c = (c<<6) + (0x3f & *(zIn++)); \ + } \ + if( c<0x80 \ + || (c&0xFFFFF800)==0xD800 \ + || (c&0xFFFFFFFE)==0xFFFE ){ c = 0xFFFD; } \ + } + +#define WRITE_UTF8(zOut, c) { \ + if( c<0x00080 ){ \ + *zOut++ = (u8)(c&0xFF); \ + } \ + else if( c<0x00800 ){ \ + *zOut++ = 0xC0 + (u8)((c>>6)&0x1F); \ + *zOut++ = 0x80 + (u8)(c & 0x3F); \ + } \ + else if( c<0x10000 ){ \ + *zOut++ = 0xE0 + (u8)((c>>12)&0x0F); \ + *zOut++ = 0x80 + (u8)((c>>6) & 0x3F); \ + *zOut++ = 0x80 + (u8)(c & 0x3F); \ + }else{ \ + *zOut++ = 0xF0 + (u8)((c>>18) & 0x07); \ + *zOut++ = 0x80 + (u8)((c>>12) & 0x3F); \ + *zOut++ = 0x80 + (u8)((c>>6) & 0x3F); \ + *zOut++ = 0x80 + (u8)(c & 0x3F); \ + } \ +} + +#endif /* ifndef SQLITE_AMALGAMATION */ + +typedef struct unicode_tokenizer unicode_tokenizer; +typedef struct unicode_cursor unicode_cursor; + +struct unicode_tokenizer { + sqlite3_tokenizer base; + int eRemoveDiacritic; + int nException; + int *aiException; +}; + +struct unicode_cursor { + sqlite3_tokenizer_cursor base; + const unsigned char *aInput; /* Input text being tokenized */ + int nInput; /* Size of aInput[] in bytes */ + int iOff; /* Current offset within aInput[] */ + int iToken; /* Index of next token to be returned */ + char *zToken; /* storage for current token */ + int nAlloc; /* space allocated at zToken */ +}; + + +/* +** Destroy a tokenizer allocated by unicodeCreate(). +*/ +static int unicodeDestroy(sqlite3_tokenizer *pTokenizer){ + if( pTokenizer ){ + unicode_tokenizer *p = (unicode_tokenizer *)pTokenizer; + sqlite3_free(p->aiException); + sqlite3_free(p); + } + return SQLITE_OK; +} + +/* +** As part of a tokenchars= or separators= option, the CREATE VIRTUAL TABLE +** statement has specified that the tokenizer for this table shall consider +** all characters in string zIn/nIn to be separators (if bAlnum==0) or +** token characters (if bAlnum==1). +** +** For each codepoint in the zIn/nIn string, this function checks if the +** sqlite3FtsUnicodeIsalnum() function already returns the desired result. +** If so, no action is taken. Otherwise, the codepoint is added to the +** unicode_tokenizer.aiException[] array. For the purposes of tokenization, +** the return value of sqlite3FtsUnicodeIsalnum() is inverted for all +** codepoints in the aiException[] array. +** +** If a standalone diacritic mark (one that sqlite3FtsUnicodeIsdiacritic() +** identifies as a diacritic) occurs in the zIn/nIn string it is ignored. +** It is not possible to change the behavior of the tokenizer with respect +** to these codepoints. +*/ +static int unicodeAddExceptions( + unicode_tokenizer *p, /* Tokenizer to add exceptions to */ + int bAlnum, /* Replace Isalnum() return value with this */ + const char *zIn, /* Array of characters to make exceptions */ + int nIn /* Length of z in bytes */ +){ + const unsigned char *z = (const unsigned char *)zIn; + const unsigned char *zTerm = &z[nIn]; + unsigned int iCode; + int nEntry = 0; + + assert( bAlnum==0 || bAlnum==1 ); + + while( zaiException,(p->nException+nEntry)*sizeof(int)); + if( aNew==0 ) return SQLITE_NOMEM; + nNew = p->nException; + + z = (const unsigned char *)zIn; + while( zi; j--) aNew[j] = aNew[j-1]; + aNew[i] = (int)iCode; + nNew++; + } + } + p->aiException = aNew; + p->nException = nNew; + } + + return SQLITE_OK; +} + +/* +** Return true if the p->aiException[] array contains the value iCode. +*/ +static int unicodeIsException(unicode_tokenizer *p, int iCode){ + if( p->nException>0 ){ + int *a = p->aiException; + int iLo = 0; + int iHi = p->nException-1; + + while( iHi>=iLo ){ + int iTest = (iHi + iLo) / 2; + if( iCode==a[iTest] ){ + return 1; + }else if( iCode>a[iTest] ){ + iLo = iTest+1; + }else{ + iHi = iTest-1; + } + } + } + + return 0; +} + +/* +** Return true if, for the purposes of tokenization, codepoint iCode is +** considered a token character (not a separator). +*/ +static int unicodeIsAlnum(unicode_tokenizer *p, int iCode){ + assert( (sqlite3FtsUnicodeIsalnum(iCode) & 0xFFFFFFFE)==0 ); + return sqlite3FtsUnicodeIsalnum(iCode) ^ unicodeIsException(p, iCode); +} + +/* +** Create a new tokenizer instance. +*/ +static int unicodeCreate( + int nArg, /* Size of array argv[] */ + const char * const *azArg, /* Tokenizer creation arguments */ + sqlite3_tokenizer **pp /* OUT: New tokenizer handle */ +){ + unicode_tokenizer *pNew; /* New tokenizer object */ + int i; + int rc = SQLITE_OK; + + pNew = (unicode_tokenizer *) sqlite3_malloc(sizeof(unicode_tokenizer)); + if( pNew==NULL ) return SQLITE_NOMEM; + memset(pNew, 0, sizeof(unicode_tokenizer)); + pNew->eRemoveDiacritic = 1; + + for(i=0; rc==SQLITE_OK && ieRemoveDiacritic = 1; + } + else if( n==19 && memcmp("remove_diacritics=0", z, 19)==0 ){ + pNew->eRemoveDiacritic = 0; + } + else if( n==19 && memcmp("remove_diacritics=2", z, 19)==0 ){ + pNew->eRemoveDiacritic = 2; + } + else if( n>=11 && memcmp("tokenchars=", z, 11)==0 ){ + rc = unicodeAddExceptions(pNew, 1, &z[11], n-11); + } + else if( n>=11 && memcmp("separators=", z, 11)==0 ){ + rc = unicodeAddExceptions(pNew, 0, &z[11], n-11); + } + else{ + /* Unrecognized argument */ + rc = SQLITE_ERROR; + } + } + + if( rc!=SQLITE_OK ){ + unicodeDestroy((sqlite3_tokenizer *)pNew); + pNew = 0; + } + *pp = (sqlite3_tokenizer *)pNew; + return rc; +} + +/* +** Prepare to begin tokenizing a particular string. The input +** string to be tokenized is pInput[0..nBytes-1]. A cursor +** used to incrementally tokenize this string is returned in +** *ppCursor. +*/ +static int unicodeOpen( + sqlite3_tokenizer *p, /* The tokenizer */ + const char *aInput, /* Input string */ + int nInput, /* Size of string aInput in bytes */ + sqlite3_tokenizer_cursor **pp /* OUT: New cursor object */ +){ + unicode_cursor *pCsr; + + pCsr = (unicode_cursor *)sqlite3_malloc(sizeof(unicode_cursor)); + if( pCsr==0 ){ + return SQLITE_NOMEM; + } + memset(pCsr, 0, sizeof(unicode_cursor)); + + pCsr->aInput = (const unsigned char *)aInput; + if( aInput==0 ){ + pCsr->nInput = 0; + }else if( nInput<0 ){ + pCsr->nInput = (int)strlen(aInput); + }else{ + pCsr->nInput = nInput; + } + + *pp = &pCsr->base; + UNUSED_PARAMETER(p); + return SQLITE_OK; +} + +/* +** Close a tokenization cursor previously opened by a call to +** simpleOpen() above. +*/ +static int unicodeClose(sqlite3_tokenizer_cursor *pCursor){ + unicode_cursor *pCsr = (unicode_cursor *) pCursor; + sqlite3_free(pCsr->zToken); + sqlite3_free(pCsr); + return SQLITE_OK; +} + +/* +** Extract the next token from a tokenization cursor. The cursor must +** have been opened by a prior call to simpleOpen(). +*/ +static int unicodeNext( + sqlite3_tokenizer_cursor *pC, /* Cursor returned by simpleOpen */ + const char **paToken, /* OUT: Token text */ + int *pnToken, /* OUT: Number of bytes at *paToken */ + int *piStart, /* OUT: Starting offset of token */ + int *piEnd, /* OUT: Ending offset of token */ + int *piPos /* OUT: Position integer of token */ +){ + unicode_cursor *pCsr = (unicode_cursor *)pC; + unicode_tokenizer *p = ((unicode_tokenizer *)pCsr->base.pTokenizer); + unsigned int iCode = 0; + char *zOut; + const unsigned char *z = &pCsr->aInput[pCsr->iOff]; + const unsigned char *zStart = z; + const unsigned char *zEnd; + const unsigned char *zTerm = &pCsr->aInput[pCsr->nInput]; + + /* Scan past any delimiter characters before the start of the next token. + ** Return SQLITE_DONE early if this takes us all the way to the end of + ** the input. */ + while( z=zTerm ) return SQLITE_DONE; + + zOut = pCsr->zToken; + do { + int iOut; + + /* Grow the output buffer if required. */ + if( (zOut-pCsr->zToken)>=(pCsr->nAlloc-4) ){ + char *zNew = sqlite3_realloc64(pCsr->zToken, pCsr->nAlloc+64); + if( !zNew ) return SQLITE_NOMEM; + zOut = &zNew[zOut - pCsr->zToken]; + pCsr->zToken = zNew; + pCsr->nAlloc += 64; + } + + /* Write the folded case of the last character read to the output */ + zEnd = z; + iOut = sqlite3FtsUnicodeFold((int)iCode, p->eRemoveDiacritic); + if( iOut ){ + WRITE_UTF8(zOut, iOut); + } + + /* If the cursor is not at EOF, read the next character */ + if( z>=zTerm ) break; + READ_UTF8(z, zTerm, iCode); + }while( unicodeIsAlnum(p, (int)iCode) + || sqlite3FtsUnicodeIsdiacritic((int)iCode) + ); + + /* Set the output variables and return. */ + pCsr->iOff = (int)(z - pCsr->aInput); + *paToken = pCsr->zToken; + *pnToken = (int)(zOut - pCsr->zToken); + *piStart = (int)(zStart - pCsr->aInput); + *piEnd = (int)(zEnd - pCsr->aInput); + *piPos = pCsr->iToken++; + return SQLITE_OK; +} + +/* +** Set *ppModule to a pointer to the sqlite3_tokenizer_module +** structure for the unicode tokenizer. +*/ +SQLITE_PRIVATE void sqlite3Fts3UnicodeTokenizer(sqlite3_tokenizer_module const **ppModule){ + static const sqlite3_tokenizer_module module = { + 0, + unicodeCreate, + unicodeDestroy, + unicodeOpen, + unicodeClose, + unicodeNext, + 0, + }; + *ppModule = &module; +} + +#endif /* !defined(SQLITE_CORE) || defined(SQLITE_ENABLE_FTS3) */ +#endif /* ifndef SQLITE_DISABLE_FTS3_UNICODE */ + +/************** End of fts3_unicode.c ****************************************/ +/************** Begin file fts3_unicode2.c ***********************************/ +/* +** 2012-05-25 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +****************************************************************************** +*/ + +/* +** DO NOT EDIT THIS MACHINE GENERATED FILE. +*/ + +#ifndef SQLITE_DISABLE_FTS3_UNICODE +#if defined(SQLITE_ENABLE_FTS3) || defined(SQLITE_ENABLE_FTS4) + +/* #include */ + +/* +** Return true if the argument corresponds to a unicode codepoint +** classified as either a letter or a number. Otherwise false. +** +** The results are undefined if the value passed to this function +** is less than zero. +*/ +SQLITE_PRIVATE int sqlite3FtsUnicodeIsalnum(int c){ + /* Each unsigned integer in the following array corresponds to a contiguous + ** range of unicode codepoints that are not either letters or numbers (i.e. + ** codepoints for which this function should return 0). + ** + ** The most significant 22 bits in each 32-bit value contain the first + ** codepoint in the range. The least significant 10 bits are used to store + ** the size of the range (always at least 1). In other words, the value + ** ((C<<22) + N) represents a range of N codepoints starting with codepoint + ** C. It is not possible to represent a range larger than 1023 codepoints + ** using this format. + */ + static const unsigned int aEntry[] = { + 0x00000030, 0x0000E807, 0x00016C06, 0x0001EC2F, 0x0002AC07, + 0x0002D001, 0x0002D803, 0x0002EC01, 0x0002FC01, 0x00035C01, + 0x0003DC01, 0x000B0804, 0x000B480E, 0x000B9407, 0x000BB401, + 0x000BBC81, 0x000DD401, 0x000DF801, 0x000E1002, 0x000E1C01, + 0x000FD801, 0x00120808, 0x00156806, 0x00162402, 0x00163C01, + 0x00164437, 0x0017CC02, 0x00180005, 0x00181816, 0x00187802, + 0x00192C15, 0x0019A804, 0x0019C001, 0x001B5001, 0x001B580F, + 0x001B9C07, 0x001BF402, 0x001C000E, 0x001C3C01, 0x001C4401, + 0x001CC01B, 0x001E980B, 0x001FAC09, 0x001FD804, 0x00205804, + 0x00206C09, 0x00209403, 0x0020A405, 0x0020C00F, 0x00216403, + 0x00217801, 0x0023901B, 0x00240004, 0x0024E803, 0x0024F812, + 0x00254407, 0x00258804, 0x0025C001, 0x00260403, 0x0026F001, + 0x0026F807, 0x00271C02, 0x00272C03, 0x00275C01, 0x00278802, + 0x0027C802, 0x0027E802, 0x00280403, 0x0028F001, 0x0028F805, + 0x00291C02, 0x00292C03, 0x00294401, 0x0029C002, 0x0029D401, + 0x002A0403, 0x002AF001, 0x002AF808, 0x002B1C03, 0x002B2C03, + 0x002B8802, 0x002BC002, 0x002C0403, 0x002CF001, 0x002CF807, + 0x002D1C02, 0x002D2C03, 0x002D5802, 0x002D8802, 0x002DC001, + 0x002E0801, 0x002EF805, 0x002F1803, 0x002F2804, 0x002F5C01, + 0x002FCC08, 0x00300403, 0x0030F807, 0x00311803, 0x00312804, + 0x00315402, 0x00318802, 0x0031FC01, 0x00320802, 0x0032F001, + 0x0032F807, 0x00331803, 0x00332804, 0x00335402, 0x00338802, + 0x00340802, 0x0034F807, 0x00351803, 0x00352804, 0x00355C01, + 0x00358802, 0x0035E401, 0x00360802, 0x00372801, 0x00373C06, + 0x00375801, 0x00376008, 0x0037C803, 0x0038C401, 0x0038D007, + 0x0038FC01, 0x00391C09, 0x00396802, 0x003AC401, 0x003AD006, + 0x003AEC02, 0x003B2006, 0x003C041F, 0x003CD00C, 0x003DC417, + 0x003E340B, 0x003E6424, 0x003EF80F, 0x003F380D, 0x0040AC14, + 0x00412806, 0x00415804, 0x00417803, 0x00418803, 0x00419C07, + 0x0041C404, 0x0042080C, 0x00423C01, 0x00426806, 0x0043EC01, + 0x004D740C, 0x004E400A, 0x00500001, 0x0059B402, 0x005A0001, + 0x005A6C02, 0x005BAC03, 0x005C4803, 0x005CC805, 0x005D4802, + 0x005DC802, 0x005ED023, 0x005F6004, 0x005F7401, 0x0060000F, + 0x0062A401, 0x0064800C, 0x0064C00C, 0x00650001, 0x00651002, + 0x0066C011, 0x00672002, 0x00677822, 0x00685C05, 0x00687802, + 0x0069540A, 0x0069801D, 0x0069FC01, 0x006A8007, 0x006AA006, + 0x006C0005, 0x006CD011, 0x006D6823, 0x006E0003, 0x006E840D, + 0x006F980E, 0x006FF004, 0x00709014, 0x0070EC05, 0x0071F802, + 0x00730008, 0x00734019, 0x0073B401, 0x0073C803, 0x00770027, + 0x0077F004, 0x007EF401, 0x007EFC03, 0x007F3403, 0x007F7403, + 0x007FB403, 0x007FF402, 0x00800065, 0x0081A806, 0x0081E805, + 0x00822805, 0x0082801A, 0x00834021, 0x00840002, 0x00840C04, + 0x00842002, 0x00845001, 0x00845803, 0x00847806, 0x00849401, + 0x00849C01, 0x0084A401, 0x0084B801, 0x0084E802, 0x00850005, + 0x00852804, 0x00853C01, 0x00864264, 0x00900027, 0x0091000B, + 0x0092704E, 0x00940200, 0x009C0475, 0x009E53B9, 0x00AD400A, + 0x00B39406, 0x00B3BC03, 0x00B3E404, 0x00B3F802, 0x00B5C001, + 0x00B5FC01, 0x00B7804F, 0x00B8C00C, 0x00BA001A, 0x00BA6C59, + 0x00BC00D6, 0x00BFC00C, 0x00C00005, 0x00C02019, 0x00C0A807, + 0x00C0D802, 0x00C0F403, 0x00C26404, 0x00C28001, 0x00C3EC01, + 0x00C64002, 0x00C6580A, 0x00C70024, 0x00C8001F, 0x00C8A81E, + 0x00C94001, 0x00C98020, 0x00CA2827, 0x00CB003F, 0x00CC0100, + 0x01370040, 0x02924037, 0x0293F802, 0x02983403, 0x0299BC10, + 0x029A7C01, 0x029BC008, 0x029C0017, 0x029C8002, 0x029E2402, + 0x02A00801, 0x02A01801, 0x02A02C01, 0x02A08C09, 0x02A0D804, + 0x02A1D004, 0x02A20002, 0x02A2D011, 0x02A33802, 0x02A38012, + 0x02A3E003, 0x02A4980A, 0x02A51C0D, 0x02A57C01, 0x02A60004, + 0x02A6CC1B, 0x02A77802, 0x02A8A40E, 0x02A90C01, 0x02A93002, + 0x02A97004, 0x02A9DC03, 0x02A9EC01, 0x02AAC001, 0x02AAC803, + 0x02AADC02, 0x02AAF802, 0x02AB0401, 0x02AB7802, 0x02ABAC07, + 0x02ABD402, 0x02AF8C0B, 0x03600001, 0x036DFC02, 0x036FFC02, + 0x037FFC01, 0x03EC7801, 0x03ECA401, 0x03EEC810, 0x03F4F802, + 0x03F7F002, 0x03F8001A, 0x03F88007, 0x03F8C023, 0x03F95013, + 0x03F9A004, 0x03FBFC01, 0x03FC040F, 0x03FC6807, 0x03FCEC06, + 0x03FD6C0B, 0x03FF8007, 0x03FFA007, 0x03FFE405, 0x04040003, + 0x0404DC09, 0x0405E411, 0x0406400C, 0x0407402E, 0x040E7C01, + 0x040F4001, 0x04215C01, 0x04247C01, 0x0424FC01, 0x04280403, + 0x04281402, 0x04283004, 0x0428E003, 0x0428FC01, 0x04294009, + 0x0429FC01, 0x042CE407, 0x04400003, 0x0440E016, 0x04420003, + 0x0442C012, 0x04440003, 0x04449C0E, 0x04450004, 0x04460003, + 0x0446CC0E, 0x04471404, 0x045AAC0D, 0x0491C004, 0x05BD442E, + 0x05BE3C04, 0x074000F6, 0x07440027, 0x0744A4B5, 0x07480046, + 0x074C0057, 0x075B0401, 0x075B6C01, 0x075BEC01, 0x075C5401, + 0x075CD401, 0x075D3C01, 0x075DBC01, 0x075E2401, 0x075EA401, + 0x075F0C01, 0x07BBC002, 0x07C0002C, 0x07C0C064, 0x07C2800F, + 0x07C2C40E, 0x07C3040F, 0x07C3440F, 0x07C4401F, 0x07C4C03C, + 0x07C5C02B, 0x07C7981D, 0x07C8402B, 0x07C90009, 0x07C94002, + 0x07CC0021, 0x07CCC006, 0x07CCDC46, 0x07CE0014, 0x07CE8025, + 0x07CF1805, 0x07CF8011, 0x07D0003F, 0x07D10001, 0x07D108B6, + 0x07D3E404, 0x07D4003E, 0x07D50004, 0x07D54018, 0x07D7EC46, + 0x07D9140B, 0x07DA0046, 0x07DC0074, 0x38000401, 0x38008060, + 0x380400F0, + }; + static const unsigned int aAscii[4] = { + 0xFFFFFFFF, 0xFC00FFFF, 0xF8000001, 0xF8000001, + }; + + if( (unsigned int)c<128 ){ + return ( (aAscii[c >> 5] & ((unsigned int)1 << (c & 0x001F)))==0 ); + }else if( (unsigned int)c<(1<<22) ){ + unsigned int key = (((unsigned int)c)<<10) | 0x000003FF; + int iRes = 0; + int iHi = sizeof(aEntry)/sizeof(aEntry[0]) - 1; + int iLo = 0; + while( iHi>=iLo ){ + int iTest = (iHi + iLo) / 2; + if( key >= aEntry[iTest] ){ + iRes = iTest; + iLo = iTest+1; + }else{ + iHi = iTest-1; + } + } + assert( aEntry[0]=aEntry[iRes] ); + return (((unsigned int)c) >= ((aEntry[iRes]>>10) + (aEntry[iRes]&0x3FF))); + } + return 1; +} + + +/* +** If the argument is a codepoint corresponding to a lowercase letter +** in the ASCII range with a diacritic added, return the codepoint +** of the ASCII letter only. For example, if passed 235 - "LATIN +** SMALL LETTER E WITH DIAERESIS" - return 65 ("LATIN SMALL LETTER +** E"). The resuls of passing a codepoint that corresponds to an +** uppercase letter are undefined. +*/ +static int remove_diacritic(int c, int bComplex){ + unsigned short aDia[] = { + 0, 1797, 1848, 1859, 1891, 1928, 1940, 1995, + 2024, 2040, 2060, 2110, 2168, 2206, 2264, 2286, + 2344, 2383, 2472, 2488, 2516, 2596, 2668, 2732, + 2782, 2842, 2894, 2954, 2984, 3000, 3028, 3336, + 3456, 3696, 3712, 3728, 3744, 3766, 3832, 3896, + 3912, 3928, 3944, 3968, 4008, 4040, 4056, 4106, + 4138, 4170, 4202, 4234, 4266, 4296, 4312, 4344, + 4408, 4424, 4442, 4472, 4488, 4504, 6148, 6198, + 6264, 6280, 6360, 6429, 6505, 6529, 61448, 61468, + 61512, 61534, 61592, 61610, 61642, 61672, 61688, 61704, + 61726, 61784, 61800, 61816, 61836, 61880, 61896, 61914, + 61948, 61998, 62062, 62122, 62154, 62184, 62200, 62218, + 62252, 62302, 62364, 62410, 62442, 62478, 62536, 62554, + 62584, 62604, 62640, 62648, 62656, 62664, 62730, 62766, + 62830, 62890, 62924, 62974, 63032, 63050, 63082, 63118, + 63182, 63242, 63274, 63310, 63368, 63390, + }; +#define HIBIT ((unsigned char)0x80) + unsigned char aChar[] = { + '\0', 'a', 'c', 'e', 'i', 'n', + 'o', 'u', 'y', 'y', 'a', 'c', + 'd', 'e', 'e', 'g', 'h', 'i', + 'j', 'k', 'l', 'n', 'o', 'r', + 's', 't', 'u', 'u', 'w', 'y', + 'z', 'o', 'u', 'a', 'i', 'o', + 'u', 'u'|HIBIT, 'a'|HIBIT, 'g', 'k', 'o', + 'o'|HIBIT, 'j', 'g', 'n', 'a'|HIBIT, 'a', + 'e', 'i', 'o', 'r', 'u', 's', + 't', 'h', 'a', 'e', 'o'|HIBIT, 'o', + 'o'|HIBIT, 'y', '\0', '\0', '\0', '\0', + '\0', '\0', '\0', '\0', 'a', 'b', + 'c'|HIBIT, 'd', 'd', 'e'|HIBIT, 'e', 'e'|HIBIT, + 'f', 'g', 'h', 'h', 'i', 'i'|HIBIT, + 'k', 'l', 'l'|HIBIT, 'l', 'm', 'n', + 'o'|HIBIT, 'p', 'r', 'r'|HIBIT, 'r', 's', + 's'|HIBIT, 't', 'u', 'u'|HIBIT, 'v', 'w', + 'w', 'x', 'y', 'z', 'h', 't', + 'w', 'y', 'a', 'a'|HIBIT, 'a'|HIBIT, 'a'|HIBIT, + 'e', 'e'|HIBIT, 'e'|HIBIT, 'i', 'o', 'o'|HIBIT, + 'o'|HIBIT, 'o'|HIBIT, 'u', 'u'|HIBIT, 'u'|HIBIT, 'y', + }; + + unsigned int key = (((unsigned int)c)<<3) | 0x00000007; + int iRes = 0; + int iHi = sizeof(aDia)/sizeof(aDia[0]) - 1; + int iLo = 0; + while( iHi>=iLo ){ + int iTest = (iHi + iLo) / 2; + if( key >= aDia[iTest] ){ + iRes = iTest; + iLo = iTest+1; + }else{ + iHi = iTest-1; + } + } + assert( key>=aDia[iRes] ); + if( bComplex==0 && (aChar[iRes] & 0x80) ) return c; + return (c > (aDia[iRes]>>3) + (aDia[iRes]&0x07)) ? c : ((int)aChar[iRes] & 0x7F); +} + + +/* +** Return true if the argument interpreted as a unicode codepoint +** is a diacritical modifier character. +*/ +SQLITE_PRIVATE int sqlite3FtsUnicodeIsdiacritic(int c){ + unsigned int mask0 = 0x08029FDF; + unsigned int mask1 = 0x000361F8; + if( c<768 || c>817 ) return 0; + return (c < 768+32) ? + (mask0 & ((unsigned int)1 << (c-768))) : + (mask1 & ((unsigned int)1 << (c-768-32))); +} + + +/* +** Interpret the argument as a unicode codepoint. If the codepoint +** is an upper case character that has a lower case equivalent, +** return the codepoint corresponding to the lower case version. +** Otherwise, return a copy of the argument. +** +** The results are undefined if the value passed to this function +** is less than zero. +*/ +SQLITE_PRIVATE int sqlite3FtsUnicodeFold(int c, int eRemoveDiacritic){ + /* Each entry in the following array defines a rule for folding a range + ** of codepoints to lower case. The rule applies to a range of nRange + ** codepoints starting at codepoint iCode. + ** + ** If the least significant bit in flags is clear, then the rule applies + ** to all nRange codepoints (i.e. all nRange codepoints are upper case and + ** need to be folded). Or, if it is set, then the rule only applies to + ** every second codepoint in the range, starting with codepoint C. + ** + ** The 7 most significant bits in flags are an index into the aiOff[] + ** array. If a specific codepoint C does require folding, then its lower + ** case equivalent is ((C + aiOff[flags>>1]) & 0xFFFF). + ** + ** The contents of this array are generated by parsing the CaseFolding.txt + ** file distributed as part of the "Unicode Character Database". See + ** http://www.unicode.org for details. + */ + static const struct TableEntry { + unsigned short iCode; + unsigned char flags; + unsigned char nRange; + } aEntry[] = { + {65, 14, 26}, {181, 64, 1}, {192, 14, 23}, + {216, 14, 7}, {256, 1, 48}, {306, 1, 6}, + {313, 1, 16}, {330, 1, 46}, {376, 116, 1}, + {377, 1, 6}, {383, 104, 1}, {385, 50, 1}, + {386, 1, 4}, {390, 44, 1}, {391, 0, 1}, + {393, 42, 2}, {395, 0, 1}, {398, 32, 1}, + {399, 38, 1}, {400, 40, 1}, {401, 0, 1}, + {403, 42, 1}, {404, 46, 1}, {406, 52, 1}, + {407, 48, 1}, {408, 0, 1}, {412, 52, 1}, + {413, 54, 1}, {415, 56, 1}, {416, 1, 6}, + {422, 60, 1}, {423, 0, 1}, {425, 60, 1}, + {428, 0, 1}, {430, 60, 1}, {431, 0, 1}, + {433, 58, 2}, {435, 1, 4}, {439, 62, 1}, + {440, 0, 1}, {444, 0, 1}, {452, 2, 1}, + {453, 0, 1}, {455, 2, 1}, {456, 0, 1}, + {458, 2, 1}, {459, 1, 18}, {478, 1, 18}, + {497, 2, 1}, {498, 1, 4}, {502, 122, 1}, + {503, 134, 1}, {504, 1, 40}, {544, 110, 1}, + {546, 1, 18}, {570, 70, 1}, {571, 0, 1}, + {573, 108, 1}, {574, 68, 1}, {577, 0, 1}, + {579, 106, 1}, {580, 28, 1}, {581, 30, 1}, + {582, 1, 10}, {837, 36, 1}, {880, 1, 4}, + {886, 0, 1}, {902, 18, 1}, {904, 16, 3}, + {908, 26, 1}, {910, 24, 2}, {913, 14, 17}, + {931, 14, 9}, {962, 0, 1}, {975, 4, 1}, + {976, 140, 1}, {977, 142, 1}, {981, 146, 1}, + {982, 144, 1}, {984, 1, 24}, {1008, 136, 1}, + {1009, 138, 1}, {1012, 130, 1}, {1013, 128, 1}, + {1015, 0, 1}, {1017, 152, 1}, {1018, 0, 1}, + {1021, 110, 3}, {1024, 34, 16}, {1040, 14, 32}, + {1120, 1, 34}, {1162, 1, 54}, {1216, 6, 1}, + {1217, 1, 14}, {1232, 1, 88}, {1329, 22, 38}, + {4256, 66, 38}, {4295, 66, 1}, {4301, 66, 1}, + {7680, 1, 150}, {7835, 132, 1}, {7838, 96, 1}, + {7840, 1, 96}, {7944, 150, 8}, {7960, 150, 6}, + {7976, 150, 8}, {7992, 150, 8}, {8008, 150, 6}, + {8025, 151, 8}, {8040, 150, 8}, {8072, 150, 8}, + {8088, 150, 8}, {8104, 150, 8}, {8120, 150, 2}, + {8122, 126, 2}, {8124, 148, 1}, {8126, 100, 1}, + {8136, 124, 4}, {8140, 148, 1}, {8152, 150, 2}, + {8154, 120, 2}, {8168, 150, 2}, {8170, 118, 2}, + {8172, 152, 1}, {8184, 112, 2}, {8186, 114, 2}, + {8188, 148, 1}, {8486, 98, 1}, {8490, 92, 1}, + {8491, 94, 1}, {8498, 12, 1}, {8544, 8, 16}, + {8579, 0, 1}, {9398, 10, 26}, {11264, 22, 47}, + {11360, 0, 1}, {11362, 88, 1}, {11363, 102, 1}, + {11364, 90, 1}, {11367, 1, 6}, {11373, 84, 1}, + {11374, 86, 1}, {11375, 80, 1}, {11376, 82, 1}, + {11378, 0, 1}, {11381, 0, 1}, {11390, 78, 2}, + {11392, 1, 100}, {11499, 1, 4}, {11506, 0, 1}, + {42560, 1, 46}, {42624, 1, 24}, {42786, 1, 14}, + {42802, 1, 62}, {42873, 1, 4}, {42877, 76, 1}, + {42878, 1, 10}, {42891, 0, 1}, {42893, 74, 1}, + {42896, 1, 4}, {42912, 1, 10}, {42922, 72, 1}, + {65313, 14, 26}, + }; + static const unsigned short aiOff[] = { + 1, 2, 8, 15, 16, 26, 28, 32, + 37, 38, 40, 48, 63, 64, 69, 71, + 79, 80, 116, 202, 203, 205, 206, 207, + 209, 210, 211, 213, 214, 217, 218, 219, + 775, 7264, 10792, 10795, 23228, 23256, 30204, 54721, + 54753, 54754, 54756, 54787, 54793, 54809, 57153, 57274, + 57921, 58019, 58363, 61722, 65268, 65341, 65373, 65406, + 65408, 65410, 65415, 65424, 65436, 65439, 65450, 65462, + 65472, 65476, 65478, 65480, 65482, 65488, 65506, 65511, + 65514, 65521, 65527, 65528, 65529, + }; + + int ret = c; + + assert( sizeof(unsigned short)==2 && sizeof(unsigned char)==1 ); + + if( c<128 ){ + if( c>='A' && c<='Z' ) ret = c + ('a' - 'A'); + }else if( c<65536 ){ + const struct TableEntry *p; + int iHi = sizeof(aEntry)/sizeof(aEntry[0]) - 1; + int iLo = 0; + int iRes = -1; + + assert( c>aEntry[0].iCode ); + while( iHi>=iLo ){ + int iTest = (iHi + iLo) / 2; + int cmp = (c - aEntry[iTest].iCode); + if( cmp>=0 ){ + iRes = iTest; + iLo = iTest+1; + }else{ + iHi = iTest-1; + } + } + + assert( iRes>=0 && c>=aEntry[iRes].iCode ); + p = &aEntry[iRes]; + if( c<(p->iCode + p->nRange) && 0==(0x01 & p->flags & (p->iCode ^ c)) ){ + ret = (c + (aiOff[p->flags>>1])) & 0x0000FFFF; + assert( ret>0 ); + } + + if( eRemoveDiacritic ){ + ret = remove_diacritic(ret, eRemoveDiacritic==2); + } + } + + else if( c>=66560 && c<66600 ){ + ret = c + 40; + } + + return ret; +} +#endif /* defined(SQLITE_ENABLE_FTS3) || defined(SQLITE_ENABLE_FTS4) */ +#endif /* !defined(SQLITE_DISABLE_FTS3_UNICODE) */ + +/************** End of fts3_unicode2.c ***************************************/ +/************** Begin file json1.c *******************************************/ +/* +** 2015-08-12 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +****************************************************************************** +** +** This SQLite extension implements JSON functions. The interface is +** modeled after MySQL JSON functions: +** +** https://dev.mysql.com/doc/refman/5.7/en/json.html +** +** For the time being, all JSON is stored as pure text. (We might add +** a JSONB type in the future which stores a binary encoding of JSON in +** a BLOB, but there is no support for JSONB in the current implementation. +** This implementation parses JSON text at 250 MB/s, so it is hard to see +** how JSONB might improve on that.) +*/ +#if !defined(SQLITE_CORE) || defined(SQLITE_ENABLE_JSON1) +#if !defined(SQLITEINT_H) +/* #include "sqlite3ext.h" */ +#endif +SQLITE_EXTENSION_INIT1 +/* #include */ +/* #include */ +/* #include */ +/* #include */ + +/* Mark a function parameter as unused, to suppress nuisance compiler +** warnings. */ +#ifndef UNUSED_PARAM +# define UNUSED_PARAM(X) (void)(X) +#endif + +#ifndef LARGEST_INT64 +# define LARGEST_INT64 (0xffffffff|(((sqlite3_int64)0x7fffffff)<<32)) +# define SMALLEST_INT64 (((sqlite3_int64)-1) - LARGEST_INT64) +#endif + +/* +** Versions of isspace(), isalnum() and isdigit() to which it is safe +** to pass signed char values. +*/ +#ifdef sqlite3Isdigit + /* Use the SQLite core versions if this routine is part of the + ** SQLite amalgamation */ +# define safe_isdigit(x) sqlite3Isdigit(x) +# define safe_isalnum(x) sqlite3Isalnum(x) +# define safe_isxdigit(x) sqlite3Isxdigit(x) +#else + /* Use the standard library for separate compilation */ +#include /* amalgamator: keep */ +# define safe_isdigit(x) isdigit((unsigned char)(x)) +# define safe_isalnum(x) isalnum((unsigned char)(x)) +# define safe_isxdigit(x) isxdigit((unsigned char)(x)) +#endif + +/* +** Growing our own isspace() routine this way is twice as fast as +** the library isspace() function, resulting in a 7% overall performance +** increase for the parser. (Ubuntu14.10 gcc 4.8.4 x64 with -Os). +*/ +static const char jsonIsSpace[] = { + 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 0, 0, 1, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, +}; +#define safe_isspace(x) (jsonIsSpace[(unsigned char)x]) + +#ifndef SQLITE_AMALGAMATION + /* Unsigned integer types. These are already defined in the sqliteInt.h, + ** but the definitions need to be repeated for separate compilation. */ + typedef sqlite3_uint64 u64; + typedef unsigned int u32; + typedef unsigned short int u16; + typedef unsigned char u8; +#endif + +/* Objects */ +typedef struct JsonString JsonString; +typedef struct JsonNode JsonNode; +typedef struct JsonParse JsonParse; + +/* An instance of this object represents a JSON string +** under construction. Really, this is a generic string accumulator +** that can be and is used to create strings other than JSON. +*/ +struct JsonString { + sqlite3_context *pCtx; /* Function context - put error messages here */ + char *zBuf; /* Append JSON content here */ + u64 nAlloc; /* Bytes of storage available in zBuf[] */ + u64 nUsed; /* Bytes of zBuf[] currently used */ + u8 bStatic; /* True if zBuf is static space */ + u8 bErr; /* True if an error has been encountered */ + char zSpace[100]; /* Initial static space */ +}; + +/* JSON type values +*/ +#define JSON_NULL 0 +#define JSON_TRUE 1 +#define JSON_FALSE 2 +#define JSON_INT 3 +#define JSON_REAL 4 +#define JSON_STRING 5 +#define JSON_ARRAY 6 +#define JSON_OBJECT 7 + +/* The "subtype" set for JSON values */ +#define JSON_SUBTYPE 74 /* Ascii for "J" */ + +/* +** Names of the various JSON types: +*/ +static const char * const jsonType[] = { + "null", "true", "false", "integer", "real", "text", "array", "object" +}; + +/* Bit values for the JsonNode.jnFlag field +*/ +#define JNODE_RAW 0x01 /* Content is raw, not JSON encoded */ +#define JNODE_ESCAPE 0x02 /* Content is text with \ escapes */ +#define JNODE_REMOVE 0x04 /* Do not output */ +#define JNODE_REPLACE 0x08 /* Replace with JsonNode.u.iReplace */ +#define JNODE_PATCH 0x10 /* Patch with JsonNode.u.pPatch */ +#define JNODE_APPEND 0x20 /* More ARRAY/OBJECT entries at u.iAppend */ +#define JNODE_LABEL 0x40 /* Is a label of an object */ + + +/* A single node of parsed JSON +*/ +struct JsonNode { + u8 eType; /* One of the JSON_ type values */ + u8 jnFlags; /* JNODE flags */ + u32 n; /* Bytes of content, or number of sub-nodes */ + union { + const char *zJContent; /* Content for INT, REAL, and STRING */ + u32 iAppend; /* More terms for ARRAY and OBJECT */ + u32 iKey; /* Key for ARRAY objects in json_tree() */ + u32 iReplace; /* Replacement content for JNODE_REPLACE */ + JsonNode *pPatch; /* Node chain of patch for JNODE_PATCH */ + } u; +}; + +/* A completely parsed JSON string +*/ +struct JsonParse { + u32 nNode; /* Number of slots of aNode[] used */ + u32 nAlloc; /* Number of slots of aNode[] allocated */ + JsonNode *aNode; /* Array of nodes containing the parse */ + const char *zJson; /* Original JSON string */ + u32 *aUp; /* Index of parent of each node */ + u8 oom; /* Set to true if out of memory */ + u8 nErr; /* Number of errors seen */ + u16 iDepth; /* Nesting depth */ + int nJson; /* Length of the zJson string in bytes */ + u32 iHold; /* Replace cache line with the lowest iHold value */ +}; + +/* +** Maximum nesting depth of JSON for this implementation. +** +** This limit is needed to avoid a stack overflow in the recursive +** descent parser. A depth of 2000 is far deeper than any sane JSON +** should go. +*/ +#define JSON_MAX_DEPTH 2000 + +/************************************************************************** +** Utility routines for dealing with JsonString objects +**************************************************************************/ + +/* Set the JsonString object to an empty string +*/ +static void jsonZero(JsonString *p){ + p->zBuf = p->zSpace; + p->nAlloc = sizeof(p->zSpace); + p->nUsed = 0; + p->bStatic = 1; +} + +/* Initialize the JsonString object +*/ +static void jsonInit(JsonString *p, sqlite3_context *pCtx){ + p->pCtx = pCtx; + p->bErr = 0; + jsonZero(p); +} + + +/* Free all allocated memory and reset the JsonString object back to its +** initial state. +*/ +static void jsonReset(JsonString *p){ + if( !p->bStatic ) sqlite3_free(p->zBuf); + jsonZero(p); +} + + +/* Report an out-of-memory (OOM) condition +*/ +static void jsonOom(JsonString *p){ + p->bErr = 1; + sqlite3_result_error_nomem(p->pCtx); + jsonReset(p); +} + +/* Enlarge pJson->zBuf so that it can hold at least N more bytes. +** Return zero on success. Return non-zero on an OOM error +*/ +static int jsonGrow(JsonString *p, u32 N){ + u64 nTotal = NnAlloc ? p->nAlloc*2 : p->nAlloc+N+10; + char *zNew; + if( p->bStatic ){ + if( p->bErr ) return 1; + zNew = sqlite3_malloc64(nTotal); + if( zNew==0 ){ + jsonOom(p); + return SQLITE_NOMEM; + } + memcpy(zNew, p->zBuf, (size_t)p->nUsed); + p->zBuf = zNew; + p->bStatic = 0; + }else{ + zNew = sqlite3_realloc64(p->zBuf, nTotal); + if( zNew==0 ){ + jsonOom(p); + return SQLITE_NOMEM; + } + p->zBuf = zNew; + } + p->nAlloc = nTotal; + return SQLITE_OK; +} + +/* Append N bytes from zIn onto the end of the JsonString string. +*/ +static void jsonAppendRaw(JsonString *p, const char *zIn, u32 N){ + if( (N+p->nUsed >= p->nAlloc) && jsonGrow(p,N)!=0 ) return; + memcpy(p->zBuf+p->nUsed, zIn, N); + p->nUsed += N; +} + +/* Append formatted text (not to exceed N bytes) to the JsonString. +*/ +static void jsonPrintf(int N, JsonString *p, const char *zFormat, ...){ + va_list ap; + if( (p->nUsed + N >= p->nAlloc) && jsonGrow(p, N) ) return; + va_start(ap, zFormat); + sqlite3_vsnprintf(N, p->zBuf+p->nUsed, zFormat, ap); + va_end(ap); + p->nUsed += (int)strlen(p->zBuf+p->nUsed); +} + +/* Append a single character +*/ +static void jsonAppendChar(JsonString *p, char c){ + if( p->nUsed>=p->nAlloc && jsonGrow(p,1)!=0 ) return; + p->zBuf[p->nUsed++] = c; +} + +/* Append a comma separator to the output buffer, if the previous +** character is not '[' or '{'. +*/ +static void jsonAppendSeparator(JsonString *p){ + char c; + if( p->nUsed==0 ) return; + c = p->zBuf[p->nUsed-1]; + if( c!='[' && c!='{' ) jsonAppendChar(p, ','); +} + +/* Append the N-byte string in zIn to the end of the JsonString string +** under construction. Enclose the string in "..." and escape +** any double-quotes or backslash characters contained within the +** string. +*/ +static void jsonAppendString(JsonString *p, const char *zIn, u32 N){ + u32 i; + if( (N+p->nUsed+2 >= p->nAlloc) && jsonGrow(p,N+2)!=0 ) return; + p->zBuf[p->nUsed++] = '"'; + for(i=0; inUsed+N+3-i > p->nAlloc) && jsonGrow(p,N+3-i)!=0 ) return; + p->zBuf[p->nUsed++] = '\\'; + }else if( c<=0x1f ){ + static const char aSpecial[] = { + 0, 0, 0, 0, 0, 0, 0, 0, 'b', 't', 'n', 0, 'f', 'r', 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 + }; + assert( sizeof(aSpecial)==32 ); + assert( aSpecial['\b']=='b' ); + assert( aSpecial['\f']=='f' ); + assert( aSpecial['\n']=='n' ); + assert( aSpecial['\r']=='r' ); + assert( aSpecial['\t']=='t' ); + if( aSpecial[c] ){ + c = aSpecial[c]; + goto json_simple_escape; + } + if( (p->nUsed+N+7+i > p->nAlloc) && jsonGrow(p,N+7-i)!=0 ) return; + p->zBuf[p->nUsed++] = '\\'; + p->zBuf[p->nUsed++] = 'u'; + p->zBuf[p->nUsed++] = '0'; + p->zBuf[p->nUsed++] = '0'; + p->zBuf[p->nUsed++] = '0' + (c>>4); + c = "0123456789abcdef"[c&0xf]; + } + p->zBuf[p->nUsed++] = c; + } + p->zBuf[p->nUsed++] = '"'; + assert( p->nUsednAlloc ); +} + +/* +** Append a function parameter value to the JSON string under +** construction. +*/ +static void jsonAppendValue( + JsonString *p, /* Append to this JSON string */ + sqlite3_value *pValue /* Value to append */ +){ + switch( sqlite3_value_type(pValue) ){ + case SQLITE_NULL: { + jsonAppendRaw(p, "null", 4); + break; + } + case SQLITE_INTEGER: + case SQLITE_FLOAT: { + const char *z = (const char*)sqlite3_value_text(pValue); + u32 n = (u32)sqlite3_value_bytes(pValue); + jsonAppendRaw(p, z, n); + break; + } + case SQLITE_TEXT: { + const char *z = (const char*)sqlite3_value_text(pValue); + u32 n = (u32)sqlite3_value_bytes(pValue); + if( sqlite3_value_subtype(pValue)==JSON_SUBTYPE ){ + jsonAppendRaw(p, z, n); + }else{ + jsonAppendString(p, z, n); + } + break; + } + default: { + if( p->bErr==0 ){ + sqlite3_result_error(p->pCtx, "JSON cannot hold BLOB values", -1); + p->bErr = 2; + jsonReset(p); + } + break; + } + } +} + + +/* Make the JSON in p the result of the SQL function. +*/ +static void jsonResult(JsonString *p){ + if( p->bErr==0 ){ + sqlite3_result_text64(p->pCtx, p->zBuf, p->nUsed, + p->bStatic ? SQLITE_TRANSIENT : sqlite3_free, + SQLITE_UTF8); + jsonZero(p); + } + assert( p->bStatic ); +} + +/************************************************************************** +** Utility routines for dealing with JsonNode and JsonParse objects +**************************************************************************/ + +/* +** Return the number of consecutive JsonNode slots need to represent +** the parsed JSON at pNode. The minimum answer is 1. For ARRAY and +** OBJECT types, the number might be larger. +** +** Appended elements are not counted. The value returned is the number +** by which the JsonNode counter should increment in order to go to the +** next peer value. +*/ +static u32 jsonNodeSize(JsonNode *pNode){ + return pNode->eType>=JSON_ARRAY ? pNode->n+1 : 1; +} + +/* +** Reclaim all memory allocated by a JsonParse object. But do not +** delete the JsonParse object itself. +*/ +static void jsonParseReset(JsonParse *pParse){ + sqlite3_free(pParse->aNode); + pParse->aNode = 0; + pParse->nNode = 0; + pParse->nAlloc = 0; + sqlite3_free(pParse->aUp); + pParse->aUp = 0; +} + +/* +** Free a JsonParse object that was obtained from sqlite3_malloc(). +*/ +static void jsonParseFree(JsonParse *pParse){ + jsonParseReset(pParse); + sqlite3_free(pParse); +} + +/* +** Convert the JsonNode pNode into a pure JSON string and +** append to pOut. Subsubstructure is also included. Return +** the number of JsonNode objects that are encoded. +*/ +static void jsonRenderNode( + JsonNode *pNode, /* The node to render */ + JsonString *pOut, /* Write JSON here */ + sqlite3_value **aReplace /* Replacement values */ +){ + if( pNode->jnFlags & (JNODE_REPLACE|JNODE_PATCH) ){ + if( pNode->jnFlags & JNODE_REPLACE ){ + jsonAppendValue(pOut, aReplace[pNode->u.iReplace]); + return; + } + pNode = pNode->u.pPatch; + } + switch( pNode->eType ){ + default: { + assert( pNode->eType==JSON_NULL ); + jsonAppendRaw(pOut, "null", 4); + break; + } + case JSON_TRUE: { + jsonAppendRaw(pOut, "true", 4); + break; + } + case JSON_FALSE: { + jsonAppendRaw(pOut, "false", 5); + break; + } + case JSON_STRING: { + if( pNode->jnFlags & JNODE_RAW ){ + jsonAppendString(pOut, pNode->u.zJContent, pNode->n); + break; + } + /* Fall through into the next case */ + } + case JSON_REAL: + case JSON_INT: { + jsonAppendRaw(pOut, pNode->u.zJContent, pNode->n); + break; + } + case JSON_ARRAY: { + u32 j = 1; + jsonAppendChar(pOut, '['); + for(;;){ + while( j<=pNode->n ){ + if( (pNode[j].jnFlags & JNODE_REMOVE)==0 ){ + jsonAppendSeparator(pOut); + jsonRenderNode(&pNode[j], pOut, aReplace); + } + j += jsonNodeSize(&pNode[j]); + } + if( (pNode->jnFlags & JNODE_APPEND)==0 ) break; + pNode = &pNode[pNode->u.iAppend]; + j = 1; + } + jsonAppendChar(pOut, ']'); + break; + } + case JSON_OBJECT: { + u32 j = 1; + jsonAppendChar(pOut, '{'); + for(;;){ + while( j<=pNode->n ){ + if( (pNode[j+1].jnFlags & JNODE_REMOVE)==0 ){ + jsonAppendSeparator(pOut); + jsonRenderNode(&pNode[j], pOut, aReplace); + jsonAppendChar(pOut, ':'); + jsonRenderNode(&pNode[j+1], pOut, aReplace); + } + j += 1 + jsonNodeSize(&pNode[j+1]); + } + if( (pNode->jnFlags & JNODE_APPEND)==0 ) break; + pNode = &pNode[pNode->u.iAppend]; + j = 1; + } + jsonAppendChar(pOut, '}'); + break; + } + } +} + +/* +** Return a JsonNode and all its descendents as a JSON string. +*/ +static void jsonReturnJson( + JsonNode *pNode, /* Node to return */ + sqlite3_context *pCtx, /* Return value for this function */ + sqlite3_value **aReplace /* Array of replacement values */ +){ + JsonString s; + jsonInit(&s, pCtx); + jsonRenderNode(pNode, &s, aReplace); + jsonResult(&s); + sqlite3_result_subtype(pCtx, JSON_SUBTYPE); +} + +/* +** Make the JsonNode the return value of the function. +*/ +static void jsonReturn( + JsonNode *pNode, /* Node to return */ + sqlite3_context *pCtx, /* Return value for this function */ + sqlite3_value **aReplace /* Array of replacement values */ +){ + switch( pNode->eType ){ + default: { + assert( pNode->eType==JSON_NULL ); + sqlite3_result_null(pCtx); + break; + } + case JSON_TRUE: { + sqlite3_result_int(pCtx, 1); + break; + } + case JSON_FALSE: { + sqlite3_result_int(pCtx, 0); + break; + } + case JSON_INT: { + sqlite3_int64 i = 0; + const char *z = pNode->u.zJContent; + if( z[0]=='-' ){ z++; } + while( z[0]>='0' && z[0]<='9' ){ + unsigned v = *(z++) - '0'; + if( i>=LARGEST_INT64/10 ){ + if( i>LARGEST_INT64/10 ) goto int_as_real; + if( z[0]>='0' && z[0]<='9' ) goto int_as_real; + if( v==9 ) goto int_as_real; + if( v==8 ){ + if( pNode->u.zJContent[0]=='-' ){ + sqlite3_result_int64(pCtx, SMALLEST_INT64); + goto int_done; + }else{ + goto int_as_real; + } + } + } + i = i*10 + v; + } + if( pNode->u.zJContent[0]=='-' ){ i = -i; } + sqlite3_result_int64(pCtx, i); + int_done: + break; + int_as_real: /* fall through to real */; + } + case JSON_REAL: { + double r; +#ifdef SQLITE_AMALGAMATION + const char *z = pNode->u.zJContent; + sqlite3AtoF(z, &r, sqlite3Strlen30(z), SQLITE_UTF8); +#else + r = strtod(pNode->u.zJContent, 0); +#endif + sqlite3_result_double(pCtx, r); + break; + } + case JSON_STRING: { +#if 0 /* Never happens because JNODE_RAW is only set by json_set(), + ** json_insert() and json_replace() and those routines do not + ** call jsonReturn() */ + if( pNode->jnFlags & JNODE_RAW ){ + sqlite3_result_text(pCtx, pNode->u.zJContent, pNode->n, + SQLITE_TRANSIENT); + }else +#endif + assert( (pNode->jnFlags & JNODE_RAW)==0 ); + if( (pNode->jnFlags & JNODE_ESCAPE)==0 ){ + /* JSON formatted without any backslash-escapes */ + sqlite3_result_text(pCtx, pNode->u.zJContent+1, pNode->n-2, + SQLITE_TRANSIENT); + }else{ + /* Translate JSON formatted string into raw text */ + u32 i; + u32 n = pNode->n; + const char *z = pNode->u.zJContent; + char *zOut; + u32 j; + zOut = sqlite3_malloc( n+1 ); + if( zOut==0 ){ + sqlite3_result_error_nomem(pCtx); + break; + } + for(i=1, j=0; i>6)); + zOut[j++] = 0x80 | (v&0x3f); + }else{ + zOut[j++] = (char)(0xe0 | (v>>12)); + zOut[j++] = 0x80 | ((v>>6)&0x3f); + zOut[j++] = 0x80 | (v&0x3f); + } + }else{ + if( c=='b' ){ + c = '\b'; + }else if( c=='f' ){ + c = '\f'; + }else if( c=='n' ){ + c = '\n'; + }else if( c=='r' ){ + c = '\r'; + }else if( c=='t' ){ + c = '\t'; + } + zOut[j++] = c; + } + } + } + zOut[j] = 0; + sqlite3_result_text(pCtx, zOut, j, sqlite3_free); + } + break; + } + case JSON_ARRAY: + case JSON_OBJECT: { + jsonReturnJson(pNode, pCtx, aReplace); + break; + } + } +} + +/* Forward reference */ +static int jsonParseAddNode(JsonParse*,u32,u32,const char*); + +/* +** A macro to hint to the compiler that a function should not be +** inlined. +*/ +#if defined(__GNUC__) +# define JSON_NOINLINE __attribute__((noinline)) +#elif defined(_MSC_VER) && _MSC_VER>=1310 +# define JSON_NOINLINE __declspec(noinline) +#else +# define JSON_NOINLINE +#endif + + +static JSON_NOINLINE int jsonParseAddNodeExpand( + JsonParse *pParse, /* Append the node to this object */ + u32 eType, /* Node type */ + u32 n, /* Content size or sub-node count */ + const char *zContent /* Content */ +){ + u32 nNew; + JsonNode *pNew; + assert( pParse->nNode>=pParse->nAlloc ); + if( pParse->oom ) return -1; + nNew = pParse->nAlloc*2 + 10; + pNew = sqlite3_realloc64(pParse->aNode, sizeof(JsonNode)*nNew); + if( pNew==0 ){ + pParse->oom = 1; + return -1; + } + pParse->nAlloc = nNew; + pParse->aNode = pNew; + assert( pParse->nNodenAlloc ); + return jsonParseAddNode(pParse, eType, n, zContent); +} + +/* +** Create a new JsonNode instance based on the arguments and append that +** instance to the JsonParse. Return the index in pParse->aNode[] of the +** new node, or -1 if a memory allocation fails. +*/ +static int jsonParseAddNode( + JsonParse *pParse, /* Append the node to this object */ + u32 eType, /* Node type */ + u32 n, /* Content size or sub-node count */ + const char *zContent /* Content */ +){ + JsonNode *p; + if( pParse->nNode>=pParse->nAlloc ){ + return jsonParseAddNodeExpand(pParse, eType, n, zContent); + } + p = &pParse->aNode[pParse->nNode]; + p->eType = (u8)eType; + p->jnFlags = 0; + p->n = n; + p->u.zJContent = zContent; + return pParse->nNode++; +} + +/* +** Return true if z[] begins with 4 (or more) hexadecimal digits +*/ +static int jsonIs4Hex(const char *z){ + int i; + for(i=0; i<4; i++) if( !safe_isxdigit(z[i]) ) return 0; + return 1; +} + +/* +** Parse a single JSON value which begins at pParse->zJson[i]. Return the +** index of the first character past the end of the value parsed. +** +** Return negative for a syntax error. Special cases: return -2 if the +** first non-whitespace character is '}' and return -3 if the first +** non-whitespace character is ']'. +*/ +static int jsonParseValue(JsonParse *pParse, u32 i){ + char c; + u32 j; + int iThis; + int x; + JsonNode *pNode; + const char *z = pParse->zJson; + while( safe_isspace(z[i]) ){ i++; } + if( (c = z[i])=='{' ){ + /* Parse object */ + iThis = jsonParseAddNode(pParse, JSON_OBJECT, 0, 0); + if( iThis<0 ) return -1; + for(j=i+1;;j++){ + while( safe_isspace(z[j]) ){ j++; } + if( ++pParse->iDepth > JSON_MAX_DEPTH ) return -1; + x = jsonParseValue(pParse, j); + if( x<0 ){ + pParse->iDepth--; + if( x==(-2) && pParse->nNode==(u32)iThis+1 ) return j+1; + return -1; + } + if( pParse->oom ) return -1; + pNode = &pParse->aNode[pParse->nNode-1]; + if( pNode->eType!=JSON_STRING ) return -1; + pNode->jnFlags |= JNODE_LABEL; + j = x; + while( safe_isspace(z[j]) ){ j++; } + if( z[j]!=':' ) return -1; + j++; + x = jsonParseValue(pParse, j); + pParse->iDepth--; + if( x<0 ) return -1; + j = x; + while( safe_isspace(z[j]) ){ j++; } + c = z[j]; + if( c==',' ) continue; + if( c!='}' ) return -1; + break; + } + pParse->aNode[iThis].n = pParse->nNode - (u32)iThis - 1; + return j+1; + }else if( c=='[' ){ + /* Parse array */ + iThis = jsonParseAddNode(pParse, JSON_ARRAY, 0, 0); + if( iThis<0 ) return -1; + for(j=i+1;;j++){ + while( safe_isspace(z[j]) ){ j++; } + if( ++pParse->iDepth > JSON_MAX_DEPTH ) return -1; + x = jsonParseValue(pParse, j); + pParse->iDepth--; + if( x<0 ){ + if( x==(-3) && pParse->nNode==(u32)iThis+1 ) return j+1; + return -1; + } + j = x; + while( safe_isspace(z[j]) ){ j++; } + c = z[j]; + if( c==',' ) continue; + if( c!=']' ) return -1; + break; + } + pParse->aNode[iThis].n = pParse->nNode - (u32)iThis - 1; + return j+1; + }else if( c=='"' ){ + /* Parse string */ + u8 jnFlags = 0; + j = i+1; + for(;;){ + c = z[j]; + if( (c & ~0x1f)==0 ){ + /* Control characters are not allowed in strings */ + return -1; + } + if( c=='\\' ){ + c = z[++j]; + if( c=='"' || c=='\\' || c=='/' || c=='b' || c=='f' + || c=='n' || c=='r' || c=='t' + || (c=='u' && jsonIs4Hex(z+j+1)) ){ + jnFlags = JNODE_ESCAPE; + }else{ + return -1; + } + }else if( c=='"' ){ + break; + } + j++; + } + jsonParseAddNode(pParse, JSON_STRING, j+1-i, &z[i]); + if( !pParse->oom ) pParse->aNode[pParse->nNode-1].jnFlags = jnFlags; + return j+1; + }else if( c=='n' + && strncmp(z+i,"null",4)==0 + && !safe_isalnum(z[i+4]) ){ + jsonParseAddNode(pParse, JSON_NULL, 0, 0); + return i+4; + }else if( c=='t' + && strncmp(z+i,"true",4)==0 + && !safe_isalnum(z[i+4]) ){ + jsonParseAddNode(pParse, JSON_TRUE, 0, 0); + return i+4; + }else if( c=='f' + && strncmp(z+i,"false",5)==0 + && !safe_isalnum(z[i+5]) ){ + jsonParseAddNode(pParse, JSON_FALSE, 0, 0); + return i+5; + }else if( c=='-' || (c>='0' && c<='9') ){ + /* Parse number */ + u8 seenDP = 0; + u8 seenE = 0; + assert( '-' < '0' ); + if( c<='0' ){ + j = c=='-' ? i+1 : i; + if( z[j]=='0' && z[j+1]>='0' && z[j+1]<='9' ) return -1; + } + j = i+1; + for(;; j++){ + c = z[j]; + if( c>='0' && c<='9' ) continue; + if( c=='.' ){ + if( z[j-1]=='-' ) return -1; + if( seenDP ) return -1; + seenDP = 1; + continue; + } + if( c=='e' || c=='E' ){ + if( z[j-1]<'0' ) return -1; + if( seenE ) return -1; + seenDP = seenE = 1; + c = z[j+1]; + if( c=='+' || c=='-' ){ + j++; + c = z[j+1]; + } + if( c<'0' || c>'9' ) return -1; + continue; + } + break; + } + if( z[j-1]<'0' ) return -1; + jsonParseAddNode(pParse, seenDP ? JSON_REAL : JSON_INT, + j - i, &z[i]); + return j; + }else if( c=='}' ){ + return -2; /* End of {...} */ + }else if( c==']' ){ + return -3; /* End of [...] */ + }else if( c==0 ){ + return 0; /* End of file */ + }else{ + return -1; /* Syntax error */ + } +} + +/* +** Parse a complete JSON string. Return 0 on success or non-zero if there +** are any errors. If an error occurs, free all memory associated with +** pParse. +** +** pParse is uninitialized when this routine is called. +*/ +static int jsonParse( + JsonParse *pParse, /* Initialize and fill this JsonParse object */ + sqlite3_context *pCtx, /* Report errors here */ + const char *zJson /* Input JSON text to be parsed */ +){ + int i; + memset(pParse, 0, sizeof(*pParse)); + if( zJson==0 ) return 1; + pParse->zJson = zJson; + i = jsonParseValue(pParse, 0); + if( pParse->oom ) i = -1; + if( i>0 ){ + assert( pParse->iDepth==0 ); + while( safe_isspace(zJson[i]) ) i++; + if( zJson[i] ) i = -1; + } + if( i<=0 ){ + if( pCtx!=0 ){ + if( pParse->oom ){ + sqlite3_result_error_nomem(pCtx); + }else{ + sqlite3_result_error(pCtx, "malformed JSON", -1); + } + } + jsonParseReset(pParse); + return 1; + } + return 0; +} + +/* Mark node i of pParse as being a child of iParent. Call recursively +** to fill in all the descendants of node i. +*/ +static void jsonParseFillInParentage(JsonParse *pParse, u32 i, u32 iParent){ + JsonNode *pNode = &pParse->aNode[i]; + u32 j; + pParse->aUp[i] = iParent; + switch( pNode->eType ){ + case JSON_ARRAY: { + for(j=1; j<=pNode->n; j += jsonNodeSize(pNode+j)){ + jsonParseFillInParentage(pParse, i+j, i); + } + break; + } + case JSON_OBJECT: { + for(j=1; j<=pNode->n; j += jsonNodeSize(pNode+j+1)+1){ + pParse->aUp[i+j] = i; + jsonParseFillInParentage(pParse, i+j+1, i); + } + break; + } + default: { + break; + } + } +} + +/* +** Compute the parentage of all nodes in a completed parse. +*/ +static int jsonParseFindParents(JsonParse *pParse){ + u32 *aUp; + assert( pParse->aUp==0 ); + aUp = pParse->aUp = sqlite3_malloc64( sizeof(u32)*pParse->nNode ); + if( aUp==0 ){ + pParse->oom = 1; + return SQLITE_NOMEM; + } + jsonParseFillInParentage(pParse, 0, 0); + return SQLITE_OK; +} + +/* +** Magic number used for the JSON parse cache in sqlite3_get_auxdata() +*/ +#define JSON_CACHE_ID (-429938) /* First cache entry */ +#define JSON_CACHE_SZ 4 /* Max number of cache entries */ + +/* +** Obtain a complete parse of the JSON found in the first argument +** of the argv array. Use the sqlite3_get_auxdata() cache for this +** parse if it is available. If the cache is not available or if it +** is no longer valid, parse the JSON again and return the new parse, +** and also register the new parse so that it will be available for +** future sqlite3_get_auxdata() calls. +*/ +static JsonParse *jsonParseCached( + sqlite3_context *pCtx, + sqlite3_value **argv, + sqlite3_context *pErrCtx +){ + const char *zJson = (const char*)sqlite3_value_text(argv[0]); + int nJson = sqlite3_value_bytes(argv[0]); + JsonParse *p; + JsonParse *pMatch = 0; + int iKey; + int iMinKey = 0; + u32 iMinHold = 0xffffffff; + u32 iMaxHold = 0; + if( zJson==0 ) return 0; + for(iKey=0; iKeynJson==nJson + && memcmp(p->zJson,zJson,nJson)==0 + ){ + p->nErr = 0; + pMatch = p; + }else if( p->iHoldiHold; + iMinKey = iKey; + } + if( p->iHold>iMaxHold ){ + iMaxHold = p->iHold; + } + } + if( pMatch ){ + pMatch->nErr = 0; + pMatch->iHold = iMaxHold+1; + return pMatch; + } + p = sqlite3_malloc64( sizeof(*p) + nJson + 1 ); + if( p==0 ){ + sqlite3_result_error_nomem(pCtx); + return 0; + } + memset(p, 0, sizeof(*p)); + p->zJson = (char*)&p[1]; + memcpy((char*)p->zJson, zJson, nJson+1); + if( jsonParse(p, pErrCtx, p->zJson) ){ + sqlite3_free(p); + return 0; + } + p->nJson = nJson; + p->iHold = iMaxHold+1; + sqlite3_set_auxdata(pCtx, JSON_CACHE_ID+iMinKey, p, + (void(*)(void*))jsonParseFree); + return (JsonParse*)sqlite3_get_auxdata(pCtx, JSON_CACHE_ID+iMinKey); +} + +/* +** Compare the OBJECT label at pNode against zKey,nKey. Return true on +** a match. +*/ +static int jsonLabelCompare(JsonNode *pNode, const char *zKey, u32 nKey){ + if( pNode->jnFlags & JNODE_RAW ){ + if( pNode->n!=nKey ) return 0; + return strncmp(pNode->u.zJContent, zKey, nKey)==0; + }else{ + if( pNode->n!=nKey+2 ) return 0; + return strncmp(pNode->u.zJContent+1, zKey, nKey)==0; + } +} + +/* forward declaration */ +static JsonNode *jsonLookupAppend(JsonParse*,const char*,int*,const char**); + +/* +** Search along zPath to find the node specified. Return a pointer +** to that node, or NULL if zPath is malformed or if there is no such +** node. +** +** If pApnd!=0, then try to append new nodes to complete zPath if it is +** possible to do so and if no existing node corresponds to zPath. If +** new nodes are appended *pApnd is set to 1. +*/ +static JsonNode *jsonLookupStep( + JsonParse *pParse, /* The JSON to search */ + u32 iRoot, /* Begin the search at this node */ + const char *zPath, /* The path to search */ + int *pApnd, /* Append nodes to complete path if not NULL */ + const char **pzErr /* Make *pzErr point to any syntax error in zPath */ +){ + u32 i, j, nKey; + const char *zKey; + JsonNode *pRoot = &pParse->aNode[iRoot]; + if( zPath[0]==0 ) return pRoot; + if( pRoot->jnFlags & JNODE_REPLACE ) return 0; + if( zPath[0]=='.' ){ + if( pRoot->eType!=JSON_OBJECT ) return 0; + zPath++; + if( zPath[0]=='"' ){ + zKey = zPath + 1; + for(i=1; zPath[i] && zPath[i]!='"'; i++){} + nKey = i-1; + if( zPath[i] ){ + i++; + }else{ + *pzErr = zPath; + return 0; + } + }else{ + zKey = zPath; + for(i=0; zPath[i] && zPath[i]!='.' && zPath[i]!='['; i++){} + nKey = i; + } + if( nKey==0 ){ + *pzErr = zPath; + return 0; + } + j = 1; + for(;;){ + while( j<=pRoot->n ){ + if( jsonLabelCompare(pRoot+j, zKey, nKey) ){ + return jsonLookupStep(pParse, iRoot+j+1, &zPath[i], pApnd, pzErr); + } + j++; + j += jsonNodeSize(&pRoot[j]); + } + if( (pRoot->jnFlags & JNODE_APPEND)==0 ) break; + iRoot += pRoot->u.iAppend; + pRoot = &pParse->aNode[iRoot]; + j = 1; + } + if( pApnd ){ + u32 iStart, iLabel; + JsonNode *pNode; + iStart = jsonParseAddNode(pParse, JSON_OBJECT, 2, 0); + iLabel = jsonParseAddNode(pParse, JSON_STRING, nKey, zKey); + zPath += i; + pNode = jsonLookupAppend(pParse, zPath, pApnd, pzErr); + if( pParse->oom ) return 0; + if( pNode ){ + pRoot = &pParse->aNode[iRoot]; + pRoot->u.iAppend = iStart - iRoot; + pRoot->jnFlags |= JNODE_APPEND; + pParse->aNode[iLabel].jnFlags |= JNODE_RAW; + } + return pNode; + } + }else if( zPath[0]=='[' && safe_isdigit(zPath[1]) ){ + if( pRoot->eType!=JSON_ARRAY ) return 0; + i = 0; + j = 1; + while( safe_isdigit(zPath[j]) ){ + i = i*10 + zPath[j] - '0'; + j++; + } + if( zPath[j]!=']' ){ + *pzErr = zPath; + return 0; + } + zPath += j + 1; + j = 1; + for(;;){ + while( j<=pRoot->n && (i>0 || (pRoot[j].jnFlags & JNODE_REMOVE)!=0) ){ + if( (pRoot[j].jnFlags & JNODE_REMOVE)==0 ) i--; + j += jsonNodeSize(&pRoot[j]); + } + if( (pRoot->jnFlags & JNODE_APPEND)==0 ) break; + iRoot += pRoot->u.iAppend; + pRoot = &pParse->aNode[iRoot]; + j = 1; + } + if( j<=pRoot->n ){ + return jsonLookupStep(pParse, iRoot+j, zPath, pApnd, pzErr); + } + if( i==0 && pApnd ){ + u32 iStart; + JsonNode *pNode; + iStart = jsonParseAddNode(pParse, JSON_ARRAY, 1, 0); + pNode = jsonLookupAppend(pParse, zPath, pApnd, pzErr); + if( pParse->oom ) return 0; + if( pNode ){ + pRoot = &pParse->aNode[iRoot]; + pRoot->u.iAppend = iStart - iRoot; + pRoot->jnFlags |= JNODE_APPEND; + } + return pNode; + } + }else{ + *pzErr = zPath; + } + return 0; +} + +/* +** Append content to pParse that will complete zPath. Return a pointer +** to the inserted node, or return NULL if the append fails. +*/ +static JsonNode *jsonLookupAppend( + JsonParse *pParse, /* Append content to the JSON parse */ + const char *zPath, /* Description of content to append */ + int *pApnd, /* Set this flag to 1 */ + const char **pzErr /* Make this point to any syntax error */ +){ + *pApnd = 1; + if( zPath[0]==0 ){ + jsonParseAddNode(pParse, JSON_NULL, 0, 0); + return pParse->oom ? 0 : &pParse->aNode[pParse->nNode-1]; + } + if( zPath[0]=='.' ){ + jsonParseAddNode(pParse, JSON_OBJECT, 0, 0); + }else if( strncmp(zPath,"[0]",3)==0 ){ + jsonParseAddNode(pParse, JSON_ARRAY, 0, 0); + }else{ + return 0; + } + if( pParse->oom ) return 0; + return jsonLookupStep(pParse, pParse->nNode-1, zPath, pApnd, pzErr); +} + +/* +** Return the text of a syntax error message on a JSON path. Space is +** obtained from sqlite3_malloc(). +*/ +static char *jsonPathSyntaxError(const char *zErr){ + return sqlite3_mprintf("JSON path error near '%q'", zErr); +} + +/* +** Do a node lookup using zPath. Return a pointer to the node on success. +** Return NULL if not found or if there is an error. +** +** On an error, write an error message into pCtx and increment the +** pParse->nErr counter. +** +** If pApnd!=NULL then try to append missing nodes and set *pApnd = 1 if +** nodes are appended. +*/ +static JsonNode *jsonLookup( + JsonParse *pParse, /* The JSON to search */ + const char *zPath, /* The path to search */ + int *pApnd, /* Append nodes to complete path if not NULL */ + sqlite3_context *pCtx /* Report errors here, if not NULL */ +){ + const char *zErr = 0; + JsonNode *pNode = 0; + char *zMsg; + + if( zPath==0 ) return 0; + if( zPath[0]!='$' ){ + zErr = zPath; + goto lookup_err; + } + zPath++; + pNode = jsonLookupStep(pParse, 0, zPath, pApnd, &zErr); + if( zErr==0 ) return pNode; + +lookup_err: + pParse->nErr++; + assert( zErr!=0 && pCtx!=0 ); + zMsg = jsonPathSyntaxError(zErr); + if( zMsg ){ + sqlite3_result_error(pCtx, zMsg, -1); + sqlite3_free(zMsg); + }else{ + sqlite3_result_error_nomem(pCtx); + } + return 0; +} + + +/* +** Report the wrong number of arguments for json_insert(), json_replace() +** or json_set(). +*/ +static void jsonWrongNumArgs( + sqlite3_context *pCtx, + const char *zFuncName +){ + char *zMsg = sqlite3_mprintf("json_%s() needs an odd number of arguments", + zFuncName); + sqlite3_result_error(pCtx, zMsg, -1); + sqlite3_free(zMsg); +} + +/* +** Mark all NULL entries in the Object passed in as JNODE_REMOVE. +*/ +static void jsonRemoveAllNulls(JsonNode *pNode){ + int i, n; + assert( pNode->eType==JSON_OBJECT ); + n = pNode->n; + for(i=2; i<=n; i += jsonNodeSize(&pNode[i])+1){ + switch( pNode[i].eType ){ + case JSON_NULL: + pNode[i].jnFlags |= JNODE_REMOVE; + break; + case JSON_OBJECT: + jsonRemoveAllNulls(&pNode[i]); + break; + } + } +} + + +/**************************************************************************** +** SQL functions used for testing and debugging +****************************************************************************/ + +#ifdef SQLITE_DEBUG +/* +** The json_parse(JSON) function returns a string which describes +** a parse of the JSON provided. Or it returns NULL if JSON is not +** well-formed. +*/ +static void jsonParseFunc( + sqlite3_context *ctx, + int argc, + sqlite3_value **argv +){ + JsonString s; /* Output string - not real JSON */ + JsonParse x; /* The parse */ + u32 i; + + assert( argc==1 ); + if( jsonParse(&x, ctx, (const char*)sqlite3_value_text(argv[0])) ) return; + jsonParseFindParents(&x); + jsonInit(&s, ctx); + for(i=0; inNode ); + if( argc==2 ){ + const char *zPath = (const char*)sqlite3_value_text(argv[1]); + pNode = jsonLookup(p, zPath, 0, ctx); + }else{ + pNode = p->aNode; + } + if( pNode==0 ){ + return; + } + if( pNode->eType==JSON_ARRAY ){ + assert( (pNode->jnFlags & JNODE_APPEND)==0 ); + for(i=1; i<=pNode->n; n++){ + i += jsonNodeSize(&pNode[i]); + } + } + sqlite3_result_int64(ctx, n); +} + +/* +** json_extract(JSON, PATH, ...) +** +** Return the element described by PATH. Return NULL if there is no +** PATH element. If there are multiple PATHs, then return a JSON array +** with the result from each path. Throw an error if the JSON or any PATH +** is malformed. +*/ +static void jsonExtractFunc( + sqlite3_context *ctx, + int argc, + sqlite3_value **argv +){ + JsonParse *p; /* The parse */ + JsonNode *pNode; + const char *zPath; + JsonString jx; + int i; + + if( argc<2 ) return; + p = jsonParseCached(ctx, argv, ctx); + if( p==0 ) return; + jsonInit(&jx, ctx); + jsonAppendChar(&jx, '['); + for(i=1; inErr ) break; + if( argc>2 ){ + jsonAppendSeparator(&jx); + if( pNode ){ + jsonRenderNode(pNode, &jx, 0); + }else{ + jsonAppendRaw(&jx, "null", 4); + } + }else if( pNode ){ + jsonReturn(pNode, ctx, 0); + } + } + if( argc>2 && i==argc ){ + jsonAppendChar(&jx, ']'); + jsonResult(&jx); + sqlite3_result_subtype(ctx, JSON_SUBTYPE); + } + jsonReset(&jx); +} + +/* This is the RFC 7396 MergePatch algorithm. +*/ +static JsonNode *jsonMergePatch( + JsonParse *pParse, /* The JSON parser that contains the TARGET */ + u32 iTarget, /* Node of the TARGET in pParse */ + JsonNode *pPatch /* The PATCH */ +){ + u32 i, j; + u32 iRoot; + JsonNode *pTarget; + if( pPatch->eType!=JSON_OBJECT ){ + return pPatch; + } + assert( iTarget>=0 && iTargetnNode ); + pTarget = &pParse->aNode[iTarget]; + assert( (pPatch->jnFlags & JNODE_APPEND)==0 ); + if( pTarget->eType!=JSON_OBJECT ){ + jsonRemoveAllNulls(pPatch); + return pPatch; + } + iRoot = iTarget; + for(i=1; in; i += jsonNodeSize(&pPatch[i+1])+1){ + u32 nKey; + const char *zKey; + assert( pPatch[i].eType==JSON_STRING ); + assert( pPatch[i].jnFlags & JNODE_LABEL ); + nKey = pPatch[i].n; + zKey = pPatch[i].u.zJContent; + assert( (pPatch[i].jnFlags & JNODE_RAW)==0 ); + for(j=1; jn; j += jsonNodeSize(&pTarget[j+1])+1 ){ + assert( pTarget[j].eType==JSON_STRING ); + assert( pTarget[j].jnFlags & JNODE_LABEL ); + assert( (pPatch[i].jnFlags & JNODE_RAW)==0 ); + if( pTarget[j].n==nKey && strncmp(pTarget[j].u.zJContent,zKey,nKey)==0 ){ + if( pTarget[j+1].jnFlags & (JNODE_REMOVE|JNODE_PATCH) ) break; + if( pPatch[i+1].eType==JSON_NULL ){ + pTarget[j+1].jnFlags |= JNODE_REMOVE; + }else{ + JsonNode *pNew = jsonMergePatch(pParse, iTarget+j+1, &pPatch[i+1]); + if( pNew==0 ) return 0; + pTarget = &pParse->aNode[iTarget]; + if( pNew!=&pTarget[j+1] ){ + pTarget[j+1].u.pPatch = pNew; + pTarget[j+1].jnFlags |= JNODE_PATCH; + } + } + break; + } + } + if( j>=pTarget->n && pPatch[i+1].eType!=JSON_NULL ){ + int iStart, iPatch; + iStart = jsonParseAddNode(pParse, JSON_OBJECT, 2, 0); + jsonParseAddNode(pParse, JSON_STRING, nKey, zKey); + iPatch = jsonParseAddNode(pParse, JSON_TRUE, 0, 0); + if( pParse->oom ) return 0; + jsonRemoveAllNulls(pPatch); + pTarget = &pParse->aNode[iTarget]; + pParse->aNode[iRoot].jnFlags |= JNODE_APPEND; + pParse->aNode[iRoot].u.iAppend = iStart - iRoot; + iRoot = iStart; + pParse->aNode[iPatch].jnFlags |= JNODE_PATCH; + pParse->aNode[iPatch].u.pPatch = &pPatch[i+1]; + } + } + return pTarget; +} + +/* +** Implementation of the json_mergepatch(JSON1,JSON2) function. Return a JSON +** object that is the result of running the RFC 7396 MergePatch() algorithm +** on the two arguments. +*/ +static void jsonPatchFunc( + sqlite3_context *ctx, + int argc, + sqlite3_value **argv +){ + JsonParse x; /* The JSON that is being patched */ + JsonParse y; /* The patch */ + JsonNode *pResult; /* The result of the merge */ + + UNUSED_PARAM(argc); + if( jsonParse(&x, ctx, (const char*)sqlite3_value_text(argv[0])) ) return; + if( jsonParse(&y, ctx, (const char*)sqlite3_value_text(argv[1])) ){ + jsonParseReset(&x); + return; + } + pResult = jsonMergePatch(&x, 0, y.aNode); + assert( pResult!=0 || x.oom ); + if( pResult ){ + jsonReturnJson(pResult, ctx, 0); + }else{ + sqlite3_result_error_nomem(ctx); + } + jsonParseReset(&x); + jsonParseReset(&y); +} + + +/* +** Implementation of the json_object(NAME,VALUE,...) function. Return a JSON +** object that contains all name/value given in arguments. Or if any name +** is not a string or if any value is a BLOB, throw an error. +*/ +static void jsonObjectFunc( + sqlite3_context *ctx, + int argc, + sqlite3_value **argv +){ + int i; + JsonString jx; + const char *z; + u32 n; + + if( argc&1 ){ + sqlite3_result_error(ctx, "json_object() requires an even number " + "of arguments", -1); + return; + } + jsonInit(&jx, ctx); + jsonAppendChar(&jx, '{'); + for(i=0; ijnFlags |= JNODE_REMOVE; + } + if( (x.aNode[0].jnFlags & JNODE_REMOVE)==0 ){ + jsonReturnJson(x.aNode, ctx, 0); + } +remove_done: + jsonParseReset(&x); +} + +/* +** json_replace(JSON, PATH, VALUE, ...) +** +** Replace the value at PATH with VALUE. If PATH does not already exist, +** this routine is a no-op. If JSON or PATH is malformed, throw an error. +*/ +static void jsonReplaceFunc( + sqlite3_context *ctx, + int argc, + sqlite3_value **argv +){ + JsonParse x; /* The parse */ + JsonNode *pNode; + const char *zPath; + u32 i; + + if( argc<1 ) return; + if( (argc&1)==0 ) { + jsonWrongNumArgs(ctx, "replace"); + return; + } + if( jsonParse(&x, ctx, (const char*)sqlite3_value_text(argv[0])) ) return; + assert( x.nNode ); + for(i=1; i<(u32)argc; i+=2){ + zPath = (const char*)sqlite3_value_text(argv[i]); + pNode = jsonLookup(&x, zPath, 0, ctx); + if( x.nErr ) goto replace_err; + if( pNode ){ + pNode->jnFlags |= (u8)JNODE_REPLACE; + pNode->u.iReplace = i + 1; + } + } + if( x.aNode[0].jnFlags & JNODE_REPLACE ){ + sqlite3_result_value(ctx, argv[x.aNode[0].u.iReplace]); + }else{ + jsonReturnJson(x.aNode, ctx, argv); + } +replace_err: + jsonParseReset(&x); +} + +/* +** json_set(JSON, PATH, VALUE, ...) +** +** Set the value at PATH to VALUE. Create the PATH if it does not already +** exist. Overwrite existing values that do exist. +** If JSON or PATH is malformed, throw an error. +** +** json_insert(JSON, PATH, VALUE, ...) +** +** Create PATH and initialize it to VALUE. If PATH already exists, this +** routine is a no-op. If JSON or PATH is malformed, throw an error. +*/ +static void jsonSetFunc( + sqlite3_context *ctx, + int argc, + sqlite3_value **argv +){ + JsonParse x; /* The parse */ + JsonNode *pNode; + const char *zPath; + u32 i; + int bApnd; + int bIsSet = *(int*)sqlite3_user_data(ctx); + + if( argc<1 ) return; + if( (argc&1)==0 ) { + jsonWrongNumArgs(ctx, bIsSet ? "set" : "insert"); + return; + } + if( jsonParse(&x, ctx, (const char*)sqlite3_value_text(argv[0])) ) return; + assert( x.nNode ); + for(i=1; i<(u32)argc; i+=2){ + zPath = (const char*)sqlite3_value_text(argv[i]); + bApnd = 0; + pNode = jsonLookup(&x, zPath, &bApnd, ctx); + if( x.oom ){ + sqlite3_result_error_nomem(ctx); + goto jsonSetDone; + }else if( x.nErr ){ + goto jsonSetDone; + }else if( pNode && (bApnd || bIsSet) ){ + pNode->jnFlags |= (u8)JNODE_REPLACE; + pNode->u.iReplace = i + 1; + } + } + if( x.aNode[0].jnFlags & JNODE_REPLACE ){ + sqlite3_result_value(ctx, argv[x.aNode[0].u.iReplace]); + }else{ + jsonReturnJson(x.aNode, ctx, argv); + } +jsonSetDone: + jsonParseReset(&x); +} + +/* +** json_type(JSON) +** json_type(JSON, PATH) +** +** Return the top-level "type" of a JSON string. Throw an error if +** either the JSON or PATH inputs are not well-formed. +*/ +static void jsonTypeFunc( + sqlite3_context *ctx, + int argc, + sqlite3_value **argv +){ + JsonParse *p; /* The parse */ + const char *zPath; + JsonNode *pNode; + + p = jsonParseCached(ctx, argv, ctx); + if( p==0 ) return; + if( argc==2 ){ + zPath = (const char*)sqlite3_value_text(argv[1]); + pNode = jsonLookup(p, zPath, 0, ctx); + }else{ + pNode = p->aNode; + } + if( pNode ){ + sqlite3_result_text(ctx, jsonType[pNode->eType], -1, SQLITE_STATIC); + } +} + +/* +** json_valid(JSON) +** +** Return 1 if JSON is a well-formed JSON string according to RFC-7159. +** Return 0 otherwise. +*/ +static void jsonValidFunc( + sqlite3_context *ctx, + int argc, + sqlite3_value **argv +){ + JsonParse *p; /* The parse */ + UNUSED_PARAM(argc); + p = jsonParseCached(ctx, argv, 0); + sqlite3_result_int(ctx, p!=0); +} + + +/**************************************************************************** +** Aggregate SQL function implementations +****************************************************************************/ +/* +** json_group_array(VALUE) +** +** Return a JSON array composed of all values in the aggregate. +*/ +static void jsonArrayStep( + sqlite3_context *ctx, + int argc, + sqlite3_value **argv +){ + JsonString *pStr; + UNUSED_PARAM(argc); + pStr = (JsonString*)sqlite3_aggregate_context(ctx, sizeof(*pStr)); + if( pStr ){ + if( pStr->zBuf==0 ){ + jsonInit(pStr, ctx); + jsonAppendChar(pStr, '['); + }else if( pStr->nUsed>1 ){ + jsonAppendChar(pStr, ','); + pStr->pCtx = ctx; + } + jsonAppendValue(pStr, argv[0]); + } +} +static void jsonArrayCompute(sqlite3_context *ctx, int isFinal){ + JsonString *pStr; + pStr = (JsonString*)sqlite3_aggregate_context(ctx, 0); + if( pStr ){ + pStr->pCtx = ctx; + jsonAppendChar(pStr, ']'); + if( pStr->bErr ){ + if( pStr->bErr==1 ) sqlite3_result_error_nomem(ctx); + assert( pStr->bStatic ); + }else if( isFinal ){ + sqlite3_result_text(ctx, pStr->zBuf, (int)pStr->nUsed, + pStr->bStatic ? SQLITE_TRANSIENT : sqlite3_free); + pStr->bStatic = 1; + }else{ + sqlite3_result_text(ctx, pStr->zBuf, (int)pStr->nUsed, SQLITE_TRANSIENT); + pStr->nUsed--; + } + }else{ + sqlite3_result_text(ctx, "[]", 2, SQLITE_STATIC); + } + sqlite3_result_subtype(ctx, JSON_SUBTYPE); +} +static void jsonArrayValue(sqlite3_context *ctx){ + jsonArrayCompute(ctx, 0); +} +static void jsonArrayFinal(sqlite3_context *ctx){ + jsonArrayCompute(ctx, 1); +} + +#ifndef SQLITE_OMIT_WINDOWFUNC +/* +** This method works for both json_group_array() and json_group_object(). +** It works by removing the first element of the group by searching forward +** to the first comma (",") that is not within a string and deleting all +** text through that comma. +*/ +static void jsonGroupInverse( + sqlite3_context *ctx, + int argc, + sqlite3_value **argv +){ + unsigned int i; + int inStr = 0; + int nNest = 0; + char *z; + char c; + JsonString *pStr; + UNUSED_PARAM(argc); + UNUSED_PARAM(argv); + pStr = (JsonString*)sqlite3_aggregate_context(ctx, 0); +#ifdef NEVER + /* pStr is always non-NULL since jsonArrayStep() or jsonObjectStep() will + ** always have been called to initalize it */ + if( NEVER(!pStr) ) return; +#endif + z = pStr->zBuf; + for(i=1; (c = z[i])!=',' || inStr || nNest; i++){ + if( i>=pStr->nUsed ){ + pStr->nUsed = 1; + return; + } + if( c=='"' ){ + inStr = !inStr; + }else if( c=='\\' ){ + i++; + }else if( !inStr ){ + if( c=='{' || c=='[' ) nNest++; + if( c=='}' || c==']' ) nNest--; + } + } + pStr->nUsed -= i; + memmove(&z[1], &z[i+1], (size_t)pStr->nUsed-1); +} +#else +# define jsonGroupInverse 0 +#endif + + +/* +** json_group_obj(NAME,VALUE) +** +** Return a JSON object composed of all names and values in the aggregate. +*/ +static void jsonObjectStep( + sqlite3_context *ctx, + int argc, + sqlite3_value **argv +){ + JsonString *pStr; + const char *z; + u32 n; + UNUSED_PARAM(argc); + pStr = (JsonString*)sqlite3_aggregate_context(ctx, sizeof(*pStr)); + if( pStr ){ + if( pStr->zBuf==0 ){ + jsonInit(pStr, ctx); + jsonAppendChar(pStr, '{'); + }else if( pStr->nUsed>1 ){ + jsonAppendChar(pStr, ','); + pStr->pCtx = ctx; + } + z = (const char*)sqlite3_value_text(argv[0]); + n = (u32)sqlite3_value_bytes(argv[0]); + jsonAppendString(pStr, z, n); + jsonAppendChar(pStr, ':'); + jsonAppendValue(pStr, argv[1]); + } +} +static void jsonObjectCompute(sqlite3_context *ctx, int isFinal){ + JsonString *pStr; + pStr = (JsonString*)sqlite3_aggregate_context(ctx, 0); + if( pStr ){ + jsonAppendChar(pStr, '}'); + if( pStr->bErr ){ + if( pStr->bErr==1 ) sqlite3_result_error_nomem(ctx); + assert( pStr->bStatic ); + }else if( isFinal ){ + sqlite3_result_text(ctx, pStr->zBuf, (int)pStr->nUsed, + pStr->bStatic ? SQLITE_TRANSIENT : sqlite3_free); + pStr->bStatic = 1; + }else{ + sqlite3_result_text(ctx, pStr->zBuf, (int)pStr->nUsed, SQLITE_TRANSIENT); + pStr->nUsed--; + } + }else{ + sqlite3_result_text(ctx, "{}", 2, SQLITE_STATIC); + } + sqlite3_result_subtype(ctx, JSON_SUBTYPE); +} +static void jsonObjectValue(sqlite3_context *ctx){ + jsonObjectCompute(ctx, 0); +} +static void jsonObjectFinal(sqlite3_context *ctx){ + jsonObjectCompute(ctx, 1); +} + + + +#ifndef SQLITE_OMIT_VIRTUALTABLE +/**************************************************************************** +** The json_each virtual table +****************************************************************************/ +typedef struct JsonEachCursor JsonEachCursor; +struct JsonEachCursor { + sqlite3_vtab_cursor base; /* Base class - must be first */ + u32 iRowid; /* The rowid */ + u32 iBegin; /* The first node of the scan */ + u32 i; /* Index in sParse.aNode[] of current row */ + u32 iEnd; /* EOF when i equals or exceeds this value */ + u8 eType; /* Type of top-level element */ + u8 bRecursive; /* True for json_tree(). False for json_each() */ + char *zJson; /* Input JSON */ + char *zRoot; /* Path by which to filter zJson */ + JsonParse sParse; /* Parse of the input JSON */ +}; + +/* Constructor for the json_each virtual table */ +static int jsonEachConnect( + sqlite3 *db, + void *pAux, + int argc, const char *const*argv, + sqlite3_vtab **ppVtab, + char **pzErr +){ + sqlite3_vtab *pNew; + int rc; + +/* Column numbers */ +#define JEACH_KEY 0 +#define JEACH_VALUE 1 +#define JEACH_TYPE 2 +#define JEACH_ATOM 3 +#define JEACH_ID 4 +#define JEACH_PARENT 5 +#define JEACH_FULLKEY 6 +#define JEACH_PATH 7 +/* The xBestIndex method assumes that the JSON and ROOT columns are +** the last two columns in the table. Should this ever changes, be +** sure to update the xBestIndex method. */ +#define JEACH_JSON 8 +#define JEACH_ROOT 9 + + UNUSED_PARAM(pzErr); + UNUSED_PARAM(argv); + UNUSED_PARAM(argc); + UNUSED_PARAM(pAux); + rc = sqlite3_declare_vtab(db, + "CREATE TABLE x(key,value,type,atom,id,parent,fullkey,path," + "json HIDDEN,root HIDDEN)"); + if( rc==SQLITE_OK ){ + pNew = *ppVtab = sqlite3_malloc( sizeof(*pNew) ); + if( pNew==0 ) return SQLITE_NOMEM; + memset(pNew, 0, sizeof(*pNew)); + } + return rc; +} + +/* destructor for json_each virtual table */ +static int jsonEachDisconnect(sqlite3_vtab *pVtab){ + sqlite3_free(pVtab); + return SQLITE_OK; +} + +/* constructor for a JsonEachCursor object for json_each(). */ +static int jsonEachOpenEach(sqlite3_vtab *p, sqlite3_vtab_cursor **ppCursor){ + JsonEachCursor *pCur; + + UNUSED_PARAM(p); + pCur = sqlite3_malloc( sizeof(*pCur) ); + if( pCur==0 ) return SQLITE_NOMEM; + memset(pCur, 0, sizeof(*pCur)); + *ppCursor = &pCur->base; + return SQLITE_OK; +} + +/* constructor for a JsonEachCursor object for json_tree(). */ +static int jsonEachOpenTree(sqlite3_vtab *p, sqlite3_vtab_cursor **ppCursor){ + int rc = jsonEachOpenEach(p, ppCursor); + if( rc==SQLITE_OK ){ + JsonEachCursor *pCur = (JsonEachCursor*)*ppCursor; + pCur->bRecursive = 1; + } + return rc; +} + +/* Reset a JsonEachCursor back to its original state. Free any memory +** held. */ +static void jsonEachCursorReset(JsonEachCursor *p){ + sqlite3_free(p->zJson); + sqlite3_free(p->zRoot); + jsonParseReset(&p->sParse); + p->iRowid = 0; + p->i = 0; + p->iEnd = 0; + p->eType = 0; + p->zJson = 0; + p->zRoot = 0; +} + +/* Destructor for a jsonEachCursor object */ +static int jsonEachClose(sqlite3_vtab_cursor *cur){ + JsonEachCursor *p = (JsonEachCursor*)cur; + jsonEachCursorReset(p); + sqlite3_free(cur); + return SQLITE_OK; +} + +/* Return TRUE if the jsonEachCursor object has been advanced off the end +** of the JSON object */ +static int jsonEachEof(sqlite3_vtab_cursor *cur){ + JsonEachCursor *p = (JsonEachCursor*)cur; + return p->i >= p->iEnd; +} + +/* Advance the cursor to the next element for json_tree() */ +static int jsonEachNext(sqlite3_vtab_cursor *cur){ + JsonEachCursor *p = (JsonEachCursor*)cur; + if( p->bRecursive ){ + if( p->sParse.aNode[p->i].jnFlags & JNODE_LABEL ) p->i++; + p->i++; + p->iRowid++; + if( p->iiEnd ){ + u32 iUp = p->sParse.aUp[p->i]; + JsonNode *pUp = &p->sParse.aNode[iUp]; + p->eType = pUp->eType; + if( pUp->eType==JSON_ARRAY ){ + if( iUp==p->i-1 ){ + pUp->u.iKey = 0; + }else{ + pUp->u.iKey++; + } + } + } + }else{ + switch( p->eType ){ + case JSON_ARRAY: { + p->i += jsonNodeSize(&p->sParse.aNode[p->i]); + p->iRowid++; + break; + } + case JSON_OBJECT: { + p->i += 1 + jsonNodeSize(&p->sParse.aNode[p->i+1]); + p->iRowid++; + break; + } + default: { + p->i = p->iEnd; + break; + } + } + } + return SQLITE_OK; +} + +/* Append the name of the path for element i to pStr +*/ +static void jsonEachComputePath( + JsonEachCursor *p, /* The cursor */ + JsonString *pStr, /* Write the path here */ + u32 i /* Path to this element */ +){ + JsonNode *pNode, *pUp; + u32 iUp; + if( i==0 ){ + jsonAppendChar(pStr, '$'); + return; + } + iUp = p->sParse.aUp[i]; + jsonEachComputePath(p, pStr, iUp); + pNode = &p->sParse.aNode[i]; + pUp = &p->sParse.aNode[iUp]; + if( pUp->eType==JSON_ARRAY ){ + jsonPrintf(30, pStr, "[%d]", pUp->u.iKey); + }else{ + assert( pUp->eType==JSON_OBJECT ); + if( (pNode->jnFlags & JNODE_LABEL)==0 ) pNode--; + assert( pNode->eType==JSON_STRING ); + assert( pNode->jnFlags & JNODE_LABEL ); + jsonPrintf(pNode->n+1, pStr, ".%.*s", pNode->n-2, pNode->u.zJContent+1); + } +} + +/* Return the value of a column */ +static int jsonEachColumn( + sqlite3_vtab_cursor *cur, /* The cursor */ + sqlite3_context *ctx, /* First argument to sqlite3_result_...() */ + int i /* Which column to return */ +){ + JsonEachCursor *p = (JsonEachCursor*)cur; + JsonNode *pThis = &p->sParse.aNode[p->i]; + switch( i ){ + case JEACH_KEY: { + if( p->i==0 ) break; + if( p->eType==JSON_OBJECT ){ + jsonReturn(pThis, ctx, 0); + }else if( p->eType==JSON_ARRAY ){ + u32 iKey; + if( p->bRecursive ){ + if( p->iRowid==0 ) break; + iKey = p->sParse.aNode[p->sParse.aUp[p->i]].u.iKey; + }else{ + iKey = p->iRowid; + } + sqlite3_result_int64(ctx, (sqlite3_int64)iKey); + } + break; + } + case JEACH_VALUE: { + if( pThis->jnFlags & JNODE_LABEL ) pThis++; + jsonReturn(pThis, ctx, 0); + break; + } + case JEACH_TYPE: { + if( pThis->jnFlags & JNODE_LABEL ) pThis++; + sqlite3_result_text(ctx, jsonType[pThis->eType], -1, SQLITE_STATIC); + break; + } + case JEACH_ATOM: { + if( pThis->jnFlags & JNODE_LABEL ) pThis++; + if( pThis->eType>=JSON_ARRAY ) break; + jsonReturn(pThis, ctx, 0); + break; + } + case JEACH_ID: { + sqlite3_result_int64(ctx, + (sqlite3_int64)p->i + ((pThis->jnFlags & JNODE_LABEL)!=0)); + break; + } + case JEACH_PARENT: { + if( p->i>p->iBegin && p->bRecursive ){ + sqlite3_result_int64(ctx, (sqlite3_int64)p->sParse.aUp[p->i]); + } + break; + } + case JEACH_FULLKEY: { + JsonString x; + jsonInit(&x, ctx); + if( p->bRecursive ){ + jsonEachComputePath(p, &x, p->i); + }else{ + if( p->zRoot ){ + jsonAppendRaw(&x, p->zRoot, (int)strlen(p->zRoot)); + }else{ + jsonAppendChar(&x, '$'); + } + if( p->eType==JSON_ARRAY ){ + jsonPrintf(30, &x, "[%d]", p->iRowid); + }else if( p->eType==JSON_OBJECT ){ + jsonPrintf(pThis->n, &x, ".%.*s", pThis->n-2, pThis->u.zJContent+1); + } + } + jsonResult(&x); + break; + } + case JEACH_PATH: { + if( p->bRecursive ){ + JsonString x; + jsonInit(&x, ctx); + jsonEachComputePath(p, &x, p->sParse.aUp[p->i]); + jsonResult(&x); + break; + } + /* For json_each() path and root are the same so fall through + ** into the root case */ + } + default: { + const char *zRoot = p->zRoot; + if( zRoot==0 ) zRoot = "$"; + sqlite3_result_text(ctx, zRoot, -1, SQLITE_STATIC); + break; + } + case JEACH_JSON: { + assert( i==JEACH_JSON ); + sqlite3_result_text(ctx, p->sParse.zJson, -1, SQLITE_STATIC); + break; + } + } + return SQLITE_OK; +} + +/* Return the current rowid value */ +static int jsonEachRowid(sqlite3_vtab_cursor *cur, sqlite_int64 *pRowid){ + JsonEachCursor *p = (JsonEachCursor*)cur; + *pRowid = p->iRowid; + return SQLITE_OK; +} + +/* The query strategy is to look for an equality constraint on the json +** column. Without such a constraint, the table cannot operate. idxNum is +** 1 if the constraint is found, 3 if the constraint and zRoot are found, +** and 0 otherwise. +*/ +static int jsonEachBestIndex( + sqlite3_vtab *tab, + sqlite3_index_info *pIdxInfo +){ + int i; /* Loop counter or computed array index */ + int aIdx[2]; /* Index of constraints for JSON and ROOT */ + int unusableMask = 0; /* Mask of unusable JSON and ROOT constraints */ + int idxMask = 0; /* Mask of usable == constraints JSON and ROOT */ + const struct sqlite3_index_constraint *pConstraint; + + /* This implementation assumes that JSON and ROOT are the last two + ** columns in the table */ + assert( JEACH_ROOT == JEACH_JSON+1 ); + UNUSED_PARAM(tab); + aIdx[0] = aIdx[1] = -1; + pConstraint = pIdxInfo->aConstraint; + for(i=0; inConstraint; i++, pConstraint++){ + int iCol; + int iMask; + if( pConstraint->iColumn < JEACH_JSON ) continue; + iCol = pConstraint->iColumn - JEACH_JSON; + assert( iCol==0 || iCol==1 ); + iMask = 1 << iCol; + if( pConstraint->usable==0 ){ + unusableMask |= iMask; + }else if( pConstraint->op==SQLITE_INDEX_CONSTRAINT_EQ ){ + aIdx[iCol] = i; + idxMask |= iMask; + } + } + if( (unusableMask & ~idxMask)!=0 ){ + /* If there are any unusable constraints on JSON or ROOT, then reject + ** this entire plan */ + return SQLITE_CONSTRAINT; + } + if( aIdx[0]<0 ){ + /* No JSON input. Leave estimatedCost at the huge value that it was + ** initialized to to discourage the query planner from selecting this + ** plan. */ + pIdxInfo->idxNum = 0; + }else{ + pIdxInfo->estimatedCost = 1.0; + i = aIdx[0]; + pIdxInfo->aConstraintUsage[i].argvIndex = 1; + pIdxInfo->aConstraintUsage[i].omit = 1; + if( aIdx[1]<0 ){ + pIdxInfo->idxNum = 1; /* Only JSON supplied. Plan 1 */ + }else{ + i = aIdx[1]; + pIdxInfo->aConstraintUsage[i].argvIndex = 2; + pIdxInfo->aConstraintUsage[i].omit = 1; + pIdxInfo->idxNum = 3; /* Both JSON and ROOT are supplied. Plan 3 */ + } + } + return SQLITE_OK; +} + +/* Start a search on a new JSON string */ +static int jsonEachFilter( + sqlite3_vtab_cursor *cur, + int idxNum, const char *idxStr, + int argc, sqlite3_value **argv +){ + JsonEachCursor *p = (JsonEachCursor*)cur; + const char *z; + const char *zRoot = 0; + sqlite3_int64 n; + + UNUSED_PARAM(idxStr); + UNUSED_PARAM(argc); + jsonEachCursorReset(p); + if( idxNum==0 ) return SQLITE_OK; + z = (const char*)sqlite3_value_text(argv[0]); + if( z==0 ) return SQLITE_OK; + n = sqlite3_value_bytes(argv[0]); + p->zJson = sqlite3_malloc64( n+1 ); + if( p->zJson==0 ) return SQLITE_NOMEM; + memcpy(p->zJson, z, (size_t)n+1); + if( jsonParse(&p->sParse, 0, p->zJson) ){ + int rc = SQLITE_NOMEM; + if( p->sParse.oom==0 ){ + sqlite3_free(cur->pVtab->zErrMsg); + cur->pVtab->zErrMsg = sqlite3_mprintf("malformed JSON"); + if( cur->pVtab->zErrMsg ) rc = SQLITE_ERROR; + } + jsonEachCursorReset(p); + return rc; + }else if( p->bRecursive && jsonParseFindParents(&p->sParse) ){ + jsonEachCursorReset(p); + return SQLITE_NOMEM; + }else{ + JsonNode *pNode = 0; + if( idxNum==3 ){ + const char *zErr = 0; + zRoot = (const char*)sqlite3_value_text(argv[1]); + if( zRoot==0 ) return SQLITE_OK; + n = sqlite3_value_bytes(argv[1]); + p->zRoot = sqlite3_malloc64( n+1 ); + if( p->zRoot==0 ) return SQLITE_NOMEM; + memcpy(p->zRoot, zRoot, (size_t)n+1); + if( zRoot[0]!='$' ){ + zErr = zRoot; + }else{ + pNode = jsonLookupStep(&p->sParse, 0, p->zRoot+1, 0, &zErr); + } + if( zErr ){ + sqlite3_free(cur->pVtab->zErrMsg); + cur->pVtab->zErrMsg = jsonPathSyntaxError(zErr); + jsonEachCursorReset(p); + return cur->pVtab->zErrMsg ? SQLITE_ERROR : SQLITE_NOMEM; + }else if( pNode==0 ){ + return SQLITE_OK; + } + }else{ + pNode = p->sParse.aNode; + } + p->iBegin = p->i = (int)(pNode - p->sParse.aNode); + p->eType = pNode->eType; + if( p->eType>=JSON_ARRAY ){ + pNode->u.iKey = 0; + p->iEnd = p->i + pNode->n + 1; + if( p->bRecursive ){ + p->eType = p->sParse.aNode[p->sParse.aUp[p->i]].eType; + if( p->i>0 && (p->sParse.aNode[p->i-1].jnFlags & JNODE_LABEL)!=0 ){ + p->i--; + } + }else{ + p->i++; + } + }else{ + p->iEnd = p->i+1; + } + } + return SQLITE_OK; +} + +/* The methods of the json_each virtual table */ +static sqlite3_module jsonEachModule = { + 0, /* iVersion */ + 0, /* xCreate */ + jsonEachConnect, /* xConnect */ + jsonEachBestIndex, /* xBestIndex */ + jsonEachDisconnect, /* xDisconnect */ + 0, /* xDestroy */ + jsonEachOpenEach, /* xOpen - open a cursor */ + jsonEachClose, /* xClose - close a cursor */ + jsonEachFilter, /* xFilter - configure scan constraints */ + jsonEachNext, /* xNext - advance a cursor */ + jsonEachEof, /* xEof - check for end of scan */ + jsonEachColumn, /* xColumn - read data */ + jsonEachRowid, /* xRowid - read data */ + 0, /* xUpdate */ + 0, /* xBegin */ + 0, /* xSync */ + 0, /* xCommit */ + 0, /* xRollback */ + 0, /* xFindMethod */ + 0, /* xRename */ + 0, /* xSavepoint */ + 0, /* xRelease */ + 0, /* xRollbackTo */ + 0 /* xShadowName */ +}; + +/* The methods of the json_tree virtual table. */ +static sqlite3_module jsonTreeModule = { + 0, /* iVersion */ + 0, /* xCreate */ + jsonEachConnect, /* xConnect */ + jsonEachBestIndex, /* xBestIndex */ + jsonEachDisconnect, /* xDisconnect */ + 0, /* xDestroy */ + jsonEachOpenTree, /* xOpen - open a cursor */ + jsonEachClose, /* xClose - close a cursor */ + jsonEachFilter, /* xFilter - configure scan constraints */ + jsonEachNext, /* xNext - advance a cursor */ + jsonEachEof, /* xEof - check for end of scan */ + jsonEachColumn, /* xColumn - read data */ + jsonEachRowid, /* xRowid - read data */ + 0, /* xUpdate */ + 0, /* xBegin */ + 0, /* xSync */ + 0, /* xCommit */ + 0, /* xRollback */ + 0, /* xFindMethod */ + 0, /* xRename */ + 0, /* xSavepoint */ + 0, /* xRelease */ + 0, /* xRollbackTo */ + 0 /* xShadowName */ +}; +#endif /* SQLITE_OMIT_VIRTUALTABLE */ + +/**************************************************************************** +** The following routines are the only publically visible identifiers in this +** file. Call the following routines in order to register the various SQL +** functions and the virtual table implemented by this file. +****************************************************************************/ + +SQLITE_PRIVATE int sqlite3Json1Init(sqlite3 *db){ + int rc = SQLITE_OK; + unsigned int i; + static const struct { + const char *zName; + int nArg; + int flag; + void (*xFunc)(sqlite3_context*,int,sqlite3_value**); + } aFunc[] = { + { "json", 1, 0, jsonRemoveFunc }, + { "json_array", -1, 0, jsonArrayFunc }, + { "json_array_length", 1, 0, jsonArrayLengthFunc }, + { "json_array_length", 2, 0, jsonArrayLengthFunc }, + { "json_extract", -1, 0, jsonExtractFunc }, + { "json_insert", -1, 0, jsonSetFunc }, + { "json_object", -1, 0, jsonObjectFunc }, + { "json_patch", 2, 0, jsonPatchFunc }, + { "json_quote", 1, 0, jsonQuoteFunc }, + { "json_remove", -1, 0, jsonRemoveFunc }, + { "json_replace", -1, 0, jsonReplaceFunc }, + { "json_set", -1, 1, jsonSetFunc }, + { "json_type", 1, 0, jsonTypeFunc }, + { "json_type", 2, 0, jsonTypeFunc }, + { "json_valid", 1, 0, jsonValidFunc }, + +#if SQLITE_DEBUG + /* DEBUG and TESTING functions */ + { "json_parse", 1, 0, jsonParseFunc }, + { "json_test1", 1, 0, jsonTest1Func }, +#endif + }; + static const struct { + const char *zName; + int nArg; + void (*xStep)(sqlite3_context*,int,sqlite3_value**); + void (*xFinal)(sqlite3_context*); + void (*xValue)(sqlite3_context*); + } aAgg[] = { + { "json_group_array", 1, + jsonArrayStep, jsonArrayFinal, jsonArrayValue }, + { "json_group_object", 2, + jsonObjectStep, jsonObjectFinal, jsonObjectValue }, + }; +#ifndef SQLITE_OMIT_VIRTUALTABLE + static const struct { + const char *zName; + sqlite3_module *pModule; + } aMod[] = { + { "json_each", &jsonEachModule }, + { "json_tree", &jsonTreeModule }, + }; +#endif + for(i=0; i */ +/* #include */ +/* #include */ + +/* The following macro is used to suppress compiler warnings. +*/ +#ifndef UNUSED_PARAMETER +# define UNUSED_PARAMETER(x) (void)(x) +#endif + +typedef struct Rtree Rtree; +typedef struct RtreeCursor RtreeCursor; +typedef struct RtreeNode RtreeNode; +typedef struct RtreeCell RtreeCell; +typedef struct RtreeConstraint RtreeConstraint; +typedef struct RtreeMatchArg RtreeMatchArg; +typedef struct RtreeGeomCallback RtreeGeomCallback; +typedef union RtreeCoord RtreeCoord; +typedef struct RtreeSearchPoint RtreeSearchPoint; + +/* The rtree may have between 1 and RTREE_MAX_DIMENSIONS dimensions. */ +#define RTREE_MAX_DIMENSIONS 5 + +/* Maximum number of auxiliary columns */ +#define RTREE_MAX_AUX_COLUMN 100 + +/* Size of hash table Rtree.aHash. This hash table is not expected to +** ever contain very many entries, so a fixed number of buckets is +** used. +*/ +#define HASHSIZE 97 + +/* The xBestIndex method of this virtual table requires an estimate of +** the number of rows in the virtual table to calculate the costs of +** various strategies. If possible, this estimate is loaded from the +** sqlite_stat1 table (with RTREE_MIN_ROWEST as a hard-coded minimum). +** Otherwise, if no sqlite_stat1 entry is available, use +** RTREE_DEFAULT_ROWEST. +*/ +#define RTREE_DEFAULT_ROWEST 1048576 +#define RTREE_MIN_ROWEST 100 + +/* +** An rtree virtual-table object. +*/ +struct Rtree { + sqlite3_vtab base; /* Base class. Must be first */ + sqlite3 *db; /* Host database connection */ + int iNodeSize; /* Size in bytes of each node in the node table */ + u8 nDim; /* Number of dimensions */ + u8 nDim2; /* Twice the number of dimensions */ + u8 eCoordType; /* RTREE_COORD_REAL32 or RTREE_COORD_INT32 */ + u8 nBytesPerCell; /* Bytes consumed per cell */ + u8 inWrTrans; /* True if inside write transaction */ + u8 nAux; /* # of auxiliary columns in %_rowid */ + u8 nAuxNotNull; /* Number of initial not-null aux columns */ +#ifdef SQLITE_DEBUG + u8 bCorrupt; /* Shadow table corruption detected */ +#endif + int iDepth; /* Current depth of the r-tree structure */ + char *zDb; /* Name of database containing r-tree table */ + char *zName; /* Name of r-tree table */ + u32 nBusy; /* Current number of users of this structure */ + i64 nRowEst; /* Estimated number of rows in this table */ + u32 nCursor; /* Number of open cursors */ + u32 nNodeRef; /* Number RtreeNodes with positive nRef */ + char *zReadAuxSql; /* SQL for statement to read aux data */ + + /* List of nodes removed during a CondenseTree operation. List is + ** linked together via the pointer normally used for hash chains - + ** RtreeNode.pNext. RtreeNode.iNode stores the depth of the sub-tree + ** headed by the node (leaf nodes have RtreeNode.iNode==0). + */ + RtreeNode *pDeleted; + int iReinsertHeight; /* Height of sub-trees Reinsert() has run on */ + + /* Blob I/O on xxx_node */ + sqlite3_blob *pNodeBlob; + + /* Statements to read/write/delete a record from xxx_node */ + sqlite3_stmt *pWriteNode; + sqlite3_stmt *pDeleteNode; + + /* Statements to read/write/delete a record from xxx_rowid */ + sqlite3_stmt *pReadRowid; + sqlite3_stmt *pWriteRowid; + sqlite3_stmt *pDeleteRowid; + + /* Statements to read/write/delete a record from xxx_parent */ + sqlite3_stmt *pReadParent; + sqlite3_stmt *pWriteParent; + sqlite3_stmt *pDeleteParent; + + /* Statement for writing to the "aux:" fields, if there are any */ + sqlite3_stmt *pWriteAux; + + RtreeNode *aHash[HASHSIZE]; /* Hash table of in-memory nodes. */ +}; + +/* Possible values for Rtree.eCoordType: */ +#define RTREE_COORD_REAL32 0 +#define RTREE_COORD_INT32 1 + +/* +** If SQLITE_RTREE_INT_ONLY is defined, then this virtual table will +** only deal with integer coordinates. No floating point operations +** will be done. +*/ +#ifdef SQLITE_RTREE_INT_ONLY + typedef sqlite3_int64 RtreeDValue; /* High accuracy coordinate */ + typedef int RtreeValue; /* Low accuracy coordinate */ +# define RTREE_ZERO 0 +#else + typedef double RtreeDValue; /* High accuracy coordinate */ + typedef float RtreeValue; /* Low accuracy coordinate */ +# define RTREE_ZERO 0.0 +#endif + +/* +** Set the Rtree.bCorrupt flag +*/ +#ifdef SQLITE_DEBUG +# define RTREE_IS_CORRUPT(X) ((X)->bCorrupt = 1) +#else +# define RTREE_IS_CORRUPT(X) +#endif + +/* +** When doing a search of an r-tree, instances of the following structure +** record intermediate results from the tree walk. +** +** The id is always a node-id. For iLevel>=1 the id is the node-id of +** the node that the RtreeSearchPoint represents. When iLevel==0, however, +** the id is of the parent node and the cell that RtreeSearchPoint +** represents is the iCell-th entry in the parent node. +*/ +struct RtreeSearchPoint { + RtreeDValue rScore; /* The score for this node. Smallest goes first. */ + sqlite3_int64 id; /* Node ID */ + u8 iLevel; /* 0=entries. 1=leaf node. 2+ for higher */ + u8 eWithin; /* PARTLY_WITHIN or FULLY_WITHIN */ + u8 iCell; /* Cell index within the node */ +}; + +/* +** The minimum number of cells allowed for a node is a third of the +** maximum. In Gutman's notation: +** +** m = M/3 +** +** If an R*-tree "Reinsert" operation is required, the same number of +** cells are removed from the overfull node and reinserted into the tree. +*/ +#define RTREE_MINCELLS(p) ((((p)->iNodeSize-4)/(p)->nBytesPerCell)/3) +#define RTREE_REINSERT(p) RTREE_MINCELLS(p) +#define RTREE_MAXCELLS 51 + +/* +** The smallest possible node-size is (512-64)==448 bytes. And the largest +** supported cell size is 48 bytes (8 byte rowid + ten 4 byte coordinates). +** Therefore all non-root nodes must contain at least 3 entries. Since +** 3^40 is greater than 2^64, an r-tree structure always has a depth of +** 40 or less. +*/ +#define RTREE_MAX_DEPTH 40 + + +/* +** Number of entries in the cursor RtreeNode cache. The first entry is +** used to cache the RtreeNode for RtreeCursor.sPoint. The remaining +** entries cache the RtreeNode for the first elements of the priority queue. +*/ +#define RTREE_CACHE_SZ 5 + +/* +** An rtree cursor object. +*/ +struct RtreeCursor { + sqlite3_vtab_cursor base; /* Base class. Must be first */ + u8 atEOF; /* True if at end of search */ + u8 bPoint; /* True if sPoint is valid */ + u8 bAuxValid; /* True if pReadAux is valid */ + int iStrategy; /* Copy of idxNum search parameter */ + int nConstraint; /* Number of entries in aConstraint */ + RtreeConstraint *aConstraint; /* Search constraints. */ + int nPointAlloc; /* Number of slots allocated for aPoint[] */ + int nPoint; /* Number of slots used in aPoint[] */ + int mxLevel; /* iLevel value for root of the tree */ + RtreeSearchPoint *aPoint; /* Priority queue for search points */ + sqlite3_stmt *pReadAux; /* Statement to read aux-data */ + RtreeSearchPoint sPoint; /* Cached next search point */ + RtreeNode *aNode[RTREE_CACHE_SZ]; /* Rtree node cache */ + u32 anQueue[RTREE_MAX_DEPTH+1]; /* Number of queued entries by iLevel */ +}; + +/* Return the Rtree of a RtreeCursor */ +#define RTREE_OF_CURSOR(X) ((Rtree*)((X)->base.pVtab)) + +/* +** A coordinate can be either a floating point number or a integer. All +** coordinates within a single R-Tree are always of the same time. +*/ +union RtreeCoord { + RtreeValue f; /* Floating point value */ + int i; /* Integer value */ + u32 u; /* Unsigned for byte-order conversions */ +}; + +/* +** The argument is an RtreeCoord. Return the value stored within the RtreeCoord +** formatted as a RtreeDValue (double or int64). This macro assumes that local +** variable pRtree points to the Rtree structure associated with the +** RtreeCoord. +*/ +#ifdef SQLITE_RTREE_INT_ONLY +# define DCOORD(coord) ((RtreeDValue)coord.i) +#else +# define DCOORD(coord) ( \ + (pRtree->eCoordType==RTREE_COORD_REAL32) ? \ + ((double)coord.f) : \ + ((double)coord.i) \ + ) +#endif + +/* +** A search constraint. +*/ +struct RtreeConstraint { + int iCoord; /* Index of constrained coordinate */ + int op; /* Constraining operation */ + union { + RtreeDValue rValue; /* Constraint value. */ + int (*xGeom)(sqlite3_rtree_geometry*,int,RtreeDValue*,int*); + int (*xQueryFunc)(sqlite3_rtree_query_info*); + } u; + sqlite3_rtree_query_info *pInfo; /* xGeom and xQueryFunc argument */ +}; + +/* Possible values for RtreeConstraint.op */ +#define RTREE_EQ 0x41 /* A */ +#define RTREE_LE 0x42 /* B */ +#define RTREE_LT 0x43 /* C */ +#define RTREE_GE 0x44 /* D */ +#define RTREE_GT 0x45 /* E */ +#define RTREE_MATCH 0x46 /* F: Old-style sqlite3_rtree_geometry_callback() */ +#define RTREE_QUERY 0x47 /* G: New-style sqlite3_rtree_query_callback() */ + + +/* +** An rtree structure node. +*/ +struct RtreeNode { + RtreeNode *pParent; /* Parent node */ + i64 iNode; /* The node number */ + int nRef; /* Number of references to this node */ + int isDirty; /* True if the node needs to be written to disk */ + u8 *zData; /* Content of the node, as should be on disk */ + RtreeNode *pNext; /* Next node in this hash collision chain */ +}; + +/* Return the number of cells in a node */ +#define NCELL(pNode) readInt16(&(pNode)->zData[2]) + +/* +** A single cell from a node, deserialized +*/ +struct RtreeCell { + i64 iRowid; /* Node or entry ID */ + RtreeCoord aCoord[RTREE_MAX_DIMENSIONS*2]; /* Bounding box coordinates */ +}; + + +/* +** This object becomes the sqlite3_user_data() for the SQL functions +** that are created by sqlite3_rtree_geometry_callback() and +** sqlite3_rtree_query_callback() and which appear on the right of MATCH +** operators in order to constrain a search. +** +** xGeom and xQueryFunc are the callback functions. Exactly one of +** xGeom and xQueryFunc fields is non-NULL, depending on whether the +** SQL function was created using sqlite3_rtree_geometry_callback() or +** sqlite3_rtree_query_callback(). +** +** This object is deleted automatically by the destructor mechanism in +** sqlite3_create_function_v2(). +*/ +struct RtreeGeomCallback { + int (*xGeom)(sqlite3_rtree_geometry*, int, RtreeDValue*, int*); + int (*xQueryFunc)(sqlite3_rtree_query_info*); + void (*xDestructor)(void*); + void *pContext; +}; + +/* +** An instance of this structure (in the form of a BLOB) is returned by +** the SQL functions that sqlite3_rtree_geometry_callback() and +** sqlite3_rtree_query_callback() create, and is read as the right-hand +** operand to the MATCH operator of an R-Tree. +*/ +struct RtreeMatchArg { + u32 iSize; /* Size of this object */ + RtreeGeomCallback cb; /* Info about the callback functions */ + int nParam; /* Number of parameters to the SQL function */ + sqlite3_value **apSqlParam; /* Original SQL parameter values */ + RtreeDValue aParam[1]; /* Values for parameters to the SQL function */ +}; + +#ifndef MAX +# define MAX(x,y) ((x) < (y) ? (y) : (x)) +#endif +#ifndef MIN +# define MIN(x,y) ((x) > (y) ? (y) : (x)) +#endif + +/* What version of GCC is being used. 0 means GCC is not being used . +** Note that the GCC_VERSION macro will also be set correctly when using +** clang, since clang works hard to be gcc compatible. So the gcc +** optimizations will also work when compiling with clang. +*/ +#ifndef GCC_VERSION +#if defined(__GNUC__) && !defined(SQLITE_DISABLE_INTRINSIC) +# define GCC_VERSION (__GNUC__*1000000+__GNUC_MINOR__*1000+__GNUC_PATCHLEVEL__) +#else +# define GCC_VERSION 0 +#endif +#endif + +/* The testcase() macro should already be defined in the amalgamation. If +** it is not, make it a no-op. +*/ +#ifndef SQLITE_AMALGAMATION +# define testcase(X) +#endif + +/* +** Macros to determine whether the machine is big or little endian, +** and whether or not that determination is run-time or compile-time. +** +** For best performance, an attempt is made to guess at the byte-order +** using C-preprocessor macros. If that is unsuccessful, or if +** -DSQLITE_RUNTIME_BYTEORDER=1 is set, then byte-order is determined +** at run-time. +*/ +#ifndef SQLITE_BYTEORDER +#if defined(i386) || defined(__i386__) || defined(_M_IX86) || \ + defined(__x86_64) || defined(__x86_64__) || defined(_M_X64) || \ + defined(_M_AMD64) || defined(_M_ARM) || defined(__x86) || \ + defined(__arm__) +# define SQLITE_BYTEORDER 1234 +#elif defined(sparc) || defined(__ppc__) +# define SQLITE_BYTEORDER 4321 +#else +# define SQLITE_BYTEORDER 0 /* 0 means "unknown at compile-time" */ +#endif +#endif + + +/* What version of MSVC is being used. 0 means MSVC is not being used */ +#ifndef MSVC_VERSION +#if defined(_MSC_VER) && !defined(SQLITE_DISABLE_INTRINSIC) +# define MSVC_VERSION _MSC_VER +#else +# define MSVC_VERSION 0 +#endif +#endif + +/* +** Functions to deserialize a 16 bit integer, 32 bit real number and +** 64 bit integer. The deserialized value is returned. +*/ +static int readInt16(u8 *p){ + return (p[0]<<8) + p[1]; +} +static void readCoord(u8 *p, RtreeCoord *pCoord){ + assert( ((((char*)p) - (char*)0)&3)==0 ); /* p is always 4-byte aligned */ +#if SQLITE_BYTEORDER==1234 && MSVC_VERSION>=1300 + pCoord->u = _byteswap_ulong(*(u32*)p); +#elif SQLITE_BYTEORDER==1234 && GCC_VERSION>=4003000 + pCoord->u = __builtin_bswap32(*(u32*)p); +#elif SQLITE_BYTEORDER==4321 + pCoord->u = *(u32*)p; +#else + pCoord->u = ( + (((u32)p[0]) << 24) + + (((u32)p[1]) << 16) + + (((u32)p[2]) << 8) + + (((u32)p[3]) << 0) + ); +#endif +} +static i64 readInt64(u8 *p){ +#if SQLITE_BYTEORDER==1234 && MSVC_VERSION>=1300 + u64 x; + memcpy(&x, p, 8); + return (i64)_byteswap_uint64(x); +#elif SQLITE_BYTEORDER==1234 && GCC_VERSION>=4003000 + u64 x; + memcpy(&x, p, 8); + return (i64)__builtin_bswap64(x); +#elif SQLITE_BYTEORDER==4321 + i64 x; + memcpy(&x, p, 8); + return x; +#else + return (i64)( + (((u64)p[0]) << 56) + + (((u64)p[1]) << 48) + + (((u64)p[2]) << 40) + + (((u64)p[3]) << 32) + + (((u64)p[4]) << 24) + + (((u64)p[5]) << 16) + + (((u64)p[6]) << 8) + + (((u64)p[7]) << 0) + ); +#endif +} + +/* +** Functions to serialize a 16 bit integer, 32 bit real number and +** 64 bit integer. The value returned is the number of bytes written +** to the argument buffer (always 2, 4 and 8 respectively). +*/ +static void writeInt16(u8 *p, int i){ + p[0] = (i>> 8)&0xFF; + p[1] = (i>> 0)&0xFF; +} +static int writeCoord(u8 *p, RtreeCoord *pCoord){ + u32 i; + assert( ((((char*)p) - (char*)0)&3)==0 ); /* p is always 4-byte aligned */ + assert( sizeof(RtreeCoord)==4 ); + assert( sizeof(u32)==4 ); +#if SQLITE_BYTEORDER==1234 && GCC_VERSION>=4003000 + i = __builtin_bswap32(pCoord->u); + memcpy(p, &i, 4); +#elif SQLITE_BYTEORDER==1234 && MSVC_VERSION>=1300 + i = _byteswap_ulong(pCoord->u); + memcpy(p, &i, 4); +#elif SQLITE_BYTEORDER==4321 + i = pCoord->u; + memcpy(p, &i, 4); +#else + i = pCoord->u; + p[0] = (i>>24)&0xFF; + p[1] = (i>>16)&0xFF; + p[2] = (i>> 8)&0xFF; + p[3] = (i>> 0)&0xFF; +#endif + return 4; +} +static int writeInt64(u8 *p, i64 i){ +#if SQLITE_BYTEORDER==1234 && GCC_VERSION>=4003000 + i = (i64)__builtin_bswap64((u64)i); + memcpy(p, &i, 8); +#elif SQLITE_BYTEORDER==1234 && MSVC_VERSION>=1300 + i = (i64)_byteswap_uint64((u64)i); + memcpy(p, &i, 8); +#elif SQLITE_BYTEORDER==4321 + memcpy(p, &i, 8); +#else + p[0] = (i>>56)&0xFF; + p[1] = (i>>48)&0xFF; + p[2] = (i>>40)&0xFF; + p[3] = (i>>32)&0xFF; + p[4] = (i>>24)&0xFF; + p[5] = (i>>16)&0xFF; + p[6] = (i>> 8)&0xFF; + p[7] = (i>> 0)&0xFF; +#endif + return 8; +} + +/* +** Increment the reference count of node p. +*/ +static void nodeReference(RtreeNode *p){ + if( p ){ + assert( p->nRef>0 ); + p->nRef++; + } +} + +/* +** Clear the content of node p (set all bytes to 0x00). +*/ +static void nodeZero(Rtree *pRtree, RtreeNode *p){ + memset(&p->zData[2], 0, pRtree->iNodeSize-2); + p->isDirty = 1; +} + +/* +** Given a node number iNode, return the corresponding key to use +** in the Rtree.aHash table. +*/ +static unsigned int nodeHash(i64 iNode){ + return ((unsigned)iNode) % HASHSIZE; +} + +/* +** Search the node hash table for node iNode. If found, return a pointer +** to it. Otherwise, return 0. +*/ +static RtreeNode *nodeHashLookup(Rtree *pRtree, i64 iNode){ + RtreeNode *p; + for(p=pRtree->aHash[nodeHash(iNode)]; p && p->iNode!=iNode; p=p->pNext); + return p; +} + +/* +** Add node pNode to the node hash table. +*/ +static void nodeHashInsert(Rtree *pRtree, RtreeNode *pNode){ + int iHash; + assert( pNode->pNext==0 ); + iHash = nodeHash(pNode->iNode); + pNode->pNext = pRtree->aHash[iHash]; + pRtree->aHash[iHash] = pNode; +} + +/* +** Remove node pNode from the node hash table. +*/ +static void nodeHashDelete(Rtree *pRtree, RtreeNode *pNode){ + RtreeNode **pp; + if( pNode->iNode!=0 ){ + pp = &pRtree->aHash[nodeHash(pNode->iNode)]; + for( ; (*pp)!=pNode; pp = &(*pp)->pNext){ assert(*pp); } + *pp = pNode->pNext; + pNode->pNext = 0; + } +} + +/* +** Allocate and return new r-tree node. Initially, (RtreeNode.iNode==0), +** indicating that node has not yet been assigned a node number. It is +** assigned a node number when nodeWrite() is called to write the +** node contents out to the database. +*/ +static RtreeNode *nodeNew(Rtree *pRtree, RtreeNode *pParent){ + RtreeNode *pNode; + pNode = (RtreeNode *)sqlite3_malloc64(sizeof(RtreeNode) + pRtree->iNodeSize); + if( pNode ){ + memset(pNode, 0, sizeof(RtreeNode) + pRtree->iNodeSize); + pNode->zData = (u8 *)&pNode[1]; + pNode->nRef = 1; + pRtree->nNodeRef++; + pNode->pParent = pParent; + pNode->isDirty = 1; + nodeReference(pParent); + } + return pNode; +} + +/* +** Clear the Rtree.pNodeBlob object +*/ +static void nodeBlobReset(Rtree *pRtree){ + if( pRtree->pNodeBlob && pRtree->inWrTrans==0 && pRtree->nCursor==0 ){ + sqlite3_blob *pBlob = pRtree->pNodeBlob; + pRtree->pNodeBlob = 0; + sqlite3_blob_close(pBlob); + } +} + +/* +** Check to see if pNode is the same as pParent or any of the parents +** of pParent. +*/ +static int nodeInParentChain(const RtreeNode *pNode, const RtreeNode *pParent){ + do{ + if( pNode==pParent ) return 1; + pParent = pParent->pParent; + }while( pParent ); + return 0; +} + +/* +** Obtain a reference to an r-tree node. +*/ +static int nodeAcquire( + Rtree *pRtree, /* R-tree structure */ + i64 iNode, /* Node number to load */ + RtreeNode *pParent, /* Either the parent node or NULL */ + RtreeNode **ppNode /* OUT: Acquired node */ +){ + int rc = SQLITE_OK; + RtreeNode *pNode = 0; + + /* Check if the requested node is already in the hash table. If so, + ** increase its reference count and return it. + */ + if( (pNode = nodeHashLookup(pRtree, iNode))!=0 ){ + if( pParent && !pNode->pParent ){ + if( nodeInParentChain(pNode, pParent) ){ + RTREE_IS_CORRUPT(pRtree); + return SQLITE_CORRUPT_VTAB; + } + pParent->nRef++; + pNode->pParent = pParent; + }else if( pParent && pNode->pParent && pParent!=pNode->pParent ){ + RTREE_IS_CORRUPT(pRtree); + return SQLITE_CORRUPT_VTAB; + } + pNode->nRef++; + *ppNode = pNode; + return SQLITE_OK; + } + + if( pRtree->pNodeBlob ){ + sqlite3_blob *pBlob = pRtree->pNodeBlob; + pRtree->pNodeBlob = 0; + rc = sqlite3_blob_reopen(pBlob, iNode); + pRtree->pNodeBlob = pBlob; + if( rc ){ + nodeBlobReset(pRtree); + if( rc==SQLITE_NOMEM ) return SQLITE_NOMEM; + } + } + if( pRtree->pNodeBlob==0 ){ + char *zTab = sqlite3_mprintf("%s_node", pRtree->zName); + if( zTab==0 ) return SQLITE_NOMEM; + rc = sqlite3_blob_open(pRtree->db, pRtree->zDb, zTab, "data", iNode, 0, + &pRtree->pNodeBlob); + sqlite3_free(zTab); + } + if( rc ){ + nodeBlobReset(pRtree); + *ppNode = 0; + /* If unable to open an sqlite3_blob on the desired row, that can only + ** be because the shadow tables hold erroneous data. */ + if( rc==SQLITE_ERROR ){ + rc = SQLITE_CORRUPT_VTAB; + RTREE_IS_CORRUPT(pRtree); + } + }else if( pRtree->iNodeSize==sqlite3_blob_bytes(pRtree->pNodeBlob) ){ + pNode = (RtreeNode *)sqlite3_malloc64(sizeof(RtreeNode)+pRtree->iNodeSize); + if( !pNode ){ + rc = SQLITE_NOMEM; + }else{ + pNode->pParent = pParent; + pNode->zData = (u8 *)&pNode[1]; + pNode->nRef = 1; + pRtree->nNodeRef++; + pNode->iNode = iNode; + pNode->isDirty = 0; + pNode->pNext = 0; + rc = sqlite3_blob_read(pRtree->pNodeBlob, pNode->zData, + pRtree->iNodeSize, 0); + } + } + + /* If the root node was just loaded, set pRtree->iDepth to the height + ** of the r-tree structure. A height of zero means all data is stored on + ** the root node. A height of one means the children of the root node + ** are the leaves, and so on. If the depth as specified on the root node + ** is greater than RTREE_MAX_DEPTH, the r-tree structure must be corrupt. + */ + if( pNode && iNode==1 ){ + pRtree->iDepth = readInt16(pNode->zData); + if( pRtree->iDepth>RTREE_MAX_DEPTH ){ + rc = SQLITE_CORRUPT_VTAB; + RTREE_IS_CORRUPT(pRtree); + } + } + + /* If no error has occurred so far, check if the "number of entries" + ** field on the node is too large. If so, set the return code to + ** SQLITE_CORRUPT_VTAB. + */ + if( pNode && rc==SQLITE_OK ){ + if( NCELL(pNode)>((pRtree->iNodeSize-4)/pRtree->nBytesPerCell) ){ + rc = SQLITE_CORRUPT_VTAB; + RTREE_IS_CORRUPT(pRtree); + } + } + + if( rc==SQLITE_OK ){ + if( pNode!=0 ){ + nodeReference(pParent); + nodeHashInsert(pRtree, pNode); + }else{ + rc = SQLITE_CORRUPT_VTAB; + RTREE_IS_CORRUPT(pRtree); + } + *ppNode = pNode; + }else{ + if( pNode ){ + pRtree->nNodeRef--; + sqlite3_free(pNode); + } + *ppNode = 0; + } + + return rc; +} + +/* +** Overwrite cell iCell of node pNode with the contents of pCell. +*/ +static void nodeOverwriteCell( + Rtree *pRtree, /* The overall R-Tree */ + RtreeNode *pNode, /* The node into which the cell is to be written */ + RtreeCell *pCell, /* The cell to write */ + int iCell /* Index into pNode into which pCell is written */ +){ + int ii; + u8 *p = &pNode->zData[4 + pRtree->nBytesPerCell*iCell]; + p += writeInt64(p, pCell->iRowid); + for(ii=0; iinDim2; ii++){ + p += writeCoord(p, &pCell->aCoord[ii]); + } + pNode->isDirty = 1; +} + +/* +** Remove the cell with index iCell from node pNode. +*/ +static void nodeDeleteCell(Rtree *pRtree, RtreeNode *pNode, int iCell){ + u8 *pDst = &pNode->zData[4 + pRtree->nBytesPerCell*iCell]; + u8 *pSrc = &pDst[pRtree->nBytesPerCell]; + int nByte = (NCELL(pNode) - iCell - 1) * pRtree->nBytesPerCell; + memmove(pDst, pSrc, nByte); + writeInt16(&pNode->zData[2], NCELL(pNode)-1); + pNode->isDirty = 1; +} + +/* +** Insert the contents of cell pCell into node pNode. If the insert +** is successful, return SQLITE_OK. +** +** If there is not enough free space in pNode, return SQLITE_FULL. +*/ +static int nodeInsertCell( + Rtree *pRtree, /* The overall R-Tree */ + RtreeNode *pNode, /* Write new cell into this node */ + RtreeCell *pCell /* The cell to be inserted */ +){ + int nCell; /* Current number of cells in pNode */ + int nMaxCell; /* Maximum number of cells for pNode */ + + nMaxCell = (pRtree->iNodeSize-4)/pRtree->nBytesPerCell; + nCell = NCELL(pNode); + + assert( nCell<=nMaxCell ); + if( nCellzData[2], nCell+1); + pNode->isDirty = 1; + } + + return (nCell==nMaxCell); +} + +/* +** If the node is dirty, write it out to the database. +*/ +static int nodeWrite(Rtree *pRtree, RtreeNode *pNode){ + int rc = SQLITE_OK; + if( pNode->isDirty ){ + sqlite3_stmt *p = pRtree->pWriteNode; + if( pNode->iNode ){ + sqlite3_bind_int64(p, 1, pNode->iNode); + }else{ + sqlite3_bind_null(p, 1); + } + sqlite3_bind_blob(p, 2, pNode->zData, pRtree->iNodeSize, SQLITE_STATIC); + sqlite3_step(p); + pNode->isDirty = 0; + rc = sqlite3_reset(p); + sqlite3_bind_null(p, 2); + if( pNode->iNode==0 && rc==SQLITE_OK ){ + pNode->iNode = sqlite3_last_insert_rowid(pRtree->db); + nodeHashInsert(pRtree, pNode); + } + } + return rc; +} + +/* +** Release a reference to a node. If the node is dirty and the reference +** count drops to zero, the node data is written to the database. +*/ +static int nodeRelease(Rtree *pRtree, RtreeNode *pNode){ + int rc = SQLITE_OK; + if( pNode ){ + assert( pNode->nRef>0 ); + assert( pRtree->nNodeRef>0 ); + pNode->nRef--; + if( pNode->nRef==0 ){ + pRtree->nNodeRef--; + if( pNode->iNode==1 ){ + pRtree->iDepth = -1; + } + if( pNode->pParent ){ + rc = nodeRelease(pRtree, pNode->pParent); + } + if( rc==SQLITE_OK ){ + rc = nodeWrite(pRtree, pNode); + } + nodeHashDelete(pRtree, pNode); + sqlite3_free(pNode); + } + } + return rc; +} + +/* +** Return the 64-bit integer value associated with cell iCell of +** node pNode. If pNode is a leaf node, this is a rowid. If it is +** an internal node, then the 64-bit integer is a child page number. +*/ +static i64 nodeGetRowid( + Rtree *pRtree, /* The overall R-Tree */ + RtreeNode *pNode, /* The node from which to extract the ID */ + int iCell /* The cell index from which to extract the ID */ +){ + assert( iCellzData[4 + pRtree->nBytesPerCell*iCell]); +} + +/* +** Return coordinate iCoord from cell iCell in node pNode. +*/ +static void nodeGetCoord( + Rtree *pRtree, /* The overall R-Tree */ + RtreeNode *pNode, /* The node from which to extract a coordinate */ + int iCell, /* The index of the cell within the node */ + int iCoord, /* Which coordinate to extract */ + RtreeCoord *pCoord /* OUT: Space to write result to */ +){ + readCoord(&pNode->zData[12 + pRtree->nBytesPerCell*iCell + 4*iCoord], pCoord); +} + +/* +** Deserialize cell iCell of node pNode. Populate the structure pointed +** to by pCell with the results. +*/ +static void nodeGetCell( + Rtree *pRtree, /* The overall R-Tree */ + RtreeNode *pNode, /* The node containing the cell to be read */ + int iCell, /* Index of the cell within the node */ + RtreeCell *pCell /* OUT: Write the cell contents here */ +){ + u8 *pData; + RtreeCoord *pCoord; + int ii = 0; + pCell->iRowid = nodeGetRowid(pRtree, pNode, iCell); + pData = pNode->zData + (12 + pRtree->nBytesPerCell*iCell); + pCoord = pCell->aCoord; + do{ + readCoord(pData, &pCoord[ii]); + readCoord(pData+4, &pCoord[ii+1]); + pData += 8; + ii += 2; + }while( iinDim2 ); +} + + +/* Forward declaration for the function that does the work of +** the virtual table module xCreate() and xConnect() methods. +*/ +static int rtreeInit( + sqlite3 *, void *, int, const char *const*, sqlite3_vtab **, char **, int +); + +/* +** Rtree virtual table module xCreate method. +*/ +static int rtreeCreate( + sqlite3 *db, + void *pAux, + int argc, const char *const*argv, + sqlite3_vtab **ppVtab, + char **pzErr +){ + return rtreeInit(db, pAux, argc, argv, ppVtab, pzErr, 1); +} + +/* +** Rtree virtual table module xConnect method. +*/ +static int rtreeConnect( + sqlite3 *db, + void *pAux, + int argc, const char *const*argv, + sqlite3_vtab **ppVtab, + char **pzErr +){ + return rtreeInit(db, pAux, argc, argv, ppVtab, pzErr, 0); +} + +/* +** Increment the r-tree reference count. +*/ +static void rtreeReference(Rtree *pRtree){ + pRtree->nBusy++; +} + +/* +** Decrement the r-tree reference count. When the reference count reaches +** zero the structure is deleted. +*/ +static void rtreeRelease(Rtree *pRtree){ + pRtree->nBusy--; + if( pRtree->nBusy==0 ){ + pRtree->inWrTrans = 0; + assert( pRtree->nCursor==0 ); + nodeBlobReset(pRtree); + assert( pRtree->nNodeRef==0 || pRtree->bCorrupt ); + sqlite3_finalize(pRtree->pWriteNode); + sqlite3_finalize(pRtree->pDeleteNode); + sqlite3_finalize(pRtree->pReadRowid); + sqlite3_finalize(pRtree->pWriteRowid); + sqlite3_finalize(pRtree->pDeleteRowid); + sqlite3_finalize(pRtree->pReadParent); + sqlite3_finalize(pRtree->pWriteParent); + sqlite3_finalize(pRtree->pDeleteParent); + sqlite3_finalize(pRtree->pWriteAux); + sqlite3_free(pRtree->zReadAuxSql); + sqlite3_free(pRtree); + } +} + +/* +** Rtree virtual table module xDisconnect method. +*/ +static int rtreeDisconnect(sqlite3_vtab *pVtab){ + rtreeRelease((Rtree *)pVtab); + return SQLITE_OK; +} + +/* +** Rtree virtual table module xDestroy method. +*/ +static int rtreeDestroy(sqlite3_vtab *pVtab){ + Rtree *pRtree = (Rtree *)pVtab; + int rc; + char *zCreate = sqlite3_mprintf( + "DROP TABLE '%q'.'%q_node';" + "DROP TABLE '%q'.'%q_rowid';" + "DROP TABLE '%q'.'%q_parent';", + pRtree->zDb, pRtree->zName, + pRtree->zDb, pRtree->zName, + pRtree->zDb, pRtree->zName + ); + if( !zCreate ){ + rc = SQLITE_NOMEM; + }else{ + nodeBlobReset(pRtree); + rc = sqlite3_exec(pRtree->db, zCreate, 0, 0, 0); + sqlite3_free(zCreate); + } + if( rc==SQLITE_OK ){ + rtreeRelease(pRtree); + } + + return rc; +} + +/* +** Rtree virtual table module xOpen method. +*/ +static int rtreeOpen(sqlite3_vtab *pVTab, sqlite3_vtab_cursor **ppCursor){ + int rc = SQLITE_NOMEM; + Rtree *pRtree = (Rtree *)pVTab; + RtreeCursor *pCsr; + + pCsr = (RtreeCursor *)sqlite3_malloc64(sizeof(RtreeCursor)); + if( pCsr ){ + memset(pCsr, 0, sizeof(RtreeCursor)); + pCsr->base.pVtab = pVTab; + rc = SQLITE_OK; + pRtree->nCursor++; + } + *ppCursor = (sqlite3_vtab_cursor *)pCsr; + + return rc; +} + + +/* +** Free the RtreeCursor.aConstraint[] array and its contents. +*/ +static void freeCursorConstraints(RtreeCursor *pCsr){ + if( pCsr->aConstraint ){ + int i; /* Used to iterate through constraint array */ + for(i=0; inConstraint; i++){ + sqlite3_rtree_query_info *pInfo = pCsr->aConstraint[i].pInfo; + if( pInfo ){ + if( pInfo->xDelUser ) pInfo->xDelUser(pInfo->pUser); + sqlite3_free(pInfo); + } + } + sqlite3_free(pCsr->aConstraint); + pCsr->aConstraint = 0; + } +} + +/* +** Rtree virtual table module xClose method. +*/ +static int rtreeClose(sqlite3_vtab_cursor *cur){ + Rtree *pRtree = (Rtree *)(cur->pVtab); + int ii; + RtreeCursor *pCsr = (RtreeCursor *)cur; + assert( pRtree->nCursor>0 ); + freeCursorConstraints(pCsr); + sqlite3_finalize(pCsr->pReadAux); + sqlite3_free(pCsr->aPoint); + for(ii=0; iiaNode[ii]); + sqlite3_free(pCsr); + pRtree->nCursor--; + nodeBlobReset(pRtree); + return SQLITE_OK; +} + +/* +** Rtree virtual table module xEof method. +** +** Return non-zero if the cursor does not currently point to a valid +** record (i.e if the scan has finished), or zero otherwise. +*/ +static int rtreeEof(sqlite3_vtab_cursor *cur){ + RtreeCursor *pCsr = (RtreeCursor *)cur; + return pCsr->atEOF; +} + +/* +** Convert raw bits from the on-disk RTree record into a coordinate value. +** The on-disk format is big-endian and needs to be converted for little- +** endian platforms. The on-disk record stores integer coordinates if +** eInt is true and it stores 32-bit floating point records if eInt is +** false. a[] is the four bytes of the on-disk record to be decoded. +** Store the results in "r". +** +** There are five versions of this macro. The last one is generic. The +** other four are various architectures-specific optimizations. +*/ +#if SQLITE_BYTEORDER==1234 && MSVC_VERSION>=1300 +#define RTREE_DECODE_COORD(eInt, a, r) { \ + RtreeCoord c; /* Coordinate decoded */ \ + c.u = _byteswap_ulong(*(u32*)a); \ + r = eInt ? (sqlite3_rtree_dbl)c.i : (sqlite3_rtree_dbl)c.f; \ +} +#elif SQLITE_BYTEORDER==1234 && GCC_VERSION>=4003000 +#define RTREE_DECODE_COORD(eInt, a, r) { \ + RtreeCoord c; /* Coordinate decoded */ \ + c.u = __builtin_bswap32(*(u32*)a); \ + r = eInt ? (sqlite3_rtree_dbl)c.i : (sqlite3_rtree_dbl)c.f; \ +} +#elif SQLITE_BYTEORDER==1234 +#define RTREE_DECODE_COORD(eInt, a, r) { \ + RtreeCoord c; /* Coordinate decoded */ \ + memcpy(&c.u,a,4); \ + c.u = ((c.u>>24)&0xff)|((c.u>>8)&0xff00)| \ + ((c.u&0xff)<<24)|((c.u&0xff00)<<8); \ + r = eInt ? (sqlite3_rtree_dbl)c.i : (sqlite3_rtree_dbl)c.f; \ +} +#elif SQLITE_BYTEORDER==4321 +#define RTREE_DECODE_COORD(eInt, a, r) { \ + RtreeCoord c; /* Coordinate decoded */ \ + memcpy(&c.u,a,4); \ + r = eInt ? (sqlite3_rtree_dbl)c.i : (sqlite3_rtree_dbl)c.f; \ +} +#else +#define RTREE_DECODE_COORD(eInt, a, r) { \ + RtreeCoord c; /* Coordinate decoded */ \ + c.u = ((u32)a[0]<<24) + ((u32)a[1]<<16) \ + +((u32)a[2]<<8) + a[3]; \ + r = eInt ? (sqlite3_rtree_dbl)c.i : (sqlite3_rtree_dbl)c.f; \ +} +#endif + +/* +** Check the RTree node or entry given by pCellData and p against the MATCH +** constraint pConstraint. +*/ +static int rtreeCallbackConstraint( + RtreeConstraint *pConstraint, /* The constraint to test */ + int eInt, /* True if RTree holding integer coordinates */ + u8 *pCellData, /* Raw cell content */ + RtreeSearchPoint *pSearch, /* Container of this cell */ + sqlite3_rtree_dbl *prScore, /* OUT: score for the cell */ + int *peWithin /* OUT: visibility of the cell */ +){ + sqlite3_rtree_query_info *pInfo = pConstraint->pInfo; /* Callback info */ + int nCoord = pInfo->nCoord; /* No. of coordinates */ + int rc; /* Callback return code */ + RtreeCoord c; /* Translator union */ + sqlite3_rtree_dbl aCoord[RTREE_MAX_DIMENSIONS*2]; /* Decoded coordinates */ + + assert( pConstraint->op==RTREE_MATCH || pConstraint->op==RTREE_QUERY ); + assert( nCoord==2 || nCoord==4 || nCoord==6 || nCoord==8 || nCoord==10 ); + + if( pConstraint->op==RTREE_QUERY && pSearch->iLevel==1 ){ + pInfo->iRowid = readInt64(pCellData); + } + pCellData += 8; +#ifndef SQLITE_RTREE_INT_ONLY + if( eInt==0 ){ + switch( nCoord ){ + case 10: readCoord(pCellData+36, &c); aCoord[9] = c.f; + readCoord(pCellData+32, &c); aCoord[8] = c.f; + case 8: readCoord(pCellData+28, &c); aCoord[7] = c.f; + readCoord(pCellData+24, &c); aCoord[6] = c.f; + case 6: readCoord(pCellData+20, &c); aCoord[5] = c.f; + readCoord(pCellData+16, &c); aCoord[4] = c.f; + case 4: readCoord(pCellData+12, &c); aCoord[3] = c.f; + readCoord(pCellData+8, &c); aCoord[2] = c.f; + default: readCoord(pCellData+4, &c); aCoord[1] = c.f; + readCoord(pCellData, &c); aCoord[0] = c.f; + } + }else +#endif + { + switch( nCoord ){ + case 10: readCoord(pCellData+36, &c); aCoord[9] = c.i; + readCoord(pCellData+32, &c); aCoord[8] = c.i; + case 8: readCoord(pCellData+28, &c); aCoord[7] = c.i; + readCoord(pCellData+24, &c); aCoord[6] = c.i; + case 6: readCoord(pCellData+20, &c); aCoord[5] = c.i; + readCoord(pCellData+16, &c); aCoord[4] = c.i; + case 4: readCoord(pCellData+12, &c); aCoord[3] = c.i; + readCoord(pCellData+8, &c); aCoord[2] = c.i; + default: readCoord(pCellData+4, &c); aCoord[1] = c.i; + readCoord(pCellData, &c); aCoord[0] = c.i; + } + } + if( pConstraint->op==RTREE_MATCH ){ + int eWithin = 0; + rc = pConstraint->u.xGeom((sqlite3_rtree_geometry*)pInfo, + nCoord, aCoord, &eWithin); + if( eWithin==0 ) *peWithin = NOT_WITHIN; + *prScore = RTREE_ZERO; + }else{ + pInfo->aCoord = aCoord; + pInfo->iLevel = pSearch->iLevel - 1; + pInfo->rScore = pInfo->rParentScore = pSearch->rScore; + pInfo->eWithin = pInfo->eParentWithin = pSearch->eWithin; + rc = pConstraint->u.xQueryFunc(pInfo); + if( pInfo->eWithin<*peWithin ) *peWithin = pInfo->eWithin; + if( pInfo->rScore<*prScore || *prScorerScore; + } + } + return rc; +} + +/* +** Check the internal RTree node given by pCellData against constraint p. +** If this constraint cannot be satisfied by any child within the node, +** set *peWithin to NOT_WITHIN. +*/ +static void rtreeNonleafConstraint( + RtreeConstraint *p, /* The constraint to test */ + int eInt, /* True if RTree holds integer coordinates */ + u8 *pCellData, /* Raw cell content as appears on disk */ + int *peWithin /* Adjust downward, as appropriate */ +){ + sqlite3_rtree_dbl val; /* Coordinate value convert to a double */ + + /* p->iCoord might point to either a lower or upper bound coordinate + ** in a coordinate pair. But make pCellData point to the lower bound. + */ + pCellData += 8 + 4*(p->iCoord&0xfe); + + assert(p->op==RTREE_LE || p->op==RTREE_LT || p->op==RTREE_GE + || p->op==RTREE_GT || p->op==RTREE_EQ ); + assert( ((((char*)pCellData) - (char*)0)&3)==0 ); /* 4-byte aligned */ + switch( p->op ){ + case RTREE_LE: + case RTREE_LT: + case RTREE_EQ: + RTREE_DECODE_COORD(eInt, pCellData, val); + /* val now holds the lower bound of the coordinate pair */ + if( p->u.rValue>=val ) return; + if( p->op!=RTREE_EQ ) break; /* RTREE_LE and RTREE_LT end here */ + /* Fall through for the RTREE_EQ case */ + + default: /* RTREE_GT or RTREE_GE, or fallthrough of RTREE_EQ */ + pCellData += 4; + RTREE_DECODE_COORD(eInt, pCellData, val); + /* val now holds the upper bound of the coordinate pair */ + if( p->u.rValue<=val ) return; + } + *peWithin = NOT_WITHIN; +} + +/* +** Check the leaf RTree cell given by pCellData against constraint p. +** If this constraint is not satisfied, set *peWithin to NOT_WITHIN. +** If the constraint is satisfied, leave *peWithin unchanged. +** +** The constraint is of the form: xN op $val +** +** The op is given by p->op. The xN is p->iCoord-th coordinate in +** pCellData. $val is given by p->u.rValue. +*/ +static void rtreeLeafConstraint( + RtreeConstraint *p, /* The constraint to test */ + int eInt, /* True if RTree holds integer coordinates */ + u8 *pCellData, /* Raw cell content as appears on disk */ + int *peWithin /* Adjust downward, as appropriate */ +){ + RtreeDValue xN; /* Coordinate value converted to a double */ + + assert(p->op==RTREE_LE || p->op==RTREE_LT || p->op==RTREE_GE + || p->op==RTREE_GT || p->op==RTREE_EQ ); + pCellData += 8 + p->iCoord*4; + assert( ((((char*)pCellData) - (char*)0)&3)==0 ); /* 4-byte aligned */ + RTREE_DECODE_COORD(eInt, pCellData, xN); + switch( p->op ){ + case RTREE_LE: if( xN <= p->u.rValue ) return; break; + case RTREE_LT: if( xN < p->u.rValue ) return; break; + case RTREE_GE: if( xN >= p->u.rValue ) return; break; + case RTREE_GT: if( xN > p->u.rValue ) return; break; + default: if( xN == p->u.rValue ) return; break; + } + *peWithin = NOT_WITHIN; +} + +/* +** One of the cells in node pNode is guaranteed to have a 64-bit +** integer value equal to iRowid. Return the index of this cell. +*/ +static int nodeRowidIndex( + Rtree *pRtree, + RtreeNode *pNode, + i64 iRowid, + int *piIndex +){ + int ii; + int nCell = NCELL(pNode); + assert( nCell<200 ); + for(ii=0; iipParent; + if( pParent ){ + return nodeRowidIndex(pRtree, pParent, pNode->iNode, piIndex); + } + *piIndex = -1; + return SQLITE_OK; +} + +/* +** Compare two search points. Return negative, zero, or positive if the first +** is less than, equal to, or greater than the second. +** +** The rScore is the primary key. Smaller rScore values come first. +** If the rScore is a tie, then use iLevel as the tie breaker with smaller +** iLevel values coming first. In this way, if rScore is the same for all +** SearchPoints, then iLevel becomes the deciding factor and the result +** is a depth-first search, which is the desired default behavior. +*/ +static int rtreeSearchPointCompare( + const RtreeSearchPoint *pA, + const RtreeSearchPoint *pB +){ + if( pA->rScorerScore ) return -1; + if( pA->rScore>pB->rScore ) return +1; + if( pA->iLeveliLevel ) return -1; + if( pA->iLevel>pB->iLevel ) return +1; + return 0; +} + +/* +** Interchange two search points in a cursor. +*/ +static void rtreeSearchPointSwap(RtreeCursor *p, int i, int j){ + RtreeSearchPoint t = p->aPoint[i]; + assert( iaPoint[i] = p->aPoint[j]; + p->aPoint[j] = t; + i++; j++; + if( i=RTREE_CACHE_SZ ){ + nodeRelease(RTREE_OF_CURSOR(p), p->aNode[i]); + p->aNode[i] = 0; + }else{ + RtreeNode *pTemp = p->aNode[i]; + p->aNode[i] = p->aNode[j]; + p->aNode[j] = pTemp; + } + } +} + +/* +** Return the search point with the lowest current score. +*/ +static RtreeSearchPoint *rtreeSearchPointFirst(RtreeCursor *pCur){ + return pCur->bPoint ? &pCur->sPoint : pCur->nPoint ? pCur->aPoint : 0; +} + +/* +** Get the RtreeNode for the search point with the lowest score. +*/ +static RtreeNode *rtreeNodeOfFirstSearchPoint(RtreeCursor *pCur, int *pRC){ + sqlite3_int64 id; + int ii = 1 - pCur->bPoint; + assert( ii==0 || ii==1 ); + assert( pCur->bPoint || pCur->nPoint ); + if( pCur->aNode[ii]==0 ){ + assert( pRC!=0 ); + id = ii ? pCur->aPoint[0].id : pCur->sPoint.id; + *pRC = nodeAcquire(RTREE_OF_CURSOR(pCur), id, 0, &pCur->aNode[ii]); + } + return pCur->aNode[ii]; +} + +/* +** Push a new element onto the priority queue +*/ +static RtreeSearchPoint *rtreeEnqueue( + RtreeCursor *pCur, /* The cursor */ + RtreeDValue rScore, /* Score for the new search point */ + u8 iLevel /* Level for the new search point */ +){ + int i, j; + RtreeSearchPoint *pNew; + if( pCur->nPoint>=pCur->nPointAlloc ){ + int nNew = pCur->nPointAlloc*2 + 8; + pNew = sqlite3_realloc64(pCur->aPoint, nNew*sizeof(pCur->aPoint[0])); + if( pNew==0 ) return 0; + pCur->aPoint = pNew; + pCur->nPointAlloc = nNew; + } + i = pCur->nPoint++; + pNew = pCur->aPoint + i; + pNew->rScore = rScore; + pNew->iLevel = iLevel; + assert( iLevel<=RTREE_MAX_DEPTH ); + while( i>0 ){ + RtreeSearchPoint *pParent; + j = (i-1)/2; + pParent = pCur->aPoint + j; + if( rtreeSearchPointCompare(pNew, pParent)>=0 ) break; + rtreeSearchPointSwap(pCur, j, i); + i = j; + pNew = pParent; + } + return pNew; +} + +/* +** Allocate a new RtreeSearchPoint and return a pointer to it. Return +** NULL if malloc fails. +*/ +static RtreeSearchPoint *rtreeSearchPointNew( + RtreeCursor *pCur, /* The cursor */ + RtreeDValue rScore, /* Score for the new search point */ + u8 iLevel /* Level for the new search point */ +){ + RtreeSearchPoint *pNew, *pFirst; + pFirst = rtreeSearchPointFirst(pCur); + pCur->anQueue[iLevel]++; + if( pFirst==0 + || pFirst->rScore>rScore + || (pFirst->rScore==rScore && pFirst->iLevel>iLevel) + ){ + if( pCur->bPoint ){ + int ii; + pNew = rtreeEnqueue(pCur, rScore, iLevel); + if( pNew==0 ) return 0; + ii = (int)(pNew - pCur->aPoint) + 1; + if( iiaNode[ii]==0 ); + pCur->aNode[ii] = pCur->aNode[0]; + }else{ + nodeRelease(RTREE_OF_CURSOR(pCur), pCur->aNode[0]); + } + pCur->aNode[0] = 0; + *pNew = pCur->sPoint; + } + pCur->sPoint.rScore = rScore; + pCur->sPoint.iLevel = iLevel; + pCur->bPoint = 1; + return &pCur->sPoint; + }else{ + return rtreeEnqueue(pCur, rScore, iLevel); + } +} + +#if 0 +/* Tracing routines for the RtreeSearchPoint queue */ +static void tracePoint(RtreeSearchPoint *p, int idx, RtreeCursor *pCur){ + if( idx<0 ){ printf(" s"); }else{ printf("%2d", idx); } + printf(" %d.%05lld.%02d %g %d", + p->iLevel, p->id, p->iCell, p->rScore, p->eWithin + ); + idx++; + if( idxaNode[idx]); + }else{ + printf("\n"); + } +} +static void traceQueue(RtreeCursor *pCur, const char *zPrefix){ + int ii; + printf("=== %9s ", zPrefix); + if( pCur->bPoint ){ + tracePoint(&pCur->sPoint, -1, pCur); + } + for(ii=0; iinPoint; ii++){ + if( ii>0 || pCur->bPoint ) printf(" "); + tracePoint(&pCur->aPoint[ii], ii, pCur); + } +} +# define RTREE_QUEUE_TRACE(A,B) traceQueue(A,B) +#else +# define RTREE_QUEUE_TRACE(A,B) /* no-op */ +#endif + +/* Remove the search point with the lowest current score. +*/ +static void rtreeSearchPointPop(RtreeCursor *p){ + int i, j, k, n; + i = 1 - p->bPoint; + assert( i==0 || i==1 ); + if( p->aNode[i] ){ + nodeRelease(RTREE_OF_CURSOR(p), p->aNode[i]); + p->aNode[i] = 0; + } + if( p->bPoint ){ + p->anQueue[p->sPoint.iLevel]--; + p->bPoint = 0; + }else if( p->nPoint ){ + p->anQueue[p->aPoint[0].iLevel]--; + n = --p->nPoint; + p->aPoint[0] = p->aPoint[n]; + if( naNode[1] = p->aNode[n+1]; + p->aNode[n+1] = 0; + } + i = 0; + while( (j = i*2+1)aPoint[k], &p->aPoint[j])<0 ){ + if( rtreeSearchPointCompare(&p->aPoint[k], &p->aPoint[i])<0 ){ + rtreeSearchPointSwap(p, i, k); + i = k; + }else{ + break; + } + }else{ + if( rtreeSearchPointCompare(&p->aPoint[j], &p->aPoint[i])<0 ){ + rtreeSearchPointSwap(p, i, j); + i = j; + }else{ + break; + } + } + } + } +} + + +/* +** Continue the search on cursor pCur until the front of the queue +** contains an entry suitable for returning as a result-set row, +** or until the RtreeSearchPoint queue is empty, indicating that the +** query has completed. +*/ +static int rtreeStepToLeaf(RtreeCursor *pCur){ + RtreeSearchPoint *p; + Rtree *pRtree = RTREE_OF_CURSOR(pCur); + RtreeNode *pNode; + int eWithin; + int rc = SQLITE_OK; + int nCell; + int nConstraint = pCur->nConstraint; + int ii; + int eInt; + RtreeSearchPoint x; + + eInt = pRtree->eCoordType==RTREE_COORD_INT32; + while( (p = rtreeSearchPointFirst(pCur))!=0 && p->iLevel>0 ){ + u8 *pCellData; + pNode = rtreeNodeOfFirstSearchPoint(pCur, &rc); + if( rc ) return rc; + nCell = NCELL(pNode); + assert( nCell<200 ); + pCellData = pNode->zData + (4+pRtree->nBytesPerCell*p->iCell); + while( p->iCellaConstraint + ii; + if( pConstraint->op>=RTREE_MATCH ){ + rc = rtreeCallbackConstraint(pConstraint, eInt, pCellData, p, + &rScore, &eWithin); + if( rc ) return rc; + }else if( p->iLevel==1 ){ + rtreeLeafConstraint(pConstraint, eInt, pCellData, &eWithin); + }else{ + rtreeNonleafConstraint(pConstraint, eInt, pCellData, &eWithin); + } + if( eWithin==NOT_WITHIN ){ + p->iCell++; + pCellData += pRtree->nBytesPerCell; + break; + } + } + if( eWithin==NOT_WITHIN ) continue; + p->iCell++; + x.iLevel = p->iLevel - 1; + if( x.iLevel ){ + x.id = readInt64(pCellData); + for(ii=0; iinPoint; ii++){ + if( pCur->aPoint[ii].id==x.id ){ + RTREE_IS_CORRUPT(pRtree); + return SQLITE_CORRUPT_VTAB; + } + } + x.iCell = 0; + }else{ + x.id = p->id; + x.iCell = p->iCell - 1; + } + if( p->iCell>=nCell ){ + RTREE_QUEUE_TRACE(pCur, "POP-S:"); + rtreeSearchPointPop(pCur); + } + if( rScoreeWithin = (u8)eWithin; + p->id = x.id; + p->iCell = x.iCell; + RTREE_QUEUE_TRACE(pCur, "PUSH-S:"); + break; + } + if( p->iCell>=nCell ){ + RTREE_QUEUE_TRACE(pCur, "POP-Se:"); + rtreeSearchPointPop(pCur); + } + } + pCur->atEOF = p==0; + return SQLITE_OK; +} + +/* +** Rtree virtual table module xNext method. +*/ +static int rtreeNext(sqlite3_vtab_cursor *pVtabCursor){ + RtreeCursor *pCsr = (RtreeCursor *)pVtabCursor; + int rc = SQLITE_OK; + + /* Move to the next entry that matches the configured constraints. */ + RTREE_QUEUE_TRACE(pCsr, "POP-Nx:"); + if( pCsr->bAuxValid ){ + pCsr->bAuxValid = 0; + sqlite3_reset(pCsr->pReadAux); + } + rtreeSearchPointPop(pCsr); + rc = rtreeStepToLeaf(pCsr); + return rc; +} + +/* +** Rtree virtual table module xRowid method. +*/ +static int rtreeRowid(sqlite3_vtab_cursor *pVtabCursor, sqlite_int64 *pRowid){ + RtreeCursor *pCsr = (RtreeCursor *)pVtabCursor; + RtreeSearchPoint *p = rtreeSearchPointFirst(pCsr); + int rc = SQLITE_OK; + RtreeNode *pNode = rtreeNodeOfFirstSearchPoint(pCsr, &rc); + if( rc==SQLITE_OK && p ){ + *pRowid = nodeGetRowid(RTREE_OF_CURSOR(pCsr), pNode, p->iCell); + } + return rc; +} + +/* +** Rtree virtual table module xColumn method. +*/ +static int rtreeColumn(sqlite3_vtab_cursor *cur, sqlite3_context *ctx, int i){ + Rtree *pRtree = (Rtree *)cur->pVtab; + RtreeCursor *pCsr = (RtreeCursor *)cur; + RtreeSearchPoint *p = rtreeSearchPointFirst(pCsr); + RtreeCoord c; + int rc = SQLITE_OK; + RtreeNode *pNode = rtreeNodeOfFirstSearchPoint(pCsr, &rc); + + if( rc ) return rc; + if( p==0 ) return SQLITE_OK; + if( i==0 ){ + sqlite3_result_int64(ctx, nodeGetRowid(pRtree, pNode, p->iCell)); + }else if( i<=pRtree->nDim2 ){ + nodeGetCoord(pRtree, pNode, p->iCell, i-1, &c); +#ifndef SQLITE_RTREE_INT_ONLY + if( pRtree->eCoordType==RTREE_COORD_REAL32 ){ + sqlite3_result_double(ctx, c.f); + }else +#endif + { + assert( pRtree->eCoordType==RTREE_COORD_INT32 ); + sqlite3_result_int(ctx, c.i); + } + }else{ + if( !pCsr->bAuxValid ){ + if( pCsr->pReadAux==0 ){ + rc = sqlite3_prepare_v3(pRtree->db, pRtree->zReadAuxSql, -1, 0, + &pCsr->pReadAux, 0); + if( rc ) return rc; + } + sqlite3_bind_int64(pCsr->pReadAux, 1, + nodeGetRowid(pRtree, pNode, p->iCell)); + rc = sqlite3_step(pCsr->pReadAux); + if( rc==SQLITE_ROW ){ + pCsr->bAuxValid = 1; + }else{ + sqlite3_reset(pCsr->pReadAux); + if( rc==SQLITE_DONE ) rc = SQLITE_OK; + return rc; + } + } + sqlite3_result_value(ctx, + sqlite3_column_value(pCsr->pReadAux, i - pRtree->nDim2 + 1)); + } + return SQLITE_OK; +} + +/* +** Use nodeAcquire() to obtain the leaf node containing the record with +** rowid iRowid. If successful, set *ppLeaf to point to the node and +** return SQLITE_OK. If there is no such record in the table, set +** *ppLeaf to 0 and return SQLITE_OK. If an error occurs, set *ppLeaf +** to zero and return an SQLite error code. +*/ +static int findLeafNode( + Rtree *pRtree, /* RTree to search */ + i64 iRowid, /* The rowid searching for */ + RtreeNode **ppLeaf, /* Write the node here */ + sqlite3_int64 *piNode /* Write the node-id here */ +){ + int rc; + *ppLeaf = 0; + sqlite3_bind_int64(pRtree->pReadRowid, 1, iRowid); + if( sqlite3_step(pRtree->pReadRowid)==SQLITE_ROW ){ + i64 iNode = sqlite3_column_int64(pRtree->pReadRowid, 0); + if( piNode ) *piNode = iNode; + rc = nodeAcquire(pRtree, iNode, 0, ppLeaf); + sqlite3_reset(pRtree->pReadRowid); + }else{ + rc = sqlite3_reset(pRtree->pReadRowid); + } + return rc; +} + +/* +** This function is called to configure the RtreeConstraint object passed +** as the second argument for a MATCH constraint. The value passed as the +** first argument to this function is the right-hand operand to the MATCH +** operator. +*/ +static int deserializeGeometry(sqlite3_value *pValue, RtreeConstraint *pCons){ + RtreeMatchArg *pBlob, *pSrc; /* BLOB returned by geometry function */ + sqlite3_rtree_query_info *pInfo; /* Callback information */ + + pSrc = sqlite3_value_pointer(pValue, "RtreeMatchArg"); + if( pSrc==0 ) return SQLITE_ERROR; + pInfo = (sqlite3_rtree_query_info*) + sqlite3_malloc64( sizeof(*pInfo)+pSrc->iSize ); + if( !pInfo ) return SQLITE_NOMEM; + memset(pInfo, 0, sizeof(*pInfo)); + pBlob = (RtreeMatchArg*)&pInfo[1]; + memcpy(pBlob, pSrc, pSrc->iSize); + pInfo->pContext = pBlob->cb.pContext; + pInfo->nParam = pBlob->nParam; + pInfo->aParam = pBlob->aParam; + pInfo->apSqlParam = pBlob->apSqlParam; + + if( pBlob->cb.xGeom ){ + pCons->u.xGeom = pBlob->cb.xGeom; + }else{ + pCons->op = RTREE_QUERY; + pCons->u.xQueryFunc = pBlob->cb.xQueryFunc; + } + pCons->pInfo = pInfo; + return SQLITE_OK; +} + +/* +** Rtree virtual table module xFilter method. +*/ +static int rtreeFilter( + sqlite3_vtab_cursor *pVtabCursor, + int idxNum, const char *idxStr, + int argc, sqlite3_value **argv +){ + Rtree *pRtree = (Rtree *)pVtabCursor->pVtab; + RtreeCursor *pCsr = (RtreeCursor *)pVtabCursor; + RtreeNode *pRoot = 0; + int ii; + int rc = SQLITE_OK; + int iCell = 0; + sqlite3_stmt *pStmt; + + rtreeReference(pRtree); + + /* Reset the cursor to the same state as rtreeOpen() leaves it in. */ + freeCursorConstraints(pCsr); + sqlite3_free(pCsr->aPoint); + pStmt = pCsr->pReadAux; + memset(pCsr, 0, sizeof(RtreeCursor)); + pCsr->base.pVtab = (sqlite3_vtab*)pRtree; + pCsr->pReadAux = pStmt; + + pCsr->iStrategy = idxNum; + if( idxNum==1 ){ + /* Special case - lookup by rowid. */ + RtreeNode *pLeaf; /* Leaf on which the required cell resides */ + RtreeSearchPoint *p; /* Search point for the leaf */ + i64 iRowid = sqlite3_value_int64(argv[0]); + i64 iNode = 0; + rc = findLeafNode(pRtree, iRowid, &pLeaf, &iNode); + if( rc==SQLITE_OK && pLeaf!=0 ){ + p = rtreeSearchPointNew(pCsr, RTREE_ZERO, 0); + assert( p!=0 ); /* Always returns pCsr->sPoint */ + pCsr->aNode[0] = pLeaf; + p->id = iNode; + p->eWithin = PARTLY_WITHIN; + rc = nodeRowidIndex(pRtree, pLeaf, iRowid, &iCell); + p->iCell = (u8)iCell; + RTREE_QUEUE_TRACE(pCsr, "PUSH-F1:"); + }else{ + pCsr->atEOF = 1; + } + }else{ + /* Normal case - r-tree scan. Set up the RtreeCursor.aConstraint array + ** with the configured constraints. + */ + rc = nodeAcquire(pRtree, 1, 0, &pRoot); + if( rc==SQLITE_OK && argc>0 ){ + pCsr->aConstraint = sqlite3_malloc64(sizeof(RtreeConstraint)*argc); + pCsr->nConstraint = argc; + if( !pCsr->aConstraint ){ + rc = SQLITE_NOMEM; + }else{ + memset(pCsr->aConstraint, 0, sizeof(RtreeConstraint)*argc); + memset(pCsr->anQueue, 0, sizeof(u32)*(pRtree->iDepth + 1)); + assert( (idxStr==0 && argc==0) + || (idxStr && (int)strlen(idxStr)==argc*2) ); + for(ii=0; iiaConstraint[ii]; + p->op = idxStr[ii*2]; + p->iCoord = idxStr[ii*2+1]-'0'; + if( p->op>=RTREE_MATCH ){ + /* A MATCH operator. The right-hand-side must be a blob that + ** can be cast into an RtreeMatchArg object. One created using + ** an sqlite3_rtree_geometry_callback() SQL user function. + */ + rc = deserializeGeometry(argv[ii], p); + if( rc!=SQLITE_OK ){ + break; + } + p->pInfo->nCoord = pRtree->nDim2; + p->pInfo->anQueue = pCsr->anQueue; + p->pInfo->mxLevel = pRtree->iDepth + 1; + }else{ +#ifdef SQLITE_RTREE_INT_ONLY + p->u.rValue = sqlite3_value_int64(argv[ii]); +#else + p->u.rValue = sqlite3_value_double(argv[ii]); +#endif + } + } + } + } + if( rc==SQLITE_OK ){ + RtreeSearchPoint *pNew; + pNew = rtreeSearchPointNew(pCsr, RTREE_ZERO, (u8)(pRtree->iDepth+1)); + if( pNew==0 ) return SQLITE_NOMEM; + pNew->id = 1; + pNew->iCell = 0; + pNew->eWithin = PARTLY_WITHIN; + assert( pCsr->bPoint==1 ); + pCsr->aNode[0] = pRoot; + pRoot = 0; + RTREE_QUEUE_TRACE(pCsr, "PUSH-Fm:"); + rc = rtreeStepToLeaf(pCsr); + } + } + + nodeRelease(pRtree, pRoot); + rtreeRelease(pRtree); + return rc; +} + +/* +** Rtree virtual table module xBestIndex method. There are three +** table scan strategies to choose from (in order from most to +** least desirable): +** +** idxNum idxStr Strategy +** ------------------------------------------------ +** 1 Unused Direct lookup by rowid. +** 2 See below R-tree query or full-table scan. +** ------------------------------------------------ +** +** If strategy 1 is used, then idxStr is not meaningful. If strategy +** 2 is used, idxStr is formatted to contain 2 bytes for each +** constraint used. The first two bytes of idxStr correspond to +** the constraint in sqlite3_index_info.aConstraintUsage[] with +** (argvIndex==1) etc. +** +** The first of each pair of bytes in idxStr identifies the constraint +** operator as follows: +** +** Operator Byte Value +** ---------------------- +** = 0x41 ('A') +** <= 0x42 ('B') +** < 0x43 ('C') +** >= 0x44 ('D') +** > 0x45 ('E') +** MATCH 0x46 ('F') +** ---------------------- +** +** The second of each pair of bytes identifies the coordinate column +** to which the constraint applies. The leftmost coordinate column +** is 'a', the second from the left 'b' etc. +*/ +static int rtreeBestIndex(sqlite3_vtab *tab, sqlite3_index_info *pIdxInfo){ + Rtree *pRtree = (Rtree*)tab; + int rc = SQLITE_OK; + int ii; + int bMatch = 0; /* True if there exists a MATCH constraint */ + i64 nRow; /* Estimated rows returned by this scan */ + + int iIdx = 0; + char zIdxStr[RTREE_MAX_DIMENSIONS*8+1]; + memset(zIdxStr, 0, sizeof(zIdxStr)); + + /* Check if there exists a MATCH constraint - even an unusable one. If there + ** is, do not consider the lookup-by-rowid plan as using such a plan would + ** require the VDBE to evaluate the MATCH constraint, which is not currently + ** possible. */ + for(ii=0; iinConstraint; ii++){ + if( pIdxInfo->aConstraint[ii].op==SQLITE_INDEX_CONSTRAINT_MATCH ){ + bMatch = 1; + } + } + + assert( pIdxInfo->idxStr==0 ); + for(ii=0; iinConstraint && iIdx<(int)(sizeof(zIdxStr)-1); ii++){ + struct sqlite3_index_constraint *p = &pIdxInfo->aConstraint[ii]; + + if( bMatch==0 && p->usable + && p->iColumn==0 && p->op==SQLITE_INDEX_CONSTRAINT_EQ + ){ + /* We have an equality constraint on the rowid. Use strategy 1. */ + int jj; + for(jj=0; jjaConstraintUsage[jj].argvIndex = 0; + pIdxInfo->aConstraintUsage[jj].omit = 0; + } + pIdxInfo->idxNum = 1; + pIdxInfo->aConstraintUsage[ii].argvIndex = 1; + pIdxInfo->aConstraintUsage[jj].omit = 1; + + /* This strategy involves a two rowid lookups on an B-Tree structures + ** and then a linear search of an R-Tree node. This should be + ** considered almost as quick as a direct rowid lookup (for which + ** sqlite uses an internal cost of 0.0). It is expected to return + ** a single row. + */ + pIdxInfo->estimatedCost = 30.0; + pIdxInfo->estimatedRows = 1; + pIdxInfo->idxFlags = SQLITE_INDEX_SCAN_UNIQUE; + return SQLITE_OK; + } + + if( p->usable + && ((p->iColumn>0 && p->iColumn<=pRtree->nDim2) + || p->op==SQLITE_INDEX_CONSTRAINT_MATCH) + ){ + u8 op; + switch( p->op ){ + case SQLITE_INDEX_CONSTRAINT_EQ: op = RTREE_EQ; break; + case SQLITE_INDEX_CONSTRAINT_GT: op = RTREE_GT; break; + case SQLITE_INDEX_CONSTRAINT_LE: op = RTREE_LE; break; + case SQLITE_INDEX_CONSTRAINT_LT: op = RTREE_LT; break; + case SQLITE_INDEX_CONSTRAINT_GE: op = RTREE_GE; break; + case SQLITE_INDEX_CONSTRAINT_MATCH: op = RTREE_MATCH; break; + default: op = 0; break; + } + if( op ){ + zIdxStr[iIdx++] = op; + zIdxStr[iIdx++] = (char)(p->iColumn - 1 + '0'); + pIdxInfo->aConstraintUsage[ii].argvIndex = (iIdx/2); + pIdxInfo->aConstraintUsage[ii].omit = 1; + } + } + } + + pIdxInfo->idxNum = 2; + pIdxInfo->needToFreeIdxStr = 1; + if( iIdx>0 && 0==(pIdxInfo->idxStr = sqlite3_mprintf("%s", zIdxStr)) ){ + return SQLITE_NOMEM; + } + + nRow = pRtree->nRowEst >> (iIdx/2); + pIdxInfo->estimatedCost = (double)6.0 * (double)nRow; + pIdxInfo->estimatedRows = nRow; + + return rc; +} + +/* +** Return the N-dimensional volumn of the cell stored in *p. +*/ +static RtreeDValue cellArea(Rtree *pRtree, RtreeCell *p){ + RtreeDValue area = (RtreeDValue)1; + assert( pRtree->nDim>=1 && pRtree->nDim<=5 ); +#ifndef SQLITE_RTREE_INT_ONLY + if( pRtree->eCoordType==RTREE_COORD_REAL32 ){ + switch( pRtree->nDim ){ + case 5: area = p->aCoord[9].f - p->aCoord[8].f; + case 4: area *= p->aCoord[7].f - p->aCoord[6].f; + case 3: area *= p->aCoord[5].f - p->aCoord[4].f; + case 2: area *= p->aCoord[3].f - p->aCoord[2].f; + default: area *= p->aCoord[1].f - p->aCoord[0].f; + } + }else +#endif + { + switch( pRtree->nDim ){ + case 5: area = (i64)p->aCoord[9].i - (i64)p->aCoord[8].i; + case 4: area *= (i64)p->aCoord[7].i - (i64)p->aCoord[6].i; + case 3: area *= (i64)p->aCoord[5].i - (i64)p->aCoord[4].i; + case 2: area *= (i64)p->aCoord[3].i - (i64)p->aCoord[2].i; + default: area *= (i64)p->aCoord[1].i - (i64)p->aCoord[0].i; + } + } + return area; +} + +/* +** Return the margin length of cell p. The margin length is the sum +** of the objects size in each dimension. +*/ +static RtreeDValue cellMargin(Rtree *pRtree, RtreeCell *p){ + RtreeDValue margin = 0; + int ii = pRtree->nDim2 - 2; + do{ + margin += (DCOORD(p->aCoord[ii+1]) - DCOORD(p->aCoord[ii])); + ii -= 2; + }while( ii>=0 ); + return margin; +} + +/* +** Store the union of cells p1 and p2 in p1. +*/ +static void cellUnion(Rtree *pRtree, RtreeCell *p1, RtreeCell *p2){ + int ii = 0; + if( pRtree->eCoordType==RTREE_COORD_REAL32 ){ + do{ + p1->aCoord[ii].f = MIN(p1->aCoord[ii].f, p2->aCoord[ii].f); + p1->aCoord[ii+1].f = MAX(p1->aCoord[ii+1].f, p2->aCoord[ii+1].f); + ii += 2; + }while( iinDim2 ); + }else{ + do{ + p1->aCoord[ii].i = MIN(p1->aCoord[ii].i, p2->aCoord[ii].i); + p1->aCoord[ii+1].i = MAX(p1->aCoord[ii+1].i, p2->aCoord[ii+1].i); + ii += 2; + }while( iinDim2 ); + } +} + +/* +** Return true if the area covered by p2 is a subset of the area covered +** by p1. False otherwise. +*/ +static int cellContains(Rtree *pRtree, RtreeCell *p1, RtreeCell *p2){ + int ii; + int isInt = (pRtree->eCoordType==RTREE_COORD_INT32); + for(ii=0; iinDim2; ii+=2){ + RtreeCoord *a1 = &p1->aCoord[ii]; + RtreeCoord *a2 = &p2->aCoord[ii]; + if( (!isInt && (a2[0].fa1[1].f)) + || ( isInt && (a2[0].ia1[1].i)) + ){ + return 0; + } + } + return 1; +} + +/* +** Return the amount cell p would grow by if it were unioned with pCell. +*/ +static RtreeDValue cellGrowth(Rtree *pRtree, RtreeCell *p, RtreeCell *pCell){ + RtreeDValue area; + RtreeCell cell; + memcpy(&cell, p, sizeof(RtreeCell)); + area = cellArea(pRtree, &cell); + cellUnion(pRtree, &cell, pCell); + return (cellArea(pRtree, &cell)-area); +} + +static RtreeDValue cellOverlap( + Rtree *pRtree, + RtreeCell *p, + RtreeCell *aCell, + int nCell +){ + int ii; + RtreeDValue overlap = RTREE_ZERO; + for(ii=0; iinDim2; jj+=2){ + RtreeDValue x1, x2; + x1 = MAX(DCOORD(p->aCoord[jj]), DCOORD(aCell[ii].aCoord[jj])); + x2 = MIN(DCOORD(p->aCoord[jj+1]), DCOORD(aCell[ii].aCoord[jj+1])); + if( x2iDepth-iHeight); ii++){ + int iCell; + sqlite3_int64 iBest = 0; + + RtreeDValue fMinGrowth = RTREE_ZERO; + RtreeDValue fMinArea = RTREE_ZERO; + + int nCell = NCELL(pNode); + RtreeCell cell; + RtreeNode *pChild; + + RtreeCell *aCell = 0; + + /* Select the child node which will be enlarged the least if pCell + ** is inserted into it. Resolve ties by choosing the entry with + ** the smallest area. + */ + for(iCell=0; iCellpParent ){ + RtreeNode *pParent = p->pParent; + RtreeCell cell; + int iCell; + + if( (++cnt)>1000 || nodeParentIndex(pRtree, p, &iCell) ){ + RTREE_IS_CORRUPT(pRtree); + return SQLITE_CORRUPT_VTAB; + } + + nodeGetCell(pRtree, pParent, iCell, &cell); + if( !cellContains(pRtree, &cell, pCell) ){ + cellUnion(pRtree, &cell, pCell); + nodeOverwriteCell(pRtree, pParent, &cell, iCell); + } + + p = pParent; + } + return SQLITE_OK; +} + +/* +** Write mapping (iRowid->iNode) to the _rowid table. +*/ +static int rowidWrite(Rtree *pRtree, sqlite3_int64 iRowid, sqlite3_int64 iNode){ + sqlite3_bind_int64(pRtree->pWriteRowid, 1, iRowid); + sqlite3_bind_int64(pRtree->pWriteRowid, 2, iNode); + sqlite3_step(pRtree->pWriteRowid); + return sqlite3_reset(pRtree->pWriteRowid); +} + +/* +** Write mapping (iNode->iPar) to the _parent table. +*/ +static int parentWrite(Rtree *pRtree, sqlite3_int64 iNode, sqlite3_int64 iPar){ + sqlite3_bind_int64(pRtree->pWriteParent, 1, iNode); + sqlite3_bind_int64(pRtree->pWriteParent, 2, iPar); + sqlite3_step(pRtree->pWriteParent); + return sqlite3_reset(pRtree->pWriteParent); +} + +static int rtreeInsertCell(Rtree *, RtreeNode *, RtreeCell *, int); + + +/* +** Arguments aIdx, aDistance and aSpare all point to arrays of size +** nIdx. The aIdx array contains the set of integers from 0 to +** (nIdx-1) in no particular order. This function sorts the values +** in aIdx according to the indexed values in aDistance. For +** example, assuming the inputs: +** +** aIdx = { 0, 1, 2, 3 } +** aDistance = { 5.0, 2.0, 7.0, 6.0 } +** +** this function sets the aIdx array to contain: +** +** aIdx = { 0, 1, 2, 3 } +** +** The aSpare array is used as temporary working space by the +** sorting algorithm. +*/ +static void SortByDistance( + int *aIdx, + int nIdx, + RtreeDValue *aDistance, + int *aSpare +){ + if( nIdx>1 ){ + int iLeft = 0; + int iRight = 0; + + int nLeft = nIdx/2; + int nRight = nIdx-nLeft; + int *aLeft = aIdx; + int *aRight = &aIdx[nLeft]; + + SortByDistance(aLeft, nLeft, aDistance, aSpare); + SortByDistance(aRight, nRight, aDistance, aSpare); + + memcpy(aSpare, aLeft, sizeof(int)*nLeft); + aLeft = aSpare; + + while( iLeft1 ){ + + int iLeft = 0; + int iRight = 0; + + int nLeft = nIdx/2; + int nRight = nIdx-nLeft; + int *aLeft = aIdx; + int *aRight = &aIdx[nLeft]; + + SortByDimension(pRtree, aLeft, nLeft, iDim, aCell, aSpare); + SortByDimension(pRtree, aRight, nRight, iDim, aCell, aSpare); + + memcpy(aSpare, aLeft, sizeof(int)*nLeft); + aLeft = aSpare; + while( iLeftnDim+1)*(sizeof(int*)+nCell*sizeof(int)); + + aaSorted = (int **)sqlite3_malloc64(nByte); + if( !aaSorted ){ + return SQLITE_NOMEM; + } + + aSpare = &((int *)&aaSorted[pRtree->nDim])[pRtree->nDim*nCell]; + memset(aaSorted, 0, nByte); + for(ii=0; iinDim; ii++){ + int jj; + aaSorted[ii] = &((int *)&aaSorted[pRtree->nDim])[ii*nCell]; + for(jj=0; jjnDim; ii++){ + RtreeDValue margin = RTREE_ZERO; + RtreeDValue fBestOverlap = RTREE_ZERO; + RtreeDValue fBestArea = RTREE_ZERO; + int iBestLeft = 0; + int nLeft; + + for( + nLeft=RTREE_MINCELLS(pRtree); + nLeft<=(nCell-RTREE_MINCELLS(pRtree)); + nLeft++ + ){ + RtreeCell left; + RtreeCell right; + int kk; + RtreeDValue overlap; + RtreeDValue area; + + memcpy(&left, &aCell[aaSorted[ii][0]], sizeof(RtreeCell)); + memcpy(&right, &aCell[aaSorted[ii][nCell-1]], sizeof(RtreeCell)); + for(kk=1; kk<(nCell-1); kk++){ + if( kk0 ){ + RtreeNode *pChild = nodeHashLookup(pRtree, iRowid); + if( pChild ){ + nodeRelease(pRtree, pChild->pParent); + nodeReference(pNode); + pChild->pParent = pNode; + } + } + return xSetMapping(pRtree, iRowid, pNode->iNode); +} + +static int SplitNode( + Rtree *pRtree, + RtreeNode *pNode, + RtreeCell *pCell, + int iHeight +){ + int i; + int newCellIsRight = 0; + + int rc = SQLITE_OK; + int nCell = NCELL(pNode); + RtreeCell *aCell; + int *aiUsed; + + RtreeNode *pLeft = 0; + RtreeNode *pRight = 0; + + RtreeCell leftbbox; + RtreeCell rightbbox; + + /* Allocate an array and populate it with a copy of pCell and + ** all cells from node pLeft. Then zero the original node. + */ + aCell = sqlite3_malloc64((sizeof(RtreeCell)+sizeof(int))*(nCell+1)); + if( !aCell ){ + rc = SQLITE_NOMEM; + goto splitnode_out; + } + aiUsed = (int *)&aCell[nCell+1]; + memset(aiUsed, 0, sizeof(int)*(nCell+1)); + for(i=0; iiNode==1 ){ + pRight = nodeNew(pRtree, pNode); + pLeft = nodeNew(pRtree, pNode); + pRtree->iDepth++; + pNode->isDirty = 1; + writeInt16(pNode->zData, pRtree->iDepth); + }else{ + pLeft = pNode; + pRight = nodeNew(pRtree, pLeft->pParent); + pLeft->nRef++; + } + + if( !pLeft || !pRight ){ + rc = SQLITE_NOMEM; + goto splitnode_out; + } + + memset(pLeft->zData, 0, pRtree->iNodeSize); + memset(pRight->zData, 0, pRtree->iNodeSize); + + rc = splitNodeStartree(pRtree, aCell, nCell, pLeft, pRight, + &leftbbox, &rightbbox); + if( rc!=SQLITE_OK ){ + goto splitnode_out; + } + + /* Ensure both child nodes have node numbers assigned to them by calling + ** nodeWrite(). Node pRight always needs a node number, as it was created + ** by nodeNew() above. But node pLeft sometimes already has a node number. + ** In this case avoid the all to nodeWrite(). + */ + if( SQLITE_OK!=(rc = nodeWrite(pRtree, pRight)) + || (0==pLeft->iNode && SQLITE_OK!=(rc = nodeWrite(pRtree, pLeft))) + ){ + goto splitnode_out; + } + + rightbbox.iRowid = pRight->iNode; + leftbbox.iRowid = pLeft->iNode; + + if( pNode->iNode==1 ){ + rc = rtreeInsertCell(pRtree, pLeft->pParent, &leftbbox, iHeight+1); + if( rc!=SQLITE_OK ){ + goto splitnode_out; + } + }else{ + RtreeNode *pParent = pLeft->pParent; + int iCell; + rc = nodeParentIndex(pRtree, pLeft, &iCell); + if( rc==SQLITE_OK ){ + nodeOverwriteCell(pRtree, pParent, &leftbbox, iCell); + rc = AdjustTree(pRtree, pParent, &leftbbox); + } + if( rc!=SQLITE_OK ){ + goto splitnode_out; + } + } + if( (rc = rtreeInsertCell(pRtree, pRight->pParent, &rightbbox, iHeight+1)) ){ + goto splitnode_out; + } + + for(i=0; iiRowid ){ + newCellIsRight = 1; + } + if( rc!=SQLITE_OK ){ + goto splitnode_out; + } + } + if( pNode->iNode==1 ){ + for(i=0; iiRowid, pLeft, iHeight); + } + + if( rc==SQLITE_OK ){ + rc = nodeRelease(pRtree, pRight); + pRight = 0; + } + if( rc==SQLITE_OK ){ + rc = nodeRelease(pRtree, pLeft); + pLeft = 0; + } + +splitnode_out: + nodeRelease(pRtree, pRight); + nodeRelease(pRtree, pLeft); + sqlite3_free(aCell); + return rc; +} + +/* +** If node pLeaf is not the root of the r-tree and its pParent pointer is +** still NULL, load all ancestor nodes of pLeaf into memory and populate +** the pLeaf->pParent chain all the way up to the root node. +** +** This operation is required when a row is deleted (or updated - an update +** is implemented as a delete followed by an insert). SQLite provides the +** rowid of the row to delete, which can be used to find the leaf on which +** the entry resides (argument pLeaf). Once the leaf is located, this +** function is called to determine its ancestry. +*/ +static int fixLeafParent(Rtree *pRtree, RtreeNode *pLeaf){ + int rc = SQLITE_OK; + RtreeNode *pChild = pLeaf; + while( rc==SQLITE_OK && pChild->iNode!=1 && pChild->pParent==0 ){ + int rc2 = SQLITE_OK; /* sqlite3_reset() return code */ + sqlite3_bind_int64(pRtree->pReadParent, 1, pChild->iNode); + rc = sqlite3_step(pRtree->pReadParent); + if( rc==SQLITE_ROW ){ + RtreeNode *pTest; /* Used to test for reference loops */ + i64 iNode; /* Node number of parent node */ + + /* Before setting pChild->pParent, test that we are not creating a + ** loop of references (as we would if, say, pChild==pParent). We don't + ** want to do this as it leads to a memory leak when trying to delete + ** the referenced counted node structures. + */ + iNode = sqlite3_column_int64(pRtree->pReadParent, 0); + for(pTest=pLeaf; pTest && pTest->iNode!=iNode; pTest=pTest->pParent); + if( !pTest ){ + rc2 = nodeAcquire(pRtree, iNode, 0, &pChild->pParent); + } + } + rc = sqlite3_reset(pRtree->pReadParent); + if( rc==SQLITE_OK ) rc = rc2; + if( rc==SQLITE_OK && !pChild->pParent ){ + RTREE_IS_CORRUPT(pRtree); + rc = SQLITE_CORRUPT_VTAB; + } + pChild = pChild->pParent; + } + return rc; +} + +static int deleteCell(Rtree *, RtreeNode *, int, int); + +static int removeNode(Rtree *pRtree, RtreeNode *pNode, int iHeight){ + int rc; + int rc2; + RtreeNode *pParent = 0; + int iCell; + + assert( pNode->nRef==1 ); + + /* Remove the entry in the parent cell. */ + rc = nodeParentIndex(pRtree, pNode, &iCell); + if( rc==SQLITE_OK ){ + pParent = pNode->pParent; + pNode->pParent = 0; + rc = deleteCell(pRtree, pParent, iCell, iHeight+1); + } + rc2 = nodeRelease(pRtree, pParent); + if( rc==SQLITE_OK ){ + rc = rc2; + } + if( rc!=SQLITE_OK ){ + return rc; + } + + /* Remove the xxx_node entry. */ + sqlite3_bind_int64(pRtree->pDeleteNode, 1, pNode->iNode); + sqlite3_step(pRtree->pDeleteNode); + if( SQLITE_OK!=(rc = sqlite3_reset(pRtree->pDeleteNode)) ){ + return rc; + } + + /* Remove the xxx_parent entry. */ + sqlite3_bind_int64(pRtree->pDeleteParent, 1, pNode->iNode); + sqlite3_step(pRtree->pDeleteParent); + if( SQLITE_OK!=(rc = sqlite3_reset(pRtree->pDeleteParent)) ){ + return rc; + } + + /* Remove the node from the in-memory hash table and link it into + ** the Rtree.pDeleted list. Its contents will be re-inserted later on. + */ + nodeHashDelete(pRtree, pNode); + pNode->iNode = iHeight; + pNode->pNext = pRtree->pDeleted; + pNode->nRef++; + pRtree->pDeleted = pNode; + + return SQLITE_OK; +} + +static int fixBoundingBox(Rtree *pRtree, RtreeNode *pNode){ + RtreeNode *pParent = pNode->pParent; + int rc = SQLITE_OK; + if( pParent ){ + int ii; + int nCell = NCELL(pNode); + RtreeCell box; /* Bounding box for pNode */ + nodeGetCell(pRtree, pNode, 0, &box); + for(ii=1; iiiNode; + rc = nodeParentIndex(pRtree, pNode, &ii); + if( rc==SQLITE_OK ){ + nodeOverwriteCell(pRtree, pParent, &box, ii); + rc = fixBoundingBox(pRtree, pParent); + } + } + return rc; +} + +/* +** Delete the cell at index iCell of node pNode. After removing the +** cell, adjust the r-tree data structure if required. +*/ +static int deleteCell(Rtree *pRtree, RtreeNode *pNode, int iCell, int iHeight){ + RtreeNode *pParent; + int rc; + + if( SQLITE_OK!=(rc = fixLeafParent(pRtree, pNode)) ){ + return rc; + } + + /* Remove the cell from the node. This call just moves bytes around + ** the in-memory node image, so it cannot fail. + */ + nodeDeleteCell(pRtree, pNode, iCell); + + /* If the node is not the tree root and now has less than the minimum + ** number of cells, remove it from the tree. Otherwise, update the + ** cell in the parent node so that it tightly contains the updated + ** node. + */ + pParent = pNode->pParent; + assert( pParent || pNode->iNode==1 ); + if( pParent ){ + if( NCELL(pNode)nDim; iDim++){ + aCenterCoord[iDim] += DCOORD(aCell[ii].aCoord[iDim*2]); + aCenterCoord[iDim] += DCOORD(aCell[ii].aCoord[iDim*2+1]); + } + } + for(iDim=0; iDimnDim; iDim++){ + aCenterCoord[iDim] = (aCenterCoord[iDim]/(nCell*(RtreeDValue)2)); + } + + for(ii=0; iinDim; iDim++){ + RtreeDValue coord = (DCOORD(aCell[ii].aCoord[iDim*2+1]) - + DCOORD(aCell[ii].aCoord[iDim*2])); + aDistance[ii] += (coord-aCenterCoord[iDim])*(coord-aCenterCoord[iDim]); + } + } + + SortByDistance(aOrder, nCell, aDistance, aSpare); + nodeZero(pRtree, pNode); + + for(ii=0; rc==SQLITE_OK && ii<(nCell-(RTREE_MINCELLS(pRtree)+1)); ii++){ + RtreeCell *p = &aCell[aOrder[ii]]; + nodeInsertCell(pRtree, pNode, p); + if( p->iRowid==pCell->iRowid ){ + if( iHeight==0 ){ + rc = rowidWrite(pRtree, p->iRowid, pNode->iNode); + }else{ + rc = parentWrite(pRtree, p->iRowid, pNode->iNode); + } + } + } + if( rc==SQLITE_OK ){ + rc = fixBoundingBox(pRtree, pNode); + } + for(; rc==SQLITE_OK && iiiNode currently contains + ** the height of the sub-tree headed by the cell. + */ + RtreeNode *pInsert; + RtreeCell *p = &aCell[aOrder[ii]]; + rc = ChooseLeaf(pRtree, p, iHeight, &pInsert); + if( rc==SQLITE_OK ){ + int rc2; + rc = rtreeInsertCell(pRtree, pInsert, p, iHeight); + rc2 = nodeRelease(pRtree, pInsert); + if( rc==SQLITE_OK ){ + rc = rc2; + } + } + } + + sqlite3_free(aCell); + return rc; +} + +/* +** Insert cell pCell into node pNode. Node pNode is the head of a +** subtree iHeight high (leaf nodes have iHeight==0). +*/ +static int rtreeInsertCell( + Rtree *pRtree, + RtreeNode *pNode, + RtreeCell *pCell, + int iHeight +){ + int rc = SQLITE_OK; + if( iHeight>0 ){ + RtreeNode *pChild = nodeHashLookup(pRtree, pCell->iRowid); + if( pChild ){ + nodeRelease(pRtree, pChild->pParent); + nodeReference(pNode); + pChild->pParent = pNode; + } + } + if( nodeInsertCell(pRtree, pNode, pCell) ){ + if( iHeight<=pRtree->iReinsertHeight || pNode->iNode==1){ + rc = SplitNode(pRtree, pNode, pCell, iHeight); + }else{ + pRtree->iReinsertHeight = iHeight; + rc = Reinsert(pRtree, pNode, pCell, iHeight); + } + }else{ + rc = AdjustTree(pRtree, pNode, pCell); + if( rc==SQLITE_OK ){ + if( iHeight==0 ){ + rc = rowidWrite(pRtree, pCell->iRowid, pNode->iNode); + }else{ + rc = parentWrite(pRtree, pCell->iRowid, pNode->iNode); + } + } + } + return rc; +} + +static int reinsertNodeContent(Rtree *pRtree, RtreeNode *pNode){ + int ii; + int rc = SQLITE_OK; + int nCell = NCELL(pNode); + + for(ii=0; rc==SQLITE_OK && iiiNode currently contains + ** the height of the sub-tree headed by the cell. + */ + rc = ChooseLeaf(pRtree, &cell, (int)pNode->iNode, &pInsert); + if( rc==SQLITE_OK ){ + int rc2; + rc = rtreeInsertCell(pRtree, pInsert, &cell, (int)pNode->iNode); + rc2 = nodeRelease(pRtree, pInsert); + if( rc==SQLITE_OK ){ + rc = rc2; + } + } + } + return rc; +} + +/* +** Select a currently unused rowid for a new r-tree record. +*/ +static int rtreeNewRowid(Rtree *pRtree, i64 *piRowid){ + int rc; + sqlite3_bind_null(pRtree->pWriteRowid, 1); + sqlite3_bind_null(pRtree->pWriteRowid, 2); + sqlite3_step(pRtree->pWriteRowid); + rc = sqlite3_reset(pRtree->pWriteRowid); + *piRowid = sqlite3_last_insert_rowid(pRtree->db); + return rc; +} + +/* +** Remove the entry with rowid=iDelete from the r-tree structure. +*/ +static int rtreeDeleteRowid(Rtree *pRtree, sqlite3_int64 iDelete){ + int rc; /* Return code */ + RtreeNode *pLeaf = 0; /* Leaf node containing record iDelete */ + int iCell; /* Index of iDelete cell in pLeaf */ + RtreeNode *pRoot = 0; /* Root node of rtree structure */ + + + /* Obtain a reference to the root node to initialize Rtree.iDepth */ + rc = nodeAcquire(pRtree, 1, 0, &pRoot); + + /* Obtain a reference to the leaf node that contains the entry + ** about to be deleted. + */ + if( rc==SQLITE_OK ){ + rc = findLeafNode(pRtree, iDelete, &pLeaf, 0); + } + +#ifdef CORRUPT_DB + assert( pLeaf!=0 || rc!=SQLITE_OK || CORRUPT_DB ); +#endif + + /* Delete the cell in question from the leaf node. */ + if( rc==SQLITE_OK && pLeaf ){ + int rc2; + rc = nodeRowidIndex(pRtree, pLeaf, iDelete, &iCell); + if( rc==SQLITE_OK ){ + rc = deleteCell(pRtree, pLeaf, iCell, 0); + } + rc2 = nodeRelease(pRtree, pLeaf); + if( rc==SQLITE_OK ){ + rc = rc2; + } + } + + /* Delete the corresponding entry in the _rowid table. */ + if( rc==SQLITE_OK ){ + sqlite3_bind_int64(pRtree->pDeleteRowid, 1, iDelete); + sqlite3_step(pRtree->pDeleteRowid); + rc = sqlite3_reset(pRtree->pDeleteRowid); + } + + /* Check if the root node now has exactly one child. If so, remove + ** it, schedule the contents of the child for reinsertion and + ** reduce the tree height by one. + ** + ** This is equivalent to copying the contents of the child into + ** the root node (the operation that Gutman's paper says to perform + ** in this scenario). + */ + if( rc==SQLITE_OK && pRtree->iDepth>0 && NCELL(pRoot)==1 ){ + int rc2; + RtreeNode *pChild = 0; + i64 iChild = nodeGetRowid(pRtree, pRoot, 0); + rc = nodeAcquire(pRtree, iChild, pRoot, &pChild); + if( rc==SQLITE_OK ){ + rc = removeNode(pRtree, pChild, pRtree->iDepth-1); + } + rc2 = nodeRelease(pRtree, pChild); + if( rc==SQLITE_OK ) rc = rc2; + if( rc==SQLITE_OK ){ + pRtree->iDepth--; + writeInt16(pRoot->zData, pRtree->iDepth); + pRoot->isDirty = 1; + } + } + + /* Re-insert the contents of any underfull nodes removed from the tree. */ + for(pLeaf=pRtree->pDeleted; pLeaf; pLeaf=pRtree->pDeleted){ + if( rc==SQLITE_OK ){ + rc = reinsertNodeContent(pRtree, pLeaf); + } + pRtree->pDeleted = pLeaf->pNext; + pRtree->nNodeRef--; + sqlite3_free(pLeaf); + } + + /* Release the reference to the root node. */ + if( rc==SQLITE_OK ){ + rc = nodeRelease(pRtree, pRoot); + }else{ + nodeRelease(pRtree, pRoot); + } + + return rc; +} + +/* +** Rounding constants for float->double conversion. +*/ +#define RNDTOWARDS (1.0 - 1.0/8388608.0) /* Round towards zero */ +#define RNDAWAY (1.0 + 1.0/8388608.0) /* Round away from zero */ + +#if !defined(SQLITE_RTREE_INT_ONLY) +/* +** Convert an sqlite3_value into an RtreeValue (presumably a float) +** while taking care to round toward negative or positive, respectively. +*/ +static RtreeValue rtreeValueDown(sqlite3_value *v){ + double d = sqlite3_value_double(v); + float f = (float)d; + if( f>d ){ + f = (float)(d*(d<0 ? RNDAWAY : RNDTOWARDS)); + } + return f; +} +static RtreeValue rtreeValueUp(sqlite3_value *v){ + double d = sqlite3_value_double(v); + float f = (float)d; + if( fbase.zErrMsg) to an appropriate value and returns +** SQLITE_CONSTRAINT. +** +** Parameter iCol is the index of the leftmost column involved in the +** constraint failure. If it is 0, then the constraint that failed is +** the unique constraint on the id column. Otherwise, it is the rtree +** (c1<=c2) constraint on columns iCol and iCol+1 that has failed. +** +** If an OOM occurs, SQLITE_NOMEM is returned instead of SQLITE_CONSTRAINT. +*/ +static int rtreeConstraintError(Rtree *pRtree, int iCol){ + sqlite3_stmt *pStmt = 0; + char *zSql; + int rc; + + assert( iCol==0 || iCol%2 ); + zSql = sqlite3_mprintf("SELECT * FROM %Q.%Q", pRtree->zDb, pRtree->zName); + if( zSql ){ + rc = sqlite3_prepare_v2(pRtree->db, zSql, -1, &pStmt, 0); + }else{ + rc = SQLITE_NOMEM; + } + sqlite3_free(zSql); + + if( rc==SQLITE_OK ){ + if( iCol==0 ){ + const char *zCol = sqlite3_column_name(pStmt, 0); + pRtree->base.zErrMsg = sqlite3_mprintf( + "UNIQUE constraint failed: %s.%s", pRtree->zName, zCol + ); + }else{ + const char *zCol1 = sqlite3_column_name(pStmt, iCol); + const char *zCol2 = sqlite3_column_name(pStmt, iCol+1); + pRtree->base.zErrMsg = sqlite3_mprintf( + "rtree constraint failed: %s.(%s<=%s)", pRtree->zName, zCol1, zCol2 + ); + } + } + + sqlite3_finalize(pStmt); + return (rc==SQLITE_OK ? SQLITE_CONSTRAINT : rc); +} + + + +/* +** The xUpdate method for rtree module virtual tables. +*/ +static int rtreeUpdate( + sqlite3_vtab *pVtab, + int nData, + sqlite3_value **aData, + sqlite_int64 *pRowid +){ + Rtree *pRtree = (Rtree *)pVtab; + int rc = SQLITE_OK; + RtreeCell cell; /* New cell to insert if nData>1 */ + int bHaveRowid = 0; /* Set to 1 after new rowid is determined */ + + if( pRtree->nNodeRef ){ + /* Unable to write to the btree while another cursor is reading from it, + ** since the write might do a rebalance which would disrupt the read + ** cursor. */ + return SQLITE_LOCKED_VTAB; + } + rtreeReference(pRtree); + assert(nData>=1); + + cell.iRowid = 0; /* Used only to suppress a compiler warning */ + + /* Constraint handling. A write operation on an r-tree table may return + ** SQLITE_CONSTRAINT for two reasons: + ** + ** 1. A duplicate rowid value, or + ** 2. The supplied data violates the "x2>=x1" constraint. + ** + ** In the first case, if the conflict-handling mode is REPLACE, then + ** the conflicting row can be removed before proceeding. In the second + ** case, SQLITE_CONSTRAINT must be returned regardless of the + ** conflict-handling mode specified by the user. + */ + if( nData>1 ){ + int ii; + int nn = nData - 4; + + if( nn > pRtree->nDim2 ) nn = pRtree->nDim2; + /* Populate the cell.aCoord[] array. The first coordinate is aData[3]. + ** + ** NB: nData can only be less than nDim*2+3 if the rtree is mis-declared + ** with "column" that are interpreted as table constraints. + ** Example: CREATE VIRTUAL TABLE bad USING rtree(x,y,CHECK(y>5)); + ** This problem was discovered after years of use, so we silently ignore + ** these kinds of misdeclared tables to avoid breaking any legacy. + */ + +#ifndef SQLITE_RTREE_INT_ONLY + if( pRtree->eCoordType==RTREE_COORD_REAL32 ){ + for(ii=0; iicell.aCoord[ii+1].f ){ + rc = rtreeConstraintError(pRtree, ii+1); + goto constraint; + } + } + }else +#endif + { + for(ii=0; iicell.aCoord[ii+1].i ){ + rc = rtreeConstraintError(pRtree, ii+1); + goto constraint; + } + } + } + + /* If a rowid value was supplied, check if it is already present in + ** the table. If so, the constraint has failed. */ + if( sqlite3_value_type(aData[2])!=SQLITE_NULL ){ + cell.iRowid = sqlite3_value_int64(aData[2]); + if( sqlite3_value_type(aData[0])==SQLITE_NULL + || sqlite3_value_int64(aData[0])!=cell.iRowid + ){ + int steprc; + sqlite3_bind_int64(pRtree->pReadRowid, 1, cell.iRowid); + steprc = sqlite3_step(pRtree->pReadRowid); + rc = sqlite3_reset(pRtree->pReadRowid); + if( SQLITE_ROW==steprc ){ + if( sqlite3_vtab_on_conflict(pRtree->db)==SQLITE_REPLACE ){ + rc = rtreeDeleteRowid(pRtree, cell.iRowid); + }else{ + rc = rtreeConstraintError(pRtree, 0); + goto constraint; + } + } + } + bHaveRowid = 1; + } + } + + /* If aData[0] is not an SQL NULL value, it is the rowid of a + ** record to delete from the r-tree table. The following block does + ** just that. + */ + if( sqlite3_value_type(aData[0])!=SQLITE_NULL ){ + rc = rtreeDeleteRowid(pRtree, sqlite3_value_int64(aData[0])); + } + + /* If the aData[] array contains more than one element, elements + ** (aData[2]..aData[argc-1]) contain a new record to insert into + ** the r-tree structure. + */ + if( rc==SQLITE_OK && nData>1 ){ + /* Insert the new record into the r-tree */ + RtreeNode *pLeaf = 0; + + /* Figure out the rowid of the new row. */ + if( bHaveRowid==0 ){ + rc = rtreeNewRowid(pRtree, &cell.iRowid); + } + *pRowid = cell.iRowid; + + if( rc==SQLITE_OK ){ + rc = ChooseLeaf(pRtree, &cell, 0, &pLeaf); + } + if( rc==SQLITE_OK ){ + int rc2; + pRtree->iReinsertHeight = -1; + rc = rtreeInsertCell(pRtree, pLeaf, &cell, 0); + rc2 = nodeRelease(pRtree, pLeaf); + if( rc==SQLITE_OK ){ + rc = rc2; + } + } + if( rc==SQLITE_OK && pRtree->nAux ){ + sqlite3_stmt *pUp = pRtree->pWriteAux; + int jj; + sqlite3_bind_int64(pUp, 1, *pRowid); + for(jj=0; jjnAux; jj++){ + sqlite3_bind_value(pUp, jj+2, aData[pRtree->nDim2+3+jj]); + } + sqlite3_step(pUp); + rc = sqlite3_reset(pUp); + } + } + +constraint: + rtreeRelease(pRtree); + return rc; +} + +/* +** Called when a transaction starts. +*/ +static int rtreeBeginTransaction(sqlite3_vtab *pVtab){ + Rtree *pRtree = (Rtree *)pVtab; + assert( pRtree->inWrTrans==0 ); + pRtree->inWrTrans++; + return SQLITE_OK; +} + +/* +** Called when a transaction completes (either by COMMIT or ROLLBACK). +** The sqlite3_blob object should be released at this point. +*/ +static int rtreeEndTransaction(sqlite3_vtab *pVtab){ + Rtree *pRtree = (Rtree *)pVtab; + pRtree->inWrTrans = 0; + nodeBlobReset(pRtree); + return SQLITE_OK; +} + +/* +** The xRename method for rtree module virtual tables. +*/ +static int rtreeRename(sqlite3_vtab *pVtab, const char *zNewName){ + Rtree *pRtree = (Rtree *)pVtab; + int rc = SQLITE_NOMEM; + char *zSql = sqlite3_mprintf( + "ALTER TABLE %Q.'%q_node' RENAME TO \"%w_node\";" + "ALTER TABLE %Q.'%q_parent' RENAME TO \"%w_parent\";" + "ALTER TABLE %Q.'%q_rowid' RENAME TO \"%w_rowid\";" + , pRtree->zDb, pRtree->zName, zNewName + , pRtree->zDb, pRtree->zName, zNewName + , pRtree->zDb, pRtree->zName, zNewName + ); + if( zSql ){ + nodeBlobReset(pRtree); + rc = sqlite3_exec(pRtree->db, zSql, 0, 0, 0); + sqlite3_free(zSql); + } + return rc; +} + +/* +** The xSavepoint method. +** +** This module does not need to do anything to support savepoints. However, +** it uses this hook to close any open blob handle. This is done because a +** DROP TABLE command - which fortunately always opens a savepoint - cannot +** succeed if there are any open blob handles. i.e. if the blob handle were +** not closed here, the following would fail: +** +** BEGIN; +** INSERT INTO rtree... +** DROP TABLE ; -- Would fail with SQLITE_LOCKED +** COMMIT; +*/ +static int rtreeSavepoint(sqlite3_vtab *pVtab, int iSavepoint){ + Rtree *pRtree = (Rtree *)pVtab; + u8 iwt = pRtree->inWrTrans; + UNUSED_PARAMETER(iSavepoint); + pRtree->inWrTrans = 0; + nodeBlobReset(pRtree); + pRtree->inWrTrans = iwt; + return SQLITE_OK; +} + +/* +** This function populates the pRtree->nRowEst variable with an estimate +** of the number of rows in the virtual table. If possible, this is based +** on sqlite_stat1 data. Otherwise, use RTREE_DEFAULT_ROWEST. +*/ +static int rtreeQueryStat1(sqlite3 *db, Rtree *pRtree){ + const char *zFmt = "SELECT stat FROM %Q.sqlite_stat1 WHERE tbl = '%q_rowid'"; + char *zSql; + sqlite3_stmt *p; + int rc; + i64 nRow = 0; + + rc = sqlite3_table_column_metadata( + db, pRtree->zDb, "sqlite_stat1",0,0,0,0,0,0 + ); + if( rc!=SQLITE_OK ){ + pRtree->nRowEst = RTREE_DEFAULT_ROWEST; + return rc==SQLITE_ERROR ? SQLITE_OK : rc; + } + zSql = sqlite3_mprintf(zFmt, pRtree->zDb, pRtree->zName); + if( zSql==0 ){ + rc = SQLITE_NOMEM; + }else{ + rc = sqlite3_prepare_v2(db, zSql, -1, &p, 0); + if( rc==SQLITE_OK ){ + if( sqlite3_step(p)==SQLITE_ROW ) nRow = sqlite3_column_int64(p, 0); + rc = sqlite3_finalize(p); + }else if( rc!=SQLITE_NOMEM ){ + rc = SQLITE_OK; + } + + if( rc==SQLITE_OK ){ + if( nRow==0 ){ + pRtree->nRowEst = RTREE_DEFAULT_ROWEST; + }else{ + pRtree->nRowEst = MAX(nRow, RTREE_MIN_ROWEST); + } + } + sqlite3_free(zSql); + } + + return rc; +} + + +/* +** Return true if zName is the extension on one of the shadow tables used +** by this module. +*/ +static int rtreeShadowName(const char *zName){ + static const char *azName[] = { + "node", "parent", "rowid" + }; + unsigned int i; + for(i=0; idb = db; + + if( isCreate ){ + char *zCreate; + sqlite3_str *p = sqlite3_str_new(db); + int ii; + sqlite3_str_appendf(p, + "CREATE TABLE \"%w\".\"%w_rowid\"(rowid INTEGER PRIMARY KEY,nodeno", + zDb, zPrefix); + for(ii=0; iinAux; ii++){ + sqlite3_str_appendf(p,",a%d",ii); + } + sqlite3_str_appendf(p, + ");CREATE TABLE \"%w\".\"%w_node\"(nodeno INTEGER PRIMARY KEY,data);", + zDb, zPrefix); + sqlite3_str_appendf(p, + "CREATE TABLE \"%w\".\"%w_parent\"(nodeno INTEGER PRIMARY KEY,parentnode);", + zDb, zPrefix); + sqlite3_str_appendf(p, + "INSERT INTO \"%w\".\"%w_node\"VALUES(1,zeroblob(%d))", + zDb, zPrefix, pRtree->iNodeSize); + zCreate = sqlite3_str_finish(p); + if( !zCreate ){ + return SQLITE_NOMEM; + } + rc = sqlite3_exec(db, zCreate, 0, 0, 0); + sqlite3_free(zCreate); + if( rc!=SQLITE_OK ){ + return rc; + } + } + + appStmt[0] = &pRtree->pWriteNode; + appStmt[1] = &pRtree->pDeleteNode; + appStmt[2] = &pRtree->pReadRowid; + appStmt[3] = &pRtree->pWriteRowid; + appStmt[4] = &pRtree->pDeleteRowid; + appStmt[5] = &pRtree->pReadParent; + appStmt[6] = &pRtree->pWriteParent; + appStmt[7] = &pRtree->pDeleteParent; + + rc = rtreeQueryStat1(db, pRtree); + for(i=0; inAux==0 ){ + zFormat = azSql[i]; + }else { + /* An UPSERT is very slightly slower than REPLACE, but it is needed + ** if there are auxiliary columns */ + zFormat = "INSERT INTO\"%w\".\"%w_rowid\"(rowid,nodeno)VALUES(?1,?2)" + "ON CONFLICT(rowid)DO UPDATE SET nodeno=excluded.nodeno"; + } + zSql = sqlite3_mprintf(zFormat, zDb, zPrefix); + if( zSql ){ + rc = sqlite3_prepare_v3(db, zSql, -1, f, appStmt[i], 0); + }else{ + rc = SQLITE_NOMEM; + } + sqlite3_free(zSql); + } + if( pRtree->nAux ){ + pRtree->zReadAuxSql = sqlite3_mprintf( + "SELECT * FROM \"%w\".\"%w_rowid\" WHERE rowid=?1", + zDb, zPrefix); + if( pRtree->zReadAuxSql==0 ){ + rc = SQLITE_NOMEM; + }else{ + sqlite3_str *p = sqlite3_str_new(db); + int ii; + char *zSql; + sqlite3_str_appendf(p, "UPDATE \"%w\".\"%w_rowid\"SET ", zDb, zPrefix); + for(ii=0; iinAux; ii++){ + if( ii ) sqlite3_str_append(p, ",", 1); + if( iinAuxNotNull ){ + sqlite3_str_appendf(p,"a%d=coalesce(?%d,a%d)",ii,ii+2,ii); + }else{ + sqlite3_str_appendf(p,"a%d=?%d",ii,ii+2); + } + } + sqlite3_str_appendf(p, " WHERE rowid=?1"); + zSql = sqlite3_str_finish(p); + if( zSql==0 ){ + rc = SQLITE_NOMEM; + }else{ + rc = sqlite3_prepare_v3(db, zSql, -1, f, &pRtree->pWriteAux, 0); + sqlite3_free(zSql); + } + } + } + + return rc; +} + +/* +** The second argument to this function contains the text of an SQL statement +** that returns a single integer value. The statement is compiled and executed +** using database connection db. If successful, the integer value returned +** is written to *piVal and SQLITE_OK returned. Otherwise, an SQLite error +** code is returned and the value of *piVal after returning is not defined. +*/ +static int getIntFromStmt(sqlite3 *db, const char *zSql, int *piVal){ + int rc = SQLITE_NOMEM; + if( zSql ){ + sqlite3_stmt *pStmt = 0; + rc = sqlite3_prepare_v2(db, zSql, -1, &pStmt, 0); + if( rc==SQLITE_OK ){ + if( SQLITE_ROW==sqlite3_step(pStmt) ){ + *piVal = sqlite3_column_int(pStmt, 0); + } + rc = sqlite3_finalize(pStmt); + } + } + return rc; +} + +/* +** This function is called from within the xConnect() or xCreate() method to +** determine the node-size used by the rtree table being created or connected +** to. If successful, pRtree->iNodeSize is populated and SQLITE_OK returned. +** Otherwise, an SQLite error code is returned. +** +** If this function is being called as part of an xConnect(), then the rtree +** table already exists. In this case the node-size is determined by inspecting +** the root node of the tree. +** +** Otherwise, for an xCreate(), use 64 bytes less than the database page-size. +** This ensures that each node is stored on a single database page. If the +** database page-size is so large that more than RTREE_MAXCELLS entries +** would fit in a single node, use a smaller node-size. +*/ +static int getNodeSize( + sqlite3 *db, /* Database handle */ + Rtree *pRtree, /* Rtree handle */ + int isCreate, /* True for xCreate, false for xConnect */ + char **pzErr /* OUT: Error message, if any */ +){ + int rc; + char *zSql; + if( isCreate ){ + int iPageSize = 0; + zSql = sqlite3_mprintf("PRAGMA %Q.page_size", pRtree->zDb); + rc = getIntFromStmt(db, zSql, &iPageSize); + if( rc==SQLITE_OK ){ + pRtree->iNodeSize = iPageSize-64; + if( (4+pRtree->nBytesPerCell*RTREE_MAXCELLS)iNodeSize ){ + pRtree->iNodeSize = 4+pRtree->nBytesPerCell*RTREE_MAXCELLS; + } + }else{ + *pzErr = sqlite3_mprintf("%s", sqlite3_errmsg(db)); + } + }else{ + zSql = sqlite3_mprintf( + "SELECT length(data) FROM '%q'.'%q_node' WHERE nodeno = 1", + pRtree->zDb, pRtree->zName + ); + rc = getIntFromStmt(db, zSql, &pRtree->iNodeSize); + if( rc!=SQLITE_OK ){ + *pzErr = sqlite3_mprintf("%s", sqlite3_errmsg(db)); + }else if( pRtree->iNodeSize<(512-64) ){ + rc = SQLITE_CORRUPT_VTAB; + RTREE_IS_CORRUPT(pRtree); + *pzErr = sqlite3_mprintf("undersize RTree blobs in \"%q_node\"", + pRtree->zName); + } + } + + sqlite3_free(zSql); + return rc; +} + +/* +** This function is the implementation of both the xConnect and xCreate +** methods of the r-tree virtual table. +** +** argv[0] -> module name +** argv[1] -> database name +** argv[2] -> table name +** argv[...] -> column names... +*/ +static int rtreeInit( + sqlite3 *db, /* Database connection */ + void *pAux, /* One of the RTREE_COORD_* constants */ + int argc, const char *const*argv, /* Parameters to CREATE TABLE statement */ + sqlite3_vtab **ppVtab, /* OUT: New virtual table */ + char **pzErr, /* OUT: Error message, if any */ + int isCreate /* True for xCreate, false for xConnect */ +){ + int rc = SQLITE_OK; + Rtree *pRtree; + int nDb; /* Length of string argv[1] */ + int nName; /* Length of string argv[2] */ + int eCoordType = (pAux ? RTREE_COORD_INT32 : RTREE_COORD_REAL32); + sqlite3_str *pSql; + char *zSql; + int ii = 4; + int iErr; + + const char *aErrMsg[] = { + 0, /* 0 */ + "Wrong number of columns for an rtree table", /* 1 */ + "Too few columns for an rtree table", /* 2 */ + "Too many columns for an rtree table", /* 3 */ + "Auxiliary rtree columns must be last" /* 4 */ + }; + + assert( RTREE_MAX_AUX_COLUMN<256 ); /* Aux columns counted by a u8 */ + if( argc>RTREE_MAX_AUX_COLUMN+3 ){ + *pzErr = sqlite3_mprintf("%s", aErrMsg[3]); + return SQLITE_ERROR; + } + + sqlite3_vtab_config(db, SQLITE_VTAB_CONSTRAINT_SUPPORT, 1); + + /* Allocate the sqlite3_vtab structure */ + nDb = (int)strlen(argv[1]); + nName = (int)strlen(argv[2]); + pRtree = (Rtree *)sqlite3_malloc64(sizeof(Rtree)+nDb+nName+2); + if( !pRtree ){ + return SQLITE_NOMEM; + } + memset(pRtree, 0, sizeof(Rtree)+nDb+nName+2); + pRtree->nBusy = 1; + pRtree->base.pModule = &rtreeModule; + pRtree->zDb = (char *)&pRtree[1]; + pRtree->zName = &pRtree->zDb[nDb+1]; + pRtree->eCoordType = (u8)eCoordType; + memcpy(pRtree->zDb, argv[1], nDb); + memcpy(pRtree->zName, argv[2], nName); + + + /* Create/Connect to the underlying relational database schema. If + ** that is successful, call sqlite3_declare_vtab() to configure + ** the r-tree table schema. + */ + pSql = sqlite3_str_new(db); + sqlite3_str_appendf(pSql, "CREATE TABLE x(%s", argv[3]); + for(ii=4; iinAux++; + sqlite3_str_appendf(pSql, ",%s", argv[ii]+1); + }else if( pRtree->nAux>0 ){ + break; + }else{ + pRtree->nDim2++; + sqlite3_str_appendf(pSql, ",%s", argv[ii]); + } + } + sqlite3_str_appendf(pSql, ");"); + zSql = sqlite3_str_finish(pSql); + if( !zSql ){ + rc = SQLITE_NOMEM; + }else if( iinDim = pRtree->nDim2/2; + if( pRtree->nDim<1 ){ + iErr = 2; + }else if( pRtree->nDim2>RTREE_MAX_DIMENSIONS*2 ){ + iErr = 3; + }else if( pRtree->nDim2 % 2 ){ + iErr = 1; + }else{ + iErr = 0; + } + if( iErr ){ + *pzErr = sqlite3_mprintf("%s", aErrMsg[iErr]); + goto rtreeInit_fail; + } + pRtree->nBytesPerCell = 8 + pRtree->nDim2*4; + + /* Figure out the node size to use. */ + rc = getNodeSize(db, pRtree, isCreate, pzErr); + if( rc ) goto rtreeInit_fail; + rc = rtreeSqlInit(pRtree, db, argv[1], argv[2], isCreate); + if( rc ){ + *pzErr = sqlite3_mprintf("%s", sqlite3_errmsg(db)); + goto rtreeInit_fail; + } + + *ppVtab = (sqlite3_vtab *)pRtree; + return SQLITE_OK; + +rtreeInit_fail: + if( rc==SQLITE_OK ) rc = SQLITE_ERROR; + assert( *ppVtab==0 ); + assert( pRtree->nBusy==1 ); + rtreeRelease(pRtree); + return rc; +} + + +/* +** Implementation of a scalar function that decodes r-tree nodes to +** human readable strings. This can be used for debugging and analysis. +** +** The scalar function takes two arguments: (1) the number of dimensions +** to the rtree (between 1 and 5, inclusive) and (2) a blob of data containing +** an r-tree node. For a two-dimensional r-tree structure called "rt", to +** deserialize all nodes, a statement like: +** +** SELECT rtreenode(2, data) FROM rt_node; +** +** The human readable string takes the form of a Tcl list with one +** entry for each cell in the r-tree node. Each entry is itself a +** list, containing the 8-byte rowid/pageno followed by the +** *2 coordinates. +*/ +static void rtreenode(sqlite3_context *ctx, int nArg, sqlite3_value **apArg){ + RtreeNode node; + Rtree tree; + int ii; + int nData; + int errCode; + sqlite3_str *pOut; + + UNUSED_PARAMETER(nArg); + memset(&node, 0, sizeof(RtreeNode)); + memset(&tree, 0, sizeof(Rtree)); + tree.nDim = (u8)sqlite3_value_int(apArg[0]); + if( tree.nDim<1 || tree.nDim>5 ) return; + tree.nDim2 = tree.nDim*2; + tree.nBytesPerCell = 8 + 8 * tree.nDim; + node.zData = (u8 *)sqlite3_value_blob(apArg[1]); + nData = sqlite3_value_bytes(apArg[1]); + if( nData<4 ) return; + if( nData0 ) sqlite3_str_append(pOut, " ", 1); + sqlite3_str_appendf(pOut, "{%lld", cell.iRowid); + for(jj=0; jjrc==SQLITE_OK ) pCheck->rc = rc; +} + +/* +** The second and subsequent arguments to this function are a format string +** and printf style arguments. This function formats the string and attempts +** to compile it as an SQL statement. +** +** If successful, a pointer to the new SQL statement is returned. Otherwise, +** NULL is returned and an error code left in RtreeCheck.rc. +*/ +static sqlite3_stmt *rtreeCheckPrepare( + RtreeCheck *pCheck, /* RtreeCheck object */ + const char *zFmt, ... /* Format string and trailing args */ +){ + va_list ap; + char *z; + sqlite3_stmt *pRet = 0; + + va_start(ap, zFmt); + z = sqlite3_vmprintf(zFmt, ap); + + if( pCheck->rc==SQLITE_OK ){ + if( z==0 ){ + pCheck->rc = SQLITE_NOMEM; + }else{ + pCheck->rc = sqlite3_prepare_v2(pCheck->db, z, -1, &pRet, 0); + } + } + + sqlite3_free(z); + va_end(ap); + return pRet; +} + +/* +** The second and subsequent arguments to this function are a printf() +** style format string and arguments. This function formats the string and +** appends it to the report being accumuated in pCheck. +*/ +static void rtreeCheckAppendMsg(RtreeCheck *pCheck, const char *zFmt, ...){ + va_list ap; + va_start(ap, zFmt); + if( pCheck->rc==SQLITE_OK && pCheck->nErrrc = SQLITE_NOMEM; + }else{ + pCheck->zReport = sqlite3_mprintf("%z%s%z", + pCheck->zReport, (pCheck->zReport ? "\n" : ""), z + ); + if( pCheck->zReport==0 ){ + pCheck->rc = SQLITE_NOMEM; + } + } + pCheck->nErr++; + } + va_end(ap); +} + +/* +** This function is a no-op if there is already an error code stored +** in the RtreeCheck object indicated by the first argument. NULL is +** returned in this case. +** +** Otherwise, the contents of rtree table node iNode are loaded from +** the database and copied into a buffer obtained from sqlite3_malloc(). +** If no error occurs, a pointer to the buffer is returned and (*pnNode) +** is set to the size of the buffer in bytes. +** +** Or, if an error does occur, NULL is returned and an error code left +** in the RtreeCheck object. The final value of *pnNode is undefined in +** this case. +*/ +static u8 *rtreeCheckGetNode(RtreeCheck *pCheck, i64 iNode, int *pnNode){ + u8 *pRet = 0; /* Return value */ + + if( pCheck->rc==SQLITE_OK && pCheck->pGetNode==0 ){ + pCheck->pGetNode = rtreeCheckPrepare(pCheck, + "SELECT data FROM %Q.'%q_node' WHERE nodeno=?", + pCheck->zDb, pCheck->zTab + ); + } + + if( pCheck->rc==SQLITE_OK ){ + sqlite3_bind_int64(pCheck->pGetNode, 1, iNode); + if( sqlite3_step(pCheck->pGetNode)==SQLITE_ROW ){ + int nNode = sqlite3_column_bytes(pCheck->pGetNode, 0); + const u8 *pNode = (const u8*)sqlite3_column_blob(pCheck->pGetNode, 0); + pRet = sqlite3_malloc64(nNode); + if( pRet==0 ){ + pCheck->rc = SQLITE_NOMEM; + }else{ + memcpy(pRet, pNode, nNode); + *pnNode = nNode; + } + } + rtreeCheckReset(pCheck, pCheck->pGetNode); + if( pCheck->rc==SQLITE_OK && pRet==0 ){ + rtreeCheckAppendMsg(pCheck, "Node %lld missing from database", iNode); + } + } + + return pRet; +} + +/* +** This function is used to check that the %_parent (if bLeaf==0) or %_rowid +** (if bLeaf==1) table contains a specified entry. The schemas of the +** two tables are: +** +** CREATE TABLE %_parent(nodeno INTEGER PRIMARY KEY, parentnode INTEGER) +** CREATE TABLE %_rowid(rowid INTEGER PRIMARY KEY, nodeno INTEGER, ...) +** +** In both cases, this function checks that there exists an entry with +** IPK value iKey and the second column set to iVal. +** +*/ +static void rtreeCheckMapping( + RtreeCheck *pCheck, /* RtreeCheck object */ + int bLeaf, /* True for a leaf cell, false for interior */ + i64 iKey, /* Key for mapping */ + i64 iVal /* Expected value for mapping */ +){ + int rc; + sqlite3_stmt *pStmt; + const char *azSql[2] = { + "SELECT parentnode FROM %Q.'%q_parent' WHERE nodeno=?1", + "SELECT nodeno FROM %Q.'%q_rowid' WHERE rowid=?1" + }; + + assert( bLeaf==0 || bLeaf==1 ); + if( pCheck->aCheckMapping[bLeaf]==0 ){ + pCheck->aCheckMapping[bLeaf] = rtreeCheckPrepare(pCheck, + azSql[bLeaf], pCheck->zDb, pCheck->zTab + ); + } + if( pCheck->rc!=SQLITE_OK ) return; + + pStmt = pCheck->aCheckMapping[bLeaf]; + sqlite3_bind_int64(pStmt, 1, iKey); + rc = sqlite3_step(pStmt); + if( rc==SQLITE_DONE ){ + rtreeCheckAppendMsg(pCheck, "Mapping (%lld -> %lld) missing from %s table", + iKey, iVal, (bLeaf ? "%_rowid" : "%_parent") + ); + }else if( rc==SQLITE_ROW ){ + i64 ii = sqlite3_column_int64(pStmt, 0); + if( ii!=iVal ){ + rtreeCheckAppendMsg(pCheck, + "Found (%lld -> %lld) in %s table, expected (%lld -> %lld)", + iKey, ii, (bLeaf ? "%_rowid" : "%_parent"), iKey, iVal + ); + } + } + rtreeCheckReset(pCheck, pStmt); +} + +/* +** Argument pCell points to an array of coordinates stored on an rtree page. +** This function checks that the coordinates are internally consistent (no +** x1>x2 conditions) and adds an error message to the RtreeCheck object +** if they are not. +** +** Additionally, if pParent is not NULL, then it is assumed to point to +** the array of coordinates on the parent page that bound the page +** containing pCell. In this case it is also verified that the two +** sets of coordinates are mutually consistent and an error message added +** to the RtreeCheck object if they are not. +*/ +static void rtreeCheckCellCoord( + RtreeCheck *pCheck, + i64 iNode, /* Node id to use in error messages */ + int iCell, /* Cell number to use in error messages */ + u8 *pCell, /* Pointer to cell coordinates */ + u8 *pParent /* Pointer to parent coordinates */ +){ + RtreeCoord c1, c2; + RtreeCoord p1, p2; + int i; + + for(i=0; inDim; i++){ + readCoord(&pCell[4*2*i], &c1); + readCoord(&pCell[4*(2*i + 1)], &c2); + + /* printf("%e, %e\n", c1.u.f, c2.u.f); */ + if( pCheck->bInt ? c1.i>c2.i : c1.f>c2.f ){ + rtreeCheckAppendMsg(pCheck, + "Dimension %d of cell %d on node %lld is corrupt", i, iCell, iNode + ); + } + + if( pParent ){ + readCoord(&pParent[4*2*i], &p1); + readCoord(&pParent[4*(2*i + 1)], &p2); + + if( (pCheck->bInt ? c1.ibInt ? c2.i>p2.i : c2.f>p2.f) + ){ + rtreeCheckAppendMsg(pCheck, + "Dimension %d of cell %d on node %lld is corrupt relative to parent" + , i, iCell, iNode + ); + } + } + } +} + +/* +** Run rtreecheck() checks on node iNode, which is at depth iDepth within +** the r-tree structure. Argument aParent points to the array of coordinates +** that bound node iNode on the parent node. +** +** If any problems are discovered, an error message is appended to the +** report accumulated in the RtreeCheck object. +*/ +static void rtreeCheckNode( + RtreeCheck *pCheck, + int iDepth, /* Depth of iNode (0==leaf) */ + u8 *aParent, /* Buffer containing parent coords */ + i64 iNode /* Node to check */ +){ + u8 *aNode = 0; + int nNode = 0; + + assert( iNode==1 || aParent!=0 ); + assert( pCheck->nDim>0 ); + + aNode = rtreeCheckGetNode(pCheck, iNode, &nNode); + if( aNode ){ + if( nNode<4 ){ + rtreeCheckAppendMsg(pCheck, + "Node %lld is too small (%d bytes)", iNode, nNode + ); + }else{ + int nCell; /* Number of cells on page */ + int i; /* Used to iterate through cells */ + if( aParent==0 ){ + iDepth = readInt16(aNode); + if( iDepth>RTREE_MAX_DEPTH ){ + rtreeCheckAppendMsg(pCheck, "Rtree depth out of range (%d)", iDepth); + sqlite3_free(aNode); + return; + } + } + nCell = readInt16(&aNode[2]); + if( (4 + nCell*(8 + pCheck->nDim*2*4))>nNode ){ + rtreeCheckAppendMsg(pCheck, + "Node %lld is too small for cell count of %d (%d bytes)", + iNode, nCell, nNode + ); + }else{ + for(i=0; inDim*2*4)]; + i64 iVal = readInt64(pCell); + rtreeCheckCellCoord(pCheck, iNode, i, &pCell[8], aParent); + + if( iDepth>0 ){ + rtreeCheckMapping(pCheck, 0, iVal, iNode); + rtreeCheckNode(pCheck, iDepth-1, &pCell[8], iVal); + pCheck->nNonLeaf++; + }else{ + rtreeCheckMapping(pCheck, 1, iVal, iNode); + pCheck->nLeaf++; + } + } + } + } + sqlite3_free(aNode); + } +} + +/* +** The second argument to this function must be either "_rowid" or +** "_parent". This function checks that the number of entries in the +** %_rowid or %_parent table is exactly nExpect. If not, it adds +** an error message to the report in the RtreeCheck object indicated +** by the first argument. +*/ +static void rtreeCheckCount(RtreeCheck *pCheck, const char *zTbl, i64 nExpect){ + if( pCheck->rc==SQLITE_OK ){ + sqlite3_stmt *pCount; + pCount = rtreeCheckPrepare(pCheck, "SELECT count(*) FROM %Q.'%q%s'", + pCheck->zDb, pCheck->zTab, zTbl + ); + if( pCount ){ + if( sqlite3_step(pCount)==SQLITE_ROW ){ + i64 nActual = sqlite3_column_int64(pCount, 0); + if( nActual!=nExpect ){ + rtreeCheckAppendMsg(pCheck, "Wrong number of entries in %%%s table" + " - expected %lld, actual %lld" , zTbl, nExpect, nActual + ); + } + } + pCheck->rc = sqlite3_finalize(pCount); + } + } +} + +/* +** This function does the bulk of the work for the rtree integrity-check. +** It is called by rtreecheck(), which is the SQL function implementation. +*/ +static int rtreeCheckTable( + sqlite3 *db, /* Database handle to access db through */ + const char *zDb, /* Name of db ("main", "temp" etc.) */ + const char *zTab, /* Name of rtree table to check */ + char **pzReport /* OUT: sqlite3_malloc'd report text */ +){ + RtreeCheck check; /* Common context for various routines */ + sqlite3_stmt *pStmt = 0; /* Used to find column count of rtree table */ + int bEnd = 0; /* True if transaction should be closed */ + int nAux = 0; /* Number of extra columns. */ + + /* Initialize the context object */ + memset(&check, 0, sizeof(check)); + check.db = db; + check.zDb = zDb; + check.zTab = zTab; + + /* If there is not already an open transaction, open one now. This is + ** to ensure that the queries run as part of this integrity-check operate + ** on a consistent snapshot. */ + if( sqlite3_get_autocommit(db) ){ + check.rc = sqlite3_exec(db, "BEGIN", 0, 0, 0); + bEnd = 1; + } + + /* Find the number of auxiliary columns */ + if( check.rc==SQLITE_OK ){ + pStmt = rtreeCheckPrepare(&check, "SELECT * FROM %Q.'%q_rowid'", zDb, zTab); + if( pStmt ){ + nAux = sqlite3_column_count(pStmt) - 2; + sqlite3_finalize(pStmt); + } + check.rc = SQLITE_OK; + } + + /* Find number of dimensions in the rtree table. */ + pStmt = rtreeCheckPrepare(&check, "SELECT * FROM %Q.%Q", zDb, zTab); + if( pStmt ){ + int rc; + check.nDim = (sqlite3_column_count(pStmt) - 1 - nAux) / 2; + if( check.nDim<1 ){ + rtreeCheckAppendMsg(&check, "Schema corrupt or not an rtree"); + }else if( SQLITE_ROW==sqlite3_step(pStmt) ){ + check.bInt = (sqlite3_column_type(pStmt, 1)==SQLITE_INTEGER); + } + rc = sqlite3_finalize(pStmt); + if( rc!=SQLITE_CORRUPT ) check.rc = rc; + } + + /* Do the actual integrity-check */ + if( check.nDim>=1 ){ + if( check.rc==SQLITE_OK ){ + rtreeCheckNode(&check, 0, 0, 1); + } + rtreeCheckCount(&check, "_rowid", check.nLeaf); + rtreeCheckCount(&check, "_parent", check.nNonLeaf); + } + + /* Finalize SQL statements used by the integrity-check */ + sqlite3_finalize(check.pGetNode); + sqlite3_finalize(check.aCheckMapping[0]); + sqlite3_finalize(check.aCheckMapping[1]); + + /* If one was opened, close the transaction */ + if( bEnd ){ + int rc = sqlite3_exec(db, "END", 0, 0, 0); + if( check.rc==SQLITE_OK ) check.rc = rc; + } + *pzReport = check.zReport; + return check.rc; +} + +/* +** Usage: +** +** rtreecheck(); +** rtreecheck(, ); +** +** Invoking this SQL function runs an integrity-check on the named rtree +** table. The integrity-check verifies the following: +** +** 1. For each cell in the r-tree structure (%_node table), that: +** +** a) for each dimension, (coord1 <= coord2). +** +** b) unless the cell is on the root node, that the cell is bounded +** by the parent cell on the parent node. +** +** c) for leaf nodes, that there is an entry in the %_rowid +** table corresponding to the cell's rowid value that +** points to the correct node. +** +** d) for cells on non-leaf nodes, that there is an entry in the +** %_parent table mapping from the cell's child node to the +** node that it resides on. +** +** 2. That there are the same number of entries in the %_rowid table +** as there are leaf cells in the r-tree structure, and that there +** is a leaf cell that corresponds to each entry in the %_rowid table. +** +** 3. That there are the same number of entries in the %_parent table +** as there are non-leaf cells in the r-tree structure, and that +** there is a non-leaf cell that corresponds to each entry in the +** %_parent table. +*/ +static void rtreecheck( + sqlite3_context *ctx, + int nArg, + sqlite3_value **apArg +){ + if( nArg!=1 && nArg!=2 ){ + sqlite3_result_error(ctx, + "wrong number of arguments to function rtreecheck()", -1 + ); + }else{ + int rc; + char *zReport = 0; + const char *zDb = (const char*)sqlite3_value_text(apArg[0]); + const char *zTab; + if( nArg==1 ){ + zTab = zDb; + zDb = "main"; + }else{ + zTab = (const char*)sqlite3_value_text(apArg[1]); + } + rc = rtreeCheckTable(sqlite3_context_db_handle(ctx), zDb, zTab, &zReport); + if( rc==SQLITE_OK ){ + sqlite3_result_text(ctx, zReport ? zReport : "ok", -1, SQLITE_TRANSIENT); + }else{ + sqlite3_result_error_code(ctx, rc); + } + sqlite3_free(zReport); + } +} + +/* Conditionally include the geopoly code */ +#ifdef SQLITE_ENABLE_GEOPOLY +/************** Include geopoly.c in the middle of rtree.c *******************/ +/************** Begin file geopoly.c *****************************************/ +/* +** 2018-05-25 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +****************************************************************************** +** +** This file implements an alternative R-Tree virtual table that +** uses polygons to express the boundaries of 2-dimensional objects. +** +** This file is #include-ed onto the end of "rtree.c" so that it has +** access to all of the R-Tree internals. +*/ +/* #include */ + +/* Enable -DGEOPOLY_ENABLE_DEBUG for debugging facilities */ +#ifdef GEOPOLY_ENABLE_DEBUG + static int geo_debug = 0; +# define GEODEBUG(X) if(geo_debug)printf X +#else +# define GEODEBUG(X) +#endif + +#ifndef JSON_NULL /* The following stuff repeats things found in json1 */ +/* +** Versions of isspace(), isalnum() and isdigit() to which it is safe +** to pass signed char values. +*/ +#ifdef sqlite3Isdigit + /* Use the SQLite core versions if this routine is part of the + ** SQLite amalgamation */ +# define safe_isdigit(x) sqlite3Isdigit(x) +# define safe_isalnum(x) sqlite3Isalnum(x) +# define safe_isxdigit(x) sqlite3Isxdigit(x) +#else + /* Use the standard library for separate compilation */ +#include /* amalgamator: keep */ +# define safe_isdigit(x) isdigit((unsigned char)(x)) +# define safe_isalnum(x) isalnum((unsigned char)(x)) +# define safe_isxdigit(x) isxdigit((unsigned char)(x)) +#endif + +/* +** Growing our own isspace() routine this way is twice as fast as +** the library isspace() function. +*/ +static const char geopolyIsSpace[] = { + 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 0, 0, 1, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, +}; +#define safe_isspace(x) (geopolyIsSpace[(unsigned char)x]) +#endif /* JSON NULL - back to original code */ + +/* Compiler and version */ +#ifndef GCC_VERSION +#if defined(__GNUC__) && !defined(SQLITE_DISABLE_INTRINSIC) +# define GCC_VERSION (__GNUC__*1000000+__GNUC_MINOR__*1000+__GNUC_PATCHLEVEL__) +#else +# define GCC_VERSION 0 +#endif +#endif +#ifndef MSVC_VERSION +#if defined(_MSC_VER) && !defined(SQLITE_DISABLE_INTRINSIC) +# define MSVC_VERSION _MSC_VER +#else +# define MSVC_VERSION 0 +#endif +#endif + +/* Datatype for coordinates +*/ +typedef float GeoCoord; + +/* +** Internal representation of a polygon. +** +** The polygon consists of a sequence of vertexes. There is a line +** segment between each pair of vertexes, and one final segment from +** the last vertex back to the first. (This differs from the GeoJSON +** standard in which the final vertex is a repeat of the first.) +** +** The polygon follows the right-hand rule. The area to the right of +** each segment is "outside" and the area to the left is "inside". +** +** The on-disk representation consists of a 4-byte header followed by +** the values. The 4-byte header is: +** +** encoding (1 byte) 0=big-endian, 1=little-endian +** nvertex (3 bytes) Number of vertexes as a big-endian integer +** +** Enough space is allocated for 4 coordinates, to work around over-zealous +** warnings coming from some compiler (notably, clang). In reality, the size +** of each GeoPoly memory allocate is adjusted as necessary so that the +** GeoPoly.a[] array at the end is the appropriate size. +*/ +typedef struct GeoPoly GeoPoly; +struct GeoPoly { + int nVertex; /* Number of vertexes */ + unsigned char hdr[4]; /* Header for on-disk representation */ + GeoCoord a[8]; /* 2*nVertex values. X (longitude) first, then Y */ +}; + +/* The size of a memory allocation needed for a GeoPoly object sufficient +** to hold N coordinate pairs. +*/ +#define GEOPOLY_SZ(N) (sizeof(GeoPoly) + sizeof(GeoCoord)*2*((N)-4)) + +/* Macros to access coordinates of a GeoPoly. +** We have to use these macros, rather than just say p->a[i] in order +** to silence (incorrect) UBSAN warnings if the array index is too large. +*/ +#define GeoX(P,I) (((GeoCoord*)(P)->a)[(I)*2]) +#define GeoY(P,I) (((GeoCoord*)(P)->a)[(I)*2+1]) + + +/* +** State of a parse of a GeoJSON input. +*/ +typedef struct GeoParse GeoParse; +struct GeoParse { + const unsigned char *z; /* Unparsed input */ + int nVertex; /* Number of vertexes in a[] */ + int nAlloc; /* Space allocated to a[] */ + int nErr; /* Number of errors encountered */ + GeoCoord *a; /* Array of vertexes. From sqlite3_malloc64() */ +}; + +/* Do a 4-byte byte swap */ +static void geopolySwab32(unsigned char *a){ + unsigned char t = a[0]; + a[0] = a[3]; + a[3] = t; + t = a[1]; + a[1] = a[2]; + a[2] = t; +} + +/* Skip whitespace. Return the next non-whitespace character. */ +static char geopolySkipSpace(GeoParse *p){ + while( safe_isspace(p->z[0]) ) p->z++; + return p->z[0]; +} + +/* Parse out a number. Write the value into *pVal if pVal!=0. +** return non-zero on success and zero if the next token is not a number. +*/ +static int geopolyParseNumber(GeoParse *p, GeoCoord *pVal){ + char c = geopolySkipSpace(p); + const unsigned char *z = p->z; + int j = 0; + int seenDP = 0; + int seenE = 0; + if( c=='-' ){ + j = 1; + c = z[j]; + } + if( c=='0' && z[j+1]>='0' && z[j+1]<='9' ) return 0; + for(;; j++){ + c = z[j]; + if( safe_isdigit(c) ) continue; + if( c=='.' ){ + if( z[j-1]=='-' ) return 0; + if( seenDP ) return 0; + seenDP = 1; + continue; + } + if( c=='e' || c=='E' ){ + if( z[j-1]<'0' ) return 0; + if( seenE ) return -1; + seenDP = seenE = 1; + c = z[j+1]; + if( c=='+' || c=='-' ){ + j++; + c = z[j+1]; + } + if( c<'0' || c>'9' ) return 0; + continue; + } + break; + } + if( z[j-1]<'0' ) return 0; + if( pVal ){ +#ifdef SQLITE_AMALGAMATION + /* The sqlite3AtoF() routine is much much faster than atof(), if it + ** is available */ + double r; + (void)sqlite3AtoF((const char*)p->z, &r, j, SQLITE_UTF8); + *pVal = r; +#else + *pVal = (GeoCoord)atof((const char*)p->z); +#endif + } + p->z += j; + return 1; +} + +/* +** If the input is a well-formed JSON array of coordinates with at least +** four coordinates and where each coordinate is itself a two-value array, +** then convert the JSON into a GeoPoly object and return a pointer to +** that object. +** +** If any error occurs, return NULL. +*/ +static GeoPoly *geopolyParseJson(const unsigned char *z, int *pRc){ + GeoParse s; + int rc = SQLITE_OK; + memset(&s, 0, sizeof(s)); + s.z = z; + if( geopolySkipSpace(&s)=='[' ){ + s.z++; + while( geopolySkipSpace(&s)=='[' ){ + int ii = 0; + char c; + s.z++; + if( s.nVertex>=s.nAlloc ){ + GeoCoord *aNew; + s.nAlloc = s.nAlloc*2 + 16; + aNew = sqlite3_realloc64(s.a, s.nAlloc*sizeof(GeoCoord)*2 ); + if( aNew==0 ){ + rc = SQLITE_NOMEM; + s.nErr++; + break; + } + s.a = aNew; + } + while( geopolyParseNumber(&s, ii<=1 ? &s.a[s.nVertex*2+ii] : 0) ){ + ii++; + if( ii==2 ) s.nVertex++; + c = geopolySkipSpace(&s); + s.z++; + if( c==',' ) continue; + if( c==']' && ii>=2 ) break; + s.nErr++; + rc = SQLITE_ERROR; + goto parse_json_err; + } + if( geopolySkipSpace(&s)==',' ){ + s.z++; + continue; + } + break; + } + if( geopolySkipSpace(&s)==']' + && s.nVertex>=4 + && s.a[0]==s.a[s.nVertex*2-2] + && s.a[1]==s.a[s.nVertex*2-1] + && (s.z++, geopolySkipSpace(&s)==0) + ){ + GeoPoly *pOut; + int x = 1; + s.nVertex--; /* Remove the redundant vertex at the end */ + pOut = sqlite3_malloc64( GEOPOLY_SZ((sqlite3_int64)s.nVertex) ); + x = 1; + if( pOut==0 ) goto parse_json_err; + pOut->nVertex = s.nVertex; + memcpy(pOut->a, s.a, s.nVertex*2*sizeof(GeoCoord)); + pOut->hdr[0] = *(unsigned char*)&x; + pOut->hdr[1] = (s.nVertex>>16)&0xff; + pOut->hdr[2] = (s.nVertex>>8)&0xff; + pOut->hdr[3] = s.nVertex&0xff; + sqlite3_free(s.a); + if( pRc ) *pRc = SQLITE_OK; + return pOut; + }else{ + s.nErr++; + rc = SQLITE_ERROR; + } + } +parse_json_err: + if( pRc ) *pRc = rc; + sqlite3_free(s.a); + return 0; +} + +/* +** Given a function parameter, try to interpret it as a polygon, either +** in the binary format or JSON text. Compute a GeoPoly object and +** return a pointer to that object. Or if the input is not a well-formed +** polygon, put an error message in sqlite3_context and return NULL. +*/ +static GeoPoly *geopolyFuncParam( + sqlite3_context *pCtx, /* Context for error messages */ + sqlite3_value *pVal, /* The value to decode */ + int *pRc /* Write error here */ +){ + GeoPoly *p = 0; + int nByte; + if( sqlite3_value_type(pVal)==SQLITE_BLOB + && (nByte = sqlite3_value_bytes(pVal))>=(4+6*sizeof(GeoCoord)) + ){ + const unsigned char *a = sqlite3_value_blob(pVal); + int nVertex; + nVertex = (a[1]<<16) + (a[2]<<8) + a[3]; + if( (a[0]==0 || a[0]==1) + && (nVertex*2*sizeof(GeoCoord) + 4)==(unsigned int)nByte + ){ + p = sqlite3_malloc64( sizeof(*p) + (nVertex-1)*2*sizeof(GeoCoord) ); + if( p==0 ){ + if( pRc ) *pRc = SQLITE_NOMEM; + if( pCtx ) sqlite3_result_error_nomem(pCtx); + }else{ + int x = 1; + p->nVertex = nVertex; + memcpy(p->hdr, a, nByte); + if( a[0] != *(unsigned char*)&x ){ + int ii; + for(ii=0; iihdr[0] ^= 1; + } + } + } + if( pRc ) *pRc = SQLITE_OK; + return p; + }else if( sqlite3_value_type(pVal)==SQLITE_TEXT ){ + const unsigned char *zJson = sqlite3_value_text(pVal); + if( zJson==0 ){ + if( pRc ) *pRc = SQLITE_NOMEM; + return 0; + } + return geopolyParseJson(zJson, pRc); + }else{ + if( pRc ) *pRc = SQLITE_ERROR; + return 0; + } +} + +/* +** Implementation of the geopoly_blob(X) function. +** +** If the input is a well-formed Geopoly BLOB or JSON string +** then return the BLOB representation of the polygon. Otherwise +** return NULL. +*/ +static void geopolyBlobFunc( + sqlite3_context *context, + int argc, + sqlite3_value **argv +){ + GeoPoly *p = geopolyFuncParam(context, argv[0], 0); + if( p ){ + sqlite3_result_blob(context, p->hdr, + 4+8*p->nVertex, SQLITE_TRANSIENT); + sqlite3_free(p); + } +} + +/* +** SQL function: geopoly_json(X) +** +** Interpret X as a polygon and render it as a JSON array +** of coordinates. Or, if X is not a valid polygon, return NULL. +*/ +static void geopolyJsonFunc( + sqlite3_context *context, + int argc, + sqlite3_value **argv +){ + GeoPoly *p = geopolyFuncParam(context, argv[0], 0); + if( p ){ + sqlite3 *db = sqlite3_context_db_handle(context); + sqlite3_str *x = sqlite3_str_new(db); + int i; + sqlite3_str_append(x, "[", 1); + for(i=0; inVertex; i++){ + sqlite3_str_appendf(x, "[%!g,%!g],", GeoX(p,i), GeoY(p,i)); + } + sqlite3_str_appendf(x, "[%!g,%!g]]", GeoX(p,0), GeoY(p,0)); + sqlite3_result_text(context, sqlite3_str_finish(x), -1, sqlite3_free); + sqlite3_free(p); + } +} + +/* +** SQL function: geopoly_svg(X, ....) +** +** Interpret X as a polygon and render it as a SVG . +** Additional arguments are added as attributes to the . +*/ +static void geopolySvgFunc( + sqlite3_context *context, + int argc, + sqlite3_value **argv +){ + GeoPoly *p; + if( argc<1 ) return; + p = geopolyFuncParam(context, argv[0], 0); + if( p ){ + sqlite3 *db = sqlite3_context_db_handle(context); + sqlite3_str *x = sqlite3_str_new(db); + int i; + char cSep = '\''; + sqlite3_str_appendf(x, ""); + sqlite3_result_text(context, sqlite3_str_finish(x), -1, sqlite3_free); + sqlite3_free(p); + } +} + +/* +** SQL Function: geopoly_xform(poly, A, B, C, D, E, F) +** +** Transform and/or translate a polygon as follows: +** +** x1 = A*x0 + B*y0 + E +** y1 = C*x0 + D*y0 + F +** +** For a translation: +** +** geopoly_xform(poly, 1, 0, 0, 1, x-offset, y-offset) +** +** Rotate by R around the point (0,0): +** +** geopoly_xform(poly, cos(R), sin(R), -sin(R), cos(R), 0, 0) +*/ +static void geopolyXformFunc( + sqlite3_context *context, + int argc, + sqlite3_value **argv +){ + GeoPoly *p = geopolyFuncParam(context, argv[0], 0); + double A = sqlite3_value_double(argv[1]); + double B = sqlite3_value_double(argv[2]); + double C = sqlite3_value_double(argv[3]); + double D = sqlite3_value_double(argv[4]); + double E = sqlite3_value_double(argv[5]); + double F = sqlite3_value_double(argv[6]); + GeoCoord x1, y1, x0, y0; + int ii; + if( p ){ + for(ii=0; iinVertex; ii++){ + x0 = GeoX(p,ii); + y0 = GeoY(p,ii); + x1 = (GeoCoord)(A*x0 + B*y0 + E); + y1 = (GeoCoord)(C*x0 + D*y0 + F); + GeoX(p,ii) = x1; + GeoY(p,ii) = y1; + } + sqlite3_result_blob(context, p->hdr, + 4+8*p->nVertex, SQLITE_TRANSIENT); + sqlite3_free(p); + } +} + +/* +** Compute the area enclosed by the polygon. +** +** This routine can also be used to detect polygons that rotate in +** the wrong direction. Polygons are suppose to be counter-clockwise (CCW). +** This routine returns a negative value for clockwise (CW) polygons. +*/ +static double geopolyArea(GeoPoly *p){ + double rArea = 0.0; + int ii; + for(ii=0; iinVertex-1; ii++){ + rArea += (GeoX(p,ii) - GeoX(p,ii+1)) /* (x0 - x1) */ + * (GeoY(p,ii) + GeoY(p,ii+1)) /* (y0 + y1) */ + * 0.5; + } + rArea += (GeoX(p,ii) - GeoX(p,0)) /* (xN - x0) */ + * (GeoY(p,ii) + GeoY(p,0)) /* (yN + y0) */ + * 0.5; + return rArea; +} + +/* +** Implementation of the geopoly_area(X) function. +** +** If the input is a well-formed Geopoly BLOB then return the area +** enclosed by the polygon. If the polygon circulates clockwise instead +** of counterclockwise (as it should) then return the negative of the +** enclosed area. Otherwise return NULL. +*/ +static void geopolyAreaFunc( + sqlite3_context *context, + int argc, + sqlite3_value **argv +){ + GeoPoly *p = geopolyFuncParam(context, argv[0], 0); + if( p ){ + sqlite3_result_double(context, geopolyArea(p)); + sqlite3_free(p); + } +} + +/* +** Implementation of the geopoly_ccw(X) function. +** +** If the rotation of polygon X is clockwise (incorrect) instead of +** counter-clockwise (the correct winding order according to RFC7946) +** then reverse the order of the vertexes in polygon X. +** +** In other words, this routine returns a CCW polygon regardless of the +** winding order of its input. +** +** Use this routine to sanitize historical inputs that that sometimes +** contain polygons that wind in the wrong direction. +*/ +static void geopolyCcwFunc( + sqlite3_context *context, + int argc, + sqlite3_value **argv +){ + GeoPoly *p = geopolyFuncParam(context, argv[0], 0); + if( p ){ + if( geopolyArea(p)<0.0 ){ + int ii, jj; + for(ii=1, jj=p->nVertex-1; iihdr, + 4+8*p->nVertex, SQLITE_TRANSIENT); + sqlite3_free(p); + } +} + +#define GEOPOLY_PI 3.1415926535897932385 + +/* Fast approximation for sine(X) for X between -0.5*pi and 2*pi +*/ +static double geopolySine(double r){ + assert( r>=-0.5*GEOPOLY_PI && r<=2.0*GEOPOLY_PI ); + if( r>=1.5*GEOPOLY_PI ){ + r -= 2.0*GEOPOLY_PI; + } + if( r>=0.5*GEOPOLY_PI ){ + return -geopolySine(r-GEOPOLY_PI); + }else{ + double r2 = r*r; + double r3 = r2*r; + double r5 = r3*r2; + return 0.9996949*r - 0.1656700*r3 + 0.0075134*r5; + } +} + +/* +** Function: geopoly_regular(X,Y,R,N) +** +** Construct a simple, convex, regular polygon centered at X, Y +** with circumradius R and with N sides. +*/ +static void geopolyRegularFunc( + sqlite3_context *context, + int argc, + sqlite3_value **argv +){ + double x = sqlite3_value_double(argv[0]); + double y = sqlite3_value_double(argv[1]); + double r = sqlite3_value_double(argv[2]); + int n = sqlite3_value_int(argv[3]); + int i; + GeoPoly *p; + + if( n<3 || r<=0.0 ) return; + if( n>1000 ) n = 1000; + p = sqlite3_malloc64( sizeof(*p) + (n-1)*2*sizeof(GeoCoord) ); + if( p==0 ){ + sqlite3_result_error_nomem(context); + return; + } + i = 1; + p->hdr[0] = *(unsigned char*)&i; + p->hdr[1] = 0; + p->hdr[2] = (n>>8)&0xff; + p->hdr[3] = n&0xff; + for(i=0; ihdr, 4+8*n, SQLITE_TRANSIENT); + sqlite3_free(p); +} + +/* +** If pPoly is a polygon, compute its bounding box. Then: +** +** (1) if aCoord!=0 store the bounding box in aCoord, returning NULL +** (2) otherwise, compute a GeoPoly for the bounding box and return the +** new GeoPoly +** +** If pPoly is NULL but aCoord is not NULL, then compute a new GeoPoly from +** the bounding box in aCoord and return a pointer to that GeoPoly. +*/ +static GeoPoly *geopolyBBox( + sqlite3_context *context, /* For recording the error */ + sqlite3_value *pPoly, /* The polygon */ + RtreeCoord *aCoord, /* Results here */ + int *pRc /* Error code here */ +){ + GeoPoly *pOut = 0; + GeoPoly *p; + float mnX, mxX, mnY, mxY; + if( pPoly==0 && aCoord!=0 ){ + p = 0; + mnX = aCoord[0].f; + mxX = aCoord[1].f; + mnY = aCoord[2].f; + mxY = aCoord[3].f; + goto geopolyBboxFill; + }else{ + p = geopolyFuncParam(context, pPoly, pRc); + } + if( p ){ + int ii; + mnX = mxX = GeoX(p,0); + mnY = mxY = GeoY(p,0); + for(ii=1; iinVertex; ii++){ + double r = GeoX(p,ii); + if( rmxX ) mxX = (float)r; + r = GeoY(p,ii); + if( rmxY ) mxY = (float)r; + } + if( pRc ) *pRc = SQLITE_OK; + if( aCoord==0 ){ + geopolyBboxFill: + pOut = sqlite3_realloc64(p, GEOPOLY_SZ(4)); + if( pOut==0 ){ + sqlite3_free(p); + if( context ) sqlite3_result_error_nomem(context); + if( pRc ) *pRc = SQLITE_NOMEM; + return 0; + } + pOut->nVertex = 4; + ii = 1; + pOut->hdr[0] = *(unsigned char*)ⅈ + pOut->hdr[1] = 0; + pOut->hdr[2] = 0; + pOut->hdr[3] = 4; + GeoX(pOut,0) = mnX; + GeoY(pOut,0) = mnY; + GeoX(pOut,1) = mxX; + GeoY(pOut,1) = mnY; + GeoX(pOut,2) = mxX; + GeoY(pOut,2) = mxY; + GeoX(pOut,3) = mnX; + GeoY(pOut,3) = mxY; + }else{ + sqlite3_free(p); + aCoord[0].f = mnX; + aCoord[1].f = mxX; + aCoord[2].f = mnY; + aCoord[3].f = mxY; + } + } + return pOut; +} + +/* +** Implementation of the geopoly_bbox(X) SQL function. +*/ +static void geopolyBBoxFunc( + sqlite3_context *context, + int argc, + sqlite3_value **argv +){ + GeoPoly *p = geopolyBBox(context, argv[0], 0, 0); + if( p ){ + sqlite3_result_blob(context, p->hdr, + 4+8*p->nVertex, SQLITE_TRANSIENT); + sqlite3_free(p); + } +} + +/* +** State vector for the geopoly_group_bbox() aggregate function. +*/ +typedef struct GeoBBox GeoBBox; +struct GeoBBox { + int isInit; + RtreeCoord a[4]; +}; + + +/* +** Implementation of the geopoly_group_bbox(X) aggregate SQL function. +*/ +static void geopolyBBoxStep( + sqlite3_context *context, + int argc, + sqlite3_value **argv +){ + RtreeCoord a[4]; + int rc = SQLITE_OK; + (void)geopolyBBox(context, argv[0], a, &rc); + if( rc==SQLITE_OK ){ + GeoBBox *pBBox; + pBBox = (GeoBBox*)sqlite3_aggregate_context(context, sizeof(*pBBox)); + if( pBBox==0 ) return; + if( pBBox->isInit==0 ){ + pBBox->isInit = 1; + memcpy(pBBox->a, a, sizeof(RtreeCoord)*4); + }else{ + if( a[0].f < pBBox->a[0].f ) pBBox->a[0] = a[0]; + if( a[1].f > pBBox->a[1].f ) pBBox->a[1] = a[1]; + if( a[2].f < pBBox->a[2].f ) pBBox->a[2] = a[2]; + if( a[3].f > pBBox->a[3].f ) pBBox->a[3] = a[3]; + } + } +} +static void geopolyBBoxFinal( + sqlite3_context *context +){ + GeoPoly *p; + GeoBBox *pBBox; + pBBox = (GeoBBox*)sqlite3_aggregate_context(context, 0); + if( pBBox==0 ) return; + p = geopolyBBox(context, 0, pBBox->a, 0); + if( p ){ + sqlite3_result_blob(context, p->hdr, + 4+8*p->nVertex, SQLITE_TRANSIENT); + sqlite3_free(p); + } +} + + +/* +** Determine if point (x0,y0) is beneath line segment (x1,y1)->(x2,y2). +** Returns: +** +** +2 x0,y0 is on the line segement +** +** +1 x0,y0 is beneath line segment +** +** 0 x0,y0 is not on or beneath the line segment or the line segment +** is vertical and x0,y0 is not on the line segment +** +** The left-most coordinate min(x1,x2) is not considered to be part of +** the line segment for the purposes of this analysis. +*/ +static int pointBeneathLine( + double x0, double y0, + double x1, double y1, + double x2, double y2 +){ + double y; + if( x0==x1 && y0==y1 ) return 2; + if( x1x2 ) return 0; + }else if( x1>x2 ){ + if( x0<=x2 || x0>x1 ) return 0; + }else{ + /* Vertical line segment */ + if( x0!=x1 ) return 0; + if( y0y1 && y0>y2 ) return 0; + return 2; + } + y = y1 + (y2-y1)*(x0-x1)/(x2-x1); + if( y0==y ) return 2; + if( y0nVertex-1; ii++){ + v = pointBeneathLine(x0,y0,GeoX(p1,ii), GeoY(p1,ii), + GeoX(p1,ii+1),GeoY(p1,ii+1)); + if( v==2 ) break; + cnt += v; + } + if( v!=2 ){ + v = pointBeneathLine(x0,y0,GeoX(p1,ii), GeoY(p1,ii), + GeoX(p1,0), GeoY(p1,0)); + } + if( v==2 ){ + sqlite3_result_int(context, 1); + }else if( ((v+cnt)&1)==0 ){ + sqlite3_result_int(context, 0); + }else{ + sqlite3_result_int(context, 2); + } + sqlite3_free(p1); +} + +/* Forward declaration */ +static int geopolyOverlap(GeoPoly *p1, GeoPoly *p2); + +/* +** SQL function: geopoly_within(P1,P2) +** +** Return +2 if P1 and P2 are the same polygon +** Return +1 if P2 is contained within P1 +** Return 0 if any part of P2 is on the outside of P1 +** +*/ +static void geopolyWithinFunc( + sqlite3_context *context, + int argc, + sqlite3_value **argv +){ + GeoPoly *p1 = geopolyFuncParam(context, argv[0], 0); + GeoPoly *p2 = geopolyFuncParam(context, argv[1], 0); + if( p1 && p2 ){ + int x = geopolyOverlap(p1, p2); + if( x<0 ){ + sqlite3_result_error_nomem(context); + }else{ + sqlite3_result_int(context, x==2 ? 1 : x==4 ? 2 : 0); + } + } + sqlite3_free(p1); + sqlite3_free(p2); +} + +/* Objects used by the overlap algorihm. */ +typedef struct GeoEvent GeoEvent; +typedef struct GeoSegment GeoSegment; +typedef struct GeoOverlap GeoOverlap; +struct GeoEvent { + double x; /* X coordinate at which event occurs */ + int eType; /* 0 for ADD, 1 for REMOVE */ + GeoSegment *pSeg; /* The segment to be added or removed */ + GeoEvent *pNext; /* Next event in the sorted list */ +}; +struct GeoSegment { + double C, B; /* y = C*x + B */ + double y; /* Current y value */ + float y0; /* Initial y value */ + unsigned char side; /* 1 for p1, 2 for p2 */ + unsigned int idx; /* Which segment within the side */ + GeoSegment *pNext; /* Next segment in a list sorted by y */ +}; +struct GeoOverlap { + GeoEvent *aEvent; /* Array of all events */ + GeoSegment *aSegment; /* Array of all segments */ + int nEvent; /* Number of events */ + int nSegment; /* Number of segments */ +}; + +/* +** Add a single segment and its associated events. +*/ +static void geopolyAddOneSegment( + GeoOverlap *p, + GeoCoord x0, + GeoCoord y0, + GeoCoord x1, + GeoCoord y1, + unsigned char side, + unsigned int idx +){ + GeoSegment *pSeg; + GeoEvent *pEvent; + if( x0==x1 ) return; /* Ignore vertical segments */ + if( x0>x1 ){ + GeoCoord t = x0; + x0 = x1; + x1 = t; + t = y0; + y0 = y1; + y1 = t; + } + pSeg = p->aSegment + p->nSegment; + p->nSegment++; + pSeg->C = (y1-y0)/(x1-x0); + pSeg->B = y1 - x1*pSeg->C; + pSeg->y0 = y0; + pSeg->side = side; + pSeg->idx = idx; + pEvent = p->aEvent + p->nEvent; + p->nEvent++; + pEvent->x = x0; + pEvent->eType = 0; + pEvent->pSeg = pSeg; + pEvent = p->aEvent + p->nEvent; + p->nEvent++; + pEvent->x = x1; + pEvent->eType = 1; + pEvent->pSeg = pSeg; +} + + + +/* +** Insert all segments and events for polygon pPoly. +*/ +static void geopolyAddSegments( + GeoOverlap *p, /* Add segments to this Overlap object */ + GeoPoly *pPoly, /* Take all segments from this polygon */ + unsigned char side /* The side of pPoly */ +){ + unsigned int i; + GeoCoord *x; + for(i=0; i<(unsigned)pPoly->nVertex-1; i++){ + x = &GeoX(pPoly,i); + geopolyAddOneSegment(p, x[0], x[1], x[2], x[3], side, i); + } + x = &GeoX(pPoly,i); + geopolyAddOneSegment(p, x[0], x[1], pPoly->a[0], pPoly->a[1], side, i); +} + +/* +** Merge two lists of sorted events by X coordinate +*/ +static GeoEvent *geopolyEventMerge(GeoEvent *pLeft, GeoEvent *pRight){ + GeoEvent head, *pLast; + head.pNext = 0; + pLast = &head; + while( pRight && pLeft ){ + if( pRight->x <= pLeft->x ){ + pLast->pNext = pRight; + pLast = pRight; + pRight = pRight->pNext; + }else{ + pLast->pNext = pLeft; + pLast = pLeft; + pLeft = pLeft->pNext; + } + } + pLast->pNext = pRight ? pRight : pLeft; + return head.pNext; +} + +/* +** Sort an array of nEvent event objects into a list. +*/ +static GeoEvent *geopolySortEventsByX(GeoEvent *aEvent, int nEvent){ + int mx = 0; + int i, j; + GeoEvent *p; + GeoEvent *a[50]; + for(i=0; ipNext = 0; + for(j=0; j=mx ) mx = j+1; + } + p = 0; + for(i=0; iy - pLeft->y; + if( r==0.0 ) r = pRight->C - pLeft->C; + if( r<0.0 ){ + pLast->pNext = pRight; + pLast = pRight; + pRight = pRight->pNext; + }else{ + pLast->pNext = pLeft; + pLast = pLeft; + pLeft = pLeft->pNext; + } + } + pLast->pNext = pRight ? pRight : pLeft; + return head.pNext; +} + +/* +** Sort a list of GeoSegments in order of increasing Y and in the event of +** a tie, increasing C (slope). +*/ +static GeoSegment *geopolySortSegmentsByYAndC(GeoSegment *pList){ + int mx = 0; + int i; + GeoSegment *p; + GeoSegment *a[50]; + while( pList ){ + p = pList; + pList = pList->pNext; + p->pNext = 0; + for(i=0; i=mx ) mx = i+1; + } + p = 0; + for(i=0; inVertex + p2->nVertex + 2; + GeoOverlap *p; + sqlite3_int64 nByte; + GeoEvent *pThisEvent; + double rX; + int rc = 0; + int needSort = 0; + GeoSegment *pActive = 0; + GeoSegment *pSeg; + unsigned char aOverlap[4]; + + nByte = sizeof(GeoEvent)*nVertex*2 + + sizeof(GeoSegment)*nVertex + + sizeof(GeoOverlap); + p = sqlite3_malloc64( nByte ); + if( p==0 ) return -1; + p->aEvent = (GeoEvent*)&p[1]; + p->aSegment = (GeoSegment*)&p->aEvent[nVertex*2]; + p->nEvent = p->nSegment = 0; + geopolyAddSegments(p, p1, 1); + geopolyAddSegments(p, p2, 2); + pThisEvent = geopolySortEventsByX(p->aEvent, p->nEvent); + rX = pThisEvent->x==0.0 ? -1.0 : 0.0; + memset(aOverlap, 0, sizeof(aOverlap)); + while( pThisEvent ){ + if( pThisEvent->x!=rX ){ + GeoSegment *pPrev = 0; + int iMask = 0; + GEODEBUG(("Distinct X: %g\n", pThisEvent->x)); + rX = pThisEvent->x; + if( needSort ){ + GEODEBUG(("SORT\n")); + pActive = geopolySortSegmentsByYAndC(pActive); + needSort = 0; + } + for(pSeg=pActive; pSeg; pSeg=pSeg->pNext){ + if( pPrev ){ + if( pPrev->y!=pSeg->y ){ + GEODEBUG(("MASK: %d\n", iMask)); + aOverlap[iMask] = 1; + } + } + iMask ^= pSeg->side; + pPrev = pSeg; + } + pPrev = 0; + for(pSeg=pActive; pSeg; pSeg=pSeg->pNext){ + double y = pSeg->C*rX + pSeg->B; + GEODEBUG(("Segment %d.%d %g->%g\n", pSeg->side, pSeg->idx, pSeg->y, y)); + pSeg->y = y; + if( pPrev ){ + if( pPrev->y>pSeg->y && pPrev->side!=pSeg->side ){ + rc = 1; + GEODEBUG(("Crossing: %d.%d and %d.%d\n", + pPrev->side, pPrev->idx, + pSeg->side, pSeg->idx)); + goto geopolyOverlapDone; + }else if( pPrev->y!=pSeg->y ){ + GEODEBUG(("MASK: %d\n", iMask)); + aOverlap[iMask] = 1; + } + } + iMask ^= pSeg->side; + pPrev = pSeg; + } + } + GEODEBUG(("%s %d.%d C=%g B=%g\n", + pThisEvent->eType ? "RM " : "ADD", + pThisEvent->pSeg->side, pThisEvent->pSeg->idx, + pThisEvent->pSeg->C, + pThisEvent->pSeg->B)); + if( pThisEvent->eType==0 ){ + /* Add a segment */ + pSeg = pThisEvent->pSeg; + pSeg->y = pSeg->y0; + pSeg->pNext = pActive; + pActive = pSeg; + needSort = 1; + }else{ + /* Remove a segment */ + if( pActive==pThisEvent->pSeg ){ + pActive = pActive->pNext; + }else{ + for(pSeg=pActive; pSeg; pSeg=pSeg->pNext){ + if( pSeg->pNext==pThisEvent->pSeg ){ + pSeg->pNext = pSeg->pNext->pNext; + break; + } + } + } + } + pThisEvent = pThisEvent->pNext; + } + if( aOverlap[3]==0 ){ + rc = 0; + }else if( aOverlap[1]!=0 && aOverlap[2]==0 ){ + rc = 3; + }else if( aOverlap[1]==0 && aOverlap[2]!=0 ){ + rc = 2; + }else if( aOverlap[1]==0 && aOverlap[2]==0 ){ + rc = 4; + }else{ + rc = 1; + } + +geopolyOverlapDone: + sqlite3_free(p); + return rc; +} + +/* +** SQL function: geopoly_overlap(P1,P2) +** +** Determine whether or not P1 and P2 overlap. Return value: +** +** 0 The two polygons are disjoint +** 1 They overlap +** 2 P1 is completely contained within P2 +** 3 P2 is completely contained within P1 +** 4 P1 and P2 are the same polygon +** NULL Either P1 or P2 or both are not valid polygons +*/ +static void geopolyOverlapFunc( + sqlite3_context *context, + int argc, + sqlite3_value **argv +){ + GeoPoly *p1 = geopolyFuncParam(context, argv[0], 0); + GeoPoly *p2 = geopolyFuncParam(context, argv[1], 0); + if( p1 && p2 ){ + int x = geopolyOverlap(p1, p2); + if( x<0 ){ + sqlite3_result_error_nomem(context); + }else{ + sqlite3_result_int(context, x); + } + } + sqlite3_free(p1); + sqlite3_free(p2); +} + +/* +** Enable or disable debugging output +*/ +static void geopolyDebugFunc( + sqlite3_context *context, + int argc, + sqlite3_value **argv +){ +#ifdef GEOPOLY_ENABLE_DEBUG + geo_debug = sqlite3_value_int(argv[0]); +#endif +} + +/* +** This function is the implementation of both the xConnect and xCreate +** methods of the geopoly virtual table. +** +** argv[0] -> module name +** argv[1] -> database name +** argv[2] -> table name +** argv[...] -> column names... +*/ +static int geopolyInit( + sqlite3 *db, /* Database connection */ + void *pAux, /* One of the RTREE_COORD_* constants */ + int argc, const char *const*argv, /* Parameters to CREATE TABLE statement */ + sqlite3_vtab **ppVtab, /* OUT: New virtual table */ + char **pzErr, /* OUT: Error message, if any */ + int isCreate /* True for xCreate, false for xConnect */ +){ + int rc = SQLITE_OK; + Rtree *pRtree; + sqlite3_int64 nDb; /* Length of string argv[1] */ + sqlite3_int64 nName; /* Length of string argv[2] */ + sqlite3_str *pSql; + char *zSql; + int ii; + + sqlite3_vtab_config(db, SQLITE_VTAB_CONSTRAINT_SUPPORT, 1); + + /* Allocate the sqlite3_vtab structure */ + nDb = strlen(argv[1]); + nName = strlen(argv[2]); + pRtree = (Rtree *)sqlite3_malloc64(sizeof(Rtree)+nDb+nName+2); + if( !pRtree ){ + return SQLITE_NOMEM; + } + memset(pRtree, 0, sizeof(Rtree)+nDb+nName+2); + pRtree->nBusy = 1; + pRtree->base.pModule = &rtreeModule; + pRtree->zDb = (char *)&pRtree[1]; + pRtree->zName = &pRtree->zDb[nDb+1]; + pRtree->eCoordType = RTREE_COORD_REAL32; + pRtree->nDim = 2; + pRtree->nDim2 = 4; + memcpy(pRtree->zDb, argv[1], nDb); + memcpy(pRtree->zName, argv[2], nName); + + + /* Create/Connect to the underlying relational database schema. If + ** that is successful, call sqlite3_declare_vtab() to configure + ** the r-tree table schema. + */ + pSql = sqlite3_str_new(db); + sqlite3_str_appendf(pSql, "CREATE TABLE x(_shape"); + pRtree->nAux = 1; /* Add one for _shape */ + pRtree->nAuxNotNull = 1; /* The _shape column is always not-null */ + for(ii=3; iinAux++; + sqlite3_str_appendf(pSql, ",%s", argv[ii]); + } + sqlite3_str_appendf(pSql, ");"); + zSql = sqlite3_str_finish(pSql); + if( !zSql ){ + rc = SQLITE_NOMEM; + }else if( SQLITE_OK!=(rc = sqlite3_declare_vtab(db, zSql)) ){ + *pzErr = sqlite3_mprintf("%s", sqlite3_errmsg(db)); + } + sqlite3_free(zSql); + if( rc ) goto geopolyInit_fail; + pRtree->nBytesPerCell = 8 + pRtree->nDim2*4; + + /* Figure out the node size to use. */ + rc = getNodeSize(db, pRtree, isCreate, pzErr); + if( rc ) goto geopolyInit_fail; + rc = rtreeSqlInit(pRtree, db, argv[1], argv[2], isCreate); + if( rc ){ + *pzErr = sqlite3_mprintf("%s", sqlite3_errmsg(db)); + goto geopolyInit_fail; + } + + *ppVtab = (sqlite3_vtab *)pRtree; + return SQLITE_OK; + +geopolyInit_fail: + if( rc==SQLITE_OK ) rc = SQLITE_ERROR; + assert( *ppVtab==0 ); + assert( pRtree->nBusy==1 ); + rtreeRelease(pRtree); + return rc; +} + + +/* +** GEOPOLY virtual table module xCreate method. +*/ +static int geopolyCreate( + sqlite3 *db, + void *pAux, + int argc, const char *const*argv, + sqlite3_vtab **ppVtab, + char **pzErr +){ + return geopolyInit(db, pAux, argc, argv, ppVtab, pzErr, 1); +} + +/* +** GEOPOLY virtual table module xConnect method. +*/ +static int geopolyConnect( + sqlite3 *db, + void *pAux, + int argc, const char *const*argv, + sqlite3_vtab **ppVtab, + char **pzErr +){ + return geopolyInit(db, pAux, argc, argv, ppVtab, pzErr, 0); +} + + +/* +** GEOPOLY virtual table module xFilter method. +** +** Query plans: +** +** 1 rowid lookup +** 2 search for objects overlapping the same bounding box +** that contains polygon argv[0] +** 3 search for objects overlapping the same bounding box +** that contains polygon argv[0] +** 4 full table scan +*/ +static int geopolyFilter( + sqlite3_vtab_cursor *pVtabCursor, /* The cursor to initialize */ + int idxNum, /* Query plan */ + const char *idxStr, /* Not Used */ + int argc, sqlite3_value **argv /* Parameters to the query plan */ +){ + Rtree *pRtree = (Rtree *)pVtabCursor->pVtab; + RtreeCursor *pCsr = (RtreeCursor *)pVtabCursor; + RtreeNode *pRoot = 0; + int rc = SQLITE_OK; + int iCell = 0; + sqlite3_stmt *pStmt; + + rtreeReference(pRtree); + + /* Reset the cursor to the same state as rtreeOpen() leaves it in. */ + freeCursorConstraints(pCsr); + sqlite3_free(pCsr->aPoint); + pStmt = pCsr->pReadAux; + memset(pCsr, 0, sizeof(RtreeCursor)); + pCsr->base.pVtab = (sqlite3_vtab*)pRtree; + pCsr->pReadAux = pStmt; + + pCsr->iStrategy = idxNum; + if( idxNum==1 ){ + /* Special case - lookup by rowid. */ + RtreeNode *pLeaf; /* Leaf on which the required cell resides */ + RtreeSearchPoint *p; /* Search point for the leaf */ + i64 iRowid = sqlite3_value_int64(argv[0]); + i64 iNode = 0; + rc = findLeafNode(pRtree, iRowid, &pLeaf, &iNode); + if( rc==SQLITE_OK && pLeaf!=0 ){ + p = rtreeSearchPointNew(pCsr, RTREE_ZERO, 0); + assert( p!=0 ); /* Always returns pCsr->sPoint */ + pCsr->aNode[0] = pLeaf; + p->id = iNode; + p->eWithin = PARTLY_WITHIN; + rc = nodeRowidIndex(pRtree, pLeaf, iRowid, &iCell); + p->iCell = (u8)iCell; + RTREE_QUEUE_TRACE(pCsr, "PUSH-F1:"); + }else{ + pCsr->atEOF = 1; + } + }else{ + /* Normal case - r-tree scan. Set up the RtreeCursor.aConstraint array + ** with the configured constraints. + */ + rc = nodeAcquire(pRtree, 1, 0, &pRoot); + if( rc==SQLITE_OK && idxNum<=3 ){ + RtreeCoord bbox[4]; + RtreeConstraint *p; + assert( argc==1 ); + geopolyBBox(0, argv[0], bbox, &rc); + if( rc ){ + goto geopoly_filter_end; + } + pCsr->aConstraint = p = sqlite3_malloc(sizeof(RtreeConstraint)*4); + pCsr->nConstraint = 4; + if( p==0 ){ + rc = SQLITE_NOMEM; + }else{ + memset(pCsr->aConstraint, 0, sizeof(RtreeConstraint)*4); + memset(pCsr->anQueue, 0, sizeof(u32)*(pRtree->iDepth + 1)); + if( idxNum==2 ){ + /* Overlap query */ + p->op = 'B'; + p->iCoord = 0; + p->u.rValue = bbox[1].f; + p++; + p->op = 'D'; + p->iCoord = 1; + p->u.rValue = bbox[0].f; + p++; + p->op = 'B'; + p->iCoord = 2; + p->u.rValue = bbox[3].f; + p++; + p->op = 'D'; + p->iCoord = 3; + p->u.rValue = bbox[2].f; + }else{ + /* Within query */ + p->op = 'D'; + p->iCoord = 0; + p->u.rValue = bbox[0].f; + p++; + p->op = 'B'; + p->iCoord = 1; + p->u.rValue = bbox[1].f; + p++; + p->op = 'D'; + p->iCoord = 2; + p->u.rValue = bbox[2].f; + p++; + p->op = 'B'; + p->iCoord = 3; + p->u.rValue = bbox[3].f; + } + } + } + if( rc==SQLITE_OK ){ + RtreeSearchPoint *pNew; + pNew = rtreeSearchPointNew(pCsr, RTREE_ZERO, (u8)(pRtree->iDepth+1)); + if( pNew==0 ){ + rc = SQLITE_NOMEM; + goto geopoly_filter_end; + } + pNew->id = 1; + pNew->iCell = 0; + pNew->eWithin = PARTLY_WITHIN; + assert( pCsr->bPoint==1 ); + pCsr->aNode[0] = pRoot; + pRoot = 0; + RTREE_QUEUE_TRACE(pCsr, "PUSH-Fm:"); + rc = rtreeStepToLeaf(pCsr); + } + } + +geopoly_filter_end: + nodeRelease(pRtree, pRoot); + rtreeRelease(pRtree); + return rc; +} + +/* +** Rtree virtual table module xBestIndex method. There are three +** table scan strategies to choose from (in order from most to +** least desirable): +** +** idxNum idxStr Strategy +** ------------------------------------------------ +** 1 "rowid" Direct lookup by rowid. +** 2 "rtree" R-tree overlap query using geopoly_overlap() +** 3 "rtree" R-tree within query using geopoly_within() +** 4 "fullscan" full-table scan. +** ------------------------------------------------ +*/ +static int geopolyBestIndex(sqlite3_vtab *tab, sqlite3_index_info *pIdxInfo){ + int ii; + int iRowidTerm = -1; + int iFuncTerm = -1; + int idxNum = 0; + + for(ii=0; iinConstraint; ii++){ + struct sqlite3_index_constraint *p = &pIdxInfo->aConstraint[ii]; + if( !p->usable ) continue; + if( p->iColumn<0 && p->op==SQLITE_INDEX_CONSTRAINT_EQ ){ + iRowidTerm = ii; + break; + } + if( p->iColumn==0 && p->op>=SQLITE_INDEX_CONSTRAINT_FUNCTION ){ + /* p->op==SQLITE_INDEX_CONSTRAINT_FUNCTION for geopoly_overlap() + ** p->op==(SQLITE_INDEX_CONTRAINT_FUNCTION+1) for geopoly_within(). + ** See geopolyFindFunction() */ + iFuncTerm = ii; + idxNum = p->op - SQLITE_INDEX_CONSTRAINT_FUNCTION + 2; + } + } + + if( iRowidTerm>=0 ){ + pIdxInfo->idxNum = 1; + pIdxInfo->idxStr = "rowid"; + pIdxInfo->aConstraintUsage[iRowidTerm].argvIndex = 1; + pIdxInfo->aConstraintUsage[iRowidTerm].omit = 1; + pIdxInfo->estimatedCost = 30.0; + pIdxInfo->estimatedRows = 1; + pIdxInfo->idxFlags = SQLITE_INDEX_SCAN_UNIQUE; + return SQLITE_OK; + } + if( iFuncTerm>=0 ){ + pIdxInfo->idxNum = idxNum; + pIdxInfo->idxStr = "rtree"; + pIdxInfo->aConstraintUsage[iFuncTerm].argvIndex = 1; + pIdxInfo->aConstraintUsage[iFuncTerm].omit = 0; + pIdxInfo->estimatedCost = 300.0; + pIdxInfo->estimatedRows = 10; + return SQLITE_OK; + } + pIdxInfo->idxNum = 4; + pIdxInfo->idxStr = "fullscan"; + pIdxInfo->estimatedCost = 3000000.0; + pIdxInfo->estimatedRows = 100000; + return SQLITE_OK; +} + + +/* +** GEOPOLY virtual table module xColumn method. +*/ +static int geopolyColumn(sqlite3_vtab_cursor *cur, sqlite3_context *ctx, int i){ + Rtree *pRtree = (Rtree *)cur->pVtab; + RtreeCursor *pCsr = (RtreeCursor *)cur; + RtreeSearchPoint *p = rtreeSearchPointFirst(pCsr); + int rc = SQLITE_OK; + RtreeNode *pNode = rtreeNodeOfFirstSearchPoint(pCsr, &rc); + + if( rc ) return rc; + if( p==0 ) return SQLITE_OK; + if( i==0 && sqlite3_vtab_nochange(ctx) ) return SQLITE_OK; + if( i<=pRtree->nAux ){ + if( !pCsr->bAuxValid ){ + if( pCsr->pReadAux==0 ){ + rc = sqlite3_prepare_v3(pRtree->db, pRtree->zReadAuxSql, -1, 0, + &pCsr->pReadAux, 0); + if( rc ) return rc; + } + sqlite3_bind_int64(pCsr->pReadAux, 1, + nodeGetRowid(pRtree, pNode, p->iCell)); + rc = sqlite3_step(pCsr->pReadAux); + if( rc==SQLITE_ROW ){ + pCsr->bAuxValid = 1; + }else{ + sqlite3_reset(pCsr->pReadAux); + if( rc==SQLITE_DONE ) rc = SQLITE_OK; + return rc; + } + } + sqlite3_result_value(ctx, sqlite3_column_value(pCsr->pReadAux, i+2)); + } + return SQLITE_OK; +} + + +/* +** The xUpdate method for GEOPOLY module virtual tables. +** +** For DELETE: +** +** argv[0] = the rowid to be deleted +** +** For INSERT: +** +** argv[0] = SQL NULL +** argv[1] = rowid to insert, or an SQL NULL to select automatically +** argv[2] = _shape column +** argv[3] = first application-defined column.... +** +** For UPDATE: +** +** argv[0] = rowid to modify. Never NULL +** argv[1] = rowid after the change. Never NULL +** argv[2] = new value for _shape +** argv[3] = new value for first application-defined column.... +*/ +static int geopolyUpdate( + sqlite3_vtab *pVtab, + int nData, + sqlite3_value **aData, + sqlite_int64 *pRowid +){ + Rtree *pRtree = (Rtree *)pVtab; + int rc = SQLITE_OK; + RtreeCell cell; /* New cell to insert if nData>1 */ + i64 oldRowid; /* The old rowid */ + int oldRowidValid; /* True if oldRowid is valid */ + i64 newRowid; /* The new rowid */ + int newRowidValid; /* True if newRowid is valid */ + int coordChange = 0; /* Change in coordinates */ + + if( pRtree->nNodeRef ){ + /* Unable to write to the btree while another cursor is reading from it, + ** since the write might do a rebalance which would disrupt the read + ** cursor. */ + return SQLITE_LOCKED_VTAB; + } + rtreeReference(pRtree); + assert(nData>=1); + + oldRowidValid = sqlite3_value_type(aData[0])!=SQLITE_NULL;; + oldRowid = oldRowidValid ? sqlite3_value_int64(aData[0]) : 0; + newRowidValid = nData>1 && sqlite3_value_type(aData[1])!=SQLITE_NULL; + newRowid = newRowidValid ? sqlite3_value_int64(aData[1]) : 0; + cell.iRowid = newRowid; + + if( nData>1 /* not a DELETE */ + && (!oldRowidValid /* INSERT */ + || !sqlite3_value_nochange(aData[2]) /* UPDATE _shape */ + || oldRowid!=newRowid) /* Rowid change */ + ){ + geopolyBBox(0, aData[2], cell.aCoord, &rc); + if( rc ){ + if( rc==SQLITE_ERROR ){ + pVtab->zErrMsg = + sqlite3_mprintf("_shape does not contain a valid polygon"); + } + goto geopoly_update_end; + } + coordChange = 1; + + /* If a rowid value was supplied, check if it is already present in + ** the table. If so, the constraint has failed. */ + if( newRowidValid && (!oldRowidValid || oldRowid!=newRowid) ){ + int steprc; + sqlite3_bind_int64(pRtree->pReadRowid, 1, cell.iRowid); + steprc = sqlite3_step(pRtree->pReadRowid); + rc = sqlite3_reset(pRtree->pReadRowid); + if( SQLITE_ROW==steprc ){ + if( sqlite3_vtab_on_conflict(pRtree->db)==SQLITE_REPLACE ){ + rc = rtreeDeleteRowid(pRtree, cell.iRowid); + }else{ + rc = rtreeConstraintError(pRtree, 0); + } + } + } + } + + /* If aData[0] is not an SQL NULL value, it is the rowid of a + ** record to delete from the r-tree table. The following block does + ** just that. + */ + if( rc==SQLITE_OK && (nData==1 || (coordChange && oldRowidValid)) ){ + rc = rtreeDeleteRowid(pRtree, oldRowid); + } + + /* If the aData[] array contains more than one element, elements + ** (aData[2]..aData[argc-1]) contain a new record to insert into + ** the r-tree structure. + */ + if( rc==SQLITE_OK && nData>1 && coordChange ){ + /* Insert the new record into the r-tree */ + RtreeNode *pLeaf = 0; + if( !newRowidValid ){ + rc = rtreeNewRowid(pRtree, &cell.iRowid); + } + *pRowid = cell.iRowid; + if( rc==SQLITE_OK ){ + rc = ChooseLeaf(pRtree, &cell, 0, &pLeaf); + } + if( rc==SQLITE_OK ){ + int rc2; + pRtree->iReinsertHeight = -1; + rc = rtreeInsertCell(pRtree, pLeaf, &cell, 0); + rc2 = nodeRelease(pRtree, pLeaf); + if( rc==SQLITE_OK ){ + rc = rc2; + } + } + } + + /* Change the data */ + if( rc==SQLITE_OK && nData>1 ){ + sqlite3_stmt *pUp = pRtree->pWriteAux; + int jj; + int nChange = 0; + sqlite3_bind_int64(pUp, 1, cell.iRowid); + assert( pRtree->nAux>=1 ); + if( sqlite3_value_nochange(aData[2]) ){ + sqlite3_bind_null(pUp, 2); + }else{ + GeoPoly *p = 0; + if( sqlite3_value_type(aData[2])==SQLITE_TEXT + && (p = geopolyFuncParam(0, aData[2], &rc))!=0 + && rc==SQLITE_OK + ){ + sqlite3_bind_blob(pUp, 2, p->hdr, 4+8*p->nVertex, SQLITE_TRANSIENT); + }else{ + sqlite3_bind_value(pUp, 2, aData[2]); + } + sqlite3_free(p); + nChange = 1; + } + for(jj=1; jjnAux; jj++){ + nChange++; + sqlite3_bind_value(pUp, jj+2, aData[jj+2]); + } + if( nChange ){ + sqlite3_step(pUp); + rc = sqlite3_reset(pUp); + } + } + +geopoly_update_end: + rtreeRelease(pRtree); + return rc; +} + +/* +** Report that geopoly_overlap() is an overloaded function suitable +** for use in xBestIndex. +*/ +static int geopolyFindFunction( + sqlite3_vtab *pVtab, + int nArg, + const char *zName, + void (**pxFunc)(sqlite3_context*,int,sqlite3_value**), + void **ppArg +){ + if( sqlite3_stricmp(zName, "geopoly_overlap")==0 ){ + *pxFunc = geopolyOverlapFunc; + *ppArg = 0; + return SQLITE_INDEX_CONSTRAINT_FUNCTION; + } + if( sqlite3_stricmp(zName, "geopoly_within")==0 ){ + *pxFunc = geopolyWithinFunc; + *ppArg = 0; + return SQLITE_INDEX_CONSTRAINT_FUNCTION+1; + } + return 0; +} + + +static sqlite3_module geopolyModule = { + 3, /* iVersion */ + geopolyCreate, /* xCreate - create a table */ + geopolyConnect, /* xConnect - connect to an existing table */ + geopolyBestIndex, /* xBestIndex - Determine search strategy */ + rtreeDisconnect, /* xDisconnect - Disconnect from a table */ + rtreeDestroy, /* xDestroy - Drop a table */ + rtreeOpen, /* xOpen - open a cursor */ + rtreeClose, /* xClose - close a cursor */ + geopolyFilter, /* xFilter - configure scan constraints */ + rtreeNext, /* xNext - advance a cursor */ + rtreeEof, /* xEof */ + geopolyColumn, /* xColumn - read data */ + rtreeRowid, /* xRowid - read data */ + geopolyUpdate, /* xUpdate - write data */ + rtreeBeginTransaction, /* xBegin - begin transaction */ + rtreeEndTransaction, /* xSync - sync transaction */ + rtreeEndTransaction, /* xCommit - commit transaction */ + rtreeEndTransaction, /* xRollback - rollback transaction */ + geopolyFindFunction, /* xFindFunction - function overloading */ + rtreeRename, /* xRename - rename the table */ + rtreeSavepoint, /* xSavepoint */ + 0, /* xRelease */ + 0, /* xRollbackTo */ + rtreeShadowName /* xShadowName */ +}; + +static int sqlite3_geopoly_init(sqlite3 *db){ + int rc = SQLITE_OK; + static const struct { + void (*xFunc)(sqlite3_context*,int,sqlite3_value**); + signed char nArg; + unsigned char bPure; + const char *zName; + } aFunc[] = { + { geopolyAreaFunc, 1, 1, "geopoly_area" }, + { geopolyBlobFunc, 1, 1, "geopoly_blob" }, + { geopolyJsonFunc, 1, 1, "geopoly_json" }, + { geopolySvgFunc, -1, 1, "geopoly_svg" }, + { geopolyWithinFunc, 2, 1, "geopoly_within" }, + { geopolyContainsPointFunc, 3, 1, "geopoly_contains_point" }, + { geopolyOverlapFunc, 2, 1, "geopoly_overlap" }, + { geopolyDebugFunc, 1, 0, "geopoly_debug" }, + { geopolyBBoxFunc, 1, 1, "geopoly_bbox" }, + { geopolyXformFunc, 7, 1, "geopoly_xform" }, + { geopolyRegularFunc, 4, 1, "geopoly_regular" }, + { geopolyCcwFunc, 1, 1, "geopoly_ccw" }, + }; + static const struct { + void (*xStep)(sqlite3_context*,int,sqlite3_value**); + void (*xFinal)(sqlite3_context*); + const char *zName; + } aAgg[] = { + { geopolyBBoxStep, geopolyBBoxFinal, "geopoly_group_bbox" }, + }; + int i; + for(i=0; ixDestructor ) pInfo->xDestructor(pInfo->pContext); + sqlite3_free(p); +} + +/* +** This routine frees the BLOB that is returned by geomCallback(). +*/ +static void rtreeMatchArgFree(void *pArg){ + int i; + RtreeMatchArg *p = (RtreeMatchArg*)pArg; + for(i=0; inParam; i++){ + sqlite3_value_free(p->apSqlParam[i]); + } + sqlite3_free(p); +} + +/* +** Each call to sqlite3_rtree_geometry_callback() or +** sqlite3_rtree_query_callback() creates an ordinary SQLite +** scalar function that is implemented by this routine. +** +** All this function does is construct an RtreeMatchArg object that +** contains the geometry-checking callback routines and a list of +** parameters to this function, then return that RtreeMatchArg object +** as a BLOB. +** +** The R-Tree MATCH operator will read the returned BLOB, deserialize +** the RtreeMatchArg object, and use the RtreeMatchArg object to figure +** out which elements of the R-Tree should be returned by the query. +*/ +static void geomCallback(sqlite3_context *ctx, int nArg, sqlite3_value **aArg){ + RtreeGeomCallback *pGeomCtx = (RtreeGeomCallback *)sqlite3_user_data(ctx); + RtreeMatchArg *pBlob; + sqlite3_int64 nBlob; + int memErr = 0; + + nBlob = sizeof(RtreeMatchArg) + (nArg-1)*sizeof(RtreeDValue) + + nArg*sizeof(sqlite3_value*); + pBlob = (RtreeMatchArg *)sqlite3_malloc64(nBlob); + if( !pBlob ){ + sqlite3_result_error_nomem(ctx); + }else{ + int i; + pBlob->iSize = nBlob; + pBlob->cb = pGeomCtx[0]; + pBlob->apSqlParam = (sqlite3_value**)&pBlob->aParam[nArg]; + pBlob->nParam = nArg; + for(i=0; iapSqlParam[i] = sqlite3_value_dup(aArg[i]); + if( pBlob->apSqlParam[i]==0 ) memErr = 1; +#ifdef SQLITE_RTREE_INT_ONLY + pBlob->aParam[i] = sqlite3_value_int64(aArg[i]); +#else + pBlob->aParam[i] = sqlite3_value_double(aArg[i]); +#endif + } + if( memErr ){ + sqlite3_result_error_nomem(ctx); + rtreeMatchArgFree(pBlob); + }else{ + sqlite3_result_pointer(ctx, pBlob, "RtreeMatchArg", rtreeMatchArgFree); + } + } +} + +/* +** Register a new geometry function for use with the r-tree MATCH operator. +*/ +SQLITE_API int sqlite3_rtree_geometry_callback( + sqlite3 *db, /* Register SQL function on this connection */ + const char *zGeom, /* Name of the new SQL function */ + int (*xGeom)(sqlite3_rtree_geometry*,int,RtreeDValue*,int*), /* Callback */ + void *pContext /* Extra data associated with the callback */ +){ + RtreeGeomCallback *pGeomCtx; /* Context object for new user-function */ + + /* Allocate and populate the context object. */ + pGeomCtx = (RtreeGeomCallback *)sqlite3_malloc(sizeof(RtreeGeomCallback)); + if( !pGeomCtx ) return SQLITE_NOMEM; + pGeomCtx->xGeom = xGeom; + pGeomCtx->xQueryFunc = 0; + pGeomCtx->xDestructor = 0; + pGeomCtx->pContext = pContext; + return sqlite3_create_function_v2(db, zGeom, -1, SQLITE_ANY, + (void *)pGeomCtx, geomCallback, 0, 0, rtreeFreeCallback + ); +} + +/* +** Register a new 2nd-generation geometry function for use with the +** r-tree MATCH operator. +*/ +SQLITE_API int sqlite3_rtree_query_callback( + sqlite3 *db, /* Register SQL function on this connection */ + const char *zQueryFunc, /* Name of new SQL function */ + int (*xQueryFunc)(sqlite3_rtree_query_info*), /* Callback */ + void *pContext, /* Extra data passed into the callback */ + void (*xDestructor)(void*) /* Destructor for the extra data */ +){ + RtreeGeomCallback *pGeomCtx; /* Context object for new user-function */ + + /* Allocate and populate the context object. */ + pGeomCtx = (RtreeGeomCallback *)sqlite3_malloc(sizeof(RtreeGeomCallback)); + if( !pGeomCtx ) return SQLITE_NOMEM; + pGeomCtx->xGeom = 0; + pGeomCtx->xQueryFunc = xQueryFunc; + pGeomCtx->xDestructor = xDestructor; + pGeomCtx->pContext = pContext; + return sqlite3_create_function_v2(db, zQueryFunc, -1, SQLITE_ANY, + (void *)pGeomCtx, geomCallback, 0, 0, rtreeFreeCallback + ); +} + +#if !SQLITE_CORE +#ifdef _WIN32 +__declspec(dllexport) +#endif +SQLITE_API int sqlite3_rtree_init( + sqlite3 *db, + char **pzErrMsg, + const sqlite3_api_routines *pApi +){ + SQLITE_EXTENSION_INIT2(pApi) + return sqlite3RtreeInit(db); +} +#endif + +#endif + +/************** End of rtree.c ***********************************************/ +/************** Begin file icu.c *********************************************/ +/* +** 2007 May 6 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** $Id: icu.c,v 1.7 2007/12/13 21:54:11 drh Exp $ +** +** This file implements an integration between the ICU library +** ("International Components for Unicode", an open-source library +** for handling unicode data) and SQLite. The integration uses +** ICU to provide the following to SQLite: +** +** * An implementation of the SQL regexp() function (and hence REGEXP +** operator) using the ICU uregex_XX() APIs. +** +** * Implementations of the SQL scalar upper() and lower() functions +** for case mapping. +** +** * Integration of ICU and SQLite collation sequences. +** +** * An implementation of the LIKE operator that uses ICU to +** provide case-independent matching. +*/ + +#if !defined(SQLITE_CORE) \ + || defined(SQLITE_ENABLE_ICU) \ + || defined(SQLITE_ENABLE_ICU_COLLATIONS) + +/* Include ICU headers */ +#include +#include +#include +#include + +/* #include */ + +#ifndef SQLITE_CORE +/* #include "sqlite3ext.h" */ + SQLITE_EXTENSION_INIT1 +#else +/* #include "sqlite3.h" */ +#endif + +/* +** This function is called when an ICU function called from within +** the implementation of an SQL scalar function returns an error. +** +** The scalar function context passed as the first argument is +** loaded with an error message based on the following two args. +*/ +static void icuFunctionError( + sqlite3_context *pCtx, /* SQLite scalar function context */ + const char *zName, /* Name of ICU function that failed */ + UErrorCode e /* Error code returned by ICU function */ +){ + char zBuf[128]; + sqlite3_snprintf(128, zBuf, "ICU error: %s(): %s", zName, u_errorName(e)); + zBuf[127] = '\0'; + sqlite3_result_error(pCtx, zBuf, -1); +} + +#if !defined(SQLITE_CORE) || defined(SQLITE_ENABLE_ICU) + +/* +** Maximum length (in bytes) of the pattern in a LIKE or GLOB +** operator. +*/ +#ifndef SQLITE_MAX_LIKE_PATTERN_LENGTH +# define SQLITE_MAX_LIKE_PATTERN_LENGTH 50000 +#endif + +/* +** Version of sqlite3_free() that is always a function, never a macro. +*/ +static void xFree(void *p){ + sqlite3_free(p); +} + +/* +** This lookup table is used to help decode the first byte of +** a multi-byte UTF8 character. It is copied here from SQLite source +** code file utf8.c. +*/ +static const unsigned char icuUtf8Trans1[] = { + 0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07, + 0x08, 0x09, 0x0a, 0x0b, 0x0c, 0x0d, 0x0e, 0x0f, + 0x10, 0x11, 0x12, 0x13, 0x14, 0x15, 0x16, 0x17, + 0x18, 0x19, 0x1a, 0x1b, 0x1c, 0x1d, 0x1e, 0x1f, + 0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07, + 0x08, 0x09, 0x0a, 0x0b, 0x0c, 0x0d, 0x0e, 0x0f, + 0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07, + 0x00, 0x01, 0x02, 0x03, 0x00, 0x01, 0x00, 0x00, +}; + +#define SQLITE_ICU_READ_UTF8(zIn, c) \ + c = *(zIn++); \ + if( c>=0xc0 ){ \ + c = icuUtf8Trans1[c-0xc0]; \ + while( (*zIn & 0xc0)==0x80 ){ \ + c = (c<<6) + (0x3f & *(zIn++)); \ + } \ + } + +#define SQLITE_ICU_SKIP_UTF8(zIn) \ + assert( *zIn ); \ + if( *(zIn++)>=0xc0 ){ \ + while( (*zIn & 0xc0)==0x80 ){zIn++;} \ + } + + +/* +** Compare two UTF-8 strings for equality where the first string is +** a "LIKE" expression. Return true (1) if they are the same and +** false (0) if they are different. +*/ +static int icuLikeCompare( + const uint8_t *zPattern, /* LIKE pattern */ + const uint8_t *zString, /* The UTF-8 string to compare against */ + const UChar32 uEsc /* The escape character */ +){ + static const uint32_t MATCH_ONE = (uint32_t)'_'; + static const uint32_t MATCH_ALL = (uint32_t)'%'; + + int prevEscape = 0; /* True if the previous character was uEsc */ + + while( 1 ){ + + /* Read (and consume) the next character from the input pattern. */ + uint32_t uPattern; + SQLITE_ICU_READ_UTF8(zPattern, uPattern); + if( uPattern==0 ) break; + + /* There are now 4 possibilities: + ** + ** 1. uPattern is an unescaped match-all character "%", + ** 2. uPattern is an unescaped match-one character "_", + ** 3. uPattern is an unescaped escape character, or + ** 4. uPattern is to be handled as an ordinary character + */ + if( !prevEscape && uPattern==MATCH_ALL ){ + /* Case 1. */ + uint8_t c; + + /* Skip any MATCH_ALL or MATCH_ONE characters that follow a + ** MATCH_ALL. For each MATCH_ONE, skip one character in the + ** test string. + */ + while( (c=*zPattern) == MATCH_ALL || c == MATCH_ONE ){ + if( c==MATCH_ONE ){ + if( *zString==0 ) return 0; + SQLITE_ICU_SKIP_UTF8(zString); + } + zPattern++; + } + + if( *zPattern==0 ) return 1; + + while( *zString ){ + if( icuLikeCompare(zPattern, zString, uEsc) ){ + return 1; + } + SQLITE_ICU_SKIP_UTF8(zString); + } + return 0; + + }else if( !prevEscape && uPattern==MATCH_ONE ){ + /* Case 2. */ + if( *zString==0 ) return 0; + SQLITE_ICU_SKIP_UTF8(zString); + + }else if( !prevEscape && uPattern==(uint32_t)uEsc){ + /* Case 3. */ + prevEscape = 1; + + }else{ + /* Case 4. */ + uint32_t uString; + SQLITE_ICU_READ_UTF8(zString, uString); + uString = (uint32_t)u_foldCase((UChar32)uString, U_FOLD_CASE_DEFAULT); + uPattern = (uint32_t)u_foldCase((UChar32)uPattern, U_FOLD_CASE_DEFAULT); + if( uString!=uPattern ){ + return 0; + } + prevEscape = 0; + } + } + + return *zString==0; +} + +/* +** Implementation of the like() SQL function. This function implements +** the build-in LIKE operator. The first argument to the function is the +** pattern and the second argument is the string. So, the SQL statements: +** +** A LIKE B +** +** is implemented as like(B, A). If there is an escape character E, +** +** A LIKE B ESCAPE E +** +** is mapped to like(B, A, E). +*/ +static void icuLikeFunc( + sqlite3_context *context, + int argc, + sqlite3_value **argv +){ + const unsigned char *zA = sqlite3_value_text(argv[0]); + const unsigned char *zB = sqlite3_value_text(argv[1]); + UChar32 uEsc = 0; + + /* Limit the length of the LIKE or GLOB pattern to avoid problems + ** of deep recursion and N*N behavior in patternCompare(). + */ + if( sqlite3_value_bytes(argv[0])>SQLITE_MAX_LIKE_PATTERN_LENGTH ){ + sqlite3_result_error(context, "LIKE or GLOB pattern too complex", -1); + return; + } + + + if( argc==3 ){ + /* The escape character string must consist of a single UTF-8 character. + ** Otherwise, return an error. + */ + int nE= sqlite3_value_bytes(argv[2]); + const unsigned char *zE = sqlite3_value_text(argv[2]); + int i = 0; + if( zE==0 ) return; + U8_NEXT(zE, i, nE, uEsc); + if( i!=nE){ + sqlite3_result_error(context, + "ESCAPE expression must be a single character", -1); + return; + } + } + + if( zA && zB ){ + sqlite3_result_int(context, icuLikeCompare(zA, zB, uEsc)); + } +} + +/* +** Function to delete compiled regexp objects. Registered as +** a destructor function with sqlite3_set_auxdata(). +*/ +static void icuRegexpDelete(void *p){ + URegularExpression *pExpr = (URegularExpression *)p; + uregex_close(pExpr); +} + +/* +** Implementation of SQLite REGEXP operator. This scalar function takes +** two arguments. The first is a regular expression pattern to compile +** the second is a string to match against that pattern. If either +** argument is an SQL NULL, then NULL Is returned. Otherwise, the result +** is 1 if the string matches the pattern, or 0 otherwise. +** +** SQLite maps the regexp() function to the regexp() operator such +** that the following two are equivalent: +** +** zString REGEXP zPattern +** regexp(zPattern, zString) +** +** Uses the following ICU regexp APIs: +** +** uregex_open() +** uregex_matches() +** uregex_close() +*/ +static void icuRegexpFunc(sqlite3_context *p, int nArg, sqlite3_value **apArg){ + UErrorCode status = U_ZERO_ERROR; + URegularExpression *pExpr; + UBool res; + const UChar *zString = sqlite3_value_text16(apArg[1]); + + (void)nArg; /* Unused parameter */ + + /* If the left hand side of the regexp operator is NULL, + ** then the result is also NULL. + */ + if( !zString ){ + return; + } + + pExpr = sqlite3_get_auxdata(p, 0); + if( !pExpr ){ + const UChar *zPattern = sqlite3_value_text16(apArg[0]); + if( !zPattern ){ + return; + } + pExpr = uregex_open(zPattern, -1, 0, 0, &status); + + if( U_SUCCESS(status) ){ + sqlite3_set_auxdata(p, 0, pExpr, icuRegexpDelete); + }else{ + assert(!pExpr); + icuFunctionError(p, "uregex_open", status); + return; + } + } + + /* Configure the text that the regular expression operates on. */ + uregex_setText(pExpr, zString, -1, &status); + if( !U_SUCCESS(status) ){ + icuFunctionError(p, "uregex_setText", status); + return; + } + + /* Attempt the match */ + res = uregex_matches(pExpr, 0, &status); + if( !U_SUCCESS(status) ){ + icuFunctionError(p, "uregex_matches", status); + return; + } + + /* Set the text that the regular expression operates on to a NULL + ** pointer. This is not really necessary, but it is tidier than + ** leaving the regular expression object configured with an invalid + ** pointer after this function returns. + */ + uregex_setText(pExpr, 0, 0, &status); + + /* Return 1 or 0. */ + sqlite3_result_int(p, res ? 1 : 0); +} + +/* +** Implementations of scalar functions for case mapping - upper() and +** lower(). Function upper() converts its input to upper-case (ABC). +** Function lower() converts to lower-case (abc). +** +** ICU provides two types of case mapping, "general" case mapping and +** "language specific". Refer to ICU documentation for the differences +** between the two. +** +** To utilise "general" case mapping, the upper() or lower() scalar +** functions are invoked with one argument: +** +** upper('ABC') -> 'abc' +** lower('abc') -> 'ABC' +** +** To access ICU "language specific" case mapping, upper() or lower() +** should be invoked with two arguments. The second argument is the name +** of the locale to use. Passing an empty string ("") or SQL NULL value +** as the second argument is the same as invoking the 1 argument version +** of upper() or lower(). +** +** lower('I', 'en_us') -> 'i' +** lower('I', 'tr_tr') -> '\u131' (small dotless i) +** +** http://www.icu-project.org/userguide/posix.html#case_mappings +*/ +static void icuCaseFunc16(sqlite3_context *p, int nArg, sqlite3_value **apArg){ + const UChar *zInput; /* Pointer to input string */ + UChar *zOutput = 0; /* Pointer to output buffer */ + int nInput; /* Size of utf-16 input string in bytes */ + int nOut; /* Size of output buffer in bytes */ + int cnt; + int bToUpper; /* True for toupper(), false for tolower() */ + UErrorCode status; + const char *zLocale = 0; + + assert(nArg==1 || nArg==2); + bToUpper = (sqlite3_user_data(p)!=0); + if( nArg==2 ){ + zLocale = (const char *)sqlite3_value_text(apArg[1]); + } + + zInput = sqlite3_value_text16(apArg[0]); + if( !zInput ){ + return; + } + nOut = nInput = sqlite3_value_bytes16(apArg[0]); + if( nOut==0 ){ + sqlite3_result_text16(p, "", 0, SQLITE_STATIC); + return; + } + + for(cnt=0; cnt<2; cnt++){ + UChar *zNew = sqlite3_realloc(zOutput, nOut); + if( zNew==0 ){ + sqlite3_free(zOutput); + sqlite3_result_error_nomem(p); + return; + } + zOutput = zNew; + status = U_ZERO_ERROR; + if( bToUpper ){ + nOut = 2*u_strToUpper(zOutput,nOut/2,zInput,nInput/2,zLocale,&status); + }else{ + nOut = 2*u_strToLower(zOutput,nOut/2,zInput,nInput/2,zLocale,&status); + } + + if( U_SUCCESS(status) ){ + sqlite3_result_text16(p, zOutput, nOut, xFree); + }else if( status==U_BUFFER_OVERFLOW_ERROR ){ + assert( cnt==0 ); + continue; + }else{ + icuFunctionError(p, bToUpper ? "u_strToUpper" : "u_strToLower", status); + } + return; + } + assert( 0 ); /* Unreachable */ +} + +#endif /* !defined(SQLITE_CORE) || defined(SQLITE_ENABLE_ICU) */ + +/* +** Collation sequence destructor function. The pCtx argument points to +** a UCollator structure previously allocated using ucol_open(). +*/ +static void icuCollationDel(void *pCtx){ + UCollator *p = (UCollator *)pCtx; + ucol_close(p); +} + +/* +** Collation sequence comparison function. The pCtx argument points to +** a UCollator structure previously allocated using ucol_open(). +*/ +static int icuCollationColl( + void *pCtx, + int nLeft, + const void *zLeft, + int nRight, + const void *zRight +){ + UCollationResult res; + UCollator *p = (UCollator *)pCtx; + res = ucol_strcoll(p, (UChar *)zLeft, nLeft/2, (UChar *)zRight, nRight/2); + switch( res ){ + case UCOL_LESS: return -1; + case UCOL_GREATER: return +1; + case UCOL_EQUAL: return 0; + } + assert(!"Unexpected return value from ucol_strcoll()"); + return 0; +} + +/* +** Implementation of the scalar function icu_load_collation(). +** +** This scalar function is used to add ICU collation based collation +** types to an SQLite database connection. It is intended to be called +** as follows: +** +** SELECT icu_load_collation(, ); +** +** Where is a string containing an ICU locale identifier (i.e. +** "en_AU", "tr_TR" etc.) and is the name of the +** collation sequence to create. +*/ +static void icuLoadCollation( + sqlite3_context *p, + int nArg, + sqlite3_value **apArg +){ + sqlite3 *db = (sqlite3 *)sqlite3_user_data(p); + UErrorCode status = U_ZERO_ERROR; + const char *zLocale; /* Locale identifier - (eg. "jp_JP") */ + const char *zName; /* SQL Collation sequence name (eg. "japanese") */ + UCollator *pUCollator; /* ICU library collation object */ + int rc; /* Return code from sqlite3_create_collation_x() */ + + assert(nArg==2); + (void)nArg; /* Unused parameter */ + zLocale = (const char *)sqlite3_value_text(apArg[0]); + zName = (const char *)sqlite3_value_text(apArg[1]); + + if( !zLocale || !zName ){ + return; + } + + pUCollator = ucol_open(zLocale, &status); + if( !U_SUCCESS(status) ){ + icuFunctionError(p, "ucol_open", status); + return; + } + assert(p); + + rc = sqlite3_create_collation_v2(db, zName, SQLITE_UTF16, (void *)pUCollator, + icuCollationColl, icuCollationDel + ); + if( rc!=SQLITE_OK ){ + ucol_close(pUCollator); + sqlite3_result_error(p, "Error registering collation function", -1); + } +} + +/* +** Register the ICU extension functions with database db. +*/ +SQLITE_PRIVATE int sqlite3IcuInit(sqlite3 *db){ + static const struct IcuScalar { + const char *zName; /* Function name */ + unsigned char nArg; /* Number of arguments */ + unsigned short enc; /* Optimal text encoding */ + unsigned char iContext; /* sqlite3_user_data() context */ + void (*xFunc)(sqlite3_context*,int,sqlite3_value**); + } scalars[] = { + {"icu_load_collation", 2, SQLITE_UTF8, 1, icuLoadCollation}, +#if !defined(SQLITE_CORE) || defined(SQLITE_ENABLE_ICU) + {"regexp", 2, SQLITE_ANY|SQLITE_DETERMINISTIC, 0, icuRegexpFunc}, + {"lower", 1, SQLITE_UTF16|SQLITE_DETERMINISTIC, 0, icuCaseFunc16}, + {"lower", 2, SQLITE_UTF16|SQLITE_DETERMINISTIC, 0, icuCaseFunc16}, + {"upper", 1, SQLITE_UTF16|SQLITE_DETERMINISTIC, 1, icuCaseFunc16}, + {"upper", 2, SQLITE_UTF16|SQLITE_DETERMINISTIC, 1, icuCaseFunc16}, + {"lower", 1, SQLITE_UTF8|SQLITE_DETERMINISTIC, 0, icuCaseFunc16}, + {"lower", 2, SQLITE_UTF8|SQLITE_DETERMINISTIC, 0, icuCaseFunc16}, + {"upper", 1, SQLITE_UTF8|SQLITE_DETERMINISTIC, 1, icuCaseFunc16}, + {"upper", 2, SQLITE_UTF8|SQLITE_DETERMINISTIC, 1, icuCaseFunc16}, + {"like", 2, SQLITE_UTF8|SQLITE_DETERMINISTIC, 0, icuLikeFunc}, + {"like", 3, SQLITE_UTF8|SQLITE_DETERMINISTIC, 0, icuLikeFunc}, +#endif /* !defined(SQLITE_CORE) || defined(SQLITE_ENABLE_ICU) */ + }; + int rc = SQLITE_OK; + int i; + + for(i=0; rc==SQLITE_OK && i<(int)(sizeof(scalars)/sizeof(scalars[0])); i++){ + const struct IcuScalar *p = &scalars[i]; + rc = sqlite3_create_function( + db, p->zName, p->nArg, p->enc, + p->iContext ? (void*)db : (void*)0, + p->xFunc, 0, 0 + ); + } + + return rc; +} + +#if !SQLITE_CORE +#ifdef _WIN32 +__declspec(dllexport) +#endif +SQLITE_API int sqlite3_icu_init( + sqlite3 *db, + char **pzErrMsg, + const sqlite3_api_routines *pApi +){ + SQLITE_EXTENSION_INIT2(pApi) + return sqlite3IcuInit(db); +} +#endif + +#endif + +/************** End of icu.c *************************************************/ +/************** Begin file fts3_icu.c ****************************************/ +/* +** 2007 June 22 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** This file implements a tokenizer for fts3 based on the ICU library. +*/ +/* #include "fts3Int.h" */ +#if !defined(SQLITE_CORE) || defined(SQLITE_ENABLE_FTS3) +#ifdef SQLITE_ENABLE_ICU + +/* #include */ +/* #include */ +/* #include "fts3_tokenizer.h" */ + +#include +/* #include */ +/* #include */ +#include + +typedef struct IcuTokenizer IcuTokenizer; +typedef struct IcuCursor IcuCursor; + +struct IcuTokenizer { + sqlite3_tokenizer base; + char *zLocale; +}; + +struct IcuCursor { + sqlite3_tokenizer_cursor base; + + UBreakIterator *pIter; /* ICU break-iterator object */ + int nChar; /* Number of UChar elements in pInput */ + UChar *aChar; /* Copy of input using utf-16 encoding */ + int *aOffset; /* Offsets of each character in utf-8 input */ + + int nBuffer; + char *zBuffer; + + int iToken; +}; + +/* +** Create a new tokenizer instance. +*/ +static int icuCreate( + int argc, /* Number of entries in argv[] */ + const char * const *argv, /* Tokenizer creation arguments */ + sqlite3_tokenizer **ppTokenizer /* OUT: Created tokenizer */ +){ + IcuTokenizer *p; + int n = 0; + + if( argc>0 ){ + n = strlen(argv[0])+1; + } + p = (IcuTokenizer *)sqlite3_malloc64(sizeof(IcuTokenizer)+n); + if( !p ){ + return SQLITE_NOMEM; + } + memset(p, 0, sizeof(IcuTokenizer)); + + if( n ){ + p->zLocale = (char *)&p[1]; + memcpy(p->zLocale, argv[0], n); + } + + *ppTokenizer = (sqlite3_tokenizer *)p; + + return SQLITE_OK; +} + +/* +** Destroy a tokenizer +*/ +static int icuDestroy(sqlite3_tokenizer *pTokenizer){ + IcuTokenizer *p = (IcuTokenizer *)pTokenizer; + sqlite3_free(p); + return SQLITE_OK; +} + +/* +** Prepare to begin tokenizing a particular string. The input +** string to be tokenized is pInput[0..nBytes-1]. A cursor +** used to incrementally tokenize this string is returned in +** *ppCursor. +*/ +static int icuOpen( + sqlite3_tokenizer *pTokenizer, /* The tokenizer */ + const char *zInput, /* Input string */ + int nInput, /* Length of zInput in bytes */ + sqlite3_tokenizer_cursor **ppCursor /* OUT: Tokenization cursor */ +){ + IcuTokenizer *p = (IcuTokenizer *)pTokenizer; + IcuCursor *pCsr; + + const int32_t opt = U_FOLD_CASE_DEFAULT; + UErrorCode status = U_ZERO_ERROR; + int nChar; + + UChar32 c; + int iInput = 0; + int iOut = 0; + + *ppCursor = 0; + + if( zInput==0 ){ + nInput = 0; + zInput = ""; + }else if( nInput<0 ){ + nInput = strlen(zInput); + } + nChar = nInput+1; + pCsr = (IcuCursor *)sqlite3_malloc64( + sizeof(IcuCursor) + /* IcuCursor */ + ((nChar+3)&~3) * sizeof(UChar) + /* IcuCursor.aChar[] */ + (nChar+1) * sizeof(int) /* IcuCursor.aOffset[] */ + ); + if( !pCsr ){ + return SQLITE_NOMEM; + } + memset(pCsr, 0, sizeof(IcuCursor)); + pCsr->aChar = (UChar *)&pCsr[1]; + pCsr->aOffset = (int *)&pCsr->aChar[(nChar+3)&~3]; + + pCsr->aOffset[iOut] = iInput; + U8_NEXT(zInput, iInput, nInput, c); + while( c>0 ){ + int isError = 0; + c = u_foldCase(c, opt); + U16_APPEND(pCsr->aChar, iOut, nChar, c, isError); + if( isError ){ + sqlite3_free(pCsr); + return SQLITE_ERROR; + } + pCsr->aOffset[iOut] = iInput; + + if( iInputpIter = ubrk_open(UBRK_WORD, p->zLocale, pCsr->aChar, iOut, &status); + if( !U_SUCCESS(status) ){ + sqlite3_free(pCsr); + return SQLITE_ERROR; + } + pCsr->nChar = iOut; + + ubrk_first(pCsr->pIter); + *ppCursor = (sqlite3_tokenizer_cursor *)pCsr; + return SQLITE_OK; +} + +/* +** Close a tokenization cursor previously opened by a call to icuOpen(). +*/ +static int icuClose(sqlite3_tokenizer_cursor *pCursor){ + IcuCursor *pCsr = (IcuCursor *)pCursor; + ubrk_close(pCsr->pIter); + sqlite3_free(pCsr->zBuffer); + sqlite3_free(pCsr); + return SQLITE_OK; +} + +/* +** Extract the next token from a tokenization cursor. +*/ +static int icuNext( + sqlite3_tokenizer_cursor *pCursor, /* Cursor returned by simpleOpen */ + const char **ppToken, /* OUT: *ppToken is the token text */ + int *pnBytes, /* OUT: Number of bytes in token */ + int *piStartOffset, /* OUT: Starting offset of token */ + int *piEndOffset, /* OUT: Ending offset of token */ + int *piPosition /* OUT: Position integer of token */ +){ + IcuCursor *pCsr = (IcuCursor *)pCursor; + + int iStart = 0; + int iEnd = 0; + int nByte = 0; + + while( iStart==iEnd ){ + UChar32 c; + + iStart = ubrk_current(pCsr->pIter); + iEnd = ubrk_next(pCsr->pIter); + if( iEnd==UBRK_DONE ){ + return SQLITE_DONE; + } + + while( iStartaChar, iWhite, pCsr->nChar, c); + if( u_isspace(c) ){ + iStart = iWhite; + }else{ + break; + } + } + assert(iStart<=iEnd); + } + + do { + UErrorCode status = U_ZERO_ERROR; + if( nByte ){ + char *zNew = sqlite3_realloc(pCsr->zBuffer, nByte); + if( !zNew ){ + return SQLITE_NOMEM; + } + pCsr->zBuffer = zNew; + pCsr->nBuffer = nByte; + } + + u_strToUTF8( + pCsr->zBuffer, pCsr->nBuffer, &nByte, /* Output vars */ + &pCsr->aChar[iStart], iEnd-iStart, /* Input vars */ + &status /* Output success/failure */ + ); + } while( nByte>pCsr->nBuffer ); + + *ppToken = pCsr->zBuffer; + *pnBytes = nByte; + *piStartOffset = pCsr->aOffset[iStart]; + *piEndOffset = pCsr->aOffset[iEnd]; + *piPosition = pCsr->iToken++; + + return SQLITE_OK; +} + +/* +** The set of routines that implement the simple tokenizer +*/ +static const sqlite3_tokenizer_module icuTokenizerModule = { + 0, /* iVersion */ + icuCreate, /* xCreate */ + icuDestroy, /* xCreate */ + icuOpen, /* xOpen */ + icuClose, /* xClose */ + icuNext, /* xNext */ + 0, /* xLanguageid */ +}; + +/* +** Set *ppModule to point at the implementation of the ICU tokenizer. +*/ +SQLITE_PRIVATE void sqlite3Fts3IcuTokenizerModule( + sqlite3_tokenizer_module const**ppModule +){ + *ppModule = &icuTokenizerModule; +} + +#endif /* defined(SQLITE_ENABLE_ICU) */ +#endif /* !defined(SQLITE_CORE) || defined(SQLITE_ENABLE_FTS3) */ + +/************** End of fts3_icu.c ********************************************/ +/************** Begin file sqlite3rbu.c **************************************/ +/* +** 2014 August 30 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** +** +** OVERVIEW +** +** The RBU extension requires that the RBU update be packaged as an +** SQLite database. The tables it expects to find are described in +** sqlite3rbu.h. Essentially, for each table xyz in the target database +** that the user wishes to write to, a corresponding data_xyz table is +** created in the RBU database and populated with one row for each row to +** update, insert or delete from the target table. +** +** The update proceeds in three stages: +** +** 1) The database is updated. The modified database pages are written +** to a *-oal file. A *-oal file is just like a *-wal file, except +** that it is named "-oal" instead of "-wal". +** Because regular SQLite clients do not look for file named +** "-oal", they go on using the original database in +** rollback mode while the *-oal file is being generated. +** +** During this stage RBU does not update the database by writing +** directly to the target tables. Instead it creates "imposter" +** tables using the SQLITE_TESTCTRL_IMPOSTER interface that it uses +** to update each b-tree individually. All updates required by each +** b-tree are completed before moving on to the next, and all +** updates are done in sorted key order. +** +** 2) The "-oal" file is moved to the equivalent "-wal" +** location using a call to rename(2). Before doing this the RBU +** module takes an EXCLUSIVE lock on the database file, ensuring +** that there are no other active readers. +** +** Once the EXCLUSIVE lock is released, any other database readers +** detect the new *-wal file and read the database in wal mode. At +** this point they see the new version of the database - including +** the updates made as part of the RBU update. +** +** 3) The new *-wal file is checkpointed. This proceeds in the same way +** as a regular database checkpoint, except that a single frame is +** checkpointed each time sqlite3rbu_step() is called. If the RBU +** handle is closed before the entire *-wal file is checkpointed, +** the checkpoint progress is saved in the RBU database and the +** checkpoint can be resumed by another RBU client at some point in +** the future. +** +** POTENTIAL PROBLEMS +** +** The rename() call might not be portable. And RBU is not currently +** syncing the directory after renaming the file. +** +** When state is saved, any commit to the *-oal file and the commit to +** the RBU update database are not atomic. So if the power fails at the +** wrong moment they might get out of sync. As the main database will be +** committed before the RBU update database this will likely either just +** pass unnoticed, or result in SQLITE_CONSTRAINT errors (due to UNIQUE +** constraint violations). +** +** If some client does modify the target database mid RBU update, or some +** other error occurs, the RBU extension will keep throwing errors. It's +** not really clear how to get out of this state. The system could just +** by delete the RBU update database and *-oal file and have the device +** download the update again and start over. +** +** At present, for an UPDATE, both the new.* and old.* records are +** collected in the rbu_xyz table. And for both UPDATEs and DELETEs all +** fields are collected. This means we're probably writing a lot more +** data to disk when saving the state of an ongoing update to the RBU +** update database than is strictly necessary. +** +*/ + +/* #include */ +/* #include */ +/* #include */ + +/* #include "sqlite3.h" */ + +#if !defined(SQLITE_CORE) || defined(SQLITE_ENABLE_RBU) +/************** Include sqlite3rbu.h in the middle of sqlite3rbu.c ***********/ +/************** Begin file sqlite3rbu.h **************************************/ +/* +** 2014 August 30 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** +** This file contains the public interface for the RBU extension. +*/ + +/* +** SUMMARY +** +** Writing a transaction containing a large number of operations on +** b-tree indexes that are collectively larger than the available cache +** memory can be very inefficient. +** +** The problem is that in order to update a b-tree, the leaf page (at least) +** containing the entry being inserted or deleted must be modified. If the +** working set of leaves is larger than the available cache memory, then a +** single leaf that is modified more than once as part of the transaction +** may be loaded from or written to the persistent media multiple times. +** Additionally, because the index updates are likely to be applied in +** random order, access to pages within the database is also likely to be in +** random order, which is itself quite inefficient. +** +** One way to improve the situation is to sort the operations on each index +** by index key before applying them to the b-tree. This leads to an IO +** pattern that resembles a single linear scan through the index b-tree, +** and all but guarantees each modified leaf page is loaded and stored +** exactly once. SQLite uses this trick to improve the performance of +** CREATE INDEX commands. This extension allows it to be used to improve +** the performance of large transactions on existing databases. +** +** Additionally, this extension allows the work involved in writing the +** large transaction to be broken down into sub-transactions performed +** sequentially by separate processes. This is useful if the system cannot +** guarantee that a single update process will run for long enough to apply +** the entire update, for example because the update is being applied on a +** mobile device that is frequently rebooted. Even after the writer process +** has committed one or more sub-transactions, other database clients continue +** to read from the original database snapshot. In other words, partially +** applied transactions are not visible to other clients. +** +** "RBU" stands for "Resumable Bulk Update". As in a large database update +** transmitted via a wireless network to a mobile device. A transaction +** applied using this extension is hence refered to as an "RBU update". +** +** +** LIMITATIONS +** +** An "RBU update" transaction is subject to the following limitations: +** +** * The transaction must consist of INSERT, UPDATE and DELETE operations +** only. +** +** * INSERT statements may not use any default values. +** +** * UPDATE and DELETE statements must identify their target rows by +** non-NULL PRIMARY KEY values. Rows with NULL values stored in PRIMARY +** KEY fields may not be updated or deleted. If the table being written +** has no PRIMARY KEY, affected rows must be identified by rowid. +** +** * UPDATE statements may not modify PRIMARY KEY columns. +** +** * No triggers will be fired. +** +** * No foreign key violations are detected or reported. +** +** * CHECK constraints are not enforced. +** +** * No constraint handling mode except for "OR ROLLBACK" is supported. +** +** +** PREPARATION +** +** An "RBU update" is stored as a separate SQLite database. A database +** containing an RBU update is an "RBU database". For each table in the +** target database to be updated, the RBU database should contain a table +** named "data_" containing the same set of columns as the +** target table, and one more - "rbu_control". The data_% table should +** have no PRIMARY KEY or UNIQUE constraints, but each column should have +** the same type as the corresponding column in the target database. +** The "rbu_control" column should have no type at all. For example, if +** the target database contains: +** +** CREATE TABLE t1(a INTEGER PRIMARY KEY, b TEXT, c UNIQUE); +** +** Then the RBU database should contain: +** +** CREATE TABLE data_t1(a INTEGER, b TEXT, c, rbu_control); +** +** The order of the columns in the data_% table does not matter. +** +** Instead of a regular table, the RBU database may also contain virtual +** tables or view named using the data_ naming scheme. +** +** Instead of the plain data_ naming scheme, RBU database tables +** may also be named data_, where is any sequence +** of zero or more numeric characters (0-9). This can be significant because +** tables within the RBU database are always processed in order sorted by +** name. By judicious selection of the portion of the names +** of the RBU tables the user can therefore control the order in which they +** are processed. This can be useful, for example, to ensure that "external +** content" FTS4 tables are updated before their underlying content tables. +** +** If the target database table is a virtual table or a table that has no +** PRIMARY KEY declaration, the data_% table must also contain a column +** named "rbu_rowid". This column is mapped to the tables implicit primary +** key column - "rowid". Virtual tables for which the "rowid" column does +** not function like a primary key value cannot be updated using RBU. For +** example, if the target db contains either of the following: +** +** CREATE VIRTUAL TABLE x1 USING fts3(a, b); +** CREATE TABLE x1(a, b) +** +** then the RBU database should contain: +** +** CREATE TABLE data_x1(a, b, rbu_rowid, rbu_control); +** +** All non-hidden columns (i.e. all columns matched by "SELECT *") of the +** target table must be present in the input table. For virtual tables, +** hidden columns are optional - they are updated by RBU if present in +** the input table, or not otherwise. For example, to write to an fts4 +** table with a hidden languageid column such as: +** +** CREATE VIRTUAL TABLE ft1 USING fts4(a, b, languageid='langid'); +** +** Either of the following input table schemas may be used: +** +** CREATE TABLE data_ft1(a, b, langid, rbu_rowid, rbu_control); +** CREATE TABLE data_ft1(a, b, rbu_rowid, rbu_control); +** +** For each row to INSERT into the target database as part of the RBU +** update, the corresponding data_% table should contain a single record +** with the "rbu_control" column set to contain integer value 0. The +** other columns should be set to the values that make up the new record +** to insert. +** +** If the target database table has an INTEGER PRIMARY KEY, it is not +** possible to insert a NULL value into the IPK column. Attempting to +** do so results in an SQLITE_MISMATCH error. +** +** For each row to DELETE from the target database as part of the RBU +** update, the corresponding data_% table should contain a single record +** with the "rbu_control" column set to contain integer value 1. The +** real primary key values of the row to delete should be stored in the +** corresponding columns of the data_% table. The values stored in the +** other columns are not used. +** +** For each row to UPDATE from the target database as part of the RBU +** update, the corresponding data_% table should contain a single record +** with the "rbu_control" column set to contain a value of type text. +** The real primary key values identifying the row to update should be +** stored in the corresponding columns of the data_% table row, as should +** the new values of all columns being update. The text value in the +** "rbu_control" column must contain the same number of characters as +** there are columns in the target database table, and must consist entirely +** of 'x' and '.' characters (or in some special cases 'd' - see below). For +** each column that is being updated, the corresponding character is set to +** 'x'. For those that remain as they are, the corresponding character of the +** rbu_control value should be set to '.'. For example, given the tables +** above, the update statement: +** +** UPDATE t1 SET c = 'usa' WHERE a = 4; +** +** is represented by the data_t1 row created by: +** +** INSERT INTO data_t1(a, b, c, rbu_control) VALUES(4, NULL, 'usa', '..x'); +** +** Instead of an 'x' character, characters of the rbu_control value specified +** for UPDATEs may also be set to 'd'. In this case, instead of updating the +** target table with the value stored in the corresponding data_% column, the +** user-defined SQL function "rbu_delta()" is invoked and the result stored in +** the target table column. rbu_delta() is invoked with two arguments - the +** original value currently stored in the target table column and the +** value specified in the data_xxx table. +** +** For example, this row: +** +** INSERT INTO data_t1(a, b, c, rbu_control) VALUES(4, NULL, 'usa', '..d'); +** +** is similar to an UPDATE statement such as: +** +** UPDATE t1 SET c = rbu_delta(c, 'usa') WHERE a = 4; +** +** Finally, if an 'f' character appears in place of a 'd' or 's' in an +** ota_control string, the contents of the data_xxx table column is assumed +** to be a "fossil delta" - a patch to be applied to a blob value in the +** format used by the fossil source-code management system. In this case +** the existing value within the target database table must be of type BLOB. +** It is replaced by the result of applying the specified fossil delta to +** itself. +** +** If the target database table is a virtual table or a table with no PRIMARY +** KEY, the rbu_control value should not include a character corresponding +** to the rbu_rowid value. For example, this: +** +** INSERT INTO data_ft1(a, b, rbu_rowid, rbu_control) +** VALUES(NULL, 'usa', 12, '.x'); +** +** causes a result similar to: +** +** UPDATE ft1 SET b = 'usa' WHERE rowid = 12; +** +** The data_xxx tables themselves should have no PRIMARY KEY declarations. +** However, RBU is more efficient if reading the rows in from each data_xxx +** table in "rowid" order is roughly the same as reading them sorted by +** the PRIMARY KEY of the corresponding target database table. In other +** words, rows should be sorted using the destination table PRIMARY KEY +** fields before they are inserted into the data_xxx tables. +** +** USAGE +** +** The API declared below allows an application to apply an RBU update +** stored on disk to an existing target database. Essentially, the +** application: +** +** 1) Opens an RBU handle using the sqlite3rbu_open() function. +** +** 2) Registers any required virtual table modules with the database +** handle returned by sqlite3rbu_db(). Also, if required, register +** the rbu_delta() implementation. +** +** 3) Calls the sqlite3rbu_step() function one or more times on +** the new handle. Each call to sqlite3rbu_step() performs a single +** b-tree operation, so thousands of calls may be required to apply +** a complete update. +** +** 4) Calls sqlite3rbu_close() to close the RBU update handle. If +** sqlite3rbu_step() has been called enough times to completely +** apply the update to the target database, then the RBU database +** is marked as fully applied. Otherwise, the state of the RBU +** update application is saved in the RBU database for later +** resumption. +** +** See comments below for more detail on APIs. +** +** If an update is only partially applied to the target database by the +** time sqlite3rbu_close() is called, various state information is saved +** within the RBU database. This allows subsequent processes to automatically +** resume the RBU update from where it left off. +** +** To remove all RBU extension state information, returning an RBU database +** to its original contents, it is sufficient to drop all tables that begin +** with the prefix "rbu_" +** +** DATABASE LOCKING +** +** An RBU update may not be applied to a database in WAL mode. Attempting +** to do so is an error (SQLITE_ERROR). +** +** While an RBU handle is open, a SHARED lock may be held on the target +** database file. This means it is possible for other clients to read the +** database, but not to write it. +** +** If an RBU update is started and then suspended before it is completed, +** then an external client writes to the database, then attempting to resume +** the suspended RBU update is also an error (SQLITE_BUSY). +*/ + +#ifndef _SQLITE3RBU_H +#define _SQLITE3RBU_H + +/* #include "sqlite3.h" ** Required for error code definitions ** */ + +#if 0 +extern "C" { +#endif + +typedef struct sqlite3rbu sqlite3rbu; + +/* +** Open an RBU handle. +** +** Argument zTarget is the path to the target database. Argument zRbu is +** the path to the RBU database. Each call to this function must be matched +** by a call to sqlite3rbu_close(). When opening the databases, RBU passes +** the SQLITE_CONFIG_URI flag to sqlite3_open_v2(). So if either zTarget +** or zRbu begin with "file:", it will be interpreted as an SQLite +** database URI, not a regular file name. +** +** If the zState argument is passed a NULL value, the RBU extension stores +** the current state of the update (how many rows have been updated, which +** indexes are yet to be updated etc.) within the RBU database itself. This +** can be convenient, as it means that the RBU application does not need to +** organize removing a separate state file after the update is concluded. +** Or, if zState is non-NULL, it must be a path to a database file in which +** the RBU extension can store the state of the update. +** +** When resuming an RBU update, the zState argument must be passed the same +** value as when the RBU update was started. +** +** Once the RBU update is finished, the RBU extension does not +** automatically remove any zState database file, even if it created it. +** +** By default, RBU uses the default VFS to access the files on disk. To +** use a VFS other than the default, an SQLite "file:" URI containing a +** "vfs=..." option may be passed as the zTarget option. +** +** IMPORTANT NOTE FOR ZIPVFS USERS: The RBU extension works with all of +** SQLite's built-in VFSs, including the multiplexor VFS. However it does +** not work out of the box with zipvfs. Refer to the comment describing +** the zipvfs_create_vfs() API below for details on using RBU with zipvfs. +*/ +SQLITE_API sqlite3rbu *sqlite3rbu_open( + const char *zTarget, + const char *zRbu, + const char *zState +); + +/* +** Open an RBU handle to perform an RBU vacuum on database file zTarget. +** An RBU vacuum is similar to SQLite's built-in VACUUM command, except +** that it can be suspended and resumed like an RBU update. +** +** The second argument to this function identifies a database in which +** to store the state of the RBU vacuum operation if it is suspended. The +** first time sqlite3rbu_vacuum() is called, to start an RBU vacuum +** operation, the state database should either not exist or be empty +** (contain no tables). If an RBU vacuum is suspended by calling +** sqlite3rbu_close() on the RBU handle before sqlite3rbu_step() has +** returned SQLITE_DONE, the vacuum state is stored in the state database. +** The vacuum can be resumed by calling this function to open a new RBU +** handle specifying the same target and state databases. +** +** If the second argument passed to this function is NULL, then the +** name of the state database is "-vacuum", where +** is the name of the target database file. In this case, on UNIX, if the +** state database is not already present in the file-system, it is created +** with the same permissions as the target db is made. +** +** With an RBU vacuum, it is an SQLITE_MISUSE error if the name of the +** state database ends with "-vactmp". This name is reserved for internal +** use. +** +** This function does not delete the state database after an RBU vacuum +** is completed, even if it created it. However, if the call to +** sqlite3rbu_close() returns any value other than SQLITE_OK, the contents +** of the state tables within the state database are zeroed. This way, +** the next call to sqlite3rbu_vacuum() opens a handle that starts a +** new RBU vacuum operation. +** +** As with sqlite3rbu_open(), Zipvfs users should rever to the comment +** describing the sqlite3rbu_create_vfs() API function below for +** a description of the complications associated with using RBU with +** zipvfs databases. +*/ +SQLITE_API sqlite3rbu *sqlite3rbu_vacuum( + const char *zTarget, + const char *zState +); + +/* +** Configure a limit for the amount of temp space that may be used by +** the RBU handle passed as the first argument. The new limit is specified +** in bytes by the second parameter. If it is positive, the limit is updated. +** If the second parameter to this function is passed zero, then the limit +** is removed entirely. If the second parameter is negative, the limit is +** not modified (this is useful for querying the current limit). +** +** In all cases the returned value is the current limit in bytes (zero +** indicates unlimited). +** +** If the temp space limit is exceeded during operation, an SQLITE_FULL +** error is returned. +*/ +SQLITE_API sqlite3_int64 sqlite3rbu_temp_size_limit(sqlite3rbu*, sqlite3_int64); + +/* +** Return the current amount of temp file space, in bytes, currently used by +** the RBU handle passed as the only argument. +*/ +SQLITE_API sqlite3_int64 sqlite3rbu_temp_size(sqlite3rbu*); + +/* +** Internally, each RBU connection uses a separate SQLite database +** connection to access the target and rbu update databases. This +** API allows the application direct access to these database handles. +** +** The first argument passed to this function must be a valid, open, RBU +** handle. The second argument should be passed zero to access the target +** database handle, or non-zero to access the rbu update database handle. +** Accessing the underlying database handles may be useful in the +** following scenarios: +** +** * If any target tables are virtual tables, it may be necessary to +** call sqlite3_create_module() on the target database handle to +** register the required virtual table implementations. +** +** * If the data_xxx tables in the RBU source database are virtual +** tables, the application may need to call sqlite3_create_module() on +** the rbu update db handle to any required virtual table +** implementations. +** +** * If the application uses the "rbu_delta()" feature described above, +** it must use sqlite3_create_function() or similar to register the +** rbu_delta() implementation with the target database handle. +** +** If an error has occurred, either while opening or stepping the RBU object, +** this function may return NULL. The error code and message may be collected +** when sqlite3rbu_close() is called. +** +** Database handles returned by this function remain valid until the next +** call to any sqlite3rbu_xxx() function other than sqlite3rbu_db(). +*/ +SQLITE_API sqlite3 *sqlite3rbu_db(sqlite3rbu*, int bRbu); + +/* +** Do some work towards applying the RBU update to the target db. +** +** Return SQLITE_DONE if the update has been completely applied, or +** SQLITE_OK if no error occurs but there remains work to do to apply +** the RBU update. If an error does occur, some other error code is +** returned. +** +** Once a call to sqlite3rbu_step() has returned a value other than +** SQLITE_OK, all subsequent calls on the same RBU handle are no-ops +** that immediately return the same value. +*/ +SQLITE_API int sqlite3rbu_step(sqlite3rbu *pRbu); + +/* +** Force RBU to save its state to disk. +** +** If a power failure or application crash occurs during an update, following +** system recovery RBU may resume the update from the point at which the state +** was last saved. In other words, from the most recent successful call to +** sqlite3rbu_close() or this function. +** +** SQLITE_OK is returned if successful, or an SQLite error code otherwise. +*/ +SQLITE_API int sqlite3rbu_savestate(sqlite3rbu *pRbu); + +/* +** Close an RBU handle. +** +** If the RBU update has been completely applied, mark the RBU database +** as fully applied. Otherwise, assuming no error has occurred, save the +** current state of the RBU update appliation to the RBU database. +** +** If an error has already occurred as part of an sqlite3rbu_step() +** or sqlite3rbu_open() call, or if one occurs within this function, an +** SQLite error code is returned. Additionally, if pzErrmsg is not NULL, +** *pzErrmsg may be set to point to a buffer containing a utf-8 formatted +** English language error message. It is the responsibility of the caller to +** eventually free any such buffer using sqlite3_free(). +** +** Otherwise, if no error occurs, this function returns SQLITE_OK if the +** update has been partially applied, or SQLITE_DONE if it has been +** completely applied. +*/ +SQLITE_API int sqlite3rbu_close(sqlite3rbu *pRbu, char **pzErrmsg); + +/* +** Return the total number of key-value operations (inserts, deletes or +** updates) that have been performed on the target database since the +** current RBU update was started. +*/ +SQLITE_API sqlite3_int64 sqlite3rbu_progress(sqlite3rbu *pRbu); + +/* +** Obtain permyriadage (permyriadage is to 10000 as percentage is to 100) +** progress indications for the two stages of an RBU update. This API may +** be useful for driving GUI progress indicators and similar. +** +** An RBU update is divided into two stages: +** +** * Stage 1, in which changes are accumulated in an oal/wal file, and +** * Stage 2, in which the contents of the wal file are copied into the +** main database. +** +** The update is visible to non-RBU clients during stage 2. During stage 1 +** non-RBU reader clients may see the original database. +** +** If this API is called during stage 2 of the update, output variable +** (*pnOne) is set to 10000 to indicate that stage 1 has finished and (*pnTwo) +** to a value between 0 and 10000 to indicate the permyriadage progress of +** stage 2. A value of 5000 indicates that stage 2 is half finished, +** 9000 indicates that it is 90% finished, and so on. +** +** If this API is called during stage 1 of the update, output variable +** (*pnTwo) is set to 0 to indicate that stage 2 has not yet started. The +** value to which (*pnOne) is set depends on whether or not the RBU +** database contains an "rbu_count" table. The rbu_count table, if it +** exists, must contain the same columns as the following: +** +** CREATE TABLE rbu_count(tbl TEXT PRIMARY KEY, cnt INTEGER) WITHOUT ROWID; +** +** There must be one row in the table for each source (data_xxx) table within +** the RBU database. The 'tbl' column should contain the name of the source +** table. The 'cnt' column should contain the number of rows within the +** source table. +** +** If the rbu_count table is present and populated correctly and this +** API is called during stage 1, the *pnOne output variable is set to the +** permyriadage progress of the same stage. If the rbu_count table does +** not exist, then (*pnOne) is set to -1 during stage 1. If the rbu_count +** table exists but is not correctly populated, the value of the *pnOne +** output variable during stage 1 is undefined. +*/ +SQLITE_API void sqlite3rbu_bp_progress(sqlite3rbu *pRbu, int *pnOne, int*pnTwo); + +/* +** Obtain an indication as to the current stage of an RBU update or vacuum. +** This function always returns one of the SQLITE_RBU_STATE_XXX constants +** defined in this file. Return values should be interpreted as follows: +** +** SQLITE_RBU_STATE_OAL: +** RBU is currently building a *-oal file. The next call to sqlite3rbu_step() +** may either add further data to the *-oal file, or compute data that will +** be added by a subsequent call. +** +** SQLITE_RBU_STATE_MOVE: +** RBU has finished building the *-oal file. The next call to sqlite3rbu_step() +** will move the *-oal file to the equivalent *-wal path. If the current +** operation is an RBU update, then the updated version of the database +** file will become visible to ordinary SQLite clients following the next +** call to sqlite3rbu_step(). +** +** SQLITE_RBU_STATE_CHECKPOINT: +** RBU is currently performing an incremental checkpoint. The next call to +** sqlite3rbu_step() will copy a page of data from the *-wal file into +** the target database file. +** +** SQLITE_RBU_STATE_DONE: +** The RBU operation has finished. Any subsequent calls to sqlite3rbu_step() +** will immediately return SQLITE_DONE. +** +** SQLITE_RBU_STATE_ERROR: +** An error has occurred. Any subsequent calls to sqlite3rbu_step() will +** immediately return the SQLite error code associated with the error. +*/ +#define SQLITE_RBU_STATE_OAL 1 +#define SQLITE_RBU_STATE_MOVE 2 +#define SQLITE_RBU_STATE_CHECKPOINT 3 +#define SQLITE_RBU_STATE_DONE 4 +#define SQLITE_RBU_STATE_ERROR 5 + +SQLITE_API int sqlite3rbu_state(sqlite3rbu *pRbu); + +/* +** Create an RBU VFS named zName that accesses the underlying file-system +** via existing VFS zParent. Or, if the zParent parameter is passed NULL, +** then the new RBU VFS uses the default system VFS to access the file-system. +** The new object is registered as a non-default VFS with SQLite before +** returning. +** +** Part of the RBU implementation uses a custom VFS object. Usually, this +** object is created and deleted automatically by RBU. +** +** The exception is for applications that also use zipvfs. In this case, +** the custom VFS must be explicitly created by the user before the RBU +** handle is opened. The RBU VFS should be installed so that the zipvfs +** VFS uses the RBU VFS, which in turn uses any other VFS layers in use +** (for example multiplexor) to access the file-system. For example, +** to assemble an RBU enabled VFS stack that uses both zipvfs and +** multiplexor (error checking omitted): +** +** // Create a VFS named "multiplex" (not the default). +** sqlite3_multiplex_initialize(0, 0); +** +** // Create an rbu VFS named "rbu" that uses multiplexor. If the +** // second argument were replaced with NULL, the "rbu" VFS would +** // access the file-system via the system default VFS, bypassing the +** // multiplexor. +** sqlite3rbu_create_vfs("rbu", "multiplex"); +** +** // Create a zipvfs VFS named "zipvfs" that uses rbu. +** zipvfs_create_vfs_v3("zipvfs", "rbu", 0, xCompressorAlgorithmDetector); +** +** // Make zipvfs the default VFS. +** sqlite3_vfs_register(sqlite3_vfs_find("zipvfs"), 1); +** +** Because the default VFS created above includes a RBU functionality, it +** may be used by RBU clients. Attempting to use RBU with a zipvfs VFS stack +** that does not include the RBU layer results in an error. +** +** The overhead of adding the "rbu" VFS to the system is negligible for +** non-RBU users. There is no harm in an application accessing the +** file-system via "rbu" all the time, even if it only uses RBU functionality +** occasionally. +*/ +SQLITE_API int sqlite3rbu_create_vfs(const char *zName, const char *zParent); + +/* +** Deregister and destroy an RBU vfs created by an earlier call to +** sqlite3rbu_create_vfs(). +** +** VFS objects are not reference counted. If a VFS object is destroyed +** before all database handles that use it have been closed, the results +** are undefined. +*/ +SQLITE_API void sqlite3rbu_destroy_vfs(const char *zName); + +#if 0 +} /* end of the 'extern "C"' block */ +#endif + +#endif /* _SQLITE3RBU_H */ + +/************** End of sqlite3rbu.h ******************************************/ +/************** Continuing where we left off in sqlite3rbu.c *****************/ + +#if defined(_WIN32_WCE) +/* #include "windows.h" */ +#endif + +/* Maximum number of prepared UPDATE statements held by this module */ +#define SQLITE_RBU_UPDATE_CACHESIZE 16 + +/* Delta checksums disabled by default. Compile with -DRBU_ENABLE_DELTA_CKSUM +** to enable checksum verification. +*/ +#ifndef RBU_ENABLE_DELTA_CKSUM +# define RBU_ENABLE_DELTA_CKSUM 0 +#endif + +/* +** Swap two objects of type TYPE. +*/ +#if !defined(SQLITE_AMALGAMATION) +# define SWAP(TYPE,A,B) {TYPE t=A; A=B; B=t;} +#endif + +/* +** The rbu_state table is used to save the state of a partially applied +** update so that it can be resumed later. The table consists of integer +** keys mapped to values as follows: +** +** RBU_STATE_STAGE: +** May be set to integer values 1, 2, 4 or 5. As follows: +** 1: the *-rbu file is currently under construction. +** 2: the *-rbu file has been constructed, but not yet moved +** to the *-wal path. +** 4: the checkpoint is underway. +** 5: the rbu update has been checkpointed. +** +** RBU_STATE_TBL: +** Only valid if STAGE==1. The target database name of the table +** currently being written. +** +** RBU_STATE_IDX: +** Only valid if STAGE==1. The target database name of the index +** currently being written, or NULL if the main table is currently being +** updated. +** +** RBU_STATE_ROW: +** Only valid if STAGE==1. Number of rows already processed for the current +** table/index. +** +** RBU_STATE_PROGRESS: +** Trbul number of sqlite3rbu_step() calls made so far as part of this +** rbu update. +** +** RBU_STATE_CKPT: +** Valid if STAGE==4. The 64-bit checksum associated with the wal-index +** header created by recovering the *-wal file. This is used to detect +** cases when another client appends frames to the *-wal file in the +** middle of an incremental checkpoint (an incremental checkpoint cannot +** be continued if this happens). +** +** RBU_STATE_COOKIE: +** Valid if STAGE==1. The current change-counter cookie value in the +** target db file. +** +** RBU_STATE_OALSZ: +** Valid if STAGE==1. The size in bytes of the *-oal file. +** +** RBU_STATE_DATATBL: +** Only valid if STAGE==1. The RBU database name of the table +** currently being read. +*/ +#define RBU_STATE_STAGE 1 +#define RBU_STATE_TBL 2 +#define RBU_STATE_IDX 3 +#define RBU_STATE_ROW 4 +#define RBU_STATE_PROGRESS 5 +#define RBU_STATE_CKPT 6 +#define RBU_STATE_COOKIE 7 +#define RBU_STATE_OALSZ 8 +#define RBU_STATE_PHASEONESTEP 9 +#define RBU_STATE_DATATBL 10 + +#define RBU_STAGE_OAL 1 +#define RBU_STAGE_MOVE 2 +#define RBU_STAGE_CAPTURE 3 +#define RBU_STAGE_CKPT 4 +#define RBU_STAGE_DONE 5 + + +#define RBU_CREATE_STATE \ + "CREATE TABLE IF NOT EXISTS %s.rbu_state(k INTEGER PRIMARY KEY, v)" + +typedef struct RbuFrame RbuFrame; +typedef struct RbuObjIter RbuObjIter; +typedef struct RbuState RbuState; +typedef struct RbuSpan RbuSpan; +typedef struct rbu_vfs rbu_vfs; +typedef struct rbu_file rbu_file; +typedef struct RbuUpdateStmt RbuUpdateStmt; + +#if !defined(SQLITE_AMALGAMATION) +typedef unsigned int u32; +typedef unsigned short u16; +typedef unsigned char u8; +typedef sqlite3_int64 i64; +#endif + +/* +** These values must match the values defined in wal.c for the equivalent +** locks. These are not magic numbers as they are part of the SQLite file +** format. +*/ +#define WAL_LOCK_WRITE 0 +#define WAL_LOCK_CKPT 1 +#define WAL_LOCK_READ0 3 + +#define SQLITE_FCNTL_RBUCNT 5149216 + +/* +** A structure to store values read from the rbu_state table in memory. +*/ +struct RbuState { + int eStage; + char *zTbl; + char *zDataTbl; + char *zIdx; + i64 iWalCksum; + int nRow; + i64 nProgress; + u32 iCookie; + i64 iOalSz; + i64 nPhaseOneStep; +}; + +struct RbuUpdateStmt { + char *zMask; /* Copy of update mask used with pUpdate */ + sqlite3_stmt *pUpdate; /* Last update statement (or NULL) */ + RbuUpdateStmt *pNext; +}; + +struct RbuSpan { + const char *zSpan; + int nSpan; +}; + +/* +** An iterator of this type is used to iterate through all objects in +** the target database that require updating. For each such table, the +** iterator visits, in order: +** +** * the table itself, +** * each index of the table (zero or more points to visit), and +** * a special "cleanup table" state. +** +** abIndexed: +** If the table has no indexes on it, abIndexed is set to NULL. Otherwise, +** it points to an array of flags nTblCol elements in size. The flag is +** set for each column that is either a part of the PK or a part of an +** index. Or clear otherwise. +** +** If there are one or more partial indexes on the table, all fields of +** this array set set to 1. This is because in that case, the module has +** no way to tell which fields will be required to add and remove entries +** from the partial indexes. +** +*/ +struct RbuObjIter { + sqlite3_stmt *pTblIter; /* Iterate through tables */ + sqlite3_stmt *pIdxIter; /* Index iterator */ + int nTblCol; /* Size of azTblCol[] array */ + char **azTblCol; /* Array of unquoted target column names */ + char **azTblType; /* Array of target column types */ + int *aiSrcOrder; /* src table col -> target table col */ + u8 *abTblPk; /* Array of flags, set on target PK columns */ + u8 *abNotNull; /* Array of flags, set on NOT NULL columns */ + u8 *abIndexed; /* Array of flags, set on indexed & PK cols */ + int eType; /* Table type - an RBU_PK_XXX value */ + + /* Output variables. zTbl==0 implies EOF. */ + int bCleanup; /* True in "cleanup" state */ + const char *zTbl; /* Name of target db table */ + const char *zDataTbl; /* Name of rbu db table (or null) */ + const char *zIdx; /* Name of target db index (or null) */ + int iTnum; /* Root page of current object */ + int iPkTnum; /* If eType==EXTERNAL, root of PK index */ + int bUnique; /* Current index is unique */ + int nIndex; /* Number of aux. indexes on table zTbl */ + + /* Statements created by rbuObjIterPrepareAll() */ + int nCol; /* Number of columns in current object */ + sqlite3_stmt *pSelect; /* Source data */ + sqlite3_stmt *pInsert; /* Statement for INSERT operations */ + sqlite3_stmt *pDelete; /* Statement for DELETE ops */ + sqlite3_stmt *pTmpInsert; /* Insert into rbu_tmp_$zDataTbl */ + int nIdxCol; + RbuSpan *aIdxCol; + char *zIdxSql; + + /* Last UPDATE used (for PK b-tree updates only), or NULL. */ + RbuUpdateStmt *pRbuUpdate; +}; + +/* +** Values for RbuObjIter.eType +** +** 0: Table does not exist (error) +** 1: Table has an implicit rowid. +** 2: Table has an explicit IPK column. +** 3: Table has an external PK index. +** 4: Table is WITHOUT ROWID. +** 5: Table is a virtual table. +*/ +#define RBU_PK_NOTABLE 0 +#define RBU_PK_NONE 1 +#define RBU_PK_IPK 2 +#define RBU_PK_EXTERNAL 3 +#define RBU_PK_WITHOUT_ROWID 4 +#define RBU_PK_VTAB 5 + + +/* +** Within the RBU_STAGE_OAL stage, each call to sqlite3rbu_step() performs +** one of the following operations. +*/ +#define RBU_INSERT 1 /* Insert on a main table b-tree */ +#define RBU_DELETE 2 /* Delete a row from a main table b-tree */ +#define RBU_REPLACE 3 /* Delete and then insert a row */ +#define RBU_IDX_DELETE 4 /* Delete a row from an aux. index b-tree */ +#define RBU_IDX_INSERT 5 /* Insert on an aux. index b-tree */ + +#define RBU_UPDATE 6 /* Update a row in a main table b-tree */ + +/* +** A single step of an incremental checkpoint - frame iWalFrame of the wal +** file should be copied to page iDbPage of the database file. +*/ +struct RbuFrame { + u32 iDbPage; + u32 iWalFrame; +}; + +/* +** RBU handle. +** +** nPhaseOneStep: +** If the RBU database contains an rbu_count table, this value is set to +** a running estimate of the number of b-tree operations required to +** finish populating the *-oal file. This allows the sqlite3_bp_progress() +** API to calculate the permyriadage progress of populating the *-oal file +** using the formula: +** +** permyriadage = (10000 * nProgress) / nPhaseOneStep +** +** nPhaseOneStep is initialized to the sum of: +** +** nRow * (nIndex + 1) +** +** for all source tables in the RBU database, where nRow is the number +** of rows in the source table and nIndex the number of indexes on the +** corresponding target database table. +** +** This estimate is accurate if the RBU update consists entirely of +** INSERT operations. However, it is inaccurate if: +** +** * the RBU update contains any UPDATE operations. If the PK specified +** for an UPDATE operation does not exist in the target table, then +** no b-tree operations are required on index b-trees. Or if the +** specified PK does exist, then (nIndex*2) such operations are +** required (one delete and one insert on each index b-tree). +** +** * the RBU update contains any DELETE operations for which the specified +** PK does not exist. In this case no operations are required on index +** b-trees. +** +** * the RBU update contains REPLACE operations. These are similar to +** UPDATE operations. +** +** nPhaseOneStep is updated to account for the conditions above during the +** first pass of each source table. The updated nPhaseOneStep value is +** stored in the rbu_state table if the RBU update is suspended. +*/ +struct sqlite3rbu { + int eStage; /* Value of RBU_STATE_STAGE field */ + sqlite3 *dbMain; /* target database handle */ + sqlite3 *dbRbu; /* rbu database handle */ + char *zTarget; /* Path to target db */ + char *zRbu; /* Path to rbu db */ + char *zState; /* Path to state db (or NULL if zRbu) */ + char zStateDb[5]; /* Db name for state ("stat" or "main") */ + int rc; /* Value returned by last rbu_step() call */ + char *zErrmsg; /* Error message if rc!=SQLITE_OK */ + int nStep; /* Rows processed for current object */ + int nProgress; /* Rows processed for all objects */ + RbuObjIter objiter; /* Iterator for skipping through tbl/idx */ + const char *zVfsName; /* Name of automatically created rbu vfs */ + rbu_file *pTargetFd; /* File handle open on target db */ + int nPagePerSector; /* Pages per sector for pTargetFd */ + i64 iOalSz; + i64 nPhaseOneStep; + + /* The following state variables are used as part of the incremental + ** checkpoint stage (eStage==RBU_STAGE_CKPT). See comments surrounding + ** function rbuSetupCheckpoint() for details. */ + u32 iMaxFrame; /* Largest iWalFrame value in aFrame[] */ + u32 mLock; + int nFrame; /* Entries in aFrame[] array */ + int nFrameAlloc; /* Allocated size of aFrame[] array */ + RbuFrame *aFrame; + int pgsz; + u8 *aBuf; + i64 iWalCksum; + i64 szTemp; /* Current size of all temp files in use */ + i64 szTempLimit; /* Total size limit for temp files */ + + /* Used in RBU vacuum mode only */ + int nRbu; /* Number of RBU VFS in the stack */ + rbu_file *pRbuFd; /* Fd for main db of dbRbu */ +}; + +/* +** An rbu VFS is implemented using an instance of this structure. +** +** Variable pRbu is only non-NULL for automatically created RBU VFS objects. +** It is NULL for RBU VFS objects created explicitly using +** sqlite3rbu_create_vfs(). It is used to track the total amount of temp +** space used by the RBU handle. +*/ +struct rbu_vfs { + sqlite3_vfs base; /* rbu VFS shim methods */ + sqlite3_vfs *pRealVfs; /* Underlying VFS */ + sqlite3_mutex *mutex; /* Mutex to protect pMain */ + sqlite3rbu *pRbu; /* Owner RBU object */ + rbu_file *pMain; /* List of main db files */ + rbu_file *pMainRbu; /* List of main db files with pRbu!=0 */ +}; + +/* +** Each file opened by an rbu VFS is represented by an instance of +** the following structure. +** +** If this is a temporary file (pRbu!=0 && flags&DELETE_ON_CLOSE), variable +** "sz" is set to the current size of the database file. +*/ +struct rbu_file { + sqlite3_file base; /* sqlite3_file methods */ + sqlite3_file *pReal; /* Underlying file handle */ + rbu_vfs *pRbuVfs; /* Pointer to the rbu_vfs object */ + sqlite3rbu *pRbu; /* Pointer to rbu object (rbu target only) */ + i64 sz; /* Size of file in bytes (temp only) */ + + int openFlags; /* Flags this file was opened with */ + u32 iCookie; /* Cookie value for main db files */ + u8 iWriteVer; /* "write-version" value for main db files */ + u8 bNolock; /* True to fail EXCLUSIVE locks */ + + int nShm; /* Number of entries in apShm[] array */ + char **apShm; /* Array of mmap'd *-shm regions */ + char *zDel; /* Delete this when closing file */ + + const char *zWal; /* Wal filename for this main db file */ + rbu_file *pWalFd; /* Wal file descriptor for this main db */ + rbu_file *pMainNext; /* Next MAIN_DB file */ + rbu_file *pMainRbuNext; /* Next MAIN_DB file with pRbu!=0 */ +}; + +/* +** True for an RBU vacuum handle, or false otherwise. +*/ +#define rbuIsVacuum(p) ((p)->zTarget==0) + + +/************************************************************************* +** The following three functions, found below: +** +** rbuDeltaGetInt() +** rbuDeltaChecksum() +** rbuDeltaApply() +** +** are lifted from the fossil source code (http://fossil-scm.org). They +** are used to implement the scalar SQL function rbu_fossil_delta(). +*/ + +/* +** Read bytes from *pz and convert them into a positive integer. When +** finished, leave *pz pointing to the first character past the end of +** the integer. The *pLen parameter holds the length of the string +** in *pz and is decremented once for each character in the integer. +*/ +static unsigned int rbuDeltaGetInt(const char **pz, int *pLen){ + static const signed char zValue[] = { + -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, + -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, + -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, + 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, -1, -1, -1, -1, -1, -1, + -1, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, + 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, -1, -1, -1, -1, 36, + -1, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, + 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 62, -1, -1, -1, 63, -1, + }; + unsigned int v = 0; + int c; + unsigned char *z = (unsigned char*)*pz; + unsigned char *zStart = z; + while( (c = zValue[0x7f&*(z++)])>=0 ){ + v = (v<<6) + c; + } + z--; + *pLen -= z - zStart; + *pz = (char*)z; + return v; +} + +#if RBU_ENABLE_DELTA_CKSUM +/* +** Compute a 32-bit checksum on the N-byte buffer. Return the result. +*/ +static unsigned int rbuDeltaChecksum(const char *zIn, size_t N){ + const unsigned char *z = (const unsigned char *)zIn; + unsigned sum0 = 0; + unsigned sum1 = 0; + unsigned sum2 = 0; + unsigned sum3 = 0; + while(N >= 16){ + sum0 += ((unsigned)z[0] + z[4] + z[8] + z[12]); + sum1 += ((unsigned)z[1] + z[5] + z[9] + z[13]); + sum2 += ((unsigned)z[2] + z[6] + z[10]+ z[14]); + sum3 += ((unsigned)z[3] + z[7] + z[11]+ z[15]); + z += 16; + N -= 16; + } + while(N >= 4){ + sum0 += z[0]; + sum1 += z[1]; + sum2 += z[2]; + sum3 += z[3]; + z += 4; + N -= 4; + } + sum3 += (sum2 << 8) + (sum1 << 16) + (sum0 << 24); + switch(N){ + case 3: sum3 += (z[2] << 8); + case 2: sum3 += (z[1] << 16); + case 1: sum3 += (z[0] << 24); + default: ; + } + return sum3; +} +#endif + +/* +** Apply a delta. +** +** The output buffer should be big enough to hold the whole output +** file and a NUL terminator at the end. The delta_output_size() +** routine will determine this size for you. +** +** The delta string should be null-terminated. But the delta string +** may contain embedded NUL characters (if the input and output are +** binary files) so we also have to pass in the length of the delta in +** the lenDelta parameter. +** +** This function returns the size of the output file in bytes (excluding +** the final NUL terminator character). Except, if the delta string is +** malformed or intended for use with a source file other than zSrc, +** then this routine returns -1. +** +** Refer to the delta_create() documentation above for a description +** of the delta file format. +*/ +static int rbuDeltaApply( + const char *zSrc, /* The source or pattern file */ + int lenSrc, /* Length of the source file */ + const char *zDelta, /* Delta to apply to the pattern */ + int lenDelta, /* Length of the delta */ + char *zOut /* Write the output into this preallocated buffer */ +){ + unsigned int limit; + unsigned int total = 0; +#if RBU_ENABLE_DELTA_CKSUM + char *zOrigOut = zOut; +#endif + + limit = rbuDeltaGetInt(&zDelta, &lenDelta); + if( *zDelta!='\n' ){ + /* ERROR: size integer not terminated by "\n" */ + return -1; + } + zDelta++; lenDelta--; + while( *zDelta && lenDelta>0 ){ + unsigned int cnt, ofst; + cnt = rbuDeltaGetInt(&zDelta, &lenDelta); + switch( zDelta[0] ){ + case '@': { + zDelta++; lenDelta--; + ofst = rbuDeltaGetInt(&zDelta, &lenDelta); + if( lenDelta>0 && zDelta[0]!=',' ){ + /* ERROR: copy command not terminated by ',' */ + return -1; + } + zDelta++; lenDelta--; + total += cnt; + if( total>limit ){ + /* ERROR: copy exceeds output file size */ + return -1; + } + if( (int)(ofst+cnt) > lenSrc ){ + /* ERROR: copy extends past end of input */ + return -1; + } + memcpy(zOut, &zSrc[ofst], cnt); + zOut += cnt; + break; + } + case ':': { + zDelta++; lenDelta--; + total += cnt; + if( total>limit ){ + /* ERROR: insert command gives an output larger than predicted */ + return -1; + } + if( (int)cnt>lenDelta ){ + /* ERROR: insert count exceeds size of delta */ + return -1; + } + memcpy(zOut, zDelta, cnt); + zOut += cnt; + zDelta += cnt; + lenDelta -= cnt; + break; + } + case ';': { + zDelta++; lenDelta--; + zOut[0] = 0; +#if RBU_ENABLE_DELTA_CKSUM + if( cnt!=rbuDeltaChecksum(zOrigOut, total) ){ + /* ERROR: bad checksum */ + return -1; + } +#endif + if( total!=limit ){ + /* ERROR: generated size does not match predicted size */ + return -1; + } + return total; + } + default: { + /* ERROR: unknown delta operator */ + return -1; + } + } + } + /* ERROR: unterminated delta */ + return -1; +} + +static int rbuDeltaOutputSize(const char *zDelta, int lenDelta){ + int size; + size = rbuDeltaGetInt(&zDelta, &lenDelta); + if( *zDelta!='\n' ){ + /* ERROR: size integer not terminated by "\n" */ + return -1; + } + return size; +} + +/* +** End of code taken from fossil. +*************************************************************************/ + +/* +** Implementation of SQL scalar function rbu_fossil_delta(). +** +** This function applies a fossil delta patch to a blob. Exactly two +** arguments must be passed to this function. The first is the blob to +** patch and the second the patch to apply. If no error occurs, this +** function returns the patched blob. +*/ +static void rbuFossilDeltaFunc( + sqlite3_context *context, + int argc, + sqlite3_value **argv +){ + const char *aDelta; + int nDelta; + const char *aOrig; + int nOrig; + + int nOut; + int nOut2; + char *aOut; + + assert( argc==2 ); + + nOrig = sqlite3_value_bytes(argv[0]); + aOrig = (const char*)sqlite3_value_blob(argv[0]); + nDelta = sqlite3_value_bytes(argv[1]); + aDelta = (const char*)sqlite3_value_blob(argv[1]); + + /* Figure out the size of the output */ + nOut = rbuDeltaOutputSize(aDelta, nDelta); + if( nOut<0 ){ + sqlite3_result_error(context, "corrupt fossil delta", -1); + return; + } + + aOut = sqlite3_malloc(nOut+1); + if( aOut==0 ){ + sqlite3_result_error_nomem(context); + }else{ + nOut2 = rbuDeltaApply(aOrig, nOrig, aDelta, nDelta, aOut); + if( nOut2!=nOut ){ + sqlite3_free(aOut); + sqlite3_result_error(context, "corrupt fossil delta", -1); + }else{ + sqlite3_result_blob(context, aOut, nOut, sqlite3_free); + } + } +} + + +/* +** Prepare the SQL statement in buffer zSql against database handle db. +** If successful, set *ppStmt to point to the new statement and return +** SQLITE_OK. +** +** Otherwise, if an error does occur, set *ppStmt to NULL and return +** an SQLite error code. Additionally, set output variable *pzErrmsg to +** point to a buffer containing an error message. It is the responsibility +** of the caller to (eventually) free this buffer using sqlite3_free(). +*/ +static int prepareAndCollectError( + sqlite3 *db, + sqlite3_stmt **ppStmt, + char **pzErrmsg, + const char *zSql +){ + int rc = sqlite3_prepare_v2(db, zSql, -1, ppStmt, 0); + if( rc!=SQLITE_OK ){ + *pzErrmsg = sqlite3_mprintf("%s", sqlite3_errmsg(db)); + *ppStmt = 0; + } + return rc; +} + +/* +** Reset the SQL statement passed as the first argument. Return a copy +** of the value returned by sqlite3_reset(). +** +** If an error has occurred, then set *pzErrmsg to point to a buffer +** containing an error message. It is the responsibility of the caller +** to eventually free this buffer using sqlite3_free(). +*/ +static int resetAndCollectError(sqlite3_stmt *pStmt, char **pzErrmsg){ + int rc = sqlite3_reset(pStmt); + if( rc!=SQLITE_OK ){ + *pzErrmsg = sqlite3_mprintf("%s", sqlite3_errmsg(sqlite3_db_handle(pStmt))); + } + return rc; +} + +/* +** Unless it is NULL, argument zSql points to a buffer allocated using +** sqlite3_malloc containing an SQL statement. This function prepares the SQL +** statement against database db and frees the buffer. If statement +** compilation is successful, *ppStmt is set to point to the new statement +** handle and SQLITE_OK is returned. +** +** Otherwise, if an error occurs, *ppStmt is set to NULL and an error code +** returned. In this case, *pzErrmsg may also be set to point to an error +** message. It is the responsibility of the caller to free this error message +** buffer using sqlite3_free(). +** +** If argument zSql is NULL, this function assumes that an OOM has occurred. +** In this case SQLITE_NOMEM is returned and *ppStmt set to NULL. +*/ +static int prepareFreeAndCollectError( + sqlite3 *db, + sqlite3_stmt **ppStmt, + char **pzErrmsg, + char *zSql +){ + int rc; + assert( *pzErrmsg==0 ); + if( zSql==0 ){ + rc = SQLITE_NOMEM; + *ppStmt = 0; + }else{ + rc = prepareAndCollectError(db, ppStmt, pzErrmsg, zSql); + sqlite3_free(zSql); + } + return rc; +} + +/* +** Free the RbuObjIter.azTblCol[] and RbuObjIter.abTblPk[] arrays allocated +** by an earlier call to rbuObjIterCacheTableInfo(). +*/ +static void rbuObjIterFreeCols(RbuObjIter *pIter){ + int i; + for(i=0; inTblCol; i++){ + sqlite3_free(pIter->azTblCol[i]); + sqlite3_free(pIter->azTblType[i]); + } + sqlite3_free(pIter->azTblCol); + pIter->azTblCol = 0; + pIter->azTblType = 0; + pIter->aiSrcOrder = 0; + pIter->abTblPk = 0; + pIter->abNotNull = 0; + pIter->nTblCol = 0; + pIter->eType = 0; /* Invalid value */ +} + +/* +** Finalize all statements and free all allocations that are specific to +** the current object (table/index pair). +*/ +static void rbuObjIterClearStatements(RbuObjIter *pIter){ + RbuUpdateStmt *pUp; + + sqlite3_finalize(pIter->pSelect); + sqlite3_finalize(pIter->pInsert); + sqlite3_finalize(pIter->pDelete); + sqlite3_finalize(pIter->pTmpInsert); + pUp = pIter->pRbuUpdate; + while( pUp ){ + RbuUpdateStmt *pTmp = pUp->pNext; + sqlite3_finalize(pUp->pUpdate); + sqlite3_free(pUp); + pUp = pTmp; + } + sqlite3_free(pIter->aIdxCol); + sqlite3_free(pIter->zIdxSql); + + pIter->pSelect = 0; + pIter->pInsert = 0; + pIter->pDelete = 0; + pIter->pRbuUpdate = 0; + pIter->pTmpInsert = 0; + pIter->nCol = 0; + pIter->nIdxCol = 0; + pIter->aIdxCol = 0; + pIter->zIdxSql = 0; +} + +/* +** Clean up any resources allocated as part of the iterator object passed +** as the only argument. +*/ +static void rbuObjIterFinalize(RbuObjIter *pIter){ + rbuObjIterClearStatements(pIter); + sqlite3_finalize(pIter->pTblIter); + sqlite3_finalize(pIter->pIdxIter); + rbuObjIterFreeCols(pIter); + memset(pIter, 0, sizeof(RbuObjIter)); +} + +/* +** Advance the iterator to the next position. +** +** If no error occurs, SQLITE_OK is returned and the iterator is left +** pointing to the next entry. Otherwise, an error code and message is +** left in the RBU handle passed as the first argument. A copy of the +** error code is returned. +*/ +static int rbuObjIterNext(sqlite3rbu *p, RbuObjIter *pIter){ + int rc = p->rc; + if( rc==SQLITE_OK ){ + + /* Free any SQLite statements used while processing the previous object */ + rbuObjIterClearStatements(pIter); + if( pIter->zIdx==0 ){ + rc = sqlite3_exec(p->dbMain, + "DROP TRIGGER IF EXISTS temp.rbu_insert_tr;" + "DROP TRIGGER IF EXISTS temp.rbu_update1_tr;" + "DROP TRIGGER IF EXISTS temp.rbu_update2_tr;" + "DROP TRIGGER IF EXISTS temp.rbu_delete_tr;" + , 0, 0, &p->zErrmsg + ); + } + + if( rc==SQLITE_OK ){ + if( pIter->bCleanup ){ + rbuObjIterFreeCols(pIter); + pIter->bCleanup = 0; + rc = sqlite3_step(pIter->pTblIter); + if( rc!=SQLITE_ROW ){ + rc = resetAndCollectError(pIter->pTblIter, &p->zErrmsg); + pIter->zTbl = 0; + }else{ + pIter->zTbl = (const char*)sqlite3_column_text(pIter->pTblIter, 0); + pIter->zDataTbl = (const char*)sqlite3_column_text(pIter->pTblIter,1); + rc = (pIter->zDataTbl && pIter->zTbl) ? SQLITE_OK : SQLITE_NOMEM; + } + }else{ + if( pIter->zIdx==0 ){ + sqlite3_stmt *pIdx = pIter->pIdxIter; + rc = sqlite3_bind_text(pIdx, 1, pIter->zTbl, -1, SQLITE_STATIC); + } + if( rc==SQLITE_OK ){ + rc = sqlite3_step(pIter->pIdxIter); + if( rc!=SQLITE_ROW ){ + rc = resetAndCollectError(pIter->pIdxIter, &p->zErrmsg); + pIter->bCleanup = 1; + pIter->zIdx = 0; + }else{ + pIter->zIdx = (const char*)sqlite3_column_text(pIter->pIdxIter, 0); + pIter->iTnum = sqlite3_column_int(pIter->pIdxIter, 1); + pIter->bUnique = sqlite3_column_int(pIter->pIdxIter, 2); + rc = pIter->zIdx ? SQLITE_OK : SQLITE_NOMEM; + } + } + } + } + } + + if( rc!=SQLITE_OK ){ + rbuObjIterFinalize(pIter); + p->rc = rc; + } + return rc; +} + + +/* +** The implementation of the rbu_target_name() SQL function. This function +** accepts one or two arguments. The first argument is the name of a table - +** the name of a table in the RBU database. The second, if it is present, is 1 +** for a view or 0 for a table. +** +** For a non-vacuum RBU handle, if the table name matches the pattern: +** +** data[0-9]_ +** +** where is any sequence of 1 or more characters, is returned. +** Otherwise, if the only argument does not match the above pattern, an SQL +** NULL is returned. +** +** "data_t1" -> "t1" +** "data0123_t2" -> "t2" +** "dataAB_t3" -> NULL +** +** For an rbu vacuum handle, a copy of the first argument is returned if +** the second argument is either missing or 0 (not a view). +*/ +static void rbuTargetNameFunc( + sqlite3_context *pCtx, + int argc, + sqlite3_value **argv +){ + sqlite3rbu *p = sqlite3_user_data(pCtx); + const char *zIn; + assert( argc==1 || argc==2 ); + + zIn = (const char*)sqlite3_value_text(argv[0]); + if( zIn ){ + if( rbuIsVacuum(p) ){ + assert( argc==2 || argc==1 ); + if( argc==1 || 0==sqlite3_value_int(argv[1]) ){ + sqlite3_result_text(pCtx, zIn, -1, SQLITE_STATIC); + } + }else{ + if( strlen(zIn)>4 && memcmp("data", zIn, 4)==0 ){ + int i; + for(i=4; zIn[i]>='0' && zIn[i]<='9'; i++); + if( zIn[i]=='_' && zIn[i+1] ){ + sqlite3_result_text(pCtx, &zIn[i+1], -1, SQLITE_STATIC); + } + } + } + } +} + +/* +** Initialize the iterator structure passed as the second argument. +** +** If no error occurs, SQLITE_OK is returned and the iterator is left +** pointing to the first entry. Otherwise, an error code and message is +** left in the RBU handle passed as the first argument. A copy of the +** error code is returned. +*/ +static int rbuObjIterFirst(sqlite3rbu *p, RbuObjIter *pIter){ + int rc; + memset(pIter, 0, sizeof(RbuObjIter)); + + rc = prepareFreeAndCollectError(p->dbRbu, &pIter->pTblIter, &p->zErrmsg, + sqlite3_mprintf( + "SELECT rbu_target_name(name, type='view') AS target, name " + "FROM sqlite_master " + "WHERE type IN ('table', 'view') AND target IS NOT NULL " + " %s " + "ORDER BY name" + , rbuIsVacuum(p) ? "AND rootpage!=0 AND rootpage IS NOT NULL" : "")); + + if( rc==SQLITE_OK ){ + rc = prepareAndCollectError(p->dbMain, &pIter->pIdxIter, &p->zErrmsg, + "SELECT name, rootpage, sql IS NULL OR substr(8, 6)=='UNIQUE' " + " FROM main.sqlite_master " + " WHERE type='index' AND tbl_name = ?" + ); + } + + pIter->bCleanup = 1; + p->rc = rc; + return rbuObjIterNext(p, pIter); +} + +/* +** This is a wrapper around "sqlite3_mprintf(zFmt, ...)". If an OOM occurs, +** an error code is stored in the RBU handle passed as the first argument. +** +** If an error has already occurred (p->rc is already set to something other +** than SQLITE_OK), then this function returns NULL without modifying the +** stored error code. In this case it still calls sqlite3_free() on any +** printf() parameters associated with %z conversions. +*/ +static char *rbuMPrintf(sqlite3rbu *p, const char *zFmt, ...){ + char *zSql = 0; + va_list ap; + va_start(ap, zFmt); + zSql = sqlite3_vmprintf(zFmt, ap); + if( p->rc==SQLITE_OK ){ + if( zSql==0 ) p->rc = SQLITE_NOMEM; + }else{ + sqlite3_free(zSql); + zSql = 0; + } + va_end(ap); + return zSql; +} + +/* +** Argument zFmt is a sqlite3_mprintf() style format string. The trailing +** arguments are the usual subsitution values. This function performs +** the printf() style substitutions and executes the result as an SQL +** statement on the RBU handles database. +** +** If an error occurs, an error code and error message is stored in the +** RBU handle. If an error has already occurred when this function is +** called, it is a no-op. +*/ +static int rbuMPrintfExec(sqlite3rbu *p, sqlite3 *db, const char *zFmt, ...){ + va_list ap; + char *zSql; + va_start(ap, zFmt); + zSql = sqlite3_vmprintf(zFmt, ap); + if( p->rc==SQLITE_OK ){ + if( zSql==0 ){ + p->rc = SQLITE_NOMEM; + }else{ + p->rc = sqlite3_exec(db, zSql, 0, 0, &p->zErrmsg); + } + } + sqlite3_free(zSql); + va_end(ap); + return p->rc; +} + +/* +** Attempt to allocate and return a pointer to a zeroed block of nByte +** bytes. +** +** If an error (i.e. an OOM condition) occurs, return NULL and leave an +** error code in the rbu handle passed as the first argument. Or, if an +** error has already occurred when this function is called, return NULL +** immediately without attempting the allocation or modifying the stored +** error code. +*/ +static void *rbuMalloc(sqlite3rbu *p, sqlite3_int64 nByte){ + void *pRet = 0; + if( p->rc==SQLITE_OK ){ + assert( nByte>0 ); + pRet = sqlite3_malloc64(nByte); + if( pRet==0 ){ + p->rc = SQLITE_NOMEM; + }else{ + memset(pRet, 0, nByte); + } + } + return pRet; +} + + +/* +** Allocate and zero the pIter->azTblCol[] and abTblPk[] arrays so that +** there is room for at least nCol elements. If an OOM occurs, store an +** error code in the RBU handle passed as the first argument. +*/ +static void rbuAllocateIterArrays(sqlite3rbu *p, RbuObjIter *pIter, int nCol){ + sqlite3_int64 nByte = (2*sizeof(char*) + sizeof(int) + 3*sizeof(u8)) * nCol; + char **azNew; + + azNew = (char**)rbuMalloc(p, nByte); + if( azNew ){ + pIter->azTblCol = azNew; + pIter->azTblType = &azNew[nCol]; + pIter->aiSrcOrder = (int*)&pIter->azTblType[nCol]; + pIter->abTblPk = (u8*)&pIter->aiSrcOrder[nCol]; + pIter->abNotNull = (u8*)&pIter->abTblPk[nCol]; + pIter->abIndexed = (u8*)&pIter->abNotNull[nCol]; + } +} + +/* +** The first argument must be a nul-terminated string. This function +** returns a copy of the string in memory obtained from sqlite3_malloc(). +** It is the responsibility of the caller to eventually free this memory +** using sqlite3_free(). +** +** If an OOM condition is encountered when attempting to allocate memory, +** output variable (*pRc) is set to SQLITE_NOMEM before returning. Otherwise, +** if the allocation succeeds, (*pRc) is left unchanged. +*/ +static char *rbuStrndup(const char *zStr, int *pRc){ + char *zRet = 0; + + if( *pRc==SQLITE_OK ){ + if( zStr ){ + size_t nCopy = strlen(zStr) + 1; + zRet = (char*)sqlite3_malloc64(nCopy); + if( zRet ){ + memcpy(zRet, zStr, nCopy); + }else{ + *pRc = SQLITE_NOMEM; + } + } + } + + return zRet; +} + +/* +** Finalize the statement passed as the second argument. +** +** If the sqlite3_finalize() call indicates that an error occurs, and the +** rbu handle error code is not already set, set the error code and error +** message accordingly. +*/ +static void rbuFinalize(sqlite3rbu *p, sqlite3_stmt *pStmt){ + sqlite3 *db = sqlite3_db_handle(pStmt); + int rc = sqlite3_finalize(pStmt); + if( p->rc==SQLITE_OK && rc!=SQLITE_OK ){ + p->rc = rc; + p->zErrmsg = sqlite3_mprintf("%s", sqlite3_errmsg(db)); + } +} + +/* Determine the type of a table. +** +** peType is of type (int*), a pointer to an output parameter of type +** (int). This call sets the output parameter as follows, depending +** on the type of the table specified by parameters dbName and zTbl. +** +** RBU_PK_NOTABLE: No such table. +** RBU_PK_NONE: Table has an implicit rowid. +** RBU_PK_IPK: Table has an explicit IPK column. +** RBU_PK_EXTERNAL: Table has an external PK index. +** RBU_PK_WITHOUT_ROWID: Table is WITHOUT ROWID. +** RBU_PK_VTAB: Table is a virtual table. +** +** Argument *piPk is also of type (int*), and also points to an output +** parameter. Unless the table has an external primary key index +** (i.e. unless *peType is set to 3), then *piPk is set to zero. Or, +** if the table does have an external primary key index, then *piPk +** is set to the root page number of the primary key index before +** returning. +** +** ALGORITHM: +** +** if( no entry exists in sqlite_master ){ +** return RBU_PK_NOTABLE +** }else if( sql for the entry starts with "CREATE VIRTUAL" ){ +** return RBU_PK_VTAB +** }else if( "PRAGMA index_list()" for the table contains a "pk" index ){ +** if( the index that is the pk exists in sqlite_master ){ +** *piPK = rootpage of that index. +** return RBU_PK_EXTERNAL +** }else{ +** return RBU_PK_WITHOUT_ROWID +** } +** }else if( "PRAGMA table_info()" lists one or more "pk" columns ){ +** return RBU_PK_IPK +** }else{ +** return RBU_PK_NONE +** } +*/ +static void rbuTableType( + sqlite3rbu *p, + const char *zTab, + int *peType, + int *piTnum, + int *piPk +){ + /* + ** 0) SELECT count(*) FROM sqlite_master where name=%Q AND IsVirtual(%Q) + ** 1) PRAGMA index_list = ? + ** 2) SELECT count(*) FROM sqlite_master where name=%Q + ** 3) PRAGMA table_info = ? + */ + sqlite3_stmt *aStmt[4] = {0, 0, 0, 0}; + + *peType = RBU_PK_NOTABLE; + *piPk = 0; + + assert( p->rc==SQLITE_OK ); + p->rc = prepareFreeAndCollectError(p->dbMain, &aStmt[0], &p->zErrmsg, + sqlite3_mprintf( + "SELECT (sql LIKE 'create virtual%%'), rootpage" + " FROM sqlite_master" + " WHERE name=%Q", zTab + )); + if( p->rc!=SQLITE_OK || sqlite3_step(aStmt[0])!=SQLITE_ROW ){ + /* Either an error, or no such table. */ + goto rbuTableType_end; + } + if( sqlite3_column_int(aStmt[0], 0) ){ + *peType = RBU_PK_VTAB; /* virtual table */ + goto rbuTableType_end; + } + *piTnum = sqlite3_column_int(aStmt[0], 1); + + p->rc = prepareFreeAndCollectError(p->dbMain, &aStmt[1], &p->zErrmsg, + sqlite3_mprintf("PRAGMA index_list=%Q",zTab) + ); + if( p->rc ) goto rbuTableType_end; + while( sqlite3_step(aStmt[1])==SQLITE_ROW ){ + const u8 *zOrig = sqlite3_column_text(aStmt[1], 3); + const u8 *zIdx = sqlite3_column_text(aStmt[1], 1); + if( zOrig && zIdx && zOrig[0]=='p' ){ + p->rc = prepareFreeAndCollectError(p->dbMain, &aStmt[2], &p->zErrmsg, + sqlite3_mprintf( + "SELECT rootpage FROM sqlite_master WHERE name = %Q", zIdx + )); + if( p->rc==SQLITE_OK ){ + if( sqlite3_step(aStmt[2])==SQLITE_ROW ){ + *piPk = sqlite3_column_int(aStmt[2], 0); + *peType = RBU_PK_EXTERNAL; + }else{ + *peType = RBU_PK_WITHOUT_ROWID; + } + } + goto rbuTableType_end; + } + } + + p->rc = prepareFreeAndCollectError(p->dbMain, &aStmt[3], &p->zErrmsg, + sqlite3_mprintf("PRAGMA table_info=%Q",zTab) + ); + if( p->rc==SQLITE_OK ){ + while( sqlite3_step(aStmt[3])==SQLITE_ROW ){ + if( sqlite3_column_int(aStmt[3],5)>0 ){ + *peType = RBU_PK_IPK; /* explicit IPK column */ + goto rbuTableType_end; + } + } + *peType = RBU_PK_NONE; + } + +rbuTableType_end: { + unsigned int i; + for(i=0; iabIndexed[] array. +*/ +static void rbuObjIterCacheIndexedCols(sqlite3rbu *p, RbuObjIter *pIter){ + sqlite3_stmt *pList = 0; + int bIndex = 0; + + if( p->rc==SQLITE_OK ){ + memcpy(pIter->abIndexed, pIter->abTblPk, sizeof(u8)*pIter->nTblCol); + p->rc = prepareFreeAndCollectError(p->dbMain, &pList, &p->zErrmsg, + sqlite3_mprintf("PRAGMA main.index_list = %Q", pIter->zTbl) + ); + } + + pIter->nIndex = 0; + while( p->rc==SQLITE_OK && SQLITE_ROW==sqlite3_step(pList) ){ + const char *zIdx = (const char*)sqlite3_column_text(pList, 1); + int bPartial = sqlite3_column_int(pList, 4); + sqlite3_stmt *pXInfo = 0; + if( zIdx==0 ) break; + if( bPartial ){ + memset(pIter->abIndexed, 0x01, sizeof(u8)*pIter->nTblCol); + } + p->rc = prepareFreeAndCollectError(p->dbMain, &pXInfo, &p->zErrmsg, + sqlite3_mprintf("PRAGMA main.index_xinfo = %Q", zIdx) + ); + while( p->rc==SQLITE_OK && SQLITE_ROW==sqlite3_step(pXInfo) ){ + int iCid = sqlite3_column_int(pXInfo, 1); + if( iCid>=0 ) pIter->abIndexed[iCid] = 1; + if( iCid==-2 ){ + memset(pIter->abIndexed, 0x01, sizeof(u8)*pIter->nTblCol); + } + } + rbuFinalize(p, pXInfo); + bIndex = 1; + pIter->nIndex++; + } + + if( pIter->eType==RBU_PK_WITHOUT_ROWID ){ + /* "PRAGMA index_list" includes the main PK b-tree */ + pIter->nIndex--; + } + + rbuFinalize(p, pList); + if( bIndex==0 ) pIter->abIndexed = 0; +} + + +/* +** If they are not already populated, populate the pIter->azTblCol[], +** pIter->abTblPk[], pIter->nTblCol and pIter->bRowid variables according to +** the table (not index) that the iterator currently points to. +** +** Return SQLITE_OK if successful, or an SQLite error code otherwise. If +** an error does occur, an error code and error message are also left in +** the RBU handle. +*/ +static int rbuObjIterCacheTableInfo(sqlite3rbu *p, RbuObjIter *pIter){ + if( pIter->azTblCol==0 ){ + sqlite3_stmt *pStmt = 0; + int nCol = 0; + int i; /* for() loop iterator variable */ + int bRbuRowid = 0; /* If input table has column "rbu_rowid" */ + int iOrder = 0; + int iTnum = 0; + + /* Figure out the type of table this step will deal with. */ + assert( pIter->eType==0 ); + rbuTableType(p, pIter->zTbl, &pIter->eType, &iTnum, &pIter->iPkTnum); + if( p->rc==SQLITE_OK && pIter->eType==RBU_PK_NOTABLE ){ + p->rc = SQLITE_ERROR; + p->zErrmsg = sqlite3_mprintf("no such table: %s", pIter->zTbl); + } + if( p->rc ) return p->rc; + if( pIter->zIdx==0 ) pIter->iTnum = iTnum; + + assert( pIter->eType==RBU_PK_NONE || pIter->eType==RBU_PK_IPK + || pIter->eType==RBU_PK_EXTERNAL || pIter->eType==RBU_PK_WITHOUT_ROWID + || pIter->eType==RBU_PK_VTAB + ); + + /* Populate the azTblCol[] and nTblCol variables based on the columns + ** of the input table. Ignore any input table columns that begin with + ** "rbu_". */ + p->rc = prepareFreeAndCollectError(p->dbRbu, &pStmt, &p->zErrmsg, + sqlite3_mprintf("SELECT * FROM '%q'", pIter->zDataTbl) + ); + if( p->rc==SQLITE_OK ){ + nCol = sqlite3_column_count(pStmt); + rbuAllocateIterArrays(p, pIter, nCol); + } + for(i=0; p->rc==SQLITE_OK && irc); + pIter->aiSrcOrder[pIter->nTblCol] = pIter->nTblCol; + pIter->azTblCol[pIter->nTblCol++] = zCopy; + } + else if( 0==sqlite3_stricmp("rbu_rowid", zName) ){ + bRbuRowid = 1; + } + } + sqlite3_finalize(pStmt); + pStmt = 0; + + if( p->rc==SQLITE_OK + && rbuIsVacuum(p)==0 + && bRbuRowid!=(pIter->eType==RBU_PK_VTAB || pIter->eType==RBU_PK_NONE) + ){ + p->rc = SQLITE_ERROR; + p->zErrmsg = sqlite3_mprintf( + "table %q %s rbu_rowid column", pIter->zDataTbl, + (bRbuRowid ? "may not have" : "requires") + ); + } + + /* Check that all non-HIDDEN columns in the destination table are also + ** present in the input table. Populate the abTblPk[], azTblType[] and + ** aiTblOrder[] arrays at the same time. */ + if( p->rc==SQLITE_OK ){ + p->rc = prepareFreeAndCollectError(p->dbMain, &pStmt, &p->zErrmsg, + sqlite3_mprintf("PRAGMA table_info(%Q)", pIter->zTbl) + ); + } + while( p->rc==SQLITE_OK && SQLITE_ROW==sqlite3_step(pStmt) ){ + const char *zName = (const char*)sqlite3_column_text(pStmt, 1); + if( zName==0 ) break; /* An OOM - finalize() below returns S_NOMEM */ + for(i=iOrder; inTblCol; i++){ + if( 0==strcmp(zName, pIter->azTblCol[i]) ) break; + } + if( i==pIter->nTblCol ){ + p->rc = SQLITE_ERROR; + p->zErrmsg = sqlite3_mprintf("column missing from %q: %s", + pIter->zDataTbl, zName + ); + }else{ + int iPk = sqlite3_column_int(pStmt, 5); + int bNotNull = sqlite3_column_int(pStmt, 3); + const char *zType = (const char*)sqlite3_column_text(pStmt, 2); + + if( i!=iOrder ){ + SWAP(int, pIter->aiSrcOrder[i], pIter->aiSrcOrder[iOrder]); + SWAP(char*, pIter->azTblCol[i], pIter->azTblCol[iOrder]); + } + + pIter->azTblType[iOrder] = rbuStrndup(zType, &p->rc); + assert( iPk>=0 ); + pIter->abTblPk[iOrder] = (u8)iPk; + pIter->abNotNull[iOrder] = (u8)bNotNull || (iPk!=0); + iOrder++; + } + } + + rbuFinalize(p, pStmt); + rbuObjIterCacheIndexedCols(p, pIter); + assert( pIter->eType!=RBU_PK_VTAB || pIter->abIndexed==0 ); + assert( pIter->eType!=RBU_PK_VTAB || pIter->nIndex==0 ); + } + + return p->rc; +} + +/* +** This function constructs and returns a pointer to a nul-terminated +** string containing some SQL clause or list based on one or more of the +** column names currently stored in the pIter->azTblCol[] array. +*/ +static char *rbuObjIterGetCollist( + sqlite3rbu *p, /* RBU object */ + RbuObjIter *pIter /* Object iterator for column names */ +){ + char *zList = 0; + const char *zSep = ""; + int i; + for(i=0; inTblCol; i++){ + const char *z = pIter->azTblCol[i]; + zList = rbuMPrintf(p, "%z%s\"%w\"", zList, zSep, z); + zSep = ", "; + } + return zList; +} + +/* +** Return a comma separated list of the quoted PRIMARY KEY column names, +** in order, for the current table. Before each column name, add the text +** zPre. After each column name, add the zPost text. Use zSeparator as +** the separator text (usually ", "). +*/ +static char *rbuObjIterGetPkList( + sqlite3rbu *p, /* RBU object */ + RbuObjIter *pIter, /* Object iterator for column names */ + const char *zPre, /* Before each quoted column name */ + const char *zSeparator, /* Separator to use between columns */ + const char *zPost /* After each quoted column name */ +){ + int iPk = 1; + char *zRet = 0; + const char *zSep = ""; + while( 1 ){ + int i; + for(i=0; inTblCol; i++){ + if( (int)pIter->abTblPk[i]==iPk ){ + const char *zCol = pIter->azTblCol[i]; + zRet = rbuMPrintf(p, "%z%s%s\"%w\"%s", zRet, zSep, zPre, zCol, zPost); + zSep = zSeparator; + break; + } + } + if( i==pIter->nTblCol ) break; + iPk++; + } + return zRet; +} + +/* +** This function is called as part of restarting an RBU vacuum within +** stage 1 of the process (while the *-oal file is being built) while +** updating a table (not an index). The table may be a rowid table or +** a WITHOUT ROWID table. It queries the target database to find the +** largest key that has already been written to the target table and +** constructs a WHERE clause that can be used to extract the remaining +** rows from the source table. For a rowid table, the WHERE clause +** is of the form: +** +** "WHERE _rowid_ > ?" +** +** and for WITHOUT ROWID tables: +** +** "WHERE (key1, key2) > (?, ?)" +** +** Instead of "?" placeholders, the actual WHERE clauses created by +** this function contain literal SQL values. +*/ +static char *rbuVacuumTableStart( + sqlite3rbu *p, /* RBU handle */ + RbuObjIter *pIter, /* RBU iterator object */ + int bRowid, /* True for a rowid table */ + const char *zWrite /* Target table name prefix */ +){ + sqlite3_stmt *pMax = 0; + char *zRet = 0; + if( bRowid ){ + p->rc = prepareFreeAndCollectError(p->dbMain, &pMax, &p->zErrmsg, + sqlite3_mprintf( + "SELECT max(_rowid_) FROM \"%s%w\"", zWrite, pIter->zTbl + ) + ); + if( p->rc==SQLITE_OK && SQLITE_ROW==sqlite3_step(pMax) ){ + sqlite3_int64 iMax = sqlite3_column_int64(pMax, 0); + zRet = rbuMPrintf(p, " WHERE _rowid_ > %lld ", iMax); + } + rbuFinalize(p, pMax); + }else{ + char *zOrder = rbuObjIterGetPkList(p, pIter, "", ", ", " DESC"); + char *zSelect = rbuObjIterGetPkList(p, pIter, "quote(", "||','||", ")"); + char *zList = rbuObjIterGetPkList(p, pIter, "", ", ", ""); + + if( p->rc==SQLITE_OK ){ + p->rc = prepareFreeAndCollectError(p->dbMain, &pMax, &p->zErrmsg, + sqlite3_mprintf( + "SELECT %s FROM \"%s%w\" ORDER BY %s LIMIT 1", + zSelect, zWrite, pIter->zTbl, zOrder + ) + ); + if( p->rc==SQLITE_OK && SQLITE_ROW==sqlite3_step(pMax) ){ + const char *zVal = (const char*)sqlite3_column_text(pMax, 0); + zRet = rbuMPrintf(p, " WHERE (%s) > (%s) ", zList, zVal); + } + rbuFinalize(p, pMax); + } + + sqlite3_free(zOrder); + sqlite3_free(zSelect); + sqlite3_free(zList); + } + return zRet; +} + +/* +** This function is called as part of restating an RBU vacuum when the +** current operation is writing content to an index. If possible, it +** queries the target index b-tree for the largest key already written to +** it, then composes and returns an expression that can be used in a WHERE +** clause to select the remaining required rows from the source table. +** It is only possible to return such an expression if: +** +** * The index contains no DESC columns, and +** * The last key written to the index before the operation was +** suspended does not contain any NULL values. +** +** The expression is of the form: +** +** (index-field1, index-field2, ...) > (?, ?, ...) +** +** except that the "?" placeholders are replaced with literal values. +** +** If the expression cannot be created, NULL is returned. In this case, +** the caller has to use an OFFSET clause to extract only the required +** rows from the sourct table, just as it does for an RBU update operation. +*/ +char *rbuVacuumIndexStart( + sqlite3rbu *p, /* RBU handle */ + RbuObjIter *pIter /* RBU iterator object */ +){ + char *zOrder = 0; + char *zLhs = 0; + char *zSelect = 0; + char *zVector = 0; + char *zRet = 0; + int bFailed = 0; + const char *zSep = ""; + int iCol = 0; + sqlite3_stmt *pXInfo = 0; + + p->rc = prepareFreeAndCollectError(p->dbMain, &pXInfo, &p->zErrmsg, + sqlite3_mprintf("PRAGMA main.index_xinfo = %Q", pIter->zIdx) + ); + while( p->rc==SQLITE_OK && SQLITE_ROW==sqlite3_step(pXInfo) ){ + int iCid = sqlite3_column_int(pXInfo, 1); + const char *zCollate = (const char*)sqlite3_column_text(pXInfo, 4); + const char *zCol; + if( sqlite3_column_int(pXInfo, 3) ){ + bFailed = 1; + break; + } + + if( iCid<0 ){ + if( pIter->eType==RBU_PK_IPK ){ + int i; + for(i=0; pIter->abTblPk[i]==0; i++); + assert( inTblCol ); + zCol = pIter->azTblCol[i]; + }else{ + zCol = "_rowid_"; + } + }else{ + zCol = pIter->azTblCol[iCid]; + } + + zLhs = rbuMPrintf(p, "%z%s \"%w\" COLLATE %Q", + zLhs, zSep, zCol, zCollate + ); + zOrder = rbuMPrintf(p, "%z%s \"rbu_imp_%d%w\" COLLATE %Q DESC", + zOrder, zSep, iCol, zCol, zCollate + ); + zSelect = rbuMPrintf(p, "%z%s quote(\"rbu_imp_%d%w\")", + zSelect, zSep, iCol, zCol + ); + zSep = ", "; + iCol++; + } + rbuFinalize(p, pXInfo); + if( bFailed ) goto index_start_out; + + if( p->rc==SQLITE_OK ){ + sqlite3_stmt *pSel = 0; + + p->rc = prepareFreeAndCollectError(p->dbMain, &pSel, &p->zErrmsg, + sqlite3_mprintf("SELECT %s FROM \"rbu_imp_%w\" ORDER BY %s LIMIT 1", + zSelect, pIter->zTbl, zOrder + ) + ); + if( p->rc==SQLITE_OK && SQLITE_ROW==sqlite3_step(pSel) ){ + zSep = ""; + for(iCol=0; iColnCol; iCol++){ + const char *zQuoted = (const char*)sqlite3_column_text(pSel, iCol); + if( zQuoted[0]=='N' ){ + bFailed = 1; + break; + } + zVector = rbuMPrintf(p, "%z%s%s", zVector, zSep, zQuoted); + zSep = ", "; + } + + if( !bFailed ){ + zRet = rbuMPrintf(p, "(%s) > (%s)", zLhs, zVector); + } + } + rbuFinalize(p, pSel); + } + + index_start_out: + sqlite3_free(zOrder); + sqlite3_free(zSelect); + sqlite3_free(zVector); + sqlite3_free(zLhs); + return zRet; +} + +/* +** This function is used to create a SELECT list (the list of SQL +** expressions that follows a SELECT keyword) for a SELECT statement +** used to read from an data_xxx or rbu_tmp_xxx table while updating the +** index object currently indicated by the iterator object passed as the +** second argument. A "PRAGMA index_xinfo = " statement is used +** to obtain the required information. +** +** If the index is of the following form: +** +** CREATE INDEX i1 ON t1(c, b COLLATE nocase); +** +** and "t1" is a table with an explicit INTEGER PRIMARY KEY column +** "ipk", the returned string is: +** +** "`c` COLLATE 'BINARY', `b` COLLATE 'NOCASE', `ipk` COLLATE 'BINARY'" +** +** As well as the returned string, three other malloc'd strings are +** returned via output parameters. As follows: +** +** pzImposterCols: ... +** pzImposterPk: ... +** pzWhere: ... +*/ +static char *rbuObjIterGetIndexCols( + sqlite3rbu *p, /* RBU object */ + RbuObjIter *pIter, /* Object iterator for column names */ + char **pzImposterCols, /* OUT: Columns for imposter table */ + char **pzImposterPk, /* OUT: Imposter PK clause */ + char **pzWhere, /* OUT: WHERE clause */ + int *pnBind /* OUT: Trbul number of columns */ +){ + int rc = p->rc; /* Error code */ + int rc2; /* sqlite3_finalize() return code */ + char *zRet = 0; /* String to return */ + char *zImpCols = 0; /* String to return via *pzImposterCols */ + char *zImpPK = 0; /* String to return via *pzImposterPK */ + char *zWhere = 0; /* String to return via *pzWhere */ + int nBind = 0; /* Value to return via *pnBind */ + const char *zCom = ""; /* Set to ", " later on */ + const char *zAnd = ""; /* Set to " AND " later on */ + sqlite3_stmt *pXInfo = 0; /* PRAGMA index_xinfo = ? */ + + if( rc==SQLITE_OK ){ + assert( p->zErrmsg==0 ); + rc = prepareFreeAndCollectError(p->dbMain, &pXInfo, &p->zErrmsg, + sqlite3_mprintf("PRAGMA main.index_xinfo = %Q", pIter->zIdx) + ); + } + + while( rc==SQLITE_OK && SQLITE_ROW==sqlite3_step(pXInfo) ){ + int iCid = sqlite3_column_int(pXInfo, 1); + int bDesc = sqlite3_column_int(pXInfo, 3); + const char *zCollate = (const char*)sqlite3_column_text(pXInfo, 4); + const char *zCol = 0; + const char *zType; + + if( iCid==-2 ){ + int iSeq = sqlite3_column_int(pXInfo, 0); + zRet = sqlite3_mprintf("%z%s(%.*s) COLLATE %Q", zRet, zCom, + pIter->aIdxCol[iSeq].nSpan, pIter->aIdxCol[iSeq].zSpan, zCollate + ); + zType = ""; + }else { + if( iCid<0 ){ + /* An integer primary key. If the table has an explicit IPK, use + ** its name. Otherwise, use "rbu_rowid". */ + if( pIter->eType==RBU_PK_IPK ){ + int i; + for(i=0; pIter->abTblPk[i]==0; i++); + assert( inTblCol ); + zCol = pIter->azTblCol[i]; + }else if( rbuIsVacuum(p) ){ + zCol = "_rowid_"; + }else{ + zCol = "rbu_rowid"; + } + zType = "INTEGER"; + }else{ + zCol = pIter->azTblCol[iCid]; + zType = pIter->azTblType[iCid]; + } + zRet = sqlite3_mprintf("%z%s\"%w\" COLLATE %Q", zRet, zCom,zCol,zCollate); + } + + if( pIter->bUnique==0 || sqlite3_column_int(pXInfo, 5) ){ + const char *zOrder = (bDesc ? " DESC" : ""); + zImpPK = sqlite3_mprintf("%z%s\"rbu_imp_%d%w\"%s", + zImpPK, zCom, nBind, zCol, zOrder + ); + } + zImpCols = sqlite3_mprintf("%z%s\"rbu_imp_%d%w\" %s COLLATE %Q", + zImpCols, zCom, nBind, zCol, zType, zCollate + ); + zWhere = sqlite3_mprintf( + "%z%s\"rbu_imp_%d%w\" IS ?", zWhere, zAnd, nBind, zCol + ); + if( zRet==0 || zImpPK==0 || zImpCols==0 || zWhere==0 ) rc = SQLITE_NOMEM; + zCom = ", "; + zAnd = " AND "; + nBind++; + } + + rc2 = sqlite3_finalize(pXInfo); + if( rc==SQLITE_OK ) rc = rc2; + + if( rc!=SQLITE_OK ){ + sqlite3_free(zRet); + sqlite3_free(zImpCols); + sqlite3_free(zImpPK); + sqlite3_free(zWhere); + zRet = 0; + zImpCols = 0; + zImpPK = 0; + zWhere = 0; + p->rc = rc; + } + + *pzImposterCols = zImpCols; + *pzImposterPk = zImpPK; + *pzWhere = zWhere; + *pnBind = nBind; + return zRet; +} + +/* +** Assuming the current table columns are "a", "b" and "c", and the zObj +** paramter is passed "old", return a string of the form: +** +** "old.a, old.b, old.b" +** +** With the column names escaped. +** +** For tables with implicit rowids - RBU_PK_EXTERNAL and RBU_PK_NONE, append +** the text ", old._rowid_" to the returned value. +*/ +static char *rbuObjIterGetOldlist( + sqlite3rbu *p, + RbuObjIter *pIter, + const char *zObj +){ + char *zList = 0; + if( p->rc==SQLITE_OK && pIter->abIndexed ){ + const char *zS = ""; + int i; + for(i=0; inTblCol; i++){ + if( pIter->abIndexed[i] ){ + const char *zCol = pIter->azTblCol[i]; + zList = sqlite3_mprintf("%z%s%s.\"%w\"", zList, zS, zObj, zCol); + }else{ + zList = sqlite3_mprintf("%z%sNULL", zList, zS); + } + zS = ", "; + if( zList==0 ){ + p->rc = SQLITE_NOMEM; + break; + } + } + + /* For a table with implicit rowids, append "old._rowid_" to the list. */ + if( pIter->eType==RBU_PK_EXTERNAL || pIter->eType==RBU_PK_NONE ){ + zList = rbuMPrintf(p, "%z, %s._rowid_", zList, zObj); + } + } + return zList; +} + +/* +** Return an expression that can be used in a WHERE clause to match the +** primary key of the current table. For example, if the table is: +** +** CREATE TABLE t1(a, b, c, PRIMARY KEY(b, c)); +** +** Return the string: +** +** "b = ?1 AND c = ?2" +*/ +static char *rbuObjIterGetWhere( + sqlite3rbu *p, + RbuObjIter *pIter +){ + char *zList = 0; + if( pIter->eType==RBU_PK_VTAB || pIter->eType==RBU_PK_NONE ){ + zList = rbuMPrintf(p, "_rowid_ = ?%d", pIter->nTblCol+1); + }else if( pIter->eType==RBU_PK_EXTERNAL ){ + const char *zSep = ""; + int i; + for(i=0; inTblCol; i++){ + if( pIter->abTblPk[i] ){ + zList = rbuMPrintf(p, "%z%sc%d=?%d", zList, zSep, i, i+1); + zSep = " AND "; + } + } + zList = rbuMPrintf(p, + "_rowid_ = (SELECT id FROM rbu_imposter2 WHERE %z)", zList + ); + + }else{ + const char *zSep = ""; + int i; + for(i=0; inTblCol; i++){ + if( pIter->abTblPk[i] ){ + const char *zCol = pIter->azTblCol[i]; + zList = rbuMPrintf(p, "%z%s\"%w\"=?%d", zList, zSep, zCol, i+1); + zSep = " AND "; + } + } + } + return zList; +} + +/* +** The SELECT statement iterating through the keys for the current object +** (p->objiter.pSelect) currently points to a valid row. However, there +** is something wrong with the rbu_control value in the rbu_control value +** stored in the (p->nCol+1)'th column. Set the error code and error message +** of the RBU handle to something reflecting this. +*/ +static void rbuBadControlError(sqlite3rbu *p){ + p->rc = SQLITE_ERROR; + p->zErrmsg = sqlite3_mprintf("invalid rbu_control value"); +} + + +/* +** Return a nul-terminated string containing the comma separated list of +** assignments that should be included following the "SET" keyword of +** an UPDATE statement used to update the table object that the iterator +** passed as the second argument currently points to if the rbu_control +** column of the data_xxx table entry is set to zMask. +** +** The memory for the returned string is obtained from sqlite3_malloc(). +** It is the responsibility of the caller to eventually free it using +** sqlite3_free(). +** +** If an OOM error is encountered when allocating space for the new +** string, an error code is left in the rbu handle passed as the first +** argument and NULL is returned. Or, if an error has already occurred +** when this function is called, NULL is returned immediately, without +** attempting the allocation or modifying the stored error code. +*/ +static char *rbuObjIterGetSetlist( + sqlite3rbu *p, + RbuObjIter *pIter, + const char *zMask +){ + char *zList = 0; + if( p->rc==SQLITE_OK ){ + int i; + + if( (int)strlen(zMask)!=pIter->nTblCol ){ + rbuBadControlError(p); + }else{ + const char *zSep = ""; + for(i=0; inTblCol; i++){ + char c = zMask[pIter->aiSrcOrder[i]]; + if( c=='x' ){ + zList = rbuMPrintf(p, "%z%s\"%w\"=?%d", + zList, zSep, pIter->azTblCol[i], i+1 + ); + zSep = ", "; + } + else if( c=='d' ){ + zList = rbuMPrintf(p, "%z%s\"%w\"=rbu_delta(\"%w\", ?%d)", + zList, zSep, pIter->azTblCol[i], pIter->azTblCol[i], i+1 + ); + zSep = ", "; + } + else if( c=='f' ){ + zList = rbuMPrintf(p, "%z%s\"%w\"=rbu_fossil_delta(\"%w\", ?%d)", + zList, zSep, pIter->azTblCol[i], pIter->azTblCol[i], i+1 + ); + zSep = ", "; + } + } + } + } + return zList; +} + +/* +** Return a nul-terminated string consisting of nByte comma separated +** "?" expressions. For example, if nByte is 3, return a pointer to +** a buffer containing the string "?,?,?". +** +** The memory for the returned string is obtained from sqlite3_malloc(). +** It is the responsibility of the caller to eventually free it using +** sqlite3_free(). +** +** If an OOM error is encountered when allocating space for the new +** string, an error code is left in the rbu handle passed as the first +** argument and NULL is returned. Or, if an error has already occurred +** when this function is called, NULL is returned immediately, without +** attempting the allocation or modifying the stored error code. +*/ +static char *rbuObjIterGetBindlist(sqlite3rbu *p, int nBind){ + char *zRet = 0; + sqlite3_int64 nByte = 2*(sqlite3_int64)nBind + 1; + + zRet = (char*)rbuMalloc(p, nByte); + if( zRet ){ + int i; + for(i=0; izIdx==0 ); + if( p->rc==SQLITE_OK ){ + const char *zSep = "PRIMARY KEY("; + sqlite3_stmt *pXList = 0; /* PRAGMA index_list = (pIter->zTbl) */ + sqlite3_stmt *pXInfo = 0; /* PRAGMA index_xinfo = */ + + p->rc = prepareFreeAndCollectError(p->dbMain, &pXList, &p->zErrmsg, + sqlite3_mprintf("PRAGMA main.index_list = %Q", pIter->zTbl) + ); + while( p->rc==SQLITE_OK && SQLITE_ROW==sqlite3_step(pXList) ){ + const char *zOrig = (const char*)sqlite3_column_text(pXList,3); + if( zOrig && strcmp(zOrig, "pk")==0 ){ + const char *zIdx = (const char*)sqlite3_column_text(pXList,1); + if( zIdx ){ + p->rc = prepareFreeAndCollectError(p->dbMain, &pXInfo, &p->zErrmsg, + sqlite3_mprintf("PRAGMA main.index_xinfo = %Q", zIdx) + ); + } + break; + } + } + rbuFinalize(p, pXList); + + while( p->rc==SQLITE_OK && SQLITE_ROW==sqlite3_step(pXInfo) ){ + if( sqlite3_column_int(pXInfo, 5) ){ + /* int iCid = sqlite3_column_int(pXInfo, 0); */ + const char *zCol = (const char*)sqlite3_column_text(pXInfo, 2); + const char *zDesc = sqlite3_column_int(pXInfo, 3) ? " DESC" : ""; + z = rbuMPrintf(p, "%z%s\"%w\"%s", z, zSep, zCol, zDesc); + zSep = ", "; + } + } + z = rbuMPrintf(p, "%z)", z); + rbuFinalize(p, pXInfo); + } + return z; +} + +/* +** This function creates the second imposter table used when writing to +** a table b-tree where the table has an external primary key. If the +** iterator passed as the second argument does not currently point to +** a table (not index) with an external primary key, this function is a +** no-op. +** +** Assuming the iterator does point to a table with an external PK, this +** function creates a WITHOUT ROWID imposter table named "rbu_imposter2" +** used to access that PK index. For example, if the target table is +** declared as follows: +** +** CREATE TABLE t1(a, b TEXT, c REAL, PRIMARY KEY(b, c)); +** +** then the imposter table schema is: +** +** CREATE TABLE rbu_imposter2(c1 TEXT, c2 REAL, id INTEGER) WITHOUT ROWID; +** +*/ +static void rbuCreateImposterTable2(sqlite3rbu *p, RbuObjIter *pIter){ + if( p->rc==SQLITE_OK && pIter->eType==RBU_PK_EXTERNAL ){ + int tnum = pIter->iPkTnum; /* Root page of PK index */ + sqlite3_stmt *pQuery = 0; /* SELECT name ... WHERE rootpage = $tnum */ + const char *zIdx = 0; /* Name of PK index */ + sqlite3_stmt *pXInfo = 0; /* PRAGMA main.index_xinfo = $zIdx */ + const char *zComma = ""; + char *zCols = 0; /* Used to build up list of table cols */ + char *zPk = 0; /* Used to build up table PK declaration */ + + /* Figure out the name of the primary key index for the current table. + ** This is needed for the argument to "PRAGMA index_xinfo". Set + ** zIdx to point to a nul-terminated string containing this name. */ + p->rc = prepareAndCollectError(p->dbMain, &pQuery, &p->zErrmsg, + "SELECT name FROM sqlite_master WHERE rootpage = ?" + ); + if( p->rc==SQLITE_OK ){ + sqlite3_bind_int(pQuery, 1, tnum); + if( SQLITE_ROW==sqlite3_step(pQuery) ){ + zIdx = (const char*)sqlite3_column_text(pQuery, 0); + } + } + if( zIdx ){ + p->rc = prepareFreeAndCollectError(p->dbMain, &pXInfo, &p->zErrmsg, + sqlite3_mprintf("PRAGMA main.index_xinfo = %Q", zIdx) + ); + } + rbuFinalize(p, pQuery); + + while( p->rc==SQLITE_OK && SQLITE_ROW==sqlite3_step(pXInfo) ){ + int bKey = sqlite3_column_int(pXInfo, 5); + if( bKey ){ + int iCid = sqlite3_column_int(pXInfo, 1); + int bDesc = sqlite3_column_int(pXInfo, 3); + const char *zCollate = (const char*)sqlite3_column_text(pXInfo, 4); + zCols = rbuMPrintf(p, "%z%sc%d %s COLLATE %Q", zCols, zComma, + iCid, pIter->azTblType[iCid], zCollate + ); + zPk = rbuMPrintf(p, "%z%sc%d%s", zPk, zComma, iCid, bDesc?" DESC":""); + zComma = ", "; + } + } + zCols = rbuMPrintf(p, "%z, id INTEGER", zCols); + rbuFinalize(p, pXInfo); + + sqlite3_test_control(SQLITE_TESTCTRL_IMPOSTER, p->dbMain, "main", 1, tnum); + rbuMPrintfExec(p, p->dbMain, + "CREATE TABLE rbu_imposter2(%z, PRIMARY KEY(%z)) WITHOUT ROWID", + zCols, zPk + ); + sqlite3_test_control(SQLITE_TESTCTRL_IMPOSTER, p->dbMain, "main", 0, 0); + } +} + +/* +** If an error has already occurred when this function is called, it +** immediately returns zero (without doing any work). Or, if an error +** occurs during the execution of this function, it sets the error code +** in the sqlite3rbu object indicated by the first argument and returns +** zero. +** +** The iterator passed as the second argument is guaranteed to point to +** a table (not an index) when this function is called. This function +** attempts to create any imposter table required to write to the main +** table b-tree of the table before returning. Non-zero is returned if +** an imposter table are created, or zero otherwise. +** +** An imposter table is required in all cases except RBU_PK_VTAB. Only +** virtual tables are written to directly. The imposter table has the +** same schema as the actual target table (less any UNIQUE constraints). +** More precisely, the "same schema" means the same columns, types, +** collation sequences. For tables that do not have an external PRIMARY +** KEY, it also means the same PRIMARY KEY declaration. +*/ +static void rbuCreateImposterTable(sqlite3rbu *p, RbuObjIter *pIter){ + if( p->rc==SQLITE_OK && pIter->eType!=RBU_PK_VTAB ){ + int tnum = pIter->iTnum; + const char *zComma = ""; + char *zSql = 0; + int iCol; + sqlite3_test_control(SQLITE_TESTCTRL_IMPOSTER, p->dbMain, "main", 0, 1); + + for(iCol=0; p->rc==SQLITE_OK && iColnTblCol; iCol++){ + const char *zPk = ""; + const char *zCol = pIter->azTblCol[iCol]; + const char *zColl = 0; + + p->rc = sqlite3_table_column_metadata( + p->dbMain, "main", pIter->zTbl, zCol, 0, &zColl, 0, 0, 0 + ); + + if( pIter->eType==RBU_PK_IPK && pIter->abTblPk[iCol] ){ + /* If the target table column is an "INTEGER PRIMARY KEY", add + ** "PRIMARY KEY" to the imposter table column declaration. */ + zPk = "PRIMARY KEY "; + } + zSql = rbuMPrintf(p, "%z%s\"%w\" %s %sCOLLATE %Q%s", + zSql, zComma, zCol, pIter->azTblType[iCol], zPk, zColl, + (pIter->abNotNull[iCol] ? " NOT NULL" : "") + ); + zComma = ", "; + } + + if( pIter->eType==RBU_PK_WITHOUT_ROWID ){ + char *zPk = rbuWithoutRowidPK(p, pIter); + if( zPk ){ + zSql = rbuMPrintf(p, "%z, %z", zSql, zPk); + } + } + + sqlite3_test_control(SQLITE_TESTCTRL_IMPOSTER, p->dbMain, "main", 1, tnum); + rbuMPrintfExec(p, p->dbMain, "CREATE TABLE \"rbu_imp_%w\"(%z)%s", + pIter->zTbl, zSql, + (pIter->eType==RBU_PK_WITHOUT_ROWID ? " WITHOUT ROWID" : "") + ); + sqlite3_test_control(SQLITE_TESTCTRL_IMPOSTER, p->dbMain, "main", 0, 0); + } +} + +/* +** Prepare a statement used to insert rows into the "rbu_tmp_xxx" table. +** Specifically a statement of the form: +** +** INSERT INTO rbu_tmp_xxx VALUES(?, ?, ? ...); +** +** The number of bound variables is equal to the number of columns in +** the target table, plus one (for the rbu_control column), plus one more +** (for the rbu_rowid column) if the target table is an implicit IPK or +** virtual table. +*/ +static void rbuObjIterPrepareTmpInsert( + sqlite3rbu *p, + RbuObjIter *pIter, + const char *zCollist, + const char *zRbuRowid +){ + int bRbuRowid = (pIter->eType==RBU_PK_EXTERNAL || pIter->eType==RBU_PK_NONE); + char *zBind = rbuObjIterGetBindlist(p, pIter->nTblCol + 1 + bRbuRowid); + if( zBind ){ + assert( pIter->pTmpInsert==0 ); + p->rc = prepareFreeAndCollectError( + p->dbRbu, &pIter->pTmpInsert, &p->zErrmsg, sqlite3_mprintf( + "INSERT INTO %s.'rbu_tmp_%q'(rbu_control,%s%s) VALUES(%z)", + p->zStateDb, pIter->zDataTbl, zCollist, zRbuRowid, zBind + )); + } +} + +static void rbuTmpInsertFunc( + sqlite3_context *pCtx, + int nVal, + sqlite3_value **apVal +){ + sqlite3rbu *p = sqlite3_user_data(pCtx); + int rc = SQLITE_OK; + int i; + + assert( sqlite3_value_int(apVal[0])!=0 + || p->objiter.eType==RBU_PK_EXTERNAL + || p->objiter.eType==RBU_PK_NONE + ); + if( sqlite3_value_int(apVal[0])!=0 ){ + p->nPhaseOneStep += p->objiter.nIndex; + } + + for(i=0; rc==SQLITE_OK && iobjiter.pTmpInsert, i+1, apVal[i]); + } + if( rc==SQLITE_OK ){ + sqlite3_step(p->objiter.pTmpInsert); + rc = sqlite3_reset(p->objiter.pTmpInsert); + } + + if( rc!=SQLITE_OK ){ + sqlite3_result_error_code(pCtx, rc); + } +} + +static char *rbuObjIterGetIndexWhere(sqlite3rbu *p, RbuObjIter *pIter){ + sqlite3_stmt *pStmt = 0; + int rc = p->rc; + char *zRet = 0; + + assert( pIter->zIdxSql==0 && pIter->nIdxCol==0 && pIter->aIdxCol==0 ); + + if( rc==SQLITE_OK ){ + rc = prepareAndCollectError(p->dbMain, &pStmt, &p->zErrmsg, + "SELECT trim(sql) FROM sqlite_master WHERE type='index' AND name=?" + ); + } + if( rc==SQLITE_OK ){ + int rc2; + rc = sqlite3_bind_text(pStmt, 1, pIter->zIdx, -1, SQLITE_STATIC); + if( rc==SQLITE_OK && SQLITE_ROW==sqlite3_step(pStmt) ){ + char *zSql = (char*)sqlite3_column_text(pStmt, 0); + if( zSql ){ + pIter->zIdxSql = zSql = rbuStrndup(zSql, &rc); + } + if( zSql ){ + int nParen = 0; /* Number of open parenthesis */ + int i; + int iIdxCol = 0; + int nIdxAlloc = 0; + for(i=0; zSql[i]; i++){ + char c = zSql[i]; + + /* If necessary, grow the pIter->aIdxCol[] array */ + if( iIdxCol==nIdxAlloc ){ + RbuSpan *aIdxCol = (RbuSpan*)sqlite3_realloc( + pIter->aIdxCol, (nIdxAlloc+16)*sizeof(RbuSpan) + ); + if( aIdxCol==0 ){ + rc = SQLITE_NOMEM; + break; + } + pIter->aIdxCol = aIdxCol; + nIdxAlloc += 16; + } + + if( c=='(' ){ + if( nParen==0 ){ + assert( iIdxCol==0 ); + pIter->aIdxCol[0].zSpan = &zSql[i+1]; + } + nParen++; + } + else if( c==')' ){ + nParen--; + if( nParen==0 ){ + int nSpan = &zSql[i] - pIter->aIdxCol[iIdxCol].zSpan; + pIter->aIdxCol[iIdxCol++].nSpan = nSpan; + i++; + break; + } + }else if( c==',' && nParen==1 ){ + int nSpan = &zSql[i] - pIter->aIdxCol[iIdxCol].zSpan; + pIter->aIdxCol[iIdxCol++].nSpan = nSpan; + pIter->aIdxCol[iIdxCol].zSpan = &zSql[i+1]; + }else if( c=='"' || c=='\'' || c=='`' ){ + for(i++; 1; i++){ + if( zSql[i]==c ){ + if( zSql[i+1]!=c ) break; + i++; + } + } + }else if( c=='[' ){ + for(i++; 1; i++){ + if( zSql[i]==']' ) break; + } + }else if( c=='-' && zSql[i+1]=='-' ){ + for(i=i+2; zSql[i] && zSql[i]!='\n'; i++); + if( zSql[i]=='\0' ) break; + }else if( c=='/' && zSql[i+1]=='*' ){ + for(i=i+2; zSql[i] && (zSql[i]!='*' || zSql[i+1]!='/'); i++); + if( zSql[i]=='\0' ) break; + i++; + } + } + if( zSql[i] ){ + zRet = rbuStrndup(&zSql[i], &rc); + } + pIter->nIdxCol = iIdxCol; + } + } + + rc2 = sqlite3_finalize(pStmt); + if( rc==SQLITE_OK ) rc = rc2; + } + + p->rc = rc; + return zRet; +} + +/* +** Ensure that the SQLite statement handles required to update the +** target database object currently indicated by the iterator passed +** as the second argument are available. +*/ +static int rbuObjIterPrepareAll( + sqlite3rbu *p, + RbuObjIter *pIter, + int nOffset /* Add "LIMIT -1 OFFSET $nOffset" to SELECT */ +){ + assert( pIter->bCleanup==0 ); + if( pIter->pSelect==0 && rbuObjIterCacheTableInfo(p, pIter)==SQLITE_OK ){ + const int tnum = pIter->iTnum; + char *zCollist = 0; /* List of indexed columns */ + char **pz = &p->zErrmsg; + const char *zIdx = pIter->zIdx; + char *zLimit = 0; + + if( nOffset ){ + zLimit = sqlite3_mprintf(" LIMIT -1 OFFSET %d", nOffset); + if( !zLimit ) p->rc = SQLITE_NOMEM; + } + + if( zIdx ){ + const char *zTbl = pIter->zTbl; + char *zImposterCols = 0; /* Columns for imposter table */ + char *zImposterPK = 0; /* Primary key declaration for imposter */ + char *zWhere = 0; /* WHERE clause on PK columns */ + char *zBind = 0; + char *zPart = 0; + int nBind = 0; + + assert( pIter->eType!=RBU_PK_VTAB ); + zPart = rbuObjIterGetIndexWhere(p, pIter); + zCollist = rbuObjIterGetIndexCols( + p, pIter, &zImposterCols, &zImposterPK, &zWhere, &nBind + ); + zBind = rbuObjIterGetBindlist(p, nBind); + + /* Create the imposter table used to write to this index. */ + sqlite3_test_control(SQLITE_TESTCTRL_IMPOSTER, p->dbMain, "main", 0, 1); + sqlite3_test_control(SQLITE_TESTCTRL_IMPOSTER, p->dbMain, "main", 1,tnum); + rbuMPrintfExec(p, p->dbMain, + "CREATE TABLE \"rbu_imp_%w\"( %s, PRIMARY KEY( %s ) ) WITHOUT ROWID", + zTbl, zImposterCols, zImposterPK + ); + sqlite3_test_control(SQLITE_TESTCTRL_IMPOSTER, p->dbMain, "main", 0, 0); + + /* Create the statement to insert index entries */ + pIter->nCol = nBind; + if( p->rc==SQLITE_OK ){ + p->rc = prepareFreeAndCollectError( + p->dbMain, &pIter->pInsert, &p->zErrmsg, + sqlite3_mprintf("INSERT INTO \"rbu_imp_%w\" VALUES(%s)", zTbl, zBind) + ); + } + + /* And to delete index entries */ + if( rbuIsVacuum(p)==0 && p->rc==SQLITE_OK ){ + p->rc = prepareFreeAndCollectError( + p->dbMain, &pIter->pDelete, &p->zErrmsg, + sqlite3_mprintf("DELETE FROM \"rbu_imp_%w\" WHERE %s", zTbl, zWhere) + ); + } + + /* Create the SELECT statement to read keys in sorted order */ + if( p->rc==SQLITE_OK ){ + char *zSql; + if( rbuIsVacuum(p) ){ + char *zStart = 0; + if( nOffset ){ + zStart = rbuVacuumIndexStart(p, pIter); + if( zStart ){ + sqlite3_free(zLimit); + zLimit = 0; + } + } + + zSql = sqlite3_mprintf( + "SELECT %s, 0 AS rbu_control FROM '%q' %s %s %s ORDER BY %s%s", + zCollist, + pIter->zDataTbl, + zPart, + (zStart ? (zPart ? "AND" : "WHERE") : ""), zStart, + zCollist, zLimit + ); + sqlite3_free(zStart); + }else + + if( pIter->eType==RBU_PK_EXTERNAL || pIter->eType==RBU_PK_NONE ){ + zSql = sqlite3_mprintf( + "SELECT %s, rbu_control FROM %s.'rbu_tmp_%q' %s ORDER BY %s%s", + zCollist, p->zStateDb, pIter->zDataTbl, + zPart, zCollist, zLimit + ); + }else{ + zSql = sqlite3_mprintf( + "SELECT %s, rbu_control FROM %s.'rbu_tmp_%q' %s " + "UNION ALL " + "SELECT %s, rbu_control FROM '%q' " + "%s %s typeof(rbu_control)='integer' AND rbu_control!=1 " + "ORDER BY %s%s", + zCollist, p->zStateDb, pIter->zDataTbl, zPart, + zCollist, pIter->zDataTbl, + zPart, + (zPart ? "AND" : "WHERE"), + zCollist, zLimit + ); + } + if( p->rc==SQLITE_OK ){ + p->rc = prepareFreeAndCollectError(p->dbRbu,&pIter->pSelect,pz,zSql); + }else{ + sqlite3_free(zSql); + } + } + + sqlite3_free(zImposterCols); + sqlite3_free(zImposterPK); + sqlite3_free(zWhere); + sqlite3_free(zBind); + sqlite3_free(zPart); + }else{ + int bRbuRowid = (pIter->eType==RBU_PK_VTAB) + ||(pIter->eType==RBU_PK_NONE) + ||(pIter->eType==RBU_PK_EXTERNAL && rbuIsVacuum(p)); + const char *zTbl = pIter->zTbl; /* Table this step applies to */ + const char *zWrite; /* Imposter table name */ + + char *zBindings = rbuObjIterGetBindlist(p, pIter->nTblCol + bRbuRowid); + char *zWhere = rbuObjIterGetWhere(p, pIter); + char *zOldlist = rbuObjIterGetOldlist(p, pIter, "old"); + char *zNewlist = rbuObjIterGetOldlist(p, pIter, "new"); + + zCollist = rbuObjIterGetCollist(p, pIter); + pIter->nCol = pIter->nTblCol; + + /* Create the imposter table or tables (if required). */ + rbuCreateImposterTable(p, pIter); + rbuCreateImposterTable2(p, pIter); + zWrite = (pIter->eType==RBU_PK_VTAB ? "" : "rbu_imp_"); + + /* Create the INSERT statement to write to the target PK b-tree */ + if( p->rc==SQLITE_OK ){ + p->rc = prepareFreeAndCollectError(p->dbMain, &pIter->pInsert, pz, + sqlite3_mprintf( + "INSERT INTO \"%s%w\"(%s%s) VALUES(%s)", + zWrite, zTbl, zCollist, (bRbuRowid ? ", _rowid_" : ""), zBindings + ) + ); + } + + /* Create the DELETE statement to write to the target PK b-tree. + ** Because it only performs INSERT operations, this is not required for + ** an rbu vacuum handle. */ + if( rbuIsVacuum(p)==0 && p->rc==SQLITE_OK ){ + p->rc = prepareFreeAndCollectError(p->dbMain, &pIter->pDelete, pz, + sqlite3_mprintf( + "DELETE FROM \"%s%w\" WHERE %s", zWrite, zTbl, zWhere + ) + ); + } + + if( rbuIsVacuum(p)==0 && pIter->abIndexed ){ + const char *zRbuRowid = ""; + if( pIter->eType==RBU_PK_EXTERNAL || pIter->eType==RBU_PK_NONE ){ + zRbuRowid = ", rbu_rowid"; + } + + /* Create the rbu_tmp_xxx table and the triggers to populate it. */ + rbuMPrintfExec(p, p->dbRbu, + "CREATE TABLE IF NOT EXISTS %s.'rbu_tmp_%q' AS " + "SELECT *%s FROM '%q' WHERE 0;" + , p->zStateDb, pIter->zDataTbl + , (pIter->eType==RBU_PK_EXTERNAL ? ", 0 AS rbu_rowid" : "") + , pIter->zDataTbl + ); + + rbuMPrintfExec(p, p->dbMain, + "CREATE TEMP TRIGGER rbu_delete_tr BEFORE DELETE ON \"%s%w\" " + "BEGIN " + " SELECT rbu_tmp_insert(3, %s);" + "END;" + + "CREATE TEMP TRIGGER rbu_update1_tr BEFORE UPDATE ON \"%s%w\" " + "BEGIN " + " SELECT rbu_tmp_insert(3, %s);" + "END;" + + "CREATE TEMP TRIGGER rbu_update2_tr AFTER UPDATE ON \"%s%w\" " + "BEGIN " + " SELECT rbu_tmp_insert(4, %s);" + "END;", + zWrite, zTbl, zOldlist, + zWrite, zTbl, zOldlist, + zWrite, zTbl, zNewlist + ); + + if( pIter->eType==RBU_PK_EXTERNAL || pIter->eType==RBU_PK_NONE ){ + rbuMPrintfExec(p, p->dbMain, + "CREATE TEMP TRIGGER rbu_insert_tr AFTER INSERT ON \"%s%w\" " + "BEGIN " + " SELECT rbu_tmp_insert(0, %s);" + "END;", + zWrite, zTbl, zNewlist + ); + } + + rbuObjIterPrepareTmpInsert(p, pIter, zCollist, zRbuRowid); + } + + /* Create the SELECT statement to read keys from data_xxx */ + if( p->rc==SQLITE_OK ){ + const char *zRbuRowid = ""; + char *zStart = 0; + char *zOrder = 0; + if( bRbuRowid ){ + zRbuRowid = rbuIsVacuum(p) ? ",_rowid_ " : ",rbu_rowid"; + } + + if( rbuIsVacuum(p) ){ + if( nOffset ){ + zStart = rbuVacuumTableStart(p, pIter, bRbuRowid, zWrite); + if( zStart ){ + sqlite3_free(zLimit); + zLimit = 0; + } + } + if( bRbuRowid ){ + zOrder = rbuMPrintf(p, "_rowid_"); + }else{ + zOrder = rbuObjIterGetPkList(p, pIter, "", ", ", ""); + } + } + + if( p->rc==SQLITE_OK ){ + p->rc = prepareFreeAndCollectError(p->dbRbu, &pIter->pSelect, pz, + sqlite3_mprintf( + "SELECT %s,%s rbu_control%s FROM '%q'%s %s %s %s", + zCollist, + (rbuIsVacuum(p) ? "0 AS " : ""), + zRbuRowid, + pIter->zDataTbl, (zStart ? zStart : ""), + (zOrder ? "ORDER BY" : ""), zOrder, + zLimit + ) + ); + } + sqlite3_free(zStart); + sqlite3_free(zOrder); + } + + sqlite3_free(zWhere); + sqlite3_free(zOldlist); + sqlite3_free(zNewlist); + sqlite3_free(zBindings); + } + sqlite3_free(zCollist); + sqlite3_free(zLimit); + } + + return p->rc; +} + +/* +** Set output variable *ppStmt to point to an UPDATE statement that may +** be used to update the imposter table for the main table b-tree of the +** table object that pIter currently points to, assuming that the +** rbu_control column of the data_xyz table contains zMask. +** +** If the zMask string does not specify any columns to update, then this +** is not an error. Output variable *ppStmt is set to NULL in this case. +*/ +static int rbuGetUpdateStmt( + sqlite3rbu *p, /* RBU handle */ + RbuObjIter *pIter, /* Object iterator */ + const char *zMask, /* rbu_control value ('x.x.') */ + sqlite3_stmt **ppStmt /* OUT: UPDATE statement handle */ +){ + RbuUpdateStmt **pp; + RbuUpdateStmt *pUp = 0; + int nUp = 0; + + /* In case an error occurs */ + *ppStmt = 0; + + /* Search for an existing statement. If one is found, shift it to the front + ** of the LRU queue and return immediately. Otherwise, leave nUp pointing + ** to the number of statements currently in the cache and pUp to the + ** last object in the list. */ + for(pp=&pIter->pRbuUpdate; *pp; pp=&((*pp)->pNext)){ + pUp = *pp; + if( strcmp(pUp->zMask, zMask)==0 ){ + *pp = pUp->pNext; + pUp->pNext = pIter->pRbuUpdate; + pIter->pRbuUpdate = pUp; + *ppStmt = pUp->pUpdate; + return SQLITE_OK; + } + nUp++; + } + assert( pUp==0 || pUp->pNext==0 ); + + if( nUp>=SQLITE_RBU_UPDATE_CACHESIZE ){ + for(pp=&pIter->pRbuUpdate; *pp!=pUp; pp=&((*pp)->pNext)); + *pp = 0; + sqlite3_finalize(pUp->pUpdate); + pUp->pUpdate = 0; + }else{ + pUp = (RbuUpdateStmt*)rbuMalloc(p, sizeof(RbuUpdateStmt)+pIter->nTblCol+1); + } + + if( pUp ){ + char *zWhere = rbuObjIterGetWhere(p, pIter); + char *zSet = rbuObjIterGetSetlist(p, pIter, zMask); + char *zUpdate = 0; + + pUp->zMask = (char*)&pUp[1]; + memcpy(pUp->zMask, zMask, pIter->nTblCol); + pUp->pNext = pIter->pRbuUpdate; + pIter->pRbuUpdate = pUp; + + if( zSet ){ + const char *zPrefix = ""; + + if( pIter->eType!=RBU_PK_VTAB ) zPrefix = "rbu_imp_"; + zUpdate = sqlite3_mprintf("UPDATE \"%s%w\" SET %s WHERE %s", + zPrefix, pIter->zTbl, zSet, zWhere + ); + p->rc = prepareFreeAndCollectError( + p->dbMain, &pUp->pUpdate, &p->zErrmsg, zUpdate + ); + *ppStmt = pUp->pUpdate; + } + sqlite3_free(zWhere); + sqlite3_free(zSet); + } + + return p->rc; +} + +static sqlite3 *rbuOpenDbhandle( + sqlite3rbu *p, + const char *zName, + int bUseVfs +){ + sqlite3 *db = 0; + if( p->rc==SQLITE_OK ){ + const int flags = SQLITE_OPEN_READWRITE|SQLITE_OPEN_CREATE|SQLITE_OPEN_URI; + p->rc = sqlite3_open_v2(zName, &db, flags, bUseVfs ? p->zVfsName : 0); + if( p->rc ){ + p->zErrmsg = sqlite3_mprintf("%s", sqlite3_errmsg(db)); + sqlite3_close(db); + db = 0; + } + } + return db; +} + +/* +** Free an RbuState object allocated by rbuLoadState(). +*/ +static void rbuFreeState(RbuState *p){ + if( p ){ + sqlite3_free(p->zTbl); + sqlite3_free(p->zDataTbl); + sqlite3_free(p->zIdx); + sqlite3_free(p); + } +} + +/* +** Allocate an RbuState object and load the contents of the rbu_state +** table into it. Return a pointer to the new object. It is the +** responsibility of the caller to eventually free the object using +** sqlite3_free(). +** +** If an error occurs, leave an error code and message in the rbu handle +** and return NULL. +*/ +static RbuState *rbuLoadState(sqlite3rbu *p){ + RbuState *pRet = 0; + sqlite3_stmt *pStmt = 0; + int rc; + int rc2; + + pRet = (RbuState*)rbuMalloc(p, sizeof(RbuState)); + if( pRet==0 ) return 0; + + rc = prepareFreeAndCollectError(p->dbRbu, &pStmt, &p->zErrmsg, + sqlite3_mprintf("SELECT k, v FROM %s.rbu_state", p->zStateDb) + ); + while( rc==SQLITE_OK && SQLITE_ROW==sqlite3_step(pStmt) ){ + switch( sqlite3_column_int(pStmt, 0) ){ + case RBU_STATE_STAGE: + pRet->eStage = sqlite3_column_int(pStmt, 1); + if( pRet->eStage!=RBU_STAGE_OAL + && pRet->eStage!=RBU_STAGE_MOVE + && pRet->eStage!=RBU_STAGE_CKPT + ){ + p->rc = SQLITE_CORRUPT; + } + break; + + case RBU_STATE_TBL: + pRet->zTbl = rbuStrndup((char*)sqlite3_column_text(pStmt, 1), &rc); + break; + + case RBU_STATE_IDX: + pRet->zIdx = rbuStrndup((char*)sqlite3_column_text(pStmt, 1), &rc); + break; + + case RBU_STATE_ROW: + pRet->nRow = sqlite3_column_int(pStmt, 1); + break; + + case RBU_STATE_PROGRESS: + pRet->nProgress = sqlite3_column_int64(pStmt, 1); + break; + + case RBU_STATE_CKPT: + pRet->iWalCksum = sqlite3_column_int64(pStmt, 1); + break; + + case RBU_STATE_COOKIE: + pRet->iCookie = (u32)sqlite3_column_int64(pStmt, 1); + break; + + case RBU_STATE_OALSZ: + pRet->iOalSz = (u32)sqlite3_column_int64(pStmt, 1); + break; + + case RBU_STATE_PHASEONESTEP: + pRet->nPhaseOneStep = sqlite3_column_int64(pStmt, 1); + break; + + case RBU_STATE_DATATBL: + pRet->zDataTbl = rbuStrndup((char*)sqlite3_column_text(pStmt, 1), &rc); + break; + + default: + rc = SQLITE_CORRUPT; + break; + } + } + rc2 = sqlite3_finalize(pStmt); + if( rc==SQLITE_OK ) rc = rc2; + + p->rc = rc; + return pRet; +} + + +/* +** Open the database handle and attach the RBU database as "rbu". If an +** error occurs, leave an error code and message in the RBU handle. +*/ +static void rbuOpenDatabase(sqlite3rbu *p, int *pbRetry){ + assert( p->rc || (p->dbMain==0 && p->dbRbu==0) ); + assert( p->rc || rbuIsVacuum(p) || p->zTarget!=0 ); + + /* Open the RBU database */ + p->dbRbu = rbuOpenDbhandle(p, p->zRbu, 1); + + if( p->rc==SQLITE_OK && rbuIsVacuum(p) ){ + sqlite3_file_control(p->dbRbu, "main", SQLITE_FCNTL_RBUCNT, (void*)p); + if( p->zState==0 ){ + const char *zFile = sqlite3_db_filename(p->dbRbu, "main"); + p->zState = rbuMPrintf(p, "file://%s-vacuum?modeof=%s", zFile, zFile); + } + } + + /* If using separate RBU and state databases, attach the state database to + ** the RBU db handle now. */ + if( p->zState ){ + rbuMPrintfExec(p, p->dbRbu, "ATTACH %Q AS stat", p->zState); + memcpy(p->zStateDb, "stat", 4); + }else{ + memcpy(p->zStateDb, "main", 4); + } + +#if 0 + if( p->rc==SQLITE_OK && rbuIsVacuum(p) ){ + p->rc = sqlite3_exec(p->dbRbu, "BEGIN", 0, 0, 0); + } +#endif + + /* If it has not already been created, create the rbu_state table */ + rbuMPrintfExec(p, p->dbRbu, RBU_CREATE_STATE, p->zStateDb); + +#if 0 + if( rbuIsVacuum(p) ){ + if( p->rc==SQLITE_OK ){ + int rc2; + int bOk = 0; + sqlite3_stmt *pCnt = 0; + p->rc = prepareAndCollectError(p->dbRbu, &pCnt, &p->zErrmsg, + "SELECT count(*) FROM stat.sqlite_master" + ); + if( p->rc==SQLITE_OK + && sqlite3_step(pCnt)==SQLITE_ROW + && 1==sqlite3_column_int(pCnt, 0) + ){ + bOk = 1; + } + rc2 = sqlite3_finalize(pCnt); + if( p->rc==SQLITE_OK ) p->rc = rc2; + + if( p->rc==SQLITE_OK && bOk==0 ){ + p->rc = SQLITE_ERROR; + p->zErrmsg = sqlite3_mprintf("invalid state database"); + } + + if( p->rc==SQLITE_OK ){ + p->rc = sqlite3_exec(p->dbRbu, "COMMIT", 0, 0, 0); + } + } + } +#endif + + if( p->rc==SQLITE_OK && rbuIsVacuum(p) ){ + int bOpen = 0; + int rc; + p->nRbu = 0; + p->pRbuFd = 0; + rc = sqlite3_file_control(p->dbRbu, "main", SQLITE_FCNTL_RBUCNT, (void*)p); + if( rc!=SQLITE_NOTFOUND ) p->rc = rc; + if( p->eStage>=RBU_STAGE_MOVE ){ + bOpen = 1; + }else{ + RbuState *pState = rbuLoadState(p); + if( pState ){ + bOpen = (pState->eStage>=RBU_STAGE_MOVE); + rbuFreeState(pState); + } + } + if( bOpen ) p->dbMain = rbuOpenDbhandle(p, p->zRbu, p->nRbu<=1); + } + + p->eStage = 0; + if( p->rc==SQLITE_OK && p->dbMain==0 ){ + if( !rbuIsVacuum(p) ){ + p->dbMain = rbuOpenDbhandle(p, p->zTarget, 1); + }else if( p->pRbuFd->pWalFd ){ + if( pbRetry ){ + p->pRbuFd->bNolock = 0; + sqlite3_close(p->dbRbu); + sqlite3_close(p->dbMain); + p->dbMain = 0; + p->dbRbu = 0; + *pbRetry = 1; + return; + } + p->rc = SQLITE_ERROR; + p->zErrmsg = sqlite3_mprintf("cannot vacuum wal mode database"); + }else{ + char *zTarget; + char *zExtra = 0; + if( strlen(p->zRbu)>=5 && 0==memcmp("file:", p->zRbu, 5) ){ + zExtra = &p->zRbu[5]; + while( *zExtra ){ + if( *zExtra++=='?' ) break; + } + if( *zExtra=='\0' ) zExtra = 0; + } + + zTarget = sqlite3_mprintf("file:%s-vactmp?rbu_memory=1%s%s", + sqlite3_db_filename(p->dbRbu, "main"), + (zExtra==0 ? "" : "&"), (zExtra==0 ? "" : zExtra) + ); + + if( zTarget==0 ){ + p->rc = SQLITE_NOMEM; + return; + } + p->dbMain = rbuOpenDbhandle(p, zTarget, p->nRbu<=1); + sqlite3_free(zTarget); + } + } + + if( p->rc==SQLITE_OK ){ + p->rc = sqlite3_create_function(p->dbMain, + "rbu_tmp_insert", -1, SQLITE_UTF8, (void*)p, rbuTmpInsertFunc, 0, 0 + ); + } + + if( p->rc==SQLITE_OK ){ + p->rc = sqlite3_create_function(p->dbMain, + "rbu_fossil_delta", 2, SQLITE_UTF8, 0, rbuFossilDeltaFunc, 0, 0 + ); + } + + if( p->rc==SQLITE_OK ){ + p->rc = sqlite3_create_function(p->dbRbu, + "rbu_target_name", -1, SQLITE_UTF8, (void*)p, rbuTargetNameFunc, 0, 0 + ); + } + + if( p->rc==SQLITE_OK ){ + p->rc = sqlite3_file_control(p->dbMain, "main", SQLITE_FCNTL_RBU, (void*)p); + } + rbuMPrintfExec(p, p->dbMain, "SELECT * FROM sqlite_master"); + + /* Mark the database file just opened as an RBU target database. If + ** this call returns SQLITE_NOTFOUND, then the RBU vfs is not in use. + ** This is an error. */ + if( p->rc==SQLITE_OK ){ + p->rc = sqlite3_file_control(p->dbMain, "main", SQLITE_FCNTL_RBU, (void*)p); + } + + if( p->rc==SQLITE_NOTFOUND ){ + p->rc = SQLITE_ERROR; + p->zErrmsg = sqlite3_mprintf("rbu vfs not found"); + } +} + +/* +** This routine is a copy of the sqlite3FileSuffix3() routine from the core. +** It is a no-op unless SQLITE_ENABLE_8_3_NAMES is defined. +** +** If SQLITE_ENABLE_8_3_NAMES is set at compile-time and if the database +** filename in zBaseFilename is a URI with the "8_3_names=1" parameter and +** if filename in z[] has a suffix (a.k.a. "extension") that is longer than +** three characters, then shorten the suffix on z[] to be the last three +** characters of the original suffix. +** +** If SQLITE_ENABLE_8_3_NAMES is set to 2 at compile-time, then always +** do the suffix shortening regardless of URI parameter. +** +** Examples: +** +** test.db-journal => test.nal +** test.db-wal => test.wal +** test.db-shm => test.shm +** test.db-mj7f3319fa => test.9fa +*/ +static void rbuFileSuffix3(const char *zBase, char *z){ +#ifdef SQLITE_ENABLE_8_3_NAMES +#if SQLITE_ENABLE_8_3_NAMES<2 + if( sqlite3_uri_boolean(zBase, "8_3_names", 0) ) +#endif + { + int i, sz; + sz = (int)strlen(z)&0xffffff; + for(i=sz-1; i>0 && z[i]!='/' && z[i]!='.'; i--){} + if( z[i]=='.' && sz>i+4 ) memmove(&z[i+1], &z[sz-3], 4); + } +#endif +} + +/* +** Return the current wal-index header checksum for the target database +** as a 64-bit integer. +** +** The checksum is store in the first page of xShmMap memory as an 8-byte +** blob starting at byte offset 40. +*/ +static i64 rbuShmChecksum(sqlite3rbu *p){ + i64 iRet = 0; + if( p->rc==SQLITE_OK ){ + sqlite3_file *pDb = p->pTargetFd->pReal; + u32 volatile *ptr; + p->rc = pDb->pMethods->xShmMap(pDb, 0, 32*1024, 0, (void volatile**)&ptr); + if( p->rc==SQLITE_OK ){ + iRet = ((i64)ptr[10] << 32) + ptr[11]; + } + } + return iRet; +} + +/* +** This function is called as part of initializing or reinitializing an +** incremental checkpoint. +** +** It populates the sqlite3rbu.aFrame[] array with the set of +** (wal frame -> db page) copy operations required to checkpoint the +** current wal file, and obtains the set of shm locks required to safely +** perform the copy operations directly on the file-system. +** +** If argument pState is not NULL, then the incremental checkpoint is +** being resumed. In this case, if the checksum of the wal-index-header +** following recovery is not the same as the checksum saved in the RbuState +** object, then the rbu handle is set to DONE state. This occurs if some +** other client appends a transaction to the wal file in the middle of +** an incremental checkpoint. +*/ +static void rbuSetupCheckpoint(sqlite3rbu *p, RbuState *pState){ + + /* If pState is NULL, then the wal file may not have been opened and + ** recovered. Running a read-statement here to ensure that doing so + ** does not interfere with the "capture" process below. */ + if( pState==0 ){ + p->eStage = 0; + if( p->rc==SQLITE_OK ){ + p->rc = sqlite3_exec(p->dbMain, "SELECT * FROM sqlite_master", 0, 0, 0); + } + } + + /* Assuming no error has occurred, run a "restart" checkpoint with the + ** sqlite3rbu.eStage variable set to CAPTURE. This turns on the following + ** special behaviour in the rbu VFS: + ** + ** * If the exclusive shm WRITER or READ0 lock cannot be obtained, + ** the checkpoint fails with SQLITE_BUSY (normally SQLite would + ** proceed with running a passive checkpoint instead of failing). + ** + ** * Attempts to read from the *-wal file or write to the database file + ** do not perform any IO. Instead, the frame/page combinations that + ** would be read/written are recorded in the sqlite3rbu.aFrame[] + ** array. + ** + ** * Calls to xShmLock(UNLOCK) to release the exclusive shm WRITER, + ** READ0 and CHECKPOINT locks taken as part of the checkpoint are + ** no-ops. These locks will not be released until the connection + ** is closed. + ** + ** * Attempting to xSync() the database file causes an SQLITE_INTERNAL + ** error. + ** + ** As a result, unless an error (i.e. OOM or SQLITE_BUSY) occurs, the + ** checkpoint below fails with SQLITE_INTERNAL, and leaves the aFrame[] + ** array populated with a set of (frame -> page) mappings. Because the + ** WRITER, CHECKPOINT and READ0 locks are still held, it is safe to copy + ** data from the wal file into the database file according to the + ** contents of aFrame[]. + */ + if( p->rc==SQLITE_OK ){ + int rc2; + p->eStage = RBU_STAGE_CAPTURE; + rc2 = sqlite3_exec(p->dbMain, "PRAGMA main.wal_checkpoint=restart", 0, 0,0); + if( rc2!=SQLITE_INTERNAL ) p->rc = rc2; + } + + if( p->rc==SQLITE_OK && p->nFrame>0 ){ + p->eStage = RBU_STAGE_CKPT; + p->nStep = (pState ? pState->nRow : 0); + p->aBuf = rbuMalloc(p, p->pgsz); + p->iWalCksum = rbuShmChecksum(p); + } + + if( p->rc==SQLITE_OK ){ + if( p->nFrame==0 || (pState && pState->iWalCksum!=p->iWalCksum) ){ + p->rc = SQLITE_DONE; + p->eStage = RBU_STAGE_DONE; + }else{ + int nSectorSize; + sqlite3_file *pDb = p->pTargetFd->pReal; + sqlite3_file *pWal = p->pTargetFd->pWalFd->pReal; + assert( p->nPagePerSector==0 ); + nSectorSize = pDb->pMethods->xSectorSize(pDb); + if( nSectorSize>p->pgsz ){ + p->nPagePerSector = nSectorSize / p->pgsz; + }else{ + p->nPagePerSector = 1; + } + + /* Call xSync() on the wal file. This causes SQLite to sync the + ** directory in which the target database and the wal file reside, in + ** case it has not been synced since the rename() call in + ** rbuMoveOalFile(). */ + p->rc = pWal->pMethods->xSync(pWal, SQLITE_SYNC_NORMAL); + } + } +} + +/* +** Called when iAmt bytes are read from offset iOff of the wal file while +** the rbu object is in capture mode. Record the frame number of the frame +** being read in the aFrame[] array. +*/ +static int rbuCaptureWalRead(sqlite3rbu *pRbu, i64 iOff, int iAmt){ + const u32 mReq = (1<mLock!=mReq ){ + pRbu->rc = SQLITE_BUSY; + return SQLITE_INTERNAL; + } + + pRbu->pgsz = iAmt; + if( pRbu->nFrame==pRbu->nFrameAlloc ){ + int nNew = (pRbu->nFrameAlloc ? pRbu->nFrameAlloc : 64) * 2; + RbuFrame *aNew; + aNew = (RbuFrame*)sqlite3_realloc64(pRbu->aFrame, nNew * sizeof(RbuFrame)); + if( aNew==0 ) return SQLITE_NOMEM; + pRbu->aFrame = aNew; + pRbu->nFrameAlloc = nNew; + } + + iFrame = (u32)((iOff-32) / (i64)(iAmt+24)) + 1; + if( pRbu->iMaxFrame